Nucl - Phys.B v.739 PDF

Nuclear Physics B 739 (2006) 159
Complex conformal spin partial wave expansion

of generalized parton distributions and distribution
amplitudes
D. Mller a,b, , A. Schfer a
a Institut fr Theoretische Physik, Universitt Regensburg, D-93040 Regensburg, Germany
b Department of Physics and Astronomy, Arizona State University, Tempe, AZ 85287-1504, USA
Received 10 October 2005; received in revised form 6 December 2005; accepted 5 January 2006
Available online 2 February 2006
Abstract
We introduce a new representation of generalized parton distributions and generalized distribution amplitudes that is based on the partial wave decomposition with respect to the complex collinear conformal spin.
This decomposition leads us to a versatile parameterization of these non-perturbative functions in terms
of conformal moments, which are measurable for integer value on the lattice. This new representation has
several advantages: basic properties and crossing relations are automatically implemented, a rather flexible
parameterization is possible, the numerical treatment of evolution is simple and analytic approximation of
scattering amplitudes can be given. We demonstrate this for simple examples. In particular, phenomenological considerations indicate that the t-dependence of Mellin moments is governed by Regge trajectories.
The new representation is vital to push the analysis of deeply virtual Compton scattering to next-to-next-toleading order.
2006 Elsevier B.V. All rights reserved.
1. Introduction
Generalized parton distributions (GPDs) [15] and their analog, obtained by crossing, generalized distribution amplitudes (GDAs) [2,6] are non-perturbative functions that are accessible in
certain hard exclusive processes such as the hard electroproduction of photons [7] and mesons
[8] off a nucleon or nucleus or hadron pair production by two photon fusion in e+ e colliders.
* Corresponding author.
E-mail address: dieter.mueller@physik.uni-regensburg.de (D. Mller).

0550-3213/$ see front matter 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.nuclphysb.2006.01.019
D. Mller, A. Schfer / Nuclear Physics B 739 (2006) 159
These functions are related to parton densities, form factors, and distribution amplitudes but contain additional non-perturbative information about the internal structure of hadrons and nuclei.
Some of this information can even not be obtained in any other way than through a global fit
of GPDs. This has been widely realized for the first time in connection with the proton spin
puzzle. Here the second moment of a certain combination of GPDs provides the orbital angular
momentum fraction carried by quarks of a given flavour [3]. Moreover, GPDs simultaneously
possess a longitudinal and transverse momentum dependence and so it has been pointed out that
they encode a three-dimensional femto-holographic picture of the probed hadron or nucleus [9].
Indeed, it could be shown that a partonic density interpretation holds in the infinite momentum
frame as long as the longitudinal momentum fraction in the t-channel is vanishing [10,11], see
also Refs. [12,13]. More precisely, in the impact parameter space GPDs are interpreted as the
probability to find a parton species i with momentum fraction x at a relative distance b from
the proton center. Even an interpretation of the three-dimensional Fourier transform of GPDs
in the rest frame has been suggested within the concept of phase space (Wigner) distributions
[14,15]. For further details we refer to the comprehensive reviews in Refs. [16,17].
At present generalized parton distributions are one of the main topics of collider and fixed target experiments at DESY and JLAB. Further experiments are planed or proposed for COMPASS
and ERIC. Also it should be mentioned that information on generalized distribution amplitudes
has, e.g., been extracted from LEP data. Unfortunately, the wealth on information encoded in
GPDs and GDAs goes along with their functional complexity. For instance, GPDs depend on
both the momentum fractions in the s- and t-channel, x and , the momentum transfer squared t,
the resolution scale Q, and the quantum numbers of the target and the probed parton. This multitude of functional dependencies is, however, very strongly constrained by their relation to parton
densities, form factors, and distribution amplitudes and even more so by crossing relations, positivity bounds and Lorentz invariance in general. The latter implies in particular that the Mellin
x-moments of GPDs must be polynomials of given order in the skewness parameter .
Unfortunately, there is still another complication. Typically experimental observables allow
only to determine convolutions including GPDs or GDAs and a formal deconvolution can practically not be done for most of the processes.1 To determine GPDs or GDAs from experimental
data one therefore needs anstze for them which involve only a minimal number of parameters.
Although, on the theoretical side GPDs have been intensively studied in the last few years, only a
few of such parameterizations have been proposed and used in phenomenology. Perhaps the most
popular parameterization is based on the ansatz suggested by Radyushkin, in which by construction the relation to parton densities, form factors, and polynomiality is assured. However, this
advantages arise from the simplicity of this ansatz which also might implement a certain rigidity.
Especially, it is widely used in combination with a factorized t-dependence although the latter is
known to be wrong. This factorization does not respect the disappearance of the t-dependence
for x 1 [21,22] and is also basically ruled out by lattice results [2325]. How far the employed
versions of this ansatz are suited for the kinematics accessible in present experiments remains an
open question. Obviously, it is highly desired to have a versatile parameterization of GPDs and
GDAs, which respects all of their formal properties.
The main idea guiding the search for a more appropriate parameterization of GPDs and GDAs
is that the relevant kinematic variables should be separated in this new representation. Let us re1 This is actually only possible if the hadron is probed with two virtual photons and the virtuality of both photons can
be independently varied, which is experimentally an extremely challenging task [1820].
mind how helpful the representation of parton densities by Mellin moments has proven to be for
the analysis of hard inclusive processes. Mellin moments are given by the analytic continuation
of the forward expectation values of leading twist-two operators with given spin J . The main advantage of the Mellin space is that operators with different spin J do not mix under evolution and
so the solution of the evolution equations is trivial. In the case of GPDs and GDAs leading twisttwo operators can contain total derivatives and so the operator basis has to be chosen differently,
in such a way that the operators again do not mix under evolution. The appropriate operator basis
is given in terms of collinear conformal operators that are labelled by the (collinear) conformal
spin and the normal spin of the operator. The former quantum number characterize the irreducible
multiplets or conformal towers of the collinear conformal algebra while the latter denotes the
members of a given multiplet. Group theoretically we are dealing with the representation of the
so-called collinear conformal algebra so(2, 1) which is a subalgebra of the full conformal algebra
so(4, 2). Let us remark that except for the trace anomaly, which is proportional to the renormalization group coefficient (s ), conformal symmetry is preserved in perturbative QCD. Even in
the case of a non-vanishing function the conformal representations can be changed in such a
way that the evolution operator is diagonal. The evolution equation can then be solved trivially.
The conformal moments at the input scale depend on the skewness parameter and can be expanded in an appropriate orthogonal polynomial basis where the expansion coefficients depend
on the momentum transfer squared. In other words, GPDs and GDAs can be represented by a
conformal partial wave expansion, where the expansion coefficients are characterized by form
factors that are labelled by the conformal spin and by an appropriate second quantum number,
e.g., the angular momentum. Most importantly, it has been demonstrated that the first few form
factors are measurable on the lattice. Moreover, the crossing relation between GPDs and GDAs
are reduced in this representation of the continuation of these form factors from the space- to the
time-like region and reverse.
Group theoretical discussions based on discrete conformal spin have a long history in QCD.
However, in practice, they seemed to be only useful for distribution amplitudes and GDAs, where
the resulting series convergence [26,27]. Combined with conformal symmetry predictions, the
perturbative corrections for (virtual) two photon processes in the generalized Bjorken limit can be
worked out, e.g., for the photon-to-pion transition form factor, to next-to-next-to-leading (NNLO)
order accuracy [28,29].
For GPDs the conformal partial wave expansion, where the conformal spin is a non-negative
integer, is represented by a series in terms of mathematical distributions, which only converges if
it is convoluted with suitable test functions. An appropriate resummation of this series has been
proposed in Ref. [30] and the details are presented here. There are also several other suggestions
in the literature to define out of this divergent series. One might insert the identity expanded in
terms of polynomials, which is however only applicable for a certain kinematical region [31,32],
and for a next-to-leading (NLO) analysis see Refs. [33,34]. One can also represent the identity by
its Fourier transform which makes contact to the group theoretical representation with complex
valued conformal spin [35], later adopted for GPDs in Ref. [36] and more recently in Ref. [37].
Also a resummation by an integral transformation has been suggested [38], which, however, under close scrutiny turned out to be unpracticable or at least rather complicated [39]. An attempt
to approximately resum the conformal partial wave expansion within a Taylor expansion of conformal moments has been suggested in Ref. [40]. Of course, all these proposals can be related
to each other. However, because of the intricate mathematics involved a correct, complete and
efficient resummation of conformal partial waves has not yet been worked out.
In this paper we employ the SommerfeldWatson transformation to resum the conformal partial wave expansion of GPDs, finally yielding a MellinBarnes integral for GPDs. The resulting
representation is similar to the one recently proposed in Ref. [37], however, not identical. The
SommerfeldWatson transformation requires the analytic continuation of the conformal spin,
which plays here the analogous role to the complex spin J in the inverse Mellin representation of
parton densities. While the analytic continuation of Mellin moments for parton distribution functions is a rather simple task, it is a highly non-trivial one for the conformal moments of GPDs.
This central problem is solved by us to an extent such that the framework can be applied as soon
as a GPD ansatz is given. Although the final GPD representation as a MellinBarnes integral over
the complex conformal spin seems to be rather complicated it has several advantages. The dependence on kinematic variables is separated in this representation, it allows a simple and stable
numerical treatment of GPDs and their convolution with hard-scattering amplitudes, and can be
used for the analytic approximation of scattering amplitudes. Moreover, the evolution equations
to leading order (LO) accuracy are trivially solved and the conformal approach in Refs. [28,29]
can be adopted for the study of higher order corrections in perturbative QCD. Especially, the
NNLO corrections to deeply virtual Compton scattering (DVCS) are calculable in a rather economic manner [41]. Also the use of the crossing relation between GDAs and GPDs is possible in
our representation and should be very valuable for phenomenology.
The paper is organized as following. Section 2 is devoted to the conformal partial wave decomposition of GPDs for complex conformal spin. We start with a review on the anatomy of GPDs
in Section 2.1 and discuss the extension of the GPD support [30]. To the best of our knowledge
this issue has not been presented in detail so far. Here it will guide us to find the correct treatment
of partial waves with complex conformal spin. We consider then the crossing relation between
GDAs and GPDs and derive in Section 2.3 the new GPD and GDA representations in terms of
MellinBarnes integrals. In Section 3 we consider evolution kernels and their convolution with
GPDs in a scheme that preserves conformal symmetry. Moreover, we present the MellinBarnes
integral for the scattering amplitude in DVCS and discuss its analytic approximation. In Section 4
we have a closer look to the analytic continuation procedure of conformal moments for a simple
GPD toy ansatz. Then we address the issue of anstze for conformal moments and explore the
features of the resulting GPDs and GDAs. Furthermore, for vanishing longitudinal momentum
fraction in the t channel we have a short look at valence quark GPDs. Motivated by lattice results [2325], we introduce a parameterization for which the experimental constraints on GPDs
indicate that leading Regge trajectories are present in conformal moments. Finally, we summarize and conclude. In Appendix A integrals, which are used in the main text, and the rotation
from ordinary Mellin moments to conformal ones are presented. Appendices B and C contain
the MellinBarnes integrals for gluonic GPDs and conformal evolution kernels, respectively.
2. Features and parameterization of GPDs
2.1. The anatomy of GPDs
GPDs are defined as Fourier transform of light-ray operators, sandwiched between the initial
and final hadronic states. There is a whole compendium of GPDs for each hadron. In addition
the initial and final states can have different quantum numbers (transition GPDs), and one even
can replace the hadrons by nuclei (nucleus GPDs). Once the initial and final states are specified,
GPDs are classified with respect to the twist of the operators and the spin content of fields. At
leading twist-two level three different types of quark and gluon GPDs can be defined (here the
gauge link is omitted):

q V

+
F

d
qF A
x, , 2 , 2 =
eixP+ P2 , S2 | qr (n) + 5 qr (n)|P1 , S1 ,
2
qF T
i+
G V
F

GF A
x, , 2 , 2
GF T

g
d ixP+
a
e
P2 , S2 |G+ (n) i+ Ga+ (n)|P1 , S1 ,
=2
P+
(1)
(2)
with P+ = n (P1 + P2 ), V = n V , n2 = (n )2 = 0, n n = 1. In the first (vector) and

second (axial-vector) entry the in- and out-going partons have the same helicities, and the sum
(vector) and difference (axial-vector) of left- and right-handed partons is taken, respectively.
For the third entry, called transversity, a helicity flip appears. GPDs depend on the momentum
fraction x, conjugated to the light-cone distance 2, the longitudinal momentum fraction =
(P1 P2 )+ /(P1 + P2 )+ in the t-channel,2 the momentum transfer 2 t = (P2 P1 )2 , and
the renormalization scale 2 . The latter is induced by the renormalization prescription of the
operators, which is part of the GPD definition. To deal with the polarization of the hadronic
states, one might introduces a form factor decomposition [42,43]. For instance, for the nucleon
GPD Dirac and Pauli-like form factors appear in the vector case [42]:

i+
U (P1 , S1 )Ei x, , 2 ,
F V = U (P2 , S2 )+ U (P1 , S1 )Hi x, , 2 + U (P2 , S2 )
2M
(3)
where i = u, d, s, . . . , G. To avoid confusion, let us note that for the process (q1 ) + p(P1 )
(q2 ) + p(P2 ) two scaling variables exist. They are denoted as and and are defined by [2]:
i
q 2
,
P q
q
P q
1
with q = (q1 + q2 ).
2
(4)
Both variable coincide, up to power suppressed corrections O(2 /Q2 ), when the outgoing photon is real, i.e., for DVCS one can simply replace by . In this paper we will treat the general
case.
The definitions (1)(3) imply the basic properties of GPDs:
In the forward limit 0 helicity non-flip GPDs reduce to parton densities [25], e.g.,

qi x, 2 = lim Hi x, , 2 , 2
(5)
0
and the helicity flip GPDs Ei decouple, but

lim Ei x, , 2 , 2 = 0.
0
(6)
2 In the literature is now denoted as , which also is the Bjorken like scaling variable in hard inelastic exclusive
processes. To be precise, we distinguish between both variables. The sign convention for is fixed, for it is changing.
For quantities which are even under reflection, i.e., , the sign convention is irrelevant. Here we define in such
a way that it corresponds to the variable , commonly used in the definition of GPDs, too.
The 2 -dependence is governed by linear evolution equations [1,2,44], which have been
evaluated in the momentum fraction representation [44] or derived from the renormalization
group equation of the light-ray operators [45,46] (for details see Ref. [17]).
Hermiticity [2] together with time reversal invariance [42] leads to a definite symmetry with
respect to the skewness parameter , e.g., Hi (x, ) = Hi (x, ).
The Mellin moments of GPDs are expectation values of local twist-two operators:

dx x n q F V x, , 2 , Q2
P+n+1
n0 nn P2 , S2 |S qr 0 i D 1 i D n qr |P1 , S1 ,
(7)
where D = D D is the covariant derivative, acting as indicated by the arrows, and the
operator S symmetrizes all indices and subtracts the traces. Lorentz covariance enforces that
this moments are polynomials in .
Furthermore, GPDs are constrained in the region x || by the positivity of the norm in the
Hilbert space of states. The most general form of such positivity bounds [4749], known so far,
are given as an infinite set of constraints [50,51]. Such constraints can be alternatively understood
within the representation of GPDs as overlap of light-cone wave functions [52].
Let us consider the support of a GPD in more detail.3 A generic quark GPD F (x, , 2 ), e.g.,
in the vector case, is related to a double distribution (DD) D(y, z, 2 ) by the integral transformation [2,54]
F x, ,
1
=
1
1|y|

dy

dz x p (x y z)D y, z, 2 ,
p = {0, 1}.
(8)
1+|y|
Here D(y, z, 2 ) is an even function in z so that F (x, , 2 ) has the proper symmetric behavior
under the exchange . Obviously, its Mellin moments, i.e., dx x n F (x, , 2 ) are even
polynomials in , since the support of D(y, z, 2 ) is restricted. Depending on the form factors
appearing in the decomposition of the GPDs (1) and (2), see, for instance, Eq. (3), the order
of the polynomial is n or n + 1. To treat both cases in a convenient and generic manner, we
have included in Eq. (8) the factor x p with p = 0 (p = 1) in the former (latter) case [55]. This
restores the correct order of the polynomials.4 We can now fix to be positive and decompose
the integration with respect to y into y > 0 and y < 0. This results into a decomposition of
F (x, , 2 ) in its quark q and anti-quark q part:

F x, , 2 = q x, , 2 q x, , 2 .
(9)
3 The GPD support might be directly derived by means of a partonic Fock state decomposition and the so-called
-representation for Feynman diagrams, see, for instance, Ref. [53]. Equivalently, a GPD can be expressed in terms of
a DD [2,54], which has a simpler structure, and one might consider it as more convenient to derive the support of the
former from that of the latter.
4 Note that within p = 0 an additive so-called D-term was proposed to generate n+1 terms [56]. It is only non-zero
in the restricted region |x| || and is contained in our representation with p = 1 as an additive term of D(y, z, 2 ) that
is proportional to (y). Our parameterization offers the possibility that the n+1 terms arise from an uniform GPD.
Here both functions separately satisfy the polynomiality condition:

q x, , 2 =
1
1y

dy

p = {0, 1}
(10)
1+y
and analogous for the anti-quark GPD q(x,

, 2 ), where D(y, z, 2 ) is replaced by D(y,
z,
2
p
2
) = (1) D(y, z, )

q x, , 2 =
1
1y

dy

p = {0, 1}.
(11)
1+y
Obviously, both quark and anti-quark GPDs have the same mathematical representation and so
we will in the following mainly deal with the quark one. The results for anti-quark GPDs are
easily obtained by replacements D D.

In the central (exclusive or ER-BL) region x , q(x, , 2 ) might be interpreted
as probability amplitude to have a meson like configuration inside the hadron, while the outer
(inclusive or DGLAP) region x 1 can be viewed as probability amplitude for emission and
x
absorbtion of a quark with momentum fraction x1 P1 = x+
1+ P1 and x2 P1 = 1+ P1 , respectively,
see Fig. 1. Remarkably, both regions have a dual interpretation, namely, as meson and parton
exchange in the t and s channel, respectively.
Lorentz invariance ties both dual regions, which can be read off from the representation (10)
that ensures polynomiality. Suppose 0, the z integration in the double distribution representation (10) can be trivially performed5 and leads to the support

q x, , 2 = ( x 1) x, , 2 + ( x 1) x, , 2 .
(12)
The function follows from the y integration in Eq. (10)
x+

1
x, , 2 =
1+

dy x p D y, (x y)/, 2 .
(13)
The GPD representation (12) is manifestly invariant under the transformation , especially, the support x 1 remains untouched.
Fig. 1. Partonic interpretation of GPDs in the central (left) and outer (right) region.
5 Since > 0, we have set (x y z) = 1/(x/ y/ z) rather than to indicate the modulus 1/||. The sign
convention of (x, ) and its transformation under reflection avoids an overall sign() factor in Eq. (12). This
allows us to treat (x, ) as a holomorphic function in the complex plane.
In the central region the GPD is given by (x, , 2 ) from which the outer region, determined
by the symmetrized function (x, , 2 ) + (x, , 2 ), can be restored, see Fig. 2. Let us have
a closer look at this continuation. In the central region the integration variable in the integral
(13) takes the values 0 y x+
1+ 1. The restriction of the second argument in the double
distribution |z| = |(x y)/| 1 y is ensured by the values of the lower and upper limit,
see solid line in Fig. 2(a). The integration path, starting at y = 0 and z = x/, lies inside of the
DD support as it must be. At the cross-over point x = it starts at the support edge y = 0 and
z = 1 and in the outer region the lower limit is y = x
1 rather than zero. We can now define an
(ambiguous) continuation of the DD support for |z| > 1 y by any smooth function D(y,
z, 2 )
symmetric in z.

D y, (x y)/, 2

x
x
D y, (x y)/, 2 y
+ D y, (x y)/, 2
y . (14)
1
1
This provides the (ambiguous) continuation of (x, , 2 ) into the outer region. (x, , 2 )
is obtained by reflection symmetry and adding both contributions leads to the integral
representation
x+

1
x, , 2 + x, , 2 =
1+

dy x p D y, (x y)/, 2 ,
(15)
x
1
in which the integration runs only over the original support of the DD. Hence, the ambiguity in
the continuation of (x, , 2 ) and (x, , 2 ) drops out in their sum.
Let us suppose that D(y, z, 2 ) can be viewed as a holomorphic function of y and z inside its
support and has branch cuts outside of it. Then the integration path in Fig. 2(a) can cross or go
along such branch cuts. To deal with a unique definition of (x, ) for all values of x we might
(a)
(b)
Fig. 2. Support of a DD (the area surrounded by thick lines) and integration path for the calculation of (x, ), see
Eq. (13), for the central region x1 < (solid) and the outer region < x2 (dashed) are shown in (a). The continuation of
the integration path over the support boundary is indicated as dotted line. In (b) the support of the resulting GPD (12) is
depicted.
define its value within its integral representation (13) for x by the principal value prescription
1
1
(x + i, ) + (x i, )
2
2
for < x.
(16)
For illustration we give a simple example for the Radyushkin ansatz

1
q(x, ) =
1y

dz (x y z)
dy
0
1+y

q(y)
|z|/(1 y) ,
1y
(17)
where the 2 dependence is disregarded. The parton density is parameterized by a toy ansatz
q(y) = y (1 y) /B( + 1, + 1), which can be considered as building block for realistic
parameterizations. A popular ansatz for the profile function is
(z) = (z|b) =
(1 z2 )b
,
B(b + 1, 1/2)
B(x, y) =
(x)(y)
,
(x + y)
(18)
where the parameter b controls the strength of the skewness effect. This GPD ansatz can be
evaluated in an analytically form in terms of hypergeometric functions for non-negative integer
value of the parameter b. We might choose here for simplicity b = 0, i.e., (z, 0) = 1/2. In the
central region this toy GPD reads

(2 + + )
x + 1+
1 + , 1 x +
for
F
2 1
1+
2+
2(2 + )(1 + ) 1 +
x .
q(x, ) =
(19)
The extension of the DD support corresponds to the analytic continuation of (x, ) into the
outer region and so we find:
q(x, ) =

(2 + + )
x + 1+
1 + , 1 x +
F
2 1
1+
2+
2(2 + )(1 + ) 1 +

x 1+
1 + , 1 x
for x .
2 F1
1
2+
1
(20)
We remark that the rescaled distribution

1y
1

X 1 2
, , = dy
q
dz X p (X yY z)D y, z, 2 ,
Y Y
p1
p = {0, 1}
1+y
(21)
has technically the same support as it appears in evolution kernels. Defining by

1+X

X, Y, 2 =
1+Y
0
we have

dy X p D y, X Yy, 2 ,
p = {0, 1},
(22)
10

X 1 2
, ,
Y
q
Y Y

Y X
1+X
X X
2
X, Y, +
.
= sign(1 + Y )
Y Y
1+Y
1+Y

p1
(23)
The evolution kernels have the form (23) in the collinear limit 2 = 0. This enables us to adopt
results for the representation of GPDs to the representation of GDAs (see next section) and of
kernels. Needless to say, the GPD (12) follows from X = x/ and Y = 1/, where (x, , 2 ),
see Eq. (13), is related to (X, Y, 2 ) by the formula

x 1 2
2
p1
, , .

x, , =
(24)

2.2. Generalized distribution amplitudes and crossing
Another ingredient we need for our derivation of a MellinBarnes representation of GPDs
is their extension to 1 < . This immediately leads us to further non-perturbative distributions,
the so-called GDAs [2,6], which are related to GPDs by crossing [57]. The GDAs, denoted as
(z, , W 2 ), are defined in analogy to the GPDs in Eqs. (1) and (2), however, the initial state is
replaced by the vacuum and the final one contains two hadrons. For instance, the crossing analog
of q F V for a spin-zero target reads

d i(12z)P+
q V
e
z, , W 2 , 2 =
0| qr (n)+ qr (n)|P1 , P2
(25)
2
with P+ = n (P1 + P2 ). Here 0 z 1 and 1 z are the momentum fractions of the quark
and anti-quark, respectively, which produce the hadron pair with invariant mass squared W 2 .
0 1 is the momentum fraction of one of the hadrons.
In the following we give a rather generic discussion of the crossing relation to GPDs, which
neglects details about form factor decomposition or quantum numbers. We consider also only
a quark GPD and its analog. For anti-quarks the only additional aspect, that one has to implement, is the sign convention in the decomposition (9). The GDA reads in terms of the rescaled
distribution (21) with X = 1 2z, Y = 1 2 , and 2 = W 2 as

X 1 2
, ,
.
z, , W 2 = Y p1 q
(26)
Y Y
X=12z, Y =12, 2 =W 2
Consequently, from Eq. (23) we read off its representation

z, , W 2 = (z ) 1 2z, 1 2, W 2 + ( z) 2z 1, 2 1, W 2 , (27)
where the function is given by the integral (22). Having in mind that in Eq. (26) a rescaled
GPD appears, it remains a trivial exercise to directly relate GPDs and GDAs6 :

1p

(z, , W 2 ),
q(x, , 2 )
,
(1 2z, 1 2, W 2 )
1p (x, , 2 )
6 Note that the upper line in Eq. (28) is only valid for 1 < , i.e., < 1/2. For 1/2 < one must replace 1p by
sign()1p . It is more convenient to use the second line, valid for all values of , together with the explicit representations (12) and (27), see also footnote 5.
1 2z
x
,
1
1 2 ,
W 2 2 ,
11
(28)
where the lower line is in fact the relation (24).

Let us shortly discuss this crossing relation. A GDA is obtained from the GPD analog by

1
1 2z
2
p1
2
,
,W .
1 2z, 1 2, W = (1 2 )
(29)
1 2 1 2
Since 0 1, the second argument of covers the region |1/(2 1)| 1. Thus, the crossing
relation requires to enter a kinematic region which is unphysical for GPDs, except for the point
= 1, i.e., = 0. However, for a given functional form of a GPD, the relation (29) allows to fix
its phenomenological parameters. So for instance, the same function appears after factorization
in hard exclusive electroproduction of a photon or mesons on a hadron target as GPD and in the
production of a hadron pair due to two photon fusion as GDA. The knowledge of the analytic
form of the GPD in the central region is sufficient to perform the symmetry transformation (29) to
obtain the corresponding GDA. Reversely, a GPD follows from a given GDA using the symmetry
transformation (24). As expected from general reasons, the physical and unphysical regions are
again connected by crossing. Moreover, as we realized above, (x, , 2 ) is not uniquely defined
in the outer region. As explained above this problem is artificial and does not affect the net
contribution in this region.
Let us stress ones more that the extension procedure of the support is unique, which was shown
in connection with the support extension of evolution kernels [1,2]. Here we adopt the same
arguments. Suppose we know the function (1 2z, 1 2, 2 ) in the region 0 z 1,
which is equivalently to the knowledge of the GDA (z, , W 2 ), see Eq. (27). Next representing
the GDA (26) in terms of the DD (21), we realize that its convolution with any holomorphic test
function (z) yields an holomorphic function in . For instance, one finds for p = 0
1
dz (z) z, , W

2
1
=
2
1
1y

dy
0
1+y

1 y(1 2 ) z
D y, z, W 2 .
dz
2

(30)
Hence, also the Fourier transform of the GDA with respect to z is a holomorphic function in
the conjugate variables and . Consequently, we can employ analytic continuation. Then the
inverse Fourier transform together with the crossing relation (28) yields the result we desire,

q x, , 2 = sign()p1

1

d ix/
e
AC 2 dz ei(12z) z, , W 2 ,
2
(31)
where = ( 1)/2, W 2 = 2 and AC denotes the analytic continuation of both variables

and . (Actually, the variables stays real and analytic continuation is only used to extend their
numerical values on the real axis, e.g., for up to infinity.)
2.3. Complex collinear conformal spin partial wave expansion
We derive now a new representation for GPDs which has several advantages, already mentioned in the introduction. In fact we will deal with a partial wave decomposition of GPDs, where
the partial waves are labelled by the complex conformal spin, the quantum number which characterize the multiplets (towers) of conformal operators. This is rather analogous to the partial wave
12
expansion of scattering amplitudes with respect to the complex angular momentum, however,
requires a more attentive consideration. Irrespectively, of whether the symmetry is preserved or
not, one can introduce such an expansion. Certainly, conformal symmetry is broken in the nonperturbative QCD sector. Fortunately, up to calculable corrections proportional the non-vanishing
function, it holds true in the perturbative sector and thus has nevertheless still predictive power.
In the following we consider only quark GPDs, the gluon case can be treated analogously and
the results are collected in Appendix B. To derive the partial wave decomposition of GPDs, the
following steps will be performed:
First the conformal moments and partial wave expansion of GPDs are introduced for discrete
conformal spin. Here GPDs are represented as a divergent series in terms of mathematical
distributions.
This series will be summed in the unphysical region, where it can be viewed as an ordinary expansion in terms of orthogonal polynomials, by means of the SommerfeldWatson
transformation, which requires the analytic continuation of the conformal spin.
Finally, we complete the SommerfeldWatson transformation and derive a representation of
GPDs in terms of a MellinBarnes integral for the central region. The outer region follows
from an suitable continuation in x that arises from the analytic structure of GPDs.
2.3.1. Conformal partial wave expansion
So-called conformal moments of quark GPDs are formed with respect to Gegenbauer poly3/2
nomials n Cn (x/) with index 3/2 and order n. These moments are given by the expectation
value of local conformal operators, see below Eq. (37). Relevant group theoretical aspects can
be found in Appendix B and in the review [58]. These moments can be viewed as an appropriate generalization of the ordinary forward Mellin moments, used in the analysis of deep inelastic
scattering. The Gegenbauer polynomials are orthogonal polynomials, possess a definite reflection
3/2
3/2
symmetry Cn (x) = (1)n Cn (x), and are the only solution of the second order differential
equation
3/2
d2
3/2
1 x 2 Cn (x) = (n + 1)(n + 2)Cn (x)
(32)
2
dx
that is finite at the singular points x = 1. These polynomials form a complete basis in the
interval [1, 1]. Here we rescale the polynomials and choose the normalization

x
(3/2)(1 + n) 3/2
n
with cn (x) = n
Cn (x)
cn (x, ) = cn
(33)
2 (3/2 + n)
in such a way that in the forward case the ordinary Mellin moments appear:
lim cn (x, ) = x n .
(34)
There are several possibilities to express cn (x, ) in terms of hypergeometric functions, which
might provide different prescriptions for the analytic continuation of the discrete variable n.
Below we will use

(3/2)(3 + j ) j
j, j + 3 x
2 F1
cj (x, ) = 1+j
(35)
2
2
2 (3/2 + j )
for complex valued j . Equivalently, we can express it in terms of (associated) Legendre functions
of the first kind.
13
The conformal moments of a GPD, separated into quark and anti-quark ones, are defined as

mn , 2 =
1

dx cn (x, )q x, , 2 ,

m
n , 2 =
1

dx cn (x, )q x, , 2 .
(36)
They are given by the expectation values of collinear conformal operators, e.g., in the vector
case,

mn , 2 (1)n m
n , 2

(3/2)(1 + n)n
3/2 i D +
r
qr |P1 , S1 .
P2 , S2 |q + Cn
= n
(37)
2 (3/2 + n)P+
P+
The rotation to the ordinary Mellin moments (7) and its inversion are given in Appendix A by
Eqs. (A.11) and (A.12). The operators are characterized by the conformal spin, which in our
case is n + 2. The (conformal) moments can be either calculated on the lattice,7 can be directly
modeled, or evaluated from a given GPD ansatz. In the latter case they are naturally decomposed
as, e.g., for quarks,

mn , 2 = n , 2 + n , 2 ,

n , 2 =
1

dx cn (x, ) x, , 2 .
(38)
This is a simple consequence of the symmetry relations cn (x, ) = cn (x, ) together with the
representation (12). The n (, 2 ) are only defined in terms of (x, , 2 ) and, thus, they can
be quite general functions of . However, after symmetrization with respect to , see first formula
in Eq. (38), one obtains the polynomial mn (, 2 ).
Now we would like to invert the transformation (36). As mentioned above the polynomials
cn (x, ), see Eqs. (33) and (34), form only a complete basis in the central region [, ]. Let us
denote by pn (x, ) the polynomials that include the weight (1 x 2 ) and an appropriate normalization

x
pn (x, ) = n1 pn
,
3/2

2n (5/2 + n)
pn (x) = 1 |x|
(39)
1 x 2 Cn (x).
(3/2)(3 + n)
The orthogonality relation for Gegenbauer polynomials reads in our notation
1
dx pn (x, )cm (x, ) = (1)n nm .
(40)
1
7 The separation of valence quarks and sea quarks is done by measuring even and odd moments. For instance, in the
vector case: if the quark and anti-quark sea is equivalent, in even moments the complete sea drops out, while for odd
moments valence and sea quarks are added.
14
The minus sign in the argument of Gegenbauer polynomials in Eq. (39) is conventionally and
induces the factor (1)n in the orthogonality relation. This sign convention is appropriate to
perform the steps which follow. Note also that the support restriction is explicitly contained in
the definition (39) and so the integration region in the integral (40) is restricted to the central
region. We might now expand a GPD in terms of such polynomials (39)

q x, , 2 =
(1)n pn (x, )mn , 2 .
(41)
n=0
It is easily to see that the conformal moments (36) of a GPD are reproduced by this series (41).
However, it is divergent as expansion in terms of polynomials.8 Especially, the restricted support property of each individual term does not imply that the GPD vanishes in the outer region.
Rather one should understand this expansion as an ill-convergent sum of distributions (in the
mathematical sense) that yields a result which is non-zero in the outer region. Indeed, pn (x, )
can be considered as the nth derivative of a smeared function:
(5/2 + n)
pn (x, ) =
n!(1/2)(2 + n)
1
n+1 (n)

du 1 u2
(x u).
(42)
Taking now the forward limit lim0 pn (x, ) = (n) (x)/n!, the series (41) turns out to be the
expansion of parton densities in terms of derivatives of (x).
On the other hand, Eq. (41) might be a convergent series for > 1 and by means of the
crossing relation (28) we find for a GDA the partial wave decomposition:

z, , W 2 =
(1)n pn (1 2z, 1)Mn , W 2 ,
(43)
n=0
where the polynomials Mn (, W 2 ) = (12 )n mn (1/(12 ), W 2 ) are of order n. If (z, , W 2 )

is a smooth function that vanishes at the end-points z = {0, 1}, the series converges and the
moments Mn (, W 2 ) behave for n as 2n n1/2 with > 0. Here the exponential suppression by 2n is a consequence of our normalization, which has been adopted from the Mellin
moments of GPDs respectively parton densities.
The series (41) and (43) are the conformal partial wave expansions with respect to the conformal spin n + 2. They have the advantage that in perturbative QCD the conformal spin is, to
some extend, a good quantum number. So, for instance, the LO evolution kernels are diagonal
with respect to Gegenbauer polynomials. Furthermore, in this expansion the x and 2 (respectively, z and W 2 ) dependence factorizes. In intrinsic x/ and a remaining (, 2 ) dependence
contained in the conformal moments. The (, 2 ) or (, W 2 ) dependencies can be separated by
an expansion of the moments with respect to an appropriate set of orthogonal polynomials. It
has been proposed to expand Mn (, W 2 ) in terms of Legendre polynomials Pl (cos( )) [40], the
eigenfunctions of the rotation group SO(3) for spinless states. Here cos( ) 1 2 = 1/ and
is the scattering angle in the center-of-mass system. Certainly, it is appealing to have such an
expansion with respect to angular momentum l, which makes contact to the SO(3) partial wave
expansion. When spin is involved, the expansion is rather given in terms of Wigner functions,
8 This is also the case in the central region [, ], since the coefficients in front or the Gegenbauer polynomials are
enhanced by the factor j 1 , which divergences for || < 1 at j .
15
expressible in terms of associated Legendre polynomials. To be not specific on the spin content
and in view of the definition in terms of conformal operators, it appears also natural to expand
conformal moments with respect to Gegenbauer polynomials with index 3/2:
n

2
mn , =
Fnk 2 n ck (1/),
n

2
Mn , W =
Fnk W 2 ck (1 2 ),
k=0
(44)
k=0
where all 2 and W 2 dependence is absorbed into the form factors9 Fnk . Obviously, the crossing of GPDs and GDAs, i.e., the transfer from the space-like to the time-like region, concerns
now only these form factors. We remark that the conformal moments, i.e., Fnk , are related to
the SO(3) partial wave amplitudes that appear in the construction of the t-channel scattering amplitude. These partial wave amplitudes are labeled by angular momentum rather than conformal
spin. To obtain this relationship one can employ the partial wave decomposition of the t-channel
scattering amplitude, which is perturbatively predicted in terms of the conformal moments (44),
see also Ref. [40]. This is easily done for scalar particles, however, we expect that due to the spin
content the derivation in our case would be somewhat tedious. We would like to leave this task
for some later publication.
2.3.2. SommerfeldWatson transformation
As already mentioned, the series (41) cannot be directly used in practical calculation. Rather,
it must be either resummed or the individual terms must be smeared by inserting the identity,
expanded with respect to an appropriate basis. However, the latter method is only applicable in a
restricted kinematical region and has been performed only approximately. So we chose to resum
the conformal partial waves series for GPDs instead. We consider it first in the unphysical region
> 1 and rewrite q(x, , 2 ) as a contour integral in the complex plane that includes the positive
real axis, see Fig. 3:

1
q x, , 2 =
2i
()
dj

1
pj (x, )mj , 2 .
sin(j )
(45)
(0)
Here we included a factor 1/ sin(j ), which has the residue Resj =n 1/ sin(j ) = (1)n / for
n = 0, 1, 2, . . . . Thus, if no other singularities are present inside the integration contour, the
residue theorem leads to the conformal partial wave expansion (41). The main difficulty is to find
an appropriate analytic continuation10 of both functions pj (x, ) and mj (, 2 ) with respect to
the conformal spin n + 2.
First let us define the analytic continuation of pj (x, ). This can be done using its definition
(39) in terms of hypergeometric functions. To include also the support restriction, we represent
the analytic continuation of the Gegenbauer polynomials by the Schlfli integral
1
(5/2 + j )
pj (x, ) =
(1/2)(2 + j ) 2i
(+1)

du
(1+)
(u2 1)j +1
.
(x + u)j +1
(46)
9 We note that in this expansion the scaling invariance and so the conformal symmetry is broken, since in general the
form factors contain now a massive parameter for dimensional reason. This is not surprising, since in non-perturbative
QCD this symmetry does not holdotherwise there would exist only massless hadrons.
10 We remind that the analytic continuation of a function that depends on a discrete variable is not unique. For instance,
a term proportional to sin(j ) could be added, which drops out for j = n = 0, 1, 2, . . . .
16
(a)
(b)
Fig. 3. (a) The integration contour in Eq. (45) enclosing the real axis in the complex j -plane. Adding semicircles results
in an integration path which is parallel to the imaginary axis. (b) The contour of the Schlfli integral (46) in the complex
u-plane.
Here the integration contour is essentially the unit circle in the complex u-plane where the points
1 and +1 are included, see Fig. 3(b). The integrand has four branch points in the complex
u-plane, namely at {, 1, x/, 1}. These points will be connected by a single branch cut
that goes along the real axis from to Max(x/, 1). It is easy to see that for non-negative
integer j = n the Schlfli integral is equivalent to the definition (39). The integrand possesses
now only a pole of order n + 1 at u = x/ (and at infinity). For |x| < the pole is inside the
integration contour and the residue theorem gives pn (x, ). On the other hand for |x| > the
pole is moved out of the contour and so the integral vanishes.
Before we evaluate the integral (46) in general, let us consider the forward case = 0. Here
the integrand is essentially reduced to the function (u2 1)j +1 and possesses for non-integer j
a discontinuity on the real axis in the interval 1 u 1. We might now deform the contour so
that the real axis is pinched. For | e u| 1 we pick up a phase factor ei(j +1) for m u 0
and so we can write
pj (x, = 0) = x j 1

1 i(j +1)
(5/2 + j )
e
ei(j +1)
(1/2)(2 + j ) 2i
1
j +1

du 1 u2
.
(47)
The remaining integral represents just (1/2)(2 + j )/ (5/2 + j ), which results in

sin([j + 1]) j 1
(48)
x
.
This is, up to the conventional factor sin([1 + j ])/ , nothing else but the integral kernel of the
inverse Mellin transform, widely used in deep inelastic scattering.
For > 0 and non-integer values j the integrand in Eq. (46) has no discontinuity for x .
For x we will pick up phase factors by surrounding the branch points at x/ and x = 1.
They appear in the following interval above or below the real axis:

< x and > 0,
[x/, 1]
for
(49)
< x.
[1, 1]
pj (x, = 0) =
17
Consequently, at the endpoint x = and for x < the integral vanishes

pj (x , ) = 0.
(50)
For both the central and the non-vanishing outer region the complex integral can be evaluated by
deforming the contour as before so that the real axis is pinched. Taking the discontinuity yield
the following integrals along the real axis
(5/2 + j ) sin([j + 1])
pj (x, ) =
(1/2)(2 + j )
(5/2 + j ) sin([j + 1])

=
(1/2)(2 + j )
1
du
x/
1
du
1
(1 u2 )j +1
,
(x + u)j +1
(1 u2 )j +1
,
(x + u)j +1
x , > 0,
0 x.
(51)
At the cross over point x = both integrals have the same value and represent a Beta function.
So pj (x = , ) is smooth in the vicinity of this point and takes the value
pj (x = , ) =
21+j (5/2 + j ) sin([j + 1])

.
j +1 (3/2)(3 + j )
(52)
The integrals in Eq. (51) define hypergeometric functions. The analytic continuation of the
mathematical distributions pn (x, ) with respect to the conformal spin is expressed by them as
following:

j 1
x
x
j 1
+ (x )
Pj
Qj
pj (x, ) = |x|
(53)
where

2j +1 (5/2 + j )
j 1, j + 2 1 + x
Pj (x) =
(1 + x) 2 F1
,
2
2
(1/2)(1 + j )

sin(j ) j 1
(j + 1)/2, (j + 2)/2 1
Qj (x) =
x
2 F1
x2 .
5/2 + j
(54)
(55)
Here, a few comments are in order. First, for j = n = 0, 1, 2, . . . only the central region contributes and the relation

2
j, j + 3 1 + x
j 1, j + 2 1 + x
(56)
=
F
F
2 1
2 1
2
2
2
2
1x
establishes the definition (39) of pn in terms of Gegenbauer polynomials, cf. Eqs. (33)(35).
Obviously, in the central region the analytic continuation is based on the definition of hypergeometric functions. In the outer region, however, the result might be surprising. To clarify its
1
1
1
meaning, we decompose the integral (51) for < x as 1 du = x/ du x/ du
and realize that it can be expressed by the function Pj . Thus, we can write Eq. (53) as

j 1
x
x
j 1
+ ( x) cos [j + 1]
pj (x, ) = ( x)
(57)
Pj
Pj .
Here it is understood that the principal value is taken at the branch cut, which starts at x = ,
i.e., we insert [Pj (x + i) + Pj (x i)]/2 for x 1. The cos([j + 1]) term in the second
18
Fig. 4. The real (left) and imaginary (right) part of the conformal partial wave pj (x, = 1) (solid), its first (dashed), and
second (dotted) derivative for j = 1/4 + i/2.
expression on the r.h.s. arise from the continuation of to , again by taking the principal
value. Hence, one realizes that this result precisely fits the structure of the representation (12) for
q(x, , 2 ) in terms of the functions . We remark that the identity

1
1
Qj (x) = Pj (x + i) + Pj (x i) + cos [j + 1] Pj (x)
2
2
for x 1,
(58)
we derived here, is a known relation between associated Legendre functions of the first and
second kind.
The function pj (x, ) is continuous at the cross-over point x = , however, the imaginary part
of the first derivative has a jump, while the real part is still continuous, see Fig. 4. It satisfies the
second order differential equation for conformal partial waves with complex valued conformal
spin j + 2:

d2
1 x2
pj (x, = 1) = (j + 1)(j + 2)pj (x, = 1).
dx 2
(59)
The expressions for the conformal partial waves in Eqs. (53), (54), and (55), turn out to be
in agreement with a representation, given recently in Ref. [37], in terms of associated Legendre
functions for the gluonic GPDs. In fact, up to some normalization factors and shift in both the
conformal spin and the index, the same functions appear in the central and outer region. For
the former one we, however, prefer to work with complex conformal spin, too. This incorporates the underlying duality between central and outer region in a manifest manner and yields a
uniform representation of scattering amplitudes, see below Section 3.2. This is essential for the
perturbative QCD analysis at higher orders.
The analytic continuation of the polynomials mn (, 2 ) is denoted as mj (, 2 ). These functions will be also analytic in , however, might have branch points at = 0, = 1, and = .
It would be desirable to have an integral representation that makes this property transparent and
might allow the continuation from 1 to 1 or even to negative values. Moreover, we will
also require that the moments mj (, 2 ) are bounded at large j . It turned out that with these
requirements the analytic or numerical calculation of moments from a given GPD is a rather
intricate task. We have to admit that this mathematical problem is also not solved here for any
conceivable GPD. In the following we give, however, some recipes to evaluate the conformal
moments for complex conformal spin in the region || 1.
19
To derive an appropriate integral representation, we decompose here, in contrast to Eq. (38),

the conformal moments into contributions that arise from the outer and the central region:

2
out
2
mn , 2 = cen
(60)
n , + n , .
Again polynomiality is only manifest for the sum but not for the separate terms on the r.h.s. The
2
analytic continuation of out
n (, ) is defined in terms of hypergeometric functions (35)
1

2
out
j , =

dx cj (x, ) x, , 2 + x, , 2 .
(61)
From the asymptotics of the hypergeometric functions

x + x 2 2 j
cj (x, )
for j , arg(j ) /2,
2
(62)
we can estimate the large j -behavior of conformal moments

out
j
1

, 2
dx
x+
x 2 2
2
j

x, , 2 + x, , 2 .
(63)
For 0 < < 1 the integrand and thus the conformal moments are exponentially suppressed for
j with | arg(j )| < /2. In the limit = 0 we arrive at the parton densities and get the
power like suppression factor j p .
The contribution from the central region reads for integer values n
cen
n

, 2 =

dx cn (x, ) x, , 2 .
(64)
The analytic continuation of the conformal spin, as done for the central region above, would
yield terms proportional to sin(j ) that exponentially grow at large m j . The presence of such
terms can also be read off from the integral (A.3), in Appendix A, together with our final GPD
representation (77), given below. In the case that this integral can be evaluated in an analytic form
for integer n these terms drop out and the analytic continuation can be performed by means of
2
the substitution n j . We remark that the explicit knowledge of out
j (, ) allows to restore
the contribution from the central region by means of the polynomiality condition. Obviously,
within such a procedure one might miss some terms that satisfy the polynomiality condition, but
contribute only in the central region. Nevertheless, this procedure offers in principle a way to
restore the polynomiality of a GPD that is only known in the outer region, e.g., from an ansatz
within the overlap representation [52].
Another possibility for the analytic continuation of the conformal spin, appropriate for numerical calculations, arises from the definition of the conformal moments in the central region
by the following contour integral (see Fig. 5):

n+1
2
cen
n , =
2i
1
1+

dz dn (z) z, , 2 .
(65)
20
Fig. 5. The integration contour in Eq. (65).
Here the function dn (z) is defined in the complex plane by the integral
1
1
dn (z) =
(1 z2 )
1
dx
1
(1 x 2 )cn (x, 1)
.
xz
(66)
It has a branch cut on the real axis in the interval [1, 1] and in addition single poles at x =
{1, 1}. Thus, we excluded in the integration contour of the integral (65) the points x = {1, 1}.
Obviously, inserting Eq. (66) into Eq. (65) yield by means of the residue theorem and taking the
limit 0 to the original definition of the conformal moments (64). The function dn (z) can be
expressed through Legendre functions of the second kind and an appropriate continuation of the
conformal spin n is given in terms of a hypergeometric function

22j 2 (1 + j )(3 + j ) x j 1
(j + 1)/2, (j + 2)/2 1
dj (x) =
(67)
2 F1
x2 .
5/2 + j
(3/2 + j )(5/2 + j ) x 2 1
The function dj (x) behaves at large x as x j 3 and has a pole at x = 1. At large j it is bound
in the whole region 1 x by
x 1j
.
(68)
x2 1
Let us suppose that the function (x, ) is a regular function that has only a branch cut along the
real axis [, 1] and has a power like behavior at infinity. We can expand the upper and lower
part of the integration contour in Fig. 5 to semicircles encompassing the complete half-planes
such that the real axis is pinched in the intervals [, 1] and [1, ]. For a certain value of
e j , the contributions at infinity are negligible and so we pick up the discontinuity along the
negative real axis and the pole at x = 1. This result serves now for the definition of the analytic
continuation in j :
dj (x, ) 2j
cen
j

, 2 | = j +1
1
dx dj (x)

1
x i, , 2 + x + i, , 2
2i
j +1
(1/2)(1 + j )
+
, , 2 .
2
(3/2 + j )
(69)
Our choice of e j guarantees the convergence of the integral and, moreover, we assumed here
that (x = , ) vanishes. The choice of e j can be relaxed by introducing appropriate subtraction terms that improve the power behavior of (x, ) at x = . On the r.h.s. of Eq. (69)
21
the last term arises from the pole at x = 1 and the first term contains a phase factor = eij ,
that depends on the continuation procedure from positive to negative real valued x. To get rid of
this factor we define for even and odd n the analytic continuation as

cen
even,

j (, 2 | = 1)
cen
2
for
n
=
n ,
(70)
2
odd.
cen
j (, | = 1)
Note that the appearance of the phase factor is strongly connected to the analytic properties of
GPDs and is only absent if the GPD has no branch point at x = .
Finally, we have for the analytic continuation of the conformal moments

even
cen

mj (, 2 )
j (, 2 | = 1)
2
out
2
for
= j , +
mn ,
cen
(, 2 | = 1)
modd
(, 2 )
j
j
even,
n=
(71)
odd.
It is obvious that meven
and modd
lead only for even and odd integer values of j = n to polyj
j
nomials, which are even in . The difference of even and odd moments arises only from the
central region and is expressed by a function in which does not degenerate into a polynomial
for integer values of j = n.
2.3.3. GPDs represented as MellinBarnes integral
Now we like to change the integration contour in Eq. (45), in such a way that it becomes parallel to the imaginary axis in the j -plane, i.e., extends from c i to c + i. Here the constant
c < 0 is chosen such that all singularities contained in the conformal moments mj (, 2 ) are on
the left-hand side of the integration path. Let us assume for the moment that mj (, 2 ) degenerates for all non-negative integer values of j to polynomials. As displayed in Fig. 3(a), we add
two quarters of circles in the first and fourth quadrant so that the integration contour includes the
imaginary axis and is closed by the arc that includes the points c + i, , and c i.
It remains to show that within our definition of conformal moments the contribution from the
arc vanishes. Suppose that 1 and that the variable x is rescaled by , i.e., x = X. Hence,
we must study the behavior of pj (X, ) for j in the interval |X| 1. The asymptotic
expansion of the partial waves for large j with | arg(j )| /2 can be read off from the behavior
of hypergeometric functions [59] and is
j

2 j arccosh(Xi)
e
ie(j +3) arccosh(Xi) for |X| 1.
pj (X, )
(72)
For 1 X 1 the function arccosh(X i) has a monotonously increasing or decreasing

imaginary part that lies in the interval [0, ] and [, 0] for the +i and i prescription,
respectively. Thus, within both prescriptions

1
pj (X, )mj , 2
sin(j )
(73)
exponentially vanishes for 1 < X < 1 on the arc, specified above, as long as mj (, 2 ) behaves
for > 1 as
j

mj , 2
(74)
for j .
2
22
This remains true, if we replace pj (x, ) by the symmetric and anti-symmetric combinations
pjeven (x, ) =

1
pj (x, ) + pj (x, ) ,
2
pjodd (x, ) =

1
pj (x, ) pj (x, )
2
(75)
and mj
by the corresponding even and odd moments, respectively. Since pn (x, ) =
(1)n pn (x, ) has definite symmetry, the residues of
(, 2 )

1
pjeven (x, )meven
, 2 and
j
sin(j )

1
pjodd (x, )modd
, 2
j
sin(j )
(76)
only contribute for even and odd values of j = n, respectively, and, thus, the polynomiality is
implemented in a manifest manner.
Unfortunately, we did not give in the previous section a representation for mj (, 2 ) that
allows the analytic continuation in to the region > 1 and simultaneously satisfies the requirement that mj (, 2 ) fulfills the bound (74). This can be read off from Eq. (63), where obviously,
we will pick up an additional phase. It seems that such a representation cannot easily be found
rather we encounter here similar problems as for the continuation to the region < 0.
This difficulty can be avoided once we replace in mj (, 2 ) the variable by with 1.
It is sufficient to consider the cross-over point, since for all other x values inside the central
region we will have an additional exponential suppression due to the phases of pj (, ), see
Eq. (72). For X =1 we find from Eq. (52) that pj (, ) behaves on the cross-over point as
(2/)j sin(j )/ j . To get rid of the exponential growth, induced by the real part of j , we
might choose > 2 and can arrange in this way even an exponential suppression.
If now the
conformal moments mj ( , 2 ) for given do not grow faster than (/2)j / j for j ,
the integral on the infinite arc does not contribute for x = 1, too. Thus, after appropriate scaling
of x with , analytic continuation to the region 0 1, and setting = we find for all
values |x| the following MellinBarnes integral representation for GPDs:
c+i

i
q x, , 2 =
2
dj

1
pj (x, )mj , 2 .
sin(j )
(77)
ci
Finally, we extend this integral into the outer region. For pj (x, ) this is done by means of the
definition (53).
The MellinBarnes integral (77) can be used when the analytic continuation of even and odd
moments leads to the same function mj (, 2 ). If this is not the case, we separately introduce
the MellinBarnes integral for the even and odd part of

q x, , 2 = q even x, , 2 + q odd x, , 2 .
(78)
Since the representation (77) remains valid if we substitute pj (x, ) by the symmetrized partial
waves pjeven (x, ) and pjodd (x, ), we have the same form of the MellinBarnes integral as before:
q
even/odd
x, ,
i
=
2
c+i

dj

1
even/odd
even/odd
pj
, 2 .
(x, )mj
sin(j )
(79)
ci
Bearing in mind that pj (x, ) is set to zero for x < , the extension of pjeven (x, ) and
pjodd (x, ), defined in Eq. (75) in terms of pj (x, ) and pj (x, ), is consistently done by the
23
procedure (53). The original GPD (78) is restored by adding its even and odd parts:

i
q x, , 2 =
2
c+i

1 even
pj (x, )meven
, 2 + pjodd (x, )modd
, 2 .
j
j
sin(j )
dj
ci
(80)
The extension into the outer region is not based on analytic continuation and the reader might
wonder whether this prescription (53) is indeed correct. Besides the arguments we gave in Section 2.3.2, we provide next some further support. Let us first verify that the GPD representation
(80) has the correct support property, i.e., it contributes only for < x. As mentioned above,
the difference of odd and even moments arise only from the central region and has thus the
functional form (/2)j +1 fj (, 2 ), where fj (, 2 ) has a power-like behavior at j , see
Eq. (69). We expect that such a term does not contribute to the outer region. Indeed this can be
read off from the MellinBarnes integral, which takes the form
i
2
c+i

dj

2j 1
Qj (x/)fj , 2 ,
sin(j )
ci
see definitions (53) and (55). For x > the term 2j 1 Qj (x/)/ sin(j ) contains an exponential damping factor (/x)j for j with | arg(j )| < /2, while the remaining factor in the
integrand is bounded. So we can close the integration contour by an infinite arc that includes the
first and forth quadrant. Since the whole integrand is analytic in these two quadrants, Cauchy
theorem gives zero. Note that Qj (x/)/ sin(j ) contains no singularities on the real positive
axis. Consequently, even and odd contributions are the same in the outer region and will add for
x > , while they cancel for x < .
Alternatively, we can represent the GPD as

q x, , 2
i
=
2
c+i

dj

1
2
2
|x| pj (x, )m
,
j , + pj (x, )mj ,
sin(j )
(81)
ci
even modd )/2. Here the polynomiality is not mani
= (meven
+ modd
j
j )/2 and mj = (mj
j
fest in the conformal moments, rather it is separately restored for m

2n + m2n and m2n+1 m2n+1 .
Let us next show that the MellinBarnes integral vanishes for |x| > 1. Here the expression
j 1 Qj (x/)/ sin(j ) in the integrand vanishes on the infinite arc, due to the damping factor
(1/x)j , see Eqs. (53) and (55). As in the previous paragraph, we can close the integration contour
and find that the integral gives zero. If we would have allowed an additional term proportional
to Pj (x, ) in the extension of the support for x > , it would in general contribute in the
unphysical region x > 1, too.
Finally, the correctness of the GPD representations as MellinBarnes integral is deduced from
the fact that for non-negative integers the conformal moments are reproduced. Employing the
symmetry with respect to x x, we find from the representation (80), e.g., for even moments,
where m
j
dx cn (x, )q x, ,

=
i
dx cn (x, )
2
c+i

dj
ci

1
pj (x, )meven
, 2 , (82)
j
sin(j )
24
for n = 0, 2, . . . . From the moments separately calculated in the outer and central region, given
in Eqs. (A.1) and (A.3), respectively, it follows11

dx cn (x, )q x, ,
1
= lim
+0 2i
n+i

dj
ni

Nnj ()
Nnj ()
meven
, 2 ,
+
j
nj + j n+
(83)
for n = 0, 2, . . . . The integration path can be parameterized by j = n + i and making use of the
identity 1/( i) 1/( + i) = 2i(), the integral yields finally the conformal moment
2
odd
2
meven
n (, ) for n = 0, 2, . . . . Analogously, the conformal moments mn (, ) arise for odd
values of n.
We complete this section with the MellinBarnes representation for GDAs, which follows
now by crossing from Eq. (77) with the constrain z , which arises from 1/ x/ 1.
Employing the symmetry transformation z 1 z and 1 we get the remaining contribution, see Eq. (27):
i

z, , W 2 =
2

1
z1z
2
(z )pj (1 2z, 1)Mj , W +
,
dj
1
sin(j )
c+i

ci
(84)
(, W 2 )
where Mj
is the analytic continuation of the conformal GPD moments, obtained by
crossing, see below Eq. (43). If Mj (, W 2 ) is bounded along the integration path for 0 1,
the integral of the first term in the square brackets in this formula exist for all values of 0 z 1.
So we can drop both the restriction (z ) and the second term in the square brackets, obtained
by symmetry. This latter term follows now from analytic continuation and so we arrive at the
representation:
i

z, , W 2 =
2
c+i

dj

1
pj (1 2z, 1)Mj , W 2 .
sin(j )
(85)
ci
Suppose that Mj (, W 2 ) vanishes for j with | arg j | /2, it is straightforward to see

that the conformal moments for non-negative integer conformal spin are reproduced. Employing
Eq. (A.3) with = 1 and k = n = {0, 1, 2, . . .}, we can then close the contour so that it now
encircles the positive real axis. The residue theorem yields then Mn (, W 2 ).
3. Evolution kernels and coefficient functions in a manifest conformal scheme
To LO accuracy the evolution kernels and coefficient functions for exclusive processes respect conformal symmetry, which is the symmetry of the QCD Lagrangian at the classical level
for massless quarks. Conformal symmetry find, in fact, many practical applications in QCD. It
11 The x-integral over the outer region only exist for e j > n. Since here the integrand has no singularities on the
positive real axis, we can first shift the MellinBarnes integration path to the imaginary axis to the right. Then with e j =
n + the x-integration is performed. On the other hand the x-integration over the central region must be performed within
the original contour. It produces a sin(j ) term, which removes the poles on the real axis and so we can shift the Mellin
Barnes integration path along the positive real axis, too. Both remaining integrals can then be combined in the limit that
is indicated in Eq. (83).
25
allows, for instance, to solve the mixing problem of light-ray operators, caused by renormalization, and by means of the conformal operator product expansion it predicts the hard-scattering
amplitude. The breaking of conformal symmetry beyond LO has been studied with perturbative
methods in great detail [28,6064]. The main lesson from these studies is that the only physical
contributions violating conformal symmetry are generated by the non-zero function. All other
terms violating conformal symmetry (in the perturbative sector) are mainly artifacts introduced
by the standard renormalization/factorization prescription for the light-ray operators, which is
based on some version of minimal subtraction within the dimensional regularization scheme.
Within the standard scheme the conformal symmetry is separately broken in the evolution kernels and the hard-scattering amplitude for two-photon processes. However, it can be restored by
a finite renormalization providing a scheme in which conformal symmetry is manifest, except for
terms proportional to that are induced by the trace anomaly of the energy momentum tensor.
Thus, we will restrict us in the following two subsections not to LO accuracy. Rather, the results
can be (at least partly) applied to all orders of perturbation theory. Of course, terms proportional
to need special consideration. A first discussion of this issue can be found in Ref. [29], see also
Ref. [65].
3.1. Convolution of GPDs with conformal kernels
In this section we consider the convolution

K q(x, )

x y
dy
K
,
q y, , 2
||

(86)
of a GPD with a generic kernel K(x, y) that respects conformal symmetry. Here the integration
region is defined by the combined restrictions for the GPDs and the kernels. The requirement
of conformal symmetry means in fact that the kernel has for |x|, |y| 1 the following spectral
representation
K(x, y) =

(1)n pn (x)kn cn (y)
for |x|, |y| 1,
(87)
n=0
where kn are the eigenvalues of K(x, y) and the polynomials pn (x) and cn (y) are defined in
Eqs. (33) and (39), respectively. Suppose that even and odd eigenvalues have the same sign and
provide after analytic continuation with respect to the conformal spin the same holomorphic
function kj so that it satisfies a bound for all values of j with arg(j ) /2. In fact the eigenvalues kn coincide with the Mellin moments of the DGLAP kernel, given by rational functions and
harmonic sums, and so we know their analytic continuation. We remark that the support in the
(x, y)-plane is defined in Eq. (C.2) and the extension to the full region is unique, see Appendix C.
The convolution of a GPD confined to [, 1] with this kernel leads to a function that has also
the support x 1.
To proceed in the simplest possible manner let us extend the representation (87) to the whole
region by the series

1
x y
K
,
=
(1)n pn (x, )kn cn (y, ).

n=0
(88)
26
Again, here we understand that pn (x, ) are mathematical distributions with the restricted support x for integer values of n, while cn (y, ) are the polynomials (33), which can be
extended to the whole y region. Taking the MellinBarnes integral (77) and the spectral representation (88), the convolution (86) leads to a divergent series

(1)n pn (x, )kn mn , 2 .
K q(x, ) =
(89)
n=0
Here we employed the integrals (A.1) and (A.3) for discrete conformal spin, as has been explained at the end of Section 2.3.3. Since the analytic continuation of kn does not spoil our
assumptions for the SommerfeldWatson transform, we can proceed as in Section 2.3.3. As expected, for a conformal kernel the convolution
c+i

i
K q(x, ) =
2
dj

1
pj (x, )kj mj , 2
sin(j )
(90)
ci
is in the Mellin momentum space given by a multiplication of the conformal GPD moments with
the corresponding eigenvalues.
In those cases in which we must distinguish between even and odd eigenvalues of the kernel,
we write K as

x y
1
K
,
=
(1)n pneven (x, )kneven cn (y, ) + pnodd (x, )knodd cn (y, ) ,
(91)

n=0
pneven (x)
where
and pnodd (x) are defined as in Eq. (75), and employ the GPD representation (80).
Since even and odd Gegenbauer polynomials and partial waves have definite symmetry under
reflection, they cannot mix. Hence, the convolution leads to
c+i

i
K q(x, ) =
2
dj

1 even
p (x, )kjeven meven
, 2 + {even odd} .
j
sin(j ) j
(92)
ci
If the support of the GPD was in the interval [, 1] and the eigenvalues are different in the
even and odd sector, the convolution certainly gives us a function that lives now in the whole
region [1, 1]. The support extension is caused by the mixing with an anti-quark GPD. With
kjeven = kj + kj and kjodd = kj kj , we can decompose the convolution as
K q(x, ) = q(x, ) + q(x,
).
(93)
Based on the representation (81), we interpret the terms on the r.h.s. as contributions to a quark
GPD, with x 1,
i
q(x, ) =
2
c+i

dj
kj
sin(j )

2
2
|x| pj (x, )m
j , + pj (x, )mj ,
ci
(94)
and an anti-quark GPD, with 1 x ,

i
q(x,
) =
2
c+i

dj
ci
kj
sin(j )

2
2
|x| pj (x, )m
.
j , + pj (x, )mj ,
(95)
27
For the convolution of a conformal kernel with a GDA we would obtain the analogous results.
The convolution is conventionally defined as

K z, , W 2 =
e
1

dy K(1 2z, 1 2y) y, , W 2 .
(96)
Note that the change of variable (1 2y) y, induces a factor 2 in comparison to the definition (86). We can now represent the GDA by the convergent series (43), and the convolution
immediately leads to

e
kn
K z, , W 2 =
(1)n pn (1 2z, 1) Mn , W 2 .
2
(97)
n=0
Within the MellinBarnes representation the solution of the evolution equation is a trivial task
to LO accuracy, where the conformal symmetry is manifest in any scheme. The restoration of this
symmetry is well understood to NLO and even the terms proportional to can be diagonalized
with respect to conformal partial waves. Relying on this symmetry and borrowing the eigenvalues
of the evolution kernels from the Mellin moments of the DGLAP kernel from Refs. [66,67] one
can even proceed to NNLO.
The simplest example is given by the flavor non-singlet sector and LO accuracy. Here the
evolution equation reads

s () (0)
d
q x, , 2 , 2 =
q x, , 2 , 2 ,
d
2
(98)
where the evolution kernel is

1+x
2
1x
2
(0) (x, y) = (x, y)
(99)
,
1+
+ (x, y)
1+
1+y
y x
1y
x y +

with the +-prescription [K(x, y)]+ = K(x, y) (x y) dz K(z, y) and the shorthand notation

y x
1+x
.
(x, y) = sign(1 + y)
1+y
1+y
Its eigenvalues can be simply calculated for discrete conformal spin n. If one did this for complex
conformal spin j , one would encounter the same problems as for moments of GPDs. Namely,
terms proportional to sin(j ) would appear. The analytic continuation of the discrete eigenvalues
can, however, be defined without such terms:

2
(0)
,
j = CF 4(j + 2) 4(1) 3
(j + 1)(j + 2)
d
4
CF = , (z) =
(100)
ln (z).
3
dz
This is just the forward anomalous dimensions of twist-two operators, well known from deep
inelastic scattering. The evolution of the conformal moments is thus governed by the ordinary
differential equation

s () (0)
d
mj , 2 , 2 =
j mj , 2 , 2 ,
d
2
(101)
28
which is easily solved. Equating the renormalization scale with the resolution scale Q and
inserting the solution into the MellinBarnes integral (81) leads to

q x, , 2 , Q2
2

c+i
(0) Q

j d s ( )

i
1
=
(102)
pj (x, ) exp
dj
mj , 2 , Q20 ,
2
sin(j )
2
2
Q20
ci
where the moments mj (, 2 , Q20 ) belongs to a given GPD at the input scale Q0 .
3.2. MellinBarnes representation for amplitudes
We study next the convolution of a GPD with a given hard-scattering amplitude. To have a
concrete example at hand we deal here with the so-called Compton form factors that appear in
the perturbative description of DVCS:

F , 2 , Q2

2

s () (1) x
dx (0) x
Cp
+
Cp
,
+ O s Fp x, , 2 , 2 .
=
2
Q
p=u,d,s,g 1
(103)
For this process the skewness parameter is equal to the Bjorken like scaling variable, i.e., = .
The same kinematic constraint appears also in the hard exclusive electroproduction of mesons.
In fact, the result for the Compton factors, we will give below in Eq. (106), can be adopted for
this process, too. This is trivial to LO and requires some additional work beyond this order.
Let us consider the partonic Compton form factors to LO accuracy,
1

Fp , 2 , Q2 =
1
dx
Q2p
x i
Q2p
+ x i

Fp x, , 2 , Q2 ,
(104)
where Qp is the electric charge of the parton. Employing the decomposition (9) of GPDs into
quark and anti-quark parts, the partonic Compton form factor (104) reads

Fp , 2 , Q2
1
=
dx
Q2p
x i
Q2p
+ x i

qp x, , 2 , Q2 + qp x, , 2 , Q2 .
(105)
Next we employ the MellinBarnes representation (81) for GPDs and perform the momentum
fraction integration by means of Eq. (A.13). Then the Compton form factors are expressed in
terms of the conformal moments
Fp =
Q2p
c+i

dj
2i
ci

cos(j ) 1
j ] , 2 , Q2 ,
i
[mj + m
(3/2)(3 + j )
sin(j )
j 1 2
j +1 (5/2 + j )
(106)
29
where the sum of mj + m

j is given by the analytic continuation of odd and even conformal
moments for the vector and axial-vector case, respectively. As we realize by comparing with
, 2 ) as it must be. The real part

Eq. (77) and Eq. (52) the imaginary part of Fp is [q + q](,
of the amplitude, given by a principal value integral in the momentum fraction representation,
contains in the integrand an additional factor tan(j/2) and cot(j/2) for the vector and axialvector case, respectively.
In a conformal subtraction scheme the inclusion of perturbative corrections is straightforward.
Higher order corrections can be written as convolution of the LO hard-scattering amplitude (104)
with certain kernels, which are conformally covariant. Hence, in analogy to the discussion in Section 3.1, this yields in the MellinBarnes representation a multiplication with the corresponding
eigenvalues, which are known from deep inelastic scattering. For instance, to NLO accuracy the
partonic form factor Hp for quarks in the parity even sector of DVCS on a nucleon target reads
for = Q
Hp =
Q2p
c+i

dj j 1
2i
ci

j
s (Q) (1)
2j +1 (5/2 + j )
i + tan
1+
Cj
(3/2)(3 + j )
2
2

[hj + h j ] , 2 , Q2
(107)
with the NLO coefficients
5 2S1 (j )
3
9
(1)
S2 (j + 1)
Cj = CF S12 (1 + j ) + S1 (j + 2) +
2
2 2(j + 1)(j + 2)

1 (0)
+ j 2S1 (2j + 3) S1 (j + 2) S1 (j + 1) ,
2
(108)
where the analytic continuation of harmonic sums is defined by derivatives of the function
S1 (z) = (z + 1) (1),
S2 (z) =
2
d
(z + 1) +
,
dz
6
(z) =
d
ln (z).
dz
(109)
The first line in Eq. (108) is up to an overall normalization factor the well-known perturbative
correction to the Wilson coefficients of the structure function F1 in deep inelastic scattering
[68], while the addenda in the second line is induced by the non-forward kinematics. This result
is verified by a direct rotation from the minimal subtraction scheme to the conformal one and
coincides with the prediction of the conformal operator product expansion [28,62].
Another advantage of the MellinBarnes representation is that it might be useful for an analytic approximation of the Compton factors at smaller values of , lets say 102 , which
should lead to a rather good approximation of the scattering amplitude for the kinematics in Collider experiments. Such an approximation has been already studied within the DD formalism,
see [69] and references therein, however, it remained restricted to the perturbative LO approximation and only the term containing the leading power in could be extracted. The main idea
here is to shift the integration path in Eq. (107) to the left, so that one picks up the leading order
contribution for 0. Suppose that for the vector case the first singularity on the l.h.s. is a pole
at j = 0 with 1 < 0 < c < 1, which might depend on 2 . We recall that in this case only
the analytic continuation of odd moments enters and, thus, c can also be chosen to be positive,
however, it must be smaller than one. The integration path can then be shifted further to the left
such that 1 < c < 0 and all other singularities remain to the left of the new integration path,
30
while the leading pole contribution (j = 0 ) is explicitly taken into account. To LO accuracy the
partonic Compton form factors thus read

0
20 +1 (5/2 + 0 )
i + tan
Res[hj + h j ] , 2 , Q2
Hp = Q2p 0 1
(3/2)(3 + 0 )
2
j =0
+
c+i
Q2p
2i
dj j 1
c i

j
2j +1 (5/2 + j )
i + tan
[hj + h j ] , 2 , Q2 .
(3/2)(3 + j )
2
(110)
This result could even be improved by further shifts of the integration contour. A more systematic
expansion, requires that also the conformal moments mj (, 2 ) are expanded in in the vicinity
of = 0. We remind that is related to the Bjorken variable by xB /(2 xB ) and is for
present fixed target experiments certainly not larger than 0.4. Hence, already the inclusion of
the O( 2 ) corrections might be sufficient to obtain a good approximation of the Compton form
factors in this kinematics.
Unfortunately, the method for a systematic approximation, pointed out above, must maybe
be refined due to the following complication. Namely, we know that in general the conformal
moments will have a branch point at = 0 and so a Taylor expansion around this point is not
possible, after the analytic continuation is performed. Even worse, it turns out that the appropriate analytic continuation, which guarantees the correctness of the MellinBarnes integral, can
induce terms that lead to a behavior j or even 2j . Certainly, the latter term requires that
we shift the integration path to the right, while the former require a consideration of all poles.
Corresponding to its behavior at j , we might close the integration path so that the positive
or negative axis is included. For specific conformal moments it is probably still possible to obtain a systematic expansion of the Compton form factors in powers of up to a certain order. In
general, however, this seems to be a serious problem.
One might naively expect that this issue can be resolved, if one goes back to the definition of
conformal moments (36); expand first cn (x, ) in 2 and take then moments with respect to x,
which in turn can be simply continued to complex valued j . Such an expansion looks like, see
Eq. (A.12),

2
4
cj (x, ) = x cj 0 + cj 2 2 + cj 4 4 + .
x
x
j
(111)
The coefficients cj m vanish for integer value j = n with m > n and so this expansion degenerates
then into a polynomial12 [70], in of degree n. Hence we can shift from the beginning for each
individual term the integration path in the MellinBarnes integral to the r.h.s., i.e., c c + m.
However, one easily realizes that this procedure will also introduce new poles in the complex j
plane that are also shifted to the right of the real axis, remaining, however, to the left of the new
integration path. Finally, it turns out that these poles will remove the power in 2 we gained by the
expansion (111). With one word, taking the limit 0 in general conformal moments, leads to

12 Obviously, for non-negative integer n we are dealing with integrals of the type 1 dx x nm q(x, , 2 ), well defined
1
for n m. To avoid divergencies for complex valued j with e j m 1, the integral should be defined as a contour
integral in the complex x plane so that it exist and can be viewed as analytic continuation with respect to the variable
n m.
31
the correct leading power behavior of the Compton form factors in , however, the normalization
might be wrong.
Let us finally comment on the representation for the scattering amplitude in the case of the
production of a hadron pair by photon fusion. For the important phenomenological situation that
one photon is on-shell, it is of course related to the DVCS amplitude by crossing. By means of
the crossing relation (28) the amplitude follows to LO accuracy from Eq. (104):

Fp , W 2 , Q2 = (1 2 )Q2p
1
dz
0

1
1
p z, , W 2 , Q2 .
z 1z
(112)
Employing the conformal partial wave decomposition (43) for GDAs, we immediately find the
following series for the scattering amplitude
n+1

2 (5/2 + n)
1 (1)n Mn , W 2 .
Fp , W 2 , Q2 = (1 2 )Q2p
(3/2)(3 + n)
(113)
n=0
As discussed in Section 2.2 the conformal moments Mn (, W 2 ) are related to the GPD ones by
crossing, cf. Eq. (43). Of course, the inclusion of perturbative corrections and evolution effects
is done in an analogous way as in the case of DVCS. The only difference is that we are now
dealing with integer n. The sum (113) converges, just as the partial wave decomposition for the
GDA itself. However, the oscillations caused by the polynomials pn (1 2z, 1) drop out and
so its numerical approximation is easier to handle. This representation can be directly used in
phenomenological studies. Knowing an appropriate analytic continuation of Mn (, W 2 ) would
also allow to rewrite the conformal partial wave expansion (113) as MellinBarnes integral:

Fp , W 2 , Q2
= (1 2 )
Q2p
c+i

dj
2i
ci
2j +1 (5/2 + j ) Mj (1 , W 2 ) Mj (, W 2 )
.
(3/2)(3 + j )
sin(j )
(114)
Here we employed the symmetry relation Mn (, W 2 ) = (1)n Mn (1 , W 2 ) and that Mj is

holomorphic in the first and forth quadrant.
4. Evaluation and parameterization of conformal moments
This section is devoted to the parameterization of conformal moments and the numerical treatment of GPDs and Compton form factors. We study first the analytic properties of conformal
moments for a simple toy GPD ansatz and give then examples for the numerical evaluation. Then
we introduce a simple parameterization of the conformal moments with respect to the skewness
dependence, i.e., we consider only a reduced GPD and its crossing analog, and discuss its dependence on the momentum transfer squared. Finally, we suggest a simple GPD model, which is
rather flexible in its parameterization.
4.1. Numerical treatment of the MellinBarnes integral
Let us consider the conformal moments in terms of the Radyushkin ansatz (17) for the reduced
GPDs. This toy example can be treated for integer values of b and in an analytic manner. To
32
represent the results here in an explicit form we choose again b = 0 and our toy model with
= 1. In this case the function (x, ) is given in terms of Eq. (12) by

2 + x 1+
2 + x + 1+
+ ( x 1)
. (115)
q(x, ) = ( x 1)
2
1+
2 1
The analytic continuation of the function n (), see Eq. (38), reads

(1/2)(3 + j )
j, 3 + j, 2 + 1 +
.
(1 )1+j 3 F2
j () = 3+j
2
2, 3 +
2 (3/2 + j )
(116)
This function has the behavior needed for the derivation of the MellinBarnes integral. However,
it is not appropriate for an analytic continuation to negative , since the argument of the hypergeometric function would then become larger than one, i.e., (1 + )/2 (1 + )/2, and so
its value is given at the branch cut of the hypergeometric function. To analyse the situation in
more detail, we can first express the hypergeometric function in Eq. (116) as a sum of two 3 F2
functions with argument 2/(1 + ). This gives

(2 + )(1 )1+j
j 1, j, 2 j 2
j () =
F
3 2
2j 2, j 1 1 +
2(2 + j + )
(2 + )2(1+j )
(1 + j )(3 + j ) tan(j )
22(2+j ) (1 )2+j (1 + j )(3/2 + j )(5/2 + j )

2 + j, 3 + j, 1 + j 2
3 F2
4 + 2j, 2 + j 1 +
(1/2)(1 + j ) (3 + )(1 + j ) sin(j )

1+j +
.
1+j
1+
(3/2 + j ) ()(3 + j + ) sin([j + ])
2
(1 )
(117)
The first term on the r.h.s. contains poles on the real axis at j = 1/2, 3/2, . . . and for < 0 at
j = , 1 , . . . . Both of them are canceled by the second and third term, respectively.
The continuation of this function to negative values of is obviously not unique, since we
have branch points at = 0, which leads to phase factors ei(j +1+) as well as ei2(j +1) .
Such phase factors in j () would, however, violate the assumptions made for the derivation
of the MellinBarnes representation. To ensure that this does not happen we define the analytic
continuation by
AC
j () =
eij j (ei ) eij j (ei )

,
2i sin(j )
(118)
were the phase factors eij compensate the phases which arise from the j terms in the definition (35) of conformal moments, while taking the discontinuity and dividing it by 2i sin(j )
ensures that no new singularities appear on the real axis and that j () is bound for j . In
fact the properties of Legendre functions imply that the prescription (118) provides the contribution from the outer region
1
AC
j () =
dx cj (x, )

2 + x + 1+
2
1+

2 1+ j +1 (1/2)(1 + j )
2+
,
+
2
1+
2
(3/2 + j )
(119)
33
where the second term on the r.h.s. arises from a pole at x = and cancels the contribution
from the lower bound of the integral. Hence, we realize that the central region is still ignored in
Eq. (119) and its contribution might be taken into account within Eq. (69). The term that appears
there from the pole at x = is already contained in Eq. (119) and so only the discontinuity on
the negative axis contributes.
Analytic continuation via Eq. (118) leads for our toy ansatz to
AC
j () =

(2 + )(1 + )1+j
j 1, j, 2 j 2
F
3 2
2j 2, j 1 1 +
2(2 + j + )
(2 + )2(1+j )
(1 + j )(3 + j ) tan(j )
2(2+j
)
2+j
(1 + j )(3/2 + j )(5/2 + j )
2
(1 + )

2 + j, 3 + j, 1 + j 2
3 F2
4 + 2j, 2 + j 1 +
1+j +
21+j (1 + )1+
(1/2)(1 + j ) (3 + )(1 + j ) sin()

.
(3/2 + j ) ()(3 + j + ) sin([j + ])
(120)
As before for j () there are no net singularities in the first and forth quadrant of the complex j plane, however, individual terms possesses poles on the real axis. The first term on the
r.h.s. is needed to restore the polynomiality in the sum n () + AC
n (). However, the third
term violates polynomiality and must be canceled by the contribution from the central region,
still missing. We can restore this missing term from the polynomiality condition in such a
way that it is free of singularities in the first and forth quadrant of the complex j plane. The
sin()/ sin([j + ]) term of the fourth line, however, generates for j = n = {0, 1, 2, . . .},
a sign alternating series and to get rid of it, we separately continue even and odd moments:
mj (| ) = j () + AC
j ()
+
(1/2)(1 + j ) (3 + )(1 + j )

1+j +
,
1+j
1+
(3/2 + j ) ()(3 + j + ))
2
(1 + )
(121)
with = 1 and = 1 in the even and odd sector, respectively. We remark that this result for
the conformal moments of our toy GPD (115) can be alternatively obtained within the framework
given in Section 2.3.2. Especially, the term we restored from the polynomiality condition arise
from the discontinuity on the negative real axis in Eq. (69).
Now let us come to the numerical treatment of the MellinBarnes integrals for GPDs (77),
(80) and Compton form factors (106). The numerical evaluation can be easily done once the
conformal moments are known in analytic or numerical form. Corresponding to the behavior of
GPDs at x 1, they should vanish at large j rather fast [21]. Hence, the integral converges
rather fast, too, and so one can in practice perform the integration over a finite interval. If one
includes higher order corrections or the evolution the numerical treatment remains stable.
For our toy model we display in Fig. 6(a) for = 1/2 and = 0.25 the GPDs arising
from odd and even conformal moments, while the sum of them (solid line) provides the original
support. The evolution with respect to the renormalization/factorization scale 2 = Q2 is taken
into account simply by including the evolution operator in the MellinBarnes integral. For the
vector case the even quark moments evolve separately, i.e., they do not mix with the gluonic
34
Fig. 6. In panel (a) the toy GPD (115) is displayed for = 1/2 and = 0.25 as solid line and its symmetric and
antisymmetric part as dashed and dotted line, respectively. Panel (b) shows the evolution of the symmetric part from the
input scale Q20 = 0.5 GeV2 (solid) to the scales Q2 = 1 GeV2 (dash-dotted), Q2 = 10 GeV2 (dashed), Q2 = 100 GeV2
(dotted), and (nearly) asymptotic limit (thin dotted).
ones, and so we can employ Eq. (102). For s (Q) we take its LO approximation for three quark
flavors and = 0.22 GeV. That the numerics is completely unproblematic is demonstrated in
Fig. 6(b), where the asymptotic case Q2 is nearly reached by setting Q2 = 101000 GeV2 .
We remind that in this limit the outer region will die out and only the lowest conformal moment
contributes leading to the asymptotic GPD

3 2 x 2
.
q asy (x, ) = |x|
4|| 2
(122)
The numerical procedure for the calculation of scattering amplitudes is even easier to handle
than that for GPDs themselves. As mentioned, a nice feature of the MellinBarnes integral is that
it can be used to derive an expansion in powers of . Let us suppose that we like to evaluate the
Compton form factor (107) to LO accuracy, where the charge Qp is set to one. Here only the
analytic continuation of the odd conformal moments is needed. Guided by the small x behavior
of parton densities, we choose the parameter to be negative and larger than 3/2. To ensure
that all singularities are on the l.h.s. of the integration path, we take Max(1/2, 1 ) <
c < Min(1/2, ). The first pole which appear in the integrand on the negative axis arises from
tan(j/2) and is at j = 1. We remark that the pole which we would have expected in the
forward case, namely, at j = 1 is absent for > 0. Rather there appears a new one at
j = 1 + , which is associated with the behavior of the GPD at the cross-over point x = .
From our explicit expression for the conformal moments, given in Eqs. (117), (120), and (121) it
is obvious that the integrand contains three different pieces, proportional to 1j , 1+j , and .
For the term proportional to 1j we can arrange a systematic expansion in by a shift of the
integration path to the left. The poles appear here at j = {1 , 2 , . . .}, j = {1, 2, . . .},
and at j = {1/2, 3/2, . . .}. In the latter case the contribution will be canceled by those from
the piece proportional to 1+j , which has poles on the positive axis only, at j = {1/2, 3/2, . . .}.
This is established by a shift of the integration path to the right. What remains are the terms
proportional to . Here we can close the integration path so that the first and forth quadrant is
included and employ then the residue theorem, where the only poles are at j = {, 1 , . . .}.
All of these poles contribute to the leading power behavior and they must be resummed. Finally,
collecting the results and neglecting power corrections of order O( 3 ) leads to the approximation
35
Fig. 7. The Compton form factor H(, 2 = 0, Q2 ), arising from the toy GPD (115), to LO accuracy. Left: exact
imaginary (solid) and real (dashed) part of the Compton form factor and their approximation by Eq. (123) (dash-dotted
and dotted lines, respectively). Right: the Compton form factor evaluated numerically at the input scale Q20 = 0.5 GeV2
(solid and dashed) and evolved with the flavor non-singlet evolution equation to Q2 = 4 GeV2 (dash-dotted and dotted).
The imaginary part is shown as solid and dash-dotted lines and the absolute value of the real part as dashed and dotted
lines.
H( ) = 2 (2 + ) i + cot
2

(1 + )(2 + ) 2
+
+ O 3
1 i(1 + ) tan
2
2

2

(2 + + ) 2
2(2 + )
1+
+ O 3 .
(1 )(2 )
(123)
This expansion coincides with that for the exact Compton form factor, which for our toy model
(115) is exactly calculable. It is remarkable that this expansion does contain odd powers of . In
fact, the term in the first line should be especially important in the small region. Numerically,
this approximation works quite well for the region that is of phenomenological interest. The
deviation from the exact expression is about 3% for = 1/2 and = 0.4 for the imaginary
part. For the real part the approximation induces a small shift of the zero of the exact expression
from 0.456 to 0.431. The accuracy of the approximation (123) grows with increasing
and rapidly with decreasing . It is amazing that this approximation remains qualitatively correct
as long as does not approaches one, see left panel in Fig. 7. In the right panel of this figure
we show the imaginary and real part of the Compton form factor, evaluated numerically with the
MellinBarnes integral, for smaller values of . Here the difference between the approximate and
exact result is even invisible. Let us stress specifically, that we do not encounter any numerical
problems in the small region.
4.2. Anstze for conformal moments
The conformal moments for complex valued conformal spin, appearing in the GPD representation (77), can be expanded in terms of ck (1, ) using a suitable integral transformation. Such
an integral transformation can be viewed as the analytic continuation of the conformal expansion
(44) for non-negative integer conformal spin. In such an representation the and 2 dependence
in the conformal moments separates and only an ansatz for the form factors Fj k (2 ) is required.
Lets have a closer look at these form factors. The GPD MellinBarnes integral representation
(77) can be interpreted as describing the effective summation of all particle exchanges in the
t-channel. These are labelled by their conformal spin. Of course, conformal symmetry is broken
36
in the non-perturbative sector. However, the conformal spin can still be used for the classification
of excitations, just like in quantum mechanics for a non-spherical potential, the partial wave expansion in terms of spherical harmonics can still be employed to solve the Schrdinger equation.
For vanishing skewness, i.e., = 0, the quantum number of the conformal spin can be replaced
by the common spin J = j + 1 and the conformal moments are given by the diagonal form
factor Fjj (2 )

mj = 0, 2 = Fjj 2 ,
(124)
cf. Eq. (44). Having in mind that the conformal partial waves pj (x, = 0) are then given by
(1/x)J , where 1/x plays the role of a (rescaled) energy, the MellinBarnes integral for GPDs
looks similar to the t-channel scattering amplitude at large energies. This suggest that there exists
a connection with Regge theory or at least with Regge phenomenology. Let us remind that in deep
inelastic scattering, i.e., for 2 = 0, this connection shows up in the small x behavior of structure
functions, which is governed by the intercept of the corresponding leading Regge trajectories
[71]. Note, however, that the small x behavior of parton densities depends on their conventions,
i.e., on the factorization scheme and scale. A more recent analysis for the unpolarized valence
quark densities can be found in Ref. [72]. Below it will be demonstrated that for = 0 there exist
phenomenological indications that the 2 dependence in Fjj (2 ) is related to Regge trajectories.
Also for general kinematics, i.e., = 0, one can take this duality between t- and s-channel
serious, see the discussion in Section 2.1. A description of DVCS in the high-energy limit is given
in [73], where the leading Regge trajectory arises from the pomeron exchange in the t-channel.
Moreover, the t-channel description provides arguments for the so-called D-term [56], appearing
in the nucleon GPDs H (x, , ) and E(x, , ), and for the so-called pion pole term [74],
appearing in E(x,
, ). Note that these effects are taken into account by a modification of GPDs
that affects only the central region and so their s-channel counterpart is absent. On the basis of
the conformal partial wave expansion, applied to LO accuracy, and the crossing relation between
GPDs and GDAs a dual description of the former ones in terms of t-channel exchanges has been
suggested in Ref. [40]. Here the conformal moments have been decomposed into contributions
with definite angular momentum. Since the concept of conformal spin has to the best of our
knowledge not been worked out for applications in hadron spectroscopy, one should focus on this
more appropriate quantum number. Nevertheless, let us point out that the form factors Fj k (2 )
for j = k, defined in Eq. (44), measure the strength of (non-perturbative) conformal symmetry
breaking. In this paper, however, we do not proceed with the spectroscopic interpretation of these
form factors (or some rotated version of them) and the search of an appropriate representation
for the conformal moments with complex valued conformal spin in terms of them.
Instead, our aim is now more pragmatic, namely we will introduce and study specific anstze
for the conformal moments. Therefore, we now introduce and explore anstze that are simpler
to handle. We start in Section 4.2.1 with a reduced variable dependence by setting 2 = 0. Then
we implement in Section 4.2.2 the 2 dependence. In Section 4.2.3 we especially consider this
dependence for the case = 0 and discuss the resulting GPD H (x, = 0, 2 ), especially, its
interpretation as three-dimensional parton density.
4.2.1. Anstze for reduced conformal moments
For our simple toy GPD example (115), studied in Section 4.1, the conformal moments appear
to be rather complicated functions. Indeed, it is not clear at all whether the complicated structure of mj (, 2 = 0) in our toy model, is an artifact of the analytic continuation procedure or
37
indicates some physics, related to the skewness dependence. So let us explore several minimal
anstze for the conformal moments.
To ensure that the forward limit is correctly reproduced, we factorize the poles in the complex
j -plane that survive the limit 0:
( + 1 + j )( + + 2) 2

mj , 2 = 0|, =
bj .
( + 1)( + + 2 + j )
(125)
Here the normalization of the function is bj (2 = 0) = 1 and, moreover, we considered it as

convenient to normalize the lowest moment according to m0 (, 2 = 0|, ) = 1. In the forward
limit we arrive after an inverse Mellin transform at

mj = 0, 2 = 0|,
( + + 2)
x (1 x) ,
( + 1)( + 1)
(126)
which is well defined for all realistic values of , except for = 1. For this special value one
can first multiply the expression with an appropriate normalization factor that cancels
the factor
1/ ( + 1). The standard parameterization of a parton density, i.e., N x (1 + A x + Bx)
(1 x) , can be obtained in Mellin space by the linear combination

qj = N mj (0, 0|, ) + A mj (0, 0| + 1/2, ) + B mj (0, 0| + 1, ) ,
(127)
where
(1 + )(1 + )
,
(2 + + )
1+
B.
B =
2++
N = N
A =
(3/2 + )(2 + + )
A,
(1 + )(5/2 + + )
(128)
It is required that the function bj (2 ) for non-negative integer values of j = n reduces to a

polynomial of order n/2 and (n 1)/2 for even and odd n, respectively. Here the order of the
odd moments depends on the specific GPD. Moreover, these functions have only singularities
in the second and third quadrant of the complex j plane. We can allow that these functions
grow with the real part of j , however, they must be bounded for large imaginary parts of j ,
| arg(j )| /2.
One can classify the conformal moments bn (2 ) in general and their analytic continuation
with respect to their dependence on both variables and j , which leads to a classification of reduced GPDs. (We do not know whether this mathematical fact is known already from some other
context.) For the time being, we concentrate on functions bj (2 ) that can be expanded around
the point 2 = 0. Several simple examples can be given in terms of hypergeometric functions

)/4,a
2 F1 j/2+(12p)(1
2
r + s2
b

bj {a, b}, {r, s}, {, p} =
(129)
j/2+(12p)(1 )/4,a ,
r
2 F1
b
where the normalization condition at = 0 is satisfied. Here the parameters a and b may depend
on j . For the analytic continuation of even moments we set as above = 1 in the case of odd
ones = 1 and the choice p = 1 leads for j = n to polynomials of order (2 )(n+1)/2 , while
p = 0 gives polynomials of order (2 )(n1)/2 . To ensure that no branch cut appears in the interval
0 2 1, the parameters r and s must fulfill the inequalities r 1 and r + s 1.
38
The definition (129) is quite general and contains a number of special cases. For instance, if
we set = 1, a = (1 j )/2, b = 2, and r = 1,

(3/2)(3 + j )
2
j/2, (1 j )/2
2

bj (1 j )/2, 2 , {1, s}, {1, 0} = j +1
2 F1
1 + s ,
2
2 (3/2 + j )
(130)
we recover for s = 1 the definition of conformal moments (35) with x = 1. Here they are
expressed in terms of hypergeometric functions that arise from the original ones by a so-called
quadratic transformation. These moments can be used for even and odd values of j . If one needs
odd moments that are of order (2 )(n+1)/2 , one should set a = (1 j )/2. We remark that the
expansion in the vicinity of 2 = 0 exists only as linear combination of two power series in 2 ,
one of them containing the overall factor (2 )3/2+j .
Besides the other parameters, s controls the strength of the skewness dependence. If we set it
to zero, the function bj (2 ) is simply one.
If a = b, bj (2 ) reduces to the simple function

(j +p)/2

s
2
.
bj 2 {a, a}, {r, s}, p = 1
(131)
1r
Here 1 r appears as a scaling factor and so in the following r can be set to zero, when simultaneously s is restricted to s 1. For negative values of s the function (129) is even analytic for
1 . To get a clue which values of a(j ) and b(j ) are allowed in the ansatz (130), we generate
associated conformal moments of bj (2 |{a, a}, {0, s}, p) by the convolution integral
1
(j +p)/2

dz fj (z) 1 zs2
,
1
dz fj (z) = 1.
with
(132)
Here it is required that fj (z) as function of j is bound for j with | arg(j )| /2 for
0 z 1. To arrive at the parameterization in terms of hypergeometric functions we choose
fj (z) =
(b(j ))
za(j )1 (1 z)b(j )a(j )1 .
(a(j ))(b(j ) a(j ))
(133)
The requirement for the bound of fj (z) is certainly satisfied for e a(j ) > 0 and e(b(j )
a(j )) > 0. Especially, if a(j ) tends to infinity for j , b(j ) has to grow as a(j ) or even
faster. The case a = (1 j )/2 and b = 2, mentioned above, can be obtained within an analogous
treatment. Here, however, one has to choose an integration contour in Eq. (132) in the complex
plane that encircles the point z = 0 and so the convergence condition for the integral can be
relaxed. We will skip this issue here and refer, for instance, to Ref. [70].
We now explore the resulting GPDs and GDAs. As example we take the reduced valence
quark GPDs and GDAs in the vector case with the rather realistic values = 1/2 and = 3.
The unpolarized parton density is normalized to one and reads, cf. Eqs. (125) and (126),
q(x) =
35 1/2
x
(1 x)3 .
32
(134)
The function bj (2 ) is the analytic continuation of conformal moments with even n and the
parameters are specified in Table 1. A few comments are in order. In case (a) we took the conformal moments itself, here in a more appropriate representation that is for > 0 equivalent to
39
Table 1
The parameters and resulting functions for our anstze concerning the analytic continuation of even conformal moments
(129), classified by case (a)(e)
Cases
(a)
Parameters of bj (2 |{a, b}, {r, s}, {1, 0})

{a, b}
{r, s}
{(1 j )/2, 2}
{1, 1}
(b)
{a, b}
{1, 0}
(c)
{199/4 + j, 2/3 + 2j }
{0, 1/4}
(d)
{199/4 + j, 2/3 + 2j }
{0, 1/4}
(e)
{a, a}
{0, 1/4}
Explicit expression

)/2
(3/2)(3+j )
1 2
F j/2,(1j
2
2j +1 (3/2+j ) 2 1
1
2
| 4
j/2,199/4+j 2
|4
2 F1
2/3+2j

2 j/2
1 4
2 F1
j/2,199/4+j
2/3+2j
cj (1, ). They possess the properties we required in the derivation of the MellinBarnes integral and of course, they reduce to polynomials for both even and odd non-integer values of n.
We will use them in the MellinBarnes integral (79) within even conformal partial waves (75).
Crossing symmetry, see Eq. (43), requires the existence of an analytic continuation to > 1 and
this is achieved here by the change of arguments cj (1, ) cj (1 2, 1), see Eq. (44). For
the numerical evaluation of the GDAs the conformal partial wave series (43) or alternatively the
MellinBarnes integral (84) can be used. The projection on the even moments can be achieved
by symmetrization:
(z, )

1
(z, ) + (1 z, ) .
2
(135)
The case (b) seems to be trivial. But after crossing it results in (1 2 )j , which leads for
odd moments with > 1/2 to an alternating series with an absolute value that converges to
one in the end-points. For the convergence of the conformal partial wave series (44) in terms of
polynomials an exponential suppression factor 2n is required. Hence, this series only converges
in the interval 1/4 < < 3/4. The analytic continuation with respect to can be achieved by
the MellinBarnes integral (85). However, outside of the convergence region, i.e., < 1/4 or
3/4 < , we will find a GDA that does not vanish at the end-points z = {0, 1}.
The cases (c) and (d) differ by the sign of the argument in the hypergeometric function and
follows from (e) by an integral transformation (132). They reduce to polynomials for even n
only. The factor 1/4 in the argument improves the convergence property after crossing. The
second parameter in the upper line of the hypergeometric function, i.e., 199/4 + j , can for
certain non-integer values of j = n < 49 be a negative integer. As explained above, this does not
generate problems. The choice of this big constant 199/4 induces a numerical enhancement
of O(2 ) terms. The parameter 3/2 + 2j in the lower line compensates for rather large values
of j the growing of the second argument in the first line. We did not include any poles on the
positive j axis. As a consequence, we have new poles on the negative one. They appear at j =
1/3, 4/3, . . . and die out in the limit 0. With our choice = 1/2, which determines
the small x behavior of the parton densities and is associated with a pole at j = 1/2, a new
leading pole at j = 1/3 arises for non-zero skewness. Its contribution, however, should be
suppressed as O(2 ). So for instance, in the Compton form factors it should only give rise to a
2 2/3 = 4/3 term. Moreover, we changed the normalization of the other poles for 0 < . For
not too large values of this should produce only a numerically small effect. Since we included
40
Fig. 8. The (left) and (right) dependence of bn (2 ) and (1 2 )2 bn (1/(1 2 )2 ), respectively, for n = 2 are shown
for the parameterizations given in Table 1: (a) solid, (b) dash-dotted, (c) dashed, and (d) dotted lines.
Fig. 9. The momentum fraction shape of reduced GPDs with = 0.5 (left) and GDAs with = 0.4 (right), same labeling
as in Fig. 8.
a numerical enhancement, this might induce some sizeable changes of the GPD and GDA shapes
for not too small values of .
In Fig. 8 we display the and dependence for the second moment, i.e., of b2 (2 ). The
dependence of the conformal moment in case (a) [(solid line)], i.e.,
c2 (, 1) = 1
2
1
c2 (1 2, 1) = (1 2 )2 ,
5
5
(136)
is rater weak and so it is similar as in case (b) [constant or (1 2 )2 (dash-dotted line)]. In

contrast we find for cases (c) [dashed line] and (d) [dotted line] a large deviation in opposite directions. For GPD moments it will die out for 0, while for the GDA moments the difference
is nearly independent and large. We will skip here the detailed discussion of the dependence
in higher moments, which are suppressed by the factor B(1/2 + n, 4 + n) 1/n4 for n .
Note, however, that only the conformal moments (a) for 2 > 0 continuously tend to zero with
increasing n. Also the rescaled conformal moments (a) for all values of possess this behaviour.
We remark that the conformal moments (e) will mostly not give quantitatively different results
as (b), so we will in the following not present them.
In Fig. 9 we show the resulting GPDs for = 0.5 and GDAs for = 0.4. As has already
been discussed the skewness dependence of the conformal moments in Fig. 8, cases (a) and (b)
can be hardly distinguished, for GPDs they are quite the same. In case (c) the area in the central
region and the magnitude of the maximum at = 0 are enhanced, while the area of the outer
region shrinks (the lowest moment of all functions is normalized to one). Consequently, also
41
Fig. 10. The reduced GPD at the cross-over point x = , multiplied with the factor , (left) versus and the convolution
of the GDA with 1/z (right) versus , same as in Fig. 8.
the magnitude at the cross-over points x = decreases. The reverse situation is observed for
case (d). This is caused by the analytic properties of the conformal moments with respect to
the variable j , which we explained above. Comparing both panels in Fig. 9, one immediately
sees that the width of GPDs shows up in the end-point behavior of GDAs. It is remarkable
that the GDAs (a) and (b) possess shapes that are governed mainly by the first partial wave of
the conformal expansion. Indeed, for case (a) it turns out that in the region 0.05 0.95
the higher partial waves give at most a 10% percent correction. In the much narrower region
0.3 0.7 this is within the same accuracy also true for the GDA (b). This region starts
slightly about or below the value, where the series turns over to be divergent, i.e., for 0.25
and 0.75 . A slightly smaller convergency radius holds for the GDA (c) and (d).
In Fig. 10 we show the quantities that enter the scattering amplitude for a hard exclusive
process, in which only the virtuality of the incoming photon can be varied. Under such circumstances the Compton form factors, i.e., Eq. (104) or their crossing analogues Eq. (112), are
measurable. The non-perturbative distributions cannot be directly determined by deconvolution,
but at least within our toy anstze the correspondence between momentum fraction dependence
and the amplitudes is clear-cut. In the left panel of Fig. 10 we depict the GPD at the point x =
multiplied by . This cross-over point trajectory is accessible in single spin asymmetry measurements. Within our anstze both the normalization and the slope encodes information about the
GPD shape. Compared with the left panel in Fig. 9, one realizes that the narrowest (widest) GPD
has the smallest (largest) value at the cross over point x = . The differences of the trajectories
diminish with decreasing . Cases (a) and (b) are only distinguishable when tends to one. Certainly, the slope of the cross-over point trajectories is dictated by the strength of the skewness
dependence of the conformal moments.
In the right panel of Fig. 10 we show the dependence of the scattering amplitude, given as
convolution of a GDA with a LO hard-scattering amplitude, see Eq. (112). It has been evaluated
within the partial wave expansion (113), taking the first thirty terms into account. The result is
only displayed for the region in which also the series for the conformal moments (b)(d) converges, i.e., for 0.25 0.75, The end-point behavior of GDAs determines the normalization
of the scattering amplitude while the dependence is in all cases (rather) flat. As already mentioned, these convergency problems arise from an exponential growth caused by the factor 2n
that is contained in the normalized conformal partial waves (39). One would expect that the conformal partial wave expansion of GDAs converges for all physical values of and z. To ensure
this, the conformal moments must be exponentially suppressed by a factor 2n for large n. We
42
remind that the normalization has been adopted from parton densities and so this factor will drop
out in the limit 0.
We come now to the dependence of the cross-over point trajectory for smaller values of .
The trajectory is represented by the MellinBarnes integral
1
q(, ) =
2i
c+i

dj j 1
ci
2j +1 (5/2 + j ) (1/2 + j )(9/2) 2

bj ,
(3/2)(3 + j ) (1/2)(9/2 + j )
(137)
which can be used to derive, as explained in Section 3.2, an analytic expansion in terms of powers
in . The leading terms
105
35
q(, ) =
(138)
+
6 2
8 2
are universal in all cases and arise from the poles at j = 1/2 and j = 3/2. The accuracy
of this approximation for the cross-over point trajectory is for all cases about 1% for = 0.05
and, of course, starts to be much better with decreasing . Corrections to the approximation
(138) depend on the specific model and will not be discussed here. For large values of they are
smaller for ansatz (a) than for (c) or (d). Remarkably, in case (b) the integral (137) can be exactly
calculated:

2
(2 )7/2
1/2, 5/2
1
.
q(, ) =
(139)
2 F1
9/2
2 2
Let us summarize the lessons from this investigation. Unluckily, after crossing it turns out
that three of our models, (b)(d), lead to convergence problems for the conformal partial waves
series. Since these series should converges, the conformal moments must for > 1 exponentially
decrease as 2n with increasing n. Viable anstze for the conformal moments are holomorphic
functions of the conformal spin j that respect the following requirements:
They should be bound for j with | arg(j )| /2 for || 1;
An expansion of these functions in 2 should exist;
They should show an exponential suppression for 2 > 1 by a factor 2n (for integer n).
In addition they should be flexible enough to be able to generate a large variety of GPD shapes.
In our examples only case (a), i.e., the analytic continuation of Gegenbauer polynomials, satisfy
the three requirements. Applying suitable integral transformations to them allows to generate
associated polynomials. We did not study in detail how flexible the GPD shapes can be parameterized within this method but got the impression that only small changes for large are
possible. We are award that these examples do not cover all possible types of conformal moments. One important lesson is that the analytic properties of conformal moments determine the
qualitative features of GPDs. Certainly, our examples here give only a first insight into this connection. For completeness, we mention the existence of a symmetry relation that arises from the
definition (35), namely, the invariance under the replacement j j 3 [37]:

21+j (3/2 + j ) j
21+j (3/2 + j ) j
cj (x, ) =
cj (x, )
.
(140)
(3/2)(3 + j )
(3/2)(3 + j )
j j 3
Since, however, perturbative QCD, e.g., anomalous dimensions, already violate this symmetry,
there is no need to implement it in a non-perturbative ansatz for conformal moments.
43
4.2.2. Implementation of momentum squared dependence

Now we address the implementation of the 2 dependence. We might include it by introducing 2 depended parameters in the ansatz (125) in such a way that no singularities in the first
and forth quadrant of the complex j -plane appear. For the sake of simplicity we write here a
factorized ansatz

( + 1 + j )( + + 2) 2 2

bj Fj ,
mj , 2 , =
(141)
( + 1)( + + 2 + j )
with the normalization condition Fj (2 = 0) = 1. In the case that the form factors Fj (2 )
are j independent the x and 2 dependence is obviously factorized, too. Lattice calculations
for the first moments of u and d quark GPDs suggest that such a factorization is actually not
correct. According to these results, the cut-off mass squared in a dipole fit increases with j = n.
Unluckily, the systematic theoretical uncertainties of these results, especially, those associated
with the chiral extrapolation are still kind of large, but the preferred fits suggest a linear growth
with (j + 1)13 [75]. A second constraint on the 2 dependence arises from the lowest moment
of GPDs, which is related to partonic form factors, e.g., for the GPD H :
p,val 2
=
F1
1

dx Hpval x, , 2 .
(142)
p,val
Where F1 (2 ) can be expressed in terms of the proton (p) and neutron (n) electromagnetic
Dirac form factors according to

p
p
F1d,val 2 = F1 2 + 2F1n 2 .
2F1u,val 2 = 2F1 2 + 2F1n 2 ,
(143)
We remark that the lowest moment of a GPD is simply obtained by setting j = 0 in the
ansatz (141).
Now we are in the position to build a minimal GPD ansatz for the valence quark GPDs H
which satisfy the theoretical constraints in the forward case and provide for the lowest moment
the correct 2 dependence. Suppose we use the parameterization of the forward parton distribution at a given input scale in the form (127), the conformal GPD moments read then, e.g., for the
u quark

u
mj val , 2

2
=
mj , 2 , + A mj , 2 + 1/2, + B mj , 2 + 1, ,

1+A +B

( + 1 + j )( + + 2) 2 u,val 2

mj , 2 , =
(144)
bj F1
/(j + 1) ,
( + 1)( + + 2 + j )
normalized to 2 for j = 0 and 2 = 0. Guided by the lattice results, we rescaled here the dipole
masses in the form factor F1u,val by j + 1, which results in a dependence on the ratio 2 /(j + 1).
13 The m2 = (n + 1)m2 dependence for the squared dipole masses m2 , taken from Table I of Ref. [75] gives a very
d,n
d,n
d,0
u+d
good fit for the generalized form factors Aud
n,0 (m = 897 MeV) and An,0 (m = 744 MeV) with n = 0, 1, 2. For the
large pion mass m = 897 MeV one might have the impression that the growth is stronger (weaker) for the (axial-)vector
case, which might be caused by a different intercept in the power behavior. Certainly, a definite conclusion cannot be
drawn from present lattice measurements.
44
The remaining free parameters A , B , , are taken from the parton density parameterization
and for bj (2 ) one can use one of the functions suggested above, where preference is given to
case (a). Analogously, one can deal with other species of GPDs that reduce in the forward limit to
parton densities. However, for these the 2 dependence is typically far less known. In particular,
unpolarized sea quark GPDs are not constraint by elastic form factor measurements.
Let us have a closer look at the parameterization (144). For j = 0 we obviously find from this
parameterization
1

dx Huval x, , 2 = 2b0 2 F1u,val 2 ,
(145)
where the -dependence drops out, b0 (2 ) = 1 and the correct normalization is ensured. Setting
2 = 0, it is also clear from Eq. (141) that we arrive by an inverse Mellin transform at the
parton density. It has been mentioned above that the GPD for 0 and small x is dominated
by the leading Regge trajectory. In the following we argue that after a small modification this
trajectory is already present in our parameterization. In our case we are dealing with the 0
and trajectories, which according to the analysis of Ref. [72] are parameterized by a linear
2 -dependence

2 = 0.42 + 2 0.95 GeV2 ,
(146)
2 = 0.48 + 2 0.88 GeV2 .
If we take for the elastic nucleon form factor the standard dipole parameterization and consider
only the first pole in the complex j -plane of the parton density Mellin-moments, i.e., the pole
that appears on the negative axis at the largest value of j , our parameterization for the conformal
moments reads

2
mval
j ,
(j + 1 + )
1
2
+ 1 + j md (j + 1 + ) 2 (1 + )

B
A
+
.
4MN2 (1 + c)2 /(j + 1 + c) m2d (1 + d)2 /(j + 1 + d)

(147)
Here MN is the nucleon mass, md is the dipole mass appearing the parameterization of the Sachs
p
form factors, e.g., in the electric one of the proton GE (2 ) = 1/(1 2 /m2d )2 . A and B are
two constants, the values of which can be read off, e.g., from Ref. [69]. Moreover, we modified
here the rescaling of the 2 dependence in such a way that the pole at j + 1 + = 0 cancels
against the scaling factor (j + 1 + ). The constants c and d, appearing in the scaling factors
of the remaining 2 dependence, are chosen to be positive so that they do not interfere with the
leading Regge trajectory. This linear trajectory is given by

2 = (0) + (0)2 ,
(0) = , (0) =
1+
m2d
(148)
and so we find with Eqs. (144) and (147)

2
mval
j , =
( + 2 + j )( + + 2) j (2 )bj (2 )
2
1 + A + B ( + 1)( + + 2 + j ) (j + 1) (2 )
(1 + A + B ),
(149)
where the remaining 2 dependence is accumulated in the function

B
A
+ 2
.
j 2 =
2
2
4MN (1 + c) /(j + 1 + c) md (1 + d)2 /(j + 1 + d)
45
(150)
As we saw, after rescaling of the dipole masses the leading pole at j = 1 in Eq. (144) is
in the parameterization (149) shifted to j = 1 + (2 ) by an amount proportional to 2 . The
numerical value of the dipole mass is md = 840 MeV and that of depends on the factorization
scale. When we choose the scale to be the intercept of the Regge trajectories (146) we find the
following slopes
(0) = 0.82 GeV2 ,
(0) = 0.74 GeV2 .
(151)
These values are only about 15% smaller than the ones given in Eq. (146). In view of the fact
that we used just the standard dipole parameterization of the elastic form factors and neglected
all non-leading Regge trajectories, this agreement is quite astonishing. We interpret it as further
evidence that Regge theory or at least Regge phenomenology is indeed applicable to GPDs.
A deeper understanding of this issue could provide the key to a dual description of GPDs in
terms of hadronic degrees of freedom and certainly warrants dedicated efforts.
4.2.3. Numerical consequences for the probabilistic interpretation of GPD H
As mentioned in the introduction, GPDs possesses for = 0 a probabilistic interpretation in
the infinite momentum frame. In particular the Fourier transform14

d 2 i b

Hq (x, b) =
e Hq x, = 0, 2 = 2
(152)
2
(2)
is the probability to find a quark species q inside the nucleon with momentum fraction x at impact
The latter is defined relative to the center of momentum of the hadron, i.e., b is the
parameter b.
distance of the active parton in the transversal direction from this center. It is worth mentioning
that the definition of such a center is based on the existence of a Galilean subgroup of transverse boosts in the infinite moment frame [76]. Within light-cone quantization such transverse
boosts have a field theoretical definition in terms of two conserved charges, expressed by the plus
components of the energymomentum tensor. Their eigenvalues are good quantum number that
labels the states of the hadron. Conveniently, they are chosen to be zero for the center.
This probabilistic interpretation of the GPD (152) has inspired several authors to build GPD
anstze to get a first glimpse of the three-dimensional tomographic picture of the nucleon. This
has often been done using an exponential ansatz for the 2 -dependence, which might serve
it purpose in the space like region, however, violates the analytic properties of scattering amplitudes, etc., such that crossing relations become meaningless. In the following we study our
conformal moments for = 0 under this aspect. The only uncertainty which is left in our GPD
representation is the j -dependence of the form factor. In Fig. 11 we present the momentum
fraction dependence for fixed 2 = 0.25 GeV2 and two different resolution scales. Obviously, the implementation of the Regge trajectory (solid and dash-dotted lines) results in a flatter
x-dependence compared to a simple rescaling of the dipole mass squared with j + 1 (dashed and
dotted lines).
14 To avoid a confusion with the definition of the GPD H (x, , 2 ) in the axial-vector case, we omit the tilde symbol
q
for the Fourier transform of Hq (x, , 2 ). The quantities in the two-dimensional impact space b are indicated by their

argument b.
46
Fig. 11. The u-valence quark GPD Hu (x, = 0, 2 , Q2 ) of the proton is displayed in panel (a) versus x for fixed
2 = 0.25 GeV2 with a 2 -dependence arising from a Regge trajectory via Eqs. (148), (149), and (150) (solid and
dash-dotted) and with a rescaled 1/(j + 1) 2 -dependence via Eq. (144) (dashed and dotted). The GPD is evolved from
the input scale Q2 = 0.5 GeV2 (solid and dashed) to Q2 = 10 GeV2 (dash-dotted and dotted). In panel (b) we show
for the 2 /(j + 1)-depended GPD Hu the expectation value for the square of the impact parameter (154). The different
lines correspond to Q2 = 0.5 GeV2 (solid), Q2 = 1 GeV2 (dash-dotted), Q2 = 10 GeV2 (dashed), Q2 = 100 GeV2
(dotted) and Q2 = 10100 GeV2 (thin dotted). We used the following parameter set: A = B = 0, = 1/2, = 3 and
c = d = 1.
As a side remark, we comment on the factorized 2 ansatz for GPDs. Although, it is wrong
in principle, this does not necessarily imply that all estimates for observables fail completely, at
least not for smaller values of 2 . For the anstze of conformal moments, we discussed in the
both previous sections, the normalization of the resulting GPDs between the different versions of
2 -dependence varies for 0.08 x 0.2 and |2 | < 0.3 GeV2 , i.e., in the fixed target kinematics, not larger than 20%. Certainly, for lower or larger values of x the differences especially in
the overall size can increase drastically, compare, e.g., the solid and dashed line in Fig. 11(a). On
the other hand the suppression introduced by the Regge motivated ansatz (see solid and dashdotted lines) is welcome to suppress sea quark and gluonic contributions, which notoriously are
overestimated in the factorized 2 ansatz for hard exclusive electroproduction processes.
The average distance from an active parton to the center of the nucleon is defined as [10,11]

Q2 )

2
d b b2 Hq (x, b,
2

= 4 2 ln Hq x, = 0, 2 , Q2 2 =0 .
b q x, Q =
(153)
2
Q )
d b Hq (x, b,
Within the parameterization (144), where for simplicity we again rescale the 2 dependence by
1/(j + 1), this average distance can be exactly calculated and expressed in terms of forward
parton distributions
1 dy
2

2
x y q(y, Q )
q
2

4 2 ln F1 2 2 =0 .
b q x, Q =
(154)
2
q(x, Q )
The 1/(j + 1) factor that arises in Mellin space from the differentiation with respect to 2 gives
1
in x space rise to the integral x dy/y . Obviously, by a more refined rescaling of the 2
dependence, see Eq. (147), we can express the resulting average by a more complex integral. For
small x this quantity tends to a constant, depending only on the resolution scale, while at large
x it vanishes as (1 x). The latter behavior is a consequence of the linear j -dependence of the
dipole masses. Such a behavior has been rejected in Ref. [22]. Namely, the quantity
d(x) =
b2 q (x)
(1 x)2
(155)
47
is interpreted, based on a partonic picture, as the distance of the active parton from the center of
momentum of the spectators and should therefore be finite for x 1. Consequently, b2 q (x)
should vanishes at least with (1 x)2 for x 1. Although this argumentation is further supported by perturbative QCD arguments [21], it is in our opinion not excluded that the simple
parton picture can be misleading for quantities, which have no well-defined field-theoretical analog. Note also that a (1 x)2 behavior of b2 q (x) for x 1 requires a (1 + j )2 growth of the
squared dipole masses. This means in terms of Regge phenomenology that the trajectories should
possesses a (small) non-linear term.
Let us come back to our ansatz. To include the scale dependence we use for the GPD
Hq (x, = 0, 2 , Q2 ) in Eq. (153) the MellinBarnes integral and insert the evolution operator, compare with Eq. (102),
2

b q x, Q2 =
c+i
ci
c+i
ci
j 1
x
(1+j +)
dj (1+j
++)(j +1) exp{
j 1
(1+j +)
dj x (1+j
++) exp{
(0)
j
2
(0)
j
2
Q2
d s ( )
}
Q20 2
Q2
d s ( )
}
Q20 2
q
ln F1 val 2 2 =0 .
2
(156)
For experimental accessible values of Q2 and for small x, both the numerator and denominator
are dominated by the leading pole at j = 1 . Shifting the integration path to the left we find
for x 0 the constant value

b 2

4
q
x = 0, Q2 =
ln F1 val 2 2 =0 .
q
2
(157)
We remark that a Regge motivated ansatz would induce a double pole in the numerator and, consequently, a logarithmic modification with respect to both the x- and Q2 -dependencies. Within
the dipole ansatz for the partonic form factors, taken from Ref. [69], we have

b 2

1
x = 0, Q2 = 0.4 fm2 ,
u
b 2

1
x = 0, Q2 = 0.53 fm2
d
(158)
for the u and d valence quarks. From Fig. 11(b), where we used the generic value = 1/2,
we can read off the qualitative x-dependence of b 2 qval (x, Q2 ) for given resolution scale Q2 . It
can be roughly approximated by a logarithmic growth with decreasing x which changes slope at
some cusp point xcusp (Q2 ) 103 . With increasing Q2 this cusp is washed out. The increase
with ln(1/x) visible in Fig. 11(b), will not continue in the limit x 0 but will saturate at a
finite value (158). Numerically, we checked this qualitative feature up to Q2 = 1010000 GeV2
and 1/x = 1040 .
So within our simple ansatz, which can be easily refined, we reproduce following well-known
grand picture. At large x the valence quarks are in the center of the nucleon. With decreasing x
they become ever more delocalized in the transverse direction and will move away in transversal
direction with decreasing momentum fraction x. In the valence quark region with x 0.3 the
average transverse distance is of the order 0.4 fm. With increasing resolution scale this value is
slowly decreasing. In the small x region their transverse distance grows, reaches (for the ansatz
we used) 0.9 fm in the limit x 0 for u valence quarks and a slightly higher value for d
quarks. Hence, the charge of the proton, which is probed in a scattering experiment, has a nontrivial transverse distribution.
48
5. Summary and conclusions

In this paper we have derived a new representation for leading twist-two GPDs and GDAs in
terms of MellinBarnes integrals over the complex conformal spin. This representation is rather
analogous to the partial wave expansion of scattering amplitudes with respect to complex angular momentum, given in terms of Legendre polynomials or functions, respectively. Indeed, also
our partial waves can be expressed in terms of associated Legendre functions of the first and
second kind. Mathematically, there should exist a one-to-one relation to other representations
that are based on conformal symmetry. The transformations between different representations,
however, still have to be worked out in detail. For instance, taking the Mellin transform of the
MellinBarnes integral must lead to the integral kernel that maps the effective forward parton
distribution to GPDs [38,39] or employing the Fourier transform one must arrive at the lightcone position space representation [3537]. We confirm and generalize the results for the gluonic
sector given in Ref. [37]. The advantage of our new representation is that the central and outer
region are obtained from the same conformal moments of a GPD and polynomiality is manifestly
implemented. This has not been done so far in the approaches, we mentioned. Sometimes, the
central region was even treated incorrectly, for comments see Ref. [39]. We derived a spectral
representation of conformal kernels with complex valued conformal spin. However, we are not
completely satisfied with this representation, since some support restriction have to be fixed explicitly by step-functions. So far an orthogonal eigenfunction basis for these kernels can only be
given as series of mathematical distributions, which are labelled by non-negative integer conformal spin.
To leading order the kernels and hard-scattering amplitudes respect conformal symmetry
and so we could provide the solution of the evolution equation and the Compton form factors
for DVCS as MellinBarnes integral. This representation allows a simple and stable numerical
evaluation of these quantities. This has some practical advantages, especially, having efficient
numerical routines at hand for the evaluation of the MellinBarnes integral, one might be able
to evaluate this quantities in real time rather than using a database of Compton form factors or
GPDs. We even demonstrated for the Compton form factors that by means of the MellinBarnes
integral a systematic analytic approximation in powers of is feasible.
Beyond LO order the conformal symmetry is broken in a subtle way by the minimal subtraction scheme, applied to the divergencies of composite operators, and the trace anomaly of
the energy momentum tensor. The first effect can be cured by a finite renormalization, while the
latter one can be absorbed into either the hard-scattering amplitudes or the evolution of GPDs.
Such symmetry breaking effects have been studied for = 1 in connection with the pion distribution amplitude and the pion elastic and 0 transition form factor. Depending on the
model for the pion distribution amplitude one finds a 10% variation of the NLO corrections [77].
For skewness < 1 one would expect an even smaller effect because the conformal symmetry
breaking is suppressed by a factor 2 . For realistic experiments typically 0.4. Further studies are desirable to clarify this issue and, hopefully might provide a systematic expansion of the
conformal symmetry breaking effects in powers of 2 .
Certainly, the evaluation of the analytic continued conformal moments from a given GPD
ansatz is a rather non-trivial task, the difficulty of which depends on the analytic properties of
the corresponding GPD. Since, however, GPDs are almost unknown non-perturbative functions,
one should rather model the conformal moments directly instead of the GPDs or DDs. This
is a non-trivial task with respect to skewness dependence, since certain analytic properties of
the conformal moments must be respected. Some effort is required to tune the internal skewness
49
dependence, arising from the conformal partial waves, accordingly. In the end, however, it turned
out that our anstze generated GDAs with non-vanishing end-point contributions. This illustrates
again that known constraints for GPDs are very difficult to fulfill. An important observation one
can draw from our examples is that strong skewness effects already show up in the first few
conformal moments for non-negative integer conformal spin. So it is worthwhile to calculate
them on the lattice to get a clue for the strength of the skewness dependence of GPDs.
A partial wave expansion of the conformal moments itself avoids the problems described
above yields a most flexible parameterization of GPDs in terms of form factors. These form factors are related to particle exchanges in the t-channel, but this connection still has to be worked
out in detail. For = 0 there is no difference between conformal spin and ordinary spin and the
conformal partial wave expansion turns over into one with respect to spin. In this kinematical
domain we found some evidence that Regge phenomenology can be used as a reliable guide
for modeling the conformal moments. Especially, for unpolarized valence quark GPDs the parameterization of parton densities and elastic electromagnetic form factors can be unified and
interpreted as a leading Regge trajectory. Since GPDs depend on the factorization conventions,
such a connection can only hold approximatively. Moreover, Regge theory is only applicable for
physical amplitudes and going beyond the leading trajectory has its own difficulties. Nevertheless, the Regge analogy provides some guidance for the modeling of conformal moments. This
is especially important for those GPDs that are difficult to extract from experiments.
The advantage of the MellinBarnes representation has been demonstrated for several analytic
and numerical examples, especially, for unpolarized valence quarks in the = 0 case. The only
unknown in this limit is the spin, i.e., j , dependence of the form factors, which for j = 0, are
measured in elastic electron proton scattering. The momentum fraction dependence of the GPD
follows then from the t- and Q2 -dependencies, where the boundary condition at the input scale
Q20 and at 2 = 0 can be simply taken from the parameterization of Mellin moments for parton
densities. Remarkably, no additional fitting procedure is needed to satisfy the GPD constraints.
However, different parameterization of the form factors with respect to j , can lead to a different
holographic picture of the nucleon. Certainly, the important task here is to pin down the remaining degrees of freedom for the j dependence. No question, improved lattice calculation with a
realistic pion mass can provide at least a partial answer.
We would like to add a speculation concerning the experimental access to this dependence
in hard exclusive reactions. Suppose it turns out from lattice measurements that the skewness
dependence of the conformal moments is weak, the -dependence for the scattering amplitude
is (approximately) known and the only degree of freedom left is the unknown j -dependence of
these form factors, which determines the shape of the trajectory of the cross-over point of a GPD
as function of , in dependence of 2 . Such a trajectory can be explored in single beam spin
experiments. Taking Mellin moments of this trajectory, which requires of course some interpolation, one directly gets the form factors, up to some normalization factor, in dependence of the
(conformal) spin.
Let us finally stress that the crossing relation between GPDs and GDAs is very simple for
conformal moments. Apart from a trivial rescaling procedure of the skewness dependence it
involves only the 2 or W 2 dependence. One has to perform an analytic continuation of the form
factors from the space- to the time-like region, which requires only a suitable parameterization
in terms of rational functions (linear combination of monopole or dipole forms), which scale for
2 such as predicted by dimensional counting rules. As there exist also non-perturbative
scales, like the hadron masses themselves or QCD , anomalous, i.e., logarithmical deviations,
from the canonical scaling should be present to some extend. Although, we were not able to
50
give for all examples of conformal moments an unified representation of GPDs and GDAs in
terms of a MellinBarnes integral this is not a restriction in practice. If one has an ansatz for
the dependence of the conformal moments on the complex conformal spin, one can employ for
GDAs the partial wave expansion with integer conformal spin.
In conclusion, we have introduced a representation that makes it easier to include GPDs and
GDAs in phenomenological studies and offers new theoretical possibilities for the investigation
of perturbative and non-perturbative aspects.
Acknowledgements
For discussions on mathematical aspects we are indebted to A. Manashov and for a general
discussion on GPDs we like to thank M. Diehl, which inspired us to include several remarks.
This project has been supported by the DFG.
Appendix A. Integrals
In this appendix we collect several integrals, which appear in the evaluation of moments or
the convolution of GPDs with the hard scattering amplitude. The partial waves pj (x, ), given
in Eqs. (53), (54), and (55),
by associated Legendre functions of the first
might be expressed
and second kind, i.e., by (1 x 2 )Pj1
(x)
and
(1 x 2 )Q1
+1
j +1 (x). The conformal moments
cj (x, ), see Eq. (35), can be represented within the same basis, however, divided by the weight
(1 x 2 ) [78]. The integrals, presented in the following, can then be read off from diverse integral
tables. Moreover, we give the relation between conformal and ordinary Mellin moments.
The conformal moments of the partial waves for complex valued conformal spin read in the
outer region

dx ck (x, )pj (x, ) =
sin(j ) Nkj ()
,
kj
e j > e k,
(A.1)
where the normalization factor is given by

Nkj () =
kj
2(3 + k)(5/2 + j )
,
(3 + j )(3/2 + k)(j + k + 3) 2
Njj () = 1.
(A.2)
Note that this integral (A.1) only converges for e j > e k. In the central region we have the
following integral

dx ck (x, )pj (x, ) =

sin(j ) (j + 1)(j + 2) sin(k) Nkj ()
(k + 1)(k + 2)
j k
(A.3)
On the r.h.s. two terms appear in the brackets. The sum of the former one and the integral (A.1)
is proportional to the identity (j k), however, the latter one gives an addendum. Fortunately,
for integer value k = m = 0, 1, 2, . . . it will not contribute. Remarkably, if the support of ck (x, )
is extended to x by means of
ck (x, ) =
51
2k+2
sin(k)(1 + k)(3 + k) k3
x
2
(3/2 + k)(5/2 + k)

(k + 3)/2, (k + 4)/2 2
2 F1
x 2 , x
5/2 + k
(A.4)
and for pj (x, ) by analytic continuation into the region x < 0, we have the following integral

dx ck (x, )pj (x, ) =
(j + 1)(j + 2) sin(k) Nkj ()

,
(k + 1)(k + 2)
j k
e k > e j.
(A.5)
Up to the sign it is equal to the sin(k) proportional contribution on the r.h.s. in Eq. (A.3).
Hence, the sum of integrals (A.1), (A.3) and (A.5) might be understand as a limit e k e j
that yields the identity, formally written as

dx ck (x, )pj (x, ) = 2i sin(j )(k j ).
(A.6)
We add that for non-negative integer values of the conformal spin the integral (A.3) establishes
the orthogonality relation (40) for Gegenbauer polynomials.
Let us also give here the Mellin moments of the partial waves pj (x, ) for integer value
n = 0, 1, 2, . . . . For the integration in the outer region we have

dx x n pj (x, )

nj sin(j )
1/2 + j/2, 1 + j/2, j/2 n/2
=
3 F2
1 ,
5/2 + j, 1 + j/2 n/2
(n j )
e j > e n,
(A.7)
while from the central region we find, up to the overall sign, the same expression

nj sin(j )
1/2 + j/2, 1 + j/2, j/2 n/2
dx x pj (x, ) =
3 F2
1 .
5/2 + j, 1 + j/2 n/2
(j n)
n
(A.8)
Neglecting the sin(j ) term, these expressions will contain single poles at j = {0, 2, . . . , n} and
j = {1, 3, . . . , n} for even and odd n, respectively. The values for the two lowest moments are:

dx pj (x, ) =
j
23 (5/2 + j ) sin(j )
2
,
j (1/2)(4 + j )
j 1
2
sin(j )
24 (5/2 + j )
.
dx xpj (x, ) =
(j 1)(4 + j )(1/2)(3 + j )
(A.9)
(A.10)
Analogous as discussed above, the sum of both integrals (A.7) and (A.8) should be understood
as a limit that results in a linear combination of -functions, which are concentrated in j =
{0, 1, . . . , n}. From the MellinBarnes representation of GPDs we explicitly find then the usual
52
Mellin moments, expressed in terms of conformal ones:

1

dx x n q x, , 2
n ni

i=0

(1 + (1)ni )n!(5/2 + i)
mi , 2 .
2i!(1 + n/2 i/2)(5/2 + i/2 + n/2)
(A.11)
The inverse relation can be brought in the form

mn ,
[n/2]

i=0
2i
(1)i n!(3/2 + n i)
i!(n 2i)!(3/2 + n)
1

dx x n2i q x, , 2 .
(A.12)
In the convolution of the hard-scattering amplitude with generalized parton distributions the
following two integrals appear:

dx
pj (x, ) =
+x
1+j
(5/2 + j )
2
,
(3/2)(3 + j )
1+j
2
dx
(5/2 + j )
pj (x, ) = eij
.
x i
(3/2)(3 + j )
(A.13)
(A.14)
We note that the reduction to non-negative integer values of the conformal spin is simply done
by the replacement j n. In this case only the central region contributes in these integrals,
especially, the imaginary part will drop out
Appendix B. Gluonic sector
Here we present the MellinBarnes integral for gluon GPDs, defined in Eq. (2). Let us first
note that the index of Gegenbauer polynomials, appearing in the definition of conformal moments, is determined by group theory. To be more general, we consider the light-ray operator
O(1 , 2 ) = (2 n)(1 n)
(B.1)
that contains two quantum fields, which live on the light-cone n2 = 0. For gluons the field is
build by the field strength tensor. We assume that these fields have definite spin projection s on
the light-cone, i.e.,
n n (n) = s(n).
(B.2)
Here is the usual generator of Lorentz transformation, acting on a field (x) at x = 0.

Moreover, the canonical dimension of the field (x) is denoted as . The conformal spin of the
field is defined as
1
j = ( + s).
2
(B.3)
53
It characterizes the behavior of the field under collinear conformal transformation, which can be
viewed as the projective transformations on a line:
=
a + b
,
c + d
ad bc = 1,

a + b

2j
(n) (n) = (c + d)
n .
c + d
(B.4)
In four-dimensional space the quantum numbers are

3
= ,
2
= 2,
1
for quark fields,
2
s = 0, 1 for gluon field strength tensor.
s=
(B.5)
(B.6)
The conformal operators are obtained by the group theoretical decomposition of the light-ray
operator (B.1) into irreducible representations. If both quantum fields have the same conformal
spin, then the (local) conformal operators which have definite conformal spin are characterized
by Gegenbauer polynomials with index = 2j 1/2:

l
l 1 2
O(1 , 2 ) = =0 , n l,
Onl = i (1 + 2 ) Cn
(B.7)
1
2
1 + 2
where the conformal spin of these operators is 2j + n. For each given conformal spin, there
appears an infinite tower of operators which are labelled by the quantum number l, related to their
spin or if one likes to their canonical dimension. For leading twist operators the spin projection s
must be maximal so that the twist of the fields t = s is minimal. Consequently, to leading twist
the quark fields have conformal spin j = 1 and gluon ones j = 3/2. The index of Gegenbauer
polynomials is in the former and latter case = 3/2 and = 5/2. We remark that the gauge link
factor connecting the fields along the light-cone does not change the construction via Eq. (B.7).
For gluonic GPDs (2) we define the conformal moments for n = 1, 2, . . .

x
(5/2)(n)
5/2
G
n1 G
with G cn (x) = n1
C (x)
cn (x, ) =
cn
(B.8)
2 (3/2 + n) n1
in such a way that in the forward limit the usual normalization of Mellin moments appear:
lim0 G cn (x, ) = x n1 . Moreover, as in the quark sector, they project on operators with conformal spin 2 + n. Compared to the Mellin moments of parton densities within our convention,
one power in x seems to be missed, however, it is included in the definition of GPDs. For instance, in the vector case we have

lim G F V x, , 2 = xg(x) for x 0,
(B.9)
0
where g(x) is the unpolarized gluon parton density. Analogous convention holds for the axialvector case. The analytic continuation of the conformal spin j in the conformal moments is again
done in terms of hypergeometric functions

(3/2)(j + 4) j 1
j + 1, j + 4 x
G
cj (x, ) = 4
(B.10)
2 F1
2 .
3
2 (3/2 + j ) 2
The conformal partial waves are analogously constructed as described in Section 2.3.1.
Namely, by including the weight (1 x 2 /2 )2 with a suitable normalization and from the requirement that the partial waves are vanishing at x = and are continuous at the cross-over
54
point x = :
G

x
x
+ (x )j G Qj
pj (x, ) = |x| j G Pj
(B.11)
where

2j (5/2 + j )
j 1, j + 2 1 + x
(1 + x)2 2 F1
,
2
3
(1/2)(j )

sin(j ) j
j/2, (j + 1)/2 1
G
Qj (x) =
x 2 F1
x2 .
5/2 + j
Pj (x) =
(B.12)
(B.13)
Moreover, Bose symmetry implies definite symmetry of the gluonic GPDs (2) under the transformation x x, i.e., in the (axial-)vector case they are always (anti-)symmetric. This property
is simply restored by forming symmetric or anti-symmetric partial waves, see Eq. (75). Corresponding to our normalization (B.9), we write
G
i

q x, , 2 =
2
c+i

dj

(1) G
pj (x, ) G pj (x, ) mj , 2 ,
sin(j )
(B.14)
ci
where only the analytic continuation of conformal moments for even (odd) integer values of n is
needed for (anti-)symmetric gluon GPDs.
Let us comment on the properties of the gluonic conformal partial waves (B.11). They are
related to the quark ones by a derivation with respect to x:
1 d G
pj (x, ).
(B.15)
j dx
As a simple consequence, the first derivative of the gluonic partial wave is smooth at the crossover point, while the second one has a jump as it is the case in the quark sector. Also at the point
x = the gluonic partial waves vanish as (x + )2 rather than (x + ) as for the quark ones.
Consequently, the convolution with the following hard-scattering amplitudes, which appear in
the electroproduction of transversely polarized vector mesons [79], exist
pj (x, ) =
dx G
pj (x, ) =
( + x)2
1+j
j (5/2 + j )
2
,
(3/2)(3 + j )
1+j
dx
j (5/2 + j )
i(j 1) 2
G
.
p
(x,
)
=
e
j
(3/2)(3 + j )
( x i)2
(B.16)
(B.17)
It has been argued already in Ref. [79] that this integrals without any further regularization exist,
which is shown here from a more general point of view. We add that in the convolution of the
gluonic GPDs with the hard-scattering amplitude at leading twist-two the integrals appear

dx G
pj (x, ) =
+x
j
4(5/2 + j )
2
,
(3/2)(4 + j )
j
dx
4(5/2 + j )
G
i(j 1) 2
pj (x, ) = e
.
x i
(3/2)(4 + j )
(B.18)
(B.19)
55
Finally, the convolution with conformal kernels is analogous done as outlined in Section 3 and
yields in the MellinBarnes integral representation of GPDs to a multiplication of the conformal
moments with its eigenvalues. Here we only mention that in correspondence with our normalization these eigenvalues follows from the Mellin transform of the forward limit, i.e., from the
Mellin moments of the usual DGLAP kernels.
Appendix C. MellinBarnes representation of conformal kernels
In the following we derive the MellinBarnes integral for a generic kernel (87), which is
conformal covariant and possess only non-negative eigenvalues kn . This is analogously done as
for GPDs in Section 2.3. Let us first remind on the support extension of the spectral representation
(87). The procedure for the kernel K(x, y) is well known and arise from the representation
1
K(x, y) =
1w
+
dw (x yw+ w )(w+ , w ).
dw+
(C.1)
1+w+
Here (w+ , w ) is the analog of the DD in the case of GPD, see Eqs. (10) and (21). Such a
kernel appears in the convolution with a light-ray operator (B.1) and it is in the mathematical
sense a distribution. The support in the whole (x, y)-plane can be read off from Eq. (23) and is
written here as

x x
K(x, y) = (y x) (x + 1) k(x, y) (x 1)k(x, y) +
,
(C.2)
y y
where the distribution k(x, y) has the integral representation
1+x
1+y
k(x, y) =
dw+ (w+ , x yw+ ).
(C.3)
Analogous as in the case of a GPD, see discussion in Section 2.1, this integral is only uniquely
defined in the central region, i.e., |x| 1, while in the outer region 1 |x| only the difference
k(x, y) k(x, y) enters. Here the ambiguities in the support extension of (w+ , x yw+ )
drop out. The support extension from the region |x|, |y| 1 to the whole support is unique [2],
see analogous discussion as in the last paragraph of Section 2.2.
Now we are in the position to derive the MellinBarnes integral representation for the kernel
K(x, y). To do so, we represent the series (87) in the region |x|, |y| < 1 as integral in the complex
plane that includes the positive real axis
1
K(x, y) =
2i
()
dj
1
pj (x)kj cj (y).
sin(j )
(C.4)
(0)
Here the functions cj (x) = cj (x, 1) and pj (x) = pj (x, 1) are defined in Eqs. (35) and (46),
respectively, and kj is the analytic continuation of the eigenvalues kn . They coincide with the
Mellin moments of the corresponding DGLAP kernel and might possesses a logarithmical growing for j . The integrand has simple poles at j = n = 0, 1, 2, . . . and so the residue theorem
leads to the series (87). Next, we deform the integration contour in such a way that it includes the
imaginary axis and is closed by an arc with infinite radius, see Fig. 3(a). It remains to show that
56
this latter contribution vanishes. The behavior of the integrand for large j with | arg(j )| /2
can be estimated from the behavior of hypergeometric functions and is in our case given by [59]
ej {arccosh(xi)arccosh(yi)}
1
pj (x + i)cj (y i)
.
sin(j )
sin(j )
(C.5)
Here the analytic continuation in the complex plane by the i prescription determines the branch
of the arccosh function, not uniquely defined for real valued x, y < 1. It is done in such a way
that the estimate (C.5) is applicable for 1 < y and x < 1 and | arg(j )| /2. Obviously, on
the infinite arc, both the denominator and numerator in Eq. (C.5) will exponentially grow and
the integrand will vanish as long as the condition | arccosh(x i) arccosh(y i)|
is satisfied. This is the case for 1 < x < y < 1. Thus, for this region we arrive at the Mellin
Barnes representation for the kernel K(x, y), i.e., for the function
i
k(x, y) =
2
c+i

dj
1
pj (x)kj cj (y).
sin(j )
(C.6)
ci
Form the spectral representation (C.1) it follows that the convolution of the kernel with a holomorphic test function (x) yield a holomorphic function, depending on y. Thus, we might now
employ analytic continuation to extend the representation (C.6) into the region 1 y. This will
not alter the convergency properties of the integral, since arg(arccosh(y)) = 0 for 1 y. We remind that pj (x) coincides for x 1 with the holomorphic function Pj (x), defined in Eq. (54).
Pj (x) has a branch cut, starting at x = 1, along the positive real axis and we might define its value
on the this cut by (Pj (x + i) + Pj (x i))/2. Hence, within this procedure the function k(x, y)
is uniquely continued into the region 1 y for all values 1 x, cf. with the prescription (16).
The MellinBarnes integral for the kernel K(x, y) follows now in an unique way from the
result (C.6) and its support, cf. Eq. (C.2). For 1 x we can write K(x, y) as the difference
K(x, y) = k(x, y) k(x, y)
for 1 x y.
(C.7)
We represent k(x, y) via the MellinBarnes integral (C.6), where pj (x) is obtained by
analytic continuation while cj (y) has now for 1 y a branch cut and a single pole at y = 1 on
the real axis. We have now to employ the principal value prescription, as above for Pj (x), and
so its value on the cut is

1
cj (y + i) + cj (y i) = cos(j )cj (y) + sin(j )dj (y),
2
(C.8)
where the function dj (y) is defined in Eq. (67). Using the asymptotic behavior of hypergeometric functions for large j , it can be shown that in the region 1 x < y the contribution Pj (x)kj dj (y), appearing in the integral (C.6), exponentially vanishes for j with
| arg(j )| /2. Hence, we can close the integration path so that the positive real axis is encircled. Since it is a holomorphic function, the corresponding integral vanishes. So we can drop this
contribution and find that for 1 x < y the extension of the integral kernel,

kj
kj
1
1
pj (x)cj (y) =
Pj (x + i) + Pj (x i) cos(j )Pj (x) cj (y),
sin(j )
sin(j ) 2
2
(C.9)
57
of the MellinBarnes integral for K(x, y)

i
K(x, y) =
2
c+i

dj
kj
pj (x)cj (y)
sin(j )
for x < y.
(C.10)
ci
The formula (C.9) defines the continuation of pj (x) for 1 x, which coincides with Eqs. (53)
(55) for = 1. The missing part of the kernel, i.e., for y < x, follows by the symmetry transformation x, y x, y from Eq. (C.10). Hence, the integral kernel can be written as
i
K(x, y) =
2
c+i

dj

kj
(y x)pj (x)cj (y) + (x y)pj (x)cj (y) .
sin(j )
(C.11)
ci
We excluded in our analyze here the line x = y. As long j kj is vanishing for j , we can
use analytic continuation to approach it. In all other cases a more advanced analyze is required.
However, this is in fact not necessary here, since it is obvious that a constant or logarithmic
behavior of kj for j is associated with -functions and +-prescriptions for singularities
at x = y. It is easy to check by forming the lowest moment that the MellinBarnes integral is
correct in that case, too.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
F.-M. Dittes, B. Geyer, D. Mller, D. Robaschik, J. Horeji, Phys. Lett. B 209 (1988) 325.
D. Mller, D. Robaschik, B. Geyer, F.-M. Dittes, J. Horeji, Fortschr. Phys. 42 (1994) 101, hep-ph/9812448.
X.-D. Ji, Phys. Rev. Lett. 78 (1997) 610, hep-ph/9603249.
A.V. Radyushkin, Phys. Lett. B 380 (1996) 417, hep-ph/9604317.
J. Collins, L. Frankfurt, M. Strikman, Phys. Rev. D 56 (1997) 2982, hep-ph/9611433.
M. Diehl, T. Gousset, B. Pire, O. Teryaev, Phys. Rev. Lett. 81 (1998) 1782, hep-ph/9805380.
A. Airapetian, et al., HERMES Collaboration, Phys. Rev. Lett. 87 (2001) 182001, hep-ex/0106068;
S. Stepanyan, et al., CLAS Collaboration, Phys. Rev. Lett. 87 (2001) 182002, hep-ex/0107043;
C. Adloff, et al., H1 Collaboration, Phys. Lett. B 517 (2001) 47, hep-ex/0107005;
S. Chekanov, et al., ZEUS Collaboration, Phys. Lett. B 573 (2003) 46, hep-ex/0305028;
C. Aktas, et al., H1 Collaboration, Eur. Phys. J. C 44 (2005) 1, hep-ex/0505061.
J. Breitweg, et al., ZEUS Collaboration, Eur. Phys. J. C 6 (1999) 603, hep-ex/9808020;
C. Adloff, et al., H1 Collaboration, Eur. Phys. J. C 13 (2000) 371, hep-ex/9902019;
S. Chekanov, et al., ZEUS Collaboration, Eur. Phys. J. C 24 (2002) 345, hep-ex/0201043;
A. Airapetian, et al., HERMES Collaboration, Phys. Lett. B 535 (2002) 85, hep-ex/0112022;
A. Airapetian, et al., HERMES Collaboration, Eur. Phys. J. C 17 (2000) 389, hep-ex/0004023;
C. Hadjidakis, et al., CLAS Collaboration, Phys. Lett. B 605 (2005) 256, hep-ex/0408005.
J. Ralston, B. Pire, Phys. Rev. D 66 (2002) 111501, hep-ph/0110075.
M. Burkardt, Phys. Rev. D 62 (2000) 071503, hep-ph/0010082.
M. Burkardt, Int. J. Mod. Phys. A 18 (2003) 173, hep-ph/0207047.
M. Diehl, Eur. Phys. J. C 25 (2002) 223, hep-ph/0205208.
A.V. Belitsky, D. Mller, Nucl. Phys. A 711 (2002) 118, hep-ph/0206306.
A.V. Belitsky, AIP Conf. Proc. 698 (2004) 607, hep-ph/0307256.
X.-D. Ji, Phys. Rev. Lett. 91 (2003) 062001, hep-ph/0304037.
M. Diehl, Phys. Rep. 388 (2003) 41, hep-ph/0307382.
A.V. Belitsky, A.V. Radyushkin, Unraveling hadron structure with generalized parton distributions, hepph/0504030.
M. Guidal, M. Vanderhaeghen, Phys. Rev. Lett. 90 (2003) 012001, hep-ph/0208275.
A.V. Belitsky, D. Mller, Phys. Rev. Lett. 90 (2003) 022001, hep-ph/0210313.
58
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
[51]
[52]
[53]
[54]
[55]
[56]
[57]
[58]
[59]
[60]
[61]
[62]
[63]
[64]
[65]
[66]
[67]
[68]
[69]
[70]
[71]
[72]
A.V. Belitsky, D. Muller, Phys. Rev. D 68 (2003) 116005, hep-ph/0307369.

