Nucl - Phys.B v.606 PDF

Nuclear Physics B 606 (2001) 317
www.elsevier.com/locate/npe
D-branes, holonomy and M-theory

Jaume Gomis a,b
a Department of Physics, California Institute of Technology, Pasadena, CA 91125, USA
b CaltechUSC Center for Theoretical Physics, University of Southern California, Los Angeles, CA 90089, USA
Received 18 May 2001; accepted 18 May 2001
Abstract
We show that M-theory on spaces with irreducible holonomy represent Type IIA backgrounds
in which a collection of D6-branes wrap a supersymmetric cycle in a manifold with a holonomy
group different from the one appearing in the M-theory description. For example, we show that
D6-branes wrapping a supersymmetric four-cycle on a manifold with G2 holonomy is described
in eleven dimensions by M-theory on a space with Spin(7) holonomy. Examples of such Type IIA
backgrounds which lift to M-theory on spaces with SU(3), G2 , SU(4) and Spin(7) holonomy are
considered. The M-theory geometry can then be used to compute exact quantities of the gauge theory
on the corresponding D-brane configuration. 2001 Published by Elsevier Science B.V.
1. Introduction
One of the important lessons of the duality era has been that string theory backgrounds
can be naturally embedded in an eleven-dimensional theory. The simplest example of this
embedding is the uplift of any background of Type IIA string theory to M-theory [1,2]. In
particular, any solution of the equations of motion described by the ten-dimensional metric
g , the RamondRamond one form C and the dilaton uplifts in eleven dimensions
to a solution of eleven-dimensional supergravity with the following eleven-dimensional
metric
2
2
4
2
ds11
(1.1)
= e 3 g dx dx + e 3 dx11 + C dx .
The uplift of the rest of the massless bosonic fields of Type IIA string theory corresponds
to turning on the three-form gauge field of eleven-dimensional supergravity. Therefore,
backgrounds of Type IIA string theory which only source the fields g , C and must
be described in eleven dimensions by a purely gravitational background without any flux.
E-mail address: gomis@theory.caltech.edu (J. Gomis).
0550-3213/01/$ see front matter 2001 Published by Elsevier Science B.V.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 2 4 7 - 4
J. Gomis / Nuclear Physics B 606 (2001) 317
On the other hand, supersymmetric M-theory vacua which are purely gravitational
are completely classified. Supersymmetric compactifications of eleven-dimensional supergravity to (d + 1)-dimensional Minkowski space R1,d appear whenever the D 10 d
dimensional compactification manifold X admits covariantly constant spinors. 1 The only
solution in this class with maximal supersymmetry is when X is a torus TD . Solutions
with reduced supersymmetry can be obtained when the compactification space X has covariantly constant spinors which do not span the full SO(D) spinor. Such D-dimensional
spaces have a holonomy group G such that at least one component of the SO(D) spinor
is left invariant by the action of G. The list [6] of irreducible holonomy groups preserving
some supersymmetry and the corresponding number of unbroken supercharges in R1,d is:
D = 4,
G = SU(2),
7-dim. N = 1 SUSY,
16 real supercharges,
D = 6,
G = SU(3),
5-dim. N = 1 SUSY,
D = 7,
G = G2 ,
4-dim. N = 1 SUSY,
D = 8,
G = SU(4),
3-dim. N = 2 SUSY,
D = 8,
G = Spin(7), 3-dim. N = 1 SUSY,
2 real supercharges.
The number of unbroken supercharges is given by the number of singlets in the

decomposition of the spinor of SO(D) under the reduced holonomy group G Spin(D).
Manifolds with SU(2), SU(3) and SU(4) holonomy are CalabiYau spaces while
manifolds with G2 and Spin(7) holonomy are real Ricci flat manifolds. 2
In this paper we show that M-theory compactified 3 on certain spaces with reduced
holonomy appear as the local eleven-dimensional description of Type IIA D6-branes
wrapping supersymmetric cycles on certain lower-dimensional spaces with a different
holonomy group than the one in the M-theory description. It is crucial that we use D6branes in the Type IIA construction since this is the only 4 brane which lifts in eleven
dimensions to pure geometry. Supersymmetry severely constraints the types of lifts that
are possible. The basic geometry that appears in eleven dimensions near the location of the
D6-branes is that of an ALE fibration over the supersymmetric cycle in which we wrapped
the D6-branes. This fibration has a different holonomy group than that of the manifold we
started with in Type IIA. We consider the possibility of adding curved orientifold six-planes
(O6-planes) by modding out by worldsheet parity combined with an appropriate involution
on the geometry. Then, the type of ALE fibration appearing in eleven dimensions depends
on whether there is an orientifold plane in the Type IIA description. The gauge theory on
the D6-branes appears in the M-theory description from singularities in the geometry. In
1 There are many other interesting supersymmetric solutions if one allows for non-trivial background fluxes.
For example, one can have AdS vacua (see [3] for a nice review of some such solutions) and warped
compactifications (see, e.g., [4,5]).
2 Eight-dimensional hyper-Kahler manifolds with Sp(2) holonomy will not be considered in this paper. They
give rise to three-dimensional theories with N = 3 supersymmetry.
3 We will consider non-compact spaces so strictly speaking we are not compactifying. Therefore, the constraint
[7] imposed by tadpole cancellation is lifted since the flux lines can escape to infinity.
4 The magnetic dual object, the D0-brane, also lifts to pure geometry.
the various examples we consider, we notice that the M-theory geometrical description can
be used to compute quantities such as prepotentials or superpotentials of the gauge
theory living on the D6-branes.
Our results extend the work of [8,9] where a configuration of D6 branes wrapping the
3
S which appears in the deformation of the conifold singularity of a CalabiYau three-fold
was represented in M-theory as compactification on a certain space with G2 holonomy. For
example, we show that M-theory on an orbifold of the spin bundle over S4 S(S4 ) which
admits a metric with Spin(7) holonomy appears as the strong coupling description of
D6-branes wrapping the supersymmetric four-cycle of the bundle of anti-self-dual twoforms over S4 (S4 ) which admits a G2 holonomy metric. This type of correspondence is
extended to all other cases. Namely, we show that M-theory on certain singular spaces with
SU(3), G2 , SU(4) and Spin(7) holonomy describe in eleven dimensions a configuration of
D6-branes wrapped on a supersymmetric cycle on a manifold with different holonomy
group. For example, we show that M-theory on the CalabiYau three-folds used in [10]
to geometrically engineer supersymmetric gauge theories theories appear as the local
description of D6-branes wrapping two-cycles in manifolds with SU(2) holonomy. We
would like to emphasize that the M-theory descriptions that we propose in this paper are
appropriate descriptions of the Type IIA brane configurations near the location of the D6branes. Globally, the eleven-dimensional geometry is more complicated. In this paper we
will provide simple examples of the local description of the wrapped supersymmetric D6branes configurations in terms of eleven-dimensional geometry.
This phenomena, apart from giving a physical rational for the existence of manifolds
with reduced holonomy, can be useful in deriving string theory dualities using geometrical
transitions. Recently, Atiyah, Maldacena and Vafa [9] have lifted the description of Type
IIA backgrounds on two different CalabiYau three-folds to eleven dimensions and have
found that the two different Type IIA backgrounds are described in M-theory by two
different G2 holonomy spaces which are related to each other by a flop transition [9].
Therefore, by going through a smooth physical transition in M-theory, like a flop in
a G2 holonomy manifold, they derive the physical equivalence of string theory in the two
different CalabiYau manifolds, which was previously conjectured by Vafa [11]. Various
aspects of this duality and generalizations to other three-folds have recently appeared in
[1215]. The lifts proposed in this paper provide quantitative information about the gauge
theory on the branes and can be useful in deriving possible new dualities among Type IIA
vacua by going through a geometrical transition in the M-theory geometries we discuss.
The rest of the paper is organized as follows. Section 2 briefly reviews the elevendimensional description of Type IIA D6-branes in flat space with and without an orientifold
six-plane. In Section 3 we discuss the gauge theories one gets by wrapping D6-branes
on supersymmetric cycles. We also exhibit the general features of the eleven-dimensional
geometry that represents the various wrapped D6-brane configurations. Sections 47 give
specific examples of lifts of Type IIA backgrounds with branes on certain manifolds with
one holonomy group in terms of M-theory on spaces with a different holonomy group.
2. The geometry of D6-branes and the O6-plane

Any solution of the Type IIA string equations of motion only sourcing the Type IIA fields
g , C and is described in eleven dimensions by a purely gravitational background.
A simple example of this phenomena is provided by a collection of D6-branes. The elevendimensional description of N separated D6-branes can be obtained by using (1.1) on the
Type IIA supergravity solution of [16]. It is given by
2
2
= dx02 + dx12 + + dx62 + dsTN
,
ds11
(2.1)
2 is the metric of the Euclidean multi-centered Taub-NUT space [17]

where dsTN

2
2
= H d r 2 + H 1 dx11 + C dx
dsTN
C = H,

H =1+
N
1 gs
,
2
|r ri |
(2.2)
i=1
which is a hyper-Kahler metric on a U (1) bundle over R3 . The x 11 coordinate isperiodic

and absence of conical singularities at r = ri requires its periodicity to be 2gs .
The
circle fiber is identified with the M-theory circle which has the expected radius R = gs
at infinity. Therefore, a collection of D6-branes in flat space can be represented in M-theory
by a four-dimensional manifold with SU(2) holonomy.
In this paper we take the D6-branes to be coincident all ri = 0 which gives
rise to the familiar enhanced SU(N) gauge symmetry. The M-theory description of this
configuration near the location of the D6-branes at r 0 is given by

2
2
H d r 2 + H 1 dx11 + C dx ,
dsALE
Ngs
,
H
(2.3)
2r
which is the metric on the AN1 ALE space [18] near an orbifold singularity. Here the
SU(N) gauge symmetry arises from membranes wrapping the collapsed two-cycles [2]. In
the limit when all ri = 0 the metric on the AN1 ALE space [18] degenerates to the metric
on the C2 /ZN orbifold, where ZN acts by
z1 e
2 i
N
z1 ,
z2 e
2 i
N
z2 .
(2.4)
Three of the coordinates which are acted on are the original transverse directions of the
D6 branes. The fourth orbifolded direction corresponds to the M-theory circle whose
asymptotic radius is now infinity. Therefore, Type IIA string theory with N coincident
D6-branes is locally described in eleven dimensions by the AN1 ALE singularity.
One can also find the M-theory realization of a collection of D6-branes sitting on top of
an orientifold six-plane (O6-plane). The O6-plane appears as the fixed locus obtained by
modding out string theory by the orientifold group G = {1, (1)FL R7 R8 R9 }, where is
worldsheet parity, FL is the left-moving spacetime fermion number and Ri is a reflection
along the x i -th coordinate. In this paper we consider M-theory lifts involving the O6-plane
Table 1
Supersymmetric cycles in irreducible holonomy manifolds
p+1
SU(2)
SU(3)
G2
SU(4)
Spin(7)
2
3
4
5
6
7
8
divisor/SLag
holomorphic
SLag
divisor
associative
coassociative
holomorphic
Cayley
divisor
Cayley
which carries 2 units of D6-brane charge 5 (the O6 plane). N coincident D6-branes on

top of this orientifold plane gives rise to enhanced SO(2N) gauge symmetry. It was shown
in [19,20] that this O6-plane is represented in M-theory by the AtiyahHitchin space [21]
and that the combined system of N D6-branes on top of the O6-plane is described locally
DN2 , where
DN2 is the
by the DN ALE singularity. This is the orbifold singularity C2 /
Dihedral group with a Z2 central extension. 6 The SO(2N) gauge symmetry arises in the
M-theory description from membranes wrapping the shrunken two-cycles at the orbifold
singularity.
To summarize, N coincident D6-branes in flat space are described in eleven dimensions
by an AN1 ALE singularity and N D6-branes on top of an O6 -plane by a DN ALE
singularity. The corresponding orbifold groups act on the normal direction of the D6-branes
and on the M-theory circle. In the rest of the paper we will explore the M-theory description
of D6-branes and D6-branes on top of an O6-plane wrapping non-trivial supersymmetric
cycles in various non-trivial backgrounds.
3. General strategy
We want to consider Type IIA supersymmetric configurations of D6-branes wrapping
cycles in spaces with reduced holonomy and to find the corresponding M-theory description. In order for the D6-branes to preserve some of the supersymmetry left unbroken by
the compactification space, the cycle which is wrapped must be supersymmetric [23,24].
The BPS condition on a supersymmetric cycle on a manifold with holonomy group G can
be written in terms of the corresponding G-structure. The following table summarizes the
classification of supersymmetric cycles (Table 3 in [25]):
All the above cycles preserve one-half of supersymmetry except the Cayley cycles
of a CalabiYau four-fold which instead preserve one quarter of the supersymmetries.
Therefore, the supersymmetric Type IIA backgrounds with wrapped D6-branes which
5 In this paper all charges are measured in the quotient space.
6 This discrete subgroup of SU(2) has order 4N 8 and has two generators a and b which satisfy a 2N4 =
DN2 on C2 can be found, for example, in Table 1 of [22].
b4 = 1 and ba = a 1 b. The action of
can be described by M-theory on a manifold with irreducible holonomy are severely

constrained by supersymmetry.
There are essentially two types of situations that need to be considered:
(1) The D6-branes wrap a supersymmetric cycle in such a way that the D6-branes fill
completely the space transverse to the IIA compactification manifold. The field theory
one obtains lives in the transverse Minkowski space. Then the local M-theory description
is given by a manifold one-dimensional higher in the list of manifolds with irreducible
holonomy. In this fashion one can find examples of the following lifts of holonomy groups
(IIA M): SU(3) G2 and G2 Spin(7).
(2) The D6-branes wrap a supersymmetric cycle in such a way that the D6-branes are
codimension one in the transverse Minkowski space to the IIA compactification manifold.
The field theory one obtains lives in codimension one on the transverse space. Then
the local M-theory description near the D6-branes is given by a manifold of two
dimensions higher in the list of manifolds with irreducible holonomy. In this fashion one
can find examples of the following lifts of holonomy groups (IIA M): SU(2) SU(3)
and SU(3) SU(4).
The effective gauge theory living on the D-branes is a topological field theory [24]. This
topological field theory is a gauge theory in which the scalars parametrizing the positions of
the D-branes are twisted and transform as sections of the normal bundle. 7 Therefore, given
a supersymmetric cycle one can construct the supersymmetric gauge theory by analyzing
the normal bundle of the supersymmetric cycle. In this paper we will concentrate on
supersymmetric cycles which are rigid so that there are no scalars describing the possible
deformations of the cycle inside the curved manifold. When the D6-branes are codimension
one, we get a massless real scalar field which is part of the vector multiplet. It describes the
positions of the D6-branes in the transverse direction. The M-theory realization of these
backgrounds with D6-branes will geometrically engineer these gauge theories.
The local geometry of the wrapped N D6-branes in M-theory is as follows. Near the
location of the D6-branes, the D6-branes are represented in M-theory by the AN1 ALE
singularity. Since some of the transverse directions of the D6-branes are curved and nontrivially fibered over the supersymmetric cycle, the lift to eleven dimensions is just given by
the AN1 ALE singularity fibered over the supersymmetric cycle. As we will show in the
sections that follow we will be able to recognize such geometries as particular manifolds
with irreducible holonomy.
We can also consider the situation in which we mod out the geometry on which the
D6-branes are embedded by an involution . We will require that the involution have
as fixed set the supersymmetric cycle on which the D-branes are wrapped and that the
fixed set satisfies the corresponding supersymmetric cycle condition. In order to obtain
a supersymmetric configuration this involution must be accompanied by an orientifold
projection. The appropriate orientifold group is given 8 by G = {1, (1)FL }. Therefore,
7 Recently, supergravity duals of these topological field theories have been constructed, see, e.g., [26].
8 In the situation where the D6-branes are codimension one in the transverse space, one must also reflect along
the transverse coordinate.
by performing this orientifold projection, one obtains a curved orientifold six-plane (O6plane) with N D-branes on top of it. Following the same argument we used to uplift the
description of the curved D6-branes to M-theory we see that the M-theory description of
the curved D6+O6 configuration is given by the DN ALE singularity fibered over the
corresponding supersymmetric cycle. In the examples we will consider we will identify
this fibration with a certain reduced holonomy manifold.
In the following sections we will provide concrete examples of these lifts.
4. From SU(2) holonomy to SU(3) holonomy via D6-branes

We lift various IIA configurations and recover the CalabiYau three-folds used, for
example, in [10] to geometrically engineer four-dimensional N = 2 gauge theories.
The simplest 9 case to consider is the M-theory description of N D6-branes wrapping the
S2 in T S2 , the cotangent bundle of S2 . This space appears when resolving or deforming
an A1 singularity of K3. For instance, T S2 is described by
z12 + z22 + z32 = r.
(4.1)
If we rewrite zj = aj + ibj for j = 1, 2, 3 and take r to be real and positive, the real
and imaginary
parts of (4.1) lead to a 2 b 2 = r and a b = 0. Since u u = 1, where

u = a / b 2 + r, u generates an S2 . Moreover, since b u = 0, bi spans the cotangent
directions of S2 , so (4.1) describes the cotangent bundle of S2 . This space is topologically
R2 S2 and admits a hyper-Kahler metric with SU(2) holonomy. The S2 is a holomorphic
cycle and branes wrapping it preserve one-half of the supersymmetries left unbroken by
the geometry.
We now consider N D6-branes wrapped on the supersymmetric S2 inside R1,5 T S2
and stretched along the x 1 , . . . , x 4 directions. Since the normal bundle of S2 N(S2 ) is
trivial, the field theory living on the branes is a supersymmetric five-dimensional pure
SU(N) gauge theory with eight real supercharges. 10 Alternatively, one can simply derive
[28] this gauge theory by representing the wrapped D6-branes by N fractional [29] D4branes at a C2 /Z2 orbifold singularity.
Now, in eleven dimensions we must obtain a geometrical background which leads to
such a five-dimensional gauge theory. As explained in the previous section the geometry
we get is that of an AN1 ALE singularity fibered over the S2 . This geometry is indeed
a CalabiYau three-fold [10] and gives rise to a five-dimensional SU(N) gauge theory.
The exact prepotential in the Coulomb branch of this gauge theory was computed from
M-theory on this three-fold geometry in [30]. The computation reduces to evaluating the
classical intersection numbers of the divisors one obtains when resolving the curve of
9 Lifting to eleven dimensions a configuration of D6-branes wrapped on the entire K3 follows trivially from the
discussion on the previous section. The corresponding eleven-dimensional geometry is just R1,2 Taub-NUT
K3.
10 The vector multiplet has a real scalar so there is a non-trivial Coulomb branch whose prepotential is
constrained by gauge invariance to be cubic in the vector superfields [27].
10
AN1 singularities which when blown up break the SU(N) gauge group to its maximal
torus.
In the Type IIA description one can perform an orientifold projection combined with the
following involution

z z i , i = 1, 2, 3,
: i
(4.2)
x5 x5 .
This involution keeps fixed the real part of Eq. (4.1) fixed which describes the
supersymmetric S2 . Therefore, we obtain an O6-plane wrapping the S2 . Then, the gauge
theory on the D6-branes is a five-dimensional SO(2N) gauge theory with eight real
supercharges.
It is simple to incorporate in the M-theory description of this background the effect of
the O6-plane. The geometry that one gets is that of a DN ALE singularity fibered over
the base S2 . This CalabiYau appears in [10] and M-theory on it gives a supersymmetric
SO(2N) gauge theory. By analysing the resolution of this singularity reference [30]
reproduced the exact prepotential in the Coulomb branch of the gauge theory from classical
geometry. 11
It is natural to ask for the M-theory description of wrapped D6-branes when the D6branes wrap a collection of S2 s of a local K3 geometry. We will consider the well known
ADE ALE spaces. The basic geometry is that of a collection of r S2 s whose intersection
form is Iab = Cab where a, b = 1, . . . , r, which is (minus) the Cartan matrix of the
corresponding ADE algebra of rank r. One can consider wrapping Na D6-branes on the
ath S2 for a = 1, . . . , r. In order to obtain a supersymmetric gauge theory on this collection
of D6-branes, one must ensure that all the S2 s are holomorphic (supersymmetric) with
respect to the same complex structure. 12 Then the gauge theory one obtains is a fivedimensional theory with eight real supercharges with gauge group
G=
r

SU(Na ),
(4.3)
a=1
and the following hypermultiplet content

1
b .
Iab Na , N
2 a=b
(4.4)
This gauge theory can be easily derived by representing the wrapped D6-branes as
fractional D4-branes probing the C2 / orbifold singularity, where is the discrete
subgroup of SU(2) whose representation theory can be associated with the extended ADE
Dynkin diagram. 13
11 Both the SU(N ) and SO(2N ) pure gauge theories have a convex prepotential along the Coulomb branch and
have a non-trivial fixed point of the renormalization group at the origin [10,27].
12 For the A ALE spaces, for which an explicit metric is known [18], one can show that the r homology
r
generators are holomorphic with respect to the same complex structure when the r vectors in R3 from which
one constructs the S2 s by fibering these vectors with the U (1) fiber are collinear.
13 A = Z

r
r+1 , Dr = Dr2 ; E6 , E7 and E8 correspond respectively to the Z2 centrally extended tetrahedral,
octohedral and isocahedral groups.
11
The M-theory description of this Type IIA background is as follows. One has a fibration
structure whose base is given by a chain of r S2 s which intersect according to a particular
ADE Dynkin diagram of rank r. Which ADE diagram appears is determined by which
ADE ALE space is used in the Type IIA description. On the ath S2 in the base one
has a ANa 1 ALE singularity fibered over it so the total geometry is that of a chain of
S2 s intersecting according to the ADE Dynkin diagram and on each S2 there is an Atype singularity fibered over it. The type of singularity on a particular S2 in the base is
determined by the number of D6-branes that wrap that cycle in the Type IIA description.
This geometry is a CalabiYau three-fold and M-theory on it results in the required quiver
gauge theory 14 and reproduces the cubic prepotential along the Coulomb branch [10].
We conclude this section by identifying the Type IIA background which lifts to Mtheory on a family of CalabiYau manifolds considered in [10]. In [10] it was shown that
one can generalize the base geometry that we have just discussed such that the base S2 s
D
E
groups. The hypermultiplets are
intersect according to Dynkin diagram of the affine A
D
E.
It is natural to try
determined like in (4.4) but now Cab is the Cartan matrix of A
to identify the Type IIA geometry with SU(2) holonomy which lifts, in the presence of
D6-branes, to this family of CalabiYau manifolds.
r Dynkin diagram, the corresponding
When the base S2 s intersect according to the A
Type IIA geometry is the deformation of the Type Ir+1 fibre singularity of an elliptically
fibered K3. This K3 is described by the Weierstrass form
y 2 = x 3 + a(z)x + b(z),
(4.5)
where (x, y) are affine coordinates on P2 and z is the coordinate on the base of the
elliptic fibration parametrizing a copy of C. The Type Ir+1 singularity appears when the
discriminant of (4.5) has a zero of order r + 1 at z = 0 and both a(z) and b(z) dont have
a zero at z = 0. The deformation of this singularity generates a chain of r + 1 S2 s which
r Dynkin diagram. Wrapping D6-branes on these two-cycles
intersect according to the A
lifts to eleven dimensions to the CalabiYau three-fold we were looking for. One can find
in a similar fashion the Type IIA realization when the S2 s in the base of the three-fold
have a different affine intersection form. This corresponds to studying Type IIA with D6branes on the deformation of certain fiber singularities see, e.g., [22] of an elliptically
fibered K3.
5. From SU(3) holonomy to G2 holonomy via D6-branes

For completeness, we briefly summarize an example of this lift which appeared recently
in [8,9]. The CalabiYau three-fold they considered is T S3 , the cotangent bundle of S3 .
This manifold appears as the deformation of the quadric on C4
z12 + z22 + z32 + z42 = r.
14 The matter appears as usual from loci of enhanced symmetry in the base.
(5.1)
12
In a similar fashion as in (4.1), one may show that for real r (5.1) describes T S3 . One can
now wrap N D6-branes branes over the S3 , which is a Special Lagrangian submanifold.
The branes completely fill the transverse R1,3 Minkowski space and give rise a fourdimensional SU(N) gauge theory with four supercharges.
In [8,9] the local description of this background was described as M-theory on a certain
space with G2 holonomy. The D6-branes lead to an AN1 ALE singularity which is fibered
over the S3 . This geometry can be understood as a ZN orbifold of an R4 fibration over
S3 . Supersymmetry dictates that this space must have G2 holonomy. It is known that the
spin bundle of S3 S(S3 ), whose fibers are topologically R4 , admits a metric with G2
holonomy [31]. Moreover, the G2 structure on this space contains an SU(2)3 symmetry
group. Therefore, modding out by any discrete subgroup of SU(2)3 will preserve the G2
structure. Indeed, it is easy to show that embedding ZN in one of the SU(2)s acts on the
fibers precisely as we expect from the lift of Type IIA D6-branes on T (S3 ). Therefore,
the Type IIA configuration is represented in eleven dimensions by a ZN orbifold of S(S3 )
which has a G2 structure.
Sinha and Vafa [12] considered the Type IIA orientifold background obtained by
modding out by the following involution
: zi z i ,
i = 1, . . . , 4.
(5.2)
This involution gives rise to an O6-plane wrapping the S3 , which is left fixed by (5.2).
The gauge theory on the D6-branes on top of this O6-plane is a four-dimensional N = 1
SO(2N) gauge theory.
From the general discussion in Section 3 such background is described in eleven
dimensions by a DN ALE singularity fibered over S3 . This is obtained by having the
DN2 SU(2), one can mod out S(S3 ) by
discrete group
DN2 act on the R4 fibers. Since

DN2 by appropriately embedding DN2 in one of the SU(2) symmetry groups of S(S3 )
such that the quotient space has a G2 structure and DN2 acts as required on the fibers.
Various aspects of this lift and generalizations to other three-folds have recently
appeared in [1215].
6. From SU(3) holonomy to SU(4) holonomy via D6-branes

In order to accomplish this lift the D6-branes must be codimension one in the transverse
Minkowski space. Therefore, the D6-branes must wrap a divisor in the CalabiYau
three-fold. In this section we will provide a simple example of this lift.
Consider the non-compact CalabiYau geometry described by the O(3) bundle
over P2 . The string sigma model on this three-fold has a simple description in terms of
the linear sigma model approach of [32]. We analyze the vacuum structure of a twodimensional N = (2, 2) U (1) gauge theory with four chiral superfields (z1 , z2 , z3 , z4 )
which carry charges (1, 1, 1, 3) under the gauge group. In the presence of a Fayet
Iliopoulos term r, the solution of the D-term equation of this model is described by
R1,3
M:
|z1 |2 + |z2 |2 + |z3 |2 3|z4 |2 = r.
(6.1)
13
The vacuum of the theory is M/U (1), where the U (1) action is specified by the U (1)
charges of zi .
The phase r > 0 describes an exceptional P2 parametrized by coordinates z1 , z2 , z3
which due to (6.1) cannot vanish simultaneously. The coordinate z4 describes a complex
line fibered over P2 . The vacuum manifold M/U (1) can be identified 15 with the O(3)
bundle over P2 . The exceptional P2 is a holomorphic cycle so branes wrapping it are
supersymmetric.
We consider N D6-branes wrapping the P2 . The field theory on the branes is a threedimensional N = 2 pure SU(N) gauge theory. 16 Classically, this theory has a Coulomb
branch parametrized by N 1 complex scalars which are formed from the real scalars
in the vector multiplet together with the real scalars one gets by dualizing the photons
one gets along the Coulomb branch. Semiclassically, YangMills instantons generate
a superpotential for the complex scalars [35] break supersymmetry.
The local M-theory description of this background is given by an AN1 ALE singularity
fibered over P2 which is a CalabiYau four-fold. Analyzing the zero mode spectrum
leads to a N = 2 three-dimensional pure SU(N) gauge theory. 17 Precisely the four-fold
geometry we have encountered can be used to exactly reproduce the superpotential along
the Coulomb branch of the gauge theory [36]. In the geometrical picture going to the
Coulomb branch corresponds to resolving the singularities on the fiber which lead in the
total geometry to a collection of divisors. The vevs of the complex scalars in the Coulomb
branch correspond to the complexified blow up parameters. The superpotential then arises,
when certain topological conditions [37] are satisfied, by wrapping Euclidean five-branes
over the divisors. It was shown in [36] that the divisors one obtains by derforming the
singularity we discussed reproduces the expected field theory superpotential of the threedimensional SU(N) gauge theory.
We can orientifold the Type IIA background by modding out by the following involution

z4 z4 ,
:
(6.2)
x3 x3 .
The action of on the three-fold (6.1) leaves the P2 fixed and gives rise to a curved
orientifold plane. Therefore, we can have N D6-branes on top of the O6-plane wrapping the
P2 inside the O(3) bundle over P2 . Thus, the gauge theory on the D6-branes is a threedimensional N = 2 pure SO(2N) gauge theory. The local M-theory description of this
Type IIA system is described by the CalabiYau four-fold given by a DN ALE singularity
fibered over P2 . This geometry was used in [36] to reproduce the superpotential on the
Coulomb branch of the gauge theory using five-brane instantons.
15 This CalabiYau is the crepant resolution of the C3 /Z orbifold singularity which appears in the linear sigma
3
model approach in the phase when r < 0; then z4 cannot vanish and the exceptional divisor is blown down. More
details on the moduli space of this model can be found in [33].
16 There is a subtlety in this example. Since P2 is not a Spin manifold we must turn on a half-integral flux on
the D-brane in order to avoid global anomalies [34]. We will return to this issue in the next section.
17 If the complex surface Z over which the singular ALE sits has non-zero Betti numbers h1,0 (Z) and h2,0 (Z)
one gets chiral multiplets [36]. Since h1,0 (P2 ) = h2,0 (P2 ) = 0 we get pure gauge theory.
14
7. From G2 holonomy to Spin(7) holonomy via D6-branes

In order to accomplish this lift the D6-branes must fill the transverse R1,2 Minkowski
space. We must then look for manifolds with G2 holonomy which have a coassociative
four-cycle. In order to avoid complications with tadpoles we will consider non-compact
geometries.
In the literature [31], there are only three complete metrics on seven manifolds which
admit a G2 structure. 18 Of these geometries, two have a coassociative four cycle. 19 They
are metrics on the bundle of anti-self-dual two-forms over the four-manifold Z (Z),
where Z = S4 or Z = P2 . Topologically, the fibers are R3 and Z is a supersymmetric
four-cycle. Therefore, wrapping a collection of D6-branes over Z gives rise to a threedimensional N = 1 gauge theory. Since the normal bundle of Z is trivial 20 one obtains
a pure SU(N) gauge theory. Theories with three-dimensional N = 1 supersymmetry do
not have the familiar constraints of holomorphy of four-dimensional N = 1 field theories,
therefore one cannot make exact statements about their non-perturbative dynamics.
We will start by identifying the M-theory description when the D6-branes wrap the S4
in (S4 ). The M-theory geometry is given by the AN1 ALE singularity fibered over
S4 or equivalently by an R4 bundle over S4 where the cyclic group ZN acts on the R4
fibers. Supersymmetry requires that this space admit a Spin(7) structure. 21 Fortunately,
we can identify this space with an orbifold of the known [31] eight-dimensional space
which admits a complete metric with Spin(7) holonomy. This space is the spin bundle
over S4 S(S4 ), whose fibers are topologically R4 and admits a Spin(7) structure with an
SU(2) Sp(2) symmetry. One can now embed ZN on SU(2) while preserving the Spin(7)
structure. Moreover, this embedding acts geometrically by natural action of ZN on the R4
fibers. Therefore, we have identified the M-theory description of D6-branes wrapping the
S4 in (S4 ) as an abelian orbifold of S(S4 ) which preserves the Spin(7) structure.
One can also consider the case when the D6-branes wrap P2 in (P2 ). Since P2 is not
a spin manifold, the field strength F of the U (N) gauge field on the D6-branes 22 does
not obey conventional Dirac quantization [34]. One has

1
F
=n+ ,
(7.1)
2
2
P1
where n is an integer and P1 generates H2 (P2 , Z). In particular, the flux cannot vanish.
Despite this flux, the effective three-dimensional theory one obtains on the branes is also
a pure SU(N) gauge theory.
18 A G structure is a globally defined three-form which is covariantly constant, closed, co-closed and
2
invariant under the group G2 . G2 is the subgroup of SO(7) which leaves the multiplication table of imaginary
octonions invariant.
19 The other metric is a G holonomy metric on the spin bundle over S3 S(S3 ) which appeared in the previous
2
section. This geometry has an associative three-cycle.
20 The normal bundle is the bundle of anti-self-dual two-forms. Since dim(H 2 (Z) = 0), there are no scalars.
21 A Spin(7) structure is a globally defined self-dual four-form which is covariantly constant, closed and
invariant under the group Spin(7).
22 More precisely F is not a connection on a vector bundle but rather a Spinc structure.
15
The eleven-dimensional description of this Type IIA background is somewhat subtle.

Naively, one gets near the location of the D6-branes an AN1 ALE singularity fibered
over P2 . Supersymmetry suggest, in analogy with the previous example, that it must be
possible to put a metric on an R4 bundle over P2 which is complete and has Spin(7)
holonomy. Moreover, such Spin(7) structure must have an SU(2) symmetry on which one
can embedd ZN in such a way that it acts on the R4 fibers in the usual fashion. We have
not found in the literature a metric such an R4 bundle over P2 with Spin(7) holonomy but
duality suggest that it must exist.
Just like in previous sections it is possible to consider the M-theory description when
there is an orientifold six-plane wrapping Z. The appropriate involution is given by acting
by 1 on the fibers such that Z is fixed by the involution. One then gets a three-dimensional
N = 1 SO(2N) gauge theory on the D-branes. The corresponding M-theory geometry is
given by a DN ALE singularity fibered over Z which admits a Spin(7) structure.
The analysis of lifts to M-theory of wrapped D6-branes on supersymmetric cycles
which are not Spin raises an interesting question. As we have remarked, absence of global
anomalies forces [34], even for a single wrapped D6-brane over P2 , to turn on a halfintegral flux (7.1). In the absence of such a flux one expects the M-theory description to
be given by Taub-NUT space. In the presence of this flux, it is not clear what must be
the appropriate modification to the M-theory solution. It would be interesting to identify 23
the M-theory origin of the constraint on Dirac quantization whenever a D6-brane wraps
a manifold which is not spin but is Spinc .
Acknowledgements
We would like to thank M. Atiyah, A. Brandhuber, R. Corrado, E. Diaconescu,
J. Gauntlett, S. Gukov, L. Motl, H. Ooguri and E. Witten for very useful discussions. Part
of this work was done during the M-theory workshop at the ITP. We would like to thank
the organizers for providing a stimulating atmosphere. This research was supported in part
by the National Science Foundation under Grant No. PHY99-07949 and by the DOE under
grant No. DE-FG03-92-ER 40701.
References
[1] P.K. Townsend, The eleven-dimensional supermembrane revisited, Phys. Lett. B 350 (1995)
184.
[2] E. Witten, String theory dynamics in various dimensions, Nucl. Phys. B 443 (1995) 85.
[3] O. Aharony, S.S. Gubser, J. Maldacena, H. Ooguri, Y. Oz, Large N field theories, string theory
and gravity, Phys. Rep. 323 (2000) 183;
M.J. Duff, TASI lectures on branes, black holes and anti-de-Sitter space, hep-th/9912164.
23 This constraint was derived [38] from an M-theory perspective when one considers instead D4-branes. Then
such a condition can be derived from an eleven-dimensional viewpoint by demanding consistency of the M-theory
five-brane partition function.
16
[4] A. Strominger, Superstrings with torsion, Nucl. Phys. B 274 (1986) 253.
[5] K. Becker, M. Becker, M-theory on eight-manifolds, Nucl. Phys. B 477 (1996) 155.
[6] M. Berger, Sur les groupes dholonomie homogne de varits connexion affines et des
varits riemanniennes, Bull. Soc. Math. France 83 (1955) 279.
[7] S. Sethi, C. Vafa, E. Witten, Constraints on low-dimensional string compactifications, Nucl.
Phys. B 480 (1996) 213, hep-th/9606122.
[8] B.S. Acharya, On realising N = 1 Super-YangMills in M theory, hep-th/0011089.
[9] M. Atiyah, J. Maldacena, C. Vafa, An M-theory flop as a large N duality, hep-th/0011256.
[10] S. Katz, P. Mayr, C. Vafa, Mirror symmetry and exact solution of 4D N = 2 gauge theories I,
Adv. Theor. Math. Phys. 1 (1998) 53.
[11] C. Vafa, Superstrings and topological strings at large N , hep-th/0008142.
[12] S. Sinha, C. Vafa, SO and Sp ChernSimons at large N , hep-th/0012136.
[13] B.S. Acharya, Confining strings from G2 -holonomy spacetimes, hep-th/0101206.
[14] B.S. Acharya, C. Vafa, On domain walls of N = 1 supersymmetric YangMills in four
dimensions, hep-th/0103011.
[15] F. Cachazo, K. Intriligator, C. Vafa, A large N duality via a geometric transition, hepth/0103067.
[16] G.T. Horowitz, A. Strominger, Black strings and p-branes, Nucl. Phys. B 360 (1991) 197.
[17] S. Hawking, Gravitational instantons, Phys. Lett. A 60 (1977) 81;
G.W. Gibbons, S.W. Hawking, Gravitational multi-instantons, Phys. Lett. B 78 (1978) 430.
[18] T. Eguchi, A.J. Hanson, Asymptotically flat self-dual solutions to Euclidean gravity, Phys. Lett.
B 74 (1978) 249;
T. Eguchi, P.B. Gilkey, A.J. Hanson, Gravitation, gauge theories and differential geometry,
Phys. Rep. 66 (1980) 214.
[19] N. Seiberg, IR dynamics on branes and spacetime geometry, Phys. Lett. B 384 (1996) 81.
[20] N. Seiberg, E. Witten, Gauge dynamics and compactification to three dimensions, hepth/9607163.
[21] M.F. Atiyah, N.J. Hitchin, Low-energy scattering of nonabelian monopoles, Phys. Lett. A 107
(1985) 2125;
M.F. Atiyah, N.J. Hitchin, Low-energy scattering of nonabelian magnetic monopoles, Philos.
Trans. R. Soc. London A 315 (1985) 459469.
[22] P.S. Aspinwall, K3 surfaces and string duality, hep-th/9611137.
[23] K. Becker, M. Becker, A. Strominger, Fivebranes, membranes and non-perturbative string
theory, Nucl. Phys. B 456 (1995) 130;
H. Ooguri, Y. Oz, Z. Yin, D-branes on CalabiYau spaces and their mirrors, Nucl. Phys. B 477
(1996) 407;
K. Becker, M. Becker, D.R. Morrison, H. Ooguri, Y. Oz, Z. Yin, Supersymmetric cycles in
exceptional holonomy manifolds and CalabiYau 4-folds, Nucl. Phys. B 480 (1996) 225.
[24] M. Bershadsky, V. Sadov, C. Vafa, D-branes and topological field theories, Nucl. Phys. B 463
(1996) 420.
[25] M. Mario, R. Minasian, G. Moore, A. Strominger, Nonlinear instantons from supersymmetric
p-branes, JHEP 0001 (2000) 005.
[26] J. Maldacena, C. Nez, Supergravity description of field theories on curved manifolds and a no
go theorem, hep-th/0007018;
J. Maldacena, C. Nuez, Towards the large N limit of pure N = 1 super YangMills, Phys. Rev.
Lett. 86 (2001) 588;
B.S. Acharya, J.P. Gauntlett, N. Kim, Fivebranes wrapped on associative three-cycles, hepth/0011190;
J.P. Gauntlett, N. Kim, D. Waldram, M-fivebranes wrapped on supersymmetric cycles, hepth/0012195;
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
17
H. Niemer, Y. Oz, Supergravity and D-branes wrapping supersymmetric 3-cycles, hepth/0011288;

C. Nez, L.Y. Park, M. Schvellinger, T.A. Tran, Supergravity duals of gauge theories from
F (4) gauged supergravity in six dimensions, hep-th/0103080.
N. Seiberg, Five-dimensional SUSY field theories, non-trivial fixed points and string dynamics,
Phys. Lett. B 388 (1996) 753.
M.R. Douglas, G. Moore, D-branes, quivers, and ALE instantons, hep-th/9603167;
C.V. Johnson, R.C. Myers, Aspects of type IIB theory on ALE spaces, Phys. Rev. D 55 (1997)
6382.
J. Polchinski, Tensors from K3 orientifolds, Phys. Rev. D 55 (1997) 6423;
M.R. Douglas, Enhanced gauge symmetry in M(atrix) theory, JHEP 9707 (1997) 004;
D.E. Diaconescu, M.R. Douglas, J. Gomis, Fractional branes and wrapped branes, JHEP 9802
(1998) 013.
K. Intriligator, D.R. Morrison, N. Seiberg, Five-dimensional supersymmetric gauge theories
and degenerations of CalabiYau spaces, Nucl. Phys. B 497 (1997) 56.
R. Bryant, S. Salomon, On the construction of some complete metrics with exceptional
holonomy, Duke Math. J. 58 (1989) 829;
G.W. Gibbons, D.N. Page, C.N. Pope, Einstein metrics on S3 , R3 and R4 bundles, Commun.
Math. Phys. 127 (1990) 529.
E. Witten, Phases of N = 2 theories in two dimensions, Nucl. Phys. B 403 (1993) 159.
D.E. Diaconescu, J. Gomis, Fractional branes and boundary states in orbifold theories,
JHEP 0010 (2000) 001.
D.S. Freed, E. Witten, Anomalies in string theory with D-branes, hep-th/9907189.
I.A. Affleck, J.A. Harvey, E. Witten, Instantons and (super)symmetry breaking in 2 + 1
dimensions, Nucl. Phys. B 206 (1982) 413.
S. Katz, C. Vafa, Geometric engineering of N = 1 quantum field theories, Nucl. Phys. B 497
(1997) 196.
E. Witten, Non-perturbative superpotentials in string theory, Nucl. Phys. B 474 (1996) 343.
E. Witten, Duality relations among topological effects in string theory, JHEP 0005 (2000) 031.

Supersymmetric non-singular fractional D2-branes

and NS-NS 2-branes
M. Cvetic a,e , G.W. Gibbons b,e , H. L c , C.N. Pope d,e
a Department of Physics and Astronomy, University of Pennsylvania, Philadelphia, PA 19104 USA
b DAMTP, Centre for Mathematical Sciences, Cambridge University, Wilberforce Road,
Cambridge CB3 OWA, UK

c Michigan Center for Theoretical Physics, University of Michigan, Ann Arbor, MI 48109, USA
d Center for Theoretical Physics, Texas A&M University, College Station, TX 77843, USA
e Institut Henri Poincar, 11 rue Pierre et Marie Curie, F 75231 Paris Cedex 05, France
Received 6 April 2001; accepted 14 May 2001
Abstract
We obtain regular deformed D2-brane solutions with fractional D2-branes arising as wrapped D4branes. The space transverse to the D2-brane is a complete Ricci-flat 7-manifold of G2 holonomy,
which is asymptotically conical with principal orbits that are topologically CP3 or the flag manifold
SU(3)/(U (1) U (1)). We obtain the solution by first constructing an L2 normalisable harmonic
3-form. We also review a previously-obtained regular deformed D2-brane whose transverse space is
a different 7-manifold of G2 holonomy, with principal orbits that are topologically S 3 S 3 . This
describes D2-branes with fractional NS-NS 2-branes coming from the wrapping of 5-branes, which
is supported by a non-normalisable harmonic 3-form on the 7-manifold. We prove that both types
of solutions are supersymmetric, preserving 1/16 of the maximal supersymmetry and hence that
they are dual to N = 1 three-dimensional gauge theories. In each case, the spectrum for minimallycoupled scalars is discrete, indicating confinement in the infrared region of the dual gauge theories.
We examine resolutions of other branes, and obtain necessary conditions for their regularity. The
resolution of many of these seems to lie beyond supergravity. In the process of studying these
questions, we construct new explicit examples of complete Ricci-flat metrics. 2001 Published
by Elsevier Science B.V.
PACS: 12.10.-g; 12.60.Jv
1. Introduction
In order to make use of the AdS/CFT correspondence [13] to study four-dimensional
YangMills theories with less than maximal supersymmetry, or with no superconformal
E-mail address: pope@physics.tamu.edu (C.N. Pope).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 2 3 6 - X
M. Cvetic et al. / Nuclear Physics B 606 (2001) 1844
19
symmetry, an extensive programme of studying D3-branes in conifold-type backgrounds

has been undertaken. In the original study [4], the flat six-dimensional transverse space of
the usual D3-brane was replaced by the conifold metric [5], which is the Ricci-flat cone
over the 5-dimensional space T 1,1 (a U (1) bundle over S 2 S 2 ). Additionally, since the
T 1,1 space (which is topologically S 2 S 3 ) has a non-trivial 2nd cohomology, it was
possible to wrap D5-branes around the S 2 cycle, giving rise to supergravity duals of the
so-called fractional D3-branes [69]. At the level of the supergravity field theory, the
wrapping of the D5-branes corresponds to having a non-trivial magnetic flux integral for
the R-R 3-form field strength; in other words the field strength is proportional to the volume
form of the 3-sphere. In fact both the R-R and NS-NS 3-forms are non-vanishing in the
solution, taking their values from the real and imaginary parts of a self-dual harmonic 3form. This leads to a non-trivial contribution in the Bianchi identity of the self-dual 5-form
that carries the original D3-brane charge.
The conifold metric is singular at the origin r = 0 (i.e., at the apex of the cone) and
furthermore, the fractional D3-brane solution based on the conifold has a naked singularity
at some positive value of r [4]. Both these problems can be eliminated if one replaces
the conifold by a deformation to a complete Ricci-flat manifold [10]. The six-manifold
is T S 3 (the cotangent bundle of the 3-sphere). Its Ricci-flat metric was first constructed
in [5], and is contained within an extensive class of generalisations obtained by Stenzel
in [11]. It achieves a smoothing-out of the apex of the cone metric, in which the
manifold locally approaches R3 S 3 at the origin. The supergravity solution is expected
to be dual to an N = 1 supersymmetric SU(N) SU(N + m) gauge theory [4,10]. This
theory will not have conformal invariance, since the large-r asymptotic structure of the
fractional D3-brane solution is modified from the usual H = 1 + Q/r 4 form to H =
1 + (Q + m2 log r)/r 4 . This modification to the leading-order fall-off behaviour is a
consequence of the non-normalisability of the harmonic 3-form in the deformed conifold,
which is used for supplying the R-R 3-form flux. As a consequence, the dual field theory
is no longer conformal and the asymptotic behaviour of the gravitational solution for
H correctly reproduces the asymptotic gauge coupling renormalisation of the two SYM
factors [8,9]. Related topics, and the implications of this type of solution, as applied to
dual gauge theories in diverse dimensions, have been studied extensively in various papers
[1221].
Solutions of a somewhat similar kind have been considered previously in different
contexts. An M2-brane in which the transverse 8-manifold is still flat, but in which the
4-form field has an additional term proportional to a singular self-dual harmonic 4-form,
was obtained in [22]. Solutions in which the 8-manifold is replaced by a complete Ricciflat metric were discussed in [23]. Large classes of explicit solutions for a variety of cases,
including M2-branes, D-branes, heterotic 5-branes and self-dual strings were constructed
in [15,18], with attention being focussed on obtaining deformations of the brane solutions
that become regular everywhere. Some warped supergravity reductions related to the M2brane examples were described in [2426].
All of the solutions obtained in [15,18] exploit the transgression terms of the form
dFn = Fp Fq that occur in the Bianchi identities (or field equations) of certain field
20
strengths in the associated supergravity theories. Specifically, the bilinear terms involving
Fp and Fq are taken to be proportional to harmonic forms on the complete Ricci-flat space
transverse to the original undeformed brane solution. Two types of solution can arise: (i)
those where the flux integral of Fp or Fq is non-vanishing, and (ii) those where it vanishes.
The first type gives rise to supergravity duals of fractional branes, with the non-vanishing
flux corresponding to a wrapping of the additional brane around the cycle in the transverse
manifold that is associated with the Hodge dual of the harmonic form. The second type
of solution, with vanishing flux integrals, is not associated with fractional branes. 1 In the
dual gauge-theory picture it was conjectured [18] that deformed solutions of this second
type were related to the Higgs branch of the dual superconformal theory. In [20] a more
extensive dual field theory interpretation of this second type of deformations was given
and evidence was presented that they correspond to perturbations by relevant operators
associated with the pseudo-scalar fields of a dual N = 1 superconformal theory. In addition,
in [20], various kinds of fractional branes were discussed in the singular limit where the
transverse manifold is a Ricci-flat cone. These are analogues of the fractional D3-brane
conifold solution of [4]. They capture the correct large-distance asymptotic behaviour, but
they become singular at short distance.
In this paper we shall focus on resolutions of various fractional branes, thus insisting
that the resolved solution has a non-zero flux integral associated with the additional
fields Fp and Fq . To obtain a regular resolution, one first needs to find a complete
Ricci-flat manifold that can provide a smoothing-out of the apex of the cone. Then, it is
necessary to find a suitable harmonic form in the complete manifold, which can give rise to
deformed solutions with regular short-distance and large-distance behaviour. Specifically,
the harmonic form should be square-integrable at short distance, and give a finite and nonvanishing flux at infinity.
The first cases we shall consider are resolved D2-branes, where the transverse space
is 7-dimensional. One example of such a solution was obtained in [15], making use of
a complete 7-dimensional manifold of G2 holonomy that was obtained in [27,28]. This
manifold is of cohomogeneity one, with level surfaces that are topologically S 3 S 3 .
The manifold is the spin bundle of S 3 . Near the origin, it locally approaches R4 S 3 .
The deformed D2-brane solution has a non-vanishing flux for the NS-NS 5-brane, which
wraps around the S 3 . Thus the solution can be viewed as a fractional NS-NS 2-brane,
together with the usual D2-brane supported by the 4-form. The fractional flux is supported
by an harmonic 3-form that is not L2 normalisable, and the solution, while regular at small
distance, corresponds at large distance to one with a linearly growing overall charge. The
result indicates that asymptotically the renomalisation of g12 g22 may grow linearly
with the energy scale, where g1 and g2 are the gauge couplings of the two SYM factors
of the dual field theory. We shall discuss this solution in more detail in the present paper,
1 We are adopting the notation in this paper that any deformed solution in which there is a non-vanishing flux
integral for an additional field, implying a wrapping of other branes around the associated homology cycle, will
be called a fractional brane, by extension of the terminology for the fractional D3-brane, where D5-branes
wrap around homology 2-cycles. By contrast, solutions without additional flux integrals will just be described as
deformed, but not fractional.
21
in particular showing that it is supersymmetric, with 1/16 of the original supersymmetry

preserved. Thus it corresponds to a dual three-dimensional N = 1 field theory.
We shall also consider the second type of complete Ricci-flat 7-metrics of G2 holonomy
that were obtained in [27,28]. They correspond to R3 bundles over four-dimensional
quaternionic-KhlerEinstein base manifolds M. These spaces are again of cohomogeneity
one, with level surfaces that are S 2 bundles over M (also known as twistor spaces
over M). The two examples of four-dimensional quaternionic-Khler space M that we
shall consider are S 4 and CP2 . Thus the two manifolds have level surfaces that are CP3
(S 2 bundle over S 4 ) or the flag manifold SU(3)/(U (1) U (1)) (S 2 bundle over CP2 ),
respectively. These two manifolds are the bundles of self-dual 2-forms over S 4 or CP2
respectively. They approach R3 S 4 or R3 CP2 locally near the origin. 2
We shall use these manifolds in order to obtain new fractional D2-brane solutions.
In these cases, the deformed D2-brane has a non-vanishing flux for a D4-brane, which
wraps around the S 2 cycle. The solution can therefore be viewed as a fractional D2brane, together with the usual D2-brane. In order to construct the solution we first need
to obtain an harmonic 3-form on the 7-manifold. Although the construction is somewhat
involved, we are able to obtain an explicit normalisable harmonic 3-form in this case. As
a consequence the solution is not only regular at small distance, but also at large distance.
At large distance, in the decoupling limit, it therefore approaches the metric of a regular
D2-brane solution with Euclidean transverse space, and so the dual field theory in the
ultraviolet regime approaches that of the regular D2-brane but with a different overall
charge. We also show that this deformed solution is supersymmetric, preserving 1/16 of
the original supersymmetry thus describing a dual three-dimensional N = 1 field theory.
The above two types of fractional D2-brane examples are discussed in Section 2.
In Section 3 we consider possible resolutions of a larger class of fractional branes,
for which the singular cone configurations were discussed in [20]. Many of the explicit
examples in [20] involved cones over compact manifolds M that are themselves U (1)
bundles over products of complex projective spaces. (Examples include the 7-manifold
Q(1, 1, 1), which is a U (1) bundle over S 2 S 2 S 2 , and M(2, 3), which is a U (1) bundle
over S 2 CP2 .) In fact in [18], classes of complete Ricci-flat metrics were constructed that
can provide resolutions of many such cone metrics. We make use of these here in order to
discuss the possibility of obtaining regular fractional D-strings.
In Section 4, we present a more extensive analysis of the resolutions of the cones
for U (1) bundles over products of EinsteinKhler spaces, obtaining additional explicit
complete Ricci-flat metrics, which go beyond those constructed in [18].
The paper ends with conclusions in Section 5.
2 In what follows, the calculations for the two cases, with the principal orbits being S 2 bundles either over S 4
or over CP2 , proceed essentially identically. We shall in general therefore just refer to the S 2 bundle over S 4
example, with the understanding that all results apply, mutatis mutandis, to the other case too.
22
2. Fractional D2-branes
1
By making use of the transgression term in the equation of motion d(e 2 F4 ) = F(4)
F(3) , one can construct the a deformed D2-brane solution of type IIA supergravity, which
is given by [15]
2
= H 5/8 dx dx + H 3/8 ds72 ,
ds10
F(4) = d 3 x dH 1 + mG(4),
F(3) = mG(3),
1
log H,
4
(2.1)
where G(3) is an harmonic 3-form in the Ricci-flat 7-metric ds72 , and G(4) = G(3) , with
the Hodge dual with respect to ds72 . The function H satisfies
1
H = m2 G2(3) ,
6
(2.2)
where denotes the scalar Laplacian with respect to the transverse 7-metric ds72 . Thus
the deformed D2-brane solution is completely determined by the choice of Ricci-flat 7manifold, and the harmonic 3-form.
The easiest way to determine whether the deformed solution is supersymmetric is to
lift it first to D = 11, and then to examine the supersymmetry of the resulting solution
of eleven-dimensional supergravity. Using the standard KaluzaKlein rules, the metric in
(2.1) becomes
2
= H 2/3 dx dx + H 1/3 ds82 ,
d s11
(2.3)
where
ds82 = ds72 + dz2 ,
(2.4)
and z is the eleventh coordinate. The 4-form in D = 11 is given by

4 = d 3 x dH 1 + mG
(4) ,
F
(2.5)
where
(4) = G(4) + G(3) dz,
G
(2.6)
where, as in (2.1), G(4) = G(3) and still means the seven-dimensional dual. Thus in
D = 11 we have a resolved M2-brane, within the general class discussed in [15], where
(4) is a 4-form that is harmonic and self-dual with respect to the Ricci-flat 8-dimensional
G
metric (2.5). Of course this 8-metric is not asymptotically conical, since the eighth direction
provides just an R factor. Nonetheless, we can apply all the standard D = 11 resolvedbrane formulae for testing the supersymmetry.
In [15,23,25], it is shown that the additional requirement for supersymmetry when the
(4) is turned on is
harmonic 4-form G
abcd bcd = 0,
G
(2.7)
where is a covariantly-constant spinor in the Ricci-flat 8-metric. Thus, in order to check

the supersymmetry of the deformed D2-brane solution, we simply have to lift the harmonic
23
3-form G(3) in the Ricci-flat 7-metric using (2.6), to get a self-dual harmonic 4-form in the
Ricci-flat 8-metric (2.4), and then check the integrability condition (2.7).
In the following subsections, we shall consider various choices for the Ricci-flat 7manifold. We begin with examples using Ricci-flat cones, which are singular at the apex.
Although the fractional D2-brane solutions will be singular, it is useful to study these first
since they capture the large-distance behaviour that will arise in resolved versions of these
manifolds. Then, we consider the two types of resolved examples using complete Ricci-flat
G2 manifolds; firstly, the S 2 bundles over S 4 or CP2 , and then the S 3 bundle over S 3 .
2.1. Fractional D2-branes over Ricci-flat cones
Here, we take the Ricci-flat 7-metric to be
ds72 = dr 2 + r 2 d62 ,
(2.8)
where d62 is the metric on a compact Einstein 6-manifold M6 satisfying Rab = 5gab . As
discussed in [20], one can obtain a fractional D2-brane solution if M6 has a non-trivial
4-cycle around whose dual 2-cycle a D4-brane can wrap. If (4) denotes the associated
harmonic 4-form in M6 we can set G(4) = (4) in (2.1) in order to obtain a solution. One
finds that the function H is given, for a suitable normalisation for (4) , by [20]
Q
m2
(2.9)
.
r 5 4r 6
The fractional D2-brane carries an electric charge Q and a magnetic charge m for F(4) ,
while the 3-form, given by F(3) = mr 4 dr 6 (4) , gives no flux integral.
The solution is analogous to the original fractional D3-brane on the 6-dimensional
conifold, constructed in [4], and it also suffers from a naked singularity at some positive
value of r where H vanishes.
A possible choice for M6 would be CP3 , in which case (4) is just given by J J , where
J is the Khler form of CP3 . In fact the cone over a manifold of CP3 topology admits a
smooth resolution, and in Section 2.2 below, we shall use this to obtain a completely regular
fractional D2-brane with wrapped D4-brane.
A completely different kind of fractional 2-brane can be obtained if M6 has a nontrivial 3-cycle, around whose dual 3-cycle a 5-brane can wrap. In this case, we let G(3) =
(3) in (2.1), where (3) is the harmonic form on M6 associated with the 3-cycle. The
function H is now given by
H = c0 +
H = c0 +
m2
Q
+ .
4r 4 r 5
(2.10)
Since there is now a term with the slower fall-off 1/r 4 , the solution no longer has a welldefined ADM mass. If one nevertheless continues to define a 4-form electric charge to
be proportional to r 6 H , then this gives the r-dependent result
charge = Q +
m2
r,
5
(2.11)
24
where m is the magnetic 5-brane charge carried by F(3) . The electric charge (2.11)
depends linearly on the distance, and this feature, and the associated ill-definement of the
ADM mass, is analogous to the logarithmic dependence of the fractional D3-brane charge
in [4]. The solution can be viewed as the usual D2-brane, together with a fractional NS-NS
2-brane that arises as an NS-NS 5-brane wrapped around the 3-cycle.
This solution is singular at r = 0, which coincides with the horizon. A possible choice
for M6 is S 3 S 3 , with (3) taken to be one of the volume forms of an S 3 factor. If the two
S 3 factors formed a direct product then the solution would be unsatisfactory for a variety of
reasons, including the absence of supersymmetry and also that it would not be resolvable.
However, one can instead take an S 3 bundle over S 3 , and although the bundle is trivial,
implying that it is still topologically S 3 S 3 , the metric will no longer be a direct sum.
This allows a resolution to a complete Ricci-flat 7-manifold, with G2 holonomy, and in fact
it was one of the examples constructed in [27,28]. Topologically, the 7-manifold is the spin
bundle of S 3 , which is R4 S 3 . The corresponding resolved D2-brane was constructed in
[15]. We shall discuss this further in Section 2.3.
An interesting feature of both the resolutions that we shall consider below is that they
cause the original D2-brane charge parameter Q to become related to the parameter m
characterising the charge of the wrapped branes. Thus although Q and m are independent
in the cone metric, they become correlated in the resolutions. In fact, this feature arises too
in the case of the deformed fractional D3-brane in [10].
2.2. Regular fractional D2-brane
In this section we consider a resolution of the fractional D2-brane on the cone over CP3 .
To do this, we need to describe the relevant complete Ricci-flat metric, and also to construct
a suitable harmonic 3-form on the 7-manifold.
The complete Ricci-flat 7-metric on the bundle of self-dual 2-forms over S 4 was
constructed in [27,28]. In the notation of [28], it is given by

2
d s72 = h2 dr 2 + a 2 Di + b2 d42 ,
(2.12)
where i are coordinates on R3 subject to i i = 1, d42 is the metric on the unit 4sphere, with SU(2) YangMills instanton potentials Ai , and
Di di + -ij k Aj k .
(2.13)
The field strengths J i dAi +

Ak satisfy the algebra of the unit quaternions,
j
k
i
J J = ij + -ij k J . A convenient orthonormal basis is
1
j
2 -ij k A
e0 = hdr,
e i = aDi ,
e = be .
(2.14)
(Note that although i runs over 3 values, there are really only two independent vielbeins
on the 2-sphere, because of the constraint i i = 1.)
The metric is Ricci-flat, with G2 holonomy, when the functions h, a and b are given by
[28]

1 1
1
1
1
h2 = 1 4
(2.15)
,
a2 = r 2 1 4 ,
b2 = r 2 .
r
4
r
2
25
The radial coordinate runs from r = 1, where the metric locally approaches R3 S 4 , to the
asymptotically-flat region at r = . The principal orbits at fixed r are CP3 , described as an
S 2 bundle over S 4 . (This is the ur twistor space.) Thus this metric provides a resolution of
the Ricci-flat cone over CP3 . It should be noted, however, that the metric at large distance
is asymptotic to the cone over the squashed Einstein metric on CP3 and not the Fubini
Study Einstein metric. 3
Before considering fractional D2-branes, we can first look for the usual type of D2brane solution but in the background of this Ricci-flat transverse metric. We find that the
radially-symmetric solution to H = 0 is given by

1
1
1
,
F
arcsin
H = c0 +
(2.16)
r(1 r 4 )1/2
r
where

F (|m)

1/2
1 m sin2
d
(2.17)
0
5
is the incomplete elliptic integral
of the first kind. This approaches a constant plus O(r )
at large r, and diverges as 1/ r 1 near r = 1. Thus the D2-brane metric has a singularity
at r = 1, which coincides with the horizon.
Now let us look for a fractional D2-brane, which requires finding a suitable harmonic 3form, which is L2 -integrable at short distance and whose dual 4-form has a non-vanishing
flux integral at infinity. In fact, as we shall see below, we are able to obtain a fully L2 normalisable harmonic 3-form in this example. 4 As in [28], we can make the following
ansatz for the harmonic 3-form:
G(3) = f1 dr X(2) + f2 dr J(2) + f3 X(3) ,
(2.18)
where
1
X(2) -ij k i Dj Dk ,
2
One can easily see that
dX(2) = X(3) ,
dJ(2) = X(3) ,
J(2) i J i ,
dX(3) = 0.
X(3) Di J i .
(2.19)
(2.20)
Imposing the harmonicity conditions dG(3) = 0 and dG(3) = 0, we obtain the equations
(correcting some typographical errors in [28])

f1 b4
f2 a 2
=
4hf
,
= 2hf3 .
f3 = f1 + f2 ,
(2.21)
3
ha 2
h
3 The usual and the squashed Einstein metrics are the members of the family of CP3 metrics ds 2 =
6
2
(Di )2 + d42 with 2 = 1 and 2 = 1/2, respectively [28]. The squashed Einstein metric is Hermitean
but not Khler. In fact, it is nearly Khler (see, for example, [29]).
4 This accords with the fact that above the middle dimension (and hence, by Hodge duality, below the middle
dimension) the L2 cohomology should be topological, and thus isomorphic to ordinary compactly-supported
cohomology. We are grateful to Nigel Hitchin for discussion on this point.
26
(Note that an easy way to determine the Hodge duals of the various forms, which handles
the constraint i i = 1, is by looking in the neighbourhood of a convenient point on the
S 2 fibres, such as the north pole where i = (0, 0, 1), implying that Di = (D1 , D2 , 0)
there.)
It is useful at this stage to note that from the covariantly-constant spinor that exists
in this manifold, we can build a covariantly-constant 3-form which must, therefore, satisfy
Eqs. (2.21). From results in [28], one can show that this has f1 = ha 2 , f2 = hb2 and
f3 = ab2 , and that this does indeed satisfy (2.21). This motivates the following field
redefinitions, namely
f1 = ha 2 u1 ,
f2 = hb2 u2 ,
f3 = ab2 u3 .
(2.22)
Note that this means that the ui functions are the coefficients of wedge-products of the
hatted vielbeins for the 3-form
1
1
i 0
G(3) = u1 i -ij k e0 ej ek + u2 i J
e e -
2
2
1 i
+ J
u3 ei e e .
2
(2.23)
The coordinate transformation x = r 4 puts the equations into the form

du3
+ (3x 1)u3 2xu2 (x 1)u1 = 0,
dx
d((x 1)u2 )
d(xu1 )
= u3 ,
= u3 .
dx
dx
From the last two we can obtain the first integral
4x(x 1)
(x 1)u2 xu1 = c0 ,
(2.24)
(2.25)
where c0 is a constant.
It is now straightforward to obtain an equation purely for u1 . It is advantageous to make
a further coordinate transformation, x = 1/y, and to define u1 = v1 c0 , after which we
find
4y(1 y)2 v1 (1 y)(3 y)v1 2v1 = 0,
(2.26)
where a dot means d/dy. The solution can be expressed in terms of elliptic integrals.
Retracing the steps, we finally arrive at the following expressions for u1 , u2 and u3 in
terms of the original r coordinate:
1
P (r)
+
,
r 4 r 5 (r 4 1)1/2
1
P (r)
,
+
u2 = 4
4
2(r 1) r(r 1)3/2
1
(3r 4 1)P (r)
u3 = 4 4
,
5 4
4r (r 1) 4r (r 1)3/2
u1 =
(2.27)
27
where
1

P (r) = F 12 1 F arcsin r 1 1 =
2
arcsin(r 1 )
d
1 + sin2
(2.28)
Note that the functions ui satisfy u1 + 2u2 + 4u3 = 0.

We have made appropriate choices for the integration constants, in order to ensure that
the functions ui are non-singular at r = 1, and that they fall off at large r. Near r = 1, the
functions have the following behaviour:
3
203
7(r 1) +
(r 1)2 + ,
2
10
7
23
1
u2 = + (r 1) (r 1)2 + ,
4 10
20
9
1 7
u3 = + (r 1) (r 1)2 + .
4 5
2
At large r, the asymptotic behaviour is
u1 =
(2.29)
1
0
1
+ 7 8 + ,
4
r
r
r
1
0
3
u2 = 4 + 7 8 + ,
r
2r
2r
30
1
u3 = 7 + 8 + ,
4r
r
where 0 = F ( 12 | 1) = ( 54 )/ ( 34 ) = 1.3110 . . ..
The magnitude of G(3) is given by

|G(3) |2 = 6 u21 + 2u22 + 4u23 .
u1 =
(2.30)
(2.31)
|2
It is evident from (2.23), and the asymptotic forms (2.29) and (2.30), that|G(3) tends to
a constant at small r, and |G(3)|2 1/r 8 at large r. Since the metric has g r 6 at large
r, it follows that the harmonic 3-form is L2 -normalisable.
We can also examine the flux associated with this harmonic from. Taking its Hodge dual
with respect to the 7-metric, we find
1
G(4) G(3) = u1 r 4 (4) + ,
(2.32)
4
where (4) is the volume form of the unit 4-sphere. We then see from (2.30) that the
integral of G(4) over the 4-sphere at infinity gives a finite and non-vanishing charge

1
1
G(4) = .
P4
(2.33)
4
4
V (S )
This implies that our D2-brane solution carries additional wrapped D4-brane charge.
Having obtained the harmonic 3-form G(3) , we can substitute into (2.1) and obtain the
solution for the resolved fractional D2-brane. It is difficult to give a fully explicit expression
for H , since the harmonic 3-form itself has a rather complicated expression. It is easy to
28
find the asymptotic forms for H at large distance, and at the origin (i.e., at r = 1). They are
given by
Large distance :
Short distance :
Q
m2
+ ,
r 5 4r 6
7m2 (r 1)
H = c1
+ ,
16
H = c0 +
(2.34)
where
m2
Q=
30

dr r 4 r 4 1|G(3) |2 .
(2.35)
The behaviour at large r is indeed the same as in the case of the cone metric, given by
(2.9). However, now we see that the short-distance behaviour is quite different from (2.9),
and in fact the metric is regular there. Note that the D2-brane charge Q, which was a
free parameter in the cone metric solution discussed in Section 2.1, is now quadratically
dependent on the charge m carried by the wrapped D4-brane.
To test the supersymmetry of this resolved D2-brane solution, we apply the procedure
described at the beginning of Section 2, of first lifting it to D = 11. The vielbein
components of the harmonic 3-form in the Ricci-flat 7-metric can be read off from (2.23).
For the purposes of the present purely algebraic calculation, a convenient way to handle
the fact that there are just two, rather than three, independent vielbeins ei , owing to the
constraint i i = 1, is to make a specific choice, such as i = (0, 0, 1), so that we shall
have only e1 and e 2 non-zero. If we let the indices in the S 4 base range of = (3, 4, 5, 6),
then by making a basis choice for the quaternionic-Khler forms, such as
1
1
2
2
3
3
= J56
= J35
= J64
= J36
= J45
= 1,
J34
(2.36)
then we read off from (2.23) that the non-vanishing vielbein components of G(3) are given
by
G012 = u1 ,
G045 = G036 = u2 ,
G156 = G134 = G264 = G235 = u3 .
(2.37)
Using (2.6), we then easily read off the 8-dimensional vielbein components of the harmonic
(4) .
self-dual 4-form G
From results in [28], the covariantly-constant spinor in the Ricci-flat 7-metric satisfies
2 ) = 0 and
integrability conditions that are completely specified by (401 J
1
(402 + J ) = 0 (after making our convenient local basis choice i = (0, 0, 1)).
This translates into
(201 + 64 + 35 ) = (202 56 34 ) = 0.
(2.38)
(4) , obtained as described above, into (2.7), we find after

Substituting the components for G
using (2.38) that the condition for supersymmetry is satisfied provided that
u1 + 2u2 + 4u3 = 0.
(2.39)
29
As noted earlier, the functions ui , obtained in (2.27), do indeed satisfy this condition, and
so we conclude that this resolved fractional D2-brane solution is supersymmetric. To be
precise, we mean by this that turning on the contribution of the harmonic 3-form leads
to no further loss of supersymmetry, over and above the reduction that already occurred
when the flat transverse 7-space of the usual D2-brane was replaced by the Ricci-flat
manifold of G2 holonomy. Thus we have obtained a completely regular supersymmetric
solution describing the usual D2-brane together with a fractional D2-brane coming from
the wrapping of D4-branes around a 2-cycle. As we remarked in footnote 1, we can also get
a completely analogous fractional D2-brane in which the Ricci-flat 7-manifold is replaced
by the related example in [27,28], where the principal orbits are S 2 bundles over CP2
instead of S 4 . We refer to [28] for further details of this example, and for the analogous
conditions on the covariantly-constant spinor, which again lead to the conclusion that the
associated fractional D2-brane will preserve supersymmetry as above.
To summarise, owing to the existence of a normalisable harmonic 3-form on these
Ricci-flat 7-manifolds with the topology of R3 bundles over S 4 or CP2 , the corresponding
fractional D2-brane solutions are regular both at small distance as well as at large distance.
Thus, in the decoupling limit they have the same asymptotic large-distance behaviour as
regular D2-branes with Euclidean transverse spaces. As a consequence, the dual asymptotic
field theory is that of a regular D2-brane (whose original charge Q N determines the
SYM factor SU(N)), but now with a different overall charge Q m2 M, which is
related to the contribution of the additional fluxes of the wrapped D4-branes, and thus
indicating a single SYM factor SU(M). These deformed solutions preserve 1/16 of the
original supersymmetry thus describing a dual three-dimensional N = 1 field theory.
2.3. Regular fractional NS-NS 2-brane
The resolved D2-brane using the 7-manifold of G2 holonomy corresponding to the spin
bundle of S 3 was constructed in [15], but its supersymmetry was not analysed there. Here,
we summarise the results, and then we shall show that it is supersymmetric. The Ricci-flat
metric on the spin bundle of S 3 is given by [27,28]

2
1
ds72 = 2 dr 2 + 2 i i + 2 i2 ,
(2.40)
2
where the functions , and are given by

1
1 1
1
2 = 1 3
,
2 = r2 1 3 ,
9
r
r
2 =
1 2
r .
12
(2.41)
Here i and i are two sets of left-invariant 1-forms on two independent SU(2) group
manifolds. The level surfaces r = constant are therefore S 3 bundles over S 3 . This bundle
is trivial, and so in fact the level surfaces are topologically S 3 S 3 . The radial coordinate
runs from r = a to r = . We define an orthonormal frame by
e0 = dr,
where i i
ei = i ,
1
2 i .
ei = i ,
(2.42)
30
The metric represents a smoothing out of a cone over S 3 S 3 . However, it should be

noted that at large distance the principal orbits approach the squashed Einstein metric
on S 3 S 3 rather than the usual one. These are both members of the homogeneous family
of S 3 S 3 metrics ds62 = 2 i2 + i2 , with 2 = 4/3 and 2 = 4, respectively (see, for
example, [28]). Again, the squashed Einstein metric is nearly Khler, although not Khler
(see, for example, [29]).
Before presenting the fractional D2-brane solution that was obtained in [15], we may
first look at the usual D2-brane solution, but in this Ricci-flat seven-dimensional transversespace background. The radially-symmetric solution to the equation H = 0 is
2

r +r +1
3r
2r + 1
+ log
3
arctan
.
+
2
H = c0 3
(2.43)
r 1
(r 1)2
3
At large r this approaches a constant plus O(r 5 ). Near to r = 1, there are divergences of
order 1/(r 1) and log(r 1). As in the previous example, the metric in the absence of
fractional branes has a singularity at r = 1, which coincides with the horizon.
It was shown in [15] that the metric (2.40) admits an harmonic 3-form given by
1
G(3) = v1 e0 ei ei + v2 -ij k ei ej ek + v3 -ij k ei ej ek ,

6
where the functions v1 , v2 and v3 are given by
v1 =
v3 =
(3r 2 + 2r + 1)
,
r 4 (r 2 + r + 1)2
v2 =
(2.44)
(r 2 + 2r + 3)
,
r(r 2 + r + 1)2
3(r + 1)(r 2 + 1)
.
r 4 (r 2 + r + 1)
(2.45)
It should be noted that these satisfy the relation

3v1 3v2 + v3 = 0.
(2.46)
This harmonic 3-form is square-integrable at short distance, but it gives a linearly

divergent integral at large distance. As shown in [15], the short-distance squareintegrability is enough to give a perfectly regular deformed D2-brane solution, even though
G(3) is not L2 -normalisable. We may also note that it gives a finite and non-vanishing flux,
when integrated over the 3-sphere associated with the metric i2 at infinity, since we have
1
G(3) = r 3 v3 (3) + ,
3 3
(2.47)
leading to a charge
1
P3
V (S 3 )
1
G(3) = .
3
(2.48)
This implies that the D2-brane solution carries additional wrapped NS-NS 5-brane charge.
The solution for the function H in the corresponding deformed D2-brane solution (2.1)
was shown to be given by [15]
31
m2 (r + 1)(16r 7 + 24r 6 + 48r 5 + 47r 4 + 54r 3 + 36r 2 + 18r + 9)

108r 3(r 2 + r + 1)3

2
8m
2r + 1
+ arctan
(2.49)
27 3
3
(after making a rescaling G(3) G(3) /108 for convenience). The function H is perfectly
non-singular for r running from the origin at 1 to infinity. At small distance and large
distance it has the forms

2
m2 2
2
7 7(r 1) + 14(r 1) + ,
r 1 : H = c0 +
27 3
3
m2
4m2
9m2
+
+ .
r : H = c0 + 4
(2.50)
4r
15r 5 28r 7
The electric charge for the 4-form supporting the usual D2-brane is given by
H = c0 +
4m2 m2
+
r.
(2.51)
15
5
As expected, this exhibits the same linear dependence on r as we saw for the cone metric in
Section 2.1. However, the resolution determines that the parameter Q in (2.11), which was
arbitrary for the cone metric, is now given in terms of the parameter m by Q = 4m2 /15.
The linear growth of the overall charge indicates an asymptotic dual field theory with two
SYM factors (which are now modified due to the fractional NS-NS 2-brane contribution),
and the renormalisation of the difference of the coupling couplings may grow linearly with
the energy scale.
Note also, that the small distance behaviour of the above solution is qualitatively the
same as that of the fractional D2-brane discussed in the previous subsection, thus indicating
the same universal infrared behaviour of dual field theories for both types of solutions.
We shall now show that the solution preserves 1/16 of the original supersymmetry, thus
describing a regular supergravity dual of a three-dimensional N = 1 field theory. It was
shown in [28] that the covariantly-constant spinor in this 7-metric of G2 holonomy satisfies
integrability conditions that are completely specified by
charge =
(04 23 ) = (05 31 ) = (06 12 ) = 0.
(2.52)
Following the same procedures as in the previous subsection, we now lift the harmonic
3-form (2.44) to D = 8 using (2.6), and then substitute into the supersymmetry criterion
(2.7). After making use of (2.52), we find that supersymmetry is indeed preserved by virtue
of the linear relation (2.46) amongst the functions v1 , v2 and v3 appearing in (2.44).
Thus we have established that the completely regular resolved D2-brane solution
obtained in [15] is supersymmetric, and that it describes the usual D2-brane together with
fractional NS-NS 2-branes coming from the wrapping of 5-branes around a 3-cycle.
2.4. Spectrum analysis
Having obtained the non-singular supergravity solutions for fractional D2-branes that
are dual to the N = 1 gauge theories in D = 3, it is of interest to study the spectrum of the
32
minimally coupled scalar wave equation in these supergravity backgrounds. In particular,

by examining whether the spectrum is discrete, one can see whether the gauge theory
exhibits confinement in the infrared regime.
Clearly, one cannot expect to be able to solve the wave equations explicitly for these
solutions, owing to their complexity. However, the characteristics of the spectrum can
be determined from the structure of the Schrdinger potential for the minimally-coupled
scalar wave equation. A simple way to obtain the Schrdinger potential is to reduce the
solutions on the principal orbits so that they become D = 4 domain walls. The resulting
4-dimensional metric can be cast into a conformal frame, of the form

ds42 = e2A(z) dx dx + dz2 .
(2.53)
The Schrdinger potential is then given by
V (z) = A + (A )2 .
(2.54)
For both our fractional D2-brane solutions, the coordinate z runs from 0 to a finite value z .
The behaviours of A(z) and V (z) near z = 0 and z = z are given by
Fractional D2-brane:
r , z 0:
r 1, z z :
1
,
z7/3
e2A z2 ,
e2A
91
,
36z2
V 0,
Fractional NS-NS 2-brane:

r , z 0:
e2A
1
,
z4
6
z2
3
.
(2.55)
4(z z )2
(These correspond to the G2 holonomy 7-metrics for the bundle of self-dual 2-forms
over S 4 , and the spin bundle of S 3 , respectively.) Thus both Schrdinger potentials are of
infinite-well type, and so the spectra for both solutions are discrete, indicating confinement
for each of the corresponding dual N = 1, D = 3 gauge theories. Again, the results of this
subsection indicate that the spectra for both cases have qualitatively the same behaviour.
r 1, z z :
e2A z3 ,
3. Resolution of other fractional branes

As we discussed in the introduction, fractional branes were constructed by making use
of the Bianchi identity (or the equation of motion) dFn = Fp Fq . The solution has a
transverse space that is a complete, non-compact Ricci-flat manifold of dimension n + 1 =
p + q. In order for Fn to carry point-like charge (as opposed to a distribution of charge),
the Ricci-flat metric at large distance should approach the conical form
2
dsn+1
= dr 2 + r 2 dn2 ,
(3.1)
where dn2 is an Einstein metric (with Rab = (n 1)gab ) on a compact n-dimensional

manifold Mn . Furthermore, if either Fp or Fq is to have a non-vanishing flux, then
33
Mn must have either a non-trivial p-cycle or q-cycle. Without loss of generality, let us
consider the case of a p-cycle, for which Fp will carry the non-trivial flux provided by an
harmonic p-form. This can always be achieved by choosing Mn = S p S q1 . However,
this configuration will not in general be supersymmetric, and in fact supersymmetric
backgrounds of this kind are uncommon.
The large-distance behaviour of the metric function H in the fractional branes is,
however, universal, given for p = 1 by

m2
Q
m2 /Q
Q
=
1
+
,
H n1 +
(3.2)
r
(q p)(p 1)r 2p2 r n1
(q p)(p 1)r pq
and for p = 1 by
H
Q
r n1
m2 log r
.
(q 1)
(3.3)
Solutions with p = 0 and p = 1 would have naked singularities at large distance, at the
radius where H vanished, and so it seems necessary to consider cases with p 2. For
these, if p > q the solutions have a well-defined ADM mass and the flux for Fn is constant.
In other words, there is no contribution from the Fp Fq transgression term to the n-form
flux. If p q, the solutions do not have a well-defined ADM mass, the transgressions terms
contribute to the n-form flux, and furthermore it is r-dependent. In particular, if p = q the
dependence is logarithmic, since now (3.2) becomes
Q + m2 log r
.
(3.4)
r n1
For p 2, all the fractional branes are well behaved at large distance. If one takes the
cone solutions as they are, they all suffer from singularities at small distance. In particular,
for p q the singularity is naked, at some positive value of r, whilst for p < q the
singularity coincides with the horizon at r = 0.
In order to avoid this singularity, it is necessary that the harmonic p-form be L2
normalisable at short distance, in the (n + 1)-dimensional Ricci-flat manifold [15,18].
This implies that the Ricci-flat metric should interpolate between the cone metric at large
distance and the following metric at small distance:
H
2
2
dsn+1
= d 2 + 2 ds2 + dqs1
+ dp2 ,
(3.5)
where s can take some value in the range 1 to q 1. In other words, there should be a noncollapsing p-surface. Indeed the resolved fractional D3-brane in [10] and the D2-branes
we discussed in the previous section satisfy these criteria.
In [14,21], alternative resolutions of the T 1,1 conifold were considered, for which the
short-distance behaviour of the Ricci-flat 6-metric is

ds62 = d 2 + 2 2 + 2 + C21 d12 + 2 + C22 d22 ,
(3.6)
where d12 and d22 are two 2-sphere metrics, and d = 1 + 2 , the sum of the volume
forms of the two 2-sphere metrics. In this resolution there is no non-collapsing 3-surface.
34
As a consequence, the resulting fractional D3-brane solutions have naked singularities at

small distance.
It is rather non-trivial to construct a fractional brane solution that satisfies both of the
above criteria, namely at large distance and at small distance. The only examples we know
so far are the deformed D3-brane in [10] and the two D2-branes that were obtained in [15]
and in this paper.
For other branes, two cases can be considered. One possibility is that we can sacrifice
the large-distance criterion that Fp carry non-vanishing flux, but still insist that the solution
is regular everywhere. A large class of supersymmetric non-singular solutions of this type
were constructed in [15,18]. Since these solutions carry only the flux of Fn , but with no
flux for Fp (or Fq ), the resulting configurations are likely to describe perturbations of
relevant operators in the dual field theories [20]. The second possibility is that one can
instead sacrifice the short-distance criterion, and hope that the resulting naked singularity
can be resolved by non-perturbative string effects, such as discussed in [30].
In the light of this discussion, we shall now examine the possibilities for deformations
of various branes in D = 10 and D = 11.
The M2-brane has an 8-dimensional transverse space. In order to construct a fractional
M2-brane, it would be necessary to have a Ricci-flat 8-manifold whose principal orbits
(the 7-dimensional level-surfaces) have a non-vanishing 3-cycle. Unfortunately, such a
configuration that is also supersymmetric does not seem to exist [20]. Although the largedistance criterion for the M2-brane cannot apparently be met, the short-distance criterion
can often be satisfied. A large class of Ricci-flat metrics with L2 -normalisable harmonic
self-dual 4-forms were obtained in [15,18], and these were used to construct various
supersymmetric, non-singular deformed M2-branes.
The discussion for the NS-NS string in type IIA is the same as for the M2-brane, since
it can be obtained by double-dimensional reduction. The situation for the NS-NS string in
type IIB is rather different. It is S-dual to the D-string, which we shall discuss presently.
Fractional D-branes were discussed in [15,20], and their large-distance asymptotic
behaviours were classified in [20]. The largest fractional D-brane in supergravity is the
D6-brane, making use of the Bianchi identity dF(2) = mF(3) , where m is the cosmological
parameter in the massive type IIA supergravity. For D5-branes and D6-branes we have p =
0 and p = 1, and hence the solutions suffer from naked singularities at large distance, as we
discussed previously. For the fractional D4-brane, the transverse space is 5-dimensional.
Since there is no suitable non-trivial (irreducible) smooth Ricci-flat manifold that admits
covariantly-constant spinors, the resolution of the singularity lies beyond the level of
supergravity. Resolutions for fractional D3-branes and D2-branes can be successfully
implemented at the level of supergravity, and were discussed in detail in [10] (for D3branes), and in [15] and this paper (for D2-branes).
3.1. Fractional D-strings
The fractional D-string has an 8-dimensional transverse space. Although there seem not
to be examples where the principal orbits are 7-manifolds with non-trivial 3-cycles, thus
35
precluding the construction of fractional M2-branes, there are examples with non-trivial
2-cycles. A D3-brane can wrap around such a 2-cycle, giving the possibility of a fractional
D-string, provided that a corresponding harmonic 3-form exists on the 8-manifold. In order
to avoid the singularity at short distance, the 8-metric should have the following shortdistance behaviour:
ds82 = d 2 + 2 d22 + d52 .
(3.7)
This condition is necessary for the harmonic 3-form (or 5-form) to be square-integrable at
short distance, which would then avoid the short-distance singularity.
Many Ricci-flat 8-manifolds are known, such as the hyper-Khler Calabi metric, the
Stenzel metric on T S 4 , the complex line bundles over CP3 or the 6-dimensional flag
manifold SU(3)/(U (1) U (1), and the Ricci-flat metric of Spin(7) holonomy on the R4
bundle over S 4 . Other examples include various cases with principal orbits that are U (1)
bundles over products of EinsteinKhler spaces, such as CP1 CP1 CP1 or CP1 CP2 .
In these cases at most one of the CPm factors collapses to zero radius at short distance,
implying that the manifold is of the form of a Ck bundle over the remaining non-collapsed
factors. The complex line bundles are particular examples, with k = 1.
The Ck bundle metrics starting from CP1 CP1 CP1 have the following short-distance
structure:

ds8 = d 2 + 2 2 + 2 + C21 d12 + 2 + C22 d22 + 2 + C23 d32,
(3.8)
where d is equal to the sum of the volume forms on the three CP1 factors. If all three
Ci parameters are non-zero, then k = 1 and the manifold is a complex line bundle over
CP1 CP1 CP1 . If in addition any two parameters Ci are equal, the corresponding
CP1 CP1 factors can be replaced by CP2 (see Section 4 for a discussion of some global
issues). If all three Ci parameters are equal, the CP1 CP1 CP1 can be replaced by
any other EinsteinKhler 6-manifold, such as CP3 or the 6-dimensional flag manifold
SU(3)/(U (1) U (1)). At most one of the Ci parameters in (3.8) could instead be zero,
giving a C2 bundle over CP1 CP1 . Again, if the remaining non-zero parameters are then
set equal, the CP1 CP1 could be replaced by CP2 . If parameters are set equal first, and
a corresponding replacement by CP2 or CP3 made, one can set this parameter to zero and
get a C3 bundle over CP1 , or C4 itself (flat space). (Again, see Section 4 for a discussion
of some global issues.)
The Ricci-flat metrics on these Ck bundles over products of EinsteinKhler spaces
M1 M2 MN provide various resolutions of the cone metrics of the 7-spaces
which are U (1) bundles over M1 M2 MN . These cone metrics were discussed
in [20], as candidates for obtaining fractional D-strings. Their large-distance asymptotic
behaviour is given by (3.2), with n = 7, p = 5, q = 2. Although the Ck bundle metrics give
smooth resolutions of the corresponding cone metrics, their short-distance behaviour is
not, however, appropriate for allowing regular short-distance behaviour in the fractional Dstring solutions. This is because, as can be seen from (3.8), they do not have non-collapsing
5-cycles at short distance. Thus they cannot admit appropriate harmonic 5-forms that could
give rise to deformed D-strings with regular short-distance behaviour. It is, however, still
36
of interest to study the short-distance singularity structure for the deformed fractional Dstrings using these resolutions of the cone metrics.
Harmonic 5-forms for the metric (3.8) are given by
G(5) = (x1 1 2 + x2 1 3 + x2 2 3 ),
(3.9)
where x1 + x2 + x3 = 0. The magnitude of G(5) is then proportional to

|G(5) |2
x12 ( 2 + C21 )
( 2 + C22 )( 2 + C23 )
+ cyclic,
(3.10)
and the determinant of the metric gives g ( 2 +C21 )( 2 +C22 )( 2 +C23 ). Thus it follows
that in the various distinct cases with vanishing or non-vanishing Ci one has
Ci = (C1 , C2 , C3 ):
H c0 m2 (log r)2 ,
m2
,
r4
m2
H c0 8 ,
Ci = (C1 , 0, 0):
r
m2
Ci = (0, 0, 0):
(3.11)
H c0 8 .
r
(If more than one Ci parameter vanishes, one would need at least to replace the associated
S 2 factors by a CP2 or CP3 in order to avoid power-law curvature singularities at r = 0.)
Presumably it could only be through non-perturbative string effects that any of the shortdistance singularities in H could be resolved.
Ci = (C1 , C2 , 0):
H c0
4. Further Ricci-flat metrics on Ck bundles

In [18], a rather general class of non-compact Ricci flat manifolds was constructed.
These are metrics of cohomogeneity one, whose principal orbits are U (1) bundles over
products of an arbitrary number N of compact EinsteinKhler manifolds Mi . Typically,
one might take each Mi factor to be a complex projective space, CPmi . The Ricci-flat
metrics constructed in [18] all had the feature that the radius of the U (1) fibres and one of
the Mi factors, say M1 , collapsed to zero at small distance. The factor M1 was therefore
necessarily CPm1 , so that the collapsing submanifold was S 2m1 +1 , and the metric could
approach R2m1 +2 M2 M3 MN locally at short distance. The manifolds were
therefore Ck bundles over M2 M3 MN , with k = m1 + 1. We refer back to [18]
for a detailed discussion of the construction of these metrics.
In this section we generalise the construction a little, to include the case where all
the Mi factors remain uncollapsed at short distance, while the U (1) fibres still shrink
to zero. The structure of the manifold will therefore now be a complex line bundle over
M1 M2 MN . The metric ansatz, and the equations for Ricci-flatness, will be the
same as in [18]. Thus we have

d s2 = e2 dr 2 + e2 2 +
(4.1)
e2i dsi2 ,
i
where dsi2 denotes the EinsteinKhler metric on Mi , and

Ai ,
= d +
37
(4.2)
with dAi = pi J i , where J i is the Khler form on Mi and pi is an appropriately-chosen

constant (see [18]). The functions i and depend only on r. (Note that we have made a
different coordinate gauge choice for the radial variable, as compared with the one called
r in [18].)
From results in [18], it follows that the metric will be Ricci flat if the following first-order
equations, derivable from a superpotential, are satisfied:
1
d
1
di
= pi e2i ,
= ke2
ni pi e2i ,
(4.3)
dr
2
dr
4
i
where ni is the real dimension of Mi . Note that k is a constant such that

i = kpi
(4.4)
for all i, where i is the cosmological constant for dsi2 [18].

We now solve the first-order equations (4.3). For i , we find

e2i = pi r + C2i ,
(4.5)
where the Ci are arbitrary constants of integration. The equation for can then be solved,
to give
e
= 2k
ni /2
+ C2i
r
dy
y + C2j
nj /2
(4.6)
The integration is elementary, giving an expression for e2 as a rational function of r for

any given choice of the integers ni , but the general expression for arbitrary dimensions ni
requires the use of hypergeometric functions. The Ricci-flat metrics are therefore given by

d s2 = e2 dr 2 + e2 2 +
(4.7)
pi r + C2i dsi2 ,
i
with given by (4.6).

The radial coordinate r always runs from r = 0 to r = . One easily sees that as r tends
to infinity, the metric (4.1) tends to
d s2 = d 2 + 2 d s2 ,
(4.8)
where
d s2 =
4k 2 2
k
+
pi dsi2 ,
D
D2
i
(4.9)

D = 2 + i ni denotes the total dimension of d s2 , and we have defined a new radial
coordinate given by r = k 2 /D. It is easy to verify that d s2 is an Einstein metric on the
38

U (1) bundle over i Mi , with R ab = (D 2)gab . Thus (4.1) approaches the cone metric
at large r.
It is appropriate at this juncture to discuss certain global and topological issues. These
concern the periodicity of the U (1) fibre coordinate , and the conditions under which one
obtains a regular manifold. The discussion divides into two parts. Firstly, we must ensure
that the principal orbits themselves are regular. These are the level surfaces with r > 0, for
which all of the metric functions e2 and e2i are positive. The level surfaces are U (1)

bundles over the product i Mi of EinsteinKhler base manifolds. Having established
when the principal orbits are regular, there is a remaining global question of whether the
metric behaves appropriately as the orbits degenerate at r = 0, so as to give a smooth
manifold.
Considering first the principal orbits, the period of the U (1) fibre coordinate must
be compatible with the integrals of the curvature of the fibre over all 2-cycles in the base

space i Mi . Specifically, using (4.2), and recalling that dAi = pi J i , this means that we
must have

pi
J i , for all the i, j,
=
(4.10)
qij
Sij
where Sij denotes the j th 2-cycle in the manifold Mi , and qij are integers. The manifold of

the U (1) bundle over i Mi will be simply connected if these integers are chosen to be as
small possible; let us denote this manifold by M. One can also have non-simply-connected
smooth manifolds in which is taken to be this maximum allowed period divided by
any integer s; these will be the manifolds M/Zs .
Since each Mi is EinsteinKhler, with Khler form J i and cosmological constant i ,
we can write its Ricci form as Ri = i J i . The Ricci form is 2 times the first Chern class,
and so

1
(4.11)
Ri = hij ,
2
Sij
where hij are a set of integers, labelled by j , characteristic of each manifold Mi , and
determined purely by its topology. Bearing in mind the relation (4.4), it is then easy to see
that the maximum allowed period for , compatible with all the integrals over 2-cycles,
will be 5
()max =
2
gcd(hij ),
k
(4.12)
where gcd(hij ) denotes the greatest common divisor of all the (non-vanishing) integers
hij . With this period, the principal orbits M will be simply-connected; one can instead
5 Note that the method we have used in order to obtain this topological information involves the use of the
metric on the base manifold. We have done this for convenience, since it provides a simple way to obtain the
results, but it is worth
emphasising that it is possible instead to obtain the same results without needing to make
use of the metric on i Mi at all.
39
take the period to be ()max /s for any integer s, giving non-simply-connected principal
orbits M/Zs .
The situation becomes simple if all the factors in the base space are taken to be complex
projective spaces, Mi = CP mi , since then there is only one 2-cycle in each factor Mi , and
furthermore we know that the associated integer hi1 is given by hi1 = mi + 1. In fact, it is
convenient in this case to make convention choices so that we have
i = 2mi + 2,
pi = mi + 1,
k = 2.
(4.13)
We therefore have that the maximum period is given by

()max = gcd(m1 + 1, m2 + 1, . . . , mN + 1).
(4.14)
The discussion above was concerned with the conditions for regularity of the principal
orbits themselves. There are further regularity considerations involving the structure of the
metric near to r = 0. The discussion bifurcates, depending on whether all the Ci are nonvanishing or not. In fact it is easy to see that regularity at r = 0 implies that at most one of
the Ci can be zero.
Consider first the case where all the Ci are non-zero. Introducing a new radial coordinate
defined by r = k 2 /2, we find that near = 0 the metric (4.1) approaches

d s2 = d 2 + k 2 2 2 +
(4.15)
pi C2i dsi2 .
i

This will be regular at = 0, approaching R2 i Mi locally, provided that the U (1)
fibre coordinate has period 2/k. From (4.12), we see that this is always possible.
Generically, the set of integers hij will have no common divisor other than 1, and so the
case with simply-connected principal orbits will be the regular one. If instead there is a
greatest common divisor s, then the principal orbits will need to be factored by Zs to get
the regular total Ricci-flat manifold.
For the special cases where there is a single EinsteinKhler base-space factor M1 , the
Ricci-flat metrics of this type fall into the class constructed in [31,32]. An example would
be the line bundle over the EinsteinKhler metric on S 2 S 2 . A new special case with
two factors, each of which is S 2 , but with parameters C1 and C2 now unequal, was recently
obtained in [21] (in a different system of coordinates).
Consider now the case where one of the constants Ci is zero; this class of metrics was
discussed in [18]. Without loss of generality, we may assume that C1 = 0. In terms of a
new radial variable , defined by r = k 2 /(n1 + 2), we find that near = 0 the metric
(4.1) approaches
d s2 = d 2 +
4k 2
p1 k 2 2
2 2
ds1 +
+
pi C2i dsi2 .
(n1 + 2)2
n1 + 2
(4.16)
i2
For this to be regular at r = 0, it is necessary that the terms involving 2 and ds12 give rise
to a sphere. A sine qua non for this to happen is that the manifold M1 must be the complex
projective space CP m1 with its FubiniStudy metric. However, it is also necessary that the
maximum allowed period for , determined by (4.12), be equal to the maximum period
40
that would be allowed if there were no additional factors M2 , M3 , . . . in the base space,
since only then will we get S 2m1 +1 itself, rather than a factoring of it. In other words,
regularity at r = 0 requires that
m1 + 1 = gcd(m1 + 1, h2j , h3j , . . . , hNj ).
(4.17)
For example, in the case where all the base-space factors are complex projective spaces,
Mi = CP mi , we must have
m1 + 1 = gcd(m1 + 1, m2 + 1, m3 + 1, . . . , mN + 1).
(4.18)
If, for instance, we have M1 = CP1 = S 2 , then we can have CP1 , CP3 , CP5 , CP7 , etc.,
for the other factors, but not CP2 , CP4 , etc. If we have M1 = CP2 , then we can have CP2 ,
CP5 , CP8 , CP11 , etc., for the other factors, but not CP1 , CP3 , CP4 , CP6 , CP7 , etc. Note
that Mi = CP m , with all factors the same, is allowed for any m.
The Ci parameters are moduli that parameterise the radii of cycles in the total manifold.
Setting one of the Ci to zero, which is a smooth transformation in the modulus space,
however corresponds to a topology-changing transformation in the manifold of the Ck

bundle over i Mi .
5. Conclusions
The principal focus of this paper was a study of the resolution of brane configurations
with additional fluxes that have non-vanishing integrals at infinity (i.e., non-vanishing
charges). These configurations, which also have an interpretation in terms of the wrapping
of branes around the associated cycles, are referred to as fractional branes. The transverse
space is a complete non-compact Ricci-flat manifold.
The examples we studied in greatest detail were associated with fractional D2-branes.
The space transverse to a D2-brane is a seven-dimensional Ricci-flat manifold. In order
to describe coincident D2-branes rather than an array, this seven-manifold should be
asymptotically conical. When the level surfaces (or principal orbits) have non-trivial
homology, the possibility arises that other kinds of brane can wrap around the cycles.
The structure of the transgression terms, and supersymmetry, imply that only D4-branes
or NS-NS 5-branes can wrap; around 2-cycles or 3-cycles, respectively. The wrapped D4branes are referred to as fractional D2-branes, and they are supported by magnetic charges
carried by the 4-form field strength of the type IIA theory. This contrasts with the electric
charges carried by the usual D2-branes. The wrapped NS-NS 5-branes can be viewed as
fractional NS-NS 2-branes, since they are supported by magnetic charges carried by the
NS-NS 3-form.
In this paper we constructed completely regular fractional D2-brane solutions, using for
the transverse space a Ricci-flat 7-manifold of G2 holonomy that was obtained in [27,
28]. This manifold has level surfaces that are topologically CP3 (arising as an S 2 bundle
over S 4 ), and can be described as the bundle of self-dual 2-forms over S 4 . (A second
possibility is to take the analogous 7-manifold in which the S 4 is replaced by CP2 , leading
41
to similar conclusions.) In order to obtain such a resolved fractional D2-brane solution we

first constructed an L2 -normalisable harmonic 3-form. At large distance, in the decoupling
limit, the solution has the same asymptotic form as that of the regular D2-brane with a
flat transverse space, however, its charge is determined by the charge of the fractional D2brane. This result indicates that the asymptotic (ultraviolet) limit of the dual field theory is
the same as for the regular D2-brane. However the SYM factor is governed by the fractional
brane charge.
We then discussed the example of a fractional NS-NS 2-brane, for which a regular
solution was obtained previously in [15]. This uses another Ricci-flat manifold of G2
holonomy that was found in [27,28], for which the level surfaces are topologically S 3 S 3 .
Note that in this case while the harmonic 3-form G(3) that supports the NS-NS 5-brane
flux is square-integrable at small distance, it is not L2 -normalisable, owing to a linear
divergence of the integral of |G(3)|2 at large distance. This means that asymptotically, the
overall charge of this resolved solution with NS-NS 5-branes wrapped over 3-cycles
grows linearly with the distance. Note that, as in the previous example with D4-branes
wrapped over 2-cycles, the constant part of the charge is determined by the fractional brane
charge (NS-NS 2-brane charge in this latter case). The linear dependence of the charge on
the distance indicates that in the dual field theory the asymptotic renormalisation of the
difference of gauge couplings may grow linearly with the energy scale.
We also showed that both the above resolved D2-brane solutions are supersymmetric.
Specifically, the fraction of preserved supersymmetry in each case is the same as the
fraction that is preserved by the usual D2-brane, with no fractionally charged branes, after
taking into account the supersymmetry reduction already implied by the replacement of the
flat transverse 7-space by the Ricci-flat manifold of G2 holonomy. Thus the supersymmetry
on the world volume of the D2-brane will be N = 1, as opposed to the usual N = 8 of a
normal D2-brane with a flat transverse space. This result implies that the supergravity
solutions are dual to N = 1 three-dimensional gauge theories. Note that the two types of
resolved D2-brane solutions, i.e., those with fractional D2-branes and those with fractional
NS-NS 2-branes, have qualitatively the same small-distance behaviour, thus indicating the
universality of the dual N = 1 field theory in the infrared regime. On the other hand
the ultraviolet behaviour of these two types of solutions is qualitatively different, and of
course neither of them has any restoration of conformal symmetry. 6 We have also analysed
the spectra of the minimally-coupled scalar equations in both the resolved D2-brane
backgrounds. The spectra are discrete and qualitatively similar, indicating confinement in
the infrared regime of the dual three-dimensional field theories.
Our subsequent additional analysis shows that regular supersymmetric fractional brane
solutions are rather rare in supergravities. The only other example known to date is the
deformed fractional D3-brane obtained in [10]. We examined possible resolutions for
other fractional branes, and derived necessary conditions for regularity, but these were
6 We should contrast these results with those for the resolved M2-branes in [15,18], for which the conformal
symmetry of AdS4 M7 is restored asymptotically, in the decoupling limit. Since the resolved M2-brane
examples have no charge associated with the additional fluxes, there are no fractional branes there and the dual
field theory interpretation is that of a three-dimensional CFT perturbed by relevant operators [20].
42
not fulfilled in any of these other examples. It is quite possible that the resolutions of these
other examples involve a resort to non-perturbative effects in string theory, such as are
discussed in [30].
For some of the other examples that we studied, we took as their starting point singular
deformations that employ the cone metrics discussed in [20]. In these examples the level
surfaces in the cone are U (1) bundles over products of complex projective spaces. In fact
a large class of complete Ricci-flat manifolds that smooth out the singularity at the apex
of the cone were discussed in [18], and in this paper we used these in order to try to
obtain resolutions of the cone deformations discussed in [20]. These did not yield regular
resolutions for fractional D-strings, however, owing to the fact that none of them has noncollapsing 5-cycles at short distance. As a by-product we also extended the class of Ricciflat metrics constructed in [18].
Various directions for further investigation remain. Firstly, it would be very interesting to
investigate further the dual field theories for the resolved D2-branes with fractional branes.
Secondly, it would be nice to establish whether further examples of regular supersymmetric
fractional branes exist at the level of supergravity. Furthermore, for cases where this is not
possible, one would like to gain a deeper understanding of how a non-perturbative string
mechanism that could resolve the problems might operate.
Finally, there also exist different types of regular supersymmetric resolved brane
solutions, that again make use of the transgression terms, and harmonic forms on the
transverse Ricci-flat manifolds, but which have no charges associated with the additional
fluxes and thus do not correspond to fractional branes. Many explicit examples, such as
resolved M2-branes, heterotic 5-brane and dyonic strings, were constructed in [15,18]
and they all restore the asymptotic conformal symmetry in the decoupling limit. Dual
field theories were conjectured to correspond to the Higgs phase [18], and evidence was
presentated [20] which shows that in the resolved M2-brane examples the dual CFT is
perturbed by relevant operators. It would be of interest to investigate whether further such
solutions exist, and to gain a better understanding of their rle in the framework of the
AdS/CFT correspondence.
Acknowledgements
We should like to thank Nigel Hitchin, Igor Klebanov and Edward Witten for
discussions. M.C. is supported in part by DOE grant DE-FG02-95ER40893 and NATO
grant 976951; H.L. is supported in full by DOE grant DE-FG02-95ER40899; C.N.P. is
supported in part by DOE DE-FG03-95ER40917. The work of M.C., G.W.G. and C.N.P.
was supported in part by the programme Supergravity, Superstrings and M-theory of the
Centre mile Borel of the Institut Henri Poincar, Paris (UMS 839-CNRS/UPMC).
References
[1] J. Maldacena, The large N limit of superconformal field theories and supergravity, Adv. Theor.
Math. Phys. 2 (1998) 231, hep-th/9711200.
43
[2] S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Gauge theory correlators from non-critical string
theory, Phys. Lett. B 428 (1998) 105, hep-th/9802109.
[3] E. Witten, Anti-de Sitter space and holography, Adv. Theor. Math. Phys. 2 (1998) 253, hepth/980215.
[4] I.R. Klebanov, A.A. Tseytlin, Gravity duals of supersymmetric SU(N) SU(N + m) gauge
theories, Nucl. Phys. B 578 (2000) 123, hep-th/0002159.
[5] P. Candelas, X.C. de la Ossa, Comments on conifolds, Nucl. Phys. B 342 (1990) 246.
[6] E.G. Gimon, J. Polchinski, Consistency conditions for orientifolds and D manifolds, Phys. Rev.
D 54 (1996) 1667, hep-th/9601038.
[7] M.R. Douglas, Enhanced gauge symmetry in M(atrix) theory, JHEP 007 (1997) 004, hepth/9612126.
[8] S.S. Gubser, I.R. Klebanov, Baryons and domain walls in an N = 1 superconformal gauge
theory, Phys. Rev. D 58 (1998) 125025, hep-th/9808075.
[9] I.R. Klebanov, N. Nekrasov, Gravity duals of fractional branes and logarithmic RG flow, Nucl.
Phys. B 574 (2000) 263, hep-th/9911096.
[10] I.R. Klebanov, M.J. Strassler, Supergravity and a confining gauge theory: duality cascades and
SB-resolution of naked singularities, JHEP 0008 (2000) 052, hep-th/0007191.
[11] M.B. Stenzel, Ricci-flat metrics on the complexification of a compact rank one symmetric
space, Manuscripta Mathematica 80 (1993) 151.
[12] M. Graa, J. Polchinski, Supersymmetric three-form flux perturbations on AdS5 , hepth/0009211.
[13] S.S. Gubser, Supersymmetry and F-theory realization of the deformed conifold with three-form
flux, hep-th/0010010.
[14] L.A. Pando Zayas, A.A. Tseytlin, 3-branes on a resolved conifold, hep-th/0010088.
[15] M. Cvetic, H. L, C.N. Pope, Brane resolution through transgression, hep-th/0011023, to appear
in Nucl. Phys. B.
[16] F. Bigazzi, L. Giradello, A. Zaffaroni, A note on regular type 0 solutions and confining gauge
theories, hep-th/0011041.
[17] M. Bertolini, P. Di Vecchia, M. Frau, A. Lerda, R. Marotta, I. Pesando, Fractional D-branes and
their gauge duals, hep-th/0011077.
[18] M. Cvetic, G.W. Gibbons, H. L, C.N. Pope, Ricci-flat metrics, harmonic forms and brane
resolutions, hep-th/0012011.
[19] O. Aharony, A note on the holographic interpretation of string theory backgrounds with varying
flux, hep-th/0101013.
[20] C.P. Herzog, I.R. Klebanov, Gravity duals of fractional branes in various dimensions, hepth/0101020.
[21] L.A. Pando Zayas, A.A. Tseytlin, 3-branes on spaces with R S 2 S 3 topology, hepth/0101043.
[22] M.J. Duff, J.M. Evans, R.R. Khuri, J.X. Lu, R. Minasian, The octonionic membrane, Phys. Lett.
B 412 (1997) 281, hep-th/9706124.
[23] S.W. Hawking, M.M. Taylor-Robinson, Bulk charges in eleven dimensions, Phys. Rev. D 58
(1998) 025006, hep-th/9711042.
[24] B.R. Greene, K. Schalm, G. Shiu, Warped compactifications in M and F theory, Nucl. Phys.
B 584 (2000) 480, hep-th/0004103.
[25] K. Becker, M. Becker, Compactifying M-theory to four dimensions, JHEP 0011 (2000) 029,
hep-th/0010282.
[26] K. Becker, A note on compactifications on Spin(7)-holonomy manifolds, hep-th/0011114.
[27] R.L. Bryant, S. Salamon, On the construction of some complete metrics with exceptional
holonomy, Duke Math. J. 58 (1989) 829.
[28] G.W. Gibbons, D.N. Page, C.N. Pope, Einstein metrics on S 3 , R3 and R4 bundles, Commun.
Math. Phys. 127 (1990) 529.
44
[29] B.S. Acharya, J.M. Figueroa-OFarrill, C.M. Hull, B. Spence, Branes at conical singularities
and holography, Adv. Theor. Math. Phys. 2 (1999) 1249, hep-th/9808014.
[30] C.V. Johnson, A.W. Peet, J. Polchinski, Gauge theory and the excision of repulson singularities,
Phys. Rev. D 61 (2000) 086001, hep-th/9911161.
[31] L. Berard-Bergery, Quelques exemples de varietes riemanniennes completes non compactes a
courbure de Ricci positive, C.R. Acad. Sci., Paris, Ser. I302 (1986) 159.
[32] D.N. Page, C.N. Pope, Inhomogeneous Einstein metrics on complex line bundles, Class.
Quantum Grav. 4 (1987) 213.

Reflection symmetry breaking scenarios with

minimal gauge form coupling in brane
world cosmology
Brandon Carter a , Jean-Philippe Uzan b
a DARC, CNRSUMR 8629, Observatoire de Paris, F-92195 Meudon, France
b Laboratoire de Physique Thorique, CNRSUMR 8627, Bt. 210, Universit Paris XI,
F-91405 Orsay Cedex, France

Received 16 January 2001; accepted 4 April 2001
Abstract
This article synthesises and extends recent work on the cosmological consequences of dropping
the usual Z2 reflection symmetry postulate in brane world scenarios. It is observed that for
a cosmological model of homogeneous isotropic type, the relevant generalised Birkhoff theorem
establishing staticity of the external vacuum in the maximally symmetric bulk outside a freely
moving world brane will remain valid for the case of motion that is forced by minimal (generalised
WessZumino type) coupling to an external antisymmetric gauge field provided its kinetic action
contribution has the usual homogeneous quadratic form. This means that the geometry on each side
of the brane worldsheet will still be of the generalised Schwarzschild anti-de-Sitter type. The usual
first integrated Friedmann equation for the Hubble expansion rate can thereby be straightforwardly
generalised by inclusion of new terms involving 2 extra parameters, respectively, measuring the
strength of the gauge coupling and the degree of deviation from reflection symmetry. Some
conceivable phenomenological implications are briefly outlined, and corresponding limitations are
derived for possible values of relevant parameters. 2001 Elsevier Science B.V. All rights reserved.
PACS: 12.10.Kt
1. Introduction of minimal coupling

The proverbially fundamental complementarity of brain versus brawn has an analogue
in the purely mathematical context associated with names such as Cartan and de Rham
wherein gauge equivalence classes of p-forms (i.e., covariant antisymmetric tensor fields)
arise as the homological complement of p-surfaces. Conversely, in the higher-dimensional
physical context associated with names such as Polchinski and Witten, (p 1)-branes
E-mail address: brandon.carter@obspm.fr (B. Carter).
0550-3213/01/$ see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 7 4 - 2
46
B. Carter, J.-P. Uzan / Nuclear Physics B 606 (2001) 4558
(i.e., p-surface supported systems) emerge naturally as concentrated source distributions

for gauge p-form fields. Thus given any (p 1)-brane one of the first questions that comes
to mind is what about the corresponding brawn, i.e., the associated gauge p-form?
In the case of a point particle (i.e., a zero brane) the answer is usually that it is the
ordinary Maxwellian electromagnetic field. In the case of the recently much studied
theories that treat our world as a 3-brane in a 5-dimensional bulk, the question of the
corresponding gauge 4-form has been generally overlooked, perhaps because the ensuing
brawn force would have to vanish anyway in the reflection symmetric scenarios that are
most commonly [1,2] considered.
Several authors have discussed models in which the usual Z2 symmetry postulate in
homogeneous isotropic cosmological scenarios treating our four-dimensional world as
a 3-brane in a 5-dimensional bulk was dropped. Ida [3] and then Krauss [4] showed that
such a reflection symmetry breaking can arise due to the choice of different integration
constants for the solution of the vacuum Einstein equations on either side of the brane,
and the general question of the observable consequences of dropping this symmetry
been addressed by Davis et al. [5]. A different kind of symmetry breaking has been
considered by Deruelle and Doleel [6] and Perkins [7], who considered bubble walls
separating two distinct five-dimensional anti-de-Sitter spacetime domains with different
cosmological constants. Finally, the possibility that both kinds of symmetry breaking occur
simultaneously has been considered by Stoica et al. [8] and by Bowcock et al. [9], who
have shown how to obtain a generalisation of the usual first integrated Friedmann equation
giving the square of the Hubble expansion rate as a sum of terms in which, as well as the
ones that are now familiar [10] in conventional (meaning force free reflection invariant)
brane world cosmology there is a new term (including the extra contribution considered
in [38]) that could have been important at some stages in the early universe.
The present article provides a synthesis that develops the underlying physics giving rise
to such scenarios by incorporating allowance for the simplest conceivable coupling to an
external antisymmetric brawn gauge field, A say (where Greek indices run from
0 to 4). This minimal coupling case [11] is governed by an action consisting just of an
external kinetic contribution of the usual purely quadratic type, together with a coupling
contribution given by an action density of the (generalised WessZumino) form
Lco = (4!)1 j A ,
(1)
for a Dirac distributional source given by some dimensionless coupling constant e{4} as
j = j [ ] with
j = e{4} E ,
(2)
where is a coordinate measuring orthogonal distance from the 4-dimensional worldsheet

of the 3-brane. Here E is the antisymmetric tangent element of the worldsheet
normalised as E E = 4! It can be used, in conjunction with the measure tensor
of the 5-dimensional background tensor, to construct the unit normal covector
= (4!)1 E . The scalar coefficient e{4} must necessarily be uniform over
the worldsheet in order for the source current to satisfy the conservation law j = 0
that is necessary for gauge invariance. The corresponding physical (i.e., gauge invariant)
47
5-form is defined by the exterior derivative of A as

F 5[ A ] = F ,
(3)
where F is a pseudo scalar (it is not strictly a scalar because its sign is parity dependent)
in terms of which the purely external Lagrangian action governing the gauge field has the
form
1 2
F ,
Lex =
(4)
2
where is a coupling constant. Varying this Lagrangian (4) with respect to the
5-dimensional spacetime metric g defines the external stress energy density tensor
(Tex = 2Lex /g + Lex g ) as
1 2
(5)
F g .
2
A coupling Lagrangian Lco of the most general kind would give rise to another contribution
Tco , to the stress energy density tensor but this can be shown [11] to vanishing in the
minimally coupled case (1) we are considering here article. Varying the coupling and
external Lagrangians with respect to the brawn field A leads to the field equation
of motion
Tex =
F = j .
(6)
It can thereby be concluded that in the source free region outside the brane the pseudo
scalar field F is uniform, i.e., that F = 0.
An important consequence of this is that outside (but not in) the world brane, the brawn
stress energy tensor (5) acts just as an effective addition to the cosmological constant
that appears in the 5-dimensional Einstein equation according to which the effect of the
total stress energy tensor T (including both external and brane contributions) is given by
G = T g ,
(7)
1
2 Rg ,
where R is the Ricci tensor and R is its trace. The

in which G R
traditional unrationalised coupling constant is related to the quantity (5) used by
several authors (e.g., [2]), and to the 5-dimensional analogue G of Newtons constant and
its associated gravitational mass scale mG as defined according to the appropriate [12]
rationalisation procedure, by
2
= (5)
= 6 2 G,
G = m3
G .
(8)
Thus, in the external region on either side of the brane, substitution of (5) for T reduces
the Einstein equation (7) to the pseudo vacuum form
2
G = g , = +
(9)
F ,
2
where F + and F are the constant values taken by the pseudoscalar brawn field in the
respective external coordinate ranges > 0 and < 0. This minimal coupling case may
be seen as the five-dimensional analogous of the standard (p = 1)-dimensional case of
a charge particle in electromagnetism in which e{1} is the particle charge, and F is
just the analog of the electromagnetic tensor F while (6) is the Maxwell equation.
48
2. The generalised (Birkhoff type) bulk staticity theorem

So long as we are concerned just with cosmological models of homogeneous isotropic
type, the consideration that the external gravitational field is of pseudo vacuum type allows
to invoke the generalised Birkhoff theorem (see the treatise of Misner et al. [13], whose
sign conventions we are following). Its original version showed that spherical symmetry
necessarily entails the staticity property that was merely assumed as a simplifying ansatz
when Schwarzschild first derived his famous solution. A recent demonstration with the
range of applicability needed in the present context has recently been provided by
Bowcock, Charmousis and Gregory [9], who deal with the case of a 5-dimensional vacuum
metric dsV2 = g dx dx that has the symmetry of a 3-sphere, a 3-plane, or an anti
3-sphere with metric given, respectively, for positive, vanishing, or negative curvature
parameter k, by
2
=
dsIII

d 2
+ 2 d 2 + sin2 d 2 .
2
1 k
(10)
The conclusion is that wherever equations of the vacuum form (9) are satisfied, and thus
on either side of the (minimally coupled) 3-brane under consideration, the metric must be
static and hence of the generalised Schwarzschild (anti)-de-Sitter form
2
+
dsV2 = r 2 dsIII
dr 2
V dt 2 ,
V
(11)
with
V =k
2 2GM
r
,
6
r2
(12)
where M+ and M are constants of integration, having the dimensions of mass, that
characterise the solution in the respective ranges > 0 and < 0. (This generalisation
is analogous to that for the charged black hole case, for which the Birkhoff theorem still
applies, giving
the static ReissnerNordstrm solution.) In the compact case k > 0 the
quantity r/ k is interpretable as a Schwarzschild type radial coordinate and the mass
constants is given by M = k 2 M where M + and M are interpretable as the masses
of distant source distributions, such as other 3-branes (or in their absence, of black holes)
on either side. For this positive curvature case it would commonly be convenient to fix
the scale so as to obtain k = 1, but such a canonical normalisation is not possible in the
spatially flat case (k = 0) that is accurate within observational uncertainties for the large
scale representation of our actual universe.
The fact that the static nature of the external metric is unaffected by the dynamical
evolution of the brane means that there is no way it can signal its presence to, and thereby
perhaps avoids collision with, any other brane that may be present outside (at a distance
whose value as a function of time is what is commonly referred to as the radion). The
avoidance of this so-called radion instability problem is one of the motivations for the
development of generalisations [14] involving the presence in the bulk of a dilatonic
scalar field that can carry information from one brane to a neighbouring brane. What we
49
have shown here is that minimal coupling to what is effectively a pseudo scalar field does
not offer any such escape from this radion instability problem. If only a single brane
is present the problem of the radion as such will not arise, but if either M+ or M is
non-zero (which, as we shall see, is necessarily the case in the kind of reflection symmetry
breaking scenarios envisaged in [36]) one might still need to worry about the possibility
of colliding with its source or (if it is source free) of falling into the corresponding black
hole.
When one is mainly interested in what is going on in the brane itself rather than in the
bulk outside, it is commonly convenient to use a brane based reference system depending
on the use of the coordinate that measures the orthogonal distance from the worldsheet.
In the homogeneous isotropic case this gives an expression of the well known form
2
+ d 2 2 d 2 ,
dsV2 = r 2 dsIII
(13)
in which the quantities r and are functions just of and . The latter is normalised
in such a way that in the limit 0 (i.e., on the worldsheet) we have 1. It means
that is interpretable as a measure of the proper cosmological time on the world brane.
The corresponding limit r a defines a function a( ) interpretable as a cosmologically
comoving scale factor, whose proper time derivative a specifies the relevant Hubble
expansion rate a/a.
Although the two versions (11) and (13) have long been separately well known,
the relationship between them was not satisfactorily clarified until the recent work by
Mukohyama, Shiromizu, and Maeda [15]. These authors showed explicitly how a metric
of the static form (11) can be transformed to the brane based version obtained (in terms
of hyperbolic functions of ) by Bintruy, Deffayet, Ellwanger and Langlois [10]. The
self contained treatment provided here avoids getting into these explicit details [10,15] by
following a more economical approach that does not depend on the particular (explicitly
integrable) form (12) of the function V and that could thus be useful in a broader context
(such as might arise if, as well as the brawn field A considered here, a lower order
gauge field, A or A was also present in the bulk).
The essential point is that (in each radial {r, t} 2-surface) the distance function
is orthogonal to lines of constant that are automatically geodesic. The stationarity
of the metric (11), therefore, implies that the unit tangent vector x (employing the
usual convention that dash and dot indicate partial differentiation with respect to
and , respectively) will contract with the time symmetry Killing vector, as specified by
k dx = V dt, to give a constant of the geodesic motion E = k x (that would be
interpretable as energy if instead of being spacelike the geodesic was timelike). Though
independent of , this quantity E depends in general on the new time variable whose
normalisation it is now convenient to fix with respect to the rate of change of the scale
length a on the worldsheet by imposing the identification E = a.
The definition of E then
provides the relation t = a/V,
and hence the unit normalisation condition g x x = 1

takes the form
a 2 = r 2 V.
(14)
50
In order for the required transformations dt = t d + t d and dr = r d + r d to preserve

coordinate orthogonality, x x = 0, it must evidently satisfy the condition r r = V 2 tt , so
substitution in (11) now leads directly to q.e.d., namely, the brane based form (13), in
which the coefficient can be seen to be given simply by
2 = r 2 /a 2 .
(15)
This short cut derivation of (13) has made no use of any specific prescription for the
dependence of V on r, but it is now to be observed that when V is given in terms of
the relevant generalised Schwarzschild AdS formula (12), the relations (14) and (15) take
precisely the form to which the (bulk) Einstein equations were shown to be reducible by
Bintruy et al. [10].
The condition of metric continuity on the brane worldsheet (where r a) is trivially
satisfied by (15), but (14) provides interesting matching conditions. We use square and
angle brackets, respectively, for the difference and average of the values limits on opposite
sides, so that in particular for the effective cosmological constant we have
[] + =
F [F ],

1 +
2
+ = +
F .
2
2
(16)
One finds that subtraction of (14) gives the relation

[] G[M]
,
r r = a 2
12
a2
(17)
which is of no interest in the Z2 reflection symmetric case where all the terms vanish, while
averaging (14) gives the more generally useful expansion rate formula
2
a
r 2 k 2GM
=
+
.
+
a
6
a2
a4
(18)
3. Solution of the junction conditions

To determine the evolution of the system we are interested in, we need to use the well
known [13] Darmois Israel junction conditions according to which the active gravitational
effect of the surface stress energy density T of the brane is governed by the equation
[K ] [K] = T ,
(19)
where and K are, respectively, the first and the second fundamental forms (K being
its the trace), defined as
g ,
K ,
(20)
= being the unit (i.e., = 1) vector normal to the brane (i.e., = 0).
It is also necessary to solve the equations governing the passive evolution for the
worldsheet, which has often been worked out ad hoc in particular applications, but which
51
can also be expressed [11] in a simple generally valid form as

T K = f,
(21)
where f is the force density acting on the brane. This will be given in terms of the
contraction of the external (bulk) stress energy density as

f = Tex .
(22)
This must of course vanish in the reflection symmetric configurations that are most
commonly considered, but need not do so in the more general context considered here.
It can be seen from (6) and from the expression (5) for F that F = e{4} [ ],
which implies that
[F ] = e{4} .
(23)
The force density is found to have a constant but generically non vanishing value given by
f = e{4} F = 1 [].
(24)
In order to apply the preceding formulae one has to work out the exact expressions of
the first and second fundamental forms as defined in Eq. (20). The first fundamental form
is obviously given by
2
2
= dx dx = a 2 dsIII
d 2 ,
dsIV
(25)
and the second fundamental form by

2
+ d 2 .
K dx dx = ar dsIII
(26)
We also need to know the form of the worldsheet stress energy density tensor T . To
be compatible with the hypothesis of a homogeneous isotropic geometry, T must itself
be of the homogeneous isotropic form, i.e.,

T = Uu u T + u u ,
(27)
with respect to the preferred unit vector defined by u = . Here U is the total energy
density and T is the total brane tension. It is normally assumed that the directly observable
energy density and pressure P (i.e., of the cosmic fluid) represent small deviations from
an isotropic (inflationary) limit state given by T = T , where T is a fixed tension
(approached by both T and U in the limit a ) in terms of which the actual tension
and energy density are given by
T T P ,
U T + .
(28)
Substitution of (27) in (19) gives explicit jump conditions of the well known [2] form

a
= (2U 3T ).
r = U,
(29)
3
3
Now, from (21) we obtain the relation
r
(30)
= f,
a
which differs from the usual version by the presence of the force density term on the right.
U + 3T
52
Using the expression (29) of [r ], we obtain, from (17), that 2r (which is the same
as the quantity d introduced as a measure of the deviation from reflection symmetry by
Davis et al. [5]) is given by

3 a[] 2G[M]

2r =
(31)
+
.
U
6
a3
This generalises the corresponding formula of Davis et al. by the inclusion of the []
term representing the effect of the brawn force, and it also provides a more specific
interpretation of their F term by showing that (as observed by Ida [3]) it is proportional
to the (reflection symmetry breaking) difference [M] between the mass terms on opposite
sides, and hence that it would have to be absent if both mass terms were zero (as was
assumed in the work of Deruelle and Doleel [6] and Perkins [7]). It is to be observed that
the gravitational coupling coefficients can be cancelled out of (31) and hence also out of
the expressions for the components of the mean value of the first fundamental form (26),
which vanish in the reflection symmetric case, and which will be given in the general case
by

[M]
a
3[M]
3T

e
+
F
,
U

=
+
1
e{4} F .
Ur =
(32)
{4}
2 2 a 3 4
2 2 a 4
4U
Using (31) and the first equation of (29), respectively, for r and [r ] and using the
decomposition r 2 = r 2 + [r ]2 /4 to express r 2 in (18), we can now proceed directly
to the relevant generalisation of the Friedmann equation, which takes the form
2

a
k
[] 6G[M] 2
U 2
2GM
1
(33)
2+
+
=
+
+
.
a
6
6
a
a4
(2U)2 2
a4
This agrees with what was obtained in a rather more circuitous manner by Bowcock
et al. [9], who were the first to undertake a systematic investigation of reflecting symmetry
breaking scenarios with the degree of generality considered here, but who used a system
in which the 5-dimensional gravitational coupling constant was set to unity, thereby
obscuring the way it affects the various terms. As well as directness, the present approach
has the advantage of showing more explicitly how the various coupling coefficients
contribute, and in particular how actually cancels out of the formulae (32) and, as will
be demonstrated immediately below, from last term of (33) which is the only one whose
presence depends on reflection symmetry breaking.
4. Investigation of the parameter space

In order to interpret the Friedmann-like equation (33), we use (28) to rewrite it in terms
of the ratio /T , as
2
k
8G4
4 2GM 2 2
a
2 +
+
=
+ G
a
3
a
3
a4

2
e{4} F 2
(1 + )2 [M] [M]
3
+
2
+
+
e
F

+
, (34)
{4}
(2T )2 2 a 4 2 a 4
2
53
in which the first three terms of the r.h.s. are those of the standard Friedmann equation,
with the effective 4-dimensional Newton constant G4 and the effective 4-dimensional
cosmological constant 4 given by

2
e{4} F 2
6 2
GT
8G4 =
(35)
,
T
4T
and
24 = + 6

GT
2
e{4} F
+
4T
2
.
(36)
In terms of the brane mass scale given by m4 = T , and of the ordinary Planck
mass scale given by m2
P G4 , it is convenient to parametrise the relative importance
in (35)(36) of the gauge field contribution F and of the bulk gravitational coupling
constant m3
G G, in terms of a dimensionless hyperbolic angle such that
1/2 6
1/2
8m sh
2 ch
,
G=
.
e{4} F =
(37)
2
3
mP
3
m2 mP
Besides the three independent parameters involved in (37), namely, , m and mP , of
which only the latter is experimentally known in advance, this model involves four more
independent parameters which are the bulk mass scales M , the ordinary cosmological
constant 4 and the cosmological curvature scale k. Only the two latter are roughly known
in advance, neither of them differing from zero by an amount that can be reliably measured
yet. Since 4 is at most of the order of the present day cosmological closure value (i.e.,
m4 1060 mP ), it is effectively negligible for our present purpose, i.e., we can set
4 = 0. This means that the effective bulk cosmological constants and the corresponding
2
curvature length and mass scales , = m1
as defined by 6, on either side of the
brane will be given (using (37) for , and using (24) and (37) for [] ) by
= 6m2
= 8
m4 2
e
.
m2P
(38)
In order to place some limits on the four independent unknown constants T , and the
bulk mass constants it is convenient to replace M by variable mass parameters

th
.
2GM 1
(39)
(1 + )2
With this notation, we can convert (34) to the form
2
a
8
k
4
+ ,
=
2+
2
a
3
3mP a
(40)
in which the final term represents the deviation from the traditional Friedmann equation
and is expressible as

2

2

(3 + 2) th2
[M]
= 2 G
(41)
+ 4,
1+
+
2
2
4
(1 + )
2 a T (1 + )
a
54
in which the first two terms are manifestly positive definite. Furthermore, if (unlike some
authors who, in scenarios involving more than one brane, have envisaged negative mass
densities) we make the traditional assumption that all external mass distributions must
be positive (hence M 0), then it can be deduced from the expression (39) that both
variables + and must also be everywhere positive, though in the case of the latter
only marginally if is large.
Before proceeding to exploit these positivity properties for the purpose of placing limits
on the parameters involved, we need to start by choosing a convenient normalisation for
the scale factor a. Since the spacial curvature factor k is too small to have been reliably
measured yet (k = 0 being a plausible possibility and certainly a good approximation),
the mathematically convenient choice k = 1 is in practice unavailable. Whatever k may
be, a physically convenient choice is to normalise the present value of the scale factor
a to agree with the length scale characterising the cosmological background radiation
temperature , i.e., to simply to take a = 1 . This normalisation has the advantage that
it holds when extrapolated backwards in time so long as photon creation processes remain
insignificant, which means ever since the temperature dropped below the electronpositron
pair creation threshold given by 2me .
In particular the temperature N of nucleosynthesis, which is of greatest interest for the
purpose of obtaining rigourous observational constraints, lies in the range 0 < < 2me
in which the product a remains constant. Moreover, throughout the radiation dominated
era during which
=
2 g 4
,
30
(42)
the order of magnitude relation a 1 will still be valid as a useful approximation so

long as the dimensionless coefficient g representing the effective number of relativistic
degrees of freedom remains comparable with its usual value g 10.75 at nucleosynthesis.
Under such conditions, with k and 4 now set to zero (since they are negligible in that era),
one can rewrite (40) as
2
8
(43)
=
+ .
3m2P
The deviation term (41) can be expressed in terms of correspondingly normalised bulk
2
mass scales m
B M /2 (representing the distant external mass associated with a small
comoving volume of size a 3 ) as

4 ch 2
(3 + 2)th2
=
1+
3T
mP
(1 + )2

[mB ] 4 2 + 4 4
+
.
+
+
(44)
T (1 + )
2
2
Since each of these terms is positive, the condition that should be small compared with
the leading term in (43) when N requires that this smallness requirement should
apply separately to each of the four terms in (44). The constraint on the first term gives the
basic requirement that

15
2
N ch
m2 ,
2 2 g
55
(45)
with the convenient corollary that the value of at the time of nucleosynthesis satisfies
N 1 throughout the range N (and considerably beyond if is large). From the
second term, one obtains the requirement

4
3

[mB ] 4 g m .
(46)
45 mP N2
One can understand these two constraints by expanding in series with respect to 1;
they simply state that the terms scaling faster than 4 (e.g., 8 . . . ) are negligible at the
time of nucleosynthesis, thus explaining the dependence on N in (45) and (46). The third
and fourth terms in (44) scale as radiation, thus behaving like extra relativistic degrees of
freedom which are constrained by the fact that g cannot deviate for more than 20% from
its expected value g = 10.75. If we choose an orientation convention so that e{4} F is
positive (thus implying that 0) then the constraint on the third term of (44) gives the
requirement

2 2 g m2
ch

.
m+
(47)
B
15
3 mP
Due to the possibility of partial cancellation, the condition obtained from the fourth (and
last) term in (44) has a less restrictive form: using (45) (and more particularly using the
fact that N ch2 1) it is expressible as

m
2 2 g m2
B
(48)
.

ch
15
3 mP
As expected, these two conditions are independent of N since the contributions of the
terms in are proportional to that of the radiation.
Using that |[mB ]| < m+

B + mB and that ch + 1/ ch < 2 ch , we deduce that |[mB ]| <
2
2
4 /3 m /(15mP ), and thus that the condition (45) on leads to the constraint (46).
This means that (46) is not needed as an independent condition, but is an automatic
consequence of the other three conditions (45), (47) and (48).
Furthermore, experimental limits on the deviation of the law of gravity with respect to
its Newtonian form imply an upper limit [17] on the length scale , , and a corresponding
lower limit on the mass scale m in the bulk. Since no such deviation has been detected
above a millimeter [16], the present limit is m 103 eV. Conveniently, this corresponds
roughly to the present temperature 0 of the cosmological microwave background
radiation and can thus be written as
m 0 .
(49)
Previous discussions [17] of this limit have been restricted to scenarios characterised by
Z2 reflection symmetry, but, in cases where this symmetry is broken, the limit on the
56
cosmological length scale must presumably be satisfied on each side separately. Since our
orientation convention is such that m+

is greater than m we deduce that in the generic
case the mass values given by (38) will be characterised by
m+
m 0 .
(50)
5. Implications and conclusions

The requirement (50) is evidently most severe for what, according to our convention,
is the negative side of the brane, so by (38) it gives an order of magnitude restriction
expressible as

3
2
0 mP ch .
m
(51)
This is evidently strong enough to ensure that (45) will be satisfied as an automatic
consequence. Nucleosynthesis and laboratory experiments on the gravitational force thus
provide just three independent constraints. One of them is given by (51), and the two others
are (47) and (48), which can be combined as

2 2 m2

(ch )1 .
m
(52)
B
15 3 mP
If m has the minimum value allowed by (51), i.e., m2 (3/)1/20 mP ch , then (52)
simply gives
2 2
2 2
0 ,
0 (ch )2 .
m

(53)
B
15
15
To estimate the orders of magnitude imposed by the three constraints (51) and (52),
we use the approximate values 0 2 1013 GeV, mP 1019 GeV and N 1 MeV,
respectively, for the cosmic background radiation temperature, the Planck mass and the
nucleosynthesis temperature. It is convenient to distinguish two regimes as follows.
m+
B
1. Type I scenarios, characterised by 1: in these scenarios the contribution e{4} F

is relatively unimportant both in (35) and (36). Using ch 1, we deduce from (51)
that the mass scale specifying the limiting value of the brane tension, T = m4 ,
must satisfy
m 106N 1 TeV,
(54)
which is the same as in the usual Z2 reflection symmetric scenarios. In this case, from
(52), the two external mass scales satisfy the same constraint

2 2 m2
,
mB
(55)
15 3 mP
10 MeV.
which in the limit m 1 TeV gives m
B 10
57
2. Type II scenarios, characterised by 1: in these scenarios the contribution e{4} F

from the gauge field is very large, so there has to be a fine tuning between its value
and that of the correspondingly large five dimensional gravitational constant G to as
to get the relatively small observed value of G4 by (35). In this case (54) is to be
replaced by a more restrictive condition of the form
m (ch )1/2 TeV.
(56)
In such scenarios the external mass scales satisfy different constraints. In particular
if the brane mass scale has the minimum value m (ch )1/2 TeV allowed by (56)
10 MeV which is the same as before, but on the other side (52)
we get m+
B 10
10 (ch )2 MeV.
gives the potentially much less restrictive condition m
B 10
To sum up, in this article we have introduced the simplest conceivable coupling to
an external gauge field, which has been shown to be compatible with staticity. Like the
effect [35] of external masses, this brawn coupling breaks the Z2 reflection symmetry
and gives a mechanism for the set up considered for instance by Deruelle and Doleel [6],
Stoica et al. [8], Perkins [7], Davis et al. [5] and Bowcock et al. [9]. We have given a simple
self contained derivation of the generalised Friedmann equation which now depends on
five adjustable parameters, and we have used its Friedmannian limit to identify the fourdimensional Newton constant and cosmological constant. The 4 independent parameters
involved, namely, the hyperbolic angle , the brane tension mass scale m , and the
external masses per thermal volume m+

B and mB (2 more than are needed for the reflection
+
symmetric case in which = 0 and mB = m
B ) have been shown to be subjet to 3
independent constraints (one more than in the symmetric case) of which two arise from
nucleosynthesis measurements and one from gravitational laboratory experiments above
a millimeter. We distinguish two kinds of regime depending on the value of the hyperbolic
angle that characterises the relative importance of the brawn field. In the Type II case
for which this parameter is large the external mass scale can (on one but not both sides) be
much larger than in the usual reflection symmetric case, but for this to be possible the brane
mass scale m must also be much larger that its usual lower limit of the order of a TeV.
References
[1] L. Randall, R. Sundrum, An alternative to compactification, Phys. Rev. Lett. 83 (1999) 4690,
hep-th/9906064.
[2] P. Bintruy, C. Defffayet, D. Langlois, Non conventional cosmology from a brane universe,
Nucl. Phys. B 565 (2000) 269, hep-th/9905012.
[3] D. Ida, Brane world cosmology, JHEP 0009 (2000) 014, gr-qc/9912002.
[4] P. Krauss, Dynamics of anti-de-Sitter domain walls, JHEP 9912 (1999) 011, hep-th/9910149.
[5] A.-C. Davis, I.R. Vernon, S. Davis, W.B. Perkins, Brane world phenomenology without the Z2
symmetry, hep-ph/0008132.
[6] N. Deruelle, T. Doleel, Brane versus shell cosmologies in Einstein and EinsteinGaussBonnet
theories, Phys. Rev. D 62 (2000) 103502, gr-qc/0004021.
[7] W.B. Perkins, Colliding bubble walls, gr-qc/0010053.
58
[8] H. Stoica, H. Tye, I. Wasserman, Cosmology in the RandallSundrum brane world scenario,
hep-th/0004126.
[9] P. Bowcock, C. Charmousis, R. Gregory, General brane cosmologies and their spacetime
structure, hep-th/0007177.
[10] P. Bintruy, C. Deffayet, U. Ellwanger, D. Langlois, Brane cosmological evolution in a bulk
with cosmological constant, Phys. Lett. B 477 (2000) 285, hep-th/9910219.
[11] B. Carter, R. Battye, General junction conditions and minimal coupling in brane world
scenarios, hep-th/0101061, Phys. Lett. B (to appear).
[12] N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phenomenology, astrophysics and cosmology of
theories with submillimeter dimensions and TeV scale quantum gravity, Phys. Rev. D 59 (1999)
086004, hep-ph/9807344.
[13] C.W. Misner, K.S. Thorne, J.A. Wheeler, Gravitation, Freeman, San Francisco, 1973.
[14] A. Mennim, R. Battye, Cosmological expansion on a dilatonic brane world, hep-th/0008192.
[15] S. Mukohyama, T. Shiromizu, K. Maeda, Global structure of exact cosmological solutions in
the brane world, Phys. Rev. D 62 (2000) 024028, hep-th/9912287.
[16] J.C. Long, A.B. Churnside, J.C. Price, Gravitational experiment below 1 millimeter and
comment on shielded casimir backgrounds for experiments in the micron regime, Proceedings
of the IXth Marcel Grossmann Meeting, Roma, July 28, 2000, hep-ph/0009062.
[17] D.J.H. Chung, H. Davoudias, L. Everett, Experimental probes of the RandallSundrum infinite
extra dimension, hep-ph/0010103.

Coannihilation effects in supergravity and

D-brane models
R. Arnowitt, B. Dutta, Y. Santoso
Center For Theoretical Physics, Department of Physics, Texas A&M University,
College Station, TX 77843-4242, USA
Received 13 March 2001; accepted 9 May 2001
Abstract
Coannihilation effects in neutralino relic density calculations are examined for a full range of
supersymmetry parameters including large tan and large A0 for stau, chargino, stop and sbottom
coannihilation with the neutralino. Supergravity models possessing grand unification with universal
soft breaking (mSUGRA), models with nonuniversal soft breaking in the Higgs and third generation
sparticles, and D-brane models with nonuniversal gaugino masses were analysed. Unlike low tan
where m0 is generally small, stau coannihilation corridors with high tan are highly sensitive to A0 ,
and large A0 allows m0 to become as large as 1 TeV. Nonuniversal soft breaking models at high tan
also allow the opening of a new annihilation channel through the s-channel Z pole with acceptable
relic density, allowing a new wide band in the m0 m1/2 plane with m1/2 400 GeV and m0 rising
to 1 TeV. The D-brane models considered possess stau coannihilations regions similar to mSUGRA,
as well as small regions of chargino coannihilation. Neutralinoproton cross sections are analysed for
all models and it is found that future detectors for halo wimps will be able to scan essentially the full
parameter space with m1/2 < 1 TeV except for a region with < 0 where accidental cancellations
occur when 5 tan 30. Analytic explanations of much of the above phenomena are given. The
above analyses include current LEP bounds on the Higgs mass, large tan NLO correction to the
b s decay, and large tan SUSY corrections to the b and masses. 2001 Elsevier Science
B.V. All rights reserved.
1. Introduction
Supersymmetry (SUSY) models with R-parity invariance offers a leading candidate for
the dark matter observed in the universe at large and locally in the Milky Way galaxy. Thus
in models of this type, the lightest neutralino, 10 , is generally the lightest supersymmetric
particle (LSP) and hence is absolutely stable. The observed dark matter is then the relic
neutralinos left over after the Big Bang. SUSY models are predictive since they can
E-mail address: b-dutta@rainbow.physics.tamu.edu (B. Dutta).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 2 3 0 - 9
60
R. Arnowitt et al. / Nuclear Physics B 606 (2001) 5983
calculate simultaneously the amount of relic density expected, Milky Way detection signals
and the production cross sections at accelerators. Thus cosmological, astronomical and
accelerator constraints simultaneously constrain the SUSY parameter space.
The procedure for calculating the relic density of neutralinos is well known. (For a
review, see [1].) Over the past decade, a number of refinements of the analysis needed
to get accurate answers have been included. Thus the treatment of threshold and s-channel
resonances in the annihilation cross section in the early universe was discussed in
[24]. In calculating the relic density, in most of the parameter space it is sufficient
to use the two body neutralino annihilation cross section, since the effects of heavier
particles are suppressed by the Boltzman factor. However, in special situations, the next
to lightest supersymmetric particle (NLSP) may become nearly degenerate with the LSP,
and the coupled annihilation channels of LSPLSP, NLSPLSP, NLSPNLSP must be
simultaneously considered [2]. This phenomena of coannihilation was considered within
the framework of the minimal supersymmetric model (MSSM) in [5] for the situation
when the lightest chargino, 1 , becomes nearly degenerate with the 10 . In this paper
we consider supergravity (SUGRA) models with grand unification of the gauge coupling
constants and soft SUSY breaking at the GUT scale MG
= 2 1016 GeV. The low energy
properties of the theory is thus obtained by running the renormalization group equations
(RGE) from MG to the electroweak scale, where SUSY breaking at MG triggers SU(2)L
U (1)Y breaking at the electroweak scale. Aside from being in accord with the LEP data
implying grand unification, radiative breaking of SU(2)L U (1)Y greatly enhances the
predictiveness of the theory as it reduces the number of SUSY parameters, and one finds
generally for such models that the 10 is mostly Bino, and the 1 is mostly Wino (when
the soft breaking masses are less than 1 TeV). Then the 10 and 1 do not become nearly
degenerate and this type of coannihilation does not generally take place. (An exception for
a class of D-brane models is discussed below.)
More recently it was pointed out that in SUGRA models, the sleptons (particularly
the lightest stau, 1 ) can become nearly degenerate with the 10 leading to a new type
of coannihilation, and this was explored for low and intermediate values of tan =
H2 /H1 [6,7]. (Here H2,1 give rise to (u, d) quark masses). In this paper we examine
this effect for the full range of tan , i.e., 2 < tan < 50 [8]. Further, in addition to 1 10
coannihilation, we find the possibility of light stop t1 10 coannihilation as well as light
sbottom b1 10 coannihilation.
We consider these effects for a range of models based on grand unfication of the
gauge coupling constants: (1) Minimal supergavity model (mSUGRA) [9] where there is
universal soft breaking masses at MG , (2) Nonuniversal supergravity models [10], where
the first two generation soft breaking masses are kept universal at MG (to suppress flavor
changing neutral currents) as well as the gaugino masses, but the Higgs and third generation
soft breaking masses are allowed to be nonuniversal, and (3) a D-brane model [11] based
on Type IIB strings [12] where soft breaking masses at MG of the SU(2)L doublet squark,
slepton and Higgs are kept degenerate but distinct from the SU(2)L singlets, and similarly
the gaugino SU(2)L doublet soft breaking mass is distinct from the SU(2)L singlets. While
the first two models have been characterized as SUGRA models, they can also be realized
61
in string models, and in fact any string model based on grand unification at a high MG
scale with the Standard Model gauge group holding below MG , phenomenologically can
be treated as a SUGRA model with an appropriate amount of nonuniversal soft breaking.
Thus the three models sample the possibilities of universal soft breaking, nonuniversal soft
breaking in the squark, slepton and Higgs sector but universal gaugino masses, and finally
nonuniversality in the gaugino sector as well.
Since the Milky Way is perhaps 90% dark matter, it is a laboratory for studying the
properties of dark matter (DM). Possible signals for DM include annihilation in the halo
of the Galaxy, annihilation in the center of the Sun or Earth, and direct detection from
scattering by terrestrial nuclear targets. Of these, the last is most promising, and we restrict
our discussion here to this. (For recent discussions of the other possibilities see [1315]).
In general, the neutralinonucleus scattering has a spin-independent and spin-dependent
part. However, for heavy nuclei (aside from exceptional situations discussed below), the
spin-independent scattering dominates, giving rise to approximately equal scattering by
neutrons (n) and protons (p) in the nucleus. It is thus possible to extract from any data
the neutralinoproton cross section, 0 p (subject to astronomical uncertainties about the
1
Milky Way). Current detectors (DAMA, CDMS, UKDMC) are sensitive to cross sections
0 p 1 106 pb
1
(1)
with a one to two order of magnitude improvement possible in the near future. Future
detectors (e.g., GENIUS, Cryoarray) plan to be sensitive down to
0 p (109 1010) pb
1
(2)
and thus it is of interest to see what parts of the SUSY parameter space can be examined
by such detectors.
In order to obtain accurate calculations of both the relic density and 0 p for large
1
tan , it is necessary to include a number of corrections in the analysis, and we list these
here. (i) In relating the theory at MG to phenomena at the electroweak scale, the two loop
gauge and one loop Yukawa renormalization group equations (RGE) are used, iterating to
get a consistent SUSY spectrum. (ii) QCD RGE corrections are further included below the
SUSY breaking scale for contributions involving light quarks. (iii) A careful analysis of
the light Higgs mass mh is necessary (including two loop and pole mass corrections) as
the current LEP limits impact sensitively on the relic density analysis. (iv) LR mixing
terms are included in the sfermion (mass)2 matrices since they produce important effects
for large tan in the third generation. (v) One loop corrections are included to mb and m
which are again important for large tan . (vi) The experimental bounds on the b s
decay put significant constraints on the SUSY parameter space and theoretical calculations
here include the leading order (LO) and NLO corrections for large tan . Note that we
have not in the following imposed b (or tb ) Yukawa unification or proton decay
constraints as these depend sensitively on unknown post-GUT physics. For example, such
constraints do not naturally occur in the string models where SU(5) (or SO(10)) gauge
symmetry is broken by Wilson lines at MG (even though grand unification of the gauge
coupling constants at MG for such string models is still required).
62
In carrying out the above calculations we have included the latest LEP bound on the
light Higgs mass, mh > 114 GeV [16] and the recent NLO corrections for large tan for
the b s decay [17,18]. (We have checked numerically that [17] and [18] give identical
results.) Since there are still some remaining errors in the theoretical calculation of mh as
well as uncertainty in the top quark mass, mt = (175 5) GeV, we will conservatively
assume here that mh > 110 GeV. In the MSSM, the constraint on mh is tan -dependent
as Ah production (A is the CP odd Higgs) when mA
= MZ is possible for tan 9
and this can be confused with Zh production. However, in SUGRA models, radiative
electroweak breaking eliminates this part of the parameter space, and so it is correct to
impose the LEP bound on mh for the full range of tan . LEP also gives the bound [19]
m > 102 GeV, and the Tevatron bound for the gluino (g)
is [20] mg > 270 GeV (for
1
squark (q)
and gluino masses approximately equal). We assume an allowed 2 range from
the CLEO data for the b s branching ratio [21]:
1.8 104 B(B Xs ) 4.5 104 .
(3)
In calculating the relic density we will assume the bounds

0.02 < 0 h2 < 0.25,
1
(4)
where 0 = 0 /c , 0 is the relic density of the 10 , c = 3H02 /(8GN ) (H0 is the

1
present Hubble constant and GN is the Newton constant) and H0 = h 100 km s1 Mpc1 .
The lower bound on 0 h2 is somewhat lower than more conventional estimates ( 0 h2 >
1
1
0.1), and allows us to consider the possibility that not all the DM are neutralinos. (In
the following, we will mention when results are sensitive to this lower bound.) Accurate
determinations by the MAP and Planck satellites of the dark matter relic density will
clearly strengthen the theoretical predictions, and already, analyses using combined data
from the CMB, large scale structure, and supernovae data suggests that the correct value of
the relic density lies in a relatively narrow band in the center of the region of Eq. (4) [22].
We will here, however, use the conservative range given in Eq. (4).
Supersymmetry theory allows one to calculate the 10 -quark cross section and we follow
the analysis of [23] to convert this to 10 p scattering. For this one needs the N term,
1
+ dd|p,
N = (mu + md )p|uu
2
(5)
0 = N (mu + md )p|s s|p and the quark mass ratio r = 2ms /(mu + md ). We use
here 0 = 30 MeV [24], and r = 24.4 1.5 [25]. Recent analyses, based on new N
scattering data gives N
= 65 MeV [26]. Older N data gave N
= 45 MeV [27]. We
will use in most of the analysis below the larger number. If the smaller number is used, it
would have the overall effect in most of the parameter space of reducing 0 p by about
1
a factor of 3. However, in the special situation for < 0 where there is a cancellation of
matrix elements, the choice of N produces a more subtle effect, and we will exhibit there
results from both values. Some of the results described below have been mentioned earlier
in conference talks [28].
63
There have been in the recent past a number of calculations of 0 p in the literature
1
[24,2934]. However, none include simultaneously all the corrections discussed above, the
large tan range and the recent LEP mh bounds. We find general numerical agreement
with other calculations for regions of parameter space where the omitted corrections are
small.
2. MSUGRA model
We consider in this section the mSUGRA model. This model is the most predictive of
the SUGRA models as it depends on only four additional parameters and one sign. These
may be chosen as follows: m0 , the universal scalar soft breaking mass at MG ; m1/2 , the
universal gaugino mass at MG (or alternately one may use m 0 or mg as these quantities
1
approximately scale with m1/2 , i.e., m 0 0.4 m1/2 and mg 2.8 m1/2); A0 , the universal
1
cubic soft breaking coupling at MG ; tan = H2 /H1 ; and /||, the sign of the Higgs
mixing parameter in the superpotential term W = H1 H2 . We restrict the parameter
space as follows:
m0 , m1/2 < 1 TeV,

A0 /m1/2 < 4,
(7)
2 < tan < 50.
(8)
(6)
The relic density is calculated by solving the Boltzman equation in the early universe. If the
excited SUSY states lie significantly above the 10 , then this takes the standard form [1]:

dn
= 3H n ann v n2 n2eq
(9)
dt
where n is the neutralino number density, neq is its equilibrium value, means thermal
average, ann is the annihilation cross section 10 + 10 f (f = final state), and v is
the relative velocity. In this case, the effects of the heavier particles are suppressed by the
Boltzman factor. If however, one or more particles, Xi , have masses mi near the neutralino
( 10 X1 , m 0 m1 ), coannihilation effects can occur. In this case Eq. (9) holds for the
1
neutralino number density with, however, neq and ann given by [2]

neq =
(10)
nieq ,
i
ann =

i,j
ri rj ij ;
ri
nieq
.
neq
(11)
In the nonrelativistic limit where freezeout occurs, the equilibrium number densities take
the form [3,35]

15T
mi T 3/2
exp(mi /T ) 1 +
.
nieq
(12)
= gi
2
8mi
64
Here ij is the cross section for Xi + Xj f (11 is the neutralino annihilation cross
section) and gi are the degeneracy factors. In our analysis, we have included all final states
that are energetically allowed.
One sees from Eqs. (11, 12) that the effects of coannihilation depend simultaneously
on the near degeneracy of any heavier particle with the neutralino, and on the size of the
annihilation cross sections. A large cross section can then compensate partially for the
Boltzman factor. In SUGRA models two effects occur which enhance the coannihilation
for the sleptons. First, since the neutralino is a Majorana particle, its annihilation cross
section is p-wave suppressed. Consequently,
ij , i1
= O[10 11];
i, j = sleptons,
(13)
and the slepton cross sections are enhanced relative to the neutralino cross sections 11 .
Second, a numerical accident allows the selectrons to be nearly degenerate with the
neutralino over a region of parameter space of increasing m0 , m1/2 . One can see this
analytically for low and intermediate tan for mSUGRA where the RGE can be solved
analytically [36]. At the electroweak scale one has
6
2
cos(2),
m
2eR = m20 + f1 m21/2 sin2 W MW
5
m 0 =
1
1
m1/2 ,
G
(14)
(15)
where fi = [1 (1 + i t)2 ]/i , t = ln(MG /MZ )2 and 1 is the U (1) function.

Numerically this gives for, e.g., tan = 5:
m
e2R = m20 + 0.15 m21/2 + (37 GeV)2 ,
m2 0 = 0.16 m21/2.
(16)
Thus for m0 = 0, the eR becomes degenerate with the 10 at m1/2 = 370 GeV,
i.e., coannihilation effects begin at m1/2
= (350400) GeV. For larger m1/2 , the near
degeneracy is maintained by increasing m0 , and there is a corridor in the m0 m1/2 plane
allowing for an adequate relic density. (This is in accord with the numerical calculations
of [30].)
For larger tan , the situation is more complicated, as the the LR mixing in the 1
(mass2 ) matrix, generally makes the 1 the lightest slepton (lighter than the eR or R ), and
hence the 1 dominates the coannihilation phenomena. The RGE must of course now be
solved numerically, and there is strong sensitivity to A0 .
The new effects are exhibited in Fig. 1 where the allowed regions in the m0 m1/2 plane
satisfying Eq. (4) for A0 = m1/2 , 2m1/2 and 4m1/2 are shown for > 0 and tan = 40.
One sees several features of the large tan region there. First, the allowed corridors are
sensitive to the value of A0 , the allowed values of m0 increasing with A0 . Further, the
corridors end at a minimum value of m1/2 due to the b s constraint. (We note that
these lower bounds on m1/2 are sensitive to the NLO corrections to the b s branching
ratio mentioned in Section 1 above.) In fact, there is only the coannihilation region left
65
Fig. 1. Allowed corridors in mSUGRA in the m0 m1/2 plane satisfying the relic density constraint
for > 0, tan = 40 for (from bottom to top) A0 = m1/2 , 2 m1/2 , 4m1/2 . The corridors terminate
at low m1/2 due to the b s constraint.
Fig. 2. ( 0 p )min in the noncoannilation region as a function of m 0 for mSUGRA for > 0,
1
1
tan = 6, obtained by varying over the range of m0 and A0 .
in the parameter space, since the b s constraint at large tan has eliminated the non
coannihilation part of the parameter space. (Actually, there can be small islands of allowed
parameter space left in the non coannihilation region (m1/2 350 GeV) for large tan due
to a cancellation that can occur between the Standard Model amplitude and the chargino
amplitude.) While for low tan , m0 is generally small [30] (m0 200 GeV), we see that
at large tan , m0 can become quite large reaching up to 1 TeV for large A0 . Finally we
mention that while for fixed A0 the allowed corridors are relatively narrow, as one allows
A0 to vary from small to large values, the allowed region due to coannihilation covers most
of the m0 m1/2 plane.
66
Fig. 3. 0 p as a function of m1/2 for mSUGRA > 0, tan = 40 and A0 = 2m1/2 (upper curve),
1
4m1/2 (lower curve).
We turn next to consider the 10 p cross section. Large cross sections can arise for large
tan and small m 0 , and in [33], it was shown that Eq. (1) could be satisfied in mSUGRA
1
only also for small 0 h2 , i.e., for

1
tan 25,
0 h2 0.1,
1
m 0 90 GeV.
1
(17)
Small cross sections generally arise for small tan , large m0 and large m1/2 . It is of interest
to see what the minimum cross section predicted by theory is in order to know how much
of the SUSY parameter space will be examined by proposed detectors. In order to see the
effects of coannihilation, we first examine the domain m1/2 < 350 GeV (m 0 < 140 GeV)
1
where coannihilation does not take place. Fig. 2 exhibits the minimum cross section for
tan = 6, > 0, satisfying all the accelerator and relic density constraints. One sees that
0 p 1 109 pb for tan > 6, and so the entire parameter space in this region would
1
be accessible to detectors planned with sensitivity of Eq. (2).
At higher m1/2 , coannihilation occurs, allowing m0 and m1/2 to become larger, and
hence the cross section should become smaller. We examine first the case > 0. The
A0 dependence at large tan shown in Fig. 1 then implies that the cross section should
also decrease with increasing A0 , since, as seen in Fig. 1, m0 then increases. This is seen
explicitly in Fig. 3.
Thus it is possible that as m1/2 gets large, a consequence of coannihilation is that large
tan and large A0 can give smaller cross sections than low tan . This is illustrated in
Fig. 4 where 0 p is plotted as a function of m1/2 for A0 = 4 m1/2, tan = 40 (upper
1
curve) and tan = 3 (lower curve). One sees that for m1/2 > 600 GeV, the tan = 40
curve actually drops below the tan = 3 curve, the A0 dependence (producing large m0 )
compensating for the tan dependence. One has however, the bound
0 p 1 1010 pb,
1
for > 0, m1/2 < 1 TeV.
(18)
67
Fig. 4. 0 p as a function of m1/2 for mSUGRA, > 0 for A0 = 4 m1/2 , tan = 40 (upper left
1
curve), tan = 3 (lower left curve).
Again, most of this parameter space will be within the reach of planned future detectors.
We turn next to the case of < 0. The situation here is more complicated. As pointed
out for low and intermediate tan in [31], an accidental cancellation can occur between
the Higgs u-quark and the Higgs d-quark scattering amplitudes which can greatly reduce
0 p and we investigate here whether this cancellation continues to occur in the high
1
tan region. Since the cancellation is somewhat subtle, we briefly describe how it comes
Wino (W
) and the two Higgsinos (H
1 , H
2 ):
about. In general, 10 , is a mixture of Bino (B),
+ N12 W
+ N13 H
1 + N14 H
2 ,
10 = N11 B
(19)
where N1i are the amplitudes. In much of the parameter space, the t-channel Higgs
exchanges (h, H ) dominate 0 p . The d and u quark Higgs amplitudes are [37]
1

sin Fh
cos FH
,
+
2MW
cos m2h cos m2H

g22 mu cos Fh
sin FH
u
A =
+
,
2MW sin m2h sin m2H
Ad =

g22 md
(20)
(21)
where is the Higgs mixing angle and

Fh = (N12 N11 tan W )(N14 cos + N13 sin ),
(22)
FH = (N12 N11 tan W )(N14 sin N13 cos ).
(23)
Since the s-quark contribution to the scattering is quite large, the d-quark amplitude Ad
will generally be quite large. However, Ad will be suppressed if the amplitudes N13 , N14
obey the equation
N14 N13
tan + (m2h /m2H ) cot

1 + m2h /m2H
(24)
68
Fig. 5. 0 p for mSUGRA for < 0, A0 = 1500 GeV, for tan = 6 (short dash), tan = 8
1
(dotted), tan = 10 (solid), tan = 20 (dot-dash), tan = 25 (dashed). Note that the tan = 6 curve
terminates at low m1/2 due to the Higgs mass constraint, and the other curves terminate at low m1/2
due to the b s constraint.
In general tan is negative and small (tan

= cot 0.1), and Eq. (24) can happen if
N14 /N13 is positive, which is the case for < 0. Once Ad is sufficiently suppressed, it can
cancel the remaining (smaller) u-quark contribution coming from Au , leading to a nearly
zero value of 0 p . For fixed tan , total cancellation occurs at a fixed value of m1/2 ,
1
though the effects of cancellation has a width. What happens is shown in Fig. 5 where
0 p is plotted in the large m1/2 region for tan = 6 (short dash), tan = 8 (dotted),
1
tan = 10 (solid), tan = 20 (dot-dash) and tan = 25 (dashed). One sees that the cross
section dips sharply when tan is increased from 6 to 8, and and goes through a minimum
for tan = 8 at m1/2
= 810 GeV. For tan = 10, the minimum recedes to m1/2
= 725 GeV,
830
GeV
for
tan = 20,
and then begins to advance for higher tan , i.e., rising to m1/2
=
and m1/2 = 950 GeV for tan = 25. Thus if we restrict m1/2 to be below 1 TeV, the cross
section will fall below the sensitivity of the planned future detectors for m1/2 450 GeV
for a restricted region of tan , i.e.,
0 p < 1 1010 pb for 450 GeV < m1/2 < 1 TeV,
1
5 tan 30.
(25)
At the minimum, the cross sections can become quite small, e.g., < 1 1013 pb,
without major fine tuning of parameters, corresponding to an almost total cancellation.
(Actually, as pointed out in [38], the true minimum of the cross section would then arise
from the spin-dependent part of the scattering, which though very small, does not possess
the same cancellation phenomena.) Further, the widths of the minima for fixed tan are
fairly broad. While in this domain the proposed detectors obeying Eq. (2) will not be able
to observe Milky Way wimps, mSUGRA would imply that the squarks and gluino would
lie above 1 TeV, but at masses that would still be accessible to the LHC (i.e., < 2.5 TeV).
Also note that this phenomena occurs only for < 0 and for a restricted range of tan , so
cross checks of the theory would still be available.
69
Fig. 6. 0 p for mSUGRA for < 0, tan = 10, A0 = 1500 GeV for N = 65 MeV (solid
1
curve) and the parameters of [27,36] (dashed curve) where N = 45 MeV was used.
The minima occurring in Fig. 5 arose from cancellations between the Higgs u-quark and
Higgs d-quark amplitudes, and as such are sensitive to the quark content of the proton. As
discussed in Section 1, N is not yet well determined, and Fig. 5 used N = 65 MeV. In
Fig. 6, we compare this choice (solid curve) for tan = 10 with the parameters used in [27,
31] (dashed curve), where N = 45 MeV. One sees that the minimum at m1/2 = 725 GeV
is shifted to m1/2 = 600 GeV, with analogous shifts occurring for the other values of tan .
Thus the extreme cancellations that occur in the coannihilation region for < 0 are quite
sensitive to the properties of the proton.
3. SUGRA models with nonuniversal soft breaking

We consider here models with nonuniversal soft breaking masses. In order to suppress
flavor changing neutral currents, we maintain the universal soft breaking slepton and squark
masses, m0 , at MG in the first two generations. However, experiment is compatible with
nonuniversal Higgs and third generation soft breaking at MG , and these can arise from
nonuniversal couplings in the supergravity Kahler potential. We parameterize this situation
as follows:
mH1 2 = m20 (1 + 1 ),
mH2 2 = m20 (1 + 2 ),
mqL 2 = m20 (1 + 3 ),
mtR 2 = m20 (1 + 4 ),
mbR 2 = m20 (1 + 6 ),
mlL 2 = m20 (1 + 7 ).
mR 2 = m20 (1 + 5 ),
(26)
where qL = (tL , bL ) squarks, lL = ( , L ) sleptons, etc. The i measure the deviations

from universality. (If one where to impose SU(5) or SO(10) symmetry then 3 = 4 =
5 10 , and 6 = 7 5 .) In the following we limit the i to obey
1 < i < +1.
(27)
70
The lower bound is necessary to prevent tachyons at MG , and the upper bound represents
a naturalness condition. We maintain gauge coupling constant unification at MG (which is
in accord with LEP data). We also assume here gaugino mass unification. (Deviations from
this can arise from nonuniversal couplings in the supergravity gauge function fab , and an
example of this is treated in the next section.)
While the nonuniversal models contain a number of additional parameters, one can
obtain an understanding of the new effects these imply from the following analytical
considerations. In the decomposition of Eq. (19), the lightest neutralino is mostly Bino,
i.e., N11 0.8. The Higgs mixing parameter to a large extent controls this mix, and
as 2 decreases (increases) the higgsino content increases (decreases). In order to see
qualitatively what happens we examine the low and intermediate tan region where the
RGE can be solved analytically. One finds for 2 the result (see, e.g., [39]):

1
t2
1 + D0
1
1 D0
1 3D0
+ 2 +
(3 + 4 )
2 + 2 m20
2 = 2
t 1
2
t
2
2
t
+ universal parts + loop corrections,
(28)
where
t = tan ,
D0
=1
mt
sin
200 GeV
2
= 0.25.
One sees that the universal contribution to the m20 term in 2 is small, making 2
quite sensitive to nonuniversal contributions. Also since D0 is small, the Higgs and
squark nonuniversalities enter coherently, approximately in the combination 3 + 4 2 .
(The RGE can also be solved analytically for the SO(10) model where tan > 40, and
numerically otherwise, with results qualitatively similar to Eq. (28).) In addition we note
here the mass of the CP odd Higgs boson, mA is

t 2 + 1 3(1 D0 ) 1 D0
1 + D0
2
+
(3 + 4 )
2 + 1 m20
mA = 2
t 1
2
2
2
+ universal parts + loop corrections.
(29)
The effects of the nonuniversalities can be seen to fall into the following categories:
(i) The spin-independent neutralinoproton cross section depends upon the interference
between the gaugino and higgsino parts of the neutralino, the larger the amount of
interference the larger the cross section. Thus 0 p can be significantly increased relative
1
to mSUGRA by decreasing 2 (and hence increasing the higgsino content) by choosing

3 , 4 , 1 < 0, 2 > 0, and be decreased (though by not so large amount) with the opposite
sign choice.
(ii) In the early universe annihilation cross section, the s-channel Z 0 pole amplitude
depends on the higgsino contribution of the neutralino (the Bino and Wino parts giving
zero coupling). Thus if 2 is lowered by choosing 3 , 4 , 1 < 0, 2 > 0 one gets an
increased amount of annihilation, and a decreased amount with the opposite choice of
signs. Thus deviations from universality can significantly effect the satisfaction of the relic
density constraints of Eq. (4).
71
Fig. 7. Maximum value of 0 p as a function of m 0 for the nonuniversal model with > 0, 1 ,
1
1
3 , 4 < 0, 2 > 0. The lower curve is for tan = 7, the upper curve is for tan = 12.
(iii) The third generation squark and slepton nonuniversalities shift the mass spectrum
of the particles. Thus the positions of the corridors of coannihilation, and their widths can
be significantly modified (also modifying the cross section 0 p ).
1
Effects of type (i) were early observed in [39], and further discussed in [33] where is
was seen that 0 p could be increased by a factor of 10 or more with the sign choice
1
3,4,1 < 0, 2 > 0. We here update this result to take into account of the latest LEP bound
mh > 114 GeV. Fig. 7 shows the maximum value of 0 p for the nonuniversal models
1
for > 0 with the above choice of signs for i for tan = 7 (lower curve) and tan = 12
(upper curve). One sees that detectors obeying Eq. (1) are sampling the parameter space
with tan 7 (compared with the mSUGRA result of tan 25).
To illustrate further some of the effects of nonuniversality, we consider first the case
where there are only Higgs mass nonuniversal soft breaking, i.e., only 1,2 are non zero.
In general for the universal mSUGRA case, both 2 and mA are determined in terms of
the other SUSY parameters. From Eqs. (28) and (29), we see that these quantities now
depend on the additional parameters 1 and 2 , and thus may be varied keeping the other
(mSUGRA) parameters m0 , m1/2 , A0 and tan fixed. In [40], it was thus assumed that 2
and mA could be chosen arbitrarily. Actually, however, both 2 and mA are constrained
by the bounds of Eq. (27), and varying 2 and mA arbitrarily can lead to points in
parameter space with unreasonably large values of 1,2 (or even tachyonic values with
1,2 < 1). Thus here we will examine the actual parameter space subject to the conditions
of Eq. (27) (as well as the usual relic density and accelerator bounds). As an example of
an effect of type (ii), we consider in Fig. 8 the case in the coannihilation region where
A0 = m1/2 , tan = 40, and chose 2 = 1. Since 2 is insensitive to 1 we set 1 = 0.
In this domain, the allowed stau coannihilation region corresponds to the lowest band in
the m0 m1/2 plane of Fig. 1, with m0 running from about 200 GeV to 400 GeV and the
corresponding corridor is only slightly effected by 2 = 0. As shown in Fig. 8, the effect of
72
Fig. 8. Effect of a nonuniversal Higgs soft breaking mass enhancing the Z 0 s-channel pole
contribution in the early universe annihilation, for the case of 2 = 1, tan = 40, A0 = m1/2 ,
> 0. The lower band is the usual 1 coannihilation region. The upper band is an additional region
satisfying the relic density constraint arising from increased annihilation via the Z 0 pole due to the
decrease in 2 increasing the higgsino content of the neutralino.
the nonuniversal soft breaking now allows the opening up of a new wide band of allowed
region at much higher values of m0 . Thus normally such large values of m0 would give rise
to too little annihilation to satisfy Eq. (4). However, by lowering 2 and hence raising the
higgsino content of the neutralino, the Z 0 channel annihilation is enhanced allowing for an
acceptable relic density. In Fig. 8, the low side of the band corresponds to 0 h2 = 0.25,
1
while the upper end to 0 h2 = 0.02. If one were to raise the relic density lower bound to
1
0 h2 = 0.1, the width of the band would be reduced by about 30% to 40%. (The small
1
dip at m1/2 = (400450) GeV (i.e., m 0 = (160180) GeV) arises due to the opening of the
1
tt channel in the annihilation cross section.) Thus we see that nonuniversal soft breaking
allows for much larger values of m0 than mSUGRA and hence a heavier mass spectrum,
and this could be an experimental signal for nonuniversalities. (Universal soft breaking
only allows for the coannihilation corridor in Fig. 8 at much lower m0 .) In addition, the
larger m0 leads to a smaller 0 p .
1
We consider a second example where both effects of type (ii) and (iii) occur simultaneously, by examining third generation squark and slepton nonuniversal masses obtained
when 10 (= 3 = 4 = 5 ) takes on the value 10 = 0.7. (All other i are chosen zero.)
We again assume A0 = m1/2 , tan = 40 and > 0. The effect of 10 is shown in Fig. 9.
The lowest band is the usual mSUGRA 1 coannihilation band (shown here for reference
only). The upper band corresponds to two phenomena: the lower half is the actual 1 coannihilation region which has been shifted upwards due to the fact that m20 for the 1 has been
replaced by m20 (1 + 10 ), and since 10 is negative, one needs a larger m0 to achieve the
coannihilation corridor. The upper part corresponds to the opening of the Z 0 channel as
in Fig. 8 since 10 < 0 reduces 2 increasing the higgsino content of the neutralino. For
73
Fig. 9. Allowed regions in the m0 m1/2 plane for the case tan = 40, A0 = m1/2 , > 0. The bottom
curve is the mSUGRA 1 coannihilation band of Fig. 1 (shown for reference). The middle band is
the actual 1 coannihilation band when 10 = 0.7. The top band is an additional allowed region
due to the enhancement of the Z 0 s-channel annihilation arising from the nonuniversality lowering
the value of 2 and hence raising the higgsino content of the neutralino. For m1/2 500 GeV, the
two bands overlap.
m1/2 500 GeV, the two bands overlap. However, at higher m1/2 , they are separated. Thus
in the high m1/2 region, the low side of the middle band (where 1 coannihilation is occurring) corresponds to 0 h2 = 0.02, while the upper side corresponds to 0 h2 = 0.25. In
1
the region between the two top bands, 0 h2 > 0.25 (and hence excluded). Finally, since
1
3 and 4 are negative (reducing 2 and also making the total m20 contribution to 2 negative), for sufficiently large m0 , the Z 0 channel opens reducing 0 h2 to 0.25 on the lower
1
side of the top band, that band terminating when 0 h2 = 0.02 on the top.
1
We note in both Figs. 8 and 9, that the effect of nonuniversalities can greatly enhance
the allowed region in the m0 m1/2 plane for large m1/2 , particularly by increasing m0 , and
these changes occur without unreasonable amounts of nonuniversality.
We consider next the neutralinoproton cross sections arising in these nonuniversal
soft breaking cases. One might have expected that the new regions in the m0 m1/2 plane
allowed by the relic density constraint would correspond to smaller values of 0 p since
1
they are bands of much larger m0 . However, this is compensated by the fact that the
nonuniversal effects considered above also reduce 2 and effects of type (i) compensate for
the largeness of m0 and in fact give a net increase to the cross section. This is illustrated for
the example of Fig. 8 in Fig. 10 where 0 p is plotted for 2 =1, tan = 40, A0 = m1/2 .
1
The solid line corresponds to the cross section for the 1 coannihilation corridor (the lower
band of Fig. 8), while the long (short) dashed curves correspond to the cross sections along
the top (bottom) of the Z 0 pole band. We see that the Z 0 band cross sections, where m0 is
very large, are quite large, and this region of parameters should soon be accessible to, e.g.,
74
Fig. 10. 0 p as a function of m1/2 for 2 = 1 (other i = 0), for tan = 40, A0 = m1/2 , > 0.
1
The long (short) dashed curves are the cross sections along the upper (lower) sides of the Z 0 bands
of Fig. 8. The solid curve is 0 p along the 1 coannihilation corridor of Fig. 8.
1
Fig. 11. 0 p as a function of m1/2 for 10 = 0.7 (other i = 0), for tan = 40, A0 = m1/2 ,
1
> 0. The long dashed curve is the cross section along the top of the upper band in Fig. 9
0
where Z annihilation is occurring. (The reduction at m1/2 500 GeV is due to the b s
constraint.) The short dashed curve is the cross section along the bottom of the middle band where 1
coannihilation effects occur for this nonuniversal model. The solid curve is included for comparison
and is the cross section in the 1 coannihilation corridor for the case of universal soft breaking
(mSUGRA).
to the next phase of the CDMS experiment in the Soudan mine. The cross section in the
lower m0 domain corresponding to the 1 coannihilation region are much smaller, and to
explore this domain would require, e.g., the GENIUS or Cryoarray detectors.
In Fig. 11, we examine the expected neutralinoproton cross section for the case of
10 = 0.7, tan = 40, A0 = m1/2 , > 0 of Fig. 9. Here the large dashed curve is the
75
cross section running along the top of the upper band of Fig. 9 (the top of the Z 0 channel
band) and the short dashed curve is the cross section along the bottom of the middle band
(the bottom of the 1 coannihilation corridor). The solid curve is shown for comparison
and would be the cross section in the 1 coannillation corridor for universal soft breaking
(mSUGRA). We see again that the Z 0 corridor with very large m0 would be accessible
to the CDMS Soudan experiment. The actual 1 coannihilation corridor cross section lies
considerably higher than the universal case, but would probably require the GENIUS or
Cryoarray detectors to explore.
In the above discussion, we have chosen the signs of the i parameters to decrease 2
to illustrate the dramatic effects such i can produce without any unusually large amount
of nonuniversality. If instead one were to chose the opposite signs, e.g., 2 < 0, or 10 > 0,
the 2 would have been increased. For this region of parameter space the Z 0 annihilation
channel would now be absent (as in mSUGRA). The usual 1 coannihilation corridors
would of course still be present, but for the 10 > 0 case at lower values of m0 . The cross
sections for > 0 would then be somewhat reduced, but still accessible to future detectors.
The cross sections discussed above were for > 0. We now consider the < 0
case. In the mSUGRA model, it was seen that for large m1/2 with m1/2 < 1 TeV
and a restricted range of tan , a special cancellation could occur which reduced
0 p to be below 1012 pb (see Fig. 5), well below what future detectors could see.
1
This cancellation involved the Higgs u-quark and Higgs d-quark amplitudes, when the
neutralino components obey Eq. (24). However, nonuniversal choices such as those of
Figs. (8)(11), involve now two region of parameter space consistent with the relic density
bounds: the coannihilation region, and the region above it dominated by the s-channel
Z 0 pole. Eq. (24) can in fact hold for the coannihilation corridors, and the minima seen in
the mSUGRA models occur here, at essentially the same points as in Fig. 5. However, for
the Z 0 bands, 2 is quite reduced, and N14 is increased violating Eq. (24). Thus no such
minima occur for m0 in the Z 0 bands of Figs. 8 and 9.
4. D-brane models
Recent studies of Type IIB orientifolds have shown that it is possible to construct a
number of string models with interesting phenomenological properties [12]. The existence
of open string sectors in Type IIB strings implies the presence of Dp-branes, and in the
simplest cases, one can consider the full D = 10 space compactified on a six torus T 6 .
Models of this type contain either 9-branes and 5i -branes, i = 1, 2, 3, or 3-branes and
7i -branes. Associated with each set of n coincident branes is a gauge group U (n). There
are a number of ways in which one can embed the Standard Model in such systems. We
consider here the model where SU(3)C U (1)Y is associated with one set of 5-branes, 51 ,
and SU(2)L with a second intersecting set 52 [11]. Strings starting on 51 and ending on 52
then carry the joint quantum numbers of the two sets of branes, the massless modes then
being the quark and lepton left doublets, and the two Higgs doublets. Strings starting and
76
ending on 51 then carry the quantum numbers of that brane, the massless modes being then
the quark and lepton right singlets.
Supersymmetry breaking can be treated phenomenologically by assuming that the F
components of the dilaton S and the moduli Ti grow VEVs [12,41]:
F S = 2 3 Re S sin b eis m3/2 ,

(30)
Ti
ii
F = 2 3 Re Ti cos b i e m3/2
(31)
where b , i are Goldstino angles (12 + 22 + 32 = 1). In the following we assume
3 = 0. T-duality determines the Kahler potential [12,41], and using Eqs. (30), (31) one
generates the soft breaking terms at MG for the gauginos,
3 = A0 = 3 cos b 1 ei1 m3/2 ,

m
1 =m
(32)

2 1/2
m3/2 ,
m
2 = 3 cos b 1 1
(33)
and for the squarks, sleptons and Higgs,

3 2
2
m12 = 1 sin b m23/2 for qL , lL , H1 , H2 ,
2

2
m1 = 1 3 sin2 b m23/2 for uR , dR , eR .
(34)
(35)
Originally this model was considered to examine the possible effects of the new CP
violating phases in Eqs. (32), (33) and in the and B0 parameters [11,42]. However, it was
subsequently seen that unless tan is small (i.e., tan 5) an unreasonable fine tuning of
the GUT scale parameters is needed if the experimental constraints on the electron and
neutron electric dipole moments are to be obeyed [42]. The LEP data has now eliminated
a considerable amount of the low tan region, and since we are interested here in effects
that arise for large tan , we will in the following set all the phases to zero. The model
then depends on three parameters: b , 1 , and m3/2 , with b < 0.615 and 0 < 1 < 1, and
contains a unique type of nonuniversal soft breaking. In terms of the notation of Eq. (26)
one has
3
1 = 2 = 3 = 7 = sin2 b ,
2
4 = 5 = 6 = 3 sin2 b ,
(36)
(37)
where we have equated m0 to m3/2 . Note that these nonuniversalities hold for all three
generations, not just the third generation.
In addition the gaugino masses at MG are no
longer universal (except when 1 = 1/ 2), and there are two effective gaugino masses
1 = m
3 , and m1/2 = m
2 . The A0 parameter in this
at MG which we label by m1/2 = m
model is then fixed by A0 = m1/2 .
We consider first the question of what parts of the parameter space might be accessible
to current detectors which obey Eq. (1). From Eqs. (34), (35) we see that the sfermion
masses decrease with increasing b , and so we expect the largest cross sections to arise
for small b , and of course for large tan . However, the b s constraint eliminates the
higher values of b . The limiting boundary obeying Eq. (1) is shown in Fig. 12, where
77
Fig. 12. 0 p for D-brane model for > 0, b = 0.2 and tan = 20. The gap in the curve is due
1
to excessive early universe annihilation through the s-channel Z pole.
0 p is plotted for tan = 20 and b = 0.2. For this case, the neutralino mass would be
1
quite small but is still consistent with the LEP bounds. 1 (The gap in the middle is due to
over annihilation in the early universe through the s-channel Z pole.) The cross section
rises, of course with higher tan , and so current detectors are sampling regions where
tan 25.
(38)
It is easy to see that slepton coannihilation effects exist in this model similar to those of
the SUGRA models. Since both the eR and 10 both evolve from the GUT scale using
the U (1) gaugino, Eq. (16) still holds with m20 replaced m23/2 (1 3 sin2 b ). Thus for
fixed b , one can adjust m3/2 so that the 10 and eR are nearly degenerate, allowing for
coannihilation to take place. In addition, however, there can be regions of parameter space
where there is chargino-neutralino coannihilation. This arises because of the nonuniversal
2 becomes smaller
gaugino masses at MG . Thus from Eqs. (32), (33), when 1 gets large, m
than m
1 , and the 10 becomes mainly Wino, and can become degenerate with the 1 .
Coannihilation becomes possible, however, only for a very small range in 1 near 0.9.
As in SUGRA models, a cancellation of matrix elements can occur for < 0, allowing
the cross sections to fall below the sensitivities of planned future detectors. This is exhibited
in Fig. 13, where 0 p is plotted for tan = 6 (solid curve), 12 (dot-dash curve), and 20
1
(dashed curve). (The tan = 6 curve terminates at low m 0 due to the mh constraint,
1
while the higher tan curves terminate at low m 0 due to the b s constraint. The
1
1 LEP determines the bound on m

by measuring lower bounds on the chargino mass and lower bounds
10
on m 0 + m 0 [43]. This combined with lower bounds on the Higgs and slepton masses and the value of the
1
2
Z width, limits the SUSY parameter space giving rise to the m 0 bound. Thus the bounds depend on the SUSY
1
model used, and LEP quotes 37 GeV for the MSSM, and 48 GeV for mSUGRA. We have checked that for the
D-brane model being considered, the LEP data still implies m 0 > 37 GeV.
1
78
Fig. 13. 0 p for the D-brane model for < 0 and tan = 6 (solid), tan = 12 (dot-dash), and
1
tan = 15 (dashed).
upper bound on m 0 , corresponding to mg < 1 TeV, arises from the h2 constraint.) One
1
sees that the cross section goes through a minimum at tan
= 12, though the expected rise
at higher m 0 does not appear since the parameter space terminates before this sets in.
1
5. Stop and sbottom coannihilation

It has been known for some time that the nearness of the t-quark Landau pole can
significantly lower the t1 mass [44], and this raises the possibility of t1 coannihilation
occurring [45]. Working against this possibility is the positive contributions to m2t from
1
the SU(3) gaugino contribution, and of course the m2t contribution. Nonuniversal soft
breaking, however, opens additional possibilities of this type, as it can allow the coefficient
of m20 to become negative, reducing further the t1 mass. To examine this in more detail, we
consider the low and intermediate tan regime where the RGE can be solved analytically.
For the tR one finds

1 D0
2 + D0
(2 + 3 ) +
4 m20
m2t = D0 +
R
3
3

H3
2
1
+ (1 D0 )D0 A0 m1/2 (1 D0 )D0 A20 + em21/2
3
F
2

G 2
1
8
2
f1 f2 + f3
m1/2 + m2t + sin2 W MZ2 cos 2.
+
(39)
3
3
4
3
The fi form factors, i = 1, 2, 3, are numerically approximately 40, 100, 700, respectively,
and e, and H3 /F given in [46], are approximately 2 and +2, respectively. One sees
that for A0 large and negative, the Landau pole A0 terms do give a significant negative
contribution and m2t would be minimized for 2,3 > 0 and 4 < 0. The universal m20 term,
R
scaled by D0
= 0.25, is already quite small, so it does not require too much nonuniversality
to achieve a degeneracy with the 10 . (In fact, this can be done even for the universal case if
79
A0 and m1/2 are chosen sufficiently large. However, the degeneracy occurs in that case for
parameters in the TeV domain.) Since early universe annihilation cross sections involving
the stop are strong interaction phenomena compared with the electroweak neutralino cross
sections, the coannihilation effect can be quite significant. However, it is necessary to
impose the usual constraints, particularly the b s bounds, and also that the stop must
lie below the stau. Thus a satisfactory coannihilation region occurs only when m0 is large
(e.g., greater than about 800 GeV). The neutralinoproton cross section in this domain
generally lies above 1010 pb, and thus should be accessible to future detectors.
Stop coannihilation phenomena are more subtle in the D-brane model, since the
nonuniversalities are related, i.e., 2 = 3 = (3/2) sin2 b m23/2 = 4 /2. The m20 term for
this case is

D0 (1 + 2D0 ) sin2 b m23/2 ,

(40)
which is negative for sin2 b 1/6. (There are also modifications in the gaugino terms
to account for their nonuniversal nature in the D-brane model.) Thus stop coannihilation
effects can occur also in this model. Here, however, the parameter space for coannihilation
is narrowed and generally occurs when m3/2 > 1 TeV.
b-squark coannihilation phenomena can also occur in these models. For low and
intermediate tan , where one may neglect the b-quark Yukawa coupling (relative to the
t-quark coupling), the bR mass is generally large (due to the SU(3) gluino contribution).
However, the bL mass can get small for nonuniversal soft breaking. Thus mbL is given by

1 D0
1 + D0
5 + D0
+
(2 + 4 ) +
3 m20
m2b =
L
2
6
6

1
1
H3
+ (1 D0 )D0 A0 m1/2 (1 D0 )D0 A20 + em21/2
3
F
2

1
8
1 1
G 2
m1/2 + + sin2 W MZ2 cos 2
+ f1 + f2 + f3
15
3
4
2 3
+ m2b ,
(41)
and the light b-squark can be achieved by choosing 2 , 4 > 0 and 3 < 0 so that the
coefficient of the m20 term is negative. In this case the light b-squark is mostly bL , and
since the t and b squarks form an SU(2) doublet, the light stop lies just above the the light
sbottom, and there can be regions where both contribute simultaneously to coannihilation.
However, these coannihilation regions again occur for m0 > 1 TeV.
6. Conclusions
We have considered in this paper coannihilation effects in dark matter neutralino-proton
detection cross sections which arise for large tan and large values of A0 for models
which have grand unification of the gauge coupling constants at the GUT scale MG .
Coannihilation can occur if the light stau (1 ), light chargino ( 1 ), light stop (t1 ) or light
sbottom (b1 ) become nearly degenerate with the neutralino. In order to examine these
80
effects for a wide range of models, we have considered SUGRA models with universal soft
breaking (mSUGRA), nonuniversal soft breaking in the Higgs and third generation squarks
and sleptons, and finally a D-brane model based on type IIB orientifolds which possess
nonuniversal gaugino masses. In all these we have imposed the cosmological neutralino
relic density constraints as well as accelerator constraints on the SUSY parameter space
(including in particular the large tan NLO corrections to the b s decay [17,18] as
well as the large tan SUSY corrections to mb and m ). Radiative breaking of SU(2)
U (1) at the electroweak weak scale also gives strong constraints on the theory.
mSUGRA illustrates some of the effects arising for large tan . Thus for low tan , the
regions in the m0 m1/2 plane allowed by the relic density constraint Eq. (4) is relatively
insensitive to A0 . However, for large tan , the b s constraint allows only the 1 10
coannihilation corridors to remain (except for some small islands when m1/2 350 GeV)
and as A0 increases, these corridors have increasing m0 . Thus m0 can rise to 1 TeV for
large m1/2 . (See, e.g., Fig. 1.) This lowers the value of 0 p and even though 0 p
1
1
increases rapidly with tan , the value of 0 p at large tan and large A0 can be smaller
1
than 0 p at tan = 3, as illustrated in Fig. 4. For > 0, however, one generally finds for
1
the parameter space of Eqs. (6)(8) that 0 p 1010 pb, and so should be accessible
1
to future detectors. In the high tan domain, the b s constraint produces a lower
bound on m1/2 , which decreases with increasing A0 . Thus one finds that m 0 120 GeV
1
(mg 740 GeV) for A0 = 4m1/2 , tan = 40, > 0.
For < 0, a special cancellation can occur in mSUGRA allowing 0 p to become
1
much less than 1010 pb, as has been previously noted for low and intermediate tan [31].
The analytic origin of this effect is discussed in Section 2. The effect, however, occurs
only for a fixed range of tan . Thus the position of these minima begins for tan
= 7 at
m1/2 = 1000 GeV, moves to lower m1/2 until tan increases to 10, and then increases to
beyond m1/2 = 1000 GeV at tan 30. (See Fig. 5.) We note that while the cancellation
occurs at a fixed value of m1/2 , the effects spread over a wide range of m1/2 and in this
region, halo dark matter would be inaccessible to current or planned detectors. However,
if the recently reported anomaly in the muon gyromagnetic ratio [47] is indeed due to
supersymmetry, then is positive and this case would be eliminated.
The nonuniversal soft breaking models illustrate new phenomena at large tan not
present in mSUGRA, and one can also understand these effects analytically. Thus the
value of 2 , obtained from radiative breaking of SU(2) U (1), is a major element in
controlling the value of 0 p , and lowering 2 increases 0 p . Thus for one sign of
1
1
the nonuniversalities, 0 p can be increased by a factor of 10 or more, allowing current
1
detectors to sample the parameter space with tan as low as
= 7. (See Fig. 7.) For
large tan a new effect can occur. The value of 2 also controls the gaugino/higgsino
content of the 10 , and lowering 2 increases the amount of Higgsino in the 10 . This then
increases the amount of early universe annihilation via an off shell s-channel Z boson,
which opens an additional allowed region in the m0 m1/2 plane, considerably wider than
the 1 coannihilation corridor and at much higher m0 . (See Fig. 8.) (Other simple choices of
nonuniversal parameters can also raise the values of m0 for the 1 coannihilation corridor,
81
as illustrated in Fig. 9.) Though m0 is large for this Z-channel annihilation, reducing
0 p , this effect is compensated by the fact that 2 is decreased (which raises 0 p ).
1
1
Thus 0 p remains relatively large (see Figs. 10, 11), and this region of parameter space
1
should become experimentally accessible in the relatively near future (e.g., when CDMS
moves to the Soudan mine).
The D-brane model examined is quite constrained as all the squark and slepton
nonuniversalities are controlled by a single parameter b . Current detectors are sampling
parts of the parameter space where tan 25. Because these models have also
nonuniversal gaugino masses, the LEP lower bound on m 0 is much smaller than in
1
SUGRA models, i.e., the same as in the MSSM: m 0 37 GeV. However, the gaugino
1
nonuniversality still allows the usual type of 1 10 coannihilation to occur. A unique

1 allowing
feature of the gaugino nonuniversality is that m
2 can become smaller than m
the 10 to become mostly Wino and a 1 10 coannihilation domain to exist. However,
the near degeneracy of 1 and 10 occurs only for a very small region of parameter space
(when 1
= 0.9). The cancellation phenomena in 0 p for < 0 also occurs for these
1
D-brane models, the minimum value of m 0 where the cancellation is complete occurring
1
at m 0
= 315 GeV for tan
= 12.
1
Nonuniversal SUGRA models also allow for t1 10 coannihilation to occur. In this case,
the light stop is mostly tR . Then for A0 < 0, the Landau pole contributions to m2t are
R
negative, and one can chose nonuniversal soft breaking so that the coefficient of m20 is
also negative. The t1 then becomes nearly degenerate with the 10 when m0 800 GeV,
coannihilation setting in for that domain. Actually, t1 10 can also occur in mSUGRA, but
to achieve the near degeneracy one needs A0 > 4m1/2 . Similarly, this coannihilation can
occur for the D-brane model but for m3/2 > 1 TeV.
In a similar fashion, b1 10 coannihilation can occur for nonuniversal models when the
b1 is mainly bL . Thus one may chose A0 < 0 and nonuniversal parameters in m2 so
bL
that the coefficient of m2 is negative, making the b 1 nearly degenerate with the 0 . This
0
coannihilation only sets in, however, for m0 > 1 TeV.
Note added
After completing this paper there appeared the paper The CMSSM Parameter Space at
Large tan by J. Ellis, T. Falk, G. Ganis, K.A. Olive and M. Srednicki, hep-ph/0102098.
This paper considers large tan effects for the mSUGRA model only, and is restricted
to A0 = 0.
Acknowledgement
This work was supported in part by National Science Foundation grant No. PHY0070964.
82
References
[1] G. Jungman, M. Kamionkowski, K. Griest, Phys. Rep. 267 (1995) 195;
E. Kolb, M. Turner, The Early Universe, AddisonWesley, Reading, MA, 1990.
[2] K. Griest, D. Seckel, Phys. Rev. D 43 (1991) 3191.
[3] P. Gondolo, G. Gelmini, Nucl. Phys. B 360 (1991) 145.
[4] R. Arnowitt, P. Nath, Phys. Lett. B 299 (1993) 58;
R. Arnowitt, P. Nath, Phys. Lett. B 303 (1993) 403, Erratum.
[5] J. Edsjo, P. Gondolo, Phys. Rev. D 56 (1997) 1879.
[6] J. Ellis, T. Falk, K.A. Olive, Phys. Lett. B 444 (1998) 367.
[7] J. Ellis, T. Falk, K.A. Olive, M. Srednicki, Astropart. Phys. 13 (2000) 181.
[8] For previous works see, e.g., M.E. Gomez, G. Lazarides, C. Pallis, Phys. Rev. D 61 (2000)
123512;
M.E. Gomez, G. Lazarides, C. Pallis, Phys. Lett. B 487 (2000) 313;
S. Khalil, G. Lazarides, C. Pallis, hep-ph/0005021;
M.E. Gomez, J.D. Vergados, hep-ph/0012020.
[9] 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. Hall, J. Lykken, S. Weinberg, Phys. Rev. D 27 (1983) 2359;
P. Nath, R. Arnowitt, A.H. Chamseddine, Nucl. Phys. B 227 (1983) 121.
[10] V. Berezinsky, A. Bottino, J. Ellis, N. Fornengo, G. Mignola, S. Scopel, Astropart. Phys. 5
(1996) 1;
V. Berezinsky, A. Bottino, J. Ellis, N. Fornengo, G. Mignola, S. Scopel, Astropart. Phys. 6
(1996) 333;
P. Nath, R. Arnowitt, Phys. Rev. D 56 (1997) 2820;
R. Arnowitt, P. Nath, Phys. Lett. B 437 (1998) 344;
A. Bottino, F. Donato, N. Fornengo, S. Scopel, Phys. Rev. D 59 (1999) 095004;
R. Arnowitt, P. Nath, Phys. Rev. D 60 (1999) 044002.
[11] M. Brhlik, L. Everett, G. Kane, J. Lykken, Phys. Rev. D 62 (2000) 035005.
[12] L. Ibanez, C. Munoz, S. Rigolin, Nucl. Phys. B 536 (1998) 29.
[13] G. Jungman et al., in Ref. [1].
[14] A. Bottino, N. Fornengo, hep-ph/9904469.
[15] A. Corsetti, P. Nath, Int. J. Mod. Phys. A 15 (2000) 905.
[16] P. Igo-Kimenes, Talk presented at ICHEP 2000, Osaka, Japan, July 27 August 2, 2000.
[17] G. Degrassi, P. Gambino, G. Giudice, JHEP 0012 (2000) 009.
[18] M. Carena, D. Garcia, U. Nierste, C. Wagner, hep-ph/0010003.
[19] I. Trigger, OPAL Collaboration, Talk presented at DPF 2000, Columbus, OH, USA;
T. Alderweireld, DELPHI Collaboration, Talk presented at DPF 2000, Columbus, OH, USA.
[20] D0 Collaboration, Phys. Rev. Lett. 83 (1999) 4937.
[21] M. Alam et al., Phys. Rev. Lett. 74 (1995) 2885.
[22] M. Fukugita, hep-ph/0012214.
[23] J. Ellis, R. Flores, Phys. Lett. B 263 (1991) 259;
J. Ellis, R. Flores, Phys. Lett. B 300 (1993) 175.
[24] A. Bottino, F. Donato, N. Fornengo, S. Scopel, Astropart. Phys. 13 (2000) 215.
[25] H. Leutwyler, Phys. Lett. B 374 (1996) 163.
[26] M. Olsson, Phys. Lett. B 482 (2000) 50;
M. Pavan, R. Arndt, I. Strakovsky, R. Workman, nucl-th/9912034.
[27] J. Gasser, M. Sainio, hep-ph/0002283.
[28] R. Arnowitt, B. Dutta, Y. Santoso, hep-ph/0010244;
R. Arnowitt, B. Dutta, Y. Santoso, hep-ph/0101020.
[29] A. Bottino et al., in Ref. [9].
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
83
J. Ellis, T. Falk, K.A. Olive, M. Srednicki, Astropart. Phys. 13 (2000) 181.

J. Ellis, A. Ferstl, K.A. Olive, Phys. Lett. B 481 (2000) 304.
J. Ellis, T. Falk, G. Ganis, K.A. Olive, Phys. Rev. D 62 (2000) 075010.
E. Accomando, R. Arnowitt, B. Dutta, Y. Santoso, Nucl. Phys. B 585 (2000) 124.
R. Arnowitt, B. Dutta, Y. Santoso, hep-ph/0005154.
M. Srednicki, R. Watkins, K.A. Olive, Nucl. Phys. B 310 (1988) 693.
L. Ibanez, C. Lopez, Nucl. Phys. B 233 (1984) 511.
J. Gunion, H. Haber, G. Kane, S. Dawson, The Higgs Hunters Guide, AddisonWesley,
Reading, MA, 1990.
V. Bednyakov, H. Klapdor-Kleingrothaus, hep-ph/0011233.
P. Nath, R. Arnowitt, Phys. Rev. D 56 (1997) 2820.
J. Ellis, A. Ferstl, K. Olive, hep-ph/0007113.
A. Brignole, L. Ibanez, C. Munoz, C. Scheich, Z. Phys. C 74 (1997) 157;
A. Brignole, L. Ibanez, C. Munoz, Nucl. Phys. B 422 (1994) 125;
A. Brignole, L. Ibanez, C. Munoz, Nucl. Phys. B 436 (1995) 747, Erratum.
E. Accomando, R. Arnowitt, B. Dutta, Phys. Rev. D 61 (2000) 075010.
ALEPH Collaboration, CERN EP/2000-139.
P. Nath, J. Wu, R. Arnowitt, Phys. Rev. D 52 (1995) 4169;
M. Carena, C.E.M. Wagner, Nucl. Phys. B 452 (1995) 45.
For earlier work see: C. Boehm, A. Djouadi, M. Drees, Phys. Rev. D 62 (2000) 035012.
L. Ibanez, C. Lopez, C. Munoz, Nucl. Phys. B 256 (1985) 218.
H.N. Brown et al., hep-ex/0102017.

Tunneling of BornInfeld strings to D2-branes

D.K. Park a,b, , S. Tamaryan a , Y.-G. Miao a , H.J.W. Mller-Kirsten a
a Department of Physics, University of Kaiserslautern, D-67653 Kaiserslautern, Germany
b Department of Physics, Kyungnam University, Masan, 631-701, South Korea
Received 25 January 2001; accepted 9 March 2001
Abstract
A BornInfeld theory describing a D2-brane coupled to a 4-form RR field strength is considered,
and the general solutions of the static and Euclidean time equations are derived and discussed. The
period of the bounce solutions is shown to allow a consideration of tunneling and quantumclassical
transitions in the sphaleron region. The order of such transitions, depending on the strength of the RR
field strength, is determined. A criterion is then derived to confirm these findings. 2001 Elsevier
Science B.V. All rights reserved.
1. Introduction
Recently it was argued [1] that a fundamental string can be viewed as a collapsed brane,
or more precisely as a p-brane (p 2) with (p 1) spatial directions of its worldvolume
(e.g., with spatial part R S p1 ) collapsed to zero size (S p1 shrunk to zero). This view is
somehow a reversal of the traditional view of considering D-branes as extended objects on
which open strings can put their ends [2], and whose low energy dynamics is described by
nonlinear BornInfeld theory on the worldvolume of the brane [3]. The latter view implies,
as has been shown [4] the existence of string-like configurations. In Ref. [1] the reversal
of the traditional view was considered and the question was asked whether by application
of external forces (here an RR-spacetime gauge potential) a string may expand back into a
D-brane, and this was in fact shown to be possible in the D2-brane case considered via a
tunneling process.
In the following our objective is to investigate in more generality the problem posed
in Ref. [1] and its solutions and the tunneling process, and to determine the order of the
* Corresponding author.
E-mail addresses: dkpark@hep.kyungnam.ac.kr (D.K. Park), sayat@physik.uni-kl.de (S. Tamaryan),

miao@physik.uni-kl.de (Y.-G. Miao), mueller1@physik.uni-kl.de (H.J.W. Mller-Kirsten).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 3 1 - 6
D.K. Park et al. / Nuclear Physics B 606 (2001) 84100
85
quantumclassical transition (in analogy with phase transitions) at nonzero temperatures

as a function of the external force.
The decay of a BornInfeld braneantibrane system by tunneling was discussed earlier
in Ref. [4] but was in that context discarded as being physically irrelevant. The D2-brane
model of [1], however, seems a fascinating and computationally manageable model which
allows the investigation of various aspects of tunneling related to strings.
In Section 2 we describe the formulation of the problem and its solutions. In Section 3
we calculate the wavelength and energy of the general static configuration. In Section 4
we determine the Euclidean version of the model for the torus brane (defined by constant
radius) and determine the period of the bounce configuration. With this and the sphaleron
configuration we can determine the phase diagrams for quantum classical transitions of the
string of vanishing radius into a brane of constant radius.
It is found that a weak RR field leads to a sharp transition described as of the first order,
whereas a stronger RR field leads to one of the second order, described as smooth. In
Sections 5 and 6 we then determine and discuss a criterion for the occurrence of transitions
of either order. Finally in Section 7 we summarize our findings in some conclusions.
2. Formulation of the problem and its solutions

Our starting point is the same BornInfeld action describing a D2-brane coupled to the
RR-background spacetime 3-form gauge potential A with field strength H = dA as that
of Ref. [1]; we also use the same conventions (e.g., = 1). Thus the action we consider is

1
ind
1
3
I = 2
det g + 2F +
A X X X . (1)
d
3!
4 g
Here , , = 0, . . . , 9 are spacetime indices, and , , = 0, 1, 2 worldvolume indices,
g is the (type IIA) string coupling and the dilaton is taken to be constant. The induced
metric is the pullback of the spacetime metric (here assumed to be flat) and thus is
X X . The U (1) gauge field tensor F contains in particular the electric field
E(t, z) := 2(0 Az z A0 ). The background spacetime gauge field H is taken to be
H0123 = h = const. As in Ref. [1] we choose the world volume to be cylindrical and hence
define
X0 = t,
X1 = z,
X2 = R(t, z) cos ,
X3 = R(t, z) sin ,
and all other Xi = const, and R > 0, 0 < < 2 . We also restrict our considerations here
only to those of the electric field component of the gauge field A . The action in terms of
the worldvolume coordinates t, z, with integrated out then becomes

h
1
R 1 R 2 E 2 + R 2 R 2 ,
I = dt dz L,
L=
(2)
2g
2
where dots and primes denote derivatives with respect to t and z, respectively. The
A 0 , A z ) yields, in particular, the Gauss law type of
Lagrangian density L(R, A0 , Az , R,
86
equation
D
= 0,
z
D=
ER
(3)
= const.
1 R 2 E 2 + R 2
In the static case to be considered below the constant D e can be looked at as a charge
per radian, i.e., 2e(z)A0 would be the appropriate source term in the static Lagrangian.
Solving Eq. (3) for E one finds

D 1 R 2 + R 2
E=
(4)
.
D2 + R2
Since D is constant and so unaffected by the form of R it is convenient to express the
action in terms of D and regard the result as a functional of R so that the variation of the
with respect to E is zero. The appropriate Legendre transform

new Lagrangian density L
of L is
= L DE
L
(5)
2g
D
with the help of Eq. (4) one obtains as in [1]

since with Eq. (3) L/E = 2g
. Evaluating L

h
= 1
L
(6)
R 2 + D 2 1 R 2 + R 2 R 2 .
2g
2
PA0 =
Since L is a functional of R, A0 , Az , we have conjugate momenta P = L/ R,
L/ A0 = 0 and PAz = D/g. As usual the vanishing of PA0 implies the Gauss law as the
constraint (3), so that the Hamiltonian H is defined by

1 D
,
H = dz P R + PAz Az L + A0
(7)
g
z
where A0 acts as Lagrange multiplier. Evaluating this, one obtains

1
(1 + R 2 ) D 2 + R 2 h 2
H=

R .
dz
2g
2
1 R 2 + R 2
The static energy E is therefore

h 2
1

2
2
2
E=
dz (1 + R )(D + R ) R .
2g
2
Variation of E with respect to R yields the equation for the static configuration

R2 + D2 h 2
R = C,
1 + R 2
2
where C is an integration constant. We rewrite this equation

h
2 R 2 )(R 2 R 2 ), R = 0,
(R+
R = () 2
hR + 2C
with

2
2
R
= 2 (1 Ch) 1 2Ch + h2 D 2 ,
h
(8)
(9)
(10)
(11)
(12)
87
and assume for simplicity that C D. We note also for later reference

4 2
4
2
2
2
(13)
C
,
R+
D
+ R
= 2 (1 Ch).
h2
h
Many special solutions have already been discussed in Ref. [1]. E.g., for h = 0 the case
C = D implies spikes R = R0 exp(z/C) for z < 0 or z > 0 respectively corresponding
2
2
to
the BIon in D3 BornInfeld theory [5], and the case C > D implies catenoids R =
2
2
C D cosh(z z0 ). Here we are interested in the general solution. However, we note
the possibility of the two signs of R in Eq. (11). The significance of these is the same as
those of the catenoid discussed in [5]. There one sign gives the spike-like solution which
minimizes the action, the other sign the solution which maximizes the action. The so-called
wormhole solution is the two-spike solution obtained by matching the two at a suitable
point.
We now consider the general solution of the static equation (10). Assuming R R
R+ and R(z0 ) = R+ , Eq. (10) becomes
2 2
R+
R =
R+
dR
R
hR 2 + 2C
2 R 2 )(R 2 R 2 )
(R+
= h(z0 z).
(14)
We observe that for real values of z the solution has turning points at R = R+ and R . The
distance between these therefore defines half a wavelength of oscillations between these.
The left hand side of Eq. (14) can be evaluated in terms of elliptic integrals. Then the
general solution is given by (cf. Ref. [6], pp. 56, 300, 301)
hR+ E(, k) +
2C
F (, k) = h(z0 z).
R+
(15)
Here F (, k) and E(, k) are the incomplete elliptic integrals of the first and second kinds
respectively, k is the elliptic modulus given by

2 R2
R+
k=
,
R+
and is the angle defined by

2 R2
R+
= sin1
.
2 R2
R+
We now consider two important limiting cases:

(a) The limit C = D. In this case
R+ =
2
1 Dh,
h
and
sin =
R = 0,

2 R2 R .
R+
+
k=1
88
Using E(, 1) = sin , F (, 1) = ln(tan + sec ) Eq. (15) becomes

2 R2

R
+
R+
+
2 R 2 + 2D ln
= (z0 z),
R+
hR+
R
(16)
which agrees with a result of Ref. [1] and describes a BI-string with spheroidal bulge
centered at z0 where R = R+ . For R R = 0 the solution becomes stringlike
(this case may be dubbed vacuum solution).
(b) The limit C = (1 + h2 D 2 )/2h is the limit of the sphaleron configuration, i.e., that
with zero separation between the turning points. In this case
2(1 Ch)
,
k = 0.
R = R+ =
h
Using
E(, 0) = F (, 0) = ,
Eq. (15) becomes

2C
hR+ +
= h(z0 z),
R+
i.e.,
= h(z0 z),
1 h2 D 2
which implies
R
2
= R+
2 h 1 h2 D 2 (z z0 )
2
2
2
= R+
R+ R sin
2
so that
2(1 Ch) 1
=
R = R+ =
1 h2 D 2 RS ,
h
h
where RS is the sphaleron configuration, in agreement with a result discussed in [1]

and of particular interest for h = 0.
3. Calculation of wavelength and energy of the general static configuration

Having prepared the ground for our considerations we now proceed to calculate the
wavelength of the general static solution of Eq. (14). As mentioned above, the points
R+ , R are turning points, so that with R in Eq. (14) replaced by R , we can put z z0 =
/2. Then
R+

hR 2 + 2C
1
= hR+ E 2 , k + 2C
dR
F 2 ,k
R+
2 R 2 )(R 2 R 2 )
(R
+
R
= hR+ E(k) +
2C
K(k) = h ,
R+
2
(17)
89
where E(k) and K(k) are the complete elliptic integrals of the second and first kinds,
respectively. The wavelength can therefore be expressed as
= 2R+ E(k) +
4C
K(k).
hR+
(18)
Using the special values E(1) = 1, K(1) = , E(0) = K(0) = /2 (cf. [6], p. 10), we can
obtain the values of of the two limiting cases discussed above. Thus for (a) with C =
D, one obtains = or frequency = 0, and for (b) with C = (1 + h2 D 2 )/(2h), the
sphaleron limit, one obtains = 2/(h 1 h2 D 2 ). With this we can define as sphaleron
frequency

sph = h 1 h2 D 2 .
Next we compute the energy of the solution (14) by replacing R 2 in E by the expression
obtained from Eq. (10), i.e.,
R 2 =
4(R 2 + D 2 )
1.
(hR 2 + 2C)2
(19)
We rewrite the expression (9) with the contribution of the fundamental BI string (the limit
C D = 0, R = R = 0) separated in view of its divergence in the linear case, i.e., we
write with a somewhat arbitrary subdivision

D
E=
dz + U0 ,
2g

1
2(R 2 + D 2 ) h 2
R D .
U0 =
(20)
dz
2g
hR 2 + 2C
2

Here the first term in E divided by the length dz gives the tension of the BI string, i.e.,
TBI = D/2g. As shown in Ref. [4] the quantization condition on the D-flux is

1
D = gn, i.e., D = gn,
2
S1
where n is the number of fundamental strings, each of tension 1/2 , adsorbed by the
BI string. Thus the first contribution to E in Eq. (20)is the energy of n fundamental strings.
R
Then by explicit calculation and using 0 dz = 2 R+ dR/R and again replacing R by

the expression of Eq. (19), we obtain (cf. Appendix A)

4
12D(D C) + (1 k 2 )h2 R+
1
K(k)
U0 =
6gh
R+

2 2 2
+ 2R+ 3 2 h(C + D) (2 k )h R+ E(k) .
(21)
2 -dependence is still contained in k 2 . Again we consider the two limiting
Of course, the R
cases of above. In case (a), the limit C = D with
lim (1 k 2 )K(k) = 0,
k1
90
one can show directly that

U0 =
h 3
R ,
6g +
R+ =
2
1 Dh,
h
k = 1,
(22)
which is consistent with Eq. (15) of Ref. [1] and represents a bulk energy. In case (b), the
sphaleron limit C = (1 + h2 D 2 )/2h, direct calculation yields
(1 hD)2
2(1 Ch)
1 h2 D 2
,
R+ =
U0 =
(23)
=
,
k = 0.
3
2R+ gh
h
h
To
show that this is physically relevant, we insert the constant value R = RS =
1 h2 D 2 / h into the expression for the static energy E of Eq. (9) and integrate over
a length L. Then E becomes
E=
1 + h2 D 2
L.
4gh
From Eq. (20) we know that this energy has to be

E=
DL
+ U0 ,
2g
i.e., this has to be satisfied. Equating the two expressions we obtain

U0 =
(1 hD)2
L.
4gh
If we put here L = = 2/h 1 h2 D 2 , we obtain

U0 =
(1 hD)2
2R+ gh3
in agreement with Eq. (23).
4. The quantumclassical decay rate transition to the torus brane

In the following we consider first the solutions to the Euclidean version of the action (1),
then specialise to the torus brane defined by the condition R = 0, obtain the corresponding
periodic bounce solutions, and then the phase diagrams.
We are interested in solutions to the Euclidean version of the action (1), i.e. (cf. Eq. (6)),

h
1
IE (D, R) =
(24)
d dz (R 2 + D 2 ) 1 + R 2 + R 2 R 2
2g
2
where R now denotes differentiation with respect to Euclidean time = it. The equation
of motion derived from (24) is

2 + D2 )
R(1 + R 2 + R 2 )
R(R

(R 2 + D 2 )(1 + R 2 + R 2 ) (R 2 + D 2 )(1 + R 2 + R 2 )
R (R 2 + D 2 )

hR = 0.
z (R 2 + D 2 )(1 + R 2 + R 2 )
(25)
91
(a)
(b)
Fig. 1. (a) The period P of the bounce as a function of the parameter C for D = 1 and h = 0.7.
(b) The sphaleron action and classical action per unit length as function of temperature T for D = 1
and h = 0.7.
Here we consider only tunneling to the torus brane (R = 0). In this case the equation
reduces to

R(1 + R 2 )
(R 2 + D 2 )(1 + R 2 )
2 + D2 )
R(R

hR = 0.
(R 2 + D 2 )(1 + R 2 )
(26)
After some manipulation one can show that Eq. (26) is exactly the same as Eq. (11) if
z there is replaced by . In fact this is easily understood from the symmmetry of the
Euclidean action (24) under the exchange z. Hence the periodic bounce solution is
in this case obtained directly from Eq. (15) by changing z0 z into 0 , i.e., from
92
(a)
(b)
Fig. 2. (a) The period P of the bounce as a function of the parameter C for D = 1 and h = 0.3.
(b) The sphaleron action and classical action per unit length as function of temperature T for D = 1
and h = 0.3.
hR+ E(, k) +
2C
F (, k) = h(0 ).
R+
(27)
With this expression for Euclidean time 0 we can calculate the period P . We first
consider the Euclidean action for this configuration with R = 0 (as stated earlier) which is

1
2
2 + D 2 )(1 + R
2 ) h Rcl
IE,cl (D, Rcl ) =
d dz (Rcl
cl
2g
2

D
=L
dz + U0 ,
2g
(28)
93
where U0 is given by Eq. (21) and L is now the circumference of the torus in Euclidean
time. We write the part with the contribution of fundamental strings subtracted out
(0)
(D, Rcl ) = LU0 (C, D, h).
IE,cl
(29)
As in statistical mechanics the period P of this configuration is to be identified with the

reciprocal of temperature T (cf. Ref. [7]). Since the wavelength = 2/, where is
the corresponding angular frequency, we can rewrite Eq. (27) here
P = 2R+ E(k) +
4C
1
K(k) = .
hR+
T
(30)
The sphaleron configuration in this tunneling is
1 h2 D 2
.
R = RS =
h
The barrier height or sphaleron energy is
Esph =
L
(1 hD)2
4gh
(31)
as observed at the end of Section 3. From this we obtain for the sphaleron action
Esph
.
(32)
T
With the formulas at hand we can now plot the phase diagrams. The integration constant
C here plays the same role as the energy E of a pseudoparticle in the usual quantum
mechanics around such a configuration (E = 0 in the latter being its value for the vacuum
solution, and E = Esph that for the sphaleron, see, e.g., [9] and [10]). In the present case
the vacuum solution has C = D, and the sphaleron C = (1 + h2 D 2 )/2h. In Fig. 1(a)
we show the behaviour of the period P as a function of C for D = 1 and h = 0.7. We
observe a monotonically decreasing behaviour characteristic of a smooth second order
transition. In Fig. 1(b) we plot for the same values of D and h the behaviour of the
(0)
(0)
/L and the Euclidean action per unit length IE,cl
/L
sphaleron action per unit length Isph
versus temperature T . We observe the expected smooth transition from one to the other. In
Fig. 2(a) and (b) we plot the same quantities again for D = 1 but for the smaller value h =
0.3 of the RR field. This time we observe a rise of the period beyond a critical value, and
a corresponding nonsmooth bifurcation of the action implying a sharp quantum classical
transition, generally described as of first order.
(0)
Isph
=
5. The criterion for a sharp transition

A criterion for the occurrence of sharp or first order quantumclassical transitions was
developed in Ref. [7]. A more explicit criterion useful for practical purposes was obtained
in Ref. [8]. Our considerations here are based on this latter reference where a detailed
explanation of steps can be found. Further applications of the criterion have been given in
94
Refs. [9] and [10]. The criterion for occurrence of a first order transition is 2 S2 > 0,
where S is the sphaleron frequency and the possible frequency different from S [8].
With some manipulations Eq. (26) can be put into the following form

2

1 + R 2
1
+
R
R R 2
(33)
+ hR 1 + R 2
= 0.
2
R +D
R2 + D2
We expand this equation around the sphaleron configuration RS by setting
1
R = RS + R( ),
RS =
1 h2 D 2 .
h
Using the following expansions

1
2
1 2h 1 h2 D 2 R + h2 3 4h2 D 2 R 2
=
h
2
2
R +D

+ 4h3 2h2 D 2 1 1 h2 D 2 R 3 + ,
(34)

1 + R 2
2
1
2h
=
h
1 h2 D 2 R + h2 3 4h2 D 2 R 2 + R 2
R2 + D2

+ 4h3 2h2 D 2 1 1 h2 D 2 R 3 2h 1 h2 D 2 R 2 R

+ ,

1 + R 2
1
=
h
1
h
1 h2 D 2 R + h2 2 3h2 D 2 R 2 + R 2
2
2
2
R +D

h
2 2 2
3
2
2
2
1 h D 2 5h D h R + R R +
(35)
2
one then obtains

1 + R 2
= h 1 h2 D 2 h 1 2h2 D 2 R
R 2
2
R +D

+ 1 h2 D 2 h2 1 4h2 D 2 R 2 + R 2

h h2 1 8h2 D 2 + 8h4 D 4 R 3 + 1 2h2 D 2 R 2 R

+
and

1 + R 2
R 1 + R 2
R2 + D2

3 4 2 2 3 2
3 2
2
2
2
2
= 1 h D + h D R + 1 h D h D R + R
2
2

5
3
+ h3 2h2 D 2 h4 D 4 R 3 + h3 D 2 R R 2 +
2
2
(36)
Eq. (33) can now be expanded as

R + h2 1 h2 D 2 R + G2 (R) + G3 (R) + = 0,
(37)
(38)
95
where

h 1 h2 D 2 2
h 2 5h2 D 2 R 2 + R 2 ,
G2 (R) =
2

h2 2
G3 (R) =
h 2 12h2 D 2 + 11h4 D 4 R 3 + 2 h2 D 2 R 2 R .
2
In lowest or linear order we have, therefore,

R + h2 1 h2 D 2 R = 0,
with solution
R = a cos S ,

S = h 1 h2 D 2 ,
(39)
(40)
(41)
where a is a small amplitude of oscillation around sphaleron. We observe that the sphaleron
period resulting from Eq. (41), i.e.,
2
2
=
S
h 1 h2 D 2
coincides exactly, as expected, with the wavelength defined after Eq. (18).
We proceed to the second order calculation by setting
Psph =
R = a cos + a 2 1 ( ),
2 = S2 + a41 2
(42)
with a new or as yet undetermined frequency . Inserting this trial solution into
Eq. (38) and reexpressing squares of cos by cos 2 we obtain at this level of the
approximation
2
1

S 2
2
h 2 5h2 D 2 2
+ S
1 =
41 2 cos S +
2
4

+ h2 2 5h2 D 2 + 2 cos 2 .
(43)
In order to avoid infinity and have a defined fluctuation, the square bracket on the right
2

2
hand side must not contain the zero mode of the operator
2 + S . Hence we must
demand 41 2 = 0. This condition therefore yields
= S
(44)
to this order of the calculation together with the fluctuation

1 = g1 + g2 cos 2S ,
(45)
where
S 1 4h2 D 2
S 1 2h2 D 2
,
g
=
.
2
4 1 h2 D 2
4 1 h2 D 2
We proceed to the third order by setting
g1 =
(46)
R = a cos + a 2 1 ( ) + a 3 2 ( ),
2 = S2 + a41 2 + a 2 42 2 .
(47)
96
Inserting this ansatz into Eq. (38) yields with a procedure as before but now at this level

1
1 2
2
2 = 3
+
S
a 2
(0)

(1)
(2)
(3)
2 + 2 cos S + 2 cos 2S + 3 cos 3S ,
(48)
where (only this is needed here)

g2
+ S3 g2
2(1) = a 2 S2 a 3 h2 (2 5h2 D 2 )S g1 +
2

h2 2

3 4
2 2
4 4
2 2
+ h 2 12h D + 11h D + S 2 h D
8
8
4

h
= a 3 42 2 a 3 1 4h2 D 2
(49)
.
2
Again we set this equal to zero to avoid infinity in (48) and so determine 42 2 . The
criterion for a sharp or first order transition 2 S2 > 0 then implies
1
(50)
hD < .
2
We observe that this result agrees with our earlier findings that the smooth second order
transition occurs for larger values of the applied force or h. Thus also the bifurcation point
in the plot of the action must disappear in the large h domain. We now investigate this point
in more detail.
42 2 > 0,
i.e.,
6. Tunneling of the string to an arbitrary D2-brane

It is much more difficult to understand the tunneling associated with the sphaleron of the
general static solution (15), i.e., for general values of in
hR+ E(, k) +
2C
F (, k) = h(z0 z).
R+
(51)
However, we can understand qualitatively this tunneling using the tunneling to the torus
investigated in the previous sections. We first compute the barrier height which is equal to
the sphaleron energy, i.e.,

1
h 2
2
2

2
Esph =
dz (R + D )(1 + R ) R c.f.s.
2g
2

2
2
4
12D(D C) + (1 k )h R+
1
=
K(k)
6gh
R+

2
E(k) ,
+ 2R+ 3 2 h(C + D) (2 k 2 )h2 R+
(52)
where c.f.s. is the contribution of the fundamental strings. In fact, the expression (52)
is exactly the same as the classical action for tunneling to the torus divided by L,
97
the circumference of the torus, as one can see by comparison with Eq. (21). With this
observation we can deduce the following facts.
(1) In the region of hD > 12 the energy Esph is minimal for tunneling to the torus
(R = 0). Hence tunneling to the torus may be the dominant process.
(2) In the region of hD < 12 the energy Esph has a minimum at the tunneling
which corresponds to the bifurcation point in the action-versus-temperature diagram (cf.
Fig. 2(b)). Thus in this region the dominant tunneling may not be tunneling to the torus but
to a wiggly distorted brane.
It is important therefore to understand when the bifurcation point appears. We can study
this, for instance, by determining those values of C = C(D, h), which correspond to the
bifurcation points in tunneling with R = 0. In order to determine this parameter C as a
function of D and h, we recall that the period P of the bounce configuration for a first
order transition has an extremum period as can also be seen from Fig. 2(a) (or Ref. [11]).
Thus this C(D, h) is determined by

dP
= 0.
dC C=C(D,h)
(53)
Here P is, of course, the same as the wavelength of Eq. (18). In Fig. 3(a) the solution
C(D, h) of this equation is plotted for D = 1 as a function of h. One can see that the
maximum of C meets the bifurcation point at C = 1.2 (cf. Fig. 2(a)) at exactly h = 0.5, the
limit set by the criterion Eq. (50). Fig. 3(b) shows a corresponding plot for D = 2. Both of
these plots therefore show how the bifurcation point disappears in the large h region, i.e.,
beyond the limit given by the criterion (50). Finally, in Fig. 4 we show the shape of a brane
at a bifurcation point.
7. Conclusions
The expansion of a string into a D2-brane by tunneling through a barrier provided by a
4-form RR field strength was already shown to be possible in Ref. [1]. In the above we have
investigated this problem in more detail by deriving the explicit solutions of the classical
equation in terms of elliptic functions. These solutions then permitted the derivation of the
period of the bounce configurations, as well as a detailed study of the sphaleron at the top
of the potential barrier. With these tools it was possible to study the quantumclassical
behaviour at nonzero temperatures in the sphaleron domain, and to determine the order
of the transitions (in analogy with phase transitions) depending on the magnitude of the
applied RR field. We then confirmed these findings with the derivation of a criterion for
the occurrence of such transitions in the present case. The case of the D2-brane discussed
here has the advantage of permitting explicit calculations with relative ease. We expect the
method, however, to be applicable also to more complicated cases.
98
(a)
(b)
Fig. 3. (a) The parameter C as solution of dP /dC = 0, P the period of the bounce, plotted as a
function of h for D = 1. The dotted line represents a maximum of C at given h. One should observe
the disappearance of the bifurcation point at C = 1.2 (cf. Fig. 2(a)) beyond h = 0.5 in agreement
with the criterion of Section 5. (b) The parameter C as solution of dP /dC = 0, P the period of the
bounce, plotted as a function of h for D = 2. The dotted line represents a maximum of C at given h.
One should observe the disappearance of the bifurcation point beyond h = 0.25 in agreement with
the criterion of Section 5.
99
Fig. 4. The shape of the brane at the bifurcation point for D = 1, h = 0.3, C = 1.20434.
Acknowledgements
D.K.P. and S.T. acknowledge support by the Deutsche Forschungsgemeinschaft (DFG),
Y.-G.M. acknowledges support by an A. von Humboldt research fellowship. D.K.P. is also
supported by Korea Research Foundation Grant (KRF-2000-D00073).
Appendix A
Here we indicate briefly how Eq. (21) is obtained. Substituting R of Eq. (11) into the
expression (20) for U0 one obtains an expression that can be written

R+
2
R 4 dR
h2

U0 =
gh
4
2 R 2 )(R 2 R 2 )
(R+
R
R+
+ D(D C)
R
dR
2
(R+
2)
R 2 )(R 2 R

R+

h(C + D)
R 2 dR

.
+ 1
2
2 R 2 )(R 2 R 2 )
(R
+
(A.1)
Using formulas 218.01 and 218.00 of Ref. [6] one can obtain
R+

R
R 2 dR
2 R 2 )(R 2 R 2 )
(R+
= R+ E(k)
(A.2)
and
R+

R
dR
2 R 2 )(R 2 R 2 )
(R+
1
K(k).
R+
(A.3)
100
Also using formulas 218.06 and 314.04 of Ref. [6] one finds
R+

R

1 3
= R+
(1 k 2 )K(k) + 2(2 k 2 )E(k) .
3
2 R 2 )(R 2 R 2 )
(R+
R 4 dR
(A.4)
Inserting these expressions into Eq. (A.1) one obtains Eq. (21).
References
[1] R. Emparan, BornInfeld strings tunneling to D-branes, Phys. Lett. B 423 (1998) 71, hepth/9711106.
[2] J. Polchinski, TASI lectures on D-branes, hep-th/9611050.
[3] R.G. Leigh, Mod. Phys. Lett. A 4 (1989) 2767.
[4] C.G. Callan, J.M. Maldacena, Brane dynamics from the BornInfeld action, Nucl. Phys. B 513
(1998) 198, hep-th/9708147.
[5] D.K. Park, S. Tamaryan, H.J.W. Mller-Kirsten, Jian-zu Zhang, D-branes and their absorptivity
in BornInfeld theory, hep-th/0005165.
[6] P.F. Byrd, M.D. Friedman, Handbook of Elliptic Integrals for Engineers and Scientists,
Springer, 1971.
[7] D.A. Gorokhov, G. Blatter, Phys. Rev. B 56 (1997) 3130;
D.A. Gorokhov, G. Blatter, Phys. Rev. B 57 (1998) 3586.
[8] H.J.W. Mller-Kirsten, D.K. Park, J.M.S. Rana, Phys. Rev. B 60 (1999) 6662.
[9] D.K. Park, H.J.W. Mller-Kirsten, J.-Q. Liang, Nucl. Phys. B 578 (2000) 728.
[10] J.-Q. Liang, H.J.W. Mller-Kirsten, D.K. Park, A.V. Shurgaia, Phys. Lett. B 483 (2000) 225.
[11] J.-Q. Liang, H.J.W. Mller-Kirsten, D.K. Park, F. Zimmerschied, Phys. Rev. Lett. 81 (1998)
216.

Precision prediction of gauge couplings and

the profile of a string theory
Dumitru M. Ghilencea a,b , Graham G. Ross a
a Department of Physics, Theoretical Physics, University of Oxford, 1 Keble Road,
Oxford OX1 3NP, United Kingdom

b Physikalisches Institut der Universitat Bonn, Nussallee 12, 53115 Bonn, Germany
Abstract
We estimate the significance of the prediction for the gauge couplings in the MSSM with an
underlying unification. The correlation between the couplings covers only (0.22)% of the a priori
reasonable region of the parameter space, while the prediction for sin2 W is accurate to 1.3%. Given
that agreement with experiment to such precision is unlikely to be fortuitous, we discuss the profile
of a string theory capable of preserving this level of accuracy. We argue that models with a low
scale of unification involving power law running in the gauge couplings do not. Even theories with a
high scale of unification are strongly constrained, requiring the compactification scale of new space
dimensions in which states transforming under the Standard Model propagate to be very close to
the string cut-off scale. As a result no new space dimensions can be larger than 1014 fm. 2001
Elsevier Science B.V. All rights reserved.
1. Introduction
The unification of gauge couplings is one of the few believable precision predictions
coming from physics beyond the Standard Model. In the Minimal Supersymmetric
Standard Model the radiative corrections to the gauge couplings coming from the states
of the MSSM causes the SU(3), SU(2) and U(1) gauge couplings to become very nearly
equal at the scale (1 3) 1016 GeV, provided one adopts the SU(5) normalisation of
the U(1) factor. It is usual to use gauge unification to predict the value of the strong
coupling given the weak and electromagnetic couplings. Doing this one finds 3 (MZ ) =
0.126 0.01 to be compared with the experimental value [1] 3 (MZ ) = 0.118 0.002.
However, the gauge unification prediction is really much more precise than this comparison
suggests. In the MSSM the precision is limited by threshold effects associated with the
E-mail address: dumitru@th.physik.uni-bonn.de (D.M. Ghilencea).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 2 3 4 - 6
102
D.M. Ghilencea, G.G. Ross / Nuclear Physics B 606 (2001) 101118
masses of the supersymmetry partners of Standard Model states affecting the prediction
for 3 (MZ ). In Section 2 we discuss these corrections and demonstrate that, expressed
as a correlation between the gauge couplings, the prediction covers only (0.22)% of the
overall (perturbative) parameter space. Moreover, the prediction for sin2 W is much more
precise than for 3 , being accurate to 1.3%. The fact that experiment and theory agree to
this precision seems to us unlikely to be just a happy accident. It is largely due to this
circumstantial evidence for unification and the fact that supersymmetry is needed to solve
the mass hierarchy problem that there has been so much interest in supersymmetric models
and in compactification schemes that preserve a stage of low energy supersymmetry. For
the same reason it is of interest to demand that the physics beyond the MSSM should not
spoil the prediction.
There are inevitably further threshold corrections coming from new states lying at
and above the unification scale. In the case of Grand Unification the states include the
heavy gauge bosons and Higgs bosons needed to make up complete GUT multiplets. The
threshold corrections associated with these states have been extensively discussed in the
context of Grand Unification [2] and the condition that the gauge unification relations
should not be significantly affected requires there should be no substantial splitting
between the heavy Higgs and gauge boson states. Provided the ratio between these masses
is less than a factor of 10 the precision is essentially unchanged in minimal GUT schemes,
although the requirement of near equality of Higgs and gauge boson masses becomes much
stronger for the case of non-minimal Higgs sectors responsible for breaking the GUT.
While at one loop order the unification scale is well defined in a SUSY GUT, at two loop
order the scale of unification is not so easy to define. We discuss this point in Section 4
and show that it is necessary to regulate the theory to determine these two loop corrections.
In a string theory this regularisation introduces new threshold corrections (associated with
the canonical normalisation of the Khler term for matter fields 1 ) and a sensitivity to the
string (cut-off) scale. As a result there are new contributions to the gauge couplings in
Grand Unified theories proportional to ln(Ms /MGUT ) with the string scale, Ms providing
the cut-off scale, which are not usually included in GUT calculations.
In the case of compactified string theories (with or without Grand Unification below
the string scale) the threshold corrections can be much larger if the compactification
scale is (significantly) different from the string scale. This may be understood (in the
effective low energy field theory) as due to KaluzaKlein states associated with gauge
non singlets propagating in the extra dimensions and with mass up to the string scale [4]
which contribute to the running of the gauge couplings. For low compactification scales
the number of such states is very large leading to very large threshold corrections
generating power-like gauge coupling running and significant threshold sensitivity of the
predictions [5]. String calculations of the threshold effects for the weakly coupled heterotic
string have been performed by several groups (for a review of the topic and references
see [6]). The regularisation of the effective field theory is provided at the (heterotic)
1 Additional corrections due to the rescaling of the metric (super-Weyl anomaly) exist, with similar implications [3].
103
string level by the geometry of the string 2 In type I models with compact in-brane
dimensions [7] one may have a similar (power-like) behaviour due to the associated
KaluzaKlein states. Finally, power-like behaviour may also be present in type I models
with anisotropic compactification. The string thresholds have been computed in [8] and
(string) regularisation is again provided by the geometry of the string. 3 As we shall discuss
in Section 3, the expectation is that all these corrections will spoil the MSSM prediction
and imply the compactification scale(s) should be very close or equal to the string scale.
One important implication of this is that one cannot appeal to non-standard power law
running to lower the unification scale and one is left only with high scale unification.
The need for a high unification (and string) scale is broadly in agreement with the
expectation in the heterotic string. However, in the weakly coupled case there is a
residual discrepancy of approximately a factor of 20 between the gauge unification scale
found in MSSM unification and the string prediction. Further, the requirement that the
compactification scale should also be close to the string scale apparently prevents large
threshold effects from coming to the rescue. However, as we discuss in Section 3.2,
in cases of string thresholds with a Wilson line background, it is possible for large
Wilson line effects to lower the unification scale to that of the MSSM while keeping the
compactification scale close to the string scale. Thus, the weakly coupled heterotic string
with Wilson line background provides an example of a string model which can preserve the
MSSM precision prediction and have an acceptable scale of unification. Another possibility
is that the string theory has a stage of Grand Unification below the compactification scale
allowing for a gauge unification scale below the compactification scale. As we discuss in
Section 4, even in this case, it is necessary that the compactification scale should be close
to the string cut-off scale due to the need to keep two-loop corrections under control. An
alternative resolution of the discrepancy is that gravity can propagate in more dimensions
than matter and gauge states, changing the string prediction for the unification scale. The
original idea for this came from the strongly coupled heterotic string in which the gauge
unification scale is lowered through the appearance of an 11th dimension [9]. Even in this
case the compactification scale of the other compactified dimensions cannot be far from
the string scale. Our conclusions are presented in Section 5 where we present a profile of a
compactified string theory capable of satisfying all the constraints needed to preserve the
accuracy of the predictions for gauge couplings.
2. Precision gauge unification in the MSSM

In the MSSM the prediction of the gauge couplings follows from the GUT or string
relation at the unification scale corrected by the renormalisation group determination of
radiative corrections involving the MSSM states. By combining the standard model gauge
couplings one may eliminate the one-loop dependence on the unification scale and the
2 I.e., the world-sheet torus [4].
3 Via cancellation of the quadratic terms between the annulus and Moebius strip contributions [8].
104
value of the unified coupling to obtain a relation which depends only on the threshold
effects associated with the unknown masses of the supersymmetric states.

15 2
3 1
1 19 Teff
31 (MZ ) =
(1)
sin W (MZ )
(MZ ) +
ln
+ two-loop.
7
7 em
2 14 MZ
Note that the leading dependence on the SUSY thresholds comes from Teff , an effective
supersymmetric scale, while two loop corrections, which have a milder dependence on the
scale, through (gauge and matter) wavefunction renormalisation ln Z ln i (scale) can be
ignored to a first approximation. In terms of the individual SUSY masses Teff is given
by [10]

28/19

4/19

ml 9/19
mW
mH 3/19 mW

Teff = mH
(2)
mg
mH
mH
mq
provided the particles have mass above MZ . The precision of the relation, Eq. (1), between
Standard Model couplings is limited by the dependence on Teff . However the uncertainty
in Teff is limited by the fact that the supersymmetry masses are bounded from below
because no supersymmetric states have been observed and from above by the requirement
that supersymmetry solve the hierarchy problem. From Eq. (1) we see the dependence on
the squark, slepton and heavy Higgs masses is very small, the main sensitivity being to the
Higgsino, Wino and gluino masses.
If all the supersymmetric partner masses are of O(1 TeV) then so is Teff . However in
most schemes of supersymmetry breaking radiative corrections split the superpartners. In
the case of gravity mediated supersymmetry breaking with the assumption of universal
scalar and gaugino masses at the Planck scale the relations between the masses imply Teff
mH(2 (MZ )/3 (MZ ))2 ||/12. For of order the weak scale Teff is approximately
20 GeV. From this we see the uncertainty in the SUSY breaking mechanism corresponds to
a wide range in Teff . In what follows we shall take 20 GeV < Teff < 1 TeV as a reasonable
estimate of this uncertainty. Using this one may determine the uncertainty in the strong
coupling for given values of the weak and electromagnetic couplings using Eq. (1). The
result (including two-loop effects) as a function of 3 and sin2 W is plotted in Fig. 1(a)
and (b). 4
The precision of the prediction is remarkable. A measure of this is given by the area
between the two curves in Fig. 1(a). If one assumes that a random model not constrained by
unification may give any value for 3 and sin2 W between 0 and 1 the relative precision is
an impressive 0.002! Of course, this estimate is sensitive to the measure chosen. Changing
to 3 and sin W makes very little difference. Changing to 31 and sin2 W (and restricting
the possible range of 31 to be 1 < 31 < 10) increases the relative precision by a factor
of 10. Using this variation as an estimate of the uncertainty associated with the measure we
conclude that a reasonable estimate for the precision of the prediction for the correlation
between the Standard Model couplings is in the range (0.22)%. One may also use the
result of Eq. (2) to predict one of the couplings given the other. However, as may be
4 Further constraints on this region, coming from small fine-tuning of other observables of the MSSM [11]
may further constrain the area of allowed values of 3 (MZ ) vs. sin2 W .
105
(a)
(b)
Fig. 1. Plots of 3 (MZ ) versus sin2 W calculated in the MSSM for two values of the effective
supersymmetric threshold, Teff = 20 GeV and Teff = 1000 GeV. The limits correspond to requiring
3 (MZ ) remains in the perturbative domain and unification occurring above the supersymmetry
threshold. The area between the two curves provides a measure of the predictivity of the theory. The
experimental range of values is also shown.
seen from Fig. 1(a), the prediction is not equally precise for each coupling. A quantitative
estimate of the accuracy of the prediction for 3 (MZ ) or sin2 W (MZ ) may be obtained
from Eq. (1) by taking the derivative with respect to Teff giving
31 (MZ ) F3 +
15 2
19
1
sin W (MZ )em
+ two-loop,
(MZ )FW
7
28
(3)
106
where

d ln(3 (MZ )
,
F3 =

d ln T
eff

d ln(sin2 W )
.
FW =

d ln T
(4)
eff
The quantities F3 and FW give the fractional threshold sensitivity of 3 and sin2 W to a
change in ln Teff . From this we see that the unification prediction for sin2 W (MZ ) is more
accurate than that for 3 (MZ ) since, close to the experimental point, we have from Eqs. (3)
and (4)
1
FW
.
F3
7.6
(5)
Given that the error in 3 is 10% for a change in Teff from 20 to 1000 GeV we see
from this equation that the corresponding error in sin2 W is 1.3%. Thus the prediction for
sin2 W provides a more realistic measure of the precision of the unification prediction. An
expansion of sin2 W (corresponding to 3 (Mz ) = 0.119) thus gives

sin2 W (MZ ) 0.2337 0.25 3 (MZ ) 0.119 0.0015.
(6)
This effect may be seen directly in Fig. 1(a) since the curve is more steeply varying in the
3 direction than in the sin2 W direction close to the experimental point. The optimal
combination of 3 and sin2 W with minimal uncertainty normal to the curve can be
determined numerically but is relatively close to sin2 W . The fact that experiment and
theory agree to this level of accuracy is impressive and is a major reason why so much
attention has been paid to supersymmetric extensions of the Standard Model.
3. High-scale threshold sensitivity of the unification prediction

Given this impressive accuracy it is clearly of interest to determine the effect on this
prediction coming from threshold effects at the unification scale in realistic string theories.
Our hope is that the requirement that the precision should not be significantly degraded will
give us information about the underlying string theory. The value of the gauge couplings
at MZ is of the form
bi
bi
(, 0 ) +
ln
2
2 MZ

() 1/3 Ti (R )
3Ti (G)
ln
ln Z (, MZ ),
+
2
i (MZ )
2
i1 (MZ ) = i + 1 () +
(7)
where is the string threshold correction corresponding to the particular string theory
considered (hereafter renamed to H , I to stand for the weakly coupled heterotic and
type I string cases, respectively). The parameter is the unification scale, () is the
unified coupling, 0 is the mass of the heavy states, often the compactification scale
in string theories and Z and (()/(MZ )) are the matter and gauge wavefunction
renormalisation coefficients respectively. In Eq. (7) there are additional effects due to low
107

energy supersymmetric thresholds i and 5 bi = 3Ti (G) + Ti (R ) are the one-loop
beta function coefficients, while bi depend on the particular compactification scheme
they are non-zero only for the N = 2 massive SUSY states and vanish for the N = 4
spectrum.
In comparing the threshold effects at the unification scale to the SUSY threshold effects
discussed above we will restrict ourselves to a measure of the sensitivity of the prediction
to these scales for either the strong coupling or sin2 W while the other is maintained fixed.
In leading order we have from Eq. (7)

15
1 b23 b31
1
2
3 (MZ )
(8)
sin W =
+ b2
+ b3 (, 0 ),
b1
7
2
b12
b12
where bij = bi bj , b1 = 33/5, b2 = 1, b3 = 3. The relative sensitivity of 3 (keeping
sin2 W fixed) to changes in 0 and Teff is, from Eqs. (9), (4) and (3), given by

(ln(3 (MZ ))) d ln 3 (MZ )

F 1

=
R
(9)
3
(ln(3 (MZ )))MSSM d ln 0

d
14 5
12
,
=
(10)
b1
b2 + b3
19 7
7
d(ln 0 )
where we have assumed that the predicted value for 3 (MZ ) in the model considered
is close to that of the MSSM. 6 One obtains the same result for the relative threshold
sensitivity if we compute the prediction for sin2 W (MZ ) (with 3 (MZ ) fixed) normalised
to FW . For this reason in our analysis of threshold sensitivity we will focus our attention
only on the value of R.
It follows immediately from Eq. (9) that models involving only complete additional GUT
representations (i.e., with b i = bi + n) do not introduce any threshold sensitivity at one
loop order. This will be the case if there is a Grand Unification below the compactification
scale. It is also true if one can find models for which bi = bi as has been suggested [12] to
lower the unification scale by power law running. In both these cases, however, there will
be significant threshold dependence at two loop order and we will discuss this separately
in Section 4.
Returning to the one-loop threshold effects of Eq. (9), we will determine the sensitivity
for the case of weakly coupled heterotic string with/without Wilson lines and type I/I
models to analyse their relative threshold sensitivity, R. In what follows , o stand for
the string and compactification scale respectively. To perform this analysis we use the
explicit string computations for the string thresholds [8,13,18].
3.1. Weakly coupled heterotic models
We first consider the case of the weakly coupled heterotic string with an N = 2 sector
(without Wilson lines present) in which six of the dimensions are compactified on an
5 The definition (2) of T
eff in terms of i is ln Teff /Mz = (28/19)(1 b23 /b12 + 2 b31 /b12 + 3 ), with
bij = bi bj .
6 For a model to be viable one must in first instance predict the right value for (M ) and only after would
Z
3
the question of threshold sensitivity be relevant.
108
orbifold, T 6 /G. In such models the spectrum splits into N = 1, N = 2 and N = 4 sectors,
the latter two associated with a T 2 T 4 split of the T 6 torus. Due to the supersymmetric
non-renormalisation theorem, the N = 4 sector does not contribute to the running of the
holomorphic couplings. As we shall discuss in Section 4 even in the case of N = 4,
two-loop corrections to the effective gauge couplings may introduce substantial threshold
corrections. The N = 1 sector gives the usual running associated with light states but does
not contain any moduli dependence. The latter comes entirely from the N = 2 sector. For
the heterotic string all states are closed string states and at one loop the string world sheet
has the topology of the torus T 2 . For the case of a six-dimensional supersymmetric string
vacuum compactified on a two torus T 2 the string corrections take the form [13,14]. Here
T R1 R2 and U R1 /R2 where T , U are moduli and R1 , R2 are the radii associated
with T 2 . We consider the case of a two torus T 2 with T = iT2 (the subscript 2 denotes
the imaginary part) and U = iU2 . Making the dimensions explicit T2 should be replaced
by T2 T2o /(2 ) 2R 2 /(2 ). Performing a summation over momentum and winding
modes and integrating over the fundamental domain of the torus gives the following result
for H [13]

4
4
8e1E
1
U2 T2 (iU2 ) (iT2 )
H = ln
(11)
2
3 3

4 Ms 2 i3 3 Ms 2 4
1
2

= ln 4 (i)
(12)

2e1
2
with the choice U2 = 1 in the last equation. Ms is the string scale, o 1/R, (x) is
the Dedekind eta function and we have replaced in terms 7 of Ms . For large 8 T2 the
eta function is dominated by the leading exponential and so one finds the power law
behaviour T2 R 2 which has a straightforward interpretation as being due to the
decompactification associated with T 2 . This contribution is basically due to the tower of
KaluzaKlein states below the string scale and can be understood at the effective field
theory level [4] whose result is regularised by the string world-sheet [4]. The presence of
power-like running, although taking place over a very small region of energy range, 9 can
significantly affect the sensitivity of the unification prediction for 3 (MZ ) with respect
to changes of the compactification scale (which, being determined by a moduli vev, is
not fixed perturbatively and hence is not presently known). To demonstrate this explicitly,
consider the variation of the threshold H with respect to T2 :

d ln |(iT2 )|
H
1
.
(13)
= 1+4
ln T2
2
d(ln T2 )
This gives the following relative threshold sensitivity (Eq. (9)) with respect to the
compactification scale 0 (x 0 /Ms ):
7 From [14] M = 2e(1E )/2 33/4 / 2 , where g is the string coupling at the unification.
s
s
8 Note that T 5.5(M R)2 in DR scheme, so one can easily have T 20 while R is still close to the string
s
2
2
length scale, to preserve the weakly coupled regime of the heterotic string. In this section large R corresponds
to values of T2 in the above range.
9 This takes place essentially between the compactification and the string scale [4].
109

14 5
12
d
2
b
b
ln
i
x
+
b
(14)
1
2x
1
2
3
,
19 7
7
dx
where = 3 3eE 1 /2. In Fig. 2 we plot the second factor in curly brackets (which
is equal to H /(ln x)). 10 To understand the importance of H and the implications
it has for the threshold sensitivity of 3 (MZ ), we note that H plays a central role in
the attempts to bridge the well-known gap (of a factor of 20) [6,12] between the
heterotic string scale and the MSSM unification scale. It has been found (for a review
see [6]) that for generic examples (Z8 orbifold) this condition requires 8.5 [6,16]
which in turn requires moduli of considerable size, T2 18.7 corresponding to a small
value of x 0.53. 11 As may be seen from Fig. 2 this leads to an even larger value for
|dH /d(ln x)| and an enhanced relative threshold sensitivity (see Fig. 3). For the Z8
orbifold [17], we find (using ba = (9/2, 5/2, 3/2)) R = 5.9 for x 0.53. Since R
gives the relative sensitivity of the gauge coupling predictions to the N = 2 threshold and
the SUSY threshold we see that even for x = 0.5 the precision of the gauge coupling
prediction is substantially reduced. The requirement that this precision should not be lost
constrains x 1 (i.e., 0 = Mstring). However, this fails to bridge the gap.
This example clearly illustrates the general problem that weakly coupled heterotic string
models have to reconcile two conflicting constraints, namely the need for a large H to
solve the scale mismatch between the MSSM unification scale and the heterotic string
scale and the need for a small derivative of H to avoid a large threshold sensitivity of the
R=
Fig. 2. The values of the derivative dH /d(ln 0 ) = 1 2x ln x k(i3 3 eE 1 /(2x 2 )) and that
of H for x close to 1.
10 This situation is somewhat similar to other (effective) models [5] where it has been shown that power-like
running brings in a significant amount of threshold sensitivity.
11 This values for T can still be on the edge of the weakly coupled regime of the 10D heterotic string as its
2
coupling is equal to = ( i Re Ti /Re S)1/2 ).
110
Fig. 3. The values of 3 (MZ ) for the model with gauge bosons in the N = 2 sector with a change
of x from 1 to 1/2, leading to an area (between dashed lines) of values available to 3 (MZ ) larger
by a factor of 5 compared to the MSSM case (continuous lines).
gauge couplings. The latter constraint is introduced by the power-law dependence 12 of the
thresholds on the compactification scale, and even though such running takes place over a
small energy range it still gives significant effects. The problem can be avoided if x 1,
corresponding to the compactification scale being very close to the string scale. On the
other hand the former constraint seems to require a small value for x. As we shall discuss
this conclusion may be evaded in theories with Wilson line breaking.
These conclusions apply to string theories with an N = 2 sector. It is possible to
construct string theories in which this sector is absent in which case there is no significant
one-loop sensitivity of the unification prediction for the gauge couplings to changes of
the high scale/moduli fields. The downside is that then one does not have the large H
needed to close the gap. Even if this problem is solved in another way the N = 4
sector necessarily present introduces strong threshold dependence at two loop order which
provides almost as restrictive a bound on the compactification scale. We shall discuss this
further in Section 4.
3.2. Weakly coupled heterotic models with Wilson lines
In the case that there are Wilson lines characterised by the moduli B the masses of the
heavy states contributing to the RG flow of the gauge couplings become B-dependent.
For this class of models the string calculation of the threshold contribution to the gauge
couplings was computed in [18]. The presence of the Wilson lines may or may not
12 This is due almost entirely to the presence of the towers of KaluzaKlein modes rather than to winding
modes [4].
111
break the symmetry group. In the latter case the threshold corrections do not seem to be
large enough [6,16] to account for the energy gap between the MSSM unification scale
and that of the weakly coupled heterotic string. The case when Wilson lines break the
symmetry gauge group is more interesting, as one can obtain a symmetry group close to
that of the standard model [19]. We will consider only the class of models with (0, 2)
compactifications which have the additional advantage of avoiding the doublet-triplet
splitting problem. For the case of the Z8 orbifold the threshold contribution is then [18]
(see also [6])
4

10

1
1

k ()
H = ln Y 10

128
20
(15)
k=1
with k the ten even, genus two theta functions. As has been shown in [6,15,16], in this
case one can obtain a large value of the threshold H which may help solve the mismatch
between the MSSM unification scale and that of the heterotic string theory, without the
need of large moduli. For example, one obtains, 13 H 8.5 for T2 = U2 = 4.5 and B =
1/2. The derivative of H with respect to T2 can easily be taken by using the expansion
of (15) [18] up to O(B 4 ), to give

1
24 d ln |(iT2 )|
H
4U2 T2
=
+
ln T2
2 4U2 T2 B 2
5
d ln T2

2
2
2
2
12 B T2 dT2 ln (iT2 )dU2 ln (iU2 )
5 1 6B 2 dT2 ln 2 (iT2 )dU2 ln 2 (iU2 )

24 d ln |(iT2 )|
1
,
1+
2
5
d ln T2
(16)
(17)
where dT2 stands for a derivative with respect to T2 and where the approximation used
above holds for the particular point in the moduli space giving the right size of the
threshold H . One may observe that this expression is very close to that of (13), with
the difference that one has smaller moduli than in the case without Wilson lines present.
For the Z8 orbifold at the point in moduli space given above we obtain R = 1.50 which
is lower than in the case without Wilson lines by a factor of 4 for the same value of the
threshold H .
The overall conclusion one may draw from this example is that the Wilson line
background can give a threshold H which is large enough to obtain the correct unification
scale with the additional benefit that its derivative with respect to T2 is not significantly
affected, leading to a significantly lower threshold sensitivity than in the other cases
without Wilson lines. Such models can preserve the precision prediction for gauge coupling
and give good agreement between the string unification scale and the MSSM value.
13 With our normalisation for I .
112
3.3. Type I/I models with a N = 2 sector

In this case the corrections to the gauge couplings come from the N = 1 (massless)
sector and from the N = 2 massive winding (type I ) sector. The massive N = 2 threshold
corrections to i1 at the string scale MI were computed in [8] giving

4

1
bi,k ln Gk Im Uk MI2 (Uk ) .
I =
(18)
2
k
Here Gk is the metric on the torus 14 T k , and G1 = R1 R2 (for a rectangular torus). The
behaviour of the thresholds when Im U = R1 /R2 is of order one is logarithmic, a
ln(R1 R2 ) and when Im U = R1 /R2 1 it is power-like, a R1 /R2 .
The interest in this class of models was initially due to the observation that, in contrast
to the heterotic case, the I s have logarithmic running which, even for a low string scale,
may continue up to the Planck scale [21] (more precisely, up to the first winding mode
close to this scale). Despite the absence of power-like terms in (18), such models still
have fine-tuning problems and this has been extensively discussed in [20]. The origin
of the fine tuning is the difficulty in keeping light the closed string state with vacuum
quantum numbers (dual to the first winding mode) which provides the cut-off for the
gauge coupling running. Its mass is MI2 /MP and to obtain running to the MSSM gauge
unification scale MI should be large (> 1010 GeV).
In the following we will determine the relative precision factor R, induced by the
structure of the thresholds (18), assuming that the linear combination of beta functions
entering the definition of R is of order O(1). The case this is not true is discussed in
Section 4. The variation of the thresholds with respect to r1,2 = MI R1,2 where R1,2 are the
compactification scale(s) gives the relative fine tuning measure
R
dI
1
d
= 2
ln (i),
d(ln r1 )
2
d
(19)
where = r1 /r2 . Quantitatively R 10 for r1 /r2 < 1/20 or r1 /r2 > 20. Since
this significantly degrades the precision we are led to the conclusion that anisotropic
compactifications are severely limited for reasonable values of the beta coefficients.
Models with large radii of compactification are apparently still allowed as it is possible
to have r1 and r2 very large while keeping
= r1 /r2 O(1). However, they suffer from
an even more severe problem in that the N = 2 beta functions have to be proportional to
the N = 1 beta functions if the MSSM predictions are to be preserved [20]. We should
stress that this sector does not correspond to KaluzaKlein excitations of the Standard
Model states and so to obtain negative beta functions it is necessary that it involves a new
massive gauge sector which also carries Standard Model gauge quantum numbers. Even
with this it is very difficult to construct a spectrum giving beta functions proportional to the
Standard Model beta functions because a N = 2 copy of the MSSM representations gives a
contribution to the beta functions which is not proportional to the N = 1 contribution [20].
14 We consider in the following only compactification on a single torus T 1 .
113
Even more implausible is the possibility that two-loop corrections should be the same as
in the MSSM. Thus in this case the good prediction of the MSSM must be viewed as a
complete accident as the running above the string scale comes from a spectrum completely
unrelated to that of the MSSM.
3.4. Type I/I models without a N = 2 sector
For models of this class we address a generic type I string inspired model [22] which
shares some similarities with the MSSM because it only has logarithmic running, but the
spectrum of the light states and the hypercharge normalisation are different. Somewhat
unexpectedly, we will show that logarithmic RG evolution is not always a sufficient
condition for a low threshold sensitivity of 3 (MZ ) prediction. This is so because the
definition of R depends not only on the scale, but also on the beta function coefficients
which in the model to discuss will play a more important role following the non-standard
hypercharge normalisation.
The model is derived from D = 4, N = 1 compact type IIB orientifolds with Dp branes
and anti-Dp branes located at different points of the underlying orbifold [22]. It has gravity
mediated supersymmetry breaking, full Standard Model gauge group, has no N = 2 sector
and the UV cut-off scale for the N = 1 sector is the string scale. The predictions of the
model for the gauge couplings unification were discussed at one loop order in [22] and at
two loop level in [20]. It involves logarithmic unification at a low value of the string scale
( 1012 GeV). Here we address the issue of the relative sensitivity R of the prediction
for 3 (MZ ) (or sin2 W ) to the heavy thresholds compared to the SUSY thresholds. The
reason a low unification scale is possible in such models is because the (string) unification
relation between gauge couplings is not that given by SU(5) due to a different hypercharge
normalisation. To compensate for this it is necessary to extend the MSSM light spectrum to
include five further pairs of Higgs doublets (of mass m
) and three vector-like right handed
colour triplets (d + d c ). When the latter have a mass equal to m
as well, the full two loop
RGE equations derived from (7) are given by [20]

Ms
Ms
bi
bi
1
1
ln
ln
+
i (MZ ) = i + g +
2
MZ
2
m

3
3
g
g
1
1 bij
+
Yij ln
ln
+
,
4
j (
m)
4
bj
j (MZ )
j =1
(20)
j =1
where Yij = bj /(bj bj )(2bj Tj (G)ij bij ), with Tj (G) = {0, 2, 3}j ; bij is the two
loop beta function as in the MSSM but with 3/11 hypercharge normalisation, bj =
{11, 1, 3}j , ( = 3/11), bj = bj + bj , with bj = {7, 5, 3} to account for the
additional five Higgs pairs and three pairs of right handed colour triplets.
The value of R is easily computed at one loop giving
R=
14
(2b1 3b2 + b3 ) 6,
19
(21)
114
where the definition of R (similar to that of (9) with 0 = m

, Ms fixed) takes into account
the different hypercharge normalisation of the model. This value for R implies a threshold
sensitivity with respect to Ms /
m significantly greater than that of the MSSM with respect
to Teff . This may also be seen in the full two loop analysis which is used in Fig. 4(a) and
m by a factor of 4, to match the
shows (at a global level) that a change of the ratio Ms /
uncertainty 15 in Teff , gives a much larger variation than that found due to SUSY threshold
uncertainty. The ratio of the two areas is 6. The problem is even worse if one relaxes the
assumption of equal bare masses for the five Higgs pairs and colour triplets.
This example illustrates the problems associated with any model which seeks to reduce
the unification scale by going to a non-SU(5) normalisation for the weak hypercharge.
While the increased threshold sensitivity and the associated loss of precision in the
prediction for gauge couplings is bad enough we think an even more significant indictment
of the scheme is that such models require that the precision agreement of the MSSM
unification prediction with experiment is just a celestial joke!
4. Two-loop limits on the compactification scale

So far we have discussed the one-loop threshold corrections coming from Eq. (7). We
now turn to the case the linear combination
b23 b31
+ b2
+ b3
f (bi , bi ) = b1
(22)
b12
b12
is zero and the one-loop corrections to the correlation between 3 and sin2 W vanish. In
such a case the heavy threshold correction only comes from two loops (and beyond).
The two loop corrections come from the coefficients Z (, Q) in Eq. (7). These are
the one-loop matter wavefunction renormalisation coefficients above the scale Q. It is
necessary to regulate the theory in order to determine the Z (, Q). In the case of a
GUT (effective) theory it is normally assumed that they are unity at the GUT scale (i.e.,
is chosen as the GUT scale). However, this is not the correct prescription if the GUT
emerges from string compactification for the string regularisation requires that be the
string cut-off scale, usually the string scale.
The coefficients Z have two contributions. One is due to the usual (gauge and Yukawa)
corrections from the N = 1 MSSM-like massless states which generate the Z (, Q)
at one loop. These lead to corrections in the gauge couplings at O( 2 /(4)). The second
contribution to Z (, Q) comes (for Q > 0 ) from towers of (N = 2, 4) KaluzaKlein
states. Such corrections of order O(N 2 /4) and will give R N/(4). The constraint
that R should not be large requires N < 4/ (equivalent to the requirement that the
theory remain perturbative). While less constraining than the one-loop constraint, this is
still very strong due to the rapid increase in the number of KaluzaKlein states, N
(/0 ) , being the number of extra dimensions. For = 2, < 100 while for =
6, < 2.50 . Thus, even if one can avoid the large threshold sensitivity (power-like )
15 Corresponding to a change of T between 250 to 1000 GeV.
eff
115
(a)
(b)
Fig. 4. (a) The values of 3 (MZ ) in function of sin2 W . The widely separated curves (dashed line)
correspond to the model investigated where the bare mass of the doublets and triplets (considered
degenerate) is changed by a factor of 4. The area between these two curves is far larger than that of
the MSSM (between the continuous lines) for a similar ratio (of 4) of the supersymmetric effective
scales (250 < Teff < 1000 GeV), meaning a less predictive power than in the MSSM case. (b) The
values of 3 (MZ ) in function of sin2 W . The dashed lines correspond to the present model when
the bare mass of the triplets 4
m meaning a less predictive power than in the previous case or than
in the MSSM (area between the continuous curves) induced by an equal factor of change of the
supersymmetric threshold from 250 to 1000 GeV. The slopes of the curves for the string model are
different from those for the MSSM case due to the different hypercharge assignment.
116
induced at one loop level by the (heterotic) string thresholds (or at the effective field
theory level by towers of KaluzaKlein states) through an (string) construction which
sets f (bi , bi ) = 0, the constraint on two-loop contributions still requires the cut-off (string
scale) and the compactification scale 0 should be very close. In particular string models
with a Grand Unified group unbroken below the compactification scale should still have
the compactification scale close to the string scale.
5. Summary: a profile of a string model

In this paper we have considered two issues. The first is the precision of the prediction
in the MSSM for low-energy gauge couplings assuming an underlying unification at a
high scale. Taking account of the uncertainties due to the unknown masses of the SUSY
partners of Standard Model states we found that the predicted range for the gauge couplings
compared to the a priori range of possible couplings is remarkably precise, between 2
and 0.2%. This gives a quantitative estimate of how significant is the gauge unification
prediction. The prediction for sin2 W itself is also impressive with an accuracy of 1.3%.
This remarkable precision led us to consider the nature of an underlying (string) theory
that can maintain this accuracy. Due to the non-decoupling of the contribution of massive
states to renormalisable terms in the low energy effective Lagrangian, the requirement that
the gauge predictions be undisturbed places strong constraints on the massive sector. We
determined the contribution of states transforming non-trivially under the Standard Model
gauge group in compactified string models. Requiring that these contributions leave the
MSSM predictions intact leads to a constraint on the magnitude of the compactification
radius associated with the propagation of such states to limit the number of states in the
KaluzaKlein tower. We found that for a wide class of string theories the contribution of
heavy states occurs at one-loop order and the radius should be very close to the inverse of
the string scale (well within a factor of 2). In the cases the one-loop contributions vanishes
the limit is only slightly relaxed.
The implication of this result is quite far-reaching. One immediate one is that unification
at a low scale through power law running of the gauge couplings cannot maintain the
precision of the MSSM predictions. Indeed, due to the different contributions to the beta
functions of massive compared to massless SUSY representations, low scale unification
requires a very different multiplet content from the MSSM in order to obtain the same
gauge unification prediction. As a result the precision prediction for gauge couplings must
be considered a fortuitous accident, something we find hard to accept given the remarkable
precision of the prediction. Even if we do accept this, and the N = 2 sector happens to give
the same beta function as the N = 1 MSSM spectrum, power law running introduces a
strong dependence on the heavy thresholds so that the precision of the prediction is lost.
For both these reasons we consider a low scale of unification due to power law running
to be unlikely. Models with a low scale of unification due to a non-SU(5) hypercharge
normalisation do not require power law running. Nonetheless it turns out that they too
117
have enhanced threshold dependence due to the need for additional light states and again
this loses the predictive power of the MSSM.
The profile of our string model which preserves the precision prediction for gauge
couplings found in the MSSM therefore requires a large scale of unification with an SU(5)
normalisation for the weak hypercharge. Even so, there is still a strong constraint on the
compactification scale because the cutoff of the contribution of heavy states is at the
string scale or higher. To avoid the same power law running corrections that degraded
the predictions in the case of low scale unification, the radius of compactification of those
dimensions in which Standard Model gauge non-singlet fields propagate must not be large
compared to the cutoff radius. This means the compactified string theory lies far from the
CalabiYau limit and close to the superconformal limit. This is quite attractive in that many
aspects of the effective low-energy field theory, such as Yukawa couplings, are amenable
to calculation in the superconformal limit.
The need for a high scale of unification broadly fits the expectation in the (weakly
coupled) heterotic string. However in detail there is a discrepancy between the MSSM
value for the unification scale of 3 1016 GeV and the prediction in the weakly
coupled heterotic string, approximately a factor of 20 larger. Three explanations have been
suggested.
The first is that heavy threshold effects raises the unification scale to the predicted value.
The requirement that the precision of the gauge coupling prediction be maintained severely
limits the magnitude of these threshold effects and precludes explanations requiring large
radii. However, in models with Wilson line breaking it is possible to have very large
threshold corrections to the unification scale while keeping the threshold contribution
to gauge couplings small. Given that such Wilson line breaking is very often needed to
break the underlying gauge symmetry of the heterotic string, this explanation seems very
reasonable.
A second possibility is to have a GUT below the string compactification scale, the GUT
breaking scale being the unification scale. Even in this case it is still necessary for the
compactification scale to be very close to the string scale to keep the two loop contributions
from the KaluzaKlein states small. If this theory comes from the weakly coupled heterotic
string the compactification and string scales must be close to the Planck scale.
Of course our analysis does not preclude the existence of large new dimensions not
associated with the propagation of Standard Model states. A particular example is the
strongly coupled heterotic string in which gravity but not the Standard Model states
propagate in the eleventh dimension. For the case the eleventh dimension is three orders
of magnitude larger than the string length, corresponding to a compactification radius of
O(1014 fm), the gauge unification scale may be reduced to that found in the MSSM. This
provides a third way to reconcile the predicted unification scale with that needed in the
MSSM. However the size of the extra dimensions is still severely constrained by the need
to have a high unification scale, the minimum occurring for just one additional dimension
as in the strongly coupled heterotic case. Thus even in the case of large new dimensions in
which Standard Model states do not propagate, the new dimension cannot be larger than
1014 fm.
118
Acknowledgements
The work of D.G. was supported by the University of Bonn under the European
Commission RTN programme HPRN-CT-2000-00131 and by the University of Oxford
(Leverhulme Trust research grant). This research is also supported in part by the EEC
research network HPRN-CT-2000-00148.
References
[1] Particle Data Group, Review of Particle Physics, Eur. Phys. J. C 15 (2000).
[2] C.H. Llewellyn Smith, G.G. Ross, J. Wheater, Nucl. Phys. B 177 (1981) 263;
L. Hall, Nucl. Phys. B 178 (1981) 263;
S. Weinberg, Phys. Lett. B 91 (1980) 51;
For a recent review see: K. Dienes, Phys. Rep. 287 (1997) 447.
[3] V. Kaplunovsky, J. Louis, Nucl. Phys. B 422 (1994) 57.
[4] D. Ghilencea, G.G. Ross, Nucl. Phys. B 569 (2000) 391.
[5] D. Ghilencea, G.G. Ross, Phys. Lett. B 442 (1998) 165.
[6] H.P. Nilles, hep-ph/0004064, TASI 1997 Lectures, Boulder, Colorado, and references therein.
[7] K. Dienes, E. Dudas, T. Gherghetta, Nucl. Phys. B 537 (1999) 47.
[8] I. Antoniadis, C. Bachas, E. Dudas, hep-th/9906039, and references therein.
[9] E. Witten, Nucl. Phys. B 471 (1996) 135.
[10] M. Carena, S. Pokorski, C.E.M. Wagner, Nucl. Phys. B 406 (1993) 59.
[11] M. Bastero-Gil, G.L. Kane, S.F. King, Phys. Lett. B 474 (2000) 103.
[12] K. Dienes, Phys. Rep. 287 (1997) 447.
[13] L. Dixon, V. Kaplunovsky, J. Louis, Nucl. Phys. B 355 (1991) 649.
[14] V. Kaplunovsky, Nucl. Phys. B 307 (1988) 145;
V. Kaplunovsky, Nucl. Phys. B 382 (1992) 436, Erratum.
[15] H.P. Nilles, S. Stieberger, Phys. Lett. B 367 (1996) 126.
[16] H.P. Nilles, hep-ph/9601241, and references therein. Lectures given at ICTP Summer School in
High Energy Physics and Cosmology, Trieste, Italy, 12 June 28 July 1995.
[17] L.E. Ibez, D. Lst, Nucl. Phys. B 382 (1992) 305.
[18] P. Mayr, S. Stieberger, Phys. Lett. B 355 (1995) 105.
[19] L.E. Ibez, H.P. Nilles, F. Quevedo, Phys. Lett. B 192 (1987) 332;
G. Aldazabal, A. Font, L.E. Ibez, A.M. Uranga, hep-th/9410206;
E. Witten, Nucl. Phys. B 258 (1985) 75;
E. Witten, Nucl. Phys. B 269 (1986) 79;
L.E. Ibez, H.P. Nilles, F. Quevedo, Phys. Lett. B 187 (1987) 25;
L.E. Ibez, J.E. Kim, H.P. Nilles, F. Quevedo, Phys. Lett. B 191 (1987) 283.
[20] D. Ghilencea, G.G. Ross, Phys. Lett. B 480 (2000) 355.
[21] C. Bachas, J. High Energy Phys. 9 (1998) 023, hep-ph/9807415.
[22] G. Aldazabal, L.E. Ibez, F. Quevedo, hep-ph/9909172.

Restoring Lorentz invariance in classical

N = 2 string
Stefano Bellucci, Anton Galajinsky 1
INFN Laboratori Nazionali di Frascati, C.P. 13, 00044 Frascati, Italy
Abstract
We study classical N = 2 string within the framework of the N = 4 topological formalism by
Berkovits and Vafa. Special emphasis is put on the demonstration of a classical equivalence of the
theories and the construction of an action for the N = 4 topological string. The SO(2, 2) Lorentz
invariance missing in the conventional BrinkSchwarz action for the N = 2 string is restored in the
N = 4 topological action. 2001 Elsevier Science B.V. All rights reserved.
PACS: 04.60.Ds; 11.30.Pb
Keywords: N = 2 string; Lorentz symmetry
1. Introduction
Nowadays, it seems to be a conventional wisdom to view string theories revealing an N extended local supersymmetry on the world-sheet as models describing a coupling of an
N -extended (conformal) d = 2 supergravity to matter multiplets. Technically, it suffices to
start with an appropriate supergravity multiplet, the component fields being connected by a
set of gauge symmetries, and require the coupling not to destroy the transformations. There
has been considerable interest in the field [19] which culminated in the classification
of the underlying N -extended superconformal algebras [1013]. The N 4 bound was
encountered as the only one compatible with the presence of a central charge in the
superconformal algebra (SCA). It should be stressed that with N growing the construction
of an action becomes technically involved. For example, the N = 4 action of Ref. [3]
appeals first to a dimensional reduction of the rigid N = 2, d = 4 nonlinear sigma model
to d = 2 which is then coupled to a supergravity multiplet. Notice also that for N > 1 the
existing actions [1,3] lack manifest Lorentz invariance in a target.
E-mail addresses: stefano.bellucci@lnf.infn.it (S. Bellucci), anton.galajinsky@lnf.infn.it (A. Galajinsky).
1 On leave from Department of Mathematical Physics, Tomsk Polytechnical University, Tomsk, Russia.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 2 3 5 - 8
120
S. Bellucci, A. Galajinsky / Nuclear Physics B 606 (2001) 119136
The N = 2 case, which is also the subject of this paper, is of particular interest. In
contrast to N = 0 and N = 1 strings, quantization of the model reveals only a finite number
of physical states in the spectrum. What is more, one encounters a continuous family of
sectors interpolating between R and NS which are connected by spectral flow [14]. The
latter fact, in particular, allows one to stick with a preferred sector. All tree-level amplitudes
with more than three external legs prove to vanish [15,16] (for details of loop calculations
see a recent work [17] and references therein), exhibiting a remarkable connection with
self-dual gauge or gravitational theory in two spatial and two temporal dimensions [16].
Turning to problematic points, in spite of being a theory of an N = 2, d = 2 supergravity
coupled to matter, there are no space-time fermions in the quantum spectrum. Furthermore,
as has been mentioned above manifest Lorentz covariance is missing in the BrinkSchwarz
action [1]. By this reason the notion of spin carried by a state is obscured and can be
clarified only after explicit evaluation of string scattering amplitudes [16].
Recently, Berkovits and Vafa introduced a new topological theory based on a small N =
4 SCA [18] (a similar extension of N = 2 SCA has been considered earlier by Siegel [19]).
After topological twisting, which incidentally does not treat all the fermionic currents
symmetrically and breaks SO(2, 2) down to U (1, 1) [18], the N = 4 formalism turns out
to be equivalent to the conventional N = 2 formalism. This has been demonstrated by
explicit evaluation of scattering amplitudes [18]. It was revealed also that the topological
prescription for calculating superstring amplitudes is free of total-derivative ambiguities
(for further developments see Ref. [20]). It is noteworthy, that a global automorphism group
of the small N = 4 SCA contains the full Lorentz group SO(2, 2) SU(1, 1) SU(1, 1)
which is larger than U (1, 1) U (1) SU(1, 1) intrinsic to N = 2 SCA.
In the present paper we study classical N = 2 string within the framework of the N = 4
topological prescription. An action functional adequate for the topological formalism is
constructed, this revealing the full Lorentz invariance in the target space. The motivation
for this work is two-fold. Firstly, it seems reasonable to use the N = 4 topological action
for restoring the Lorentz invariance missing in the previous analysis of the N = 2 string
scattering amplitudes. Secondly, being reduced to a chiral sector, the action is appropriate
for describing a chiral half of a recently proposed N = 2 heterotic string with manifest
spacetime supersymmetry [21].
The organization of the work is as follows. In Section 2 we outline the main features
concerning a conventional Lagrangian formulation for the N = 2 string. An extension by
topological currents and the issue of the classical equivalence are addressed in Section 3.
The global automorphism group of the N = 4 SCA is discussed in Section 4 which will
give us a key to the construction of the action. In Section 5 we apply Noether procedure to
install an extra U (1, 1) global invariance in the conventional N = 2 string action, thus
raising the global symmetry to the full Lorentz group. Beautifully enough, the worldsheet gauge fields fall in a multiplet of an N = 4, d = 2 supergravity. Section 6 contains
Hamiltonian analysis for the model at hand which proves to be a necessary complement to
that in Section 5. We summarize our results and discuss possible further developments in
Section 7. Our notations are gathered in Appendix A.
121
2. Classical N = 2 string in the Lagrangian framework

As originally formulated in Ref. [1], the action of the classical N = 2 string describes
a coupling of an N = 2 world-sheet supergravity (containing a graviton en (n stands for
a flat index), a complex gravitino A , A = 1, 2, and a real vector field A ) to a complex
(on-shell) matter multiplet. The latter involves a complex scalar zn and a complex Dirac
spinor An , with the target index n taking values n = 0, 1 in the critical dimension. The
BrinkSchwarz action reads

1
n en + i
n en
d d g g z z i
S0 =
2

1
n A en + z
n m en em
+
2

1
n m
+ z
(1)
,
en em
2
which explicitly lacks manifest Lorentz covariance. Due to the complex structure intrinsic
to the formalism, the (spin cover of the) full target-space Lorentz group Spin(2, 2) =
SU(1, 1) SU(1, 1) turns out to be broken down to U (1) SU(1, 1) U (1, 1) which
is the global symmetry group of the formalism.
Since in two spacetime dimensions (on the world-sheet) all irreducible representations
of the Lorentz group are one-dimensional, it seems convenient to get rid of the -matrices.
This, in particular, allows one to avoid the extensive use of cumbersome Fierz identities
in checking local symmetries of the action. Besides, the study of various heterotic theories
taking a chiral half from the N = 2 string [21,22] inevitably appeals to an action of such a
type. For the case at hand this yields (our notations are gathered in Appendix A)

1
d d g g z z + i 2 (+) (+) + (+) (+) e
S0 =
2

+ i 2 () () + () () e+ 2 (+) (+) A e
2 () () A e+ + 2i z () (+) e e+ 2i z (+) () e+ e
2i z (+) () e+ e + 2i z () (+) e e+
2 (+) () (+)() e+ e 2 () (+) () (+) e+ e

+ (+) (+) () () e+ e + e+ e

+ () () (+) (+) e+ e + e+ e .
(2)
Here the () indices indicate the chirality of an irreducible spinor representation with
respect to the world-sheet (local) Lorentz group (see Appendix A). These can also be used
to mark the right and left moving fermionic modes on the world-sheet of the string.
Apart from the usual reparametrization invariance, local Lorentz transformations
and Weyl symmetry, which allow one to bring locally the world-sheet metric to the
Minkowskian form, the action (2) reveals two more bosonic transformations with real
parameters a and b
122
A = a,
i
() = a() ,
2
i
() = a() ;
2
(3)
i
i
(+) = b(+) ,
() = b() ,
2
2
i
i
(+) = b(+) ,
(4)
() = b(),
2
2
where e1 = (det(en ))1 = g, these being sufficient to remove the corresponding

gauge field A , as well as the super Weyl transformation with two complex fermionic
parameters ()
A = e1 g b,
(+) = g e+ () ,
() = g e (+) ,

1
A = g e+ e (+) () + () (+)
2

1
+ g e e+ () (+) + (+) () ,
2
(5)
and an N = 2 local world-sheet supersymmetry with fermionic complex parameter ()

z = i () (+) ,
() = 0,
(+) = 0,
e+ = 0,

i
i
() = () + () A + () () + () () () e+
2
2 2
i
+ () () () e+ ,
2
1
i
(+) = () ze+ () () (+) e+
2
2

i
+ (+) () () () () e+ ,
2 2

i
e = e+ () () + () () e ,
2

1
A = () () () () () () e+
2

i
+ A () () + () () A () () + () () e+
2 2

i
(6)
() () + () () () () e+ e+ ,
4
and (+)
z = i (+) () ,
(+) = 0,
() = 0,
e = 0,

i
i
(+) = (+) (+) A (+) (+) + (+) (+) (+) e
2
2 2
i
(+) (+) (+) e ,
2
123
1
i
() = (+) ze + (+) (+) () e
2
2

i
() (+) (+) (+) (+) e ,
2 2

i
e+ = e (+) (+) + (+) (+) e+ ,
2

1
A = (+) (+) (+) (+) (+) (+) e
2

i
A (+) (+) + (+) (+) A (+) (+) + (+) (+) e
2 2

i
(7)
+ (+) (+) + (+) (+) (+) (+) e e .
4
Altogether these allow one to gauge away the gravitino fields () . In the equations above
stands for the conventional world sheet covariant derivative B = B B .

After the gauge fixing (for a more rigorous analysis see Section 6)
1
e+ 0 = e 0 = e+ 1 = e 1 = ,
2
A = 0,
() = 0,
(8)
only the first three terms survive in the action (2) and one has to keep track of the N = 2
superconformal currents which arise from a variation of the action with respect to the
world-sheet supergravity fields one had before fixing the gauge. We write these explicitly
in the Hamiltonian form (see also Section 6)

1
1
i
1 z pz +
1 z
(+) 1 (+) + (+) 1 (+) = 0,
T = pz +
2
2
2
2

1
1

G = pz +
G = pz +
1 z (+) = 0,
1 z (+) = 0,
2
2
J = (+) (+) = 0,
(9)
where (zn , pzn ), n = 0, 1, form a canonical pair, (zn , pz n ), are complex conjugates and
n , n are a couple of complex conjugate fermions
n

nm

z ( ), pzm ( ) = nm ( ),
z ( ), pzm
( ) = ( ),
n

( ), m ( ) = inm ( ),
(10)
(+)
(+)
with
= diag(, +) the Minkowski metric. There is also an analogous set where
1
1
(pz + 2
1 z ) and (+) are to be exchanged with (pz 2
1 z ), () , these for the right
movers.
nm
3. Adding topological currents

It is worth mentioning further that the N = 2 superconformal currents given in Eq. (9)
above are not the maximal closed set one can realize on the space of the matter fields. As
124
was pointed out in Refs. [18,19], two more bosonic currents

nm n(+) m(+) = 0,
nm n(+) m(+) = 0,
(11)
and two more fermionic ones

1
1
nm pz n +
(12)
1 zn m(+) = 0,
1 z n m(+) = 0,
nm pzn +
2
2
extend the algebra up to a small N = 4 SCA. After a topological twist, which does not
treat all the fermionic currents symmetrically and breaks SO(2, 2) down to U (1, 1) [18],
the N = 4 extension turns out to be equivalent to the N = 2 formalism. This has
been demonstrated by explicit computation of scattering amplitudes [18]. Guided by the
quantum equivalence, it seems quite natural to expect that a similar correspondence should
hold also at the classical level. We now proceed to show that the extra currents (11), (12) do
not contain an additional information as compared to that implied by the N = 2 currents.
Since equations of motion in both the cases are free and have the same form, this provides
the classical equivalence. It seems convenient first to break a target space vector index into
1
the lightcone components (see Eq. (A.3) of Appendix A). Denoting z = (pz + 2
1 z )
1
and z = (pz + 2 1 z), one finds for the currents (9) (for brevity we omit spinor indices
carried by the fermions)
T = z+ z z z+

i +
1 + 1 + + + 1 + 1 + = 0,
+
2
2
= + + = 0,
G = z+ z + = 0,
G
z
z
J = + + = 0,
(13)
while the newly introduced ones acquire the form

+ = 0,
+ = 0,
z+ z + = 0,
z+ z + = 0.
(14)
Assuming now that z+ = 0 2 one can readily solve for , by making use of the
second line of Eq. (13)
=
+
+ ,
z+
+ .
Being substituted in the expression for J these yield

+
z z + z z+ + + = 0.
(15)
(16)
It is trivial to observe now that the bosonic currents from Eq. (14) follow from Eq. (15).
The same turns out to be true for the fermionic ones, provided Eq. (16), T + = 0 and
T + = 0 have been used.
2 This can be achieved, for example, by imposing a light-cone gauge z+ + z + = z+ + z + + (p + + p + ). The
z
z
0
0
real part of z+ (and hence the modulus) then does not vanish and one can safely divide by z+ . It should be
stressed that, due to the presence of two temporal dimensions in the target, the light cone analysis for the N = 2
string seems to be less rigorous than in the Minkowski space.
125
Thus, one can conclude that at the classical level the extra currents of the N = 4
topological description do not contain any new information. Our analysis here is in
agreement with that of Siegel [19]. At the quantum level, as an alternative to the proof
by Berkovits and Vafa which appeals to the explicit evaluation of string amplitudes [18],
one could proceed directly from the small N = 4 SCA to verify that the positive (halfinteger) modes of the fermionic currents (12) kill all physical states, provided so do the
zero modes of bosonic ones (11). Since even for the smaller N = 2 SCA a proper analysis
shows the absence of excited states [23] and because the zero modes of the new bosonic
currents annihilate the ground state, one arrives at the same spectrum as for the ordinary
N = 2 string.
4. U (1, 1) and U (1, 1)outer

As has been mentioned above, after the topological twist the N = 4 prescription is
equivalent to the N = 2 formalism. There is an intimate connection between this point
and automorphism groups corresponding to the algebras. Since this will give us a key to
construction of the action, we turn to discuss this issue in more detail.
As we have seen earlier, the global symmetry group of the conventional N = 2 string,
which is also a trivial automorphism of an N = 2 SCA, is given by U (1, 1). It seems
instructive to give here the explicit realization. Combining zn and z n in a single column
n
Z A = zz n with A = 1, 2, 3, 4, one has for the infinitesimal U (1, 1) transformation
B
Z A = ii LA
i BZ ,
L Ti = Li ,
AB = diag(, +, , +),
(17)
where i are four real parameters and Li form a basis in the u(1, 1) algebra,
0
i
L1 =
0
0
1
0
L3 =
0
0
i
0
0
0
0
0
0
i
0
0
,
i
0
0
0
0
+1 0
0
,
0 +1 0
0
0 1
1
0
L2 =
0
0
0
0
0
1 0
0
,
0 +1 0
0
0 +1
0 +1
1 0
L4 =
0
0
0
0
0
0
0
0
.
0 1
+1 0
(18)
Because the matrices are block-diagonal the corresponding representation of the u(1, 1)
algebra is obviously decomposed into a direct sum and the fields zn and z n do not get
mixed under these transformations.
Beautifully enough, one can find another realization of the group on the space of the
fields at hand that do mix zn and z n (generators are off-diagonal) and, what is more
important, it automatically generates the currents of the small N = 4 SCA when applied
126
Table 1
U (1, 1)outer

0
0
0
)
2 (
2 ( )
2 ( )
0
2i3 ()
)
2i3 (
)
i4 ( +
i4 ( )
i4 ( )
Table 2
U (1, 1)

0
0
0
0
2i2 ()
)
2i2 (
0
0
0
0
0
0
to those of the N = 2 SCA [21]. We stick with the terminology of Ref. [20] and call this
U (1, 1)outer . The explicit form of the generators is
0 i 0
i 0
0
L1 =
0
0
0
0
0 i
1 0
0
0 1 0
L3 =
0
0 +1
0
0
0
0
0
,
i
0
0
0
,
0
+1
0
0
L2 =
0
i
0
0
L4 =
0
+1
0
0
i
0
0
0
+1
0
i
0
,
0
0
0 1
1 0
.
0
0
0
i
0
0
(19)
Notice that a combination of the form z n nm y m , which is trivially invariant under the
action of U (1, 1), does not hold invariant under the action of U (1, 1)outer , and a more
general object like Z A AB Y B is to be handled with in a formalism respecting the latter.
This, in particular, causes certain difficulties, when trying to construct vertex operators for
a recent N = 2 heterotic string with manifest spacetime supersymmetry [21].
In the next section we shall install U (1, 1)outer in the N = 2 string action by making use
of the Noether procedure. Before concluding this section we collect some technical points
regarding the action of U (1, 1) and U (1, 1)outer on the elementary objects like n nm m
and n nm m .
Independently of the statistics of the fields and , one finds for the outer group see
Table 1, while the variations with respect to the ordinary group prove to be much simpler
(Table 2).
When realizing U (1, 1)outer in the N = 2 string action, a successive inspection of the
terms entering Eq. (2) simplifies considerably with the use of Tables 1, 2.
127
5. Restoring Lorentz invariance

No work has to be done with the kinetic terms, these prove to be invariant with respect
to both U (1, 1) and U (1, 1)outer . The next two contributions involving the vector field A
are to be accompanied by

1
S1 =
2 () () e+ + (+) (+) e (B + iC )
d d g
2

2 () () e+ + (+) (+) e (B iC ) ,
(20)
where B and C are two new real vector (gauge) fields. The whole piece is inert under
U (1, 1)outer provided the transformation rules (Table 3) for the gauge fields.
For the ordinary U (1, 1) the triplet of vector fields proves to be invariant under 1 , 3
and 4 , while for 2 one has to set
2 A = 0,
2 B = 22 C ,
2 C = 22 B .
(21)
An important point to notice is that the transformations on the gauge fields do satisfy
the u(1, 1) algebra. Furthermore, one infers from Table 3 that they transform as a triplet
of SU(1, 1) subgroup of the full U (1, 1)outer, so that one can set a single field AI =
(A , B , C ), I = 1, 2, 3.
Let us now proceed to the terms linear in z, z . In order to make them U (1, 1)outer
invariant it suffices to introduce a couple of complex fermions (+) and () which
bring the following contribution to a full action

1
S2 =
d d g 2i z() (+) e e+ + 2i z () (+) e e+
2
2i z(+) () e+ e

2i z (+) () e+ e .
(22)
These should transform according to the rules presented in Table 4.
The action of U (1, 1) proves to be nontrivial only for 2 and looks like
2 (+) = 2i2 (+),
2 () = 2i2 () .
(23)
It is noteworthy, that the fermionic fields () , () furnish a twodimensional complex

representation of SU(1, 1)outer (to be more precise, the representation is realized on a four
component Majorana spinor (, , ,
);
the indices and () hold inert). One can
i
, i = 1, 2, this composed of () , ()
again set a single complex fermionic field ()
and forming a doublet representation of SU(1, 1)outer. As we shall see below no extra fields
are necessary to make the full action U (1, 1) and U (1, 1)outer invariant. One eventually
i
).
comes to the conclusion that on the world-sheet the theory is described by (en , AI , A
Beautifully enough, this set coincides with the N = 4, d = 2 supergravity multiplet (see,
e.g., [5]). Thus, within the framework of the N = 4 topological formalism the action
describes a coupling of the N = 4, d = 2 supergravity to matter multiplets.
128
Table 3
U (1, 1)outer
A
B
C
0
0
0
42 B
2 A
0
0
23 C
23 B
44 C
0
4 A
Table 4
U (1, 1)outer
(+)
()
(+)
()
0
0
0
0
2 ( )
(+)
2 ( )
()
2 ( )
(+)
2 ( )
()
2i3 (+)
2i3 ()
0
0
i4 ( )(+)
i4 ( )()
i4 ( + )
(+)
i4 ( + )
()
We now turn to discuss the last two terms in Eq. (2). Introducing a further amendment
to the action

1
() ( + )
()
S3 =
d d g (+) (+) (+) (+) ( )
2

() ( + )
+ (+) (+) + (+) (+) ( + )
()

(+) ( + )
+ () () () () ( )
(+)

(+) ( + )
+ () () + () () ( + )
(+)

1
(24)
e+ e + e+ e ,
4
one can compensate their variations. It remains to discuss two terms involving both (+)
and () and entering the fifth line in Eq. (2). Here the analysis turns out to be more
intricate. The most general form of the compensators is

1
d d g () (+) () (+)
S4 =
2

+ () (+) + () (+)

(25)
+ 2() (+) + 2(+) () e+ e ,
where , , and are some as yet unspecified tensors. In the following, we

shall assume that , , , 0 as () 0. This seems natural because in the limit
the action we are searching for should reduce to the N = 2 string action. In order that the
action be real the compensators must obey
( ) = ,
( ) = ,
( ) = .
(26)
A variation of the whole piece with respect to U (1, 1) and U (1, 1)outer yields a set of
constraints on the newly introduced fields. For example, the 2 transformation implies

2 + 22 + + (+) () + () (+) = 0,

() () ( )
(+) = 0,
2 + 2 (+) ( )

(+) (+) ( )
() = 0,
2 + 2 () ( )
129
(27)
with being inert. With the help of Table 4 and Eq. (23) one can readily solve
the variational equations. Surprisingly enough, the global invariance does not fix the
compensators uniquely. One encounters the following solution
() ( + )
(+) + c1 ( )
(+) ( + )
()
= c1 ( )
+ c2 ( )
() ( + )
(+) + c2 ( )
(+) ( + )
()
+ (+) () (+) () + () (+) () (+) ,
= c1 ((+) () + () (+) ) + c2 (() (+) + (+) () )
(+) () ,
= c1 ( )
() ( )
(+) + c1 ( )
(+) ( )
()
+ c2 ( )
() ( )
(+) + c2 ( )
(+) ( )
()
+ (+) () + (+) () + () (+) + () (+) ,
(28)
with c1 and c2 being arbitrary real numbers. Obviously, the missing information needed to
fix them is encoded in local symmetries an ultimate action must possess. As is clear from
our discussion above, one expects the action to exhibit N = 4 local supersymmetry.
Alternatively, one could proceed to the Hamiltonian formalism and fix the constants
there from the requirement that only the set of currents generating a small N = 4 SCA
remains after completion of the Dirac procedure. Extra constrains are not allowed. Because
the Lagrangian and Hamiltonian methods are in one-to-one correspondence, this will yield
a correct answer. It should be mentioned that, generally, it may happen to be an intricate
task to reveal all the local symmetries just from the form of an action at hand. For example,
the b-symmetry of the N = 2 string action displayed above in Eq. (4) was missing in
the original paper [1] and has been discovered only four years later in Ref. [24]. As is
well known, a straightforward way to discover how many local symmetries one should
expect to find in a Lagrangian theory is prompted by the Hamiltonian framework. For
models with irreducible constraints it suffices to count the number of Lagrange multipliers
corresponding to primary first class constraints (see, e.g., [25,26]). In order to elucidate
this point for the model under consideration and for the sake of coherence we then proceed
to the Hamiltonian formalism to fix the missing constants.
6. Hamiltonian analysis
Introducing momenta associated to the configuration space variables one finds the
following primary constraints (we define a momentum conjugate to a fermionic variable to
be the right derivative of a Lagrangian with respect to velocity)
pe = 0,
pA = 0,
pB = 0,
pC = 0,
p = 0,
p = 0,
p = 0,
p = 0,
(29)
130
and
i 2
i 2
(+)
0
(+) e 0 = 0,
(+)
+
(+) p +
(+) e = 0,
2e
2e
i 2
i 2
()
()
0
() e+ = 0,
() e+ 0 = 0,
() p +
(30)
() p +
2e
2e
where e = det(en ) and pq stands for a momentum canonically conjugate to a configuration space variable q. It is worth mentioning now that, a Hamiltonian of a diffeomorphism
invariant theory is given by a linear combination of constraints (Diracs theorem). Remarkably, the form of the coefficients c1 and c2 is uniquely fixed already from the form of the
Hamiltonian. The key point to notice is that, since the currents under consideration do no
involve (+) and () simultaneously, such terms should disappear from the Hamiltonian.
This automatically holds in the Hamiltonian of the N = 2 string

HN=2 = d pe e + pA A + p + p
(+)
p
+ (+) (+) + () () + (+) (+) + () ()

i 2
+
(+) 1 (+) + (+) 1 (+) e 1
2e

1
i 2
e
1
() 1 () + () 1 () e+ + 0 0 pz pz +
+
1 z1 z
2e
e+ e
(2)2
0 1

1
e+ e + e 0 e+ 1 (pz 1 z + pz 1 z )
0
0
2e+ e

e+
e
1
1
1 z () (+) 0 i pz +
1 z (+) () 0
+ i pz
2
e+
2
e

e+
e
1
1
1 z () (+) 0 i pz +
1 z (+) () 0
+ i pz
2
e+
2
e
2
2
(+) (+) A e
() () A e+
2e
2e

0
1
e+
+
(+) (+) () () e+ e + e e+ 2e e
2e
e 0

1
e 0
+
() () (+) (+) e+ e + e e+ 2e+ e+ 0 . (31)
2e
e+
For the full Hamiltonian this turns out to be true only if the constants take the following
specific value
c1 = 1,
c2 = 0.
In this case the full Hamiltonian acquires the form

HN=4 = HN=2 + d pB B + pC C + p + p

2
() () e+ + (+) (+) e (B + iC )
+
2e
(32)
131

2
() () e+ + (+) (+) e (B iC )
2e

e+
e
1
1
+ i pz
1 z () (+) 0 i pz +
1 z (+) () 0
2
e+
2
e

e+
e
1
1
1 z () (+) 0 i pz +
1 z (+) () 0
+ i pz
2
e+
2
e

1
(+) (+) (+) (+) ( )
+
() ( + )()
8e

() ( + )()
+ (+) (+) + (+) (+) ( + )

(+) ( + )(+)
+ () () () () ( )

(+) ( + )(+)
+ () () + () () ( + )

e+ e + e e+
1
(+) (+) () ()
+
e

e+ 0
(+) (+) () () (+) (+) () () e e 0
e
1
() () (+) (+) () () (+) (+)
+
e

e 0
,
() () (+) (+) e+ e+
(33)
e+ 0
which, as we shall shortly see, is indeed a linear combination of the constraints. In the
relations above s stand for the Lagrange multipliers associated to the primary constraints.
Given the form of the constants c1 and c2 , the solutions (28) simplify considerably
= (+) () ,
= (+) () + () (+) (+) () () (+) ,
= (+) () ,
= (+) () + () (+) + (+) () + () (+) .
(34)
An ultimate form of the N = 4 topological string has thus been fixed and is given by the
sum
S = S0 + S1 + S2 + S3 + S4 .
(35)
It has to be stressed that SO(2, 2) global invariance missing in our starting point S0 is
restored in the full action S.
We now turn to outline the Dirac procedure. That only the currents of a small N = 4
SCA remain as essential constraints on the system provides a consistency check for the
formalism developed so far.
Following Diracs method, one requires primary constraints to be conserved in time.
In other words, a point constrained to lie on a surface at some initial instant of time is
not allowed to leave the latter in the process of time evolution. Being applied to pA = 0,
132
pB = 0 and pC = 0 from Eq. (29) this gives

(+) (+) = 0,
(+) (+) = 0,
(+) (+) = 0,
(36)
plus analogous equations where a (+) should be exchanged with a (). These are to be
viewed as secondary constraints. In deriving the relations above Eq. (A.5) of Appendix A
proves to be helpful. Analogously, the fermionic primary constraints from Eq. (29) induce
further relations

1
1
pz +
1 z (+) = 0,
1 z () = 0,
pz
2
2

1
1
1 z (+) = 0,
1 z () = 0,
pz
pz +
(37)
2
2
as well as their complex conjugates. The remaining equation pe = 0 in (29) yields two
more secondary constraints

1
1 z pz
pz
2

1
1 z pz +
pz +
2

1
i 2
1 z +
() 1 () + () 1 () e+ 0 = 0,
2
2 2 e

1
i 2
1 z
(+) 1 (+) + (+) 1 (+) e 0 = 0. (38)
2
2 2 e
In their turn the -equations from Eq. (30) do not lead to any new constraints but determine
(indicating the presence of second
the value of the Lagrange multipliers () and ()
class constraints). These prove to be complicated expressions. Of direct relevance to the

forthcoming discussion are some of their consequences which we list below
(+) (+) = (+) 1 (+)
e 1
,
e 0
() () = () 1 ()
e+ 1
,
e+ 0
() () 0,
(+) (+) 0,

e+ 1
1
1
1 z () = pz
1 z 1 () 0
pz
2
2
e+

e
1
1
1 z pz
1 z (+) e+ ,
pz
2
2
2e+ 0 e+ 0

1
1
e 1
pz +
1 z (+) = pz +
1 z 1 (+) 0
2
2
e

e
1
1
1 z pz +
1 z () e ,
+
pz +
2
2
2e 0 e 0

e+ 1
1
1
pz
1 z () = pz
1 z 1 () 0
2
2
e+

e
1
1
1 z pz
1 z (+) e+ ,
+
pz
2
2
2e+ 0 e+ 0
133

e 1
1
1
1 z (+) = pz +
1 z 1 (+) 0
pz +
2
2
e

e
1
1
1 z pz +
1 z () e ,
pz +
2
2
2e 0 e 0
(39)
plus their complex conjugates (notice that (() ) = () ). The symbol stands for an
equality up to a linear combination of the constraints.
An important next point to check is that the secondary constraints (36), (37), (38) are
preserved in time. It is straightforward, although a bit tedious, exercise to verify that this
is indeed the case and no tertiary constraints arise or Lagrange multipliers get fixed. The
Dirac algorithm thus ends here.
To fix the gauge freedom one encounters in solutions of equations of motion due to the
presence of the unspecified Lagrange multipliers, it is customary to impose extra gauge
conditions. For the case at hand one can choose them in the form
1
e+ 0 = e 0 = e+ 1 = e 1 = ,
2
B = 0,
C = 0,
A = 0,
() = 0,
() = 0,
() = 0,
() = 0.
(40)
The conservation in time of the gauges indeed specifies the Lagrange multipliers
e = A = B = C = = = = = 0,
(41)
and singles out a unique trajectory from the bunch of those connected by gauge
transformations. Notice also that the complicated expressions for , simplify
considerably in the gauge chosen
(+) = 1 (+) ,
= 1 (+) ,
(+)
() = 1 () ,
= 1 () .
()
(42)
It remains to notice that the second class constraints from Eq. (30) can be considered to be
fulfilled strongly, provided a conventional Dirac bracket has been introduced. This removes
()
()
the momenta p , p , while leads , to obey
n

m
m
(+) , (+)
(43)
= inm ,
() , ()
= inm .
D
D
Finally, the full Hamiltonian boils down to

i
gauge
(+) 1 (+) + (+) 1 (+)
HN=4 = d
2

i
() 1 () + () 1 ()
+
2

1
+ 2 pz pz +
1 z1 z ,
(2)2
(44)
which also guarantees the free dynamics for the physical fields z, z , , remaining in the
question.
134
Now let us comment on the number of local symmetries one can expect to reveal within
the Lagrangian framework. Restricting ourselves first to the case of the conventional N = 2
string Hamiltonian HN=2 , one sees that, before the gauge fixing, the Lagrange multipliers
e , A , () , () are not determined by the Dirac procedure and bring degeneracy
into solutions of equations of motion. This is known to be in a one-to-one correspondence
with the presence of local symmetries in the Lagrangian framework. One indeed finds two
local diffeomorphisms, the Weyl symmetry and the local Lorentz symmetry corresponding
to e . Associated to A are the local a and b symmetries we considered in Section 2.
The remaining four complex fermionic Lagrange multipliers () correspond to the
super Weyl transformations (two complex parameters) and the N = 2 local world-sheet
supersymmetry (two complex parameters). Turning to the full Hamiltonian HN=4 , one
concludes that the action (35) must exhibit four more bosonic symmetries associated to
B and C , the corresponding parameters presumably forming a triplet with a and b from
Eqs. (3), (4) under the action of SU(1, 1)outer. The latter point is also prompted by the fact
that the corresponding gauge fields A , B , C do fall in the triplet. Besides, one expects
to reveal a set of fermionic symmetries with four complex parameters, these corresponding
to () . Because the gauge fields () , () proved to form a spinor representation of
SU(1, 1)outer , one can indeed expect a doubling of the super Weyl transformations and the
N = 2 local supersymmetry intrinsic to the conventional (U (1, 1) covariant) formalism.
Although we did not perform the Lagrangian analysis explicitly, on the basis of the
Hamiltonian analysis outlined above one can conclude that the full action (35) must exhibit
an N = 4 local supersymmetry.
7. Conclusion
Thus, in the present paper we extended the N = 2 string action to the one adequate
for the N = 4 topological prescription by Berkovits and Vafa. The major advantage
of the new formulation is that the Lorentz invariance holds manifest. The approach
proposed in this work differs from the previous ones. We neither use gauging of global
supersymmetry [1] nor appeal to dimensional reductions [3]. Based on a simple observation
that an automorphism group of the small N = 4 SCA involves an extra U (1, 1) we
constructed the N = 4 topological string action just by installing the latter in the N = 2
string. To guarantee the new global invariance extra world-sheet fields are to be introduced.
Remarkably, they proved to complement the N = 2, d = 2 supergravity multiplet to that of
the N = 4, d = 2 world-sheet supergravity.
Turning to possible further developments, the obvious point missing in this paper is
to explicitly reveal extra local symmetries indicated in Section 6. This seems to be a
technical point rather than an ideological one. A more tempting question is whether it is
possible to make use of the action for a covariant calculation of scattering amplitudes.
It is worth mentioning also, that the action functional, when reduced to a chiral half,
describes the right movers of a recent N = 2 heterotic string with manifest space time
135
supersymmetry [21]. Adding a Lagrangian describing the left movers might give a heterotic
action missing in Ref. [21] and suggests another interesting problem.
Acknowledgements
We would like to thank Nathan Berkovits for useful discussions. This work was
supported by the Iniziativa Specifica MI12 of the Commissione IV of INFN and by INTAS
grant No. 00-00254.
Appendix A
It seems customary to use purely imaginary -matrices to describe spinors on the world
sheet of a string. A conventional basis is

0 i
0 i
0 =
,
1 =
,
i 0
i 0

1 0
3 = 0 1 =
(A.1)
.
0 1
It is trivial to check the algebraic properties
{m , n } = 2mn ,
m n = mn mn 3 ,
(0 m )T = 0 m ,
(A.2)
where mn = diag(, +) and mn is the 2d Levi-Civita totally antisymmetric tensor, 01 =
1.
Since in d = 2 irreducible representations of the Lorentz group are one-dimensional, it
is convenient to use the light-cone notation for vectors and spinors

1
(+)
A = (A0 A1 ),
, (A.3)
An Bn = A+ B A B+ ,
A =
()
2
which makes the latter point more transparent. Actually, the Lorentz transformation
acquires the form
1
(A.4)
() = () ,
2
and the invariance is kept, for example, by contracting a + with a (one could fairly
well contract a (+) with a () or a + with two ()). In the light cone notation one
can get rid of -matrices and work explicitly in terms of irreducible components of tensors
under consideration.
Introducing the zweibein em on the world sheet, nm = en g em , g = en nm em ,
where stands for a curved index and m for a flat one, one can easily verify the relations
A = A ,
g = e+ e e e+ ,
mn em en = e+ e + e e+ = e ,
++ = = 0,
where e = det(em
).
+ = + = 1,
(A.5)
136
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
L. Brink, J.H. Schwarz, Nucl. Phys. B 121 (1977) 285.

P. van Nieuwenhuizen, Int. J. Mod. Phys. A 1 (1986) 155.
M. Pernici, P. van Nieuwenhuizen, Phys. Lett. B 169 (1986) 381.
K. Schoutens, Nucl. Phys. B 292 (1987) 150.
S.J. Gates Jr., Y. Hassoun, P. van Nieuwenhuizen, Nucl. Phys. B 317 (1989) 302.
S.J. Gates Jr., L. Lu, R. Oerter, Phys. Lett. B 218 (1989) 33.
C. Kounnas, M. Porrati, B. Rostand, Phys. Lett. B 258 (1991) 61.
S.J. Gates Jr., Phys. Lett. B 338 (1994) 31.
S. Bellucci, E. Ivanov, Nucl. Phys. B 587 (2000) 445.
P. Ramond, J. Schwarz, Phys. Lett. B 64 (1976) 75.
M. Bershadsky, Phys. Lett. B 174 (1986) 285.
R. Gastmans, A. Sevrin, W. Troost, A. Van Proeyen, preprints KUL-TH-86/6, KUL-TH-86/7.
K. Schoutens, Phys. Lett. B 194 (1987) 75.
A. Schwimmer, N. Seiberg, Phys. Lett. B 184 (1987) 191.
H. Ooguri, C. Vafa, Mod. Phys. Lett. A 5 (1990) 1389.
H. Ooguri, C. Vafa, Nucl. Phys. B 361 (1991) 469.
G. Chalmers, O. Lechtenfeld, B. Niemeyer, Nucl. Phys. B 591 (2000) 39.
N. Berkovits, C. Vafa, Nucl. Phys. B 433 (1995) 123.
W. Siegel, Phys. Rev. Lett. 69 (1992) 1493.
N. Berkovits, C. Vafa, E. Witten, JHEP 9903 (1999) 018.
S. Bellucci, A. Galajinsky, O. Lechtenfeld, A heterotic N = 2 string with spacetime
supersymmetry, hep-th/0103049.
J. de Boer, K. Skenderis, Nucl. Phys. B 500 (1997) 192.
J. Bienkowska, Phys. Lett. B 281 (1992) 59.
E. Fradkin, A. Tseytlin, Phys. Lett. B 106 (1981) 63.
M. Henneaux, C. Teitelboim, J. Zanelli, Nucl. Phys. B 332 (1990) 169.
A. Deriglazov, K. Evdokimov, Int. J. Mod. Phys. A 15 (2000) 4045.

The Galilean nature of V-duality

for noncommutative open string
and YangMills theories
Rong-Gen Cai a , J.X. Lu b , Yong-Shi Wu c
a Department of Physics, Osaka University, Toyonaka, Osaka 560-0043, Japan
b Michigan Center for Theoretical Physics, Randall Physics Laboratory, University of Michigan,
Ann Arbor MI 48109-1120, USA

c Department of Physics, University of Utah, Salt Lake City, UT 84112, USA
Received 13 February 2001; accepted 9 May 2001
Abstract
A V-duality conjecture for noncommutative open string theories (NCOS) that result from
decoupling D-branes in Lorentz-boost related backgrounds was put forward recently in hepth/0006013. The aim of this paper is to test the Galilean nature of this conjecture in the gravity
dual setup. We start with an (F, D3) bound state Lorentz-boosted along one D3-brane direction
perpendicular to the F-string, and show that insisting a decoupled NCOS allows only infinitesimal
Lorentz boosts. In this way, it is shown that the V-duality relates a family of NCOS by Galileo
boosts. Starting with a Lorentz-boosted (D1,D3) bound state, we show that a similar V-duality works
for noncommutative YangMills (NCYM) theories as well. In addition, we deduce by a holography
argument that the running string tension, as a function of the energy scale, for NCOS (or NCYM)
remains unchanged under V-duality. 2001 Elsevier Science B.V. All rights reserved.
PACS: 12.10.-g
1. Introduction
There is a surge of interest recently in seeking dynamical theories without gravity in
string/M theory. The motivation is two-fold. On one hand, we try to understand better the
Standard Model physics using the knowledge of string/M theory. One good example is the
AdS/CFT correspondence and its variations. On the other hand, we intend to collect more
theoretical data for the eventual formulation of M-theory since these decoupled theories are
part of the big M-theory and we have a better hand on them due to the absence of gravity.
E-mail addresses: cai@het.phys.sci.osaka-u.ac.jp (R.-G. Cai), jxlu@umich.edu (J.X. Lu),
wu@physics.utah.edu (Y.-S. Wu).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 2 3 1 - 0
138
R.-G. Cai et al. / Nuclear Physics B 606 (2001) 137150
Among these decoupled theories, for the purpose of this paper, we will focus on the
so-called noncommutative YangMills theory (NCYM) 1 and noncommutative open string
theory (NCOS). 2 In the simplest context, a (1 + p)-dimensional noncommutative Yang
Mills with spacespace noncommutativity arises in the decoupling limit as the decoupled
theory of Dp branes in a constant magnetic background (or constant B-field with only
spatial components). However, a decoupled noncommutative field theory does not exist
for Dp branes in a constant electric background. Instead, one can have a decoupled open
string theory living on the Dp brane worldvolume with spacetime noncommutativity in
the critical electric field limit. This is consistent with the fact that a unitary field theory
cannot exist with spacetime noncommutativity [20,21]. The basic picture here is that in
the decoupling limit the near critical electric force stretches the open string ending on the
Dp brane, and balances the usual string tension to end up with almost tensionless string
confined on the brane. 3
It is well-known that we have various dualities such as S- and T -, or in general U -dualities in the big M-theory. Up to this point, these dualities are global discrete transformations
which relate different but physically equivalent vacua in M-theory. These dualities are
believed to be inherited in the so-called little m-theory without gravity which is obtained
as a decoupled theory of M-theory. NCOS and NCYM are among this little m-theory.
Because of the absence of gravity (or in general the closed strings) in these decoupled
theories, one may wonder whether there exist some new global transformations, 4 also
called dualities, which connect physically equivalent theories. It has been suggested in [14]
that there exists a new spacetime duality relating different but physically equivalent
NCOS which result from decoupling D-branes in Lorentz-boost related backgrounds. This
can be understood as follows: the open string metric and spacetime (and/or spacespace)
noncommutativity are both dictated by constant electric (and/or magnetic) background
on the D-brane worldvolume. Some background configurations are related by the
worldvolume Lorentz boosts. The corresponding decoupled worldvolume noncommutative
theories 5 are conjectured to be physically equivalent to each other, since the Lorentz
boosts are expected to be a symmetry of the parent perturbative open string theory in the
decoupling limits. In this way, some decoupled theories with different noncommutativities
and/or (open-string) metrics can be mapped to each other. For example, by first applying
a Lorentz boost on a purely electric background and then imposing the corresponding
decoupling limit, one can obtain a (1 + p)-dimensional NCOS with both spacetime and
spacespace noncommutativities as a decoupled theory of Dp branes with constant electric
and magnetic backgrounds. This NCOS is related to the original NCOS with only space
time noncommutativity [14] by the new spacetime duality. This conjectured duality was
1 An incomplete list of Refs. is [18].
2 An incomplete list of Refs. for NCOS is [919].
3 We can keep the tension for the NCOS fixed by scaling the original tension to infinity.
4 We exclude those transformations which belong to the symmetry group of the underlying theory.
5 The nature of theory (either NCOS or NCYM) cannot be changed since the Lorentz boost leaves E2 B2
invariant.
139
named V-duality in Ref. [14], because it originates from boosts, characterized by a relative
velocity, and alphabetically it follows the existing S-, T - and U -dualities.
Naturally raised is the following question: what is the action of the V-duality on metric
and noncommutativities? Namely, what is the transformation that connects the decoupled
theories arising from decoupling Dp branes in backgrounds related by a Lorentz boost? 6 It
is not necessarily a Lorentz boost, because the decoupling limit involves intriguing scalings
of the background and the (closed-string) metric. For example, consider a NCOS theory
resulting from decoupling D-branes in a background which is related to a purely electric
background by a Lorentz-boost. In Ref. [14] an analysis using SeibergWitten relations [5]
indicated that the result is a Galilean boost for the Lorentz boost in a direction orthogonal
to that of the electric background. This is a particular case with orthogonal electric and
magnetic backgrounds.
In summary, the V-duality can be stated clearly in two parts: (a) the NCOSs (or
NCYMs) resulting from decoupling D-branes in Lorentz-related backgrounds are related
to each other by Galileo boosts; 7 and (b) such related NCOS or NCYM theories are
physically equivalent to each other. In this paper, we will test the V-duality from the
gravity dual description of NCOS, and will find that indeed the V-duality holds exactly
as suggested in [14]. Further, we will extend this analysis of V-duality to NCYM theories
that result from decouplings of Dp branes in backgrounds Lorentz-boost related to a
purely magnetic background and show that the V-duality holds for NCYM as well. For
concreteness, we will focus on the case of p = 3 from now on, but the conclusion drawn is
general.
In this Letter we limit ourselves to the particular cases in which the backgrounds are
Lorentz-boost related to a purely electric or a purely magnetic background. However,
our investigation indicates that the V-duality holds in general, for example, for NCOS
and NCYM resulting from their respective decouplings of D3 branes with backgrounds
Lorentz-boost related to a given parallel electric and magnetic background as discussed
in [14,15,19]. We plan to report the general discussion of V-duality and its structure,
significance and relations to other theories such as Matrix string theory in a separate
paper [22].
This paper is organized as follows. In Section 2, we first present a gravity configuration
of (F, D3) bound state boosted along a brane direction orthogonal to the F-strings. This
system provides the gravity dual description of NCOS resulting from the decoupling of D3
branes in an orthogonal electric and magnetic background (time-like). We show that the
gravity description is related to that of NCOS resulting from a purely electric background
by a Galilean transformation. In other words, we show that the V-duality proposed in [14]
holds also true in the gravity description. In Section 3, we present a gravity configuration
of (D1,D3) bound state boosted along a brane direction orthogonal to the D-strings. This
system provides a gravity dual description of NCYM resulting from the decoupling of
6 We consider only those Lorentz boosts which do not leave the background fields invariant.
7 Recall that the noncommutative geometry for either NCOS or NCYM is described by the (open string) metric
and the (anti-symmetric) coordinate noncommutative matrix.
140
D3 branes in an orthogonal electric and magnetic background (space-like). We show that

similar V-duality holds here almost trivially. In Section 4, we give a detailed explanation of
holographic correspondence proposed in [14,28] for NCYM and NCOS. Then we give
a holographic derivation of the NSNS fields in the gravity dual description of NCOS
discussed in Section 2. We conclude this paper in Section 5 where some issues on the
validity of V-duality are also addressed.
2. The Galilean nature of NCOS: V-duality

We test the V-duality for NCOS using its gravity description in this section. The relevant
gravity configuration can be obtained by a Lorentz boost of the (F,D3) bound state [23,24]
along a D3 brane direction orthogonal to the F-strings in the bound state. The explicit form
is

0 2 1 2 2 2
2
1/2 H
ds = H
dx
+ dx
+ dx
H

3 2

4ngs 2
2
0 2
sinh
dx
+
dx
sin
tan
cosh
dx
+
r 4H

+ H 1/2 dr 2 + r 2 d52 ,

e = gs H /H ,
F5 = 16n 2 (5 + 5 ),

2 B = sin H 1 cosh dx 0 dx 1 + sinh dx 1 dx 2 ,

A2 = gs1 tan H 1 sinh dx 0 dx 3 + cosh dx 2 dx 3 ,
(2.1)
where we have the metric in string frame, 5 denotes the volume form of unit 5-sphere,
denotes the Hodge dual in 10 dimensions, and the harmonic functions are
4ngs 2 1
4ngs 2

(2.2)
,
H
=
1
+
cos .
cos
r4
r4
In the above, the integer n is the number of D3 branes in the bound state and gs is the
asymptotic string coupling. The boost is along the x 2 direction with boost parameter .
When = 0, we recover the (F,D3) bound state. We can read the electric and magnetic
=
background fields from the asymptotic B-field (i.e., r ) from the above as E1 = B01
sin cosh , B3 = B12 = sin sinh . For = 0, we have only electric field. We have here
E2 B2 = sin2 0 and E B = 0 which are invariant under a Lorentz transformation
as is evident. For NCOS, we need sin
= 0, therefore E2 B2 > 0, i.e., time-like as it
should be. The gravity dual description can be obtained from the above with the following
decoupling limit:
H =1+

= eff
,
1/2
r = eff
u,
cos = ,
= v,
x 0,1
x 0,1 = ,

x 2,3 =
gs =
G2o
,

2,3
x ,
(2.3)
, v,
where 0 while parameters eff
Go , u, x with ( = 0, 1, 2, 3) remain fixed. We
would like to point out that insisting a decoupled gravity dual of NCOS allows only an
141
infinitesimal boost = v.

This in turn says that a finite boost will not give a decoupled
NCOS in the decoupling limit. Therefore, a Lorentz boost on a given NCOS will render
it undecoupled. This is entirely consistent with the conclusion drawn in [14] in a different
analysis. With the above, the gravity description is then
2

2

2
1/2 u
d x 02 + d x 12 + h1 d x32 + d x 2 v d x 0
ds = h
R2
2

2 2 du
2
+ eff
R
+
d
,
5
u2
e = G2o h1/2 ,
F0123u = 2
16nu3
,
2 R8 h
eff
u4
d x 0 d x 1 ,
R4

u4
A2 = 2 4 d x 2 v d x 0 d x 3 ,
Go R h
2 B =
(2.4)
where
h=1+
u4
,
R4
R4 =
4nG2o
.
2
eff
(2.5)
It is not difficult to examine that the parameter v looks like a velocity and this
gravity description is related to that of NCOS with only spacetime noncommutativity
(corresponding to v = 0 or resulting from the background without boost) by a Galilean
transformation
x 0 x 0 ,
x 2 x 2 v x 0 .
(2.6)
Further, if we calculate the noncommutative parameters, we have

01 = 2eff
,

12 = 2eff
v,
(2.7)
which is consistent with the Galilean transformation given in Eq. (2.6) above. Therefore,
we in a gravity setup show that the V-duality conjecture proposed in [14] holds true.
3. The Galilean nature of NCYM: V-duality

The field theory analysis of NCYM for this case has been given in [14]. It concluded
there that one cannot have NCYM with spacetime noncommutativity in addition to
the spacespace one. This is also confirmed in other independent analysis, for example,
in [15,19]. This clearly indicates that one cannot perform a Lorentz transformation
on NCYM since it will in general lead to spacetime noncommutativity. Any allowed
transformation on the decoupled theory must be induced from the transformation of string
theory before the decoupling. We will show in this section that the allowed one is again
a Galilean transformation, describing the V-duality action. This is entirely consistent with
142
the fact that we cannot have spacetime noncommutativity for NCYM since a Galilean
transformation on a spacespace noncommutativity cannot give rise to a spacetime one.
In this section, we will spell out the V-duality action for NCYM using its gravity dual
description. The relevant gravity configuration is the (D1, D3) bound state [23,25] boosted
along a D3 brane direction perpendicular to the D-strings in the state. 8 Its explicit form is

2
2
2 H
2
2
1/2
dx 0 + dx 1 + dx 2 + dx 3
ds = H
H

H H
2
0 2
+ H 1/2 dr 2 + r 2 d52 ,
+
cosh dx sinh dx

H

F5 = 16n 2 (5 + 5 ),
e = gs H /H ,

2 B = H 1 tan sinh dx 0 dx 3 cosh dx 2 dx 3 ,

A2 = gs1 H 1 sin cosh dx 0 dx 1 + sinh dx 1 dx 2 ,
(3.1)
where ds 2 is the string metric and the harmonic functions H and H are also given by
Eq. (2.2). We can read the D3 brane worldvolume electric and magnetic background fields
= tan sinh , B = B = tan cosh . Now
from the asymptotic B-field as E3 = B03
1
23
2
2
2
we have E B = tan 0 and E B = 0 which are Lorentz invariant. Since we are
interested in NCYM, tan is nonzero. So E2 B2 < 0, i.e., space-like as it should be for
NCYM. The gravity dual of NCYM can be obtained from the above with the following
decoupling limit:

,
= eff
cos = ,

r = eff
u,
x 0,1 = x 0,1 ,
= v,
gs = G2o ,
x 2,3 = x 2,3,
(3.2)
, v,
Go , x with ( = 0, 1, 2, 3) remain fixed. With the
where 0 while parameters eff
above decoupling limit, the gravity dual of NCYM is
ds 2 / =
2 2

u2
2
2
1
x 0 2
d
x
d
x
+
d
x
+
h
+
d
x
d
0
1
3
R2
2

2 2 du
2
+ eff
R
+
d
5 ,
u2
e = G2o h1/2 ,
2 B =
A2 =
F0123u = 2
16nu3
,
2 R8 h
eff

u4 2
0
d x 3 ,
d
x
d
x
R4 h
u4
d x 0 d x 1 ,
G2o R 4
(3.3)
where h and R are defined the same as those in Eq. (2.5). We would like to point out again
that the above decoupled gravity dual description of NCYM requires infinitesimal small
boost parameter = v.
A finite boost would imply that we cannot have the decoupled
8 This boosted configuration was also given in [27] for the discussion of light-like NCYM.
143
theory. This further implies that a Lorentz boost acting on a given decoupled NCYM will
render it undecoupled.
One can examine that the above gravity dual of NCYM is related to that of NCYM
resulting from the decoupling in a purely magnetic background (corresponding to
v = 0) again by a Galilean transformation defined in Eq. (2.6). If we calculate the
noncommutativity parameters, we have, as expected,
0i = 0,

23 = 2eff
,
(3.4)
where i = 1, 2, 3. This result is consistent with the Galilean transformation Eq. (2.6) as
discussed at the outset of this section. So we show that V-duality also works for NCYM.
We are not surprised by the infinitesimal boost requirement for NCYM given that for
NCOS discussed in the previous section. This is because the two are S-dual to each
other which can be explicitly checked from the gravity description for NCYM given in
this section and that for NCOS in the previous section. So the V-duality here is also a
consequence of S-duality. However, the physical reason behind is not explained up to this
point. We try to provide this before turning to the next section.
As we know that to have a decoupled NCOS, we need to have a background electric field
present from the world volume view. This electric field represents the presence of F-strings
inside D3 branes. The stable BPS configuration is the so-called non-threshold bound state
of (F, D3). On the other hand, for NCYM we need to have magnetic background which
represents the presence of D-strings inside D3 branes such that they form stable BPS nonthreshold (D1, D3) bound state. So for D3 branes, the presence of D-strings favors the
decoupled NCYM while that of F-strings favors the decoupled NCOS. We know that the
decoupling limit for one against that for other. One cannot form a consistent hybrid theory
which contains both NCOS and NCYM while at the same time it decouples the bulk closed
strings. The effect of a Lorentz boost, for example, acting on (F, D3) is to create D-strings
orthogonal to the F-strings in the bound state with its charge density proportional to the
Lorentz boost. Given the previous discussion, we know that if the charge density for Dstrings is comparable to that of F-strings in the bound state, we cannot have a decoupled
NCOS. 9 Only for infinitesimally small D-string charge, a decoupled NCOS has a chance
to exist. This is precisely what we achieved in the previous section and this is the physical
reason why only a Galilean transformation or V-duality is allowed for the decoupled theory.
The discussion for NCYM follows the same line but now the F-string charge density needs
to be infinitesimal small, implying an infinitesimal boost.
9 The best we can do is to infinitely boost a vanishing small electric background to end up with a new light-like
NCYM [26,27]. The same applies for infinitely boosting a vanishing small magnetic background. As discussed
in [26,27], the light-like NCYM, even though a noncommutative field theory, is qualitatively different from the
usual NCYM with space-like noncommutativity, therefore a new kind of noncommutative field theory. One can
say that NCOS, light-like NCYM and space-like NCYM result respectively from the decoupling of D3 branes in
time-like, light-like and space-like electric and magnetic backgrounds. We stress that the aforementioned infinite
boost is used just for defining a finite light-like background from a vanishing either time-like or space-like one
so that a decoupling limit for light-like NCYM can be addressed. This infinite boost is not directly related to
V-duality. We will address the related V-duality issue in [22].
144
4. Holography and NCOS

A holographic correspondence between NCYM and its gravity dual was proposed
in [28]. This was further extended in [14] to NCOS with pure timespace noncommutativity. It basically states that the radial profile of the on-shell closed string moduli (string
metric, NSNS B-field and dilaton) in the dual gravity description of NCYM or NCOS can
be derived using SeibergWitten relations [5] between fixed open string moduli (effective flat open string metric, noncommutative parameters and open string coupling) and the
closed string moduli, provided a simple ansatz for the running string tension as the function of energy scale is assumed. Given V-duality, the holographic correspondence [14,28]
is expected to hold for NCOS and NCYM discussed in this paper as well. In this section
we provide a direct demonstration of this correspondence for the case of NCOS as an independent and direct check for V-duality. The check for NCYM is even easier and we will
not repeat it here.
Physically the holographic correspondence proposed in [28] for NCYM and in [14] for
NCOS is a natural consequence of D-brane picture in the decoupling limit. We know that
there are two physically equivalent descriptions of D-branes, one is that of the open string
ending on the D-branes and the other is the closed string one. In general, from the closed
string perspective, D-branes themselves interact with bulk closed string modes and from
the open string perspective, the open string ending on the D-branes interacts with the same
bulk modes. However, in the so-called decoupling limit, these two descriptions describe
the dynamics of the same D-branes, therefore they must be physically equivalent. From
the open string perspective, the worldvolume must be flat since the bulk gravity (or closed
string) decouples. However, the closed string description of D-branes (i.e., the solitonic
profile of D-branes in the decoupling limit) gives a curved background (loosely called the
near-horizon geometry of D-brane). 10 It is the decoupled theory in a flat worldvolume
obtained from the open string perspective that gives a holographic description of the above
curved background.
A better description for a decoupled theory can be obtained when there are electric
and/or magnetic background (or NSNS B-background) on the D-branes. In terms of this
description, the D-brane metric seen by the end-points of the effective open strings does
not coincide with the closed string metric. And this description gives rise to coordinate
noncommutativity. In [5] Seiberg and Witten have established the general relations between
closed string moduli and the effective open string ones. In general, these relations hold only
to the leading order since a flat rigid geometry for D-branes is assumed there. However,
in the decoupling limit for D-branes, these relations become exact since the bulk modes,
which render the D-brane world volume curved, decouple. Originally SeibergWitten
relations are suggested in the first description of the D-branes (perturbative open-strings
ending on flat D-branes). The important recognition made in [14,28] is that these relations
10 The original spacetime is separated to two regions: one is occupied by the bulk closed strings and the other
describes the D-branes. The bulk closed strings cannot enter into the region of D-branes since its size is substringy.
This is the picture of the decoupling of D-branes from the bulk closed strings in terms of the closed string (or
bulk) description.
145
apply also to the closed string moduli in the dual gravity description. This provides a
better understanding of how the gravity dual describes noncommutative theories in a
holographical way. We can understand this as follows.
The open string description is non-gravitational (since the bulk closed strings decouple).
Therefore the worldvolume geometry should be independent of energy scale, so are the
metric and the noncommutativity parameters. 11 As we know, the energy scale in the open
string description is just the radial coordinate u in the closed string description of D-branes.
So for a fixed u, we can ask for the effective description of open string. This effective open
string coupling should be the same as before, since it is dimensionless, but the effective
string tension should be determined properly. We therefore expect that SeibergWitten
relations should hold between the closed string moduli at a fixed but arbitrary u and the
same open string moduli, but now for the effective open string with its yet unknown
effective open string tension. The success of this prescription depends crucially on how
. Given that the closed string moduli in
to determine the effective string tension or run
original SeibergWitten relations are constants and independent of the u coordinate, they
must correspond to those outside of the boundary of the gravity description. This fact helps
which was also discussed in [14,28] in a different
us to set the boundary condition for run
context. In other words, it must approach to its boundary value as u .
In the following, we derive the NSNS fields for the gravity dual of NCOS given in
Eq. (2.4) as an example. The SeibergWitten relations [5] between open string moduli
(G , , Go ) and closed string ones (g , B , gs ) in their original forms are

G = g (2 )2 Bg 1 B ,
(4.1)

1
= 2
(4.2)
,
g + 2 B A
and

Gs = G2o = gs
det G
det(g + 2 B)
1/2
.
(4.3)
In the above, ( )A denotes the anti-symmetric part and = 0, 1, 2, 3.

The open string metric G and the noncommutative parameters can be calculated
using the asymptotic values for closed string metric and NSNS B-field from Eq. (2.1) with
respect to the scaled coordinates x as
(1 v 2 ) 0 v 0
0
1 0 0
,
G =
(4.4)
v
0 1 0
0
0 0 1
and

01 = 2eff
,

12 = 2eff
v.
(4.5)
11 One may wonder if the noncommutative parameters can be so, since they are dimensionful. The correct
derivations given in [14,28] ensure this.
146
We know that the open string metric G is defined with respect to the string tension
1/(2 ). We can redefine a new open string metric G = G / with a finite tension
) since = / . We now have G = G = fixed as expected. The
1/(2eff
eff
eff
new metric G can be obtained from Eq. (4.1) simply by setting closed string metric
. In other words, we now work with the metric ds 2 / and string
g g/ and eff
. So is the boundary value for the . The question is how to obtain a
constant eff
run
eff
.
correct ansatz for run
Consider a type IIB F-string probe in a given background, its kinetic term is

1
(4.6)
d 2 x M x N gMN ,
2
with gMN the bulk metric. Now consider this string to probe the D3 brane background.
Recall that the NCOS tension is the same with respect to its metric in any direction. But
the story is different if we try to read the running tension from the closed string side.
The reason that we have a decoupled NCOS is because the near-critical electric force
balances the original string tension to end up with a finite tension if the original tension
is sent to infinity. This NCOS is now along x 1 direction. It is in this direction that the
open string metric is different from the asymptotic closed string one. One cannot use the
probe F-string to read the running tension for the effective open string along this direction.
Examining the open string metric G given in Eq. (4.4), we can see that the open string
metric G22 , G33 are the same as the corresponding asymptotic closed metric g22 , g33 with
respect to the scaled coordinates x . We should be able to read the running tension for the
effective open string using the probe along, for example, x 3 direction. In other words, from

= /g . From
1/(2 ) x 3 x 3 g33 (u) = g33 (u)/(2 ) x 3 x 3 , we have run
33

as u , i.e.,

2
1/2
2

Eq. (2.4), we have run = eff R h /u . It is easy to see that run eff
satisfying the boundary condition as expected. The physical picture above is: the closed

string has a string constant but moving in a curved background, for example, g33 (or eff

but in g33 /) while the open string has a flat metric G but with a running run (or G
).
with run
With the above understanding and given

= eff
R 2 h1/2 /u2 ,
run
(4.7)
we now present the holographic derivation for the NSNS fields in the gravity dual of
NCOS given in Eq. (2.4). Given that the gravity dual of NCOS in (2.4) is obtained from a
background with orthogonal electric and magnetic fields and the magnetic background is
brought about by a Lorentz boost, we expect that the only nonvanishing metric along the
brane directions are g00 , g11 , g02 = g20 , g22 = g33 . We also have possible nonvanishing
B01 and B12 . In the following, we use open string metric G = G /, this implies that
, it implies that we know g / =
we use closed string metric g/. Given the ansatz for run
22
/ as is evident from the above discussion. Now the in Eqs. (4.1)(4.3) is
g33 / = eff
run
. This further implies that the closed string metric used in SeibergWitten
replaced by run
relations should be g = g /g33 . With the above in mind, we set g00 = f1 , g11 = f2 ,
B = h , 2 B = h . Then
g02 = f3 and g22 = g33 = 1. For simplicity, we set 2run
01
1
2
run 12
147
SeibergWitten relations give, from the metric,

1+
h22
= 1,
f2
f3 = v,
f1
h21
= 1 v 2 ,
f2
f2
h21
f1 + f32
= 1,
(4.8)
and, from the noncommutative matrix,

u2
h1
=
.
R 2 h1/2
h21 f2 (f1 + f32 )
(4.9)
One can solve from the above to have

u2 1/2
h ,
h2 = 0.
(4.10)
R2
One can check that these solutions give the correct metric and NSNS B-field in Eq. (2.4).
B = h which gives 2 B = u4 /R 4 , the correct
For example, we have 2run
01
1
eff 01
answer. Using the above results, we can calculate the effective closed string coupling e ,
B)]1/2 , as
from e = G2o [ det(g + 2run
f1 = h v 2 ,
f2 = h,
e = G2o h1/2 ,
f3 = v,
h1 =
(4.11)
again the correct answer. We note that the running string tension for NCOS in our case,
Eq. (4.7), as a function of the radial coordinate u is the same as that given in Ref. [14]
for NCOS with pure spacetime noncommutativity. This means that the NCOSs related
by V-duality have the same running string tension. Or the latter remains unchanged under
V-duality. This may be viewed as an additional evidence for V-duality.
5. Conclusions and discussions

Before concluding this paper, let us now address an issue regarding the supporting
argument for the equivalence of the noncommutative theories that arise from decoupling
D-branes in Lorentz-boost related backgrounds. This issue arises as follows: in the usual
perturbative open string description of D-branes, from the outset, one assumes that Dbrane are rigid and flat. This is known to be not exactly true, since the coupling to bulk
closed strings will render the D-branes curved. In this sense, the Poincare symmetry along
the brane directions in this description should be considered true only approximately.
However, in a limit in which the bulk closed strings decouple, this symmetry is expected
to become exact. Moreover, the decoupling for a noncommutative theory (either NCOS
or NCYM for the purpose of this paper) in general requires a D-brane worldvolume
electric and/or magnetic background field. The presence of such a background field breaks
the worldvolume Lorentz symmetry, but this breakdown is only spontaneous. Namely,
if we simultaneously Lorentz transform all backgrounds, including the electromagnetic
one, we end up with a physically equivalent situation. This can be explicitly seen in our
present study: in the gravity dual description in Sections 2 and 3 it is obvious that the
Lorentz boost on a relevant D-brane bound state solution before decoupling leads to a
physically equivalent solution. Thus, given that the worldvolume Lorentz symmetry for the
148
parent open string theory should be exact in decoupling limits and that the introduction of
backgrounds breaks the symmetry only spontaneously, we expect that the noncommutative
theories of Dp branes resulting from decoupling in Lorentz-boost related background fields
are physically equivalent.
In conclusion, we test the V-duality conjecture proposed in [14] for NCOS using its
gravity dual description. We also extend the V-duality for NCYM and find that this seems
to be true in general. The implication of this V-duality is that one can no longer use a
Lorentz transformation other than a Galilean one, for example, to choose a particular
frame in performing calculations in NCOS and NCYM. We present an explanation for
the holographic correspondence given in [14,28] for noncommutative theories, and give a
holographic derivation for the gravity dual of NCOS discussed in this paper. In addition,
our holographic derivation shows that the NCOSs (or NCYMs) related by V-duality
share the same running string tension as a function of the radial coordinate (or energy
scale).
The Galilean nature of this V-duality is, in a sense, implied already by the noncommutative relation [x , x ] = i , which implies spacetime nonlocality. This in turn implies
that we have action at distance. In other words, we do not have a limit on the speed of
light, 12 i.e., we are dealing with Newtonian-like physics. With the V-duality discussed
in this paper and the above simple physical implication behind the noncommutative relations, one can easily understand the recent finding in [29,30] that solitons can travel faster
than the speed of light in NCYM and also the infinitely large speed of light as discussed
in [3133]. We will elaborate this and other related points further in the forthcoming paper [22].
Acknowledgements
The work of RGC was supported in part by the Japan Society for Promotion of Science
and in part by Grants-in-Aid for Scientific Research Nos. 99020 and 12640270, and by
a Grant-in-Aid on the Priority Area: Supersymmetry and Unified Theory of Elementary
Particles. JXL acknowledges the support of U.S. Department of Energy. The research
of YSW was supported in part by the U.S. National Science Foundation under Grant
No. PHY-9907701.
References
[1] P.M. Ho, Y.S. Wu, Noncommutative geometry and D-branes, Phys. Lett. B 398 (1997) 52, hepth/9611233.
[2] A. Connes, M. Douglas, A. Schwartz, Noncommutative geometry and matrix theory,
JHEP 9802 (1998) 003, hep-th/9711162.
12 There are some subtleties involved when the boost is not orthogonal to the spacetime noncommutative
directions. We will address this in detail in [22].
149
[3] M.R. Douglas, C. Hull, D-branes and the noncommutative torus, JHEP 9802 (1998) 008, hepth/9711165.
[4] P.M. Ho, Y.S. Wu, Noncommutative gauge theories in Matrix theory, Phys. Rev. D 58 (1998)
026006, hep-th/9712201.
[5] N. Seiberg, E. Witten, String theory and noncommutative geometry, JHEP 9909 (1999) 032,
hep-th/9908142.
[6] A. Hashimoto, N. Itzhaki, Non-commutative YangMills and the AdS/CFT correspondence,
Phys. Lett. B 465 (1999) 142, hep-th/9907166.
[7] J. Maldacena, J. Russo, Large N limit of noncommutative gauge theories, JHEP 9909 (1999)
025, hep-th/9908134.
[8] M. Alishahiha, Y. Oz, M. Sheikh-Jabbari, Supergravity and large N noncommutative field
theories, JHEP 9911 (1999) 007, hep-th/9909215.
[9] N. Seiberg, L. Susskind, N. Toumbas, Strings in background electric field space/time
noncommutativity and a new noncritical string theory, JHEP 0006 (2000) 021, hep-th/0005040.
[10] R. Gopakumar, J. Maldacena, S. Minwalla, A. Strominger, S-duality and noncommutative
gauge theory, JHEP 0006 (2000) 036, hep-th/0005048.
[11] J. Barbon, E. Rabinovici, Stringy fuzziness as the custodian of timespace noncommutativity,
Phys. Lett. B 486 (2000) 202, hep-th/0005073.
[12] T. Harmark, Supergravity and spacetime noncommutative open string theory, JHEP 0007
(2000) 043, hep-th/0006023.
[13] I. Klebanov, J. Maldacena, (1 + 1)-dimensional NCOS and its U (N) gauge theory dual, hepth/0006085.
[14] G.-H. Chen, Y.-S. Wu, Comments on noncommutative open string theory: V-duality and
holography, hep-th/0006013.
[15] J.X. Lu, S. Roy, H. Singh, ((F,D1),D3) bound state, S-duality and non-commutative open
string/YangMills theory, JHEP 0009 (2000) 020, hep-th/0006193.
[16] J. Russo, M. Sheikh-Jabbari, On noncommutative open string theories, JHEP 0007 (2000) 052,
hep-th/0006202.
[17] S.-J. Rey, R. von Unge, S-duality, noncritical open string and noncommutative gauge theory,
hep-th/0007089.
[18] R.-G. Cai, N. Ohta, (F1,D1,D3) bound state, its scaling limits and SL(2, Z) duality, Prog.
Theor. Phys. 104 (2000) 1073, hep-th/0007106.
[19] J.X. Lu, S. Roy, H. Singh, SL(2, Z) duality and 4-dimensional noncommutative theories, hepth/0007168, to appear in NPB.
[20] N. Seiberg, L. Susskind, N. Toumbas, Space/time noncommutativity and causality, JHEP 0006
(2000) 044, hep-th/0005015.
[21] J. Gomis, T. Mehen, Spacetime noncommutative field theories and unitarity, Nucl. Phys. B 591
(2000) 265, hep-th/0005129.
[22] R.-G. Cai, J.X. Lu, Y.-S. Wu, in preparation.
[23] J.G. Russo, A.A. Tseytlin, Waves, boosted branes and BPS states in M-theory, Nucl. Phys.
B 490 (1997) 121.
[24] J.X. Lu, S. Roy, Non-threshold (F,Dp) bound states, Nucl. Phys. B 560 (1999) 181, hepth/9904129.
[25] J. Breckenridge, G. Michaud, R. Myers, More D-brane bound states, Phys. Rev. D 55 (1997)
6438.
[26] O. Aharony, J. Gomis, T. Mehen, On theories with light-light noncommutativity, JHEP 0009
(2000) 023, hep-th/0006236.
[27] M. Alishahiha, Y. Oz, J. Russo, Supergravity and light-like non-commutativity, JHEP 0009
(2000) 002, hep-th/0007215.
[28] M. Li, Y.-S. Wu, Holography and noncommutative YangMills, Phys. Rev. Lett. 84 (2000)
2084, hep-th/9909085.
150
[29] A. Hashimoto, N. Itzhaki, Traveling faster than the speed of light in noncommutative geometry,
hep-th/0012093.
[30] D. Bak, K. Lee, J.-H. Park, Noncommutative vortex solitons, hep-th/0011099.
[31] U. Danielsson, A. Goijosa, M. Kruczenski, Newtonian gravitons and D-brane collective
coordinates in Wound string theory, hep-th/0012183.
[32] U. Danielsson, A. Goijosa, M. Kruczenski, IIA/B wound and wrapped, JHEP 0010 (2000) 020,
hep-th/0009182.
[33] J. Gomis, H. Ooguri, Non-relativistic closed string theory, hep-th/0009181.

EDM constraints in supersymmetric theories

S. Abel a,b , S. Khalil a,c , O. Lebedev a
a Centre for Theoretical Physics, University of Sussex, Brighton BN1 9QJ, UK
b IPPP, University of Durham, South Rd., Durham DH1 3LE, UK
c Ain Shams University, Faculty of Science, Cairo, 11566, Egypt
Abstract
We systematically analyze constraints on supersymmetric theories imposed by the experimental
bounds on the electron, neutron, and mercury electric dipole moments. We critically reappraise the
known mechanisms to suppress the EDMs and conclude that only the scenarios with approximate
CP-symmetry or flavour-off-diagonal CP violation remain attractive after the addition of the mercury
EDM constraint. 2001 Elsevier Science B.V. All rights reserved.
PACS: 11.30.Er; 13.40.Em
1. Introduction
There are a number of reasons to suspect that there are additional sources of CP violation
beyond those of the Standard Model (given by and KM ). The most compelling one
is that the SM is unable to explain the cosmological baryon asymmetry of our universe.
Also, the Standard Model is very unlikely to be the ultimate theory of nature. Most
extensions of the SM bring in new sources of CP-violation. In particular, the most attractive
one the Minimal Supersymmetric Standard Model (MSSM) allows for new sources of
CP violation in both supersymmetry-breaking and supersymmetry-conserving sectors, and
there are no compelling arguments for them to be zero. In addition, the improving precision
in the measurements of the CP-observables such as ACP (B Ks ) [1] may soon reveal
deviations from the SM predictions.
The most stringent constraints on models with additional sources of CP violation come
from continued efforts to measure the electric dipole moments (EDM) of the neutron [2],
electron [3], and mercury atom [4]
dn < 6.3 1026 e cm (90% CL),
E-mail address: kafg9@pact.cpes.susx.ac.uk (O. Lebedev).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 2 3 3 - 4
152
S. Abel et al. / Nuclear Physics B 606 (2001) 151182
de < 4.3 1027 e cm,

dHg < 2.1 1028 e cm.
(1)
With the expected improvements in experimental precision, the EDM is likely to be one
of the most important tests for physics beyond the Standard Model for some time to come,
and EDMs will remain a difficult hurdle for supersymmetric theories if they are to allow
sufficient baryogenesis. Indeed it is remarkable that the SM contribution to the EDM of the
neutron is of order 1030 e cm, whereas the generic supersymmetric value is 1022 e cm.
In this paper we analyze neutron, electron, and mercury EDMs in the context of R-parity
conserving supersymmetric theories. In particular, we reconsider the known mechanisms
to suppress EDMs in light of the recently reported bound on the mercury EDM [4].
These include SUSY models with small CP phases, models with heavy sfermions, the
cancellation scenario, and models with flavour off-diagonal CP violation. We also study to
what extent different scenarios rely on assumptions about the neutron structure, i.e., chiral
quark model vs parton model.
The paper is organized as follows. In Section 2 we present general formulae for the
EDMs and discuss their model-dependence. In Section 3 we define our supersymmetric
framework and present all relevant supersymmetric contributions to the EDMs. In
particular, we analyze the importance of the two-loop BarrZee and Weinberg type EDM
contributions. Section 4 is devoted to the study of the EDM suppression mechanisms.
First we consider in detail the canonical scenarios: suppression due to small CP phases
or heavy sfermions. Second, in the context of the cancellation scenario, we analyze the
possibility of the EDM cancellations in two classes of models: mSUGRA-like models with
nontrivial gaugino phases and D-brane models. Finally, we discuss the EDM suppression
in models with flavour-off-diagonal CP violation. In addition, we present new modelindependent bounds on the sfermion mass insertions imposed by the electron, neutron,
and mercury electric dipole moments. In Section 5 we overview and discuss our results.
2. Electron, neutron, and mercury electric dipole moments

Let us first summarize the contributions to the three most significant EDMs, beginning
with the most reliable, the electron EDM.
2.1. Electron EDM
The electron EDM is defined by the effective CP-violating interaction
i
L = de e 5 eF ,
(2)
2
where F is the electromagnetic field strength. The experimental bound on the electron
EDM is derived from the electric dipole moment of the thallium atom and is given by [3]
de < 4 1027 e cm.
(3)
153
In supersymmetric models, the electron EDM arises due to CP-violating 1-loop diagrams
with the chargino and neutralino exchange:
+
de = de + de .
(4)
Since the EEDM calculation involves little uncertainty it allows to extract reliable bounds
on the CP-violating SUSY phases.
2.2. Neutron EDM
The neutron EDM has contributions from a number of CP-violating operators involving
quarks, gluons, and photons. The most important ones include the electric and chromoelectric dipole operators, and the Weinberg three-gluon operator:
i
i
L = dqE q 5 qF dqC q 5 T a qGa
2
2
1 G
d fabc Ga Gb Gc " ,
(5)
6
where Ga is the gluon field strength, T a and fabc are the SU(3) generators and group
structure coefficients, respectively. Given these operators, it is however a nontrivial task
to evaluate the neutron EDM since assumptions about the neutron internal structure are
necessary. In what follows we will study two models, namely the quark chiral model
and the quark parton model. Neither of these models is sufficiently reliable by itself [5],
however a power of the combined analysis should provide an insight into implications of
the bound on the neutron EDM and in particular comparing them gives some indication of
the importance of these systematic errors in, for example, cancellations. A better justified
approach to the neutron EDM based on the QCD sum rules has appeared in [8] and earlier
work [9,12]. We note that in any case the NEDM calculations involve uncertain hadronic
parameters such as the quark masses and thus these calculations have a status of estimates.
The major conclusions of the present work are independent of the specifics of the neutron
model.
(i) Chiral quark model. This is a nonrelativistic model which relates the neutron EDM
to the EDMs of the valence quarks with the help of the SU(6) coefficients:
4
1
dd du .
3
3
The quark EDMs can be estimated via Naive Dimensional Analysis [6] as
dn =
dq = E dqE + C
e C
e G
d + G
d ,
4 q
4
(6)
(7)
where the QCD correction factors are given by E = 1.53, C G 3.4, and
1.19 GeV is the chiral symmetry breaking scale. We use the numerical values for these
coefficients as given in [7]. The parameters C,G involve considerable uncertainties
stemming from the fact that the strong coupling constant at low energies is unknown.
Another weak side of the model is that it neglects the sea quark contributions which play
an important role in the nucleon spin structure.
154
The supersymmetric contributions to the dipole moments of the individual quarks result
from the 1-loop gluino, chargino, neutralino exchange diagrams
g(E,C)
dqE,C = dq
+ dq
+ (E,C)
+ dq
0 (E,C)
(8)
and from the 2-loop gluinoquarksquark diagrams which generate d G .

(ii) Parton quark model. This model is based on the isospin symmetry and known
contributions of different quarks to the spin of the proton [10]. The quantities q defined as
5 q|n = q S , where S is the neutron spin, are related by the isospin symmetry
n| 12 q
to the quantities (q )p which are measured in the deep inelastic scattering (and other)
experiments, i.e., u = (d )p , d = (u )p , and s = (s )p . To be exact, the neutron
EDM depends on the (yet unknown) tensor charges rather than these axial charges. The
main assumption of the model is that the quark contributions to the NEDM are weighted
by the same factors i , i.e., [10]

dn = E d ddE + u duE + s dsE .
(9)
In our numerical analysis we use the following values for these quantities d = 0.746,
u = 0.508, and s = 0.226 as they appear in the analysis of Ref. [11]. As before, we
have
g(E)
dqE = dq
+ dq
+ (E)
+ dq
0 (E)
(10)
The major difference from the chiral quark model is a large strange quark contribution
(which is likely to be an overestimate [5]). In particular, due to the large strange and charm
quark masses, the strange quark contribution dominates in most regions of the parameter
space. This leads to considerable numerical differences between the predictions of the two
models.
The current experimental limit on the neutron EDM is [2]
dn < 6.3 1026 e cm.
(11)
2.3. Mercury EDM

The EDM of the mercury atom results mostly from T-odd nuclear forces in the mercury
nucleus, which induce the effective interaction of the type (I )(r) between the electron
and the nucleus of spin I [5]. In turn, the T-odd nuclear forces arise due to the effective
4-fermion interaction pp
ni
5 n. It has been argued [5] that the mercury EDM is primarily
sensitive to the chromoelectric dipole moments of the quarks and the limit [4]
dHg < 2.1 1028 e cm
can be translated into
C

d d C 0.012d C /gs < 7 1027 cm,
u
s
d
(12)
(13)
where gs is the strong coupling constant. As in the parton neutron model, there is a
considerable strange quark contribution. The relative coefficients of the quark contributions
in (13) are known better than those for the neutron, however the overall normalization is
still not free of uncertainties [12].
155
3. The EDMs in SUSY models

We will study supersymmetric models with the following high energy scale soft breaking
potential
+ Ad Yijd H1 qLi dRj

VSB = m20 (BH1 H2 + h.c.) + Al Yijl H1 lLi eRj

1
a W
a + m1 B
B
,
Au Yiju H2 qLi u Rj + h.c. + m3 g g + m2 W
(14)
2
where denotes all the scalars of the theory. We generally allow for Al = Au = Ad
as well as nonuniversal gaugino and scalar masses, which is important for the analysis
of D-brane models. The , B, A , and mi parameters can be complex, however two of
their phases can be eliminated by the U (1)R and U (1)PQ transformations under which
these parameters behave as spurions. The PecceiQuinn transformation acts on the Higgs
doublets and the right-handed superfields in such a way that all the interactions but
those which mix the two doublets are invariant. The PecceiQuinn charges are QPQ () =
QPQ (B), QPQ (A) = QPQ (mi ) = 0. The U (1)R transforms the Grassmann variable
ei and the fields in such a way that the integral of the superpotential over the
Grassmann variables is invariant, i.e., the U (1)R charge of the superpotential is 2. As a
result, QR (B) = QR () 2, QR (A) = QR (mi ) = 2. The six physical CP-phases of
the theory are invariant under both U (1)R and U (1)PQ , and can be chosen as

Arg Ad mi , Arg (B) A ,
(15)
where i = 1, 2, 3 and = d, u, l. All other CP-phases can be expressed as their linear
combinations. If the A-terms are universal, there are four physical phases Arg(A mi ),
Arg((B) A).
It is customary to choose the phase convention in which the Higgs potential parameter
B is real. In this case, the physical phases become Arg(Ad mi ) and Arg(A ). If
universality is assumed, the number of physical phases reduces to two. In what follows
we will set m2 to be real by a U (1)R rotation.
Nonuniversality will play a crucial role in the D-brane and flavour models analysis, but
otherwise does not lead to different conclusions for the models we study. Thus we will
assume universal A-terms and gaugino masses unless otherwise specified.
In what follows we use tan , m0 , A, mi as input parameters and obtain low energy
quantities via the MSSM renormalization group equations (RGE). We also assume
radiative electroweak symmetry breaking, i.e., that the magnitude of the parameter is
given (at tree level) by
||2 =
m2H1 m2H2 tan2

tan2 1
1 2
m .
2 Z
(16)
The phase of is an input parameter and is RG-invariant. Our numerical results are
sensitive to the quark masses which we fix at the MZ scale to be: mui = (0.005,
1.40, 165) GeV and mdi = (0.010, 0.194, 3.54) GeV. The light quark masses are poorly
determined (in fact mu = 0 is not excluded) which results in the uncertainties of the EDM
156
Fig. 1. Leading SUSY contributions to the EDMs. The photon and gluon lines are to be attached to
the loop in all possible ways.
normalization; for definiteness, we have chosen the light quark masses as they appear in [7].
The GUT scale is assumed to be 2 1016 GeV.
It is well known that O(1) supersymmetric CP phases generally lead to unacceptably
large electric dipole moments which constitutes the SUSY CP problem. In this paper we
consider different mechanisms for suppressing EDMs and analyze them in detail.
3.1. Leading SUSY contributions to the EDMs
In this subsection we list formulae for individual supersymmetric contributions to the
EDMs due to the Feynman diagrams in Fig. 1. In our presentation we follow the work of
Ibrahim and Nath [7].
Neglecting the flavour mixing, the electromagnetic contributions to the fermion EDMs
are given by [7]:
g(E)
dq
/e =
m2
2
2s 1k mg
g
Im q
Q
B
,
q
3
Mq2
Mq2
k=1
du
+ (E)
/e =
2
2

em
2
4 sin W
Im(uik )
k=1 i=1
m +
i
M 2
dk

m2 +
m2 +
i
i
+ (Qu Qd ) A
,
Qd B
2
M
M 2
dk
+ (E)
dd
/e =
2
2

em
4 sin2 W
k=1 i=1
Im(dik )
dk
m +
i
Mu2 k

m2 +
m2 +
i
i
+ (Qd Qu ) A
,
Qu B
2
Mu k
Mu2 k
+
de /e
em
4 sin2 W
2 m +

i=1
m2
m2 +
i
,
Im(ei ) A
m2
0 (E)
df
/e
157
m2 0
i
Qf B
.
M 2
(17)
em
2
4

4 sin2 W
k=1 i=1
Im(f ik )
m 0
i
M 2
fk
fk
Here
q1k = ei3 Dq2k Dq1k

,
(18)
with 3 being the gluino phase and Dq defined by Dq Mq2 Dq = diag(Mq2 , Mq2 ). The
1
sfermion mass matrix M 2 is given by

ML 2 + mf 2 + Mz2 12 Qf sin2 W cos 2
Mf2 =
mf (Af Rf )
mf (Af Rf )
2 + m 2 + M 2 Q sin2 cos 2
MR
f
W
z f

,
where Rf = cot (tan ) for I3 = 1/2 (1/2). The chargino vertex f ik is defined as

uik = u Vi2 Dd1k Ui1
Dd1k d Ui2
Dd2k ,

dik = d Ui2
(19)
Du1k Vi1 Du1k
u Vi2 Du2k
and analogously for the electron; here U and V are the unitary matrices diagonalizing the
chargino mass matrix: U M + V 1 = diag(m + , m + ). The quantities f are the Yukawa
1
2
couplings
u =
mu
2 mW sin
md,e
d,e =
.
2 mW cos
The neutralino vertex f ik is given by

f ik = 2 tan W (Qf I3f )X1i + I3f X2i Df 1k f Xbi Df 2k

2 tan W Qf X1i Df 2k f Xbi Df 1k ,
(20)
(21)
where I3 is the third component of the isospin, b = 3(4) for I3 = 1/2(1/2), and X is
the unitary matrix diagonalizing the neutralino mass matrix: XT M 0 X = diag(m 0 , m 0 ,
1
2
m 0 , m 0 ). In our convention the mass matrix eigenvalues are positive and ordered as
3
4
m 0 > m 0 > (this holds for all mass matrices in the paper). The loop functions A(r),
1
2
B(r), and C(r) are defined by

1
2 ln r
A(r) =
,
3
r
+
1r
2(1 r)2

1
2r ln r
B(r) =
1
+
r
+
,
2(r 1)2
1r

2r ln r 18 ln r
1
10r 26 +
.
C(r) =
(22)
6(r 1)2
1r
1r
The chromoelectric contributions to the quark EDMs are given by
g(C)
dq
m2
2
gs s 1k mg
g
,
Im q
C
2
2
4
M
M
q k
q k
k=1
158
+
dq (C)
dq
0 (C)
2
2
2
m + m +
g 2 gs
i
i
,
=
Im(qik ) 2 B
2
16 2
M
M
q k
q k
k=1 i=1
2
4
2
m 0 m 0
gs g 2
i
i
.
Im(
)
B
qik
2
2
16 2
M
M
q k
q k
k=1 i=1
(23)
Finally, the contribution to the Weinberg operator [13] from the two-loop gluino-topstop and gluino-bottom-sbottom diagrams reads
3
z1 z2
gs
d G = 3s mt
(24)
Im(t12 )
H(z1 , z2 , zt ) + (t b),
4
m3g
Mt
mt 2
where zi = ( m i )2 , zt = ( m
) . The two-loop function H (z1 , z2 , zt ) is given by [14]
g
1
H (z1 , z2 , zt ) =
2
1
1
dx
1
dy x(1 x)u
du
0
N1 N2
,
D4
(25)
where

N1 = u(1 x) + zt x(1 x)(1 u) 2ux z1 y + z2 (1 y) ,
1
N2 = (1 x)2 (1 u)2 + u2 x 2 (1 u)2 ,
9

D = u(1 x) + zt x(1 x)(1 u) + ux z1 y + z2 (1 y) .
(26)
The numerical behaviour of this function was studied in [14]. We emphasize that the bquark contribution is significant and often exceeds the top one.
Before we proceed to the discussion of the EDM suppression mechanisms, let us
consider the effect of other potentially nonnegligible two-loop contributions.
3.2. BarrZee type EDM contributions
In view of considerable recent interest in the subject we will consider the two-loop
BarrZee type contributions separately. We will follow the work of Chang, Keung, and
Pilaftsis [15].
In Ref. [16] Barr and Zee have presented two-loop Higgs-mediated EDM contributions
which can be competitive with the Weinberg three-gluon operator. A supersymmetric
version of the BarrZee graphs (Fig. 2) was studied in [15]. In what follows we will analyze
only the leading contributions to the EDMs presented in [15]. The EDMs arise due to CPviolating couplings of the (s)fermions to the CP-odd Higgs boson a0 . The EDM and the
CEDM of a light fermion f are computed to be [15]
M2
M 2
3em Rf mf
q 1
q 2
2
dfE /e = Qf
Q
F
F
,
q q
32 3 Ma2
Ma2
Ma2
q=t,b
dfC
gs s Rf mf
=
64 3 Ma2

q=t,b
M2
M 2
q 1
q 2
F
,
q F
Ma2
Ma2
(27)
159
Fig. 2. BarrZee type contributions to the EDMs.
where Rf = cot (tan ) for I3 = 1/2 (1/2) and the two-loop function F (z) is
1
F (z) =
dx
0

x(1 x)
x(1 x)
ln
.
z x(1 x)
z
(28)
The CP-violating couplings t,b are given by

t =
sin 2tmt Im(eit )
,
2v 2 sin2
sin 2b mb Im(Ab eib )
b =
,
2v 2 sin cos
(29)
with t,b being the standard stop and sbottom mixing angles; q = Arg(Aq Rq ) and
v = 174 GeV. Note the difference from [15] in the definitions of and v, also see the
corresponding Erratum.
In our numerical analysis, besides assuming the radiative electroweak symmetry
breaking, we use the following tree level mass of the CP-odd Higgs [17]
Ma2 = m2H1 + m2H2 + 2||2 ,
(30)
which is a function of tan and other GUT scale input parameters. Strictly speaking, this
formula is valid for a CP-conserving case, however the EDMs are not very sensitive to the
exact value of Ma2 and an inclusion of loop corrections and CP-phases does not alter our
results.
In Fig. 4 we present a typical BarrZee type EDM behaviour as a function of tan for
A = /2, = 0. The other parameters are fixed to be m0 = m1/2 = A = 200 GeV. The
values of tan beyond 42 are not displayed for this parameter set since the CP-odd scalar
becomes massless in this region and the pattern of the EW symmetry breaking becomes
unacceptable. We observe that generally these EDM contributions by themselves do not
impose significant constraints on the GUT scale A-term phases even at large tan ; as can
be seen from the plot, the BarrZee contributions typically are one-two orders of magnitude
below the experimental limit. One of the reasons for this is that the third generation A-term
phases reduce by an order of magnitude due to the RG evolution at large tan . Also the
value of the parameter is typically below 500 GeV owing to the radiative electroweak
symmetry breaking. Other factors which distinguish our results from those of [15] are the
160
Fig. 3. Embedding the SM gauge group within different sets of D-branes. The extra Dq brane (52 ) is
marked by a cross.
Fig. 4. BarrZee and Weinberg operator induced EDMs as a function of tan . 1 electron
BarrZee EDM, 2 neutron BarrZee EDM, 3 neutron EDM due to the Weinberg operator.
Here m0 = m1/2 = A = 200 GeV, = 0, A = /2.
imposition of Eq. (30) and utilization of the chiral quark neutron model. 1 For comparison,
in Fig. 4 we provide the contribution of the Weinberg operator which also arises at the
two-loop level.
The constraints become more restrictive for larger A-terms ( 3m0 ) and larger m0 /m3
ratios. In Fig. 5 we set m0 = 500 GeV, m1/2 = 200 GeV, and A = 600 GeV. In this case
1 We observe similar behaviour in the parton quark model.
161
Fig. 5. BarrZee and Weinberg operator induced EDMs as a function of tan . 1 electron
BarrZee EDM, 2 neutron BarrZee EDM, 3 neutron EDM due to the Weinberg operator.
Here m0 = 500 GeV, m1/2 = 200 GeV, A = 600 GeV, = 0, A = /2.
the A-term phase is not diluted as much as before and for some parameters the BarrZee
EDMs can be close to the experimental limit. A similar effect can be achieved in models
with non-universal gaugino masses by introducing a O(1) gluino phase. With the parameter
set of Fig. 5 the CP-odd Higgs becomes unacceptably light around tan 36.
From the point of view of low energy theory, the BarrZee type contributions can provide
useful constraints on the phases of the third generation A-terms [15,18]. One can imagine
a situation in which the first two generation CP-violating effects are suppressed (as in the
decoupling scenario), then the EDMs would constrain the third generation phases. We find
however that typically the Weinberg three-gluon operator is considerably more sensitive to
such phases and provides more severe constraints even at large tan . 2 For example, for
the parameters of Figs. 4 and 5 the contribution of the Weinberg operator exceeds that of
the BarrZee type graphs by one-two orders of magnitude.
Finally, the Z- and W-mediated BarrZee type graphs have been analyzed in [19]
and found to be significantly smaller than those considered above. A number of other
subleading two loop contributions such as the gluino CEDM-induced quark EDMs, etc.
have been studied in [20].
Although taken into account, this entire class of diagrams is numerically unimportant in
our analysis.
2 The Weinberg operator contribution can be suppressed by increasing the gluino mass. However, in mSUGRA
the BarrZee type contributions will also get a suppression factor due to the RG dilution of the phases of the
third generation A-terms.
162
4. Suppression of the EDMs in SUSY models

4.1. Small CP-phases
For a light (below 1 TeV) supersymmetric spectrum, the SUSY CP phases have to be
small in order to satisfy the experimental EDM bounds (unless EDM cancellations occur).
In Figs. 68 we illustrate the EDMs behaviour as a function of the CP-phases in the
mSUGRA-type models, where we have set m0 = m1/2 = A = 200 GeV. At low tan ,
the EDM constraints impose the following bounds (at the GUT scale):
A 102 101 ,
103 102 ,
gaug. 102 .
(31)
For tan > 3, these bounds become even stricter (for and gaug. the bounds are roughly
inversely proportional to tan ). We note that A is less constrained than and gaug. .
There are two reasons for that: first, A is reduced by the RG running from the GUT scale
d )
down to the electroweak scale and, second, the phase of the (11
LR mass insertion which
gives the dominant contribution to the EDMs is more sensitive to and gaug. due to
|A| < tan .
SUSY models with small CP-phases can be motivated by the approximate CP
symmetry [21]. The well established experimental results exhibit small degree of CP
violation. Thus it is conceivable that all existing CP-phases are small, including the CKM
phase. The smallness of CP violation can be explained, for example, by a small ratio of
Fig. 6. EDMs as a function of A . 1 electron, 2 neutron (chiral model), 3 neutron (parton

model), 4 mercury. The experimental limit is given by the horizontal line. Here tan = 3,
m0 = m1/2 = A = 200 GeV.
163
Fig. 7. EDMs as a function of . 1 electron, 2 neutron (chiral model), 3 mercury,

4 neutron (parton model). The experimental limit is given by the horizontal line. Here tan = 3,
m0 = m1/2 = A = 200 GeV.
Fig. 8. EDMs as a function of the gluino phase 3 . 1 electron, 2 neutron (chiral model), 3
neutron (parton model), 4 mercury. The experimental limit is given by the horizontal line. Here
tan = 3, m0 = m1/2 = A = 200 GeV.
the scale at which the CP symmetry gets broken spontaneously and the scale at which it is
communicated to the observable sector.
Currently the CKM phase is consistent with zero and it could be supersymmetry that is
responsible for the observed values of and . This does not require large supersymmetric
164
phases. In fact in models with non-universal A-terms and can even be saturated with
d )
d
O(102 ) phase of the mass insertion (12
LR [22]. In this case, |(12 )LR | is required to be
3
O(10 ) which naturally appears in models with matrix-factorizable A-terms of the form
B Y C, where B and C are flavour matrices. The EDM bounds serve to constrain the
d
)LR is to use
flavour structure of B and C. Another possibility to produce the desired (12
asymmetric A-term textures in string-motivated models (where the standard supergravity
relation A ij = Aij Yij is assumed).
Encouragingly, the phase of the -term of order 102 may be sufficient to produce the
observed baryon asymmetry [23], which is in marginal agreement with the EDM bounds.
The hypothesis of the approximate CP symmetry is currently being tested in the B physics
experiments where the Standard Model predicts large CP-asymmetries. It is noteworthy
that the smallness of the CP-phases in this picture does not constitute fine-tuning according
to the t Hoofts criterion [24] since setting them to zero would increase the symmetry of
the theory.
We remark that small CP-phases may also arise due to the dynamics of the system. For
instance, in weakly coupled heterotic string models, small soft and CKM phases arise when
the T-moduli get VEVs close to the edge of the fundamental domain which is often the
case [25]. This mechanism however relies on the assumption that the dilaton has a real
VEV, so this model as it stands does not solve the SUSY CP problem. Nevertheless, it may
serve as a step toward a consistent string model with naturally small CP phases.
Note that EDMs constrain only the physical phases (15). One can imagine a situation
when the individual phases are O(1) whereas the physical ones are small. This occurs for
example in gauge mediated SUSY breaking models, see [21] and references therein.
4.2. Heavy SUSY scalars
This possibility is based on the decoupling of heavy supersymmetric particles. Even if
one allows O(1) CP violating phases, their effect will be negligible if the SUSY spectrum
is sufficiently heavy [26]. Generally, SUSY fermions are required to be lighter than the
SUSY scalars by, for example, cosmological considerations. So the decoupling scenario
can be implemented with heavy sfermions only. Here the SUSY contributions to the EDMs
are suppressed even with maximal SUSY phases because the squarks in the loop are very
heavy and the mixing angles are small.
In Fig. 9 we display the EDMs as functions of the universal scalar mass parameter m0 for
the mSUGRA model with maximal CP-phases = A = /2 and m1/2 = A = 200 GeV.
We observe that all EDM constraints except for that of the electron require m0 to be around
5 TeV or more. The mercury constraint is the strongest one and requires
(m0 )decoupl. 10 TeV.
(32)
This leads to a serious fine-tuning problem. Recall that one of the primary motivations for
supersymmetry was a solution to the naturalness problem. Certainly this motivation will
be entirely lost if a SUSY model reintroduces the same problem in a different sector, i.e.,
for example a large hierarchy between the scalar mass and the electroweak scale.
165
Fig. 9. EDMs as a function of the universal mass parameter m0 . 1 electron, 2 neutron (chiral
model), 3 neutron (parton model), 4 mercury. The experimental limit is given by the horizontal
line. Here tan = 3, m1/2 = A = 200 GeV, = A = /2.
The degree of fine-tuning can be quantified as follows. The Z boson mass is determined
at tree level by
m2H1 m2H2 tan2
1 2
mZ =
2 .
2
tan2 1
One can define the sensitivity coefficients [27,28]

ln m2Z
,

ci
ln ai2
(33)
(34)
where ai are the high energy scale input parameters such as m1/2 , m0 , etc. Note that is
treated here as an independent input parameter. A value of ci much greater than one would
indicate a large degree of fine-tuning. The Higgs mass parameters are quite sensitive to m0 ,
so for m0 = 10 TeV we find
cm0 5000
(35)
for the parameters of Fig. 9. For the universal scalar mass of 5 (3) TeV the sensitivity
coefficient reduces to 1300 (500). This clearly indicates an unacceptable degree of finetuning.
This problem can be mitigated in models with focus point supersymmetry, i.e., when
mH2 is insensitive to m0 [28]. However, this mechanism works for m0 no greater than
23 TeV which is not sufficient to suppress the EDMs. Another interesting possibility is
presented by models with a radiatively driven inverted mass hierarchy, i.e., the models in
which a large hierarchy between the Higgs and the first two generations scalar masses
is created radiatively [29]. However, a successful implementation of this idea is far from
166
trivial [30]. One can also break the scalar mass universality at the high energy scale [31].
In this case, either a mass hierarchy appears already in the soft breaking terms or certain
relations among the soft parameters must be imposed (for a review see [32]). These
significant complications disfavour the decoupling scenario as a way to solve the SUSY
CP problem, yet it remains a possibility.
Note that the decoupling scenario rules out supersymmetry as a possible explanation of
the recently observed 2.6 deviation in the muon g 2 from the SM prediction [33].
This scenario may also lead to cosmological difficulties, in particular with the relic
abundance of the LSPs since the LSP annihilation cross section falls rapidly as msfermion
increases. Concerning the other phenomenological consequences, we remark that the
SUSY contributions to the CP-observables involving the first two generations (such as
, ) are negligible, although those involving the third generation may be considerable.
The corresponding CP-phases are constrained through the Weinberg operator contribution
to the neutron EDM, which typically prohibits the maximal phase At,b |GUT = /2 while
still allowing for smaller O(1) phases (Fig. 4).
4.3. EDM cancellations
The cancellation scenario is based on the fact that large cancellations among different
contributions to the EDMs are possible in certain regions of the parameter space [7,11,35,
36] which allow for O(1) flavour-independent CP phases. Although not particularly well
motivated, this possibility is interesting since, when supplemented with a real non-trivial
flavour structure, it allows for significant supersymmetric contributions to CP observables
including those in the B system. In this case, the SM predictions can be significantly
altered leading for instance to a nonclosure of the unitarity triangle (see M. Brhlik et al.
in [22]). Given the appropriate A-terms flavour structures, the parameters and can
be of completely supersymmetric nature [22]. Large flavour-independent SUSY phases
may also be responsible for electroweak baryogenesis [34]. Thus the cancellation scenario
presents a interesting alternative to the decoupling and approximate CP solutions.
For the case of the electron, the EDM cancellations occur between the chargino
and the neutralino contributions. For the case of the neutron and mercury, there are
cancellations between the gluino and the chargino contributions as well as cancellations
among contributions of different quarks to the total EDM. A number of approximate
formulae quantifying the cancellations are presented in Refs. [7,35].
In what follows we examine the cancellation scenario in the universal and nonuniversal
cases, with the latter being motivated by Type I string models. We note that the CP phases
are to be understood modulo .
4.3.1. EDM cancellations in mSUGRA-type models
These are mSUGRA-type models allowing different phases for different gaugino masses
while keeping their magnitudes the same. As we will see, mSUGRA models with zero
gaugino phases can realize the cancellation scenario. An introduction of the gaugino phases
makes the cancellations much harder to achieve (if possible at all) and leads to a much
167
Fig. 10. Phases allowed by simultaneous electron, neutron, and mercury EDM cancellations in
mSUGRA. The chiral quark neutron model is assumed. Here tan = 3, m0 = m1/2 = 200 GeV,
A = 40 GeV.
higher degree of fine-tuning. The parameters allowing the EDM cancellations strongly
depend on the neutron model. For example, in the parton model, it is more difficult to
achieve these cancellations due to the large strange quark contribution. Therefore, one
cannot restrict the parameter space in a model-independent way and caution is needed
when dealing with the parameters allowed by the cancellations.
In mSUGRA, the EDM cancellations can occur simultaneously for the electron, neutron,
and mercury along a band in the (A , ) plane (Figs. 10 and 11). However, in this case
the mercury constraint requires the phase to be O(102 ) and the magnitude of the Aterms to be suppressed ( 0.1m0 ) which results in only a small effect of the A-terms on
the phase of the corresponding mass insertion. This is to be contrasted with simultaneous
EEDM/NEDM cancellations which allow for O(101 ) and A m0 [11,35]. In that
case the individual contributions to the EDMs exceeded the experimental limit by an
order of magnitude (or more). Owing to the addition of the mercury constraint, the EDM
cancellations become much milder as illustrated in Fig. 12. For example, without these
cancellations the EDMs would exceed the experimental limit only by a factor of a few.
Obviously, the border between the cancellation and the small phases scenarios becomes
blurred. This is even more so at larger tan (Fig. 13), in which case the cancellation band
becomes narrower and the limits on become tighter. We note that there could exist
some points which allow for O(101 ) [37], however these are rare exceptions and
such points do not form a band.
As noted in Ref. [11], in the case of zero gaugino phases, the band in the (A , ) plane
where the cancellations occur can approximately be described by a relation
a sin A ,
(36)
168
Fig. 11. Phases allowed by simultaneous electron, neutron, and mercury EDM cancellations in
mSUGRA. The parton quark neutron model is assumed. Here tan = 3, m0 = m1/2 = 200 GeV,
A = 20 GeV.
Fig. 12. Chargino neutralino cancellations for the EEDM. 1 neutralino, 2 chargino, 3 total
EDM. The SUSY parameters are the same as for Fig. 10, which allow for simultaneous electron,
neutron, and mercury EDM cancellations.
where a is a constant depending on the parameters of the model which represents the
maximal allowed phase . For example, for the chiral quark neutron model and tan = 3,
m0 = m1/2 = 200 GeV, and |A| = 40 GeV (parameters of Fig. 10), a = 0.017. Of course,
the value of a depends on the input parameters such as the quark masses, the GUT scale,
169
Fig. 13. Phases allowed by simultaneous electron, neutron (chiral model), and mercury EDM
cancellations in mSUGRA at larger tan . Here tan = 10, m0 = m1/2 = 200 GeV, A = 40 GeV.
the SUSY scale, etc. and also involves numerical uncertainties, so caution is needed when
treating this number. The maximal allowed is roughly inversely proportional to tan ,
e.g. for tan = 10 we have a = 0.005 (Fig. 13). In the case of the parton model, the
cancellations occur, for example, with a = 0.006 and tan = 3, m0 = m1/2 = 200 GeV,
and |A| = 20 GeV (Fig. 11). As mentioned earlier, the cancellations are harder to achieve
in the parton model, which results in tighter limits on ( O(103 )). In both cases
is restricted to be of order 103 102 , whereas A can be arbitrary (however physical
effects due to A will be suppressed because of the small magnitude of A). The GUT
parameters given above imply that the squark masses at low energies are about 500 GeV,
so these bounds can be relaxed only if the squark masses are over 1 TeV.
If the gluino phase is turned on, simultaneous EEDM, NEDM, and mercury EDM
cancellations are not possible. The gluino phase affects the NEDM cancellation band by
altering the relation (36):
a sin(A + ) c,
(37)
while leaving the EEDM cancellation band almost intact (Fig. 14). An introduction of the
bino phase 1 qualitatively has the same off-setting effect on the EEDM cancellation
band as the gluino phase does on that of the NEDM (Fig. 15). Note that the bino phase has
no significant effect on the neutron and mercury cancellation bands since the neutralino
contribution in both cases is small. When both the gluino and bino phases are present (and
fixed), simultaneous electron, neutron, and mercury EDMs cancellations do not appear
to be possible along a band. The reason is that the mercury EDM depends on 3 more
strongly than the NEDM does, so that an introduction of the gluino phase splits the bands
where the mercury and neutron EDM cancellations occur as illustrated in Figs. 14 and 15.
170
Fig. 14. Bands allowed by the electron (rightmost), neutron (middle), and mercury (leftmost)
EDMs cancellations in the mSUGRA-type model with a nonzero gluino phase. Here tan = 3,
m0 = m1/2 = 200 GeV, A = 40 GeV, 3 = /10.
Fig. 15. Bands allowed by the electron (rightmost), neutron (middle), and mercury (leftmost) EDMs
cancellations in the mSUGRA-type model with nonzero gluino and bino phases. Here tan = 3,
m0 = m1/2 = 200 GeV, A = 40 GeV, 1 = 3 = /10.
Thus we see that zero gaugino phases are much more preferred by the cancellation
scenario.
The cancellation scenario involves a significant fine-tuning. Indeed, restricting the
phases to the band where the cancellations occur does not increase the symmetry of the
171
model and thus is unnatural according to the t Hoofts criterion [24]. It is a non-trivial
task to quantify the degree of fine-tuning in this case. One possibility is to define an
EDM sensitivity coefficient with respect to the CP-phases, in analogy with Eq. (34). We
typically find it to be between 30 and 100 on the cancellation band. This however represents
only a local behaviour of the EDMs. In other words it shows how easy it is to spoil the
cancellations without a reference to how improbable it is to achieve such cancellations in
the first place. An alternative way to quantify the degree of fine-tuning is simply to estimate
the probability that a random small area in the ( , A ) plane will satisfy the cancellation
condition. Since any point is not preferred over any other point by the underlying theory,
this should give a fairly good idea of the minimal degree of fine-tuning needed. From
Fig. 10 we obtain
fine-tuning (probability)1 102 .
(38)
Note that this estimate does not take into account the fine-tuning of the other soft breaking
parameters which is necessary to allow for simultaneous EEDM, NEDM, and mercury
EDM cancellations. Other estimates give a similar number for the universal case, whereas
for the nonuniversal case the degree of fine-tuning drastically increases: only one out of
105 random points in the parameter space satisfies the cancellation condition [37].
4.3.2. EDM cancellations in D-brane models
Let us first briefly review basic ideas of D-brane models (see also Refs. [38] and [39]).
Recent studies of type I strings have shown that it is possible to construct a number
of models with non-universal soft SUSY breaking terms which are phenomenologically
interesting. Type I models can contain 9-branes, 5i -branes, 7i -branes, and 3-branes where
the index i = 1, 2, 3 denotes the complex compact coordinate which is included in the
5-brane world volume or which is orthogonal to the 7-brane world volume. However,
to preserve N = 1 supersymmetry in D = 4 not all of these branes can be present
simultaneously and we can have (at most) either D9-branes with D5i -branes or D3-branes
with D7i -branes.
Gauge symmetry groups are associated with stacks of branes located on top of each
other. A stack of N branes corresponds to the group U (N). The matter fields are
associated with open strings which start and end on the branes. These strings may be
attached to either the same stack of branes or two different sets of branes which have
overlapping world volumes. The ends of the string carry quantum numbers associated with
the symmetry groups of the branes. For example, the quark fields have to be attached to the
U (3) set of branes, while the quark doublet fields also have to be attached to the U (2) set of
branes. Given a brane configuration, the Standard Model fields are constructed according
to their quantum numbers.
The SM gauge group can be obtained in the context of D-brane scenarios from U (3)
U (2) U (1), where the U (3) arises from three coincident branes, U (2) arises from two
coincident D-branes and U (1) from one D-brane. As explained in detail in Ref. [39], there
are different possibilities for embedding the SM gauge groups within these D-branes.
It was shown that if the SM gauge groups come from the same set of D-branes, one
172
cannot produce the correct values for the gauge couplings j (MZ ) and the presence of
additional matter (doublets and triplets) is necessary to obtain the experimental values of
the couplings [40]. On the other hand, the assumption that the SM gauge groups originate
from different sets of D-branes leads in a natural way to intermediate values for the string
scale MS 101012 GeV [39]. In this case, the analysis of the soft terms has been done
under the assumption that only the dilaton and moduli fields contribute to supersymmetry
breaking and it has been found that these soft terms are generically nonuniversal. The
MSSM fields arising from open strings are shown in Fig. 3. For example, the up quark
singlets uc are states of the type C 953 , the quark doublets are C 951 , etc. The presence of
extra (Dq ) branes which are not associated with the SM gauge groups is often necessary to
reproduce the correct hypercharge and to cancel non-vanishing tadpoles.
Recently there has been a considerable interest in supersymmetric models derived from
D-branes [41,42]. In a toy model of Ref. [41], the gauge group SU(3)c U (1)Y was
associated with 51 branes and SU(2)L was associated with 52 branes. It was shown that in
this model the gaugino masses are non-universal (M1 = M3 = M2 ) so that the physical
CP phases are 1 = 3 , A and . Non-universality of the gaugino masses allowed
to enlarge the regions of the parameter space where the NEDM/EEDM cancellations
occurred. However, such a model is oversimplified and the relation 1 = 3 = 2 does
not appear to hold in more realistic models [43]. In what follows we will consider the
EDM cancellations in this and a more realistic D-brane models.
The model in which U (3), U (2), and U (1) originate from different sets of branes is
phenomenologically interesting. In this case one naturally obtains an intermediate string
scale (10101012 GeV), although higher values up to 1016 GeV are still allowed. Both the
up and the down type Yukawa interactions are allowed, while that for the leptons typically
vanishes (depending on further details of the model) [39]. The (anomaly-free) hypercharge
is expressed in terms of the U (1) charges Q1,2,3 of the U (1)1,2,3 groups:
1
1
Y = Q3 Q2 + Q1 ,
3
2
(39)
with the following (Q3 , Q2 , Q1 ) charge assignment:

q = (1, 1, 0),
l = (0, 1, 0),
H2 = (0, 1, 1),
uc = (1, 0, 1),
d c = (1, 0, 0),
ec = (0, 0, 1),
H1 = (0, 1, 0).
The gaugino masses in this model are given by
M3 = 3 m3/2 sin eis ,
M2 = 3 m3/2 1 cos ei1 ,

2
1
3 cos ei3 +
1 cos ei1
MY = 3 m3/2 Y (MS )
1 (MS )
2 (MS )

2
sin eis ,
+
33 (MS )
(40)
(41)
173
where
1
2
1
2
=
+
+
.
Y (MS ) 1 (MS ) 2 (MS ) 33 (MS )
(42)
Here k correspond to the gauge couplings of the U (k) branes. As shown in Ref. [39],
1 (MS ) 0.1 leads to the string scale MS 1012 GeV. The parameters and i
parameterize supersymmetry breaking in the usual way [38]:

F S = 3 S + S m3/2 sin eis ,

F i = 3 Ti + Ti m3/2 cos i eii ,
(43)
and i = 1, 2, 3 labels the three complex compact dimensions.
The soft scalar masses are given by

3
m2q = m23/2 1 1 12 cos2 ,
2

2
3
2
2
2
md c = m3/2 1 1 2 cos ,
2

2
3
2
2
2
muc = m3/2 1 1 3 cos ,
2

3
m2ec = m23/2 1 sin2 + 12 cos2 ,
2

2
3
2
2
2
2
ml = m3/2 1 sin + 3 cos ,
2

3 2
2
2
2
2
mH2 = m3/2 1 sin + 2 cos ,
2
2
2
mH1 = ml ,
and the trilinear parameters are

3
Au =
m3/2 2 ei2 1 ei1 3 ei3 cos sin eis ,
2

3
m3/2 3 ei3 1 ei1 2 ei2 cos sin eis ,
Ad =
2
Ae = 0.
(44)
(45)
(46)
(47)
We observe that the angles i and are quite constrained if we are to avoid negative
mass-squareds for squarks and sleptons. In what follows we set 3 = 0 and 1 = 2 ; then
the soft terms are parameterized in terms of the phase 1 s .
In Fig. 16 we display the bands allowed by the electron (red), neutron (green), and
mercury (blue) EDMs. In this figure, we set m3/2 = 150 GeV, tan = 3, 12 = 22 = 1/2,
cos2 = 2 sin2 = 2/3, and 1 (MS ) 1 with MS being the GUT scale. For the plot to be
more illustrative, we do not impose any additional constraints besides the EDM ones (i.e.,
bounds on the chargino and slepton masses, etc.). It is clear that even though simultaneous
EEDM/NEDM cancellations allow the phase to be O(1), an addition of the mercury
constraint requires all phases to be very small (modulo ) and thus practically rules out
174
Fig. 16. Bands allowed by the electron (1), neutron (2), and mercury (3) EDMs in the D-brane model.
Here tan = 3, m3/2 = 150 GeV, 12 = 22 = 1/2, cos2 = 2 sin2 = 2/3, and MS 1016 GeV.
Fig. 17. Bands allowed by the electron (1), neutron (2), and mercury (3) EDMs in the
D-brane model with an intermediate scale. Here tan = 3, m3/2 = 150 GeV, 12 = 22 = 1/2,
cos2 = 2 sin2 = 2/3, and MS 1012 GeV.
the cancellation scenario in this context. The situation becomes even worse in the case of
an intermediate string scale 1012 GeV (i.e., 1 (MS ) 0.1), see Fig. 17. We find it quite
generic that the mercury EDM behaviour in D-brane models is very different from that of
the electron and neutron and thus is crucial in constraining the parameter space. The major
difference from the mSUGRA-type models with fixed Y and 3 is that the phase of the
175
Fig. 18. Bands allowed by the electron (1), neutron (2), and mercury (3) EDMs in the model of
Ref. [41]. Here tan = 2, m3/2 = 150 GeV, 1 = 0.9, = 0.4.
A-terms is correlated with the gaugino phases resulting in the cancellation bands which are
not described by a simple relation a sin(A + ) c.
Next we consider the model of Ref. [41]. The (corrected) soft terms for this model read
(for 3 = 0)
MY = M3 = A = 3 m3/2 cos 1 ei1 ,
M2 = 3 m3/2 cos 2 ei2 ,

3
m2L = m23/2 1 sin2 ;
2

2
2
mR = m3/2 1 3 cos2 22 .
(48)
To illustrate the EDM constraints, we choose the parameters which allow for simultaneous
EEDM/NEDM cancellations, namely m3/2 = 150 GeV, tan = 2, 1 = 0.9, and = 0.4
as given in Ref. [41]. Fig. 18 shows that the mercury constraint has the same behaviour as
in the model considered above and rules out large CP-phases.
We see that the cancellation scenario in simple models faces a number of difficulties.
Presently available string-motivated models with non-universal gaugino masses cannot
accommodate simultaneous electron, neutron, and mercury EDM cancellations. In the
mSUGRA, such cancellations are possible but require a significant fine-tuning. The
addition of the mercury EDM bound restricts the phase of the term to be O(102 ) if
we are to achieve the EDM cancellations along a band in the (A , ) plane. Without an
additional SUSY flavour structure, the CP-phases allowed by the cancellations will have
very small observable effects. Even in a more general situation (unconstrained MSSM), the
phases allowed by the EDM cancellations typically lead to small CP-asymmetries ( 1%)
176
in collider experiments [37]. Testing the cancellation scenario experimentally may prove
to be a challenge.
4.4. Flavour-off-diagonal CP violation
This is one of the more attractive possibilities to avoid overproduction of EDMs in SUSY
models. Nonobservation of EDMs may imply that CP-violation has a flavour-off-diagonal
character just like in the Standard Model. The origin of CP-violation in this case is closely
related to the origin of the flavour structures rather than the origin of supersymmetry
breaking. While models with flavour-off-diagonal CP violation naturally avoid the EDM
problem, they have testable effects in K and B physics.
This class of models requires hermitian Yukawa matrices and A-terms, which forces the
flavour-diagonal phases to vanish (up to small RG corrections) in any basis. The flavourindependent quantities such as the -term, gaugino masses, etc., are real. This is naturally
implemented in left-right symmetric models [44] and models with a horizontal flavour
symmetry [46].
In the left-right models, the hermiticity of the Yukawas and A-terms as well as the reality
of the -term is forced by the left-right symmetry. The SU(2)L gaugino mass is in general
complex, so in order to suppress the EDMs the additional assumption of its reality is
needed. The phenomenology of such models in the context of up-down unification has been
studied in [45]. The left-right symmetry appears to be too restrictive in this case to satisfy
all of the phenomenological constraints; however the decisive test will be undertaken at the
B factories.
Another possibility is based on a horizontal U (3)H symmetry [46]. Hermitian Yukawa
matrices may appear due to a (gauged) horizontal symmetry U (3)H which gets broken spontaneously by the VEVs of the real adjoint fields Ta (a = 1, . . . , 9; =
u, d, l, . . .) [47]. Some of these real VEVs also break CP since some of the components
of Ta are CP-odd. As a result, CP violation appears in the superpotential through complex Yukawa couplings, whereas the -term is real since it arises from a U (3)H invariant
combination of the type Ta Ta . An effective U (3)H -invariant superpotential of the type
gH
a
d
a
a a
= gH H
W
M 1 Qi (Td )ij Dj produces the Yukawa matrix (Y )ij = M Td ( )ij , where
0 is proportional to the unit matrix and 18 are the Gell-Mann matrices. Note that the
real fields Ta may only come from a non-supersymmetric (anti-brane) sector. If Ta are
the scalar components of chiral multiplets, they are intrinsically complex and hermitian
Yukawas in this case arise only if the VEVs of Ta are real.
The gaugino masses in this model are not forced to be real by symmetry. One has
to make an additional assumption that either the gaugino masses are universal and the
corresponding phase can be rotated away or that the SUSY breaking dynamics conserve
CP which seems natural if CP breaking is associated with the origin of flavour structures. In
other words, the SUSY breaking auxiliary fields get real VEVs as a result of the underlying
dynamics such as the dilaton stabilization in Type I string models [48] or the effective
potential minimization in heterotic string models [49]. Further, if the Khler potential is
either generation independent or left-right symmetric, the A-terms are hermitian [46]. This
177
leads to very small phases in the flavour-diagonal mass insertions which are responsible
for generating EDMs and the EDM bounds are easily satisfied.
Both the left-right model and the model with a horizontal symmetry mitigate the strong
CP-problem (under the additional assumptions given above). The parameter vanishes at
both the tree and the leading log one-loop levels. However neither of these models solves
the problem completely due to the significant one loop finite corrections which appear
mostly due to a non-degeneracy of the squark masses and a misalignment between the
Yukawas and the A-terms [50].
In this class of models we have the following setting at the high energy scale:
Y = Y ,
A = A ,
Arg(Mk ) = Arg() = 0.
(49)
Generally, the off-diagonal elements of the A-terms can have O(1) phases without
violating the EDM constraints. Due to the RG effects, large phases in the soft trilinear
couplings involving the third generation generate small phases in the flavour-diagonal mass
insertions for the light generations, and thus induce the EDMs. For example, the A-terms
of the form 3
1 a12 a13
Ad = Au = m0 a12
1 a23
(50)
a13 a23 a33

and the following GUT-scale hermitian Yukawa matrices
4.1 104
6.9 104 i 1.4 102
Y u = 6.9 104 i 3.5 103 1.4 105 i ,
6.9 101
1.4 102 1.4 105 i
1.3 104
2.0 104 + 1.8 104 i 4.4 104
Y d = 2.0 104 1.8 104 i
9.3 104
7.0 104 i
4
4
7.0 10 i
1.9 102
4.4 10
typically induce the mercury EDM of the order of the experimental limit if a33 is of order 1,
whereas the induced NEDM is 12 orders of magnitude below the experimental limit. If
, this RG effect will be suppressed rendering the
one uses Yukawa textures with smaller Y13
induced mercury EDM far below the experimental limit.
The supersymmetric contribution to the parameter is suppressed in models with
hermitian flavour structures [46]. 4 This occurs due to severe cancellations between
d
d
)LR and (12
)RL mass insertions (we use the standard
the contributions involving (12
d
d
)RL , whereas they contribute to
definitions of [52]). Due to the hermiticity (12 )LR (12
d
d
with opposite signs. Typically we find | Im[(12 )LR (12 )RL ]| < 106 which produces
an order of magnitude below the experimental limit [52]. On the other hand, similar
cancellations do not occur for the parameter and the SUSY contribution to can be
3 The standard supergravity relation (A
)ij = (A )ij (Y )ij for the trilinear soft couplings is assumed.
4 For phenomenology of models with non-universal (but non-hermitian) A-terms see also [51].
178
Table 1
Bounds on the imaginary parts of the mass insertions. The chiral quark model for the neutron is
assumed. Here x = m2g /m2q = m2 /m2 , mq = 500 GeV, ml = 100 GeV. For different squark/slepton
B
l
masses the bounds are to be multiplied by mq /500 GeV or ml/100 GeV
x
d ) )|
| Im(11
LR
u ) )|
| Im(11
LR
l ) )|
| Im(11
LR
0.1
0.3
1
3
5
10
1.0 106
9.2 107
1.1 106
1.8 106
2.4 106
4.1 106
2.0 106
1.8 106
2.2 106
3.6 106
4.9 106
8.2 106
1.4 107
1.3 107
1.6 107
2.6 107
3.5 107
5.9 107
Table 2
Bounds on the imaginary parts of the mass insertions for the parton neutron model. For the squark
masses different from 500 GeV, the bounds are to be multiplied by mq /500 GeV
x
d ) )|
| Im(11
LR
u ) )|
| Im(11
LR
d ) )|
| Im(22
LR
0.1
0.3
1
3
5
10
1.8 106
1.6 106
1.9 106
3.2 106
4.4 106
7.3 106
1.3 106
1.2 106
1.5 106
2.3 106
3.2 106
5.4 106
5.9 106
5.4 106
6.6 106
1.1 105
1.4 105
2.4 105
d
even dominant. The value of (12
)LR which saturates the observed || 2.26 103 is

d )2 | 3.5 104 for the gluino and squark masses of 500 GeV [52].
given by |Im(12
LR
For completeness, we provide the bounds on the imaginary parts of the mass insertions

1 d(u)
d(u)
ASCKM ii v1(2) Yi v2(1)
(51)
2
m
derived from the leading gluino contributions to the electromagnetic (NEDM) operator
and the bino contribution to the electron EDM. We update the bounds of Ref. [52] and
also include the QCD correction factor. The advantage of the mass insertion approach is
that it allows to obtain model independent bounds and thus is quite useful when dealing
with complicated flavour structures. For comparison we present the bounds for the chiral
(Table 1) and the parton (Table 2) neutron models.
A drastic improvement of the bounds comes from the addition of the mercury EDM
constraint. Using the expressions for the chromomagnetic moments of Ref. [36], in Table 3
we present the bounds on the mass insertions from the gluino contributions to the mercury
d
u
)LR )| and | Im(11
)LR )| these bounds turn out to be very strict, more than
EDM. For | Im(11
an order of magnitude stricter than those imposed by the NEDM. This severely restricts the
CP-asymmetry ACP (b s ) in models with hermitian flavour structures. The reason is
d(u)
LR
ii
179
Table 3
Bounds on the imaginary parts of the mass insertions imposed by the mercury EDM. For the squark
masses different from 500 GeV, the bounds are to be multiplied by mq /500 GeV
x
d ) )|
| Im(11
LR
u ) )|
| Im(11
LR
d ) )|
| Im(22
LR
0.1
0.3
1
3
5
10
2.6 108
3.6 108
6.7 108
1.5 107
2.5 107
5.3 107
2.6 108
3.6 108
6.7 108
1.5 107
2.5 107
5.3 107
2.2 106
3.0 106
5.6 106
1.3 105
2.1 105
4.4 105
that in order to obtain a large ( 10%) SUSY contribution to this observable, the elements
of the A-terms involving the third generation have to be larger than 1. This induces via
d,u
)LR )|, often in conflict with the bounds of Table 3.
the RG running a considerable | Im(11
We find that with the above Yukawa textures ACP (b s ) is allowed to be no more than
23%. One can relax this constraint by using different hermitian textures, especially with
. In this case the CP-asymmetry can be as large as 67%.
suppressed Y13
To conclude this section, we remark that the problem of baryogenesis in this class
of models requires careful investigation and at the moment it is unclear whether or not
large flavour off-diagonal SUSY phases can produce sufficient baryon asymmetry of the
universe.
5. Discussion and conclusions

We have systematically analyzed constraints on supersymmetric models imposed by the
experimental bounds on the electron, neutron, and mercury electric dipole moments. We
find that the EDMs can be suppressed in SUSY models with:
(1) Small SUSY CP-phases ( 102 ). This possibility can be motivated by the
approximate CP-symmetry which also implies that the CKM phase is small. This provides
testable signatures for B-factories experiments.
(2) Heavy SUSY scalars, msfermion 10 TeV. In this class of models there is a large
hierarchy between the SUSY and electroweak scales, which is hard to realize without an
extreme fine-tuning.
(3) EDM cancellations. We have analyzed the possibility of such cancellations in Dbrane and mSUGRA-like models with nontrivial gaugino phases. We find that, with the
addition of the mercury EDM constraint, only the EDM cancellations in mSUGRA (1
3 0) survive in any considerable part of the parameter space. Even in this case the
cancellations require small 102 and suppressed |A| ( 0.1m0 ). As a result, the
border between the small phases and the cancellation scenarios fades away. In addition to
the finetuning problem, models with the EDM cancellations lack predictive power as it is
unclear whether the allowed CP-phases can have observable effects.
180
(4) Flavour-off-diagonal CP violation. This can occur in models with CP-conserving

SUSY breaking dynamics and hermitian flavour structures. Such models allow for O(1)
flavour off-diagonal phases which can have significant effects in K and B physics.
It is also possible to combine different mechanisms to suppress the EDMs. For example,
a hybrid of the decoupling and the cancellation scenarios was considered in Ref. [53].
Such models seem to share shortcomings of both parents without an apparent advantage
over either of them.
There is also a rather radical proposal that all gaugino masses and the A-terms are
vanishingly small [54]. Clearly this eliminates all of the physical phases in Eq. (15) and
thus produces no CP violation. Such a strong assumption requires a firm motivation such
as the continuous R-symmetry. However, by the same token, the R-symmetry eliminates
the B term thereby leading to a very light CP-odd Higgs boson. Such a scenario faces a
number of difficulties with experimental results.
To summarize, we have studied the supersymmetric CP problem taking into account all
of the current EDM constraints. Our conclusion is that there remain two attractive ways
to avoid overproduction of the EDMs in SUSY models. The first possibility is that CP is
an approximate symmetry of nature and the second one is that CP violation has a flavour
off-diagonal character just as in the Standard Model.
Acknowledgements
S.A. and O.L. were supported by the PPARC Opportunity grant, S.K. was supported by
the PPARC SPG. The authors are grateful to D. Bailin, M. Brhlik, T. Ibrahim, C. Muoz,
and M. Pospelov for useful discussions, and to M. Gomez and D. Cerdeo for their help in
the computing aspects of this work.
References
[1] T. Affolder et al., CDF Collaboration, Phys. Rev. D 61 (2000) 072005;
B. Aubert et al., BaBar Collaboration, hep-ex/0102030;
A. Abashian et al., Belle Collaboration, hep-ex/0102018.
[2] P.G. Harris et al., Phys. Rev. Lett. 82 (1999) 904;
see also the discussion in S.K. Lamoreaux, R. Golub, Phys. Rev. D 61 (2000) 051301.
[3] E.D. Commins et al., Phys. Rev. A 50 (1994) 2960.
[4] M.V. Romalis, W.C. Griffith, E.N. Fortson, Phys. Rev. Lett. 86 (2001) 2505;
J.P. Jacobs et al., Phys. Rev. Lett. 71 (1993) 3782.
[5] T. Falk, K.A. Olive, M. Pospelov, R. Roiban, Nucl. Phys. B 560 (1999) 3.
[6] H. Georgi, A. Manohar, Nucl. Phys. B 234 (1984) 189;
R. Arnowitt, J.L. Lopez, D.V. Nanopoulos, Phys. Rev. D 42 (1990) 2423;
R. Arnowitt, M.J. Duff, K.S. Stelle, Phys. Rev. D 43 (1991) 3085.
[7] T. Ibrahim, P. Nath, Phys. Rev. D 57 (1998) 478;
T. Ibrahim, P. Nath, Phys. Rev. D 58 (1998) 019901, Erratum;
T. Ibrahim, P. Nath, Phys. Rev. D 60 (1999) 019901;
T. Ibrahim, P. Nath, Phys. Lett. B 418 (1998) 98;
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
181

T. Ibrahim, P. Nath, Phys. Rev. D 60 (1999) 099902, Erratum.
M. Pospelov, A. Ritz, Phys. Rev. D 63 (2001) 073015.
V.M. Khatsimovsky, I.B. Khriplovich, A.R. Zhitnitsky, Z. Phys. C 36 (1987) 455;
V.M. Khatsimovsky, I.B. Khriplovich, A.S. Yelkhovsky, Ann. Phys. 186 (1988) 1;
V.M. Khatsimovsky, I.B. Khriplovich, Phys. Lett. B 296 (1992) 219.
J. Ellis, R. Flores, Phys. Lett. B 377 (1996) 83.
A. Bartl, T. Gajdosik, W. Porod, P. Stockinger, H. Stremnitzer, Phys. Rev. D 60 (1999) 073003.
I.B. Khriplovich, S.K. Lamoreaux, CP Violation Without Strangeness, Springer, 1997.
S. Weinberg, Phys. Rev. Lett. 63 (1989) 2333;
E. Braaten, C. Li, T. Yuan, Phys. Rev. Lett. 64 (1990) 1709.
J. Dai, H. Dykstra, R.G. Leigh, S. Paban, D. Dicus, Phys. Lett. B 237 (1990) 216.
D. Chang, W. Keung, A. Pilaftsis, Phys. Rev. Lett. 82 (1999) 900;
D. Chang, W. Keung, A. Pilaftsis, Phys. Rev. Lett. 83 (1999) 3972, Erratum.
S.M. Barr, A. Zee, Phys. Rev. Lett. 65 (1990) 21.
J.F. Gunion, H.E. Haber, Nucl. Phys. B 272 (1986) 1.
S. Baek, P. Ko, Phys. Rev. Lett. 83 (1999) 488.
D. Chang, W. Chang, W. Keung, Phys. Lett. B 478 (2000) 239;
A. Pilaftsis, Phys. Lett. B 471 (1999) 174.
A. Pilaftsis, Phys. Rev. D 62 (2000) 016007.
M. Dine, E. Kramer, Y. Nir, Y. Shadmi, hep-ph/0101092;
G. Eyal, Y. Nir, Nucl. Phys. B 528 (1998) 21, and references therein.
A. Masiero, H. Murayama, Phys. Rev. Lett. 83 (1999) 907;
S. Khalil, T. Kobayashi, Phys. Lett. B 460 (1999) 341;
M. Brhlik, L. Everett, G.L. Kane, S.F. King, O. Lebedev, Phys. Rev. Lett. 84 (2000) 3041.
M. Carena, M. Quiros, A. Riotto, I. Vilja, C.E. Wagner, Nucl. Phys. B 503 (1997) 387.
G. t Hooft, in: G. t Hooft et al. (Eds.), Recent Advances in Gauge Theories, Proceedings of
the Cargese Summer Institute, Cargese, France, 1979, NATO Advanced Study Institute Series
B Physics, Vol. 59, Plenum, New York, 1980.
D. Bailin, G.V. Kraniotis, A. Love, Nucl. Phys. B 518 (1998) 92;
J. Giedt, Nucl. Phys. B 595 (2001) 3.
P. Nath, Phys. Rev. Lett. 66 (1991) 2565;
Y. Kizukuri, N. Oshimo, Phys. Rev. D 46 (1992) 3025.
R. Barbieri, G.F. Giudice, Nucl. Phys. B 306 (1988) 63;
J. Ellis, K. Enqvist, D.V. Nanopoulos, F. Zwirner, Mod. Phys. Lett. A 1 (1986) 57;
G.G. Ross, R.G. Roberts, Nucl. Phys. B 377 (1992) 571.
J.L. Feng, K.T. Matchev, T. Moroi, Phys. Rev. D 61 (2000) 075005;
J.L. Feng, K.T. Matchev, T. Moroi, Phys. Rev. Lett. 84 (2000) 2322.
J.L. Feng, C. Kolda, N. Polonsky, Nucl. Phys. B 546 (1999) 3.
H. Baer, C. Balazs, M. Brhlik, P. Mercadante, X. Tata, Y. Wang, hep-ph/0102156.
S. Dimopoulos, G.F. Giudice, Phys. Lett. B 357 (1995) 573.
H. Baer, M.A. Diaz, P. Quintana, X. Tata, JHEP 0004 (2000) 016.
L. Everett, G.L. Kane, S. Rigolin, L.-T. Wang, hep-ph/0102145;
J.L. Feng, K.T. Matchev, hep-ph/0102146;
E.A. Baltz, P. Gondolo, hep-ph/0102147;
U. Chattopadhyay, P. Nath, hep-ph/0102157.
M. Brhlik, G.J. Good, G.L. Kane, Phys. Rev. D 63 (2001) 035002.
T. Falk, K.A. Olive, Phys. Lett. B 375 (1996) 196;
Phys. Lett. B 439 (1998) 71;
S. Pokorski, J. Rosiek, C.A. Savoy, Nucl. Phys. B 570 (2000) 81.
182
[37] V. Barger, T. Falk, T. Han, J. Jiang, T. Li, T. Plehn, hep-ph/0101106.