F. Yuan, Phys. Rev. D 69 (2004) 051501, hep-ph/0311288.
M. Burkardt, Phys. Lett. B 595 (2004) 245, hep-ph/0401159.
P. Hagler, et al., LHPC Collaboration, SESAM Collaboration, Phys. Rev. D 68 (2003) 034505, hep-lat/0304018.
M. Gockeler, et al., QCDSF Collaboration, Phys. Rev. Lett. 92 (2004) 042002, hep-ph/0304249.
P. Hagler, et al., LHPC Collaboration, Eur. Phys. J. A 24S1 (2005) 29, hep-ph/0410017.
S. Brodsky, Y. Frishman, G. Lepage, C. Sachradja, Phys. Lett. B 91 (1980) 239.
A.V. Efremov, A.V. Radyushkin, Phys. Lett. B 94 (1980) 245.
D. Mller, Phys. Rev. D 58 (1998) 054005, hep-ph/9704406.
B. Melic, D. Mller, K. Passek-Kumericki, Phys. Rev. D 68 (2003) 014013, hep-ph/0212346.
D. Mller, Nucl. Phys. A 755 (2005) 71, hep-ph/0501158.
A.V. Belitsky, B. Geyer, D. Mller, A. Schfer, Phys. Lett. B 421 (1998) 312.
L. Mankiewicz, G. Piller, T. Weigl, Eur. Phys. J. C 5 (1998) 119, hep-ph/9711227.
A.V. Belitsky, D. Mller, L. Niedermeier, A. Schfer, Phys. Lett. B 437 (1998) 160, hep-ph/9806232.
A.V. Belitsky, D. Mller, L. Niedermeier, A. Schfer, Nucl. Phys. B 546 (1999) 279, hep-ph/9810275.
I. Balitsky, V. Braun, Nucl. Phys. B 311 (1989) 541.
N. Kivel, L. Mankiewicz, Nucl. Phys. B 557 (1999) 271, hep-ph/9903531.
A. Manashov, M. Kirch, A. Schfer, Phys. Rev. Lett. 95 (2005) 012002, hep-ph/0503109.
A. Shuvaev, Phys. Rev. D 60 (1999) 116005, hep-ph/9902318.
J.D. Noritzsch, Phys. Rev. D 62 (2000) 054015, hep-ph/0004012.
M. Polyakov, A. Shuvaev, On dual parametrizations of generalized parton distributions, hep-ph/0207153.
D. Mller, Next-to-next-to leading order corrections to deeply virtual Compton scattering: the non-singlet case,
Phys. Lett. B 634 (2006) 227, hep-ph/0510109.
X.-D. Ji, J. Phys. G 24 (1998) 1181, hep-ph/9807358.
E.R. Berger, F. Cano, M. Diehl, B. Pire, Phys. Rev. Lett. 87 (2001) 142302, hep-ph/0106192.
A.P. Bukhvostov, G.V. Frolov, L.N. Lipatov, E.A. Kuraev, Nucl. Phys. B 258 (1985) 601.
M. Bordag, D. Robaschik, Theor. Math. Phys. 49 (1982) 1063.
I.I. Balitsky, Phys. Lett. B 124 (1983) 230.
A.D. Martin, M.G. Ryskin, Phys. Rev. D 57 (1998) 6692, hep-ph/9711371.
B. Pire, J. Soffer, O. Teryaev, Eur. Phys. J. C 8 (1999) 103, hep-ph/9804284.
A.V. Radyushkin, Phys. Rev. D 59 (1999) 014030, hep-ph/9805342.
P.V. Pobylitsa, Phys. Rev. D 66 (2002) 094002, hep-ph/0204337.
P.V. Pobylitsa, Phys. Rev. D 67 (2003) 034009, hep-ph/0210150.
M. Diehl, T. Feldmann, R. Jakob, P. Kroll, Nucl. Phys. B 596 (2001) 33, hep-ph/0009255.
B. Geyer, D. Robaschik, M. Bordag, J. Horeji, Z. Phys. C 26 (1985) 591.
A.V. Radyushkin, Phys. Rev. D 56 (1997) 5524, hep-ph/9704207.
A.V. Belitsky, D. Mller, A. Kirchner, A. Schfer, Phys. Rev. D 64 (2001) 116002, hep-ph/0011314.
M.V. Polyakov, C. Weiss, Phys. Rev. D 60 (1999) 114017, hep-ph/9902451.
O.V. Teryaev, Phys. Lett. B 510 (2001) 125, hep-ph/0102303.
V.M. Braun, G.P. Korchemsky, D. Mller, Prog. Part. Nucl. Phys. 51 (2003) 312, hep-ph/0306057.
Y.L. Luke, The Special Functions and their Approximations, vol. 2, Academic Press, New York, 1969.
R. Crewther, Phys. Lett. B 397 (1997) 137, hep-ph/9701321.
A.V. Belitsky, D. Mller, Phys. Lett. B 417 (1997) 129, hep-ph/9709379.
A.V. Belitsky, D. Mller, Nucl. Phys. B 527 (1998) 207, hep-ph/9802411.
A.V. Belitsky, D. Mller, Nucl. Phys. B 537 (1999) 397, hep-ph/9804379.
S.J. Brodsky, G.T. Gabadadze, A.L. Kataev, H.J. Lu, Phys. Lett. B 372 (1996) 133, hep-ph/9512367.
S. Moch, J.A.M. Vermaseren, A. Vogt, Nucl. Phys. B 688 (2004) 101, hep-ph/0403192.
A. Vogt, S. Moch, J.A.M. Vermaseren, Nucl. Phys. B 691 (2004) 129, hep-ph/0404111.
E.B. Zijlstra, W.L. van Neerven, Nucl. Phys. B 383 (1992) 525.
A.V. Belitsky, D. Mller, A. Kirchner, Nucl. Phys. B 629 (2002) 323, hep-ph/0112108.
Y.L. Luke, The Special Functions and their Approximations, vol. 1, Academic Press, New York, 1969.
P.V. Landshoff, J.C. Polkinghorne, R.D. Short, Nucl. Phys. B 28 (1971) 225.
M. Diehl, T. Feldmann, R. Jakob, P. Kroll, Eur. Phys. J. C 39 (2005) 39, hep-ph/0408173.
[73]
[74]
[75]
[76]
[77]
[78]
[79]
59
I. Balitsky, E. Kuchina, Phys. Rev. D 62 (2000) 074004, hep-ph/0002195.

M. Penttinen, M.V. Polyakov, K. Goeke, Phys. Rev. D 62 (2000) 014024, hep-ph/9909489.
P. Hgler, et al., LHPC Collaboration, SESAM Collaboration, Phys. Rev. Lett. 93 (2004) 112001, hep-lat/0312014.
J. Kogut, D. Soper, Phys. Rev. D 1 (1970) 2901.
M. Abramowitz, I. Stegun, Handbook of Mathematical Functions, Dover, New York, 1970.
L. Mankiewicz, G. Piller, Phys. Rev. D 61 (2000) 074013, hep-ph/9905287.
Scaling test of fermion actions in the Schwinger model

N. Christian a, , K. Jansen a , K. Nagai a , B. Pollakowski a,b
a John von Neumann Institute for Computing, Platanenallee 6, 15738 Zeuthen, Germany
b Institut fr Physik, Humboldt Universitt zu Berlin, Newtonstr. 15, 12489 Berlin, Germany
Received 12 October 2005; received in revised form 4 January 2006; accepted 5 January 2006
Abstract
We discuss the scaling behaviour of different fermion actions in dynamical simulations of the 2dimensional massive Schwinger model. We have chosen Wilson, hypercube, twisted mass and overlap
fermion actions. As physical observables, the pion mass and the scalar condensate are computed for the
above mentioned actions at a number of coupling values and fermion masses. We also discuss possibilities to simulate overlap fermions dynamically avoiding problems with low-lying eigenvalues of the overlap
kernel.
1. Introduction
Besides the interest in the 2-dimensional Schwinger model [1] in its own right as a quantum
field theory, it can be considered as a test laboratory for new theoretical concepts and ideas that
aim at eventual applications in more demanding situations such as lattice QCD. In particular, for
numerical simulations the lattice Schwinger model is most suitable to perform test studies since
the computations are much cheaper than in four dimensions and precise results at many parameter
values can be obtained, see, e.g., Refs. [25] for a selection of recent work in the Schwinger
model. In this paper, we want to address the scaling properties of a number of fermion actions
for Nf = 2 flavours of dynamical fermions. To this end, we will compare standard Wilson [6],
Wilson twisted mass [7,8], hypercube [9] and overlap fermions [10] in their approach to the
continuum limit. We will use throughout the paper the Wilson plaquette gauge action with a
coupling = 1/a 2 e2 , with a the lattice spacing and e the dimensionful coupling.
E-mail address: nils.christian@desy.de (N. Christian).

doi:10.1016/j.nuclphysb.2006.01.029
N. Christian et al. / Nuclear Physics B 739 (2006) 6084
61
Each of the above mentioned fermion actions has certain advantages and are used in present
simulations of lattice QCD. Understanding the scaling properties when using different lattice
actions and to check the expected scaling behaviour as a function of the lattice spacing a is
certainly one of the most important questions in lattice calculations. However, if we think of
dynamical fermion simulations in lattice QCD a scaling analysis is, at least nowadays, far too
computer time consuming to be addressed, see, e.g., Ref. [11] and Ref. [12] for recent estimates
of the simulation costs in lattice QCD. On the other hand, for the 2-dimensional Schwinger
model, such simulations are perfectly possible and give important insight into the properties of
the above mentioned actions.
In order to study the scaling behaviour, we will fix the scaling variable z (mf )2/3 , where
mf is the fermion mass in lattice units. We have chosen z = 0.2, 0.4, 0.8. The fermion mass is
determined from the PCAC relation where we employ local as well as conserved currents. At
each of these fixed values of z we compute the pseudoscalar mass and the scalar condensate and
follow its behaviour with decreasing value of the lattice spacing. Performing finally a continuum limit of our results allows us to compare to analytical predictions that are available from
approximations of the massive Schwinger model which cannot be solved exactly.
The paper is organized as follows. In Section 2 we give the definition of the Wilson, hypercube, maximally twisted mass and overlap fermion operators which we have employed in
this work. Section 3 is devoted to the observables we have used and provides a description of
the numerical simulations, in particular our attempts to simulate overlap fermions dynamically.
Section 4 contains our results and Section 5 the conclusions.
2. Lattice fermions
In this section we give the definitions of the different kind of fermions we have used, i.e.,
Wilson, hypercube, Wilson twisted mass and overlap fermions. Throughout the work we have
employed only the Wilson plaquette gauge action with gauge fields Un, U (1), n denoting a
lattice point and = 1, 2 a direction of the 2-dimensional lattice
SG [U ] =
1

UP + UP
2
(1)
with UP standing for the plaquette
UP = Un,1 Un+1,2
U
n+2,1
Un,2
,
(2)
where 1 and 2 denote shifts in direction 1 and 2 respectively. The coupling multiplying the plaquette gauge action is = 1/g 2 . Denoting by e the physical coupling, dimensions are introduced
by = 1/a 2 e2 . We restrict ourselves to Nf = 2 flavours of dynamical fermions.
2.1. Wilson fermions
As a first action that can serve as a kind of benchmark action, we have chosen standard Wilson
fermions [6] with the Wilson operator DW given by
n,m
DW
= (m0 + 2r)n,m

1
(r )Un, n,m + (r + )Um,

n,m+ .
2
(3)
62
Table 1
0
1
2
(1)
(1)
+1.5 + mhyp
0.25
0.125
+0.334
+0.083
Here the sum goes over the two directions = 1 and = 2 and we use the standard Paulimatrices , = 1, 2, 3 with 3 = diag(1, 1), n, m denote lattice points with space and time
coordinates (x, t) and we suppress the Dirac indices. The Wilson parameter r is chosen to be
r = 1 throughout the paper. The Wilson action is expected to lead to large discretization errors
in physical observables linear in the lattice spacing a and hence approaches the continuum limit
rather slowly. We will compare results obtained with the Wilson action to corresponding results
from actions where these O(a) lattice artefacts are expected to vanish and changed to an O(a 2 )
behaviour.
2.2. Hypercube fermions
Perfect actions [9] are completely free of lattice artefacts. However, since such actions cannot
be realized in practice, truncated versions are used which are expected to inherit many of the
properties of the perfect actions. For this work, we have chosen a particular ansatz, the hypercube
action [13]. In this approach the interaction among fields on the lattice is extended from a purely
nearest neighbour interaction to a hypercube of the lattice.
A general ansatz for the hypercube operator is
n,m
Dhyp
=
mn + mn 1,
(4)
where mn and mn are real parameters, the values of which have to be optimized according to some criterion, e.g., by demanding an improved continuum limit behaviour of physical
observables.
Furthermore, symmetry requirements of the action leads, in the case of the Schwinger model,
to a reduction of these couplings to only 5 free parameters that will depend on the gauge coupling
constant g and the bare fermion mass mhyp . The values of and are determined for free
fermions, setting g = 0 and then they are taken over to the interacting case. As an alternative, the
values of and may be determined by requiring that the violations of the GinspargWilson
relation are minimized. In this work we will, however, only work with the scaling optimized
set of parameters as given in Ref. [14] since we are interested mainly in the scaling behaviour.
We give in Table 1 the values of the parameters that we have used in this work.
The full hypercube Dirac operator is given by
Dhyp = 0 + H (1) + H (2) ,
2
(1) nm

H
Un, n,m 1 + (1) + Un
=
n,m+ 1 (1) ,
,
=1
63

(2) nm 1
(2)
= (Un,1 Un+1,2
H
+ Un,2 Un+2,1
)n,m1
2 2 + (1 + 2 )
2

(2)
n,m1+
+ Un,1 U + U Un2,1
2 2 + (1 2 )
n+12,2
n2,2

(2)
n,m+1
+ U Un1,2
+ Un,2 Un+2
2 2 + (1 + 2 )
1,1
n1,1

(2)
.
+ U U + U U n,m+1+
2 2 + (1 2 )
n1,1
n12,2
n2,2
n12,1
(5)
2.3. Twisted mass fermions
The twisted mass formulation of lattice fermions has been introduced originally to regulate
small, unphysical eigenvalues of the Wilson lattice Dirac operator [7]. In order to keep an O(a)
improvement, the twisted mass setup has been first developed in the O(a) Symanzik improved
theory. However, it then was realized that by a careful tuning of the parameters of the Wilson
twisted mass action, an automatic, full O(a)-improvement can be reached, leading to lattice
discretization errors that appear only in O(a 2 ) and hence allow for a much accelerated continuum
limit [8]. The Wilson twisted mass formulation has received a lot of attention recently and a
number of tests in the quenched [1519] and partly also for full dynamical fermions [2023] has
been performed.
In order to introduce twisted mass fermions, let us start with the continuum Euclidean action

D + m + if 3 3 ),
] = d 2 x (
S[,
(6)
where 3 is the third Pauli matrix, acting in flavour space and f represents the twisted mass
parameter. The transformation
3
= ei3 2 ,
i3
= e
3
2
(7)
leaves this form of the action invariant with rotated mass parameters,
m = m cos + f sin ,
f = m sin + f cos ,
(8)
with the rotation (twist) angle ,

f
.
tan =
(9)
m
On the lattice, the Wilson term is not invariant under the rotation in Eq. (7) and the twisted
mass operator takes the form
0
Dtm = DW
+ [m0 + if 3 3 ],
where
0
DW
(10)
denotes the Wilson operator without the mass term, i.e.,
0
DW
= DW [m0 = 0].
(11)
It can be shown that the twisted mass action leads to an O(a)-improvement when the angle
= /2. It goes beyond the scope of this paper to provide the arguments for this remarkable
result the derivation of which can be found in Ref. [8]. We only would like to remark that in
64
order to obtain this value of the twist angle is equivalent to tune the bare fermion mass parameter
m0 to a critical value mcrit
0 .
One very important aspect of twisted mass fermions is that there is a particular definition
of mcrit
0 from the vanishing of the PCAC fermion mass that does not only lead to an O(a)improvement [8,16,17,24,25] but also substantially reduces cut-off effects that appear in O(a 2 )
as has been demonstrated in [18,26]. In this paper, we will use a value of mcrit
0 that was obtained
in the pure Wilson fermion theory without twisted mass term from the PCAC relation. The value
of mcrit
0 was tuned in such a way that the corresponding PCAC fermion mass vanishes. After
having determined mcrit
0 in this way, we varied the twisted mass parameter f to realize the
fermion and pion masses we are interested in.
2.4. Overlap fermions
A lattice Dirac operator DGW that satisfies the GinspargWilson relation [27]
DGW 5 + 5 DGW = 2aDGW 5 RDGW ,
(12)
where R is a local term, leads to an action that has an exact (lattice) chiral symmetry eliminating
thus automatically O(a) cut-off effects. The realization of an operator DGW that we use here is
n,m
the overlap fermion, which is characterized by the overlap Dirac operator [10]. For Rn,m = 2
it takes the form

mov a
(0)
Dov = 1
(13)
Dov
+ mov ,
2

1
(0)
1 + D0 / D0 D0 ,
Dov
(14)
=
a
where D0 is the so-called overlap kernel operator and a a/. For the kernel D0 , there is a
large choice. In the following, however, we will only use the hypercube operator, i.e., D0 = Dhyp
with fixed 0 = 0.5, i.e., setting mhyp = 1. This corresponds to setting the parameter = 1 in
Eq. (14) and guarantees the locality of the overlap operator [28].
We remark that we have realized the square root operator 1/ D0 D0 in Eq. (14) by Chebyshev polynomials Pn, with degree n and a lower bound . This Chebyshev polynomial shows
an exponential convergence rate in the interval [, 1]. Setting to the lowest eigenvalue of the
overlap kernel operator D0 D0 which is normalized to one, we always have chosen the degree n
such that we reach machine precision for the evaluation of 1/ D0 D0 . We used eigenvalue deflation and projected out a number of low-lying eigenvalues. In this case, was set to the lowest
non-projected eigenvalue of D0 D0 .
3. Observables and simulations
In this section we give the operators that we have used to determine the physical observables
we have computed and describe the numerical simulations. In particular, we discuss some of
our attempts to perform dynamical overlap simulations avoiding problems with very low-lying
eigenvalues of the overlap kernel operator.
65
3.1. Observables
2-point functions
Generally, the bi-linear operators are given by
O n = n n ,
(15)
where stands for certain combinations of the Pauli matrices acting in Dirac-space, while the
Pauli matrices act in flavour space. We will consider the following operators,
(Dirac) (flavour)
P na = n 3
a
n ,
Sn = n n ,
(17)
(Dirac) (Dirac) a(flavour) n ,

A a
n = n 3
(18)
V na
(19)
= n (Dirac) a(flavour) n ,
(16)
with P the pseudoscalar, S the scalar, A the axial-vector and V the vector operator.
For the twisted mass fermions, the operators take a modified form as can be obtained from the
field transformation according to Eq. (7), leading to
a(tb)
a(ph)
P n ,
if a = 1, 2,
=
Pn
(20)
3(tb)
(tb)
cos P n + sin Sn ,
if a = 3,

a(tb)
b(tb)
cos A n
+ sin ab V n
, if a = 1, 2,
a(ph)
=
An
(21)
3(tb)
A n
,
if a = 3,

a(tb)
b(tb)
cos V n
+ sin ab A n
, if a = 1, 2,
a(ph)
V n
(22)
=
3(ph)
Vn
,
if a = 3.
Here ph denotes the physical and tb the twisted basis. Note that in the special case = /2
for a = 1, 2, V and A just interchange their role while P remains invariant. In particular, the
scalar operator is given by the 3rd component of the pseudoscalar operator.
For the Wilson, hypercube and Wilson twisted mass fermions we computed the 2-point correlation functions by standard techniques using a conjugate gradient solver. For overlap fermions
we calculated the correlators both from using a conjugate gradient solver and from eigenvectors i and eigenvalues i of the overlap operator in Eq. (13). A generic non-singlet correlator
C(n, m) is then expressed in terms of the fermion propagator S(n, m) and a suitable Dirac structure , corresponding to the operators listed above, as

C(n, m) = tr S(n, m) S(m, n)
1

j (n) i (n) i (m) j (m) .

=
(23)
i j
i ,j
PCAC fermion mass

The bare mass parameter for Wilson (m0 ) and hypercube (mhyp ) fermions receives an additive
mass renormalization, necessitating the determination of a critical fermion mass mcrit which we
computed by the vanishing of the PCAC fermion mass, extracted from the PCAC relation

O .
A (n)O = 2m P(n)
(24)
66
Choosing O = P and projecting to zero momentum (denoting by A and P the currents summed
over space), we arrive at
mf (t) =
2 A2t P0
.
2Pt P0
(25)
For the currents that appear in Eq. (25) one may use the local currents of Eqs. (16)(19). An
alternative is to employ also conserved currents. A very general way to derive the conserved
currents is given in [29]. The vector and axial currents are then given by
V na = JnR + JnL ,
R
L
A a
n = Jn Jn ,
(26)
(27)
where
1
3 )K,n (1 + 3 ),
JnR = (1
2
1
+ 3 )K,n (1 3 )
JnL = (1
2
and
K
,n

D(U () )
= i
,
n, =0
()
Un,
= ein, Un, ,
(28)
(29)
(30)
()
where D(U ) denotes the lattice Dirac operator used. This method of constructing the conserved currents is very useful when complicated lattice Dirac operators are considered since their
construction from the current conservation condition can be very cumbersome. We followed this
prescription to compute the conserved currents, only for the overlap fermions we used the local
point currents. We remark that for the twisted mass and the overlap case in principle also the bare
fermion masses, f and mov can be used. However, since this could lead to very different lattice
artefacts, we employed also in these cases the fermion mass derived from the PCAC relation for
all physical results presented in the following in order to be able to directly compare the different
fermion actions.
Scalar condensate
Besides the 2-point functions which will provide the pseudoscalar mass, we considered also
the scalar condensate, .

A first method to compute is by calculating Tr D 1 using
Gaussian noise sources. We will denote the so computed values of as direct . This quantity
develops in the case of Wilson and hypercube fermions a divergent piece 1/a. A second way,
which avoids the appearance of the divergent piece from the beginning, is to use the integrated
axial Ward identity leading to a subtracted scalar condensate, sub [30]. We remark that in the
(0)
a Dov
case of overlap fermions we computed from the improved scalar operator (1
)
2
which we evaluated from the eigenvalues of the overlap operator. More precise definitions and a
further discussion will be provided in Section 4.
3.2. Simulations
In our work, we have used for the Wilson, the hypercube and the twisted mass fermion action a standard Hybrid Monte Carlo algorithm (HMC) [31]. In the case of overlap fermions
67
a straightforward implementation of the HMC algorithm is not suitable since the variation of
the Hamiltonian with respect to the gauge field can lead to very large forces proportional to
(D0 D0 )3/2 when the overlap kernel operator D0 develops exceptionally small eigenvalues. See
Refs. [3234] for variants of the HMC algorithm that circumvent this difficulty.
For the overlap fermion results we used configurations that were generated in the pure gauge
theory only and performed a reweighting with the overlap fermion determinant to obtain physical
observables for dynamical Nf = 2 flavours of fermions. However, we also tried several possibilities to avoid the problem with low-lying eigenvalues of the overlap kernel operator D0 by
approx
that is a good approximation to Dov but which is safe
replacing Dov by some operator Dov
against these low-lying eigenvalues. The simulation can then be made exact again by adding a
approx
correction step employing the ratio Dov /Dov .
The general idea is to write (we use for simplicity only a single operator here, in practice one
would have to use the operator D D, of course)

Dov
approx
det
Rdet .
approx det Dov
Dov

approx
det Dov = det Dov
(31)
approx
While det Dov

would be used in the HMC algorithm, the remaining determinant ratio Rdet
could be implemented either in an additional accept/reject step or it could be included as a
reweighting factor in the computation of a given observable.
approx
can be found such that
A crucial question in such an approach is, whether an operator Dov
the fluctuations in Rdet are small enough to obtain statistically significant results. We decided
therefore to test this idea by computing Rdet stochastically using n = 10 Gaussian noise vectors
on a number of gauge field configurations generated in the pure gauge theory at = 3 on a 162
approx
lattice. In the following we will describe the results of these tests for three choices of Dov .
approx
= Dhyp
Case of Dov
As a first trial, we used the hypercube operator Dhyp as an approximation to Dov . This choice
has been motivated by the fact that the operator Dhyp is constructed to be approximating the
GinspargWilson relation, resulting in a very similiar eigenvalue spectrum for both operators
[14]. Note also that we have used the hypercube operator itself as an overlap kernel operator D0 .
(0)
Let us remark that we re-write the massive overlap operator as Dov + mov /(1 mov /2) in order
to better match the eigenvalue spectra of both operators.
We computed Rdet as a function of the hypercube bare fermion mass mhyp in a range 0.25 <
mhyp < 0.25. However, we found that the fluctuations of Rdet were extremely large for all values
of mhyp we have tested. Thus we had to conclude that Dhyp cannot serve as an infrared safe,
approximate operator for the overlap simulations when Rdet is used in a stochastic correction
step. We cannot exclude, of course, that by using an improved hypercube operator, e.g., by adding
the clover term or performing smearing of the link variables, the situation could be improved.
However, given the negative findings of our investigation, we did not pursue this direction. What
we tried instead, is to use Dhyp as the guidance Hamiltonian in the molecular dynamics part
of the Hybrid Monte Carlo algorithm while keeping Dov as the exact overlap operator for the
accept/reject Hamiltonian. Amazingly, despite the negative results described here for Rdet , we
found that this led to reasonable acceptance rates, at least for not too large systems, see Ref. [35]
for a further discussion of this point.
68
Modified Chebyshev polynomial

As a second attempt, we tried to modify the range of the Chebyshev polynomial employed in
the construction of the overlap operator. There, the square root of the kernel is computed by a
Chebyshev polynomial Pn, of degree n in the interval [, 1],

Pn, D0 D0 1 D0 D0 .
(32)
We have chosen the approximation accuracy in Eq. (32) to be very high, compatible with machine
(64-bit) precision. The value of is chosen to correspond to the lowest eigenvalue of D0 D0 .
In principle, it is possible to use instead of the exact polynomial with parameters n and a
modified polynomial with parameters n < n and > . Clearly, if could be chosen to be well
above the smallest eigenvalues of D0 D0 then using an overlap operator with a polynomial Pn,

in the molecular dynamics part would lead to an infrared safe simulation.
As a start situation, we had chosen = and reduced only the degree of the polynomial
successively. We found that it is indeed possible to reduce n substantially to n 1/3 n without
having large fluctuations in the determinant ratio Rdet . This is a very positive outcome since it
gives rise to a substantial acceleration of simulations with the overlap operator. However, when
we tried to change the value of only slightly, we observed immediately large fluctuations in
Rdet . For = 1.1 the fluctuations in Rdet became already so large that there is no chance to
obtain a statistically significant result in such a setup.
Explicit infrared regularization
As a third (and last) attempt, we tried to use an approximate overlap operator that has an
infrared regulator built in. In particular, we studied the situation where we use a modified sign
function

D0 D0 D0
D0 D0 + .
D0
(33)
The (optimistic) expectation here is that by choosing large enough, without having too large
fluctuations in the determinant ratio Rdet , the overlap operator with such a modified sign function
would be infrared safe for the HMC simulation.
Such an optimistic expectation is not completely unfounded when one inspects the eigenvalues
approx
of the exact overlap operator Dov and the sign function modified one, Dov . Indeed, we show
in Fig. 1 the eigenvalues when choosing = 0, 0.0001, 0.001, 0.01 for a typical configuration.
As can be seen, all the eigenvalues lie on the expected circle and a difference for various
choices of is not visible. In Fig. 2 we show the difference of the eigenvalues between the
spectra using = 0 and = 0. In building the difference of the eigenvalues we used an ordering
of the eigenvalues with respect to their real part. The scale in the plot is chosen for the situation
of = 0.01. We have rescaled the difference in the eigenvalues by a factor of 30 for = 0.0001
and 7 for = 0.001 in order to plot all difference spectra in one common plot. It appears that the
difference spectra do not build a perfect circle shape but are rather lemon shaped. Nevertheless
the lemon distortion of the circle happens at a rather small scale of O(102 ) for the case of
= 0.01 while for smaller values of the difference becomes even smaller. Note that the smallest
eigenvalue is min (D0 D0 ) = 0.24 for this configuration.
The smallness of the distortions in the spectrum as observed in Fig. 2 appears to be quite
promising for this choice of an approximate overlap operator. However, the effects on the fluctuations are considerable as can be seen in Fig. 3. For a value of = 0.0001 (circles) the determinant
ratio Rdet 1 and has very small fluctuations until the degree of the polynomial is lowered to
69
Fig. 1. The eigenvalues of the exact overlap operator and a modified overlap operator when the sign function is
changed according to Eq. (33). The plot contains the eigenvalues for = 0 (the original overlap operator) and for
= 0.0001, 0.001, 0.01. For all these cases, the eigenvalues are so close to each other that they cannot be distinguished
in the graph.
n = 8 which is a factor of three smaller than the original degree of the polynomial. However,
already for = 0.001 (upward triangles) Rdet starts to develop some fluctuations and finally, for
= 0.01 (downward triangles) the fluctuations are becoming so large that realistic simulations
with such a value of cannot be performed.
As a conclusion, we found that none of the above described attempts led to a satisfactory
solution for an infrared safe dynamical overlap simulation. As said above, we therefore have
performed the overlap simulations finally by generating pure gauge configuration and using a
reweighting with the determinant, computed exactly from the eigenvalues.
Summary of simulation setup
In Table 2 we shortly summarize which technique we have used for the different lattice fermions for the simulations and for the observables.
Let us end this section by a small technical remark. For the computations of the eigenvalues
and eigenvectors we used the LAPACK [36] routine. In the course of our work we found that
this routine does not compute correctly the eigenvectors in the case of degenerate zero modes. It
rather gives linear combinations of the exact solutions which led to a problem in the computation
of the pseudoscalar correlator using the operator P if configurations with topological charge
|Q| > 1 are considered. The effect shows up in such a way that the pseudoscalar correlator
Eq. (16) develops a plateau like behaviour for values of Euclidean time t close to the middle of
70
Fig. 2. The difference spectra of the overlap operator with = 0 and the operators with = 0.01 (most outer spectrum),
= 0.001 (middle spectrum) and = 0.0001 (most inner spectrum). The difference spectra for = 0.001 and = 0.0001
have been scaled with a factor 7 and 30, respectively, in order to be able to include them in the plot.
Fig. 3. The determinant ratio Rdet as a function of the degree n of the Chebyshev polynomial. We plot the cases for
= 0.0001 (circles), = 0.001 (upward triangles) and = 0.01 (downward triangles).
the lattice. This behaviour made it practically impossible to extract the pseudoscalar mass and the
PCAC fermion mass. By re-diagonalizing the zero mode sector we could verify that this effect
goes away.
As a simple solution for our simulations we decided to always take the correlator P(t)P(0)
S(t)S(0) in which this zero mode contribution is cancelled out [37]. The only exception is
the case z = 0.8, where the pseudoscalar masses are so high that it became very difficult to
71
Table 2
Wilson/tm/hypercube
Overlap
simulations
operators
standard HMC
conjugate gradient solver
currents
pseudoscalar correlator
scalar condensate
local/conserved
PP
WardTakahashi-identity
determinant reweighting
eigenvalues & eigenvectors/
conjugate gradient solver
local
PP SS
a D)
(1
2
disentangle them from the scalar masses when the S(t)S(0) was subtracted of. Therefore we
used for this large value of z the conjugate gradient solver to compute the correlators.
4. Results
It is the main aim of this paper to check the scaling behaviour of the fermion actions described in
Section 2. To this end, we have chosen the simulation parameters such that we fix the scaling variable

z (mf /g)2/3 = (mf )2/3 .
(34)
We performed simulations for a wide range of -values, 0.1 6. For Wilson, hypercube and twisted
mass fermions our lattices were mainly of size V = L2 = 322 . Only at = 5, 6 and z = 0.2 we went up to
V = 482 lattices. For the overlap fermion simulations we mainly used V = 202 and, for z = 0.2, V = 242
lattices. Only at = 5 and z = 0.2 we used a V = 282 lattice. It is our experience that for larger lattices
the determinant reweighting technique used for the overlap fermion simulations are not practical since they
lead to large fluctuations which spoil the signal to noise ratio. We finally mention that all our error analyzes
are based on the method described in Ref. [38] including thus the autocorrelation times in the computation
of the errors.
As a prime quantity we will test the scaling behaviour of m , performing finally a continuum limit
for all actions used. In the continuum, there are two approximate calculations for mcont
/e for the massive
Schwinger model.
The first is performed by Smilga [39] who finds for strong coupling and small fermion mass

(3/4) 2/3 (1/6) m 2/3
mcont
25/6 e /3
(35)
e
(1/4)
(2/3) e
2/3
m
= 2.008
.
(36)
e
For large masses, Gattringer in Ref. [40] finds from a semi-classical analysis

25/6 m 2/3
mcont
e2 /3 1/6
e
e
2/3
m
= 2.163
.
e
(37)
(38)
Note that we give these results in terms of the dimensionful quantities, coupling e, fermion mass m and
pseudoscalar mass mcont
. Both expression are approximate computations and it is interesting to compare
these against our non-perturbative calculations.
For completeness, we also give here analytical expressions for the scalar condensate as available in the
literature. The first is again from Ref. [39],
1/3
m
cont
0.388
.
(39)
e
e
72
Table 3
hyp
, and hypercube, mcrit ,
Critical hopping parameters at the values of used in our simulations for the Wilson, mWilson
crit
fermions
hyp
mWilson
crit
mcrit
1.0
2.0
3.0
4.0
5.0
6.0
0.3204(7)
0.1968(9)
0.1351(2)
0.1033(1)
0.0840(1)
0.0719(1)
0.335(1)
0.203(1)
0.1392(4)
0.1050(2)
0.0856(1)
0.0727(1)
A second expression is derived in Ref. [41],

2
1 (mcont
cont
/e)
=
.
e
4 (m/e)
(40)
Critical fermion mass

For Wilson fermions and for hypercube fermions, the fermion mass receives an additive renormalization.
Therefore, the critical values of the bare fermion mass, where, e.g., the PCAC fermion mass vanishes has
to be determined by non-perturbative simulations. We computed mPCAC , extracted from Eq. (25) using the
conserved currents of Eqs. (26), (27) as a function of the bare fermion masses, m0 for Wilson and mhyp for
hypercube fermions. We determined at each value of those values of m0 and mhyp where mPCAC = 0.
The value of these bare fermion masses can be found in Table 3. Note that the actual critical mass values
used for the twisted mass simulations in order to realize full twist, though compatible, differ slightly from
the ones in Table 3 since they were obtained with less statistics.
Fermion mass dependence of the pion mass
As described above, in the scaling analysis we will fix the scaling variable z = (mf )2/3 . To this end,
we first explored the dependence of the pion mass as a function of the PCAC fermion mass for fixed values
of . We give one example for the case of hypercube fermions in Fig. 4 using the conserved currents of
Eqs. (26), (27) to compute the PCAC fermion mass. We also plot in the graph the analytic predictions of
Eqs. (35), (37) for the fermion mass dependence of the pion mass.
The data in Fig. 4 are for = 0.1, = 0.5 and = 1.0. Clearly, for = 0.1 the data do not follow the
theoretical expectation while for the other values of there seems to be some agreement with the analytical
formulae. However, in order to make more definite statements, a closer look to the scaling behaviour is
clearly needed. Anyhow, the main purpose of Fig. 4 is to show that we have collected data that are concentrated around the values of z = 0.2, z = 0.4 and z = 0.8 that we are interested in for our final scaling
analysis. When the values of z did not coincide directly with the desired ones, we performed an only very
small linear interpolation of our data to achieve the exact values of z.
Finite size effects

One source of a systematic error in the determination of m are possible finite size effects. For two
dimensions the asymptotic finite size corrections for the pseudoscalar mass were computed in Ref. [42] and
studied numerically in the case of the Schwinger model in Ref. [4]. A very good agreement to the theoretical
prediction was found and it was observed that the variable m L needs to be surprisingly large to suppress
finite size corrections.
73
Fig. 4. Pseudoscalar mass m versus z = (mf )2/3 for three values of . The solid lines are the analytical predictions of Eqs. (35), (37). Crosses represent our data for = 1, triangles for = 0.5 and diamonds for = 0.1.
Taking this as a warning, we followed the procedure of Ref. [4] and performed a finite size correction of
our values for the pseudoscalar masses when necessary. We used the analytical formula

m
+
A
m (L) = m
(41)
eLm ,
where m
denotes the pseudoscalar mass in the infinite volume limit. Rescaling Eq. (41) by this
formula can be applied to all our finite volume data at various, large enough values of to keep the effects
of the lattice spacing small. Inspecting our data, we decided to perform a global fit to our simulation data
obtained with Wilson, hypercube and Wilson twisted mass fermions at z = 0.2 and z = 0.4 for 4.
As a result we found a universal constant A = 12.4(6) which is compatible with the value computed in
Ref. [4]. In Fig. 5 we give an example for the resulting fit for a subset of our data in the case of Wilson
fermions.
We then examined all pseudoscalar masses that we will use in the detailed scaling analysis described
below and used Eq. (41) to analytically correct for finite size effects. We give in Tables 4, 5, 6 and 7 the
infinite volume values for the pseudoscalar mass as obtained from this finite size correction. Note that in
most of the cases, this correction is negligibly small and stays anyhow on the percent level.
4.1. Results for the pseudoscalar mass

In Tables 4, 5, 6 and 7 we give our results for the pseudoscalar masses at fixed scaling variable z of
Eq. (34), where z is fixed either by using local or conserved currents to extract the PCAC fermion mass.
While in the tables we give only the results for 1, we have performed simulations
also for smaller
values of for all actions, except overlap fermions. We show the result for m as a function of in
Fig. 6. The graph is done for the example of z = 0.4 where we used the conserved currents to determine
the PCAC fermion mass. Clearly, large lattice cut-off effects are observed when small values of < 1 are
taken. From the figure it is obvious that for an asymptotic scaling analysis values of 1 should be taken.
In Figs. 7, 8 and 9 we show the results of our scaling test for fixed z = 0.8, z = 0.4 and z = 0.2 respectively. For the values of = 1 and = 2 at z = 0.2, the determinant used for reweighting the overlap
results induced large fluctuations such that no reliable values of physical observables could be extracted.
For all figuresthe value of z was determined using local currents to compute the PCAC fermion mass.
We show m as a function of 1/ a 2 . We first performed linear fits in 1/
independently for each
lattice fermion used. Since these fits gave consistent continuum values for m for all actions, see the
74
Fig. 5. We show m as a function of L at the example of Wilson fermions and three values of (downward triangles:
= 4, circles: = 5, upward triangles: = 6). The fit curves, represented by solid lines, are obtained by performing
a common fit to Eq. (41) to our data for z = 0.2 and z = 0.4 and for Wilson, hypercube and twisted mass fermions,
choosing always 4.
Table 4
Results for the Wilson fermion simulations. The scaling variable (mf )2/3 is fixed by using either the local or the
conserved currents to determine the PCAC fermion mass resulting in the two different values of the pseudoscalar mass.
Typical statistics of the runs were between 4000 and 10000 configurations
Wilson fermion results
z
mlocal
0
mcons
0
mlocal
,
mcons
,
cons
sub
0.2
0.2
0.2
0.2
0.2
0.2
0.4
0.4
0.4
0.4
0.4
0.4
0.8
0.8
0.8
0.8
0.8
0.8
1.0
2.0
3.0
4.0
5.0
6.0
1.0
2.0
3.0
4.0
5.0
6.0
1.0
2.0
3.0
4.0
5.0
6.0
0.231367
0.132316
0.082626
0.057374
0.043199
0.034249
0.075308
0.017495
0.014505
0.026358
0.032526
0.036146
0.352443
0.314017
0.298677
0.276568
0.255009
0.239625
0.230193
0.131842
0.081935
0.056979
0.043010
0.034009
0.059286
0.008352
0.019789
0.031078
0.035391
0.037693
0.537801
0.403487
0.350958
0.314133
0.285628
0.262974
32
32
32
32
48
48
32
32
32
32
32
32
32
32
32
32
32
32
0.388(5)
0.402(3)
0.404(4)
0.406(2)
0.403(2)
0.404(2)
0.75(1)
0.783(8)
0.799(7)
0.80(1)
0.808(4)
0.811(3)
1.36(1)
1.52(1)
1.60(1)
1.63(1)
1.65(1)
1.686(6)
0.391(3)
0.406(2)
0.408(2)
0.408(2)
0.404(1)
0.406(2)
0.779(5)
0.813(6)
0.819(3)
0.820(3)
0.824(2)
0.820(2)
1.560(5)
1.690(5)
1.729(4)
1.745(5)
1.758(3)
1.770(3)
0.277(7)
0.252(6)
0.232(7)
0.231(8)
0.219(6)
0.225(4)
0.339(6)
0.339(3)
0.334(4)
0.340(3)
0.345(2)
0.345(2)
0.329(1)
0.400(1)
0.438(1)
0.469(1)
0.494(1)
0.518(1)
example for z = 0.4 in Ref. [35],

we finally performed constraint fits, demanding that all actions give the
same continuum value for m at fixed value of z. We show these constraint fits in Figs. 7, 8 and 9 as
the solid lines. The data for all kind of fermions, Wilson, hypercube, maximally twisted mass and overlap,
are nicely consistent with a linear behaviour in a 2 . While this is expected for maximally twisted mass and
75
Table 5
Results for the hypercube fermion simulations. The scaling variable (mf )2/3 is fixed by using either the local or the
conserved currents to determine the PCAC fermion mass resulting in the two different values of the pseudoscalar mass.
Typical statistics of the runs were between 4000 and 10000 configurations
Hypercube fermion results
z
mlocal
0
mcons
0
mlocal
,
mcons
,
cons
sub
0.2
0.2
0.2
0.2
0.2
0.2
0.4
0.4
0.4
0.4
0.4
0.4
0.8
0.8
0.8
0.8
0.8
0.8
1.0
2.0
3.0
4.0
5.0
6.0
1.0
2.0
3.0
4.0
5.0
6.0
1.0
2.0
3.0
4.0
5.0
6.0
0.237698
0.137038
0.085462
0.059364
0.044387
0.035528
0.083223
0.025253
0.012877
0.025787
0.031525
0.034082
0.311570
0.309020
0.293918
0.272609
0.255540
0.237752
0.231044
0.133369
0.083199
0.058095
0.043885
0.034893
0.045098
0.004337
0.021147
0.031202
0.035539
0.037392
0.613589
0.433230
0.365237
0.321868
0.290832
0.266771
32
32
32
32
48
32
32
32
32
32
32
32
32
32
32
32
32
32
0.39(1)
0.400(3)
0.400(7)
0.399(9)
0.408(2)
0.413(7)
0.78(2)
0.78(1)
0.806(6)
0.809(4)
0.812(3)
0.816(3)
1.39(2)
1.55(2)
1.61(1)
1.65(1)
1.672(7)
1.68(1)
0.416(3)
0.416(3)
0.415(2)
0.416(2)
0.410(2)
0.411(7)
0.844(9)
0.846(4)
0.836(3)
0.834(4)
0.831(2)
0.835(2)
1.742(6)
1.792(6)
1.793(4)
1.792(5)
1.792(4)
1.796(5)
0.290(6)
0.264(7)
0.250(6)
0.240(5)
0.246(8)
0.247(4)
0.375(2)
0.381(2)
0.382(2)
0.385(2)
0.392(3)
0.392(3)
0.366(1)
0.466(1)
0.517(1)
0.557(1)
0.588(1)
0.616(1)
Table 6
Results for the Wilson maximally twisted mass fermion simulations. The scaling variable (mf )2/3 is fixed by using
either the local or the conserved currents to determine the PCAC fermion mass resulting in the two different values of
the pseudoscalar mass. Typical statistics of the runs were between 4000 and 10000 configurations
Twisted mass fermion results
z
local
f
cons
f
mlocal
,
mcons
,
cons
sub
0.2
0.2
0.2
0.2
0.2
0.2
0.4
0.4
0.4
0.4
0.4
0.4
0.8
0.8
0.8
0.8
0.8
0.8
1.0
2.0
3.0
4.0
5.0
6.0
1.0
2.0
3.0
4.0
5.0
6.0
1.0
2.0
3.0
4.0
5.0
6.0
0.050499
0.045007
0.036965
0.177303
0.144653
0.125186
0.113557
0.102819
0.095373
0.471381
0.376485
0.334417
0.306509
0.282663
0.262653
0.071591
0.054990
0.046886
0.041533
0.037643
0.034651
0.208213
0.157190
0.133453
0.117956
0.106827
0.098326
0.654028
0.459309
0.384597
0.337947
0.305270
0.280421
32
32
32
32
48
48
32
32
32
32
32
32
32
32
32
32
32
32
0.408(8)
0.411(2)
0.409(2)
0.78(1)
0.81(1)
0.819(8)
0.824(4)
0.820(4)
0.821(3)
1.47(2)
1.60(1)
1.66(2)
1.695(6)
1.719(6)
1.731(6)
0.434(2)
0.428(2)
0.422(2)
0.421(1)
0.414(2)
0.414(2)
0.874(3)
0.867(3)
0.851(2)
0.846(2)
0.844(2)
0.842(2)
1.799(5)
1.856(3)
1.838(3)
1.835(3)
1.829(3)
1.833(3)
0.303(7)
0.271(10)
0.249(6)
0.226(6)
0.237(5)
0.233(7)
0.446(7)
0.412(4)
0.394(4)
0.385(5)
0.388(4)
0.393(5)
0.571(2)
0.628(2)
0.638(2)
0.659(2)
0.671(2)
0.683(2)
76
Table 7
Results for the overlap fermion simulations. The scaling variable (mf )2/3 is fixed by using the local currents to
determine the PCAC fermion mass. We give for comparison also the values of the scaling variable z when the bare
overlap fermion mass mov is used. Typical statistics of the runs were 4000 configurations
Overlap fermion results
mov
(mov )2/3
2/3
(mlocal
)
f
mlocal
,
3.0
4.0
5.0
1.0
2.0
3.0
4.0
5.0
1.0
2.0
3.0
4.0
5.0
0.0464758
0.0447214
0.0420000
0.2529822
0.1788854
0.1460595
0.1264911
0.1142685
0.5724334
0.4232769
0.3553500
0.3152474
0.2916839
0.1864
0.2000
0.2066
0.4000
0.4000
0.4000
0.4000
0.4027
0.6894
0.7103
0.7236
0.7353
0.7521
0.194(15)
0.200(8)
0.20(3)
0.4061(14)
0.4069(19)
0.399(2)
0.383(2)
0.396(5)
0.797(2)
0.8047(5)
0.7934(2)
0.7916(1)
0.8010(2)
24
24
28
20
20
20
20
24
20
20
20
20
20
0.37(4)
0.42(4)
0.46(4)
0.807(16)
0.824(14)
0.851(12)
0.833(10)
0.837(10)
1.433(10)
1.568(4)
1.604(3)
1.638(2)
1.682(2)
0.176(4)
0.177(4)
0.138(2)
0.1950(16)
0.2340(12)
0.2518(7)
0.2618(7)
0.2415(8)
0.2363(3)
0.2905(3)
0.3270(3)
0.3549(3)
0.3811(6)
Fig. 6. We show m as a function of at fixed value of z = 0.4 as determined from the PCAC fermion mass
employing conserved currents. We represent results from Wilson fermions by circles, from twisted mass fermions by
downward and for hypercube fermions by upward triangles. The arrows represent the theoretical prediction, the upper
arrow is from Eq. (37) the lower from Eq. (35).
overlap fermions, this outcome is somewhat surprising for hypercube and Wilson fermions. Note that our
finding for Wilson fermions is, however, consistent with the results in Ref. [43].
In general it is very difficult to decide which kind of lattice fermion shows the best scaling behaviour and
we cannot draw a definite conclusion here. First of all, all kind of fermions show only small lattice artefacts.
Second, if one would use the PCAC fermion mass from the conserved currents, the picture changes. We
give an example for z = 0.4 in Fig. 10 which reveals again an O(a 2 ) scaling for the here used Wilson,
maximally twisted mass and hypercube fermions. When compared to Fig. 8 where maximally twisted mass
fermions show the smallest scaling violations, in Fig. 10 hypercube fermions seem to do better.
77
Fig. 7. Scaling test of m as function of 1/ a 2 at fixed value of z = 0.8. The solid lines represent linear constraint
fits in 1/ demanding the same continuum limit value for all kind of lattice fermions used. For Wilson, hypercube and
twisted mass fermions we used data obtained at 3, while for overlap fermions data with 2 were used.
Fig. 8. Scaling test of m as function of 1/ a 2 at fixed value of z = 0.4. The solid lines are fits explained in Fig. 7.
Continuum comparison to theory

As was shown above, all kind of fermions show
a nice scaling behaviour that is linear in a 2 and give
a universal continuum limit. In Fig. 11 we show m as a function of z in the continuum and compare
to the theoretical expectations of Eq. (35) (lower line) and Eq. (37) (upper line). As an inlay we plot the
theor of our non-perturbatively obtained data extrapolated to the continuum and the
ratio Rm = mdata
/m
two theoretical predictions. Triangles represent the data divided by the corresponding value computed from
Eq. (37) while circles represent the data divided by the corresponding value computed from Eq. (35). To the
precision we could obtain in this work, the theoretical predictions do not describe the non-perturbatively
obtained simulation data satisfactory at all values of z. Only for small values of the fermion mass (z = 0.2)
there seems to be some agreement with Eq. (35) while at z = 0.8 Eq. (37) seems to hold only.
78
Fig. 9. Scaling test of m as function of 1/ a 2 at fixed value of z = 0.2. The solid lines are fits explained in Fig. 7.
Fig. 10. Scaling of m as function of 1/ using conserved currents to compute the PCAC fermion mass to fix
z = 0.4. We denote by circles Wilson, by downward triangles twisted mass and by upward triangles hypercube fermions.
The solid lines represent linear fits in 1/ for 3. The arrows represent the theoretical prediction, the upper arrow is
from Eq. (37) the lower from Eq. (35).
4.2. Scalar condensate
Another physical quantity we will consider in this work is the scalar condensate ,
for which
analytical predictions exist, see Eqs. (39) and (40). A very simple way to calculate the scalar condensate is
to compute
=
1
TrD 1 (x, x)
V x
(42)
using a stochastic method. We denote the so computed values of as direct . A severe drawback of this
definition of is that, at least in the case of Wilson and hypercube fermions, from the mixing with the
identity operator a divergent piece 1/a appears that needs to be subtracted non-perturbatively.
79
Fig. 11. Continuum behaviour of m as function of z. The lower curve represents Eq. (35), the upper curve Eq. (37).
The inlay gives the ratio Rm of our non-perturbatively obtained data and the two theoretical calculations. Triangles use
Eq. (37) and circles use Eq. (35).
3 3 , i.e.,
In the case of the twisted mass fermions at full twist, it is possible to use the operator
the 3rd component of the pseudoscalar operator (see Eq. (16)), for the calculation of the scalar condensate.
This operator does not mix with the identity operator and thus the power divergent piece does not appear.
(0)
a Dov
) which subtracts of the
Also for the overlap operator there is a definition ov (1
2
divergent piece automatically. For overlap fermions, we calculated therefore directly the scalar condensate
from the eigenvalues of Dov

1 1
ov =
(1 i (0)/2) ,
(43)
V
i
i
where i denotes an eigenvalue of the massive overlap operator and i (0) denotes an eigenvalue at zero
(0)
fermion mass, i.e., the eigenvalues of Dov .
A second way, which avoids the appearance of the divergent piece from the beginning, is to compute a
so-called subtracted scalar condensate sub using the integrated axial WardTakahashi identity [30],

P x P 0 .
sub = 2mPCAC
(44)
x
In the twisted mass case, this corresponds to the integrated PCVC relation (using the charged correlators)

3 3 sub = 2mPCVC
sub =
(45)
P x P 0 .
x
Comparison of direct and sub

To just illustrate the effect of using a direct and a subtracted definition of the scalar condensate, we plot
direct and sub for Wilson and maximally twisted mass fermions in Fig. 12.
In two dimensions, there is in general a relation between direct and sub given by
sub = direct c0 /a.
(46)
The coefficient c0 multiplying the 1/a divergence is to be determined non-perturbatively. In the upper
panel of Fig. 12, the 1/a dependence is clearly visible for direct , as for increasing values of , the values
80
Fig. 12. The condensate from the direct direct (open symbols) and the subtracted sub (full symbols) determinations.
Error bars are within the size of the symbols. Upper graph: direct and sub for Wilson fermion. We indicate the values
of where direct is obtained. Lower graph: direct and sub for twisted mass fermions. In this case, the results from
both condensates are consistent up to scaling violations.
of direct increase accordingly. It is clear from the figure that an extraction of a physical value for the scalar
condensate will be very difficult since the term c0 /a dominates the signal.
According to [30], c0 comes from the explicit breaking term of chiral symmetry for Wilson fermions.
Therefore, one can expect that for twisted mass fermions at maximal twist this term is absent and direct
behaves like sub . This is shown in the lower panel of Fig. 12. In the case of twisted mass fermions both
direct and sub are comparable (modulo scaling violations). In fact, we find a tendency that the ratio
sub /direct approaches one when is increased. We finally remark that also for the improved definition of the scalar condensate of Eq. (43) in case of overlap fermions the divergence term is automatically
subtracted of. As a result of this discussion we will calculate the scalar condensate from the subtracted
scalar condensate in the case of Wilson, hypercube and maximally twisted mass fermions while for overlap
fermions we will use the direct calculation with the improved definition.
Scaling of the scalar condensate

Let us now turn to the results of the scaling behaviour of the scalar condensate for the various fermion
actions. We show the results in Fig. 13. sub is calculated
for Wilson, hypercube and twisted mass fermion,
and direct is done for overlap fermions, at z = (mf )2/3 = 0.2, 0.4 and 0.8.
81
Fig. 13. Scaling of the scalar condensate as a function of 1/ at z = 0.2 (circles), z = 0.4 (upper triangles) and z = 0.8
(downward triangle). The solid lines represent the fits using Eq. (48) for 2.
The figures show a strong dependence of the scalar condensate on 1/, in particular for the heavy
fermion region. This behaviour can be explained by the fact that in two dimensions, the scalar condensate
develops a logarithmic divergence in when the lattice spacing is sent to zero.1 This is most easily seen in
the free theory where a simple computation of the scalar condensate leads to a behaviour (see also Ref. [5])

m
1/a 2 + m2
+ (m/e)2
m
free
log
log
=
,
(47)
e
e
e
m2
(m/e)2
where we have inserted = 1/a 2 e2 and m denotes the continuum fermion mass. Fixing m/e, as we do
in this work, and approaching the continuum limit by letting , a logarithmic increase of the scalar
condensate in will appear. Fig. 13 shows clearly such an behaviour for all actions we have employed.
It is therefore natural to use a fit ansatz of the form
=A+
1
B
+ C log .
(48)
In Fig. 13 we show also the fit to the data using Eq. (48). For z = 0.2 we could not extract reliable values
for at = 1, 2 in the case of overlap fermions since the determinant induced very large fluctuations.
Since we were then left with only three data points for a three parameter fit, and we were also not sure
whether the values for are possibly affected by finite size effects, we did not use the fit in Eq. (47) for
overlap fermions at z = 0.2.
As can be seen, this fit function provides a nice description of the numerical data. In principle, for the
cases of twisted mass and overlap fermions the divergent piece can be subtracted of from the evaluation of
this term in the free theory when the fermion mass is matched. This is, however, very difficult for the cases
of Wilson and hypercube fermions since there the matching of the fermion mass is not unique. We give in
Table 8 the fit results using Eq. (48).
1 In 4 dimensions, there is also a logarithmic divergence. But the origin is different and this divergence can be removed
renormalising the scalar condensate by a multiplicative renormalisation factor.
82
Table 8
Fit results for the scalar condensate using Eq. (48)
1
=A+ B
+ C log( )
2/3
(m )
= 0.2
B
0.215(150)
0.259(134)
0.371(179)
2 /dof
0.0397(447)
0.0610(389)
0.0767(518)
0.56
0.52
1.7
Wilson
HYP
TM
(m )2/3 = 0.4
0.118(103)
0.0938(898)
0.0352(1189)
A
2 /dof
Wilson
HYP
TM
OV
(m )2/3 = 0.8
0.258(52)
0.316(38)
0.218(69)
0.249(19)
0.107(76)
0.0790(543)
0.293(98)
0.0565(271)
0.0396(229)
0.0357(174)
0.0698(307)
0.0200(83)
0.67
0.48
0.47
0.13
2 /dof
Wilson
HYP
TM
OV
0.235(14)
0.311(14)
0.465(31)
0.128(10)
0.145(6)
0.162(6)
0.106(14)
0.160(4)
0.62
0.19
2.4
0.022
0.129(21)
0.0843(207)
0.178(45)
0.136(18)
5. Conclusions
In this paper we have tested four different lattice fermions in their approach to the continuum limit in the 2-dimensional massive Schwinger
model with Nf = 2 flavours of dynamical fermions. At fixed
scaling variable z = (mf )2/3 = 0.2, 0.4, 0.8 we have computed the
pseudoscalar mass m and the scalar condensate for various values of = 1/e2 a 2 .
For all kind of fermions
used, Wilson, hypercube, twisted mass and overlap fermions, the
scaling behaviour of m appears to be linear in a 2 . While this is expected for twisted mass
fermions at full twist as realized here and overlap fermions, this result is somewhat surprising for
Wilson and hypercube fermions where a linear dependence on the lattice spacing was expected.
Of course, it might be that another quantity can show different lattice artefacts and the scaling
behaviour shows another dependence on the lattice
spacing.
In Fig. 11 we show our final results for m computed non-perturbatively and extrapolated
to the continuum such that a direct comparison to analytical predictions can be made. Only for
a value of z = 0.2 there seems to be a consistency with Eq. (35) valid for strong couplings and
small masses while at z = 0.8 Eq. (37), obtained from a large mass expansion, seems to describe
the data. In general our conclusion is that to the precision we could compute our
results here, the
analytical formulae do not describe the non-perturbatively obtained values of m satisfactory
at all values of z.
As a second quantity we looked at the scalar condensate. We demonstrated that in our 2dimensional setup the use of a subtracted scalar condensate as derived from the integrated Ward
identity is very useful to compute the scalar condensate since the so defined scalar condensate is
free of divergence terms 1/a. Our data are also consistent with a logarithmic divergence in a
as can be derived in the free theory.
We also discussed some attempts to simulate the overlap operator dynamically by using an
infrared safe kernel to construct an approximate overlap operator. This approximate overlap operator is then used in the simulation and physical observables are corrected by reweighting with
the determinant ratio of the exact to the approximate operator. We tested this idea by computing
83
stochastically this determinant ratio. For our best candidate for an approximate overlap operator
in Eq. (33), for which we use an explicit infrared regulator in the sign function, we found that
the eigenvalue spectra of the exact and approximate operators are very similar, see Fig. 2, which
shows the lemon shaped difference spectra for various accuracies of the approximation corresponding to different values of in Eq. (33). Nevertheless, even in this case, the fluctuations in
the determinant ratio appeared to be very large even for a small value of the parameter .
Although this led us to conclude that this stochastic way of incorporating the correction of
the determinant ratio is not successful, we are still exploring to use the approximate overlap
operator as the guidance Hamiltonian in the molecular dynamics part of the Hybrid Monte Carlo
simulation while for the accept/reject Hamiltonians the exact overlap operator is used. We tested
this setup employing the hypercube operator as an approximate overlap operator as the guidance Hamiltonian in the molecular dynamics part. We found that even with this operator, which
appeared to be the worst choice in the case of the stochastic estimate of the determinant ratio
tested here, the acceptance rates look reasonable, see Ref. [35]. We are investigating this promising result further at the moment with the other approximate overlap operators discussed in this
paper.
Acknowledgements
We thank W. Bietenholz, V. Linke, G. Rossi, C. Urbach, U. Wenger for many useful discussions. In particular we thank C.B. Lang for pointing out to us to use the modified sign function of
Eq. (33). The computer centers at DESY, Zeuthen, and at the Freie Universitt in Berlin supplied
us with the necessary technical help and computer resources.
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]
J.S. Schwinger, Phys. Rev. 128 (1962) 2425.

F. Farchioni, C.B. Lang, M. Wohlgenannt, Phys. Lett. B 433 (1998) 377, hep-lat/9804012.
F. Farchioni, I. Hip, C.B. Lang, Phys. Lett. B 443 (1998) 214, hep-lat/9809016.
C. Gutsfeld, H.A. Kastrup, K. Stergios, Nucl. Phys. B 560 (1999) 431, hep-lat/9904015.
S. Durr, C. Hoelbling, Phys. Rev. D 71 (2005) 054501, hep-lat/0411022.
K.G. Wilson, Phys. Rev. D 10 (1974) 2445.
Alpha Collaboration, R. Frezzotti, P.A. Grassi, S. Sint, P. Weisz, JHEP 0108 (2001) 058, hep-lat/0101001.
R. Frezzotti, G.C. Rossi, JHEP 0408 (2004) 007, hep-lat/0306014.
P. Hasenfratz, hep-lat/9803027.
H. Neuberger, Phys. Lett. B 417 (1998) 141, hep-lat/9707022.
K. Jansen, Nucl. Phys. B (Proc. Suppl.) 129 (2004) 3, hep-lat/0311039.
C. Urbach, K. Jansen, A. Shindler, U. Wenger, hep-lat/0506011.
W. Bietenholz, U.J. Wiese, Nucl. Phys. B 464 (1996) 319, hep-lat/9510026.
W. Bietenholz, I. Hip, Nucl. Phys. B 570 (2000) 423, hep-lat/9902019.
L F Collaboration, K. Jansen, A. Shindler, C. Urbach, I. Wetzorke, Phys. Lett. B 586 (2004) 432, hep-lat/0312013.
A.M. Abdel-Rehim, R. Lewis, R.M. Woloshyn, Phys. Rev. D 71 (2005) 094505, hep-lat/0503007.
L F Collaboration, K. Jansen, M. Papinutto, A. Shindler, C. Urbach, I. Wetzorke, hep-lat/0503031.
L F Collaboration, K. Jansen, M. Papinutto, A. Shindler, C. Urbach, I. Wetzorke, hep-lat/0507010.
L F Collaboration, K. Jansen, et al., hep-lat/0507032.
F. Farchioni, et al., Eur. Phys. J. C 42 (2005) 73, hep-lat/0410031.
F. Farchioni, et al., Nucl. Phys. B (Proc. Suppl.) 140 (2005) 240, hep-lat/0409098.
F. Farchioni, et al., Eur. Phys. J. C 39 (2005) 421, hep-lat/0406039.
F. Farchioni, et al., hep-lat/0506025.
S. Aoki, O. Br, Phys. Rev. D 70 (2004) 116011, hep-lat/0409006.
S.R. Sharpe, J.M.S. Wu, hep-lat/0411021.
84
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
R. Frezzotti, G. Martinelli, M. Papinutto, G.C. Rossi, hep-lat/0503034.

P.H. Ginsparg, K.G. Wilson, Phys. Rev. D 25 (1982) 2649.
P. Hernandez, K. Jansen, M. Lscher, Nucl. Phys. B 552 (1999) 363, hep-lat/9808010.
P. Hasenfratz, S. Hauswirth, T. Jorg, F. Niedermayer, K. Holland, Nucl. Phys. B 643 (2002) 280, hep-lat/0205010.
M. Bochicchio, L. Maiani, G. Martinelli, G.C. Rossi, M. Testa, Nucl. Phys. B 262 (1985) 331.
S. Duane, A.D. Kennedy, B.J. Pendleton, D. Roweth, Phys. Lett. B 195 (1987) 216.
Z. Fodor, S.D. Katz, K.K. Szabo, JHEP 0408 (2004) 003, hep-lat/0311010.
N. Cundy, et al., hep-lat/0502007.
T. DeGrand, S. Schaefer, hep-lat/0506021.
N. Christian, K. Jansen, K.-i. Nagai, B. Pollakowski, hep-lat/0509174.
E. Anderson, et al., LAPACK Users Guide, third ed., Society for Industrial and Applied Mathematics, Philadelphia,
PA, 1999.
T. Blum, et al., Phys. Rev. D 69 (2004) 074502, hep-lat/0007038.
ALPHA Collaboration, U. Wolff, Comput. Phys. Commun. 156 (2004) 143, hep-lat/0306017.
A.V. Smilga, Phys. Rev. D 55 (1997) 443, hep-th/9607154.
C. Gattringer, hep-th/9503137.
J.E. Hetrick, Y. Hosotani, S. Iso, Phys. Lett. B 350 (1995) 92, hep-th/9502113.
M. Lscher, Commun. Math. Phys. 104 (1986) 177.
C. Hoelbling, C.B. Lang, R. Teppner, Nucl. Phys. B (Proc. Suppl.) 73 (1999) 936, hep-lat/9809075.
Supersymmetric compactifications of heterotic strings

with fluxes and condensates
Pantelis Manousselis a , Nikolaos Prezas b, , George Zoupanos c
a Department of Engineering Sciences, University of Patras, GR-26110 Patras, Greece
b Institut de Physique, Universit de Neuchtel, CH-2000 Neuchtel, Switzerland
c Physics Department, National Technical University of Athens, GR-15780 University Campus, Athens, Greece
Received 22 November 2005; received in revised form 15 December 2005; accepted 5 January 2006
Available online 19 January 2006
Abstract
We discuss supersymmetric compactifications of heterotic strings in the presence of H-flux and general
condensates using the formalism of G-structures and intrinsic torsion. We revisit the examples based on
nearly-Khler coset spaces and show that supersymmetric solutions, where the Bianchi identity is satisfied,
can be obtained when both gaugino and dilatino condensates are present.
1. Introduction
In string theory compactifications on CalabiYau (CY) manifolds and the corresponding dimensional reductions [1], the resulting low-energy field theory in four dimensions typically
contains a number of massless chiral fields, characteristic of the internal geometry, known as
moduli. These fields, arising in four dimensions as massless modes of the higher-dimensional
matter fields and from the gauge-independent variations of the metric of the compact space,
correspond generically to flat directions of the four-dimensional effective potential. Therefore,
the values taken by the moduli in the vacuum, which in turn specify the masses and couplings
of the four-dimensional theory, are left undetermined. Hence, the theory is without predictive
power.
E-mail addresses: pantelis@upatras.gr (P. Manousselis), nikolaos.prezas@unine.ch (N. Prezas),

zoupanos@mail.cern.ch (G. Zoupanos).
doi:10.1016/j.nuclphysb.2006.01.008
86
P. Manousselis et al. / Nuclear Physics B 739 (2006) 85105
Fortunately, the moduli problem in the form described above appears only in the simplest
choice of string backgrounds, where out of the plethora of closed-string fields only the metric is
assumed to be non-trivial. By considering more general backgrounds involving fluxes [2,3] as
well as non-perturbative effects [4,5], the four-dimensional theory can be provided with potentials for some or all moduli. The terminology fluxes refers to the inclusion of non-vanishing
field strengths for the ten-dimensional antisymmetric tensor fields with directions purely inside
the internal manifold. Therefore, the present day problem is the choice of the appropriate background which could lead to realistic and testable four-dimensional theories.
The presence of fluxes has a dramatic impact on the geometry of the compactification space.
Specifically, the energy carried by the fluxes back-reacts on the geometry of the internal space
and the latter is deformed away from Ricci-flatness. Then, the CY manifolds used so often in
string theory compactifications cease to be true solutions of the theory. For example, the requirement that some supersymmetry is preserved implies that the internal manifold is a non-Khler
space for heterotic strings with NSNS fluxes [69], while it can be a non-complex manifold
for type IIA strings [1012]. For type IIB strings, instead, the deviation is mild since the overall
effect due to the fluxes is a conformal rescaling of the original CY solution [13,14]. We should
mention at this point that in principle we can consider compactifications on CY manifolds with
fluxes too, but these are reliable only in the large-volume limit where the flux back-reaction can
be consistently ignored.
A considerable amount of literature has been devoted to the problem of including appropriately the back-reaction of the fluxes on the internal manifold and constructing examples of
manifolds which are true solutions of the theory. In general, these manifolds have non-vanishing
torsion. Consequently, demanding that the low-energy theory is supersymmetric implies that the
internal manifold admits a G-structure [15]. The existence of a G-structure is a generalization of
the condition of special holonomy. Subsequently, the allowed G-structures can be classified in
terms of their intrinsic torsion classes [16].
For heterotic strings, requiring that supersymmetry is preserved in the presence of nonperturbative effects such as gaugino condensation [17], leads to AdS4 spacetimes with noncomplex internal manifolds as potential solutions [18].1 The non-complex manifolds that we
will consider here are simple homogeneous nearly-Khler coset spaces. They were identified as
interesting possible solutions of heterotic string theory in Refs. [19,20], whereas supersymmetric solutions of the form AdS4 G/H were first obtained in Refs. [21,22]. In order to obtain
such solutions the authors in Refs. [21,22] assumed the presence of a gaugino condensate and
performed a case by case analysis. More recently, the 10-dimensional supersymmetry conditions in the presence of a non-vanishing gaugino condensate were examined in the language of
G-structures in Ref. [18].
Here, we generalize the setup of Ref. [18] by considering more exotic condensates. The consideration of other condensates besides that due to the gaugino is imposed upon us for both
technical and aesthetical reasons. At a technical level, the source-free Bianchi identity cannot
be solved in a supersymmetry-preserving manner if only the gaugino condensate is present. One
of the objectives of the present work is to show that supersymmetric solutions of the Bianchi
identity can be obtained if, in addition to the gaugino condensate, a dilatino condensate acquires
a non-vanishing vacuum expectation value (vev). On aesthetic grounds and although there is no
1 Originally, gaugino condensation had been suggested as a supersymmetry breaking mechanism in a CalabiYau
compactification [4,5]. Here, instead, we consider supersymmetric solutions on manifolds that are not CalabiYau.
87
known mechanism leading to condensation of the dilatino,2 we adopt here a more broad point
of view which considers the condensates on equal footing with the fluxes. Following this approach, we first formulate the supersymmetry conditions in the formalism of G-structures and
then we revisit the solutions of Refs. [21,22], while completing also their list by one more example.
We should mention that, as in most works that study the supersymmetry constraints on the
geometry, we do not check explicitly the equations of motion. In our case this would be a nontrivial task because the presence of the condensates renders the derivation of the equations of
motion subtle [18]. Instead, we rely on the fact that backgrounds that preserve some supersymmetry and where the Bianchi equations and the equations of motion for the matter fields are
satisfied, are automatically solutions of the Einstein equations [12,27]. It is straightforward to
show that this is indeed true for heterotic strings by specializing the results of [27] to heterotic
M-theory.
A key feature of the solutions we consider is that since they are not Ricci-flat, the fourdimensional spacetime will not be Minkowski but AdS (at least in the case where some supersymmetry is preserved). For this reason, the Ricci-flat CY manifolds were originally more
attractive candidates. Recently, however, this has become less relevant since whenever the inclusion of fluxes produces a stable vacuum without moduli, this vacuum turns out to be anti-de
Sitter. Therefore, it is not a serious drawback to start from an anti-de-Sitter solution in the first
place and hope that eventually non-perturbative effects will lift this to a Minkowski or de Sitter
vacuum according, for example, to the scenario proposed in [28] (see also [29] for an explicit
construction of this type in heterotic M-theory). This process may result in interesting GUTs in
four dimensions [3035].
The layout of this paper is as follows. In Section 2 we establish our notation and we present
the supersymmetry transformations of the fermionic fields of heterotic supergravity when various
condensates are non-vanishing. In Section 3 we examine the conditions imposed on the external
and internal geometries under the requirement of preserving N = 1 supersymmetry in the external space. The conditions on the internal geometry are formulated in terms of the intrinsic
torsion classes of the SU(3)-structure. In Section 4 we study some specific solutions based on
nearly-Khler coset spaces. We show that the Bianchi identity can be satisfied if and only if both
gaugino and dilatino condensates are present. Finally, in Section 5 we present a few concluding
remarks. Also, in two appendices we provide our gamma matrix conventions and some details
on the SU(3)-structure of the cosets under consideration.
2. Heterotic strings with condensates
The fields of heterotic supergravity, which is the low-energy limit of heterotic superstring
N,
theory, consists of the N = 1, D = 10 supergravity multiplet which contains the fields eM
M , BMN , , (i.e., the metric, the gravitino which is a RaritaSchwinger field, the twoform potential, the dilatino which is a MajoranaWeyl spinor, and the dilaton which is a
scalar), coupled to a N = 1, D = 10 vector supermultiplet which contains the gauge field
AM and the corresponding gaugino . The gauge field and the gaugino transform in the ad2 See, however, Refs. [23,24] where condensation of fermions in the gravity sector is considered in a different context
and Ref. [25] which discusses a similar effect in a 5-brane background. Moreover, Ref. [26] considered Minkowski vacua
in 11-dimensional supergravity with gravitino condensates.
88
joint of E8 E8 . The Lagrangian and the full supersymmetry variations can be found in Refs.
[3638].
The supersymmetry variations of the fermionic fields, including the relevant for our discussion
fermion bilinears, are [37]3

2 3/4
N PQ
M N P Q 12M
HN P Q
M = M +
32

1
1
NP Q
N PQ
M
Tr( N P Q ) + (N P Q )
8M
256
2
2
+
( M KL ) KL ,
96
1
= 3/8 MN FMN
4

3
2
1
3( ) ( MN ) MN ( MNKL ) MNKL ,
+
64
2
24
3 2 1 M
2
1 3/4 MNP
=
M +
Tr( MNP ) MNP . (2.1)
HMNP +
8
8
384
We have set the gravitational and YangMills coupling constants equal to 1. This in turn implies that we work with units where = 4. The supersymmetry parameter is a 10-dimensional
MajoranaWeyl spinor with 16 real components. The hats denote the supercovariant generalization of the corresponding fields,
FMN = FMN 3/8 ( [M N ] ),
(2.2)

1
H MNK = HMNK 3/4 2( [M N K] ) ( [M N K] ) .
4
(2.3)
It is well known that at the supergravity level the gauge-invariant 3-form field strength is
HMNP = 3[M BN P ] 1 A[M FN P ] . The corresponding Bianchi identity reads as
2
2
tr(F F ).
dH =
6
(2.4)
The full Bianchi identity includes one more term that is a string-theoretic correction and will be
added later.
In the above supersymmetry transformations we have assumed that some of the possible condensates between the fermionic fields of heterotic string theory have non-vanishing vacuum
expectation values. Our motivation for keeping only those appearing above is that, as we will
see later, they permit supersymmetric solutions without rendering the analysis computationally
challenging. We will also assume that the condensates ( [M N ] ) and ( [M N K] ), which appear in the supercovariant field strengths, are vanishing. We postpone the presentation of a more
complete analysis incorporating all possible condensates for future work [39].
3 Note that the coefficient of the last term in was corrected in Ref. [4].
89
An issue that should be addressed at this point concerns the supersymmetry variations of the
bosonic fields in the presence of fermion condensates. These read [37]
1
1
1
L
= L M ,
=
2,
AM = 3/8 M ,
eM
2
3
2

1 3/4
1
M N N M
BMN =
2
2MN
4
2
1 3/8
+
2 [M Tr(AN ] ),
2
(2.5)
and it is obvious that had the fermions acquired non-zero vevs, these variations would not vanish
in general. In such a case, of course, checking the supersymmetry variations would be redundant since the non-vanishing vevs for the fermions would imply anyway broken spacetime (e.g.,
Lorentz) symmetry. The crucial point here is that in quantum field theory the fermion condensates
can be non-zero while maintaining a vanishing fermion vev. Therefore, although the fermion condensates are assumed to be non-vanishing, the vevs of the corresponding fermions are taken to
be zero, as is necessary for preserving maximal symmetry in spacetime. It should be emphasized
that this effect has been explicitly demonstrated in N = 1 super-YangMills theory where gaugino condensation indeed takes place non-perturbatively without leading to non-zero vevs for the
gauginos [40,41]. Here, we assume that this is precisely what happens in the strongly-coupled
microscopic theory that underlies our effective supergravity description. This is in line with the
original treatment of gaugino condensation in heterotic supergravity where the fermion vevs are
zero while a non-trivial condensate is generated [4,5]. The punchline is that the supersymmetry
variations of the bosons are trivially vanishing since the fermion vevs are consistently zero.
In our setup, the only extra assumption is that some unknown quantum effects can lead to nontrivial condensates for the gravitinos. Actually, such a point of view was already taken in [26]
while [25] provided some evidence that this effect indeed takes place in NS5-brane backgrounds.
Furthermore, we should stress that here we simply work in an effective approach where we ask
what would happen if such condensates were generated. Hence, our treatment is entirely analogous to the usual chiral Lagrangian approach to hadron physics. In this approach, one simply
assumes that in the IR a non-vanishing quark condensate breaking chiral symmetry is formed,
although the microscopic theory (i.e., QCD) governing this effect is out of reach in this regime.
Moreover, in order to maintain Lorentz invariance the quark vevs are assumed to be zero despite
the presence of the non-perturbative quark condensate.
We now make the following field redefinitions:
= e8/3 ,
gMN = e2 gMN ,
/2

2 (0) (0)
(0)
M
,
M
4
M = e
(0)
= e/2 (0) ,
(0)
FMN = FMN ,
1
= e/2 (0) ,
2
= e/2 (0) ,
M = e M ,
(0)
3 (0)
HMNP = HMNP ,
2
(2.6)
with the quantities bearing the superscript (0) referring to those in (2.1). The supersymmetry
variations (2.1) become

1
1
1
1
4
+ ,
M = M
(2.7)
HM 2M M M
4
3
4
3
4
90

3
1 MN
1
1
MN
MNKL
,
3 MN
=
MNKL
(2.8)
FMN +
4
32
2
24
1
= /
+ (H + ),
(2.9)
24
[] ). Notice that since these condensates are not singlets of the E8 E8
where [] = (
gauge group, turning on vevs for them can break part of the original gauge symmetry. Of course,
this is similar to the (partial) breaking of the original gauge symmetry due to a non-vanishing
background gauge field strength.
MKL ) because the difference in M KL ) = (
In the field-redefinitions above we set (
M ) which are zero due to the assumption of maximal symmetry
volves terms of the form (
in the external space and due to the absence of globally-defined vector fields on six-dimensional
manifolds with SU(3)-structure. In the new variables the Bianchi identity (2.4) reads as
1
dH = tr(F F ),
2
with the supercovariant field strengths (2.3) being
FMN = FMN ,
H MNK = HMNK 6MNP ,
(2.10)
(2.11)
(2.12)
since after the field redefinitions we set ( [M N K] ) = 0. We use the standard shorthand notation
HM = HMNP N P , H = HMNP MNP , M = MNP N P , = MNP MNP with MNP =
1
1
16 tr( MNP ) and MNP = 16 (MNP ).
We emphasize that the metric in the above equations is the sigma-model metric and is re0
0 . We have also used the fact that the
as gMN = e2 gMN
lated to the Einstein metric gMN
covariant derivatives of a spinor with respect to these two metrics are related as M =
(0)
M 12 M N N .
3. Conditions for 4-dimensional N = 1 vacua
3.1. Metric ansatz and SU(3)-structure
0
describes the warped product M1,3 w K6 of a 4-dimensional maximally
We assume that gMN
symmetric spacetime M1,3 with a compact 6-dimensional internal space K6 . Explicitly

0
ds 2 = gMN
(3.1)
(x, y) dx M dx N = e2D(y) g (x) dx dx + g mn (y) dy m dy n .
Since we are interested in 4-dimensional vacua with N = 1 supersymmetry, we demand that

on K6 there exists a globally-defined complex spinor + (and its conjugate with opposite
chirality). In other words, K6 should be equipped with an SU(3)-structure. Our ansatz for the
10-dimensional MajoranaWeyl spinor is
(x, y) = f (y)+ (x) + (y) f (y) (x) (y),
(3.2)
where f (y) is an arbitrary complex function. This ansatz yields N = 1 supersymmetry in

M1,3 expressed in terms of the Weyl spinors . The 6-dimensional spinors are normalized4
4 Our spinor conventions and some useful formulae can be found in Appendix A.
91
as +
+ =
= 1. As usual the four-dimensional spinors are taken to be Grassmann while
the six-dimensional ones are commuting.
The SU(3)-structure is characterized by an almost complex structure and5 the associated 2form Jmn , and by a (3, 0)-form mnp . These forms are globally-defined and non-vanishing and
they are subject to the following compatibility conditions:
4i
(3.3)
J J J.
3
Furthermore, the spinors determine (up to a phase which we fix to a convenient value) the
SU(3)-structure forms as
J = 0,
Jmn = i+
mn + = i
mn ,
(3.4)
mnp + ,
mnp =
(3.5)
= +
mnp .
mnp
(3.6)
n
Using Fierz identities one can show that indeed Jm is an almost complex structure, i.e., it satisp
fies Jm n Jn p = m . Then, the projectors ( )m n = 12 (m n iJm n ) can be used to separate the
holomorphic and antiholomorphic parts of a generic form.
Now we can decompose the condensates in 4- and 6-dimensional pieces. First, notice the
gaugino field is Majorana and of the same chirality as the supersymmetry parameter . Hence,
it admits a decomposition = + + with the 4-dimensional gaugino fields.
Notice that this is not a zero-mode decomposition but the usual decomposition in terms of SU(3)
singlets. Then we obtain
mnp =

1
Tr( mnp ) =
= 3 mnp + c.c. , (3.7)
+ mnp + +
mnp
16
+ (5) )) =
where 3 = 12 Tr((1
+ is the 4-dimensional condensate. We see that the
condensate consists only of (3, 0) and (0, 3) pieces. This expansion is valid for mnp as well.
+ (5) )) the vev of the 4-dimensional dilatino condensate, we have
By denoting 3 = 12 Tr((1
3
mnp = mnp + c.c.
In order to expand correctly the dilatino-gaugino condensates [] we have to take into
account that and have opposite chiralities. Hence, is expanded as = +
+ with the 4-dimensional dilatinos. Then we find that is a real scalar =
(+ + + ), mn is a real 2-form mn = i(+ + )Jmn 0 Jmn , and mnkl
is a real 4-form mnkl = 3J[mn Jkl] . Also, all of them have an adjoint E8 E8 index that we
suppress.
The supercovariant H-flux (2.3) can be expanded in terms of the SU(3)-invariant forms as

1
3 (1,0)
o(2,1)
+ H [m Jnp] + c.c.,
H mnp = mnp H (3,0) + H mnp
(3.8)
48
4
where H (3,0) = mnp H mnp , H m(1,0) = ( + )m s H snp J np and H o(2,1) the primitive (2, 1) piece
of H mnp which satisfies H o(2,1) J = 0. Notice that due to (2.12) and the fact that mnp consists
(2,1)
(2,1)
of only (3, 0) and (0, 3) pieces, we have H mnp = Hmnp .
5 As is common, we will use the same symbol for both the almost complex structure tensor and the associated 2-form.
92
Now, we are ready to proceed to the analysis of the supersymmetry variations (2.7), (2.8) and
(2.9) for the setup under consideration.
3.2. Conditions on the external geometry
The 4-dimensional part of the gravitino variation reads as
1
(3.9)
mnp mnp = 0,
12
where we have introduced the combination mnp = mnp + 34 mnp = ( 3 mnp + c.c.). Since
gMN = e2(D) g MN and M = e(D) M we can rewrite Eq. (3.9) as
=
1
1
= + n n (D ) e(D) ,
2
12
(3.10)
and its integrability condition gives = 0 [2].

By assumption the 4-dimensional metric g (x) is maximally symmetric. Hence, the corresponding Riemann curvature tensor is of the form
R
R = (g g g g )
12
with R the constant scalar curvature. For this metric we can show easily that =
Then the integrability condition yields
R
4 .
R
1
3 n (D ) n (D ) + 3 m m e(D) e2(D) 2 = 0.
(3.11)
4
12
For vanishing condensates Eq. (3.11) reduces to Eq. (2.8) of [2]. In this case one finds that
R = 0 since the only possible constant value for n (D ) n (D ) on a compact manifold
is zero. Hence, M1,3 is Minkowski and the warp factor satisfies D(y) = (y) up to an additive
constant.
At first sight a non-zero condensate mnp seems to open up a much wider range of possibilities. However, we will see in a moment that the supersymmetry conditions imply the constancy
of D . Then M1,3 is AdS4 with constant negative curvature given by
1
R = e2(D) 2 .
(3.12)
3
Notice that 2 is negative because is antihermitian.
Another way of computing the dependence of R on the vev of the condensate is the following. The 4-dimensional spinors in a maximally symmetric spacetime satisfy + = W
and = W + . The curvature is then R = 48|W |2 . Using the fact that
1
1
+ = + + m m (D )+ = W e(D) + (5) + m m (D )
2
2
and inserting the latter expression in (3.9) we obtain
f W e(D) + +
f
f
m (D ) m
mnp mnp = 0.
2
12
(3.13)
93
Multiplying now Eq. (3.13) with +

from the left we obtain
mnp
f
mnp +
= 0
12
f
( )3 2 = 0,
f W e(D)
12
f W e(D)
(3.14)
mnp )1/2 . Using Fierz identities one can find that 2 = 48. For vanishwhere = (mnp
ing total condensate 3 = 0, we obtain W = 0 and hence M1,3 is flat. However, in general it
holds that
6.
R e2(D) ||
(3.15)
n from the left

Now we can prove the constancy of D . By multiplying Eq. (3.13) with
mnpq
[mn
pq]
= 3J J
we obtain
and using the equality
i
f n (D ) f m (D )Jn m
2
3 mpq [mn pq]
3i
J J
+
(3.16)
f
f nmp J mp = 0.
256
256
Since is a (3, 0) + (0, 3) form, we have nmp J mp = 0 and mpq J [nm J pq] = 0. This immediately shows that
n (D ) = 0,
(3.17)
and demonstrates that the constancy of D is actually imposed upon us from the supersymmetry conditions.
3.3. Conditions on the internal geometry and intrinsic torsion classes
3.3.1. Gravitino variation
Next we do the analysis of the 6-dimensional part of the gravitino variation. This reads as

1
1
1
m = m
(3.18)
Hmpq 2mpq + mpq pq npq m npq = 0.
4
6
12
Inserting the spinor ansatz (3.2) in Eq. (3.18) we obtain

1
1
m + = m log f + +
Hmpq 2mpq + mpq pq +
4
6
1
+ npq m npq + .
12
Using the identity
(3.19)
m npq = mnpq + mq np + mp qn + mn pq
we can rewrite Eq. (3.19) as
m + = m log f + +

1
1
1
Hmpq mpq + mpq pq + + npq m npq + .
4
6
12
(3.20)
94
Expressing m npq as m npq = 2i m npqrs rs 7 we then obtain

1
1
m + = m log f + +
Hmpq mpq + mpq + i mpq pq + ,
4
6
(3.21)
where mpq = 3!1 klr klr mpq .

However, the total contribution of mnp in the above equations drops out. Indeed, due to the
) we have
mpq = i 3 mpq + c.c. Using also
imaginary (anti)-self-duality of mnp (mnp
pq
+
the fact that mpq = 0, we finally obtain
1
m + = m log f + + H mpq pq + ,
(3.22)
4
where we have defined H mnp = H mnp + 16 mnp . Consequently, the spinor + is parallel under
()
()
the connection with torsion () = 1 H . Defining the curvature through [m , n ]+ =
1 ()
pq +
4 Rmnpq ,
the integrability condition for Eq. (3.22) implies
()
Rmnpq
J pq = 0.
(3.23)
Now using Eq. (3.22) and its conjugate we find that |f | = const. Hence, in general f = ei . For
our purposes, however, taking f = 1 is not restrictive and we will do so.
It is straightforward to compute the covariant derivatives of the SU(3)-structure forms:

m Jn p = H sm p Jn s + H s mn Js p ,
(3.24)
p
p
p
m kln = H mn pkl + H lm pkn + H mk pln .
(3.25)
These relations imply that the internal space has contorsion given by H mnp , which in turn has a
contribution due to the H-flux and another due to the dilatino condensate.
3.3.2. Dilatino variation
In the case that a non-zero gaugino condensate is present, the vanishing of the dilatino variation demands that H mnp has a non-zero (3, 0) + (0, 3) piece. Indeed, the dilatino variation is
1
(Hmnp + mnp ) mnp = 0,
24
and for the spinor ansatz (3.2) it becomes
= m m +
m m + +
1
(Hmnp + mnp ) mnp + = 0.
24
(3.26)
(3.27)
from the left gives

Multiplying Eq. (3.27) by
(H mnp + mnp ) mnp = 0.
(3.28)
This fixes the (3, 0) and (0, 3) parts of the supercovariantized flux in terms of the gaugino condensate vev
(3,0)
= 3 mnp .
H mnp
(3.29)
If the dilatino condensate is zero, so that H mnp = Hmnp , the above equations imply that in order
to preserve supersymmetry in the presence of a non-vanishing gaugino condensate an H-flux
with non-vanishing (3, 0) and (0, 3) components should be present too.
95
Multiplying now Eq. (3.27) by +

k from the left gives
1
i
iJk n n + k H mnp Jk m J np + H kmn J mn = 0.
(3.30)
8
8
The contribution of mnp drops out since the gaugino condensate has only (3, 0) + (0, 3) components. From Eq. (3.30) we obtain an equation for the dependence of the dilaton on the internal
space coordinates

1
i
k = H mnp Jk m J np = H (1,0) H (0,1) .
8
8
(3.31)
We see that only the non-primitive (2, 1) + (1, 2) piece of H mnp (and hence of Hmnp since
they only differ in their (3, 0) + (0, 3) parts), contributes to the variation of the dilaton. In particular, if we have only non-zero the (3, 0) and (0, 3) components of the H-flux, the dilaton is
constant, k = 0. Then due to Eq. (3.17) the warp factor D(y) is constant too. This shows that
the assumptions of Refs. [21,22] on the constancy of D and are not just sufficient but also
necessary for unbroken supersymmetry.
3.3.3. Gaugino variation
Finally we consider the gaugino variation. This is given by

3
1
1
1
mn
mn
mnkl
,
3 mn mnkl
= Fmn +
4
32
2
24
while inserting the spinor ansatz and the decomposition of the condensates yields

1
3
= Fmn + 0 Jmn mn + ,
4
16
(3.32)
(3.33)
1
where the terms proportional to are zero due to the identity (1 + 24
Jmn Jkl mnkl )+ = 0.
A vanishing gaugino variation demands that the field strength satisfies Fmn = Jm k Jn l Fkl , i.e.,
(2,0)
= F (0,2) = 0. Furthermore, the non-primitive part of Fmn is compensated by the vev of the
F
dilatino-gaugino condensate 0 as 0 = 89 Fmn J mn . In the usual case where this condensate is
taken to be zero, we end up with the standard result that a supersymmetry preserving background
gauge field has to be a primitive (1, 1) form.
3.4. Torsion classes

We can summarize now our findings in the language of intrinsic torsion classes which are
defined as
dJ =
3i
1 ) + W4 J + W3 ,
(W1 W
4
(3.34)
and
d = W1 J J + W2 J + W5 ,
(3.35)
satisfying J W3 = J J W2 = W3 = 0. The classes W1 and W2 can be decomposed

in real and imaginary parts as W1 = W1+ + W1 and W2 = W2+ + W2 .
The classes W1 and W2 are vanishing when the almost complex structure is integrable, i.e.,
when the manifold is complex. A Khler manifold has furthermore W3 = W4 = 0. Finally, CY
96
manifolds have in addition W5 = 0. Hence, one can think of the five torsion classes as parameterizing the deformation away from SU(3)-holonomy.
Using Eqs. (3.24) and (3.25) we obtain

1
1
W1 = H (0,3) = 8 ( )3 ( )3 ,
6
6
W2 = 0,

W3 = i H o(2,1) H o(1,2) ,

i
W4 = H (1,0) H (0,1) ,
2

W5 = i H (1,0) H (0,1) .
(3.36)
In order to derive the class W1 , we took into account Eq. (3.29) and the definition H mnp =
H mnp + 16 mnp . Using Eq. (3.31) we can express the torsion classes W4 and W5 in terms of the
dilaton gradient as
W4 = 4 d,
W5 = 8 d.
(3.37)
Now we summarize our findings in order to be easy to contrast them to the case with H-flux
but without condensates [6] and to the case with H-flux and only the gaugino condensate [18].
We have found that:
The gaugino condensate mnp induces a (3, 0) + (0, 3) piece to the supercovariant H -flux.
The dilatino and gaugino condensate yield a non-vanishing W1 class, rendering the internal
spaces non-complex.
The class W2 is zero for the H-flux and for all the condensates we turned on.
The intrinsic torsion classes Wi , i = 3, 4, 5 are determined only in terms of the (2, 1) + (1, 2)
pieces of the flux and hence do not depend on the condensates.
The spacetime curvature depends on the dilatino and gaugino condensates in such a way that
one can tune them to obtain a Minkowski vacuum with a non-complex internal space.6
The dilatino-gaugino condensates in the gaugino variation allow for background gauge fields
that can have a non-primitive (1, 1) piece.
4. Supersymmetric solutions on nearly-Khler spaces
In the previous section we presented a set of conditions on the intrinsic torsion classes of a sixdimensional manifold with SU(3)-structure, which are necessary for obtaining supersymmetric
vacua of heterotic string theory in the presence of H-flux and several condensates. The most
general manifolds with torsion classes specified by Eq. (3.36) are known as G1 manifolds [16].
In the absence of condensates, only an H-flux with non-trivial (2, 1) + (1, 2) components can
be present if some supersymmetry is to be preserved. In particular, if only the primitive part of
the flux is non-zero we have W1 = W2 = W4 = W5 = 0. These manifolds are known as specialhermitian and are well-studied in the mathematical literature. However, so far it has been proved
that is difficult to satisfy the Bianchi identity for compactifications of this type.
6 A similar conclusion was reached in Ref. [26] for compactifications of 11-dimensional supergravity with gravitino
condensates.
97
Another natural choice is to assume that only the condensates and the corresponding pieces
of the H-flux induced by them are non-vanishing. In this case, the H-flux is of (3, 0) + (0, 3)
type and, according to Eq. (3.36), the appropriate six-dimensional manifolds will have only the
class W1 different from zero. Notice that supersymmetry specifies further the H-flux in terms
of the condensates, effectively determining W1 in terms of 3 and 3 . Setups with 3 = 0 have
been already considered in Refs. [21,22]. One of our purposes here is to revisit them in the
framework of SU(3)-structures and intrinsic torsion. Besides that, the analysis of the previous
section showed that some of the assumptions of Refs. [21,22] are actually necessary and not just
sufficient for preserving spacetime supersymmetry. Furthermore, the authors of Refs. [21,22] did
not take into account the modifications to the heterotic Bianchi identities due to the non-trivial
torsion. As we shall see, to satisfy the Bianchi identity, the presence of the gaugino condensate
is not enough.
Six-dimensional manifolds with non-zero W1 have been well-studied too in the mathematical
literature. They are known as nearly-Khler and they have certain special properties. Among
others, they are Einstein spaces of positive scalar curvature, their almost complex structure is
never integrable since the Nijenhuis tensor is non-zero, their first Chern class vanishes, and they
admit a spin structure. In the ensuing we will discuss in detail some examples of nearly-Khler
spaces which are based on coset spaces.
Let us just mention for the moment that it would be interesting to relax the condition
H (2,1) = 0 and turn-on a primitive (2, 1) + (1, 2) piece of the H-flux (in addition to the
(0, 3) + (3, 0) piece induced by the condensates). Then the conditions on the intrinsic torsion
would imply that the compactification manifold should have W2 = W4 = W5 = 0 and (choosing
purely imaginary condensates) W1+ = 0. Such manifolds are particular cases of half-flat manifolds and one simple but interesting realization is provided by twisted toroidal orbifolds [42] (see
also [43] for the 7-dimensional analogue). It would be worthwhile to investigate if more general
solutions can be provided by such spaces, but we postpone that for future work.
4.1. Nearly-Khler coset spaces
The only known examples of compact nearly-Khler spaces are 3-symmetric spaces that can
be described as homogeneous cosets. These are (i) G2 /SU(3), which is an S 6 but less symmetric
than the usual 6-sphere, (ii) Sp(4)/(SU(2) U (1))non-max. which is similarly a less symmetric version of CP3 , (iii) SU(3)/(U (1) U (1)) which is the flag manifold F (1, 2), and (iv)
SU(2)3 /SU(2) which is isomorphic to S 3 S 3 . All these spaces admit an SU(3)-structure. Although in general they are half-flat, there are special values of their moduli for which only W1
is non-zero. For these special values these cosets become nearly-Khler.
In Appendix B we present the SU(3)-structures of the 4 cosets and compute their intrinsic
torsion classes. Inspection of the results shows that all possible radii of the cosets have to be the
same if we want to keep a vanishing W2 class. Then, the only non-zero class is W1 and it takes
the general form W1 = i wa . The actual values of w for each coset can be found in Appendix B.
Furthermore, using the corresponding curvature 2-forms [55], we can compute the first Pontryagin classes:
G2 /SU(3):
tr R R = 0,
Sp(4)/(SU(2) U (1)):
tr R R =
18
J J,
a2
98
18
tr R R = 2 J J,
a
2
8
tr R R =
J J.
9a
SU(3)/(U (1) U (1)):

SU(2)3 /SU(2):
(4.1)
Hence, for all cosets under consideration and for metrics with nearly-Khler structure we have a
general formula for the first Pontryagin class given by tr R R = pa 21 J J . Since J J is exact,
this class is cohomologically trivial. Using the general expression for W1 we can also write
tr R R =
p1
|W1 |4 J J.
w4
(4.2)
The supersymmetry conditions demand that the torsion class W1 is fixed in terms of the
condensates. This stabilizes the radial modulus to a value

1 2
w2
a = 3 3 .
(4.3)
8
6
Since W1 is imaginary for the cosets under consideration, the condensate combination ( )3
1 3
6 ( ) has to be imaginary too.
4.2. Bianchi identity
Let us now consider the Bianchi identity for the H-flux. The full Bianchi identity includes a
correction due to the gravitational ChernSimons term required for anomaly cancellation [44]
and it reads

1
1
(+)
(+)
dH =
(4.4)
tr R R
Tr F F ,
2
30
where R (+) is the curvature 2-form of the connection (+) = + 14 H . We have also adopted the
usual normalization for the trace of the gauge field strengths.
The torsion of the connection whose curvature appears in the Bianchi identity is the opposite
of the torsion of the connection () = 14 H appearing in the supersymmetry variation of
the gravitino in the internal space. Although this fact is well-established only for the case where
the torsion comes entirely from the H-flux [45], it is quite natural to expect that the proper generalization of the gravitational ChernSimons correction to the Bianchi involves the full torsion
tensor appearing in the gravitino variation. Let us also mention that demanding only supersymmetry and anomaly freedom does not specify completely the torsion relevant for the gravitational
ChernSimons term [46]. From this point of view our choice is perfectly consistent. Notice, furthermore, that the ambiguity is fixed by the additional requirement of conformal invariance on
the worldsheet [45] or, equivalently, of satisfying the spacetime equations of motion. It would
be extremely interesting and important to actually verify the validity of our choice with an explicit calculation. However, such a computation is bound to be subtle due to the presence of the
condensates.
We can now proceed with the computation of the quantities that appear in the Bianchi identity.
The curvature tensor for (+) = + 14 H is

(+)
= Rmnpq + 2[m H n]pq H mp r H rnq H np r H rmq ,
Rmnpq
(4.5)
99
and the corresponding curvature 2-form is

1 (+) p
(+)
Rmn
= Rmnpq
e eq .
2
(4.6)
For nearly-Khler spaces the torsion H mnp has only (3, 0) + (0, 3) parts which are fixed by
the class W1 as
H mpq = 8(W1 mpq + c.c.).
(4.7)
Using Eq. (3.25) and the identity

pmn pkl = 16 + [m [k + n] l]
(c.f., for instance, [12]), we obtain
[m H n]pq = 24|W1 |2 J[mn Jpq] ,

H mp r H rnq H np r H rmq = 4|W1 |2 (gmp gnq gmq gnp ) Jmn Jpq 3J[mn Jpq] .
Subsequently, the generalized curvatures are found to be

(+)
Rmnpq
= Rmnpq + 60|W1 |2 J[mn Jpq] 4|W1 |2 (gmp gnq gmq gnp ) Jmn Jpq ,

()
= Rmnpq 36|W1 |2 J[mn Jpq] 4|W1 |2 (gmp gnq gmq gnp ) Jmn Jpq .
Rmnpq
(4.8)
(4.9)
(4.10)
(4.11)
The left-hand side of the Bianchi identity (4.4) can be derived by using the fact that the
H mnp Hmnp 6mnp is fixed, by the vanishing of the dilatino variation, to be H mnp =
3 mnp + c.c. Since for nearly-Khler spaces holds that d = W1 J J , we find that

dH = 3 6 3 W1 J J + c.c.
(4.12)
Then we can eliminate the explicit dependence on and its conjugate by using the supersymmetry condition W1 = 8(( )3 16 ( )3 ). Finally we obtain

1
|W1 |2 Re 3 W1 J J,
dH =
(4.13)
4
3
where we have rescaled 3 18
35 .
The right-hand side of the Bianchi identity (4.4) depends on the first Pontryagin class of (+) .
After a slightly tedious calculation we obtain
tr R (+) R (+) = tr R R 4608|W1 |4 J J.
(4.14)
Next, to obtain a feeling of the implications of the Bianchi identity (4.13), we assume that there
is no background gauge field. Using Eq. (4.2) we find that the Bianchi identity reduces to the
following equation

p1
1
|W1 |2 Re 3 W1 =
2304
|W1 |4 .
(4.15)
4
2w 4
In addition, for simplicity let us consider the case G2 /SU(3) for which p1 = 0. Recall that for
all cosets the class W1 is imaginary. Then we find the following solution for 3 , which in the
present case is purely imaginary:

1
3
2
+ 2304|W1 | .
= i|W1 |
(4.16)
4
100
Moreover, the gaugino condensate is fixed to a purely imaginary value as well given by

1
3
Im 3 =
Im 3 |W1 |,
35
8
which in terms of the class W1 becomes

3
1
41
2
+ 3 2304|W1 | .
Im = |W1 |
35
8
(4.17)
(4.18)
It is straightforward to find similar solutions for the other three cosets.

It is clear that the previous solution retains in four dimensions the full gauge group E8 E8
and thus it is undesirable as a realistic GUT. However, we can turn on a background gauge field
that breaks part of the gauge symmetry in a supersymmetric way. The condition for supersymmetry dictates that the gaugino variation (3.33) is zero. A simple way to deal with this is to assume
that all the condensates [] are vanishing and to consider further the curvature 2-form corresponding to the SU(3) connection with torsion () . Embedding this connection in the gauge
()
group, yields a background SU(3) gauge field (Fab )mn = Rabmn which leads to a vanishing gaug()
ino variation as a consequence of the integrability condition Rabmn mn + = 0. For such a choice
1
() R () , where the latter is found to be
30 Tr F F = tr R
tr R () R () = tr R R + 1536|W1 |4 J J.
The Bianchi identity leads now to the relation

1
|W1 |2 Re 3 W1 = 3072|W1 |4 ,
4
(4.19)
(4.20)
which fixes the dilatino condensate and, in turn, also the gaugino condensate by the supersymmetry condition (4.3). The background gauge field breaks the gauge group down to E6 E8 as
does the standard embedding in the case of CY compactifications.
) is imaginary (anti)-self-dual, one can easily check that
Using now the fact that mnp (mnp
for all the above solutions the NSNS flux is coclosed:
d H =0
(4.21)
and hence it solves the source-free equation of motion. Accordingly, since the Bianchi equation
and the equation of motion for H are satisfied and the background preserves some supersymmetry, the Einstein equations are also verified as a consequence of the results of [27].
5. Conclusions
In the present work we have examined in detail some examples of supersymmetric AdS compactifications of heterotic strings on non-complex manifolds, which we believe will be eventually
phenomenologically interesting. The considered six-dimensional manifolds include the coset
spaces G2 /SU(3), Sp(4)/(SU(2) U (1))non-max , SU(3)/(U (1) U (1)) and SU(2)3 /SU(2).
These have been examined some time ago in the context of Coset Space Dimensional Reduction [4749], in attempts to obtain realistic GUTs from extra dimensions [3133,50,51], and
also in the context of heterotic sting theory in studies of possible supersymmetry preserving
backgrounds in Refs. [1922]. These solutions have been analyzed here in the framework of
G-structures while a new example based on the coset SU(2)3 /SU(2) has been added.
101
The conjecture that nearly-Khler spaces have no complex-structure moduli [52] and the fact
that the supersymmetry preserving conditions fix the radii to specific values, implies that one
should expect supersymmetric AdS vacua with most moduli stabilized in this approach. Subsequently, it would be worthwhile to analyze the possibilities for lifting the AdS vacuum to
Minkowski or de Sitter [28,30] and study supersymmetry breaking in this context. Work in this
direction is currently in progress.
Acknowledgements
We would like to thank K. Behrndt, G.L. Cardoso, G. DallAgata, J.-P. Derendinger, A. Frey,
P. Forgacs, J. Gauntlett, M. Lippert, D. Lst, A. Micu, A. Nagy, G. Papadopoulos, K. Sfetsos,
A. Uranga and D. Zoakos for helpful discussions and correspondence. We are also grateful to
A. Frey, M. Lippert and D. Tsimpis for interesting comments and questions on the first version
of this paper, as well as to T. Kimura for pointing out a numerical mistake.
The work is supported by the EPEAEK programme Pythagoras and co-founded by the
European Union (75%) and the Hellenic state (25%). The work of N.P. is supported by the
Swiss National Science Foundation and by the Commission of the European Communities under
contract MRTN-CT-2004-005104.
Appendix A. Gamma matrix conventions
The 10-dimensional gamma matrices M with M = 0, 1, . . . , 9 satisfy { M , N } = 2MN
where MN = (1, 1, . . . , 1). Splitting to 4 + 6 indices , = 0, 1, 2, 3 and m, n = 4, . . . , 9, we
can write them as = I and m = 5 m , where the usual 4-dimensional gamma
matrices, (5) is the chirality matrix satisfying ( (5) )2 = 1 and { , (5) } = 0, and m are the
gamma matrices in 6-dimensional Euclidean space.
The 6-dimensional spinors have opposite chiralities (7) = , where (7) =
TC
=
i4 9 . They satisfy
(6) where C(6) is the 6-dimensional charge conjugation matrix. The 4-dimensional Weyl spinors are also chiral (5) = with (5) = i0 1 2 3 .
Moreover, = T C(4) with C(4) the 4-dimensional charge conjugation matrix. As usual, for
a Dirac spinor we define = 0 .
The 10-dimensional chirality and charge conjugation matrix are (11) = (5) (7) and
C(10) = C(4) (5) C(6) . With these definitions the spinor (3.2) satisfies (11) = and =
T C(10) , i.e., it is chiral and Majorana. For the gamma matrix manipulations we used [53].
Finally, we present some spinor bilinears that were used in the considerations of Section 3:
+
+ =
= 1,
(A.1)
+ = +
= 0,
m
m
=
= 0,
mn
mnpq
=
= 0,
mn
mn
= iJ ,
mnpq
= 3J [mn J pq] ,
mnp
mnpqrs
= 0,
mnpqr
mnpqr

=
= 0.
(A.2)
(A.3)
(A.4)
(A.5)
(A.6)
(A.7)
(A.8)
102
Appendix B. SU(3)-structures on coset spaces

In [54] the intrinsic torsion classes for the coset SU(3)/(U (1) U (1)) were calculated in
terms of the radii a, b, c. Here we extend this computation to the rest of the coset spaces that
admit a nearly-Khler structure. The data required for this exercise can be found in [55,56].
B.1. SU(3)/(U (1) U (1))
Metric:

ds 2 = a e1 e1 + e2 e2 + b e3 e3 + e4 e4 + c e5 e5 + e6 e6 .
(B.1)
SU(3)-structure:
J = ae1 e2 + be3 e4 ce5 e6 ,

dJ = (a + b + c) e1 e3 e5 + e1 e4 e6 e2 e3 e6 + e2 e4 e5 ,

= (abc) e1 + ie2 e3 ie4 e5 + ie6 ,

d = 4i abc e1 e2 e3 e4 e1 e2 e5 e6 + e3 e4 e5 e6 .
(B.2)
(B.3)
(B.4)
(B.5)
Non-vanishing intrinsic torsion classes:

2i a + b + c
,
3
abc

4i 1
W2 =
a(2a b c)e12 b(2b a c)e34 + c(2c a b)e56 .
3 abc
W1 =
(B.6)
(B.7)
B.2. SU(2) SU(2)

Metric:

1 2
1 3
1 1
1
6
6
2
5
5
3
4
4
ds = a e e + 3e e + b e e + 3e e + c e e + 3e e .
3
3
3
(B.8)
SU(3)-structure:
J = ae1 e6 be2 e5 + ce3 e4 ,

1
1
dJ = (a + b c)e1 e2 e4 (a b + c)e1 e3 e5
3
3
1
+ (a b c)e2 e3 e6 + 3(a + b + c)e4 e5 e6 ,
3

abc 1
e 3ie6 e2 + 3ie5 e3 3ie4 ,
=i
27

4i
d =
abc e1 e2 e5 e6 e1 e3 e4 e6 + e2 e3 e4 e5 .
3
2i
(a + b + c),
W1 =
9 abc
(B.9)
(B.10)
(B.11)
(B.12)
(B.13)
4i
a(2a b c)e1 e6 + b(2b a c)e2 e5
27 abc

c(2c a b)e3 e4 ,
2
W3 = (a + b 2c)e1 e2 e4 + (a 2b + c)e1 e3 e5
3 3

+ (2a + b + c)e2 e3 e6 .
103
W2 =
(B.14)
(B.15)
B.3. Sp(4)/(SU(2) U (1))non-max

Metric:

ds 2 = a e1 e1 + e2 e2 + b e3 e3 + e4 e4 + a e5 e5 + e6 e6 .
(B.16)
SU(3)-structure:
J = ae1 e2 + be3 e4 ae5 e6 ,

dJ = (2a + b) e1 e3 e5 + e1 e4 e6 e2 e3 e6 + e2 e4 e5 ,

= a 2 b e1 + ie2 e3 ie4 e5 + ie6 ,

d = 4i a 2 b e1 e2 e3 e4 e1 e2 e5 e6 + e3 e4 e5 e6 .
(B.17)
(B.18)
(B.19)
(B.20)

2i 2a + b
,
3 a2b

4i 1
W2 =
a(a b)e12 2b(b a)e34 + a(a b)e56 .
3 a2b
W1 =
(B.21)
(B.22)
B.4. G2 /SU(3)
Metric:

ds 2 = a e1 e1 + e2 e2 + e3 e3 + e4 e4 + e5 e5 + e6 e6 .
(B.23)
SU(3)-structure:
J = ae1 e2 ae3 e4 ae5 e6 ,

dJ = 2 3a e1 e3 e5 e1 e4 e6 + e2 e3 e6 + e2 e4 e5 ,

= a 3 e1 ie2 e3 + ie4 e5 + ie6 ,

1
d = 8i a 3 e1 e2 e3 e4 + e1 e2 e5 e6 e3 e4 e5 e6 .
3
(B.24)
(B.25)
(B.26)
(B.27)
Non-vanishing intrinsic torsion class:

4i
W1 = .
3a
(B.28)
104
References
[1] P. Candelas, G.T. Horowitz, A. Strominger, E. Witten, Vacuum configurations for superstrings, Nucl. Phys. B 258
(1985) 46.
[2] A. Strominger, Superstrings with torsion, Nucl. Phys. B 274 (1986) 253.
[3] B. de Wit, D.J. Smit, N.D. Hari Dass, Residual supersymmetry of compactified D = 10 supergravity, Nucl. Phys.
B 283 (1987) 165.
[4] M. Dine, R. Rohm, N. Seiberg, E. Witten, Gluino condensation in superstring models, Phys. Lett. B 156 (1985) 55.
[5] J.P. Derendinger, L.E. Ibanez, H.P. Nilles, On the low-energy D = 4, N = 1 supergravity theory extracted from the
D = 10, N = 1 superstring, Phys. Lett. B 155 (1985) 65.
[6] G.L. Cardoso, G. Curio, G. DallAgata, D. Lust, P. Manousselis, G. Zoupanos, Non-Khler string backgrounds and
their five torsion classes, Nucl. Phys. B 652 (2003) 5, hep-th/0211118.
[7] G. Curio, A. Krause, Four-flux and warped heterotic M-theory compactifications, Nucl. Phys. B 602 (2001) 172,
hep-th/0012152.
[8] K. Becker, M. Becker, K. Dasgupta, P.S. Green, Compactifications of heterotic theory on non-Khler complex
manifolds I, JHEP 0304 (2003) 007, hep-th/0301161.
[9] K. Becker, M. Becker, P.S. Green, K. Dasgupta, E. Sharpe, Compactifications of heterotic strings on non-Khler
complex manifolds II, Nucl. Phys. B 678 (2004) 19, hep-th/0310058.
[10] G. DallAgata, N. Prezas, N = 1 geometries for M-theory and type IIA strings with fluxes, Phys. Rev. D 69 (2004)
066004, hep-th/0311146.
[11] K. Behrndt, M. Cvetic, General N = 1 supersymmetric fluxes in massive type IIA string theory, Nucl. Phys. B 708
(2005) 45, hep-th/0407263.
[12] D. Lust, D. Tsimpis, Supersymmetric AdS(4) compactifications of IIA supergravity, JHEP 0502 (2005) 027, hepth/0412250.
[13] S.B. Giddings, S. Kachru, J. Polchinski, Hierarchies from fluxes in string compactifications, Phys. Rev. D 66 (2002)
106006, hep-th/0105097.
[14] G. DallAgata, On supersymmetric solutions of type IIB supergravity with general fluxes, Nucl. Phys. B 695 (2004)
243, hep-th/0403220.
[15] J.P. Gauntlett, D. Martelli, D. Waldram, Superstrings with intrinsic torsion, Phys. Rev. D 69 (2004) 086002, hepth/0302158.
[16] A. Gray, L. Hervella, The sixteen classes of almost hermitian manifolds and their linear invariants, Ann. Math. Pure
Appl. 123 (1980) 35.
[17] G.L. Cardoso, G. Curio, G. DallAgata, D. Lust, Heterotic string theory on non-Khler manifolds with H-flux and
gaugino condensate, Fortschr. Phys. 52 (2004) 483, hep-th/0310021.
[18] A.R. Frey, M. Lippert, AdS strings with torsion: Non-complex heterotic compactifications, hep-th/0507202.
[19] L. Castellani, D. Lust, Superstring compactification on homogeneous coset spaces with torsion, Nucl. Phys. B 296
(1988) 143.
[20] D. Lust, Compactification of ten-dimensional superstring theories over Ricci-flat coset spaces, Nucl. Phys. B 276
(1986) 220.
[21] T.R. Govindarajan, A.S. Joshipura, S.D. Rindani, U. Sarkar, Coset spaces as alternatives to CalabiYau spaces in
the presence of gaugino condensation, Int. J. Mod. Phys. A 2 (1987) 797.
[22] T.R. Govindarajan, A.S. Joshipura, S.D. Rindani, U. Sarkar, Supersymmetric compactification of the heterotic string
on coset spaces, Phys. Rev. Lett. 57 (1986) 2489.
[23] K. Konishi, N. Magnoli, H. Panagopoulos, Spontaneous breaking of local supersymmetry by gravitational instantons, Nucl. Phys. B 309 (1988) 201.
[24] K. Konishi, N. Magnoli, H. Panagopoulos, Generation of mass hierarchies and gravitational instanton induced
supersymmetry breaking, Nucl. Phys. B 323 (1989) 441.
[25] N. Kitazawa, Gravitino condensation in fivebrane backgrounds, Phys. Rev. D 65 (2002) 086004, hep-th/0108232.
[26] M.J. Duff, C.A. Orzalesi, The cosmological constant in spontaneously compactified D = 11 supergravity, Phys.
Lett. B 122 (1983) 37.
[27] J.P. Gauntlett, S. Pakis, The geometry of D = 11 Killing spinors, JHEP 0304 (2003) 039, hep-th/0212008.
[28] S. Kachru, R. Kallosh, A. Linde, S.P. Trivedi, De Sitter vacua in string theory, Phys. Rev. D 68 (2003) 046005,
hep-th/0301240.
[29] G. Curio, A. Krause, G-fluxes and non-perturbative stabilisation of heterotic M-theory, Nucl. Phys. B 643 (2002)
131, hep-th/0108220.
105
[30] C.P. Burgess, R. Kallosh, F. Quevedo, De Sitter string vacua from supersymmetric D-terms, JHEP 0310 (2003) 056,
hep-th/0309187.
[31] D. Lust, G. Zoupanos, Dimensional reduction of ten-dimensional E8 gauge theory over a compact coset space S/R,
Phys. Lett. B 165 (1985) 309.
[32] P. Manousselis, G. Zoupanos, Dimensional reduction of ten-dimensional supersymmetric gauge theories in the
N = 1, D = 4 superfield formalism, JHEP 0411 (2004) 025, hep-ph/0406207.
[33] P. Manousselis, G. Zoupanos, Dimensional reduction over coset spaces and supersymmetry breaking, JHEP 0203
(2002) 002, hep-ph/0111125.
[34] P. Manousselis, G. Zoupanos, Soft supersymmetry breaking due to dimensional reduction over non-symmetric coset
spaces, Phys. Lett. B 518 (2001) 171, hep-ph/0106033.
[35] P. Manousselis, G. Zoupanos, Supersymmetry breaking by dimensional reduction over coset spaces, Phys. Lett.
B 504 (2001) 122, hep-ph/0010141.
[36] E. Bergshoeff, M. de Roo, B. de Wit, P. van Nieuwenhuizen, Ten-dimensional MaxwellEinstein supergravity, its
currents, and the issue of its auxiliary fields, Nucl. Phys. B 195 (1982) 97.
[37] G.F. Chapline, N.S. Manton, Unification of YangMills theory and supergravity in ten-dimensions, Phys. Lett. B 120
(1983) 105.
[38] E.A. Bergshoeff, M. de Roo, The quartic effective action of the heterotic string and supersymmetry, Nucl. Phys.
B 328 (1989) 439.
[39] P. Manousselis, N. Prezas, G. Zoupanos, in preparation.
[40] V.A. Novikov, M.A. Shifman, A.I. Vainshtein, V.I. Zakharov, Instanton effects in supersymmetric theories, Nucl.
Phys. B 229 (1983) 407.
[41] V.A. Novikov, M.A. Shifman, A.I. Vainshtein, M.B. Voloshin, V.I. Zakharov, Supersymmetry transformations of
instantons, Nucl. Phys. B 229 (1983) 394.
[42] P.G. Camara, A. Font, L.E. Ibanez, Fluxes, moduli fixing and MSSM-like vacua in a simple IIA orientifold, hepth/0506066.
[43] G. DallAgata, N. Prezas, ScherkSchwarz reduction of M-theory on G2 -manifolds with fluxes, JHEP 0510 (2005)
103, hep-th/0509052.
[44] M.B. Green, J.H. Schwarz, Anomaly cancellation in supersymmetric D = 10 gauge theory and superstring theory,
Phys. Lett. B 149 (1984) 117.
[45] C.M. Hull, Compactifications of the heterotic superstring, Phys. Lett. B 178 (1986) 357.
[46] C.M. Hull, Anomalies, ambiguities and superstrings, Phys. Lett. B 167 (1986) 51.
[47] P. Forgacs, N.S. Manton, Spacetime symmetries in gauge theories, Commun. Math. Phys. 72 (1980) 15.
[48] Y.A. Kubyshin, I.P. Volobuev, J.M. Mourao, G. Rudolph, Dimensional Reduction of Gauge Theories, Spontaneous
Compactification and Model Building, Lecture Notes in Physics, vol. 349, Springer-Verlag, Berlin, 1989.
[49] D. Kapetanakis, G. Zoupanos, Coset space dimensional reduction of gauge theories, Phys. Rep. 219 (1992) 1.
[50] F.A. Bais, K.J. Barnes, P. Forgacs, G. Zoupanos, Dimensional reduction of gauge theories yielding unified models
spontaneously broken to SU(3) U (1), Nucl. Phys. B 263 (1986) 557.
[51] D. Kapetanakis, G. Zoupanos, Constructing unified models based on E(8) in ten dimensions, Z. Phys. C 56 (1992)
91.
[52] A. Micu, Heterotic compactifications and nearly-Khler manifolds, Phys. Rev. D 70 (2004) 126002, hepth/0409008.
[53] U. Gran, GAMMA: A M ATHEMATICA package for performing gamma-matrix algebra and Fierz transformations
in arbitrary dimensions, hep-th/0105086.
[54] T. House, E. Palti, Effective action of (massive) IIA on manifolds with SU(3) structure, hep-th/0505177.
[55] F. Mueller-Hoissen, R. Stuckl, Coset spaces and ten-dimensional unified theories, Class. Quantum Grav. 5 (1988)
27.
[56] J. Gutowski, S. Ivanov, G. Papadopoulos, Deformations of generalized calibrations and compact non-Khler manifolds with vanishing first Chern class, math.DG/0205012.
Lepton flavour violation in realistic non-minimal

supergravity models
J.G. Hayes, S.F. King, I.N.R. Peddie ,1
School of Physics and Astronomy, University of Southampton, Southampton SO17 1BJ, UK
Received 10 October 2005; received in revised form 23 December 2005; accepted 6 January 2006
Abstract
Realistic effective supergravity models have a variety of sources of lepton flavour violation (LFV) which
can drastically affect the predictions relative to the scenarios usually considered in the literature based on
minimal supergravity and the supersymmetric see-saw mechanism. We investigate a string inspired effective supergravity model arising from intersecting D-branes supplemented by an additional U (1) family
symmetry. In such theories the magnitude of the D-terms is predicted, and we calculate the branching ratios for e and for different benchmark points designed to isolate the different non-minimal
contributions. We find that the D-term contributions are generally dangerously large, but in certain cases
such contributions can lead to a dramatic suppression of LFV rates, for example by cancelling the effect of
the see-saw induced LFV in models with lop-sided textures. In the class of string models considered here we find the surprising result that the D-terms can sometimes serve to restore universality in the
effective non-minimal supergravity theory.
2006 Published by Elsevier B.V.
1. Introduction
The flavour problem has long been a key puzzle for high-energy physics, especially in the
context of supersymmetry (SUSY), where the strong hierarchy in the Yukawa matrices contrasts
strongly with the approximate universality required in the soft mass matrices of the squarks
and the sleptons. Models must now account for massive neutrinos [1,2]. SUSY models with a
E-mail address: inrp@gen.auth.gr (I.N.R. Peddie).

1 Address after 1st September, 2005: Physics Division, School of Technology, Aristotle University of Thessaloniki,
Thessaloniki 54124, Greece.

0550-3213/$ see front matter 2006 Published by Elsevier B.V.
doi:10.1016/j.nuclphysb.2006.01.017
J.G. Hayes et al. / Nuclear Physics B 739 (2006) 106119
107
see-saw explanation for neutrino masses will have lepton flavour violation (LFV) inspired by
renormalisation group running, even if there is none at the GUT scale [3,4]. There is a large
literature on this area [5].
There are two parts to the flavour problem. The first is understanding the origin of the Yukawa
couplings (and heavy Majorana masses for the see-saw mechanism), which lead to low-energy
quark and lepton mixing angles. In low-energy SUSY, we also need to understand why flavour
changing and/or CP violating processes induced by SUSY loops are so small. A theory of
flavour must address both problems simultaneously. For a full discussion of this see the review [6].
Flavour violation will be generated by one-loop SUSY corrections in the MSSM (and all of its
minimal extensions), if the soft trilinear matrices A f or the scalar soft mass matrices m2f are not
diagonal in the basis where the Yukawa matrices are diagonal. This basis is called the SCKMbasis. The experimental constraints in the lepton sector are strong here, and the off diagonal
elements in the SCKM basis must be small.
In models with a family symmetry, we can generate a D-term correction to the squark and
slepton soft masses. The correction for each generation will be dependent on its charge, and
therefore the whole correction will be non-diagonal.
We take a model considered previously [7], which has regions in its parameter space
where various contributions to RG-inspired flavour violation contribute separately. There are
two reasons for wishing to update the results. Firstly, the previous analysis neglected the
D-term corrections to the scalar masses. Secondly, an assumption was made which was inconsistent with the underlying string model, which has been corrected in the current results.
The outline of this paper is as follows. In Section 2 we summarise the relevant details of the
model, and define our conventions. Then in Section 3, we present numerical results for LFV in
five benchmark points of the model, and discuss their implications. We conclude in Section 4.
2. The model
The model is a PatiSalam model with an extra U (1)F family symmetry, as discussed in [7].
The model comes from a brane setup with two intersecting D5-branes, 51 and 52 . To summarise
1, 2)
the field notation, the matter content is unified into the representations F (4, 2, 1) and F (4,
of the PatiSalam group SU(4) SU(2)L SU(2)R . The MSSM Higgs doublets are unified into
1, 2).
h(1, 2, 2). The symmetry is broken by two GUT scale vevs of the fields H (4, 1, 2) and H (4,
The family symmetry is broken by two fields and which carry charges of +1, 1 under
U (1)F , respectively. The charges of the field F i are qi and the charges of F i are qi .
The soft supersymmetry breaking terms can be derived from the string model, as described
in [7]. The only difference relates to the F-terms F , FH and FH . These F-terms depend on the
vevs of the moduli fields S and Ti , which are related to the brane gauge couplings, g9 and g5i .
Previously, we had assumed that we could set the unseen brane couplings g9 and g53 as we liked.
However, all of the couplings are related by the equation

M 2
= g9 g51 g52 g53 .
32 2
(1)
MPl
Previously, we assumed that g9 = g53 = g52 , and took M to be at the supersymmetric GUT
scale, which disagrees with Eq. (1). Instead we take g9 = g53 , which makes Re(S) an order of
magnitude smaller.
108
The string assignments of the fields Fi , Fj , h, H , H , , lead to unique results for the
soft SUSY breaking parameters in terms of the vevs of the auxiliary fields. In our notation, the
auxiliary fields are parameterised as:
S,
FS = 3m3/2 (S + S)X
(2)
FTi = 3m3/2 (Ti + Ti )XTi ,

(3)
1/2
FH = 3m3/2 H (S + S)
(4)
XH ,
FH = 3m3/2 H (T3 + T3 )1/2 XH ,

(5)
1/4
1/4
F = 3m3/2 (S + S) (T3 + T3 ) X .
(6)
2
The parameters Xi are Goldstino parameters, and we require i Xi = 1. The gravitino mass,
m3/2 is also a free parameter. The moduli vevs are determined from the brane couplings in the
usual way [8]. The other exotic vevs are all equal H = H = = M , where is Wolfenstein
parameter, and M is the string scale, taken here to be the SUSY GUT scale.
2.1. Soft scalar masses
The SUGRA contribution to the scalar mass matrices gives a unified soft matrix, m2L for the
left-handed matter contained in F , and a unified m2R for the right-handed matter contained in F .
It also gives mass m2h for hd and hu :

3
3
m2L = m23/2 Diag 1 XS2 + XT23 , 1 XS2 + XT23 , 1 3XT23 ,
(7)
2
2

3 2
3 2
2
2
2
2
2
mR = m3/2 Diag 1 XS + XT3 , 1 XS + XT3 , 1 3XT2 ,
(8)
2
2

m2h = m23/2 1 3XS2 .
(9)
There are two D-term contributions to the scalar masses. The first is the contribution from
the breaking of the PatiSalam group to the MSSM group [9]. However, the result is modified
when a U (1)F D-term is also switched on, due to the U (1)F charges of H , H being non-zero.
The second D-term comes solely from the breaking of the U (1)F . The corrections lead to the
following masses:
2

2
mQL ij = m2L ij + g42 DH
(10)
ij + gF2 D2 qi ij ,
2

2
2
2
2
mU R ij = mR ij g4 2g2R
(11)
DH
ij + gF2 D2 qi ij ,
2

2
2
2
2
mDR ij = mR ij g4 + 2g2R
(12)
DH
ij + gF2 D2 qi ij ,
2
2
2 2
mLL ij = mL ij 3g4 DH ij + gF2 D2 qi ij ,
(13)
2

2
2
2
2
mER ij = mR ij + 3g4 2g2R
(14)
DH
ij + gF2 D2 qi ij ,
2

2
2
2
2
2 2
mN R ij = mR ij + 3g4 + 2g2R DH ij + gF D qi ij ,
(15)
2
2
m2hu = m2h 2g2R
DH
,
(16)
2
2
m2hd = m2h + 2g2R
DH
.
(17)
109
2 and D 2 are:
The expressions DH
2
DH
=
D2 =
1
2 + 6g 2
4g2R
4
m2 m2
.
2gF2
2

mH m2H + qH m2 m2 ,
(18)
(19)
The second term in Eq. (18) is new. This is a contribution from the breaking of the U (1)F
symmetry, and the new contribution vanishes when D2 = 0. Since SUGRA predicts the soft
masses of H , H , and , the D-terms are predicted, as a function of m3/2 and the Xi .
With a family PatiSalam model, it is possible to pick any charges for qi , qi , and there will
be an anomaly free set of charges that lead to the same Yukawa texture.2 The charges appearing
in Eqs. (10)(15) are the anomaly free charges.
The gauge couplings and mass parameters in Eqs. (18), (19) are predicted from the model, in
terms of the X parameters and m3/2 . The D-terms will be zero if XS = XTi , or if the brane
assignment is the same as . Choosing the second of these conditions is useful since it gives a
comparison case where there are no U (1)F D-terms.
2.2. Other soft terms
The soft gaugino masses and the soft trilinear couplings are the same as in [7].

3m3/2
M3 =
(T1 + T1 )XT1 + (T2 + T2 )XT2 ,
(T1 + T1 ) + (T2 + T2 )
M2 = 3m3/2 XT1 ,

3m3/2
5
2
(T1 + T1 )XT1 + (T2 + T2 )XT2 .

M1 = 5
2
3
3
3 (T1 + T1 ) + 3 (T2 + T2 )
(20)
(21)
(22)
The values of T1 + T1 and T2 + T2 are proportional to the brane gauge couplings g51 and g52 ,
which are related in a simple way to the MSSM couplings at the unification scale. This is discussed in [11].
The contributions to the soft masses and gaugino masses from the FN and heavy Higgs auxiliary fields is completely negligible due to the small size of their F-terms. However for the soft
trilinear masses these contributions are of order O(m3/2 ) despite having small F-terms [12]:
d1 + dH + p(i, j )d d1 + dH + p(i, j )d d2 + dH + p(i, j )d
A = 3m3/2 d1 + dH + p(i, j )d d1 + dH + p(i, j )d d2 + dH + p(i, j )d ,

d3 + dH + p(i, j )d
d3 + dH + p(i, j )d
d4
(23)
where
p(i, j ) |qi + qj + qh |,
(24)
d1 = XS XT1 XT2 ,
(25)
2 This is discussed in more detail in, e.g. [10].
110
1
1
d2 = XS XT1 XT2 ,
2
2
1
1
d3 = XS XT1 XT2 + XT3 ,
2
2
d4 = XT1 ,
dH = (S + S)
1/2
XH + (T3 + T3 )
1/2
(26)
(27)
(28)
XH ,
(29)
1/4 (T3 + T3 )1/4 X .

d = (S + S)
(30)
These results are independent of which string assignment we give to .