[38] L.E. Ibez, C. Muoz, S. Rigolin, Nucl. Phys. B 553 (1999) 43;
A. Brignole, L.E. Ibez, C. Muoz, Nucl. Phys. B 422 (1994) 125.
[39] D.G. Cerdeo, E. Gabrielli, S. Khalil, C. Muoz, E. Torrente-Lujan, hep-ph/0102270, to appear
in Nucl. Phys. B.
[40] G. Aldazabal, L.E. Ibez, F. Quevedo, A.M. Uranga, JHEP 0008 (2000) 002;
D. Bailin, G.V. Kraniotis, A. Love, hep-th/0011289.
[41] M. Brhlik, L. Everett, G.L. Kane, J. Lykken, Phys. Rev. Lett. 83 (1999) 2124;
M. Brhlik, L. Everett, G.L. Kane, J. Lykken, Phys. Rev. D 62 (2000) 035005.
[42] E. Accomando, R. Arnowitt, B. Dutta, Phys. Rev. D 61 (2000) 075010;
T. Ibrahim, P. Nath, Phys. Rev. D 61 (2000) 093004.
[43] S. Abel, S. Khalil, O. Lebedev, hep-ph/0103031.
[44] R.N. Mohapatra, G. Senjanovic, Phys. Lett. B 79 (1978) 283;
R.N. Mohapatra, A. Rasin, Phys. Rev. Lett. 76 (1996) 3490;
R.N. Mohapatra, A. Rasin, Phys. Rev. D 54 (1996) 5835.
[45] K.S. Babu, B. Dutta, R.N. Mohapatra, Phys. Rev. D 61 (2000) 091701.
[46] S. Abel, D. Bailin, S. Khalil, O. Lebedev, hep-ph/0012145, to appear in Phys. Lett. B.
[47] A. Masiero, T. Yanagida, hep-ph/9812225.
[48] S.A. Abel, G. Servant, Nucl. Phys. B 597 (2001) 3;
S.A. Abel, G. Servant, in preparation.
[49] D. Bailin, G.V. Kraniotis, A. Love, Phys. Lett. B 432 (1998) 343;
T. Dent, hep-th/0011294.
[50] M. Dine, R.G. Leigh, A. Kagan, Phys. Rev. D 48 (1993) 2214.
[51] S.A. Abel, J.-M. Frre, Phys. Rev. D 55 (1997) 1623;
S. Khalil, T. Kobayashi, A. Masiero, Phys. Rev. D 60 (1999) 075003;
S. Khalil, T. Kobayashi, O. Vives, Nucl. Phys. B 580 (2000) 275.
[52] F. Gabbiani, E. Gabrielli, A. Masiero, L. Silvestrini, Nucl. Phys. B 477 (1996) 321.
[53] U. Chattopadhyay, T. Ibrahim, D.P. Roy, hep-ph/0012337.
[54] L. Clavelli, T. Gajdosik, W. Majerotto, Phys. Lett. B 494 (2000) 287.

Electroweak baryogenesis: concrete in a SUSY

model with a gauge singlet
S.J. Huber a,b , M.G. Schmidt a
a Institut fr Theoretische Physik, Philosophenweg 16, D-69120 Heidelberg, Germany
b Bartol Research Institute, University of Delaware, Newark, DE 19716, USA
Received 28 June 2000; accepted 18 May 2001
Abstract
SUSY models with a gauge singlet easily allow for a strong first order electroweak phase transition
(EWPT) if the vevs of the singlet and Higgs fields are of comparable size. We discuss the profile
of the stationary expanding bubble wall and CP-violation in the effective potential, in particular
transitional CP-violation inside the bubble wall during the EWPT. The dispersion relations for
charginos contain CP-violating terms in the WKB approximation. These enter as source terms in
the Boltzmann equations for the (particleantiparticle) chemical potentials and fuel the creation of
a baryon asymmetry through the weak sphaleron in the hot phase. This is worked out for concrete
parameters. 2001 Published by Elsevier Science B.V.
PACS: 05.60.Cd; 11.15.Kc; 11.30.Pb; 12.60.Fr
1. Introduction
Different from models starting at the Grand Unified scale, the ingredients for a creation
of the baryon asymmetry of the universe in a first order electroweak phase transition
(PT) have a fair chance to be tested experimentally in the near future. Besides nonequilibrium the other two necessary criteria of Sakharov for baryogenesis, CP and baryon
number violation are naturally fulfilled by the CP violating bubble wall of a first order PT.
The bubble wall separates the hot symmetric phase with rapid sphaleron transitions
from the Higgs phase, where sphaleron transitions are suppressed by exp(v(t)/T ).
However it turned out that the electroweak Standard Model (SM) does not have
a strong first order PT, indeed there is a crossover behavior and no PT at all for
Higgs masses beyond the present experimental bound [1]. Furthermore, CP-violation
by the CKM matrix is very small. In spite of tremendous successes of the SM it is
E-mail addresses: shuber@udel.edu (S.J. Huber), m.g.schmidt@thphys.uni-heidelberg.de (M.G. Schmidt).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 2 5 0 - 4
184
S.J. Huber, M.G. Schmidt / Nuclear Physics B 606 (2001) 183230
commonly believed that it has to be embedded/enlarged to a more fundamental theory.

Supersymmetry is supposed to be an important facet of such a theory but it is still
not backed by experiments. In concrete SUSY models, in order to have a strong first
order PT one has to strengthen the (one loop) 3 term of the effective potential
or to have such a term already on the tree level. In the minimal SUSY extension of
the SM, the MSSM, the superpartner of the right-handed top with a mass below
that of the top gives such a strong 3 loop correction. With a negative SUSY
breaking scalar mass it is almost massless in the hot phase and has a strong Yukawa
coupling to the Higgs fields. This is confirmed by detailed analytical [2] and lattice
calculations [3], showing a strong first order PT even for Higgs masses as large as
110 GeV just beyond the experimental bound for MSSM Higgses. There are also possible
CP-violating phases in the Higgs Lagrangian if 1-loop effects are taken into account [4].
They are restricted much by the measurements of the neutron electric dipole moment
unless one invokes special conditions. A recent investigation of a possible spontaneous
CP-violation just in the bubble wall at the temperatures of the PT gave negative results [5].
In this paper 1 we want to discuss baryogenesis in a strong first order electroweak PT
(EWPT) in much detail starting from a model where all the three criteria given above can
be fulfilled without much problem. This is a supersymmetric model with an additional
gauge singlet superfield. In its original form (NMSSM) is was designed to substitute the
problematic H1 H2 term of the MSSM superpotential by a coupling SH1 H2 [6]
k
WZ3 = SH1 H2 + S 3
3
with a Z3 -symmetry whose spontaneous breaking causes dangerous domain walls in the
early cosmos [7]. The soft SUSY breaking potential contains a term A SH1 H2 which can
act as a 3 term if both the vevs of the Higgses and of the singlet are of the same order of
magnitude [8]. But in a Z3 -symmetric model with universal SUSY breaking this turns out
not to be possible if one wants to have a reasonable spectrum of particles.
In agreement with Ref. [9] we thus add further terms to the superpotential which now
obtains the form
k
W = SH1 H2 + S 3 + H1 H2 + rS
(1.1)
3
and is not Z3 -symmetric anymore. The soft SUSY breaking Lagrangian contains scalar
masses, gaugino masses and A-terms
k
3
c
c 1 + Yu Au u c qH
Lsoft
2 + h.c.
A = A SH1 H2 + Ak S + Ye Ae e lH1 + Yd Ad d qH
3
(1.2)
Thus we have reintroduced a -term and we have the associated fine-tuning problem
known from the MSSM [10]. Additionally, there is the danger of quadratically divergent
singlet tadpoles [11,12]. Such tadpole diagrams require three ingredients: (i) a singlet field
1 Whose content was presented at the TMR network meeting Finite Temperature Phase Transitions in Particle
Physics in Korfu, September 1999.
185
(ii) non-renormalizable interactions and (iii) soft SUSY breaking terms. (i) Giving the
singlet a charge under some discrete symmetry one can remove the tadpoles from the
very beginning. But the discrete symmetry cannot be exact in order to avoid unacceptable
domain walls at the EWPT [13]. (ii) In Refs. [12,14] models with gauged R-symmetry
or duality symmetry, both broken at some superheavy scale have been proposed to
forbid dangerous non-renormalizable operators. On the renormalizable level these models
typically have no discrete symmetries, and therefore are not plagued by the domain wall
problem. While such models do not solve the -problem, they are still very interesting
with respect to Higgs phenomenology and the EWPT. For that reason we will study this
type of singlet model in the following. (iii) Another way to evade the tadpole divergences
restricts the soft SUSY breaking terms. Gauge mediated SUSY breaking (GMSB) in the
context of singlet models does not have domain wall problems. A -parameter is generated
by radiative corrections and the singlet vev [15]. However one of the special properties of
GMSB models seems to be the strong suppression of A-terms. These however also contain
the -type term responsible for a strong first order PT.
In order not to have too many parameters we use universality of SUSY breaking at the
GUT scale and run renormalization group equations to get down to the electroweak scale.
With such a Lagrangian and adding the temperature corrections we can demonstrate [16]
that one can easily get a strong first order PT for Higgs masses as high as 115 GeV and
superpartner masses beyond present experimental limits. The effective potential in H1 ,
H2 and S can be used to derive differential equations for the bubble wall profile and for
varying CP-violating phases in the bubble wall arising from constant explicit phases in the
theory or, more interesting, from spontaneous CP-violation. This is described in Sections 4
and 5. As worked out in Section 6, in the WKB approximation the CP-violating phases in
the bubble wall create terms in the dispersion relations which differ between particles and
antiparticles. This gives rise to a driving term in the Boltzmann equation for the difference
of particle and antiparticle chemical potentials, in particular for left-handed quarks and
their superpartners. This difference is converted into a baryon asymmetry by the hot phase
sphalerons in front of the proceeding bubbles.
2. The model
Weak scale supersymmetric models have many unknown parameters related to supersymmetry breaking. A considerable part of this parameter space is excluded because of
FCNC or the appearance of charge and color breaking vacua. These problems are partially evaded by imposing universality of the soft terms at the GUT scale, MGUT 2.6
1016 GeV, where the SM gauge couplings unify. At the GUT scale there is a common gaugino mass M0 , a universal scalar mass squared m20 and a universal trilinear coupling A0 .
Thus, universal SUSY breaking drastically reduces the parameters of the model to
yt 0 , 0 , k0 , M0 , A0 , m20 , 0 , r0 , B0 .
(2.1)
186
The subscript 0 indicates that the masses and couplings are evaluated at the GUT
scale. yt denotes the top Yukawa coupling, 2 and BH1 H2 is the soft Higgs mass term
corresponding to the -term in the superpotential. Since the masses of the top quark and
the Z-boson mass are known, only seven parameters of (2.1) are independent.
To evolve the parameters from the GUT scale to the weak scale we use the
renormalization group equations (RGEs) in the 1-loop approximation. For the Z3 symmetric case the relevant RGEs have already been given in Ref. [17]. Defining t = ln Q2
the Z3 -breaking terms , r and B are easily included with help of Ref. [18]

d
1
=
3yt2 + 22 + 2k 2 3g22 g12 ,
2
dt
32

1 2
d
r=
r + k2 ,
dt
16 2

1 2
d
B=
2 B + 3yt2At + 22 A + 3g22 M2 + g12 M1 .
(2.2)
2
dt
16
Even with the simple universal pattern of soft breaking terms we are left with a 9dimensional parameter space. Additionally, we must satisfy two constraints coming
from the Z-boson and top quark masses. In the literature (see, e.g., Ref. [19]), usually
random shooting is applied to deal with such a situation. This means randomly
chosen sets of GUT scale parameters are evolved to the electroweak scale, where their
phenomenological implications are investigated. Although the physical Z-boson mass can
easily be reproduced by an appropriate rescaling of the dimensionful parameters, one
typically is still plagued by some light unobserved SUSY particles or an unphysical top
quark mass. Therefore this procedure is rather inefficient. Furthermore, random shooting
only provides statistical averages and correlations. To avoid these shortcomings we use
the more systematic approach which we introduced in Ref. [16]. It allows the elimination
of , r and B by the Higgs and singlet vevs, while maintaining universal SUSY breaking.
It relies on the observation that the Z3 -breaking terms , r and B do not enter the RGEs for
the remaining parameters. They can therefore be calculated without specifying the former
ones. Our procedure consists of the following steps and is also sketched in Fig. 1:
Fix the values of x S
, tan = v2 /v1 , and k at the weak scale. The choice of
tan automatically fixes yt by the relation yt v sin = mtop (= 175 GeV).
Evolve yt , and k to the GUT scale.
Choose a set of the parameters M0 , A0 and m20 at the GUT scale.
Run all parameters down to the weak scale, with exception of , r and B whose initial
conditions have not yet been specified at this point.
At the weak scale calculate , r and B by using the saddle point conditions for the
1-loop Higgs potential I V (H1 , H2 , S) = 0 (I = H1 , H2 , S) at H1 = H2+ = 0,
0 = v , S = x.
H1,2
1,2
Finally one can use the corresponding RGEs to determine the values of , r and B at
the GUT scale. Appropriate values of 0 , r0 and B0 can always be found because of the
2 We neglect the Yukawa couplings of the leptons and light quarks. This approximation is well justified in the
range of small and mediate tan which we consider in the following.
187
Fig. 1. Sketch of our procedure to fix the weak scale parameters using RGEs and saddle point
conditions.
effectively linear structure of Eqs. (2.2). Our procedure has the additional benefit that the
constraints coming from the top quark and Z-boson masses, MZ2 = (g12 + g22 )(v12 + v22 )/2,
are automatically built in. This reduces the number of free parameters by two so we are
finally left with the 7-dimensional parameter space
tan , x, , k, M0 , A0 , m20 .
(2.3)
Of course, not every parameter set leads to a viable phenomenology. The constraints
on the model parameters are related to the particle spectrum and the vacuum structure. In
the elimination procedure discussed before we assumed that the extremum parametrized
by v, tan and x is indeed the global minimum of the scalar potential. Thus we must
(numerically) verify that there are no deeper minima in V (H1 , H2 , S). If there appears
a deeper minimum the parameter set has to be discarded.
Additional constraints arise from the required absence of charge and color breaking
minima (CCB minima) deeper than the standard minimum, i.e., from the absence of
squark and slepton vevs. We check slepton and squark vevs induced by large trilinear
couplings [20,21]. Furthermore, there are CCB minima which come from negative scalar
mass squares, typically m2H2 . These dangerous directions involve Higgs, squark and slepton
fields (UFB directions). The condition for such a minimum not to be deeper than the
standard minimum implies a lower limit on the ratio m20 /M02 of O(1) [22]. However,
the decay rate of the standard vacuum into a minimum in the UFB direction is usually
negligible compared to the age of the universe. We therefore allow for a meta-stable
standard vacuum with respect to the UFB directions and disregard the corresponding
constraints.
Up to now, no superpartners of the SM particles have been detected. From the
experimental lower limits on the SUSY particle masses [23] various constraints on the
parameter space can be derived. The experimental limit on the chargino mass M >
1
90 GeV [24] translates into bounds on the universal gaugino mass and the parameter
|M0 | 100 GeV,
| + x| 80 GeV.
(2.4)
188
If the bound on M0 is satisfied the gluino mass M3 is automatically above its experimental
limit [23]. In the case of m0 > 100 GeV also the squark and slepton masses are compatible
with the experimental data.
Of particular importance are the properties of the lightest neutral CP-even Higgs
mass eigenstate h, which is a mixture of the Higgses and the singlet. If h has a large
singlet content its coupling to the Z-boson and thus its production cross section at LEP is
significantly reduced [25]. Parameter sets (2.3) cannot be ruled out by simply calculating
the lightest Higgs mass Mh . For example, if the Higgs production cross section is reduced
by a factor of 10 compared to the SM, Higgs masses down to about 60 GeV are still in
agreement with the data [26].
In Fig. 2 we present a scan of the M0 A0 plane for two different sets of (x, tan , k) [16].
The parameters M0 and A0 are of particular interest since they determine the values of the
trilinear couplings at the weak scale which will play a prominent role in the discussion of
the EWPT in Section 3. We have taken m0 > 100 GeV in order avoid light squarks and
sleptons. Choosing a small value of and a large value of k prevents the appearance of
very light Higgs bosons. In Fig. 2b we increased tan and the singlet vev to obtain large
Higgs masses up to 115 GeV. The increase of Mh with decreasing A0 can be traced to
the singlet diagonal entry in the Higgs boson mass matrix which is diminished by the Ak
contribution. On the same lines the increasing singlet content of the lightest CP-even Higgs
state with increasing A0 can be understood. In Fig. 2b this effect is reduced by the larger
values of |x| and k, which render the singlet state rather heavy.
The most important constraints on M0 and A0 arise from the chargino mass and from the
vacuum structure. The lower bound on M0 is a consequence of Eq. (2.4). The requirement
of the standard minimum being the global minimum of the Higgs potential implies the
upper and lower bounds on A0 displayed in Fig. 2. Constraints from the required absence
of A-term induced CCB minima are then automatically satisfied. Note that M0 is also
bounded from above. In Fig. 2 we excluded parameter sets which predict Mh < 65 GeV.
This leads to the upper bound on A0 in Fig. 2a. In Fig. 2b no additional constraint arises,
because of the larger Higgs masses. On the other hand, if we would allow for smaller Higgs
masses in Fig. 2a, still an upper bound on A0 would be implied by the vacuum structure,
as is the case in Fig. 2b. Taking into account the data from the Higgs boson search further
restricts the upper bounds on M0 and A0 . Roughly, the regions M0 > 600 GeV in the
tan = 5 set, and M0 > 1200 GeV in the tan = 10 set are additionally excluded
by the Higgs boson data [26]. These parameter regions are characterized by a very heavy
spectrum of SUSY particles. For that reason they are not very promising candidates for the
implementation of electroweak baryogenesis anyway, as will be discussed in Section 6.
We close this section by emphasizing some important differences between our model
and the Z3 -symmetric NMSSM, which is usually considered in the literature, e.g.,
in Refs. [19,25]. By setting = 0 Eq. (2.4) can be translated into a lower bound
on the singlet vev, |x| 125 GeV. However, the analysis of the Z3 -symmetric case
carried out in Ref. [19] shows that the actual lower bound on |x| is much larger,
once all phenomenological constraints are taken into account. Typically one has a singlet vev in the multi-TeV range and couplings 1 and k 1, with x and
189
(a)
(b)
Fig. 2. Scan of the M0 A0 plane for two different sets of (x, tan , k). The gray regions represent
the phenomenologically viable range of the parameters before cuts from the Higgs boson search are
applied. The dashed lines are curves of constant mass of the lightest CP-even Higgs boson. In the
dark-gray (light-gray) areas, the lightest Higgs boson is predominantly a singlet (Higgs) state.
kx = O(MZ ). In the next section we will discuss that in such a scenario the singlet is just a spectator during the EWPT, which proceeds more or less in the standard (i.e., MSSM) way. This property additionally motivates our inclusion of the
-term in the model, while it seems questionable whether the idea of electroweak baryogenesis can be realized in the Z3 -symmetric NMSSM. An additional peculiarity of the
Z3 -symmetric model is that large A-terms are required for successful electroweak symmetry breaking, leading to A20 > 9m20 . However, this parameter range is severely constrained
from the absence of CCB minima [21]. Some of the special features reported in Ref. [19]
may be due to the assumption of universal soft SUSY breaking used in that work. It seems
questionable, however, if the problematic bound, |x| MZ , can be evaded in the case of
non-universal soft terms [27], since the required large values of typically lead to small
190
values of the Higgs mass due to large Higgs singlet mixing. The NMSSM with and without Z3 symmetry also have different properties with respect to CP-violation, as will be
discussed in Section 5.
3. Strength of the electroweak phase transition

Order and strength of the EWPT are central questions in electroweak baryogenesis. Only
in the case of a first order phase transition (PT) the associated departure from equilibrium is
sufficient to induce a relevant baryon number production. To avoid baryon number washout
after the PT the even stronger criterion vc /Tc 1 has to be satisfied, where vc denotes
the Higgs vev at the critical temperature Tc . A first order phase transition is triggered by
cubic terms in the finite temperature effective potential [28]. In the (MS)SM these terms
arise from 1-loop thermal corrections of bosons and therefore are small from the very
beginning. Thus it is difficult to satisfy vc /Tc > 1. In the NMSSM, on the other hand,
trilinear terms enter already the tree-level Higgs potential due to Higgs singlet couplings,
leading to a significantly stronger EWPT [8,9,16]. These contributions stem from the soft
SUSY breaking couplings A and Ak , and from the -term

0 2 0 2
k
0 0
3

( S + h.c.) H1 + H2
(3.1)
+ A SH1 H2 + Ak S + h.c. .
3
All these trilinear terms explicitly contain the singlet field. Therefore, in order to induce
deviations from the (MS)SM behavior also the singlet vev must change during the phase
transition. Since at the EWPT thermal contributions to the effective potential are of the
order O(T 4 ) MZ4 , this requires the mass and the vev of the singlet also to be of the order
of the electroweak scale.
At finite temperature the effective potential of the neutral Higgs and singlet fields is
modified by the interaction with the hot plasma

(1)
VT H10 , H20 , S = Vtree H10 , H20 , S + V (1) H10 , H20 , S + VT H10 , H20 , S . (3.2)
In the 1-loop zero temperature corrections V (1) we include tops, stops and gauge bosons,
while in the 1-loop finite temperature part VT(1) also Higgs bosons, neutralinos and
charginos are taken into account. We do not make a high temperature expansion, as some
(1)
of the particles can be heavy in part of the field space. Rather, VT is evaluated using
a spline interpolation between the high and low temperature regions.
We stress that in the range of parameters we will study a strong first order EWPT is the
consequence of the tree-level terms in the Higgs potential (3.1). Since the stop mass will
turn out to be always larger than 200 GeV, the thermally induced cubic terms are too small
to account for vc /Tc > 1 [2]. The most important finite temperature effect is the appearance
of thermal effective masses
m2 m2 (T ) = m2 (T = 0) + const T 2 ,
(3.3)
where the constant encodes the couplings of the Higgs and singlet fields to the particles
in the plasma. It is this positive thermal contribution to the Higgs mass that makes
191
the symmetric phase stable at high temperatures, and thus causes the restoration of the
electroweak symmetry. Using the complete 1-loop expression rather than the simple high
temperature approximation, is just a convenient prescription to handle the decoupling of
heavy particles. Because of the dominance of the tree potential, we neglect contributions to
the thermal potential stemming from daisy resummation [29,30] and 2-loop diagrams [2].
In order to determine the strength of the PT one has to compute the critical
temperature Tc . We define Tc as being the temperature where the symmetric and the
broken minimum of the finite temperature Higgs potential (3.2) become degenerate. We
denote the Higgs and singlet vevs in the broken phase by (v1c , v2c , xc ). In general, the
singlet vev is different from zero even in the symmetric phase. We refer to a PT as being
strongly first order, if it avoids baryon number washout, according to the condition
vc /Tc > 1. Since both Higgs fields contribute
to the sphaleron energy, vc is given by vc =
2(|v1c |2 + |v1c |2 )1/2 , where the factor of 2 is due to our choice of field normalization.
Although CP-violation is essential for baryogenesis, we assume CP-conservation in this
section, i.e., all vevs, mass parameters and coupling constants are taken real valued. Of
course, we have to verify that VT has no deeper CP-violating minimum. Turning on phases
much smaller than one would induce only marginal changes in our results.
In this paragraph we study again the two parameter sets already discussed in the context
of Fig. 2 in Section 2. We determine the critical temperature by numerical minimization
of VT . After having checked that no CP-violating minima are present we disregard the
imaginary parts of the Higgs and singlet fields. 3 We then only have to minimize VT in the
real valued fields Re(H10 ), Re(H20 ) and Re(S).
Our investigations of the strength of the PT are summarized in Fig. 3 where the regions
of strong and weak PT in the M0 A0 plane are displayed for the two parameter sets of
Fig. 2 [16]. In case of tan = 5 and x = 100 GeV (Fig. 3a) the PT is strongly first
order in most part of the phenomenologically allowed range of parameters. However, the
corresponding Higgs masses up to 90 GeV are compatible with the experimental Higgs
mass bounds only because of the reduced Higgs production cross section due to Higgs
singlet mixing. In the parameter set of Fig. 3b we increased | tan | and |x| in order to obtain
larger Higgs masses. As a consequence, the region of weakly first order PT is enlarged.
However, a strong PT occurs still for a wide range of the parameters, while Higgs masses
up to 115 GeV are consistent with vc /Tc > 1.
In contrast to the SM or the MSSM where the PT definitely becomes weaker with
increasing Higgs masses the situation in the NMSSM is more complicated. Larger Higgs
masses can be related to a stronger phase transition. This may happen for example if the
broken and symmetric minima of the effective potential are almost degenerate already at
zero temperature. The strong PT at negative A0 and Mh > 100 GeV in Fig. 3b results
precisely from this effect.
Our results [16] are in reasonable agreement with those of Ref. [9], where the strength
of the EWPT in the general NMSSM has also been studied. The authors found that about
50% of their parameter sets were compatible with vc /Tc > 1, and also advocated for
3 Bubble walls in the presence of CP-violation will be discussed in Section 5.
192
(a)
(b)
Fig. 3. Scan of the M0 A0 plane for the two different sets of (x, tan , k) which have already been
considered in Fig. 2. In the dark-gray (light-gray) areas the PT is strongly (weakly) first order,
i.e., vc /Tc > 1 (vc /Tc < 1). The dotted lines are curves of constant mass of the lightest CP-even
Higgs boson. In the regions below (above) the dashed lines the lightest Higgs boson is predominantly
a Higgs (singlet) state.
x v in order to obtain a strong phase transition. While in Ref. [9] no restrictions with
respect to the SUSY breaking were made, we demonstrated that even with universal SUSY
breaking a strongly first order PT is quite natural in the general NMSSM. Furthermore,
we used updated (more restrictive) experimental bounds to constrain the parameter space.
A further difference comes from the procedure used in scanning the parameter space.
In Ref. [9] random shooting was applied, which only allowed for statistical statements.
Moreover, from the initially tested 105000 parameter sets just 2% gave a viable zero
temperature phenomenology, leaving merely about 2000 sets for the study of the PT. This
again demonstrates the usefulness of the systematic procedure for scanning the parameter
space, which we described in Section 2.
193
4. Shape of the phase boundary

In the previous section we defined Tc as being the temperature where the symmetric
and the broken minimum of VT become degenerate. However, tunneling with formation
of bubbles of the broken phase starts at some lower nucleation temperature, when the
symmetric phase is already meta-stable. The bubble wall profile will crucially enter the
calculation of the baryon production during the PT which will be discussed in Section 6.
Here we concentrate on CP-conserving bubble walls. The very important case of CPviolating wall profiles will be discussed in the next section.
At high temperatures the probability for thermal tunneling is proportional to eS3 /T ,
x ) describing
where S3 is the three-dimensional action of the static field configuration (
tunneling [31]. Here collectively denotes the Higgs and singlet fields. Assuming
spherical symmetry the bubble configuration (critical bubble) obeys the equation of
motion
d 2 2 d
VT () = 0.
+
(4.1)
2
dr
r dr
Further simplifications of Eq. (4.1) are justified if the tunneling occurs between two almost
degenerate minima of the potential with an energy difference +VT much smaller than
the height of the potential barrier. In such a case the radius of the bubble becomes much
larger than the thickness of the bubble wall, which thus is referred to as thin wall limit.
Neglecting therefore the d/dr term in (4.1) we are left with
d 2
VT () = 0.
(4.2)
dz2
We have replaced the spatial coordinate r by z, indicating that the solution to (4.2) may be
viewed as a planar domain wall with translational invariance in the directions perpendicular
to the z axis. During the period of stationary expansion the pressure induced by the energy
difference of the minima, +VT , is compensated by friction [67,68]. In the following we
model this effect by taking the effective potential at the critical temperature.
Notice that (4.2) is just the classical equation of motion of a particle moving from one
maximum of the turned around potential V () to the other, where is comes at rest. In
this picture z takes the role of the time coordinate and represents the configuration space
variable. Obviously, this a very delicate process, especially if more than one scalar field is
involved. Small changes of the initial conditions lead to a completely different shape of the
trajectory.
For a general effective potential the bubble wall equations have to be solved numerically.
In the case of only one scalar field the so-called overshooting undershooting procedure
can by applied: the initial value (r0 ) is tuned such that the trajectory approaches =
sym in the limit r , which then gives the bubble shape. This procedure can be used
for the critical bubble as well as for the domain wall configuration.
The situation is completely different once there are additional directions in field space.
Although in principle the shooting procedure is still applicable, in practice the required
high accuracy of the initial conditions cannot be achieved. Thus one has to devise
194
other numerical methods [5,3337]. They determine the bubble wall configuration as the
minimum of a functional F [] which is built from the squared equations of motion [35].
Details concerning the numerical algorithms used to minimize F on a lattice are given in
Refs. [33,35]. It turns out that the minimization procedure is reliable only if the starting
configuration, i.e., the ansatz for the wall shape, differs not too much from the actual
solution. Otherwise one keeps being stuck to local minima, which arise from discretizing
the functionals on a lattice [33]. Thus minimization starting from an arbitrary initial
configuration is not possible up to now, and finding an appropriate ansatz is very important.
In the case of the domain wall the kink solution fits the exact bubble shape reasonably well,
especially if every field is allowed for having its own wall thickness

z i
vi
1 tanh
.
i (z) =
(4.3)
2
Li
In general, different off-sets i which shift the fields against each other are possible as
well. For the domain wall case the mechanical analogue to energy conservation, E(z) =
1
2
2 (d/dz) V () = const, provides a very sensitive criterion to check the quality of the
numerically obtained solution.
In general, the numerical methods developed in Refs. [5,3336] are inevitable to
determine the bubble shape in multi-field models. But if the problem is effectively one
dimensional some improved shooting method can still be applied. The investigation
of CP-conserving bubble wall shapes in the MSSM [33,35] revealed that for realistic
Higgs masses the variation of tan in the bubble is very small, O(102 103).
Thus the bubble is very accurately described by taking only the combination of the Higgs
fields, H , which corresponds to the direction of the broken minimum. Simple shooting
along this direction provides excellent values for the wall thickness, the surface tension or
the action of the critical bubble. Even the small variation in can be reliably determined
by minimizing the effective potential along the direction perpendicular to H . The line
which connects these minima gives a good approximation to the trajectory in field space
the actual bubble solution is corresponding to. This can be checked by using the energy
conservation criterion. The kink ansatz turns out to be a good fit for bubble profile.
In case of the NMSSM the straight connection between the broken and the symmetric
minimum is no longer a reasonable approximation to the true bubble wall trajectory.
Ignoring the variation of tan , which turns out to be small and hence can be included
afterwards, we effectively have a two field system, H and S. In a large part of the parameter
space (as long as A0 is significantly above its lower bound shown in Fig. 2) the effective
potential is characterized by a distinct smooth ridge which connects the symmetric and the
broken minimum. This ridge S = F (H ) can be determined by minimizing the potential
along the direction perpendicular to the straight connection of the symmetric and broken
minima. 4 Using VT (H, F (H )) the set of bubble wall equations (4.2) reduces to only one
differential equation which can again be solved via the shooting method. However, not
4 Numerically, the following method leads to better results: determine the saddle point along the ridge, i.e.,
H VT = S VT = 0. The ridge is then defined as the trajectory of an over-damped particle rolling from the saddle
point down to the (symmetric or broken) minimum, satisfying d/dz + VT = 0.
195
every function F is a possible candidate for the actual bubble wall. Since dH /dz vanishes
in the symmetric and broken phase, integrating the bubble wall equation results in the
constraint
H
brk
dH

VT H, F (H ) = 0.
H
(4.4)
(H ) = F (H ) + f H (H vc ),
In order to evade this inconsistency we deform our ansatz by F
where the free parameter f is fixed by Eq. (4.4). This ansatz is motivated by the mechanical
analogue of the bubble wall problem: if a ball rolls from one maximum of VT to the other,
it moves a bit below the ridge in order to compensate for the centrifugal force. On half

way between the two maxima the centrifugal force is most efficient and the trajectory F

deviates maximally from the ridge F . In practice, F F is small compared to the deviation
of F from the straight line.
(H ) one could just as well eliminate H by F
1 (S).
Instead of expressing S in terms of F
corresponds to the trajectory of the actual bubble configuration the two prescriptions
If F
the two prescriptions will lead to different results for
are equivalent. For a general F
the wall profile. The difference between the two solution indicates the quality of the
. We emphasize that although several steps are necessary to compute the bubble
chosen F
wall shape via this improved shooting approach, the required numerics is fairly simple. It
can by carried out using computer algebra systems like M APLE or M ATHEMATICA. The
results are competitive to those obtained with the sophisticated minimization algorithms
of Refs. [33,35], which on the other hand were very helpful to check the validity of our
approximations.
Let us discuss a specific example from the parameter set of Fig. 2a with M0 = 300 GeV
and A0 = 0. It corresponds to a lightest Higgs mass Mh = 80 GeV, which because of the
reduced coupling to the Z-boson is compatible with the experimental data. We have Tc =
109 GeV and vc /Tc = 1.12, i.e., the washout of baryon number after the phase transition is
avoided. Using our improved shooting method we find the bubble wall profile displayed in
Fig. 4a. In Fig. 4b we show the corresponding trajectory in field space which considerable
deviates from a straight line. Fitting the numerical solutions by the kink ansatz (4.3) we
obtain the wall thicknesses for the Higgs and singlet fields, Lh = 0.13 GeV1 = 14/Tc and
Ls = 0.10 GeV1 = 11/Tc , respectively. For the surface tension we find = 46300 GeV3 .
The energy conservation check gives +E = 6000 GeV4 which has to be compared to
the barrier height Vb 125000 GeV4 . This is the same level of accuracy [37] which
is accessible in the minimization approach of Ref. [33]. On the other hand, taking the
, the dashed straight line in Fig. 4b, results in Lh = Ls = 8/Tc and
simplest ansatz for F
3
= 82200 GeV , which is already off by about a factor of two.
Up to now we neglected the variation of tan in the bubble wall which has a strong
impact on some sources for baryogenesis, as will be discussed in Section 6. In order to
include this effect in our calculation we simply minimize the effective potential in the
direction orthogonal to H while keeping H and S fixed, implicitly assuming that it is only
(H ). In Fig. 5a we show the variation of tan for
a small perturbation to the trajectory F
196
Fig. 4. (a) Bubble wall profile at the critical temperature for the parameter set of Fig. 2a with
M0 = 300 GeV and A0 = 0. (b) The corresponding trajectory in the H S plane (solid line) and
the straight connection between the symmetric and the broken minimum (dashed line). (All units in
GeV.)
Fig. 5. (a) Variation of tan in the bubble wall at the critical temperature for the parameter set of
Fig. 4. (b) Shape of the critical bubble for the same parameter set at T = 108.5 GeV. (Units in GeV.)
the bubble wall of Fig. 4. We obtain = 1.2 103 , i.e., the assumption of being a
small perturbation is very well justified. Our result is in complete agreement [37] with the
one computed by using the minimization technique of Ref. [33]. Moreover, we find that
the variation of tan in the NMSSM and MSSM [35] are of the same order of magnitude.
Particularly, the singlet field provides no additional sources for which can be traced to
its equal couplings to both Higgs fields.
Finally, our improved shooting method can be used to determine the shape of the critical
bubble (4.1). The first bubbles nucleate when the action of the critical bubble satisfies
S3 (Tn )/Tn 130140 [32]. In the computation of the Tn the small tan variation in the
wall can be safely ignored. Considering again the parameter set of Fig. 4 we obtain the
bubble configuration shown in Fig. 5b, where T = 108.5 GeV. Since Tc = 109.2 GeV
the super-cooling amounts to 0.7 GeV, the same order of magnitude as in the MSSM
[35]. Notice that the wall shape is still very similar to a kink solution (4.3), which is cut
off at r = 0. Calculating the corresponding action we get S3 /T = 134. Thus we are at
197
i 3
2
the nucleation temperature. Note that the thin wall approximation S3 = 16P
3 /(+VT ) ,
leading to S3 /T = 94 is not very accurate in this case. Here +VT = 4.04 105 GeV4 is the
potential barrier. Since in our example bubble radius and wall thickness are of comparable
size this difference is not surprising.
We emphasize that our method to determine bubble shapes is restricted to cases where
the bubble trajectory is associated with a smooth ridge in the effective potential, which
correspond to values of A0 significantly above its lower bound. Otherwise one is referred
to the sophisticated minimization methods of Refs. [33,35,36].
5. CP-violating bubble walls

Having established a strongly first order EWPT in the NMSSM, successful electroweak
baryogenesis still depends crucially on the available amount of CP-violation. In models
containing two Higgs doublets, such as the MSSM or the NMSSM, complex vevs of
the Higgs (and singlet) fields provide additional sources of CP-violation. Complex vevs
can either arise spontaneously or are induced by explicitly CP-violating couplings. In this
section we discuss their impact on bubble wall profiles in the NMSSM. Most interestingly,
we find that CP-violation may be restricted to the phase transition itself, a phenomenon
which is called transitional CP-violation. In particular, this scenario allows for a large
amount of CP-violation during the process of baryon production, while being completely
unconstrained by experiment.
To start with, let us summarize the case of CP-violation at zero temperature. In the
MSSM the tree-level Higgs potential automatically conserves CP. However, explicit CPviolation emerges from possible complex phases of the soft SUSY breaking A-terms and
gaugino masses, and from the -parameter [4]. These phases contribute to the electric
dipole moments (EDMs) of quarks and electrons [38]. From the experimental upper limits
[39] on the neutron EDM, dn < 1.1 1025 e cm, and the electron EDM, de < 4.3
1027e cm, constraints on the supersymmetric phases can be derived. Conservatively,
rather small supersymmetric phases of the order O(102 103) are required to satisfy
the experimental bounds [38]. Larger phases may only be tolerated if the first and second
generation squarks have masses in the TeV range [40], or if accidental cancellations do
occur. Recently it was realized that these cancellations are more generic than thought
previously [41]. Explicit CP-violation occurring in the SUSY breaking sector of the MSSM
is communicated to the Higgs sector by radiative corrections. However, the complex
phases induced in the Higgs vevs are much too small to have any phenomenological
implications [42]. Nevertheless, Higgs phenomenology may be significantly changed due
to mixing of the CP-even and CP-odd Higgs states induced by complex mass parameters
and coupling constants [4].
The Higgs sector of the Z3 -symmetric NMSSM contains one not removable phase which
can be chosen to be the phase of k . This phase is not very much constrained provided ||
is significantly smaller than one [43]. If we allow for Z3 -symmetry breaking, six possibly
198
complex parameters appear in the tree-level Higgs potential. 5 Two phases can be absorbed
by a redefinition of the Higgs and singlet fields. Thus we can take B + r and kr
being real valued, without loss of generality. We parametrize the Higgs and singlet fields
according to
H10 = h 1 ei1 ,
H20 = h 2 ei2 ,
S = seiS ,
(5.1)
i , etc. Furthermore, we introduce

and define = e
i , = e
= 1 + 2 ,
= 1 2 .
(5.2)
Because of gauge invariance the Higgs potential is independent of the phase combination .
Using these definitions the tree-level Higgs potential takes the form

s cos(S + ) h 21 + h 22 + 2 h 21 h 22 + k 2 s4
Vtree = 2 + 2 s 2 + 2
2
g12 + g22 2
h 1 h 22
8
+ 2(r + B)h 1 h 2 cos + 2kr s 2 cos(2S ) + m2H1 h 21 + m2H2 h 22 + m2S s2
2
+ 2 A s h 1 h 2 cos( + S + + A ) + k A k s3 cos(3S + k + Ak ).
3
(5.3)
Note that in contrast to the MSSM the phase of the -parameter enters Vtree . Explicit
CP-violation automatically generates complex vevs for the Higgs and singlet fields. In
the general NMSSM this is a tree-level effect which should be taken into account. An
example will be discussed below. EDM constraints on CP-violating phases in the general
NMSSM depend on the coupling of these phases to the (MS)SM sector. Phases which
directly enter the squark, chargino and neutralino mass matrices, e.g., and
, have
to obey the MSSM bounds which have been discussed above. The other phases, e.g., k
and S
, are less constrained, provided the coupling to the MSSM sector, , is sufficiently
small [43].
Even when the Lagrangian is CP-conserving, the CP-symmetry may be spontaneously
violated by complex scalar vevs, i.e., CP and the SU(2) U (1) gauge symmetry are
broken together at the EWPT. In the MSSM radiative corrections to the Higgs potential
can induce a CP-violating vacuum. However, the required Higgs boson with mass of a few
GeV is obviously ruled out by experiment [42]. The Z3 -symmetric NMSSM provides only
limited improvement on the minimal model: radiative corrections generate spontaneous
CP-violation, provided the lightest neutral Higgs mass is smaller than about 40 GeV [44].
In the general NMSSM the situation is very different. Spontaneous CP-violation (SCPV) is
possible even at tree-level. However, Higgs spectra consistent with the experimental Higgs
mass bounds require nearly maximal CP-violation, i.e.,
, S
of the order O(1) [45].
Phases 0.1 are only possible for Higgs masses smaller than 30 GeV. This finding is
related to the fact that CP is a discrete symmetry. If CP is spontaneously broken there arise
+ 2 k s2 h 1 h 2 cos( 2S + k ) +
5 In this section we relax the assumption of universal soft SUSY breaking which would induce correlations
between the various phases at the weak scale.
199
two degenerate minima 6 with phases

, which are nearby when
becomes small.
From the SM Higgs potential it is known that the squared mass of the Higgs boson is
proportional to h
2 , where denotes the quartic coupling. In the case of small SCPV
we can identify h
2 v 2
2 , which up to numerical prefactors of order unity is the
mass of the emerging light, almost CP-odd Higgs boson. Obviously, for small
this state
becomes arbitrarily light.
Thus SCPV in the NMSSM does not appear to be very promising. Small phases are ruled
out because they require a light Higgs boson, while large phases are tightly constrained by
the EDM experiments. We will see below that this conclusion does not apply to SCPV
which is only present during the PT.
We now derive the equations of motion for moduli and phases of the Higgs fields. 7
Using the definitions (5.1) and (5.3) we can rewrite the kinetic terms for the Higgs bosons
in the Lagrangian according to
L = h 1 h 1 + h 2 h 2 +
+
h 21 h 22
+ .
2

h 21 + h 22
+
4
Since VT = 0 the EulerLagrange equation for implies

2
h 1 + h 22 + h 21 h 22 = c = const.
(5.4)
(5.5)
cannot be set to zero without introducing a pure

In general, the integration constant
gauge field. However, a non-vanishing c would enhance the energy of a bubble
configuration, so we will set c = 0 in the following. After elimination of by help

of (5.5) the equations of motion for h 1 , h 2 and take the form
2h 1 h 42 d 2
d2
VT = 0,
h1 + 2
2
dz
(h 1 + h 22 )2 dz2
h 1
2h 2 h 41 d 2
d2
VT = 0,
2 2 h 2 + 2
dz
(h 1 + h 22 )2 dz2
h 2

d 2h 21 h 22 d
VT = 0,
dz h 21 + h 22 dz
(5.6)
(5.7)
(5.8)
where we restricted ourselves to the case of the static domain wall perpendicular to the
z-direction (4.2). The equations for the critical bubble follow along the same lines. In
Eqs. (5.6)(5.8) the phase is dynamical only if both moduli, h 1 and h 2 , are different
from zero. This demonstrates that in the bubble wall cannot be assigned to one of the
Higgs fields.
6 For that reason explicit CP-violation has also to be present in the model in order to avoid domain walls when
CP is broken at the EWPT. However, this explicit phase may be much smaller than one.
7 The singlet field is most conveniently split into real and imaginary part which can be treated along the lines
discussed in the previous section.
200
In the previous section we noted that tan varies only slightly in the bubble wall.
Neglecting this small perturbation Eqs. (5.6)(5.8) can be further reduced and we obtain
for the variation of the phases of the Higgs fields H10 and H20
1 = sin2 (T ),
2 = cos2 (T ).
(5.9)
Here tan T represents the ratio of the Higgs field moduli in the broken minimum at the
critical temperature. In the limit of large tan(T ) the variation of is almost completely
absorbed by 1 , while 2 remains more or less constant. For example, in case of tan T = 5
we obtain 2 = 0.038 , which is small even for = O(1). In Section 6 we will discuss
that the variations of complex phases fuel baryon production, rather than the complex
phases themselves. As a consequence, baryon number generation resulting from a varying
i2 , will turn out to be highly suppressed
phase in the top quark mass, mt = yt sin(T )he
in the limit of large tan(T ). Since in the region of the parameter space considered in this
work the Higgs vev ratio changes only by a few percent when T is raised from zero to the
critical temperature, this is the case in the regime of large tan .
We now discuss the implications of the general equations of motion derived above. Let
us start with explicit CP-violation. In the MSSM the complex phases which are induced
in the Higgs vevs by CP-violating couplings are completely negligible for realistic Higgs
masses. In Ref. [5] we found this behavior confirmed at finite temperature: O(103 ),
even for explicit phases of the order of one. The variation of in the bubble is of the same
order of magnitude. We conclude that in the MSSM with explicit CP-violation, the Higgs
vevs can be taken real to very good approximation.
In the NMSSM the situation is different, because explicit CP-violation is possible in the
tree-level Higgs potential (5.3). In general, the behavior of the complex valued Higgs and
singlet fields in the bubble wall can only be reliably determined by using the minimization
techniques of Ref. [33]. In order to make this possible, the equations of motion which
describe s = Re(S), c = Im(S), h 1 , h 2 and have to be included in the F . In Ref. [5]
this approach was applied to compute CP-violating bubble wall profiles in the MSSM and
NMSSM. It turned out that also the phase is rather accurately described by a kink-ansatz
(4.3). Moreover, it was verified that in the calculation of the wall shape small CP-violating
phases O(101 ) can indeed be treated as perturbations of a CP-conserving solution.
In this case the bubble wall profile can be conveniently determined in two steps: first, one
computes the CP-conserving profile of the fields h 1 , h 2 and s, where the CP-violating fields
and c are set to zero. In a second step, the CP-violating constituents of the bubble wall
may be computed by simply minimizing the effective potential VT (h 1 , h 2 , , s, c) with
respect to and c, while keeping h 1 , h 2 and s fixed. The variation of the Higgs vev ratio
can be neglected in this calculation. We emphasize that this prescription works only in case
of small CP-violation in the Higgs and singlet fields.
If VT (h 1 , h 2 , = 0, s, c = 0) has a smooth ridge, the profile of the CP-conserving
fields can be obtained from the improved shooting method discussed in Section 4.
Hence the complete CP-violating bubble profile can be computed without minimizing the
functional F .
201
Fig.
6. Variation of the CP-violating fields (a) and c (b) in the bubble wall as a function of
h = h 2 + h 2 at the critical temperature for = 0.1, A0 = 100 GeV and M0 = 125 GeV. The
1
2
remaining parameters are chosen as in Fig. 2a. (Units in GeV.)
In Fig. 6 we display an example of a bubble wall in presence of explicit CP-violation

using = 0.1, A0 = 100 GeV and M0 = 125 GeV, the remaining parameters are chosen
as in Fig. 2a. The lightest Higgs boson has a mass Mh = 86 GeV. We find Tc = 101 GeV
and vc /Tc = 1.77, so the PT is strongly first order. For the wall thickness we obtain Lw =
5/T . The CP-even singlet field evolves from s = 115 GeV in the symmetric phase to
s = 79 GeV in the Higgs phase. From Fig. 6a we take = 0.0182 in the broken phase,
and = 0.0065. In the MSSM, 1 would be required to induce an effect of the same
magnitude. Also the CP-odd component of the singlet field, c, varies in the bubble wall, as
is shown in Fig. 6b. In the symmetric phase we have c = 0, since couples to the singlet
only in case of non-vanishing h 1,2 .
We now turn to the case of spontaneous CP-violation. Although this scenario is
disfavored at zero temperature, the Higgs and singlet fields may still acquire large complex
phases at high temperatures, or even only during the EWPT. Since it will turn out in
Section 6 that explicit CP-violation can account for the observed baryon asymmetry only
in special regions of the NMSSM parameter space, spontaneous CP-violation at finite
temperature becomes an interesting alternative scenario.
In the Z3 -symmetric NMSSM spontaneous CP-violation at finite temperature has
already been investigated some years ago in Ref. [46], where an effective potential
truncated at the renormalizable level was used. It was found that spontaneous CP-violation
cannot occur in the finite temperature broken phase while having a viable temperature
zero phenomenology, since thermal corrections to the effective potential have the tendency
to restore symmetries. The authors suggested that spontaneous CP-violation may take
place during the EWPT (transitional CP-violation), i.e., the Higgs and singlet fields
may acquire complex phases in the bubble wall, although no definite example was given.
However, the truncated effective potential used in their analysis proved to give misleading
results in case of the MSSM [5], as will be discussed below.
In our analysis of finite temperature spontaneous CP-violation in the general NMSSM
we use the full 1-loop effective potential (3.2) without any truncations. It turns out that
202
Fig. 7. Transitionally CP-violating bubble wall profile from the parameter set of Table 1. The position
variable z is given in units of GeV1 .
Table 1
Parameter set used in Fig. 7 (all dimensionful parameters in GeV)
x
tan
M0
A0
m0
Ak
m2S
150
0.05
0.4
100
100
200
150
50
2000
spontaneous CP-violation does not occur for the universal pattern of SUSY breaking
discussed in Section 2, which induces too large masses for the CP-odd Higgs bosons.
However, spontaneous CP-violation can take place if universality is violated in the singlet
sector. In this case A , Ak and m2S are free parameters. 8 Fig. 7, which is taken from
Ref. [5], shows an example of a transitionally CP-violating bubble wall for the parameters
given in Table 1, which was computed using the minimization algorithm of Ref. [33]. More
precisely, the scenario corresponds to a phase transition between an electroweak symmetric
high temperature phase, where CP is violated by a complex singlet vev, and an electroweak
broken, CP-conserving low temperature phase, i.e., (s, c)sym = (50 GeV, 99 GeV)
(s, c)brk = (122 GeV, 0). The critical temperature is found to be 101 GeV, and we have
a strong first order PT with vc /Tc = 1.6. The mass of the lightest CP-even Higgs boson is
84 GeV which because of its reduced coupling to the Z-boson is still compatible with the
experimental data. There is also a rather light CP-odd state with a mass of 105 GeV, which
is almost a pure singlet. The bubble wall is found to be rather thin with Lw 3/T . In our
example spontaneous CP-violation is triggered in the singlet sector. The small coupling
communicates it to the Higgs fields. As a consequence, the almost maximal phase of the
singlet field, s , induces only a small phase 1/20 s in the Higgs sector. Up to now
we only identified one particular region in the NMSSM parameter space where transitional
8 It is even sufficient to violate universality only via m2 .
S
203
CP-violation occurs. A more systematic study would be very desirable. Especially it would
be interesting to find an example where transitional CP-violation originates from the Higgs
fields rather than the singlet field. In that case is expected to be of the order one.
Let us finally comment on transitional CP-violation in the MSSM, where there also have
been claims in the literature that transitional may occur [47]. In Ref. [48] even an example
of a transitionally CP-violating parameter set has been presented. All these studies were
carried out using an effective potential that has been truncated at the renormalizable level,
i.e., the logarithmic terms, etc., were neglected. In Ref. [5] we performed a systematic
search for transitional CP-violation in the MSSM using the full 1-loop thermal effective
potential without finding any viable parameter set. In particular, the example of transitional
CP-violation presented in Ref. [48] could not be confirmed. This discrepancy appears to
be due to the different approximations which have been made concerning the effective
potential. Our results indicate that the truncated effective potential is not appropriate to
discuss transitional CP-violation.
6. Baryon asymmetry in the semi-classical limit

6.1. Generalities
Between the initial nucleation and the completion of the phase transition expanding
bubbles convert the symmetric phase into the broken phase. The early universe is filled
with a hot plasma of particles, most of them have rather different masses and mixings inside
and outside the bubble. Thus the bubble wall behaves like a potential on which the particles
scatter. At the bubble surfaces the plasma is thrown out of equilibrium by the motion of the
phase boundary. The higher pressure inside the bubble tends to accelerate the wall, while
the interaction between the wall and the particles in the plasma dissipates energy and slows
the wall down. Finally, a stationary situation is reached, where the different forces balance
and the wall propagates with constant velocity vw . In general, the wall velocity depends on
the shape of the effective potential and on the composition of the plasma.
In the symmetric phase baryon number violation occurs frequently due to hot sphaleron
processes. A baryon asymmetry is produced, if the departure from equilibrium at the phase
boundary biases the baryon number violating processes in a CP-violating fashion. The most
efficient mechanisms of baryon number generation rely on transport. If CP is violated in
the interaction between the bubble wall and the particles in the plasma, different population
densities for particles and antiparticles are induced in front of the wall. The difference
between the particle and antiparticle populations is then transported into the symmetric
phase, where it biases rapid baryon number violation. The total amount of baryon number
which gets produced during the phase transition depends crucially on the shape and motion
of the bubble wall.
Different methods have been suggested in the literature in order to describe the effects
of CP-violating interactions between particles in the plasma and the propagating bubble
wall, which finally generate CP-violating source terms that enter the diffusion equations
204
for particle transport. Depending on the properties of the bubble wall, in the first place
its velocity vw and thickness Lw , different approximations may be applied. The most
rigorous approach is based on the closed time-path (CTP) formulation of non-equilibrium
quantum field theory [50]. It leads to a set of quantum Boltzmann equations [51] which
describe the temporal evolution of particle densities including particle number changing
interactions and CP-violating source terms [52]. However, the approximations and results
are still controversely discussed. There have also been efforts to use the CTP formalism
only in calculating the CP-violating source terms which in turn are inserted into classical
Boltzmann equations [49].
If the thickness of the bubble wall Lw is smaller than the mean free path l of the
particle under consideration (thin wall), one may neglect the influence of the plasma
during the scattering of the particle off the wall. The interaction of a fermion or boson with
the wall can then approximately be described by using a free Dirac or KleinGordon
equation, respectively. CP-violation is encoded in different reflection and transmission
coefficients for particles and antiparticles [5357]. In Ref. [58] also the effects of thermal
scattering were taken into account, which due to decoherence have a negative impact on
the generation of a CP-violating observable [59]. Baryon production is fueled by a net
flux of charge into the symmetric phase (charge transport mechanism [53]), which is
induced by the reflection asymmetry. This current is subsequently inserted into a set of
classical Boltzmann equations which determines the evolution of the particle distributions.
Recently, reflection and transmission probabilities have also been calculated by using
standard quantum field theory methods [60]. There, thermal scattering, i.e., damping, was
included via an imaginary part of the self-energy.
As the bubble wall becomes thicker, Lw l, interactions with the plasma must
inevitably be taken into account, when a particle encounters the propagating wall. In
general, this requires non-equilibrium quantum field theory methods [52]. However, if
Lw 1/T , most particles have inverse momenta 1/p Lw and may therefore be
treated semi-classically [6164]. Using the WKB approximation one can derive dispersion
relations which in the presence of CP-violation are different for particles and antiparticles.
The dispersion relations then enter classical Boltzmann equations, leading to a unified
description of CP-violating source terms, particle scattering and transport. Recently, it was
clarified how the semi-classical description arises from the quantum Boltzmann equations
in the limit of a slowly varying background field (thick wall regime) [65]. Of course,
even in the thin wall limit the high-momentum particles may be described semi-classically.
However, in that case they only give a sub-leading contribution to the CP-violating source,
which in this case is dominated by the low-momentum particles. Even in case of thick
walls there may be important contributions from low-momentum modes neglected in the
semi-classical treatment, which according to Ref. [52] significantly enhance the produced
amount of baryon number. There is also a very recent calculation [66] showing that
gauge fields in the hot phase lower significantly the wall velocity, such also leading to a
more efficient baryon production. In Ref. [67] the semi-classical approximation to particle
dynamics has been applied to calculate the bubble wall velocity in the SM. In Ref. [68] it
was shown that the stops in the MSSM considerably lower the wall velocity.
205
Notice that different particle species with different (gauge) interaction have different
mean free paths. Thus whether one is in the thin or thick wall regime depends on the
particle species under consideration. In the (N)MSSM especially particles with strong
interactions (quarks and squarks) require a thick wall treatment.
There is some disagreement among the groups that have estimated the baryon
asymmetry generated during the EWPT in the MSSM. In Refs. [49,52,58] the CP-violating
source terms are proportional to the variation of tan in the wall, which according to the
discussion of Section 4 causes a suppression of at least O(102 ). This dependence is due to
taking into account only the leading order of the Higgs insertion expansion, which has been
used to calculate source terms. At higher orders in the expansion there arise contributions,
which escape the suppression, as was recently shown in Ref. [60]. However, these
corrections turn out very small and are competitive to the leading order contributions only
in case of < O(103 ). In the recent paper [71] it was argued convincingly that the
(tan ) source cancels and that one has to consider closely a source term symmetrical in
the two Higgs fields.
In this section we generalize the method introduced in Ref. [64] to calculate the baryon
asymmetry in the NMSSM, which has not been considered so far. Different from Ref. [64]
our formulas also cover the case of CP-violating bubble walls. We obtain a non-vanishing
baryon asymmetry even in case of constant tan , which is due to the variation of the singlet
field or the presence of CP-violating bubble walls.
6.2. WKB approximation and dispersion relations
The idea of deriving CP-violating source terms using a semi-classical approximation was
developed in context of the two Higgs doublet model [63] and afterwards applied to the
charginos in the MSSM [64]. In the following we review the derivation of semi-classical
dispersion relations and generalize this method to account for the effects of CP-violating
bubble walls, which potentially are present in the NMSSM. These additional sources of CPviolation will turn out to be very helpful in order to generate a baryon to entropy ratio in the
observed range: B = nB /s 27 1011 [69]. We also give the dispersion relations in
the bosonic case. In the next section these dispersion relations enter the classical Boltzmann
equations that describe particle transport induced by the propagating bubble wall.
6.2.1. The fermionic case
We start the derivation of fermionic dispersion relations in presence of CP-violation
with a simple example: A single (Dirac-)fermion DT = ( , ) that couples only to one
of the two Higgs doublets, which we denote by H . The most prominent realizations are
the quarks and leptons of the (N)MSSM in the presence of a CP-violating bubble wall.
Due to its coupling to the Higgs, y, the fermion obtains a mass proportional to the Higgs
vev, M = yH . During the passage of the bubble wall the fermion mass becomes spacetime
dependent. Assuming that the bubble has grown to macroscopic size and reached its final
velocity, we can neglect the curvature of the wall and boost to the rest frame of the bubble.
Then the fermion mass only depends on one position coordinate, which we denote by z,
206
i.e., M = M(z). As a consequence, the energy of the particle, E, and its momentum
perpendicular to the z-direction, p , are constants of motion. Writing M = mei , possible
CP-violation is encoded in a non-vanishing phase . 9 Any constant can be absorbed by
a redefinition of the fermion field. When couplings to additional fields are neglected only
a varying phase contributes to CP-violation.
In the presence of a complex mass term the fermion field is described by the free Dirac
equation

M
i
= 0.
i PL M PR M D
(6.1)
i M
Notice that at this level all interactions between the particle and the plasma are neglected.
The scattering effects will be accounted for in the next section when Boltzmann equations
are written down that describe the local phase space distributions. We are working in the
chiral representation of the -matrices. Exploiting conservation of energy and boosting to
the Lorentz frame where p = 0 we can take the ansatz D = eiEt (z) and are left with
a one-dimensional problem. The interaction between the fermion and the wall conserves
the z-component of the spin, Sz . Thus Eq. (6.1) splits into two equations [55]

E
m(z)ei(z)
,
iz = Q(z) , Q(z) =
(6.2)
m(z)ei(z)
E
where + = (1 , 3 ) and = (2 , 4 ) are the Sz = 12 components of . To solve Eq. (6.2)
one brings the z-dependent matrix Q(z) = D(z)QD (z)D(z)1 into a diagonal form,
where [64]

2
E m2
0
cosh X
ei sinh X
,
D
=
QD =
(6.3)
cosh X
0
E 2 m2
ei sinh X
and tanh 2X = m/E. In the local helicity basis, = D 1 , the Dirac equation (6.2)
takes the form

i h z = QD D 1 i h z D
(6.4)
which still is an exact equation. In general, the correction term D 1 i h z D caused by the
position dependent field redefinition is not of diagonal form. The two components of
are still coupled. However, the off-diagonal part is proportional to z D D/Lw . Typical
momenta of the particles in the plasma are of the order of the temperature T , which is
much larger than 1/Lw for the bubbles under consideration. We therefore expand Eq. (6.4)
in powers of z or more precisely in powers of h (WKB approximation) that we already
reintroduced for that reason.
To order (h )0 we can neglect the D 1 hi
z D contribution. Thus the two components of
decouple in Eq. (6.4). Inserting the WKB ansatz for the fermion field

1 i z pz (z ) dz
0 i z pz (z ) dz
(2)
(1) =
h
e
,
=
e h
(6.5)
0
1
9 Here denotes the phase of the Higgs boson the fermion is coupling to. It should not be mixed up with the
common phase of both Higgs fields defined in Eq. (5.3) which is denoted by the same symbol.
into (6.4), we obtain the dispersion relations pz (E)

+(1) , (2) : pz = E 2 m2 ,

+(2) , (1) : pz = E 2 m2 .
207
(6.6)
The momenta pz are the eigenvalues of the matrix entering the RHS of Eq. (6.4). The
eigenfunctions may more transparently be labeled by the chirality states they correspond to
(1)
in the limit m 0,
e.g., + 1 1 , etc. The dispersion relations may then be combined
to pz = sgn(pz ) E 2 m2 which holds for left-handed particles and right-handed
particles . Obviously, in the classical limit the -dependence completely disappears,
demonstrating that CP-violation is indeed a quantum-mechanical phenomenon.
To solve the Dirac equation (6.4) to order h we have to take into account the D 1 hi
zD
term which reintroduces a coupling between the two components of . The dispersion
relations pz (E) are obtained from the eigenvalues of the matrix QD D 1 i h z D. Since
to order h the off-diagonal terms do not contribute, we are left with [64]

2
L(): pz = sgn(pz ) E 2 m2 h
sinh X,

R( ): pz = sgn(pz ) E 2 m2 + h sinh2 X,
(6.7)
where = z and
E E 2 m2
sinh X =
(6.8)
.
2 E 2 m2
Again the states are labeled by their asymptotic chirality properties. Notice that the
variation of m, which is encoded in z X, drops in the dispersion relations. The CP-violating
part of the dispersion relation, +pz = h sinh2 X, is proportional to the derivative of the
phase . Thus only a varying phase contributes to CP-violation in the semi-classical limit.
Furthermore, CP-violation is proportional to sinh2 X, which guarantees that its effect is
turned off in the limit m 0, where is no longer well defined. Because of the different
dispersion relations, left- and right-handed particles feel a different (semi-classical) force in
and (R),
the CP-violating part
their interaction with the wall. For the antiparticles, (L)
comes with the opposite sign. Besides the force term there is a second manifestation of
CP-violation: The phase enters also the transformation matrix to the helicity basis D.
Since the interaction eigenstates are different from the helicity states, particles and
antiparticles interact in a different way with the surrounding plasma. This generates a CPviolating source term which drives spontaneous baryogenesis [70].
As pointed out recently in Ref. [71], one should better use the kinetic momentum pkin =
mvgroup = mE/p instead of the canonical momentum p (which we used up to now) in
the quasi-classical limit of particles in the Boltzmann transport equations. This is quite
in the spirit of the correspondence principle of basic quantum mechanics. Calculating the
(inverse) group velocity from (6.7), and using (6.8), the kinetic moment beyond the zeroth
order contains (order h ) correction terms
2
+pkin =
h
m2
.
2E E 2 m2
(6.9)
208
In the following we will need the dispersion relation for energy in terms of momentum
which to order h is obtained as
m2
2 + m2 sign(p )
E E0 +E = pkin
(6.10)
.
z
2 + m2 )
2(pkin
E+ is the energy of left-handed particles and antiparticles, whereas E corresponds to
right-handed particles and antiparticles. 10 In the derivation of Eq. (6.10) we transformed
to a general Lorentz frame with non-zero momentum parallel to the wall. When boosting
to the plasma frame the dispersion relation (6.10) is preserved to linear order in vw and .
In our derivation of the dispersion relations we followed basically Ref. [64]. CPviolation arises because of a position dependent phase in the transformation from the
interaction (i.e., chirality) states to the local mass (i.e., helicity) eigenstates. In Ref. [63]
a slightly different approach was used to obtain the dispersion relation of a fermion
(e.g., top quark) in the presence of a CP-violating Higgs field background, H , in context
of the 2HD model: the complex phase in the fermion mass was removed by a gauge
transformation, which induced a gauge field in the kinetic term of the fermion. The
dispersion relation obtained with this technique is completely analogous to our result (6.7).
However, it is not clear how to generalize this method to cover the case where several
species mix, as occurs with the gauginos and higgsinos in the (N)MSSM, or where the
NMSSM singlet field background contributes to CP-violation. In both situations the CPviolating phase cannot be removed by a gauge transformation. On the other hand, the
method described above is still applicable [64].
We are now in the position to address the problem of mixing Dirac fermions DI , where
the flavor index I = 1, . . . , N . In order to solve the corresponding Dirac equation we
diagonalize the matrix

1E M (z)
Q(z) =
(6.11)
M(z)
1E
which is a straight forward generalization of (6.2). Here 1 denotes the unity matrix in flavor
space and M is a position dependent complex mass matrix for the fermions, without further
specifications. As a first step we write M and M in terms of two unitary matrices U and V,
and a diagonal matrix MD = mei
M = VMD U ,
M = UMD V .
(6.12)
U and V may be obtained by diagonalizing the hermitian matrices M M and MM ,

respectively. The transformation to the helicity basis takes the form

U
cosh X
ei sinh X
0
=
(6.13)
T1 ,
cosh X
ei sinh X
0 V
where X is a diagonal matrix in flavor space, obeying tanh(2X) = m/E. Like in the single
fermion case the CP-violating part of the dispersion relation is encoded in the diagonal
10 Notice that in Ref. [64] the phase has been defined with the opposite sign.
elements of T1 i h z T. We find

(1)
: pz = E 2 m2I I sinh2 XI cosh2 XI U iz U I

+ sinh2 XI V iz V I ,

(2)
: pz = E 2 m2I + I sinh2 XI cosh2 XI V iz V I

+ sinh2 XI U iz U I .
209
(6.14)
Here the subscript I on the RHS denotes the I th diagonal element of the corresponding
matrix, i.e., mI = mI I , and we dropped the factor h.
In addition to the contribution,
which we already encountered in the case of a single Dirac fermion, the dispersion relations
receive corrections due to the position dependent rotations in flavor space, U and V.
In Eqs. (6.13) and (6.14) we allowed for complex phases in the diagonal matrix MD .
Since U and V contain 2N 2 2 real parameters it would be sufficient to take real values
in MD in order to reproduce the 2N 2 real parameters of M. However, we have to omit the
N 1 rotations in U and V which belong to the Abelian subgroup of SU(N). These are
not related to CP-violation, but rather correspond to artificial redefinitions of the interaction
(chirality) states, as can be deduced from the case vanishing fermion mixing. Thus allowing
for complex values for the masses in MD we end up with a one by one correspondence
between the N 2 parameters in M and the ones contained in U, V and MD .
Having resolved the ambiguity in the definition of the helicity basis, it is possible to
determine the matrices U, V and MD numerically. The dispersion relations are then easily
obtained by evaluating the expressions (6.14). Here we will concentrate on the charged
winos and higgsinos in the (N)MSSM. Since they mix via the 2 2 chargino mass matrix
+

g2 (H20 )
M2
iW
, h
L = + iW
(6.15)
1
h +
g2 (H10 ) + S
2
analytic formulas can be obtained. The charged winos and higgsinos can be combined
+ , h + )T and R = (W
, h )T . We
to a Dirac spinor D = (L , R )T , where L = (W
2
1
parametrize the SU(2) matrices U and V by

cos b
ei sin b
cos a
ei sin a
,
V=
.
U=
(6.16)
cos a
cos b
ei sin a
ei sin b
According to the discussion above we dismissed phases multiplying the cos-terms. The
diagonal elements of U iz U which enter the dispersion relations (6.14) read

U iz U 1 = U iz U 2 = sin2 a.
(6.17)
We observe that only the derivative of the complex phase contributes, whereas the
derivative of a drops. Similar relations hold for V iz V. Inserting these expressions into
Eq. (6.14) we obtain the dispersion relations for left- and right-handed particles and
antiparticles

LI : pz = sgn(pz ) E 2 m2I I + sin2 b sinh2 XI + sin2 a cosh2 XI ,
210
L I :
RI :
I :
R

pz = sgn(pz ) E 2 m2I + I + sin2 b sinh2 XI sin2 a cosh2 XI ,

pz = sgn(pz ) E 2 m2I + I sin2 a sinh2 XI + sin2 b cosh2 XI ,

pz = sgn(pz ) E 2 m2I I sin2 a sinh2 XI sin2 b cosh2 XI .
(6.18)
+

In the symmetric phase L2 and R2 evolve to the left-handed higgsinos states h2 and
h
1 , respectively. The flavor transformations U and V are related to the parameters of the
chargino mass matrix (6.15):
sin2 a = 2|A|2/( + )
(6.19)
with

A = g2 M2 H20 + ( + S)H10 ,
2 2
= |M2 |2 | + S|2 + g22 H10 H20 ,
1/2

= 2 + 4|A|2
and = arg A. This gives
sin2 a = 2 Im(A A )/( + )
and there are similar relations for sin2 b, and sin2 b exchanging a and b, H10 and H20 ,
and . The mass eigenvalues read (in non-symmetric notation)
sin a i
cos a
g2 H20
e ,
cos b
cos b
sin a i
cos a
[MD ]22 = ( + S)
(6.20)
+ g2 H10
e .
cos b
cos b
For = 0 these expressions agree with the results of Ref. [71]. In this case one is left
with the MSSM, and the phases and only vary due to a change in the Higgs vev ratio
tan or because of transitional CP-violation in the bubble wall. The first contribution is
highly suppressed, since the variation of is at most 102 for realistic Higgs masses
[33,35], while transitional CP-violation most probably does not occur at all in the MSSM
[5]. On the other hand, the contribution to the chargino dispersion relations stemming from
the variation of the complex phases in MD requires only explicit CP-violating phases in
or M2 . Eq. (6.20) demonstrates that even though the phases in the two terms entering
[MD ]11,22 are position independent, their contribution to the resulting phase varies due
to the change in the (real) Higgs vevs [64]. According to the discussion in Section 5, the
corrections due to complex Higgs vevs induced by the explicitly CP-violating phases can
also be safely neglected in the MSSM [5].
In the NMSSM CP-violation enters the dispersion relations in several new ways. As
already discussed in Section 5, spontaneous CP-violation in the bubble wall occurs for
specific values of the SUSY parameters, which leads to a variation of , and 1,2 . If
this effect is present, it dominates the CP-violating part of the dispersion relations. Even in
[MD ]11 = M2
211
Fig. 8. CP-violating contributions to the chargino dispersion relation for the explicitly CP-violating
parameter set of Fig. 6, (a) as a function of z for p = Tc and (b) as a function of p for z = 0. The
dashed-dotted, dashed and solid lines represent the helicity and flavor contributions (6.18) and the
result using the kinetic momentum (6.21), respectively.
the absence of transitional CP-violation and with constant tan there are contributions to
and : The complex phases induced in the vevs by the explicitly CP-violating phases
may considerably change in the bubble wall according to Section 5. Also the phase of the
effective -term, + S, is position dependent because of the variation of the singlet vev.
However, this effect is suppressed when the coupling is small.
Again we calculate the (inverse) group velocity, now from (6.18). E-independent
terms drop out, i.e., cosh2 XI can be substituted by sinh2 XI . The kinetic momenta
2
beyond the zeroth order (6.7) then contain (order h)
correction terms (I + sin b
sin2 a)m2/2E(E 2 m2 )1/2 . The dispersion relation can be inverted, and to leading
order in the derivatives the CP-violating part of the dispersion relation for the eigenstate
L2 corresponding to h +
2 in the symmetric phase is

+E = sign(pz ) 2 + sin2 b sin2 a
m22
,
2
2(pkin + m22 )
(6.21)
where m22 = |[MD ]22 |2 from Eq. (6.20). For pkin m this is 2|pz |/p times the result one
would obtain with canonical momentum after the substitution of cosh2 by sinh2 in (6.18)
(see Fig. 8b). Note that +E is now totally symmetric under the exchange of H1 and H2 .
This destroys the most prominent source term H1 H2 H1 H2 of older work.
In Fig. 8 we compare the CP-violating contributions to the chargino dispersion relations
stemming from the local helicity and flavor transformations (6.18) with the result from
using kinetic variables (6.21). We assume that p = 0, and approximate the bubble wall
profile by a kink-ansatz with a common wall thickness for all fields, i.e., we take the
straight connection between the symmetric and the broken minimum in field space. In
this approximation tan is automatically constant along the wall. We study the case of
explicit CP-violation induced by = 0.1, which has already been considered in Fig. 6 in
Section 5. It is characterized by Lw = 5/Tc and Tc = 101 GeV. Fig. 8a shows the position
dependence of +E for p = pz = Tc . In Fig. 8b we present +E as a function of momentum
212
for z = 0. One observes that +E is dominated by the helicity contribution proportional

to 2 (dashed-dotted line), while the flavor contribution (dashed line) only gives a small
correction. The result from the kinetic approach is larger by a factor between 1 and 2. For
the transitionally CP-violating parameter set of Fig. 7 the results are similar. The amplitude
of +E, however, is by a factor of about 7 larger, due to the larger amount of CP-violation
and the thinner bubble wall (Lw = 3/T ). In any case, +E provides only a small correction
to dispersion relation.
We close this paragraph by briefly considering the case of Majorana fermions. Majorana
spinors have the special property of being invariant under charge conjugation, i.e., particles
and antiparticles are contained in the same four-component spinor. The mass matrix of
Majorana particles is symmetric, and the entries are complex numbers, in general. In
the NMSSM, the neutralinos are Majorana particles with a 5 5 mass matrix. From the
symmetry property of the mass matrix we conclude MM = MM = (M M) . The flavor
transformations therefore obey V = U . Inserting this result into the general dispersions
relations (6.14) (or into its two-dimensional version (6.18)) we find that left-handed and
right-handed states acquire exactly the opposite CP-violating contribution, in contrast to
the general Dirac case. This is not surprising, since in case of Majorana fermions the two
helicity states describe particles and antiparticles, respectively.
6.2.2. The bosonic case
In supersymmetric models the scalar superpartners of the top quarks may give important
contributions to the CP-violating source that fuels baryon production. The stops efficiently
interact with the bubble wall via the large top Yukawa coupling. They contain many degrees
of freedom, and due to renormalization group flow the right-handed stop is probably lighter
than any other squark. In the following we consider the interaction between the squarks and
the bubble wall in the semi-classical limit. Our approximation is valid if the thickness of
the bubble wall is much larger than the typical wavelength of particles in the plasma which
is of order 1/T . As in the fermionic case, we derive dispersion relations in which the
CP-violating part is generated by varying phases in the scalar mass matrix.
Consider N complex scalar fields A = (A1 , . . . , AN )T with mass term A M2 A. The
entries of the hermitian mass matrix M2 may be complex. Due to the interaction with the
wall, M2 depends on the position variable z. We represent M2 by a unitary matrix U and
2 , . . . , M2 )
a real diagonal matrix M2D = diag(MD1
DN
M2 = UM2D U .
(6.22)
Like in the fermionic case we omit the N 1 rotations in U which belong to the Abelian
subgroup of SU(N). These transformations artificially redefine the interaction eigenstates.
Their phases are not related to CP-violation, as can be deduced from the case of vanishing
mixing. We then have a one to one correspondence between the N 2 real parameters in M2 ,
and in U and M2D .
From the KleinGordon equation for the scalar fields (h 2 + M2 )A = 0 we obtain
= U A
the equation of motion for the local mass eigenstates A

2 2

h z + 2U z Uz + U z2 U + M2D E 2 A(z)
(6.23)
= 0.
213
To derive this equation we implemented energy conservation in the wall frame by the
iEt and boosted to the Lorentz frame where there is no momentum
z) = A(z)e
ansatz A(t,
parallel to the wall, i.e., p = 0.
U
We solve Eq. (6.23) by the WKB method. To order O(h 0 ) we can neglect
z
z the U
2
(I
)

(z) = eI exp(i pz (z ) dz ),
and U z U contributions and find the wave functions A
where eI is the I th unit vector in flavor space. Furthermore, we obtain the dispersion
2 . As we already found in the fermionic case, CP-violation vanishes
relation pz2 = E 2 MDI
in the classical limit.
We now include the order O(h 1 ) corrections. We still can neglect the U z2 U term,
while the U z U contribution becomes relevant now. Its off-diagonal entries reintroduce
a coupling between the N components of Eq. (6.23). In the dispersion relation only the
diagonal part of U z U enters. We obtain

(I ) : E 2 = M 2 + p2 + 2pz U iz U ,
A
(6.24)
z
DI
I
where [U iz U ]I denotes the I th diagonal entry of the corresponding matrix. To transform
the dispersion relations to a general Lorentz frame with non-vanishing p one has to
2.
replace E 2 by E 2 p
Let us evaluate the general expression (6.24) for the case N = 2. Using the representation of U from Eq. (6.16), we obtain
(1,2): E 2 = M 2 + pz2 2pz sin2 a.
A
D1,2
(6.25)
Again, CP-violation arises due to the varying phase in the transformation to the local
mass eigenstates. A variation in a gives no contribution to the dispersion relations. For the
(1,2) , the CP-violating part in Eq. (6.25) enters with the opposite sign. If
antiparticles, A
we apply these expressions to the top squarks of the (N)MSSM with mass matrix

m2LR
m2LL
t
,
L = t , tc
(tc )
(m2LR ) m2RR
2

g12 0 2 0 2
g2
2
2
2 0 2
mLL = mQ3 + yt H2 +
H1 H2 ,
4
12
2 g 2 2 2
m2RR = m2U3 + yt2 H20 + 1 H10 H20 ,
3

m2LR = yt ( + S)H10 + yt At H20 ,
(6.26)
the parameters of U are given similarly to (6.19) by sin2 a = 2|A|2/( + ), = arg A
now with A = yt (( + S)H10 + At H20 ). Solving (6.25) for p, one obtains p(E) = (E 2
2 )1/2 + sin2 a to first order in . The group velocity (p/E)1 is independent
MD
1,2
of , thus stops do not contribute to CP-violation in this order in the kinetic approach.
6.3. Diffusion equations
6.3.1. The fluid approximation
In this section we study the coupled differential equations that describe particle
interactions and transport during the phase transition. We treat the plasma as consisting
214
of quasi-classical particles with definite canonical position and momentum. This is an

approximation to quantum Boltzmann equations which have to be discussed in principle.
This picture is justified for thick walls (p 1/Lw ) if it predicts a sizable effect, not
dominated by non-leading terms in the derivative expansion [65]. The dynamics of the
particles is then governed by the dispersion relations E(
x , p)
derived in the previous
section.
The information about the particle distributions is encoded in the phase space densities
x , p,
t). Their temporal evolution follows from the Boltzmann equation
fi (

dt fi = t + x x + p p fi = Ci [f ].
(6.27)
The time derivatives of position and momentum obey the Hamilton equations x =
p E(
x , p)
and p = x E(
x , p).
The Boltzmann equation can in principle be solved
numerically. However, to make it analytically tractable we use the fluid-type truncation
[63]
fi (
x , p,
t) =
1
e(Ei vi pz i ) 1
(6.28)
for the phase space densities of fermions (+) and bosons () in the rest frame of the
plasma. Here vi and i denote the velocity perturbations and chemical
potentials for
each fluid, respectively. We also split Ei into a dominant part E0i = p2 + m2i and a
perturbation +Ei z which is related to CP-violation. The chemical potentials are
the central quantities that finally will determine the baryon asymmetry. The velocity
perturbation on the other hand, is only introduced to allow the particles to move in response
to the force, giving rise to chemical potential perturbations. The fluid truncation is valid as
long as perturbations beyond the ansatz (6.28) are attenuated faster than chemical potential
perturbations. As discussed in Ref. [63] this requires vw < Lw /(3D), where the diffusion
constant D will be introduced below.
We are looking for a stationary solution of the Boltzmann equation, because at
late times the wall moves with constant velocity vw . This means that any explicit time
dependence enters in the combination z z vw t. Inserting the fluid ansatz into the
Boltzmann equation (6.27) we obtain to linear order in the perturbations +Ei , i and
vi
f

pz
(m2i )

v pz + i +
vi = Ci [f ],
vw +Ei vi pz i
E0i i
2E0i
(6.29)
where f = df /dE0 = eE0 /(eE0 1)2 denotes the derivative of the unperturbed
FermiDirac or BoseEinstein distribution. The remaining primes in Eq. (6.29) denote
z . More precisely, Eq. (6.29) is the difference of the corresponding equations
for particles and antiparticles. The parameters +Ei , i and vi therefore represent the
differences of these quantities for particles and antiparticles. For that reason the term
, which provides the main contribution to friction in the calculation of the wall
f vw E0i
velocity [67], cancels out. In order to obtain differential equations for the perturbations we
215
average over momentum, weighting Eq. (6.29) by 1 and pz , respectively, and receive
2
p
z vi + i vw i = Ci
,
(6.30)
E0i
2

pz

i vw pz +E0i
= pz Ci
.
vw pz2 vi
(6.31)
E0i
The average is defined according to
3
3
d p f ()
d p f ()

3
.
i 3

d p f+ (m = 0)
d p f
(6.32)
The statistical factor is 1 for massless fermions, 2 for massless bosons and exponentially
small for particles much heavier than T . In the derivation of Eqs. (6.30) and (6.31) one uses
that the energy perturbations +Ei are odd in pz , according to the results of the previous
section. Furthermore, we neglected the contribution of the last term in the curly brackets
of Eq. (6.29) which turns out to be numerically less important than the pz2 /E0i
vi term
taken into account in Eq. (6.30).
To linear order in the perturbations the collision terms on the RHS of the Boltzmann
equation take the form [64]

pd
j ,
pz C i
= vi p z2
pe ,
Ci
=
(6.33)
p
where p denotes the rate of the process p. In the sum over chemical potentials, incoming
particles enter with positive sign, outgoing particles with negative sign. We distinguish
between elastic interactions with rates e and decays (inelastic interactions) with rates d .
Since elastic scattering conserves the number of particles these processes do not contribute
to Ci
. In the hot electroweak plasma the elastic processes dominated by gauge boson
exchange reactions are much more efficient than the particle decays coming from Yukawa
interactions, sphalerons, etc. It is therefore justified to neglect the contribution of inelastic
processes to pz C i
.

In the collision integral a delta function ( pi ) usually represents the conservation of
energy and momentum in the interactions. However, in the plasma frame this is no longer
true, because of the moving bubble wall. The Boltzmann equation is formulated in terms
of mass eigenstates which change in space and time due to the varying Higgs (and singlet)
vevs. As a result, additional CP-violating contributions to the collision term arise, which
are related to what has been dubbed spontaneous baryogenesis [70]. The final form of
the collision term is therefore given by [64]

d
p
j + +Esp,p ,
Ci
=
(6.34)
p
where +Esp,p denotes the bubble wall induced deviation from energy conservation in the
corresponding process.
Finally, we reduce the two coupled transport equations (6.30) and (6.31) to a single one
by differentiating (6.31) and eliminating vi in favor of i . We approximate the thermal
216
averages according to pz2 /E0

i pz2 /E0
0 , pz2
i pz2
0 , where the subscript 0
denotes averaging with the massless, unperturbed FermiDirac distribution. Defining the
diffusion constant Di = i pz2 /E0
20 /(pz2 ie ) we find the diffusion equation [64]

i (Di i + vw i ) +
pd
j = Si ,
p

Di vw
Si = 2
pz +Ei

pd +Esp,p
.
pz /E0
0
p
(6.35)
In order to obtain Eqs. (6.35) we neglected derivatives of the CP-conserving thermal

averages and rates, and left aside ratios of inelastic to elastic scatterings. We also ignored
terms beyond leading order in the wall velocity, an approximation that certainly
breaks
down if the wall velocity approaches the speed of sound in the plasma, vs = 1/ 3 0.58.
If the wall moves faster than vs , perturbations cannot propagate in the region in front of
the wall any more. Notice that the first contribution to the CP-violating source term Si in
(6.35), which is due to the semi-classical force, is proportional to the diffusion constant.
This is simply because particles must move in order to build up perturbations. The second,
spontaneous source term is independent of transport properties of the corresponding
particles and therefore dominates in the limit of inefficient transport. In Si the thermal
averages over the CP-violating energy perturbations +E are performed using the massive
distribution functions in order to account for Boltzmann suppression of heavy particles.
6.3.2. Diffusion equations for supersymmetric models
In the previous paragraph we derived a complicated network of diffusion equations
(6.35) that couples all particle species in the hot plasma. In principle, after specification
of decay rates, diffusion constants and CP-violating sources, it is possible to solve
the transport equations numerically. However, analytic progress can be made by using
conservation laws and neglecting interactions that are slow compared to the relevant time
scale. An interaction with rate can be neglected if the typical interaction time is large
compared to the average time a particle spends diffusing in front of the wall before being
caught by the wall, which is equivalent to [63]
D
1 .
2
vw
(6.36)
The electroweak (weak) sphaleron interaction with rate ws is slow in precisely the
sense of (6.36) (unless the wall velocity is particularly small, i.e., vw 0.01). In the
following we will therefore assume baryon and lepton number conservation and include
the weak sphalerons only at the end of the calculation. The neglect of the weak sphalerons
allows us to completely forget about leptons in our transport equations and compute only
the quark and Higgs densities.
The processes we do take into account are the supergauge interactions, the strong
sphaleron interactions, and those described by the Lagrangian
Lint = yt t c q3 H2 + yt tc q3 h 2 + yt t c q3 h 2 yt tc q3 H1 + yt At tc q3 H2 + h.c. (6.37)
217
Via terms of type s h 1 H2 , the singlino, s , is coupled to the quarkHiggs system. In the case
of transitional CP-violation triggered in the singlet sector (see Fig. 7), the singlino receives
the largest CP-violating source term of all particles, even though its impact on baryogenesis
is suppressed by the small coupling . For the sake of simplicity we ignore this interesting
contribution in the following. We assume the supergauge interactions to be in equilibrium.
The chemical potential of any particle is then equal to that of its superpartner, with
exception of the singlet field. It is convenient to define the chemical potentials U =
(uc + u c )/2, Q1 = (u + d + u + d )/4, H1 = (H 0 + H + h 0 + h )/4,
etc. In this notation the interaction terms take the form 11
(y + yA )(H2 + Q3 + T ),
y (H1 Q3 T ),
ss (2Q3 + 2Q2 + 2Q1 + T + B + C + S + U + D ),

hf (H1 + H2 ),
m (Q3 + T ),
H1 H1 ,
H2 H2 .
(6.38)
The rates in the first line are related to the interactions (6.37). ss denotes the strong
sphaleron rate. hf is due to higgsino helicity flips induced by the h 1 h 2 term. H1,2
and m correspond to Higgs and axial top number violating processes, present only in the
phase boundary and the broken phase.
Because of the small Yukawa couplings of the first and second family quarks, these
particles are in very good approximation only produced by strong sphalerons. Hence their
number densities are algebraically constrained. If the system is near thermal equilibrium,
number densities and chemical potentials are related by
1
ni = ki i T 2 ,
(6.39)
6
where ki is the appropriate sum over statistical factors introduced in (6.32), e.g., kQ1 =
Nc (u + d + u + d ), kU = Nc (uc + u c ), kH1 = (H 0 + H + h 0 + h ), etc. Nc = 3
1
1
1
1
denotes the number of colors. In the massless limit used in Ref. [64] one obtains kQ1,2,3 =
18, kU = kD = = kT = 9, kH1,2 = 6. Using baryon number conservation the strong
sphaleron rate reads

kT
kQ
ss (2Q3 + + D ) = ss 2 + 9 3 Q3 + 1 9
(6.40)
T .
kB
kB
To arrive at this expression we assumed that all the squark partners of the light quarks are
degenerate in mass. Assuming equilibrium for the strong sphalerons we obtain
2kB + 9kQ3
(6.41)
Q 3 .
9kT kB
The validity of this assumption will be discussed below.
We are now able to write down the reduced set of diffusion equations for the relevant
particle species Q3 , H1 and H2
T =
ADq Q3 + (y + yA )[H2 + BQ3 ] y [H1 BQ3 ] + Bm Q3 = SQ3 ,

11 In contrast to Refs. [58,64] we count all left-handed particles and the corresponding superpartners with
positive signs.
218
kH1 Dh H1 + y [H1 BQ3 ] + hf [H1 + H2 ] + H1 H1 = SH1 ,

kH2 Dh H2 + (y + yA )[H2 + BQ3 ] + hf [H1 + H2 ] + H2 H2 = SH2 ,
(6.42)
where
9kT kQ3 + 9kB kQ3 + 4kB kT
,
9kT kB
kB + 9kT + 9kQ3
B=
9kT kB
A=
(6.43)
and Di Di d 2 /d z 2 + vw d/d z . These equations result from summing the diffusion

equations of particles which belong to the same color and SU(2) multiplets. We have taken
a common diffusion constant, Dq , for the (s)quarks, as well as one for the two Higgs
doublets, Dh . Notice that the effects of hypercharge screening have been neglected, which
can be shown to affect the created baryon asymmetry at most by a factor of order one [72].
We keep the rates related to the top Yukawa interactions finite. If these interaction
are in equilibrium, the resulting diffusion equations are sourced only by the combination
SH1 SH2 , because of the constraint H1 + H2 = 0. As a result the dominant contribution
+
to the chargino source terms cancels, because the corresponding terms for h
1 and h2 are
exactly of the same size. This would not be true for the , contribution if we used the
dispersion relations for canonical momenta.
In the MSSM the full diffusion equations (6.42) have already been studied in Ref. [71]
and later in Ref. [73]. Applying the closed time path formalism to calculate the source
terms, the diffusion equations (6.42) were analyzed recently in Ref. [74].
6.4. The baryon asymmetry
6.4.1. Solution of the diffusion equations
In the previous section we derived differential equations for the chemical potentials
of the various particle species contained in the hot electroweak plasma. Baryon number
violation has been neglected throughout this calculation. However, what we set out to
compute was the total baryon asymmetry created during the phase transition. So before
solving the network of diffusion equations, we turn to baryon number generation by weak
sphaleron processes, which is fueled by the chemical potential of left-handed quarks,
BL Q1 + Q2 + Q3 . Using baryon number conservation and Eq. (6.41) we find

kQ3 + 2kT 2kB 2kB
Q3 CQ3 .
+
BL = 1
(6.44)
9kT kB kQ1
kQ2
The evolution of the baryon number density nB is governed by

Dq nB + 3(z)ws T 2 BL anB = 0,
(6.45)
where we have assumed identical diffusion constants for all quarks and squarks, and
neglected contributions of leptons. The position dependence of the weak sphaleron rate
is modeled by a step function (z): anomalous baryon number violation is unsuppressed
219
in the symmetric phase (z > 0) and suddenly switched off in the broken phase (z < 0). The
second term in Eq. (6.45) describes damping of the baryon asymmetry by weak sphalerons
in the symmetric phase. The parameter a depends on the degrees of freedom present in the
hot plasma. Taking only the right-handed stop to be light gives a = 48/7 [71].
From Eq. (6.45) one can easily obtain the baryon to entropy ratio in the broken phase
135ws
nB
=
B
s
2 2 g vw T
d z BL (z)e z ,
(6.46)
where we have taken the entropy density s = (2 2 g /45)T 3 and = 3aws /(2vw ) [71].
g 126 is the effective number of degrees of freedom at the phase transition temperature.
Eq. (6.46) shows that the integral over the left-handed quark number, nL BL in the
symmetric phase determines the final baryon asymmetry.
We now return to Eqs. (6.42) in order to compute BL . These linear second order
differential equations can be solved by finding the appropriate Greens function. We keep
the discussion general and consider the following set of N coupled diffusion equations
1
k11 D11 + 11 k1N D1N + 1N
S1
..
..
.. ..
..
(6.47)
. = . ,
.
.
.
SN
kN1 DN1 + N1 kNN DNN + NN
N
where Dab = Dab d 2 /d z 2 + vw d/d z . The corresponding boundary conditions read
a (|z| ) = 0. The matrix valued Greens function Gab is defined by
N

(kac Dac + ac )Gcb (z) = ab (z).
c=1
In the transport equations (6.42) also position dependent rates are present, e.g., m . They
typically vanish in the symmetric phase and become maximal in the broken phase. In order
to keep the problem analytically tractable we simply model the position dependence of
these rates by step functions, i.e., ab (z) = +ab (z) + ab (z).
The general structure of the Greens function then reads
Gab (z) = (z)
N

i=1
ci+ab ei+ z + (z)
N

ciab ei z .
(6.48)
i=1
The constants i can be computed from det[kab (Dab 2 + vw ) + ab ] = 0. Of

course, this procedure supplies us with 4N solutions. However, 2N of them (i+ < 0 for
z > 0 and i > 0 for z > 0) correspond to exponentially growing solutions of Eq. (6.47)
and have to be discarded. The coefficients ciab can then be computed from the 2N 3
dimensional set of linear equations
N

kab Dab 2i+ + kab vw i+ + +ab ci+bc = 0,
b=1
(6.49)
220

N

kab Dab 2i + kab vw i + ab cibc = 0,
(6.50)
b=1
N

[ci+ab ciab ] = 0,
(6.51)
i=1
N

kab Dab [i+ ci+bc i cibc ] = ac ,
(6.52)
i,b=1
where ab denote the rates for positive and negative z . Eqs. (6.49) and (6.50) result from
solving the homogeneous version of Eq. (6.47) in the range of positive and negative z ,
respectively. Only 2N 2 (N 1) of them are independent. The continuity of the Greens
function at z = 0 is guaranteed by Eq. (6.51). Finally, Eq. (6.52) is obtained by integrating
the definition equation for the Greens function on an infinitesimally small interval around
z = 0. Diffusion constants Dab and statistical factors kab differing in broken and symmetric
phase can be treated along the same lines.
Once the Greens function is known one can easily compute the chemical potentials.
Applying the general formulas to the diffusion equations (6.42) we identify 1 Q3 ,
2 H1 , 3 H2 , S1 SQ3 , S2 SH1 and S3 SH2 . Using Eqs. (6.46) and (6.44) we
obtain for the baryon to entropy ratio
135ws
B =
C
2 2 g vw T

d z e
d z G1a (z z )Sa (z ).
(6.53)
Since the Greens function consists only of exponentials, the z -integration can be
performed analytically. The evaluation of this expressions is performed in the next
paragraph. Also the various approximations we used in its derivation will be discussed.
6.4.2. Numerical evaluation and discussion
Before starting to calculate a numerical value for the emerging baryon asymmetry we
discuss the validity of the assumptions and approximations made in the derivation of
Eq. (6.53). This requires the specification of the various parameters that enter the diffusion
equations. For the diffusion constants of quark and Higgs fields we take [74]
Dq =
6
,
T
Dh =
110
.
T
(6.54)
We use the rates [74]

y + yA = 0.015T ,
hf = 0.016T ,
y = 0,
m = 0.05T (z),
H1 = H2 = 0.0036T (z).
(6.55)
y is strongly suppressed because it involves heavy left-handed stop states (6.37). m ,

H1 and H2 are present only broken phase. We model their position dependence by a step
221
function. The weak and strong sphaleron rates are given by [76]
4
ws 6w
T 2.2 105 T ,
ss 1500ws 0.033T .
(6.56)
It has been shown recently that parametrically ws =

[77], but lattice
measurements of the rate are consistent with C 1/w . The thickness of the bubble
wall varies considerably in the NMSSM, 1/T Lw 20/T , i.e., the wall may become
much thinner than in case of the MSSM, where 20/T Lw 30/T has been found [35].
Calculations of the wall velocity in the SM lead to 0.36 < vw < 0.44 [67]. Gauge fields in
the hot plasma diminish this result [66]. In supersymmetric models there arise additional
friction terms from the SUSY particles, in the first place from a light top squark. In the
MSSM, this can bring down the wall velocity to vw 102 [68]. In the following we treat
vw as a free parameter and examine its impact on the emerging baryon asymmetry.
Let us now summarize the approximations leading to Eq. (6.53) for the baryon to entropy
ratio:
5
C ln(1/g 2 )w
Assumption 1. Lw > 1/T so that most particles in the plasma are indeed accurately
described by the WKB approximation in their interaction with the bubble wall. For
typical wall thicknesses in the NMSSM, 5/T Lw 15/T , this assumption is very well
justified, although for the very thinnest walls Lw 1/T the WKB approximation becomes
questionable.
Assumption 2. vw < Lw /(3D), the thermalization condition that guarantees the applicability of the fluid ansatz. In case of (s)quarks this condition is satisfied for v < 0.4 even
for rather thin walls with Lw 5/T . Higgs particles thermalize much slower, as can be
deduced from their large diffusion constant. Even for the largest wall thicknesses, Lw
20/T , rather small velocities, vw < 0.1, are required.
Assumption 3. vw < 1/ 3. We work to linear order in the wall velocity, which is only
justified if the wall moves slower than the speed of sound in the plasma. Otherwise,
diffusion in the region in front of the wall, giving
rise to non-local baryogenesis, is
no longer possible. However, in case of vw > 1/ 3 the fluid approximation would break
down anyway, i.e., from assumption 3 no new constraints result.
2 /D, so that the back-reaction of the baryon number violating
Assumption 4. ws < vw
processes can be neglected in the diffusion equations (6.42). According to Eqs. (6.54) and
(6.56) this requires vw > 0.01 (quarks) and vw > 0.04 (Higgs particles).
2 /D in order to put the strong sphaleron interaction to equilibrium.
Assumption 5. ss > vw
In this case the strongest constraint, vw < 0.45, result from the quarks, which is easily
satisfied.
The calculation of the final baryon asymmetry still requires the specification of the
source terms (6.35) which enter Eq. (6.53). We concentrate on the source terms induced
222
by the semi-classical force and neglect those related to spontaneous baryogenesis. We

include the source terms of the top quark (6.10) and the charged higgsinos (6.18). For
sake of simplicity we neglected the neutral higgsinos, which should affect the final baryon
asymmetry at most by a factor of order one. Also the singlino is disregarded.
In order to evaluate the source terms (6.35) the bubble wall profile is required. We
approximate our numerical solutions by a kink-ansatz with a common wall thickness for
all fields present in the bubble. However, we allow for a variation in tan in the bubble
wall by taking |H20 (z)| = |H10(z)|(tan 0 + const |H10 (z)|2 ), in the spirit of Fig. 5a. The
relevant source terms are obtained from combining Eqs. (6.10), (6.18) and (6.35)

Nc Dq vw
A
t + B
m2t t ,
2
pz /E
0

Dh vw
= 2
A
h h sin2 (ah ) + h sin2 (bh )
pz /E
0

+ B
m2h h h sin2 (ah ) + h sin2 (bh ) ,
St =
Sh 1
Sh 2 = Sh 1
with

A
=
|pz ||m|2
2E 2
(6.57)
(6.58)
(6.59)

,
+
B
=
|pz |(E 2 |m|2 )

2E 4

.
(6.60)
2 =p
Here Nc = 3 is the number of colors and E 2 = p2 + m2i , Ez2 = pz2 + m2i and p
2 pz2 .
2
With exception of pz /E
0 all thermal averages in Eqs. (6.57)(6.59) are performed with
massive distribution functions for fermions (6.32), which ensures the decoupling of heavy
particles. Taking into account only the source terms of Eqs. (6.57)(6.58), we have SQ3 =
St , SH1 = Sh 1 and SH2 = Sh 2 .
In the approach with canonical momenta (which we had in the first version of this
paper), there was a source St for the stop and Sh 1 = Sh 2 . The latter leads to an important
contribution to the (H1 H2 )-combination, the well-known term proportional (tan ) and
to an additional term for the singlet field. These are absent now. The brackets A
and B
in the approach with canonical momenta are

2
can
can
|pz |Ez pz2
p |pz | + pz2 Ez
=
=
,
B
.
A
(6.61)
2E
4E 3 ez
+
+
For p m they differ by a factor 12 |p|/|pz | from the ones for kinetic momenta (6.60).
The generated baryon asymmetry depends on the squark spectrum because of the
potential suppression due to strong sphalerons. In the following we consider four different
squark spectra which are listed in Table 2. Case A corresponds to the massless limit, where
strong sphaleron suppression is most efficient. In case B all squarks are assumed to be
heavy, with exception of the right-handed stop, which is taken massless. The spectra C and
D are obtained from the universal pattern of SUSY breaking discussed in Section 2. The
first and second generation squarks are almost degenerate in mass with b2 , thus 2kB =
kQ1 = kQ2 = 2kU = . Case C corresponds to the example of explicit CP-violation
223
Table 2
Squark spectra and corresponding statistical factors used in the discussion of the baryon asymmetry.
For the sets C and D we used = 0.05, k = 0.4, m0 = 200 GeV and tan = 5. Furthermore, we
took x = 100 (150) GeV in case C (D). The sets A and B do not follow the universal pattern of
SUSY breaking considered in Section 2. (Units in GeV)
A
B
C
D
M0
A0
Mt1
Mt2
Mb
Mu
kQ 3
kT
kB
kQ 1
125
100
100
100
418
371
0
0
238
201
334
217
375
325
18
6
7.26
8.43
9
9
4.56
4.98
9
3
3.60
3.87
18
6
7.20
7.74
0
0.385
0.125
0.102
discussed in context of Fig. 6, while case D is the squark spectrum of the transitionally
CP-violating example of Fig. 7.
According to Eq. (6.53) the produced baryon asymmetry is proportional to the
parameter C, which is related to the chemical potential of left-handed quarks (6.44) and
also given in Table 2. It turns out that the results for different spectra can indeed be obtained
by rescaling with the relevant C parameters, i.e., the indirect impact of the statistical factors
in the diffusion equations (6.42) is small. For the massless case A, and more general for
the case of degenerate squarks, C vanishes. Baryon production is completely suppressed
by rapid strong sphalerons transitions, which is a well known result [75]. The cancellation
disappears for non-degenerate squarks. Even for the realistic spectra C and D, resulting
from universal SUSY breaking, there is only a mild suppression by a factor of 34 relative
to the idealized spectrum B (usually assumed in context of the MSSM).
To begin with let us consider the source term of the top quarks (6.57), which in the
first place depends on the variation of the phase of the top quark mass along the wall,
St t , and the wall thickness Lw . From Eqs. (5.1) and (5.9) we obtain t = 2 =
cos2 () , where 2 denotes the phase of H20 and is the (gauge invariant) sum of
the phases of the two Higgs fields, that entered the bubble equations in Section 5. As a
consequence, St is rather sensitive to the Higgs vev ratio. Baryon production from the top
quark source is particularly efficient for small values of tan(T ). In Fig. 9 we present the
baryon asymmetry induced by St as a function of the wall thickness and velocity, where
we used vc /Tc = 1.60, Tc = 101 GeV and yt = 1.015 (corresponding to tan = 5).
Our results 12 would be slightly enhanced by larger values of vc /Tc . We measure the
baryon asymmetry in units of 2 1011 , which is the lower observational bound [69]
=
(6.62)
.
2 1011
In Fig. 9 we display the dependence of the baryon asymmetry on the wall velocity for
different values of the wall thickness Lw = 20/T , 10/T , 5/T , 3/T . We used the squark
12 In numerical evaluations we still used the brackets Acan
, B can
of Eq. (6.61) instead of A
, B
of
Eqs. (6.60) in order not to be forced to completely repeat the whole analysis. This introduces a deviation of
order (1), in the case |p| m a factor 1/2 instead of |pz |/|p| in the brackets A
and B
. This is well within the
accuracy of the present investigation.
224
Fig. 9. Top quark contribution to the baryon asymmetry for the spectrum D in units of 2 1011
normalized by t as a function of the wall velocity for different values of the wall thickness
Lw = 20/T , 10/T , 5/T , 3/T (from below).
spectrum D in the evaluation. Results for other spectra can be easily obtained by rescaling
with the relevant C parameters of Table 2. For Lw 10/T baryon production is most
efficient for vw 0.020.03. Interesting enough, this is precisely the range of wall
velocities found in recent calculations in the MSSM [66,68]. becomes small for large
values of vw because transport in front of the wall gets more and more inefficient. For
small wall velocities deviation from equilibrium becomes too small to generate a relevant
baryon asymmetry. We observe an approximate 1/L2w dependence in the generated baryon
number. This behavior is expected from the explicit expression for the source term (6.57),
which contains three derivatives with respect to z . Since the calculation of (6.53) requires
one (numerical) integration of the source over z , approximating z 1/Lw we obtain
1/L2w . As a result baryon production from the top quark source is most efficient in the case
of slow (vw 102 ) and thin walls, and for non-degenerate squark spectra. It turns out that
the Higgs source term behaves in a similar way. In Fig. 9 we assumed = 103 in the
bubble wall which is a typical value according to Section 4. However, even large values,
= 102 , change the result only by a few percent. From Fig. 9 we approximately obtain
for the maximal value of the baryon asymmetry

2
5
47
t .
max
(6.63)
Lw T
However, for medium and large Higgs vev ratios even small values of t O(102 ) are
difficult to achieve, since t = cos2 () . In the case of tan = 5, for instance, we obtain
t = /26, which leads to a significant suppression. In the example of transitional CP
violation (Fig. 7) we have tan = 5, Lw = 3 and 1/20 leading to max
1/4. In
case of explicit CP-violation, the top quark contribution to the baryon asymmetry is much
smaller even, and can be safely neglected.
225
Fig. 10. (a) The chargino contribution to the baryon asymmetry in units of 2 1011 as a function
of the wall velocity for different values of the wall thickness Lw = 20/T , 10/T , 5/T , 3/T (from
below). We use the squark spectrum C and the example of explicit CP-violation considered in the
context of Fig. 6. (b) The same quantity for the transitionally CP-violating bubble wall of Fig. 7 and
the squark spectrum D.
We now come to the charged higgsino contribution to the baryon asymmetry fueled
by the source terms (6.58) and (6.59) for the examples of explicit and transitional CPviolation already discussed in the context of Figs. 6 and 7. Again, we present our
results12 as a function of the wall velocity for different values of the wall thickness Lw =
20/T , 10/T , 5/T , 3/T . Concerning the dependence of the generated baryon asymmetry
on the wall velocity and the wall thickness we find a similar behavior as in the case of the
top quark source: becomes large for thin walls and vw 102 .
The baryon asymmetry generated from the chargino dispersion relation is shown in
Fig. 10. If the top Yukawa interactions were in equilibrium, as was assumed in elder work
on the subject, this contribution is completely erased, because of the equal source terms for
both higgsinos (6.58), (6.59). Even with finite top Yukawa rates no baryon asymmetry is
generated from this source, if both Higgs fields have equal rates in the diffusion equations
(6.42), i.e., if y + yA = y and H1 = H2 . In our scenario an asymmetry in the
top Yukawa rates is inevitably induced by the heavy left-handed stop, leading to a strong
suppression of y (6.55). In the evaluations of Fig. 10 we used = 103 . If we increase
the change in the Higgs vev ratio by a factor of 10, the results only change by about 15
percent.
From Fig. 10a for the case of explicit CP-violation we read of
max
0.8
5
Lw T
2
sin( ).
(6.64)
Thus CP-violating phases of order 101 are required to account for the observed baryon
asymmetry in this scenario. Phases of this size are compatible with the EDM experiments
only if the first and second generation squarks are heavy or in the case of accidental
cancellations. Our findings for baryogenesis from explicit CP-violation in the NMSSM
226
Fig. 11. Chargino contribution to the baryon asymmetry in units of 2 1011 as a function of the
gaugino mass parameter M2 (a) and as a function of the soft scalar mass parameter mU3 . In both
cases we otherwise used the parameter set of explicit CP-violation (Fig. 6), = 0.1, Lw = 5/T ,
vw = 0.03 and = 103 .
are very similar to the results in the MSSM [73]. Still, baryogenesis in the NMSSM is
slightly more effective because of the thinner bubble walls.
The scenario of transitional CP-violation is a very attractive alternative, being not in
conflict with the EDM experiments. As shown in Fig. 10a, it can generate the baryon
asymmetry for a considerable range of wall velocities, especially for Lw = 3/T . It even
allows for a certain baryon over-production to compensate for a partial washout of the
generated baryon asymmetry due to the existence of bubbles with opposite CP-properties.
The introduction of small explicit phases lifts the energetic degeneracy of the two sort of
bubbles, leading to a more rapid nucleation of bubbles of the true vacuum and a net baryon
production.
Up to now we centered the discussion of baryon production around two particular sets
of NMSSM parameters. However, the results may considerably change if different regions
in the parameter space are considered. Of course, baryon production becomes inefficient if
the particles which supply the CP-violating source terms, in the first place the higgsinos,
are heavy and decouple from the plasma. The thermal averages, which enter the sources
(6.57)(6.59), already decrease by a factor of 10 if m/T is raised from one to five. In case
of m/T = 10 the suppression is of order O(103 ) and successful baryogenesis becomes
extremely difficult. If universal SUSY breaking is assumed this occurs in the regime M0
700 GeV.
The chargino source term is rather sensitive to the relative size of the higgsino mass
parameter, , and the mass of the SU(2)-gauginos, M2 . In Fig. 11a we present the
dependence of the generated baryon asymmetry on M2 , while the remaining parameters
as well as the bubble wall profile are taken from the example of explicit CP-violation
considered in Fig. 6, i.e., the effect of a change in M2 on the bubble wall is ignored.
Moreover, we use the squark spectrum C, = 0.1 Lw = 5/T and vw = 0.03. We find
a resonance structure for , with the peak located at M2 || = 226 GeV. Fig. 10a
corresponds to M2 = 103 GeV. In Fig. 11b we present the dependence of the baryon
227
asymmetry on the soft breaking scalar mass parameter mU3 . Even though this parameter
has no direct impact on the chargino source terms (6.58), (6.59) it alters the statistical
factors kQ3 and kT due to varying stop masses. Baryogenesis becomes less efficient if the
two stop stated are more or less degenerate. In the case under consideration this happens
for mU3 mQ3 = 329 GeV. In Fig. 10a we used mU3 = 256 GeV.
7. Conclusions
In our computation we used the semi-classical approximation to describe the interaction
of the particles in the plasma and the propagating bubble wall which in context of the
MSSM has been introduced in Ref. [64]. Our work differs in various aspects from that
analysis. We included contributions from top quarks which have previously been neglected
and took into account also position-dependent mixing in the chargino mass matrix (flavor
contribution). In Ref. [64] different results were obtained due to an error in the transport
equations, which prevented the cancellation between the dominating helicity parts in the
+
h
1 and h2 source terms. With dispersion relations for canonical momenta, the charged
higgsinos provide in the MSSM CP-violating source terms which are proportional to the
variation of tan in the bubble wall. This result would agree well with Refs. [49,52,58]
where different methods where used to determine the CP-violating source terms. Recently
independent contributions to the source terms in the MSSM have been calculated [60].
However, these arise from higher orders in a Higgs insertion expansion and should not be
included in our semi-classical approach. In the NMSSM non-vanishing source terms for
stops and charginos (and for the top quark) are generated even for constant tan because
of the changing singlet field, and due to CP-violating bubble walls. These arise either from
explicit CP-violation, which in case of the NMSSM is possible already in the tree-level
Higgs potential, or from transitional CP-violation.
Using kinetic momenta in the dispersion relations as suggested convincingly in Ref. [64]
h1 and h 2 sources are exactly equal and tan effects cancel. Thus one has to consider
a source symmetric in the Higgses. This contribution can be also sizeable, if the Yukawa
interactions are not in equilibrium [64] and if the left-handed stop is heavy [73,74]. This
statement is unchanged in our singlet model. The WKB results and their difference to other
work should be further discussed using quantum transport equations.
In our work we investigated the impact of different squark spectra on the emerging
baryon asymmetry, while in Ref. [64] only the massless approximation was considered. As
a consequence, we obtained a non-vanishing baryon asymmetry, while assuming the strong
sphalerons to be in equilibrium. Baryogenesis becomes most efficient for non-degenerate
squarks. The generated baryon asymmetry is roughly proportional to 1/L2w . Since in the
NMSSM the bubble walls can be considerably thinner than in the MSSM, more baryons
can be produced. The baryon asymmetry strongly depends on the velocity of the bubble
wall, which in the NMSSM has not been computed so far. Interesting enough, we find that
the maximal baryon number is generated for vw 102 . This is a typical value for the wall
velocity in the MSSM.
228
Finally, we studied the dependence of the chargino source terms on , M2 , m2Q3 and
m2U3 . We found an enhancement if and M2 become degenerate. This resonance behavior
has also been reported in Refs. [49,52,60,64]. If m2Q3 m2U3 , baryon production is partially
suppressed by strong sphalerons. Concluding, the NMSSM can account for the observed
baryon asymmetry, especially if M2 and if the squark spectrum is rather nondegenerate, or if transitional CP-violation occurs. Both scenarios point to a non-universal
pattern of SUSY breaking.
Acknowledgements
We thank P. John for useful discussions. This work was supported in part by the
TMR network Finite Temperature Phase Transitions in Particle Physics, EU contract no.
ERBFMRXCT97-0122.
References
[1] K. Kajantie, M. Laine, K. Rummukainen, M.E. Shaposhnikov, Phys. Rev. Lett. 77 (1996) 2887.
[2] D. Bdeker, P. John, M. Laine, M.G. Schmidt, Nucl. Phys. B 497 (1997) 387;
M. Carena, M. Quiros, C.E.M. Wagner, Nucl. Phys. B 523 (1998) 3;
M. Losada, Nucl. Phys. B 537 (1999) 3.
[3] M. Laine, K. Rummukainen, Nucl. Phys. B 535 (1998) 423.
[4] See, e.g., A. Pilaftsis, C.E.M. Wagner, Nucl. Phys. B 553 (1999) 3.
[5] S.J. Huber, P. John, M. Laine, M.G. Schmidt, Phys. Lett. B 475 (2000) 104.
[6] H.-P. Nilles, M. Srednicki, D. Wyler, Phys. Lett. B 120 (1983) 346;
J.-M. Frere, D.R.T. Jones, S. Raby, Nucl. Phys. B 222 (1983) 11;
L.E. Ibanez, J. Mas, Nucl. Phys. B 286 (1987) 107.
[7] S.A. Abel, S. Sarkar, P.L. White, Nucl. Phys. B 454 (1995) 663.
[8] M. Pietroni, Nucl. Phys. B 402 (1993) 27.
[9] A.T. Davies, C.D. Froggatt, R.G. Moorhouse, Phys. Lett. B 372 (1996) 88.
[10] See, e.g., N. Polonsky, in: Proc. of Supersymmetry, Supergravity and Superstring, Seoul, 1999,
pp. 100124, hep-ph/9911329.
[11] U. Ellwanger, Phys. Lett. B 133 (1983) 187;
J. Bagger, E. Poppitz, L. Randall, Nucl. Phys. B 455 (1995) 59.
[12] S.A. Abel, Nucl. Phys. B 480 (1996) 55.
[13] C. Panagiotakopoulos, K. Tamvakis, Phys. Lett. B 446 (1999) 224.
[14] S.A. Abel, hep-ph/9603301.
[15] U. Ellwanger, Phys. Lett. B 349 (1995) 57;
H.-P. Nilles, N. Polonsky, Phys. Lett. B 412 (1997) 69;
A. de Gouvea, A. Friedland, H. Murajama, Phys. Rev. D 57 (1998) 5676;
T. Han, D. Marfatia, R.-J. Zhang, Phys. Rev. D 61 (2000) 013007.
[16] S.J. Huber, M.G. Schmidt, Eur. Phys. J. C 10 (1999) 473;
S.J. Huber, in: Proc. of Strong and Electroweak Matter, Copenhagen, 1998, pp. 339343, hepph/9902325.
[17] J.-P. Derendinger, C.A. Savoy, Nucl. Phys. B 237 (1984) 307.
[18] S. Martin, M. Vaughn, Phys. Rev. D 50 (1994) 2282.
[19] U. Ellwanger, M. Rausch de Traubenberg, C.A. Savoy, Nucl. Phys. B 492 (1997) 21.
[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]
229
L. Alvarez-Gaume, J. Polchinski, M. Wise, Nucl. Phys. B 221 (1983) 495.

U. Ellwanger, C. Hugonie, Phys. Lett. B 457 (1999) 299.
S. Abel, C.A. Savoy, Nucl. Phys. B 532 (1998) 3.
For a summary see, e.g., C. Caso et al., Particle Data Group, Eur. Phys. J. C 3 (1998) 1.
ALEPH Collaboration, R. Barate et al., CERN-EP/99-014.
J. Ellis, J. Gunion, H. Haber, L. Roszkowski, F. Zwirner, Phys. Rev. D 39 (1989) 844;
U. Ellwanger, M. Rausch de Traubenberg, C.A. Savoy, Phys. Lett. B 315 (1993) 331;
U. Ellwanger, C. Hugonie, hep-ph/9909260.
OPAL Collaboration, G. Abbiendi et al., Eur. Phys. J. C 7 (1999) 407.
P. Brax, U. Ellwanger, C.A. Savoy, Phys. Lett. B 347 (1995) 269.
L. Dolan, R. Jackiw, Phys. Rev. D 9 (1974) 3320.
M. Dine, R.G. Leigh, P. Huet, A. Linde, D. Linde, Phys. Rev. D 46 (1992) 550.
M.E. Carrington, Phys. Rev. D 45 (1992) 2933;
J.R. Espinosa, M. Quiros, F. Zwirner, Phys. Lett. B 314 (1993) 206.
A.D. Linde, Phys. Lett. B 70 (1977) 306;
A.D. Linde, Nucl. Phys. B 216 (1983) 421.
L. McLerran, M.E. Shaposhnikov, N. Turok, M. Voloshin, Phys. Lett. B 256 (1991) 451.
P. John, Phys. Lett. B 452 (1999) 221.
P. John, in: Proc. of Strong and Electroweak Matter, Copenhagen, 1998, pp. 349353, hepph/9901326.
J.M. Moreno, M. Quirs, M. Seco, Nucl. Phys. B 526 (1998) 489.
J.M. Cline, G.D. Moore, G. Servant, Phys. Rev. D 60 (1999) 105035.
P. John, Ph.D. Thesis, University of Heidelberg, 1999.
W. Buchmller, D. Wyler, Phys. Lett. B 121 (1983) 321;
J. Polchinski, M.B. Wise, Phys. Lett. B 125 (1983) 393.
I.S. Altraev et al., Phys. Lett. B 276 (1992) 242;
E. Commins et al., Phys. Rev. A 50 (1994) 2960.
Y. Kizukuri, N. Oshimo, Phys. Rev. D 45 (1992) 1806.
A. Pomerol, Phys. Lett. B 287 (1992) 331.
M. Matsuda, M. Tanimoto, Phys. Rev. D 52 (1995) 3100.
K.S. Babu, S.M. Barr, Phys. Rev. D 49 (1994) 2156;
N. Haba, M. Matsuda, M. Tanimoto, Phys. Rev. D 54 (1996) 6928.
A.T. Davies, C.D. Frogatt, A. Usai, in: D. Lellouch, G. Mikenberg, E. Rabinovic (Eds.), Proc. of
the International Europhysics Conference on High Energy Physics, Jerusalem, 1997, SpringerVerlag, 1998, p. 891, hep-ph/9712501;
A.T. Davies, C.D. Frogatt, A. Usai, in: Proc. of Strong and Electroweak Matter, Copenhagen,
1998, pp. 227231, hep-ph/9902476.
D. Comelli, M. Pietroni, A. Riotto, Phys. Rev. D 50 (1994) 7703.
D. Comelli, M. Pietroni, Phys. Lett. B 306 (1993) 67;
J.R. Espinosa, J.M. Moreno, M. Quiros, Phys. Lett. B 319 (1993) 505;
K. Funakubo, A. Kakuto, S. Otsuki, F. Toyoda, Prog. Theor. Phys. 99 (1998) 1045.
K. Funakubo, A. Kakuto, S. Otsuki, F. Toyoda, Prog. Theor. Phys. 102 (1999) 389.
M. Carena, M. Quirs, A. Riotto, I. Vilja, C.E.M. Wagner, Nucl. Phys. B 503 (1997) 387.
For reviews see, e.g., P.A. Henning, Phys. Rep. 253 (1995) 235;
C. Greiner, S. Leupold, Ann. Phys. 270 (1998) 328.
L.P. Kadanoff, G. Baym, Quantum Statistical Mechanics, Benjamin, New York, 1962.
A. Riotto, Phys. Rev. D 53 (1996) 5834;
A. Riotto, Phys. Rev. D 58 (1998) 095009.
230
[53] A.G. Cohen, D.B. Kaplan, A.E. Nelson, Phys. Lett. B 245 (1990) 561;
A.G. Cohen, D.B. Kaplan, A.E. Nelson, Nucl. Phys. B 349 (1991) 727;
A.E. Nelson, D.B. Kaplan, A.G. Cohen, Nucl. Phys. B 373 (1992) 453.
[54] K. Funakubo, A. Kakuto, S. Otsuki, K. Takenaga, F. Toyoda, Phys. Rev. D 50 (1994) 1105.
[55] M. Joyce, T. Prokopec, N. Turok, Phys. Rev. Lett. 75 (1995) 1695;
M. Joyce, T. Prokopec, N. Turok, Phys. Rev. D 53 (1996) 2930.
[56] M. Aoki, N. Oshimo, A. Sugamoto, Progr. Theor. Phys. 98 (1997) 1179;
M. Aoki, N. Oshimo, A. Sugamoto, Progr. Theor. Phys. 98 (1997) 1325.
[57] H. Davoudiasl, K. Rajagopal, E. Westphal, Nucl. Phys. B 515 (1998) 384.
[58] P. Huet, A.E. Nelson, Phys. Rev. D 53 (1996) 4597.
[59] M.B. Gavela, P. Hernandez, J. Orloff, O. Pene, C. Quimbay, Nucl. Phys. B 430 (1994) 382;
P. Huet, E. Sather, Phys. Rev. D 51 (1995) 379.
[60] N. Rius, V. Sanz, Nucl. Phys. B 570 (2000) 155.
[61] N. Turok, J. Zadrozny, Nucl. Phys. B 358 (1991) 471.
[62] A.G. Cohen, D.B. Kaplan, A.E. Nelson, Phys. Lett. B 263 (1991) 86.
[63] M. Joyce, T. Prokopec, N. Turok, Phys. Rev. D 53 (1996) 2958.
[64] J.M. Cline, M. Joyce, K. Kainulainen, Phys. Lett. B 417 (1998) 79.
[65] M. Joyce, K. Kainulainen, T. Prokopec, Phys. Lett. B 468 (1999) 128;
M. Joyce, K. Kainulainen, T. Prokopec, in: Proc. of Strong and Electroweak Matter,
Copenhagen, 1998, pp. 232237, hep-ph/9906413;
M. Joyce, K. Kainulainen, T. Prokopec, JHEP 0010 (2000) 029.
[66] G.D. Moore, JHEP 0003 (2000) 006.
[67] G. Moore, T. Prokopec, Phys. Rev. D 52 (1995) 7182.
[68] P. John, M.G. Schmidt, hep-ph/0002050.
[69] K. Olive, astro-ph/9901231.
[70] A.G. Cohen, D.B. Kaplan, Phys. Lett. B 199 (1987) 257;
A.G. Cohen, D.B. Kaplan, Nucl. Phys. B 308 (1988) 913;
A.G. Cohen, D.B. Kaplan, A.E. Nelson, Phys. Lett. B 336 (1994) 41.
[71] J.M. Cline, M. Joyce, K. Kainulainen, JHEP 0007 (2000) 018;
K. Kainulainen, hep-ph/0002273;
M.J. Cline, K. Kainulainen, Phys. Rev. Lett. 85 (2000) 5519.
[72] J.M. Cline, K. Kainulainen, Phys. Lett. B 356 (1995) 19.
[73] S.J. Huber, P. John, M.G. Schmidt, hep-ph/0101249.
[74] M. Carena, J.M. Moreno, M. Quiros, M. Seco, C.E.M. Wagner, hep-ph/0011055.
[75] G.F. Giudice, M. Shaposhnikov, Phys. Lett. B 326 (1994) 118.
[76] G.D. Moore, Phys. Lett. B 412 (1997) 359;
G.D. Moore, N. Turok, Phys. Rev. D 56 (1997) 6533.
[77] D. Bdeker, Phys. Lett. B 426 (1998) 351.

Background field method in the Wilson formulation

M. Bonini, E. Tricarico
Universit degli Studi di Parma and I.N.F.N., Gruppo collegato di Parma, viale delle Scienze,
43100 Parma, Italy
Abstract
A cutoff regularization for a pure YangMills theory is implemented within the background field
method keeping explicit the gauge invariance of the effective action. The method has been applied to
compute the beta function at one loop order. 2001 Elsevier Science B.V. All rights reserved.
PACS: 11.10.Gh; 11.15.-q; 11.15.Bt
Keywords: Renormalization group formalism; Gauge field theory; Ward identity; Background field method
1. Introduction
The Wilson, or exact, renormalization group (RG) approach regards an interacting field
theory as an effective theory, i.e., the higher modes of the fields, with respect to some
scale , generate effective interactions for the lower modes [1]. This division of momenta
conflicts with gauge invariance as it is easy to see in the case of an homogeneous gauge
transformation acting on some matter field (x):
(x) (x)(x).
(1)
In the momentum space the gauge transformed field is mapped into a convolution with the
gauge transformation and then any division of momenta is lost. From the functional point
of view this conflict appears as a breaking term in the SlavnovTaylor (ST) identities for the
Wilsonian effective action [2,3], which originates from the introduction of a cutoff function
into the propagators. Although one can rewrite this as modified SlavnovTaylor [4,5],
it is necessary to prove that the physical effective action satisfies the ST identities. As
shown in Refs. [2,6] this can be achieved, in perturbation theory, by properly fixing the
boundary conditions for the non-invariant couplings in the Wilsonian effective action at the
ultraviolet (UV) scale 0 . This is the so-called fine tuning procedure and it is equivalent
to solve the modified SlavnovTaylor identities at the scale 0 .
E-mail address: bonini@pr.infn.it (M. Bonini).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 2 2 2 - X
232
M. Bonini, E. Tricarico / Nuclear Physics B 606 (2001) 231244
It has been proved [7] for a pure YangMills theory that by using a covariant gauge
fixing function depending on some classical external field, named the background field, it
is possible to define a gauge invariant effective action. In [7] explicit calculations at two
loops order have been performed using the dimensional regularization. A definite proof
that the S-matrix elements can be obtained from the background gauge invariant effective
action has been recently established [8].
These results have led to apply this background field method to the RG formulation, with
the aim of keeping the gauge invariance explicit [9]. To implement this requirement one
allows the cutoff function to depend on the background field in a covariant way. However,
it is easy to realize, by simple one loop computations, that the cutoff vertices are not well
regularized. A possible way out consists in adding a mass term for the quantum gauge fields
and for the ghost [10]. In this way the BRS invariance is lost and the need of restoring the
ST identities requires a fine tuning procedure. Therefore the advantages coming from the
background gauge method are partially lost when the cutoff regularization is used.
The aim of this paper is to show, by an explicit example, how the cutoff regularization
can be successfully implemented within the background field method. In Section 2 we
give the notation. In Section 3 we work out the cutoff regularization preserving the explicit
gauge invariance and we determine the Feynman rules needed to compute the one-loop two
point background amplitude. In Section 4 we show the transversality of this amplitude and
we compute the beta function at one loop order. The detailed calculation of the background
field wave function renormalization is given in Appendix A. Section 5 contains some
remarks and final comments.
2. The background field

The gauge-fixed classical action for a pure YangMills theory is given by
S = Scl + Sgf + SFP ,
(2)
4 a a
1
a
a
a
abc
b
where Scl = 4 d xF F is the classical action, F = A A + gf A Ac
is field strength tensor and f abc are the structure constants of the group. As it has been
proposed in [7], one can choose the gauge fixing term depending on a non-dynamical
field 1

ab b
1
A A b 2 ,
Sgf =
(3)
d 4x D
2
where A a is the background field and the covariant derivative is
ab = ab + gf acb A c .
D
(4)
As a consequence of (3) the ghost term also depends on the background field and reads

ab Dbd cd ,
SFP = d 4 x ca D
(5)
1 We set the gauge fixing parameter = 1 corresponding to the Feynman gauge.
233
with Dab = ab + gf acb Ac . The particular choice of the gauge fixing term makes the
action (2) invariant under the following background gauge transformation:
Aa = Dab b ,
ab b ,
A a = D
(6)
c = gf
c = gf
(7)
abc b c
c ,
c ,
abc b c
where = (x) is the infinitesimal gauge parameter. The invariance of the classical action
under the gauge transformation of the field A is implemented at the quantum level by the
BRS symmetry:
1
(A A)
a ,
ca = gf abc cb cc ,
ca = D
(8)
2
where is a Grassmann parameter. Adding to the action (2) the source term associated to
the BRS variations (8) of A and c one has

g
SBRS [, ] = S + d 4 x ua Dab cb f abc v a cb cc ,
(9)
2
Aa = Dab cb ,
where we have denoted by = (Aa , ca , ca ) and = (ua , v a ) the fields and the BRS
sources.
In the conventional functional approach one defines the generating functional

Z[J, ] = eiW = D eiSBRS [, ]+i(J ) ,
(10)
where J = (ja , a , a ) are the field sources and we introduced the notation

(J, ) = d 4 x ja Aa + a ca + ca a .
(11)
The symmetry of the BRS action with respect to the background gauge transformation (6)
translates for the effective action
[cl , ] = W [J, ] (J cl ),
cl =
W
,
J
(12)
into
a [cl , ] = 0,
Ga [cl , ] + G
(13)
where
Ga = Dab
Ab,cl
b
+ gf abd ccl
d
ccl
b
+ gf abd ccl
d
ccl
(14)
and
a = D
ab
G
A b
(15)
,cl
Similarly the BRS transformation translates into the ST identities. The Ward identity (13)
expresses the gauge invariance of the effective action [cl , ] for A a = Aa,cl .
234
In the generating functional (10) one can set Qa = Aa A a obtaining the new
related to W by
functional W

J, ; A = W [J, ] d 4 x ja A a
W
(16)
and the corresponding effective actions are related by

cl , ; A = [cl , ],
A a,cl = Aa,cl A a .

(17)
From this relation it is evident that the gauge invariant effective action [cl , ]A=A
cl

cl , ; A]
can be obtained from [
,
i.e.,
by
evaluating
1PI
Green
functions
with
Acl =0
background fields on the external legs and the quantum fields Q, c and c inside loops.
A rigorous proof of this background gauge equivalence has been recently given in [8].
One has to realize that the derivation of the Ward identities (13) is purely formal since
the regularization procedure has not been taken into account. If one uses the dimensional
regularization all the symmetries are preserved (for a pure YM theory) and the Ward
identity also holds for the renormalized background effective action. In particular one can
derive the -function from the two point background amplitude, since the gauge coupling
and wave function renormalizations are related [7].
In the following sections we will use the cutoff regularization, by adjusting the Wilson
Polchinski RG approach to the background field method. To maintain the background
gauge invariance we will introduce a covariant cutoff function as proposed in [9].
Moreover, we will be forced to add a mass term to the bare action, to properly regularize
the one loop contribution.
3. Covariant regularization
In the following we will specify to the SU(2) case and, to simplify the notation, we will
use the dot and wedge SU(2) products. The BRS action SBRS expressed in terms of the
quantum gauge field Q is

1
1
2
2

S=
F F + Q D Q + Q F Q c D c
4
2
x

,
+ Sint Q, c, A;
(18)
a is the field strength tensor of the background field and we have explicitly written

where F
the terms quadratic in the quantum field Q and in the ghosts. The remaining terms,
which have been collected in Sint , do not contribute to the one-loop vertex functions with
background fields on the external legs and for this reason it is not necessary to work out
their expression. Notice that the field Q transforms according to the adjoint representation
under the background gauge transformation (6).
To select the modes of the quantum fields below the UV cutoff 0 we introduce a cutoff

function K0 and make the following change of variables in the generating functional W
Q K0 Q ,
c K0 c,
(19)
235
(it is not necessary to introduce a new c field). The invariance of the action with respect
to the background gauge transformation (6) can be maintained if the regularized fields
transform according to the adjoint representation. This can be achieved if K is a function
2 and then we choose
of an appropriate covariant operator, such us D
n
1
n=

D
2 /20 =
K (n) (0)
K0 K D
(20)
= K 2 /20 + ,
n!
0
n=0
d n K(x)
(n)
where K (0) = dx n x=0 and the dots refer to the terms containing the A field. After
the substitution (19), the gauge propagator is multiplied by the factor K(p2 /20 )2 while
the ghost one by the factor K(p2 /20 )1 . As usual in the Wilson RG formulation, one
chooses the cutoff function such as to suppress the propagation of the modes with p2 > 20 .
However, the choice (20) of the cutoff function produces new interactions among the quantum and the background fields which are multiplied by the cutoff function K(p2 /20 ) (or
its derivatives, see later). Thus the loop momenta which are suppressed by the inverse of
the cutoff function in the propagator are enhanced by the cutoff function in these new
vertices and the regularization fails. This fact is not surprising and is essentially a consequence of the Ward identity, which relates the vertex with the inverse of the propagator. 2
To overcome this obstacle we introduce in the action a mass term for the quantum fields

1
Q MQ Q + c Mc c ,
Sm =
(21)
2
x
where the matrices MQ and Mc depend on the cutoff function and must satisfy the requirement that this mass term does not generate relevant interactions in the 0 limit. For
instance, by choosing the exponential covariant cutoff function:

2 /(22 )
2 /20 = eD
0 ,
K D
(22)
the matrices MQ and Mc have the structure

2 )cd db
ab
(D
ab
MQ
= 20 ab K 2
+ K ac
K
,
20

2 )ac cb
(D
ab
2
ab
ab
.
K
Mc = 20 K +
220
(23)
Notice that this structure holds for every cutoff function satisfying the conditions
K(0) = 1,
K (0) = 1/2.
(24)
After using (18), (19) and (21), the regularized action becomes

1
1 2 Q 1 K 2 Q g F
(KQ ) (KQ )
S0 =
F
F
0
4
2
x

.
+ 220 c (1 K)c + Sint KQ, Kc, A;
(25)
2 We thanks Prof. Carlo Becchi for fruitful discussions on this point.
236
This action preserves the background gauge invariance (6) but breaks the BRS symmetry
(7) and a fine tuning procedure must be imposed in order to restore the ST identities. This
analysis can be performed as in the standard Wilsonian approach for gauge theories [2,3].
First one introduces a cutoff dependent BRS transformation, studies the modified ST
identities and determines the non-invariant couplings which compensate the breaking
introduced by the cutoff. From the quantum action principle these couplings are order h
and, therefore, affect the background field amplitudes starting from the second loop order.
In this paper we are only interested in one-loop computations and, therefore, this finetuning problem can be ignored. However, this procedure is unavoidable in the complete
analysis.
From (25) the gauge and ghost propagators read
ab
D
(q) =
ab g
,
20 (1 K 2 (q))
Dab (q) =
ab
,
220 (1 K(q))
(26)
where

2
2
K(q) K q 2 /20 = eq /(20 ) .
(27)
Notice that for large q (i.e., q 0 ) these propagators become constant, and the UV
finiteness of the loop integral will be ensured by the cutoff function in the vertices.
In order to evaluate the Feynman rules coming from (25), we first expand (KQ)a in
terms of the A field. For instance the terms with one and two A fields are given by

(KQ)a 1A =

(KQ)a 2A =
n1

2 k ab 2 nk1 b
1
Q
2n
n! 2n 0 k=0
n=1
(28)
and
n=1
n! 2n 2n
0
n1

2 k ab 2 nk1 b
2
Q
k=0
n=2
n! 2n 2n
0
n2 nk2

k
2 l cb 2 nkl2 b
2 ac
1
Q ,
1
(29)
k=0 l=0
where

acb c
ab
A + A c
1 = g4
(30)
2 ace edb c d
ab
4 A A .
2 =g 4
(31)
and
In the following section we will compute the one loop two point function for the
background field. For this purposes we need to evaluate the Feynman rules with at most
two A fields.
By inserting (28) in (25) the Qa (q)A a11 (p1 )Qb (p)-vertex is given by

aa1 b
(q, p1 , p) = ig4 aa1 b 2K(q)K(p) (g1 p1 g1 p1 )
V
1

+ 20 K(q) + K(p) F (q, p)g (q p)1 ,
(32)
237
where

F (q, p) = K(q) K(p) / q 2 p2 .
The vertex with two Q and two A fields receives contribution from the term
KQ KQ in (25) and from the covariant cutoff functions (29). We do not need
F
to compute the former since it does not contribute to the one loop two point functions (i.e.,
to the tadpole diagram). The remaining terms of the Qa (q)A a11 (p1 )A a22 (p2 )Qb (p)vertex are
aa1 a2 b
V
(q, p1 , p2 , p)
1 2

= g 2 20 g a1 a2 ab a1 b a2 a

K(q) + K(p) F (q, p)g1 2

F (q, p) F (p, q + p1 )
F (q, p) F (q, p + p2 )
K(q) +
K(p)
p2 (p + p2 )2
q 2 (q + p1 )2

+ F (q, q + p1 )F (p, p + p2 ) (2q + p1 )1 (2p + p2 )2 + (1 2). (33)
The interactions of the ghosts with the background fields can be obtained from (25) and
2 /2 )c in powers of A as we have done for KQ . The ca (q)A a11 (p1 )
expanding K(D
0
cb (p)-vertex is given by
Vaa1 1 b (q, p1 , p) = 2i20g4 aa1 b (q p)1 F (q, p).
The
ca (q)A a11 (p1 )A a22 (p2 )cb (p)-vertex is
(34)
given by
Vaa1 1 a22 b (q, p1 , p2 , p)

= 220 g 2 ab a1 a2 a1 b a2 a (2q + p1 )1 (2p + p2 )2

F (q, p) F (q, p + p2 )
g
F
(q,
p)
+
(1
2)
.
1 2
p2 (p + p2 )2
(35)
Though the vertices (32)(35) have been computed using the cutoff function (22), their
expression in term of the functions K(q) and F (q, p) holds for every cutoff function
satisfying (24).
From the above expression for the vertices and the propagators, it is clear that the UV
finiteness of the loop integrals is ensured if the function K(q 2 /20 ) decreases rapidly
enough in the region q 2 20 .
4. One loop computations
As briefly discussed in Section 2, one is only interested to discuss vertices with
background external fields. We will explicitly compute the two point amplitude for the
background field and we will verify that is transverse, as it must be since the regularization
238
(a)
(b)
(c)
(d)
Fig. 1. Graphical contribution to the two point function with background external legs which are
depicted as curly lines. The wavy and the full lines in the loops refer to the quantum gauge and ghost
field, respectively.
preserves the background gauge invariance. There are four Feynman graphs contributing
to this amplitude, which are depicted in Fig. 1(a)(d). The corresponding loop integrals
are
[1a]ab
G
(p; 0 )

16K(q)F (q, q)g + (2q + p) (2q + p) 4F 2 (q, q + p)

= g 2 ab
q

F (q, q) F (q, q + p)
1
+8
K(q) + (p p)
,
2
2
q (q + p)
1 K 2 (q)
[1b]ab
G
(p; 0 ) = g 2 ab
(36)

2
2
4(2q + p) (2q + p) F (q, q + p) K(q) + K(q + p)

q
K 2 (q)K 2 (q + p)

2
+ 8 g p p p
40
1
(37)
,
2
(1 K (q))(1 K 2 (q + p))
239
[1c]ab
G
(p; 0 )

F (q, q) F (q, q + p)
2 ab
= 2g
2F (q, q)g + (2q + p) (2q + p)
q 2 (q + p)2

q
1
(38)
,
+(p p)
1 K(q)

F 2 (q, q + p)
[1d]ab
G
,
(p; 0 ) = 2g 2 ab (2q + p) (2q + p)
(1 K(q))(1 K(q + p))
q
(39)
where in the first two contribution the symmetry factor 1/2 has been included and
F (q, q) = lim F (q, q + p) =

p0
2 2 K (q)
d
K
q /0
.
dq 2
20
One can easily verify the transversality of the two-point function. Indeed form the gauge
field loop (Fig. 1(a) and (b)) one finds
[1a]ab

[1b]ab
p p G
+ G

K 2 (q) K 2 (q + p)
(K 2 (q) K 2 (q + p))2
2 ab
2
+
= 4g
1 K 2 (q)
(1 K 2 (q))(1 K 2 (q + p))
q

= 8g 2 ab
q

K 2 (q) K 2 (q + p)
K 2 (q)
= 0,
1
1 K 2 (q)
1 K 2 (q + p)
where the last two equalities has been obtained by performing the change of the
integration variable q p q. Similarly from the ghost loop (Fig. 1(c) and (d)) one
finds
[1c]ab

[1d]ab
p p G
+ G

(K(q) K(q + p))2
K(q) K(q + p)
2 ab
+
= 2g
2
= 0.
1 K(q)
(1 K(q))(1 K(q + p))
q
Therefore, the four graphs (36)(39) sum up to

ab
G
(p, 0 ) = ab p2 g p p G(p, 0 ),
where
G(p, 0 ) =
1 aa
G (p, 0 ).
6p2
The renormalized background two point amplitude

ab
(p)p2 =2 = 0,
(p) = ab p2 g p p (p),
at one loop level is given by

(p) = lim G(p, 0 ) 1 (/0 ) ,
0
(40)
240
where the relevant coupling 1 is

1 (/0 ) = G(p, 0 )p2 =2
(41)
and is the renormalization scale. This relevant coupling, which depends on the
regularization, will be computed in the appendix in the case of a polynomial cutoff
function.
4.1. One-loop beta function
In the Wilsonian approach one can show [11] that the beta function can be obtained from
the bare coupling constant gB by
(g) g =
0 0 gB
.
g gB
(42)
Due to the background gauge invariance the bare coupling constant gB is related to the
relevant coupling 1 by
gB = g(1 + 1 )1/2
from which, at one-loop order, one obtains
1
(g) = 0
(43)
1 (/0 ).
2
0
In Appendix A, we will compute this coupling by using a polynomial cutoff function.
However, the first coefficient of the beta function is independent of the regularization and
therefore it should be computed without specifying the cutoff function.
In order to determine the beta function we first consider the contribution to (40)
originating from the two diagrams with the gauge field loop (see Figs. 1(a) and (b)).
From (36) and (37) one obtains
[1a]aa
[1b]aa
G
+ G

8K(q)K (q)
3p2 K 2 (q)K 2 (q + p)
= 16g 2
+
40 (1 K 2 (q))(1 K 2 (q + p)) 20 (1 K 2 (q))
q

(2q + p)2
K(q)K (q) K(q + p)K (q + p)
2
.
1 K 2 (q + p)
0 p (2q + p) 1 K 2 (q)
(44)
Similarly, from the two diagrams with the ghost loop (see Figs. 1(c) and (d)) one gets
[1c]aa
[1d]aa
+ G
G

(2q + p)2
K (q + p)
4g 2
8K (q)
K (q)
= 2
.
1 K(q) p (2q + p) 1 K(q) 1 K(q + p)
0
(45)
Because of the infrared divergence 3 one cannot set p2 = 2 = 0 in these integrals,

however, as far as the beta function is concerned, in (42) one can take /0 0 [11].
3 Notice that the mass term (21) does not change the infrared behavior of the propagators.
Therefore, from (40) and (43), the ghost contribution to the beta function is

1 [1c]aa
1
[1d]aa

0
G
+
G
2
2
0 62
=0
241
(46)
and, by Taylor expanding in p the integrand of (45), it becomes

1 g2
3 16 2
2 2
1

dq q 0
k (x) + 3xk (x) + x k (x) ,
0 40
3
2 2
(47)
where k(x) [K (x)/(1 K(x))] and x = q 2 /20 . After integrating by parts and
using (24), this contribution is
g2 2
.
16 2 3
Performing a similar analysis for the gauge loop contribution, one obtains

1 [1a]aa
1
[1b]aa

+ G
0
G
2
2
0 62
=0
4 g2
1
3K 4 (x)
2 2
=
dq
q
+ h (x)
0
3 16 2
0 40 (1 K 2 (x))2

0
2 2

+ 3xh (x) + x h (x)
3
20 g 2
,
3 16 2
(48)
with h(x) [K(x)K (x)/(1 K 2 (x))]. Summing up the two contributions one obtains
the well-known one-loop result
(g) =
22 1
g3 .
3 16 2
5. Conclusions
In this paper we have implemented a cutoff regularization which maintains the gauge
invariance explicit. It is based on the introduction of a background field and a cutoff
regulator covariantly depending on this field. The finiteness of the loop integrals is ensured
by the presence of a cutoff function which multiplies all interactions and suppresses the
loop momenta in the UV region. Moreover, by adding a mass term for the quantum fields
the propagators remain finite in this region. This mass term is necessary if one wants to
keep gauge invariance and, at the same time, to have an efficient regularization. The only
request we made is that this mass term does not introduce relevant interactions into the
action when removing the cutoff. The explicit gauge invariance of the effective action has
been exploited to compute the beta function at one loop order from the wave function
renormalization of the background field. The fact that this result is independendent of the
cutoff function choice is a check of the consistency of our computation.
242
All the background amplitudes can be made finite to all loops by using an appropriate
cutoff function, such as the exponential one (22). The main problem one has to address
before extending this analysis to higher loops lies on the explicit BRS symmetry breaking
introduced by the mass term. Although this symmetry only affects the quantum fields,
the ST identities are necessary to show the background gauge equivalence [8]. Therefore
a complete analysis requires the determination of the symmetry breaking counterterms.
As in the standard (i.e., without the background field) Wilsonian formulation of gauge
theories, these counterterms can be calculated by introducing a generalized BRS symmetry,
dependent on the cutoff function, and studying the corresponding ST identities. Also in
this case one can show that by imposing these modified ST identities, the renormalized
effective action satisfies the ST, in the limit in which the cutoff is removed. The solution
of this fine tuning problem is outside the aim of the present work and is left to a further
publication. We only remarque that the background gauge invariance greatly simplifies the
determination of the symmetry breaking counterterms. Moreover, in order to determine
background gauge amplitudes only few of these counterterms are needed. For instance
for computing the second coefficient of the beta function one only needs to determine the
couplings of the interactions with at most two quantum gauge fields.
Acknowledgements
We have greatly benefited from discussions with Carlo Becchi and Giuseppe Marchesini.
Appendix A
In this appendix we compute the relevant coupling 1 given in (41). To performe this
calculation one needs to specify the cutoff function. We use the polynomial cutoff function
K(q) =
1
(1 + q 2 /(420 ))2
(A.1)
which satisfies the conditions (24) and allows to evaluate the Feynman integrals by
exploiting the Feynman parameterization.
We first concentrate on the ghost loop contribution (45), which becomes
[1c]aa
[1d]aa
+ G
G

= 128g 2 40
q
8
q 2 (q 2
+ 420 )(q 2
+ 820 )

(2q + p)2
+
q 2 (q 2
q 2 (q
+ p)2 (q 2
+ 420 )(q 2
1
+ 420 )(q 2 + 820 )
1
+ 820 )((q + p)2 + 820 )
243
1
(q 2 + 420 )(q 2 + 820 )(q + p)2 ((q + p)2 + 420 )
One can easily see that the q-integral is UV. The first term in the integrand is independent
of p and can be easily evaluated obtaining

1620 2
1
=
g log 2.
1024 g 240
(A.2)
2
2
2
q 2 (40 + q 2 )(80 + q 2 )
q
For the remaining terms one uses the Feynman parameterization. For instance one has

(2q + p)2
q 2 (420 + q 2 )(820 + q 2 )(q + p)2
q

= 3!

q x,y,z
[2q + p(1 2z)]2

q 2 + z(1 z)p2 + 420 (x + 2y)
4 ,
(A.3)
1 1z 1xy

dx. After integrating with respect to q and the
where x,y,z 0 dz 0 dy 0
Feynman parameters the ghost loop contribution gives

p2
p2 2 1
p2
+ log 2 + O p4 /20 .
2 log 2 + 2
(A.4)
4
0 3 4
The gauge loop contribution (44), can be evaluated in a similar way. The presence of the
square of the cutoff function makes the Feynman integrals more involved. By inserting the
cutoff function (A.1) in (44) one obtains
[1a]aa
[1b]aa
+ G
G

= 512 g 280
q
+
+
+
6144 p240
((420 + q 2 )4 (4 20 )4 )((420 + (q + p)2 )4 (4 20 )4 )
+ q 2 )(820

+ p)2
q 2 (420
8
+ q 2 )(3240 + 820 q 2 + q 4 )
(2q
1
2
2
4
2
2
2
2
(80 + q ) (40 + q )(320 + 80 q + q 4 )(q + p)2 (420 + (q + p)2 )
820 + q 2 + (q + p)2
q 2 (420 + q 2 )(3240 + 820 q 2 + q 4 )(3240 + 820 (q + p)2 + (q + p)4 )
1
+ (q + p)2 )(820 + (q + p)2 )

1
+
.
q 2 (q + p)2 (420 + q 2 )(3240 + 8q 2 20 + q 4 )
(3240
+ 820 q 2
+ q 4 )(q
+ p)2 (420
(A.5)
One can see that this integral is UV finite and that it vanishes for p = 0. We only compute
the contribution originating from the first and the last terms since the others do not generate
log p2 /20 , as can be seen by taking the 0 limit in the integrand. After applying the
244
Feynman parameterization and integrating on the whole set of variables one obtains

5p2
p2
log
+ O p2 .
2
2 2
0
(A.6)
Adding the two contributions from the ghost and the gauge loop one obtains
G(p/0 ) =

g 2 22
log p2 /20 + O(1).
2
16 3
(A.7)
References
[1] K.G. Wilson, Phys. Rev. B 4 (1971) 3174, 3184;
K.G. Wilson, J.G. Kogut, Phys. Rep. 12 (1974) 75;
J. Polchinski, Nucl. Phys. B 231 (1984) 269.
[2] C. Becchi, On the construction of renormalized quantum field theory using renormalization
group techniques, in: M. Bonini, G. Marchesini, E. Onofri (Eds.), Elementary Particles, Field
Theory and Statistical Mechanics, Parma University, 1993.
[3] M. Bonini, M. DAttanasio, G. Marchesini, Nucl. Phys. B 421 (1994) 429;
M. Bonini, M. DAttanasio, G. Marchesini, Nucl. Phys. B 437 (1995) 163;
M. Bonini, M. DAttanasio, G. Marchesini, Phys. Lett. B 346 (1995) 87.
[4] U. Ellwanger, Phys. Lett. B 335 (1994) 364.
[5] M. DAttanasio, T.R. Morris, Phys. Lett. B 378 (1996) 213.
[6] M. Bonini, E. Tricarico, Nucl. Phys. B 585 (2000) 253.
[7] L.F. Abbott, Nucl. Phys. B 185 (1981) 189.
[8] C. Becchi, R. Collina, Nucl. Phys. B 562 (1999) 412.
[9] M. Reuter, C. Wetterich, Nucl. Phys. B 417 (1994) 81;
F. Freire, C. Wetterich, Phys. Lett. B 380 (1996) 337;
F. Freire, D.F. Litim, J. M Pawlowski, Phys. Lett. B 495 (2000) 256.
[10] C. Becchi, Introduction to the background gauge, Lecture notes delivered at the School
IX Seminario Nazionale di Fisica Teorica, Parma, September 2000, to be published.
[11] M. Bonini, G. Marchesini, M. Simionato, Nucl. Phys. B 483 (1997) 475.