3. Results
In order to reduce the large dimension of the parameter space, from this point in we shall
assume that the Goldstino parameters XTi are degenerate:
XT1 = XT2 = XT3 = XT .
(31)
3.1. Two models

We will study two models differ only in the U (1)F charges. The anomaly-free [10] family
charges are laid out in Table 1. This first model is almost the model studied in [7,13], but with an
e and Y e . The second model is defined to have the same Yukawas
operator to allow non-zero Y12
13
as the first model, but with the charges of left-handed matter are the same. This caused the U (1)F
D-term to left-handed scalar mass matrices to not lead to extra flavour violation. This choice is
a small change to the model, since two of the left-handed charges are the same anyway, and is
made because normally the left-handed contribution dominates over the right-handed contribution. The family charges for both models are laid out in Table 1.
The numerical values of the Yukawa matrices are given by3 :
Y u (MX ) =
2.159 1006
5.606 1004
5.090 1003
,
0.000
1.105 1003
0.000
03
01
0.000
6.733 10
5.841 10
04
04
1.661 10
5.606 10
1.018 1002
Y d (MX ) = 7.683 1004 5.343 1003 1.216 1002 ,
1.769 1004 3.133 1002 3.933 1001
1.246 1004
0.000
0.000
Y e (MX ) = 1.537 1003 2.432 1002 3.649 1002 ,
1.327 1004
2.159 1006
Y (MX ) =
0.000
0.000
3.133 1002
1.525 1003
8.290 1004
5.050 1003
(32)
(33)
(34)
5.469 1001
0.000
3.923 1001 .
5.469 1001
3 But note that in part of the later analysis, we will set Y e = 0 and Y e = 0.
12
13
(35)
111
Table 1
The family charges for Model 1 and Model 2
Field
Model 1 charge
F1
11
6
5
6
5
6
19
6
7
6
56
56
5
6
F2
F3
F1
F2
F3
H
H
Model 2 charge
1
1
1
3
1
1
1
1
Table 2
Values of the X parameters for the five benchmark points, AE
Point
XS
XT
XH
XH
A
B
C
D
E
0.500
0.535
0.270
0.270
0.290
0.500
0.488
0.270
0.270
0.264
0.000
0.000
0.000
0.595
0.000
0.000
0.000
0.000
0.595
0.000
0.000
0.000
0.841
0.000
0.841
0.000
0.000
0.000
0.000
0.000
The RH Majorana neutrino mass matrix for Models 1 and 2 has the numerical values:
3.508 108 3.686 109 3.345 1011
9
8.313 1010 5.886 1012
MRR (MX ) = 3.686 10
.
11
12
14
3.345 10
5.886 10
5.795 10
(36)
3.2. Benchmark points

We consider here again the same benchmark points, AD as considered previously [7]. Benchmark point A corresponds to a mSUGRA scenario, with universal scalar mass matrices and
aligned trilinear matrices at the GUT scale. Point B corresponds to F-term non-universality in
the scalars and non-alignment in the trilinears. Points C and D correspond to non-aligned trilinears dominated by the F-terms of the and H fields, respectively. The U (1)F D-terms are
only non-zero at benchmark point B. We are motivated to introduce a new benchmark point, E,
where F dominated non-aligned trilinears and with U (1)F D-terms are switched on together.
This point allows us to investigate two different effects from the U (1)F sector of the models. The
values of the X parameters for these five benchmark points are laid out in Table 2.
e and Y e
3.3. Varying Y12
13
Normally, the chargino contribution to LFV dominates. Since the Feynman diagram for this
includes the left-handed sfermions, we would expect the D-term corrections to the left-handed
slepton mass matrix to dominate the flavour violation. However, Model 2 is set up to have universal left-handed charges, so the left-handed D-term correction from the breaking of U (1)F will not
112
contribute to flavour violation except by modifying slightly the mass suppression. The D-term is
limited in magnitude by the difference of m2 and m2 , and although this is not a strong correction
to the soft masses, it can contribute significantly to the lepton flavour violating branching ratios.
e = Y e = 0.
The difference in e between Model 1 and Model 2 is negligible for Y12
13
e
e
This should not be surprising, since the texture zero coming from Y12 = Y13 = 0 will yield small
mixing angles, resulting in small lepton flavour violation.
In order to get a picture of how great an effect the D-term contributions could have on the soft
e and Y e .
masses, it is necessary to have non-zero Y12
13
3.4. Varying brane assignments for
The field is fixed to reside on the C 51 52 brane, but we allow the brane assignment of to
vary over the possibilities C 51 52 , C151 , C251 and C351 . This gives us D-terms that are calculable in
each case, rather than being free parameters. The assignment of to C 51 52 , which is the same
assignment as the field ensures a zero U (1)F D-term. The other possibilities, will highlight the
contribution of the D-terms to lepton flavour violation.
3.5. Numerical procedure
The code used to generate all the data here is a modified version of SOFTSUSY [14]. The
changes were to introduce three right-handed neutrino fields, and apply a see-saw mechanism to
generate left-handed Majorana masses and mixing angles.
Electroweak symmetry breaking provides a significant constraint on the results. The breakdown of electroweak symmetry breaking was responsible for the spike feature that was shown
in the plots for benchmark points A and B in [7]. For the data above the spike, radiative electroweak symmetry breaking does not work properly, as the Z-boson mass becomes tachyonic.
In the present paper such bad regions where electroweak symmetry breaking fails are cut-off,
however there is still a remnant of the spike left, which is why one can see a slight rise at the
ends of the plots for our benchmark points A and B, as can be seen in Section 3.6.
3.6. Numerical results
We have now defined our two models, Model 1 and Model 2, and a set of five benchmark
points in Table 2 to examine within them. We have also set up what we will be varying apart
e , Y e and the string assignment of
from the gravitino mass in these modelsthe values of Y12
13
which gives different D-terms. We are now in a position to present our results. We shall focus
on the branching ratios for e and . In the following plots, we do not consider
the assignment to C351 as this is exactly the same as C251 , due to our choice to make the XTi
degenerate. Were we to allow the XTi to be non-degenerate, the phenomenological results of
assigning to C251 and C351 would not be the same. The detailed spectrum will look different at
each parameter point, but the general trend is for the physical masses to increase in magnitude
as the gravitino mass increases. Thus too-high gravitino masses will start to reintroduce the finetuning problem resulting from the gluino mass being too high [15].
Fig. 1 shows numerical results for BR( e ) for Model 1, plotted against the gravitino
mass m3/2 , where each of the four panels (i)(iv) correspond to each of the four benchmark
113
e = Y e = 0. Panel (i) corresponds to benchFig. 1. Plots showing the branching ratio for e for Model 1 with Y12
13
mark point A, panel (ii) is for B, panel (iii) is for C and panel (iv) is for D. The assignments are shown with the
5
5
separate lines: C 51 52 (solid), C1 1 (dashed), and C2 1 (dotdash). The solid curve corresponds to zero D-terms, and
the other curves correspond to different models for the D-terms. The 2002 experimental limit [16] is also given by the
horizontal line.
points AD.4 Panel (i) of Fig. 1 refers to benchmark point A. Panel (ii) of Fig. 1 refers to benchmark point B, In this case one can clearly distinguish the additional contributions to LFV arising
from the D-terms. This makes benchmark point B the most phenomenologically interesting for
the purposes of this study. The differing contributions stem from the string assignments, which
are shown by the separate lines: C 51 52 (solid), C151 (dashed), and C251 (dotdash). The C 51 52
case shows the zero-D-term limit. The other two locations for then turn on the D -term contributions. With the D-terms switched on, Model 1 is experimentally ruled out over all parameter
space shown here. Panel (iii) of Fig. 1, referring to benchmark point C, is the FroggattNielsen
benchmark point, and for this case we see that the experimental limit is satisfied for m3/2 over
1400 GeV. Panel (iv) of Fig. 1, benchmark point D, shows the heavy Higgs point, for which the
experimental limit is satisfied everywhere above 800 GeV.
Fig. 2 shows analogous results for BR( ) for Model 1, plotted against the gravitino
mass m3/2 . All benchmark points come below the experimental limit for a substantial amount of
the parameter space. The experimental limit here is more recent, and subsequently much more
stringent than the previous limit. For these models in which there is a large (2, 3) element in the
neutrino Yukawa matrix the branching ratio for is essentially as constraining as that for
e , as first pointed out in [18]. The D-term coupling to right-handed scalars has a Yukawa
mixing angle of order 3 for BR( e ), compared to 2 for e . So the right-handed
sector is of equal importance to the left-handed sector. We note that the see-saw effect enters
4 The solid line on each plot in Fig. 1 corresponds to the solid lines in Fig. 1 of [7].
114
e = Y e = 0. Panel (i) corresponds to benchFig. 2. Plots showing the branching ratio for for Model 1 with Y12
13
mark point A, panel (ii) is for B, panel (iii) is for C and panel (iv) is for D. The assignments are shown with the
5
5
separate lines: C 51 52 (solid), C1 1 (dashed), and C2 1 (dotdash). The solid curve corresponds to zero D-terms, and
the other curves correspond to different models for the D-terms. The 2005 experimental limit [17] is also given by the
horizontal line.
prominently in the left-handed sector, and by considering Eqs. (10) and (13) for the soft scalar
masses in the (23) sector for , one can show that there is little effect coming from the Dterms coupling to left-handed scalars, since we have universal family charges for the left-handed
(23) sector, qL2 = qL3 .
The solid line in panel (ii) of Fig. 2 for the C 51 52 string assignment of has zero contribution from the U (1)F D-terms, and shows just the effect of non-minimal SUGRA. This actually
suppresses the flavour violation arising from the see-saw effect alone, showing an interesting
cancellation between the LFV from the see-saw mechanism and the LFV from the non-universal
D-terms. On the other hand the dashed line in panel (ii) for the C151 case is very similar to the
see-saw scenario of benchmark point A shown in panel (i). This is due to the D-terms actually
conspiring to restore universality in the scalar masses, turning non-minimal SUGRA back into
the minimal form. It is an amazing consequence of this string assignment for that in this model
the effects of the non-universal U (1)F D-terms can exactly cancel the effects of the non-universal
SUGRA for the branching ratio of , leading to universal scalar mass matrices, even with
SUGRA turned on. This is a string effect that directly affects the amount of flavour violation
predicted in this scenario. For the C251 case shown by the dotdash line, applying D-terms in
this case actually enhance the effect of non-minimal SUGRA, causing the scalar mass matrices to become even more non-universal. One can understand this purely right-handed effect by
considering the different mass insertion diagrams for the left- and right-handed sectors. The lefthanded sector involves charginos, whereas the right-handed sector only involves neutralinos, so
the right-handed masses scale differently with the gravitino mass as compared to the left-handed
masses. Thus we do not have a universal mass scaling between the left- and right-handed sectors.
115
e =0
Fig. 3. Plots showing the branching ratio for e for benchmark point B for Model 1 only. Panel (i) has Y12
e = 0. Panel (ii) has Y e = 1.5 103 and Y e = 0. Panel (iii) has Y e = 0 and Y e = 1.5 102 . Panel (iv)
and Y13
12
13
12
13
5
has Y e = 1.5 103 and Y e = 1.5 102 . The assignments are shown with the separate lines: C 51 52 (solid), C 1
12
13
(dashed), and C2 1 (dotdash). The solid curve corresponds to zero D-terms, and the other curves correspond to different
models for the D-terms. The 2002 experimental limit [16] is also given by the horizontal line.
This leads to the observed smooth cancellation of flavour violation between the two competing
contributions from the see-saw mechanism and from SUGRA with additional U (1)F D-terms.
e and Y e electron
Fig. 3 shows benchmark point B for Model 1. The four panels show the Y12
13
e
e = 0.
Yukawa elements being turned on and off. The results for Fig. 3(i) are for Y12 = 0 and Y13
This is the same as in panel (ii) of Fig. 1, and is the base from which we start. Panel (ii) of
e = 1.5 103 and Y e = 0, so we can clearly see the effect of turning Y e on.
Fig. 3 has Y12
13
12
It only affects the C 51 52 line, as the D-terms dominate over this effect for the other two string
e = 0 and Y e = 1.5 102 , highlighting the effect of
assignments. Panel (iii) of Fig. 3 uses Y12
13
e alone. Again the zero D-term line of C 51 52 is the only one that is sizably affected by
just Y13
this change in Yukawa texture. Panel (iv) of Fig. 3 shows the effect of turning on both Yukawa
e = 1.5 103 and Y e = 1.5 102 . We see that the shape of the solid line is
elements: Y12
13
determined by both Yukawa texturesthey seem to have an equal impact on it.
Fig. 4 shows benchmark point E, which combines the features of benchmark points B and
C, thus both U (1) D-term and FroggattNielsen (FN) flavour violation appear in the predicted
branching ratios shown in this figure. There is some interesting interplay between the FN F-terms
in the benchmark point C region of parameter space, and the D-terms in benchmark point B,
and benchmark point E is designed to show this difference. In Fig. 4 the results are shown for
e and Y e Yukawa elements on and off. The shape of the curves are as in
Model 1 with both Y12
13
benchmark point C due to X being turned on, and the results are numerically similar to those
of benchmark point C, showing that the dominant contribution comes from FroggattNielsen
flavour violation. However, as in benchmark point B, the different D-terms corresponding to
116
e =0
Fig. 4. Plots showing the branching ratio for e for benchmark point E for Model 1 only. Panel (i) has Y12
and Y13
12
13
12
13
5
12
13
e = 0 and
different assignments leads to noticeable shifts in the results. In panel (i) with Y12
e
Y13 = 0 it is seen that the presence of non-zero D-terms actually reduces the LFV rate somewhat
compared to the solid curve with zero D-terms, corresponding to a region of parameter space
where there is some cancellation between the flavour violation from the FroggattNielsen fields
and that caused by the U (1)F D-terms. The other panels show variation of the Yukawa elements
e and Y e Yukawa elements being
as in Fig. 3, for example panel (iv) corresponds to both Y12
13
non-zero. Note that the solid curve in panel (iv) is slightly lower than the solid curve in panel (ii),
showing that sometimes a non-zero Yukawa coupling can reduce LFV.
Fig. 5 shows the effects of the Yukawa elements for benchmark point E using Model 2, where
Model 2 has the same Yukawa structure as Model 1, but has the feature that the left-handed
family charges are the same for all three families, resulting in universal D-terms, at least in the
left-handed sector. Comparing Fig. 5 to Fig. 4, we see that in panel (i) of both figures with
e = 0 and Y e = 0 there is no observable difference between the predictions of the two models.
Y12
13
e and Y e
However comparing panels (iv) of Fig. 5 and Fig. 4 we see that with non-zero Y12
13
Model 2 has the effect of reducing the LFV resulting from the D-terms.
4. Conclusions
We have performed a numerical analysis of LFV in a string-inspired PatiSalam model enhanced with a U (1) family symmetry. This model allows for two distinct sources of flavour
violation from the family symmetry. The first comes from the F-terms associated with the
Higgs which breaks a family symmetry, and the second is from the D-terms that contribute
117
e =0
Fig. 5. Plots showing the branching ratio for e for benchmark point E for Model 2 only. Panel (i) has Y12
and Y13
12
13
12
13
5
12
13
when the family symmetry is spontaneously broken. The model in question has points in
it parameter space where these effects occur separately, together and not at all; this allows
an investigation of the interplay, and an understanding of the significance of their contributions.
The first conclusion is the confirmation of how dangerously large the calculable D-term contribution to flavour violation can be. However it should be emphasised that while the D-terms
are calculable in these models they are also model dependent, and it is always possible to simply
switch off the D-terms by selecting the field to have the same string assignment as the intersection state. However other choices will lead to non-zero but calculable D-terms, which can
be dangerously large, or can massively suppress flavour violation. For example the curves with
non-zero D-terms in Fig. 3 all exceed the experimental limit for the branching ratio for e ,
showing that D-term effects have the potential to greatly exceed the other contributions from
SUGRA, the see-saw mechanism, FN and Higgs, depending on the choice of Yukawa textures.
However in some cases the D-terms generated by breaking the U (1)F family symmetry can also
suppress the branching ratio for e , as shown in panels (i) and (iii) of Figs. 4 and 5.
Another notable feature is the effect of Yukawa texture on the results. The Yukawa texture has
e = Y e = 0, and we have shown that turning on non-zero values of these Yukawa couplings
Y12
13
can greatly enhance the branching ratio for e almost arbitrarily. The reason is that the
rotations to the SCKM basis are controlled by these Yukawa elements and the larger these rotations the larger will be the off-diagonal soft masses in the SCKM basis. The non-zero magnitudes
e and Y e were therefore chosen to be large enough to show the variations in the branchof Y12
13
118
ing ratios of the different models, but small enough to keep within the currently experimentally
allowed range.
In this paper we have worked with a particular Yukawa texture in which there is a large (2, 3)
element in the neutrino Yukawa matrix leading to large see-saw induced LFV and a branching
ratio for which is as constraining as that for e [18]. However we have seen that in
some cases the D-terms can lead to a large suppression of the rate for particular values of m3/2 ,
as seen in panel (ii) of Fig. 2. For other cases the effects of the non-universal U (1)F D-terms can
exactly cancel the effects of the non-universal SUGRA model leading to universal scalar mass
matrices, thereby restoring universality even for a non-minimal SUGRA model. Such effects are
only possible in certain string set-ups and thus LFV provides an observable signal which may
discriminate between different underlying string models.
In conclusion, we have seen that within realistic non-minimal supergravity models there can
be several important effects leading to much larger LFV than in the case usually considered
in the literature of minimum flavour violation corresponding to just mSUGRA and the see-saw
mechanism, and considered here as benchmark point A. We find that the D-term contributions
are generally dangerously large, but in certain cases such contributions can lead to a dramatic
suppression of LFV rates, for example by cancelling the effect of the see-saw induced LFV in
models with lop-sided textures. In the class of string models considered here we find
the surprising result that the D-terms can sometimes serve to restore universality in the effective
non-minimal supergravity theory. Thus D-terms can give very large and very surprising effects
in LFV processes.
In general there will be a panoply of different sources of LFV in realistic non-minimal
SUGRA models, and we have explored the relative importance of some of them within a particular framework. The results here only serve to heighten the expectation that LFV processes such
as e and may be observed soon, although it is clear from our results that the
precise theoretical interpretation of such signals will be more non-trivial than is apparent from
many previous studies in the literature.
Acknowledgements
J.H. thanks PPARC for a studentship.
References
[1] G.L. Fogli, E. Lisi, A. Marrone, D. Montanino, A. Palazzo, A.M. Rotunno, hep-ph/0212127;
P.C. de Holanda, A.Y. Smirnov, hep-ph/0212270;
V. Barger, D. Marfatia, Phys. Lett. B 555 (2003) 144, hep-ph/0212126;
A. Bandyopadhyay, S. Choubey, R. Gandhi, S. Goswami, D.P. Roy, hep-ph/0212146;
M. Maltoni, T. Schwetz, J.W. Valle, hep-ph/0212129.
[2] Y. Fukuda, et al., Super-Kamiokande Collaboration, Phys. Lett. B 433 (1998) 9;
Y. Fukuda, et al., Super-Kamiokande Collaboration, Phys. Lett. B 436 (1998) 33;
Y. Fukuda, et al., Super-Kamiokande Collaboration, Phys. Rev. Lett. 81 (1998) 1562.
[3] J. Hisano, T. Moroi, K. Tobe, M. Yamaguchi, Phys. Rev. D 53 (1996) 2442, hep-ph/9510309.
[4] S.F. King, M. Oliveira, Phys. Rev. D 60 (1999) 035003, hep-ph/9804283.
[5] J. Hisano, T. Moroi, K. Tobe, M. Yamaguchi, Phys. Lett. B 391 (1997) 341;
J. Hisano, T. Moroi, K. Tobe, M. Yamaguchi, Phys. Lett. B 397 (1997) 357, Erratum;
J. Hisano, D. Nomura, Y. Okada, Y. Shimizu, M. Tanaka, Phys. Rev. D 58 (1998) 116010;
J. Hisano, D. Nomura, T. Yanagida, Phys. Lett. B 437 (1998) 351;
J. Hisano, D. Nomura, Phys. Rev. D 59 (1999) 116005;
S. Davidson, A. Ibarra, JHEP 0109 (2001) 013;
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
119
S. Davidson, A. Ibarra, Phys. Lett. B 459 (1999) 171;

M.E. Gomez, G.K. Leontaris, S. Lola, J.D. Vergados, Phys. Rev. D 59 (1999) 116009;
J.R. Ellis, M.E. Gomez, G.K. Leontaris, S. Lola, D.V. Nanopoulos, Eur. Phys. J. C 14 (2000) 319;
W. Buchmller, D. Delepine, L.T. Handoko, Nucl. Phys. B 576 (2000) 445;
D. Carvalho, J. Ellis, M. Gomez, S. Lola, Nucl. Phys. B 618 (2001) 171;
W. Buchmller, D. Delepine, F. Vissani, Phys. Lett. B 515 (2001) 323;
F. Deppisch, H. Pas, A. Redelbach, R. Ruckl, Y. Shimizu, hep-ph/0206122;
J. Sato, K. Tobe, Phys. Rev. D 63 (2001) 116010;
J. Hisano, K. Tobe, Phys. Lett. B 510 (2001) 197;
S. Lavignac, I. Masina, C.A. Savoy, Phys. Lett. B 520 (2001) 269, hep-ph/0106245;
J. Hisano, hep-ph/0204100;
J.A. Casas, A. Ibarra, J.R. Ellis, J. Hisano, M. Raidal, Y. Shimizu, Phys. Lett. B 528 (2002) 86, hep-ph/0111324;
J.R. Ellis, J. Hisano, S. Lola, M. Raidal, Nucl. Phys. B 621 (2002) 208, hep-ph/0109125;
S. Lavignac, I. Masina, C.A. Savoy, Nucl. Phys. B 633 (2002) 139, hep-ph/0202086;
I. Masina, C.A. Savoy, hep-ph/0211283;
S. Pascoli, S.T. Petcov, C.E. Yaguna, Phys. Lett. B 564 (2003) 241, hep-ph/0301095;
S. Pascoli, S.T. Petcov, W. Rodejohann, hep-ph/0302054;
A. Masiero, S.K. Vempati, O. Vives, New J. Phys. 6 (2004) 202, hep-ph/0407325.
D.J.H. Chung, L.L. Everett, G.L. Kane, S.F. King, J. Lykken, L.T. Wang, Phys. Rep. 407 (2005) 1, hep-ph/0312378.
S.F. King, I.N.R. Peddie, Nucl. Phys. B 678 (2004) 339, hep-ph/0307091.
A. Brignole, L.E. Ibanez, C. Munoz, hep-ph/9707209.
S.F. King, M. Oliveira, Phys. Rev. D 63 (2001) 015010, hep-ph/0008183.
G.L. Kane, S.F. King, I.N.R. Peddie, L. Velasco-Sevilla, hep-ph/0504038.
L.L. Everett, G.L. Kane, S.F. King, S. Rigolin, L.T. Wang, Phys. Lett. B 531 (2002) 263, hep-ph/0202100.
S. Abel, S. Khalil, O. Lebedev, Phys. Rev. Lett. 89 (2002) 121601, hep-ph/0112260.
T. Blazek, S.F. King, J.K. Parry, hep-ph/0303192.
B.C. Allanach, Comput. Phys. Commun. 143 (2002) 305, hep-ph/0104145.
G.L. Kane, S.F. King, Phys. Lett. B 451 (1999) 113, hep-ph/9810374.
K. Hagiwara, et al., Particle Data Group Collaboration, Phys. Rev. D 66 (2002) 010001.
B. Aubert, et al., BaBar Collaboration, Phys. Rev. Lett. 95 (2005) 041802, hep-ex/0502032.
T. Blazek, S.F. King, hep-ph/0211368.
Gauge fixing in causal dynamical triangulations

Fotini Markopoulou a,b , Lee Smolin a,b,
a Perimeter Institute for Theoretical Physics, 35 King Street North, Waterloo, Ontario N2J 2W9, Canada
b Department of Physics, University of Waterloo, Waterloo, Ontario N2L 3G1, Canada
Received 14 November 2005; received in revised form 8 January 2006; accepted 9 January 2006
Abstract
We relax the definition of the AmbjrnLoll causal dynamical triangulation model in 1 + 1 dimensions
to allow for a varying lapse. We show that, as long as the spatially averaged lapse is constant in time, the
physical observables are unchanged in the continuum limit. This supports the claim that the time slicing of
the model is the result of a gauge fixing, rather than a physical preferred time slicing.
1. Introduction
A widely accepted strategy to construct a quantum theory of gravity is to define the theory by
means of a background independent path-integral, in which one sums over microscopic spacetime histories, each weighed by a quantum amplitude. Such a theory can be a candidate for a
fundamental theory of spacetime because it takes over into the quantum realm the most basic,
and experimentally confirmed, principle of general relativity, namely, that the geometry of space
and time is entirely dynamical. Approaches of this kind include causal sets, loop quantum gravity, spin foam models, Regge calculus and dynamical triangulations. All path-integral approaches
have the same ultimate aim, to derive general relativity as a low energy, coarse grained approximation to the fundamental discrete and quantum description. This is still an open problem and
we will call it the low energy problem.
In almost all cases, the sum-over-histories is a formal path-integral. At the mathematical level,
that path integral suffers from being ill-defined, and in several ways. At the practical level, it is
very complicated, making the low energy problem a formidable one. It is no less obscure at the
E-mail addresses: fotini@perimeterinstitute.ca (F. Markopoulou), lsmolin@perimeterinstitute.ca (L. Smolin).

doi:10.1016/j.nuclphysb.2006.01.009
F. Markopoulou, L. Smolin / Nuclear Physics B 739 (2006) 120130
121
physical level, where its formal interpretation as quantum superpositions of spacetimes needs to
be understood and related to observations.
The overall motivation for the present work is to investigate a basic questions raised by sumover-histories quantum gravity: which, if any, geometric features are shared between the quantum
histories, the paths, and the classical spacetime geometry that is to emerge in the low energy
limit? In particular, what is the relationship between the fundamental microscopic causal structure and the macroscopic causal structure that emerges in the low energy limit? Similarly, what
is the relationship between microscopic and macroscopic locality?
These questions can only be investigated in the context of a path-integral model where explicit
results on the low energy limit can be calculated. Of all the different approaches, the one that has
been most successful at addressing the low-energy problem is Causal Dynamical Triangulations
(CDT). This is a non-perturbative gravitational path-integral for Lorentzian spacetimes, proposed
by Ambjrn and Loll in [2]. The histories are described, in the presence of an ultraviolet regulator, as , as a triangulated Lorentzian manifold. Each spacetime history is built by pasting together
fundamental building blocks with timelike and spacelike faces whose interiors are isomorphic to
subsets of Minkowski spacetime. Each history has a causal structure, which means that the histories are restricted to contain arrangements of simplices whose spacelike faces form a sequence
of spatial slices, the triangulated equivalent of a globally hyperbolic spacetime with equal-time
slices.
In 1 + 1 dimensions, causal dynamical triangulations have been shown, both analytically and
numerically, to have a continuum limit when as 0 which describes physics in an effective 1 + 1
geometry, which is large compared to the Planck scale. One infers this from measurements of
the Hausdorff dimension, which is near 2 (instead of the problematic 4 of Euclidean dynamical
triangulations [1]). The good behavior is maintained when matter is added, and it is even possible to exceed the barrier of c = 1 that prevented the Euclidean versions of such models from
providing a non-perturbative approach to string theory [3]. Recently published numerical results
in 3 + 1 dimensions provide evidence for a continuum limit which consists of universes which
are (3 + 1)-dimensional, whose spatial slices have volumes which are large and slowly varying,
and whose spatial geometries have Hausdorff dimension dh 3 [4].
Compared to other path-integral approaches, causal dynamical triangulations have three distinct features:
(1) Each history is a fixed foliation of spacelike surfaces and thus has topology R, with
a fixed spatial topology.
(2) Since each triangle has fixed spatial edge length as and fixed timelike edge
length at , the
a2
fixed foliation implies that the lapses are fixed to be a constant, N0 = at2 4s , in the
(1 + 1)-dimensional model. The same is true in higher dimensions.
(3) The limit as at 0 is taken, so that the discreteness scale is considered an ultraviolet
cutoff to be removed rather than a fixed physical scale.
To understand the results of CDT models, and compare them to results that may be expected
from other approaches, it is essential to know the extent to which these three features are necessary conditions for the good results obtained. In particular, we would like to know to what
extend the CDT histories have a geometrical interpretation. The model can be understood as a
statistical model of random surfaces, and the inventors of causal dynamical triangulations warn
their readers not to think of the individual histories as spacetimes. However, if we strictly follow
122
their advice, then in what sense should the constraints they impose on the random surfaces that
are allowed to enter the sum be interpreted as causality? Can different causality conditions be
imposed that lead to the same results?
The specific task for this paper is the following: Given a global foliation, and hence a R
topology, the choice of equal lapses can be conjectured to be a gauge condition. If this is the
case it should be possible to show explicitly that the results of CDT are preserved if the gauge
conditions is modified.
We will work in the 1 + 1 case, and show that the same continuum limit results are obtained if
the lapses of the simplices are allowed to vary over positive values, subject only to two conditions.
(1) The average lapse in each time slice is fixed. (2) Two edges of each triangle remain timelike.
This applies both to the pure gravity model and to gravity coupled to matter. This supports the
claim that, once the R topology is fixed, the condition of fixed lapses is just a gauge condition
that can be modified without affecting the physical results.
In a related paper [10], Konopka shows that similar results can be obtained for the 2 + 1 case.
Together with the results presented here, this gives strong support for the claim that the restriction
to constant lapse in CDT models can be considered as a gauge condition.
One of the main issues with the preferred foliation of this model is the implication of nonlocality in the causal evolution. The rule that constructs the histories is non-local because the
spacelike slices that define the preferred foliation are first fixed, after which one sums over all
triangulations that connect them, subject to the condition that two edges of each triangle remain
timelike. This can be contrasted with other causal approaches to the path integral in quantum
gravity, such as causal spin networks [5] or percolation causal sets [7]. There, histories are constructed by local evolution rules, meaning that the rule for adding a triangle and computing its
amplitude depends only on its causal past. The existence of the global foliation imposed in CDT
models requires either the application of a non-local rule, or the truncation of the ensemble of
histories constructed by local rules.
Our results indicate that the non-locality is still present even when the lapse gauge is changed,
since it is required to keep the average lapse constant. The class of histories with local causal
rules is much larger than the causal dynamical triangulations, and there is little reason at this
point to believe that the good CDT results are compatible with local causal rules. It appears
that, while other features of sum-over-histories quantum gravity, such as locality or fundamental
discreteness, appear desirable, they are not shared by causal dynamical triangulations. As it is
the latter that possesses the good low-energy behavior, this implies either that these desirable
features are misguided, or that our previous expectations for the geometrical interpretation of the
path-integral histories as the microscopic description of spacetime were naive.
The outline of this paper is as follows: In Section 2 we review the 1 + 1 causal dynamical
triangulations. In Section 3, we generalize the 1 + 1 pure gravity model to the varying lapse
gauge. In Section 4, we apply the varying lapse gauge to gravity coupled to matter. In Section 5
we compare causal dynamical triangulations to local causal models. We discuss our conclusions
in Section 6.
2. Pure gravity 1 + 1 causal dynamical triangulations
Causal dynamical triangulations is a non-perturbative gravitational path integral for Lorentzian spacetimes, proposed by Ambjrn and Loll in [2]. Computing the path integral is done by
counting geometries at the regularized level weighted by the Lorentzian Einstein action. The regularized spacetimes are histories of flat Lorentzian simplices with certain causality conditions.
123
In pure gravity in 1 + 1 dimensions, the only attribute a spatial slice has is its length L. In the
regulated model L = as l, where as is the spacelike edge length, which serves as an ultraviolet
regulator in the model, and l is the number of spacelike edges that make up the slice, an integer.
The central object in the AmbjrnLoll pure gravity model is the amplitude to go from a
universe of length l0 to a universe with length lt in t time steps:

G(l0 , lt , t) =
(1)
eiSAL [T (t)] ,
T (t):l0 lt
where T (t) refers to all the triangulations with in boundary a circle of length l0 , out boundary of
length lt and consist of t time steps.
A history with t time steps has t + 1 time slices i = 0, . . . , t, each of length li > 0. Between
slice i and i + 1 is a band of ni+1 = li + li+1 triangles, with li triangles oriented up with base
on the bottom slice and li+1 triangles oriented down, with base on the top slice:
Each triangle has one spacelike edge of length ls2 = a 2 and two equal time like edges of length
lt = a 2 where a is the regularization cutoff:
The height of the triangles is t = 1. Note that the causality conditions of the model is not
restricted to the Minkowskian interior of the triangles, but also require that there is no spatial
topology change or closed timelike curves.
In two dimensions, the Einstein action is a topological invariant,

(2)
dx | det g|R = 2,
M
where is the Euler characteristic of the spacetime M. In the CDT model, only trivial topologies
are allowed so weighing the histories by this action simply gives a constant overall factor to the
path integral. Since in a dynamical triangulation scheme each triangle has the same area, the
action for each history is
SAL [T ] = An(T ),
(3)
where n(T ) is the number of triangles in the history. The area is defined by A = N0 as /2 where
N0 the fixed lapse.
Note that it is part of the definition that the length of the universe can change by any amount
in a single time step (but cannot shrink to less than l = 1). We also note that the physical amplitude should be divided by l0 to divide out by a discrete residue of spatial diffeomorphism
124
invariance at the boundaries (i.e. histories that differ by an overall rotation of the initial universe
are equivalent).
The propagation amplitude G(l0 , lt , t) satisfies
G(l0 , lt , t) =
G(l0 , l, 1)G(l, lt , t 1),
(4)
l=1
where the single time step amplitude G(l0 , l, 1) is the one-step transfer matrix. It can be written
as follows. Each vertex r of the initial ring is attached to kr vertices of the final ring, which we
call the future set of the vertex r. However we identify the leftmost vertex in the future set of
l0
kr
r with the rightmost vertex in the future set of r 1. There are then n0 = l0 + l1 = r=1
triangles in the strip, and hence the one-step transfer matrix is
l0

eiA r=1 kr .
G(l0 , l, 1) =
(5)
k1 ,...,kl0
It is convenient to define the amplitude for a single triangle by

g = eiA ,
(6)
g n(T ) .
in which case the amplitude of a history is

To solve Eq. (4) we invent the generating function
G(x,
y, t) =
x k y l G(k, l, t),
(7)
k,l=1
so that the recursion relation (4) is written for the generating function as

dz 1
G(x,
y, t1 + t2 ) =
G x, z , t1 G(z, y, t2 ).
2iz
(8)
One then arrives at the transfer matrix G(x,

y, 1) for the transform by a counting argument.
We use (7) to construct the generating function for one time step, which is the amplitude to go
from an initial loop to a next loop. We sum over the number of edges in the initial loop, and
associate a factor of x with each edge, coming from the definition (7). Each initial edge can give
rise to any number of edges in the next loop. To each of these we associate a factor y. We then
multiply each triangle by a factor g. There is one triangle (pointed up) for each edge in the initial
loop, and one triangle, (pointed down) for each triangle in the next loop. Thus each x and each y
is multiplied by a factor of g. The amplitude is given by the double sum

g 2 xy
l
.
(gy)
(gx)k =
gx
G(x,
y, 1) =
(9)
(1 gx)(1 gx gy)
k=0
l=0
k=0
The second factor excludes the possibility that the next loop has zero edges.
The analysis proceeds from here by the solution of the recursion relation

dz 1
G(x,
y, t) =
G x, z , 1 G(z, y, t 1).
2iz
(10)
This leads to the construction of the full generating function, which is then a function only of
x, y and g (for more details see [2]). It is then shown that the continuum limit is possible because
125
of the existence of a fixed point at

1
gc = .
2
If we define the dimensionless quantity
= A,
(11)
(12)
the critical point (11) corresponds to a critical 0 = i ln 2.

3. Pure gravity CDT in varying lapse gauge
In the quantization of a diffeomorphism invariant theory the lapses must be allowed to vary
without affecting the physical results. If that is the case in CDT, then the fixing of the lapses is
merely a gauge condition and not a physical reduction of the degrees of freedom of the theory.
To examine whether this is the case, we now consider a simple generalization of CDT in which
the lapses and areas of the different triangles in a triangulation are allowed to vary.
We denote Ni the lapse of triangle i. If we keep the spatial edge lengths as fixed for all
triangles, corresponds to area Ai = 12 Ni as . We require that
as
2
below which the timelike edges in the triangles become null, then spacelike.
With this change the action for a history T becomes
S[T ] =
Ai = An(T
),
Ni > Nmin =
(13)
(14)
iT
where A = Ai is the average triangle area and n(T ) is the number of triangles in triangulation T . The new weight for a triangle becomes
gi = eiAi .
(15)
The CDT model is solved using a one-step transfer function. To reproduce this solution for the
varying lapses, one has to work with average lapse and area in a single slice, instead of the average history lapse N = Ni . A spacetime solution of the 1 + 1 CDT that would accommodate
the average history lapse is underway elsewhere [9]. For the present note, we will be restricted
to a generalization of the CDT model that is easily constructed within the present transfer matrix
solution. For this, we need to restrict to histories in which the average lapse of each slice remains
constant in time:
N t = N .
(16)
At this point, it is clear that our modest generalization cannot change the CDT partition function. The action changes by an overall scaling.
S[T ] =
N
N aS n(T )
=
SAL [T ],
2
N0
reflecting the fraction by which N differs from the lapse of CDT.
(17)
126
In the continuum limit, we will have the same gc = 12 and the change in (17) just goes into
rescaling of the renormalization group trajectory used to define the continuum limit, so that the
dimensionless scales by = N /N0 .
Quantities of physical interest, such as the Hausdorff dimension, are computed by counting
features of a typical triangulation, and taking the continuum limit. For example, one counts the
number of spatial edges a fixed number of causal steps into the future and shows that this grows
by the appropriate power. What is relevent for the continuum limits is the behavior of large sets,
for a large number of steps. Allowing the lapses to vary does not change the physical interpretation as long as the sets inolved are large enough that average quantities may be substituted for
sums.
It is important to emphasize that varying the lapse as proposed here is not the same as summing over lapses. What we have done amounts to changing the gauge fixing constraint in the
path integral, but not removing it. The original causal dynamical triangulation model amounts to
a gauge fixed path integral, which is formally

a
dqab dN dN a (N 1) N a FP eiS[qab ,N,N ] ,
Z=
(18)
x,t
where qab is the spatial part of the metric, N is the lapse and N a is the shift. What we have
shown is that this may be modified to

a
dqab dN dN a N f (x, t) N a FP eiS[qab ,N,N ]
Z=
(19)
x,t
so long as, for all slices,

qf (x, t)

= const.
q
(20)
Furthermore, substituting unit lapse with constant average within each physical region involved in the measurement of a macroscopic physical observable will not affect the physical
quantities that describe the macroscopic behavior such as the Hausdorff dimension or the leading low energy behavior of the propagators of matter fields. We will discuss some examples of
this in the next section.
4. Matter coupling in 1 + 1 dimensions
We now consider the consequences of varying the lapse for models where matter is coupled.
Following [3] we consider coupling gravity to a spin system by adding a spin i to each triangle.
The CDT coupled to gravity partition function is
AL ALspin
[T ]
eiS [T ]
eS
,
G(, , t) =
(21)
T (t)
where the initial and final lengths and spin states are fixed and sum is over all the internal spins
and histories with t time steps. The matter action is given by

S ALspin [T ] =
(22)
i j ,
ij T
where is the coupling constant of the spin system.
If the lapse varies, then the effect will be to modify (22) to

S spin [T , N] =
i t Rij [Ni , Nj ],
127
(23)
ij T
where the new factor Rij [Ni , Nj ] takes into account the varying metric. However, note that, for
each fixed triangulation and set of lapses and each fixed set of spins,

(24)
i j Rij =
1+
1 Rij =: P ( )Rij ,
ij T
{ij :i j =1}
{ij :i j =1}
where P ( ) is the number of nearest neighbor pairs whose spins agree minus the number whose
spins disagree.
For the continuum limit we are interested in large triangulations, so that the averages will be
fixed to good accuracy over the sum over triangulations. Hence, the effect of changing the gauge
condition so as to let the lapses vary can only be to modify the effective matter coupling so

i j ,
S spin [T ] =
(25)
ij T
where

= Rij .
(26)
Hence, the properties of matter found in [3] for the continuum limit will be unchanged.
5. CDT and locality
We turn now to the question of whether the non-locality of the CDT model is necessary for
the existence of a good continuum limit. Certainly, what is desired for a discete path integral
approach to quantum gravity is: (1) A good low energy or coarse grained limit, which is both
local and causal macroscopically, and leads to a recovery of Lorentz invariant spacetime physics
at low energies, but also (2) microscopic causality, and (3) microscopic locality are desirable
properties. At present there are several models whose investigation is relevent for the question
of whether there exists a theory with all three good properties. It is interesting to compare the
results concerning the following models.
CDT models [2]. As we discussed above, these are causal microscopically, but not local, because the evolution rule preserves the existentence of a global slicing. However, the evidence
points to the existence of a good continuum limit.
Causally evolving spin network models (CSN) [5]. In this construction all edges are spacelike, and each triangle has a causal orienation. There are two kinds of triangles: future
pointing triangles with one past edge and two future edges and past pointing triangles, with
two past edges and one future edge. Histories are constructed by a local rule which adds
simplices locally. The amplitude of a history is a product of the amplitude for each simplex
added, and this is a function only of the simplices in the immediate past.
Euclidean dynamical triangulation models (EDT) [1]. These are constructed by a rule which
is local, but not causal. One sums over all 2d random triangulations, weighed only by the
number of triangles. They do not lead to a continuum limit which reproduces physics in
relativistic spacetimes.
128
Causal set models [6]. Histories are partial orders of events. A generic causal set does not
correspond to a triangulation of a manifold, nor is there an obvious notion of locality.1
There are other models, such as Regge calculus models and alternate versions of causal spin
foam models, but they are not needed for the present discussion. We now discuss these, restricting to the (1 + 1)- or 2-dimensional case, except in the causal set model, where there is no
microscopic notion of dimension.
We note that the causal set histories differ from the histories in the other causal models, both
CDT and CSN, in that the antichains have no structure. In both CDT and CSN the antichains are
spatial slices in a triangulation.
We next note that, since CDT does not allow topology change, there are many more EDT
histories than CDT histories (for given initial and final states, specified by l0 and lt , and given
N ). The former do not have continuum limits which reproduce physics which is macroscopically
(1 + 1)-dimensional, while the latter do. Thus, the restriction to causal histories seems necessary
for a good continuum limit. A key question is whether such a restriction must be non-local
microscopically, as is the case with CDT. Since the continuum limit properties are known only
for CDT and Euclidean histories, we will attempt a preliminary comparison of CSN with these
two theories.
The local CSN rules mean that it is possible to add many more simplices in some part of
a spatial slice, while keeping other parts of it unchanged. While this is not a topology change
as such (it causes no branching of the spatial slice), such a CSN history can be mapped to a
Euclidean triangulation of a branching history. (Note that there is no map of CSN histories to
topologies with branching and rejoining spatial slices.) In addition, since a CSN simplex also
carries a causal structure, multiple CSN histories (where the same triangulation carries different
causal structures) can be mapped to the same Euclidean one.
There are then more CSN histories than Euclidean ones. This means that, if the CSN histories
are weighed equally by eN as in CDT and EDT histories, there can be no good continuum limit
which reproduces (1 + 1)-dimensional physics. It has been argued (see for example [8]) that,
since the growth in volume of the number of inequivalent triangulations of a given volume is
factorial, no adjustments of the weights can stop models with topology change from having the
continuum limit properties of the Euclidean dynamical triangulations.
Unfortunately, that is how far the implications of CDT for other path integral models can be
taken. The above argument is correct for a statistical path integral, namely only for EDT and the
Wick rotated CDT. For a quantum path integral, the possibility of interference and cancellations
remains. There is no Wick rotation for CSN, so the mapping of CSN histories to EDT ones does
not imply anything conclusive about the CSN weights.
To summarize, at present, the only of these models for which there is evidence that they reproduce (1 + 1)-dimensional physics in the low energy or continuum limit is the CDT model,
whose construction, however, appears to be non-local. The other models are local microscopically, but either they do not reproduce low-dimensional causal physics macroscopically or there
is no method to solve them. Thus, from the present results, it appears as if there is a conflict
between microscopic and macroscopic local physics as there is no model that has both.
1 Percolation causal set, in which the partial order is constucted by a percolation process [7] are microscopically causal
and satisfy a version of microscopic locality.
129
6. Conclusions
We began by mentioning three features which distinguish the AmbjrnLoll causal dynamical triangulations model from other background independent approaches to the path integral in
quantum gravity. Of these, we have seen that the restriction to unit lapse, stressed in point 2 can
be somewhat relaxed, providing support for the view that the preferred time slicings of the model
reflect a gauge fixing, rather than a physical preferred time.
There are a number of further steps, which are currently under investigation.
We expect that the relaxation of the gauge fixing described here, and in [10] for 2 + 1 dimensions, will hold also in the (3 + 1)-dimensional model [4].
The need for a constant time average lapse may depend on the setup of the transfer matrix
solution of the CDT. To check this, an alternative spacetime solution of CDT in 1 + 1 is
currently in progress [9].
Given the recent results concerning the CDT model in 3 + 1 dimensions which support the
conjecture that the model has a good continuum limit which yields a theory of spacetime
geometry in 3 + 1 dimensions, it becomes of interest to see if that model can be elaborated
to yield a spin foam model with similar properties. This can be done by noting that spatial
slicings exist whose duals are graphs, and then extending the model by introducing labels as
in spin foam models.
It is of interest to understand better the relationship between causal spin foam models such
as those studied in [5] and causal dynamical triangulation models. This raises the issue of
non-locality, as the latters rules are non-local in ways the formers are not. It is of particular
interest to establish whether the good results gotten with the CDT models are dependent on
the presence of non-locality in the microscopic evolution rules.
Acknowledgements
We are grateful to Jan Ambjrn, Renate Loll and members of Perimeter Institute for discussions on these issues.
References
[1] J. Ambjrn, Z. Burda, J. Jurkiewicz, C.F. Kristjansen, Quantum gravity represented as dynamical triangulations,
Acta Phys. Pol. B 23 (1992) 991;
J. Ambjrn, Quantum gravity represented as dynamical triangulations, Class. Quantum Grav. 12 (1995) 2079;
M.E. Agishtein, A.A. Migdal, Simulations of four-dimensional simplicial quantum gravity, Mod. Phys. Lett. A 7
(1992) 1039.
[2] J. Ambjrn, A. Dasgupta, J. Jurkiewicz, R. Loll, A Lorentzian cure for Euclidean troubles, hep-th/0201104;
J. Ambjrn, R. Loll, Nucl. Phys. B 536 (1998) 407, hep-th/9805108;
J. Ambjrn, J. Jurkiewicz, R. Loll, Phys. Rev. Lett. 85 (2000) 924, hep-th/0002050;
J. Ambjrn, J. Jurkiewicz, R. Loll, Nucl. Phys. B 610 (2001) 347, hep-th/0105267;
R. Loll, Nucl. Phys. B (Proc. Suppl.) 94 (2001) 96, hep-th/0011194;
J. Ambjrn, J. Jurkiewicz, R. Loll, Phys. Rev. D 64 (2001) 044011, hep-th/0011276;
J. Ambjrn, J. Jurkiewicz, R. Loll, G. Vernizzi, Lorentzian 3d gravity with wormholes via matrix models, JHEP 0109
(2001) 022, hep-th/0106082;
B. Dittrich, R. Loll, A hexagon model for 3D Lorentzian quantum cosmology, hep-th/0204210.
[3] J. Ambjrn, K.N. Anagnostopoulos, R. Loll, Crossing the c = 1 barrier in 2d Lorentzian quantum gravity, Phys.
Rev. D 61 (2000) 044010, hep-lat/9909129.
[4] J. Ambjrn, J. Jurkiewicz, R. Loll, Emergence of a 4D world from causal quantum gravity, hep-th/0404156.
130
[5] F. Markopoulou, Dual formulation of spin network evolution, gr-qc/9704013;

F. Markopoulou, L. Smolin, Quantum geometry with intrinsic local causality, Phys. Rev. D 58 (1998) 084032,
gr-qc/9712067.
[6] L. Bombelli, J.H. Lee, D. Meyer, R. Sorkin, Spacetime as a causal set, Phys. Rev. Lett. 59 (1987) 521.
[7] X. Martin, D. OConnor, D.P. Rideout, R.D. Sorkin, On the renormalization transformations induced by cycles of
expansion and contraction in causal set cosmology, Phys. Rev. D 63 (2001) 084026, gr-qc/0009063;
D.P. Rideout, R.D. Sorkin, Evidence for a continuum limit in causal set dynamics, Phys. Rev. D 63 (2001) 104011,
gr-qc/0003117;
D. Rideout, Dynamics of causal sets, gr-qc/0212064;
D.P. Rideout, R.D. Sorkin, A classical sequential growth dynamics for causal sets, Phys. Rev. D 61 (2000) 024002,
gr-qc/9904062.
[8] R. Loll, W. Westra, Spacetime foam in 2D and the sum over topologies, hep-th/0309012.
[9] M. Ansari, F. Markopoulou, in preparation.
[10] T. Konopka, Foliations and 2 + 1 causal dynamical triangulation models, hep-th/0505004.
Small-x QCD effects in forward-jet and

MuellerNavelet jet production
C. Marquet a, , C. Royon b,c
a Service de Physique Thorique, CEA/Saclay, 91191 Gif-sur-Yvette cedex,
URA 2306, Unit de Recherche Associe au CNRS, France

b DAPNIA/Service de Physique des Particules, CEA/Saclay, 91191 Gif-sur-Yvette cedex, France
c Fermi National Accelerator Laboratory, Batavia, IL 60510, USA
Abstract
We investigate small-x QCD effects in forward-jet production in deep inelastic scattering in the kinematic
regime where the virtuality of the photon and the transverse momentum of the jet are two hard scales of
about the same magnitude. We show that the data from HERA published by the H1 and ZEUS Collaborations are well described by leading-logarithmic BFKL predictions. Parametrizations containing saturation
effects expected to be relevant at higher energies also compare well to the present data. We extend our
analysis to MuellerNavelet jets at the LHC and discuss to what extent this observable could test these
small-x effects and help distinguishing between the different descriptions.
1. Introduction
The Regge limit of perturbative QCD comes about when the centre-of-mass energy in a collision is much bigger than the fixed hard scales of the problem. In this limit usually called the
small-x regime, parton densities inside the projectiles grow with increasing energy, leading to
the growth of the scattering amplitudes. As long as the densities are not too high, this growth
is described by the BalitskyFadinKuraevLipatov (BFKL) equation [1] that resums the leading logarithms. As the parton density becomes higher and the scattering amplitudes approach
E-mail addresses: marquet@spht.saclay.cea.fr (C. Marquet), royon@hep.saclay.cea.fr (C. Royon).

doi:10.1016/j.nuclphysb.2006.01.024
132
C. Marquet, C. Royon / Nuclear Physics B 739 (2006) 131155
the unitarity limit, one enters a regime called saturation [25] where the BFKL evolution breaks
down and parton densities saturate.
In the past years, as colliders started to explore the small-x regime, proposals were made to
test the relevance of the BFKL equation at the available energies. In this paper we concentrate
on two of the proposed measurements: forward jets [6] in deep inelastic scattering (DIS) and
MuellerNavelet jets [7] in hadronhadron collisions. Forward-jet production is a process in
which the virtual photon interacts with the proton and a jet is detected in the forward direction
of the proton. The virtuality of the photon and the squared transverse momentum of the jet are
hard scales of about the same magnitude. In the case of MuellerNavelet jets, a proton interacts
with another proton or antiproton and a jet is detected in each of the two forward directions;
the transverse momenta of the jets are as well hard scales of about the same magnitude. If the
total energy in the photonproton (for forward jets) or protonproton (for MuellerNavelet jets)
collision is large enough, these processes feature the kinematics corresponding to the Regge
limit.
The description of forward jets with fixed-order perturbative QCD in the Bjorken limit
amounts in the following. Large logarithms coming from the strong ordering between the soft
proton scale and the hard forward-jet scale are resummed using the DokshitzerGribovLipatov
AltarelliParisi (DGLAP) evolution equation [8] and the hard cross-section is computed at fixed
order in the coupling constant. In the small-x regime, due to the extra ordering between the total energy and the hard scales, other large logarithms arise and should be resummed within the
hard cross-section itself. In other words, the inclusion of small-x effects aims at improving QCD
predictions by replacing fixed-order hard cross-sections with resummed hard cross-section, using the BFKL equation or, at even higher energies, using resummations that include saturation
effects.
To study different observables in this small-x regime, a convenient approach is to formulate
the cross-sections in terms of scattering amplitudes for colorless combinations of partons. The
simplest of those is the q q dipole, a quarkantiquark pair in the color singlet state; it describes
for instance the interaction of a virtual photon. Any colorless gg, q qg,
. . . multiplets can a priori
be involved, for instance the gluongluon (gg) dipole is what describes gluon emissions.
To compute the evolution of those scattering amplitudes with energy, the QCD dipole
model [9] has been developed. This formalism constructs the light-cone wavefunction of a q q
dipole in the leading logarithmic approximation. As the energy increases, the original dipole
evolves and the wavefunction of this evolved dipole is described as a system of elementary q q
dipoles. When this system of dipoles scatters on a target, the scattering amplitude has been shown
to obey the BFKL equation. Interestingly enough, the dipole formalism was shown to be also
well-suited to include density effects and nonlinearities that lead to saturation and unitarization
of the scattering amplitudes [10,11]. This is why the dipole picture is suitable for investigating
the small-x regime of QCD, it allows to study both BFKL and saturation effects within the same
theoretical framework.
The formulation of the forward-jet and MuellerNavelet jet processes in terms of dipole
amplitudes has been addressed in [12]. In both cases the problem is similar to the one of onium
onium scattering: the growth with energy of the total cross-section due to BFKL evolution is
damped by saturation effects which arise purely perturbatively. For instance in the large-Nc
limit, that involves multiple pomeron exchanges. In our study, we consider both the BFKL energy regime and saturation effects. We shall implement saturation in a very simple way, using
a phenomenological parametrization inspired by the Golec-Biernat and Wsthoff (GBW) ap-
133
proach [13] which gave a good description of the proton structure functions with a very few
parameters.
First comparisons of small-x predictions with forward-jet data from HERA were quite successful: the first sets of data published by the H1 [14] and ZEUS [15] Collaborations are well
described by the leading-logarithmic (LL) BFKL predictions [16] and also show compatibility with saturation parametrizations [17,18]. In both cases, the DGLAP resummation associated
accounted for leading logarithms.
Very recently, new forward-jet experimental results have been published [19,20]. They involve a broader range of observables with several differential cross-sections and go to smaller
values of x than the previous measurements. As QCD at next-to-leading order (NLO) is not sufficient to describe the small-x data, we shall address the following issues: whether the BFKL-LL
predictions keep being in good agreement and whether saturation parametrizations still show
compatibility. The first part of the present work is devoted to those questions.
In the second part of the paper, we deal with MuellerNavelet jets. We display BFKL-LL predictions in the LHC energy range for different differential cross-sections. We compare them with
saturation predictions obtained from our parametrizations of saturation effects constrained by the
forward-jet data. We propose different measurements and discuss their potential for identifying
BFKL and saturation behaviors.
The paper will be organized as follows. In Section 2, we compute the forward-jet cross-section
in the high-energy regime and express it in terms of a dipoledipole cross-section. We compare the BFKL-LL predictions and the saturation parametrization with the new H1 and ZEUS
data for several differential cross-sections. In Section 3, we compute the MuellerNavelet jet
cross-section and show the BFKL and saturation predictions for LHC energies and for several
differential cross-sections. The final Section 4 is devoted to conclusion and outlook.
2. Forward-jet production
Forward-jet production in a leptonproton
collision is represented in Fig. 1 with the different
kinematic variables. We denote s the total energy of the leptonproton collision and Q2 the
virtuality of the intermediate photon that undergoes the hadronic interaction. We shall use the
usual kinematic variables of deep inelastic scattering: x = Q2 /(Q2 + W 2 ) and y = Q2 /(xs)
where W is the center-of-mass energy of the photonproton collision. In addition, kT QCD
is the jet transverse momentum and xJ its longitudinal momentum fraction with respect to the
proton. In the following, we compute the forward-jet cross-section in the high-energy limit, recall
the BFKL predictions and give our formulation of the saturation model.
2.1. Formulation
The QCD cross-section for forward-jet production reads
pJ X
pJ X
pJ X 2
dT
dT
dL
d (4)
y
em
(1 y) +
,
=
+
2
2
2
2
2
2
xQ
dx dQ dxJ dkT
dxJ dkT
dxJ dkT
dxJ dkT 2
(1)
pJ X
dT ,L
/dxJ
dkT2
where
is the cross-section for forward-jet production in the collision of the
transversely (T ) or longitudinally (L) polarized virtual photon with the target proton.
We now consider the high-energy regime x 1. In an appropriate frame called the dipole
frame, the virtual photon undergoes the hadronic interaction via a fluctuation into a dipole. The
134
Fig. 1. Production of a forward jet in a protonlepton collision. The kinematic variables of the problem are displayed.
Q2 is the virtuality of the photon that undergoes the hadronic interaction. s and W 2 are the total energies squared in the
leptonproton and photonproton collisions respectively. kT is the transverse momentum of the forward jet and xJ is
its longitudinal momentum fraction with respect to the incident proton. Y is the rapidity interval between the two hard
probes.
dipole then interacts with the target proton and one has the following factorization

pJ X
dT ,L
dxJ dkT2
1
2
d r
0

2 dq q
dz T ,L (r, z; Q)
(r).
dxJ dkT2
(2)
The wavefunctions T and L describe the splitting of the photon on the dipole and
dq q /dxJ dkT2 is the cross-section for forward-jet production in the dipoleproton collision.
T and L are given by

(r, z; Q)2 = em Nc
ef2 z2 + (1 z)2 z(1 z)Q2 K12 z(1 z)Q|r| ,
T
2
2
(3)

(r, z; Q)2 = em Nc
ef2 4Q2 z2 (1 z)2 K02 z(1 z)Q|r|
L
2
2
(4)
for a transversely (3) and longitudinally (4) polarized photon where ef is the charge of the
quark1 with flavor f . The integration variable r is the transverse size of the q q pair and z is
the longitudinal momentum fraction of the antiquark with respect to the photon. In the leading
logarithmic approximation we are interested in, the cross-section dq q /dxJ dkT2 does not depend
on z but only on the dipole size r. This cross-section has been computed in [12] where it was
shown that the emission of the forward jet can be described through the interaction of an effective
1 We consider massless quarks and sum over four flavors in (3) and (4). This is justified considering the rather high
values of the photon virtuality (Q2 > 5 GeV2 ) used for the measurement.
135
gluonic (gg) dipole:

Nc
(r) =
feff xJ , kT2
2
2
dxJ dkT
16kT
dq q

0

r (q q)(gg)
d r J0 (kT r )
(r, r , Y )
r r
(5)
(r, r , Y ) is the cross-section in

with Y = log(xJ /x) the rapidity assumed to be very large. (q q)(gg)
the collision of a q q dipole of size r with a gg dipole of size r with total rapidity Y . feff (xJ , kT2 )
is the effective parton distribution function and resums the leading logarithms log(kT2 /2QCD ). It
has the following expression

CF

q xJ , kT2 + q xJ , kT2 ,
feff xJ , kT2 = g xJ , kT2 +
Nc
(6)
where g (respectively q, q)
is the gluon (respectively quark, antiquark) distribution function in
the incident proton.
Let us comment formula (5). Since the forward jet measurement involves perturbative values
of kT and moderate values of xJ , it is not surprising that formula (5) features the collinear factorization of feff ; note also that kT2 has been chosen as the factorization scale. The remaining hard
interaction is between a gg dipole and the incident q q dipole of size r. The gg dipole emerges
as the effective degree of freedom for the gluon emission at high energies [12]. This feature has
been pointed out several times [21].
Formulae (1)(6) express the forward-jet observable (1) in terms of the cross-section (q q)(gg)
which contains the high-energy QCD dynamics: the problem is analogous to the one of onium
onium scattering. In Section 2.2, we deal with the BFKL energy regime for which the interaction
between the q q dipole and gg dipole is restricted to a pomeron exchange. In that case of course,
our formulation is equivalent to the kT -factorization approach. In Section 2.3, we go beyond
a priori contains any
kT -factorization and investigate the saturation regime in which (q q)(gg)
number of gluon exchanges.

2.2. The BFKL energy regime
The BFKL q qgg
dipoledipole cross-section reads (see for instance [22])

BFKL
(q
(r, r , Y ) = 2s2 r 2
q)(gg)

s Nc
d (r /r)2
( )Y
exp
2i 2 (1 )2
(7)
with the complex integral running along the imaginary axis from 1/2 i to 1/2 + i and
with the BFKL kernel given by
( ) = 2(1) (1 ) ( )
(8)
where ( ) is the logarithmic derivative of the gamma function. It comes about when the interaction between the q q-dipole
and the gg-dipole is restricted to a two-gluon exchange. Summing

the leading-logarithmic contributions of ladders with any number of real gluon emissions, one
obtains the BFKL pomeron and the resulting growth of the cross-section with rapidity.
136
Inserting (7) in (5) and (2), one obtains

pJ X
dT ,L
dxJ dkT2

2 Nc s2
feff xJ , kT2
2
2
4kT Q

d Q2 4 ( )T ,L ( )
s Nc
exp
(
)Y
,
2i kT2
(1 ) (2 )
(9)
where we have defined the following Mellin-transforms
T ,L ( ) =

2

2 1
1
d r r Q

2
dz T ,L (r, z; Q)
(10)
which are given by

T ( )
L ( )

=

2em Nc 2 1 2 (1 + ) 2 (1 ) 2 (2 ) (1 + )(2 )
.
eq
2 (1 )
4 (2 2 ) (2 + 2 )(3 2 )
q
(11)
Inserting formula (9) into (1) gives the forward-jet cross-section in the BFKL energy regime. One
can easily show that the result is identical to the one obtained using kT -factorization [16,23]. The
only undetermined parameter is s Nc / (with s the strong coupling constant kept fixed)
which appears in the exponential in formula (9).
2.3. The saturation regime
Contrary to the BFKL case, the oniumonium cross-section in the saturation regime has not
yet been computed from QCD. Studies are being carried out to identify the dominant terms in
sat
the multiple gluon exchanges [11,2426] but the cross-section (q
remains unknown. To
q)(gg)
take into account saturation effects, we are led to use a phenomenological parametrization. We
consider the following model introduced in [17] which is inspired by the GBW approach:

2

reff (r, r )
sat
2
(q
1
exp
.
(r,
r
,
Y
)
=
4
s 0
q)(gg)
4R02 (Y )
(12)
2 (r, r ) is defined through the two-gluon exchange:

The dipoledipole effective radius reff

2 2
max(r, r )
.
r , r 1 + log
min(r, r )
2
BFKL
4s2 reff
(r, r ) (q
(r, r , 0) = 4s2 min
q)(gg)
(13)
For the saturation radius we use the parametrization R0 (Y ) = e 2 (Y Y0 ) /Q0 with Q0 1 GeV.
Let us express the cross-section in terms of a double Mellin-transform:

sat
(q
(r, r , Y ) = 4s2 0
q)(gg)
d
2i
d
2i
r2
4R02 (Y )
1
r 2
4R02 (Y )

g( , )
(14)
137
with

g( , ) =

2
du
0

2
d u 2 u2 4 u 2 2 1 ereff (u,u)

2 ( )
(1, 3 + , 2 ) + (1, 3 + , 2 2 ) ,
=
1+
0 < Re( ), Re( ), Re( ) < 1,
(15)
where the confluent hypergeometric function of Tricomi (1, a, b) can be expressed [27] in
terms of incomplete gamma functions. Inserting (14) in (5) and (2), one obtains
pJ X
dT ,L
dxJ dkT2

d 2 2
2 Nc s2 0
2
4Q R0 (Y ) T ,L ( )
f
x
g(
,
)
,
k
eff
J
T
2
2
2i
8Q2 kT R0 (Y )

d 2 2 4 2 ( )
4kT R0 (Y )
g( , ).
2i
(1 )
(16)
Inserting formula (16) into (1) gives our parametrization of the forward-jet cross-section in the
saturation regime. The parameters are , Y0 and the normalization 0 .
2.4. Fixing the parameters
The first sets of data published by the H1 [14] and ZEUS [15] Collaborations regarded the
measurement of d/dx. In previous studies, we fitted the BFKL-LL [16] and saturation parametrization [7] on those data with the cut x < 102 . Despite corresponding different energy regimes,
in both cases we obtained good descriptions with 2 values of about 1. The obtained values of
the parameters and the 2 of the fits are given in Table 1.
In the BFKL-LL case, the only parameter is and the value obtained was 4 log(2) = 0.430.
For the saturation fit, the two relevant parameters are and Y0 and the fit showed two 2 minima
for ( = 0.402, Y0 = 0.82) and ( = 0.370, Y0 = 8.23). We shall refer to the first (respectively second) solution as a strong (respectively weak) saturation parametrization. Indeed, the
first saturation minimum corresponds to strong saturation effects as, for typical values of Y , the
saturation scale 1/R0 is about 5 GeV which is the value of a typical kT . The second saturation
minima corresponds to small saturation effects and rather describes BFKL physics.
Along with formulae (9) and (16), the values of the parameters given in Table 1 completely
determine the BFKL-LL predictions and two parametrizations for the saturation model. We are
now going to compare these with the very recent data without any adjustment of the parameters.
This will provide a strong test of those small-x effects.
Table 1
Results of the BFKL and saturation fits to the first HERA forward-jet data. The saturation fits shows two independent
solutions showing either strong or weak saturation parameters (see text)
Fit
Parameters
1/R0 (Y = 0)
2 (/d.o.f.)
BFKL-LL
Strong sat.
Weak sat.
4 log(2) = 0.430
= 0.402 and Y0 = 0.82
= 0.370 and Y0 = 8.23
1.18 GeV
0.22 GeV
12 (/13)
6.8 (/11)
8.3 (/11)
138
2.5. Comparison to the 2005 HERA data

We want to compare the cross-section (1) obtained from the BFKL-LL prediction (9) and
the saturation parametrization (16) with the new data coming from measurements performed at
HERA [19,20]. On one side, our theoretical results are for the cross-section (1) which is differential with respect to all four kinematic variables x, Q2 , xJ , and kT . On the other side, HERA data
concern observables which are less differential: d/dx, d/dQ2 , d/dkT2 , and d/dx dQ2 dkT2 .
Therefore, on the top of the Mellin integrations in (9) and (16), one has to carry out a number
of integrations over the kinematic variables which have to be done taking into account the kinematic cuts applied by the different experiments. A detailed description of how we performed
those integrations is given in Appendix A and the resulting cross-sections that can be compared
to the data are given in Appendix B. The method allows for a direct comparison of the data with
theoretical predictions but it does not allow to control the overall normalization. In the following studies, we therefore compare only spectra and will not refer to normalizations anymore. As
already mentioned, one does not adjust any of the parameters of Table 1.
Let us start with the observable d/dx which has been measured by both the H1 and ZEUS
Collaborations and which now features lower values of x than the first measurements. The comparison is displayed in Fig. 2, the three small-x parametrizations describe very well the data. One
cannot really distinguish between the three curves, except at small values of x where one starts
to see a difference: the BFKL curve is above the weak-saturation curve which is itself above the
strong-saturation curve. However the main conclusion is that the data seem to feature the BFKL
growth, when going to small values of x. For comparison, the fixed-order QCD predictions at
NLO computed in [19,20] with the DISENT Monte Carlo program [28] are reproduced in Fig. 2.
At the lowest values of x, they do not reproduce the data as they are about a factor 1.5 to 2.5 below depending on the experiment and the error bars. Even adding a resolved-photon component
to the NLO predictions [20] does not pull them within the uncertainties, contrary to what happened for the previous data [29]. This is an interesting difference with the forward-pion case [30]
for which it seems that no higher-order effect other than a NLO resolved-photon contribution is
needed.
In Fig. 3 are represented the two other single differential cross-sections that we shall briefly
discuss: d/dQ2 and d/dkT measured by the ZEUS Collaboration. One can see again that the
three small-x parametrizations agree well with the data, it is a strong result that one is able to
describe the Q2 and kT spectra without any adjustment of the parameters as they were only fitted
to describe the x dependence.
We shall finally compare our predictions with the triple differential cross-section d/
dx dQ2 dkT2 measured by the H1 Collaboration. The interesting part of this measurement is that
it has been carried out in 9 different bins of r kT2 /Q2 from 0.1 < r < 1.8 to 9.5 < r < 80.
This allows to test the limits of our parametrizations which are supposed to be valid only when
r 1 as they do not take into account any transverse momentum ordering of the gluons emitted
in rapidity between the forward jet and the photon. The comparisons with the data are shown
on Fig. 4 and one sees the expected trend. The bins which have r 1 are well described by the
small-x parametrizations while the others are not: for the latter, we overshoot the data as the
BFKL rise towards small values of x is too steep. Interestingly enough, the trend is reversed for
QCD predictions at NLO: they describe better the data which feature large values of r. These
observations favor the need of the BFKL resummation to describe the r 1 data. The large-r
bins also exhibit a limitation of the saturation model as one can see that the strong saturation
139
Fig. 2. The forward-jet cross-section d/dx. The points are measurement by the H1 (upper plot) and ZEUS (lower plot)
Collaborations. The lines are comparisons with BFKL-LL predictions (full lines) and the two saturation parametrizations (dotted and dashed lines). In both cases, there is good agreement with the data. For comparison, fixed-order QCD
predictions at NLO are also displayed.
140
Fig. 3. The forward-jet cross-sections d/dQ2 (upper plot) and d/dkT (lower plot). The points are measurement
by the ZEUS Collaboration. The lines are comparisons with BFKL-LL predictions (full lines) and the two saturation
parametrizations (dotted and dashed lines). For the two observables, there is good agreement with the data.
141
Fig. 4. The forward-jet cross-section d/dx dQ2 dkT2 . The points are measurement by the H1 Collaboration. The lines
are comparisons with BFKL-LL predictions (full lines) and the two saturation parametrizations (dotted and dashed
lines for strong and weak saturation respectively). In the regime where on expects small-x effects to be important
(r kT2 /Q2 1), there is a good description of the data. In the regime where r 1, the small-x parametrizations
do not reproduce the data as expected from the hierarchy of the hard scales.
parametrization lies above the weak saturation parametrization. Such a behavior indicates that
the model should not be used when kT2 Q2 .
Let us comment further on the two saturation parametrizations. While the BFKL formula (9)
is a QCD prediction as it is computed from Feynman diagrams [1], the saturation parametrization (16) is a phenomenological model. The fact that it describes well the data does not call for
the same conclusions as in the BFKL case. It only exhibits that, as it is the case for a number
of observables, data are compatible with saturation effects even at energies which do not require
them. In other words, the forward-jet measurement at the present energies cannot distinguish
142
between saturation and BFKL effects; one would start seeing a significant difference at higher
energies.
It is the purpose of the next section to look for such differences, by studying another process
similar to forward-jets, namely MuellerNavelet jets, at LHC energies.
3. Towards the LHC: MuellerNavelet jets
MuellerNavelet jet production in a proton-proton
collision is represented in Fig. 5 with the
different kinematic variables. We denote S the total energy of the collision, k1 and k2 the transverse momenta of the two forward jets and x1 and x2 their longitudinal fraction of momentum
with respect to the protons as indicated on Fig. 5. In the following, we compute the Mueller
Navelet jet cross-section in the high-energy limit, recall the BFKL predictions and formulate
our saturation model. We then display predictions for observables which can be measured at the
LHC.
3.1. Formulation
As in the original paper [7], we consider the cross-section differential with respect to x1 and
x2 and integrated over the transverse momenta of the jets with k1 > Q1 and k2 > Q2 . Q1 and Q2
represent then experimental kT -cuts. Considering the high energy limit, the QCD cross-section
for MuellerNavelet jet production reads [12,31]:

2 Nc2
d ppJ XJ
feff x1 , Q21 feff x2 , Q22
=
dx1 dx2
64

dr d r Q1 J1 (Q1 r)Q2 J1 (Q2 r )(gg)(gg) (r, r , Y )
0
(17)
Fig. 5. MuellerNavelet jet production in a protonproton collision. The kinematic variables of the problem are displayed.
S is the total energy squared, k1 and k2 are the transverse momenta of the jets and x1 and x2 are their longitudinal
momentum fraction with respect to the incident protons. Y is the rapidity interval between the hard probes.
143
with Y = log(x1 x2 S/Q1 Q2 ) the rapidity assumed to be very large. (gg)(gg) (r, r , Y ) is the crosssection in the collision of two gg dipoles of size r and r with rapidity total Y . As before feff is
the effective parton distribution function (6).
Let us comment formula (17). As before, each forward jet involves perturbative values of
transverse momenta and moderate values for x1 and x2 . This explains the collinear factorization
of the two functions feff ; here we have taken the factorization scales to be Q21 and Q22 . The
remaining hard interaction is between two gg dipoles: as we have seen in the previous section,
each of them describes a gluon emission at high energies. Formula (17) expresses the Mueller
Navelet jet observable in terms of the cross-section (gg)(gg) which contains the high-energy
QCD dynamics. This is the similarity with the forward-jet case: the problem is also analogous to
the one of oniumonium scattering.
Let us first consider the BFKL energy regime, the gggg dipoledipole cross-section reads
BFKL
(gg)(gg)
(r, r , Y ) =
2Nc s2 2
r
CF

s Nc
d (r /r)2
( )Y
exp
2i 2 (1 )2
(18)
which combined with (17) gives

d BFKL 3 Nc3 s2
=
2
dx1 dx2
8CF Q1

s Nc
d (Q1 /Q2 )2
exp
( )Y .
2i (1 )
(19)
One can easily show that the result is identical to the one obtains using kT -factorization [7]. As
in the forward-jet case, the only undetermined parameter is which appears in the exponential
in formula (19). We shall consider in this study the same value that was used for forward jets,
that is = 0.16.
For the saturation parametrization, we use the following gggg dipoledipole cross-section:
sat
(gg)(gg)
(r, r , Y ) =

2

r (r, r )
4Nc s2
0 1 exp eff 2
CF
4R0 (Y )
(20)
which up to the normalization is the same as (12). The effective radius reff is defined by for
mula (13) and the saturation radius by R0 (Y ) = e 2 (Y Y0 ) /Q0 with Q0 1 GeV. Inserting (20)
into (17), one obtains [31]

d sat
3 Nc3 s2 0
=
dx1 dx2
16CF

2Q1 Q2 uR02 (Y )
du
I1
1 2R02 (Y )Q1 Q2
1 + log(u)
1 + log(u)
1

Q21 + u2 Q22 2
Q22 + u2 Q21 2
exp
.
R (Y ) + exp
R (Y )
1 + log(u) 0
1 + log(u) 0

(21)
144
In the following we consider only the strong saturation parametrization to display what could
be the maximal expected effects at the LHC. The parameters are = 0.402 and Y0 = 0.82. The
normalization 0 is a priori not determined but we have fixed it so that at large momenta and
small Y , one obtains the BFKL result.
3.2. Phenomenology
We are going to study the dependence of the cross-sections (19) and (21) as a function of the
different kinematic variables x1 , x2 , Q1 , and Q2 . We want to consider large rapidities Y which
implies very forward jets and therefore large values of x1 and x2 . The well-known problem is that
the cross-section is then damped by the parton distribution functions which at large x become
very small. This prevents one to see the BFKL enhancement of the hard part of the cross-section
with rapidity.
A way out of this problem is to consider the following observables RS/S :
RS/S
d ppJ XJ
d ppJ XJ
(Q1 , Q2 , S)
(Q1 , Q2 , S),
dx1 dx2
dx1 dx2
(22)
in other words, cross-section ratios for same jet kinematics and two different values of the to The advantage of such observables is that they are independent
tal energy squared (S and S).
of the parton densities and allow to study more quantitatively the influence of small-x effects
[16,17,31]. For instance the BFKL-LL prediction is (via a saddle point approximation):
4 log(2)
S
RS/S
(23)
.
S
The experimental verification of this at the Tevatron [32] was not conclusive. The data were found
above the prediction (23), however it has been argued [33] that the measurement was biased by
the use of upper kT -cuts, the choice of equal lower kT -cuts, and hadronization corrections. The
ratios (22) also display in a clear way the saturation effects [17,31] which lead to ratios that, as
a function of Q1 and Q2 , go from the value (23) to 1 as the momentum cuts decrease into the
saturation region (see Fig. 4 in Ref. [17] and Figs. 3, 4 in Ref. [31]).
There is however an important experimental limitation to carry out the measurement (22):
it would require to run the LHC at two different center-of-mass energies. If it turns out not
to be possible, then one should settle for the cross-section d/dx1 dx
2 . We shall now exhibit
S = 14 TeV. Also the
some of its characteristics, fixing the LHC center-of-mass energy
at
absolute normalization is fixed to reproduce the Tevatron data at S = 1.8 TeV published in [32].
These data feature somewhat large error bars which leads to a significant uncertainty on the
normalization for the LHC predictions.
Let us introduce the rapidities of the two jets:

x1 S
x2 S
,
y2 = log
.
y1 = log
(24)
Q1
Q2
We first considered the case where one of the two jets has fixed kinematics Q2 = 30 GeV and
y2 = 4.5. We looked at the dependence of the cross-section d/dy1 dy2 = x1 x2 d/dx1 dx2 as
a function of the other jet kinematic variables. In Fig. 6, we plotted the results for the BFKL-LL
prediction (19) and the saturation parametrization (21) where the different plots feature the y1
dependence for different values of Q1 . As expected, the cross-sections decrease quickly as y1
145
Fig. 6. The MuellerNavelet jet cross-section d/dy1 dy2 as a function of y1 for different values of Q1 . The kinematics
of the other jet are fixed at Q2 = 30 GeV and y2 = 4.5. The full lines are BFKL-LL predictions and the dashed lines
are the saturation parametrization.
gets large which corresponds to x1 getting closer to one. For each value of Q1 , one cannot really
see a difference between the behaviors of the BFKL and saturation curves as a function of y1 .
However the relative normalization between the two curves is quite sensitive to the value of Q1 .
This is better exhibited on Fig. 7 where one displays the ratio of the saturation and BFKL results
of Fig. 6. The ratio goes down to about 0.3 for Q1 = 10 GeV which represents a significant
difference between the BFKL and saturation predictions. Note that this difference does not appear
to be that large on Fig. 6 where the cross-sections are plotted.
The second case we considered is the symmetric case Q Q1 = Q2 and y y1 = y2 which
allows to go to bigger values of Y . We looked at the dependence of the cross-section d/dy1 dy2
as a function of Q and y. In Fig. 8, we plotted the results for the BFKL prediction (19) and the
saturation parametrization (21) where the different plots feature the Q dependence for different
values of y. In this case, the cross-section falls even faster when y gets big as both x1 and x2 get
close to 1. Again, because of that, one does not see on the plot the difference between the BFKL
and the saturation curves, yet it is still quite big as shown on Fig. 9 where we have displayed
146
Fig. 7. Ratio of the saturation and BFKL-LL MuellerNavelet jet cross-sections d/dy1 dy2 as a function of y1 for
different values of Q1 . The kinematics of the other jet are fixed at Q2 = 70 GeV and y2 = 3.5.
the ratio of the saturation and BFKL results of Fig. 8. For y = 5.5 and Q decreasing down to
10 GeV, the ratio goes down to about 0.4.
We did not include in this study the weak saturation parametrization, the corresponding curves
would lie in between the BFKL and strong saturation curves which are displayed, and even is
closer to the BFKL curve. There is a number of other plots one could study showing other dependences of d/dy1 dy2 but they are not needed for drawing our conclusions: testing BFKL effects
and saturation effects with the observable d/dy1 dy2 at the LHC will be a major experimental
challenge as one will have to measure cross-sections with a great precision. We insist that this
is due to the fact that the parton distribution functions at large x really damp the cross-section.
Obtaining a high accuracy is not unfeasible because of the high luminosity at the LHC but this
will require a very good understanding of the systematics errors. However we would like to emphasize the fact that better tests of small-x effects could be realized with the measurement of the
ratio RS/S , see formula (22).
4. Conclusions
Let us summarize the main results of the paper. The first part of the work was devoted to
the study of the forward-jet measurement. We started by computing the QCD cross-section for
forward-jet production (1) in the high-energy (small-x) limit. We recalled the BFKL-LL predic-
147
Fig. 8. The MuellerNavelet jet cross-section d/dy1 dy2 as a function of Q Q1 = Q2 for different values of
y y1 = y2 . The full lines are BFKL-LL predictions and the dashed lines are the saturation parametrization.
tions (9) and also formulated the phenomenological model (16) that takes into account saturation
effects. We then compared the BFKL and saturation-model predictions to the recent data from
HERA for a number of observables: d/dx, d/dQ2 , d/dkT2 , and d/dx dQ2 dkT2 .
We obtained a very good agreement with the BFKL predictions and saturation parametrizations also show compatibility with the data. Along with the fact that QCD at NLO predictions
do not reproduce the small-x data, this observation leads us to the conclusion that the present
forward-jet data display the BFKL enhancement when going to small values of x.
However, to make a definitive statement, one would have to make comparisons with BFKL
predictions at next-to-leading-logarithmic (NLL) accuracy. The latter are under investigations
and will hopefully be available soon. In the mean time, let us discuss the expected qualitative
impact of these BFKL-NLL corrections. Because we are describing a kinematic regime in which
kT2 Q2 , one can infer that they should be small. For instance, the contribution coming from
the running of the coupling between those two scales would be unimportant. By comparison,
the BKFL-NNL corrections seem to be very large for the proton structure function measurement [34], in which case one easily understands why: the evolution takes place in a large range,
148
Fig. 9. Ratio of the saturation and BFKL-LL MuellerNavelet jet cross-sections d/dy1 dy2 as a function of
Q Q1 = Q2 for different values of y y1 = y2 .
from the soft proton scale up to the hard scale Q2 . The situation is much different for forward
jets.
In the second part of the paper, we investigated small-x effects for MuellerNavelet jets in the
LHC energy range, using the parameters that successfully describe forward-jets at HERA. We
compared the BFKL-LL predictions (19) with those of the saturation model (21) and concluded
that the measurement of the simple cross-section will require a great precision to test the different
scenarios. We argued that a better option to look for small-x effects was to measure the ratio of
cross-sections (22) which implies running the LHC at two different energies.
On longer time scales, the international linear collider would give the opportunity to measure
the virtual photonvirtual photon total cross-section at very high energies. This would also offer
great possibilities [35,36] for testing the BFKL enhancement and the saturation regime of QCD.
Acknowledgement
The authors would like to thank Robi Peschanski for useful comments and fruitful discussions.
Appendix A. On the integration method
To compare the forward-jet cross-section (1) obtained from the BFKL-LL prediction (9) and
the saturation parametrization (16) with the data for observables which are less differential
149
(d/dx, d/dQ2 , d/dkT2 , and d/dx dQ2 dkT2 ), one has to carry out a number of integrations over the kinematic variables. They have to be done while properly taking into account the
kinematic cuts applied by the different experiments. This appendix deals with these issues.
Let us start from the quadruple differential cross-section d/dx dQ2 dxJ dkT2 , see formula (1). First one performs the Mellin integrations of (9) and (16). Then we choose the appropriate variables for the remaining integrations: to avoid numerical problems in the integral
calculations, we chose variables which lead to the weakest possible dependence of the differential cross-section. We noticed that the best choice is 1/Q2 , 1/kT2 , log(1/xJ ), and log(1/x).
Since the experimental measurements are not differential with respect to xJ , we carry out the
integration

d (3)
xJ d (4)
1
.
=
d
log
2
2
2
2
xJ
dx dQ dkT
dx dQ dxJ dkT
With the BFKL-LL formula (9), the convergence is fast enough so that one can perform all
the remaining integrations to obtain any of the four observables mentioned above. With the saturation formula (16), because of the extra Mellin integration and the time it takes to compute the
functions , performing all the remaining integrations would require important numerical work.
We chose to use another method to obtain the cross-sections in that case. We shall describe it
now with the example of d/dx.
For a given value of x, the first step is to compute the differential cross-section

d BFKL
1
d (3) BFKL
1
= Q4 kT4
d 2
d
dx
Q2
dx dQ2 dkT2
kT
for the BFKL case. The second step is to compute the bin center (kT2 C , Q2C ) defined as follows:
d (3) BFKL 2 2 d BFKL /dx
QC , kT C
.
dQ2 dkT2
dx dQ2 dkT2
The bin center is thus the point in the (kT2 , Q2 ) phase space where the differential cross-section
in kT2 and Q2 is equal to the integral over the bin divided by the bin size (we will specify the
integration limits later on). The third step is to obtain the cross section for the saturation case.
We compute the cross-section at the bin center (kT2 C , Q2C ):
d sat
d (3) sat 2 2
=
QC , kT C
dx
dx dQ2 dkT2

dQ2 dkT2 .
This procedure is valid if the bin center does not change much between the BFKL and saturated
cross-sections. In other words, it means that the difference between the BFKL and saturated
cross-sections is small. We saw in Section 2.5 that this is indeed the case. The method is easily
adapted to the case of d/dQ2 for which one finds a bin center (xC , kT2 C ) for each value of Q2
and to the case of d/dkT2 with a bin center (xC , Q2C ) for each value of kT .
For the triple differential cross-section d/dx dQ2 dkT2 which is measured as a function of x,
integrating over xJ is not enough since one does not know the (Q2 , kT2 ) bin-center. Instead, for
a given value of x, one integrates also over Q2 and kT2 and then divide the result by the Q2 and
kT2 bin sizes to obtain d/dx dQ2 dkT2 . This is done for the BFKL case and one uses again the
method described above to compute the cross-section in the saturation case.
150
The other difficulty arises when setting the integration limits as one has to take into account the
correlations between the kinematic variables (for instance y = Q2 /sx < 1) and the cuts applied
by the experiments (for instance cuts on the forward jet phase space). There are two ways to take
these into account: either appropriately set the limits of integration while computing the integrals
or evaluate later the phase space correction due to the experimental cuts. We are going to use
both, since it is not possible to include all experimental cuts while computing the integrals.
Table 2 is a list of the different set of cuts used by the H1 and ZEUS experiments to carry
out their measurements. For the ZEUS cuts, we only consider what they call the forward-BFKL
phase space which corresponds the most to the Regge limit kinematics. Referring to this table,
let us enumerate the integration limits which are used for the different integral calculations:
d/dx for H1: we integrate over 1/kT2 with the limit on kT2 defined by 0.5 < kT2 /Q2 < 5
(this is an extra cut that H1 applies to this measurement only), over 1/Q2 with Q2 < sx, and
over log(1/xJ ) with 0.035 < xJ < 1
d/dx for ZEUS: we integrate over 1/kT2 with the limit on kT2 defined by 0.5 < kT2 /Q2 < 2,
over 1/Q2 with Q2 < sx, and over log(1/xJ ) with 2 < log(1/xJ ) < 3
d/dQ2 for ZEUS: we integrate over 1/kT2 with the limit on kT2 defined by 0.5 < kT2 /Q2 <
2, over log(1/x) with x Q2 /s, and over log(1/xJ ) between 2 and 3
d/dkT2 for ZEUS: we integrate over log(1/x) with x Q2 /s, over 1/Q2 with the limits on
Q2 defined by 0.5 < kT2 /Q2 < 2, and over log(1/xJ ) between 2 and 3
d/dx dQ2 dkT2 for H1: the 1/kT2 and 1/Q2 limits of the integrals are defined by the bin values measured by the H1 Collaboration with also the kinematic constraint 0.1 < y = Q2 /sx <
0.7. The log(1/xJ ) limits are obtained taking into account the cuts on the forward-jet angle
which leads to 1.7354 < log(1/xJ ) < 2.7942.
The effects of the cuts defined in Table 2 which are not used above need to be computed using
a toy Monte Carlo. They are modeled by bin-per-bin correction factors that multiply the crosssections obtained as described above. This is how one proceeds: we generate flat distributions
in the variables 1/kT2 , 1/Q2 , log(1/xJ ), and log(1/x) using reference intervals which include
the whole experimental phase-space (the azimuthal angle of the jet is not used in the generation since all the cross-section measurements are independent of that angle). In practice, we get
the correction factors by counting the numbers of events which fulfill the experimental cuts for
Table 2
ZEUS (forward-BFKL phase space) [19] and H1 [20] cuts to define the
forward-jet phase space. Ee is the energy of the outgoing electron and J =
log(1/xJ ) = log tan(J /2). The other kinematic variables have been defined
in the text
H1
ZEUS
Ee 10 GeV
0.1 y 0.7
104 x 4.103
5 < Q2 < 85 GeV2
kT > 3.5 GeV
7 J 20 degrees
xJ > 0.035
Ee 10 GeV
0.04 y 1
Q2 > 25 GeV2
kT > 6 GeV
2 < J < 3
0.5 < kT2 /Q2 < 2
151
each x-bin when we compute d/dx, each Q2 -bin when we compute d/dQ2 and so on. The
correction factors are obtained by the ratio of the number of events which pass the experimental cuts and the kinematic constraints to the number of events which fulfill only the kinematic
constraints, i.e., the so-called reference bin. Of course the experimental or kinematic cuts which
have been applied already while computing the integrals are not applied in this study to avoid
double counting effects.
This method allows for a direct comparison of the data with theoretical predictions but it does
not allow to control the overall normalization. This would require a full Monte Carlo. Note that
we did not use one in order to avoid any strong model dependence of the correction factors as
they are only due to kinematic-cut effects. The derivation of the correction factors is independent
of the theoretical input. They are given in Appendix B and they can be used to test any model
suitable for the forward-jet cross-section, providing the same integration method as described
above is used.
Appendix B. Tables with correction factors and resulting cross-sections
In this section, we list the corrections factors that we obtained for the observables d/dx
(H1 and ZEUS), d/dQ2 (ZEUS), d/dkT2 (ZEUS), and d/dx dQ2 dkT2 (H1). We also give
the resulting cross-sections for the different points that we used to draw the curves on Figs. 24
(Tables 37).
Table 3
Correction factors due to experimental cuts and the resulting corrected cross-sections for d/dx in nb for BFKL-LL,
weak saturation and strong saturation (see Fig. 2). Left table: for H1 cuts, right table: for ZEUS cuts
x
Factor
BFKL-LL
Weak sat.
0.00015
0.0005
0.0010
0.0015
0.0020
0.0025
0.0030
0.0035
0.0040
0.24
0.81
0.86
0.80
0.66
0.54
0.45
0.38
0.31
1200
805
371
205
117
72.0
47.1
32.3
22.5
1046
785
365
202
114
70.4
45.9
31.4
21.8
Strong sat.
897
722
360
203
116
71.7
46.9
32.1
22.3
Factor
BFKL-LL
Weak sat.
Strong sat.
0.00075
0.0017
0.004
0.01
0.025
0.13
0.43
0.48
0.34
0.13
39.3
62.5
25.4
3.33
0.106
34.5
58.1
24.7
3.38
0.109
31.1
56.0
24.6
3.34
0.106
Table 4
Correction factors due to experimental cuts and the resulting corrected cross-sections for BFKL-LL, weak saturation and
strong saturation. Left table: for d/dQ2 in pb/GeV2 , right table: for d/dkT in pb/GeV (see Fig. 3)
Q2 (GeV2 ) Factor BFKL-LL Weak sat. Strong sat.
kT (GeV)
Factor
BFKL-LL
Weak sat.
Strong sat.
35
65
150
350
950
7
9
12
17.5
25
0.32
0.31
0.28
0.22
0.15
71.9
21.1
6.23
1.01
0.119
72.4
20.2
5.59
0.828
0.0875
71.6
20.4
5.63
0.828
0.0875
0.60
0.70
0.88
0.88
0.96
6.75
1.40
0.184
0.0127
0.000611
5.36
1.03
0.139
0.0112
0.000549
3.81
0.843
0.137
0.0112
0.000549
152
Table 5
Correction factors due to experimental cuts and the resulting corrected cross-sections for d/dx dkT2 dQ2 in nb/GeV4
(bins with 12.25 < kT2 < 35 GeV2 ) for BFKL-LL, weak saturation and strong saturation (see Fig. 4)
kT2 (GeV2 )
Q2 (GeV2 )
Factor
BFKL-LL
Weak sat.
Strong sat.
12.25 < kT2 < 35
5 < Q2 < 10
12.25 < kT2 < 35
10 < Q2 < 20
12.25 < kT2 < 35
20 < Q2 < 85
0.0002
0.0004
0.0006
0.0008
0.001
0.0002
0.0004
0.0006
0.0008
0.001
0.0012
0.0014
0.0016
0.0018
0.002
0.001
0.0015
0.002
0.0025
0.003
0.0035
0.004
0.065
0.065
0.065
0.065
0.065
0.042
0.065
0.065
0.065
0.065
0.065
0.065
0.065
0.065
0.065
0.060
0.060
0.063
0.065
0.065
0.065
0.065
8.28
3.86
2.04
0.705
0.160
0.571
1.15
0.752
0.544
0.416
0.289
0.171
9.90e2
5.70e2
2.20e2
4.79e2
3.50e2
2.68e2
1.86e2
1.20e2
8.08e3
5.60e3
7.85
3.69
1.96
0.674
0.151
0.551
1.12
0.735
0.530
0.406
0.280
0.166
9.53e2
5.07e2
2.11e2
4.72e2
3.45e2
2.63e2
1.82e2
1.17e2
7.90e3
5.47e3
4.40
2.53
1.51
0.591
0.151
0.332
0.826
0.612
0.473
0.376
0.275
0.169
9.95e2
5.34e2
2.25e2
3.97e2
3.09e2
2.46e2
1.76e2
1.15e2
7.90e3
5.52e3
Table 6
(bins with 35 < kT2 < 95 GeV2 ) for BFKL-LL, weak saturation and strong saturation (see Fig. 4)
kT2 (GeV2 )
Q2 (GeV2 )
Factor
BFKL-LL
Weak sat.
Strong sat.
35 < kT2 < 95
5 < Q2 < 10
35 < kT2 < 95
10 < Q2 < 20
35 < kT2 < 95
20 < Q2 < 85
0.0002
0.0004
0.0006
0.0008
0.001
0.0002
0.0004
0.0006
0.0008
0.001
0.0012
0.0014
0.0016
0.0018
0.002
0.001
0.0015
0.002
0.0025
0.003
0.0035
0.004
0.22
0.22
0.22
0.22
0.22
0.14
0.22
0.22
0.22
0.22
0.22
0.22
0.22
0.22
0.22
0.20
0.20
0.21
0.22
0.22
0.22
0.22
3.93
1.86
1.05
0.314
7.41e2
0.272
0.551
0.360
0.261
0.204
0.142
8.40e2
4.82e2
2.57e2
1.08e2
2.49e2
1.85e2
1.42e2
9.97e3
6.53e3
4.47e3
3.15e3
3.35
1.57
0.836
0.250
6.72e2
0.250
0.519
0.340
0.245
0.188
0.131
7.80e2
4.52e2
2.42e2
1.02e2
2.44e2
1.81e2
1.38e2
9.67e3
6.33e3
4.34e3
3.04e3
3.28
1.82
1.08
0.321
7.58e2
0.240
0.521
0.355
0.263
0.213
0.149
8.79e2
5.04e2
2.68e2
1.12e2
2.55e2
1.91e2
1.46e2
1.02e2
6.63e3
4.47e3
3.13e3
153
Table 7
(bins with 95 < kT2 < 400 GeV2 ) for BFKL-LL, weak saturation and strong saturation (see Fig. 4)
kT2 (GeV2 )
Q2 (GeV2 )
Factor
BFKL-LL
Weak sat.
Strong sat.
95 < kT2 < 400
5 < Q2 < 10
95 < kT2 < 400
10 < Q2 < 20
95 < kT2 < 400
20 < Q2 < 85
0.0002
0.0004
0.0006
0.0008
0.001
0.0002
0.0004
0.0006
0.0008
0.001
0.0012
0.0014
0.0016
0.0018
0.002
0.001
0.0015
0.002
0.0025
0.003
0.0035
0.004
0.31
0.31
0.31
0.31
0.31
0.20
0.31
0.31
0.31
0.31
0.31
0.31
0.31
0.31
0.31
0.29
0.29
0.30
0.31
0.31
0.31
0.31
0.539
0.253
0.135
4.47e2
1.02e2
4.28e2
7.71e2
5.04e2
3.78e2
3.07e2
2.08e2
1.17e2
6.55e3
3.49e3
1.48e3
3.65e3
2.78e3
2.12e3
1.49e3
9.84e4
6.76e4
4.79e4
0.384
0.181
9.67e2
3.43e2
7.87e3
2.99e2
6.34e2
4.14e2
2.99e2
2.28e2
1.60e2
9.62e3
5.61e3
3.03e3
1.28e3
3.28e3
2.47e3
1.89e3
1.33e3
8.86e4
6.13e4
4.36e4
0.540
0.246
0.127
4.35e2
9.62e3
4.00e2
8.01e2
5.13e2
3.59e2
2.62e2
1.82e2
1.09e2
6.38e3
3.40e3
1.42e3
3.69e3
2.69e3
2.03e3
1.41e3
9.20e4
6.30e4
4.44e4
References
[1] L.N. Lipatov, Sov. J. Nucl. Phys. 23 (1976) 338;
E.A. Kuraev, L.N. Lipatov, V.S. Fadin, Sov. Phys. JETP 45 (1977) 199;
I.I. Balitsky, L.N. Lipatov, Sov. J. Nucl. Phys. 28 (1978) 822.
[2] L.V. Gribov, E.M. Levin, M.G. Ryskin, Phys. Rep. 100 (1983) 1.
[3] A.H. Mueller, J. Qiu, Nucl. Phys. B 268 (1986) 427;
E. Levin, J. Bartels, Nucl. Phys. B 387 (1992) 617.
[4] L. McLerran, R. Venugopalan, Phys. Rev. D 49 (1994) 2233;
L. McLerran, R. Venugopalan, Phys. Rev. D 49 (1994) 3352;
L. McLerran, R. Venugopalan, Phys. Rev. D 50 (1994) 2225;
A. Kovner, L. McLerran, H. Weigert, Phys. Rev. D 52 (1995) 6231;
A. Kovner, L. McLerran, H. Weigert, Phys. Rev. D 52 (1995) 3809;
R. Venugopalan, Acta Phys. Pol. B 30 (1999) 3731.
[5] E. Iancu, A. Leonidov, L. McLerran, Nucl. Phys. A 692 (2001) 583;
E. Iancu, A. Leonidov, L. McLerran, Phys. Lett. B 510 (2001) 133;
E. Iancu, L. McLerran, Phys. Lett. B 510 (2001) 145;
E. Ferreiro, E. Iancu, A. Leonidov, L. McLerran, Nucl. Phys. A 703 (2002) 489;
H. Weigert, Nucl. Phys. A 703 (2002) 823.
[6] A.H. Mueller, Nucl. Phys. B (Proc. Suppl.) 18C (1990) 125;
A.H. Mueller, J. Phys. G 17 (1991) 1443.
[7] A.H. Mueller, H. Navelet, Nucl. Phys. B 282 (1987) 727.
[8] G. Altarelli, G. Parisi, Nucl. Phys. B 126 (1977) 298, 18C;
V.N. Gribov, L.N. Lipatov, Sov. J. Nucl. Phys. (1972) 438;
V.N. Gribov, L.N. Lipatov, Sov. J. Nucl. Phys. (1972) 675;
Yu.L. Dokshitzer, Sov. Phys. JETP. 46 (1977) 641;
154
[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]
For a review: Yu.L. Dokshitzer, V.A. Khoze, A.H. Mueller, S.I. Troyan, in: J. Tran Thanh Van (Ed.), Basics of
Perturbative QCD, Editions Frontires, 1991.
A.H. Mueller, Nucl. Phys. B 415 (1994) 373;
A.H. Mueller, B. Patel, Nucl. Phys. B 425 (1994) 471;
A.H. Mueller, Nucl. Phys. B 437 (1995) 107.
I. Balitsky, Nucl. Phys. B 463 (1996) 99;
Y.V. Kovchegov, Phys. Rev. D 60 (1999) 034008;
Y.V. Kovchegov, Phys. Rev. D 61 (2000) 074018.
E. Iancu, A.H. Mueller, Nucl. Phys. A 730 (2004) 460;
E. Iancu, A.H. Mueller, Nucl. Phys. A 730 (2004) 494.
C. Marquet, Nucl. Phys. B 705 (2005) 319;
C. Marquet, Nucl. Phys. A 755 (2005) 603c.
K. Golec-Biernat, M. Wsthoff, Phys. Rev. D 59 (1999) 014017;
K. Golec-Biernat, M. Wsthoff, Phys. Rev. D 60 (1999) 114023.
H1 Collaboration, Phys. Lett. B 356 (1995) 118;
H1 Collaboration, C. Adloff, et al., Nucl. Phys. B 538 (1999) 3.
ZEUS Collaboration, J. Breitweg, et al., Eur. Phys. J. C 6 (1999) 239;
ZEUS Collaboration, Phys. Lett. B 474 (2000) 223.
J.G. Contreras, R. Peschanski, C. Royon, Phys. Rev. D 62 (2000) 034006;
R. Peschanski, C. Royon, Pomeron intercepts at colliders, in: Workshop on Physics at LHC, hep-ph/0002057.
C. Marquet, R. Peschanski, C. Royon, Phys. Lett. B 599 (2004) 236.
C. Marquet, Small-x effects in forward-jet production at HERA, hep-ph/0507108.
ZEUS Collaboration, S. Chekanov, et al., Forward jet production in deep inelastic ep scattering and low-x parton
dynamics at HERA, hep-ex/0502029.
H1 Collaboration, A. Aktas, et al., Forward jet production in deep inelastic scattering at HERA, hep-ex/0508055.
B.Z. Kopeliovich, A.V. Tarasov, A. Schafer, Phys. Rev. C 59 (1999) 1609;
A. Kovner, U. Wiedemann, Phys. Rev. D 64 (2001) 114002;
Y.V. Kovchegov, K. Tuchin, Phys. Rev. D 65 (2002) 074026.
H. Navelet, S. Wallon, Nucl. Phys. B 522 (1998) 237.
J. Bartels, A. De Roeck, M. Loewe, Z. Phys. C 54 (1992) 635;
W.-K. Tang, Phys. Lett. B 278 (1992) 363;
J. Kwiecinski, A.D. Martin, P.J. Sutton, Phys. Rev. D 46 (1992) 921.
E. Iancu, D.N. Triantafyllopoulos, Nucl. Phys. A 756 (2005) 419;
E. Iancu, D.N. Triantafyllopoulos, Phys. Lett. B 610 (2005) 253;
J.P. Blaizot, E. Iancu, K. Itakura, D.N. Triantafyllopoulos, Phys. Lett. B 615 (2005) 221;
Y. Hatta, E. Iancu, L. McLerran, A. Stasto, Color dipoles from bremsstrahlung in QCD evolution at high energy,
hep-ph/0505235.
A. Kovner, M. Lublinsky, Phys. Rev. D 71 (2005) 085004;
A. Kovner, M. Lublinsky, JHEP 0503 (2005) 001;
A. Kovner, M. Lublinsky, Phys. Rev. Lett. 94 (2005) 181603;
A. Kovner, M. Lublinsky, Densedilute duality at work: Dipoles of the target, hep-ph/0503155.
A.H. Mueller, A.I. Shoshi, S.M.H. Wong, Nucl. Phys. B 715 (2005) 440;
C. Marquet, A.H. Mueller, A.I. Shoshi, S.M.H. Wong, On the projectile-target duality of the color glass condensate
in the dipole picture, hep-ph/0505229.
A. Prudnikov, Y. Brychkov, O. Marichev, Integrals and Series, Gordon and Breach, New York, 1986.
S. Catani, M.H. Seymour, Nucl. Phys. B 485 (1997) 291;
S. Catani, M.H. Seymour, Nucl. Phys. B 510 (1997) 503, Erratum.
H. Jung, L. Jnsson, H. Kster, Eur. Phys. J. C 9 (1999) 383.
P. Aurenche, R. Basu, M. Fontannaz, R.M. Godbole, Eur. Phys. J. C 34 (2004) 277;
P. Aurenche, R. Basu, M. Fontannaz, R.M. Godbole, Eur. Phys. J. C 42 (2005) 43.
C. Marquet, R. Peschanski, Phys. Lett. B 587 (2004) 201;
C. Marquet, in: Deep Inelastic Scattering, Strbske Pleso, 2004, p. 352, hep-ph/0406111.
D0 Collaboration, B. Abbott, et al., Phys. Rev. Lett. 84 (2000) 5722.
J.R. Andersen, V. Del Duca, S. Frixione, C.R. Schmidt, W.J. Stirling, JHEP 0102 (2001) 007.
R. Peschanski, C. Royon, L. Schoeffel, Nucl. Phys. B 716 (2005) 401.
J. Bartels, A. De Roeck, H. Lotter, Phys. Lett. B 389 (1996) 742;
S.J. Brodsky, F. Hautmann, D.E. Soper, Phys. Rev. D 56 (1997) 6957;

M. Boonekamp, A. De Roeck, C. Royon, S. Wallon, Nucl. Phys. B 555 (1999) 540.
[36] N. Timneanu, J. Kwiecinski, L. Motyka, Eur. Phys. J. C 23 (2002) 513;
M. Kozlov, E. Levin, Eur. Phys. J. C 28 (2003) 483.
155
Gauge and modulus inflation from 5D orbifold SUGRA

Filipe Paccetti Correia a, , Michael G. Schmidt b , Zurab Tavartkiladze c
a Centro de Fsica do Porto, Faculdade de Cincias da Universidade do Porto,
Rua do Campo Alegre 687, 4169-007 Porto, Portugal

b Institut fr Theoretische Physik, Universitt Heidelberg, Philosophenweg 16, D-69120 Heidelberg, Germany
c Physics Department, Theory Division, CERN, CH-1211 Geneva 23, Switzerland
Received 20 October 2005; accepted 12 January 2006

Abstract
We study the inflationary scenarios driven by a Wilson line fieldthe fifth component of a 5D gauge field
and corresponding modulus fieldwithin S (1) /Z2 orbifold supergravity (SUGRA). We use our off shell
superfield formulation and give a detailed description of the issue of SUSY breaking by the F -component
of the radion superfield. By a suitably gauged U (1)R symmetry and including couplings with compensator
supermultiplets and a linear multiplet, we achieve a self consistent radion mediated SUSY breaking of no
scale type. The inflaton 1-loop effective potential has attractive features needed for successful inflation. An
interesting feature of both presented inflationary scenarios are the red tilted spectra with ns 0.96. For
gauge inflation we obtain a significant tensor to scalar ratio (r 0.1) of the density perturbations, while for
the modulus inflation r is strongly suppressed.
1. Introduction
Inflation is the only candidate which naturally evades numerous cosmological problems [1].
In order to have a sufficiently flat universe, a de Sitter type expansion with a slowly rolling scalar
inflaton field is needed. This requires a flat inflaton potential and for that supersymmetry (SUSY)
is believed to be crucial in a realistic model building [2]. A different possibility for realizing
a flat potential is that the inflaton field is a pseudo NambuGoldstone boson (PNGB) field [3].
E-mail addresses: paccetti@fc.up.pt (F. Paccetti Correia), m.g.schmidt@thphys.uni-heidelberg.de (M.G. Schmidt),

zurab.tavartkiladze@cern.ch (Z. Tavartkiladze).
doi:10.1016/j.nuclphysb.2006.01.023
F. Paccetti Correia et al. / Nuclear Physics B 739 (2006) 156168
157
However, this idea seems to be difficult to realize since it usually requires VEVs much higher than
the Planck scale: at such large VEVs, one might not trust the results obtained in the framework of
an (effective) quantum field theory (see a more detailed discussion in Ref. [4]). A nice and elegant
realization of the PNGB inflation scenario was proposed in [4] (and subsequent works [5,6]),
where an extra-dimensional construction was suggested and the PNGB inflaton is the Wilson
line field corresponding to the fifth component of a 5D U (1) gauge field. In this setting, the
flatness of the inflaton potential does not require unnatural assumptions and the model turns out
to be fully self consistent with an effective 4D quantum field theory setting. Although the idea of
Ref. [4] works without invoking SUSY, we think that (with its phenomenological and theoretical
motivations) it is worthwhile to study this type of scenarios in the framework of SUSY. Our recent
work [6] was dedicated to this issue and can be considered as a step towards the construction of
the SUSY gauge inflation scenario. The setting which we have proposed there was based on
a rigid on shell SUSY 5D construction of Ref. [7] with the fifth dimension compactified on a
circle S (1) . The radion superfield T was used for SUSY breaking by its auxiliary component
FT = 0.
The aim of the present paper is to extend studies to 5D orbifold supergravity (SUGRA). Our
construction is based on the off shell formulation of 5D conformal SUGRA developed by Fujita,
Kugo and Ohashi (FKO) [810] and uses the superfield approach suggested recently by us in
Ref. [11] (see also the subsequent Ref. [12]).1 This superfield approach turned out to be very
economical and powerful for studying various phenomenological and theoretical issues in 5D
[15]. Here, we first present in Section 2 the minimal setting which is needed in order to realize
gauge inflation. Then in Section 3 we give full account of the issue of SUSY breaking by the
radions auxiliary F -component. We show that to obtain flatness and selfconsistency, gauging of
a U (1)R part of the global SU(2)R symmetry and a linear multiplet with appropriate couplings
with the U (1)R gauge supermultiplet play an important role. In Section 4 we turn to the calculation of the one loop inflaton effective potential including the Wilson line and modulus fields. We
separately study two inflationary scenarios: in Section 5 the gauge inflation and in Section 6 the
inflation driven by the modulus field. We discuss features allowing to realize a natural inflation
and give a detailed study of some properties for both inflationary scenarios.
2. The setting
In this paper we will deal with two types of hypermultiplets. One is a compensator (denoted
by h) and is necessary for gauge fixing of the conformal symmetry. The second type of hypermultiplet is a physical onereferred to as a matter hypermultiplet (denoted by H). It will
play an important role in the generation of the inflaton (the Wilson line field) potential. In gen , H2 ), where
eral, 5D hypermultiplets H = (Ai , , Fi ) can be ordered into r pairs (H21
= 1, 2, . . . , r (see [810] for a detailed discussion). In the following, we will use the notation

H (H1 , H2 ) = H, H c ,
(1)
for such a pair (omitting the index )
and similar for the compensator h = (h1 , h2 ) = (h, hc ).
In the general discussion of the hypermultiplet case we will use H and understand that this
similarly applies to the compensator. Essential differences for the compensator hypermultiplet,
1 For original papers on off shell 5D SUGRA formulation see Ref. [13]. This formulation was used in many phenomenologically oriented papers [14].
158
will be pointed out throughout the text. The 5D hypermultiplet of Eq. (1) decomposes into a pair
of N = 1 4D chiral superfields with opposite orbifold parities [10]:

2

, 2iR2 , iM A + D 5 A 1 ,
H = A22 = A211

21

H c = A221
(2)
,
= A21 , 2iR21
, iM A + D 5 A 1
with
M Ai = igM I (tI ) Ai + Fi ,
1
Ai W50 Fi V5ij Aj 2i 5i ,
D 5 Ai = 5 Ai igW5
(3)
where (tI ) is the generator of the gauge group GI . In this paper we will deal only with Abelian
gauge groups. In this case the gauge coupling g should be replaced by g/2. The components
(, , F ) of a 4D chiral superfield are assumed to be of right handed chirality. Therefore,
the superfield with a left chirality in a two component notation is given by
= (, , F ) = + L 2 F .
(4)
We will use this basis during the calculations.

Besides the hypermultiplets, in this discussion, three 5D gauge supermultiplets (V , )I will
be considered. (i) The U (1)Z central charge symmetry corresponds to I = 0: (V , )I =0
(V0 , 0 ). This is a compensating gauge supermultiplet and, as was observed in [11], in the rigid
limit accounts for the radion superfield. In the covariant derivatives of Eq. (3) I = 0 does not
, but acts only on an auxiliary component F . This is a particular
participate in M I (tI ) and W5
i
property of the compensating I = 0 gauge supermultiplet. (ii) Our construction is based on a
gauged U (1)R symmetry whose corresponding gauge supermultiplet is (V , )I =R (VR , R ).
Only the compensator hypermultiplet h is charged under this group. (iii) Finally, we introduce
an U (1) gauge supermultiplet (V , )I =1 (V1 , 1 ), where
1 contains the fifth component of
a vector field which generates the Wilson line field = dy A5 . The latter being a 4D scalar
will play the role of the inflaton field in the following. Note that only the matter hypermultiplet
H is charged under (V1 , 1 ).
The orbifold Z2 parities (y y) of the introduced gauge supermultiplets are given as:
Z2 :
(V0 , VR , V1 ) (V0 , VR , V1 ),
(0 , R , 1 ) (0 , R , 1 ).
(5)
Therefore all 4D gauged U (1) symmetries are broken on the orbifold fixed points. As far as the
hypermultiplets are concerned, without loss of generality we can consider the following orbifold
parity prescriptions:
Z2 :
H H,
H c H c .
(6)
For all gauge fields (VI , I ) we introduce the following parameterization

Vab = gV q ba ,
ab = g q ba ,
with |
q | = 1.
With this, the hypermultiplet Lagrangian is given by [15]

ab
1
e(4)
L(hyper) = d 4 Wy 2Ha e2V Hb d 2 (H )a (y )ab Hb + h.c.
(7)
(8)
In case of the compensator, the Lagrangian Eq. (8) should come with opposite sign. The superoperator y is obtained by promoting y to an operator containing odd (under orbifold parity)
159
elements of the 5D Weyl multiplet (see [11] for a more detailed discussion), which do not have
any relevance for our purposes and can be ignored.
With the orbifold parity assignments given in (5) and (6) we should gauge the U (1) symmetries of (7) in 1 , 2 direction, i.e. q3 = 0.2 In (8) Wy is a real general type 4D supermultiplet
which contains part of the radion chiral superfield as [11]

1
(9)
T + T + .
2
This relation is useful to account for the radion coupling with hypermultiplets. Since all 4D gauge
superfields V have negative orbifold parities in this setting, they do not contain zero mode states
and will be irrelevant for us. Therefore, we will set further V = 0. Taking all this into account,
the action (8) for matter and compensator hypermultiplets can be written as:
Wy =
L(hyper)|V =0 = L(H ) + L(h),

1
e(4) L(H ) = d 4 T + T H H + H c H c

+ d 2 2H c y H + g1 1 ei1 H 2 ei1 H c2 + h.c.,

1
e(4) L(h) = d 4 T + T h h + hc hc

d 2 2hc y h + gR R eiR h2 eiR hc2 + h.c.,
(10)
where cos 1 = q11 , sin 1 = q21 , cos R = q1R , sin R = q2R .

3. SUSY breaking through the radion superfield
In our model, for SUSY breaking we will use a non-zero F component of the radion superfield T . As it was pointed out in Ref. [11], to obtain a flat tree-level potential for FT we need
to introduce a linear multiplet L which couples with the V I =R vector multiplet. As we will see
below, the rle of L is to insure a self consistent SUSY breaking. Assuming that L is neutral
under V I =R , its field content is [9]

L = Lij , i , E , N ,
(11)
where E is an unconstrained antisymmetric tensor field. The coupling action of L V I =R is
given by

ij
e1 L V I =R , L = YR Lij + 2 Ri i + 2i ia a Rj Lij

1
1
j
+ MR N 2i b b 2i ai ab b Lij + e1 F (WR )E .
2
4
(12)
2 The other possibility would be to gauge in the 3 -direction which implies the introduction of an odd gauge coupling.
This was used to obtain supersymmetric RandallSundrum models (see [11] and references therein, also [24] for a recent
study). From couplings in the 3 -direction we obtain effective potentials which are flat in the Wilson-line -direction.
For this reason we do not consider such couplings in this work.
160
L plays the role of a Lagrange multiplier. A variation with respect to the components of L leads
to
Ri = 0,
MR = 0,
ij
YR = 0,
F (WR ) = 0.
(13)
The last equation of (13) has the solution

W5R = const.
(14)
This is enough to insure a non-zero F component of T .

Consider the part of the action (10) which involves the compensator hypermultiplet. The relevant bosonic couplings have the form

2

1
1
c
iR

e(4) L(h) 2Fh + 5 h gR R e h + FT h
2

2

1
+ 25 hc gR R eiR h + FT h

2
2

1
R c
c

2Fhc 5 h + gR R e
h + FT h
2
2

1
iR c
c

+ 25 h gR R e
h FT h
2
2
2

+ gR F R eiR h eiR hc + h.c.,
(15)
where for the lowest components of the hypermultiplet we have used the same notation as for the
corresponding superfield. From (2), (3) we have
gR
1
W5R eiR h + FT h,
2
2
g
1
R
Fhc = Fhc + 5 h + i W5R eiR hc + FT hc ,
2
2

W50
W50
21
2
c
,
with Fh = i
Fh1 , Fh = i
Fh1
Fh = Fh 5 hc i
(16)
where the relations

iFT
V51 + iV52 = 5 ,
ey
i
R = W5R ,
2
with ey5 = 1
(17)
have been used. Taking into account all this and the constraints h = 1 , hc = 0, (15) reduces to

2
1
L(h) 2Fh igR W5R eiR 1 + FT 1 2|Fhc |2
e(4)
+

2
2
FT igR W5R eiR + gR 2 FR eiR + h.c. .

2
(18)
L
= 0 then have the solutions
The on-shell equations FLh = FLc = F
T
h
FT = igR W5R eiR ,
Fh = Fhc = 0.
(19)
161
Therefore, gauging U (1)R we have obtained a non-zero FT with a flat potential. This is a noscale SUSY breaking scenario with SUSY breaking mediated by the radion superfield.3 For a
discussion of this phenomenon within a 5D on shell construction see [16,17]. An FT = 0 is
important for transmitting the SUSY breaking into the matter sector. All states which carry an
SU(2)R index, couple with FT through the covariant derivative and obtain a soft SUSY breaking
mass. For instance, the zero mode of the 4D gravitino obtains a soft mass through the 5D grav1 + 2 ). The latter is a goldstinothe fermionic
itino kinetic term by mixing with 5 i(5L
5R
component of the radion superfield.
Concluding this section, let us comment on another role of the linear multiplet. It insures that
all other F -terms are zero. The first term of (12) can be written as
ij
YR Lij = 2YRa La = (FR L + h.c.) + ,
(20)
where L = L1 + iL2 and ellipses stand for terms which are irrelevant for us. Collecting together
all couplings containing FR , we have

1
L(FR ) = 2|FR |2 + FR gR 2 eiR + L + h.c.
e(4)
The conditions
FR = 0,
L
FR
(21)
L
= L
= 0 are satisfied by the solutions
L = gR 2 eiR .
(22)
Note, that without the coupling to the lowest component of the linear multiplet, we would not be
able to have FR = 0. The latter is needed for the F -flatness and a vanishing vacuum energy on
the classical level.
Together with the gauge inflation, below we will also study the inflation driven by the modulus
field M 1 . With M 1
= 0 the corresponding F1 will have the potential4 : de4 2Wy P (V5 )
N11 |F1 12 M 1 FT |2 . This gives
1
F 1 = M 1 FT ,
2
(23)
which will play an important role for calculation of the masses of KK states.
4. KK decomposition and inflaton potential
Now we are ready to derive the inflation effective potential. Relevant for us is the 4D chiral
superfield 1 which contains the fifth component A15 of U (1) and the corresponding (real) modulus M 1 as 1 = 12 (M 1 iA15 ). The superfield 1 has positive Z2 orbifold parity and therefore
the Wilson line field

= dy A5 = 2RA15 ,
(24)
3 In the FKO treatment the role of a F -VEV is played by the gauge field components W
T
5R VEV after a redefinition
N
N
i
R
with FT
= 0.
FT = FT igR W5R e
4 For the definition of prepotential P (V ) see [11].
5
162
and (the zero mode of) M 1 are y-independent 4D scalars. Taking all this into account, from (10)
with (23) one can easily derive the potential for the scalar components H , H c with phase redefi
nition H ei1 /2 H , H c ei1 /2 H c :

2

g1
i
1
M1
H c FT H c
V (H ) = 25 H
2
2R
2

g1
i
1 2
c
1

+ 25 H
M
H + FT H
2
2R
2

1

1
+ g1 M 1 FT H 2 H c2 + g1 M 1 FT H 2 H c2 .
2
2
With the parity assignment (6), the KK decomposition for H , H c is given by
1 (n)
ny
1
H cos ,
H=
H (0) +
R
2 R
2R n=1
Upon integration along the fifth dimension L(4) =
of the two real zero modes are

2
(0) 2
1 g1 1
R|FT | .
m = 2
4
2
R
2R
0
1 (n)
ny
Hc =
H sin .
R
2R n=1
The
mass2
(27)
(n +
(n)
(n)
(H H+ ), H (n) =
(n)
1 (H
2
matrices for appropriate nth KK modes are

(n)
(n)
H(n)
(n)
H
(26)
dy L(5) , one can easily see that the mass2
For KK states it is convenient to choose the basis H (n) =

(n)
H+ ).
(25)
g1 2
4 )
+ 14 |RFT |2 + 14 (g1 RM 1 )2
1
i(n + g4
)RFT
(n +
1
i(n + g4
)RFT
1 ,
2
g1 2
1
1
2
1 2 2R
4 ) + 4 |RFT | + 4 (g1 RM )
(28)
and similar for
get
H+(n)
states. For
mass2 s
of four real scalar states (per n = 0 KK state) we thus

2
2
2 (n)
2
(n)
g1 1
1
1
|RFT | + g1 M 1 .
m (H ) = m (H+ ) = 2 n +
4
2
2
R
(29)
A non-zero FT does not affect the masses of the fermionic components (H , H c ) coming
c because they are blind with respect of SU(2) . Therefore the masses of Majorana
from HM , HM
R
fermionic components are

1
(n)
n g1
, n = , . . . , ,
m (H ) =
R
4

1
m(n) ( H c ) =
(30)
n + g1
, n = , . . . , , n = 1.
R
4
Note that the spectrum in (29), (30) is equivalent to one obtained within the ScherckSchwarz
SUSY breaking scenario.
163
As we have already mentioned, integration of the states with -dependent masses induces
an effective 1-loop potential for . The potential will also depend on the modulus M 1 . Using
Poisson resummation for each KK modes contribution to the effective potential (or starting from
a worldline expression, see an Appendix in [6]), we finally obtain
V eff ( ) =

1
3
1 cos kR|FT | cos(kg4 R )
6
4
5
16 R
k
k=1

2
1
kg4 R|M |
1 + kg4 R|M | + kg4 R|M |
.
e
3
(31)
The effective
potential in
(31) is written in terms of canonically normalized 4Dscalar fields
= / 2R, M = 2RM 1 and dimensionless 4D gauge coupling g4 = g1 / 2R (for a
general cubic norm function the field M is canonically normalized in the global minimum with
M
= 0).
Notice that the divergent bosonic and fermionic contributions at k = 0 cancel exactly because of SUSY. In the limit FT 0 (unbroken SUSY) the effective one-loop potential van2
ishes. The potential in (31) is invariant under the shifts + kg2k41R , |FT | |FT | + 2k
kR
(k1,2 = integer) reflecting the invariance under 5D gauge symmetries. Besides the (M )dependent part, the potential gets a constant contribution by integration of states which are neutral
under (V1 , 1 ) but feel SUSY breaking through FT . These kind of states are for example the 4D
gravitino, the gauginos and (V1 , 1 ) neutral bulk hypermultiplets. Thus we add a constant part to
the potential in Eq. (31) and tune the former in such a way that the potential is zero in the global
minimum (this is the usual fine tuning of the 4D cosmological constant). Keeping the dominant
terms of (31), the inflaton potential will have the form

V = V eff k=1 + V0 ,
with V0 =

3
1
cos
R|F
|
.
T
16 6 R 4
Fig. 1. Effective potential as a function of X = g4 RM and Y = g4 R .
(32)
164
For FT = 0 and g4 R = 1, M = 0 the potential V is positive, and thus it drives de Sitter

expansion. Since V depends on two dynamical fields we will have inflation driven by these two
fields. Below we will study the two extreme cases where one of the fields lies in its minimum
and the inflation is driven by only one field. This allows an analytical study of the inflation
and spectral properties of the density perturbations. The analysis for two field inflation will be
presented elsewhere.
5. Gauge inflation
First we consider the inflation driven by and set M = 0. In this case the two slow roll
parameters are

(MPl )2 V 2 2
g4 R
(g4 RMPl )2 tan2
,
=
=
2
V
2
2

2
2 g4 R
2 V
2
|| = (MPl ) =
(33)
(g4 RMPl ) tan
1.
V
2
2
[V , V denote derivatives with respect to , and MPl = 2.4 1018 GeV.] For moderate values
of tan2 g42R the slow roll conditions , || 1 can be easily satisfied by properly suppressed
f
g4 . Therefore, this is a good framework for a natural inflation. The value at which the inflation
ends is determined from the conditions , || 1
f
g4 R
2

tan
(34)
2
g4 RMPl
(we are considering the interval 0 g4 R 1). The fulfillment of the slow roll conditions
allows to determine analytically the number of e-foldings during the corresponding time interval

1
f
N = 2
MPl

f
V
d
V

1
1
2 g4 R
2
.
1 + (g4 RMPl )
=
ln sin
2
2
(g4 RMPl )2
(35)
With this expression one can calculate the value which corresponds to the epoch when the
present horizon scale crossed outside the inflationary horizon scale. From the present observations we have NQ = 5560 and therefore we need g4 RMPl 1. We will use the latter relation
for approximating the exact expressions. Using (35) we get

0.79
NQ
2
Q
arcsin exp
(g4 RMPl )2

g4 R
2
g4 R
with NQ = 60, RMPl = 10, g4 = 1.4 103 .
We see that for this value tan2
Q
1
,
2NQ
Q
g4 R
2
(36)
is not small. The slow roll parameters
1
2
(g4 RMPl )2
2NQ
2
however are small enough and we therefore the relation 3H = V .
(37)
165
The quadrupole anisotropy of the temperature fluctuations due to the scalar perturbations can
be calculated according to expression [21]

T
5 V 3/2
(38)
=
,
3 V Q
T QS 60 MPl
and for our case is given by

T
1
V0 1/2
1
NQ
2
(g
=
sinh
RM
)
4
Pl
4
T QS
2
3 2 10 g4 RMPl MPl
g4
NQ
,

3
8 30 RMPl
(39)
MPl V 2
T 2
2
while the tensor to scalar ratio r = ( T
T )QT /( T )QS 2.16( V ) is
r 4.32 Q
6.8
.
NQ
(40)
From (39), (40) with NQ = 60, g4 = 1.4 103 , RMPl = 10 one obtains the measured value
6
( T
T )Q 6 10 . For the same values of the parameters one has relatively large r 0.11.
This is one of the remarkable feature of this inflationary scenario. The planned measurements of
the Planck satellite could detect such a value of the tensor contribution. It is interesting to note
that a quadratic inflaton potential gives a similar relation (r 6.8/NQ ), although the scenario
considered there differs from ours in various aspects. The spectral index ns = 1 + 2Q 6 Q for
the scenario considered here is
ns = 1
2
2 (g4 RMPl )2 .
NQ
(41)
We see that the spectrum is red-tilted. For parameters given in (36) we have ns 0.96, which is
compatible with WMAP data [22]. Combining WMAP and the Ly data [23] gives the restriction ns 0.96. In Table 1 we present the results for several cases satisfying this data. As we see
the tensor to scalar ratio r is significant while the spectral index practically shows no running.
Q
One can check that for presented cases > MPl . However, since corresponds to the gauge
1
field A5 one can be sure that 5D gauge invariance and locality will guarantee that there is no
undesirable corrections to the inflaton potential. The non-local operators, not respecting the shift
symmetry + g2k
, are suppressed by a factor e2RM5 [4], where M5 is the 5D Planck
4R
scale. For R M55 the suppression factor is e2RM5 1013 . Therefore, all kind of non-local
contributions can be safely ignored.
Table 1
6
Spectral properties from gauge inflation with different values of parameters and ( T
T )Q 6 10
NQ
RMPl
g4
ns
55
55
60
60
5
10
5
10
7.4 104
1.5 103
7 104
1.4 103
0.96
0.96
0.97
0.96
0.12
0.12
0.11
0.11
s
103 ddn
ln k
0.67
0.78
0.56
0.65
166
6. Modulus field driven inflation

Now we consider the case in which the inflation is only due to modulus field M , assuming
that is settled in its minimum g4 R = 1. For simplicity we will consider the norm function
1 N = (M 0 )3 M 0 (M 1 )2 . Using the constraint N = 2 , the field M is not canonically
normalized when M
= 0. For parameterization of the very special manifold we introduce a
new variable t such that
sinh t
M
M 0 = 1 cosh2/3 t,
M1 =
,
= 1
1/3
cosh
t
2R
1
with =
(42)
.
=
2R MPl
Then the kinetic term has the form

1
1
1
(43)
g(t)( t)2 = ( F )2 , with g(t) dt = dF, g(t) = 1 + tanh2 t,
2
2
3
where F is a canonically normalized field playing the role of the modulus inflaton. Using the
dimensionless variables
1
tP =
t,
R = RM ,
(44)
MPl
the derivatives can be written as
1 1 V R
dV
=
,
dF
MPl g R tP

d 2V
1 1 2 V R 2
V 2 R
1 R g
(45)
=
+
.
2 g 2
tP
R tP2
2g tP tP
dF 2 MPl
R
Using these relations we can calculate the slow roll parameters , which allows to determine
f
numerically the point tP corresponding to the end of inflation. The tPQ can be determined through
the number of e-foldings through the relation
tP
Q
NQ =
g(tP )
dtP .
2 (tP )
(46)
f
tP
Having determined tP we can calculate the quantities T /T , ns and r. The selection of g4 , R

6
should be done in such a way as to have T
T 6 10 . Numerical study shows that one can
have inflation both for large and small values of a g4 RMPl . For a 1 we have
g4 RM 1,
tp 1.
(47)
This means that the metric NI J is nearly diagonal and certain approximations can be done.
Namely, using (47) we obtain for a 1

g4
T
NQ
(48)

,
3
T Q 16 15 RMPl
2
2.16
ns 1
(49)
,
r
.
2
NQ
(g4 RMPl )2 NQ
167
Table 2
6
Spectral properties from modulus inflation with different values of parameters and ( T
T )Q 6 10
NQ
60
55
60
55
60
55
60
RMPl
g4
77
100
100
600
600
3000
3500
1.3 104
2.5 102
2.15 102
0.136
0.13
0.7
0.7
ns
0.96
0.96
0.96
0.96
0.97
0.96
0.97
1.4 104
5 105
5 105
0
0
0
0
s
103 ddn
ln k
1.05
0.86
0.75
0.54
0.54
0.64
0.54
In the limit a 1 we have

g4 RM 1,
tp 1,
and we obtain the following approximate values

T
10
NQ

,
4
T Q 48 5 (RMPl )2
2
6.5
,
r
.
ns 1
2
NQ
102 NQ
(50)
(51)
(52)
The exact numerical results are summarized in Table 2. They confirm that the approximations
which led to (48), (49), (51) and (52) work well. As we see, the spectrum here is also red tilted.
However, the tensor to scalar ratio r is strongly suppressed.
7. Discussion
We presented in the previous sections two different scenarios for inflation in the potential (32) plotted in Fig. 1. The first case, inflation in the M 1 = 0 axis, is essentially the gauge
inflation model of [4]. As we pointed out there, successful inflation in this direction requires
a Rg4 MPl 1 and g4 1 (see Table 1). The scenario we called modulus inflation does not
share these constraints. In fact it is possible to realize modulus inflation for both small and large a
(see Table 2), and since g4 102 a 1/2 , this scenario allows for a not too suppressed (4D) gauge
coupling and relatively large compactification radius. This opens up the possibility to embed the
modulus inflation scenario in orbifold GUTs. Note also that for a 1 we have Q < MPl and
therefore quantum gravity corrections should not play a rle.
Concluding, let us remark that within our analysis we have assumed that during inflation the
size of the extra dimension (R) is fixed. In our treatment R is related to the lowest component
(ey5 ) of the radion superfield. Its stabilization is needed and may be realized by one of the mechanisms which have been widely discussed in the literature [18,17,19,20]. If the extra-dimension
is stabilized in a way that our inflation scenario is not modified significantly, the above analysis
should remain valid. However this issue goes beyond the scope of this paper.
Acknowledgements
We thank Qaisar Shafi for discussion and interesting comments. The research of F.P.C. is
supported by Fundao para a Cincia e a Tecnologia (grant SFRH/ BD/4973/2001).
168
References
[1] A.H. Guth, Phys. Rev. D 23 (1981) 347;
A.D. Linde, Phys. Lett. B 108 (1982) 389.
[2] E.J. Copeland, A.R. Liddle, D.H. Lyth, E.D. Stewart, D. Wands, Phys. Rev. D 49 (1994) 6410, astro-ph/9401011;
G.R. Dvali, Q. Shafi, R.K. Schaefer, Phys. Rev. Lett. 73 (1994) 1886, hep-ph/9406319.
[3] K. Freese, J.A. Frieman, A.V. Olinto, Phys. Rev. Lett. 65 (1990) 3233;
F.C. Adams, J.R. Bond, K. Freese, J.A. Frieman, A.V. Olinto, Phys. Rev. D 47 (1993) 426, hep-ph/9207245.
[4] N. Arkani-Hamed, H.C. Cheng, P. Creminelli, L. Randall, Phys. Rev. Lett. 90 (2003) 221302, hep-th/0301218;
N. Arkani-Hamed, H.C. Cheng, P. Creminelli, L. Randall, JCAP 0307 (2003) 003, hep-th/0302034.
[5] D.E. Kaplan, N.J. Weiner, JCAP 0402 (2004) 005, hep-ph/0302014;
B. Feng, M.Z. Li, R.J. Zhang, X.M. Zhang, Phys. Rev. D 68 (2003) 103511, astro-ph/0302479.
[6] R. Hofmann, F. Paccetti Correia, M.G. Schmidt, Z. Tavartkiladze, Nucl. Phys. B 668 (2003) 151, hep-ph/0305230.
[7] D. Marti, A. Pomarol, Phys. Rev. D 64 (2001) 105025, hep-th/0106256.
[8] T. Fujita, K. Ohashi, Prog. Theor. Phys. 106 (2001) 221, hep-th/0104130.
[9] T. Fujita, T. Kugo, K. Ohashi, Prog. Theor. Phys. 106 (2001) 671, hep-th/0106051.
[10] T. Kugo, K. Ohashi, Prog. Theor. Phys. 108 (2002) 203, hep-th/0203276.
[11] F. Paccetti Correia, M.G. Schmidt, Z. Tavartkiladze, Nucl. Phys. B 709 (2005) 141, hep-th/0408138.
[12] H. Abe, Y. Sakamura, JHEP 0410 (2004) 013, hep-th/0408224;
H. Abe, Y. Sakamura, hep-th/0501183.
[13] M. Zucker, Nucl. Phys. B 570 (2000) 267, hep-th/9907082;
M. Zucker, JHEP 0008 (2000) 016, hep-th/9909144;
M. Zucker, Phys. Rev. D 64 (2001) 024024, hep-th/0009083;
M. Zucker, Fortschr. Phys. 51 (2003) 899.
[14] G. von Gersdorff, M. Quiros, A. Riotto, Nucl. Phys. B 634 (2002) 90, hep-th/0204041;
R. Rattazzi, C.A. Scrucca, A. Strumia, Nucl. Phys. B 674 (2003) 171, hep-th/0305184, and references therein.
[15] F. Paccetti Correia, M.G. Schmidt, Z. Tavartkiladze, Phys. Lett. B 613 (2005) 83, hep-th/0410281.
[16] Z. Chacko, M.A. Luty, JHEP 0105 (2001) 067, hep-ph/0008103.
[17] M.A. Luty, N. Okada, JHEP 0304 (2003) 050, hep-th/0209178.
[18] E. Ponton, E. Poppitz, JHEP 0106 (2001) 019, hep-ph/0105021;
S. Nasri, P.J. Silva, G.D. Starkman, M. Trodden, Phys. Rev. D 66 (2002) 045029, hep-th/0201063;
G. von Gersdorff, M. Quiros, A. Riotto, Nucl. Phys. B 689 (2004) 76, hep-th/0310190;
P. Bucci, B. Grzadkowski, Z. Lalak, R. Matyszkiewicz, JHEP 0404 (2004) 067, hep-ph/0403012;
T. Kobayashi, K. Yoshioka, JHEP 0411 (2004) 024, hep-ph/0409355, and references therein.
[19] E. Dudas, M. Quiros, Nucl. Phys. B 721 (2005) 309, hep-th/0503157.
[20] N. Maru, N. Okada, Phys. Rev. D 70 (2004) 025002, hep-th/0312148.
[21] A.R. Liddle, D.H. Lyth, Phys. Rep. 231 (1993) 1, astro-ph/9303019;
A.R. Liddle, D.H. Lyth, Cosmological Inflation and Large-Scale Structure, Cambridge Univ. Press, Cambridge,
2000;
G. Lazarides, in: Lecture Notes in Physics, vol. 592, Springer-Verlag, Berlin, 2002, p. 351, hep-ph/0111328, see
also references therein.
[22] H.V. Peiris, et al., Astrophys. J. Suppl. 148 (2003) 213, astro-ph/0302225.
[23] U. Seljak, et al., Phys. Rev. D 71 (2005) 103515, astro-ph/0407372.
[24] T. Flacke, B. Hassanain, J. March-Russell, hep-ph/0503255.
A quantization of topological M theory

Lee Smolin
Perimeter Institute for Theoretical Physics, 35 King Street North, Waterloo, Ontario N2J 2W9, Canada
Department of Physics, University of Waterloo, Waterloo, Ontario N2L 3G1, Canada
Received 9 May 2005; received in revised form 10 January 2006; accepted 16 January 2006
Abstract
A conjecture is made as to how to quantize topological M theory. We study a Hamiltonian decomposition
of Hitchins 7-dimensional action and propose a formulation for it in terms of 13 first class constraints. The
theory has 2 degrees of freedom per point, and hence is diffeomorphism invariant, but not strictly speaking
topological. The result is argued to be equivalent to Hitchins formulation. The theory is quantized using
loop quantum gravity methods. An orthonormal basis for the diffeomorphism invariant states is given by
diffeomorphism classes of networks of two-dimensional surfaces in the six-dimensional manifold. The
Hamiltonian constraint is polynomial and can be regulated by methods similar to those used in LQG.
To connect topological M theory to full M theory, a reduction from 11-dimensional supergravity to
Hitchins 7-dimensional theory is proposed. One important conclusion is that the complex and symplectic structures represent non-commuting degrees of freedom. This may have implications for attempts to
construct phenomenologies on CalabiYau compactifications.
2006 Published by Elsevier B.V.
1. Introduction
Approaches to quantum gravity have so far fallen into two broad classes, according to whether
they are background independent or background dependent. So far most work on string and
M theory has been based on background dependent methods and ideas. But it has long been
acknowledged that this was a temporary expedient and that the ultimate principles of string theory
must be formulated in background independent terms. Meanwhile, a great deal of progress has
been made on background independent approaches, including loop quantum gravity [1,2], causal
sets [3] and Lorentzian dynamical triangulations [4].
E-mail address: lsmolin@perimeterinstitute.ca (L. Smolin).
0550-3213/$ see front matter 2006 Published by Elsevier B.V.
doi:10.1016/j.nuclphysb.2006.01.016
170
L. Smolin / Nuclear Physics B 739 (2006) 169185
The results of these, especially loop quantum gravity (LQG), have inspired a few attempts to
approach string or M theory from a background independent perspective [5,6]. These make use
of one of the most powerful observations of LQG, which is that theories of gravity are closely
related to topological field theories [2]. The precise relation is that gravitational theories are
constrained topological field theories. This means that their action is a sum of the action for a
BF theory, plus quadratic constraints. These are sometimes called theories of forms, because the
metric information is coded into the dynamics of forms [7,8].1 This is true of general relativity
in all dimensions [9], as well as of supergravity in 11 dimensions [5], so it is a fact that must be
relevant for how we formulate M theory.
Recently Dijkgraaf et al. [12] proposed a form of topological M theory, which is a sevendimensional theory which is hypothesized to unify two six-dimensional theories called topological string theories.2 This theory is defined by an action proposed by Hitchin [10,11], and
involves only the dynamics of a three-form in seven dimensions. Dijkgraaf et al. in fact propose
that this theory is related by dimensional reduction to topological field theories relevant for three
and four-dimensional theories. This makes it natural to suggest that the quantization of Hitchins theory may be accomplished by using background independent methods which have been
successfully applied to topological theories and theories of forms in lower dimensions.
In this paper we make a first attempt at such a background independent quantization of
topological M theory. In the next section we propose a form of the theory as a constrained
Hamiltonian theory. We find that the dynamical variables are coded into a two form and densitized bivector, , on a compact six manifold . These are canonically conjugate to each other
and are associated with the specification of two structures that go into the definition of a Calabi
Yau manifold, which are, respectively, a complex and symplectic structure. We find a system
of first class constraints relating them, which we argue is equivalent to the dynamics described
earlier by Hitchin in [10,11].
In Section 3 we count the local degrees of freedom, using standard methods. We find there
are two local degrees of freedom per point. Thus, if the proposal made in this paper is correct,
topological M theory is not actually a topological field theory.
In Section 4 we then quantize the local degrees of freedom, following the methods of LQG. We
find a theory of extended objects living in the six-dimensional manifold, . These are described
by observables parameterized by membranes and four-dimensional branes in . These involve,
respectively, the complex structure and symplectic structures on . We find that the quantum
states of the theory have a separable basis in one-to-one correspondence with the diffeomorphism
equivalence classes of the membranes embedded in the six manifold.
In the classical theory of Hitchin, the complex and symplectic structures each give a volume
to , and these are required to be equal to each other. In the Hamiltonian formulation presented here, this condition is expressed by a Hamiltonian constraint. Its quantization leads to
analogues of the WheelerdeWitt equations. This has a form not seen before, being cubic rather
than quadratic in momenta. We are able to use LQG methods to express the WdW operator as
a limit of a sequence of regulated operators. Unlike LQG, the operator is the sum of two terms,
and no easy solutions are apparent.
Finally, in Section 5, we show how the degrees of freedom of Hitchins theory arise from a
dimensional reduction of 11-dimensional supergravity in which the frame fields are set to zero.
1 In 4 spacetime dimensions, these turn out to be the self-dual two forms of a metric.
2 Related papers are [13].
171
While these results may be seen as a first sketch of a quantum theory, there is one intriguing question, raised in [12], that confronts us. The definition of a CalabiYau manifold requires
fixing both the complex and symplectic structures. Here we find that those structures do not commute with each other quantum mechanically. Thus, the use of CalabiYau manifolds to describe
compactifications of string and M theory can only be sensible at a semiclassical level in which
one works on a fixed, classical background geometry. Once quantum gravity effects are turned
on, an uncertainty principle may prevent a quantum state as being identified as a CalabiYau
manifold. This is true in Hitchins theory, as pointed out in [12], but the fact that the degrees of
freedom of that theory arise from a compactification of 11-dimensional supergravity suggest it
will be true also in M theory.
This gives rise to several fascinating questions that future work may address.
Might there be quantum effects of order lPl that arise from the quantum fluctuations of the
CalabiYau geometry? Could this lead to new kinds of effects, perhaps observable in experiments such as AUGER and GLAST?
Might the quantum fluctuations in the CalabiYau geometries help to stabilize then quantum
mechanically against decay to the negative energy density states found by [15]?
If the CalabiYau compactifications do not correspond to quantum states of the fundamental
theory, but only arise in the classical limit, there are implications for how they are to be
counted in considerations of the landscape of theories.
2. Hamiltonian formulation of Hitchins theory
Hitchin described a seven-dimensional theory [10,11], which Dijkgraaf et al. propose is a
formulation of topological M theory [12]. We begin by reviewing their proposal.
2.1. Review of topological M theory
The theory is defined on a 7-dimensional manifold, M. There is only one field, which is a
real three form , with fixed cohomology class.3
Analogously to how the metric in LQG in 4d is formed from a set of two forms, we can
construct a metric on M, h() depending only on . As in the 4d case [7,8], the densitized
metric is cubic in the form field. We have
h ab =
hhab = acd bef ghi cdefghi .
The action of Hitchin is

N
h().
I =
(1)
(2)
This gives rise to a non-trivial theory when the cohomology class of is frozen. One then has
= 0 + d
3 [12] study the theory in a dual formulation written in terms of a four-form on M.
(3)
172
for a two form. With 0 fixed the action is a functional of d and hence generates an interesting dynamics.

I N [] =
(4)
h 0 + d .
M
Put in this form, the action is invariant under gauge transformations parameterized locally by
a one form
= + d.
(5)
There are six of these per point of M, because and generate the same gauge transform on
when = + df .
2.2. Hamiltonian constrained systems
To quantize any theory we must first cast it into Hamiltonian form.4 Dirac long ago discovered
how to construct a Hamiltonian system for a theory invariant under the diffeomorphisms of a
(d + 1)-dimensional manifold [1]. One considers the manifold to have the form R where
is called the spatial manifold.5 There are d constraints that generate the diffeomorphisms
of , called Di where i, j = 1, . . . , d is a spatial index. There is a Hamiltonian constraint H
that generates the remaining diffeomorphisms in R. Any additional gauge symmetries are
generated by constraints G. These constraints must form a first class algebra, which means that
they close under Poisson brackets.
In the spatially compact case, which we will assume here, the Hamiltonian must be a linear combination of these constraints. Hence, it must be possible, by a change of variables, to
transform the action to the following form,

ij

I H = dt
(6)
ij a G a N i Di N H ,
where ij is the momenta conjugate to ij , while a , N i and N are Lagrange multipliers.

We next proceed to construct such a theory that we conjecture is equivalent to the theory of
Hitchin [10].
2.3. Topological M theory as a Hamiltonian constrained system
The action given by Hitchin can be rewritten in a form suggested by Hawkins [14]. Let gab
be an arbitrary metric on M. Then

E
I [g, ] =
(7)
g g ab h ab ()
M
gives the same equations of motion as I N when both g and are varied.
4 Some elements of the Hamiltonian formulation were described by Hitchin in [11].
5 There is no need to assume the degrees of freedom include a metric, so there is not necessarily a distinction between
timelike and spacelike.
173
Now choose the manifold to be of the form M = R with a compact 6 manifold. Form
indices in will be denoted, i, j = 1, . . . , 6. We fix a time coordinate and hence a slicing of M
and define canonical momenta
ij =
I E
,
ij
(8)
where dot denotes as usual derivative by the coordinate on R, called t.

There are 6 initial primary constraints given by
0i =
I E
= 0.
0i
(9)
The Poisson algebra is generated by

ij (x), kl (y) = ijkl 6 (x, y)
(10)
from which we see that the momenta carry density weight one. This means that the dual is a
four-form, = . In six dimensions, a generic four form is stable (see [10,12] for the meaning
of this term) and can be written equivalently in terms of a two-form k, as
= k k.
(11)
In LQG there is an analogous situation, and one chooses a single sector to construct the quantum
theory. We will then restrict ourselves in the following to the positive sector, in which
= +k k.
(12)
In this, generic configurations may then be expressed equivalently in terms of k or ij . For the
canonical quantum theory, the latter is more convenient, as we will see below.
We can see how the action depends on velocities by noting that, in an obvious notation,
h 00 = (ij d0i )3 .
(13)
Hence the action (7) is roughly of the form,

IE
dt (ij d0i )3 A3 + (ij d0i )2 A2 + (ij d0i )A1 + potential ,
(14)
h ij = (ij d0i )(dij )2 ,
h i0 = (ij d0i )2 dij ,
where the AI are polynomials of spatial derivatives of ij . As a result we will find an equation
of the form
ij (kl )2 A3 + (kl )A2 + constants.
(15)
It is not straightforward to invert this relation to find kl as a function of ij . It may be possible

to do this, but for the present we proceed by making an educated guess for the form of the
Hamiltonian theory based on our experience with other diffeomorphism invariant systems, and
checking its internal consistency as well as its agreement with known results about Hitchins
action. We find such a conjecture, and describe it here. I believe, but have not shown, that the
system of constraints described here, is a restatement of previous results on this system [10,12].
We expect that the inversion of (15) is only possible modulo a system of constraints. This
system of constraints will include generators of all local gauge invariances of the theory.
174
We expect a total of 13 first class constraints. Six will generate the gauge transformations (5)
in . These must have the form,
G i = k ik = 0.
(16)
These form an Abelian algebra.

Six constraints will generate local diffeomorphisms of .6
They will be given by
Di = ij k j k = 0.
(17)
Let us integrate these against a vector field v i , with compact support on a contractible region
of .

D(v) = v i ij k j k .
(18)
It is straightforward to express this as

D(v) =
(Lv j k ) j k 2v i ik G k + v i ij0 k j k .
(19)
If we ignore the last term, then we see that D(v) generate a linear combination of diffeomorphisms and gauge transformations (5) on . However, in a compact, topologically trivial region,
we can take 0 = d 0 , so that the last term is included in the previous terms. It is then straightforward to show that the algebra of gauge and diffeomorphism constraints (16), and (17), closes,
so long as the constraints are multiplied by functions with support on contractible regions.
Now we come to the dynamics. For a diffeomorphism invariant theory on a spatially compact
manifold without boundary, this is going to be specified by a Hamiltonian constraint H, which
must be a local density on . For such a theory it is a general result that the Hamiltonian must
be proportional to constraints. The only exception is that there can be a non-vanishing boundary
term, but we are considering here the case of a manifold without boundary.
As the action contains terms up to cubic in ij we expect H to have terms up to cubic in ij .
By analogy with the Ashtekar formalism, we may expect that the Hamiltonian constraint will be
polynomial in the fields when written as a density of weight two. There are two monomials in
and that give us a scalar of weight two. The first is the simplest scalar density polynomial in
the ij , which is,
K = ij kl mn ij klmn .
(20)
This is a kind of kinetic energy. For a potential energy we seek a scalar of density weight two
j
polynomial in the ij k . One exists, defined by Hitchin as follows. Let i be the densitized, (1, 1)
tensor,
j
i = ikl mno klmnoj .
(21)
6 In [12] and [10] another form of the diffeomorphism constraint is proposed. It is plausible, but not yet shown, that
the two proposed forms are equivalent, at least on the space of solutions to (16).
175
Note that the trace ii = 0. However the trace of the square is not zero, and it gives a scalar
density of weight two,
j
V = i ji .
(22)
Combining them, we have a natural candidate for the Hamiltonian constraint,7 which is
H = K aV,
(23)
where a is a dimensionless factor.

We can check this guess by seeing if it leads to a constraint algebra that closes. The fact that
H is a scalar density of weight two determines that its Poisson brackets with (16) and (17) closes,
so long as the gauge transformations and diffeomorphisms have compact support on contractible
regions. To compute the rest of the Poisson algebra we smear against a test function N of density
weight minus one, again with compact support in a topologically trivial region.

j
H(N ) = N ij kl mn ij klmn a i ji .
(24)
It is straightforward to check that the algebra closes

j
H(N ), H(M) = wN M Dj = D(wN M ),
(25)
where
j
wN M = 18a(Ni M Mi N ) ik k .
(26)
Di and H make a closed system of

Thus, we see that the combination of the 13 constraints,
first class constraints.
In fact, we can argue that its solutions are identical to the solutions of Hitchins theory. When
H = 0 we have locally
Gi ,
a i ji = ij kl mn ij klmn .
We can take the square root of each side to find that
a j i
=
ij kl mn ij klmn
.
i j
(27)
(28)
We can find a geometric interpretation of the Hamiltonian constraint. To do so we note that the
is known to characterize the complex structure of [10,12]. The densitized bivector ij
provides a symplectic structure. These fields allow us to form two different volume elements on
the six manifold .
There is a volume element associated with the symplectic structure,
=
ij kl mn ij klmn
.
(29)
There is similarly a volume element associated with the complex structure, given by the three
form metric h(), pulled back into the six manifold .
h =
i ji
.
(30)
7 This is related to a form of the Hamiltonian studied by Hitchin in [11].
176
The Hamiltonian constraint says that the two volume forms are equal to each other, up to the
constant a. In [10,12] we see that Hitchins theory implies that

2 h = .
(31)
We see that this condition is implied by the guess for the Hamiltonian constraint we gave, (23)
so long as a = 1/4. Hence, the diffeomorphism classes of solutions to the theory given here will
coincide with the solutions of Hitchins theory.
Hitchin [10] also provides a translation to the complex geometry of 6 manifolds. He shows
(Proposition 2) that when
V <0
(32)
one can define a complex structure

j
=i ,
(33)
V
such that is the real part of a complex holomorphic three-form.
To summarize we have argued that the Hitchin action can be rewritten as a constrained Hamiltonian system of form, (6) with constraints given by (16), (17) and (23). We have not constructed
an explicit map between Hitchins theory and the one described here, but we note that both
formulations have diffeomorphism constraints that generate the diffeomorphisms of the spatial
surfaces, and that every solution to Hitchins equation (31) is a solution to our Hamiltonian constraint (23). This plus the consistency of the constraint algebra makes it very plausible that there
is at least a local equivalence between the two formulations.
We also reach the important conclusion mentioned in the introduction, that the complex and
symplectic structures are coded by canonically conjugate degrees of freedom, so long as (32) is
imposed.
j
Ji
3. Counting of degrees of freedom

It is straightforward to count the local degrees of freedom. There are 15ij which have 15
conjugate momenta ij . We have 13 first class constraints, which will require 13 gauge fixing
conditions. This leaves 2 + 2 canonical degrees of freedom. Thus the theory is not topological,
there are two local degrees of freedom per point of .
There are, of course, also global degrees of freedom, that correspond to integration of
around non-contractible cycles of .
4. Quantization
Dirac proposed a method to quantize Hamiltonian constrained systems. With some refinements to take into account issues of regularization and ordering that arise in field theories, this
is the method that all background independent approaches to Hamiltonian quantization follow.
Diracs method can be further specialized to the case of diffeomorphism invariant theories whose
configuration variables are connections or p-forms with local gauge invariance [1,2]. This specification of Diracs method to theories invariant under both diffeomorphisms and local gauge
invariances is the essence of the Hamiltonian part of loop quantum gravity. We first briefly summarize the procedure, then we apply it to the constrained system just introduced.
177
4.1. Brief review of Dirac quantization

We begin by specifying the kinematical configuration space, C. In the case of topological M
theory, this is the space of two forms on . By imposing invariance under the action of the
gauge and spatial diffeomorphism constraints, in this case (16) and (17), we then go down to a
gauge and (spatially) diffeomorphism invariant configuration space
C diffeo =
C
.
local gauge transformations Diff()
(34)
The aim of the quantization procedure is to first, construct the corresponding Hilbert spaces and,
second, construct the Hamiltonian constraint as an operator on diffeomorphism invariant states.
This is accomplished in three steps:
Step 1
Step 2
Step 3
Find an algebra A of observables on the kinematical phase space which has a representation A on a Hilbert space H kinematical such that
(1) The reality conditions of the classical theory, i.e. which variables are real, are realized by the inner product on H kinematical . That is, the inner product is chosen so that
real classical observables are represented by Hermitian operators.
(2) H kinematical carries an exact, non-anomalous unitary representation of Diff(). This
is given by unitary operators, U (), where Diff().
, which are
Construct a space of diffeomorphism invariant states H diffeo Hkinematical
invariant under the action of U (). These are the diffeomorphism invariant states and
they live inside the dual of the kinematical Hilbert space.
Construct a sequence of regularized operators, H (x) to represent the Hamiltonian constraints, in H kinematical . Prove that the limit as 0 takes diffeomorphism invariant
states to diffeomorphism invariant states, and thus defines a finite operator in H diffeo .
Prove that the limit has a kernel in H diffeo that is infinite-dimensional. This kernel
H physical H diffeo is the physical Hilbert space.
When carried out in LQG there are four key observations, that may extend to the present case
There is no known way to realize the second condition of Step 1 when A is the usual local
canonical algebra defined by the gauge connection and conjugate electric fields. In particular, Fock representations fail because they depend on a background metric, which breaks
diffeomorphism invariance. To proceed one must base A on extended observables, such as
Wilson loops.
When A is taken to include the Wilson loops of the connection, together with conjugate
operators linear in the momenta of the connection, there is a theorem [17] that says that
there is a unique way to realize the first two steps. It is not known if this extends to the
present case, but if it does there would appear to be only one way to successfully carry out
this program for topological M theory.
The kinematical Hilbert space H kinematical is not separable, because any two distinct, nonoverlapping, Wilson loop operators create orthogonal states. However, this non-separability
is exactly cancelled by imposing diffeomorphism invariance. Hence H diffeo is separable,
assuming only a technical condition, which is that it is defined in terms of piecewise smooth
diffeomorphisms.
When applied to general relativity, all three steps have been carried out rigorously [16].
178
4.2. Quantum topological M theory

We here sketch how the program just described may be applied to topological M theory. We
do not attempt to give a rigorous treatment, but we find that at a particle physics level of rigor we
can follow the same program as was originally used in constructing LQG.
We begin by finding the algebra of observables analogous to Wilson loops and their conjugate
variables, to represent the local degrees of freedom. We start with the analogue of Wilson loops.
Given any closed and contractible two surface S in we define a function of C,
T [S] = e
(35)
Similarly, given a four-dimensional surface A we define momentum flux operators

[A] = ,
(36)
where

is a four formequivalent to the momenta . They have a simple Poisson algebra

T [S], [A] = Int[S, A]T [S],
(37)
where Int[S, A] is the intersection number of the surfaces S and A.

We note that these observables commute with the action of local gauge transformations generated by (16).
Following the strategy just outlined, we seek a representation of (37) on a Hilbert space,
H kinematical that carries a nonanamalous representation of Diff().
Let be a network of two surfaces S , whose faces are labeled by integers. The integers
count elementary closed surfaces, out of which the network is formed. This implies that the
triangle inequalities are satisfied at every trivalent edge where surfaces meet.
States are functionals of , so we have
|
= ( ).
(38)
The operator representing T [S] is defined by

| T [S] = S|,
(39)
where S is the network with the surface S added. This gives us

T [S] [ ] = [ S].
The conjugate momentum operator [A] is defined by

Int[S, A] |.
| [A] = h
(40)
(41)
One can check explicitly that the commutator

T [S], [A] = h Int[S, A]T [S].
(42)
The kinematical inner product is

|
= .
(43)
179
This gives rise to a non-separable Hilbert space, as in LQG. The unitary representation of
Diff() is defined by
|U () = |.
(44)
This is easily shown to be unitary under (43).

We then define the diffeomorphism invariant Hilbert space, H diffeo to be those states in the
dual of H kin such that
|U () = |.
(45)
Following the standard method of LQG, these can be shown to have a countable, orthonormal basis, given by |{ }
, where { } are diffeomorphism classes8 of networks of labeled two-surfaces
embedded in .
We now want to introduce a regularized Hamiltonian constraint operator, H expressed in
terms of elements of the surface algebra, in the kinematical Hilbert space. This should have
several properties:
(1) On the classical counterpart, lim0 H = H.
(2) The limit lim0 H acts on H diffeo in that it takes diffeomorphism invariant states to diffeomorphism invariant states.
Here are some steps towards the construction of such a regularized operator. A regularization
procedure is going to break diffeomorphism invariance in . So let us introduce in a local region R, a flat metric qij0 in and a set of coordinates y i . At a point p we can have a box
jk axis. We have, to leading order
B (p) of volume 3 in q 0 alongside the i,
ijk
ij

T B (p) = 1 + 3 Fijk (p),
ij k
(46)
where T [B (p)] takes the intergral of around the surface of the box.
ij k
We can then write a regularized three form operator as

1
T B (p) 1 .
ij k
3
We can then write a regulated Hamiltonian constraint operator
=
(47)
H = K + V
(48)
ij k
with
j
V = i ji ,
(49)
where the regulated operator ji is

j
i
klm

n o j
.
= m
n o
i kl
(50)
8 To eliminate continuous labels on states coming from labeling diffeomorphism equivalence classes of complicated
intersections, the diffeomorphisms are extended to piecewise smooth diffeomorphisms, after which the basis is countable [1].
180

ij
Similarly, let us define a surface A (p) to be a four-dimensional hypercube of size on a side,

orthogonal to the ij directions, all with respect to the background metric qij0 , at the point p. We
then can define
i j =
1 ij
A (p) .
4
(51)
We then have for the regularized kinetic energy

K = ijk lm n i j k l m n .
(52)
There remains much to do, but the outline is clear from here, by analogy with the development
of LQG. For example, one can define a path integral by exponentiation. It will be defined as a
spin foam model, based on labeled triangulations of M. The three-simplices of the triangulation
will be labeled with integers, corresponding to the evolutions of the graphs. There will also be
labels on the four-simplices, corresponding to .
5. Down from 11 dimensions
We do not have a background independent formulation of M theory, so the existence of the
theory remains a conjecture. But part of that conjecture is that a classical limit of M theory is
given by 11-dimensional supergravity. Hence it is of interest to see if Hitchins 7-dimensional
theory might be derived from a suitable reduction of 11-dimensional supergravity. If it can be,
then it may be that we can identify the quantum states just described as the actual quantum
degrees of freedom corresponding to the membranes of M theory.
As is the case in all known versions of general relativity and supergravity, the action and field
equations for 11-dimensional supergravity can be written in a polynomial form [5]. This makes
it possible to take a consistent reduction in which the frame field, connection and gravitino fields
(with certain density weights) are taken to zero, leaving only the three form aABC .9 Since the
field equations are polynomial, these provide a subset of solutions to the full equations of 11d supergravity. It can be shown that the supersymmetry transformations, which are also polynomial,
are trivially satisfied for such solutions.
In this reduction, the action is

11
I =
(53)
da da a.
M11
This is a version of higher-dimensional ChernSimons theory. Its dynamics and quantization

were studied in detail in [5]. It is important to note that higher dimension and higher form Chern
Simons theories have local degrees of freedom. This theory is diffeomorphism invariant, but it is
not topological.
Let us see if the degrees of freedom of Hitchins 7-dimensional theory can be found imbedded
in this metric-less reduction of 11-dimensional supergravity.
The field equations of (53) are
da da = 0.
9 A, B, . . . = 0, 1, . . . , 10, while ten-dimensional spatial indices are given by I, J, . . . = 1, . . . , 10.
(54)
181
The gauge transformation is of course

a = d
(55)
with a two form.

A solution to (54) is given by the following ansatz. Let A = a, , with a = 0, . . . , d and
= d + 1, . . . , 10, Then the ansatz is
daABC = 0
(56)
so long as d 6. It is interesting that the largest non-trivial case is d = 6, which gives us a

reduction to a seven-dimensional theory.
In fact, we can find a simple set of solutions, that are locally but not globally flat. Let the
topology be chosen to be the standard one proposed for a reduction from M theory to string
theory,
M11 = R S 1 R 3 .
(57)
Here is a compact six manifold, and the R is time, as in previous sections. These are coordinatized as before by x a , a = 0, i, with i = 1, . . . , 6. Let y 7 = be the coordinate around
the S 1 . This is the standard circle around which membranes are wrapped to get strings. The three
remaining dimensions in the R 3 can as usual be taken to be ordinary, uncompactified space, coordinatized by y with = 1, 2, 3 from now on. We will assume that everything is constant in
space, so that
aABC
= 0.
y
(58)
This, physically, means that we are studying the geometry of string compactifications that might
arise from M theory.
Let us take a solution which is locally pure gauge, of the form of (55), with (locally on the S 1 )
ab = ab (x).
However globally, we will have

ab (x) = d aab = ab (x).
(59)
(60)
S1
Since the solution is locally trivial, da = 0 everywhere, so this is a solution to 11d supergravity.
From (55) we see that there is still a gauge invariance, given by
= = d,
where is a one form.
The integral around three-cycles CI of are given by

a = d.
CI
CI
These are constants, as they do not evolve in time under the equations of motion (54).
(61)
(62)
182
The canonical momenta for aI J K is I J K (a da) , where the duality is in the tendimensional manifold S 1 R 3 . We have

aI J K (x), LMN (y) = 10 (x, y)ILMN
(63)
JK .
The dimensionally reduced momenta is

= ,
(64)
R3
which is a four-form on R. It can be pulled back to a four form on . We have

i i
= 6 x i , y i ij klmn .
ij x , klmn y = d d 3 x aij , klmn
S1
(65)
R3
Thus, the canonical degrees of freedom of topological M theory can be seen to arise from the
reduction of supergravity from 11 dimensions.
The reduced theory then has degrees of freedom (ab , cd ), with fixed cohomology on R .
In the quantum theory there will arise an effective action to describe the low energy dynamics
of these degrees of freedom. The effective action will be dominated by the lowest dimension
term that can be made from d on R. One can conjecture that this will be given by the
Hitchins action, which is a cosmological constant term, and hence should dominate the low
energy limit.
It is possible we can proceed further in this direction. Let g be a flat metric on the R 4 parameterized by x 0 and y and let e0 , e be four one form orthonormal frame fields. Given the
imbedding of R 4 into M11 we can pull these back to a degenerate set of 11-dimensional frame
fields. We may conjecture that these, together with any aABC such that locally da = 0, give solutions of 11-dimensional supergravity. If this is true, then there is a sector of M theory with a
conventional geometry on the four uncompactified spacetime dimensions, but where the geometry on the compactified dimensions is entirely based on a forms theory.
There should be much more in this sector. We should be able to add other degrees of freedom
coming from the fields of 11-dimensional supergravity to systematically expand Hitchins theory
to a reduction of M theory, with a full set of local degrees of freedom.
6. Conclusions
What is described here is a first step towards a background independent quantization of topological M theory. Many issues remain open. While the conjecture that the constrained system
here is equivalent to Hitchins seven-dimensional theory is plausible, it still needs to be proved.
The results on the quantum theory are just a first sketch, along the lines of early papers on LQG.
It is likely that the quantization can be made rigorous, along the lines of [16]. Of great interest is
whether there is an extension of the LOST uniqueness theorem [17] to this context. Further exploration of this direction can be expected to shed light both on the key question of what M theory
may be as well as on the interpretation of the results in LQG concerning (3 + 1)-dimensional
physics. The idea proposed in [12] that Hitchins theory may open the way to a unification of
string theory and LQG is intriguing and the results obtained here give us a common language
within which the precise relationship between the two approaches can be elucidated.
183
But we can already draw a few interesting conclusions from the results obtained here. First
we see a possible non-perturbative origin for D-brane states in a background independent formulation. Second, as pointed out in [12], there are implications for string compactifications. In the
standard string compactifications on CalabiYau manifolds, the and ij are fixed. However,
we see that in topological M theory these are conjugate variables. Moreover, we see that these
variables can be understood to descend from full M theory, where they are still conjugate variables. If so, then there can be no quantum states of M theory corresponding to fixed CalabiYau
geometries on . Thus, any phenomenology that depends on the fixed background structure of
a CalabiYau manifold can only be meaningful in the semiclassical limit in which conjugate
variables can both have definite values.
Finally, it is interesting to note that the real variables on which Hitchins theory is based
only define a complex manifold when the condition (32) is satisfied. We also have imposed
the positivity condition (12). These are analogous to the condition that the determinant of the
spatial metric be positive. It means that the part of the configuration space that corresponds to
complex geometries is not a vector space, but satisfies a non-linear inequality. There is then
the issue of how this inequality is to be satisfied in the quantum theory. Just as the metric may
have an amplitude to be non-degenerate in any first order formulation of quantum gravity, so we
must consider the possibility that a quantum state can give a non-zero amplitude to a region of
configuration space where (32) is violated, leading to quantum fluctuations in which fails to
have a complex structure.
Acknowledgements
I am grateful to Robert Dijkgraaf, Sergei Gukov, Nigel Hitchin, Andrew Neitzke and Cumrun
Vafa for taking the time to explain to me the basics of topological string and M theory, and especially to Andrew Neitzke for comments on the manuscript. I am also very grateful to Christian
Romelsberger and Jaume Gomis for very helpful comments and especially to Eli Hawkins for
many perceptive and useful suggestions. Finally, thanks to Stephon Alexander for suggestions
which improved the manuscript.
References
[1] A. Ashtekar, New Perspectives in Canonical Gravity, Bibliopolis, Naples, 1988;
A. Ashtekar, in: Lectures on Non-Perturbative Canonical Gravity, in: Advanced Series in Astrophysics and Cosmology, vol. 6, World Scientific, Singapore, 1991;
R. Gambini, J. Pullin, Loops, Knots, Gauge Theories and Quantum Gravity, Cambridge Univ. Press, 1996;
C. Rovelli, Quantum Gravity, Cambridge Univ. Press, 2004.
[2] L. Smolin, An invitation to loop quantum gravity, hep-th/0408048.
[3] L. Bombelli, J.H. Lee, D. Meyer, R. Sorkin, Spacetime as a causal set, Phys. Rev. Lett. 59 (1987) 521;
X. Martin, D. OConnor, D.P. Rideout, R.D. Sorkin, On the renormalization transformations induced by cycles of
expansion and contraction in causal set cosmology, Phys. Rev. D 63 (2001) 084026, gr-qc/0009063;
D.P. Rideout, R.D. Sorkin, Evidence for a continuum limit in causal set dynamics, Phys. Rev. D 63 (2001) 104011,
gr-qc/0003117;
D.P. Rideout, Dynamics of causal sets, gr-qc/0212064;
D.P. Rideout, R.D. Sorkin, A classical sequential growth dynamics for causal sets, gr-qc/9904062;
D.P. Rideout, R.D. Sorkin, Phys. Rev. D 61 (2000) 024002.
[4] J. Ambjorn, A. Dasgupta, J. Jurkiewiczcy, R. Loll, A Lorentzian cure for Euclidean troubles, hep-th/0201104;
J. Ambjorn, R. Loll, Nucl. Phys. B 536 (1998) 407, hep-th/9805108;
J. Ambjorn, J. Jurkiewicz, R. Loll, Phys. Rev. Lett. 85 (2000) 924, hep-th/0002050;
184
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
J. Ambjorn, J. Jurkiewicz, R. Loll, Nucl. Phys. B 610 (2001) 347, hep-th/0105267;

R. Loll, Nucl. Phys. B (Proc. Suppl.) 94 (2001) 96, hep-th/0011194;
J. Ambjorn, J. Jurkiewicz, R. Loll, Phys. Rev. D 64 (2001) 044011, hep-th/0011276;
J. Ambjorn, J. Jurkiewicz, R. Loll, G. Vernizzi, Lorentzian 3d gravity with wormholes via matrix models, JHEP 0109
(2001) 022, hep-th/0106082;
B. Dittrich, R. Loll, A hexagon model for 3D Lorentzian quantum cosmology, hep-th/0204210;
J. Ambjorn, J. Jurkiewiczcy, R. Loll, Emergence of a 4D world from causal quantum gravity, hep-th/0404156.
Y. Ling, L. Smolin, Eleven-dimensional supergravity as a constrained topological field theory, Nucl. Phys. B 601
(2001) 191, hep-th/0003285;
L. Smolin, ChernSimons theory in 11 dimensions as a non-perturbative phase of M theory, hep-th/9703174.
L. Smolin, Strings as perturbations of evolving spin-networks, Nucl. Phys. B (Proc. Suppl.) 88 (2000) 103113,
hep-th/9801022;
L. Smolin, M theory as a matrix extension of ChernSimons theory, Nucl. Phys. B 591 (2000) 227242, hepth/0002009;
L. Smolin, The cubic matrix model and a duality between strings and loops, hep-th/0006137;
L. Smolin, The exceptional Jordan algebra and the matrix string, hep-th/0104050;
T. Azuma, M. Bagnoud, Curved-space classical solutions of a massive supermatrix model, hep-th/0209057;
M. Bagnoud, L. Carlevaro, A. Bilal, Supermatrix models for M-theory based on osp(1|32, R), Nucl. Phys. B 641
(2002) 6192, hep-th/0201183;
T. Azuma, S. Iso, H. Kawai, Y. Ohwashi, Supermatrix models, Nucl. Phys. B 610 (2001) 251279, hep-th/0102168;
T. Azuma, Investigation of matrix theory via super-Lie algebra, hep-th/0103003.
M. Plebanski, On the separation of Einsteinian substructures, J. Math. Phys. 18 (1977) 2511;
H. Urbantke, J. Math. Phys. 25 (1983) 2321;
R. Capovilla, J. Dell, T. Jacobson, Phys. Rev. Lett. 21 (1989) 2325;
R. Capovilla, J. Dell, T. Jacobson, Class. Quantum Grav. 8 (1991) 59;
R. Capovilla, J. Dell, T. Jacobson, L. Mason, Class. Quantum Grav. 8 (1991) 41.
T. Sano, J. Shiraishi, The nonperturbative canonical quantization of the N = 1 supergravity, Nucl. Phys. B 410
(1993) 423, hep-th/9211104;
T. Sano, J. Shiraishi, The Ashtekar formalism and WKB wave functions of N = 1, 2 supergravities, hep-th/9211103;
T. Kadoyoshi, S. Nojiri, N = 3 and N = 4 two form supergravities, Mod. Phys. Lett. A 12 (1997) 11651174,
hep-th/9703149;
K. Ezawa, Ashtekars formulation for N = 1, N = 2 supergravities as constrained BF theories, Prog. Theor. Phys. 95
(1996) 863882, hep-th/9511047.
L. Freidel, K. Krasnov, R. Puzio, BF description of higher-dimensional gravity theories, Adv. Theor. Math. Phys. 3
(1999) 12891324, hep-th/9901069.
N. Hitchin, The geometry of three-forms in six and seven dimensions, math.DG/0010054.
N. Hitchin, Stable forms and special metrics, math.DG/0107101.
R. Dijkgraaf, S. Gukov, A. Neitzke, C. Vafa, Topological M-theory as unification of form theories of gravity, hepth/0411073.
V. Pestun, E. Witten, The Hitchin functionals and the topological B-model at one loop, hep-th/0503083;
N. Nekrasov, Z theory, hep-th/0412021;
P.A. Grassi, P. Vanhove, Topological M theory from pure spinor formalism, hep-th/0411167;
A.A. Gerasimov, S.L. Shatashvili, Towards integrability of topological strings I: Three-forms on CalabiYau manifolds, hep-th/0409238.
E. Hawkins, private communication.
T. Hertog, G.T. Horowitz, K. Maeda, Negative energy density in CalabiYau compactifications, JHEP 0305 (2003)
060, hep-th/0304199.
T. Thiemann, Introduction to modern canonical quantum general relativity, gr-qc/0110034, Living Rev., in press;
A. Ashtekar, J. Lewandowski, D. Marlof, J. Mourau, T. Thiemann, Quantization of dieomorphism invariant theories
of connections with local degrees of freedom, J. Math. Phys. 36 (1995) 519, gr-qc/9504018.
H. Sahlmann, Some comments on the representation theory of the algebra underlying loop quantum gravity, grqc/0207111;
H. Sahlmann, When do measures on the space of connections support the triad operators of loop quantum gravity?,
gr-qc/0207112;
H. Sahlmann, T. Thiemann, On the superselection theory of the Weyl algebra for diffeomorphism invariant quantum
gauge theories, gr-qc/0302090;
185
H. Sahlmann, T. Thiemann, Irreducibility of the AshtekarIshamLewandowski representation, gr-qc/0303074;

A. Okolow, J. Lewandowski, Diffeomorphism covariant representations of the holonomy flux star-algebra, Class.
Quantum Grav. 20 (2003) 35433568, gr-qc/0302059;
J. Lewandowski, A. Okolow, H. Sahlmann, T. Thiemann, Kinetical quantum algebra for theories of connections and
uniqueness of its representations in the diffeomorphism invariant context, in preparation;
C. Fleischhack, Representations of the Weyl algebra in quantum geometry, math-ph/0407006.
A model of electroweak symmetry breaking from

a fifth dimension
Giuliano Panico a , Marco Serone a, , Andrea Wulzer a,b
a ISAS-SISSA and INFN, Via Beirut 2-4, I-34013 Trieste, Italy
b IFAE, Universitat Autnoma de Barcelona, 08193 Bellaterra, Barcelona
Received 11 November 2005; accepted 16 January 2006

Abstract
We reconsider the idea of identifying the Higgs field as the internal component of a gauge field in the flat
space R 4 S 1 /Z2 , by relaxing the constraint of having unbroken SO(4, 1) Lorentz symmetry in the bulk.
In this way, we show that the main common problems of previous models of this sort, namely the prediction
of a too light Higgs and top mass, as well as of a too low compactification scale, are all solved. We mainly
focus our attention on a previously constructed model. We show how, with few minor modifications and
by relaxing the requirement of SO(4, 1) symmetry, a potentially realistic model can be obtained with a
moderate tuning in the parameter space of the theory. In this model, the Higgs potential is stabilized and the
hierarchy of fermion masses explained.
1. Introduction
The Electroweak Symmetry Breaking (EWSB) mechanism and the hierarchy of fermion
masses are among the most obscure aspects of the Standard Model (SM). The minimal set-up
of a single doublet scalar field (the SM Higgs field) which drives the EWSB is affected by a
stability problem at the quantum level, since the Higgs mass term is quadratically sensitive to
the scale of new physics. In the SM, moreover, the observed fermion masses are obtained by an
unnatural choice of Yukawa couplings. Even leaving aside the three neutrinos, their values range
from 105 106 for the electron up to 1 for the top quark.
E-mail address: serone@sissa.it (M. Serone).

doi:10.1016/j.nuclphysb.2006.01.025
G. Panico et al. / Nuclear Physics B 739 (2006) 186207
187
Looking for alternative theories in which these problems are solved has been one of the main
guidelines for new ideas and models beyond the SM. Supersymmetry (SUSY) is certainly the
most interesting and well motivated possibility. It predicts the unification of gauge couplings,
it naturally incorporates a good candidate to explain the dark matter abundance in the universe
and, if assumed to be broken at energy scales few TeV, it can also give rise to a natural
EWSB. Last, but not least, it is a weakly coupled theory. The simplest model of this sort is
the Minimal Supersymmetric Standard Model (MSSM). Despite the above important positive
aspects, superparticles have not been discovered yet and, in fact, most of the parameter space
of the MSSM is already experimentally ruled out, resulting in an unwanted fine-tuning in the
model. Moreover, the MSSM does not provide a sensible explanation for the hierarchy of SM
fermion masses. This motivates the quest for other ideas and models, also alternative to SUSY,
which can explain the stability of the EWSB scale and the hierarchy of fermion masses.
Models where the Higgs field is identified with the internal component of a gauge field in
TeV-sized extra dimensions [1] (also known as models with gauge-Higgs unification) are an
example of this sort [2,3] (see [4] for earlier references and [5] for a brief overview). The higherdimensional gauge symmetry, rather than SUSY, provides the stabilization of the Higgs mass
term. Consequently, the quadratic divergencies in the Higgs mass due to the SM particles are
cancelled by states with the same statistic, and not opposite as in SUSY. This is analogous to
what typically happens in little Higgs models [6], which indeed arose from the deconstructed
version of gauge-Higgs unification models [7]. The five-dimensional (5D) case, with one extra
dimension, is the simplest one and also the one which seems phenomenologically more appealing. It is by now clear how to embed the SM fermions and to break the flavour symmetry in such
framework, despite the fact that the Yukawa couplings are gauge couplings: one can either put the
SM fermions on the boundaries and couple them to massive bulk fermions [8,9] or one can identify the SM fields as the (chiral) zero modes of bulk fermions with jumping mass terms [3,10]. In
both cases one ends up with a concrete realization of the idea of getting small Yukawa couplings
by means of exponentially small overlaps of wave functions in the internal space [11]. Models
defined in flat space seem to have common drawbacks. Namely, one obtains too low Higgs, top
and compactification masses. To solve these problems, one has to find a gauge-invariant way to
increase the (gauge) couplings of the Higgs with the bulk fermions. Two known possibilities are
the introduction of large localized gauge kinetic terms [8] and warped compactification [12]. In
both cases, however, the bulk wave functions are distorted in a non-trivial way, resulting in potentially too large deviations from the SM coming from the Electroweak Precision Tests (EWPT)
and the universality of gauge couplings. Implementing a custodial symmetry improves the situation, but some fine tuning is still necessary to get viable models. An interesting proposal along
this direction has been provided in [13].
In this paper we propose a different approach to get a potentially realistic model with gaugeHiggs unification in flat space. The essential ingredient which we advocate is an explicit tree-level
breaking of the Lorentz SO(4, 1) symmetry. More precisely, we notice that another possible way
to increase the couplings between the Higgs field and the fermions in a 5D gauge-invariant way
is achieved by breaking the SO(4, 1)/SO(3, 1) symmetry (so that the usual SO(3, 1) Lorentz
symmetry is unbroken), which is the one that obliges us to couple the fermions in the same way
to the gauge bosons and to the Higgs field. In light of this symmetry breaking, we reconsider
the minimal 5D model constructed in [8], to which we also add a new antiperiodic bulk fermion.
The latter state plays an important role to get a substantial hierarchy between the SM scale and
the scale of new physics. As we will see, such a proposal allows to stabilize the electroweak
scale, explain the hierarchy of fermion masses, get the correct top mass and high enough Higgs
188
mass and compactification scale, then resulting in a potentially very interesting model.1 As in
other models of gauge-Higgs unification, the EWSB is radiatively induced. The Higgs mass
is completely finite at one-loop level. At higher-loops, mainly due to the Lorentz symmetry
breaking, divergencies could be reintroduced, but they would not spoil the stability of the Higgs
potential. The Higgs mass can range from 125 GeV up to 600 GeV (see Fig. 5) depending on
the particular set-up of the model. The lightest non-standard particle is a colored fermion with
mass M 12 TeV. Interestingly enough, the neutral component of the lightest KaluzaKlein
(KK) state of the bulk antiperiodic fermion is a stable weakly interacting particle with a mass of
a few TeV, which is a potential Dark Matter (DM) candidate, along the lines of [15].
The EWPT, the observed suppression of Flavour Changing Neutral Currents (FCNC) and the
universality of the gauge couplings are typically the most severe tests that any theory which
claims to be a realistic extension of the SM should pass. We do not perform a detailed analysis
of these effects, which are left for future work, but we quantitatively show that our model can
pass all these tests. We argue that FCNC effects should be acceptable, since we estimate higher
derivative operators mediating FCNC to be governed by dimensionless couplings which are naturally small, particularly for the first two generations. As in several models based on warped extra
dimensions (see e.g. [13,16,17]), a dangerous and worrisome effect is a deviation from the SM
ZbL bL coupling. The resulting bound in our case is about the same one would get from corrections to the four fermion operators, as computed for theories similar to ours (Higgs and the gauge
fields in the bulk, SM fermions on the brane) [18]. Another important deviation is due to a custodial breaking mixing between the gauge bosons, which leads to a non-vanishing tree-level .
The resulting bound is, once again, approximately the same as those coming from Zbb and four
fermion operators. Our theory is compatible with the above phenomenological constraints for a
compactification scale 1/R 5 TeV. Such values for 1/R can be obtained, but at the price of
some fine-tuning in the parameter space of the model, which is hard to quantify in a meaningful
way, depending on the prescription used.
From a more theoretical point of view, we show that the SO(4, 1) Lorentz symmetry breaking
which we advocate can have a natural origin as a spontaneous breaking induced by a Scherk
Schwarz [19] twist on a shift symmetry, which can also be seen as a constant flux for a four-form
field strength. This interpretation indicates that the Lorentz violation we consider can have a
natural origin in a 5D framework.
The structure of the paper is as follows. In Section 2 we present our model which, as mentioned, is mainly based on the model of [8]. In Section 3 we show the predictions of the model,
mostly based on numerical results. In particular, we focus on the values of the Higgs and top
masses, and of the compactification scale 1/R, since the too low values of these quantities were
the main obstructions in constructing a realistic model of this sort in 5D. In Section 4 we roughly
FCNC and the EWPT. We also point out
quantify the bounds imposed on our model by Zbb,
the difficulty in giving a solid estimate of the amount of the fine-tuning necessary in our model
to pass all these tests. In Section 5 we discuss the possible microscopic origin of the SO(4, 1)
Lorentz symmetry breaking as a spontaneous breaking induced by a ScherkSchwarz twist. Finally, in Section 6 we report our conclusions.
1 Another interesting model with gauge-Higgs unification in flat space is provided in [14], where it has been shown
that other variants of the model of [8] can also be made realistic.
189
2. The model
The model we consider is mainly based on the one built in [8], namely a 5D gauge
theory

on the S 1 /Z
2 orbifold, with group G = SU(3)c SU(3)w U (1) . We denote by g5 g 2R
and g5 g 2R the SU(3)w and U (1) gauge couplings respectively. The extra U (1) and its
coupling g have to be introduced in order to get the correct weak mixing angle. The Z2 orbifold
projection is embedded non-trivially in the electroweak SU(3)w group only, by means of the
matrix

1 0 0
2i 3t8
P =e
(2.1)
= 0 1 0 ,
0
0 1
where ta are the standard SU(3) generators, normalized as Tr ta tb = 1/2ab . The twist (2.1)
breaks the electroweak gauge group to SU(2) U (1) U (1) in 4D. The massless 4D fields are
the gauge bosons in the adjoint of SU(2) U (1) SU(3), the gauge field A and a charged
scalar doublet H , the Higgs field, arising from Aa5 . The hypercharge generator Y , such that
Y = 1/2 for the Higgs field, is taken to be the linear combination Y = 1 t8 + t of the U (1)
3
and U (1) generators. The gauge field AY associated to the hypercharge and its orthonormal
combination AX are
3gA8 g A
g A8 + 3gA
AY =
(2.2)
,
AX =
.
3g 2 + g 2
3g 2 + g 2

The
Y coupling gY is related to the 4D SU(2) and U (1) couplings g and g as gY =
U (1)

2

2
3gg / 3g + g . By suitably choosing g we can adjust the weak angle to the correct value,
according to the relation
sin2 W =
gY 2
3
=
.
2
2
g + gY
4 + 3g 2 /g 2
(2.3)
As we will better explain in Section 2.2, the zero mode of AX will get a large mass and decouple
from the theory, leaving only its KK excitations. A vacuum expectation value (VEV) for Aa5
induces an additional spontaneous symmetry breaking to U (1)EM . We can take

2 a7
Aa5
,
g5 R
(2.4)
corresponding to an imaginary VEV for the Higgs field: H = 2i/(gR). The parameter
[0, 1/2] in Eq. (2.4) is a Wilson line phase, and thus the EWSB in this model is equivalent to a
Wilson line symmetry breaking [20].
2.1. The matter Lagrangian
Introducing matter fields in this set-up is a non-trivial task. One possibility is to include massive 5D bulk fermions and massless localized chiral fermions, with a mixing between them, so
that the matter fields are identified with the lowest KK mass eigenstates. In this way, Yukawa couplings are exponentially sensitive to the bulk mass terms, and the observed hierarchy of fermion
190
masses is naturally explained.2 Here we focus on the third generation of quarks (top and bottom),
since light quarks and leptons do not significantly contribute to the Higgs potential.3 The bulk 5D
fermions that have the correct quantum numbers to couple with the (conjugate) top and with the
bottom are respectively the symmetric (6) and fundamental (3) representations of SU(3)w , neutral under the U (1) group. In addition to that, we also add a symmetric representation, antiperiodic on the covering circle S 1 , with U (1) charge 2/3. We impose a global U (1)A symmetry under which only the antiperiodic bulk fermions transform. This symmetry forbids any mixing between these fields and the localized ones.4 Such state was not present in the original model of [8].
The matter fermion content of this basic construction is more precisely the following. We
introduce a couple of periodic bulk fermions (t , t ) with opposite Z2 parities, in the repre 6) and (b , b ) in the (3, 3) of SU(3)c SU(3)w and a couple of antiperiodic
sentation (3,
bulk fermions A and A with opposite Z2 parities, in the (1, 6). All these fermions have unconventional SO(4, 1) Lorentz violating kinetic terms. At the orbifold fixed points, we have
a left-handed doublet QL = (tL , bL )T and two right-handed fermion singlets tR and bR of
SU(2) U (1). They are located at y1 and y2 , equal to 0 or R, the two boundaries of the
segment. The parity assignments for the bulk fermions allow for a bulk mass term M mixing
and , as well as boundary couplings e1,2 with mass dimension 1/2 mixing the bulk fermions
to the boundary fields QL , tR and bR . The matter Lagrangian reads5
Lmat =

j iD
/ 4 kj D5 5 j + j iD
/ 4 kj D5 5 j + ( j Mj j + h.c.)
j =t,b,A

/ 4 QL + e1b Q L b + e1t Q cR t + h.c.
+ (y y1 ) Q L iD

/ 4 tR + bR iD
/ 4 bR + e2b bR b + e2t tLc t + h.c. ,
+ (y y2 ) tR iD
(2.5)
where t,b and t,b are the doublet and singlet SU(2) components of the bulk fermions t,b .
For simplicity, in the following we take kj = kj . The metric is mostly minus and ( 5 )2 = 1.
All bulk fermion modes are massive
and, neglecting the bulk-to-boundary couplings, their mass
spectrum is given by Mn,j = m2n,j + Mj2 , where mn,j = kj n/R. When the EWSB induced
by (2.4) is considered, a new basis has to be defined for the bulk fermion modes in which they
have diagonal mass terms, with a shift in the KK masses mn,j mn,j (). The procedure is
outlined in the appendix of [8].
In the following, it will be convenient to take the size R of the orbifold as reference length
scale and use it to define dimensionless
quantities. In particular, it will be useful to introduce the
parameters i = RMi and ia = R/2eia .

2 Note that bulk-brane systems of fermions of this kind could naturally originate from a single bulk field on a resolved
orbifold, along the lines of [21]. The bulk-brane spectra considered here are however not compatible with those found
in [21].
3 See however [14] for the study of a different set-up, in which other quarks and leptons can significantly contribute to
the Higgs potential.
4 This U (1) symmetry, as well as the U (1) charge we have chosen for the antiperiodic fermions, are introduced
A
uniquely to possibly get a viable DM candidate out of these states.
5 The flavour structure of the full model, including all quarks and leptons, is obtained exactly as in [8], with the only
difference that now one could introduce an SO(4, 1) Lorentz violating matrix kij , which provides an additional source of
flavour mixing. An interesting alternative would be to introduce a flavour symmetry in the model, along the lines of [22].
191
2.2. Gauge bosons and anomaly cancellation

The localized chiral fermions in our model induce gauge and gravitational anomalies that must
be cancelled. As already discussed in [8], the precise pattern of anomaly cancellation depends
on the position of the localized fermions. For all distributions of matter, all anomalies can be
cancelled, but at the price of introducing two localized axions (at y = 0 and at y = R) and
a ChernSimons term with a jumping coefficient [23], which has to be introduced anytime the
SM anomalies (the ones which do not involve the field AX ) do not locally cancel in the internal
direction. When a ChernSimons term is needed, couplings which are Z2 odd cannot be anymore
consistently neglected. For this reason, for simplicity, we focus in the following on two special
set-ups, that we shortly denote = 0 and = 1, in which all the SM anomalies are locally
cancelled. We call = 0 the set-up in which all SM fermions are located at the same fixed-point
(say, at y = 0). Among all the various set-ups in which the matter is located in both fixed-points,
but in such a way that the SM anomalies cancel locally, we call = 1 the ones in which the
(tL , bL ) doublet is, say, at y = 0, whereas tR and bR are at y = R, without specifying in detail
the location of the other SM fermions.
The anomalies which are left are those involving AX and can be cancelled by means of
a 4D version of the GreenSchwarz mechanism (GS) [24]. One introduces one ( = 0) or two
( = 1) localized axions, transforming non-homogeneously under the U (1)X symmetry, with
non-invariant 4D WessZumino couplings that compensate for the one-loop anomaly. In this
way all mixed SU(3)c SU(2)L U (1)Y U (1)X gauge and gravitational anomalies can be
cancelled. When = 0, the single axion is eaten by the gauge field AX , whereas for = 1 one
combination of them is eaten, while the orthogonal one remains massless. In a suitable gauge,
the net effect of the anomaly on the gauge bosons is the appearance of localized quadratic terms
in AX , with a mass term MX whose natural size is the cut-off scale of the model. For MX 1/R,
the localized mass terms simply result in an effective change (from Neumann to Dirichlet) of the
boundary condition of AX at the points where they are located. In the limit MX and for
R y R, the KK expansion of AX is given by
1
[n + (1 + )/2]|y|
AX (x, y) =
.
AX, n (x) sin
R
R n=0
(2.6)
The zero-mode of AX decouples while its KK tower is still at low energy and can have sizable
effects.6 The remaining gauge bosons are insensitive to the localized mass terms and retain their
original expansion in cosines or sines as given by Eq. (2.1).
When the EWSB occurs, the diagonal mass eigenstates, taking into account of the localized
mass terms for AX , are the following:
n+
n + h ()
,
m(2)
,
n =
R
R
n
n+1
n + (1 + )/2
m(3)
,
m(4)
,
m(5)
,
n =
n =
n =
R
R
R
m(1)
n =
n [, ],
n [0, ].
(1)
(2)
(3)
The SM gauge bosons W , Z and are associated to the n = 0 modes of mn , mn and mn , so

that the W mass equals
mW = .
(2.7)
R
6 Ref. [8] overlooked these effects, considering only the mixing of the SM Z boson with the zero mode of A .
X
192
The functions h () appearing in Eq. (2.7) originate from the localized mass terms of AX . In
the limit MX , mZ = h () is defined by the transcendental mass equations
1
sin2 (2),
4 cos2 W
1
1
sin2 ()
sin4 (),
sin2 (mZ R) =
cos2 W
4 cos4 W
sin2 (mZ R) =
= 0,
= 1.
(2.8)
By expanding the sines in Eq. (2.8) one finds at leading order the SM relation mZ = mW / cos W ,
as expected. Corrections due to the localized mass terms for AX are however present, so that
= 1. We will better quantify such corrections in Section 4.2.
By 5D gauge symmetry, the Higgs mass vanishes at tree-level and is radiatively induced. It
equals

m2H (min ) =
g4 R
2
2

2 V
,
2 =min
(2.9)
with V () the (radiatively induced) Higgs effective potential and min its minimum.
2.3. Higgs potential and induced fermion masses
The 5D SU(3)w gauge symmetry, which is not broken by the Lorentz violating couplings kj ,
forbids the appearance of any local Higgs potential in the bulk. A Higgs potential localized at the
orbifold fixed points is also forbidden by a non-linearly realized symmetry which is left unbroken
by the orbifold boundary conditions. This symmetry acts on the Higgs field components Aa5
(a = 4, 5, 6, 7) as [25]
Aa5 Aa5 + 5 a .
(2.10)
The symmetry (2.10) is not broken in our model and hence we expect that the Higgs potential is
still radiatively induced by non-local operators and thus finite. Since the field A5 couples only
to the gauge fields and to the bulk fermions, its potential depends indirectly on the boundary
couplings through diagrams in which the virtual bulk fermions temporarily switch to a virtual
boundary fermion.
The one loop contribution to the potential given by the 5D gauge bosons and ghosts is easily
computed from the explicit form of the KK mass spectrum (2.7). It is given by7

Vg () = 2VA () + VA h () ,
(2.11)
where
VA () =

1
9
cos(2k).
64 6 R 4
k5
(2.12)
k=1
7 Eq. (2.11) is obtained by replacing 2 h () in Eq. (29) of [8], that overlooked the corrections due to the U (1)
X
anomaly.
193
The one loop contribution from a massive 5D fermion with mass and given k is also easily
found.8 For a pair of modes with charge q, one has

4 2 2 2n
3k 4 5
k cos 2nq +
1
+
2n
e
,
n
V (q) =
(2.14)
+
n
k 3 k2
2q
8 6 R 4
n=1
where = 0 for periodic fermions and = 1 for antiperiodic fermions.

The full Higgs effective potential is obtained by summing the gauge and fermion contributions, including also the contributions of the fermion boundary terms. The explicit formulae for
the latter ones can be derived exactly as in [8], modulo the changes due to the SO(4, 1) breaking
parameters, and are given by (see [8] for the notation)9
2
b
t

1b2
2t2
x
x
1
3
Vt () =
dx x ln
Re 1 + i1
f0
, 0 + i2
f0
,0
b
t
6
4
kb x
kb
2kt x
kt
4 R
i=1

t2

t
2
i
it2
xt
x
f0
, 2 +
Im
, 2
,
f
+
(2.15)
2 i2 kt x t
kt
2 i2 kt x
kt
i=1
2
b
t

ib2
1t2
x
x
1
3
Vb () =
dx
x
ln
Re
1
+
f
,
f
,
0
i1
0
kb x b
kb
kt x t
kt
4 6 R 4
0
2

i=1
i=1
b
ib2
x
f
Im
,
.
kb x
kb

(2.16)
We find that the presence of antiperiodic fermions is necessary to obtain small enough values
of min . Indeed, as it can be seen from Eq. (2.14), they permit a partial cancellation of the leading
cosine in the fermion contribution to the potential to be enforced, then lowering the position of
its global minimum [26] (see also [27]). Note that for this cancellation to take place a certain
correlation among the parameters, mainly between kt and kA , is required. We better quantify it
in Section 3. The Higgs mass, however, is generically too low in this set-up for ki = 1. Higher
values of ki considerably help in getting higher Higgs masses. This is particularly clear in the
rough approximation in which one neglects the boundary contributions (2.15) and (2.16) (as well
as the gauge contribution) to the Higgs potential, and takes kb = kt = kA = k and massless 5D
bulk fermions: i = 0. In this case, the total Higgs potential is given by the sum of the bulk contributions of the form (2.14) (with = 0). This is exactly of the same form as the usual SO(4, 1)
invariant case, except for an overall k 4 factor in front of the potential. According to Eq. (2.9), the
8 As far as the contribution of a single fermion is concerned, the factor k can be eliminated by a redefinition of the
y coordinate, which results in the following rescaling of the parameters:
R R/k,
/k,
(2.13)
i i /k,
and a rescaling / k. This procedure can be used to derive the bulk fermion contributions in Eq. (2.14). However,
the boundary contributions in Eqs. (2.15) and (2.16), in which two fields with different ks are involved, cannot be
obtained by such simple scaling argument.
9 In Eq. (25) of [8] there is a typo in the last term: f (x u , ) should be replaced by f (x u , 2).
194
Fig. 1. The Higgs potential in the = 0 set-up, obtained with input parameters t = 0.99, b = 6.9, A = 0.24, kt = 2.42,
kb = 2.26, kA = 3.14, 1t = 1.9, 2t = 1.6, 1b = 2.9, 2b = 3.4.
Higgs mass is k 2 times the Higgs mass evaluated in the standard case with k = 1. As we will see
below, the factors ki s are also crucial to get reasonable top masses.
In Fig. 1, as an illustrative example, the effective potential is shown for a suitable choice of
the free microscopic parameters, in the set-up with = 0. The minimum is at min = 9 103 ,
corresponding to a compactification mass R 1 = 8.9 TeV. The value of the top and bottom quark
masses are mt = 176 GeV and mb = 1 GeV. The Higgs mass is 370 GeV.
When the bulk-to-boundary couplings are included, the exact spectrum of the bulk-boundary
fermion system defined by the Lagrangian (2.5) is determined by solving a complicated transcendental equation (whose form can however be deduced from Eqs. (2.15) and (2.16)). The lightest
states are identified with the top and bottom quarks and are in general a mixture of localized and
bulk fermion states. When the physical mass induced for the boundary fields is much smaller than
the masses of the bulk fields, to a very good approximation the top and bottom quark Yukawa
couplings (and hence their masses) are found by integrating out the massive bulk fermions and
neglecting the momentum dependence induced by higher derivative operators. The relations giving the top and bottom masses are given by appropriate modifications of Eqs. (15)(18) of [8].
One has

ma
ma = a0 a , a = t, b,
(2.17)
Z Z
1
where

t
Im f
=
, 2 ,
kt
2kt R
b
bb
, ,
mb0 = 1 2 Im f
kb R
kb
b
t
t

it2
1b2
2t2
t
Re f0
, 0 + i2
Re f0
,0 +
Re f0
, 2 ,
Zi = 1 + i1
kb b
kb
2kt t
kt
2 i2 kt t
kt

b2
t2
b
t
, + i1 1 t Re f0
, .
Zib = 1 + i b Re f0
(2.18)
kb
kb
kt
kt
The changes induced by the ki s are better seen in the limit in which one takes large bulk-toboundary mixing 1t , 2t 1 and min 1. For simplicity, we also take 1b = 2b = 0, since we
are mainly interested on the top mass formula. In these approximations one finds

mt 2kt mW F (2 )t /kt ,
(2.19)
mt0
1t 2t
195
where
x
(2.20)
.
sinh x
The function F (x) has a maximum for x = 0, where F (0) = 1, and is monotonically decreasing
for x 0. Thus
mt 2kt mW ,
(2.21)
F (x) =
for both = 0 and = 1. It is clear from Eq. (2.21) that kt 2 is enough to get the correct
top mass. Another possible way to increase the top mass is obtained by increasing the rank of
the SU(3) representation of the bulk fermion which couples to the localized fields. In this way
one can get a larger group-theoretical factor multiplying Eq. (2.19), at the cost of introducing
large representations of SU(3), lowering the Nave-Dimensional Analysis (NDA) estimate of the
cut-off.
2.4. Estimate of the cut-off
We estimate the cut-off using NDA, as the value at which the first fundamental coupling in
the theory gives rise to one-loop diagrams of the same sizeas the tree-level ones. For simplicity,
we consider the non-compact limit R , but with g5 / 2R = g fixed. The 5D loop factor
is 24 3 , so one gets
1 g52
g 2 R
(2.22)
=
1,
2 24 3
24 2
where the factor 1/2 in the first expression of Eq. (2.22) is due to the Z2 orbifold projection. We
should be careful since g5 k is effectively a new coupling constant. The most stringent bounds
arise indeed from this coupling, when g5 is the strong SU(3)c coupling constant. One finds
Rc 24 2 /(kgs2 ) 100/(kR), namely that the cut-off scales as 1/k. This rescaling can easily be understood in the non-compact case, by noting that k enters not only in the coupling,
g5 g5 k, but also in the propagators of the virtual states running in the loop. The latter is reabsorbed by sending q5 q5 /k, q5 being the momentum along the fifth direction, so that the
loop factor scales as 1/k. We see that NDA does not give strong bounds on the allowed values
of k, as long as k 10, which is above the values we have considered. If one takes instead the
electroweak coupling constant, w 1000/(kR) and no significant constraint arises.
Although k is practically not constrained by perturbativity, it is important to recall that the
explicit breaking of the SO(4, 1) Lorentz symmetry presents the drawback of generating several
counterterms in the effective action which are no longer constrained by SO(4, 1) to be absent
or equal between each other. This would result in a less constrained model and would also lead
to the appearance of additional radiative corrections, absent in the SO(4, 1) invariant case. As
an example of an effect of this sort, we would expect that at two-loop level the Higgs mass
will develop a linear divergence. Indeed, although we think that the Higgs mass term would still
be finite, being associated to non-local operators [28], the wave function renormalization of the
field A5 is no longer exactly cancelled by the gauge coupling constant renormalization, as in the
SO(4, 1) invariant case, giving rise to a divergence for the physical Higgs mass. Since the Higgs
mass term is one-loop induced, such divergence occurs at two-loop level. It is important to stress
that this two-loop linear divergence does not significantly destabilize the Higgs mass. It would
be interesting to better quantify how higher loop corrections, in general, modify the predictions
we have given for the Higgs mass, compactification scale and other parameters.
196
3. Results
In the present section we report the predictions of our model, as obtained by a numerical study. The
analysis is performed by randomly extracting the microscopic parameters within suitable ranges, and
computing the resulting values for the relevant observables, namely the Higgs and top masses and the
compactification scale. Obviously, we restrict to configurations for which the electroweak symmetry is
spontaneously broken, so that we discard points for which min = 0. Moreover, a cut min < 0.05 is applied
in order for the compactification scale 1/R to be sufficiently high. As described in Section 2, we consider
two variants of the model ( = 0 and = 1) which differ in the location of the boundary fields. The two
cases have many qualitative features in common, but they give rise to different quantitative predictions. In
particular, different ranges are obtained for the Higgs mass.
3.1. Set-up with = 0

As already mentioned, the cancellation of the quadratic term in the effective potential, which permits to
obtain small enough values of min , basically results in a correlation between kA and kt . Our numerical
study reveals indeed acceptable points (with min < 0.05) to be only found when 1.1 kt < kA < 1.5 kt .
For realizing the plots which follow, the bounds
t

0.5 < < 1.5,
2 < kt < 3,
0.75 < it < 7.5,
b
2 < kb < 3,
5 < < 7,
2 < ib < 7,
2.6 < kA < 3.9,
0.13 < A < 3.3,
have been used for the input microscopic parameters. First of all, in order to appreciate the effect of the
Lorentz violating parameters ki , let us see how the various observables depend on kt . Clearly, due to the
aforementioned cancellation condition, the behaviour in kA is similar to the latter, while the results are
weakly sensitive to kb (and to all others b parameters as well), due to the high value of b , which suppresses their contributions. In Fig. 2, the dependence on kt of the Higgs and top masses is shown. As
expected, the upper bound on the top mass linearly increases with kt and correct values (between the black
lines in the figure) are obtained for kt 2. On the other hand, as expected from Eq. (2.14), the Higgs mass
grows quadratically with kt .
It can be inferred from Fig. 2 that, at fixed kt , a certain correlation between the Higgs and top masses
exists. This is shown in Fig. 3, in which the Higgs mass is plotted versus the top one, and different colors
correspond to different values of kt . Fig. 4 shows mH and mt as a function of min . Higher Higgs and top
masses are favoured at small values of min , even though realistic values of mt can always be obtained. The
Fig. 2. Top (red in the web version) and Higgs (green in the web version) masses versus the input value of kt for = 0
(left) and = 1 (right). The interval between the two black lines corresponds to the physical top mass.
197
Fig. 3. Correlation between Higgs and top masses for = 0 (left) and = 1 (right). Different colors label input values of
kt in different ranges and the region among the two vertical black lines corresponds to the physical top mass.
Fig. 4. Dependence of the Higgs (green in the web version) and top (red in the web version) masses on min for = 0
(left) and = 1 (right). The interval between the two black lines corresponds to the physical top mass.
dependence on min of the upper bound for the top mass can be derived from Eqs. (2.17) and (2.18) in the
large regime.
Let us now restrict to realistic values for the top quark mass, in the range 169 GeV < mt < 180 GeV.10
With this cut (see Fig. 5) the Higgs mass is found to be in the range 250600 GeV, independently of the
value of min . In Fig. 6, finally, the mass (Mt = t / R) of the lightest non-standard fermions in the model,
coming from the KK towers of t and t , is plotted versus min . As we will discuss in Section 4, this mass
is important for estimating new physics effects arising in our model.
3.2. Set-up with = 1

As in the case of the previous subsection, acceptable vacua are found only if a certain correlation between
kt and kA is imposed. We take 1.1 kt < kA < 1.5 kt , and we restrict the other microscopic parameters
10 Due to the small statistics of our data, a cut on the bottom mass cannot be applied. In the present set of data, m goes
b
from 0.2 GeV up to 10 GeV, and is more or less uniformly distributed. As expected, no quantity is found to be correlated
with mb , so that realistic values of mb can be easily obtained.
198
Fig. 5. Higgs mass range as a function of min for = 0 (left) and = 1 (right). The top mass has been fixed to the
physical value 169 GeV < mt < 180 GeV.
Fig. 6. The mass Mt of the first non-standard fermions, as a function of min for = 0 (left) and = 1 (right). The top
mass has been fixed to the physical value 169 GeV < mt < 180 GeV.
to the ranges:

1.5 < kt < 2.5,

1.25 < kb < 2.25,
t
0.5 < < 1.5,
5 < b < 7,
0.75 < A < 3.5,
0.75 < it < 7.5,

2 < ib < 7.
In Fig. 2, the dependence on kt of the Higgs and top masses is shown. As in the previous case, the upper
bound on the top mass linearly increases with kt , while the Higgs mass grows quadratically. Note that,
differently from the = 0 case, configurations with a Higgs mass smaller or equal to the top one can be
found (see also Fig. 3). Fig. 4 shows the dependence of the Higgs and top masses on min . The behaviour
is similar to the one for = 0.
We now restrict our analysis to configurations with realistic top mass: 169 GeV < mt < 180 GeV. The
dependence of the Higgs mass on min is reported in Fig. 5. Allowed Higgs masses are in the range 125
400 GeV, and the distribution favours small values. Finally, the mass of the lightest non-standard fermions
is shown in Fig. 6. The dependence on min is analogous to the one found in the case in which all localized
fields are at the same fixed point.
3.3. The EW phase transition and a dark matter candidate

We have also studied the behaviour of our model at finite temperature, focusing in particular to the study
of how (if any) an EW Phase Transition occurs. This analysis is relevant to establish whether baryogenesis
199
at the electroweak scale could be a viable possibility or not. As known, this requires a first-order phase
transition where the order parameter H (TC )/TC 1, TC being the critical temperature of the transition.
The analysis is a simple generalization of [29], so that we will be very brief here and report only the final
results. The model develops a first-order phase transition at a temperature of order TC (0.11.5)/(2 R).
We get 0.01 H (TC )/TC 0.05 for = 0 and 0.02 H (TC )/TC 0.14 for = 1. The phase transition
strength, as expected, is approximately proportional to 1/m2H and this explains why the = 1 set-up appears
to have a stronger phase transition than the = 0 case. In both cases, however, the latter seems to be too
weak to open the possibility of achieving a baryogenesis at the electroweak phase transition.
Interestingly enough, there is a potential DM candidate particle in our model. Thanks to the global
U (1)A symmetry that we have imposed to our theory, under which only the antiperiodic fermions transform,
the lowest KK modes of both A and A are absolutely stable particles. After EWSB, with the U (1)
assignment we have given, the 6 of SU(3)w gives rise to a couple of four different towers of KK modes (see
the appendix of [8] for details), one for A and one for A :

k 2 (n + 1/2 + )2
(1)
2 + A
mn = MA
,
with q = +1, n [, +],
R2

k 2 (n + 1/2 + 2)2
(2)
2 + A
, with q = 0, n [, +],
mn = MA
R2

k 2 (n + 1/2)2
(3)
2 + A
mn = M A
,
with q = +2, n [0, +],
R2

k 2 (n + 1/2)2
(4)
2 + A
mn = MA
(3.1)
,
with q = 0, n [0, +],
R2
where q is the electromagnetic U (1)
charge of the state. The lightest particles in Eq. (3.1) are a couple
2 + k 2 (1 4)2 /(4R 2 ). Since m
of neutral states with mass mDM = MA
DM 1/R for the typical input
A
values of MA and kA , which corresponds to a mass of several TeV, such states are potential candidates to
explain the observed DM abundance in the Universe.
4. Estimate of phenomenological bounds

In this section we give an order-of-magnitude estimate of the main physical effects which we
believe to provide the most stringent bounds on our model. The purpose of this analysis is to
show that it is not trivially ruled out and to roughly estimate the allowed range in the parameter
space of the model.
4.1. Direct corrections: the ZbL bL vertex and FCNC
The first effect one should worry about is the non-universality of the EW couplings which is
present in our model, after EWSB, since the physical SM fermions (with diagonal propagators)
are a complicated mixture of fields in different representations of the underlying 5D SU(3)w EW
group. The eit,b couplings in Eq. (2.5), indeed, make the localized fields QL and bR , tR 11 to mix
with the KaluzaKlein towers of t,b and t,b in the 6 and 3 representations of SU(3), which
contain singlets, doublets and triplets of SU(2)w . Due to gauge invariance, the localized fields
only couple to the components of the bulk ones with the right (21/6 , 11/3 and 12/3 ) quantum
11 The (t, b) couple should now be thought to represent any of the three families of quarks.
200
numbers. After EWSB, however, mixing among fields in different representations are generated.
The latter give rise to tree-level corrections to the EW couplings through tree-level diagrams such
as in Fig. 7, in which all standard and non-standard fermions qn , belonging to the KaluzaKlein
towers of , , propagate. The couplings of qn to the SM gauge bosons is diagonal, since the
wave function of the latter in the extra dimension is flat. We focus in the following only on the
corrections to the vertex of the Z gauge boson, since this is the one which is experimentally more
constrained.12 Diagrams such as the one depicted in Fig. 7 give rise at the same time to a vertex
and a propagator correction. The physical correction to the vertex is obtained only after having
canonically normalized the kinetic terms for the external SM fermions. Gauge invariance allows
Yukawa couplings of the SM Higgs and the bottom quark bL only through triplets of SU(2),
arising from the 6 of SU(3)w . As it is clear from Fig. 7, the distortion is inversely proportional
to the mass of the triplets and proportional to the mixing between the doublet and the triplet.
Computing the overlap of the wave functions of the quark doublet with all the KK tower of the
triplets and then diagonalizing the resulting mass matrix is not straightforward. To a very good
approximation, however, the distortion is dominated by the first massive state of the KK tower,
with mass Mt . This is not only the lightest state of the tower, but also the one which mixes more
with the quark doublet. By considering only such a state, an explicit computation shows that for
the down quarks one has

1t 2 kt2 mW 2
1
gb

,
(4.1)
gb
1 23 sin2 W t 2 Z1 Mt
where Z1 is the factor appearing in Eq. (2.18), evaluated at = 0, for which Z1 = Z1t = Z1b . We
expect that a similar estimate will also apply for up quarks. The distortion caused by Eq. (4.1)
is always safely below current experimental bounds for all light quarks (including all leptons),
in which the bulk fermions are very massive and/or one can consider moderately small mixing
1u 0.1. The only exception is represented by the bottom quark, because the requirement of
getting a reasonable mass for the top quark obliges us to take 1t 1 and t 1. It turns out,
indeed, that Eq. (4.1) represents a strong constraint on the parameter space of the model, as can
be seen from Fig. 8, where gb /gb is reported as a function of min , the most relevant parameter.
Considering that (gb /gb )exp 102 103 , essentially all values of min 2 102 (1/R
Fig. 7. The leading tree-level correction to the ZbL bL vertex and to the bL propagator. The distortion of the ZbL bL
coupling is generated through the Higgs-mediated mixing of bL with one component (3 ) of the 31/3 triplet contained
in t .
12 The mixing between the KK modes of A and the Z boson induces a further correction to the Zb b vertex, which
X
L L
however turns out to be negligible.
201
4 TeV) are ruled out. It is interesting to notice that the constraint imposed by ZbL bL also plays
an important role in the warped model of [13,30]. In the latter case, as in ours, the requirement
of having an acceptable top mass forbids to lower this distortion.
Another important issue to consider, closely related to the non-universality of the EW couplings, is the suppression of the FCNC which are typically generated at tree-level when integrating out the massive KK modes. For simplicity, consider here the case in which the bulkto-boundary couplings i are diagonal in flavour space and the non-trivial flavour structure, as
in [8], is totally encoded in non-trivial bulk mass matrices, which can now involve not only the
bulk mass terms Mi but also the Lorentz violating factors ki . The FCNC are induced in our model
by tree-level couplings which arise from diagrams such as the one in Fig. 7, in which the nonstandard triplet 3 switches at some point to the KK mode of a different family. Their structure
can then be inferred from Eq. (4.1), which is in fact the leading correction to a flavour preserving
neutral current or to a FCNC, modulo the flavour textures which we will not specify here and
conservatively take to be of order one. The typical bound on the couplings of FCNC involving
b quarks is 102 . Since these couplings are equal to or smaller than the value estimated by
Eq. (4.1), they do not represent any problem. In other words, the strongest bounds in b-physics
arise from the ZbL bL correction.
The bounds on the couplings of FCNC involving d and s quarks are instead much stronger,
105 . In particular, one should worry that, in presence of a generic flavour mixing, a light
quark (d or s) can first switch to a triplet of the heavy KK tower of the bulk fermion of the
corresponding up quark (u or c), which then switches to the much lighter triplet of the KK tower
of the top quark. The latter emits a Z boson and then switches to another heavy KK tower and
thus eventually to another light quark (s or d), resulting in a FCNC. We can estimate the treelevel coupling gFCNC of this FCNC vertex from Eq. (4.1). Neglecting the factor Z1 , which is
typically of order 1, one has

1c 1u mW 2
.
gFCNC c u
(4.2)

Mt
Considering that for the c and the u quarks one can take c u 10 and, at the same time, one can
naturally take 1u,c 0.1, it is reasonable to expect that gFCNC can be made smaller than 105
or 106 .
FCNC are also induced by the exchange of the massive KK modes of the Z gauge boson
and gluons [31], due to the non-universality of their couplings to different families. This effect,
which is present even in the absence of EWSB, comes from the fact that in our model, as required
Fig. 8. Correction to the ZbL bL coupling gb /gb as a function of min for = 0 (left) and = 1 (right).
202
for explaining the mass hierarchy, quarks of different families have different wave functions.
By choosing for different generations the same distribution of localized fields, we can however
strongly suppress this effect.13 Flavour non-universality in the KK couplings, indeed, arises in
this case only from diagrams in which the brane quark q is changed to a bulk KK fermion, which
emits a KK gauge boson, and then back to the brane. The effective coupling of this diagram can
be estimated as
gKK
q 2
q2.
gKK
(4.3)
For the light families, if q is moderately small ( 101 ), we expect the coupling (4.3) to be
naturally of order of 103 104 , since q 2 10. In this way, the resulting FCNCdue to their
stronger couplings, gluons give the dominant contributionis of the same order of magnitude of
that estimated for the Z and thus within the current limits.
4.2. Oblique corrections:
The leading worrisome universal deviation from the SM in our theory is due to the localized
mass terms for the gauge field AX , which are necessarily present, as explained in Section 2.2, for
anomaly cancellation. They induce tree-level mixing between the Z boson and the KK modes
of AX , which are conveniently expressed in the following term:

2
2
1
Z
3 4 sin2 W AX (y) ,
Lm =
(4.4)
2
2
2 cos W R
where Z = cos W A30 sin W AY,0 is the usual 4D SM gauge boson, whereas AX (y) is the 5D
field in Eq. (2.6). Since, due to the anomaly, the wave functions (2.6) are not regular cosines,
the AX, n fields couple to the zero-mode of the Z via the ZAX term in Eq. (4.4), see also Fig. 9.
The physical mass eigenstate of the Z, given by Eq. (2.8), is then a complicated mixture of the
full KK towers of AX , which lead to a deviation from the SM relation mZ = mW / cos W or
= 1. The deviation = 1 (or T ) is easily computed by expanding in series the sines in
Eq. (2.8), or diagrammatically, by considering the ZAX mixing terms in Eq. (4.4). The result is
=
4
8 2
2
min
.
+ O min
9(1 + 3)
(4.5)
The current experimental limit on is exp 103 , so that Eq. (4.5) leads to a bound on min
which is min 102 (1/R 8 TeV) for = 0 and min 2 102 (1/R 4 TeV) for = 1.
Fig. 9. Leading tree-level correction to .
13 Notice that this is not possible in the set-up = 1, which is then disfavored, as far as FCNC are concerned.
203
For = 0, Eq. (4.5) represents the strongest bound on the model, whereas for = 1 it equals the
bound found in the last subsection from Zbb.

We have not computed the corrections induced by four fermion operators, since we expect
that the associated bounds in our model are roughly the same as the ones estimated in [18]
for universal theories in which the gauge and Higgs field are bulk fields and the SM fermions
are localized states. As such, the resulting bound on 1/R is approximately the same as the one
coming from ZbL bL or .14 Other universal corrections arise at one-loop level. The leading
ones are given by vector-like massive Dirac fermions and are safely small.
4.3. The fine-tuning
The phenomenological bounds estimated before essentially result in a bound on min which
is min 102 . As can be seen from, say, Fig. 5, low values of min are not so uncommonly
obtained. It is however important to better quantify how much fine-tuning is necessary to impose
in the microscopic parameters of our theory to get min 102 . The exact determination of such
tuning is actually a very challenging task due to the difficulty of choosing a precise definition of
the tuning itself.
The fine-tuning, according to a commonly used definition [33], is related to the sensitivity of
the physical observables to the microscopic parameters of the theory. In such a view, the finetuning can be estimated by computing the logarithmic derivative of the observables with respect
to the parameters. In our case the most sensitive observable is the Higgs VEV (namely min ),
while the most relevant parameters are the Lorentz violating couplings kt and kA or, better, their
ratio = kA /kt . Computing the derivative
C()
log min
,
log
(4.6)
one finds C 103 for min 102 . If one takes 1/C as an estimate of fine-tuning in the model,
this should translate in a fine-tuning of O().
In [34], an improved definition of fine-tuning was proposed. According to this prescription,
the value of the logarithmic derivative C() at a given point must be divided by the average
value of C in a suitable range of the microscopic parameters. This should allow to distinguish
spurious high sensitivity to the parameters from real fine-tuning due to cancellations. In this
procedure, however, an appropriate definition of the range of the microscopic parameters, in
which the average will be performed, must be chosen. In the present case, the result crucially
depends on what we assume to be the natural values of , i.e. on what we decide to be its
natural interval of variation. If we take values of for which the EWSB is realized (roughly
0 1), i.e. all values for which 0 < min < 1/2, and average over all the resulting vacua, the
fine-tuning turns out to be roughly as before of O().
However, we notice that most of the vacua in this ensemble are either at min = 0 or at large
values min 1/3. If one disregards these points, by taking for instance a range of for which
0 < min < 1/4, one finds a relevant spurious sensitivity of min on . The range of for which
0 < min < 1/4 is quite small and the real fine-tuning is found now, applying the proposal
of [34], to be of O(10%).
14 Notice that in [18] all the oblique parameters, including the effects of four fermion operators, are encoded in four
T , W and Y . The parameters S and T , modulo a normalization, are defined as in [32], but with
parameters, denoted S,
respect to gauge fields which are a mixture of the SM vector bosons with their non-standard KK modes.
204
Although it is hard to draw a conclusion about the fine-tuning in our model, in the light of
these different estimates, we think that it might be fair to say that the bound min 102 could
be translated to a fine-tuning of O(%).
5. Is our model really a 5D theory?
In this work we have essentially shown how it is possible to get a potentially realistic model
with gauge-Higgs unification at the price of explicitly breaking the SO(4, 1)/SO(3, 1) Lorentz
generators in the fermionic sector. In light of this breaking, one could wonder whether it is correct
to consider our model as a canonical 5D theory or not. Indeed, contrary to the usual spontaneous breaking of the SO(4, 1)/SO(3, 1) Lorentz symmetry induced by the compactification,
which implies that at short distances x R the model is effectively a 5D Lorentz invariant
theory (in the bulk), the explicit breaking we advocate implies that at arbitrarily short scales the
SO(4, 1) symmetry is not recovered. This is clearly a theoretical issue, which is mainly related
to the possible existence and form of an underlying UV completion of our model. Moreover,
the concept of gauge-Higgs unification itself relies on the existence of a 5D interpretation. It
is clear that we can always consider our model as an IR effective description of a 4D moose
theory for which the accidental SO(4, 1) Lorentz symmetry is not recovered in the fermionic
sector [7]. From this point of view, our model would resemble more a moose-based little Higgs
model rather than a gauge-Higgs unification model. We would like to point out, however, that the
SO(4, 1) Lorentz breaking we advocate in this paper can have a simple origin in the context of
a purely 5D theory. A particularly elegant and interesting explanation is the following. Consider
an axion-like field , which for simplicity we take to be dimensionless, invariant under the shift
+ 2 . In light of this shift symmetry, one can take twisted periodicity conditions for ,
which reads
(y + 2R) = (y) + 2.
(5.1)
ScherkSchwarz reductions of the form (5.1) are not new, appearing in supergravity as a way to
obtain gauged SUGRA or theories with fluxes (see e.g. [35]). Consistency of Eq. (5.1) with the
Z2 orbifold action y y requires that should be Z2 odd. This is welcome, implying that all
the excitations of are massive. Due to the twisted condition (5.1), the VEV of is non-trivial.
The background configuration 0 which satisfies the field equations of motion and Eq. (5.1) is
y
,
(5.2)
R
which clearly induces a spontaneous breaking of the SO(4, 1)/SO(3, 1) Lorentz symmetry. The
Lorentz violating factors k introduced in Eq. (2.5) are then reinterpreted as due to couplings
involving and fermion bilinears. It turns out that if we also impose a Z2 global symmetry
under which , the lowest-dimensional operator which couples and bulk fermions
read
(5.3)
M N M D N ,
f2
0 (y) =
where is a dimensionless coupling and f is the decay constant. When = 0 , the

operator (5.3) precisely induces the Lorentz violating terms which appear in the Lagrangian (2.5).
Since we have considered in our model values of k which are not close to 1, the effective coupling
constant of the operator (5.3) is strong (of order 1) and thus insertions of this operator have to be
205
resummed. This is what we have effectively done in our previous analyses.15 Notice that Eq. (5.2)
can also be interpreted as a non-vanishing flux for the 1-form field-strength H1 = d dy
or else for a non-vanishing flux for the Hodge dual 4-form field-strength H4 dx 0 dx 1
dx 2 dx 3 .
We think that the above picturein no way necessary for the model we have presented
shows that the Lorentz violating factors kj can have a natural origin in a 5D framework.
In light of the rescaling (2.13), the factors kj effectively imply that different fermions see
a different radius of compactification. Their effect is then quite similar to recent ideas in the
context of Higgless models in 5D warped models, in which it has been advocated that different
fields could propagate in internal spaces with different sizes [36,37].
6. Outlook
In this paper we have shown that realistic models based on gauge-Higgs unification in 5D
flat space can be constructed, but at the price of breaking the SO(4, 1) Lorentz symmetry in the
bulk. Our key observation is that the stability of the Higgs potential is mostly provided by the
5D gauge symmetry rather than the SO(4, 1) symmetry. Breaking the latter results in additional
divergencies and in an increasing number of independent operators to be considered, which however do not significantly destabilize the Higgs potential. Somehow, the SO(4, 1) breaking models
we propose represent a sort of middle course between little Higgs models and the previously considered SO(4, 1) invariant models with gauge-Higgs unification. For simplicity, we have focused
our attention on a variant of the minimal model of [8], where an additional antiperiodic bulk
fermion is introduced. The latter state is crucial to increase the value of the compactification
scale above the TeV scale and, as a by product, its lightest neutral KK state is a possible DM
candidate. Clearly, several other models, already constructed or not, could be considered in this
Lorentz non-invariant scenario. We have also shown that our model could pass various phenomenological tests, such as the universality of the couplings, EWPT and FCNC.
An important issue that we have not considered at all in this paper regards the experimental
signatures of our model. In the light of the forthcoming Large Hadron Collider (LHC), it is
particularly important to address which are (if any) the distinct collider signatures of our model.
We plan to address in a future work the latter issue, as well as a detailed study of the viability of
our theory as a realistic proposal to go beyond the SM.
Acknowledgements
We would like to thank G. Cacciapaglia, C. Csaki, A. Pomarol, A. Romanino, C.A. Scrucca,
L. Silvestrini, A. Strumia and P. Ullio for useful discussions and comments. We also thank
G. Cacciapaglia, C. Csaki and S.C. Park for sharing a draft of [14] with us prior to publication. This work is partially supported by the European Communitys Human Potential Programme under contract MRTN-CT-2004-005104 and by the Italian MIUR under contract PRIN2003023852.
15 Of course, operators similar to (5.3) involving more s should be taken into account. However, if we assume that
is large so that one can take Rf 1, these are naturally suppressed.
206
References
[1] I. Antoniadis, Phys. Lett. B 246 (1990) 377.
[2] G.R. Dvali, S. Randjbar-Daemi, R. Tabbash, Phys. Rev. D 65 (2002) 064021, hep-ph/0102307;
L.J. Hall, Y. Nomura, D.R. Smith, Nucl. Phys. B 639 (2002) 307, hep-ph/0107331;
I. Antoniadis, K. Benakli, M. Quiros, New J. Phys. 3 (2001) 20, hep-th/0108005;
M. Kubo, C.S. Lim, H. Yamashita, Mod. Phys. Lett. A 17 (2002) 2249, hep-ph/0111327;
N. Haba, M. Harada, Y. Hosotani, Y. Kawamura, Nucl. Phys. B 657 (2003) 169, hep-ph/0212035;
N. Haba, M. Harada, Y. Hosotani, Y. Kawamura, Nucl. Phys. B 669 (2003) 381, Erratum;
N. Haba, Y. Shimizu, Phys. Rev. D 67 (2003) 095001, hep-ph/0212166;
K.w. Choi, et al., JHEP 0402 (2004) 037, hep-ph/0312178;
I. Gogoladze, Y. Mimura, S. Nandi, Phys. Lett. B 560 (2003) 204, hep-ph/0301014;
I. Gogoladze, Y. Mimura, S. Nandi, Phys. Lett. 562 (2003) 307, hep-ph/0302176;
I. Gogoladze, Y. Mimura, S. Nandi, Phys. Rev. D 69 (2004) 075006, hep-ph/0311127;
C.A. Scrucca, M. Serone, L. Silvestrini, A. Wulzer, JHEP 0402 (2004) 049, hep-th/0312267;
A. Wulzer, Gauge-Higgs unification in six dimensions, hep-th/0405168;
C. Biggio, M. Quiros, Nucl. Phys. B 703 (2004) 199, hep-ph/0407348.
[3] G. Burdman, Y. Nomura, Nucl. Phys. B 656 (2003) 3, hep-ph/0210257.
[4] D.B. Fairlie, Phys. Lett. B 82 (1979) 97;
D.B. Fairlie, J. Phys. G 5 (1979) L55;
N.S. Manton, Nucl. Phys. B 158 (1979) 141;
P. Forgacs, N.S. Manton, Commun. Math. Phys. 72 (1980) 15;
S. Randjbar-Daemi, A. Salam, J. Strathdee, Nucl. Phys. B 214 (1983) 491;
N.V. Krasnikov, Phys. Lett. B 273 (1991) 246;
H. Hatanaka, T. Inami, C. Lim, Mod. Phys. Lett. A 13 (1998) 2601, hep-th/9805067.
[5] M. Serone, AIP Conf. Proc. 794 (2005) 139, hep-ph/0508019.
[6] N. Arkani-Hamed, et al., JHEP 0208 (2002) 021, hep-ph/0206020;
N. Arkani-Hamed, A.G. Cohen, E. Katz, A.E. Nelson, JHEP 0207 (2002) 034, hep-ph/0206021.
[7] N. Arkani-Hamed, A.G. Cohen, H. Georgi, Phys. Lett. B 513 (2001) 232, hep-ph/0105239.
[8] C.A. Scrucca, M. Serone, L. Silvestrini, Nucl. Phys. B 669 (2003) 128, hep-ph/0304220.
[9] C. Csaki, C. Grojean, H. Murayama, Phys. Rev. D 67 (2003) 085012, hep-ph/0210133.
[10] Y. Grossman, M. Neubert, Phys. Lett. B 474 (2000) 361, hep-ph/9912408;
T. Gherghetta, A. Pomarol, Nucl. Phys. B 586 (2000) 141, hep-ph/0003129.
[11] N. Arkani-Hamed, M. Schmaltz, Phys. Rev. D 61 (2000) 033005, hep-ph/9903417.
[12] R. Contino, Y. Nomura, A. Pomarol, Nucl. Phys. B 671 (2003) 148, hep-ph/0306259;
K.y. Oda, A. Weiler, Phys. Lett. B 606 (2005) 408, hep-ph/0410061;
Y. Hosotani, M. Mabe, Phys. Lett. B 615 (2005) 257, hep-ph/0503020.
[13] K. Agashe, R. Contino, A. Pomarol, Nucl. Phys. B 719 (2005) 165, hep-ph/0412089.
[14] G. Cacciapaglia, C. Csaki, S.C. Park, Fully radiative electroweak symmetry breaking, hep-ph/0510366.
[15] G. Servant, T.M.P. Tait, Nucl. Phys. B 650 (2003) 391, hep-ph/0206071.
[16] K. Agashe, A. Delgado, M.J. May, R. Sundrum, JHEP 0308 (2003) 050, hep-ph/0308036.
[17] G. Burdman, Y. Nomura, Phys. Rev. D 69 (2004) 115013, hep-ph/0312247;
G. Cacciapaglia, C. Csaki, C. Grojean, J. Terning, Phys. Rev. D 71 (2005) 035015, hep-ph/0409126.
[18] R. Barbieri, A. Pomarol, R. Rattazzi, A. Strumia, Nucl. Phys. B 703 (2004) 127, hep-ph/0405040.
[19] J. Scherk, J.H. Schwarz, Phys. Lett. B 82 (1979) 60;
J. Scherk, J.H. Schwarz, Nucl. Phys. B 153 (1979) 61.
[20] Y. Hosotani, Phys. Lett. B 126 (1983) 309;
Y. Hosotani, Phys. Lett. B 129 (1983) 193;
Y. Hosotani, Ann. Phys. 190 (1989) 233.
[21] M. Serone, A. Wulzer, Class. Quantum Grav. 22 (2005) 4621, hep-th/0409229;
A. Wulzer, Orbifold resolutions with general profile, hep-th/0506210.
[22] G. Martinelli, M. Salvatori, C.A. Scrucca, L. Silvestrini, JHEP 0510 (2005) 037, hep-ph/0503179.
[23] C.A. Scrucca, M. Serone, L. Silvestrini, F. Zwirner, Phys. Lett. B 525 (2002) 169, hep-th/0110073;
L. Pilo, A. Riotto, Phys. Lett. B 546 (2002) 135, hep-th/0202144;
R. Barbieri, R. Contino, P. Creminelli, R. Rattazzi, C.A. Scrucca, Phys. Rev. D 66 (2002) 024025, hep-th/0203039;
C.A. Scrucca, M. Serone, M. Trapletti, Nucl. Phys. B 635 (2002) 33, hep-ph/0203190.
207
[24] M.B. Green, J.H. Schwarz, Phys. Lett. B 149 (1984) 117;
E. Witten, Phys. Lett. B 149 (1984) 351;
M. Dine, N. Seiberg, E. Witten, Nucl. Phys. B 289 (1987) 589.
[25] G. von Gersdorff, N. Irges, M. Quiros, Nucl. Phys. B 635 (2002) 127, hep-th/0204223;
G. von Gersdorff, N. Irges, M. Quiros, Finite mass corrections in orbifold gauge theories, hep-ph/0206029.
[26] G. Cacciapaglia, Talk given at the 13th International Conference on Supersymmetry and Unification of Fundamental
Interactions, IPPP Durham, 1823 July 2005.
[27] N. Haba, Y. Hosotani, Y. Kawamura, T. Yamashita, Phys. Rev. D 70 (2004) 015010, hep-ph/0401183.
[28] N. Arkani-Hamed, et al., Nucl. Phys. B 605 (2001) 81, hep-ph/0102090;
A. Masiero, C.A. Scrucca, M. Serone, L. Silvestrini, Phys. Rev. Lett. 87 (2001) 251601, hep-ph/0107201.
[29] G. Panico, M. Serone, JHEP 0505 (2005) 024, hep-ph/0502255.
[30] K. Agashe, R. Contino, The minimal composite Higgs model and electroweak precision tests, hep-ph/0510164.
[31] A. Delgado, A. Pomarol, M. Quiros, JHEP 0001 (2000) 030, hep-ph/9911252.
[32] M.E. Peskin, T. Takeuchi, Phys. Rev. D 46 (1992) 381.
[33] R. Barbieri, G.F. Giudice, Nucl. Phys. B 306 (1988) 63.
[34] G.W. Anderson, D.J. Castano, Phys. Lett. B 347 (1995) 300, hep-ph/9409419.
[35] E. Bergshoeff, et al., Nucl. Phys. B 470 (1996) 113, hep-th/9601150.
[36] G. Cacciapaglia, et al., Top and bottom: A brane of their own, hep-ph/0505001.
[37] R. Foadi, C. Schmidt, An effective Higgsless theory: Satisfying electroweak constraints and a heavy top quark,
hep-ph/0509071.
The see-saw mechanism, neutrino Yukawa couplings,

LFV decays li lj + and leptogenesis
S.T. Petcov a,b,1 , W. Rodejohann c , T. Shindou a,b, , Y. Takanishi d,2
a Scuola Internazionale Superiore di Studi Avanzati, I-34014 Trieste, Italy
b Istituto Nazionale di Fisica Nucleare, Sezione di Trieste, I-34014 Trieste, Italy
c Physik-Department, Technische Universitt Mnchen, James-Franck-Strae, D-85748 Garching, Germany
d The Abdus Salam International Centre for Theoretical Physics, Strada Costiera 11, I-34100 Trieste, Italy
Received 21 November 2005; received in revised form 9 January 2006; accepted 17 January 2006
Abstract
The LFV charged lepton decays e + , e + and + and thermal leptogenesis are
analysed in the MSSM with see-saw mechanism of neutrino mass generation and soft SUSY breaking
with universal boundary conditions. The case of hierarchical heavy Majorana neutrino mass spectrum,
M1 M2 M3 , is investigated. Leptogenesis requires M1 109 GeV. Considering the natural range of
values of the heaviest right-handed Majorana neutrino mass, M3 5 1013 GeV, and assuming that the
soft SUSY breaking universal gaugino and/or scalar masses have values in the range of few 100 GeV,
we derive the combined constraints, which the existing stringent upper limit on the e + decay rate
and the requirement of successful thermal leptogenesis impose on the neutrino Yukawa couplings, heavy
Majorana neutrino masses and SUSY parameters. Results for the three possible types of light neutrino mass
spectrumnormal and inverted hierarchical and quasi-degenerateare obtained.
1. Introduction
The experiments with solar, atmospheric, reactor and accelerator neutrinos [15] have provided during the last several years compelling evidence for the existence of non-trivial 3-neutrino
E-mail addresses: shindou@sissa.it (T. Shindou), yasutaka@sissa.it (Y. Takanishi).

1 Also at: Institute of Nuclear Research and Nuclear Energy, Bulgarian Academy of Sciences, 1784 Sofia, Bulgaria.
2 Address after 22 November 2005: Scuola Internazionale Superiore di Studi Avanzati, I-34014 Trieste, Italy.
doi:10.1016/j.nuclphysb.2006.01.034
S.T. Petcov et al. / Nuclear Physics B 739 (2006) 208233
209
mixing in the weak charged-lepton current (see, e.g., [6]):

lL =
3
Ulj j L ,
l = e, , ,
(1)
j =1
where lL are the flavour neutrino fields, j L is the field of neutrino j having a mass mj and U is
the PontecorvoMakiNakagawaSakata (PMNS) mixing matrix [7], U UPMNS . The existing
data, including the data from the 3 H -decay experiments [8] imply that the massive neutrinos j
are significantly lighter than the charged leptons and quarks: mj < 2.3 eV (95% C.L.).3 A natural explanation of the smallness of neutrino masses is provided by the see-saw mechanism of
neutrino mass generation [11]. The see-saw mechanism predicts the light massive neutrinos j
to be Majorana particles. An integral part of the mechanism are the heavy right-handed (RH)
Majorana neutrinos [12]. In grand unified theories (GUT) the masses of the heavy RH Majorana neutrinos are typically by a few to several orders of magnitude smaller than the scale of
unification of the electroweak and strong interactions, MGUT
= 2 1016 GeV. In this case the
CP-violating decays of the heavy RH Majorana neutrinos in the Early Universe could generate,
through the leptogenesis scenario, the observed baryon asymmetry of the Universe [13].
The existence of the flavour neutrino mixing, Eq. (1), implies that the individual lepton
charges, Ll , l = e, , , are not conserved (see, e.g., [14]), and processes like e + ,
e + e+ + e , e + , + , + (A, Z) e + (A, Z), etc., should
take place. Stringent experimental upper limits on the branching ratios and relative cross sections
of the indicated |Ll | = 1 decays and reactions have been obtained [1517] (90% C.L.):
B( e + ) < 1.2 1011 ,
B( 3e) < 1.2 1012 ,
R( + Ti e + Ti) < 4.3 1012 ,

B( + ) < 6.8 108 ,
B( e + ) < 1.1 107 .
(2)
Future experiments with increased sensitivity can reduce the current bounds on B( e + ),
B( + ) and on R( + (A, Z) e + (A, Z)) by a few orders of magnitude (see,
e.g., [18]). In the experiment MEG under preparation at PSI [19] it is planned to reach a sensitivity to

B( e + ) 1013 1014 .
(3)
It has been noticed a long time ago that in SUSY (GUT) theories with see-saw mechanism
of neutrino mass generation, the rates and cross sections of the LFV processes can be strongly
enhanced [20]. If the SUSY breaking occurs via soft terms with universal boundary conditions
at a scale MX above the RH Majorana neutrino mass scale MR , MX > MR ,4 the renormalisation
group (RG) effects transmit the LFV from the neutrino mixing at MX to the effective mass terms
of the scalar leptons at MR even if the soft SUSY breaking terms at MX are flavour symmetric
and conserve the lepton charges Ll . As a consequence of these RG-induced new LFV terms in the
3 More stringent upper limit on m follows from the constraints on the sum of neutrino masses obtained from cosj
mological/astrophysical observations, namely, the CMB data of the WMAP experiment combined with data from large

scale structure surveys (2dFGRS, SDSS) [9]: j mj < (0.72.0) eV (95% C.L.), where we have included a conservative
estimate of the uncertainty in the upper limit (see, e.g., [10]).
4 The possibility of flavour-blind SUSY breaking of interest is realised, e.g., in gravity-mediation SUSY breaking
scenarios (see, e.g., [21]).
210
effective Lagrangian at MR < MX , the LFV processes can proceed with rates and cross sections
which are within the sensitivity of presently operating and future planned experiments [20,22]
(see also, e.g., [2332]). In contrast, in the non-supersymmetric case, the rates and cross sections
of the LFV processes are suppressed by the factor [33] (see also [34]) (mj /MW )4 < 6.7 1043 ,
MW being the W mass, which renders them unobservable.
One of the basic ingredients of the see-saw mechanism is the matrix of neutrino Yukawa
couplings, Y . Leptogenesis depends on Y as well [13] (see also [35,36] and the references
quoted therein). In the large class of SUSY models with see-saw mechanism and SUSY breaking
mediated by flavour-universal soft terms at a scale MX > MR we will consider, the probabilities
of LFV processes also depend strongly on Y (see, e.g., [23,24]). The matrix Y can be expressed
in terms of the light neutrino and heavy RH neutrino masses, the neutrino mixing matrix UPMNS ,
and an orthogonal matrix R [23]. Leptogenesis can take place only if R is complex. The matrix
Y depends, in particular, on the Majorana CP-violation (CPV) phases in the PMNS matrix
UPMNS [37].5 It was shown in [28,32] that if the heavy Majorana neutrinos are quasi-degenerate
in mass, the Majorana phases can affect significantly the predictions for the rates of LFV decays
e + , e + , etc., in the class of SUSY theories of interest.
The matrix Y can be defined, strictly speaking, only at scales not smaller than MR . The
probabilities of LFV processes depend on Y at the scale MR , Y = Y (MR ). In order to evaluate
Y (MR ) one has to know, in general, the light neutrino masses mj and the mixing matrix UPMNS
at MR , i.e., one has to take into account the RG running of mj and UPMNS from the scale
MZ 100 GeV, at which the neutrino mixing parameters are measured, to the scale MR (see,
e.g., [32,42] and the references quoted therein). However, if the RG running of mj and UPMNS
is sufficiently small, Y (MR ) will depend on the values of the light neutrino masses mj and the
mixing angles and CP-violation phases in UPMNS at the scale MZ .
Working in the framework of the class of SUSY theories with see-saw mechanism and soft
SUSY breaking with flavour-universal boundary conditions at a scale MX > MR , we investigate in the present article the combined constraints, which the existing stringent upper limit of
the e + decay rate and the requirement of successful thermal leptogenesis impose on
the neutrino Yukawa couplings, heavy Majorana neutrino masses and on the SUSY parameters.
The case of hierarchical heavy Majorana neutrino mass spectrum, M1 M2 M3 , is considered. Leptogenesis requires M1 109 GeV. The analysis is performed assuming that the heaviest
RH Majorana neutrino has a mass M3 51013 GeV, and that the soft SUSY breaking universal
gaugino and/or scalar masses (at the scale MX ) have values in the range of few 100 GeV. One
gets naturally M3 5 1013 GeV in GUT (SUSY GUT) theories with see-saw mechanism of
neutrino mass generation,6 see, e.g., [43] and the references quoted therein. If the SUSY breaking
universal gaugino and/or scalar masses have values in the few 100 GeV range, supersymmetric
particles will be observable in the experiments under preparation at the LHC (see, e.g., [44]).
We find, in accordance with the results of earlier studies (see, e.g., [23,25,27,30]), that under the
5 Obtaining information about the Majorana CPV phases in the PMNS matrix U
PMNS if the massive neutrinos are
proved to be Majorana particles would be a remarkably challenging problem. The oscillations of flavour neutrinos,
l l and l l , l, l = e, , , are insensitive to the two Majorana phases in UPMNS [37,38]. The only feasible
experiments that at present have the potential of establishing the Majorana nature of light neutrinos j and of providing
information on the Majorana phases in UPMNS are the experiments searching for neutrinoless double beta (()0 )decay, (A, Z) (A, Z + 2) + e + e (see, e.g., [14,3941]).
2
6 A naive explanation of this fact follows from the simplified see-saw formula m
= mD /MR : using mD 175 GeV
2
14
and m 5 10 eV one finds MR 6 10 GeV.
211
indicated assumptions, the existing stringent upper limit on the e + decay rate cannot be
satisfied, unless the terms proportional to M3 in the e + decay amplitude are absent or
strongly suppressed. The requisite decoupling of the terms M3 from the e + decay
amplitude is realised if the matrix R has a specific form which admits a parametrisation with just
one complex angle. Using the latter we obtain results for the three types of light neutrino mass
spectrumnormal and inverted hierarchical (NH and IH), and quasi-degenerate (QD). For each
of the three types of spectrum we derive the leptogenesis lower bound on the mass of the lightest
RH Majorana neutrino M1 . The lower bounds thus found in the cases of IH and QD spectrum are
1013 GeV. The upper limit on B( e + ) in these two cases can be satisfied for specific
ranges of values of the soft SUSY breaking parameters implying relatively large masses of the
supersymmetric particles. Using these soft SUSY breaking parameters we derive predictions for
B( e + ), B( e + ) and B( + ) which are compatible with the requirement of
successful leptogenesis.
Our analysis is performed under the condition of negligible RG effects for the light neutrino
masses mj and the mixing angles and CP-violation phases in UPMNS . The RG effects in question
are negligible in the class of SUSY theories we are considering in the case of hierarchical light
neutrino mass spectrum (see, e.g., [32,42]). The same is valid for quasi-degenerate j mass spectrum provided the parameter tan < 10, tan being the ratio of the vacuum expectation values
of the up- and down-type Higgs doublet fields in SUSY extensions of the standard theory.
2. General considerations
2.1. Neutrino mixing parameters from neutrino oscillation data
We will use the standard parametrisation of the PMNS matrix UPMNS (see, e.g., [40]):

c12 c13
UPMNS = s12 c23 c12 s23 s13 ei
s12 s23 c12 c23 s13 ei
M
diag 1, ei 2 , ei 2 ,
s12 c13
c12 c23 s12 s23 s13 ei
c12 s23 s12 c23 s13 ei
s13 ei
s23 c13
c23 c13
(4)
where cij = cos ij , sij = sin ij , the angles ij = [0, /2], = [0, 2] is the Dirac CPviolating phase and and M are two Majorana CP-violation phases [37,45]. One can identify
the neutrino mass squared difference responsible for solar neutrino oscillations, m2 , with
m221 m22 m21 , m2 = m221 > 0. The neutrino mass squared difference driving the dominant ( ) oscillations of atmospheric ( ) is then given by |m2A | = |m231 |
=
2
2
|m32 | m21 . The corresponding solar and atmospheric neutrino mixing angles, and A ,
coincide with 12 and 23 , respectively. The angle 13 is limited by the data from the CHOOZ
and Palo Verde experiments [46].
The existing neutrino oscillation data allow us to determine m221 , |m231 |, sin2 12 and
2
sin 223 with a relatively good precision and to obtain rather stringent limits on sin2 13 (see,
e.g., [2,47,48]). The best fit values of m221 , sin2 12 , |m231 | and sin2 223 read7 :
7 The data imply, in particular, that maximal solar neutrino mixing is ruled out at 6 ; at 95% C.L. one finds
cos 2 0.26 [47], which has important implications [49].
212
m2 = 8.0 105 eV2 ,

212
m = 2.1 103 eV2 ,
31
sin2 12 = 0.31,
(5)
sin 223 = 1.0.

2
(6)
A combined 3 oscillation analysis of the solar neutrino, KamLAND and CHOOZ data gives [47]
sin2 13 < 0.024 (0.044),
at 95% (99.73%) C.L.

m221 ,
12 , |m231 |
(7)
2
The neutrino oscillation parameters

sin
and sin 223 are determined by the
existing data at 3 with an error of approximately 12%, 24%, 50% and 16%, respectively. These
parameters can (and very likely will) be measured with much higher accuracy in the future (see,
e.g., [6]). In all further numerical estimates we use the best fit values of m221 , sin2 12 , |m231 |
and sin2 223 . Whenever the parameter sin 13 is also relevant in the calculations, we specify the
value used.
The sign of m2A = m231 , as it is well known, cannot be determined from the present (SK
atmospheric neutrino and K2K) data. The two possibilities, m231(32) > 0 or m231(32) < 0 correspond to two different types of -mass spectrum:
with normal hierarchy m1 < m2 < m3 , m2A = m231 > 0, and
with inverted hierarchy m3 < m1 < m2 , m2A = m232 < 0.
Depending on the sign of m2A , sgn(m2A ), and the value of the lightest neutrino mass, min(mj ),
the -mass spectrum can be
(m2 )1/2 0.009 eV, m3
Normal hierarchical: m1 m2 m3 , m2 =
= |m2A |1/2

0.045 eV;
Inverted hierarchical: m3 m1 < m2 , with m1,2
= |m2A |1/2 0.045 eV;
Quasi-degenerate (QD): m1
= m2
= m3
= m, m2j |m2A |, m 0.10 eV.
2.2. The see-saw mechanism and neutrino Yukawa couplings
We consider the minimal supersymmetric standard model with RH neutrinos and see-saw
mechanism of neutrino mass generation (MSSMRN). In the framework of MSSMRN one can
always choose a basis in which both the matrix of charged lepton Yukawa couplings, YE , and
the Majorana mass matrix of the heavy RH neutrinos, MN , are real and diagonal. Henceforth,
we will work in that basis and will denote by DN the corresponding diagonal RH neutrino mass
matrix, DN = diag(M1 , M2 , M3 ), with Mj > 0 and M1 < M2 < M3 . The largest mass M3 will be
standardly assumed to be of the order of, or smaller than, the GUT scale MGUT
2 1016 GeV.
Below the see-saw scale, MR = min(Mj ), the heavy RH neutrino fields Nj are integrated out,
and as a result of the electroweak symmetry breaking, the left-handed (LH) flavour neutrinos
acquire a Majorana mass term:
1 C
(m )j k Lk + h.c.,
Lm = Rj
2
C C(
where Rj
Lj )T and
ik 1 kl
MN
(m )ij = vu2 YT
(Y )lj .
(8)
(9)
Here vu = v sin , where v = 174 GeV and tan is the ratio of the vacuum expectation values
of up-type and down-type Higgs fields, and Y is the matrix of neutrino Yukawa couplings. The
213
neutrino mass matrix m is related to the light neutrino masses mj and the PMNS mixing matrix
as follows
ik kj
(m )ij = U mk U .
(10)
Using (9) and (10), we can rewrite the matching condition at the energy scale MR in the form
U D U = vu2 YT M1
N Y ,
(11)
where D = diag(m1 , m2 , m3 ). Thus, in the basis in which the RH neutrino mass matrix is diagonal MN = DN , the matrix of neutrino Yukawa couplings at MR can be parametrised as [23]
Y (MR ) =

1
DN R D U .
vu
(12)
Here R is a complex orthogonal matrix8 RT R = 1.

In what follows we will investigate the case when the RG running of mj and of the parameters
in UPMNS from MZ to MR is relatively small and can be neglected. This possibility is realised in
the class of theories under discussion for sufficiently small values of tan and/or of the lightest
neutrino mass min(mj ) [32], e.g., for tan 10 and/or min(mj ) 0.05 eV. Under the indicated
condition D and U in Eq. (12) can be taken at the scale MZ , at which the neutrino mixing
parameters are measured.
As is well-known and we shall discuss further, in the case of soft SUSY breaking mediated
by soft flavour-universal terms at MX > MR , the predicted rates of LFV processes such as
e + decay are very sensitive to the off-diagonal elements of
Y (MR )Y (MR ) =

1
U D R DN R D U ,
2
vu
(13)
while leptogenesis depends on [13] (see also [35,36] and the references quoted therein)
Y (MR )Y (MR ) =

1
DN RD R DN .
2
vu
(14)
In such a way, the matrix of neutrino Yukawa couplings Y connects in the see-saw theories
the light neutrino mass generation with leptogenesis; in SUSY theories with SUSY breaking
mediated by soft flavour-universal terms in the Lagrangian at MX > MR , Y links the light
neutrino mass generation and leptogenesis with LFV processes (see, e.g., [28,29]).
2.3. The LFV decays li lj +
As was indicated in the introduction, in the class of theories we consider, one of the effects of
RG running from MX to MR < MX is the generation of new contributions in the amplitudes of
the LFV processes [20,22]. In the mass insertion and leading-log approximations (see, e.g., [22,
24,30]), the branching ratio of li lj + decay due to the new contributions has the following
8 Eq. (12) represents the so-called orthogonal parametrisation of Y . In certain cases it is more convenient to use
diag
diag
the bi-unitary parametrisation [29] Y = UR Y UL , where UL,R are unitary matrices and Y is a real diagonal
matrix. The orthogonal parametrisation is better adapted for our analysis and we will employ it in what follows.
214
form

2
3
(3 + a02 )m20 2

(li e ) em
MX

tan2 ,
Y
ln
(Y
)
kj

ik
8
2
total (li ) G2F mS
Mk
8
k
(15)
where i = j = 1, 2, 3, l1 , l2 , l3 e, , , m0 and A0 = a0 m0 are the universal scalar masses and
trilinear scalar couplings at MX and mS represents SUSY particle mass. It was shown in [30]
that in most of the relevant soft SUSY breaking parameter space, the expression

2
m8S
0.5m20 m21/2 m20 + 0.6m21/2 ,
(16)
B(li lj + )
=
m1/2 being the universal gaugino mass at MX , gives an excellent approximation to the results
obtained in a full renormalisation group analysis, i.e., without using the leading-log and the mass
insertion approximations. It proves useful to consider also the double ratios,
B( e + )
B( e e ),
B( e + )
B( e + )
B( e e ),
R(21/32)
B( + )
R(21/31)
(17)
which are essentially independent of the SUSY parameters.

To get an estimate for the typical predictions of the schemes with heavy Majorana neutrinos
with hierarchical spectrum we will consider further, we introduce a benchmark SUSY scenario
defined by the values of the SUSY parameters
m0 = m1/2 = 250 GeV,
A0 = a0 m0 = 100 GeV,
(18)
and tan (510). In this scenario the lightest supersymmetric particle is a neutralino with
a mass of 100 GeV. The next to the lightest SUSY particles are the chargino and a second
neutralino with masses 200 GeV. The squarks have masses in the range of (400600) GeV.
Supersymmetric particles possessing the indicated masses can be observed in the experiments
under preparation at the LHC.
The benchmark values of m0 , m1/2 and A0 in Eq. (18) correspond to

2
B(li lj + )
9.1 1010 Y LY ij tan2 ,
(19)
where (L)kl = kl (L)k , Lk ln(MX /Mk ). Since tan2 will typically enhance B(li lj + )
by at least 1 order of magnitude, the quantity |(Y LY )21 | has to be relatively small to be in
agreement with the existing experimental upper limit on B( e + ). For given values of the
heavy Majorana neutrino masses, this will lead to certain constraints on the parameters in R.
Regarding the masses of the heavy Majorana neutrinos, we shall assume that M1 M2 M3 ,
with M3 having a value M3 (1013 1014 ) GeV, M3 MX . Constraints from thermal leptogenesis require that M1 109 GeV [35,36]. This would indicate a hierarchy, e.g., of the form
M1
(109 1011 ) GeV, M2
(1012 1013 ) GeV and M3
(1014 1015 ) GeV.
2.4. Leptogenesis
In the case of M1 M2 M3 , the baryon asymmetry of interest is given by
YB
102 1 ,
(20)
215
where 1 is the CP-violating asymmetry in the decay of the lightest RH Majorana neutrino N1
having the mass M1 , and is an efficiency factor calculated by solving the Boltzmann equations
(see, e.g., [35]). A simple approximate expression for the efficiency factor in the case of thermal
leptogenesis we will assume in what follows, was found in [36]:
1.16

1 3.3 103 eV
m
1
(21)
+
,
m
1
0.55 103 eV
where the neutrino mass parameter m
1 is given by
m
1

vu2
Y Y 11 .
M1
(22)
The CP-violating decay asymmetry 1 has the form

1

2 M1

1
3
Im Y Y 21
.
8 (Y Y )11
M2
(23)
Extensive numerical studies have shown [35,36] that in MSSM and for hierarchical spectrum
of masses of the heavy Majorana neutrinos under discussion, successful thermal leptogenesis is
possible only for
m
1 0.12 eV.
(24)
For typical values of (103 101 ) one gets for YB a value compatible with the observations [9],
YB = (6.15 0.25) 1010 ,
(25)
(105 107 ).
if 1
As it follows from Eqs. (13) and (15), the branching ratios of li lj + decays in the case
of interest depend on the orthogonal matrix R. Successful leptogenesis can take place only if R
is complex, so we will consider (R) = R. In what follows we will use a parametrisation of R
with complex angles (see, e.g., [23,24]):
R = R12 R13 R23
or
R = R12 R23 R12 ,
(26)
=
where Rij (R12 ) describes now the rotation with a complex angle ij = ij + iij (12

12 + i12 ), ij and ij (12 and 12 ) being real parameters. These parametrisations prove particularly convenient for investigating the case of hierarchical spectrum of masses of the heavy
RH neutrinos.
3. The see-saw mechanism, neutrino Yukawa couplings, LFV decays li lj + and

leptogenesis
There has been a considerable theoretical effort in recent years to understand possible connections between the neutrino mass and mixing data, LFV charged lepton decays and leptogenesis.
Here we shall focus on the combined constraints which the existing stringent upper limit on the
e + decay rate and the requirement of successful leptogenesis impose on MSSMRN. We
will be interested, in particular, in the possible implications of these constraints for the form of
the matrix R, the heavy Majorana neutrino masses, the predicted rates of the decays e + ,
e + and + , and the basic SUSY parameters.
216
3.1. Normal hierarchical light neutrino mass spectrum

Given the inequalities m1 m2,3 and M1 M2 M3 , we assume first that the terms
m1 and M1,2 give subleading contributions to (Y LY )ij , i = j , and we neglect them with
respect to those m2,3 and M3 , respectively, which give the dominant contribution. In this
approximation we find using R = R12 R13 R23 :

L3 M3 cos 13 cos 13
Y LY 21
2vu2

s12 sin 23
m2 ei(M )/2 m3 cos 23
m2 c12 sin 23

m3 cos 23 cos 23
s13 ei .
(27)
In deriving Eq. (27) we have set for simplicity 23 = /4 and neglected the terms m2 s13
2 in the square brackets. The corresponding expressions for (Y LY )
and m2,3 s13
31,32 are
very similar to that for (Y LY )21 . In particular, both are proportional to L3 M3 cos 13 cos 13
14
15
16
For the
values of M3 = (10 10 ) GeV and MX = 2 10 GeV, one finds that
plausible
2
2
M3 L3 mA /( 2vu ) 0.66. Barring accidental cancellations between the terms in the square
brackets in Eq. (27), we get from Eq. (19) that the predicted value of B( e + ) will
be larger at least by a factor of 103 than the existing upper
boundB( e + ). For
13
16
M3 = 10 GeV and MX = 2 10 GeV one finds M3 L3 m2 /( 2v 2 )

= 0.09, which
A
for tan2 = 25(100) and the chosen benchmark values of the SUSY parameters leads to
B( e + )
= 1.8 1010 (7.4 1010 ). The values obtained are still larger than the current
limit. This might suggest that the SUSY parameter m1/2 has a bigger value than the benchmark
scenario value we have assumed, or that m0 m1/2 500 GeV. We will pursue, however, an alternative hypothesis. We will suppose that m1/2 few 100 GeV. If indeed M3 5 1013 GeV,
M3 MX , where MX MGUT , the existing stringent experimental upper limit on B( e + )
might suggest that 13
= /2 and we will explore this interesting possibility in what follows. For
13 = /2, the R matrix has the form

R
0 sin
0 cos
1
0

cos
sin ,
0
(28)
where 12 23 . Thus, only the combination (12 23 ) of the complex angles 12 and 23
appears in the expression for R. It is not difficult to convince oneself that if m1 is sufficiently
small, and R has the form given in Eq. (28), we have (Y )3j
= 0 (j = 1, 2, 3).9 This means that
the heaviest RH Majorana neutrino N3 effectively decouples and Y has practically the form of
the matrix of neutrino Yukawa couplings in the so-called 32 see-saw model [50]. We will
keep, however, the elements (Y )1j (2j ) = 0 (j = 1, 2, 3) in our further analysis.
With m1 m2,3 , M1 M2 and R having the form (28), the terms M2 L2 give the dominant
9 It follows from Eq. (15) that the contribution of the heaviest RH Majorana neutrino N to the e + decay
3
amplitude will be suppressed if [23,25] just (Y )32 = 0 and/or (Y )31 = 0.
217
contribution in |(Y LY )ij | (i = j ). We get:

Y LY
L2 M2 ei/2 m2 c13 s12 c ei(M 2)/2 m3 s13 s
21
vu2

eiM /2 m3 c13 s23 s ei/2 m2 c12 c23 c ,
(29)
where c cos , s sin and we have neglected terms s13 which can give a correction
10
12
not exceeding approximately 13%,
and terms m1 . Setting, for instance, M2 = 10 GeV,
we find |(Y LY )21 | M2 L2 m2 /v 2
102 , and correspondingly B( e + )
= 2.3
1012 for tan2 = 25. This is the range that will be explored by the experiment MEG [19]
currently under preparation. Similarly, we obtain for (Y LY )31,32 :

Y LY
L2 M2 ei/2 m2 s12 c ei(M 2)/2 m3 s13 s
31
vu2

eiM /2 m3 c23 s + ei/2 m2 c12 s23 c ,
(30)

Y LY
L2 M2 ei/2 m2 c12 c23 c + eiM /2 m3 s23 s
32
2
vu
i /2

e M
m3 c23 s + ei/2 m2 c12 s23 c .
(31)
and
As it follows from Eqs. (29)(31), for = 0, the li lj + decay branching ratios of interest depend on the Majorana phase difference ( M ). The effective Majorana mass | m|
in ()0 -decay (see, e.g., [14,39]) depends on the same Majorana phase difference (see, e.g.,
[40,41]):

m
= m221 sin2 12 ei(M ) + m231 sin2 13 .
(32)
If s13 |s | is not negligibly small, B( e + ) and B( e + ) will depend also on the Dirac
phase .
For the double ratio R(21/31) we find from Eqs. (29) and (30):

m3 s23 s m2 c12 c23 c ei 2 M 2
R(21/31) =
(33)
.

m3 c23 s + m2 c12 s23 c ei 2 M 2

2 / m2 , R(21/31) depends only on ( ) and . If
Given 12 , 23 and m2 /m3
m
=
M
21
31
the terms m3 ( m2 ) in Eq. (33) dominate, we have R(21/31)

= 1.
The expression for the double ratio R(21/32) can be obtained from Eqs. (29) and (31). For
2 /c2 0.1, while if m |s | m |c |s ,

m3 s13 |s | m2 |c | we get R(21/32)
= s13
3
2 12
23
2 0.9.
one finds R(21/32)
= (tan2 12 )/s23
=
10 We assume that m /m
1
2,3 1.
218
3.1.1. Leptogenesis constraints

We shall analyse next the constraints on the parameter in the matrix R, Eq. (28), which
follow from the requirement of successful thermal leptogenesis.11 With 13 = /2 we find that
in the case of NH light neutrino mass spectrum we are considering
2

2
Im c2 + m21
s
2
3 m3 M1
m
1

31
8
vu2
m221
|c |2 +
|s |2
m231

sin 2 sinh 2
3 m3 M1
,
m2
2
2
8
(1 m3 ) cos 2 + (1 + m
vu
m3 ) cosh 2
(34)
(35)
m2 , m2 =
m2 and the
where and are determined by = + i and we have used m22 =
21
3
31
relation Im c2 = Im s2 . Thus, in the case under discussion, the mass M2 governs the magnitude
of the li lj + decay branching ratios, whereas M1 determines the value of the leptogenesis
decay asymmetry. The conditions |c |2 0, |s |2 0 imply that and (in Eq. (35)) should
satisfy cosh 2 | cos 2|, which is always valid since cosh 2 1. As can be easily shown, we
have
2 m2 2
21
Im c +
s
| Im c2 |
m2
(36)

1,
31
|c |2
m221
2
2
|c | +
2 |s |
m31
leading to the well-known [51] upper limit

m3
M1
174 GeV 2
3 m3 M1
7
1.97
10
| 1 |
.
8
0.05 eV
vu
vu2
109 GeV
(37)
The requirement of a non-zero asymmetry, 1 = 0, implies, as it follows from Eq. (35), that both
the real and imaginary parts of have to be non-zero, = k/2, k = 0, 1, 2, . . . , = 0, i.e., that
R has to be complex. Moreover, we should have sin 2 sinh 2 < 0 since the decay asymmetry
1 has to be negative in order to generate a baryon asymmetry of the correct sign. The maximal asymmetry | 1 | is obtained for | Im c2 | = |c |2 , which is satisfied for cos 2 cosh 2 = 1,
cos 2 = 1, cosh 2 = 1.
The neutrino mass parameter m
1 , Eq. (22), can also be easily found:
1
1
m
1
m3 |c |2 + m2 |s |2 = (m3 + m2 ) cosh 2 + (m3 m2 ) cos 2 m2 .
(38)
2
2

The minimal value of m
1 is m2
= m221
= 9 103 eV, corresponds to cosh 2 = 1 and
1 0.12 eV, where we have taken into

cos 2 = 1, for which | 1 | = 0. For 9 103 eV < m
account Eq. (24), the efficiency factor lies in the interval 1.9 103 < 3.9 102 . For this
range of values of successful leptogenesis is possible for M1 1010 GeV. We will consider
values of M1 in the interval M1 = (1010 1012 ) GeV, which is compatible with the assump11 Using the requirement of successful leptogenesis in order to constrain the high-energy parameters of the see-saw
model is not new, of course, see, e.g., [26].
(a)
219
(b)
Fig. 1. (Colour online.) The leptogenesis constraints on the parameters and for M1 = 1010 GeV; 1011 GeV;
1012 GeV (blue, green, red areas) in the case of NH light neutrino mass spectrum. The two panels correspond to two different intervals of values of the baryon asymmetry of the Universe, YB , considered: (a) 5.0 1010 YB 7.0 1010 ,
1 0.12 eV:
and (b) 3.0 1010 YB 9.0 1010 . The solid lines show the limit associated with the upper bound m
outside the region between the solid lines the wash-out effects are too strong and leptogenesis cannot produce the observed baryon asymmetry.
tion we made about the hierarchical mass spectrum of the heavy Majorana neutrinos. Thus,
for given M1 , the requirement of successful leptogenesis implies a constraint on the two parameters and of the theory. In Fig. 1 we show the leptogenesis constraint on and for
M1 = 1010 GeV; 1011 GeV; 1012 GeV. As we see from Fig. 1, the requirement of successful leptogenesis severely limits the allowed ranges of values of and . Moreover, the values
of the two parameters are strongly correlated. We note, in particular, that as | | increases, the
wash-out effects become stronger and for | | 1, the observed baryon asymmetry cannot be
reproduced. The maximal asymmetry | 1 | for a given M1 is obtained for values of cos 2 and
cosh 2 close, but not equal, to (1) and (+1), respectively. Given the interval of allowed values of m
1 , we
can write m
1 = f m3 with f
= [0.2, 2]. The condition of maximal | 1 | implies
2
1 = m3
cos 2 = (f 1 + f )m3 /(m3 m2 ). Choosing f = 1, i.e., m
= 0.05 eV, for instance,
we get cos 2 = 0.5 and correspondingly cosh 2 = 2. In this example

= 5.4 103 and for
2
tan 3 we get the requisite value of the baryon asymmetry for M1 = 6 1010 GeV.
In Fig. 2 we show the relation between the predicted values of YB in the thermal leptogenesis
scenario and of B(li lj + ). The figure was obtained for the benchmark values of the soft
SUSY breaking parameters m0 = m1/2 = 250 GeV, a0 m0 = 100 GeV and the minimal value
of tan = 5. For the relevant heavy Majorana neutrino masses we used M1 = 6 1010 GeV
and M2 = 1012 GeV. Results for two values of the Majorana phase difference ( M ) = 0; ,
are shown. For the values of and ensuring successful leptogenesis we find that typically
1014 B( e + ) 5 1013 , 1013 B( + ) 5 1012 and 1015 B(
e + ) 5 1013 . However, we have B(li lj + ) tan2 and, e.g., for tan = 20 we get
typically 1.6 1013 B( e + ) 8 1012 , which is entirely in the range of sensitivity
of the MEG experiment. As Fig. 2 indicates, the dependence of B(li lj + ) on the Majorana
phase ( M ), on which the rate of ()0 -decay depends, is relatively weak.
220
(a)
(b)
(c)
Fig. 2. (Colour online.) The correlation between the predicted values of the LFV decay branching ratios B( e + )
(a), B( e + ) (b), B( + ) (c) and of the baryon asymmetry YB in the thermal leptogenesis scenario, for
M1 = 6 1010 GeV, M2 = 1012 GeV and NH light neutrino mass spectrum. The figure was obtained for the benchmark values of the soft SUSY breaking parameters m0 = m1/2 = 250 GeV, a0 m0 = 100 GeV and the minimal
value of tan = 5. The region between the two vertical dashed lines corresponds to the observed baryon asymmetry:
5.0 1010 YB 7.0 1010 . Results for two values of the Majorana phase ( M ) equal to 0 (red + green areas)
and (green areas) are shown.
3.2. Inverted hierarchical light neutrino mass spectrum

We will perform next a similar analysis assuming that the light neutrino mass spectrum is of
the inverted hierarchical type. We set m2
= m1 and neglect terms m3 /m1,2 1. We neglect
first also terms M1 /M3 and M2 /M3 . Setting for simplicity s13 = 0 and 23 = /4, we find
that B( e + ) will depend on

L3 M3 m2
Y LY 21
c12 sin 13 + s12 ei/2 cos 13 sin 23
2

i/2
.
c12 cos 13
sin 23
s12 sin 13
e
(39)
This serves to underline thatas in the case of a NH

light neutrino spectrum discussed in
Section 3.1we have typically |(Y LY )21 | M3 m2A /vu2 . This leads for M3
(1014
1015 ) GeV and m0 , A0 and m1/2 in the few 100 GeV range to a e + decay branching
ratio which exceeds the existing limit by approximately 3 orders of magnitude. Looking again for
simplifications with interesting phenomenological consequences, we can reduce the magnitude
221
of |(Y LY )21 | by setting 13 = 23 = 0 and, correspondingly, R13 = R23 = 1 in Eq. (26).12

The corresponding form of R is

R
cos 12
sin 12
0
sin 12
cos 12
0

0
0 .
1
(40)
With R given by Eq. (40) and negligible m3 /m1,2 , the heaviest RH Majorana neutrino N3 effectively decouples and we have again (Y )3j
= 0 (j = 1, 2, 3). Neglecting further the splitting
between m1 and m2 we find:

L2 M2 |m231 |

i/2
c
sin
e
s
cos
Y LY 21

12
12
12
12
vu2
i/2

e
(41)
,
c12 c23 cos 12
+ s12 c23 + ei c12 s23 s13 sin 12

where we have used m1,2
= |m231 | and have neglected terms s13 which give a correction

not bigger than approximately 13%. Being of the order M2 |m231 |/vu2 , the expression (41) for
(Y LY )21 will lead for M2
1012 GeV and values of the soft SUSY breaking parameters in
the few100 GeV range to B( e + ) close to the existing limits. For (Y LY )31,32 we
similarly get

L2 M2 |m231 |

i/2
Y LY 31

e
s
cos
c
sin
12
12
12
12
vu2
i/2

e
(42)
,
c12 s23 cos 12
+ s12 s23 ei c12 c23 s13 sin 12

L2 M2 |m231 |

Y LY 32

c23 s12 + ei c12 s23 s13 sin 12 + ei/2 c12 c23 cos 12
2
vu
i/2

e
(43)
.
c12 s23 cos 12
+ s12 s23 ei c12 c23 s13 sin 12
We see that, as in the case of NH light neutrino mass spectrum, (Y LY )21 and (Y LY )31 are
rather similar in structure, whereas (Y LY )32 differs somewhat. The Majorana phase M does
not appear in the expressions (41)(43) because we have set m3 /m1,2 = 0. As the phase factor
including the Dirac phase appears always multiplied by the small parameter s13 , for s13 < 0.1
the branching ratios depend essentially only on the Majorana phase , which enters also into the
expression for the effective Majorana mass | m| in ()0 -decay [40,41,52]:

m
(44)
= m213 cos2 12 + ei sin2 12 .
We will give next the ratios of B( e + ) and B( e + ) (B( + )) in the case
of negligible contribution of the terms s13 :13
12 If one uses a somewhat different parametrisation of R, namely, R = R R R , the same result is achieved by
12 23 12
setting just 23 = 0.
13 For s < 0.10 the correction due to the terms in question can be shown to be smaller than approximately 15%.
13
222
R(21/31)
cot2 23 ,
i/2 c sin s cos |2 |c sin ei/2 s cos |2
12
12
12
12
12
12
12
12
2 |e
R(21/32)
s23
.
|ei/2 s12 sin 12 + c12 cos 12 |2 |s12 sin 12 + ei/2 c12 cos 12 |2
(45)
(46)
Hence, as in the case of NH light neutrino mass spectrum, R(21/31) is rather close to one,
whereas R(21/32) can have a wide range of values. Most interestingly, R(21/32) can have a
value close to two or even be as large as 10.
We can again work out possible constraints from the requirement of successful leptogenesis. Using expression (40) for the matrix R and Eq. (23) we find that the CP-violating decay
asymmetry 1 of interest has the form

Im[sin2 12 ]
3 m2 M1 m221
1

2
8
vu2
|m231 | 1 + m21 | sin |2 + | cos |2
12
12
2|m231 |

m2 M1 m221
3
sin 2 tanh 2,
(47)
16
vu2
|m231 |

where 12 = + i , m2
= |m231 | and we have neglected corrections m221 /|m231 |. We
see that in order to have 1 = 0, both and should be different from zero: = k/2, k =
0, 1, 2, . . . , = 0. Since 1 < 0, we should have sin 2 tanh 2 > 0.
It follows from Eq. (47) that in the case of IH light neutrino mass spectrum under discussion, the CP-asymmetry 1 is suppressed by the factor m221 /|m231 |. The expression for the
CP-asymmetry we have found for the NH spectrum, Eq. (35), does not contain the indicated
suppression factor. It is not difficult to show that one always has
| Im[sin2 12 ]|
1
.

m221
1+
| sin 12 |2 + | cos 12 |2 2
2
(48)
2|m31 |
For the asymmetry 1 we get the upper limit

m2
m2 M1 m221
M1
174 GeV 2
3
9
3.2 10
| 1 |
, (49)
16
0.05 eV
vu
vu2
109 GeV
|m231 |
where we have used m221 /|m231 | = 3.2 102 . The maximal value of 1 is reached for
= /4 and 0.5.
For the neutrino mass parameter m
1 , Eq. (22), we find:

m
1
m1,2 | cos 12 |2 + | sin 12 |2 = m1,2 cosh 2 m1,2 .
(50)

The minimal value of m
1 = m1,2
= 5 102 eV, corresponds to cosh 2 = 1
= |m231 |
for which | 1 | = 0. For 5 102 eV < m

1 0.1 eV, the efficiency factor lies in the interval
2.4 103 < 5.4 103 . It is not difficult to convince oneself that for a given M1 , the max1
imal value of | 1 | is reached for
= 0.5, for which m
= 7.8 102 eV and, correspondingly,
3
= 3.210 . Thus, successful leptogenesis can take place for M1 6.71012 GeV, where we
have used Eq. (25). In Fig. 3 we show the regions of values of and , favored by requirement of
successful thermal leptogenesis, for three fixed values of M1 = 7 1012 GeV; 1.5 1013 GeV;
(a)
223
(b)
Fig. 3. (Colour online.) The same as in Fig. 1, but for IH light neutrino mass spectrum and (a) M1 = 7 1012 GeV;
1.5 1013 GeV; 3 1013 GeV (blue, green, red areas), (b) M1 = 4 1012 GeV; 7.0 1012 GeV; 1.5 1013 GeV
(blue, green, red areas).
3 1013 GeV. We find that | | 0.75, | | = 0. Given the minimal value of M1 determined by
the leptogenesis constraint, we will consider further in this subsection the following hypothet13
ical heavy Majorana neutrino mass spectrum: M1 = 7.0 1012 GeV,
M2 = 4.0 10 GeV,
M3 = 2.0 1014 GeV. For M2 = 4.0 1013 GeV, we find M2 |m2 |L2 /v 2
= 0.4. At
31
the same time the upper limit on B( e + ) implies that for the benchmark values of
the soft SUSY breaking parameters we have specified earlier and tan = 5, we should have
|(Y LY )21 |2 4.8 104 . It follows from the explicit expression for |(Y LY )21 |2 , Eq. (41),
that this upper limit is impossible to satisfy for the values of the parameters and satisfying the
leptogenesis constraint (Fig. 3). This is clearly seen in Fig. 4, which shows that the requirement
of successful leptogenesis and the existing experimental upper limit on B( e + ) are incompatible in the case of the benchmark values of the SUSY parameters, m0 = m1/2 = 250 GeV,
a0 m0 = 100 GeV, and of tan = 5. The result we have obtained indicates that in the case
of IH light neutrino mass spectrum, the SUSY parameters m0 and/or m1/2 should have values
considerably larger than the benchmark values we consider. More specifically, we can have
m0 (250300) GeV, but m20 m21/2 . This possibility is illustrated in Fig. 5, where the predicted values of B(li lj + ) for m0 = 300 GeV, m1/2 = 1400 GeV, a0 m0 = 0, and tan = 5,
are shown as functions of the predicted value of the baryon asymmetry. The figure corresponds
to M1 = 7 1012 GeV and M2 = 4.0 1013 GeV. Now the requirement for successful leptogenesis is compatible with the existing constraint on B( e + ): for the values of and
ensuring successful leptogenesis we find that typically 3 1014 B( e + ) 5 1013 .
Significantly larger values of B( e + ) are possible if tan 10. As Fig. 5 also shows, the
predicted branching ratios B(li lj + ) exhibit weak dependence on the Majorana phase .
If the light neutrino mass spectrum is of the IH type, the results we have obtained in this
subsection can have important implications for the predicted spectrum of SUSY particles in the
few 100 GeV1 TeV region, to be probed by the experiments at the LHC. For m0 = 300 GeV,
m1/2 = 1400 GeV, a0 m0 = 0, and tan = 5, the lightest SUSY particle is still a neutralino and its
mass is approximately 600 GeV. The mass of next to the lightest SUSY particle, which is a stau,
224
Fig. 4. The correlation between the predicted baryon asymmetry of the Universe YB and the predicted branching ratio
of e + decay B( e + ) in the case of IH light neutrino mass spectrum and for M1 = 7.0 1012 GeV
and M2 = 4.0 1013 GeV. The figure was obtained for the benchmark values of the soft SUSY breaking parameters
m0 = m1/2 = 250 GeV, a0 m0 = 100 GeV and tan = 5. The horizontal line indicates the experimental upper limit
on B( e + ), while the region between the two vertical dashed lines is favored by the observed value of the baryon
asymmetry of the Universe, 5.0 1010 YB 7.0 1010 .
(a)
(b)
(c)
Fig. 5. (Colour online.) The correlation between the predicted YB and the predicted B( e + ) (a), B( e + )
(b), B( + ) (c), for IH light neutrino mass spectrum and M1 = 7.0 1012 GeV, M2 = 4.0 1013 GeV. The
results shown are obtained for two values of the Majorana phase = 0; and the following set of values of the SUSY
parameters: m0 = 300 GeV, m1/2 = 1400 GeV, a0 = 0 and tan = 5. The region between the two vertical dashed lines
corresponds to 5.0 1010 YB 7.0 1010 and is favored by the measured value of the baryon asymmetry. The
green (red + green) areas correspond to the Majorana phase = (0).
225
is very close to the mass of the lightest neutralino. At the same time the squarks are predicted to
be relatively heavy, having masses (23) TeV.
3.3. Quasi-degenerate light neutrinos
In this case one has m1
= m2
= m3 m, with m 0.1 eV. It is easy to see that (Y LY )21
will be proportional to M3 m/vu2 , and therefore a too large branching ratio for e + decay
will be predicted. Indeed, neglecting the terms M1 /M3 and M2 /M3 and using the complex
Euler angle parametrisation R = R12 R23 R12 , we find:

L3 mM3
s12 + ei s13 c12 sin 23

sin 12
ei 2 c12 cos 12
sin 23
Y LY 21

2
2vu
M

+ ei 2 c13 cos 23

i
c13 c12 sin 12 ei 2 s12 cos 12 sin 23 + e 2 (M 2) s13 cos 23 ,
(51)
where for simplicity we have set 23 = /4 and have neglected the sub-dominant terms s13 .
There are two possibilities for suppression of |(Y LY )21 |: (i) If sin 23 = 0, the contribution
due to M3 in (Y LY )21 remains, but is proportional to s13 . The necessary suppression can take
place if s13 is sufficiently small. (ii) The parameters 12 and 23 can have values such that the
different terms M3 in (Y LY )21 cancel (completely or partially) each other. The latter seems
to require fine tuning between the values of several very different parameters.
In what follows we shall consider the case (i) and we set 23 = 0. In this case R has the
form given in Eq. (40). The quantities of interest (Y LY )ij (i = j ), including the contributions
M2 , are given by:

L3 mM3 i
Y LY 21 =
e s23 c13 s13
vu2

L2 mM2
+
c13 c23 s12 + ei s23 c12 s13 s + ei 2 c23 c12 c
2
vu

c12 s + ei 2 s12 c ,

L3 mM3 i
e c23 c13 s13
Y LY 31 =
vu2

L2 mM2
+
c13 s23 s12 + ei c23 c12 s13 s ei 2 s23 c12 c
2
vu

c12 s + ei 2 s12 c ,

L3 mM3 2
c13 c23 s23
Y LY 32 =
vu2

L2 mM2
+
c13 s23 s12 + ei c23 c12 s13 s ei 2 s23 c12 c
vu2

(c23 s12 + ei s23 c12 s13 )s + ei 2 c23 c12 c ,
(52)
(53)
(54)
. It is interesting to note that the quantity |(Y LY ) |2 , and correwhere = 12 + 12

32
2 . The effective Majorana mass

spondingly B( + ), is not suppressed by the factor s13
226
in ()0 -decay depends in the case of QD spectrum strongly on the CP-violation Majorana phase [40,41], present in the expressions (52)(54) for (Y LY )ij (i = j ): | m|
=
m| cos2 12 + ei sin2 12 |.
For the CP-violating decay asymmetry 1 we find

Im[s2 ]
3 mM1 m221
.
1 =

2
2
8
vu
m |c |2 + 1 + m2 |s |2
2m2
(55)
It is not difficult to show that

Im(s2 )
1
Im(s2 )
1
= sin 2 tanh 2 .
2

2
2
m
2
2
|c |2 + 1 + 2m2 |s |2 |c | + |s |
Thus, the maximal asymmetry | 1 | is given by

0.1 eV
M1
174 GeV 2
| 1 | 1.6 109
.
m
vu
109 GeV
One can easily find also the mass parameter m
1:

m
1
= m |c |2 + |s |2 = m cosh 2 m.
(56)
(57)
(58)
Since successful leptogenesis is possible for [36] m

1 0.12 eV, while for QD light neutrino
mass spectrum m 0.1 eV, we get from Eq. (58) that m
= 0.1 eV. Therefore in all further
analysis and numerical calculations in this subsection we set m = 0.1 eV.
As it follows from the preceding discussion, we have m
1
= (0.100.12) eV. Correspondingly,
the wash-out effect in the case under consideration is relatively strong. Taking into account the
1 0.12 eV, we get for the corresponding effiprecise upper limit on m
1 given in Ref. [36], m
1 0.12 eV implies 0.3, for
ciency factor 1.9 103 2.4 103 . The condition m
which tanh 2 0.5. Thus, using Eqs. (20) and (57) we obtain the minimal value of M1 ensur 1 0.1 eV, we can expect that
ing successful leptogenesis: M1 3.0 1013 GeV. Since m
lies in the interval
= (0.20.3). This is confirmed by a more detailed numerical analysis. The
values of the parameters and allowed by the leptogenesis constraint are shown in Fig. 6 for
M1 = 3.0 1013 GeV; 5.0 1013 GeV.
Given the lower bound M1 3.0 1013 GeV, a possible mildly hierarchical heavy Majorana
neutrino mass spectrum would correspond to, e.g., M1 = 3.0 1013 GeV, M2 = 1.2 1014 GeV,
and M3 = 4.8 1014 GeV. For this spectrum, L3 mM3 /vu2
= 6.0 and L2 mM2 /vu2
= 2.0. Using
2
the lowest possible value for tan = 10 we find that even if s13 = 0 and the term M3 does
not contribute to |(Y LY )21 |, the contribution of the terms M2 is so large in the case of the
benchmark values of the soft SUSY breaking parameters, m0 = m1/2 = 250 GeV, a0 m0 =
100 GeV, that the predicted B( e + ) exceeds the existing upper limit on B( e + )
by more than a factor of 103 . The indicated incompatibility between the leptogenesis and
B( e + ) constraints is illustrated in Fig. 7.
Similarly to the case of IH light neutrino mass spectrum, the requirement of successful thermal
leptogenesis and the upper limit on B( e + ) can be simultaneously satisfied only if the
scale of masses of supersymmetric particles is significantly larger than that predicted for the
benchmark values of the soft SUSY breaking parameters we have adopted. In Fig. 8 we show
(a)
227
(b)
Fig. 6. (Colour online.) The same as in Fig. 1, but for QD light neutrino mass spectrum and (a) M1 = 3.0 1013 GeV;
5.0 1013 GeV (green, red areas), (b) M1 = 1.7 1013 GeV; 3.0 1013 GeV (green, red areas).
Fig. 7. The correlation between the predicted YB and the predicted B( e + ) in the case of QD light neutrino mass
spectrum and for M1 = 3.0 1013 GeV, M2 = 1.2 1014 GeV and M3 = 4.8 1014 GeV. The benchmark values
of the soft SUSY breaking parameters m0 = m1/2 = 250 GeV, a0 m0 = 100 GeV and tan = 5, have been used. The
horizontal line indicates the experimental upper limit on B( e + ), while the region between the two vertical dashed
lines corresponds to 5.0 1010 YB 7.0 1010 and is favoured by the observed value of YB .
the correlation between the predicted values of B(li lj + ) for sin 13 = 0.05 and tan = 5,
m0 = 300 GeV, m1/2 = 1400 GeV and a0 = 0, and the predicted value of the baryon asymmetry.
The results presented in this figure have been obtained for the spectrum of the heavy Majorana
neutrino masses specified above. We note, in particular, that the predicted interval of values of
B( e + ) which is compatible with the observed baryon asymmetry is in the range of
sensitivity of the ongoing MEG experiment: B( e + ) can have a value just below the
present experimental upper limit. As in the cases of NH and IH light neutrino mass spectra, we
find that the dependence of B(li lj + ) on the relevant Majorana phase is rather weak.
In Fig. 9, the correlations between the predicted value of YB and those of the double ratios R(21/31) and R(21/32) are displayed. In the approximation we use, the double ratios
228
(a)
(b)
(c)
Fig. 8. The correlation between the predicted YB and the predicted B( e + ) (a), B( e + ) (b), B( + )
(c), for QD light neutrino mass spectrum and M1 = 3.0 1013 GeV, M2 = 1.2 1014 GeV, M2 = 4.8 1014 GeV.
The SUSY parameters used to obtain the figure are m0 = 300 GeV, m1/2 = 1400 GeV, a0 = 0 and tan = 5. Results
for two values of the Majorana phase = 0; are shown. The region between the two vertical dashed lines corresponds
to 5.0 1010 YB 7.0 1010 .
are independent of SUSY parameters and are determined only by the off-diagonal elements
of Y Y . When the constraint of successful leptogenesis is imposed, we get for the allowed
range of values of R(21/31) for the NH, IH and QD light neutrino mass spectrum respectively
10 R(21/31) 100, 20 R(21/31) 50 and R(21/31)
20; 100. Similarly, for the double ratio R(21/32) we get in the three cases 102 R(21/32) 10, 10 R(21/32) 103 and
50 R(21/32) 103 , respectively. We find, in particular, that R(21/32) can be much smaller
than 1 only for NH light neutrino mass spectrum.
4. Conclusions
We have considered the LFV decays e + , e + and + and leptogenesis
in the MSSM with see-saw mechanism of neutrino mass generation and soft SUSY breaking
with universal boundary conditions at a scale MX > MR , MR being the heavy RH Majorana
neutrino mass scale. The heavy Majorana neutrinos were assumed to have hierarchical mass
spectrum, M1 M2 M3 , while the scale MX was taken to be the GUT scale, MX = 2
1016 GeV. We have analysed the combined constraints, which the existing stringent upper limit
on the e + decay rate and the requirement of successful leptogenesis impose on the
neutrino Yukawa couplings, heavy Majorana neutrino masses and SUSY parameters in the cases
of the three types of light neutrino mass spectrumnormal and inverted hierarchical (NH and
229
Fig. 9. The correlation between the predicted YB and the double ratios R(21/31) and R(21/32), defined in Eq. (17). The
heavy neutrino mass spectrum in the NH, IH, and QD cases is taken to be same as in Figs. 2, 5 and 8, respectively. The
region between the two vertical dashed lines corresponds to 5.0 1010 YB 7.0 1010 .
230
IH), and quasi-degenerate (QD). A basic quantity in these analyses is the matrix of neutrino
Yukawa couplings, Y . In the present work we have used the orthogonal parametrisation of Y ,
in which Y is expressed in terms of the light neutrino and heavy RH neutrino masses, the PMNS
neutrino mixing matrix UPMNS , and an orthogonal matrix R. Leptogenesis can take place only if
R is complex.
The constraints from thermal leptogenesis require, in general, that M1 109 GeV. This
would indicate a hierarchy, e.g., of the form M1
(109 1011 ) GeV, M2
(1012 1013 ) GeV
and M3 1013 GeV M2 , M3 < ()MX . In our analysis we have considered a benchmark
SUSY scenario defined by the values of the soft SUSY breaking parameters in the range of
few 100 GeV: m0 = m1/2 = 250 GeV, a0 m0 = 100 GeV, and tan (510). In this scenario the lightest supersymmetric particle is a neutralino with a mass of 100 GeV. The next
to the lightest SUSY particles are the chargino and a second neutralino with masses 200 GeV.
The squarks have masses in the range of (400600) GeV. For the indicated set of benchmark values of the soft SUSY breaking parameters, the typical values of the heaviest Majorana
neutrino mass M3
= (5 1013 1015 ) GeV, and barring accidental cancellations, the predicted
e + decay branching ratio B( e + ) exceeds the existing upper limit by few orders of
magnitude independently of the type of light neutrino mass spectrum. The requisite suppression
of the contribution of the heaviest RH Majorana neutrino N3 to the e + decay amplitude
can take place, as it is well-known [23,25], if (Y )32 = 0 and/or (Y )31 = 0.
For each of the three types of neutrino mass spectrumNH, IH and QD, we find simple forms
of the matrix R which lead to a suppression of the dominant contributions due to the terms M3
in B( e + ). In all three cases the matrix R ensuring the requisite suppression admits a
parametrisation by one complex angle. In the case of NH spectrum R is given by Eq. (28), while
for IH and QD spectra, R
= R12 , R12 being the matrix of (complex) rotations in the 12 plane. In
this case the dominant contribution in B( e + ) comes from terms M2 . For QD spectrum
the terms M3 are suppressed by the factor sin 13 and can be comparable to those M2 .
The requirement of successful leptogenesis leads to a rather stringent constraint on the complex mixing angle in R. For IH and QD spectra it also implies a relatively large lower limit on the
mass of the lightest RH Majorana neutrino: M1 7.0 1012 GeV and M1 3.0 1013 GeV,
respectively. With such values of M1 and hierarchical heavy Majorana neutrino mass spectrum,
the upper bound on B( e + ) can be satisfied only if the scale of masses of SUSY particles
is considerably higher than that implied by the benchmark values of the soft SUSY breaking parameters we have considered. We have analysed a specific case of such SUSY scenario:
m0 = 300 GeV, m1/2 = 1400 GeV and a0 = 0. In this scenario the lightest SUSY particle is a
neutralino with a mass of approximately 600 GeV, the next to the lightest SUSY particle is a
stau and its mass is very close to the mass of the lightest neutralino, while the squarks are relatively heavy, having masses (23) TeV. The predictions for B( e + ) are now largely
in the range of sensitivity of the ongoing MEG experiment. If more stringent upper limits on
B( e + ) will be obtained in the future, it would be rather difficult to reconcile the IH
and QD light neutrino mass spectra with the e + and leptogenesis constraints and SUSY
particle masses in the TeV range. Our results may have important implications for the search of
SUSY particles in the few 100 GeV1 TeV region, to be performed by the experiments at the
LHC.
Satisfying the combined constraints from the existing upper limit on the e + decay rate
and the requirement of successful thermal leptogenesis proves to be a powerful tool to test the
viability of supersymmetric theories with see-saw mechanism of neutrino mass generation and
231
soft flavour-universal SUSY breaking at a scale above the heavy RH Majorana neutrino mass
scale.
Acknowledgements
We would like to thank P. Di Bari and M. Raidal for useful correspondence. This work was
supported in part by the Italian MIUR and INFN under the programs Fisica Astroparticellare
(S.T.P. and T.S.). The work of W.R. was supported by the Deutsche Forschungsgemeinschaft
in the Sonderforschungsbereich 375 fr Astroteilchenphysik and under project number RO2516/3-1.
References
[1] B.T. Cleveland, et al., Astrophys. J. 496 (1998) 505;
Y. Fukuda, et al., Kamiokande Collaboration, Phys. Rev. Lett. 77 (1996) 1683;
J.N. Abdurashitov, et al., SAGE Collaboration, J. Exp. Theor. Phys. 95 (2002) 181;
T. Kirsten, et al., GALLEX, GNO Collaborations, Nucl. Phys. B (Proc. Suppl.) 118 (2003) 33;
C. Cattadori, et al., Nucl. Phys. B (Proc. Suppl.) 143 (2005) 3.
[2] S. Fukuda, et al., Super-Kamiokande Collaboration, Phys. Lett. B 539 (2002) 179;
Y. Ashie, et al., Phys. Rev. Lett. 93 (2004) 101801;
Y. Ashie, et al., Phys. Rev. D 71 (2005) 112005.
[3] Q.R. Ahmad, et al., SNO Collaboration, Phys. Rev. Lett. 87 (2001) 071301;
Q.R. Ahmad, et al., SNO Collaboration, Phys. Rev. Lett. 89 (2002) 011301;
Q.R. Ahmad, et al., SNO Collaboration, Phys. Rev. Lett. 89 (2002) 011302;
S.N. Ahmed, et al., Phys. Rev. Lett. 92 (2004) 181301;
B. Aharmim, et al., nucl-ex/0502021.
[4] K. Eguchi, et al., KamLAND Collaboration, Phys. Rev. Lett. 90 (2003) 021802;
T. Araki, et al., hep-ex/0406035.
[5] E. Aliu, et al., K2K Collaboration, hep-ex/0411038.
[6] S.T. Petcov, Nucl. Phys. B (Proc. Suppl.) 143 (2005) 159.
[7] B. Pontecorvo, Zh. Eksp. Teor. Fiz. 33 (1957) 549;
B. Pontecorvo, Zh. Eksp. Teor. Fiz. 34 (1958) 247;
B. Pontecorvo, Zh. Eksp. Teor. Fiz. 53 (1967) 1717;
Z. Maki, M. Nakagawa, S. Sakata, Prog. Theor. Phys. 28 (1962) 870.
[8] V. Lobashev, et al., Nucl. Phys. A 719 (2003) 153c;
K. Eitel, et al., Nucl. Phys. B (Proc. Suppl.) 143 (2005) 197.
[9] D.N. Spergel, et al., WMAP Collaboration, Astrophys. J. Suppl. 148 (2003) 175.
[10] S. Hannestad, astro-ph/0303076;
O. Elgaroy, O. Lahav, astro-ph/0303089.
[11] P. Minkowski, Phys. Lett. B 67 (1977) 421;
M. Gell-Mann, P. Ramond, R. Slansky, in: F. van Nieuwenhuizen, D. Friedman (Eds.), Supergravity, North-Holland,
Amsterdam, 1979, p. 315;
T. Yanagida, in: O. Sawada, A. Sugamoto (Eds.), Proceeding of the Workshop on Unified Theories and the Baryon
Number of the Universe, KEK, Japan, 1979;
R.N. Mohapatra, G. Senjanovic, Phys. Rev. Lett. 44 (1980) 912.
[12] B. Pontecorvo, Zh. Eksp. Teor. Fiz. 53 (1967) 1717;
S.M. Bilenky, B. Pontecorvo, Lett. Nuovo Cimento 17 (1976) 569.
[13] M. Fukugita, T. Yanagida, Phys. Lett. B 174 (1986) 45.
[14] S.M. Bilenky, S.T. Petcov, Rev. Mod. Phys. 59 (1987) 671.
[15] M.L. Brooks, et al., MEGA Collaboration, Phys. Rev. Lett. 83 (1999) 1521.
[16] S. Eidelman, et al., Particle Data Group, Phys. Lett. B 592 (2004) 1.
[17] B. Aubert, et al., BaBar Collaboration, hep-ex/0502032;
B. Aubert, et al., BaBar Collaboration, hep-ex/0508012;
K. Hayasaka, et al., Belle Collaboration, Phys. Lett. B 613 (2005) 20.
232
[18] Y. Kuno, Y. Okada, Rev. Mod. Phys. 73 (2001) 151;

A.G. Akeroyd, et al., SuperKEKB Physics Working Group, hep-ex/0406071.
[19] L.M. Barkov, et al., MEG proposal, http://meg.psi.ch.
[20] F. Borzumati, A. Masiero, Phys. Rev. Lett. 57 (1986) 961.
[21] R. Barbieri, S. Ferrara, C. Savoy, Phys. Lett. B 119 (1982) 343;
L. Hall, J. Lykken, S. Weinberg, Phys. Rev. D 27 (1983) 2359.
[22] J. Hisano, et al., Phys. Lett. B 357 (1995) 579;
J. Hisano, et al., Phys. Rev. D 53 (1996) 2442;
J. Hisano, D. Nomura, Phys. Rev. D 59 (1999) 116005.
[23] J.A. Casas, A. Ibarra, Nucl. Phys. B 618 (2001) 171.
[24] J. Ellis, et al., Nucl. Phys. B 621 (2002) 208.
[25] J. Ellis, et al., Phys. Rev. D 66 (2002) 115013.
[26] J. Ellis, M. Raidal, Nucl. Phys. B 643 (2002) 229.
[27] S. Lavignac, I. Masina, C.A. Savoy, Phys. Lett. B 520 (2001) 269;
A. Kageyama, et al., Phys. Rev. D 65 (2002) 096010;
A. Kageyama, et al., Phys. Lett. B 527 (2002) 206;
F. Deppisch, et al., Eur. Phys. J. C 28 (2003) 365;
X.-J. Bi, Eur. Phys. J. C 27 (2003) 399;
T. Blazek, S.F. King, Nucl. Phys. B 662 (2003) 359;
B. Dutta, R.N. Mohapatra, Phys. Rev. D 68 (2003) 056006;
J.I. Illana, M. Masip, Eur. Phys. J. C 35 (2004) 365;
A. Masiero, S.K. Vempati, O. Vives, New J. Phys. 6 (2004) 202;
M. Bando, et al., hep-ph/0405071;
I. Masina, C.A. Savoy, Phys. Rev. D 71 (2005) 093003;
K.S. Babu, J.C. Pati, P. Rastogi, hep-ph/0502152;
P. Paradisi, hep-ph/0505046.
[28] S. Pascoli, S.T. Petcov, C.E. Yaguna, Phys. Lett. B 564 (2003) 241.
[29] S. Pascoli, S.T. Petcov, W. Rodejohann, Phys. Rev. D 68 (2003) 093007;
W. Rodejohann, Eur. Phys. J. C 32 (2004) 235.
[30] S.T. Petcov, et al., Nucl. Phys. B 676 (2004) 453.
[31] S. Kanemura, et al., hep-ph/0501228;
S. Kanemura, et al., hep-ph/0507264.
[32] S.T. Petcov, T. Shindou, Y. Takanishi, Nucl. Phys. B 738 (2006) 219.
[33] S.T. Petcov, Sov. J. Nucl. Phys. 25 (1977) 340.
[34] S.M. Bilenky, S.T. Petcov, B. Pontecorvo, Phys. Lett. B 67 (1977) 309;
T.P. Cheng, L.-F. Li, Phys. Rev. Lett. 45 (1980) 1908.
[35] W. Buchmller, P. Di Bari, M. Plmacher, New J. Phys. 6 (2004) 105.
[36] G.F. Giudice, et al., Nucl. Phys. B 685 (2004) 89.
[37] S.M. Bilenky, J. Hosek, S.T. Petcov, Phys. Lett. B 94 (1980) 495.
[38] P. Langacker, et al., Nucl. Phys. B 282 (1987) 589.
[39] C. Aalseth, et al., hep-ph/0412300.
[40] S.M. Bilenky, S. Pascoli, S.T. Petcov, Phys. Rev. D 64 (2001) 053010.
[41] S.T. Petcov, New J. Phys. 6 (2004) 109;
S.T. Petcov, Talk given at the Nobel Symposium 129 on Neutrino Physics, Haga Slott, Enkping, Sweden, 1924
August, 2004, hep-ph/0504166;
S. Pascoli, S.T. Petcov, L. Wolfenstein, Phys. Lett. B 524 (2002) 319;
S. Pascoli, S.T. Petcov, T. Schwetz, hep-ph/0505226;
S. Choubey, W. Rodejohann, hep-ph/0506102.
[42] S. Antusch, et al., Nucl. Phys. B 674 (2003) 401.
[43] Y. Nomura, T. Yanagida, Phys. Rev. D 59 (1999) 017303;
M. Bando, T. Kugo, Prog. Theor. Phys. 101 (1999) 1313;
K.S. Babu, J.C. Pati, F. Wilczek, Nucl. Phys. B 566 (2000) 33;
W. Buchmuller, D. Wyler, Phys. Lett. B 521 (2001) 291;
S.F. King, G.G. Ross, Phys. Lett. B 574 (2003) 239;
B. Stech, Z. Tavartkiladze, Phys. Rev. D 70 (2004) 035002;
K.S. Babu, T. Enkhbat, I. Gogoladze, Nucl. Phys. B 678 (2004) 233;
[44]
[45]
[46]
[47]
[48]
[49]
[50]
[51]
[52]
M. Bando, et al., hep-ph/0405071;

R. Dermisek, S. Raby, hep-ph/0507045;
C.H. Albright, Phys. Rev. D 72 (2005) 013001;
See also the review articles: R.N. Mohapatra, et al., hep-ph/0510213;
G. Altarelli, F. Feruglio, New J. Phys. 6 (2004) 106.
A. Airapetian, et al., ATLAS Collaboration, Report CERN-LHCC-99-15;
S. Abdullin, et al., CMS Collaboration, J. Phys. G 28 (2002) 469.
M. Doi, et al., Phys. Lett. B 102 (1981) 323;
J. Schechter, J.W.F. Valle, Phys. Rev. D 22 (1980) 2227;
J. Bernabeu, P. Pascual, Nucl. Phys. B 228 (1983) 21.
M. Apollonio, et al., Phys. Lett. B 466 (1999) 415;
F. Boehm, et al., Phys. Rev. Lett. 84 (2000) 3764.
A. Bandyopadhyay, et al., Phys. Lett. B 608 (2005) 115;
A. Bandyopadhyay, et al., 2005, unpublished;
See also: A. Bandyopadhyay, et al., Phys. Lett. B 583 (2004) 134.
J.N. Bahcall, M.C. Gonzalez-Garcia, C. Pea-Garay, JHEP 0408 (2004) 016.
S. Pascoli, S.T. Petcov, Phys. Lett. B 544 (2002) 239;
S. Pascoli, S.T. Petcov, Phys. Lett. B 580 (2004) 280.
P.H. Frampton, S.L. Glashow, T. Yanagida, Phys. Lett. B 548 (2002) 119.
S. Davidson, A. Ibarra, Phys. Lett. B 535 (2002) 25.
S.M. Bilenky, et al., Phys. Rev. D 56 (1996) 4432.
233
From 3-geometry transition amplitudes to graviton

states
Federico Mattei, Carlo Rovelli, Simone Speziale , Massimo Testa
Dipartimento di Fisica dellUniversit La Sapienza, and INFN Sez. Roma1, I-00185 Roma, Italy
Centre de Physique Thorique de Luminy, Universit de la Mditerrane, F-13288 Marseille, Italy
Received 29 August 2005; received in revised form 16 January 2006; accepted 19 January 2006
Abstract
In various background independent approaches, quantum gravity is defined in terms of a field propagation
kernel: a sum over paths interpreted as a transition amplitude between 3-geometries, expected to project
quantum states of the geometry on the solutions of the WheelerdeWitt equation. We study the relation
between this formalism and conventional quantum field theory methods. We consider the propagation kernel
of 4d Lorentzian general relativity in the temporal gauge, defined by a conventional formal Feynman path
integral, gauge fixed la FaddeevPopov. If space is compact, this turns out to depend only on the initial
and final 3-geometries, while in the asymptotically flat case it depends also on the asymptotic proper time.
We compute the explicit form of this kernel at first order around flat space, and show that it projects on the
solutions of all quantum constraints, including the WheelerDeWitt equation, and yields the correct vacuum
and n-graviton states. We also illustrate how the Newtonian interaction is coded into the propagation kernel,
a key open issue in the spinfoam approach.
1. Introduction
The tentative quantum theories of gravity that are currently better developed, such as for
instance strings and loops, are very different from one another in their assumptions and in the
formalism utilized. In particular, the relation between the background-independent methods used
in canonical quantum gravity and in the spinfoam formalism, and conventional perturbative quan* Corresponding author.
E-mail addresses: federico.mattei@roma1.infn.it (F. Mattei), rovelli@cpt.univ-mrs.fr (C. Rovelli),

simone.speziale@roma1.infn.it (S. Speziale), massimo.testa@roma1.infn.it (M. Testa).
doi:10.1016/j.nuclphysb.2006.01.026
F. Mattei et al. / Nuclear Physics B 739 (2006) 234253
235
tum field theory (QFT), is far from transparent [1]. Besides clouding the communication between
research communities, these differences hinder the clarification of a number of technical and conceptual problems. For instance, there are open questions concerning the physical interpretation
of the spinfoam formalism [2], in particular concerning its low-energy limit, the comparison
with quantities computed in the standard QFT perturbative expansion, and the derivation of the
Newtonian interaction.
Here we contribute to the effort of bridging between different languages, by studying the field
propagator, or Feynman propagation kernel, or Schrdinger functional, in 4d Lorentzian general
relativity (GR). The field propagator (and its relativistic extension [1]) is not often utilized in
conventional QFT (but see [3,4]), but it is a central object in background-independent quantum
gravity [58]. Here we analyze this object starting from a conventional path integral quantization
of GR andneglecting its non-renormalizabilityin a perturbative expansion of this integral.
On the one hand, this provides a clean interpretation of a basic non-perturbative tool in terms of
conventional and well understood quantum field theoretical quantities. On the other hand, this
provides the precise expression of the low-energy limit of the field propagator, to which the nonperturbative one must be compared. In other words, we study the explicit relation between the
3-geometry to 3-geometry transition amplitude and the graviton-state language.
The key to bridge between a conventional path integral formulation and the non-perturbative
framework is the use of the temporal gauge, with a careful implementation of the FaddeevPopov
(FP) gauge-fixing. Conventional perturbation theory is usually studied in covariant gauges such
as the Lorentz gauge in YangMills (YM) theory or the harmonic gauge in GR [10], but the
temporal gauge is naturally closer to the Hamiltonian formalism. The propagation kernel of YM
theory in the temporal gauge is well understood [11]. It can be computed as a Feynman path
integral for finite time, with fixed initial and final field configurations. In the temporal gauge
there are no dynamical ghosts, and the kernel has a clear physical interpretation: it gives the
matrix element of the evolution operator between eigenstates of the YM connection. Namely, the
transition amplitude between the states defined on the boundaries. But it is also a projector on the
physical states, which satisfy the YM quantum constraints. At the zeroth order in perturbation
theory, the kernel can be explicitly computed with a Gaussian integration in the Euclidean regime,
and it nicely codes the form of the perturbative vacuum as well as all the n-particle states [1113].
Furthermore, it is straightforward to express the n-point functions in terms of it. In spite of key
differences, the structure of GR is similar to a YM theory in many respects. We can apply to GR
the techniques used for YM theory, and, in particular, study the propagation kernel of quantum
GR in the temporal gauge. This is what we do in this paper. Notice that the use of non-covariant
gauges has been considered in the literature (see for instance [14]).
First, we consider the formal path integral that defines the propagation kernel in the temporal gauge, and study its properties. This object has been considered in the literature (see [8,9]
and references therein), though with different techniques. The expression is formal because the
measure is unknown, and the usual perturbative definition is not viable because the perturbative
expansion on a background is non-renormalizable. We do not consider background-independent
definitions of the path integral, such as the spinfoam one, because our interest here is the interpretation and the low energy limit of the kernel, not its ultraviolet divergences. We consider
the two possibilities of compact and asymptotically-flat space [15]. We carefully discuss the FP
gauge fixing and in this context we identify the integration implementing the quantum constraints
as the one on the FP gauge parameters. We discuss how the propagation kernel turns out to be
independent from the coordinate time in both casesa feature that drastically distinguishes GR
from YM theory, but to depend on the asymptotic proper-time in the second case. We use
236
the conventional metric formalism for GR. The expression we obtain is formal, and we do not
discuss topological and ultraviolet aspects.
Second, we consider the zeroth order term in the perturbative expansion of the integral on a
flat background, and we compute the propagation kernel explicitly. We show that it projects on
the solutions of all the constraints, and that it correctly codes the perturbative GR vacuum state
[16,17] and the n-particle states. This provides an explicit bridge between 3-geometry transition
amplitudes and perturbative graviton states. The non-perturbative boundary amplitudes of the
spinfoam formalism must reduce to this expression for boundary metrics close to flat space.
Finally, we couple an external matter source to the theory, and derive the expression for the energy of the field in the presence of matter. This expression codes the Newtonian interaction. Our
hope is that this could open the way for extracting the Newtonian interaction from the spinfoam
amplitudes, hence providing a key missing check of their physical viability.
We fix the speed of light and the Newton and Planck constants by c = 16G = h = 1. Greek
indices range from 0 to 3; Latin indices from 1 to 3. We use coordinates x = (t, x). A (x) is the
YM field, Ai (
x ) the vector potential; g (x) is the gravitational field and the spacetime metric;
while gij (
x ) is the metric of a spacelike surface. (Indices are intended in Gerochs abstract index
and so on.)
notation: Ai means A,
1.1. YM propagation kernel in temporal gauge
Before turning to GR, we recall the properties of the field propagation kernel in YM theory. As
we shall see, the gravitational case will present substantial analogies, as well as key differences.
Formally, the propagation kernel is given by the functional integral over the configurations of
the YM field defined on the spacetime region bounded by the initial and final surfaces t = 0 and
t = T , and restricted to given initial and final configurations A (
x , 0) = A (
x ) and A (
x, T ) =

A (
x)

DA eiS[A ] ,
W [A , A , T ] =
(1)
A
A
where
T
S[A ] =
dt
d 3 x LYM
(2)
and LYM is the YM Lagrangian. The integral (1) contains an infinity due to the integration over
the group of the gauge transformations A A , where (
x , t) is the gauge parameter. We fix
this by gauge fixing (1) la FP in the temporal gauge A0 = 0 [11]. That is, we insert in (1) the
identity

1 = FP (A ) D A
(3)
0 .
The FP determinant FP is actually a constant, because the YM gauge transformations do not
mix the different components of the 4-vector A , hence FP can only depend on A0 ; but this,
in turn, is fixed to be zero by the -function appearing in the integral. The propagation kernel
237
between boundary values of the 3d YM connection Ai is thus given by

iS[A ]
.
W [Ai , Ai , T ] = FP DA D A
0 e
(4)
Ai
Ai
Changing the order of the two integrations, changing variables A A

and integrating over
A0 , we obtain

DAi eiS[Ai ;A0 =0] .
W [Ai , Ai , T ] = FP D
(5)
(T )
Ai
(0)
Ai
The only components of the integration variable entering the integrand are its values at t =
0 and t = T . We can therefore drop the bulk integration on (
x , t), 0 > t > T , discarding a
trivial infinity. Furthermore, since DAi and S[Ai , A0 = 0] are invariant under time independent
x , 0) and (
x , T ) is
gauge transformations, the DAi one of the two remaining integrals on (
redundant. Dropping the second, we have therefore

W [Ai , Ai , T ] = D W A
(6)
i , Ai , T ,
where
W [Ai , Ai , T ] = FP
DAi eiS[Ai ,A0 =0] ,
(7)
Ai
Ai
and (
x ) is the gauge parameter of the residual time-independent gauge transformations Ai
Ai of the A0 = 0 gauge. The propagator W [Ai , Ai , T ] is invariant under simultaneous gauge
transformations on the two boundaries. The integral over in (6) makes W [Ai , Ai , T ] invariant
under independent gauge transformations on the two boundaries.
The field propagator is the matrix element of the evolution operator between eigenstates of
the field operator Ai
W [Ai , Ai , T ] = Ai |eiH T |Ai ,
(8)
where H is the Hamiltonian; up to the difficulties in defining fixed-time operators in an interacting QFT (see [3]), it can be interpreted as the Feynman probability amplitude of having the
field configuration Ai at time T , given the field configuration Ai at time 0. Equivalently, it timepropagates the quantum state in the Schrdinger functional representation of the quantum field
theory

t+T [Ai ] = DAi W [Ai , Ai , T ]t [Ai ].
(9)
However, notice that, because of its gauge-invariance, any state obtained by propagating with the
kernel is invariant under gauge transformations of Ai . Therefore the kernel is also a projector on
the gauge-invariant, or physical, states, which satisfy [Ai ] = [Ai ]. That is, the states that
satisfy the Gauss-law quantum constraint. Thus, the DAi integral (7) takes care of the gaugevariant dynamics, and the D integral (6) imposes gauge invariance.
238
If n (Ai ) is a basis of gauge-invariant physical states that diagonalize the energy, (8) implies

W [Ai , Ai , T ] =
(10)
eiEn T n [Ai ]n [Ai ].
n
In particular, once subtracted the zero-point energy, we can read out the form of the vacuum state
from the propagator
+
dT W [Ai , Ai = 0, T ] = const 0 [Ai ].
(11)
Thus, the temporal-gauge propagation kernel nicely bridges between the functional integral formalism and the Hamiltonian one.
The propagation kernel W [Ai , Ai , T ] can be computed explicitly order by order in perturbation theory, by a Gaussian integration in the Euclidean regime. For this to be well defined, we
need to assume appropriate boundary conditions for the field at spacial infinity. In particular, we
assume that the field vanishes at spacial infinity, and therefore so has to do the gauge parameter.
In the lowest order (or exactly in the Maxwell case) the propagator kernel turns out to be [11]

i
d 3 p (|AT |2 + |AT |2 ) cos pT 2AT AT

. (12)
W [Ai , Ai , T ] = N (T ) exp
p
2
sin pT
(2)3

x A (
Here p = |p| = pi p i , the Fourier transform of the potential is Aj (p)
= d 3 x ei p
j x ) and
T
its transverse component is defined as Ai (p)
Dij (p)A
j (p),
where
Dij = ij
pi pj
.
p2
(13)
We can read out from this kernel all the free n-particle states with momenta p , energy E = p0
and polarizations , = 0, . . . , n, from the expression
W [Ai , Ai , T ] =
3
1
d 3 pn i n E T
d p1
=1
e
3
3
n!
(2)
(2)

n=0
p1 1 ,...,pn n [Ai ]p1 1 ,...,pn n [Ai ].
(14)
Let us now turn to general relativity.