QCD factorization in B K, decays and

extraction of Wolfenstein parameters
M. Beneke a , G. Buchalla b , M. Neubert c , C.T. Sachrajda d
a Institut fr Theoretische Physik E, RWTH Aachen, D-52056 Aachen, Germany
b Theory Division, CERN, CH-1211 Geneva 23, Switzerland
c Newman Laboratory of Nuclear Studies, Cornell University, Ithaca, NY 14853, USA
d Department of Physics and Astronomy, University of Southampton, Southampton SO17 1BJ, UK
Abstract
In the heavy-quark limit, the hadronic matrix elements entering nonleptonic B-meson decays
into two light mesons can be calculated from first principles including nonfactorizable stronginteraction corrections. The B K, decay amplitudes are computed including electroweak
penguin contributions, SU(3) violation in the light-cone distribution amplitudes, and an estimate of
power corrections from chirally-enhanced terms and annihilation graphs. The results are then used to
reduce the theoretical uncertainties in determinations of the weak phases and . In that way, new
constraints in the (,
)
plane are derived. Predictions for the B K, branching ratios and
CP asymmetries are also presented. A good global fit to the (in part preliminary) experimental data
on the branching fractions is obtained without taking recourse to phenomenological models. 2001
PACS: 13.25.Hw
1. Introduction
The study of nonleptonic two-body decays of B mesons is of primary importance for the
exploration of CP violation and the determination of the flavour parameters of the Standard
Model. Because of the interference of several competing amplitudes, these processes allow
for the presence of different weak- and strong-interaction phases, which play a crucial
role for CP violation. In the Standard Model, all CP-violating observables are related to
the complex phase of the quark mixing matrix, which in turn implies nontrivial angles
+ V V + V V = 0. With the standard choice of
in the unitarity triangle Vud Vub
cd cb
td tb
), as well
phase conventions, one defines the weak phases = arg(Vt d ) and = arg(Vub
E-mail address: mbeneke@physik.rwth-aachen.de (M. Beneke).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 2 5 1 - 6
246
M. Beneke et al. / Nuclear Physics B 606 (2001) 245321
as = 180 . In the Standard Model, sin 2 can be extracted in a theoretically

0 J /KS .
clean way by measuring the time-dependent rates for the decays B 0 , B
The measurement of (or ) is more difficult, since it requires controlling the hadronic
dynamics in nonleptonic B decays.
A promising strategy for the determination of is based on rate measurements for the
charged modes B (K) and B 0 [1]. Hadronic uncertainties in this method
can be reduced to a minimum by exploiting the structure of the effective weak Hamiltonian
and using isospin and SU(3) flavour symmetries [25]. If only measurements of CPaveraged branching ratios are available, it is still possible to derive bounds on [3,6], or to
determine under the assumption of only a moderate strong-interaction phase between
penguin and tree amplitudes. A different strategy for extracting uses SU(3)-symmetry
d + and
relations between the various contributions to the time-dependent Bd , B
+
s K K decay amplitudes [7]. The main theoretical limitation of these methods

Bs , B
is in the accuracy with which the effects of SU(3) breaking can be estimated.
The angle can be determined from the time-dependent CP asymmetry in the decays
0 + , if the subdominant penguin contribution to the decay amplitudes can
B0, B
be subtracted in some way. This can be achieved by an isospin analysis [8]; however, in
practice that route is extremely challenging due to the difficulty in measuring the very
0 0 0 branching ratios. One must therefore rely on dynamical input for
small B 0 , B
the penguin-to-tree ratio [911]. Alternatively, can be measured in related decays such
as B , for which the penguin contribution can be eliminated using a time-dependent
analysis of the B + 0 Dalitz plot [12].
Most of the above-mentioned determinations of or have theoretical limitations,
which would be reduced if some degree of theoretical control over two-body nonleptonic
B decays could be attained. However, in the past this has proven to be a very difficult
problem. Even advanced methods such as lattice gauge theory, QCD sum rules, or the largeNc expansion have little to say about the QCD dynamics relevant to hadronic B decays.
In recent work, we have developed a systematic approach to this problem. It is based on
the observation that, in the heavy-quark limit mb
QCD , a rigorous QCD factorization
formula holds for the two-body decays B M1 M2 , if the emission particle M2 (the
meson not obtaining the spectator quark from the B-meson) is a light meson. (It has been
argued that perhaps the large-Nc limit may be more relevant to factorization than the heavyquark limit [13]. However, the dramatic decrease of nonfactorizable effects seen when
comparing K , D K and B K decays shows that the heavy-quark limit is
of crucial importance.) We have previously applied the factorization formula to B
decays and obtained results for the decay amplitudes at next-to-leading order in s and to
leading power in QCD /mb [14]. The conceptual foundations of our approach have been
discussed in detail in [15], which focused on the simpler case of decays into heavylight
final states (where M1 is a charm meson). In the present work, the QCD factorization
formula is applied to the general case of B decays into a pair of light, flavour-nonsinglet
pseudoscalar mesons. (Preliminary results of this analysis have been presented in [16].)
The present analysis contains three new theoretical ingredients in addition to a much more
detailed phenomenological analysis:
247
1. Matrix elements of electroweak penguin operators are included, which play an important role in charmless decays based on b s qq
transitions. This is a straightforward
extension of our previous analysis. However, a sensible implementation of QCD corrections to electroweak penguin matrix elements implies that one departs from the usual
renormalization-group counting, in which the initial conditions for the electroweak penguin operators at the scale = MW are treated as a next-to-leading order effect.
2. Hard-scattering kernels are derived for general, asymmetric meson light-cone distribution amplitudes. This is important for addressing the question of nonfactorizable SU(3)breaking corrections, since the distribution amplitudes of strange mesons are, in general,
not symmetric with respect to the quark and antiquark momenta.
3. The leading power corrections to the heavy-quark limit are estimated by analyzing
chirally-enhanced power corrections related to certain twist-3 distribution amplitudes
for pseudoscalar mesons. We also discuss potentially large power corrections arising from
annihilation topologies, noted first in [17]. This is essential for controlling the theoretical
uncertainties of our approach.
The second and third items have not been considered in previous generalizations of
the results of [14] to the case of B K decays [1821]. Chirally-enhanced power
corrections were discussed in [22,23], but we disagree with some of the results obtained
by these authors.
The QCD factorization approach provides us with model-independent predictions for
the decays amplitudes including, in particular, their strong-interaction phases. The same
method can also be applied to other charmless decays, such as vectorpseudoscalar [19]
or vectorvector modes. Our main focus here is on the development of the new conceptual
aspects of the approach that are important for a comprehensive phenomenological analysis.
This includes a detailed discussion of various sources of potentially large power corrections
to the heavy-quark limit. We then perform a comprehensive study of CP-averaged
branching fractions and direct CP asymmetries in decays to K and final states,
including a detailed discussion of the theoretical uncertainties from all inputs to the QCD
factorization approach. In many of the phenomenological applications discussed in this
work the dynamical information obtained using the QCD factorization formalism is used
in a minimal way, to reduce the hadronic uncertainties in methods that are theoretically
clean up to nonfactorizable SU(3)-breaking effects. Most importantly, these strategies
do not suffer from uncertainties related to weak annihilation contributions. In this way, it
is possible to reduce the hadronic uncertainties in the strategies for determining from
B K, decays proposed in [1,4] to the level of nonfactorizable corrections
that simultaneously violate SU(3) symmetry and are power suppressed in the heavyquark limit. These corrections are parametrically suppressed by the product of three small
quantities: 1/Nc , (ms md )/QCD , and QCD /mb . As a consequence, we argue it will
eventually be possible to determine with a theoretical accuracy of about 10 (unless
is much different from its expected value in the Standard Model). More accurately, the
strategies discussed here can constrain the Wolfenstein parameters and with accuracies
248
mixing. Given
similar those obtained from the standard global fit to |Vub |, K , and BB
the theoretical input discussed in this paper, even the present, preliminary data on the rare
hadronic decays exclude at 95% confidence level half of the parameter space obtained from
the standard fit.
The QCD factorization approach provides a complete theoretical description of all B
P P decay amplitudes. This allows for a large variety of predictions, which go far beyond
those explored in the present work. In the future, this will offer the possibility of several
nontrivial experimental tests of the factorization formula. A more detailed discussion of
these predictions will be presented elsewhere.
The remainder of this paper is organized as follows: In Section 2, we collect some basic
formulae and express the B K, decay amplitudes in terms of parameters ai and
bi appearing in the effective, factorized transition operators for these decays (including
weak annihilation contributions). Section 3 contains the technical details of the calculations
based on the factorization formula, a discussion of annihilation effects, and a compilation
of the relevant formulae for the numerical evaluation of our results. Readers not interested
in the technical aspects of our work can proceed directly to Sections 4 and 5, where
we present numerical values for the amplitude parameters ai and bi (Section 4) and
discuss phenomenological applications of our results (Section 5). Specifically, we consider
strategies to bound and determine the weak phase and to extract sin 2 from mixinginduced CP violation in B + decay. We also present predictions for CP-averaged
branching fractions and CP asymmetries, and perform a global fit in the (,
)
plane to all
measured B K, branching fractions. A critical comparison of our formalism with
other theoretical approaches to hadronic B decays is performed in Section 6.
2. Parameterizations of the decay amplitudes

The effective weak Hamiltonian for charmless hadronic B decays consists of a sum of
local operators Qi multiplied by short-distance coefficients Ci and products of elements of
or = V V . Below we will focus on B K
the quark mixing matrix, p = Vpb Vps
pb pd
p
decays to be specific; however, with obvious substitutions a similar discussion holds for all
other B decays into two light, flavour-nonsinglet pseudoscalar mesons. Using the unitarity
relation t = u + c , we write

GF
p
p
Heff =
p C1 Q1 + C2 Q2
2 p=u,c

+
Ci Qi + C7 Q7 + C8g Q8g + h.c.,
(1)
i=3,...,10
p
where Q1,2
Q3,...,6 and
are the left-handed currentcurrent operators arising from W -boson exchange,

Q7,...,10 are QCD and electroweak penguin operators, and Q7 and Q8g are
the electromagnetic and chromomagnetic dipole operators. They are given by

p
p
V A s p V A ,
Q2 = pi bj V A sj pi V A ,
Q1 = pb
249

Q3 = sb V A
qq
V A ,

Q4 = si bj V A
qj qi V A ,

qq
V +A ,
Q5 = sb V A

qj qi V +A ,
Q6 = si bj V A

3
V +A ,
Q7 = sb V A
2 eq qq

3
Q8 = si bj V A
2 eq q j qi V +A ,

3
V A ,
Q9 = sb V A
2 eq qq
Q10 = si bj

V A
3
2 eq

qj qi V A ,
e
gs
Q7 =
(2)
mb s (1 + 5 )F b,
Q8g =
mb s (1 + 5 )G b,
8 2
8 2
where (q1 q2 )V A = q1 (1 5 )q2 , i, j are colour indices, eq are the electric charges of
the quarks in units of |e|, and a summation over q = u, d, s, c, b is implied. (The definition
of the dipole operators Q7 and Q8g corresponds to the sign convention iD = i +
gs Aa ta for the gauge-covariant derivative.) The Wilson coefficients are calculated at a high
scale MW and evolved down to a characteristic scale mb using next-to-leading
order renormalization-group equations. The essential problem obstructing the calculation
of nonleptonic decay amplitudes resides in the evaluation of the hadronic matrix elements
of the local operators contained in the effective Hamiltonian.
Applying the QCD factorization formula and neglecting power-suppressed effects, the
matrix elements of the effective weak Hamiltonian can be written in the form [14,15]
GF

B =
B ,
p K|Tp + Tpann
K|Heff
(3)
2 p=u,c
where

Tp = a1 (K)pu ub
V A s u V A + a2 (K)pu s b V A uu
V A

p
+ a3 (K)
V A + a4 (K)
s b V A qq
qb
V A s q V A
q

p
+ a5 (K)
V +A + a6 (K)
(2) qb
SP s q S+P
s b V A qq

s b V A 32 eq qq
V +A
+ a7 (K)
q
p
+ a8 (K)

(2) qb
SP 32 eq s q S+P
q

s b V A 32 eq qq
V A
+ a9 (K)
q
p
+ a10 (K)

qb
V A 32 eq s q V A .
(4)
Here (q1 q2 )SP = q1 (1 5 )q2 , and a summation over q = u, d is implied. The symbol
indicates that the matrix elements of the operators in Tp are to be evaluated in the
|j1 |BK|j

factorized form K|j1 j2 |B
2 |0 or K|j1 |B|j
2 |0, as appropriate.
250
Nonfactorizable corrections are, by definition, included in the coefficients ai . The matrix
elements for B mesons (i.e., mesons containing a b-antiquark)

are obtained from (3) by
CP conjugation. A corresponding result, with obvious substitutions, holds for other decays
such as B .
The term Tpann in (3) arises from weak annihilation contributions and introduces a
set of coefficients bi (K), which we shall define and discuss in detail in Section 3.5.
Annihilation contributions are suppressed by a power of QCD /mb and not calculable
within the QCD factorization approach. Nevertheless, we will include the coefficients bi
in the amplitude parameterizations in this section.
The coefficients ai multiplying products of vector or axial-vector currents are renormalization-scheme invariant, as are the hadronic matrix elements of these currents. For
the coefficients a6 and a8 a scheme dependence remains, which exactly compensates the
scheme dependence of the hadronic matrix elements of the scalar or pseudoscalar densities
associated with these coefficients. These matrix elements are power suppressed by the ratio
rK () =
2m2K
,
m
b ()(
mq () + m
s ())
(5)
which is formally of order QCD /mb but numerically close to unity. In the following
we shall use the same notation for charged (q = u) and neutral kaons (q = d), since the
difference is tiny. A corresponding ratio
r () =
2m2
,
m
b ()(
mu () + m
d ())
(6)
appears in the discussion of the B decay amplitudes. For a phenomenological

analysis of nonleptonic B decays it is necessary to estimate these chirally-enhanced
corrections despite the fact that they are formally power suppressed. A detailed discussion
of these corrections will be presented in Section 3.2, and their numerical importance will
be investigated in Section 4.
In terms of the parameters ai , the B K decay amplitudes (without annihilation
contributions) are expressed as

0 = p a p 1 a p + rK a p 1 a p AK ,
A B K
4
6
2 10
2 8

p
p
p
p
2 A B 0 K = u a1 + p a4 + a10 + p rK a6 + a8 AK

3
+ u a2 + p (a7 + a9 ) AK ,
2
0

p

+
K = u a1 + p a + a p + p rK a p + a p AK ,
A B
4
10
6
8
0

0 = A B K
0 + 2 A B 0K
0K
2A B
0

+K .
A B
(7)
, a a (K), and a summation over p = u, c is implicitly understood
Here p = Vpb Vps
i
i
p
in expressions like p ai . The last relation follows from isospin symmetry. The CPconjugate decay amplitudes are obtained from the above by replacing p p . We have
251
defined the factorized matrix elements

GF
AK = i m2B m2 F0B m2K fK ,
2

GF 2
AK = i mB m2K F0BK m2 f ,
2
(8)
where F0BM (q 2 ) are semileptonic form factors. Weak annihilation effects contribute
further terms to the decay amplitudes, which can be parameterized as

0 = u b2 + (u + c ) b3 + bEW BK ,
Aann B K
3

0 ,
2 Aann B 0 K = Aann B K

0

1 EW
+

Aann B K = (u + c ) b3 b3
BK ,
2
0

0

0K
+K ,
0 = Aann B
(9)
2 Aann B
where
GF
BK = i fB f fK .
2
(10)
The coefficients bi bi (K) will be defined in Section 3.5, but we may note here that b1,2
p
p
are related to the currentcurrent operators Q1 and Q2 in the effective Hamiltonian (1),
EW
and b3,4 (b3,4 ) are related to QCD (electroweak) penguin operators.
The B decay amplitudes are given by

0
+ = u a1 + p a p + a p + p r a p + a p A ,
A B
4
10
6
8

3
p
0

p
2 A B = u (a1 + a2 ) + p a7 + r a8 + a9 + a10 A ,
2
0

0

0 0
0

+ ,
A B = 2A B A B
(11)
, a a (), and
where now p = Vpb Vpd
i
i

GF
A = i m2B m2 F0B m2 f .
2
(12)
The additional annihilation contributions are

0

+ = u b1 + (u + c ) b3 + 2b4 1 b EW + 1 b EW B ,
Aann B
2 3
2 4

2 Aann B 0 = 0,
0

0

0 0 = Aann B
+ ,
Aann B
(13)
where bi bi (), and
GF
B = i fB f2 .
2
(14)
252
Neglecting tiny mass corrections of order (m,K /mB )2 ,

RK
F BK (0)f
AK
0B
,
AK
F0
(0)fK
AK
fK

.
A
f
(15)
These ratios will play an important role in the discussion of SU(3) violations in Section 5.
The expressions collected above provide a complete description of the decay amplitudes
in terms of the parameters ai and bi for the various processes. They are the basis
for most of the phenomenological applications discussed in this work. However, in the
literature several alternative parameterizations of the B K decay amplitudes have
been introduced, which are sometimes useful when considering CP asymmetries or ratios
of branching fractions. We briefly elaborate on one such parameterization here, adopting
the notations of [4]. The dominant contributions to the B K decay amplitudes come
from QCD penguin operators. Because the corresponding operators in the effective weak
Hamiltonian preserve isospin, this contribution is the same (up to trivial ClebschGordan
coefficients) for all decay modes. Isospin-violating contributions to the decay amplitudes
are subdominant and arise from the currentcurrent operators Qu1 and Qu2 (so-called tree
contributions), and from electroweak penguins. The latter are suppressed by a power of
/s , whereas the former are suppressed by the ratio
KM e
u
= tan2 C Rb ei ,
c
where C is the Cabibbo angle,

|Vub |
= 2 + 2 ,
Rb = cot C
|Vcb |
(16)
(17)
is one of the sides of the unitarity triangle, and and are the Wolfenstein parameters.
A general parameterization of the decay amplitudes is

0 = P 1 + a eia ei ,
A B K

2 A B 0 K = P 1 + a eia ei 3/2ei ei qei ,
0

+ K = P 1 + a eia ei T eiT ei qC eiC .
A B
(18)
0 0K
0 ) is then
(The phase eia was denoted ei in [4].) The amplitude A(B
determined by the isospin relation shown in the last line of (7). The dominant penguin
0 amplitude that
amplitude P is defined as the sum of all contributions to the B K
are not proportional to ei . This quantity cancels whenever one takes ratios of decay
amplitudes, such as CP asymmetries or ratios of branching fractions. The parameters
3/2 and T measure the relative strength of tree and QCD penguin contributions, q and
qC measure the relative strength of electroweak penguin and tree contributions, and a
0 amplitude arising from upparameterizes a rescattering contribution to the B K
quark penguin topologies. Moreover, , T , , C , and a are strong rescattering phases.
(The strong-interaction phase of P is not observable and can be set to zero.) All parameters
except qei receive weak annihilation contributions.
253
It is straightforward to express the various amplitude parameters in terms of the

parameters ai ai (K) and bi bi (K) defined earlier. We obtain

1 c
1
P = c
(19)
a4c a10
+ rK a6c a8c AK + b3 + b3EW BK ,
2
2
for the leading penguin amplitude, and for the remaining parameters
3/2 ei =
KM
T eiT =
KM
a eia =
qei =
KM
u + r K au + R
(a1 + RK a2 ) + 32 [a10
K (a9 a7 )]
8
c
(a4c + rK a6c ) 12 (a10
+ rK a8c ) + rA (b3 + b3EW )
u
a1 + 32 (a10
+ rK a8u ) rA (b2 + 32 b3EW )
c
(a4c + rK a6c ) 12 (a10
+ rK a8c ) + rA (b3 + b3EW )
qC eiC =
c
(a4c + rK a6c ) 12 (a10
+ rK a8c ) + rA (b3 + b3EW )
c
a10
+ rK a8c + RK (a9 a7 )
KM
u
(a1 + RK a2 ) + 32 [a10
+ rK a8u + RK (a9 a7 )]
3
2
u
(a4u + rK a6u ) 12 (a10
+ rK a8u ) + rA (b2 + b3 + b3EW)
3
2
KM
c
a10
+ rK a8c rA b3EW
u
a1 + 32 (a10
+ rK a8u ) rA (b2 + 32 b3EW )
(20)
where RK = AK /AK is the ratio of the two factorized amplitudes given in (15), and
rA =
BK
fB f
2 B
.
AK
mB F0
(0)
(21)
Our notation for amplitude ratios is such that the ratio RK (denoted by a capital R)
deviates from 1 only by SU(3)-breaking corrections, whereas the ratios rA and rK
(denoted by a lower-case r) are formally of order QCD /mb in the heavy-quark limit.
However, whereas rA 0.003 is indeed very small, rK 0.7 (at a scale 1.45 GeV) is
numerically large for real B mesons. Finally, note that the electroweak penguin coefficients
a7,...,10 could be safely neglected in all quantities other than qei and qC eiC , because
they are tiny compared with the other coefficients a1,...,6 . In (20), they are included only
for completeness. (The systematics of including electroweak penguin contributions will be
discussed in more detail later.)
An important quantity affecting the determination of sin 2 from the time-dependent CP
0 + is the ratio of penguin to tree amplitudes. In
asymmetry in the decays B 0 , B
this case the tree contribution is no longer CKM suppressed, since
u
= Rb ei ,
c
(22)
0 + decay amplitude in (11) can be written as

is of order unity. The B
0

+ ei + P ,
A B
T
(23)
254
where
c

P
1 c
=
+ r a8c + rA b3 + 2b4 12 b3EW b4EW
a4 + r a6c + a10
T
Rb

u

a1 + a4u + r a6u + a10
+ r a8u

1
+ rA b1 + b3 + 2b4 12 b3EW b4EW
(24)
,
and rA = B /A . In this case the electroweak penguin terms are very small, because
they are not CKM enhanced with respect to the tree contribution.
This concludes the discussion of parameterizations of the decay amplitudes. The following section is devoted to a detailed description of the QCD factorization formalism, the calculation of the parameters ai for the various nonleptonic decay amplitudes, and an estimation of the annihilation parameters bi . The reader mainly interested in the phenomenological applications of our results can proceed directly to Section 4, where we present numerical results for these parameters, which will be used in the subsequent analysis in Section 5.
3. QCD factorization in B K decays
Based on the underlying physical principle of colour transparency (see [2426] for
early discussions in the context of decays to heavylight final states), supported by a
detailed diagrammatic analysis of infrared cancellations at leading power in the heavyquark expansion, we have shown in previous work that the complexity of the hadronic
matrix elements governing energetic, two-body hadronic decays of B mesons simplifies
greatly in the heavy-quark limit mb
QCD [14,15]. Consider B K decays as an
example. To leading power in QCD /mb , but to all orders in perturbation theory, the
matrix elements of the local operators Qi in the effective weak Hamiltonian in (1) obey
the factorization formula
I
I
K|Qi |B = F0B TK,i
fK K + F0BK T,i
f
+ TiII fB B fK K f ,
(25)
where M are leading-twist light-cone distribution amplitudes, and the -products imply
an integration over the light-cone momentum fractions of the constituent quarks inside
the mesons. A graphical representation of this result is shown in Fig. 1. Because the
energetic, collinear light-quark pair that ultimately evolves into the emission particle
at the upper vertex is created by a point-like source, soft gluon exchange between
this pair and the other quarks in the decay is power suppressed in the heavy-quark
limit (colour transparency). In other words, whereas the hadronic physics governing the
semileptonic B M1 transition and the formation of the emission particle M2 is genuinely
nonperturbative, nonfactorizable interactions connecting the two systems are dominated
by hard gluon exchange.
I
The hard-scattering kernels TiI,II in (25) are calculable in perturbation theory. TM,i
starts at tree level and, at higher order in s , contains nonfactorizable corrections from
hard gluon exchange or light-quark loops (penguin topologies). Hard, nonfactorizable
255
Fig. 1. Graphical representation of the factorization formula. Only one of the two form-factor terms
in (25) is shown for simplicity.
I (first two rows) and T II (last row). In

Fig. 2. Order s corrections to the hard-scattering kernels TM,i
i
I , the spectator quark does not participate in the hard interaction and is not drawn. The
the case of TM,i
two lines directed upwards represent the quarks that make up one of the light mesons (the emission
particle) in the final state.
interactions involving the spectator quark are part of TiII . The relevant Feynman diagrams
contributing to these kernels at next-to-leading are shown in Fig. 2. Although individually
these graphs contain infrared-sensitive regions at leading power, all soft and collinear
divergences cancel in their sum, thus yielding a calculable short-distance contribution.
Annihilation topologies are not included in (25) and Fig. 2, because they do not contribute
at leading order in QCD /mb . These power-suppressed contributions will be discussed
separately in Section 3.5.
We stress that the factorization formula does not imply that hadronic B decays
are perturbative in nature. Dominant soft contributions to the decay amplitudes exist,
which cannot be controlled in a hard-scattering approach. However, at leading power all
these nonperturbative effects are contained in the semileptonic form factors and lightcone distribution amplitudes. Once these quantities are given, the nonleptonic decay
amplitudes can be derived using perturbative approximations to the hard-scattering kernels.
This allows us to compute perturbative corrections to naive factorization estimates of
nonleptonic amplitudes, which is crucial for obtaining results that are independent of the
renormalization scheme adopted in the calculation of the effective weak Hamiltonian.
256
The hard-scattering kernels also contain imaginary parts, which determine the strong
rescattering phases of the decay amplitudes. At leading power in QCD /mb these
imaginary parts are of perturbative origin.
In the remainder of this section, we discuss in detail the various issues to be addressed
in the evaluation of the factorization formula. In Section 3.1, a modified renormalizationgroup treatment of electroweak penguin effects is introduced, which is more appropriate
than the standard scheme as far as applications to rare hadronic B decays are concerned.
Each of the diagrams in Fig. 2 contains a leading-power contribution relevant to (25)
and power-suppressed terms, which do not factorize in general. An important class
of such power-suppressed effects is related to certain higher-twist meson distribution
amplitudes. These amplitudes are defined in Section 3.2, and their leading, chirallyenhanced contributions to the nonleptonic decay amplitudes are evaluated. Section 3.4
contains a compendium of the relevant formulae for the calculation of the parameters ai .
Section 3.5 is devoted to annihilation topologies and the definition of the parameters bi .
3.1. Wilson coefficients of electroweak penguin operators
In the conventional treatment of the effective weak Hamiltonian (1), the initial conditions
for the electroweak penguin coefficients at the scale = MW are considered a next-toleading order effect, because they are proportional to the electroweak gauge coupling
(see [27,28] for a detailed discussion). For our purposes, however, it is preferable
to deviate from this standard power counting and introduce a modified approximation
scheme. The reason is that the electroweak penguin contributions in B K decays
and other rare processes based on b s qq
transitions are important only because they
compete with strongly CKM-suppressed tree topologies. Electroweak penguins and tree
topologies together are responsible for the isospin-violating contributions to the decay
amplitudes [29]. Therefore, their effects are important even though they are suppressed
with respect to the leading QCD penguin amplitude, which conserves isospin. The ratio
of electroweak penguin to tree amplitudes scales like /2 1, where = 0.22 is the
Wolfenstein parameter. Moreover, the dominant electroweak penguin effects are enhanced
by a factor of (mt /MW )2 and 1/ sin2 W . Hence it is not appropriate to count as a
small parameter in the renormalization-group evolution, if the effect of interest is related
to isospin breaking.
We now describe a systematic modification of the usual leading and next-to-leading
approximations, in which the dominant part of the electroweak penguin coefficients at the
scale = MW is treated as a leading-order effect. It is then consistent to include the QCD
radiative corrections to the enhanced terms in the initial conditions for the electroweak
penguin coefficients and, at the same time, the corrections of order s to the matrix
elements of the electroweak penguin operators, which represent the next-to-leading order
corrections to the hard-scattering kernels in the factorization formula.
Using a compact matrix notation, the solution to the renormalization-group equation for
the Wilson coefficients C1 , . . . , C10 in (1) can be written as
257

s ()
s (MW )
4

W ).
J U0
U 0J +
R 0 + R 1 C(M
C()
= U0 +
4
4
4 s ()
(26)
The matrices U 0 , J , R 0 and R 1 depend on the ratio s ()/s (MW ) and on the anomalous
dimensions and -function. At leading order, the evolution matrix reduces to U 0 +
(/s )R 0 . The remaining terms shown above are the next-to-leading corrections.
W ) at the weak scale as
We now expand the coefficients C(M

s (MW ) (1)
(0) s (MW ) (1) (0)
(0)

C(MW ) = Cs +
(27)
Cs +
Ce +
Ce + Re ,
4
4
4
where superscripts indicate the order in the strong coupling constant s (MW ). The term
proportional to represents the electroweak contribution originating from photon-penguin,
Z-penguin and box diagrams. We split this term into a contribution C e containing all terms
enhanced by the large top-quark mass and/or a factor of 1/ sin2 W , and a remainder Re . As
explained above, we treat Ce as a leading effect and hence include the first two terms in its
expansion in powers of s (MW ). The remainder Re is considered a next-to-leading effect,
and so we only keep the first term in its perturbative expansion. Explicitly, the nonvanishing
contributions to the initial conditions in the electroweak sector (i = 7, 8, 9, 10) are
(0)
Ce,7 =
xt
,
3
(0)
Ce,9 =

2
xt
+
10B0 (xt ) 4C0 (xt ) ,
2
3
3 sin W
(28)
and
8
2
xt
(0)
(0)
,
Re,7 = Re,9 = C0 (xt ) + D
0 (xt )
3
3
3
(29)
2 with m = m
where xt = m2t /MW
t (mt ). The InamiLim functions B0 (x), C0 (x) and
t

D0 (x) can be found, e.g., in [30]. Numerically, the remainder Re(0) is indeed much
smaller than Ce(0) , justifying our approximation scheme. Note that Ce(0) is gauge and
(0)
renormalization-scheme independent. The remainder Re is gauge-independent, but it
carries the usual next-to-leading order scheme dependence of the electroweak coefficients.
(1)
Explicit expressions for the QCD corrections contributing to C e have been obtained in
[31]. Using these results, we obtain the approximate expressions (valid for a high-energy
matching scale W = MW )
(1)
Ce,7
29.60xt1.142 + 28.52xt1.148,
(1)
Ce,9 571.62xt0.580 + 566.40xt0.590,
(1)
Ce,8
0.94xt0.661,
(1)
Ce,10 5.51xt1.107.
(30)
In the conventional treatment C e(1) would be absent, while the sum (C e(0) + Re(0) ) would be
the usual initial condition for the electroweak coefficients at the scale = MW , counted as
a next-to-leading order effect.
In addition to the modified counting scheme for powers of coupling constants, we make
a further, numerically excellent approximation, which greatly simplifies the systematic
evaluation of the Wilson coefficients. In essence, it amounts to neglecting QED effects
in the calculation of the Wilson coefficients C1 , . . . , C6 . The values of these coefficients
258
are obtained using the standard next-to-leading order approximation including only stronginteraction effects. At the same time, we neglect QED corrections to the matrix elements of
the operators Q1 , . . . , Q6 . (In fact, the virtual corrections of order are infrared divergent
and require the inclusion of photon bremsstrahlung contributions in order to obtain physical
results. Our approximation scheme avoids this complication.) This treatment can be
justified by noting that QED and electroweak contributions to the decay amplitudes in
(18) are only important if they contribute to the parameters qei and qC eiC . From (20),
it follows that the terms of order contained in the coefficients a7,...,10 are enhanced by
the prefactor 1/ KM . It is thus sufficient for all practical purposes to only include the order
corrections from the coefficients a7,...,10 . On the contrary, QED corrections to the other
amplitude parameters can be safely neglected (i.e., the coefficients a1,...,6 do not contain
terms proportional to ). Precisely this is achieved by our approximation scheme.
At the technical level, the approximation described above can be explained in terms of
the 10 10 anomalous-dimension matrix for the operators Q1 , . . . , Q10 , written in block
form as

A66 B 64
.
=
(31)
C 46 D 44
We set C = 0, thereby neglecting the mixing of the electroweak penguin operators into the
operators Q1 , . . . , Q6 , and ignore contributions of order to A. At the same time, we drop
the terms of order in the matching conditions (27) for C1 , . . . , C6 . We also omit terms of
order in D, which would yield second-order corrections in . For the matrix B, we use
the complete next-to-leading order result including terms of order .
Numerical results for the Wilson coefficients obtained at leading and next-to-leading
order in our modified approximation scheme are given in Table 1. Throughout this work,
we use the naive dimensional regularization (NDR) scheme with anticommuting 5 , as
defined in [27]. The matrix elements of the dipole operators Q7 and Q8g enter the decay
amplitudes only at next-to-leading order. Consequently, the standard leading-logarithmic
approximation is sufficient for the coefficients C7 and C8g . In practice, it is advantageous
to work with so-called effective coefficients, which in the NDR scheme are defined
1
eff = C
eff
as C7
7 3 C5 C6 and C8g = C8g + C5 . In the numerical analysis of (26) we
consistently drop all terms of higher than next-to-leading order according to our modified
counting scheme. Throughout we use the two-loop expression for the running coupling
s () evaluated with nf = 5 light quark flavours.
3.2. Meson distribution amplitudes and twist-3 projections
Referring to the factorization formula shown graphically in Fig. 1, we denote by x
the longitudinal momentum fraction of the constituent quark in the emission meson M2
(the meson at the upper vertex), and by y the momentum fraction of the quark in

the meson M1 . For a B-meson
decaying into two light mesons, we define light-cone
distribution amplitudes by choosing the + direction along the decay path of the light
emission particle and denote by the light-cone momentum fraction of the light spectator
antiquark. A massive pseudoscalar meson has two leading-twist light-cone distribution
259
Table 1
Wilson coefficients Ci in the NDR scheme and based on our modified approximation scheme (see
(5)
text). Input parameters are
= 0.225 GeV, mt (mt ) = 167 GeV, mb (mb ) = 4.2 GeV, MW =
MS
80.4 GeV, = 1/129, and sin2 W = 0.23

NLO
= mb /2
= mb
= 2mb
C1
C2
C3
C4
C5
C6
1.137
1.081
1.045
0.295
0.190
0.113
0.021
0.014
0.009
0.051
0.036
0.025
0.010
0.009
0.007
0.065
0.042
0.027
eff
C7
eff
C8g
C7 /
= mb /2
= mb
= 2mb
LO
= mb /2
= mb
= 2mb
C9 /
C10 /
0.024
0.011
0.011
0.096
0.060
0.039
1.325
1.254
1.195
0.331
0.223
0.144
C1
C2
C3
C4
C5
C6
1.185
1.117
1.074
0.387
0.268
0.181
0.018
0.012
0.008
0.038
0.027
0.019
0.010
0.008
0.006
0.053
0.034
0.022
eff
C7
eff
C8g
0.364
0.318
0.281
0.169
0.151
0.136
C7 /
= mb /2
= mb
= 2mb
C8 /
0.012
0.001
0.018
C8 /
0.045
0.029
0.019
C9 /
1.358
1.276
1.212
C10 /
0.418
0.288
0.193
amplitudes [15,32], but only one of them enters our results. This amplitude is called
B ( ) and coincides with the function B1 ( ) defined in Section 2.3.3 of [15]. The meson
distribution amplitudes are normalized to 1 once the decay constants are factored out as in
(25). For a light meson, we define the leading-twist amplitude (x) in the usual way and
assume that (x) = O(1) if both x and (1 x) are of order unity, and (x) = O(x) for
x 0 (and similarly for x 1). For the B meson, almost all momentum is carried by the
heavy quark, and hence B ( ) = O(mb /QCD ) and = O(QCD /mb ).
Higher-twist light-cone distribution amplitudes for the light mesons give power-suppressed contributions in the heavy-quark limit. However, as has been explained in
Section 2, these can sometimes be large if they appear in conjunction with the chiral
enhancement factors rM () defined in (5) and (6). The corresponding terms are associated
with twist-3 quarkantiquark distribution amplitudes and can be identified completely.
Since the calculation of these contributions in momentum space is less straightforward
than for the leading-twist contributions, we summarize the relevant projection operators
below, following the discussion in [33].
The relevant definitions of the light-cone distribution amplitudes of a light pseudoscalar
meson P in terms of bilocal operator matrix elements are [34]
260
1
P (p)|q(z
2 ) 5 q(z1 )|0 = ifP p
1)
dx ei(xpz2+xpz
(x),
1
P (p)|q(z
2 )i5 q(z1 )|0 = fP P
1)
dx ei(xpz2+xpz
p (x),
1
(x)
,
6
0
(32)
where fP is the decay constant, and we have defined z = z2 z1 and x = 1 x. The
parameter P = m2P /(m1 + m2 ), where m1,2 are the current quark masses of the meson
constituents, is proportional to the chiral quark condensate. (This definition does not hold
for the 0 meson, in which case 0 = m2 /(mu + md ) as for the charged pions [15].)
(x) is the leading-twist (twist-2) distribution amplitude, whereas p (x) and (x) have
subleading twist (twist-3). All three distribution amplitudes are normalized to 1, as follows
by taking the limit z1 z2 . The above definitions can be combined into the matrix
P (p)|q(z
2 ) 5 q(z1 )|0 = ifP P (p z p z )
1)
dx ei(xpz2 +xpz
P (p)|q (z2 )q (z1 )|0

ifP
=
4
1
0
1)
dx ei(xpz2+xpz

(x)
p
/ 5 (x) P 5 p (x) p z
.
6
(33)
We implicitly assume that the bilocal matrix elements are supplied with the appropriate
path-ordered exponentials of gluon fields so as to make the definitions of the lightcone distribution amplitudes gauge invariant. (These exponentials are absent in lightcone gauge.) The distribution amplitudes depend on the renormalization scale . This
scale dependence is compensated by higher-order corrections to the hard-scattering
kernels, which however are beyond the accuracy of the present calculation. We thus
suppress the argument in the distribution amplitudes, because it is irrelevant to our
discussion.
To obtain the corresponding projector of the quarkantiquark amplitude in momentum
space, the transverse components of the coordinate z must be taken into account. The
collinear approximation can be taken only after the projection has been applied. We
therefore assign momenta
k1 = xp + k +
k 2
p ,
2xp p
k2 = xp
k +
k 2
p ,
2xp
p
(34)
to the quark and antiquark in the light meson, where p is a vector whose 3-components
point into the opposite direction of p.
Then the exponential in (33) becomes ei(k1 z2 +k2 z1 ) .
(Meson mass effects are neglected, so that p and p can be considered as light-like.) The
Note that
transverse components k are defined with respect to the vectors p and p.
261
k12 = k22 = 0. In general, the projector (33) is part of a diagram expressed in configuration
space. Transformation to momentum space is achieved by performing the integrations over
zi , which reduce to momentum-space -functions after substituting

+
z (i)
(35)
= (i)
+ .
k1
p p x k
The ellipses denote a term proportional to p , which does not contribute to the result. We
also omit terms of order k 2 , which cannot contribute in the limit k 0. As written above,
the derivative acts on the hard-scattering amplitude in the momentum-space representation.
Using an integration by parts, the derivative with respect to x can be made to act on the
light-cone distribution amplitude. The second term, which involves the derivative with
respect to the transverse momentum, must be evaluated before the collinear limit k1
is taken. The light-cone projection operator of a light pseudoscalar meson in
xp, k2 xp
momentum space, including twist-3 two-particle contributions, then reads

p p (x)
ifP
P
M =
p
/ 5 (x) P 5 p (x) i
4
p p
6

(x)
(36)
+ i p
.
6 k
It is understood that, after the derivative is taken, the momenta k1 and k2 are set equal to
xp and xp,
respectively. A complete description of the pseudoscalar meson at the twist-3
level would also include three-particle quarkantiquarkgluon contributions (see [34] for
a detailed discussion), which do not involve the large normalization factor P and thus are
omitted here. (Note that the overall sign of (36), as well as of (41) below, depends on the
convention of ordering the quark fields in (33).)
The asymptotic limit of the leading-twist distribution amplitude, valid for ,
is (x) = 6x(1 x). For finite value of the renormalization scale, it is convenient and
conventional to employ an expansion in Gegenbauer polynomials of the form

(3/2)
nM ()Cn (2x 1) .
M (x, ) = 6x(1 x) 1 +
(37)
n=1
In numerical evaluations it will be sufficient to truncate this expansion at n = 2, using

(3/2)
(3/2)
C1 (u) = 3u and C2 (u) = 32 (5u2 1). The Gegenbauer moments nM () are
multiplicatively renormalized. The scale dependence of these coefficients enters our results
only at order s2 , which is beyond the accuracy of a next-to-leading order calculation.
The twist-3 two-particle distribution amplitudes are determined by the three-particle distributions via the equations of motion, except for a single term [34]. In the approximation
adopted here, where only terms proportional to P are kept and all three-particle distributions are neglected, the twist-3 amplitudes must obey the equations of motion

(x)
1x
(x)
(x)
(x)
x
p (x) +
=
,
p (x)
=
.
(38)
2
6
6
2
6
6
These equations enforce that we must use the asymptotic forms p (x) = 1 and (x) =
6x(1 x). It will be important below that p (x) and (x) do not vanish at the endpoints
262
x = 0 or 1. We finally observe that the k -derivative in (36) can be substituted by

2k
2 .
k
k
(39)
This is because we may expand the amplitude to first order in k (higher powers do not
contribute in the k 0 limit), and use
k
2k
k =
= g
,
2
k
k
(40)
which holds after averaging k over the azimuthal angle (denoted by . . .). We use the
= diag(0, 1, 1, 0). The twist-3 terms in (36) can now be combined into
definition g
the projector [35]
k/2 k/1
ifP P
5
p (x),
4
k2 k1
(41)
where k1,2 are the quark and antiquark momenta defined in (34), and the factor of
p (x) = 1 is simply there to remind us that this is a twist-3 projection. In our analysis
below, we will quote the results of the twist-3 projections in this form, i.e., after eliminating
()
(x) using the equations of motion. Expressions in terms of two functions p (x) and
(x) are ambiguous, but reduce to the same expression upon substituting the asymptotic
forms of the distribution amplitudes.
3.3. Comments on the calculation
We now describe some technical aspects of the calculation of the various diagrams in
more detail. The complete results for the parameters ai will be given in Section 3.4.
Vertex corrections
The calculation of the four one-loop vertex diagrams (first four diagrams in Fig. 2)
involves the trace

(2k2 + /C )
(2k1 + /C)
,
tr M M2
(42)
(xq + C)2
(xq
+ C)2
where M M2 is the projector from (36), q the momentum of the meson M2 , k1 (k2 ) the
momentum of the quark (antiquark) in this meson, and C the momentum of the gluon.
We have used that M M2 k/ 1 = k/ 2 M M2 = 0 by the equations of motion, and that we can put
in the denominator. A Fierz transformation may be necessary in order
k1 = xq and k2 = xq
to arrive at the trace (42). The possible structures are therefore V A (contributing to
a1,...,4,9,10 ), V + A (contributing to a5,7 ), and S + P (contributing to a6,8 ).
If = V A, only the leading-twist light-cone distribution amplitude (x) in (36)
contributes under the trace. We then recover the results of [14,15] and find that there
exist no chirally-enhanced power corrections to a1,...,5,7,9,10 from the vertex diagrams. If
= S + P , the leading-twist contribution to the trace vanishes, while the twist-3 result is
proportional to rM2 = 2M2 /mb . This gives an order s correction to the coefficients a6,8 ,
263
Fig. 3. The two different penguin contractions.
which are multiplied by rM2 already in naive factorization. We can now exploit the fact
that tr = 0 to show that the term involving the transverse-momentum derivative in (36)
does not contribute to the trace. At this stage, the collinear limit can be taken to compute
the kernel in the usual way. The projection on p (x) yields a result containing symmetric
and antisymmetric parts under the exchange of x and (1 x). As explained above, in
the approximation of keeping only chirally-enhanced terms we are forced to assume the
asymptotic form for p (x), so that the antisymmetric part of the kernel integrates to zero.
The symmetric part turns out to be a scheme-dependent constant and is responsible for
the 6 in the expressions for a6,8 in (46) below. The kernel resulting from the (x)
projection is symmetric under x (1 x) and thus vanishes after integration with (x).
Penguin diagrams
We now consider the penguin contractions (fifth diagram in Fig. 2), restricting our
attention first to the twist-2 part of the projector (36). The corresponding contributions
to the hard-scattering kernels have been given in our previous work [14]. Because there
have been conflicting results for the penguin terms in the recent literature (see Section 6),
we wish to clarify the origin of the discrepancies here.
The point to note is that, depending on the structure of the (V A) (V A) four
fermion operators in the effective weak Hamiltonian (without Fierz transformation!), there
exist two distinct penguin contractions with different contractions of spinor indices, as
indicated in Fig. 3. Although the two diagrams are related by Fierz transformations in four
dimensions, in dimensional regularization they give different results in the NDR scheme
with anticommuting 5 , because this scheme does not preserve the Fierz identities in d
dimensions [27]. The results for the two contractions shown in the figure involve

m2
2
2 m2b
ln 2 G(s),
right:
left:
(43)
ln 2b + 1 G(s),
3
3
where the function G(s) is given in (50) below. The first contraction appears in matrix
elements of the operators Q4,6 , while the second one enters in matrix elements of Q1,3 .
This assumes the standard Fierz-form of the effective Hamiltonian, which is employed in
the calculation of the Wilson coefficients in the NDR scheme. The operator Q5 is special,
because its contribution is a pure ultraviolet effect which, by definition, is absorbed into
eff = C + C of the chromomagnetic
the definition of the effective Wilson coefficient C8g
8g
5
dipole operator [30]. The discrepancies between our results and some of the papers
discussed in Section 6 seem to arise from the fact that Q4 and Q6 are treated like Q1,3 .
Including now the twist-3 part of the projector (36), we still find that the penguin
contractions can be straightforwardly evaluated in terms of four-quark operators, before
the actual projection is made. The on-shell conditions for the external quarks connected to
264
the gluon need to be used in this step. The projection is then very simple and only the 5
projector multiplied by p (x) contributes. The resulting kernel is identical to that obtained
with the twist-2 projection.
The calculation of the matrix element of the chromomagnetic dipole operator (sixth
diagram in Fig. 2) is more interesting in this respect. In this case it is no longer possible to
reduce the amplitude to the usual structures involving four-quark operators. This is related
to the fact that at the twist-3 level the k momenta cannot immediately be put to zero. The
complete twist-3 projection (36) has to be performed to evaluate the diagram, which leads
to the expression
1
u q (pq ) M M2 (P
/ P/ )(1 + 5 )ub (pb ),
(1 x)m2b
(44)
where P = pq + k2 = pb k1 denotes the momentum of the gluon, and the prefactor comes
from the gluon propagator. M M2 projects on the emission particle M2 with momentum q,
and pb , pq denote the momenta of the b quark and the quark in M1 , respectively. We find
that all four terms of the projector contribute when evaluated on this expression, giving the
result

(x)
ifP u q (pq ) (1 + 5 )
x

3
1 (x) 1 (x)
P
ub (pb )
+
(1 5 ) p (x) +
+
mb
2
2 6
x 6

(x) 2P
+
(1 5 )p (x) ub (pb ).
= ifP u q (pq ) (1 + 5 )
(45)
x
mb
The second line is obtained after using the equations of motion (38) for the twist-3
distribution amplitudes, so that the asymptotic form p (x) = 1 is understood. We observe
that the factor 1/x from the gluon propagator is cancelled in the twist-3 term, and so the
convolution integral has no endpoint divergence.
One can see from (45) that the matrix element of the chromomagnetic operator is
obtained incorrectly at the twist-3 level if the incomplete projector containing only p (x)
is used. Note that, in general, this leads to gauge-dependent results, since the equations of
motion are not respected.
Hard spectator interaction
The calculation of the two diagrams in the third row of Fig. 2 leads to the same trace as
in (42). It therefore follows that, for = S + P , the projection on M2 can result at most in
a constant multiplying the distribution amplitude p (x). However, the constant obtained
for the vertex diagrams resulted entirely from a term of order = (2 d/2) in the trace
multiplying the ultraviolet-divergent loop diagram. Since the hard spectator contributions
result from tree diagrams, the trace can be evaluated in four dimensions, and this constant
is absent. We thus conclude that there is no hard spectator correction to the parameters a6,8
at order s .
If = V A, only the twist-2 distribution amplitude contributes for the emission
particle M2 , but all four terms in the projector for M1 contribute to the result. We also
265
find that both terms in the B-meson projection (as given in [15]) contribute, but one
of the two B-meson light-cone distribution amplitudes drops out after implementing the
equations of motion (38). Contrary to the other corrections, the kernels TiII resulting from
hard spectator interactions have logarithmic endpoint singularities at twist-3 level. They
1
where y is the momentum of the antiquark in the
arise from integrals of the form 0 dy/y,
meson that picks up the spectator antiquark from the B meson. These endpoint singularities
prevent a reliable perturbative calculation of the chirally-enhanced power corrections to the
hard spectator interactions. (The endpoint singularities are missed if the incomplete lightmeson projector with only the 5 projection at twist-3 is employed.)
3.4. Results for the parameters ai
After these preliminaries, we now present the results for the coefficients ai obtained
at next-to-leading order in s , and including the complete set of chirally-enhanced power
corrections to the heavy-quark limit. We focus on the case of B K decays, but similar
results (with obvious substitutions) hold for all other B decays into two flavour-nonsinglet
pseudoscalar mesons. For later convenience, every coefficient ai (K) is split into two
terms: ai (K) = ai,I (K) + ai,II (K). The first term contains the naive factorization
contribution and the sum of vertex and penguin corrections (the form-factor terms in the
factorization formula (25)), while the second one arises from the hard spectator interactions
(the hard-scattering term in the factorization formula). Weak annihilation effects are not
included here; they will be discussed separately in Section 3.5. The calculation of the
kernels described above results in

CF s
C2
C2 CF s
VK ,
1+
a1,II =
HK ,
a1,I = C1 +
Nc
4
Nc Nc

CF s
C1
C1 CF s
V ,
a2,I = C2 +
1+
a2,II =
HK ,
Nc
4
Nc Nc

C4
C4 CF s
CF s
HK ,
V ,
a3,I = C3 +
1+
a3,II =
Nc
4
Nc Nc

p
C3
C3 CF s
CF s
CF s PK,2
p
VK +
a4,I = C4 +
, a4,II =
HK ,
1+
Nc
4
4 Nc
Nc Nc

C6
C6 CF s
CF s

V ,
a5,I = C5 +
HK
,
1+
a5,II =
Nc
4
Nc Nc

p
CF s PK,3
CF s
C5
p
a6,I = C6 +
+
16
, a6,II = 0,
Nc
4
4 Nc

CF s
C8
C8 CF s

V ,
a7,I = C7 +
1+
a7,II =
HK
,
Nc
4
Nc Nc

p,EW
C7
CF s
PK,3
p
a8,I = C8 +
, a8,II = 0,
16
+
Nc
4
9 Nc

C10
C10 CF s
CF s
V ,
a9,I = C9 +
HK ,
1+
a9,II =
Nc
4
Nc Nc
266
a10,I = C10 +

p,EW
CF s
PK,2
C9
C9 CF s
VK +
1+
, a10,II =
HK ,
Nc
4
9 Nc
Nc Nc
(46)
where Ci Ci (), s s (), CF = (Nc2 1)/(2Nc ), and Nc = 3. The quantities

p
p
p,EW
p,EW
()
()
VM , HM2 M1 , PK,2 , PK,3 , PK,2 , and PK,3 are hadronic parameters that contain all
nonperturbative dynamics. These quantities consist of convolutions of hard-scattering
()
result from the
kernels with meson distribution amplitudes. Specifically, the terms VM
p
p
p,EW
p,EW
vertex corrections (first four diagrams in Fig. 2), PK,2 and PK,3 (PK,2 and PK,3 )
arise from QCD (electroweak) penguin contractions and the contributions from the dipole
()
are due to hard gluon exchange
operators (fifth and sixth diagrams in Fig. 2), and HM
2 M1
involving the spectator quark in the B meson (last two diagrams in Fig. 2). For the penguin
terms, the subscript 2 or 3 indicates the twist of the corresponding projection.
In the numerical evaluation of these expressions we consistently drop higher-order
terms in the products of the Wilson coefficients with the next-to-leading order corrections.
Also, in the computation of the O() corrections we confine ourselves to the penguin
contractions of the currentcurrent operators Q1 and Q2 and to the contribution of
the electromagnetic dipole operator, which have the largest Wilson coefficients. These
p,EW
p,EW
contributions, which are contained in the quantities PK,2 and PK,3 , suffice to cancel
the renormalization-scheme dependence of the electroweak penguin coefficients C7,...,10 at
next-to-leading order. Additional O() corrections proportional to the Wilson coefficients
C3,...,6 of the QCD penguin operators exist, but because these coefficients are very small
their effects can be safely neglected.
Vertex and penguin contributions
We now collect the relevant formulae needed for the calculation of the coefficients ai,I .
The vertex corrections are given by (M = , K)
VM
mb
18 +
= 12 ln
1
dx g(x)M (x),
0

= 12 ln
VM
mb
6+
1
dx g(1 x)M (x),
0

1 2x
ln x i
1x

2 ln x
+ 2 Li2 (x) ln2 x +
(3 + 2i) ln x (x 1 x) ,
1x
g(x) = 3
(47)
where Li2 (x) is the dilogarithm. The constants 18 and 6 are specific to the NDR scheme.
0
The function g(x) can be obtained from the corresponding function relevant to, e.g., B
D + K decays [15] by taking the limit mc 0. (Recently, a partial two-loop result for the
vertex corrections has been obtained in [36], where the terms of order 0 s2 were calculated
analytically. To be consistent with the next-to-leading order analysis of this paper, we will
not make use of this result.)
267
If the leading-twist light-cone distribution amplitudes are expanded in Gegenbauer

polynomials as shown in (37), the relevant convolution integral can be evaluated
analytically, giving
1

11
1
21
3i 1M 2M + .
dx g(x)M (x) = 3i +
2
2
20
(48)
The integral with g(x) g(1 x) is obtained by changing the sign of the odd Gegenbauer
coefficients. Since the pion distribution amplitude (x) is symmetric under the exchange
x (1 x), it follows that V = V + 12.
Next, the penguin contributions are

4 mb 2
8 mb 4
p
+ GK (sp ) + C3 ln
+ GK (0) GK (1)
PK,2 = C1 ln
3
3
3
3

4nf mb
ln
(nf 2)GK (0) GK (sc ) GK (1)
+ (C4 + C6 )
3
1
eff
2C8g
0
dx
K (x),
1x

p,EW
PK,2

1
dx
4 mb 2
eff
+ GK (sp ) 3C7
K (x),
= (C1 + Nc C2 ) ln
3
3
1x
(49)
where nf = 5 is the number of light quark flavours, and su = 0, sc = (mc /mb )2 are mass
ratios involved in the evaluation of the penguin diagrams. Small electroweak corrections
p
from C7 , . . . , C10 are consistently neglected in PK,2 within our approximations. In
p,EW
principle, in PK,2 also contributions from C3 , . . . , C6 appear. Their impact is extremely

small numerically and we drop them for simplicity. Similar comments apply to (54) below.
The function GK (s) is given by
1
GK (s) =
dx G(s i , 1 x)K (x),
(50)
1
G(s, x) = 4

du u(1 u) ln s u(1 u)x

2(12s + 5x 3x ln s) 4 4s x (2s + x)
x
.
arctan
9x
4s x
3x 3/2
Its expansion in terms of Gegenbauer moments reads

GK (sc ) =

K
K
5 2
4
ln sc + 1 + 2 + 8 + 91K + 92K sc
3 3
2
5
3

+ 2 8 + 631K + 2142K sc2 24 91K + 802K sc3 + 28802K sc4
(51)
268

2
1 4sc 1 + 2sc + 6 4 + 271K + 782K sc2
3

36 91K + 702K sc3 + 43202K sc4 (2 arctanh 1 4sc i)

4
+ 12sc2 1 + 31K + 62K 1 + 91K + 362K sc
3

K
+ 18 1 + 102K sc2 2402K sc3 (2 arctanh 1 4sc i)2 + ,
K
5 2i 1K
+
+
+ 2 + ,
3
3
2
5

2
155
85
4
GK (1) =
6 3 +
36 3 + 12 2 1K
3
9
2

7001
504 3 + 136 2 2K + .
+
5
GK (0) =
(52)
The contributions of the dipole operators in (49) involve the integral

1
0

dx
M (x) = 3 1 + 1M + 2M + .
1x
(53)
The twist-3 terms from the penguin diagrams are related to the twist-2 terms by the
eff and C eff ,
simple replacement K (x) pK (x) = 1. For the terms proportional to C7
8g
however, the twist-3 projection yields an additional factor of (1 x), which cancels the
denominator in (53). We therefore find

4 mb 2
8 mb 4
p
K (1)
+ GK (sp ) + C3 ln
+ GK (0) G
PK,3 = C1 ln
3
3
3
3

4nf mb
K (0) G
K (sc ) G
K (1) 2C eff ,
+ (C4 + C6 )
ln
(nf 2)G
8g
3

4 mb 2
p,EW
eff
+ GK (sp ) 3C7
,
PK,3 = (C1 + Nc C2 ) ln
(54)
3
3
with
K (s) =
G
1
dx G(s i , 1 x)pK (x).
(55)
Inserting
pK (x) = 1,
this integral is evaluated to

K (sc ) = 16 (1 3sc ) 2 ln sc + (1 4sc )3/2 2 arctanh 1 4sc i ,

G
9
3
2
32
K (1) =
K (0) = 16 + 2 i,
G
G
.
9
3
9
3
(56)
The typical parton off-shellness in the loop diagrams contributing to the vertex and penguin
contributions to the hard-scattering kernels is of order mb , and hence it is appropriate to
269
choose a value mb for the renormalization scale in the Wilson coefficients Ci and in the
kernels TiI when evaluating the quantities ai,I . The -dependent terms in expressions (47),
(49) and (54) cancel the renormalization-scale dependence of the Wilson coefficients Ci ()
at next-to-leading order. Similarly, the constants accompanying the various logarithms
in these expressions are renormalization-scheme dependent and combine with schemedependent constants in the expressions for the Wilson coefficients to give renormalizationgroup invariant results.
The coefficients ai,I contain strong-interaction phases via the imaginary parts of the
functions g(x) and G(s, x). At next-to-leading order and to leading power in QCD /mb ,
these phases yield the asymptotic contributions to the final-state rescattering phases of the
B K decay amplitudes. The presence of a strong-interaction phase in the penguin
function G(s, x) is well known and commonly referred to as the BanderSilvermanSoni
mechanism [37]. It was included in many phenomenological investigations of nonleptonic
B decays; however, there was always an argument as to how to choose the gluon
momentum, kg2 , in the fifth diagram in Fig. 2. In our approach there is no ambiguity to this
choice, because the distribution of kg2 = (1 x)m2b is determined by the kaon distribution
amplitudes. The imaginary part of the function g(x) is another source of rescattering
phases, which arises from hard gluon exchanges between the two outgoing mesons [14,
15]. The appearance of this phase is a new element of the QCD factorization approach.
Hard-scattering contributions
The hard spectator interactions shown in the last two diagrams in Fig. 2 give leadingtwist and chirally-enhanced twist-3 contributions to the kernels TiII . We include these hardscattering contributions as parts of the coefficients ai , although they are not related to
factorized matrix elements in the usual sense. Only the twist-2 terms are dominated by hard
gluon exchange and thus calculable. Nevertheless, for consistency with our treatment of
the penguin coefficients we should also include the chirally-enhanced terms of subleading
twist, which however have logarithmic endpoint singularities.
At tree level, the integrals over meson distribution amplitudes for the hard spectator
contributions factorize. We find (the momentum fractions x and y are defined in
Section 3.2)
HK
fB f
= 2 B
mB F0
(0)
1
d
B ( )
1
0
dx
K (x)
x
1
0

dy
2 x
p (y)
(y) +
y
mb x
1 1

fB f
x K y + r x 1 K XH
=
,
B
mB B F0
(0)
HK
fB fK
= 2 BK
mB F0
(0)
1
0
d
B ( )
1
0
dx
(x)
x
1
0

dy
2K x K
(y)
K (y) +
y
mb x p

K
1 1
fB fK
=
,
x y K + rK x 1 XH
BK
mB B F0
(0)
270

HK
fB fK
= 2 BK
mB F0
(0)
1
d
B ( )
1
0
dx
(x)
x
1
0

dy
2K x K
K (y) +
p (y)
y
mb x

K
1 1
fB fK
,
=
x y K + rK x 1 XH
BK
mB B F0
(0)
(57)
where we have defined the moments

1
d
mB
B ( )
,
1

dx x n M (x) x n M .
(58)
The quantity B parameterizes our ignorance about the B-meson distribution amplitude
[14]. Not much is known about this parameter except for the upper bound 3B 4
[38], where = mB mb is a scheme-dependent parameter. In practice, this just means
that B is expected to be less than 600 MeV or so. The ratios 2K /mb and 2 /mb
multiplying the twist-3 terms coincide with the parameters rK and r introduced in (5) and
(6), respectively. The twist-3 contribution involves the logarithmically divergent integral
(M = or K)
1
M
XH
dy
M (y) =
1y p
1
0
dy
.
1y
(59)
As previously we have to use the asymptotic form for pM (y), so no distinction between
pion and kaon is necessary. (Consequently, the superscript M will often be dropped
from now on.) The divergence results from the region where the spectator quark in the B
meson enters the light final-state meson at the lower vertex in Fig. 1 as a soft quark. The
twist-3 hard-scattering kernels do not provide sufficient endpoint suppression to render this
contribution subleading. In practice, the singularity will be smoothed out by soft physics
related to the intrinsic transverse momentum and off-shellness of the partons, which
unfortunately does not admit a perturbative treatment (see also the discussion in Section 6).
In particular, the resulting contribution may be complex due to soft rescattering in higher
orders. For the purpose of power counting, we note that the effect of transverse momentum
and off-shellness would be to modify (1 y) (1 y) + with = O(QCD /mb ) in the
M ln(m /
denominator in (59). We thus expect that XH
b
QCD ), however, with a potentially
complex coefficient.
Considering the off-shellness of the gluon in the last two diagrams in Fig. 2, it is natural
to associate a scale h (QCD mb )1/2 , rather than mb , with the hard-scattering
contributions of leading power. Hence, we set s = s (h ) (and also evaluate the Wilson
coefficients at the scale h ) for the twist-2 contributions to the quantities ai,II in (46).
Specifically, we use h = h with h = 0.5 GeV for the scale in the hard-scattering
diagrams. However, because of the endpoint divergence the contributions proportional
M must be considered as nonperturbative effects dominated by small-momentum
to XH
interactions. For this reason, we should more properly write
HK
271

s (s )r (s ) 1
1 1
fB f
x K XH ,
x K y +
=
s (h )
mB B F0B (0)
(60)
and similarly for the other two quantities in (57). Here s may be a soft scale. In other
words, in principle there is no reason to expect that the twist-3 contributions should
be governed by a perturbative coupling constant. However, it turns out that the product
s (s )r (s ) is almost renormalization-group invariant (it scales only as [s ()]1/25 with
Nc = 3 and four flavours), and therefore we may evaluate it at the scale h , so that the ratio
of running couplings in (60) equals 1.
The expressions for the hard-scattering contributions in (57) contain many poorly known
parameters. Besides the divergent quantity XH and the wave-function parameter B , they
depend on the B-meson decay constant and heavy-to-light form factors. Fortunately,
it turns out that ratios of the different hard-scattering contributions have very small
uncertainties. Using the symmetry of the pion distribution amplitude, we find that

= HK ,
HK
HK RK HK ,
(61)
where RK = AK /AK is the ratio of factorized matrix elements defined in (15).

As argued in Section 3.2, in the approximation where only chirally-enhanced power
corrections are included one is forced to employ the asymptotic forms of the twist3 distribution amplitudes p (x) and (x). It is then not necessary to keep other
nonasymptotic effects in the twist-3 terms. That is, we are free to replace the moments
of twist-2 amplitudes multiplying XH by their asymptotic values. In this approximation,
the twist-3 contributions in (57) reduce to a universal, multiplicative correction of the twist2 term, and we obtain the approximate form of the second relation in (61). In other words,
up to small SU(3) violations the main uncertainties in the description of the hard-scattering
terms combine into a single poorly-known quantity HK .
Finally, note that in (57) we have not assumed any symmetry properties of the twist-2
light-meson distribution amplitudes. Therefore, with obvious substitutions our results can
be applied directly to other decays, such as B and Bs K + K .
3.5. Weak annihilation contributions bi
Weak annihilation contributions to charmless hadronic B decays are power suppressed
in the heavy-quark limit and hence do not appear in the factorization formula (25).
Nevertheless, as emphasized in [17,39], these contributions may be numerically important
for realistic B-meson decays. Besides their power suppression, weak annihilation effects
differ from the hard spectator interactions discussed earlier in that they exhibit endpoint
singularities even at twist-2 order in the light-cone expansion for the final-state mesons, and
therefore cannot be computed self-consistently in the context of a hard-scattering approach.
In the following discussion, we will ignore the soft endpoint divergences and derive results
for the annihilation contributions in terms of convolutions of hard-scattering kernels
with light-cone distribution amplitudes, including again the chirally-enhanced twist-3
projections. Despite the fact that such a treatment is not entirely self-consistent, it is
nevertheless useful to estimate the importance of annihilation for particular final states.
272
Fig. 4. Annihilation diagrams.
At leading order in s , the annihilation kernels follow from the diagrams shown in Fig. 4.
They result in a further contribution to the hard-scattering term in the factorization formula,
in addition to the hard spectator scattering discussed in Section 3.2. It will be convenient
for future applications to keep the discussion of annihilation contributions general. We
thus consider a generic b-quark decay and use the convention that M2 contains a quark
from the weak decay vertex with longitudinal momentum fraction x, and M1 contains an
antiquark from the weak vertex with momentum fraction y.
The four-quark operators in the
effective weak Hamiltonian are Fierz-transformed into the form (q1 b)1 (q2 q3 )2 , such that

the quarks in the first bracket refer to the constituents of the B-meson.
If the colour indices
of this bracket are of the form q1i bi , diagrams (c) and (d) do not contribute, while diagrams
(a) and (b) give rise to a colour factor CF /Nc . If the colour indices are of the form q1i bj ,
then the colour factor of all four diagrams is CF /Nc2 . The projections onto the light-cone
distribution amplitudes are done in the same manner as for the hard spectator scattering
described in Section 3.2. At leading power (and assuming that x, y
) the integration
over the B-meson distribution amplitude is trivial and yields the B-meson decay constant,
since the kernels are independent. The remainders of the diagrams can be expressed in
terms of the following building blocks: (The leading-twist contribution to Ai1 has already
been discussed in [39].)
1
Ai1
= s

dx dy M2 (x)M1 (y)

1
1
4M1 M2 2
+ 2 +
,
y(1 x y)
x y
xy
m2b

dx dy M2 (x)M1 (y)

1
4M1 M2 2
1
+ 2 +
,
x(1
x y)
xy
xy
m2b
0
f
A1
= 0,
1
Ai2
= s
0
f
A2
= 0,

1
Ai3
= s
dx dy
0
f
A3
1
= s
dx dy
0

2M2
2M1
2y
2x
M2 (x)
M1 (y)
,
mb
xy(1
x y)
mb
xy(1
x y)

2M2
2M1
2(1 + x)
2(1 + y)
+
M2 (x)
(y)
.
M1
mb
x 2 y
mb
xy
2
(62)
273
Here the superscripts i and f refer to gluon emission from the initial- and final-state quarks,
i,f
respectively. The subscript k on Ak refers to one of the three possible Dirac structures
1 2 , namely k = 1 for (V A) (V A), k = 2 for (V A) (V + A), and
k = 3 for (2)(S P ) (S + P ). As always, M (x) denotes the leading-twist light-cone
distribution amplitude of a pseudoscalar meson M, and the asymptotic forms of the twist-3
amplitudes have been used. Note that in the limit of symmetric (under x x)
distribution
i
i
amplitudes, and assuming SU(3) flavour symmetry, we have A1 = A2 and Ai3 = 0. In this
approximation the annihilation contributions can be parameterized by only two quantities
f
(Ai1 and A3 ). For an estimate of these annihilation contributions we use the asymptotic
form of the leading-twist distribution amplitudes to obtain

2
i
2 2
+ 2r XA ,
A1 s 18 XA 4 +
3
2

f
A3 12s r 2XA
(63)
XA ,
1
where XA = 0 dy/y parameterizes the divergent endpoint integrals. Similar to the case of
the twist-3 hard-scattering contributions parameterized by XH , we will treat the quantity
XA as a phenomenological parameter. Clearly, taking the same value of XA for all
annihilation terms is a crude model. We shall see below, however, that the parameter
f
A3 (contributing to penguin annihilation topologies) gives the dominant contribution.
Therefore, our treatment effectively amounts to defining a model for this particular
parameter.
To complete the calculation we need to account for the flavour structure of the various
operators. It is convenient to introduce the compact notation

M1 M2 |j1 j2
(64)
Bq icfBq fM1 fM2 ,
where the constant c takes into account factors of (1) or 1/ 2 appearing in the quark
wave-functions of some of the mesons, and c = 0 only if the flavours of the currents j1
and j2 match those of the mesons M1 and M2 , respectively. It is apparent from Fig. 4 that
the products j1 j2 have the flavour structure

q
q12 =
(65)
q q2 q1 q ,
q
where the sum over q = u, d, s arises from the g q q vertex in the annihilation
q
diagrams. Effectively, the flavour structure q12 creates a quark q1 (lower index)
and an antiquark q2 (upper index), together with a flavour-singlet q q pair. It is then
straightforward to find the set of meson final states to which a given flavour structure
contributes. All operators contributing to charged B decays have the structure du or
su . The corresponding two-particle final states with light (flavour-nonsinglet) pseudoscalar
mesons are
du :
0 , 0, K K 0,
su :
0.
0K , K
(66)
r decays (with r = d, s) have the structure r , sr , uu ,

Operators contributing to neutral B
d
q

q
or one of the penguin structures tr( ) q q and tr(Q ) q eq q . Here Q is the
274
charge operator for the quarks, and the sum (trace) over q is inherited from the QCD and
electroweak penguin operators in the effective weak Hamiltonian. The final states to which
these operators contribute are
dd :
sd :
0K 0,
+ , 0 0, K
0,
+K , 0K
ss :
K + , K 0 0,
0,
K +K , K 0K
uu :
0 0, +, K K +,
tr( ), tr(Q ):
0K 0.
0, K
0 0, + , + , K K +, K +K , K 0K
ds :
(67)
It follows from the above discussion that the weak annihilation contributions to the decay
amplitudes can be parameterized in terms of the coefficients

CF
CF
f
f
C1 Ai1 ,
b3 = 2 C3 Ai1 + C5 Ai3 + A3 + Nc C6 A3 ,
2
Nc
Nc

CF
C
F
b2 = 2 C2 Ai1 ,
b4 = 2 C4 Ai1 + C6 Ai2 ,
Nc
Nc

C
F
f
f
b3EW = 2 C9 Ai1 + C7 Ai3 + A3 + Nc C8 A3 ,
Nc

C
F
b4EW = 2 C10 Ai1 + C8 Ai2 .

Nc
b1 =
(68)
They correspond to currentcurrent annihilation (b1 , b2 ), penguin annihilation (b3 , b4 ), and

electroweak penguin annihilation (b3EW, b4EW ), where within each pair the two coefficients
correspond to different flavour structures. The quantities bi depend on the final-state
i,f
mesons through the light-cone distribution amplitudes entering the expressions for Ak
and thus should be written as bi (M1 M2 ). We suppress this notation when confusion cannot
arise. As for the hard spectator terms, we will evaluate the various quantities in (68) at the
scale h = h .

The effective weak Hamiltonian for B-meson
decays contains a strangeness-conserving
part (HHS=0 ) and a strangeness-changing part (HHS=1 ). Using the above definitions, the
annihilation contributions to the matrix elements of HHS=1 can be written as
GF

p M1 M2 |Tpann
M1 M2 |HHS=1
(69)
B =
B ,
2 p=u,c
, and
where p = Vpb Vps

Tpann = up rn b1 uu + ru b2 su + b3 sr + rn b4 tr( )
3
3
+ b3EW er sr + rn b4EW tr(Q ).
(70)
2
2
The index r refers to the flavour of the spectator quark inside the B meson in (69), and
mesons and 0 for B . The matrix elements of
rn = rd + rs equals 1 for neutral B
275
Table 2
Summary of theoretical input parameters
QCD scale and running quark masses
(5)
MS
225 MeV
mb (mb )
mc (mb )
ms (2 GeV)
(mu + md ) (2 GeV)
4.2 GeV
(1.3 0.2) GeV
(110 25) MeV
(9.1 2.1) MeV
F0B (0)
0.28 0.05
RK
0.9 0.1
2
0.1 0.3
B
(350 150) MeV
Parameters related to hadronic matrix elements

f
131 MeV
fK
160 MeV
fB
(180 40) MeV
Parameters of distribution amplitudes

1K
0.3 0.3
2K
0.1 0.3
1
0
in this case,
HHS=0 take an identical form, except that p is replaced with p = Vpb Vpd
q
q
and s must be replaced with d .
It is now straightforward to derive the annihilation contribution to a particular final state
in terms of the coefficients bi (M1 M2 ). The expressions for the decay modes discussed in
this paper have been given earlier in (9) and (13). We will later use the approximation (63)
i,f
for the quantities Ak to estimate the annihilation coefficients bi numerically. We should
recall, however, that the annihilation kernels have been derived under the assumption of
hard scattering. Specifically, we have neglected the momentum fraction of the spectator
quark in the B meson compared to x, x,
y, y in deriving the kernels. This is the reason why
i,f
the results for Ak turned out to be independent of the form of the B-meson distribution
amplitude. In the endpoint regions, one or two of the variables x, x,
y, y can be of
order , invalidating this approximation. Therefore, our numerical results for the weak
annihilation contributions presented in the next section must be considered as modeldependent estimates.
4. Numerical analysis of amplitude parameters

In this section we summarize the numerical values of the parameters ai and bi entering
the B K decay amplitudes and perform detailed estimates of various sources of
theoretical uncertainties. Other decays such as B will be discussed later.
The theoretical input parameters used in our analysis, together with their respective
ranges of uncertainty, are summarized in Table 2. The quark masses are running masses
in the MS scheme. Note that the value of the charm-quark mass is given at = mb .
The ratio sc = (mc /mb )2 needed for the calculation of the penguin contributions is
scale independent. The values of the light quark masses are such that rK = r . We
hold (mu + md )/ms fixed and use ms as an input parameter. (This implies that in our
276
error estimation procedure |P /T | indirectly depends on ms .) The value of the QCD

scale parameter corresponds to s (MZ ) = 0.118 in the MS scheme. The values for the
B-meson decay constant fB , the semileptonic form factor F0B (0), and the hadronic
parameter RK in (15) are consistent with recent determinations of these quantities using
light-cone QCD sum rules [40,41], form-factor models (see, e.g., [42]), and lattice gauge
theory (see, e.g., [43]). The last row in the table contains our values for the Gegenbauer
moments of the pion and kaon light-cone distribution amplitudes, and for the B-meson
wave-function parameter B . The Gegenbauer moments for the light mesons are adopted
with a conservative error estimate that encompasses most of the parameter ranges obtained
from phenomenological or QCD sum-rule determinations of these quantities. The value of
B is an educated guess guided by the model determinations B 23 300 MeV [32]
and B = (380 120) MeV [38]. For comparison, using the models of [17] we obtain
B = (410 170) MeV.
4.1. Vertex and penguin contributions
We begin the discussion with the vertex and penguin contributions, i.e., the terms
ai,I (K) in (46). In this case all convolution integrals are finite, even for the powersuppressed coefficients rK a6,I (K) and rK a8,I (K). Table 3 contains the values of
the various coefficients for three different values of the renormalization scale, and
with theoretical uncertainties due to the variation of the pion and kaon light-cone
distribution amplitudes and the charm-quark mass, as specified in Table 2. We vary the
Gegenbauer moments independently and quote the maximal variation in the table. The
main characteristics of theoretical uncertainties are as follows:
Renormalization-scale dependence The residual scale dependence at next-to-leading
order depends on the size of the leading-order Wilson coefficients and the magnitude of
the Wilson coefficient that multiplies the next-to-leading order correction. In general, we
find a significant reduction of scale dependence compared to the ai parameters obtained at
leading order (corresponding to naive factorization). Only for the parameters a2 and a10
a sizeable scale dependence remains at next-to-leading order, even though it is reduced
by about a factor of 2 relative to the leading order. In general, the imaginary parts of the
coefficients ai,I , which occur first at order s , have a larger scale dependence than the real
parts.
Light-cone distribution amplitudes The explicit expressions in Section 3.4 show that the
second Gegenbauer moments 2K and 2 enter the results for the vertex and penguin
contributions typically with small coefficients. Therefore, the main uncertainty comes
from the first moment 1K of the kaon distribution amplitude, which affects only a1 , a4
and a10 . The real part of a10 is particularly uncertain. There is also some uncertainty in
a2 , because the dependence on the second moment of the pion light-cone distribution
amplitude is amplified by the large Wilson coefficient C1 . From Table 3, we conclude
that the dependence on the distribution amplitudes is almost always smaller than the
277
Table 3
Next-to-leading order results for the coefficients ai,I (K) for three different choices of the
renormalization scale. Numbers in parentheses show the maximal change in the last digit(s) under
variation of the Gegenbauer moments of the light-cone distribution amplitudes; if present, numbers
in square brackets show the change under variation of the charm-quark mass
Real part
mb /2
a1,I
1.073(8)
0.039(4)
a2,I
0.008
a3,I
u
0.031(4)
a4,I
c
a4,I
0.036(9)[2]
0.011
a5,I
u
0.052
rK a6,I
c
0.056
rKa6,I
a7,I /
0.007
u / 0.090
rKa8,I
c / 0.090
rKa8,I
a9,I / 1.258(1)
u /
0.062(27)
a10,I
c /
0.062(27)
a10,I
mb
Imaginary part
2mb
mb /2
mb
2mb
1.054(4)
1.037(2)
0.048(11)
0.026(6)
0.015(3)
0.005(3)
0.045(2)
0.113
0.084
0.066
0.006
0.004
0.004
0.002
0.001
0.029(3)
0.027(2)
0.023(0)
0.017
0.014
0.033(6)[1]
0.030(4)[1]
0.005(3)[4] 0.004(3)[3]
0.004(2)[2]
0.007
0.004
0.005
0.003
0.001
0.052
0.052
0.017
0.018
0.019
0.056
0.056
0.005[3]
0.007[3]
0.008[3]
0.011
0.025
0.004
0.002
0.001
0.077
0.059
0.001
0.009
0.020
0.075
0.055
0.000
0.005[1]
0.010[3]
1.222(1)
1.181
0.040
0.022
0.012
0.020(20) 0.025(16)
0.168(39)
0.116(29)
0.084(22)
0.018(21)[1] 0.028(17)[1] 0.168(39)
0.121(30)[1] 0.093(25)[2]
scale dependence. Since, for consistency, we must use the asymptotic twist-3 distribution
amplitudes, the coefficients a6 and a8 show no dependence on the Gegenbauer moments.
This is clearly an approximation, which is valid only if the chirally-enhanced power
corrections dominate over the remaining power corrections. (This is a questionable
approximation, because the quarkantiquarkgluon distribution amplitude impacts on
p (x) with a large numerical coefficient [34].) As a consequence, we cannot control
SU(3)-breaking effects at twist-3 order. In practice, we expect such SU(3) violations to
have a similar (hence, small) effect as those at leading twist.
Charm and strange-quark masses The value of the charm-quark mass affects the penguin
contributions with a charm-quark loop. This leads to a significant uncertainty in the
imaginary parts of a4c and a6c . The real parts of the coefficients are much less affected. Note
that the chiral enhancement factor rK multiplying the coefficients a6 and a8 in Table 3 is
inversely proportional to the strange-quark mass. The 25 MeV uncertainty in the value
of ms leads to a sizeable uncertainty in the values of the products rKa6,8 which, except for
c , is a much larger effect than the dependence on the charmthe imaginary parts of rKa6,8
quark mass.
Unknown power corrections A source of theoretical uncertainty that is difficult to
estimate arises from power corrections which cannot be computed using the QCD
factorization approach. Naive dimensional analysis suggests that such corrections are
278
Table 4
values
Coefficients ai,II (K) for three different choices of the renormalization scale and fixed default
of the quantity HK . All scale-dependent quantities are evaluated at the scale h = h with
h = 0.5 GeV. Numbers in parentheses show the maximal change in the last digit(s) under variation
of the Gegenbauer moments; numbers in square brackets show the dependence under variation
of RK
h
default
HK
a1,II
a2,II
a3,II
a4,II
a5,II
a7,II /
a9,II /
a10,II /
mb /2
1.02 GeV
0.92
0.087(16)[10]
0.231
0.010
0.004(1)
0.016
0.014
0.112
0.221(42)[25]
mb
1.45 GeV
0.99
0.061(14)[7]
0.192
0.007
0.003(1)
0.010
0.009
0.080
0.182(42)[20]
2mb
2.05 GeV
1.05
0.045(12)[5]
0.167
0.005
0.002(1)
0.008
0.006
0.060
0.157(42)[17]
of order QCD /mb (1020)%, but they could be enhanced, e.g., by large Wilson
coefficients. This situation may potentially be realized for the parameters a2 and a10 .
Recently, some power corrections to QCD factorization for the final state have been
investigated in the framework of QCD sum rules [44] and using the renormalon calculus
[36]. (Power corrections for final states with one heavy meson were also considered in [15,
45].) In none of these cases particularly large corrections have been identified.
4.2. Hard spectator interactions
We have argued in Section 3.4 that the description of the hard-scattering contributions
to the coefficients ai suffers from large theoretical uncertainties, which however can be
parameterized in terms of a single (complex) quantity HK defined in (57). If this quantity
is fixed, then the hard-scattering contributions ai,II can be calculated with relatively small
uncertainties. In Table 4, we show the results for these coefficients obtained by keeping
HK fixed at its central value (using central values for all input parameters and setting
XH = ln(mB /h ) 2.4). Because the hard-scattering terms arise first at order s , they
exhibit a relatively strong scale dependence. In addition, the coefficients a1 , a4 and a10
have some dependence on the Gegenbauer moments and on the value of the ratio RK .
Although the twist-3 correction is sizeable, it does not dominate the result for the hard
spectator term. With our default value for XH we obtain an enhancement of the leading
twist-2 term by about 40%.
Comparison of the results for ai,I and ai,II in Tables 3 and 4 shows that in most cases
the hard spectator terms are of a similar magnitude as the vertex corrections. Notable
exceptions are the coefficients a2 and a10 , for which the hard-scattering contributions are
the dominant effects. The predictions for these coefficients are correspondingly uncertain.
279
Fig. 5. Ranges for the complex parameter HK . The dot shows the default value used in obtaining
the results in Table 4.
On the other hand, the hard spectator contributions are very small (or absent) in the case of
the coefficients a4 , a6 and a8 .
So far we have ignored the large overall uncertainty in the hard-scattering terms resulting
from the uncertainty in the value of HK . For an estimate of this quantity, we parameterize
the divergent integral XH in (59) in the form

mB
XH = 1 + IH eiH ln
(71)
, IH 1,
h
with an arbitrary phase H , which may be caused by soft rescattering. In other words, we
assign a 100% uncertainty to the default value XH = ln(mM /h ) 2.4. If, in addition,
the parameters B , fB and F0B (0) are varied within the ranges shown in Table 2, the
result for HK is confined to the interior of a region in the complex plane shown in Fig. 5.
(Variations of the Gegenbauer moments or the parameter RK have a minor effect and can
be safely neglected in this plot.) The obtained values for HK are of order unity, but with
an uncertainty of at least a factor 2 and a potentially significant strong-interaction phase
(of up to about 17 with our choice of parameters).
4.3. Annihilation contributions
As emphasized earlier, the results for the weak annihilation contributions derived in
Section 3.5 are based on the assumption of hard scattering, which is invalidated by the
presence of endpoint singularities. Nevertheless, Eqs. (63) and (68) can be employed as a
model for the annihilation terms, which we expect to give the correct order of magnitude
of the effects. In analogy with the previous section, we parameterize the divergent integral
XA in the form
mB

, IA 1,
XA = 1 + IA eiA ln
(72)
h
with an arbitrary phase A . Table 5 shows the results for the annihilation contributions
obtained with the default value XA = ln(mB /h ). They have an overall uncertainty of
about 30% due to the error in the value of the ratio rA = (3.0 0.9) 103 defined in
(21).
280
Table 5
(EW)
for three different choices of the renormalization scale and fixed
Annihilation coefficients rA bi
default values of all input parameters
h
rA b1
rA b2
rA b3
rA b4
rA b3EW /
rA b4EW /
mb /2
1.02 GeV
mb
1.45 GeV
2mb
2.05 GeV
0.025
0.011
0.008
0.003
0.021
0.014
0.021
0.008
0.006
0.002
0.018
0.010
0.018
0.006
0.005
0.001
0.016
0.007
We observe that the default values for the annihilation contributions are rather small,
compatible with being first-order power corrections of a canonical size. Specifically, from
the relations for the amplitude parameters in (20) it follows that for an estimate of the
most important annihilation effects in B K decays we should compare rA b3 with
(a4c + rKa6c ) (denominator of 3/2 , T , a ), rA (b2 + b3 ) with (a4u + rKa6u ) (numerator of
c + r Ka c ) (numerator of q ). Similarly, from (24) it follows that
a ), and rA b3EW with (a10
C
8
in B decays we should compare rA (b3 + 2b4 ) with (a4c + r a6c ) (numerator of
P /T ). In all cases, annihilation effects can be neglected in comparison with a1 1.
With the default values from Table 5 the annihilation contributions are always a moderate
correction of less than 25% to the leading terms obtained from the QCD factorization
formula. However, these estimates have a large uncertainty.
Phenomenologically most relevant are the penguin annihilation effects parameterized
by b3 and b4 . They tend to increase the penguin amplitudes P in B K decays and
P in B decays, thereby reducing the values of the tree-to-penguin ratios 3/2 and
T , and increasing the value of the penguin-to-tree ratio P /T . The fact that penguin
annihilation graphs can significantly enhance the penguin amplitude has been noted first
in [17]. We confirm this effect; however, with our default parameter variations we find
a more moderate enhancement than these authors. In order to illustrate the effect and its
dependence on the value of the quantity XA , we show in Fig. 6 the combinations rA b3 and
rA (b3 + 2b4 ) parameterizing the penguin annihilation contributions in B K and B
decays. The default values for the leading penguin coefficients (a4c + r a6c ) are shown
for comparison. The regions bounded by the solid lines refer to our standard choice IA 1
in (72). In this case, the annihilation contribution can increase the penguin amplitudes
by up to 3040%. The dashed curves show the accessible parameter space in the more
extreme case where we let IA 2 (corresponding to a 200% uncertainty in the value of
the divergent integral XA ). Then the annihilation contributions can be almost as large as
the leading penguin terms. We will see later that there are no experimental indications of
such large annihilation effects.
281
Fig. 6. Ranges for rA b3 (left) and rA (b3 + 2b4 ) (right) parameterizing, respectively, penguin
annihilation effects in B K and B decays. Solid lines refer to IA = 1, dashed ones to
IA = 2. The dots show the default values. For comparison, the central values for the leading penguin
coefficients (a4c + r a6c ) are shown by the double circles.
4.4. Amplitude parameters for B K, decays

We are now in a position to combine the results for the parameters ai,I , ai,II and bi
discussed in the previous subsections into complete predictions for the decay amplitudes.
We focus first on the amplitude parameters defined in (20). They are sufficient to calculate
any ratio of B K decay amplitudes, such as CP asymmetries and ratios of CP-averaged
branching fractions. We also study the ratio P /T in B decays defined in (24).
A more extensive phenomenological analysis will be performed in Section 5.
It will be important to distinguish two types of theoretical uncertainties: those arising
from the variation of input parameters to the factorization formula, and those associated
with power corrections to factorization. Uncertainties of the first kind have a well-defined
meaning and can, at least in principle, be reduced in a systematic way. They include the
dependence on the renormalization scale, quark masses, moments of light-cone distribution
amplitudes, and hadronic quantities such as fB and F0B (0). The errors in the input
parameters can be reduced, e.g., by using experimental data or lattice calculations. The
residual dependence on the renormalization scale can be reduced by calculating higherorder corrections to the hard-scattering kernels. Our predictions also depend on the value
of |Vub /Vcb |; however, this should not be considered a theoretical uncertainty. Ultimately,
the goal is to use hadronic B decays to learn about such CKM parameters.
Theoretical uncertainties related to power corrections to the factorization formula are
of a different quality. Since factorization does, in general, not hold beyond leading power,
there is no systematic formalism known that would allow us to analyze power corrections
in a model-independent way. This is a general problem of QCD factorization theorems
in cases where no operator product expansion can be applied. (Another familiar example
are event-shape variables in e+ e annihilation to hadrons.) In the present work, we have
identified sources of potentially large power corrections (chirally-enhanced contributions
282
Table 6
Predictions for the amplitude parameters including all theoretical uncertainties of the first kind (see
text), but using default values for the power corrections to QCD factorization. The first error on
the central value is the sum of all theoretical uncertainties added in quadrature. The second error
(if present) shows the dependence on |Vub /Vcb |. The last column indicates the two most important
contributions to the theoretical uncertainty. For each quantity, the second line shows the result without
weak annihilation contributions (except for q and , which do not receive annihilation terms)
Parameter
Central value
Dominant errors
3/2 (%)
23.9 4.5 4.8

25.7 4.8 5.1
3.5(ms )
3.6(ms )
1.4()
1.6(2K )
(deg)
9.6 3.8
3.5(mc )
1.4(1K )
10.2 4.1
3.7(mc )
1.5(1K )
20.6 3.5 4.1

22.0 3.6 4.4
5.7 4.4
6.2 4.6
3.2(ms )
3.3(ms )
3.5(mc )
3.7(mc )
0.9()
0.8(2K )
2.3()
2.2()
2.0 0.1 0.4

1.9 0.1 0.4
13.6 4.4
16.6 5.2
0.1(mc )
0.1(mc )
3.7(mc )
3.9(mc )
0.1()
1.7(1K )
2.8()
q (%)
(deg)
58.8 6.7 11.8

2.5 2.8
6.4(RK )
1.9()
1.3()
1.8(1K )
qC (%)
8.3 4.5 1.7
2.7(B )
2.3(1K )
8.9 4.9 1.8

60.2 49.5
54.2 44.2
3.1(B )
31.7()
29.5()
2.3(1K )
27.9(B )
24.1(B )
28.5 5.1 5.7

25.9 4.3 5.2
8.2 3.8
9.0 4.1
4.6(ms )
4.1(ms )
3.3(mc )
3.6(mc )
1.8()
0.8()
2.0()
1.8()
T (%)
T (deg)
a (%)
a (deg)
C (deg)
|P /T | (%)
arg(P /T ) (deg)
and weak annihilation terms) and estimated their effects. These estimates are uncertain due
to logarithmically divergent endpoint contributions, which indicate the dominance of soft
gluon exchange. Without significant conceptual progress in the understanding of power
corrections to observables that do not admit a local operator product expansion, it will be
difficult to reduce these uncertainties in a systematic way.
Our results for the amplitude parameters including all theoretical uncertainties of the
first kind are shown in Table 6. They are obtained by keeping the parameters XH and
XA entering the power corrections fixed at the default value XH = XA = ln(mB /h ). All
other input parameters are varied within the ranges shown in Table 2. Following common
practice, we vary the renormalization scale between mb /2 and 2mb . The individual
283
contributions to the error are then added in quadrature to obtain the total theoretical
uncertainty of the first kind. The two most important contributions to the total error are
shown in the last column of the table. The sign convention is such that the upper (lower)
sign corresponds to increasing (decreasing) the value of an input parameter. Finally, the
second error on the central value (if present) indicates the sensitivity to the uncertainty in
the ratio |Vub /Vcb | = 0.085 0.017, which we assume to be 20%. This is not a hadronic
uncertainty and therefore should not be combined with the first error. The amplitude
parameters are either proportional (or inversely proportional) to |Vub /Vcb | or independent
of this parameter, so that this error can easily be readjusted if needed.
In almost all cases the theoretical uncertainty is dominated by a single source. With
the exception of the strong-interaction phase C , our next-to-leading order results for
the amplitude parameters are very stable under variation of the renormalization scale.
The uncertainty in the values of the tree-to-penguin ratios 3/2 , T , and |P /T | is
dominated by the error on the strange-quark mass, whereas the corresponding stronginteraction phases , T , and arg(P /T ), as well as the phase a , are most sensitive
to the error on the charm-quark mass. The largest uncertainty in the electroweak-penguin
parameter q comes from the SU(3) violations parameterized by the amplitude ratio RK
in (15). The uncertainty in the value of the wave-function parameter B is the dominant
source of uncertainty for the electroweak-penguin parameter qC . Note that the Gegenbauer
moments of the pion and kaon light-cone distribution amplitudes are never the dominant
contribution to the error. This shows that the precise shapes of these amplitudes are of
minor importance for phenomenological applications of QCD factorization.
In order to study the theoretical uncertainties of the second kind, related to our estimate
of potentially large power corrections to the factorization formula, we show in Fig. 7
results for amplitude parameters obtained using central values for all input parameters,
but varying the complex quantities XH and XA according to the parameterizations (71)
and (72). The dots with error bars show the central results for the amplitude parameters
as given in Table 6 (for fixed |Vub /Vcb | = 0.085). The dashed curves (visible only for
3/2 and qC ) bound the parameter space obtained by variation of XH , whereas the solid
curves (present for all parameters except q) define the region obtained by variation of XA .
It is evident that the chirally-enhanced twist-3 corrections to the hard spectator interactions
do not lead to a dominant uncertainty. Their effect is always smaller than the uncertainty
due to parameter variations of the first kind. The uncertainty in the description of weak
annihilation is significant for the tree-to-penguin ratios 3/2 , T , P /T and their
phases. The resulting variation is typically as large as the total theoretical uncertainty
of the first kind. Although the modelling of weak annihilation thus introduces sizeable
uncertainties, we stress that their impact on the values of the amplitude parameters is still
a moderate correction (see also Table 6). Moreover, it is important that the dominant effect
is a universal contribution to the leading penguin amplitudes P and P , which leads
to correlations between the various amplitude parameters. Ultimately, this will help to
constrain the annihilation contributions using experimental data (see Section 5.3 below).
284
Fig. 7. Results for the amplitude parameters in the complex plane. Dots with error bars show the
default values and uncertainties of the first kind, while the regions bounded by curves show the
variation of the central values under variation of XH with IH = 1 (dashed), and XA with IA = 1
(solid).
5. Phenomenological applications
This section, which can be read without studying the technical details of our analysis,
illustrates several applications of our results to the phenomenology of B K and B
decays. The main goal will be to obtain information about the Wolfenstein parameters
and defining the unitarity triangle. After some general remarks in Section 5.1, we
start by considering methods for constraining and that depend on a minimal amount
of theoretical input about QCD dynamics. The simplest such strategy is the Fleischer
Mannel bound [6] discussed in Section 5.2. In the following Section 5.3, we review analysis
strategies developed by Rosner and one of us [1,3,4], which are based on CP-averaged rate
measurements for the charged modes B K, . These methods provide powerful
constraints in the (,
)
plane with minimal theoretical uncertainties. The third method,
0 +
the determination of sin 2 from the time-dependent CP asymmetry in B 0 , B
decays, is explored in Section 5.4. Towards the end of our discussion we will then rely more
heavily on the new theoretical results obtained in the present work. In Section 5.5, we give
predictions for the absolute values of branching fractions and results for various ratios
of CP-averaged branching ratios as a function of . We then perform a global fit to the
experimental data on the branching ratios and extract the corresponding allowed region in
the (,
)
plane. In the final Section 5.6, we show predictions for the direct CP asymmetries
in the various B K and B decay modes. Because of their sensitivity to stronginteraction phases, these results have the largest theoretical uncertainties.
285
5.1. General observations

Several points following from the numerical analysis in Section 4.4 are worth repeating
here, because they will have a direct impact on some of the analysis strategies mentioned
below:
1. The rescattering effects parameterized by a are very small and not much affected by
0 in (18) is, to a
theoretical uncertainties. Therefore, the decay amplitude for B K
very good approximation, given by the pure penguin amplitude P . This has two important
consequences: first, it is a safe approximation to neglect terms of order a2 in the squared
decay amplitudes; secondly, it follows that the direct CP asymmetry in the decays B
K 0 ,

ACP + K 0 = 2a sin a sin + O a2 1% sin ,
(73)
is tiny and unobservable in the foreseeable future. (We define the CP asymmetries as

the difference of the B-meson minus B-meson
decay rates divided by their sum.) An
experimental finding of a sizeable asymmetry in this decay mode would have to be
interpreted as a sign of physics beyond the Standard Model, or as a gross failure of the QCD
factorization formula, indicating the presence of large, uncontrollable power corrections.
For completeness, we note that the prediction that a is small can be tested experimentally
by measuring the CP-averaged branching ratio for the decays B K K 0 [4,46].
2. The strong-interaction phase = (2.5 2.8) is accurately predicted and tiny,
consistent with zero within errors. Also, the value of q is not affected by annihilation
contributions. These observations confirm a theoretical argument presented in [2,3], which
uses Fierz identities and top-quark dominance in the electroweak penguin diagrams to
show that 0, and that q can be calculated in a model-independent way up to SU(3)breaking corrections. The argument is based on the fact that in the limit of V-spin (s u)
symmetry the leading electroweak penguin contributions parameterized by qei can be
related to the currentcurrent contributions from the operators Qu1 and Qu2 using Fierz
identities. Neglecting the small contributions from the operators Q7 , Q8 , and Q7 , one
then obtains

xt
3 C9 + C10
1
3 ln xt
qei
(74)

1
+
,
2
2 KM C1 + C2
xt 1
KM 8 sin W
where the ratio of Wilson coefficients is renormalization-scheme invariant and can thus
be evaluated at the electroweak scale. The strong-interaction phase vanishes in this
approximation. When SU(3)-breaking corrections and small electromagnetic contributions
are included, the above result is rescaled by a factor Rq = (0.84 0.10)ei(2.52.8) .

About half of the deviation from 1 is due to (mostly factorizable) SU(3) violations. The
important point to note is that the smallness of is a model-independent result that does
not rely on the QCD factorization formula. It follows that terms of second order in can
be safely neglected.
286
Table 7
Experimental results for the CP-averaged B K and B branching ratios in units of 106 .
The BaBar and Belle results are preliminary. Our averages ignore correlations
Decay mode
CLEO [48]
BaBar [49]
Belle [50]
Average
B 0 +
4.3+1.6
1.4 0.5
4.1 1.0 0.7
5.9+2.4
2.1 0.5
4.4 0.9
B 0
B 0 K
B 0K
B K 0
B0 0K 0
5.6+2.6
2.3 1.7
5.1+2.0
1.8 0.8
+3.6+0.9
7.13.01.2
< 12.7 (90% C.L.)
< 9.0 (90% C.L.)
< 12.6 (90% C.L.)
17.2+2.5
2.4 1.2
16.7 1.6+1.2
1.7
18.7+3.3
3.0 1.6
+3.0+1.4
11.62.71.3
18.2+4.6
4.0 1.6
+5.9+2.4
14.65.13.3
+2.1+1.0
10.81.91.2
+3.3+1.6
18.23.02.0
+3.1+1.1
8.22.71.2
+3.7+2.0
17.03.32.2
13.1+5.5
4.6 2.6
14.6+6.1
5.1 2.7
5.6 1.5
17.2 1.5
12.1 1.7
17.2 2.5
10.3 2.5
0 + K decay amplitude,
3. The electroweak penguin contributions to the B
parameterized by qC in (18), are small and can also be treated to first order to good
approximation. In the literature, these effects are sometimes referred to as coloursuppressed electroweak penguin contributions.
4. The strong-interaction phases , T , and arg(P /T ) are small. We find central
values of order 10 or less in magnitude, with an uncertainty of about a factor 2 due to
potentially large annihilation contributions and higher-order perturbative corrections to the
hard-scattering kernels. It follows that the cosines of these phases deviate from 1 by only
a few percent, and the direct CP asymmetries are suppressed by a factor | sin i | 0.1
0.3. We note, in this context, that the smallness of the phase difference T 4
(see Table 6) can be understood based on simple physical arguments and implies a strong
correlation between the direct CP asymmetries in the decays B 0 K and B 0
K [47].
In the following sections, we explain how these general observations can be put to work
in different analysis strategies. We start with those strategies that avoid theoretical input
on tree-to-penguin ratios such as 3/2 and T . This eliminates the sensitivity to weak
annihilation effects and hence the main uncertainty of the QCD factorization approach.
As a result, these strategies are particularly clean from a theoretical point of view.
Experimental data for the CP-averaged B K and B branching fractions as
reported by several experimental groups are collected in Table 7. The last column shows
our naive averages of these results neglecting correlations. We use the average result for
the B 0 branching ratio in spite of the fact that the individual measurements of
this mode have less than 3 significance.
5.2. FleischerMannel bound
Based on a few plausible assumptions, Fleischer and Mannel have proposed a very
simple method for obtaining a bound on the weak phase from the measurement of a
287
single ratio
R=
0 +K )
(B + ) Br(B 0 K + ) + Br(B
0)
(B 0 ) Br(B + + K 0 ) + Br(B K
(75)
of CP-averaged branching ratios [6] (see [5] for more sophisticated generalizations of
this method). The current value of this ratio obtained from Table 7 is R = 1.06 0.18.
Neglecting the small rescattering effects parameterized by a and the colour-suppressed
electroweak penguin contribution qC (observations 1 and 3), it follows from (18) that
R 1 2T cos T cos + T2 sin2 .
(76)
The bound excludes a region near | | = 90 provided that R < 1. It is valid for any
values of T and the strong-interaction phase T and is thus independent of theoretical
assumptions about the tree-to-penguin ratio in these decays.
Our calculations confirm the dynamical assumptions that go into the derivation of this
bound. The neglected terms in the exact formula for R are of order a T or qC T . Both
are at the few percent level and can be safely neglected until the experimental error on R
is much below 10%. (More accurately, we find that the bound (76) is violated by at most
1.5% and only in the region 72 < < 86 .)
5.3. Strategies based on charged modes
The theoretical analysis of the charged decays B K profits from the fact that, with
the exception of the strong-interaction phases and a , the hadronic parameters entering
the parameterization of the corresponding decay amplitudes in (18) can be constrained in
the limit of SU(3) symmetry [35]. Therefore, the QCD factorization formula is needed
only to reduce the uncertainties in the estimate of SU(3)-breaking corrections. We have
already seen above how SU(3) symmetry and Fierz identities help to calculate the quantity
qei with small theoretical uncertainties. In addition, the decay amplitudes for the charged
modes B K in (18) depend on the tree-to-penguin ratio 3/2 and the very small
rescattering parameter a (and the corresponding phases and a ). It is apparent from
Fig. 7 that our prediction for 3/2 has a large uncertainty due to weak annihilation
contributions as well as parameter variations. The advantage of the charged B K
modes is that the tree-to-penguin ratio can be determined, up to small SU(3) violations,
using experimental data. Specifically, a certain combination of the parameters 3/2 and a
can be measured by comparing the CP-averaged branching fractions for the decays B
0 and B K 0 . The relation is
3/2
3/2 =
(77)
Rth exp,
1 + 2a cos a cos + a2
where

fK 2[Br(B + + 0 ) + Br(B 0 )] 1/2
(78)
0)
f Br(B + + K 0 ) + Br(B K
is an observable, and Rth = 1 in the limit of U-spin (d s) symmetry [3]. In the
theoretical analysis of B K decays it is convenient to replace the parameter 3/2
exp = tan C
288
by 3/2 . Since a = O( KM ) is very small (observation 1), both quantities take very similar
values. The experimental uncertainty in the current value exp = 0.223 0.034 is still
large. Ultimately, however, the accuracy in the determination of 3/2 is only limited by
the theoretical uncertainty in the calculation of the SU(3)-breaking corrections to Rth .
The QCD factorization approach helps reducing the model dependence in this calculation.
Neglecting the tiny contributions from electroweak penguins,

a1 (K) + RK a2 (K) F0B (m2K )

.
Rth =
(79)
a1 () + a2 () F0B (m2 )
The form-factor ratio can be safely set equal to 1. (Deviations are of order (m2K
m2 )/m2B 1%.) Note that there are no annihilation contributions to Rth . If we keep
the parameter XH governing the twist-3 contributions to the hard spectator interactions
fixed and vary all other input parameters over their respective ranges of uncertainty, we
find that Rth = 0.98 0.02. Next, if we keep the input parameters fixed but vary XH as
shown in (71), we find a variation of about 0.01. However, since our focus here is on
K and X vary
SU(3) violations we should be more conservative and let the quantities XH
H
independently. This gives a larger variation of about 0.03. Taken altogether, we obtain
Rth = 0.98 0.05.
(80)
Combining this with the experimental value of exp gives 3/2 = 0.218 0.034exp
0.011th.
Our results for the SU(3)-breaking corrections parameterized by Rth and Rq (the
quantity that corrects (74)) are valid up to nonfactorizable corrections that simultaneously
violate SU(3) symmetry and are power-suppressed in QCD /mb . The potentially most
important power corrections are included in the error estimate. The remaining uncertainties
are of order

1 ms md
,
O
(81)
Nc
mb
and, by naive power counting, should not amount to more than a few percent.
Whereas our theoretical predictions for the parameter 3/2 were affected by large
uncertainties (see Fig. 7), relations (77) and (80) can be combined to obtain a much more
accurate value for the related parameter 3/2 provided, of course, that the experimental
value of exp has a small error. This parameter can then be used in the phenomenological
analysis of B K decays. Moreover, the comparison of the so-determined value
of this parameter with our prediction for the tree-to-penguin ratio 3/2 provides a
nontrivial test of the QCD factorization approach, and ultimately could help to reduce
the uncertainties in our modelling of weak annihilation contributions.
To illustrate this last point, we show in Fig. 8 our prediction for 3/2 ei (corresponding to
the first plot in Fig. 7) and underlay as a gray band the experimental value for exp with its
1 and 2 errors. (The two quantities should agree if we neglect the small deviation of Rth
from 1, and the tiny contribution of a in (77).) It is pleasing that the central experimental
value is in excellent agreement with our theoretical prediction. The smallness of the
289
Fig. 8. Comparison of the prediction for the tree-to-penguin ratio 3/2 in B K decays with
the experimental value (at 1 and 2 ) of the quantity exp defined in (78). The region bounded by
the solid line refers to our default annihilation model with IA = 1. The dashed line corresponds to
IA = 2.
tree-to-penguin ratio in B K decays has often been interpreted as evidence for large,
nonfactorizable contributions to the decay amplitudes. Here we find that this effect is
reproduced in the QCD factorization approach without any tuning of parameters. This
is an important result. A deviation of our prediction from the experimental result could
have been considered as an indication of large corrections to QCD factorization, such as
enhanced weak annihilation effects not covered by our simple model estimates. As an
example, the dashed curve in the figure shows the allowed region obtained by increasing the
value of the parameter IA in (72) from 1 to 2. Fortunately, the data provide no evidence for
the existence of such a large deviation from our central prediction. (However, the evidence
from exp alone does also not exclude a potentially large annihilation contribution, if it has
a large phase.)
Model-independent bound on
A key observable in the study of the weak phase is the ratio of the CP-averaged
branching ratios in the two B K decay modes, defined as
R =
0)
Br(B + + K 0 ) + Br(B K
.
2[Br(B + 0 K + ) + Br(B 0 K )]
(82)
Its current value is R = 0.71 0.14. The theoretical expression for this ratio obtained
using the parameterization in (18) is

2
1 2q cos + q 2
R1 = 1 + 23/2 cos (q cos ) + 3/2

23/2a sin2 cos cos a + (1 q cos ) sin sin a

23/2q sin sin + O a2 , 2 , a .
(83)
290
(The parameter q is also called EW in the literature on this ratio.) Note that the rescattering
effects described by a , as well as the terms linear in the small phase , are suppressed by
a factor of 3/2 and thus reduced to the percent level. Higher-order terms in these small
parameters can be safely neglected (observations 1 and 2).
From a measurement of the ratio R alone a bound on cos can be derived, which for
R < 1 implies a nontrivial constraint on the Wolfenstein parameters and [3]. Only
CP-averaged branching ratios are needed for this purpose. The idea is to allow the stronginteraction phases and a in (83) to take any value between 0 and 2 . (Of course, the
QCD factorization approach predicts rather restricted ranges for these strong-interaction
phases. For the moment, however, we will not make use of these predictions.) This gives
[3,4]

2

R1 < 1 + 3/2|q cos | + 3/2 3/2 + 2a sin2 + O a2 , a , 2 .
(84)
Note that there is no term linear in on the right-hand side. Provided R is significantly
smaller than 1, the bound implies an exclusion region for cos , which becomes larger the
smaller the values of R and 3/2 turn out to be. The effect of the rescattering contribution
proportional to a on the right-hand side of the bound is numerically very small.
For fixed value of q, Eq. (84) excludes a region in cos provided that R < 1. However,
we should take into account that the value of q itself depends on the Wolfenstein parameters
and .
Hence, it is more appropriate to display the relation (84) as a constraint in the
(,
)
plane. We use
,
cos =
2
+ 2
q=
q
0.222 0.025
=
,
Rb
2 + 2
(85)
where the numerical value q = qRb = 0.222 0.025 corresponds to |Vub /Vcb | = 0.085
0.017. The experimental inputs to the bound are the measured ratios R and exp of CPaveraged branching ratios. The theoretical inputs are the value of q,
the parameter Rth in
the relation 3/2 = Rth exp , and a value for the rescattering parameter a . The accuracy
of our predictions for q and Rth is intrinsically limited only by effects that have a strong
parametric suppression, as shown in (81). As far as a is concerned, the bound becomes
weaker the larger the value of a . In our analysis we take a < 0.04, corresponding to an
upper bound that is twice as large as predicted by the QCD factorization approach.
Fig. 9 illustrates the resulting constraint in the (,
)
plane obtained for some
representative upper bounds on R and exp . For comparison, the dashed circles show the
constraint arising from the measurement of the ratio |Vub /Vcb | in semileptonic B decays.
(Note that the information from kaon CP violation excludes < 0 in the Standard Model.)
It is evident that, depending on the values of R and exp , the constraint may be very
nontrivial. If R < 0.7, then only values | | > 90 are allowed, which are significantly
larger than those favoured by the global analysis of the unitarity triangle (see [5154] for
some recent discussions of the standard analysis). For yet smaller values R < 0.6, the
arising constraint would become inconsistent with the global analysis. This would be an
indication of some new flavour physics beyond the Standard Model. On the other hand,
if R > 0.9 then the excluded region is too small (or absent) to be of phenomenological
291
Fig. 9. Theoretical constraints on the Wolfenstein parameters (,

)
implied by (still hypothetical)
experimental upper bounds on the ratios R and exp . The left-hand plot refers to exp < 0.19,
the right-hand one to exp < 0.25. For a given upper bound on R , and are restricted to lie
inside the corresponding shaded region. The dashed circles show the allowed region implied by the
measurement of |Vub /Vcb | in semileptonic B decays.
significance. The present uncertainty in the value of R is too large to tell which of these
possibilities is realized.
Determination of and constraint in the (,
)
plane
Ultimately, the goal is of course not only to derive a bound in the (,
)
plane, but
to determine the Wolfenstein parameters (and thus the unitarity triangle). This requires
obtaining information about the strong-interaction phase in (83), which can be achieved
either through the measurement of a CP asymmetry or with the help of theory. A strategy
for a model-independent determination of from B K, decays has been
suggested in [1]. It generalizes a method proposed by Gronau, Rosner and London [55]
to include the effects of electroweak penguins. The approach has later been refined to
account for rescattering contributions to the B K 0 decay amplitudes [4]. However,
this method relies on the measurement of a direct CP asymmetry in addition to R and
hence requires very high statistics. Here, we suggest an easier strategy for a theory-guided
determination of and ,
which does not require the measurement of a CP asymmetry.
Instead, we will exploit the prediction of the QCD factorization approach that the stronginteraction phase is small, i.e.,

sin = O s (mb ), QCD /mb .

(86)
This implies that the deviation of cos from 1 is a second-order effect in s (mb ) and/or
QCD /mb . More specifically, we found that (10 15) , where the uncertainty is
dominated by the weak annihilation contribution. Taking this value literally we would
conclude that cos > 0.9; however, to be conservative we allow for a larger phase such that
cos > 0.8. (This prediction can be tested experimentally once the direct CP asymmetry
in the decays B 0 K has been measured [1,4].) With the help of this result, a
292
measurement of the ratio R can be used to obtain a narrow allowed region in the (,
)
plane, which for fixed value of |Vub /Vcb | corresponds to a determination of that is unique
up to a sign. To this end, we rewrite (83) as
cos = q +
2 (1 q 2 ) 2
3/2
1 R1 + 3/2
23/2(cos + 3/2q)

+ O a2 , a , 2 ,
(87)
where

= a sin2 cos cos a + (1 q cos ) sin sin a + q sin sin .
(88)
It is safe to take 0 < < 0.05 and treat this as an independent parameter. Then, in addition
to q and Rth , we must specify a range for cos .
In evaluating the result (87), we scan the theory input parameters in the ranges 0.197 <
q < 0.247, 0.93 < Rth < 1.03, 0 < < 0.05, and 0.8 < cos < 1 (corresponding to || <
)
plane obtained
37 ). The left-hand plot in Fig. 10 shows the allowed regions in the (,
for some representative values R and the current central value exp = 0.22. We stress that
with this method a useful constraint on the Wolfenstein parameters is obtained for any value
of these parameters. When combined with a measurement of |Vub /Vcb |, this determines the
weak phase up to a sign ambiguity. Note that the theoretical accuracy of the method is
high, especially if the ratio R turns out to be close to 1, corresponding to a value of
as suggested by the global analysis of the unitarity triangle. In that case the theoretical
uncertainty in the determination of is about 10 (for fixed value of |Vub /Vcb |). Also, the
resulting constraint is then very weakly dependent on the value of the parameter exp . We
stress that the width of the band in the (,
)
plane corresponding to R = 1 is narrower
than the widths of the theoretical error bands corresponding to the standard constraints
mixing, and
on the unitarity triangle derived from charmless semileptonic B decays, BB

CP violation in KK mixing. Only the measurement of sin 2 is theoretically cleaner. The
uncertainty in the extraction of increases as one considers values of R significantly less
than 0.8 or larger than 1.2. However, such values would be inconsistent with the global
analysis of the unitarity triangle [5154] and thus provide evidence for physics beyond the
Standard Model.
With the current values of the branching ratios collected in Table 7, the method proposed
here is at the verge of providing a useful constraint in the (,
)
plane. This is shown in
the right-hand plot in Fig. 10, where we indicate the resulting allowed regions at 68% and
95% confidence level and compare them with the allowed region (light gray area) obtained
from the standard global fit of the unitarity triangle. Here and below we use the most recent
result for the standard fit obtained in [54], which includes the measurements of sin 2 at the
B-factories and adopts a conservative treatment of theoretical uncertainties that is similar
in spirit to the one adopted here. In evaluating the 1 and 2 domains of the quantities R
and exp , we take into account the correlation implied by the fact that the B K 0
branching ratio enters both quantities. Specifically, we vary the branching fractions for
B 0 , K 0 and 0 K independently such that 2 1 (for 1 ) or 4 (for 2 ). In
this way we find that the minimum and maximum values of R at 95% confidence level are
293
Fig. 10. Left: Allowed regions in the (,

)
plane corresponding to exp = 0.22 and different values of
R as indicated. The widths of the bands reflect the total theoretical uncertainty, obtained by scanning
the input parameters inside the ranges specified in the text. Right: Allowed regions at 68% and 95%
confidence level obtained from the current experimental results on the branching ratios (theoretical
input parameters are scanned as before). The dark band shows the theoretical uncertainty for the
central experimental values. The light gray area is the allowed region obtained from the standard
global fit of the unitarity triangle [54].
0.47 and 1.07, respectively. If in the future the upper value can be reduced, the resulting
allowed region in the (,
)
plane will no longer fully overlap with the standard domain.
5.4. Determination of sin 2 from B + decays
The methods described so far in this section provide constraints in the (,
)
plane that,
in essence, correspond to a determination of the magnitude of = arg(Vub ). Independent

information about the unitarity triangle can be obtained from a measurement of the time 0 + , which is sensitive to the Bd B
d
dependent CP asymmetry in the decays B 0 , B
mixing phase e2i . We define
0 (t) + )
Br(B 0 (t) + ) Br(B
0 (t) + )
Br(B 0 (t) + ) + Br(B
= S sin(HmB t) + C cos(HmB t),
A
CP (t) =
(89)
where
ei + P /T
.
ei + P /T
(90)
The coefficient C , which is a function of the weak phase , coincides with the direct CP
asymmetry to be discussed later. The mixing-induced asymmetry S depends on and .
In fact, in the limit where P /T is set to zero it follows that = e2i(+ ) = e2i ,
and hence S = sin 2.
S =
2 Im
,
1 + | |2
C =
1 | |2
,
1 + | |2
= e2i
294
Fig. 11. Relation between sin 2 and the mixing-induced CP asymmetry S , assuming
sin 2 = 0.48. The dark band reflects parameter variations of the first kind, the light band shows
the total theoretical uncertainty. The lower portion of the band refers to values 45 < < 135 , the
upper one to 0 < < 45 (right branch) or 135 < < 180 (left branch).
When the penguin contributions to the B decay amplitudes are included, the
relation between the coefficient S and sin 2 receives hadronic corrections [811], which
can be calculated using the QCD factorization approach [14]. To illustrate the effect,
we first assume that |Vub /Vcb | and the weak phase have been determined accurately.
Then using = 180 the expression for in (90) becomes a function of
and our prediction for the penguin-to-tree ratio P /T . If we further assume that the
unitarity triangle lies in the upper half of the (,
)
plane, then a measurement of S
determines sin 2 with at most a two-fold discrete ambiguity. Fig. 11 shows the relation
between the two quantities for the particular case where |Vub /Vcb | = 0.085 and =
14.3 , corresponding to sin 2 = 0.48 (the current world average is sin 2 = 0.48 0.16
[56]). The dark band shows the theoretical uncertainty due to input parameter variations
as specified in Table 2, whereas the light band indicates the total theoretical uncertainty
including the effects of weak annihilation and twist-3 hard spectator interactions. We
observe that for negative values of sin 2, as preferred by the global analysis of the
unitarity triangle [5254], a measurement of the coefficient S could be used to determine
sin 2 with a theoretical uncertainty of about 0.1. Interestingly, for such values of sin 2
the penguin pollution effect enhances the value of the mixing-induced CP asymmetry,
yielding values of S between 0.5 and 1. Such a large asymmetry should be relatively
easy to observe experimentally.
Although it illustrates nicely the effect of penguin pollution on the determination of
sin 2, Fig. 11 is not the most appropriate way to display the constraint on the unitarity
triangle implied by a measurement of S . In general, there is a four-fold discrete
ambiguity in the determination of sin 2, which we have reduced to a two-fold ambiguity
by assuming that the triangle lies in the upper half-plane. Next, and more importantly, we
295
Fig. 12. Allowed regions in the (,

)
plane corresponding to constant values of the mixing-induced
asymmetry S , assuming the Standard Model (left), or using a fixed value sin 2d = 0.48 (and
d mixing phase (right). The widths of the bands reflect the total theoretical
cos 2d > 0) for the Bd B
uncertainty. The corresponding bands for positive values of S are obtained by a reflection about
the axis. The light circled area in the left-hand plot shows the allowed region obtained from the
standard global fit of the unitarity triangle [54].
have assumed that |Vub /Vcb | and are known with precision, whereas is undetermined.
However, in the Standard Model |Vub /Vcb | and the angles , , are all functions of the
Wolfenstein parameters and .
It is thus more appropriate to represent the constraint
implied by a measurement of S as a band in the (,
)
plane. To this end, we write
i
ei =
,
2 + 2
r ei
P
=
,
T
2 + 2
e2i =
(1 )
2 2 2i (1
,
2
2
(1 )
+
(91)
where r ei parameterizes the large fraction in (24) without the prefactor 1/Rb and is
thus independent of and .
We now insert these relations into (90) and draw contours
of constant S in the (,
)
plane. The result is shown by the bands in the left-hand
plot in Fig. 12. The widths of the bands reflect the total theoretical uncertainty (including
power corrections). For clarity we show only bands for negative values of S ; those
corresponding to positive S values can be obtained by a reflection about the axis
(i.e., ).
Note that even a rough measurement of S would translate into a rather
narrow band in the (,
)
plane (this was also noted in [57]), which intersects the ring
representing the |Vub /Vcb | constraint at almost right angle. This would therefore provide
a very powerful constraint on the Wolfenstein parameters. We also stress that the sign
of S , when combined with a measurement of | | (using, e.g., the method described in
Section 5.3), can potentially determine whether the unitarity triangle lies in the upper or
lower half-plane, and thus provide a nontrivial test of the CKM model of CP violation.
296
d
The strategy outlined above remains useful even in the hypothetical case where Bd B
mixing is affected by new physics beyond the Standard Model (this scenario has recently
been discussed in [5862]). Then the factor e2i in the expression for in (90) must be
replaced with the mixing phase e2id , where d = due to the presence of new physics. It
is the value of sin 2d that is measured in the time-dependent CP asymmetry in the decays
0 J /KS . From (90), we then obtain
B0, B
i + r ei

2
=
sin
2
i
sin
2
,
NP
(92)
d
d
+ i + r ei
which still implies a constraint in the (,
)
plane for each pair of experimental values S
and sin 2d . The sign refers to the sign of cos 2d . The result obtained for sin 2d = 0.48
is shown in the right-hand plot in Fig. 12. In the figure we assume that cos 2d > 0; the
resulting bands for cos 2d < 0 are, once again, obtained by a reflection about the axis.
We observe that the bands in the first quadrant of the (,
)
plane intersect the rings from
the |Vub /Vcb | constraint at almost the same places as in the Standard Model case, implying
d mixing have
that for S < 1 (and cos 2d > 0) potential new physics effects in Bd B
a minor impact on the results. On the other hand, the impact would be significant if S
turned out to be positive, or if cos 2d < 0.
5.5. Predictions for CP-averaged branching ratios
Whereas so far in this section we have focused on methods that require minimal input
from the QCD factorization approach, we now discuss in detail our theoretical predictions
for the B K, branching ratios. These predictions follow from the theory described
in this work without relying on further phenomenological input. All branching fractions
discussed in this section are averaged over CP-conjugate modes (even though for neutral
K and B mesons this is not indicated by the notation). We use B + = 1.65 ps and B 0 =
1.56 ps for the B-meson lifetimes.
Absolute predictions for branching fractions
Two out of the seven decay modes, B 0 and B K 0 , are (almost)
independent of the CKM phase , since the corresponding decay amplitudes have to a
very good approximation only a single weak phase. The predicted branching fractions for
these modes are (setting = 55 for concreteness)

+0.8

|Vub | F0B (0) 2
6
0
10 Br B = 5.30.4 B , 2 0.3(XH )
,
0.0035 0.28
B

F0
(0) 2
+6.4
+8.1
6
0
10 Br B K = 14.14.0(ms )3.6 (XA )
(93)
,
0.28
where the first error is due to parameter variations as shown in Table 2, whereas the second
one accounts for the uncertainty due to power corrections from weak annihilation and
twist-3 hard spectator contributions. The dominant contributions to the uncertainty are
shown in parentheses. Note that the decays B 0 do not receive weak annihilation
297
contributions. The largest uncertainty for the 0 final state is a 50% normalization
uncertainty from the current errors on |Vub | and the B form factor. The systematics
of theoretical errors is different for the K final states. Their absolute branching fractions
are (approximately) proportional to the square of the penguin amplitude a4 + rKa6 + rA b3 ,
which is sensitive to the penguin annihilation coefficient b3 (see the left-hand plot in
Fig. 6). The dominant theoretical errors therefore come from the annihilation parameter
XA and from the strange-quark mass (through rK ). The B K 0 branching ratio is to
a good approximation proportional to (ms /110 MeV)1.35 . The theoretical errors detailed
here are characteristic for the absolute branching fractions of all and K final states,
respectively.
The central values in (93) agree well with the current data summarized in Table 7. In
particular, we find that the QCD factorization approach prefers large branching fractions
for the K final states. This arises due to an enhancement of the QCD penguin amplitude
relative to naive factorization. Weak annihilation contributions play only a minor role in
this enhancement. Without annihilation the B K 0 branching fraction would be
reduced to 12.1 106. Hence the effect is not negligible, but there is no need for a largely
enhanced annihilation contribution (this is in contrast to the findings of [17]). In fact, the
good agreement of our prediction with the data provides circumstantial evidence against
the idea that annihilation effects could be much enhanced with respect to our estimates. For
instance, increasing the parameter IA in (72) from 1 to 2 would increase the corresponding
+26.4
error on the branching ratio from +8.1
3.6 to 9.2 , in which case it would require considerable
fine-tuning of the strong-interaction phase of XA to reproduce the experimental value of
the branching ratio.
It has also been suggested in the literature that one needs a large enhancement from
penguin diagrams with a charm loop (so-called charming penguins) to understand
the overall K branching fractions [63]. The leading perturbative contribution from the
penguin diagrams in Fig. 3, however, turns out to be very small. There exists also a
power-suppressed contribution from these diagrams, when one of the quarks the gluon
decays into becomes soft. In this case, the gluon is semi-hard and probes the charmquark loop at a scale of order h (QCD mb )1/2 . The penguin function G(s, x) in (51)
tends to a constant for small x, which implies that the contribution from the semi-hard
region to the function GK (s) in (50) is suppressed by two (not one!) powers of the heavyquark mass relative to the leading perturbative contribution. Hence, although the strong
coupling constant in the semi-hard region is larger than s (mb ), a large nonperturbative
enhancement of the charm-penguin contribution appears implausible. (It is possible to
obtain a first-order power correction by invoking a higher Fock component in the wave
function of the emission meson, as discussed in [15]. However, in this case the penguin
loop continues to be a hard subprocess, and so contributes a factor of s (mb ).) In a
recent article [64], a nonperturbative charming-penguin contribution has been fitted to
experimental data under the assumption that this effect is responsible for any deviation
of the measured branching fractions from those expected within the Standard Model with
and determined by the standard unitarity-triangle fit. It it worth noting that such a
fit is technically equivalent to fitting the annihilation contribution to the QCD penguin
298
Fig. 13. Absolute CP-averaged branching fractions as functions of . Central values are shown by
the solid line. In each plot, the inner (dark) band corresponds to the variation of the theory input
parameters (including |Vub /Vcb |), whereas the outer (light) band also includes the uncertainty from
weak annihilation and twist-3 hard spectator contributions (see text for details). The long-dashed
lines indicate the sensitivity to |Vub /Vcb |. The short-dashed curve shows the result obtained using
naive factorization.
amplitude, indirectly related to XA (Eq. (72)) in our notation. If a modification of this

amplitude were required (for which the present data does not provide strong motivation,
as will become more evident below), we would attribute its physical origin to weak
annihilation rather than charm penguins, given the power counting detailed above.
In Fig. 13, we show the dependence of the absolute branching fractions of the various
B K, decay modes (except for B 0 ) on the weak phase . In each plot,
299
Table 8
Predicted ratios of CP-averaged branching fractions for selected values of . The last column
shows the experimental values deduced from Table 7. Our averages ignore correlations between
the individual measurements
Ratio
2 Br( 0 K )
Br( K 0 )
Br( K )
2 Br( 0 K 0 )
B + Br( K )
B 0 Br( K 0 )
Br( + )
Br( K )
B + Br( + )
B 0 2 Br( 0 )
40
70
100
130
Experiment
0.94 0.07
1.16 0.07
1.44 0.16
1.70 0.25
1.41 0.29
0.92 0.08
1.17 0.08
1.50 0.19
1.83 0.34
0.83 0.22
0.74 0.07
0.91 0.04
1.12 0.07
1.32 0.12
1.06 0.18
0.96 0.60
0.67 0.38
0.43 0.25
0.27 0.18
0.26 0.06
0.96 0.25
0.83 0.20
0.66 0.15
0.50 0.13
0.42 0.14
the solid line gives the central prediction of the QCD factorization approach at next-toleading order in s . For comparison, the short-dashed line shows the result at leading
order, corresponding to naive factorization. The dark-shaded band is obtained by varying
all input parameters as specified in Table 2, and by varying the renormalization scale
between mb /2 and 2mb . It also includes the uncertainties due to the errors on the CKM
parameters |Vub /Vcb | = 0.085 0.017 and |Vcb | = 0.041 0.003. The variation due to
|Vub /Vcb | alone is indicated by the long-dashed lines. The light-shaded band adds to
this the uncertainties inherent to our modelling of power corrections due to twist-3 hard
spectator and weak annihilation corrections, as discussed in Sections 4.2 and 4.3. The
annihilation contributions dominate the uncertainty in all cases (see Fig. 7). They imply a
considerable uncertainty in the overall normalization of the K modes. For the purposes of
our discussion here the different sources of theoretical uncertainty are added in quadrature.
Later, when our focus is on constraining CKM parameters, we will be more conservative
and scan over the entire theory parameter space (corresponding to linear addition of errors).
The next-to-leading order effects included in the QCD factorization approach significantly enhance the branching fractions for the B K modes with respect to their values obtained in the naive factorization model. No such enhancement occurs for the decays
B 0 + and B 0 . As a result, the empirical finding that B K branching
ratios are larger than the B branching ratios is reproduced in our approach without
any tuning of parameters.
Predictions for ratios of branching fractions
The uncertainty in the overall magnitude of the penguin amplitude due to weak
annihilation effects, and the overall scale uncertainty in the predictions for the branching
fractions due to hadronic form factors and CKM elements, are largely eliminated by taking
ratios of branching fractions. The dependence of the six relevant ratios on the weak phase
is displayed in Fig. 14, in which the curves and bands have the same interpretation
as in the previous figure. (The results presented here are consistent with the preliminary
300
Fig. 14. Ratios of CP-averaged branching fractions as functions of . The meaning of the curves and
bands is the same as in Fig. 13. The horizontal bands show the 1 and 2 error bands for the current
experimental results.
results published in [16], which did not include annihilation contributions. However, the
treatment of theoretical uncertainties in [16] differs from the present work.) Table 8 gives
numerical predictions and theoretical uncertainties for some selected values of the angle .
It is evident that annihilation effects are less important for the ratios. The small error on
the ratios involving K final states shows the potential of these ratios to constrain .
Comparing our results with the preliminary experimental data collected in Tables 7 and 8
(and shown by the horizontal bands), we note a preference for large values of ; however,
the experimental errors are still too large to assign much significance to this observation.
301
An important remark in this context concerns the ratio involving the neutral mode B 0
0 K 0 , which seems to prefer a small value of . Our theoretical results indicate that to a
good approximation (compare the two plots in the first row in Fig. 14)
Br(B 0 K )
2 Br(B 0 K )
.
Br(B K 0 )
2 Br(B 0 0 K 0 )
(94)
This relation is in slight disagreement (albeit by no more than 2 ) with present data.
Some authors have interpreted this fact as an indication of large rescattering phases (see,
e.g., [65]). However, the validity of (94) is a model-independent consequence of isospin
symmetry, which is valid to linear order in the small tree-to-penguin ratios [4]. Therefore,
we expect that with more precise data the values of the two ratios in (94) will come closer
to each other.
Finally, we wish to stress that the difference between the QCD factorization results and
naive factorization are largest for the ratio of the + and K final states. Whereas
we can accommodate the low experimental value of this ratio for > 50 (at the 2 level),
this would not be possible for any value of if one were to use the naive factorization
model.
It is evident from Fig. 14 that in many cases the error on |Vub /Vcb | constitutes one
of the largest uncertainties in the predictions for the ratios of branching fractions. This
is true, in particular, for the interesting ratio of the + and K final states. For
instance, if |Vub /Vcb | is 20% lower than its current central value, then our prediction for
the B 0 + branching ratio is reduced by 40%, whereas the result for the B 0
K branching fraction remains almost unaffected. This shows that for future analyses
of rare B decays it will be important to reduce the theoretical uncertainty in the value of
|Vub |. We can unfold the strong dependence of the ratios on |Vub /Vcb | by presenting the
predictions for the branching fractions as functions of the Wolfenstein parameters and
.
This is illustrated in Fig. 15 for the two cases of the ratios 2 Br( 0 K )/ Br( K 0 )
and Br( + )/ Br( K ). The other two ratios involving K final states exhibit a
dependence similar to the first of these two examples. For not too large values of ,
any of
the three ratios involving only K final states is a direct measure of ,
since the dominant
For positive the
dependence on the CKM parameters arises from KM cos = tan2 C .
theoretical uncertainty is relatively small, as can be deduced by comparing the three panels
of Fig. 15 that refer to 2 Br( 0 K )/ Br( K 0 ). (Note that the contour lines in this plot do
not coincide with those of constant R1 inferred from Fig. 10, since there the parameter
3/2 is assumed to be fixed and extracted from data, whereas here it is determined from
theory and hence proportional to |Vub /Vcb |.) The ratio Br( + )/ Br( K ) exhibits
a rather different behaviour. The circular contours reflect the increase of the +
branching fraction with |Vub |. The offset of the center to negative values of results
from the interference of the two contributions with different weak phases to the decay
amplitudes, which for > 0 is constructive (destructive) for the (K) final states (and
vice versa for < 0).
The current experimental values for the CP-averaged branching fractions are still
afflicted by large errors, so a fit of (,
)
to these ratios may appear premature. Nonetheless,
302
Fig. 15. Two ratios of CP-averaged branching fractions as functions of the Wolfenstein parameters
and .
The upper row corresponds to our central values; the middle (lower) row corresponds to
the upper (lower) theoretical value including all uncertainties. We only show the upper half-plane
( > 0). The current standard best-fit unitarity triangle and the annulus given by |Vub /Vcb | are
overlaid to guide the eye.
it is useful to carry out such an analysis at the present stage, not only to exhibit its potential
once the errors decrease, but also to gauge the limitations eventually set by the theoretical
uncertainties. Fig. 16 shows the results separately for each of the five ratios of CP-averaged
branching fractions that have been measured so far. The dark-shaded band in each plot
covers the region allowed by theory given the current central experimental value. The lightshaded bands represent the theoretically allowed regions obtained by using experimental
values for the ratios shifted up or downwards by two standard deviations. The width of
each band thus reflects the (largely irreducible) theory uncertainty given a certain value
of the observable, while the current constraint on and provided by each observable
separately can roughly be taken to be the entire region bounded by the two light-shaded
bands.
303
Fig. 16. Regions in the (,

)
plane allowed by theory given the current central experimental values
of the ratios of branching fractions (dark bands), and the experimental values 2 standard deviations
(light bands). Only the upper half-plane ( > 0) is shown. The dashed circles indicate the allowed
region implied by the measurement of |Vub /Vcb | in semileptonic B decays. The light circled area
shows the allowed region obtained from the standard global fit of the unitarity triangle [54].
Global fit in the (,

)
plane
The information from all CP-averaged B K, branching fractions (and, in
the future, from the corresponding CP asymmetries) can be combined into a single
global fit, giving regions for the Wolfenstein parameters and that are allowed by
theory. We will now determine these regions, using two slightly different treatments of
theoretical uncertainties. Because our focus is to determine fundamental parameters of
the Standard Model, we adopt a more conservative error analysis than in the previous
paragraphs of this section and scan all theory input parameters over their respective
ranges of uncertainty (rather than adding theoretical uncertainties in quadrature). The same
conservative treatment was used in the analysis of the weak phase in Section 5.3.
It is evident from our previous discussion that ratios of CP-averaged branching fractions
provide the most powerful constraints on the Wolfenstein parameters, since they are
afflicted with relatively small theoretical uncertainties. However, there are two reasons
for not performing a global fit to the ratios directly (see also the discussion in Appendix C
304
Fig. 17. 95% confidence region in the (,

)
plane obtained from a fit to the CP-averaged
B K, branching fractions. The dark curve gives the envelope of the 95% confidence level
2 points. The solid lines refer to a cut on
contours, the lighter curve the envelope of the set of min
2
the min value which corresponds to a 5% confidence level of the fit for a given model. The dashed
2 per degree of freedom of less than 1. See text for further
lines refer to all models which have a min
explanations.
of [54]): first, with the present large experimental errors there are significant differences in
the results of the fit depending on whether the ratios or inverse ratios are used, leading to an
element of arbitrariness; secondly, even if two quantities have Gaussian errors, their ratio
does not. It is therefore preferable to perform the global fit for the individual branching
ratios, despite the fact that the predictions for the branching fractions have larger (and
correlated) theoretical uncertainties.
We now present two versions of such a fit. In the first method, we find the allowed region
for and by determining first the 95% confidence level contour for a given theoretical
input, and then scanning over all models in the theory parameter space that survive a 5%
confidence cut. The theory parameter space is given by the set of all input parameters
confined to the ranges specified in Table 2, variation of the renormalization scale between
mb /2 and 2mb , and scanning of the parameters XH and XA in the ranges described in (71)
and (72). The resulting 95% confidence region is bounded by the dark solid lines in Fig. 17.
2 points of all models that pass
The gray solid line inside this region envelopes the set of min
the 5% confidence cut. The dashed dark and gray lines have the same interpretation except
2 per degree of freedom of less than 1 are selected. The
that only theory models with a min
(,
)
range so determined is compared with the standard CKM fit (also at 95% confidence

level) [54], which uses information from semileptonic B decays (|Vub | and |Vcb |), KK

mixing ( K ), and BB mixing (HmBd , HmBs , and sin 2). The constraint corresponding
to |Vub /Vcb | = 0.085 0.017 is indicated by the solid annulus.
A fit to CP-averaged branching fractions cannot distinguish between and ,
and so
for simplicity only the upper half of the (,
)
plane is shown in the figure. At present our
contours are symmetric about the axis. In the future, data on direct CP asymmetries in
the various B K, modes should be included in the fit, in which case it will become
possible to obtain information about the sign of .
305
Before interpreting the results of this analysis, we discuss a second method for
performing the global fit, which follows the strategy outlined in [54]. It differs from the
method explained above in two ways. First, the large overall scale uncertainty (of almost
40%) due to the error on the value of the form factor F0B (0), which is common to
the theoretical predictions for all branching fractions (see, e.g., Eq. (93)), is eliminated
by fitting the form factor to the data. In that way, the sensitivity to the Wolfenstein
parameters in increased in the same way as it would be by using ratios; however, the
statistical problems of using ratios (see above) are avoided. Specifically, we start from
the 2 function
Ei STi 2
2
(S) =
(95)
,
i
i
where Ei are the experimental values for the branching ratios, i are their errors, and Ti are
the corresponding theoretical predictions obtained with the fixed form factor F0B (0) =
0.28. The scale factor S parameterizes the deviations of the product (F0B (0)|Vcb |)2 from
its central value. For a given set of theory parameters, we determine

Ei Ti /i2
S = i
(96)
2
i (Ti /i )
such that 2 (S) is minimized, provided that the value of S is allowed by the theory ranges
for the form factor (see Table 2) and |Vcb |. For most choices of theory parameters this
condition is satisfied. If it is not, we choose the value of S inside the allowed range that is
closest to the optimal value in (96). Next, we scan over all theory parameters and determine
)
plane. We then plot contours of constant H 2 =
the global minimum of 2 in the (,
2
2
min corresponding to the 68, 90 and 95% confidence levels. The second difference
with respect to our first scanning method is that these contours refer to a constant 2 ,
whereas in the first method the 2 is different for each of the theory models considered.
The results of this analysis are shown in Fig. 18. The best fit has an excellent 2 = 2.1 (for
3 degrees of freedom), corresponding to a confidence level of 54%. The parameters of the
particular theory model corresponding to this fit are all very reasonable. In particular, the
quantity XA parameterizing the weak annihilation terms is even smaller in magnitude than
our default value. (Neglecting weak annihilation contributions altogether we still obtain
an excellent fit with 2 = 2.7, and with the location of the minimum almost unchanged.)
Hence, a good global fit to the data can be obtained without using extreme choices of input
parameters, or invoking phenomenological recipes such as largely enhanced annihilation
or charm-penguin contributions. The default parameter set in Table 2 also yields a good
fit, which has 2 = 4.6 and 21% confidence level. Note that at 90% confidence level the
allowed range obtained by combining our results with the measurement of |Vub /Vcb | in
semileptonic B decays excludes = 0, thus establishing the existence of a CP-violating
phase in b u transitions.
The results of the two scanning methods shown in Figs. 17 and 18 are similar (though
the range obtained in the first method is somewhat more conservative). The allowed region
for and obtained from our analysis of rare hadronic B decays is compatible with the
306
Fig. 18. 95% (solid), 90% (dashed) and 68% (short-dashed) confidence level contours in the (,
)
plane obtained from a global fit to the CP-averaged B K, branching fractions, using the
scanning method of [54]. The dark dot shows the overall best fit, whereas the light dot indicates the
best fit for the default parameter set. See text for further explanations.
mixing and BB

standard global fit using information from semileptonic B decays, KK
mixing. However, the best fits to the rare-decay branching fractions prefer a larger value
of (of about 90 ). This has been noted by many other authors in the past (see, e.g.,
[66]), usually based on the naive factorization approximation and without a theoretical
error estimate. Here we have put this analysis on a firmer theoretical footing. On the other
hand, we see that small values of are also allowed, provided the value of |Vub | is lower
than its standard central value. Combining our results with the standard fit would reduce
the allowed region in the (,
)
plane by about a factor 2, indicating that even with present
experimental (and theoretical) errors the information obtained from rare hadronic decays
provides important constraints on the Wolfenstein parameters.
It will be interesting to follow how the comparison of the standard fit and the fit to
rare hadronic B decays will develop as the data become more precise. If the two fits
remain consistent with each other even as the allowed regions shrink in size, this would
constitute a highly nontrivial test of the CKM model, in which for the first time the phase
of the Vub matrix element (which enters through treepenguin interference) is restricted
by both direct and indirect measurements. At the same time, this would yield credibility to
the QCD factorization approach as a useful, quantitative tool to analyze rare hadronic B
decays. If, on the other hand, the two regions would not overlap at a reasonable confidence
level, this may suggest that the standard theory of weak interactions does not account
mixing amplitude (since this is what determines the leftcompletely for either the BB
most side of the allowed region in the standard fit) or the loop-induced flavour-changing
amplitudes in B K, decays. (Another option would be to abandon the theoretical
framework advocated here, but we understandably leave it to others to pursue this avenue.
Either way, the case for New Physics should be reinforced using the methods that are least
susceptible to large theoretical uncertainties, such as the strategies described in Sections 5.2
and 5.3.)
307
5.6. Predictions for CP asymmetries

The QCD factorization approach predicts the strong-interaction phases and hence
the direct CP asymmetries in the heavy-quark limit. In the following, we define the
asymmetries as
ACP (f ) =
f)
Br(B f ) Br(B
,
f)
Br(B f ) + Br(B
(97)
= B
0 , B contains a b quark (rather than antiquark). The dependence of the CP
where B
asymmetries on the weak phase is displayed in Fig. 19. The asymmetries are typically
predicted to be small, concurrent with the fact that strong-interaction phases are suppressed
in the heavy-quark limit [14]. Even the largest asymmetries for the K final states are
predicted to be 15% or less. An exception to this rule is the final state f = 0 0 , for
which the asymmetry is sensitive to QCD penguins and to the coefficient a2 , both of which
have large uncertainties and a potentially large relative phase.
While our approach predicts the generic magnitude of the CP asymmetries, their precise
values remain uncertain. This is not surprising, since in contrast to the branching ratios the
asymmetries are more sensitive to and in fact generated by corrections to naive
factorization. Hence, they are subject to larger relative uncertainties. In particular, the
CP asymmetries are proportional to the sines of strong-interaction phases, which are of
order s (mb ) or QCD /mb . Only the leading perturbative contributions to these phases are
calculable. On the contrary, the CP-averaged branching fractions depend on the cosines of
strong-interaction phases, which are equal to 1 in the heavy-quark limit.
First measurements of CP asymmetries have been published by the CLEO [67] and Belle
[50] Collaborations. All results are compatible with no asymmetry, with typical 1 errors
of about 20%. These errors are not yet small enough to draw meaningful conclusions,
except that very large asymmetries appear already improbable. In the future, accurate
measurements of CP asymmetries will test the generic prediction of the QCD factorization
approach that strong-interaction phases are small. If this prediction is confirmed, further
analysis of the CP asymmetries should help to determine the sign of , which cannot be
probed using CP-averaged branching fractions. In addition, data on CP asymmetries may
help to further constrain annihilation contributions and other theoretical input parameters
in the QCD factorization approach.
6. Comparison with other work

A general overview of the various approaches to the evaluation of the matrix elements in
exclusive hadronic B decays has been given in [15]. In that reference we have commented
on the strategies and methods that had been employed previously and explained how they
are related to QCD factorization. In this section we discuss recent papers on the subject
and clarify the differences with our work. We divide these papers into four categories:
those using our QCD factorization formalism, those using a perturbative PQCD approach,
308
Fig. 19. Direct CP asymmetries as functions of , assuming > 0. For negative , the signs of
the asymmetries are reversed. The meaning of the curves and bands is the same as in Fig. 13. The
asymmetries vanish in naive factorization, so no corresponding line is drawn.
those in which nonperturbative effects are estimated using light-cone sum rules or model
calculations, and finally those containing phenomenological analyses. We consider each of
these in turn.
Before discussing other approaches to nonleptonic B-decays it may be helpful to
summarize again the conceptual basis of the QCD factorization approach. The key
ingredient is the systematic analysis of Feynman graphs in the heavy-quark limit, from
which we deduce the factorization of infrared singularities into hadronic light-cone
distribution amplitudes and form factors. This enables us to establish the factorization
formula (25). The separation of long- and short-distance contributions to the decay
309
amplitudes, necessary to establish factorization, holds only to leading power in QCD /mb
and is based on considerations analogous to those used to demonstrate factorization
in other applications of QCD to hard processes (such as deep inelastic scattering,
DrellYan production, and electromagnetic form factors of hadrons at large momentum
transfer). The factorization formula leads to a model-independent treatment of exclusive
hadronic B decays in the heavy-quark limit. A consistent counting scheme for powers
of QCD /mb and a systematic identification of all the leading contributions are crucial
for establishing this result. The framework proposed in [14,15] is general and provides
a starting point for further theoretical developments, such as the improved understanding
of the nonperturbative input (e.g., the B form factor and the light-cone distribution
amplitudes) and estimates of power corrections.
6.1. Analyses within QCD factorization
Several authors have used the framework of QCD factorization for applications to twobody hadronic B decays [1823]. We now compare the results reported in these papers
with ours and comment on apparent differences and discrepancies.
Muta et al. have generalized the results of our previous work [14] by including
electroweak penguin contributions, and have applied the QCD factorization approach to the
decays B K, . We would like to point out the following differences with respect to
the present work:
The hard-scattering kernels are derived only for symmetric light-cone distribution
amplitudes.
In evaluating the s corrections to the penguin coefficients a4 and a6 , the existence
of the two distinct types of penguin contractions shown in Fig. 3 is not taken into
account. As discussed in Section 3.3, this leads to incorrect terms proportional to
s (C4 + C6 ) in these coefficients.
An incomplete projector for the twist-3, two-particle distribution amplitude of the
eff in a .
pion is employed. This gives an incorrect contribution proportional to C8g
6
We also observe a disagreement with the remaining terms in the correction of order
s to a6 , which concerns the function GM2 (sq ) in [18]. All three components, the
sq -dependent part, the constant, and the coefficient of the logarithm, differ from our
findings. We note that the -dependence of GM2 is in conflict with the requirement
of renormalization-group invariance of the product rM ()a6 (). In addition, the
contributions from the radiative vertex corrections to a6 and a8 are missing.
A minor difference comes from the fact that we neglect electroweak penguin
contributions to a4 and a6 , while these are included in [18]. On the other hand, Muta
eff to the electromagnetic
et al. omit the contributions proportional to C1 , C2 and C7
penguin coefficients a8 and a10 .
The twist-3 contributions to the spectator hard-scattering amplitudes and annihilation
effects are not discussed. As we have seen, these corrections are the potentially most
important source of theoretical uncertainty.
310
The formulae given in [18] have also been applied to discuss B decays into vector
pseudoscalar final states [19] and final states containing and [20]. The previous
comments apply also to these papers, as they do to [21], which relies on the same
expressions except for dropping the formally power-suppressed terms proportional to a6
and a8 .
In a series of papers, Du et al. have discussed B decays into two light pseudoscalar
mesons [22]. We focus on the two last papers in [22], which contain the most complete
and updated results. Similarly to the present work, electroweak penguin contributions and
chirally-enhanced twist-3 components of the pion distribution amplitude are included in
these papers, but no weak annihilation effects are considered. The hard-scattering kernels
are given only for symmetric distribution amplitudes, and explicit results are presented
for the case of B decays. We will therefore not distinguish between and
K final states in the present discussion. The formulae for the coefficients ai can be
directly compared with our results. A minor difference concerns the tiny electroweak
penguin contributions to the coefficients a1 , . . . , a6 , which we decided to neglect in our
approximation scheme, but which are retained in [22]. Next we recall from Section 3.3
that in the approximation of including twist-3 contributions only when they are chirally
enhanced, the equations of motion require the use of asymptotic distribution amplitudes
p , . Du et al. have considered a more general situation, which would affect the
cancellation of endpoint singularities and the renormalization-scale dependence of some
results. Use of the appropriate asymptotic distribution amplitudes eliminates these spurious
ambiguities. Finally, we note two minor discrepancies:
Our expression for the twist-3, hard spectator interaction in (57) contains a factor
2P M (x) p (y)
,
mb
x
y
(98)
in the integrand, whereas the corresponding result given in Eq. (37) of the third paper
in [22] has
2P M (x) (y)
.
mB
x
6y 2
(99)
Using the asymptotic forms of the twist-3 distribution amplitudes p and , we find
that the two results differ by a factor of y.
p
In our result for the twist-3 penguin contribution P3 in (54), the coefficient of C1 , for
example, is

mb
2
4
ln
+ G(s
(100)
p ).
3
3
The equivalent expression given in Eq. (23) of the third paper in [22] reads

7
1
mb
2
+
+ A G (sp ) G (sp ).
1 + A ln
3
12 2
(101)
Using asymptotic twist-3 distribution amplitudes, one may check that G (sp ) +
p ) and A = 1/2. The two results then agree up to a constant 1/6. This
G (sp ) = G(s
311
difference can be traced back to an inconsistent use of twist-3 projectors in four space
time dimensions within a d-dimensional loop calculation, before the subtraction of
ultraviolet poles is performed (see the last reference in [22]). Analogous differences
occur in the other terms in a6 and a8 .
Apart from these discrepancies, the expressions agree with our results for symmetric
distribution amplitudes.
Cheng and Yang have applied the QCD factorization approach to a study of the
decays B K [23]. Annihilation topologies are discussed and argued to be important
numerically. A few minor discrepancies with our results occur in the expressions for the
coefficients a3,...,10 . The penguin contraction of the operators Q4,6 is treated incorrectly, as
discussed in Section 3.3. Photonic penguin contractions of the operators Q1,2 contributing
to a8,10 are omitted. Also, QCD corrections to a6,8 are not considered, presumably because
the corresponding amplitudes are formally power suppressed. Similarly to the results of
[22], the twist-3 kernel for the hard spectator interaction contains an additional factor of y.
Finally, we briefly comment on the analysis of weak annihilation contributions. In [23] the
final state consists of a vector meson and a kaon rather than two pseudoscalar mesons. The
decay amplitudes are estimated in the approximation of using leading-twist distribution
amplitudes for the meson, but including chirally-enhanced twist-3 contributions for the
kaon. Taking this systematic difference into account, we can still compare the results of
[23] with our expressions. The two agree in the case of (V A) (V A) operators,
but the results for (S P ) (S + P ) operators are different. The latter are the ones that
depend on the twist-3 contributions, but we do not have sufficient information to trace back
the origin of the discrepancy.
6.2. The PQCD approach
Recently, Keum, Li and Sanda have presented an extensive study of B K decays
within a perturbative hard-scattering (or PQCD) approach [17]. While the underlying
goal of a separation of soft and hard physics in the B-decay matrix elements is similar in
spirit to QCD factorization, there are fundamental differences in the implementation of this
idea.
In the PQCD approach, the B and B K form factors are assumed to be
perturbatively calculable. This assumption is justified by invoking Sudakov effects to
regulate the infrared-sensitive contribution from the endpoint region in the integral over
the light-cone momentum of the outgoing spectator quark. In other words, the contribution
from the region where the spectator quark is soft is supposed to be strongly suppressed, and
therefore the exchange of a hard gluon is always required. Instead of being of leading order
(in the QCD coupling constant), as in the generic case of a B form factor dominated
by soft physics, the form factor is now counted as being of order s in perturbation theory.
Thus, the hierarchy of the various contributions to the decay amplitudes in PQCD is very
different from that in our approach:
312
In the PQCD approach, the B K matrix element is a quantity of order s . To

this order, therefore, all the Wilson coefficients can then be taken in the leading
logarithmic approximation.
The nonfactorizable hard gluon-exchange contributions to the kernels TiI in (25)
enter at order s2 and are therefore omitted. Note, however, that it is essential to
consider such effects in order to establish factorization. This has never been done
explicitly in the PQCD framework.
The form-factor terms containing the kernels TiI and the hard-scattering terms
containing the kernels TiII in the factorization formula are treated as being of the
same order in the PQCD approach. The former are called factorizable and the
latter nonfactorizable. Because both terms vanish simultaneously in the limit s
0, naive factorization is not recovered in any limit of QCD. This is in contrast to
our interpretation of the QCD factorization formula, for which naive factorization is
recovered in the heavy-quark limit.
After this brief synopsis of the key features of the PQCD scheme, we now examine
critically the main assumptions of this approach in the context of B K decays.
Importance of Sudakov effects and calculability of the B form factor
In the PQCD approach the transverse momenta of the quarks are kept explicitly when
evaluating contributions that are potentially infrared sensitive. For instance, the gluonexchange kernel for the B form factor is written as
H
(x3 m2b
1
+ ,
2

+ k 3 )[x1 x3 m2b + (k1 k3 )2 ]
(102)
where the ellipses represent less singular terms. Here x1 (x3 ) is the light-cone momentum
fraction of the spectator quark in the B meson (pion). In the conventional collinear
would be treated
expansion adopted in our work, the transverse momenta k 2 2
i
QCD
as higher-twist effects and dropped (more precisely, the amplitude would be expanded in
powers of transverse momenta). It then follows that H 1/x32 , and so the convolution
with theleading-twist pion distribution amplitude (x3 ) x3 (1 x3 ) would lead to an
integral dx3 /x3 , which is infrared divergent. This well-known result can be interpreted as
a sign of the dominance of soft endpoint contributions in the B form factor [68,69].
In [17] (see also [70]) this divergence is regulated by keeping the transverse momenta.
Then a Fourier transform from transverse momentum space into impact parameter space
(b space) is performed, and a Sudakov factor is included for each of the meson distribution
amplitudes. This factor strongly suppresses the region of large b, and the B form
factor is therefore assumed to be perturbatively calculable.
It should be noted that retaining the transverse momentum dependence in H , which
is only part of the complete higher-twist contribution, is a model-dependent procedure.
It is precisely in the critical endpoint region x3 0 that the leading two-particle Fockstate description that underlies this analysis breaks down. A further puzzle to us is the
assumption made in [70] that the spectator quark in the
B meson may have a large
momentum component in the direction, k1 = x1 mb / 2, when the pion momentum is
313
in the + direction. This is used to derive the Sudakov factor for the B meson in analogy
to the case of a fast light meson. Since the spectator quark in the B meson is intrinsically
soft, we see no justification for such an assumption.
In conclusion, we believe that the perturbative evaluation of the B form factor,
which is one of the central ingredients of the PQCD analysis in [17], is not justified. The
relevance of Sudakov form factors in hadronic B decays should be investigated in a more
systematic way. However, it seems unlikely to us that Sudakov logarithms are sufficiently
important at the scale mb to eliminate the soft contributions to heavy-to-light form factors
and thus render them calculable in a model-independent way. A complete description
of the B transitions at large recoil in the heavy-quark limit and the derivation of
a (hypothetical) factorization theorem for these processes has, to our knowledge, never
been presented. For recent progress in this direction and further critical discussions of the
problem see [33,71].
Dynamical enhancement of penguin contributions (fat penguins)
Keum et al. claim a dynamical enhancement of (factorizable) contributions to the matrix
elements of penguin operators (such as Q4 and Q6 ) in the effective weak Hamiltonian.
The first reason given in support of this assertion is a dynamical enhancement of the
B form factor of the scalar density ub
contained in (the Fierz-transformed form of)
Q6 , relative to the form factor of the vector current u
b relevant for other operators.
A dynamically different structure for the form factor of the scalar density in PQCD is
claimed, and a strong sensitivity of the effect to the model employed for the B-meson
light-cone distribution amplitude is noted. However, an enhancement of this nature is not
possible, because the form factor of the scalar density is related to that of the vector current
by the equations of motion, leading to
0
m2

B = F0B q 2 B .
+ ub
(103)
mb
The erroneous conclusion in [17] is the consequence of an incorrect treatment of the twist-3
contribution to the pion distribution amplitude (see below).
The second reason given in favor of a penguin enhancement is the choice of a low scale
in the (leading-order) penguin coefficients a4 () and a6 (). This is motivated by arguing
that should be a typical scale intrinsic to the dynamics of the B form factor in
the PQCD evaluation. However, the scale is unphysical and must be canceled by vertex
corrections to the operator matrix elements, and in the case of a6 () also by the running
quark masses. Such corrections have been properly included in the present work, but they
were neglected in [17]. The scale in a4,6 may depend on the scale of nonfactorizable
hard spectator interactions, but it is clearly independent of the internal dynamics of the
form factor. The arguments made in favour of a low scale and correspondingly increased
penguin coefficients are therefore not well justified.

Relevance of annihilation topologies

In their analysis, Keum et al. find that penguin annihilation contributions are numerically
very important and give a dominant contribution to some of the B K decay
314
amplitudes. We have investigated these effects in the present work and found them to give
a significant correction (compatible with being a power correction of canonical size), but
not a dominant contribution. Nevertheless, it must be emphasized that weak-annihilation
diagrams contribute at subleading power in the heavy-quark expansion and, in general, are
not calculable in a QCD-based factorization approach. Therefore, the question about the
relevance of these effects warrants further investigation.
Generation of large, calculable strong-interaction phases
It is claimed that large strong-interaction phases are generated by hard gluon exchange
with the spectator quark in the B meson, as well as by weak annihilation diagrams. Let
us consider the hard spectator interactions in detail. In that case the effect is calculable
within the QCD factorization approach and found to be real to leading order. The source
of the imaginary part found in [17] is as follows. The quark propagator entering the hardscattering diagram is written as
Hq
x2 x3 m2b
1
.

(k2 + k3 )2 + i
(104)
Working to leading power we would drop the transverse components and find a real
contribution to the kernel. Keum et al., on the other hand, keep the transverse momenta
and hence generate an imaginary part proportional to (k 2 x2 x3 m2b ), where k = k2 +
k3 . Again, this treatment is model dependent. The Fourier transform of the kernel into
impact parameter space results in a Bessel function with the imaginary part proportional
to mb bJ0 ( x2 x3 mb b). In the evaluation of the matrix element this function is convoluted
with the Sudakov factor. Since the Bessel function is oscillating rapidly, with the amplitude
of mb bJ0 growing like mb b, the result of this convolution is very sensitive to the details
of the b-space cut-off. This implies that the estimate of the strong-interaction phase is both
model dependent and numerically sensitive to effects that are poorly under control. Similar
comments hold for the estimate of the strong-interaction phases from the annihilation
contributions, which are generated in an analogous way.
Finally, we remark that the simple 5 structure of the twist-3 projection for the
pion employed by Keum et al. is incomplete. The proper treatment is discussed in
Section 3.2. The wrong twist-3 projection is, in particular, inconsistent with gauge
invariance. Moreover, the correct asymptotic behaviour of the twist-3 pion distribution
amplitude p (x) is proportional to a constant, whereas the functional form x(1 x)
is assumed in [17]. This problem affects all decay amplitudes, including the B form
factor, the corresponding spurious penguin enhancement, and the nonfactorizable spectator
interactions and annihilation contributions.
Other works using the PQCD approach
Other analyses in the spirit of the PQCD approach were presented in [7275], to which
similar comments apply. In [74], the presence of a recoil-phase effect was advocated,
which was claimed to originate within the PQCD framework. This phase should affect,
e.g., the B form factor, which was assumed to be dominated by hard gluon exchange.
315
It was argued that, when the gluon is exchanged between the spectator quark and the b
quark, the b-quark propagator could go on the mass shell because mB > mb + mspect . This
in turn would lead to a complex phase in the form factor. In our opinion this conclusion is
unwarranted, since bound-state effects have to be factorized before a perturbative treatment
can be justified and thus cannot influence the hard-scattering process. Another fundamental
objection to the recoil phase is that it would contradict the fact that the B form
factor is a real quantity.
6.3. Estimates of nonperturbative effects
Recently, Khodjamirian has suggested to study hadronic matrix elements for two-body
B decays in the framework of light-cone QCD sum rules [44]. As an example, he discusses
the matrix elements of the currentcurrent operators Qu1 and Qu2 for B + decays.
The idea is to generalize the light-cone QCD sum-rule analysis of the B form factor
directly to the case of hadronic two-body modes. The starting point is the correlation
function

F (p, q, k) = d 4 x ei(pq)x

d 4 y ei(pk)y 0|T j() (y)Qui (0)j (B) (x) (q) , (105)
()
5 d are interpolating currents for the emission
5 d and j (B) = mb bi
where j = u
pion and the B meson, respectively. The explicit pion state (q) represents the recoil
pion that absorbs the spectator quark. According to the QCD sum rule philosophy, the
correlator is evaluated in two ways: by a direct calculation in QCD, and by inserting
complete sets of hadronic states between the operator Qui and the interpolating currents,
extracting the desired ground-state contribution with the help of quarkhadron duality. The
two sides of the equation are expressed in the form of dispersion relations, and the usual
Borel transformation is applied. To leading order in s , the factorized result for the matrix
element is recovered. A particular type of power correction is then estimated as a further
illustration. This contribution comes from higher twist, three-particle (quarkantiquark
gluon) Fock components of the recoil pion, where the (nonfactorizable) gluon couples to
the quark lines of the emission current. In the heavy-quark limit the resulting expression
is demonstrated to scale as QCD /mb relative to the leading contribution. It is estimated
to be of relative size E /mB with E 0.050.15 GeV. Note that this correction has
no rescattering phase. Its numerical effect is small, but comparable to the perturbative
corrections at leading power. Additional nonfactorizable contributions exist but have not
yet been investigated. Examples are the gluon-exchange effects that correspond to the
vertex corrections and the spectator scattering diagrams in the QCD factorization formula.
In the sum-rule approach these effects are in principle contained in similar diagrams,
together with certain hadronic corrections of subleading power. A potential difficulty will
be to disentangle power corrections to the asymptotic result given by the QCD factorization
formula from uncertainties intrinsic to the sum rule method, such as the assumption of
316
quarkhadron duality and the approximation of the emission pion by an interpolating

current. It is not fully clear how this can be achieved in a controlled and systematic fashion.
Some recent papers [7678] have tried to model soft final-state interactions via the
(in Ref. [77] the author
rescattering of certain hadronic channels such as B D D
models inelastic rescattering in more general terms). We consider such an approach to
be problematical for a variety of reasons. There are many more intermediate channels
beyond those usually taken into account. Systematic cancellations among these channels,
which are predicted to occur in the heavy-quark limit, are missed when only a subset
of intermediate states is retained (see the discussion in Sections 3.2 and 7.2 of [15]).
Moreover, the hadronic dynamics of multi-body decays is very complicated and in general
not under theoretical control. The main problem of such models is the lack of a systematic
approximation scheme based on parametric expansions. In our opinion, the use of a purely
hadronic language, suitable for kaon decays, is not very helpful in the case of B decays,
where the number of channels and the energy release are large.
In [76], the coefficients ai obtained from the QCD factorization approach have been used
in conjunction with a hadronic description of final-state interactions. The general caveats
concerning hadronic rescattering models apply also here. Moreover, such an approach faces
a manifest double-counting problem, since rescattering effects in the heavy-quark limit are
already contained in the QCD coefficients ai .
A nonperturbative treatment of charming penguins, matrix elements with charm loops
of the type shown in Fig. 3, was proposed in [79]. In this calculation operators with flavour
structure (cb)(
s c) were split into nonlocal products of currents (cb)
and (s c). These were
connected by intermediate virtual D-meson states to yield transitions such as B D
K under the separate action of the currents. However, the operators (cb)(
s c) are local
operators in the effective theory at the b-mass scale. They would become nonlocal only if
probed at the far higher scale of MW , but not at a scale of order mD , as implied in [79]. We
therefore see no justification for the method adopted in this paper.
6.4. Phenomenological analyses
There are several recent analyses in the literature that approach hadronic B decays in a
more phenomenological way. Some of the most extensive studies of this type have been
presented by Ali et al. [80], who elaborate on the concept of generalized factorization
[81]. In these analyses, part of the order-s vertex and penguin contributions calculated
in the present work are included. As a consequence, the decay amplitudes receive stronginteraction phases due to the BanderSilvermanSoni mechanism [37], and nonzero CP
asymmetries are obtained. However, no systematic attempt to include all such effects (or
to prove factorization) is made. Also, these authors rely on ad hoc model assumptions
such as an effective number of colours Nc = 3, which is introduced to parameterize
nonfactorizable corrections.
A recent study of B K decays presented in [82] emphasizes general, modelindependent parameterizations of the decay amplitudes such as our parameterization in
(18), avoiding as much as possible the use of dynamical input. For instance, strong-
317
interaction phases are a priori allowed to take any value in the construction of bounds
on CKM parameters. Information on these phases may then be inferred indirectly. Such
an approach is complementary to the more ambitious goal of exploiting theoretical
insight into the hadronic matrix elements. In fact, the smallness of the rescattering effects
parameterized by a eia in (18), which is a prediction of the QCD factorization approach,
represents a useful input to such a phenomenological analysis.
Many other phenomenological studies of two-body hadronic B decays have been
performed in the literature. Recent examples are [8388], where simplifying assumptions
are made and no systematic estimates of theoretical uncertainties are undertaken.
7. Conclusions
In this paper we have presented a detailed study of B K and B decays
based on the QCD factorization formula. This approach allows us to perform a systematic
and model-independent calculation of two-body hadronic B decays in the heavy-quark
limit. We have evaluated the hard-scattering kernels entering the matrix elements for
B K, decays at next-to-leading order in s , thus obtaining predictions for the
decay amplitudes including the leading, nonfactorizable corrections. We have included
the contributions from electroweak penguins, taking into account the corresponding
2 or 1/ sin2 ) in
order-s corrections to the dominant terms (enhanced by m2t /MW
W
a consistent approximation scheme. All hard-scattering kernels have been derived for
general, asymmetric distribution amplitudes, as appropriate for K mesons.
In addition to computing the model-independent leading-twist results, we have identified
and estimated those power corrections which are expected to be the largest. We have
analyzed the complete set of contributions from light-cone wave functions of twist-3
with a chiral enhancement factor m2 /(mu + md ) or m2K /(mq + ms ), as well as the
contributions from weak annihilation topologies, including both twist-2 and twist-3 twoparticle components in the light-meson wave functions. We distinguish three classes of
power corrections: the contributions from operators with (pseudo-) scalar currents ( a6,8 ),
twist-3 effects in the hard spectator interactions, and weak annihilation corrections. While
the first two effects are proportional to the chiral enhancement factor, the annihilation terms
are quadratic polynomials in this factor. Since these are power-suppressed effects, we do
not expect the factorization formula to work in these cases. In fact, naively evaluating these
terms in a hard-scattering framework one encounters logarithmic endpoint singularities.
(Incidentally, the terms of order s related to a6,8 are free of such infrared singularities,
but divergent terms are present for the other two contributions.) Therefore, while our
results at leading twist are model-independent predictions of QCD in the heavy-quark
limit, some model dependence is currently unavoidable in the description of power
corrections. We choose to regulate these divergences by introducing a simple cut-off,
leading to the complex phenomenological parameters XH and XA . The motivation for this
model-dependent approach to power corrections is twofold: we obtain systematic orderof-magnitude estimates, while automatically keeping track of relative suppressions or
318
enhancements from Wilson coefficients and CKM factors. In addition, we can investigate
the sensitivity of the observables on the model parameters. It turns out that annihilation
graphs may in principle give sizable corrections. On the other hand, twist-3 effects in the
hard spectator contributions appear generally to be less prominent.
Using this framework, we have performed a detailed comparison with the available
experimental data on B K and B branching ratios. The data are beginning
to be sufficiently precise to allow for detailed phenomenological analyses. An important
result is that in the QCD factorization approach the B K branching fractions can
be larger than the B ones without any tuning of the theoretical input parameters,
and without invoking large phenomenological power corrections. An acceptable fit to the
branching fractions is obtained even if we impose that < 90 as implied by the standard
constraints on the unitarity triangle. This is in contrast to the naive factorization model,
in which the observed branching fractions can either not be reproduced at all or require a
large CKM angle , in conflict with other indirect determinations. Encouraged by this
success, we have derived constraints in the (,
)
plane from a global fit to the B
K, branching ratios, which already provide useful additional information on the
unitarity triangle. Clearly, the experimental situation will continue to improve and allow
us to further test our framework and exploit it in phenomenological studies of hadronic
B decays. An important role in these analyses will be played by the tree-to-penguin ratio
3/2 , which can be reliably determined from B 0 and B K 0 decays and
can be unambiguously confronted with the theoretical predictions. This will be a valuable
check that the power corrections are not surprisingly large. The absence of penguin and
annihilation contributions in the B 0 mode is a crucial feature in this analysis.
A more precise measurement of this decay will thus be particularly useful.
We would like to conclude by emphasizing again the strategy underlying our treatment
of weak annihilation contributions and nonleading, but chirally-enhanced spectator
interactions. While the present data do not require these effects to be larger than expected,
the data also do not definitively exclude this possibility. Our error analysis is based on
allowing the parameters IA and IH to be smaller than 1 (reflecting the range of our
expectations), but the error estimates sometimes depend sensitively on the precise choice
of this upper limit, in particular in the case of IA . With more precise data one may
contemplate fitting IA to data (in practice this implies fitting the QCD penguin amplitude)
in order to decide upon the plausibility of the adopted range of values. The discovery of
large annihilation contributions would by itself constitute valuable insight into the stronginteraction dynamics of nonleptonic decays, but would evidently limit the utility of the
QCD factorization approach. The present data make this scenario appear unlikely, but we
look forward to further information to validate our error estimation. In this case, important
and reliable information on flavour physics will become available from B K and
B decays, as well as from a large class of other two-body hadronic decays into
light pseudoscalar and vector mesons.
319
Acknowledgements
We are grateful to A. Hcker and H. Lacker for helpful discussions and for providing us
with a data file of their CKM fit results. We also thank G. Martinelli and L. Silvestrini
for useful discussions. The research of M.N. is supported in part by the National
Science Foundation. C.T.S. acknowledges partial support from PPARC through Grant No.
GR/K55738.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
M. Neubert, J.R. Rosner, Phys. Rev. Lett. 81 (1998) 5076, hep-ph/9809311.

R. Fleischer, Phys. Lett. B 365 (1996) 399, hep-ph/9509204.
M. Neubert, J.R. Rosner, Phys. Lett. B 441 (1998) 403, hep-ph/9808493.
M. Neubert, JHEP 02 (1999) 014, hep-ph/9812396;
M. Neubert, Nucl. Phys. B Proc. Suppl. 86 (2000) 477, hep-ph/9909564.
R. Fleischer, Eur. Phys. J. C 6 (1999) 451, hep-ph/9802433;
A.J. Buras, R. Fleischer, Eur. Phys. J. C 11 (1999) 93, hep-ph/9810260.
R. Fleischer, T. Mannel, Phys. Rev. D 57 (1998) 2752, hep-ph/9704423.
R. Fleischer, Phys. Lett. B 459 (1999) 306, hep-ph/9903456.
M. Gronau, D. London, Phys. Rev. Lett. 65 (1990) 3381.
J.P. Silva, L. Wolfenstein, Phys. Rev. D 49 (1994) 1151, hep-ph/9309283.
Y. Grossman, H.R. Quinn, Phys. Rev. D 58 (1998) 017504, hep-ph/9712306.
J. Charles, Phys. Rev. D 59 (1999) 054007, hep-ph/9806468.
A.E. Snyder, H.R. Quinn, Phys. Rev. D 48 (1993) 2139.
Z. Ligeti, M. Luke, M.B. Wise, Preprint LBNL-47551, hep-ph/0103020.
M. Beneke, G. Buchalla, M. Neubert, C.T. Sachrajda, Phys. Rev. Lett. 83 (1999) 1914, hepph/9905312.
M. Beneke, G. Buchalla, M. Neubert, C.T. Sachrajda, Nucl. Phys. B 591 (2000) 313, hepph/0006124.
M. Beneke, G. Buchalla, M. Neubert, C.T. Sachrajda, Preprint PITHA-00-13, hep-ph/0007256,
in: Proceedings of the 30th International Conference on High-Energy Physics (ICHEP 2000),
Osaka, Japan, 27 July2 August, 2000, in press.
Y.Y. Keum, H.-N. Li, A.I. Sanda, Preprint KEK-TH-642, hep-ph/0004004;
Y.Y. Keum, H.-N. Li, A.I. Sanda, Phys. Rev. D 63 (2001) 054008, hep-ph/0004173;
Y.Y. Keum, H.-N. Li, Phys. Rev. D 63 (2001) 074006, hep-ph/0006001.
T. Muta, A. Sugamoto, M.Z. Yang, Y.D. Yang, Phys. Rev. D 62 (2000) 094020, hepph/0006022.
M.Z. Yang, Y.D. Yang, Phys. Rev. D 62 (2000) 114019, hep-ph/0007038.
M.Z. Yang, Y.D. Yang, Preprint OCHA-PP-166, hep-ph/0012208.
X.G. He, J.P. Ma, C.Y. Wu, Phys. Rev. D 63 (2001) 094004, hep-ph/0008159.
D.-S. Du, D.-S. Yang, G.-H. Zhu, Phys. Lett. B 488 (2000) 46, hep-ph/0005006;
D.-S. Du, D.-S. Yang, G.-H. Zhu, hep-ph/0008216;
D.-S. Du, D.-S. Yang, G.-H. Zhu, hep-ph/0102077;
D.-S. Du, D.-S. Yang, G.-H. Zhu, hep-ph/0103211.
H.-Y. Cheng, K.-C. Yang, hep-ph/0012152.
J.D. Bjorken, Nucl. Phys. B Proc. Suppl. 11 (1989) 325.
M.J. Dugan, B. Grinstein, Phys. Lett. B 255 (1991) 583.
H.D. Politzer, M.B. Wise, Phys. Lett. B 257 (1991) 399.
320
[27] A.J. Buras, M. Jamin, M.E. Lautenbacher, Nucl. Phys. B 400 (1993) 75, hep-ph/9211321.
[28] M. Ciuchini, E. Franco, G. Martinelli, L. Reina, Phys. Lett. B 301 (1993) 263, hep-ph/9212203;
M. Ciuchini, E. Franco, G. Martinelli, L. Reina, Nucl. Phys. B 415 (1994) 403, hep-ph/9304257.
[29] N.G. Deshpande, X.-G. He, Phys. Rev. Lett. 74 (1995) 26;
N.G. Deshpande, X.-G. He, Phys. Rev. Lett. 74 (1995) 4099, Erratum, hep-ph/9408404.
[30] G. Buchalla, A.J. Buras, M.E. Lautenbacher, Rev. Mod. Phys. 68 (1996) 1125, hep-ph/9512380.
[31] A.J. Buras, P. Gambino, U.A. Haisch, Nucl. Phys. B 570 (2000) 117, hep-ph/9911250.
[32] A.G. Grozin, M. Neubert, Phys. Rev. D 55 (1997) 272, hep-ph/9607366.
[33] M. Beneke, T. Feldmann, Nucl. Phys. B 592 (2001) 3, hep-ph/0008255.
[34] V.M. Braun, I.E. Filyanov, Z. Phys. C 44 (1989) 157;
V.M. Braun, I.E. Filyanov, Z. Phys. C 48 (1990) 239.
[35] B.V. Geshkenbein, M.V. Terentev, Phys. Lett. B 117 (1982) 243;
B.V. Geshkenbein, M.V. Terentev, Sov. J. Nucl. Phys. 40 (1984) 487, Yad. Fiz. 40 (1984) 758,
(in Russian);
B.V. Geshkenbein, M.V. Terentev, Yad. Fiz. 39 (1984) 873.
[36] T. Becher, B.D. Pecjak, M. Neubert, Preprint CLNS 01-1723, hep-ph/0102219.
[37] M. Bander, D. Silverman, A. Soni, Phys. Rev. Lett. 43 (1979) 242.
[38] G.P. Korchemsky, D. Pirjol, T.-M. Yan, Phys. Rev. D 61 (2000) 114510, hep-ph/9911427.
[39] V. Chernyak, A. Zhitnitsky, Nucl. Phys. B 201 (1982) 492;
V. Chernyak, A. Zhitnitsky, Nucl. Phys. B 214 (1983) 547, Erratum.
[40] A. Khodjamirian et al., Phys. Rev. D 62 (2000) 114002, hep-ph/0001297;
A. Khodjamirian, R. Rckl, in: A.J. Buras, M. Lindner (Eds.), Heavy Flavours II, World
Scientific, Singapore, 1998, p. 345, hep-ph/9801443.
[41] P. Ball, JHEP 09 (1998) 005, hep-ph/9802394.
[42] D. Melikhov, B. Stech, Phys. Rev. D 62 (2000) 014006, hep-ph/0001113.
[43] A. Abada et al., Nucl. Phys. B Proc. Suppl. 83 (2000) 268, hep-lat/9910021.
[44] A. Khodjamirian, Preprint CERN-TH-2000-325, hep-ph/0012271.
[45] C.N. Burrell, A.R. Williamson, Preprint UTPT-01-02, hep-ph/0101190.
[46] A.F. Falk, A.L. Kagan, Y. Nir, A.A. Petrov, Phys. Rev. D 57 (1998) 4290, hep-ph/9712225.
[47] M. Gronau, J.L. Rosner, Phys. Rev. D 59 (1999) 113002, hep-ph/9809384.
[48] D. Cronin-Hennessy et al., CLEO Collaboration, Phys. Rev. Lett. 85 (2000) 515.
[49] G. Cavoto, BaBar Collaboration, Talk at the 36th Rencontres de Moriond on QCD and Hadronic
Interactions, Les Arcs, France, 1724 March, 2001, to appear in the Proceedings.
[50] T. Iijima, Belle Collaboration, Talk at the 4th International Conference On B Physics And CP
Violation (BCP4), Ise-Shima, Japan, 1923 February, 2001, to appear in the Proceedings.
[51] A. Ali, D. London, Eur. Phys. J. C 9 (1999) 687, hep-ph/9903535.
[52] S. Plaszczynski, M.-H. Schune, Talk given at 8th International Symposium on Heavy Flavour
Physics, Southampton, UK, 2529 July, 1999, Preprint LAL-99-67, hep-ph/9911280.
[53] M. Ciuchini et al., Preprint LAL-00-77, hep-ph/0012308.
[54] A. Hcker, H. Lacker, S. Laplace, F. Le Diberder, Preprint LAL-01-06, hep-ph/0104062.
[55] M. Gronau, J.L. Rosner, D. London, Phys. Rev. Lett. 73 (1994) 21, hep-ph/9404282.
[56] J. Beringer, BaBar Collaboration, Talk at the 36th Rencontres de Moriond on QCD and
Hadronic Interactions, Les Arcs, France, 1724 March, 2001, to appear in the Proceedings.
[57] G. Buchalla, A.J. Buras, Phys. Rev. D 54 (1996) 6782, hep-ph/9607447.
[58] A.L. Kagan, M. Neubert, Phys. Lett. B 492 (2000) 115, hep-ph/0007360.
[59] J.P. Silva, L. Wolfenstein, Phys. Rev. D 63 (2001) 056001, hep-ph/0008004.
[60] G. Eyal, Y. Nir, G. Perez, JHEP 0008 (2000) 028, hep-ph/0008009.
[61] Z.-Z. Xing, hep-ph/0008018.
[62] A.J. Buras, R. Buras, Phys. Lett. B 501 (2001) 223, hep-ph/0008273.
[63] M. Ciuchini, E. Franco, G. Martinelli, L. Silvestrini, Nucl. Phys. B 501 (1997) 271, hepph/9703353.
[64]
[65]
[66]
[67]
[68]
[69]
[70]
[71]
[72]
[73]
[74]
[75]
[76]
[77]
[78]
[79]
[80]
[81]
[82]
[83]
[84]
[85]
[86]
[87]
[88]
321
M. Ciuchini et al., hep-ph/0104018.

W.-S. Hou, K.-C. Yang, Phys. Rev. Lett. 84 (2000) 4806, hep-ph/9911528.
W.-S. Hou, J.G. Smith, F. Wrthwein, Preprint NTU-HEP-99-25, hep-ex/9910014.
S. Chen et al., CLEO Collaboration, Phys. Rev. Lett. 85 (2000) 525, hep-ex/0001009.
A. Szczepaniak, E.M. Henley, S.J. Brodsky, Phys. Lett. B 243 (1990) 287.
G. Burdman, J.F. Donoghue, Phys. Lett. B 270 (1991) 55.
H.-N. Li, H.-L. Yu, Phys. Rev. D 53 (1996) 2480, hep-ph/9411308.
A.P. Szczepaniak, A. Radyushkin, Phys. Rev. D 59 (1999) 034017, hep-ph/9810438.
C.-D. Lu, K. Ukai, M.-Z. Yang, Phys. Rev. D 63 (2001) 074009, hep-ph/0004213.
C.-H. Chen, H.-N. Li, Phys. Rev. D 63 (2001) 014003, hep-ph/0006351.
B.F.L. Ward, hep-ph/0006357.
C.-D. Lu, M.-Z. Yang, Preprint HUPD-0012, hep-ph/0011238.
Z.-Z. Xing, Phys. Lett. B 493 (2000) 301, hep-ph/0007136.
P. Zenczykowski, Phys. Rev. D 63 (2001) 014016, hep-ph/0009054.
K. Terasaki, Preprint YITP-00-65, hep-ph/0011358.
C. Isola et al., Preprint BA-TH-404-00, hep-ph/0101118.
A. Ali, C. Greub, Phys. Rev. D 57 (1998) 2996, hep-ph/9707251;
A. Ali, G. Kramer, C.-D. Lu, Phys. Rev. D 58 (1998) 094009, hep-ph/9804363;
A. Ali, G. Kramer, C.-D. Lu, Phys. Rev. D 59 (1999) 014005, hep-ph/9805403.
M. Bauer, B. Stech, M. Wirbel, Z. Phys. C 34 (1987) 103;
M. Neubert, B. Stech, in: A.J. Buras, M. Lindner (Eds.), Heavy Flavours II, World Scientific,
Singapore, 1998, p. 294, hep-ph/9705292.
A.J. Buras, R. Fleischer, Eur. Phys. J. C 16 (2000) 97, hep-ph/0003323.
H.-Y. Cheng, K.-C. Yang, Phys. Rev. D 62 (2000) 054029, hep-ph/9910291.
Y.-L. Wu, Y.-F. Zhou, Phys. Rev. D 62 (2000) 036007, hep-ph/0002227.
Y.F. Zhou, Y.L. Wu, J.N. Ng, C.Q. Geng, Phys. Rev. D 63 (2001) 054011, hep-ph/0006225.
X.H. Guo, O. Leitner, A.W. Thomas, Phys. Rev. D 63 (2001) 056012, hep-ph/0009042.
C. Isola, T.N. Pham, Preprint CPHT-S-085-0900, hep-ph/0009210.
W.N. Cottingham, H. Mehrban, I.B. Whittingham, hep-ph/0102012.

Probability distribution function of the diquark

condensate in two colours QCD
R. Aloisio a,d , V. Azcoiti b , G. Di Carlo c,d , A. Galante a,d , A.F. Grillo d
a Dipartimento di Fisica dellUniversit di LAquila, 67100 LAquila, Italy
b Departamento de Fsica Terica, Facultad de Ciencias, Universidad de Zaragoza, 50009 Zaragoza, Spain
c Istituto Nazionale di Fisica Nucleare, Laboratori Nazionali di Frascati, P.O.B. 13 - 00044 Frascati, Italy
d Istituto Nazionale di Fisica Nucleare, Laboratori Nazionali del Gran Sasso, 67010 Assergi (LAquila), Italy
Received 30 November 2000; accepted 9 May 2001
Abstract
We consider diquark condensation in finite density lattice SU(2). We first present an extension
of VafaWitten result, on spontaneous breaking of vector-like global symmetries, that allows us to
formulate a no-go theorem for diquark condensation in a region of the chemical potentialmass
parameter space. We then describe a new technique to calculate diquark condensate in unquenched
lattice simulations with any number of flavours directly at zero external source, without using any
potentially dangerous extrapolation procedure. We apply it to the strong coupling limit and find
compelling evidences for a second order transition to a phase with nonzero diquark condensate.
Our results are in quantitative agreement with low-energy effective Lagrangian calculations. 2001
PACS: 11.15.Ha
1. Introduction
Recently the expected scenario for the phase diagram of QCD in the chemical potential
temperature plane has changed: besides the hadronic and quarkgluon plasma phases, the
existence of a new state of matter has been proposed by several groups [1]. This new phase
is characteristic of the high density, low temperature regime: the asymptotic freedom of
QCD and the known instability of large Fermi spheres, in presence of (whatever weak)
attractive forces, results in a pairing of quarks with momenta near the Fermi surface (in
analogy with the Cooper pairing in solid state systems at low temperature). A condensation
of quark pairs should be the distinctive signal of the new phase and has been indeed
predicted using simplified phenomenological models of QCD.
E-mail address: galante@lngs.infn.it (A. Galante).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 2 3 2 - 2
R. Aloisio et al. / Nuclear Physics B 606 (2001) 322336
323
Unfortunately the lattice approach, the most powerful tool to perform first principles,
nonperturbative studies, is affected in the case of finite density QCD by the well known
sign problem that has prevented until now any step towards the understanding of this new
phase. This is not the case for the SU(2) theory where the fermions are in the pseudo-real
representation and finite density numerical simulations are feasible [2].
In this paper we present a detailed study of diquark condensation for the unquenched
two colours model. This work follows a previous one where we considered chiral and
diquark susceptibilities and found strong evidence for a phase transition separating the
ordinary low density phase from a high density one where the chiral condensate vanishes
and the baryon number symmetry is broken [3]. In the present work we mainly focus on an
approach, based on the analysis of the probability distribution function (p.d.f.) of the order
parameter for Grassmann fields [6], to compute the value of the order parameter directly
at zero external source without using any potentially dangerous extrapolation procedure,
mandatory in the standard approach.
This technique has also another relevant feature. Since configurations are generated
directly at zero diquark source, the sign ambiguity in the definition of the fermionic
partition function is avoided, with the consequence that we are not restricted to nf = 8
but any number of flavours can be in principle simulated.
The study of diquark condensation in two colours QCD is an interesting topic by itself
and we can hope to use some of the results to get insights about the three colours case. We
are interested in the similarities between the phase diagram of SU(3) and SU(2), however
we have to be aware of the differences between the two models:
The quarkquark condensate qq is coloured for the Nc = 3 theory and colourless
for Nc = 2. This implies that in the SU(3) case the diquark condensation has to be
interpreted as a Higgs-like mechanism leading to the breaking of the local colour
symmetry, while in the SU(2) case we have spontaneous symmetry breaking (SSB) of
the UB (1) global symmetry associated with conservation of baryon number.
The zero temperature critical chemical potential, related to the mass of the lightest
baryonic state, is different in the two cases: it is expected to be 1/3 of the nucleon
mass for SU(3) and 1/2 of the pion mass for SU(2) (i.e., zero at vanishing quark
mass). For Nc = 2, in order to have a phase diagram similar to that of SU(3), it is
appropriate to consider a nonzero bare quark mass.
The lightest baryonic state is a fermion for the tree colours model and a boson for the
two colours one.
Our results can be used to do a highly nontrivial, quantitative check of the continuum
predictions for the SU(2) theory, as for instance those from low energy effective Lagrangian
calculations [4].
In the next section we present an extension, to SU(2) gauge theory at finite chemical
potential, of the VafaWitten theorem on the impossibility to break spontaneously baryon
number conservation in a vector-like theory [5], finding an analytical bound, useful to
check approximate models. In Section 3 we give a review of the p.d.f. formalism for
Grassmann degrees of freedom, emphasizing on the details of its application to the analysis
324
of the order parameter for the U(1) symmetry associated to baryon number conservation.
In Section 4 we present some details of the numerical algorithm used for the unquenched
simulation. The last section is devoted to the presentation and discussion of numerical
results.
2. The VafaWitten theorem at finite

In their original paper [5] Vafa and Witten proved that the two point correlation function
of any operator with nonzero isospin or baryon number (or any other conserved quantity
carried by fermions but not by gauge bosons) falls off exponentially at large distances.
From this result they concluded that any order parameter for such vector-like global
symmetries is zero and that massless bound states can only be built by zero bare mass
constituents.
The two main ingredients that are necessary to the proof are (i) the antihermiticity of
the massless Dirac operator (to put a bound on the fermion propagator for a given gauge
background) and (ii) the positivity of the integration measure (used to extend the above
bound to the averaged correlation function). These conditions limit the applicability of the
theorem to vector-like theories (Yukawa coupling generally invalidates the argument) at
= 0 and zero chemical potential.
We are here interested to the case in which the CP-violating vacuum angle vanishes
but the chemical potential is nonzero. While for SU(3) this breaks both conditions, for
two colours QCD only condition (i) is not fulfilled. We will show how the theorem can
be extended to the latter case for a range of values of that depends on the bare quark
mass. Such new version of the theorem will allow us to exclude spontaneous breaking of
baryon number (i.e., the formation of a diquark condensate that breaks baryon number
conservation) for some region of the parameter space.
In this work we consider the theory regularized on the lattice. In the following we assume
an infinite lattice because we are interested in finding a bound for the quark propagator
at large distances, in arbitary background gauge field. The quark propagator is an object
which has a well defined thermodynamical limit as the limit of a finite box with periodic
(or antiperiodic) boundary conditions. Therefore our results hold for both periodic and
antiperiodic boundary conditions, provided one takes first the thermodynamic limit and
then compute the bound for the quark propagator.
On the lattice there is no need to consider smeared operators as in the original paper and
we have to prove a uniform upper bound for the fermion propagator

x|1 ()|0 exp(|x|),
(1)
where () is the Dirac operator on the lattice in presence of a chemical potential term
and , are constants. () can be written as the sum of the Dirac operator at zero
chemical potential ((0) = i + m) plus a term containing all the dependence: () =
(0) + (). The matrix () = (e 1)G + (1 e )G contains only the forward
(G) and backward (G ) temporal links and is bounded: () 2 sinh . Expanding the
325
inverse of the Dirac operator 1

1
1 () = I + 1 (0)() 1 (0)
= 1 (0) 1 (0)()1 (0) + 1 (0)()1 (0)()1 (0)
+
(2)
and using an integral representation for 1 (0) we get the following series expansion for
the fermion propagator:
x|

()|0 =
d0 em0 x|ei0 |0
d0 d1 em(0 +1 ) x|ei0 ()ei1 |0

+
0
d0 d1 d2 em(0 +1 +2 )
x|ei0 ()ei1 ()ei2 |0 + .
(3)
The expectation values in the r.h.s. of (3) can be bounded depending on the value of the
integration variables: since
n k

Pn (x, 0) = x|ei0 () ein |0 ()
(0 + + n )k
k!
kx
we get the relations

()n ex ,
Pn (x, 0)

()n
0 + + n xe2 ,
(4)
otherwise.
If we take the modulus of expression (3) and use relations (4) we get

x|1 ()|0
qn +
rn ,
(5)
where
() n x
e
,
mn+1
n

rn ()
d0 dn em(0 +1 ++n ) ,
qn
Dn
and Dn is the set of points that satisfy the condition 0 + + n x exp(2). Clearly

qn is bounded by a geometric series

1
if () < m.
qn ex
(6)
m ()
1 The expansion is valid if 1 (0)() < 1. Using 1 (0) m1 and the bound on () the above
condition is fulfilled if m > 2 sinh . At the end we will check that our result is inside this region.
326
We have to find an (exponential) bound for the other term in the r.h.s. of (5). Since

1
2
I0 = d0 em0 = emxe
m
D0
it is possible to calculate

In = d0 dn em(0 ++n )
Dn
using the change of variables

0 + 1 + + n = t0 ,
1 + + n = t1 ,
..
.
n = tn .
It is easy to check that
In =
(1)n d n I0
,
n! dmn
and finally

n
rn
(1)k
k=0
() k d n I0
.
k!
dmn
(7)
Note that I0 is an analytic function of m for m = 0 so the r.h.s. of (7) is equal to the Taylor
expansion of I0 (m ()) around I0 (m) provided m > ().
Putting together the last result with (6) and the condition for the Taylor expansion in (2)
we get the final bound

x

1
x|1 ()|0
e
+ e(m())x exp(2) ,
(8)
m ()
valid for any m > (). Since () 2 sinh we conclude that the fermionic
correlation function decays exponentially at least for any
m > 2 sinh .
(9)
Note that in the 0 limit we recover the original result of Vafa and Witten, i.e., the
propagator goes to zero exponentially for any nonzero mass.
The formation of a diquark condensate, as follows from (9), is excluded for < =
sinh1 (m/2). Near the chiral limit, is smaller than half the pion mass so condition (9) is
not fulfilled for the expected critical point c = m /2.
Our result is interesting because it possess a rigorous limit to diquark condensation and
this limit can be used as a check to phenomenological approximations to the model. It
also shows how the VafaWitten theorem on spontaneous breaking of vector-like global
symmetries can be extended, for some region of the parameter space, to a theory at nonzero
chemical potential.
327
3. The probability distribution function of the diquark condensate

The use of the probability distribution function to analyse the spontaneous symmetry
breaking in spin systems or Quantum Field Theories with bosonic degrees of freedom is a
standard procedure. Less standard is its application to QFT with Grassmann fields where
the fermionic degrees of freedom have to be integrated analytically.
The version of this method we use has been developed to extract the chiral condensate
in the chiral limit from simulations of QFT with dynamical fermions [6], and will be used
here to study the vacuum structure of two colours QCD at nonzero baryon density with
Nf = 1, 2, 4 flavours of dynamical KogutSusskind fermions and specifically to extract
the diquark condensate at j = 0. We refer to the original paper [6] for a full description of
the p.d.f. technique; we present here a brief introduction focusing on the peculiarities of
the diquark condensate case.
In analogy with the study of chiral symmetry, we can construct a two component diquark
i2
which transforms under U(1)B as
2 ,
2 )
condensate vector (2 +
a vector under rotations in the plane. Therefore we can take any of these two diquark
condensates as order parameter c for the U(1)B symmetry associated to baryon number
conservation. In the following we will use the first component. Then let be an index
which characterizes all possible (degenerate) vacuum states and w the probability to get
the vacuum state when choosing randomly an equilibrium state. If c is the value taken
by the order parameter in the state, we can write

1
2 (x)
2 (x) +
,
c =
(10)
V x
where V is the lattice volume and the sum is over all lattice points.
P (c), the p.d.f. of the diquark order parameter c, will be given by

w (c c )
P (c) =
1
V Z
= lim
[d]eSG (U )+
[dU ] [d ]

1
2 (x) + 2 (x) c .
V x
The main point is that while P (c) is not directly accessible with a numerical simulation
its Fourier transform

P (q) = dc eiqc P (c)
(11)
can be easily computed. Inserting in (11) the definition of P (c) and using an integral
representation for the -function, we can compute the integral over the Grassmann
variables:

iq
2 )
(2 +
[d] e+iq/V
[d ]
(12)
= Pf B
,
V
328
where we have defined

j
B(j ) =
1 T
2 2
1
2 2

.
(13)
and is the usual lattice Dirac operator (it contains the mass and dependence).
After some algebra we obtain:
iq

Nf
1
SG (U ) Pf B V
[det ] 4 ,
PV (q) =
(14)
[dU ] e
Z
det
where PV (q) is the Fourier transformed p.d.f. of the diquark order parameter at zero
external source with Nf flavours of dynamical fermions at finite volume V ; Z is the
standard partition function of the full theory.
To take advantage of (14) we should be able to compute correctly the Pfaffian involved.
This indeed turns out to be easy. The only ambiguity is related to the sign of Pf B( iq
V )=
det B(iq/V ). In the next section we will analyse some properties of the matrix B and
show how it is possible to compute Pf B(iq/V ) once we have the eigenvalues of B(0).
As we will discuss in the next section, in standard simulations it is practically impossible
to determine the sign of the Pfaffians involved and one is forced to choose Nf equal to a
multiple of 8. A remarkable property of our approach is that, since we are able to compute
Pf B(iq/V ), any value of Nf can be considered in the simulations. In fact Nf enters
only in the computation of the fermionic factor det (positive for any ) and not in the
evaluation of the Pfaffian.
Once we have the PV (q) and hence, by simple Fourier transform, the p.d.f. of the order
parameter we need to extract the correct value for the order parameter. To do that we
first have to recall that diquark condensate is the order parameter of the UB (1) symmetry
2
and i2
associated with baryon number conservation and that 2 +
2 are the components of a vector (in a plane) which rotates by an angle 2 when we
do a global phase transformation of parameter on the fermionic fields. Therefore, if c0
is the vacuum expectation value of the diquark condensate (in the infinite volume limit)
corresponding to the -vacuum selected by a diquark source term after taking the zero
source limit, P (c) can be computed as

1
P (c) =
(15)
d c c0 cos(2) ,
2
which gives P (c) = 1/((c02 c2 )1/2 ) for c0 c c0 and P (c) = 0 otherwise [6]. In
the symmetric phase c0 = 0 and P (c) reduces to a -function in the origin.
The above results are valid in the thermodynamic limit while, at finite volume, the
nonanalyticities of the p.d.f. are absent. Even without performing a detailed finite size
scaling analysis, we expect, for the finite volume p.d.f. PV (c), a function peaked in the
origin in the symmetric phase and peaked at some nonzero value in the broken phase.
This is indeed the behaviour we can observe in Fig. 1 where the the p.d.f. of the smallest
volume simulated (see Sections 4 and 5 for the details of the simulations) is reported at
different values of . It is clear that, increasing the chemical potential, the vacuum starts
to be degenerate signalling a spontaneous breaking of the baryon number conservation.
329
Fig. 1. P (c) for the 44 lattice, m = 0.2 at = 0.2, 0.4, 0.475, 0.5 (from top to bottom).
We can also compare PV (c) for two lattice volumes in the symmetric and broken phase.
This is done in Fig. 2 where we see clearly as, increasing the volume, the peak of the p.d.f.
becomes sharper. To determine the value of the diquark condensate we used the position
of the peak: a definition that clearly converges to the correct value in the thermodynamic
limit.
From Fig. 2 we see also that data of the larger volume are more noisy (indeed we also
get negative values for the p.d.f. in the broken phase). To have an estimate of the errors we
calculated several distribution functions for the 64 lattice using the jacknife procedure. We
saw that, except for the critical region, the position of the peak was very stable.
4. Simulation scheme
The standard way to study SSB is to introduce first an explicit symmetry breaking term
in the action. If we do that for the diquark in the SU(2) model we have to add a term [8]
2 )
and, after integrating the Grassmann field, the fermionic contribution
j (2 +
for Nf = 4 quark flavours becomes proportional to the Pfaffian of a 4V 4V matrix [3]:

Zferm (j ) = Pf B(j ) = det B(j ),
(16)
where B(j ) is defined in the previous section and j is real.
Using the relation 2 2 = we can easily prove that B(0) is antihermitian and
det B(j ) 0 for any j . It can also be shown that the eigenvalues of B(0) are doubly
330
Fig. 2. P.d.f. for m = 0.2: 44 (dashed line) and 64 (continuous line) in the symmetric (top) and broken
(bottom) phase.
degenerate and B 2 (0) is block diagonal with two hermitian blocks on the diagonal having
the same eigenvalues. It follows that to compute det B(j ) for any j (i.e., to obtain all the
eigenvalues of B(0)) it is sufficient to diagonalize only one block of B 2 (0) (reducing the
problem to the diagonalization of a 2V 2V hermitian matrix) and then take the two
(imaginary) square roots of its (real and negative) eigenvalues.
To avoid the sign ambiguity in (16) it is customary to consider a theory with Nf = 8
quark flavours where the fermionic partition function becomes Zferm (j ) = det B(j ). This
limitation can be overcome by exploiting our ability to work directly at zero diquark source.
331
In the j = 0 limit the sign ambiguity disappears and the Pfaffian is positive definite since
Pf B(0) det and the last quantity is real and positive for any value of . Then we can
easily consider any value of Nf writing Zferm (j = 0) = (det B(0))Nf /8 .
If we are interested in the p.d.f. of the diquark order parameter at zero external source,
the Pfaffian in (14) can also be easily computed. The nondegenerate eigenvalues of B(0)
can be written as in (n = 1, . . . , V ) with real and positive n . Using the relation (16)
we arrive at the following expression

V

iq
q2
=
Pf B
(17)
2n 2 .
V
V
n=1
The sign ambiguity is solved noticing that (17) has to be positive at q = 0 (indeed
Pf B(0) = det ). Increasing q, (17) changes sign each time q/V is equal to one of the n ,
except possibly in case of degeneracy in the eigenvalues of B(0) resulting in the Pfaffian
not crossing zero but tangent to the horizontal axis. This situation never occurred in our
simulations.
The procedure we presented in previous section can be used for any value of the gauge
coupling but we have studied the phase structure of the theory in the limit of infinite gauge
coupling ( = 0). The main reason lies in the possibility to check our results with standard
numerical simulations as well as analytical calculations. In this way we can restrict our
efforts to the exploration of the phase space in the masschemical potential plane.
To simulate the theory we have used a procedure inspired from MFA scheme for
unquenched lattice simulations [7] (extensively tested in the past both for abelian and
nonabelian gauge theories).
We generate gauge configurations with a value of the pure gauge action in the range
relevant for the values of the thermodynamical parameters we choose. The matrix B is
constructed and diagonalized and the eigenvalues of B are then used for the evaluation
of both the Pfaffian of B and the determinant of in (14); note that we can use the
same configurations for all the values of Nf we considered, owing to the fact that Nf
enters only as an exponent of the fermionic weight in (14). Repeating the procedure
for a large number of gauge configurations this gives us a statistical extimation of the
PV (q) and, via Fourier (anti)trasformat, the probability distribution function of the diquark
condensate.
Being interested in the strong coupling limit of the theory we choose to measure
fermionic observables on gauge configurations generated randomly, i.e., with only the Haar
measure of the gauge group. This choice implies a Gaussian distribution of the plaquette
energy around zero which, according to the results of Morrison and Hands [8], has a net
overlap with the importance sample of gauge configurations for the full theory at the values
of , m and Nf used in our calculations. The validity of this procedure has also been
tested comparing different physical observables (number density and chiral condensate)
with Hybrid Montecarlo results [9].
We have considered the theory in a 44 and 64 lattice diagonalizing 300 gauge
configurations in the first lattice volume and 100 in the second one. As we pointed out
in the introduction the simulations have been performed at nonzero quark mass, in order
332
to study a physical situation closer to SU(3). We have chosen m = 0.025, 0.05, 0.20 and
values of the chemical potential ranging from = 0 to = 1.0. The values of the diquark
condensate are determined from the position of the maximum of the probability distribution
function and the errors have been determined using a jacknife procedure.
All numerical simulations have been performed on a cluster of PCs at the INFN Gran
Sasso National Laboratory.
5. Results and conclusions

Here we present the results for the diquark condensate at j = 0 as a function of the
chemical potential.
Fig. 3 contains a comparison of 64 and 44 results for Nf = 1 and m = 0.2. We can
clearly distinguish two symmetric phases separated by a broken one and two, possibly
continuous, transition points. The high density symmetric phase has no physical relevance,
being consequence of the saturation of all lattice sites with quarks. This phenomenon
has nothing to do with continuum physics, being a pure lattice artifact. The physically
interesting phase transition, i.e., the transition that has a continuum counterpart, is the first
one [3] and it is the only one we will consider in the following.
With only two available volumes we do not have the possibility to perform a serious
finite size scaling analysis. Anyway it is evident that increasing the lattice volume the
behaviour near the (physical) critical point is more singular. A similar qualitative result
Fig. 3. Diquark condensate for 44 (diamonds) and 64 (squares) lattices at m = 0.2, Nf = 1.
333
Fig. 4. Diquark condensate for Nf = 1 (diamonds), 2 (squares) and 4 (stars) in a 64 lattice at m = 0.2.
holds for all the available masses and flavour numbers thus strongly supporting the picture
of singular behaviour in the infinite volume limit.
We have also taken advantage of our simulation scheme considering different values
of flavours. Increasing Nf our operators become more noisy, especially near the transition
point, but we can safely extend our calculations up to Nf = 4. In Fig. 4 we plot the diquark
condensate for the largest mass and Nf = 1, 2, 4. We see clearly that the three data set are
almost coincident and we conclude that no dependence on Nf is evident. Once again this
result is valid also for the smaller masses.
The existence of spontaneous symmetry breaking with a nonvanishing diquark condensate has already been predicted by Mean Field calculations [10] as well as numerical calculations [3,11] and effective models [4]. We can use our relatively large data set to take a
step forward and make a quantitative comparison with available analytical predictions.
In previous work we used low energy effective Lagrangian calculations to check j = 0
results finding remarkable agreement [3]. This motivated us to repeat the same procedure
for our j = 0 results. In this case the authors of [4] provide the following formula for the
diquark condensate:

0 1 m 4 , m /2,

2
=
(18)
0,
otherwise,
where 0 , m are the chiral condensate and the pion mass at zero chemical potential.
We used (18) to fit our data for the larger lattice near the critical point: plotting 4 2
versus 4 we expect a linear dependence. We used all the available masses and the Nf = 1
334
Fig. 5. 4 2 vs 4 for 64 lattice, m = 0.025 and linear best fit.

Table 1
Parameters for the low energy effective Lagrangian predictions
m
0.025
0.05
0.2
1 m (fit)
2
1m
2
0 (fit)
0
2
d.o.f.
0.165(2)
0.236(5)
0.484(4)
0.1696(11)
0.2405(9)
0.4841(7)
1.325(10)
1.29(2)
1.438(13)
1.31(2)
1.29(2)
1.23(2)
1.1
2.2
3.3
results that have smaller errors. In Fig. 5 the m = 0.025 case is considered showing both
the data points and the resulting linear fit. In all cases the 2 is good (see Table 1) and we
conclude that (18) gives a good description of the numerical results.
In Fig. 6 we have reported the diquark condensate as a function of for all the available
masses. Clearly, away from the critical point, we can appreciate the dramatic effect of
lattice discretization that prevents the diquark condensate to stay at the = 0 chiral
condensate value (18).
In Table 1 we compare the results of the fits with the = 0 determinations of

3
and m /2 performed on a 6 12 lattice at = 0. The determination of the critical point
and chiral condensate is in good agreement in the two data set, at least for the two smaller
masses. Increasing m the 2 becomes larger and the determination of the chiral condensate
is less accurate. This however is not surprising since we expect the validity range of the low
energy Lagrangian prediction (18) to be restricted to the small quark mass region where
335
Fig. 6. Diquark condensate for 64 lattice, Nf = 1, m = 0.025 (diamonds), 0.05 (squares), 0.2 (stars)
and corresponding fits using expression (18).
the gap between the pion mass and the first non-Goldstone excitation is large [4].
What is more surprising is that, like in the j = 0 case [3], our = 0 calculations have
a incredibly good agreement with a continuum prediction. Since the analytical predictions
are, for the values of presented, well inside the validity region of the low energy
approximation we can conclude that this gives indication of a very small dependence of
lattice results on .
It would be very interesting to test this prediction by performing finite coupling
calculations. This can be done using our approach to extract directly the diquark
order parameter at zero external source together with a HMC algorithm for generating
configurations at = 0. In this case we can compute PV (q) in (14) considering the
ratio Pf B(iq/V )/ det as an observable which, for each gauge configuration and q <
V , is a number of order one (or smaller). This program has to face two difficulties: at
zero temperature we need to increase our lattice temporal extent and correspondingly the
computer time; on the other hand only unconclusive results are available for the = 0
thermodynamics of SU(2).
Acknowledgements
The authors thank M.P. Lombardo for providing the code necessary to extract the pion
mass in = 0 simulations. This work has been partially supported by CICYT (Proyecto
336
AEN97-1680) and by a INFN-CICYT collaboration. The Consorzio Ricerca Gran Sasso

has provided part of the computer resources needed for this work.
References
[1] M. Alford, K. Rajagopal, F. Wilczek, Phys. Lett. B 422 (1998) 247;
M. Alford, K. Rajagopal, F. Wilczek, Nucl. Phys. B 537 (1999) 443;
R. Rapp, T. Schaefer, E.V. Shuryak, M. Velkovsky, Phys. Rev. Lett. 81 (1998) 53;
T. Schafer, F. Wilczek, Phys. Rev. Lett. 82 (1999) 3956.
[2] M.P. Lombardo, hep-lat/9907025;
S.J. Hands, J.B. Kogut, S.E. Morrison, D.K. Sinclair, hep-lat/0010028;
E. Bittner, M.P. Lombardo, H. Markum, R. Pullirsch, hep-lat/0010018;
E.B. Gregory, S.H. Guo, H. Kroger, X.Q. Luo, Phys. Rev. D 62 (2000) 054508;
X.Q. Luo, E.B. Gregory, S.H. Guo, H. Kroger, hep-ph/0011120;
S.J. Hands, I. Montvay, S.E. Morrison, M. Oevers, L. Scorzato, J. Skullerud, hep-lat/0006018;
S.J. Hands, I. Montvay, M. Oevers, L. Scorzato, J. Skullerud, hep-lat/0010085.
[3] R. Aloisio, V. Azcoiti, G. Di Carlo, A. Galante, A.F. Grillo, Phys. Lett. B 493 (2000) 189.
[4] J.B. Kogut, M.A. Stephanov, D. Toublan, J.J.M. Verbaarschot, A. Zhitnitsky, hep-ph/0001171.
[5] C. Vafa, E. Witten, Nucl. Phys. B 234 (1984) 173.
[6] V. Azcoiti, V. Laliena, X.Q. Luo, Phys. Lett. B 354 (1995) 111.
[7] V. Azcoiti, G. Di Carlo, A.F. Grillo, Phys. Rev. Lett. 65 (1990) 2239;
V. Azcoiti, A. Cruz, G. Di Carlo, A.F. Grillo, A. Vladikas, Phys. Rev. D 43 (1991) 3487;
V. Azcoiti, G. Di Carlo, L.A. Fernandez, A. Galante, A.F. Grillo, V. Laliena, X.Q. Luo, C.E.
Piedrafita, A. Vladikas, Phys. Rev. D 48 (1993) 402;
V. Azcoiti, G. Di Carlo, A. Galante, A.F. Grillo, V. Laliena, Phys. Rev. D 50 (1994) 6994;
R. Aloisio, V. Azcoiti, G. Di Carlo, A. Galante, A.F. Grillo, Phys. Rev. D 61 (2000) 111501(R).
[8] S. Morrison, S. Hands, in: Strong and Electroweak Matter 98, Copenhagen, Dec. 1998, heplat/9902012.
[9] R. Aloisio, V. Azcoiti, G. Di Carlo, A. Galante, A.F. Grillo, Nucl. Phys. B 564 (2000) 489.
[10] E. Dagotto, A. Moreo, U. Wolff, Phys. Lett. B 186 (1987) 395.
[11] S. Hands, S. Morrison, in: Understanding Deconfinement in QCD, Trento, March 1999, heplat/9905021.