2. Propagation kernel in general relativity
We foliate spacetime with a family of 3d surfaces t , with t R, and focus on the region
t [0, T ]. We fix initial and final positive definite metrics gij on 0 and gij on T as boundary
data, and we want to describe the quantum dynamics of the gravitational field in terms of the
functional integral

Dg eiS[g ]
W [gij , gij , T ] =
(15)

gij

gij
239
over the 4d spacetime metrics inducing the given 3d metrics on the two boundaries. The action
of the gravitational field is
T
S[g ] =
dt
0
d x gg R +
0 T
T
d xK
3
d 3 x L.
dt
0
(16)
Here R is the Ricci tensor and g the determinant of the metric. K is the extrinsic curvature
of the boundary surfaces. The presence of the boundary term, sometimes called the Gibbons
Hawking term, is needed in order to have only first-order time derivatives in the action L, so that
the convolution property of the propagation kernel is guaranteed [18].
There are various sources of infinities in (15). First, there are ultraviolet divergences. As discussed in the introduction, we disregard them here, under the assumption that an appropriate
non-perturbative definition of the integral, such as in the spinfoam formalism, could take care of
them.
Second, we must be sure that a sufficient number of boundary conditions are fixed. In general,
we may reasonably demand that a classical solution is uniquely selected by these conditions by
minimizing the action. We focus on two cases: (i) The compact case, in which the Riemaniann
manifolds (0 , gij ) and (T , gij ) have finite volume and no boundary. In this case, initial and
final 3-metrics may determine a unique classical solution. The thin sandwich conjecture [6,19],
indeed, states that, at least for small times, generically there is only one spacetime metric between
two given spacelike metrics. Exceptions are known, generally characterized by pathologies such
as singularities [20]. In the following we restrict to the cases where the conjecture holds. Amongst
the exceptions is the remarkable case of flat space, as emphasized in [21]. (ii) The asymptotically
flat case, in which is homeomorphic to R3 and gij and gij appropriately converge to ij at
infinity. In this case, for (15) to be well defined, we demand g to converge asymptotically to
the Minkowski metric for all t. The consequences for the uniqueness of the solution will be
discussed in details below.1
Third, the invariance under diffeomorphisms of GR makes the integral (15) infinite. This
situation is analogous to the YM case, and can be cured with a gauge-fixing, as we do below.
2.1. Gauge-fixing
General relativity is invariant under diffeomorphisms, namely under the pull back g g =
g of the gravitational field by a map : M M from the spacetime to itself. These are the
gauge transformations of GR.2 Explicitly, g is defined by
g (x) =

(x) (x)
g (x) ,
x
x
(17)
1 , a form we will use later

where : x (x). (Or g (x) =
x x g ((x)) where =
on.) For the action (16) to be invariant, must not change the boundaries of the spacetime region
1 It has been recently suggested that the most interesting case for nonperturbative quantum gravity is when has
t
boundaries [22]. For a discussion of the classical solutions in this case, see [23].
2 They should not be confused with the freedom of choosing coordinates on M: once coordinates are fixed, the integral (15) still gets contributions from distinct fields g and g .
240
considered, that is
0 (0, x) = 0,
0 (T , x) = T .
(18)
Because of this gauge invariance, the integral (15) has an infinite contribution from the integration
over the gauge group. We take care of this as we did for YM, by introducing a (non-covariant)
gauge-fixing. The GR analogue of the temporal gauge is known as Gaussian normal coordinates, or the Lapse = 1, Shift = 0, or proper-time gauge
g00 = 1,
g0i = 0.
(19)
As for YM, this is not a complete gauge-fixing, but we expect that additional gauge-fixing is not
required in the path integral. In the linearized case we shall explicitly see that the remaining part
of the gauge is taken care by the integration over the gauge parameters. Thus, we gauge-fix the
path integral by inserting in (15) the FP identity

1 = FP [g ] D g00 + 1 g0i .
(20)
D is a formal measure over the group of the 4d diffeomorphisms. We can see here a first difference with YM theory: the GR gauge transformations mix the different components of g , hence
the FP factor FP is not a constant anymore. Since the four g0 are fixed by the -functions, it
will depend only on the spacial components of the metric.
The integration (20) is over all (
x , t) with t [0, T ], including the boundaries, and it is not
restricted by (18), therefore it includes that change the action (16). To understand this delicate
point, observe that the FP integral must include sufficient gauge transformations for transforming
any field to one satisfying (19).We cannot fix g00 = 1, and also the coordinate time between
initial and final surface. This would amount to discard all four-metrics yielding a proper time
between the two surfaces different from T ; but these field configurations do contribute to (15)
and cannot be discarded. In other words, we cannot gauge transform all fields contributing to
the integral to the gauge (19) without changing the action. However, nothing prevents us from
transforming them to fields satisfying (19) and changing the action when needed. (See also [8]
on this.)
Introducing (20) in (15) we obtain

Dg FP [gij ]

gij

gij

T

3
D exp i d x dt L[g ] g00 + 1 g0i .
0
(21)
We now evaluate the integrals over g0 . As we did in the YM case, we exchange the order of
the integrations and change variables g g , together with a change of coordinates x

(x) in the action. As a consequence, the boundary data will now depend on . We perform the
integrals over g00 and g0i , obtaining

W [gij , gij , T ] = D W [gij , gij , , T ],
(22)
241
where
W [gij , gij , , T ] =

0(
x ,T )
3
Dgij FP [gij ] exp i d x
dt L[gij , g00 = 1, g0i = 0] .

gij
0 (
x ,0)

gij
(23)
Let us analyze these expressions. The Dgij integral in (23) is over the fields in the space
x , 0) and
time region bounded by the surfaces 0 and T , defined respectively by t = 0 (

0
t = (
x , T ), having boundary value gij on 0 and gij on T . Notice that gij depends
precisely on gij and , since, using (19), the transformation of gij reads
gij (x) =

0 0 k l

(x)
=
g
+ i j gkl (x) ;
i
j
i
j
x x
x x
x x
(24)
that is, the value of gij on 0 is determined by gij on 0 and by the map : 0 0 .
Eq. (23) depends only on the boundary diffeomorphisms ini (

x ) = (
x , 0) and fin (
x) =
(
x , T ). The integral over the bulk diffeomorphisms depends in general on the boundary
surfaces.3 The result of this integration is an appropriate functional such that the convolution
property of the kernel will be satisfied. However, since we are interested in explicitly computing the linear approximation, this term is not relevant and it will be discarded in the following.
The boundary gauge transformations play a more subtle role than in YM theory. Indeed, notice

that we cannot write W [gij , gij , , T ] in the form W [gij , gij , T ] as we did for YM, because
0 and 0
affects also the action. Let us distinguish the temporal boundary diffeomorphisms ini
fin
i
i
from the spacial ones, ini and fin . Truly, (23) is invariant under a spacial diffeomorphism which
acts identically on both boundaries. This fact is analogous to the YM case, and allows us to drop
the dependence on i of one of the boundaries, say at t = 0. On the other hand, this property is
not true for temporal diffeomorphisms as emphasized in [8]. Thus we can write, introducing the
shorthand notation L[gij ] L[gij , g00 = 1, g0i = 0],

0(
x ,T )
3
dt L[gij ] .

i
0
0
Dfin
Dfin
Dini

gij
0 (
x ,0)
0
gij
(25)
This is the exact expression for the propagation kernel of GR. We expect the integration over the
gauge parameters to implement the constraints of the theory. In Section 3, we show explicitly
how this happens in the linearized case. Before that, however, we discuss a feature of the kernel
which is characteristic of GR.
2.2. The disappearance of time
Let us first focus on the case (i) in which (, gij ) is a compact space with finite volume,
0 integration in (25). This is an integration over all possible coordinate
and consider the Dfin
3 We thank our referee for pointing this out.
242
positions of the final surface. This ensemble is the same whatever is T , hence the right-hand side
of (25) does not depend on T . Therefore in this case the propagation kernel of GR is independent
from T , and we can simply write
W [gij , gij ] =

i
0
0
Dfin
Dfin
Dini

gij

0(
x ,1)
3
dt L[gij ] . (26)
0 (
x ,0)
0
gij
This is what we mean by disappearance of time. Notice however that a proper time is (generically) determined by the initial and final 3-metrics themselves. In fact, the integral (26) is likely
to be picked on the classical solution g bounded by gij and gij . But recall that in classical
GR the initial and final 3-metrics are expected to generically determine a classical time lapse between them. To see how this can happen, consider the theory in the partial gauge-fixing g0i = 0.
x , t) are second order equations that we expect to generically
The six evolution equations for gij (
admit a solution for given initial and final data at t = 0 and t = 1. The timetime component of
the Einstein equationsthe scalar constraintcan then be written in the form

gij gkl
1
g00 det gij R[gij ] = 0,
g ik g j l g ij g kl
(27)
2
t t
where R is the Ricci scalar of the 3-metric gij . Once a solution gij (
x , t) is given, this equation
x , t) and therefore it determine the physical
can be immediately solved algebraically for g00 (
proper time

1
1
(g ik g j l 1 g ij g kl ) gij gkl

2
t t
dt
T (
x) =
(28)
g00 (
x , t) dt =
det gij R[gij ]
0
between initial and final surface, along the x = const lines, which in this gauge are geodesics normal to the initial surface. Hence in general this proper time is determined by the initial and final
3-geometries. On the other hand, notice that as an equation for g00 , (27) becomes indeterminate
when R[gij ] = 0, and in particular on flat space.
The disappearance of the time coordinate in (26), and in general in the transition amplitudes
of quantum gravity, and its physical interpretation, have been amply discussed in the literature.
Its physical meaning is that GR does not describe physical evolution with respect to a time
variable representing an external clock, but rather the relative evolution of an ensemble of partial
observables. See for instance [1] for a detailed discussion, and the references therein.
The invariance of GR under coordinate time reparametrization implies that the Hamiltonian
H in (8) vanishes, and therefore we expect to have

n [gij ]n [gij ]
W [gij , gij ] =
(29)
n
instead of (10), where again n [gij ] is a complete basis of physical states, satisfying all the Dirac
constraints of GR, including, in particular, the WheelerDeWitt equation, which codes the quantum dynamics of the theory in the Hamiltonian framework. Eq. (29) indicates that W [gij , gij ] is
the kernel of a projector on the physical states of the theory [24]. Roughly speaking, the D i integration implements the invariance under spacial diffeomorphisms, while the D 0 integrations
243
project on the solutions of the WheelerDeWitt equation. We will see that this is indeed the case
in the linear approximation.
An explicit perturbative computation of the propagation kernel in the compact case would be
very interesting. The problem of expanding around a flat background in the compact case is that
this is precisely one of the degenerate cases where the thin-sandwich conjecture fails; essentially
because, as we have seen, (27) fails to determine g00 in this case. However, this can probably be
simply circumvented, for instance by adding a small cosmological constant. We leave this issue
for further developments, and we turn, instead, to the asymptotically flat case, where the explicit
computation of the propagation kernel can be performed in a more straightforward fashion.
Thus, consider the case (ii) in which (, gij ) is asymptotically flat. In this case, again (27)
does not determine T . However, the argument above for the disappearance of time fails, for the
following reason. As we have already mentioned, for the path integral to be well defined we must
x , t). In order to be well
require flat (i.e., gij = ij ) boundary conditions at spacial infinity for gij (
defined on this space of fields, the gauge transformations must vanish at infinity accordingly, as
they are required to do in the YM case. Therefore 0 (
x , T ) must converge to 0 (
x, T ) = T
for large x , and the right-hand side of (25) is not independent from T . In fact, what is physically
relevant is obviously not the coordinate time, but rather the proper time separation between initial
and final surfaces at infinity. That is, in the asymptotically flat case, the GR field propagator
depends on the asymptotic proper time at infinity.4
3. Linearized theory
We now focus on the asymptotic case, write
g (x) = + h (x)
(30)
and consider the field h (x) as a perturbation. Since we restrict to small h fields, we restrict,
accordingly, to small diffeomorphisms, that preserve the form (30). The gauge condition (19)
implies h0 (x) = 0. In this gauge, the action (16) reads
T

3

1
dt L = d x dt i j hij 2 h + hii hii hij hij
4
0

+ k hij j hik + k hij i hj k i hij j h j hij i h .
T
d x
0
(31)
Indices are now raised and lowered with the Minkowski metric and the 3d Euclidean metric. It
is convenient to separate the four Poincar-invariant components of the linearized field hij : the
spin-zero trace of the transverse components h0 , the spin-one and spin-zero longitudinal components hi = hTi + ppi hL , and the spin-two traceless and transverse components hTT
ij . In Fourier
space, this decomposition is
pj
pi pj
pi
+ hTj
+ hL 2 + hTT
hij = 2h0 Dij + hTi
(32)
ij
p
p
p
4 A suggestive way of understanding the presence of asymptotic time in the asymptotically flat case, based on the
boundary interpretation of the time evolution developed in [1,22] is the following. According to [1,22], the observable
proper time in GR can be determined by the boundary value of the propagation kernel of a finite spacetime region, namely
a 4d ball. In the limit in which the spacial dimensions of the region go to infinity, the boundary proper time converges to
the asymptotic proper time.
244
where Dij is the projector on the transverse modes given in (13) and
p i hTT
ij = 0,
ij hTT
ij = 0,
p i hTi = 0.
(33)
See Appendix C for more details.

The action is invariant under infinitesimal spatial diffeomorphisms, x i i (x) x i + i (x),
which induce the transformation hij (x) h ij (x) = hij (x)+i j (x)+j i (x) on the linearized
field. Or
hi (x) h i (x) = hi (x) + i (x).
(34)
Note, from this last expression, that i (x) is of the same order of hij (x).
As mentioned, the role of the temporal infinitesimal diffeomorphism is subtle: the Lagrangian
is insensitive to them, but they affect the extrema in the t integral in the action.5 The term
i j hij 2 h = 2 2 h0 appearing in the Lagrangian is a boundary term, and does not contribute to the equations of motion, but it does contribute to the integral. It is the linear approximation to the left-hand side of the timetime component of the Einstein equations (27), namely
of the GR scalar constraint.
x ) and hij (T , x) = hij (
x ), vanish at spacial infinity.
The boundary data hij (0, x) = hij (
Hence (25) becomes

T + 0 (T ,x )
i d3x
dt L[hij ,h0 =0]

i
0
0
0+ 0 (0,
x)
W [hij , hij , T ] = D fin D fin D ini
Dhij FP [hij ]e
.
hij
hij
(35)
i
D fin
The
integration implements the invariance (34) separately on initial and final data. It thus
makes W [hij , hij , T ] independent from the spinone and longitudinal spinzero longitudinal
components hi and hi . How about the integrations over the 0 s? Since the background metric
is static, 0 enters only the actions boundaries.6 Taylor expanding the action we have
T + 0 (
x ,T )
dt L[hij , h0 = 0]
d x
0+ 0 (
x ,0)
1

4
T

3
d x

dt h h hij hij + k hij j hik + k hij i hj k

i hij j h j hij i h

+ d 3 x 0 (T , x) i j hij 2 h t=T 0 (0, x) i j hij 2 h t=0 .
(36)
Thus, the integration over the 0 s gives two -functions of the linear term i j hij h on the
boundary data
5 Condition (18) for the invariance of the action now reads 0 (
x , 0) = 0, 0 (
x , T ) = 0.
6 On a generic background g 0 , the transformation under diffeomorphisms of the perturbation is, at first order, h (x)
ij
ij
0 (x). Therefore, 0 would enter the boundary data.
h ij (x) = hij (x) + i j (x) + j i (x) + gij
W [hij , hij , T ]

i
2 i

2
i
= i j hij h
i j hij h
D
245
Dhij FP (hij )
i
hij
hij

T

i
3
exp
d x dt h h hij hij + k hij j hik + k hij i hj k
4
0

ij
j
i
i h j h hij h .
(37)
The argument of the first -function does not depend on i , since, from the transformation propi
i
erties of hij we have i j h ij 2 h = i j hij 2 h . Recalling also that the constant
Fourier component of the boundary data vanishes because of the conditions at infinity, and absorbing a constant in the normalization factor, we can write the -functions simply as (h0 )(h0 ).
These two -functions impose the linearized scalar constraint h0 = 0 on the boundary data, and
project on the solution of the quantum scalar constraint, namely the WheelerDeWitt equation.
As emphasized by Kuchar in [16], indeed, in the linear approximation the scalar constraint is not
anymore a relation between momenta and configuration variables, but rather a condition on the
configuration space: the solution of the linearized WheelerDeWitt equation are the wave functions with support on h0 = 0. Using this argument, in [17] Hartle imposes these two -functions
on the ground state functional for Euclidean quantum gravity by hand. Here, instead, we obtain
them as a result of the FP integration calculations.
Notice the two different ways in which the linearized temporal and spacial diffeomorphisms
eliminate, respectively, the spin-zero h0 and spin-one and spin-zero longitudinal components hi
of the linearized field hij . The physical states are independent from hi , while they are concentrated on h0 = 0. Finally, we are left only with the physical spin-two field hTT
ij .
The FP factor FP is cubic in the fields [9]. Since in the following we are interested in computing the Gaussian approximation, we neglect it. The Gaussian integral over hij is straightforward;
we give the details of the integration in the appendix. Writing
H TT (p)
= hTT ij (p)h
TTij (p)
+ hTT ij (p)h
TTij (p),

H TT (p)
= hTT ij (p)h
TTij (p)
+ hTT ij (p)h
TTij (p),
(38)
the result of the integration is

i
W [hij , hij , T ] = N (T )(h0 )(h0 )e 4
TT cos pT H TT (p)
d3p

p H (p)
sin pT
(2)3
(39)
where the function N (T ) is a normalization factor. This is the field propagation kernel of linearized GR.
3.1. Ground-state and graviton states
We are now ready to read the vacuum state and the n-graviton states from the propagation
kernel (39). To do so, we expand (39) in power series of eipT . To first order, we obtain
246

1
d3p
pH TT (p)
W [hij , hij , T ] = N (T )(h0 )(h0 )e 4 (2)3

1
d 3p
TT
ipT
1+
pH
(
p)e

+
.
2
(2)3
(40)
Using (11), the (non-normalized) vacuum state can be read from the zeroth order of (40): we
have

1
d 3p
TT
0 [hij ]0 [hij ] = (h0 )(h0 ) exp
pH
(
p)

,
4
(2)3
and therefore

1
d 3p
TT
TT
0 [hij ] = (h0 ) exp
ph ij (p)h

.
ij (p)
4
(2)3
(41)
This is in agreement with the literature [16,17].

The graviton states can be obtained from the analog of (14), namely
3
1
d 3 pn i n Em T
d p1
m=1
W [hij , hij , T ] =
e
3
n!
(2)
(2)3

n=0
p1 1 ,...,pn n [hij ]p1 1 ,...,pn n [hij ].

This expression can be matched with (40) to extract the n-graviton states. The (non-normalized)
wave functional of the one-graviton state with momentum p and polarization , for example,
reads
0 [hij ]
p, [hij ] = (h0 ) p ij hTT ij (p)
(42)
and so on.
4. Newton potential from the propagation kernel
In the presence of external static sources, the energy of the lowest energy state must be the
Newton self-energy of the external source. Therefore the Newton self-energy can be extracted
from the field propagation kernel as its lowest Fourier component in T . This procedure has
proven effective in YM theory [11]. In particular, in the Abelian case, the external source can
be taken to be static, and the lowest energy state is characterized by the Coulomb self-energy.
In the non-Abelian case, on the other hand, the external sources cannot be static, reflecting the
exchange of colour charges between the external sources and the system. Inserting an external
source in GR is a more delicate procedure, because the Newton potential emerges not only in the
non-relativistic limit, as the Coulomb potential in YM theory, but also in the low gravity limit.
This means that we can follow the same procedure one uses in YM theory, but we expect for the
Newton potential to emerge only in the linear approximation.
x , t), we consider the Lagrangian
To introduce an external source J (
Lm [g ; J ] = L[g ] g J .
(43)
The source is the densitized matter energymomentum tensor J = gT . In the linear

approximation around Minkowski space, the Lagrangian (43) becomes
Lm [h ; J ] = L[h ] J + h J .
(44)
247
We characterize a static external source with the condition that only the component J00 = be
different from zero. (
x , t) is thus the energy density of the source. Covariance is broken since
the source itself defines a preferred frame. The source term in (44) reads then + h00 . Note
that the conservation law T = 0 at first order reads 0 = 0, consistently with the fact that
the source is static. In the temporal gauge h0 = 0 the Lagrangian becomes
Lm [hij , h0 = 0; J ] = L[hij ] + .
(45)
Notice that we seem to lose the coupling between the gravitational field and the external source.
This is analogous to what happens in YM, where the coupling term A0 to an external static
source is killed by the temporal gauge A0 = 0. This seems to prevent us from coupling external
sources to the field in the temporal gauge. But the coupling can nonetheless be obtained, because
part of the gauge degrees of freedom turn out to describe the source [11]. Below we show how this
happens. Indeed, staring from Lagrangian (45) and following the same steps as in the previous
sections we arrive at the following expression for the propagation kernel,
W [hij , hij , T ]

= N (T ) i j hij 2 h + i j hij 2 h + exp iT d 3 x

2Hij kl (p)
+ H ij kl (p)

i
d 3p
p2 T
Dij Dkl
exp
3
4
12
(2)

Hij kl (p)
H ij kl (p)
2
p Dij Dkl 2pi pj Dkl 2pk pl Dij
+
2p 2 T

Hij kl (p)
cos pT H ij kl (p)

(Dik Dj l + Dil Dj k Dij Dkl ) ,
+p
2 sin pT
(46)
where
x , y) = hij (
y )hkl (
x ) + hij (
y )hkl (
x ),
Hij kl (
x , y) = hij (
y )hkl (
x ) + hij (
y )hkl (
x ).
H ij kl (
(47)
As before, the integrals over the 0 s implement the WheelerDeWitt constraint, but in presence
of matter this is now i j hij 2 h + , or, in momentum space, 2p 2 h0 (p).
Using these
constraints we get
W [hij , hij , T ] = N (T )ei(E0 m)T
where
1
E0 =
32

3
d x

HijTTkl (p)
cos pT H ijTTkl (p)

i
d 3p
p
,
exp
4
sin pT
(2)3
(48)
(
x )(
y)
,
d y
|
x y|
3

m=
(
x ) d 3 x.
(49)
Therefore, the ground state is now weighted by the classical expression for the Newtonian selfenergy of the external source and by the its rest mass. In the presence of sources, the field
propagator nicely codes the Newtonian interaction.
248
5. Summary
We have studied the field propagation kernel of general relativity in the temporal gauge,
W [gij , gij , T ]. This is given in Eq. (25). It can be interpreted as a 3-geometry to 3-geometry
transition amplitude, or as a projector on the solutions of the quantum constraints. It is independent from T when gij and gij are defined on compact spaces; it depends on the asymptotic time
if they are asymptotically flat. When gij and gij are close to ij , it can be computed explicitly
to first order: the resulting expression is given in (39). It is independent from the spin-one and
spin-zero longitudinal components hi of the linearized field, and concentrated on the spin-zero
component h0 = 0 values. From its form, the linearized vacuum, given in (41), and n-graviton
states, as in (42), follow immediately. By coupling a static source to the field, we can extract the
Newtonian self-energy (49) from the linearized expression of the kernel.
These results are based on an accurate implementation of the FP procedure. The projection on
the solutions of the constraints is implemented by the integration over the FP gauge parameters.
This happens in a more subtle way than in YM theory, since temporal diffeomorphisms affect
the boundary of the action.
We have disregarded ultraviolet divergences. There are tentative background-independent definitions of quantum gravity that define 3-geometry to 3-geometry transition amplitudes free from
ultraviolet divergences. In order for these to have an acceptable low energy limit, they should
reduce to the propagator kernel computed here, for values of their arguments close to flat space.
Finally, showing that the background-independent definitions of quantum gravity lead to the
correct Newtonian interaction at low energy has proven so far surprisingly elusive. Here we have
shown how to extract this interaction from the field propagation kernel, by adding an external
source.
Acknowledgements
M.T. thanks the MIUR (Italian Minister of Instruction University and Research) and INFN
(Italian National Institute for Nuclear Physics) for partial support. C.R. thanks the physics department of the University of Rome La Sapienza and F.M., M.T. and S.S. thank the CPT for
hospitality during the preparation of this work.
Appendix A. Hamilton function of linearized GR
In this appendix we evaluate the Hamilton function of linearized GR. This is the value of the
gravitational action on a classical solution with given boundary values, expressed as a function
of these boundary values [1]. This will be used in the next appendix, to obtain the expression (39)
of the propagation kernel. To simplify the calculations, we work with Euclidean signature, using
analytic continuation to switch back to Lorentzian signature at the end of the calculations. In this
appendix, x 0 is the Euclidean time, defined by x 0 = it.
The quadratic part of the action of the linearized field, in the temporal gauge is

1
S=
d 4 x k hij j hik + k hij i hj k hij hij i hij j h j hkj k h
4

+ h h .
(A.1)
249
In the following calculations the linear term in (31) plays no role. The equations of motion
obtained by varying this action are
hij i m hmj j m him + i j h + ij m n hmn ij h = 0.
(A.2)
We are interested in evaluating the action on the classical solution with boundary data hij and
hij . Using the field equations and the boundary data, the action can be written as

1
Scl = ak bl ab kl
4

ab
ab
d 3 x h (
x )0x hkl (x)x 0 =T h (
x )0x hkl (x)x 0 =0 , (A.3)
where the hij (

x , t) is the classical solution interpolating between hij and hij . This classical
solutions can be constructed by means of a Green function Gij kl (x, y), defined by
Gij kl (x, y) i m Gmj kl (x, y) j m Gimkl (x, y) + i j Gm mkl (x, y)
1
+ ij m n Gmnkl (x, y) ij Gm mkl (x, y) = (ik j l + il j k ) (4) (x y),
2
(A.4)
x , 0; y) = Gij kl (
x , T ; y) = 0. To find it, we Fourier transform in
with boundary conditions Gij kl (
the spacial components,

Gij kl (x, y) =
x y )
d 3p
x 0 , y 0 ei p(
;
Gij kl p;
3
(2)
ij kl into covariant tensors and match the corresponding terms in (A.4). We

then decompose G
obtain

ij kl p;
x 0 , y 0 = A p 2 , x 0 , y 0 pi pj pk pl + B p 2 , x 0 , y 0 pi pj pk pl + p 2 (pi pk j l
G

+ pj pl ik + pi pl j k + pj pk il pi pj kl pk pl ij )

+ C p 2 , x 0 , y 0 p 4 (ik j l + il j k ij kl )
p 2 (pi pk j l + pj pl ik + pi pl j k + pj pk il pi pj kl pk pl ij )

+ pi pj pk pl ,
(A.5)
where the functions A, B, C are defined as follows:
3
2
2

y 0 + 2T 2 y 0 3y 0 T + (y 0 T )x 0 0 (x 0 y 0 )3 0
A p2 , x 0 , y 0 =
x
x y0 ,
2
2
12p T
12p

x0
(x 0 y 0 ) 0
+
B p2 , x 0 , y 0 = y 0 T
x y0 ,
4
4
2p T
2p

2 0 0 sinh px 0 sinh (py 0 pT ) 0
y x0
C p ,x ,y =
5
2p sinh pT

sinh(px 0 pT ) sinh py 0 0
+
x y0 .
2p 5 sinh pT
The Green function allows us to write the solution of the field equations with given boundary
data
250
hkl (y) = (im j n ij mn )

d 3 x hmn (
x )0x Gij kl (x, y)x 0 =T hmn (
x )0x Gij kl (x, y)x 0 =0 .
(A.6)
By inserting (A.5) into (A.6) and this into (A.3), we obtain the Hamilton function of unconstrained linearized GR
Scl [hij , hij , T ]

2Hij kl (p)
Hij kl (p)
+ H ij kl (p)

H ij kl (p)

1
d 3p
p 2 T
Dij Dkl +
=
3
2
4
12
(2)
2p T
2
p Dij Dkl + pi pk Dj l + pj pl Dik + pi pl Dj k + pj pk Dil 2pi pj Dkl

2pk pl Dij

cosh pT H ij kl (p)

Hij kl (p)
[Dik Dj l + Dil Dj k Dij Dkl ] ,
+p
2 sinh pT
(A.7)
where Dij and Hij kl are given in (13) and (47).

Appendix B. Evaluation of the propagation kernel
Here we illustrate the derivation of (39) from (37). To perform the Gaussian integral over hij ,
we write the field as the solution of the classical equations of motion h0ij with the given boundary
data, plus a fluctuation ij . The boundary data are then
ij (0) = ij (T ) = 0.
(B.1)
Linear terms in ij vanish because of the equations of motion, the integration over ij yields a
normalization factor N (T ) depending of T , but independent of the boundary data. The remaining
exponential, independent of the perturbation field, is the Hamilton function, computed above.
Using this, we obtain for the Euclidean propagation kernel
WE [hij , hij , T ]

= N (T ) i j hij 2 h i j hij 2 h

1
d 3p
3
y
x)
D i (T , x) exp
x
d 3 y ei p(
d
4
(2)3

2Hij kl (
x , y) + H ij kl (
x , y)
T Dij Dkl
p 2
12
x , y) H ij kl (
x , y) 2
Hij kl (
p Dij Dkl + pi pk Dj l + pj pl Dik + pi pl Dj k
+
2
2p T

+ pj pk Dil 2pi pj Dkl 2pk pl Dij

x , y) cosh pT H ij kl (
x , y)
Hij kl (
(Dik Dj l + Dil Dj k Dij Dkl ) .
+p
2 sinh pT
(B.2)
Here the apex indicates a spatial diffeomorphism i only on the field hij ,
Hij kl (
x , y) = h
i
i
y )h kl (
x ) + hij (
y )hkl (
x ),
ij (
(B.3)

i
i
H ij kl (
x , y) = h ij (
y )hkl (
x ) + hij (
y )h kl (
x ).
251
(B.4)
The term in the last line of the expression (B.2) has the structure of the propagation kernel for a
harmonic oscillator (see for instance [1]).
The next step is to consider the integration over the spatial diffeomorphisms; the integrals can
be performed separately for the longitudinal and the transversal parts7 of i . When we perform
the integration over the transversal spatial diffeomorphisms, the result we obtain is that only the
term with the structure of a harmonic oscillator survives. On the other hand, the integration over
the longitudinal part i i gives again an implementation of the scalar constraint, in the form

i j hij 2 h i j hij + 2 h .
(B.5)
The redundance of the integration over the 0 s and the i i can be understand as follows: in
the Hamiltonian formalism, the scalar constraint is proportional to the lapse function N . In the
Lapse = 1, Shift = 0 gauge we are working it, the transformation of the lapse under a diffeomorphism is given by (see for instance [9]) N = 1 + 0 0 + i i . Therefore, both the integration
over 0 and over i i provide variation of the lapse and consequently an implementation of the
scalar constraint.
In the case of matter, the redundant implementation of the scalar constraint (B.5), coming
from the integration over the longitudinal spatial diffeomorphisms, is consistent with the other
-functions, because the source is static and thus (T ) = (0).
The final expression can be extended to Lorentzian signature by means of the analytic continuation T iT , and we obtain
W [hij , hij , T ]

= N (T ) i j hij 2 h i j hij 2 h

Hij kl (
x , y) cos pT H ij kl (
x , y)
i
d 3 p i p(
exp
e y x ) p
d 3x d 3y
3
4
2 sin pT
(2)

(Dik Dj l + Dil Dj k Dij Dkl ) ,
(B.6)
which coincides with (39), once the spacial indices are contracted.
Appendix C. Linearized Einstein equations in the temporal gauge
For completeness, we write here the linearized Einstein equations in the temporal gauge h0 =
0, extensively utilized in this paper. To do so it is convenient to work in the Fourier space with
the Poincar-irreducible components introduced in (32). These can be obtained from the field as
follows
1
h0 = hkl Dkl ,
2
pk
hTi = hkl Dil ,
p
pk pl
,
p2

1
D
D
hTT
=
h
D
D
kl
ik j l
ij kl .
ij
2
hL = hkl
7 We have the usual decomposition of a 3-vector, i = i + i = 1 i j + i 1 i j .

j
j
L
T
2
2
(C.1)
252
We consider the case of a dust distribution with energy density (

x ). So the only non-vanishing
component of the energymomentum tensor is T00 = . As pointed out in Section 4, the source
has to be static in the linearized case because of the continuity equation. We will found this result
also in the study of the Einstein equations; this is not surprising because the continuity equation
comes from the Bianchi identities, which are properties of the Einstein equations themselves.
The vacuum case is easily reconstructed setting = 0.
The linearized equations for a dust distribution are
h h h + h = .
(C.2)
2
They can be written in the temporal gauge as follows
h i i = ,
(C.3)
2
j h ij + i h k k = 0,
(C.4)
hij i k hkj j k hj i + i j hk k = ij ,
(C.5)
2
where the dot indicates the derivative respect to the zero component and = . From the
time-equations (C.3), (C.4) we obtain the constraints in terms of the quantities defined in (C.1)
h 0 = 0,
(C.6)
h Ti = 0,
h L = .
2
On the other hand, imposing the constraints on the motion equations (C.5) we have
hTT
ij = 0,
2p 2 h0 = .
(C.7)
The first is a wave equation for the traceless transverse components, which are the only physical degrees of freedom. The second is the WheelerdeWitt equation in the Fourier space, as
can be easily seen from the definition of h0 in (C.1). As it was previously said, the first of the
constrains (C.6) together with the WheelerdeWitt equation imposes the source to be static.
In terms of the Poincar-irreducible field components, the quadratic term of the Lagrangian
can be written as
TT

T T
4L(2) = hTT
ij hij + 2h0 h0 + 4h0 hL 2hi hi .
(C.8)
References
[1] C. Rovelli, Quantum Gravity, Cambridge Univ. Press, Cambridge, 2004.
[2] A. Perez, Spin foam models for quantum gravity, Class. Quantum Grav. 20 (2003) R43.
[3] K. Symanzik, Schrdinger representation and Casimir effect in renormalizable quantum field theory, Nucl. Phys.
B 190 (1981) 1.
[4] M. Luscher, R. Narayanan, P. Weisz, U. Wolff, The Schrdinger functional: A renormalizable probe for non-Abelian
gauge theories, Nucl. Phys. B 384 (1992) 168.
[5] C. Misner, Feynman quantization of general relativity, Rev. Mod. Phys. 29 (1957) 497.
[6] C. Misner, K.S. Thorne, J.A. Wheeler, Gravitation, Freeman, San Francisco, 1973.
[7] S. Hawking, The path integral approach to quantum gravity, in: S. Hawking, W. Israel (Eds.), General Relativity:
An Einstein Centenary Survey, Cambridge Univ. Press, Cambridge, 1979;
G.W. Gibbons, S.W. Hawking (Eds.), Euclidean Quantum Gravity, World Scientific, Singapore, 1993.
[8] C. Teitelboim, Quantum mechanics of the gravitational field, Phys. Rev. D 25 (1982) 31593179.
[9] C. Teitelboim, Proper-time gauge in the quantum theory of gravitation, Phys. Rev. D 28 (1983) 297309.
253
[10] M.J.G. Veltman, Quantum theory of gravitation, in: Proceedings, Methods In Field Theory, Les Houches, 1975, Les
Houches, Amsterdam, 1976, pp. 265327.
[11] G.C. Rossi, M. Testa, The structure of YM theories in the temporal gauge, 1: General formulation, Nucl. Phys.
B 163 (1980) 109132.
[12] G.C. Rossi, M. Testa, Ground state wave function from Euclidean functional integral, 1: Quantum mechanics, Ann.
Phys. 148 (1983) 144.
[13] F. Cesi, G.C. Rossi, M. Testa, Non-symmetric double well and Euclidean functional integral, Ann. Phys. 206 (1991)
318333.
[14] R. Delbourgo, B.D. Winter, Matter field infinities in axial gauge gravity, J. Phys. A 15 (1982) L165.
[15] C. Teitelboim, Quantum mechanics of the gravitational field in asymptotically flat space, Phys. Rev. D 28 (1983)
310316.
[16] K. Kuchar, Ground state functional of linearized gravitational field, J. Math. Phys. 11 (1970) 3322.
[17] J.B. Hartle, Ground-state wave function of linearized gravity, Phys. Rev. D 29 (1984) 2730.
[18] G.W. Gibbons, S.W. Hawking, Action integrals and partition functions in quantum gravity, Phys. Rev. D 15 (1977)
27522756.
[19] R.F. Baierlein, D.H. Sharp, J.A. Wheeler, Three-dimensional geometry as carrier of information about time, Phys.
Rev. 126 (1962) 1864.
[20] R. Bartnik, G. Fodor, On the restricted validity of the thin sandwich conjecture, Phys. Rev. D 48 (1993) 35963599.
[21] A. Peres, N. Rosen, On Cauchys problem in general relativity, Nuovo Cimento XII (1959) 430.
[22] F. Conrady, L. Doplicher, R. Oeckl, C. Rovelli, M. Testa, Minkowski vacuum in background independent quantum
gravity, Phys. Rev. D 69 (2004) 064019.
[23] H. Friedrich, G. Nagy, The initial boundary value problem for Einsteins vacuum field equation, Commun. Math.
Phys. 201 (1999) 619655.
[24] C. Rovelli, The projector on physical states in loop quantum gravity, Phys. Rev. D 59 (1999) 104015.
Discrete flavor symmetry, dynamical mass textures,

and grand unification
Naoyuki Haba a , Koichi Yoshioka b,
a Physik-Department, Technische Universitt Mnchen, James-Franck-Strae, 85748 Garching, Germany
b Department of Physics, Kyushu University, Fukuoka 812-8581, Japan
Received 20 November 2005; received in revised form 22 December 2005; accepted 19 January 2006
Abstract
Discrete flavor symmetry is explored for an intrinsic property of mass matrix forms of quarks and leptons.
In this paper we investigate the S3 permutation symmetry and derive the general forms of mass matrices in
various types of S3 theories. We also exhibit particular realizations of previous ansatze of mass matrices,
which have often been applied in the literature to the standard model Yukawa sector. Discrete flavor symmetry is also advantageous for vanishing matrix elements being dynamically generated in the vacuum of
scalar potential. This is due to the fact that group operations are discrete. While zero elements themselves
do not explain mass hierarchies, we introduce an Abelian flavor symmetry. A non-trivial issue is whether
successful quantum numbers can be assigned so that they are compatible with other (non-Abelian) flavor
symmetries. We show typical examples of charge assignments which not only produce hierarchical orders
of mass eigenvalues but also prohibit non-renormalizable operators which disturb the hierarchies in firstorder estimation. As an explicit application, a flavor model is constructed in grand unification scheme with
S3 and U (1) (or ZN ) flavor symmetries.
1. Introduction
One of the most important issues in and beyond the standard model is the masses and mixing angles of the three-family quarks and leptons. After the electroweak symmetry breaking, the
observed values of masses and mixing angles are to be produced from the structures of Yukawa
couplings. However the Yukawa couplings generally have redundancy in explaining the experi* Corresponding author.
E-mail address: yoshioka1scp@mbox.nc.kyushu-u.ac.jp (K. Yoshioka).

doi:10.1016/j.nuclphysb.2006.01.027
N. Haba, K. Yoshioka / Nuclear Physics B 739 (2006) 254284
255
mental results: apparently different forms of Yukawa matrices lead to the same physical masses
and mixing angles. Therefore it has been a long outstanding problem which patterns of Yukawa
couplings are relevant from phenomenological and theoretical viewpoints. Various progresses
have been made in the literature by applying additional principles to the standard model Yukawa
sector. The two well-known examples of such principles are to adopt the unification hypothesis
of matter multiplets and to assume specific forms of Yukawa couplings (textures). The former
is a topdown approach to the problem. The grand unification principle relates the properties of
quarks and leptons, and reduces the degrees of freedom of Yukawa couplings in the theory. On
the other hand, the latter approach is rather a bottomup one. Available forms of Yukawa textures are explored so that they are consistent with the experimental observations. It is interesting
that the number of successful textures is found to be highly limited. This fact is revealed from
a simplifying assumption that some elements of Yukawa matrices vanish, called texture zeros.
Along this line, phenomenologically possible forms of mass matrices have been proposed in the
literature (for example, [13], and also [4,5] for systematic analyses of zero textures).
As for the neutrino sector, the recent experimental results suggest that one of the most likely
forms of Majorana mass textures of light neutrinos is proportional to
O(1)
O(1)
O(1)
O(1)
(1.1)
in the basis where the generation mixing has been rotated out in the charged-lepton side. This is
the dominant part of mass matrix and other small entries have not explicitly been written down.
It is clearly seen that the large leptonic mixing between the second and third generations [6] is
predicted from (1.1). The observed large 12 mixing [7] requires an additional condition for the
above texture: the dominant 2 2 submatrix of (1.1) has a reduced rank and its determinant
is of the same order of small off-diagonal elements neglected in (1.1). Given that, the texture
form (1.1) is considered as a promising candidate consistent to the present experimental data.
The condition (the vanishing determinant of the dominant submatrix) may be realized without
fine tuning of model parameters, e.g., with the right-handed neutrino dominance [8], R-parity
violation [9], and the lopsided form of charged-lepton mass matrix [10]. It is interesting that the
lopsided mass textures can be naturally embedded in grand unified theory. However if the theory
is supersymmetrized, large off-diagonal elements in the lepton Yukawa matrices generally induce
sizable rates of flavor-violating processes to excess the present experimental bounds [11]. While
the details depend on superparticle mass spectrum, a natural way to avoid this flavor problem is
to consider the case that the lepton as well as quark Yukawa matrices take hierarchical forms,
which lead to small generation mixing. If this is the case, an interesting possibility to have large
lepton mixing is to suppose asymmetric forms of mass textures (zeros).1 It seems that Abelian
flavor symmetry may be difficult to generate such asymmetric zeros without expense of model
complexity. Furthermore non-Abelian continuous flavor symmetry is not suitable for handling
mass textures since texture zeros are rotated to other arbitrary forms by continuous symmetry
rotations and do not have physical implications.
Motivated by these results, in this paper we investigate the power of non-Abelian discrete
flavor symmetry for constructing mass matrix models. We particularly focus on the minimal discrete non-Abelian symmetry S3 . (For fermion mass models based on other non-minimal discrete
1 Leptonic mixing angles may be enhanced, e.g. by integrating out heavy fields such as right-handed neutrinos [12].
256
non-Abelian flavor symmetries, see [13].) The S3 operation is the permutation of three objects,
which has a simple geometrical interpretation, i.e. the symmetry of an equilateral triangle. While
it is the smallest non-Abelian discrete symmetry, it might be regarded as a remnant of flavor symmetry of fundamental theory in high-energy regime. The purposes of this paper are the following
two points: (i) non-Abelian flavor symmetries such as S3 are so effective that various types of
phenomenological textures are handled and (ii) these symmetries can also generate asymmetric
forms of Yukawa matrices in dynamical ways. Further it should be noted that texture zeros themselves do not explain fermion mass hierarchies. Previous approaches to fermion masses with
S3 have assumed hierarchical values of Yukawa couplings and/or involved symmetry-breaking
patterns such as vacuum expectation values (VEVs) of Higgs scalars. In this paper we show that
realistic values of masses and mixing angles are dynamically achieved by introducing U (1) symmetry. A non-trivial problem arises whether successful U (1) quantum numbers can be assigned
so that they are compatible with other non-Abelian (flavor) symmetries. It is noticed that mass
hierarchy is also realized in a similar way with a discrete subgroup of the flavor U (1) symmetry
such as ZN with appropriate (enough large) N and the same quantum numbers as in the U (1)
case. A smaller choice of N would be possible and interesting from a viewpoint of brevity. In
this case, the problem of fermion masses can be handled with flavor symmetries that are entirely
discrete. While the U (1) charge assignments are presented in this paper, they can always be read
as the charges in flavor ZN theory. We finally present an explicit model where the S3 flavor symmetry is incorporated consistently to unified gauge symmetry and hierarchical forms of mass
matrices.
This paper is structured as follows. In Section 2, we discuss some fundamental issues of the
S3 group, which are needed in Section 3 to study symmetry-invariant forms of mass matrices. In
Section 4, S3 is applied to supply dynamical justifications to realistic candidates of mass textures
which have often been discussed in the literature. Based on these results, we present in Section 5
a toy flavor model in grand unification scheme where Yukawa textures are controlled by a single
flavor S3 , assisted by U (1) (or ZN ) symmetry. In Section 6, we analyze the invariant scalar
potentials of S3 doublet whose VEV form is a key ingredient of the approach developed in this
paper. Section 7 is devoted to summarizing our results.
2. The S3 group
2.1. Representations and representation matrices
The S3 symmetry is the smallest non-Abelian symmetry, the permutations which an equilateral triangle has. The S3 group therefore contains six elements T1 , . . . , T6 , half of which are the
circulations of triangle apices and the other half corresponds to the exchanges of two of three
apices while the other is fixed. There are only few numbers of irreducible representations; a twodimensional representation and two different one-dimensional representations. Throughout of
this paper we denote them as 2 (doublet), 1S (singlet), and 1A (pseudo singlet), respectively. The
non-trivial one-dimensional representation 1A distinguishes the group elements in two parts. The
matrix representations of group elements are given in Table 1. It is easily seen that the even permutations T1 , T2 and T3 constitute the subgroup Z3 . This means that the S3 symmetry is broken
down to Z3 when a pseudo singlet field 1A develops an expectation value. Thus the S3 group has
simple but non-trivial structures, and is suitable for applying it to the flavor problems of threegeneration fermions in the standard model. A number of models have been proposed to explain
Yukawa coupling structures of quarks and leptons with the S3 flavor symmetry [1416].
257
Table 1
The representation matrices of the S3 elements. The symbol is the third root of unity ( = e2 i/3 )
2
1S
1A
T1

1 0
0 1
1
1
T2

0
2
T3
2
1
1
T4

0 1
1 0
1
1

1
1
T5

0

2
0
1
1
T6

0
2
1
1
It may be convenient to introduce the reducible three-dimensional representations for discussing the three-generation flavor physics. Corresponding to 1S and 1A , there are two types of
three-dimensional representations:
3S 2 + 1 S
(triplet),
(2.1)
3A 2 + 1 A
(pseudo triplet).
(2.2)
If S3 is regarded as a remnant of some gauge symmetry in fundamental theory, 3A should be

applied not to induce discrete gauge anomaly. This is understood from the fact that the threedimensional vector representation of SO(3) is decomposed as 3 = 2 + 1A in terms of its subgroup
S3 . One may also use the 3S representation at the expense that S3 is assumed to be a global
symmetry or the anomaly is cancelled by introducing appropriate numbers of pseudo singlet
fermions. Further the aforementioned geometrical interpretation of S3 operations is made clear
for triplet representations. Such an interpretation is seen in a different basis of S3 , as will be
discussed later in this section. The matrix representations of 3S and 3A are read from Table 1 and
given by
,
,
,
T1 =
T2 =
T3 =
1
2
1
1
1
1
2
,
,
,
T5 =
T6 = 2
T4 = 1
(2.3)
1
1
1
for the 3S representation. As for 3A , the matrices Ti are given by (2.3) except that the 33 elements in the odd permutation matrices T4,5,6 are replaced with 1.
2.2. Tensor products and 2 representation
The tensor products involving 1A are given by 1A 1A = 1S and 1A 2 = 2. The only
remaining non-trivial product is that of two doublets: 2 2 = 2 + 1A + 1S . In the basis where
the group elements are given by Table 1, the product of two doublets = (1 , 2 )t and =
(1 , 2 )t is explicitly written as follows:

t

= + , 2 + 3 1 + 1 ,
(2.4)
A
S

t

= 1 2 , 2 1 2 + 1 1 2 2 1 + 1 1 + 2 2 1 ,
(2.5)
A
1 i2
2 .
where i (i = 1, 2, 3) are the Pauli matrices and

The subscripts in the right-handed
sides denote S3 representations. Note that (and also ) is generically complex-valued, while
S3 is a real group. This fact is important when transforms as some complex representation
258
under other symmetries than S3 . The complex conjugate belongs to the 2 representation
for which the representation matrices become Ti . In practical use, however, it is convenient to
define a doublet from an anti-doublet:

2
,
C 1 =
(2.6)
1
which transforms as C 1 (Ti ) = Ti 1 = Ti C , and indeed acts as the 2 representation
of S3 . The tensor product involving C is hence given by

t

C = t L , t R 2 + t i2 1 + t 1 1 ,
(2.7)
A
S

t
= 2 2 , 1 1 2 + (1 2 2 1 )1A + (1 2 + 2 1 )1S ,
(2.8)
1+3
3
where L 1
2 and R 2 , respectively. It is noticed that this product does not contain
any complex conjugates. Such a type of tensor product is necessary for describing, e.g., matter
interaction terms (F terms) in supersymmetric theory and Majorana masses for neutrinos. On the
other hand, the product (2.4) is applied to the usual Dirac mass terms of quarks and leptons. The
form of tensor product depends on the group basis, and therefore the S3 model construction and
its physical consequences also do. For details, see the next subsection and Section 3.
It may be useful for later discussion to explicitly write down what types of singlet components
are contained in the products of more than two doublets. In the basis discussed here, one finds
1 1 + 2 2 ,
1 2 + 2 1 ,
1 2 1 + 2 1 2 ,
1 2 2 + 2 1 1 ,
1 1 2 + 2 2 1 ,
1 1 1 + 2 2 2 ,

1 1 1 1 + 2 2 2 2 ,
1 1 1 1 + 2 2 2 2 ,
1 1 1 1 + 2 2 2 2 ,
1 1 1 2 + 2 2 2 1 ,
1 1 1 2 + 2 2 2 1 ,
1 1 1 2 + 2 2 2 1 ,
1 2 2 2 + 2 1 1 1 ,
1 2 1 1 + 2 1 2 2 ,
1 1 2 1 + 2 2 1 2 ,
1 1 2 2 + 2 2 1 1 ,
1 2 1 2 + 2 1 2 1 ,
(2.9)
(2.10)
1 2 2 1 + 2 1 1 2 .
(2.11)
Their Hermitian conjugates are also in the 1S representations.
2.3. Group basis dependence
We have presented the S3 algebra in the complex basis where the representation matrices are
given by those in Table 1. There are, however, several bases of the S3 matrices which are often
used in the literature. So it may be instructive here to describe the relation among these group
bases, compared to the complex basis used in the previous subsections.
2.3.1. Democratic basis
The democratic basis is adopted to produce a flavor-democratic mass matrix in which all
matrix elements are equal [14]. In this basis, the S3 operations generate the permutations of
three objects, for example, the exchange of the first and second indices. The invariance under
such transformations require the universal size of couplings for three generations if they belong
259
to a three-dimensional representation of S3 . Notice however that, since the three-dimensional

representation is reducible, the Higgs profile is appropriately chosen, namely should be in the
singlet such that the democratic form of mass matrix is reproduced. If this is not the case or
due to radiative corrections, the S3 invariance does not necessarily lead to the universal value of
couplings. We will mention this point in later sections.
Unlike in the complex basis, the representation matrices have apparently non-trivial (not
block-diagonal) structure for three-dimensional representations. Different group bases are converted to each other by unitary transformations. It is worth noting that there are two types of
democratic basis which correspond to the existence of two three-dimensional representations; 3S
and 3A . The elements in the democratic basis for 3S are defined by the unitary matrix V as
TiS = V Ti V ,
where Ti s in the right-handed side are given in the complex basis, and V U PS with
1/ 6 1/ 3
1/ 2 1/ 2
1/ 2
.
U = 1/ 2 1/ 6 1/ 3 ,
PS = i/ 2
i/ 2
0
2/ 6 1/ 3
1
Thus the representation matrices for 3S are given by the label-changing matrices:
1
1
1
,
,
T2S =
T3S = 1
T1S =
1
1,
1
1
1
1
1
1
,
,
T4S = 1
T5S =
T6S =
1
1.
1
1
1
(2.12)
(2.13)
(2.14)
S
For the pseudo triplet representation 3A , the matrices T1,2,3
have the same forms as above, but
S
T4,5,6 become rather complicated, as can be seen by the definition (2.12).
Another democratic basis is defined so that the representation matrices for 3A are expressed
by label-exchanging matrices like (2.14). The matrices TiA are given by a similar unitary rotation
to (2.12), except that the matrix PS is now replaced by
1/ 2 1/ 2
.
PA = i/ 2 i/ 2
(2.15)
1
A
take the same forms as those in (2.14) both for triplet and
One can see that the matrices T1,2,3
A (3 ) = T S (3 ) and
pseudo triplet representations. The differences appear for T4,5,6 ; T4,5,6
A
4,5,6 S
S
A
T4,5,6 (3S ) = T4,5,6 (3A ), the former matrices exchanges the three objects as (2.14) and the latter
have some complicated forms.
2.3.2. Real basis

In the discussion of the democratic basis, PS,A rotate only the first and second indices. This
means that the representation matrices TiS,A are rotated to block-diagonal forms only by the
U rotation, namely, the U -rotated matrices are decomposed into the matrices for irreducible
representations. (Note that U is the unitary rotation which diagonalizes the so-called flavordemocratic mass matrix.) The real basis TiU is defined by rotating TiS,A with the U matrix, for
260
example,
TiU = U TiS U.
(2.16)
These group elements take the following forms for the 3S representation:
1
1/2
3/2
,
,
T2U = 3/2 1/2
T1U =
1
1
1
1/2 3/2
1
,
,
T3U = 3/2 1/2
T4U =
1
1
1
3/2
1/2
3/2
1/2
,
,
T5U = 3/2 1/2
T6U = 3/2 1/2
1
1
(2.17)
which are block diagonal. It is clear from the definition that the T U basis is obtained from the
complex basis by the PS or PA phase rotation. Note that, unlike in the complex basis, all the
representation matrices are real-valued. This is why we call it the real basis. Due to this reality
of matrix elements of TiU , two types of tensor products are possible for S3 doublets. In the real
basis of (2.16), the one tensor product of two doublets is

t

= 1 , 3 2 + i2 1 + 1 ,
(2.18)
A
S

t
= 1 2 + 2 1 , 1 1 2 2 2 + 1 2 2 1 1 + 1 1 + 2 2 1 .
A
S
(2.19)
Another consistent product can be defined by using transposition of , instead of . This choice
is possible since all the group elements are expressed in terms of real numbers. In the language of
the complex basis, the product defined with daggers corresponds to (2.4) and that with transpositions to (2.7). The difference between these two types of products becomes important in the case
that S3 -doublet fields transform non-trivially under other symmetries than S3 . In particular, this is
indeed the case for the standard model fermions and Higgs bosons. The real basis has often been
used in the literature [15]. For completeness, we comment on another real basis which is defined
from TiA , instead of TiS in (2.16). The matrices TiU are now given by TiU = U TiA U and their
U
U
U ). Therefore
explicit forms for 3A are (2.17) by changing the signs of T4,5,6
(i.e., T4,5,6
T4,5,6
the non-trivial tensor product becomes = ( 3 , 1 )t2 + ( i2 )1A + ( )1S
or that with t .
Finally we show the singlet components contained in the products of more than two doublets.
In the real basis, they are given by (up to the fourth order)

1 1 + 2 2 ,
(2.20)
1 1 2 + 1 2 1 + 2 1 1 2 2 2 ,
(2.21)
(1 1 + 2 2 )(1 1 + 2 2 ),
(1 1 + 2 2 )(1 1 + 2 2 ),
(1 2 2 1 )(1 2 2 1 ),
(1 1 + 2 2 )(1 1 + 2 2 ),
(1 2 2 1 )(1 2 2 1 ),
(1 2 2 1 )(1 2 2 1 ).
(2.22)
261
In the right-handed sides, the components i , j , . . . can be replaced with i , j , . . . . Namely,

(1 1 + 2 2 ), (1 1 2 + 1 2 1 + 2 1 1 2 2 2 ), and others are also S3 singlets. This
is due to the fact that the representation matrices TiU are real in the real basis.
3. Invariant matrices
Various forms of mass textures of quarks and leptons have been known to be phenomenologically viable [15]. As discussed in the previous section, S3 has three types of irreducible
representations; 2, 1A and 1S . We would like here to examine what combinations of S3 representations for matter fields lead to those mass textures. Higgs fields may also transform non-trivially
under the S3 symmetry and play significant roles for realizing textures forms. Since there are
three repetitions of matter fields, various assignments of S3 representations are available, where
the first two generations constitute a doublet and so on. These include a possibility to realize asymmetrical forms of mass textures and give rise to physically observable effects such as
flavor-violating processes in future experiments. In particular, as mentioned in the introduction,
vanishing matrix elements lead to interesting phenomenological consequences. Such zero elements could be obtained in the framework of non-Abelian discrete flavor symmetry by suitably
taking Higgs representations and their VEV forms, which are calculable by analyzing scalar potential. Natural forms of VEVs generally depend on the S3 group basis in which mass matrices
are described. At this stage, explicit forms of mass textures are governed only by discrete flavor
symmetry. In some cases, however, additional implementation might be needed to have more control, in particular, for fermion mass hierarchy being naturally realized. In the present framework,
the key ingredients for model construction are to select (i) S3 group structure, (ii) representations
of matter fields, (iii) Higgs profiles (representations and VEVs), and (iv) extra symmetries. Let
us first comment on these issues in some details.
(i) S3 group structures: As mentioned in Section 2.3, physical consequences of a mass texture
potentially depend on the group basis adopted in constructing S3 models. It is noted that a choice
of specific basis does not affect physical results as long as the flavor symmetry is unbroken: Apparent basis dependence, e.g. different forms of mass matrices, is only due to a choice of flavor
basis. The S3 invariance guarantees the same spectrum without regard to basis choices. However the predictions for generation mixing might be different. This is because, in the standard
model, there already exists a basis which defines the generation structure, namely, the interaction basis where the weak current interaction is flavor diagonal. Once one picks up a group basis
from some model-building perspective, the relabeling of flavor indices by S3 transformation
gives physical meaning on generation structure. Another important basis dependence appears in
breaking the flavor symmetry. Since any kind of flavor symmetry has not been observed in the
low-energy regime, in principle any form of symmetry-breaking patterns is possible. In realistic
model construction, some guiding principles are often adopted, such as simplicity and/or dynamical justification. In either case, symmetry-breaking parameters depend on the S3 basis. While
the breaking parameters take a simple form in one basis, they are rotated to a complicated form
in other general bases, which form seems to be completely unnatural. In this way, the choice of
flavor-group basis may have physical consequences if three-generation fermions are assigned to
(pseudo) triplets, and therefore is an important factor in constructing models with flavor symmetry.
Another issue is what types of S3 symmetries are involved into the theory. For example, with
only one S3 symmetry, both left- and right-handed fermions (and also Higgs bosons) transform
under the same S3 . On the other hand, one may easily imagine that three-generation fermions
262
have non-trivial charges of separate S3 groups. A well-known example of the latter case is the
democratic mass texture [14] realized by S3L S3R symmetries.
(ii) Representations of matter fields: In this paper, S3 is introduced to control the flavor
structure of quarks and leptons. Three-generation matter fields generally belong to non-trivial
representations of such flavor symmetries. A charge assignment often assumed in the literature is that the first and second generations make up a doublet. This assignment is adopted to
account for several phenomenological issues. First, the flavor symmetry invariance suppresses
flavor-violating effects between the first and second generations, which effects have been tightly
constrained by various experimental results in supersymmetric theory [17]. Second, if the third
generation has different flavor charges from the other twos, they are appropriate to have larger
masses. It is however noted that all other flavor charge assignments are equally allowed. For
example, a phenomenologically interesting alternative is that the second and third generation
leptons make an S3 doublet. This is motivated by the recent observations of neutrino flavor mixing between the second and third generations. Moreover it could account for the lightness of the
first-generation fermions. In what follows, we show that representations of matter fields, including leftright asymmetric assignments, are useful for obtaining various types of mass matrices.
(iii) Higgs profiles: Phenomenologically indirect but important possibilities arise for the profile of Higgs fields. If SU(2)W -doublet Higgses are in non-trivial representations of S3 , Yukawa
couplings can be described by renormalizable operators. Higher-dimensional operators including multiple Higgs fields are suppressed by a large cutoff scale and give negligible corrections
to Yukawa couplings, which is a nice feature in a sense that everything is described within the
renormalizable level. This approach however requires that hierarchically small values of Yukawa
couplings for the first two generations must be fixed by hand as in the standard model. Moreover,
due to the existence of multiple SU(2)W -doublet Higgses, naive gauge coupling unification is
spoiled, and Higgs-mediated flavor-violating effects might not be negligible even at tree level.
An alternative choice of Higgs charges is that SU(2)W -doublet Higgses belong to the flavor singlet. In this case, Yukawa interactions are effectively derived from higher-dimensional operators
which can be made invariant by introducing appropriate scalars s in non-trivial S3 representations. The VEVs of break the flavor symmetry and generate trilinear Yukawa couplings
below the breaking scale. Since the ratio to a cutoff scale gives a unit of Yukawa hierarchy
of quarks and leptons, S3 should be broken at a high-energy scale below the cutoff. This situation resolves the above-mentioned problems of S3 -charged SU(2)W Higgs doublets: Yukawa
hierarchy is explained by controlled higher-dimensional operators, gauge coupling unification is
preserved, and Higgs-mediated flavor violation is suppressed by a large S3 -breaking scale. As
an imprint of such high-scale flavor symmetry, new sources of flavor violation could arise from
renormalization-group evolution below the symmetry-breaking scale. For example, if the theory
is supersymmetrized, flavor-changing couplings are generally induced, and their magnitudes depend on the dynamics of supersymmetry breaking. While the flavor violation tends to be small as
Khler terms are limited by flavor symmetry, it may be observable, e.g. in the gravity mediation
scenario [18]. Note that this type of flavor violation is negligible in the case of S3 -charged Higgs
bosons as long as the flavor symmetry remains intact at low-energy regime.
The S3 property of Higgs fields is also relevant to the group basis. That is the form of VEVs
and their naturalness in the sense of t Hooft. For example, if a pseudo singlet field develops
a non-vanishing VEV, S3 is broken down to a subgroup Z3 . In the limit of other VEVs being
zeros, low-energy effective theory still has the residual Z3 invariance. In fact, for realistic cases,
some S3 doublets have non-zero VEVs in order to give non-trivial flavor structure. Then the
maximal residual subgroup is S2 . Notice here that the S2 -invariant forms of VEVs depend on the
263
group basis. It is found from (2.3) and (2.17) that such technically natural VEV of S3 doublet is
proportional to (1, 1)t in the complex basis and (0, 1)t [or (1, 0)t ] in the real basis. The former
is available to realize large flavor mixing and the latter is useful for generating textures with
vanishing elements. Since the flavor symmetry is completely broken at low energy, the group
basis might be chosen so that symmetry-breaking forms seem as natural as possible. Finally,
there is another form of doublet VEV which is often utilized in the literature. That is (x, 1)t with
x 1, which arises from a linear combination of the above two types of VEVs, but it is nothing
but parameter tuning.
(iv) Extra symmetries: The S3 flavor symmetry does not fully explain the mass hierarchies
of quarks and leptons. A hierarchical order of S3 -breaking VEVs requires different order of
couplings whose origin is generally unclear. An attractive way to dynamically justify hierarchical couplings is to introduce extra symmetries into the theory. As for Yukawa couplings,
a well-known example is the FroggattNielsen mechanism [19] where extra symmetry controls
the orders of Yukawa couplings. They are generated in low-energy effective theory via decoupling heavy fields, i.e. higher-dimensional operators whose coefficients are naturally given by
the fundamental scale of the theory.
In the following, we present several examples of mass matrices which stem from the S3 flavor
symmetry. In almost the cases, three-generation fermions belong to non-trivial triplet representations: two of them make up an S3 doublet and the other is a (pseudo) singlet. On the other
hand, appropriate representations of Higgs fields are chosen to have non-vanishing elements in
S3 -invariant Yukawa matrices. Mass textures will be expressed in the complex basis unless particularly mentioned. The expressions in the other bases are easily obtained by the unitary basis
rotations defined in the previous section.
3.1. A single S3
First, we study the case that both left-handed fermions Li and right-handed ones Rj
(i, j = 1, 2, 3) transform under a single S3 group. This case is also straightforwardly applied
to Majorana mass terms with the identification L = R . In the following, we take a charge assignment that (L1 , L2 ) and (R1 , R2 ) are S3 doublets and the other fermions are in (pseudo)
singlet representation. This does not lose any generalities since mass matrices for other charge assignments are obtained by flavor rotations. As for Higgs fields, all the three types of irreducible
representations are available for giving non-zero matrix elements; a doublet HD = (H1 , H2 ),
a pseudo singlet HA , and a singlet HS . Here HD,S,A are interpreted as either SU(2)W -doublet elementary scalars or products of some numbers of fields [see, e.g. the comment (iii) Higgs profiles
in the above].
3.1.1. The general case
Dirac mass terms flip the chirality of fermions. Given that L3 and R3 are singlets, the most
general S3 -invariant Dirac mass matrix is described as
LDirac = Ri Mij Lj + h.c.,
aHS + a HS + bHA + b HA
M =
cH1 + c H2
eH2 + d H1
(3.1)
cH2 + c H1
aHS + a HS bHA b HA
eH1 + d H2
dH1 + e H2
dH2 + e H1 ,
f HS + f HS
(3.2)
264
where a, a , b, . . . , f, f are independent coupling constants. The generic form of mass matrix
(3.2) is simplified if HD,S,A belong to complex representations of other groups than S3 : either
Hx or corresponding Hx is dropped out in each element of the generic matrix. This is indeed the
case for the SU(2)W Higgs doublets in the standard model. Another charge assignment is that the
pseudo singlet representation 1A is adopted for the third-generation fermions. For example, when
L3 is a pseudo singlet, the third column of (3.2) is modified so that H2 and HS are replaced
with H2 and HA , respectively.
The general form of Majorana mass matrix is described by identifying L = R . It is
noted that the tensor product (2.7) should be used in constructing Majorana mass term in the
complex basis. Thus the S3 invariance leads to
1
LMajorana = ic Mij j + h.c.,
2
aH1 + bH2 cHS + c HS

M = cHS + c HS aH2 + bH1
dH2 + eH1 dH1 + eH2
(3.3)
dH2 + eH1
dH1 + eH2 .
f HS + f HS
(3.4)
Similar to the above case of Dirac masses, either Hx or Hx is removed in each matrix element
when Hx has some (complex) quantum charge other than that of S3 . Due to the difference of S3
tensor products used for Dirac and Majorana mass terms, the resulting flavor structures in (3.2)
and (3.4) are rather different. This fact could provide an interesting possibility for realistic model
construction of flavor.
3.1.2. Supersymmetric case
Supersymmetry invariant fermion masses come from the superpotential, which is a holomorphic function of chiral superfields. That is, superpotential terms are written in terms of chiral superfields, which contain fermionic components with a certain chirality, e.g. left-handed fermions.
In this case, right-handed fermions are introduced as charge conjugations of left-handed ones. As
in the general case, suppose that the chiral superfields of first two generations, (L1 , L2 ) and
(R1 , R2 ), are S3 doublets, and R3 , L3 are singlets. It should be noted that this assignment
leads to the right-handed fermions (R1 , R2 ) being an anti-doublet. Therefore the product (2.7)
is applied to constructing S3 -invariant superpotential. The most general supersymmetric mass
term is thus given by
WDirac = Ri Mij Lj ,
aH1
bHS + cHA
M = bHS cHA
aH2
eH2
eH1
dH2
dH1 .
f HS
(3.5)
(3.6)
In the case that L3 is assigned to a pseudo singlet, one needs the replacement H2 H2 and
HS HA in the third column of (3.6). Similarly, the S3 -invariant Majorana mass term is easily
found by symmetrizing the matrix (3.6), namely, by setting the couplings as c = 0 and d = e.
This does not significantly modify the flavor structure, unlike in the non-supersymmetric case.
3.1.3. Relation to the democratic mass matrix
We have descried the mass matrices in the complex basis of the S3 group. Let us here comment on the relation to the so-called democratic form of mass matrix in which all the matrix
elements have equal magnitude. It has a simple S3 derivation in the democratic basis defined
265
in Section 2.3.1. As we explained before, these two group bases are connected by the unitary
rotation U in (2.13). A simple way to recover the democratic mass matrix is to introduce only an
S3 -singlet Higgs field HS . As seen from (3.2), that corresponds to a flavor-diagonal matrix in the
complex basis. By rotating the matrix to the democratic basis, it turns out to be
1
1 1 1
h S ,
M = 1 1 1 hS +
(3.7)
1
1
1 1 1
where hS = [(f a)HS + (f a )HS ]/3 and h S = aHS + a HS . It is not hard to see the S3
invariance of the above two matrices in the democratic basis. The democratic mass matrix is thus
found to be derived from an assumption that only one matrix element is dominant in the complex
basis. Therefore a single S3 symmetry cannot ensure the flavor democracy: even with a single
HS field, the democratic ansatz is generally disturbed. As seen from the representation matrices
(2.14), the democratic basis deals with the three indices of a (pseudo) triplet equivalently, and is
not suitable to discuss the decomposition to irreducible representations.
3.2. S3L S3R
For controlling chirality-flipping operators, one can utilize a flavor symmetry under which
left-handed and right-handed fermions transform separately, that is, the S3L S3R symmetry. Without loss of any generalities, we assume that the first two generations (L1 , L2 ) and
(R1 , R2 ) belong to the doublet representations of S3L and S3R , respectively. Since there exist
three irreducible representations for the S3 group, fermion bilinear terms transform in nine ways
under S3L S3R . The corresponding nine types of scalars which ensures the flavor invariance
are denoted by Hij (i, j = D, S, A), where an obvious notation has been used, e.g. HDA means a
doublet under S3L and a pseudo singlet of S3R . As mentioned before, the symbols Hij stand for
either SU(2)W -doublet elementary scalars or products of some fields with appropriate charges.
3.2.1. The general case
For an illustration, we assume that the third-generation fermions L3 and R3 are singlets of
S3L and S3R , respectively. In case that they are pseudo singlets, the subscripts S of Higgs fields
should be replaced with A in the following expressions. The most general Dirac mass matrix is
made symmetry invariant by introducing HDD , HDS , HSD , and HSS , and is given by
a(HDD )11 + b(HDD )12
+ c(HDD )21 + d(HDD )22
M = a(HDD )21 + b(HDD )22

+ c(H ) + d(H )
DD 11
DD 12
)
g(HDS )2 + h(HDS
1
(3.8)
a(HDD )12 + b(HDD )11

) + d(H )
+ c(HDD
22
DD 21
e(HSD )1 + f (HSD )2
a(HDD )22 + b(HDD )21

) + d(H )
+ c(HDD
12
DD 11
)
e(HSD )2 + f (HSD
1
)
g(HDS )1 + h(HDS
2
j HSS + j HSS
(3.9)
where
are the coupling constants. The Higgs fields either with or without asterisks
are dropped out in each matrix element if they have some quantum numbers of other symmetries
than S3L S3R . The possible Majorana mass term of L (or R ) is written in the same way as
(3.4) by including other H s with appropriate charges.
The Dirac mass term in supersymmetric theory is given by the superpotential which is a analytic function of superfields with definite chirality. If we take the same flavor charge assignment
a, b, . . . , j, j
266
as above for the three-generation superfields Li and Rj , the most general form of mass matrix is
WDirac = Ri Mij Lj ,
a(HDD )22 a(HDD )21
M = a(HDD )12 a(HDD )11

c(HDS )2
c(HDS )1
b(HSD )2
b(HSD )1 .
dHSS
(3.10)
(3.11)
3.2.2. Examples
It is found from (3.9) that the general form of Dirac mass term is rather complicated. Combining with simple assumptions, we here present several examples where phenomenologically
interesting forms of mass matrices are obtained, in particular, by choosing relevant Higgs contents.
The first example is the introduction of a single elementary scalar H in the (pseudo) singlet
representation of both S3L and S3R symmetries. According to the S3 charges of L3,R3 , a relevant
representation of the scalar becomes that of Hij with i, j = S or A. In any case, only the 33
element of mass matrix is allowed;
M =
(3.12)
aH
If turning to the democratic basis, we have
1 1 1
1
M = 1 1 1 aH.
3
1 1 1
(3.13)
Both of these matrices have only one non-zero eigenvalue and the same mass spectrum. However
the resulting flavor mixing structures are clearly different: the matrix (3.13) induces large flavor
mixing, while (3.12) has no mixing between the third and the first-two generations. As we noted
before, this may be a physical difference because, in realistic models, there exists more than one
sector whose relative basis differences do affect flavor structure such as the quark mixing angles.
The second example is not to consider that elementary scalars Hij give all trilinear Yukawa
couplings, but to work with some products of scalars in lower-dimensional representations. As
an example, let us introduce three scalar fields HD = (H1 , H2 ), HD = (H1 , H2 ) and HS whose
representations of (S3L , S3R ) are (2, 1S ), (1S , 2), and (1S , 1S ), respectively. With this field content
and the matter representation as above, all entries in the Dirac mass matrix can be filled up with
non-renormalizable operators. In supersymmetric theory, the mass matrix is now given by
aH2 H2 aH1 H2 bH2

M = aH2 H1 aH1 H1 bH1 .
(3.14)
cH2
cH1
dHS
On top of economical field content, this example has several useful properties for constructing flavor models of quarks and leptons. First, the matrix (3.14) has a vanishing determinant and
therefore provides a compelling dynamical reason for the observed tiny masses of first-generation
fermions. Secondly, the effective Yukawa couplings for the first-two generations become naturally small. This is because the S3 invariance requires that they come from higher-dimensional
267
operators suppressed by some large mass scale. Thirdly, the hierarchical flavor structure is easily attained with a smaller number of scalar VEVs. It might often lead to some relations among
mass eigenvalues and mixing angles of quarks and leptons. One can make use of these interesting
features with a smaller number of representations of scalar fields than (3.9).
3.3. Singlet flavor
We have so far discussed the case that both left- and right-handed fermions have non-trivial
charges of discrete flavor symmetry. In this subsection, we comment on a possibility that either
left- or right-handed generations is insensitive to flavor transformation.
Let us consider a single S3 symmetry under which left-handed generations transform nontrivially but right-handed ones do not. In a similar way to the discussion in Section 3.1.1, the
generic types of representations, HD = (H1 , H2 ), HS , and HA , are taken into account. If one
takes a charge assignment that (L1 , L2 ) is a doublet, the most general Dirac mass matrix is
given by
aH2 + bH1 aH1 + bH2

M = cH2 + dH1 cH1 + dH2
eH2 + f H1 eH1 + f H2
gHS + g HS
hHS + h HS .
j HS + j HS
(3.15)
(3.16)
In each matrix element, a symbol H either with or without asterisk should be dropped if H is
accompanied with some charges other than that of S3 . In supersymmetric theory, the mass matrix
is simplified to
aH2 aH1 gHS

M = cH2 cH1 hHS .
(3.17)
eH2 eH1 j HS
If only a singlet Higgs field exists, that results in
g g g
(3.17) h h h HS
j j j
(3.18)
in the democratic basis. This form of Dirac mass matrix has been discussed in the so-called
lopsided models. Similar results are also obtained with the 1A representation instead of 1S .
4. Mass textures from S3
On the prescription described in the previous sections, we will perform several constructions
of mass textures by use of flavor S3 symmetry. That includes well-established forms of mass
matrices which have been discussed in the literature. In realistic flavor models for quarks and
leptons, some combinations of texture forms are usually assumed, and it is therefore meaningful
to examine whether they can be simultaneously reproduced by horizontal symmetry.
It is found in Section 3 that the key ingredients for S3 model building are to define matter and
Higgs profiles and, if needed, to introduce additional symmetry. Among such further symmetries,
we will focus in this work on U (1) flavor symmetry to dynamically realize fermion mass hierarchy [19,20]. It is, however, highly non-trivial to assign relevant U (1) quantum numbers. This is
268
because an S3 doublet, e.g. the first and second generations in the above matrix examples, must
have the same U (1) charge and it does not provide mass hierarchies between them. Moreover,
a naive charge assignment often leads to a prediction that classical mass hierarchy realized by
flavor symmetry is disturbed by higher-dimensional operators. These facts are typical features of
S3 flavor models accompanied by additional U (1). In models without discrete flavor symmetry,
U (1) charge assignment generally has wider flexibility [20]. On the other hand, models with
only S3 symmetry have to do with fermion mass hierarchy by arbitrarily tuning parameters such
as Yukawa couplings and Higgs VEVs. In the following, we illustrate several examples of mass
textures derived from S3 and assistant U (1) flavor dynamics, and also mention how to cure the
above problems of charge assignment. It is noticed that mass hierarchy is dynamically realized
in a similar way with a discrete subgroup of the flavor U (1) symmetry. For example, all the
hierarchical mass textures we discuss below can also be obtained by ZN subgroup with appropriate (enough large) N and the same quantum numbers as in the U (1) case. A smaller choice
of N would be possible and interesting from a viewpoint of brevity. In this case, the problem of
fermion masses can be handled with flavor symmetries that are entirely discrete.
In what follows, we consider the cases that the flavor symmetries are broken at some highenergy scale, and renormalization-group running down to low energy should be taken into
account if one obtains precise values of coupling constants, once the model below the symmetrybreaking scale is specified. However, it is not hard to see that, in all examples we discuss below,
the running effects do not change qualitative results and can safely be dropped. The analysis
in this section is performed in the models with a single flavor S3 symmetry and the results are
described in the complex basis, unless we particularly mention it.
4.1. Nearest neighbor form
The first example of S3 models includes the mass texture proposed by Fritzsch [1]. We assign
non-trivial S3 charges to three-generation left- and right-handed fermions Li and Rj (i =
1, 2, 3). They are given by (L1 , L2 ) + L3 and (R1 , R2 ) + R3 , namely, the first and second
generations make up S3 doublets and the third ones are singlets.
For the Higgs profile, we introduce an S3 -doublet HD = (H1 , H2 ) as only a scalar with
non-trivial S3 charge. As mentioned in Section 3, HD has two possibilities concerned with
the electroweak charge. We here take HD as a singlet of the electroweak gauge symmetry,
and accordingly utilize the usual SU(2)W -doublet Higgs h. Advantages of this choice are the
suppression of flavor-changing rare processes, the preservation of gauge coupling unification in
supersymmetric theory, and so on. A key point is that the VEV of the S3 doublet scalar is assumed to take the form HD = (H1 , 0). This form can be dynamically justified by analyzing
the scalar potential, as we will show in later section. Another type of VEV, HD = (0, H2 ),
just gives the case easily found by exchanging the up and down components in S3 doublets.
The next step is to impose extra symmetries. As a simple example, we here assume a Z2
parity acting non-trivially on S3 doublets (doublet parity). We also have flavor U (1) symmetry
in order to control mass hierarchy and mixing among the three generations. Notice that due to
the assignment of these non-vanishing charges the resultant mass matrices are simplified. That
is, either a Higgs field or its complex conjugate can appear in each matrix element. At this stage,
the 3 3 mass matrix for L,R is found to be
a
a
(4.1)
bH1 h,
cH1
d
269
where a, b, c, and d are the O(1) coefficients which include coupling constants and, if any, scalar
fields with non-vanishing VEVs. Throughout this paper, O(1) coefficients are denoted where the
fundamental mass scale of theory is taken to be unity. The texture form (4.1) was first adopted by
Fritzsch [1] for quark mass matrices. It is noticed that the parity leads to some of vanishing matrix
elements at tree level. An alternative parity assignment is viable where L3 , R3 and HD have
negative parity. A more interesting possibility is to incorporate supersymmetry in the model.
In this case, vanishing matrix elements could be due to the holomorphicity of superpotential,
combined with U (1) quantum charges for realizing mass hierarchy, i.e. supersymmetric texture
zeros.
To generate fermion mass hierarchy, we use the flavor U (1) symmetry whose quantum
charges are assigned as in Table 2.
We have introduced a scalar field to describe symmetry-invariant higher-dimensional operators which effectively induce mass terms below the U (1) breaking scale . The U (1) charges
are normalized by letting take a unit charge, and h is assumed to be neutral in order to fix
overall mass scale. As a result, the orders of magnitude of the matrix elements in (4.1) becomes
x+y

x+y
(4.2)
x +y+z+ h,

x+y +z+
x +y
where we have defined and H1 ( 0 for a consistent theory below the fundamental scale).
The eigenvalues of (4.2) are easily found; m1 hx+y2(z+) , m2 hx+y+2(z+) , and

m3 hx +y . To obtain a hierarchy m1 m2 , the condition z + < 0 is naively needed.
It should be, however, noticed that this condition causes a problem that higher-dimensional
operators involving HD give larger effects and are not necessarily negligible. We will in the
below enumerate possible resolutions to the problem; (i) to set z + > 0, (ii) to take z < 0
and make higher-dimensional operators negligible by supersymmetry, (iii) to tolerate some of
higher-dimensional operators, (iv) to impose extra symmetry, (v) to change the S3 representations of fermions, and (vi) to introduce S3 -singlet scalars.
(i) If one chooses z + > 0, problematic non-renormalizable operators become irrelevant.
This may be the simplest solution in the viewpoint of U (1) charge assignment. It is found from
the matrix form (4.2) that the mixing angle between the first and second generations becomes
/2. Note here that the S3 freedom cannot be used to reorder the mass eigenvalues because the
S3 group basis has already been fixed such that HD takes a particular form of VEV. While we
know the generation mixing in the quark sector is small, the S3 realization of the Fritzsch ansatz
can be consistent (for z + > 0), provided that the up and down quark sectors have almost the
same flavor structure. Namely, If both the up and down sectors employ the matrix form (4.2), the
label exchanging effects are cancelled out between the two sectors, and the quark mixing angles
become small of powers of . Such a situation is similar to the case of the democratic quark mass
matrices [14].
Table 2
U (1)
Z2
L1,2
L3
x
+
R1,2
R3
HD
0
+
1
+
270
(ii) If one chooses z < 0, higher-dimensional operators can be forbidden by holomorphicity

in supersymmetric theory. For this being achieved, it is a simple assignment that all the U (1)
charges except for HD and are positive in order not to give any higher-dimensional terms.
Since the doublet scalar HD now develops a VEV in the up component, the down component of
doublet in the product (HD )2 becomes non-zero, and the singlet component in (HD )3 does [see
the tensor product (2.7)]. Consequently, higher-order terms involving the products of HD might
spoil the first-order analysis, because HD has a negative U (1) charge (z < 0). In fact, the 13,
31, and 22 elements in the matrix (4.1) receive non-negligible contributions from (HD )2 . The
contributions to the 13 and 31 elements are forbidden by the doublet parity. We however find
that higher-order contribution to the 22 element is difficult to be suppressed as long as the 23
and 32 matrix elements are allowed. A simple way to remedy this last problem is to add some
extra symmetry, otherwise to apply the option (iii).
(iii) One, in some sense, negative choice is to abandon the exact Fritzsch ansatz and put
up with some contribution from non-renormalizable operators. As mentioned above, a nonvanishing 22 element often appears even if the parity invariance is imposed. One can therefore
choose as the third option that higher-order terms are generally forbidden by parity and/or supersymmetry while the 22 element is not. For example, in supersymmetric approach, the U (1)
charges are needed to satisfy the mild constraints x + y + z < 0, x + y + z 0, x + y + z 0,
and z < 0. If this is the case, we obtain a mass matrix of the form (4.1) corrected by a non-zero
22 element from a higher-dimensional operator involving (HD )2 . Such a type of mass texture
has been recently discussed [21] to be suitable for solving fermion mass problems including the
neutrino physics.
(iv) If one chooses to introduce more additional symmetries, harmful higher-dimensional operators might be removed. However such an operation generally reduces to complicate the models
and involve uncontrollable factors, which make the models unfavorable.
(v) Contrary to the above options (i)(vi), one can choose to extend the model to include
more fields in other representations of flavor symmetry. Let us consider an additional Higgs
HA of pseudo S3 singlet. The representations of matter fields are accordingly changed to
(L2 , L3 ) + L1 and (R2 , R3 ) + R1 , i.e. the second and third generations are S3 doublets. It is easily found by constructing S3 -invariant terms that one still has a mass matrix of the
Fritzsch ansatz, provided that the non-vanishing VEVs are given by HD = (0, H2 ) and HA .
An important difference between this and the above models is whether the parity symmetry is
needed or not to suppress undesired matrix elements. In the model here, the Fritzsch ansatz is
obtained without imposing any parities. To make the 11 element negligible, it is sufficient to
take appropriate U (1) charge assignment, since the first-generation fermions now belong to different S3 representations from the others, and have different U (1) charges. Note that the example
here may not be applied to leftright symmetric cases due to the anti-symmetric matrix elements
generated by a pseudo singlet (i.e. M23 = M32 ). However asymmetrical forms of mass texture
(zeros) could provide phenomenologically interesting possibility.
(vi) Another choice of additional scalar fields is a singlet Higgs HS . A reason to introduce
such singlet scalars is to suppress bare Yukawa couplings [e.g. a and d in the matrix (4.1)]. They
are induced from other operators involving S3 singlet scalars with non-vanishing U (1) charges.
Once the singlet scalars obtain VEVs, the resultant mass matrix takes the same form as (4.1),
but flavor-invariant non-renormalizable operators are different due to the non-vanishing charges
271
of singlet scalars.2 Let s be the U (1) charge of HS , which must be positive as will be seen
below. There are three possible ways to include the HS scalar: (a) The 33 element comes from
an operator involving HS . That needs the charge conditions x + y < 0 and x + y + s 0.
Moreover, for the mass hierarchy being realized, an additional condition 2(z + ) < s + s
is required, where HS s . It is interesting that the operators which contribute to the 22
element is automatically suppressed, and one does not need to rely on any additional symmetries.
Furthermore if one chooses s + s < 52 (z + ) [or s + s < 3(z + )], corrections to the 13 and
31 [or 11] elements become negligibly small. It can be checked that all the above charge
conditions are easily satisfied. (b) In this case, the charge conditions x + y < 0 and x + y + s 0
are needed for HS to appear in the 12 and 21 elements. A proper mass hierarchy is realized
when 2(z + ) < s + s , and the 13 and 31 elements are negligibly small if s + s < 3(z + ).
Unlike in the case (a), these charge conditions do not suppress higher-dimensional contributions
to the 11 and 22 elements, so some symmetry should be imposed. (c) All the 12, 21, and
33 elements contain the HS field. That requires x + y < 0, x + y < 0, x + y + s 0, and
x + y + s 0. Moreover one should take z + < s + s in order for fermion mass hierarchy to
be preserved. With these charge conditions, all the operators concerned with the 22 element are
automatically suppressed. It is, however, found that the 11 element generally receives sizable
contribution unless extra symmetry is imposed.
In this way, the nearest-neighbor form of mass matrices, including the well-known Fritzsch
ansatz, is realized in the framework of S3 flavor symmetry. Moreover the mass hierarchy among
the three generations is also achieved by incorporating an additional U (1) symmetry. An important and non-trivial issue is whether U (1) charges can be assigned so that they are compatible
with the non-Abelian flavor symmetry. We have shown typical examples of assignments which
not only produce mass hierarchy but also suppress non-renormalizable operators which tend to
disturb the mass hierarchy in the first-order estimation.
4.2. Next-nearest neighbor form
In most models with S3 flavor symmetry and also in the previous subsection, the first two light
generations are assumed to compose S3 doublets. While such an assignment may be favorable to
some phenomenological issues e.g. for suppressing flavor-changing processes, it is not necessarily the unique choice for S3 charges. One may easily imagine different, but somewhat unfamiliar,
S3 flavor structures. In the following, we show that S3 models with such twisted generations are
also relevant to constructing realistic flavor theory of quarks and leptons. As a simple example,
we here discuss the ansatz for mass texture proposed in [3], which type of texture suggests that
the first and third generation fermions make up S3 doublets.
The procedure is completely parallel to that in the previous subsection. The key ingredients
are the representations of matter and Higgs fields, the profile of Higgs VEVs, and additional
symmetries. Consider the matter representation (L1 , L3 ) + L2 and (R1 , R3 ) + R2 , i.e.
the first and third generations are S3 doublets. We also introduce an S3 -doublet Higgs scalar
HD = (H1 , H2 ), whose down component H2 is assumed to develop a non-vanishing VEV. The
electroweak gauge invariance is implemented by an SU(2)W -doublet scalar h, which is a singlet
2 In case that SU(2) -doublet Higgses belong to non-trivial representations of S , such effects of H (the suppression
W
S
3
of bare Yukawa terms) can always be taken into account.
272
of S3 . The setup leads to the following form of mass matrix;
h,
(4.3)
b
a
cH2
where 12 and 21 elements have been suppressed by imposing a Z2 parity concerning the second generation; L2 , R2 L2 , R2 . Similar suppression can also be obtained by use of
holomorphicity in supersymmetric theory. At this stage, the coefficients a, b, and c are supposed
to contain coupling constants and scalar VEVs, whose natural sizes are O(1). The texture (4.3)
has the form of the Giudice ansatz [3] for the up quark mass matrix. The coefficients a and b
should be smaller than cH2 to properly describe mass hierarchy of fermions. That is realized
in the present work by introducing flavor U (1) symmetry and a charge-compensating scalar .
Defining U (1) quantum charges as in Table 3, we have the orders of magnitude of the matrix
elements
x+y

h,
(4.4)
x +y
x+y
x+y+z+
where , and parametrizes the VEV of S3 doublet H2 . It is obvious that a

charge condition z + < 0 is necessary for mass hierarchy without inducing maximal generation
mixing. Notice that this is similar to the case of the Fritzsch ansatz where higher-dimensional operators involving the S3 doublet scalar lead to significant modification of matrix form. Therefore
also in the present case, one can apply the resolutions discussed in the previous subsection to
have natural hierarchy of mass eigenvalues and mixing angles.
4.3. Asymmetric textures
A more unfamiliar but interesting case is that flavor charges are asymmetrically assigned to
left- and right-handed fermions, which generally lead to asymmetrical forms of mass textures. In
grand unification schemes, the mass matrix of up-type quarks is often assumed to be symmetric
(exactly speaking, hermitian), which comes from the fact that, even in the minimal SU(5) model,
one-generation up-type quarks with both chiralities belong to a single multiplet of unified gauge
symmetry. However this is not generally the case for other Dirac mass matrices. In particular,
the present experimental data suggest that the leptonic flavor mixing is quite un-parallel to the
quark mixing. This asymmetrical observation can be compatible with quarklepton unification,
if fermion mass textures take asymmetrical forms in the generation space [10]. There are also
some classes of non-Hermitian ansatze for quark mass matrices [22], which are consistent with
the experimental data and cannot be transformed to the symmetric solutions previously found in
Ref. [4]. It is therefore worthwhile to investigate the dynamical realization of asymmetric mass
textures with discrete flavor symmetry. A systematic study of asymmetric mass matrices for the
up and down quarks has recently been performed, particularly paying attention to the connection
to leptonic flavor mixing [23].
Table 3
U (1)
Z2
L1,3
L2
x
+
R1,3
R2
HD
y
+
0
+
z
+
1
+
273
Among various types of viable asymmetric textures, we here present an S3 flavor model which
predicts the mass matrix proposed in [24]. This texture is relevant to the neutrino Dirac mass matrix and is interesting in that large lepton mixing is realized without any tuning of couplings, if
there is a suitable hierarchy among right-handed neutrino masses. The texture indicates neither
that the first-two generations are in doublet representation nor that left- and right-handed fermions have parallel assignments of S3 charges. That leads us to consider highly asymmetric flavor
structure: (L1 , L3 ) + L2 and (R2 , R3 ) + R1 . We also have SU(2)W -singlet scalars, HS
and HD = (H1 , H2 ), which are S3 singlet and doublet, respectively. The down component H2 of
the doublet HD is assumed to develop a non-vanishing VEV. The electroweak gauge invariance
is maintained by introducing an SU(2)W -doublet standard model Higgs h, which is a singlet of
S3 . Consequently we obtain the following form of mass texture:
d
aH2
(4.5)
bH2 eHS h.
eHS
cH2
It is interesting to note that no additional symmetry is required to eliminate unwanted non-zero
matrix elements, contrary to the previous two examples. The coefficients a, . . . , e contain coupling constants whose natural sizes are O(1). Splitting the sizes of the coefficients is easily
obtained, for example, by introducing flavor U (1) symmetry and a charge-compensating scalar .
A typical U (1) charge assignment is given in Table 4 with x > y 0 and z 0, and we obtain
the mass texture of the form
z
z
(4.6)
x+y x+y h.
x+y
y
( )
The parameter denotes the ratio of the VEV to the fundamental scale. Compared to the
matrix in [24], a non-vanishing 31 element is generated. However one can easily find that such
a tiny entry does not contribute both to the mass eigenvalues and mixing angles, though it is
obliged to be included to respect the flavor symmetry.
It was discussed [24] that the texture (4.6) (without the 31 element) could be derived by
making use of more than two continuous flavor symmetries. By contrast, the present approach
with discrete flavor symmetry seems somewhat simpler. The texture form is totally controlled by
a single S3 symmetry, and then, the mass hierarchy is determined by extra assumption (here the
U (1) flavor symmetry).
Finally let us comment on the corrections from non-renormalizable operators. It is found that
the above charge assignment suppresses the contribution of higher-dimensional operators to the
13 and 32 matrix elements, and the texture zeros are not destabilized. On the other hand, the
correction to the 21 element could be as large as the other components in the second rows, i.e.
O(x+y ). This result itself is not a disaster but the above zero texture should be modified. That
is similar to the cases in the previous subsections where higher-dimensional operators involving
HD could give significant effects on the first-order approximation. Therefore one is able to apply
Table 4
U (1)
L1,3
L2
R2,3
R1
HD
HS
2x
274
similar resolutions to remedy the problem. For example, if the S3 doublet scalar HD has a nontrivial charge of some other symmetry, any problematic deformation of mass matrix is completely
avoidable.
4.4. Missing a right-handed neutrino
Referring to the S3 algebra presented in Section 2, the maximal irreducible representation
is doublet. The triplet representation is reducible, while it may be suitable to describe threegeneration fermions. Two fermions with the same standard model charges are also well handled
with S3 symmetry. While we have already known the existence of three repetitions of the standard
model fermions, there is an idea that only two right-handed neutrinos are effectively included
[25]. In this framework, a missing right-handed neutrino might be thought to decouple at a superheavy scale and regarded as a flavor singlet.3 The remaining two right-handed neutrinos are
hence treated as a doublet. In this subsection, we present a model with three left-handed and two
right-handed neutrinos in the framework of S3 flavor symmetry.
We would like here to stress that another important issue of flavor symmetry is to simultaneously control different sectors of flavor. For example, physical quark mixing angles are
determined with a conspiracy of the up and down mass matrices, and low-energy Majorana
neutrino masses are derived from Dirac and right-handed Majorana masses with the seesaw
mechanism [26]. In the following, we illustrate simultaneous treatment of two different types
of mass textures, i.e. the Dirac and Majorana masses of neutrinos. A more realistic flavor model,
including quarks and charged leptons, will be discussed in the next section.
Among various models with two right-handed neutrinos, we here focus on the texture ansatz
proposed in [27] and its realization with S3 flavor symmetry in the neutrino sector. The model
contains three left-handed leptons L1,2,3 and two right-handed ones R1,2 . We assign the flavor
charges to these fermions such that (L1 , L2 ) and (R1 , R2 ) are S3 doublets and L3 a singlet
(either of the S3 singlet representations is possible). As for the Higgs sector, at least two types
of scalars are introduced for controlling two different sectors of mass textures, though not necessarily required. We take, as an example, two scalar fields which are S3 doublets and the standard
gauge singlets; HD = (H1 , H2 ) and HD = (H1 , H2 ), whose VEVs generate Dirac and Majorana
masses, respectively. For compensating electroweak gauge invariance, a usual SU(2)W -doublet
Higgs h is also included. A more economical choice might be to have SU(2)W -doublet and S3 doublet scalar fields. In this case, however, there may exist some problems, as discussed before,
that gauge coupling unification must be non-trivially realized and large rates of flavor-changing
processes generally spoil the models.
The S3 invariant terms induce the neutrino Dirac mass M and the right-handed Majorana
mass MR , which read from the generic expressions (3.2) and (3.4),

aH2 + bH1 cH1 + dH2
h,
M =
(4.7)
aH1 + bH2
cH2 + dH1

eH1 + f H2
,
MR =
(4.8)
eH2 + f H1
where a, . . . , f are the coupling constants. Thus we have obtained the textures discussed in [27],
if the generation indices are properly exchanged while physical consequences are unchanged
3 Such a singlet should be 1 to preserve S symmetry in low-energy regime. The existence of the third right-handed
S
3
neutrino would be needed to reconcile the lepton sector to the quark one in high-energy unification schemes.
275
(except for a sign reversion of as yet unobserved leptonic CP-violating quantity). The bi-large
generation mixing in the neutrino sector is naturally established, provided that (M )12 (M )13
and (M )21 (M )23 , and that there is a little hierarchy between these two combinations or
between the mass eigenvalues of MR [27].
It is rather straightforward to incorporate supersymmetry into the above picture. Every field
is promoted to a superfield which is in the same representations of the standard model gauge
and flavor symmetries. The Dirac and Majorana mass terms are described by superpotential.
Notice that the right-handed neutrinos Ri are embedded into the corresponding superfields in
the form Rc i . As we explained in Section 3, this embedding modifies the tensor product and
slightly changes the texture form. We thus find that the S3 -invariant superpotential induces the
mass matrices

bH2
aH1
cH1
,
MR =
.
M =
(4.9)
aH2 bH1
cH2
Again the neutrino mass textures presented in [27] are dynamically realized in a supersymmetric
S3 model. It is easy to see that the coupling constants and scalar VEVs can be appropriately
chosen for the neutrino physics without any fine tuning.
5. Grand unification with flavor S3
Based on the above prescription for mass texture, in this section, we present a grand unified
model of quarks and leptons with S3 flavor symmetry. It is the most important point of flavor symmetry that any matrix form from flavor symmetry is valid only in the case that there are more than
two sectors governed by the symmetry. A well-known example is the up and down quark mass
textures whose forms are simultaneously altered by flavor rotation of SU(2)W -doublet quarks. In
other words, any realization of zero textures for a single sector cannot be physically distinguished
from other freely-rotated, generally complicated, matrix forms. In the previous subsection, we
discussed a model in which two types of neutrino mass matrices are controlled in the same fashion. To include the charged lepton sector is straightforward, referring to the general discussion of
S3 -invariant matrix forms. Towards realistic flavor models of quarks and leptons, we here construct an illustrative grand unified model with S3 flavor symmetry. The model has asymmetrical
forms of mass textures due to non-trivial assignment of flavor charges, which particularly leads
to bi-large lepton mixing. While the model presented here is not so complete, it suggests the
validity of discrete symmetry for flavor physics and would give a large step to understand the
origin of flavor.
Let us consider SU(5) as a minimal candidate for unified gauge group. For matter Yukawa
couplings, we introduce as usual the three-generation fermions, i (10), i (5 ), i (1) (i = 1, 2, 3)
and two scalars, hu (5) and hd (5 ). Needless to say, one needs some different types of scalar fields
to break the unified gauge symmetry, but they are generally irrelevant to the Yukawa sector.
We define the flavor S3 charges of matter and scalar fields so that the combinations (1 , 2 ),
(1 , 3 ) and (1 , 2 ) are S3 doublets and the others in the 1S representation. To implement the S3
invariance, a doublet and a singlet scalars are also introduced, called HD and HS , respectively.
The total field content is listed in Table 5. We assume that S3 is broken by the scalar VEVs;
HD = (H1 , 0) and HS , which are of O(1) and can be the result of analyzing symmetryinvariant scalar potential without fine tuning.
While the right-handed neutrinos are assigned to the S3 representations (1 , 2 ) + 3 , that is
not essential as long as the low-energy left-handed Majorana masses are concerned. It is because,
276
Table 5
SU(5)
S3
U (1)
1,2
1,3
1,2
hu
hd
HD
HS
10
2
4
10
1
0
5
2
2
5
1
0
1
2
x +2
1
1
y
5
1
2
5
1
2
1
2
0
1
1
2
1
1
1
in the seesaw mechanism, arbitrary rotations of right-handed neutrinos can be inserted while
preserving left-handed Majorana mass matrix. In fact, the S3 flavor charges (i.e. the flavor basis)
of i turn out to be fixed, once some other dynamics than S3 is incorporated.
In addition to these, we have a U (1) (or ZN ) symmetry to produce natural hierarchies among
matrix elements (and also to help to discriminate i ). A key of the mechanism is the existence
of a scalar field with a non-vanishing VEV, which is a bit smaller than the fundamental scale
of the theory. The whole S3 representations and U (1) quantum numbers are shown in the table
(x 0 and 2 > y 0 for phenomenological requirements).
Let us first discuss the Majorana masses of right-handed neutrinos. They are given by the
following form of S3 -invariant mass operator ic (MR )ij j :
M
,
MR = M
(5.1)

M
where the mass parameters M and M are generally free. They contain some compound factors
such as suppressions by flavor U (1) invariance, but such suppression factors are cancelled out
via the seesaw mechanism and irrelevant for low-energy quantity. The orders of magnitudes of
M and M would be determined by experimental data. Theoretically this seems natural since MR
cannot be directly induced by hu,d , nor come from HD,S , as the Majorana mass term violates
lepton number symmetry. Otherwise one easily obtains MR by assuming VEVs of S3 -singlet
scalar fields with lepton numbers. In any case, the most important point here is that a single S3
flavor symmetry controls both Dirac and Majorana mass textures simultaneously.
We also find that the S3 and U (1) flavor invariance leads to the following Dirac mass textures
Mu,d,e, for quarks and leptons:
6
4
4
6
hd ,
Mu = 6
(5.2)
Md = 4 2
2 hu ,
2
1
1
x+2
x+2
hu ,
Me = MdT ,
(5.3)
M = x+2 x
y
y
where we have neglected coupling constants and defined the expansion parameter as the ratio
of to the fundamental scale. Our first prediction obtained from the above textures is that the
VEV ratio of two SU(2)W doublets should be large, that is, hu / hd mt /mb 60. Secondly, we
find the Majorana mass matrix ML of left-handed neutrinos derived from the seesaw formula

2y h2u
.
ML =
(5.4)
1
1+
M
1+
1
277
Here the parameter has been defined as 2x2y+2 (M /M). The induced matrix elements
in the 11, 13 and 31 positions are negligibly small and have been dropped in the expression
(5.4). It is interesting to note that all the symbols 1 in the matrix (5.4) are exactly 1 due to
the seesaw mechanism, and consequently, the lower-right 2 2 submatrix has a reduced determinant of O(), not O(1). For a small ( 12 ), the matrix (5.4) automatically realizes the
bi-large generation mixing with hierarchical neutrino spectrum. The model also predicts a small
lepton mixing of O() between the first and third generations. The planned improvements in the
sensitivity to such a small angle are expected to reach O(102 ) [28], and the above model will
be testable in near future. On the other hand, the neutrinoless double beta decay [29] cannot be
observed due to the negligible value of (ML )11 .
It is found in the above Dirac mass matrices that some distorted values are induced for light
generations, while the coefficients of matrix elements could be appropriately chosen for reproducing the observed values of fermion masses and mixing angles. We checked that, for the
present simple field content, the flavor charge assignment of S3 and U (1) symmetries is uniquely
determined for realizing the two conditions that (i) the right-handed neutrino masses MR come
from the mass operator (5.1) and (ii) the rank of 2 2 submatrix of ML is reduced via the seesaw mechanism without tuning of couplings. Therefore an improvement of the above toy model
is made by extending it to include additional scalar fields and modifying the texture forms. For
example, the equivalence of the down and charged-lepton mass spectrum may be split by making
use of Higgs fields in higher-dimensional representations. Another simple extension is to change
the S3 group basis, e.g. to the real basis. This might be a possible amelioration because the texture analysis and its physical consequences depend on the group basis of flavor symmetry, as
explained before. We leave these issues to future investigation.
6. Analysis of scalar potentials
We have so far discussed mass textures of fermions in the framework of S3 flavor symmetry,
where the S3 invariance is violated only by scalar VEVs. In particular, the assumption that either
of components in a doublet has a vanishing VEV leads to texture forms for quark and lepton
mass matrices, which result in various types of phenomenologically viable flavor models. In
this section, we present, as an existence proof, the analysis of scalar potentials which generate
desired forms of symmetry-breaking VEVs. We also discuss model parameters and symmetry for
obtaining such VEVs. It should be noted that, unlike in the case of continuous flavor symmetry,
a VEV of doublet cannot be transformed to an arbitrary form by S3 rotations because of the
discreteness of group operations (see Table 1). Even if there is a case that such rotation is useful,
it is physically meaningful only when other symmetries in the theory do not commute with S3 .
This is not necessarily satisfied, for example, in the presence of U (1) flavor symmetry.
6.1. The general case
We consider a single S3 doublet scalar H = (H1 , H2 ) and analyze the structure of its generic
scalar potential. The potential analysis becomes more involved if there are some numbers of
scalar fields. However since we have shown that mass textures of quarks and leptons can be
derived from a single doublet scalar, the analysis for one doublet would provide a step towards
more practical cases with more than one doublets. Furthermore, when several scalar fields have
VEVs of enough separated scales, they can be independently analyzed if the most generic terms
are taken into account in scalar potentials.
278
The most general and renormalizable scalar potential for a doublet H = (H1 , H2 ) is given by

2
2
V (H1 , H2 ) = 2 H1 H1 + H2 H2 + H1 H1 + H2 H2

+ 2 H1 H2 + H13 + H23 + H12 H2 + H22 H1 + h.c.

+ H1 H1 H2 H2 + H12 H22 + H1 H12 H2 + H2 H22 H1 + h.c.
(6.1)
with respect to the S3 invariance in the complex basis. The minimization of the potential leads to
the following two conditions:
0=
0=
V
H1
V
H2
= 2 H1 + 2H1 H12 + 2 H2 + 3 H1 2 + H22 + 2 H1 H2 + H1 H2 H2

+ 2 H1 H2 2 + H12 H2 + 2H1 H1 H2 + H2 2 H2 ,
(6.2)
= 2 H2 + 2H2 H22 + 2 H1 + 3 H2 2 + H12 + 2 H1 H2 + H1 H1 H2

+ 2 H1 2 H2 + H1 H22 + H1 2 H1 + 2H1 H2 H2 .
(6.3)
2 V /Hi Hj
(i, j = 1, 2)
The minimum is ensured by examining whether the 2 2 matrix
has positive eigenvalues at the extreme. Generally, the stationary conditions (6.2) and (6.3) are
satisfied at various points in the field space. The origin H1 = H2 = 0 is a solution of these two
condition, but at this point, the S3 flavor symmetry is unbroken and the result is unrealistic.
Parallel to the well-known electroweak symmetry breaking, one easily avoids this trivial solution
by assuming a negative value of 2 , which in turn makes the origin being a locally maximum
point and destabilized.4 A possible minimum we are interested in is that either of VEVs in a
doublet scalar is vanishing, namely H1 = 0 and H2 = 0, or vice versa. For H2 = 0, the stationary
conditions (6.2) and (6.3) become
2 H1 + 3 H1 2 + 2H1 H12 = 0,
(6.4)
2 H1 + H1 2 + H1 H12 = 0.
(6.5)
It is found that the existence of non-trivial (at least local) minimum with H1 = 0 requires a set of
conditions for the coupling constants. For example, = = = 0 or 2 / 2 = 3/ = 2/,
and so on. The former condition could easily be obtained if the scalar has some other quantum
charges than S3 . There exist various possibilities of coupling constants for stabilizing the scalar
potential at non-trivial minimum. It is an intriguing task to find the conditions of couplings which
are naturally realized with symmetries. Finally we comment on the VEV form H1 = 0, H2 = 0.
Such a form of doublet VEVs is needed to achieve the texture form discussed in Section 4.4. If
the coupling constants in the above scalar potential have the same order of magnitude (in the unit
of fundamental scale), two VEVs H1 and H2 are on a similar order and generally have different
values. In this case, it is difficult to write down the generic solution to (6.2) and (6.3) for the
potential minimum. We numerically investigated that there indeed exists the minimum of the
potential (6.1). As an example, a set of natural values of coupling constants, = = = m
and = = = = 1, leads to the stationary point H1 = 0.61m and H2 = 1.62m. It can
easily be checked that this point corresponds to a potential minimum by analyzing the second
order derivatives of the potential. It is also interesting to explore the region of coupling constants
for which such a form of doublet VEVs appears.
4 Exactly speaking, if 2 < | 2 |, the origin of field space is destabilized, but only along with limited directions.
279
6.2. Supersymmetric case

The analysis is similarly performed in supersymmetric theory. For an S3 -doublet superfield
H = (H1 , H2 ), the symmetry-invariant superpotential W is given by
w

y
W = mH1 H2 + H13 + H23 + H12 H22 + xH1 H2 H13 + H23
3
2
z2
z1 6
H1 + H26 + H13 H23 ,
+
(6.6)
6
3
where m, y, . . . , z1 , z2 are the coupling constants and we have included the most general operators with mass dimensions up to seven. In this subsection, we assume that supersymmetry
is unbroken, that is, supersymmetry-breaking soft terms do not fix the breaking scale of flavor
symmetry. The scalar potential is then given by

2

V (H1 , H2 ) = mH1 + yH22 + wH12 H2 + x 4H1 H23 + H14 + z1 H25 + z2 H13 H22

2

+ mH2 + yH12 + wH1 H22 + x 4H13 H2 + H24 + z1 H15 + z2 H12 H23 .
(6.7)
Comparing to the general case (6.1), we have the relations from supersymmetry:
2 = |m|2 ,
= |y|2 ,
= m y,
= m w,
2 = = = = 0.
(6.8)
We here focus on some simple cases obtained by introducing discrete symmetry on the superpotential. The vacuum is where one of the VEVs of doublet is vanishing, e.g. H = (H1 , 0). Let us
consider a Z3 symmetry of which the superfield H has a charge +1. In this case, it permits only
the y, z1 and z2 terms in the superpotential (6.6). As in the general case without supersymmetry,
the origin H1 = H2 = 0 is a trivial solution without flavor symmetry breaking. We here assume
for simplicity that the origin is destabilized if supersymmetry-breaking (and S3 -breaking) soft
scalar terms are included. We then find a potential minimum where the scalars develop their
VEVs of the form H = ((y/z1 )1/3 , 0). At this vacuum, a vanishing VEV can give rise to mass
textures (zeros) for quarks and leptons as explained before. Another simple example is a discrete R symmetry Z3R under which the superfield H has a charge +1. In this case, only the m
and x terms are allowed in the superpotential (6.6). Similarly assuming that the origin does not
correspond to the minimum, we find a vacuum at which the doublet has the one-sided form of
VEVs H = ((m/x)1/3 , 0). It may be interesting to note that the Z3(R) symmetry is suitable for
discussing trilinear Yukawa terms and also for suppressing non-renormalizable operators involving more than one S3 doublets. We finally comment that the analysis becomes more involved if
higher-dimensional operators in Khler terms were included.
7. Summary and discussions
Discrete flavor symmetry is a powerful instrument in classifying and constructing mass matrix
forms of quarks and leptons. It is greatly anticipated that such discrete symmetry is practically
inherited from continuous symmetries in more fundamental theory in high-energy regime. In
this paper, we focus on the S3 group as a promising candidate of such symmetry. The S3 is
the smallest non-Abelian discrete symmetry and also has a simple geometrical interpretation.
After discussing some fundamental issues of the S3 group (representations, tensor products and
group basis dependence), we have presented the general forms of 3 3 mass matrices which
are derived from various types of S3 theories. Based on that prescription, we have performed
280
the construction of particular ansatze of mass textures which have been often discussed in the
literature. That includes asymmetrical matrices in generation space, which would be appropriate
to discussing the leptonic flavor mixing.
Furthermore, discrete flavor symmetry can control the structure of mass matrices (texture)
in dynamical ways. In particular, we have shown that vanishing matrix elements (texture zeros)
are dynamically realized in the vacuum of scalar potential. It is however noted that texture zeros themselves do not give any explanation for fermion mass hierarchies. The observed values
of masses and mixing angles are obtained in our scheme by introducing a U (1) or ZN symmetry. A non-trivial issue here is whether Abelian charges can be assigned so that they are
compatible with non-Abelian flavor symmetry. We have described typical examples of charge
assignments which not only produce mass hierarchy among the generations but also suppress
higher-dimensional operators which tend to disturb mass hierarchies in the first-order estimation.
Unlike previous approaches with the S3 group, we do not assume hierarchical coupling constants
nor sequential breaking of flavor symmetry. In realistic flavor models for quarks and leptons, it
is important that several different sectors are simultaneously controlled by flavor symmetry. As
an illustrative example, a grand unified model has been constructed with a single S3 symmetry.
Finally, we have also analyzed the scalar potentials for S3 -doublet scalars, whose form of VEV
is a key ingredient of our approach.
In this paper (especially below Section 4), we have focused on the models with a single S3
symmetry. It may be an interesting task to explore other types of flavor extensions, e.g. S3L S3R
and other discrete symmetries larger than S3 . They would generally lead to different types of mass
textures and physical consequences such as flavor-violating processes and their characteristic experimental signatures. More detailed study, including these issues, is left to future investigations.
Acknowledgements
We would like to thank M. Tanimoto for helpful discussion. This work is supported by scientific grants from the Ministry of Education (Nos. 16028214, 16540258, 17740150) and by
grant-in-aid for the scientific research on priority area Progress in elementary particle physics
of the 21st century through discoveries of Higgs boson and supersymmetry (No. 16081209).
References
[1]
[2]
[3]
[4]
[5]
[6]
H. Fritzsch, Phys. Lett. B 73 (1978) 317.

H. Georgi, C. Jarlskog, Phys. Lett. B 86 (1979) 297.
G.F. Giudice, Mod. Phys. Lett. A 7 (1992) 2429.
P. Ramond, R.G. Roberts, G.G. Ross, Nucl. Phys. B 406 (1993) 19.
P.H. Frampton, S.L. Glashow, D. Marfatia, Phys. Lett. B 536 (2002) 79.
Y. Fukuda, et al., Phys. Lett. B 433 (1998) 9;
Y. Fukuda, et al., Phys. Lett. B 436 (1998) 33;
Y. Fukuda, et al., Phys. Rev. Lett. 81 (1998) 1562;
Y. Fukuda, et al., Phys. Rev. Lett. 82 (1999) 2644;
Y. Fukuda, et al., Phys. Rev. Lett. 85 (2000) 3999.
[7] S. Fukuda, et al., Phys. Rev. Lett. 86 (2001) 5651;
S. Fukuda, et al., Phys. Rev. Lett. 86 (2001) 5656;
S. Fukuda, et al., Phys. Lett. B 539 (2002) 179;
Q.R. Ahmad, et al., Phys. Rev. Lett. 89 (2002) 011301;
Q.R. Ahmad, et al., Phys. Rev. Lett. 89 (2002) 011302;
[8]
[9]
[10]
[11]
[12]
K. Eguchi, et al., Phys. Rev. Lett. 90 (2003) 021802;

T. Araki, et al., Phys. Rev. Lett. 94 (2005) 081801.
S.F. King, Phys. Lett. B 439 (1998) 350;
S.F. King, JHEP 0209 (2002) 011;
S. Davidson, S.F. King, Phys. Lett. B 445 (1998) 191.
A. Santamaria, J.W.F. Valle, Phys. Lett. B 195 (1987) 423;
R. Hempfling, Nucl. Phys. B 478 (1996) 3;
F.M. Borzumati, Y. Grossman, E. Nardi, Y. Nir, Phys. Lett. B 384 (1996) 123;
M. Drees, S. Pakvasa, X. Tata, T. ter Veldhuis, Phys. Rev. D 57 (1998) 5335;
E.J. Chun, S.K. Kang, C.W. Kim, U.W. Lee, Nucl. Phys. B 544 (1999) 89;
V. Bednyakov, A. Faessler, S. Kovalenko, Phys. Lett. B 442 (1998) 203;
B. Mukhopadhyaya, S. Roy, F. Vissani, Phys. Lett. B 443 (1998) 191;
O.C.W. Kong, Mod. Phys. Lett. A 14 (1999) 903;
J. Ferrandis, Phys. Rev. D 60 (1999) 095012;
D.E. Kaplan, A.E. Nelson, JHEP 0001 (2000) 033;
A.S. Joshipura, S.K. Vempati, Phys. Rev. D 60 (1999) 111303;
A. Abada, M. Losada, Nucl. Phys. B 585 (2000) 45;
M. Hirsch, M.A. Diaz, W. Porod, J.C. Romao, J.W.F. Valle, Phys. Rev. D 62 (2000) 113008;
M. Hirsch, M.A. Diaz, W. Porod, J.C. Romao, J.W.F. Valle, Phys. Rev. D 65 (2002) 119901, Erratum;
S. Davidson, M. Losada, Phys. Rev. D 65 (2002) 075025.
K.S. Babu, S.M. Barr, Phys. Lett. B 381 (1996) 202;
J. Sato, T. Yanagida, Phys. Lett. B 430 (1998) 127;
C.H. Albright, K.S. Babu, S.M. Barr, Phys. Rev. Lett. 81 (1998) 1167;
J.K. Elwood, N. Irges, P. Ramond, Phys. Rev. Lett. 81 (1998) 5064;
Y. Nomura, T. Yanagida, Phys. Rev. D 59 (1999) 017303;
N. Haba, Phys. Rev. D 59 (1999) 035011;
G. Altarelli, F. Feruglio, Phys. Lett. B 451 (1999) 388;
W. Buchmuller, T. Yanagida, Phys. Lett. B 445 (1999) 399;
Z. Berezhiani, A. Rossi, JHEP 9903 (1999) 002;
M. Bando, T. Kugo, Prog. Theor. Phys. 101 (1999) 1313;
S. Lola, G.G. Ross, Nucl. Phys. B 553 (1999) 81;
Y. Nir, Y. Shadmi, JHEP 9905 (1999) 023;
K. Yoshioka, Mod. Phys. Lett. A 15 (2000) 29;
P.H. Frampton, A. Rasin, Phys. Lett. B 478 (2000) 424;
M. Bando, T. Kugo, K. Yoshioka, Prog. Theor. Phys. 104 (2000) 211;
M. Bando, T. Kugo, K. Yoshioka, Phys. Lett. B 483 (2000) 163;
J.M. Mira, E. Nardi, D.A. Restrepo, J.W.F. Valle, Phys. Lett. B 492 (2000) 81;
M. Bando, et al., Phys. Rev. D 63 (2001) 113017;
N. Haba, H. Murayama, Phys. Rev. D 63 (2001) 053010;
M.S. Berger, K. Siyeon, Phys. Rev. D 63 (2001) 057302;
A. Kageyama, M. Tanimoto, K. Yoshioka, Phys. Lett. B 512 (2001) 349;
K.S. Babu, S.M. Barr, Phys. Lett. B 525 (2002) 289;
X.J. Bi, Y.B. Dai, Eur. Phys. J. C 27 (2003) 43;
R. Kitano, T.j. Li, Phys. Rev. D 67 (2003) 116004;
T. Asaka, Phys. Lett. B 562 (2003) 291;
K.S. Babu, T. Enkhbat, I. Gogoladze, Nucl. Phys. B 678 (2004) 233.
J. Sato, K. Tobe, T. Yanagida, Phys. Lett. B 498 (2001) 189;
J. Sato, K. Tobe, Phys. Rev. D 63 (2001) 116010.
A.Y. Smirnov, Phys. Rev. D 48 (1993) 3264;
A.Y. Smirnov, Nucl. Phys. B 466 (1996) 25;
M. Tanimoto, Phys. Lett. B 345 (1995) 477;
M. Bando, T. Kugo, K. Yoshioka, Phys. Rev. Lett. 80 (1998) 3004;
M. Jezabek, Y. Sumino, Phys. Lett. B 440 (1998) 327;
M. Bando, K. Yoshioka, Prog. Theor. Phys. 100 (1998) 1239;
D. Falcone, Phys. Rev. D 61 (2000) 097302;
E.K. Akhmedov, G.C. Branco, M.N. Rebelo, Phys. Lett. B 478 (2000) 215;
281
282
[13]
[14]
[15]
[16]
T.K. Kuo, G.H. Wu, S.W. Mansour, Phys. Rev. D 61 (2000) 111301;
A. Datta, F.S. Ling, P. Ramond, Nucl. Phys. B 671 (2003) 383;
M. Honda, S. Kaneko, M. Tanimoto, Phys. Lett. B 593 (2004) 165;
R. Dermisek, Phys. Rev. D 70 (2004) 073016;
M. Lindner, M.A. Schmidt, A.Y. Smirnov, JHEP 0507 (2005) 048.
E. Derman, H.S. Tsao, Phys. Rev. D 20 (1979) 1207;
Y. Yamanaka, H. Sugawara, S. Pakvasa, Phys. Rev. D 25 (1982) 1895;
Y. Yamanaka, H. Sugawara, S. Pakvasa, Phys. Rev. D 29 (1984) 2135, Erratum;
G. Ecker, Z. Phys. C 24 (1984) 353;
D. Chang, W.Y. Keung, G. Senjanovic, Phys. Rev. D 42 (1990) 1599;
D.B. Kaplan, M. Schmaltz, Phys. Rev. D 49 (1994) 3741;
P.H. Frampton, T.W. Kephart, Int. J. Mod. Phys. A 10 (1995) 4689;
M. Schmaltz, Phys. Rev. D 52 (1995) 1643;
P.H. Frampton, O.C.W. Kong, Phys. Rev. Lett. 77 (1996) 1699;
A. Aranda, C.D. Carone, R.F. Lebed, Phys. Lett. B 474 (2000) 170;
E. Ma, G. Rajasekaran, Phys. Rev. D 64 (2001) 113012;
K.S. Babu, E. Ma, J.W.F. Valle, Phys. Lett. B 552 (2003) 207;
E. Ma, Mod. Phys. Lett. A 17 (2002) 2361;
W. Grimus, L. Lavoura, Phys. Lett. B 572 (2003) 189;
C.I. Low, R.R. Volkas, Phys. Rev. D 68 (2003) 033007;
M. Hirsch, J.C. Romao, S. Skadhauge, J.W.F. Valle, A. Villanova del Moral, Phys. Rev. D 69 (2004) 093006;
W. Grimus, A.S. Joshipura, S. Kaneko, L. Lavoura, M. Tanimoto, JHEP 0407 (2004) 078;
M. Frigerio, S. Kaneko, E. Ma, M. Tanimoto, Phys. Rev. D 71 (2005) 011901;
K.S. Babu, J. Kubo, Phys. Rev. D 71 (2005) 056006;
G. Altarelli, F. Feruglio, Nucl. Phys. B 720 (2005) 64;
A. Zee, Phys. Lett. B 630 (2005) 58.
H. Harari, H. Haut, J. Weyers, Phys. Lett. B 78 (1978) 459;
Y. Koide, Phys. Rev. D 28 (1983) 252;
Y. Koide, Phys. Rev. D 39 (1989) 1391;
P. Kaus, S. Meshkov, Mod. Phys. Lett. A 3 (1988) 1251;
P. Kaus, S. Meshkov, Mod. Phys. Lett. A 4 (1989) 603, Erratum;
L. Lavoura, Phys. Lett. B 228 (1989) 245;
M. Tanimoto, Phys. Rev. D 41 (1990) 1586;
G.C. Branco, J.I. Silva-Marcos, M.N. Rebelo, Phys. Lett. B 237 (1990) 446;
H. Fritzsch, J. Plankl, Phys. Lett. B 237 (1990) 451;
H. Fritzsch, Z.Z. Xing, Phys. Lett. B 372 (1996) 265;
M. Fukugita, M. Tanimoto, T. Yanagida, Phys. Rev. D 57 (1998) 4429;
R.N. Mohapatra, S. Nussinov, Phys. Lett. B 441 (1998) 299;
P.F. Harrison, W.G. Scott, Phys. Lett. B 557 (2003) 76;
F. Caravaglios, S. Morisi, hep-ph/0503234.
S. Pakvasa, H. Sugawara, Phys. Lett. B 73 (1978) 61;
E. Derman, Phys. Rev. D 19 (1979) 317;
L.J. Hall, H. Murayama, Phys. Rev. Lett. 75 (1995) 3985;
C.D. Carone, L.J. Hall, H. Murayama, Phys. Rev. D 53 (1996) 6282;
C.D. Carone, L.J. Hall, H. Murayama, Phys. Rev. D 54 (1996) 2328;
R.N. Mohapatra, A. Perez-Lorenzana, C.A. de Sousa Pires, Phys. Lett. B 474 (2000) 355;
J. Kubo, A. Mondragon, M. Mondragon, E. Rodriguez-Jauregui, Prog. Theor. Phys. 109 (2003) 795;
T. Kobayashi, J. Kubo, H. Terao, Phys. Lett. B 568 (2003) 83;
J. Kubo, H. Okada, F. Sakamaki, Phys. Rev. D 70 (2004) 036007;
T. Araki, J. Kubo, E.A. Paschos, hep-ph/0502164.
D. Wyler, Phys. Rev. D 19 (1979) 330;
E. Ma, Phys. Rev. D 43 (1991) 2761;
N.G. Deshpande, M. Gupta, P.B. Pal, Phys. Rev. D 45 (1992) 953;
R. Dermisek, S. Raby, Phys. Rev. D 62 (2000) 015007;
S.L. Chen, M. Frigerio, E. Ma, Phys. Rev. D 70 (2004) 073008;
S.L. Chen, M. Frigerio, E. Ma, Phys. Rev. D 70 (2004) 079905, Erratum;
283
W. Grimus, L. Lavoura, JHEP 0508 (2005) 013.

[17] S. Dimopoulos, H. Georgi, Nucl. Phys. B 193 (1981) 150;
J.R. Ellis, D.V. Nanopoulos, Phys. Lett. B 110 (1982) 44;
J.F. Donoghue, H.P. Nilles, D. Wyler, Phys. Lett. B 128 (1983) 55;
L.J. Hall, V.A. Kostelecky, S. Raby, Nucl. Phys. B 267 (1986) 415;
F. Gabbiani, A. Masiero, Nucl. Phys. B 322 (1989) 235;
J.S. Hagelin, S. Kelley, T. Tanaka, Nucl. Phys. B 415 (1994) 293;
F. Gabbiani, E. Gabrielli, A. Masiero, L. Silvestrini, Nucl. Phys. B 477 (1996) 321.
[18] A.H. Chamseddine, R. Arnowitt, P. Nath, Phys. Rev. Lett. 49 (1982) 970;
R. Barbieri, S. Ferrara, C.A. Savoy, Phys. Lett. B 119 (1982) 343;
L.J. Hall, J. Lykken, S. Weinberg, Phys. Rev. D 27 (1983) 2359;
H.P. Nilles, Phys. Rep. 110 (1984) 1.
[19] C.D. Froggatt, H.B. Nielsen, Nucl. Phys. B 147 (1979) 277.
[20] M. Leurer, Y. Nir, N. Seiberg, Nucl. Phys. B 420 (1994) 468;
L.E. Ibanez, G.G. Ross, Phys. Lett. B 332 (1994) 100;
E. Dudas, S. Pokorski, C.A. Savoy, Phys. Lett. B 356 (1995) 45;
P. Binetruy, S. Lavignac, P. Ramond, Nucl. Phys. B 477 (1996) 353;
R.N. Mohapatra, A. Riotto, Phys. Rev. D 55 (1997) 4262;
K. Choi, E.J. Chun, H.D. Kim, Phys. Lett. B 394 (1997) 89;
A.E. Nelson, D. Wright, Phys. Rev. D 56 (1997) 1598;
Y. Grossman, Y. Nir, Y. Shadmi, JHEP 9810 (1998) 007;
S.F. King, Nucl. Phys. B 576 (2000) 85;
Q. Shafi, Z. Tavartkiladze, Phys. Lett. B 482 (2000) 145;
N. Maekawa, Prog. Theor. Phys. 106 (2001) 401;
N. Maekawa, T. Yamashita, Prog. Theor. Phys. 107 (2002) 1201;
M. Kakizaki, M. Yamaguchi, JHEP 0206 (2002) 032;
H.K. Dreiner, H. Murayama, M. Thormeier, Nucl. Phys. B 729 (2005) 278;
P.H. Chankowski, K. Kowalska, S. Lavignac, S. Pokorski, Phys. Rev. D 71 (2005) 055004.
[21] D.s. Du, Z.z. Xing, Phys. Rev. D 48 (1993) 2349;
H. Fritzsch, Z.z. Xing, Phys. Lett. B 353 (1995) 114;
P.S. Gill, M. Gupta, Phys. Rev. D 56 (1997) 3143;
J.L. Chkareuli, C.D. Froggatt, Phys. Lett. B 450 (1999) 158;
H. Nishiura, K. Matsuda, T. Fukuyama, Phys. Rev. D 60 (1999) 013006;
D. Falcone, F. Tramontano, Phys. Rev. D 63 (2001) 073007;
R. Rosenfeld, J.L. Rosner, Phys. Lett. B 516 (2001) 408;
W. Buchmuller, D. Wyler, Phys. Lett. B 521 (2001) 291;
J.L. Chkareuli, C.D. Froggatt, H.B. Nielsen, Nucl. Phys. B 626 (2002) 307;
M. Bando, M. Obara, Prog. Theor. Phys. 109 (2003) 995.
[22] G.C. Branco, J.I. Silva-Marcos, Phys. Lett. B 331 (1994) 390;
T.K. Kuo, S.W. Mansour, G.H. Wu, Phys. Rev. D 60 (1999) 093004;
G.C. Branco, D. Emmanuel-Costa, R. Gonzalez Felipe, Phys. Lett. B 477 (2000) 147.
[23] N. Uekusa, A. Watanabe, K. Yoshioka, Phys. Rev. D 71 (2005) 094024.
[24] G. Altarelli, F. Feruglio, I. Masina, Phys. Lett. B 472 (2000) 382.
[25] R. Kuchimanchi, R.N. Mohapatra, Phys. Rev. D 66 (2002) 051301;
T. Endoh, S. Kaneko, S.K. Kang, T. Morozumi, M. Tanimoto, Phys. Rev. Lett. 89 (2002) 231601;
M. Raidal, A. Strumia, Phys. Lett. B 553 (2003) 72;
S.F. King, Phys. Rev. D 67 (2003) 113010;
S. Raby, Phys. Lett. B 561 (2003) 119;
B. Dutta, R.N. Mohapatra, Phys. Rev. D 68 (2003) 056006;
V. Barger, D.A. Dicus, H.J. He, T.j. Li, Phys. Lett. B 583 (2004) 173;
W.l. Guo, Z.z. Xing, Phys. Lett. B 583 (2004) 163;
W. Rodejohann, Eur. Phys. J. C 32 (2004) 235;
A. Ibarra, G.G. Ross, Phys. Lett. B 591 (2004) 285;
S.h. Chang, S.K. Kang, K. Siyeon, Phys. Lett. B 597 (2004) 78.
[26] T. Yanagida, in: O. Sawada, A. Sugamoto (Eds.), Proceedings of the Workshop on Unified Theories and Baryon
Number in the Universe, KEK report 79-18, 1979;
284
M. Gell-Mann, P. Ramond, R. Slansky, in: P. van Nieuwenhuizen, D.Z. Freedman (Eds.), Supergravity, NorthHolland, Amsterdam, 1979.
[27] P.H. Frampton, S.L. Glashow, T. Yanagida, Phys. Lett. B 548 (2002) 119.
[28] A. Cervera, A. Donini, M.B. Gavela, J.J. Gomez Cadenas, P. Hernandez, O. Mena, S. Rigolin, Nucl. Phys. B 579
(2000) 17;
P. Huber, M. Lindner, W. Winter, Nucl. Phys. B 645 (2002) 3;
P. Huber, M. Lindner, T. Schwetz, W. Winter, Nucl. Phys. B 665 (2003) 487.
[29] For a recent review, see, S.R. Elliott, P. Vogel, Annu. Rev. Nucl. Part. Sci. 52 (2002) 115, hep-ph/0202264.
Nuclear Physics B 739 [FS] (2006) 285310
Bubbling 1/4 BPS solutions in type IIB and

supergravity reductions on S n S n
James T. Liu, Diana Vaman , W.Y. Wen
Michigan Center for Theoretical Physics, Randall Laboratory of Physics, The University of Michigan,
Ann Arbor, MI 48109-1120, USA
Received 13 September 2005; received in revised form 15 December 2005; accepted 22 December 2005
Abstract
We extend the construction of bubbling 1/2 BPS solutions of Lin, Lunin and Maldacena (H. Lin,
O. Lunin, J. Maldacena, JHEP 0410 (2004) 025, hep-th/0409174) in two directions. First we enquire
whether bubbling 1/2 BPS solutions can be constructed in minimal 6d supergravity and second we construct
solutions that are 1/4 BPS in type IIB. We find that the S 1 S 1 bosonic reduction of (1, 0) 6d supergravity
to 4d gravity coupled to 2 scalars and a gauge field is consistent only provided that the gauge field obeys
a constraint (F F = 0). This is to be contrasted to the case of the S 3 S 3 bosonic reduction of type IIB
supergravity to 4d gravity, 2 scalars and a gauge field, where consistency is achieved without imposing any
such constraints. Therefore, in the case of (1, 0) 6d supergravity we are able to construct 1/2 BPS solutions,
similar to those derived in type IIB, provided that this additional constraint is satisfied. This ultimately prohibits the construction of a family of 1/2 BPS solutions corresponding to a bubbling AdS3 S 3 geometry.
Returning to type IIB solutions, by turning on an axiondilaton field we construct a family of bubbling 1/4
BPS solutions. This corresponds to the inclusion of back-reacted D7 branes to the solutions of Lin, Lunin
and Maldacena.
1. Introduction and summary

A recent paper by Lin, Lunin and Maldacena [1] provided a nice supergravity realization of
chiral primary operators in N = 4 super-YangMills. These operators, with conformal dimension
equal their U (1)R charge, form a decoupled sector of BPS states which can be identified with
E-mail addresses: jtliu@umich.edu (J.T. Liu), dvaman@umich.edu (D. Vaman), wenw@umich.edu (W.Y. Wen).
doi:10.1016/j.nuclphysb.2005.12.026
286
J.T. Liu et al. / Nuclear Physics B 739 [FS] (2006) 285310
a gauged quantum mechanics matrix model, with a harmonic oscillator potential [2,3]. By going
to the eigenvalue basis, the path integral measure acquires a Van der Monde determinant factor
which makes the eigenvalues behave as fermions which fill the energy levels inside a harmonic
potential well. There is yet another perspective on the dynamics of eigenvalues: they correspond
to the electrons in a magnetic field which fill the lowest Landau levels; by dropping the kinetic
term, the positions of the electrons in the plane become non-commutative/canonical conjugates
(quantum Hall effect) [1,4].
There appears to be a one-to-one correspondence between the phase space regions occupied
by the eigenvalues, and a similar picture that characterizes the supergravity solutions. Specifically, the 1/2 BPS family of solutions of [1] is constructed in terms of an auxiliary function;
the boundary conditions which must be enforced on this auxiliary function in order for the supergravity solution to be non-singular reproduce precisely the phase space configuration of the
eigenvalues. On the supergravity side, the incompressibility of the phase space has to be tied
to the requirement that the 5-form RR-flux be fixed. In particular, the ground state on the matrix
model side is a circular quantum Hall droplet in phase space, while on the supergravity side, the
same droplet corresponds to the AdS5 S 5 ground state, with the radius of the droplet related to
2
the RAdS
radius. Small excitations of the Fermi sea on the matrix side correspond to AdS5 S 5
5
perturbations by gravitons (ripples on the ground state droplet) or giant gravitons (small holes
inside the ground state droplet, or small droplets outside the ground state droplet) [1].
It is worth asking whether a similar picture might carry through for the case of AdS3 S 3
which is the near horizon geometry of a D1D5 brane configuration, and whether there is a bubbling AdS3 configuration corresponding to perturbation by giant gravitons. The six-dimensional
giant gravitons are also configurations with = J , but they have certain peculiar features: they
exist only for a discrete set of values of the angular momentum (J = nN5 + mN1 , where N5 , N1
are the numbers of D5 , respectively D1 branes), and the potential governing their size is flat. The
dual gauge theory in this case is a (1 + 1)-dimensional CFT living on the boundary of AdS3 .
To address this question we look for 1/2 BPS solutions to minimal six-dimensional supergravity, which have S 1 S 1 isometry. More precisely we consider the following ansatz:
ds 2 = g (x) dx dx + eH (x)+G(x) d 2 + eH (x)G(x) d 2 ,
2H(3) = F(2) d + F(2) d ,
(1.1)
where x = {t, y, x 1 , x 2 }. This is a natural extension of the ansatz used in [1], and similarly
reduces the problem to that of an effective four-dimensional theory. Requiring that this ansatz
preserves supersymmetry, we found two possible sets of Killing spinors: one which is independent on the , coordinates, and which yields a conventional KaluzaKlein dimensional
reduction, and a second set which carries KaluzaKlein momentum on the two circles. The latter set of Killing spinors gives rise to six-dimensional solutions which include the AdS3 S 3
and the maximally symmetric plane wave solutions, as well as the multi-center D1D5 solutions. The metric and the self-dual 3-form are expressed in terms of the same auxiliary
function as in the 10-dimensional case, namely x21 z + x22 z + yy (1/yy z) = 0. The corresponding metric is non-singular provided that the same boundary conditions as in [1] are imposed
on the auxiliary function: z(x1 , x2 , y = 0) = 1/2. However, this time the field equations are
satisfied only if an additional constraint is enforced as well: F F = 0, which can be easily seen by inspecting the component of Einstein equations. This constraint appears as a
non-linear first order differential equation which the auxiliary function must satisfy as well:
(x1 z)2 + (x2 z)2 + (y z)2 = (1 4z2 )2 /(4y 2 ). One can check that for the case of AdS3 S 3 and
287
that of the maximally symmetric plane wave the corresponding auxiliary functions do satisfy this
additional constraint. However, the image of a bubbling AdS3 appears to be incompatible with
the additional constraint, and we have to conclude that the starting ansatz is too restrictive to describe giant gravitons on AdS3 S 3 . In fact, therein lies the resolution: in order to eliminate this
constraint, we must allow for (at least) an off-diagonal metric component g = 0 in the ansatz.
This would correspond in the 4-dimensional reduction language to keeping a non-vanishing axion field besides the two scalars and the gauge field. We hope to report more on this in a future
work. We also considered the case when a tensor multiplet (dilaton, dilatino and anti-selfdual
tensor field) is coupled to six-dimensional minimal supergravity. With the same metric ansatz
as before, we arrived yet again at the same auxiliary function z(x 1 , x 2 , y), same metric and the
same constraint F F = 0.
Finally, we return to the 10-dimensional type IIB supergravity to investigate a family of 1/4
BPS solutions which preserve the same set of SO(4) SO(4) isometries as in [1] and which
correspond to turning on an axion-dilaton field , or equivalently, including the back-reaction
of a stack of Nf D7 branes. We found it possible to embed the D7 branes in a way compatible with the metric ansatz of [1], provided that they are transverse to x 1 , x 2 . The presence of
the D7 branes manifests in the metric by curving the x 1 , x 2 directions with an additional factor
1 2
e(x ,x ) . The auxiliary function z(x 1 , x 2 , y) obeys a slightly modified second order differential equation, but one still imposes the same boundary conditions at y = 0. By setting up a
perturbative expansion in Nf , we notice that to first order in Nf , and near the D7 branes, their
effect on the geometry is to create a deficit angle in the x 1 , x 2 plane. Given that the fluctuations in the Z = x 1 + ix 2 direction transverse to the stack of D3 branes correspond to the BPS
chiral operators that are singled out in the gauged quantum mechanics matrix model, and that
their eigenvalues correspond to the electrons in the quantum Hall effect, we argue that the deficit
angle in the supergravity x 1 , x 2 phase space can be interpreted as the fractional statistics of the
electrons in a fractional quantum Hall system.
It is worth noting that, despite recent advances, the complete classification of supersymmetric
backgrounds with fluxes remains a technically challenging issue, especially in higher dimensions
or with a large number of supersymmetries. The analysis of [1] is striking in this regards, in that
by postulating a minimal set of isometries, the problem of classification may be greatly simplified. In particular, by reducing a (4 + 2n)-dimensional system on S n S n , we end up having
to work with only two scalars and a gauge field in four dimensions. Of course, it is tempting to
view this reduction as a KaluzaKlein reduction on S n S n . In general, care must be taken when
introducing states in the massive KaluzaKlein tower. Here, however, by truncating to the singlet
sector on S n S n , one is able to obtain a consistent bosonic truncation. This is sufficient for
investigating the supersymmetry of the original system, provided the full supersymmetry transformations are used. One unusual feature of this analysis is the possibility of obtaining bosonic
backgrounds which do not appear supersymmetric in four dimensions, but which are nevertheless
supersymmetric when viewed as a solution of the original higher-dimensional theory. This phenomenon was noted in [5], where it was referred to as supersymmetry without supersymmetry.
Furthermore, this method of obtaining gravitational solutions by solving a harmonic equation
in an auxiliary space is similar in spirit to the work of Weyl [6] (see also [7] for a generalization to
higher dimensions), who found all static axisymmetric solutions to the four-dimensional vacuum
Einstein equations by mapping the problem to a cylindrically symmetric Laplacian problem in
an auxiliary flat three-dimensional space. What makes the problem tractable in the Weyl case is
the presence of a sufficient number of commuting Killing symmetries. Of course, one of the interesting features of the Weyl solutions is that they represent non-supersymmetric configurations
288
of black holes held together by rods or struts. Thus, in that case, it is rather surprising that they
may be described by solutions of a harmonic equation. For BPS configurations this is somewhat
less of a surprise, as they are expected to obey the principle of linear superposition. Nevertheless,
it is suggestive that new results in the classification of supersymmetric vacua may be obtained by
revisiting some of the Weyl solution techniques in the present context. In this sense, it may also
be worth looking at the M-theory compactification of [1] on S 2 S 5 from the four-dimensional
perspective.
The paper is structured as follows: Section 2 is dedicated to a proof that the bosonic reduction
of type IIB supergravity on S 3 S 3 by retaining the breathing modes of the two 3-spheres and a
gauge field in the reduced 4-dimensional theory is a consistent bosonic reduction. We also review
the supersymmetry analysis of [1], using a slightly different representation of the 10-dimensional
Clifford algebra which allows for more streamlined expressions. Section 3, which is organized
in the same fashion as Section 2, contains the analysis of the six-dimensional supergravity reduction on S 1 S 1 to the same set of 4-dimensional fields, and arrives at the conclusion that the
reduction is a consistent bosonic reduction provided that the constraint F F = 0 is satisfied.1
In Section 4 we construct the six-dimensional solutions which are compatible with supersymmetry. The additional constraint does not allow for solutions corresponding to bubbling AdS3 S3 .
Section 5 details the construction of the 1/4 BPS family of solutions corresponding to a bubbling
AdS5 S 5 in the presence of D7 branes.
We have included the most technical parts of our investigation in a set of appendices. Appendix A contains a unified treatment of both the type IIB and six-dimensional supergravity bosonic
reductions on S n S n , where n = 3(1) for the 10(6)-dimensional case respectively. Appendix B
discusses the integrability condition of the supersymmetry variations of both type IIB and sixdimensional supergravity, and highlights the difference between the two, in the sense that the
constraint F F = 0 shows up in all S n S n reduction cases other than n = 3. Finally, Appendix C contains the full set of differential identities for spinor bilinears implied by supersymmetry.
2. S 3 S 3 compactification of IIB supergravity
The bosonic fields of IIB supergravity are given by the NSNS fields gMN , BMN and as well
+
. In the Einstein frame, the IIB action has the form
as the RR field strengths F(1) , F(3) and F(5)
e1 L = R
2
(3)
(3)
F(5)
M M G(3) G
1 C(4) G(3) G
+
,
2 3!
4 5! 4i

2( )2
(2.1)
where the self-duality F(5) = F(5) must still be imposed by hand on the equations of motion.
Here the field strengths are defined by
F(3) = dC(2) ,
H(3) = dB(2) ,
G(3) = F(3) H(3) ,
1
1
F(5) = dC(4) C(2) H(3) + B(2) F(3) ,
2
2
(2.2)
and = C(0) + ie is the familiar axiondilaton.

1 In this paper we address the reduction of six-dimensional supergravity on a rectangular torus T 2 . As a consequence,
in the reduced four-dimensional theory we have only two scalar fields, corresponding to the breathing modes of the
two T 2 circles. The ensuing constraint F F = 0 can be eliminated by turning on an additional four-dimensional field,
an axionic scalar, corresponding to the reduction of six-dimensional supergravity on a generic torus [8].
289
Although we are not directly concerned with the entire fermionic sector, since we are interested in the Killing spinor equations, we will need the IIB gravitino and dilatino variations

i
1
M = DM +
FN P QRS N P QRS M
M N P Q + 2 N P Q M GN P Q ,
16 5!
96
i
= i M PM GMNP MNP .
(2.3)
24
Here DM = M 2i QM where PM and QM are the scalar kinetic and composite U (1) connection, respectively.
Note that the NSNS sector of the IIB model can be reduced on S 3 S 3 [9,10] to yield the
gauged N = 4 FreedmanSchwarz theory [11]. What we are interested in at present, however,
is a reduction with additional degrees of freedom, in particular the self-dual 5-form as well as
metric breathing modes. Here we note that, from a KaluzaKlein point of view, the breathing
modes are actually part of the massive KaluzaKlein tower. In general, in the FreedmanSchwarz
supergravity context, massive KaluzaKlein supermultiplets necessarily involve charged modes
on the spheres. As a result, it would be inconsistent to retain a single massive multiplet without
retaining the entire tower.
Nevertheless, it is always possible to obtain a consistent bosonic breathing mode reduction by
retaining only singles on S 3 S 3 [12,13]. While the truncated theory is non-supersymmetric, we
may still explore the original ten-dimensional Killing spinor equations obtained from (2.3), even
in the context of the reduced bosonic fields. In this fashion, bosonic solutions of the compactified
theory may be lifted to supersymmetric backgrounds of the original IIB theory, so long as the
original Killing spinor equations are satisfied.
We now follow [1], and turn to the sector where only F(5) is turned on (in addition to the
metric). In this case, the equations of motion obtained from (2.1) admit a consistent truncation,
so the relevant ten-dimensional Lagrangian is of the form
1
(2.4)
F2 ,
4 5! (5)
where F(5) = dC(4) and F(5) = F(5) is to be imposed on the equations of motion. This system is
now that of a single self-dual form-field coupled to gravity, admitting a straightforward reduction
on S 3 S 3 . In particular, we take a reduction ansatz preserving an SO(4) SO(4) isometry of
the form

2
= g (x) dx dx + eH (x) eG(x) d32 + eG(x) d 32 ,
ds10
F(5) = F(2) 3 + F(2) 3 .
(2.5)
e1 L = R
The details of this reduction are given in Appendix A. In the end, we obtain an effective fourdimensional Lagrangian of the form

3
1
15
2
+ 12eH cosh G .
e1 L = e3H R + H 2 G2 e3(H +G) F
(2.6)
2
2
4
At this point it is worth noting that the model of [1] may be extended, not just by turning on
the axiondilaton, but also by retaining the 3-form field-strength with an ansatz of the form
G(3) = G(3) (x) + a(x)3 + a(x)
3 .
(2.7)
While this may be of interest for obtaining additional supersymmetric backgrounds, we will not
further pursue this direction at present.
290
2.1. Supersymmetry variations

We now turn to the reduction of the IIB supersymmetry variations, (2.3). At present, since
we only turn on the metric and self-dual 5-form, the only non-trivial variation is that of the IIB
gravitino, which has the form

i
FN P QRS N P QRS M .
M = M +
(2.8)
16 5!
Writing this out in components, and using the reduction ansatz (2.5), we find

3
1
= e 2 (H +G) F (3) ,
16

3
1
1
a = a + a (H + G) e 2 (H +G) F (3) a ,
4
16

3
1
1
a = a + a (H G) e 2 (H +G) F (3) a ,
4
16
(2.9)
where (3) = 6i abc abc , and we have taken into account the chirality of IIB spinors, 11 =
1
M1 ...M10 M1 ...M10 .
, where 11 = 10!
To proceed, we choose a Dirac decomposition respecting the S 3 S 3 symmetry
= 1 1 1 ,
a = 1 a 1 2 ,
a = 5 1 a 1 .
Here the s are ordinary Pauli matrices, while 5 =

that, in this representation, the respective chirality matrices are
i
.
4!
(2.10)
It is straightforward to see
i
= 5 1 1 1,
4!
i
(3) = abc abc = 1 1 1 2 ,
3!
i
(3) = a b c a bc = 5 1 1 1 ,
3!
1
11
=
M ...M M1 ...M10 = i (5) (3) (3) = 1 1 1 3 .
10! 1 10
With this decomposition, the gravitino transformation becomes

3
i
= + e 2 (H +G) F ,
16

i
i 3
a = a a (H + G) e 2 (H +G) F ,
4
4

1
i 3
a = a + 5 a (H G) + e 2 (H +G) F .
4
4
(5) =
(2.11)
(2.12)
We now write the complex IIB spinor as = where and are Killing spinors on
the respective unit three-spheres. Since they satisfy the Killing spinor equations

i
i
a + a = 0,
(2.13)
a + a = 0,
2
2
where = 1 and = 1, the transformations (2.12) may be rewritten as

3
i
= + e 2 (H +G) F ,
16

1
1
1
H = H + e 2 H e 2 G i
5 e 2 G ,

1
1
1
i 3 (H +G)
G
G
.
F + e 2 e 2 + i
5e 2
G = G e 2
4
291
(2.14)
(2.15)
(2.16)
Here we have defined

i
1
a = 5 a (H G ).
a = a (H + G ),
(2.17)
4
4
To summarize, we have achieved a consistent bosonic breathing-mode reduction of the truncated IIB theory on S 3 S 3 . The resulting four-dimensional Lagrangian is given by (2.6),
while the reduction of the supersymmetry variations results in the system (2.14)(2.16). Note
that, while the variations have a typical form associated with four-dimensional N = 2 supergravity, they nevertheless should not to be thought of as supersymmetries of the effective
four-dimensional theory. This is because the bosonic truncation to singlets on S 3 S 3 does not
(and cannot) retain the complete supermultiplet content in the massive KaluzaKlein sector, as
indicated previously.
By reducing to an effective four-dimensional theory, we have ended up with a fairly simple
system to investigate. Of course, we are mainly concerned with solving the Killing spinor equations derived from (2.14)(2.16). In contrast to the approach of [1], we may now work directly in
a four-dimensional context, even though the equations originated from the full IIB gravitino variation, (2.3). In addition, by reducing the six-dimensional N = (1, 0) solution to four dimensions,
we would similarly obtain an effective theory of the same general form as (2.6) and (2.14)(2.16).
Thus, using the four-dimensional picture, we will be able to solve both model simultaneously.
3. S 1 S 1 compactification of minimal D = 6 supergravity
We now turn to the case of minimal D = 6, N = (1, 0) supergravity which admits an AdS3
S 3 solution corresponding to the near horizon of the D1D5 system. The field content of this
+
+
theory is given by the gravity multiplet (gMN , M , BMN
), where BMN
denotes a two-form
+
potential with self-dual field strength, H(3) = H(3) where H(3) = dB(2) . This is often extended
, , ), as the inclusion of both chiralities of BMN

by the addition of a dilaton multiplet (BMN
then allow a covariant Lagrangian formulation. Furthermore, the SalamSezgin model [14] may
be obtained by coupling to an Abelian vector multiplet (A , ). In the following section, we
will consider the addition of the dilaton multiplet. Here, however, we only focus on the minimal
supergravity without dilaton or vector multiplet.
The bosonic sector of the minimal theory may be described by the Lagrangian
1
(3.1)
H2 ,
2 3! (3)
where the self-duality condition on H(3) remains to be imposed after deriving the equations of
motion. Note that, for convenience, we have chosen to normalize H(3) as if it were an unconstrained form-field. The resulting equations are simply
e1 L = R
1
RMN = HMP Q HN P Q ,
4
H(3) = H(3) ,
dH(3) = 0.
(3.2)
292
In addition, we note that the gravitino variation is given by

1
M = M + HN P Q N P Q M .
(3.3)
48
These are the starting points of the reduction.
Analogous to the S 3 S 3 reduction of IIB supergravity, we reduce the N = (1, 0) theory
on S 1 S 1 . This is, of course, a familiar situation, as it is simply an ordinary KaluzaKlein
reduction on T 2 , specialized to a rectangular torus. Before proceeding, it is worthwhile recalling
the standard KaluzaKlein result. Since this supergravity involves eight real supercharges, it
reduces to a N = 2 theory in four dimensions. Since the six-dimensional metric reduces to a fourdimensional metric, two vectors and three scalars (two dilatonic and one axionic), and the self+
dual BMN
reduces to a vector and axionic scalar, the resulting four-dimensional theory consists
of N = 2 supergravity coupled to two vector multiplets.
In the present case, however, we specify a reduction ansatz of the form

ds62 = g (x) dx dx + eH (x) eG(x) d 2 + eG(x) d 2 ,
1
1
H(3) = F(2) d F(2) d ,

(3.4)
2
2
which preserves the SO(2) SO(2) isometry.2 The 1/2 factors in the H(3) ansatz are physically unimportant, but are chosen for later notational convenience in the reduced supersymmetry
variations. In the framework of a torus reduction, this ansatz corresponds to setting both metric
gauge fields and both axionic scalars to zero. As a result, this ends up corresponding to an inconsistent truncation of the original N = (1, 0) theory. Nevertheless, as we demonstrate below, this
inconsistency manifests itself solely in one additional constraint that must be imposed by hand
on the effective four-dimensional theory.
Given the reduction ansatz (3.4), it is straightforward to apply the results of Appendix A
(while taking into account the normalization of F(2) ) to arrive at the effective four-dimensional
Lagrangian

1
1
1
2
e1 L = eH R + H 2 G2 e(H +G) F
(3.5)
,
2
2
8
G
where we
note that F(2) = 4 e F(2) . If desired, F(2) may be canonically normalized by taking
F(2) 2F(2) . However, it will be clear below when considering the supersymmetry variations
why we avoid this last step. In addition, the reduction of the R component of the Einstein
equation results in a constraint F(2) F(2) = 0. This is the manifestation of the inconsistency in
the reduction alluded to above. Ordinarily F(2) F(2) will source one of the axions. However, by
truncating them away, we can no longer allow such a source. Nevertheless, so long as we satisfy
this constraint, all solutions to the effective four-dimensional theory may be lifted to provide
solutions to the original N = (1, 0) model.
This reduction differs from the S 3 S 3 case since, unlike the spheres, the circles are flat. Thus
no potential is generated in the reduced theory. For the same reason, the scalar H , which plays
the rle of a breathing mode in the sphere reduction, is instead an ordinary massless scalar in the
present case.
2 As mentioned in the introduction, we are concerned here only with the reduction on a rectangular torus T 2 . The
reduction ansatz on a generic torus, which has been addressed in [8], includes an additional four-dimensional pseudoscalar field, (x): ds62 = g (x) dx dx + eH (x)+G(x) d 2 + eH (x)G(x) (d 2 + (x) d)2 .
293
3.1. Supersymmetry variations

Having completed the reduction of the bosonic sector, we now proceed to reduce the gravitino variation, (3.3). Although the N = (1, 0) theory is generally formulated with symplecticMajorana Weyl spinors, here the six-dimensional spinors may be taken to be simply complex Weyl, satisfying the left-handed projection 7 = as well as 7 M = M , where
7 = 6!1 M1 ...M6 M1 ...M6 .
To obtain an effective four-dimensional description of the supersymmetry, we find it convenient to decompose the six-dimensional Dirac matrices according to
= 1 ,
4 = 1 2 ,
5 = 5 1 .
(3.6)
respectively. It is straightforward to see that

Indices 4 and 5 correspond to coordinates and ,
(5) =
i
= 5 1,
4!
7=
1
M ...M M1 ...M6 = 1 3 .
6! 1 6
(3.7)
As a result, left-handed six-dimensional spinors may be written as = [ 10].

Noting that

3 1
HMNP MNP = e 2 (H +G) F 4 1 + 7 ,
(3.8)
2
(taking into account the unusual normalization of (3.4)) and using the above Dirac decomposition, the gravitino variation becomes

i 1
= + e 2 (H +G) F ,
16

i 1 (H +G)
1
2
(H + G) F
,
= e
4
16

1 1 (H G) 5
i G
5
2
(H G) + e F
.
= + e
(3.9)
4
16
Up to this point, we have followed a conventional KaluzaKlein reduction which involves truncation to zero-modes only on S 1 S 1 . However, we are not necessarily interested in obtaining a
consistent truncation to four dimensions, but rather wish to obtain supersymmetric configurations
of the original N = (1, 0) theory. We thus relax the KaluzaKlein condition on the supersym
= ei(q+q )
metry parameter . In particular, we take (x, , )
(x), where the KaluzaKlein
momenta q and q are quantized in half-integer units (as appropriate for a spinor on a circle).
For convenience of notation, we redefine the (half-integer) KaluzaKlein charges as q = /2
and q = /2.
Then, for spinors charged on the two circles, we may make a simple replacement
i ,
i .
2
2
As a result, the above transformations may be rewritten as

i 1 (H +G)
F ,
= + e 2
16

1G
1
12 H
G
5e 2
e 2 i
,
H = H + e
(3.10)
294

1
1
1
i 1 (H +G)
G
G
,
G = G e 2
F + e 2 e 2 + i
5e 2
4
(3.11)
where we have defined the linear combinations

i 1
= e 2 (H +G) (H + G ),
4
1 1
= e 2 (H G) 5 (H G ).
4
(3.12)
In this form, the supersymmetry variations (3.11) resemble those of the S 3 S 3 reduced IIB
theory, (2.14)(2.16). However, in this case, the parameters and may take on any integer
values include zero. Ordinary Killing spinors of the massless sector KaluzaKlein reduction are
obtained for = = 0, while charged Killing spinors are obtained otherwise. We keep in mind,
however, that the bosonic sector is unchanged and always has the form of (3.5) regardless of the
structure of the Killing spinors.
This reduction framework provides yet another example where the supersymmetry of spaces
such as AdS3 S 3 , viewed as a fibration, involve Killing spinors charged along U (1) fibers.
Proper supersymmetry counting then involves proper identification of the fiber U (1) charges [5,
13,15].
4. Supersymmetry analysis
In the previous two sections, we have demonstrated that both the reduction of IIB theory on
S 3 S 3 and the reduction of N = (1, 0) supergravity on S 1 S 1 lead to similar effective fourdimensional descriptions. In particular, this similarity is evident not only in the bosonic sectors
(2.6) and (3.5) but also in the supersymmetry transformations (2.14)(2.16) and (3.11). At first
sight, this is actually somewhat surprising, as the details of the fermionic sector, and in particular
the mechanics of the supersymmetry algebra, are very much dimension dependent. However,
given that both theories truncate to an identical field content, it is perhaps not unreasonable to
expect that the effective four-dimensional descriptions of the supersymmetry transformations
necessarily have a similar form.
In fact, comparing (2.14)(2.16) with (3.11), we find that they may both be written in the form

i n (H +G)
= + e 2
F ,
16

1
1
1
5 e 2 G ,
H = H + e 2 H e 2 G i

1G
1
i n (H +G)
12 H
G
2
2
2
G = G e
(4.1)
e
,
F + e
+ i
5e
4
where n = 3 or 1 corresponds to the S 3 S 3 or S 1 S 1 cases, respectively. This description
of supersymmetry allows for a unified analysis of half-BPS solutions, using the methods of [1,
1619].3 There are several differences to note about the two cases, however. Firstly, for n = 3,
the choice of = 1 and = 1 is required based on satisfying the Killing spinor equations
on the two 3-spheres. However, for n = 1, the parameters and refer to U (1) charges along
the two circles, and may be chosen to be arbitrary integers including zero (to be consistent with
charge quantization). Secondly, although F(2) shows up identically in (4.1), there is actually a
3 Note that general solutions of minimal N = (1, 0) supergravity have been classified in [18]. Solutions to IIB theory
on S 3 S 3 were of course examined in [1].
295
factor of two difference in the field strength terms in the bosonic Lagrangians, (2.6) and (3.5).
n
This factor conspires with the n-dependence in the exponential prefactors e 2 (H +G) in (4.1)
so as to yield the appropriate (distinct) bosonic equations of motion based on integrability of
the supersymmetry variations. This issue of integrability is investigated in Appendix B. Finally,
while not directly evident above, it is important to realize that when n = 1 there is an added
constraint, F(2) F(2) = 0, that must be checked for the solution.
Regardless of the differences mentioned above, we begin with a unified treatment of both
cases. Our analysis parallels that of [1]. Thus we first assume that is a Killing spinor and
proceed by forming the spinor bilinears
f1 = 5 ,
f2 = i ,
Y = i 5 .
(4.2)
The factors of i are chosen to make these tensor quantities real. Then, by Fierz rearrangement,
we may prove the algebraic identities
L2 = K 2 = f12 + f22 ,
K = ,
K L = 0.
L = 5 ,
(4.3)
Next we turn to the differential identities. While the complete set of such identities are provided in Appendix C, only a subset suffices for the present analysis. Without yet making any
assumptions about the metric, we may first fix the form of the scalar quantities f1 and f2 . Combining the differential identities for f1 and f2 in (C.1) with the L identities in (C.2) and
(C.3), we obtain
1 n
1
f1 = e 2 (H +G) F K = f1 (H G),
4
2
1 n (H +G)
1
F K = f2 (H + G),
f2 = e 2
4
2
which may be integrated to obtain
1
f1 = be 2 (H G) ,
f2 = ae 2 (H +G) .
(4.4)
(4.5)
G f1 of (C.2). In particular
The constants a and b are related through the identity f2 = e
a + b = 0.
(4.6)
Note that at this stage we still allow and to be arbitrary integers (including zero).
Having fixed the scalars f1 and f2 , we now turn to the vectors K and L . Here, we observe
from (C.1) that the equations for K and L indicate that K(;) = 0 (so that K is a Killing
vector) and dL = 0. As done in [1], this allows us to specialize the metric ansatz below. Before
doing so, however, we note that these vectors are necessarily normalized by (4.3) to be

L2 = K 2 = f12 + f22 = eH a 2 eG + b2 eG .
(4.7)
In particular, this indicates that K is in fact a time-like Killing vector.4 Furthermore, the L
equations of (C.2) provide the constraints
L = b eH ,
L
= a eH .
(4.8)
4 It is useful to observe that this is in agreement with [18] where the 6-dimensional Killing vector M , M = 0, . . . , 5
was shown to be null. The latter statement can be rephrased using our representation of 6-dimensional Dirac matrices as
( )2 + f12 + f22 = 0.
296
Following [1], we now choose a preferred coordinate basis so that the Killing vector K
corresponds to /t and the closed one-form L dx to dy, where t and y are two of the four
coordinates. In particular, we write down the four-dimensional metric as

2

ds42 = h2 dt + Vi dx i + h2 dy 2 + ij dx i dx j ,
(4.9)
where i, j = 1, 2. Note that we have already taken the spatial metric to be conformally flat based
on the identical reasoning as [1]. The remaining components of the metric are Vi , to be determined below, and h2 , given from (4.7) to be

h2 = eH a 2 eG + b2 eG .
(4.10)
In addition, for L = dy, (4.8) yields the constraints
= by eH ,
= ay eH ,
(4.11)
where we still allow for any of these constants , ,

a or b to be zero.
At this stage, we have sufficient information to fix the form of the field strengths F(2) as well
as dV . For F(2) , we use the component relations
n+1
n+1
4a
4b
F K =
e 2 (H +G) ,
e 2 (H G) ,
n+1
n+1
obtained from (4.4) as well as the explicit form of the metric (4.9) to find
F K =
(4.12)

n+1
4b 2 nG
4a n+1 (H +G)
(dt + V )
d e 2
h e 3 d e 2 (H G) ,
n+1
n+1

n+1
n+1 (H G)
4a 2 nG
4b
F(2) =
(4.13)
(dt + V ) +
3 d e 2 (H +G) ,
d e 2
h e
n+1
n+1
where 3 denotes the Hodge dual with respect to the flat spatial metric. For dV , we take the
antisymmetric part of K in (C.1), written in form notation as
F(2) =
1 n
dK = e 2 (H +G) (f2 F(2) f1 F(2) ),
2
(4.14)
and substitute in the expressions for the Killing vector K = h2 (dt + V ) as well as for F(2) .
This gives both the known expression for h2 , namely (4.10), as well as the relation
dV = 2abh4 eH 3 dG.
(4.15)
4.1. Specialization of and

Until now, we have allowed for all possible values of and .
For sphere, as opposed to circle,
compactifications (n > 1) the only possible choices of and are 1, as this is dictated by the
Killing spinor equations on the sphere, (2.13). On the other hand, when n = 1, we may consider
three distinct possibilities: both non-vanishing, one vanishing, and both vanishing.
4.1.1. Both and non-vanishing
We begin with the case of both and non-vanishing. In particular, to satisfy (4.6), we take
a = and b = with a 2 = b2 = 1. The actual choice of a = b corresponds to the case of
chiral primaries with J = 0 in the dual field theory. For this case, (4.11) yields the simple
297
result eH = y, so that (4.10) becomes h2 = 2y cosh G [1]. As in [1], the consistency condition
d 2 V = 0 following from (4.15) yields the second order differential equation d(y 1 3 dz) = 0 or

1
i2 + yy y z = 0,
(4.16)
y
where z = 12 tanh G.
Of course, it is not surprising that the analysis of [1] may be generalized away from n = 3, as
the Killing spinor equations (4.1) have a relatively straightforward dependence on n. However, it
is important to examine the complete consistency of the solution generated above, as in general
solving the Killing spinor equations does not automatically yield a complete solution to the
equations of motion, but only guarantees that a subset of the equations are solved. This issue
of integrability is examined in detail in Appendix B. Here, it is sufficient to note that the F(2)
equations of motion are not obviously satisfied. Instead, by combining (4.13) with (4.15), we
find that
dF(2) = (n 3)bh2 ye
n1
2 (H +G)
(dG 3 dG dH 3 dH ).
(4.17)
This demonstrates that, at least for n = 3, we must impose the additional constraint dG3 dG =
dH 3 dH , or
i zi z + y zy z =
(1 4z2 )2
.
4y 2
(4.18)
That this constraint shows up for n = 3 is related to the fact that the bosonic equations pick up a
F(2) F(2) = 0 constraint in this case as well. Unfortunately, this non-linear constraint is highly
restrictive for functions z(x1 , x2 , y) already satisfying the Laplacian equation of motion (4.16).
While we have not made an exhaustive search, we have only found the maximally symmetric
AdS3 S 3 and plane-wave solutions to satisfy this constraint.
Without loss of generality, we choose a = b = 1 for 1/2 BPS solutions. In this case, the
generalization of [1] for n = 1 as well as 3 may be summarized as follows:

2

ds62 = h2 dt + Vi dx i + h2 dy 2 + ij dx i dx j + y eG dn2 + eG d n2 ,
(4.19)

n+1 n+1
n+1
n+1
4
F(2) =
(4.20)
d y 2 e 2 G (dt + V ) + h2 enG 3 d y 2 e 2 G ,
n+1
where
1
1
z tanh G,
dV = 3 dz.
h2 = 2y cosh G,
(4.21)
2
y
Note that F(2) is only canonically normalized for n = 3. Furthermore, the function z must satisfy
(4.16) for any n, and additionally the constraint (4.18) for n = 3.
4.1.2. Only one of and non-vanishing
This and the subsequent possibility only applies when n = 1. Without loss of generality, we
take = 0, = 1. In this case, the constraint (4.6) indicates that b = 0, so that in particular
f1 = 0 or 5 = 0. To avoid the degenerate situation, we assume that a = 0. Taking a = ,

we see that (4.11) again gives eH = y. This time, however, the relation (4.10) yields a single
exponential, h2 = yeG . In addition, the field strength F(2) is then given by (4.13)

F(2) = 2a d yeG (dt + V ),
(4.22)
298
indicating that it is of pure electric type.

For both and non-vanishing, the second order equation giving the bubbling picture was
obtained from the consistency condition d 2 V = 0. Here, however, dV is trivially closed, as may
be seen by setting b = 0 in (4.15). Nevertheless, we must still satisfy the equation of motion for
F(2) , most conveniently expressed as d F(2) = 0 where

F(2) = 2ay 1 e2G 3 d yeG .
(4.23)
The resulting equation is simply d(y 3 d( y1 eG )) = 0, or

1
2
i + y yy H = 0,
y
(4.24)
where H = h2 = y1 eG is a function of (x 1 , x 2 , y). It is now evident that H is a harmonic function

in a four-dimensional space R2 R2 where (x 1 , x 2 ) span the first R2 and y corresponds to the
radial direction in the second (auxiliary) R2 . Since there is no angular dependence in the second
R2 , the harmonic function is restricted to s-wave only solutions in the auxiliary space.
Putting together the above relations (and taking a = 1), we find that the solution has the form

ds62 = H1 dt 2 + d 2 + H ij dx i dx j + dy 2 + y 2 d 2 ,
1
F(2) = 2 dt d .
(4.25)
H
Note that, because dV = 0, we have set Vi = 0 since this may always be obtained by a suitable
gauge transformation (diffeomorphism). It is now evident that we have reproduced the familiar
multi-centered string solution in six dimensions, restricted to singlet configurations along the
direction, under the assumption that the S 1 parameterized by has decompactified. This configuration arises naturally from the D1D5 system with N1 = N5 .
4.1.3. Both and vanishing
Finally, for n = 1, we could have directly performed a standard KaluzaKlein reduction on the
circles, which would correspond to setting = = 0. Here, the constraint (4.6) becomes trivial,
so that a and b may take on arbitrary values. Assuming at least one of the two is non-vanishing,
then (4.11) implies that H is a constant, which we take to be zero. With this choice of H = 0, we
then solve (4.10) for
h2 = a 2 eG + b2 eG ,
(4.26)
as well as (4.13) for

F(2) = 2aeG dG (dt + V ) + 2bh2 3 dG.
(4.27)
In addition, (4.15) gives

1
1 a 2 eG b2 eG
,
3 dz, z
(4.28)
ab
2 a 2 eG + b2 eG
provided ab = 0, or simply dV = 0 otherwise.
For ab = 0, the consistency condition d 2 V = 0 yields a three-dimensional Laplacian,
d3 dz = 0, or
2

i + y2 z = 0.
(4.29)
dV = 2abh4 3 dG =
299
An additional constraint similar to (4.18), which arises from the F(2) equation of motion, is
still present. This time, however it simply states that dG 3 dG = 0 in the three-dimensional
Euclidean space, so that G is necessarily a constant. As a result, we only obtain the Minkowski
vacuum in this case.
For, say b = 0, on the other hand, the above relations reduce to
h2 = a 2 eG ,
F(2) = 2adeG (dt + V ),
dV = 0.
(4.30)
The equation of motion for F(2) then gives d3 deG = 0, resulting in a solution of the form
(setting a = 1)

ds62 = eG dt 2 + d 2 + eG ij dx i dx j + dy 2 + d 2 ,
1
F(2) = 2 dt d G ,
(4.31)
e
where H = eG is harmonic in R3 spanned by (x 1 , x 2 , y). This solution is in fact of the same
form as (4.25), and corresponds to a multi-centered string solution. This time, however, the
Killing symmetry / is not of an angular type, and both circles have decompactified. As a
result, we have unfortunately been unable to obtain any new 1/2 BPS solutions of the minimal
N = (1, 0) system beyond the already familiar multi-centered string solutions.
4.2. S 1 S 1 reduction in the presence of a tensor multiplet
We now explore the possibility of evading the previous conclusion about the non-existence
of a bubbling AdS3 S 3 solution, within the boundary of our ansatz, by enlarging the set of
fields, from minimal 6-dimensional supergravity, to supergravity coupled to a tensor multiplet.
and dilatino .
The field content of the tensor multiplet is: dilaton , anti-selfdual tensor H
The dilatino is Weyl, with opposite chirality than that of the gravitino.
We continue to work with the same metric ansatz as before, while the 3-form ansatz becomes
H(3) = (F(2) + K(2) ) d1 + (F(2) + K (2) ) d2 ,
(4.32)
+
where F(2) = eG 4 F(2) and K (2) = eG 4 K(2) . Thus F, F form the self-dual 3-form H(3)
,
while K, K define the anti-selfdual 3-form H(3)

. The Bianchi identity and equation of motion
read

dH(3) = 0,
(4.33)
d e 6 H(3) = 0.
Assuming that = (x), in terms of the reduced form fields, these equations become
d(F(2) + K(2) ) = d(F(2) + K (2) ) = 0,
d (F(2) K(2) ) + d(F(2) K(2) ) = d (F(2) K (2) ) + d(F(2) K (2) ) = 0. (4.34)
The 1 2 component of the Einstein equation yields the constraint
F(2) F(2) K(2) K(2) = 0.
(4.35)
The supersymmetry variations of the supergravity multiplet are the same as before, (3.9),
with the exception that F , F must be replaced by e/2 F , e/2 F . Correspondingly, the spinor
bilinear equations (C.1)(C.3) are modified by means of the same replacement. The immediate consequence of this observation is that we obtain as before f2 = exp((H + G)/2),
300
f1 = exp((H G)/2), K is a Killing vector and L dx is still a closed form. Also, the
relation h2 = f12 + f22 holds as well. Therefore the metric has become once more ds 2 =
h2 (dt + V )2 + h2 (dy 2 + 2i,j =1 hij dx i dx j ) + eH +G d12 + eH G d22 . We have learned also

that
iK F(2) = eH +G e/2 d(H + G),
iK F(2) = eH G e/2 d(H G),
(4.36)
where iK denotes the inner contraction with the Killing vector K = h2 (dt + V ), and

1 H +G /2
(4.37)
e F(2) + eH G e/2 F(2) .
e
2
Substituting K as well as (4.36) into the previous equation we find the same differential equation
defining z = 12 tanh(G) as (4.16). Thus the metric is identical to the one derived previously in the
absence of the tensor multiplet.
Consistency of (4.37), namely d 2 K = 0, combined with Bianchi and equations of motion
(4.34) leads to
dK =
K(2) K(2) = 0,
(4.38)
which in turn implies that F(2) F(2) = 0. We see that despite our efforts to avoid the F(2) F(2)
constraint which translates into the additional non-linear differential equation that z(x1 , x2 , y)
had to satisfy, we have to conclude that turning on the tensor multiplet did not achieve, as one
might have hoped, a bubbling AdS3 S 3 picture. As mentioned in the introduction, a possible
way to evade the negative conclusion on bubbling AdS3 S 3 solutions is to allow for a 4d axion
field (arising from g1 2 ). In fact, a rather large class of supersymmetric 6d solutions of conical
defect type [20] fall into this class of metrics.
5. Bubbling 1/4 BPS solutions: turning on an axiondilaton
In this section we show how the 1/2 BPS family of solutions discovered recently by Lin,
Lunin and Maldacena [1] can be modified to accommodate a holomorphic axiondilaton field.
Of course, in doing so we break the amount of supersymmetry that the new solutions preserve
by half. We will end up with a family of 1/4 BPS solutions which have the same SO(4) SO(4)
isometries inherited from the 1/2 BPS family.
Our interest in this class of 1/4 BPS solutions resides in its implications for the dual gauge
theory. We expect that turning on the axiondilaton field , which amounts to adding D7 branes
by appropriately including their back-reaction, will lead to the addition of flavor degrees of freedom to the dual gauge theory. By embedding Nf D7 branes in the initial AdS5 S 5 geometry,
one adds Nf N = 2 hypermultiplets, Q, in the fundamental of Nc to the N = 4SU(Nc ) dual
gauge theory. The gauge theory superpotential is accordingly modified by the addition of the
hypermultiplets to Tr X[Y, Z] + QZQ.

More precisely, we begin our construction of 1/4 BPS solutions in type IIB supergravity with
the following ansatz:
ds 2 = g dx dx + eH +G d32 + eH G d 32 ,
F(5) = F dx dx d3 + F dx dx d 3 ,

= x 1 + ix 2 , with x = t, y, x 1 , x 2 .
(5.1)
(5.2)
(5.3)
301
As in [21], we will be able to exploit the fact that the D3D7 problem separates, with the D7
branes curving the space transverse to them, and the warping due to the D3 branes modified to
accommodate the D7 branes backreaction. The self-duality condition and the Bianchi identity
of the 5-form imply for the reduced form fields
F = e3G 4 F ,
F = dB,
F = d B.
(5.4)
Requiring that this solution is supersymmetric, we impose

i
i
FM M M M M M1 M2 M3 M4 M5 = 0,
M = M QM +
2
480 1 2 3 4 5
= iPM M = 0,
(5.5)
(5.6)
where M and are the complex gravitino and dilatino, whose U (1) charges are 1/2 and 3/2,
respectively. The axion and dilaton fields parameterize a scalar coset SL(2, R)/U (1), with the
U (1) connection given by
QM =
1 M
,
2
(5.7)
and where g MN PM PN represents the kinetic term of the sigma-model Lagrangian, with
1 M
.
(5.8)
2
Notice that the previous supersymmetry variations of the gravitino along the sphere directions,
(2.15), (2.16) are not modified by the presence of the scalar fields, and that (2.14) contains a new
term due to the Q-connection. The new constraint following from the supersymmetry variation
of the dilatino only enforces

1
+ i 2 = 1 + i 2 = 0.
(5.9)
PM =
The spinor bilinear equations derived previously (C.1) are unchanged, because a bilinear of the
type ( . . . ) is U (1) neutral. However, the one-form
= T C dx ,
(5.10)
=
where C is the charge conjugation matrix
case in the absence of the axiondilaton field; rather it obeys
(
C ,T C),
is no longer closed as it was the
d = iQ .
(5.11)
Given that K is still a Killing vector and L dx is still an exact form, we can choose as
before K as the generator of time translations, K t = 1, and we choose L = dy, with = 1.
Therefore we arrive at the same form of the metric

2

ds42 = h2 dt + Vi dx i + h2 dy 2 + h ij dx i dx j .
(5.12)
We can see now that the constraint (5.9) has become a projection condition on

1 + i 12 = 0.
(5.13)
Using the Killing spinor equations we find the following set of equations
(H G)/2 L ,
f2 H = e
(5.14)
302
1
f2 = e3(H +G)/2 F K = f2 (H + G),
2
which allows the integration of both the spinor bilinear f2 and H as
f2 = 4e(H +G)/2 ,
Bt = e2(H +G) ,
eH = y,
(5.15)
(5.16)
where we fix the sign of such that

= 1. Similarly we find
f1 = 4e(H G)/2 ,
B t = e2(H G) .
(5.17)
With the choice 4 = 1, we end up by fixing using on the one hand (4.10)
h2 = f12 + f22 ,
(5.18)
and
2
y e = h

f22
f12
=
,
4 4
(5.19)
on the other hand. The latter equation is obtained from (2.15) by multiplication with 5 . We
conclude that the last two equations imply 4 = 1, and = .
Substituting H into the Killing
spinor equation (2.15) we identify another projector

1 3
5 e(H G)/2
+ e(H +G)/2 i
0=
yh

= 1 + e2G 3 ieG 5 + 1 .
(5.20)
Moreover, using that
K t = h 0 = h = 1,
Ly = h 3 5 = ,
(5.21)
one derives the last projector

0 3 5 = i 12 = 0.
(5.22)
However, we do not have the freedom of two choices of sign for , because of the first projection
condition that we encountered (5.9) from the susy variation of the dilatino which identifies
= 1.
(5.23)
Thus our solution ends up preserving only 1/4 of the 32 supersymmetries.

Finally, using the projectors we can solve for the Killing spinor
= e(H +G)/4 exp(i5 3 )0 ,
sinh = eG ,
0 = 1.
(5.24)
Substituting the Killing spinor in (5.11), where still the only non-vanishing components are 1
and 2 , we realize that we end up with a conformally flat two-dimensional space in the x 1 , x 2
directions:
2
2
1 2
h ij dx i dx j = e (x ,x ) dx 1 + dx 2 ,
(5.25)

(x 1 ,x 2 )
2
=e
(5.26)
d x1 + i dx .
Moreover, the conformal factor e is related to the axiondilaton field, because we argued earlier
that the U (1) connection Q becomes the spin connection in this two-dimensional space
(x1 , x2 ) = (x1 + ix2 ).
(5.27)
303
In fact, there is even more freedom in defining (x1 , x2 ) in terms of multiplication by a holomorphic and an antiholomorphic function exp( ) = exp(f (x1 + ix2 ) + f (x1 ix2 )). This
stems from the freedom of multiplying 0 by a phase: exp((f f )/2)0 . Imposing modular invariance, with Nf D7 branes located at Zi = (x1 + ix2 )i = 0, uniquely determines the conformal
factor as
N
f

4
e (Z,Z ) = ( )
|Z Zi |1/6 .
(5.28)
i=1
The corresponding geometry is non-singular provided that Nf < 24. Near the D7 branes, the
axiondilaton field behaves as a logarithm, and its equation of motion has delta-function singularities at the location of the D7 branes
i
1
(Z) +
(5.29)
ln(Z Zi ),
g 2i
i

1
1
ln |Z Zi | .
exp( )
(5.30)
g 2
i
We are left only with determining the 5-form flux: given the similarity of our equations to
those of [1], it is easy now to see that the bubbling 1/4 BPS solutions read

1
2
e i z + yy
y z = 0,
y
1
1
dV = 3 dz, z = tanh G,
y
2
F = d B (dt + V ) + B dV + d B,
F = dB (dt + V ) + B dV + d B,
t
1
1
B t = y 2 e2G ,
Bt = y 2 e2G ,
4
4
1
z
+
1/2
z 1/2
1 3
3
d B =
,
d B =
,
y 3 d
y 3 d
2
4
4
y
y2
(5.31)
where one should keep in mind that the
Hodge symbol 3 in the three-dimensional space is

computed relative to the metric dy 2 + e i=1,2 (dx i )2 .
To gauge the effect of the axiondilaton field on the geometry, we can in a first order of
approximation solve the differential equation which defines the auxiliary function z(x1 , x2 , y)
perturbatively in Nf . We assume that all D7 branes are overlapping and we approximate the
conformal factor by (5.30). We define polar coordinates in the (x 1 , x 2 ) plane, and we redefine
the radial coordinate by r = e/2 . The effect of this rescaling is to map the line element ds22 =
N
e (d 2 + 2 d 2 ) into ds 2 (1 2f )(dr 2 + r 2 d 2 ). This is nothing else but a 2-dimensional

space with a deficit angle. Therefore z(x1 , x2 , y)/y 2 is to a first order of approximation still a
harmonic function, but it is a harmonic function of a 6-dimensional space, with a deficit angle in
the 2-plane defined by x1 , x2 .
Therefore the presence of the D7 branes, while not affecting to a dramatic degree the bubbling
AdS5 S 5 picture, so that in particular it does not change the interpretation of z(x1 , x2 , y = 0) =
1/2 as a phase space, nevertheless induces a deficit angle in this plane. Since the fluctuations
of the D3 branes in the direction Z = x1 + ix2 become in the decoupling limit the BPS chiral primary operators defining the gauged quantum mechanics matrix model of [2,3], a deficit angle in
304
the (x1 , x2 ) planes translates into a non-trivial monodromy of the chiral primary operators. This
ultimately implies that the eigenvalues of Z have a non-trivial monodromy, or equivalently, the
electrons participating in the quantum Hall effect (i.e. the eigenvalues) have fractional statistics.
Acknowledgements
We would like to thank F. Larsen for many illuminating discussions throughout the course of
this project, as well as A. Sinkovics for comments. We are also grateful to M. Perry for bringing
Weyl solutions to our attention. This work was supported in part by the US Department of Energy
under grant DE-FG02-95ER40899.
Appendix A. Bosonic reductions on S n S n
The reduction of the bosonic fields (metric and form-field) may be performed in arbitrary dimensions. For a reduction to four dimensions on S n S n , we start with a D = 4 + 2n dimensional
bosonic action of the form
e1 L = R
1
F2 .
2 (n + 2)! (n+2)
The resulting equations of motion are simply

2
1
1
F MN
gMN F 2 ,
RMN =
2(n + 1)!
2(n + 2)
dF = dF = 0.
(A.1)
(A.2)
Note that here we have taken F(n+2) to be canonically normalized. Furthermore, at this stage we
do not impose self-duality on the form-field, although below we will show what modifications
would be necessary to cover the self-dual case.
The reduction of the equations of motion, (A.2), proceeds by taking an ansatz of the form

ds 2 = g (x) dx dx + eH (x) eG(x) dn2 + eG(x) d n2 ,
F(n+2) = F(2) n + F(2) (x) n ,
(A.3)
where n and n are volume forms on the respective unit n-spheres. Our goal is now to obtain
the four-dimensional effective theory for the fields g , H , G, F(2) and F(2) .
At this point, it is worth recalling that, for sphere compactifications, the fields H and G
are actually breathing mode scalars which live in the massive KaluzaKlein tower. In general, it
would be inconsistent to retain only a subset of the massive KaluzaKlein states, as they typically
source each other ad infinitum. However, the scalars H and G themselves are uncharged on the
spheres, and hence this breathing mode reduction remains consistent by virtue of retaining only
singlets on the spheres [12,13].
We begin by reducing the form-field equation of motion. From the ansatz (A.3), we may
obtain the Hodge dual
F(n+2) = 4 enG F(2) n + ()n 4 enG F(2) n .
(A.4)
At this point, we note that, since 4 4 = 1, we may only impose a self-dual condition on F(n+2)
for odd n dimensions (i.e. D = 6 or 10). In such dimensions, self-duality yields the relation
305
F(2) = 4 enG F(2) . In any case, we see that the F(n+2) equation of motion reduces simply to
dF(2) = 0,
d F(2) = 0,

d enG 4 F(2) = 0,

d enG 4 F(2) = 0.
(A.5)
Turning to the Einstein equation, we first compute the Ricci tensor for the metric ansatz (A.3)
n
(D)
= R ( H H + G G) n H,
R
2
n
1
(D)
Rab = R ab gab H (H + G) gab (H + G),
2
2
n
1
(D)
R = R a b ga b H (H G) ga b (H G).
a b
2
2
(A.6)
Here R ab = (n 1)g ab and R a b = (n 1)g a b are the curvatures on the unit n-spheres with
metrics g ab and g a b , respectively. The D-dimensional metric components in the sphere directions
are gab = eH +G g ab and ga b = eH G g a b .
For the form-field, we compute

1
2
F(n+2)
= (n + 2)! en(H +G) F 2 + en(H G) F 2 ,
2

2
F(n+2) = (n + 1)! en(H +G) F 2 + en(H G) F 2 ,
2
1
F(n+2) ab = (n + 1)!en(H +G) gab F 2 ,
2
2
1
F(n+2) a b = (n + 1)!en(H G) ga b F 2 ,
2
2
F(3) a a = ea ea F F for n = 1.
(A.7)
These expressions allow us to work out the source for the Einstein equations. They may be
combined with (A.6) to obtain the four-dimensional equations of motion
n
R = ( H H + G G) + n H
2

1
1
1
1
2
2
g F 2 + en(H G) F
g F 2 ,
+ en(H +G) F
2
4
2
4
H + n H H = 2(n 1)eH cosh G,
1
1
G + n H G = en(H +G) F 2 + en(H G) F 2 2(n 1)eH sinh G.
4
4
(A.8)
The scalar equations were separated by taking appropriate linear combinations of the Rab and
Ra b equations. In addition, for n = 1 (D = 6), the reduction of the Ra a Einstein equation yields a
constraint F F = 0. In general, this signifies an inconsistency in the reduction. However, so
long as we satisfy this constraint, we are ensured that solutions to the effective four-dimensional
theory may be lifted to solutions of the original six-dimensional theory.
The equations of motion, (A.5) and (A.8), may be derived from an effective four-dimensional
Lagrangian
306

1
1
1
L=e
R + n(2n 1)H 2 nG2 en(H +G) F 2
2
2
4

1 n(H G) 2
H
e
F + 2n(n 1)e cosh G .
4
nH
(A.9)
If desired, this may be transformed into the Einstein frame by the Weyl transformation g =
enH g . The resulting Einstein frame action has the form
1
1
1
1
e1 L = R n(n + 1)H 2 nG2 enG F 2 enG F 2
2
2
4
4
+ 2n(n 1)e(n+1)H cosh G.
(A.10)
Note that the scalar fields are not canonically normalized. Nevertheless,
we find it convenient to
retain this convention, so as to avoid unpleasant factors of n and n(n + 1).

Finally, for the reductions we have considered, the form field F(n+2) is taken to be self dual
in D = 10 or 6. Reducing the self-dual field follows the procedure given above, so long as we
impose the self-dual condition after obtaining the equations of motion, (A.2). In this case, the
F 2 term vanishes, and we are left with an Einstein equation of the form
RMN =
2
1
F MN .
2(n + 1)!
(A.11)
Note that, if canonical normalization is desired, we ought to include an additional factor of 1/2
in the field-strength term of the original Lagrangian, (A.1), in which case the right-hand side of
(A.11) must also be multiplied by 1/2. This is indeed what we do for the IIB theory. However,
we forego this factor of 1/2 for the N = (1,
0) model in six dimensions. This choice of a noncanonically normalized 3-form H(3) avoids 2 factors in the supersymmetry transformation (3.3)
of Section 3 (and furthermore keeps canonical normalization in the case where the N = (1, 0)
theory is coupled to a single tensor multiplet).
Regardless of normalization, for the self-dual case, we impose the condition F(2) =
4 enG F(2) to eliminate F(2) in favor of F(2) in the reduced equations of motion. For canonical
self-dual normalization, which incorporates the additional factor of 1/2 introduced above, this
simply amounts to erasing all F terms from the expressions in (A.8). The resulting effective
Lagrangians are identical to (A.9) and (A.10), except with the F 2 terms removed. If we instead
leave out the factor of 1/2, the resulting F 2 terms are twice as large (and the F 2 terms are absent
as usual). Here, we see the familiar feature that while F cannot be dualized in the Lagrangian, it
is valid to do so in the equations of motion.
In addition, for self-duality in D = 6, the constraint F F = 0 is replaced by F(2) F(2) =
0. Here it is clear that F(2) F(2) would ordinarily source an axionic field. However, by truncating away all axions, we can no longer allow such a source. Again, so long as we impose this
constraint by hand, we will still be able to obtain solutions to the original six-dimensional model.
Appendix B. Integrability of the Killing spinor equations
Since we have found the somewhat surprising result that the Killing spinor equations resulting
from both S 3 S 3 compactification of IIB supergravity and S 1 S 1 compactification of the
N = (1, 0) theory have very similar forms, it is interesting to see how they can lead to different
equations of motion, namely (A.8) with either n = 3 or n = 1. In order to see how this works, we
may consider the integrability of the Killing spinor equations, (4.1), repeated here as
= D ,
H = H ,
G = G ,
307
(B.1)
where
i n (H +G)
F ,
e 2
16
1
1
1
H = H + e 2 H e 2 G i
5e 2 G ,
1
1
1
i n
G = G + e 2 H e 2 G + i
(B.2)
5 e 2 G e 2 (H +G) F .
4
In the original theory (either in D = 10 or 6), there is only one object to examine for first
order integrability, namely [DM , DN ]. However, viewed in the effective four-dimensional point
of view, we may consider the various commutators of D , H and G . We begin with [D , D ].
After straightforward although tedious manipulations, we find

1
n + 1 n(H +G) 2
1
[D , D ] = R
e
F g F 2
2
8
4

n
( H H + G G) n H
2

n
i
i n
+ e 2 (H +G) [ F ] + e 2 (H G) enG F
16
8
n
n
n
+ [D , H ] + H H + GG
2
4
4
in n (H +G)
e 2
F (H G ).
(B.3)
32
Since the last two lines above vanish on Killing spinors, we see that this integrability yields the
Einstein equation as well as Bianchi and equation of motion for F(2) . In particular, if the latter two
conditions are imposed on F(2) , then the Einstein equation is guaranteed by supersymmetry. Note
also that the Einstein equation of (A.8) is reproduced with proper dimension dependent (n = 1
or 3) coefficients. This also shows the curious fact that, starting from an identical normalization
of F(2) in the supersymmetry variations, (B.2), one in fact obtains different normalizations in the
bosonic equations involving F(2) .
Turning to the [D , H ] condition, we find

[D , H ] = H + nH 2 (n 1)eH 2 eG + 2 eG

i n
+ n H e 2 (H +G) F
8

1
1 1H 1G
+ n
5 e 2 G H
e 2 e 2 + i
2
1
1 1H 1G
e 2 e 2 i
5 e 2 G G ,
(B.4)
2
which simply reproduces the H equation of motion in (A.8). In particular, for n = 1 we are
required to choose 2 = 2 = 1, while for n = 1 these U (1) charges are irrelevant. This demonstrates that the identical bosonic equations are satisfied, regardless of the KaluzaKlein charges
carried by the Killing spinors.
At this stage, it is also worth noting that we may form the combination

1
1
1
H e 2 H e 2 G + i
(B.5)
5 e 2 G H = H 2 eH 2 eG + 2 eG .
D = +
308
When acting on Killing spinors, this demonstrates that supersymmetry further imposes the condition

H 2 = eH 2 eG + 2 eG .
(B.6)
Combining this with the equation of motion of H yields the simple expression
H + H 2 = 0,
(B.7)
which must be satisfied on supersymmetric backgrounds.

The final integrability condition we obtain is the one between D and G . In this case, we
obtain
[D , G ] = G + nH G +
+
+
+
+

n + 1 n(H +G) 2
F (n 1)eH 2 eG 2 eG
e
16
3 n n(H +G)
F F
e
32
i n (H +G)
i n
[ F] e 2 (H G) (enG F )
e 2
4
2
1
1
1
1
5 e 2 G H
n G + (n 1)e 2 H e 2 G i
2
1
1
1
1
5e 2 G
n H + (n 1)e 2 H e 2 G + i
2

i(n 1) n (H +G)
F G .
e 2
4
(B.8)
In addition to the G equation as well as Bianchi and equation of motion for F(2) , we see that the
F(2) F(2) = 0 constraint shows up in this integrability condition provided n = 3. So, at least for
the N = (1, 0) theory, supersymmetry implies not just the equations of motion of (A.8), but also
the F(2) F(2) = 0 constraint.
More precisely, for partially broken supersymmetry, the Killing spinor equations often yield
only linear combinations of the equations of motion. In this case, we see from (B.4) that both the
H equation as well as the H constraint (B.7) are automatically satisfied independent of the rest of
the fields. However, from (B.3) we see that the Einstein equation is only satisfied in combination
with the F(2) equations, and similarly from (B.8), that the G equation of motion is satisfied
in combination with the F(2) equations. We may conclude that, for obtaining supersymmetric
backgrounds, it would be sufficient to satisfy the F(2) Bianchi identity and equation of motion in
addition to the Killing spinor equations themselves.
Finally, while the supersymmetry transformations (4.1) were only obtained for the cases n = 3
and 1, they may nevertheless be formally extended to any value of n. Examination of the integrability conditions (B.3), (B.4) and (B.8) indicate consistency with a bosonic sector described
by an effective Lagrangian

1
n + 1 n(H +G) 2
1
1
nH
e L=e
F
e
R + n(2n 1)H 2 nG2
2
2
16

+ n(n 1)eH 2 eG + 2 eG .
(B.9)
For n = 3 this system must be extended with the constraint F(2) F(2) = 0.
309
Appendix C. Differential identities for the spinor bilinears

The supersymmetric construction of [1619] proceeds by postulating the existence of a
Killing spinor and then forming the tensors f1 , f2 , K , L and Y from spinor bilinears
(4.2). The algebraic identities of interest were given in the text in (4.3). Here we tabulate the
differential identities obtained by demanding that solves the Killing spinor equations (4.1).
First, by assuming = 0, we may demonstrate that
1 n
f1 = e 2 (H +G) F K ,
4
1 n
f2 = e 2 (H +G) F K ,
4
1 n (H +G)
(f2 F f1 F ),
K = e 2
4

1 n (H +G) 1
2
L = e
g F Y 2F( Y) ,
4
2

1 n
Y = e 2 (H +G) 2g[ F] L 2F[ L] + F L .
(C.1)
4
In particular, the equation for K indicates that K(;) = 0, so that K is Killing. This is in fact
a generic feature of constructing a Killing vector from Killing spinors.
In addition, the H = 0 condition allows us to derive the additional relations
K H = 0,
f2 = e
G f1 ,
1
L H = e 2 (H +G) f1 e
2 (H G) f2 ,
1
e 2 (H +G) L = f1 H,
1
e 2 (H +G) K = Y H,
2L[ ] H = 0,
e
2 (H G) L = f2 H,
1
e
2 (H G) K = Y H,
1
2K[ ] H = e 2 (H +G) Y + e
2 (H G) Y .
(C.2)
Similarly, the G = 0 condition yields the relations

1 1n (H +G)
F Y = f2 e
G f1 ,
e 2
4
1
1
1 n
2 (H G) f2 e 2 (H +G) F Y ,
L G = e 2 (H +G) f1 + e
4
1
1 n
e 2 (H +G) L = f1 G + e 2 (H +G) F K ,
2
1 n (H +G)
12 (H G)
L = f2 G + e 2
F K ,
e
2
1
1 n
e 2 (H +G) K = Y G + e 2 (H +G) F L ,
2
1 n
12 (H G)
K = Y G + e 2 (H +G) F L ,
e
2
n2 (H +G)
2L[ ] G = 2e
F[ Y] ,
1
1
1 n
2 (H G) Y e 2 (H +G) (f1 F + f2 F ).
2K[ ] G = e 2 (H +G) Y e
2
K G = 0,
(C.3)
310
Although the above identities are algebraic and not differential on the spinor bilinears, they
originate from the supersymmetry variations along the internal directions of the KaluzaKlein
reduction. So in this sense, they form a generalized set of differential identities. However, as
they are only algebraic, they prove extremely useful in determining much of the geometry, as is
evident from the analysis of [1].
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
H. Lin, O. Lunin, J. Maldacena, JHEP 0410 (2004) 025, hep-th/0409174.

S. Corley, A. Jevicki, S. Ramgoolam, Adv. Theor. Math. Phys. 5 (2002) 809, hep-th/0111222.
D. Berenstein, JHEP 0407 (2004) 018, hep-th/0403110.
D. Berenstein, hep-th/0409115.
M.J. Duff, H. Lu, C.N. Pope, Phys. Lett. B 409 (1997) 136, hep-th/9704186.
H. Weyl, Ann. Phys. 54 (1917) 117.
R. Emparan, H.S. Reall, Phys. Rev. D 65 (2002) 084025, hep-th/0110258.
D. Martelli, J.F. Morales, JHEP 0502 (2005) 048, hep-th/0412136;
J.T. Liu, D. Vaman, hep-th/0412242.
A.H. Chamseddine, M.S. Volkov, Phys. Rev. D 57 (1998) 6242, hep-th/9711181.
M. Cvetic, H. Lu, C.N. Pope, Nucl. Phys. B 574 (2000) 761, hep-th/9910252.
D.Z. Freedman, J.H. Schwarz, Nucl. Phys. B 137 (1978) 333.
M.S. Bremer, M.J. Duff, H. Lu, C.N. Pope, K.S. Stelle, Nucl. Phys. B 543 (1999) 321, hep-th/9807051.
J.T. Liu, H. Sati, Nucl. Phys. B 605 (2001) 116, hep-th/0009184.
A. Salam, E. Sezgin, Phys. Lett. B 147 (1984) 47.
M.J. Duff, H. Lu, C.N. Pope, Nucl. Phys. B 544 (1999) 145, hep-th/9807173.
J.P. Gauntlett, D. Martelli, S. Pakis, D. Waldram, Commun. Math. Phys. 247 (2004) 421, hep-th/0205050.
J.P. Gauntlett, J.B. Gutowski, C.M. Hull, S. Pakis, H.S. Reall, Class. Quantum Grav. 20 (2003) 4587, hepth/0209114.
J.B. Gutowski, D. Martelli, H.S. Reall, Class. Quantum Grav. 20 (2003) 5049, hep-th/0306235.
J.P. Gauntlett, D. Martelli, J. Sparks, D. Waldram, Class. Quantum Grav. 21 (2004) 4335, hep-th/0402153.
O. Lunin, S.D. Mathur, A. Saxena, Nucl. Phys. B 655 (2003) 185, hep-th/0211292.
B.A. Burrington, J.T. Liu, L.A. Pando Zayas, D. Vaman, hep-th/0406207.
Low temperature correlation functions in integrable

models: Derivation of the large distance and time
asymptotics from the form factor expansion
B.L. Altshuler a,b,c , R.M. Konik d , A.M. Tsvelik d,
a Physics Department, Princeton University, Princeton, NJ 08544, USA
b Physics Department, Columbia University, New York, NY 10027, USA
c NEC-Laboratories America, Inc., 4 Independence Way, Princeton, NJ 085540, USA
d Department of Physics, Brookhaven National Laboratory, Upton, NY 11973-5000, USA
Received 29 August 2005; received in revised form 29 October 2005; accepted 17 January 2006
Abstract
We propose an approach to the problem of low but finite temperature dynamical correlation functions in
integrable one-dimensional models with a spectral gap. The approach is based on the analysis of the leading
singularities of the operator matrix elements and is not model specific. We discuss only models with welldefined asymptotic states. For such models the long time, large distance asymptotics of the correlation
functions fall into two universality classes. These classes differ primarily by whether the behavior of the
two-particle S-matrix at low momenta is diagonal or corresponds to pure reflection. We discuss similarities
and differences between our results and results obtained by the semi-classical method suggested by Sachdev
and Young [S. Sachdev, A.P. Young, Phys. Rev. Lett. 78 (1997) 2220].
PACS: 71.10.Pm; 72.80.Sk
1. Introduction
In this paper we discuss the problem of finite temperature time-dependent correlation functions in integrable models. We do not consider here one-dimensional models with gapless linear
E-mail address: tsvelik@bnl.gov (A.M. Tsvelik).

doi:10.1016/j.nuclphysb.2006.01.022
312
B.L. Altshuler et al. / Nuclear Physics B 739 [FS] (2006) 311327
spectra, for which there exists a general solution: the Matsubara n-point correlation functions
at finite temperature can be obtained from the corresponding T = 0 correlators by a conformal
transformation. Instead we focus on models with spectral gaps. Finite temperature dynamics for
that class of models is not yet fully understood. The spectral gap, M, provides an energy scale
separating the high temperature (T M) and low temperature (T M) regimes. In the former
regime the leading order physics reduces to that of a conformal field theory at high temperatures.
We thus concentrate on the low temperature regime T M. In this limit we will argue that the
relevant physics is governed by the zero momentum limit of the interactions between excitations
in the theory. These interactions reflect the integrable nature of the models: the N -particle scattering matrix is a product of the two-particle ones. It is known that the zero momentum limit
of the two-particle scattering matrix, S(0), in an integrable system can be either diagonal, as it
occurs, for instance, in the quantum Ising model, or equal to the permutation operator. The latter
alternative is realized for the solitons of the sine-Gordon model at general values of the coupling
constant. This dichotomy suggests the existence of two universality classes for the low temperature dynamics. As we will demonstrate in this paper, this expectation is met for a class of order
parameter-like operators.
The most advanced methods of evaluating correlation functions for integrable models with
spectral gaps are based on the form factor technique (see [13] for reviews). This technique
allows one to calculate matrix elements (form factors) of the operators between the exact eigenstates entering the Lehmann expansion for the correlation functions. At finite temperature this
Lehmann expansion for the two-point correlation function of the operator, O, has the following
form:

1
GO (x, t) = Tr eH O(x, t)O(0, 0)
Z

Esn n, s |O(x, t)O(0, 0)|n, s
n
n
nsn e

=
Esn n, s |n, s
e
n
n
nsn

E
s
n
n, sn |O(x, t)|m, sm m, sm |O(0, 0)|n, sn
n,sn ;m,sm e

.
=
(1)
Esn n, s |n, s
n
n
nsn e
Here the state, |n, sn , denotes a set of n-particles carrying quantum numbers, {sn }. The final
expression in the above involves a double sum. The first sum appears as a sum weighted by
Boltzmann coefficients while the second sum arises from inserting a resolution of the identity
between the two fields. The two-point correlation function has thus been reduced to a question
of computing and summing form factors.
The form factor approach turns out to be highly effective for T = 0 two-point functions. This
is particularly so in computing spectral functions where at finite energies the sum corresponding
to that above reduces to an expression with a finite number of readily computable terms. However, its generalization to multi-point functions, as well as for T = 0 faces difficulties caused
by singularities that appear in the needed form factors [1]. In this paper we propose a method
of summation of these singularities that is suited to the extraction of the long distance, long
time, low temperature asymptotics of the correlation functions. In a particular case, our results
reproduce the semi-classical formulae obtained in Refs. [5,6,21]. In other instances, we find
divergences between our formulae and those derived using the semi-classical methodology proposed in Refs. [5,6,21].
The outline of the paper is as follows. In Section 2 we consider the simplest of the integrable
models, the Ising model. Using expressions obtained in Ref. [10] from a lattice analysis of this
313
model, we analyze the asymptotics of the spinspin correlation functions. In doing so we are able
to cast the spinspin correlation function in a form nearly identical to what would be found using
the Lehmann expansion. We take this similarity as instructive: it both indicates how to handle
singularities that appear in the Lehmann expansion and it tells us how to proceed with more general integrable models. Thus in Section 3, we generalize our computations to arbitrary integrable
models. We find two universality classes of behavior. One class of behavior is reminiscent of the
Ising model. Here we find no discrepancies between our approach and that of the semi-classical
approach. For the second universality class we find novel scaling behavior that is not reproduced
semi-classically. We consider the implications of these two universality classes for the physics of
the sine-Gordon model in Section 4. Finally in Section 5 we devote an in-depth discussion to the
discrepancy between methods seen for models and operators that fall into the second universality
class.
2. Finite temperature spinspin correlation functions in the Ising model
To understand how to deal with the singularities that appear in the Lehmann expansion, we
first turn to the quantum Ising model. In the following section we will extend our approach to
other integrable models.
We first review the basics of the model (see Ref. [7], for example). The Hamiltonian, H , of
the Ising model is equivalent to a Hamiltonian of noninteracting massive fermions

z
J nz n+1
hzn1/2 zn+1/2 + J xn
+ hnx
(p)Fp+ Fp ,
H=
n
(J h)2 + 4J h sin2 (p/2),

(2)

where zn+1/2 = j <n jx is the disorder operator and F, F + are fermion annihilation and creation operators. The fermions represent walls between domains with different orientations of the
magnetization and so are to be thought of as solitons.
Calculations simplify in the continuum limit, when the spectral gap
M = |J h| is much
smaller than the bandwidth J and the spectrum is relativistic (p) = c2 p 2 + M 2 , c2 = J h.
In this case, energy and momentum of a quasi-particle are conveniently parameterized by a rapidity, (cp = M sinh ). Then the eigenstates of Hamiltonian (2) are labeled by sets of rapidities,
{i }, such that the energy and momentum of the system are equal to
(p) =
E=M
n

cosh i ,
P = c1 M
i=1
n

sinh i .
(3)
i=1
z (we
will call them and ) have infinitely many

Below we set c = 1. Operators z and
matrix elements. By Lorentz invariance, the form factors depend on rapidity differences [4]
1 , . . . , n | |1
, . . . , m

= AM 1/8
tanh(ij /2)
tanh(pq
/2)
coth (i p
i)/2 ,
i<j
p<q
(4)
i,p
where at h < J (the ordered phase) n + m is even for and odd for (for h > J it is the other
way around). Here ij = i j and A is a known numerical constant. The singularities in (4)
appear as soon as some in and out rapidities coincide. Singularities of this kind are known as
kinematical poles and appear routinely in integrable models. There are operators however where
314
the matrix elements are free of singularities. In the Ising model such an operator is the energy
density operator

+
x (x) =
(5)
eiqx (k) (k q)Fk Fqk , Fk = Fk
, (k) = 1 + k/(k).
k
We will not in general consider the correlation functions of such operators in the paper. We can
say however their long time, large distance behavior will generally be ballistic.
2.1. Possible approaches
In the Ising model the form factors are known explicitly and it has been a common belief that
their analysis would lead to a better understanding of the general problem. Several approaches to
the problem of the singularities that appear in the Lehmann expansion are possible.
One can try to regularize the singularities by considering a finite system. Understanding
the singularities as arising from an infinite sized system has been advocated both in Ref. [17]
and in Ref. [9]. In the Ising case it is clear how this regulation would proceed. In the infinite
system the singularities arise when the momenta of the in and out states coincide. However if the
system is treated as on a ring of finite size, the in and out states must be quantized differently. The
spin operator in the Ising model connects the Ramond and NeveuSchwarz sectors of the Hilbert
space. In the Ramond sector, the fermionic modings are periodic while in the NeveuSchwarz
sector they are antiperiodic. (These features of the Ising model Hilbert space are reviewed in
Ref. [8].) Thus at finite system size the momenta simply cannot coincide and the expressions are
singularity free. However the summation of form factors is then discrete not continuous. To the
best of our knowledge no one has then succeeded in explicitly performing the necessary double
summation.
In a similar spirited approach, one can interchange imaginary time and space axes using
a Wick rotation. The finite T Matsubara Greens functions of the infinite system then become
zero temperature correlators for a system on a circle of circumference, 1/T . While the corresponding form factor expansion is manifestly free of singularities, one faces the problem of
calculating form factors and excitation energies in a system of finite size. For the Ising model
this was achieved in Ref. [10]. The two-point correlation functions of s and s were obtained
rigorously in the form of series. Below we use these series to extract the long time and distance
asymptotics of the dynamical correlation functions.
One can calculate correlations by re-expressing them as the determinant of an integral operator of the Fredholm type. In this way results were obtained for the spin-1/2 XX magnet [12]
and the Bose gas [13]. For a general exposition on this methodology see Ref. [14].
Exploiting determinantal expressions of the matrix elements arising from explicit expressions for the Bethe eigenfunctions, Refs. [15,16] have obtain expressions for the zero temperature
correlation functions in the Heisenberg XXZ chain. It is likely this approach can be readily extended to finite temperature computations.
In an approach adopted in Ref. [9] one can imagine treating the computation of correlation
functions at finite temperature in terms of a renormalized thermal vacuum and a renormalized
set of thermal excitations. The computation of correlation functions then proceeds in a fashion
similar to that of the zero temperature calculation but with the form factors obeying a revised set
of axioms. More recently, the notion of a thermal vacuum was explored in the work of Ref. [11].
315
The sole method that does not trade on the underlying integrability of the models of concern
was suggested by Sachdev and Young [5] in the form of a renormalized classical approach valid
at finite T and long time. The result for the causal correlation function of the spin field in the
ordered phase valid for T M is

(p)
dp (p)
x
t
e
D(x, t) (t, x) (0, 0) = CM 1/4 exp
(6)
,

p
where = M 2 + p 2 M + p 2 /2M and C is a numerical constant. This expression suggests that the typical correlation lengthscales as t . One can approximate (6) as D(x, t)
C
)max(|x|, v|t|)] where v = T /2M is the thermal velocity of kinks and n(T ) =
exp[n(TM/T
T M/2 e
is the average number of solitons (fermions). For the case of the Ising model,
we show below that Eq. (6) also follows directly from the form factor approach using the results
of Bugrij [10].
The final possible approach involves attacking directly the double sum Lehmann expansion
(Eq. (1)) through the use of infinite volume form factors. To understand how this is to be done in
general, we employ the series expansions from Ref. [10] which will provide the necessary clues
to understanding the singularity structure in this case.
2.2. Long space and time asymptotics of the spinspin correlation function
In the section we use the results of Ref. [10] to understand the asymptotics of the spinspin
correlation function in the Ising model. From this we will be able to follow a natural path to
generalizing these results to integrable models at large.
The result for the Matsubara Greens function of the Ising model order parameter field obtained in Ref. [10] is

(, x) (0, 0)

2N |x|i i qi (qi )

q i qj 2
T 2N
e
1/4 2|x|n(T )
1+
,
= CM e
(2N )! q ,...,q
i
i + j
i>j
N=1
1
2N i=1
(7)
where is imaginary time,

dp (p)
T M/2 eM/T ,
e
n(T ) =
(8)
2
is the soliton density, q = 2T m (m integer), and (q) = M 2 + q 2 . The term in the exponent,
2(q)
(q) =

0
dx
2 (q) + x 2

ln coth (x)/2 ,
(9)
at T M is exponentially small and will be neglected from hereon. The symmetry breaking
transition at T = 0 leads to a finite magnetization, = [CM 1/4 ]1/2 . This is reflected in the
zeroth order term in Eq. (7). We see that this expression for the correlator has the form of a zerotemperature Lehmann expansion: the sum over N represents intermediate states with N particles
while the sum over the qi s are momentum sums for a given set of N particles. The rightmost
term in Eq. (7) can then be interpreted as a finite volume form factor squared. We can see that
the correlation function, Eq. (7), is periodic in , as it must be. The entire expression can then
316
be thought of as a Wick rotation of a computation done in the picture with periodic boundary
conditions in the space direction.
The correlation function of the disorder parameters can be presented in a similar fashion,

(, x)(0, 0)

e|x|i i qi (qi ) qi qj 2
2N+1
T 2N +1
= CM 1/4 e2|x|n(T )
,
(2N + 1)! q ,...,q
i
i + j
i>j
N=0
i=1
1
2N+1
(10)
where now we have a sum over an odd number of qs.
Eqs. (7) and (10) are well placed to rewrite the correlation functions as an expansion in powers
of exp[M]. Focusing on the order-parameter correlation function, we rewrite Eq. (7) representing the N th terms in Eq. (7) as a sequence of contour integrals:

1
e cosh 1 +ix sinh 1
e cosh N +ix sinh N
(11)
d1
dN
tanh2 (i j )/2 ,
cosh
cosh
1 1
1 1
N!
e
e
C
i>j
with the contour C shown on Fig. 1. One can deform this contour into the superposition
C Clow + Cup , where Clow lies just above real axis while Cup corresponds to m = 0.
As a result one obtains a sum of particle (lower contours) and anti-particle (upper contours)
contributions:

N
n
Nn

1
di f (i )e i +ixpi
dj
f (j )e( )j ixpj
n!(N n)!
n=1
i=1
j =1

tanh2 (i k )/2
tanh2 (i
p
)/2
coth2 (i p
+ i0)/2 ,
(12)
i>k
j >p
i,p
1]1
is the Bose distribution function and p = M sinh , = M cosh . In

where
Eq. (12), n and N n are numbers of particles and antiparticles. Note that the singularities of the
form factors are now removed from the real axis to one side of the contour, as was suggested in
Ref. [17]. A formula similar to Eq. (12) was proposed in Ref. [18]. However, there is a noticeable
f () = [e
Fig. 1. The integration contours in the plane. The circles are poles at = i/2 + sinh1 (2 nT /M).
317
difference: the distribution functions turn out to be of the Bose type, although the quasi-particles
of the Ising model are fermions.
To obtain the casual Greens functions, one has to replace with it . From now on we will
be interested in the behavior of the Greens functions at longest real times t M 1 and |x|
M 1 . Only those terms in Eq. (12) which contain both particles and antiparticles are singular.
The leading contribution to the Greens function in each order comes from the vicinity of each
singularity. Each in carries the thermal exponent eM 1. It is thus sufficient to restrict the
summation in Eq. (12) by terms with equal numbers of in and out rapidities, each in rapidity
i being close to one of the out rapidities j
. There are n! ways to choose a pair for each i ;
this factor cancels one of the n! terms from the denominator of Eq. (12). The integral over the
difference i j
can be evaluated with the resulting series summing up to the exponent,

(x, t) (0, 0) = CM 1/4 (t)e2|x|n(T ) exp R(t, x) ,
(13)
where

iMt (12 22 )/2+iMx12
1
M(1+12 /2) e
d
d
e
1 2
2
(12 + i0)2

MeM
2
d eM /2 |x| |x t| .
=
This result reproduces the semi-classical computation found in Eq. (6).

We now turn to how this result determines how we should proceed in the case of general
integrable models.
R(t, x > 0) =
3. The general case: two universality classes

The form of Eq. (12) has a form intimately related to the double summed Lehmann expansion of Eq. (1). Apart the bosonic distribution function f () (unimportant in the limit T M),
Eq. (12) looks for all intent purpose as those subset of terms in Eq. (1) involving form factors (as
given by Eq. (4)) of exactly N particles or excitations. This then implies that the double summed
Lehmann expansion is tractable. Or put another way, we can generalize to correlation functions
involving an operator O in other integrable models by using Eq. (12) with the Ising model form
factors replaced by the form factors of O. At temperatures much smaller than the gap, we replace
the distribution functions by Boltzmann exponents. The asymptotics of the correlation function
is then determined solely by the behavior of the form factors in the vicinity of the kinematic
poles. As we have already pointed out, the existence of kinematic poles is a general consequence
of the theory of form factors [1] rather than a specific feature of the Ising model. These poles
are due to annihilation processes and occur when rapidities of a particle and an anti-particle coincide. The behavior of form factors in the vicinity of the poles divide all integrable models into
two universality classes. Accordingly the resolution of the problem of singularities should not be
model-specific.
Consider a generic matrix element of operator O:
F O (i , a|i b) = 1 , . . . , n |O|1 , . . . , m a1 ,...,an ;b1 ,...,bm ,
(14)
where i and j are rapidities and ai and bj are isotopic indices of in and out particles. One can
O be the irreducible
express the form factor (14) in terms of canonical matrix elements. Let Firr
O
part of F given by the connected diagrams. Only this part is necessary for our calculations done
318
Fig. 2. A graphical representation of Eq. (17) written in terms of the irreducible form factors in Eq. (14) (the rectangles). The up arrows represent rapidities i , while the down ones correspond to the rapidities i + j . Circles at the
intersections of the solid lines represent two-particle S matrices.
O is equal to the matrix element

to lowest order in exp(M). According to Refs. [1719], Firr
between the vacuum and a certain excited state of n + m particles:
F O (i , a|i , b)irr = 0|O|1 , . . . , m ; i + n i0, . . . , i + 1 i0b1 ,...,bm ;a n ,...,a 1 ,

(15)
where a indices are obtained from a ones by charge conjugation. The choice of whether the
small imaginary part, i0, is positive or negative determines the exact form of the disconnected
parts (here ignored) of the form factor in Eq. (14).
The axiom for the annihilation poles in the standard formulation is expressed in terms of the
form factors of the type
FaO1 ,...,an (1 , . . . , n ) = 0|O|n , . . . , 1 an ,...,a1 .
(16)
Namely, as was shown in Ref. [1] (see also Ref. [3] in review), these form factors have poles at
n j = i , j < n, and the residues satisfy the following relation (see Fig. 2):
i Resn =n1 +i FaO1 ,...,an (1 , 2 , . . . , n )

= FaO
,...,a
(1 , . . . , n2 )Can ,a
n1
n2
n3

i1 ,a
Si ,ai i (n1

i )
,a1
I e2ian ,O S1n1
,a1
n3 ,an2
San1 ,an2
(n1
(n1 1 )

n2 ) .
(17)
i=2
Here S is the two-particle scattering matrix, C is the charge conjugation matrix, and a,O is the
semi-locality index between the particle creation operator Aa and O. We will define this index
later.
To demonstrate how Eq. (17) can be used to extract the most singular behavior of the form
factors, we combine it with (14) to obtain the following behavior of two- and three-particle form
factors at the poles:
i(1 e2ib,O )
0|O|0a,b + disconnected part,
1 i0
1 , a1 ; 2 , a2 |O|, a
, a|O|1 , b =
a
,a
= i
a,a2 a1 ,a1
ei2a,O a,a2
Sa 22,a 11 (2 1 )
2 i0
0|O|1 i, a 1

(18)
319
a
,a
a,a1
Sa 22,a 11 (2 1 ) ei2a,O a,a1 a2 ,a2
1 i0
+ disconnected part.
0|O|2 i, a 2

(19)
Now it is appropriate to define the index . We will call two operators, O1 and O2 , mutually nonlocal if the Euclidean correlator O1 (r)O2 (0) is a multi-valued function of r. In particular,
if under analytic continuation z ze2i , z z e2i (z = + ix and z = ix), the correlator
acquires only a phase, 21,2 , the two operators are called semi-local. The remarkable advantage
of this definition is that the index 1,2 can be evaluated at small distances, where all correlation
functions have a simple power law form.
From Eq. (18) it follows that an operator O with a finite vacuum expectation has annihilation
poles only if a,O is not an integer, that is when O is nonlocal. For instance, and in the Ising
model are nonlocal with respect to the fermions (they acquire a phase, , after circling around
the fermions in the x plane). However, if the operator can create an odd number of particles,
the annihilation poles may appear provided the S matrix is nontrivial. For instance, the vector
field n in the O(3) nonlinear sigma model belongs to the later category. Its form factors do have
annihilation poles, although its vacuum expectation value is zero [1].
We proceed with the calculation of the most singular parts of the form factors. We will do
it explicitly for operators with 0|O|0 = 0. The maximal singularities come from the matrix
elements which contain equal numbers of in and out particles. The contributions from all other
matrix elements contain additional powers of either exp[M] or exp[itM]. Since form factors
with different order of rapidities are related to one another (Refs. [13]), it is sufficient to know
the residues for one particular arrangement. For the form factors of the type in Eq. (14), we focus
on the residues at the poles at i = i . In this case one can use the crossing symmetry of the Smatrix and obtain from Eq. (17) the relation for the residues of (14). This allows the convenient
graphical representation of Fig. 2.
At this point, the computation divides itself into two cases. Each case marks out a separate
universality class. We deal with each in turn.
3.1. Ising-like universality class
In the first case, Ising-like behavior results. This occurs if either of following conditions hold
(1) The S matrix is diagonal.
(2) The S matrix is not diagonal, but the semi-locality index, a,O = 1/2, a = 1, . . . , , is independent of the particle index, a.
In these cases the residue is diagonal in the isotopic indices: ai = bi . For the diagonal S-matrix
this is obvious. For the case = 1/2 with a nondiagonal S-matrix, some additional reasoning is
required. Namely, one needs to apply Eq. (17) in successive steps. For the first step we obtain the
residue in the two-particle form factor. This yields Eq. (18) diagonal in the isotopic indices. Then
we use this result to calculate the coefficient at [(1 1 )(2 2 )]1 in the four-particle form
factor FO (1 i i0, 2 i i0, 2 , 1 )a2 a1 b2 b1 . As follows from Eq. (17), the S-matrices
enter into the expression for this coefficient in the vicinity of the double pole in the combination
a
,a
a
,a
a
,a
Sa 22,a 11 (21 )Sb 2,a

1 (i + 21 ) = Sa 22,a 11 (21 )Sab
2,,ab
1 (21 ) = a1 ,b1 a2 ,b2 .
2
(20)
320
Here we have used the crossing symmetry and unitarity of the S-matrices. As we see, this residue
is again diagonal in the isotopic indices and equal to (i)2 (1 e2i )2 = 4.
Thus for these two cases, it is clear that a multiple application of formula (17) yields the main
singularity for general n as
O
Fsing
(1 , . . . , n ; 1 , . . . , n )
= (i)n

1 e2ia,O
Pi
= (2i)n
n
n
(i Pi i0)1 0|O|0
i=1
n

(i Pi i0)1 0|O|0,
(21)
Pi i=1
where the sum includes all permutations of {1, . . . , n}. The result for the correlation function of
operators, O, with finite vacuum expectation values is then essentially the same as Eq. (6) but
with

exp[]
(22)
sin2 (j ) exp[j ] =
exp[j ],
j =1
j =1
where the sum runs over isotopic indices of the excitations.

3.2. Non-Ising like universality class
When both the S-matrix is nondiagonal and the semi-locality index, a,O , is not simply 1/2,
the behavior of the correlators is considerably more complex. Attendantly, the calculations are
more difficult due to the fact that a depends on the particle index. Nevertheless the maximal
residue can be calculated, although only at small rapidities (in the given context, that is all we
need). We take advantage of the fact that in integrable models with a non-diagonal S-matrix, the S
matrix has the universal limit S( 0) = P (the permutation operator). This limit corresponds
to a pure reflection. In general the crossover to pure reflectivity occurs at some model-dependent
scale 0 . It may happen that this scale is much smaller than one; then universal dynamics occurs
only for T M02 < M when rapidities | | 0 determine the expressions for the correlation
functions. At these temperatures, one can replace all S-matrices by permutation operators which
leads to simplifications in Eq. (17) (see Fig. 3):
As we see from Fig. 3, the isotopic indices of in and out particles must coincide. This allows
us to simplify the notational representation of the form factor:
Resi =i F irr (1 , . . . , n ; a1 , . . . , an |n , . . . , 1 ; an , . . . , a1 ) (i)n f (a1 , . . . , an )0|O|0.
Fig. 3. A graphical representation of Eq. (17) for S = P .
321
Then the equation depicted in Fig. 3 becomes

f (a1 , . . . , an ) = f (a2 , . . . , an ) eian f (a1 , . . . , an1 ).
(23)
This recurrence relation can be solved explicitly with the result

n

n

n1
ik
(nk)
exp i
j
f (a1 , . . . , an ) =
1 e (1)
,
k1
(24)
j =k+1
k=1
where a 2a,O .
In this case the correlator, O(x, t)O(0, 0), takes the form

O(x, t)O(0, 0)

2 d1
1
dn d1
dn
= O2
f (a1 , . . . , an )
n! a
2
2 2
2
n=0
i

n

cosh(j )
exp M

exp iMt
j =1
n

n

cosh(j ) cosh(j ) iMx
sinh(j ) sinh(j )
j =1
j =1
k
n
nk

1 k
1
1
(25)
C
,
n
n
2
2
( q i0) p=k+1 (p p + i0)2
k=0
q=1 q

where Cnk = nk . Again this expression arises either from using the form factors of O in conjunction with Eq. (12) or directly from the double summed Lehmann expansion (Eq. (1)).
Notice the particular arrangement of double poles in Eq. (25). As we have indicated the irreducible part of the form factor offers some flexibility in whether to deform the poles into the
upper or lower half plane (at the cost of changing the disconnected pieces of the form factors).
With this particular choice of poles, the integral can be straightforwardly evaluated (under the
same assumption as before that the most important contribution comes from small i and i ).
The result is

2

(S(x, t))n
O(x, t)O(0, 0) = O2
f (a1 , . . . , an ) ,
2n
2 n!
a
n=0
(26)
where

M 2
M M
d e 2 |x t |.
S(x, t) = e
(27)
This particular arrangement of poles can be justified by examining the structure of the disconnected pieces.1 Within the disconnected pieces lurk potential infinities (as regulated by the i0
terms). However with our choice of poles, the infinities ultimately cancel one another. Thus this
1 This is relatively easy to show for the case of the Ising model. Here we can directly rederive the result of Section 3
starting directing from Eq. (1) using infinite volume form factors. This arrangement of poles coincides with the form
factors yielding a finite expression. With correlators of the sine-Gordon model, it can be demonstrated that at least the
leading infinities contained in the disconnected pieces of the form factors vanish. To show sub-leading pieces vanish is
322
arrangement of poles is consistent with the requirement the form factors in their entirety (connected and disconnected pieces) yield a finite expression.
We now turn to the computation of the residues, f (a1 , . . . , an ), of the poles. To simplify the
discussion, we will consider only the case where the excitation spectrum of the theory constitutes
a doublet: a = = 1. This situation includes many important models, in particular the sineGordon model. In that case the semi-locality index of operator O with a soliton/antisoliton equals
. We have, for instance,
2

f (+) = 2 1 ei ,
f (++) = 1 ei ,

2
f (+) = 2 1 ei ,
f () = 1 ei .
In this case Eq. (24) reduces to

f (1 , . . . , n )2
i =1

j
1
n1
k1
= 2n+2 sin2
+ 2 sin2
Cn1
Cn1 (cos )kj 1 .
C2n2
2
2
(28)
k>j
To evaluate the sum, k < j , we rewrite it as follows:

2
k1 j 1
Cn1
Cn1 (cos )kj 1
k>j
n
1
1 k1 j 1
n1
=
Cn1 Cn1 (cos )|kj |

C
cos() 2n2 cos
k,j
C n1 .
cos() 2n2
(29)
For cos() > 0, we use the identities

|kj |
cos()
1
=

d
ei(kj ) ,
+ 2
1
sin2 ()
=
,
4( + m)2 + 2 8 sin2 () cos() + sin4 (/2)
m=
(30)
where = ln cos , to rewrite as

22n4
=
cos()

d sin2n2 ()
0
sin2 ()
sin2 () cos() + sin4 (/2)
(31)
n1
, in a similar integral form
If we finally use the identity to write the combinatorial factor, C2n2
n1
C2n2
22n2
=

2n2
d cos(/2)
,
(32)
more difficult as the full structure of the (highly non-trivial) sine-Gordon form factors are needed rather than merely the
behavior of the form factor in the immediate vicinity of its annihilation poles. This work will be reported shortly by the
authors together with F.H.L. Essler.
323
we can, using Eq. (26), represent the correlator as

1
O(x, t)O(0, 0) =
2
/2

d exp 2S(x, t) sin2 ()
0
sin2 ()
sin2 () cos() + sin4 (/2)
(33)
It turns out that this relation holds even if cos() < 0. To establish this we must use instead the
identity
1
sin2 ()
=
,
4( + m)2 + 2 8 cos2 () cos() + sin4 (/2)
m=
(34)
where now = ln | cos()|. We point out that as , the expression for the asymptotics of
the correlator (Eq. (33)) reproduces the result for the Ising universality class.
For x = 0 and large |t|, this correlation function behaves as

O(0, t)O(0, 0) 1/t 1/2 ,
(35)
i.e., it demonstrates a comparatively slow, nonexponential, decay. This behavior is similar to the
cases of a diagonal S-matrix or a nondiagonal S-matrix with trivial nonlocality phase. In the latter
two cases, we find exponential decay at large t .
These results are also qualitatively similar to the recent semi-classical analysis in Refs. [6,21].
In Ref. [6] the sine-Gordon model is explicitly studied while in Ref. [21] the three state Potts
model outside of the scaling regime is examined. Here it appears the S-matrix is nondiagonal.2
The two semi-classical analyses coincide (after a certain identification of parameters).
4. Low temperature charge dynamics in the sine-Gordon model
To provide an example of a physical problem which can be solved by the method developed
in this paper, we consider the problem of charge counting statistics in the sine-Gordon model.
The Lagrangian corresponding to this theory is
1
( )2 m2 cos( ).
16
The cosine term is relevant at < 1. The conserved soliton charge is defined as

dx x .
Q=
2
L=
(36)
(37)
Therefore Q(x, t) = ( /2)[(x, t) (x, 0)] is the charge that has passed through a point x
during time t. One can calculate the probability distribution function of Q(x, t) using the results
for the correlation function of operators exp[i]. In this case = (/ ).
The sine-Gordon model is a good example to study as it encompasses both universality classes
as is varied. At the so-called reflectionless points n2 = 1/(n + 1), n > 1, the sine-Gordon
model has a diagonal S-matrix. At these points the model belongs to the same universality class
as the Ising model. Away from these values of , the solitonsoliton scattering is not diagonal
and the model falls into the second universality class (although for operators ei /2 where the
2 In the scaling regime it is firmly believed that the S-matrix for the 3-state Potts model in the paramagnetic regime is
diagonal. See for example [22].
324
semi-locality index is 1/2 we revert to the Ising universality class). In this case the physics is
governed by the second universality class provided we are at a sufficiently low temperature. The
rapidity scale governing this crossover is

2
1
0 cosh
+ 2sin
cos
,
2

where = 2 /(1 2 ). If is close to one of the reflectionless points, i.e., 2 = 1/(n + 1) + ,
the crossover rapidity simplifies to
0
n + 1
.
n
2
.
The corresponding temperature at which the crossover occurs is T M (n+1)
2n2
We now consider the counting statistics present in the sine-Gordon model when it falls into
each of the two universality classes.
4.1. Counting statistics at the reflectionless points
At 2 = 1/(n + 1) we are in the Ising-like universality class. The correlation function then
takes the form, as follows from Eq. (22),

i(x,t) i(0,0)
(p)
dp (p)
2
2
e
(38)
= C ( , ) exp 2 sin (/ )
e
e
x t p ,
where (p) is the soliton energy and the factor C( , ) is equal to the vacuum expectation value
of exp[i]. This formula remains valid even for 2 = (n+1)1 provided = /2. The quantity
C( , ) was calculated in Ref. [20]:

C( , ) =
2
aM (1/(2 2 2 )) ((2 3 2 )/(2 2 2 )) 2
4

dy
sinh2 (2 y)
22 e2y .
exp
y 2 sinh( 2 y) sinh y cosh[(1 2 )y]
(39)
Here we have assumed that the correlator, ei(x,t) ei(0,0) , satisfies the short distance normalization,

i(x,t) i(0,0)
4a
|x y|4
where a 1 is the effective cutoff in the theory. M m1/(1 ) is the soliton mass. In the field
2
theory limit, Ma 1, and so the dependence in Eq. (39) is dominated by the term (Ma)2 .
Thus the magnitude of the propagator has a strong maximum at small .
We now examine the generating functional exp[iQ(x, t)]. This is given by
iQ(x,t) i w (x,t) i w (x,0)
e
= e 2
e 2

dp (p)
w 2
= C ,
(40)
exp 2 sin2
tp (p) .
e
2
2
325
In the limit of large time the above integral finds its predominant contribution at small . We
then can write

iQ(x,t)
2
= (aM)( ) exp 2 |t|/0 ,
e
(41)
2
( )
2
for some constant and where 1/0 =
where we have written C( , w
2 ) (aM)
M/T
(T /)e
.
The Fourier transform of Eq. (41) then yields a Gaussian distribution function for the charge,

2 /4 ,
P (Q) (Q Q) = d eiQ eiQ(x,t) 1/2 exp Q
(42)
where = (t/0 ) +
ln[1/(aM)].
4.2. Charge counting statistics at 2 = 1/(n + 1)

Here we consider the sine-Gordon model when it falls into the second universality class. As
we have already mentioned, at 2 = 1/(n + 1) there is a crossover temperature T below which
one can replace the S-matrix of slow solitons by the permutation operator. Then we may use
Eq. (33), finding that

iQ(x,t)

2
w 1
.
= C ,
d exp 42 t/0 2
2 2
+ 4 /16
2
(43)
The Fourier transform of this expression, after some manipulation, can be written as
= t
P (Q)
1/2
dy
Q
4(A+|y|)
y 2 /t
A + |y|
(44)
where t = t/0 and A = 2 ln(a 1 /(M)). At small times, the integral sees its main contribution
at small y, thus allowing us to write
2
2 t
3/2
Q
Q
P (Q)t1 exp
(45)
1
.
4A
32A3
2A
with small time dependent corIn this case the distribution is that of a static Gaussian in Q
rections. At large times we can evaluate this integral in the saddle point approximation which
yields

4/3

Q
1 1/6
3
exp 5/3 1/4
.
P Q(t) t1
(46)
2
2
t
tQ
The reader may notice that quantum fluctuations play a significant role in modifying the Gaussian
distribution of the counting statistics seen in the simpler Ising-like reflectionless case.
5. Conclusions
In this paper we have found universal expressions for the asymptotics of correlation functions
of order parameter operators. These operators have T = 0 nonzero vacuum expectation values
326
thus breaking some discrete symmetry present in the theory. We restricted our consideration to
theories with well defined asymptotic particle states (this excludes such models as the super
sine-Gordon or the O(2N + 1) GrossNeveu model). Among such models, we have found two
characteristic types of order parameter dynamics. One type appears in the models with either
diagonal S-matrices (these include, among other models, the quantum Ising model and the sineGordon model at the reflectionless points) or for operators of nondiagonal theories with trivial
semi-locality factors. These dynamics essentially coincide with the semi-classical dynamics described in [5]. Another type of dynamics is described by Eq. (33) and appears in the models
where the zero momentum S-matrix corresponds to pure reflection and the order parameter has
a nontrivial semi-locality factor.
Acknowledgements
This research was supported by DARPA under the QuIST Program, by the Institute for
Strongly Correlated and Complex Systems at BNL (B.L.A.), and by the DOE under contract
number DE-AC02-98 CH 10886 (R.M.K., A.M.T.). All the authors are grateful to the International Centre for Theoretical Physics in Trieste, Italy, for its hospitality during which this project
was completed. A.M.T. is grateful to Princeton University for hospitality and to S.A. Lukyanov
for attracting his attention to Ref. [10]. We are also grateful to H. Babujian, M.J. Bhaseen,
K. Damle, F.H.L. Essler, V.A. Fateev, V.E. Korepin, S. Sachdev, F.A. Smirnov, G. Zarand, and
A. Chitov for fruitful discussions and interest in the work.
References
[1] F.A. Smirnov, Form Factors in Completely Integrable Models of Quantum Field Theory, Advanced Series in Mathematical Physics, vol. 14, World Scientific, Singapore, 1992.
[2] H. Babujian, A. Fring, M. Karowski, A. Zapletal, Nucl. Phys. B 538 (1999) 535;
H. Babujian, M. Karowski, Nucl. Phys. B 620 (2002) 407.
[3] F.H.L. Essler, R.M. Konik, in: M. Shifman, A. Vainshtein, J. Wheather (Eds.), From Fields to Strings: Circumnavigating Theoretical Physics, World Scientific, Singapore, 2005, cond-mat/0412421.
[4] B. Berg, M. Karowski, P. Weisz, Phys. Rev. D 19 (1979) 2477.
[5] S. Sachdev, A.P. Young, Phys. Rev. Lett. 78 (1997) 2220;
S. Sachdev, Quantum Phase Transitions, Cambridge Univ. Press, Cambridge, 1999.
[6] K. Damle, S. Sachdev, cond-mat/0507380.
[7] A.M. Tsvelik, Quantum Field Theory in Condensed Matter Physics, Cambridge Univ. Press, Cambridge, 2003.
[8] P. Fonseca, A. Zamolodchikov, J. Stat. Phys. 110 (2003) 527.
[9] A. Leclair, F. Lesage, S. Sachdev, H. Saleur, Nucl. Phys. B 482 (1996) 579.
[10] A.I. Bugrij, in: S. Pakuliak, G. von Gehlen (Eds.), Integrable Structures of Exactly Solvable Two-Dimensional
Models of Quantum Field Theory, in: NATO Series, Kluwer Academic, Dordrecht, 2006, hep-th/0107117;
A.I. Bugrij, O. Lisovyy, JETP 94 (2002) 1140.
[11] B. Doyon, Finite temperature form factors in the free Majorana theory, hep-th/0506105.
[12] A.R. Its, A.G. Izergin, V.E. Korepin, N.A. Slavnov, Phys. Rev. Lett. 70 (1993) 1704.
[13] V. Korepin, N. Slavnov, Phys. Lett. A 236 (1997) 201.
[14] V.E. Korepin, N.M. Bogoliubov, A.G. Izergin, Quantum Inverse Scattering Method and Correlation Functions,
Cambridge Univ. Press, Cambridge, 1993.
[15] N. Kitanine, J.-M. Maillet, V. Terras, Nucl. Phys. B 554 (1999) 647;
N. Kitanine, J.-M. Maillet, V. Terras, Nucl. Phys. B 567 (2000) 554.
[16] J.-S. Caux, J.-M. Maillet, Phys. Rev. Lett. 95 (2005) 077201.
[17] J. Balog, Nucl. Phys. B 419 (1994) 480.
[18] A. LeClair, G. Mussardo, Nucl. Phys. B 552 (1999) 624.
[19] R. Konik, Phys. Rev. B 68 (2003) 104435.
[20] S. Lukyanov, Mod. Phys. Lett. A 12 (1997) 2543.
327
[21] A. Rapp, G. Zarand, Dynamical correlations and quantum phase transitions in the quantum Potts model, condmat/0507390.
[22] R. Koberle, J.A. Swieca, A.M. Tsvelick, Nucl. Phys. B 305 (1988) 675;
R. Koberle, J.A. Swieca, A.M. Tsvelick, Phys. Lett. B 86 (1979) 209;
A.B. Zamolodchikov, Int. J. Mod. Phys. A 3 (1988) 743;
A.B. Zamolodchikov, Nucl. Phys. B 342 (1990) 695.
Twisted boundary states and representation

of generalized fusion algebra
Hiroshi Ishikawa a, , Taro Tani b,1
a Department of Physics, Tohoku University, Sendai 980-8578, Japan
b Interdisciplinary Center of Theoretical Studies, Institute of Theoretical Physics, Chinese Academy of Sciences,
Beijing 100080, China

Received 15 November 2005; accepted 23 January 2006
Abstract
The mutual consistency of boundary conditions twisted by an automorphism group G of the chiral algebra is studied for general modular invariants of rational conformal field theories. We show that a consistent
set of twisted boundary states associated with any modular invariant realizes a non-negative integer matrix
representation (NIM-rep) of the generalized fusion algebra, an extension of the fusion algebra by representations of the twisted chiral algebra associated with the automorphism group G. We check this result for
several concrete cases. In particular, we find that two NIM-reps of the fusion algebra for su(3)k (k = 3, 5)
are organized into a NIM-rep of the generalized fusion algebra for the charge-conjugation automorphism
of su(3)k . We point out that the generalized fusion algebra is non-commutative if G is non-Abelian and
provide some examples for G
= S3 . Finally, we give an argument that the graph fusion algebra associated
with simple current extensions coincides with the generalized fusion algebra for the extended chiral algebra,
and thereby explain that the graph fusion algebra contains the fusion algebra of the extended theory as a
subalgebra.
E-mail addresses: ishikawa@tuhep.phys.tohoku.ac.jp (H. Ishikawa), tani@itp.ac.cn, tani@kurume-nct.ac.jp

(T. Tani).
1 Address after September 2005: Kurume National College of Technology, 1-1-1 Komorino, Kurume 830-8555, Japan.
doi:10.1016/j.nuclphysb.2006.01.031
H. Ishikawa, T. Tani / Nuclear Physics B 739 [FS] (2006) 328388
329
1. Introduction
The classification of conformally invariant boundary states in two-dimensional conformal
field theories (CFTs) is an interesting subject of research, both from the point of view of purely
field theoretical problems and applications to condensed matter physics or string theory. In string
theory, conformally invariant boundary states correspond to D-branes, which are considered to
be submanifolds embedded in the target space of strings. The classification of boundary states is
therefore the classification of D-branes, which might be helpful in understanding the nature of
stringy geometry.
This classification problem of boundary states is hard to treat in full generality, and one would
need some restriction on the problem to make it tractable. A natural choice for such a restriction
is to consider only rational CFTs (RCFTs)2 and boundary states that preserve the full chiral
algebra of RCFTs. Actually, in a RCFT, every consistent set of boundary states should satisfy
a set of simple algebraic equations, the so-called Cardy equation [1], if the full chiral algebra
is preserved in the open-string channel. Finding a set of Cardy states, i.e., the states satisfying
the Cardy equation, is equivalent to finding a non-negative integer matrix representation (NIMrep) of the fusion algebra [2]3 assuming that the set of boundary states is complete in a sense
of [3]. The classification of boundary states for this case is therefore related to the classification
of NIM-reps of the fusion algebra [7].
Clearly, the boundary states preserving the full chiral algebra constitute only a subset of
conformally invariant boundary states, since the chiral algebra in general contains the Virasoro
algebra as a proper subalgebra. In order to obtain, and classify, more states other than those preserving the full chiral algebra, we need some way to reduce the symmetry preserved by boundary
states in a controlled manner. One way to accomplish this is to modify (or twist) the boundary
condition of the chiral currents by an automorphism group G of the chiral algebra. The corresponding twisted boundary states preserve the Virasoro algebra if G keeps it fixed. The states
preserving the full chiral algebra (the untwisted states) correspond to the identity element of G
and are considered to be a particular case of the twisted states. In this sense, the twisted states
associated with the automorphism group G realize a generalization of untwisted states.
The classification problem of twisted boundary states is systematically studied by several
groups [811].4 In particular, it is pointed out in [10] (see also [9,15]) that the mutual consistency of twisted boundary states in the charge-conjugation modular invariant follows from the
integrality of the structure constants of a certain algebra called the generalized fusion algebra.
This algebra is defined by the fusion of representations of the twisted chiral algebra and hence
contains, as a subalgebra, the ordinary fusion algebra consisting of only representations of the
(untwisted) chiral algebra. This relation of twisted boundary states with the generalized fusion
algebra is quite analogous to the relation of untwisted boundary states with the ordinary fusion
algebra, and suggests that the generalized fusion algebra also plays some role in the classification
problem of twisted boundary states.
In this paper, we give an answer to the above question concerning the role of the generalized
fusion algebra in the classification problem of twisted boundary states. Namely, we show that
2 For the current status of the research about non-rational cases, see, e.g., [21].
3 It should be noted that the Cardy condition is a necessary condition on boundary states [46]. Actually, there is a
NIM-rep which does not give rise to any consistent boundary states, such as the tadpole NIM-rep of su(2)2n1 (see, e.g.,
[2,7]).
4 For the other approaches to boundary states with less symmetry, see, e.g., [1214].
330
a set of mutually consistent twisted boundary states realizes a NIM-rep of the generalized fusion algebra. Our result is valid for any finite automorphism group including non-Abelian ones.
We point out that the generalized fusion algebra for a non-Abelian automorphism group is noncommutative and provide some examples for the case of the symmetric group S3 . We develop
the representation theory of the generalized fusion algebras and find their irreducible representations, which generalize quantum dimensions for the case of the ordinary fusion algebras. Unlike
quantum dimensions, irreducible representations of the generalized fusion algebras are possible
to have dimension greater than one, reflecting the non-commutativity of the generalized fusion
algebras. We also point out that a NIM-rep of the generalized fusion algebra is decomposed into
NIM-reps of the ordinary fusion algebra, since the former contains the latter as a subalgebra. As
a check of our argument, we explicitly construct twisted boundary states in several concrete cases
and show that their overlap matrices indeed form a NIM-rep of the corresponding generalized
fusion algebra. In particular, we find that two NIM-reps of su(3)k (k = 3, 5) are combined into a
NIM-rep of the generalized fusion algebra of su(3)k for the automorphism group {1, c }, where
c is the charge-conjugation automorphism of su(3)k . We also give an argument that some graph
fusion algebra [2,19,20] can be regarded as the generalized fusion algebra. More precisely, we
show that the graph fusion algebra associated with boundary states in a simple current extension
[16] coincides with the generalized fusion algebra of the extended chiral algebra for an appropriate automorphism group. This result naturally explains the observation [19] that the graph fusion
algebra contains the fusion algebra of the extended theory as a subalgebra.
An algebraic structure of twisted boundary states is also studied in [8,9] and we should clarify
the difference of our analysis from that of [8,9]. In [8], the case that a chiral algebra A is obtained
by a simple current extension is considered. In this setting, the unextended chiral algebra A0 is
characterized as the fixed point algebra of some finite Abelian automorphism group G of A,
which is the dual of the simple current group used in the extension. The authors of [8] classify
boundary states that preserve A0 in the charge-conjugation invariant of A using simple current
techniques, and show that the states preserving A0 can be regarded as the G-twisted states. They
also show that the G-twisted boundary states correspond to one-dimensional representations of
some commutative algebra called the classifying algebra [17]. In [9], the structure of twisted
boundary states in the charge-conjugation invariant is investigated for an arbitrary chiral algebra,
in particular untwisted affine Lie algebras, with the automorphism group G
= Z2 , and the corresponding generalized fusion algebra together with the classifying algebra are studied. Our stance
on the problem is slightly different from these studies. We do not intend to construct twisted
boundary states explicitly; rather we clarify a consistency condition of twisted states assuming
their existence. We impose no restrictions on the model we consider. The charge-conjugation
modular invariant, as well as an Abelian automorphism group, is a particular case of our analysis. Finally, the expression [9] for the case of the Z2 automorphism suggests that the classifying
algebra in [8] is the dual of the generalized fusion algebra for Abelian automorphism groups in
the sense of C-algebras [1820], although we have no proof for this statement. Since this duality exchanges the irreducible representations with the elements of the algebra, the result of [8] is
consistent with the observation [10,15] that the twisted boundary states in the charge-conjugation
invariant realize the generalized fusion algebra, which is the starting point of our analysis.
This paper is organized as follows. In Section 2, we give the definition of the generalized
fusion algebras using the twisted boundary states in the charge-conjugation modular invariant
along the lines of [10]. The case of the affine Lie algebra so(8)1 and its triality automorphism
group is treated in detail. In Section 3, we show that the fusion coefficients of the generalized
fusion algebras can be expressed in a form similar to the Verlinde formula even in the case of
331
non-Abelian automorphisms, and obtain the irreducible representations of the generalized fusion
algebras. Based on this result, we show in Section 4 that the overlap matrices of mutually consistent twisted boundary states in any modular invariant form a NIM-rep of the generalized fusion
algebra. Some examples including the case of non-Abelian automorphisms are presented in Section 5. In Section 6, we argue the relation of graph fusion algebras with the generalized fusion
algebras. The final section is devoted to discussions. In the appendices, we present the details of
the calculation in our examples. In Appendix A, we give an explicit realization of the algebra
embedding su(3)3 so(8)1 and show that the charge-conjugation automorphism group {1, c }
of su(3) has a lift to so(8), which is isomorphic to the symmetric group S3 . In Appendix B, the
generalized fusion algebra of su(3)k (k = 3, 5) for the charge-conjugation automorphism group
is determined and it is shown that the twisted boundary states associated with non-trivial modular invariants yield a NIM-rep of the generalized fusion algebra. The same calculation is done in
Appendix C for su(3)3
1 and its permutation automorphism of three factors.
2. Generalized fusion algebra
In this section, we give a definition of the generalized fusion algebra for the chiral algebra
A with the automorphism group G by using the cylinder amplitude for the (twisted) boundary
states in the charge-conjugation modular invariant of A.
2.1. Untwisted boundary states and ordinary fusion algebra
We start from the construction of the untwisted boundary states and their relation with the
ordinary fusion algebra [1].
Let I be the set of all the irreducible representations of the chiral algebra A. Since we consider
only rational cases, I is a finite set. The vacuum representation of A is denoted by 0 I. For
each I, the character (q) of the representation is defined as
c
(q) = TrH q L0 24 ,
(2.1)
where q = e2i , c is the central charge of the theory and the trace is taken in the irreducible representation space H with the highest-weight . Under the modular transformation 1/ ,
the characters (q) transform as follows

S (q)
q = e2i/ .
(q) =
(2.2)
I
The modular transformation matrix S is unitary and symmetric. The (ordinary) fusion algebra is
an associative commutative algebra defined by the fusion of two representations in I,

() () =
(2.3)
N () (, I),
I
where the fusion coefficients N take values in non-negative integers and are related with the
modular transformation matrix via the Verlinde formula [22],
N =
S S S
.
S0
(2.4)
A boundary state | is a coherent state in the closed-string channel that preserves a half of
the (super) conformal symmetry on the worldsheet. We also require that | keep the full chiral
332
algebra A. In terms of modes, the boundary condition can be written as

Wn (1)h W n | = 0,
(2.5)
where Wn and W n are respectively the modes of the current W (z) of the holomorphic chiral
algebra A with conformal dimension h and its counterpart W (z) of the anti-holomorphic chi For each I, we can construct a building block |(, ) of boundary states
ral algebra A.
satisfying (2.5),

, = 1
|; N |; N H H ,
(2.6)
S0 N
which is called the Ishibashi state [23]. Here |; N is an orthonormal basis of the representation
space H and we denote by I the conjugate representation of I. This notation for
Ishibashi states is slightly different from that used in other references, such as | . We adopt
|(, ) , instead of | , to indicate explicitly that this state is composed of the states in H H
Since the boundary condition (2.5) is linear,
with the highest-weight state | | of A A.
multiplying |(, ) by some constant still gives a solution to (2.5). The above normalization
(2.6) is chosen so that the following equation is satisfied,
1

c
1
1

; N |q L0 24 |; N =
(q).
, q 2 Hc , =
(2.7)
S0
S0
N
Here Hc is the closed string Hamiltonian

c
Hc = L0 + L 0
(2.8)
12
and we used (L0 L 0 )|(, ) = 0 to eliminate L 0 .
In order to find solutions to (2.5), we have to specify the spectrum of bulk fields, or equivalently the modular invariant

M ,
Z=
(2.9)
,I
where corresponds to a bulk field (, ) which carries a representation (, ) of A A.
In (2.9), there are M fields with representation (, ) and we should distinguish them by
putting an index like (, )i if M > 1. In order to keep the notation simple, however, we
omit the extra index i and simply use (, ) understanding that they appear in (2.9) with the
multiplicity M . We denote by Spec(Z) the set of labels for bulk fields available in (2.9),

Spec(Z) = , (, ) exists in (2.9) .
(2.10)
Note that a symbol (, ) appears M times in Spec(Z) corresponding to M independent
bulk fields with the label (, ). In this section, we restrict ourselves to the case of the chargeconjugation modular invariant

| |2 ,
Zc =
(2.11)
I
for which the spectrum of bulk fields reads

Spec(Zc ) = , I .
(2.12)
333
We have one Ishibashi state |(, ) for each bulk field (, ) . A general boundary state |
satisfying (2.5) are therefore a linear combination of |(, ) ( I),

(, ) , .
| =
(2.13)
I
A consistent set of boundary states is obtained by considering the cylinder amplitude Z

with boundary conditions and . This amplitude can be calculated in two ways: a closedstring propagation between two boundary states, and an open-string one-loop amplitude. In the
closed-string channel, the cylinder amplitude can be expressed as
1
Z = |q 2 Hc | ,
(2.14)
= e2/t
and t is the circumference of the cylinder. On the other hand, the calculation
where q
in the open-string channel yields

Z =
(2.15)
(n ) (q),
I
= e2t
and (n ) is the multiplicity of representation in the spectrum

where q
of open string
with boundary condition [, ]. Note that this form of open-string spectrum, I (n ) H ,
follows from the requirement that the boundary conditions and of open string preserve the
chiral algebra A. Comparing two expressions for Z , one obtains

1
(n ) (q).
|q 2 Hc | =
(2.16)
I
The multiplicity (n ) takes values in non-negative integers. In particular, (n0 ) = 1, since

an open string with the same boundary condition at both ends should have the unique vacuum.
We therefore obtain
1
|q 2 Hc | = 0 (q) + ,
(2.17)
for any boundary state | . Clearly, the simplest situation is that the right-hand side contains only
the vacuum character 0 . We denote such a state by |0 ,
1
0|q 2 Hc |0 = 0 (q).
(2.18)
This state can be realized in the charge-conjugation modular invariant using the Ishibashi states,

S0 , .
|0 =
(2.19)
I
One can check that this state has the desired overlap with itself,
1
0|q 2 Hc |0 =
S0 S0

1
(q)
=
S0 (q)
= 0 (q).
S0
(2.20)
The remaining states satisfying the boundary condition (2.5) are determined by requiring that
their overlap with |0 contain a single character ,
1
0|q 2 Hc | = (q) ( I).
(2.21)
334
Fig. 1. Joining two open strings with boundary condition [ , 0] and [0, ] yields a string with [ , ].
One can solve this equation to obtain | in the form

S , .
| =
(2.22)
Actually, the overlap of this state with |0 yields ,

1
0|q 2 Hc | =
S0 S

1
(q)
=
S (q)
= (q).
S0
(2.23)
Taking the complex conjugation of this equation, we obtain the conjugate representation of
I,
1
|q 2 Hc |0 = (q).
(2.24)
One can relate the overlap of two boundary states with the fusion rule coefficients. Suppose
that we have two open strings, one of which has the boundary condition [ , 0] and the other has
[0, ] (see Fig. 1). The spectrum of these two can be calculated as
1 H
1
0|q 2 Hc | = (q).
q 2 c |0 = (q),
(2.25)
Since the boundary condition 0 is common to both strings, one can join two strings at the
boundary 0 to yield a string with the boundary condition [ , ]. The spectrum of this string
can be determined by the fusion of the spectra of the initial strings. In the present case, this is
nothing but the fusion of and . Hence we obtain

1 H
q 2 c | =
(2.26)
N (q).
I
On the other hand, we can explicitly calculate the left-hand side using the Ishibashi states,

1 H
1
q 2 c | =
S S
(q)
S0
I
S S
1 1
S (q)
S0
I
S S S
=
(q),
S0
, I
(2.27)
335
where we used the unitarity of S and the property

S = S .
(2.28)
1
Comparison of these two expressions for |q 2 Hc | yields [1] the Verlinde formula (2.4).
2.2. Twisted boundary states and generalized fusion algebra
We next turn to the case of twisted boundary states and define the generalized fusion algebra
[9,10,15].
Let G be a group of automorphisms of the chiral algebra A, which is in general a subgroup
of the full automorphism group Aut(A). We restrict ourselves to the case that G is finite (but
not necessarily Abelian). We also require that G fixes the Virasoro algebra in A. Then we can
modify the boundary condition of the chiral algebra using G as follows,

(Wn ) (1)h W n |
(2.29)
= 0,
which we call twisted boundary conditions while the condition (2.5) the untwisted one. This
boundary condition preserves the conformal symmetry since G fixes the Virasoro algebra by
assumption.
An automorphism G determines a unitary operator R() on the representation space

I H of A through the relation
(Wn ) = R()Wn R()1 .
(2.30)
The operator R() in general maps the representation space H to the different one, which we
denote by H() ,
R() : H H() .
(2.31)
Using this R(), the basis of solutions to (2.29) can be constructed from the Ishibashi states (2.6)
for the untwisted condition (2.5) in the following manner,

, ; = R() , = 1
R()|; N |; N H H
S0 N

= () ,
(2.32)
where R() acts only on the holomorphic sector and = (). One can verify that this state
satisfies the twisted boundary condition (2.29) as follows

(1)h W n , ; = R()(1)h W n ,

= R()Wn ,

= R()Wn R()1 R() ,

= (Wn ) , ; .
(2.33)
We call these states |(, ); twisted Ishibashi states. The untwisted one (2.6) can be considered as the particular case = 1 of twisted states. Note that the -twisted Ishibashi state
|(, ); exists if and only if = ().
336
The overlap of twisted Ishibashi states can be calculated in the same way as the untwisted
case,

1

1

, ; q 2 Hc , ; = , R() q 2 Hc R( ) ,

c
1
=
(2.34)
; N |R 1 q L0 24 |; N .
S0
N
Clearly, this expression vanishes unless

S() G is the stabilizer of I,
1 ()
= , or equivalently 1 S(), where
S() = { G | () = }.
(2.35)
Using the relations = () and = ( ) = (), we can rewrite the condition 1

S() into that for and ,
= () = 1 () = () = .
Conversely,
implies
(2.36)
S(),
1 () = 1 ( ) = 1 () = .
(2.37)
Putting these things together, we can express the overlap of twisted Ishibashi states in the following form

where
1

1 1

, ; q 2 Hc , ; = (, )( , )
(q)
S0

= (), = ( ) ,
(2.38)
S()) is the twining character [24] of A,

c
(q) = TrH R()q L0 24 .
(2.39)
for a given I vanishes unless S(). Equivalently, for a

As we have noted above,
given G, exists if and only if () = . To express the condition for non-vanishing
for a given G, we introduce a set I() which consists of representations of A fixed by

I() = I | () = .
(2.40)
It is useful to note that the following two conditions are equivalent,

I() S().
(2.41)
|(, ); composed from the representation (, )
Since the twisted Ishibashi state

exists
only when = (), the set of -twisted Ishibashi states available in the charge-conjugation
invariant (2.11) have the form

, ; I() .
(2.42)
A general boundary state |
satisfying the boundary condition (2.29) in the charge-conjugation
modular invariant is therefore expressed as follows,

(,
|
=
(2.43)
) , ; .
I ()
337
Fig. 2. Joining two open strings with boundary condition [ , 0] and [0, ]
yields a string with [ , ].
The thick line
stands for a twist by the automorphism G.
1
In order to determine the coefficients (,

2 Hc |
) , we consider the cylinder amplitude 0|q
with boundary conditions 0 I and .

Since the boundary condition of the corresponding open
string is twisted by at only one end (see Fig. 2), the chiral algebra in the open-string channel
is the twisted chiral algebra A of A associated with G, which is generated by the currents
of A with the twisted boundary condition

W e2i z = W (z) .
(2.44)
1
is therefore expanded into the characters of A .

The cylinder amplitude 0|q 2 Hc |
We denote by I the set of all the irreducible representations of the twisted chiral algebra
A and by the character of I . The modular transformation of can be expanded into
1
the twining characters (2.39) of A. Actually, evaluating the cylinder amplitude 0|q 2 Hc |
in
1
the closed-string channel yields a linear combination of the overlap (, )|q 2 Hc |(, ); ,
as we have shown in (2.38).5 Accordingly,
which in turn provides the twining character (q)
we can express the modular transformation of the character in the following form,
(q) =

I ()
S (q)

I .
(2.45)
The matrix S relates the representations of A with those of A fixed by . We assume that S
is unitary. In particular, two sets, I and I(), have the same order, |I | = |I()|. This holds
for many examples including the case of an affine Lie algebra of untwisted type with the diagram
automorphism of its horizontal Lie algebra as an automorphism .6
We can determine the coefficients of twisted Ishibashi states in a way parallel to the untwisted
one, namely we require the condition
= (q)
0|q 2 Hc |

I .
(2.46)
In the same way as the untwisted case, twisted boundary states satisfying the above condition can
be constructed using the modular transformation matrix of the chiral algebra in the open-string
5 This argument gives an explanation for the fact [24] that the twining character for an affine Lie algebra g(1) with the
diagram automorphism of its horizontal Lie algebra as an automorphism of order r can be expressed by characters of
a twisted affine Lie algebra g (r) , since the modular transformation of characters of a twisted affine Lie algebra g(r) is
expanded into characters of a (generally different) twisted affine Lie algebra g (r) [32].
6 A detailed study of the matrix S for the case of 2 = 1 is given in [9].
338
channel, this time A ,

=
|
S , ; .
(2.47)
I ()
We can check this expression yields the desired overlap (2.46) with |0 ,

1
1
S0 S
(q)
=
S (q)
= (q).
0|q 2 Hc | =
S0
I ()
(2.48)
I ()
Having obtained the explicit form of the twisted states, one can calculate the overlap of the
twisted state | (
I ) with generic untwisted states,

1 H
1
q 2 c |
=
S S
(q)
S0
I ()
I ()
S S
1 1
S (q)
S0
S
S S
S0
I , I ()
(q).
(2.49)
In the same way as the case of untwisted states, one can regard this as the fusion of two open
strings, [ , 0] and [0, ]
(see Fig. 2). The spectrum of these two strings contains only one
can be
representation, and ,
respectively. Therefore the spectrum of the joined string [ , ]
considered to express the fusion () ()
of two representations, I and I . Since the
string [ , ]
has a twist by at only one end ,
its spectrum consists of representations of the
twisted chiral algebra A . Hence one can conclude that the product () ()
is expanded into
representations of A ,

N ( ) I, I .
() ()
=
(2.50)
I
The coefficients N express the multiplicity of representation ( ) in the product () ()

and
take values in non-negative integers. In terms of these coefficients, the overlap of untwisted states
with twisted states can be written in the following form,

1 H
q 2 c |
(2.51)
=
N (q).
I
Comparing this with the closed-string channel calculation (2.49), one can obtain a formula for
N ,
N =
S
S S

I ()
S0

I; ,
I .
(2.52)
The above result can be readily generalized to the fusion of two generic representations. To
state the result in a concise form, we treat the untwisted states as the particular case = 1 of the
twisted states, and introduce a set I of all the representations of twisted chiral algebras A for
G,

I =
(2.53)
I,
G
339
which contains I = I =1 as a subset. We use a capital letter such as L for expressing an element
of I and denote the automorphism type of L I by putting a subscript like L G. In this
notation, the modular transformation, Eqs. (2.2) and (2.45), of characters of A can be written
as follows,

SLL L (q)
(L I).
L (q) =
(2.54)
I (L )
If L is a representation of A, L I, the modular transformation matrix S L is the ordinary Smatrix and the twining character L is nothing but the character of A. The boundary states
are labeled by the element of I and take the form given in Eqs. (2.22) and (2.47),

L
|L =
(2.55)
SL
, ; L (L I).
I (L )
The overlap of |L with the untwisted state |0 yields the character L in the open-string channel,
(L I),
0|q 2 Hc |L = L (q)
(2.56)
which means that the open string [0, L] has a spectrum consisting of only one representation
L I.
Taking the complex conjugation of this equation, one obtains the spectrum of the string [L, 0].
Since the string [L, 0] has an orientation opposite to that of [0, L], its chiral algebra has a
1
instead of L . The spectrum of the string [L, 0] is thereboundary condition twisted by L
1
fore expanded into characters of AL . Since the string [0, L] has only one representation, the
spectrum of the string [L, 0] also consists of only one representation. We call this the conjugate
1
representation of L and denote it by L I L I,

1
L|q 2 Hc |0 = L (q) (L I).
(2.57)
This generalizes the complex conjugation of ordinary representations (see (2.24)). Calculating
the left-hand side of Eq. (2.57) in terms of the Ishibashi states yields
1

L L
SL
(q).
L (q) =
(2.58)
I (L )
Comparing this with Eq. (2.54), we obtain the following formula

1
L
SL
= SLL
(L I),
(2.59)
which reduces to Eq. (2.28) for the case of L = 1.

Since the spectrum of two open strings, [L , 0] and [0, M], consists of only one representation, L and M, respectively, the spectrum of the joined string [L , M] is given by the fusion
of two representations L and M. From the expression given in Eq. (2.55), one can calculate the
overlap of |L and |M as follows,
1 H
L q 2 c |M =
1 1 H

1 M
q 2 c , ; M
, ; L
SLL SM
I (L )I (M )
I (L )I (M )
L M
SL
SM
1 L M
(q)
S0
340
L M
SL
SM
I (L )I (M )
1
S0
NI L M I (L )I (M )

NI L M
S L M
1
(q)
N N
L M L M
SL
SM SN
N (q),
S0
(2.60)
where we used the formula (2.38) and the unitarity of S L M . This equation shows that the
can be expanded into representations of the
fusion of two representations L and M (L, M I)
Therefore the set I is closed
automorphism type L M G, which are also the elements of I.
under the fusion of representations and forms an algebra

(L) (M) =
(2.61)
N LM N (N ),
NI
which generalizes the fusion algebra of ordinary untwisted representations. We call this the
generalized fusion algebra and denote it by F (A; G).7 The coefficients N LM N take values in
non-negative integers and we call them the generalized fusion coefficients. As is shown in the
overlap (2.60), the generalized fusion coefficients are given by the following formula
N LM N =

I (L )I (M )
L M N
SL
SM SN
L M ,N ,
S0
(2.62)
which generalizes the Verlinde formula (2.4) for the ordinary fusion coefficients.
As we have argued above, the fusion of a representation L with M has the automorphism
type L M and is given by a linear combination of the elements in I L M . An immediate consequence of this fact is that (L) (M) = (M) (L) in general, since two sets I L M and
I M L are distinct from each other if L M = M L . Therefore the generalized fusion algebra
F(A; G) is non-commutative if the automorphism group G is non-Abelian. In contrast to this,
there seems to be no special reason for a generalized fusion algebra to be non-commutative for an
Abelian G. Actually, it follows from the definition (2.62) of the generalized fusion coefficients
that (L) (M) = (M) (L) if L M = M L . In particular, F (A; G) is commutative if G
is Abelian. The ordinary fusion algebra F (A) corresponds to the trivial case F (A; {1}) and is
hence commutative.
2.3. An example: generalized fusion algebras for so(8)1
As an illustration of our argument, we give the explicit form of the generalized fusion algebra
for the affine Lie algebra so(8)1 and its triality automorphism group, which is isomorphic to the
symmetric group S3 . Since S3 is non-Abelian, the generalized fusion algebra F(so(8)1 ; S3 )8 is
non-commutative.
There are four irreducible representations for so(8)1 ,
I = {O = 0 , V = 1 , S = 3 , C = 4 },
(2.63)
where i is the fundamental weight of so(8)1 and each symbol stands for the vacuum, vector, spinor and conjugate spinor representation, respectively. The modular transformation matrix
7 F stands for fusion.
8 For brevity, we denote the automorphism group itself by S .
3
reads
341
1 1
1
1
1 1 1 1 1
S=
,
2 1 1 1 1
1 1 1 1
(2.64)
where the rows and the columns are ordered as (2.63).