Evolution equations for the effective four-quark

interactions in QCD
Enrico Meggiolaro a , Christof Wetterich b
a Dipartimento di Fisica, Universit di Pisa, Via Buonarroti 2, I-56127 Pisa, Italy
b Institut fr Theoretische Physik, Universitt Heidelberg, Philosophenweg 16, D-69120 Heidelberg, Germany
Received 8 December 2000; accepted 12 March 2001
Abstract
A nonperturbative renormalization group equation describes how the momentum dependent fourquark vertex depends on an infrared cutoff. We find a quasilocal one-particle irreducible piece
generated by (anomaly-free) multi-gluon exchange. It becomes important at a cutoff scale where
scalar and pseudoscalar meson-bound states are expected to play a role. This interaction remains
subleading as compared to the effective one-gluon exchange contribution. The local instanton
induced four-quark interaction becomes dominant at a scale around 800 MeV. In absence of a gluon
mass the strong dependence of the one-gluon exchange on the transferred momentum indicates that
the pointlike interactions of the NambuJona-Lasinio model cannot give a very accurate description
of QCD. A pointlike effective four-quark interaction becomes more realistic in case of spontaneous
color symmetry breaking. 2001 Elsevier Science B.V. All rights reserved.
PACS: 12.38.-t; 12.38.Lg; 11.10.Hi
1. Introduction
The NambuJona-Lasinio (NJL) model [1] has often been used as a simple model
for low-momentum strong interactions. After inclusion of the chiral anomaly it gives a
relatively successful description of the chiral symmetry aspects of QCD and the associated
properties of mesons [2]. A renormalization-group treatment based on exact flow equations
has provided a consistent picture of the chiral aspects of the high temperature phase
transition for two light quark flavors [3]. On the other hand it is well known that crucial
properties of long-distance QCD such as confinement cannot be accounted for by this
model. The NJL-model shares the chiral symmetries of QCD whereas color symmetry
appears only as a global symmetry. It is based on a pointlike four-quark interaction
E-mail address: enrico.meggiolaro@df.unipi.it (E. Meggiolaro).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 3 0 - 4
338
E. Meggiolaro, C. Wetterich / Nuclear Physics B 606 (2001) 337356
(NJL)
=
NJL

i a j b
Lb Ri Ra Lj ,
d4 x i ai ia + 2(NJL)
(1.1)
where NJL indicates that only fluctuations with momenta q 2 < 2NJL should be included
in the functional integral. (Here i, j = 1, 2, 3, and a, b = 1, 2, 3, are the color and flavor
indices, respectively.) The quartic interaction may be supplemented by a local anomaly
term which is sixth order in the quark fields for three light flavors. Also extensions with
four-quark interactions in the vector channel have been investigated [2].
From the perspective of QCD, integrating out the gluon fields as well as all fermion
fluctuations with (covariant) momenta larger than NJL should lead to an effective action
containing pieces like (1.1), together with possibly quite complicated additional multifermion interactions. The question arises under which circumstances and for which
problems the NJL model (1.1)can be considered as a good approximation to the multiquark interactions obtained from QCD. As a first step towards an answer we compute
in this paper the four-quark vertex with the index structure appearing in Eq. (1.1). An
investigation of its momentum dependence serves as a test if the pointlike interaction of
the NJL model is reasonable.
Our method to explore the physics at the nonperturbative scale NJL (500800) MeV
is based on truncations of the exact nonperturbative renormalization group equation for
the effective average action [48]. An earlier study [9] has followed the running of the
momentum-dependent effective four-quark interaction with initial conditions mimicking
QCD. In contrast, we start in this paper at short distances from standard perturbative QCD.
We observe that a one-particle-irreducible (1PI) four-quark interaction is generated by boxtype diagrams and becomes indeed approximately pointlike for scales of the order NJL .
Nevertheless, the contribution of the one-gluon exchange to the effective four-quark vertex
retains the characteristic momentum dependence of a particle exchange in the t-channel.
This contribution dominates quantitatively over the NJL-interaction. Another pointlike
interaction arises from instanton effects. It becomes dominant at a cutoff scale around
800 MeV. We show the different contributions to the four-quark vertex in Fig. 2, where
we present the case of massless gluons. For massless gluons a pointlike NJL-interaction of
the type (1.1) seems not to be a realistic approximation to the effective four-quark vertex
in QCD. The hypothesis that gluons acquire a mass MV 850 MeV from spontaneous
color symmetry breaking [10] is discussed later (Section 5). For massive gluons the
approximation of pointlike four-quark interactions can be defended. Nevertheless, any
quantitatively reliable approximation should retain the one-gluon exchange in addition to
the instanton interaction and the NJL-interaction (1.1). This is the setting used for gluon
meson duality [10].
The paper is organized as follows. In Section 2 we derive a truncated nonperturbative
flow equation for the scalar-exchange-like effective momentum-dependent four-quark
interaction in QCD. This effective one-particle irreducible interaction arises from boxtype diagrams. The box interaction is relevant for the physics of scalar and pseudoscalar
mesons and, therefore, for spontaneous chiral symmetry breaking. In the limit where it
becomes independent of the momentum exchanged in the t-channel it can be described by
meson exchange. In Section 3 we study, by numerical methods, the above-mentioned flow
339
equation. We also present the perturbative expansion of the flow equation and its solution
and comment about some checks on the validity of our approximations. In Section 4
we compare the box interaction and the local instanton-induced four-quark interaction.
In Section 5 we discuss the hypothesis of spontaneous color symmetry breaking, and a
summary and conclusions are drawn in Section 6.
2. Nonperturbative flow equations
2.1. Effective average action
The effective average action k [4] generalizes the generating functional for the 1PI
correlation functions (effective action) for a situation where only fluctuations with
(covariant) momenta q 2 > k 2 are included in the functional integral. It obtains by
introducing an infrared cutoff k into the original functional integral such that for k 0
one recovers the effective action = k0 . For a given k at nonzero k the computation
of from k involves only the fluctuations with q 2 < k 2 . This is the setting one needs for
the effective NJL model. Our aim is, therefore, the computation of k for k = NJL . Since
we do not know the precise value of NJL we will consider arbitrary k in the appropriate
nonperturbative range. More precisely, the NJL model is formulated in terms of quark
fields. Since we start with the QCD action involving both quarks and gluons, the gluons
should be integrated out. This may be done by retaining the infrared cutoff only for the
fermionic fields. One then has to solve the gluon field equations obtained from k as a
functional of the quark fields and reinsert the solution into k [11,12]. Typically, we will
not integrate out the gluon fluctuations with momenta smaller than k in this paper. Our
results will nevertheless be a good guide for the qualitative characteristics of the effective
four-quark interactions.
for gluons A, ghosts c and quarks
, ]
The effective average action k [A, c, c,
obtains by a Legendre transform from a functional integral involving the QCD classical
action and an infrared regulator:
= Sgauge [A, c, c] + Sk,gauge[A, c, c]
+ SF [A, , ]
Sk [A, c, c, , ]
+ SkF [, ].
Here the fermionic action SF is given by

= d4 x ai i D [A] j ja ,
SF [A, , ]
i
(2.1)
(2.2)
with (D [A])i j = i igAz (T z )i j the covariant derivative in the fundamental

representation. Here a denotes the flavor and we use two sorts of color indices z =
1, . . . , Nc2 1, and i, j = 1, . . . , Nc , for a gauge group SU(Nc ). We consider, for simplicity,
Nf massless quarks: a = 1, . . . , Nf . Throughout this paper we work in four-dimensional
Euclidean space, so that the matrices in the previous equation are the (hermitean)
Euclidean -matrices. The gauge part Sgauge consists of three pieces
= SYM [A] + Sgf [A] + Sgh [A, c, c],
Sgauge [A, c, c]
(2.3)
340
where SYM is the YangMills action, given by:

1
z
z
F
,
SYM [A] =
d4 x F
4
(2.4)
with
z
= Az Az + gf zyw Ay Aw
F
.
(2.5)
The gauge-fixing part Sgf reads

1
Sgf [A] =
d4 x Az Az ,
2
(2.6)
where is the gauge parameter, and the ghost action Sgh is given by

Sgh [A, c, c]
= d4 x cz zw + gf zyw Ay cw .
(2.7)
The infrared cutoff modifies the propagators. For the fermions, the explicit expression
reads in momentum space [9,12]
SkF [, ]

d4 q i
=
(q) RkF (q) ia (q),
SkF [, ]
(2.8)
(2)4 a
where we choose here

RkF (q) = q rkF q 2 ,

rkF q 2 =
eq
2 / k2
.
2 2
1 eq / k
In particular, the (regularized) free fermion propagator turns out to be:

)
G(F
2,k (q)
ab
ij
= ab ij
q
1 eq
= ab ij q
2
2
q [1 + rkF (q )]
q2
(2.9)
2 / k2
(2.10)
is the sum of a gluon infrared cutoff

The gauge infrared cutoff term Sk,gauge[A, c, c]
The
term and a ghost infrared cutoff term: Sk,gauge [A, c, c] = SkA [A] + Sk,gh [c, c].
explicit expression of the gluon infrared cutoff term SkA [A] is given by [6,8]

1
d4 p z
kA
SkA [A] =
(2.11)
A (p) R
(p) Az (p),
2 (2)4
where

p p
1
kA
R
1
(p) = +
RkA p2 ,
2
(2.12)
with
RkA p
=p
ep
2/ k2
.
2 2
1 ep / k
Finally, the ghost infrared cutoff term Sk,gh [c, c]
reads

d4 p z
c (p) Rk,gh p2 cz (p),
Sk,gh [c, c] =
4
(2)
(2.13)
(2.14)
341
where

Rk,gh p2 = p2
ep
2 / k2
1 ep
2 / k2
(2.15)
With this choice, the regularized ghost propagator becomes

(gh)
G2,k (q)
zy
p
zy
zy 1 e
=
p2 + Rk,gh (p2 )
p2
2 / k2
(2.16)
It is well known [68] that the presence of the infrared cutoff explicitly breaks gauge
and hence BRS invariance: in particular, a gluon mass term appears. A background field
identity [6] or modified SlavnovTaylor identities [8,13] can be derived, which are valid for
k > 0. They guarantee the BRS invariance of the theory for k 0. Within a perturbative
expansion of k , one can derive an equation for a k-dependent mass term for the gauge
fields implied by the modified SlavnovTaylor identities. For pure YangMills theories
(i.e., with no fermions), regularized with the above infrared cutoff term (11)(13), one
finds the following result, at the leading perturbative order [8,11]:
3Nc gk2 2
(2.17)
k ( 1).
128 2
Here is the gauge parameter, introduced in Eq. (2.6), and gk is the renormalized
(k-dependent) coupling constant. In the theory with regularized fermions that we are
considering, a contribution to the gluon mass also appears from the fermion part of the
gluon self-energy. One finds, at the leading perturbative order:
m2kA =
3Nf gk2 2
k .
(2.18)
32 2
When including the gluon mass, the regularized gluon propagator becomes, in the Feynman
gauge = 1:
m2kF =
G(A)
2,k (p)
zy
zy
,
p2 + m2k + RkA (p2 )
(2.19)
where
m2k = m2kA + m2kF .
(2.20)
(Indeed, m2kA = 0 at the perturbative order O(gk2 ) in the Feynman gauge.) In other words,

using the expression (2.13) for RkA p2 :

G(A)
2,k (p)
zy
= zy
1 ep
2 / k2
p2 + m2k (1 ep
(F )
We shall concentrate on the part k

depends only on the fermion fields:
2 / k2
(2.21)
of the effective average action which

[, ]
+ (A) [A, c, c,
, ].
k [A, c, c, , ] = k(F ) [, ]
k
(2.22)
342
In general, it can be expanded [11,12] as

(F )
= [, ]
+ [, ]
+ ,
[, ]
2,k
4,k
(F )
(F )
(2.23)
(F )
contains terms with 2n fermion fields, i.e., of the form ()
n . For
where 2n,k [, ]
n = 1, we have the bilinear term 1

d4 q
(F )
=
i (q) q ia (q),
2,k
[, ]
(2.24)
(2)4 a
(F )
with n 3
Similar to Refs. [9,12], we truncate the effective action by omitting 2n,k
[, ]
(F )
and we concentrate on the behavior of the four-quark interaction (n = 2). For [, ]

4,k
we adopt a chirally-invariant parametrization [12]:

4
4
4
4
d p1
d p2
d p3
d p4
(F )
4,k [, ] =
4
4
4
(2)
(2)
(2)
(2)4
(2)4 (4) (p1 + p2 p3 p4 )

(p1 , p2 , p3 , p4 ) M (p1 , p2 , p3 , p4 )
+ (p1 , p2 , p3 , p4 ) M (p1 , p2 , p3 , p4 )

+ p (p1 , p2 , p3 , p4 ) Mp (p1 , p2 , p3 , p4 ) + ,
(2.25)
where the four-fermion couplings i have the property:

i (p1 , p2 , p3 , p4 ) = i (p4 , p3 , p2 , p1 ),
(2.26)
and the usual invariant quantities s and t read

s = (p1 + p2 )2 = (p3 + p4 )2 ,
t = (p1 p3 )2 = (p2 p4 )2 .
(2.27)
The four-fermion operators Mi are defined as:

j

1 i
a (p1 )ib (p2 ) b (p4 )ja (p3 )
2
j

1
+ ai (p1 ) 5 ib (p2 ) b (p4 ) 5 ja (p3 ) ,
2
j

1 i
M (p1 , p2 , p3 , p4 ) = + a (p1 ) ib (p2 ) b (p4 ) ja (p3 )
4
j

1
+ ai (p1 ) 5 ib (p2 ) b (p4 ) 5 ja (p3 ) ,
4
j

1 i
Mp (p1 , p2 , p3 , p4 ) =
(p1 ) ia (p3 ) b (p4 ) jb (p2 ) . (2.28)
2Nc a
If does not depend on t, the term M describes the exchange of color-singlet, flavornonsinglet spin-0 scalar quarkantiquark states in the s channel: they correspond to the
spin-0 pseudoscalar and scalar mesons. Similarly, M describes the exchange of colorsinglet, flavor-nonsinglet spin-1 vector states in the s channel: they correspond to the spin1 vector and axial vector mesons. Finally, Mp describes the propagation of a color- and
M (p1 , p2 , p3 , p4 ) =
1 We omit in this paper the running of the fermion wave function renormalization.
343
flavor-singlet spin-1 vector state in the t channel: these are the quantum numbers of the
pomeron. We concentrate on the analysis (both analytical and numerical) of the evolution
equation for the scalar-like effective four-fermion coupling , which multiplies M and
is relevant for the physics of scalar and pseudoscalar mesons and, therefore, for spontaneous chiral symmetry breaking. The dots in Eq. (2.25) stand for more complicated momentum structures, the contribution of the chiral anomaly (to be discussed in Section 4)
and terms breaking the chiral flavor symmetry.
2.2. Flow equations
The exact flow equation for k can be written as a formal derivative of a renormalization
group improved one-loop expression [4]
k k =

(2)
1
STr k ln k + Rk ,
2
(2.29)
where k(2) is the second functional derivative with respect to the fields and k acts only on
Rk and not on k(2) , i.e.:
Rk
k
.
k Rk
(2.30)
Here Rk is to be interpreted as a matrix Rk = diag(RkA , Rk,gh , RkF ). The trace involves a

summation over indices and a momentum integration and STr indicates an additional
minus sign for the fermionic part. Flow equations for n-point functions follow from
appropriate functional derivatives of Eq. (2.29) with respect to the fields. For their
derivation it is sufficient to evaluate the corresponding one-loop expressions in presence
of the infrared cutoff (with the vertices and propagators derived from k ) and then to take
a formal k derivative. This permits the use of (one-loop) Feynman diagrams and standard
perturbative techniques.
Using these techniques, we have derived the flow equation for the four-fermion (1PI)
coupling by computing the relevant one-loop diagrams from the effective average action
these diagrams are shown in Fig. 1. Our truncation includes only two
, ]:
k [A, c, c,
vertices: (i) the usual quarkgluon vertex and (ii) the four-fermion (1PI) vertex M .
We have extracted from the relevant diagrams the contributions to the -type four-fermion
vertex M and then performed a formal k derivative in the sense of Eq. (2.30). In this
picture (which is different from the one used in Ref. [9]) the gluons are not integrated out.
Therefore, in addition to the contributions to the flow equation coming from the infrared
cutoff in the fermion lines inside a given Feynman diagram, one has to consider also the
ones from the infrared cutoff in the gluon lines.
2.3. Truncations and the flow equation for
In general, the function can depend on six independent Lorentz-invariant products
of the momenta = (s, t, p12 , p22 , p32 , p42 ). We have derived the evolution equation of
for the particular configuration of momenta (p1 , p2 , p3 , p4 ) = (p, p, p, p), so that
344
(a)
(b)
(c)
(d)
Fig. 1. One-loop diagrams for the flow equation (2.31) for the four-fermion (1PI) coupling . Five
more diagrams similar to (c) with gluon lines attached at different quark lines are not shown.
s = 0, t = 4p2 . In order to simplify the momentum structure, we have made a further

truncation, namely that depends only on the invariant t: (t). Furthermore,
our truncation restricts the quarkgluon interactions to minimal coupling with a coupling
constant depending on the gluon momentum. We use the classical form of the (regularized)
gluon propagator. 2 Our truncation omits the influence of all multiquark interactions except
those proportional M (2.28). Moreover, we neglect the possible effects from a q 2 - and
k-dependent fermion wave-function renormalization Z,k .
With these approximations, we find the following result:
kk (t)

16Nc
d4 q
3
2
= 2
4 1
k (p q)2 k (p + q)2
k
(2)4
2Nc2
2 2

q 2 Fk (q)Gk (p q) eq / k Gk (p + q) + Fk (q)k (p + q)

2 2
1
k (p + q)2 q 2
4p2 eq / k Fk (2p + q)Gk (p + q)
2
2Nc

2
2
+ e(2p+q) / k Fk (q)Gk (p + q) + Fk (q)Fk (2p + q)k (p + q)

q 2 / k 2

3
2
2
(q)
2e
G
(p
+
q)
+
F
(q)
(p
+
q)
(p
q)
F
k
k
k
k
2Nc2

2 2
(2.31)
+ (p q)2 (p + q)2 q 2 eq / k Fk (q) .
2 See [13,14] for an investigation of the q 2 -dependence of the gluon propagator by means of nonperturbative
flow equations.
345
Here we use the following shorthands:

Fk (q) =
1
1 eq
=
q 2 [1 + rkF (q 2 )]
q2
2 / k2
1 eq / k
1
=
,
Gk (q) =
q 2 + m2k + RkA (q 2 ) q 2 + m2k (1 eq 2 / k 2 )
2
k (q) =
(q 2 )2 eq
(2.32)
2 / k2
[q 2 + m2k (1 eq
2 / k2
)]2
and employ
rkF (q 2 ) Fk (q)
2
2 2

= 3 eq / k ,
k Fk (q) =
2
k
rkF (q )
k
2 ) G (q)
(q
R
2
kA
k

k Gk (q) =
= 3 k (q).
2
k
RkA (q )
k
(2.33)
The first two terms in the r.h.s. of Eq. (2.31) (i.e., those proportional to k2 or k )
reflect the exchange of gauge bosons (see Fig. 1(a)(c)). The last term (i.e., the one
proportional to 2 ) corresponds to the result of Ref. [9] for the truncation = (t)
(see Fig. 1(d)).
2.4. RG-improvement of the gauge coupling
The flow Eq. (2.31) is RG-improved in the sense that we use everywhere a running
coupling constant k (q 2 ) = gk2 (q 2 )/4 . Instead of computing k (q 2 ) by its own flow
equation, we use here an ad hoc relation to the two-loop running gauge coupling:

2
2
1
1 ln[ln( /QCD )]
() =
(2.34)
1 2
.
40 ln(2 /2QCD )
0 ln(2 /2QCD )
Here 0 and 1 are the two first coefficients of the QCD -function
(g) =
dg
= 0 g 3 1 g 5 + ,
d
(2.35)
which, for a gauge group SU(Nc ) and Nf quark flavors, are given by:

11Nc 2Nf
34 2
1
1
13Nc2 3
,
1 =
N
Nf , (2.36)
0 =
3
3Nc
(4)2
(4)4 3 c
such that, for Nc = 3 and Nf = 3:

2
2
64 ln[ln( /QCD )]
4
1
.
() =
81 ln(2 /2QCD )
9 ln(2 /2QCD )
(2.37)
For the relation between q 2 , k 2 and 2 , we use two different ansatze: (A) 2 = q 2 +k 2 , and
(B) 2 = k 2 . The first one (A) is motivated by the fact that both q 2 and k 2 act as effective
infrared cutoffs such that essentially the larger one counts. The second one (B) neglects
346
the dependence of the vertex on the gluon momentum. The overall mass scale needs in
addition the unknown ratio between QCD in the scheme relevant for the flow equations
(ERGE-scheme) to the MS scheme. We, therefore, report all quantities with dimension of
(3)
(3)
[mass]n in units of (GeV QCD / )n , where
is the fundamental mass scale of
MS
MS
QCD in the MS renormalization scheme (two-loop), with Nf = 3. The experimental value

(3)
s (MZ ) = 0.118(3) corresponds to 360 MeV. For comparison we will also use a
MS
(3)
fixed value of corresponding to the scale = kc = 1.5 QCD / GeV, i.e., kc 0.34
MS
[ansatz (C)].
2.5. One-gluon-exchange contribution
The term M describes only the contribution to the effective -type vertex coming
from the 1PI graphs. A four-quark interaction with a similar structure arises also from onegluon exchange which is not one-particle irreducible and therefore does not contribute 3 to
in the present formulation. Indeed, the four-fermion vertex induced by the one-gluon
exchange is given by [12]
4
4
4
4
d p1
d p2
d p3
d p4
(1g) =
(2)4 (4) (p1 + p2 p3 p4 )
4
4
4
(2)
(2)
(2)
(2)4

2k (p1 p3 )2 Gk (p1 p3 ) M(p1 , p2 , p3 , p4 ),
(2.38)
where

M(p1 , p2 , p3 , p4 ) = ai (p1 ) (T z )i j ja (p3 ) bk (p4 ) (Tz )kl lb (p2 ) .
(2.39)
By an appropriate Fierz transformation, one can show that
M = M + M + Mp ,
(2.40)
where M , M and Mp have been defined in Eq. (2.28). The total contribution in the
(1g)
-channel is, therefore ( + )M with
(1g)
(t) = 2k (t)
1 et / k
t + m2k (1 et / k )
2
(2.41)
This contribution remains nonlocal for k 0 if m2k tends to zero in this limit. The size of
the 1PI contribution should be compared with the effective coupling (2.41).
3. Scale dependence of the four-quark interaction
3.1. Perturbation theory
In perturbation theory the 1PI four-quark interactions arise from box and cross twogluon exchange diagrams (see Figs. 1(a), (b)). For the part with the structure M we find
the perturbative value (t = 4p2 )
3 In the formulation [9,11,12] the one-gluon exchange part is included in .
347
3
2
(p)
(t)
=
16
N
1
2Nc2

2
d4 q
k (p q)2 k (p + q)2 q 2 Fk (q) Gk (p q) Gk (p + q).
4
(2)
(3.1)
In particular, for t = 0 and momentum-independent k [ansatz (B)], the momentum
integration can be performed explicitly. One finds, for Nc = 3 and in the limit of a vanishing
gluon mass term m2k = 0:
(p)
(t = 0) =
Here the constant
5k2
K .
k2
(3.2)
2
2
d4 q 2
q Fk (q) G0k (q) ,
4
(2)
2 2

1 eq / k
1
0

Gk (q) = Gk (q) m2 =0 = 2
=
,
k
q + RkA (q 2 )
q2
K = 8 2 k 2
(3.3)
depends on the precise form 4 of the infrared cutoff. For our choice (see Eq. (2.13)) one
finds
K = 2 [5 ln 2 3 ln 3] 0.34.
(3.4)
For the purpose of comparison with our results in the following figures, we quote the value
corresponding to k = kc = 1.5 QCD /(3) GeV (kc 0.34):

(p) (t = 0)
k=kc
8.7 10

2
MS
(3) 2
.
MS
GeV QCD /
(3.5)
In order to get a feeling for the relative strength of the -channel as compared to other
types of 1PI four-quark interactions, we also present here the full one-loop 1PI interaction
for zero external momenta (i.e., s = 0, t = 0). For an arbitrary gauge parameter one
obtains the effective (local) four-quark interaction
gk4
K
64 2k 2

1
j

3 2 i a j
9 9
a j b ib ai ia b jb
+ +
4 2
4
3

11 i 5 a j

5 i 5 a j
5 b
5 b
i b j .
a j b i
2
6 a
(3.6)
We note that the first term has precisely the structure of the one-gluon-exchange diagram
M (2.39). We also emphasize that there are several ways how to Fierz-transform the
expression (3.6). Our truncation of keeping only corresponds more precisely to state
which invariants are neglected.
(F )(p)
L4,k
4 For the fermion cutoff used in [5] one has to replace F 2 F /q 2 .
348
It is straightforward to verify that the first term in the flow equation (2.31) for
precisely corresponds to the formal k derivative of the perturbative value (3.1). For t = 0
the contribution to kk obtains by replacing in the r.h.s. of Eq. (3.2) the constant K by
2K 0.68.
3.2. Initial condition for the flow equation
In order to solve explicitly the flow equation (2.31), we need to fix the initial condition,
i.e., the value of (t) at a certain initial value k = kc . We assume that at the scale
(3)
k = kc = 1.5 QCD / GeV, where perturbation theory is assumed to work well, the
MS
value of (t)|k=kc is well approximated by the leading perturbative contribution. In other
words, we use as initial condition for the flow equation (2.31):

(t)k=kc = (p)
(3.7)
(t) k=kc .
We have then solved numerically the flow equation (2.31) with the initial condition
(3.1), (3.7).
3.3. Results
In Fig. 2 we plot the resulting (t) (for Nc = 3 and Nf = 3) as a function of k, for
the particular case t = 0. The continuous line corresponds to a running gauge coupling
according to the ansatz (A). The dotted line in the same figure represents (t = 0) as a
function of k with the initial condition (t)|k=kc = 0, instead of (3.1), (3.7). We observe
the insensitivity with respect to the precise choice of the initial condition. The dot-dashed
line is the value of (t = 0) as a function of k with the initial condition (3.1), (3.7), but
using everywhere (both in the flow equation and in the initial condition) the momentumindependent coupling constant k defined by Eq. (2.37) with 2 = k 2 [ansatz (B)]. We
observe a strong increase of at a scale k 400600 MeV.
(p)
Comparison with perturbation theory is made in Fig. 3. We plot the value of (t = 0),
defined by Eq. (3.1), as a function of k (dotted line). The dashed line represents the value of
(p)
(t = 0), defined by Eq. (3.1), as a function of k, but using the momentum-independent
coupling constant k defined by Eq. (2.37) with 2 = k 2 [ansatz (B)]. In this case, the
integration in Eq. (3.1) can be performed explicitly (neglecting also the gluon mass mk
appearing in the function Gk ) and one obtains (for Nc = 3) the result (3.2)(3.4). Finally,
(p)
the dot-dashed line is obtained by using this expression for (t = 0), but using the fixed
coupling constant k = k=kc 0.34 [ansatz (C)].
In Fig. 4 we plot (t) (continuous line) as a function of t at the fixed value k = 0.42
QCD /(3) GeV. At this scale the approximation of the box-type four-quark interaction
MS
by a local NJL-vertex (1.1) is reasonable.
3.4. Checking the approximations
When inserting the effective action (2.22), (2.25) into the flow equation (2.29) and
extracting only the contributions to the -type four-fermion vertex M , one finds that
349
Fig. 2. Scale dependence of the quartic four-quark coupling. Continuous line: (t = 0) as a

function of k with the initial condition as in Eqs. (3.1), (3.7). Dotted line: (t = 0) as a function
of k with the initial condition (t)|k=kc = 0 [ansatz (A)]. Dot-dashed line: the same as for the
continuous line, but using everywhere (both in the flow equation and in the initial condition) the
momentum-independent coupling constant [ansatz (B)]. Dashed line: the one-gluon-exchange
(1g)
contribution (t = 0), defined by Eq. (2.41), as a function of k. Long-dashed line: the instanton
(3)
contribution inst , defined by Eq. (4.3), as a function of k.
in the flow equation (2.31) for the effective coupling also other pieces proportional to
i j appear: these additional pieces are generated by the four-fermion structures M
and Mp and they are put to zero in our truncation. Since only the retained contribution
2 is proportional to the color factor Nc , while all other terms of the form i j are
suppressed, our truncated evolution equation becomes exact in the leading order in a 1/Nc
expansion. Moreover, we have checked that a multiplication of the term proportional to 2
in the flow equation (2.31) by a factor of ten does not strongly influence the results. The
solution (t) of the modified flow equation is very near to the real solution of Eq. (2.31)
reported in Figs. 2, 3 and 4. In other words, the terms of the form i j do not seem to
play a fundamental role in the flow equation at the scales of k reported here. Neglecting the
contributions from the and pomeron channels does therefore not seem to be a too brutal
approximation.
We have also tried to test the approximation involved in using the two-loop expression
for the running coupling constant. In order to do this, we have considered a different
350
Fig. 3. Different approximations for in perturbation theory. Continuous line: the same as in Fig. 2.
(p)
(p)
Dotted line: (t = 0) versus k, momentum-dependent [ansatz (A)]. Dashed line: (t = 0)
(p)
versus k, momentum-independent [ansatz (B)]. Dot-dashed line: (t = 0) versus k, constant
[ansatz (C)].
version of the running coupling constant in the flow equation (2.31): we have taken for
gk2 (p2 ) the usual two-loop expression when gk2 (p2 )|2-loop < 10, while gk2 (p2 ) = 10 when
gk2 (p2 )|2-loop > 10. In other words, the new gk2 (p2 ) stops running when it reaches the value
gk2 = 10. Also in this case, the solution (t) of the modified flow equation does not show
any dramatic difference when compared to the solution of Eq. (2.31) reported in Figs. 2, 3
and 4.
Finally, since the one-particle irreducible four-point function remains small in the
momentum range of interest, a neglection of the higher-order anomaly-free effective
multiquark vertices seems well motivated.
4. Comparison with the instanton contribution

(F )
by using U (Nf ) U (Nf )
In Eq. (2.25) we have chosen to parametrize 4,k
[, ]
chirally symmetric operators M , M and Mp . In this section we supplement the effect
of the chiral anomaly. We add to the effective four-quark interactions an anomaly-induced
term [15], which is SU(2) SU(2) U (1)V invariant, but not U (1)A invariant.
351
Fig. 4. Dependence of the four-quark interaction on the transferred momentum t . We show (t)
(1g)
(continuous line) and the one-gluon-exchange contribution (t) (dashed line) as functions of t at
(3)
the fixed value k = 0.42 QCD /
GeV.
MS
In particular, for a nonvanishing mass ms of the strange quark, the instanton-induced

fermion effective action for Nf = 3 includes a four-fermion contribution of the form [16]

(4)
(Nc )
c)
= d4 x (N
inst
(4.1)
inst O4,inst (x),
where
(Nc )
O4,inst
1 l j
j l
= i k
Nc i k

u i uj dk dl u i dj dk ul + u i 5 uj d k 5 dl

u i 5 dj d k 5 ul .
(N )
(N )
(4.2)
This operator comes from L3,Ic + L3,Ac , where I = instanton and A = anti-instanton,
and the effective coupling constant involves an integral over instanton sizes :
(N )
instc
(1.34)3ms
=
2(Nc2 1)
1/ k
0
1
d (Nc ) 2 3 2 (1/) 4 2 0
d
()
4
4 0
()
352
ms
= 0.35 3
k
1
dz z2
0
(k/z)
()
4/9
2
(k/z)

2
exp
.
(k/z)
Here = 1 GeV is the renormalization scale for the fermion fields and [16,17]:
2Nc

2
2
,
exp
d0(Nc ) () = CNc
(1/)
(1/)
(4.3)
(4.4)
with
CNc =
4.6 exp(1.68Nc )
.
2 (Nc 1)! (Nc 2)!
(4.5)
For (1/) we take the (two-loop) running coupling constant (2.34) with = 1/. For the
mass of the strange quark we have used the value ms 150 MeV. For Nc = 3 and Nf = 3
one has C3 1.51 103 and we use (2.37). We have integrated the instanton length scale
between 0 and 1/k, since k is an infrared cutoff momentum scale. The long-dashed line
in Fig. 2 is the value of (3)
inst (which is t-independent!) as a function of k. One concludes
that the instanton-induced four-quark interaction becomes dominant at a scale k around
(somewhat below) 800 MeV.
5. Spontaneous color symmetry breaking?

It has been suggested [10] that confinement can be described by the Higgs phenomenon.
In this picture the expectation value of a color-octet quarkantiquark condensate 0 gives
a mass to the gluons
2
02 .
M 2 = geff
(5.1)
In the limit of three massless quarks the mass of all eight gluons is equal. Due to
spontaneous color symmetry breaking the physical electric charge of the gluons is integer
and they carry precisely the quantum numbers of the known vector mesons 5 , K , .
For this reason M is identified with the average mass of the vector meson octet MV
850 MeV. Confinement arises in this picture in complete analogy to magnetic flux
tubes in superconductors. A nonlinear reformulation of the Higgs phenomenon preserves
manifestly the gauge symmetry and shows the equivalence of the confinement and Higgs
pictures [18]. Instanton-induced six-quark interactions destabilize the color symmetric
vacuum and have been proposed as the origin of dynamical color symmetry breaking.
The results in the preceding sections correspond to a vanishing octet condensate 0 = 0.
In order to get some information on the 0 -dependence of the effective four-quark
interactions we present in this section results for a nonvanishing value of the induced gluon
mass term Mk2 . Since the precise relation between geff in Eq. (5.1) and the running gauge
coupling in our approximation (2.37) is nontrivial [10], we present results for two different
settings.
5 See [10] for a discussion of the singlet vector meson state.
353
Fig. 5. Quartic coupling with spontaneous color symmetry breaking. Continuous and dashed lines:
the same as in Fig. 2. Lower (A) and upper (B) dotted lines: (t = 0) versus k using the running
gluon mass term m2k given by Eq. (5.2) (A) and Eq. (5.3) (B). Lower (A) and upper (B) dot-dashed
(1g)
lines: (t = 0) versus k using the running gluon mass term m2k given by Eq. (5.2) (A) and
Eq. (5.3) (B).
For the first setting (A) we simply add a constant gluon mass term Mk2 = MV2 =
(850 MeV)2 in the propagator such that:
m2k = MV2 + m2kA + m2kF .
(5.2)
Moreover, we use a momentum-dependent running gauge coupling (), with 2 = q 2 +

k 2 + MV2 . For the second setting (B) we work at constant 0 = 140 MeV with a running
perturbative gauge coupling (2.34) gk2 = 4, i.e., Mk2 = gk2 02 . This replaces the gluon
mass term in Eq. (2.20) by
m2k = gk2 02 + m2kA + m2kF .
(5.3)
In this case, we use a momentum-dependent running gauge coupling (), with 2 =

2 , where M
2 = 4( = 600 MeV) 2 .
q 2 + k2 + M
k
k
0
(1g)
In Figs. 5 and 6 we compare and for different running gluon mass terms m2k
given by (5.2) (A), (5.3) (B), and (2.20) (C). We observe that the gluon mass damps the
(1g)
increase of both and . On the other hand, a pointlike approximation to the anomalyfree four-quark interaction becomes now better justified, especially for (A). Thanks to the
354
Fig. 6. Momentum dependence of quartic couplings with spontaneous color breaking. Continuous
and dashed lines: the same as in Fig. 4. Lower (A) and upper (B) dotted lines: (t) versus t using
the running gluon mass term m2k given by Eq. (5.2) (A) and Eq. (5.3) (B). Lower (A) and upper (B)
(1g)
dot-dashed lines:
Eq. (5.3) (B).
(t) versus t using the running gluon mass term m2k given by Eq. (5.2) (A) and
additional infrared cutoff Mk2 the flow can formally be followed towards k 0 without
encountering a problem from a diverging gauge coupling: the growth of the gauge coupling
is reduced by the presence of the (nonvanishing) gluon mass term. One should not expect,
however, that our results remain valid for very small k. Indeed, it has been argued that
the quark wave function renormalization Z,k multiplying the quark kinetic term 6 (2.24)
drops by a factor of about three as k reaches a few hundred MeV. This reflects the binding
of three quarks into a baryon. As a result, the running gauge coupling gk increases by a
(1g)
factor of three and the renormalized four-quark couplings and become multiplied
by a factor of nine. The running of the wave function renormalization becomes therefore
important for low k. We have not included it here (except for its contribution to the running
gauge coupling), and this is the reason why we do not explore very low values of k even
though the infrared problems are formally absent in presence of a nonzero gluon mass.
We finally note that, in presence of six quark interactions reflecting the axial anomaly,
an octet condensate 0 = 0 will also result in additional effective four-quark interactions
6 This wave function has been set Z
,k = 1 for the present work.
355
linked to the anomaly. They are not included in the present investigation. We estimate their
order of magnitude to be comparable to (or even larger than) the instanton contribution
discussed in Section 4.
6. Summary and conclusions

In this paper we have explored the effective NambuJona-Lasinio-type four-quark
interactions at the nonperturbative scale NJL (500800) MeV. For this purpose we
have used a method based on truncations of the exact nonperturbative renormalization
group equation for the effective average action [4,6,8]. More precisely, we have derived a
truncated nonperturbative flow equation for the scalar-like effective momentum-dependent
four-quark interaction in QCD. This type of interaction is relevant for the physics of scalar
and pseudoscalar mesons and therefore for spontaneous chiral symmetry breaking.
The nonperturbative renormalization group equation describes how the momentum
dependent four-quark vertex depends on an infrared cutoff k. The initial value of the
vertex at short distances is computed from perturbative QCD. We have studied the abovementioned flow equation both analytically, in the perturbative regime, and numerically,
in the nonperturbative regime. For the nonperturbative regime we have derived the
dependence of the scalar-like four-quark interaction on the infrared cutoff scale k and on
the momentum exchanged in the t-channel. The results are summarized in Figs. 2, 3 and 4.
We find that a quasilocal one-particle-irreducible (1PI) four-quark interaction is
generated and becomes indeed approximately pointlike for scales of the order NJL , where
scalar and pseudoscalar meson-bound states are expected to play a role. The quasilocality
of this box-interaction is put in evidence in Fig. 4, where the dependence on the
momentum-scale is shown.
As discussed in Section 4, another pointlike interaction arises from instanton effects. In
Fig. 2(a) a comparison is made between the strength of the box interaction and that of the
local instanton-induced four-quark interaction. At scales of the order NJL , the instanton
interaction dominates.
Nevertheless, the contribution of the one-gluon exchange to the effective four-quark
vertex retains the characteristic momentum dependence of a particle exchange in the
t-channel. We show the different contributions to the four-quark vertex in Figs. 2, 3
and 4. The effective one-gluon exchange contribution remains substantially larger than
the pointlike box-type interactions. In consequence, the pointlike interactions (1.1) of the
NambuJona-Lasinio model cannot give a very accurate description of QCD if the gluons
are massless. Models based on a pointlike instanton induced four-quark interaction look
more favorable. Still, the one-gluon exchange is not negligible.
The results shown in Figs. 2, 3 and 4 have been obtained with the assumption of
massless gluons (in the limit of vanishing infrared cutoff scale, k 0). In Section 5
we have discussed the intriguing hypothesis that gluons acquire a mass MV 850 MeV
from spontaneous color symmetry breaking [10]. Our results for a nonvanishing gluon
mass are shown in Figs. 5 and 6. In particular, from Fig. 6, where the dependence on the
356
momentum scale is shown, it is clear that for this scenario pointlike four-quark interactions
can be defended as a much better approximation. Nevertheless, any quantitatively reliable
approximation should retain the one-gluon exchange in addition to the instanton interaction
and the NJL-interaction (1.1).
We emphasize that our approximations for the flow equation in the nonperturbative range
remain relatively crude, even if in Section 3 we have done some checks in order to test the
validity of some of the many approximations involved. Our results should, therefore, only
be trusted on a semi-quantitative level. In our opinion, they are nevertheless a good guide
for the qualitative characteristics of the effective four-quark interactions at scales between
a few hundred MeV and 1 GeV.
References
[1] Y. Nambu, G. Jona-Lasinio, Phys. Rev. 122 (1961) 345.
[2] T. Hatsuda, T. Kunihiro, Phys. Rep. 247 (1994) 221;
J. Bijnens, Phys. Rep. 265 (1996) 369.
[3] J. Berges, D. Jungnickel, C. Wetterich, Phys. Rev. D 59 (1999) 034010.
[4] C. Wetterich, Phys. Lett. B 301 (1993) 90;
C. Wetterich, Nucl. Phys. B 352 (1991) 529;
C. Wetterich, Z. Phys. C 57 (1993) 451;
C. Wetterich, Z. Phys. C 60 (1993) 461.
[5] J. Berges, N. Tetradis, C. Wetterich, hep-ph/0005122.
[6] M. Reuter, C. Wetterich, Nucl. Phys. B 408 (1993) 91;
M. Reuter, C. Wetterich, Nucl. Phys. B 417 (1994) 181.
[7] M. Bonini, M. DAttanasio, G. Marchesini, Nucl. Phys. B 418 (1994) 81.
[8] U. Ellwanger, Phys. Lett. B 335 (1994) 364.
[9] U. Ellwanger, C. Wetterich, Nucl. Phys. B 423 (1994) 137.
[10] C. Wetterich, Phys. Lett. B 462 (1999) 164;
C. Wetterich, hep-ph/9908514, to appear in Eur. Phys. J. C;
C. Wetterich, hep-ph/0008150;
C. Wetterich, hep-ph/0011076.
[11] C. Wetterich, hep-th/9501119.
[12] C. Wetterich, Z. Phys. C 72 (1996) 139.
[13] U. Ellwanger, M. Hirsch, A. Weber, Z. Phys. C 69 (1996) 687.
[14] B. Bergerhoff, C. Wetterich, Phys. Rev. D 57 (1998) 1591.
[15] G. t Hooft, Phys. Rev. Lett. 37 (1976) 8;
G. t Hooft, Phys. Rev. D 14 (1976) 3432;
G. t Hooft, Phys. Rev. D 18 (1978) 2199, Erratum.
[16] M.A. Shifman, A.I. Vainshtein, V.I. Zakharov, Nucl. Phys. B 163 (1980) 46.
[17] E.V. Shuryak, Nucl. Phys. B 203 (1982) 93.
[18] T. Banks, E. Rabinovici, Nucl. Phys. B 160 (1979) 349;
E. Fradkin, S. Shenker, Phys. Rev. D 19 (1979) 3682;
G. t Hooft, in: Recent Developments in Gauge Theories, Plenum, New York, 1980, p. 135;
S. Dimopoulos, S. Raby, L. Susskind, Nucl. Phys. B 173 (1980) 208;
T. Matsumoto, Phys. Lett. B 97 (1980) 131;
M. Yasu, Phys. Rev. D 42 (1990) 3169.

A proper-time cure for the conformal sickness

in quantum gravity
A. Dasgupta, R. Loll 1
Max-Planck-Institut fr Gravitationsphysik, Am Mhlenberg 1, D-14476 Golm, Germany
Abstract
Starting from the space of Lorentzian metrics, we examine the full gravitational path integral
in 3 and 4 space-time dimensions. Inspired by recent results obtained in a regularized, dynamically
triangulated formulation of Lorentzian gravity, we gauge-fix to proper-time coordinates and perform
a non-perturbative Wick rotation on the physical configuration space. Under certain assumptions
about the behaviour of the partition function under renormalization, we find that the divergence due to
the conformal modes of the metric is cancelled non-perturbatively by a FaddeevPopov determinant
contributing to the effective measure. We illustrate some of our claims by a 3d perturbative
calculation. 2001 Elsevier Science B.V. All rights reserved.
PACS: 04.60.-m; 04.60.Gw; 04.60.Kz; 04.60.Nc
1. Path integrals for quantum gravity

It is the central aim of path integral formulations of quantum gravity to give a physical
and mathematical meaning to the formal expression

1
Z=
(1)
D[g ]eiS ,
S[g ] =
d 4 x | det g| (R 2),
16GN
Metrics(M)
Diff(M)
for the gravitational propagator, subject to boundary conditions on the metric fields g (x).
The earliest attempts to construct a Feynman propagator for gravity [1,2] go back to a
time when neither of the present authors had been born or, well, barely. The perturbation
series for (1) around flat Minkowski space is non-renormalizable and thus cannot
serve as a fundamental definition of the theory. Assuming that a quantum theory of
E-mail addresses: dasgupta@aei-potsdam.mpg.de (A. Dasgupta), loll@aei-potsdam.mpg.de (R. Loll).
1 Address as from 1 September 2001: Institute for Theoretical Physics, Utrecht University, Minnaertgebouw,
Leuvenlaan 4, NL-3584 CE Utrecht, The Netherlands.

PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 2 2 7 - 9
358
A. Dasgupta, R. Loll / Nuclear Physics B 606 (2001) 357379
gravity does indeed exist, one is therefore forced to consider non-perturbative methods
for constructing Z. However, a non-perturbative evaluation of (1) in the continuum meets
with a number of well-known problems:
(i) Explicit field coordinates on the physical configuration space Metrics(M)
Diff(M) of diffeomorphism-equivalence classes of metrics [g ] (the so-called geometries) must be
found;
(ii) A measure D[g ] on the space of paths (the set of all d-dimensional space-time
geometries interpolating between an initial and a final spatial geometry) must be
given;
(iii) Since we are dealing with a field theory, a regularization and renormalization
respecting the diffeomorphism symmetry of the gravitational theory must be
found.
Even if good candidates (i)(iii) have been identified, we still expect difficulties with the
evaluation of the non-perturbative integral since
(iv) The action is not quadratic in the fundamental metric fields;
(v) The integral is unlikely to converge because of the imaginary factor i in front of the
Einstein action.
In ordinary quantum field theory on a fixed Minkowskian background, problem (v)
is usually solved by rotating to imaginary time, evaluating n-point functions in the
Euclidean sector and invoking the OsterwalderSchrader axioms. It is much less obvious
how to proceed in gravity, where the metric field is a dynamical variable. A generic metric
g has no geometrically distinguished notion of time t, and it is therefore unclear how to
perform a Wick rotation of the form t = it.
This difficulty has motivated some researchers to change the configuration space of the
theory, from Lorentzian space-time metrics g with signature ( + + +) to Euclidean
eu with signature (+ + + +), and simultaneously to replace the complex
metrics g
eu ]). It is important to
amplitudes exp(iS[g ]) by real Boltzmann weights exp(S[g
realize that this substitution is ad hoc in the sense of replacing one physical problem by
another one which without a non-perturbative generalization of the Wick rotation is
a priori unrelated and potentially inequivalent.
Unfortunately, because of the so-called conformal-factor problem, such a procedure
does still not guarantee the convergence of the regularized path integral. This property
is visualized most easily by decomposing the (Euclidean or Lorentzian) metric g into a
product of a conformal factor and a metric g of constant curvature according to g =
the kinematic term
e2 g . Rewriting the Einstein action as a function of and g,
(0 )2 for the conformal field is seen to contribute with the wrong sign, making the action
unbounded from below, and the functional -integration in the Euclidean case potentially
divergent.
This conformal sickness has been known since the early days of the Euclidean path
integral [3,4]. Following the suggestion of performing a conformal rotation i [3]
for asymptotically Euclidean metrics (and = 0), the typical cure consists in a suitable
integration over complex instead of real metrics g . A place where Euclidean amplitudes
359
are essential is the no-boundary proposal of Hartle and Hawking [5]. Extensive studies
of cosmological models with compact slices have been conducted in search of definite
prescriptions of complex integration contours, satisfying certain criteria of physicality and
semi-classicality [6,7]. For simple mini-superspace models such contours can be found,
but it seems very difficult to come up with a prescription for selecting a contour uniquely
which at the same time could claim some generality.
Other authors, again in a perturbative context, have insisted that the proper physical
starting point for any analysis should be Lorentzian gravity. Either by working backwards
from a continuum phase-space path integral in terms of reduced, physical variables [8] or
by gauge-fixing the configuration-space path integral and properly including the ensuing
FaddeevPopov determinants [9], they have argued that the conformal divergence is
spurious. These arguments highlight the potential importance of the measure D[g ] in
(1), an issue to which we will return in due course.
Because of the ill-definedness of the perturbative path integral, the relevance of these
considerations for a full theory of quantum gravity is not immediately clear. To our
knowledge, the conformal problem of the path integral has not been addressed in a
genuinely non-perturbative setting. This has to do with the general lack of regularizations
for gravity within which this issue could be treated in a mathematically meaningful way. In
addition, going beyond the perturbative case, a Wick rotation on the space of all metrics is
needed if one believes as we do that the Lorentzian signature and the causal structure
of space-time are of fundamental physical importance, and should therefore be built into
any quantization from the outset.
Our interest in gravitational path integrals is motivated by the recent construction of
a non-perturbative regularized path integral for gravity based on simplicial Lorentzian
geometries [1012] (see [13] for recent reviews), a Lorentzian version of previously
investigated so-called dynamically triangulated models. The model can be defined in any
dimension d, possesses a well-defined notion of Wick rotation and a set of causality
constraints reflecting the properties of the discrete Lorentzian structure.
This formulation of quantum gravity goes some way in addressing the list of problems
mentioned earlier. In the spirit of Regges old idea of describing geometries without
coordinates [14], it is defined directly on the physical space of geometries. The
(discretized) geometries are described in terms of the combinatorial data of how a set of
flat d-dimensional simplicial building blocks (whose metric properties are encoded in their
geodesic edge lengths) is glued together. This amounts to a definite prescription for (i)(iii)
above. The non-perturbative Wick rotation gets rid of the factor of i of problem (v), and the
Wick-rotated path integral can be shown to converge for a suitable choice of bare coupling
constants.
It is remarkable that a regularization for quantum gravity with such properties should
exist and it is of great interest to understand whether the path integral can be evaluated
explicitly, and simultaneously the diffeomorphism-invariant cut-off be removed to give rise
to a well-defined continuum theory. Instead of performing Gaussian continuum integrals as
in (iv), solving the model means the evaluation of the discrete combinatorial state sum
over distinct gluings. This program can be carried out exactly by analytical methods in
360
dimension d = 2 [10], yielding a well-defined propagator (1), in agreement with a (formal)

continuum calculation in proper-time gauge [15].
These results are reassuring as far as the consistency of Lorentzian dynamically
triangulated gravity is concerned, but more serious problems are expected to appear in
higher dimensions, in the case of the conformal-factor problem for d 3. Although the
discrete model always possesses a phase where Z converges, this may be attributed to the
effective curvature bounds inherent in the regularization. It does not necessarily exclude a
dominance of the unphysical conformal mode in the state sum. One can indeed identify
simplicial geometries (whose spatial volumes oscillate rapidly in proper time) with a
large and negative Euclidean action. Nevertheless, it has been established by numerical
simulations that for the 3d model there is a large range of the gravitational coupling
constant where such modes do not play a role [16,17]. This entails a win of entropy over
energy, that is, well-behaved geometries outnumber completely the potentially dangerous
ones associated with conformal excitations.
It would be very significant if the same behaviour persisted in four spacetime
dimensions, since it would suggest a resolution of the conformal-factor problem at the
non-perturbative level, where a quantum theory of gravity has a chance of existing. The
question we will address in the present work is whether and how such a behaviour can
be understood from a continuum point of view. The evidence from Lorentzian dynamical
triangulations so far suggests that a crucial contribution in the cancellation of the conformal
divergence must come from the path integral measure.
To imitate the discretized formulation as closely as possible, we will use a configuration
space path integral in terms of metric fields g . Our calculations will be done for d = 3, 4.
In order to gauge-fix, we will work with proper-time (or Gaussian) coordinates. This
is motivated by the presence of a preferred notion of (discrete) proper time in the lattice
model (although it should be pointed out that in this case there is no gauge-fixing proper
time is simply selected from the combinatorial data characterizing each geometry).
We do not expect to be able to perform the non-perturbative functional integrals
explicitly (this is problem (iv) from above), but we will show that under certain
plausible assumptions about the behaviour of the path integral under renormalization the
conformal divergence is cancelled by a compensating term in the measure, arising as
a FaddeevPopov determinant during the gauge-fixing. Our treatment will concentrate
on the conformal factor-dependence and will remain formal in the sense that we will
not spell out the details of the regularization and renormalization. However, the results
from the simplicial formulation make us confident that suitable diffeomorphism-invariant
regularization schemes do indeed exist.
The cancellation mechanism we uncover is a non-perturbative version of the one found
by Mazur and Mottola [9], and requires that C < 2/d for the constant C appearing
in the DeWitt measure, exactly the range where the DeWitt metric is indefinite. It
leads us to conjecture that the natural measure given by the dynamical triangulations
approach corresponds to a value of C < 2/d. This is quite plausible, given that the only
distinguished value of C (inherent in the action and appearing explicitly in a canonical
treatment of three- and four-dimensional gravity) is C = 2, which lies in the required
361
range.
The contents of the remainder of our paper is as follows: in Section 2 we will separate
out the gauge components of the metric tensor and discuss some properties of the propertime gauge. We also introduce our conventions for various scalar products. In Section 3,
we explicitly isolate the negative-definite part of the action responsible for the conformal
divergence. The FaddeevPopov determinants associated with the variable changes on the
space of metrics are computed in Section 4. We then show that under certain assumptions
on the renormalization behaviour of the state sum (borrowed from 2d Liouville gravity),
a piece of these determinants exactly cancels the leading conformal divergence in the
action. Section 5 contains a perturbative evaluation of the complete proper-time path
integral around a fixed constant-curvature torus metric in 3d, to illustrate the cancellation
mechanism at work in a complete and explicit calculation. In the final Section 6, we
summarize our findings.
2. Implementing the proper-time gauge

Our first task will be to split the metric degrees of freedom into physical and gauge
components, and to divide the gravity partition function by the (infinite) volume of the
diffeomorphism group. We will work with d-dimensional spacetimes M, d = 3, 4, with
topology (d)M = [0, 1] (d1), where denotes a compact spatial manifold.
In an attempt to follow as closely as possible the discrete construction of [11,12], we will
represent the physical configuration space of geometries on M (i.e., the quotient space of
spacetime metrics M = Metrics(M) and spacetime diffeomorphisms Diff(M)) by the
space of metrics in proper-time gauge, 2 which are of the block-diagonal form

0
pt
g =
(2)
, , = 0, . . . , d 1, i, j = 1, . . . , d 1.
0 gij
Our Wick rotation consists in substituting
= 1 in the Lorentzian case by = +1 in
the Euclidean case (where we define 1 := +i). For the case that the spatial (d 1)dimensional metric gij is time-independent as, for instance, in the case of the flat
Minkowski metric this prescription is equivalent to an analytical continuation in proper
time t. It is not straightforward to define an exact analogue of the discrete Wick rotation
of [11,12], which is given as an operation on discrete geometries, without the need to
introduce any coordinates. In that case, one can nevertheless choose a coordinate system
on each individual flat simplicial building blocks in which the metric tensor takes the form
(2), and the discrete Wick rotation (up to a constant rescaling of proper time) corresponds
to a sign flip of the (00)-component. Another property our Wick rotation shares with the
discrete case is the fact that it maps real Lorentzian metrics to real Euclidean metrics.
Note that unlike its discrete counterpart, our prescription does not in general map
solutions of Lorentzian gravity to Euclidean solutions. (For the dynamically triangulated
2 Note that this gauge was used by Leutwyler in one of the first path-integral treatments [2]. In the context of
the canonical path integral, a similar proper-time gauge was employed in [18].
362
Lorentzian models this is ensured in the sense that the two actions are mapped into each
other.) However, this is no obstacle to our non-perturbative construction, where
simply gives us a 1-to-1 map from Lorentzian to Euclidean geometries. All computations
are then performed in the Euclidean sector where up to regularization they are welldefined. We will not address the question of what is the most suitable way of rotating back
the result, since this will ultimately be dictated by the physical interpretation of the final,
non-perturbative partition function (obvious candidates are an inverse flip = 1 = 1
or an analytic continuation in proper time).
One may wonder why we have not adopted a prescription of the form of an analytic
continuation in time, t it. The problem is that although such prescriptions work for a
handful of metrics g with special symmetries (flat space, static solutions, etc.), they do
not exist in general. Firstly, a generic spacetime does not have a physically preferred timedirection, and the prescription is clearly not invariant under diffeomorphisms. Secondly, if
by some gauge choice one does distinguish a preferred system of coordinates (like the
Gaussian coordinates we are using), the substitution t it will in general lead to complex
metrics, defeating the purpose of making the non-perturbative path integral better defined.
Keeping track of the signature is particularly convenient in proper-time gauge and we
will work throughout with factors of . The metric gij is taken to be positive definite.
Locally on a spacetime one can always find so-called Gaussian normal coordinates in
which the metric tensor assumes the form (2), but in general one expects obstructions to
introducing such coordinates globally.
As with any gauge choice the gauge must be attainable and unique, that is, any point
g M must lie on a gauge orbit that cuts the constraint surface C (in our case defined by
the vanishing of the gauge condition, g0 0 = 0) exactly once. A necessary condition
which is easier to prove is that any g in the vicinity of C can be uniquely decomposed
pt
into a configuration g C and an infinitesimal diffeomorphism. This is demonstrated in
Appendix A.
Potential difficulties with the global implementation of the proper-time gauge have to
do with the focussing properties of time-like geodesics. Anti-de-Sitter space in 3 and
4 dimensions is an example of a solution to the classical Einstein equations where propertime coordinates do not cover the entire spacetime. The 4d metric in these coordinates
assumes the form

ds 2 = dt 2 + cos2 t d 2 + sinh2 d 2 + sin2 d 2 ,
(3)
with coordinate singularities at t = /2, where the time-like geodesics orthogonal to
the hypersurfaces t = const converge to points, as is illustrated by the Penrose diagram
in Fig. 1. This is therefore an example of a metric albeit one with rather bizarre
causality properties [19] that cannot be reached from the constrained surface C by a
diffeomorphism. (By contrast, no such problems occur for the de Sitter space, say.)
The existence of such configurations in an infinite-dimensional context is not surprising.
For example, it is well-known from the Riemannian case [20] that configurations g
with special symmetries must be excised from M to make quotient spaces of the kind
M/Diff(M) well-defined. Similarly, for the purposes of the non-perturbative path integral
363
Fig. 1. Penrose diagram of anti-de-Sitter space, illustrating the convergence of time-like geodesics.
we are only interested in capturing the properties of generic metrics, and not of
pt
sets of measure zero. In our work we assume that the diffeomorphism orbits f g
through metrics of the special form (2) are in a suitable, function-theoretic sense dense
in the space M of all metrics. 3 Since we do not have a precise definition of a suitable
quantum analogue of the space M/Diff(M) beyond formal continuum calculations, such
an assumption can ultimately only be justified by the results of a properly regularized
formulation of the theory.
We now must implement our gauge choice to isolate the physical degrees of freedom.
pt
This requires a change of coordinates g (g , f ) on M, where f labels spacetime
diffeomorphisms that map any boundaries of M into themselves. (We do not specify any
detailed boundary conditions because our main argument will not depend on them.) Such
a coordinate transformation must be accompanied by a Jacobian [2,22], whose explicit
functional form depends on the gauge condition imposed on the metric. We will determine
this Jacobian in proper-time gauge using the methods of Mottola et al. [9,23] (see also [24]
for a pedagogical introduction). We decompose an arbitrary tangent vector g |g h |g
in a point g M according to
pt
pt
h = h + + =: h + (L ) ,
where
pt
h
(4)
is the gauge-invariant piece of h defined by

pt
pt
pt
(F hpt ) = F h 0 h = h0 = 0,
(5)
and the vector field generates an infinitesimal diffeomorphism of M. Note that we are
not separating out the trace-free part of the metric at this stage. A natural scalar product for
3 As far as we can see, this assumption is not in contradiction with the well-known difficulty of using propertime coordinates in an initial-value formulation of the classical Einstein equations: in this case, given initial data
for the spatial metric gij and the extrinsic curvature Kij , caustics in the proper-time coordinate system will
generically develop at some finite (positive or negative) time if Tr K 0 initially (see, for example, [21]).
364
tangent vectors to M is given by

h, h T g = d d x | det g| h G(C) h ,
(6)
where the T stands for tensor and we will from now on suppress the dependence on the
base point g M. The DeWitt metric is

1
g g + g g + Cg g ,
(7)
2
for an arbitrary real constant C. Similarly, the scalar product for vector fields on M is

V
, = d d x | det g| g ,
(8)
G(C) =
and for scalars

S
, =
dd x
| det g| .
(9)
Note in passing that Lorentz-invariance is not an issue in defining expressions like (6), (8),
(9), since the metric manifold (M, g ) does not carry a global action of the Lorentz group
unless g is the flat Minkowski metric. (It would be a requirement if we were considering a
perturbative formulation around flat space.) On the other hand, diffeomorphism invariance
of the entire path integral will be maintained throughout our construction.
The (base-point dependent) Jacobian J [g ] is defined by
pt

[Dh ] = J Dh D ,
(10)
and can be computed by assuming Gaussian normalization conditions of the form

[Dh ] exp
h, hT = 1,
, V
[D ] exp
= 1,
2
2
(11)
pt
and similarly for h . The diffeomorphism-invariance of the measure [Dh ] has been
shown in [23]. We have introduced the subscript for the measures and scalar products
to indicate their dependence on the signature. Analogously, functional determinants of
suitable operators O are computed according to

1

T
.
h, Oh =
[Dh ] exp
(12)
2
det O
The way the -dependence is to be interpreted in the functional integrals above is as
follows. All computations are to be performed for the Euclidean value = 1 and then
continued. This continuation can be non-trivial in a relation like (12) only if the original
operator O had an explicit -dependence.
The measure [Dg ] for the full path integral (1) must be diffeomorphism-invariant and
is usually assumed to be closely related to the tangent space measure [Dh ]. Since both
the measure and the Einstein action are invariant under diffeomorphisms, we can factor out
365
the volume of the d-dimensional diffeomorphism group and are left with the path integral
(cf. [9,23])

pt pt i S [g pt ]
()
Z =
(13)
Dg J g e
.
The gravitational action in terms of the gauge-fixed metrics appearing in (13) is given by

ij kl
d
(d1)
S [g] =
(14)
R 2 G(2) (0 gij )(0 gkl ) ,
d x det gij
16G
4
pt
where (d1)R denotes the scalar curvature of the spatial metric gij , and where we have
now dropped the explicit superscript indicating proper-time gauge. The Jacobian has the
form

J = detV (F F )1 (F L) (F L),
(15)
where L defined in (4) maps vectors to symmetric tensors, and F the gauge
condition according to (5) maps symmetric tensors into vectors. We tacitly assume
that zero eigenvectors have been removed in the computation of determinants like (15).
Adjoints and determinants are defined with respect to the scalar product on d-vectors
induced by the DeWitt metric (7) at g = g pt , namely,

pt

, = d d x det gij g pt .
(16)
3. Isolating the conformal divergence

As we have already described in the introduction, the Euclidean gravity path integral in
d 3 suffers from a conformal sickness arising because of the unboundedness of the
action from below [3,9]. It is straightforward to see that the same is true for the action
(14) in proper-time gauge. To isolate the relevant kinetic terms, we decompose the timederivatives according to
0 gij = (0 gij ) + (0 gij )
ij kl (0 gkl ) + G
ij kl (0 gkl ),
: = (1 G)
(17)
onto the trace-free subspace is

into a trace part and a trace-free part, where the projector G
given by

1
2
kl
l k
k l
kl

Gij =
(18)
gij g .
i j + i j
2
d 1
The kinetic term in (14) can be rewritten as
d 2
ij kl
(0 gij ) g ij g kl (0 gkl ) + (0 gij ) g ik g j l (0 gkl )
G(2) (0 gij )(0 gkl ) =
d 1
d 2 (0 det g)2
=
+ (0 gij ) g ik g j l (0 gkl ) .
(19)
d 1 (det g)2
The first term on the right-hand side is the negative-definite trace-part. This is precisely
the kinetic term of the conformal mode one isolates in perturbative expansions around
366
Ricci-flat metrics, and which leads to the conformal divergence. The second term in (19),
coming from the trace-free directions, is positive-definite (cf. [25]). In order to make the
-dependence more explicit, we decompose the metric according to gij = e2 g ij , where
gij is a constant-curvature metric. This is always possible for the Riemannian metrics we
are considering [26], but note that in our case gij has an additional time-dependence.
We will deal with the Jacobian accompanying the coordinate change gij (, gij ) in
the next section. The complete action (14) becomes

i

d
+ (d 2)(d 3)
i

S =
d x det gij e(d3) R
16G

log det g 2
+ e(d1) 2 + (d 1)(d 2) 0 +
2(d 1)

(0 gij ) g ik g j l (0 gkl ) ,
(20)
4
and is unbounded below (for either signature) because of the wrong sign for the kinetic
term in the shifted scaling parameter
= +
1
log det g.
2(d 1)
(21)
This presents a potential problem

for the Euclidean approach where the exponentiated

2
2
action contains a term e (0 ) ( 2 is a real positive constant) which can become
arbitrarily large for strongly oscillating conformal factors. For later convenience, we
rewrite the action in terms of the shifted variable as

i
1

d

i
S =
+ (d 2)(d 3)
d x det gij d1 e(d3) R
16G

2
+ e(d1) 2 + (d 1)(d 2) 0

1
ik j l
ij kl
g g (0 gkl ) ,
(0 gij ) g g
4
(d 1)
(22)
where we have now written out the positive-definite kinetic term explicitly. Note also that
S is non-polynomial in both and gij . In three dimensions, the expression (22) simplifies
further to

2

d=3
2
2
2 + 2 0
S =
dt 4 + d x e
16G

1 ij kl
ik j l
(0 gij ) g g g g (0 gkl ) ,
(23)
4
2
with denoting the Euler characteristic of the two-dimensional spatial manifold.
367
From the point of view of the canonical formulation of gravity the presence of the
conformal divergence is rather puzzling. In that case it is clear that the conformal factor is
not a propagating degree of freedom, since it is canonically conjugate to a gauge variable
and becomes fixed by imposing the Hamiltonian constraint. In metric path integrals of the
kind we are considering, this property is not at all obvious. The natural place to look for a
cancellation of the divergence is the path-integral measure, which is a central ingredient in
any non-perturbative formulation. As we have mentioned earlier, this scenario seems to be
realized in the non-perturbative approach based on piece-wise linear Lorentzian geometries
[1012], which is one of the few well-defined regularized path integrals available that do
not rely on any fixed background geometry. Numerical investigations of the corresponding
continuum theory in d = 3 have shown that for sufficiently large bare Newton constant
there is a phase whose ground state has a stable and extended geometry, without the large
fluctuations indicative of conformal excitations [16].
This clearly non-perturbative effect has motivated us to re-examine the continuum
gravitational path integral, to understand how such a cancellation may occur when the
measure is properly taken into account. Having identified the explicit form of the conformal
divergence in proper-time gauge, we will now look for potentially compensating terms
in the measure, more precisely, relevant contributions in the form of FaddeevPopov
determinants.
4. Computing the measure and cancelling the divergence

Next we determine the measure contributions arising as a result of the coordinate
pt
pt
transformations of the previous section, g (g , f ) and g gij (gij , ).

The two functional determinants appearing in the Jacobian J in (15) are vector
determinants. The operators in their arguments are formally self-adjoint, because they are
of the form of products of operators with their adjoints. Computing the explicit operators,
one finds

1
C
0 g
= ( g0 + g0 )
2
dC + 2
(24)
for the adjoint of F , leading to the diagonal operator

F F
1
C(d 2) + 2
g0 0
= +
2
2(dC + 2)

2(d1)C+4
dC+2
,
=
1d1
2
(25)
where 1d1 denotes the (d 1)-dimensional unit matrix. We will be needing the
determinant of the inverse of this operator which for later convenience we factorize into a
scalar and a spatial (d 1)-dimensional vector determinant according to

1
2 + dC
= detS
detV F F
(26)
detVd1 (2).
2 + (d 1)C
368
The remaining terms in the Jacobian J depend on the operator

(F L) = 0 + 0
(27)
together with its adjoint,

.
0 + 0
(F L) = g0 0
(28)
Substituting in the expression 0 = (0 + 0 ) for the adjoint of 0 , we finally obtain

(F L) (F L) = g0 0 0 0
+ 0 0
g0 0 0
0 + 0 0 .
(29)
Note that the determinant of this operator can be written as a product of two determinants
of operators which are separately self-adjoint, namely,

1

detV (F L) (F L) = detV 0 0 detV 0 (F L) (F L)01

= detV 0 0 detV F L 01 F L 01

= : detV 0 0 detV K K .
(30)
We have separated out the time derivatives since we are particularly interested in
identifying terms that can cancel the divergence associated with the conformal kinetic
terms in (20), (22). The FaddeevPopov operator (29) contains terms of the same kind,
coming from eigenfunctions (x) that are rapidly oscillating in time. In the region
where |i | |0 |, this behaviour is captured by the factorized operator (0 0 ). The
factorization (30) will be used in the cancellation argument below.
We proceed similarly for the second Jacobian J, which comes from isolating the
divergent mode in the action (cf. the discussion in Section 3). For the purposes of this
section, it is not necessary to specify explicitly which variables (gij ) are used on the
and 1 G
as in (17), (18),
remainder of the configuration space. Using the projectors G
we decompose the tangent vectors as
gij = (gij ) + (gij ) ,
(31)
where the trace part is given by

kl gkl =
(gij ) = 1 G
ij
1
gij log det g 2gij
d 1
and has already appeared in (21). The Jacobian J is now defined through

1=
Dg
Dg e 2 g,gC

= J
D
Dg e 2 (g ( ),g ( )C +g ,g C ) ,
(32)
(33)
where the scalar products are taken with respect to the DeWitt metric (7) restricted to the
spatial components,

ij kl
g, gC = d d x det gij gij G(C) gkl .
(34)
369
As in Section 2 above, we assume separate Gaussian normalizations for the two functional
integrals, leading to

d

2
J1 =
(35)
D e (d1)(2+(d1)C) d x det g ( ) ,
and, therefore,

J = detS 2(d 1)(2 + (d 1)C).
(36)
We now collect all determinants to obtain

J J = detS (2 + dC)(d 1)200 detVd1 20 0 detV K K ,
(37)
where we now have decomposed also the vector determinant of the time derivatives into a
scalar and a spatial vector piece. This combined Jacobian appears in our final form of the
non-perturbative continuum path integral in proper-time gauge,

Z =
(38)
Dgij
D J J e S (gij ,) .
Unfortunately, but not unexpectedly, there is no immediate way in either d = 3 or d = 4
of evaluating these integrals since they are not of Gaussian form. For the time being we are
content with less, namely, understanding the role played by the conformal factor. For this
purpose, let us concentrate on the leading divergence in in the action (22),

2

d
(d1)
0 = k d d x e(d1) 0 + 0 0
SD = k d x e

(d1)

0 ,
= k d d x e(d1)
(39)
=
k
d d x ge
0 0
0
with the positive constant k = (d 1)(d 2)/(16G), neglecting all other terms (including
boundary contributions) in the action.
What still stands in the way of our doing the -integration in (38) is the -dependence
of the measure in the action and of the various Jacobians. Following the example of Distler
and Kawai in two-dimensional Liouville gravity [27], we make the unrigorous, but wellmotivated assumption that all measures with respect to the metric gij = e2 gij can by
suitable field redefinitions be turned into measures with respect to the constant-curvature
metric gij , where the Jacobian accompanying this variable change is assumed to be of the
same functional form as the (exponentiated) original action. 4 Applying this philosophy to
the truncated path integral, we can pull all functional determinants out of the -integral
(since they are now defined with respect to the metric gij ), resulting in

Dgij detS (2 + dC)(d 1)2 00 detVd1 2
Z =
0 0 detV K K

e k ren d d x g 0 0 +
[D]
(40)
,
4 In principle other (higher-curvature) terms may be generated in the process, but they will not affect our
cancellation argument for the conformal divergence.
370
where we have absorbed the effect of the new Jacobian in a renormalization of the
gravitational coupling constant contained in k (and we are assuming that k ren is still
positive). 5
Since takes values on the entire

real line, we can set D = D and perform the
-integral to formally obtain 1/ detS ( 0 ). As can be seen, this term is cancelled

0
by the scalar determinant in (40) provided that its prefactor is negative, that is, if C <
2/d. Obviously the determinants involved here are badly divergent and must in principle
be regularized. However, since the two terms have the same functional form we expect
the cancellation to go through independent of the regularization chosen. (What we have
in mind as a typical non-perturbative regularization of the partition function (40) is a
common frequency cutoff n0 for the entire expression Z (not just for the -integration).

0)
The regularized determinants then take the form det(n
nn0 en , where en is the
S O =

eigenvalue of the nth eigenfunction of the operator O, and where we are suppressing
degeneracies of the eigenfunctions associated with the spatial directions.)
Thus we conclude that under the plausible assumption that the conformal part of the
volume element det g can be absorbed into a redefinition of the coupling constants and
provided that the DeWitt metric is chosen with C < 2/d, the conformal divergence of
the Euclidean wick-rotated gravitational path integral is cancelled non-perturbatively by a
corresponding term in the measure, coming from a FaddeevPopov determinant.
The condition C < 2/d was found previously in the perturbative treatment of Mazur
and Mottola [9], with a similar cancellation mechanism. We differ from their and other
authors treatments by obtaining the configuration space of Euclidean gravity through
a non-perturbative Wick rotation of the gauge-fixed Lorentzian path integral Z (1) ,
expression (13). Wick rotation in proper-time gauge corresponds to a straightforward
substitution in our formulas. In other gauges, there is no immediate relation
between the Euclidean and the Lorentzian sectors beyond the perturbative regime around a
fixed (typically flat) background metric. In these cases, even if any non-perturbative results
could be obtained for Euclidean signature, their implications for the physical, Lorentzian
sector would be unclear.
By contrast, we have shown under what conditions a cancellation of the conformal
divergence may take place in the full, non-perturbative theory of Lorentzian space
times. This lends further support to the finding of the quantum gravity model obtained
from Lorentzian dynamical triangulations, whose effective measure (including entropy
contributions) apparently leads to a non-perturbative cancellation of the conformal
sickness of the action. The restriction C < 2/d found in the continuum is not unnatural
in the sense that this parameter region contains the only dynamically distinguished value
C = 2 of the constant C (found after Legendre-transforming the gravitational action in
d = 3 and 4).
5 There is a somewhat related path-integral treatment by Mazur, based on a conformal decomposition of
Riemannian spacetime metrics [28]. However, he concentrates on boundary rather than bulk terms in the
effective action.
371
Even with our assumption of the absorption of the scaling factors e2 , it is unlikely
that one can make much progress in computing the continuum path integral in proper-time
gauge explicitly, at least in four dimensions. Note that substantial simplifications occur in
the case d = 3, where there are no further explicitly -dependent terms in the part of the
action indicated by the dots in formula (40). In that case, it remains to evaluate

K
ei k ren S ,
0 detV K
Zd=3 =
(41)
Dgij detVd1 2
0
with the action

d=3
S = dt 4 + d 2 x det gij

1 ij kl
ik j l
2 (0 gij ) g g g g (0 g kl ) .
4
2
(42)
Although this expression looks now tantalizingly simple, it is still non-polynomial in the
remaining metric components. We will not attempt here to evaluate (41) further, but it
would clearly be interesting to relate it to any one of the known exact results obtained in
other approaches to three-dimensional quantum gravity (see, for example, [30]).
5. Perturbative evaluation of the 3d path integral

Although it was not our main motivation, one can take our formulation as the starting
point for a perturbative expansion around a given classical solution. Depending on the
solution one is interested in, the proper-time gauge is not necessarily the most convenient
gauge choice perturbatively. Also, we do not expect to find anything new compared with
previous calculations using other gauges. Nevertheless, the calculation we will perform
illustrates the general procedure outlined in the main part of the paper and gives an explicit
example of the cancellation mechanism at work.
We will study the path integral (38) by computing its perturbative expansion Z (2) around
a particular classical solution. For the sake of simplicity, we choose the spatial slices to
be flat two-tori (corresponding to a vanishing cosmological constant ), such that M =
[0, 1] T 2 . There is a two-parameter set of classical solutions,

0
0
1
V 12
g = 0 V 2
(43)
,
0 V
1
2
12 +22
2
in a proper-time coordinate system (t, x1 , x2 ), where the xi are periodic and rescaled to
have period 1 [30,31]. The metrics are parametrized by two modular parameters , and
V denotes a constant spatial area. We will perform our perturbative calculation around any
one of the straight torus solutions with (1 , 2 ) = (0, ), > 0, where the metric takes
the diagonal form

0
g
(44)
= diag , V 1 , V .
372
Our starting point is the partition function

S(2) (h
(2)
ij , ) ,
Z =
Dhij
D J J e
with the action given by

ik j l

2 0 h
g
S(2) =
g
h
dt d 2 x det gij 2(0 )
,
0
ij
kl
16G
4
(45)
(46)
and where both expressions are to be evaluated at the base metric (44). Since our end
result will not depend on in a non-trivial way, we will from now on simply work with
the Euclidean value = 1 and drop the subscripts . We are using the notation hij = gij
0 M, and decompose them according to (31), (32). We will
for the tangent vectors at g
parametrize the directions h
ij perpendicular to the trace part explicitly as functions of the
spatial vector fields i and infinitesimal modular parameters , such that
()

k
, () ( )ij
h
ij = i j + j i gij k +

gij
()
= : L
+ () , () ( )ij , with ij :=
.
(47)
ij
This variable change is motivated by the standard decomposition of two-dimensional

Riemannian metrics [29,30]
x , t) = e2( x,t )fx ,t gij (
x , t),
gij (
(48)
followed by a shift of the conformal factor, cf. (21). In the decomposition (48),
gij (
x , t) is one of a set of constant-curvature metrics, labelled by Teichmller parameters
, and f a spatial diffeomorphism, with generators i . Note that as a consequence of our
gauge-fixing procedure, all quantities in (48) carry an additional proper-time dependence.
To avoid misunderstandings, let us also point out that the diffeomorphisms f (which
act in a standard way on the coordinates and gij s, and map surfaces of constant t into
themselves) are of course no longer invariances of the gauge-fixed action.
The ()ij form a basis for ker L , that is, for the transverse traceless tensors, and we
have chosen them to be orthonormal with respect to the scalar product , (which involves
an integration over the spatial directions only). Explicitly, they are given by

V 0 1
V 1/
0
,
(2) =
.
(1) =
(49)
0
2 1 0
2
Associated with the variable change h

ij (i , ) is another Jacobian J which takes the
form

J = detV2 L L det () , () .
(50)
A straightforward calculation yields

detV2 L L = detV2 2g ij i j detV2 (2)
(51)
and

2V
det () , () = 2 .
(52)
373
After the coordinate transformation on the tangent space of metrics, the partition function
reads

(2)
Z (2) = J JJ
(53)
Di
D
D eS (, , ) ,
with the action

2 1 ij
V
2
g
i j
dt d x 2 0 +
16G
2

2
1
V
dt 0 .
+
16G 2
S (2) =
(54)
We have pulled out the determinants from under the functional integrations in (53), because
they depend only on the fixed background metric g 0 .
We can now perform the integrations over i , and , since the action is quadratic in
these variables. Up to irrelevant (positive) constant terms, the two integrations yield:

1/2
,
i -integral: detV2 (02 )()
(55)
2 1/2
-integral:
detS (0 )
(56)
,

2 1
-integral: det (0 ) ,
(57)
where by definition all zero-modes have been excluded, and the last determinant is that
1 V
of a free two-dimensional particle with mass m = 16G
. The term coming from the
-integration is of course the ill-defined determinant associated with the conformal

divergence.
since the flat
A well-known subtlety arises in the evaluation of the determinant of (L L)
torus possesses Killing vectors, leading to zero-modes of this operator [24,30,32]. This can
be taken care of by writing

detV2 L L = VK1 detV2 L L ,
(58)
where VK denotes the (infinite) volume of the diffeomorphism subgroup generated by the
Killing vectors. Since we do not keep track of positive constant and infinite factors, we
will simply drop this term. Also, it is not our aim here to investigate the possible physical
significance of such zero-modes and we will remove them from now on wherever they
occur.
Collecting all the determinants, the partition function is now given by
Z (2)
1

2
detV2 02 detV2 ()detS 02 det 02

det, detS 4(2 + 3C)detV (F L) (F L)

.
=

det 02
detV2 02 detS 02
J JJ
(59)
The only non-trivial task remaining is the evaluation of the FaddeevPopov determinant
detV (F L) (F L). At the metric g 0 M, we can compute this determinant explicitly,
374
so there is no need to split off the time derivatives as we did in (30). As a warm-up, let us
calculate the functional determinant detV (0 0 ) detV (02 ). In order not to have to deal
with non-trivial boundary terms we demand that the eigenfunctions on M = [0, 1] T 2 be
periodic in the x1 - and x2 -direction, as well as in the time direction. A complete set of
eigenfunctions is then given by
(1 ,2 ,,) = () ei(1 x1 +2 x2 ) eit ,
with

(0) =

1
2
0 ,
V
0
(60)
0
2
(1) =
1 ,
(2) =

0
2 0 .
(61)
The frequency takes discrete values = 2k, k = 1, 2, 3, . . . , and similarly the i

are given by i = 2ki , ki = 0, 1, 2, . . . . The eigenvectors are orthonormal with respect
to the scalar product (16), with discrete eigenvalues
(1 ,2 ,,) = 2 .
(62)
The functional determinant is therefore the infinite product

detV 02 =
6 .
(63)
,1 ,2
Next we determine the eigenfunctions of the FaddeevPopov operator in

(F L) (F L) = .
(64)
Making an ansatz of the form

= (t)ei x ,
(65)
for the three-vectors, one obtains a coupled set of eigenvalue equations, namely,

1
1 2
1
2
2
40 +
1 + 2 0 i1 0 1 i2
0 2 = 0 ,
V
V
V
(66)
i10 0 02 1 = 1 ,
(67)
i20 0 02 2
(68)
= 2 .
(We note in passing that it is clear from (67), (68) that in the presence of boundaries at
t = 0, 1 it would not be consistent to require a simultaneous vanishing of all components
of at the boundaries.) One third of the eigenfunctions is easily found by setting 0 = 0.
The eigenvectors are of the form
0
(1 ,2 ,,0) = 1 k2 ei x eit ,
(69)
k1
with eigenvalues
(1 ,2 ,,0) = 2 .
(70)
For the remaining eigenfunctions one finds

(1 ,2 ,,) = 2 1 ei x eit ,
375
(71)
2 2
with the associated eigenvalues

1
1
1
(k1 ,k2 ,,) =
52 +
12 + 22
2
V

1
1
1 2 2
2
2
164 .
5 +
1 + 2
2
V
(72)
For fixed 1 , 2 and , there are therefore three orthogonal eigenfunctions. The entire
functional determinant is thus given by

(1 ,2 ,,0) (1 ,2 ,,+) (1 ,2 ,,)
detV (F L) (F L) =
=0,1 ,2
46 .
(73)
=0,1 ,2
This coincides (up to a constant factor) with the determinant of the kinetic term we
calculated earlier, that is, detV 0 0 . Substituting the results for the determinants back
into (59), we observe again an exact cancellation of the infinite products provided that
C < 2/3, in agreement with our earlier non-perturbative results. Since the regularized
determinant (57) gives a term proportional to V / , our final result for the perturbative
partition function around flat torus metrics of the type (44) is given by
Z (2) =
det, V 2
3.
det (02 )
(74)
6. Conclusions
Inspired by recent attempts of constructing a non-perturbative propagator for gravity by
discrete methods, we have investigated the continuum gravitational propagator in propertime gauge, concentrating on the role played by the conformal mode of the metric. Our
starting point was the space of physical spacetime metrics of Lorentzian signature.
After performing a generalized Wick rotation, the partition function becomes real, but the
Euclideanized action is seen to suffer from the usual conformal sickness: as a result of
conformal excitations it is unbounded below.
We then proceeded to determine the FaddeevPopov determinants that arise during the
coordinate changes on the space of metrics, the first from splitting off the gauge degrees
of freedom associated with the diffeomorphisms of the base manifold, the second from
isolating the part of the metric that has a negative-definite kinetic term. Although an explicit
evaluation of the functional determinants and the non-perturbative path integral seem
376
technically out of reach, we showed that under certain assumptions about the behaviour
of the partition function under renormalization the conformal divergence in the action
is cancelled by a corresponding FaddeevPopov term in the measure. This conclusion
also required that the signature of the DeWitt metric was chosen to be indefinite, i.e., the
constant C in the DeWitt measure (used to define inner products on the tangent space
of metrics) satisfied C < 2/d which we argued was a natural condition. Our work can
therefore be seen as a non-perturbative generalization of earlier findings by Mazur and
Mottola [9] which although acknowledged by the authorities [33] are maybe not
widely appreciated.
Our results reinforce the evidence coming from dynamically triangulated formulations
of quantum gravity [16] that in a fully non-perturbative path integral the conformal mode
is not a propagating degree of freedom and therefore the conformal divergence is simply
absent, contrary to what one may have expected from just looking at the action or from
considering reduced, cosmological models. Although geometric configurations with large
and negative action exist, they are effectively suppressed by the non-trivial path-integral
measure. This is an attractive proposition because it implies that the unboundedness of the
action is no obstacle in principle to the construction of a gravitational path integral. Also,
at a kinematical level (that is, before quantum gravity proper has been solved), it brings the
covariant formulation of gravity into line with canonical treatments where the conformal
degree of freedom is fixed by imposing the Hamiltonian constraint. Although in the path
integral this constraint is not enforced explicitly, it seems that the measure nevertheless
does know about it.
All our calculations were done in proper-time gauge, mimicking a similar procedure
in the discrete Lorentzian gravity models of [10,11]. Note that there is nothing wrong in
principle with choosing a gauge in quantum gravity. Our choice of proper time does have
an invariant geometric meaning, but this by no means entails that proper-time correlators
assume a simple form in other gauges or that any interesting physical quantity will be
easily expressible in proper-time gauge. This is simply an inevitable feature of gauge
theories. Obviously we expect our result about the absence of the conformal sickness to
be gauge-independent. In practice, an explicit check may not be straightforward, since
our construction of a Wick rotation was closely tied to the proper-time gauge. There may
be other gauge choices and other prescriptions of Wick-rotating, but we are currently not
aware of any concrete proposals. As usual in quantum gravity, one is not exactly faced with
an embarrassment of riches when trying to quantize the theory.
Clarifying the role of the conformal factor is only one step in analyzing the properties
of non-perturbative path integrals for Lorentzian gravity. We do not expect that much
progress can be made in a continuum formulation in evaluating these quantities explicitly
and showing how the non-renormalizability of the perturbative approach is resolved nonperturbatively. For a solution of these more fundamental problems we must rely on the
properly regulated discrete quantum gravity models that are currently under investigation.
377
Acknowledgement
We are grateful to J. Ambjrn for discussion and comments. R.L. acknowledges the
support of EU network grant HPRN-CT-1999-00161.
Appendix A
In this appendix we show that every metric g in an infinitesimal neighbourhood of the
constraint surface C can be uniquely decomposed into an element in C and an infinitesimal
diffeomorphism, parametrized by a vector field . Writing the original metric as g =
pt
g + h , we would like to understand under which conditions the tangent vector h can
be uniquely decomposed as
pt
h = h + + ,
where
pt
h
pt
(A.1)
is tangential to C, i.e., of the form

0 0
.
=
0 hij
(A.2)
This is tantamount to solving the set of equations

20 0 = h00 ,
(A.3)
0 i + i 0 = h0i
(A.4)
for , where the covariant derivatives refer to the base point metric g . Since this metric
is in proper-time gauge, its Christoffel symbols take the form
pt
0
= 0,
00 = 0
1
lj0 = glj,0 ,
2
1
i
0j
= g ik gj k,0 ,
2
1
lji = g ik (glk,j + gj k,l glj,k ).
2
Using these explicit expressions we obtain
1
0 (t, x) =
2
t
dt h00 (t , x),
(A.5)
(A.6)
t i + aij j = bi (t, x),
(A.7)
i
. The system of differential equations determining
where bi = h0i i 0 , and aij = 20j
i has unique solutions if
(1) bi and aij are smooth and continuous in the interval of interest, namely, [0,t];
(2) the aij satisfy a Lipschitz condition
d1

|aij | < kj ,
i=1
where the kj are arbitrary constants, greater than zero.
(A.8)
378

pt
These conditions are satisfied for all metrics g , which we assume to be nondegenerate and sufficiently differentiable. The boundedness property (A.8) follows from
the compactness of the base space (d) M = [0, t] (d1).
References
[1] C.W. Misner, Feynman quantization of general relativity, Rev. Mod. Phys. 29 (1957) 497509.
[2] H. Leutwyler, Gravitational field: equivalence of Feynman quantization and canonical quantization, Phys. Rev. 134 (1964) B11551182.
[3] G.W. Gibbons, S.W. Hawking, M.J. Perry, Path integrals and the indefiniteness of the
gravitational action, Nucl. Phys. B 138 (1978) 141150.
[4] S.W. Hawking, in: S.W. Hawking, W. Israel (Eds.), General Relativity: an Einstein Centenary
Survey, Cambridge Univ. Press, Cambridge, 1979, pp. 746789.
[5] J.B. Hartle, S.W. Hawking, Wave function of the universe, Phys. Rev. D 28 (1983) 29602975.
[6] J.J. Halliwell, J. Louko, Steepest descent contours in the path integral approach to quantum
cosmology. The de Sitter minisuperspace model, Phys. Rev. D 39 (1989) 22062215;
J.J. Halliwell, J. Louko, Steepest descent contours in the path integral approach to quantum
cosmology. Microsuperspace, Phys. Rev. D 40 (1989) 18681875;
J.J. Halliwell, J. Louko, Steepest descent contours in the path integral approach to quantum
cosmology. A general method with applications to anisotropic minisuperspace models, Phys.
Rev. D 42 (1990) 39974031.
[7] J.J. Halliwell, J.B. Hartle, Integration contours for the no-boundary wave function of the
universe, Phys. Rev. D 41 (1990) 18151834.
[8] K. Schleich, Conformal rotation in perturbative gravity, Phys. Rev. D 36 (1987) 23422363.
[9] P.O. Mazur, E. Mottola, The path integral measure, conformal factor problem and stability of
the ground state of quantum gravity, Nucl. Phys. B 341 (1990) 187212.
[10] J. Ambjrn, R. Loll, Non-perturbative Lorentzian quantum gravity, causality and topology
change, Nucl. Phys. B 536 (1998) 407434, hep-th/9805108.
[11] J. Ambjrn, J. Jurkiewicz, R. Loll, A nonperturbative Lorentzian path integral for gravity, Phys.
Rev. Lett. 85 (2000) 924927, hep-th/0002050.
[12] J. Ambjrn, J. Jurkiewicz, R. Loll, Dynamically triangulating Lorentzian quantum gravity,
preprint Golm AEI-2001-049.
[13] J. Ambjrn, J. Jurkiewicz, R. Loll, Lorentzian and euclidean quantum gravity analytical and
numerical results, in: L. Thorlacius, T. Jonsson (Eds.), M-Theory and Quantum Geometry,
NATO Science Series, Kluwer Academic, 2000, pp. 382449, hep-th/0001124;
R. Loll, Discrete Lorentzian quantum gravity, Nucl. Phys. B Proc. Suppl. 94 (2001) 96107,
hep-th/0011194.
[14] T. Regge, General relativity without coordinates, Nuovo Cimento A 19 (1961) 558571.
[15] R. Nakayama, 2-d quantum gravity in the proper time gauge, Phys. Lett. B 325 (1994) 347353,
hep-th/9312158.
[16] J. Ambjrn, J. Jurkiewicz, R. Loll, Non-perturbative 3d Lorentzian quantum gravity, preprint
Golm AEI-2000-074, hep-th/0011276, to appear in Physical Review D.
[17] J. Ambjrn, J. Jurkiewicz, R. Loll, Computer simulations of 3d Lorentzian quantum gravity,
Nucl. Phys. B Proc. Suppl. 94 (2001) 689692, hep-lat/ 0011055.
[18] C. Teitelboim, Proper-time gauge in the quantum theory of gravitation, Phys. Rev. D 28 (1983)
297309.
[19] S.W. Hawking, G.F.R. Ellis, The Large Scale Structure of SpaceTime, Cambridge Univ. Press,
Cambridge, 1973.
379
[20] A.E. Fischer, The theory of superspace, in: M. Carmeli et al. (Eds.), Relativity, Proc. of the
Relativity Conf. in the Midwest, Plenum, New York, 1970, pp. 303357.
[21] L.D. Landau, E.M. Lifshitz, The Classical Theory of Fields, 4th rev. English edn., Butterworth
Heinemann, Oxford, 1998.
[22] L.D. Faddeev, V.N. Popov, Covariant quantization of the gravitational field, Usp. Fiz. Nauk 111
(1973) 427, Sov. Phys. Usp. 16 (1974) 777788.
[23] Z. Bern, S.K. Blau, E. Mottola, General covariance of the path integral for quantum gravity,
Phys. Rev. D 43 (1991) 12121222.
[24] E. Mottola, Functional integration over geometries, J. Math. Phys. 36 (1995) 24702511.
[25] B. DeWitt, Quantum theory of gravity. The canonical theory, Phys. Rev. 160 (1967) 11131148.
[26] R. Schoen, Conformal deformation of a Riemannian metric to constant scalar curvature, J. Diff.
Geom. 20 (1984) 479495.
[27] J. Distler, H. Kawai, Conformal field theory and 2D quantum gravity, Nucl. Phys. B 321 (1989)
509527.
[28] P.O. Mazur, Quantum gravitational measure for three-geometries, Phys. Lett. B 262 (1991)
405410.
[29] A.M. Polyakov, Quantum geometry of bosonic strings, Phys. Lett. B 103 (1981) 207210.
[30] S. Carlip, Quantum Gravity in 2 + 1 Dimensions, Cambridge Univ. Press, Cambridge, 1998.
[31] V. Moncrief, Reduction of the Einstein equation in 2 + 1 dimensions to a Hamiltonian system
over Teichmller space, J. Math. Phys. 30 (1989) 29072914.
[32] M. Seriu, Partition function for (2 + 1)-dimensional Einstein gravity, Phys. Rev. D 55 (1997)
781790.
[33] S.W. Hawking, private communication to A.D.

Covariant quantization of the CBS superparticle

P.A. Grassi a , G. Policastro a,b , M. Porrati a
a Physics Department, New York University, 4 Washington Place, New York, NY 10003, USA
b Scuola Normale Superiore, Piazza dei Cavalieri 7, Pisa 56100, Italy
Abstract
The quantization of the CasalbuoniBrinkSchwarz superparticle is performed in an explicitly
covariant way using the antibracket formalism. Since an infinite number of ghost fields are required,
within a suitable off-shell twistor-like formalism, we are able to fix the gauge of each ghost sector
without modifying the physical content of the theory. The computation reveals that the antibracket
cohomology contains only the physical degrees of freedom. 2001 Elsevier Science B.V. All rights
reserved.
1. Introduction
The correct quantization of systems with an infinitely reducible gauge symmetry is a
longstanding problem and, recently, several new efforts (see, e.g., [1]) have been made
toward the construction of a covariant quantization procedure for the GreenSchwarz
superstring model [2]. Indeed, the GS superstring is the most important and interesting
model enjoying the feature of infinite reducibility, but since the model is very difficult
to handle in its complexity, the study of simpler models provides a good test for the
quantization techniques. This is essentially the reason why, in the past decades, people
devoted several efforts trying to quantize the superparticle model of CasalbuoniBrink
Schwarz type [3].
In the case of GS superstring and superparticle, there are first- and second-class
constraints [47]. The occurrence of second-class constraints arises from the fact that
the Grassman momenta P , conjugate to the fermionic variables , are non-independent
phase-space variables. If, as a formal procedure, one attempts to construct Dirac brackets,
treating all the fermionic constraints as if they were second class, the resulting expressions
are singular. An alternative procedure would be a careful separation of first- and secondE-mail address: pag5@nyu.edu (P.A. Grassi).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 2 2 5 - 5
P.A. Grassi et al. / Nuclear Physics B 606 (2001) 380400
381
class constraints, but, in that way, a covariant quantization procedure is impossible to

achieve.
In order to maintain manifest Lorentz covariance, one has to exploit the -symmetry
[2,8] of the model, which cancels half of the fermionic degrees of freedom realizing the
matching between the bosonic and the fermionic states. Unfortunately, the -symmetry
is a reducible local fermionic symmetry. This means that, in reality, only four degrees of
freedom are effectively canceled at the first stage. Pursuing the analysis, it is easy to show
that an infinite tower of ghosts is necessary to match the correct number of degrees of
freedom. In terms of the Hamiltonian formalism [6], this is equivalent to the statement that
the constraints are infinitely reducible: there exist not only linear vanishing combinations
of constraints, but also zero modes of those relations, and so on to infinitely many levels
(for further details see, for example. [9,10]).
Several attempts to a solution of the problem can be found in the literature. In particular,
we would like to mention the idea of changing the classical constraints in order that all
the second-class constraints are transformed into first-class ones [11,12]. This essentially
yields an extension of the phase-space where the -symmetry is gauged by means of
suitable fermionic gauge fields. Other approaches to the superparticle quantization are
the harmonic superspace [13,14] and the construction of models based on a given BRST
operator which selects the correct physical subspace [15]. Nevertheless, all of these
approaches share the common feature of infinite number of classical and ghost fields.
Finally, the computation of the gauge-fixed BRST characteristic cohomology [16,17]
for the superparticle model received considerable attention [15,1820]. It has been shown
in [20] that, due to the infinitely many interacting fields, the canonical transformations
performed to implement the gauge fixing turn out to be ill-defined. Moreover, as a
consequence of further gauge symmetries of the gauge-fixed action, a two-step gaugefixing procedure is required and it may cause problems, as argued in [15].
The aim of our paper is to present a new solution of the application of the
BVBRST formalism ([21,22]; for a review, see [5,23,24]) to the problem of the
superparticle. In particular, we take the advantage of the existing literature to make
essential steps towards the complete solution. The main issue here is the construction of
a procedure to quantize the CasalbuoniBrinkSchwarz classical action, computing the
correct antibracket map and the corresponding cohomology. The present technique is based
on canonical transformations which implement the gauge fixing of the nth order ghosts
without affecting the (n 1)th order.
The structure of the superparticle model involves word-line diffeomorphisms, onshell closure of the algebra of symmetries and an infinite tower of ghost fields. To
take into account the open algebra, the BVBRST formalism is implemented and the
solution of the master equations contains quadratic terms in the antifields. On the
other hand, the -symmetry of the model entails field-dependent transformations which
produce interactions among the ghost fields of different levels and the superparticle fields.
Therefore, after the gauge fixing, an infinite number of ghosts fields interact, among
themselves and with the physical fields. In this situation, two types of problems arise:
382
(i) the renormalizability of the model may be lost due to an infinite number of
parameters;
(ii) the computation of the gauge-fixed BRST cohomology and the definition of the
physical spectrum of the model are ill-defined.
In order to solve the problem of interactions among the ghost fields of different levels, we
apply the idea of P. Townsend [25] for the chiral superfield to the superparticle model. The
basic idea is a suitable redefinition of ghost fields, in such a way that they appear decoupled
from the rest of the theory. The complete formulation of Townsends idea has already been
discussed in [26]. There, it was shown that, in the case of the gauged, complex, linear
superfield model [27], a convenient redefinition of ghost fields realizes the decoupling and
the quantization can be performed. In addition, it was shown that a careful application
of the technique of [27] provides a quantization procedure with the correct antibracket
cohomology. The realization of Townsends idea in the superparticle framework requires
further efforts.
Following the twistor-like formulation [28], we introduce twistor variables to re-express
the ghost fields in a decoupled fashion. In the literature only the on-shell twistor-like
formalism is available. Unfortunately, this is not suitable for our purposes. Therefore,
we extended the twistor-like formalism in order that the ghost fields can be conveniently
redefined off-shell. In past years, some interesting formulations of the superparticle model
have been discussed in the context of the N = 8 word-line supergravity framework [28].
However, we do not follow these directions and our twisted-ghost formalism only amounts
to a redefinition of classical variables.
Even after the twisting, the zero modes of the -symmetry (the ghost fields) turn
out to interact with the physical fields; however, the interaction terms are proportional
to the einbein equation of motion (Virasoro constraints) and, therefore, these couplings
can be eliminated by a simple canonical transformation (cf. Ref. [23]). Although the
fermion generator of these transformations contains a formal sum on infinite ghost fields,
it can be easily verified that each individual canonical transformation involves a finite
number of fields [15,20] and it has a well-defined inverse. In this way, the cohomology
of the antibracket is not changed and these transformations do not affect the physical
observables. After twisting and canonical transformations, the ghost fields are free and
decoupled from the rest of the fields. In addition, the ghost fields are off-shell zero modes
of the -symmetry. Comparing with [26,27], we stress again the fact that the quantization
procedure, in the presence of an infinite number of ghosts, can be performed only in the
case of off-shell decoupling.
Although the technique of [27] appears very promising for the superparticle models, as
it has been shown in [26], it cannot be easily implemented in the present context. This is
due to the structure of the antifield-dependent part of the action. The diagonalization of
the fields necessary to achieve the decoupling among the ghosts of different levels induces
new couplings in the antifield terms which cannot be removed. Therefore, in order to apply
safely the quantization procedure of [26,27], one has to reach the third order of ghost field,
at least, to circumvent the problem. Unfortunately, despite several efforts, it seems that
there is no way to apply the aforementioned procedure to the superparticle case. However,
383
we can follow a similar strategy: we quantize the nth ghost field without affecting the
(n 1)th level, but the actual realization is different.
We choose the conventional Lorentz invariant gauge fixing proposed in [11,15,18] to fix
the gauge of the spinor field and for the ghost fields. In particular, following [15,29], we
take into account the Stckelberg symmetry arising in the non-minimal sector. The latter
contains Lagrangian multipliers, anti-ghost fields and the non-minimal extra-ghost fields.
Contrary to [15,30], we use the Stckelberg symmetry at each step of the quantization
procedure in such a way that all degrees of freedom are gauge fixed and their dynamics is
precisely prescribed.
Along the quantization procedure, we compute the antibracket cohomology and we
check that the sequence of canonical transformations does not modify the physical sector
of the Hilbert space. In addition, we compute the cohomology group H ( , H ()) of the
completely gauge-fixed theory where and are defined in [16,24] and the result
coincides with the correct physical spectrum.
The outline of the paper is the following. In Section 2, we briefly review the antifield and
the antibracket formalism with particular attention to the definition of physical observables.
In Section 3, we present some details about the strategy of the quantization procedure.
In Section 4, we recall the solution of the master equation in order to establish our
conventions. Next, in Section 5 we introduce the twistor-like variables and we define
the off-shell zero modes. Finally, in Section 6 the gauge fixing of the complete tower of
ghost fields is implemented. Section 7 contains the computation of the cohomology and
concluding remarks. An Appendix A details our notations.
2. Symmetries and the anti-bracket formalism

Essential ingredients of the BatalinVilkovisky (BV) formalism are anti-fields and anti with opposite statistics.
brackets [2224]. For any field A , one introduces an anti-field A
A ghost number is assigned to every field, in such a way that for the classical fields
) =
gh() = 0, the (extended) action has ghost number zero, and for all fields gh(A
A
gh( ) 1.
The anti-bracket between two functions F and G of the fields and anti-fields is defined
by
(F, G) =
l
r
l
r
F G F A G,
A
A
A
(2.1)
where r/ l denote the derivative from the left or from the right. The antibrackets satisfy
graded commutation, distribution, and Jacobi relations [22]. With these brackets, fields
and anti-fields behave as coordinates and their canonically conjugate momenta

A
A B
A , B = 0,
, B = BA .
, = 0,
(2.2)
The extended action S(, ) is a solution of the master equation
(S, S) = 0,
(2.3)
384
which satisfies certain boundary conditions: (1) it coincides with the classical action Scl ()
are coupled to the gauge
when the ghost fields are set to zero; (2) the anti-fields A
transformations of the classical fields; (3) it satisfies the properness requirement, i.e., the
2N 2N matrix
l r S
,
A B
A, B = 1, . . . , N,
S
S
has rank N on the stationary surface, defined by the equations of motion
=
= 0.
Notice that there is a natural grading among the fields and the anti-fields, namely, the
antighost number [24]. This allows a convenient decomposition of the extended action and,
accordingly, the master equation can be easily solved. Because of the boundary conditions,
a solution of the master equation does not only give the dynamics of the system, as it
generates the equations of motion; it also encodes the gauge structure, as it generates the
BRST transformations on a function X of fields and antifields via
sX = (X, S).
The master equation implies at once that S itself is invariant under this transformation,
and that s is a nilpotent differential: s 2 = 0. The second boundary condition ensures
that s starts as the classical BRST differential, when acting on fields, but it is
automatically equipped with the modifications that are necessary when the classical BRST
transformations do not define a nilpotent operator; this is the case of an open algebra, where
the commutator of two gauge symmetries yields a gauge symmetry only on-shell.
Finally, the properness condition tells us that the extended action has N on-shell
independent gauge symmetries, and this is just the number that is necessary to eliminate
unphysical anti-fields and quantize only the fields. This is essential, since in the end one
wants to perform a path integral only on the space of fields configurations; still, the
antifields do not fall completely out of the scene at this point: they can play the role of
classical sources for the BRST transformations.
It is useful, in the analysis of the cohomology of the BRST operator, to introduce another
grading, called anti-field number, whose value is 0 for the fields, 1 for the anti-fields of
classical fields, 2 for anti-fields of ghosts, 3 for anti-fields of ghosts for ghosts and so on.
The BRST operator can be decomposed accordingly:

s (k) .
s =+ +
k0
The terms in the extended action all have non-negative antifield number, and the antibracket
operation lowers it by one, so that the decomposition starts with degree 1, and
the corresponding operator which is necessarily nilpotent is called the KoszulTate
differential. It acts trivially on fields, while its action on an anti-field yields the equation of
motion of the corresponding field:
=
A
l S
.
A
385
The next term, , has antifield number 0. Its action on fields is given by

A = A , S =0 .
Canonical transformations are an important part of the formalism [23]. They preserve
the anti-bracket structure (2.1): calculating the anti-brackets in the old or new variables is
the same, or in other words the new variables also satisfy (2.2). Therefore, also the master
equation (S, S) = 0 is preserved.
B
Canonical transformations from {, } to { , }, for which the matrix r A
is invertible, can be obtained from a fermionic generating function F (, ) with

gh(F ) = 1. The transformations are defined by
A =
l F (, )
,

A
A
=
r F (, )
.
A
(2.4)
Infinitesimal canonical transformations are produced by

A
F = A
+ f , ,
the first term is the identity operator in the space of fields-antifields. In the following, we
will have to use different types of transformations.
Point transformations are the easiest ones. These are just fields redefinitions
f A () which thus determines the
A = f A (). They are generated by F = A
corresponding transformations of the anti-fields. The latter replace the calculations
of the variations of the new variables.
Adding the BRST transformation of a function s () to the action is obtained by a
A + (). The latter gives
canonical transformation with F = A
A = A ,

A
= A
+ A (),
(2.5)
and by means of these transformations it is possible to implement the gauge fixing

conditions as a canonical change of variables.
It is possible to redefine the symmetries by adding equations of motion (trivial
symmetries). This is obtained by
A
AB
+ A
B h ().
F = A
(2.6)
The canonical transformations leave by definition the master equation invariant, and
because they are non-singular, they also preserve the properness requirement on the
extended action. This point will be discussed at length in the following. Of course, in the
new variables, we do not see the classical limit anymore. But the most important property
is that the anti-bracket cohomology [16] (and the BRST cohomology) is not changed.
3. Strategy
Before entering into the details of the analysis of the CBS superparticle [3], we illustrate
the main steps of the quantization procedure of systems with an infinitely reducible gauge
386
symmetry. In particular, we describe each single canonical redefinition of the field variables
needed to bring the action in a form which is suitable for path integral quantization. The
main point is to find out a correct gauge fixing procedure which fixes the gauge degrees
of freedom, and to construct the physical subspace as the characteristic cohomology of the
BRST operator. In the present section, we denote by the physical variables X , , e
and P (in the first order formalism), by kn the ghosts of the -symmetry and by CA the
ghost for the remaining local and rigid symmetries. We refer to the index n as level; the
zero level is associated to the fields A .
According to [11,15,18,19,31], the general solution of the master equation in the case of
the CBS superparticle and of the GS superstring has the form

S = I A + A AAB ()CB + B A ()k1

+ CA D AB (, C)CB + F A ()k1 k1

n,
+ kn, G ()kn+1 + F (C)kn + k

AB

AB

+ A B H AB
()k1 k1 + J ()k2 + CA B M ()k1 k2
B

B
+ kn, B N
(3.1)
()kn+2 + P
(C)k1 kn+1
+ k
n, ,
where the terms quadratic in the anti-fields are necessary to close the algebra and to take
are of higher
n , k
into account the fact that the symmetry is reducible only on-shell, and k
n
order in the ghosts. In the case of the CBS superparticle there are further simplifications:

, and k
are absent since the ghosts are diffeomorphic
(i) the terms F (C)k , k
n,
n,
invariant and they have no quartic interactions;

(ii) the terms with powers of the anti-fields are restricted to one type, that is when one
of the anti-fields happens to be the anti-field of the einbein e.
(1) We begin the quantization procedure by introducing the twisting of the ghost
fields. The aim of this redefinition is to decouple the ghost fields kn from the
physical degrees of freedom described by X , P and . For that purpose, we
will follow the ideas of P. Townsend [25] introducing redefined ghost fields kn
and showing that their dynamics is independent of the physical variables. In
particular, the idea is to redefine the fields in such a way that kn, G kn+1 becomes
independent of the field . Consequently, provided that the gauge fixing procedure
does not introduce a dependence on A , the ghost fields are decoupled from A .
As it will be explained in the next section, in order to implement the suggested
redefinition of ghost fields [25], we adopt a twistor-like (off-shell) formalism.
Obviously, this change of variables can be performed by a canonical transformation.
Here, we have to point out that, as recognized by [15,20], some canonical
transformations involving infinite number of fields are ill-defined and, therefore,
they might change drastically the BRST cohomology. It will be our main concern to
avoid any such redefinition.
(2) Furthermore, although the twisting provides a convenient coordinatization for the
ghost degrees of freedom, there are remaining couplings between ghost fields and
387
A . After the twisting, we have

I
kn, G ()kn+1 kn, G + G,A A kn+1 .

(3.2)
The second term in the bracket can be eliminated by a further canonical transformation of type (2.6)

kn, G,A kn+1 .
kn, kn + A
= A A +
(3.3)
n
n

I
In the present case (and in the case of GS superstring), G,A
A is proportional
to the equations of motion for the world-line (-sheet) metric e and, therefore, the
canonical transformation involves the anti-field e . The latter is an anticommuting
field and, thus, (e )2 = 0. Consequently, it is easy to show that the redefinition of
the fields effected by (and the corresponding inverse transformation) involves
only a finite number of fields kn . Obviously, the master equation and the classical
cohomology of the theory remain unchanged by these transformations and we can
thus proceed with the quantization procedure.
(3) Following the BV formalism [22], one introduces a non-minimal sector of fields,
that is the anti-ghosts, the extra-ghosts and the Lagrange multipliers

, = p,q , cA
= pq , cA ,
C
,
C
q A
= p , ,
(3.4)
q + p = n, n 1,
and the non-minimal part to the action

Sn.m. = cA A +
,q=p
,q
p p .
(3.5)
p=0,q=1
However, since the application of the common BV formalism for an infinitely

reducible theory seems to generate wrong results [15,20], we follow a different
procedure.
(4) The main problem of a complete gauge fixing lies in the fact that after having fixed
the minimal and extra ghosts, the theory is still invariant under a local symmetry of
Stckelberg type [29]. The latter needs a further gauge fixing. As will be shown in
Section 6, some of the extra fields can be identified with the ghosts of Stckelberg
symmetry and, therefore, they can be used to fix completely the gauge invariances.
As a consequence, the non-minimal sector of ghosts and extra ghosts is partially
redundant and it can be reduced to a smaller set. Moreover, the gauge fixing of
the entire ghost sector is performed in a step-by-step procedure by using canonical
transformations which leave the cohomology unchanged.
4. The minimal solution of the master equation

The general strategy is here applied to the specific model of the CBS superparticle.
The analysis of the classical action, the constraints coming from the general coordinates
388
invariance and the supersymmetry on the target space imply that the action is invariant also
under the -symmetry. We associate to the world-line diffeomorphism the local ghost c,
and to the fermionic local -symmetry the infinite tower of ghosts kn . This fixes the field
content to (without considering the global ghosts for the super-Poincar invariance)
A

= e, X , , gh A = 0,
A
C = {c},
gh C A = +1,
A
K = {kn },
(4.1)
gh kn = n,
and the corresponding antifields

, ,
A = e , X ,
, gh A = 1,

CA = c ,
gh CA = 2,

KA = kn ,
gh kn = n 1.
(4.2)
The classical action for CBS superparticle [3] is given, in first order formalism, by
1
/ eP 2 ,
S = P X P
(4.3)
2
it is invariant under world-line reparametrizations, target-space Poincar transformations
and the kappa-symmetry transformations
P
/ ,
X =
= P
/ ,
e = 4 ,
(4.4)
where , the parameter of the transformation, is an anticommuting MW spinor, and is

replaced by a commuting spinor ghost k in the BRST version of the symmetry. The symmetry is closed on-shell (see, e.g., [2]) and one can see easily that it is reducible.
The transformation has 16 parameters, the number of components of a MW spinor in 10
/ the symmetry is
dimensions, but on-shell they are not all independent, since for = P
2
trivial due to the equation of motion P = 0. Thus only half of the components of can
be used to gauge away degrees of freedom of , and on-shell one achieves the matching
between bosonic and fermionic degrees of freedom that is required by supersymmetry. The
other components are zero-modes of the symmetry. There is no way to get rid of these if one
wants to keep the manifest Lorentz-invariance of the model; instead, one has to consider
this additional redundancy as arising from a gauge symmetry in the ghost sector, namely,
/ k2 ,
k = P
(4.5)
where an additional ghost has been introduced, which is a MW spinor with chirality
opposite to that of k. The relation between k and k2 is the same as the one between
and k, so the new symmetry is again reducible, a third generation ghost is required, and so
on: the symmetry is infinitely reducible. According to the BV procedure [22], the minimal
solution of the master equation takes the form of an action depending on an infinite number
of fields. The matching of degrees of freedom is obtained in this way: 8 components of
are physical, the remaining 8 are canceled by half of the components of the first ghost,
389
the remaining components of k are canceled by half of those of k2 and so on. This can be
written as
8 = 8 + (8 8) (8 8) + ,
and clearly the sum makes sense only if it is performed with the parenthesis in a
certain position, or if it is regularized in a proper way; it is customary to use the Euler
regularization of an alternate sum:

n=0
(1)n = lim
x1

n=0
1
(x)n = .
2
(4.6)
This gives a hint that the theory must be handled very carefully to avoid problems arising
from formal manipulations of non-convergent series. This kind of problems invalidate
solutions given so far to the problem of covariantly quantizing the GS string and the CBS
superparticle, as was first pointed out by Bastianelli et al. [20].
The minimal solution of the master equation is [18]

1
1
/ eP 2 + X cP + P/ k1 + e c + 4 k
L = P X P
2

2c k1P
/ k1 + P/ k1 +
/ kp+1
kp P
p1

+ 2e X k2 k1 k1 4c k1 k2 + k2 +
kp kp+2 .
(4.7)
p1
This action satisfies the requirements requested to a minimal solution. As expected, it is

quadratic in the antifields since the gauge algebra is open, but there are no terms with
higher powers of antifields.
5. Twisting the ghosts

The ghosts zero mode condition, as we have seen, is field-dependent and is only
satisfied on-shell. These are the two unpleasant features that we would like to eliminate, in
order to get a well-defined quantization procedure. Our strategy is to look for a redefinition
of the ghosts such that the new ghosts have off-shell zero modes, after possibly some
further manipulations that turn out to be necessary to eliminate terms proportional to the
equations of motion. All the transformations are performed via canonical transformations,
so that the master equation is satisfied at every step. The first observation to be made is
that a twistor-like formulation [28] is possible for the massless particle or superparticle,
in which bosonic degrees of freedom are traded for world-line spinors, and for
example, in Galperin et al. the kappa-symmetry is replaced by an extended world-line
supersymmetry. We adopt here a formulation that differs from the previous ones in that the
twistor-like variables are not regarded as fundamental, but just as (non-linear) functions of
the fundamental fields.
390
The idea of the twisting is to replace, in the -symmetry transformation laws, the
operator P
/ with another operator whose nilpotency is independent of the equations of
motion. The simplest choice is to take a 2 2 matrix, and the only nilpotent matrices
are the Pauli matrices and + , up to a multiplicative factor. The fundamental equations
that define the twistors are the following:

(P
/ ) = ,i + P 2 + i j j, ,
j
j, ,i = i ,
(5.1)
where we have defined = C21 C10 . As a consequence of the second equation of (5.1),
the two- and ten-dimensional charge-conjugation matrices are related:
C2 = C10 .
(5.2)
On shell, P
/ is simply replaced by ; the choice of + is of course possible, and is related
to the other choice by a discrete symmetry, namely, C2 . From the ten-dimensional
point of view, the twistors are commuting spinors (it is worthwhile mentioning that
/ is real, the twistor can be complex), but they carry also internal indices =
although P
1, 2; i = 1, . . . , 8, in order to be invertible matrices, as required by the second equation in
(5.1). We will never need to display explicitly the whole index structure.
Due to the anticommuting properties of the matrices, one can verify that if Eq. (5.1)
is read as the definition of a matrix P
/ , that matrix satisfies P/ 2 = P 2 . But we need to show
that the equation can be solved, with respect to , for any P . It is clear from the outset that
the solution can never be unique: for instance, can be multiplied by an orthogonal matrix
j
Oi in the internal indices. It can be proved that these are the only internal transformations
that leave Eq. (5.1) invariant. By a rotation, we can bring P to have only the P + and P
components in light-cone coordinates. Then, using the well-known iterative construction
of matrices starting from the two-dimensional ones, we have

0 1
0 0
+
=
1,
=
1.
0 0
1 0
In this frame, + and can be identified with + and .
On-shell, P
/ is a nilpotent matrix. We can choose a reference frame in which P + is the only
non-vanishing component of P ; in this frame, P
/ is proportional to + , whose kernel has
a dimension equal to one half of the dimension of the whole space. The Jordan canonical
/ is a matrix with diagonal 2 2 blocks of the form of , and eventually a block
form of P
of zeroes, but the argument on the dimension of the kernel forbids the latter. Then P
/ can
be transformed with a change of basis into 1, and this amounts to say that a solution
to the twisting equation does exist.
Off-shell, P
/ is diagonalizable with eigenvalues P 2 . Since P/ is a linear combination of
/ = 0. Then its Jordan form is
matrices, Tr P
1 0
(5.3)
P2
,
0 1
which is the same Jordan form of + P 2 + , so we conclude again that Eq. (5.1) can be
solved.
391
We have to stress that, even though the twistor is a non-linear function of the momentum
P , the Lorentz symmetry is linearly realized in the target space. The presence of an internal
structure carried by the twisted ghosts does not affect the 10-dimensional covariance. In
addition, all degrees of freedom are quantized in a covariant manner, and the physical
spectrum is identified with the antibracket cohomology. In appearance, our formalism
could remind the reader of a light-cone formulation of the superparticle, however the
choice of has been adopted only for the internal indices of the twistors ,i and
it is harmless for the covariance in the target space. 1
After the twisting, the minimal solution has the form

1 2
P
Lmin = P X P
/ P e 2 k p + kp+1 + X cP +
/ k1
2

/ k1 + P/ k1 + k p kp+1
+ e c + 4 k1 2c k 1 P

+ 2e X k2 k 1 k1 4c k 1 k2 + k2 +
k p kp+2 .
(5.4)

The twisted ghosts have off-shell zero modes, given by k = k (from now on the
tilde shall be understood, and all the ghosts are twisted). These zero-modes coincide onshell with the original ones. We can now eliminate completely the coupling of the non-zero
+ k, simply by a redefinition of the einbein
mode part of the ghosts, given by the terms P 2 k
field. This is performed by means of a canonical transformation with generating function

kp + kp+1 .
F = 2e
(5.5)
p1
The trivial part of the generating function, corresponding to the identity operator, has
been understood, as will be henceforth. The corresponding shifts of the fields are listed
below:

kp + kp+1 ,
e = e 2
kp
p1
= kp 2e ()p + kp+1 ,

kp+1
= kp+1
2e kp + .
(5.6)
These redefinitions as well as their inverses do not involve an infinite number of fields due
to the presence of the nilpotent factor e , so they are allowed transformation in the sense
discussed in [15], that is, the redefined field are not subject to any constraint. As a partial
check of this statement, we shall now verify that the BRST cohomology is not affected by
the transformation. The action is

1
Lmin = P X eP 2 P
/ + X cP + P/ k1 + e (c + 4 k1 )
2

/ k1 + P/ k1 + kp kp+1
2c k1 P

k1 8e c k1 + k2 + 2e + k2 .
+ 2e X k2 k1
(5.7)
1 We thank I. Bandos for pointing out that our formulation is reminiscent of spinor Lorentz harmonics [14].
392
One can notice that, as was to be expected, the terms e kn have disappeared.
The physical sector of the theory is given by the zero-ghost number characteristic
cohomology, that is, the cohomology of the BRST operator modulo the equations of
motion. More precisely, two classical physical observables are identified if they coincide
on the equations of motion; the identification is enforced by considering the cohomology
of the KoszulTate differential at antifield number 0, i.e., H0 (). The BRST operator
defined by (2) anticommutes with so that it defines a nilpotent operator in the cohomology
of , and one can consider the group H ( , H0()). This group is actually isomorphic to
H (s) [17].
The relevant BRST transformations are
/ k1 ,
X = cP + P
1.
/ k1 ,
e = c + 4 k
=P
P = 0,
(5.8)
Manifestly P is in the cohomology, and on-shell is a constant light-like vector. It is easy

to verify that the vector

1
1
U = (X P )X X2 P X[ P]
2
2
is BRST invariant, is light-like on-shell, and vanishes for X proportional to P , so it
accounts for 8 bosonic degrees of freedom. The fermionic variables in the cohomology
/.
are P
6. Gauge-fixing
We have now written the minimal extended action, i.e., a minimal proper solution of
the master equation, in a form that is suitable for further manipulations. The next task
to be performed is the gauge-fixing of the -symmetry. This requires the introduction
of a set of non-minimal fields and non-minimal terms in the extended action. We shall
follow a completely standard procedure: the non-minimal fields include extra-ghosts
and Lagrange multipliers , with opposite statistics, and the non-minimal terms have
generically the form . We shall work by steps, at each step fixing the gauge of a given
level of fields. At level 0 (the classical fields), we have
L0 = Lmin + Lnm,0 = Lmin + 11 11 ,
(6.1)
the gauge-fixing of is achieved by the gauge fermion

0 = 11 ( ),
(6.2)
the canonical transformation generates new terms in the action:

L0 L0 + 11 11 P
/ k1 2e 11 + k2 ,
(6.3)
/ in the
and the term proportional to the equations of motion which appears after rewriting P
twisted form can be eliminated by the canonical transformation
= 2e 11 + k1 ,
(6.4)
393
that gives
1
1
L0 = L0 +
1 1 k1 + 2e k1 + 1 .
(6.5)
The propagator for the physical field is now invertible, as is required for a gauge-fixed
action.
At level 1,
1 = 21 k1 ,
Lnm,1 = 21 21 ,

L1 = L0 + 21 k1 21 21 k2 ,
(6.6)
we have fixed the gauge of k1 . A crucial remark is now in order [15]: the ghost 11 not only
has the usual and expected -symmetry, but also a more general one, of Stckelberg type,
namely,
11 = ;,
21 = ;.
The gauge-fixed action (the part independent of the antifields) is invariant under this
symmetry. We take into account this new symmetry introducing a new ghost, 1 , and new
terms in the action. These have the form of minimal terms for the new symmetry, but it
turns out that other terms are necessary to close the master equation:
1

21 + 33 1 21 33 .
L(1)
(6.7)
St = 1 +
It can be verified that L1 + LSt satisfies the master equation, and the cohomology is not
altered. This is hardly surprising, when one realizes that L(1)
St is just a non-minimal term
written in transformed variables. Indeed, the non minimal term is
(1)
33 1
and L(1)
St is recovered via the transformation

= 11 + 21 33 .
With a canonical transformation we now redefine the new ghost 1 1 11 ; as a
result, the non-minimal term 11 11 , previously introduced, is now canceled by the shift
of the first term in L(1)
St . We can now proceed to fix the Stckelberg symmetry; although it
allows for an algebraic fixing, we choose to use a Lorentz gauge-fixing, in order to have
more uniformity in the treatment of the various extra-ghosts. We then add further nonminimal terms
1 = 22 11 + 31 k2 ,
Lnm,1 = 22 22 + 31 31 ,

2

1
11 22 22 1
L2 = L1 + L(1)
St 1 1 1 + 2

+ 31 k2 31 31 k3 .
(6.8)
(6.9)
It is now apparent that the role of the ghost 1 is that of a Lagrange multiplier, which
fixes the gauge of 22 ; thus there is no need to introduce other non-minimal terms and
gauge-fixings for 22 that would bring in the action a mixing of 11 with higher-level ghosts.
394
The first step is completed; at this point, the ghost k2 is fixed, while 21 has the Stckelberg
symmetry. As before, we introduce non-minimal terms 43 2 followed by

= 21 31 43 ,
that yields

(2)
LSt = 21 31 + 43 2 + 31 43 ,
(6.10)
then we perform a canonical transformation

2 2 21 + 33 ,
21 21 + 2 ,
33 33 2 .
(6.11)
It is actually sufficient to substitute in the action the shift of 2 , since the shifts of the
antifields do not generate any new term, as they cancel each other. But this canonical
transformation leads also to the cancellation of the terms 21 21 and 21 33 . The gaugefixing is done in perfect analogy with the previous step:
2 = 22 11 + 31 k2 ,
Lnm,2 = 32 32 + 41 41 ,

2
2
1
3
1
2
L3 = L2 + L(2)
St 2 2 2 + 3 + 3 2 3 3 2
1
1
+ 4 k3 4 41 k4 .
(6.12)
(6.13)
Now the ghosts k3 , 21 and 32 are fixed, while 31 has a Stckelberg symmetry.
We are now able to give the algorithmic procedure for obtaining the gauge-fixed action
at any level: given the action at level n, when the last ghost fixed is kn and n1 has the
Stckelberg symmetry, one has to perform the following operations:
(1) introduce the non-minimal terms corresponding to the Stckelberg symmetry

(n)
1
3
1
3
n + ()n n+1
+ n+2
n+2
;
LSt = n1 + ()n+1 n+1
(6.14)
(2) redefine the Stckelberg ghost:
3
n n n1 + ()n n+1
,
(6.15)
this has to be done via a canonical transformation, so there are also other
redefinitions which, however, do not have any effect on the action;
(3) add non-minimal terms
1
1
2
2
n+1
+ n+1
n+1
;
Lnm,n = n+1
(6.16)
(4) fix the gauge with the canonical transformation

1
2
n = n+1
kn+1 + n+1
n1 .
(6.17)
The procedure just consists of canonical operations which can not change the cohomology.
The situation at level 3 is depicted in Table 1. We find in this way that only a subset
of the whole pyramid of extra-ghosts as is given, e.g., in [11] is really necessary for the
quantization procedure: an entire sub-pyramid with vertex in 33 can be omitted. It is easy
to count the degrees of freedom and verify that the counting is correct at each step.
395
Table 1
Fields after the 3rd step
(0)
11
(1)
22
(0)
k1
(1)
21
(2)
k2
(2)
32
(1)
31
(3)
k3
(3)
41
(4)
k4
(4)
It is useful to look at the Lagrangian at a generic step n. Its form is given by

Ln = Lmin + 2e k1 + 11 + 11
n+1
p1 kp
p=1
n1

p=1
n
2
p1 p+1
n
p2 p1 +
p=2
n1
p1 p +
p=1

3
1
1
+ n1 n1 n+1
n+1
.
+ n+1
1
kp p+1
p=1

1
3
+
p3 p2 p2
+ ()p p1
p=3
n
n
p2 p2
p=2
n

1
()p p1 p1 p1
p=2
(6.18)
The complete action, fixed at every level of ghosts, is now obtained simply as the formal
limit n : all the sums in (6.18) run up to infinity, and the two terms in the last line
must be dropped. The equations of motion fix all fields to constant values.
Finally, the invariance under diffeomorphisms is algebraically fixed by introducing the
non-minimal fields (a Lagrangian multiplier d and the antighost b), a non-minimal term in
the action and performing the canonical transformation generated by diff :
Ldiff = b d,
diff = b(e 1).
(6.19)
They form a trivial pair under BRST, namely, sb = d and sd = 0, so they do not belong to
the cohomology.
396
7. The BRST cohomology

We will now compute the BRST cohomology for the model in the limit n . At each
step of our procedure, it is guaranteed that the cohomology is correct. The limit is the only
operation which is not canonical; nevertheless, we will show now that the cohomology
is still the correct one. The relevant parts of the BRST transformations for the fermionic
sector, in the limit n , are listed below:
s = P
/ + 11 ,

1
+ p+1
skp = ()p p1 kp1
,
/ k1 + 2b+ k2 ,
s = P
skp = kp+1 ,
sp1 = p ,
2
sp1 = kp + p+1
,
sp2 = p2 ,
sp2 = p1 ,
1
3
3
+ ()p p1
,
sp3 = p+1
,
sp3 = p2 p2

1
3
1
sp1 = ()p p1 p1
,
,
sp1 = ()p kp1 + p+2
+ p+1

2
2
p
1
2
sp = () p1 p ,
sp = 0,
2

3
1
sp = 0,
p+1
sp = ()p p+1
p1 p+2
,
where we have omitted the terms coming from the part of the action quadratic in the
antifield in Eq. (5.7), as well as the terms dependent from the antifields of the bosonic
sector and of the diffeomorphism sector. These terms do not change the results of the
analysis. In the above list of transformations, one can recognize at once some trivial pairs,
e.g., (p1 , p ), (p2 , p2 ); then one can define a reduced cohomology, setting to zero the
trivial pairs in the above transformations. Splitting the spinor
in two components,
+fields
1
1+
+
are
= ( , ), so that = ( , 0), one can see that also kp , kp+1 , p , p+1
3 3+
1+ 3+
trivial pairs, as well as p , p+1 , and finally 1 , 3 . We conclude that the only
/ . Moreover, a careful analysis of the antifield sector
field in the physical spectrum is P
/ is actually BRST-exact, so that only its zero mode
along the same lines shows that P
survives in the cohomology. In this way, we recover the correct superparticle spectrum.
In the limit, we obtain an action that is free, i.e., quadratic in all fields, except for nonpolynomial interactions involving the twistor-like fields and cubic interactions involving
only the physical fields, the diffeomorphism ghost, and the first -symmetry ghost k1 , as
well as the corresponding Lagrange multipliers. All the other ghosts are decoupled, so they
do not enter in the computation of any amplitude involving only external physical states.
8. Conclusions
Using the conventional BVBRST methods, we provide the complete gauge-fixed action
for the CasalbuoniBrinkSchwarz superparticle. Despite several difficulties, by means
of a suitable redefinition of the minimal sector fields in terms of twisted variables, we
are able to construct a quantization procedure of the model. A direct computation of the
antibracket cohomology shows that the BRST charge selects the correct physical spectrum
397
of the theory. In addition, it is also verified that the characteristic cohomology H ( , H ())
confirms the result of the antibracket analysis.
Acknowledgement
We thank Peter van Nieuwenhuizen and Warren Siegel for illuminating comments and
suggestions. The research has been supported by the NSF grants no. PHY-9722083 and
PHY-0070787.
Appendix A. Notations and conventions

Our conventions are the following. Xm , m = 0, 1, . . . , 9, denotes the 10d spacetime
coordinates and m are 10-d Dirac matrices

m , n = 2mn ,
where = ( + + +).
(A.1)
These s differ by a factor of i from those of Ref. [2]. For this choice of gamma matrices
the massive Dirac equation is ( M) = 0. In fact, our conventions are such that
the quantity i = 1 will not appear in any equations. As it is quite standard, we also
2
introduce 11 = 0 1 9 , which satisfies {11, m } = 0 and 11 = 1 and the totally
antisymmetrized Dirac matrices mnk = [m n k] .
The Grassman coordinates are spacetime spinors and world-line scalars. They can
be decomposed as = R + L , where
1
R = (1 + 11 ),
2
1
L = (1 11 ).
2
(A.2)
For anticommuting spinors, the rule to lower and to rise the spinor indices implies
= C ,
= C
,

= = = = ,
(A.3)
where C is the antisymmetric charge conjugation matrix. With these conventions, the
fermionic degrees of freedom of the CBS superparticle are carried by a MajoranaWeyl
(MW) spinor R . In the same way all the -symmetry ghosts are MW spinors.
In general, the symmetry properties of bilinears built out of MW spinors are the
following [32]:
1 n = (1)1+;();()+n(n+1)/2
1 n .
(A.4)
There is only one independent non-vanishing bilinear of a single MW-spinor, i.e.,

.
398
References
[1] N. Berkovits, Super-Poincar covariant quantization of the superstring, JHEP 0004 (2000) 018,
hep-th/0001035;
N. Berkovits, B.C. Vallilo, Consistency of super-Poincar covariant superstring tree amplitudes,
JHEP 0007 (2000) 015, hep-th/0004171;
N. Berkovits, Cohomology in the pure spinor formalism for the superstring, JHEP 0009 (2000)
046, hep-th/0006003.
[2] M.B. Green, J.H. Schwarz, E. Witten, Superstring Theory, Vol. 1, Cambridge Univ. Press, 1987,
Chapter 4;
M.B. Green, J.H. Schwarz, Covariant description of superstrings, Phys. Lett. B 136 (1984) 367;
M.B. Green, J.H. Schwarz, Properties of the covariant formulation of superstring theories, Nucl.
Phys. B 243 (1984) 285.
[3] R. Casalbuoni, Phys. Lett. B 293 (1976) 49;
R. Casalbuoni, Nuovo Cimento A 33 (1976) 389;
L. Brink, J.H. Schwarz, Phys. Lett. B 100 (1981) 310.
[4] P.A.M. Dirac, Lectures on quantum mechanics, Yeshiva University, New York, 1964.
[5] M. Henneaux, C. Teitelboim, Quantization of Gauge Systems, Princeton Univ. Press, Princeton,
1992.
[6] D.M. Gitman, I.V. Tyutin, Quantization of Fields with Constraints, Springer-Verlag, Springer
Series in Nuclear and Particle Physics.
[7] L. Brink, M. Henneaux, C. Teitelboim, Covariant Hamiltonian formulation of the superparticle,
Nucl. Phys. B 293 (1987) 505.
[8] W. Siegel, Hidden local supersymmetry in the supersymmetric particle action, Phys. Lett. B 128
(1983) 397.
[9] S. Bellucci, A. Galajinsky, Phys. Rev. D 62 (2000) 027501, hep-th/9909190;
A.A. Deriglazov, A.V. Galajinsky, S.L. Lyakhovich, Nucl. Phys. B 473 (1996) 245, hepth/9512036.
[10] O.F. Dayi, Phys. Rev. D 44 (1991) 1239;
O.F. Dayi, D 45 (1991) 1239, Erratum;
O.F. Dayi, Int. J. Mod. Phys. A 7 (1992) 2531.
[11] M.B. Green, C.M. Hull, QMC/PH/89-7, Presented at Texas A and M Mtg. on String Theory,
College Station, TX, March 1318, 1989;
M.B. Green, C.M. Hull, Mod. Phys. Lett. A 5 (1990) 1399;
M.B. Green, C.M. Hull, Nucl. Phys. B 344 (1990) 115;
M.B. Green, C.M. Hull, The Brst cohomology of an N = 1 superparticle, in: Strings 90,
Proceedings, College Station, TX, 1990, pp. 133147;
C.M. Hull, J. Vazquez-Bello, Nucl. Phys. B 416 (1994) 173, hep-th/9308022;
J. Feinberg, M. Moshe, Ann. Phys. B 206 (1991) 272;
J. Feinberg, M. Moshe, Phys. Lett. B 247 (1990) 509.
[12] F. Essler, E. Laenen, W. Siegel, J.P. Yamron, Phys. Lett. B 254 (1991) 411;
F. Essler, M. Hatsuda, E. Laenen, W. Siegel, J.P. Yamron, T. Kimura, A. Mikovic, Nucl. Phys.
B 364 (1991) 67;
J.L. Vazquez-Bello, Int. J. Mod. Phys. A 7 (1992) 4583;
A. Mikovic, M. Rocek, W. Siegel, P. van Nieuwenhuizen, J. Yamron, A.E. van de Ven, Phys.
Lett. B 235 (1990) 106.
[13] E. Nissimov, S. Pacheva, S. Solomon, Nucl. Phys. B 296 (1988) 462;
E. Nissimov, S. Pacheva, S. Solomon, Nucl. Phys. B 297 (1988) 349;
R. Kallosh, M. Rakhmanov, Phys. Lett. B 209 (1988) 233.
[14] A. Galperin, P. Howe, K. Stelle, Nucl. Phys. 368 (1992) 248;
A. Galperin, F. Delduc, E. Sokatchev, Nucl. Phys. 368 (1992) 143;
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
399
I.A. Bandos, A.A. Zheltukhin, Phys. Lett. B 288 (1992) 77;

I.A. Bandos, A.A. Zheltukhin, Phys. Part. Nucl. 25 (1994) 453477;
I. Bandos, T. Bandos, hep-th/0010044.
E. Bergshoeff, R. Kallosh, A. Van Proeyen, Superparticle actions and gauge fixings, Class.
Quantum Grav. 9 (1992) 321;
R. Kallosh, Nucl. Phys. Proc. Suppl. B 18 (1990) 180.
J.M.L. Fisch, M. Henneaux, Homological perturbation theory and the algebraic structure of the
antifield-antibracket formalism for gauge theories, Commun. Math. Phys. 128 (1990) 627;
G. Barnich, F. Brandt, M. Henneaux, Local BRST cohomology in the antifield formalism:
I. general theorems, Commun. Math. Phys. 174 (1995) 93.
G. Barnich, M. Henneaux, T. Hurth, K. Skenderis, Cohomological analysis of gauge-fixed
gauge theories, hep-th/9910201.
U. Lindstrm, M. Rocek, W. Siegel, P. van Nieuwenhuizen, A.E. van de Ven, Phys. Lett. B 224
(1989) 285;
U. Lindstrm, M. Rocek, W. Siegel, P. van Nieuwenhuizen, A.E. van de Ven, Phys. Lett. B 228
(1989) 53;
M. Rocek, W. Siegel, P. van Nieuwenhuizen, A.E. van de Ven, Phys. Lett. B 227 (1989) 87.
S.J. Gates, M.T. Grisaru, U. Lindstrm, M. Rocek, W. Siegel, P. van Nieuwenhuizen, A.E. van
de Ven, Lorentz covariant quantization of the heterotic superstring, Phys.Lett. B 225 (1989) 44;
U. Lindstrm, M. Rocek, W. Siegel, P. van Nieuwenhuizen, A.E. van de Ven, Construction of
the covariantly quantized heterotic superstring, Nucl. Phys. B 330 (1990) 19.
F. Bastianelli, G.W. Delius, E. Laenen, Phys. Lett. B 229 (1989) 223.
C. Becchi, A. Rouet, R. Stora, Phys. Lett. B 52 (1974) 344;
C. Becchi, A. Rouet, R. Stora, Commun. Math. Phys. 42 (1976) 127;
C. Becchi, A. Rouet, R. Stora, Ann. Phys. 98 (1976) 287;
I.V. Tyutin, Lebedev preprint FIAN No. 39 (1975).
I.A. Batalin, G.A. Vilkovisky, Quantization of gauge theories with linearly dependent
generators, Phys. Rev. D 28 (1983) 2567;
I.A. Batalin, G.A. Vilkovisky, Closure of the gauge algebra, generalized Lie equations and
Feynman rules, Nucl. Phys. B 234 (1984) 106.
A. Van Proeyen, BatalinVilkovisky Lagrangian quantization, hep-th/9109036;
W. Troost, P. van Nieuwenhuizen, A. Van Proeyen, Anomalies and the BatalinVilkovisky
Lagrangian formalism, Nucl. Phys. B 333 (1990) 727.
J. Gomis, J. Paris, S. Samuel, Antibracket, antifields and gauge theory quantization, Phys.
Rep. 259 (1995) 1, hep-th/9412228.
P.K. Townsend, Three lectures on quantum supersymmetry and supergravity: 1. Counterterms.
2. Nonrenormalization theorems. 3. Chiral anomalies, Print-84-0712 (Cambridge), in: Supersymmetry and Supergravity 84, Proceedings, Trieste, 1974, pp. 400425.
P.A. Grassi, G. Policastro, M. Porrati, Notes on the quantization of the complex linear
superfield, Nucl. Phys. B 597 (2001) 615, hep-th/0010052.
M. Grisaru, A. Van Proeyen, D. Zanon, Nucl. Phys. B 502 (1997) 345;
S. Penati, A. Refolli, A. Sevrin, D. Zanon, Nucl. Phys. B 533 (1998) 593;
S. Penati, A. Refolli, A. Van Proeyen, D. Zanon, Nucl. Phys. B 514 (1998) 460.
D.P. Sorokin, V.I. Tkach, D.V. Volkov, Superparticles, twistors and Siegel symmetry, Mod.
Phys. Lett. A 4 (1989) 901908;
D.V. Volkov, A.A. Zheltukhin, On the equivalence of the Lagrangians of massless Dirac and
supersymmetrical particles, Lett. Math. Phys. 17 (1989) 141147;
D.V. Volkov, A.A. Zheltukhin, Lagrangians for massless particles and strings with local and
global supersymmetry, Phys. Lett. B 335 (1990) 723739;
400
[29]
[30]
[31]
[32]
D. Sorokin, V. Tkach, D.V. Volkov, A.A. Zheltukhin, E. Sokatchev, From the superparticle
Siegel symmetry to the spinning particle proper time supersymmetry, Phys. Lett. B 216 (1989)
302306;
A. Galperin, E. Sokatchev, A Twistor like D = 10 superparticle action with manifest N = 8
worldline supersymmetry, Phys. Rev. D 46 (1992) 714, hep-th/9203051;
M. Tonin, Twistor-like formulation of heterotic strings, in: General relativity and gravitational
physics, Bardonecchia, 1992, pp. 435452, hep-th/9301055;
F. Delduc, E. Ivanov, E. Sokatchev, Twistor like superstrings with D = 3, D = 4, D = 6 target
superspace and N = (1, 0), N = (2, 0), N = (4, 0) world sheet supersymmetry, Nucl. Phys.
B 384 (1992) 334, hep-th/9204071;
E. Sezgin, Spacetime and world volume supersymmetric super p-brane actions, hepth/9411055.
A. Dresse, J.M. Fisch, P. Gregoire, M. Henneaux, Nucl. Phys. B 354 (1991) 191;
J.M. Fisch, M. Henneaux, J. Stasheff, C. Teitelboim, Commun. Math. Phys. 120 (1989) 379;
J.M. Fisch, M. Henneaux, A note on the covariant BRST quantization of the superparticle,
ULB-TH2-89-04-REV.
I.A. Batalin, R.E. Kallosh, A. Van Proeyen, KUL-TF-87-17 Presented at Seminar on Quantum
Gravity, Moscow, USSR, May 2529, 1987;
R. Kallosh, W. Troost, A. Van Proeyen, Phys. Lett. B 212 (1988) 428;
E.A. Bergshoeff, R. Kallosh, A. Van Proeyen, Phys. Lett. B 251 (1990) 128.
M. Tonin, Superstrings, K symmetry and superspace constraints, Int. J. Mod. Phys. A 3 (1988)
1519;
M. Tonin, Consistency condition for kappa anomalies and superspace constraints in quantum
heterotic superstrings, Int. J. Mod. Phys. A 4 (1983) 1989;
M. Tonin, Int. J. Mod. Phys. A 6 (1991) 315.
E. Bergshoeff, M. de Roo, B. de Wit, P. van Nieuwenhuizen, Ten-dimensional Maxwell
Einstein supergravity, its currents, and the issue of its auxiliary fields, Nucl. Phys. B 195 (1982)
97, reprinted in: A. Salam, E. Sezgin (Eds.), Supergravities in Diverse Dimensions, Vol. 1,
Elsevier, 1989, p. 195.

Coherent states for canonical quantum general

relativity and the infinite tensor product extension
H. Sahlmann, T. Thiemann, O. Winkler
MPI f. Gravitationsphysik, Albert-Einstein-Institut, Am Mhlenberg 1, 14476 Golm, Germany
Abstract
We summarize a recently proposed concrete programme for investigating the (semi)classical limit
of canonical, Lorentzian, continuum quantum general relativity in four spacetime dimensions. The
analysis is based on a novel set of coherent states labelled by graphs. These fit neatly together with
an Infinite Tensor Product (ITP) extension of the currently used Hilbert space. The ITP construction
enables us to give rigorous meaning to the infinite volume (thermodynamic) limit of the theory which
has been out of reach so far. 2001 Elsevier Science B.V. All rights reserved.
1. Introduction
Canonical quantum general relativity has matured over the last decade into a serious
candidate theory of quantum gravity which is manifestly background independent. The
most important developments include a rigorously defined mathematical framework, nonperturbative operator regularization methods, the prediction of a discrete or distributional
(rather than smooth) quantum geometry at Planck scale, a microscopic explanation for
the BekensteinHawking black hole entropy and a fully diffeomorphism covariant state
sum formulation. See [1] for pedagogical reviews, more or less covering these developments.
Recently, some effort [2] has also been devoted to the extraction of semi-classical
physics from the non-perturbative quantum theory. By this one means that if the distances
we probe are large compared to the Planck scale p = h ( is Newtons constant up to

a numerical factor in units where c = 1) then the fluctuations of geometry can be neglected
and we are close to a situation that one can treat with the methods of quantum field theory
on curved spacetimes, which in turn also is valid only as long as the backreaction of
E-mail addresses: sahlmann@aei-potsdam.mpg.de (H. Sahlmann), thiemann@aei-potsdam.mpg.de
(T. Thiemann), winkler@aei-potsdam.mpg.de (O. Winkler).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 2 2 6 - 7
402
H. Sahlmann et al. / Nuclear Physics B 606 (2001) 401440
geometry to quantum matter can be discarded. In particular, if the gravitational field is

in a semi-classical state approximating the Minkowski metric then we should arrive at the
quantum field theory describing the standard model plus corrections.
What is a semi-classical gravitational state? It is not entirely straight forward to
answer this question as the manifest background independence of the quantization scheme
forces us to adopt non-standard representations of the canonical commutation relations,
in particular, the powerful Fock Hilbert space techniques are not available which would
immediately allow us to construct the standard coherent states used in the literature. In
what follows, the contents of a programme based on a new set of gravitational coherent
states [3] is summarized which provides a proposal for a systematic construction of the
classical limit.
The organization of the article is as follows. In Section 2 we describe in detail and
in a non-technical way the main ideas of our proposal which is based on a new class of
kinematical coherent states, that is, minimal uncertainty states for both connection and
electric field operators labelled by a points in the classical, not necessarily constrained or
reduced, phase space. Here we use the gauge theory canonically conjugate coordinates for
this phase space described in [1]. As such, they are in sharp contrast to the weave states
already constructed in the literature which are are electric field operator eigenstates and
thus completely outspread in the connection operator. For every classical observable we
construct an approximate (effective) quantum operator which has the correct expectation
values and fluctuations with respect to our coherent states. The term effective refers to
the fact that these operators are, in general, different from those already constructed in the
literature. This poses no problem because to say that certain operator algebras correspond
to certain classical Poisson algebras is an empty statement without specifying with respect
to which semi-classical states the correspondence is made. Our coherent states are simply
not always adapted to the operators already constructed but we will point out how they can
be modified in that respect.
We will see that three different scales emerge in the semi-classical analysis, the
microscopic Planck scale, a mesoscopic scale defined by the edge length of an embedded
graph (as measured by the three metric defined by a point in the classical phase space) and
a macroscopic scale associated with the curvature of the four metric of that phase space
point. When we ask that fluctuations be minimized we learn that the mesoscopic scale is
a geometric mean of the other two scales. This result can be interpreted in the following
way: since the Planck length is proportional to a positive power of h and if we understand
the classical limit roughly as the limit h 0 while the continuum limit corresponds to
0, then we may say that the continuum and classical limit merge into only one limit!
More precisely, since h is a small but finite number, also is a small but finite number and
the actual continuum limit 0 is unphysical, however, in an expansion in powers of
the zeroth order term gives the correct classical continuum expression.
An important step needed is the combination of model coherent states labelled by
points in a model phase space (the cotangent bundle over a compact gauge group) with
the continuum phase space that we are actually interested in. We discuss advantages
and shortcomings of our proposal at the end of Section 2 and we supplement it by
403
Appendices A, B where we exhibit the derivation of an approximate version of the area

operator and the appearance of optimal scales in fluctuation calculations.
In Section 3 we sketch the construction of model coherent states in a hopefully
pedagogical way, bearing, as much as possible on an analogy with the harmonic oscillator
coherent states of which the present ones are a non-Abelian version. In particular, we try
to provide some intuition based on geometric quantization and heat kernel methods.
In Section 4 we introduce a technical tool, the Infinite Tensor Product construction which
is well known in statistical physics, in order to tackle the infinite volume limit of the theory
which has been out of reach so far. We outline just the most basic mathematical definitions
and discuss how these mathematical termini translate into physical concepts. This section
is supplemented by Appendix C which employs the mathematical notions used in a simple
system, the infinite spin chain.
In the final section we summarize and list future applications of the concepts introduced,
one of the most important being how to make contact with quantum field theory on curved
spacetimes.
2. A concrete programme for the construction of (semi)classical canonical quantum

general relativity
To say that a state is semi-classical makes sense only if one specifies a set of elementary
observables ge (e belongs to some label set) which are to behave almost classically, not
every possible operator on the Hilbert space can have a good classical limit in the sense
of approximating a given function on the classical phase space (see the third reference
in [2] for an illustration in the context of canonical quantum general relativity). The
specification of a such a subset of the operator algebra is sometimes referred to as coarsegraining. One starts with a classical phase space M which in general depends also on the
three-manifold underlying the canonical formulation, for example, through boundary
or fall-off conditions or through the differentiable structure (we refrain from displaying
this dependence of M on in what follows). A state m labelled by a point m in this
j
classical phase space (say a canonical pair consisting of a connection Aa and an electric
field Eja ) is said to be semi-classical for a set of elementary classical observables ge if the
expectation values of the corresponding self-adjoint operators assume their classical values
at the point m and if their relative fluctuations are small, that is, m , ge m = ge (m) and

(ge )quant(m) := m , [ge ]2 m /[ge (m)]2 1 1.
(2.1)
Clearly, as remarked by Ashtekar, if ge (m) = 0 then (2.1) blows up but this is just because
the value zero is smaller than the fundamental absolute fluctuation of the operator which is
bounded away from zero. Nobody would say that the harmonic oscillator coherent states
break down at the origin of the phase space. If we replace ge by the quantity ge + c (which
classically captures the same information as ge ) for some constant c much larger than
the absolute fluctuation then (2.1) is finite and small. Alternatively, we can work with
dimensionless quantities ge and consider only absolute fluctuations. In what follows we
404
assume that one has taken care of these subtleties in either way and that (2.1) is well
defined. Notice that the four-metric determined by the initial data set m introduces a first
length scale into the analysis, namely its typical curvature radius L(m) as measured by the
three-metric determined by m. As we will see in Appendix B there are actually two such
radii (an electric and magnetic one) but we will symbolize them by a single label L(m)
for the descriptional purposes of this section. If needed, we may want to restrict the phase
space by allowing only those points m for which L(m) L0 is bounded from below which
could induce a restriction on the (Gauss, diffeomorphism and Hamiltonian) gauge choice
involved in specifying m.
The states of the gravitational Hilbert space that we will consider are labelled by
piecewise analytic graphs with an at most countably infinite number of edges. These
graphs are embedded into the three-dimensional spatial manifold of given topology
which underlies the canonical formulation classical general relativity. For a given state it
is only at the locus of the corresponding embedded graph where the excitations of the
gravitational field are probed. This raises immediately the question: what are these degrees
of freedom, called graph degrees of freedom in what follows, that are probed? As far
as the configuration degrees of freedom are concerned, the answer is easy: they are the
holonomies of the gravitational connection along the edges of the graph. However, for
the momentum degrees of freedom this is less obvious. There is some freedom in choosing
them and we will specify our choice below. Taken together, the observables associated with
these degrees of freedom for given graph are good candidates for a set of elementary
observables ge with respect to which a state ,m , now also labelled by the graph, is semiclassical and e will run through the set of edges of the graph.
Now the following problem appears: the above comments suggest to define semiclassical states over one single arbitrary but fixed graph . However, an arbitrary classical
observable O generically cannot be written as a function of the elementary graph degrees of
freedom ge for the given graph and therefore it is not guaranteed that the corresponding
continuum quantum operator will display semi-classical behaviour in the state ,m . A
way out is suggested by the observation that any O can be approximated by a function
O = O ({ge }) of the graph degrees of freedom in the sense that

O (m)
1 1
(O)class(m) :=
(2.2)
O(m)
for all m M. This can be seen as follows: notice that given , m automatically a second
length scale is introduced, namely the typical length (measured by the three-metric
determined by m) of an edge of . Since L is the scale on which the four-metric varies, the
classical approximation will be good as long as L. If L L0 we just need to choose
small enough to obtain a small value of (2.2) even uniformly in m.
This observation motivates to use the corresponding operators O as substitutes for the
There are two worrisome issues about this proposal:
continuum operator O.
(I) The operators O will, in general, not coincide with the cylindrical projections
of the continuum operators already constructed in the literature, say the area
operators [4]. Thus, even if one manages to produce diffeomorphism covariant,
405
cylindrically consistently defined families of operators O corresponding to O one

will get a different continuum operator this way.
(II) The substitute function O necessarily must make explicit use of a coordinate
system. Thus, unless one is very careful in constructing it, it is conceivable that
one breaks diffeomorphism covariance and background independence. This would
be unsatisfactory because one of the strongest motivations and guidelines for
this whole approach to quantum gravity is to keep background independence at
every stage of the construction. To see how this might happen, consider the socalled staircase problem which we discovered in discussions with Ashtekar,
Lewandowski and Pullin and which we discuss more explicitly in Appendix A:
suppose, we would like to measure the area O = AS of a surface S in the state ,m
where is topologically a cubic lattice. If the surface is built from the elementary
surfaces S e that come with the graph degrees of freedom as specified below then
the cylindrical projection of the operator constructed in [4], which is manifestly
diffeomorphism covariant, actually coincides with O and its expectation values
agree with the classical values. However, if it lies transversally to them then its
expectation value will be drastically off the classical value AS (m). The replacement
AS, will have to take this error into account by using a coordinate system which
enables us to tell how the elementary surfaces are embedded relative to S. (This
coordinate system, however, does not depend on the point m.) Now, while ge
transforms covariantly under diffeomorphisms, there could be an extra coordinate
dependence which is guaranteed to disappear only in the limit 0 and O might
not transform covariantly any longer. Thus, the operator O could be not only state
dependent (it depends explicitly on but not on m) but could also be coordinate
dependent. The former dependence can be removed if one can show that the O
are cylindrical projections of one and the same operator O which can always be
achieved by introducing suitable projections, see below. However, it is possible that
the extra coordinate dependence cannot be removed, at least not obviously, at any
finite .
Of course, one could ignore this problem altogether and measure only operators that
actually can be written in terms of the ge . In fact, as pointed out by Lewandowski, if one
agrees that all physical information can be extracted, up to a certain accuracy set by a
length scale, from electric and magnetic fluxes through surfaces of the size of that scale
or bigger then, up to a negligible error, one can actually write these quantities purely
in terms of the ge due to the Gauss law and the Bianchi identity, see Appendix B, in
a manifestly coordinate independent way. While this is a working proposal, we want
to be more ambitious and measure any operator quantum mechanically and not by first
measuring elementary operators quantum mechanically and then assembling the values
of these measurements by a classical formula into a value for the observable that we are
interested in. Thus we have to deal with both problems (I), (II) in what follows.
We will show below that one can actually solve problem (II). Then one would still have
problem (I), that is, for every classical observable O one has two candidate quantum
the first one of the kind already constructed in the literature using
observables O , O,
406
background independent regularization techniques and the second one being constructed
by the method outlined below. One may object that given a classical function O there
must be a unique operator O such that its classical limit is given by O and not several
of them. However, in order to tell that a quantum theory given by a Hilbert space and a
quantum commutator algebra has a given phase space and Poisson algebra as its classical
limit one needs one more input: a selection of semi-classical states with respect to which
the semi-classical limit is to be performed. It is perfectly possible to have two systems of
operators and two systems of semi-classical states such that the classical limit of the first
set of operators with respect to the first system of states agrees with the classical limit
of the second set of operators with respect to the second system of states. There is no
claim that this will continue to hold after exchanging the two systems of states. In fact, our
current situation in quantum general relativity is the following: what we have are candidate
operators O already constructed in the literature and we have candidate semi-classical
states constructed in [3]. The staircase problem alluded to above reveals that these states
are not always semi-classical for the operators O . Thus there exist two avenues to arrive
at a satisfactory semi-classical analysis :
Either (A) one keeps the operators O and modifies the states to arrive at
better behaved semi-classical states or (B) one keeps the states and modifies
These are complimentary
the operators O to arrive at better behaved operators O.
programmes aiming at the same goal whose methods can be fruitfully combined so that one
meets somewhere in the middle. The advantage of avenue (A) is that the O are explicitly
known while the are not, for avenue (B) the are explicitly known while the O have
to be constructed graph by graph.
In this paper we will mainly focus on avenue (B) but we will devote this short paragraph
to describe avenue (A) in some detail in order to reveal that both avenues can actually be
combined.
Currently, avenue (A) is being studied in the context of the statistical geometry
approach [5]. The idea is to construct semiclassical states without making reference to
a particular, arbitrary but fixed, graph or any additional extra structure at all. The state
should be constructed just from one datum, the phase space point m, in such a way that
a sufficiently coarse-grained and complete subset of the set of the operators O already
constructed in the literature behaves semi-classically. Since these operators are manifestly
diffeomorphism covariantly defined, problem (II) never arises. In order to remove effects
similar to the staircase problem which are associated with a direction dependence on a
given graph, one obviously has to sum or average over a huge number of graphs. One could
consider either pure or mixed states but at this stage it is more natural to consider mixed
states because then we do not need to choose the phases of the probability amplitudes
involved but only their absolute values. Choosing this option, instead of looking at one
pure coherent state ,m one constructs a mixed state by averaging the one-dimensional
projections P ,m onto ,m over a subset m of the set of all allowed graphs (say piecewise
analytic, compactly supported) which may depend on m M with a probability measure
dm ( ) on m . Here, one can be very general to begin with so that ,m is not necessarily
one of our coherent states, see [5]. Thus, one tries to construct a density matrix (a trace
class operator of unit trace)

m := dm ( )P ,m
407
(2.3)
and now expectation values are given by O m := Tr(m O ) where O is an operator on

the currently used Hilbert space and not necessarily a function of the ge . Notice that, if
as in [5] the measure m is absolutely continuous with respect to a Lebesgue measure,
then as an operator on the continuum Hilbert space, m is (almost) the zero operator
(and has actually zero trace) due to the diffeomorphism invariance of the inner product,
however, the idea is to construct a new representation Hm of the operator algebra via the
GNS construction from this state. This should be compared with the strong equivalence
class Hilbert spaces based on pure coherent states as cyclic vectors that we will discuss in
Section 4.
Now, if the set m is large enough then due to the averaging performed in (2.3) the above
pointed out direction dependence (staircase problem) disappears. In fact, the explicit choice
made in [5] for flat initial data m and compact is such that m is Euclidean invariant
(using a weave state [7] or one of our coherent states for ,m ). The interested reader is
referred to this beautiful statistical geometry construction (based on the DirichletVoronoi
method [6] for Euclidean spatial metrics) for more details. In general, there is certainly a
lot of freedom in the choice of m , m and it has to be understood how O m depends on
these choices.
However, whatever choice is made, our coherent state construction fits neatly in with
the idea to construct density matrices as is obvious from the general formula (2.3). For
instance, using the peakedness properties derived in the third reference of [3] it is easy
to see that for flat initial data the coherent state is sharply peaked in the momentum
representation exactly at the value of the spin j that has been previously derived in the
weave construction! In other words, if we expand our coherent states in terms of spinnetwork states, then practically all the coefficients except for the one with spin label je 2p
Ar(Se ) are vanishing, where e is an edge of , Se is a face of a polyhedronal decomposition
of dual to (see below) which is intersected only by e and Ar(Se ) its area measured
by the flat three-metric. This dual polyhedronal decomposition, an extra structure which is
necessary to introduce in our coherent state construction below and which comes with some
freedom to choose from, is actually naturally fixed by the DirichletVoronoi construction!
This is a very surprising coincidence because the need for the polyhedronal decomposition
in the DirichletVoronoi construction on the one hand and in the coherent state construction
on the other hand are logically completely unrelated. Roughly then, if a macroscopic
surface S is assembled from the Se , the area expectation value can be practically taken
with respect to a weave spin-network state at the above specified values of je and gives
the correct result already reported earlier in the literature. If S is not assembled from the
Se it still might be possible, although no proof exists so far, that by using some of the
freedom just mentioned in our coherent state construction (specifically in the the map
,,X derived below) we can simply use our coherent states in this averaging method
408
in order to arrive at a satisfactory semi-classical state. These issues are subject to future
research in collaboration with the authors of [5].
Let us now come to avenue (B) for which at this point one will work with a single,
sufficiently large, gaph only rather than with an average. Let us stress that restricting
to a single graph is only due to mathematical convenience and by no means crucial. In
the future we might want to relax the condition to work with a single graph only and
to use some kind of average over a countably infinite number of graphs in the fashion
of (2.3). The measure m would in this case become a discrete (counting) measure
which would not result in a precisely Euclidean invariant density operator peaked on the
Minkowski metric but this is maybe not bad because fundamentally there should not exist a
precisely Euclidean invariant state in nature anyway if anything about the idea of a quantum
spacetime foam at Planck scale is correct. On the other hand, using a countable number of
graphs in the average would have the advantage that we obtain a non-vanishing trace class
operator of unit trace on the continuum Hilbert space and do not need to pass to a new
GNS Hilbert space.
The easiest single graphs to use that come to ones mind are regular lattices (say of cubic
topology). This has the obvious disadvantage that such graphs have preferred directions
and do not look isotropic at any scale larger than . One idea to remove this direction
dependence motivated by [5] and suggested to us by Bombelli is to use for our not a graph
that is embedded as and/or is topologically a regular lattice but rather a (generic) random
graph of the DirichletVoronoi type so that there are no preferred directions on scales larger
than or equal to an isotropy scale I which in turn is much larger larger than . As a result,
in measuring macroscopic observables (those that depend on a large number of elementary
ones) one does not detect the particulars of the graph any more. The isotropy scale I gets
absorbed in the particulars of the classical error (2.2) as we will see in Appendix B and
as a net effect influences the size of the curvature scale L. Having said this, we will not
explicitly display the dependence of L on I any more in this paper.
Our first task is to show that one can indeed solve problem (II), that is, the O have to
be constructed in diffemorphism covariant fashion. To do this, recall that in the canonical
approach to quantum gravity one assumes that the four-dimensional manifold M has
topology R where is an analytic three-dimensional manifold of given topology. The
graphs are unions of edges which themselves are one-dimensional analytic submanifolds of
intersecting at most in their endpoints. Likewise, classical functions OS on M typically
depend on analytic submanifolds S of like curves, surfaces and regions. Also, as we
will see below, in order to define the graph degrees of freedom we must actually also
specify a polyhedronal decomposition P dual to and a set of paths inside the
corresponding faces. One can describe these structures mathematically by introducing a
coordinate system X (to which we will refer as an embedding in what follows) and model
graphs , faces P , sets of paths and surfaces S in the model space of (that is, R3 )
In what follows, model
such that = X( ), P = X(P ), = X( ), S = X(S).
objects like these will carry a check sign as compared to their embedded counterparts.
Given , fix once and for all a coordinate system X0 . Given a pair , S, choose once
and for all a representant (0 , S0 ) in their orbit [( , )] := {(( ), (S)); Diff()}
409
under diffeomorphisms and a diffeomorphism S, such that S, (0 ) = , S, (S0 ) = S.

Furthermore, we require that 0 is the same graph for pairs ( , S), ( , S ) such that ,
are diffeomorphic. Notice that 0 , S0 , S, are far from unique and in order to choose them
we make use of the axiom of choice.
Define the model graph and surface , S respectively by 0 = X0 ( ), S0 = X0 (S)

and choose once and for all model faces and paths P , for . Pick a prescription
to construct a model function OS,
,P , , which depends only on the model graph degrees
of freedom ge . The model graph degrees of freedom are labelled explicitly by a model
edge e (but no dual face and corresponding path system S e , e respectively since we
fixed them once and for all) and take values in a model phase space M as we will see
below. By definition, the numerical coefficients that the model function involves do not
explicitly depend either on X0 or on m! Rather they depend only on , P , , S and
Now, given X0 , there is a natural map ,,X
possibly additional extra structures in .
0
from M into M defined by ge := ge0 (m) =: ,,X0 (m) where ge0 depends explicitly
S e0 = X0 (S e ), e0 = X0 (e ) respectively. The
on the embedded structures e0 = X0 (e),
point is now that if OS was diffeomorphism covariantly and background independently
defined then there is a natural way to construct this model function in such a way that
OS,
,P , ,,X0 is close to OS0 pointwise in M as we will explicitly demonstrate in
Appendices A, B.
We can then finish the construction of OS, through the definition

OS, (m) := OS,
(2.4)
S, ge0 (m) e E( ) ,
,P ,
where ( ge )(m) := ge ( 1 m) is the natural action of Diff() on the ge induced by

that on M and E( ) denotes the set of edges of . Now it is easy to see that due to
the diffeomorphism covariance of the functions ge we actually have S, ,,X0 =
,,XS, where X := X. We can therefore write (2.4) more compactly as
OS, := OS,
,P
,,XS, ,
(2.5)
which shows we have automatically a natural action of the diffeomorphism group on OS,
given by

( OS, )(m) = OS, 1 (m)
(2.6)
as expected from a function on the classical phase space. However, in general
OS, = O(S),( ) in contrast to OS = O(S) for the exact classical function.
Since lim0 OS, (m) = OS (m) pointwise on M by construction, this behaviour under
diffeomorphisms is expected only in the continuum limit although it will be close to it
for sufficiently small . This violation of a continuum property of the classical function is
the price to pay for making the function graph dependent. There are two complementary
points of view, reached in a discussion with Ashtekar, Lewandowski and Pullin: either one
regards the approximate operators as effective ones, good only for a semi-classical regime.
Or one takes the opposite point of view, that since we want to quantize general relativity
non-perturbatively and background-independently, we are forced to adopt a non-standard
410
Hilbert space whose elements are labelled by graphs. Therefore, the OS, are the more
fundamental building blocks from which everything else must be derived and as long as a
continuum property is regained in the limit of infinitely fine graphs, this is acceptable. This
issue will be explored in more depth in future publications.
As we will show below, all that is important about the map ,,XS, for quantization
purposes is that XS, ( ) = so that the correct quantization of the functions (2.5) is given
by

O S, := OS,
(2.7)
,P , {g e }eE( ) .
Namely, we will show below that this operator, interpreted as an operator on the closed
subspace H of the full continuum Hilbert space H spanned by spin-network functions
over [8] with dependence on every edge by non-trivial irreducible representations of
SU(2), can be written in terms of holonomy operators h e and momentum operators Pje
acting by multiplication and differentiation with respect to he , respectively, where e
E( ). Then, interpreting ge as the pair (h e , Pje ) there is a natural quantum action of the
diffeomorphism group on these operators defined by
U ()ge U ()1 = g (e)
and U ()O S, U ()1 = OS,

,P
(g(e) ),
(2.8)
so that it is completely diffeomorphism covariantly defined. Moreover, it leaves the

subspace H := H invariant (the sum running over subgraphs of ). The
corresponding family of operators O S := {O S, } may not yet be consistently defined
in the sense that (O S, )|H = (O S, )|H for any . If that is not already the case

we replace O S, by O S, P (or P O S, P to make it explicitly selfadjoint) where P is an orthogonal projection onto H . From now on we assume that
O S has been consistently defined like this and thus we have arrived at a new definition
of continuum operators O which are diffeomorphism covariantly defined without going
through a regularization procedure. This finishes our task associated with problem (II).
In what follows, we will only work with this new kind of operators and drop the label S
for OS again, O wil be a general classical function on M. Then the next question is: given
, m, for which graph should one construct a semi-classical state in order to arrive at the
desired semi-classical interpretation, in particular, what is the size of ? This is connected
with the question of the continuum limit 0. Certainly, should be small enough so that
the classical error (2.2) is small. But there are also quantum errors: first of all, unless the
corresponding quantum operator O := O (ge ) is normal ordered we also have a normal
ordering error

,m , O ,m

1,
(O )normord(m) :=
(2.9)
O (m)
which in applications is usually of the same order as the the quantum fluctuation of O
given by
2
1/2

O
1 ,m
.
(O )quant(m) := ,m ,
(2.10)
O (m)
411
These quantum errors are not yet the quantum errors of our consistently defined continuum
However, since when computing expectation values or fluctuations of O on
operator O.
cylindrical functions we actually project O to one of the O we can introduce the following
notion of fluctuation of a real valued observable O for which we assume O to be selfadjoint:
2
2

1/2
1/2
O
O
1 ,m
1 ,m
= ,m ,
,
(O)total(m) := ,m ,
O(m)
O(m)
(2.11)
which measures the fluctuation of the substitute operator O compared to the exact
classical value O(m). It is important to realize that due to cylindrical consistency we could
This quantity can be written in terms of the classical, normal
replace O in (2.11) by O.
ordering and quantum error as

O 2 O (m) 2
2
(O )quant(m)2
(O)total(m) =
O (m)
O(m)

2

O (m) O
O (m)
1 +
1 .
+
(2.12)
O(m) O (m)
O(m)
Eq. (2.12) allows us to derive the following important inequality

(O)total(m) 1 + (O )normord(m) 1 + (O)class(m) (O )quant(m)

+ 1 + (O)class(m) (O )normord(m) + (O)class(m)
(2.13)
and thus will be small if both (O )quant (m) and /L are small. If practically possible
and physically motivated one should of course normal order the operator O . These
fluctuations will be the smaller the more elementary observables, say N , O involves
additively (see first reference of [2]), that is, the more macroscopic O is. This
is because
of the law of large numbers for fluctuation errors which decreases as 1/ N . The least
macroscopic observables that we can measure are the elementary ones, O = ge for which
we require still good semi-classical behaviour and which we will call the mesoscopic
ones. We obviously will not be allowed to choose as small as p or lower because then
(O )quant (m) will be large since we would probe the geometry at the microscopic Planck
scale.
These considerations suggest the relation p L(m) or p L0 if we want
to be independent of m. We think of as a parameter for graphs which we can adjust in such
a way as to minimize fluctuations. We will see later in Appendix B that minimization fixes
to be of the order of = rp L(m)s or = rp Ls0 where r, s are positive rational numbers
adding up to one, because there is a competition between the classical error which becomes
large for large and the quantum error which becomes large for small . Moreover, the
right-hand side of (2.13) turns out to be (not surprisingly) of the form (p /L(m))
(p /L0 ) for some positive rational number . We see that the continuum limit 0 and
the classical limit h 0 are no longer separate limits but happen simultaneously. This
at least restricts our freedom to choose a graph, given , m in order to enforce maximum
412
semi-classical behaviour. Of course, if L(m) = , which happens if m corresponds to flat

initial data, then there is no optimal value for because the classical error vanishes, there
is only a quantum error which becomes the smaller the larger and we will choose of
the order of the smallest scales that we wish to resolve, say of the order of the wave length
of a ray burst [25].
We are then left with the specification of the degrees of freedom ge and the maps
,,X , or equivalently, the ge . Once we have singled them out, we will choose the states
,m to be coherent states for the ge , that is, an overcomplete set of minimal uncertainty
states for the ge , see Section 3. Given an embedded graph and point m M we simply
take the holonomies he (A) along its edges as the configuration degrees of freedom where
A is the connection datum specified by m = (A, E). As already announced, in order to
specify momentum degrees of freedom we have to blow up the label set somewhat: we
consider a polyhedronal decomposition P of the three-manifold such that the graph
is dual to it. Thus, for each edge e of there is a unique, open, face S e of the polyhedronal
decomposition which it intersects transversally in an interior point pe of both e and S e
and whose orientation agrees with that of e. Next, for each x S e we choose a non-selfintersecting path e (x) within S e which starts at pe and ends at x without winding around
any point in S e . The collection of these paths will be denoted by . Finally, let a be a
fixed parameter of the dimension of a length. Then we define the dimensionless quantity

1
e
Pj (A, E) := 2 tr j Adhe (pe )
(2.14)
,
Adhe (x) E(x)
2a
Se
where he (pe ) is the holonomy along the segment of e starting at the starting point of e and
ending at pe , denotes the metric independent Hodge dual and E = Ej j with respect
to generators j of su(2) with the properties tr(j k ) = 2j k , [j , k ] = 2j kl l . Notice
that (2.14) is gauge covariant, that is, under local gauge transformations it transforms
in the adjoint representation of SU(2) at the starting point of e. It is also background
independently and diffeomorphism covariantly defined. Moreover, Pje (m) depends linearly
on the electric field datum E specified by m. The interpretation of and the motivation for
the quantitity a with the dimension of a length is not clear at this point but it will become
obvious later on when we compute fluctuations where it will be fixed at the order of L(m)
or L0 . Also, tuning a will tune the Gaussian width of the coherent states that we construct.
Together with the holonomies he (m) of the connection along edges we obtain the
elementary graph degrees of freedom

ge,,P , (m) := he (m), Pje (m) ,
(2.15)
which, abusing the notation, are specific functions on M that depend on , , P , . It
is therefore very surprising at first that they satisfy a Poisson algebra (see the first reference
in [3]), derived from the canonical Poisson brackets {., .} between the A, E on M, which
is completely independent of , P , , specifically
{he , he }(m) = 0,
e

j
he (m),
Pj , he (m) = 2 ee
2
a
413

Pje , Pke (m) = ee 2 j kl Ple (m).
(2.16)
a
With a little experience in background independent quantum field theory, however, this is
not so surprising, it just mirrors the fact that no metric, distances, angles etc. can appear
in (2.16).
Let us now forget about the structures , P , altogether and consider a model
graph which can be embedded as into . Let us also forget the continuum phase
space (M, {., .}) and rather consider a model phase space (M , {., .} ) with coordinates
he SU(2), Pje R3 which specify ge = (he , Pje ) and the point m := {ge }eE(
) where
E( ) is the set of directed edges of . These coordinates enjoy the basic brackets inherited
from (2.16)
{he , he } = 0,
e

j
Pj , he = 2 ee he ,
a
2
e e
ee
j kl Ple ,
Pj , Pk =
(2.17)
a2
which defines the natural symplectic structure of the |E( )| fold copy of the cotangent bundle T SU(2), displaying M as the cotangent bundle over the |E( )| fold copy of SU(2).
We now want to construct coherent states for M . These will be states on the model
Hilbert space (which is of course isomorphic with H )

L2 SU(2), dH ,
H :=
(2.18)
eE(
where H is the Haar measure on SU(2), that is, they will functionally depend on the point
h := {he }e of the configuration subspace of M . On the other hand, we want them to be
peaked at a point m of the phase space M and therefore they will carry a label m . The
tensor product structure of the Hilbert space makes it obvious that these states will be of
the form of a direct product

gt e (he ),
,m (h ) =
(2.19)
e
where gt (h) is a coherent state on the Hilbert space L2 (SU(2), dH ) with label g
T SU(2) functionally depending on h SU(2). These states depend explicitly on the
dimensionless classicality parameter
t=
2p
a2
(2.20)

which naturally arises in quantizing the above Poisson bracket, e.g., [Pje , Pke ] =

it ee j kl Ple . It is important to see that the Hilbert space (2.18) carries a faithful
representation of these commutation relations and the adjointness conditions (Pje ) =
P e , (h AB ) = (h BA )1 following from the classical reality (unitarity) respectively if h e
acts by multiplication and Pje is it/2 times the right invariant vector field on the copy of
SU(2) corresponding to he generating left translations in the direction of j in su(2).
414
More explicitly these states are labelled by points in SL(2, C) which is possible since
one can identify ge with the following element of SL(2, C) (we abuse the notation in using
the same symbol)

ge := exp ij Pje /2 he ,
(2.21)
which is actually a diffeomorphism from T SU(2) to SL(2, C) (the inverse map being
given by polar decomposition). Since the former is a sympletic manifold and the latter
a complex manifold it is not surprising that SL(2, C) is actually a Khler manifold (the
complex structure is compatible with the symplectic structure) which suggests to use well
known methods from geometric quantization (in particular, heat kernel methods) for the
construction of the coherent states. These methods will be described in some detail in
Section 3, here it is sufficient to contemplate that the situation is actually quite analogous
to the harmonic oscillator coherent states which are labelled by a point z in the complex
plane, whose real and imaginary part respectively can be interpreted as configuration q and
momentum p degree of freedom respectively, and which depend functionally on a point
x R in the configuration space of the phase T R. The complexification of the real line
is the complex plane and the complexification of the group SU(2) is given by SL(2, C).
Thus, (2.21) is nothing else than a non-Abelian and exponentiated version of z.
This finishes the construction of a coherent state on the model Hilbert space H labelled
by a point in a model phase space. We now must make contact with the Hilbert space
H and the phase space M appropriate to describe a semi-classical situation for a given
topology . The formulae (2.14), (2.16) provide the necessary clue: given a model graph
, a three manifold , an embedding X of into as = X( ), a polyhedronal
decomposition P of dual to and a choice of paths within its faces we can
construct the concrete maps (2.15) which provide us with a map (we display only its
dependence on but the dependence on the other structures should be kept in mind)
(m)}eE( ) ,
,,X : M M ; m {ge := gX(e),,X
(2.22)
which, according to (2.16), (2.17), is actually a symplectomorphism (more precisely, it

leaves the Poisson brackets invariant but it is not invertible and not even necessarily
surjective if is not compact since then m needs to obey certain fall-off conditions). The
map (2.22) is the final ingredient to obtain a semi-classical state appropriate for a given
and phase space point m
,m (h ) := , ,,X (m) (h ),
(2.23)
where we have identified H with H , in particular, he with he whenever X(e)

= e.
Since the model Hilbert space is the same for any , this opens the possibility that
one might be able to to describe topology change within canonical quantum general
relativity: if we can define consistently a new representation of the canonical commutation
relations based on model graphs rather than embedded ones, then the transition amplitude
, ,,X (m) , , , ,X (m ) is well-defined since the scalar product is computed on
the common model Hilbert space and not on the embedded one and one and the same
model graph can be embedded differently into different . We elaborate more on this
415
idea in future publications [15] where methods from algebraic graph theory will become
crucial [16].
By construction, the state (2.23) is sharply peaked at Pje = Pje (m), he = he (m) (these
values depend explicitly on , X, P , ) and enjoys further important semi-classical
properties as we will describe in detail in the next section. They thus are good candidate
states to base a semi-classical analysis on.
Let us summarize:
(i) Classical Continuum Phase Space
For any given three manifold we start from a classical, continuum phase space
(M, {., .}) based on with points m and a continuum Poisson algebra O of
classical observables O.
(ii) Classical Discrete Phase Space
Given a model graph we have a classical, model phase space (M , {., .} ) (direct
sum of copies of T SU(2), one for each edge of ) with points m = {ge } and
classical model Poisson algebra O of observables O respectively. No reference to
a particular manifold is made. For any , a subset of M can be obtained as the
image of M based on under a symplectomorphism ,,X of the form (2.22).
(iii) Discrete Quantum Hilbert Space
The model symplectic manifold is easy to quantize and gives a discrete quantum
theory given by the model graph Hilbert space (H , ., . ) (direct product of of
copies of L2 (SU(2), dH ), one for each edge of ) with coherent states t ,m and
model quantum commutator algebra O of model quantum observables O . By

construction, the classical limit of (H , ., . , O ) gives us back (M , {., .} , O )
which means that for all m , O the zeroth order term in h of the quantities
t

,m , O t ,m O (m ) and
t ,m ,
[O , O ]
it

t ,m
{O , O } (m )
(2.24)
and of their fluctuations, e.g., (O )quant(m ), vanishes (so-called Ehrenfest

property).
(iv) Continuum Quantum Hilbert Space
Given a model graph and an embedding X into some we can identify the model
and he := he . Likewise, given a model
Hilbert space H with H where e = X(e)
operator O on H we obtain an operator O on H by substituting multiplication
by and differentiation with respect to he for multiplication by and differentiation
with respect to he . By this method we obtain diffeomorphism covariant families
of operators O and we consider only those that are cylindrically consistent. The
operators that we have constructed above are of this type and the algebra of these
operators will be called O.

Given , M and a classical observable O we produce by the method described
above such a consistently defined family of operators O = {O } based on classical
416
functions O of the graph degrees of freedom. Among all possible graphs , by

construction there exists at least one 1-parameter family of graphs such that
O(m) = lim0 O (m (m)) for any m M. Thus, while for a generic graph O
will not have a semi-classical interpretation, the operators O do. The point is now
that H H, ., . = ., .|H , O = O |H . Therefore, since the continuum limit
is coupled to the classical limit (no separate continuum limit to be taken) we can
make the following definition:
is given by
We will say that for given the classical limit of (H, ., ., O)
(M, {., .}, O) if for all m, O the zeroth order of the h expansion of the quantities

[O , O ]
,m {O, O }(m)
,m , O ,m O(m) and ,m ,
it
(2.25)
and of their fluctuations, e.g., (O)total(m), vanishes (and that the corrections are
in agreement with experiment). Here = (p , L(m)), a = a(p , L(m)) or =
(p , L0 ), a = a(p , L0 ) take their optimal value as specified above so that they
depend only on p , , m or p , , M. Notice that in a label for a coherent state we
may take a to be m-dependent although non-trivial m-dependence of a would spoil
the Poisson brackets (2.16). This is because in the model Poisson brackets (2.17) we
can take a to be any function of m without changing them, the model phase space
and the phase space M have nothing to do with each other a priori. Alternatively, a
is an m-independent real parameter to begin with, it just gets fixed by optimization
through a quantum computation in a label for a coherent state at an m-dependent
value a(m), thus Poisson brackets are to be evaluated by taking a as m-independent
and after computing them one should set it equal to a(m). This is also obvious
from the fact that expectation values of commutators divided by it reproduce the
Poisson brackets of the classical function by treating a as m-independent. In the
same manner we could also turn a into a function of edges (paths) in order to bring
in some locality and in order to increase our flexibility in fluctuation calculations
but we will not do that in the present paper for the sake of simplicity.
We have thus arrived at a closed diagramme of how to pass from a given continuum,
classical theory to a continuum, quantum theory and back by taking a route through a
discrete regime of both theories. Moreover, we have derived an explicit criterium for how
to decide whether canonical quantum general relativity has classical general relativity as its
classical limit. The Ehrenfest theorems proved already in [3] and the explicit construction
of the operators O which guarantees that for given , m such that = , = (p , m)
both its classical and quantum error are small indicate that this might actually be the case.
A couple of remarks are in order:
(1) Our scheme works only if the operators O leave the Hilbert space H invariant.
However, the diffeomorphism and Hamiltonian constraint operators that have been
constructed in the literature (e.g., [911]) do not have this property. Either one must
modify them in such a way that their cylindrical projections leave the cylindrical subspaces
invariant or we have to modify our scheme as follows for such operators: if O maps H
417
into H for some = then find the minimal (in the sense of numbers of edges) such
that , and replace ,m in (2.25) by ,m . Strictly speaking then, (2.25) does
not really test the continuum operator O but only its cylindrical projections O which is
not what we really want but which is the best we can do at the moment.
(2) All the coherent states that we defined are labelled by a specific and although all
of them belong to the continuum Hilbert space it would be nicer and it would solve the
problem just mentioned in (1), if we could find a state m such that its projection onto H
coincides with ,m . This could be achieved if our family of states ,m would solve the
consistency condition that ,m projected onto H coincides with ,m for any .
It is easy to see that this would require (see also [13] for a related observation) (a) to make
the parameter t in (2.19) a function of the edge e, that is, t = t (e) such that t (e1 ) =
t (e) and t (e e ) = t (e) + t (e ) and (b) that ge1 = ge1 and gee = ge ge which are
precisely the defining algebraic relations for a length function and an SL(2, C) holonomy
respectively. There is no problem to generalize our scheme and to make t a function of
edges. However, our coherent states are constructed in such a way that the operators Pje ,
h e are peaked at P e , he where ge = exp(iP e j )he and the holonomy property is in
j
conflict with the interpretation Pje = Pje (A, E), he = he (A) given in (2.14). One can,
of course, turn the argument the other way around and just define a classical complex
j
connection, say AC
b (m) := Ab ieb /a where eb is the co-triad. Then for given and
phase space point m we can define ge := he (AC (m)) where the latter means the holonomy
of the complex connection determined by m along the embedded edge e = X(e).
Then, if
p L0 the coherent state is peaked at the value

1
he he (A),
(2.26)
Pje pje (A, E) := tr j
Adhe (A,x) e(x)
,
2a
e
where he (A, x) denotes the holonomy of A along e from the starting point of e until the
point x e. The advantage of this new map ,,X is that we need less structure (no
polyhedronal decomposition, no choices of paths). The disadvantage is that (a) (2.26) holds
only approximately while (2.23) holds exactly independent of the choice of and (b) there
is no closed classical Poisson algebra underlying the functions he (A), pe (A, E): first of
all, the electrical field wants to to be smeared in two rather than in one dimension so
j
the algebra becomes necessarily distributional. Secondly, since ea or any other Lie algebra
valued co-vector that can be built from Eja is a non-polynomial function of the electric field,
the Poisson algebra actually does not close. It follows that the relation (2.26) between the
model graph phase space and the classical phase space does not survive this interpretation
of ge at the level of Poisson brackets. Therefore, the expectation value of the commutator
between the model elementary graph operators divided by it (which is non-distributional)
cannot be given by the Poisson brackets of the functions on the right-hand side of (2.26)
(which is distributional) as computed with the symplectic structure of M.
(3) The coherent states that we constructed are kinematical coherent states, i.e., they are
not annihilated by the constraint operators of the theory (in fact, they are not even gauge
invariant). One can rightfully ask how such states can possibly be used in the semiclassical
418
analysis. The answer is the following. First of all, one of our aims is find out whether
the Hamiltonian constraint operators constructed in [11] (the alternative proposal of [12]
cannot yet be treated because it is so far not embedded in a Hilbert space setting) really
do have the correct classical limit and obviously it is then not allowed to consider physical
coherent states which would be automatically annihilated by them.
Secondly, the gauge and diffeomorphism group are unitarily implemented on the
Hilbert space and therefore expectation values and fluctuations of gauge and spatially
diffeomorphism invariant operators are in fact gauge and spatially diffeomorphism
invariant. This leaves us with the issue of observables which are invariant under the
motions of the Hamiltonian constraint which famously cannot be implemented unitarily.
The problem is, of course, that we do not know any such observable explicitly so that it
is impossible to test the usefulness of kinematical coherent states in full quantum general
relativity directly. However, one can study exactly solvable model systems and it turns
out that for a large class of such systems [14] including those with a non-polynomial
Hamiltonian constraint and for which the Dirac observables are non-polynomial functions
of kinematical observables, the expectation value of these Dirac observables and their
commutators divided by it as measured by kinematical coherent states precisely coincides,
to zeroth order in t, with those as measured by dynamical coherent states provided
we choose m on the constraint surface of the phase space, no matter in which gauge.
Moreover, the fluctuations of these operators are of the same order in h . Thus, while
corrections to the classical theory are not independent of ones choice of kinematical versus
dynamical coherent states, one can control the size of their difference and the classical limit
itself seems to be unaffected. This is of course no proof that the same will be true in full
quantum general relativity but it is a non-trivial consistency check. Of course, since we
only check a finite number of commutators in the classical limit h 0 we have only
control over the infinitesimal dynamics and no a priori control on the finite time evolution
of the system. Fortunately, in principle the finite time evolution is not necessarily required
in order to compute Dirac observables and this is all we need since we do not have a
Hamiltonian but only a Hamiltonian constraint.
Finally, if we consider the GNS representations to be discussed in Section 4 with respect
to a coherent state labelled by a point on the constraint surface of the phase space then
constraint operators will have expectation value zero (or at least very close to if not
normal ordered) although their fluctuations do not vanish. Thus, such representations can
be considered as approximately physical representations, a point of view independently
reached also by Ashtekar (they would be exactly physical if the constraint operators would
be the zero operators).
(4) The careful reader may object, as pointed out by Ashtekar and Lewandowski, that in
order to make our family of operators cylindrically consistent we must in general introduce
the projection operators P which have no classical counterpart and thus it is actually
no longer clear that the classical limit as defined in (2.25) still gives the desired result.
Fortunately, there is no problem: at the optimal value of , a derived in Appendix B we find
for the projection Pe on spin-network states over a single edge e the expectation value 1
e
1
t 2/3
where t 2p /L20 is a tiny quantity. Thus the expectation value is 1 + O(t ). For a
419
general graph a similar argument holds provided that the projection operators needed in the
definition of O are finite products of the Pe . But this is the case for operators which come
from classical functions defined as integrals over regions, surfaces and curves provided
that the valence of the vertices of is bounded from above, even if has a countably
infinite number of edges. It should be clear from our exposition that on pathological graphs
which lack this boundedness property a good semi-classical behaviour cannot be expected
anyway so that this is not a restriction.
3. Coherent states on cotangent bundles over compact gauge groups

As explained in the previous section, a first step towards constructing coherent states
for quantum general relativity is to construct such states for SU(2). This was achieved
in the general setting of compact Lie groups by Hall [17]. To make our presentation more
pedagogic, we start by reviewing the situation for the harmonic oscillator in one dimension.
Here the wave function is given by
1
2
e(zx) /2t ,
zt (x) =
2t
where z = x0 ip0 and t = h /m, and zt is an element of L2 (R, dx). As is well-known
zt (x) has a Gaussian shape with peak at x0 , or, more generally, at (z). Similarly, by
transforming zt (x) to momentum space one finds that it is peaked there around p0 ,
or, again more generally, around (z). Furthermore, the sharpness of the peak in each
representation is given by t. For that reason it is usually stated that coherent states
approach the delta function as one considers the limit h and therefore t to zero. What
is really meant by this limit, of course, is that the physical quantities characterizing the
system under consideration here, for example, m and are much, much larger than the
value of h . In that sense, the classical limit of, e.g., expectation values with respect to these
coherent states is given by the zeroth order and semiclassical corrections by the first and
higher orders of a power expansion in t. Further important properties of these states are
that they are overcomplete, they satisfy the Heisenberg uncertainty relation and they are
eigenstates of the annihilation operator: (zzt 0 )(x) = z0 zt 0 (x).
To make the transition from the harmonic oscillator to the compact Lie group case,
a more technical derivation of the former is helpful. We start with the classical phase
space P , coordinatized by (x, p). By choosing a positive function C on P the so-called
complexifier we change to a complex polarization of P by defining
z1 :=
z2 :=
n

i
n!
n=0
n

i
n=0
n!
{x, C}n ,
{p, C}n ,
(3.1)
where {, }n stands for a multiple Poisson bracket of order n. The phase space P is now
coordinatized by (in general) complex coordinates (z1 , z2 ). In the case of the harmonic
420
oscillator, choosing C = p2 /2 results in the usual z1 = x ip and z2 = p. One proceeds

with quantization by replacing Poisson brackets by commutators divided by i h and by
promoting classical functions to operators. For the harmonic oscillator this can be done
without problems:
z =
n

h
n!
n=0
n = W t x W t1
[x,
C]
(3.2)
C
with W t = e t , which for the harmonic oscillator is just the heat kernel as then C is, up
to factors, the Laplacian on R. This operator W t also establishes the connection to the
coherent state transform which is a unitary transformation U t from the position Hilbert
space L2 (R, dx) to the phase space Hilbert space (SegalBargmann space) HL2 (ZC, d)
of holomorphic L2 -functions over C. The new measure is determined by the unitarity
requirement. The transform is defined by

(U t f )(z) := dx t (y, x)f (x)|yz =: zt , f ,
(3.3)
where t (x, y) := (W t y )(x) and y z means analytic continuation from y to z. From

these relations one can thus deduce the following definition of the harmonic oscillator
coherent states:

zt (x) := et /2y (x)yz ,
(3.4)
which one can check to be identical to the form given at the beginning of this section. It
should be kept in mind that, while the coherent state transform leads to a Hilbert space
over C, the associated coherent states themselves are still elements of the position Hilbert
space over R.
Halls idea was to follow this construction for compact Lie groups as closely as possible.
Instead of wavefunctions over R one considers wavefunctions over a compact Lie group
G, dx is replaced by Haar measure dH , the complex label z = x ip by an element of
the complexified group GC that can be written in polar decomposition as g = H h with
h G and H positiv definite hermitian, and finally the Laplacian on R is replaced by the
LaplaceBeltrami operator G on G. Coherent states are then analogously defined by

gt (h) := et G /2 H ,h (h)h g .
(3.5)
Using the expression for the delta function on Lie groups one arrives at the explicit
expression

t
d e 2 gh1 ,
gt (h) =
(3.6)
where is (a representant of) an irreducible representation of G, d its dimension, the

eigenvalue of G in the representation , and denotes the trace in that representation.
The trace is to be understood in the following sense: One first calculates (h h1 ) with
h G and then performs the analytical continuation from h to g. Hall was also able to
421
show that these coherent states form an overcomplete system of states on the Hilbert space
L2 (G, dH ).
To use Halls construction for quantum general relativity we specialize to G = SU(2).
Then three additional inputs are required:
One has to establish the connection between the complex group element parameterizing the coherent state and the coordinates on the classical phase space of general
relativity, that is connections and electric fields. This is the analogue of z = x ip in
the case of the harmonic oscillator.
One has to show that the coherent states so defined have the properties important
for a (semi-)classical interpretation. Most importantly, they should be peaked in the
configuration, momentum and phase space (SegalBargmann) representations around
the points in phase space specified by g, respectively. Also, Ehrenfest theorems should
hold, that is expectation values of operators and their commutators divided by it
should approach the corresponding classical functions and their Poisson brackets in
the limit t 0.
Finally, one has to extend Halls framework from single copies of the gauge group to
multiple ones. In the language of quantum general relativity this means the extension
from coherent states defined over edges to those defined over graphs [3].
We begin by defining the complexifier for an edge e by Ce = 12 Pie Pje ij where Pie
are the momentum variables for general relativity as defined in Section 2. The complex
holonomies are then given by
hC
e := ge =
n

i
n=0
n!
{he , Ce }n = e
ij Pje /2
he = He he ,
(3.7)
where He is positive definite Hermitian and he is an element of SU(2). Notice that ge now
carries information on the connection as well as on the electric field (over the edge e). The
t
e = 1 ij X e X e , where the X e are the derivative
heat kernel operator is W t,e := e 2 e with
i j
j
4
operators coming from the quantization of the Pje . The coherent states for a single edge e
have now the form

gt e (he ) =
(3.8)
(2j + 1)etj (j +1)/2j ge h1
e ,
j =0, 21 ,1,...
where ge given by (3.7) can be computed, given a point in the classical phase space of
general relativity as in the previous section and j runs over SU(2) spin representations.
l2
The peakedness or classicality parameter t is given by t = ap2 where a has dimension of

length. Like the mass and frequency for the harmonic oscillator it characterizes the system
under consideration and in practice is fixed as outlined in Section 2. Finally we can also
define the annihilation operator
ge := e 2 e h e e 2 e = e 8 ei Pj j /2 h e .
t
3t
(3.9)
Here we have used quotation marks because, although the coherent states are eigenstates
of ge , up to now it is not yet clear whether this operator can be used to derive a Fock
422
space structure. Other properties of the coherent states that follow immediately from these
definitions are that expectation values of normal ordered polynomials of the ge and their
adjoints are given by the corresponding classical functions (without quantum corrections)
and that the self-adjoint operators defined by xe := 12 (ge + ge ) and ye := 2i1 (ge ge )
saturate the unquenched Heisenberg uncertainty relations. Much more work is required
to prove the peakedness properties so essential for a semiclassical interpretation. In the
configuration and thus connection representation, e.g., one has to prove that
pgt (h) :=
|gt (h)|2
"gt "2
(3.10)
is sharply peaked at h = h where g = H h is the polar decomposition of g, with sharpness

being given by the parameter t. More explicitly, one needs to show that an upper bound
for pgt (h) can be given that (roughly) has the form of a Gaussian times a quantity that
decays to 1 exponentially fast with t 0. A first look at (3.8) reveals that this is not
straightforward as the convergence producing term etj (j +1)/2 converges very slowly for
t 0. To circumvent this, we employed in the third reference of [3] the so-called Poisson
transformation that transforms the sum in (3.8) into a sum where the exponential contains t
in the denominator. This speeds up convergence dramatically and thus allows one to derive
the desired results, even though calculations are still very tedious to perform.
By similar methods peakedness was shown in the momentum and phase space
representation. In addition, as a further check-up and for illustration purposes, these
calculations were also performed numerically with a confirmation of the analytical results,
see [3] for some graphics. As a result, the coherent states defined above have been shown
to be peaked sharply around a point in the classical phase space (in contrast, e.g., to the
so-called weaves [7], which so-far had been used to approximate classical spacetimes, but
were peaked only in the momentum configuration). They thus serve as a good starting point
to approximate classical spacetimes (remember that quantum states can never reproduce a
classical state exactly, due to the uncertainty relations), even more so as we were able
to derive Ehrenfest theorems for them in the fourth reference of [3]. This means that
j
expectation values of arbitrary polynomials in the h e and pe with respect to the coherent
j
states approach the corresponding polynomial of he and pe on phase space in the limit
t 0. Also, expectation values of commutators divided by it approach the value of
the corresponding Poisson bracket, which means that with respect to the coherent states,
infinitesimally, the quantum dynamics reproduces the classical dynamics.
Finally, coherent states over graphs are constructed by simply tensoring together those
over the respective edges, that is

t
{h} :=
{g},
(3.11)
gt e (he ),
eE( )
where E( ) is the set of all edges of the graph and {g}, {h} denote the collections
ge1 , . . . , gen , he1 , . . . , hen of the associated group elements, respectively. The peakedness
and Ehrenfest properties discussed above for the single-edge case also go through for the
423
case of graphs. This follows from the fact that peakedness on single edges implies peakedness on the whole graph and vice versa, and from the fact that operators on different edges
commute. This tensor product structure of the coherent states brings us naturally to the
next subject.
4. The infinite tensor product extension

Quantum field theory on curved spacetimes is best understood if the spacetime is actually flat Minkowski space on the manifold M = R4 . Thus, when one wants to compute the
low energy limit of canonical quantum general relativity to show that one gets the standard
model (plus corrections) on a background metric one should do this first for the Minkowski
background metric. Any classical metric is macroscopically non-degenerate. Since in the
Hilbert space that has been constructed for loop quantum gravity (see [2] for a review and
[18] for more details) the quantum excitations of the gravitational field are concentrated
on the edges of a graph, in order that, say, the expection values of the volume operator for
any macroscopic region is non-vanishing and changes smoothly as we vary the region, the
graph must fill the initial value data slice densely enough, the mean separation between
vertices of the graph must be much smaller than the size of the region (everything is measured by the three-metric, determined by the four metric to be approximated, in this case
the Euclidean one). Now R4 is spatially non-compact and therefore such a graph must necessarily have an at least countably infinite number of edges whose union has non-compact
range.
The Hilbert spaces in use for loop quantum gravity have a dense subspace consisting of
finite linear combinations of so-called cylindrical functions labelled either by a piecewise
analytic graph with a finite number of edges or by a so-called web, a piecewise smooth
graph determined by the union of a finite number of smooth curves that intersect in a
controlled way [19], albeit possibly a countably infinite number of times in accumulation
points of edges and vertices. Moreover, in both cases the edges or curves respectively are
contained in compact subsets of the initial data hypersurface. These categories of graphs
will be denoted by 0 and 0 , respectively, where , , 0 stands for analytic, smooth
and compactly supported, respectively. Thus, the only way that the current Hilbert spaces
can actually produce states depending on a countably infinite graph of non-compact range
is by choosing elements in the closure of these spaces, that is, states that are countably
infinite linear combinations of cylindrical functions.
The question is whether it is possible to produce semi-classical states of this form, that

is, = n zn n where n is either a finite piecewise analytic graph or a web, zn is a
complex number and we are summing over the integers. It is easy to see that this is not
the case: Minkowski space has the Poincar group as its symmetry group and thus we will
have to construct a state which is at least invariant under (discrete) spatial translations. This
forces the n to be translations of 0 and zn = z0 . Moreover, the dependence of the state
on each of the edges has to be the same and therefore the n have to be mutually disjoint.
It follows that the norm of the state is given by
424

2

2
2

"" = |z|
1 1 1, 0 +
1 1, 0 ,
2
(4.1)
where we assumed without loss of generality that "0 " = 1 and we used the diffeomorphism invariance of the measure and 1 is the normalized constant state. By the Schwarz
inequality the first term is non-negative and convergent only if 0 = 1 while the second is
non-negative and convergent only if 1, 0 = 0. Thus the norm diverges unless z = 0.