The chiral algebra so(8)1 has an automorphism group consisting of all the permutations of
three legs of the Dynkin diagram for the horizontal subalgebra so(8) so(8)1 , which is isomorphic to the symmetric group S3 . The automorphism group S3 is generated by two elements
and , which act on the elements of I as
: V S C V , O O,
: O O, V V , S C S.
(2.65)
In terms of and , the elements of S3 can be expressed as follows,

S3 = , = 1, , 2 , , , 2 .
(2.66)
In order to obtain the generalized fusion algebra, we need to know twisted representations
and their modular transformation for each element of S3 . For and 2 , the fixed point set of the
automorphism consists of a single element

I() = I 2 = {O}.
(2.67)
Correspondingly, there is only one representation for each twisted chiral algebra, which is iso(3)
morphic to the twisted affine Lie algebra D4 at level 1. (The irreducible representations and
(r)
their modular transformation matrix for the affine Lie algebra D4 (r = 1, 2, 3) are presented,
2
e.g., in [25].) We denote the twisted representations for and as follows,

2
I = (0) ,
(2.68)
I = (0) 2 .
The modular transformation matrix in this case is simply a number
S = (1),
S = (1).
(2.69)
There are two representations fixed by S3 ,

I( ) = {O, V }.
(2.70)
(2)
The twisted chiral algebra for this case is isomorphic to D4 at level 1. There are two representations of the twisted chiral algebra for , which we denote as

I = (0) , (1) .
(2.71)
The modular transformation matrix reads

1 1 1
S =
.
2 1 1
For the other elements and 2 , one can proceed in the same way as to obtain

I( ) = {O, C},
I = (0) , (1) ,
(2.72)
(2.73)
342

I 2 = {O, S},

= (0) 2 , (1) 2 .
(2.74)
The modular transformation matrix is the same as that for ,

S = S
(2.75)
= S .
Consequently, the set I of all the twisted representations consists of 12 elements,

= |I| + I + I 2 + I + I + I 2 = 4 + 1 + 1 + 2 + 2 + 2 = 12. (2.76)
|I|
Since all the modular transformation matrices are real, the relation (2.59) implies that the repre2
sentations in I, I , I and I are self-conjugate. On the other hand, for the representations
2
in I and I , the conjugation changes the automorphism type, (0) = (0) 2 , since 1 = 2 .
Having obtained twisted representations and their modular transformation matrices, it is
straightforward to calculate the generalized fusion coefficients of F (so(8)1 ; S3 ) using the formula (2.62). For example, the coefficient N (0) (0) (0) can be obtained as follows,
N (0) (0) (0) =
S(0)
S S
(0) (0)
SO
I ()I ( )
S(0)
S
S
O (0) O (0) O
SOO
= 1.
(2.77)
The other cases can be calculated in the same manner; we give the result in Table 1. As this table
shows explicitly, the generalized fusion algebra F (so(8)1 ; S3 ) is non-commutative. For instance,
Table 1
Multiplication table of the generalized fusion algebra F (so(8)1 ; S3 ). The subscripts stand for the automorphism type of
twisted representations. Since = 2 = , this algebra is non-commutative
O
V
S
C
(0)
(0) 2
(0)
(1)
(0)
(1)
(0) 2
(1) 2
O
V
S
C
(0)
(0) 2
(0)
(1)
(0)
(1)
(0) 2
(1) 2
(0)
(0) 2
O
V
S
C
(0)
(0) 2
(0)
(1)
(0)
(1)
(0) 2
(1) 2
V
O
C
S
(0)
(0) 2
(0)
(1)
(1)
(0)
(1) 2
(0) 2
S
C
O
V
(0)
(0) 2
(1)
(0)
(1)
(0)
(0) 2
(1) 2
C
S
V
O
(0)
(0) 2
(1)
(0)
(0)
(1)
(1) 2
(0) 2
(0)
(0)
(0)
(0)
2(0) 2
O +V +S +C
(0) 2 + (1) 2
(0) 2 + (1) 2
(0) + (1)
(0) + (1)
(0) + (1)
(0) + (1)
(0) 2
(0) 2
(0) 2
(0) 2
O +V +S +C
2(0)
(0) + (1)
(0) + (1)
(0) 2 + (1) 2
(0) 2 + (1) 2
(0) + (1)
(0) + (1)
(0)
(1)
(0)
(1)
(0) 2
(1) 2
(0)
(0)
(1)
(1)
(0) + (1)
(0) 2 + (1) 2
O +V
S +C
(0)
(0)
(0) 2
(0) 2
(1)
(1)
(0)
(0)
(0) + (1)
(0) 2 + (1) 2
S +C
O +V
(0)
(0)
(0) 2
(0) 2
(0)
(1)
(1)
(0)
(0) 2 + (1) 2
(0) + (1)
(0) 2
(0) 2
O +C
V +S
(0)
(0)
(1)
(0)
(0)
(1)
(0) 2 + (1) 2
(0) + (1)
(0) 2
(0) 2
V +S
O +C
(0)
(0)
(0) 2
(1) 2
(0) 2
(1) 2
(0) + (1)
(0) + (1)
(0)
(0)
(0) 2
(0) 2
O +S
V +C
(1) 2
(0) 2
(1) 2
(0) 2
(0) + (1)
(0) + (1)
(0)
(0)
(0) 2
(0) 2
V +C
O +S
343
the fusion of (0) with (0) yields

(0) (0) = (0) + (1) ,
(0) (0) = (0) 2 + (1) 2 .
(2.78)
Since 2 = , these two expressions are distinct from each other, namely (0) (0) =
(0) (0) , which shows the non-commutativity of this algebra. One can also check the associativity of this algebra, for example,

(0) (0) (0) = (0) (0) + (0) (1) = 2(0) 2 + 2(1) 2 ,

(0) (0) (0) = 2(0) 2 (0) = 2(0) 2 + 2(1) 2 .
(2.79)
A proof for the associativity of the generalized fusion algebras is given in the next section.
The generalized fusion algebra F (so(8)1 ; S3 ) contains non-trivial subalgebras other than the
ordinary fusion algebra F (so(8)1 ). Actually, a subalgebra of F (so(8)1 ; S3 ) exists for each subgroup of S3 ,

{1}, , , 2 , = A3
(2.80)
= Z3 .
For example, corresponding to the three elements {1, , 2 } of the alternating group A3 S3 , the
2
representations in I, I and I form a subalgebra which we denote as F(so(8)1 ; A3 ). The generalized fusion algebra F(so(8)1 ; S3 ) is then decomposed into representations of F (so(8)1 ; A3 ).
2
Indeed, from Table 1, one can check that the set I I I is invariant under the action
of F(so(8)1 ; A3 ) and hence gives a representation of it. Together with the regular representation
2
of F(so(8)1 ; A3 ) on I I I , we eventually obtain two representations of F (so(8)1 ; A3 )
from the decomposition of the regular representation of F (so(8)1 ; S3 ).
3. Irreducible representations
For ordinary fusion algebras, the formula (2.62) for the generalized fusion coefficients reduces
to the ordinary Verlinde formula
NLM N =
SL SM SN
I
S0
(3.1)
from which one can show that the generalized quantum dimension
b (L) =
SL
S0
( I)
realizes a one-dimensional representation of the fusion algebra F(A),

b (L)b (M) =
NLM N b (N ).
(3.2)
(3.3)
NI
The aim of this section is to extend this result to the case of the generalized fusion algebra
F(A; G) for a general finite group G. As we shall show below, we have irreducible representations with dimension greater than 1 due to the non-commutativity of F (A; G) if G is
non-Abelian.
344
We start our analysis by introducing a new basis for twining characters. Namely, we define
the following linear combination

dim ab
ab
() (q)
(;) (q) =
|S()|
S ()

Lc
dim
ab
0
24
=
(3.4)
R () q
Tr
.
|S()|
S ()
Here S() G is the stabilizer of defined in (2.35), and Irr(S()) is a unitary irreducible
representation of S(),

() = ab () 1a,bn U (n; C) (n = dim ).
(3.5)
ab
The set {(;)
| Irr(S()); 1 a, b dim } forms a basis for the twining characters of
ab ,
I. Actually, one can express in terms of (;)
1
(q) =
|S()|
ab
dim ab ()(;)
(q).
(3.6)
Irr(S ())
1a,bdim
Here we used the orthogonality relations for matrix elements of irreducible representations
ij ac bd
1 ab
i (h)jcd (h) =
|H |
dim i
hH
1
|H |

i , j Irr(H ) ,
(dim ) ab (h) ab (h ) = hh
(h, h H ),
(3.7a)
(3.7b)
Irr(H )
1a,bdim
which hold for any finite group H .

ab , the modular transformation of the character (L I)
can be expressed
In terms of (;)
L
as follows,

L (q) =
SLL L (q)
I (L )
I (L )
1
L
SL
|S()|
ab
dim ab (L )(;)
(q)
Irr(S ())
1a,bdim
L
SL
I (L ) Irr(S ())
1a,bdim
We write this in the following form,

S ab
L (q) =
dim ab
ab
(q).
(L )(;)
|S()|
ab
L(;) (;) (q).
(;)I 1a,bdim
(3.8)
(3.9)
345
Here the set I and the matrix S are defined as follows,

I = (; ) | I; Irr S() ,

L
dim ab
S
ab
L |S ()| (L ) ( I(L ), Irr(S())),
SL(;) =
0
(
/ I(L )).
(3.10)
(3.11)
We call S the generalized S-matrix of the chiral algebra A. This matrix S is square since

I =
I() =
S() =
=
|I|
(dim )2 =
(dim )2 .
G
I Irr(S ())
(;)I
(3.12)
Moreover, S is unitary,

ab
ab
SL(;)
SM(;)
(;)I ;a,b
I (L )I (M )
L M
SL
SM
dim
ab (L ) ab (M )
|S()|
Irr(S ()) a,b
L L
SL
SM L M
I (L )
= LM ,
(3.13)
where we used the orthogonality relation (3.7b). The same matrix for the case of G
= Z2 is found
in [9,10].
One can express the generalized fusion coefficients (2.62) in a form completely parallel to the
ordinary Verlinde formula (3.1) using the generalized S-matrix (3.11) in place of the ordinary
S-matrix. Namely, we can prove the following formula
ab
bc
ac
SL(;)
SM(;)
SN
(;)
N LM N =
11
S0(;)
(;)I ;a,b,c
(3.14)
This can be readily checked by a straightforward calculation,

r.h.s. of (3.14)

=
I (L )I (M )I (N )
L M
SL
SM SNN
S0
L M N
SL
SM SN
I (L )I (M )I (N )

I (L )I (M )
= N LM N ,
S0
L M N
SL
SM SN
S0
dim
ab (L ) bc (M ) ac (N )
|S()|
Irr(S ()) a,b,c

Irr(S ())
dim

1
aa L M N
|S()|
a
L M ,N
(3.15)
where we used the orthogonality relation (3.7b) for h = 1. We call Eq. (3.14) the generalized
Verlinde formula.
346
The similarity of Eq. (3.14) to the ordinary Verlinde formula (3.1) suggests that the quantum
dimension (3.2) is also generalized to the case of generalized fusion algebras. Actually, one can
prove the following

b(;) (L)b(;) (M) =
(3.16)
N LM N b(;) (N ),
NI
where b(;) (L) is a matrix defined as

b(;) (L) =
ab

SL(;)
11
S0(;)

n = dim , (; ) I .
Mn (C)
(3.17)
1a,bn
The proof of Eq. (3.16) is straightforward,

N LM N bab
(;) (N ) =
NI
a b
b c
a c
ab
SL(
; ) SM( ; ) SN ( ; ) SN (;)
NI ( ; )I ;a ,b ,c
a b
b c
SL(
; ) SM( ; )
11
11
S0(
; ) S0(;)
( ; )I ;a ,b ,c

11
S0(;)
a a c b
ab
bb
SL(;)
SM(;)
b
11
S0(
; )
11
11
S0(;)
S0(;)

bb
bab
(;) (L)b(;) (M).
(3.18)
The above equation (3.16) means that b(;) defined in (3.17) realizes a representation of the
generalized fusion algebra F (A; G) by linearly extending its action on the basis (L) to the entire
algebra. For the case of ordinary fusion algebras, b(;) reduces to the quantum dimension b
since the generalized S-matrix S coincides with the ordinary S-matrix for ordinary fusion algebras. Therefore b(;) actually gives a generalization of the one-dimensional representations b
to the case of generalized fusion algebras.
Since we define the generalized fusion algebra by products of the boundary vertex operators,
we can expect that the generalized fusion algebra is associative. However the associativity is
not manifest from the definition (2.62) of the generalized fusion coefficients and its proof is
desirable. Let N L be a matrix (N L )M N = N ML N . The associativity of the generalized fusion
algebra is equivalent with the following condition for N L ,

N L N M =
(3.19)
N LM N N N .
NI
One can easily see that this follows from the generalized Verlinde formula (3.14). First note that
(N L )M N = N ML N =
(;);a,b ( ; );a ,b
(;);a,b ( ; );a ,b

ab
aa
SM(;)
bb
SL(;)
11
S0(;)
a b
SN
( ; )

ab
a b
aa bbb
SM(;)
(;) (L)SN ( ; )
347

N

b(;) (L) b(;) (L) S ,
= S

M
(;)
(3.20)
(dim ) terms
which shows that the matrix N L is similar to a direct sum of b(;) (L). Eq. (3.19) is an immediate
consequence of this fact since b(;) (L) satisfies the generalized fusion algebra (3.16) for all
(; ) I . The generalized fusion algebra is therefore associative and the matrix N L realizes
the (right) regular representation of the generalized fusion algebra.
We next show the mutual independence, in particular irreducibility, of b(;) for (; ) I .
To see this, we use the explicit form of b(;) (L),
L
SL
b(;) (L) = S0 (L ) (L S()),
(3.21)
0
(otherwise),
which follows from Eq. (3.11). Suppose that b(;) is reducible. Then there exists a matrix X
such that

1
Xb(;) (L)X =
(3.22)
for all L I.
0
However this implies that Irr(S()) is also reducible, since there exists at least one represen = 0 and () = S0 b
tation L I for each S() such that SL
(;) (L). This contradiction

SL
proves that b(;) is irreducible for all (; ) I .
In order to show the mutual independence of b(;) , we introduce the character of b(;) ,
L
SL

aa
b(;) (L) =
b(;) (L) = S0 (L ) (L S()),
(3.23)
0
(otherwise),
a
where is the group character of the representation Irr(S()). These characters satisfy the
following orthogonality relation

LI
b(;) (L)b( ; ) (L) =
bb
aa
SL(;)
SL(
; )
a,b
11
11
S0(;)
S0(
; )
Suppose that two representations b(;) and b( ; )

that
b(;) (L) = Y b( ; ) (L)Y 1
for all L I.
aa
a
(S0 )2 |dim
S ()|

1
2
|S ()| (S0 )
(3.24)
are similar, i.e., there exists a matrix Y such
(3.25)
Then the corresponding characters are the same, b(;) (L) = b( ; ) (L). However this implies
that (; ) = ( ; ) since

b(;) (L)2 > 0
(3.26)
b(;) (L)b( ; ) (L) =
LI
which is not possible for (; ) = ( ; ) from the orthogonality relation (3.24). Therefore two
representations b(;) and b( ; ) are distinct if (; ) = ( ; ). To summarize, we have shown
that {b(;) | (; ) I } is a set of mutually independent irreducible representations of the
generalized fusion algebra.
348
Table 2
2i
Irreducible representations 0 , 1 , 2 of the symmetric group S3 = , . is a cube root of 1, = e 3
0
1
2
1
1
1 0
1
1
0
1
1
0 1
1
1
0
10
1
1
0
01
1
1
0
We illustrate our results by the generalized fusion algebra F(so(8)1 ; S3 ) presented in Section 2.3. The stabilizer (2.35) for the representations (2.63) of so(8)1 is determined from the
action (2.65) of S3 on the representations,

S(O) = S3 ,
(3.27)
S(V ) = {1, },
S(S) = 1, 2 ,
S(C) = {1, }.
The stabilizers for V , S, C are isomorphic to Z2 , for which there are two irreducible representations. We denote them as Irr(Z2 ), which are defined for the case of S(V ) as
(1) = 1,
( ) = 1.
(3.28)
For the symmetric group S3 , there are three irreducible representations, 0 , 1 and 2 , two of
which are one-dimensional while the remaining one is two-dimensional (see Table 2). Accordingly, the set I in (3.10) takes the form

I = (; ) | I; Irr S()

= (O; 0), (O; 1), (O; 2), (V ; +), (V ; ), (S; +), (S; ), (C; +), (C; ) ,
(3.29)
where we write a (a = 0, 1, 2; ) as a for simplicity. The generalized S-matrix (3.11), which is
a 12 12 matrix in the present case, is then obtained in the following form,
1
S =
2 6
O
V
S
C
(0)
(0) 2
(0)
(1)
(0)
(1)
(0) 2
(1) 2
(O; 0) (O; 1) (O; 2)11 (O; 2)22 (O; 2)12 (O; 2)21 (V ; +)
1
1
0
0
2
2
3
2
2
0
0
1
1
3
0
0
3
1
1
2
2
1
1
0
0
3
2
2
2
2
0
0
0
8
8
2
8
8
0
0
2
0
0
0
2
2
2 2
6
0
0
2
2
6
2 2
0
0
2
2
0
2 2
0
0
2
2
0
2 2
0
0
2
2
0
2 2
2 2
0
0
2
2
0
(V ; )
3
3
3
3
0
0
6
6
0
0
0
0
(S; +)
3
3
3
3
0
0
0
0
0
0
6
6
(S; )
3
3
3
3
0
0
0
0
0
0
6
6
(C; +)
3
3
3
3
0
0
0
0
6
6
0
0
(C; )
3
3
3
3
0
0
0
0
6
6
0
0
(3.30)
2i
3
where = e . One can readily check that this matrix is unitary and that the generalized Verlinde formula (3.14) reproduces the generalized fusion coefficients of F (so(8)1 ; S3 )
one can construct the irreducible representations
given in Table 1. From this matrix S,
(3.17) of F (so(8)1 ; S3 ). We give the result in Table 3. From this table, one can check that
b(;) (L)((; ) I ) indeed satisfies the generalized fusion algebra in Table 1 and realizes a
representation of F(so(8)1 ; S3 ).
Before concluding this section, we comment on the case that the automorphism group G
is Abelian. Since all the irreducible representations are one-dimensional for an Abelian G, the
representation matrix () has only one component 11 (). Therefore we can omit the suffix
and write () instead of 11 (). This enables us to express the formulas we have given above
349
Table 3
2i
Irreducible representations of the generalized fusion algebra F (so(8)1 ; S3 ). is a cube root of 1, = e 3 . Unlike ordinary fusion algebras, there is an irreducible representation of dimension 2 due to the non-commutativity of F (so(8)1 ; S3 )
O
(0)
(0) 2
(0)
b(O;0) 1
1
1
1
2
2
2
b(O;1) 1
1
1
1
2
2
2
1 0 1 0 1 0 1 0 0 0 0 1
b(O;2)
2
2
2
01
01
01
01
0
0
10
b(V ;+) 1
1
1
1
0
0
2
b(V ;) 1
1
1
1
0
0
2
b(S;+) 1
1
1
1
0
0
0
b(S;) 1
1
1
1
0
0
0
b(C;+) 1
1
1
1
0
0
0
b(C;) 1
1
1
1
0
0
0
(1)
2
0 1
2
10
2
2
0
0
0
0
(0)
2
0
2
(1)
2
0
2
(0) 2
2
0
2
(1) 2
2
0
2
0
0
0
0
0
0
0
0
2
2
0
0
0
0
2
2
0
0
2
0
0
in a simple form. First, the basis (3.4) of twining characters and its modular transformation can
be written as follows,

1
(;) (q) =
() (q),
|S()|
S ()

L (q) =
SL(;) (;) (q),
(3.31)
(3.32)
(;)I
where the generalized S-matrix is defined as

SLL 1 (L ) ( I(L ), Irr(S())),
|S ()|
SL(;) =
0
(
/ I (L )).
(3.33)
The generalized Verlinde formula (3.14) then takes the form

N LM N =

(;)I
SL(;) SM(;) SN (;)

.
S0(;)
(3.34)
For G
= Z2 , the generalized S-matrix (3.33) and the generalized Verlinde formula (3.34) reproduce the result given in [9,10]. The matrix S is denoted as S in [9], which is considered to be a
particular case of the same matrix in [8] for a general finite Abelian G. This suggests that two
matrices S in (3.33) and S in [8] are the same, although we have no proof for this statement. If
this is the case, the classifying algebra in [8] is the dual of the generalized fusion algebra (3.34)
in the sense of C-algebras [1820] with the structure constants
M(;)( ; ) (
; )
SL(;) SL( ; ) SL( ; )

,
SL(0;0 )
(3.35)
LI
where 0 is the identity representation 0 () = 1 ( G) of G. Clearly, the irreducible repreS
sentation of this algebra takes the form (; ) L(;) and is labeled by L I.

SL(0;0 )
350
4. Boundary states as a NIM-rep of generalized fusion algebras

It is now well understood that boundary states preserving a chiral algebra A form a nonnegative integer matrix representation (NIM-rep) of the fusion algebra F (A) [2]. Since the
generalized fusion algebra F (A; G) is defined through the twisted boundary states associated
with the charge-conjugation modular invariant, it is natural to expect a close relationship between
F(A; G) and twisted boundary states for general modular invariants, such as simple current invariants [16] or exceptional ones. As we shall show below, this expectation turns out to be true;
a set of mutually consistent boundary states associated with any modular invariant realizes a
NIM-rep of the generalized fusion algebra.
4.1. Charge-conjugation invariants
We begin our analysis by rewriting the twisted boundary states for the charge-conjugation
modular invariant in the form that the relation to the generalized fusion algebra is more apparent.
We call them the regular states since their mutual overlap is expressed by the generalized fusion
coefficients, which realize the regular representation of the generalized fusion algebra.
ab
Corresponding to the character (;)
defined in Eq. (3.4), we introduce a new basis for the
twisted Ishibashi states,

dim ab
, ;
(4.1)
=
Irr S() .
() , ;
ab
|S()|
S ()
One can express the original basis in terms of |((, ); )ab as

, ; = 1
|S()|
Irr(S ())
1a,bdim

dim ab () , ; ab ,
(4.2)
where we used the orthogonality relation (3.7b) for the matrix elements of . We denote the set
of labels for the new basis |((, ); )ab of twisted Ishibashi states by E0 ,

E0 = , ; I; Irr S() .
(4.3)
This is essentially the same set as I defined in (3.10). We have introduced the new symbol E0
for the labels of Ishibashi states in order to reserve I for expressing the chiral quantity. The
b b ,
overlap of |((, ); )ab with |((, ); )a b yields (;)

1

, ; a b q 2 Hc , ; ab
dim dim
1 1

a b ( ) ab ()
(q)
=
|S()|
S
0
, S ()
dim dim
1

a b ( ) ab ( )
(q)
=
|S()|
S0

= aa

S ()
, S ()

b b ()
1
(q)
S0

=
1
S0
aa
= aa
|S()| b b
(q)
dim (;)
1
b b
(;)
(q).
11
S0(;)
351
(4.4)
as follows
Using |((, ); )ab , we can rewrite the regular state |L (L I)

L
SL
, ; L
|L =
I (L )
I (L )
1
L
SL
|S()|
((, );)E0
a,b
Irr(S ())
1a,bdim

dim ab (L ) , ; ab

ab
, ;
.
SL(;)
ab
(4.5)
The boundary state coefficients of the regular state |L are therefore given by the generalized
It is instructive to calculate the overlaps of |L starting from the form given above,
S-matrix S.
1
M|q 2 Hc |L =

(;)I a,b a ,b

(;) a,b b
=
=

ab
a b
aa
SM(;)
SL(;)
ab
ab
SM(;)
SL(;)
1
b b
(;)
(q)
11
S0(;)
11
S0(;)
ab
bb
ab
SL(;)
SN
(;) SM(;)
(;);a,b,b
11
S0(;)
b b
SN
(;) N (q)
N (q)
N LN M N (q)

=
(N N )L M N (q),
(4.6)
where we used the formula

ab
= SLba (;)
SL(;)
(4.7)
which follows from Eq.(2.59). In this way, we reproduce the regular representation N L of the
generalized fusion algebra as the overlap matrices of the regular states.
4.2. General cases
We next turn to the case of twisted boundary states associated with a general modular invariant Z (see (2.9)) of the chiral algebra A. In order to express twisted boundary states, we have
to determine what kinds of twisted Ishibashi states are available in the modular invariant (2.9).
As we have argued in Section 2.2, the twisted Ishibashi state |(, ); exists if and only if
= (). Accordingly, the set of labels for -twisted Ishibashi states, which we denote by E(),
352
takes the following form

E() = , , Spec(Z); = () ,
(4.8)
where Spec(Z) is the spectrum of bulk fields in Z (see (2.10)). A general boundary state |A
satisfying the boundary condition (2.29) is then written as

|A =
(4.9)
A V V .
A(,
) , ;
(, )E ()
Here V is the set of labels for the -twisted boundary states. We have also introduced the set V
of all the labels of twisted boundary states,

V =
(4.10)
V.
G
As in the case of the regular states, we assume that the number of twisted boundary states is equal
to that of twisted Ishibashi states [3],

V = E() ( G).
(4.11)
This condition together with the mutual consistency of boundary states implies that the boundary states labeled by V form a NIM-rep of the (ordinary) fusion algebra F(A) of A [2]. In
particular, the matrix ( )A(, ) = A(,

) is unitary.
In the same way as the regular states (2.60), the overlap of two twisted boundary states can be
expanded into a sum of the characters for representations of twisted chiral algebras,
B|q 2 Hc |A =
(, )E (A )E (B )
(, )E (A )E (B )

1
NI A
A
B
A(,
) B(, )
1 B1 A
(q)
S0
A
B
A(,
) B(, )
1
S0

B
(, )E (A )E (B )
1 A
SNB
N (q)
1
NI B A
1 B
SNA
A
A(, )
S0
B
B(,
) N (q)
(n N )A B N (q),
(4.12)
1
NI A B
|V|
matrix
where A is the automorphism type of |A and we used the formula (2.38). A |V|
n N is defined for each N I as
(n N )A B =

(, )E (A )E (B )
A
A(,
)
N
SN
S0
B
B(,
) A N ,B .
(4.13)
The entry of n N should be non-negative integer for the mutual consistency of boundary states,
since it represents the multiplicity of representation N I in the open-string spectrum. In the
satisfy the generalized fusion
rest of this section, we show that these matrices {n N | N I}
algebra F (A; G).
353
We first rewrite the boundary states (4.9) in a form similar to the regular case (4.5). To do
so, we have to generalize the basis (4.1) of Ishibashi states for (, ) Spec(Zc ) to that for
(, ) Spec(Z). In contrast with the case of the charge-conjugation invariant Zc , the label
(, ) of bulk fields is in general not symmetric; there exists some (, ) Spec(Z) with
= . Although the untwisted Ishibashi state can not be constructed for (, ) with = , it is
possible to obtain twisted states for (, ) if = () for some G. In other words, twisted
Ishibashi states are available for (, ) if belongs to the G-orbit G of ,

G = () | G I.
(4.14)
/ G does
In the following, we consider only the case of G, since the label (, ) with
not appear in the twisted boundary states for the automorphism group G.
Let be an element of G that relates G with , namely = (). In general,
this element cannot be determined uniquely due to the stabilizer S() of . Suppose that
G satisfying = (). Then 1 is an element of S()
we have another element

1
1
belongs to the coset
() =
() = . In other words,
since

S() = | S() G.
(4.15)
This result shows that, for a given (, ) Spec(Z), the twisted Ishibashi state |(, );
exists if and only if S().
Based on this fact, one can generalize the basis (4.1) of twisted Ishibashi states in the following
manner,

dim ab
, ;
=
() , ;
ab
|S()|
S ()

= (); Irr S() .
(4.16)
Similarly to Eq. (4.2), one can express the original basis |(, ); in terms of |((, ); )ab ,

, ; = 1
|S()|

S() .

Irr(S ())
1a,bdim
1

dim ab
, ; ab
(4.17)
The overlap of |((, ); )ab with |((, ); )a b can be calculated in exactly the same way
as |((, ); )ab ,

1

, ; a b q 2 Hc , ; ab
1

dim dim

=
a b ( ) ab () , ; q 2 Hc , ;
|S()|
, S ()
dim dim
1 1

=
a b ( ) ab ()
(q)
|S()|
S
0
, S ()
dim dim
1

a b ( ) ab ( )
(q)
=
|S()|
S
0

, S ()
354
= aa
b b ()
S ()
aa
1
(q)
S0
1
b b
(;)
(q).
11
S0(;)
(4.18)
The overlap between two states with different labels, ( , ) = (, ), vanishes from the formula (2.38).
Substituting (4.17) in (4.9), the boundary state |A can be brought to the following form,

A

A(,
|A =
) , ; A
(, )E ()
(, )E ()
1
A
A(,
)
|S()|
((, );)E
a,b

Irr(S ())
1a,bdim
1

dim ab
A , ; ab

ab
, ;
,
A((,
);)
ab
(4.19)
where we denote by E the set of labels for the basis |((, ); )ab available in the modular
invariant (2.9),

!

E=
, ; ,
(4.20)
E(); Irr S() .
G
It should be noted that some (, ) may appear more than once in E corresponding to the
multiple occurrence of bulk fields with representation (, ) in the modular invariant (2.9). The
A
ab
boundary state coefficient A((,
);) is related to the original coefficient A(, ) as follows

A
dim ab 1
ab
A(, ) |S ()| ( A ) ((, ) E(A ), Irr(S())), (4.21)
A((, );) =
0
((, )
/ E(A )).
This form is completely parallel to the generalized S-matrix (3.11). Actually, from the unitarity
of for all G, one can prove that the matrix is also unitary,

V =
E() =
S() =
=
(dim )2 ,
|V|
(4.22)
G
(, )
"
G E ()
((, );)E
ab
ab
A((,
);)
B((, );)
((, );)E ;a,b
(, )E (A )E (B )
A
B
A(,
) B(, )
dim
1 1
A ab B
ab
|S()|
Irr(S ()) a,b
(, )E (A )
= AB ,
A
A
A(,
) B(, ) A B
(4.23)
355
1
1
where we used the fact that
A =
B in S() G implies A = B in G.
Using the form given in (4.19) together with the formula (4.18), one can calculate the overlap
of two boundary states |A and |B in the following way,

1
1
aa
b b
ab
ab
B|q 2 Hc |A =
(;)
(q)
B((,
);) A((, );)
11
S

0(;)
a,b
a ,b
((, );)E

((, );)E a,b b
=
=
((, );)E
a,b b
ab
ab
B((,
);) A((, );)
1
b b
SN
(;) N (q)
11
S0(;)
NI
bb
SN
(;)
ab
ab
B((,
);) A((, );)
NI
ab
A((,
);)
aa

NI ((, );)E a,b a b
11
S0(;)
N (q)
bb
a b
b(;) (N ) B((,
);) N (q).
(4.24)
Comparing this with Eq. (4.12), we obtain the following expression for the overlap matrix n N ,

n N =
(4.25)
b(;) (N ) b(;) (N ) .

((, );)E
(dim ) terms
This has exactly the same structure as Eq. (3.20) for the regular representation matrix N N , from
which we have shown N N satisfies the generalized fusion algebra. We can repeat the same thing
here since the matrix is unitary from the assumption (4.11) of completeness. For a unitary ,
the above expression (4.25) for the overlap matrix n N means that n N is similar to a direct sum of
b(;) (N ). Since b(;) (N ) satisfies the generalized fusion algebra for all (; ) I , the matrix
n N also satisfies the generalized fusion algebra. The mutual consistency of twisted boundary
states requires that the coefficients (n N )A B of N in the cylinder amplitude take values in non Hence we are led to the conclusion that the overlap matrix n N for
negative integers for all N I.
a set of mutually consistent boundary states realizes a NIM-rep of the generalized fusion algebra
F (A; G) if the condition (4.11) is satisfied.
5. Examples
As we have shown in the previous section, a consistent set of twisted boundary states in any
modular invariant form a NIM-rep of the generalized fusion algebra. In this section, we check this
for three concrete chiral algebras, u(1)k , su(3)k and su(3)3
1 , by explicitly constructing twisted
boundary states for each case.
5.1. u(1)k
The simplest chiral algebra with non-trivial automorphisms is u(1)k , where the level k is a
positive integer. There are 2k irreducible representations labeled by positive integer n mod 2k,

I = (n) | n = 0, 1, . . . , 2k 1 ,
(5.1)
for which the modular transformation matrix takes the form
356
i
1
Smn = e k mn .
(5.2)
2k
Therefore, for the charge-conjugation modular invariant, we have 2k untwisted boundary states,

(m I).
Smn n, n
|m =
(5.3)
nI
The charge-conjugation c is an automorphism of u(1)k , which acts on I as

c : (n) n = (n) = (2k n) (n I).
There are two representations fixed by c ,

I(c ) = (0), (k) .
(5.4)
(5.5)
Corresponding to this, we have two twisted boundary states, which we denote by | ,

1
| = (0, 0); c (k, k); c .
(5.6)
2
The overlap of | with the untwisted state |0 yields the character of the twisted chiral
algebra u(1)k c (see Eq. (2.46)),

1
1
(q) = 0|q 2 Hc | = 0c (q)
kc (q)
,
2
(5.7)
where nc is the twining character of (n) for c . From this equation, one obtains the modular
transformation matrix S c

1 1 1
c
S =
(5.8)
,
2 1 1
where the rows and the columns are ordered as I c = {+, } and I(c ) = {0, k}, respectively.
Since c1 = c and S c is real, the relation (2.59) implies that the twisted representations ()
I c are self-conjugate, ( ) = ().
Using these data and the formula (see Eq. (2.60))

1 H
N LM N N (q) (L, M I),

L q 2 c |M = (L) (M) =
(5.9)
NI
one can determine the generalized fusion coefficients N LM N for the chiral algebra u(1)k and the
automorphism group Gc = {1, c }
= Z2 . The result is as follows,
(l) (m) = (l + m),
() (l = 0 mod 2),
() (l = 1 mod 2),
(+) (+) = () () = (0) + (2) + + (2k 2),
(l) () = () (l) =
(+) () = () (+) = (1) + (3) + + (2k 1),
(5.10)
where (l), (m) I and ()

This defines the generalized fusion algebra F (u(1)k ; Gc )
for Gc = {1, c }
= Z2 . As is easily seen, this algebra is generated by (1) and (+). Hence
F(u(1)k ; Gc ) is an extension of the ordinary fusion algebra F (u(1)k ), which is the group
algebra of Z2k , by the twisted representation (+). Note that F(u(1)k ; Gc ) is commutative,
(L) (M) = (M) (L), reflecting the fact that the automorphism group Gc
= Z2 is Abelian.
I c .
357
We next turn to the case of modular invariants other than the charge-conjugation one and check
that the associated boundary states form a NIM-rep of the generalized fusion algebra (5.10). We
consider two cases: a simple current extension of k = 4 and a non-diagonal invariant of k = 6.
5.1.1. k = 4
For k = 4, we have the following block-diagonal invariant
Z = |0 + 4 |2 + |2 + 6 |2 ,
(5.11)
which is a simple current extension [16] of u(1)4 by (4) I. The chiral algebra of the extended
theory is u(1)1 and each block corresponds to an irreducible representation of u(1)1 . Since there
are two irreducible representations for u(1)1 , we have two untwisted boundary states that keep
u(1)1 (see Eq. (5.3))

1
(5.12)
(0 , 0 ) (1 , 1 ) ,
2
where we distinguish representations of u(1)1 from those of u(1)4 by putting a prime. Since
u(1)1 contains u(1)4 as a subalgebra, we can regard these states as untwisted boundary states of
u(1)4 associated with the modular invariant (5.11). From the branching rule
(0 ) = (0) (4),
(1 ) = (2) (6),
(5.13)
and our normalization (2.7) for Ishibashi states, we can express Ishibashi states of u(1)1 by those
of u(1)4 ,

(1 , 1 ) = 1 2, 2 + 6, 6 .
(0 , 0 ) = 1 (0, 0) + (4, 4) ,
(5.14)
2
2
Substituting this into Eq. (5.12), we obtain two untwisted boundary states of u(1)4 ,

= 1 (0, 0) + (4, 4) + 1 2, 2 + 6, 6 ,
|0
2
2

1
1

=
.
(0, 0) + (4, 4) 2, 2 + 6, 6
|2
(5.15)
2
2
Since there are four untwisted Ishibashi states in (5.11),

E(1) = (0, 0), 2, 2 , (4, 4), 6, 6 ,
(5.16)
we need two more boundary states for completeness. The remaining two can be constructed by,
e.g., the fusion with representations of the unextended chiral algebra [2,26,27]. (See [27] for a
detailed exposition of this procedure.) The result is as follows,

= 1 (0, 0) (4, 4) i 2, 2 6, 6 ,
|1
2
2

i
1

.
(0, 0) (4, 4) + 2, 2 6, 6
|3 =
(5.17)
2
2
Consequently,the boundary state coefficients for untwisted boundary states associated with
(5.11) take the following form
1 1
1
1
1 1 i 1 i
=
(5.18)
,
2 1 1 1 1
1 i
1 i
where the column is ordered as in (5.16).
358
The restriction of the charge-conjugation of u(1)1 to the subalgebra u(1)4 is clearly the
charge-conjugation of u(1)4 . In other words, the charge-conjugation of u(1)4 has a lift to the
extended chiral algebra u(1)1 . (The lifting of automorphisms for RCFTs is studied in [28].) This
enables us to construct twisted boundary states of u(1)4 starting from those of u(1)1 . Since
two representations of u(1)1 are fixed by the charge-conjugation, we have two twisted boundary
states of u(1)1 ,

1
(5.19)
(0 , 0 ); c (1 , 1 ); c .
2
This is completely the same form as the untwisted states. Hence one can proceed in the same
way as the untwisted states to obtain four twisted states of u(1)4 . The boundary state coefficients
c of twisted states coincide with given in (5.18) with the spectrum

E(c ) = (0, 0), 6, 2 , (4, 4), 2, 6
(5.20)
instead of E(1).
From these data, one can calculate the mutual overlap of boundary states and the associated
open string spectrum n defined in (4.13). Since we have 4 + 4 = 8 boundary states, n is a 8 8
matrix and takes the form
(4) l

(n1 )
O
O n
n l =
(5.21)
=
,
(l
=
0,
1,
.
.
.
,
7),
n
(4)
n O
O
(n1 )l
where O is a 4 4 matrix with all entries being 0 and n1, are defined as
0 1 0 0
1 0 1 0
0
0 0 1 0
0 1 0 1
1
(4)
n1 =
n+ =
n =
,
,
0 0 0 1
1 0 1 0
0
1 0 0 0
0 1 0 1
1
1
0
.
1
0
(5.22)
One can confirm that these 10 matrices actually satisfy the generalized fusion algebra (5.10) for
k = 4, namely,
(n 1 )8 = 1,
n 1 n + = n ,
n + n + = n 0 + n 2 + n 4 + n 6 .
1
0
1
0
0
1
0
1
(5.23)
The boundary states associated with the simple current extension (5.10) therefore realize a NIMrep of the generalized fusion algebra F (u(1)4 ; Gc ). Note that this 8-dimensional NIM-rep is
distinct from the regular one, which is 8 + 2 = 10 dimensional. Hence we have obtained two
NIM-reps of F (u(1)4 ; Gc ); one of them corresponds to the charge-conjugation invariant whereas
the other originates from the block diagonal invariant (5.10).
Since the generalized fusion algebra contains the ordinary fusion algebra as a subalgebra,
one can decompose a NIM-rep of the former into NIM-reps of the latter. Conversely, several
NIM-reps (more precisely, |G| NIM-reps) of the ordinary fusion algebra may be combined to
yield a NIM-rep of the generalized fusion algebra. For u(1)4 , there are four NIM-reps of the
ordinary fusion algebra F (u(1)4 ) corresponding to four subgroups of the simple current group
Z8 [7]. Two of them, the regular NIM-rep and the two-dimensional one, form the regular NIMrep of the generalized fusion algebra. The four-dimensional one (5.18) is paired with itself to
form a 8-dimensional NIM-rep of the generalized fusion algebra, as we have seen above. The
remaining one, which is one-dimensional and unphysical, is also paired with itself to yield a
two-dimensional NIM-rep of the generalized fusion algebra. Consequently, there are three NIMreps for the generalized fusion algebra F (u(1)4 ; Gc ); two of them are physical and correspond to
359
the charge-conjugation modular invariant and the simple current extension (5.10), respectively,
while the remaining one is unphysical and has no associated modular invariant.
5.1.2. k = 6
For k = 6, we have the following non-diagonal invariant,
Z = |0 |2 + |2 |2 + |4 |2 + |6 |2 + |8 |2 + |10 |2
+ 1 7 + 3 9 + 5 11 + 7 1 + 9 3 + 11 5 ,
(5.24)
which is the Z2 -orbifold of u(1)6 . Let |m (m I = {0, 1, 2, . . . , 11}) be the untwisted boundary
state in the charge-conjugation modular invariant of u(1)6 defined in (5.3). The orbifold action
on |m reads |m |m + 6 , and the untwisted state in (5.24) can be obtained by averaging |m
over the Z2 -orbit,

i
1
1
|m Z2 = |m + |m + 6 =
(5.25)
e 3 mn 2n, 2n .
6 n=0,1,...,5
2
The c -twisted boundary states in (5.24) can be constructed as the untwisted states in the Z3 orbifold of u(1)6 , since the Z3 -orbifold is the T-dual of (5.24). The untwisted states in the Z3 orbifold are obtained by averaging |m over the Z3 -orbit,

1

i
1
e 2 mn 3n, 3n .
|m Z3 = |m + |m + 4 + |m + 8 =
(5.26)
2
3
n=0,1,2,3
The c -twisted states in the Z2 -orbifold (5.24), which we denote by |m; c Z2 , are constructed
by acting c on the holomorphic sector of |m Z3 ,
|m; c Z2 = R(c )|m Z3 =
1
2

i
e 2 mn 12 3n, 3n ; c .
(5.27)
n=0,1,2,3
In this way, we obtain 6 untwisted states and 4 twisted states in (5.24). The overlap matrix n is a
10 10 matrix and takes the form,
(6) l
(6,6)

(n1 ) O (6,4)
O
n
=
(l
=
0,
1,
.
.
.
,
11),
n
,
n l =
(5.28)
l
nT
O (4,4)
O (4,6) (n(4)
1 )
(4)
where n1 is defined in (5.22), O (m,n) is a m n zero matrix and the other matrices are defined
as
0 1 0 0 0 0
1 0 1 0
0 0 1 0 0 0
0 1 0 1
0 0 0 1 0 0
1 0 1 0
(6)
n1 =
n+ =
,
,
0 0 0 0 1 0
0 1 0 1
0 0 0 0 0 1
1 0 1 0
1 0 0 0 0 0
0 1 0 1
0 1 0 1
1 0 1 0
0 1 0 1
n =
(5.29)
.
1 0 1 0
0 1 0 1
1 0 1 0
360
One can easily check that these matrices satisfy the generalized fusion algebra (5.10) for k = 6.
The simple current group of u(1)6 is isomorphic to Z12 . Hence we have 6 NIM-reps of the
ordinary fusion algebra F (u(1)6 ) corresponding to 6 divisors {1, 2, 3, 4, 6, 12} of 12 [7]. In the
same way as the case of k = 4, two of them, the regular and the two-dimensional ones, form
the regular NIM-rep of the generalized fusion algebra F (u(1)6 ; Gc ). NIM-reps of dimension 6
and 4 are combined to yield another NIM-rep of F (u(1)6 ; Gc ) associated with the Z2 -orbifold
(or equivalently, the Z3 -orbifold) of u(1)6 , as we have shown above. In contrast with the case of
k = 4, however, the remaining two NIM-reps of dimension 1 and 3 cannot form a NIM-rep of
F (u(1)6 ; Gc ). Actually, one can show that the overlap matrices n containing the one-dimensional
NIM-rep of F(u(1)6 ) as a factor and satisfying the generalized fusion algebra F(u(1)6 ; Gc )
necessarily have non-integral entries and cannot be a NIM-rep. The case of the three-dimensional
NIM-rep is shown in the same way. Consequently, we have two NIM-reps of F (u(1)6 ; Gc ), one
of which corresponds to the charge-conjugation (or the diagonal) invariant while the other is
associated with the Z2 -orbifold (or its T-dual) of u(1)6 .
5.2. su(3)k
Our next example is su(3)k with the charge-conjugation c as an automorphism of the chiral
algebra. Since the automorphism group Gc = {1, c } consists of two elements, we can expect
that two NIM-reps of the ordinary fusion algebra F(su(3)k ) are combined to yield a NIM-rep of
the generalized fusion algebra F (su(3)k ; Gc ). We check this for the case of k = 3 and k = 5, in
which there are respectively four and six independent NIM-reps of F (su(3)k ). We present here
only the outline of our analysis and give the details in Appendix B.
5.2.1. k = 3
For k = 3, there are four NIM-reps (A(6) , A(6) , D (6) and D (6) 9 ) of the ordinary fusion algebra F (su(3)3 ) [2,19]. Two of them, A(6) and A(6) , correspond to respectively the untwisted
and the twisted boundary states in the charge-conjugation modular invariant and constitute the
regular NIM-rep of the generalized fusion algebra F(su(3)3 ; Gc ). The remaining two, D (6) and
D (6) , express respectively the untwisted and the twisted boundary states corresponding to the
following block diagonal modular invariant,
ZD (6) = |(0,0) + (3,0) + (0,3) |2 + 3|(1,1) |2 ,
(5.30)
which is a simple current extension of su(3)3 by (3, 0). Here the subscripts of characters denote
the Dynkin label of su(3). The extended chiral algebra of this invariant is so(8)1 and one can
construct the boundary states associated with (5.30) using the data of so(8)1 . Actually, as is
c to so(8), which is isomorphic
shown in Appendix A, the automorphism group Gc has a lift G
to the symmetric group S3 . Accordingly, the boundary states for D (6) and D (6) are obtained
c -twisted states in so(8). By an explicit calculation of the overlap of the twisted states
by the G
with the untwisted ones in su(3), we can show that two NIM-reps, D (6) and D (6) , of F(su(3)3 )
form a NIM-rep of F (su(3)3 ; Gc ). In this way, four NIM-reps of the ordinary fusion algebra
F(su(3)3 ) are organized into two NIM-reps of the generalized fusion algebra F(su(3)3 ; Gc );
one corresponds to the charge-conjugation modular invariant while the other is originated from
the simple current extension (5.30).
9 The superscript stands for the sum k + h = k + 3, where h is the dual Coxeter number of su(3).
361
5.2.2. k = 5
For k = 5, there are six NIM-reps (A(8) , A(8) , D (8) , D (8) , E (8) and E (8) ) of the ordinary fusion algebra F(su(3)5 ) [2,19]. Similarly to the case of k = 3, A(8) and A(8) express the boundary
states in the charge-conjugation invariant and form the regular NIM-rep of the generalized fusion
algebra F(su(3)5 ; Gc ). The remaining four correspond to other modular invariants. Two of them,
D (8) and D (8) , represent the boundary states in the simple current invariant
ZD (8) = |(0,0) |2 + |(1,1) |2 + |(2,2) |2 + |(3,0) |2 + |(0,3) |2 + |(4,1) |2 + |(1,4) |2
+ ((5,0) (0,5) + (3,1) (1,3) + (1,2) (2,1)
+ (2,3) (0,2) + (2,0) (3,2) + (0,4) (1,0) + (0,1) (4,0) + c.c.),
while
E (8)
and
E (8)
(5.31)
correspond to the exceptional invariant
ZE (8) = |(0,0) + (2,2) |2 + |(0,2) + (3,2) |2 + |(1,2) + (5,0) |2

+ |(3,0) + (0,3) |2 + |(2,1) + (0,5) |2 + |(2,0) + (2,3) |2 ,
(5.32)
which originates from the conformal embedding su(3)5 su(6)1 .

The invariant (5.31) is the Z3 -orbifold of su(3)5 and we can obtain untwisted states by averaging the Z3 -orbit of untwisted states in the charge-conjugation invariant. On the other hand,
the twisted states in the charge-conjugation invariant are fixed points of the orbifold action and
we need an appropriate resolution of them by the Ishibashi states from the twisted sector of the
orbifold to obtain twisted states in (5.31). For the invariant (5.32), one can construct twisted
boundary states from the data of su(6), since the charge-conjugation of su(3) is lifted to that of
su(6). One obtains six untwisted states together with two twisted states using the data of su(6).
The remaining states of E (8) and E (8) can be obtained by, e.g., the fusion with representations
of su(3).
From the explicit form of boundary states in (5.31) and (5.32), we can calculate the overlap of
twisted states with untwisted ones to show that two NIM reps of F (su(3)5 ) corresponding to the
same modular invariant form a NIM-rep of F (su(3)5 ; Gc ). Consequently, we obtain three NIMreps of the generalized fusion algebra F (su(3)5 ; Gc ) corresponding to three modular invariants
available for k = 5.
5.3. su(3)3
1
The final example we consider is su(3)3
1 = su(3)1 su(3)1 su(3)1 . The details are given in
Appendix C. This chiral algebra has the automorphism group consisting of all the permutations
of three factors, which is isomorphic to S3 , and we obtain a non-commutative generalized fusion
3
algebra F (su(3)3
1 ; S3 ). For the chiral algebra su(3)1 , we have a non-trivial block-diagonal
modular invariant
ZE6 = |(0,0,0) + (1,1,1) + (2,2,2) |2
+ |(0,2,1) + (1,0,2) + (2,1,0) |2 + |(0,1,2) + (2,0,1) + (1,2,0) |2 ,
(5.33)
where the subscripts stands for the label of representations of su(3)3

1 and the numbers 0, 1, 2
correspond respectively to the fundamental weights 0 , 1 , 2 of su(3)1 . This invariant orig10 Since the automorphism group S of
inates from the conformal embedding su(3)3
3
1 E6,1 .
10 E
6,1 means E6 at level 1.
362
su(3)3
1 has a lift to E6,1 , we can construct twisted boundary states for S3 [11,29] in this invariant. In the same way as the case of su(3)k , we can calculate the overlap of twisted states
to check that the overlap matrices indeed form a NIM-rep of the generalized fusion algebra
F(su(3)3
1 ; S3 ).
6. Graph fusion algebras
Given a boundary state coefficient matrix ( V, E)11 and 0 V such that 0 = 0,
one can define the graph fusion algebra [2,19,20]

() () =
(6.1)
G ( ) (, V)
V
with the structure constant

G =

E
(6.2)
Since this has the same form as the Verlinde formula (3.1), one can consider the graph fusion
algebra as a generalization of ordinary fusion algebras. Actually, for the case of boundary states
associated with a block diagonal invariant, it is observed [19] that the graph fusion algebra contains the fusion algebra of the extended theory as a subalgebra.
For the case of simple current extensions, this relation of graph fusion algebras and the fusion
algebra of the extended theory can be explained from the point of view of the generalized fusion
algebra. As an illustration, we consider the D-type modular invariant of su(2)4 ,
Z = |0 + 4 |2 + 2|2 |2 ,
(6.3)
where the subscript stands for the Dynkin label of su(2). This invariant is obtained from a simple
current extension of su(2)4 by the simple current group = {(0), (4)}
= Z2 . The extended chiral
algebra is su(3)1 and one can regard (6.3) as the charge-conjugation invariant of su(3)1
Z = |(0,0) |2 + |(1,0) |2 + |(0,1) |2 ,
(6.4)
where representations are again denoted by their Dynkin labels. One obtains (6.3) from (6.4)
using the branching rule for the representations of su(3)1
(0, 0) = (0) (4),
(1, 0) = (2),
(0, 1) = (2).
(6.5)
The unextended chiral algebra in simple current extensions can be characterized by the fixed
point algebra of a certain automorphism group G of the extended chiral algebra [8], where
G is the character group for the simple current group used in the extension and hence
isomorphic to ; G =
= . For the case of su(2)4 su(3)1 , G is nothing but the chargeconjugation automorphism group Gc of su(3)1 , since su(2)4 is the fixed point algebra of the
charge-conjugation c . The eigenvalue of c is 1, since c has order 2. The vacuum representation of su(3)1 is therefore decomposed into two parts according to the eigenvalue of c .
Since c fixes su(2)4 su(3)1 , this decomposition of (0, 0) coincides with the decomposition
(6.5) into representations of su(2)4 . The representation (0) is su(2)4 itself and corresponds to
11 In this section, we omit the label for the anti-holomorphic sector of Ishibashi states for simplicity.
363
c = 1. The representation (4) contains the highest root of su(3), whose sign is reversed under
the exchange of two simple roots of su(3), i.e., the charge-conjugation c . The representation (4)
therefore corresponds to c = 1,
c : (0) (0), (4) (4).
(6.6)
From this action, one can express the Ishibashi states of su(3)1 in terms of those of su(2)4 as
follows,

(0, 0) = 1 |0 + |4 ,
2

(0, 0); c = 1 |0 |4 ,
2
(6.7)
where we have used the definition (2.32) of twisted Ishibashi states. Conversely, one obtains

1
|0 = (0, 0) + (0, 0); c = (0, 0); + ,
2

1
|4 = (0, 0) (0, 0); c = (0, 0); ,
2
(6.8)
where stands for the irreducible representations of Gc = {1, c }

= Z2 . The remaining two
representations of su(3)1 have the trivial stabilizer S((1, 0)) = S((0, 1)) = {1} and we have only
one basis for each of them

|21 = (1, 0) ,
(6.9)
|22 = (0, 1) .
We have distinguished two bulk fields with representation (2) in (6.3) by putting the subscript.
In this way, the basis (4.1) of twisted Ishibashi states in su(3)1 can be regarded as the (untwisted) Ishibashi states of su(2)4 . This implies that the regular states (4.5) of the generalized
fusion algebra F (su(3)1 ; Gc ) yield a complete set of boundary states in (6.3), whose coefficient
matrix is given by the generalized S-matrix (3.11)

1
1
2 2

2i
1 1
1
2 2
= S =
(6.10)
=e 3 .
2
1
1
6
3 3
0
0
Here the columns are ordered as {((0, 0); +), ((0, 0); ), (1, 0), (0, 1)}. The first three rows correspond to three representations {(0, 0), (1, 0), (0, 1)} of su(3)1 , respectively, while the fourth
(2)
row stands for the representation of the twisted chiral algebra A2 at level 1. The boundary
states of (6.3) that preserve su(2)4 are therefore labeled by L I for the generalized fusion
algebra F (su(3)1 ; Gc ).
The structure constants (6.2) of the graph fusion algebra associated with is defined by
summing over the column indices of . Since = S in the present case, this is nothing but the
generalized Verlinde formula (3.14) if we choose (0, 0) I as the node 0 in (6.2). Namely, the
structure constants of the graph fusion algebra are given by the generalized fusion coefficients of
F(su(3)1 ; Gc ),
GLM N = N LM N .
(6.11)
The graph fusion algebra associated with (6.3) therefore coincides with the generalized fusion
algebra F (su(3)1 ; Gc ). Since the generalized fusion algebra contains the ordinary fusion algebra
as a subalgebra, this identification of the graph fusion algebra with the generalized fusion algebra
364
naturally explains the observation that the graph fusion algebra contains the fusion algebra of the
extended theory as a subalgebra.
This argument for the relation of graph fusion algebras with the generalized fusion algebras of
the extended theory is readily extended to simple current extensions other than su(2)4 su(3)1 .
For instance, the case of su(3)3 so(8)1 in (5.30) can be treated in completely the same way.
First, we give the branching rule for the representations (2.63) of so(8)1 ,
O = (0, 0) (3, 0) (0, 3),
V = (1, 1),
S = (1, 1),
C = (1, 1).
(6.12)
As is shown in Appendix A, the unextended chiral algebra su(3)3 is characterized as the fixed
point algebra of an automorphism group
= Z3 of so(8)1 . Since has order 3, the eigen2i
value of takes the form n ( = e 3 ; n = 0, 1, 2). The decomposition (6.12) of the vacuum
representation O of so(8)1 into those of su(3)3 corresponds to the decomposition according to
the eigenvalue of , since fixes su(3)3 . From the explicit realization of su(3)3 so(8)1 given
in Appendix A, one can show that acts on the vacuum representation O of so(8)1 as follows
0), (0, 3) (0, 3).
: (0, 0) (0, 0), (3, 0) (3,
(6.13)
The twisted Ishibashi states of so(8)1 are then expressed in terms of the (untwisted) Ishibashi
states of su(3)3 ,

1
|O = (0, 0) + (3, 0) + (0, 3) ,
3

1
|O; = (0, 0) + (3, 0) + (0, 3) ,
3

1
O; 2 = (0, 0) + (3, 0) + (0, 3) ,
(6.14)
3
where we have used again the definition (2.32) of twisted Ishibashi states. From this expression,
one obtains

(0, 0) = 1 |O + |O; + O; 2 = (O; ) ,
0
3

1
(3, 0) = |O + |O; + O; 2 = (O; ) ,
1
3

1
(0, 3) = |O + |O;
(6.15)
+ O; 2 = (O; 2 ) ,
3
where n (n = 0, 1, 2) are the irreducible representations of defined as n ( ) = 2n . The
other three representations of so(8)1 have the trivial stabilizer {1} and we can relate the
corresponding Ishibashi states with those of su(3)3 as follows,

(1, 1)1 = |V ,
(1, 1)2 = |S ,
(1, 1)3 = |C .
(6.16)
We have distinguished three bulk fields with representation (1, 1) in (5.30) again by putting the
subscript. The six untwisted Ishibashi states of su(3)3 are thus identified with the six twisted
Ishibashi states of so(8)1 for the automorphism group
= Z3 . Hence the boundary states
of su(3)3 associated with the modular invariant (5.30) are identified with the twisted boundary
states of so(8)1 for
= Z3 .
Since is composed of the triality automorphism and an inner automorphism of so(8)1 ,
(3)
the twisted chiral algebra for is also isomorphic to the twisted affine Lie algebra D4 at level 1
365
and we have only one representation for . The same argument holds for 2 . The generalized
fusion algebra F (so(8)1 ; ) therefore consists of 4 + 1 + 1 = 6 representations. From the
definition (3.11), one obtains its generalized S-matrix in the form
1 1
1
3
3
3
1
3
1 1
1
3
3
1
1
1
S =
(6.17)
.
3
1 1
1 3 3
2 3
2 2 2
0
0
0
2 2 2
0
0
0
Here the columns are ordered as {(O; 0 ), (O; 1 ), (O; 2 ), (V ), (S), (C)}. The first four rows
express respectively the untwisted representations O, V , S and C, while the last two rows cor2

respond to representations in I and I . This matrix gives the boundary state coefficients

for the regular states of F (so(8)1 ; ). Using the correspondence of Ishibashi states given in
(6.15) and (6.16), we can regard this as the boundary state coefficient matrix D (6) for the untwisted boundary states associated with (5.30). The graph fusion algebra corresponding to D (6)
is therefore identified with the generalized fusion algebra F (so(8)1 ; ). Since F (so(8)1 ; )
contains the ordinary fusion algebra F (so(8)1 ), which is the group algebra of Z2 Z2 , the graph
fusion algebra of D (6) also contains it as a subalgebra.
In the modular invariant (5.30), in addition to the untwisted boundary states, we have the
twisted boundary states for the charge-conjugation automorphism group Gc = {1, c } of su(3)3 .
c = , to so(8)1 , which
As is shown in Appendix A, the automorphism group Gc has a lift G
is isomorphic to the symmetric group S3 . Using this fact, we can construct the c -twisted states
c -twisted states of so(8)1 .
in (5.30) starting from the G
c is composed of the chirality flip and an inner automorphism of so(8)1 , the
Since G
twisted chiral algebra for is isomorphic to D4(2) at level 1 and the generalized fusion algebra
c ) has exactly the same form as F (so(8)1 ; S3 ) presented in Section 2.3. ConseF(so(8)1 ; G
c , of which six states for = {1, , 2 }
quently, one obtains 12 twisted boundary states for G
correspond to the untwisted states of su(3)3 while the remaining six for the coset =
{ , , 2 } are regarded as the c -twisted ones, since the restriction of to su(3)3
yields c . For example, two -twisted states of so(8)1 take the form

1
|O; |V ; .
2
(6.18)
The -twisted Ishibashi states of so(8)1 can be expressed in terms of the c -twisted Ishibashi
states of su(3)3 ,

1
|O; = R( )|O = R(c ) (0, 0) + (3, 0) + (0, 3)
3

1
= (0, 0); c + (0, 3); c + (3, 0); c ,
3

|V ; = R( )|V = R(c )(1, 1)1 = (1, 1)1 ; c .
(6.19)
(6.20)
Substituting these expressions into (6.18), we obtain two c -twisted states of su(3)3 ,

1
(0, 0); c + (3, 0); c + (0, 3); c 3(1, 1)1 ; c .
6
(6.21)
366
The case of and 2 can be treated in the same way and we obtain four more c -twisted
states,

1
|O; |C;
2

1
= (0, 0); c + (3, 0); c + (0, 3); c 3(1, 1)3 ; c ,
(6.22)
6

1
O; 2 S; 2
2

1
= (0, 0); c + (3, 0); c + (0, 3); c 3(1, 1)2 ; c .
(6.23)
6
The c -twisted boundary states obtained in this way form a complete set of twisted states, since
the number of the twisted boundary states is the same as the number |E(c )| of the twisted
Ishibashi states available in (5.30). The resulting boundary states have the following coefficient
matrix
0
0
1 1 1
3
0
1 1 1 3 0
1
0
3
0
1
D (6) =
(6.24)
,
1
0
3 0
6
1
0
0
3
1
0
0
3
where the columns are ordered as {(0, 0), (3, 0), (0, 3), (1, 1)1 , (1, 1)2 , (1, 1)3 }. This matrix
D (6) realizes the D (6) NIM-rep of the ordinary fusion algebra F(su(3)3 ). By appropriately
mixing three columns of D (6) corresponding to (1, 1)n (n = 1, 2, 3), one can make the first
row of D (6) consist of only non-vanishing entries. The graph fusion algebra for D (6) is then
constructed by taking the first row as the node 0. It would be interesting to relate this algebra
c ).
with some generalized fusion algebra, possibly F (so(8)1 ; G
7. Summary and discussion
In this paper, we have studied a consistency condition of boundary states satisfying the boundary condition twisted by an automorphism group G of the chiral algebra A and clarified the
relation of twisted boundary states with the generalized fusion algebra. We have shown that the
fusion coefficients of the generalized fusion algebra is expressed by a formula analogous to the
Verlinde formula even for non-Abelian cases and thereby determined irreducible representations
of the generalized fusion algebra, which generalize quantum dimensions of the ordinary fusion
algebras. For a non-Abelian G, some irreducible representations have dimension greater than 1
reflecting the fact that the generalized fusion algebra is non-commutative. Based on these results, we have shown that a consistent set of twisted boundary states forms a NIM-rep of the
generalized fusion algebra. As a check of our argument, we have considered twisted boundary
states in several models, which includes non-diagonal modular invariants as well as the case of
non-Abelian automorphisms. We have seen that several NIM-reps of the ordinary fusion algebra
are organized into a single NIM-rep of the generalized fusion algebra. In particular, the D and
E type NIM-reps of su(3)k (k = 3, 5) are paired with their counterpart D and E , respectively,
to yield a NIM-rep of the generalized fusion algebra for the charge-conjugation automorphism
group of su(3)k . Finally, we have given an argument that the graph fusion algebra associated with
367
a simple current extension can be regarded as the generalized fusion algebra of the extended chiral algebra, which naturally explains the observation that the graph fusion algebra contains the
fusion algebra of the extended theory as a subalgebra.
Having obtained these results, a natural problem is the classification of NIM-reps of the generalized fusion algebra, which generalizes the corresponding problem [2,7] for the case of ordinary
fusion algebras. These two problems are related with each other, since the generalized fusion
algebra contains the ordinary fusion algebra as a subalgebra and a NIM-rep of the former is decomposed into NIM-reps of the latter. More generally, if the automorphism group G of the chiral
algebra A has a subgroup H G, the generalized fusion algebra F (A; G) has a subalgebra
F(A; H ), and a NIM-rep of F (A; G) is decomposed into NIM-reps of F (A; H ). On the other
hand, as we have pointed out in Section 5.1, there exist some unphysical NIM-reps of the ordinary fusion algebra F (u(1)6 ) that are not originated from those of the generalized fusion algebra
F (u(1)6 ; Gc ). This example shows that not all of the NIM-reps for F (A; H ) can be obtained
from those of F(A; G). It is interesting to find other examples with this property and understand
its significance. In particular, any relation with the notion of physical NIM-reps [7] would be
desirable.
The case of block diagonal modular invariants is of particular interest. In that case, the chiral
algebra A has an extension Aext and one can consider a lift of the automorphism group G of A to
Aext . If a lift Gext of G exists,12 one can construct Gext -twisted boundary states of the extended
theory, which can also be regarded as G-twisted states of the unextended theory. This suggests
that there is some relation between two generalized fusion algebras F (A; G) and F (Aext ; Gext ).
Indeed, for the case of simple current extensions, NIM-reps of F (Aext ; Gext ) can be considered
as NIM-reps of F(A; G), as we have seen in Section 6. Probably, this correspondence of NIMreps is a common feature of the cases in which the unextended chiral algebra A is characterized
as the fixed point algebra of a certain automorphism group Hext Gext of Aext . This is exactly
the setting of the Galois theory for vertex operator algebras (see [28] and references therein) and
the mutual relation of generalized fusion algebras in such cases might be described also by the
Galois theory. For exceptional invariants, however, the unextended chiral algebra A cannot be
obtained as the fixed point algebra for any automorphism group of Aext in general, and one would
need another tool, such as the operator-algebraic methods [12,30,31], for a deep understanding
of the mutual relation of generalized fusion algebras associated with an exceptional invariant.
Appendix A. Algebra embedding su(3)3 so(8)1
In this appendix, we give an explicit realization of the algebra embedding su(3)3 so(8)1 and
c
show that the charge-conjugation automorphism group Gc = {1, c }
= Z2 of su(3)3 has a lift G
to so(8)1 , which is isomorphic to the symmetric group S3 .
The affine Lie algebra so(8)1 can be expressed by four pairs of complex fermions i (i =
1, 2, 3, 4). The Cartan element H i and the simple root E i of so(8) have the form
H i = i+ i ,
E 3 = 3+ 4 ,
E 1 = 1+ 2 ,
E 4 = 3+ 4+ .
E 2 = 2+ 3 ,
(A.1)
12 This is actually an assumption. There is in general an obstruction to lifting the automorphism group G of A to the
extended chiral algebra Aext [28].
368
The other roots are written as i j or i j with i = j . The subalgebra su(3) so(8) with
embedding index 3 can be realized as follows,
#
3
1
H1 = (H 1 2H 2 + H 3 ),
H2 =
(H1 + H 3 ),
2
2
E 1 = 1 3+ + 4 2 +
4+ 2 ,
E 2 = 2+ 3+ + 1+ 4+ +
1+ 4 ,
E = E 1 + E 3 + E 4 = 1+ 2 + 3+ 4 + 3+ 4+ ,
(A.2)
where Hi (i = 1, 2) and E i (i = 1, 2) stand for the Cartan elements and the simple roots of su(3),
2i
respectively. E = [E 1 , E 2 ] is the highest root of su(3) and = e 3 . The 28 elements of so(8)
can be decomposed into irreducible representations of this su(3) as 28 = 8 10 10, where
10 (10) is the highest weight representation with the Dynkin label (3, 0) ((0, 3)) and contains
2 3+ (1+ 3+ ) as the highest weight state.
One can construct an automorphism of so(8) that fixes the su(3) subalgebra (A.2) from the
triality of so(8). Let be the triality automorphism of so(8), which generates an automorphism
group = {1, , 2 }
= Z3 . The action of on the simple roots of so(8) reads
: E 1 E 3 E 4 E 1 , E 2 E 2 ,
(A.3)
which in turn implies the following action on the Cartan elements,

1
: H 1 (H 1 + H 2 + H 3 H 4 ),
2
1
H 2 (H 1 + H 2 H 3 + H 4 ),
2
1
H 3 (H 1 H 2 + H 3 + H 4 ),
2
1
H 4 (H 1 H 2 H 3 H 4 ).
(A.4)
2
The action on the other elements of so(8) can be determined by the commutation relations and
one obtains
: 1+ 3 4 2+ 4+ 2+ 1+ 3 , 2+ 3+ 1+ 4+ 1+ 4 2+ 3+ ,
1+ 3+ 1+ 3+ , 1+ 2+ 1+ 2+ .
(A.5)
Here we give only the action on the positive roots; the action on the negative roots can be obtained
by taking the hermitian conjugation of this equation. Since the order of is 3, its eigenvalue is of
the form n (n = 0, 1, 2), and so(8) is decomposed into the eigenspaces of as 28 = 14 7 7.
The fixed point algebra of has dimension 14 and coincides with G2 so(8) with embedding
index 1.
acts on su(3) in (A.2) as follows,
: E 1 E 1 , E 2 E
2 , E E , Hi Hi .
(A.6)
Hence almost fixes su(3). One can construct an automorphism fixing su(3) in the form
= Adh , where Adh is an inner automorphism J Adh (J ) = hJ h1 with the element h =
369
2i
3 (H1 +H2 )
SO(8). From the action of Adh on i ,
Adh : 1 1 1 , 2 1 2 , 3 3 , 4 4 ,
(A.7)
one can readily check that = Adh actually fixes su(3). Since Adh commutes with and
(Adh )3 = 1, the automorphism has also order 3 and hence generates an automorphism group
= {1, , 2 }
= Z3 . The elements of so(8) are again decomposed according to the eigenvalue n (n = 0, 1, 2) of as 28 = 8 10 10, which coincides with the decomposition into
irreducible representations of su(3) mentioned above. In this way, the su(3) subalgebra (A.2) is
characterized as the fixed point algebra of the automorphism group
= Z3 of so(8).
The fact that su(3) in (A.2) is characterized as the fixed point algebra of implies that the
identity automorphism of su(3) has a lift
= Z3 to so(8). One can show that the chargeconjugation automorphism c of su(3) also has a lift to so(8). Consider the following automorphism of so(8),
: 1 2 , 2 1 , 3 3 , 4 4 ,
(A.8)

which has order 2. This is an outer automorphism of so(8), since acts on eight real fermions
as an element of O(8) with determinant 1. This automorphism acts on su(3) in (A.2) as the
exchange of two simple roots
: E 1 E 2 , E E .
(A.9)
In particular, keeps su(3) so(8) invariant and can be restricted to su(3). Clearly, the restriction of to su(3) is the charge-conjugation c of su(3). In other words, is a lift of c to so(8).
c of Gc = {1, c }. By explicit calculations, one can show that
Together with , forms a lift G
c = , is isomorphic to the symmetric group S3 ; the
1 = 1 , which means that G
c to so(8), which is
charge-conjugation automorphism group Gc of su(3) so(8) has a lift G
c can be identified with the quotient group of G
c by
isomorphic to S3 . The restriction Gc of G
c /
the stabilizer of su(3) so(8) as G
= S3 /Z3
= Z2 .
Appendix B. Twisted boundary states for su(3)k
In this appendix, we describe in detail the construction of twisted boundary states for the
chiral algebra su(3)k (k = 3, 5) and the charge-conjugation automorphism group Gc = {1, c },
and check that the resulting boundary states realize a NIM-rep of the generalized fusion algebra
F (su(3)k ; Gc ).
Let us first calculate the explicit form of the generalized fusion algebra F(su(3)k ; Gc ) for
k = 3 and 5. The set I of the irreducible representations of su(3)k reads

I = P+k su(3) = = (1 , 2 ) | 1 + 2 k .
(B.1)
The modular transformation matrix of the characters of these representations is given by the
KacPeterson formula [32]. The charge conjugation c acts on I as
c : (1 , 2 ) (2 , 1 ).
(B.2)
(2)
A2 .
Since its horiThe corresponding twisted chiral algebra is the twisted affine Lie algebra
(2)
zontal subalgebra is so(3), the irreducible representations of A2 are labeled by a single Dynkin
label which we denote by ,

I c = P+k A(2)
(B.3)
= { | 2 k}.
2
370
The set I(c ) of representations fixed by c reads

I(c ) = = (1 , 1 ) | 21 k .
(B.4)
The modular transformation matrix S c of the twisted representation I c takes the form [32]

2
2
sin
( + 1)(1 + 1) .
S c =
(B.5)
k+3
k+3
By using these data, one can construct untwisted and c -twisted boundary states compatible
of boundary states
with the charge conjugation modular invariant of su(3)k . The total number |I|
is

10 + 2 = 12 for k = 3,
c

|I| = |I| + I =
(B.6)
21 + 3 = 24 for k = 5.
As we have shown in Section 2, the generalized fusion algebra F (su(3)k ; Gc ), or equivalently the
regular NIM-reps of F (su(3)k ; Gc ), is obtained from the overlaps of these boundary states. We
give only the part of F (su(3)k ; Gc ) concerning the twisted representations. The part containing
only the untwisted representations, i.e., the ordinary fusion algebra F(su(3)k ), is found in, e.g.,
[33].
k=3
(1 , 2 ) (0) = (1 , 2 ) (1)

(0),
(1 , 2 ) {(0, 0), (3, 0), (0, 3)},
=
(0) + (1), (1 , 2 ) {(1, 0), (0, 1), (2, 0), (0, 2), (2, 1), (1, 2)},
(1, 1) (0) = (0) + 2 (1),
(1, 1) (1) = 2 (0) + (1),
(0) (0) = (1) (1) =
(1 , 2 ),
(1 ,2 )I
(0) (1) = (1, 0) + (0, 1) + (2, 0) + (0, 2) + (2, 1) + (1, 2) + 2 (1, 1).
k=5

= (),
(1 , 2 ) (0, 0), (5, 0), (0, 5) , I c = {0, 1, 2},

(1 , 2 ) ()
(0) = (0) + (1),

(1 , 2 ) (1) = (0) + (1) + (2),
(2) = (1) + (2),

(1 , 2 ) (1, 0), (0, 1), (4, 0), (0, 4), (4, 1), (1, 4) ,
(0) = (0) + (1) + (2),

(1 , 2 ) (1) = (0) + 2(1) + (2),
(2) = (0) + (1) + (2),

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

(1 , 2 ) (1) = 2(0) + 2(1) + 2(2), (1 , 2 ) (1, 1), (3, 1), (1, 3) ,
(2) = (0) + 2(1) + (2),
(B.7)
371
(0) = (0) + 2(1) + 2(2),

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

(0) (0) = (2) (2) =
(1 , 2 ),
(1 ,2 )I
(1) (1) = (0, 0) + (5, 0) + (0, 5) + (1, 0) + (0, 1) + (4, 0) + (0, 4) + (4, 1) + (1, 4)

+ 2 (2, 0) + (0, 2) + (3, 0) + (0, 3) + (3, 2) + (2, 3) + (1, 1)

+ (3, 1) + (1, 3) + 3 (2, 1) + (1, 2) + (2, 2) ,
(0) (1) = (1) (2) = (1, 0) + (0, 1) + (4, 0) + (0, 4) + (4, 1) + (1, 4)
+ (2, 0) + (0, 2) + (3, 0) + (0, 3) + (3, 2) + (2, 3)

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

+ 2 (2, 1) + (1, 2) + (2, 2) .
(B.8)
B.1. su(3)3
The modular invariant
ZD (6) = |(0,0) + (3,0) + (0,3) |2 + 3|(1,1) |2
(B.9)
is a simple current extension of su(3)3 by (3, 0) and the corresponding extended chiral algebra
is so(8)1 . As we have shown in Section 6, the twisted boundary states for the automorphism
c , a lift of Gc to so(8)1 . The resulting boundary state
group Gc are obtained from those for G
coefficients are given in (6.17) and (6.24). The set V of all the twisted boundary states consists
of 12 elements

= |V| + V c = 6 + 6 = 12.
|V|
(B.10)
By calculating the overlaps of these 12 boundary states, we obtain a set of 12 12 matrices
defined in (4.12). These matrices turn out to be non-negative integer valued. The
{n N | N I}
explicit form is as follows,

I O
,
n (0,0) = n (3,0) = n (0,3) =
O I

O
T
T
T
n (1,0) = n (0,2) = n (2,1) = n (0,1) = n (2,0) = n (1,2) =
,
O

(1,1)
O
,
n (1,1) =

O
(1,1)

O 0T
,
n (0) =
0 O

O 1T
,
n (1) =
(B.11)
1 O
where the first six rows and columns stand for the states of D (6) in (6.17) and the last six for
those of D (6) in (6.24). O and I are the 6 6 zero and the unit matrices, respectively, and the
372
other matrices are defined as
0 0 0 0 1
0 0 0 0 1
0 0 0 0 1
=
0 0 0 0 1
0 0 0 0 0
1 1 1 1 0
0 1 1 1
1 0 1 1
1 1 0 1
(1,1) =
1 1 1 0
0 0 0 0
0 0 0 0
1 1 0 0 1
0 0 1 1 1
1 0 1 0 1
0 =
0 1 0 1 1
1 0 0 1 1
0 1 1 0 1
0
0 0 0
0
0 0 0
0
1 1 0
=
,
0
1 1 0
2
0 0 1
0
0 0 1
0 0
1
0 0
2
0 0
0

=
(1,1)
,
0 0
0
3 0
0
0 3
0
1
0 0 1
1
1 1 0
1
0 1 0
1 =
,
1
1 0 1
1
0 1 1
1
1 0 0
0
0
0
0
1
1
2
1
0
0
0
0
1
1
0
0
0
0
0
0
1
2
0
0
1
0
1
0
0
1
1
1
1
1
1
1
1
1
0
,
0
0
0
0 0
0 0
2 0
1 0
0 1
0 2
1
1
1
.
1
1
1
0
0
0
,
0
2
1
(B.12)
One can check that these matrices n form a NIM-rep of F (su(3)3 ; Gc ) defined in (B.7). For
example, n satisfies the following relation
n (1,1) n (0) = n (0) + 2n (1) ,
n (1,1) n (1) = 2n (0) + n (1) ,
n (0) n (1) = n (1,0) + n (0,1) + n (2,0) + n (0,2) + n (2,1) + n (1,2) + 2n (1,1) ,
etc.
(B.13)
In this way, two NIM-reps, D (6) and D (6) , associated with the simple current modular invariant
(B.9) are combined to yield a NIM-rep of the generalized fusion algebra F (su(3)3 ; Gc ).
B.2. su(3)5
We next turn to the construction of twisted boundary states associated with a non-trivial modular invariant of su(3)5 and check that the resulting boundary states realize a NIM-rep of the
generalized fusion algebra F(su(3)5 ; Gc ).
B.2.1. The case of ZE (8)
We first consider the exceptional modular invariant of su(3)5 ,
ZE (8) = |(0,0) + (2,2) |2 + |(0,2) + (3,2) |2 + |(1,2) + (5,0) |2
+ |(3,0) + (0,3) |2 + |(2,1) + (0,5) |2 + |(2,0) + (2,3) |2 .
(B.14)
This invariant is originated from the charge conjugation modular invariant of su(6)1 through the
conformal embedding su(3)5 su(6)1 , in which the representations of su(6)1 branch to those of
su(3)5 as follows,
0 (0, 0) (2, 2),
1 (0, 2) (3, 2),
2 (1, 2) (5, 0),
373
3 (3, 0) (0, 3),

4 (2, 1) (0, 5),
5 (2, 0) (2, 3).
(B.15)
available,13
In the modular invariant (B.14), there are twelve untwisted Ishibashi states

E = (0, 0), (5, 0), (0, 5), (2, 2), (1, 2), (2, 1), (3, 0), (2, 3), (0, 2), (0, 3), (2, 0), (3, 2) .
(B.16)
Correspondingly, we have twelve untwisted boundary states. Six of them are identified with
the untwisted states of su(6)1 and the other six states are generated by using the fusion with
the representations of su(3)5 . The resulting boundary state coefficient matrix E (8) takes the
following form
a1 K a2 K
K
K
1 a K a2 K
K K
E (8) = 1
(B.17)
,
2 a2 K a1 K iK iK
a2 K a1 K iK iK

where a1 = 2 sin 8 = 22 2 , a2 = 2 cos 8 = 2+2 2 and the matrix K is defined as

1
K=
3
'
1
1
1 e2i/3
1 e2i/3
e2i/3
(B.18)
e2i/3
Here the columns of E (8) are ordered as in (B.16). The first six rows correspond to the states
of the su(6)1 theory. One can check that these untwisted states yield the E (8) NIM-rep of the
ordinary fusion algebra F(su(3)5 ).
The labels of the available c -twisted Ishibashi states are obtained from (B.14),

E(c ) = (0, 0), (2, 2), (0, 3), (3, 0) .
(B.19)
As is seen from the branching rule (B.15), the charge conjugation c can be lifted to the charge
conjugation c of su(6)1 , which acts on the fundamental weights of su(6)1 as follows,
c : 0 0 , i 6i
(i = 1, 2, 3, 4, 5).
(B.20)
The c -twisted boundary states in the su(6)1 theory take the form

1
| = |0 ; c |3 ; c .
2
(B.21)
One can express these states in terms of the -twisted Ishibashi states of su(3)5 in the same way
as the D (6) states considered in Section 6. First, from the branching rule (B.15) and the modular
transformation matrix of su(3)5 , the Ishibashi state for the vacuum representation of su(6)1 can
be decomposed as follows,

1
|0 = a1 (0, 0) + a2 (2, 2) ,
2
13 For simplicity, we omit in this appendix the label for the anti-holomorphic representation of Ishibashi states.
(B.22)
374
where a1 , a2 are the constants used in (B.17). The c -twisted Ishibashi state |0 ; c is then
obtained by the action of c ,

1
|0 ; c = R( c )|0 = a1 (0, 0); c a2 (2, 2); c .
2
(B.23)
Here we use the fact that the c acts on (2, 2) as 1, which can be shown, for example, by
1
the calculation of 0 |q 2 Hc |+ . In this way, we obtain two c -twisted states. The remaining two
states are generated by the fusion of su(3)5 and we obtain four c -twisted states, whose boundary
state coefficient matrix takes the form
E (8)
a1
1 a1
=
2 a2
a2
a2
a2
a1
a1
1
1
i
i
1
1
.
i
i
(B.24)
Here the columns are ordered as in (B.19) and the first (second) row corresponds to |+ (| ).
These states form the E (8) NIM-rep of F (su(3)5 ).
The number of all the twisted boundary states is thus

= |V| + V c = 12 + 4 = 16.
|V|
(B.25)
From the explicit form of the boundary state coefficients, one can calculate the corresponding
overlap matrices n,
which are 16-dimensional. The result is as follows,
n (0,0) = I1616 ,
O O
O O
O
n (1,0) =
O
O I
I +
O
t
O
n (2,0) =
t
t
O
t
O
t
O
t
2 t
O
t
2 t
t
O
O
I
n (1,1) =
I
O
O
I
I
O

I

I + 2
I
I
2I
2I
I
I
2I
2I
O
I +
I + 2(I + )
O
I
I
n (3,0) =
O
I
O
O
I
O
I
I
2I
I
O
2I
I

I I + 2
n (2,1) =
3
2
O
O
n (4,0) =
O
O
O
O
O
O
t
n (3,1) =
t
O
O
t
t
O
I I +
t
t
2 t
2 t
I
O
I
n (2,2) =
I
O
I
I
I
I
I
2I
3I
t
O
O
n (5,0) =
O
O
t
O
O
O
O
t
O
O
O
O
t
I
O
I
I
O
I
I
I
O
O
O
n (4,1) =
I
O
O
I
O
O
O
I +

+ 2(I + )
I
I
I
3I
2I
t
t
2 t
2 t
I
I +

+ 3I + 2
I
O
2
3
I
I +

+ 3I + 2

I O
O I
O I
I +
375
376
O
n (3,2) =
O

O
n (0) = n (2) =
A
At
O

I

I + 2
n (1) =
O
B
Bt
O

.
(B.26)
The matrices not presented above are obtained by n (2 ,1 ) = n T(1 ,2 ) . The first 12 rows and
columns of these matrices n are ordered as in the rows of E (8) in (B.17) and the last four
are ordered as in E (8) defined in (B.24). O and O (I and I ) are respectively 3 3 and 2 2
zero (unit) matrices while the other matrices are defined as
'
(
0 1 0
= 0 0 1 ,
1 0 0
06 16
0 16
A= 6
,
16 26
16 26
0
1
16
16
B =
16
16

1
,
0
16
16
,
36
36
(B.27)
where m6 (m = 0, 1) is a row vector with all entries being m, m6 = (m, m, m, m, m, m). One
can explicitly check that these non-negative integer valued matrices {n N } satisfy the generalized
fusion algebra F (su(3)5 ; Gc ) in (B.8), for example,
n (2,2) n (0) = n (0) + 2n (1) + 2n (2) ,
n (2,2) n (1) = 2n (0) + 3n (1) + 2n (2) ,
n (2,2) n (2) = 2n (0) + 2n (1) + n (2) ,
n (0) n (2) = n (2,0) + n (0,2) + n (3,0) + n (0,3) + n (3,2) + n (2,3) + n (1,1) + n (3,1) + n (1,3)
+ 2(n (2,1) + n (1,2) + n (2,2) ),
etc.
(B.28)
B.2.2. The case of ZD (8)

We next consider twisted boundary states associated with the simple current invariant
ZD (8) = |(0,0) |2 + |(1,1) |2 + |(2,2) |2 + |(3,0) |2 + |(0,3) |2 + |(4,1) |2 + |(1,4) |2
+ ((5,0) (0,5) + (3,1) (1,3) + (1,2) (2,1)
+ (2,3) (0,2) + (2,0) (3,2) + (0,4) (1,0) + (0,1) (4,0) + c.c.).
(B.29)
Since this invariant is the Z3 -orbifold of su(3)5 , one can obtain the boundary states for (B.29) by
averaging the Z3 -orbit of boundary states for the charge-conjugation invariant. For the untwisted
377
states, this procedure yields the following coefficient matrix,
1 1
2 1 + 1
1
1
2
2
1
1
2
i
i
2
2
1
0
1
1
+
i
1
i
1

D (8) =
2
0
2
0
0
2 2
1
1 i 1 + i
1
0
1
1
i
i
2
2
1
1
1+ 2 1
1
1
2
1
2
1
2
1
2
2
i
i 1
where the columns are ordered as

E = (0, 0), (1, 1), (2, 2), (3, 0), (0, 3), (4, 1), (1, 4) .
i 1
2
2 ,
1
i
2
1
(B.30)
(B.31)
Since all of three c -twisted states for the charge-conjugation invariant are the fixed point of the
Z3 action, we obtain 3 3 = 9 twisted states for (B.29) by an appropriate resolution of the fixed
points. The resulting coefficient matrix takes the following form
(
(
'
'
1
2
1
1
1
2i/3
2i/3
D (8) =
with =
e
e
2
0
2 , (B.32)
2
3 e2i/3 e2i/3
1 2
1
where the columns are ordered as

E(c ) = (0, 0), (1, 1), (2, 2), (5, 0), (3, 1), (1, 2), (0, 5), (1, 3), (2, 1) .
(B.33)
In this way, we obtain

= |V| + V c = 7 + 9 = 16
|V|
(B.34)
boundary states associated with the modular invariant (B.29). By calculating the overlap of these
states, we obtain a set of 16 16 matrices
n (0,0) = I1616 ,

AD (8) O
n (1,0) =
,
O AD (8)
(B.35)
where AD (8) and AD (8) are the adjacency matrix for the graphs of type D (8) and D (8) , respectively,
0 1 0 0 0 0 0
0 0 1 0 0 1 0
(
'
O O
0 0 0 1 1 0 0
AD (8) = 0 1 0 0 1 0 1 ,
AD ( 8) = O O ,
O O
0 0 0 0 0 1 1
1 0 0 1 0 0 0
0 0 1 1 0 0 1
'
(
1 1 0
= 1 1 1 .
(B.36)
0 1 1
378
The remaining matrices for the untwisted representations take the following form,
O
A(2,0)
O (2,0) O
,
n (2,0) =
O
O
O (2,0)
n (1,1) =
n (3,0) =
n (2,1) =
n (4,0) =
(2,0)
A(1,1)
O
At(2,0)
O
O
O
O
O
AtD (8)
O
(1,1)
O
,
O (1,1) O
O
O (1,1)
O
(2,0) O
O
,
O
O
(2,0)
O
A(2,2)
O
O
O
O
(2,0)
O
(2,2)
,
(2,2)
O
O (2,2) O
O
O O
,
O O
O O
A(1,1)
(1,1) O
,
O (1,1)
O
(1,1) O
O
O
A(2,2)
(2,2) O
O
,
n (2,2) =
O
O (2,2) O
O
O (2,2)
O
I77
O 133 O
,
n (5,0) =
O O
O 133
133 O
O
O
AD (8)
O
O
n (4,1) =
O O O ,
n (3,1) =
n (3,2) =
O
O
O O
A(2,0)
O
O
O
O
O
(2,0)
O (2,0)
(2,0)
,
O
O
(B.37)
379
where
0
0
A(2,0) = 0
0
0
0
0
A(2,2) = 0
0
1
'
(2,0) =
1
1
1
0
0
0
0
0
1
1
0
0
1
1
0
0
1
1
0
0
1
0
0
1
0
1
1
1
1
0
1
0
1
0
1
1
0
1
0
1
1
1
1
1
2
1
2
1
(
1
1 ,
1
0
1
1
0
0
0
1
0
0
1
1
1
1
1
0
0
1
1
0
0
0
0
0
0
1
1
0
1
0
0
1,
1
1
1
1
2,
1
2
'
(1,1) =
0
0
A(1,1) = 1
0
0
0
1
0
0
1
0
1
0
0
1
1
0
1
1
1
0
1
2
1
0
1
0
1
0
1
1
0
1
0
0
1
0
0
1
1
0
1
1,
1
2
(B.38)
(
1 2 1
2 2 2 ,
1 2 1
'
(2,2) =
(
1 2 2
2 3 2 .
2 2 1
(B.39)
For the twisted representations, we obtain
O B0t B0t B0t
0
,
n (0) =
B
O
0
B0
O B2t B2t B2t

B
2
,
n (2) =
B
O
2
B2
O B1t B1t B1t
B
1
n (1) =
B
1
B1
(B.40)
where
'
1
B0 = 0
0
'
0
B2 = 0
1
1
1
0
0
1
1
1
1
1
1
1
1
1
2
1
1
2
1
1
1
1
1
1
1
1
1
0
0
1
1
(
1
2 ,
2
(
2
2 .
1
'
B1 =
(
0 1 1 2 1 1 2
1 1 2 2 2 1 3 ,
0 1 1 2 1 1 2
(B.41)
We can check that these matrices form a NIM-rep of the generalized fusion algebra F (su(3)5 ;
Gc ) given in (B.8).
In summary, we have checked that the twisted boundary states of su(3)5 realize a NIM-rep of
the generalized fusion algebra F(su(3)5 ; Gc ). In other words, we have constructed three NIMreps of F (su(3)5 ; Gc ). Besides the regular one, which has dimension 24, we have obtained a
12-dimensional one associated to the exceptional modular invariant ZE (8) and a 16-dimensional
one associated to the simple current invariant ZD (8) .
380
Appendix C. Twisted boundary states for su(3)3

1
In this appendix, we show that the S3 -twisted boundary states associated with the modular
3
invariant (5.33) of su(3)3
1 yield a NIM-rep of the generalized fusion algebra F (su(3)1 ; S3 ).
C.1. Generalized fusion algebra F (su(3)3
1 ; S3 )
For su(3)1 , there are three integrable representations corresponding to three fundamental
weights 0 , 1 and 2 . In the following, we denote them as (0) = (0 ), (1) = (1 ) and
(2) = (2 ). In this notation, the modular transformation matrix of su(3)1 can be written as
1 2i
su(3)1
= e 3 mn
Smn
3
(m, n = 0, 1, 2).
(C.1)
Since this expression is invariant under the shift m m + 3 (and also n n + 3), we
can consider that the representations of su(3)1 are labeled by an integer modulo 3, namely,
(m + 3) = (m). Actually, the fusion algebra of su(3)1 is the group algebra of Z3 ,
(m) (n) = (m + n).
(C.2)
In this notation, the action of the charge-conjugation is expressed as

(m) = (m).
(C.3)
The representations of su(3)3

1 are then labeled by three integers (n1 , n2 , n3 ) and the set I of
integrable representations reads

I = (n1 , n2 , n3 ) | ni = 0, 1, 2 .
(C.4)
Hence there are 33 = 27 representations for su(3)3
1 . The modular transformation matrix S is
simply the tensor product of three S su(3)1 ,
1 2i
su(3)1 su(3)1 su(3)1
Sm2 n2 Sm3 n3 = e 3 (m1 n1 +m2 n2 +m3 n3 ) .
S(m1 ,m2 ,m3 )(n1 ,n2 ,n3 ) = Sm
1 n1
3 3
(C.5)
The fusion algebra of su(3)3

1 is the group algebra of Z3 Z3 Z3 ,
(m1 , m2 , m3 ) (n1 , n2 , n3 ) = (m1 + n1 , m2 + n2 , m3 + n3 ).
(C.6)
The automorphism group S3 is generated by two elements and , which act on the elements
of I as
: (n1 , n2 , n3 ) (n3 , n1 , n2 ),
: (n1 , n2 , n3 ) (n1 , n3 , n2 ).
(C.7)
In terms of and , the elements of S3 can be expressed as follows,

S3 = 1, , 2 , , , 2 .
(C.8)
The fixed point of S3 is of the form (n, n, n),

I() = (n, n, n) | n = 0, 1, 2 .
(C.9)
381
Therefore we have three twining characters for , which can be expressed by those of su(3)1 as
follows

c
(n,n,n)
(C.10)
(q)
= Tr(n,n,n) q L0 24 = nsu(3)1 q 3 .
As is shown in Eq. (2.45), the -twisted representations are obtained by the modular transformation of the twining characters ,

1

su(3)1 su(3)1 3
su(3)1
q =
Smn
n
Smn
(n,n,n) (q).
msu(3)1 q 3 =
(C.11)
n=0,1,2
n=0,1,2
Since the left-hand side is labeled by an integer m (mod 3), we denote by (m) the -twisted
representation of su(3)3
1 ,

I = (m) | m = 0, 1, 2 .
(C.12)
From the above equation, the characters and the modular transformation matrix for (m) read
1
su(3)1
S(m)
= Smn
.
(m) (q) = msu(3)1 q 3 ,
(C.13)
(n,n,n)
For S3 , the fixed points and their twining characters have the form

I( ) = (n1 , n2 , n2 ) | n1 , n2 = 0, 1, 2 ,
2
c
1 (q)
1 q
.
(q)
= Tr(n1 ,n2 ,n2 ) q L0 24 = nsu(3)
nsu(3)
(n
1
2
1 ,n2 ,n2 )
(C.15)
We can therefore label the -twisted representations by a pair of integers,

I = (m1 , m2 ) | m1 , m2 = 0, 1, 2 ,
(C.16)
for which the characters and the modular transformation matrix read
1
su(3)1 su(3)1
1 (q) su(3)1 q 2 ,
S(m
= Sm
Sm2 n2 .
(m1 ,m2 ) (q) = msu(3)
m2
1
1 n1
1 ,m2 ) (n1 ,n2 ,n2 )
(C.17)
The remaining cases are treated in the same way and we give only the results below,

2
2
su(3)1
S(m)
= Smn
,
I = (m) 2 | m = 0, 1, 2 ,
2 (n,n,n)
(C.18)
(C.14)

I = (m1 , m2 ) | m1 , m2 = 0, 1, 2 ,
su(3)1 su(3)1
= Sm
Sm2 n2 ,
S(m
1 n1
1 ,m2 ) (n1 ,n1 ,n2 )

2
I = (m1 , m2 ) 2 | m1 , m2 = 0, 1, 2 ,
2

S(m
1 ,m2 )
2 (n1 ,n2 ,n1 )
su(3)1 su(3)1
= Sm
Sm2 n2 .
1 n1
The set I of all the representations eventually consists of 60 elements,

= |I| + I + I 2 + I + I + I 2 = 33 + 3 2 + 32 3 = 60.
|I|
(C.19)
(C.20)
(C.21)
From the formula (2.59), the conjugation acts on the twisted representations as follows,
(m1 , m2 , m3 ) = (m1 , m2 , m3 ),
(m) = (m) 2 ,
(m1 , m2 ) = (m1 , m2 )

, , 2 .
(C.22)
382
Table 4
Multiplication table of the generalized fusion algebra F(su(3)3
1 ; S3 ). The subscripts stand for the automorphism type
of twisted representations. li , mi and ni take values in Z3 = {0, 1, 2}. Since = 2 = , this algebra is noncommutative
(m1 , m2 , m3 )
(m)
(m) 2
(l1 , l2 , l3 )
(l)
(l) 2
(l1 , l2 )
(l1 , l2 )
(l1 , l2 ) 2
(l1 + m1 , l2 + m2 , l3 + m3 )
(l + m1 + m2 + m3 )
(l + m1 + m2 + m3 ) 2
(l1 + m1 , l2 + m2 + m3 )
(l1 + m1 + m2 , l2 + m3 )
(l1 + m1 + m3 , l2 + m2 ) 2
(l1 + l2 + l3 + m)
3(l + m) 2
)
(n , n , n )
)n1 +n2 +n3 =l+m 1 2 3
(n , n )
)n1 +n2 =l1 +l2 +m 1 2 2
(n1 , n2 )
n
+n
=l
+l
+m
) 1 2 1 2
n1 +n2 =l1 +l2 +m (n1 , n2 )
(l1 + l2 + l3 + m) 2
)
n1 +n2 +n3 =l+m (n1 , n2 , n3 )
3(l + m)
)
(n , n )
)n1 +n2 =l1 +l2 +m 1 2
(n1 , n2 ) 2
n
+n
=l
+l
+m
) 1 2 1 2
n1 +n2 =l1 +l2 +m (n1 , n2 )
(m1 , m2 )
(m1 , m2 )
(m1 , m2 ) 2
(l1 , l2 , l3 )
(l)
(l) 2
(l1 , l2 )
(l1 , l2 )
(l1 , l2 ) 2
(l1 + m1 , l2 + l3 + m2 )
)
(n , n )
)n1 +n2 =l+m1 +m2 1 2
(n , n )
)n1 +n2 =l+m1 +m2 1 2 2
(l
+
m1 , n2 , n3 )
1
n2 +n3 =l2 +m2
(l1 + l2 + m1 + m2 )
(l1 + l2 + m1 + m2 ) 2
(l1 + l2 + m1 , l3 + m2 )
)
(n , n )
)n1 +n2 =l+m1 +m2 1 2 2
n1 +n2 =l+m1 +m2 (n1 , n2 )
(l + l2 + m1 + m2 ) 2
)1
n1 +n2 =l1 +m1 (n1 , n2 , l2 + m2 )
(l1 + l2 + m1 + m2 )
(l1 + l3 + m1 , l2 + m2 ) 2
)
(n , n )
)n1 +n2 =l+m1 +m2 1 2
n1 +n2 =l+m1 +m2 (n1 , n2 )
(l1 + l2 + m1 + m2 )
(l + l2 + m1 + m2 ) 2
)1
n1 +n3 =l1 +m1 (n1 , l2 + m2 , n3 )
Having obtained twisted representations and their modular transformation matrices, it is

straightforward to calculate the generalized fusion coefficients of F (su(3)3
1 ; S3 ) using the for(n1 ,n2 ) 2
can be obtained as follows,

mula (2.62). For example, the coefficient N(l1 ,l2 ) (m)
N (l1 ,l2 ) (m)
(n1 ,n2 ) 2
S(l1 ,l2 ) S(m)

S
(n1 ,n2 )
I ( )I ()
S(0,0,0)
2
S
S(l1 ,l2 ) (p,p,p) S(m)
(p,p,p) (n1 ,n2 )
p=0,1,2
(p,p,p)
1 su(3)1 su(3)1 su(3)1 su(3)1

Slsu(3)
Sl2 p Smp Sn1 p Sn2 p
1p
p=0,1,2
S(0,0,0)(p,p,p)
su(3)1 3
)
(S0p
1 2i p(l1 +l2 +mn1 n2 )

e 3
3
p=0,1,2
(3)
= l1 +l2 +m,n1 +n2 ,
(C.23)
where (3) is the Kronecker delta for Z3 . The other cases can be calculated in the same manner;
we give the result in Table 4.
C.2. Non-trivial NIM-rep of F (su(3)3
1 ; S3 )
For the chiral algebra su(3)3
1 , the following block diagonal invariant is available,
ZE6 = |(0,0,0) + (1,1,1) + (2,2,2) |2
+ |(0,2,1) + (1,0,2) + (2,1,0) |2 + |(0,1,2) + (2,0,1) + (1,2,0) |2 ,
(C.24)
383
Fig. 3. Extended Dynkin diagram of E6 . stands for the highest root of E6 . Each box expresses the simple roots of
su(3) E6 .
14
which originates from the conformal embedding su(3)3
1 E6,1 (see Fig. 3). For E6,1 , there
are three integrable representations, ( 0 ), ( 1 ) and ( 5 ),15 and the charge-conjugation invariant
reads
Z = | 0 |2 + | 1 |2 + | 5 |2 .
(C.25)
From this together with the branching rule

( 0 ) = (0, 0, 0) (1, 1, 1) (2, 2, 2),
( 1 ) = (0, 2, 1) (1, 0, 2) (2, 1, 0),
( 5 ) = (0, 1, 2) (1, 2, 0) (2, 0, 1),
(C.26)
one obtains the invariant (C.24).

There are three boundary states preserving E6,1 corresponding to three integrable representations,16
E
| i =
S 6 | j (i = 0, 1, 5).
(C.27)
j =0,1,5
i j
S E6 is the modular transformation matrix of E6,1
1
1
1
1
2i
2i
S E6 = 1 e 3
e 3 ,
2i
2i
3
3
1 e
e 3
(C.28)
where the rows and the columns are ordered as {( 0 ), ( 1 ), ( 5 )}. From the branching rule
(C.26), one can express the Ishibashi states of E6,1 in terms of those of su(3)3
1 ,

1
| 0 = (0, 0, 0) + (1, 1, 1) + (2, 2, 2) ,
3

1
| 1 = (0, 2, 1) + (1, 0, 2) + (2, 1, 0) ,
3
14 This invariant is also considered to be a simple current extension by (1, 1, 1) I.
15 We distinguish the fundamental weights of E
6,1 from those of su(3)1 by putting a tilde.
16 For simplicity, we omit in this appendix the label for the anti-holomorphic representation of Ishibashi states and
denote by |; instead of |(, ); .
384

1
| 5 = (0, 1, 2) + (1, 2, 0) + (2, 0, 1) .
3
(C.29)
Substituting this into (C.27), one obtains three boundary states preserving su(3)3
1 . The remaining 6 (= 9 3) states can be constructed by the fusion with (1, 0, 0) I. Since (1, 0, 0)3 =
(3, 0, 0) = (0, 0, 0), we obtain three boundary states for each state preserving E6,1 . We therefore
label the resulting states as follows

V = (0, 0) = ( 0 ), (0, 1), (0, 2),

(1, 0) = ( 1 ), (1, 1), (1, 2), (2, 0) = ( 5 ), (2, 1), (2, 2) .
(C.30)
The boundary state coefficient takes the form
K
K
K
1
1
1
2i
2i
= K e 3 K e 3 K = 1
2i
2i
3
3
K e 3 K e 3 K
1
1
2i
e 3
2i
e 3
2i
e 3 K,
2i
e 3
where K is a 3 3 unitary matrix
1
1
1
1
2i
2i
K=
1 e 3
e 3 .
2i
3
2i
1 e 3
e 3
(C.31)
(C.32)
The row of is ordered as (C.30) while the column is ordered as

E = (0, 0, 0), (1, 1, 1), (2, 2, 2), (0, 2, 1), (1, 0, 2), (2, 1, 0), (0, 1, 2),

(1, 2, 0), (2, 0, 1) .
(C.33)
The overlap matrices n can be calculated using the formula (4.13), and the result is

(n (m1 ,m2 ,m3 ) )(,a) (,b) = P m2 +2m3 P m1 +m2 +m3 a b ,
(C.34)
where (, a), (, b) V and P is a 3 3 permutation matrix

'
(
0 1 0
P= 0 0 1 .
1 0 0
(C.35)
These matrices {n (m1 ,m2 ,m3 ) | (m1 , m2 , m3 ) I} satisfy the ordinary fusion algebra F(su(3)3
1 )
n (l1 ,l2 ,l3 ) n (m1 ,m2 ,m3 ) = P l2 +2l3 +m2 +2m3 P l1 +l2 +l3 +m1 +m2 +m3
= n (l1 +m1 ,l2 +m2 ,l3 +m3 ) ,
(C.36)
which means that the untwisted boundary states (C.31) form a NIM-rep of F (su(3)3
1 ).
We turn to the construction of twisted boundary states in the invariant (C.24). The automorphism group S3 of su(3)3
1 has a lift to E6,1 (see Fig. 4). The automorphism group Aut(E6 ) of E6
has a normal subgroup Aut0 (E6 ) consisting of all the inner automorphisms. The quotient group
Aut(E6 )/Aut0 (E6 ) has two elements: one is the identity and corresponds to Aut0 (E6 ) while the
other comes from the outer automorphisms that contains the charge conjugation c of E6 ,
c : ( 0 ) ( 0 ), ( 1 ) ( 5 ).
(C.37)
385
Fig. 4. The automorphism group S3 of su(3)3 has a lift to E6 .
As is seen from the definition (C.7) of S3 and the branching rule (C.26), does not change
any representation of E6 . Hence the lift of to E6 is an inner automorphism of E6 . On the other
hand, S3 exchanges ( 1 ) with ( 5 ) and its lift is an outer automorphism of E6 .
Let be the lift of to E6 . Since is inner, the corresponding twisted boundary states of E6
are expressed by the same boundary state coefficient as the untwisted ones, namely the modular
transformation matrix (C.28). Therefore, applying the fusion (1, 0, 0) I to the -twisted states,
we obtain exactly the same boundary state coefficient for the -twisted states as the untwisted
ones,
= .
Accordingly the labels of the -twisted boundary states has the same structure as V,

E() = E.
V = (, a) | = 0, 1, 2; a = 0, 1, 2 ,
(C.38)
(C.39)
We can construct the 2 -twisted states in the same way as the case of and obtain the result

2
2
= ,
(C.40)
V = (, a) 2 | = 0, 1, 2; a = 0, 1, 2 ,
E 2 = E.
The lift of S3 to E6 is outer. Hence we have to start from a non-trivial boundary state
coefficient in E6 instead of the modular transformation matrix. Since fixes only ( 0 ) among
the integrable representations of E6,1 , there is only one -twisted boundary state17

(0) = | 0 ; = 1 (0, 0, 0); + (1, 1, 1); + (2, 2, 2); .
3
(C.41)
Applying the fusion with (1, 0, 0) I yields the remaining two states. The result is as follows,

V = (a) | a = 0, 1, 2 ,
= K,

E( ) = (0, 0, 0), (1, 1, 1), (2, 2, 2) ,
(C.42)
where K is the matrix defined in (C.32). The case of and 2 can be treated in the same way
as and yields the result

= K,
(C.43)
V = (a) | a = 0, 1, 2 ,
E( ) = E( ),
17 One can regard the boundary state coefficient of |(0) as the modular transformation matrix of the twisted chiral
(2)
algebra E6 at level 1.
386
= K,

= (a) 2 | a = 0, 1, 2 ,

E 2 = E( ).
In this way, we obtain 36 boundary states in the block diagonal invariant (C.24),

= |V| + V + V 2 + V + V + V 2 = 9 3 + 3 3 = 36.
|V|
(C.44)
(C.45)
From the boundary state coefficients obtained above, we can calculate the overlap matrices n
by the formula (4.13). In expressing n,
which is a 36 36 matrix, it is convenient to factorize V
in the manner similar to (C.34),

V = (0), (1), (2), (0) , (1) , (2) , (0) 2 , (1) 2 , (2) 2 , (0) , (0) , (0) 2

(a) | a = 0, 1, 2 ,
(C.46)
which is related to the original notation as, e.g., () (a) = (, a) V , (0) (a) = (a)
V . In this basis, one can show that the matrices n take the following form,
m2 +2m3
P
O
O
O
m
+2m
3
O
O
O
P 2
m +m +m
n (m1 ,m2 ,m3 ) =
P 1 2 3,
O
O
P m2 +2m3 O
O
O
O
I
O T O O
O O T O
O
O O T
T O O O
m
m
n (m) 2 =
n (m) =
P ,
P ,
T O O O
O T O O
O O O 3P 2
O O O 3P
O
O
O E1
O
O E2
O
m +m
n (m1 ,m2 ) =
P 1 2,
O
O
O E3
E1T E2T E3T O
O
O
O E2
O
O E3
O
m +m
n (m1 ,m2 ) =
P 1 2,
O
O
O E1
E2T E3T E1T O
O
O
O E3
O
O E1
O
m +m
n (m1 ,m2 ) 2 =
(C.47)
P 1 2,
O
O
O E2
E3T E1T E2T O
where P is a permutation matrix of (C.35) while O and I are the zero and the unit matrices,
respectively. The matrices T , E1 , E2 and E3 are defined as follows,
'
(
'
(
1 1 1
1 0 0
T= 1 1 1 ,
E1 = 1 0 0 ,
1 1 1
1 0 0
(
(
'
'
0 1 0
0 0 1
E3 = 0 0 1 .
E2 = 0 1 0 ,
(C.48)
0 1 0
0 0 1
satisfy the generalized fusion algebra
We have checked that these 60 matrices {n N | N I}
3
F(su(3)1 ; S3 ). For example, (0, 0) (0) = (0, 0) 2 + (1, 2) 2 + (2, 1) 2 is satisfied as
follows
n (0,0) n (0)
O T
O
O
O E1
O
O E2 O O
O
=
T O
O
O
O E3
T
T
T
E1 E2 E3 O
O O
O
O
O
3E3
O
O
3E1
O
=
I
O
O
O
3E2
T
T
T
3E3 3E1 3E2
O
= n (0,0) 2 + n (1,2) 2 + n (2,1) 2 .
O
T
O
O
387
O
O
I
O
3P 2
(C.49)
The matrices {n N } therefore form a 36-dimensional NIM-rep of F(su(3)3

1 ; S3 ). Together
with the regular NIM-rep, which is 60-dimensional, we have obtained two NIM-reps of
3
F(su(3)3
1 ; S3 ) corresponding to two modular invariants of su(3)1 .
References
[1] J.L. Cardy, Boundary conditions, fusion rules and the Verlinde formula, Nucl. Phys. B 324 (1989) 581.
[2] R.E. Behrend, P.A. Pearce, V.B. Petkova, J.B. Zuber, Boundary conditions in rational conformal field theories, Nucl.
Phys. B 570 (2000) 525;
R.E. Behrend, P.A. Pearce, V.B. Petkova, J.B. Zuber, Nucl. Phys. B 579 (2000) 707, hep-th/9908036.
[3] G. Pradisi, A. Sagnotti, Y.S. Stanev, Completeness conditions for boundary operators in 2D conformal field theory,
Phys. Lett. B 381 (1996) 97, hep-th/9603097.
[4] J.L. Cardy, D.C. Lewellen, Bulk and boundary operators in conformal field theory, Phys. Lett. B 259 (1991) 274.
[5] D.C. Lewellen, Sewing constraints for conformal field theories on surfaces with boundaries, Nucl. Phys. B 372
(1992) 654.
[6] I. Runkel, Boundary structure constants for the A-series Virasoro minimal models, Nucl. Phys. B 549 (1999) 563,
hep-th/9811178;
I. Runkel, Structure constants for the D-series Virasoro minimal models, Nucl. Phys. B 579 (2000) 561, hepth/9908046.
[7] T. Gannon, Boundary conformal field theory and fusion ring representations, Nucl. Phys. B 627 (2002) 506, hepth/0106105.
[8] J. Fuchs, C. Schweigert, Orbifold analysis of broken bulk symmetries, Phys. Lett. B 447 (1999) 266, hepth/9811211;
J. Fuchs, C. Schweigert, Symmetry breaking boundaries, I: General theory, Nucl. Phys. B 558 (1999) 419, hepth/9902132;
J. Fuchs, C. Schweigert, Symmetry breaking boundaries, II: More structures, examples, Nucl. Phys. B 568 (2000)
543, hep-th/9908025.
[9] L. Birke, J. Fuchs, C. Schweigert, Symmetry breaking boundary conditions and WZW orbifolds, Adv. Theor. Math.
Phys. 3 (1999) 671, hep-th/9905038.
[10] M.R. Gaberdiel, T. Gannon, Boundary states for WZW models, Nucl. Phys. B 639 (2002) 471, hep-th/0202067.
[11] A. Recknagel, Permutation branes, JHEP 0304 (2003) 041, hep-th/0208119.
[12] J. Fuchs, C. Schweigert, Solitonic sectors, alpha-induction and symmetry breaking boundaries, Phys. Lett. B 490
(2000) 163, hep-th/0006181.
[13] J.M. Maldacena, G.W. Moore, N. Seiberg, Geometrical interpretation of D-branes in gauged WZW models,
JHEP 0107 (2001) 046, hep-th/0105038.
[14] T. Quella, V. Schomerus, Symmetry breaking boundary states and defect lines, JHEP 0206 (2002) 028, hepth/0203161.
[15] H. Ishikawa, Boundary states in coset conformal field theories, Nucl. Phys. B 629 (2002) 209, hep-th/0111230.
[16] A.N. Schellekens, S. Yankielowicz, Extended chiral algebras and modular invariant partition functions, Nucl. Phys.
B 327 (1989) 673;
A.N. Schellekens, S. Yankielowicz, Simple currents, modular invariants and fixed points, Int. J. Mod. Phys. A 5
(1990) 2903.
388
[17] J. Fuchs, C. Schweigert, A classifying algebra for boundary conditions, Phys. Lett. B 414 (1997) 251, hepth/9708141.
[18] E. Bannai, T. Ito, Algebraic Combinatorics, I: Association Schemes, BenjaminCummings, Redwood City, CA,
1984.
[19] P. Di Francesco, J.B. Zuber, SU(N ) lattice integrable models associated with graphs, Nucl. Phys. B 338 (1990) 602;
P. Di Francesco and J.B. Zuber, SU(N ) lattice integrable models and modular invariance, in: S. Randjbar-Daemi,
E. Sezgin, J.-B. Zuber (Eds.), Recent Developments in Conformal Field Theories, Trieste Conference, 1989;
P. Di Francesco, Integrable lattice models, graphs and modular invariant conformal field theories, Int. J. Mod. Phys.
A 7 (1992) 407.
[20] V.B. Petkova, J.B. Zuber, From CFTs to graphs, Nucl. Phys. B 463 (1996) 161, hep-th/9510175.
[21] V. Schomerus, Non-compact string backgrounds and non-rational CFT, hep-th/0509155.
[22] E. Verlinde, Fusion rules and modular transformations in 2D conformal field theory, Nucl. Phys. B 300 (1988) 360.
[23] N. Ishibashi, The boundary and crosscap states in conformal field theories, Mod. Phys. Lett. A 4 (1989) 251.
[24] J. Fuchs, B. Schellekens, C. Schweigert, From Dynkin diagram symmetries to fixed point structures, Commun.
Math. Phys. 180 (1996) 39, hep-th/9506135.
[25] H. Ishikawa, A. Yamaguchi, Twisted boundary states in c = 1 coset conformal field theories, JHEP 0304 (2003)
026, hep-th/0301040.
[26] I. Affleck, M. Oshikawa, H. Saleur, Quantum Brownian motion on a triangular lattice and c = 2 boundary conformal
field theory, Nucl. Phys. B 594 (2001) 535, cond-mat/0009084.
[27] H. Ishikawa, T. Tani, Novel construction of boundary states in coset conformal field theories, Nucl. Phys. B 649
(2003) 205, hep-th/0207177.
[28] J. Fuchs, C. Schweigert, J. Walcher, Projections in string theory and boundary states for Gepner models, Nucl. Phys.
B 588 (2000) 110, hep-th/0003298.
[29] M.R. Gaberdiel, S. Schafer-Nameki, D-branes in an asymmetric orbifold, Nucl. Phys. B 654 (2003) 177, hepth/0210137.
[30] F. Xu, New braided endomorphisms from conformal inclusions, Commun. Math. Phys. 192 (1998) 349.
[31] J. Bckenhauer, D.E. Evans, Modular invariants, graphs and -induction for nets of subfactors, I, Commun. Math.
Phys. 197 (1998) 361, hep-th/9801171;
J. Bckenhauer, D.E. Evans, Modular invariants, graphs and -induction for nets of subfactors, II, Commun. Math.
Phys. 200 (1999) 57, hep-th/9805023;
J. Bckenhauer, D.E. Evans, Modular invariants, graphs and -induction for nets of subfactors, III, Commun. Math.
Phys. 205 (1999) 183, hep-th/9812110;
J. Bckenhauer, D.E. Evans, Y. Kawahigashi, Chiral structure of modular invariants for subfactors, Commun. Math.
Phys. 210 (2000) 733, math.QA/9907149.
[32] V.G. Kac, Infinite Dimensional Lie Algebras, Cambridge Univ. Press, Cambridge, 1990.
[33] P. Di Francesco, P. Mathieu, D. Senechal, Conformal Field Theory, Springer-Verlag, New York, 1997.
On generalized gauge fixing in the fieldantifield

formalism
I.A. Batalin a , K. Bering b, , P.H. Damgaard c
a I.E. Tamm Theory Division, P.N. Lebedev Physics Institute, Russian Academy of Sciences,
53 Leninsky Prospect, Moscow 11999, Russia

b Institute for Theoretical Physics and Astrophysics, Masaryk University, Kotlrsk 2, CZ-611 37 Brno, Czech Republic
c The Niels Bohr Institute, Blegdamsvej 17, DK-2100 Copenhagen, Denmark
Received 4 January 2006; accepted 23 January 2006

Abstract
We consider the problem of covariant gauge fixing in the most general setting of the fieldantifield formalism, where the action W and the gauge-fixing part X enter symmetrically and both satisfy the quantum
master equation. Analogous to the gauge-generating algebra of the action W , we analyze the possibility of
having a reducible gauge-fixing algebra of X. We treat a reducible gauge-fixing algebra of the so-called first
stage in full detail and generalize to arbitrary stages. The associated square root measure contributions are
worked out from first principles, with or without the presence of antisymplectic second-class constraints.
Finally, we consider an W X alternating multi-level generalization.
PACS: 02.40.-k; 02.40.Hw; 04.60.Gw; 11.10.-z; 11.10.Ef; 11.15.-q
Keywords: BV fieldantifield formalism; Odd Laplacian; Antisymplectic geometry; Second-class constraints; Reducible
gauge algebra; Gauge fixing
1. Introduction
The fieldantifield quantization formalism [13] has been given a substantial reformulation,
which shows how it fits into a much more general scheme [47]. The essential ingredient is a
E-mail addresses: batalin@lpi.ru (I.A. Batalin), bering@physics.muni.cz (K. Bering), phdamg@nbi.dk

(P.H. Damgaard).
doi:10.1016/j.nuclphysb.2006.01.030
390
I.A. Batalin et al. / Nuclear Physics B 739 [FS] (2006) 389440
Grassmann-odd and nilpotent differential operator that is symmetric,

= T ,
(1.1)
with respect to a transposition defined by

d F (G) = (1)F d T F G,
(1.2)
where d is a functional measure, whose explicit form we will return to in great detail below.
The partition function is then given by

i
X
Z = d e h (W +X) ,
(1.3)
where both W and X satisfy the quantum master equations
i
e h W = 0 and e h X = 0,
(1.4)
respectively. The important observation is that W , which through the boundary conditions incorporates the classical action, and X, which does the required fixing of gauge symmetries, enter
symmetrically. Viewing as a generalized odd Laplacian which may potentially have quantum corrections that consist of higher-order differential operators, we see that the integrand in
(1.3) is required to be a product of two superharmonic functions. One may argue on general
i
grounds that an arbitrary infinitesimal variation of the gauge-fixing part e h X has the form
i
e h X = [, ]e h X ,
(1.5)
or equivalently, using (1.4), the variation X is BRST-exact,

h i X i X
(1.6)
e h e h X ( ),
i
where X is a quantum BRST-operator. Surprisingly, the independence of gauge-fixing X for the
partition function Z X can formally be demonstrated [7] by just using the above ingredients (1.1),
(1.4) and (1.5), without reference to the detailed form of :

i
i
X+X
X
Z
Z = d e h W [, ]e h X

i
i i
i
= d e h W e h X + e h W e h X = 0.
(1.7)
X =
The analogous statement (with X replaced by W ) to show independence of the choice of gauge
generating functional W was first studied by Tyutin and Voronov [8].
The fact that the most general description of gauge-fixing X puts it on equal footing with the
construction of the action W means that there are situations in which gauge fixing must require
special attention. This happens when, be it for reasons of, for example, locality or unitarity, the
gauge-fixing function X itself contains gauge degrees of freedom. It is of interest to clarify what
happens in such a situation. What, in this formalism, fixes gauge symmetries of X? Remarkably,
it turns out that the machinery is ready to tackle this more general situation, and provide the
solution to the gauge-fixing problem. How this is achieved will be described in detail in this
paper; the principle is simply that gauge symmetries of X are fixed by what used to play the role
of only action, W . We thus introduce the notion of a gauge-fixing algebra in the X-part of the
391
fieldantifield formalism, very similar to the usual gauge-generating algebra inside the W -part.
In order to give a very specific example where the more general quantization problem needs to
be faced, we consider in Section 5 in detail the case where the gauge-fixing X is described in
terms of a set of reducible gauge-fixing conditions G . The process may not stop there, since
also the new gauge fixings in W in turn may contain additional symmetries that need to be fixed
as well. Then the formalism allows X to take over the gauge fixing again, and so forth, for any
finite number of steps in an alternating manner. In this way a multi-level construction is induced
naturally. This is the subject of Section 6. We take the viewpoint that the existence of these new
classes of theories must be taken seriously, and that their formal properties with respect to the
quantization program therefore must be established.
The present paper is devoted to an exposition of this more general situation. However, it is
necessary first to establish a systematic and condensed formalism before approaching these new
and interesting possibilities. We therefore begin in Section 2 with a discussion of some of the
geometrical aspects of the fieldantifield formalism from a covariant perspective. In particular,
we describe the properties of the measure density , and the use of anticanonical transformations,
i.e., the antisymplectic analogue of canonical transformations. As an immediate application we
extend the semi-density theory of Khudaverdian et al. [9] to the degenerate case. We next show
how deformations of solutions to the quantum master equation can be understood in terms of anticanonical transformations and associated measure function changes. This establishes a compact
formula for the changes of gauge in the action, which is essential for all subsequent developments
in the W X multi-level formalism. While we shall give the necessary definitions below, we here
just briefly remind the reader that the multi-level formalism must be introduced if one wishes
to secure the most general and covariant construction that in particularly simple gauges reduce
to the well-known fieldantifield prescription that was presented in the original papers [1,2]. In
Section 3 we then turn to the situation where gauge fixings in X are irreducible, exploring gauge
fixing at the first level, and providing a new compact derivation of the form of X in that case.
In Section 4 we return to the possibility of having antisymplectic second-class constraints in the
path integral, a situation quite analogous to the more conventional case of symplectic secondclass constraints with respect to the Poisson bracket in the Hamiltonian formalism. In particular,
filling out a gap in the existing literature, we first establish a reduction theorem which explicitly
demonstrates that the final gauge fixed path integral can be expressed, on a physical subspace of
antisymplectic coordinates, in precisely the same form as the partition function (1.3). Secondly,
we show that the second-class construction is manifestly invariant under reparametrizations of
the second-class constraints, a vital investigation that taps into the very foundation of the antisymplectic Dirac construction. In Section 5 we consider a reducible gauge-fixing algebra and work
out a general first-stage reducible theory in detail, and determining the associated path integral
measure by solving the master equation. We perform several consistency checks by reduction
techniques, linking reducible and irreducible descriptions of the gauge-fixing constraints, and
comparing minimal and non-minimal approaches. Section 6 discusses the generalization of the
first-level formalism to the above mentioned multi-level formalism. Finally, Section 7 contains
our conclusions.
2. Antisymplectic geometry revisited
Let us start with a covariant, odd -operator of second order,
= + V ,
() = 1,
(2.1)
392
l
(1)A l

E AB
,
A
2
B
(2.2)
V =VA
(2.3)
where A denotes local coordinates with Grassmann parity A ( A ). The -operator is build
with the help of covariant structure functions E AB = E AB ( ), V A = V A ( ), and = ( )
that transform under general coordinate transformations as a bi-vector, a vector and a density,
respectively. We shall assume that E AB has a Grassmann-graded skewsymmetry
E BA = (1)(A +1)(B +1) E AB .
(2.4)
Locally, Eq. (2.1) describes the most general second-order odd -operator such that
(1) = 0.
(2.5)
The condition (2.5) is not vital for the construction below, but since currently there are no applications that would require (1) = 0, we shall not pursue such a possibility here. The antibracket
of two functions F = F ( ) and G = G( ) is defined via a double commutator1 with the operator, acting on the constant function 1,
r

AB
(F, G) [F, ], G 1 = F
(2.6)
G
,
E
A
B
where use was made of Eq. (2.4). The square 2 = 12 [, ] is generally a third-order operator
with no zero-order term 2 (1) = 0. It becomes of second order if and only if a Grassmann-graded
Jacobi identity

(2.7)
(1)(F +1)(H +1) F, (G, H ) = 0
F,G,H cycl.
for the antibracket holds. We shall assume this from now on. A bi-vector E AB that satisfy
skewsymmetry (2.4) and the Jacobi identity (2.7) is called a possibly degenerate antisymplectic bi-vector. There is an antisymplectic analogue of Darbouxs theorem that states that locally,
if the rank of E AB is constant, there exist Darboux coordinates A = { ; ; a }, such that
the only non-vanishing antibrackets between the coordinates are ( , ) = = ( , ).
In other words, the Jacobi identity is the integrability condition for the Darboux coordinates.
The variables , and a are called fields, antifields and Casimirs, respectively. Granted the
Jacobi identity (2.7), the square of the -operator is a first-order differential operator,
2 (F G) = 2 (F )G + F2 (G),
if and only if there is a Leibniz rule for the interplay of and the antibracket

(F, G) = (F ), G (1)F F, (G) .
(2.8)
(2.9)
We shall also assume this to be the case. It is interesting to note that the Leibniz rule (2.9) holds
automatically for a conventional odd Laplacian (still assuming the Jacobi identity (2.7)), so
1 Here, and throughout the paper, [A, B] denotes the graded commutator [A, B] = AB (1)A B BA.
the Leibniz rule (2.9) actually reduces to

V (F, G) = V (F ), G (1)F F, V (G) .
393
(2.10)
We see that V is a generating vector field for an anticanonical transformation.

In the degenerate case, one would usually proceed by investigating an antisymplectic leaf/orbit
where the values of the Casimirs a are kept fixed. Then seen from within such leaf the antisymplectic structure will appear non-degenerate. An example of this is the case of antisymplectic
second-class constraints, which will be the subject of Section 4. On the other hand, if E AB is nondegenerate, the V in Eq. (2.10) becomes locally an Hamiltonian vector field, i.e., there exists a
bosonic Hamiltonian H such that V = (H, ). It follows that one can locally absorb the V -term
into a rescaling of the measure density = e2H . Since and H are intimately related
through this mechanism one may regard the Leibniz rule (2.9) as an integrability condition for
the local existence of . In any case, we shall from now on only consider the conventional odd
Laplacian without the V -term.
2.1. Compatible structures
A measure density and a possibly degenerate antisymplectic E AB are called compatible if
and only if the odd Laplacian is nilpotent,
2 = 0.
(2.11)
Constant and constant E AB are the most important example of compatible structures. It is
interesting to classify the compatible structures within the set of all pairs (, E). To this end,
consider two -operators sharing the same antisymplectic structure E, and with two different
measure densities and , respectively, that are not necessarily compatible with E. They differ
by a Hamiltonian vector field,

= ln /, .
(2.12)
Also the difference in their squares is a Hamiltonian vector field [10,11]:

2 2 = ( ; , E), .
Here we have introduced a Grassmann-odd function

1
1

( ; , E) / / = 1 1
(2.13)
(2.14)
of a measure density with respect to a reference system (, E). The quantity acts as a scalar
under general coordinate transformations, and satisfies the following 2-cocycle condition [9]:
(1 ; 2 , E) + (2 ; 3 , E) + (3 ; 1 , E) = 0,
(2.15)
and, as trivial consequences thereof,

(1 ; 1 , E) = 0,
(1 ; 2 , E) + (2 ; 1 , E) = 0.
In fact,
( ; , E)
(2.16)
can be written globally as a difference of a scalar function (; E),
( ; , E) = ( ; E) (; E).
(2.17)
394
To see Eq. (2.17), go to a coordinate system where E AB becomes equal to a constant antisymplectic reference matrix E0AB , which we for simplicity take to be the Darboux matrix. It follows
from the antisymplectic analogue of Darbouxs theorem that one may cover the manifold with
such coordinate charts, except for singular points where the rank of E jumps. In this section we
shall denote a -operator corresponding to constant = 1 and E AB = E0AB as 0 . Now define
1
(; E0 ) (; 1, E0 ) = 0 ,
(2.18)
where the arguments , 1 and E0 all refer to the above Darboux coordinate system. For the
definition (2.18) to be well-defined, one should justify that two different choices of Darboux
coordinates lead to same value of . By definition, any two Darboux coordinate systems are
connected by an anticanonical transformation. According to Lemma 2.1 below the Jacobian
A
f
Jf i sdet
(2.19)
iB
associated to an anticanonical transformation iA fA has a vanishing :
(Jf i ; E0 ) (Jf i ; 1, E0 ) =

1
0 Jf i = 0.
Jf i
(2.20)
Hence, it follows from the 2-cocycle condition (2.15) that the -definition (2.18) does not depend
on the particular choice of Darboux coordinate system:

i
(f ; 1, E0 ) =
; 1, E0 = (i ; Jf i , E0 ) = (i ; 1, E0 ) (Jf i ; 1, E0 )
Jf i
= (i ; 1, E0 ).
(2.21)
In this way one achieves a well-defined function (; E0 ) on the set of all Darboux coordinate
charts. It is assumed that the definition can be extended uniquely to singular points by continuity.
One generalizes the definition of (; E) to an arbitrary coordinate system A by requiring that
(; E) is a scalar under general coordinate transformations, i.e.,

(; E)
(2.22)
; E0 ,
J
A
A
0
where J sdet(
) denotes the Jacobian of a transformation 0 into some Darboux
A
coordinate system 0 . One may easily check that this definition fulfills Eq. (2.17). Moreover,
the definition is independent of the constant reference matrix E0AB . We shall from now on use
a shorthand notation (; E). Next, define an operator E that takes semi-densities to
semi-densities [9]

E ,
(2.23)
i.e., for Darboux coordinates it is simply

E 0 .
(2.24)
We emphasize that the constructions of and E rely heavily on Lemma 2.1. The operator E
is nilpotent,
2E = 0.
(2.25)
395
Eq. (2.25) encodes precisely the antisymplectic data (2.4) and (2.7) without information about
any particular . On the other hand, the odd Laplacian , which takes scalars to scalars, consists
of both the E-structure in E and a measure density ,
1
(F ) = [E , F ] = (F 2 )F,
(F ) = 0.
(2.26)
The nilpotency of E implies

( + )2 = 0,
(2.27)
( ) = 0,
(2.28)
2
= ( , ).
(2.29)
We summarize the above information in diagram (2.30).

The following diagram holds for an arbitrary pair of measure density and possibly degenerate antisymplectic structure E, cf. [9]:
Darboux coordinate
system such that = 1.
Darboux coordinate system and

anticanonical transformation
such that = J, the Jacobian
2 = 0
is a Casimir
= 0

Darboux coordinate system
such that eigenvector

for 0 with eigenvalue
(2.30)
The above eigenvalue is constant within an antisymplectic leaf/orbit. In the non-degenerate case
the Casimir is a Grassmann-odd constant. Evidently, Grassmann-odd constants cannot be
non-zero, if the theory does not have any external Grassmann-odd parameters. In practice, this
is the case. The diagram contains two implication arrows , that are not bi-implications . One
is the possibility of a non-zero Casimir ; the other is a non-trivial -cohomology obstruction,
cf. Section 2.3.
2.2. Anticanonical transformations
An anticanonical transformation preserves by definition the antisymplectic structure E. Infinitesimally, it is generated by a bosonic vector field X such that

X(F, G) = X(F ), G + F, X(G) .
(2.31)
A Hamiltonian vector field X = ad , where ad (, ) denotes the adjoint action with
respect to the antibracket, and where is a Grassmann-odd generator, is an example of an
infinitesimal anticanonical transformations (2.31). This follows directly from the Jacobi identity
(2.7). It is natural to call an infinitesimal anticanonical transformation X in Eq. (2.31) for adclosed, and a Hamiltonian vector field X = ad for ad-exact. If E is non-degenerate, then all
396
ad-closed vector fields are locally of the ad-exact type. Here we shall elaborate on ad-closed
vector fields in a possibly degenerate antisymplectic manifold. To this end, let
(1)A l A
div X
X
A
denote the divergence of a vector field X with respect to a measure density . Then
1
X ad-closed [ , X] = (div X, ),
2
(2.32)
(2.33)
and
1
(2.34)
( div X) = X( ).
2
Eqs. (2.33) and (2.34) are ad-closed versions of the Leibniz rule (2.9) and the relation (2.29),
respectively. They reduce to those relations, if X is ad-exact, because the odd Laplacian is the
divergence of a Hamiltonian vector field [10,11]
X ad-closed
1
= div (ad ), ( ) = 1.
2
In Darboux coordinates with = 1 Eq. (2.34) becomes
X ad-closed (1 div1 X) = 0.
(2.35)
(2.36)
This non-covariant result will be needed for the Lemma 2.1 below, which in turn is used to justify
the definition (2.18) of . To avoid circular logic, we mention that the special case Eq. (2.36)
can also be proven directly without relying on the concept of .
Consider now a one-parameter family of (not necessarily anticanonical) passive coordinate
transformations A (t) for some parameter t [ti , tf ], and governed by a one-parameter generating vector field Xt = Xt ( ),
d A (t)
= XtA .
(2.37)
dt
We are here and below guilty of infusing some active picture language into a passive picture, i.e.,
properly speaking, the active vector field is minus X, and so forth. The solution
A (t) = U (t; ti ) A (ti )
(2.38)
can be expressed with the help of a path-ordered exponential

tf
U (tf ; ti ) P exp
(2.39)
dt Xt .
ti
The Jacobian
J (tf ; ti ) sdet
A (tf )
B (ti )

(2.40)
is given by
tf
ln J (tf ; ti ) =
(t )
dt U (tf ; t) div1 i Xt ,
ti
(2.41)
397
(t)
where div1 refers to the divergence with = 1 in the coordinates A (t). The formula (2.41)
can be deduced from the differential equation

d
l d A (t)
(t)
(t )
ln J (t; ti ) = (1)A
= div1 Xt = div1 i Xt + Xt ln J (t; ti ) .
A
dt
(t) dt
(2.42)
The measure density transforms in the passive picture with the Jacobian
(tf ) =
(ti )
.
J (tf ; ti )
(2.43)
Therefore, satisfies the following differential equation:

d
Xt = Xt ln (t) div Xt .
ln (t) = div(t)
1
dt
(2.44)
Next, put back into the divergence in Eq. (2.41). Then

tf

J (tf ; ti )
dt U (tf ; t) div Xt .
U (tf ; ti )(ti ) = exp
(ti )
(2.45)
ti
This equation can be understood both in the active and the passive picture, and it will play a key
role later on. In deriving it we have used that
tf

dt U (tf ; t)Xt ln (ti ) =
ti
tf
dt

d
U (tf ; t) ln (ti ) = U (tf ; ti ) 1 ln (ti ).
dt
ti
(2.46)
Finally, recall the remarkable fact that the -operatordespite being a second-order differential
operatorhas a Leibniz type interplay with ad-closed vector fields, cf. Eq. (2.33). For this reason we can form identities that look very similar to well-known identities in ordinary quantum
mechanics. For instance,

1
Xt ad-closed , U (tf ; ti ) =
2
tf
dt U (tf ; t) ad(div Xt )U (t; ti ).
(2.47)
ti
We may now give a short direct proof of a result used in Eq. (2.20):
Lemma 2.1. Let A (ti ) A (tf ) be a finite anticanonical transformation between Darboux
coordinates in a possibly degenerate antisymplectic manifold. Then the Jacobian J (tf ; ti ) satisfies
(t )
1 i
(t )
J (tf ; ti ) = 0.
(2.48)
Here 1 i refers to the odd Laplacian with = 1 in the initial Darboux coordinates A (ti ).
398
Proof of Lemma 2.1.

f

(t )
(t )
1
J (tf ; ti ) =
dt 1 i , U (tf ; t) div1 i Xt
2
t
(t )
1 i ln
ti
1
=
4
tf
tf
dt
ti
1
=
4

(t )
(t )
dt U (tf ; t ) div1 i Xt , U (t ; t) div1 i Xt
tf
dt U (tf ; t
(t )
) div1 i Xt ,
ti
tf
(t )
dt U (tf ; t) div1 i Xt
(t t)
ti

1
= ln J (tf ; ti ), ln J (tf ; ti ) ,
2
where use has been made of Eqs. (2.36), (2.41) and (2.47).
(2.49)
Lemma 2.1 is a degenerate generalization of a well-known result [3,9] for the non-degenerate
case. The covariant version of Lemma 2.1 reads

U (tf ; ti )(ti )
(tf )
1/2

U (tf ; ti )(ti )
(tf )
1/2

= U (tf ; ti )(ti ); (tf ), E(tf )

= U (tf ; ti ) 1 ,
(2.50)
where A (ti ) A (tf ) U (tf ; ti ) A (ti ) is a finite anticanonical transformation.

We observe that it was never necessary in this subsection to assume that and E are compatible, i.e., that is nilpotent. In the remaining part of the paper we shall assume that the
-operator is nilpotent, except for a subtlety (4.36) concerning second-class constraints.
2.3. Varying the solutions to the quantum master equation
Let us now consider solutions to the quantum master equation.
i
We are mainly interested in deformations of the quantum master action W in e h W = 0

for nilpotent , where W satisfies certain rank and boundary conditions [13]. Here both
and E are kept fixed.
As a precursor for the above problem, it is of interest to vary the semi-density in

(t )
(t )
1 i = 0 for nilpotent 1 i . Here the antisymplectic structure E is kept fixed.
Let us collectively write ( ) = 0 to represent both types of problems, so that denotes a solution in the space of solutions to the quantum master equation. Now consider a
one-parameter family of solutions (t), where t [ti , tf ]. Obviously, the difference of neighboring solutions is -closed: (d/dt) = 0. If the difference is furthermore -exact, we may
399
write2

d (t)
(2.51)
= , (t) (t),
dt
where (t) is a one-parameter family of fermionic functions. Next, introduce a path-ordered
exponential
tf

V (tf ; ti ) P exp dt , (t) .
(2.52)
ti
Integrating Eq. (2.51) along the path, we find

(tf ) = V (tf ; ti ) (ti ) = V (tf ; ti ) (ti )V (ti ; tf )V (tf ; ti )1
= U (tf ; ti ) (ti ) v(tf ; ti ),
(2.53)
where we have used the identity
e[, ] e[, ] = e[[, ], ] = ead ,
(2.54)
and defined
tf

J (tf ; ti )
v(tf ; ti ) V (tf ; ti )1 = exp dt U (tf ; t) (t) =
U (tf ; ti )(ti ).
(ti )
ti
(2.55)
Use has been made of Eq. (2.35), Eq. (2.45) and the differential equation

d
v(t; ti ) = , (t) v(t; ti ) = (t), v(t; ti ) (t) v(t; ti ).

(2.56)
dt
In the third equality of (2.55) we reinterpret the (t) family, which originates from a -exact
variation, as a generator of an ad-exact anticanonical transformation. There is thus a one-to-one
correspondence between ad-exact anticanonical transformations and -exact variations. Moreover, as anticanonical transformations can be understood passively, one may also give -exact
variations a passive interpretation, i.e., one is not changing the solution ; only the coordinates
A . The detailed mechanism for this one-to-one correspondence is of great interest, both conceptionally and in practice.
Definition 2.2. We shall say that an anticanonical transformation U (tf ; ti ) acts on a pair (ti )
and (ti ) according to the following twisted transformation rules:

J (tf ; ti )
(ti ) (tf ) =
(2.57)
U (tf ; ti ) (ti ) (ti ) ,
(ti )
(ti ) (tf ) =
(ti )
.
J (tf ; ti )
(2.58)
2 In general, there is non-trivial -cohomology. In finite dimensions, for a constant non-degenerate antisymplectic matrix E AB , whose fermionic blocks vanish, the non-trivial 0 -cohomology is one-dimensional, generated by the fermionic
top-monomial, i.e., the monomial of all fermionic and no bosonic variables [12]. In local field theory, cohomology may
arise from locality requirements. Furthermore, our treatment obviously only applies to a path-connected solution space.
400
To be more precise, it is the stabilizer subgroup {U G|U ( ) = } that acts on solutions to

the quantum master equation = 0. The full group G of anticanonical transformations acts
on solutions to the modified quantum master equation ( + ) = 0.
Letting the anticanonical generator (t) depend on t is somewhat academic, because one
may always find an equivalent constant generator (and choose the parameter interval to be
[ti , tf ] = [0, 1]), such that U (tf = 1; ti = 0) = ead . While t -dependent s provide a deeper
theoretical understanding, it is preferred in practice to work with such t -independent s where
path-ordering issues are absent. In the latter case the above one-parameter solution is of the form
(t) = et[, ] i ,
(2.59)
and Eq. (2.45) reduces to

J (tf ; ti )
U (tf ; ti )(ti ) = E(ad ) ,
ln
(ti )
(2.60)
where
t
f =1
dt ext =
E(x) =
ti =0
ex 1
.
x
(2.61)
It follows from Eqs. (2.53) and (2.55) that

Proposition 2.3. A finite -exact transformation
i
e h Wf = e[, ] e h Wi
(2.62)
deforms the quantum action W according to

Wf = ead Wi + (i h)E(ad
) = ead Wi + (i h)
ead 1
.
ad
(2.63)
This important deformation formula will be used repeatedly throughout the remainder of the
paper. By expanding in Plancks constant,
W =S+

n
(i h)
Wn ,
n=1

n
(i h)
n ,
(2.64)
n=0
one sees that the classical action S undergoes a classical anticanonical transformation
Sf = ead 0 Si ,
while the leading quantum correction W1 transforms as

W1,f = ead 0 W1,i + E(ad 0 )0 + E(ad 0 )1 , Sf .
(2.65)
(2.66)
To summarize, the deformations of the classical solutions S to the classical master equation
(S, S) = 0 are generated by the group of anticanonical transformations ead 0 , cf. (2.65). This
should be compared to the quantum situation, where deformations of solutions W to the quani
tum master equation e h W = 0 are similarly generated by the group of quantum anticanonical
transformations ead . Here depends on h in accordance with (2.64), but with the important
401
difference that the group action is applied in a non-standard way, twisted by the semi-density
, cf. Eq. (2.57). This twisting effect is not felt at the classical level.
There are important exceptions where the above twisting is not present at all. This happens for
instance in Darboux coordinates A = { ; } when = () and = () are independent
of the antifields , so that = 0. Then the formula (2.63) reduces to a purely anticanonical
transformation

ad
Wi = Wi ; +
,
Wf = e
(2.67)
a formula that is intimately tied to the original way of gauge fixing in the fieldantifield formalism
[1].
i
i
i
i
The dilation transformation e h W Ce h W (or e h X Ce h X ), where C is a constant factor, is clearly a symmetry of the quantum master equation. From a mathematical standpoint,
granted that the action satisfies pertinent rank conditions, the scaling represents non-trivial
-cohomology, which is excluded from our reasoning (1.5) of gauge independence for Z. It
obviously does change the partition function Z CZ. On the other hand, from a physics perspective such an overall constant rescaling is totally trivial and plays no role whatsoever.
As another simple application, let us briefly mention the second type of problem. We have a
(t )
nilpotent -operator 1 i with (ti ) = 1 in the coordinates A (ti ). The constant semi-density
(t )
(ti ) = 1 is a trivial solution to (1 i ) = 0. Now act with an anticanonical transformation
U (tf ; ti ) on (ti ) = 1 according to the transformation rule Eq. (2.57). Then
(tf ) =
(ti )
1
=
,
J (tf ; ti ) J (tf ; ti )
(2.68)
so the transformed semi-density becomes
U (tf ; ti )[ (ti ) (ti ) ]
= J (tf ; ti ).
(tf ) =
(tf )
(2.69)
This in turn provides a descriptive proof of Lemma 2.1.

3. Irreducible first-level gauge-fixing formalism
3.1. Review of original gauge-fixing formalism
Our starting point is an action W = W ( ; h ) that possesses N gauge symmetries that should
be fixed. In the original gauge-fixing prescription of the fieldantifield formalism [13] there
is no X-part. In non-degenerate Darboux coordinates A = { ; } with a density that is
independent of the antifields , we may reformulate gauge fixing in the following way that is
easy to generalize later:
i
1a. First, change the Boltzmann factor e h W with a -exact transformation generated by a gauge
fermion function ,
i
e h W = e[, ] e h W .
(3.1)
This formula obviously preserves the quantum master equation, and it generalizes readily to
a gauge fermion operator , but we shall not pursue such a generalization here. According
402
to Eq. (2.63) the transformed action W becomes

W = ead W + (i h)E(ad
).
(3.2)
0 to zero,
1b. Next, put the antifields

i
i

Z = [d]e h W =0 = [d ]( )e h W .
(3.3)
1c. The partition function Z defined this way does not depend on the gauge fermion .
Proof. Start with exponentiating the -function in Eq. (3.3):

i
( ) = [d]e h X
(3.4)
with a trivial action X = that obviously satisfies the quantum master equation
i
[1] e h X = 0, where [1] +(1) (/ )(/ ) is the suitably extended -operator.

Then the partition function
i
i
Z = [d ][d]e h X e[[1] , ] e h W
(3.5)
becomes of the W X-form discussed in the introduction. The independence of follows
straightforwardly from the symmetry (1.1) of the -operator,
Z
1

Z =
[d ][d]
dt e(1t)[[1] , ] e h W [[1] , ]et[[1] , ] e h X = 0.
(3.6)
The s and the [1] can be viewed as part of the so-called first-level formalism, cf. Section 3.2. 2
2a. One may reach an alternative version of the partition function (3.3) by using the symmetry
(1.1) of the -operator to write Eq. (3.3) as

i
i
W [ , ]
( ) = [d ]e h W e ad eE( ad ) .
Z = [d ]e e
(3.7)
2b. If furthermore the gauge fermion is independent of the antifields , as is normally assumed, the transformation (3.2) reduces to a purely anticanonical transformation,
W = ead W,
(3.8)
and one arrives at the familiar prescription of the original fieldantifield formalism, where
gauge fixing is done by an explicit substitution of the antifields / with a field
gradient of a gauge fermion :

i
Z = [d]e h W = .
(3.9)
The above gauge-fixing procedure with explicit removal of the N antifields obviously refers
to a particular set of coordinates on the supermanifold, and is therefore not covariant. The Xpart of the new formulation is precisely introduced [6] as a covariantization of the gauge-fixing
prescription.
403
3.2. First-level formalism

The gauge-fixing procedure was considerably generalized in the nineties into a so-called
multi-level formalism [46] to allow for covariant and more flexible gauge-fixing choices. For
systematic reasons we shall retroactively call the original non-degenerate antisymplectic phase
space variables A = { ; } for zeroth-level fields, the expansion parameter h h (0) for the
zeroth-level Planck constant, and the gauge-fixing procedure of the last subsection for zerothlevel gauge fixing.
In the (irreducible) first-level formalism one introduces N Lagrange multipliers (1) of
Grassmann parity and N antifields (1) , which we collectively call the first-level
fields. The phase space variables

A
[1]
(3.10)
A ; ,
(1)
for the first-level formalism thus consist of the zeroth and the first-level fields.3 The first-level
odd Laplacian
[1] + (1)
(3.11)
gives rise to an extended antisymplectic structure in the standard way. We shall always assume
there is a trivial measure density associated with the first-level sector.4
Furthermore, one introduces a first-level Planck constant h (1) as a new expansion parameter
for the quantum action

X = + (i h (1) ) + (i h (1) )2 + O h 3(1) ,
(3.12)
where the dependence of the s, the s and the previous (zeroth) level objects is implied.
At the end of the calculations one substitutes back h (1) h.
The (first-level) Planck number
grading Pl Pl(1) is defined as [4]
A
Pl(F G) = Pl(F ) + Pl(G),
Pl(h
(0) ) = Pl = 0,

Pl(h
(3.13)
(1) ) = Pl = Pl = 1.
One may compactly write the Planck number grading as a Planck number operator

Pl = , [1] + h
.
(3.14)
(1)
h (1)
We remark that the introduction of two different expansion parameters h (0) and h (1) is spurred on
one hand by the wish to limit the number of terms in X by imposing (first-level) Planck number
etc., to depend on h (0) . If the latter is not
conservation, and on the other hand to allow , , ,
an issue, one only needs one Planck constant.
At the first level one is guided by the following
3 Notation: We use capital roman letters A, B, C, . . . from the beginning of the alphabet as upper index for both A
A , respectively. Usually a quantity Q
and [1]
(n) with a soft-bracket index (n) is associated with the nth level only, while
a quantity Q[n] with a hard-bracket index [n] accumulates all the levels n.
4 In the multi-level formalism the previous levels are treated covariantly and the present level non-covariantly. This
means at the first level that general zeroth-level coordinate transformations A A are allowed, while the first-level
fields and are considered to be fixed from the onset. This implies, for instance, that it is consistent to choose a
trivial measure in the first-level sector.
404
Principle 3.1. The gauge-fixing action X satisfies three requirements:

1. Planck number conservation: Pl( hX(1) ) = 0.
i
2. The quantum master equation: [1] exp[ h (1)
X] = 0.
A , i.e., 2N in the
3. The Hessian of X should have rank equal to half the number of fields [1]
irreducible case.
Strictly speaking, it is enough that the rank conditions are met only on stationary field configurations, a technicality we shall assume implicitly from now on. Planck number conservation
limits the lowest-order terms in X to

X = G + (i h (1) )H R + O ( )2 , Pl(X) = 1,
(3.15)
where

1
, Pl R = 2.
(1) +1 + (i h (1) )V + (i h (1) )2 G
R = U
2
The quantum master equation generates a tower of equations; the first few read
(G , G ) = G U ,
(3.16)
(3.17)
(G ) (H, G ) = (1)

(3.18)
+ G V ,

1
.
eH eH = (H ) + (H, H ) = V G G
(3.19)
2
The most important Eq. (3.17) is a non-Abelian involution of the N gauge-fixing constraints G .
There are essentially two equivalent ways of performing first-level gauge fixing: one may
gauge-fix either the X- or the W -part.
1a. To gauge-fix the X-part, first change the Boltzmann factor e h X with a [1] -exact transformation generated by a gauge fermion ,
i
e h X = e[[1] , ] e h X .
(3.20)
0 to zero,
1b. Next, put the Lagrange multiplier antifields

i
i
= [d ][d]e h (W +X ) =0 = [d[1] ]( )e h (W +X ) .
Z[1]
(3.21)
defined this way does not depend on the gauge fermion .

1c. The partition function Z[1]
Proof. Use second-level techniques, cf. Section 6, and exponentiate the -function in
Eq. (3.21):

i
hi W
= [d(2) ]e h W[2] ,
( )e
(3.22)
i
with an action W[2] = (2) + W that satisfies the master equation [2] e h W[2] = 0. Then
the partition function
i
i
Z[1] = [d[1] ][d(2) ]e h W[2] e[[2] , ] e h X

(3.23)
becomes of the W X-form discussed in the introduction.
405
1d. If, furthermore, the gauge fermion is independent of the Lagrange multiplier antifields
, as is normally assumed, the gauge-fixing action (3.20) reduces to
ad ),
X = e ad[1] X (i h)E(
(3.24)
where in detail,
e
ad[1]

ad
X=X e
; , E( ad )
.
(3.25)
2a. To alternatively gauge-fix the W -part, one may use the symmetry (1.1) of the [1] -operator
to re-write Eq. (3.21) as

i

i
i
i
= [d[1] ]e h X e[[1] , ] ( )e h W = [d[1] ]e h X ead[1] e h W[1] ,

Z[1]
(3.26)
where we have defined
i
e h W[1] = e[[1] , ] e h W .
(3.27)
,
2b. If the gauge fermion is independent of the Lagrange multiplier antifields

the action
in Eq. (3.27) reduces to W in Eq. (3.2), and the Lagrange multiplier antifields are
W[1]
gauge-fixed as
i.e., the partition function reads

i
Z[1] = [d ][d]e h (W +X) =E(ad ) .