This caveat is the source of its removal: we notice that the formal state := n n
really depends on an infinite graph and has unit norm if we formally compute it by

limN " N
n=N n " = 1 using disjointness of the n . The only problem is that this
state is not any longer in our Hilbert space, it is not the Cauchy limit of any state in the

2|NM| so that
Hilbert space: defining N := N
n=N n we find |N , M | = |1, 0 |
N is not a Cauchy sequence unless 0 = 1. However, it turns out that it belongs to the
Infinite Tensor Product (ITP) extension of the Hilbert space.
To construct this much larger Hilbert space we must first describe the class of embedded
graphs that we want to consider. We will consider graphs of the category where now
stands for countably infinite. More precisely, an element of is the union of a countably
infinite number of analytic, mutually disjoint (except possibly for their endpoints) curves
called edges of compact or non-compact range which have no accumulation points of
edges or vertices. In other words, the restriction of the graph to any compact subset of
the hypersurface looks like an element of 0 . These are precisely the kinds of graphs that
one would consider in the thermodynamic limit of lattice gauge theories and are therefore
best suited for our semi-classical considerations since it will be on such graphs that one
can write actions, Hamiltonians and the like.
The construction of the ITP of Hilbert spaces is due to von Neumann [20] and already
more than sixty years old. We will try to outline briefly some of the mathematical notions
involved (see the fifth Ref. in [3] for a concise summary of the most important definitions
and theorems).
Let for the time being I be any index set whose cardinality |I | = takes values in the set
of non-standard numbers (Cantors alephs). Suppose that for each e I we have a Hilbert
space He with scalar product ., .e and norm "."e . For complex numbers ze we say that

eI ze converges to the number z provided that for each positive number > 0 there
exists a finite set I0 () I such that for any other finite J with I0 () J I it holds

that | eJ ze z| < . We say that eI ze is quasi-convergent if eI |ze | converges.

If eI ze is quasi-convergent but not convergent we define eI ze := 0. Next we say
that for fe He the ITP f := e fe is a C0 vector (and f = (fe ) a C0 sequence) if

" f " := eI "fe "e converges to a non-vanishing number. Two C0 sequences f, f are
said to be strongly resp. weakly equivalent provided that

fe , f e 1 resp.
fe , f e 1
(4.2)
e
e
e
converges. The strong and weak equivalence class of f is denoted by [f ] and (f ),

respectively, and the set of strong and weak equivalence classes by S and W, respectively.
We define the ITP Hilbert space H := e He to be the closed linear span of all C0 vectors.
425
Likewise we define H[f

] or H(f ) to be the closed linear spans of only those C0 vectors
which lie in the same strong or weak equivalence class as f . The importance of these
notions is that the determine much of the structure of H , namely:
(1)
(2)
(3)
(4)
(5)
All the H[f

] are isomorphic and mutually orthogonal.

Every H(f ) is the closed direct sum of all the H[f
] with [f ] S (f ).
The ITP H is the closed direct sum of all the H(f

) with (f ) W.
Every H[f ] has an explicitly known orthonormal von Neumann basis.

If s, s are two different strong equivalence classes in the same weak one then there
exists a unitary operator on H that maps Hs to Hs , otherwise such an operator
does not exist, the two Hilbert spaces are unitarily inequivalent subspaces of H .
Notice that two isomorphic Hilbert spaces can always be mapped into each other such that
scalar products are preserved (just map some orthonormal bases) but here the question is
whether this map can be extended unitarily to all of H . Intuitively then, strong classes
within the same weak classes describe the same physics, those in different weak classes
describe different physics such as an infinite difference in energy, magnetization, volume,
etc. See [21] and references therein as well as Appendix C for illustrative examples.
Next, given a bounded operator ae on He (notice that closed unbounded operators have a
polar decomposition into an unitary and a self-adjoint piece and that a self-adjoint operator
is completely determined by its bounded spectral projections so that restriction to bounded
operators is no loss of generality) we can extend it in the natural way to H by defining
a e densely on C0 vectors through a e f = f with fe = fe for e = e and fe = ae fe . It
turns out that the algebra of these extended operators for a given label is automatically a
von Neumann algebra for H and we will call the weak closure of all these algebras the
von Neumann algebra R of local operators.
Given these notions, the strong equivalence class Hilbert spaces can be characterized
further as follows. First of all, for each s S one can find a representant s s such
that " s " = 1. Moreover, one can show that Hs is the closed linear span of those C0
vectors f such that fe = es for all but finitely many e. In other words, the strong
equivalence class Hilbert spaces are irreducible subspaces for R , s is a cyclic vector
for Hs on which the local operators annihilate and create local excitations and thus, if I
is countable, Hs is actually separable. We see that we make naturally contact with Fock
space structures, von Neumann algebras and their factor type classification [22] (modular
theory) and algebraic quantum field theory [23]. The algebra of operators on the ITP which
are not local do not have an immediate interpretation but it is challenging that they map
between different weak equivalence classes and thus change the physics in a drastic way.
It is speculative that this might in fact enable us to incorporate dynamical topology change
in canonical quantum gravity.
A number of warnings are in order:
(1) Scalar multiplication is not multi-linear! That is, if f and z f are C0 sequences

where (z f )e = ze fe for some complex numbers ze then f = ( e ze ) f is in

general wrong, it is true if and only if e ze converges.
426
(2) Unrestricted use of the associative law of tensor products is false! Let us subdivide

the index set I into mutually disjoint index sets I = I where runs over
some other index set A. One can now form the different ITP H = H , H =
eI He . Unless the index set A is finite, a generic C0 vector of H is orthogonal
to all of H . This fact has implications for quantum gravity which we outline below.
After this mathematical digression we now come back to canonical quantum general
relativity. In applying the above concepts we arrive at the following surprises:
(i) First of all, we fix an element and choose the countably infinite index set
I := E( ), the edge set of . If |E( )| is finite then the ITP Hilbert space H :=

AL of the Ashtekar
eE( ) He is naturally isomorphic with the subspace H
Lewandowski Hilbert space obtained as the closed linear span of cylinder functions
over . However, if |E( )| is truly infinite then a generic C0 vector of H is
orthogonal to any possible HAL

, 0 . Thus, even if we fix only one ,
the total HAL is orthogonal to almost every element of H .
(ii) Does H have a measure theoretic interpretation as an L2 space ? By the Kolmogorov theorem [24] the infinite product of probability measures is well defined and thus one is tempted to identify H = e L2 (SU(2), dH ) with HAL :=
L2 (e SU(2), e dH ). However, this cannot be the case, the ITP Hilbert space is
non-separable (as soon as dim(He ) > 1 for almost all e and |E( )| = ) while the
latter Hilbert space is separable, in fact, it is the subspace of HAL consisting of the
closed linear span of cylindrical functions over with 0 E( ).
(iii) Yet, there is a relation between H and HAL through a construction called the inductive limit of Hilbert spaces: we can find a directed sequence of elements n
0 E( ), that is, m n for m n, such that is its limit in . The sub HAL are isometric isomorphic with the subspaces of H given by
spaces HAL
n
the closed linear span of vectors of the form n [eE( n ) 1] where n
Hn which provides the necessary isometric monomorphism to display H
HAL
n
as the inductive limit of the HAL
.
n
(iv) So far we have looked only at a specific . We now construct the total Hilbert
space

H
H :=
(4.3)

equipped with the natural scalar product derived in the fifth reference of [3]. This
is to be compared with the AshtekarLewandowski Hilbert space

HAL =
HAL .
HAL :=
(4.4)
0
The identity in the last line enables us to specify the precise sense in which HAL
H : for any the space HAL is isometric isomorphic as specified in (iii)
with the strong equivalence class Hilbert subspace H,[1] where 1e = 1 is the constant function equal to one. Thus, the the AshtekarLewandowski Hilbert space
427
describes the local excitations of the AshtekarLewandowski vacuum AL with

eAL = 1 for any possible analytic path e.
Notice that both Hilbert spaces are non-separable, but there are two sources of nonseparability: the AshtekarLewandowski Hilbert space is non-separable because
0 has uncountable infinite cardinality. This is also true for the ITP Hilbert since
0 but it has an additional character of non-separability: even for fixed
with an infinite number of edges the Hilbert space H splits into an uncountably
infinite number of mutually orthogonal strong equivalence class Hilbert spaces and
HAL is only one of them.
(v) Recall that spin-network states [8] form a basis for HAL . The result of (iv) states
that they are no longer a basis for the ITP. The spin-network basis is in fact the
von Neumann basis for the strong equivalence class Hilbert space determined by
[ AL] but for the others we need uncountably infinitely many other bases, even for
fixed . The technical reason for this is that, as remarked above, the unrestricted
associativity law fails on the ITP.
We would now like to justify this huge blow up of the original AshtekarLewandowski
Hilbert space from the point of view of physics. Clearly, there is a blow up only when the
initial data hypersurface is non-compact as otherwise 0 = .
(a) Let us fix a three manifold and a graph in order to describe semi-classical
physics on that graph as outlined in Section 2. Given a classical initial data set m we
can construct a coherent state ,,X,m which in fact is a C0 vector fm for H

of unit norm. This coherent state can be considered as a vacuum or background
state for quantum field theory on the associated spacetime. As remarked above,
the corresponding strong equivalence class Hilbert space H,[fm ] is obtained by
acting on the vacuum by local operators (where local means that only finitely
many edges are affected), resulting in a space isomorphic with the familiar Fock
spaces and which is separable. In this sense, the fact that H is non-separable,
being an uncountably infinite direct sum of strong equivalence class Hilbert spaces,
simply accounts for the fact that in quantum gravity all vacua have to be considered
simultaneously, there is no distinguished vauum as we otherwise would introduce a
background dependence into the theory.
(b) The Fock space structure of the strong equivalence classes immediately suggests
to try to identify suitable excitations of ,m as graviton states propagating on a
spacetime fluctuating around the classical background determined by m [25].
Also, it is easy to check whether for different solutions of Einsteins equations the
associated strong equivalence classes lie in different weak classes and are thus physically different. For instance, preliminary investigations indicate that Schwarzschild
black hole spacetimes with different masses lie in the same weak class. Thus, unitary
black hole evaporation and formation seems not to be excluded from the outset.
(c) From the point of view of HAL the Minkowski coherent state is an everywhere excited state like a thermal state, the strong classes [ AL ] and [fm ] for Minkowski
data m are orthogonal and lie in different weak classes. The state AL has no obvi-
428
ous semi-classical interpretation in terms of coherent states for any classical spacetime.
(d) In the fourth Ref. of [2] the GNS construction is employed for one special background state and one special graph in order to arrive at a state describing Minkowski
space. Unfortunately, the semi-classical properties of this state were not explored
there and it remains unclear if and in which sense this state really describes an approximation to a Minkowski background state. However, we want to describe the
connection of the resulting representation with our ITP Hilbert space:
The background state employed there is actually not a state of the Ashtekar
Lewandowski Hilbert space and thus what one can do is to consider it as an expectation value functional on a subalgebra of local operators in the bulk of a non-compact
three manifold providing a new Hilbert space through the GNS construction. On the
other hand, this background state can be considered as a true vector state of our ITP
Hilbert space and the GNS Hilbert space constructed there is isomorphic with the
strong equivalence class Hilbert space based on the given background state. This
shows that the GNS construction is contained as a special case in our ITP construction.
But the ITP construction provides powerful additional structure: it organizes all
these Hilbert spaces, even for arbitrary graphs, into one big Hilbert space, establishes the relation between them and gives concrete mathematical criteria for when
two such GNS Hilbert spaces are physically equivalent or not. In other words, it
equips us with a much more general, universal, enveloping framework that is absent
from the individual GNS Hilbert spaces and is a rigorous framework that allows us
to take the infinite volume (thermodynamic) limit of the theory.
Finally, clearly the ITP construction fits neatly together with out coherent state construction via formulas (2.22), (3.11), thus providing a powerful, general starting
point for the semi-classical analysis in spatially compact or non-compact situations.
We do not have to start constructing state on a given graph again and again as we
vary the point in the classical phase space which it should approximate, rather, we
have a coherent state machine which does this job for us with all the desirable semiclassical properties already established.
(e) While non-local operators, such as Hamiltonian constraint operators smeared
against test fields of rapid decrease rather than of compact support (as required in order to discuss supertranslations), matter Hamiltonians coupled to gravity (important
in order to obtain effective matter Hamiltonians of the standard model in a gravitational state fluctuating around Minkowski space), electromagnetic charges and
Poincar charges (in fact, all Dirac observables of general relativity are non-local),
etc., can by definition only be treated through a limiting procedure within the GNS
construction, they immediately belong to the algebra of operators on the ITP Hilbert
space.
The graviton annihilation and creation operators labelled by a specific momentum
(and helicity) mode obviously also belong to this class of non-local operators. Once
we have identified suitable graviton Fock states propagating on a fluctuating quan-
429
tum metric as defined through the excitations of a unit norm representant of the
corresponding strong equivalence class, we should be able, in principle to make
contact with perturbative quantum field theory and string theory and thus to deliver
significant evidence that loop quantum gravity is not just some completely spurious
sector of quantum gravity.
5. Applications and outlook

Given the tools provided in [3,5] one can start investigating many fascinating physical
problems starting from non-perturbative quantum general relativity. We will list a few of
them.
(i) Generalization to Higher Ranks
In [3] peakedness and Ehrenfest theorems were analytically proved for rank one
gauge groups and a generalization of the proof to higher ranks was sketched. It
would be important to fill in the details at least for G = SU(3), if not for arbitrary
compact gauge groups.
(ii) Quantum Field Theory on Curved Spacetimes
As already mentioned in the main text, choosing a coherent state peaked on the
initial values of a solution to Einsteins equations one obtains a cyclic state for a
Fock-like representation. It would be important to isolate suitable graviton states in
such a representation in order to make contact with perturbative quantum gravity
and string theory. The same can be done when coupling matter and one could, for
instance, study photons propagating on fluctuating quantum spacetimes.
(iii) Renormalization Group and Diffeomorphism Group
Given a Fock like structure one can try to set up an analog of the conventional
renormalization techniques. One expects that from non-perturbative quantum
gravity no UV singularities arise (see [11,12]) for hint in that direction). But then
one would like to see how they get absorbed as compared to the conventional
perturbative approaches and the obvious guess is that it is the diffeomorphism
invariance of the theory that lets us get rid of them (since there is no background
metric, one cannot tell from a fundamental point of view whether a momentum
gets large or small). It should be possible to nail down the precise mechanism for
UV-finiteness with our tools.
(iv) Avoidance of Classical Singularity Theorems
One can plug into the label of a coherent state a solution m( ) of Einsteins
equations (in some gauge) which becomes singular after some time lapse 0 . It
would be challenging to see whether the classically singular function remains finite
in the sense of expectation values of the corresponding operator at = 0 . The
recent regularity findings obtained within loop quantum cosmology [26] which uses
the techniques developed in [11,12] are an indication that this actually could be the
case in the full theory as well.
430
(v) Corrections to the Standard Model

One of the major motivations for the construction of coherent states is, of course,
the question whether our non-perturbative quantum theory has the correct classical
limit. But beyond that one can start computing corrections to the standard model,
a development that has just started. See for instance the two first references in [2]
discussing Poincar invariance violating effects that are not yet ruled out to lie in
the detectable regime. In order to improve these calculations it would be desirable
to have a fast diagonalization computer code for the volume operator [4,27] since it
plays an important role in the quantization of the Hamiltonian (constraints) [11,12].
Acknowledgements
We thank Alexandro Corichi, Rodolfo Gambini, Brian Hall, Bei-Lok Hu, Ted Jacobson,
Hugo Morales-Tecotl and, especially, Luca Bombelli and Jurek Lewandowski as well as
the members of the quantum gravity division of the Center for Gravitational Physics and
Geometry Martin Bojowald, Olaf Dreyer, Stephen Fairhurst, Amit Gosh, Badri Krishnan,
Jacek Wisniewski and, especially, Abhay Ashtekar and Jorge Pullin for very productive
discussions about semiclassical states, which strongly clarified our understanding of the
appearance of different scales in quantum gravity, the difference between coherent and
semi-classical states and the difference between kinematical and dynamical semi-classical
states. Their inputs and criticisms influenced our point of view and lead us to our
construction of approximate operators and our fluctuation optimization procedure.
We also thank Brian Hall for clarifying discussions about infinite tensor products.
H. Sahlmann and O. Winkler thank the Studienstiftung des Deutschen Volkes for
financial support.
Appendix A. Construction of approximate operators
In this appendix we construct an approximate area operator for a specific class of
surfaces.
Suppose we are given some embedded graph = X( ) (here given as the image
under
the embedding X of a model graph ) and look at the area function AS (m) :=

Y 1 (S) d y det(Y q) of an embedded surface S (here given as the image under the
embedding Y of a model two surface S into ) where q denotes the three metric
determined by m. We choose a polyhedronal decomposition P of dual to and a
corresponding system of paths inside its faces. Our first task is to write down a function
of the correspeonding graph degrees of freedom ge = (he , Pje ) that approximates AS . As
compared to the construction in the main text we choose here for simplicity X0 = X,
,S = id .
Without specifying exactly the topology of , S it is difficult to write down an explicit
formula, so let us assume for simplicity that has the topology of a (possibly infinite) cubic
can be labelled by a direction index I = 1, 2, 3 and
lattice. In this case model edges eI (v)
a vertex index v Z3 and we think of them as given by unit intervals along the coordinate
431
axes of R3 . Furthermore, we take as P the image under X of the faces S I (v) (dual to eI (v)
of the model graph given by the translation of by the vector (1/2, 1/2, 1/2) in R3 and for
we take the image under X of the natural radial paths in each of the faces. Likewise,
for definiteness, we assume that S has the topology of a square so that u, v range over the
interval (N, N) where N is an integer. The embedding X : R3 ; t' X(t') provides a
(local) coordinate chart so that we can think of Y as a composition Y a (u, v) = Xa (t'(u, v)).
The coordinates t', u, v are taken to be dimensionless. We can then write the area function
explicitly as

2
a
AS (q) =
(A.1)
du dv
Ej (X(t' ))nIa (t' ) t'=t'(u,v)nI (u, v) ,
[N,N]2
b (t' )X c (t' )
nIa (t' ) = 12 abc IJK X,J
,J
J (u, v)t K (u, v).

where
and nI (u, v) = IJK t,u
,v
Let us now partition S into the maximal, connected, open pieces Sk which lie completely
3
within a coordinate
cube of R , let vk be the vertex of that cube, choose anyI(uk , vk )
Sk and let k := Sk du dv. Then, given any point m M we denote by Pj (v, m) :=
e (v)
Pj I (m) and nI (k) := nI (uk , vk ). With these choices we arrive at the gauge invariant
approximate function

2
AS, (m) := a 2
(A.2)
k PjI (vk , m)nI (k) ,
k
where vk = X(vk ) which now depends explicitly on our choices (collectively denoted by )
as well as the connection and so is a function of m and not only of qab . By construction,
for fixed S (A.2) is a Riemann sum for the integral (A.1) and thus converges to it pointwise
in M in the limit that the sum over k involves infinitely many terms (the model lattice
becomes finer and finer within the model surface). Notice that the coefficients k , nI (k)
and the sum over the Sk do not depend on X or m and thus (A.2) defines a well-defined
function on M. Moreover, (A.2) is diffeomorphism covariant since the only dependence
on X rests in the P I (v, m) while the k , nI (k), Sk are diffeomorphism invariant. Eq. (A.2)
thus serves as a suitable departure point for quantization.
According to our general programme we define now the substitute operator

2
k PjI (vk )nI (k) ,
A S, = a 2
(A.3)
k
which is to be thought of as an operator on H since it is already consistently defined

on every subgraph of . By doing a similar construction for an arbitrary graph and by
enforcing cylindrical consistency we in principle (we do not know it explicitly but must
construct it graph by graph) arrive at a continuum operator A S .
Let us now consider in a specific example for , S how the expectation values are
computed. Consider the embeddings into := R3
Xa (t') := Ia t I ,
Y a (u, v) = (u, v, u),
t'(u, v) = (u, v, u),
(A.4)
where 0 1 is a fixed parameter and u, v [N, N]. Eq. (A.4) defines a plane parallel
to the t 2 direction and which is tilted against the t 1 direction by an angle defined by
432
tan() = (for /4 interchange the directions of t 1 , t 2 ). We will not consider the most
general case for which it is difficult to write down an explicit formula but rather consider
the case that = 1/M where M is an integer such that N/M is also a natural number.
In this case we are looking at the vertices v(n
1 , n2 ) = (nI ) = (n1 , n2 , n3 (n1 )), n1 , n2 =
N, N + 1, . . . , N 1 where n3 (n1 ) = N + [(n1 + N) ] (Gauss bracket). The Sk
are simply model unit squares so that k = 1. We easily compute (nI (k)) = (, 0, 1) so
that (A.3) becomes

2
A S, = a 2
(A.5)
Pj3 (v(n1 , n2 )) Pj1 (v(n1 , n2 )) ,
n1 ,n2
where v(n1 , n2 ) = X(v(n

1 , n2 )).
This expression is to be compared with the projection onto H of the operator of [4]
!

2

1
3,up

A S, =
Jj Jj3,down X(v(n
1 , n2 ))
2 n
n1 =N+kM
2
"

3,up
3,right2
3,down
1,left
+
Jj Jj
X(v(n
1 , n2 )) (A.6)
+ Jj
Jj
,
n1 =N+kM
where Jje = i2p /2tr((j he )T /(he )) and J I, (v)j = JjeI (v) where eI (v) is the segment
of eI (v) outgoing from the intersection of eI (v) with S into the direction =
A simple calculation reveals that for e =

up, down, left, right and eI (v) = X(eI (v)).
(edown)1 eup or e = (eleft )1 eright we have JjI, (v)f (he ) = a 2 P I (v)j f (he ) if
= up, right and J I, (v)f (he ) = a 2 Oj k (he )P I (v)k f (he ) if = down, left where the
orthogonal matrix Oj k (h) is defined by Adh (j ) = Oj k (h)k . It follows that
!

2

2
a

A S, =
j k + Oj k (he3 (v(n1,n2 )) ) Pk3 (v(n1 , n2 ))
2 n
n1 =N+kM
2
#

j k + Oj k (he3 (v(n1 ,n2 )) )
+
n1 =N+kM
"

1
2 $1/2
3
. (A.7)
Pk v(n1 , n2 ) j k + Oj k (he1 (v(n1 ,n2 )) ) Pk (v(n1 , n2 ))
j
Let us test these expressions in a coherent state for flat data m = (Aa = 0, Eja = ja ).
Notice that then the parameter is really the edge length determined by m. Then heI (v) =
1, PjI (v) = jI 2 and the Ehrenfest theorems proved in [3] reveal that the zeroth order in h
of the expectation values is given by
%
lim ,m , A S, ,m = (2N)2 1 + 2 ,
t 0

lim ,m , A S, ,m = (2N)2 1 + ( 2 1) .
(A.8)
t 0
The value in the first line of (A.8) is in fact the exact classical area of the surface while
the second line is off (equal or bigger than) that value. This is the staircase problem. We
433
get coincident results only for = 0, 1 corresponding to M = , 0. The relative mistake

is zero for M = 1, then reaches its maximum of 8% at M = 2 and then monotonously
decreases as 1/M to zero. An error of eight percent is inacceptible which is why we are
forced to use A S instead of A S which is guaranteed to give results of sufficiently high
accuracy. Notice that the dual faces of the graph under consideration in this example are
not at all tangential to the surface S and still we get a good (even exact) result.
Yet, this example was special in the sense that the coefficients k , nI (k) were
independent of k so that one would have gotten the same result even if there was only one
cube needed (replace N by 1 and by N) which is due to the fact that in this case (A.2)
gives the exact classical value. If the surface would not be a plane, then the coefficients
would vary and (A.2) no longer gives the exact classical value. If the surface S wiggles at
a scale much lower than then (A.2) will be a very bad approximation. In that case we can
improve the result by increasing the number of edges of the lattice and thus this error is a
finite size effect. Next, in replacing (A.1) by (A.2) it was crucial that m does not vary too
much on an embedded edge of the graph or the dual embedded face. An error due to this
is related to the curvature of m and will be called a curvature effect.
In what follows we will exclude finite size effects by the assumption that the surfaces
that we are interested in are wiggling at most at the scale of the graph. More precisely, the
Sk = X(Sk ) are allowed not to lie at all inside the dual faces of the embedded graph but
the tangent space should be approximately constant over Sk . For instance, we allow for a
surface which is parallel to the t 2 direction as above but such that jumps between 1
1 3
from cube
to cube (its projection into the t , t plane is a zig-zag curve with triangle edge
length 2). However, we do not allow to jump within one cube. We are then left with
curvature effects to which we turn in the next section.
Appendix B. Minimization of fluctuations through optimization of scales

In this appendix we will sketch the computation of fluctuations in our scheme leading to
a fixing of the free parameters a, through optimization. We will assume that the scale L
is finite.
Given the fact that we are in a gauge field theory context, it is motivated to consider
as an elementary set of classical continuum observables the non-Abelian analogues of the
electric and magnetic fluxes through embedded surfaces S. At this point we could say
that we measure only the quantum fluxes associated with the dual faces of a given graph
since this information is enough in order to reconstruct any other classical observable in
the fashion outlined in the previous Appendix A. In that case and after normal ordering
there would be no classical or normal ordering error at all. However, one would like to be
more general and consider fluxes through arbitrary surfaces and not only elementary ones
because one would like to measure these more general observables directly at the quantum
level and not only through an indirect procedure using a classical formula. Actually, if the
Gauss law a Eja = 0 holds and due to the Bianchi identity, we can replace S by any surface
S built entirely from dual faces of the polyhedronal decomposition such that S S = R
434
is the boundary of a region R (possibly up to some negligible piece at the boundary of S)

without changing the classical flux functions. Thus, for the fluxes (on the constraint surface
of the Gauss constraint) we could actually keep the discussion at the level of surfaces built
from elementary faces which would simplify the discussion, however, we will not do that
for the sake of generality and in view of applications to other sets of basic variables which
are not fluxes and for which an off-shell approximation might be desirable.
In order to keep the presentation at an elementary level we will replace SU(2) by U (1)3
as otherwise the non-Abelianess blows up the effort to obtain all the estimates by an order
of magnitude (see especially the fourth Ref. in [3]) without changing the end result. We are
thus concerned with the following classical (kinematical) observables

Bj (S) = Fj (x) and Ej (S) = (Ej )(x),
(B.1)
S
Aj .
where Fj is the curvature two-form of

Notice that Fj , Ej are gauge invariant. In the
non-Abelian case one would need to choose a point p inside S and a system of paths
between p and any other point x S and integrate instead of Ej (x), Fj (x) their image
under the adjoint action of SU(2) evaluated at the holonomies along those paths in order
to obtain a gauge-covariant result. The non-Abelian case will be discussed in more detail
elsewhere.
Using the same notation as in the previous section we have a model graph of cubic
topology and a model surface S of square topology which are embedded into via
embeddings X and Y = X t' respectively so that

Bj (S) = du dv nIa (t' )Bja (X(t' )) t'=t'(u,v)nI (u, v)
(B.2)
S
j
and similar for Ej (S) where Bja = 12 abc Fbc . Using the same dual polyhedronal
decomposition and choices of paths as in the previous section we arrive at the classical
substitute functions
Bj, (S) =

k
Ej, (S) = a 2
h I (v ) (h I (v ) )1
j
k nI (uk , vk )
2i
k nI (uk , vk )PjI (vk ),
,
(B.3)
where I (v) = X( I (v))

is the image under X of the model plaquette loop I ((v)) =
eK (v + bJ )1 eJ (v + bK )1 eK (v)
1 on in the J , K plane with I J K = 1
eJ (v)
1
and with vertex at v = X (v). Here, bI denotes the standard orthonormal basis in R3 .
Notice that the path system is actually not needed in order to define PjI (v) since U (1)3
is Abelian. Due to the relation
j
gI (v) = e
PjI (v) j
heI (v)
(B.4)
it is possible to replace (B.3) by functions which are linear combinations of products of

the gI (v), (gI (v))1 , gI (v), (gI (v))1 . These functions are the classical analogues of
j
435
the creation operators gI (v), (gI (v))1 and annihilation operators (gI (v)) , ((gI (v))1 )
of Section 3 and with the help of these functions it becomes possible to normal order
j
the corresponding operators in the quantum theory. For example, gI (v) + (gI (v))1 ) =
hI (v)(1 + O(( a )4 )) and gI (v)gI (v) (gI (v))1 (gI (v))1 = 4PjI (v)(1 + O(( a )4 )). By
using (finitely many) higher powers of these four functions one can suppress the subleading
terms to any desired order in (/a).
We can then quantize (B.3) as
j
B j, (S) :=
E j, (S) = a
k nI (uk , vk ):
k

2
j
j
h I (v ) (h I (v ) )1
k
2i
:,
k nI (uk , vk ) :PjI (vk ):,
(B.5)
where the normal ordering symbol is to be understood after substituting the just mentioned
approximations up to (/a)2n for some desired n. There is then no normal ordering error.
Next, we are concerned with the classical error given by

a2
k nI (uk , vk ) :P I (vk ): P I (vk )
Ej (S) Ej (S)
class
k

I

+ Ej (S)
k nI (uk , vk )Ej S (vk ) ,
(B.6)
where S I (v) is the image under X of the surface S I (v)

dual to eI (v)
and similar for
(Bj (S))class . The normal ordering symbols in (B.6) just mean that one shoul replace
the functions inside the normal ordering symbols by the above mentioned functions of
j
the gI (v). Let us estimate these terms. If we denote by AS (E) the classical area of S

determined by E then the number of terms in the sum k will be of the order of AS (E)/ 2
j
j
and so the first term can be bounded by AS (E)(E S )n (/a)2(n1) where E S is the maximum

of Ej (S )/AE (S ) as we vary S S. For the second term we employ the EulerMacLaurin
estimates [28] for the numerical difference between an integral and a Riemann sum. We
will use only the coarsest estimate resulting from the identity

y+N

N1
F (y) + F (y + N)
+
dx F (x)
F (y + k)
2
y
3
=
2
1
dt 2 (t)
0
N1

k=1

y + (m + t)
(B.7)
k=0
where 2 (t) = t 2 t is the Bernoulli polynomial of the second degree and F is a twice
differentiable function on the interval [y, y + N)]. Using 2 (t)
1/4 for 0 t 1 the
y+N
2
right-hand side of (B.7) can be bounded from above by /8 y
dx|F (x)| and thus
involves the second derivative of F . Let us introduce the curvature scale

1

2 0 dt 2 (t) N1

1
k=0 F (y + (m + t))

(B.8)
:=
y+N
.
2
LF
dx F (x)
y
436
Using this formula and iterating it since we are dealing with a two-dimensional integral we
find that the second term can be bounded by |Ej (S)| 4 /L4E where LE is a lower bound on
the electric curvature radius of the phase space point m and thus independent of j , S.
We will now make the assumption that a, LE and that n is sufficiently high so that
the first term in (B.6) is sub-leading (since we will relate a to LE later on, this is a selfconsistency assumption). Then we obtain (estimates are in orders, so we neglect numerical
factors of order one)
Ej (S)

class
4
L4E
and
Bj (S)

class
4
,
L4B
(B.9)
where the magnetic curvature radius will be of the same order as LE since both types of
curvature come from the same four-dimensional curvature tensor.
Finally, we are left with the quantum mechanical error. Notice that the operators (B.5)

are of the form O = k O k where the O k mutually commute for the electric flux while
for the magnetic flux the next neighbour terms are non-commuting. However, the number
of these non-commuting terms per vertex is of the same order as the number of the O k
and since we are just concerned with orders of magnitude here their contribution to the
following estimates gives just a multiplicative numerical factor of order one which we do
not display. We can thus assume that all the O k are mutually commuting and obtain for the
fluctuations in the coherent state ,,X,m

AE (S)
2 (O)(m)2quantum =
O
,
(B.10)
O k2 O k 2 t
2
k
where the Ehrenfest theorems for O k = PjI (v), h I (v) of the third reference in [3] have
been used in the estimate. For the total error we therefore find
j

Ej (S)
Bj (S)
where E S =
(m)2 t
total
2p a 2
a 4 AE (S)
8
8
+
+
,
2
2
8
Ej (S)
2 AS (E)E S L8E
LE
(m)2 t
total
2p L4B
8
AE (S)
8
+ 8,
+
2
2
8
2
2
Bj (S)
a AS (E) LB
LB
(B.11)
Ej (S)2 /AE (S) is of order unity and in the last line we have used the
approximate identity Bj (S) AE (S)/L2B .

In order to fix the parameters , a we notice that by assumption AS (E) 2 (N
=
AS (E)/ 2 is the number of contributing faces and gives the law of large numbers 1/ N
for the quantum mechanical fluctuations). Thus we can further estimate
2p a 2

8
Ej (S) (m)2
+ 8,
4 E S
LE
2p L4B

8
Bj (S) (m)2 2 4 + 8 ,
a
LB
(B.12)
and we require that these two errors be equal (yielding an unquenched uncertainty
relation for the fluxes) and minimal. One can solve E = B analytically for a 2 resulting
437

in a 2 = x 3 + (x 3 )2 + L4B E S where = E S (1/L8B 1/L8E )/(22p ) and x = 4 . The
remaining optimization problem can then only be solved numerically. We will not treat
the most general%case in this paper but make the physical assumption that LE LM
L so that a 2 E S L2 . Notice that provided L our self-consistency assumption
from above is indeed
satisfied. The optimization then yields an absolute minimum at
%
12 = 2p L10 /(2 E S ). Thus, is locked at a geometric mean of the microscopic and
macroscopic scale. We verify that indeed p L as long as p L, finishing up
our self-consistency check. The total fluctuation becomes up to a numerical factor equal to

p 4/3
) where L0 is the phase space bound mentioned in the main text and
( LEp )4/3 ( L E
S
0 S
depends on M but not on m. Locking a at L0 from the outset would lead to qualitatively
similar results.
We finish this section with a couple of remarks:
(1) Our analysis has optimized only an upper bound, it may well be that one can get
even better estimates depending on m.
(2) One may wonder why the total fluctuation of our operator should depend on
a classical error which then leads to an optimization problem in which one
is not used to from electrodynamics. However, there actually the same effect
happens [14]: if one considers in free Maxwell theory the fluctuations of electric
and magnetic flux operators then one simply finds infinity. This is due to the fact
that in the corresponding Fock representation one has to smear the fields over threedimensional regions rather than two-dimensional surfaces in order to arrive at a welldefined operator. Thus, in order to make the fluctuation calculation well-defined one
has to fatten the surfaces by a transversal thickness (which is a regularization
comparable to our replacing continuum functions by discretized objects) and then
divides the fattened flux by . When one now computes the fluctuation of these
operators compared to exact classical flux value one finds a similar competition
between the quantum mechanical fluctuation blowing up as 0 and the classical
error vanishing as 0. The optimization thus takes place since one is interested
in measuring classical obervables quantum mechanically which are smeared by
singular smearing functions. Whether this is a sensible thing to do or not is a physical
question. If one insists to measure fluxes quantum mechanically then one has no
choice and fluxes are precisely the kind of objects that are natural in our non-Fock
representation.
Appendix C. The infinite spin chain

In this final appendix we consider a trivial but instructive physical system in order to
exemplify infinite tensor product concepts, namely an infinite chain of uncoupled spin
degrees of freedom.
Our label set will be the integers I = Z and for each n Z we have the Hilbert space
Hn = C2 with standard inner product fn , fn n = fn+ fn+ + fn fn . In each Hilbert space
we have the standard orthogonal basis of vectors en and spin operators n = 3 (Pauli
438
matrix) so that n en = en corresponds to spin up/down. We also have ladder operators

n = 12 [1 i2 ] so that n en = 0, n en = en . Consider the positive semi-definite,
self-adjoint Hamiltonian

1
n n
[1 + n ] =
H :=
2 n
n
(C.1)
on the ITP Hilbert space H = n Hn which is non-separable even though each Hn has
finite dimension two.
We will first consider a C0 vector f with "fn "n = 1 and a second one f with fn =
fn . Are the correspeonding C0 sequences in the same strong (weak) equivalence class?

Since fn , fn n = 1 we see that n |fn , fn n 1| = n 2 = but n | |fn , fn n |

1| = n 0 = 0, thus they are in different strong classes within the same weak class. In fact,
the unitary operator U on H defined densely on arbitrary C0 vectors by U g = g with
gn = gn maps the two unit C0 vectors into each other and thus the strong equivalence
class Hilbert spaces built from them will be unitarily equivalent subspaces of the whole

ITP. Notice that indeed f , f = n (1)n = 0 since the product of numbers zn = 1
is only quasi-convergent.
Consider now the unit C0 vectors := f with fn := en (the total spin up/down

state). We have n | |fn+ , fn n | 1| = n 1 = , thus they are in different weak
classes. It is true that the non-local operator A := n n+ maps into + , however,
f with
A is not a unitary operator as one easily verifies by computing the norm of A
+
+
fn = 0 for at least one n. In fact, is the ground state of H while is an infinite
energy (infinitely excited) state for H and thus their strong equivalence class Hilbert spaces
describe drastically different physics. It is therefore not to be expected on physical grounds
that these representations should be unitarily equivalent.
Finally, it is easy to see that the orthonormal spin-network C0 vector system {n } :=
n enn with n = is not a basis: for instance the C0 vector with tensor product factors
fn := 1 [en+ + en ] is orthonormal to all of them, a nice example in which one sees that
2
the unrestricted associative law is false.
With the help of simple systems like the infinite spin chain one can study a lot of more
phenomena of the ITP such as the occurrence of von Neumann algebra factor types I , II 1 ,
III s , s (0, 1). The von Neumann algebra of local operators R considered in the main
text is of that type if we restrict it to the strong equivalence class Hilbert spaces defined by
the cyclic vectors

s := n
1+s +
n +
2
1s
n
2

(C.2)
of the ITP Hilbert space H := n [C2 C2 ] for s {1}, {0}, (0, 1), respectively. These
representations can be interpreted as zero, infinite and finite temperature representations,
respectively, see [21] for details.
439
References
[1] A. Ashtekar, Non-Perturbative Canonical Gravity, World Scientific, Singapore, 1991;
A. Ashtekar, in: B. Julia (Ed.), Gravitation and Quantization, Elsevier, Amsterdam, 1995;
C. Rovelli, Loop quantum gravity, Review written for electronic journal Living Reviews,
gr-qc/9710008;
C. Rovelli, Strings, loops and others: a critical survey of the present approaches to quantum
gravity, Plenary lecture given at 15th Int. Conf. on Gen. Rel. and Gravitation (GR15), Pune,
India, Dec. 1621, 1997, gr-qc/9803024;
A. Ashtekar, C. Beetle, O. Dreyer, S. Fairhurst, B. Krishnan, J. Lewandowski, J. Wisniewski,
Phys. Rev. Lett. 85 (2000) 35643567, gr-qc/0006006;
J. Baez, Lect. Notes Phys. 543 (2000) 25.
[2] R. Gambini, J. Pullin, Phys. Rev. D 59 (1999) 124021;
J. Alfaro, H.A. Morales-Tecotl, L.F. Urrutia, Phys. Rev. Lett. 84 (2000) 23182321;
M. Varadarajan, J.-A. Zapata, Class. Quantum Grav. 17 (2000) 4085, gr-qc/0001040;
M. Arnsdorf, Approximating connections in loop quantum gravity, gr-qc/9910084;
M. Arnsdorf, S. Gupta, Nucl. Phys. B 577 (2000) 529;
A. Corichi, A Gaussian weave for kinematical loop quantum gravity, gr-qc/0006067.
[3] T. Thiemann, Symplectic structures and continuum lattice formulations of gauge field theories,
hep-th/0005232;
T. Thiemann, Gauge field theory coherent states (GCS): I. General framework, hep-th/0005233;
T. Thiemann, O. Winkler, Gauge field theory coherent states (GCS): II. Peakedness properties,
hep-th/0005237;
T. Thiemann, O. Winkler, Gauge field theory coherent states (GCS): III. Ehrenfest theorems,
hep-th/0005234;
T. Thiemann, O. Winkler, Gauge field theory coherent states (GCS): IV. Infinite tensor product
and thermodynamic limit, hep-th/0005235.
[4] C. Rovelli, L. Smolin, Nucl. Phys. B 442 (1995) 593;
C. Rovelli, L. Smolin, Nucl. Phys. B 456 (1995) 734, Erratum;
A. Ashtekar, J. Lewandowski, Class. Quantum Grav. 14 (1997) A5581.
[5] A. Ashtekar, L. Bombelli, Statistical Geometry and Semiclassical Quantum Gravity, to appear.
[6] C. Itzykson, J.-M. Drouffe, Statistical Field Theory, Vol. 2, Cambridge Univ. Press, Cambridge,
1997.
[7] A. Ashtekar, C. Rovelli, L. Smolin, Phys. Rev. Lett. 69 (1992) 237.
[8] C. Rovelli, L. Smolin, Phys. Rev. D 52 (1995) 5743;
J. Baez, Adv. Math. 117 (1996) 253.
[9] T. Thiemann, Quantum spin dynamics (QSD): VIII. The classical limit, in preparation.
[10] A. Ashtekar, J. Lewandowski, D. Marolf, J. Mouro, T. Thiemann, J. Math. Phys. 36 (1995)
64566493, gr-qc/9504018.
[11] T. Thiemann, Phys. Lett. B 380 (1996) 257264, gr-qc/960688;
T. Thiemann, Class. Quantum Grav. 15 (1998) 839873, gr-qc/9606089;
T. Thiemann, Class. Quantum Gravity 15 (1998) 12071247, gr-qc/9705017;
T. Thiemann, Class. Quantum Grav. 15 (1998) 14871512, gr-qc/9705021.
[12] C. Di Bartolo, R. Gambini, J. Griego, J. Pullin, Phys. Rev. Lett. 84 (2000) 2314;
C. Di Bartolo, R. Gambini, J. Griego, J. Pullin, Class. Quant. Grav. 17 (2000) 3211;
C. Di Bartolo, R. Gambini, J. Griego, J. Pullin, Class. Quant. Grav. 17 (2000) 3239.
[13] A. Ashtekar, J. Lewandowski, D. Marolf, J. Mouro, T. Thiemann, J. Funct. Analysis 135
(1996) 519551, gr-qc/9412014.
440
[14] A. Ashtekar, L. Bombelli, O. Dreyer, S. Fairhurst, A. Gupta, H. Sahlmann, T. Thiemann,

O. Winkler, Kinematical versus Dynamical Coherent States, in preparation.
[15] H. Sahlmann, T. Thiemann, O. Winkler, Canonical Quantum General Relativity and Algebraic
Graph Theory, in preparation.
[16] N. Biggs, Algebraic Graph Theory, 2nd edn., Cambridge Univ. Press, Cambridge, 1993;
D. Stauffer, A. Aharony, Introduction to Percolation Theory, Taylor and Francis, 2nd edn.,
London, 1994;
D.M. Cvetovic, M. Doob, H. Sachs, Spectra of Graphs, Academic Press, New York, 1979.
[17] B.C. Hall, J. Funct. Analysis 122 (1994) 103;
B.C. Hall, Commun. Math. Phys. 184 (1997) 233250.
[18] A. Ashtekar, C.J. Isham, Class. Quantum Grav. 9 (1992) 1433;
A. Ashtekar, J. Lewandowski, Representation theory of analytic holonomy C O algebras, in:
J. Baez (Ed.), Knots and Quantum Gravity, Oxford Univ. Press, Oxford, 1994;
A. Ashtekar, J. Lewandowski, J. Geo. Physics 17 (1995) 191;
A. Ashtekar, J. Lewandowski, J. Math. Phys. 36 (1995) 2170.
[19] J. Baez, S. Sawin, Functional integration on spaces of connections, q-alg/9507023;
J. Baez, S. Sawin, Diffeomorphism invariant spin-network states, q-alg/9708005;
J. Lewandowski, T. Thiemann, Class. Quantum Grav. 16 (1999) 22992322, gr-qc/9901015.
[20] J. von Neumann, Comp. Math. 6 (1938) 177.
[21] W. Thirring, Lehrbuch der Mathematischen Physik, Vol. 4, Springer-Verlag, Wien, 1994.
[22] A. Connes, Noncommutative Geometry, Academic Press, 1994.
[23] R. Haag, Local Quantum Physics, 2nd edn., Springer-Verlag, Berlin, 1996.
[24] Y. Yamasaki, Measures on Infinite Dimensional Spaces, World Scientific, Singapore, 1985.
[25] H. Sahlmann, T. Thiemann, Quantum Field Theory on Curved Spacetimes from Canonical
Quantum General Relativity: I. Scalar Fields, II. Photons and the Ray Burst Effect, III.
Gravitons, to appear.
[26] M. Bojowald, H. Kastrup, Class. Quantum Grav. 17 (2000) 3009, hep-th/9907042;
M. Bojowald, H. Kastrup, Class. Quantum Grav. 17 (2000) 3044, hep-th/9907043;
M. Bojowald, J. Math. Phys. 41 (2000) 4313, hep-th/9908170;
M. Bojowald, Class. Quantum Grav. 17 (2000) 1489, gr-qc/9910103;
M. Bojowald, Class. Quantum Grav. 17 (2000) 1509, gr-qc/9910104;
M. Bojowald, gr-qc/0008052;
M. Bojowald, gr-qc/0008053;
M. Bojowald, Absence of singularity in loop quantum cosmology, gr-qc/0102069.
[27] A. Ashtekar, J. Lewandowski, Adv. Theor. Math. Phys. 1 (1997) 388.
[28] E.T. Whittaker, G.N. Watson, A Course of Modern Analysis, Cambridge Univ. Press,
Cambridge, 1969, Chapter VII;
H. Jeffreys, B.S. Jeffreys, Methods of Mathematical Physics, Cambridge Univ. Press, Cambridge, 1992.

Electrically charged topological solitons

J.F. Gomes, E.P. Gueuvoghlanian, G.M. Sotkov, A.H. Zimerman
Instituto de Fsica Terica IFT/UNESP, Rua Pamplona 145, 01405-900, So Paulo, SP, Brazil
Received 5 January 2001; accepted 24 April 2001
Abstract
Two new families of T-dual integrable models of dyonic type are constructed. They represent
(1)
specific An singular non-abelian affine Toda models having U (1) global symmetry. Their 1-soliton
spectrum contains both neutral and U (1)-charged topological solitons sharing the main properties
of 4-dimensional YangMillsHiggs monopoles and dyons. The semiclassical quantization of these
solutions as well as the exact counterterms and the coupling constant renormalization are studied.
2001 Elsevier Science B.V. All rights reserved.
PACS: 02.20.Tw; 11.25.Hf; 11.27.+d; 11.10.Kk; 11.15.Kc; 11.25.-w
1. Introduction
In search for new nonperturbative methods of quantization of four-dimensional
(nonsupersymmetric) SU(n + 1) QCD (and its bosonic part the YangMillsHiggs
(YMH) model), the relationship between 4-D selfdual YangMills (SDYM) theories and
certain 2-D integrable models (IMs) [1] deserves special attention. As it is well known
the dimensional (and symmetries) reduction of 4-D SDYM to lower-dimensional (D =
1, 2, 3) IMs (see [2] for a review) provides effective methods for the construction of a large
class of exact (classical) solutions of the 4-D gauge theory, including spherical symmetric
monopoles [3,4], instantons [5], domain walls (DW) [6,7], etc. Among them the solitons
of relativistic 2-D IMs are of particular interest:
They are expected to describe (at least in large n limit) DWs solutions of 4-D
SU(n + 1) YMH theory.
Their exact quantization is known to be related to the centerless affine quantum group
(1)
(2)
(2)
Uq (An ) (or Uq (A2n ), Uq (A2n1 )) [8].
E-mail addresses: jfg@ift.unesp.br (J.F. Gomes), gueuvogh@ift.unesp.br (E.P. Gueuvoghlanian),
sotkov@ift.unesp.br (G.M. Sotkov), zimerman@ift.unesp.br (A.H. Zimerman).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 2 1 1 - 5
442
J.F. Gomes et al. / Nuclear Physics B 606 (2001) 441482
Whether Uq (A(1)
n ) (or its dual) takes place in the description of the quantum DWs and
the other strong coupling solutions of SU(n + 1) YMH theory monopoles, dyons and
strings is an open problem. There exists however a hint in favour of such conjectute,
namely: the space of classical solutions of 4-D SU(n + 1) SDYM (and BPS solutions of
+ 1) loop algebra as dynamic symmetry [9], which coincides
YMH) has the affine SU(n
(1)
with the classical limit q 1(h 0) of the quantum soliton symmetry group Uq (An )
(see [8] and references therein).
The domain walls that appear in the string (i.e., D-branes) description of 4-D gauge
theories [6,7] manifest properties a bit more involved than the simplest topological solitons
of say, sine-Gordon or abelian affine Toda models. For instance, they may carry certain
U (1) charges or require nonmaximal breaking of SU(n + 1) gauge group, say SU(k)
U (1)nk+1 (i.e., k-coincident D-branes) [7]. The question arises whether one can find 2-D
IMs whose soliton solutions possess all DWs characteristics. The problem we address
in the present paper is to construct such dyonic integrable models 1 admitting U (1)charged topological solitons and CPT-violating term. We shall study the solitonic spectrum
(classical and quantum) of the following new family of (CPT-noninvariant) A(1)
n -dyonic
IMs (n > 1)
n1
1
1
k + e
Va ,
La =
ik i
2
i,k=1

n1

m2 (i1 +i+1 2i )
(1 +n1 )
2
1
Va = 2
1 + e
n ,
e
+e
(1.1)
i=1
1 , =
where ik = 2i,k i,k1 i,k+1 , i, k = 1, . . . , n 1, = 1 + 2 n+1
0
2n e
2
n = 0, = t + x , = t x and = 2/K. The IMs (1.1) are invariant under
global U (1) transformation,
= eia ,
2
= eia ,
2
i = i ,
where a is a real constant. They represent a specific mixture of the LundRegge model
(1)
interacting with the An -abelian Toda theory. It is worthwhile to mention that the
above dyonic models appear to be an appropriate integrable deformation of the recently
(1)
constructed Vn+1
-algebra minimal models [14] (i.e., the singular An -non-abelian (NA)
Toda theories):
L = Lconf
m2
Vdef ,
2

Vdef = e(1 +n1 ) 1 + 2 e1 .
(1.2)
The CPT-violating term e1 1 is hidden in the second term of (1.1)

e1
e1
g + .
=
1 The sine-Gordon (SG) and the G

n -abelian affine Toda model are examples of magnetic type IM having
neutral topological solitons. The complex sine-Gordon (or LundRegge) model [1012] and its homogeneous
space SG generalizations [13] represent the electric type IMs, whose solitons are U (1) (or U (1)r ) charged but
nontopological.
443
It has the same origin as in the conformal theory Lconf , i.e., the axial gauging of the U (1)
subgroup spanned by 1 H . The CPT-invariant partner of (1.1) is the following vector
(1)
type An -dyonic model:
n1

1 1
Lv = 2
ln ci ln(ci ci+1 cn1 ) + ln ci ln(ci ci+1 cn1 )
2
i=1

+ B A
1 AB
2
Vv ,
2
1 AB

m2
B
c2 c3
cn1
Vv = 2 Ac12 c2 cn1 +
+
+
+
n
.
2
c1 c2
cn2
c1 c2 cn2 cn1
(1.3)
It is obtained by vector gauging of the same U (1) subgroup and appears to be T-dual to the
axial model (1.1) [15]. The ungauged integrable model giving rise to both models (1.1)
and (1.3) has the following Lagrangian 2
(n+1)R
1
+ e
k + (n + 1) R R
( n 1 ) Vu ,
Lu = ik i
2
2n
n1

2

m
(i1 +i+1 2i )
(1 +n1 )
2
( (n+1)R
)
1
n
Vu = 2
e
+e
1 + e
n .
i=1
(1.4)
The corresponding action Su is invariant under chiral U (1) gauge transformations:

R = R + w(z) + w(
z),
(n+1)
2n R
= e
(n+1)w(z)
n
= e
(n+1)w(
z)
n
(n+1)
2n R
where = e
, = e
.
The zero curvature representation and the proof of the classical integrability of all
models (1.1), (1.3) and (1.4), is derived in Section 2. Special attention is devoted to the
discussion of discrete symmetries of (1.1) and (1.3). They are crucial in the definition of
top
the vacuum lattice, the spectra of the topological charges Qk and Q as well as in the
derivation of the first order 1-soliton equations in Section 3. It is important to note that the
first order system (3.7) has to be completed with a chain of algebraic relations (3.8) in order
to determine a consistent vacua Backlund transformation. They represent 2-D analog of the
YMH-model Bogomolny equations and constitute our main tool in constructing 1-soliton
solutions of the dyonic models (1.1) and (1.3) in Sections 4 and 5. Compared to the other
methods of construction of soliton solutions as vertex operators (or -functions, etc.)
[1719] they have the advantage in providing a simple proof of the topological character
of the solitons energy, momenta and action S1-sol (A(1)
n ). It is shown in Section 3 that all
ax , Qtop , s ax ) characterizing U (1)-charged
,
Q
conserved quantities (E ax , P ax , M ax , Qax
k
el
topological 1-solitons depend only on the asymptotics of the fields at x . Therefore
their values can be calculated without the knowledge of the explicit form of the solutions.
2 Note that the n = 2 (i.e., SL(3, R)) IM is nothing but the (thermal operator) integrable perturbation of the
(2)
BershadskyPolyakov W3 minimal models [16].
444
The main result of this paper is the explicit construction of the charged topological 1solitons of the axial and vector models (1.1) and (1.3) presented in Sections 4 and 5. Their
semiclassical spectrum (derived in Sections 3, 6 and 7) is shown to have the following
promising dyonic 3 form
20
ax
2
j ,
Qel = 0 jel +
j ,
Qax
=
2
8/02
4
Qmag = 2 j , j = 0, 1, . . ., (n 1),
0

2

4mn Qax
el 0 Qmag
ax
Mjel ,j = 2 sin
,
4n
0

(n 1) ax
s ax =
(1.5)
Qel + 02 Qmag , jel = 0, 1, . . .
2
20
( and 0 are arbitrary real numbers; = i0 ) for the axial model (1.1) and a similar
ax
vec
vec
form with (Qax
el , Q ) (Q , Qel ) interchanged for its T-dual vector model (1.3). It
ax
is worthwhile to mention that j (and , 0 )-dependent shifts of Qax
el and Q come from
the topological -terms (2.28) and (2.33) added to La and Lv , respectively. Our proof
that the axial and vector models 1-solitons have equal masses, energies and spins (see
Section 5) represents an important step in testing the off-critical (i.e., nonconformal)
T-duality. An interesting property of the mass spectrum (1.5) is that its large n limit

m
2

M ax (n ) = 2 Qax
el 0 Qmag
0
(2)
(1)
coincides with the BPS bound for the masses of particular (Qel = 0 = Qmag ) dyons of
(2)
4-D N = 2 super YM theory, i.e., M ax (n ) Z = Q(1)
el Qm [21].
(1)
Section 7 contains preliminary results concerning the exact quantization of the An dyonic models in consideration. Our starting point is the path integral formulation of the
+ two loop WZW model (see Refs. [22]).
\A(1)
models (1.1) and (1.3) as gauged H
n /H
(1)
We next derive the exact effective action for the An -dyonic integrable models following
the method [23] developed for the corresponding conformal -models (i.e., m = 0 case).
It turns out that the relevant counterterms as well as the finite renormalization of the
2
coupling constants renorm
are identical to those of the conformal models (i.e., Vdef = 0 in
2
2
Eq. (1.2)). One can further speculate that Eqs. (1.5) with 02 0,renorm
= kn1
represent
the exact quantum 1-soliton spectrum. Few hints in favour of such conjecture are given
in Section 7.3. The complete answer of the question about the exact quantization of the
electrically charged topological solitons of the dyonic integrable models (1.1) and (1.3),
as well as whether they can be described as representations of the affine quantum group
Uq (A(1)
n ) requires further investigations.
An important feature of all A(1)
n -abelian affine Toda models (n > 1) is known to be
that their action (as well as the Hamiltonian) becomes complex for imaginary coupling
3 To be compared with the semiclassical dyonic spectrum of YMH model (see, for example, Section 6 of Ref.
[20]).
445
constants. The corresponding soliton solutions are also complex functions but their energy
and momentum are real [24]. The only exception is the sine-Gordon model (i.e., A(1)
1 )
whose action, Hamiltonian, solitons and breathers are all real. A similar phenomena takes
place for the A(1)
n -dyonic IM (n > 1). There exist only two exceptional cases with real
action, Hamiltonian, etc. The first is the A(1)
1 vector (or axial) model (i.e., all i = 0)
which is known as the complex sine-Gordon (or LundRegge) model [10,12]. However
it is not of dyonic type since its solitons are either charged but nontopological or neutral
topological in the T-dual picture. The A(1)
2 -vector model represents the unique member of
(1)
the An family of models (1.1) and (1.3) that has real actions, soliton solutions, etc. After
an appropriate change of variables, A = ei0 sinh(0 r), B = ei0 sinh(0 r), c = ei0
we end up with the following real Lv
+ r r
+ tanh2 (0 r)
+
Lv =
2m2
02
sinh(0 r) cos 0 ( + 2).
The properties of its soliton solutions can be extracted from Eqs. (5.8).
It is important to note that the U (1)-charged topological solitons do not exhaust all
the particle-like finite energy solutions of the IMs (1.1) and (1.3). The dyonic IMs in
consideration admit also d species (d = 1, 2, . . . n 1) of neutral 1-solitons that turns
(1)
out to coincide with the An1 -abelian affine Toda solitons [17,25]. The complete list of
solutions also includes three type of breathers representing all the possible bound states
of the charged and neutral 1-soliton and antisolitons (i.e., dyondyon, dyon-monopole
condensates, etc.). The explicit construction of all these solutions and the discussion of
their spectrum are presented in our work [22].
Section 8 contains preliminary discussion of the S- and T-duality properties of the
considered integrable models.
2. A(1)
n -dyonic model
2.1. Zero curvature and graded structures
(1)
The integrability of the axial An -model is a consequence of the fact that its equations
of motion
e
n + 1 e21
+ m2 en1 = 0,
+ 2
2
2n
21
1

2 n + 1 e
e
+ m2 en1 = 0,
+
2
2n
m2 (l+1 +l1 2l )
l + n l e

e

l
m2 (1 +n1 )
e
+
+ 2 en1 = 0,
(2.1)
n
l = 1, . . . , n 1, can be written as zero curvature condition
446
[ + A, + A ] = 0.
(2.2)
The Lax connections A and A (i.e., A ) lies in the A(1)

n centerless affine KacMoody
algebra (loop algebra) 4 and are given by
n1
1 1 H (0)
1 1 (0)
e
e 2 E1
+
j h(0)
+
j
+1
2
2
2
1
j =1
n1
1
1
(0)
(1)
+m
e 2 (2j j1 j+1 ) Ej+1 + e 2 n1 E1 +2 ++n
A = 2
j =1
+e
1
2 (1 +n1 )
E(1)
2 ++n
n1
(0)
1 (0)

(0)
2 1 1 H
e
e 2 1 E1
A =
j
j
+1
2
2
21
j =1
n1
1
1
(1)
m
e 2 (2j j1 j+1 ) E(0)
+ e 2 n1 E(1 +2 ++n )
j+1
j =1
+e
1
2 (1 +n1 )
(1)
E(2 ++n )
(2.3)
(1)
where 1 is the first fundamental weight of An , 21 = n/(n + 1). In the above particular
choice for the connection A is hidden an interesting and rich algebraic structure known
as the LeznovSaveliev method [3] for constructing 2-D integrable models. In order to
give an idea of how it works in our case of the singular non-abelian affine Toda (dyonic)
models, we transform A and A by an appropriate gauge transformation (dressing)
1
+ W W 1 ,
AW
= W A W
n1

1 H
1
1
(0)
1
(0)
i hi+1 + E1 + R 2
W = exp
2
2
1
i=1
that leaves Eqs. (2.1) unchanged and transforms the Lax connection into the following
suggestive form
AW = D 1 D + ,
where
= m
n
A W = D 1 + D,
(2.4)

(0)
El
(1)
+ E(2 ++n )
l=2
The group element D SL(2) U (1)(n1) is parametrized as follows

4 All the algebraic definitions concerning A(1) are as in Refs. [17,22].
n
(2.5)
447

n1

R
R
(0)
(0)
(0)
(0)
(0)
Da = exp
1 H + E1 +
l hl+1 + E1 + 2 1 H
221
21
l=1

n1

R
(0)
(0)
(0)
+
H
+
l h(0)
= exp E

(2.6)
1
l+1 + E1 .
1
21
l=1

As explained in Refs. [14,22] the -invariant subgroup g00 U (1), [ , g00 ] = 0, is

spanned by 1 H (0) , i.e., g00 = exp[ R2 1 H (0)], where l , l = 1, . . . , n, are the funda1
mental weights of An . The appearance of the nonlocal field R defined by
1
= e1
e
R =
(2.7)
,
R
is a consequence of the subsidiary constraint

1
J = Tr DD
J = Tr D 1 D1 H (0) = 0,
1 H (0) = 0
= 0) of the original two loop
imposed on the zero grade conserved currents ( J = J
(1)
(1)
gauged H \An An /H+ WZW model [22,26]. We remind the algebraic recipe for
constructing generic integrable models (see, for example, [1719,22]): given an (finite- or
infinite-dimensional) algebra Gn(a) . Introduce a graded structure (Gn(a) , Q) by means of the
grading operator Q,

[Q, Gl ] = lGl ,
Gn(a) =
Gl ,
[Gl , Gk ] Gl+k ,
l, k = 0, 1, . . . .
(a)
A family of grade one integrable models {Gn , Q, , g0 G0 /G00 } is defined by:

an appropriate choice of grade one constant elements G1 .
The 0-grade group element D = exp(G0 ) contains all fields R(z, z ), l (z, z ), (z, z ),
(z, z ), etc. appearing in (1.1), (2.1), (2.3).
When G0 has an invariant subspace G00 G0 , such that [ , G00 ] = 0, one can consider
the subfamily of singular integrable models by imposing the subsidiary constraints

1 0
Tr D 1 DG00 = Tr DD
(2.8)
G0 = 0
allowing to eliminate the degree of freedom associated to G00 .
Finally, with , D = exp(G0 ) and the subsidiary condition (2.8) we can construct the
desired Lax connections A, A according to Eqs. (2.4).
The answer to the questions concerning the derivation of this recipe, its equivalence
to the Hamiltonian reduction (DrinfeldSokolov) procedure and the two loop gauged
(a)
(a)
H \Gn Gn /H+ WZW models as well as the classification of the grade one
(1)
integrable models can be found in Refs. [22]. The An -dyonic model (1.1) in consideration
corresponds to the following specific choice of the grading operator
Q = nd +

n
l H (0),

E(p) = pE(p) ,
d,
l=2

h(p) = ph(p) ,
d,
i
i
p = 0, 1, . . . .
The grade one constant generators are those given in Eq. (2.5) and G00 = 1 H (0) .
448
2.2. A(1)
n -vector model
As we have mentioned, there exist two inequivalent ways (axial and vector) of gauge
fixing the local U (1) symmetry generated by the currents (J1 H , J1 H ) G00 . The details
concerning the derivation of the axial and vector Lagrangians, (1.1) and (1.3) from the
ungauged one (1.4) are presented in our recent paper [15]. The problem we address here
is about the nonlocal change of the axial variables , , i into the vector ones A, B,
ci . Observe that both, the axial model Eqs. (2.1) as well as the vector model equations of
motion

2 c2
2
Ac1 c2 cn1 ,
c1 = m
c1

cn1
B
2
cn1 = m
,
2
cn2
c1 cn1

k = m2 ck+1 ck , k = 2, 3, . . . , n 2,
c
ck
ck1
A
m2
AAB
+
,
=
2
1 AB
(1 AB)2 c1 cn1
B
B AB
(2.9)
+ m2 c12 c2 cn1
=
1 AB
(1 AB)2
can be written in a compact form

1
+ , D D 1 = 0. (2.10)
DD
D 1 D + , D 1 + D = 0,
For the axial case we take D = Da g0 in the form (2.6), or equivalently in the following
matrix representation

0
d2
Da =
,
dn1 = diag e(2 1 R/n) , . . . , e(n1 +R/n) ,
0 dn1
n1

e 2n R
eR
.
d2 =
(2.11)
n1
R
e 2n R e(1 n ) (1 + 2 e1 )
Eliminating further the field R according to Eqs. (2.7) we derive Eqs. (2.1) from (2.10) and
(2.7). The parametrization of D SU(2) U (1)n1 appropriate for the vector case is

d2
0
Dv =
,
dn1 = diag(c1 , . . . , cn1 ),
0 dn1

A
u
d2 =
(2.12)
,
AB1
B
uc1 c2 cn1
c1 c2 cn1
where the nonlocal field u (the vector model analog of R) is defined by the following first
order equations
B A
AB
(2.13)
,
ln u =
.
1 AB
1 AB
Starting from Eqs. (2.10) with D = Dv and taking into account (2.13) the result now
is the vector model equations (2.9). It is then clear that comparing axial and vector
ln(uc1 cn1 ) =
449
parametrizations of D, i.e., Da = Dv we arrive at the desired axial-vector change of

variables
ek = ck ck+1 cn1 A n , k = 1, . . . , n 1,
n1
n1
AB 1
=uA 2n ,
=
A 2n .
uc1 c2 cn1
nk
eR =A,
Equivalently the reverse vector-axial transformations are given by

ck = e(k+1 k R/n) ,
A = eR ,
B = eR 1 + 2 e1 ,
c1 c2 cn1 1/2
.
u 2 = , u = u
AB 1
(2.14)
(2.15)
One can easily check that inserting (2.14) in Eqs. (2.1) gives the vector model equations
(2.9). It is worthwhile to mention that axial gauge fixing corresponds to the elimination
of the field R from G00 constraints, J1 H = J1 H = 0, i.e., Eqs. (2.7) in this case. In this
language, the vector gauge fixing is equivalent to the elimination of another field / = u 2
from the same constraint equations.
Performing the above change of variables in Lv we find the following relation including
the so-called generating function [15] F :

dF
1
dF
,
=
Lv = La +
(2.16)
R ln
R ln
.
dt
dt
2
It is not surprising that the same relation (2.16) appears as a result of abelian T-duality
transformation between the axial and vector IMs in consideration (see Section 3 of Ref.
[15]). Denote by = 2i 0 ln(/) (for the axial model (1.1)) and = i20 ln A = 2R
(for the vector model) the corresponding isometric coordinates 5 and by and their
conjugate momenta. As it is well known [15,28] the following canonical transformation
= x ,
= x
(2.17)
(and all remaining i , i unchanged) acts as T-duality transformation with generating

function F [28]. Then, the relation (2.16) between vector and axial Lagrangians is a
simple consequence of the fact that both Hamiltonians are equal, i.e., Ha = Hv . Therefore
the nonlocal change of variables (2.14) represent an integrated form of the T-duality
transformations (2.17) accompanied by certain point transformations of the rest of the
(nonisometric) variables k ck = fk (i , R).
2.3. Symmetries and vacua
It is important to note that both, the Noether symmetries of Eqs. (2.1) (and La in (1.1)),
as well as the multiple zeros of the potential (1.1)

1
Va = 2 Tr + D D 1 m2 n
(2.18)
5 + a and + a are symmetries of L and L , that coincides with their global U (1) symmetry.
a
v
450
are encoded in the algebraic data: (Q, , G00 ). Writing the field equations (2.1) in a
compact form (2.10) and taking traces with hi0 , we verify that all the elements hi0 of G00
G0 ([hi0 , ] = 0, i = 1, . . . , K0 = dim G00 ) generate continuous symmetries of Eq. (2.10)

K0

2
l l
a h0 D.
D = exp i0
(2.19)
l=1
(1)
In our case K0 = 1, h0
= 1 H (0) and the above transformation is a global U (1) (electric)

symmetry of (1.1) mentioned in Section 1. For imaginary coupling = i0 , the A(1)
n dyonic Lagrangian (1.1) (as well as the potential (2.18)) are invariant under the following
discrete group transformations
l = l +
2 l
N,
0 n
= ei( n +s1 ) ,
N
l = 1, . . . , n 1,
= ei( n +s2 ) ,
N
(2.20)
for N an arbitrary integer and sa , a = 1, 2, are both even (odd) integers (i.e., s1 + s2 = 2S ,
s1 s2 = 2L ). It is also invariant under CP transformations (P : x x)
l = l ,
= ,
= .
(2.21)
It is convenient to parametrize and as

1 i0 ( 1 1 )
2n
2n
1 i0 ( 1 1 +)
2
2
= e
sinh(0 r),
sinh(0 r)
= e
0
n+1
0
n+1
in terms of one noncompact (r) and two compact (1 , ) fields following the tradition of
the 3-D black string constructions. 6 The , discrete transformations (2.20) and (2.21)
then take the form
= + L ,
r = r,
0
2
r = r,
(2.22)
= +
L
0
with L and L arbitrary integers. As a consequence of (2.20) and (2.21) our dyonic model
(1.1) possesses together with the trivial classical vacuum solution D = 1 (N = 0, l = 0,
= 0, ln(/) = const, V (D = 1) = 0) an infinite set of new distinct vacua
2 l
N,
r (0) = 0,
0 n
(L)
L
i
(L)
=
=
ln
,
(L)
20
20
l(N) =
L = 2L L .
(2.23)
This provides a new set of allowed boundary conditions for the fields at x :
l(N) () =
2 lN
,
0 n
r() = 0,
(L) () =
L
20
6 Our n = 2 free L (i.e., V = 0) after an appropriate field redefinitions (reflecting another U (1) gauge fixing)
and with the counterterms (7.9) and (7.10) included coincides with the Euclidean 3-D black string of Ref. [27].
451
(i.e., () = () = 0, but () = 0). The multiply degenerate vacuum of our

axial integrable model (1.1) is an indication that it admits finite energy topological solitons
solutions interpolating between two different vacua (N , L ): j = N+ N (mod n),
j = L+ L . Due to the U (1) symmetry (2.19) (K0 = 1) such solutions might also carry
nontrivial electric charge:

Qax
el

dx Jel0,ax
= 20
,ax
dx x R,
Jel
= 20 R,
(2.24)
where R is defined in (2.7). The topological current
j
,ax
,
Qax
,
= 0 = ln
J
=
2i
j = 0, 1, . . .
(2.25)
,vec
Jel
(see Section 2.5 and Ref. [15]). It

appears to be T-dual to the electric current
plays the role of electric current in the T-dual model (1.3) that generates the global U (1)
a 2
2
2
transformation: A = eia0 A, B = eia0 B, ck = ei n 0 ck . Similarly to the free case
,ax
,ax
(and their conjugate coordinates and 2R )
(V = 0), [28] the currents Jel , J
form a canonical pairs

1 0,ax
1 0,ax
J
(x, t0 ), (y, t0 ) = (x, t0 ), J (y, t0 ) = (x y)
0 el
0
ax
but {Qax
el , Q } = 0. Note that the other topological currents
Jk
2n
= k ,
0
top
Qk

=
Jk0 dx = kQmag ,
Qmag =
4
j
02
(2.26)
(Qmag stands for the magnetic charge) are not T-duals to the corresponding currents Jk
(5.14) of the vector model (1.3).
The origin of the discrete symmetry (2.20) is in the following continuous symmetry of
the V = 0 free model (1.1)
J1 :
Jk :
Jel :
1 = 1 + a1 ,
k = k + ak ,
l = l ,
i0 a1
i0 a1
= e 2 ,
= ,
= e 2 ,
= , k = 1,
= eia0 ,
= eia0 ,
l = 1, 2, . . . , n 1
broken to (2.23) when V = 0 is added. It is important to mention an interesting relation
between the two U (1) currents J1 , Jel and the topological current I1 = 1 which,
say for n = 2, reads
Jel = 2 J1 2I1 .
(2.27)
2.4. Topological -term

Although Eq. (2.27) takes place in the V = 0 model only, the fact that the electric charge
top
gets contributions from the topological charge Q1 persists in the general case where
452
V = 0 (1.1) as we shall show on the example of the 1-soliton solutions in Section 3. As

one can expect the dyonic properties of our model (1.1) are consequences of the CPT
breaking term
i0 1
e
= LCPT
which, by the following change of variables, 0 = ei 2 1 , 0 = ei

in
0
LCPT =
0 0
1 ln
0 (n + 1)
0

in
0
1 ln
+
.
0 (n + 1)
0
0
2 1
reads,
The pure topological term

in
0
2n
ln
1
1
0 (n + 1)
0
n+1
(being a total derivative) does not contribute to the equations of motion (2.1), but makes
2n
0 1 .
evident the top-charge contribution to the electric current (2.24), Jel = n+1
We further observe that the specific value of the coefficient multiplying the topological
current is irrelevant since one can make it arbitrary by adding to the original Lagrangian
(1.1) certain -type topological terms 7
Latop = i

n1
0
ln
,
k
k
2
8
(2.28)
k=1
where k are arbitrary real constants. As a result, the improved electric current (calculated
from Limpr = L + Ltop ) has the form
03

k k .
4 2
n1
,ax
Jel,impr = 20 R
(2.29)
k=1
Therefore we find that Qax

el is shifted by j :

0,ax
Jel,impr
dx = Qax
el
1
kk ,
n
02
j ,
2
n1
j = 0, 1, . . . , (n 1).
(2.30)
k=1
Similar phenomenon takes place in the open string with one compactified dimension (say
1
ln(/)) and boundary (ChanPaton) term included (see Section 8.6 of
X25 = 2
Ref. [30]). The string momenta P25 gets contribution from the boundary (Wilson line)
7 We restrict ourselves to consider La in a particular form (2.28), although one can include more -terms,
top
say, ij i j + i0 i ln + . The only contribution to Qel comes from (2.28).
453

term GSstring = i A25 (Xi ) dX25 . Such similarity is not occasional, since in our case the
electric current Jelax coincides with the momenta conjugate to the field coordinate ,
which for 1-soliton solutions (4.9) turns out to be periodic. Then the -term (2.28) we have
added to the original action (1.1) can be rewritten as the Wilson line of certain background
gauge field Aa (Xk ) (a = 1, 2, . . . , n + 1 and Xk = {i , , ln }, k = 1, 2, . . . , n + 1):
2

a
2
a
Stop = d z Ltop =
(2.31)
dt i i t = d A (Xk ),
4
2 n1
n1 linear
)
i.e., A = ( 4
i=1 i i and Aa = 0 for a = . Therefore our specific U (1)
gauge potential Aa (X) corresponds to constant electromagnetic field Fab = a Ab
2
) i and all the other components vanish.
b Aa , i.e., Fi = ( 4
2.5. Axial vs. vector -terms
2
ax
0
The shift of the electric charge Qax
el Qel 2 j induced by the topological term
ax
(2.28) together with the fact that Q is equal to the vector model electric charge Qvel
ax
ax
v
v
(i.e., Qax
el , Q Q , Qel by T-duality) addresses the question about the changes in Q
caused by (2.28). Its answer requires the explicit form of the vector model -term, that
corresponds to the axial one (2.28) via T-duality transformation. Applying the axial-vector
change of variables (2.14) in Eq. (2.28) results in a nonlocal -term (including u)
for the
vector model. This is an indication that one should consider new canonical transformations
(more general than (2.17)) in order to have local -terms for the vector model. Therefore
we choose
n1
impr = +
k=1
= 2R,
k k ,
impr = +
n1

ak k + a0 ln 1 + 2 e1 ,
k=1
1
ln
=
2
(2.32)
as new isometric coordinates. Next, we define the standard (T-duality) canonical transformation (2.17) in terms of the new isometric coordinates impr , impr and their conjugate
momenta (together with the point transformations in the first line of Eq. (2.15)). In order
to simplify the calculation we seek for the vector -term in a form:

n1

k
A

ln(c
c
)
ln
Lvtop =
(2.33)
.
k
n1
2
B
k=1
The consistency with the improved T-duality transformation (with impr , impr ) determines the unknown constants k , ak , a0 and k as follows:
k =
k =
k 2
,
4 2
8 2 n
2
ak = k ,
a0 =
n1
1
k (k n),
2n
k=1
n1
k=1 k (n k)
k
l
a0 3
= =
,
k
l
4 2
454
k, l = 1, 2, . . . , n 1.
(2.34)
As it is well known the U (1)-currents and the and topological currents of the axial and
vector models are related as follows:
0,ax
= 0 x impr = J 0,vec,
Jel,impr
0,vec
Jel,impr
= 0 x impr = J0,ax .
(2.35)
The corresponding charges calculated from the Lagrangians with -terms and by using the
explicit form (2.32) of impr (and impr ) indeed coincide,
vec
ax
Qax
el,impr = Q = Qel
02
j ,
2
ax
Qvec
el,impr = Q = Q + 2 j .
(2.36)
The new parameters , 0 , 0 and are defined by

1
kk ,
=
n
n1
0 =
k=1
n1
1
=
kk ,
n
n1
k ,
k=1
0 =
k=1
n1
k=1
and satisfy the following relations

0 = +
8 2
,
2
8 2
,
2 0
0
.
+ 8 2 / 2
(2.37)
Eqs. (2.37) are direct consequence of the above definitions and Eqs. (2.34). Finally, the
shifts in the charges due to the -terms take the form:
ax
Qax
el,impr = Qel
2
j ,
2
Qax
,impr = Q +
20
,
+ 8 2 / 2
Q =
j .
2
(2.38)
2.6. More discrete symmetries

The translational type (l , )-symmetries (2.20) that allows us to determine the vacua
(N)
(1)
lattice (l , (N) ) (2.23) does not exhaust all the discrete symmetries of the An -dyonic
Lagrangian (1.1). By analogy with the Sn symmetries of the An -abelian Toda models one
expects certain (l , )-rotational symmetries to take place. It is easy to verify that the
vector model (1.3) is invariant under the following transformation
A = B,
B = A,
ck =
1
cnk
k = 1, 2, . . . , n 1.
(2.39)
The derivation of their counterpart that leaves invariant the axial gauged model (1.1) is
however far from obvious. We start with the observation that the Eqs. (2.10) as well as
the potential (2.18) remain invariant (V (D ) = V (D)) under the transformation : D =
(D 1 ) with the following properties:
(+ ) = ,
(E1 ) = E(1 ) = E1 ,
2 = 1
(2.40)
P 2 = 1. The next step is to explicitly

combined with the space reflection P : P = ,
construct with the above properties in terms of the elements wk = wk (wk =
455
(k )k ) of the Weyl group of A(1)

n . Define a composite transformation (n 3),
j +2 j +3
t4
j +1 j +2
t4
(n=2j )
= tj +1 tj
(n=2j 1)
= tj +1 tj
2j 1 2j
t3 ,
2j 2 2j 1
t3
,
j = 2, 3, . . . ,
where tkl = wl wl1 wk+1 wk wk+1 wl1 wl , l k, tkk = wk . One can prove by
induction its main property, namely
(n)
(n)
0 (1 ) = 1 ,
0 (2 ) = 2 + 3 + + n ,
0(n) (i ) = ni+3 ,
(k)
i = 3, 4, . . . , n.
(k)
Since 0 (E ) = E0 () we realize that

(1)

(0)
(0)
(0)
0 (+ ) = m E
+ E
+ + E
+ E
+ E(0)
.
n
2
3
n1
2 +3 ++n
Remember that E(l) can be represented as E(l) = l E(0) and d = d/d. We next define
an operator T3 = 2 H (0) + 2d such that
(1)
1 (1)
(1)
(0)
E2 = e ln()T3 E2 eln()T3 = E2 = E2 ,
1
(0)
(1)
E2 ++n = E2 ++n = E2 ++n ,
(0)
(0)
Ek = Ek , k = 3, 4, . . . , n.
Therefore
e ln()T3 0 (+ )eln()T3 =
and as a result we find that the operator has the form = 0 .
Taking into account the explicit parametrization (2.6) and (2.11) of Da we derive the
following field transformations obtained from D = (D 1 ),
k = nk+1 1
R = R +
1
ln 0 ,
= e1 0
nk
ln(0 ),
n
k = 1, 2, . . . , n 1, n = 0,
0 = 1 + 2 e1 ,
(1n)/2n
= e1 0
(1n)/2n
(2.41)
They leave invariant the potential: V (D ) = V (D) and combined with P (i.e., P ),
generate symmetries of the action (1.1). The Lagrangian (1.1) transforms modulo total
derivatives L(D ) = L(D) + L , (see Section 8 of Ref. [14]). We have to mention that
(2.41) can be obtained from the vector model transformations (2.39) applying the axialvector (nonlocal) change of variables (2.14). As we shall see in Section 3, the discrete
symmetries (2.41) play a crucial role in the derivation of the first order solitonic equations
(3.5).
456
2.7. Weyl families of IMs

The algebraic (Weyl group) constructions used in the discussion of the discrete
symmetries of (1.1) addresses the question of whether the remaining Weyl group elements
(or their specific combinations including, say w1 , etc.) act as symmetries of our model
(1.1) and if not, whether the Lagrangians obtained represent new integrable models.
This problem appears to be the nonconformal generalization of the Weyl families of
conformal non-abelian Toda models constructed in Ref. [14] (see Section 8). The simplest
transformation D = w1 (D) is not a symmetry of (1.1). It has the following components
form (D = (0 , 0 , 0i ))

nk
ln 1 + 2 0 0 e01 ,
n

(n1)/2n
,
= 0 e01 1 + 2 0 0 e01

01
2
01 (n1)/2n
= 0 e
.
1 + 0 0 e
k = 0k
(2.42)
As a result of this change of variables the w1 -image of (1.1) is an integrable model with
Lagrangian Lw1,
0 e01
1
0
0j +
V0 ,
Lw1 = ij 0i
01
2
1 + 2 n+1
2n 0 0 e

m2 (201 +02 )
V0 = 2 e
1 + 2 0 0 e01 + e(01 +0n1 )

n1

(0k1 +0k+1 20k )
+
e
n ,
k=2
where 0n = 0. The other Weyl group elements wi and their combinations lead to
transformations similar to (2.42), thus producing new families of classically equivalent
integrable models Lwi . Hence we can conclude that P -transformation given by Eqs.
(2.41) is the unique (affine) Weyl group transformation leaving (1.1) invariant. Contrary to
the abelian affine Toda theories all the other Weyl transformations (different from ) are
not symmetries of (1.1), thus giving rise to new phenomena Weyl families of IMs.
3. Soliton equations
3.1. Vacua Backlund transformations
Consider two arbitrary solutions D1 and D2 of Eqs. (2.10). The corresponding Lax
W
connections AW
(s) = A (Ds ), s = 1, 2, are related by appropriate dressing (gauge)
transformations :
1
1
W
AW
(2) = A (1) + ( ) .
(3.1)
They leave invariant Eq. (2.2) together with the linear problem

AW
(Ds ) T (Ds ) = 0.
457
(3.2)
The relation between the monodromy matrices Ts = T (Ds ) = P exp( AW

dx ) has the
well-known form [18]
T2 = T1 ,
(1)
+ T1 = T1 g0 ,
(3.3)
(1)
where g0
is a constant element of the corresponding affine group. The strategy in deriving

the infinitesimal Backlund transformations D1 D2 consists in the following (a) first
solve Eq. (3.2) for (Ds ) and (b) find first order differential equations for Ds by
(1)
substituting these (Ds ) in (3.1). In the case of An -abelian affine Toda theories [25],
(1)
and for all non-abelian Toda theories based on A1 , the realization of the above recipe is
quite straightforward. We find that

+ = X 1 + (D1 X)1 Y D2
(3.4)
and the corresponding Backlund transformations take the following compact form

D11 D1 X XD21 D2 = D11 Y D2 , ,

1 )D 1 Y Y (D
2 )D 1 = D1 XD 1 , + .
(D
1
(a)
(3.5)
(1)
The constant elements X(, ai ), Y (, bi ) of the universal enveloping algebra of Gn (An

or A(1)
1 in our case), contain all the parameters ai , bi of the Backlund transformation. They
also have to satisfy the following conditions
[X, ] = 0,
[Y, + ] = 0.
(3.6)
It is not difficult to check that the second order equations (2.10) are indeed the integrability
conditions for the first order equations (3.5) if requirements (3.6) are fullfilled. The
verification of the above statement does not depend on the explicit form of (and on
(1)
the parametrization of Ds ) and therefore is valid for our dyonic An model (i.e., and
Ds given by Eqs. (2.5) and (2.6)). It turns out however that the simple form (3.4) of +
(1)
takes place only for specific g0 giving rise to 1-soliton solutions [22], i.e., when D1 =
i
H
e 0 1 = const. The derivation of the explicit form of the vacua Backlund transformation
(D1 = const) for our model (1.1) includes one more complication. Taking X and Y in the
form
X = X01 I + nX02 1 H +
n1
ak ( )k ,
k=1
Y = Y01 I + nY02 1 H +
n1
bk (+ )k
k=1
and D as in Eq. (2.6) we have to further impose on Eqs. (3.5) the requirement of symmetry (2.41) in order to get a complete system of equations. The result is

m (k1 k +R/n)
nk 2
eR/n ,
e(n1 +R/n) +
k =
e
458
k = m e(1 R/n) e(k+1 k R/n) + k 2 eR/n ,
n
k = 1, 2, . . . , n 1, 0 = n = 0
= m eR/n ,
R

n1 2
= m e(n1 +R/n) 1
e(n1 +2R/n) ,
2n
= m e(1 R/n) ,
= m e(1 R/n) ,

= m e(n1 +R/n) 1 n 1 2 e(n1 +2R/n) ,
2n
R = meR/n ,
(3.7)
where
=
b1
Y01 Y02
b2
=
=
= = eb
a1
X01 X02
a2
is the Backlund transformation parameter and the following chain of algebraic relations
should take place,
e1 e2 = e2 e3 = = ek ek+1
= en1 en = en e1 2 .
(3.8)
We have introduced new variables k , and defined by

k = k k1
R
= e 2n ,
R
,
n
k = 1, . . . , n, 0 = n = 0,
R
= e 2n ,
1 + 2 + + n = R
(3.9)
in order to make evident the parallel with the abelian affine Toda case [25]. The algebraic
equations (3.8) are crucial in the proof of the statement that the second order differential
equations (2.1) are the integrability conditions for the first order system (3.7). In the case
of the A(1)
n1 abelian affine Toda model ( = = 0) the analog of Eqs. (3.8) appears again
as a result of the requirement of the abelian analog of the -symmetry (2.41), i.e., (k =
nk+1 ) of the first order equations. They do not play however the same role as Eqs.
(3.8) in the non-abelian model (1.1), but are indeed essential in the derivation of 1-soliton
solutions 8 [25].
It is important to mention that, although our first order system (3.7) has rather
complicated form (including the nonlocal field R) it can be obtained from the simple
solitonic equations of the vector model (5.1) and (5.2) applying the integrated form of
T-duality transformation (2.15). The same is indeed true for the corresponding 1-soliton
solutions.
8 In Ref. [25] they have been used in the form of first integrals ek ek+1 = const.
459
3.2. Soliton spectrum

The main virtue of the vacua Backlund transformation (D1 = const) (3.5) is to provide
an elementary proof [25] of the topological character of the soliton energy and momenta
(and also of the electric charge and the spins, in our case). The key point is that the
top
derivation of the particle like soliton spectrum (M, Qel , Q , Qk , s) does not require
the explicit knowledge of the 1-soliton solutions. The conserved charges depend only on
the asymptotics of the fields at x and the specific solitonic conservation laws 9
encoded in Eqs. (3.7). In order to extend the arguments used in the abelian Toda model [25]
to the A(1)
n -dyonic model (1.1) it is convenient to rewrite Eqs. (3.7) in terms of variables
p , and defined in (3.9)

m p
p =
e
p,1 ,
ep1 + 2

p = m ep ep+1 2
p = 1, 2, . . . , n,
p,n ,
n n ,
= me
1 1 ,
= me
m n
m 1
1 ,
= e
n ,
= e
(3.10)
n . We next observe that the first order

where 1 = 1 + 12 2 e1 , n = 1 12 2 e
system (3.10) (and (3.7) as well) admits the following solitonic conservation laws:
nonchiral

ek + 1 ek+1 = 0,

n
1
1
e
e 1 = 0,
n +
k = 1, . . . , n 1,
chiral

1
+ ep = 0, p = 1, 2, . . . , n,

1
+ ( ) = 0,
1
= 0.
ln
(3.11)
(3.12)
(3.13)
Note that the algebraic relations (3.8) have been used in the derivation of the chiral
conservation currents (3.12) and (3.13). The conclusion is that the complete algebra
of symmetries of the 1-soliton solutions of the axial model (1.1) is generated by
M , = x T
the Poincar currents T (with charge the 2-momenta
P ) and0,01
= s), the electric

x T + spin matrix (the 2-d spin-orbital momenta is i dx M
,ax
,ax
and magnetic currents Jel , J , Jk given by Eqs. (2.24), (2.25) and (2.26) and by
9 The precise statement is that the soliton spectrum is determined by the boundary conditions (b.c.) and by the
algebra of symmetries of the first order (BPS-like) equations (3.7).
460
the internal currents (3.11), (3.12) and (3.13). Leaving aside an interesting problem of
deriving the explicit form of this algebra, we will restrict ourselves to consider few simple
consequences of Eqs. (3.11) and (3.12), that allows us to find the mass spectrum of the
U (1)-charged 1-solitons.
Taking into account Eqs. (3.10) and (3.11) we realize that the potential (2.18)

n
m2 (k+1 k )
e
+ 2 e
n n
V= 2
k=1
and the stress-tensor T :

1

1
k )2 + e
)
T00 =
(k )2 + (
( +
+ V,
4
21
n
k=1
n
1
T01 =
k=1
1

k )2 + e
)
(k )2 (
(
21
(3.14)
are total derivatives:

V =

1
F + F + ,
2

T00 = x F F + ,

T01 = x F + F + .
(3.15)
we have introduced in Eq. (3.15) turn out to be the light cone components F =
The
+
(F , F ) of certain linear combination of the conserved currents (3.11):
n

n
m k
m k
+
2
F = 2
(3.16)
e
,
F =
e
+ .
2
k=1
k=1
Hence the energy and the momentum of the 1-soliton described by Eqs. (3.10) receive
contributions from the boundary terms only:

E=

T00 dx = F F + ,

P=

T01 dx = F + F + . (3.17)
The same is true for the electric charge (2.24), (2.30)

Qax
el

Jel0,ax dx = 20 R() R() ,

0,ax
Jel,impr
dx = Qax
el
n1

03
k k () k ()
2
4
(3.18)
k=1
,ax
as well as for the topological currents J

and Jk by construction. We fix the asymptotics
of the fields p , , (and R) at x by requiring V |x = 0, i.e., the solutions
of (3.10) we seek for to interpolate between two nontrivial vacua (2.23). More precisely,
461
we choose,
l () =
2 l
N ,
0 n
()
= 0,
and 10
() =
l = 1, 2, . . . , n 1,
R() =
2
f
0
ln
=
L ,
20
20
(3.19)
where N and L are arbitrary integers and f are real numbers. According to Eq. (3.9)
we also get
p () =
2
(N f ),
n0
(3.20)
Substituting (3.20) into (3.17) for = i0 we derive the 1-soliton energymomentum

spectrum: 11

4m
1 ax
Qel 02 Qmag sin a,
n sin
2
4n
0

4mi
1 ax
Qel 02 Qmag cos a,
P = 2 n sin
4n
0
E=
(3.21)
where
(N+ + N f+ f ) ib,
n
The soliton charges are given by:
a=
= eb .
Qax
el = 4(f+ f ),
4
Qmag = 2 j , j = N+ N = 0, 1, 2, . . ., (n 1).
0
It turns out that the masses of the charged 1-solitons of the axial model (1.1):

1 ax
4m
M ax = E 2 P 2 = 2 n sin
Qel 02 Qmag
4n
(3.22)
does not depend on the topological charge Qax

. As we shall show in Section 5 the same
formula (3.22) takes place
for
the
1-solitons
of
the vector
but with Qax
el replaced

model (1.3)
vec
vec
2
2
by its dual Q = i x ln(A ) and Qel = i x ln(u ) = 0. Note that the 1
soliton spectrum of the dyonic model (1.1) (i.e., M, Qel , Q , Qk , s, E) depends upon
arbitrary parameters f (i.e., on the b.c. of the nonlocal field R). They are not related to
10 The precise definition [22] of the topological charge Q for arbitrary complex and is Q = Re( ()
. For the 1-soliton solutions it coincides with (3.19).

()), Re () = 21 (Arg Arg )|
0
11 The complex form of E and P is misleading. It will be shown below that the chain relations (3.8) together
with the conservation laws (3.11), (3.12) and (3.13) impose conditions on N and f that ensures the reality of
E and P , (E 0, as well).
462
the b.c. of the physical fields l , , and it does not seems to be one of the new internal
conserved charges. We are going to show now that f represent the value of the following
chain of first integrals of Eq. (3.10)
y = ei0 p ei0 p+1 = ei0 n ei0 1 + 02 ,
p = 1, . . . , n 1.
It is easy to check that as a consequence of Eqs. (3.8), (3.11) and (3.12) we have

ei0 n ei0 1 + 02 = 0, etc.
ei0 1 ei0 2 = 0,
(3.23)
(3.24)
Combining them with the chiral conservation laws (3.11) and (3.12) we realize that all the
members of the chain relations (3.8) represent first integrals of the system (3.23). Thus, the
= 0) conserved currents y = 2i sin
only (zero) mode of the double chiral (i.e., y = y
describes an internal conserved charge. Since (3.23) is valid for all x and t applying them to
the boundary case (t fixed and x ) we find the following relations between f , N
and :
2
(f N ) = sin .
sin
(3.25)
n
The general solution of (3.25) is given by
2
(a)
(f N ) = + 2S
,
n
(3.26)
2
(b)
(f N ) = Sign() + 2S ,
n
(3.27)
or
(a,b)
S
= 0, 1, 2, . . . . We chose f (for cos > 0), such that they provide a nontrivial
-dependence of Qel :
n
(a)
+ 2S+ + N+ ,
f+ =
2
n
(b)
Sign() + 2S + N
f =
(3.28)
2
(and f+ f , N+ N for cos < 0) and therefore

ax
Qel = 4(f+ f ) = 4n Sign() + 4j ,
2

02
0,ax
(3.29)
j ,
Jel,impr
dx = Qax
el
2
(a)
(b)
neglecting the term n(S+ + S ) in Qax

el since j is defined as integer mod n.
Taking into account (3.28) we realize that
sin a = Sign() cosh b,
cos a = i Sign() sinh b
and therefore
E = M ax cosh b,
P = M ax sinh b,
M ax =
4mn
| cos |
02
463
(i.e., E 0 as it should be). To make the discussion of the spectrum of the U (1)-charged
topological soliton complete we anticipate the semiclassical quantization of Qel (see
Section 7.1):
2
j
j
jel = 0, 1, . . . .
Qax
(3.30)
=
+
el
,
0
el
2
This form of Qax
el confirms the arguments presented in Section 2 that the electric charge
of the solitons (and breathers as well [22]) of the axionic model (1.1) gets contributions
from the magnetic charge Qmag = 42 j . It is important to note that the 1-soliton charges
0
(Qax
el , Qmag ) (3.30) coincide with the electric and magnetic charges of the dyonic solutions
of 4-D SU(n + 1) YMH model with CP-breaking -term [29]. The 2-D soliton mass
2
= (Q2el +
spectrum (1.5) however is different from the semiclassical masses Mdyon
Q2mag ) of 4-D YMH dyons. It is worthwhile to mention that the n limit of our mass
(2)
formula (3.22) coincides with the BPS bounds for the masses of particular dyons (Qel =
0 = Q(1)
mag ) of four-dimensional N = 2 supersymmetric YM theory [21].
4. Electrically charged topological solitons
4.1. Soliton solutions
In our derivation of the dyonic properties (1.5) of the 1-solitons of (1.1) we left
unanswered the important question: whether Eqs. (2.1) and (3.10) possess soliton solutions
with both charges Qel and Qm different from zero, i.e., N+ = N and f+ = f . It is indeed
the case as we will show by explicit construction of solutions of Eqs. (3.10) and (3.8). We
first consider the p , p = 1, . . . , n, equations. Taking into account the algebraic relations
(3.8) and (3.23) we realize that they can be rewritten in the following compact form:

ei0 p = meb 1 e2i0 p 2i sin ei0 p ,

ei0 p = meb 1 e2i0 p 2i sin ei0 p .
We next introduce the solitonic variables
+ = cosh(b)x sinh(b)t,
+ = sinh(b)t + cosh(b)x ,
= cosh(b)t sinh(b)x,
= cosh(b)t + sinh(b)x
and taking into account the chiral conservation laws (3.12), (3.13) we get

ei0 p = 0.
+ ei0 p = m 1 e2i0 p 2i sin()ei0 p ,
(4.1)
The general solution of Eqs. (4.1) is given by

ei0 p = ei
Sp e2i+2m cos + 1
,
Sp e2m cos + + 1
where Sp are certain integration constants satisfying the recursive relations

Sp+1 = e2ii Sp
(4.2)
464
as a consequence of Eqs. (3.23) and (4.2). Therefore

Sp = (1)p1 e2i(p1)S1 (1)pn e2i(pn)
and Eq. (4.2) acquires the final form
(1)p1S1 e2ip e2m cos + 1
.
(4.3)
(1)p1 S1 e2i(p1)e2m cos + + 1
With Eq. (4.3) at hand we can write the 1-solitons in the original variables l and R,
ei0 p = ei
R=
n

l
l = R +
p
n
l
p ,
p=1
(4.4)
p=1
as follows
ef + (1)n e2in ef
,
ef + ef
+ (1)ln e2i(nl)ef
ei0 R = ein( Sign())

ei0 l =
ef
(ef + ef ) n (ef + (1)n e2in ef )
nl
n
(4.5)
,
l = 1, . . . , n 1,
(4.6)
where is a complex constant, f = m+ cos + 12 ln(). The next step is to derive solutions
for and . Eqs. (4.5) and (4.6) together with the algebraic relation (3.23) allows us to
determine the product :

(n+1)/n f
(n1)/n
2 = N 2 ef + ef
e + (1)n e2in ef
,

2
2i
n 2in
(1) e
1 .
N = 1+e
(4.7)
= / satisfies the following simple equations
It turns out that the ratio /

b i0 2
i0 1
,
ln
e
ln
= me e
= meb ei0 2 ei0 1 .
Applying once more Eq. (3.23) we find that ln(/ ) is independent of + while the
dependence is linear
= 2im sin .
ln
(4.8)
Therefore the ratio / is given by
= e2i(m sin +q) ,

(4.9)
where q is an arbitrary complex constant. Thus, Eqs. (4.7) and (4.9) completely determine
the solution of the first order equations (3.7)

(n+1)/2n f
(n1)/2n
N
e + (1)n e2in ef
= ei(m sin +q) ef + ef
,

(n+1)/2n f
(n1)/2n
N
e + (1)n e2in ef
= ei(m sin +q) ef + ef
.
(4.10)
The 1-soliton solution of the dyonic model (1.1) is presented by the set of functions (of +
and ) (4.6) and (4.10).
465
4.2. Soliton charges

It remains to be shown that these solitons carry nontrivial electric and magnetic charges.
The simplest way to do this consists in calculating the asymptotics of the fields R, l , ,
from Eqs. (4.5), (4.6) and (4.10) and further comparing with the values (3.19) and (3.28)
proposed in Section 3. We have to distinguish three cases: 12 (i) cos > 0; (ii) cos < 0;
(iii) sin = 0, n odd. Taking the limit x in Eqs. (4.5), (4.6) and (4.10) we obtain
for cos > 0:

2 l l
l
N + K ,
l () =
() = (),
j = 0,
0 n

2
2 n
+ S+ ,
R() =
f+ =
0
0 2

2
2
n( Sign())
+ S .
R() =
(4.11)
f =
0
0
2
l
and S are further restricted by the conditions
The arbitrary integers Nl , K
(1) to provide nontrivial zeros of the potential (2.18),

(2) chain relations (3.8) and (3.23).
The simplest solution satisfying (1) and (2) that leads to nontrivial dyonic spectra (3.29)
(with j = 0) is given by
Nl = N (mod n),
S = N (mod n),
l
K
=0
for all l = 1, 2, . . . , n 1 (N being new arbitrary integers). The important fact is that
the 1-soliton solution given by (4.5), (4.6) and (4.10) (for cos > 0) are topological and
electrically charged, i.e., with both charges Qax
el and Qm different from zero. Such dyonic
type soliton combines the properties of the LundRegge (complex SG) solitons [11] with
the An -abelian affine Toda solitons [17,25]. The case cos < 0 leads again to (4.11) but
with N+l Nl and f+ f , i.e., Qel Qel and Qmag Qmag . Therefore the
corresponding particles can be interpreted as antisolitons.
4.3. Periodic lumps
The main feature of the 1-solitons with sin = 0 (and cos = 0) is that in the rest
frame v = tanh b = 0, they represent periodic in time t, ( = 2/, = m sin ) bounded
solutions. Such periodic lumps behaviour is familiar from the CSG solitons [11] and the
breathers (doublet solutions) of the SG theory [31]. It reflects the angular property of the
of the fields , (see Eq. (4.9)):
)
phase 2i0 = ln(/
(t + , x) = (t, x)
2
0
(4.12)
12 We are not considering here two limiting cases: (a) cos = 0 since there are no true 1-soliton and (b)
(1)
sin = 0, n even which leads to N 2 = 0 and = = 0 (i.e., to the An -abelian Toda model).
466
and therefore t S1 and S1 . The particle interpretation of the 1-soliton data: center of
mass X = Re 12 ln(), U (1)-moduli q, its velocity v = tanh(b) and the angular velocity =
m sin can be borrowed from the CSG model [11] as particle with internal coordinate q,
rotating in the q space with constant angular velocity at the rest frame v = 0. Another
interpretation comes from the SG breather, as bounded motion of charged particles [31].
The singular case sin = 0, i.e., = s , s = 0, 1, 2, . . . and n-odd (i.e., N 2 = 0) is an
example of static solitons (v = 0) with charges

j
4
ax
1
Qel = 4n s 2 sign(cos ) +
Qmag = 2 j ,
n
0
and degenerate (independent of charges) mass M =
4m
n.
02
4.4. Breathers
It is important to note that the electrically charged topological 1-soliton solution of
our model do not exhaust all the particle-like solutions of (1.1). The complete list of
(topologically) stable 1-solitons, breathers and breathing (or excited solitons) includes
together with the above constructed U (1)-charged topological 1-soliton (2-D dyon) also
ax
the neutral 1-soliton of specie d, d = 1, . . . , n 1 (Md , Qmag = 42 d, Qax
el = 0, Q = 0,
i.e., monopole) and their bound states:
The A(1)
n1 -abelian affine Toda breather (two neutral vertices [17]) and the excited
solitons.
The (NA Toda) three vertex breathing [22] describing the one charged 1-soliton and
one neutral (d) 1-soliton bound state.
The four vertex charged breathers [22] describing the bound state of two charged
topological 1-solitons.
The explicit construction of all these solutions for the axial model (1.1) as well as their
spectrum are presented in our forthcoming publication [22]. The method employed in
[22] is a slight modification of the standard (abelian Toda) vertex operators [18] (or
soliton specialization [17] or Hirota function [26]) method adapted to the case of the
singular non-abelian affine Toda models (1.1) and (1.3). The 3-vertex breather represent
an interesting example of U (1) charged particle-like solution carrying both topological
charges j and j = 1 (remember that j = 0 for our charged solitons (4.5), (4.6), (4.10))
[22]:

4
ax
2
(j + d) ,
Qmag = 2 (j + d),
Qel = 0 jel +
2
0
Q = (j + 4 j ), j = 1, j , d = 1, . . . , (n 1).
(4.13)
2
vec
Due to the fact that Qax
= 0 is the T-dual of the vector model electric
charge Qel = 0
ax
vec
(Qel is the dual of the vector model topological charge Q = 2i dxx ln(A)), this
solution plays an important role in the understanding of the T-duality relations between the
discrete spectra of the axial and vector IMs.
467
5. Vector model solitons

5.1. First order equations

k
Taking D1 = Dv , D2 = e1 H and X = X01 I + nX02 1 H + n1
k=1 ak ( ) , Y =
n1
k
Y01 I + nY02 1 H + k=1 bk (+ ) in Eq. (3.5) we derive the following incomplete system
of first order equations
B = m (1 AB)c1 c2 cn1 ,
= m (1 AB) ,
A
c1 c2 cn1

1
ln(c1 ) = m
Ac1 c2 cn1 ,
c1
m
ln(c1 ) = (c1 c2 ),
1
1
ln(ck ) = m
,
ck ck1
m
ln(ck ) = (ck ck+1 ),

1
1
ln(cn1 ) = m
,
cn1 cn2

B
ln(cn1 ) = m cn1
,
c1 c2 cn2
(5.1)
k = 2, . . . , n 2, where
=
b2
b1
= = eb
=
a1
(X01 X02 )
is an arbitrary parameter related to the soliton velocity v. We next impose the condition of
invariance of (5.1) under the discrete symmetries (2.39). It requires that the equations
= m (1 AB)c1 c2 cn1 ,
B
A = m
(1 AB)
c1 cn1
(5.2)
and the following chain of algebraic relations

Ac1 c2 cn1 c1 =
1
1
1
B
c2 = =
cn1 =
c1
cn2
cn1 c1 c2 cn1
(5.3)
to be satisfied. An easy check confirms that each set of A, B, ci that solves (5.1) and (5.2)
and
1
B
+ Ac1 c2 cn1 =
+ c1
c1 c2 cn1
cn1
(5.4)
satisfies the second order equations (2.9) as well. The remaining algebraic relations ensure
the compatibility of (5.4) and (5.1), (5.2), i.e., they can be obtained by differentiating (5.4)
and then simplifying with help of (5.1) and (5.2).
468
5.2. Solitons
An important property of the first order system (5.1) and (5.2) is the existence of specific
solitonic conservation laws:
nonchiral

( Ac1 c2 cn1 ) + 1 c1 = 0,
B
1

+
= 0,
cn1
c1 c2 cn1

1
ck+1 = 0,
ck1 +
chiral

+ d = 0,
u = 0,
k = 1, 2, . . . , n 1,
(5.5)
d = A, B, ck ,
u =
u(c1 c2 cn1 )1/2

.
(AB 1)1/2
As a consequence of (5.5), (5.6) and (5.3) we find that
c2 = 0,
(Ac
1 c2 cn1 c1 ) = 0,
c1
(5.6)
etc.
and therefore all the members of the chain (5.3) represent first integrals of the system (5.1)
and (5.2):
1
ck+1 = y,
Ac1 c2 cn1 c1 = y,
ck
1
B
(5.7)
= y,
cn1 c1 c2 cn1
where y = 2i sin is an arbitrary constant. We next substitute (5.6) and (5.7) in Eqs. (5.1)
and (5.2) resulting the following set of 2(n + 2) first order differential equations for A, B,
ck , u:

ck = 0,
+ ck = 2m 1 ck2 bck ,

2m
A = 0,
+ A = (n1) 1 (1)n1 A2 A 0(n2) + 0(n) ,
0
my
u,
u =
+ u = 0,
2
where
sin[(s + 1)( + 2 )]
0(s) = i
.
cos
The equation for B is similar to the one for A. The following algebraic relation
B=
(1)n1 A + 0(n)
1 0(n2) A
469
holds. In fact, one has to solve equations for ck only since A and B can be obtained from
(5.3). Applying once more the methods used for the axial model (1.1) in Section 4, we get
the 1-soliton solutions for the vector model (1.3):
ck = ei( Sign())
ef (1)kn e2i(nk1)ef
,
ef + (1)kn e2i(nk)ef
ef + (1)n e2in ef
,
ef + ef
ef e2i ef
B = ein( Sign()) f
,
e + (1)1n e2i(n1)ef
A = ein( Sign())
u = ei(m sin +q) ,
(5.8)
where f = + m cos + 12 ln().

5.3. Soliton spectrum
The U (1) symmetry
A = eia0 A,
B = eia0 B,
ck = e
ia02
n
ck
of Lv (1.3) gives rise to the following electric conserved current

A B B A
,vec
+ ln(c1 c2 cn1 ) .
Jel
=i
1 AB
(5.9)
,ax
As in the axial model (Jel = 20 R) it can be realized in terms on the nonlocal

,vec
= i ln(u 2 ). The corresponding
field u (or u)
from Eqs. (2.13) and (5.6), i.e., Jel
electric charge

Qvec
el

Jel0,vec dx
= i

dx x ln u 2
(5.10)
vanishes for the 1-soliton solutions (5.8), i.e., Qvec

el = 0. Observe that according to the
axial-vector change of variables (2.14) we have
u 2 =
u =
u(c1 c2 cn1 )1/2

(AB 1)1/2
u = ei0
,vec
and therefore the electric current Jel

of the vector gauged model is T-dual (see Ref.
,ax
,vec
[15]) of the topological current J
= i 2 ln(/) = 1 2 Jel
of the axial
20
20
model (1.1) (remember that Qax

= 0 for the 1-soliton solutions of the axial model (4.6)
and (4.10)). Next step is to recognize that the b.c. (i.e., ) dependence of the vector model
vec
vec ) comes now from the topological
soliton spectrum (M vec , E vec , Qvec
el , Q , Qmag , s
current (for cos > 0)
,vec
= i ln A2 ,
470

Qvec
= i
dxx ln A = 4n Sign() + 4j .
2
2
(5.11)
,ax
It turns out to be T-dual [15] to the axial model electric current Jel = 20 R. It
reflects the fact that the vector model fields A and B have nontrivial asymptotics at x
,
A(+) = ein =
1
,
B(+)
A() = ein( Sign()) =

,vec
The origin of the symmetries generated by the current J
model (2.20),
A = eis1 A,
B = eis2 B,
1
.
B()
(5.12)
is similar to one of the axial
ck = eis1 /n ck ,
s1 + s2 = 2K, s1 s2 = 2L. One might find contradictory the continuous dependence
of the topological charge Qvec

(5.11). In fact should be quantized, i.e., 2 Sign() =
02
4n (jel
4
j ),
02
as we shall see in Section 7.
k (i.e., the vector analogs of Qtop , J =

The remaining (true) topological charges Q
k
k

cannot be defined as Qk = dx x ln(ck ) since the asymptotics of ck include
certain dependence, as one can see from Eq. (5.8),
2n
0 k ),
ck (+) = ei ,
ck () = ei( Sign()).
(5.13)
Taking into account the axial-vector transformation (2.14) and the vector model 1-soliton
asymptotics (5.12) and (5.13) it is easy to check that

i
2 k
ln ck ck+1 cn1 A(nk)/n =
N ,
0
0 n
k = 1, 2, . . . , n 1
are arbitrary integers. Hence we can take

and N
top = 2n
Q
k
0
+

4k
top
x ln ck ck+1 cn1 A(nk)/n = 2 j = Qk .
0
(5.14)
It is important to note that although the zeros of the vector model potential (i.e., its vacua)
N
) the vector 1-solitons (5.8) interpolating between
are again labeled by two integers (L,
different from zero, j = L + L = 0 (see (5.18)), j =
two vacua with both L and N
= 0 contrary to the axial model charged solitons (4.6) and (4.10) that have
+ N
N
Qax
L+ L = 0, i.e., relates the trivial vacua (0, 0) to (0, N) (for = 0, i.e., without
topological -term).
The derivation of the 1-solitons energy E vec and momentum P vec is similar to the axial
case presented in Section 3. The main tools are again the soliton conservation laws (5.5)
vec
vec
and T00
are indeed total derivatives. Thus, we find
and (5.6) that allow to show that T01
vec
vec
are certain functions of the asymptotics of fields A, B, ck (5.12)
that the E and P
471
vec
similar to the axial E and P (3.21) and (3.22) replacing Qax
el by its dual Q ,

1 vec
4nm
vec
2
M = 2 sin
Q 0 Qmag = M ax .
4n
0
(5.15)
In order to complete the analogy with the axial model soliton spectrum (including
and ) we have to add a purely topological CPT-breaking term to the CPT-invariant vector
Lagrangian (1.3) Limpr = Lvec + Lvtop ,
Lvtop =

n1

k
A

ln(c
c
)
ln
.
k
n1
(5.16)
k=1
,vec
Although the equations of motion remains unchanged Lvtop contributes to Jel

,ax
to J ) as it was explained in Section 2.5:
(i.e.,

Qvec
el,impr
= 0
Qvec
= 0
,impr
ax
x impr dx = Qax
+ 4 j = Q,impr ,
x impr dx = Qax
el
Taking into account that Qvec
,impr
2 vec
Q
= 2j ,
02 ,impr
02
j .
2
(5.17)
/02 is a topological charge, i.e.,
j = 0, 1, 2, . . .
(5.18)
we find
2
ax
j
=
+
j
Qvec
= Qel (jel j ).
0
(5.19)
The complete discussion of the T -duality relations between the solitons and breathers
spectrum of the axial and vector models requires further investigation.
6. Soliton spin
6.1. Weak coupling spectrum
The fields (and particles) in 2-D relativistic theories belong to certain representations of
the 2-D Poincar group, P2 = O(1, 1) T2 . As one can see from its algebra:

+

P ,P = 0
M01 , P = P ,
the Lorentz boosts M01 and the mass operator 2P + P provide a set of mutually commuting operators. Their eigenvalues (s, M 2 ) are used to characterize the P2 representations.
472
For example, the field (s) (z, z ) of Lorentz spin s transforms under O(1, 1) Lorentz rotations: z = e z, z = e z ( real) as

(s)
(z, z ) = eM01 eM01 = es (s)(z , z )
or infinitesimally,
[M01 , (s) (z, z )] = (z z + s)(s) (z, z ).
(6.1)
In P2 invariant
theories M01 appears as O(1, 1) Noether charge in the well known form

iM01 = xT00 dx. Taking the explicit form of T00 (say, (3.15)) in terms of fields and
their momenta and using the canonical commutation relations it is straightforward to derive
the canonical spins of all fields:
(n 1)
.
2
Similarly we find for their U (1) charges
sl = 0,
s = s =
Qell = 0,
(6.2)
Qel = Qel = 02 .
(6.3)
It is instructive to write Eqs. (6.2), (6.3) in a compact form, relating spins with electric
charges:
s=
(n 1)
Qel .
202
(6.4)
We complete the discussion of the spectrum of weak coupling (02 0) fields (particles)
(1)
by noting that the bare masses of the weak fields of the An model (1.1) are given by

l
ml = 2m sin
m = m = m,
(6.5)
.
n
6.2. Strong coupling particles
The main characteristic of the weak coupling particles is that they do not carry
top
topological charges, Q = 0, Ql = 0, as one can easily check by substituting the weak
coupling free solutions into (2.25) and (2.26). It reflects the fact that all weak fields
have vanishing asymptotics at x . The analysis of the classical vacua structure (i.e.,
(1)
nontrivial constant solutions) of the axial An model presented in Section 2.3 allows to list
all the admissible boundary conditions
(N)
() =
2 lN
,
0 n
r() = 0,
(L) () =
L
.
20
(6.6)
The solutions of Eqs. (2.1) with nonvanishing asymptotics (6.6) such that j = N+ N
= 0 and/or j = L+ L = 0 are by construction topologically charged. The simplest
example j = 1, j = 0 of finite energy electrically charged topological soliton is given
by Eqs. (4.6) and (4.10). As it is expected its semiclassical spectrum (1.5) shows the
characteristic strong coupling dependence on the coupling constant 02 . They define a
strong coupling set of particles labeled by (M, s; Qel , Qmag , Q ). The knowledge of the
473
explicit values of these conserved charges (classical, semiclassical and quantum) is crucial
in answering the question about strong coupling symmetry group of the model as well
as in the construction of the dual model in terms of the strong fields (carrying the strong
coupling particles quantum numbers). It is clear that the spin of the strong fields is the main
ingredient in writing the kinetic part of the S-dual Lagrangian.
6.3. Lorentz spin of electrically charged topological solitons
As we have shown in Section 3.2, the fact that 1-soliton stress-tensor is a total derivative

T00 = x F F + ,
T01 = x F + F +
is crucial in the demonstration that solitons energy (and mass) get contributions from the
boundary terms only. One expects that similar arguments take place in the calculation of
the soliton spin. What we need in this case is to represent T00 (and T01 ) as certain second
derivatives

n1
2 2
n1
R + n .
T00 = x
(6.7)
k +
2
k=1
An auxiliary field we have introduced in (6.7) is defined as solution of the following first
order equations:
m
= e(n1 +R/n) ,

= m e(1 R/n) + 2 eR/n .
(6.8)
is proportional to the
As a consequence of (6.8) and Eqs. (3.7), (3.8) we realize that
trace T of the stress-tensor T : 13

m2
m2 (1 +n1 )
e
zz.
+ 2 en1 , = +
The proof of Eq. (6.7) is based on the following identity

n1

n1
R + n = F + F ,
x
k +
2
2
= T =
(6.9)
(6.10)
k=1
which can be easily checked taking into account Eqs. (3.7), (3.8) and (6.8). With the explicit
1-soliton solutions (4.5)(4.7) at hand we find that satisfying Eqs. (6.8) has the form

1
m
m
= ln 1 + e2f 2 (cos )+ + 2i (sin ) .
(6.11)
Note that for imaginary coupling = i0 the stress tensor is complex, say for T00 we have
n1

2 2
n1
k +
R + n .
T00 = 2 x , = i0
(6.12)
2
o
i=1
13 The field with the above properties together with the free field are familiar from the conformal affine
extension [19] of the affine NA-Toda models.
474
Although the energy (3.21) of the 1-solitons is real and finite, their spin M01 is in general
complex and also contains an infinite part

M01 = i
xT00 dx =
2i
(xx )|
.
02
Its real part, however is finite

s = Re(M01 ) =
(n 1)
202
2
Qax
el + 0 Qmag
(6.13)
and we take it as a definition of the spins of 1-solitons in consideration. It is worthwhile to

mention that Eqs. (6.7)(6.13) with R = 0 (Qel = 0) remain valid for the solutions of the
A(1)
n1 -abelian affine Toda, i.e., the spins of the 1-solitons of this model turns out to be
sab. Toda =
n1
Qmag .
2
(6.14)
7. Towards exact quantization

7.1. Semiclassical quantization
Classical periodic motions are known to correspond to quantum bound states (discrete
energy spectrum Ej ) such that the frequency of the emition Ej Ej +1 coincides with
the frequency of the classical motion. The energy eigenvalues Ej can be found from the

BohrSommerfeld quantization rule 0 dt pq = 2j (j an integer), i.e.,

Ej + S = 2j,
S=
dt L.
(7.1)
As we have shown in Section 4 (see Eq. (4.12)) our charged 1-solitons (4.6), (4.10) at
the rest frame v = tanh(b) = 0 represent periodic particle-like motion = 2/(m sin )
similar to the SG breather. Hence, the semiclassical soliton spectrum can be derived from
the field theory analog of Eq. (7.1)
H = i i L,
i =
L
,
i
i = (, , k ),

S + E(v = 0) =
i i = 2jel .
0
Taking into account the relations between and x R we realize that
i
+ = x Rt ln
+ t Rt ln().
0
(7.2)
475
At the rest frame (cosh b = 1, sinh b = 0) our 1-soliton solutions (4.6), (4.7) and (4.9) have
a simple t-dependence:
t ln
t i = t i = 0.
= 2im sin ,
t ln = 0,
Therefore we find (for = 0)

i i =
0
2 ax
Q .
02 el
In the case = 0 the corresponding momenta of the fields i acquire improvements

resulting in

0
2
i i = 2
0

0,ax
Jel,impr
dx.
This leads to the following semiclassical quantization of Qel (see Eq. (2.29) for the
0,ax
definition of Jel,impr
):
ax
2
j + jel .
Qel = 0
(7.3)
2
We find instructive to present an alternative derivation
of the quantization rule (7.2),
(7.3) by explicit calculation of the action S( ) = 0 dxLa on the 1-soliton solutions.
The key point is to demonstrate that La is a total derivative due to the first order equations
(3.7) and (3.10) and the soliton conservation laws (3.11), (3.12) and (3.13). The calculation
is a bit more involved than the similar one in (3.15) concerning T00 and T01 . That is why
we consider the simplest case of A(1)
2 only ( = 1, cosh(b) = 1, v = 0, = 0)
+ e 2 Va = T00 + 2m sin x R,
La =
where we have used

2 R
= 2 eR + eR 2 y 2 .
e
R
R =
+ eR 2 y 2 ,
y
y
Therefore we have
2
SM ( ) = E(v = 0) + 2 Qax
,
0 el
(7.4)
that together with (7.2) leads again to (7.3). It is important to note that repeating this
calculation for the Euclidean case t it we find
2
SE ( ) = 2 Qax
(7.5)
el ,
0
i.e., the Euclidean charged 1-solitons (they are not static) have properties similar to the
instantons, i.e., the Euclidean action is bounded below by the electric charge. In the case
of the vector gauged model (1.3) all the results have the same form as above with Qax
el
Qvec
.
476
7.2. Exact counterterms and renormalization

(1)
The quantum properties of the An -dyonic integrable modes (1.1) and (1.3) turns out to
be quite similar to the CSG model [12]. While the main effect of the soliton quantization
of the SG and A(1)
n -abelian affine Toda theories is the coupling constant renormalization
[24,31] in the CSG, as well as in the models in consideration the 1-loop calculation shows
that new counterterms are needed in order to define a consistent quantum theory.
The same phenomenon takes place in the free limit V = 0 (i.e., m = 0) of (1.1) and
(1.3) known to describe strings in curved background: 2-D blackhole (n = 1), 3-D black
string (n = 2), etc. [27]. Since they can be realized as gauged SL(2) U (1)n1 /U (1)
WZW models, the best way 14 of deriving the corresponding quantum theories (i.e.,
renormalization, counterterms, etc.) is by applying the functional integral methods [23].
The path integral formalism for the conformal limit of our dyonic models (i.e., Vdef = 0 in
(1.2)) [14] was extended in our recent paper [22] to the case of a large family of integrable
+ =
\G(1)
models, that can be represented as two loop gauged WZW model H
n /H
n1

G0 /U (1), G0 = SL(2) U (1) . Integrating over the infinite H but keeping U (1)
(spanned by 1 H (0) ) ungauged, we find the following effective action (of the intermediate
model (1.4) for A0 = A 0 = 0)

1
G0
G0
dx 2 Tr + Dv Dv1
Seff = SWZW (Dv ) 2
0,ren

1
v Dv1 A 0 Dv1 Dv
+ 2
dx 2 Tr A0 D
0,ren

Dv1 A0 Dv A 0 + (1 + 0 )A0 A 0 .
(7.6)
G
,
Extending Tseytlins arguments (see Sections 3 and 4 of the first of Refs. [23]) to the Seff
given by (7.6), we realize that
2
2
,
02 =
,
kn1
k
(n + 1) 2
n+1
0 = 2
=
0,ren
kn1
2
0,ren
=
(7.7)
(and 1 + 0 = 1, i.e., 0 = 0 is the unrenormalized classical coefficient of A0 A 0 term

responsible for the cancellation of the U (1) gauge anomaly). Integrating out A0 , A 0 in the
partition function

G
G0 /U(1)
v
Seff0
Z (1) = DA0 DA0 DDv e

= N DDv eSeff
An
we find the renormalized quantum action for the dyonic model (1.3)
14 An alternative way is to use the representation theory of the parafermionic extensions of the Virasoro
(1)
algebra. The quantum theory of the conformal non free A2 NA-Toda model is shown to be equivalent to
(1,1)
the representation theory of the V3
-algebra [14].

G /U (1)
Seff0
477

n+1
(A B B A)2
= Lv d x +
8
(1 AB)(1 + 0 AB)
(2)
n
R
k ln(ck ) d 2 x.
+i
0
2
(7.8)
k=1
It differs from the classical action (1.3) by

(a) the counterterm
(A B B A)2
n+1
,
8 (1 AB)(1 + 0 AB)
(b) the dilaton contribution
Lct =
i (2)
R
k ln ck ,
0
n
Ldil =
ln cn =
k=1
1 = 3,
(7.9)
i0
ln(1 AB),
2
k = 1, k = 2, . . . , n
(7.10)
(R (2)
is the worldsheet curvature),

(c) the renormalization (7.7).
2
) result
Expanding denominators in (7.9) we find it consistent with the 1-loop (in 0,ren
(1)
(see Ref. [12] for n = 1 case and [32] for the Bn T-self dual Fateevs models).
+ 1)) symmetries of the soliton spectrum
7.3. Uq (SL(n
The most important question concerning the quantum solitons and breathers spectrum is
(1)
about the solitons symmetry group. Following the parallel with the An -abelian affine Toda
models where the quantum solitons are related to the representations with q k+n+1 = 1 of
+ 1)) [8] one expects that neutral and charged 1-solitons
the affine quantum group Uq (SL(n
of our dyonic models to have similar properties whether it is true is an open question,
although there exist few hints that this is indeed the case:
Neutral 1-solitons and their conservation laws coincides with the A(1)
n1 -abelian affine

symmetry (but
Toda solitons, that is why neutral sector should manifest Uq (SL(n))
+ 1, q)).
not SL(n
The form of the classical conservation laws (3.11) for the charged solitons is similar
to the abelian case and the classical braiding relations of the currents components
+ 1))
encoded in the classical r-matrix [39] are supporting the conjecture the Uq (SL(n
is the charged solitons group of symmetries.
The n = 1 case (i.e., the quantum CSG model) corresponds to certain thermal
perturbations of the parafermionic conformal models [33]. Its quantum S-matrix

symmetry in this case.
confirms the Uq (SL(2))
(1)
The n = 2 (i.e., A2 ) dyonic model can be realized as an appropriate perturbation of
(1,1)
the V3 -algebra conformal minimal models constructed in [14] (see Section 9 of
[14]). Our preliminary calculation based on bosonization of the vertices representing
(1)
quantum solitons shows that the A2 dyonic models possess a nontrivial set of
quantum conserved currents, whose charges span the SL(3)q quantum affine algebra.
478
8. Concluding remarks
8.1. S-duality
The particle-like nonperturbative solutions of SU(n + 1) YMH model-monopoles and
dyons are believed to provide the fields degrees of freedom relevant for the description of
the strong coupling phase of YMH (and of QCD in general). Similar phenomena is known
to take place in 2-D IM: topological solitons serve as strong coupling variables dual to
the fundamental (weak coupling) particles (fields), presented in the IMs weak coupling
actions (say, (1.1)) [32,34]. The simplest and best understood example is the strongweak
coupling (S-)duality between the massive Thirring and the sine-Gordon models [34,35].
The knowledge of the exact quantum dyons spectrum is crucial in the derivation of the
YMH model strong coupling symmetry group and its representations to be used in the
construction of the strong coupling YMH effective action. The MontonenOlive duality
conjecture [36] states that the monopoles of the Gn -YMH model belongs to fundamental
representations of the dual group Gvn (say, Gn = SU (n + 1), and Gvn = SU(n + 1)/Zn+1 ).
It has been proved for N = 4 SUSY Gn -YM theories and certain N = 2 SUSY YM-matter
models [37].
For N = 1 and for the non-SUSY YMH models, the exact dyonic spectrum is unknown.
As it was argued in the introduction, the study of the U (1)-charged topological solitons
of appropriate 2-D IM of dyonic type and their exact quantization may contribute to the
understanding of the nature of strong coupling symmetry groups of non-SUSY YMH
theory. The conjectured form of the exact quantum 2-D dyons spectrum and the arguments
presented in Section 7.3 makes feasible the role of the centerless quantum affine group
2
(1)
(1)
2
2
= Kn1
) as strong coupling symmetry group of the An An (q) (q = ei h 0,ren , 0,ren
dyonic IMs (1.1) and (1.3). The question of whether certain affine quantum group appear
in the description of the 4-D YMH quantum dyons and domain walls is far from being
+ 1) symmetries of the classical
answered. The observation concerning the affine SU(n
solutions of the SU(n + 1) selfdual YM (and the YMHBogomolny) equations [9] is an
+ 1)) =
indication that their finite energy quantum solutions could have the Uq (SU(n
(1)
Uq (An ) as an algebra of symmetries. The problem to be solved before any attempt
for affine quantum group improvement of the original MontonenOlive conjecture is
about the integrable models S-dual to the A(1)
n -dyonic models (1.1) and (1.3). The
fundamental fields used in writing their Lagrangians have to carry the quantum numbers
(1)
(Qel , Qm , M, s, j , . . .) of charged and neutral solitons of the An models. Its soliton (i.e.,
strong coupling) spectrum should have the quantum numbers of the fundamental fields of
the dyonic models, i.e., , , i . Similarly to the solitons of all known 2-D integrable
models they must have certain affine quantum group structure behind. The most natural
i h 2 , 2 = 1/ 2 . The partition functions
candidate is the algebra dual to A(1)
n (q) with q = e
of such pair of S-dual IMs are expected to be related by the S-duality transformations

n1 + m1
,
n2 + m2
2
+i 2,
2
r
479

1
where nn12 m
m2 SL(2, Z). We have not a recipe of how to construct the S-dual of a given
IM satisfying all the requirements listed above. An important experience in this direction
are the families of S-dual pairs of IM introduced and studied by Fateev [32]
{, , i }weak , 2 ,

4
(2)
2
{,
,
,
A(2)
}
,
}
,
{,
,
A
i strong
i weak
2n
2n
2

4
(2)
i }weak ,
Dn+1 {, ,
i }strong, 2 Cn(1) {, ,
.
2

(1)
{, ,
i }strong, 2 A(2)
Bn+1
2n+1
(8.1)
The strong coupling models G(a)

n represent the complex SG (LundRegge) model
(, )
interacting with G(a)
-abelian
affine Toda theories (i ). Their actions are similar
n1
to (1.3) and as it is shown in [22] they can be derived following the general Hamiltonian
reduction procedure (or from gauged two loop WZW model) discussed in Section 2. The
explicit form of the Gn(1) valued flat connections depend on the specific choice of the
grading operator Q, constant elements and in the way the chiral U (1) symmetry G00
(a)
is gauged fixed. The weak coupling models G
n represent massive Thirring fermions
v
, interacting with the abelian affine Toda model based on the dual algebra Gn1
of
(1)
(1)
(2)
(1)
Gn1
, i.e., Bn+1
is dual to A(2)
2n1 , Dn+1 is dual to Cn , etc. It is important to note that
only the case of real coupling constant has been considered in Ref. [32]. Therefore, the
corresponding potentials admit only trivial zero i = 0 (but not i 2
, is the
coroot lattice of G(a)

n1 ). As a consequence, their finite energy nonperturbative solutions are
the G(a)
n analogs of the U (1)-charged nontopological solitons of the CSG model and their
(a)
quantization is known to be related to Uq (SL(2)) (and not to Gn (q)). This explains the
pair of Thirring fermions representing these solitons in the weak coupling model. Whether
the methods for quantization of these models, for construction of their exact S-matrix, etc.
[32] can be applied to the imaginary coupling constant case is an open question. Due
to more complicated soliton spectrum of both models for i0 their S-duality for
imaginary couplings has to be reconsidered. Although the S-duality is a property of the
quantum IM (i.e., counterterms, renormalization, etc. are essential), an important problem
to be addressed is whether exists an algebraic recipe (the 2-d analog of MontonenOlive
0
conjecture) that for any given IM {G(1)
n , Q, , G0 } prescribes its S-dual IM (the classical
(1)
0
}. The simplest problem to be solved is to find the graded
, G
n , Q,
limit only) {G
0
structure behind the weak coupling Fateevs models for real and to construct their
zero curvature representations. The presence of the Thirring fermions interacting with the
(1)V
abelian Toda bosons is an indication that they are from the family of Gn (the group dual
(1)
of Gn ) non-abelian Toda-matter models of higher grade, say to be of grade 2 (or
some halfinteger grade) and the physical fields to lie in the lowest grades |s| < 2, as for the
IMs constructed in Ref. [38].
480
8.2. T-duality
(1)
As we have mentioned in Sections 1, 2 and 5 the two An -dyonic IMs (1.1) and
(1.3) studied in the present paper are T-dual by construction. The abelian T-duality
transformation (2.14) and (2.16) is specific for the models with target space metrics
for (1.1) and = 2 ln(A) for (1.3),
admitting one isometric coordinate = 2i 0 ln(/
)
i0
(i.e., our models are invariant under global U (1), + a, + a).
The particular
canonical transformation (, ) ( , )
= x ,
= x ,
(8.2)
which integrated form is Eq. (2.14), are familiar from the string T-duality [28] relating
certain conformal -models representing different curved string backgrounds of 3-D
blackstring type [27]. As it was pointed out in Section 2, the free limits m = 0, i.e.,
V = 0 of our dyonic models are nothing but the conformal -models considered in Ref.
[27] in a specific parametrization (of Gauss type) of the An group elements and without
the counterterm contributions (7.8) and (7.9) essential for the exactness (in 2 ) of the
target-space conformal metrics (gij , bij , dil ). The important difference is that in this
m = 0 case there exists more isometries i = i + ai , together with = + a . Adding
the conformal (and nonconformal) vertices representing nontrivial (bounded for i0 )
potentials results in breaking U (1)n to U (1), i.e., the free conformal T-duality group
O(n, n|Z) is broken to O(1, 1|Z). An important and new feature of the nonconformal Tduality specific for the dyonic IMs is the relation between their soliton (and breathers)
vec
ax
vec
solutions and the interchanges Qax
el Q and Q Qel that are IM generalization of
the well known maps between the momenta and winding numbers in the conformal (string)
case. The complete discussion of the abelian T-duality for a large class of U (1) symmetric
axial and vector IM of dyonic type including the formal proofs of Eqs. (2.14) and (2.16)
are presented in our work [15].
Since the T-dual IMs (1.1) and (1.3) are expected to be appropriate symmetry reductions
of the 4-D SU(n + 1)YMH models and their (charged) topological solitons to be related to
certain charged domain walls solutions, the natural question is about the 4-D consequences
of the T-duality observed in 2-D soliton spectra (see Section 5). It is clear that they have to
represent specific residual (discrete) gauge symmetries as in the string case [28] reflecting
the independence of the strong coupling spectra of the manner of the local (unbroken) U (1)
gauge symmetry is fixed.
The problem of the S- and T-dualities of the soliton spectra of 2-D IMs of dyonic type
deserves special attention and more complete investigation.
Acknowledgements
We are grateful to FAPESP, UNESP and CNPq for financial support.
481
References
[1] R.S. Ward, Phil. Trans. R. Soc. London A 315 (1985) 451;
R.S. Ward, Lecture Notes Phys. 280 (1987) 106;
R.S. Ward, London Math. Soc. Lecture Note Ser. 156 (1990) 246.
[2] M.J. Ablowitz, S. Chakravarty, L. Takhtajan, PAM Rep. 108 (1991);
M.J. Ablowitz, S. Chakravarty, L. Takhtajan, PAM Rep. 113 (1991);
M.J. Ablowitz, S. Chakravarty, L. Takhtajan, Commun. Math. Phys. 158 (1993) 289;
T.A. Ivanova, A.D. Popov, Theor. Math. Phys. 102 (1995) 280.
[3] A.N. Leznov, M.V. Saveliev, Group Theoretical Methods for Integration of Nonlinear
Dynamical Systems, Progress in Physics, Vol. 15, BirkhauserVerlag, Basel, 1992.
[4] N. Ganoulis, P. Goddard, D. Olive, Nucl. Phys. B 205 (1982) 601.
[5] E. Witten, Phys. Rev. Lett. 38 (1977) 121.
[6] A. Kovner, M. Shiffman, A. Smilga, Phys. Rev. D 56 (1997) 79787989;
Also, G. Dvali, M. Shiffman, Phys. Lett. B 396 (1997), hep-th/9612128;
A. Hanany, K. Hori, Nucl. Phys. B 513 (1998) 119, hep-th/9706089;
E. Witten, Nucl. Phys. B 507 (1997) 658.
[7] J. Gauntlett, D. Tong, P. Townsend, Supersymmetric intersecting domain walls in massive
hyperkahler sigma models, DAMTP-2000-69, hep-th/0007124;
J. Gauntlett, R. Portugues, D. Tong, P. Townsend, D-brane solitons in supersymmetric models,
DAMTP-2000-68, hep-th/0008221.
[8] D. Bernard, A. LeClair, Commun. Math. Phys. 142 (1991) 99;
H.J. de Vega, H. Eichenherr, J.M. Maillet, Nucl. Phys. B 240 (1984) 377;
H.J. de Vega, H. Eichenherr, J.M. Maillet, Commun. Math. Phys. 92 (1984) 507.
[9] L. Dolan, Phys. Rep. 109 (1984) 1;
L.-L. Chau, M.-L. Ge, Y.-S. Wu, Phys. Rev. D 25 (1982) 1086.
[10] F. Lund, Phys. Rev. Lett. 38 (1977) 1175;
B.S. Getmanov, JETP Lett. 25 (1977) 119.
[11] N. Dorey, T.J. Hollowood, Nucl. Phys. B 440 (1995) 215.
[12] H.J. de Vega, J.M. Maillet, Phys. Rev. D 28 (1983) 1441.
[13] C.R. Fernandez-Pousa, M.V. Gallas, T.J. Hollowood, J.L. Miramontes, Nucl. Phys. B 484
(1997) 609;
C.R. Fernandez-Pousa, M.V. Gallas, T.J. Hollowood, J.L. Miramontes, Nucl. Phys. B 499
(1997) 673;
C.R. Fernandez-Pousa, J.L. Miramontes, Nucl. Phys. B 518 (1998) 745.
[14] J.F. Gomes, G.M. Sotkov, A.H. Zimerman, Ann. Phys. 274 (1999) 289, hep-th/9803234;
J.F. Gomes, G.M. Sotkov, A.H. Zimerman, Phys. Lett. B 435 (1998) 49, hep-th/9803112.
[15] J.F. Gomes, E.P. Gueuvoghlanian, G.M. Sotkov, A.H. Zimerman, Torsionless T selfdual affine
non abelian Toda models, hep-th/0002173, to appear in the Proc. of the VI Int. Wigner
Symposium, Istambul, Turkey, 2000;
See also, J.F. Gomes, E.P. Gueuvoghlanian, G.M. Sotkov, A.H. Zimerman, T-duality of axial
and vector dyonic integrable models, hep-th/0007116, Ann. Phys. 289 (2001) 232.
[16] A.M. Polyakov, Int. J. Mod. Phys. A 5 (1990) 833;
M. Bershadsky, Commun. Math. Phys. 139 (1992) 71.
[17] D. Olive, N. Turok, J. Underwood, Nucl. Phys. B 409 (1993) 509;
D. Olive, N. Turok, J. Underwood, Nucl. Phys. B 401 (1993) 663.
[18] O. Babelon, D. Bernard, Int. J. Mod. Phys. A 8 (1993) 507;
O. Babelon, D. Bernard, Commun. Math. Phys. 149 (1992) 297.
[19] L.A. Ferreira, J.L. Miramontes, J.S. Guillen, Nucl. Phys. B 449 (1995) 631.
482
[20] L. Alvarez-Gaum, F. Zamora, Duality in quantum field theory (and string theory), hepth/9709180, Workshop on Fundamental particles and Interactions, Vanderbilt University and
CERNLa Plata Santiago de Compostela School of Physics, May 1997.
[21] J. Gauntlett, N. Kim, J. Park, P. Yi, Phys. Rev. D 61 (2000) 125012.
[22] J.F. Gomes, E.P. Gueuvoghlanian, G.M. Sotkov, A.H. Zimerman, Dyonic integrable models,
Nucl. Phys. B 598 (2001) 615, hep-th/0011187.
[23] A.A. Tseytlin, Nucl. Phys. B 399 (1993) 601;
A.A. Tseytlin, Nucl. Phys. B 411 (1994) 509.
[24] T.J. Hollowood, Nucl. Phys. B 384 (1992) 523.
[25] H.C. Liao, D. Olive, N. Turok, Phys. Lett. B 298 (1993) 95.
[26] H. Aratyn, L.A. Ferreira, J.F. Gomes, A.H. Zimerman, Phys. Lett. B 254 (1991) 372;
L.A. Ferreira, J.F. Gomes, A. Schwimmer, A.H. Zimerman, Phys. Lett. B 274 (1992) 65.
[27] J.H. Horn, G.T. Horowitz, Nucl. Phys. B 368 (1992) 444;
P. Ginsparg, F. Quevedo, Nucl. Phys. B 385 (1992) 527.
[28] A. Giveon, M. Porrati, E. Rabinovici, Phys. Rep. 244 (1994) 77;
E. Alvarez, L. Alvarez-Gaum, Y. Lozano, Nucl. Phys. Proc. Suppl. 41 (1995) 1.
[29] E. Witten, Phys. Lett. B 86 (1979) 283.
[30] J. Polchinski, String Theory, Vol. 1, Cambridge Univ. Press, 1998.
[31] R.C. Dashen, B. Hasslacher, A. Neveu, Phys. Rev. D 10 (1974) 4114;
R.C. Dashen, B. Hasslacher, A. Neveu, Phys. Rev. D 10 (1974) 4130;
R.C. Dashen, B. Hasslacher, A. Neveu, Phys. Rev. D 10 (1974) 4138;
R.C. Dashen, B. Hasslacher, A. Neveu, Phys. Rev. D 11 (1975) 3424.
[32] V.A. Fateev, Nucl. Phys. B 479 (1996) 594.
[33] V.A. Fateev, H. deVega, J. Phys. A 25 (1992) 2693;
V. Fateev, Int. J. Mod. Phys. A 6 (1991) 2109.
[34] S. Coleman, Phys. Rev. D 11 (1975) 2088.
[35] S. Mandelstam, Phys. Rev. D 11 (1975) 3026.
[36] C. Montonen, D. Olive, Phys. Lett. B 72 (1977) 117.
[37] C. Vafa, E. Witten, Nucl. Phys. B 431 (1994) 3;
N. Seiberg, E. Witten, Nucl. Phys. B 426 (1994) 16;
N. Seiberg, E. Witten, Nucl. Phys. B 431 (1994) 484.
[38] L.A. Ferreira, J.-L. Gervais, J. Sanchez-Guillen, M.V. Saveliev, Nucl. Phys. B 470 (1996) 236.
[39] J.F. Gomes, E.P. Gueuvoghlanian, G.M. Sotkov, A.H. Zimerman, in: G. Pogosyan et al. (Eds.),
Classical Integrability of Non Abelian Affine Toda Models, to appear in the Proc. of the XXIII
Int. Colloquium on Group Theoretical Methods in Physics, Dubna, 2000, hep-th/0010257.

Local realistic theories and quantum mechanics for

the two-neutral-kaon system
R.H. Dalitz a , G. Garbarino a,b
a Department of Theoretical Physics, University of Oxford, 1 Keble Rd, Oxford OX1 3NP, UK
b Grup de Fsica Terica, Universitat Autnoma de Barcelona, 08193 Bellaterra, Barcelona, Spain
Received 28 November 2000; accepted 3 May 2001
Abstract
0
The predictions of local realistic theories for the observables concerning the evolution of a K 0 K
quantum entangled pair (created in the decay of the -meson) are discussed. It is shown, in agreement
with Bells theorem, that the most general local hidden-variable model fails in reproducing the
whole set of quantum-mechanical joint probabilities. We achieve these conclusion by employing
two different approaches. In the first approach, the local realistic observables are deduced from the
most general premises concerning locality and realism, and Bell-like inequalities are not employed.
The other approach makes use of Bells inequalities. In the first approach, under particular conditions
for the detection times, the discrepancy between quantum mechanics and local realism for the timedependent asymmetry turns out to be not less than 20%. A similar incompatibility can be made
evident by means of a Bell-type test by employing both Wigners and (once properly normalized
probabilities are used) ClauserHorneShimonyHolts inequalities. Because of its relatively low
experimental accuracy, the data obtained by the CPLEAR collaboration for the asymmetry parameter
do not yet allow a decisive test of local realism. Such a test, both with and without the use of Bells
inequalities, should be feasible in the future at the Frascati -factory. 2001 Elsevier Science B.V.
All rights reserved.
PACS: 03.65.-w; 03.65.Ud; 14.40.-n
1. Introduction
In 1935 Einstein, Podolsky and Rosen (EPR in the following) [1] advanced a strong
criticism concerning the interpretation of quantum theory. They arrived at the conclusion
that the description of physical reality given by the quantum wave function is not
complete. EPRs argumentation was based on a condition for a complete theory (every
element of physical reality must have a counterpart in the physical theory) and on a
E-mail address: garbarin@ifae.es (G. Garbarino).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 2 1 9 - X
484
R.H. Dalitz, G. Garbarino / Nuclear Physics B 606 (2001) 483517
criterion which defines physical reality (if, without in any way disturbing a system, we
can predict with certainty (i.e., with probability equal to unity) the value of a physical
quantity, then there exists an element of physical reality corresponding to this quantity).
They also assumed the quantum world to be local: this requirement was introduced in
order to express relativistic causality, which prevents any action-at-a-distance. Starting
from these premises, by considering the behaviour of a correlated and non-interacting
system composed by two separated entities, EPR arrived at the following conclusion:
contrary to what the indetermination principle states, two non-commuting observables
can have simultaneous physical reality, then the description of physical reality given
by Copenhagens interpretation, which does not permit such a simultaneous reality, is
incomplete. At the very heart of their logical conclusion is the following fact: their
assumption, that a quantum system has real and well defined properties also when does not
interact with other systems (including a measuring apparatus), is contradicted by quantum
mechanics.
This was the point attacked by Bohr in his famous response [2] to EPRs paper. Here
he noticed that EPRs criterion of reality contained an ambiguity if applied to quantum
phenomena. Starting from the complementarity point of view, Bohr stated that quantum
mechanics within its scope (namely, in its form restricted to human knowledge) would
appear as a completely rational description of the physical phenomena. In the opinion
of Bohr the conclusion of EPR was not justified since they contradicted quantum theory
at the beginning, through their criterion of physical reality: following Copenhagens
interpretation, quantum reality has to be defined by the experimental observation of
phenomena.
The probabilistic meaning of the quantum wave function is the main assumption that
originated criticisms and debate for a broader interpretation of quantum theory. In fact, the
wave function provides a description of the microscopic world in accordance with the laws
of chance, namely, it is non-deterministic: the actual result of a measurement is selected
from the set of possible outcomes at random. It is this interpretation of the quantum state
that led Einstein to pronounce the historical sentence: God does not play dice.
Within Copenhagens interpretation, the measurement process changes the state of the
measured system through the reduction of the wave packet. The description of this (nondeterministic and non-local) process given by the Hermitian operator associated to the
observable one measures is mathematically different from the (deterministic) evolution of
the statistical predictions of the wave function, which is accounted for by the Schrdinger
equation and its unitary time evolution operator. This matter of fact is also the origin of
different paradoxical conclusions of quantum mechanics. It is important to stress that the
collapse of the wave function is a non-local aspect of quantum mechanics. It arises from
the fact that the theory does not provide a causal explanation of the anti-correlations which
exist between the probabilities of finding a system (say a particle) in two separated regions
of space. The EPR-type correlations of two-particle entangled states clearly exhibits a
non-locality. To avoid this feature, interpretations of quantum mechanics which do not
incorporate the reduction of the wave packet have been introduced (see, for instance,
Bohmian mechanics [3] and Everetts many-world interpretation [4]). However, we have
485
to stress that the non-local features exhibited by EPRs states do not contradict the theory
of relativity, since they do not allow for faster-than-light communications [5].
Another puzzling question concerns the subdivision of the physical world into quantum
system and classical apparatus, the latter being directly controllable and needed to define
(through the measurement process) the properties of quantum phenomena. Actually,
strictly speaking, real physical properties are possessed only by the combined system
of quantum object plus measuring device. This dualistic approach, which leaves the
measuring devices out of the world treated by the mathematical formalism of the theory,
leads to a description of the physical universe which is not unified, namely, to a theoretical
framework which is not fully coherent.
The first hypothesis for the solution of the paradoxical conclusion of EPR concerning
quantum correlations was proposed by Furry [6] in 1935. He assumed that the quantummechanical description of many-body systems could break-down when the particles are
sufficiently distant one from another (practically when their wave functions do not overlap
any more). This means that in presence of EPR correlations between two quantum
subsystems which are very far away one from each other, the state of the global system is no
longer given by a superposition of tensorial products of states but it is simply represented
by a statistical mixture of products of states (namely it is factorizable). However, Furrys
hypothesis revealed to be incorrect: an old experiment concerning polarization properties
of correlated photons [7,9], as well as more recent tests [1012], excluded a possible
separability of the many-body wave function even in the case of space-like separated
particles.
In 1952 Bohm [3] suggested an interpretation of quantum theory in terms of hiddenvariables, in which the general mathematical formulation and the empirical results of
the theory remained unchanged. In Bohms interpretation the paradoxical behaviour of
correlated and non-interacting systems revealed by EPR find an explanation. However, for
such systems Bohms theory exhibits a non-local character, which cannot be reconciled
with relativity theory.
This result is consistent with what Bell obtained in 1964 [13]. He proved that any
deterministic local hidden-variable theory is incompatible with some statistical prediction
of quantum mechanics. This is the content of Bells theorem in its original form, which
has been then generalized [14] to include non-deterministic theories. EPRs paradox was
interpreted as the need for the introduction of additional variables, in order to restore
completeness, relativistic causality (namely locality) and realism in the theory (the point
of view of realism asserts that quantum systems have intrinsic and well defined properties
even when they are not subject to measurements). In line with this requirement, Bell and
other authors [1518] derived different inequalities suitable for testing what has been called
local realism.
Once established the particularity of Bells local realism in connection with the predictions of quantum mechanics, different experiments have been designed and carried out to
test these theories. The oldest ones [18,19] measured the linear polarization correlations
of photon pairs created in radiative atomic cascade reactions or in electronpositron annihilations, whereas, more recently, parametric down-conversion photon sources have been
486
employed [11,12,20]. Essentially, all the experiments performed until now (in optics and
atomic physics) have proved that the class of theories governed by Bells theorem are unphysical: they showed the violation of Bells inequalities and were in good agreement with
the statistical predictions of quantum mechanics. Actually, to be precise, because of apparata non-idealities and other technical problems, supplementary assumptions are needed
in the interpretation of the experiments and consequently, no test employed to refute local
realism has been completely loophole free [12,18,21]. It is then important to continue performing experiments on correlation properties of many particle systems, possibly in new
0 and B 0 B
0 pairs are considsectors, especially in particle physics, where entangled K 0 K
erable examples. If future investigations will confirm the violation of Bells inequalities,
it is clear that, under the philosophy of realism, the locality assumption would be incompatible with experimental evidence. Then, if this were the case, maintaining realism one
should consider as a real fact of Nature a non-local behaviour of quantum phenomena. This
fact is not in conflict with the theory of relativity. Actually, there is no way to use quantum
non-locality for faster-than-light communication: for a correlated system of two separated
entities, according to quantum mechanics, the result of a measurement on a subsystem is
always independent of the experimental setting used to measure the other subsystem.
In this paper we discuss the predictions of local realistic schemes for a pair of correlated
neutral kaons created in the decay of the -meson. The two-neutral-kaon system is
the most interesting example of massive two-particle system that can be employed to
discuss descriptions of microscopic phenomena alternative to quantum mechanics (for a
0 system
discussion concerning possible violations of quantum mechanics in the K 0 K
see Ref. [22]). Unlike photons, kaons are detectable with high efficiency (by observing
0 strong interactions with the nucleons of absorbers).
KS and KL decays or K 0 and K
0
0

Moreover, for K K pairs, which can be copiously produced at a high luminosity
-factory, additional assumptions regarding detection not implicit in local realism (always
implemented in the interpretation of experiments with photon pairs [18]) are not necessary
to derive Bells inequalities suitable for experimental tests of local realism [23]. Finally,
the two-kaon system offers the possibility for tests on unexplored time and energy scales.
A correlation experiment discriminating between local realism and quantum mechanics
could be performed at the Frascati -factory in the future [24]. Indeed, being designed to
0 system, such a factory employs high precision
measure direct CP violation in the K 0 K
detectors. Unlike the other papers in the literature [2532] which treated the two-kaon
correlated system within local realistic models, we shall discuss tests of local realism both
with and without the use of Bells inequalities.
The work is organized as follows. In Section 2 we introduce, starting from the
original EPRs program, the point of view of local realism for the two-kaon system. The
quantum-mechanical expectation values relevant for the evolution of the system are briefly
summarized in Section 3. Section 4 is devoted to the presentation of the local realistic
scheme we use to describe the observable behaviour of the pair: the philosophy of realism is
implemented in our discussion by means of the most general hidden-variable interpretation
of the two-kaon evolution. Then, in Section 5 we study the compatibility among the
local realistic expectation values and the statistical predictions of quantum mechanics. In
487
agreement with Bells theorem, we show how any local hidden-variable theory for the
two-kaon entangled state is incompatible with certain predictions of quantum mechanics.
0 state by employing
In Section 6 the difficulties of testing local realism for the K 0 K
Bell-type inequalities are discussed. We show that, contrary to what is generally believed
in the literature, a Bell-type test at a -factory is possible. Our conclusion are given in
Section 7.
2. From EPRs argument to local realism

The starting point of EPRs argumentation was the following condition for a complete
theory: every element of physical reality must have a counterpart in the physical theory.
They defined the physical reality by means of the following sufficient criterion: if, without
in any way disturbing a system, we can predict with certainty (i.e., with probability equal
to unity) the value of a physical quantity, then there exists an element of physical reality
corresponding to this quantity. In addition, for a system made of two correlated, spatially
separated and non-interacting entities, EPR introduced the following locality assumption:
since at the time of measurement the two systems no longer interact, no real change can
take place in the second system in consequence of anything that may be done to the first
system.
EPR assumed that the physical world is analyzable in terms of distinct and separately
existing elements of reality, which are represented, in the supposed complete theory,
by well defined mathematical entities. The previous criterion of reality supports the
anthropocentric point of view nowadays called realism: it asserts that quantum systems
have intrinsic and well defined properties even when they are not subject to measurements.
Under this philosophy, the existence of quantum world is (as in classical physics) objective:
thus, any measurement performed on a quantum system must produce a result with a
definite and predetermined value.
To exemplify EPRs argumentation, consider the case of a particle with total angular
momentum zero which decays, at rest, into two spin 1/2 particles, 1 and 2, with zero
relative orbital angular momentum, which fly apart with opposite momenta. After a certain
time (when the particles are separated by a macroscopic distance) suppose they do not
interact any more (this situation corresponds to the EPRBohms gedanken experiment
[8,9]). At this time, the normalized spin wave function of the global system, which does
not depend (because of the spherical symmetry of the singlet state) on the quantization
direction of the spin, is:

1
|S = 0, Sz = 0 = |+1 |2 |1 |+2 .
(2.1)
2
For particles 1 and 2, |+ and | represent spin-up and spin-down states, respectively,
along a direction chosen as z-axis. Because of the entangled nature of this wave function,
the two particles do not have definite values of the spin component along any direction. The
superposition of two product states (2.1) produces then non-factorizable joint probabilities.
The paradoxical behaviour of correlated and non-interacting systems originates from the
488
fact that the wave function of the global system is not a tensorial product of superpositions
of states of the component systems.
When 1 and 2 do not interact any more, a measurement of the spin component of one
particle produces a given outcome (which is not predetermined by the quantum state (2.1))
and forces, immediately, the spin of the other particle along the opposite direction; notice
that this is independent of whether or not any measurement is then performed on the other
particle. For instance, if the result of a measurement along the z-axis finds particle 1 in
the spin-up state, we conclude that at the same time particle 2 (which is supposed not to
interact with particle 1 nor with the measuring device) has spin-down along z; the wave
packet reduction has led to the disentanglement of the superposition (2.1):
|S = 0, Sz = 0 |+1 |2 ,
(2.2)
and the total angular momentum of the pair is indefinite after the measurement. The
instantaneous response (due to the collapse of the wave function) of the particle which
is not observed is what Einstein called spooky action-at-a-distance.
The two particles of Bohms gedanken experiment are perfectly correlated, and,
following EPR, the spin component of particle 2 is an element of physical reality, since
it is predicted with certainty and without in any way disturbing particle 2. Moreover, in
order to fulfil the locality assumption (no action-at-a-distance), EPR assumed that such
an element of reality existed independently of any measurement performed on particle 1.
Following EPRs argumentation, the interpretation of the above experiment by means of
quantum mechanics leads to a difficulty. In fact, if we had performed a measurement of
the spin component of particle 1 along another direction, say along the x-axis, this would
have defined the x component of the spin of particle 2 as another element of reality, again
independent of measurement. Obviously, this is also valid for any spin component; then it
should be possible, in the supposed complete theory, to assign different spin wave functions
to the same physical reality. Therefore, one arrives at the conclusion that two or more
physical quantities, which correspond to non-commuting quantum operators, can have
simultaneous reality. However, in quantum mechanics two observables corresponding to
non-commuting operators cannot have simultaneous reality. Therefore, there exist elements
of physical reality for which quantum mechanics has no counterpart, and, according to
EPRs completeness definition, quantum theory cannot give a complete description of
reality.
Actually, one could object, with Bohr [2], that in connection with a correlated system of
non-interacting subsystems (described, in quantum mechanics, by Eq. (2.1)), EPRs reality
criterion reveals the following weak point: it is not correct to assert that the measurement
on subsystem 1 does not disturb system 2; in fact, in quantum mechanics the measurement
do separate systems 1 and 2, which are not separated entities before the reduction of
the wave packet. It is the measurement on system 1 that fixes (in a way that, however,
does not depend only on the experimental setting one uses but contains an element of
randomness) the quantum state (previously undetermined) of system 2. Any measurement
on system 1 is, therefore, a measurement on the entire system 1 + 2. Moreover, in quantum
mechanics two or more physical quantities can be considered as simultaneous elements of
489
reality only when they can be simultaneously measured. Then, from the point of view
of orthodox quantum mechanics, EPRs argumentation ceases to be a paradox: EPRs
proof of incompleteness is mathematically correct but is founded on premises which are
inapplicable to microphenomena.
However, one has to remember that, since in quantum mechanics the elements of reality
of quantum systems are our knowings (and not elements concerning the actual behaviour
of matter), this interpretation only provides an incomplete description of the dynamics of
quantum world, because each knowing originates a collapse of the wave function; this
process affects the future behaviour of the system and randomly selects among different
and alternative possibilities, whose only known characteristic is the statistical distribution.
Thus, in quantum mechanics the reality of two non-commuting observables, which cannot
be defined simultaneously, depends on the measurement one performs. In this way, reality
is in part created by the observer.
EPRs paradox was interpreted as the need for the introduction, in quantum mechanics,
of additional variables, in order to restore completeness, causality and realism. Then, Bell
and other authors developed different inequalities suitable for testing what has been called
local realism.
2.1. Local realism for the two-neutral-kaon system
Now we come to the entangled system of two neutral kaons. In this paper we neglect the
effects of CP violation. Then, the CP eigenstates are identified with the short and long
living kaons (mass eigenstates): |K+ |KS (CP = +1), |K |KL (CP = 1). In
0 are given by:
this approximation the strong interaction eigenstates |K 0 and |K
0

K
= 1 |KS |KL .
2
0

K = 1 |KS + |KL ,
2
(2.3)
The time evolution of the mass (weak interaction) eigenstates is:

KS,L ( ) = eiS,L |KS,L ,
(2.4)
where |KS,L |KS,L (0), = t 1 v 2 is the kaon proper time (t (v) being the time
(kaon velocity) measured in the laboratory frame) and:
i
S,L = mS,L S,L,
2
(2.5)
mS,L denoting the KS and KL masses and S,L the corresponding decay widths: S,L
1/S,L (we use natural units: h = c = 1).
Consider now the strong decay of the (1020)-meson, whose relevant quantum numbers
0 [BR( K 0 K
0 ) 34.1%]. With good approximation
are J P C = 1 , into K 0 K
the process is non-relativistic: in the center of mass system, the kaons correspond to
a Lorentzian factor 1.02. Just after the decay, at proper time = 0, the quantummechanical state is given by the following superposition:
490
0 0

1
1 0
K
K
= |KL 1 |KS 2 |KS 1 |KL 2 ,
|(0) = K 0 1 K
2
1
2
2
2
(2.6)
written in both bases we have introduced. Since the kaon is a spinless particle and the
has spin 1, angular momentum conservation requires the kaons to be emitted in a
spatially antisymmetric state. The state is also antisymmetric under charge conjugation.
The second equality in (2.6) is only approximated when one includes the (small) effects of
CP violation. Moreover, in the above equation, 1 and 2 denote the directions of motion of
the two kaons. From Eqs. (2.3) and (2.4) the time evolution of state (2.6) is obtained in the
following form:

(1 , 2 ) = 1 ei(L 1 +S 2 ) |KL 1 |KS 2 ei(S 1 +L 2 ) |KS 1 |KL 2
2
0 0
0
1 i(L 1 +S 2 )
K
K
e
=
+ ei(S 1 +L 2 ) K 0 1 K
2
1
2
2 2

i( + )
i(S 1 +L 2 )
L 1
S 2
e
+ e

0 0
(2.7)
K

.
K 0 1 K 0 2 K
1
2
In the following we introduce, within local realism, the elements of physical reality for
the two-kaon system. Before doing this, it is important to remember again that within the
philosophy of realism, quantum systems have intrinsic and well defined properties, even
when they are not subject to measurements. The existence of the quantum world is then
(like in classical physics) objective and independent of our observations. As a consequence,
any measurement performed on a quantum system produces a result with a definite and
predetermined value. We shall assume locality by requiring that physical phenomena in
a spacetime region cannot be affected by what occurs in all spacetime regions which
are space-like separated from the first one. This means that when the two kaons are
space-like separated, the elements of reality belonging to one kaon cannot be created
nor influenced by a measurement made on the other kaon. This amounts to expressing
relativistic causality, which prevents any action-at-a-distance. Implicit in our description
is also the non-existence, in any reference frame, of influences acting backward in time:
a measurement performed on one kaon cannot influence the elements of reality possessed
by this kaon for times preceding the measurement.
Quantum mechanics predicts (and we know it is a well tested property) a perfect anticorrelation in strangeness and CP values when both kaons are considered at the same
time (see Eq. (2.7)). If an experimenter observes, say along direction 1, a K 0 (KL ), at the
same time 1 , along direction 2, because of the instantaneous collapse of the two-kaon
0 (KS ). Thus, at time 1 to the kaon
wave function, one can predict the presence of a K
moving along direction 2 we assign an element of reality (since, following EPRs reality
criterion the value of the corresponding physical quantity is predicted with certainty and
without in any way disturbing the system), the value 1 (+1) of strangeness (CP ). The
0 (or KS ) as well as
same discussion is valid when the state observed along direction 1 is K
when one exchanges the kaon directions: 1 2. For times 2 following the observation at
time 1 along direction 1 of a KL (KS ), a CP measurement on the other kaon will give with
491
certainty the same result CP = +1 (CP = 1) one expects at time 1 . This expresses CP
conservation. Obviously, because of the instability of the KL and KS components, along
direction 2 the experimenter could observe either CP = +1 or CP = 1 decay products
at time 2 , but what is important in the present discussion is that for any pair of times
(1 , 2 ) there exists perfect anti-correlation on CP . In the case in which neither kaon has
0 ), at times 2 > 1 along direction 2
decayed, when the kaon detected at time 1 is K 0 (K
0 (K 0 ) as well as a K 0 (K
0 ):
quantum mechanics predicts the possibility to observe a K
since strangeness is not conserved during the evolution of the system (governed by the
weak interaction), perfect anti-correlation on strangeness only exists when both particles
are considered at the same time.
Following EPRs argument, in the local realistic approach one then associates to
both kaons of the pair, at any time, two elements of reality, which are not created by
measurements eventually performed on the partner when the particles are space-like
separated (locality): one determines the kaon CP value, the other one supplies the kaon
strangeness S. They are both well defined also when the meson is not observed (realism)
and can take two values, 1, which appear at random with the same frequency in a
statistical ensemble of kaons. Because of strangeness non-conservation, a particular value
of the element of reality S is defined instantaneously (in fact, instantaneous oscillations
between S = 1 and S = 1 occur), but what is important in the realistic approach is that S
has an objective and well defined existence at any instant time. For a pair, the instantaneous
and simultaneous |S| = 2 oscillations are compatible with locality only if one introduces
a hidden-variable interpretation of the pair evolution which predetermines the times of the
strangeness jumps.
In conclusion, neglecting CP violation, within local realism a kaon is characterized
by two different elements of physical reality, which can both take two values with
equal frequency; thus, four different single kaon states can appear just after the
decay, with the same frequency (25%). They are quoted in Table 1. It is clear that this
classification is incompatible with quantum mechanics: in fact, under local realism a kaon
has, simultaneously, defined values of strangeness and CP , whereas in quantum mechanics
these quantities are described by non-commuting operators, in which case they cannot be
measured simultaneously.
Table 1
Kaon realistic states
State
Strangeness
CP
K1 KS0
0
K2 K
+1
+1
+1
0
K3 KL
0
K4 K
+1
492
3. Quantum-mechanical expectation values

By introducing the shorthand notation:
ES,L ( ) = eS,L ,
(3.1)
and the mass difference:

m = mL mS ,
(3.2)
from Eq. (2.7) the quantum-mechanical (QM) probability

0

(1 , 2 ) 2
0 (2 ) K 0 K
PQM K 0 (1 ), K
1
2
0 0

0
(1 ), K 0 (2 ) K
K (1 , 2 ) 2 ,
PQM K
1
2
0 ) at time 1 along direction 1 and a K
0 (K 0 ) at time
that a measurement detects a K 0 (K
2 along direction 2 is:

0

(1 ), K 0 (2 )
0 (2 ) = PQM K
PQM K 0 (1 ), K
1
= EL (1 )ES (2 ) + ES (1 )EL (2 )
8

+ 2 EL (1 + 2 )ES (1 + 2 ) cos m(2 1 ) .
(3.3)
The other probabilities relevant for our discussion are the following ones:

0

(1 ), K
0 (2 )
PQM K 0 (1 ), K 0 (2 ) = PQM K
1
= EL (1 )ES (2 ) + ES (1 )EL (2 )
8

2 EL (1 + 2 )ES (1 + 2 ) cos m(2 1 ) ,
(3.4)

1
PQM KL (1 ), KS (2 ) = EL (1 ) ES (2 ),
2

1
PQM KS (1 ), KL (2 ) = ES (1 ) EL (2 ),
2

PQM KS (1 ), KS (2 ) = PQM KL (1 ), KL (2 ) = 0,

0 (2 ) = PQM K 0 (1 ), KL (2 )
PQM KS (1 ), K 0 (2 ) = PQM KS (1 ), K

0
(1 ), KL (2 ) = 1 ES (1 ) EL (2 ),
= PQM K
4

0 (2 ) = PQM K 0 (1 ), KS (2 )
PQM KL (1 ), K 0 (2 ) = PQM KL (1 ), K

0
(1 ), KS (2 ) = 1 EL (1 ) ES (2 ),
= PQM K
4
Eq. (3.7) expressing CP conservation.
(3.5)
(3.6)
(3.7)
(3.8)
(3.9)
In the particular case of 1 = 2 :

0

( ), K 0 ( ) = 1 EL ( )ES ( ),
0 ( ) = PQM K
PQM K 0 ( ), K
2

0

( ), K 0 ( ) = 0.
PQM K 0 ( ), K 0 ( ) = PQM K
493
(3.10)
(3.11)
These relations, together with Eq. (3.7), show the perfect anti-correlation of the quantummechanical state (2.7) concerning strangeness and CP .
Starting from probabilities (3.3) and (3.4) it is useful to introduce a time-dependent
asymmetry parameter, defined by the following relation for a generic theory:

0

0 (2 ) + P K
(1 ), K 0 (2 )
A(1 , 2 ) P K 0 (1 ), K

0

(1 ), K
0 (2 )
P K 0 (1 ), K 0 (2 ) P K

0

0 (2 ) + P K
(1 ), K 0 (2 )
P K 0 (1 ), K

0

1

(1 ), K
0 (2 )
(3.12)
.
+ P K 0 (1 ), K 0 (2 ) + P K
The quantum-mechanical expression of this quantity is a function of 2 1 only:
EL (2 1 )ES (2 1 )
cos m(2 1 ),
AQM (1 , 2 ) = 2
(3.13)
EL (2 1 ) + ES (2 1 )
0K
0 ) and
and measures the interference term appearing in like-strangeness (K 0 K 0 or K
0
0
0
0
or K
K ) events.
unlike-strangeness (K K
4. Local realistic expectation values

In this section we discuss the widest class of local hidden-variable models for the twokaon state and their predictions for the observables provided, in quantum mechanics,
by Eqs. (3.3)(3.9). Following the derivation of Ref. [31], we start considering how the
quantum-mechanical expectation values for the single kaon evolution can be reproduced by
a realistic approach. Then, we extend the description to the interesting case of an entangled
kaon pair.
4.1. Evolution of a single kaon
In the realistic approach one introduces the four kaonic states of Table 1. K1 is a state
with defined strangeness (+1) and CP (+1), and the same is true for the other states.
Introduce the notation:

pij ( | 0) p Kj (0) Ki ( ) ,
(4.1)
for the conditional probability that a state Ki is present at time if the original state at
time = 0 was Kj . We can immediately write down the time = 0 probabilities; they are:
p11 (0 | 0) = p22 (0 | 0) = p33 (0 | 0) = p44 (0 | 0) = 1,
pij (0 | 0) = 0
for i = j.
(4.2)
494
When the evolution of the four states is considered, CP conservation requires that, for all
times:
p13 ( | 0) = p14 ( | 0) = p23 ( | 0) = p24 ( | 0) = p31 ( | 0)
= p32 ( | 0) = p41 ( | 0) = p42 ( | 0) 0.
(4.3)
In quantum mechanics, assuming CP conservation, the mass eigenstates |KL and |KS
are exactly orthogonal to each other, then:

KL (0)KS ( ) = KS (0)KL ( ) = 0.
(4.4)
During the time evolution, strangeness jumps between S = +1 (1) and S = 1 (+1)
states occur. Thus, only transitions K1 K2 and K3 K4 are permitted, and Eq. (4.3) is
valid.
In order to fix the time evolution of the four states of Table 1 we have to determine 8
probabilities pij ( | 0). As we are going to show, it is possible to fix these quantities and
reproduce all the quantum-mechanical predictions relevant for the single kaon propagation
[31].
From quantum mechanics (Eqs. (2.3), (2.4)) one obtains:
0

0

K (0)KS ( ) 2 = K
(0)KS ( ) 2 = KS (0)K 0 ( ) 2

0 2 1
( ) = ES ( ),
= KS (0)K
(4.5)
2
where the different terms have obvious significance. These restrictions correspond to
requiring the following equalities, that we write in the same order as before, among the
realistic probabilities:
1

1
p11 ( | 0) + p12 ( | 0) = p21 ( | 0) + p22 ( | 0)
2
2
1
1
1
= p11 ( | 0) + p21 ( | 0) = p12 ( | 0) + p22 ( | 0) = ES ( ),
2
2
2
which correspond to fix:
(4.6)
p21 ( | 0) = p12 ( | 0),
(4.7)
p22 ( | 0) = p11 ( | 0),
(4.8)
p11 ( | 0) + p12 ( | 0) = ES ( ).
(4.9)
The first two equalities are compatible with time-reversal invariance, which follows from
CP T theorem, having adopted CP conservation. In the same way, the equalities:
0

0

K (0)KL ( ) 2 = K
(0)KL ( ) 2 = KL (0)K 0 ( ) 2

0 2 1
( ) = EL ( ),
= KL (0)K
2
(4.10)
require:
p43 ( | 0) = p34 ( | 0),
(4.11)
495
p44 ( | 0) = p33 ( | 0),
(4.12)
p33 ( | 0) + p34 ( | 0) = EL ( ).
(4.13)
At this point, two of the 8 pij s are independent. However, other constraints come from
quantum mechanics. In fact, one can write:

0 0 2 1
K (0)K ( ) = EL ( ) + ES ( ) + 2 EL ( ) ES ( ) cos m
4

1
= p11 ( | 0) + p33 ( | 0) ,
(4.14)
2
where the first (second) equality follows from quantum mechanics (realism) and analogously:

0 0 2 1
K
(0)K ( ) = EL ( ) + ES ( ) 2 EL ( ) ES ( ) cos m
4

1
= p12 ( | 0) + p34 ( | 0) ,
(4.15)
2
where, in the last equality, we have taken into account of Eqs. (4.7) and (4.11).
0 ( )|2 , |K
0 (0)|K
0 ( )|2 ,
The other equations one gets for the quantities |K 0 (0)|K
2
2
|KS (0)|KS ( )| and |KL (0)|KL ( )| do not supply new constraints but are compatible
with the conditions written above. Thus, assuming Eqs. (4.7), (4.8), (4.11) and (4.12),
among p11 , p12 , p33 and p34 we have the system of equations:
p11 ( | 0) + p12 ( | 0) = ES ( ),
p ( | 0) + p ( | 0) = E ( ),
33
34
L

(4.16)
(
|
0)
+
p
(
|
0)
=
EL ( ) + ES ( )Q+ ( ),
p
11
33
p12 ( | 0) + p34 ( | 0) = EL ( ) + ES ( ) Q ( ),
where the shorthand notation:

EL ( )ES ( )
1
12
cos m
Q ( ) =
2
EL ( ) + ES ( )
(4.17)
has been employed. Since Q+ ( ) + Q ( ) = 1, in Eq. (4.16) only three conditions out of
four are independent. A symmetrical choice of p11 , p12 , p33 and p34 leads to the following
realistic probability matrix [31]:
p( | 0)
ES ( )Q+ ( ) + ( )
ES ( )Q ( ) ( )
=
0
0
ES ( )Q ( ) ( )
ES ( )Q+ ( ) + ( )
0
0
0
0
EL ( )Q+ ( ) ( )
EL ( )Q ( ) + ( )
EL ( )Q ( ) + ( )
EL ( )Q+ ( ) ( )
(4.18)
where the degree of freedom is given by the function ( ). The requirement that all the
matrix elements are well defined (0 pij ( | 0) 1) can be satisfied if one chooses
properly the function ( ). A particular solution correspond to keep ( ) 0. Actually,
as we shall see in Section 4.2.2, an identically vanishing ( ) function is the only solution
compatible with the local realistic evolution of a correlated pair of kaons.
496
Table 2
Realistic states for the kaon pair at initial time = 0
Direction 1
Direction 2
K1 KS0 , (S = +1, CP = +1)

0 , (S = 1, CP = +1)
K2 K
S
0,
K3 KL
0,
K4 K
L
0 , (S = 1, CP = 1)
K4 K
L
(S = +1, CP = 1)
0 , (S = +1, CP = 1)
K3 KL

K2 KS0 , (S = 1, CP = +1)
(S = 1, CP = 1)
K1 KS0 , (S = +1, CP = +1)
Table 3
Local realistic states for the kaon pair at times 2 1
Probabilities
Direction 1 (left) time 1
Direction 2 (right) time 2
P1 (1 , 2 ; )
K1 KS0
0
K4 K
L
CP = 1 DP
0
K4 K
P2 (1 , 2 ; )
K1 KS0
P3 (1 , 2 ; )
CP = +1 DP
P4 (1 , 2 ; )
K1 KS0
0
K2 K
P5 (1 , 2 ; )
P6 (1 , 2 ; )
0
K2 K
S
P7 (1 , 2 ; )
CP = +1 DP
0
K2 K
P8 (1 , 2 ; )
P9 (1 , 2 ; )
0
K3 KL
P10 (1 , 2 ; )
0
K3 KL
P11 (1 , 2 ; )
P12 (1 , 2 ; )
P13 (1 , 2 ; )
P14 (1 , 2 ; )
P15 (1 , 2 ; )
P16 (1 , 2 ; )
P17 (1 , 2 ; )
P18 (1 , 2 ; )
CP = 1 DP
0
K3 KL
0
K4 K
L
0
K4 K
L
0
K3 KL
0
K3 KL
CP = 1 DP
0
K3 KL
0
K4 K
L
0
K2 K
S
CP = +1 DP
0
K2 K
S
K1 KS0
K1 KS0
CP = +1 DP
CP = 1 DP
0
K4 K
K1 KS0
0
K2 K
CP = +1 DP
CP = 1 DP
CP = 1 DP
CP = +1 DP
4.2. Evolution of a correlated kaon pair

0 pair
Now we come to the time evolution of a correlated and non-interacting K 0 K
emitted in the decay of a -meson.
From quantum theory (Eq. (2.7)) we know that for any time the joint observation of the
mesons finds them perfectly correlated. At time = 0, immediately after the decay, in
the realistic picture there are four possible states for the kaon pair, each appearing with
a probability equal to 1/4: they are listed in Table 2. Kaon K1 is created together with
497
a K4 : we assume, as in quantum mechanics, since it is a well tested property, a perfect

anti-correlation in strangeness and CP when both kaons are considered at equal time. The
other three initial states show, obviously, the same correlation property.
When the system evolves, the kaons fly apart from each other, and at two generic times 1
and 2 (corresponding to opposite directions of propagation labeled 1 and 2, respectively)
the kaon pair is in one of the states reported in Table 3. The first row refers to the state
with a K1 at time 1 along direction 1 (which we define as left direction) and a K4 at
time 2 along direction 2 (right direction; we have in mind, here, the kaon pair propagation
in the center of mass system). Given the classification of the table, in our discussion we
consider 2 1 : the isotropy of space guarantees the invariance of the two-kaon states by
exchanging the directions 1 and 2. In the second row the state corresponds to a left going
K1 at time 1 and CP = 1 decay products (DP) at time 2 on the right. These decay
products originate from the instability of the K3 and K4 pure states, which are both long
living kaons, namely, CP = 1 states (the corresponding physical processes are: KL
3, , ee ). At time 1 the state correlated with a left going K1 is necessarily either
a K4 or a state containing CP = 1 decay products, K3DP or K4DP . Then, at time 2 (> 1 )
on the right we can have: (i) a K4 (state in the first row), (ii) CP = 1 decay products
(state in the second row) or (iii) a K3 (state in the fourth row). The former case refers to
the transition K4 (1 ) K4 (2 ), the latter to K4 (1 ) K3 (2 ), both along direction 2.
Occurrence (ii) takes contributions from the following transitions: K3DP (1 ) K3DP (2 ),
K4DP (1 ) K4DP (2 ), K4 (1 ) K4DP (2 ) and K4 (1 ) K3 (1 < < 2 ) K3DP (2 ).
The other states in Table 3 have similar meaning, the last two rows corresponding to
the situation in which both left and right going kaons are decayed at times 1 and 2 ,
respectively.
4.2.1. Interpretation of the states with local hidden-variables
At this point it is important to stress that the states listed in Table 3 are assumed to
be well defined for all times 1 and 2 with 1 2 : this is the main requirement of the
realistic approach (analogous discussion is valid for states in Tables 1 and 2). For a given
kaon pair we assume that only one of the 18 possibilities of Table 3 really occurs for
fixed 1 and 2 . This means that we are making the hypothesis (realism) that there exist
additional variables, usually called hidden-variables (with respect to orthodox quantum
mechanics these variables are hidden in the sense that they are uncontrollable), that
provide a complete description of the pair, which is viewed as really existing and with well
defined properties independently of any observation. The state representing the meson pair
for given times (1 , 2 ) is completely defined by these hidden-variables: they are supposed
to determine in advance (say when the two kaons are created) the future behaviour of the
pair. Thus, the times in correspondence of which the instantaneous |S| = 2 jumps and
the decay occur for a given kaon are predetermined by its hidden-variables. Under this
hypotheses there is no problem concerning a possible causal influence acting among the
different entities of entangled systems when a measurement takes place on one subsystem.
However, the new variables, which we denote with the compact symbol , are unobservable
because they are averaged out in the measuring processes, and unobservable are the states
498
of Table 3. In principle, also the measuring apparata could be described by means of

hidden-variables, which influence the results of measurement. Besides, hidden-variables
associated to the kaon pair could show a non-deterministic behaviour. It is important to
stress that in the approach with hidden-variables, the probabilistic character of quantum
mechanics is viewed as a practical necessity for treating problems at the observation level,
but (and this is a strong difference compared to the orthodox interpretation) does not
originate from the intrinsic behaviour of microphenomena: the indetermination principle
is supposed to act only during the observation process.
The realistic probabilities listed in Table 3:
Pi (1 , 2 ; ) Pi (1 , 2 | ) (),
(4.19)
correspond to the situation in which a single meson pair, described by the value of the
hidden-variables, is considered. Once 1 and 2 are fixed, the state of the kaon pair (we
1 ,2 ]
,
can think it is fixed at the time of the pair creation) can take values in the set {[
i
i = 1, . . . , 18} (however, we stress again, is fixed when a single pair is considered), and
for a deterministic theory we have:

1 ,2 ]
.
Pi 1 , 2 ; j[1 ,2 ] = ij [
(4.20)
i
In the previous
relations,

is the probability distribution of the kaon pair hidden-variables
1 ,2 ]
and Pi 1 , 2 | [
(= ij ) is the probability of the ith state of Table 3 conditional
j
1 ,2 ]
on the presence of a pair in the state [
. For a single meson pair, only one of the
j
probabilities of Table 3 is different from zero at instants (1 , 2 ) in a deterministic model:
[ , ]
if k 1 2 , the non-vanishing probability is the kth of the table. As far as different

times 1 and 2 are considered, Eq. (4.20) is valid for the same set of hidden-variables, but
1 ,2 ]
[ , ]
in general with the variables appearing in a different permutation, i 1 2 P [
j
(indexes i and j always refer to the classification of Table 3). Therefore, in the model
we are describing the pair can be created in 18 different realistic states, namely with
18 different values of the hidden-variables. We stress again: the hidden-variable sets at
different pair of times contain the same objects, but in one of the 18! different orderings;
and the notation used in Eq. (4.20) does not mean that the hidden-variables are timedependent. From Eq. (4.20) it follows that in a model with deterministic kaon pair hiddenvariables, the normalization of the local realistic probabilities corresponds to that of the
hidden-variables:
18

i=1
18

1 ,2 ]
= 1.
Pi 1 , 2 ; i[1 ,2 ] =
[
i
(4.21)
i=1
In the realistic interpretation, the quantum state (2.7) corresponds to a statistical

ensemble of meson pairs, which are further specified by different values of the hiddenvariables. To exemplify, let us consider such a (large) ensemble of identical kaon pairs
specified by different s (which we suppose now to be continuous variables), whose
distribution we assume to be independent of the apparatus parameters 1 and 2 , the
kaons being emitted in a way which does not depend on the adjustable times 1 and 2
499
(we assume, here, no retroactive causality). We can give a statistical characterization of

this ensemble by means of the set of observables (3.3)(3.9). Considering, as an example,
0 (2 )], within a general deterministic local hidden-variable interpretation:
P [K 0 (1 ), K

0

0

0 , 2 |
PLR K (1 ), K (2 ) d () P K 0 , 1 ; K

0

, 2 | ,
= d ()P Left K 0 , 1 | P Right K
(4.22)
0 , 2 | ), which is
where LR stands for local realism. The joint probability P (K 0 , 1 ; K
conditional on the presence of the particular value of the hidden-variables, has been
assumed to be locally explicable, then it appears in the factorized form in the last equality.
0 , 2 | )] is the conditional probability that, once
The function P Left (K 0 , 1 | ) [P Right (K
is fixed, the left (right) going kaon at time 1 (2 ), which is fully specified by , is K 0
0 ). As required by locality, given and the apparatus parameters 1 and 2 , P Left and
(K
P Right are independent. They only take two values:

1, when the state at time 1 is K 0 ,
P Left K 0 , 1 | =
(4.23)
0, when the state at time 1 is not K 0 .
The knowledge (impossible, we emphasize again) of the hidden-variables associated to an
individual kaon pair emission would permit the determination of the instants the KS and
KL components decay, then Eq. (4.23) follows. The locality condition is motivated by the
requirement of relativistic causality, which prevents faster-than-light influences between
space-like separated events. In the present case, assuming there is no delay among the
times at which the experimenters choose to perform their observations and the real kaon
measurement times 1 and 2 , the locality requirement is fulfilled when the two observation
events are separated by a space-like interval (see Eq. (5.4)). However, to be precise, as we
shall explain in Section 5, a loophole that is impossible to block exists and could permit, in
principle and without requiring the existence of action-at-a-distance, information to reach
both the measuring devices for any choice of the detection times 1 and 2 .
In the above we have restricted our argumentation to deterministic theories only, but
it is possible to extend the same description given by Eq. (4.22) to non-deterministic
(namely stochastic) theories as well as to deterministic theories in which additional hiddenvariables correspond to the measurement devices [18]. Let us make the hypothesis that also
the experimental apparata are described in terms of hidden-variables, which influence the
measurement outcomes. In this case, denoting with ( ) the hidden-variables specifying
the behaviour of the apparatus measuring on the left (right), in the new local hiddenvariable theory the expectation value of Eq. (4.22) is given by:

0 (2 )
PLR K 0 (1 ), K

0 , 2 | , ,
d d d , , P K 0 , 1 ; K

= d d d () p( | ) p( | )

0

, 2 | , ,
(4.24)
P Left K 0 , 1 | , P Right K
500
where,

, , p , | () = p | p | ().
(4.25)
In the second equality of Eq. (4.24), locality has been assumed for both kaon pair and
apparata hidden-variables; p( | ) [p( | )] is the conditional probability that, when
the kaon pair is specified by the variables , the device measuring the left (right) going
kaon is described by the variables ( ). Since the distribution () of the kaon pair
hidden-variables is normalized to unity, the same occurs for p( | ) and p( | ) for
any . It is than clear, by comparing Eqs. (4.22) and (4.24), that:

0
0

0 , 2 | , , ,
P K , 1 ; K , 2 | = d d p , | P K 0 , 1 ; K
(4.26)
and when one implements locality for the kaon pair and apparata hidden-variables, equality
(4.23) is replaced by:

0

Left
0P
(4.27)
K , 1 | d p | P Left K 0 , 1 | , 1.
The difference compared to the deterministic case without apparata hidden-variables is
now clear, and in the new picture an equality like (4.23) is valid for P Left (K 0 , 1 | , ).
It is important to stress here that for the most general non-deterministic local hiddenvariable theory, elements of randomness entering the probabilities P Left (K 0 , 1 | ) and
0 , 2 | ) could be related not only to apparata hidden-variables, but also with
P Right (K
other unknown mechanisms. Nevertheless, the above discussion that led to Eqs. (4.24)
(4.27) also applies to the most general non-deterministic local hidden-variable theory.
In our local realistic theory (Table 3) the set of hidden-variables describing the kaon pair
forms a discrete set, and Eq. (4.22) reduces to:

0 (2 ) = [1 ,2 ] P KS0 , 1 ; K
L0 , 2 | [1 ,2 ]
PLR K 0 (1 ), K
1
1

[ , ]
0 , 2 | [1 ,2 ]
+ 9 1 2 P KL0 , 1 ; K
S
9

P1 1 , 2 ; 1[1 ,2 ] + P9 1 , 2 ; 9[1 ,2 ] ,
(4.28)
where in the last line the notation of Table 3 has been introduced. It is important to stress
that, contrary to what holds for quantum-mechanical probabilities, in the description with
hidden-variables of Eqs. (4.22) and (4.28), the transitions that lead to two-kaon states which
0 (2 )] do not interfere with one another.
contribute to PLR [K 0 (1 ), K
Within the scheme of Table 3 it is easy to obtain the predictions of local realism for
measurements concerning an individual kaon. For instance, the probability for observing a
left going K 0 at time 1 is given by:

0

Left
1 ,2 ]
K (1 ) =
.
PLR
(4.29)
Pi 1 , 2 ; [
i
i=1,2,4,9,10,12
The hidden-variable interpretation of the observables we have discussed in this section

has also been assumed, even if not explicitly declared, when, in Section 4.1, we treated the
propagation of a single kaon from the point of view of realism.
501
4.2.2. Evaluation of the observables

Now we proceed to discuss the range of variability of the meson pair observables
compatible with the most general local realistic model. We shall use the rules of classical
probability theory.
Start considering the probabilities of Table 3. In particular, we concentrate on the state
in the fourth row. At time 1 the left going kaon is K1 , then, requiring CP conservation,
at time = 0 the initial state was either a K1 or a K2 . Since either a K1 or a K2 must
be present as initial state for the left going kaon (both with equal frequency 1/4), as
we see from matrix (4.18), the probability that at time 1 the state is K1 equals [p11
(1 | 0) + p12 (1 | 0)]/4 = ES (1 )/4. Correlated with this K1 , at the same time 1 , there
is either a K4 or CP = 1 decay products on the right. Since the two-kaon state we are
considering corresponds to a K3 at time 2 , at time 1 we must require the presence of a
K4 : the probability that at this time the K4 has not yet decayed is p43 (1 | 0) + p44 (1 |
0) = EL (1 ). Finally, from 1 the K4 must evolve into K3 at time 2 . This transition occurs
with (conditional) probability that we denote:

p34 (2 | 1 ) p K4 (1 ) K3 (2 ) .
(4.30)
Then, probability P4 of Table 3 is given by:

1
P4 1 , 2 ; 4[1 ,2 ] = ES (1 ) EL (1 ) p34 (2 | 1 ).
(4.31)
4
By using the same line of reasoning one obtains the other probabilities. They have the
following expressions:

1
P1 1 , 2 ; 1[1 ,2 ] = ES (1 )EL (1 )p44 (2 | 1 ),
4

1

[1 ,2 ]
P2 1 , 2 ; 2
= P6 1 , 2 ; 6[1 ,2 ] = ES (1 ) 1 EL (2 ) ,
4

1
P3 1 , 2 ; 3[1 ,2 ] = 1 ES (1 ) EL (1 ) p43 (2 | 1 ) + p44 (2 | 1 ) ,
4

1
[1 ,2 ]
P5 1 , 2 ; 5
= ES (1 )EL (1 ) p33 (2 | 1 ),
4

1
[1 ,2 ]
P7 1 , 2 ; 7
= 1 ES (1 ) EL (1 ) p33 (2 | 1 ) + p34 (2 | 1 ) ,
4

1
[ , ]
P8 1 , 2 ; 8 1 2 = ES (1 )EL (1 ) p43 (2 | 1 ),
4

1
[1 ,2 ]
P9 1 , 2 ; 9
= ES (1 )EL (1 ) p22 (2 | 1 ),
4

1
[1 ,2 ]
[1 ,2 ]
P10 1 , 2 ; 10
= P14 1 , 2 ; 14
= EL (1 ) 1 ES (2 ) ,
4

1
[1 ,2 ]
P11 1 , 2 ; 11
= ES (1 ) 1 EL (1 ) p21 (2 | 1 ) + p22 (2 | 1 ) ,
4

1
[1 ,2 ]
P12 1 , 2 ; 12
= ES (1 )EL (1 ) p12 (2 | 1 ),
4
502

1
[1 ,2 ]
= ES (1 )EL (1 ) p11 (2 | 1 ),
P13 1 , 2 ; 13
4

1
[1 ,2 ]
= ES (1 ) 1 EL (1 ) p11 (2 | 1 ) + p12 (2 | 1 ) ,
P15 1 , 2 ; 15
4

1
[1 ,2 ]
= ES (1 )EL (1 ) p21 (2 | 1 ),
P16 1 , 2 ; 16
4

1
[1 ,2 ]
= 1 ES (1 ) 1 EL (2 ) ,
P17 1 , 2 ; 17
2

1
[1 ,2 ]
= 1 EL (1 ) 1 ES (2 ) .
P18 1 , 2 ; 18
2
(4.32)
The description of Eqs. (4.28) and (4.29) and Table 3 corresponds to the most general hidden-variable theory. Actually, the local realistic probabilities of Eqs. (4.31), (4.32)
must be interpreted by means of equations like (4.26), namely, they contain elements
of randomness related both to apparata hidden-variables and, in general, to other un

[1 ,2 ]

1 ,2 ]
1 ,2 ]
known mechanisms: Pi 1 , 2 ; [
= Pi 1 , 2 | [
i
, where 0 Pi 1 ,
i
i

1 ,2 ]
2 | [
1.
i
When 1 = 2 = 0, only the probabilities for the four states of Table 2:

1
[0,0]
P1 0, 0; 1[0,0] = P5 0, 0; 5[0,0] = P9 0, 0; 9[0,0] = P13 0, 0; 13
= ,
4
(4.33)
are non-vanishing. Moreover, for 1 = 2 = 0, four probabilities of our set are still
zero:

]
]
]
]
= P8 , ; [,
= P12 , ; [,
= P16 , ; [,
= 0, (4.34)
P4 , ; [,
4
8
12
16
because of the requirement of perfect anti-correlation on strangeness at equal times.
]
Consider now the contribution to P1 (, ; [,
) coming from the transitions K1 (0)
1
K1 ( ) on the left and K4 (0) K4 ( ) on the right. It can be written in the following two
equivalent ways: (1) the probability that the left going kaon is created in the state K1 and is
then subject to the transition K1 (0) K1 ( ) is p11 ( | 0)/4; in order to obtain the required
probability we have to multiply this quantity by the probability EL ( ) that the right going
kaon at time , which is correlated with the left going K1 , is a K4 not yet decayed; (2) the
probability that the right going kaon is created in the state K4 and is then subject to the
transition K4 (0) K4 ( ) is p44 ( | 0)/4; to obtain the required probability we have to
multiply this quantity by the probability ES ( ) that the left going kaon at time is a K1
not yet decayed. Therefore, the following equality is valid:
p11 ( | 0)EL ( ) = p44 ( | 0)ES ( ),
(4.35)
and from Eq. (4.18) we obtain that it is verified only when ( ) 0. This property can
also be proved starting from the other probabilities of Eqs. (4.31), (4.32) which are nonvanishing when 1 = 2 .
The independent observables relevant for the problem (given, in quantum mechanics, by
Eqs. (3.3)(3.6)) can be written, within local realism, as follows:

0 (2 ) P1 1 , 2 ; [1 ,2 ] + P9 1 , 2 ; [1 ,2 ]
PLR K 0 (1 ), K
1
9

1
= ES (1 )EL (1 ) p22 (2 | 1 ) + p44 (2 | 1 ) ,
4
0

(1 ), K 0 (2 ) P5 1 , 2 ; [1 ,2 ] + P13 1 , 2 ; [1 ,2 ]
PLR K
5
13

1
= ES (1 )EL (1 ) p11 (2 | 1 ) + p33 (2 | 1 ) ,
4

0

[1 ,2 ]
0
PLR K (1 ), K (2 ) P4 1 , 2 ; 4[1 ,2 ] + P12 1 , 2 ; 12

1
= ES (1 )EL (1 ) p12 (2 | 1 ) + p34 (2 | 1 ) ,
4
0

(1 ), K
0 (2 ) P8 1 , 2 ; [1 ,2 ] + P16 1 , 2 ; [1 ,2 ]
PLR K
8
16

1
= ES (1 )EL (1 ) p21 (2 | 1 ) + p43 (2 | 1 ) ,
4

[1 ,2 ]
PLR KL (1 ), KS (2 ) P9 1 , 2 ; 9[1 ,2 ] + P12 1 , 2 ; 12

[1 ,2 ]
1 ,2 ]
+ P13 1 , 2 ; 13
+ P16 1 , 2 ; [
16

1
= ES (1 )EL (1 ) p11 (2 | 1 ) + p12 (2 | 1 )
4

+ p21 (2 | 1 ) + p22 (2 | 1 ) ,

1 ,2 ]
PLR KS (1 ), KL (2 ) P1 1 , 2 ; 1[1 ,2 ] + P4 1 , 2 ; [
4

1 ,2 ]
+ P5 1 , 2 ; 5[1 ,2 ] + P8 1 , 2 ; [
8

1
= ES (1 )EL (1 ) p33 (2 | 1 ) + p34 (2 | 1 )
4

+ p43 (2 | 1 ) + p44 (2 | 1 ) .
503
(4.36)
The probabilities of Eqs. (3.8) and (3.9) do not supply new information since they are
not independent of the other ones just considered, whereas Eq. (3.7), which ensures CP
conservation, was assumed, in Section 2.1 and then in Table 3, when we introduced local
realism for the two-kaon system.
In order to determine the observables of Eq. (4.36), we now ask whether it is possible
to derive useful relations among the pij (2 | 1 )s and the probabilities pij ( | 0) of matrix
(4.18). By introducing three-time probabilities:
pij k (2 , 1 , 0) = pij k (2 , 1 | 0) pk (0) = pij k (2 | 1 , 0) pj k (1 | 0) pk (0),
(4.37)
and using multiplication theorem, p11 (2 | 1 ), p12 (2 | 1 ), p21 (2 | 1 ) and p22 (2 | 1 )

can be written as follows:

1
p11 (2 , 1 ) p11 (2 | 1 )p1 (1 ) = p111 (2 , 1 | 0) + p112 (2 , 1 | 0) ,
(4.38)
4

1
p12 (2 , 1 ) p12 (2 | 1 )p2 (1 ) = p121 (2 , 1 | 0) + p122 (2 , 1 | 0) ,
(4.39)
4

1
p21 (2 , 1 ) p21 (2 | 1 )p1 (1 ) = p211 (2 , 1 | 0) + p212 (2 , 1 | 0) ,
(4.40)
4

1
p22 (2 , 1 ) p22 (2 | 1 )p2 (1 ) = p221 (2 , 1 | 0) + p222 (2 , 1 | 0) .
(4.41)
4
504
In the previous relations, p1 (1 ) = [p11 (1 )+p12 (1 )]/4 = ES (1 )/4 (p2 (1 ) = [p21 (1 )+

p22 (1 )]/4 = ES (1 )/4) is the probability to observe a K1 (K2 ) along direction 1 at time
1 (analogous relations are valid for CP = 1 states, and p3 ( ) = p4 ( ) = EL ( )/4), the
pj k ( | 0)s are given in Eq. (4.18) with ( ) 0, whereas pij (2 , 1 ) denote standard
(namely non-conditional) two-times probabilities. Moreover, pij k (2 , 1 , 0) is the probability to have states Kk , Kj and Ki at times 0, 1 and 2 , respectively, pij k (2 , 1 | 0) is
the probability that at times 1 and 2 the states are Kj and Ki , respectively, if the state at
time 0 was Kk , and, finally, pij k (2 | 1 , 0) is the probability of a Ki at time 2 conditional
on the presence of a Kk at time 0 and a Kj at 1 . It is then clear that, in Eqs. (4.38)(4.41),
the two-times probabilities pij (2 , 1 ) are obtained by summing over the possible states
appearing at time = 0.
Let us now consider probability p11 (2 | 0). Introduce a time 1 in the interval [0, 2 ]:
at instant 1 the state can be either a K1 or a K2 , then the contributions to p11 (2 | 0)
come from two transitions with different intermediate states. They are K1 (0) K1 (1 )
K1 (2 ) and K1 (0) K2 (1 ) K1 (2 ), thus:
p11 (2 | 0) = p111 (2 , 1 | 0) + p121 (2 , 1 | 0),
(4.42)
for any 1 [0, 2 ]. Limiting again the discussion to probabilities relevant for the evolution
of CP = +1 states, one obtains the remaining relations:
p12 (2 | 0) = p112 (2 , 1 | 0) + p122 (2 , 1 | 0),
(4.43)
p21 (2 | 0) = p211 (2 , 1 | 0) + p221 (2 , 1 | 0),
(4.44)
p22 (2 | 0) = p212 (2 , 1 | 0) + p222 (2 , 1 | 0).
(4.45)
Now, the sum of two three-times probabilities corresponding to the same states at times
0 and 1 but with different states at 2 provides a known result; in fact:
p111 (2 | 1 , 0) + p211 (2 | 1 , 0) = ES (2 1 ),
(4.46)
p112 (2 | 1 , 0) + p212 (2 | 1 , 0) = ES (2 1 ),
(4.47)
p121 (2 | 1 , 0) + p221 (2 | 1 , 0) = ES (2 1 ),
(4.48)
p122 (2 | 1 , 0) + p222 (2 | 1 , 0) = ES (2 1 ).
(4.49)
Each of these equalities accounts for the contributions to transitions into final states K1
and K2 once the kaonic states at times 0 and 1 are fixed: these probabilities equal the
probability ES (2 1 ) that a CP = +1 kaon does not decay during the time interval
between 1 and 2 .
By using the shorthand notation pij k pij k (2 , 1 | 0), it follows from (4.18) that the
above Eqs. (4.42)(4.49) can be written in the equivalent form:
p111 + p121 = p222 + p212 = ES (2 )Q+ (2 ),
(4.50)
p112 + p122 = p221 + p211 = ES (2 )Q (2 ),
(4.51)
p111 + p211 = p222 + p122 = ES (2 )Q+ (1 ),
(4.52)
p112 + p212 = p221 + p121 = ES (2 )Q (1 ).
(4.53)
505
These conditions on the 8 CP = +1 three-times probabilities supplies two system of

equations:
p111 + p121 = ES (2 )Q+ (2 ),

p221 + p211 = ES (2 )Q (2 ),
(4.54)
p111 + p211 = ES (2 )Q+ (1 ),
p222 + p212 = ES (2 )Q+ (2 ),

p112 + p122 = ES (2 )Q (2 ),
(4.55)
p222 + p122 = ES (2 )Q+ (1 ),

each containing three independent conditions and four unknown probabilities.
From previous results one obtains:

1 p111 + p112 p121 + p122
+
p11 (2 | 1 ) + p12 (2 | 1 ) =
4
p1 (1 )
p2 (1 )
= ES (2 1 ).
(4.56)
Analogously:
p21 (2 | 1 ) + p22 (2 | 1 ) = p11 (2 | 1 ) + p21 (2 | 1 )
= p12 (2 | 1 ) + p22 (2 | 1 ) = ES (2 1 ),
(4.57)
thus:
p21 (2 | 1 ) = p12 (2 | 1 ),
(4.58)
p22 (2 | 1 ) = p11 (2 | 1 ).
(4.59)
Exactly the same derivation can be repeated for the CP = 1 probabilities: the relations
valid in this case are obtained from (4.38)(4.59) simply by replacing ES with EL and
1 3, 2 4 for the state indexes. Obviously, the normalization of the local realistic

[1 ,2 ]
= 1, is guaranteed by the above
probabilities (4.31) and (4.32), 18
i=1 Pi 1 , 2 ; i
results.
From Eqs. (4.36) and previous analysis one thus obtains the following expression for the
observables within the local realistic approach:

0

(1 ), K 0 (2 )
0 (2 ) = PLR K
PLR K 0 (1 ), K

1
= ES (1 )EL (2 ) + EL (1 )ES (2 ) 1 + ALR (1 , 2 ) ,
8
(4.60)
0
0

(1 ), K
0 (2 )
PLR K (1 ), K 0 (2 ) = PLR K

1
= ES (1 )EL (2 ) + EL (1 )ES (2 ) 1 ALR (1 , 2 ) ,
8
(4.61)

1
PLR KL (1 ), KS (2 ) = EL (1 )ES (2 ),
(4.62)
2

1
PLR KS (1 ), KL (2 ) = ES (1 )EL (2 ),
(4.63)
2
506
written directly in terms of the asymmetry parameter [see definition (3.12)]:

[p11 (2 | 1 ) + p33 (2 | 1 )]
1
ES (2 1 ) + EL (2 1 )
(p111 + p112 )/ES (1 ) + (p333 + p334 )/EL (1 )
=2
1,
ES (2 1 ) + EL (2 1 )
ALR (1 , 2 ) = 2
(4.64)
where, compatibly with constraints (4.54) and (4.55), the three-times probabilities can vary
in the following intervals:

p111
p333
,
Min Q+ (1 ); Q+ (2 ) ,
Max 0; Q+ (2 ) Q (1 )
ES (2 ) EL (2 )

p334
p112
,
Min Q (1 ); Q (2 ) . (4.65)
Max 0; Q (1 ) Q+ (2 )
ES (2 ) EL (2 )
5. Compatibility between local realism and quantum mechanics
From Eqs. (4.29), (4.31), (4.32), (4.56), and (4.57) we obtain that local realism
reproduces the single kaon quantum-mechanical expectation values:

0
0 1
( ) PQM K
( ) = ES ( ) + EL ( ) ,
PLR K
4
0
0 1

PLR K ( ) PQM K ( ) = ES ( ) + EL ( ) ,
4

1
PLR KS ( ) PQM KS ( ) = ES ( ),
2

1

PLR KL ( ) PQM KL ( ) = EL ( ).
(5.1)
2
This result is of general validity [13]: for all EPR-like particle pairs it is always possible to
take into account of the single particle observables by employing a local hidden-variable
model.
Results (4.62) and (4.63) reproduce the quantum-mechanical predictions (3.5) and (3.6);
it is easy to see that expectation values (3.8) and (3.9) are obtained too. The same
conclusion would be true for the joint observables (4.60) and (4.61) involving KS KL
mixing if the time-dependent local realistic asymmetry parameter had the same expression
it has in quantum mechanics. Thus:
Local realism equivalent to quantum mechanics ALR (1 , 2 ) AQM (1 , 2 ).
(5.2)
From Eqs. (4.64) and (4.65) it follows that the asymmetry corresponding to the most
general local realistic theory satisfies the following inequality:

2Q+ (2 ) Q (1 ) 1 ALR (1 , 2 ) 1 2Q+ (2 ) Q+ (1 ).
(5.3)
If one considers the special case in which 2 = 1 , local realism is compatible
Max
with quantum mechanics: in fact, AMin
LR (, ) AQM (, ) = ALR (, ) 1 for all times.
This is obvious, since within the class of local realistic theories supplying asymmetry
507
Fig. 1. Local-realistic and quantum-mechanical asymmetry parameters for 2 = 1.51 plotted vs

1 /S .
parameters in the interval (5.3), a model that reproduces the perfect anti-correlation
properties of the two-kaon state at equal times must exist. When 1 = 0, both descriptions
lead to the same asymmetry: ALR (0, ) = AQM (0, ) Q+ ( ) Q ( ). Another special
case is when, for instance, 2 = 1.51 : in this situation, the local realistic asymmetry does
not satisfy the compatibility requirement (5.2). This is depicted in Fig. 1. The maximum
values of the local realistic asymmetry stands below the quantum-mechanical ones for
0 < 1 2.3S . The largest incompatibility corresponds to 1 1.5S , where [AQM
AMax
LR ]/AQM 20%. In general, local realism and quantum mechanics are incompatible
when 2 = 1 with > 1. The degree of incompatibility increases for increasing . For
instance, when 2 = 21 2.4S , AQM is 27% larger than AMax
LR . The large differences
among quantum-mechanical and local realistic predictions justify our approach, which
neglected CP violation.
However, it is important to stress the following restriction concerning the choice (which
must be at free will) of the detection times 1 and 2 . In order to satisfy the locality
condition, namely to make sure that the measurement event on the right is causally
disconnected from that on the left, these events must be space-like separated. For a twokaon system in which the kaons fly back-to-back in the laboratory frame system, this
requirement corresponds to choose detection times which satisfy the inequality [27]:
1
2 1 + v
= 1.55,
<
1 1 v
(5.4)
where v 0.22 is the kaon velocity (in units of c) in the laboratory frame system.
Nevertheless, concerning the locality assumption, a loophole that is impossible to
avoid could allow, in principle (it is completely unknown, however, in which way),
an information to reach both devices at the instants of measurement, whatever the
508
Table 4
Asymmetry parameter measured by CPLEAR Collaboration [10]
Time difference: 2 1
0
1.37S
Experiment
0.88 0.17
0.56 0.12
Quantum mechanics
Local realism
1
0.64
0.861
0.340.48
choice of these times is. In fact, events in the overlap region of the two backward
light-cones corresponding to the measurements at 1 and 2 might be responsible for
the choice of the times 1 and 2 as well as for the experimental outcomes. If this
were the actual case, even for causally disconnected measurement events one could
not infer that the non-occurrence of action-at-a-distance implies locality. Thus, a nonlocal behaviour of microscopic phenomena could be still compatible with relativistic
causality.
An experiment that measured the asymmetry parameter was performed by the CPLEAR
0 pairs were produced by protonantiproton
Collaboration at CERN [10]. The K 0 K
annihilation at rest, while the kaon strangeness was detected through kaon strong
interactions with bound nucleons of absorber materials. The data, corrected for a
comparison with pure quantum-mechanical predictions (Eq. (3.13)), are reported in
Table 4. The temporal uncertainty of data is not considered here. Asymmetry values
compatible with local realism depend on the detection times 1 and 2 separately: the
CPLEAR set-up corresponds to the following corrected times: 1 = 2 = 0.55S when
2 1 = 0 and 1 = 0.55S , 2 = 1.92S when 2 1 = 1.37S . We notice that also in
the second case the two observation events were space-like separated: in fact, 2 /1 = 3.5
(also when uncorrected times are considered) and, since the kaon velocity in the center of
0 is v 0.85, condition (5.4) gives 1 2 /1 < 12.2. It is
mass system for pp K 0 K
evident from Table 4 that the data are in agreement, within one standard deviation, with
both quantum mechanics and local realism. For a decisive test of local realistic theories
more precise data are needed.
In agreement with Bells theorem, in this section we have seen that local realism
contradicts some statistical predictions of quantum mechanics concerning the evolution
of the two-neutral-kaon system. Local realism has already been tested against quantum
mechanics (by employing Bell-type inequalities) in optics and atomic physics: neglecting
existing loopholes, apart from some irrelevant exception, all the experimental results
revealed incompatible with the local realistic viewpoints and were in good agreement with
quantum mechanics. For the two-kaon correlated system one avoids the detection loophole
and, in particular situations, the differences among the predictions of local realism and
quantum mechanics are so large that a future measurement at the Frascati -factory (say
for 2 = 1.51 , with 1 around 1.5S ) should be able to decide which of the two pictures
holds in Nature.
509
6. On the possibility to test local realism with Bells inequalities for the
two-neutral-kaon system
When CP non-conservation is taken into account, a Bells inequality violated by
quantum mechanics has been derived in the special case of a gedanken experiment [28].
Unfortunately, the magnitude of violation of this inequality is very small, of the order of
0 CP violating parameter, ., thus representing a problem from the experimental
the K 0 K
point of view. A similar inequality, which is violated by a non-vanishing value of the direct
CP and CP T violating parameter, . , is discussed in Ref. [29]. Moreover, experimental
set-up exploiting KS KL regeneration processes have also been proposed in order to
formulate Bells inequalities that show incompatibilities with some statistical predictions
of quantum theory [26,27,30]. Unfortunately, in order to avoid a tiny violation of the
inequalities that one obtains for thin regenerators, this kind of Bell-type test requires large
amount of regenerator materials. Moreover, the test proposed in Ref. [26] can be performed
only at asymmetric -factories.
In Ref. [25] the authors concluded that, under the hypothesis of CP conservation,
because of the specific properties of the kaon, it is impossible to test local realism by
using Bells inequalities, since whatever inequality one considers, a violation by quantummechanical expectation values cannot be found. In this section we consider again this
question in order to prove how such a test is actually feasible with Wigners inequalities
[16]. Moreover, in agreement with the discussion of Ref. [32], we shall also show that a
Bell-test is possible when properly normalized observables and ClauserHorneShimony
Holts (CHSHs) inequalities [15,17] are employed.
The two-kaon system presents some analogies but also a significant difference compared
to the case of the singlet state of two spin-1/2 particles (2.1). The (free) choice of the times
1 and 2 at which a strangeness measurements on the kaon pair (2.7) is performed is
analogous to the (free) choice of the orientation along which the spin is observed in the
case of the singlet state (2.1). If we consider the ideal limit in which the weak interaction
eigenstates KS and KL are stable (L = S = 0), quantum-mechanical kaon probabilities
(3.3) and (3.4) have exactly the same expressions (proportional to 1 cos 12 ) of the
spin-singlet case, provided one replaces the angle between the two spin analyzers with
12 m(2 1 ). Then, a strangeness measurement on the two-kaon system is perfectly
equivalent to a spin measurement on the singlet state (2.1). It is then obvious that if the
above hypothesis were realized in Nature, one could find violations of Bells inequalities
of the same magnitude of the ones that characterize the spin system.
However, this hypothesis is far from being realistic, and the kaon joint probabilities
decreases with time because of the KS and KL weak decays. This leads to an important
difference with respect to the spin case. Because of the particular values of the kaon
lifetimes (S and L ) and of the quantity m mL mS , which controls the quantummechanical interference term (i.e., the KS KL mixing), Ref. [25] concluded that no choice
of the detection times is able to show a violation, by quantum mechanics, of Bells
inequalities. The authors of Ref. [25] reached this conclusion on the basis of CHSHs
inequalities.
510
Actually, the interplay between kaon exponential damping and strangeness oscillations
only makes it more difficult (but not impossible) a Bell-type test. The reason of this
behaviour lies in the very short KS lifetime (S ) compared with the typical time (2/m
13S ) of the strangeness oscillations. The situation would be different (namely the
discrimination between quantum mechanics and local realistic theories would be easier)
0 system (this is due to the fact that the states analogous to KS and
if one treated the B 0 B
KL for the B 0 -meson have the same lifetime).
We start discussing Wigners inequalities. We must recall that these inequalities are
derivable for deterministic theories only, therefore they are less general than CHSHs. Let
us consider the following Wigners inequality involving KS KL mixing:
0
0
0

(1 ), K
(1 ), K
(3 ), K
0 (2 ) PLR K
0 (3 ) + PLR K
0 (2 ) ,
PLR K
(6.1)
0K
0 states are
0 joint detection since K
where 1 3 2 . It has been written for K
easier to detect than K 0 states. Obviously, the same conclusions that we shall obtain in the
following are valid for the inequality corresponding to K 0 K 0 detection. Inequalities that
0 joint probabilities turn out to be useless for Bell-type tests.
contain K 0 K
In the limit S = L = 0, Eq. (6.1) reduces to the analogous inequality for the spinsinglet case:
PLR (sa = , sb = ) PLR (sa = , sc = ) + PLR (sc = , sb = ).
(6.2)
Since:
1
PQM (s = , s = ) = (1 cos ),
(6.3)
4
being the angle between the spin measurement directions characterized by the unitary
inequality (6.2) is violated by quantum mechanics when one chooses
vectors and ,
ab = 2ac = 2cb 2 , with in the interval [0, /2] (see Fig. 2). The greatest violation
of Eq. (6.2) (0.375 > 0.250) is for = /3 and is significant, since it corresponds to
Fig. 2. Violation of Wigners inequality (6.2) for the spin-singlet state (2.1). The function
WQM () PQM (sa = , sb = ) PQM (sa = , sc = ) PQM (sc = , sb = ) is plotted
versus ( ab /2 = ac = cb ). The inequality is violated by quantum mechanics when is in the
interval [0, /2].
511
Fig. 3. Violation of Wigners inequality (6.1) for 1 = 2 /p = 23 /(p + 1) . The function

0 (1 ), K
0 (2 )] PQM [K
0 (1 ), K
0 (3 )] PQM [K
0 (3 ), K
0 (2 )] is plotted
WQM ( ) PQM [K
versus /S . From the top to the bottom the curves correspond to p = 1.5, 1.3, and 1.1, respectively.
PQM (sa = , sb = ) = 0.375 and PQM (sa = , sc = ) = PQM (sc = , sb = ) =

0.125.
Coming back to the two-kaon system in the real case with S /L 579, we must
require the three detection times of inequality (6.1) to satisfy restriction (5.4), dictated
by the necessity to avoid any causal connection between the measurements that could be
present if the two observation events were not be space-like separated. By introducing the
relation:
2 1 = 2(3 1 ) = 2(2 3 ) (p 1),
(6.4)
among the observation times and choosing 1 = , one obtains 2 = p and 3 = (p +

1)/2, and the locality requirement (5.4) is fulfilled when 1 p < 1.55. Apart from the
case with p = 1, for all values of p in this range, inequality (6.1) is incompatible with
quantum mechanics at small : in Fig. 3 this is shown for p = 1.5, 1.3, and 1.1. In
the case with p = 1.5, the largest violation of Eq. (6.1) (0.0051 > 0.0026) corresponds
0 (1 ), K
0 (2 )] = 0.0051, PQM [K
0 (1 ), K
0 (3 )] = 0.0016 and
to 1.6S , PQM [K
0 (3 ), K
0 (2 )] = 0.0010. An experimental test of a tiny violation like this one
PQM [K
is difficult to perform since, relatively to the typical experimental accuracy, the function
WQM ( ) of Fig. 3 is too close to 0 in the region of maximum violation. In particular, a
precision like that of the CPLEAR experiment (see Table 4) is insufficient to achieve this
purpose, since W ( ) would be measured with an error larger than the maximum violation
shown in Fig. 3.
When one considers joint probabilities normalized to undecayed kaon pairs:
0
0
0

0
( ), K
0 ( ) P ren K
0 ( ) P [K ( ), K ( )]
( ), K
P K
P [( ), ( )]

1
= 1 A(, ) ,
4
(6.5)
512
since these quantities are less damped than the original ones, a Bell-type test can be
performed also with CHSHs inequalities. In the previous equation, the probability that
at times (on the left) and (on the right) both kaons are undecayed is:

0

0

0 ( ) + P K
0 ( )
( ), K
( ), K 0 ( ) + P K 0 ( ), K
P ( ), ( ) = P K

+ P K 0 ( ), K 0 ( )

1
= ES ( )EL ( ) + EL ( )ES ( ) ,
(6.6)
2
the last equality being valid both in the local realistic description (Eqs. (4.60), (4.61)) and in
0 oscillations.
quantum mechanics (Eqs. (3.3), (3.4)), since it is independent of the K 0 K
The same derivation that supplies the CHSHs inequality in the unrenormalized case can be
applied to the renormalized observables of Eq. (6.5). By introducing four detection times
(1 and 2 for the left going meson, 3 and 4 for the right going meson), the CHSHs
inequality for strangeness 1 detection is then:
1 SLR (1 , 2 , 3 , 4 ) 0,
(6.7)
with:
SLR (1 , 2 , 3 , 4 )

ren 0
ren 0
ren 0
0 (3 ) PLR
0 (4 ) + PLR
0 (3 )
K (1 ), K
K (1 ), K
K (2 ), K
PLR

0
0
ren 0
0 (4 ) P ren K
(2 ) P ren K
(3 ) ,
K (2 ), K
+ PLR
(6.8)
LR
LR
where the single meson observables are given by (see Eq. (5.1)):
0 ( )] 1

PLR [K
ren 0
PLR
K ( )
= .
PLR [( )]
2
(6.9)
Consider the special case in which the four times are related by:
1
3 1 = 2 3 = 4 2 = (4 1 ) .
3
Thus, in quantum mechanics quantity (6.8) reduces to (see Eq. (6.5)):
(6.10)

1
2 3AQM ( ) + AQM (3 ) 1.
(6.11)
4
If we choose 1 , the other times become: 2 = 3 , 3 = 2 and 4 = 4 , and, in
the limit of stable kaons, both side of inequality (6.7) are violated by quantum mechanics
in periodical intervals of (see curve marked spin in Fig. 4): the largest violations are:
1.21 < 1, 0.21 > 0.
As far as the real case for kaons is considered, quantum mechanics does not violate
inequality (6.7) when unrenormalized expectation values are used (see curve unren
in Fig. 4). The conclusion is different once one employs probabilities normalized to
undecayed kaon pairs: as it is shown in Fig. 4 (curve ren), for 0 < 1.4S quantummechanical expectation values are incompatible with the left hand side of inequality (6.7).
The largest violation of the inequality (1.087 < 1) corresponds to 0.81S and
ren [K
0 (1 ), K
0 (3 )] 0.036, P ren [K
0 (1 ), K
0 (4 )] 0.195.
PQM
QM
SQM ( ) =
513
Fig. 4. Violation of CHSHs inequality (6.7) for 1 /p = 2 /(p + 2) = 3 /(p + 1) = 4 /(p + 3) .