= E(ad )
(3.28)
(3.29)
3.3. Going on-shell with respect to the constraints

A tractable gauge is the 0 gauge, i.e., a trivial gauge fermion 0. Then X reduces to
only two terms

X =0 = G + (i h (1) )H.
(3.30)
The s becomes Lagrange multipliers for the constraints G , which are in turn enforced directly through -functions in the path integralhence the name. Moreover, the set of constraints
G has to be irreducible in order for the rank condition on the Hessian of X to be met, i.e.,
X : G X = 0 A = (1) A :
(3.31)
Let
=
= + 1
be arbitrary zeroth-level coordinate functions of statistics
such that A {F ; G } forms a coordinate system in the zeroth-level sector, and let J =
A
A A . Then
sdet(
B ) denote the Jacobian of the transformation
F
F ( ; h)
X = G A .
(F )
Theorem 3.2. The quantum master equation implies that the quantum correction H depends on
the constraints G modulo terms that vanish on-shell with respect to the G s according to the
following square root formula [5,13]:

J sdet(F , G )
+ O(G),
H = ln
(3.32)
406
up to an overall unphysical integration constant that may be discarded. Moreover, in the Abelian
case U = 0, it is possible to solve off-shell as well:

H = ln
J sdet(F , G )
G
1

V (F, G)1 (F ;G)(tF ;G) dt F .
(3.33)
Proof of Theorem 3.2. First note that the N N matrix (F , G ) is invertible, at least
in a neighborhood of the constrained surface G 0. Use the Jacobi identity and the involution
(3.17)

F , G , G (1)( +1)( +1) ( ) = F , (G , G ) = F , G U

(3.34)
to deduce that
1
( +1)( +1)
( )
( , G ) (1)

1
= U + (1) G
F , U .
(3.35)
Now supertrace to get

(str ln , G ) + (1)( +1) , G 1

= (1) U
+ (1) ( +1) G 1 F , U .
(3.36)
Next use the new coordinates A {F ; G } to rewrite
(1)A l
(
A , G )
2 A
(1) +1 l
(1) l
= ln ,
F , G +
G +
(G , G )
2
F
2 G

(1)( +1)

r
(1) l

G U
G +

+
= ln ,
2
F
2 G

( +1)
G
U
, G + (1) U + (1)
= ln
sdet
G

r
(1)( +1)

(G , G ) 1
2
G

(
+1)
1 l
(1)
+
U ,
(3.37)
G
F ,F
2
F
where = /J denotes the transformed density, and where Eq. (3.36) has been used in the last
equality. The last three terms are of order O(G). Then one of the consequences of the master
equation (3.18) is

sdet
r
= O(G),
+H
ln
(3.38)
F
(G ) =
and the square root formula (3.32) follows by integration. The integration constant represents the
trivial dilation symmetry H H + c, where c is a constant. Clearly, H only appears in differentiated form in the quantum master equation, so any integration constant is a priori allowed.
407
Restricting ourselves to consider only classes of solutions that are [1] -exactly connected, it is
consistent to always set this integration constant to zero. 2
The square root formula (3.32) gives H as a function of the F s. But since the F s are
arbitrary, this dependence must be trivial. This fact may also be shown directly:
Lemma 3.3 (See [5,13]). The factor J sdet(F , G ) is independent of the F s up to terms that
vanish on-shell with respect to G , if the G s satisfy the involution equation (3.17).
Proof of Lemma 3.3. Exponentiate the two determinants by introducing a ghost pair C A and
C A of statistics A + 1, and another ghost pair B and B of statistics + 1, so that the product
of the two determinants inside the square root can be written as a partition function

i
h Sdet ,
Zdet = J sdet F , G = [d C][dC][d
(3.39)
B][dB]e
with a determinant action given as

r

A
C B + B F , G B .
Sdet = CA
B
(3.40)
Now there are several ways to proceed. Perhaps the most enlightening treatment is to rewrite this
as a mini fieldantifield system within our theory. Consider a classical action

r
CA,
S0 = C G
(3.41)
A
where we have split C A = {C ; C } of Grassmann parity (C ) = and (C ) = + 1, respectively. On-shell with respect to G the classical action S0 is invariant S0 0 under the
following BRST-like symmetry

C A = A , G B ,
(3.42)
because of the involution (3.17). Here is a Grassmann-odd parameter. The standard field
antifield recipe [1,2] now instructs us to construct a minimal proper action as

Smin = S0 + CA A , G B ,
(3.43)
and a non-minimal proper action
S = Smin + C B ,
(3.44)
where we have identified C with NakanishiLautrup auxiliary fields. It is easy to check that
(S, S)BC 0 and BC S = 0, where
BC
l
l
l
l
l
+1
A +1
(1)
+
(1)
+
(1)
B B
C A CA
B B
l
l
+ (1)A +1
.
C A C A
+1
(3.45)
So, the quantum master equation for S is satisfied on-shell. Now choose a gauge fermion as

r

= B F
(3.46)
CA.
A
408
The gauge-fixed action

S CA =
;B =
C A
B

r
r

= C G
C + C F
C A + B F , G B = Sdet
A
A
(3.47)
is precisely the determinant action Sdet . Hence, the partition function (3.39) does not depend on
, and since the F s only appear inside the gauge fermion , we conclude that the partition
function (3.39) does not depend on the F s as well. 2
The following Theorem 3.4 is in some respect a reversed statement of Theorem 3.2.
Theorem 3.4. For an arbitrary set of irreducible constraints G and structure functions U that
, etc., such that X is a solution to
satisfy the involution (3.17), there exist functions H , V , G
the quantum master equation.
Proof. This relies on Abelianization, i.e., there exist Abelian constraints G0 with (G0 , G0 ) = 0,
and an invertible rotation matrix , such that G = G0 . 2
Corollary 3.5 (See [5,13]). The partition function5

i
G
Z[1]
= [d ]e h W (G) J sdet(F , G )
(3.49)
is independent of the G s satisfying the involution equation (3.17).

Sketched Proof. Corollary 3.5 does not explicitly refer to an X-part, but we may always assume an underlying X-part because of Theorem 3.4. Therefore, the broad strategies concerning
independence of the gauge-fixing X-part mentioned in the introduction apply. 2
Another interesting result is the following
Theorem 3.6. The on-shell square root formula (3.32) for H , viewed as part of X, is form
invariant under finite [1] -exact deformations of X, even if X does not solve the quantum master
equation.
Proof of Theorem 3.6. Assume that the gauge-fixing action Xf that contains the investigated
quantum correction Hf , is a [1] -exact deformation
Xf = ead[1] Xi + (i h (1) )E(ad[1] )[1] ,
(3.50)
5 To recover the zeroth-level gauge fixing (3.9) one first goes to Darboux coordinates A = { ; } with = 1, and
then substitute F and G / . In our conventions the Lagrange multipliers and the antifields
(of the previous level) carry the same Grassmann parity. In detail, we define

(0)
,
and so forth.

(0)
(1)
+ 1 ,

(1)
,
(3.48)
409
of an initial gauge-fixing action Xi with a one-loop correction Hi that obeys the on-shell square
root formula (3.32), i.e.,

{Fi ; Gi } sdet(Fi , Gi, )
Hi = c ln sdet
(3.51)
+ O(Gi ),
where we included the integration constant c. We have to show that a similar formula holds for
Hf . Order by order in h (1) the (3.50) implies that
f = ead[1] 0 i ,
(3.52)
f = e
(3.53)
ad[1] 0

i + E(ad[1] 0 )[1] 0 + E(ad[1] 0 )1 , f [1] ,
f = O( ).
(3.54)
The generator = 0 + (i h (1) )1 + O(h 2(1) ) conserves the Planck number, Pl( ) = 0. This
restricts the possible lowest terms to

0 = (0) + (1) + (1) + O ( )2 ,
(3.55)
1 = O( ),
(3.56)
where the sign factors in front of the matrix (1) of Grassmann grading + are introduced
for later convenience. First we look at Eq. (3.52) for f :

l

+

Gf, = f =0 = exp ad (0) (1)
(3.57)
(1)
Gi, .
The constraints Gf, are a composition of a rotation and an anticanonical transformation,

i, ,
Gf, = G
(3.58)
where tilde denotes the anticanonical transformation

A A exp[ad (0) ] A .
(3.59)
We shall later need an expression for the superdeterminant of the rotation matrix ,
ln sdet = str ln = E(ad (0) ) str (1) .
(3.60)
To prove Eqs. (3.58) and (3.60) one may use one-parameter techniques, i.e., let the generator
t be proportional to a parameter t [0, 1] to study the transition
G (t = 0) Gi, G (t = 1) Gf, .
(3.61)
It follows from Eq. (3.57) that the constraints obey the differential equation

dG (t)
(3.62)
= (0) , G (t) (1) G (t).
dt
The first term on the right-hand side represents an anticanonical transformation, while the second
term is a rotation. The solution to Eq. (3.62) is
G (t) = (t) exp[t ad (0) ]Gi, ,
(3.63)
410
where the rotation matrix (t) is a path-ordered matrix expression in the parameter t [0, t],

t

(t) = P exp dt e(tt ) ad (0) (1) .
(3.64)
This leads immediately to Eqs. (3.58) and (3.60).

Next we look at the right-hand side of Eq. (3.53) for f . The third term is proportional to
either the constraints Gf, or to because of Eqs. (3.56) and (3.57), so only the first two terms
contribute on-shell:

Hf = f =0 = exp[ad (0) ]Hi + E(ad (0) )((0) + str (1) ) + O(Gf ).
(3.65)
Combining Eqs. (3.65), (3.51), (2.60), (3.60) and (3.58), one deduces the theorem:

i, )
{Fi ; Gi } sdet(Fi , G
i)
Hf = c + ln sdet
+ O(G

( )

A
( )

sdet
+ ln sdet + O(Gf )
+ ln
B

i, )
sdet(Fi , G
{Fi ; Gi }
sdet 2
+ O(Gf )
= c + ln sdet

{Ff ; Gf } sdet(Ff , Gf, )

+ O(Gf ),
= c + ln sdet
with Ff = Fi .
(3.66)
Since one may in principle create every X-solution through [1] -exact deformations of some
trivial X-action like that of the = 0 gauge, one may interpret Theorem 3.6 as generating the
square root formula (3.32) via first-level anticanonical transformations ead . Theorem 3.6 also
shows that the integration constant from Theorem 3.2 is invariant under [1] -exact deformations,
so one may consistently discard it. The proof shows that only the classical part 0 of the underlying generator plays an active role in the transformation of H on-shell. Here the word classical
is used in the first-level sense, i.e., for objects independent of h (1) . Moreover, we have seen that
0 generates rotations and zeroth-level anticanonical transformations of the G s.
4. Second-class constraints
It is of interest to extend the irreducible first-level construction of a gauge-fixed Lagrangian
path integral to include antisymplectic second-class constraints [4,14]. Consider therefore a set
of 2ND second-class constraints a with Grassmann parity ( a ) = a that reduce the 2N dimensional antisymplectic manifold down to a physical submanifold of dimension 2(N ND ).
This proceeds quite analogous to the Poisson-bracket treatment of second-class constraints in the
Hamiltonian formalism.6 The antibracket matrix

E ab a , b
411
(4.1)
of the second-class constraints a has by definition an inverse matrix Eab ,

Eab E bc = ac ,
(4.2)
so that one can introduce a Dirac antibracket completely analogous to the Dirac Poisson bracket
[4]

(F, G)D (F, G) F, a Eab b , G ,
(4.3)
where F = F ( ) and G = G( ) are arbitrary functions. The bracket satisfies a Jacobi identity

(4.4)
(1)(F +1)(H +1) (F, G)D , H D = 0
F,G,H cycl.
everywhere in the extended phase space A . The projection property ensures that the Dirac
antibracket vanishes,

adD a F a , F D = 0,
(4.5)
when taken of any function F with any of the constraints a . In addition, there exists a nilpotent
Dirac -operator D ,
2D = 0,
(4.6)
so that the Dirac antibracket (4.3) equals the failure of D to act as a derivation, in complete
analogy with the usual -operator. It reads
l
(1)A l
AB
D =
D ED
,
A
2D
B
with a degenerated antisymplectic metric

AB
ED
A, B D ,
(4.7)
(4.8)
and with a compatible Dirac measure density D = D ( ). By definition, the Dirac measure
density D transforms as

D
=
D
A
sdet
B
(4.9)
6 Here, we work partly at the gauge-generating zeroth level and partly at the gauge-fixing first level. Therefore,
the second-class constraints a = a ( ; h (1) ) and several of the Dirac constructions to be introduced below could in
principle depend on a Planck expansion parameter h (1) , which we assign to a previous minus first level. Moreover, we
postpone for simplicity the issue of reparametrizations of the constraints a a = a b ( ) b to Section 4.5, i.e.,
the defining set of constraints are kept fixed for now. In fact, we derive in Section 4.5 that a reparametrization invariant
formulation necessitates off-shell corrections to Eqs. (4.5), (4.6) and (4.11). Finally, let us mention that antisymplectic
first-class constraints, and moreover, the conversion from second to first-class antisymplectic constraints, have been
addressed in [15].
412
under change of coordinates A A , while the Dirac measure density D is required to

transform as

= D sdet a b + O()
D
(4.10)
under reparametrization of the constraints a a = a b ( ) b . Finally, the D -operator
annihilates the constraints

D a = 0,
(4.11)
because of Eq. (4.5), and independently of the choice of D . In the case of higher-order
-operators the a s become operators, and the condition (4.11) should be replaced with
[D , a ] = 0, cf. [14].
4.1. First-level partition function
With the above ingredients, the corresponding irreducible first-level Lagrangian path integral
formulation can be carried out very analogous to the case without second-class constraints [4].
The appropriate path integral in the = 0 gauge is

i
Z[1]D = [d ][d]D e h (WD +XD )
(4.12)
a ,
a
with both WD and XD satisfying the corresponding quantum master equations

i
i
D exp
(4.13)
WD = 0,
XD = 0.
[1]D exp
h (0)
h (1)
At the first level there are N ND Lagrange multipliers and N ND corresponding antifields
. One may again expand the action
XD = G + (i h (1) )H + O( )
(4.14)
in terms allowed by the Planck number conservation. The quantum master equation for XD
shows that the N ND gauge-fixing functions G are in involution with respect to the Dirac
antibracket,
(G , G )D = G U .
(4.15)
As in the case with no second-class constraints, an on-shell closed-form expression for the oneloop correction H has been found [14]. Let F = F ( ; h ) be arbitrary zeroth-level coordinate
functions such that

A F ; G ; a
(4.16)
A
forms a coordinate system in the zeroth-level sector, and let JD = sdet(

B ) denote the Jacobian
A
A
of the transformation . Then
Theorem 4.1. The quantum master equation implies that the one-loop correction H depends
on the constraints G modulo terms that vanish on-shell with respect to the G s and the a s
according to the following square root formula:

JD sdet(F , G )D
+ O(G; ).
H = ln
(4.17)
D
413
One may check that

Lemma 4.2. The factor JD sdet(F , G )D is independent of the F s up to terms that vanish
on-shell with respect to G , if the G s satisfy the involution (4.15) with respect to the Dirac
antibracket.
Proof of Lemma 4.2. We may also this time build an auxiliary fieldantifield system. It is almost
identical to the case without second-class constraints, so we shall only point out some of the
differences. The classical action S0 now reads

r
r
A
a a
+
C
C
CA,
S0 = C G
(4.18)
A
A
where we have split the antighost C A = {C ; C ; C a } in three parts that reflects the splitting
in (4.16). On-shell with respect to G the classical action S0 is invariant S0 0 under the
following BRST-like symmetry

C A = A , G D B ,
(4.19)
because of the involution (4.15). Note that in Lemma 4.2 we can work off-shell with respect to
the a s. Hence the minimal proper action of this auxiliary fieldantifield system is

Smin = S0 + CA A , G D B ,
(4.20)
and the non-minimal proper action is
S = Smin + C B .
(4.21)
There are at least three very good reasons to impose the D transformation rule (4.10). First
of all, it is precisely what is needed to make the partition function (4.12) invariant under reparametrization of the constraints a a = a b ( ) b . Secondly, note that the rule (4.10) also
render the expression inside the square root of (4.17) reparametrization invariant, up to terms that
vanish on-shell with respect to the a s. Thirdly, we shall show in Section 4.5 below that the rule
(4.10) is needed to make the Dirac odd Laplacian D reparametrization invariant on-shell.
4.2. Unitarizing coordinates
The 2N zeroth-level variables A = A ( ; ) can be viewed as functions of 2(N ND )
physical variables A and 2ND second-class variables a such that the second-class constraints
satisfy a ( ( ; )) = a . In other words, we may choose so-called unitarizing coordinates
A that split A = { A ; a } directly into a physical and a second-class subsector. We use
capital roman letters A, B, C, . . . from the beginning of the alphabet as upper index for both the
full and the reduced variables A and A , respectively. A change of unitarizing coordinates

A = A ; a A = A ; a
(4.22)
has in general the form
A = A ( ),
a = a b ( ) b ,
(4.23)
414
where the matrix a b is invertible. This causes the Jacobian J of the coordinate transformation
to factorize on-shell,
A
A

a
J = sdet
(4.24)
= sdet
sdet
+ O().
B
B
b
The Dirac antibracket becomes
r

A B
l
(F, G)D = F A , D
G
(4.25)
in unitarizing coordinates.
4.3. Reduction to physical submanifold
In unitarizing coordinates A = { A ; a }, we may assign reduced tilde objects that live on
the physical submanifold. In order of appearance,

D =0 ,
E AB A , B D =0 ,
l

(1)A l
AB = D
E
,
=0
2 A
B

G
X XD =0 ,
G
,
W WD =0 ,
=0

F F =0 , etc.
H H =0 ,
(4.26)
The antibracket
r

l
(F, G) F A E AB
G
,
(4.27)
etc., are all independent of the

the actions W and X,
the measure [d
], the odd Laplacian ,

a
defining set of constraints (and of the unitarizing coordinates) used in the Dirac construction,
is nilpotent,
2 = 0, because D does not contain cf. Section 4.5. The odd Laplacian
derivatives (when using unitarizing coordinates). Furthermore, the actions W and X satisfy the
quantum master equations

i
exp i W = 0,
[1] exp

(4.28)

X = 0.
h (0)
h (1)
The first-level partition function (4.12) reduces to

i
h (W +X)
Z[1]D = [d ][d]e
(4.29)
on the physical submanifold. The resulting partition function is precisely of the general form
(1.3) for fields entirely living on the physical submanifold. Moreover, the coordinates A
; a } are both examples
{F ; G ; a } used in (4.16), and in particular the coordinates {F ; G
of unitarizing coordinates. Therefore, the square root formula (4.17) reduces to

)
J sdet (F , G
H = ln
(4.30)
+ O(G).
Here we have used that the Jacobian

A

{F ; G }

=
sdet
J
JD =0 = sdet
B =0
B
415
(4.31)
satisfies the factorization property (4.24). We summarize the above observations in the following
Theorem 4.3 (Reduction Theorem). A first-level fieldantifield theory (4.12) with second-class
constraints a may always be written in a set of unitarizing coordinates A = { A ; a }. In
these coordinates the theory reduces to a physical theory (4.29) with physical coordinates A on
the physical submanifold. The reduction is independent of the parametrization of the constraints
a and the choice of unitarizing coordinates A = { A ; a }.
4.4. Transversal coordinates
Let us define transversal coordinates as unitarizing coordinates A = { A ; a } with the additional property that
A a
, = 0,
(4.32)
so that the second-class variables a and the physical variables A are perpendicular to each
other in the antibracket sense. For every system of unitarizing coordinates A = { A ; a } there
exist unique deformation functions XaA = XaA ( ) such that a unique primed set of coordinates
A = { A ; a }, defined as

A = A XaA a ,
(4.33)
a = a,
is a set of transversal coordinates: ( A , a ) = 0. In fact, XaA satisfy the following fixed-point
equation:

XaA = A , b Eba (1)c (1+A ) c XcA , b Eba ,
(4.34)
that may be solved recursively XaA = ( A , b )Eba + O() to all orders in .
We conclude that each set of second-class constraints a may be complemented with variables A into a system of transversal coordinates A = { A ; a }. In transversal coordinates the
Dirac antibracket matrix

A B
, D = A, B
(4.35)
becomes the original antibracket matrix. This may be used to give a short proof of the remarkable
fact that the Jacobi identity (4.4) for the Dirac antibracket holds everywhere in the extended
phase space A . Clearly, for all physical purposes it would have been enough to have the Jacobi
identity (4.4) satisfied just on the physical submanifold. Nevertheless, the Dirac construction
(4.3) provides the stronger Jacobi identity (4.4) for free.
Similarly, we may always impose strong nilpotency (4.6) of D when considering an arbitrary but fixed set of second-class constraints a . However, we shall see in the next Section 4.5
that strong nilpotency (4.6) and the transformation rule (4.10) cannot both be maintained under
reparametrization of the second-class constraints a . We have already seen the necessity of the
transformation rule (4.10), so instead we would surprisingly have to relax the nilpotency requirement (4.6) for D . A manifestly reparametrization invariant ansatz turns out to be that the square
416
of the Dirac odd Laplacian,

2D
=O
(4.36)
is a first-order differential operator with coefficient functions O A = O() that vanish on-shell
with respect to the second-class constraints a .
4.5. Reparametrization of the second-class constraints
Let us now reparametrize the defining set of second-class constraints
a a = a b ( ) b
(4.37)
in Eq. (4.1), and build the Dirac antibracket (, )D and odd Laplacian D from the primed set
of constraints a . Our aim is dual: first of all, we must check that the different choices of the
second-class constraints do not lead to different physical quantities on the physical submanifold.
Secondly, it is of interest to know whether a relation can be maintained strongly everywhere in
the extended phase space, or whether there appear additional contributions of order O().
The Dirac antibracket does not transform on-shell under reparametrization,
(F, G)D (F, G)D = (F, G)D + O().
(4.38)
In fact, one may calculate the above transformation to any order of precision in . This is important because higher-order terms in (4.38) that naively appear to play no physical role, can be
exposed by -differentiations. The calculations are simplified by choosing coordinates A such
that A = { A ; a } are transversal coordinates. To second order in one finds
(F, G)D (F, G)D
r

r

l
l
a
b
a
b
= F
,
G
F,
D
D a
a
b
b

r

d
l
b
c
F, a D
,G D
bc
a
d
r

r

l
l
a
b
c
d
+ F
G + O 3 .
, D
a
b
c
d
In particular, the analogue of Eq. (4.5) becomes

r

a
a
b, F D
,F D =
b

r

l
l
a
b
c
d
+
F + O 3
, D

b

c
d
= O().
Similarly, to first order in , the Dirac odd Laplacian transforms as

(1)A l A
(1)a l a
1

, F D + ln D
,
F
+
,F D
D F =
D
2 A
2 a
2

A a
(1)A l
l
l

b
= D F
F
A D ,
D a
2D
b
(4.39)
(4.40)
(1)A

2D
a

a
417
A

D
,F D

(1)a
l
l
a
b
F + R(F ),
,
b
2
a
D
(4.41)
where the remainder R(F ) = O( 2 ) is a second-order differential operator consisting of terms

that contain at least as many powers of s as -derivatives. It vanishes to the second order
O( 2 ) in when it is normal-ordered. We have furthermore defined

D
and

D
,
sdet
(4.42)
D
(1)A l A
A D , D
2D
(4.43)
is the Dirac odd Laplacian in unprimed transversal coordinates A = { A ; a } and equipped

as Dirac measure. To make sure that the Dirac odd Laplacian does not transform
with D
D
on-shell,

(D F ) D F = (D F ) + O(),
(4.44)
= + O(), which is just the transformation rule
we would clearly have to impose D
D
D
(4.10). In general, the Dirac odd Laplacian D does change when we leave the physical submanifold.
On the other hand, the transformation rule (4.10) provides us with a limited freedom in choos , or rather, in choosing . It is by construction clear that the squares 2 , 2 and 2 are
ing D
D
D
D
D
all first-order differential operators, and let us a priori assume that D is strongly nilpotent, i.e.,
= + O() implies that is at least nilpotent on-shell, because
Eq. (4.6). Then the rule D
D
D
D does not contain -derivatives (in transversal coordinates). Also D becomes nilpotent
on-shell,

2 2
2
l
A
F , O A = O(),
D F D F = D F + O
(4.45)
A
because each -derivative in (4.41) is accompanied with at least one power of .
The analogue of Eq. (4.11), derived using transversal coordinates, becomes

A b
a
l
(1)A l
a
D =
D
D b
2D A

(1)b
l
a
+ O 2
b,

b
2

l

l
b
a
b
b
a
= D
(1) ,
+ O 2
b
b
D

l

= (1)b b D
(4.46)
a + O 2 = O().
b
Applying D one more time one gets

2 a
l
b
b
a
+ O 2 = O().
D = (1) D D

b
(4.47)
418
From this we conclude somewhat surprisingly that D is in general not nilpotent away from the
= + O(). A manifestly reparaphysical submanifold, independently of the choice of D
D
metrization invariant formulation is to assume the weaker nilpotency (4.36) from the beginning.
This has of course no consequences for the physics, which only lives on-shell.
5. Reducible gauge fixing
We consider in this section an interesting generalization, where the gauge-fixing functions
G0 in the X-part become reducible. This is quite analogous to reducibility among the gaugegenerators in the zeroth-level W -part. Recall that originally the stage of reducibility in the W sector was introduced so that a zeroth-stage gauge theory corresponds to an irreducible gauge
algebra, i.e., if the ghosts do not carry gauge symmetry. Similarly, first-stage gauge theories have
ghosts-for-ghosts that do not carry gauge symmetry, and so forth [2]. We shall here adjust this
terminology to the gauge-fixing X-part in the first-level formalism.
The motivation to work with an overcomplete set of constraints is a well-known theme in the
theory of constrained dynamics: often the independent set of constraints breaks symmetries (such
as, e.g., Lorentz covariance) or locality that one would like to preserve during the quantization
process. Here an overcomplete set of constraints can provide an immediate remedy.
One starts as usual with a zeroth-level theory W = W ( ; h ) that has N gauge symmetries
that should be fixed. Next one introduces N0 Lagrange multipliers 0 and N0 antifields 0 .
For each positive integer i one chooses a number Ni of so-called (first-level) ith-stage ghosts
i
(1)i
i , of
, with Grassmann parity i + i, where i i , and Ni antifields ci c(1)
ci c(1)i
i
opposite statistics, where the index i runs through i = 1, . . . , Ni . The integers N0 , N1 , N2 , . . . ,
can be chosen at will, as long as all of the following alternating sums are non-negative:
i 1
(1)ij Nj 0,
(5.1)
j =1
where N1 N . In particular, one may easily check from the above inequalities that
i 1 Ni 0,
(5.2)
as it should be. The stage s of reducibility is defined as the maximum

s max {i 0|Ni > 0} {1} ,
(5.3)
over the non-empty set {i 0|Ni > 0} {1}. A zeroth-stage theory with s = 0 requires
N = N 0 > 0 = N 1 = N2 = N3 = .
(5.4)
This is precisely the irreducible case of Section 3. Similarly, a first-stage theory with s = 1
requires
N 0,
N0 N = N1 > 0 = N2 = N3 = N4 = ,
while a higher-stage theory requires

s = 2:
N0 N 0,
N 1 N 0 + N = N2 > 0 = N 3 = N 4 = N 5 = ,
s = 3:
N3 N2 + N1 + N = N0 N 0,
N 2 N3 > 0 = N 4 = N 5 = ,
(5.5)
s = 4:
..
.
N4 N3 + N2 = N1 N0 + N 0,
..
.
419
N0 N 0, N3 N4 > 0 = N5 = ,

(5.6)
and so forth. The stage s of reducibility could be . If s < , then i=0 N2i1 = i=0 N2i .
Summarizing, the minimal field content in the first-level formalism is

A
A
s
0
1
2
[1]
(5.7)
min ; , 0 ; c , c1 ; c , c2 ; . . . ; c , cs .
s
Inaddition to the 2 i=1 Ni minimal fields and antifields, there is a triangular tower of
4 si=1 iNi non-minimal fields and antifields, in complete analogy with reducibility at the zeroth level, cf. Ref. [2] and Section 5.7. The first-level minimal odd Laplacian becomes
[1] min + (1)
0
0 0
s

l l
+
(1)i +i
,
c i ci
(5.8)
i=1
while the first-level minimal Planck-number operator is

s

0
i
Plmin = 0 +
(i + 1)ci c ,
+ h (1)
.
h (1)
i=1
[1]
(5.9)
The gauge-fixing action Xmin should again satisfy Principle 3.1, i.e., (1) Planck number conservation, (2) the quantum master equation and (3) rank requirements. Although it is straightforward
to expand Xmin in action terms allowed by Planck number, it quickly becomes space consuming.
Instead, we shall focus on a few important terms
Xmin = G0
+ 0
+
+ (i h (1) )H
+ 0 Z 0 1 c1
s1

ci Z i i+1 ci+1
i=1

1 0 0
0 +1
0
0
(1)
+ (i h (1) )V 0 0 + (i h (1) )2 0 G
U
2 0 0

ci Uii0 0 (1)i +i+1 + (i h (1) )V i i ci + .
(5.10)
i=1
In particular, the FaddeevPopov term 0 Z 0 1 c1 and its higher-stage counterparts ci

Z i i+1 ci+1 will be important new ingredients (as compared to the irreducible case). The struc
ture functions Z i1 i Zi i1 i ( ; h ) will carry Grassmann parity i + i1 . We stress that
Eq. (5.10) should not be read as a systematic expansion of the action Xmin . Rather the terms in
Eq. (5.10) were selected simply because they enter the first few consequences of the quantum
master equation for Xmin ,
(G0 , G0 ) = G0 U000 ,
(5.11)
1 = 0,
i
i Z i+1 = O(G),
(5.12)
G0 Z
i1
Z

i+1
(0 +1)(i +i+1 ) i
Z i i+1 , G0 = Z i i+1 Ui+1
Ui 0 Z i i+1 + O(G),
0 (1)
(G0 ) (H, G0 ) =
s

(1)i +i Uii0 + G0 V 0 0 ,
i=0
(5.13)
(5.14)
(5.15)
420

1
0 .
(H ) + (H, H ) =
V i i G0 G
2
s
(5.16)
i=0
The first equation (5.11) is just the usual non-Abelian involution of the gauge-fixing functions
G0 . The second equation (5.12) and the third equation (5.13) show that G0 and Z i1 i , respectively, are in general reducible. They imply that the action X exhibit gauge symmetries on-shell
with respect to the G0 s,
ci = Z i i+1 i+1 ,
(5.17)
where i+1 are gauge parameters.

Note that an irreducible gauge-fixing action X corresponding to s = 0 has no gauge symmetry
at the first level.7 This is why we could choose a trivial first-level gauge 0 in Section 3.3. In
the reducible case the gauge fermion should meet certain rank requirements.
5.1. First-stage reducibility
In Sections 5.15.6 we shall work out the simplest case of reducible gauge fixing in detail,
namely first-stage reducibility. In this setting the gauge-fixing action X has N1 gauge symmetries
in the 0 -variables due to reducibility equation (5.12) among the gauge-fixing functions G0 .
Besides the two minimal first-level fields 0 and c1 , there are also two non-minimal fields, c1
of statistics 1 + 1, and 1 of statistics 1 , i.e.,

A
[1]
(5.18)
A ; 0 , 0 ; c1 , c1 ; c1 , c1 ; 1 , 1 .
The Planck-number operator is chosen to be

Pl = 0 0 + 2c1 c1 c1 c1 , [1] + h
,
(1)
h (1)
or equivalently,

Pl 0 = 1,
Pl c1 = 2,
Pl(c1 ) = 1,

Pl 0 = 1,
Pl c1 = 2,
Pl c1 = 1,

Pl A = 0,
Pl(h
Pl(h
(1) ) = 1,
) = 0.
(5.19)
Pl(1 ) = 0,
Pl 1 = 0,
(5.20)
have vanishing Planck number, so they appear on the same

Note, in particular, that 1 and
footing as the original variables A in a Planck number expansion. This is just one of many
reasons to enlarge the 2N -dimensional zeroth-level phase space A into a 2N0 -dimensional
phase space

Aext A ; 1 , 1 ,
(5.21)
where N0 = N + N1 . We shall see in Section 5.4 that this space plays a profound role. The odd
Laplacian reads

l
l
l l
l
l
0
1 +1
+ (1)
+
,
[1] ext + (1)
(5.22)
0 0
c1 c1
c1 c 1
7 Compare this with the zeroth-level terminology, where a trivial gauge algebra, i.e., with no gauge symmetry, is strictly
speaking of stage 1. In other words, the definition of stage of the first-level X-action has been shifted by one unit as
compared to the original definition of stage of the zeroth-level W -action [2]. This shift is introduced to avoid speaking
of negative stages.
where
ext + (1)
1
1 1
421
(5.23)
The subscript ext will everywhere in this section refer to the extended space Aext .
However, let us first take a more traditional route. Recall that in the original fieldantifield
approach the minimal sector is introduced to obtain solutions to the master equation satisfying the appropriate rank condition, and the non-minimal sector is only added to have welldefined gauge-fixing choices at hand [1,2]. So, ignoring for the moment the non-minimal fields
{c1 , c 1 ; 1 , 1 }, the rank of the Hessian of Xmin in the minimal sector
A
A
0
1
[1]
(5.24)
min ; , 0 ; c , c1
should be 2N0 , where N0 = N + N1 . This implies that the two rectangular matrices G0 / A
and Z 0 1 have maximal rank, i.e.,

r
rank G0
(5.25)
= N,
rank Z 0 1 = N1 ,
A
and therefore the Z 0 1 matrix does not have zero-eigenvalue right eigenvectors.
5.2. First-level gauge fixing
The standard ansatz for the non-minimal action is
A
X = Xmin [1]
[1] + 1 c 1 ,
min ; h
(5.26)
while a simple choice for reads

= c1 1 ,
Pl( ) = 0,
(5.27)
with N1
) of Grassmann parity 1 . Planck
0; h
number conservation restricts us to a linear dependence of 0 ,
first-level gauge-fixing conditions 1
1 = 1 0 0
= 1 ( ;
(5.28)
(up to an inessential constant proportional to h (1) ). This gauge-fixing choice is based on a matrix
1 0 = 1 0 ( ; h ) of Grassmann parity 0 + 1 , such that the FaddeevPopov matrix
1 1 1 0 Z 0 1
(5.29)
is invertible, i.e.,

rank 1 1 = N1 .
(5.30)
One could in principle let 1 0 depend on 1 , but one may prove a constraint 1 0 that
would turn this idea into a vacuous exercise. According to the general theory outlined in Section 3.2 the first-level partition function is given by

i
i

= d e h (W +X ) ,c ,c , =0 = d e h (W +X) ,
Z[1]
(5.31)
0
with a measure
d = [d ][d0 ][dc1 ][d c1 ][d1 ],
(5.32)
422
and a gauge-fixing surface specified by
= c1 E(ad )1 0 ,
0
c1 = E(ad ) = 0,
c 1
c 1 = E(ad )
= E(ad ) 1 ,
c1
1 = E(ad )
= 0,
1
0 = E(ad )
(5.33)
(5.34)
(5.35)
(5.36)
cf. the prescription (3.28).

Lemma 5.1. The path integrations over the FaddeevPopov ghost pair {c1 ; c1 } can be performed explicitly. The first-level partition function (5.31) thereby simplifies to

i
i
0
Z[1] = [d ][d0 ][d1 ]e h W + h G,0 H sdet(1 1 ),
(5.37)
where
G,0 G0 + 1 1 0
(5.38)
are -deformed gauge-fixing constraints, and 1 1 is the FaddeevPopov matrix (5.29).

Proof of Lemma 5.1. It is convenient to split the gauge-fixed action

h (1)
S=
W + X = S0 + SFP + V ,
h
(5.39)
into a part
S0
h (1)
W + G0 0 + (i h (1) )H +
h
(5.40)
that is independent of the ghosts and antighosts {c1 ; c1 }, a FaddeevPopov term SFP
c1 1 1 c1 that is quadratic in {c1 ; c1 }, and a part V that contains all interaction terms, tadpole terms and terms quadratic in the antighost c1 . At this point the only quantities left that carry
Planck number, are h (1) , 0 , c1 and c1 . Since the action S has Planck number Pl(S) = 1, the
multiplicities mh , m , m1 , and m
1 of h (1) , 0 , c1 and c1 , respectively, must obey
mh + m + 2m1 m
1 =1
(5.41)
in any given term in the action. Equivalently,

m
1 m1 = mh + m + m1 1 RHS.
(5.42)
Clearly, the right-hand side RHS 1. There are no terms with RHS = 1, and the terms with
RHS = 0 are precisely the free part S0 + SFP . This implies that all the terms in the V -part have
fewer ghosts c1 than antighosts c1 , i.e.,
m1 < m
1.
(5.43)
423
Next, one scales the ghosts and antighosts,

1
c1 c1 ,
c1 c1 ,
(5.44)
and let 0. The V -term drops out because of the rule (5.43), while the free part S0 + SFP and
the path integral measure are unchanged. Hence the integration over {c1 ; c1 } can be explicitly
performed. 2
5.3. A square root formula for H
The -deformed constraints G,0 have two very important properties:
1. First, the N0 constraints G,0 are irreducible on the 2N0 -dimensional extended space

Aext A ; 1 , 1 .
(5.45)
Besides containing the original reducible constraints G0 0 of rank N , the -deformed
constraints G,0 0 in addition contain N1 conditions 1 0, as is clear from the formula

1 = G,0 Z 0 1 1 1 1 ,
(5.46)
where use has been made of Eq. (5.12). See Section 5.5 below for more details.
2. Second, the G,0 are in non-Abelian involution
0
(G,0 , G,0 )ext = G,0 U,
,
0 0
(5.47)
with respect to the extended ( , )ext bracket. The -deformed structure functions read
0
U,
=
0 0

1 0
0 Z 0 1 1 1 1 1 0 U00 0 + Z 0 1 1 1 1 1 0 , G,0 ext
2
(1)(0 +1)(0 +1) (0 0 ) + O(G )
(5.48)
to the zeroth order in G .

In the following Theorem 5.2 the fundamental role played by the 2N0 -dimensional extended
space Aext is displayed further. Let
F 0 F 0 + 0 1 1 + 0 1 1 +
(5.49)
be arbitrary zeroth-level coordinate functions of statistics 0 + 1, where denotes terms that

are at least quadratic in 1 and 1 , and where
F 0 = F 0 ( ; h ),
0 1 = 0 1 ( ; h ),
0 1 = 0 1 ( ; h ),
such that the transformation

0
Aext
Aext A ; 1 , 1 ,
F ; G,0
(5.50)
(5.51)

Aext
)
is a coordinate transformation of the 2N0 -dimensional space Aext , and let J, = sdet( ,Bext
denote the Jacobian. Then
424
Theorem 5.2. The quantum master equation for Xmin implies that the quantum correction H is
given the following square root formula:

J, sdet(F 0 , G,0 )ext

H = ln
(5.52)
+ O(G).

sdet(1 1 )2
1 ,1 =0
There are essentially two new features in the reducible case that we would like to emphasize as compared to the irreducible case, cf. Theorem 3.2. First and most pronounced, almost
all reference to the original space A has been replaced with the extended space Aext . The
two conditions G0 0 and 1 0 enter slightly asymmetrically, because H = H ( ; h ) conventionally does not depend on 1 , and hence this asymmetry is due to notation rather than
substance, cf. Section 5.4. Secondly, the FaddeevPopov determinant sdet(1 1 ) makes an interesting appearance in formula (5.52).
In more details the Jacobian is given as

r
i
Aext
J, sdet ,
(5.53)
=
[d C ext ][dCext ]e h SC ,
B
ext
with a Jacobian action

SC = C Aext Aext

r
C Bext
,
Bext

r
l
0
Aext +1 Aext
0
= (1)
C
G,0 C + C0 F
C Aext
Aext
Aext

l
1 +1
A +1 A
1
= (1)
C
G + (1)
C1 0 C 0
A 0

r
A
0
1
0 1
+ C 0 F 0
+
C
+
C
C
1
1 + O 1 ; 1 .
A
Similarly, one may exponentiate the other superdeterminant

0
i
sdet F , G,0 ext = [d B0 ][dB0 ]e h SB
(5.54)
(5.55)
with action

SB = B 0 F 0 , G,0 ext B 0

= B 0 F 0 , G0 0 1 1 0 B 0 + O 1 ; 1 .
(5.56)
Here the ghost pair (C Aext ; C Aext ) has statistics Aext + 1 and the ghost pair (B 0 ; B 0 ) has
statistics 0 + 1. We have split C Aext = {C 0 ; C 0 } of Grassmann parity (C 0 ) = 0 and
(C 0 ) = 0 + 1, respectively. The ghosts C Aext of the 2N0 -dimensional extended space decompose as C Aext = {C A ; C1 , C 1 }, with Grassmann parity (C A ) = A + 1, (C1 ) = 1 + 1
and (C 1 ) = 1 , respectively.
Due to the above mentioned properties of G,0 , it is clear that one may re-use the Lemma 3.3
from Section 3.3 for this situation:
Lemma 5.3. The factor J, sdet(F 0 , G,0 )ext is independent of F 0 up to terms that vanish
on-shell with respect to G,0 , if the G,0 s are in involution with respect to the ( , )ext bracket.
425
Proof of Theorem 5.2. We would like to derive an analogue of Eq. (3.18) with G replaced
with the -deformed G,0 in the extended space Aext . From the master equation, we are provided with a reducible version (5.15). Let us first contract an index on the -deformed structure
functions (5.48):

0
= (1)0 00 Z 0 1 1 1 1 1 0 U000
(1)0 U,
0 0

+ (1)0 Z 0 1 1 1 1 1 0 , G,0 ext + O(G ),
(5.57)
where use has been made of Eq. (5.12). Next, one derives

(1)1 U110 = (1)1 1 1 1 1 0 Z 0 1 , G,0 ext

+ (1)0 Z 0 1 1 1 1 1 0 U000 + O(G )
(5.58)
from Eq. (5.14). Combining Eqs. (5.15), (5.57) and (5.58), one gets the sought-for equation

0
(ext G,0 ) H ln sdet 1 1 , G,0 ext = (1)0 U,
(5.59)
+ O(G ).
0 0
Besides working in the extended space Aext , the only real difference from Eq. (3.18) is that H
has been shifted by the FaddeevPopov determinant. Hence, one may proceed as in the proof of
Theorem 3.2. 2
5.4. From irreducible to reducible constraints
The partition function for reducible gauge-fixing constraints becomes

i
G
Z[1] = [d ][d1 ]e h W (G,0 ) J, sdet F 0 , G,0 ext
1 =0
(5.60)
by combining the two previous results (5.37) and (5.52). Note that the FaddeevPopov determinant have completely cancelled out from the partition function (5.60)! This suggests that there
is a much simpler and broader approach as follows: if you want N0 reducible gauge-fixing constraints in the original 2N -dimensional zeroth-level phase space A , then introduce first-level
variable {0 , 0 ; 1 , 1 }, and an action

X = G0 0 + (i h (1) )H 0 R0 + O ( )2 ,
(5.61)
where all the structure functions G0 = G0 (ext ; h ), H = H(ext ; h ), etc., are allowed to depend
on 1 and 1 as well. Hence the full constraints G0 live in the extended Aext space and
should be irreducible, because of rank requirements, while the reducible constraints are simply
the restriction

G0 = G0 , =0
(5.62)
1
into the original A space. According to Theorem 3.2 applied on the extended space Aext , the
quantum master equation for X will carve out the square root measure factor directly,

J sdet(F 0 , G0 )ext
H = ln
(5.63)
+ O(G),
Aext
with J sdet( Bext ) and Aext {F 0 ; G0 }. There is thus no need to introduce a Faddeev
Popov ghost pair {c1 ; c1 }. Therefore, by appealing to the first-level irreducible theory from
426
Sections 3.2 and 3.3, one derives the following version of Corollary 3.5, written in the so-called
0 = 0 = 1 gauge.
Corollary 5.4. The partition function

i
G
Z[1]
= [dext ]e h W 1 (G0 ) J sdet F 0 , G0 ext
(5.64)
is independent of G0 s that are in involution with respect to the ( , )ext -bracket.

We conclude the following
Theorem 5.5 (Reduction Theorem). The irreducible partition function (5.64) on the 2N0 dimensional extended space Aext reduces to the reducible partition function (5.60) on the
2N -dimensional original space A , if one chooses the gauge-fixing conditions G0 to be the
-deformed constraints G,0 G0 + 1 1 0 .
In the next subsection we will present a Reduction Theorem 5.6 that in some respect is opposite to the above Reduction Theorem 5.5.
5.5. From reducible to irreducible constraints
One may always assume that the reducible constraints G0 , 0 = 1, . . . , N0 , can be written as
linear combinations of irreducible constraints G , = 1, . . . , N ,
G0 = G P 0 ,
(5.65)
such that the irreducible constraints G are in involution, cf. Eq. (3.17). Therefore, the theory
can be set up purely within the irreducible framework of Section 3. In this subsection we check
that the reducible and the irreducible approach agree.
Theorem 5.6 (Reduction Theorem). The reducible partition function (5.60) coincides with the
irreducible partition function (3.49), when re-writing the reducible quantities in their irreducible
counterparts.
Proof. First of all, let us note that the reducible gradients

l
l

G =
G P 0 + O(G )
A 0
A
(5.66)
are linear combinations of irreducible gradients on-shell due to Eq. (5.65). The rectangular matrix
P 0 has rank(P 0 ) = N . It follows from Eqs. (5.12) and (5.65) that there exists an antisymmetric matrix A 1 = A 1 (1) such that
X 1 P 0 Z 0 1 = G A 1 = O(G ).
Because of the rank condition (5.30), one may combine
N0 matrix

P
,
N N
0
(5.67)
P
and
into an invertible N0
(5.68)
427
at least in the vicinity of the constrained surface G 0. Next, define a rectangular matrix P 0
via

1
P
[P ]N0 N =
,
(5.69)
0 N N
N N
0
and a square matrix R 0 0 as

[R]N0 N0 [ P
Z ]N0 N0 .
It follows that

1
P
[R]N0 N0 =
0
N N
0
(5.70)
X

,
(5.71)
N0 N0
cf. (5.29), so that R 0 0 must be invertible as well. Next change the coordinates 0 = P 0 +
Z 0 1 1 , or equivalently,

.
0 N 1 = [R]N0 N0 1
(5.72)
0
N 1
0
Therefore, one can decompose the reducible -function

i

i
0
1
1
[d ] d1 e h G +1 1
(G,0 ) = [d0 ]e h G,0 = sdet R 0 0
=
sdet(R 0 0 )
(G )(1 )
sdet(1 1 )
(5.73)
in its irreducible components.

There is an almost identical story for the reducible partners F 0 , 0 = 1, . . . , N0 , in terms
of irreducible primed functions F , = 1, . . . , N , although with some important differences.
For instance, we shall assume directly that the reducible gradients are linear combinations of
irreducible gradients,

r
r
0

F 0
(5.74)
=
Q
F
.
A
A
The rectangular matrix Q0 has rank N . This means there exists a matrix Z 1 0 of rank N1
such that
Z 1 0 Q0 = 0.
(5.75)
Then 0 1 is chosen such that

1 1 Z 1 0 0 1
(5.76)
is invertible, i.e.,

1 1 = N1 .
rank
One may combine
[Q
Q0
and
(5.77)
0
into an invertible N0 N0 matrix
]N0 N0 .
(5.78)
Next, define a rectangular matrix Q 0 via

N N0 [ Q
[Q]
]N0 N0 = [ 1 0 ]NN0 ,
(5.79)
428
and a square matrix R 0 0 as

N0 N0 Q
[R]
.
Z N N
0
(5.80)
Finally change the coordinates

C
,
C 0 N 1 = [R]N0 N0 1
0
C
N0 1

N0 N0 ,
[C 0 ]1N0 = C C 1 1N [R]
0

B
B 0 N 1 = [R]N0 N0
,
0
B 1 N 1
0

N0 N0 .
[B 0 ]1N0 = B B 1 1N [R]
(5.81)
The two superdeterminant actions (5.54) and (5.56) become

l

SC = (1)A +1 C A
G
C + (1)1 +1 C1 1 1 C 1
A

r

1 1 C 1 + C R 0 0 0 1 C1 + O G ; 1 ; 1 ,
+ C F
C A + C 1
0
A
(5.82)
and

1 1 1 1 B 1 + O G ; 1 ; 1 ,
SB = B F , G B B 1
(5.83)
respectively. Integration over C 1 yields a -function (C1 ). Hence one may drop the 0 1
term from the SC action (5.82), so that the Jacobian J, factorizes on-shell,
J, = J

sdet(1 1 ) sdet(R 0 0 )
+ O G ; 1 ; 1 ,
sdet(R 0 ) sdet( 1 1 )
(5.84)
A
A {F ; G }. Similarly,
with J sdet(
B ) and
1 1 )

sdet(1 1 ) sdet(

+
O
G
.
;
sdet F 0 , G,0 ext = sdet F , G

1
1
sdet(R 0 0 ) sdet(R 0 0 )
Therefore, formula (5.52) becomes

J sdet(F , G )
H = ln
+ O(G ),
sdet(R 0 0 )2
(5.85)
(5.86)
and by combining Eqs. (5.37), (5.73) and (5.86), one arrives at the irreducible partition function
(3.49) of Corollary 3.5, i.e.,

i
G
2
Z[1] = [d ]e h W (G ) J sdet F , G .
(5.87)
In the notation of the above proof one may summarize the bijective correspondence between
the constraints {G,0 } {G ; 1 } as follows:

P
,
[G,0 ]1N0 = G 1 1N
(5.88)
0
N N
0
G

1N0

= [G,0 ]1N0 P
Z1

N0 N0
(5.89)
Other useful observations are

P
1 X1
[R]N0 N0
[1]N0 N0 =
N N
0
1
N0 N0
0
0

P
=
P (Z P X)1 N N ,
0
0
N N
0
0 N1
1
R N
0 N0

=
(5.90)
which follows from Eq. (5.71). Therefore, one also has

P
1
[1]N0 N0 = P (Z P X)
N0 N0
N0 N0

= [P ]N0 N [P ]NN0 + (Z P X)1 N
and in particular,
429
P X1
1
[]N1 N0 ,
(5.91)

.
(5.92)
N0 N0
5.6. Non-minimal approach

With the standard non-minimal ansatz (5.26) it is mandatory to choose a non-trivial gauge
fermion = 0. In this subsection we shall study the most general non-minimal solution
1
A
X = GA0 A0 + (i h (1) )H + A0 Z A0 1 c1 + A0 U0 0 B0 C0 C0 B0 (1)B0 +1
2
+ c1 U11 1 A0 A0 c1 (1)1 + (i h (1) )A0 V A0 B0 B0 + ,
(5.93)
and therefore we may put = 0 without loss of generality. In the above Eq. (5.93), which should
not be read as a systematic Planck expansion, we have grouped together fields and antifields of
the same Planck number, i.e.,

A0 0 ; c 1 ,
A0 0 ; c1 ,
Aext A ; 1 , 1 ,
(5.94)
of Planck number 0, 1 and 1, respectively. The minus in front of c 1 in Eq. (5.94) is introduced
A
so that (A0 , B0 )[1] = B00 . The extended index A0 runs over A0 = 1, . . . , N0 + N1 . It is interesting to see how the quantum master equation determines the one-loop correction H = H(ext ; h )
in this general case. The first few consequences of the quantum master equation read
C
(GA0 , GB0 )ext = GC0 U0 0 A0 B0 ,
(5.95)
GA0 Z 1 = 0,
A

A
Z 0 1 , GB0 ext = Z A0 1 U11 1 B0 + (1)(B0 +1)(1 +1) U0 0 B0 C0 Z C0 1 + O(G),
(5.96)
A0
(ext GB0 ) (H, GB0 )ext = (1)A0 U0 0 A0 B0 (1)1 U11 1 B0 + GA0 V A0 B0 .
(5.97)
(5.98)
The extended constraints GA0 = GA0 (ext ; h ) and the extended generators Z A0 1 = Z A0 1 (ext ;
h ) are both reducible sets of functions,

Z A0 1 Z 0 1 ; 1 1 .
GA0 {G0 ; G1 },
(5.99)
430
Here 1 1 is defined as the matrix from the quadratic {c1 ; c1 } term in the action (5.93). A firststage theory has a total of 4(N0 + N1 ) fields and antifields, and we know that the Hessian for
X has half rank on stationary field configurations. Putting all the antifields to zero, the Hessian
must have full rank (= 2(N0 + N1 )) in the fieldfield quadrant. Hence, it follows that G0 is
irreducible and 1 1 has maximal rank. Eqs. (5.95) and (5.96) therefore show that
G1 = O(G0 ),
Z 0 1 = O(G0 ),
(5.100)
respectively. We arrive at the following version of Lemma 5.1:

Lemma 5.7. The path integrations over the FaddeevPopov ghost pair {c1 ; c1 } can be performed explicitly. The first-level partition function (5.31) thereby simplifies to

i
i
0
Z[1] = [d ][d0 ][d1 ]e h W + h G0 H sdet 1 1 .
(5.101)
One can also prove a version of Theorem 3.2. For that purpose, let Aext {F 0 ; G0 } and
Aext
J sdet( Bext ).
Theorem 5.8. The quantum master equation for the non-minimal X in Eq. (5.93) implies that
the quantum correction H is given by the following square root formula

J sdet(F 0 , G0 )ext
H = ln
(5.102)
+ O(G).
sdet(1 1 )2
Proof of Theorem 5.8. Eq. (5.97) can be rewritten as

ln sdet 1 1 , GB0 ext = (1)1 U11 1 B0 (1)1 U01 1 B0 + O(G),
(5.103)
which with the help of Eq. (5.98) becomes

(ext GB0 ) H ln sdet 1 1 , GB0 ext = (1)0 U0 0 0 B0 + O(G).
(5.104)
Next, one proceed as in the proof of Theorem 3.2.
Eqs. (5.101) and (5.102) leads to Eq. (5.64) in Corollary 5.4. So the non-minimal approach
agrees with the previous approach of Section 5.4.
5.7. Higher-stage reducibility and second-class constraints
Adapting
the reducible zeroth-level recipe of Ref. [2] to a first-level theory of stage s, one
introduces 2 si=0 (2i + 1)Ni first-level fields and antifields as indicated in Table 1. It is useful
to group together fields with the same Planck number assignment. Hence, we write

aext 0 ; 2 ; 4 ; . . . ,
(5.105)
a {1 ; 3 ; 5 ; . . .},

cai cii ; ci i+1 ; ci i+2 ; . . . ; cis ,
(5.106)
i = 1, . . . , s,
cai {ci,i ; ci,i+1 ; ci,i+2 ; . . . ; ci,s },
i = 1, . . . , s.
(5.107)
(5.108)
431
Table 1
First-level fields in the general reducible case. Antifields are not shown. The last field in each row is a minimal field
Grassm.
a5 + 1
a4
a3 + 1
a2
a1 + 1
a
aext
a1 + 1
a2
a3 + 1
a4
Pl
a5 + 1
6
ca5
ca4
ca3
ca2
ca1
aext
ca1
ca2
ca3
ca4
ca5
0
c1,1
1
2
c3,3
3
4
c5,5
.
..
5
.
..
c2,2
c1,2
c2,3
c1,3
c4,4
c3,4
c2,4
c1,4
c4,5
.
..
c3,5
.
..
c2,5
.
..
c1,5
.
..
Stage
Antighosts
c1 1
1
2
c2 2
c1 3
c2 3
3
4
5
c1 2
c1 4
c1 5
.
..
c2 4
c2 5
.
..
Lagr. mult.
c3 3
c3 4
c4 4
.
..
c4 5
.
..
c3 5
c5 5
.
..
Ghosts
The total space is then

A
Aext ; aext , aext ; ca1 , ca1 , ca1 , c a1 ; ca2 , ca2 , ca2 , c a2 ; . . . ; cas , cas , cas , c as ,
[1]
(5.109)

Aext A ; a , a .
(5.110)
Here the indices run
Aext = 1, . . . , 2Modd ,
a = 1, . . . , Modd N,
aext = 1, . . . , Meven ,
ai = 1, . . . , Ms Mi1 ,
(5.111)
where
Modd
Meven
Ni ,
i odd
Mi
Nj ,
Ni ,
Modd = Meven + ND ,
i=0,...,s
i even
i=1,...,s
N1 N.
(5.112)
j =1
The Planck-number operator is

s

(i + 1)cai cai i cai c ai ,
Pl = aext aext +
i=1
or equivalently,

Pl aext = 1,

Pl aext = 1,

Pl Aext = 0,
+ h (1)
[1]

Pl cai = i + 1,
Pl(cai ) = i,
Pl(h
(1) ) = 1,
Pl(h
) = 0.

Pl cai = (i + 1),
,
h (1)
(5.113)

Pl c ai = i,
(5.114)
432
Table 2
a
The simplest choice of a rectangular gauge-fixing matrix i i bi1 , i = 1, . . . , s
ai \bi1
i1
i
i+1
i+2
i+3
i+4
..
.
s
i i i1
i+1
i+2
i+3
i+4
i i+2 i+3
i i+4 i+3
The standard ansatz for the non-minimal action is

A
XD = XD,min [1]
j c1 j +
[1] +
min ; h

i i+2 i+1
i i i+1
j =1,...,s
j odd
j c1 j +
j =2,...,s
j even
i,j =2,...,s
j i odd, >0
..
..
ci1,j ci
i,j =2,...,s
j i even, 0
ci1,
c j.
j i
(5.115)
A choice of the gauge fermion reads

= ca1 1a1 aext aext +
cai iai ai1 cai1 ,
Pl( ) = 0,
(5.116)
i=2
where a simple choice for the matrices iai ai1 , i = 1, . . . , s, is indicated in Table 2. For notational
reasons it is convenient to trivially extend the matrix 1a1 aext 1a1 a0 with zero columns such
that the column index reads a0 {0 ; 1 ; 2 ; 3 ; . . .} rather than aext {0 ; 2 ; 4 ; 6 ; . . .}. The
first-level partition function is given by

a
i
i
)
(W
+X
D

D
h
Z[1]D = dD e
= dD e h (WD +XD ) a , (5.117)
,c ,c , =0
with a measure
dD = D [d ][d][dc][d c][d],
(5.118)
and a gauge-fixing surface specified by
= ca1 E(adD )1a1 aext ,

aext
ai+1
cai = E(adD ) a = cai+1 E(adD )i+1
ai , i = 1, . . . , s 1,
c i
cas = E(adD ) a = 0,
c s
= E(adD )1a1 aext aext ,

c a1 = E(adD )
ca1
= E(adD )iai ai1 cai1 , i = 2, . . . , s,

c ai = E(adD )
cai
aext = E(adD )
(5.119)
(5.120)
(5.121)
(5.122)
(5.123)
433
Table 3
a
a
A square i i bi matrix corresponding to the gauge-fixing matrices i i bi1 of Table 2. The entry corresponding to the
i1
i
i
first row and the first column is i i i i1 Zi
i
ai \bi
i i i
i+1
i+1
i+1
i
i+1
i i+2 i+1
i+2
i+2
i+3
i+4
i i i+1
i+3
i+1
i+1
i+2
i+3
i+1
i+2
i+4
.
..
s
i i+2 i+3
i i+4 i+3
i+3
i+1
i+4
..
..
a = E(adD )
= 0,
a
(5.124)
cf. the prescription (3.28).
Similar to the first-stage case the generators Zi i1 i and the gauge-fixing matrices iai bi1
a
give rise to FaddeevPopov matrices i i bi as indicated in Table 3. The pertinent rank conditions
ai
to the gauge-fixing matrices i bi1 are such that the sequence of matrices ai bi , i = 1, . . . , s,
becomes invertible. Explicit rank conditions for the iai bi1 matrices for theories of stage 1, 2
and 3 can be found in the original paper [2].
Lemma 5.9. The path integrations over the ghost pairs {ca1 ; ca1 ; . . . ; cas ; cas } can be performed
explicitly. The first-level partition function thereby simplifies to

s
(1)i

i
i
aext
sdet ai bi
, (5.125)
ZD[1] = [d ][d][d]D e h WD + h G,aext H a
i=1
where G,aext are the -deformed gauge-fixing constraints,

G,aext {G,0 ; G,2 ; G,4 ; G,6 ; . . .},
(5.126)
G,0 G0 + 1 11 0 ,
G,i i+1 1 i+1 i , i = 2, . . . , s,
(5.127)
i even,
(5.128)
and G0 are the original reducible constraints.

Sketched Proof of Lemma 5.9. The gauge-fixed action

h (1)
S=
WD + XD = S0 + SFP + V ,
h
splits into a constant part

h (1)
WD + G,aext aext + (i h (1) )H
h
that is independent of the ghosts and antighosts {cai ; cai }, a FaddeevPopov part
S0
SFP
s

i=1
cai ai i bi cbi
(5.129)
(5.130)
(5.131)
434
that is quadratic in {cai ; cai }, and an part V that contains all interaction terms, tadpole terms and
terms quadratic in the antighosts cai . The multiplicity condition that generalizes the rule (5.43)
states, that for each term in V there exists an i = 1, . . . , s, such that there are fewer ghosts cai
than antighosts cai . 2
Aext

Aext
aext
a
Aext { A ; , a }.
Next, let JD,, sdet( ,Bext
), ,
a
{F ; G,aext ; } and
Then
Theorem 5.10. The quantum master equation for XD,min implies that the quantum correction H
is given by the following square root formula:

s

a (1)i
JD,, sdet(Fa ext , G,bext )D,ext
sdet i bi
H = ln
+ O(G0 ; ).

D
, =0 i=1
(5.132)
Aext
In addition, let JD sdet( Bext ) and Aext {F aext ; Gaext ; a }. Then
Corollary 5.11. The partition function of stage s

i
G
= [dext ]e h WD a a (Gaext ) D JD sdet(F aext , Gbext )D,ext
ZD[1]
(5.133)
is independent of Gaext s that are in involution with respect to the ( , )D,ext -bracket.
Note that the alternating product of superdeterminants [16] has cancelled out of the final
expression. A proof of Theorem 5.10 will appear elsewhere.
5.8. Non-minimal approach for higher stages
We consider the most general non-minimal solution
XD = GA0 A0 + (i h (1) )H + A0 Z A0 A1 cA1 +
s1

cA
Z Ai Ai+1 cAi+1
i
i=1
1
A
+ A0 U0 0 B0 C0 C0 B0 (1)B0 +1 +
2
cA
U A1
C0 cBi (1)Bi +i+1
i i Bi C0
i=1
+ (i h (1) )A0 V A0 B0 B0 + .
In Eq. (5.134) we have grouped together fields of the same Planck number, i.e.,
a

ext
cai
Ai
A0
,
c
,
cAs cas ,
a
1
i+1
c
c

cai
A0 aext ,
cas .
cA
, i = 1, . . . , s 1,
cA
s
i
ca1
cai+1
(5.134)
(5.135)
The first few consequences of the quantum master equation read

C
(GA0 , GB0 )D,ext = GC0 U0 0 A0 B0 ,
(5.136)
435
GA0 Z A0 A1 = 0,
(5.137)
Z
Ai Z Ai+1 = O(G),
A

Ai+1
Z i Bi+1 , GB0 D,ext = Z Ai Ai+1 Ui+1
Bi+1 B0
(5.138)
Ai1
Ai
(1)(B0 +1)(Ci +Bi+1 ) UiAi Ci B0 Z Ci Bi+1 + O(G),

s

(1)Ai +i UiAi Ai B0 + GA0 V A0 B0 .
(D,ext GB0 ) (H, GB0 )D,ext =
(5.139)
(5.140)
i=0
Ignoring at first the second-class constraints, the Hessian for X has half rank on stationary field
configurations. Putting all the antifields to zero, the Hessian must have full rank in the fieldfield
quadrant. It follows that the other three quadrants of the Hessian must vanish on stationary field
configurations. In the presence of second-class constraints the same reasoning can be used in
the physical subsector. We conclude that the extended constraints GA0 = GA0 (ext ; h ) and the
A
extended generators Z Ai1 Bi = Zi i1 Bi (ext ; h ) are both reducible sets of functions,

O(G; ) O(G; )
O(G; )
Ai1
As1
GA0 [ Gaext Ga1 ],
Z
,
,
Z
Bi =
Bs =
ai bi
as bs
O(G; )
Ga1 = O(Gaext ; ), i = 1, . . . , s 1.
(5.141)
Here ai bi is defined as the matrix from the quadratic {cai ; cbi } term in the action (5.134). Hence
it follows that Gaext is irreducible and ai bi , i = 1, . . . , s, have maximal rank. We arrive at the
following version of Lemma 5.9:
Lemma 5.12. The path integrations over the ghost pairs {ca1 ; ca1 ; . . . ; cas ; cas } can be performed
explicitly. The first-level partition function thereby simplifies to

s
(1)i

i
i
aext
sdet ai bi
.
ZD[1] = [d ][d][d]D e h WD + h Gaext H a
(5.142)
i=1
Aext
One can also prove a version of Theorem 5.10. For that purpose, let JD sdet( Bext ),
Aext {F aext ; Gaext ; a }, and Aext { A ; a , a }. Then
Theorem 5.13. The quantum master equation for the non-minimal XD in Eq. (5.134) implies
that the quantum correction H = H(ext ; h ) is given by the following square root formula:

s
a (1)i
JD sdet(F aext , Gbext )D,ext
i
sdet bi
H = ln
(5.143)
+ O(G; ).
D
i=1
Proof of Theorem 5.13. Eq. (5.139) can be rewritten as

ai
ln sdet ai bi , GB0 D,ext = (1)ai Uiai ai B0 (1)ai Ui1
which with the help of Eq. (5.140) becomes

s

a
i
i
(D,ext GB0 ) H +
(1) ln sdet bi , GB0
i=1
ai B0
+ O(G),
(5.144)
= (1)aext U0aext aext B0 + O(G).

D,ext
(5.145)
436
Next, one proceed as in the proof of Theorem 3.2.
Eqs. (5.142) and (5.143) leads to Eq. (5.133) in Corollary 5.11. So the non-minimal approach
agrees with the approach of Section 5.7.
6. Higher-level formalism
6.1. Recursive construction
In the irreducible n th-level formalism one introduces N Lagrange multipliers (n) of Grassmann parity + n and N antifields (n) , which we collectively call the n th-level fields.
The phase space variables

A
A
[n]
(6.1)
[n1]
; (n) , (n)
(n)
(0)
for the n th-level formalism thus consists of fields of levels n. The idea is roughly that the n thlevel Lagrange multipliers should gauge-fix the (n 1) th-level first-level antifields (n1) ,
although this is just one gauge-fixing choice out of infinitely many.
A
In the n th-level formalism, one first lifts the previous (n 1) th-level phase space [n1]
to a
A
A =
fully covariant status, i.e., one allows for general coordinate transformations [n1] [n1]
A (
[n1]
[n1] ) that preserve the Planck number symmetries of the previous levels, so that
[n1] ; h
the (n 1) th-level odd Laplacian becomes of the covariant form
l
l
(1)A
AB
[n1] =
E
[n1]
[n1]
A
B
2[n1] [n1]
[n1]
(6.2)
AB
AB (
with a symplectic metric E[n1]
= E[n1]
[n1] ) and a measure density [n1] =
[n1] ; h
[n1] ([n1] ; h [n1] ). Secondly, one defines a n th-level odd Laplacian
[n] [n1] + (1)
(n)
(n) (n)
(6.3)
a n th-level Planck constant h (n) , and a n th-level Planck operator

Pl(n) = (n) (n) , [n] + h

(n)
h
(6.4)
(n)
At even levels n, one has the following n th-level generalization of Principle 3.1:
Principle 6.1. The W[n] -action should satisfy three principles:
W
[n]
) = 0.
1. Planck number conservation: i = 1, . . . , n Pl(i) ( h (n)
i
2. The quantum master equation: [n] exp[ h (n)
W[n] ] = 0.
A , i.e., (n + 1)N in the

3. The Hessian of W[n] has rank equal to half the number of fields [n]
irreducible case.
The n th-level partition function is defined as

i
Z[n]
= [d[n1] ][d(n) ][n1] e h (W[n] +X[n1] )
(n) =0
(6.5)
437
with
i
e h W[n] = e[[n] , ] e h W[n] .
(6.6)
does not depend on the gauge

It follows from standard arguments that the partition function Z[1]
fermion . At odd levels n, there is a similar story where the roles of W and X are exchanged,
up to a relative sign,
i
e h X[n] = e[[n] , ] e h X[n] .
(6.7)
Again, the Planck number conservation limits the number of possible

tions in the action

W[n] = G(n1) (n) + (i h (n) )H[n1] + O (n) .
n th-level
structure func(6.8)
When one decomposes the n th-level quantum master equation in terms of the above (n 1) thlevel structure functions one generates a tower of equations; the first few equations read:
(G(n1) , G(n1) )[n1] = G(n1) U(n1) ,
(6.9)
(n)
([n1] G(n1) ) (H[n1] , G(n1) )[n1] = (1) U(n1)

+ G(n1) V(n1)
, (6.10)
([n1] H[n1] ) + (H[n1] , H[n1] )[n1] = V(n1)

(6.11)
G(n1) G
(n1) .
2
There is in general no covariant on-shell expression for the higher-level H[n1] with n 2,
although partial results exist [5].
6.2. Recursive reduction
The reduction from n th to (n 1) th level can be demonstrated as follows:
1. Go to coordinates

A
A
[n]
[n2]
; (n1) , (n1) ; (n) , (n)
(6.12)
with Darboux coordinates at the last two levels n and n 1, and with a measure density
[n1] [n2] .
h
2. Choose the gauge-fixing functions G(n1) = (n1)

h (n1) , so that U(n1) = 0. In this
case Eq. (6.10) yields that
H[n1] = K[n1] (n1) V(n1)

(n1) ,
(6.13)
where
V(n1)
1

dt V(n1)
[n2] ; t(n1) , (n1) ; h [n1] ,
(6.14)
and where the integration constant K[n1] = K[n1] ([n2] ; (n1) ; h [n1] ) is independent
of (n1) .
438
3. Define the (n 2) th-level action as

W[n2] = (i h (n2) )H[n1]
(n1) =0
(6.15)
if n is even. (If n is odd, exchange X W .) The action W[n2] defined this way does
not depend on h (n1) , because of Planck number conservation Pl(n1) (W[n2] ) = 0, and
moreover, W[n2] satisfies the (n 2) th-level quantum master equation

i
W[n2] = 0,
[n2] exp
(6.16)
h (n2)
as a result of (one of the consequences of) the n th-level quantum master equation (6.11).
Note that other consequences of the n th-level quantum master equation would in general
impose other conditions on W[n2] . (If the (n 2) th-level action W[n2] is already known
independently, Eq. (6.15) should be interpreted as a boundary condition.)
4. Choose the gauge fermion independent of the n th-level Lagrange multiplier antifields
(n) . Then is independent of the Lagrange multipliers (n) as well, because of Planck
number conservation Pl(n) ( ) = 0. (Here we have implicitly assumed that there are only
A ; h } present inside .) Using the symmetry
non-negative powers of n th-level objects {(n)
(n)
(1.1) of the [n] -operator, one derives
i

i
= [d[n] ][n2] e h W[n] e[[n] , ] (n) e h X[n1]

Z[n]

i
i
= [d[n] ][n2] e h ((n1) (n) +W[n2] ) (n) e[[n] , ] e h X[n1]

i
= [d[n2] ][d(n1) ][n2] e h (W[n2] +X[n1] ) =0 = Z[n1]

,
(6.17)
(n1)
which completes the reduction step.

The n th-level formalism can also be set up in the case of reducible gauge-fixing constraints
and second-class constraints. In the reducible case one introduces the relevant stages of minimal
and non-minimal fields, in accordance with the general fieldantifield prescription, as we saw in
Section 5.
Hence, the multi-level formalism consists in recursively building master actions W[0] W ,
X[1] X, W[2] , X[3] , . . . . By zig-zagging through the W - and X-parts it becomes a simple matter
to re-use the constructions of the previous levels and to create a manifestly gauge-independent
formalism.
7. Conclusion
Driven by the wish to develop the Lagrangian quantization program into its most general and
axiomatized formulation, we have in this paper focused on three aspects. First, we have given
a more fully geometric description of the multi-level formalism, in all generality. Second, we
have explored the new symmetric formulation which puts the action W on the same footing
as the gauge-fixing X to yield the full quantum action that enters in the Boltzmann factor of
the functional integral. One particular aspect of this symmetry concerns the algebras behind the
master actions W and X. On one hand there is the gauge-generating algebra behind W , which
is associated with gauge symmetries of the classical action, and is known to accommodate both
439
open and reducible gauge-algebras. On the other hand, there is the gauge-fixing algebra behind
X, which is a quantum mechanical feature, with no analog at the level of the classical theory,
that usually carries just an irreducible algebra. We have in detail demonstrated how to permit
a reducible gauge-fixing algebra, and calculated the associated measure or gauge volume from
first principles, namely by solving the quantum master equation. Several consistency checks on
the formalism were performed along the way by reduction methods. Third, we included an extensive discussion of antisymplectic second-class constraints in the fieldantifield formalism,
and demonstrated manifest invariance under reparametrization of the second-class constraints.
Second-class constraints hold surprisingly many features in common with the gauge-fixing constraints G , and often do they appear side-by-side in the formulas. In this way, the second-class
constraints merge effortlessly with the multi-level formalism, even in the case of reducible gaugefixing algebras.
While there are still many more aspects of the new and more axiomatic formulation of the
Lagrangian quantization prescription that need to be explored, it is already now apparent that
there is a rich and beautiful algebraic structure behind. This algebra has its root in the one single
object from which all is derived: the Grassmann-odd nilpotent -operator, and its associated
quantum master equation.
Note added
After the paper appeared on the archive we became aware of Ref. [17] where a reducible
gauge-fixing X arose in a superfield context.
Acknowledgements
I.A.B. and K.B. would like to thank the Niels Bohr Institute for the warm hospitality extended
to them there. The work of I.A.B. is supported partially by grants RFBR 05-01-00996 and RFBR
05-02-17217, and the work of K.B. is supported partially by the Ministry of Education of the
Czech Republic under the project MSM 0021622409.
References
[1] I.A. Batalin, G.A. Vilkovisky, Phys. Lett. B 102 (1981) 27.
[2] I.A. Batalin, G.A. Vilkovisky, Phys. Rev. D 28 (1983) 2567;
I.A. Batalin, G.A. Vilkovisky, Phys. Rev. D 30 (1984) 508, Erratum.
[3] I.A. Batalin, G.A. Vilkovisky, Nucl. Phys. B 234 (1984) 106.
[4] I.A. Batalin, I.V. Tyutin, Int. J. Mod. Phys. A 8 (1993) 2333.
[5] I.A. Batalin, I.V. Tyutin, Mod. Phys. Lett. A 8 (1993) 3673;
I.A. Batalin, I.V. Tyutin, Mod. Phys. Lett. A 9 (1994) 1707;
I.A. Batalin, I.V. Tyutin, in: American Mathematical Society Translations, Series 2, vol. 177, 1996, p. 23.
[6] I.A. Batalin, R. Marnelius, A.M. Semikhatov, Nucl. Phys. B 446 (1995) 249.
[7] I.A. Batalin, K. Bering, P.H. Damgaard, Phys. Lett. B 389 (1996) 673.
[8] B.L. Voronov, I.V. Tyutin, Theor. Math. Phys. 50 (1982) 218;
B.L. Voronov, I.V. Tyutin, Theor. Math. Phys. 52 (1982) 628.
[9] O.M. Khudaverdian, Th. Voronov, Lett. Math. Phys. 62 (2002) 127;
O.M. Khudaverdian, Contemp. Math. 315 (2002) 199;
O.M. Khudaverdian, Commun. Math. Phys. 247 (2004) 353.
[10] A. Schwarz, Commun. Math. Phys. 155 (1993) 249.
[11] O.M. Khudaverdian, A.P. Nersessian, Mod. Phys. Lett. A 8 (1993) 2377.
440
[12] K. Bering, Ph.D. Thesis, Uppsala University, Sweden, 1997;

I.A. Batalin, I.V. Tyutin, Theor. Math. Phys. 114 (1998) 198.
[13] O.M. Khudaverdian, prepared for the International Workshop On Geometry And Integrable Models, Dubna, Russia, 48 October 1994, in: P.N. Pyatov, S. Solodukhin (Eds.), Geometry and Integrable Models, World Scientific,
Singapore, 1996.
[14] I.A. Batalin, K. Bering, P.H. Damgaard, Phys. Lett. B 408 (1997) 235.
[15] I.A. Batalin, R. Marnelius, Nucl. Phys. B 511 (1998) 495.
[16] A. Schwarz, Lett. Math. Phys. 2 (1978) 247;
A. Schwarz, Commun. Math. Phys. 67 (1979) 1.
[17] A.A. Reshetnyak, Russ. Phys. J. 47 (2004) 1026.
A gradient flow for worldsheet nonlinear sigma models

T. Oliynyk a , V. Suneeta b,c , E. Woolgar c,d,
a Max-Planck-Institut fr Gravitationsphysik (Albert Einstein Institute), Am Mhlenberg 1, D-14476 Potsdam, Germany
b Department of Mathematics and Statistics, University of New Brunswick, Fredericton, NB, E3B 5A3, Canada
c Theoretical Physics Institute, University of Alberta, Edmonton, AB, T6G 2J1, Canada
d Department of Mathematical and Statistical Sciences, University of Alberta,
Edmonton, AB, T6G 2G1, Canada

Abstract
We discuss certain recent mathematical advances, mainly due to Perelman, in the theory of Ricci flows
and their relevance for renormalization group (RG) flows. We consider nonlinear sigma models with closed
target manifolds supporting a Riemannian metric, dilaton, and 2-form B-field. By generalizing recent mathematical results to incorporate the B-field and by decoupling the dilaton, we are able to describe the 1-loop
-functions of the metric and B-field as the components of the gradient of a potential functional on the
space of coupling constants. We emphasize a special choice of diffeomorphism gauge generated by the
lowest eigenfunction of a certain Schrdinger operator whose potential and kinetic terms evolve along the
flow. With this choice, the potential functional is the corresponding lowest eigenvalue, and gives the order
correction to the Weyl anomaly at fixed points of (g(t), B(t)). The lowest eigenvalue is monotonic along
the flow, and since it reproduces the Weyl anomaly at fixed points, it accords with the c-theorem for flows
that remain always in the first-order regime. We compute the Hessian of the lowest eigenvalue functional
and use it to discuss the linear stability of points where the 1-loop -functions vanish, such as flat tori and
K3 manifolds.
E-mail addresses: todd.oliynyk@aie.mpg.de (T. Oliynyk), suneeta@math.unb.ca (V. Suneeta),

ewoolgar@math.ualberta.ca (E. Woolgar).
doi:10.1016/j.nuclphysb.2006.01.036
442
T. Oliynyk et al. / Nuclear Physics B 739 [FS] (2006) 441458
1. Introduction
A standard approach to renormalization group (RG) flow for a quantum field theory consists of
deriving differential equations governing the behaviour of the coupling constants under changes
in the renormalization scale ([13]). These RG flow equations are written in terms of -functions
which are components of a vector field tangent to the flow on the space of coupling constants
of the theory. The -functions can be computed in a loop expansion. For the worldsheet (i.e.,
2-dimensional) bosonic sigma model, the loop expansion parameter is , the square of the string
scale.
A basic result is Zamolodchikovs c-theorem [4,5], which implies that certain RG flows are irreversible. This theorem asserts the existence of a function on the space of coupling constants of
certain 2-dimensional quantum field theories called the C-function, which decreases monotonically along any renormalization group trajectory from an unstable to a stable fixed point, and
equals the Weyl anomaly (the central charge of the Virasoro algebra) at fixed points. The ctheorem was extended to nonlinear sigma models with compact target spaces by Tseytlin [6].
The C-function obeys the monotonicity formula dC/dt = (, ) for t a parameter along the
flow, a nonnegative quadratic form on the space of coupling constants, and the array of
-functions of the coupling constants. The c-theorem is not contingent on the loop expansion
for .
A longstanding question is whether RG flow is a gradient flow: is the vector field defined
by the -functions orthogonal to level surfaces of a potential function on the space of coupling
constants. Recent advances in mathematics have shed light on this matter. Consider the special
case of a nonlinear sigma model whose target space is purely gravitational (in particular, the
antisymmetric B-field is absent), and with replaced by its 1-loop (order ) approximation.
Then the 1-loop RG flow for the target space metric is known in the mathematics literature as
Ricci flow [7]. This flow was long known to be gradient on a space of coupling constant endowed
with a metric of indefinite sign [8]. But the above discussion of the c-theorem and suggests
there may be a positive-definite metric for 1-loop gradient flow, from which one could obtain a
monotonicity formula. Perelman [9] has now shown that this flow on closed (target) manifolds
of arbitrary dimension is in fact a gradient flow on a space of coupling constants with positivedefinite metric.
Then what do these recent mathematical advances mean for RG flows of sigma models more
generally, when B is not held to zero? This paper is intended to address this issue. We restrict
our attention to the first of many potentially relevant results announced in [9], the gradient nature
of the flow for g and related monotonicity. We endeavour to set out this result in some detail in
language appropriate to the RG setting. Our first task is therefore to generalize it to incorporate
not only the target manifolds Riemannian metric g but also the antisymmetric 2-form field B
and dilaton (though the dilaton can often be ignored by decoupling it from the system using a
suitably chosen diffeomorphism). We find that for sigma models whose target space is a closed
manifold, the order RG flow of (g, B) is gradient on a space of coupling constants with
positive-definite metric, and thus the flow is irreversible.1 The irreversibility argument is easiest
when we choose a certain diffeomorphism gauge along the flow, and then we call the potential
function . We note that the full RG flow (i.e., not the order truncation) on closed target
manifolds is known to be irreversiblethis is a consequence of the c-theorem. However, in
1 Elsewhere, we have extended Perelmans work to noncompact asymptotically flat manifolds with somewherenegative scalar curvature and zero B-field [10].
443
the absence of knowledge of the higher loop corrections to the -function, it is not practically
possible to compute the C-function on closed target manifolds. Our result can be considered
valid when the 1-loop truncation of the -function is a good approximation. In this case, we can
explicitly compute the potential function that generates the 1-loop RG flow without requiring
knowledge about regimes where stringy (higher order in ) corrections become important.
We find the value of the potential function at fixed points to be nonnegative, and zero at any
Ricci-flat fixed point. This raises the undesirable possibility that a Ricci-flat fixed point such as a
flat torus might flow to a non-Ricci-flat one. To preclude this possibility, we compute the Hessian
of the potential and use the resulting second variations to discuss the linear stability, as well as
the rigidity (or isolation), of Ricci-flat fixed points. We confirm linear stability for the particular
examples of flat tori and K3 manifolds. Our considerations lead us to briefly discuss the possible
rigidity of Ricci-flat perturbative string vacua, at which the -functions vanish to all orders in
. This may be of interest when viewed in the light of investigations into the topology of the
configuration space of string theory [11,12].
In Section 2, we discuss the sigma model under consideration, recalling its RG flow equations
at order and decoupling the dilaton from the flow of g and B.
Section 3 is the heart of the paper, particularly the first subsection, wherein we describe and
generalize Perelmans approach. Perelmans remarkable insight (see also [13,14]) is that the Ricci
flow of g, modified by the correct choice of diffeomorphism, becomes the gradient flow of a potential which is actually the lowest eigenvalue of a certain Schrdinger operator on the target
manifold,2 and we show that this carries over mutatis mutandis to the (g, B) flow as well. This
leads to an easy proof of the absence of periodic or homoclinic behaviour (i.e., the irreversibility) of the 1-loop flow. In the second subsection, we evaluate this eigenvalue at a fixed point.
We show that it is 0 and is 4 times the order correction to the Weyl anomaly at fixed
points. The final subsection contains a brief aside on the gradient nature of RG flow with an arbitrary diffeomorphism; i.e., not a diffeomorphism chosen as above to connect to the Schrdinger
problem.
Section 4 contains a derivation of the second variation of the potential function at a fixed
point. This section builds on a similar result for Ricci flow presented in [15]. In Section 5 we
apply the second variation formula to discuss linear stability of fixed points, including some
particular Ricci-flat examples. We conclude Section 5 with some speculative remarks concerning
the potential applicability of our results to an issue in string theory. Section 6 contains remarks
on higher-order flows and the C-theorem.
Appendix A contains some calculations related to Section 3 that we believe would unnecessarily clutter the main text. That said, we have tried to provide a certain level of calculational
detail, especially when such detail has not been provided in the mathematics literature.
Throughout, the target manifold is closed (i.e., compact and without boundary) and is assumed
to carry a positive-definite Riemannian metric. Our RG flow equations agree with those that
appear in [16].
2. RG flow of the worldsheet nonlinear sigma model
The 2-dimensional, or worldsheet, nonlinear sigma model is a quantum field theory of maps
X i from a 2-dimensional Riemannian manifold (, h), the worldsheet, to another Riemannian
2 In fact, the relevance of this eigenvalue problem for RG flow was proposed in [13], in the special case of zero B-field
and a closed, 2-dimensional target manifold. It was noticed in [14] that the work of [9] validates this proposal.
444
manifold (M, g), the target manifold or target space, which herein we take to be a closed manifold of dimension at least 3. We let denote the volume element on the worldsheet and let R(h)
be the worldsheet scalar curvature. The sigma model action is (using coordinates = ( 1 , 2 )
on )

1
S = d 2
(2.1)
hh gij X i X j + Bij X i X j hR(h) ,
where gij , Bij = Bj i , and are the target space metric, B-field, and dilaton, respectively.
This model describes the motion of a bosonic string in a background wherein the massless string
modes have acquired vacuum expectation values gij , Bij , and . The action is invariant under
reparametrizations and conformal rescalings on the worldsheet and under the addition B B +
d of an exact form d to B, which is a target space 2-form. The gauge-invariant 3-form field
strength for B is H := dB.
Cut-off independence of the regulated quantum theory leads to renormalization group flow
equations (see, e.g., [16])

gij
1
kl

(2.2)
= Rij + 2i j Hikl Hj ,
t
4

H
1
(2.3)
= LB H dH, grad ,
t
2

1
1
(2.4)
= A + ||2 + |H |2 ,
t
2
24
where is the Laplacian, LB H := (d + d)H is the LaplaceBeltrami operator acting on
the 3-form H := dB, A is a constant whose value depends on the target manifold dimension (and
the number of ghost fields when they are present), and t is the log of the renormalization scale.3
Eq. (2.3) is actually derived by taking the curl of the flow equation

Bij
1
= k Hkij Hkij k + i j j i
t
2

1
=
(2.5)
H + H, grad + d .
2
ij
The terms arise because B is determined only up to the exterior derivative of an arbitrary 1form . Then B
t acquires a contribution which, because exterior differentiation commutes with
t , can be written as the exterior derivative of the 1-form := t .

These equations are written with respect to a coordinate basis fixed with respect to t . In a basis
that changes with t , extra terms will be introduced into the evolution equations. We will exploit
this now to decouple , and later to demonstrate the monotonicity formula.
Pulling back by the t-dependent diffeomorphism t generated by the vector field grad
adds a Lie derivative term to each of the flow equations. For example, the left-hand side of
the equation for gij becomes t t gij = ( t g ij gij ) t = ( t g ij 2 i j ) t , where
g ij := t gij . The right-hand side is natural under diffeomorphisms and becomes, schematically,
3 We do not presume t to be in any way related to physical time here, although it has been conjectured that RG flow
could model real time evolution in certain situations.
445
t (RHS) = RHS t . The other evolution equations transform similarly, yielding

g ij
1
(2.6)
= R ij H ikl H j kl ,
t
4

H
= LB H ,
(2.7)
t
2

1
1
(2.8)
= A + + |H |2 .
t
2
24
We say these equations are expressed in Hamilton gauge in recognition of the relationship of
(2.6) to the Ricci flow equation of Hamilton [7]. Notice that has now decoupled from the
evolution equations for g ij and H ij k .
It will prove convenient to express these equations in an arbitrary t -dependent coordinate system; i.e., to pull back by an arbitrary t -dependent diffeomorphism. To do so, we simply add Lie
derivative terms to each equation. We are interested in particular in diffeomorphisms generated
by the gradient of a scalar, so for later convenience we choose to write the generator as
where is arbitrary. Then the evolution equations become

gij
1
g
= Rij + i j Hikl Hj kl =: ij ,
(2.9)
t
4

Hij k

(2.10)
=
LB H dH, grad ij k =: ijHk ,
t
2

1
= A +
+ |H |2 =: .
(2.11)
t
2
12
The right-hand sides of these equations define what are called -functions. We have dropped the
tildes now, but note that there is of course a distinction between quantities such as gij appearing
in Eqs. (2.2)(2.4) and those in (2.9)(2.11). Namely, the former are obtained from the latter by
choosing the gauge = 2. Hamilton gauge (denoted by the tildes) is the choice = 0.
In the arbitrary gauge, (2.5) becomes

Bij
k
1
=
Hkij Hkij k + 2 [i j ] Bj ]k k ,
(2.12)
t
2
2
with square brackets on indices indicating antisymmetrization. However, we will fix the evolution
of the internal gauge by imposing the flow equation

Bij
k
=
Hkij Hkij k =: ijB .
(2.13)
t
2
This forces

1
i := j Bj k k
(2.14)
2
to be a closed 1-form. We will now show that the flow given by (2.9), (2.13) is a gradient flow.
3. The gradient flow
3.1. The monotonicity formula
In this section we elucidate the first two sections of [9] (see also the detailed notes [17]) and
generalize that work to sigma models with B-field.
446
The space of coupling constants G (gij (x), Bij (x), (x)), x M, factors as G = G
C (M), where (gij (x), Bij (x)) G. The C factor4 will be used to accommodate both the
dilaton and the diffeomorphism generating function , but it is important not to equate these.5
Consider now a section (gij , Bij , ) of G, where we take g, B, and to be related by (2.9) and
(2.13) respectively, but the t -evolution of is arbitrary. If a choice of t -evolution for is made,
Eqs. (2.9) and (2.13) will then determine an evolving section (g(t), B(t)) in G. Because the
dilaton is decoupled, we do not need to compute its t -evolution simultaneously. Rather, we can
compute (t) a posteriori from (2.11), once the t-evolutions for g(t) and B(t) are determined.
In this way, each choice of evolution (t) gives a distinct t -evolving parametrization of the
coupling constants (gij (t, x), Bij (t, x), (t, x)).
In Section 3.3 we describe, for each choice of parametrization of the coupling constants,
a potential that generates a gradient flow for (g(t), B(t)) on the space G G endowed with a
natural choice of inner product. However, by choosing in a certain very natural way (which
we dub Perelman gauge), the resulting gradient flow is particularly useful, and it is that choice
which we concern ourselves with first. Define the functional

1
2
2
R + || |H | .
F [g, B, ] := dV e
(3.1)
12
M
Integrating by parts, we can write

1
/2
2
F [g, B, ] = dV e
R |H | 4 e/2 .
12
(3.2)
M
1
Now let u(t, x), x M, be the lowest eigenfunction of the Schrdinger operator R 12
|H |2
4. The operator depends on t through the flowing metric g(t), and thus so do the eigenfunctions.
Normalize u (and the other eigenfunctions) to unity:

u2 dV = 1.
(3.3)
M
Since the lowest eigenfunction u has no nodes, it has a well-defined logarithm. We use this to
define a function P by
eP (t,x)/2 := u(t, x).
Then, for the eigenvalue belonging to u, we have

1
1
R |H |2 4 u =: u R |H |2 + 2P |P |2 = .
12
12
(3.4)
(3.5)
Clearly,
the choice P = minimizes the functional (3.2) over all C (M) functions obeying
dV = 1. Thus on G there is a new functional [g, B], equal in value at (g(t), B(t)) to
Me
(t), defined by

F [g, H, ] (t) = g(t), B(t)
[g, B] := inf
(3.6)
{|
e dV =1}
4 According to the way we have defined points of G, strictly speaking it is not C (M) that splits off from G but rather
a trivial bundle whose sections belong to C (M).
5 That would be a gauge choice; e.g., the choice = 2 produces Hamilton gauge (2.2)(2.4).
447
and the infimum is realized by the choice = P .

We can compute the gradient of [g, B] on G by evaluating the first variation of F [g, B, ],
requiring to be a solution = P of (3.5) all along the variation. The first step is to compute
the free variation in which is not constrained. This variation is familiar in physics. If we
momentarily think of as the dilaton6 for purposes of computing this variation, then F is the
low energy effective action in string theory and its variational derivative is well known. For
completeness, we provide a derivation in Appendix A; cf. Eq. (A.9). The variation gives

1 i
dF
ij
i j
j kl gij
=
R + H kl H
ds
4
s
M

1 ij gij
1
2
2
g
+ R |H | + 2 ||
12
2
s
s
Bij
1
+ k H kij H kij k
(3.7)
e dV .
2
s
Now impose the constraint that, for each value of s along the variation, is not freely varied
but rather is fixed to obey (3.4). That is, (s) = P (s) = 2 log u(s), where u(s) is the lowest
eigenfunction of the Schrdinger operator determined by the varied metric and B-field. Then
(3.7) becomes the first variation formula for [g, B]. As well, on the right-hand side of (3.7) we
use first (3.5) and then (3.3) and (3.4) to write

P P
1
1 ij gij
g
e dV
R |H |2 + 2P |P |2
12
2
s
s
M

d
P P
1 ij gij
eP dV = 0.
g
e dV =
=
(3.8)
2
s
s
ds
M
Thus, the middle line of (3.7) vanishes and we are left with

d
1 i
ij
i j
j kl gij
=
R P + H kl H
ds
4
s
M
Bij P
1
e dV .
+ k H kij H kij k P
(3.9)
2
s
Now we can consider the elements of G to be 2-index tensors with symmetric part g and skew
part B. Then the inner product is given by

T , T := eP g ik g j l Tij Tkl dV
M

eP g ik g j l [Sij Skl
+ Aij Akl ] dV ,
(3.10)
where S is the symmetric part of T T G and A is the antisymmetric part. Then (3.9) is the
B
directional derivative T , Grad of in the direction T := ( g
s , s ). Then we can read off the
gradient.
6 Keeping in mind of course that it is , not , that obeys the dilaton RG flow; indeed, when is constrained in the
manner of Section 3.3, it can be shown to evolve in t according to a backwards parabolic evolution equation.
448
Proposition 3.1. The 1-loop RG flow for (g, B), with diffeomorphism gauge fixed to be a
solution P of (3.5), is the gradient flow in G generated by the potential [g, B]:

(R + P 1 H H kl )
ij
i j
gij
4 ikl j
(3.11)
= Grad [g, B],

k
k
t Bij
[ Hkij Hkij P ]
2
and (t) is monotone increasing along the gradient flow:

2

d
Rij + i j P 1 Hikl Hj kl + 1 k Hkij Hkij k P 2 eP dV .
=

dt
4
4
(3.12)
Furthermore, fixed points of (3.11) (where H = dB) are stationary points of .

Proof. The gradient formula (3.11) can be read off from (3.9). To obtain (3.12), consider the
special case in (3.9) of a variation in (g, B) produced by evolving (g, B) along the RG flow. That
B
is, let ( g
s , s ) in (3.9) be given by the flow equations (2.9), (2.13) with s = t . Finally, for fixed
points of (3.11), the right-hand side of (3.12) vanishes. 2
Corollary 3.2. is monotonic along 1-loop RG flow (2.9), (2.13).
Proof. Starting from the same initial data, the resulting solutions of the gradient flow (3.11) and
the RG flow with arbitrary (2.9), (2.13) are related by at worst a time-dependent diffeomorphism. But (t) is diffeomorphism-invariant, and monotonic along the gradient flow. 2
We now discuss briefly the absence of periodic 1-loop RG flows. We will call a solution
of the flow (2.9), (2.13) a breather if it is periodic up to gauge and diffeomorphism; i.e., if
there is a diffeomorphism , a 1-form , and parameter values t1 < t2 such that (g(t1 ), B(t1 )) =
( g(t2 ), B(t2 ) + d). A solution that is not a breather is a homoclinic orbit if it is eternal
(i.e., defined for all t (, )) with (g(t), B(t)) converging to (g0 , B0 ) for t and to
( g0 , B0 + d) for t +.
Proposition 3.3. There are no periodic or homoclinic orbits of the 1-loop RG flow other than the
fixed points of (3.11).
Proof. For a periodic orbit of the flow (2.9), (2.13), there will be some t1 < t2 such that (t1 ) =
(t2 ) =: . Then by monotonicity, (t) = for all t [t1 , t2 ]. But by (3.12), this can only
happen if the right-hand side of (3.11) vanishes throughout [t1 , t2 ]. This is the condition for a
fixed point (with = P ). For the homoclinic case, note that the sequence (nT ) (nT ),
n Z, T > 0, is increasing. However, for a homoclinic orbit, this sequence must converge to
zero. Therefore (nT ) (nT ) = 0 for all n. Setting t1 = nT , t2 = nT , we see as before that
the flow on [t1 , t2 ] = [nT , nT ] is at a fixed point. But we can take n arbitrarily large. 2
Note that this result implies that if the flow equations with any potential have a fixed point
in which the right-hand sides of (2.9) and (2.13) vanish, then the diffeomorphism generator
must be a solution P of (3.5).
Perelman, in Section 2 of [9], was able to go farther. Using as we have defined it, but of
course with B = 0, he was able to prove the nonexistence of nontrivial expanding breathers,
449
metrics that are equal at two different t-values up to a diffeomorphism and a homothety such that
the metric with greater t -value has greater volume. However, the argument does not go through
generally if the B-field is permitted to have nonzero field strength H at some time during the
flow.
3.2. and the Weyl anomaly at fixed points
We can now evaluate the eigenvalue at a stationary point, thus at a fixed point of the flow for
which H is an exact 3-form. Since we consider a one-parameter family of flows with parameter s,
we write s for the eigenvalue and denote the stationary point by s = 0. At such a point, each
term in (3.12) vanishes, and thus in particular we must have
1
R + P |H |2 = 0.
4
When this holds, (3.5) takes the form
1
0 = P |P |2 + |H |2 .
6
(3.13)
(3.14)

Multiply this by eP and integrate. On the left-hand side, M 0 eP dV = 0 M eP dV = 0 ,

and on the right-hand side the derivatives of P vanish upon integration by parts (since M is
closed). This yields

1
0 =
(3.15)
|H |2 eP dV .
6
M
Notice now that, from the right-hand side of (2.11) with the diffeomorphism now chosen so that
= P as must be the case at a fixed point, we can write

eP dV = A
eP |H |2 dV = A 0 .
(3.16)
24
4
M
(Recall that A is a constant depending on the dimension of the target manifold and the number
of ghost fields, if any are present.) We compare this expression to Tseytlin [6] by considering the
combination7
1
g
:= g ij ij ,
4
(3.17)
g
which equals the Weyl anomaly at fixed points of g and B [19], where of course ij vanishes
and thus we obtain

eP dV = A 0 .
(3.18)
4
M
But at fixed points, is constant on M ([19]; for the case where no B-field is present, this follows
g
as an integrability condition for the fixed point equation ij = 0 as discussed in [20]; for the case
7 Tseytlins definition of t differs from ours by a sign, but the signs in his RG equations are such that his -functions
agree with ours.
450
with B-field, see the discussion in [21]). Then (3.18) reduces to the relation

(3.19)
0 .
4
An example is provided by a fixed point of (g(t), B(t)) with |H |2 := Hij k H ij k constant over
the closed manifold M and with dilaton linear in the scale t . We must first note that (3.14) can
now be written as

1 P
P
2
2 P
e = e
(3.20)
|H | |H | e dV .
6
= A
is constant over M (and so by (3.15) |H |2 = 6), the right-hand side of (3.20) vanSince
P
ishes and e is harmonic. Then P is constant. (For the B = 0 case, the result was already known
from work of Bourguignon [18].) Then (2.11) reduces to

1
(3.21)
= A +
+ = .
t
2
2
|H |2
A solution is given by

=
(3.22)
A t + 0 ,
4
0
(3.23)
= 0 ,
t
2
where the operator has no time dependence, in consequence of our being at a fixed point with
trivial diffeomorphism P = 0. Then if we choose that 0 is constant in time, it also will be
harmonic and thus spatially constant at all times, and we obtain, as claimed, that
= = A

.
4
(3.24)
3.3. A gradient flow with arbitrary diffeomorphism term

The 1-loop RG flow (2.9), (2.13) with arbitrary diffeomorphism not necessarily determined
by (3.5) is nonetheless a gradient flow as well. To see this, return to (3.7) but do not require
now that = P = 2 log u at each s. Instead, to eliminate the middle line, choose any fiducial
measure dm := e dV and hold it fixed pointwise along the variation. That is,
Proposition 3.4. The 1-loop RG flow (2.9), (2.13) is the gradient flow of F [g, B, ] along the
surface in G determined by the fixed fiducial measure dm := e dV on M.
g
Proof. If dm is fixed then g ij sij

s = 0 and the middle line on the right-hand side of (3.7)
again vanishes. We obtain (3.9), from which the gradient can be read off. Since dm can be chosen
arbitrarily, is now arbitrary as well. 2
While the approach of Proposition 3.4 may seem more general and perhaps simpler than the
approach of Proposition 3.1, it is in fact far less powerful, because F depends on the arbitrary
diffeomorphism potential . To prove results such as Proposition 3.3, one must remove the
dependence by passing to which, in contrast to F , is geometrically meaningfuland has a
clear physical interpretation.
451
4. The second variation of

In the case of Ricci flow (B = 0) the second variation formula for s was written down in
[15]. We generalize the formula here for arbitrary B-field. The next section applies this formula
in a special case.
In an endeavour to minimize clutter, we establish a convention. If a quantity below is to be
considered as a function of s, we write s explicitly as an argument or subscript. If s does not
appear, the quantity is evaluated at s = 0 (possibly after s differentiation; this should be clear
from context). We will use k to denote the covariant derivative compatible with gij (s) and Dk
for the covariant derivative compatible with gij gij (0).
The second variation formula about an arbitrary point is complicated. Fortunately, for most
purposes, we only need the second variation about a stationary point of s . Thus we require the
s = 0 fields (g, H ) to obey
1
Rij + Di Dj P Hikl Hj kl = 0,
4
D k Hkij Hkij D k P = 0.
Thus from (3.9) the second variation formula about a stationary point is

d 2 s
1
kl
ij
=
Rij i j P + Hikl Hj
h
s
4
ds 2 s=0
M

k
1
Hkij Hkij k P eP dV ,
+ ij
2
s
(4.1)
(4.2)
(4.3)
where we define
gkl
,
s
Bkl
kl :=
,
s
and for use below
hkl :=
hij := g ik g j l hkl ,
(4.4)
ij := g ik g j l kl ,
(4.5)
P
.
s
The standard formulas
Q :=
(4.6)
(4.7)
Rij = k ijk i jkk ,
s
s
s
k 1 kl
(4.8)
= g (i hj l + j hil l hij )
s ij 2
yield the easy identity (writing D for |s=0 whenever we can and using a standard result for the
variation of a Christoffel symbol)
(Rij Di Dj P )
s

P
k
P
P
k
Di j k
= e Dk e
Di Dj
s ij
s
s
452

k
1 kl
= e Dk e
g hkl Q
+ Di D j
s ij
2

1 P kl P
1 kl
= e g Dk e (Di hj l + Dj hil Dl hij ) + Di Dj
g hkl Q
2
2

1
= Di D k hj k + Dj D k hik Rkilj hkl Rkj li hkl + Rik hj k + Rjk hik hij
2

1 kl
+ Di D j
g hkl Q
2

1
1 kl
= Di D k hj k + Dj D k hik L hij + Di Dj
g hkl Q ,
2
2

(4.9)
where L denotes the Lichnerowicz Laplacian

L hij := hij + Rkilj hkl + Rkj li hkl Rik hj k Rjk hik .
(4.10)
Thus we obtain (round brackets around indices indicate symmetrization):

The Second Variation Formula:

d 2 s
1
=
eP hij L hij Hikm Hj l m hkl dV

2
ds s=0 2
M

1 kl
P ij
k
g hkl Q dV
+ e h D(i D hj )k + Di Dj
2
M

1
k
+
eP ij
Hkij Hkij k P dV ,
2
s
(4.11)
for variations about a general stationary point.

As a check on our results, we restrict to variations (h, 0) in which B is fixed. We determine
Q in the second variation formula by noting that (3.5) must hold all along the variation (i.e., for
d
s |s=0 = 0
all s). Differentiate it and evaluate the derivative at s = 0. On the left-hand side, ds
since s = 0 is a stationary point, while the right-hand side can be simplified by using (4.7), etc.,
to obtain

D i eP Di g ij hij 2Q = D i D j eP hij .
(4.12)
We write the solution of this equation (one always exists and is unique modulo an inconsequential
additive constant, since M is closed) as
vh := g ij hij 2Q.
(4.13)
|H |2
This determines Q, given hij . Now we further restrict to the case where
is constant on the
manifold. Then by (3.19), we can set P = 0. Under these circumstances, the second variation
formula becomes

1
d 2 s
ij
m kl
div(h)2 1 |Dvh |2 dV ,

dV
+
=
h
h
H
H
h
L
ij
ikm
j
l
2
ds 2 s=0 2
M
(4.14)
and this yields agreement with the Ricci flow result of [15] when
where (div h)i :=
Hij k = 0. For hij transverse (i.e., if D i hij = 0), both terms in the second integrand vanish.
D k hik ,
453
5. H = 0 fixed points
Consider a family of solutions of the RG flow with t the parameter along each flow and s the
family parameter. In the sequel, we always choose s such that s = 0 is a stationary point of s
and thus a fixed point of the flow. If the second variation of s about s = 0 is positive at t = 0,
then C := s (0) > 0 (0) for some s > 0. Since s (t) is monotonic in t and 0 (t) is constant, then
s (t) C > 0 (0) for all t . Since s is continuous on G, the coupling constants cannot approach
d2
the fixed point couplings along the flow. Thus, ds
2 s |s=0 > 0 indicates an instability of the fixed
point.
Now Ricci-flat fixed points have = 0, which is the least possible value of at a fixed point.
A particularly worrisome scenario would occur if certain Ricci-flat manifolds, such as flat tori
and K3 manifolds, were unstable against small perturbations and could flow to other, non-Ricciflat fixed points. To preclude this possibility, we must study the eigenvalues of the Hessian of s
about s = 0. This notion of stability is called linear stability.
It is difficult to study general variations (h, ) in both g and B about an arbitrary fixed point,
owing to the difficulty in diagonalizing the Hessian. However, Ricci-flat fixed points necessarily
have H = 0, and for a fixed point with H = 0 things simplify considerably, making it possible.
As well as the direct simplification of setting H = 0, the argument surrounding (3.19) then gives
that P can be set to zero as well, and so the fixed point condition gives that Rij = 0. We use
(4.11) to write

2
1
1
d 2 s
ij
2
2

=
h L hij + 2 div(h) |Dvh | |d| dV ,
(5.1)
3
ds 2 s=0 2
M
where now vh solves

D k Dk vh D k Dk g ij hij 2Q = D i D j hij .
Eq. (5.1) has the form

d 2 s
1
1
2
=
d
dV ,
|d|
,
d
h,
Lh
dV
=
h,
Lh
6
6
ds 2 s=0
M
(5.2)
(5.3)
1
Lh := L h + div div h + DDvh ,
2
(5.4)
where we have suppressed the indices, written (div h)i := D k hik , and let div denote the adjoint of the divergence and d denote the adjoint of d with respect to the inner product (3.10).
It is known [2224] that eigenvectors of L belonging to positive eigenvalues, if any exist, are
transverse and traceless, and moreover for hij transverse traceless then
1
L|N = L |N ,
2

N := hij | (divh)i := D k hik = 0, trg h := g ij hij = 0 .
(5.5)
(5.6)
Furthermore, on the complement of N , L < 0 unless hij arises from the action of a diffeomorphism or homothetic rescaling on gij , in which case L = 0.
The |d|2 term in (5.3) is obviously negative semi-definite, and negative if is not closed.
With that and the above comments concerning L in mind, we define a fixed point (g, B) with
Rij = 0, Hij k = 0 to be linearly stable if L |N 0. If the tangent directions at (g, B) for which
454
L |N = 0 can be exponentiated to give a smooth submanifold U of Ricci-flat fixed points in

the space of coupling constants, then we say that the fixed point (g, B) is integrable. We will
call this an integral submanifold of Ricci-flat fixed points. Since by construction L < 0 in the
complement in N of T U at each point along U , we call U strongly linearly stable. If U is strongly
linearly stable and if neither the spectrum of L nor that of d : 2 (M) 2 (M) accumulates
at zero, we will call U strictly linearly stable.
Lemma 5.1. If a strictly linearly stable integral submanifold U consists entirely of a disjoint
set of one or more Ricci-flat fixed points, then those points are rigid (isolated): there are no
neighbouring fixed points, Ricci-flat or not, except those obtained by diffeomorphism and/or
homothety.
Proof. At p U , consider any submanifold S whose tangent space at p is contained in N .
Because is zero and stationary at p and L is bounded below zero on T S, and d is bounded
below zero on 2 (M) modulo closed forms, then < 0 on some neighbourhood of p in S. But
fixed points have 0 by (3.15). 2
We have no examples of rigid fixed points. However, the proof generalizes to the case of
submanifolds U of nonzero dimension provided what is meant by isolated is interpreted to
mean that neighbouring fixed points which also belong to U are allowed. This brings us to the
cases of greatest interest here.
Flat tori
For flat manifolds, it helps to write (5.3) as

1
1
d 2 s
2
2
=
3|D(i hj k) | + |d| dV ,
2
3
ds 2 s=0
(5.7)
for hij N . This is strictly negative unless is closed and D(i hj k) = 0, and then hij is called a
Killing tensor. For tori, the Killing tensors are always linear combinations of outer products of
translation Killing vectors. These modes correspond to the relative rescaling of distinct cycles,
holding the torus volume fixed. These relative rescalings give rise to the moduli space of flat
structures on the torus, which is clearly a strongly linearly stable integral submanifold of Ricciflat fixed points in the space of coupling constants. Moreover, it is evident from the triviality of
the eigenvalue problem in this case that the moduli space is in fact strictly linearly stable.
K3 manifolds
The infinitesimal deformations (meaning in this situation the hij N such that L hij = 0) of
Khler Ricci-flat metrics on K3 manifolds are known to actually correspond to Ricci-flat metrics
[23,25]. Once again, these metrics form a submanifold U of Ricci-flat fixed points. It was shown
in [24] that L < 0 on the complement of T U in N . Thus U is a strongly linearly stable integral
submanifold of Ricci-flat fixed points.
Although these results on linear stability are strongly suggestive, they do not demonstrate
dynamical stability without further technical argument. In the case of Ricci flow (i.e., no Bfield), Sesum [23] has found linear and dynamical stability to be equivalent when the fixed point
satisfies the integrability condition. Thus flat tori and K3 manifolds are dynamically stable under
Ricci flow. We expect similar results will hold when a B-field is present. The issue is presently
under investigation. Nonetheless, the picture that emerges is one where flat tori and K3s are final
455
and not initial endpoints of the flow (except of course for the trivial case of a flow that remains
always at the fixed point). If flows that end at these points begin at unstable fixed points, then
monotonicity of and the fact that = 0 at the final endpoint would imply that those initial fixed
points would have < 0, contradicting (3.15). The resolution of this apparent paradox is simply
that higher order terms are significant for such flows and cannot be neglected.
Finally, consider the stability of the subset of fixed points of the first-order flow that remain
fixed points of the flow to all orders in . That is, they receive no stringy corrections at any
order in perturbation theory. These are called perturbative string vacua, and are described by
conformal field theories. Questions concerning the topology of the space of such vacua, the
dimension of the moduli space, etc., have been discussed in the string theory literature primarily
in the language of CFTs and their operator content. Because our results for general order
fixed points certainly obviously descend to perturbative string vacua, we have a complementary
picture in which these questions can be phrased in the language of target manifold geometry. The
clearest case would be that of a zero-dimensional integral submanifold of Ricci-flat perturbative
string vacua. Then Lemma 5.1 would apply directly.8 Checking stability would then be a matter
of checking the eigenvalues of L . If this could be done and if stability were confirmed, we
could infer that the related CFT should have no relevant or marginal operators. Now while we
presently know of no specific example of such a zero-dimensional manifold of vacua, we have
seen for flat tori that it need not be difficult to draw conclusions about relevant operators in the
CFT even when the vacuum belongs to a nontrivial integral submanifold.
In [11], Vafa addresses the question of whether the C-function can serve as a Morse function
for the configuration space of string theory, the hope being that this would shed light on the
topology of this configuration space. In particular, he points out that the possible existence of
nontrivial fixed points with no relevant or marginal directionsrigid perturbative string vacua in
the language aboveraises the question of whether the configuration space of string theory is
connected.
6. Concluding remarks
Throughout, we have limited attention to the most elementary of the monotone quantities for
Ricci flow in [9] and to order -functions. To do more would have made this article unwieldy.
It is, however, interesting to ponder whether Perelmans W -entropy, reduced length, and reduced
volume have useful analogues for the flows we have considered. An analogue of the scale invariant W -entropy would allow one to address and probably rule out the possibility that there may
be solutions of the RG flow which are periodic except for an overall homothetic rescaling [26].
It is also natural to ask whether these techniques might show the RG flow to be gradient
and have a monotonicity formula at higher order in . Higher-order RG flow equations have
a significant difficulty, which is that nonlinear combinations of leading-order spatial derivative
terms appear in the flow PDEs. Then the question may be vacuous, in that these PDEs might
not admit any solutions at all. Physics does not require that they do, since the exact RG flow, by
which we mean that the -functions are not approximated using a truncated loop expansion, can
still exist nonetheless. However, if a gradient flow is found for, say, the order 2 RG flow, then
this would be an important step in showing existence of solutions of the PDEs (as streamlines
8 KhlerRicci-flat K3 manifolds are perturbative string vacua. A version of Lemma 5.1 could be proved for the larger
integral submanifolds that occur there. However, we do not know if these manifolds are strictly linearly stable, even
though they are strongly linearly stable.
456
of the gradient flow). Thus, in the absence of a separate existence proof for solutions there is no
reason to expect to find a suitable gradient flow, but it is the very absence of such a proof that
makes the question more interesting.
Thinking beyond the loop expansion, we return to the C-theorem. In the present work, we
utilized recent mathematical breakthroughs to say something about order physics. These
advances occurred in the context of Ricci flow and attempts to prove conjectures concerning
3-manifold topology, but seem to echo the C-theorem (cf. Section 3.2 and [13]). It is intriguing
to ask what may come from a reverse strategy. For example, might the C-theorem lead to a tower
of geometric flows with monotonicity properties and interesting fixed points, order-by-order in
?9 Can the Ricci flow with surgery be given a physics interpretation and might the C-theorem
have something to say about topology of manifolds? To realize this potential, it seems very important to understand more deeply the relationship between the C-function and , or perhaps the
other monotonic quantities known for Ricci flow [9] but which we have not discussed herein.
Acknowledgements
This research was partially supported by a Discovery Grant from the Natural Sciences and
Engineering Research Council of Canada. V.S. was supported by a fellowship from the Pacific
Institute for the Mathematical Sciences. We thank the organizers of the Dark Side of Extra Dimensions meeting and the staff of the Banff International Research Station (BIRS), where this
work was begun. E.W. thanks the Albert Einstein Institute, the Isaac Newton Institute, and the
organizers of the Global Problems in Mathematical Relativity (GMR) workshop for hospitality. E.W. also thanks M.T. Anderson, Gerhard Huisken, H.-P. Knzle, and Kostas Skenderis for
discussions, Jim Isenberg for help with K3 manifolds, and Joe Polchinski and especially Hugh
Osborn for advice and references concerning the C-theorem for 2-dimensional sigma models.
Appendix A
Here we compute the first variation in the functional F [g, B, ] that results from a 1parameter variation of g, B, and . We denote the parameter by s and compute the first variation
term-by-term, starting from the formula

1
R
dF
=
+ ||2
|H |2 e dV
ds
s
s
12 s
M

1

e dV .
R + ||2 |H |2
12
s
(A.1)
Lemma A.1. For M compact, then

ij

gij
R
R + i j i j g ij ||2
e dV =
e dV .
s
s
M
9 This intriguing idea arose in discussion with Gerhard Huisken.
(A.2)
Proof. Use the standard formula

R
ij gij
i
j gij
kl gkl
= R
+
i g
s
s
s
s
and integrate by parts twice.
(A.3)
Lemma A.2. Assume that (3.1) holds. Then

gij i

2
j
|| e dV =
e dV .
2 ||
s
s
s
M
457
(A.4)
Proof.

gij i
||2 e dV =
j e dV
2 i i
s
s
s
(A.5)
and integrate by parts.
Lemma A.3.

1
2
|H | e dV
12
s
M

=
B ij

1 i
j kl gij
kij
kij
H kl H
+ k H H k
e dV .
4
s
s
(A.6)
Proof. The first term on the right-hand side is obvious. The second term follows from the fact
[i Bj k] = [i s
Bj k] = [i s
Bj k] . Then
that s
1
H ij k Hij k e dV
6
s
M

Bql
1
ip j q kl
e dV
Hij k g g g p
=
2
s
M

ip j q Bpq
k
1
k
2
e dV .
Hkij Hkij g g
=
(A.7)
2
s
M
Lemma A.4.

1

e dV
R + ||2 |H |2
12
s
M

1

1 ij gij
g
e dV .
R + ||2 |H |2
=
12
2
s
s
(A.8)
Proof. Follows from the formula
g
1
g s
= 12 g ij
gij
s
for the derivative of a determinant.
458
Proposition A.5. For any arbitrary smooth 1-parameter variation of g, B, and , then

gij
1
dF
=
R ij i j + H i kl H j kl
ds
4
s
M

1 ij gij
1
g
+ R |H |2 + 2 ||2
12
2
s
s

B
1
ij
+ k H kij H kij k
e dV .
2
s
Proof. Follows immediately from Lemmata A.1A.4.
(A.9)
References
[1] K. Wilson, Phys. Rev. B 4 (1971) 3174;
K. Wilson, Phys. Rev. B 4 (1971) 3184.
[2] K. Wilson, J.B. Kogut, Phys. Rep. C 12 (1974) 75.
[3] F.J. Wegner, A. Houghton, Phys. Rev. A 8 (1973) 401.
[4] A.B. Zamolodchikov, JETP Lett. 43 (1986) 730.
[5] J. Cardy, Scaling and Renormalization in Statistical Physics, Cambridge Univ. Press, Cambridge, 1996, p. 223.
[6] A.A. Tseytlin, Phys. Lett. B 194 (1987) 63.
[7] R.S. Hamilton, J. Differential Geom. 17 (1982) 255.
[8] D.H. Friedan, Ann. Phys. (N.Y.) 163 (1985) 318.
[9] G. Perelman, math.DG/0211159.
[10] T. Oliynyk, V. Suneeta, E. Woolgar, Phys. Lett. B 21739 (2005), hep-th/0410001.
[11] C. Vafa, Phys. Lett. B 212 (1988) 28.
[12] S.R. Das, G. Mandal, S.R. Wadia, Mod. Phys. Lett. A 4 (1989) 745.
[13] V.A. Fateev, E. Onofri, A.B. Zamolodchikov, Nucl. Phys. B 406 (1993) 521.
[14] V.A. Fateev, E. Onofri, J. Phys. A 36 (2003) 11881, math-ph/0307010.
[15] H.-D. Cao, R. Hamilton, T. Ilmanen, math.DG/0404165.
[16] J.G. Polchinski, String Theory, vol. I, Cambridge Univ. Press, Cambridge, 1998, p. 111.
[17] B. Kleiner, J. Lott, Notes on Perelmans papers, unpublished, http://www.math.lsa.umich.edu/research/ricciflow/
perelman.html.
[18] J.-P. Bourguignon, in: D. Ferus (Ed.), Global Differential Geometry and Global Analysis, in: Lectures in Mathematics, vol. 838, Springer, Berlin, 1981.
[19] C.G. Callan, D. Friedan, E.J. Martinec, M.J. Perry, Nucl. Phys. B 262 (1985) 593.
[20] G. Curci, G. Paffuti, Nucl. Phys. B 286 (1987) 399.
[21] H. Osborn, Ann. Phys. (N.Y.) 200 (1990) 1.
[22] A.L. Besse, Einstein Manifolds, Springer, Berlin, 1987, p. 132.
[23] N. Sesum, math.DG/0410062.
[24] C. Guenther, J. Isenberg, D. Knopf, Commun. Anal. Geom. 10 (2002) 741.
[25] A. Todorov, Invent. Math. 61 (1980) 251;
A. Todorov, in: Arithmetic and Geometry: Papers dedicated to I.R. Shafarevich on the Occasion of His Sixtieth
Birthday, vol. 2, Birkhuser, Basel, 1983, p. 451.
[26] V. Dolotin, A. Morozov, hep-th/0501235, and references therein.
CUMULATIVE AUTHOR INDEX B731B739
Ableev, V.
Ahmadi, N.
Aissaoui, H.
Albino, S.
Alcaniz, J.S.
Alexandrov, S.
Aliev, T.M.
Alimohammadi, M.
Altshuler, B.L.
Ambjorn, J.
Ambjrn, J.
Ammosov, V.
Anguelova, L.
Anisimov, A.
Antoniadis, I.
Apollonio, M.
Arakawa, G.
Arce, P.
Artamonov, A.
B732 (2006) 1
B738 (2006) 269
B738 (2006) 76
B734 (2006) 50
B732 (2006) 379
B731 (2005) 242
B732 (2006) 291
B733 (2006) 123
B739 (2006) 311
B734 (2006) 287
B736 (2006) 288
B732 (2006) 1
B733 (2006) 132
B737 (2006) 176
B731 (2005) 164
B732 (2006) 1
B732 (2006) 401
B732 (2006) 1
B732 (2006) 1
Blossier, B.
Bobisut, F.
Bogdan, A.V.
Bogomilov, M.
Bolzoni, P.
Bondarev, A.L.
Bonesini, M.
Booth, C.
Borghi, S.
Boucaud, Ph.
Bozi, D.
Bozzi, G.
Broncano, A.
Brmmer, F.
Buchel, A.
Buchel, A.
Bunyatov, S.
Burguet-Castell, J.
Buttar, C.
B734 (2006) 138

B732 (2006) 1
B732 (2006) 169
B732 (2006) 1
B731 (2005) 85
B733 (2006) 48
B732 (2006) 1
B732 (2006) 1
B732 (2006) 1
B734 (2006) 138
B737 (2006) 351
B737 (2006) 73
B737 (2006) 176
B738 (2006) 283
B731 (2005) 109
B733 (2006) 160
B732 (2006) 1
B732 (2006) 1
B732 (2006) 1
Babu, K.S.
Babujian, H.
Bachri, A.
Bagulya, A.
Barnes, E.
Barr, G.
Batalin, I.A.
Bauer, R.O.
Becirevic, D.
Becker, M.
Behrndt, K.
Belitsky, A.V.
Bellorn, J.
Beneke, M.
Benincasa, P.
Berg, M.
Bering, K.
Bjrnsson, J.
Blondel, A.
B738 (2006) 76
B736 (2006) 169
B738 (2006) 76
B732 (2006) 1
B732 (2006) 89
B732 (2006) 1
B739 (2006) 389
B733 (2006) 91
B734 (2006) 138
B738 (2006) 124
B732 (2006) 200
B735 (2006) 17
B737 (2006) 190
B736 (2006) 34
B733 (2006) 160
B736 (2006) 82
B739 (2006) 389
B736 (2006) 156
B732 (2006) 1
Calvi, M.
Campanelli, M.
Canfora, F.
Cao, J.
Carlevaro, L.
Carmelo, J.M.P.
Carmelo, J.M.P.
Catanesi, M.G.
Catani, S.
Caudrelier, V.
Cervera-Villanueva, A.
Chan, C.-T.
Chen, C.-M.
Chen, W.
Chimenti, P.
Chizhov, M.
Chong, Z.-W.
Christian, N.
Chukanov, A.
B732 (2006) 1
B732 (2006) 1
B731 (2005) 389
B731 (2005) 352
B736 (2006) 1
B737 (2006) 237
B737 (2006) 351
B732 (2006) 1
B737 (2006) 73
B738 (2006) 351
B732 (2006) 1
B738 (2006) 93
B732 (2006) 224
B732 (2006) 118
B732 (2006) 1
B732 (2006) 1
B732 (2006) 118
B739 (2006) 60
B732 (2006) 1
0550-3213/2006 Published by Elsevier B.V.

doi:10.1016/S0550-3213(06)00156-8
460
Ciechanowicz, S.
Coney, L.
Cramp, N.
B734 (2006) 203

B732 (2006) 1
B738 (2006) 351
Dai, J.
DAlessandro, M.
Damgaard, P.H.
da Providncia, J.
Das, S.R.
Dasgupta, K.
DAuria, R.
Dedovitch, D.
De Fazio, F.
de Florian, D.
de Haro, S.
Delfino, G.
Delfino, G.
de Medeiros Varzielas, I.
De Min, A.
Denner, A.
Derendinger, J.-P.
Derkachov, S.
De Santo, A.
Dev, A.
Di Capua, E.
Dittmaier, S.
Dolan, B.P.
Dolgov, A.
Dore, U.
Dubath, F.
Dumarchez, J.
B731 (2005) 285

B732 (2006) 64
B739 (2006) 389
B737 (2006) 337
B733 (2006) 297
B738 (2006) 124
B732 (2006) 389
B732 (2006) 1
B733 (2006) 1
B737 (2006) 73
B731 (2005) 225
B736 (2006) 259
B737 (2006) 291
B733 (2006) 31
B732 (2006) 1
B734 (2006) 62
B736 (2006) 1
B738 (2006) 368
B732 (2006) 1
B732 (2006) 379
B732 (2006) 1
B734 (2006) 62
B737 (2006) 153
B734 (2006) 208
B732 (2006) 1
B736 (2006) 302
B732 (2006) 1
Eden, B.
Edgecock, R.
Elagin, A.
Ellis, M.
Endo, I.
B738 (2006) 409

B732 (2006) 1
B732 (2006) 1
B732 (2006) 1
B732 (2006) 426
Fehr, L.
Feldmann, T.
Ferrara, S.
Ferri, F.
Foerster, A.
Forte, S.
Fr, P.
Fr, P.
Friedrich, R.M.
Frolov, S.A.
B734 (2006) 304

B733 (2006) 1
B732 (2006) 389
B732 (2006) 1
B736 (2006) 169
B731 (2005) 85
B733 (2006) 334
B737 (2006) 1
B733 (2006) 91
B731 (2005) 1
Gaillard, M.K.
Galleas, W.
Gapienko, V.
Garbrecht, B.
Gargiulo, F.
Gastaldi, U.
Genovese, L.
B734 (2006) 116

B732 (2006) 444
B732 (2006) 1
B736 (2006) 133
B737 (2006) 1
B732 (2006) 1
B732 (2006) 64
Giani, S.
Giannini, G.
Gibbons, G.W.
Gibin, D.
Gilardoni, S.
Gimnez, V.
Giovannangeli, P.
Giribet, G.
Giusto, S.
Giusto, S.
Gmez-Cadenas, J.J.
Gorbatov, E.
Gorbunov, P.
Gling, C.
Gostkin, M.
Grange, P.
Grant, A.
Graulich, J.S.
Grazzini, M.
Grgoire, G.
Grichine, V.
Grinza, P.
Gromov, N.
Grossheim, A.
Gruber, P.
Guglielmi, A.
Guo, L.
Guralnik, Z.
Guskov, A.
B732 (2006) 1
B732 (2006) 1
B732 (2006) 118
B732 (2006) 1
B732 (2006) 1
B734 (2006) 138
B738 (2006) 23
B737 (2006) 209
B733 (2006) 297
B738 (2006) 48
B732 (2006) 1
B732 (2006) 89
B732 (2006) 1
B732 (2006) 1
B732 (2006) 1
B732 (2006) 366
B732 (2006) 1
B732 (2006) 1
B737 (2006) 73
B732 (2006) 1
B732 (2006) 1
B737 (2006) 291
B736 (2006) 199
B732 (2006) 1
B732 (2006) 1
B732 (2006) 1
B738 (2006) 243
B732 (2006) 46
B732 (2006) 1
Haack, M.
Haba, N.
HARP Collaboration
Hayato, Y.
Hayes, J.G.
Hebecker, A.
Hegedus, .
Hodgson, P.
Hollik, W.
Howlett, L.
Hurth, T.
Hwang, S.
B736 (2006) 82
B739 (2006) 254
B732 (2006) 1
B732 (2006) 1
B739 (2006) 106
B738 (2006) 283
B732 (2006) 463
B732 (2006) 1
B731 (2005) 213
B732 (2006) 1
B733 (2006) 1
B736 (2006) 156
Ichikawa, A.
Ichinose, I.
Imeroni, E.
Intriligator, K.
Ishikawa, H.
Ivanchenko, V.
Ivanov, D.A.
Ivanov, D.Yu.
B732 (2006) 1
B732 (2006) 401
B731 (2005) 242
B732 (2006) 89
B739 (2006) 328
B732 (2006) 1
B737 (2006) 304
B732 (2006) 183
Jacob, P.
Jacobsen, J.L.
Jain, D.
Janik, R.A.
Janke, W.
B733 (2006) 205

B731 (2005) 335
B732 (2006) 379
B736 (2006) 288
B736 (2006) 319
Jansen, K.
Jantzen, B.
Jarvis, P.D.
Johnston, D.A.
Jung, E.
B739 (2006) 60
B731 (2005) 188
B734 (2006) 272
B736 (2006) 319
B731 (2005) 171
Kain, B.
Kakizaki, M.
Kant, E.
Karakhanyan, D.
Karowski, M.
Kato, I.
Katz, S.
Kaufhold, C.
Kawano, T.
Kayis-Topaksu, A.
Kazakov, V.
Kenna, R.
Khachatryan, Sh.
Khartchenko, D.
Khorrami, M.
King, S.F.
Kirsanov, M.
Kirschner, R.
Klimov, O.
Klinkhamer, F.R.
Klinkhamer, F.R.
Knauf, A.
Kniehl, B.A.
Kniehl, B.A.
Kobayashi, T.
Koibuchi, H.
Kokorelis, C.
Kolev, D.
Konik, R.M.
Konstandin, T.
Korchemsky, G.P.
Koreshev, V.
Kotikov, A.V.
Kramer, G.
Krasnoperov, A.
Kristjansen, C.
Khn, J.H.
Kulaxizi, M.
Kustov, D.
B734 (2006) 116

B735 (2006) 84
B731 (2005) 125
B738 (2006) 368
B736 (2006) 169
B732 (2006) 1
B738 (2006) 124
B734 (2006) 1
B735 (2006) 1
B732 (2006) 1
B736 (2006) 199
B736 (2006) 319
B734 (2006) 287
B732 (2006) 1
B733 (2006) 123
B739 (2006) 106
B732 (2006) 1
B738 (2006) 368
B732 (2006) 1
B731 (2005) 125
B734 (2006) 1
B738 (2006) 124
B734 (2006) 50
B738 (2006) 306
B732 (2006) 1
B732 (2006) 426
B732 (2006) 341
B732 (2006) 1
B739 (2006) 311
B738 (2006) 1
B735 (2006) 17
B732 (2006) 1
B738 (2006) 306
B734 (2006) 50
B732 (2006) 1
B736 (2006) 288
B731 (2005) 188
B738 (2006) 317
B732 (2006) 1
Laveder, M.
Lee, J.-C.
Lee, R.N.
Li, T.
Linssen, L.
Liu, J.T.
Liu, X.
Lopes Cardoso, G.
L, C.-D.
L, H.
B732 (2006) 1
B738 (2006) 93
B732 (2006) 169
B732 (2006) 224
B732 (2006) 1
B739 (2006) 285
B738 (2006) 243
B732 (2006) 200
B738 (2006) 243
B732 (2006) 118
461
Lubicz, V.
Lst, D.
B734 (2006) 138

B732 (2006) 243
Maeda, K.-i.
Maeda, T.
Mahapatra, S.
Manousselis, P.
Manvelyan, R.
Markopoulou, F.
Marquet, C.
Martelo, L.M.
Martins, M.J.
Martins, M.J.
Mass, M.
Mathieu, P.
Mathur, S.D.
Mathur, S.D.
Matone, M.
Matsui, T.
Matsumoto, S.
Mattei, F.
Mayr, P.
Meier, U.
Menegolli, A.
Mescia, F.
Metlitski, M.A.
Meusburger, C.
Mezzetto, M.
Mills, G.B.
Minahan, J.A.
Minamitsuji, M.
Minasian, R.
Misiaszek, M.
Morariu, B.
Morita, H.
Morone, M.C.
Muciaccia, M.T.
Mck, W.
Mller, D.
Mller, D.
Mussardo, G.
Mussardo, G.
B738 (2006) 184

B735 (2006) 96
B732 (2006) 200
B739 (2006) 85
B733 (2006) 104
B739 (2006) 120
B739 (2006) 131
B737 (2006) 237
B732 (2006) 444
B738 (2006) 391
B732 (2006) 1
B733 (2006) 205
B733 (2006) 297
B738 (2006) 48
B732 (2006) 321
B732 (2006) 401
B735 (2006) 84
B739 (2006) 234
B732 (2006) 243
B731 (2005) 213
B732 (2006) 1
B734 (2006) 138
B731 (2005) 309
B738 (2006) 425
B732 (2006) 1
B732 (2006) 1
B735 (2006) 127
B737 (2006) 121
B732 (2006) 366
B734 (2006) 203
B734 (2006) 156
B737 (2006) 337
B732 (2006) 1
B732 (2006) 1
B736 (2006) 82
B735 (2006) 17
B739 (2006) 1
B736 (2006) 259
B737 (2006) 291
Nagai, K.
Nakatsu, T.
Nakaya, T.
Nanopoulos, D.V.
Nardi, E.
Naylor, W.
Nikolaev, K.
Nishikawa, K.
Nishiyama, S.
Noma, Y.
Nouri-Zonoz, M.
Novella, P.
B739 (2006) 60
B735 (2006) 96
B732 (2006) 1
B732 (2006) 224
B731 (2005) 140
B737 (2006) 121
B732 (2006) 1
B732 (2006) 1
B737 (2006) 337
B735 (2006) 96
B738 (2006) 269
B732 (2006) 1
Ohnishi, H.
Okuda, T.
B737 (2006) 337

B733 (2006) 59
462
Oliynyk, T.
Onishchenko, A.I.
Ookouchi, Y.
Ookouchi, Y.
Orestano, D.
Ozpineci, A.
B739 (2006) 441

B738 (2006) 306
B733 (2006) 59
B735 (2006) 1
B732 (2006) 1
B732 (2006) 291
Paccetti Correia, F.
Paganoni, M.
Paleari, F.
Palladino, V.
Panico, G.
Panman, J.
Pantev, T.
Papa, A.
Papadopoulos, I.
Papageorgakis, C.
Parameswaran, S.L.
Park, D.K.
Parkhomenko, S.E.
Pascoli, S.
Pasquali, M.
Pasternak, J.
Pastore, F.
Pattison, C.
Peddie, I.N.R.
Pelliccia, D.N.
Penc, K.
Penc, K.
Penin, A.A.
Penin, A.A.
Petcov, S.T.
Petcov, S.T.
Petcov, S.T.
Piperov, S.
Plmacher, M.
Pollakowski, B.
Polukhina, N.
Pope, C.N.
Popov, B.
Prezas, N.
Prior, G.
Prokopec, T.
Prokopec, T.
Pusztai, B.G.
B739 (2006) 156

B732 (2006) 1
B732 (2006) 1
B732 (2006) 1
B739 (2006) 186
B732 (2006) 1
B733 (2006) 233
B732 (2006) 183
B732 (2006) 1
B731 (2005) 45
B737 (2006) 49
B731 (2005) 171
B731 (2005) 360
B734 (2006) 24
B732 (2006) 1
B732 (2006) 1
B732 (2006) 1
B732 (2006) 1
B739 (2006) 106
B734 (2006) 208
B737 (2006) 237
B737 (2006) 351
B731 (2005) 188
B734 (2006) 185
B734 (2006) 24
B738 (2006) 219
B739 (2006) 208
B732 (2006) 1
B737 (2006) 176
B739 (2006) 60
B732 (2006) 1
B732 (2006) 118
B732 (2006) 1
B739 (2006) 85
B732 (2006) 1
B736 (2006) 133
B738 (2006) 1
B734 (2006) 304
Radicioni, E.
Ramgoolam, S.
Rasmussen, J.
Reffert, S.
Restuccia, A.
Ribeiro, G.A.P.
Richard, J.-F.
Ridolfi, G.
Riva, V.
Robbins, S.
B732 (2006) 1
B731 (2005) 45
B736 (2006) 225
B732 (2006) 243
B737 (2006) 190
B738 (2006) 391
B731 (2005) 335
B731 (2005) 85
B736 (2006) 259
B732 (2006) 1
Rodejohann, W.
Roiban, R.
Ross, G.G.
Rovelli, C.
Royon, C.
Rhl, W.
Rulik, K.
B739 (2006) 208

B731 (2005) 1
B733 (2006) 31
B739 (2006) 234
B739 (2006) 131
B733 (2006) 104
B737 (2006) 1
Sakakibara, K.
Saleur, H.
Santin, G.
Sanz-Cillero, J.J.
Sasaki, M.
Sato, Y.
Schfer, A.
Schmidt, M.G.
Schmidt, M.G.
Schmidt, M.G.
Schmitz, D.
Schomerus, V.
Schroers, B.J.
Schroeter, R.
Schwetz, T.
Seco, M.
Sedrakyan, A.
Semak, A.
Senami, M.
Serdiouk, V.
Serone, M.
Sharpe, E.
Shindou, T.
Shindou, T.
Simone, S.
Simula, S.
Singh, H.
Skvortsov, M.A.
Smirnov, V.A.
Smolin, L.
Smolin, L.
Sobkw, W.
Soler, F.J.P.
Sorel, M.
Sorin, A.S.
Sotkov, G.
Speziale, S.
SPQCD R Collaboration
Srivastava, Y.
Starinets, A.O.
Stieberger, S.
Suneeta, V.
Sviridov, Yu.
B732 (2006) 401

B734 (2006) 221
B732 (2006) 1
B732 (2006) 136
B737 (2006) 121
B735 (2006) 84
B739 (2006) 1
B736 (2006) 133
B738 (2006) 1
B739 (2006) 156
B732 (2006) 1
B734 (2006) 221
B738 (2006) 425
B732 (2006) 1
B734 (2006) 24
B738 (2006) 1
B734 (2006) 287
B732 (2006) 1
B735 (2006) 84
B732 (2006) 1
B739 (2006) 186
B733 (2006) 233
B738 (2006) 219
B739 (2006) 208
B732 (2006) 1
B734 (2006) 138
B734 (2006) 169
B737 (2006) 304
B731 (2005) 188
B739 (2006) 120
B739 (2006) 169
B734 (2006) 203
B732 (2006) 1
B732 (2006) 1
B733 (2006) 334
B736 (2006) 259
B739 (2006) 234
B734 (2006) 138
B733 (2006) 297
B733 (2006) 160
B732 (2006) 243
B739 (2006) 441
B732 (2006) 1
Tachikawa, Y.
Tachikawa, Y.
Takanishi, Y.
Takanishi, Y.
Takashima, S.
B733 (2006) 188

B735 (2006) 1
B738 (2006) 219
B739 (2006) 208
B732 (2006) 401
Tamakoshi, T.
Tanabe, M.
Tani, T.
Tarantino, C.
Tasinato, G.
Tatar, R.
Tavartkiladze, Z.
Taylor, T.R.
Tcherniaev, E.
Temnikov, P.
Tereshchenko, V.
Testa, M.
Tierz, M.
Tirziu, A.
Tonazzo, A.
Tornero, A.
Tortora, L.
Tran, N.-K.
Trigiante, M.
Trincherini, E.
Troquereau, S.
Tsenov, R.
Tseytlin, A.A.
Tseytlin, A.A.
Tsuboi, Z.
Tsukerman, I.
Tsvelik, A.M.
B735 (2006) 96
B738 (2006) 184
B739 (2006) 328
B734 (2006) 138
B737 (2006) 49
B738 (2006) 124
B739 (2006) 156
B731 (2005) 164
B732 (2006) 1
B732 (2006) 1
B732 (2006) 1
B739 (2006) 234
B731 (2005) 225
B735 (2006) 127
B732 (2006) 1
B732 (2006) 1
B732 (2006) 1
B734 (2006) 246
B732 (2006) 389
B738 (2006) 283
B732 (2006) 1
B732 (2006) 1
B731 (2005) 1
B735 (2006) 127
B737 (2006) 261
B732 (2006) 1
B739 (2006) 311
Uccirati, S.
Ueda, S.
B731 (2005) 213

B732 (2006) 1
Vaman, D.
Vaman, D.
van Holten, J.W.
Vannucci, F.
Veenhof, R.
B733 (2006) 132

B739 (2006) 285
B734 (2006) 272
B732 (2006) 1
B732 (2006) 1
463
Veretin, O.L.
Vergeles, S.N.
Vidal-Sitjes, G.
Volpato, R.
B738 (2006) 306

B735 (2006) 172
B732 (2006) 1
B732 (2006) 321
Wang, H.
Wang, X.-J.
Wang, X.R.
Wang, Y.
Wen, W.Y.
Wiebusch, C.
Woolgar, E.
Wright, J.
Wu, X.
Wu, Y.-S.
Wulzer, A.
B738 (2006) 243

B731 (2005) 285
B731 (2005) 352
B731 (2005) 352
B739 (2006) 285
B732 (2006) 1
B739 (2006) 441
B732 (2006) 89
B733 (2006) 297
B731 (2005) 285
B739 (2006) 186
Xiao, Z.
Xie, X.C.
Xiong, G.
B738 (2006) 243

B731 (2005) 352
B731 (2005) 352
Yagi, F.
Yang, D.
Yang, Y.
Yoshioka, K.
B735 (2006) 1
B736 (2006) 34
B738 (2006) 93
B739 (2006) 254
Zaets, V.
Zavala, I.
Zhemchugov, A.
Zhitnitsky, A.R.
Zhou, C.
Zoubos, K.
Zoupanos, G.
Zuber, K.
Zucchelli, P.
Zuluaga, J.I.
B732 (2006) 1
B737 (2006) 49
B732 (2006) 1
B731 (2005) 309
B733 (2006) 297
B738 (2006) 317
B739 (2006) 85
B732 (2006) 1
B732 (2006) 1
B731 (2005) 140

Nucl - Phys.B v.739 PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Nucl - Phys.B v.739 PDF

Uploaded by

Copyright:

Available Formats

Nuclear Physics B 739 (2006) 159

Complex conformal spin partial wave expansion

E-mail address: dieter.mueller@physik.uni-regensburg.de (D. Mller).

D. Mller, A. Schfer / Nuclear Physics B 739 (2006) 159

D. Mller, A. Schfer / Nuclear Physics B 739 (2006) 159

D. Mller, A. Schfer / Nuclear Physics B 739 (2006) 159

D. Mller, A. Schfer / Nuclear Physics B 739 (2006) 159

gauge link is omitted):

with P+ = n (P1 + P2 ), V = n V , n2 = (n )2 = 0, n n = 1. In the first (vector) and

and the helicity flip GPDs Ei decouple, but

D. Mller, A. Schfer / Nuclear Physics B 739 (2006) 159

D. Mller, A. Schfer / Nuclear Physics B 739 (2006) 159

Here both functions separately satisfy the polynomiality condition:

and analogous for the anti-quark GPD q(x,

easily obtained by replacements D D.

D. Mller, A. Schfer / Nuclear Physics B 739 (2006) 159

D. Mller, A. Schfer / Nuclear Physics B 739 (2006) 159

For illustration we give a simple example for the Radyushkin ansatz

We remark that the rescaled distribution

has technically the same support as it appears in evolution kernels. Defining  by

D. Mller, A. Schfer / Nuclear Physics B 739 (2006) 159

D. Mller, A. Schfer / Nuclear Physics B 739 (2006) 159

where the lower line is in fact the relation (24).

where = ( 1)/2, W 2 = 2 and AC denotes the analytic continuation of both variables

D. Mller, A. Schfer / Nuclear Physics B 739 (2006) 159

D. Mller, A. Schfer / Nuclear Physics B 739 (2006) 159

D. Mller, A. Schfer / Nuclear Physics B 739 (2006) 159

where the polynomials Mn (, W 2 ) = (12 )n mn (1/(12 ), W 2 ) are of order n. If (z, , W 2 )

D. Mller, A. Schfer / Nuclear Physics B 739 (2006) 159

D. Mller, A. Schfer / Nuclear Physics B 739 (2006) 159

The remaining integral represents just (1/2)(2 + j )/ (5/2 + j ), which results in

D. Mller, A. Schfer / Nuclear Physics B 739 (2006) 159

Consequently, at the endpoint x = and for x < the integral vanishes

(5/2 + j ) sin([j + 1])

21+j (5/2 + j ) sin([j + 1])

D. Mller, A. Schfer / Nuclear Physics B 739 (2006) 159

D. Mller, A. Schfer / Nuclear Physics B 739 (2006) 159

To derive an appropriate integral representation, we decompose here, in contrast to Eq. (38),

From the asymptotics of the hypergeometric functions

we can estimate the large j -behavior of conformal moments

D. Mller, A. Schfer / Nuclear Physics B 739 (2006) 159

Fig. 5. The integration contour in Eq. (65).

D. Mller, A. Schfer / Nuclear Physics B 739 (2006) 159

For 1 X 1 the function arccosh(X i ) has a monotonously increasing or decreasing

D. Mller, A. Schfer / Nuclear Physics B 739 (2006) 159

D. Mller, A. Schfer / Nuclear Physics B 739 (2006) 159

fest in the conformal moments, rather it is separately restored for m

D. Mller, A. Schfer / Nuclear Physics B 739 (2006) 159

Suppose that Mj (, W 2 ) vanishes for j with | arg j | /2, it is straightforward to see

D. Mller, A. Schfer / Nuclear Physics B 739 (2006) 159

for |x|, |y| 1,

D. Mller, A. Schfer / Nuclear Physics B 739 (2006) 159

and an anti-quark GPD, with 1 x ,

D. Mller, A. Schfer / Nuclear Physics B 739 (2006) 159

where the evolution kernel is

D. Mller, A. Schfer / Nuclear Physics B 739 (2006) 159

D. Mller, A. Schfer / Nuclear Physics B 739 (2006) 159

where the sum of mj + m

, 2 ) as it must be. The real part

with the NLO coefficients

D. Mller, A. Schfer / Nuclear Physics B 739 (2006) 159

D. Mller, A. Schfer / Nuclear Physics B 739 (2006) 159

Here we employed the symmetry relation Mn (, W 2 ) = (1)n Mn (1 , W 2 ) and that Mj is

has technically the same support as it appears in evolution kernels. Defining by

The remaining integral represents just (1/2)(2 + j )/ (5/2 + j ), which results in

(5/2 + j ) sin([j + 1])

21+j (5/2 + j ) sin([j + 1])

For 1 X 1 the function arccosh(X i) has a monotonously increasing or decreasing

(1/2)(1 + j ) (3 + )(1 + j ) sin(j )

(1/2)(1 + j ) (3 + )(1 + j ) sin()

(1/2)(1 + j ) (3 + )(1 + j )

2j +1 (5/2 + j ) (1/2 + j )(9/2) 2

(0) = 0.74 GeV2 .

the scalar condensate, .