The function SQM of Eq. (6.11) is plotted versus . The curve unren corresponds to the
case employing unrenormalized probabilities and p = 1: the inequality is not violated by
quantum-mechanical observables. Values of p compatible with the locality assumption (p > 5.45)
provide unrenormalized functions SQM more damped. Both curves valid in the limit S = L = 0
(spin) and in the real case of unstable kaons with probabilities normalized to undecayed kaons (ren)
violate CHSHs inequality and are independent of p.
With the previous choice of the four detection times the locality condition (5.4) is
not satisfied, since: 4 /1 = 4 > 1.55. In order to fulfil this requirement when relation
(6.10) is used, one can introduce times 1 = p , 2 = (p + 2) , 3 = (p + 1) and 4 =
(p + 3) (p 0) and require 4 /1 = (p + 3)/p < 1.55, thus p > 5.45. However, since
the renormalized quantum-mechanical probabilities only depend on the difference between
the observation times (see Eqs. (6.5), (3.13)), the result ren of Fig. 4 is independent of p,
and the locality condition is satisfied. Thus, experimentally one could choose to use, for
instance, p = 6, namely, 1 = 6 , 2 = 8 , 3 = 7 , 4 = 9 , and the largest violation
of the inequality would be again for 0.81S . However, as p increases, even if the
renormalized probabilities are unchanged, the strangeness detection becomes more and
more difficult, because of the kaon decays, thus small p are preferable. Also the curve
corresponding to the limit S = L = 0 is the same for any p. Of course, the curve
corresponding to the inequality that makes use of unrenormalized probabilities depends
on p, but this case is not interesting since it cannot be used for a discriminating test
whatever the choice of p is.
Assuming the same relative error P ren /P ren in the measurement of all joint probabilities entering quantity S of Eq. (6.8) and disregarding the uncertainties on the single kaon
detection, the need for an error on S much smaller than the maximum violation of Fig. 4
514
(S 0.087) is satisfied if P ren /P ren 40%. 1 Practically, with a detection accuracy

P ren /P ren = 510%, a clear test with CHSHs inequality (6.7) would be possible. The
CPLEAR data did not fulfil this condition. The experimental accuracy required to perform
a conclusive test of local realism with CHSHs inequality (6.7) is of the same order of
magnitude of that needed in the use of Wigners inequality (6.1). To give an idea of the
comparison between the potentialities of a test with Bells inequalities and a test through
the measurement of the asymmetry parameter, let us consider the situation of Fig. 1 with
0 joint
2 = 1.51 = 2.25S . In this case, the requirement P ren /P ren 40% for K 0 K
detection corresponds to an accuracy on the asymmetry A/A 6%, 2 which would allow a clear test of local realism. In conclusion, an experiment measuring the asymmetry
parameter seems easier to perform than a test with Bells inequalities.
We conclude with a remark concerning the negligible role, in the measurement of
0 produced in the decay of the -meson (here, an
the asymmetry, of final states K 0 K
undetectable (soft) photon is produced together with the kaon pair). Indeed, on the one
0 has not been observed until now. Theoretical estimations
hand, the decay K 0 K
0 ) of the order of 108 . On the other hand,
[34] supply a branching ratio BR( K 0 K
0 final states: it corresponds to two
a different decay mechanisms could contribute to K 0 K
0
0
and a0 K 0 K
0 . However, the BRs measured
channels, f0 K K
for f0 and a0 are small and, besides, the subsequent decays f0 (980)
0 and a0 (980) K 0 K
0 are expected to be very unprobable for kinematical reasons.
K 0K
1 Indeed, employing the standard formula for the error propagation:
S( ) =
!

0 (1 ), K
0 (3 )] 2 + P ren [K
0 (1 ), K
0 (4 )] 2
3P ren [K
(6.12)
(see Eqs. (6.8) and (6.10)), assuming:

0 ( ), K
0 (3 )]
0 ( ), K
0 (4 )]
P ren
P ren [K
P ren [K
ren 0 1
= ren 0 1
,
(1 ), K
0 (3 )]
(1 ), K
0 (4 )]
P ren
P [K
P [K
(6.13)
and using the quantum-mechanical values corresponding to the maximum discrepancy between quanren [K
0 (1 ), K
0 (3 )] = 0.036 and
tum mechanics and local realism showed in Fig. 4 [ = 0.81S , PQM
ren
0
0

P [K (1 ), K (4 )] = 0.195], from previous two equations one easily obtains that the requirement S
QM
0.087 corresponds to P ren /P ren 40%.

2 From Eq. (6.5), the asymmetry parameter is obtained as:

0
0 (2 ) ,
(1 ), K
A(1 , 2 ) = 1 4P ren K
(6.14)
therefore,

0
A(1 , 2 )
4
0 (2 ) .
(1 ), K
=
P ren K
A(1 , 2 )
A(1 , 2 )
(6.15)
ren [K
0 (1 ), K
0 (2 )] = 0.031, therefore, when P ren /P ren 40%,
When 1 = 1.5S and 2 = 1.51 , one has PQM
the asymmetry is measured with a precision A/A 6%.
515
7. Conclusions
In agreement with Bells theorem, in this paper we have shown that quantum mechanics
for the two-neutral-kaon system cannot be completed by a theory which is both local and
realistic: the separability assumed in Bells local realistic theories for the joint probabilities
contradicts the non-separability of quantum entangled states. Although both the locality
condition and the realistic viewpoint seem reasonable, they are not prescribed by any
first principle. Any local realistic approach is only able to reproduce the non-paradoxical
predictions of quantum mechanics like the perfect anti-correlations in strangeness and
CP and the single meson observables. On this point it is important to recall that the
authors of Ref. [33] showed how for entangled systems of three or more particles the
incompatibility between local realism and quantum mechanics is even deeper: in fact, for
these systems, a contradiction already arises at the level of perfectly correlated quantum
states, the premises of local realism being in conflict with the non-statistical predictions
of quantum mechanics. For EPRs pairs, maintaining the realistic viewpoint, in order to
reproduce the prediction of quantum mechanics (which, up to now, have been strongly
supported by experimental evidence), one is forced to consider as a real fact of Nature a
non-local behaviour of microscopic phenomena.
In the present paper, the incompatibility proof among quantum mechanics and local
realistic models has been carried out by employing two different approaches. We
started discussing the variability of the expectation values deduced from the general
premises concerning locality and realism. The realistic states have been interpreted
within the widest class of hidden-variable models. As far as the process e+ e
0 is considered, under particular conditions for the experimental parameters
K 0K
(the detection times), the discrepancies among quantum mechanics and local realistic
models for the time-dependent asymmetry are not less than 20%. The data collected by
0 ) do not allow for
the CPLEAR Collaboration (which used the reaction pp K 0 K
conclusive answers concerning a refutation of local realism: these data are compatible
not only with quantum-mechanical asymmetries, but with the range of variability of local
realistic predictions. Therefore, a decisive test of local realism needs for more precise
data.
The other approach we followed in this paper makes use of Bell-like inequalities
involving KS KL mixing. Contrary to what is generally believed in the literature, we
have shown that a Bell-type test is feasible at a -factory, both with Wigners and
(once probabilities normalized to undecayed kaons are used) CHSHs inequalities. For an
experiment at a -factory, the degree of inconsistency between quantum mechanics and
local realism shown by a Bell test is of the same order of magnitude of that obtained in the
first part of the paper through the comparison of the asymmetry parameters.
Concluding, by employing an experimental accuracy for joint kaon detection considerably higher than that corresponding to the CPLEAR measurement, a decisive test of local
realism vs quantum mechanics both with and without the use of Bells inequalities will be
feasible in the future at the Frascati -factory.
516
Acknowledgements
We are grateful to Albert Bramon for many valuable discussions. We also acknowledge
the hospitality of Prof. D. Sherrington of the Department of Theoretical Physics of Oxford
University. One of us (G.G.) acknowledge financial support by the EEC through TMR
Contract CEE-0169.
References
[1] A. Einstein, B. Podolsky, N. Rosen, Phys. Rev. 47 (1935) 777.
[2] N. Bohr, Phys. Rev. 48 (1935) 696.
[3] D. Bohm, Phys. Rev. 85 (1952) 166;
D. Bohm, Phys. Rev. 85 (1952) 180.
[4] H. Everett, Rev. Mod. Phys. 29 (1957) 454.
[5] A. Shimony, in: Proceedings of the International Symposium on Foundation of Quantum
Mechanics, Physical Society of Japan, Tokyo, 1984, p. 25.
[6] W.H. Furry, Phys. Rev. 49 (1936) 393;
W.H. Furry, Phys. Rev. 49 (1936) 476.
[7] C.S. Wu, I. Shaknov, Phys. Rev. 77 (1950) 136.
[8] D. Bohm, Quantum Theory, PrenticeHall, New York, 1951, p. 614.
[9] D. Bohm, Y. Aharanov, Phys. Rev. 108 (1957) 1070.
[10] A. Apostolakis et al., Phys. Lett. B 422 (1998) 339.
[11] D. Bouwmeester, J.-W. Pan, M. Daniell, H. Weinfurther, A. Zeilinger, Phys. Rev. Lett. 82
(1999) 1345.
[12] W. Tittel, J. Brendel, H. Zbinden, N. Gisin, Phys. Rev. Lett. 81 (1998) 3563;
W. Tittel, J. Brendel, H. Zbinden, N. Gisin, Phys. Rev. A 59 (1999) 4150.
[13] J.S. Bell, Physics 1 (1964) 195.
[14] J.S. Bell, in: B. dEspagnat (Ed.), Proceedings of the International School of Physics Enrico
Fermi, Course XLIX, Academic, New York, 1971, p. 171;
Reprinted: J.S. Bell, in: Speakable and Unspeakable in Quantum Mechanics, Cambridge Univ.
Press, Cambridge, 1987, p. 29.
[15] J.F. Clauser, M.A. Horne, A. Shimony, R.A. Holt, Phys. Rev. Lett. 23 (1969) 880.
[16] E.P. Wigner, Am. J. Phys. 38 (1970) 1005.
[17] J.F. Clauser, M.A. Horne, Phys. Rev. D 10 (1974) 526.
[18] J.F. Clauser, A. Shimony, Rep. Prog. Phys. 41 (1978) 1881.
[19] A. Aspect, P. Grangier, G. Roger, Phys. Rev. Lett. 47 (1981) 460;
A. Aspect, P. Grangier, G. Roger, Phys. Rev. Lett. 49 (1982) 91;
A. Aspect, J. Dalibard, G. Roger, Phys. Rev. Lett. 49 (1982) 1804.
[20] G. Weihs, T. Jennewein, C. Simon, H. Weinfurter, A. Zeilinger, Phys. Rev. Lett. 81 (1998) 5039.
[21] E. Santos, Phys. Rev. A 46 (1992) 3646;
E. Santos, Phys. Lett. A 212 (1996) 10;
L. De Caro, A. Garuccio, Phys. Rev. A 46 (1994) R2803.
[22] P. Huet, M.E. Peskin, Nucl. Phys. B 434 (1995) 3.
[23] V.L. Lepore, F. Selleri, Found. Phys. Lett. 3 (1990) 203.
[24] G. Pancheri (Ed.), Proceedings of the Workshop on Physics and Detectors for Da6ne, INFN,
LNF, 1991;
L. Maiani, G. Pancheri, N. Paver (Eds.), The Da6ne Handbook, INFN, LNF, 1992;
L. Maiani, G. Pancheri, N. Paver (Eds.), The Second Da6ne Physics Handbook, INFN, LNF,
1995.
517
[25] G.C. Ghirardi, R. Grassi, T. Webern, in: Proceedings of the Workshop on Physics and Detectors
for Da6ne, G. Pancheri (Ed.), INFN, LNF, 1991, p. 261;
G.C. Ghirardi, R. Grassi, R. Regazzon, in: L. Maiani, G. Pancheri, N. Paver (Eds.), The Da6ne
Handbook, INFN, LNF, 1992, p. 283.
[26] P.H. Eberhard, Nucl. Phys. B 398 (1993) 155;
P.H. Eberhard, in: L. Maiani, G. Pancheri, N. Paver (Eds.), The Second Da6ne Physics
Handbook, INFN, LNF, 1995, p. 99.
[27] A. Di Domenico, Nucl. Phys. B 450 (1995) 293.
[28] F. Uchiyama, Phys. Lett. A 231 (1997) 295.
[29] F. Benatti, R. Floreanini, Phys. Rev. D 57 (1998) R1332.
[30] A. Bramon, M. Nowakowski, Phys. Rev. Lett. 83 (1999) 1;
B. Ancochea, A. Bramon, M. Nowakowski, Phys. Rev. D 60 (1999) 094008.
[31] F. Selleri, Phys. Rev. A 56 (1997) 3493;
R. Foadi, F. Selleri, Phys. Lett. B 461 (1999) 123;
R. Foadi, F. Selleri, Phys. Rev. A 61 (2000) 012106.
[32] N. Gisin, A. Go, Am. J. Phys. 69 (2001) 246.
[33] D.M. Greenberger, M.A. Horne, A. Shimony, A. Zeilinger, Am. J. Phys. 58 (1990) 1131.
[34] A. Bramon, A. Grau, G. Pancheri, Phys. Lett. A 283 (1992) 416;
A. Bramon, A. Grau, G. Pancheri, Phys. Lett. A 289 (1992) 97.

Thermal production of gravitinos

M. Bolz, A. Brandenburg, W. Buchmller
Deutsches Elektronen-Synchrotron DESY, Hamburg, Germany
Received 8 December 2000; accepted 9 March 2001
Abstract
We evaluate the gravitino production rate in supersymmetric QCD at high temperature to leading
order in the gauge coupling. The result, which is obtained by using the resummed gluon propagator,
depends logarithmically on the gluon plasma mass. As a byproduct, a new result for the axion
production rate in a QED plasma is obtained. The implications for the cosmological dark matter
problem are briefly discussed, in particular the intriguing possibility that gravitinos are the dominant
part of cold dark matter. 2001 Elsevier Science B.V. All rights reserved.
1. Introduction
Supersymmetric theories, which contain the standard model of particle physics and
gravity, predict the existence of the gravitino [1], a spin- 23 particle which acquires a mass
from the spontaneous breaking of supersymmetry. Since the couplings of the gravitino
with ordinary matter are strongly constrained by local supersymmetry, processes involving
gravitinos allow stringent tests of the theory.
It was realized long ago that standard cosmology requires gravitinos to be either very
light, mG
< 1 keV [2], or very heavy, mG
> 10 TeV [3]. These constraints are relaxed if the
standard cosmology is extended to include an inflationary phase [4,5]. The cosmologically
relevant gravitino abundance is then created in the reheating phase after inflation in which
a reheating temperature TR is reached. Gravitinos are dominantly produced by inelastic
2 2 scattering processes of particles from the thermal bath. The gravitino abundance is
essentially linear in the reheating temperature TR .
The gravitino production rate depends on mg /mG
, the ratio of gluino and gravitino
masses. The ten 2 2 gravitino production processes were considered in [5] for mg
mG
. The case mg mG
, where the goldstino contribution dominates, was considered in
[6]. Four of the ten production processes are logarithmically singular due to the exchange
of massless gluons. As a first step this singularity can be regularized by introducing either
E-mail address: brandenb@mail.desy.de (A. Brandenburg).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 3 2 - 8
M. Bolz et al. / Nuclear Physics B 606 (2001) 518544
519
a gluon mass or an angular cutoff [5]. The complete result for the logarithmically singular
part of the production rate was obtained in [7]. The finite part depends on the cutoff
procedure.
To leading order in the gauge coupling the correct finite result for the gravitino
production rate can be obtained by means of a hard thermal loop resummation. This has
been shown by Braaten and Yuan in the case of axion production in a QED plasma [8].
The production rate is defined by means of the imaginary part of the thermal axion selfenergy [9]. The different contributions are split into parts with soft and hard loop momenta
by means of a momentum cutoff. For the soft part a resummed photon propagator is used,
and the logarithmic singularity, which appears at leading order, is regularized by the plasma
mass of the photon. The hard part is obtained by computing the 2 2 scattering processes
with momentum cutoff. In the sum of both contributions the cutoff dependence cancels and
the finite part of the production rate remains. For the gravitino production rate the soft part
has been considered in [10] and the expected logarithm of the gluon plasma mass has been
obtained.
Constraints from primordial nucleosynthesis imply an upper bound on the gravitino
number density which subsequently yields an upper bound on the allowed reheating
temperature TR after inflation [1115]. Typical values for TR range from 107 1010 GeV,
although considerably larger temperatures are acceptable in some cases [16]. In models of
baryogenesis where the cosmological baryon asymmetry is generated in heavy Majorana
neutrino decays [17], temperatures TR 108 1010 GeV are of particular interest [18].
Further, it is intriguing that for such temperatures gravitinos with mass of the electroweak
scale, i.e., mG
100 GeV can be the dominant component of cold dark matter [7]. In all
these considerations the thermal gravitino production rate plays a crucial role. In this paper
we therefore calculate this rate to leading order in the gauge coupling, extending a previous
result [7] and following the procedure of Braaten and Yuan [8].
The paper is organized as follows. In Section 2 we summarize some properties of
gravitinos and their interactions which are needed in the following. In order to illustrate
how the hard thermal loop resummation is incorporated we first discuss the axion case
in Section 3. The most important intermediate steps and the final result for the gravitino
production rate are given in Section 4. Using the new results the discussion in [7] on
gravitinos as cold dark matter is updated in Section 5, which is followed by an outlook
in Section 6. The calculation of the hard momentum contribution to the production rates is
technically rather involved. We therefore give the relevant details in the appendices.
2. Gravitino interactions
In the following we briefly summarize some properties of gravitinos which we shall need
in the following sections. More detailed discussions and references can be found in [19
21].
Gravitinos are spin-3/2 particles whose properties are given by the Lagrangian for the
vector-spinor field (x),
520

1
1
1
L = 5 mG
S.
(1)
,
2
4
2M
1/2 is the Planck mass and S is the
Here mG
is the gravitino mass, M = (8GN )
supercurrent corresponding to supersymmetry transformations. and S are Majorana

fields, so that S = S .
Free gravitinos satisfy the RaritaSchwinger equation,

1
1
5 mG
(2)
, = 0,
2
4
which, using
(x) = 0,
(x) = 0,
(3)
reduces to the Dirac equation

(i/
mG
) (x) = 0.
(4)
Consider as matter sector first a non-abelian supersymmetric gauge theory with

Lagrangian

1 a a 1 a
L = F
(5)
F
+ i Dab mg ab b
4
2
for the vector boson Aa and the gluino a . Supersymmetry is explicitly broken by the
gluino mass term. Hence, the supercurrent is not conserved,

i
a
S = , a F
4

1
a
= mg , a F
4
= mg S.
(6)
The calculation of the gravitino production rate in Section 4 will involve squared matrix
elements which are summed over all four gravitino polarizations. The corresponding
polarization tensor for a gravitino with momentum P reads

(P ) =
l (P ) l (P )
l

P P
P
P
1
= (P
/ + mG) g 2 + (P/ mG ) + .
3
mG
mG
mG

(7)
Since (x) is a solution of the RaritaSchwinger equation one has for the polarization
tensor
(P ) = 0,
P (P ) = 0,
/ m) (P ) = 0.
(P
(8)
(9)
We shall be interested in the production of gravitinos at energies much larger than the
gravitino mass. In this case the polarization tensor simplifies to
2 P P
/ g + P/ 2 ,
(P ) P
3 m
G
(10)
521
where we have used S = 0. Clearly, the first term corresponds to the sum over the
helicity 32 states and the second term represents the sum over the helicity 12 states
which represent the goldstino
part of the gravitino. Eq. (10) is the basis for the familiar
substitution rule 23 m1 which is used to obtain the effective Lagrangian

G
describing the interaction of goldstinos with matter [2224].
The gravitino production rate at finite temperature can be expressed in terms of the
imaginary part of the gravitino self-energy [9] which takes the form,

(P ) = tr (P )G
(P )

1
2 tr (P )S (P ) S (P )
M
2m2

1
/ )S (P ) S (P ) + 2 g 2 tr P/ S(P ) S(P ) .
2 tr (P
(11)
M
3mG
M
Here we have used Eq. (6) for the divergence of the supercurrent. Note, that F = 3 mG
M
is the scale of spontaneous supersymmetry breaking, which gives the strength of the
goldstino coupling to the supercurrent. The dots denote the sum over the contributions
to the self-energy in the loop expansion. In S (P ) = 4i [ , ] a (P1 )F (K1 ) and
S(P ) = 14 [ , ]a (P1 )F (K1 ), with P = P1 + K1 , a (P1 ) represents one end of an
internal gluino line and F (K1 ) stands for the end of one or two internal gluon lines. For
mG
mg the goldstino part dominates the gravitino production cross section.
Eq. (11) is useful to derive relations between the helicity 32 and the helicity 12
contributions to the self-energy. As shown in Appendix A, one obtains to two-loop order

m2g
g2
(P ) 2 1 +
(12)
.
M
3m2G

We shall exploit this fact to perform the hard thermal loop resummation, which is necessary
because of the infrared divergences, just for the helicity 12 part of the rate.
The full supercurrent also involves quarks and squarks in addition to gluons and
gluinos. The divergence of the additional part of the supercurrent that involves the cubic
gravitinoquarksquark coupling is again proportional to the parameter of supersymmetry
breaking, i.e., m2q , the squark mass squared. The corresponding goldstino contribution
to the production rate is then proportional to m4q , which is suppressed at high energies
compared to the gluino contribution for dimensional reasons.
3. Axion production
Let us now consider the thermal production of axions in a relativistic QED plasma of
electrons and photons. According to the procedure outlined in the introduction the thermal
production rate can be obtained as sum of two terms, a soft momentum contribution which
is extracted from the axion self-energy evaluated with a resummed photon propagator and
a hard momentum contribution which is computed from the 2 2 processes.
522
Fig. 1. Axion self energy; the blob denotes the resummed photon propagator.
The axionphoton interaction is described by the effective Lagrangian

L=
1
,
aF F
4f
(13)
where f is the axion decay constant. The axion self-energy a (P ) (cf. Fig. 1) can be
evaluated in the imaginary-time formalism for external momentum P = (p0 , p), with p0 =
i2nT and p = |p|. In covariant gauge the resummed photon propagator has the form [25,
26]
i (K) = i(A T + B L + C ),
(14)
with the tensors

1 2
K v v K v(K v + K v ) + K K ,
2
k

K v
K v 2
= v v
(K v + K v ) +
K K ,
K2
K2
K K
=
,
(K 2 )2
A = g
B
C
(15)
and the transverse and longitudinal propagators

1
,
k02 k 2 T (k0 , k)
1
L (k0 , k) = 2
.
k L (k0 , k)
T (k0 , k) =
(16)
Here K = (k0 , k) with k0 = i2nT and k = |k|, is a gauge-fixing parameter, v is the

velocity of the thermal bath and T /L are the transverse and longitudinal self-energies of
the photon. The corresponding propagators T /L have the spectral representation
T /L (k0 , k) =
1
L/T (, k).
k0
For || < k the spectral densities L/T are given by [26],

T (, k) =
x
3
,
4m2 (1 x 2 )(AT (x)2 + (z + BT (x))2 )
(17)
L (, k) =
2x
3
,
2
2
4m AL (x) + (z + BL (x))2
where m = eT /3 is the plasmon mass of the photon, x = /k, z = k 2 /m2 and

x2
3
3
1+x
AT (x) = x,
BT (x) =
2
,
+ x ln
4
4 1 x2
1x

3
3
1+x
BL (x) =
AL (x) = x,
2 x ln
.
2
2
1x
523
(18)
(19)
The contribution to the axion production rate from soft virtual photons is obtained by
analytically continuing the axion self energy function a (P ) from the discrete imaginary
value p0 to the continuous real value E = p [8],

Im a (E + i9, p)
asoft (E) =

E
k<kcut

kcut
k
d
2
2 2
3
L (, k) 1 2 + T (, k) 1 2
.
dk k
k
k
0
k
(20)
The axion production rate depends logarithmically on kcut . The corresponding coefficient
can be obtained analytically. The remaining constant has to be evaluated numerically. This
yields the result, first obtained in [8],

2
3m2 T
kcut
soft
ln 2 1.379 .
a (E) =
(21)
16f 2
m
T
=
8f 2
The corresponding collision term in the Boltzmann equation is
d 3p
Casoft (T ) =
nB (E)asoft (E)
(2)3

e2 (3)T 6
kcut
=
ln
0.689 ,
24 3 f 2
m
(22)
where
nB (E) =
1
exp (E/T ) 1
(23)
is the BoseEinstein distribution.

The dependence of asoft (E) on the cutoff kcut is cancelled by the contribution from hard
virtual photons to the self-energy. This part of the axion production rate can be obtained
directly from the processes e e a (cf. Fig. 2) [9],

3
dp 1
d 3 pi
hard
nB (E)a (E) = 2
(2)4 4 (P1 + P2 P P3 )
4 2E
(2)3 2Ei
i=1
nF (E1 )nB (E2 )(1 nF (E3 ))|M|2 (|p1 p3 | kcut ).
(24)
524
Fig. 2. Axion production in electronphoton scattering.
Here we have used rotational invariance by averaging over the directions of the
axion momentum; |M|2 is the photonaxion matrix element squared for e (P1 ) (P2 )
e (P3 )a(P ),

2s 2
e2
2
2s t ,
|M| = 2
(25)
t
f
where s = (P1 + P2 )2 , t = (P1 P3 )2 , and nF (E) is the FermiDirac distribution,
nF (E) =
1
.
exp(E/T ) + 1
(26)
The phase space integration has to be carried out under the constraint on the virtual photon
momentum k |p1 p3 | > kcut . For the angular integrations it turns out to be convenient
to define all momenta with respect to k. Some details of this calculation are given in
Appendix B. One finally obtains,
nB (E)ahard (E)

1
3e2
= 8 3 2 2 dE1 dE3 dk nF (E1 )nB (E2 )(1 nF (E3 )) (E1 E3 )2 k 2
2 f E

2 E12 + E32 + 2EE2 (E3 + E1 )2 (E + E2 )2
1 +
(27)
,
3
k2
k4
where E2 = E + E3 E1 and the integrations are restricted by ,

= (k kcut )(k |E1 E3 |)(E1 + E3 k)(2E + E3 E1 k)
(E1 )(E3 )(E + E3 E1 ).
(28)
After performing the k-integration one is left with several domains for the E1 - and E3 integrations. The logarithmic dependence on kcut can be extracted by means of a partial
integration in E1 . In the remaining part of the integral kcut can be set equal to zero. The
final result reads
ahard (E)

2 2 3
(2)
e2
2T
17
T
+
=
ln
+
16 3 f 2
3
kcut
6
(2)
+ e
E/T

|E1 E3 |
dE1 ln
nF (E1 )nB (E2 )
E3
E+E
dE3 [1 nF (E3 )]
0
525

E32 E 2 nF (E1 ) + nB (E2 )
E22
E12
2
(E E1 ) 2 2E1 2 + E1
E
E2
T
E22

nF (E1 ) + nB (E2 )
E 2 + E32 2
E2
+ E1 + E32
(E1 E3 ) 22 2E1 2 1
E2
T
E

2
nF (E1 ) + nB (E2 )
+ (E3 E1 ) 2E1 + E1 + E32
.
T
(29)
This result agrees with the one obtained in [8] except for the first expression (E E1 )
in the double integral. Integration over the axion energy E yields for the collision term in
the Boltzmann equation
d 3p
hard
nB (E)ahard (E)
Ca (T ) =
(2)3

(2)
e2 (3)T 6
2T
17
1.280
.
=
(30)
ln
+
24 3f 2
kcut
6
(2)
The numerical constant is about 20% smaller than the one obtained from the axion rate
given in [8].
Consistency requires that the dependence on the cutoff kcut cancels in the total
production rate. Comparison of Eqs. (22) and (29) shows that this is indeed the case. The
result for the total axion collision term reads
2

T
e2 (3)T 6
soft
hard
ln
+ 0.8194 .
Ca (T ) = Ca (T ) + Ca (T ) =
(31)
48 3 f 2
m2
4. Gravitino production
The rate for the thermal production of gravitinos can be calculated in complete analogy
to the axion production rate. It is dominated by QCD processes since the strong coupling is
considerably larger than the electroweak couplings. The contribution due to soft virtual
gluons can again be extracted from the gravitino self-energy with a resummed gluon
propagator to which the contribution from hard 2 2 processes has to be added.
Properties and interactions of the gravitino have been discussed in Section 2. For
supersymmetric QCD with gluons, gluinos, quarks and squarks one obtains [1921],

i
D PL (D )PR
L =
2M

i
a
, a F
.
8M
(32)
526
Here denotes a left-handed quark or antiquark and the corresponding squark. For
light gravitinos one can use a simpler effective Lagrangian [2224]. The corresponding
goldstinogluongluino coupling can be read off from Eqs. (6) and (11)

mg
a
, a F
+ .
Leff =
2 6 MmG
(33)
Here is the goldstino, the spin-1/2 component of the gravitino. The effective theory
contains the same vertices as the full theory, except for the gravitinoquarksquarkgluon
vertex. Instead, there is a new four particle vertex, the gravitinogluinosquarksquark
vertex [24]. All vertices are proportional to supersymmetry breaking mass terms, i.e., m2q
and mg . At high energies and temperatures, with mq , mg T , contributions involving the
cubic goldstinoquarksquark coupling are suppressed by m2q /T 2 relative to the gluino
contribution because of the higher mass dimension of the coupling.
In the imaginary-time formalism one obtains for the goldstino self-energy (cf. Fig. 3)
with momentum P summed over helicities:

l
l
(P ) (P )G
(P ) = tr
(P )
l=1/2
=T
k0

d 3k
1
tr P
/ V2 (K)V1 .
(2)3
Q
/
(34)
Here we have neglected gluino and gravitino masses since mq , mg T ; V1,2 are the
vertices, Q = P K is the momentum of the gluino, and (K) is the resummed
gluon propagator, which is obtained from the resummed photon propagator (14) by the
substitution m mg . The thermal gluon mass for N colours and nf colour triplet and
anti-triplet chiral multiplets is given by
m2g =
g2 T 2
(N + nf ).
6
(35)
This result is easily obtained from the expressions for the gluon vacuum polarization [27]
by adding up the contributions from gluons, gluinos, quarks and squarks.
Fig. 3. Gluongluino loop diagram, the leading contribution to the imaginary part of the gravitino
self energy. The blob denotes a resummed gluon propagator.
527
Inserting gluon propagator and vertices in Eq. (34) yields the gauge-independent result
2

4 mg T 2
1
d 3k
(P ) =
(36)
1
(DL L + DT T ) 2 ,
N
2
3
2
3 M m
(2)
Q
k0
where

1
tr P
/ [K,
/ ] Q[
/ K,
/ ]A ,
32

1
tr P
/ [K,
/ ]Q[
/ K,
/ ]B .
DL (k0 , k, E, p, pk) =
(37)
32
After a straightforward calculation, analogous to the one for the axion self-energy, one
finds for the gravitino production rate

Im (E + i9, p)
soft
G
(E) =

E
DT (k0 , k, E, p, pk) =
k<kcut
m2g (N 2 1)T
6M 2 m2G
kcut
dk k
0
k
3
k

2
2 2
L (, k) 1 2 + T (, k) 1 2
(38)
.
k
k
The momentum integral depends logarithmically on the cutoff kcut . The integrand, which
is identical with the one for the axion production rate (20), agrees with the result obtained
in [10]. After performing the momentum integrations one finally obtains

2
(N 2 1)m2g m2g T kcut
soft
G
(39)
ln
1.379
.
(E)
=

m2g
4M 2 m2
G
Note, that the overall normalization differs from the expression given in [10] by the factor
4(N 2 1).
The dependence of the soft part of the gravitino production rate on the cutoff kcut is again
cancelled by the cutoff dependence of the contribution from the hard 2 2 processes.
There are ten processes denoted by A to J [5]:

A: g a + g b g c + G
(crossing of A)
B: g a + g b g c + G
a

C: qi + g qj + G
528
(crossing of C)
D: g a + qi qj + G
(crossing of C)
E: qi + qj g a + G

F: g a + g b g c + G

G: qi + g a qj + G

H: qi + g a qj + G
(crossing of G)
I: qi + qj g a + G
a
(crossing of H)
J: qi + qj g + G
The corresponding matrix elements have been evaluated in [7]. As discussed in Section 2,
they must have the form
|Mi |2

m2g
1
1
+
M2
3m2G
(40)
in the high energy limit.

In Table 1 the squared matrix elements of all ten processes are listed. Sums over initial
and final spins have been performed. For quarks and squarks the contribution of a single
chirality is given. One easily checks that the matrix elements satisfy the relevant crossing
symmetries. The particle momenta P1 , P2 , P3 , and P used in the calculations correspond
to the particles in the order in which they are written down in the column process i of
Table 1. This fixes the energies of Bose and Fermi distribution. The matrix elements in the
table correspond to the definitions s = (P1 + P2 )2 and t = (P1 P3 )2 .
The different processes fall into three classes depending on the number of bosons and
fermions in initial and final state. A, C and J are BBF processes with two bosons in the
initial and a fermion in the final state; correspondingly, B, D, E and H are BFB processes,
and F, G and I are FFF processes. Only four processes, B, F, G and H contribute to the
529
Table 1
production in two-body processes involving left-handed
Squared matrix elements for gravitino (G)
a
quarks (qi ), squarks (qi ), gluons (g ) and gluinos (g a ). The values are given for the specified choice
of colors and summed over spins in the initial and final state. f abc and Tjai are the usual SU(3)
colour matrices
Process i
2
m2
|Mi |2 g 2 1 + g2

g a + g b g c + G
4(s + 2t + 2 ts )|f abc |2

g a + g b g c + G
a

qi + g qj + G
4(t + 2s + 2 st )|f abc |2

2t|Tjai |2

g a + qi qj + G
a

q i + qj g + G

g a + g b g c + G
+t )
abc |2
8 (s st+st
(s+t ) |f

qi + g a qj + G
2
4(s + st )|Tjai |2

qi + g a qj + G

qi + qj g a + G
2(t + 2s + 2 st )|Tjai |2
2
4(t + ts )|Tjai |2

qi + q j g a + G
2(s + 2t + 2 ts )|Tjai |2
C
D
3mG
2s|Tjai |2
2t|Tjai |2
2
2 2
logarithmic cutoff dependence. The gravitino production rate is then given by (cf. [9]),

3
3p
d
d
1
p
i
hard
nF (E)G
(2)4 4 (P1 + P2 P P3 )
(E) =
4 2E
(2)3 2Ei
i=1

nBBF |MBBF |2 + nBFB |MBFB |2 + nF F F |MF F F |2
(|p1 p3 | kcut ).
(41)
Here, nBBF , nBFB and nF F F are the products of number densities for the corresponding
processes, e.g.,

nBBF = nB (E1 )nB (E2 ) 1 nF (E3 ) .
(42)
The matrix elements |MBBF |2 etc. are obtained by summing the corresponding matrix
elements in Table 1 with the appropriate multiplicities and statistical factors. Angular and
momentum integrations can now be carried out as in the case of axion production. One
finally obtains the result

m2g g 2 (N 2 1)
hard
G
(E)
=
1
+

8 3 M 2
3m2G
530

(2)
2T
17
+
+
2 2 (N + nf )T 3 ln
kcut
6
(2)
E+E

3

|E1 E3 |
+ (N + nf ) eE/T + 1
dE3
dE1 ln
E3
0

2 2
E1 E2
d
2
(nBFB + nF F F )
(E E1 )
E3
dE1
E2

E22 2
d
2
(E1 E3 )
(nBFB + nF F F ) 2 E1 + E3
dE1
E

2
d
2
(nBFB + nF F F ) E1 + E3
+ (E3 E1 )
dE1

+ IBBF + IBFB + IF F F .
(43)
Performing the differentiations with respect to E1 yields expressions analogous to the one
given in Eq. (29). IBBF , IBFB and IF F F , which are not all proportional to N + nf , are
hard
given in Appendix C; they contribute to the cutoff-independent part of G
.
The dependence on kcut cancels in the sum of soft and hard contributions to the
production rate. From Eqs. (39) and (43) one obtains for the collision term
soft

d 3p
hard
nF (E) G
(E) + G
(E)
(2)3

m2g 3 (3)g 2 (N 2 1)T 6
= 1+
32 3 M 2
3m2
CG
(T ) =
2

T
ln
+ 0.3224 (N + nf ) + 0.5781nf .
m2g
(44)
This is the main result of this paper. It allows to calculate the gravitino abundance to leading
order in the gauge coupling g(T ), contrary to previous estimates which depended either on
ad hoc cutoffs [5,6] or on an unknown scale of the logarithmic term [7].
An important question concerns the size of higher-order corrections. Note, that g 0.85
for T 1010 GeV. This is much better than at the electroweak scale T 100 GeV where
g 1.2, or for the quarkgluon plasma at T 1 GeV where g 2.5. However, one still
has to worry about the usually assumed separation of scales g 2 T gT T , which would
correspond to g mg T , where g g 2 T is the magnetic screening mass. Note, that
for the supersymmetric standard model with N = 3 and nf = 6 one has mg T . For the
static Debye and magnetic screening masses the separation of scales has recently been
studied in detail for the case of non-supersymmetric QCD [28]. For real-time processes
almost nothing is presently known about non-perturbative effects related to the magnetic
screening mass. This is a challenging theoretical problem.
531
5. Gravitinos as cold dark matter

We can now study the cosmological implications of our result Eq. (44) for the Boltzmann
collision term of gravitino production. We are particularly interested in the case of large
reheating temperatures after inflation, i.e., TR 108 1010 GeV, which are relevant for
models of leptogenesis. In the following we shall concentrate on the possibility that the
gravitino is the lightest supersymmetric particle (LSP), updating the discussion in [7],
where it was pointed out that a large gravitino mass mG
100 GeV is compatible with
such reheating temperatures. We shall ignore the non-thermal production of gravitinos [29,
30] which depends on the model of inflation.
From the Boltzmann equation,
dnG

+ 3H nG
(45)
= CG
,
dt
one obtains for the gravitino abundance at temperatures T < TR , assuming constant
entropy,
YG
(T ) =
nG
CG
gHS (T )
(T )
(TR )

,
nrad (T ) gHS (TR ) H (TR )nrad (TR )
(46)
where gHS (T ) is the number of effectively massless degrees of freedom [31]. For T <
1 MeV, i.e., after nucleosynthesis, gHS (T ) = 43/11, whereas gHS (TR ) = 915/4 in the
supersymmetric standard model. With H (T ) = (gH (T ) 2 /90)1/2T 2 /M one obtains in the
case of light gravitinos (mG
mg (), 100 GeV) from Eqs. (46) and (44) for the
gravitino abundance and for the contribution to h2 ,

TR
100 GeV 2 mg () 2
10
,
YG
(47)
= 1.1 10
1010 GeV
mG
1 TeV

2
2 1
G
h = mG
YG
(T )nrad (T )h c

mg () 2
TR
100 GeV
= 0.21
.
1010 GeV
mG
1 TeV
(48)
Here we have used g(TR ) = 0.85, nrad (T ) = (3)T 3 / 2 , and mg (T ) = g 2 (T )/

g 2 ()mg (); c = 3H02 M 2 = 1.05h2 105 GeV cm3 is the critical energy density.
2
The new result for G
h is smaller by a factor of 3 compared to the result given in [7].
Due to the large value of the plasma mass mg an estimate of the gravitino production rate,
which is based just on the logarithmic term of the 2 2 cross sections as in [7] is rather
uncertain.
2
It is remarkable that reheating temperatures TR 108 1010 GeV lead to values G
h =
0.01 . . .1 in an interesting gravitino mass range. This is illustrated in Fig. 4 for a gluino
mass mg = 700 GeV. As an example, for TR 1010 GeV, mG
80 GeV and h 0.65
=
0.35,
which
agrees
with
recent
measurements
of M [31]. In general,
[31] one finds G

to find a viable cosmological scenario one has to avoid two types of gravitino problems:
for unstable gravitinos their decay products must not alter the observed abundances of
light elements in the universe, which is referred to as the big bang nucleosynthesis (BBN)
532
2
Fig. 4. The density parameter G
h for different gravitino masses mG
as function of the reheating
temperature TR . The gluino mass has been set to mg = 700 GeV.
constraint. For stable gravitinos this condition has to be met by other superparticles, in
particular the next-to-lightest superparticle (NSP), which decay into gravitinos; further,
the contribution of gravitinos to the energy density of the universe must not exceed the
closure limit, i.e., G
= G
/c < 1. Consider first the constraint from the closure limit.
The condition G
= YG
mG
nrad /c 1 yields an allowed region in the mG
mg plane
which is shown in Fig. 5 for three different values of the reheating temperature TR . The
allowed regions are below the solid lines, respectively.
With respect to the BBN constraint, consider a nonrelativistic particle X decaying into
electromagnetically and strongly interacting relativistic particles with a lifetime X . X
decays change the abundances of light elements the more the longer the lifetime X and
the higher the energy density mX YX nrad are. These constraints have been studied in detail
by several groups [1113]. They rule out the possibility of unstable gravitinos with mG

100 GeV for TR 1010 GeV.
For stable gravitinos the NSP plays the role of the particle X. The lifetime of a fermion
decaying into its scalar partner and a gravitino is
NSP = 48
2
m2G
M
m5NSP
(49)
For a sufficiently short lifetime, NSP < 2 106 s, the energy density which becomes free in
NSP decays is bounded by mX YX < 4 1010 GeV, which corresponds to X h2 < 0.008.
The lifetime constraint yields a lower bound on superparticle masses which is represented
by the dashed line in the mG
mNSP/g plane in Fig. 5.
533
Fig. 5. Upper and lower bounds on the gluino mass and the NSP mass as functions of the gravitino
mass. The full lines represent the upper bound on the gluino mass mg > mNSP for different reheating
temperatures from the closure limit constraint. The dashed line is the lower bound on mNSP which
follows from the NSP lifetime.
In order to decide whether the second part of the BBN constraint, NSP h2 < 0.008,
is satisfied, one has to specify which particle is the NSP. The case of a higgsino-like
neutralino as NSP has been discussed in [7]. A detailed discussion of the case where a
scalar -lepton is the NSP has been given in [14,15].
A complete treatment of gravitinos as cold dark matter has to include non-thermal
contributions. The situation is analogous to leptogenesis where, in principle, non-thermal
contributions also have to be added to the thermal part. However, non-thermal contributions
depend on assumptions about the state of the early universe before the hot thermal phase,
for instance the type of inflationary phase, and they are therefore strongly model dependent.
6. Outlook
The main result of the paper is the production rate of gravitinos for supersymmetric
QCD at high temperature to leading order in the gauge coupling. The result is valid for
gravitino masses larger or smaller than the gluino mass.
As expected the gravitino production rate depends logarithmically on the gluon plasma
mass which regularizes an infrared divergence occurring in leading order. Following the
procedure of Braaten and Yuan, the result is obtained by matching contributions to the
gravitino self-energy with soft and hard internal gluon momenta and by using a resummed
gluon propagator for the soft part. As a byproduct a new result for the axion production rate
in a QED plasma is obtained which is slightly smaller than a previously published result.
534
The QCD coupling is large, and even at temperatures T 1010 GeV the usually assumed
separation of scales, g 2 T gT T appears problematic. Hence, higher-order corrections
to the gravitino production rate may be sizeable. Further, it is of crucial importance to gain
some understanding of the influence of the non-perturbative magnetic mass scale g on
real-time processes in general.
The thermal gravitino production rate plays a central role in cosmology since it is closely
related to the dark matter problem. For many supersymmetric extensions of the standard
model this rate defines a limiting temperature beyond which the standard hot big bang
picture becomes inconsistent. At present supersymmetric theories offer several interesting
candidates for cold or hot dark matter. It is an intriguing possibility that the gravitino itself
is the dominant component of cold dark matter.
Acknowledgements
We would like to thank T. Asaka, O. Br, D. Bdeker, O. Philipsen and M. Plmacher
for helpful discussions. The work of A.B. has been supported by a Heisenberg grant of the
D.F.G.
Appendix A
In the following we shall derive the prefactor of the self-energy,

m2g
,
(P ) 1 +
3m2G
(A.1)
extending the discussion in Section 2.

The gravitino self-energy takes the form (cf. Eq. (11)),
1
M2
1
2
M
(P )

tr (P )S(P ) S (P )

a
tr (P
/ ) , a (P1 )F
(K1 ) Fb (K1 ) b (P1 ) ,
2m2g
2
3m2G
M

a

tr P
/ , (P1 )F (K1 ) F (K1 ) a (P1 ) , .
(A.2)
The one- and two-loop contributions for the pure gauge theory in resummed perturbation
theory are depicted in Fig. 6. The contributions (a) and (b) represent for the gluino line the
first two terms of the gluino propagator. This corresponds to the substitution,
(P1 ) (P1 ) P
/ 1 A(P1 , v) + v/B(P1 , v).
(A.3)
This is the general form of the gluino propagator for mg = 0 because of chiral symmetry
and the fact that the velocity v appearing in the gluon propagator is the only other vector
535
Fig. 6. Contributions to the gravitino self energy.
available apart from the momentum P1 . With

= 2
(A.4)
one then reads off the factor (A.1) for the contributions (a) and (b). The same arguments
apply for the contribution Fig. 6(d).
For Fig. 6(c) one obtains for the gluino line,
/ 1 P/ 2 .
(P1 ) (P1 ) P
(A.5)
/ 1 P/ 2 = 2P/ 2 P/ 1 ,
P
(A.6)
With
the interchange P1 P2 and the property of the gluon propagator (K) = (K) one
obtains the factor (A.1) also for this contribution.
Appendix B
In this appendix we explain the calculation of the contribution of hard virtual photons to
the axion production rate ahard (E). We start by reconsidering the defining equation (24):

3
dp 1
d 3 pi
hard
(2)4 4 (P1 + P2 P P3 )
nB (E)a (E) = 2
4 2E
(2)3 2Ei
i=1
FBF
2
ntotal |M| (|p1 p3 | kcut ),
e a
where the matrix element squared for

nFBF
total = nF (E1 )nB (E2 ) 1 nF (E3 ) .
(B.1)
is given in Eq. (25) and

(B.2)
536
A great simplification is achieved in the computation of (B.1) if one uses as reference

momentum the difference vector k = p1 p3 , i.e., we write

d 3 p1
= P12 (E1 ) dE1 d 3 p1
2E1

= d 3 k 3 (k + p3 p1 ) P12 (E1 ) dE1 d 3 p1

= E12 |k + p3 |2 (E1 ) dE1 d 3 k.
(B.3)
Further,
d 3 p2 4
(P1 + P2 P P3 )
2E2

= P22 (E2 ) d 4 P2 4 (P1 + P2 P P3 )

= (E + E3 E1 )2 (p k)2 (E + E3 E1 ).
(B.4)
We now use rotational invariance to choose

k = k(0, 0, 1),
p = E(0, sin , cos ),
p3 = E3 (cos sin , sin sin , cos ),
(B.5)
which implies
s = (P1 + P2 )2 = (P + P3 )2 = 2EE3 (1 sin sin sin cos cos ),
t = (P1 P3 )2 = (E1 E3 )2 k 2 ,
(B.6)
|k + p3 |2 = E32 + k 2 + 2E3 k cos ,
|p k|2 = E 2 + k 2 2Ek cos .
(B.7)
and
It follows that

(E + E3 E1 )2 (p k)2

E 2 + k 2 (E + E3 E1 )2
1
cos
=
,
2kE
2kE

E 2 E32 k 2
1
cos 1
E12 |k + p3 |2 =
.
2kE3
2kE3
(B.8)
The integrations over the -functions yield the following -functions (where we use also
the -functions (E1 ), (E + E3 E1 ) and E = p > 0, E3 = p3 > 0:
(1) From the integration over cos we get:
cos < 1
k > E1 E3 ,
cos > 1
E1 > |E3 k|.
(B.9)
537
The second of these constraints is equivalent to

E3 E1 < k < E1 + E3 .
(B.10)
(2) From the integration over cos we get:

cos < 1
cos > 1
|E k| < E + E3 E1 ,
k > E3 E1 .
(B.11)
The first of these constraints is equivalent to

E1 E3 < k < 2E + E3 E1 .
(B.12)
After integrating out the -functions we therefore have:
1 1
2
nB (E)ahard (E) = 7 4 2 dE1 dE3 nFBF
(B.13)
total dk d |M| ,
2 E
where is the product of all -functions that restrict the integrations over E1 , E3 and k,
= (k kcut )(k |E1 E3 |)
(E1 + E3 k)(2E + E3 E1 k)
(E1 )(E3 )(E + E3 E1 ).
(B.14)
Since only s depends on , we can integrate out also this angle without difficulty:

e2
2s 2
2
2s t
d |M| = 2 d
t
f

3e2
=
(E1 E3 )2 k 2
2
2f

2 E12 + E32 + 2EE2 (E3 + E1 )2 (E + E2 )2
g.
1+
3
k2
k4
(B.15)
We thereby obtain the result given in Eq. (27) in the main text. We now rewrite the
expression for (B.14) using
(E1 + E3 k) = 1 (k E1 E3 ).
We use (k E1 E3 )(k |E1 E3 |) = (k E1 E3 ) and thus get

= (k kcut )(k |E1 E3 |)(2E + E3 E1 k)

(k kcut )(k E1 E3 )(2E + E3 E1 k)
(E1 )(E3 )(E + E3 E1 ).
(B.16)
(B.17)
We multiply the second term in the brackets of Eq. (B.17) with 1:

1 = (kcut E1 E3 ) + (E1 + E3 kcut ),
(B.18)
and note that

(k kcut )(k E1 E3 )(kcut E1 E3 )
= (k kcut )(kcut E1 E3 ),
(B.19)
538
and
(k kcut )(k E1 E3 )(E1 + E3 kcut )
= (k E1 E3 )(E1 + E3 kcut ).
(B.20)
The contribution from the first term on the r.h.s. of Eq. (B.18) is zero in the limit kcut 0.
We see this by integrating over k from kcut to 2E + E3 E1 . The resulting expression
3 , 1/k 1 . Since from k
has terms 1/kcut
cut > E1 + E3 it follows that both E1 and E3
cut
are smaller than kcut it is easy to see by power counting that the expression after the k
integration is of order kcut . Then we are left to consider:
nB (E)ahard (E) = g1 + g2 ,
(B.21)
where
1 1
g1 = 7 4 2 dE3 dE1 nFBF
total (E + E3 E1 )
2 E
0
0
dk (k kcut )(k |E1 E3 |)(2E + E3 E1 k)g,
1 1
g2 = 7 4 2 dE3 dE1 nFBF
total (E + E3 E1 )(E1 + E3 kcut )
2 E
0
0
(B.22)
dk (k E1 E3 )(2E + E3 E1 k)g,
The integral over k in g2 is nonzero only if
E1 + E3 < 2E + E3 E1 E1 < E.
(B.23)
In the limit kcut 0 we therefore get:

1
e2
g2 =
3
2
16 f E 2
E
dE1 nFBF
total (E1 E)[(E + E1 )E3 + E1 (E E1 )],
dE3
0
(B.24)
We rewrite this result for later use as follows:

1
e2
g2 =
3
2
16 f E 2

|E1 E3 |
dE1 ln
E3
E+E
dE3
0

d FBF 2 2
(E E1 )
ntotal E1 E2 E 2 E32 .
dE1
We now turn towards the computation of g1 . We first multiply by 1:
g1 = g1 (kcut |E1 E3 |) + g1 (|E1 E3 | kcut ) g11 + g12 .
(B.25)
(B.26)
Note that
(k kcut )(k |E1 E3 |)(kcut |E1 E3 |)
= (k kcut )(kcut |E1 E3 |)
(B.27)
539
and
(k kcut )(k |E1 E3 |)(|E1 E3 | kcut )
= (k |E1 E3 |)(|E1 E3 | kcut ).
(B.28)
We therefore have
1 1
g11 = 7 4 2 dE3 dE1 nFBF
total (E + E3 E1 )(kcut |E1 E3 |)
2 E
0
0
dk (k kcut )(2E + E3 E1 k)g,

(B.29)
1 1
g12 = 7 4 2 dE3 dE1 nFBF
total (E + E3 E1 )(|E1 E3 | kcut )
2 E
0
0
dk (k |E1 E3 |)(2E + E3 E1 k)g.
(B.30)
Consider first g11 . The integration of g over k can be carried out easily. We do not write
3 and 1/k . The
down the result explicitly but note that it contains terms 1/kcut
cut
integration over E1 is done next. From kcut > |E1 E3 | we get
E3 kcut < E1 < E3 + kcut .
In the limit kcut 0 we can therefore set E1 = E3 in the distribution functions
get
e2
g11 =
nB (E)
3 3 f 2
(B.31)
nFBF
total
and
dE3 E32 nF (E3 )(1 nF (E3 ))

0
e2 T 3
nB (E).
=
18f 2
Now we turn towards the computation of g12 . We insert
1 = (E1 E3 ) + (E3 E1 ).
(B.32)
(B.33)
Then g12 = g121 + g122 with
1 1
g121 = 7 4 2 dE3 dE1 nFBF
total (E + E3 E1 )(E1 E3 kcut )
2 E
0
0
dk (k E1 + E3 )(2E + E3 E1 k)g,
1 1
g122 = 7 4 2 dE3 dE1 nFBF
total (E3 E1 kcut )
2 E
0
0
dk (k E3 + E1 )(2E + E3 E1 k)g.
(B.34)
540
The integration over k gives

1
e2
g121 =
16 3 f 2 E 2
E
3 +E
dE1 nFBF
total
dE3
E3 +kcut
1
e2
g122 =
16 3 f 2 E 2
(E12 + E32 )E22

,
E1 E3
E3
kcut
dE1 nFBF
total
dE3
0
E 2 (E12 + E32 )
.
E1 E3
(B.35)
The logarithmic dependence on kcut is extracted by a partial integration with f (E1 ) =

1/(E1 E3 ), f (E1 ) = ln(|E1 E3 |/E3 ).
For the surface term we get:
e2
gsurface = 3 2 nB (E)
4 f

E32 exp(E3 /T )
kcut
dE3 ln
E3 (exp(E3 /T ) + 1)2

(2)
3
2T
,
+
+
=
n
(E)
ln
(B.36)
B
kcut
2
(2)
24f 2

In the remaining term, which is given by dE1 f (E1 )g (E1 ), kcut can be set to zero.
Writing g12 = gsurface + gpartial we obtain:
e2 T 3

1
e2
|E1 E3 |
gpartial =
dE3 dE1 (E + E3 E1 ) ln
E3
16 3f 2 E 2
0
0

d FBF 2 2
(E1 E3 )
ntotal E2 E1 + E32
dE1

d FBF 2 2
(E3 E1 )
ntotal E E1 + E32 .
dE1
(B.37)
Performing the differentiation and combining the results for g1 and g2 leads to the final
result Eq. (29) given in the main text.
Appendix C
In this appendix we describe in some detail the calculation of the hard virtual gluon
contribution and of the other non-singular contributions to the gravitino production
hard
rate G
(E). We start by considering the defining equation (41). By summing the
corresponding squared matrix elements of Table 1 with the appropriate multiplicities and
statistical factors, we get

m2g 2g 2 (N 2 1)
2
|MBBF | = 1 +
M2
3m2G

2t 2
s + 2t +
(C.1)
(N + nf ) + 2snf ,
s
541

m2g 4g 2 (N 2 1)
|MBFB |2 = 1 +
M2
3m2G

2s 2
t 2s
(C.2)
(N + nf ) 2tnf ,
t

m2g 4g 2 (N 2 1)
s2
t2
s2
+
(N + nf ).
2s
|MF F F |2 = 1 +
t
t +s
s
M2
3m2G

(C.3)
First we note that since s = t u we may write s + 2t = t u and
s2
s2 s2
s2
+
= .
(C.4)
t
s+t
t
u
The difference t u and 1/t 1/u is odd under exchanging P1 and P2 . If the remaining
integrand and the measure is even under this transformation, the integral over such terms
will be zero. Therefore in |MBBF |2 , the contribution of s + 2t will give zero. Further we
may trade s in |MBBF |2 with 2t. In |MF F F |2 , we may likewise substitute
s2
2s 2
s2
+
.
t
s+t
t
Therefore only the following squared matrix elements have to be considered:
|M1 |2 = t 2s
(C.5)
2s 2
,
t
|M2 |2 = t,
t2
|M3 |2 = ,
s
and we replace the matrix elements in (41) by

m2g 4g 2 (N 2 1)

|MBBF |2 1 +
|M3 |3 (N + nf ) 2|M2 |2 nf ,
2
2
M
3mG

m2g 4g 2 (N 2 1)

|M1 |2 (N + nf ) 2|M2 |2 nf ,
|MBFB |2 = 1 +
2
2
M
3mG

2
mg 4g 2 (N 2 1)

|MF F F |2 1 +
|M1 |2 |M3 |2 (N + nf ).
M2
3m2
(C.6)
(C.7)
(C.8)
(C.9)
|M1 is the axion matrix element which has been discussed in Appendix B. The different
statistical factors in the case of gravitino production do not change the structure of
the contribution from |M1 |2 as compared to the axion case. Again we can extract the
logarithmic dependence on the cutoff kcut by a partial integration. In the case of BFB, the
surface term contains an additional term which depends on the energy of the gravitino, see
Eq. (C.14) below. The other two matrix elements do not induce a logarithmic dependence
on the cutoff kcut , i.e., one can set kcut = 0 to compute their contribution to the gravitino
production rate. The contribution from |M2 |2 = t can be obtained easily using the same
|2
542
methods as in the axion case, where now no partial integration is needed. We obtain

3
dp 1
d 3 pi
t
(2)4 4 (P1 + P2 P P3 )
IBBF (BFB) =
4 2E
(2)3 2Ei
i=1
nBBF (BFB) |M2 |2
E+E
3
1
=
dE3
dE1 nBBF (BFB)
96 3
0
0

E E1 2
2E + (3E3 E1 )(E + E1 )
(E E1 )
2
E
E2
(E1 E3 ) 22 (2E E3 + E1 )
E

+ (E3 E1 )(3E3 + 3E1 2E) .
(C.10)
To compute the contribution from |M3 |2 = ts it is convenient to choose different

coordinates to perform the angular integrations, namely
q p + p3 = q(0, 0, 1),
p = E(0, sin , cos ),
p2 = E2 (cos sin , sin sin , cos ).
(C.11)
hard
The calculation of this contribution to G
(E) then goes along similar lines as for the
axion, i.e., the integration of the angular variables cos , cos can be trivially performed
using the -functions. This leads to several constraints for the integration over q, which
can be performed without problems. The final result is rather compact:

3
dp 1
d 3 pi
t 2 /s
IBBF (F F F ) =
(2)4 4 (P1 + P2 P P3 )
4 2E
(2)3 2Ei
i=1
nBBF (F F F ) |M3 |2
1
=
32 3
E+E
dE3
0
dE2 nBBF (F F F )
0

E22
E3 E2
+ (E2 E3 )
[E
(E
E
)
+
E(E
+
E
)]
.
3 3
2
3
2
E + E3
E2
(C.12)
hard
The full result for G
(E) can be written as in Eq. (43) with

t 2 /s
t
,
IBBF = 32 3 eE/T + 1 (N + nf )IBBF 2nf IBBF

2
E/T

3
IBFB = T (N + nf ) Li2 e
1 + 8 ln(2)
6

t
,
64 3 nf eE/T + 1 IBFB
(C.13)
(C.14)

t 2 /s
IF F F = 32 3 eE/T + 1 (N + nf )IF F F .
543
(C.15)
The hard contribution to the collision term can be obtained by a numerical integration.
Adding all the contributions we finally find:
d 3p
hard
hard
CG
(T
)
=
nF (E)G

(E)
(2)3

m2g 3 (3)g 2 (N 2 1)T 6
= 1+
32 3 M 2
3m2G

2
T
+ 1.7014 + 0.5781nf ,
(N + nf ) ln 2
(C.16)
kcut
which yields, after adding the soft contribution, our final result for the gravitino collision
term (44).
References
[1] D.Z. Freedman, P. van Nieuwenhuizen, S. Ferrara, Phys. Rev. D 13 (1973) 3214;
S. Deser, B. Zumino, Phys. Lett. B 62 (1976) 335.
[2] H. Pagels, J.R. Primack, Phys. Rev. Lett. 48 (1982) 223.
[3] S. Weinberg, Phys. Rev. Lett. 48 (1982) 1303.
[4] M.D. Khlopov, A.D. Linde, Phys. Lett. B 138 (1984) 265.
[5] J. Ellis, J.E. Kim, D.V. Nanopoulos, Phys. Lett. B 145 (1984) 181.
[6] T. Moroi, H. Murayama, M. Yamaguchi, Phys. Lett. B 303 (1993) 289.
[7] M. Bolz, W. Buchmller, M. Plmacher, Phys. Lett. B 443 (1998) 209.
[8] E. Braaten, T.C. Yuan, Phys. Rev. Lett. 66 (1991) 2183.
[9] H.A. Weldon, Phys. Rev. D 28 (1983) 2007.
[10] J. Ellis, D.V. Nanopoulos, K.A. Olive, S.-J. Rey, Astropart. Phys. 4 (1996) 371.
[11] J. Ellis, G.B. Gelmini, J.L. Lopez, D.V. Nanopoulos, S. Sarkar, Nucl. Phys. B 373 (1992) 399.
[12] M. Kawasaki, T. Moroi, Progr. Theor. Phys. 93 (1995) 879.
[13] E. Holtmann, M. Kawasaki, K. Kohri, T. Moroi, Phys. Rev. 60 (1999) 023506.
[14] T. Gerghetta, G.F. Giudice, A. Riotto, Phys. Lett. B 446 (1999) 28.
[15] T. Asaka, K. Hamaguchi, K. Suzuki, Phys. Lett. B 490 (2000) 136.
[16] T. Asaka, T. Yanagida, hep-ph/0006211.
[17] M. Fukugita, T. Yanagida, Phys. Lett. B 174 (1986) 45.
[18] For a recent review and references, see W. Buchmller, M. Plmacher, Neutrino masses and the
baryon asymmetry, hep-ph/0007176.
[19] J. Wess, J. Bagger, Supersymmetry and Supergravity, Princeton University Press, Princeton,
New Jersey, 1992.
[20] T. Moroi, Ph.D. thesis, hep-ph/9503210.
[21] S. Weinberg, The Quantum Theory of Fields, Vol. III, Cambridge University Press, Cambridge,
2000.
[22] P. Fayet, Phys. Lett. B 84 (1979) 421.
[23] T.E. Clark, S.T. Love, Phys. Rev. 54 (1996) 5723.
[24] T. Lee, G.-H. Wu, Phys. Lett. B 447 (1999) 83.
[25] V.P. Silin, Sov. Phys. JETP 11 (1960) 1136;
O. Kalashnikov, V.V. Klimov, Sov. J. Nucl. Phys. 31 (1980) 699;
V.V. Klimov, Sov. Phys. JETP 55 (1982) 199;
544
[26]
[27]
[28]
[29]
[30]
[31]
H.A. Weldon, Phys. Rev. D 26 (1982) 1394;

H.A. Weldon, Ann. Phys. 271 (1999) 141.
R.D. Pisarski, Physica A 158 (1989) 146.
E. Braaten, R.D. Pisarski, Nucl. Phys. B 337 (1990) 569.
A. Hart, M. Laine, O. Philipsen, Nucl. Phys. B 586 (2000) 443.
R. Kallosh, L. Kovman, A. Linde, A. van Proeyen, Phys. Rev. D 61 (2000) 103503.
G.F. Giudice, A. Riotto, I. Tkachev, JHEP 9911 (1999) 036.
Review of Particle Physics, Eur. Phys. J. C 15 (2000) 1.
Nuclear Physics B 606 [PM] (2001) 547582

Supersymmetric CalogeroMoserSutherland
models and Jack superpolynomials
Patrick Desrosiers a , Luc Lapointe b , Pierre Mathieu c,
a Dpartement de Physique, Universit Laval, Qubec, Canada, G1K 7P4
b Department of Mathematics and Statistics, McGill University, Montral, Qubec H3A 2K6, Canada
c Dpartement de Physique, Universit Laval, Qubec, Canada, G1K 7P4
Received 29 March 2001; accepted 23 April 2001
Abstract
A new generalization of the Jack polynomials that incorporates fermionic variables is presented.
These Jack superpolynomials are constructed as those eigenfunctions of the supersymmetric extension of the trigonometric CalogeroMoserSutherland (CMS) model that decomposes triangularly
in terms of the symmetric monomial superfunctions. Many explicit examples are displayed. Furthermore, various new results have been obtained for the supersymmetric version of the CMS models:
the Lax formulation, the construction of the Dunkl operators and the explicit expressions for the conserved charges. The reformulation of the models in terms of the exchange-operator formalism is a
crucial aspect of our analysis. 2001 Elsevier Science B.V. All rights reserved.
PACS: 11.10.Lm; 11.30.Pb; 03.65.Fd
Keywords: CalogeroMoserSutherland model; Supersymmetry; Jack polynomials; Lax formalism; Dunkl
operators
1. Introduction
CalogeroMoserSutherland (CMS) models [13] have been studied extensively in
the decade following their discovery. Apart from the pioneer works, devoted mainly to
the study of the energy spectrum, the ground-state wave functions and their correlators,
roughly speaking, the initial interest of this first wave of activity was mainly centered
around their integrability and the formulation of their various (Lie algebraic) extensions
(see, e.g., [4]). This subject was developed in parallel to the soliton theory. 1 Although the
* Corresponding author. Tel: 418-656-3416, fax: 418-656-2040.
E-mail addresses: pdesrosi@phy.ulaval.ca (P. Desrosiers), lapointe@scylla.math.mcgill.ca (L. Lapointe),

pmathieu@phy.ulaval.ca (P. Mathieu).
1 Curiously, soliton equations were first defined classically while mechanical models were initially formulated
as quantum systems.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 2 0 8 - 5
548
P. Desrosiers et al. / Nuclear Physics B 606 [PM] (2001) 547582
structure of both classes of models is quite different, they have some common properties,
the existence of a Lax formalism being a good example. There are however deeper and
curious connections: for instance, it has been observed that the time evolution of the
movable poles of KdV rational solutions is governed by a dynamics of the CMS type
(with a 1/r 2 potential) [5].
The renewal in the interest for the CMS models that occurred in the nineties has
many sources. One of it is rooted in the interest for systems having fractional statistics
and the realization that the particles subject to CMS dynamics obey fractional statistics
(see, for instance, [68]). This motivation was triggered by two important problems in
condensed matter. The first is the quantum Hall effect; it has been suggested in the
mid-eighties that the quasiparticles obey fractional statistics (cf., for instance, [9], and
see [10] for a relation between the quantum Hall effect and CMS models). The early
nineties brought a second and somewhat stronger motivation in relation with high-Tc
superconductivity and its possible realization as a gas of anyons (see, e.g., [11] and [12]).
The formulation of Haldanes generalized Pauli principle [13] has also motivated further
theoretical considerations on the issue of fractional statistics, this time for one-dimensional
systems.
Another discovery of the late eighties which also partly accounts for the revival of the
CMS models is that of a new integrable spin-chain model with long-range interaction, the
HaldaneShastry model [14,15]. This model proves to have quite remarkable properties,
among which a Yangian symmetry [16]. It turns out to be closely related to the
trigonometric CMS (tCMS) model. Such a connection, already observed in the original
papers, has been made precise by Polychronakos in the context of its seminal formulation
of the exchange-operator formalism [17]. He showed that the HaldaneShastry spin model
can be recovered from the tCMS model augmented with spin degrees of freedom by
freezing the dynamical degrees of freedom, thereby fixing the sites of the chains at the
minima of the tCMS potential. This observation has led to the discovery of new integrable
spin-chain models with long-range order (for instance, [18,19]). It has also stimulated
the interest for the study of symmetries of the Yangian-type in these CMS spin models.
On the other hand, the physical motivations underlying the formulation of the Haldane
Shastry model, related to Andersons resonating-valence-bond model at the time an
alternative to the Nel state in high-Tc superconductivity and the 1D Hubbard models,
have nourished the interest of CMS models in condensed matter physics. 2 Many more
physical applications have been found in the last decade, ranging from quantum chaos
and matrix models [21], mesoscopic systems [22], black holes [23] to supersymmetric
integrable gauge theories [24].
The return of CMS models on the hot seat in physics, mainly through the discoveries of
many applications, has stimulated a wave of activities on the models intrinsic properties.
2 In that vein, we could also mention that the Jack polynomials, the eigenfunctions of the tCMS model
see below have proved to be a rather convenient basis for performing some computations in specific field
theoretical problems of condensed matter; see, for instance, [20].
549
Two objects, discovered in quite different contexts by mathematicians, have played a key
role in that regard: these are the Dunkl operators [25] and the Jack polynomials [26].
Dunkl operators play a crucial role in various branches of mathematics, e.g., in the
theory of affine Hecke algebras [27] and for Schubert calculus (see, e.g., [28]). In the
context of CMS models, they were a crucial ingredient in the formulation of the exchangeoperator formalism, as they allowed a very simple and direct construction of the commuting
charges [29]. They are also at the heart of the transfer matrix formulation of the CMS
models [30].
The rebirth of CMS models coincides with a period of intense activity in mathematics
regarding the theory of symmetric functions, centered mainly on the Jack and Macdonald
polynomials, and in particular on their combinatorial applications. Jack polynomials are
symmetric polynomials that were shown to be eigenfunctions of a simple differential
operator (see, e.g., Section 5 in [31]). Then, Forrester pointed out that this operator
is precisely the gauge transformation, by the ground-state wave function, of the tCMS
hamiltonian [32], bringing the subject of Jack polynomials on the physicists desktable.
At the time, there were still no explicit expressions for these polynomials. The search for
an explicit description has led to two noteworthy discoveries by physicists. The first one is
a Rodrigues type formula, namely, a recursive construction via the action of a differential
operator built up from shifted products of Dunkl operators [33]. Independently, integral
formulas for the Jack polynomials have been found in [34]; they led to the discovery of a
fascinating but still mysterious connection between these polynomials and special singular
vectors of the Virasoro and W algebras (see also [35]).
On the mathematical side, explicit expressions for the Jack polynomials have been
obtained using a non-symmetric version of these polynomials [36]. And more recently,
a rather simple determinant formula for the Jack polynomials has been presented [37].
Some of these results in particular, the creation-operator formalism have been
extended to the construction of the hi-Jack polynomials [38], which are eigenfunctions of
the CMS model with an inverse-square interaction augmented by an harmonic confining
term, and of the Macdonald polynomials [39,40], which are eigenfunctions of the
trigonometric RuijsenaarsSchneider model [41], a relativistic version of the tCMS model.
In the latter case, integral formulas have also been obtained (see, e.g., [42] and references
therein).
A quite natural extension of these studies is to consider their supersymmetric generalizations. The supersymmetric version of the rational CMS model has been considered by
Freedman and Mende [43], with emphasis on the study of supersymmetry breaking, the
very physical problem that has motivated the development of supersymmetric quantum
mechanics [44]. The revival in the interest for the CMS models has stimulated a number of studies of their supersymmetric counterparts [4550]. However, there remain many
open problems. For instance, there are no concise Lax formulation. Moreover, the suitable
generalization of the Dunkl operators is known only for the rational model with confinement [46]. And more importantly, there are absolutely no known results concerning the
proper superextension of the Jack polynomials. Indeed, it is only for the rational case with
harmonic term that the solutions have been constructed in [43,50] out of fermionic and
550
bosonic creation operators related to those of a supersymmetric harmonic oscillator by a

similarity transformation.
The initial goal of this work was to launch the study of the Jack superpolynomials. As an
offshoot, we have obtained a number of new results on the supersymmetric tCMS (stCMS)
models per se.
The first step in the construction of the supersymmetric model is the introduction of
the fermionic variables i and their conjugates i . Out of these variables, two generic
expressions for fermionic charges the possible supersymmetric charges can be
constructed. The point we want to stress at this level is that this construction, rooted in
the presence of two fermionic charges, leads necessarily to two supersymmetries. These
charges are then used to build an hamiltonian. Explicitly, the hamiltonian is written as
the anticommutator of these two charges, which thereby make the latter automatically
conserved with respect to the dynamics generated by this hamiltonian. 3 We then adjust
the precise expression of the charges in order to recover, when the fermionic variables are
dropped, the bosonic hamiltonian to be supersymmetrized. The complete hamiltonian is
thus the supersymmetric hamiltonian we are looking for. In our case, this is the stCMS
hamiltonian [45]. This analysis is presented in Section 3.1
An observation that proves to be central for our subsequent analysis is that the part of the
hamiltonian that contains the fermionic variables can be described in terms of a fermionic
exchange operator cf. Section 3.1 (and we found afterwards that the same observation
had been made before in [45]). This allows us to use the projection formalism developed
in [17,30] for the description of the CMS models with spin degrees of freedom. The key
point of this projection technique is that by restricting the space of functions on which
the operators act (namely, functions that are completely symmetric with respect to both
the fermionic and the bosonic variables), we can trade the fermionic-exchange operator
hence, the fermionic degrees of freedom for a standard position-exchange operator.
In particular, this method leads us to a novel but quite natural construction of the Dunkl
operators in either their covariant or their commuting version cf. Section 3.3. This
leads us to a direct proof of the integrability via the construction of commuting conserved
bosonic charges. In Section 3.2, another proof of the integrability is presented, this one
based on the Lax formalism. Although we arrived at this Lax formulation independently,
we realized that the same Lax operators, expressed in terms of exchange operators, had
been presented in [35], albeit in a different context.
Before pursuing the presentation of the papers content, let us pause to discuss
briefly the meaning of integrability for supersymmetric mechanical systems. In the nonsupersymmetric case, this amounts to demonstrate the existence of N the number of
particles commuting independent bosonic charges. A working criterion for an integrable
supersymmetric extension of an integrable mechanical system could be the existence of N
commuting independent bosonic charges that reduce to their non-supersymmetric version
when the fermionic variables are dropped. For all the cases we can think of (including field
3 It is clear that this supersymmetrization process is quite different from the one used in classical field theory
based on superspace techniques. There is no natural analogue of the superfield here, for instance.
551
theoretical models), this appears to be sufficiently restrictive. However, we could argue

that having introduced N new degrees of freedom (the fermionic variables being split into
a set of generalized variables, the i s, and their conjugates, the i s, i = 1, . . . , N ), we
should expect, in the spirit of the Liouville theorem, that N additional conserved charges
are required. Actually, the mere supersymmetry invariance appears to supply automatically
further conserved charges. As already pointed out, the built in supersymmetry implies the
existence of two conserved charges, denoted by Q and Q . Recall that the hamiltonian
is given by their anticommutator. But this turns out to be true for all the higher-order
hamiltonians of the system, i.e., they can all be expressed as anticommutators of higherorder fermionic charges. Indeed, for the stCMS model, we can construct rather directly
(using the Dunkl operators, for instance) 2N conserved fermionic charges. However,
it should be stressed that these do not anticommute among themselves. Moreover, by
inspection, we readily find N additional bosonic conserved charges that commute with
the bosonic ones previously constructed. Afterwards, this appears to be somewhat natural
given that we have two supersymmetries, suggesting heuristically that the charges get
organized in multiplets of four, two bosonic and two fermionic. 4
In Section 4 we turn to the main subject of this work: the formulation of the Jack
superpolynomials. They are defined as eigenfunctions of the stCMS model. Notice that by
a superpolynomial we refer to a polynomial in bosonic and fermionic variables without
imposing a supersymmetric invariance constraint (i.e., these are not supersymmetric
polynomials). We first unravel, in Section 4.1, the mixed symmetry properties, with respect
to the bosonic variables, that are induced by the presence of the fermionic variables
on any symmetric superpolynomials. This leads us naturally to the central concept of
superpartitions introduced in Section 4.2 and which appears to be original. Superpartitions
are used in turn to define the monomial superfunctions. Jack superpolynomials are then
defined in Section 4.3 as those stCMS eigenfunctions that are triangular with respect to a
monomial superfunction decomposition. Many examples are presented.
Various straightforward extensions of the results presented in this paper, directions for
future research and conclusions are collected in the final section. Some auxiliary sections
complete the article. A brief review of the basic definitions pertaining to the usual Jack
polynomials and some associated concepts is presented in Section 2. The remaining
complementary material is spread in three appendices. Excited states can be built from
a vacuum state free of fermions, as it is done in the main body of the paper, or from
a vacuum filled by N fermions. This second option is considered in Appendix A. In
Appendix B, we introduced creation operators analogous to those introduced in [33] for
the standard Jack polynomials. Finally, a simple combinatorial expression counting the
number of superpartitions of a given degree and a given fermionic number is presented in
Appendix C.
4 This can be compared to the case of the classical N = 1 supersymmetric Korteweg de Vries equation [51],
which is probably the best studied supersymmetric integrable system. In that case, we find that in addition to
the bosonic supersymmetric extension of the usual KdV charges, there are (twice as many) nonlocal fermionic
charges whose Poisson brackets yield a local bosonic charge if dimensionally allowed and vanish otherwise.
552
2. Background
2.1. CalogeroMoserSutherland models
The CMS-models describe systems of N particles interacting pairwise through longrange potentials. The classical and quantum versions of these models are integrable. In this
article, we focus on the supersymmetric extension of the quantum tCMS model in which
the identical particles of mass m lie on a circle of circumference L. If we set m = h = 1,
the hamiltonian of the tCMS model is the following [3]:
2
N

1
1 2
pi +
( 1)
,
H=
(1)
2
2
L
sin (xij /L)
i=1
1i<j N
where is a dimensionless real coupling constant. In this equation, and for the remainder
of the article, double indicing stands for the difference between two variables, i.e.,
xij xi xj .
(2)
Position and momentum variables obey the usual commutation relations:

[xj , pk ] = ij k .
(3)
Two other models can be obtained from the tCMS model: the replacement L iL yields
the hyperbolic model, whereas the limit L gives the rational model (on an infinite
line).
The hamiltonian (1) is semi-positive, i.e.,
1
Aj Aj + E0 ,
H=
(4)
2
j
with
Aj = pj i
Xj k
(5)
k=j
and

xj k
cot
Xj k =
.
L
L
(6)
Hence, the minimal value in the spectrum of H is given by

2 N(N 2 1)
.
E0 =
L
6
(7)
The ground state, which is annihilated by every operator Aj , corresponds to the following
Jastrow-type function:

xj k
0 (x) = e j<k dxj Xjk =
(8)
sin
(x).
L
j <k
The simplest way of showing the integrability of the quantum CMS models is by
displaying a Lax pair, namely two N N Hermitian matrices, denoted by L and M,
553
satisfying the relations:

L j k = i[Lj k , H ] = i[L, M]j k ,
Mj k =
Mj k = 0.
(9)
The constraint on M is essential for the following N independent quantities to be

conserved:
H(n) =

1
1
tr Ln tr Ln ,
n
n
(10)
where is the matrix whose entries are all 1s. Therefore, tr A denotes the total trace
of A, that is, the sum of all the entries of A. For the tCMS model, the Lax pair reads [2]:
Lj k = pj j k + i(1 j k )Xj k ,

Xj l (1 j k )Xj k ,
Mj k = j k
(11)
l=j
where Xj k = dX(xj k )/dxj k . From Eq. (10), we see that the first and second conserved

quantities correspond, respectively, to the momentum P = i pi and the hamiltonian H
of the system.
In order to solve the Schrdinger equation associated to the CMS model, it is convenient
to set:
zj = e2ixj /L .
(12)
The variable zj thus gives the position of the j th particle on a circle of circumference L in
the complex plane. In this notation, H becomes:
2

zi zj
2
H =2
(13)
2( 1)
.
zi
2
L
zi
zij
i
i<j
The eigenfunctions of the excited states of the hamiltonian (13) are written in the form
(x) = (x)0 (x) where (x) is required to be symmetric in order for to behave like
0 under the exchange of particles. It is thus natural to conjugate the hamiltonian with the
ground-state wave function 0 (x):

1 L 2

H=
(H E0 ) ,
2
zi + zj

(zi i )2 +
(zi i zj j ),
=
zij
i
(14)
i<j
and look for the eigenfunctions (x) of this conjugated hamiltonian. In the following, bar
operators will stand for operators that have been obtained by a similar conjugation of the
ground state.
The symmetric eigenfunctions (x) of (14) are known as the Jack polynomials.
554
2.2. Symmetric functions and Jack polynomials

We now summarize some basic results concerning symmetric functions [39]. This will
allow us to define properly the Jack polynomials [26,31], and thereby, to present the
solutions of the tCMS model [3].
2.2.1. Symmetric functions and exchange operators
Symmetric functions are invariant under the action of the symmetric (or permutation)
group SN . If z = (z1 , . . . , zN ) denotes the set of variables, then a function F (z) is
symmetric if it remains invariant under the exchange of its variables:
Kij F (z) = F (z)
i, j,
(15)
where Kij is a transposition of SN , i.e., an exchange operator, whose action is defined as

follows:
Kij f (zi , zj ) = f (zj , zi , )Kij ,
(16)
with f (zi , zj ) standing for a function or an operator. The fundamental properties of the
exchange operators are
Kij = Kj i ,
Kij = Kij ,
Kij Kj k = Kik Kij = Kj k Kki ,
Kij2 = 1.
(17)
(n)
Let N denote the ring of symmetric polynomials in the variables z1 , . . . , zN , and let N
be the subspace of symmetric polynomials of degree n. Before introducing various bases
for N , we need to introduce the notion of partitions.
2.2.2. Partitions
A partition is a weakly-decreasing sequence of non-negative integers. More precisely, a
partition of weight (or degree) n is defined as follows:
= (1 , 2 , . . . , l ),
1 2 l 1,
n = 1 + 2 + + l = ||,
(18)
where l = l() is the length of the partition, that is, the number of its non-zero parts. We
use p(n) for the number of partitions of n, e.g., the number of partitions of 4 is p(4) = 5.
We can also represent a partition as:
= (1m1 , 2m2 , . . . , i mi , . . .),
(19)
where mi is the number of parts of equal to i. This allows us to define the following
constant
z = 1m1 m1 ! 2m2 m2 ! . . . ,
(20)
555
which enters in the definition of the Jack polynomials. There exists a natural partial order
on partitions called the dominance ordering. It is defined in the following way:
if 1 + 2 + + i 1 + 2 + + i ,
i.
(21)
The dominance ordering is a total order only for weights up to n = 5. Note finally that to
we can associate a Young tableau having i boxes in the ith row. The conjugate partition,
denoted corresponds to the partition resulting from the interchange of the rows and
columns in the Young tableau associated to .
2.2.3. Power sums
The symmetric polynomials

zin ,
pn =
(22)
where the sum extends over the N variables, are called power sums. The set of all products
of power sums, i.e.,
p = p1 p2 pl ,
(23)
forms a basis of N .
2.2.4. Elementary symmetric functions
The elementary symmetric functions are:

en =
zi1 zin .
(24)
i1 <i2 <<in
Again, the set of all products of elementary functions

e = e 1 e l ,
(25)
is a basis of N .
2.2.5. Monomial symmetric functions
The monomial symmetric functions are defined as follows:

P (1) P (2)
zP () =
z1 z2 zNP (N) ,
m =
P SN
(26)
P SN
where here and below, the prime on the sum is used to indicate that it is done only over
distinct permutations, which means that no monomial is repeated. The p(n) possible
(n)
monomial functions of degree n constitute another basis of N . Also, the monomial
symmetric functions generalize en and pn :
m(n) = pn and m(1n ) = en .
The simplest monomial functions are given in Table 1.
(27)
556
Table 1
Monomial functions of weight || 3 for N = 4 variables
Weight
||
Partition
Monomial function
m (z)
(0)
(1)
z1 + z2 + z3 + z4
(11)
(2)
z1 z2 + z1 z3 + z1 z4 + z2 z3 + z2 z4 + z3 z4
z12 + z22 + z32 + z42
(111)
(21)
z1 z2 z3 + z1 z2 z4 + z1 z3 z4 + z2 z3 z4
z12 z2 + z12 z3 + z12 z4 + z22 z1 + z22 z3 + z22 z4
+ z32 z1 + z32 z2 + z32 z4 + z42 z1 + z42 z2 + z42 z3
3
z1 + z23 + z33 + z43
(3)
2.2.6. Jack polynomials

The Jack polynomials, J (z1 , . . . , zN ; ), are symmetric polynomials depending on a
parameter that also form a basis of N . They belong to the ring Q()[z1 , . . . , zN ]SN of
symmetric polynomials with rational coefficients in . They are uniquely characterized by
the following two conditions (see, e.g., [39]):
J , J = 0
and = (orthogonality),

J (z; ) = m +
v ()m (unitriangularity),
(28)
(29)
<
where the scalar product is defined in the following way, with respect to the power sums:
p , p = , z l() ,
(30)
where z has been introduced in Eq. (20). The Jack polynomials generalize several types
of symmetric polynomials:
1,
s (z) (Schur functions),
J (z; ) m (z) (monomial functions), ,
(31)
e (z) (elementary functions), 0

(recall that refers to the conjugate partition).
A few examples of Jack polynomials expanded in terms of monomial symmetric
functions are shown in Table 2. In this notation, the number of variables is irrelevant as
long as it is not smaller than the degree of the polynomial.
2.2.7. Jack polynomials and the tCSM model
Jack polynomials are related to the tCMS model for:
1/,
(32)
where is the models coupling constant. In this context, we have to replace the scalar
product A, B by the physical one:
557
Table 2
Jack polynomials of weight || 4
Weight Partition
||
Eigenvalue
(, N)
Jack polynomials
J (z; 1/)
(0)
m(0)
(1)
1 + N
m(1)
(12 )
2 + 2N 4
m(12 )
(2)
4 + 2N 2
2
m(2) + 1+
m(12 )
(13 )
3 + 3N 9
(21)
5 + 3N 5
m(13 )
6
m(21) + 1+2
m(13 )
(3)
9 + 3N 3
3 2
6 2
m(3) + 2+
m(21) + (1+)(2+)
m(13 )
(14 )
4 + 4N 16
m(14 )
(212 )
6 + 4N 10
(22 )
8 + 4N 8
(31)
10 + 4N 6
(4)
16 + 4N 4

A(x), B(x) =
12
m(212 ) + 1+3
m(14 )
2
12 2
m(22 ) + 1+
m(212 ) + (1+)(1+2)
m(14 )
2
12 2
m(31) + 1+
m(22 ) + (3+5)
2 m(212 ) +
2 m(14 )
(1+)
(1+)
2
6(1+)
m(4) + 3+ m(31) + (2+)(3+) m(22 )
12 2
24 3
+ (2+)(3+)
m(212 ) + (1+)(2+)(3+)
m(14 )

dzN
dz1
zi
A(1/z)B(z),
1
2i
2i
zj
i=j
2
dx1 dxN |0 |2 A(x) B(x),
(33)
defined
where 0 is the ground state of the trigonometric model. The hamiltonian H
in (14) is self-adjoint with respect to the scalar product (33). This physical scalar product
can equally be used to characterize the Jack polynomials as the unique polynomials
satisfying (29) that are orthogonal with respect to (33).
It is also known that the Jack polynomials are eigenfunctions of the transformed
[31,32]:
hamiltonian H
J (z; 1/) = J (z; 1/)
H
(34)
with eigenvalues:

2j + (N + 1 2j )j .
=
(35)
The wave functions of the original trigonometric model are now simply (z) =
J (z; 1/) , with eigenvalues E = 2(/L)2 + E0 . Therefore, if we introduce the
558
quasi-momenta

2
i + (N + 1 2i) ,
i =
L
(36)
we observe that the spectrum of the model is that of a system of N free quasi-particles,
each of these with quasi-momentum i :
E =
2
i
(37)
The quasi-momenta of two neighboring quasi-particles satisfies:

4
(38)
.
L
The excited states of the tCMS model thus obey a generalized exclusion principle [7,8,13].
In particular, we recover free bosons if = 0 and free fermions if = 1.
i i+1
3. Supersymmetric CalogeroMoserSutherland models

3.1. Supersymmetric quantum mechanics
Consider a quantum model that contains both bosonic and fermionic variables and
whose hamiltonian is denoted H. Following the usual methods of supersymmetric quantum
mechanics [44,52,53], we consider, in addition to the 2N bosonic variables (x, p), the 2N
fermionic variables (, ). 5 The bosonic and fermionic variables satisfy, respectively, a
Heisenberg and a Clifford algebra:

[xj , pk ] = ij k ,
(39)
j , k = j k ,
with all other commutators or anticommutators equal to zero. We will usually work with a
differential realization of these algebras:
pj = i
,
xj
j =
.
i
(40)
To construct a supersymmetric hamiltonian, we will first construct two supersymmetric

charges, denoted Q and Q , and define the hamiltonian as their anticommutator:
1
H = {Q, Q }.
(41)
2
By construction, the hamiltonians eigenvalues are non-negative. The hamiltonian is
invariant under a supersymmetric transformation if:
2
Q2 = Q = 0.
(42)
5 This amounts to considering N = 2 supersymmetries i.e., there will be two conserved supersymmetric
charges. Extensions to more supersymmetries are discussed in the conclusion.
559
By writing the charges under the form

Q=
N
i Ai (x, p),
Q =
i=1
N
i Ai (x, p),
(43)
i=1
we find that Eq. (42) requires:

[Ai , Aj ] = 0 = Ai , Aj ,
i, j.
(44)
The generic supersymmetric hamiltonian is thus:

1

H=
Ai Ai +
i j Ai , Aj .
2
i
(45)
i,j
For non-relativistic models, the hamiltonian is proportional to the square of the particles
speed. We therefore write Ai as a linear function of the momentum pi :

Q=
(46)
j pj ij (x) ,
Q =
j pj + ij (x) .
j
From Eq. (42), the potential j (x) must be of the form

j (x) = xj W (x),
(47)
where W (x) (called the prepotential), is an arbitrary function of the variables x1 , . . . , xN .

The supersymmetric hamiltonian now takes the form [43]:

1 2
H=
(48)
pi + (xi W )2 + x2i W
i j xi xj W .
2
i
i,j
This hamiltonian is an extension of the purely bosonic model whose potential is

2
2
i [(xi W ) + xi W ].
Since the hamiltonian is semi-positive, any state annihilated by the charges Q and Q is
a ground state (vacuum). Obviously, only the vacuum is supersymmetric since an excited
state cannot be simultaneously annihilated by both charges. The charges defined in (46)
naturally lead to two ground states:
0 = eW |0 and 0 = eW |0,
(49)
belong to the fermionic Fock space and are defined as

where the ground states |0 and |0
follows:
= 0,
i |0
i |0 = 0,
i.
(50)
and |0 correspond to the

If we interpret the i s as fermionic creation operators, then |0
N -fermion and the 0-fermion states, respectively. In the realization (40), the ground states
must then be of the form:
|0 1,
1 N .
|0
(51)
To be physically meaningful, the functions 0 and/or 0 must be normalizable. If this

is not the case, the supersymmetry is said to be broken. It should be noted that Eq. (49)
560
provides a natural way to supersymmetrize a model. Knowing the ground state 0 of that
model, it suffices to let W = ln 0 to get its supersymmetric extension [53].
We now specialize to a prepotential of the form:

w(xij ).
W (x) =
(52)
i<j
The comparison of Eqs. (8) and (49) immediately gives the right choice of W for the CMS
models:
w (xij ) = Xij .
(53)
Consequently, the stCMS hamiltonian reads:

2

1 2 2

Xij + Xij 1 ij ij N(N 1)(N 2)
pi +
,
H=
2
L
i
i<j

2
N
1 + ij ij
1 2
=
pi +
E0 ,
2
L
sin2 (xij /L)
(54)
i<j
i=1
where E0 is as given in (7). The two ground states

0 (x) = (x) and 0 (x, ) = (x)1 N
(55)
are invariant under supersymmetric transformations and are normalizable for any value
of .
The supersymmetric model can be solved much more easily if we notice that the term
ij 1 ij ij = 1 (i j )(i j )
(56)
is a fermionic-exchange operator [45], that is,

ij f i , j , i , j = f j , i , j , i ij
(57)
for any monomial function f . Moreover, the ij s satisfy the usual properties (17) of
exchange operators.
As in the non-supersymmetric case, it is convenient to use zj = e2ixj /L , in terms of
which the hamiltonian reads:
2
zi zj
2
(zi i ) 2
( ij ) E0 .
H=2
(58)
2
L
zij
i
i<j
Removing the ground-state contribution leads to:

1 L 2

H
H
2

zi + zj
zi zj
=
(zi i )2 +
(zi i zj j ) 2
(1 ij ),
2
zij
zij
i
i<j
i<j
(59)
561
which is still supersymmetric because it is invariant under the action of the transformed
fermionic charges:
Q = Q
and Q = Q .
(60)
A complete set of eigenfunctions of the hamiltonian (59) is given in Section 4. Excited

states built from the second ground state are considered in Appendix A.
3.2. Lax formalism in the supersymmetric CMS models
Quite remarkably, knowing the CMS models Lax pair is enough to guarantee the
existence of their supersymmetric extensions. Moreover, the supersymmetric Lax pair is
a simple extension of the non-supersymmetric one.
The first statement is proved as follows. Recall that the supersymmetric hamiltonian is
the anticommutator of the two supersymmetric charges. Comparing Eqs. (11), (46), (47)
and (53), we see that the supersymmetric charges, hence the supersymmetric hamiltonian,
can easily be built from the Lax matrices [45]:

Q=
(61)
j Lij ,
Q =
i Lij .
i,j
i,j
Therefore, the Lax matrices of the quantum CMS models guarantee the existence of their
supersymmetric extensions!
In order to prove the integrability of the supersymmetric models, we introduce the four
matrices that will provide the Lax formulation of the supersymmetric system:
Lj k = pj j k + i(1 j k )Xj k j k ,

Mj k = j k
Xj l j l (1 j k )Xj k j k .
l=j
jk = j j k .
j k = j j k ,
These matrices obey the relations:

L j k = i Lj k , H = i[L, M]j k ,

j k = i j k , H = i[, M]j k ,

jk = i jk , H = i[ , M]j k .
(62)
(63)
To verify the equivalence between theses relations and the equations of motion, we use the
properties
Lj k k = j Lj k ,
and

i
Mij =
Lj k k = j Lj k
Mij = 0 M = M = 0,
(64)
(65)
where ij = 1.
We see that L and M are obtained from L and M by simply multiplying each Xj k or
Xj k factor by the exchange operator j k .
562
Using the matrices L and , one can construct the following independent quantities that
can easily be shown to be conserved:
1
1 n
H(n) = tr Ln =
Lj k , n = 1, 2, . . . , N,
n
n
jk
Q(n) =
1
1
tr (Ln ) =
j Lnjk ,
n
n
n = 0, 1, . . . , N 1,
jk
Q(n) =
1
1 n
tr ( Ln ) =
j Lj k ,
n
n
n = 0, 1, . . . , N 1,
jk
I(n) =

1
1
tr Ln =
j j Lnjk ,
n
n
n = 0, 1, . . . , N 1.
(66)
jk
More generally, any operator that can be written as the total trace of a polynomial function
F only depending on the matrices , and L is conserved:

d
tr F , , L = 0.
dt
(67)
The quantities Q(1) and Q(1) are simply the generators of the supersymmetric transformations. However, the fermionic charges are not in involution: their anticommutators generate
the hamiltonians H(n) (see below).
We stress that the results of this subsection apply to all types of supersymmetric CMS
models and not just the stCMS one.
3.3. Dunkl operator formalism and the supersymmetric CalogeroMoserSutherland
models
In this section we construct the Dunkl operators of the stCMS model. With these
operators in hands, the integrability of the model can be very easily (re)established.
We first introduce a new exchange operator that acts on the bosonic and fermionic
variables:
Kij ij Kij ,
where [ij , Kij ] = 0.
(68)
A function of the variables zi and i is said to be a symmetric superfunction if it is invariant

under the action of the Kij s. It is worth noticing that the action of ij on symmetric
superfunctions is equivalent to the action of Kij on those functions. For instance, if FK is
a symmetric superfunction, that is,
ij FK = Kij FK ,
(69)
we can rewrite the hamiltonian H H as:

H FK = HK FK ,
2
zi zj
(zi i )2 2
( Kij ) E0 .
HK = 2
2
L
zij
i
i<j
(70)
563
This remark holds for any supersymmetric model whose hamiltonian contains a fermionicexchange term.
Let us denote by E (O) the projection of an operator O on a vector space invariant
under the action of E [17,30]. For instance,
K (Kij ) = 1,
K (Kij ) = ij ,
K (ij kl ) = Kkl Kij .
(71)
We can consider the hamiltonian with exchange term Kij as the most fundamental one.
Indeed, by an appropriate choice of projection, various models can be generated from
it. For instance, the tCMS model and its supersymmetric generalization are obtained
respectively from: 6
K (HK ) = H,
(72)
K (HK ) = H .
(73)
We now present a simple way to derive the various types of Dunkl operators for the
CMS models found in the literature out of the Lax operator (cf. [8,17,30]). The covariant
Dunkl operator is simply given by

Dj =
Lj k (Xj k Xj k Kj k )
k
= pj + i
Xj k Kj k .
(74)
k=j
This Dunkl operator satisfies the following properties:

Kij Di = Dj Kij
(covariance),

(Kik Kj k )Kij
[Di , Dj ] = (/L)2
(non-commutativity),
k=i,j
[Di , HK ] = 0
(conservation),
(75)
where covariance means that Di behaves like the variable zi under the action of the
symmetric group SN . 7 It should be noted that this Dunkl operator can be viewed as the
square of fermionic derivatives, i.e.,
2
Di = Ci2 = Ci ,
(76)
where
+ i Di ,
Ci = i + Di
.
i
i
It is useful to introduce another Dunkl operator:

Kij
Kij
Di = Di
L
Ci =
j <i
(77)
(78)
j >i
6 We could also choose projections on antisymmetric spaces, e.g., K = . This is considered in

ij
ij
Appendix A. In a similar vein, spin degrees of freedom can be introduced in that way cf. the conclusion.

7 Another simple covariant Dunkl operator is D
j = Dj (/L)
j =i Kij . It has the following, somewhat
j ] = (2/L)(D
i D
j )Kij .
i , D
more natural, commutation property: [D
564
that satisfies
[Di , Dj ] = 0
2
L
(commutativity),
[Di , HK ] = 0
(conservation).
Ki,i+1 Di+1 Di Ki,i+1 =
(degenerated Hecke algebra),
(79)
In addition to commuting among themselves, the Di have the nice property that the
hamiltonian with exchange term K lies in their universal algebra:
1
(Di )2 = HK + E0 .
2
(80)
The supersymmetric hamiltonian is recovered by a simple projection.

Using these two versions of the Dunkl operators, it is now fairly easy to prove
the integrability of the supersymmetric trigonometric model by constructing explicitly
conserved charges from sums of powers of the Dunkl operators. First, the N commuting
conserved bosonic quantities which generalize those of the non-supersymmetric model are
simply: 8

n
H(n) = K
Di , n = 1, 2, . . . , N,

H(n) , H(m) = 0
(81)
n, m.
The proof of the commutativity relies on a simple property of the projections [54]:

K [A, B] = K (A), K (B) if [Kij , A] = [Kij , B] = 0.
(82)
n
The operators H(n) meet this requirement since [Kij , ( i Di )] = 0. The latter property

implies also [Di , ( j Djn )] = 0. In addition to these bosonic conserved quantities, charges
with fermions can be constructed as:

i Din , n = 0, 1, . . . , N 1,
Q(n) = K

Q(n)
= K

i Din
I(n) = K

n = 0, 1, . . . , N 1,
,

i i Din
n = 0, 1, . . . , N 1.
(83)
Note that here we use the covariant Dunkl operators rather than the commuting ones.
This is imposed by the presence of the factors. Indeed, proving that those quantities are
conserved still requires (82) and the covariant character of the Dunkl operators ensures the
8 We use the same notation for the charges constructed from the Lax operators and from the Dunkl operators.
Although the lowest order charges calculated from both expressions agree, this may not be so for the higher-order
ones. Nevertheless, they are equivalent sets of independent charges.
565

commutativity of the operators i i Din with an arbitrary exchange operator Kij (which
would not be true if the commuting Dunkl operators were used instead).
The fermionic charges are not in involution: their anticommutators generate bosonic
quantities, e.g., H(n) . Take for instance the rational case (on the infinite line) where Di
Di ; the conserved quantities constructed from the Dunkl operators (like those defined from
the Lax matrices) satisfy the following algebra:

Q(n) , Q(m) = H(n+m) ,

Q(n) , I(m) = Q(n+m) ,
Q(n) , I(m) = Q(n+m) ,

Q(n) , Q(m) = Q(n) , Q(m) = I(n) , I(m) = 0,

Q(n) , H(m) = Q(n) , H(m) = I(n) , H(m) = H(n) , H(m) = 0.
(84)
Only the last line remains true for the trigonometric and the hyperbolic models. In fact, it
seems that the algebra of {H, Q, Q , I} does not close linearly for those models.
Moreover, we could also replace the N independent hamiltonians H(n) by the following
conserved quantities:

J(n) = K
(85)
Din , n = 0, 1, . . . , N 1.
i
However, the supersymmetric hamiltonian H would not belong to this set. There is thus
some freedom in the way we choose a set of independent conserved charges. But quite
generally, the projection K of any quantity, made out of either the Di s or the Di s as
well as out of the fermionic quantities i and i , invariant under the action of the exchange
operator Kij , is always conserved in the supersymmetric model.
We end this section by mentioning that the Dunkl operators of the transformed
are obtained as follows:
hamiltonian H

i = L
L ij (Xij Xij Kij ),
D
2
j
zi + zj
(1 Kij ),
2
zij
j =i

i = D
i
D
Kij
Kij ,
2
= zi i +
j <i
(86)
j >i
and we verify that:

1 L 2
2

K
Di = H +
E0 .
2
(87)
All the quantities constructed in this section can thus be directly transposed to the case
where the ground-state wave function is factored out.
566
4. Jack superpolynomials
4.1. Symmetry of the tCMS models eigenfunctions
It is known that, when defining the Jack polynomials, we can replace condition (28)

by the condition that the Jack polynomials be eigenfunctions of the hamiltonian H
of the tCMS model. Similarly, one of the conditions entering the definition of their
superanalogues will be that they be eigenfunctions of the stCMS model. We thus begin
by making general observations regarding the symmetry properties of the eigenfunctions
of the stCMS model.
We are looking for functions of and z that are invariant under the transpositions Kij
and that are eigenfunctions of the hamiltonian:

zi + zj
zi zj
=
H
(88)
(zi i )2 +
(zi i zj j ) 2
(1 ij ).
2
zij
zij
i
i<j
i<j
Since the hamiltonian is of degree 0 in both and z, the eigenfunctions have to be

homogeneous in both variables. Moreover, since the underlying mechanical problem
describes the dynamics of a system of particles on a circle, the solutions must be invariant
under the transformation xi xi + 2 ; therefore, only integral powers of the variables z
must be considered. Moreover, because the product of an eigenfunction of degree n by a
Galilean boost,
q
Gq =
(89)
zi , q Z,
i
gives another eigenfunction, now of degree n + Nq, we can restrict ourselves to nonnegative powers of z.
We are thus seeking polynomial eigenfunctions that are invariant under the action of
Kij . This operator commutes with the superhamiltonian, which is not the case with the
operators Kij and ij taken separately. As already pointed out, the polynomials need to
be homogeneous in and z, lets say with degree m and n respectively. These degrees are
good quantum numbers. Indeed, the total momentum

=
P
(90)
zi i
i
and its eigenvalue is the degree in z of the monomial on which it acts:

commutes with H

n1

nN
n

P z z
(91)
=
ni zn1 z N .
i
Likewise, the quantity

=
i i =
i
i
i
(92)
commutes with the supersymmetric hamiltonian and counts the number of fermions in a
monomial:
(i1 im ) = m(i1 im ).
(93)
567
We say that the above monomial belongs to the m-fermion sector.

We can thus solve the supersymmetric Schrdinger equation in a fixed fermionic sector
at a time. The independent eigenfunctions in a given fermionic sector will be denoted
A(m)
(z, ; 1/); they are indexed by a set of integers , called a superpartition, whose
norm refers to the degree in z: || = n (see the following subsection for their actual
definition).
We now clarify the symmetry properties, with respect to the z variables, of any
(m)
symmetric superpolynomials and, in particular, of the eigenfunctions A . The key
(m)
observation is that the solutions A must necessarily be of the form:

A(m)
(94)
i1 ...im Ai1 ...im (z; 1/),
(z, ; 1/) =
1i1 <i2 <<im N
where
i1 ...im = i1 im .
(95)
Indeed, the various terms in the m-fermion sector can always be rearranged as sums of z
i ...i
polynomials with a monomial prefactor in the i s. A1 m is a homogeneous polynomial
in z indexed by a superpartition . The solutions A(m)
being symmetric superpolynomials,
must be invariant under the action of the exchange operators Kij . Given that the products
are antisymmetric, i.e.,
ia ib i1 ...im = i1 ...im
if ia , ib {i1 , . . . , im },
(96)
the superpolynomials Ai1 ...im must be partially antisymmetric to ensure the complete
i1 ...im
must satisfy the following
symmetry of A(m)
. More precisely, the functions A
relations:
Kij Ai1 ...im (z; 1/) = Ai1 ...im (z; 1/) i and j {i1 . . . im },
/ {i1 . . . im }.
Kij Ai1 ...im (z; 1/) = Ai1 ...im (z; 1/) i and j
(97)
Note that the case m = 1 is special:

A(1)
i Ai (z; 1/),
=
i
Kij Ak = Ak
if and only if i, j = k.
(98)
We have thus established that any symmetric eigenfunction of the stCMS model contains
terms of mixed symmetry in z: each polynomial Ai1 ...im is completely antisymmetric in the
variables {zi1 , . . . , zim }, and totally symmetric in the remaining variables z/{zi1 , . . . , zim }.
Appendix B presents a simple way of generating such eigenfunctions by acting with
appropriate operators on the Jack polynomials. However, this method does not lead to a
unique characterization of the eigenfunctions. Later in this section will be presented an
approach free from this drawback. But first, we need to define properly the superpartitions
and some related concepts.
568
4.2. Symmetric superpolynomials

4.2.1. The ring of symmetric superfunctions
The symmetric superpolynomials are polynomials in z and that commute with
the generators Kij of the symmetric group SN of all possible permutations of the N
variables i = (zi , i ). As in the symmetric polynomial case, the set of all symmetric
superpolynomials in the N variables i forms a ring over the field of integers:

N = Z 1 , . . . , N
(99)
S = Z z1 , . . . , zN ; 1 , . . . , N S .
N
It is clear that the set of superpolynomials of degrees m in and n in z is a Z-module,

which we will denote in the following way:

(m;n) = Z 1 , . . . , N (m;n) .
(100)
N
S
N
The ring of symmetric superpolynomials is thus bigraded:

(m;n)
N =

.
N
(101)
m,n
4.2.2. Superpartitions
In the case of symmetric polynomials, the basis elements of N are indexed by partiN can be indexed by superpartitions. To
tions. In the same manner, basis elements of
motivate the following definition of superpartitions, recall that the symmetric superpolynomials in the m-fermion sector are antisymmetric in the m variables {zi1 , . . . , zim } and
symmetric in the remaining ones. We thus define a superpartition of a m-fermion sector as
a sequence of integers that generates two partitions separated by a semicolon:

= (1 , . . . , m ; m+1 , . . . , L ) = a ; s ,
(102)
the first one being associated to an antisymmetric function
a = (1 , . . . , m ),
i > i+1
i = 1, . . . , m 1,
i 0 1 i m,
(103)
and the second one, to a symmetric function:

s = (m+1 , . . . , L ),
i i+1
i > m,
i 0 if i = m + 1,
i > 0 i > m + 1. (104)
In the zero-fermion sector (m = 0), the semicolon disappears and we recover the
partition s .
The length L N of a superpartition corresponds to the total number of its parts, of
which at most two can be zero: one on the antisymmetric side and one on the symmetric
side. The weight (or degree) of a superpartition is simply the sum of its parts:
|| =
L

i=1
i = |a | + |s |.
(105)
569
For instance, the only possible superpartitions of weight 2 in the one-fermion sector are:
(2; 0),
(0; 2),
(1; 1),
(0; 1, 1).
(106)
For 2 fermions, we have instead:

(2, 0; 0) and (1, 0; 1).
(107)
In order to specify explicitly the fermionic sector, we will sometimes denote the degree of
a superpartition as:

degree (m; n) fermionic sector m; weight n = || .
(108)
Summation formulas giving the number of superpartitions of degree (m; n) are presented
in Appendix C.
We mention finally that to any superpartition there corresponds a single standard
partition obtained by rearranging the parts of the superpartition in decreasing order:
= {i |i {1 , . . . , L }, i i+1 }.
(109)
4.2.3. Monomial symmetric superpolynomials

(m;n) that generalizes the symmetric monomial basis
We can now introduce a basis of
N
of N(n) :

(m)
P (1,...,m) zP ()
m (z, ) = m(1 ,...,m ;m+1 ,...,L ) (z, ) =
(110)
P SN
(recall that the prime indicates that the summation is restricted to distinct terms). It is
understood that the action of the permutations on a superpartition is not affected by the
semicolon:

P () = P (1) , . . . , P (m) ; P (m+1) , . . . , P (L) .
(111)
(m)
(0)
The functions m are called monomial symmetric superfunctions and m m (z).

More explicitly, the monomial superfunctions can be written in the following way:
m(1 ,...,m ;m+1 ,...,L )
= m(a ;s )

=

i1 ,...,im aa (zi1 , . . . , zim )ms z/{zi1 , . . . , zim } ,
i1 <i2 <<im
i1 <<im P a Sm P s SNm
sgn(P a )i1 im zi1 P
a (1)
zimP
a (m)
s (L)
(112)
where we have introduced the antisymmetric monomial function

a (z1 , . . . , zN ) =
sgn(P )zP () =
sgn(P )z1 P (1) zNP (N) .
P SN
zimP+1(m+1) ziLP
P SN
Many examples of monomial superfunctions are shown in Table 3.
(113)
570
Table 3
List of all the monomial superfunctions of weight || 3 for N = max(3, L) variables
Weight
||
Sector Superpartition
m
Monomial superfunction
(m)
m (z, )
(0;0)
1 + 2 + 3
(1;0)
(0;1)
(1,0;0)
1 z1 + 2 z2 + 3 z3
1 (z2 + z3 ) + 2 (z1 + z3 ) + 3 (z1 + z2 )
1 2 (z1 z2 ) + 1 3 (z1 z3 ) + 2 3 (z2 z3 )
(1;1)
(0;1,1)
(1,0;1)
1 z1 (z2 + z3 ) + 2 z2 (z1 + z3 ) + 3 z3 (z1 + z2 )

1 (z2 z3 ) + 2 (z1 z3 )3 (z1 z2 )
1 2 (z1 z2 )z3 + 1 3 (z1 z3 )z2 + 2 3 (z2 z3 )z1
(2;0)
1 z12 + 2 z22 + 3 z32
(0;2)
1 (z22 + z32 ) + 2 (z12 + z32 ) + 3 (z12 + z22 )
(2,0;0)
1 2 (z12 z22 ) + 1 3 (z12 z32 ) + 2 3 (z22 z32 )
(1;1,1)
(0;1,1,1)
(1,0;1,1)
1 z1 (z2 z3 + z2 z4 + z3 z4 ) + 2 z2 (z1 z3 + z1 z4 + z3 z4 )
+ 3 z3 (z1 z2 + z1 z4 + z2 z4 ) + 4 z4 (z1 z2 + z1 z3 + z2 z3 )
1 (z2 z3 z4 ) + 2 (z1 z3 z4 ) + 3 (z1 z2 z4 ) + 4 (z1 z2 z3 )
1 2 (z1 z2 )(z3 z4 ) + 1 3 (z1 z3 )(z2 z4 )
+ 1 4 (z1 z4 )(z2 z3 ) + 2 3 (z2 z3 )(z1 z4 )
+ 2 4 (z2 z4 )(z1 z3 ) + 3 4 (z3 z4 )(z1 z2 )
(2;1)
(1;2)
1 z12 (z2 + z3 ) + 2 z22 (z1 + z3 ) + 3 z32 (z1 + z2 )

1 z1 (z22 + z32 ) + 2 z2 (z12 + z32 ) + 3 z3 (z12 + z22 )
(0;2,1)
1 (z22 z3 + z2 z32 ) + 2 (z12 z3 + z1 z32 ) + 3 (z12 z2 + z1 z22 )
(2,1;0)
1 2 (z12 z2 z1 z22 ) + 1 3 (z12 z3 z1 z32 ) + 2 3 (z22 z3 z2 z32 )
(2,0;1)
(1,0;2)
1 2 (z12 z22 )(z3 ) + 1 3 (z12 z32 )(z2 ) + 2 3 (z22 z32 )(z1 )

1 2 (z1 z2 )(z32 ) + 1 3 (z1 z3 )(z22 ) + 2 3 (z2 z3 )(z12 )
(3;0)
1 z13 + 2 z23 + 3 z33
(0;3)
1 (z23 + z33 ) + 2 (z13 + z33 ) + 3 (z13 + z23 )
(3,0;0)
1 2 (z13 z23 ) + 1 3 (z13 z33 ) + 2 3 (z23 z33 )
2
2
1
2
4.3. Jack superpolynomials: monomial expansion

We now define the Jack superpolynomials in the m-fermion sector as the unique
that can be decomposed in terms
eigenfunctions of the supersymmetric hamiltonian H
of monomial superfunctions in the following way:
(m)
(m)
J (z, ; 1/) = m (z, ) +

<
(m)
c, ()m (z, ),
(114)
571
where and are the partitions associated to the rearrangements of and ,

respectively. 9
The relation between Jack superpolynomials and the usual Jack polynomials can now be
stated precisely: J(0) = Js .
The coefficients c, () in (114) are rational functions in . We can easily verify, from
is the same as
the leading terms, that the spectrum of the supersymmetric hamiltonian H

that of H , i.e.,
(m) (z, ; 1/) = J (m) (z, ; 1/) = J (m) (z, ; 1/),
HJ
where the eigenvalues are given by:

=
2j + (N + 1 2j )j ,
(115)
(116)
being the rearrangement of . Observe that the eigenvalue is independent of the

fermionic sector, i.e., independent of the value of m. 10
Tables 4 and 5 present simple examples whose degrees (m; n) are not larger than
(3; 4). The coefficients c, are obtained by simply diagonalizing the hamiltonian.
Polynomials with the same partition have the same eigenvalue. Given that the eigenvalues
are independent of m, they can be read in Table 2.
We should stress that the decomposition is not simply a straightforward extension of the
triangular decomposition of the Jack polynomials, where the ordering is on partitions. Here
the ordering that allows a triangular decomposition is on superpartitions. The existence of
such an ordering seems to us quite remarkable (even though it is not a genuine ordering,
as it is shown in [55]).
A closer look at those results shows that expression (114) is not restrictive enough:
certain monomials allowed by the dominance ordering of the rearranged superpartitions do
not appear in the actual expansion of the Jack superpolynomials. For instance, no monomial
9 The reader is referred to [55] for a proof that this definition does in fact characterize a family of polynomials
(m;n) .
that forms a basis of
N
10 Therefore, the most general eigenfunction having energy is a linear combination of all the eigenfunctions
(m)
J whose eigenvalue is = :
J (z, ; 1/) =
N
(m)
mJ
(z, ; 1/)
m=0
= J (z; 1/) +

1 i J i (z; 1/) + 2 i,j J i,j (z; 1/) +
N 1...N J 1...N (z; 1/)

+
where an ordered summation on repeated indices is understood:

i ...i
i ...i
i1 im J1 m .
i1 ...im J1 m =
(117)
(118)
1i1 <i2 <<im N
m stands for a commuting (anticommuting) constant when m is even (odd). These constants are auxiliary:
Here
they only guarantee the homogeneity of the statistics of the superpolynomials. Eq. (117) thus corresponds in some
way to the Taylor series of a generalized Jack superpolynomial around = 0. The first term in the expansion is
simply a Jack polynomial.
572
Table 4
The Jack superpolynomials of weight || 3
Weight Partition
||
Sector Superpartition
m
Jack superpolynomial
(m)
J (z, ; 1/)
(0)
(0; 0)
m(0;0)
(1)
(1; 0)
(0; 1)
(1, 0; 0)
m(1;0)
m(0;1)
m(1,0;0)
(12 )
(1; 1)
(0; 12 )
(1, 0; 1)
m(1;1)
m(0;12)
m(1,0;1)
(2; 0)
m(2;0) + 1+
m(1;1)
(0; 2)
m(0;2) + 1+
m(0;12 )
(2, 0; 0)
m(2,0;0) + 1+
m(1,0;1)
1
2
(1; 12 )
(0; 13 )
(1, 0; 12 )
m(1;12)
m(0;13)
m(1,0;12)
(2; 1)
2
m(2;1) + 1+2
m(1;12)
(1; 2)
2
m(1;2) + 1+2
m(1;12)
(0; 2, 1)
(2, 1; 0)
2
6
m(0;2,1) + 1+2
m(1;12 ) + 1+2
m(0;13 )
m(2,1;0)
(2, 0; 1)
2
m(2,0;1) + 1+2
m(1,0;12)
(1, 0; 2)
2
m(1,0;2) + 1+2
m(1,0;12)
(3; 0)
m(3;0) + 2+
m(2;1) + 2+
m(1;2)
(0; 3)
2
+ (1+)(2+)
m(1;12)
3
2
m(0;3) + 2+ m(2;1) + 2+
m(0;2,1) + 2+
m(1;2)
(3, 0; 0)
4
6
+ (1+)(2+)
m(1;12) + (1+)(2+)
m(0;13)
2
m(3,0;0) + 2+ m(2,1;0) + 2+ m(2,0;1)
(2)
(13 )
(2, 1)
(3)
2 2
+ (1+)(2+)
m(1,0;12) + 2+
m(1,0;2)
superfunction associated to a superpartition with a 0 on the antisymmetric side appears

in the expansion of a Jack superpolynomial whose superpartition does not contain any
0 to the left of the semicolon. This information is not however encoded in Eq. (114). A
partial ordering formulated directly among superpartitions would lead to a more precise
formulation of the monomial expansion of the Jack superpolynomials. Such an ordering
has indeed been found [55]. But because its formulation is somewhat technical, it will be
presented elsewhere.
573
Table 5
Jack superpolynomials of weight || = 4
Partition Sector Superpartition
(14 )
(2, 12 )
(1; 13 )
(0; 14 )
(1, 0; 13 )
m(1;13)
m(0;14)
m(1,0;13)
(2; 12 )
3
m(2;12) + 1+3
m(1;13)
(1; 2, 1)
6
m(1;2,1) + 1+3
m(1;13 )
(0; 2, 12 )
(2, 1; 1)
(2, 0; 12 )
m(0;2,12) +
m(2,1;1)
m(2,0;12) +
(1, 0; 2, 1)
(2, 1, 0; 1)
m(2,1,0;1)
(2; 2)
(22 )
(m)
J (z, ; 1/)
12
12
1+3 m(1;13 ) + 1+3 m(0;14 )
3
1+3 m(1,0;13 )
6
m(1,0;2,1) + 1+3
m(1,0;13)
(0; 22 )
6 2
m(2;2) + 1+
m(2;12 ) + 1+
m(1;2,1) + (1+)(1+2)
m(1;13 )
2
m(0;2,2) + 1+ m(1;2,1) + 1+ m(0;2,12)
(2, 0; 2)
12
6
+ (1+)(1+2)
m(0;14) + (1+)(1+2)
m(1;13)
m(2,0;2) + 1+ m(2,1;1) + 1+ m(2,0;12) + 1+

m(1,0;2,1)
6
+ (1+)(1+2)
m(1,0;13)
(31)
(3; 1)
(1+2)
m(3;1) + 1+
m(2;2) + (2+3)
2 m(2;12 ) +
2 m(1;2,1)
2
+ 3 2 m(1;13)
(1+)
(1; 3)
m(1;3) + 1+
m(2;2) +
2
+ 3 2 m(1;13)
(1+)
2(1+)
2
m 2 + (3+4)2 m(1;2,1)
(1+)2 (2;1 )
2(1+)
(1+)
(0; 3, 1)
m(0;3,1) + 1+
m(0;2,2) + (1+)
m(2;12 ) + (1+2)2 m(1;2,1)
2(1+)
6 2
12 2
+ (3+5)
2 m(0;2,12 ) +
2 m(1;13 ) +
2 m(1;13 )
(1+)
(3, 1; 0)
(3, 0; 1)
(1,0;3)
(3, 1, 0; 0)
(1+)
(1+)
m(3,1;0) + 1+ m(2,1;1)
m(3,0;1) + 1+
m(2,0;2) + (1+2)2 m(2,1;1)
2(1+)
2
(1+2)
+ (2+3)
m
+
m
+ 3 2 m(1,0;13)
2
(1+)2 (2,0;1 )
2(1+)2 (1,0;2,1)
(1+)
m(1,0;3) + 1+
m(2,0;2)
m
2(1+)2 (2,1;1)
2
2
(3+4)
+
m
m
+ 3 2 m(1,0;13)
2 +
(1+)2 (2,0;1 )
2(1+)2 (1,0;2,1)
(1+)
m(3,1,0;0) + 1+
m(2,1,0;1)
(continued on the next page)
574
Table 5 continued
Partition Sector Superpartition
(4)
(4; 0)
(m)
J (z, ; 1/)
3
3(1+)
m(4;0) + 3+
m(3;1) + 3+
m(1;3) + (2+)(3+)
m(2;2)
2
3
6
+ (2+)(3+)
m(1;2,1) + (2+)(3+)
m(2;12 )
3
(0; 4)
6
+ (1+)(2+)(3+)
m(1;13 )
3
m(0;4) + 3+ m(3;1) + 3+
m(1;3)
4
3(1+)
6(1+)
+ 3+
m(0;3,1) + (2+)(3+)
m(2;2) + (2+)(3+)
m(0;2,2)
2
9
6
+ (2+)(3+)
m(1;2,1) + (2+)(3+)
m(2;12 )
2
12
+ (2+)(3+)
m(0;2,12)
3
(4, 0; 0)
18
24
+ (1+)(2+)(3+)
m(1;13 ) + (1+)(2+)(3+)
m(0;14 )
2
3
2
m(4,0;0) + 3+ m(3,1;0) + 3+ m(3,0;1) + 3+ m(1,0;3)
2(1+)
3 2
+ (2+)(3+)
m(2,0;2) + (2+)(3+)
m(2,1;1)
2
6
+ (2+)(3+)
m(2,0;12)
2
3
6
+ (2+)(3+)
m(1,0;2,1) + (1+)(2+)(3+)
m(1,0;13)
5. Conclusion
In this work, we have presented a number of results concerning the stCMS model: its
reformulation in terms of the exchange-operator formalism, the Lax formalism, the Dunkl
operators and an explicit construction for the conserved charges. In fact, 4N conserved
charges have been constructed, 2N bosonic and 2N fermionic ones.
However, our most important results pertain to the construction of the stCMS eigenfunctions, with particular emphasis on the subclass which we call the Jack superpolynomials
and which is a natural generalization of the Jack polynomials. In view of defining them
properly, we have introduced the pivotal concept of superpartitions providing the natural
labelling of the Jack superpolynomials. The Jack superpolynomials are then naturally defined by further imposing that they decompose in a specific manner in terms of monomial
superfunctions, a procedure that extends the standard way of defining the Jack polynomials.
In a forthcoming publication, we will present a dominance ordering on the superpartitions that suggests an exact expression for the coefficients c, in (114) and a related
determinantal formula. These results can in turn be used to demonstrate the actual existence of the Jack superpolynomials.
Numerous extensions of this work can be contemplated. The most immediate one
concerns the generalization from 2 to an arbitrary even number 2M of supersymmetries.
This is rather straightforward in the exchange-operator formalism: it suffices to set
Kij ij Kij ,
ij ij1
575
(119)
. . . ijM ,
(120)
where ija is the operator that exchanges the Grassmannian variables a and a , where
a = 1, . . . , M. The 2M generators of the supersymmetric transformations are then:

a
a
i Di ,
Q = K
(121)
i
= K

ia Di
(122)
The construction of the conserved quantities is analogous to the one we have presented
in the case with 2M = 2 supersymmetries. The construction of the eigenfunctions is also
rather direct: the Jack superpolynomials are then indexed by M fermionic sectors and a
superpartition involving M antisymmetric partitions:
(m)
(m1 ,...,mM )
J J
J(1 ,...,m1 ;m1 +1 ,...,m1 +m2 ;...;m1 ++mM +1 ,...,L ) .
(123)
Another simple way of extending the model is by adding spin degrees of freedom. Again,
this is very simple in the framework of the exchange-operator formalism, where we only
need to add an extra piece in the total exchange operator, i.e., set:
Kij = ij Kij ij .
(124)
ij interchanges the spins of particles i and j . This leads to a supersymmetric model

with spin degrees of freedom. The Lax formalism and the construction of the Dunkl
operators can be extended directly to this more general case. Note that the spectrum of a
supersymmetric model, with or without spin, is always the same as its non-supersymmetric
relative. The effect of supersymmetry on the spectrum is to increase the degeneracy of each
eigenvalue, in addition to generate fermionic states. But this does not change the fractional
statistics, i.e., the existence of a generalized Pauli principle.
Another natural generalization concerns the application of the method developed here,
to the construction of the eigenfunctions of the supersymmetric rational CMS model with
harmonic term, thereby generating the Hi-Jack (or generalized Hermite) superpolynomials.
Similarly, one could work out the super extension of the Macdonald polynomials as the
eigenfunctions of the supersymmetric RuijsenaarsSchneider model.
Coming back to the Jack superpolynomials per se, we can also point out various axes
of research stemming from this work. At first, it would be interesting to reformulate
their definition in a more abstract way. Mimicking the mathematical definition of the
Jack polynomials, this would amount to define the superpolynomials in terms of two
requirements: triangularity and orthogonality. A natural way of defining a scalar product
is not difficult to figure out. However, one would need to lift the degeneracy of the Jack
superpolynomials. A natural idea for this would be to look for combinations of the Jack
superpolynomials that are eigenfunctions of the charges In .
576
Another interesting issue would be to try to formulate a sort of superfield formalism,

by summing the Jack polynomials associated to a given superpartition over the different
fermionic sectors. This would construct what we could call a super-Jack polynomial. At
this point, this summation over the different sectors appears to be loosely defined, however.
How could we constraint the value of the relative coefficients associated to the different
fermionic sectors?
We also expect that there exists creation operators that provide the super analogues
of the operators constructed in [33]. A first trial in that direction has been presented in
Appendix B. This indeed maps simple stCMS eigenfunctions in fact, Jack polynomials, hence Jack superpolynomials specialized to the zero-fermion sector to other stCMS
eigenfunctions. However, as we have already pointed out, the action of these creation operators does not close within the set of Jack superpolynomials labelled by superpartitions.
Two other issues, both rooted in our initial motivations for studying sCMS models, are
worth mentioning. Given the rather intriguing relation that exists between the Virasoro
singular vectors and the Jack polynomials, one can naturally ask whether there is a similar
relation at the supersymmetric level. But this would presumably require the definition of a
proper super-Jack polynomial (i.e., an appropriate sum over the different fermionic sectors
of the Jack superpolynomials). Note that the potentially related singular vectors would be
those of the N = 2 superconformal algebra.
On the other hand, we have recalled in the introduction the remarkable connection
relating the dynamics of the rational CMS models and the time evolution of the rational
solutions of the Korteweg de Vries equation. A quite natural quest would be to try to find
a supersymmetric counterpart to this phenomenon.
We expect to report elsewhere on some of these other issues.
Acknowledgements
We would like to thank Luc Vinet for his collaboration at the initial stage of this work
and for his continuous support and David Snchal for some clarifications. P.D. would like
to thank the Fondation J.A Vincent for a student fellowship. Finally, we acknowledge the
financial support of FCAR and CRSNG.
Appendix A. Antisymmetric eigenfunctions

In this appendix, we construct excited states related to the second ground state
The differential representation |0
1 N makes clear the complete
0 = |0.
Thus, the exited states that behave like 0
antisymmetry of the fermionic ground state |0.
under the exchange of bosonic and fermionic variables take the following form:

) = z, 0 = (z, / ) 1 N , where Kij = Kij . (A.1)
(z,
Equivalently, we can define as:
) = (z,
) ,
(z,
ij .
where Kij = K
(A.2)
577
The antisymmetric states are eigenfunctions of the supersymmetric hamiltonian:

1 L 2

H ,
H=
2

zi + zj
zi zj
=
(A.3)
(zi i )2
(zi i zj j ) + 2
(1 + ij ),
2
zij
zij
i
i<j
i<j
, ij ij ).
= H(
the superfunctions are also eigenfunctions of the (modified)
Since ij = Kij ,
hamiltonian with an exchange term:
= H
,
H
K
(A.4)
where the prime symbol indicates that the sign of the coupling constant is reversed:
f () = f ().
Like the symmetric superfunctions, the antisymmetric superfunctions of degree (m; n)
can be indexed by a superpartition and denoted J(m) . These solutions will be called
the antisymmetric Jack superpolynomials. They can be defined in terms of a triangular
decomposition in antisymmetric monomial superfunctions:

(m)
(m)
(z, ) +
c, ()a
(z, ),
J(m) (z, ; 1/) = a
(A.5)
<
where
(m)
a = a(s ;a ) ,

=
i1 ,...,im ms (zi1 , . . . , zim )aa (z/{zi1 , . . . , zim })
(A.6)
i1 <i2 <<im
where a is the antisymmetric monomial function defined in (113). The eigenvalues are
given by:

=
=
2j (N + 1 2j )j .
(A.7)
Appendix B. Eigenfunction generators

by applying some differential operators
It is possible to generate eigenfunctions of H
(m)
directly on Jack polynomials. Let us consider the operator B , indexed by a positive
integer m and a partition , given by:

B(m) =
P (1,...,m) D
P (1,...,N) ,
P SN

1
N
D
(j1 ,...,jN ) = (Dj1 ) (DjN ) .
(B.1)
i is the Dunkl operator defined in (86). The partition is used to identify

In this formula, D
i . The B(m) s commute with the hamiltonian with an
the independent polynomials in D
578
exchange term and are invariant under the action of the exchange operators Kij :
(m)

B , HK = 0 = B(m), Kij .
(B.2)
(m)
These properties ensure that the application of B on the Jack polynomial J (z; 1/)
gives a solution to the supersymmetric model, the solution being indexed by two partitions
and :
(m)
A , (z, ; 1/) B(m) J (z; 1/),

(m)
(m)
H A , (z, ; 1/) = HK B(m) J = B(m) HK J = A , (z, ; 1/).
(B.3)
In the previous equation, we have used the equivalence between the supersymmetric
hamiltonian H H and the hamiltonian with an exchange term HK when they act on
(m)
a function which is invariant under the operators Kij . The solutions A , are of degree
(m, ||) in z. We can easily check that these symmetric superfunctions are non-zero only
if the partitions are strictly decreasing with respect to their m first parts.
(m)
The operators B can thus generate any H eigenfunctions. They play the double role
of fermionic creation operators and of symmetrizer in and in z. Moreover, they generalize
the supersymmetric charges introduced in Section 3.3. For example:
(1)
(n) .
K B(n) = Q
(B.4)
Given the infinite number of partitions , there exists an infinite number of such operators,
even in a specific fermionic sector. Obviously, since the number of Jack superpolynomials
(m)
is finite for a given degree m, most of the solutions A , are linearly dependent.
(m)
There is however no precise relation between a given eigenfunction A , and the Jack
superpolynomials defined in (114) and indexed by a superpartition.
Appendix C. Combinatorics of superpartitions

We now determine the number p(m; n) of independent states of degree (m; n). This
amounts to counting the number of superpartitions of degree (m; n). As we will show, this
number is simply given by the following sum:

(m)
pD (na )p(ns ),
p(m; n) =
(C.1)
na +ns =n
where na = |a | and ns = |s |. As usual p(n) is the number of partitions of n so that p(ns )

(m)
counts the number of partitions of s . The quantity pD (na ) is a restricted partition whose
restriction symbols agree with the usual definitions of combinatorial analysis:
p(m) (n) = number of partitions of n of length less or equal to m,
pD (n) = number of partitions of n whose parts are all distinct.
(C.2)
For instance, p(2) (4) = 3 since there are 3 partitions of 4 with at most 2 parts: (4), (3, 1)
(m)
(n) gives the number of partitions (without zero) of n whose parts
and (2, 2). Hence, pD
579
are strictly decreasing and whose length is smaller or equal to m. This quantity counts
the number of antisymmetric partitions a . The summation takes into account the various
ways of splitting n into the two integers na and ns . Note that, since the superpartitions of
the form (; s ) = (s ) are acceptable, we have to adopt the following conventions:
p(0) (0) = 1
p(0) (n 1) = 0.
and
(C.3)
For example, we find 5 superpartitions of weight 5 in the 2-fermion sector:

(3, 0; 0),
(2, 1; 0),
(2, 0; 1),
(1, 0; 2),
(1, 0; 1, 1).
(C.4)
This agrees with Eq. (C.1):

(2)
pD
(na )p(ns ),
p(2; 3) =
na +ns =3
(2)
(2)
(2)
(2)
= pD (3)p(0) + pD (2)p(1) + pD (1)p(2) + pD (0)p(3),

= 2 1 + 1 1 + 1 2 + 0 3 = 5.
(C.5)
If we compare the number of superpartitions of degree (m; n) for n fixed but for different
values of m, we see that this number is maximum in the one-fermion sector and minimum
in the N -fermion sector (since N corresponds to the maximal length of the superpartition).
Since the antisymmetric partitions are strictly decreasing, we have:
(N 1)(N 2)
.
(C.6)
2
For the superpolynomials of degree 3 in z and in , for instance, only one state exists
because there is only one possible superpartition, (2, 1, 0). This is less than the p(3) = 3
possible partitions of 3. On the other hand, if the fermionic sector is m = 1, we have p(1; 3)
independent states, each associated to one of the 7 possible superpartitions:
p(N; n) = 0
(3; 0),
if n
(0; 3),
(2; 1),
(1; 2),
(0; 2, 1),
(1; 1, 1),
(0; 1, 1, 1).
(C.7)
References
[1] F. Calogero, Ground state of one-dimensional N body system, J. Math. Phys. 10 (1969) 2197;
F. Calogero, Solution of a three body problem in one dimension, J. Math. Phys. 10 (1969) 2191;
F. Calogero, Solution of the one-dimensional N body problems with quadratic and/or inversely
quadratic pair potentials, J. Math. Phys. 12 (1971) 419.
[2] J. Moser, Three integrable Hamiltonian systems connected with isospectral deformations, Adv.
Math. 16 (1975) 197.
[3] B. Sutherland, Quantum many body problem in one dimension: ground state, J. Math. Phys. 12
(1971) 246;
B. Sutherland, Exact results for a quantum many body problem in one dimension, Phys.
Rev. A 4 (1971) 2019;
B. Sutherland, Exact results for a quantum many body problem in one dimension II, Phys.
Rev. A 5 (1972) 1372.
580
[4] M.A. Olshanetsky, A.M. Perelomov, Classical integrable finite dimensional systems related to
Lie algebras, Phys. Rep. 71 (1981) 313;
M.A. Olshanetsky, A.M. Perelomov, Quantum integrable systems related to Lie algebras, Phys.
Rep. 94 (1983) 313.
[5] H. Airault, H.P. McKean, J. Moser, Rational and elliptic solutions of the Korteweg de Vries
equation and a related many-body problem, Commun. Pure Appl. Math. 30 (1) (1977) 95.
[6] M.V.N. Murthy, R. Shankar, Thermodynamics of a one-dimensional ideal gas with fractionnal
exclusion statistics, Phys. Rev. Lett. 73 (1994) 3331.
[7] D. Bernard, Some simple (integrable) models of fractional statistics, hep-th/9411017.
[8] A.P. Polychronakos, Generalized statistics in one dimension, hep-th/9902157.
[9] R.E. Prange, S.M. Girvin, The quantum Hall effect, 2nd edn., Springer Verlag, 1990.
[10] H. Azuma, S. Iso, Explicit relation of the quantum Hall effect and the CalogeroSutherland
model, Phys. Lett. B 331 (1994) 107.
[11] E. Fradkin, Field theory of condensed matter systems, AddisonWesley, 1991.
[12] S. Ouvry, On the relation between anyon and Calogero models, cond-mat/9907239.
[13] F.D. Haldane, Fractional statistics in arbitrary dimensions: a generalization of the Pauli
principle, Phys. Rev. Lett. 67 (1991) 937.
[14] F.D. Haldane, Exact JastrowGutzwiller resonating valence bond ground state of the spin 1/2
antiferromagnetic Heisenberg chain with 1/R 2 exchange, Phys. Rev. Lett. 60 (1988) 635.
[15] B. Sriram Shastry, Exact solution of an S = 1/2 Heisenberg antiferromagnetic chain with long
ranged interactions, Phys. Rev. Lett. 60 (1988) 639.
[16] F.D.M. Haldane, Z.N.C. Ha, J.C. Talstra, D. Bernard, V. Pasquier, Yangian symmetry of
integrable quantum chains with long-range interactions and a new description of states in
conformal field theory, Phys. Rev. Lett. 69 (1992) 2021.
[17] A.P. Polychronakos, Exchange operator formalism for integrable systems of particles, Phys.
Rev. Lett. 69 (1992) 703;
J.A. Minahan, A.P. Polychronakos, Integrable systems for particles with internal degrees of
freedom, Phys. Lett. B 302 (1993) 299, hep-th/9206046;
J.A. Minahan, A.P. Polychronakos, hep-th/9202057.
[18] H. Frahm, Spectrum of a spin chain with inverse square exchange, J. Phys. A 26 (1993) L473.
[19] A.P. Polychronakos, Exact spectrum of SU(n) spin chain with inverse square exchange, Nucl.
Phys. B 419 (1994) 553, hep-th/9310095.
[20] P. Fendley, F. Lesage, H. Saleur, Solving the 1D log gas using Jack polynomials and functional
relations, J. Stat. Phys. 79 (1995) 799, hep-th/9405008.
[21] B.D. Simons, P.A. Lee, B.L. Altshuler, Matrix models, one-dimensional fermions, and quantum
chaos, Phys. Rev. Lett. 72 (1994) 64.
[22] M. Caselle, Distribution of transmission eigenvalues in disordered wires, Phys. Rev. Lett. 74
(1995) 2776.
[23] G.W. Gibbons, P.K. Townsend, Black holes and Calogero models, Phys. Lett. B 454 (1999)
187.
[24] E. DHoker, D.H. Phong, SeibergWitten theory and CalogeroMoser systems, Prog. Theor.
Phys. Suppl. 135 (1999) 75, hep-th/9906027.
[25] C.F. Dunkl, Differential-difference operators associated to reflection groups, Trans. Am. Math.
Soc. 311 (1989) 1.
[26] H. Jack, A class of symmetric polynomials with a parameter, Proc. R. Soc. Edinburgh Sect.
A 69 (1970/1971) 118;
H. Jack, A surface integral and symmetric functions, Proc. R. Soc. Edinburgh Sect. A 69 (4)
(1972) 347364.
[27] I. Cheredenik, Double affine Hecke algebras and Macdonalds conjectures, Ann. Math. 141
(1995) 191;
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
581
I.G. Macdonald, Affine Hecke algebras and orthogonal polynomials, Sminaire Bourbaki,
Vol. 1994/95, Astrisque 247 (1996) 189.
S. Fomin, A.N. Kirillov, Quadratic algebras, Dunkl elements and Schubert calculus, Prog.
Math. 172 (1999) 147.
M. Flower, J.A. Minahan, Invariants of the HaldaneShastry SU(N) chain, Phys. Rev. Lett. 70
(1993) 2325, cond-mat/9208016.
D. Bernard, M. Gaudin, F.D. Haldane, V. Pasquier, YangBaxter equation in long range
interacting system, J. Phys. A 26 (1993) 5219, hep-th/9301084.
R.P. Stanley, Some combinatorial properties of Jack symmetric functionsl, Adv. Math. 77
(1988) 76115.
P.J. Forrester, Selberg correlation integrals and the 1/r 2 quantum many-body system, Nucl.
Phys. B 388 (1992) 671.
L. Lapointe, L. Vinet, A Rodrigues formula for the Jack polynomials and the Macdonald
Stanley conjecture, Int. Math. Res. Notices 1995 (9) (1995) 419424;
L. Lapointe, L. Vinet, Exact operator solution of the CalogeroSutherland model, Commun.
Math. Phys. 178 (2) (1996) 425452.
H. Awata, Y. Matsuo, S. Odake, J. Shiraishi, A note on CalogeroSutherland model, W (n)
singular vectors and generalized matrix models, hep-th/9503028;
H. Awata, Y. Matsuo, S. Odake, J. Shiraishi, Excited states of CalogeroSutherland model and
singular vectors of the W (N) algebra, Nucl. Phys. B 449 (1995) 347, hep-th/9503043.
K. Hikami, M. Wadati, Infinite symmetry of the spin systems with inverse square interactions,
J. Phys. Soc. Japan 62 (12) (1993) 42034217;
K. Hikami, M. Wadati, Integrable systems with long range interactions, w-infinity algebra and
energy spectrum, Phys. Rev. Lett. 73 (1994) 1191.
F. Knop, S. Sahi, A recursion and a combinatorial formula for the Jack polynomials, Invent.
Math. 128 (1997) 9.
L. Lapointe, A. Lascoux, J. Morse, Determinantal formula and recursion for Jack polynomials,
Electron. J. Comb. 7 (2000) 467.
H. Ujino, M. Wadati, Rodrigues formula of the hi-Jack symmetric polynomials associated with
the quantum Calogero model, J. Phys. Soc. Japan 65 (1996) 2423.
I.G. Macdonald, Symmetric functions and Hall polynomials, 2nd edn., Clarendon Press, Oxford
Univ. Press, 1995;
Symmetric Functions and Orthogonal Polynomials, Dean Jacqueline B. Lewis Memorial
Lectures presented at Rutgers University, American Mathematical Society, 1998.
L. Lapointe, L. Vinet, Rodrigues formula for the Macdonald polynomials, Adv. Math. 130
(1997) 261.
S.N. Ruijsenaars, H. Schneider, A new class of integrable systems and its relation to solitons,
Ann. Phys. 170 (1986) 370;
S.N. Ruijsenaars, Complete integrability of relativistic CalogeroMoser systems and elliptic
function identities, Commun. Math. Phys. 110 (1987) 191.
H. Awata, Hidden algebraic structure of the CalogeroSutherland model, integral formula for
Jack polynomial and their relativistic analog, in: J.F. van Diejen, L. Vinet (Eds.), Calogero
MoserSutherland Models, Springer, 2000, p. 23.
D.Z. Freedman, P.F. Mende, An exactly solvable N particle system in supersymmetric quantum
mechanics, Nucl. Phys. B 344 (1990) 317.
E. Witten, Dynamical breaking of supersymmetry, Nucl. Phys. B 188 (1981) 513;
E. Witten, Constraints on supersymmetry breaking, Nucl. Phys. B 202 (1982) 253.
B. Sriram Shastry, B. Sutherland, Superlax pairs and infinite symmetries in the 1/r 2 system,
Phys. Rev. Lett. 70 (1993) 4029, cond-mat/9212029.
L. Brink, T.H. Hansson, S. Konstein, M.A. Vasiliev, The Calogero model: anyonic representation, fermionic extension and supersymmetry, Nucl. Phys. B 401 (1993) 591, hep-th/9302023.
582
[47] L. Brink, A. Turbiner, N. Wyllard, Hidden algebras of the (super) Calogero and Sutherland
models, J. Math. Phys. 39 (1998) 1285, hep-th/9705219.
[48] N. Wyllard, (Super)conformal many-body quantum mechanics with extended supersymmetry,
J. Math. Phys. 41 (2000) 2826, hep-th/9910160.
[49] A.J. Bordner, N.S. Manton, R. Sasaki, CalogeroMoser models. V: Supersymmetry and
quantum lax pair, Prog. Theor. Phys. 103 (2000) 463, hep-th/9910033.
[50] P.K. Ghosh, Super-CalogeroMoserSutherland systems and free super-oscillators: a mapping,
hep-th/0007208.
[51] P. Dargis, P. Mathieu, Nonlocal conservation laws for supersymmetric KdV equation, Phys.
Lett. A 176 (1993) 67, hep-th/9301080.
[52] H. Nicolai, Supersymmetry and spin systems, J. Phys. A 9 (1976) 1497;
H. Nicolai, Extensions of supersymmetric spin systems, J. Phys. A 10 (1977) 2143.
[53] F. Cooper, A. Khare, U. Sukhatme, Supersymmetry and quantum mechanics, Phys. Rep. 251
(1995) 267, hep-th/9405029.
[54] P. Mathieu, Y. Xudous, Conserved charges of non-yangian type for the FrahmPolychronakos
spin chain, hep-th/0008036.
[55] P. Desrosiers, L. Lapointe, P. Mathieu, Jack superpolynomials, superpartition ordering and
determinantal formulas, hep-th/0105107.

Nonlinear supersymmetry, quantum anomaly

and quasi-exactly solvable systems
Sergey M. Klishevich a,b , Mikhail S. Plyushchay a,b
a Departamento de Fsica, Universidad de Santiago de Chile, Casilla 307, Santiago 2, Chile
b Institute for High Energy Physics, Protvino, Russia
Received 11 December 2000; accepted 23 April 2001
Abstract
The nonlinear supersymmetry of one-dimensional systems is investigated in the context of the
quantum anomaly problem. Any classical supersymmetric system characterized by the nonlinear
in the Hamiltonian superalgebra is symplectomorphic to a supersymmetric canonical system with
the holomorphic form of the supercharges. Depending on the behaviour of the superpotential, the
canonical supersymmetric systems are separated into the three classes. In one of them the parameter
specifying the supersymmetry order is subject to some sort of classical quantization, whereas the
supersymmetry of another extreme class has a rather fictive nature since its fermion degrees of
freedom are decoupled completely by a canonical transformation. The nonlinear supersymmetry
with polynomial in momentum supercharges is analysed, and the most general one-parametric
Calogero-like solution with the second order supercharges is found. Quantization of the systems
of the canonical form reveals the two anomaly-free classes, one of which gives rise naturally to
the quasi-exactly solvable systems. The quantum anomaly problem for the Calogero-like models is
cured by the specific superpotential-dependent term of order h 2 . The nonlinear supersymmetry
admits the generalization to the case of two-dimensional systems. 2001 Elsevier Science B.V. All
rights reserved.
PACS: 03.65.-w; 11.10.Lm; 11.30.Pb
1. Introduction
After introducing the supersymmetric quantum mechanics as a toy model for studying
the supersymmetry breaking mechanism [1,2], it was applied for solving many problems
in theoretical and mathematical physics [3,4]. The most recent applications of the
supersymmetric quantum mechanics can be found in the dynamics of D-branes and black
holes [5,6], in M-theory and matrix models [7], in the theory of integrable systems [4] and
fluid mechanics [8,9].
E-mail addresses: sklishev@lauca.usach.cl (S.M. Klishevich), mplyushc@lauca.usach.cl (M.S. Plyushchay).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 9 7 - 3
584
S.M. Klishevich, M.S. Plyushchay / Nuclear Physics B 606 [PM] (2001) 583612
Some time ago it was observed that the pure parabosonic [10] (and parafermionic [11])
systems possess the supersymmetry characterized by the nonlinear superalgebra. Such nonlinear supersymmetry takes place in the simple quantum mechanical system generalizing
the usual superoscillator [10], and earlier it was revealed in a similar but particular form in
the fermion-monopole system [12,13], and in the P , T -invariant systems of planar fermions [14] and ChernSimons fields [15]. The algebraic structure of the nonlinear supersymmetry resembles the structure of the finite W -algebras [16] for which the commutator of
any two generating elements is proportional to a finite order polynomial in them.
As it was noted in Ref. [10], in a generic case under attempt of constructing the quantum
analogue of the classical systems possessing the nonlinear supersymmetry one faces the
problem of the quantum anomaly [1719]. To resolve this problem, in the present paper
we investigate the nonlinear supersymmetry of one-dimensional systems at the classical
and quantum levels. This will allow us to reveal the unexpected very close relation of the
nonlinear supersymmetry with associated quantum anomaly problem to the quasy-exactly
solvable systems [2024] being related, in turn, to the conformal field theory [2529].
The results will give also a new perspective on the known quantum supersymmetry
characterized by the second order supercharges [30].
The paper is organized as follows. Section 2 is devoted to the detailed investigation
of the various aspects of the classical one-dimensional systems possessing the nonlinear
supersymmetry of the most general form. In Section 3 we consider their quantization,
and reveal the two classes of anomaly-free quantum systems possessing the nonlinear
supersymmetry of arbitrary order k Z+ . One of them turns out to be closely related
to the quasy-exactly solvable systems and this aspect of the nonlinear supersymmetry
is investigated in Section 4. Section 5 discusses the general quantum case of the k = 2
supersymmetry in the context of anomaly-free quantization of the class of k = 2 Calogerolike supersymmetric systems found and analyzed in Section 2. We show also how this k = 2
supersymmetry can be used for constructing new exactly solvable systems. In Section 6
the brief summary of the obtained results is presented and some open problems to be
interesting for further investigation are discussed. In particular, we point out how the
nonlinear supersymmetry can be generalized to the case of the two-dimensional classical
and quantum systems.
2. Classical supersymmetry
In this section we investigate the classical supersymmetry of the most general form
realizable in one-dimensional bosonfermion system. We shall show that in generic case
the supersymmetry is characterized by a nonlinear Poisson algebra and includes the usual
supersymmetry as a particular case. Analysing the structure of the supersymmetry from
the viewpoint of canonical transformations, we shall observe the existence of the three
essentially different classes: in the first class the parameter characterizing the order of
superalgebra is subject to the classical quantization, in the second (intermediate) class
the supercharges Poisson bracket can be equal to any real nonnegative degree of the
585
Hamiltonian, whereas the systems of the third class allow ones the complete classical
decoupling of the fermion from the boson degrees of freedom.
2.1. General structure of supersymmetry
Following Ref. [31], let us consider a non-relativistic particle in one dimension described
by the Lagrangian
1
L = x 2 V (x) L(x)N + i + ,
2
(2.1)
where are the Grassmann variables, ( + ) = , N = + , and V (x) and L(x) are
two real functions. The nontrivial PoissonDirac brackets for the system are {x, p} = 1
and { + , } = i, and the Hamiltonian is
1
H = p2 + V (x) + L(x)N.
2
The latter generates the equations of motion
x = p,
p = V (x) L (x)N,
(2.2)
= iL(x) .
The Hamiltonian H and the nilpotent quantity N are the even integrals of motion for any
choice of the functions V (x), L(x), whereas the odd quantities
+
Q = B (x, p) ,
(2.3)
B
= B ,
are the integrals of motion when the differential equations

V (x)
iL(x) B (x, p) = 0
p
x
p
(2.4)
have the solutions being regular functions in the corresponding domain of the phase space
defined by V (x) and L(x). It is obvious that such odd integrals can exist only for a special
choice of the functions V (x) and L(x). Let us investigate this question in detail and
restrict ourselves to the physically interesting class of the systems given by the potential
V (x) bounded from below. Such a potential can generally be represented in terms of a
superpotential W (x) and real constant v:
1
V (x) = W 2 (x) + v.
2
The condition of regularity of B (x, p) at p = 0 leads to the relation L(x) = W (x)(x)
with some function (x). Having in mind that for the functions B (x, p) the substitution
p p is equivalent to the complex conjugation, one can represent them as B (x, p) =
B(W (x), ip). Then the Hamiltonian and Eq. (2.4) take the form
1
1
)N,
H = p2 + W 2 (x) + v + W (x)(W
2
2

) B(W, ip) = 0,
W
+ i (W
p
W
p
(2.5)
(2.6)
586
(x)) = (x). In terms of the complex variables

where (W
z = W (x) + ip,
z = W (x) ip
(2.7)
Eq. (2.6) is represented as

z + z
z z zz +
B(z, z ) = 0.
2
(2.8)
General solution to Eq. (2.8) is given by

cos ) d ,
B(, ) = f () exp i (
(2.9)
where f () is an arbitrary function and z = ei . The simplest case (x) = 0 with

f () = corresponds to the holomorphic solution of Eq. (2.8), B(z) = z , whereas the
case 0 with f () = gives the antiholomorphic solution B(z) = z , both regular
at z = z = 0. If the superpotential W (x) is the unbounded function, one of the functions
B(z) = z or B(z) = z is well defined on the whole complex plane only for Z, i.e.,
we have here some sort of classical quantization [32]. This simplest solution with Z
corresponds to the nonlinear supersymmetry investigated in Ref. [10], and includes the
usual linear supersymmetry with || = 1 as a particular case.
According to the definition (2.3), the functions B (x, p) are defined up to an additive
nilpotent term proportional to 1 N . This allows ones to represent the supercharges in the
equivalent form
Q = f (H )e
i(p,W (x))
p, W (x) =
cos ) d.
(
(2.10)
We suppose here that like in the case (x) = , the function f (H ) is chosen in the simplest
form compatible with the requirement of regularity of the supercharges (see below). The
supercharges (2.10), the Hamiltonian (2.5) and the nilpotent integral N form generally the
nonlinear superalgebra

+
N, Q = iQ ,
Q , Q = if 2 (H ),

Q , H = 0,
{N, H } = 0.
2.2. Three cases of supersymmetry and classical quantization
In the case of unbounded superpotential W (x), the dynamics of the system projected on
the unity of the Grassmann algebra is defined on the whole complex plane C. In order to
have the supercharges (2.10) to be well defined single-valued observables on C, we have
1 At the quantum level the terms N , N 2 , etc., do not vanish and may be essential for preserving the
supersymmetry.
587
to impose the condition

2
cos ) d = 2k,
(
k Z.
The function is supposed to be regular and can be represented in the form

W (x) = k + W (x)M W 2 (x) ,
(2.11)
where M(W 2 ) is an arbitrary function, |M(0)| < . As a consequence, the supercharge

can be written as
Q+ = zk ei
p
0
M(p 2 y 2 +W 2 (x)) dy +
(2.12)
Here we suppose that k is nonnegative; in the case of negative k Eq. (2.12) with substitutions k k, + gives the supercharge Q = (Q+ ) . The exponential factor in
Eq. (2.12) can be removed by the transformation
= eiG(p, x) ,
X = x + p G(p, x)N,
P = p x G(p, x)N, (2.13)
that is the canonical transformation with generating function G obeying the differentiability condition
p x G(p, x) = x p G(p, x).
In the case of general superpotential W (x), the transformation (2.13) can be used to
reduce the classical Hamiltonian (2.5) to the most simple form, e.g., to the form with
L(x) = W (x), = const. This gives the following equation for the function G:

W (x)W (x)
p
(2.14)
G(x, p) + W (x) W (x) = 0.
x
p
Though Eq. (2.14) has exactly the form of that for log B with shifted (see Eq. (2.6)),
there is a difference: we require for G to be regular and single-valued function on the whole
physical domain, whereas the same condition of regularity is imposed on the function B,
but not on log B. The general solution to Eq. (2.14) is

G(p, x) =

cos ) d,
(
and its behaviour depends on the physical domain for z defined, in turn, by the properties
of the superpotential W (x). Here it is necessary to separate the three different cases and
the results can be summarized as follows.
1. The physical domain in terms of z includes the origin (a < W (x) < b, a < 0, b >
0). This, in particular, corresponds to the case of unbounded superpotential with
a = , b = +. From the regularity of G in such a domain it follows that =
) can be decomposed into the Taylor series at

(0).
We assume that the function (W
W = 0. From the regularity of B we arrive at the classical quantization condition
588
). Thus, the most

= k, k Z, and at the restriction (2.11) on the function (W
general form of the Hamiltonian admitting the nonlinear supersymmetry is

p2 1 2
+ W (x) + v + W (x) k + W (x)M W 2 (x) N,
k Z, (2.15)
2
2
whose associated supercharges have
been described above. By the canonical
p
transformation (2.13) with G(p, x) = 0 M(p2 y 2 + W 2 (x)) dy (supplemented by
the transformation in the case k < 0) we can always reduce the system
with this Hamiltonian and the supercharge (2.12) to the form of the supersymmetric
system possessing the holomorphic supercharge:
H=
1
1
H = P 2 + W 2 (X) + v + kW (X) + ,
2
2
Q+ = Z k + , k Z+ ,
(2.16)
where Z = W (X) + iP . The presence of the quantized, integer number k in the

Hamiltonian (2.16) means that the instant frequencies of the oscillator-like odd, ,
variables are commensurable. Only in this case the regular odd
and even, Z, Z,
integrals of motion can be constructed, and the factor Z k in the supercharge Q+
corresponds to the k-fold conformal mapping of the complex plane (or the strip a <
Re Z < b) on itself (or on the corresponding region in C).
2. The physical domain is defined by the condition Re z 0 (or Re z 0) and also
includes the origin of the complex plane. But unlike the previous case, there are no
closed contours around z = 0. As a consequence, though the regularity of G results in
the same relation = (0),

no quantization condition appears from the regularity
of B. The most general form of the Hamiltonian and the supercharge is
H=

p2 1 2
+ W (x) + v + W (x) + R W (x) N,
2
2
Q+ = z e
R( cos ) d +
(2.17)
where we assume that R, and the function R(W ) is analytical at W = 0 and

R(0) = 0. The singularity in Q+ at the origin z = 0 for < 0 is not physical and
can be removed multiplying Q+ by (zz) , that results in changing the holomorphic
function B(z, z ) = z for the antiholomorphic function B(z, z ) = z to be regular
at
G(p, x) =
z = 0. After the canonical transformation (2.13) with the function
0 R( cos ) d (supplemented by the transformation for < 0), the

Hamiltonian and the supercharge can be reduced to the form
1
1
H = P 2 + W 2 (X) + v + W (X) + ,
2
2
Q+ = Z + , R+ .
(2.18)
3. The physical domain is defined by the condition Re z > 0 (or Re z < 0), i.e., the origin
of the complex plane is not included. In this case and (0)

are not related since the
function G admits in such a domain the terms proportional to arg z. Therefore, though
589
the general form of the Hamiltonian and the supercharge is

H=
p2 1 2
)N,
+ W + v + W (W
2
2
Q+ = f (H )e
cos ) d +
(
by the canonical transformations (2.13) with function G(p, x) =

one can reduce the Hamiltonian to the form
1
1
H = P 2 + W 2 (X) + v
2
2
(2.19)
cos ) d,
(
(2.20)
with trivial dynamics for the Grassmann variables , = 0, playing the role of
the supercharges. Thus, classically the supersymmetry of any system with bounded
nonvanishing superpotential has a fictive nature.
The obtained classification of the classical supersymmetric systems emerged from the
aim to present the Hamiltonian in the most simple form when the superpotential W (x) has
the definite behaviour defining the type of the physical domain in the complex plane. On
the other hand, one can consider the supersymmetry from the viewpoint of the functional
dependence of the Hamiltonian on W (x) without specifying the superpotentials type. Then
the Hamiltonian and the supercharges
1
1
H = p2 + W 2 (x) + v + kW (x) + ,
2
2
Q = z k , k Z+
Q+ = zk + ,
(2.21)
give the intersection of the described three classes of the systems, and can be treated
as the representatives of the more broad classes of the supersymmetric systems (2.15),
(2.17) and (2.19), with which (2.21) is related by the corresponding symplectomorphism.
For (2.21) the constant k characterizes the degree of nonlinearity of the associated
superalgebra and one can refer to it as to the system with k-supersymmetry. At the same
time, it is necessary to bare in mind that in the case of the superpotential of the third class
all the systems (2.21) with different k = 0 are symplectomorphic to the system with k = 0.
The three types of supersymmetric Hamiltonians (2.15), (2.17), (2.18) are defined up to
the additive constant v. This arbitrariness could be used to present the potential V (x) in
terms of other superpotential via the relation
2 (x) = W 2 (x) + const.
W
(2.22)
(x) can correspond to different types of

Generally, the superpotentials W (x) and W
nonlinear supersymmetry. Then the natural question is: can such a transition change the
type of supersymmetry when the form of the nilpotent term of the Hamiltonian has been
already fixed? In other words, the question is if the described classification of classical
supersymmetric Hamiltonians has an invariant sense. The invariance of the classification
is demonstrated in Appendix A. At the same time, it is necessary to stress that the systems
related by the canonical transformation have not to be equivalent on the quantum level
590
due to the ordering problem and because the canonical transformations are, as a rule,
nonpolynomial in momenta.
For the sake of completness, let us discuss the Lagrangian formulation for the ksupersymmetric system (2.21). Its Lagrangian is
1
1
L = x 2 W 2 (x) v kW (x) + + i + .
2
2
(2.23)
In the Hamiltonian formulation the supertransformations of the variables x and are

generated canonically by the supercharges:

x = x, Q+ x, Q + = ik zk1 + + z k1 + ,

+ = + , Q + = izk + ,
= , Q+ = i z k .
Using the equations of the motion, we obtain the corresponding supertransformations at
the Lagrangian level:
k
k1 + + k1 +
+ A
,
= i A ,
x = ik A
The Lagrangian (2.23) is quasi-invariant under these supertranswhere A = W (x) i x.
formations:
d k1 + + k1 +
ik x A
+ A
.
L =
dt
It is worth noting that on shell the commutator of the two supertransformations for the
physical variables is proportional to the translation in time:

d
[1 , 2 ]X = i2k kE k1 x, x,
1 2+ 2 1+
X,
dt
) is the energy function of the system. Therefore, the
where X is x or , and E(x, x,
supertransformations form an open algebra with the structure functions depending on the
physical variables.
2.3. Supersymmetry with polynomial supercharges
We have arrived at the three types of Hamiltonians (2.15), (2.17), (2.18), which after
appropriate canonical transformations can be reduced to the more simple form with the
associated supercharges represented in the holomorphic or antiholomorphic form. On
the other hand, as it was mentioned above, the quantization of canonically equivalent
classical systems can give in some cases the quantum systems with different types
of supersymmetry, even of different order k. Therefore, the search for other special
representations for the Hamiltonian and associated supercharges is important. Based on
this remark, let us look for the representation in which the function B(x, ip) defining the
supercharges is the polynomial in p of the degree k, i.e.,
B(x, ip) =
k

n=0
bkn (x)(ip)n .
591
Substituting this into Eq. (2.4), we obtain the recurrent equation

bn (x) + (k n + 2)bn2 (x)V (x) L(x)bn1 (x) = 0,
(2.24)
where bn (x) = 0 for n < 0 and n > k is assumed. Due to the equation b0 (x) = 0, one can
fix b0 (x) = 1. Then the part of Eq. (2.24) can be solved giving for L(x) and V (x) the
relations

1 b1 (x)2
b2 (x) + v.
V (x) =
L(x) = b1 (x),
k
2
To simplify the notation, we put b1 (x) = y(x) and realize the change of the variables

x = x(y),
n (y) = bn x(y) , for n > 1.
(2.25)
If the function inverse to y(x) does not exist globally, we can perform this transformation
separately on each interval where the function y(x) is monotonic. Under the transformation (2.25), Eq. (2.24) acquires the form of the system of differential equations in the
variable y:

kn+2
n2 (y) 2 (y) y n1 (y) = 0.
(2.26)
k
A general solution to this system has k 1 real parameters. If we know solution to the
system (2.26), we could find the form of the functions bn (x), at least implicitly. This
means that the general solution to Eq. (2.24) depends on arbitrary function b1 (x) but
not on its derivatives, and which can be called the superpotential. One notes also that the
system (2.26) is invariant under the transformation y y if the functions n obey the
relation
n (y)
n (y) = (1)n n (y).
(2.27)
The solution corresponding to the holomorphic case is of this type.

The simplest case k = 1 corresponds to the usual linear supersymmetry, and we turn to
the case k = 2. For k = 2, from (2.26) we obtain the equation for 2 :
y2 + 22 = y 2 .
This equation has the solution
2 =
y 2 4c
+ 2,
4
y
where c is an arbitrary real constant. Considering W (x) = y(x)/2 as a superpotential, one

arrives at the supercharges of the form

2
1
c
,
Q =
(2.28)
ip + W (x) + 2
2
W (x)
which together with the Hamiltonian

1
c
H = p2 + W 2 (x) 2
+ 2W (x)N + v
2
W (x)
(2.29)
592
form the nonlinear superalgebra

+
Q , Q = i (H v)2 + c ,

Q , H = 0.
(2.30)
Note that the Hamiltonian (2.29) has the Calogero-like form: at W (x) = x its projection
to the unit of Grassmann algebra takes the form of the Hamiltonian of the two-particle
Calogero system. The functions B (p, x) = B(x, ip) and Hamiltonian (2.29) form the
nonlinear Poisson algebra
+

B , B = 4iW H,
B , H = iW B ,
which does not depend on the constant c, and is reduced to the sl(2, R) algebra at
W (x) = x.
For c = 0 the obtained k = 2 supersymmetric system (2.28), (2.29) is reduced to the
k = 2 supersymmetric system of the form (2.21) characterized by the holomorphic form
of the supercharge. Let us investigate the relationship of the system (2.28), (2.29) with ksupersymmetry (2.21) in the case c = 0. First we show that for c < 0 (c = 2 , > 0) the
system (2.29), (2.28) can be reduced to the linear (k = 1) supersymmetric system of the
form (2.21) with the corresponding holomorphic supercharge multiplied by some function
of the Hamiltonian. To show this we represent the potential of the Hamiltonian (2.29) in
the form

2
1 2
1
W (x)
+v+ = W
(x) + v + .
V (x) =
2
W (x)
2
Without loss of generality we can consider W (x) > 0 and get the relation

1
2 (x) + 4 .
W (x) = W
(x) + W
2
(x) the function L(x) is represented as
Therefore, in terms of W

2
(x)M W
(x) ,
(x) 1 + W
L(x) = W
2 2

(x) = W
(x) + 4 1/2 .
M W
Comparing this with the factor at the nilpotent term in Hamiltonians (2.15), (2.17) and
(2.18), we find that our system corresponds to the system (2.21) with k = 1 since the term
2 (x)) can be removed by the appropriate canonical transformation.
with function M(W
Therefore, we conclude that the system (2.29) is canonically equivalent to the k = 1
supersymmetric system (2.21).
Analogously, applying the canonical transformation (2.13) to the system (2.29) with
c > 0 (c = 2 , = 0), one can reduce it to the k = 0 supersymmetric system with the
(x) by the
Hamiltonian of the form (2.20). Indeed, let us define the new superpotential W
relation
2 (x) + 2v = W 2 (x)
W
2
+ 2v = 2V (x),
W 2 (x)
where without loss of generality we assume that W (x) > 0, and the constant v has to be
2 (x) 0. As a result, one can represent W (x) in terms
chosen to provide the inequality W
of the new superpotential as

1/2
W (x) = 2
D + D2 + 4 2 ,
593
2 (x) 25v,
D=W
, we obtain L(x) W
(x)W
(x)
where 5v = v v.
Expressing L(x) in terms of W
2
2
2
2
1/2
1/2

]W
(W (x)) to be
M(W (x)), with the function M(W (x)) = [1 [D + 4 ]
regular due to the inequality W (x) > 0. Therefore the nilpotent term can be removed from
the Hamiltonian by the canonical transformation 2 (2.13) since the generating function
) is regular in this case.
G(p, W
Let us turn now to the next, k = 3 case given by the system of equations

2
1
3 y 2 y 2 = 0,
(2.31)
3 + 2 2 y = 0.
3
3
Substituting 3 from the second equation into the first one, we arrive at the nonlinear
differential equation of the second order
2 2 + (2 )2 + y2 + 22 2y 2 = 0,
which can equivalently be represented as the first order nonlinear differential equation

1 2
y 2 y 2 + 2 + y 2 2 y 2 C = 0.
2
When the integration constant C = 0, the solution to the equation is a root of the 4th order
algebraic equation for 2 (y 2 ):

3

32 y 2 y 2 + 2 C1 y 2 = 0.
y2 2 +
At C1 = 0, its solutions have a simple form: 2 = y 2 /3, 2 = y 2 . The first solution

corresponds to the holomorphic case with k = 3, while the second one does to the case of
k = 1 supersymmetry with the supercharge multiplied by the Hamiltonian. Even in the case
C = 0, C1 = 0, the corresponding solutions 2 (y 2 ) have a complicated form of solutions in
radicals of the 4th order equation, whereas for C = 0 we have not succeeded in finding any
analytical solution of the nontrivial nature. The same complications appear with finding
the nontrivial solutions to the system (2.26) for k > 3.
3. Quantum anomaly and k-supersymmetry

As we have seen in the previous section, the supersymmetry in classical one-dimensional
system is defined by the arbitrary function W (x) and in general case the supercharges
together with the Hamiltonian form a nonlinear superalgebra. According to the results of
Ref. [10] on the supersymmetry in pure parabosonic systems, a priori one can not exclude
the situation characterized by the supercharges to be the nonlocal operators represented
d
. Since such nonlocal supercharges
in the form of some infinite series in the operator dx
have to anticommute for some function of the Hamiltonian being a usual local differential
2 See Appendix A for the details.
594
operator of the second order, they have to possess a very peculiar structure. 3 Due to
this reason, we restrict ourselves by the discussion of the supersymmetric systems with
the supercharges being the differential operators of order k. Classically this corresponds
to the system (2.21) with the holomorphic supercharges or to the systems discussed in
Section 2.2.
In Ref. [10] it was observed that just in the simplest case of the superoscillator possessing
the nonlinear k-supersymmetry and characterized by the holomorphic supercharges of
the form (2.21) with the simplest superpotential W (x) = x, the form of the classical
k
superalgebra {Q+
)(H 2h ) (H
k , Qk } = H is changed for {Qk , Qk } = H (H h
h (k 1)) due to the quantum noncommutativity. Moreover, it was also observed that
for W (x) = ax + b in generic case we have a global quantum anomaly [17,18]: the
direct quantum analogue of the superoscillators loose the property of the conservation,
[Q
k , Hk ] = 0. Therefore, we arrive at the problem of looking for the classes of
superpotentials and corresponding quantization prescriptions leading to the quantum ksupersymmetric systems without quantum anomaly. This and next two sections are devoted
to the solution of such a problem.
3.1. The k-supersymmetry: quadratic superpotential
In this and next subsections we consider the quantization of the nonlinear supersymmetry characterized by the holomorphic form of the supercharges (2.21). As we shall see, the
straightforward quantization without special quantum corrections leads to the rigid restrictions on the form of the superpotential W (x).
Let us fix the quantum supercharges in the holomorphic form corresponding to the
classical k-supersymmetry,
k
Q = 2k/2 A ,
(3.1)
with
d
+ W (x).
dx
With realization
A = h
(3.2)
h
(3.3)
(3 + 1),
2
the operator N has a sense of the fermionic quantum number and being normalized for h is
the projector onto the fermionic subspace, whereas the complimentary projector onto the
bosonic subspace is + /h = 1 N h 1 . Then choosing the quantum Hamiltonian in the
form (2.2), from the requirement of conservation of the supercharges, [Q , H ] = 0, we
arrive at the equations
N = + =
L(x) = kW (x),
(3.4)
3 In pure parabosonic systems revealing k-supersymmetry both the supercharges and Hamiltonian have a
nonlocal structure [10].

1

V (x) = W 2 (x) k hW
(x) + v,
2

h 3 k k 2 1 W (x) = 0.
Therefore, the quantum system given by the Hamiltonian

1
d2

H=
h 2 2 + W 2 (x) + 2v + k h
3W
2
dx
595
(3.5)
(3.6)
(3.7)
possesses the nonlinear supersymmetry of order k 2 characterized by the holomorphic

supercharges (3.1) only when
W (x) = w2 x 2 + w1 x + w0 .
(3.8)
For any other form of the superpotential the nilpotent operators (3.1) are not conserved
that can be treated as a quantum anomaly. Below we shall see that the quantum anomaly
for the system (3.7) can be cured for some superpotentials if to modify appropriately (by
h -dependent terms) the supercharges. It is necessary to stress that the relation (3.6) fixing
the form of the superpotential for k 2 has a purely quantum nature. One also notes that
with the prescription (3.3), V (x) given by Eq. (3.5) plays the role of the total potential
for the bosonic sector, V (x) = VB (x), whereas the potential for the fermionic sector is
VF (x) = V (x) + h kW (x). The appearance in V (x) of the term proportional to h can
obviously be associated with the ordering ambiguity under construction of the operator N :
under the choice N = 12 [ + , ] instead of (3.3), we again arrive at the Hamiltonian (3.7),
but the term linear in h disappears from V (x). In what follows we shall have in mind the
quantum prescription (3.3).
The anticommutator of supercharges Q+ and Q gives the polynomial of order k in H ,
e.g., for the simplest cases k = 2, k = 3 and k = 4 we have

1
Q , Q+ = H 2 w12 4w0 w2 ,
4
+

Q , Q = H 3 w12 4w0 w2 H + 2w22 ,

2
+
5
9
Q , Q = H 4 w12 4w0 w2 H 2 + 12w22 H + w22 w12 4w0 w2 ,
2
16
where for the sake of simplicity we have put v = 0 and h = 1. The family of
supersymmetric systems (3.8) is reduced to the superoscillator at w2 = 0 with the
associated exact k-supersymmetry [10]. For w2 = 0 the k-supersymmetry is realized
always in the spontaneously broken phase since in this case the supercharges (3.1) have
no zero modes (normalized eigenfunctions of zero eigenvalue).
It is worth noting that the nonlinear supersymmetry with quadratic superpotential (3.8)
was found earlier in Ref. [33] as a by-product in the context of the discussion of the nonrenormalization theorem for supersymmetric theories.

3.2. The k-supersymmetry: exponential superpotential

Let us take the supercharges in the form of polynomials of order k in the oscillator-like
variables A (3.2):
596
Q =2
k/2

k
k1

qkn A

n

,
(3.9)
n=0
where qn are real parameters have to be fixed. If we treat the supercharges (3.9) classically,
then the condition of their conservation by the Hamiltonian of the general form (2.2) results
in

1/ k

q1 k
k/2
Q =2
A +
,
H = Q+ , Q
+ v,
k
i.e., we arrive again at the k-supersymmetric system (2.21) with the arbitrary function
W (x) + q1 /k playing the role of the superpotential. However, as we shall see, quantum
mechanically ansatz (3.9) gives us a nontrivial family of k-supersymmetric systems related
to the so-called quasi-exactly solvable problems [2024]. First one notes that the parameter
q1 can be removed by a simple shift of the superpotential both on the classical and
quantum levels and we can put q1 = 0. Then, as in the case of the supercharges (3.1),
the requirement of conservation of (3.9) results in the Hamiltonian (3.7) as well as in some
k 2 algebraic equations fixing the parameters q3 , q4 , . . . , qk in terms of q q2 , whereas
the condition (3.6) for k 2 is changed now for
h 3 W k2 h W = 0,
k2 =
24q
.
k(k 2 1)
(3.10)
This means that the supercharges (3.9) contain only the one-parameter arbitrariness and the
case of quadratic superpotential (3.8) is included here as a particular case corresponding to
q = 0. For q = 0 the solution of Eq. (3.10) acquires the exponential form
W (x) = w+ ek x + w ek x + w0 ,
(3.11)
where all the parameters w,0 are real, while the parameter k is real or pure imaginary
depending on the sign of q, and for the sake of simplicity we put h = 1. In the limit k 0
this superpotential is reduced to the quadratic form (3.8) via the appropriate rescaling of
the parameters w,0 .
For k = 2 the anticommutator of the supercharges has the form
+
Q , Q = (H c )(H c+ ),
where

c = 2 w02 4w+ w + q + v.
For k 2 the superalgebra for the system with superpotential (3.11) can be represented as

k

cn H kn ,
Q , Q+ = H k +
cn R.
(3.12)
n=1
In principle, the coefficients cn can be found explicitly since the system of linear equations
arises for them. The values of energy of the supercharges singlets are the roots of the
polynomial on the right-hand side of Eq. (3.12) and to find them one has to solve the
algebraic equation of the corresponding order. In next section we analyse in detail the class
597
of k-supersymmetric systems given by the superpotential (3.11) in the context of the partial
algebraization of the spectral problem.
4. The k-supersymmetry and partial algebraization of the spectral problem

In the late eighties a new class of spectral problems was discovered [20,21]. It occupies
the intermediate position between the exactly solvable problems and all others. According
to Ref. [34], the quantum-mechanical system is quasi-exactly solvable or admits partial
algebraization of the spectrum if its potential depends explicitly on the natural parameter n
in such a way that exactly n levels in the spectrum can be found algebraically. The unique
nature of such systems is based on the hidden symmetry of the Hamiltonian. The part of
the spectrum that can be found algebraically is related to finite-dimensional representations
of the corresponding Lee group (algebra). For one-dimensional systems the quasi-exact
solvability is associated with nonunitary finite-dimensional representations of the sl(2, R).
Such representations are characterized by a parameter j referred to as a spin, which can
take integer and half-integer values. The number of the eigenstates of the Hamiltonian that
can be found algebraically is equal to 2j + 1.
Here we argue in favour of existence of the intimate relation between nonlinear
supersymmetry and the partial algebraization scheme [21,34]. For example, if in a given
system with the nonlinear supersymmetry of the order k,
+
Q , Q = (H E1 ) (H Ek ),
there are k singlets in the bosonic or fermionic sectors, i.e., k zero modes of Q+ or
Q , then the eigenvalues of the corresponding states are equal to Ei . Then it is quite
obvious that if such a system is not exactly-solvable, it admits the partial algebraization
of its spectrum. Having in mind these preliminary comments, let us show that the
supersymmetric system with the superpotential (3.11) can be related to some known
families of quasi-exactly solvable problems.
Let us put h = 1 and consider the potentials
2 1
2
1
V (x) = aex + b 2j + cex + b + 1 ,
(4.1)
2
2

a2
1
V (x) =
(4.2)
sin x,
cos2 x + a 2j +
2
2

a2
1
2
V (x) =
(4.3)
sinh x a 2j +
cosh x,
2
2
a2
a
cosh4 x (a + 4j + 2) cosh2 x,
(4.4)
2
2
admitting the partial algebraization of the spectrum [21,34]. The potential (3.5) with the
superpotential of the from (3.11) coincides with the potential (4.1) when
V (x) =
k = 1,
w+ = a,
w = c,
w0 = 1 + 2(b j ),
3
v = + b + b2 2ac + j 2bj + 3j 2 ,
4
k = 2j + 1,
598
or with the potential (4.2) when

a
w = w+ = ,
v = 0,
k = 4j + 1,
(4.5)
w0 = 0,
2
a
w0 = 0,
k = 1,
(4.6)
w = ,
v = 0,
k = 4j + 1,
2

a
k = 2j + 1.
k = 2,
w = ,
v = a j + 12 ,
w0 = 0,
4
In the cases (4.5) and (4.6), there is a complete correspondence with the quasi-exactly
solvable potentials (4.2) and (4.3) for odd k only. Having in mind this observation, we first
analyse in detail the correspondence between the nonlinear supersymmetry and the partial
algebraization scheme for the case of quasi-exactly solvable potential (4.3). Writing down
this potential as
k = i,
a
a2
sinh2 x k cosh x, k N,
(4.7)
2
2
we see that it has exactly the form of the potential (3.5) of the nonlinear supersymmetry
system with the corresponding superpotential
V (x) =
W (x) = a sinh x.
(4.8)
Since the potential of the form (4.7) corresponds to the potential (4.3) for odd k only, let
us investigate the k-supersymmetric systems with even k and start from the case k = 2.
Formally this corresponds to the spin j = 1/4 in the partial algebraization scheme [21]
for the potential (4.3). In this case q = 1/4 and the supercharges read as

1 2 1
Q =
A
,
2
4
where A are defined in (3.2). Zero modes of the supercharge Q+ belong to the bosonic
sector and have the form
(x) = ex/2ea cosh x ,
where for definiteness we have supposed that a > 0. The anticommutator of the
supercharges is

+
a 1
a 1
Q ,Q = H +
H+ +
,
2 8
2 8
and we find that the linear combinations () = + are the eigenfunctions of the
bosonic Hamiltonian,
HB = E,
corresponding to the eigenvalues
a 1
E = .
2 8
(4.9)
599
The fermionic Hamiltonian HF is characterized by the potential

a2
a
sinh2 x + k cosh x,
2
2
which is the superpartner of the potential (4.7). Since this differs from (4.7) in the sign
before the second term, the corresponding solutions in the fermionic sector in the case of
a < 0 can be obtained from the bosonic solutions with a > 0 by a simple change a a.
Therefore, we see that there are two bound states in the bosonic (fermionic) sector for
a > 0 (a < 0), that means that the potential (4.7) is quasi-exactly solvable for k = 2 as
well. As we shall see, the same conclusion is also true for any even k.
Let us consider the case k = 3 corresponding to the spin j = 1/2. In this case the
supercharges Q acquire the form
2

Q = 23/2 A A 1 .
VF (x) =
The zero modes of the operator Q+ are

1 = ea cosh x ,
2,3 = ex ea cosh x .
Looking for the solutions to Eq. (4.9) in the form of the linear combination of the
zero modes, = c1 1 + c2 2 + c3 3 , we obtain the following three (not normalized)
eigenfunctions
= (1 + c cosh x)ea cosh x ,
0 = sinh x ea cosh x ,
where c = (4a)1 (1 16a 2 + 1 ). The energies of these states are

1
1
E = 16a 2 + 1 1 ,
E0 = ,
2
4
and the anticommutator of the supercharges is
+
Q , Q = (H E0 )(H E )(H E+ ).
(4.10)
It is necessary to stress that though in the case k = 3 the direct application of the
partial algebraization scheme to the potential allows ones to find only two eigenstates
and corresponding eigenvalues ( , E ), the concept of nonlinear supersymmetry gives
the information on one more exact eigenstate (0 ) of the Hamiltonian. The same is also
valid for any odd k: the algebraization scheme gives for the potential (4.7) the information
on k+1
2 eigenvalues corresponding to the even (in x) eigenstates, whereas the nonlinear
supersymmetry allows ones to find in addition k1
2 eigenvalues corresponding to the odd
eigenfunctions.
The eigenfunctions and eigenvalues in the fermionic sector for a < 0 can be obtained via
the formal change a a, and we conclude that in the case k = 3 there are three bound
states in bosonic (fermionic) sector for a > 0 (a < 0).
Now let us generalize the k-supersymmetry of the system given by the superpotential (4.8) for the case of arbitrary k. Using the explicit form of the supercharges for k = 2, 3,
one can suppose that the supercharges Q
k obey the recurrent relation
600

1 2
k+1 2
A
Q
=
(4.11)
k.
2
2
Let us prove the conservation of the supercharges (4.11) by the method of mathematical
induction using this conjecture. As the first step we suppose that for some given k the
supercharge is conserved. Thus, calculating [Q+
k , Hk ], we obtain the relation
Q
k+2

1
+
+
W , Q+
(4.12)
k + kW A Qk + (k 1)kW Qk2 = 0.
4
The Hamiltonian Hk+2 is related to the Hk as Hk+2 = Hk W + 2W N.
Then we have
+

1
+
+
Qk+2 , Hk+2 = W , Q+
k+2 + (k + 2)W A Qk+2 + (k + 1)(k + 2)W Qk .
4
Using the relations (4.11) and (4.12), after rather cumbersome algebraic manipulations
one can reveal that the expression vanishes. Since the corresponding supercharges are
conserved for k = 2, 3, we conclude that the supercharges Q
k are conserved for any k
as well. This also proves the conjecture (4.11).
The relation (4.11) gives us the following form of the supercharges for arbitrary k:

k1
k3
k3
k1
k/2
Q
=
2
+
+
A
A
A
A
.
k
2
2
2
2

Using this representation, we find all the zero modes of the supercharge Q+
k :
k3 k1
k1 k3
,
,...,
,
.
2
2
2
2
Since these modes belong to the bosonic sector, we can build k eigenfunctions of the
bosonic part of the Hamiltonian as a linear combination of them:
m
nx
k = 2m + 1,
n=m cn e ,
a cosh x
(x) = e
m1
1
(n+ 2 )x
, k = 2m.
n=m cn e
s (x) = esxa cosh x ,
s=
If we put this into the corresponding stationary Schrdinger equation, we arrive at the
recurrent system of algebraic equations on energy E and the coefficients cn

1 2
n + 2E cn + cn+1 (m + n + 1) + cn1 (m n + 1) = 0, for odd k,
a
where cn = 0 for |n| > m, and

1 2
1
n+
+ 2E cn + cn+1 (m + n + 1) + cn1 (m n) = 0, for even k,
a
2
where cn = 0 for n < m or n m. Thus, we have demonstrated how the concept
of nonlinear supersymmetry allows us to reduce the problem of finding the part of the
spectrum for the potential (4.7) to the pure algebraic problem.
Let us discuss shortly the potential (4.4), to which the superpotential W (x) = a2 sinh 2x
corresponds within the framework of the nonlinear supersymmetry. From the form of the
superpotential one can immediately conclude that the potential (4.4) can be reduced to the
form (4.7) just by rescaling the argument and the parameter a. Indeed, with the help of the
relation
601
a2
a
cosh4 x (a + 4j + 2) cosh2 x
2
2
2
2j + 1
2j + 1
a
sinh2 2x
a cosh 2x
a,
=
8
2
2
after rescaling x 2x, a a/4, we arrive at the form (4.3). But in the case of the
potential (4.4), there is a difference in comparison with (4.3). As we have mentioned
above, according to Ref. [21,34], only the even eigenstates of Hamiltonian with the
potential (4.3) can be found following the sl(2, R) partial algebraization scheme, whereas
for the potential (4.4) all the lowest 2j + 1 states can be found within the framework
of the very scheme. But we see that though the potentials (4.3) and (4.4) have different
representations in the algebraization scheme, in reality they correspond to the same
physical system. Moreover, we see that all the potentials (4.1)(4.4) are particular cases
of the potential (3.5) with the exponential superpotential (3.11). For the potential (3.5)
the first k states can be found algebraically with the help of the associated nonlinear
supersymmetry. In other words, the nonlinear supersymmetry provides us with a universal
point of view on the quasi-exactly solvable potentials (4.1)(4.4).
Now let us discuss the superpotential (3.11) of the general form, which, as we have seen,
allows ones to unify all the cases (4.1)(4.4). The potential (3.5) with this superpotential
has the form

k
k
1 2 2x
x
+ w0 w+ e + w0 + w ex
V (x) = w+ e
2
2
2
1 2 2x
1
+ w e
+ v + w+ w + w0 .
2
2
The superpartner of this potential can be obtained by the formal substitution k k.
Consider in detail the simplest case k = 2. The corresponding supercharge

1 2 2 +
+
Q =
A
2
4
has the zero modes

1
x
x
w+ e w e
= exp (/2 w0 )x
,
(4.13)
and for the equation HB = E one finds the two eigenfunctions

1
() = + c + ,
w0 w02 4w+ w ,
c =
2w
with the energy

1
2
E = w02 4w+ w + v
.
2
8
In the fermionic sector, the zero modes of the supercharge Q are

1
w+ ex w ex
= exp (/2 + w0 )x +
(4.14)
602
and all the corresponding formulas can by reproduced via the changes w w , w0
w0 . As it follows from (4.13) and (4.14), in the case w+ w < 0 there are two bound
states in bosonic (fermionic) sector when w+ > 0 (w+ < 0), whereas for w+ w > 0 there
are no such bound states at all. The corresponding k = 2 superalgebra is written as

+
1 2
Q , Q = H q + q w02 4w+ w .
2
Using the obtained results for the potential (4.7) and realizing the substitution x x,
we arrive at the supercharges obeying the recurrent relation

1 2
k+1 2 2
Q
Q
=
A
k.
k+2
2
2
From the consideration of the nonlinear supersymmetry with the superpotential (3.11), it
follows that for any k the supercharges depend on the parameter k and do not depend
explicitly on other parameters of the superpotential. This means that the supercharges are
conserved and obey the same recurrent relation if we rescale the parameter q to yield the
relation k = . Then in the same way as for the potential (4.7), one can look for the wave
functions of the k singlets in the form
m
nx ,
x
k = 2m + 1,
n=m cn e
W (y) dy
(x) = e
m1
1
(n+
)x
2
, k = 2m.
n=m cn e
The corresponding stationary Schrdinger equation gives us the recurrent system of
algebraic equations on the energy and on the coefficients cn :

1
cn E v + n(n 2w0 ) + (m n + 1)w+ cn1
2
(m + n + 1)w c1+n = 0,
for odd k, where cn = 0 for |n| > m, and

1
1 1
+n
+ n 2w0
+ (m n)w+ cn1
cn E v +
2 2
2
(m + n + 1)w cn+1 = 0,
for even k, where cn = 0 for n < m or n m. Excluding the coefficients cn from the
recurrent systems, one can obtain the algebraic equation for E of the order k. Normalizing
the corresponding polynomial to the form E k + and substituting E H , one can get
the exact form of the polynomial specifying the k-supersymmetry via the anticommutator
of the supercharges Q . Similarly to the case k = 2, one can find that if w+ w < 0, then
there are k bound states in bosonic (fermionic) sector when w+ > 0 (w+ < 0). If w+ w >
0, such bound singlets do not exist at all.
To conclude this section, we note that the choice of the parameters in the superpotential (3.11)
k = ,
w+ = 0,
w = B,
w0 = A
(k 1),
2
603

1
v = (k 1) (k 1) 4A
8
leads to the exactly solvable Morse potential [4]
V (x) = A2 + B 2 e2x B(2A + )ex ,
(4.15)
where A, B are real positive parameters. As it is well known, from the Morse potential a
vast number of famous exactly solvable potentials can be reproduced by the point canonical
transformations and suitable limiting procedures [4]. This illustrates an intimate though
indirect relation of the nonlinear supersymmetry not only to the quasi-exactly solvable
systems but to the exactly solvable models as well.
5. Anomaly-free k = 2 supersymmetry
In Subsection 2.3, we have analysed the classical supersymmetry from the viewpoint
of the polynomial in momentum structure of the supercharges. Since for the second order
supercharges the solution was found by us in the general form, let us analyse this case at
the quantum level too. On the other hand, there is no sense to analyse in such a context
the orders higher then 2 since we do not know the general solution even at classical
level. Different aspects of supersymmetry with the general second order supercharges
were considered in Ref. [30]. Here we discuss the general quantum case of the k = 2
supersymmetry in the context of curing the quantum anomaly problem and from the
viewpoint of all the possible types of spectra. Then, as an example of application of the
k = 2 supersymmetry, we construct a new exactly solvable nontrivial quantum model
associated with the infinitely deep well.
5.1. Quantum k = 2 supersymmetry vs. classical supersymmetry
Let us consider the second order supercharge of the general form

1 2 d2
d
+
Q =
+ b(x) + ,
+ 2h f (x)
h
2
dx 2
dx
and the most general quantum Hamiltonian
(5.1)
h 2 d 2
+ V (x) + L(x) + .
2 dx 2
The condition [Q+ , H ] = 0 partially determines the unknown functions in Q and H :
H =
L(x) = 2f (x),

h f 2
1
f
V (x) =
2
2f
(5.2)

hf
h
c
+ v,
(5.3)
2
2f
2f 2
2
c
h 2 f
h f
+
b(x) = 2 + f 2 + h f
(5.4)
,
2f
2f
f
where c and v are real constants. The function (5.3) plays the role of the potential for
the bosonic sector. Taking into account Eqs. (5.2)(5.4) and identifying f (x) = W (x), we
604
Fig. 1. The four types of the spectra for the k = 2 supersymmetry.
obtain the following most general form for the Hamiltonian and the supercharge of the
quantum k = 2 supersymmetry:

2
c
1
2 d
2

+ W 2 + 2v + 2h W 3 + 5(W ) ,
H = h
(5.5)
2
dx 2
W

2

1
c
d
+ W + 2 5(W ) + ,
Q+ =
(5.6)
h
2
dx
W
5=

h 2
2W W W 2 .
2
4W
(5.7)
Looking at the quantum Hamiltonian (5.5) and supercharge (5.6), we see that their
form is different from the direct quantum analogue constructed proceeding from the
classical quantities (2.29) and (2.28) via the quantization prescription N = 12 [ + , ]:
the presence of quadratic in h 2 term (5.7) in both operators H and Q+ is crucial for
preserving the supersymmetry at the quantum level. Therefore, one can say that the
quantum correction (5.7) cures the problem of the quantum anomaly since without it the
supercharge would not be the integral of motion. It is interesting to note that the quantum
term of the form (5.7) appears also in the method of constructing new solvable potentials
from the old ones via the operator transformations [4].
The supercharges Q+ and Q = (Q+ ) satisfy the relation
+
Q , Q = (H v)2 + c
(5.8)
being the exact quantum analogue of the first classical relation from Eq. (2.30). This
superalgebra defines the form of the spectrum of bounded states. For the sake of
definiteness, let us put the constant v = 0. Then in general the spectra of the k = 2
supersymmetric system can be of the four types presented on Fig. 1. For c > 0, there are
no singlets in the system and we have completely broken k = 2 SUSY, see Fig. 1a. This
case corresponds, in particular, to the k = 2 SUSY given by the quadratic superpotential
discussed in Section 3.1.
In the case c = 0, there are three possibilities: (i) the completely broken phase (there are
no singlet states), see Fig. 1a; (ii) there is one singlet state in either bosonic or fermionic
sector, see Fig. 1b; (iii) there are two singlet states with equal energy, one in bosonic and
605
another in fermionic sector, see Fig. 1c. The types of the spectra b and c are represented by
the new exactly solvable model in the next subsection.
In the case c < 0, all the three mentioned types of the spectra, a, b and c, can
be realized and, in addition, another situation with two singlet states in one of the
two (bosonic or fermionic) sectors can exist, see Fig. 1d. Examples of the type d
were represented above exhaustively by the supersymmetric systems associated with the
exponential superpotential (3.11).
5.2. The k = 2 supersymmetry in action: a new exactly solvable model
Via the appropriate choice of the superpotential, one can construct the k = 2
supersymmetric systems associated with the exactly solvable potentials. E.g., the simplest
choice of the superpotential W (x) = x leads to the famous (two-particle) Calogero model.
The Morse potential (4.15) can also be reproduced and there are two possibilities to realize
it here:
1
W (x) = A Bex ,
2
c=
1 2
(2A )2 ,
16
or
1
3
c = 2 (2A + 3)2 .
W (x) = Bex A ,
2
16
It is well-known that the usual linear supersymmetry allows ones to obtain new exactly
solvable potentials from the given ones. Here we show that the k = 2 supersymmetry
can be exploited in the same way. For example, one can find that in the case c 0 the
potential (5.3) has exactly the form of the potential of the linear supersymmetry with the
superpotential
f (x) 2 c
.
W(x) = f (x)
2f (x)
Here and in what follows we put h = 1. If the superpotential W(x) is given, we can
consider this relation as differential equation. Solving this equation, we obtain the oneparametric solution for k = 2 superpotential f (x) = W (x). Though the initial potential
is the same for linear (k = 1) and k = 2 supersymmetric systems, the corresponding
superpartners are different: V2k=1 = V + W , while V2k=2 = V + 2f . Therefore, in the case
k = 2 one can construct the one-parametric family of isospectral potentials as superpartner
to V (x).
We will illustrate this by the example of the infinite square potential well. Without loss
of generality, we can assume that the width of the potential is equal to 1. This problem is
equivalent to the equation

(x) + 2E 2 (x) = 0
with the boundary conditions
(0) = (1) = 0,
606
and we can think that x runs over the segment [0, 1] only. The eigenfunctions have the
form
n (x) = 2 sin(n + 1)x

(5.9)
with the energy
2
n(n + 2),
(5.10)
2
where the energy of the ground state has been chosen to be equal to 0.
Let us consider the infinite square potential well as a bosonic potential of the system
with the k = 2 SUSY. In order to obtain the superpartner of the potential, we have to find
the superpotential f (x). For simplicity we put c = 0. Then to find the superpotential, we
have to solve the equation
En =
f (x)
(x)
f (x)
= 0 .
2f (x)
0 (x)
The general solution has the form

f (x) =
2 sin2 x
.
c0 2x + sin 2x
(5.11)
In order the superpotential to be well defined function on (0, 1), we have to assume that
the real constant c0 can take any value in R except the interval (0, 2). Let us note that
the superpartner in the framework of the usual (k = 1) supersymmetry is proportional to
cosec2 x and has a pure trigonometric nature, while this is not the case for the k = 2
SUSY superpartner.
The superpartner of the potential is defined as Vf (x) = V (x) + 2f (x), and in this case
it acquires the form
Vf (x) =
sin x(sin x (x 12 c0 ) cos x)

2
+ 16 2
.
2
(c0 2x + sin 2x)2
(5.12)
Acting by the supercharge Q+ of the form (5.1) with the superpotential (5.11) on the wave
functions (5.9) of the square potential well, we obtain the eigenfunctions of the potential
(5.12),

4(1 + n) sin x sin nx
(f )
n (x) = N
c0 2x + sin 2x

2n(c0 2x)
+ n2 +
sin(1 + n)x ,
c0 2x + sin 2x
(f )
given here up to a normalization constant N . One can verify that 0 (x) is identically
equal to zero.
(f )
It is worth noting that the function n (x) has n nodes when c0 = 0, 2 and n 1 nodes
(f )
when c0 = 0 or c0 = 2 . This means that in the latter case 1 (x) is the ground state of
the potential (5.12), but this is not valid for the former case. This situation is illustrated on
(f )
Figs. 2, 3 by the example of the function 1 (x). The arrows indicate direction of the node
607
(f )
Fig. 2. The plots of the function 1 (x) for the cases c0 < 0 and c0 = 0.
(f )
Fig. 3. The plots of the function 1 (x) for the cases c0 > 2 and c0 = 2 .
motion when c0 goes to 0 (Fig. 2) or to 2 (Fig. 3). For the case c0 = 0, 2 , the ground
(f )
state can be found by calculating the limit limn0 n (x)/n. Having done this, one can
verify that the function obtained in such a way,
0 (x) =
2c0 (c0 2)
sin x
,
c0 2x + sin 2x
0 = 1,
is indeed the ground state of the potential. The function vanishes when c0 = 0 or c0 = 2
that corresponds to the statement above. Therefore, by a choice of the parameter c0 we can
obtain the unbroken k = 2 SUSY of the two types, namely, with one singlet for c0 = 0 or
c0 = 2 , and with two singlets with equal energies for other admissible values of c0 .
Thus, we have illustrated the special case of the k = 2 supersymmetric system by the
example of the infinite square potential well. We have also shown that in the case c = 0,
the two types of the spectra represented on Fig. 1b and c can be realized. The spectrum
of the first type has exactly the form of that of a system with the usual supersymmetry
in the exact phase. The spectrum of the second type has rather unusual properties. It has
the form characteristic to a system with k = 1 broken supersymmetry (with coinciding
two lowest energy levels), but here the two lowest states with equal energies are the true
supersymmetric singlets and, therefore, the k = 2 supersymmetry is unbroken.
6. Discussion and outlook

Let us summarize briefly the obtained results and then discuss some open problems that
deserve further attention.
608
Classical supersymmetry is characterized by the Poisson algebra being nonlinear in

the Hamiltonian, and includes the usual linear supersymmetry as a particular case.
Any supersymmetric system is symplectomorphic to the supersymmetric system of
the canonical form with the holomorphic supercharges.
The canonical supersymmetric systems are separated into the three classes defined by
the behaviour of the superpotential. In the first class the parameter characterizing
the order of the superalgebra is subject to a classical quantization: = k Z+ ; in the
second class it can take any non-negative value: R+ ; in the systems of the third
class the fermion degrees of freedom can be decoupled completely by a canonical
transformation and, hence, their supersymmetry has a rather fictive nature.
We have investigated the nonlinear supersymmetry with supercharges being polynomial of order k in the momentum, and for k = 2 have found the most general oneparametric solution of the Calogero-like form (2.29). Depending on the value of the
parameter c, classically the Calogero-like k = 2 supersymmetric system can be reduced to the k = 0, k = 1 or k = 2 supersymmetric system of a canonical form.
The quantization of a system with nonlinear supersymmetry is a nontrivial problem due
to the quantum anomaly taking place in general case. We have shown that
The anomaly-free quantization of the classical k-supersymmetric system with the
holomorphic supercharges is possible for the superpotential of the quadratic (3.8) and
exponential forms (3.11) only;
The k-supersymmetric systems with the exponential superpotential are closely related
to the well-known families of the quasi-exactly solvable systems [21];
The k = 2 supersymmetric Calogero-like systems can be quantized in the anomalyfree way. The problem of the quantum anomaly is cured here by the specific
superpotential-dependent term of order h 2 . Such a quantum term appeared earlier
in the operator transformations method of constructing new solvable potentials from
the known ones [4];
The general k = 2 supersymmetry associated with the Calogero-like systems can be
used for producing new exactly solvable potentials.
The supersymmetric systems (2.21) given by the superpotentials of the third type are
canonically equivalent to the system (2.20) with the completely decoupled fermion degrees
of freedom. On the other hand, the k = 1 quantum analogues of (2.21) with superpotential
of the third type are the systems with spontaneously
broken linear supersymmetry. In this
= Q / 2H are well defined and their anticommutator

case the nilpotent operators Q
look like k = 0 quantum supercharges. However, these operators
is equal to 1, i.e., Q
are not decoupled from the even operators x and p. Therefore, the question is whether
the quantum analogue of the above mentioned canonical transformations exists. If so,
the quantum fermion degrees of freedom could be completely decoupled and the initial
k = 1 supersymmetric system would be reduced to the k = 0 supersymmetric system in
correspondence with the classical picture. On the other hand, if such a transformation does
not exist (at least for some superpotentials W (x)), we face a sort of quantum transmutation.
The classical equivalence of the Calogero-like k = 2 supersymmetric systems to the
609
k = 0 (for c > 0) and to the k = 1 (for c < 0) supersymmetric systems of the canonical
form (2.21) is also based on the existence of the corresponding canonical transformations.
If for the quantum Calogero-like k = 2 supersymmetric systems (5.5)(5.7) the quantum
analogues of the above mentioned canonical transformations do not exist, we again have a
quantum transmutation.
The supersymmetric system (2.21) represents the whole class of the symplectomorphic
systems (2.15). However, the quantization apparently breaks the equivalence of the
systems (2.15) with different functions M(W 2 ). Therefore the quantization of the systems
(2.15) may lead to different nontrivial quantum systems with k-supersymmetry. Most
likely, the anomaly-free quantization could be possible only for some special cases of
the function M(W 2 ) and the superpotential W (x). In this context it is necessary to stress
that the quantum anomaly is specific not only for the nonlinear supersymmetry. The
quantization of the general system (2.15) with k = 1 and M(W 2 ) 0 gives rise to the
quantum anomaly. From this view-point one can say that the system of the form (2.21)
underlies the quantum linear supersymmetry since the quantization of such a system with
arbitrary superpotential does not reveal any anomaly. However, it would be interesting
to find at least one example of the quantum-mechanical system of the nonstandard form
(2.15) with the linear supersymmetry. Since for the nontrivial cases the quantum analogue
of the transformation (2.13) is nonlocal, the corresponding system has to operate with
nonlocal objects. It is possible that the supersymmetry of the pure parabosonic systems
[10] can be arrived at in this way. We hope that further investigations will shed light on the
relation between the classical and quantum supersymmetries in the context on the described
hypothetical quantum transmutations and the quantum anomaly problem.
Though we have considered the concept of the nonlinear supersymmetry for onedimensional systems, it can be extended to the higher dimensional systems as well. For
example, the two-dimensional system describing the motion of a charged particle in
external magnetic field admits the nonlinear supersymmetries. The Pauli Hamiltonian of
such a system is given by (h = m = e = 1) H = 12 (P12 + P22 + gij i Aj N), where Pi =
ii + Ai , and Ai , i = 1, 2, form the 2D vector gauge potential. For the gyromagnetic ratio
g = 2 this system reveals the usual linear supersymmetry both at the classical and quantum
levels [4]. It turns out that for g = 2k the system possesses the k-supersymmetry. To show
this let us introduce the complex variables z = P1 iP2 . In terms of these variables
the Hamiltonian can be rewritten as H = 12 (z+ z + ik{z+ , z }N). But this is exactly the
form of the Hamiltonian (2.21) with the variables z, z changed for z + , z . Therefore, the
Hamiltonian commutes with the supercharges Q = (z )k and, as a consequence, the
system possesses the classical k-supersymmetry. The quantization of this system reveals
the properties similar to those for the k-supersymmetric system with the holomorphic
supercharges: the requirement of conservation of the nonlinear supersymmetry leads to the
restrictions on the form of the magnetic field B = ij i Aj . One can show that as in the onedimensional case, the form of the magnetic field turns out to be restricted by the quadratic
case and by the sum of exponential functions in the variables xi [35]. We suppose that such
systems are also closely related to the two-dimensional quasi-exactly solvable problems.
610
To conclude, we note that our investigation of the nonlinear supersymmetry was

motivated by the existence of nonlinear supersymmetry in pure parabosonic systems [10],
where the reflection (parity) operator R plays the role of the Z2 -grading operator. Such
a special role of the reflection operator was the main idea in the construction of the
minimally bosonized k = 1 supersymmetric quantum mechanics [36,37], where the role
of the superpotential is played by the arbitrary odd function. Therefore, the construction of
the bosonized nonlinear supersymmetry is an attractive problem which we hope to consider
elsewhere [35].
When this paper was finished, the interesting papers [38,39] devoted to the development
of the 1D nonlinear supersymmetry of Ref. [33] have appeared. We note that the
generalization of the k-supersymmetry with the holomorphic supercharges found in
Ref. [39] and possessing the typical Calogero-like structure of the k = 2 supersymmetry
from Section 5 could also be treated in the context of the quasi-exactly solvable systems
discussed here.
Acknowledgements
The work was supported by the grants 1980619, 1010073 and 3000006 from FONDECYT (Chile) and by DYCIT (USACH).
Appendix A
2 (x) =
Here we show that the arbitrariness in the definition of the superpotential, W
+ const, can not be used to change the type of the classical supersymmetry of
the given system, i.e., we demonstrate the invariance of the classification obtained in
Section 2.2. For the purpose, it is enough to prove that the redefinition of the superpotential
can not provoke a transition between the adjacent classes of supersymmetry.
Let us start by analysing the possibility of the transition from the supersymmetry of
the first type with the superpotential unbounded from below to the supersymmetry of
the second type. For the sake of simplicity we assume that the potential V (x) has finite
number of minima, the lowest one is at the origin and is equal to zero. For such a
potential, the superpotentials of the first and second types obey the relation W 2 (x) =
V (x). Suppose that the bounded and unbounded superpotentials Wu (x) and Wb (x) with
the associated supersymmetries of the first and second types correspond to the potential.
The superpotential Wb (x) is a continuously differentiable function at x = 0 only when
the potential has the local behaviour V (x) = x 4 + O(x 5 ). Indeed, if V (x) = x 2 + O(x 3 ),
the bounded superpotential locally is Wb (x) = |x| + O(x 2 ), and as a consequence, its
derivative is not regular at the origin. The case of the potential admitting the regular
superpotential Wb (x) is presented on the Fig. 4. In this case the bounded and unbounded
superpotentials Wb (x) and Wu (x) coincide for positive values of x and have different sign
for negative x: Wu (x) = sign(x)Wb (x). If we have the nonlinear supersymmetry of the
first type, i.e., L(x) = kWu (x), k N, then rewriting this function in terms of Wb we
W 2 (x)
611
Fig. 4. The plots of a potential V (x) and the corresponding superpotentials Wu (x) and Wb (x).
obtain L(x) = k sign(x)Wb (x). The function L(x) is not good in terms of Wb in the sense
that there exist no canonical transformation that could reduce this system to that with the
supersymmetry of the second type given by Eq. (2.18). The potential with several minima
can be considered in a similar way.
The case of supersymmetry of the first type with a bounded superpotential merits the
of the second type, the
special analysis. After the transition (2.22) to the superpotential W

2
1/2

, w = const. The function

function L(x) can be rewritten as L(x) = k W W /(W w)
2 w)
1/2 has singularities at the points where the initial superpotential vanishes.
(W
Therefore, the type of the supersymmetry can not be changed either.
Now let us consider the possibility of transition from the supersymmetry of the
second type to that of the third type by means of the change (2.22). We suppose
that the superpotential W (x) of a given system has a minimum equal to zero at
the origin of coordinates. The superpotential of the third type defined by W 2 (x) =
2 (x) w has the minimum equal to w at the origin. Rewriting L(x) in terms of
W
/(W
2 w)1/2 . There exists no
W
the new superpotential, we obtain L(x) = W = W
canonical transformation (2.13) that could remove the corresponding nilpotent part of the
/(W
2 w)1/2 is singular at the origin. The case of the
Hamiltonian since the function W
superpotential with several minima can be treated in a similar way. This completes our
proof of the invariance of the obtained classification of the classical supersymmetries.
References
[1]
[2]
[3]
[4]
[5]

L.E. Gendenshtein, I.V. Krive, Sov. Phys. Usp. 28 (1985) 645.
F. Cooper, A. Khare, U. Sukhatme, Phys. Rep. 251 (1995) 267, hep-th/9405029.
P. Claus, M. Derix, R. Kallosh, J. Kumar, P.K. Townsend, A. Van Proeyen, Phys. Rev. Lett. 81
(1998) 4553, hep-th/9804177.
[6] J. Michelson, A. Strominger, JHEP 9909 (1999) 005, hep-th/9908044.
[7] T. Banks, W. Fischler, S.H. Shenker, L. Susskind, Phys. Rev. D 55 (1997) 5112, hepth/9610043.
[8] R. Jackiw, A particle field theorists lectures on supersymmetric, non-Abelian fluid mechanics
and D-branes, physics/0010042.
612
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
R. Jackiw, A.P. Polychronakos, Phys. Rev. D 62 (2000) 085019.

M. Plyushchay, Int. J. Mod. Phys. A 23 (2000) 3679, hep-th/9903130.
S. Klishevich, M. Plyushchay, Mod. Phys. Lett. A 14 (1999) 2739, hep-th/9905149.
F. De Jonghe, A.J. Macfarlane, K. Peeters, J.W. van Holten, Phys. Lett. B 359 (1995) 114,
hep-th/9507046.
M.S. Plyushchay, Phys. Lett. B 485 (2000) 187, hep-th/0005122.
G. Grignani, M. Plyushchay, P. Sodano, Nucl. Phys. B 464 (1996) 189, hep-th/9511072.
Kh.S. Nirov, M.S. Plyushchay, Nucl. Phys. B 512 (1998) 295, hep-th/9803221.
J. de Boer, F. Harmsze, T. Tjin, Phys. Rept. 272 (1996) 139, hep-th/9503161.
E. Witten, Phys. Lett. B 117 (1982) 324.
S. Elitzur, Y. Frishman, E. Rabinovici, A. Schwimmer, Nucl. Phys. B 273 (1986) 93.
S.B. Treiman, E. Witten, R. Jackiw, B. Zumino, Current Algebra and Anomalies, Singapore,
World Scientific, 1985.
A. Turbiner, Commun. Math. Phys. 118 (1988) 467.
M.A. Shifman, Int. J. Mod. Phys. A 4 (1989) 2897.
A. Ushveridze, Quasi-Exactly Solvable Models in Quantum Mechanics, IOP Publishing,
Bristol, 1994.
F. Finkel, A. Gonzalez-Lopez, N. Kamran, P.J. Olver, M.A. Rodriguez, Lie algebras of
differential operators and partial integrability, hep-th/9603139.
C.M. Bender, G.V. Dunne, M. Moshe, Phys. Rev. A 55 (1997) 2625, hep-th/9609193.
A.Y. Morozov, A.M. Perelomov, A.A. Roslyi, M.A. Shifman, A.V. Turbiner, Int. J. Mod. Phys.
A 5 (1990) 803.
A.S. Gorsky, Pisma Zh. Eksp. Teor. Fiz. 54 (1991) 296.
V.V. Bazhanov, S.L. Lukyanov, A.B. Zamolodchikov, J. Statist. Phys. 102 (2001) 567, hepth/9812247.
P.E. Dorey, C. Dunning, R. Tateo, Ordinary differential equations and integrable models, hepth/0010148.
J. Suzuki, Functional relations in Stokes multipliers fun with x 6 + x 2 potential, quantph/0003066.
A.A. Andrianov, M.V. Ioffe, V.P. Spiridonov, Phys. Lett. A 174 (1993) 273, hep-th/9303005.
M. Plyushchay, Lecture Notes in Physics, Vol. 524, Springer-Verlag, Berlin, 1999, p. 270, hepth/9808130.
K.S. Nirov, M. Plyushchay, Phys. Lett. B 405 (1997) 114, hep-th/9707070.
H. Aoyama, H. Kikuchi, I. Okouchi, M. Sato, S. Wada, Nucl. Phys. B 553 (1999) 644, hepth/9808034.
M.A. Shifman, Quasi-exactly solvable problems, Preprint UMN-TH-1701-98.
S. Klishevich, M. Plyushchay, in progress.
M.S. Plyushchay, Mod. Phys. Lett. A 11 (1996) 397, hep-th/9601141.
J. Gamboa, M. Plyushchay, J. Zanelli, Nucl. Phys. B 543 (1999) 447, hep-th/9808062.
H. Aoyama, M. Sato, T. Tanaka, M. Yamamoto, Phys. Lett. B 498 (2001) 117, quantph/0011009.
H. Aoyama, M. Sato, T. Tanaka, Phys. Lett. B 503 (2001) 423, quant-ph/0012065.

Noncommutative linear sigma models

Bruce A. Campbell, Kirk Kaminsky
Department of Physics, University of Alberta, Edmonton, Alberta, T6G 2J1, Canada
Received 9 February 2001; accepted 12 April 2001
Abstract
We examine noncommutative linear sigma models with U (N) global symmetry groups at the
one-loop quantum level, and contrast the results with our previous study of the noncommutative
O(N) linear sigma models where we have shown that NambuGoldstone symmetry realization
is inconsistent with continuum renormalization. Specifically we find no violation of Goldstones
theorem at one-loop for the U (N) models with the quartic term ordering consistent with possible
noncommutative gauging of the model. The difference is due to terms involving noncommutative
commutator interactions, which vanish in the commutative limit. We also examine the U (2), and
O(4) linear sigma models with matter in the adjoint representation, and find that the former is
consistent with Goldstones theorem at one-loop if we include only trace invariants consistent with
possible noncommutative gauging of the model, while the latter exhibits violations of Goldstones
theorem of the kind seen in the fundamental of O(N) for N > 2. 2001 Elsevier Science B.V. All
rights reserved.
PACS: 12.38.Bx; 12.60.-i; 12.90.+b
1. Introduction
Recently field theories on noncommutative spacetime backgrounds have been the subject
of intense scrutiny [1]. Part of this motivation stems from the fact that noncommutative
U (N) gauge theories arise on D-branes in the presence of a constant NSNS B-field
background, in the zero-slope, field theoretic limit of string theory [2,3]. A second
motivation, independent of string theory, is the question of whether the world we live
in is based on a noncommutative spacetime. In order to construct realistic models of
particle physics on noncommutative spacetimes, one needs to be sure that noncommutative
theories preserve the features that underlie the standard model, including perturbative
renormalizability in the presence of spontaneous symmetry breaking [4,5].
The general scheme for defining field theories with the noncommutative spacetime
structure defined by [x , x ] = i , real, constant and antisymmetric, is to
E-mail address: kaminsky@phys.ualberta.ca (K. Kaminsky).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 8 9 - 4
614
B.A. Campbell, K. Kaminsky / Nuclear Physics B 606 [PM] (2001) 613635
invoke WeylMoyal correspondence. This has the effect of replacing the underlying
noncommutative spacetime with a commutative spacetime at the expense of replacing
the ordinary pointwise product of spacetime dependent functions with an infinitely
nonlocal star product. The induced momentum space Feynman rules for interaction vertices
associated with a given field theory then involve momentum-dependent phases, which
generically split a graph (at least at one-loop) into planar and nonplanar parts. The former
are identical to the usual commutative graphs (up to a total phase depending only on
the external momenta, and a combinatorial reweighting), and in particular possess the
usual divergence structure associated with a commutative quantum field theory. The latter,
nonplanar components are explicitly finite (at least at one-loop) because of oscillatory
damping due to the phases, and replace an ultraviolet divergence with an infrared
divergence in the external momenta [6,7].
Superficially, as a consequence of the finiteness of nonplanar graphs, and of the similar
divergence structure of the planar graphs, one might conclude that the renormalization of
noncommutative field theories proceeds as in the commutative theory, because the counterterm structure is formally the same. However, as is well-known, the renormalization of
spontaneously broken theories, with either underlying global or gauge symmetries, is more
subtle because the number of counterterm vertices exceeds the number of renormalization
parameters. As a result, the renormalizability of (commutative) spontaneously broken theories hinges in general on intricate graphwise cancellations [4,5] order by order in perturbation theory. Thus it is of obvious interest to examine whether or not these cancellations
persist in noncommutative field theories.
In a previous paper [8] we studied the spontaneous symmetry breaking of a global
O(N) symmetry in the noncommutative deformation of the linear sigma model with
scalars in the fundamental representation. We found that one-point tadpoles of the sigma
at one-loop were insensitive to the noncommutativity because no external momentum
flows into the trilinear tadpole vertex. Thus the one-point sigma counterterm is identical
to the one in the commutative limit, which in turn fixes the pion mass counterterm to
be the same as its commutative limit. On the other hand, the planar components of the
1PI graphs contributing to the one-loop pion (inverse) propagator renormalization are
re-weighted with respect to the corresponding commutative graphs. As a consequence,
there is an unavoidable UV cutoff dependence (for nonzero external momentum) after
renormalization, signalling the nonexistence of a continuum limit, and noncommuting UV
(UV ) and IR (p 0) limits. Specifically we found that the sum of the 1PI graphs
and the counterterm contributing to the pion mass renormalization yielded quadratic and
logarithmic UV cutoff dependences as:

2
ij
1
=
N(1 f ) + f 1
16 2
1 + 2 (p p)
1-loop

2 f m2 log 1 + 2 (p p) + finite p2 ,
(1)
2 q /4,
respectively. Here N is the dimension of the fundamental of O(N), p q p
and f takes into account the two possible quartic orderings for and terms:
615

k
f k l l + (1 f ) k l k l + f k k
4
4
2
k
k

+ (1 f ) L.
(2)
2
For nonzero and p, the only circumstance under which we can take the continuum
limit is when f = 2 and N = 2, where both logarithmic and quadratic dependences on
vanish. This corresponds to the Abelian O(2) model, and if written in terms of a complex
scalar corresponds precisely to the ordering 1 . Otherwise, the conditions
f = N/(N 1) and f = 2 required for the cancellations of quadratic and logarithmic
dependences on respectively, cannot be simultaneously satisfied, Goldstones theorem
fails at the one-loop level, and the continuum limit of the model fails to exist. Thus for
general N > 2 and for all possible orderings consistent with the global O(N) symmetry,
the noncommutative O(N) linear sigma model does not exist in the continuum limit.
Prima facie, this incompatibility of continuum renormalizability with spontaneous symmetry breaking for O(N) linear sigma models appears to present severe difficulties for attempts to make realistic models of particle physics on noncommutative spacetimes. First,
it is clear that models with spontaneously broken gauge symmetries must have consistent
spontaneously broken global limits (as the gauge couplings vanish); the absence of such
a global limit with spontaneous symmetry breaking would preclude its subsequent gauging (at least perturbatively). Second, the standard model of the fundamental interactions
(and unified theories which encompass it) depends, for electroweak symmetry breaking,
on a complex Higgs doublet. As is well known, resolved into real components, the purely
scalar sector of the standard model is O(4) invariant (and not just SU(2) U (1) invariant), with the real components in the fundamental representation; our previous results then
appear to preclude noncommutative deformation of the standard model. We will argue below that this is not necessarily the case. In particular, noncommutative theories with N
complex scalars, i (i = 1, . . . , N , N > 1), and with U (N) invariant self-interactions, are
not invariant under an O(2N) symmetry acting on their real components, due to purely
noncommutative commutator interactions arising from the noncommutativity of the spacetime. Thus we will first undertake an analysis for the case of a U (N) symmetry group with
the scalars in the fundamental representation, choosing the quartic invariant .
The spontaneous breaking of this group to U (N 1) leaves N 1 complex pions, and one
real pion. We find that the one-loop 1PI graphs contributing to the mass renormalization of
the complex pions, like the one-point tadpoles, do not see the noncommutativity at this order, and so Goldstones theorem holds. The 1PI one-loop graphs contributing to the mass
renormalization of the real pion (which arises through the breaking of the U (1) O(2)
subgroup of the U (N)), are split into divergent planar, and finite nonplanar pieces in such
a way that Goldstones theorem holds at one-loop. The essential difference between the
U (N) models and the corresponding O(2N) models (N > 1) is the presence of the purely
noncommutative commutator interactions in the former.
1 The interested reader may verify that the N = 2, f = 2 case of our general analysis, is precisely the global
limit of the Abelian U (1) O(2) model subsequently considered in [9] as can be easily seen by comparing
Lagrangian densities, and Feynman rules; the conclusions of Ref. [9] agree with our previous analysis [8].
616
We will also begin to explore how our present, and previous [8], results might depend
on the scalar field representation responsible for spontaneously breaking the symmetry.
In particular, we consider both an O(4) and a U (2) model, with scalars in the adjoint
representation, to see if our previous results depended on our scalars being in the respective
fundamentals. For the U (2) model with matter in the adjoint representation, we will
find that Goldstones theorem holds if we include only interactions involving a single
trace operator, which we will in turn demonstrate are the only ones consistent with
noncommutative gauge invariance in the case that we gauge the U (2) symmetry. In this
model Goldstones theorem holds due to a notable cancellation of a purely noncommutative
graph involving couplings to the U (1) component of the field. For the O(4) model with
matter in the adjoint, we find violations of Goldstones theorem at one-loop of the type
found in the O(N) fundamental representation studied in [8]. Finally we discuss the
implications of these results for model building, and comment on the nature of the IR
divergences found by [7] in the context of noncommutative theories with matter in the
adjoint representation.
2. NC U (N ) linear sigma model: fundamental rep

In this section we examine Goldstones theorem in the noncommutative deformation of
the linear sigma model with a global U (N) symmetry group, and contrast the results with
our previously discovered violations of Goldstones theorem in the O(N) linear sigma
model.
The noncommutative U (N) linear sigma model is defined by the Lagrangian density
given by
L = + 2 ,
(3)
where is an N -vector of complex fields i (i = 1, . . . , N ), where the star product is

y
defined as usual by f (x) g(x) = exp (i z )f (y)g(z)|y,zx , and where we have
included the star ordering of the quartic term consistent with noncommutative gauge
invariance of a possible gauging of the model (see below). For 2 > 0, the symmetry
is spontaneously broken to U (N 1). Throughout the remainder of this paper, we
will consider only translationally invariant vacua. 2 By an SU(N) transformation, we
can rotate the resulting VEV into the last field of , and by a U (1) rotation we can
identify this VEV with a constant shift, a, in the real partof this field. Thus we define
i = i for i = 1, . . . , N 1, while N = ( + a + i0 )/ 2; there are N 1 complex
Goldstone bosons, and one real Goldstone mode. The minimization of the potential for this
configuration implies:
V (a) =
2
2 2 4
a a a 2 =
.
2
4
(4)
2 As Gubser and Sondhi have argued [11], more exotic vacua such as stripe phases are possible in
noncommutative theories.
617
Writing (3) in terms of these variables yields:

1
1
1
L = ( )2 + ( 0 )2 + i i 22 2 a 3
2
2
2

a02 2ai i i i j j 4 + 04
4
2

2 2
2
0 + 0 0 i i + 0 i i [, 0 ].
2
(5)
For notational brevity all star products will be suppressed henceforth, unless there is danger
of confusion. Furthermore, we will implicitly use the identity
A1 An = A (1) A (n)
(6)
(where { (1), . . . , (n)} represents any cyclic permutation of {1, . . . , n}), with the
understanding that all Lagrangian density terms sit under a spacetime integral. This identity
means that quadratic terms in the action, and hence propagators, are identical to their
commutative counterparts.
To simplify the discussion relative to that occurring in [8], we will not a priori impose
the vanishing of the tadpole as a renormalization condition. Instead we will include the
one-point tadpole contributions, and their counterterm directly in calculating the mass
renormalization of the pion. In this completely equivalent language, the two counterterms
present cancel each other, up to the wavefunction renormalization, so the sum of the oneparticle irreducible (1PI) graphs and the one-point tadpole insertions must be automatically
finite up to wavefunction renormalization (and for Goldstones theorem to hold at oneloop, must vanish in the p 0 limit) [10]. Furthermore, to exhibit the essentially algebraic
nature of the result, we will expand the non-phase part of the integrands about zero-external
momentum, in the cases where there are two propagators in the loop using the Taylor
expansion
1
k 2 [(p + k)2
m2 ]
k 2 (k 2
2k
1
+
p 2 2
2
m )
k [k m2 ]2
(7)
and then note that the p-dependent terms yield finite loop-momentum integrals (for all p),
and vanish as p 0. We then define the momentum integrals

d 4k
1
,
(2)4 k 2 m2

d 4 k cos(k p)
d 4 k eikp
=
,
I,p m2 =
(2)4 k 2 m2
(2)4 k 2 m2
I m
(8)
where k p = k p .
The vertices for the theory are listed in Appendix A, and the propagators are the usual
ones. Dashes denote complex pions, dots the real pion associated with the sigma, and solid
lines denote the sigma. The 1PI one-loop graphs contributing to the mass renormalization
618
of the N 1 complex pions are
(a),
(b),
(c),
(d).
They are given, respectively, by
d 4 k kl [ ij kl e0 + il j k e0 ]
(a) = 2ii
= 2N ij I (0),
(2)4
k2
d 4 k e0 cos(0)
2ii ij
(b) =
= ij I (0),
2
(2)4
k2

2ii ij
d 4 k e0 cos(0)
= ij I 22 ,
(c) =
4
2
2
2
(2) k 2
i
i
d 4 k e 2 (kp) e 2 (pk)
(d) = (2ia)2i 2 ik kl lj
(2)4 [(p + k)2 22 ]k 2

2
2
4
4 a ij
1
d k
1
=
+ ij A p
22
(2)4 k 2 22 k 2

= 2 ij I 22 I (0) + ij A p ,
(9)
(10)
A (p)
is finite for all p. Evidently these 1PI graphs do not see the noncommutativwhere
ity. Meanwhile the one-point tadpoles insertions, as in [8], also do not see the noncommutativity at one-loop. They are given by
(e),
(f ),
(g).
Their values are given by
i

2ia kk iI 22 = 2(N 1) ij I (0),
(e) = 2ia ij
2
2
2

i
i 2
ij
ij
I
2
(6ia)
I
2 ,
(f ) = 2ia
=
3
2
22
i

i
(g) = 2ia ij
(2ia) I 22 = ij I (0),
2
2
2
where all noncommutative vertices manifestly collapse.
(11)
(12)
619
The sum of these seven graphs is equal to zero (modulo the finite term which
itself vanishes as p 0), independently of a regulator, and of the noncommutativity,
whence Goldstones theorem holds at one-loop; the complex pions undergo no mass
renormalization. Now consider the one-loop mass renormalization of 0 . The 1PI graphs
contributing are given by
(h),
(i),
(j ),
(k), (13)
with values given by
2i
(h) =
2
2i
(i) =
2
kp
d 4 k i[2 cos2 ( 2 ) + 1]
= 2I (0) + I,p (0),
(2)4
k2

d 4 k i[2 cos2 (0) cos(k p)]
= 2I 22 I,p 22 ,
4
2
2
(2)
k 2
4
0
d k ie cos(0)
(j ) = 2i kk
= 2(N 1)I (0),
(2)4
k2
i 2 cos2 ( pk
d 4k
2
2 )
(k) = (2ia)
4
2
(2) [(p + k) 22 ]k 2

2
2
4

4 a
1
d k 1
1
1
+
cos(k
p)
+ B p
22
(2)4 2
k 2 22 k 2

= I 22 I (0) + I,p 22 I,p (0) + B p ,
(14)
B (p)
is finite for all p. Evidently, the nonplanar contributions due to the

where
noncommutativity cancel between these graphs. For completeness, the one-point tadpole
insertions are
(l),
(n),
(l) = (2ia)

i
2ia kk iI (0) = 2(N 1)I (0),
2
2
(m),
(15)
620

i
i
(6ia) I 22 = 3I 22 ,
2
2
2
i
i
(n) = (2ia)
(2ia) I (0) = I (0).
2
2
2
(m) = (2ia)
(16)
The sum of these seven graphs also vanishes (again modulo the finite p dependent term,
which vanishes as p 0); so that Goldstones theorem also holds for the neutral pion of
this model.
Let us reflect on these results. First, had we included the other ordering of the quartic
term i j i j , we would again find violations of Goldstones theorem of the type
found in [8]. Secondly, we contrast these results with those of the general O(N) model
studied in [8], where we showed violations of Goldstones theorem at one-loop for all
orderings consistent with the O(N) global symmetry (except for the trivial Abelian case).
The difference here of course is that we are working with a U (N) group, which we now
show exhibits crucial algebraic differences with the O(N) case in noncommutative scalar
theories.
Matter in the fundamental of O(N) is described by a real N -vector of fields, which we
denote by . As such, the invariant term T merely is the sum of squares of the real
components. Then, the expansion of the quartic invariant can yield cross-terms only of the
form i i j j , or i j i j corresponding to the two possible star product
orderings of such an invariant. Note that no more than two distinct fields can occur.
On the other hand, the fundamental of U (N) is described by a complex N -vector of
fields, . Now however, the quadratic invariant , written in terms of real fields
picks up the commutator of each fields real part with its imaginary part due to the
noncommutativity since
(R iI )(R + iI ) = R 2 + I 2 + i[R, I ].
(17)
While such commutators in the quadratic term vanish when integrated over spacetime,
the quartic invariant now picks up products of such commutators with other fields or
commutators which, for N > 1, constitute new interactions between real components of
two complex s, not present in, and incompatible with, the O(2N) symmetry.
Let us make this argument manifest. Expanding the quartic term in the U (N) theory in
terms of its real components yields
i i j j = (iR iiI )(iR + iiI )(j R ij I )(j R + ij I )
2 2
2 2
2 2
2 2
= iR
j R + iR
j I + iI
j R + iI
j I [iR , iI ][j R , j I ]
2

2
+ i iR + iI
[j R , j I ] + i j2R + j2I [iR , iI ].
(18)
We note for i = j (which can occur for N > 1), the presence of interactions (those
involving three or four distinct real fields) which cannot occur in the O(2N) case by
the general argument above. We emphasize this is a purely noncommutative effect. 3
3 For i = j (or N = 1), the last two terms vanish under the spacetime integral, and the product of commutators
merely induces the orderings of the O(2) model studied in [8] with f = 2.
621
The presence of these extra, purely noncommutative interactions is responsible for the
the differing behaviour of the spontaneously broken phase at the quantum level for these
models.
To conclude, we have found that one cannot in general spontaneously break a
fundamental representation NC O(N) linear sigma model, while one can break a
fundamental representation NC U (N) linear sigma model for the noncommutative gauge
invariant quartic ordering. This latter theme is one that will arise again in a more dramatic
fashion in the adjoint representation model to which we now turn.
3. NC U (2) sigma model: adjoint representation

We now examine the status of Goldstones theorem in the noncommutative deformation
of the linear sigma model with scalars in the adjoint representation of U (2). There are
several reasons for this: first, we wish to compare the results for adjoint representation
scalars with our results from the previous section for fundamental scalars, in a tractable
case. Secondly, adjoint matter naturally arises in noncommutative world-volume theories
on D-branes. Thirdly, grand unified theories embedding the standard model commonly rely
on adjoints for the first stage of symmetry breaking.
We write the scalars in the adjoint of U (2) as

1 4 + 3
2
,
= a T a =
(19)
2
4 3
2
where T a are the canonical generators of U (2): T a = a /2, for a = 1, 2, 3 and T 4 = I2 /2.
The global U (2) symmetry transformation acts as
U U
(20)
and as before does not involve the star product because the symmetry is global. For
simplicity we impose invariance under . The Lagrangian density for the global
model we consider is defined by

L = Tr + 2 Tr( ) 1 Tr( )

2
2 Tr( ) ,
(21)
where we define

j
Tr 4 ji k lk il ,

2
j
Tr( ) ji i lk kl ,
(22)
and where we discuss the remaining, omitted trace invariants and star product orderings at
the end of this section.
Let us now consider spontaneous symmetry breaking which occurs for 2 > 0 (we take
i > 0). Then acquires a vacuum expectation value, say 0 , and since it is a Hermitian
622
(but not necessarily traceless) matrix, we analyze it by diagonalization to the form

1 a 0
0 =
,
2 0 b
(23)
whence the potential becomes

1 4
2 4

2 2
a + b2 +
a + b4 +
a + 2a 2b2 + b4 .
4
16
16
This is minimized for
V (a, b) =
a 2 = b2 =
2
1
2
+ 2
(24)
2
.
(25)
The states corresponding to a = b, which are degenerate in energy with the states
corresponding to a = b, and admitted because we are considering U (2) and not simply
SU(2), do not reflect spontaneously broken states, because 0 is then proportional to the
identity and so manifestly commutes with all of the generators. Furthermore, since they
correspond to constant shifts in the U (1) component 4 , they are forbidden by the discrete
symmetry. On the other hand, the states corresponding to a = b do yield spontaneously
broken vacua, since they do not commute with the T 1 and T 2 generators and reflect a
vacuum expectation value for the field 3 .
In notation suggestive of the linear sigma model, we expand around the vacuum b =
a < 0 (without any loss of generality), defining and through

1 4 +
2

=
(26)
0 ,
2 4
2
so that 3 = + a. Expanding the scalar potential in terms of these variables yields

1
1
1 + 2
2
+
V = 22 2 + 1 a 2 42 +
2
2
2
2

1 + 2
+
2
1
1
+ 2 +
+ 42
+
2
2
2

1
+ 4 4 + a + + 4 + 44 + a 3
2
4
1 + 2 2 2 1
4 + 4 4 + ( + 1 )a42
+
2
4

1
4 4 + a 4 4
+
(27)
2
using = 1 /2 + 2 .
The symmetrized vertices for this theory are listed in Appendix A. In the following,
solid lines denote the , dots denote the 4 , and dashes denote the . Excluding the
purely noncommutative interactions for separate consideration, there are four 1PI graphs
contributing to the mass renormalization of the complex pion (Goldstone mode) in this
model:
(a),
(b),
(c),
623
(d)
(28)
kp
d 4k i
d 4 k i cos2 ( 2 )
2i
(a) = 2i(1 + 2 )
2
(2)4 k 2
(2)4
k2
= (21 + 32 )I (0) + 2 I,p (0),
1

(b) = (1 + 2 )I 22 I,p 22 ,
2
1

2
(c) = (1 + 2 )I 1 a + I,p 1 a 2 ,
2

4k i
i
d
2
2 kp
(d) = (2ia)
cos
(2)4 k 2 (p + k)2 22
2

4
1
d k 1
1
2 2
2 kp
= 4 a
cos
+ C (p)p
(2)4 22 k 2 22 k 2
2

= I 22 I (0) + I,p 22 I,p (0) + C (p)p ,
(29)
is finite for all p.

where
The one-point tadpole contributions are given by
(e),
(g),
(f ),
(30)
which are respectively given by

i
iI (0) = 2I (0),
22
2
i
i 2
I
2
=
3I
2 ,
(f ) = (2ia)
(6ia)
22
2

i
i
(g) = (2ia)
2i( + 1 )a I 1 a 2
2
2
2

2
= ( + 1 )I 1 a .
(e) = (2ia)2
(31)
The sum of these seven graphs is given by

1

1

I (0) I,p (0) 2 I 22 I,p 22
I 1 a 2 I,p 1 a 2
=
2
2
+ C (p)p .
(32)
624
In the commutative limit 0, this degenerates to the finite term C (p)p (which itself
vanishes as p 0), so the mass counterterm vanishes and this is a demonstration of
Goldstones theorem for this model. However for nonzero , the I (m2 ) terms are divergent
and require regularization, say by an ultraviolet cutoff . But there is no counterterm
freedom to cancel the dependence, so for nonzero p and nonzero we cannot take the
continuum limit; that is, UV ( ) and IR (p 0) limits do not commute.
However, we have (intentionally) neglected a purely noncommutative graph due to the
last interaction in (27). The purely noncommutative interaction generated by (
)4 yields a graphical contribution given by
pk
kp
d 4 k sin( 2 ) sin( 2 )
(2)4 k 2 [(p + k)2 1 a 2 ]

kp
1
1
d 4 k sin2 ( 2 )
+ D (p)p
= 21 a 2
(2)4 1 a 2
k 2 1 a 2 k 2

1
1
1
d 4k
+ D (p)p
1 cos(k p)
=
2
(2)4
k 2 1 a 2 k 2

1

1
I 1 a 2 I,p 1 a 2
I (0) I,p (0) + D (p)p ,
=
2
2
= (1 a) i
2 2
(33)
where again D is finite for all p, and vanishes also in the limit 0. Rather
unexpectedly, this graph, which manifestly vanishes in the commutative limit, and involves
the U (1) component of the matter field, cancels the 1 pieces in (32), leaving behind a
residual divergence (for nonzero p) that depends only on the coupling to the Tr( 2 )2 term
in the potential.
However, in the corresponding gauge theory, the term

Tr 2 Tr 2
(34)
is not gauge invariant even under the spacetime integral. In fact no term involving the
product
D of more than one trace in the adjoint representation is gauge invariant (even under
d x) in noncommutative theories for N > 1. To see this write (34) in terms of its internal
indices (first choosing the canonical ordering with respect to the star product) and gauge
transform:
2 2
j
Tr = ji i lk kl
i

j 1
j
j2
Ui1
ji11 Uj Uj 2 i2 Uii2

k
k1
l
l2
l1
Ull1 Ul2
k2
Ukk2
Uk1
j1
i
k
k1
l1
ji11 i2 Uii2 Uk1
l1
k2
Ukk2 .
= Ui1
(35)
625
The presence of the star product does not allow us to use Uji Uk = ki on the remaining
local U and U factors which are separated by two factors of , even if we use the
cyclicity property of the star product under the spacetime integral. This is to be contrasted
with a single internal index trace term (with canonical internal index ordering), and the
commutative limit where the ordering of components is immaterial.
It is clear this argument applies both to the other internal index ordering ji lk
j
i kl (whose gauge transformation does not allow the use of U U = I anywhere),

and to any product of (internal index) traces in the adjoint representation. Thus if we forbid
[Tr( 2 )]2 from the scalar potential, by regarding the global theory as the limit of a gauge
theory, we have no remaining violation of Goldstones theorem for this model. Incidentally,
this argument also forbids the other terms still allowed by the imposition of the discrete
symmetry that we neglected when we wrote the scalar potential for this theory; namely

4
2

Tr() , Tr 2 Tr()
Tr() Tr 3 , Tr() Tr(),
(36)
as well as other star product orderings of the Tr( 4 ) term.
An immediate consequence of the preceding argument is that for U (N) gauge theories
with adjoint scalar matter, the symmetry breaking pattern is restricted to only one of
the two possible patterns that would be allowed by the commutative limit of the theory.
Specifically, because noncommutative gauge invariance forbids Tr(2 ) Tr(2 ), vacuum
stability now requires 1 > 0, and thus allows only the breaking pattern U (N) U (n1 )
U (N n1 ) (with n1 = N/2, N even; or n1 = (N +1)/2, N odd) [12], and forbids U (N)
U (N 1) [12].
This argument has another consequence for noncommutative theories in general. As
van Raamsdonk and Seiberg [7] demonstrated in considering scalar theories with scalars
represented by N N matrices, all infrared divergences of the type found in [6] are
proportional at one-loop to
Tr(O1 ) Tr(O2 ),
(37)
where Oi are operators built out of . Furthermore we have seen above that an operator of
this form (Tr(2 ) Tr(2 )) appearing in the scalar potential, would induce violations
of Goldstones theorem by renormalization effects. However, the preceding argument
indicates that these are precisely the form of operators that are not gauge invariant in
an adjoint representation gauge theory. So if we regard these theories as embedded in a
corresponding gauge theory where we must forbid such terms, then we would expect that
infrared divergences for N > 1 4 of the form observed in [7], no longer appear.
4. NC O(4) sigma model: adjoint representation

In this section we repeat the analysis of the previous section for the noncommutative
O(4) sigma model in the adjoint representation; again this will allow us to study,
4 For the N = 1 case considered in [6], corresponding to a single scalar, the above argument fails, since the
index structure becomes degenerate.
626
in a simple context, scalar representation (in)dependence of our results on Goldstone

renormalization, this time in the context of orthogonal symmetry groups.
We consider the classical symmetry breaking O(4) U (2). Now is a real
antisymmetric matrix, whence the vacuum state 0 can be put in standard form
0 1 0 0
a 1 0 0 0
.
0 =
(38)
2 0 0 0 1
0 0 1 0
The scalar potential is given by
2 2 1 2 2 2 4
(39)
Tr +
Tr +
Tr ,
2
4
4
where we note that the sign of the quadratic term is opposite that of the U (2) model of
the previous section because of the antisymmetry (as opposed to Hermicity) of , and
where we have normalized differently for later convenience (we now assume the canonical
internal index ordering with respect to the star product as per the conclusions of the
previous section). Thus the minimization of the potential with respect to the vacuum (38),
yields
V () =
V (0 ) =
2 2 1 a 4 2 a 4
42
2
a +
+
a2 =
=
.
2
4
16
41 + 2 1 + 2
4
(40)
The suitable parametrization of relevant to a discussion of spontaneous symmetry

breaking is given by
0
+ 1
+ 2

( + a) +
1
0
2
1
( + a) +
,
(41)
2
0
( + a)
2 ( + 1 )
( + 2 )
1
( + a)
0
where the is the field acquiring the VEV, and 1 , 2 are the two Goldstone modes.
Focussing now on the one-loop mass renormalization of one of the s, say 1 , the
expansion of the potential reads

1
1
2
1
1 +
14
V = 22 2 + 2 a 2 /2 2 + 2 + 2 +
2
2
4
4

1 2
+
+
12 22 + 2 + 2 + 2 + 2
2
4
2
+ [1 2 1 2 + 1 1 + 1 1 + 1 1 1 1 ]

8

2
32
a 2 + 12 + 22 + 1 +
a 2 + 2 + 2
+ 1 +
4
4
+ ,
(42)
where the ellipsis represents (four-field) terms that do not contribute to the one-loop mass
renormalization of 1 . The Feynman rules for these vertices are in the appendix. Now that
627
we have six distinct fields, we simply use dotted lines to denote the s, and use solid lines
for the other four fields, and instead explicitly label the lines.
The 1PI graphs contributing are
(a),
(c),
(b),
(d),
(e),
(43)
and are given respectively by

kp
2 i
d 4 k 1 + cos2 ( 2 )
(a) = 2i 1 +
4 2 (2)4
k2

2
2I (0) + I,p (0) ,
= 1 +
4

2
2
I (0) I,p (0),
(b) = 1 +
2
4

2

2
(c) = 3 1 +
I 2 a 2 /2 + I,p 2 a 2 /2 ,
2
4

2 2 2
I 2 I,p 22 ,
(d) = 1 +
2
4
2

cos2 ( pk
2
d 4k
2 )
a i2
(e) = 2i 1 +
4
2
2
4
(2) k [(p + k) 22 ]

2(1 + 42 )2 a 2
1
1
d 4k
=
+ D p
1 + cos(p k)
22
(2)4
k 2 22 k 2

2 2
2
= 1 +
I 2 I (0) + 1 +
I,p 22 I,p (0)
4
4
+ D p ,
(44)
where D is finite for all p, and where the factor of three in the third graph originates from
having three species of particle with the same contribution to this calculation.
The one-point tadpoles contributions are
(f ),
(g),
628
(h)
(45)

2
2
i
i
(f ) = 2 (2i) 1 +
(2i) 1 +
a
a I (0)
2
4
2
4
2

2
= 2 1 +
I (0),
4

2
2
i
i
(6i)
+
a
a I 22
(g) = 2i 1 +
1
2
4
2
4
2

2 2
= 3 1 +
I 2 ,
4

2 a 2
2
32
i
i
I
(h) = 3 (2i) 1 +
(2i)
+
a
a
1
4
22
4
2
2

32
2 a 2
= 3 1 +
I
,
4
2
(46)
where the overall factors of two and three in the first and third graphs, respectively, again
come from the multiplicity of particle species with the same contribution.
Thus the total one-loop contribution to the mass renormalization of the 1 (or 2 ) in this
model is

2
32
2 a 2
2 a 2
I (0) I,p (0)
I
I,p
=
4
4
2
2
2
2
1 I 2 I,p 2 + D p .
(47)
Unlike the U (2) adjoint representation model, there is no purely noncommutative graph
that saves us for either quartic invariant, and so again we cannot take the continuum limit
(UV ) and Goldstones theorem fails for this model.
5. Discussion
To summarize: in noncommutative field theory, U (N) (N > 1) linear sigma models
with complex scalars in the fundamental representation, do not have O(2N) global
invariance due to noncommutative commutator interactions between the real components,
which vanish in the commutative limit. As a result of these commutator interactions,
noncommutative linear U (N) sigma models with fundamental matter can be continuum
renormalized while preserving NambuGoldstone symmetry realization, at least at
one-loop. This contrasts with our previous results [8], where we demonstrated that
for noncommutative linear O(N) sigma models with fundamental matter, continuum
renormalization is inconsistent with NambuGoldstone symmetry realization already at
one loop (except for the degenerate Abelian case O(2) U (1)).
629
To investigate possible scalar representation dependence of these contrasting results,

we have considered linear sigma models with adjoint matter. For the adjoint U (2) linear
sigma model, we again find that NambuGoldstone symmetry realization survives at oneloop, provided we drop interaction terms (and star product orderings) which would be
inconsistent with the gauging of the symmetry; noncommutative restrictions on the allowed
operators in a U (N) gauge theory Lagrangian also act to restrict the allowed symmetry
breaking patterns. For the adjoint O(4) linear sigma model, we find violations of Nambu
Goldstone symmetry realization at one-loop order, as in the fundamental O(N) models.
These results suggest that the difference in behaviour is determined by the symmetry group,
as opposed to the scalar representation thereof.
Some of the results in this paper are suggested by D-brane dynamics. Consider the
coincidence of N D3-branes in type IIB string theory, in a constant NSNS background.
It describes a U (N) gauge theory in the decoupling limit. The separation of k < N
D3-branes from the other branes spontaneously breaks the U (N) gauge symmetry down
to U (N k) U (k), the global limit (gYM 0) of which is the process described in
Section 3. On the other hand, since orthogonal groups are realized on branes by orientifold
projections, which project out the NSNS field responsible for noncommutativity on
the brane, we expect that similar constructions with O(N) groups should encounter
difficulties, and our arguments bear this out. 5
Our results for the noncommutative linear U (N) sigma models now open the
possibility of building models of the elementary particles and their interactions based on
noncommutative non-Abelian theories with spontaneous symmetry breaking. Clearly, to
make particle physics models, it is necessary that the spontaneous symmetry breaking be
consistent with the renormalization not just of the global limits of these theories, but also
with their gaugings; we see two reasons to be sanguine on this point, at least at one-loop.
First, the gauging of the U (1)(= O(2)) model [9] is consistent with spontaneous symmetry
breaking for precisely the star orderings uniquely picked out 6 by our [8] calculation of
the Goldstone violating effects in the general noncommutative O(N) fundamental linear
sigma model. Second, in our treatment of the non-Abelian U (2) model with adjoint scalars,
violations of NambuGoldstone symmetry realization vanish when one restricts to the
subset of couplings which would be allowed, were the symmetry to be gauged; so the
limited evidence suggests that global theories may be a good guide to the behaviour of the
local theories, much as in the case of commutative field theories [4,5].
However, to go from models to actual theories would require demonstration of all-order
consistency of continuum renormalization of noncommutative theories with spontaneous
symmetry breaking. While failure of NambuGoldstone symmetry breaking can be
demonstrated at one-loop, demonstrating consistency requires an all order analysis; this
remains a major open issue in this field.
5 See however [13] for a discussion of the possibility of working with non-Lie algebra valued NC SO(N ) gauge
groups.
6 To make the anomalous effects vanish, which can happen only in the Abelian O(2) case.
630
Acknowledgements
We would like to thank the referee for useful suggestions. We would also like to thank
Andrzej Czarnecki for a useful conversation. This research is supported in part by the
Natural Sciences and Engineering Research Council of Canada.
Appendix A
A.1. Scalar potential Feynman rules, U (N) fundamental
All momenta flow into the vertices:

i
= 2i ij kl e 2 (p1 p2 +p3 p4 )

i
+ il j k e+ 2 (p1 p2 +p3 p4 ) ,

p1 p2
p3 p4
= 2i cos
cos
2
2

p1 p3
p2 p4
+ cos
cos
2
2

p1 p4
p2 p3
+ cos
cos
,
2
2

p1 p2
p3 p4
= 2i 2 cos
cos
2
2

p1 p3 p2 p4
,
+
cos
2
2
ij 2i (p1 p2 )
= 2i e
= 2i ij e 2 (p1 p2 ) sin

p3 p4
cos
,
2

p3 p4
,
2
(A.1)
(A.2)
(A.3)
(A.4)
(A.5)

p1 p2
p1 p3
= 2ia cos
+ cos
2
2

p2 p3
,
+ cos
2
631
(A.6)
= 2ia ij e 2 (p1 p2 ) ,
(A.7)

p1 p2
= 2ia cos
.
2
(A.8)
A.2. Scalar potential Feynman rules, U (2) adjoint


p1 p2 p3 p4
+
= 2i(1 + 2 ) cos
2
2

p2 p4
p1 p3
cos
,
2i2 cos
2
2

p1 p2
p3 p4
= 2i(1 + 2 ) cos
cos
2
2

p1 p3 p2 p4
+
,
+ i1 cos
2
2

p1 p2
p3 p4
= 2i(1 + 2 ) cos
cos
2
2

p1 p3 p2 p4
+
i1 cos
,
2
2

p1 p2
p3 p4
= 2i(1 + 2 ) cos
cos
2
2

p1 p3 p2 p4
+
i1 cos
,
2
2
(A.9)
(A.10)
(A.11)
(A.12)
632

p3 p4
p1 p3
p2 p4
p1 p2
cos
+ cos
cos
= 2i cos
2
2
2
2

p1 p4
p2 p3
+ cos
(A.13)
cos
,
2
2

p1 p2 p3 p4
,
+
= 1 sin
2
2
(A.14)

p1 p2
,
= 2ia cos
2
(A.15)

p1 p2
p1 p3
= 2ia cos
+ cos
2
2

p2 p3
,
+ cos
2

p1 p2
,
= 2i( + 1 )a cos
2
(A.16)
(A.17)

p1 p2
= 1 a sin
.
2
(A.18)
633
A.3. Scalar potential Feynman rules, O(4) adjoint (partial)


2
p1 p2
p3 p4
= 2i 1 +
cos
cos
4
2
2

p2 p4
p1 p3
cos
+ cos
2
2

p1 p4
p2 p3
+ cos
(A.19)
cos
,
2
2

p1 p2
p3 p4
2
cos
cos
= 2i 1 +
2
2
2

i2
p1 p3 p2 p4
+
cos
+
,
2
2
2

2
p1 p2
p3 p4
= 2i 1 +
cos
cos
2
2
2

p1 p3 p2 p4
i2
,
cos
+
2
2
2

p1 p2
p3 p4
2
cos
cos
= 2i 1 +
2
2
2

p1 p3 p2 p4
i2
cos
+
+
,
2
2
2
(A.20)
(A.21)
(A.22)

p1 p2
2
a cos
,
= 2i 1 +
4
2

2
p1 p2
p1 p3
= 2i 1 +
a cos
+ cos
4
2
2

p2 p3
,
+ cos
(A.23)
2
634

p1 p2
32
a cos
.
= 2i 1 +
4
2
(A.24)
References
[1] An incomplete list is given by:
T. Eguchi, R. Nakayama, Phys. Lett. B 122 (1983) 59;
A. Gonzalez-Arroyo, M. Okawa, Phys. Rev. D 27 (1983) 2397;
A. Gonzalez-Arroyo, C.P. Korthals Altes, Phys. Lett. B 131 (1983) 396;
T. Filk, Phys. Lett. B 376 (1996) 5358;
J.C. Varilly, J.M. Gracia-Bondia, Int. J. Mod. Phys. A 14 (1999) 1305, hep-th/9804001;
M. Chaichian, A. Demichev, P. Presnajder, hep-th/9812180, hep-th/9904132;
C.P. Martin, D. Sanchez-Ruiz, Phys. Rev. Lett. 83 (1999) 476479, hep-th/9903077;
M.M. Sheikh-Jabbari, JHEP 9906 (1999) 015, hep-th/9903107;
T. Krajewski, R. Wulkenhaar, hep-th/9903187;
S. Cho, R. Hinterding, J. Madore, H. Steinacker, hep-th/9903239;
D. Bigatti, L. Susskind, hep-th/9908056;
H.B. Benaoum, hep-th/9912036;
C.S. Chu, F. Zamora, JHEP 0002 (2000) 022, hep-th/9912153;
S. Iso, H. Kawai, Y. Kitazawa, hep-th/0001027;
H. Grosse, T. Krajewski, R. Wulkenhaar, hep-th/0001182;
W. Fischler, J. Gomis, E. Gorbatov, A. Kashani-Poor, S. Paban, P. Pouliot, JHEP 0005 (2000)
024, hep-th/0002067;
J.M. Gracia-Bondia, C.P. Martin, Phys. Lett. B 479 (2000) 321328, hep-th/0002171;
R. Oeckl, Nucl. Phys. B 581 (2000) 559574, hep-th/0003018;
W. Fischler, E. Gorbatov, A. Kashani-Poor, R. McNees, S. Paban, P. Pouliot, JHEP 0006 (2000)
032, hep-th/0003216;
F. Zamora, JHEP 0005 (2000) 002, hep-th/0004085;
J. Gomis, K. Landsteiner, E. Lopez, hep-th/0004115;
I. Mocioiu, M. Pospelov, R. Roiban, hep-ph/0005191;
A. Armoni, hep-th/0005208;
S. Kar, hep-th/0006073, hep-th/9911251;
C.P. Martin, F. Ruiz Ruiz, hep-th/0007131;
W.-H. Huang, hep-th/0009067;
I. Chepelev, R. Roiban, JHEP 0005 (2000) 037, hep-th/9911098;
I.Ya. Arefeva, D.M. Belov, A.S. Koshelev, Phys. Lett. B 476 (2000) 431436, hep-th/9912075;
M. Hayakawa, hep-th/9912094, hep-th/9912167;
C.S. Chu, F. Zamora, JHEP 0002 (2000) 022, hep-th/9912153;
I.Ya. Arefeva, D.M. Belov, A.S. Koshelev, hep-th/0001215;
S. Ferrara, M.A. Lledo, JHEP 0005 (2000) 008, hep-th/0002084;
S. Terashima, Phys. Lett. B 482 (2000) 276282, hep-th/0002119;
A.A. Bichl, J.M. Grimstrup, H. Grosse, L. Popp, M. Schweda, R. Wulkenhaar, hep-th/0007050;
A. Matusis, L. Susskind, N. Toumbas, hep-th/0002075;
I.Ya. Arefeva, D.M. Belov, A.S. Koshelev, O.A. Rytchkov, hep-th/0003176;
O. Andreev, H. Dorn, hep-th/0003113;
Y. Kiem, S. Lee, hep-th/0003145;
A. Bilal, C.S. Chu, R. Russo, hep-th/0003180;
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
635
J. Gomis, M. Kleban, T. Mehen, M. Rangamani, S. Shenker, hep-th/0003215;

K. Matsubara, Phys. Lett. B 482 (2000) 417, hep-th/0003294;
C.-S. Chu, R. Russo, S. Sciuto, hep-th/0004183;
Y. Kiem, S. Lee, J. Park, hep-th/0008002;
H.O. Girotti, M. Gomes, V.O. Rivelles, A.J. da Silva, hep-th/0005272;
A. Connes, M.R. Douglas, A. Schwarz, JHEP 9802 (1998) 003, hep-th/9711162;
N. Nekrasov, A. Schwarz, Commun. Math. Phys. 198 (1998) 689703, hep-th/9802068;
H. Aoki, N. Ishibashi, S. Iso, H. Kawai, Y. Kitazawa, T. Tada, Nucl. Phys. B 565 (2000) 176
192, hep-th/9908141;
M. Li, Nucl. Phys. B 499 (1997) 149158, hep-th/9612222;
N. Ishibashi, S. Iso, H. Kawai, Y. Kitazawa, Nucl. Phys. B 573 (2000) 573593, hep-th/
9910004;
I. Bars, D. Minic, hep-th/9910091;
J. Ambjorn, Y.M. Makeenko, J. Nishimura, R.J. Szabo, JHEP 9911 (1999) 029, hep-th/9911041;
S. Iso, H. Kawai, Y. Kitazawa, Nucl. Phys. B 576 (2000) 375398, hep-th/0001027;
R. Gopakumar, S. Minwalla, A. Strominger, JHEP 0005 (2000) 020, hep-th/0003160;
N. Ishibashi, S. Iso, H. Kawai, Y. Kitazawa, hep-th/0004038;
J. Ambjorn, Y.M. Makeenko, J. Nishimura, R.J. Szabo, JHEP 0005 (2000) 023, hep-th/0004147;
L. Alvarez-Gaume, J.L.F. Barbon, hep-th/0006209;
A.S. Gorsky, Y.M. Makeenko, K.G. Selivanov, hep-th/0007247;
C.-G. Zhou, hep-th/0007255;
I. Chepelev, R. Roiban, hep-th/0008090;
J.L. Hewett, F. Petriello, T.G. Rizzo, hep-ph/0010354;
C.P. Martin, D. Sanchez-Ruiz, hep-th/0012024.
V. Schomerus, JHEP 9906 (1999) 030, hep-th/9903205.
N. Seiberg, E. Witten, JHEP 9909 (1999) 032, hep-th/9908142.
B.W. Lee, Nucl. Phys. B 9 (1969) 649;
J.-L. Gervais, B.W. Lee, Nucl. Phys. B 12 (1969) 627;
K. Symanzik, Lett. Nuovo Cimento 1 (1969) 10;
K. Symanzik, Commun. Math. Phys. 16 (1970) 48.
M. Veltman, Nucl. Phys. B 7 (1968) 637;
G. t Hooft, Nucl. Phys. B 35 (1971) 167;
G. t Hooft, M. Veltman, Nucl. Phys. B 44 (1972) 189;
G. t Hooft, M. Veltman, Nucl. Phys. B 50 (1972) 318.
S. Minwalla, M.V. Raamsdonk, N. Seiberg, JHEP 0002 (2000) 020, hep-th/9912072.
M.V. Raamsdonk, N. Seiberg, JHEP 0003 (2000) 035, hep-th/0002186.
B.A. Campbell, K. Kaminsky, Nucl. Phys. B 581 (2000) 240, hep-th/0003137.
F. Petriello, hep-th/0101109.
T.-P. Cheng, L.-F. Li, Gauge Theory of Elementary Particle Physics, Oxford Univ. Press, 1984.
S.S. Gubser, S.L. Sondhi, hep-th/0006119.
L.-F. Li, Phys. Rev. D 9 (1974) 1723.
L. Bonora, M. Schnabl, M.M. Sheikh-Jabbari, A. Tomasiello, hep-th/0006091.

Analytic loops and gauge fields

E.K. Loginov
Physics Department, Ivanovo State University, Ermaka St. 37, Ivanovo 153377, Russia
Received 27 October 2000; accepted 3 May 2001
Abstract
In this paper the linear representations of analytic Moufang loops are investigated. We prove
that every representation of semisimple analytic Moufang loop is completely reducible and find all
nonassociative irreducible representations. We show that such representations are closely associated
with the (anti-)self-dual YangMills equations in R8 . 2001 Elsevier Science B.V. All rights
reserved.
1. Introduction
The pure YangMills theory defined in the four-dimensional Euclidean space has a rich
and interesting structure even at the classical level. The discovery of regular solutions to the
YangMills field equations, which correspond to absolute minimum of the action (Belavin
et al.) [1], has led to an intensive study of such a classical theory. One hopes that a deep
understanding of the classical theory will be invaluable when one tries to quantize such
a theory.
In the past few years, increased attention has been paid to gauge field equations in
spacetime of dimension greater than four, with a view to obtaining physically interesting
theories via dimensional reduction [2]. A telling illustration of this is the geometrical
Higgs mechanism due to Englert and others [3]. At the same time, Corrigan et al. [4]
have obtained an analog of (anti-)self-dual YangMills equations in R8 . Later there were
found its several solutions [5,6] which were then used to construct string and membrane
solitons [6,7]. In recent papers [8], the 8D equations have been applied to construct a
topological YangMills theory on Joyce manifolds as an 8D counterpart of the Donaldson
Witten theory [9]. It is also recently discussed that self-dual YangMills gauge fields
depending only upon time play a role in the context of M-theory [10].
From the viewpoint of mathematical physics, the above works has made most
conspicuous the possibly central role played by octonions [11] and their attending Lie
E-mail address: loginov@ivanovo.ac.ru (E.K. Loginov).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 2 1 8 - 8
E.K. Loginov / Nuclear Physics B 606 [PM] (2001) 636646
637
groups [12]. The algebra of octonions (Cayley numbers) is the most known example of
nonassociative alternative algebras. The alternative algebras are closely associated with
the Malcev algebras and analytic Moufang loops. These algebraic structures are actively
investigated and applied in physics [13]. Our work is a step in this direction.
The paper is organized as follows. Section 2 contains well known facts of the structural
theory of alternate algebras, Malcev algebras, and analytic Moufang loops which we use.
Sections 3 and 4 present our main results. In Section 3 the theorem on connection of
analytic Moufang loop representations and its tangent Malcev algebra representations is
proved. The complete reducibility of semisimple analytic Moufang loop representations
is proved and all irreducible nonassociative representations of analytic Moufang loops are
found. In Section 4 we show that such representations are closely associated with the (anti)self-dual YangMills equations in R8 and find solutions of the latter.
2. Analytic Moufang loops

A loop is a binary system G with an unity element 1, in which the equations ax = b and
ya = b are uniquely solvable for all a, b G. Moufang loops are distinguished from the
class of all loops by the identities
ab ca = a(bc a),
(cb c)a = c(b ca),
a(b cb) = (ab c)b,
(1)
which are called the central, left, and right Moufang identity accordingly. It is well
known [14], that two of them are a corollary of third. This paper is concerned with analytic
Moufang loops; that is, analytic manifolds equipped with the Moufang loop structure, in
which the binary operations are analytic.
Moufang loops are closely associated with the alternative algebras which are defined by
the identities
x 2 y = x(xy),
yx 2 = (yx)x.
(2)
It is evident from the definition that any associative algebra is alternative. We shall now a
construction of the most important alternative algebras which are not associative, namely,
the CayleyDickson algebras [15,16].
Let A be an algebra with an involution x x over a field F (here and below we consider
finite-dimensional algebras of characteristic 0 with identity element). Given a nonzero
F we define a multiplication on the vector space (A, ) = A A by

(x1 , y1 )(x2 , y2 ) = x1 x2 y2 y1 , y2 x1 + y1 x2 .
(3)
This makes (A, ) an algebra over F . It is clear that A is isomorphically embedded into
(A, ) and dim(A, ) = 2 dim A. Let e = (0, 1). Then e2 = end (A, ) = A Ae.
Given any z = x + ye in (A, ) we suppose z = x ye. Then the mapping z z is an
involution in (A, ).
Starting with the base field F the CayleyDickson construction leads to the following
tower of alternative algebras:
638
1. F , the base field.

2. C() = (F, ), a field if x 2 + is the irreducible polynomial over F ; otherwise,
C() F F .
3. H(, ) = (C(), ), a generalized quaternion algebra. This algebra is associative
but not commutative.
4. O(, , ) = (H(, ), ), a CayleyDickson algebra. Since this algebra is not
associative the CayleyDickson construction ends here.
The algebras in 14 are called composition. Any of them has the nondegenerate quadratic
form (norm) n(x) = x x,
such that n(xy) = n(x)n(y). In particular, over the field R of real
numbers, the above construction gives 3 split algebras (if = = = 1) and 4 division
algebras (if = = = 1): the fields of real R and complex C numbers, the algebras of
quaternions H and octonions O, taken with the Euclidean norm n(x) = |x|. Note also that
any simple nonassociative alternative algebra is isomorphic to the CayleyDickson algebra
O(, , ) [16].
The set of all regular elements of alternative algebra A over R is an analytic Moufang
loop. Its tangent algebra is isomorphic to the commutator algebra A() of A. If A is
nonassociative algebra, then A() is not a Lie algebra. Instead of the Jacobi identity, A()
satisfies the Malcev identity
J (x, y, xz) = J (x, y, z)x,
(4)
where J (x, y, z) (xy)z + (yz)x + (zx)y is the Jacobian of x, y, z, and the Sagle identity
(xy z)t + (yz t)x + (zt x)y + (tx y)z = (xz)(yt).
(5)
The identities (4) and (5) are equivalent [17]. An anticommutative algebra whose
multiplication satisfies these identities is called a Malcev algebra.
Malcev algebras and alternative algebras are closely associated. Any simple non-Lie
Malcev algebra is isomorphic to the commutator algebra of CayleyDickson algebra
[18,19]. In particular, there exists a unique simple compact non-Lie Malcev algebra over R.
It is isomorphic to the seven-dimensional algebra O() . Any semisimple Malcev algebra
A is decomposed in the direct sum A = N(A) J (A) of Lie center N(A) and ideal J (A).
Besides, N(A) is a semisimple Lie algebra and J (A) is a direct sum of simple non-Lie
Malcev algebras [19,20].
There exists a correspondence between local analytic Moufang loops and real Malcev
algebras [21]. It generalizes the classical correspondence between local Lie groups and
Lie algebras. The correspondence is completely transferred for (global) analytic Moufang
loops [22]. Namely, there exists a unique simply connected analytical Moufang loop G
with the given tangent Malcev algebra, and any connected analytic Moufang loop with the
same tangent algebra is isomorphic to the quotient algebra G/H where H is a discrete
central normal subgroup in G. Any simply connected semisimple analytic Moufang loop
is decomposed in a direct product of semisimple Lie group and simple nonassociative
Moufang loops each of which is analytically isomorphic to one of the spaces S7 , S3 R4 ,
or S7 R7 . Actually, any simply connected simple nonassociative Moufang loop is
isomorphic to the loop of elements of norm 1 in the CayleyDickson algebra over R or C.
639
3. Bimodules and birepresentations

We recall the concepts of bimodule and birepresentation of linear algebra [23]. Let K
be a class of linear algebras over a field F . Suppose we have a pair of linear mappings
(, ) : A End M of A K into a vector space M over F . Then we can define a
multiplication on the vector space direct sum A M by
(a, x)(b, y) = (ab, xb + ya ).
(6)
It is clear that this operation is bilinear, so A M is an algebra. Moreover, A is a subalgebra

and M is an ideal in A M such that M 2 = 0. The algebra A M is called the split null
extension of A. If A M K, then M is called a bimodule for A (or A-module) and the
pair of mappings (, ) a birepresentation of A in the class K.
It is comfortable to use the designations xa and ax instead of xa and xa . Clearly, we
can consider the algebra A as A-bimodule if xa and ax are the multiplication in A. Such
bimodules and corresponding birepresentation a Ra , a La are called regular.
A bimodule M for an alternate algebra A is alternative A-module if and only if
(a, b, x) = (x, b, a) = (a, x, b) = (b, x, a),
(7)
where a, b A, x M and (a, b, x) = ab x a bx. Examples of alternative bimodules

are the regular bimodule for an alternative algebra and the CayleyDickson bimodule for
an associative composition algebra. The latter is the A-submodule Ae of the composition
algebra (A, ) = A Ae where and are induced by the left and right multiplications
of A on (A, ). There is a second way of representing the CayleyDickson bimodule: to
taking M = A, a and a the right multiplications in A by a and a, respectively.
It is known [24], that every alternative bimodule for a semisimple alternative algebra
is completely reducible. The irreducible bimodule for an alternative algebra is either
associative or a submodule of the regular bimodule for the CayleyDickson algebra or
a submodule of the CayleyDickson bimodule for the generalized quaternions algebra.
By (5) the linear mapping : A End M is called a (right) representation of Malcev
algebra A if
xyz = x y z z x y + y zx yz x ,
(8)
for all x, y, z A. In this case M is called a Malcev A-module. Since the algebra A
is anticommutative, the concepts of Malcev A-module and bimodule are equivalent. An
special case of the Malcev representation is the regular representation x Rx .
It is known [19,25], that every representation of semisimple Malcev algebra is
completely reducible. The irreducible bimodule is either Lie or the regular bimodule
for a simple nonassociative Malcev algebra or sl(2, F )-module of dimension 2 such that
(a) = a where a is the adjoint matrix to the matrix a sl(2, F ). Since every semisimple
Malcev algebra A can be embedded in the commutator algebra of appropriate semisimple
alternative algebra, the representation of A is connected to the representation (, ) of
the alternative algebra by the equality = .
The concepts of bimodule and birepresentation of linear algebra can be extend to
Moufang loops. Let M be a linear space over a field F and (a , a ) invertible linear
640
transformations of M where a G. Define the multiplication (6) on the set G M. If the

groupoid G M is a Moufang loop, then we say that the vector space M is G-module and
the pair of mappings (, ) : G Aut M a linear representation of G. If G is an analytical
Moufang loop, we additionally require G M to be also analytic loop.
Suppose F [G] is the formal linear envelope of the Moufang loop G and M is G-module.
Extend the representation (, ) : G Aut M from G to F [G] by F -linearity. We will have
the representation (,
) : F [G] End M of F [G]. The sets
Ker(, ) = {a G | a = a = id},

= a F [G] | a = a = 0 ,
Ker(,
)
accordingly. The representation
is called kernels of the representation (, ) and (,
)
(, ) is called faithful if its kernel coincides with the unity element of G. In the previous
paper [26], we proved the following
Proposition 1. The mappings (, ) : G Aut M are a linear representation of the
Moufang loop G if and only if:
(i) The associator (a, b, x) is skew-symmetric.
(ii) x(b ab) = (xb a)b and (ab a)x = a(b ax) for all a, b F [G] and x M.
Proposition 2. The following statements are true:
(i) Ker(, ) is a normal subloop of G.
(ii) Ker(,
) is an ideal in F [G].
is alternative.
(iii) The quotient algebra F [G]/Ker(,
)
Proposition 3. Every irreducible G-module of Moufang loop G is a submodule of an
irreducible alternative bimodule.
Now let G be a subloop of the loop of invertible elements of an alternative algebra A.
Mark out the subalgebra B in A generated by elements of G. The operators Ra and La of
the right and left multiplications by the element a G are invertible, satisfy Proposition 1,
and coincide with the identity operator only if a = e where e is unity element of G.
Therefore the pair of mappings (R, L) : G Aut B is a faithful representation of G.
Conversely, let the Moufang loop G has the faithful representation (, ) : G
Aut M. Extend it from G to F [G] by F -linearity. The homomorphism : F [G]
F [G]/Ker(,
) of algebras induces the homomorphism : G G/H of loops. We will
prove that the subloop H coincides with the kernel Ker(, ) of the representations (, ).
and (e) (g) = (e g) = 0. Therefore
Let g Ker(, ). Then e g Ker(,
)
g H and Ker(, ) H . Let g H . Then (e g) = (e) (g) = 0 end e g
Therefore g Ker(, ) and H Ker(, ). Thus, H = Ker(, ).
Ker(,
).
Since the representation (, ) is faithful, the loop G is isomorphically embedded into
the loop of invertible elements of F [G]/Ker(,
). Using Proposition 2, we get
641
Proposition 4. The Moufang loop G has a faithful representation if and only if it is

isomorphically embedded into a loop of invertible elements of alternative algebra.
Now let G be a simply connected analytical Moufang loop. In the neighborhood U of
the unity element the multiplication operation in G is expressed through the addition and
multiplication operations in the tangent Malcev algebra AG by the CampbellHausdorff
series

1
1
1
[xy]y +
x[xy] + .
xy = x + y + [xy] +
(9)
2
12
12
Define the mappings (, ) : U Aut AG by
x =

Bn
n=0
n!
x =
(x )n ,

Bn
n=0
n!
(x )n ,
(10)
where Bn are the Bernoulli numbers and yx = [yx]. Obviously, the operators x and x
are analytic, invertible, and satisfy the conditions of Proposition 1. Therefore the pair of
mappings (, ) is a representation of the local loop U . It is clear, that this representation
is faithful.
Further, let G AG be a simply connected analytic Moufang loop which is locally
isomorphic to the local loop U AG . Obviously, G is a subloop of G AG . It follows
from the left and right Moufang identities (1) that the relations
gi gj = gi gi gj g 1 ,
i
gj gi = gi gi gj g 1 ,
(11)
are valid for all gi , gj U . Since the connected loop is algebraic generated by elements of
any neighborhood of the unity element, we see that the multiplication in G AG can be
represented as (6) where the elements a, b are in G and the operators b , a are generated
by the operators gi , gi . Thus, we have the representation (, ) : G Aut AG .
Further, it follows from Proposition 2 and faithfully of representation of U that the kernel
H = Ker(, ) is a discrete normal subloop of G. It follows from connectedness of G that
H is in the center of G. Therefore there exists the representation G Aut F [H ] regular
on H and identical on G \ H (i.e., ab = ba = ab if a, b H and a = a = id if
a, b
/ H ). Obviously, the induced representation G Aut (AG F [H ]) is faithful. Using
Proposition 4, we get
Theorem 1. Every simply connected analytic Moufang loop is isomorphically embedded
into a loop of invertible elements of alternative algebra over the field of real numbers.
Now let G be an analytic Moufang loop locally isomorphic to the simply connected
analytic Moufang loop G and suppose that the tangent algebra AG has the representation
: AG M. It means that the split null extension AG M of AG is a Malcev algebra.
Expressing the multiplication operation in the local Moufang loop U M through the
addition and multiplication operations in the tangent algebra AG M, we will have the
representation U Aut M as was shown above. Continuing this representation from U
to G , we will receive the representation G Aut M.
642
Conversely, let the analytical Moufang loop G has the representation G Aut M.
This representation induces the representation (, ) : G Aut M. We embed G into
the loop A of invertible elements of alternative algebra A. The enclosure G A
induces the enclosure : AG A() . Suppose = ; then the mappings composition
: AG End M is a representation of AG . Thus, we have the following
Theorem 2. Every G-module of the analytic Moufang loop G is AG -module of its tangent
algebra. Conversely, every AG -module of the Malcev algebra AG over the field of real
numbers is G -module of simply connected analytic Moufang loop G locally isomorphic
to G.
Using Theorem 2 and Proposition 3, we get
Theorem 3. Every representation of semisimple analytic Moufang loop is completely
reducible. Every irreducible G-module of analytic Moufang loop G is either alternative
or Malcev irreducible bimodule.
4. Self-dual gauge fields in R8
Now we shall study the irreducible nonassociative representations of analytic Moufang
loops. It follows from Theorem 3 that it is enough to investigate the regular birepresentation
of complex and real CayleyDickson algebras and their commutator Malcev algebras.
Suppose A is a complex (real) CayleyDickson algebra, M is its commutator Malcev
algebra, and L(A) is the enveloping Lie algebra of regular representation of A. It is obvious
that the algebra L(A) is generated by the operators Rx and Lx , where x A. We select in
L(A) the subspaces R(A), L(A), S(A), P (A) and D(A) generated by the operators Rx ,
Lx , Sx = Rx + 2Lx , Px = Lx + 2Rx and Dx,y = [Tx , Ty ] + T[x,y] , where Tx = Rx Lx ,
accordingly. Eq. (7) imply that
[Rx , Sy ] = R[x,y] ,
(12)
[Lx , Py ] = L[y,x] .
(13)
It follows from the identities (12) and (13) that the algebra L(A) is decomposed into the
direct sums
L(A) = D(A) S(A) R(A),
(14)
L(A) = D(A) P (A) L(A),
(15)
of the Lie subalgebras D(A) S(A), D(A) P (A) and the vector spaces R(A), L(A)
(see [15]). In addition, the map x Sx from M into S(A) is a linear representation of the
algebra M, which transforms the space R(A) into M-module that is isomorphic, by (12),
to the regular Malcev M-module. We prove the following
Proposition 5. The direct summands in (14) and (15) are orthogonal with respect to the
scalar product tr{XY } on L(A).
643
Proof. Let A be the complex CayleyDickson algebra. Then A supposes the base
1, e1 , . . . , e7 such that
ei ej = ij + cij k ek ,
where the structural constants cij k are completely antisymmetric and different from 0 only
if
c123 = c145 = c167 = c246 = c257 = c374 = c365 = 1.
It is easy to see that in such base the operators
1
1
Rei = e[i0] cij k e[j k] ,
Lei = e[i0] + cij k e[j k] ,
2
2
where e[] are skew-symmetric matrixes 8 8 with the elements (e ) =
. Using the identity
cij k cmnk = im j n in j m + cij mn ,
where the completely antisymmetric tensor cij kl is defined by the equality
(ei , ej , ek ) = 2cij kl el ,
we have
Dei ,ej = 8e[ij ] + 2cij mn e[mn] .
Finally, using the identities
cmj k cnj k = 6mn ,
cij mn ckmn = 4cij k ,
we obtain
tr{Rei Sej } = tr{Rei Dej ,ek } = 0,
(16)
tr{Lei Pej } = tr{Lei Dej ,ek } = 0.
(17)
As these equalities are invariant with respect to choice of the base in A, equalities (16)
and (17) are valid for real forms of the Lie algebra L(A). This completes the proof of
Proposition 5.
Let A be the real CayleyDickson algebra with division (octonion algebra). Its
enveloping Lie algebra L(A) (in fixed base) consists of real skew-symmetric 8 8
matrixes. Therefore we can connect every element F = F e[] of L(A) with the 2-form
F = F dx dx . It follow from (16) and (17) that the factors of F are such that

1
F S(A) D(A),
F0i + cij k Fj k = 0, if
(18)
F P (A) D(A),
2

F R(A),
F0i = cij k Fj k ,
(19)
if
F L(A),
644
where there is no summing over j, k in (19), cij k = 0, and

= 0,
if F D(A),
= 1,
if F S(A) D(A), F
/ D(A) or F R(A),
= 1,
if F P (A) D(A), F
/ D(A) or F L(A).
For = 1 these are precisely the (anti-self-dual) equations of Corrigan et al. [4]. It
follows from Proposition 5 that the expressions (18) and (19) are not depend with respect
to choice of base in A. In addition, R(A) and L(A) are not Lie algebras. Therefore the
Eq. (19), in contrast to (18), are not YangMills equations.
Nevertheless, there are a solution of the Eq. (19), which generalizes the known
(anti-)instanton solution of Belavin et al. [1]. We find this solution. For this purpose we
consider L(A)-valued 1-forms
1 Rxb R x R x Rx
b
dx ,
2
2 + |x b|2
1 Lxb L x L x Lx
b
dx ,
B,b (x) =
2
2 + |x b|2
where x, b A and is a positive number. The following proposition is true.
A,b (x) =
Proposition 6. There exists a smooth pair of mappings (f, g) : A S7 , determined near

, such that A,b (x) f 1 (x) df (x) and B,b (x) g 1 (x) dg(x) as |x| for any
and b. The difference F,b G,b of corresponding 2-form F,b = dA,b + A,b A,b
and G,b = dB,b + B,b B,b is a sum of self-dual and anti-self-dual 2-forms.
Proof. Using the identities

x x(
x)x
x
x(
x)x
x
=
2
+
+
,
|x|
|x|
|x|3
|x|3
|x|5
Rx Rx = |x|2 id,
Rx R x Rx = Rx( x)x
,
Lx Lx = |x| id,
Lx L x Lx = Lx( x)x
,
we get
A,b (x) f 1 (x) df (x),
B,b (x) g 1 (x) dg(x),
where f (x) = Rx |x|1 , g(x) = Lx |x|1 , and

F,b = 2
R x R x R x R x
dx dx ,
(2 + |x b|2)2
L x L x L x L x
G,b = 2
dx dx .
(2 + |x b|2 )2
Let x = x0 + x and x = x0 x. Then, by (12) and (13), we get that F,b G,b =
R,b + L,b , where
R,b =
(2
0

22 Rei
dx x i + cij k dx j dx k ,
2
2
+ |x b| )
L,b =
645
0

22 Lei
dx x i cij k dx j dx k .
(2 + |x b|2)2
The self-dual of R,b and anti-self-dual of L,b are obvious. To conclude the proof, it
remains to note that the map f , g translates S7 in itself and are the symmetries x Rx ,
x Lx on S7 .
References
[1] A.A. Belavin, A.S. Polyakov, A.S. Schwarz, Yu.S. Tyupkin, Phys. Lett. B 59 (1975) 85.
[2] E. Cremmer, B. Julia, J. Scherk, Phys. Lett. B 76 (1978) 409;
[3] F. Englert, Phys. Lett. B 119 (1982) 339;
R. DAuria, P. Fre, P. Van Nieuwenhuizen, Phys. Lett. B 122 (1983) 22;
F. Gursey, C.-H. Tze, Phys. Lett. B 127 (1983) 22.
[4] E. Corrigan, C. Devchand, D.B. Fairlie, Nucl. Phys. B 214 (1982) 452.
[5] B. Crossman, T.W. Kephart, J.D. Stasheff, Commun. Math. Phys. 96 (1984) 431;
D.B. Fairlie, J. Nuyts, J. Phys. A 17 (1984) 2867;
S. Fubini, H. Nicolai, Phys. Lett. B 155 (1985) 369;
A.H. Bilge, T. Dereli, S. Kocak, J. Math. Phys. 38 (1997) 4804;
E.G. Floratos, A. Kehagias, Phys. Lett. B 427 (1998) 283.
[6] T.A. Ivanova, Phys. Lett. B 315 (1993) 277;
M. Gunaydin, H. Nicolai, Phys. Lett. B 351 (1995) 169;
E.G. Floratos, G.K. Leontaris, A.P. Polychronakos, R. Tzani, Phys. Lett. B 421 (1998) 125.
[7] J.A. Harvey, A. Strominger, Phys. Rev. Lett. 66 (1991) 549;
E.G. Floratos, G.K. Leontaris, Nucl. Phys. B 512 (1998) 445.
[8] B.S. Acharya, M. OLoughlin, B. Spence, Nucl. Phys. B 503 (1997) 657;
L. Baulieu, H. Kanno, I.M. Singer, Commun. Math. Phys. 194 (1998) 149.
[9] S.K. Donaldson, Topology 29 (1990) 257;
E. Witten, Commun. Math. Phys. 117 (1988) 353.
[10] T. Curtright, D.B. Fairlie, C.K. Zachos, Phys. Lett. B 405 (1997) 37;
J.M. Figueroa-OFarrill, C. Kohl, B. Spence, Nucl. Phys. B 521 (1998) 419;
C.M. Hull, Adv. Theor. Math. Phys. 2 (1998) 619.
[11] R. Dundarer, F. Gursey, C.-H. Tze, J. Math. Phys. 25 (1984) 1496.
[12] J.R. Faulkner, J.C. Far, Bull. London Math. Soc. 9 (1977) 1.
[13] J. Lohmus, E. Paal, L. Sorgsepp, Nonassociative Algebras in Physics, Hadronic Press, Palm
Harbor, FL, 1994.
[14] R.H. Bruck, A Survey of Binary System, Springer, Berlin, 1971.
[15] R.D. Schafer, An Introduction to Non-Associative Algebras, Academic Press, New York, 1966.
[16] K.A. Zhevlakov, A.M. Slinko, I.P. Shestakov, A.I. Shirsov, Rings that are Nearly Associative,
Academic Press, New York, 1982.
[17] A.A. Sagle, Trans. Amer. Math. Soc. 101 (1961) 426.
[18] A. A Sagle, Pacific J. Math. 12 (1962) 1057;
V.T. Filippov, Algebra i Logika 15 (1976) 235 (in Russian).
[19] E.N. Kuzmin, Algebra i Logika 7 (1968) 48 (in Russian).
[20] A. Elduque, Proc. Amer. Math. Soc. 107 (1989) 73.
[21] A.I. Malcev, Mat. Sb. 36 (1955) 569 (in Russian);
E.N. Kuzmin, Algebra i Logika 10 (1971) 3 (in Russian).
[22] F.S. Kerdman, Algebra i Logika 18 (1979) 523 (in Russian).
646
[23] S. Eilenberg, Ann. Soc. Math. Polon. 21 (1948) 125;

N. Jacobson, Structure and Representations of Jordan Algebras, Providence, RI, 1968.
[24] R.D. Schafer, Trans. Amer. Math. Soc. 72 (1952) 1;
K. McCrimmon, Proc. Amer. Math. Soc. 18 (1966) 480.
[25] A. Elduque, Comm. Algebra 18 (1990) 1551.
[26] E.K. Loginov, Comm. Algebra 21 (1993) 2527.

Wick type deformation quantization

of Fedosov manifolds
V.A. Dolgushev, S.L. Lyakhovich, A.A. Sharapov
Department of Physics, Tomsk State University, Lenin ave. 36, 634050 Tomsk, Russia
Received 23 January 2001; accepted 14 May 2001
Abstract
A coordinate-free definition for Wick-type symbols is given for symplectic manifolds by means of
the Fedosov procedure. The geometry of the symplectic manifolds admitting the symbol construction
is explored and a certain analogue of the NewlanderNirenberg theorem is presented. The 2-form is
explicitly identified which cohomological class coincides with the Fedosov class of the Wick-type
star-product. For the Khler manifolds this class is shown to be proportional to the first Chern class
of a complex manifold. We also show that the symbol construction admits canonical superextension,
which can be thought of as the Wick-type deformation of the exterior algebra of differential forms on
the base (even) manifold. Possible applications of the deformed superalgebra to the noncommutative
field theory and strings are discussed. 2001 Elsevier Science B.V. All rights reserved.
PACS: 02.09.+p; 02.40.+k
Keywords: Deformation quantization; Wick symbol; Supersymplectic manifolds
1. Introduction
The deformation quantization as it was originally defined in [1,2] has now been well
established for every symplectic manifold through the combined efforts of many authors
(for review see [3]). The question of existence of the formal associative deformation for
the commutative algebra of smooth functions, so-called star product, has been solved by
DeWilde and Lecomte [4]. The classification of the equivalence classes of deformation
quantization by formal power series in the second De Rham cohomology has been carried
out in several works [58]. Finally, in the seminal paper [9] Fedosov has given an explicit
geometric construction for the star product on an arbitrary symplectic manifold making use
of the symplectic connection. As it was shown by Xu [10] every deformation quantization
on a symplectic manifold is equivalent to that obtained by Fedosovs method. Recently, the
E-mail addresses: vald@phys.tsu.ru (V.A. Dolgushev), sll@phys.tsu.ru (S.L. Lyakhovich),
sharapov@phys.tsu.ru (A.A. Sharapov).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 2 4 1 - 3
648
V.A. Dolgushev et al. / Nuclear Physics B 606 [PM] (2001) 647672
relationship has been established between the Fedosov quantization and BRST method for
the constrained dynamical systems [11].
In parallel with the general theory of deformation quantization some special types of
star-products, possessing additional algebraic/geometric properties, have been studied as
well. Motivated by the constructions of geometric quantization and symbol calculus on the
Khler manifolds, a particular emphasis has been placed on the deformation quantization of
symplectic manifolds admitting two transverse polarizations. The respective quantization
constructions may be thought of as a generalization of the Wick or qp-symbol calculus,
known for the linear symplectic spaces, much as the Fedosov deformation quantization
may be regarded as a generalization of the WeylMoyal star-product construction. At
present, there is a large amount of literature concerning the deformation quantization
on polarized symplectic manifolds [1221] beginning with the pioneering paper by
Berezin [22] on the quantization in complex symmetric spaces.
It should be emphasized that in all the papers cited above the construction of the Wicktype star-products is based on the explicit use of a special local coordinate system adapted
to the polarization (separation of variables in terminology of work [15]). This does not
seem entirely adequate for the physical applications as the most of interesting physical
theories are formulated in general covariant way, i.e., without resorting to a particular
choice of coordinates. So, it is desirable to relate the polarization with an additional
geometric structure (tensor field) on the symplectic manifold in such a way that the
resulting star-product construction would not imply to use any particular choice of the
coordinates. One may further treat this structure, if necessary, as dynamical field encoding
all the polarizations, just as the Einsteins equations for metric field govern the geometry
of Riemannian manifolds, and try to assign this structure with a physical interpretation.
In this form the Wick symbols may enjoy some interesting physical applications, two
of which (noncommutative field theories on curved symplectic manifolds and nonlinear
sigma-models) are discussed in the concluding section.
Although it is a commonly held belief that, leaving aside global geometrical aspects,
all the quantizations are (unitary or formally) equivalent to each other, this equivalence
can be spoiled because of quantum divergences which might appear as soon as one deals
with infinite-dimensional symplectic manifolds (field theories). This is best illustrated by a
concrete physical example of the free bosonic string. Formally, the theory may be quantized
by means of both Weyl and Wick symbols, but the critical dimension of space-time and
nontrivial physical spectrum are known to arise only for the Wick symbol. This is due to
the infinities that appear when one actually tries to apply the equivalence transformation to
the operators of physical observables. Thus the different quantization schemes may lead to
essentially different quantum mechanics and the Wick symbols are recognized to be more
appropriate for the field-theoretical problems. On the other hand, the global aspects of
quantization cannot be surely ignored in a mathematically rigor treatment of the question.
It is the account of a phasespace topology that gives a deep geometrical insight into the
quantization of such paramount physical characteristics as the spin, magnetic and electric
charge of the particle, energy levels of hydrogen atom and harmonic oscillator and so on,
both in the frameworks of geometric or deformation quantizations [23,24]. Finally, even
649
though two quantization are equivalent in a mathematical sense this does not yet imply their
physical equivalence since, for instance, the same classical observables being quantized in
either case will have different spectra in general [25].
In this paper we give a constructive coordinate-free definition for the star-product of
Wick type in the framework of the Fedosov deformation quantization. After the paper [26],
a symplectic manifold equipped by a torsion-free symplectic connection is usually called
the Fedosov manifold because precisely these data, symplectic structure and connection, do
enter the Fedosov star-product [9]. The Wick deformation quantization involves one more
geometric structure a pair of transverse polarizations, and, by analogy to the previous
case, the underlying manifold may be called as the Fedosov manifolds of Wick type or
FW-manifold for short.
Let us briefly outline the key idea underlying our approach to the construction of Wicktype symbols on general symplectic manifolds. Hereafter the term Wick-type will be
used in a reference to a broad class of symbols incorporating, along with the ordinary
(genuine) Wick symbols, the so-called qp-symbols as well as various mixed possibilities
commonly regarded as the pseudo-Wick symbols. To give a more precise definition of
what is meant here, consider first the linear symplectic manifold R2n equipped with the
canonical Poisson brackets {y i , y j } = ij . Then the usual WeylMoyal product of two
observables, defined as

i h ij
a(y)b(z)|z=y ,
a b(y) = exp
2
y i zj
turns the space of smooth functions in y to the noncommutative associative algebra with a
unit, which is called the algebra of Weyl symbols. Note that all the coordinates ys enter
uniformly in the above formula. Contrary to this, the construction of Wick-type symbols
always implies some (real, complex or mixed) polarization [23] splitting the ys into two
sets of (canonically) conjugated variables. For example, the qp-symbol construction is
based on separation of phasespace variables on the coordinates q and momenta p
(that corresponds to the choice of some real polarization) and the standard ordering
prescription, first q, then p, for any polynomial in ys observable. For the complex
polarization the same role is played by pairs of oscillatory variables q ip. Formally, the
transition from the Weyl to Wick-type symbols is achieved by adding a certain complexvalued symmetric tensor g to the Poisson tensor in the formula for the WeylMoyal
-product,

i h ij
a g b(y) = exp
a(y)b(z)|z=y ,
2
y i zj
ij = ij + g ij ,
= ,
i, j = 1, 2, . . . , 2n.
(1)
Although the associativity of the modified product holds for any constant g, the Wick-type
symbols are extracted by the additional condition
rank = corank = n.
(2)
In particular, the genuine Wick symbol corresponds to a pure imaginary g while the real
gives a qp-symbol. In either case (including the mixed polarization) the square of the
650
matrix
Iji = ik gkj
(3)
is equal to 1 and hence the operator I generates a polarization splitting the complexified
phase space C2n in a direct sum of two transverse subspaces related to the eigen values 1.
The formula (1) may serve as the starting point for the covariant generalization of the
notion of Wick symbol to general symplectic manifolds. Turning to the curved manifold M,
dim M = 2n, we just replace the constant matrix by a general complex-valued bilinear
form (x) = (x) + g(x) with the closed and non-degenerate antisymmetric part (x)
and satisfying to the above half-rank condition (2). Then each tangent space Tx M, x
M, turns to the symplectic vector space w.r.t. (x) and may be quantized by means of
the Wick-type product (1). Taking the union of all tangent spaces equipped with the starproduct we get the bundle of Wick symbols, which is a sort of quantum tangent bundle.
Then, following Fedosovs idea, we introduce a flat connection on it by adding some
quantum correction to the usual affine connection preserving (x). The flat sections of this
connection can be naturally identified with the space of quantum observables C (M)[[h ]].
Finally, the pull back of the bundle star-product via the Fedosov connection induces a
star-product on C (M)[[h ]]. The only crucial point of this program is the existence of
a torsion-free linear connection preserving . As we show bellow the necessary and
sufficient condition for such a connection to exist is the integrability of the right and left
kernel distribution of (x). When the latter condition is fulfilled is just the Levi-Civita
connection associated to the symmetric and non-degenerate form g(x) and the right and left
kernel distributions define the transverse polarization of the symplectic manifold (M, ).
Under specified conditions we refer to the pair (M, ) as the FW-manifold.
The paper is organized as follows. In Section 2 we define the Fedosov manifolds of
Wick type, discuss their geometry and give some examples. The main tool we use here
is the integrable involution structure (3) associated to any FW-manifold structure .
The deformation quantization on the FW-manifolds by Fedosovs method is presented in
Section 3. In Section 4 we pose the question about equivalence between original Fedosov
star-product (generalized Weyl symbols) and star-product of Wick-type. The answer is
follows: The only obstruction for the equivalence is associated with a non-zero De Rham
class of a certain (in general complex) 2-form, which explicit expression is given by
contraction of the curvature tensor of the FW-manifold and the corresponding involution
tensor. In the Khler case, this 2-form represents the first Chern class c1 of a complex
manifold. Section 5 is devoted to the superextension of the previous constructions to
supersymplectic manifold, which, as we show, can be canonically associated to the tangent
bundle of any FW-manifold. This superextension is a particular example of more general
construction of super-Poisson brackets and their quantization proposed some time ago by
Bordemann [27]. Here we also study the relationship between the algebra of quantum
observables on initial FW-manifold and that on its superextension. This relationship is not
so evident because the canonical projection of superextended FW-manifold to the base
(M, ) or the natural embedding of the FW-manifold to its superextension do not induce
a homomorphism of the corresponding algebras.
651
2. Fedosov manifolds of Wick type

Consider a 2n-dimensional real manifold M equipped with a complex-valued bilinear
form (not necessarily symmetric or antisymmetric). In local coordinates {x i } on M the
form is completely determined by its components ij = (i , j ), where i /x i .
Having the form one can define two maps from the comlexified tangent bundle to the
comlexified cotangent one: for any X T M the corresponding linear forms are
(, X),
t (, X) = (X, ),
(4)
ker t
being transpose to . Denote by ker and

the right and left kernel distributions
of . Obviously, dim ker = dim ker t . It is convenient to decompose into the sum of
symmetric and antisymmetric parts

1
1
= + g, = t , g = + t .
(5)
2
2
Definition 1.1. The pair (M, ) will be called by an almost Fedosov manifold of Wick
type (or almost FW-manifold for short) if at each point p M
(i) = 12 ( t ) is a real non-degenerate two-form,
(ii) dimC ker = 12 dim M.
The first condition merely implies that the antisymmetric part of defines on M an
almost symplectic structure on M. From the second condition it follows, that at each
point x M, the left and right kernel distributions span the tangent space TxC M. Indeed, the
conditions the vector fields X(Y ) belong to the left (right) kernel distributions, (X, ) =
0, (, Y ) = 0, can be rewritten in the form:
I X = X,
I Y = Y,
(6)
where the smooth field of automorphisms I (x): TxC M TxC M is defined, in any
coordinate chart on M, by the matrix
Iji = ik gkj ,
with ij being inverse to ij , ik kj = ji . Belonging to the different eigenvalues (1)
of the map I , the vectors X and Y are linearly independent and hence ker ker t =
0. Accounting the dimensions of the kernel distributions one concludes that ker (x)
ker t (x) = TxC M, x M. Additionally, we have proved that the form g is nondegenerate and the automorphism I is involutive, i.e.:
I 2 = id (identical transformation).
(7)
That is why we will call I by an almost involution structure. Note that for an anti-Hermitian
the components of tensor I are imaginary
, that is t = ,
I = 1J,
(8)
J 2 = id,
and hence they define (and defined by) an almost complex structure J . As is well
known the almost complex structure becomes the complex one if it defines a structure
652
of the complex manifold. Due to the NewlanderNirenberg identification theorem for the
complex manifolds [28] it is equivalent to vanishing of the Nijenhuis tensor associated
to J providing the latter is sufficiently smooth. The vanishing of the Nijenhuis tensor is
known to be equivalent,
in turn, to the integrability of vector distributions belonging to the
eigenvalues 1 of J . As we will see bellow the analog of the latter condition can be
advocated for the almost involution structure I as well, with precisely the same definition
of the Nijenhuis tensor.
Definition 1.2. The Nijenhuis tensor associated to the almost involution structure I is a
smooth tensor field of skew-symmetric bilinear maps N : TxC M TxC M TxC M, which
can be defined in two equivalent ways:
(i) for every pair of smooth vector fields X and Y
N(X, Y ) = [X, Y ] I [I X, Y ] I [X, I Y ] + [I X, I Y ],
(9)
where the brackets [, ] stand for the commutator of vector fields;

(ii) let be an arbitrary torsion-free connection, then in local coordinates {x i } the
components of N are given by

Nijk = Iil l Ijk Ijl l Iik Ilk i Ijl j Iil .
(10)
One may easily check that the relations (9), (10) do really define (the same) tensor as
they do not actually depend on derivatives of X, Y and on the choice of connection .
Before examining the geometric consequences of the condition N = 0, let us introduce the
key ingredient of our construction.
Definition 1.3. The almost FW-manifold (M, ) will be called the FW-manifold if there
exist a torsion-free connection preserving , i.e., = 0.
Obviously, the form is covariantly constant iff both its symmetric and antisymmetric
parts are constant, i.e.:
= 0 = g = 0.
As there is the only torsion-free connection compatible with a given non-degenerate
symmetric form g, the existence of such a connection implies its uniqueness. On the
other hand, the totally antisymmetric part of the equation = 0 written in some local
coordinates suggests that the form is closed, i.e., d = 0, and therefore any FW-manifold
is a symplectic manifold as well. Emphasize that in general we deal with the complexvalued form g, so the connection , when exists, is supposed to act in the comlexified
tangent bundle and determined, in each coordinate chart, by complex-valued Christoffel
symbols. The following theorem gives the explicit criteria for an almost FW-manifold
(M, ) to admit a torsion-free connection preserving .
Theorem 1.1. Given an almost FW-manifold (M, ) with the closed antisymmetric part
of , then the following statements are equivalent:
653
(i) defines the structure of FW-manifold,

(ii) the involution I associated to has the vanishing Nijenhuis tensor,
(iii) the kernel distributions ker and ker t are integrable.
Although these assertions are quite elementary we prove them below as they have a
direct bearing on the deformation quantization of FW-manifolds, which will be considered
in the next sections.
Proof. We will proceed following the scheme: (i) (ii), (ii) (iii).
Implication (i) (ii) straightforwardly follows from the second definition of Nijenhuis
tensor (10) in which is taken to be compatible with (and hence with I ). Conversely, let
be the torsion-free connection preserving g. Accounting that d = 0, the expression (10)
for the Nijenhuis tensor can be reduced to
Nji k = 2il l j k ,
from which the implication (ii) (i) is immediately follows. Now let X and Y be two
eigenvector fields of the involution tensor with the same eigenvalue ; of course, 2 = 1.
Evaluating N on these vector fields with the help of first definition (9) we get

N(X, Y ) = 2 [X, Y ] I [X, Y ] .
(11)
So, if N = 0 then I [X, Y ] = [X, Y ] and the eigen distributions of I are involutive. This
proves the implication (iii) (ii). The relation (11) also implies that N comes to zero
on each pair of tangent vectors belonging to the same kernel distribution, ker or ker t ,
whenever they are integrable. In the case of X and Y belonging to the different distributions
associated to the eigenvalues and , respectively, the value of the Nijenhuis tensor on
them is equal to zero identically:
N(X, Y ) = [X, Y ] I [I X, Y ] I [X, I Y ] + [I X, I Y ]
= [X, Y ] I [X, Y ] + I [X, Y ] [X, Y ] = 0.
Since the vectors from left and right kernel distributions form the basis of Tx M, x M,
the last conclusion proves the implication (ii) (iii) and the theorem.
Examples. The extended list of examples is provided by the Khler manifolds. In this
case the form is anti-Hermitian and
structure I is identified
the integrable involution
a
with the complex one multiplied by 1. In the frame {/z , / z b } associated to the
local holomorphic coordinates {za }, a = 1, . . . , n, on M the matrix of the form looks
like

0 ha b
=
,
0 0
where h = ha b dza d z b is the Khler (1, 1)-form, dh = 0. The form is preserved by
a =
the Khler connection with non-zero components of the Christoffel symbols being bc
ha d hbd /zc = (bac ), the matrix hda is inverse to ha b . In particular, all two-dimensional
654
orientable manifolds admit the structure of FW-manifold as they known to admit the Khler
structure. More generally, let g be a (pseudo-)Riemannian metric on a two-dimensional
real manifold and let = | det g| dx 1 dx 2 be the corresponding volume form. Then the
form
= g +
is obviously preserved by the metric connection for any C. In order for to define the
structure of FW-manifold we have to require
det = det g + 2 | det g| = 0,
that fixes = 1 for the Riemannian metric (det g > 0) and = 1 for the pseudoRiemannian one (det g < 0). The former case corresponds to the Khler manifolds, while
the latter is concerned with the manifolds endowed by a real structure. The last situation
is strikingly illustrated by the example of one-sheet hyperboloid embedded into threedimensional Minkowsky space as the surface
x 2 + y 2 z2 = 1
(x, y, z) being linear coordinates in R 2,1 . The induced metric is of the pseudo-Euclidean
type and it has a constant negative curvature. For the corresponding real structure, the
integral leaves of the eigen distributions coincide with the two transverse sets of linear
elements of the hyperboloid. Obviously, these linear elements are nothing but the isotropic
geodesics of the metric g.
Let us recall that the integrable complex distribution P on a symplectic manifold (M, )
is called polarization if dimC P = dim M/2 and |P = 0. In other words, at each point x
M, P extracts a (complex) Lagrangian subspace Px TxC M in the complexified tangent
bundle. It is easy to see that the right and left kernel distributions on the FW-manifold
(M, ) are Lagrangian and hence define a pair of transverse polarizations PR , PL . These
polarizations are highly important for physical applications both in the framework of the
deformation quantization and geometric one as it allows one to introduce the notion of a
state of quantum-mechanical system. The quantization on symplectic manifolds endowed
with a pair of transverse polarizations was intensively studied in two limiting cases:
PR = PL , PR PL = 0. The first possibility is realized for the Khler manifolds with
holomorphicantiholomorphic polarizations, while the second implies the existence of a
pair of transverse Lagrangian foliations on M, as a consequence of the Frobenius theorem.
It is interesting to note that for the real = g + the leaves of foliations PR , PL , being
Lagrangian submanifolds with respect to the symplectic structure , turn out to be totally
geodesic submanifolds with respect to pseudo-Riemannian structure g. The proof of the
last fact is straightforward and we omit it.
3. Deformation quantization on FW-manifolds

Let, as before, (M, ) be an FW-manifold of dimension 2n. We are reminded that
can be splitted into the sum of symmetric and antisymmetric parts (5) which are both non-
655
degenerate. Introduce the second-rank contravariant tensor field ij (x)i j defined as

follows:
ij = im mn nj = g ij + ij ,
where ij and g ij are matrices inverse to ij and gij , respectively. By construction,
rank(ij ) = rank(ij ) = n
and ij is a Poisson tensor. Under the deformation quantization of the FW-manifold
(M, ) we will mean the construction of an associative multiplication operation of two
functions, which is an one-parametric deformation of the ordinary pointwise multiplication
in the algebra C (M) and which meets the boundary condition:
i h ij
i aj b + ,
(12)
2
where h is the formal deformation parameter (Plank constant), and dots mean the terms
of higher orders in h . The condition (12) is compatible with so-called correspondence
principle of quantum mechanics:
a b = ab +
lim
h 0
1
(a b b a) = {a, b},
i h
(13)
where {, } means the Poisson brackets associated to ij . The boundary condition should
also be added by the requirement of locality
supp(f g) = supp(f ) supp(g).
(14)
To meet the latter condition the coefficients of each power of h in (12) are restricted to be
finite-order differential expressions, being bilinear in a and b. Note that when h is treated
as a formal (not numerical) parameter the -product of two functions is not a function but
is an element of a more wide space C (M)[[h ]] consisting of the formal series: a(x, h ) =
a0 (x) + h a1 (x) + h 2 a2 (x) + , ai (x) C (M). The space C (M)[[h ]] is closed
already with respect to -multiplication and it may be regarded as the algebra of quantum
observables, much as the Poisson algebra of smooth functions C (M) on symplectic
manifold (M, ) is identified with the space of classical observables. The problem now is
to construct an associative -product starting from the ansatz (12). In this section we show
how this the problem can be resolved by a minimal modification of Fedosovs geometric
approach to the deformation quantization [9].

Definition 2.1. The formal symbol algebra A =
m,n=0 Am,n is the bi-graded associative
algebra over C with a unit whose elements are formal series

a(x, y, dx, h ) =
(15)
h k ak,i1 ...ip j1 ...jq (x)y i1 y ip dx j1 dx jq .
2k+p,q0
Here the expansion coefficients ak,i1 ...ip j1 ...jq (x) are the components of covariant tensor
fields on M symmetric with respect to i1 , . . . , ip and antisymmetric in j1 , . . . , jq , h is
a formal (deformation) parameter; and {y i } are variables transforming as components
656
of tangent vector, hence the whole expression (15) does not depend on the choice of
coordinates. The general term of the expansion (15) is assigned by a bi-degree (2k + p, q)
and thus belongs to the subspace A2k+p,q . The product of two elements a, b A is defined
by the rule

i h ij

a b = exp
(16)
(x) i j a(x, y, dx, h ) b(x, z, dx, h )|z=y ,
2
y z
where stands for the ordinary exterior product of differential forms.
It is easily seen that the multiplication (16) is associative and bi-graded, i.e., Am,n
Ak,l Am+k,n+l . In what follows we will refer to subscripts m and n labelling the graded
subspace Am,n as the first and second degree respectively. The natural filtration in A with
respect to the first degree,

Ak,n ,
A A1 A2 , Am
(17)
n
km
defines the topology and convergence in the space of infinite formal series (15). Being
associative algebra, A can be turned to a differential graded Lie algebra with respect to the
second degree: the commutator of two homogeneous elements a Am,n , b Ak,l is given
by [a, b] = a b (1)nl b a, and extends to the whole A by linearity; the nilpotent
differential : Am,n Am1,n+1 acts as
a = dx k
a
,
y k
(a) = 0,
a A.
(18)
Alternatively, one can write

1
ij y i dx j , a ,
i h
and hence is an inner derivation of the superalgebra and of the -product as well:
a =
(a b) = (a) b + (1)n a (b),
a A,n ,
b A.
(19)
(20)
Note that acts algebraically in a sense that it does not involve derivatives with respect to
x. The nontrivial cohomologies of corresponds to the subspace of quantum observables
C (M)[[h ]] A whose elements are independent of y and dx; for the complementary
subspace one can construct a homotopy operator 1 : Am,n Am+1,n1 of the form
a=y i
x k

1
a(x, ty, t dx, h )
dt
,
t
(21)
i(/x k )
where
means the contraction of the vector field /x k and a form. Extending
the action of 1 to C (M)[[h ]] by zero, we get the HodgeDe Rham decomposition
holding for any a A,
a = (a) + 1 a + 1 a,
where (a) = a(x, 0, 0, h ) denotes the projection of a onto
(22)
C (M)[[h]].
657
Given the torsion-free connection preserving FW-manifold structure , it induces the

covariant derivative on elements of A which will be denoted by the same symbol,

j k
: An,m An,m+1 ,
(23)
= dx i
(x)
,
ij
x i
y k
ijk are Christoffel symbols of the connection . Definition 1.3 implies the following
property of the covariant derivative:
(a b) = a b + (1)n a b,
a A,n ,
b A.
(24)
The next lemma is a counterpart of Lemma 2.4 in [9] for FW-manifolds.

Lemma 2.1. Let be the covariant derivative of A. Then
a + a = 0,
(25)
1
1
[R, a], R = Rij kl y i y j dx k dx l ,
i h
4
m
= im R j kl is the curvature tensor of the connection .
2 a = (a) =
where Rij kl
(26)
Proof. The first identity holds because is a symmetric connection. It follows from the
definition (23) that
a
1
j
2 a = dx k dx l Rikl y i j .
2
y
(27)
So, 2 is an algebraic operator. On the other hand,

1
1
i j n a
i nj a
[R, a] = Rij kl y
Rij kl y
dx k dx l
i h
4
y n
y n

2a
2a
i h
+
Rij kl j m in m n Rij kl mj ni m n dx k dx l
8
y y
y y
a
1 k
= dx dx l Rj ikl y i nj n = 2 a.
2
y
n , and its symplectic analog
Here we have used the Ricci identity, gin Rjnkl = gj n Rikl
m
m
im Rj kl = j m Rikl (for the proof see [26]). The lemma proved suggests that the square
of external derivative is a derivative again and, moreover, it is an inner derivative of the
algebra A. This fact is of primary importance for the -product construction along the line
of the Fedosov approach.
Following Fedosov, we define a more general derivative in A of the form
D= +

1
1
[r, ] = +
ij y i dx j + r, ,
i h
i h
(28)
where r = ri (x, y, h ) dx i belongs to A31 and satisfies the Weyl normalization condition
ri (x, 0, h ) dx i = 0. A simple calculation yields that
D2 a =
1
[, a],
i h
a A,
(29)
658
where
1
1
= ij dx i dx j + R r + r + r r
(30)
2
i h
is the curvature of D. A connection of the form (28) is called Abelian if two-form does
not contain ys, i.e., belongs to the center of A. It is clear, that the kernel subspace of an
Abelian connection D is automatically a subalgebra in A. Denote AD = ker D A,0 . The
following two theorems are pivot for the -product construction and they can be proved
completely analogously to Fedosovs original theorems in [9, Theorems 3.2, 3.3].
Theorem 2.1. There exists a unique r A31 obeying condition 1 r = 0 such that D, given
by (28), is an Abelian connection with the curvature = (1/2)ij dx i dx j .
Theorem 2.2. For any observable a C (M)[[h ]] there is a unique element a AD such
that (a)
= a. Therefore, establishes an isomorphism between AD and C (M)[[h ]].
Corollary 2.1. The pull-back of -product via 1 induce an associative -product on the
space of physical observables C (M)[[h ]], namely

a b = 1 (a) 1 (b) .
(31)
Besides the fact of existence these theorems provide an effective procedure for the
construction of the lifting map 1 by iterating a pair of coupled equations

1
1
r =
R + r + r r ,
i h

1 1
1
1
1
(a) = a +
(32)
(a) +
r, (a) .
i h
Since the operator preserves the filtration and 1 raises it by 1, the iteration of the
system (32) converges in the topology (17) and define the unique solution for 1 (a).
Thus the -product of two functions can be computed with any prescribed accuracy in h.
It should be noted that for the anti-Hermitian s (the case of Khler manifolds) the
introduced multiplication possesses the following property of reality:
a b = b a,
a, b C (M)[[h ]].
(33)
In particular, the formal functions whose coefficients at each power of h are real-valued
smooth functions form a closed Lie subalgebra with respect to the -commutator multiplied
by i. In the standard quantum-mechanical interpretation, this algebra is usually referred to
as an algebra of physical observables corresponding to the algebra of self-adjoint operators.
The formula (33) trivially follows from the analogous relation for -product,
a b = (1)mn b a,
a A,m , b A,n ,
and from the structure of Eqs. (31), (32).
(34)
659
Remark. As is seen, the rank condition imposed on by Definition 1.1 is certainly

inessential for the construction of associative -product obeying (12). The only fact we
have used here is that of existence of a torsion-free connection preserving . So, the
construction described above works in a more general situation including the case of
degenerate g, when g = 0 we get the Fedosov quantization. It would be interesting to
formulate the necessary and sufficient conditions for a torsion-free connection preserving
a given form to exist and to describe all such connections.
In conclusion let us introduce one more important ingredient of deformation quantization a trace functional on the algebra of quantum observables.
Let C0 (M)[[h ]] C (M)[[h ]] be an ideal consisting of compactly supported quantum
observables. Recall, that linear functional on C0 (M)[[h ]] with values in formal constants
C[[h ]] is called a trace if it vanishes on commutators, 1 that is
tr(a) =
h k ck ,
ck C,
(35)
k=0
and
tr(a b) = tr(b a).
(36)
Let d = d0 + h d1 + be a formal smooth density on M, then

integration by d
delivers a continuous functional of the form C0 (M)[[h ]] ! a M a d. A formal

density is called a trace density for -product if the functional M a d is a trace. The work
of Nest and Tsygan [5] suggests that any continuous trace functional is defined by a trace
density. For the deformation quantization on FW-manifolds a more profound statement is
true.
Theorem 2.3. Up to an overall constant factor there exist a unique trace density
associated to the algebra of quantum observables C (M)[[h ]]. It has the form

2
d = 1 + ht
(37)
1 (x) + h t2 (x) + ,
where = n /n! = | det g| dx 1 dx 2n is symplectic (Riemannian) volume form

on M and coefficients ti (x) are polynomials in curvature tensor Rij kl and its covariant
derivatives.
Proof. This theorem is the analogue of Theorem 5.6.6 in [24] and it can be proved
following the same idea based on a localization principle. Besides, in the next section we
will present the explicit form for an operator establishing a local isomorphism between the
deformation quantization on FW-manifolds and that on the Fedosov quantization on the
corresponding symplectic manifold (M, ). Using this local isomorphism one may derive
the trace density (37) from Fedosovs trace density for symplectic manifold (M, ).
1 Our definition differs from the standard one by the lack of inessential normalization multiplier 1/(2 h)n .
660
It is pertinent to note that for a homogeneous FW-manifold the corresponding symmetry

group, acting on M transitively by symplectomorphisms, defines the symmetry group of
deformation quantization. In this case one may see that d = .
4. The question of equivalence

The rich geometry of the FW-manifolds offers at least two different schemes for their
quantization: the Fedosov quantization, which exploits only the skew-symmetric part of
the form , and the deformation quantization involving entire form via ansatz (12).
For the reasons mentioned in the Introduction we refer to these quantizations as those of
Weyl and Wick type, respectively. The natural question aroused is that of whether we have
essentially different quantizations or an equivalence transform may be found to establish a
global isomorphism between both algebras of quantum observables. 2 Below we formulate
the necessary and sufficient conditions for such an isomorphism to exist. As in the general
case, the obstruction for equivalence of two star-products lies in the second De Rham
cohomology of symplectic manifold and we identify a certain 2-form as its representative.
In order to distinguish Wick-type star product from the Weyl one, all the constructions
related to the former product will be attributed by the additional symbol g (pointing
on non-zero symmetric part g in ). In particular, through this section the fibrewise
multiplication (16) will be denoted by g , while will be reserved for the Fedosov
-product [9] resulting from (16) if put g = 0.
We start by noting that fibrewise and g products are equivalent in the following sense:
Lemma 3.1. For a, b A we have
a g b = G1 (G a G b),
where the formal operator G and its formal inverse are given by

i h ij
i h
1
g
,
G
.
=
exp
G = exp g ij i
4
y y j
4
y i y j
(38)
(39)
In other words, the map G : A A, being considered as an automorphism of a linear

space, establishes the isomorphism of algebras (A, ) and (A, g ).
The proof is obvious from the direct substitution (39) to (38).
The operator G satisfies following simple properties:
G = G,
G = G.
(40)
2 Of course, a local equivalence for the deformation quantizations of symplectic manifolds is a trivial fact due to
Darbouxs theorem. For general Poisson manifolds the classification theorem (even locally) is much more subtle
problem which was solved only recently by Kontsevich [29]: All the deformation quantizations of a Poisson
manifold are proved to be in one-to-one correspondence with nontrivial deformations of the respective Poisson
bracket.
661
The automorphism G defines a new Abelian connection D = GDg G1 , which in virtue of

relations (40) can be written as
1
r = Grg ,
D = + [r , ],
i h
and the bracket stands for -commutator. The elements r and r fulfill the equations
GR + r r +
1
r r = 0,
i h
(41)
(42)
1
(43)
r r = 0.
i h
Thus we have two star-products and corresponding to the pair of Abelian connections
Since D #= D,
in general, the action of the fibrewise isomorphism G establishing
D and D.
the equivalence between - and g -products (and hence between star products and g ) is
not automatically followed by the equality = . Indeed, evaluating lowest orders in h we
get:
R + r r +

h 2
i h ij
i aj b ik j l i j ak l b + O h 3 ,
2
2

h 2
i h
a b = ab + ij i aj b ik j l i j ak l b + O h 3 ,
2
2
where
a b = ab +
ij
ij = ij + h 1 ,
(44)
(45)
and
i
1
(GR R) = Rij kl g ij dx k dx l .
(46)
8
h
The 2-form is closed in virtue of the Bianchi identity for the curvature tensor. In
fact, relations (44), (45) say that the second star product is a so-called 1-differentiable
deformation of the first one [3]. This deformation is known to be trivial iff the 2-form is
exact [3,30]. Now supposing that = d let us try to establish an equivalence between
and by means of a fibrewise conjugation automorphism
ij
1 = ik kl lj ,
a U a U 1 ,
(47)
where U is an invertible element of A,0 . The element U is so chosen that (47) turns D
In other words, we subject U to the condition
to D.

U 1 , a A,
D U a U 1 = U (Da)
or, equivalently,

1
U 1 DU, a = [5r, a],
i h
where 5r = r r. The last condition means that
U 1 DU =
1
1
5r + ,
i h
i h
(48)
(49)
662
where is a globally defined 1-form on M. The compatibility condition for Eq. (49)
resulting from the identity D 2 = 0 requires that
DU 1 DU = DU 1 U U 1 DU =
=
1
5r 5r
(i h )2
1
1
D5r + d.
i h
i h
(50)
That is
1
5r 5r + d = 0.
i h
The analogous relation is obtained if we subtract (42) from (43)
D5r +
(51)
1
5r 5r i h = 0.
(52)
i h
Comparing (51) with (52) we conclude that the compatibility condition holds provided
is exact. Now rewrite (49) in the form
D5r +
U = U +
1
1
[r, U ] U (5r + )
i h
i h
(53)
and apply the operator 1 to both sides of the equation. Using the HodgeDe Rham
decomposition (22) and taking (U ) = 1, we get

1
1
1
U =1+
(54)
U + [r, U ] U (5r + ) .
i h
i h
In [9, Theorem 4.3] it was proved that iterations of the last equation yield a unique solution
for (53) provided the compatibility condition (51) is fulfilled. Starting from 1, this solution
defines an invertible element of A,0 . Then the equivalence transform T : (C (M), )
(C (M), g ) we are looking for is defined as the sequence of maps

T a(x) = U G g1 (a) U 1 y=0 ,
(55)
so that
a g b = T 1 (T a T b).
The main results of this section can be summarized as follows:
Theorem 3.1. The obstruction to equivalence between Weyl- and Wick-type deformation
quantizations lies in the second De Rham cohomology H 2 (M). The quantizations are
equivalent iff the 2-form Rij kl g ij dx k dx l is exact.
Remark. For the anti-Hermitian the 2-form is nothing but the Ricci form of the
Khler manifold (M, ). In this case the cohomology class of , being proportional
to the first Chern class c1 (M), is known to depend only on the complex structure of
the manifold [31]. Since, for example, c1 (CP n ) #= 0 and for any Khler manifold M
a topological equivalence M CP n implies a bi-holomorphic one [32], the Weyl and
Wick quantizations on CP n are not equivalent for any .
663
5. Superextension
In this section we show that the Wick-type deformation quantization of FW-manifolds
described above admits a surprisingly simple generalization to a certain class of supersymplectic supermanifolds. According to the Rothstein theory [33] any supersymplectic supermanifold can be completely specified (up to isomorphism) by the set (M, , E, g, g ),
where (M, ) is the symplectic manifold, and E is a vector bundle over M with metric g
and g-compatible connection g . When E = T M and M is an FW-manifold all Rothsteins
data are already contained in the FW-structure , and thus, one may speak about canonical
superextension for the FW-manifolds.
As is well known, the geometry of a manifold can be recovered from the commutative
algebra of functions on it. A supermanifold is defined by extending of such an algebra to a
supercommutative superalgebra denoted below by C. Geometrically, the elements of C can
be viewed as the sections of a Grassmann bundle associated to some vector bundle E M
over a given manifold M, that is C = (E), and the supercommutative multiplication is
given by the pointwise wedge product,
a b = (1)d1 d2 b a.
(56)
Here a, b (E), a of degree d1 and b of degree d2 . The elements of superalgebra

C will be called superfunctions depending on the commuting coordinates {x i } (i.e., local
coordinates on the base (M, )) and anticommuting coordinates { i }, so that the general
element of superalgebra looks like
C ! a = a(x, ) =
2n

a(x)i1 ...ik i1 ik ,
(57)
k=1
where ak = a(x)i1 ...ik dx i1 dx ik transforms as a k-form on M. As usual the symbols
/ i and / i will denote the left and right partial derivatives in s, which definition
is as follows:
a
=
ka(x)j i2 ...ik i2 ik ,
j
2n
k=1
a
=
ka(x)i1 ...ik1 j i1 ik1 .
j
2n
(58)
k=1
In terms of supercoordinates (x i , j ) the Rothstein supersymplectic form on the superextended FW-manifold (M, ) is given by
1
j
= ij dx i dx j + gij Rmkl m i dx k dx l + gij D i D j ,
2
D i d i jik k dx j .
Hereafter we use the following conventions:
dx i dx j = dx j dx i ,
d i d j = d j d i ,
i j = j i ,
i d j = j d i .
dx i j = j dx i ,
(59)
The deformation quantization of general supersymplectic supermanifolds defined by

Rothsteins data was first performed by Bordemann [27]. The remarkable feature of this
664
construction is that the deformation quantization for the algebra of superfunctions is

performed first and the super-Poisson bracket arises here a posteriori as h -linear term of
the supercommutator. Below we show how this quantization can be extended to generate
a super-Wick symbols associated to the super-FW-manifold. In view of natural bijection
between superfunctions and inhomogeneous differential forms,
a(x)i1 ...ik i1 ik a(x)i1 ...ik dx i1 dx ik ,
this construction may also be thought of as a deformation of the exterior algebra of
differential forms on M. 3
To begin with, we define a superextension of the formal symbol algebra A introduced in
the Definition 2.1.
Definition 4.1. The bi-graded superalgebra SA =
a space of formal series,
m,n=0 (SA)m,n
with unit over C is
a(x, , y, dx, h )

h k aki1 ...ip j1 ...jq & l1 ...lq && (x)y i1 y ip j1 jq & dx l1 dx lq && ,
=
(60)
2k+p0
multiplied with the help of associative -product of the form

i h ij

(x)
a b = exp
+ i
i
j
2
y z
j
a(x, y, , dx, h )b(x, z, , dx, h )|y=z,= .
(61)
A general term of the series (60) is assigned by the bi-degree (2k + p + q & , q && ).
The associativity of this -product is well known (see, e.g., [27, Proposition 2.1]) and
may be checked by straightforward computation. As before the expansion coefficients
aki1 ...ip j1 ...jq & l1 ...lq && (x) are considered to be components of covariant tensors on M
symmetric in i1 , . . . , ip and antisymmetric with respect to j1 , . . . , jq & and l1 , . . . , lq && . In
accordance with (59) one may regard dx i and j as the set of 2n anticommuting variables,
that turns SA to Z2 -graded algebra with respect to the Grassmann parity q = q & + q && (60).
The supercommutator of two homogeneous elements a, b SA with the parities q1 and q2
is defined as
[a, b] = a b (1)q1 q2 b a.
(62)
The nilpotent operator defined by Eq. (18) is still the inner derivation of the
superalgebra SA and 1 (21) is the partial homothopy operator for in the sense
of HodgeDe Rham decomposition (22), where now (a) = a(x, , 0, 0, h ). Thus
the nontrivial cohomology of coincides with the space of superobservables C[[h ]].
Obviously, SA contains A as a subalgebra. The superextension of covariant derivative (23)
3 It is pertinent to note that in the flat case the deformation of exterior form calculus was previously considered
in [34,35].
665
from A to SA is defined as follows:

: (SA)n,m (SA)n,m+1 ,

i
j k
j k
y ij (x) k ij (x) k ,
= dx
i
x
y
(63)
ijk are Christoffel symbols of the connection associated to . Then

(a b) = a b + (1)q a b,
(64)
with q being the Grassmann parity of a. It is easy to check that anticommutes with
and that its curvature is given by
2 a = (a) =
1
[R, a],
i h
(65)
1
1
R = Rij kl y i y j dx k dx l + Rij kl i j dx k dx l ,
(66)
4
4
where we have used the notations Rij kl = im Rjmkl and Rij kl = gim Rjmkl . By analogy with
nonsupercase, one can combine with an inner derivative to get the Abelian connection
of the form

1
1
ij y i dx j + r, , r = ri (x, , y, h ) dx i ,
D = + [r, ] = +
i h
i h
1
1
D 2 a = [, a] = 0, a SA, = ij dx i dx j .
(67)
i h
2
Denote SAD = ker D SA,0 . The next assertion is the supercounterpart of that stated in
Theorems 2.1, 2.2 and Corollary 2.1.
Theorem 4.1. With the above definitions and notations we have:
(i) there is a unique Abelian connection D (67) for which
1 r = 0,
ri (x, , 0, h ) = 0
and r consists of monomials whose first degree is no less then 3;

(ii) SAD is a subalgebra of SA and the map being restricted to SAD defines a
linear bijection onto C[[h ]];
(iii) the formula

a b = 1 (a) 1 (b)
defines the associative multiplication on C[[h ]].
Proof can be directly read off from [27, Theorems 2.1, 2.2].
As in nonsupercase the explicit construction for the lifting map 1 results from
iterations of two coupled Eqs. (32) with respect to the first degree, with the only difference
that R is now determined by (66) and a C[[h ]]. The first equation of (32) implies that
666
the element r is an odd one and, as the result, the lift 1 : C[[h ]] SAD preserves the
Grassmann parity.
Using two transverse polarizations on (M, ), associated to the left and right kernel
distributions of , one may introduce yet another graded structure on SA. In order to
present this gradation in explicit form, let us introduce a frame of (complex) 1-forms {e =
ei dx i , f = fi dx i }, , = 1, . . . , dim M/2, spanned the complexified cotangent bundle

over a contractible coordinate chart U . The 1-forms can be chosen to satisfy
I e = e ,
If = f ,
(68)
I being the integrable involution structure defined by (6). Then one can introduce the new
(polarized) basis in the space of Grassmann generating elements i :
= ei i ,
= fj j .
(69)
With respect to this basis the superalgebra is decomposed onto the direct sum of its
subspaces
SA =

k=
SA(k) ! a =
SA(k) ,

(70)
a1 ...m 1 ...n (x, y, h , dx) 1 m 1 n .
mn=k
The easiest way to see that the decomposition (70) does really define a Z-gradation with
respect to -product is to introduce an inner derivative of the form
a = 1 [N, a],
N
i h
1
N = ij i j .
2
(71)
are collected in the following

The main properties of the derivative N
Proposition 4.1. Let all notations be as above, then
(i)
(ii)
(iii)
(iv)
a = i I k a/ k ;
N
i
= , N
= ;
a = na, a SA(n) , in particular N
N
(0)
D = D N
r SA ;
N
1

1 .
N = N
The second assertion presents the intrinsic definition of the Z-gradation (70) without
resorting to the special local frame (68).
Proof. (i) Straightforward computation leads to

a = 1 [N, a] = i Iik a i h ij j k il kj li a.
N
k
i h
8
l k
The second term in (72) vanishes due to the identities ij j k il = ij kj li = 0.
(ii) This immediately follows from (i) and definition (69).
(72)
667
(iii) We have to show that DN = 0 or, more explicitly,

1
(73)
[r, N] = 0.
i h
The first two terms vanishes since preserves the symplectic structure and N does
not depend on ys. So, it remains to prove that N commutes with r or, in other words,
that r SA(0) . The element r is determined from Eq. (32), where R is given by (66).

Expending r with respect to the first degree, r =
i=3 ri , ri SAi,1 , and substituting to
the first equation (32) we get a recursive definition for r:
N N +
1
1
r3 = 1 R = Rij kl y i y j y k dx l + Rij kl i j y k dx l ,
8
4

k1
i
1
r3+k =
r2+k +
rj +2 rkj +2 , k = 1, 2, . . . .
h
(74)
(75)
j =1
commutes with and 1 and differentiates -product the assertion may be

Since N
proved by induction on the degree. The starting point of induction is

r3 = 1 N, 1 R = 1 Rij kl y k i j dx l = 0.
N
(76)
i h
2
The latter equality holds because Rij kl = Rj ikl .
(iv) Let a = 1 (a). We expand a and a with respect to the first degree:
a=
ak ,
a =
k=0
a k ,
ak = h k k ,
k C,
a k SAk,0 .
(77)
k=0
Given r, then the sequence {a k } is recursively determined via {ak } from Eq. (32) as follows:

k2
i
1
(a k1 ) +
a k = ak +
(78)
[rn+2 , a kn1 ] , k = 1, 2, . . . ,
h
n=1
to both
where of course an empty sum is assumed to be zero and a 0 = a0 . Applying N
= 0 we get a chain of equations determining the
sides of (78) and using the fact that Nr
a. But this means (iv).
lift of N
Now let us clarify the relationship between the algebra of quantum observables on FWmanifold (C (M)[[h ]], & ), introduced in Section 3, and its superextension (C[[h ]], ). In
order to avoid a confusion all the constructions related to the former algebra will be labelled
by prime. First of all, we note that the subspace C (M)[[h ]] C[[h ]] is not closed with
respect to -product. The reason is that even though elements a and b do not depend on
s there lifts 1 (a) and 1 (b), defined with respect to (67), are in general do and so
does their product. Thus, the natural embedding C (M)[[h ]] C[[h ]] does not induce a
homomorphism of algebras; instead, the following relation holds true:
a & b = (a b),
a, b C (M)[[h ]],
C (M)[[h]]
(79)
where : C[[h ]]
is the canonical projection defined by the rule:
a(x, y, h , ) = a(x, y, h , 0). This relation is the particular case of a more general.
668
Proposition 4.2. For a C (M)[[h ]] and for b C[[h ]] we have:

a & (b) = (a b),
(b) & a = (b a).
(80)
Before proving this proposition consider first the fibrewise analog of relations (80).
Proposition 4.3. Let : SA A be a canonical projection defined as a(x, y, , dx, h )
= a(x, y, 0, dx, h ). Consider the algebras SA and A as left (right) moduli over SA(0)
and A, respectively. Then defines a homomorphism of the moduli introduced, that is for
a SA(0) and b SA
(a b) = (a) (b),
(b a) = (b) (a).
Proof. In view of (70) we have the expansion a =
n=0 an ,
(81)
where
an = a1 ...n 1 ...n (x, y, h , dx) 1 n 1 n .

Now it suffices to prove that for n > 0, (an b) = (b an ) = 0. But this directly follows
from the identities
b = b ,
b = b,
b SA.
(82)
To prove Proposition 4.2, we need the following:

Lemma 4.1. 1 (a) = ( 1 )& a, a C[[h ]].
Proof. Denote as before a = 1 (a). The element a,
being expanding as in (77), is
recursively determined from the pair of coupled equations (75), (78). Note that r3 = r3&
and commutes with and 1 . Now apply the canonical projection to both sides
of Eqs. (75), (78). In doing so, we can use the identities (81) since r SA(0) in view of
Proposition 4.1. By induction we have

k1
i &
&
1
&
&
rj +2 rkj +2 ,
r3+k =
(83)
r2+k +
h
j =1

n2
i &
1
a n = an +
(84)
( a n1 ) +
[rk+2 , a n1k ] .
h
k=1
But this is precisely the recursive definition of ( 1 )& a. Thus the theorem is proved.
Now Proposition 4.2 follows from the sequence of equalities:

(a b) = 1 a 1 b = 1 a 1 b = 1 a 1 b
&
&
&
&

= 1 a 1 b = 1 a 1 b = a b
and analogous relations for the second equality of (80).
669
6. Discussion
The deformation quantization, as it was originally formulated, was aimed to provide
the rigor mathematical groundwork for what physicists called quantization. Nowadays
the methods of deformation quantization go far beyond the quantization problem by
itself and constitute an integral and the most elaborated part of more general concept
noncommutative geometry [36]. The fresh interest to the subject was provoked by the
resent developments in the M-theory [37,38], where a certain limit of nonperturbative
string dynamics was recognized as the noncommutative YangMills theory (NYM) [39].
By now, however, only the restricted class of such models has been intensively studied,
namely, the models formulated on linear symplectic space Rn or tori Tn endowed with
canonical Poisson brackets and Euclidean metric. Turning to the more general manifolds
(such as ALE and K3 spaces, CalabiYau orbifolds, higher genus Riemann surfaces, etc.,
which all of physical interest) one has to deal with the nonconstant symplectic and metric
structures. It is clear, that the naive covariantization of the flat NYM Lagrangian destroys
the associativity of star-multiplication and thus it may break the gauge invariance of the
theory. So, the problem is to unify consistently three different ingredients of the NYM
theory: the noncommutativity of the coordinates, governed by some Poisson structure; the
metric properties of the manifold and the tensor nature of the YM fields. This can be also
considered as a part of more general mathematical problem: given a Poisson manifold, how
to define the associative deformation of the corresponding tensor algebra and fundamental
tensor operations? In the special case of functions, forming the subalgebra in the full tensor
algebra, it is just the problem which is studied by the deformation quantization.
In the seminal paper [37] the Fedosov deformation quantization was suggested as a
possible tool for constructing the NYM on general symplectic manifolds. So far as we
know, only two attempts have been undertaken to proceed in this direction [40,41], and
both are not too successful. Without going into details we note that the action functional
for NYM proposed in [40] is not in fact gauge invariant (except the trivial flat case)
because the Fedosov covariant derivative has not been properly extended to the tensor
observables. In the work [41] a local frame of closed 1-forms is introduced to convert the
tensor observables into the scalar functions, which may be then quantized by the Fedosov
method. However, the technical restrictions imposed by the authors actually imply the
existence of global Darboux coordinates on the manifold that makes all the construction
trivial.
We hope that the super-FW-manifolds introduced in this work may provide a geometrical
background for constructing NYM theories in nontrivial symplectic manifolds. Indeed, the
natural bijection between differential forms and superfunctions,
a(x)i1 ...ik i1 ik a(x)i1 ...ik dx i1 dx ik ,
give rise to the associative deformation of the exterior algebra on the base FW-manifold.
The NYM fields are embedded in this algebra as 1-forms. The gauge transformations are
then associated with the internal automorphisms of the superalgebra,
a = [a, ].
670
Note, that subspace of 1-forms is not in general invariant under these transformations even
when the gauge parameter is restricted to be a function (0-form). Thus a hypothetical
NYM theory being consorted in this way would necessary incorporate the entire multiplet
of antisymmetric tensor fields. In the flat limit, however, the dynamics of NYM field
is expected to decouple from the dynamics of higher-rank forms. The main advantage
of this approach is that the symplectic and metric structures enter to the formalism on
equal footing from the very beginning, that ensures the general covariance of the theory in
question. Finally, the construction of action functional for NYM theory requires to define
the trace functional on the algebra of superobservables. This is still an open question and
we postpone it to a future work.
The covariant Wick-type symbol construction, given in this paper, may probably find an
application in the problems related to quantization of the nonlinear field theory models like
strings in the curved spaces. The naive canonical quantization, being based on the Weyltype symbol, is usually inadequate in this case (as is seen even from the simple case of the
bosonic string in the flat space), while the nontrivial operator formulation is known only
for special examples of nonlinear theories, like WZWN sigma models. The perturbative
quantization of the nonlinear theory might be sometimes possible by means of the
expansion over the linear background making use of the Wick symbol defined with respect
to the linear approximation (examples of the perturbative construction of the Wick symbols
see in Ref. [4244]). However, the expansion over the linear vacuum cannot always be an
efficient method, e.g., it would be a hopeless attempt to construct the string theory in the
AdS space taking it as a weak perturbation to the Minkowskian string vacuum. As one
can expect, the presented Wick symbol formalism may have a potential for constructing an
adequate deformation quantization procedures for the nonlinear field theoretical models.
However, the practice of the general method application should be developed in separate
works for the actual problems of this type. In particular, the physically relevant models
are usually subject to the phase space constraints, so the symbol construction should be
first adapted to the case of a constrained system. Second, it should be understood in the
nonlinear field theory how one can identify the symmetric tensor which defines the Wick
symbol in the phase space. We expect that in many of physically relevant cases, the second
problem can be reduced to the first one, because it frequently occurs that a nonlinear theory
can be equivalently represented as a linear one subject to constraints (e.g.: the AdS string
can be thought of as a string on the constrained hypersurface in the flat space with an
additional time dimension).
Acknowledgements
We are grateful to I. Gorbunov, M. Grigoriev, M. Henneaux, A. Karabegov, R. Marnelius, A. Nersessian and M. Vasiliev for valuable discussions. We also appreciate the
partial financial support from INTAS under the grant N 00-262 and Russian Ministry of
Education under the grant E-00-33-184. The work of SLL is partially supported by RFBR
under the grant N 99-01-00980. The work of VAD and AAS is partially supported by
671
RFBR grant N 00-02-17-956 and by RFBR Grants for Young Scientists N 01-02-06418
and N 01-02-06420.
References
[1] F.A. Berezin, Izv. Akad. Nauk 38 (1974) 1116;
F.A. Berezin, Commun. Math. Phys. 40 (1975) 153.
[2] F. Bayen, M. Flato, C. Fronsdal, A. Lichnerowicz, D. Sternheimer, Ann. Phys. (NY) 111 (1978)
61;
F. Bayen, M. Flato, C. Fronsdal, A. Lichnerowicz, D. Sternheimer, Ann. Phys. (NY) 110 (1978)
111.
[3] D. Sternheimer, in: Proc. of the 1998 Lodz Conference Particles, Fields and Gravitation,
math/9809056.
[4] M. DeWilde, P.B.A. Lecomte, Lett. Math. Phys. 7 (1983) 487.
[5] R. Nest, B. Tsygan, Adv. Math. 113 (1995) 151.
[6] M. Bertelson, M. Cahen, S. Gutt, Class. Quant. Grav. 14 (1997) A93A107.
[7] P. Deligne, Selecta Mathematica, New Series 1 (1995) 667.
[8] M. DeWilde, P.B.A. Lecomte, Note di Matematica (suppl.) 10 (1) (1992) 205.
[9] B.V. Fedosov, J. Diff. Geom. 40 (1994) 213.
[10] P. Xu, Commun. Math. Phys. 197 (1998) 167.
[11] M.A. Grigoriev, S.L. Lyakhovich, Fedosov deformation quantization as a BRST theory,
Commun. Math. Phys. 218 (2001) 437.
[12] V.F. Molchanov, Func. Anal. Appl. 14 (2) (1980) 73.
[13] M. Cahen, S. Gutt, J. Rawnsley, Trans. Am. Math. Soc. 337 (1993) 73.
[14] M. Bordemann, S. Waldmann, Lett. Math. Phys. 41 (1997) 243, q-alg/9605012.
[15] A.V. Karabegov, Commun. Math. Phys. 180 (1996) 745, hep-th/9508013.
[16] A.V. Karabegov, Pseudo-Khler quantization on flag manifolds, dg-ga/9709015.
[17] A.V. Karabegov, On Fedosovs approach to deformation quantization with separation of
variables, math/9903031.
[18] C. Moreno, Lett. Math. Phys. 11 (1986) 361.
[19] M.J. Pflaum, New York J. Math. 4 (1998) 97.
[20] N. Reshetikhin, L. Takhtajan, Deformation quantization of Khler manifolds,
math.QA/9907171.
[21] M. Bordemann, E. Meinrenken, M. Schlichenmaier, Commun. Math. Phys. 165 (1994) 281
296.
[22] F.A. Berezin, Izv. Akad. Nauk 37 (1972) 1134.
[23] N.J.M. Woodhouse, Geometric Quantization, Oxford University Press, Oxford, 1992.
[24] B.V. Fedosov, Deformation Quantization and Index Theory, Akademia Verlag, Berlin, 1996.
[25] M. Cahen, M. Flato, S. Gutt, D. Sternheimer, J. Geom. Phys. 2 (1985) 35.
[26] I. Gelfand, V. Retakh, M. Shubin, Fedosov manifolds, dg-ga/9707024.
[27] M. Bordemann, On the deformation quantization of super-Poisson brackets, q-alg/9605038.
[28] A. Newlander, L. Nirenberg, Ann. Math. 65 (1957) 3;
A. Newlander, L. Nirenberg, Ann. Math. 65 (1957) 391.
[29] M. Kontsevich, Deformation quantization of Poisson manifolds, q-alg/9709040.
[30] P. Bonneau, Fedosov star-products and 1-differiable deformations, math/9809032.
[31] S.S. Chern, Complex Manifolds without Potential Theory, 2nd edn., Universitext Springer,
Berlin, 1979.
[32] K. Kodaira, F. Hirzebruch, J. Math. Pures et App. 36 (1957) 201.
672
[33] M. Rothstein, The structure of supersymplectic supermanifolds, in: C. Bartocci, U. Bruzzo,

R. Cianci (Eds.), Differential Geometric Methods in Mathematical Physics, Lecture Note in
Physics, Vol. 375, Springer-Verlag, Berlin, 1991, p. 331.
[34] E. Gozzi, M. Reuter, Mod. Phys. Lett. A 8 (1993) 1433;
E. Gozzi, M. Reuter, Int. J. Mod. Phys. A 9 (1994) 2191.
[35] M. Reuter, Int. J. Mod. Phys. A 11 (1996) 1253.
[36] A. Connes, Noncommutative Geometry, Academic Press, 1994.
[37] A. Connes, M.R. Douglas, A. Schwarz, JHEP 9802 (1998) 003, hep-th/9711162.
[38] N. Seiberg, E. Witten, JHEP 9909 (1999) 032, hep-th/9908142.
[39] A. Connes, M. Rieffel, YangMills for noncommutative two-tori, in: Operator Algebras and
Mathematical Physics, Iowa City, Iowa, 1985, Contemp. Math. Oper. Alg. Math. Phys. 62
(1987) 237.
[40] H. Garcia-Compean, Nucl. Phys. B 541 (1999) 651, hep-th/9804188.
[41] T. Asakawa, I. Kishimoto, Nucl. Phys. B 591 (2000) 611, hep-th/0002138.
[42] I.L. Buchbinder, E.S. Fradkin, S.L. Lyakhovich, V.D. Pershin, Int. J. Mod. Phys. A 6 (1991)
1211.
[43] A.Yu. Kamenshchik, S.L. Lyakhovich, Nucl. Phys. B 495 (1997) 309, hep-th/9608130.
[44] I.Yu. Karataeva, S.L. Lyakhovich, Teor. Mat. Fiz. 105 (1995) 55, hep-th/9504051.

Primary currents and Riemannian geometry

in W algebras
G. Bandelloni a , S. Lazzarini b
a Department of Physics, Genoa University, Istituto Nazionale di Fisica Nucleare,
Sezione di Genova Via Dodecaneso 33, I-16146 Genova, Italy

b CPT-CNRS Luminy, Case 907, F-13288 Marseille Cedex, France
Abstract
It is proved that general consistency requirements of stability under complex analytic change
of charts show that primary currents in finite chiral W-algebras are described in terms of pure
gravitational variables. 2001 Elsevier Science B.V. All rights reserved.
PACS: 11.10; 11.25.Hf
Keywords: W-algebras; Symplectic geometry
1. Introduction
The description of a physical phenomenon is independent of any parametrization used
for its representation. Without this belief any physical law loses its credibility.
This degree of freedom helps us in many complex sets of circumstances where
symmetries or work simplification conditions are present. Mathematically speaking, any
switch of parametrization is realized by the change of charts procedure, that is a finite (or
infinitesimal) reparametrization procedure which modifies the frame leaving the physics
unchanged. This recipe defines so to speak a geometrical stability for all of kinds of
symmetries which are locally realized. This stability requirement must not be confused
with the algebra of stability for the redefinitions of the physical parameters. In this paper
we discuss the former aspect in the framework of the W algebras [1], realized over
Riemann surfaces [24]. The intricacies encountered are well known and can be found
in the literature [59], and the outstanding mathematical books [10] as well.
Even if the W algebras arise from many sources [1118], the proper definition of all
the W observables over the whole Riemann surface is not a simple affair. For sake of
E-mail addresses: beppe@genova.infn.it (G. Bandelloni), sel@cpt.univ-mrs.fr (S. Lazzarini).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 2 5 2 - 8
674
G. Bandelloni, S. Lazzarini / Nuclear Physics B 606 [PM] (2001) 673688
truth this problem occurs in the usual conformal models [19], but in the context of W
algebras it becomes relevant [2028]. The capital role payed by a W symmetry has been
pointed out in many physical fields, such as integrable models [29,30], string theory [31]
and supersymmetry [3235]. Many efforts have been made in that direction and important
results have been reached [3646]. The physical symmetry constituents are currents whose
O.P.E. (or different recipes), generate extensions of the conformal symmetry [4749]: for
this reason they are usually defined as primary fields. Their equations of motion define
the dynamics of the system and, since they are defined on a Riemann surface [5052], they
must be promoted to the status of geometrical objects. Since any physical investigation
cannot be separated from any geometrical consideration, the same kind of symmetry arises
in the context of writing down welldefined differential equations on Riemann surfaces [20
62].
The main difficulty lies in the definition of differential operators on Riemann surfaces
[5]; in particular, their equipment in a conformal covariant frame has been widely
discussed in the literature [3,10,45]. In particular, the more general question of classifying
all holomorphically covariant differentials operators amounts to the study of operators
which are covariant with respect to projective transformations. At the end, the most
general holomorphic covariant operators become a superposition of ordinary derivative
operators with functions coefficients, whose reparametrization properties ensure the full
covariantization of the operator. The latter (or some of their combinations), within a well
defined mathematical framework, become generators of W-algebras [7,8,2729].
A natural question which is now to be asked concerns the link between these geometrical
coefficients and the physical primary fields currents arising in W -symmetry. There are
other way of controlling the intricacies of covariance and associativity of the symmetry in
W geometry by the use of appropriate holomorphic bundles, such as the jet bundles [41],
the DrinfeldSokolov bundle [42,43], and the Toda bundle [44]. However, we believe that
W-algebras describe a local symmetry, for this reason we have recently proposed [6365]
an approach to W-algebras in a symplectic framework. First, the chiral W -algebra [69]
emerges from the conspiracy of the infinitesimal action of symplectomorphisms together
with suitable reparametrizations of the two-dimensional local complex (z, z ) coordinates
over a compact Riemann surface [63]. In this approach the complex structure plays, as
usual, an essential role [64], and the geometry of the outstanding of finite W-algebras can
be studied [65].
The W-algebras govern [63,64] a hierarchy of smooth changes of local complex
(r)
(r) (z, z )), r = 1, . . . , n from the (z, z ) background
coordinates (z, z ) (Z0 (z, z ), Z
0
(r) (z, z )) ones through a
local complex coordinates to a sequence of suitable (Z0(r) (z, z ), Z
0
symplectic scenario. We shall follow a B.R.S. approach in order to fully exploit the locality
of the theory. Accordingly, instead of considering nonlocal commutators (or Poisson
brackets) between some primary fields, we shall analyse the B.R.S. transformations of
the FadeevPopov ghosts.
The so-called chiral B.R.S. ghost [19], K(r) associated to the hierarchy of smooth
changes of complex coordinates can be decomposed into a sum of the other ghost fields
C (j ) (z, z ) pertaining to the infinite W-symmetry and whose coefficients are fixed by the
675
geometrical space content. In our scheme all the elements of the algebra acquire a well
defined geometrical meaning, and each of the ghosts C (r) (z, z ) behaves as a (r, 0)differential. So this approach includes together both a hierarchy of smooth changes of
complex coordinates and the W transformations. We have pointed out in [65] that there
are two different physical situations: the first one occurs when the hierarchy realizes a
physical symmetry, the second one when this symmetry is broken, but the associativity
requirements are preserved. In this set of circumstances arise the primary fields, and the
broken symmetry can be ruled by means of a constant FaddeevPopov field which,
notwithstanding the breaking terms, secures the validity of the Jacobi identity [66]. So
the whole symplectic space must be doubled in a 0 component (where the physics lives)
and a partner (which guarantees the algebra closure). The coordinate transformations
will be written as:
(r)
(r)
(r)
(r)
SW Z0 (z, z ) = K0 (z, z )Z0 (z, z ) + Z (z, z ),

SW Z(r) (z, z ) = K0(r) (z, z )Z(r)(z, z ) K(r) (z, z )Z(r) (z, z ),
with the holomorphic ghosts transformations:
(r)

(r)
SW K0 (z, z ) + K (z, z )
(r)

(r)
(r)
(r)
= K0 (z, z ) + K (z, z ) K0 (z, z ) + K (z, z ) ,
(1.1)
(1.2)
and SW = 1.
In our symplectic formalism all the quantities acquire a well defined geometric meaning.
In particular, the holomorphic ghosts give an origin to a decomposition which generate the
finite W-algebra generators C (j ) (z, z ) and X (j ) (z, z ), by setting
(r)
(r)
K0 (z, z ) + K (z, z )
=
r

(r)

(r)
(j 1),0 (z, z ) + (j 1), (z, z ) C (j ) (z, z ) + X (j ) (z, z ) ,
(1.3)
j =1
(r)
(r)
where the intertwining coefficients (j
1),0 (z, z ) and (j 1), (z, z ) can be recursively
expressed in terms of the coordinates as
a
mj
(p )

Z0 i (z, z ) i
(r)
(j 1),0 (z, z ) = j !

(r)
) i ai =j, i ai pi =r
i=1 ai ! Z0 (z, z

Z (pi ) (z, z ) ai
p1 >p2 >>pmj
mj
(r)
(j
) = j !
1), (z, z
(r)
i=1
ai ! Z0 (z, z )

a
(pi )
(r)
Z0 (z, z ) i Z (z, z )

2
i ai =j,i ai pi =r
ai ! Z0(r)(z, z )
p1 >p2 >>pmj
and in turn are completely fixed by the geometry and behave as true tensors.
(1.4)
676
At the end the symplectic B.R.S. transformations imply that the variations of the
C (l)(z, z ) ghost fields describe the chiral part of W-algebras [63,65]:
SW C (z, z ) =
(l)
l
sC (s) (z, z )C (ls+1)(z, z ) + X (l) (z, z ),
l = 1, . . . , n,
(1.5)
s=1
where the X (l) (z, z ) breaking terms represent, in the B.R.S. approach, the contribution to
the finite W-algebras coming from the O.P.E. between primary fields, and indeed break
the chiral representation invariance, but still preserve associativity [65]. In the finite W3
instance we have:
16
SW3 C (1)(z, z ) = C (1) (z, z )C (1)(z, z ) T (z, z )C (2)(z, z )C (2)(z, z )
3

2
+ C (2)(z, z ) 2 C (2) (z, z ) C (2)(z, z ) 3 C (2)(z, z ) ,
3
SW3 C (2)(z, z ) = C (1) (z, z )C (2)(z, z ) + 2C (2)(z, z )C (1)(z, z ).
(1.6)
In order to insure the nilpotency of the BRS operation, X (l) (z, z ) in Eq. (1.5) must
transform as:
SW X (l) (z, z ) =
l

sC (s)(z, z )X (lr+1)(z, z ) sX (s) (z, z )C (ls+1)(z, z ).
s=1
(1.7)
The full completion of the last equation (1.7) amounts to introducing together with the
W-variations all the set of the primary fields (including fields not appearing in X (l) (z, z ) )
which belong to the algebra. In the W3 case we obtain:
SW3 T (z, z ) = C(z, z )T (z, z ) + 2T (z, z )C(z, z ) W(z, z )C (2)(z, z )
2
C (2)(z, z )W(z, z ) + 3 C(z, z ),
3
(1.8)
which force to introduce a cubic differential W(z, z ) as a field not involved in X (1) (z, z );
the minimal nilpotency chain closes by:

SW3 W(z, z ) = C(z, z )W(z, z ) + 3C(z, z )W(z, z ) + 16T (z, z ) C (2)(z, z )T (z, z )

+ 5 C (2)(z, z ) + 2C (2)(z, z ) 3 T (z, z ) + 10T (z, z ) 3 C (2) (z, z )

+ 15T (z, z ) 2 C (2)(z, z ) + 9 2 T (z, z )C (2)(z, z ) .
(1.9)
The structure of Eqs. (1.5), (1.7) fixes a hierarchy such that the lowest orders are fixed by
the higher ones; so the highest breaking term X (n) (z, z ) is so to speak related to the lowest
order terms.
In the paper we aim to study the properties of the X (l) (z, z ) quantities under finite
and infinitesimal change of charts in order to guarantee a global definition of the finite
W-algebra over a Riemann surface. Our investigation will lead to the result that primary
fields are related to the function coefficients required to build up the conformal covariant
derivatives and as such they describe gravitational degrees of freedom. This will be shown
677
first in Section 2 through a finite change of charts. Anyhow this procedure does not fully
exploit the locality values. To utilize them we further use the infinitesimal local change
of charts in a B.R.S. way. The derived consistency requirements in this approach generate
a connection between functions containing primary fields and the coefficients necessary
to define the action of the holomorphic derivative in an intrinsic way. This link proves
the inverse of the result, previously cited, found by Refs. [7,8,2729], and it is obtained
within a fully general covariance calculations. As a concluding remark we stress that this
result greatly takes some benefits from the local approach to W algebras we have recently
provided in Refs. [6365].
2. The stability of W-algebra under holomorphic change of charts

The purpose of this section is to investigate the consequences of the background
holomorphic change of coordinates
(w, w
) (z, z ),
w = w(z)
(2.1)
for the two Eqs. (1.5), (1.7). To hold its validity over the whole surface, each member of
each equation must verify the same transformation law. It is obvious that changes of charts
and W-symmetry will commute. Since the fields C (l)(z, z ) behave as (r, 0)-differentials
) = (w
) C (l)(z, z ),
C (l) (w, w
l
(2.2)
each term of Eqs. (1.5) and (1.7) must transform in the same way.
Thanks to the ghost property, a simple calculation shows that the chiral summand
l
(s)
)C (ls+1)(z, z ) behaves under finite change of charts as a (l, 0)s=1 sC (z, z
differential, so that problems come from the presence of the X (l) (z, z ) breaking term which
by covariance must transform also as a (l, 0)-differential
) = (w
) X (l) (z, z ).
X (l) (w, w
l
(2.3)
According to the FaddeevPopov grading and for l = 1, . . . , n, the X (l) breaking term is
decomposed over the ghost monomials but C (1) as
X (l) (z, z ) =
n
1 r (p)
C (z, z ) s C (q) (z, z )(l)
),
p+qrsl (r, p|s, q|l)(z, z
2
r,s0 p,q=2
(2.4)
(l)
where the zero graded coefficient functions p+qrsl (r, p|s, q|l) is skewsymmetric
under the permutation of the pairs (r, p) (s, q) and contain primary fields.
2.1. Behaviour under finite holomorphic change of complex coordinates
As is well known troubles come from the derivative operators: in fact the rth derivative
of a pth order tensor is no more a tensor and transforms in a convoluted way. For this
678
reason we shall use normalized covariant Bol operators Lr [2,5]:

Lr (z, z ) =
r
(r)
aj (z, z ) rj ,
(r)
a0 (z, z ) = 1,
(r)
a1 (z, z ) = 0,
(2.5)
j =0
which under any change of holomorphic charts transform as

) = (w
)
Lr (w, w
1+r
2
Lr (z, z )(w
)
1r
2
(2.6)
1+r
Lr transforms functions of conformal weight 1r
2 into ones of weight 2 . In particular, the
) is r dimensional linear space which is stable under any holomorphic
kernel of Lr (w, w
change of charts (w, w
) (z, z ).
This covariance property enables a (nonunique) construction of the aj(r)(z, z ) coefficients
[8] within a coordinate description. How this construction can be carried out in our
coordinate scheme is discussed in Appendix A.
(2)
The explicit expressions in terms of the coordinates for the coefficients a2 (z, z ),
(3)
(3)
a2 (z, z ) and a3 (z, z ) (the latter will be useful to discuss the W3 -algebra) are:
2
1
ln Z(z, z ) ,
2

2 v(z, z )
1
(3)
a2 (z, z ) = 2 ln w(z, z ) ln w(z, z ) +
,
3
w(z, z )
3
1
2
v(z, z )
(3)
ln w(z, z ) +
ln w(z, z ),
a3 (z, z ) = 3 ln w(z, z ) +
3
27
3w(z, z )
(2)
a2 (z, z ) = 2 ln Z(z, z )
where we have set the determinants

Z(z, z ) Z (2)(z, z )

,
w(z, z ) = 2
Z(z, z ) 2 Z (2) (z, z )
2

Z(z, z ) 2 Z (2) (z, z )
.
v(z, z ) = 3
Z(z, z ) 3 Z (2) (z, z )
(2.7)
(2.8)
Turning back to the gluing problem of the breaking term X (l) , without any loss of
generality, (2.4) rewrites for l = 1, . . . , n
X (l) (z, z ) =
n
1
Lr C (p) (z, z )Ls C (q) (z, z )A(l)
p+qrsl (r, p|s, q|l)(z, z ),
2
r,s0 p,q=2
(2.9)
(l)
where Ap+qrsl (r, p|s, q|l) is of zero ghost grading, skewsymmetric under the permutation of the pairs (r, p) (s, q) and constructed over some primary fields.
It is easy to see that the derivative troubles can be overcome by inverting Eq. (2.5)
iteratively,
r =
r

j =0
(r)
bj (z, z )Lrj (z, z ),
(r)
b0 (z, z ) = 1,
(r)
b1 (z, z ) = 0,
(2.10)
679
and where each of the bj(r) for j 2 is a polynomial expression in the ak(s), namely for
j 2,
k1
Jk :=
J1 = 0,
jm ,
m=0
(r)
bj
r
(1)
i=1
rJ
i

j =1
=1
(rJ )
aj
j,Ji+1 .
(2.11)
Eq. (2.3) thus writes:

n

Lr (w, w
)C (p) (w, w
)Ls (w, w
)C (q)(w, w
)
r,s0 p,q=2
A(l)
)
p+qrsl (r, p|s, q|l)(w, w
= (w )
n

Lr (z, z )C (p)(z, z )Ls (z, z )C (q) (z, z )
r,s0 p,q=2
A(l)
p+qrsl (r, p|s, q|l)(z, z )
n

(w
)

(1+r)
2
Lr (z, z )(w
)
(1r)
2 +p

(1+s)
C (p) (z, z ) (w
) 2
r,s0 p,q=2
(1s)

(l)
+q (q)
Ls (z, z )(w
) 2
C (z, z ) Ap+qrsl (r, p|s, q|l)(w, w
).
(2.12)
Using successively (2.5), Leibniz rule and (2.10) one gets after some combinatoric
manipulations

1+r
1r
(w
) 2 Lr (z, z ) (w
)( 2 +p) C (p) (z, z )
=
r
(r, m, p)(z, z )Lm C (p) (z, z ),
(2.13)
m=0
(s)
with the polynomial in the ak

(r, m, p)(z, z ) =
rj
rm

j =0 k=m
r j
k
(w
)
1+r
2
rj k (w
)
(k)
aj(r)(z, z )bm
(z, z )
1r
2 +p
(2.14)
Inserting twice (2.13) into (2.12) we finally deduce a set of algebraic equations

(l)
Ap+qrsl (r, p|s, q|l)(z, z ) =
w
l (m, r, p)(z, z )(m
, s, q)(z, z )
mr m
s
).
A(l)
p+qmm
l (m, p|m , q|l)(w, w
(2.15)
In particular, this system gives for each value of p, q, r, s, l the gluing properties of the
680
function A(l)
p+qrsl (r, p|s, q|l). Restricting the system (2.15) to the scalar sector, that is
for p + q r s l = 0 and
(l)
(l)
)|r+spql=0
A0 (r, p|s, q|l)(z, z )|r+spql=0 = A0 (r, p|s, q|l)(w, w
(l)
the scalar component of Ap+qrsl (r, p|s, q|l) cancels out and the system reduces to an
homogeneous one containing Ar+smm
(m, p|m
, q|l), m + m
> r + s with therefore
negative lower indices. The finite degree n of the algebra guarantees a finite number of
equations l = 1, . . . , n, and the determinant of the system will be nonzero in general leading
(l)
to Ah 0 for h < 0 and l = 1, . . . , n. We reach the conclusion:
(l)
Theorem 1. The stability under change of charts requires that the only nonvanishing
A(l)
p+qrsl (r, p|s, q|l) have a positive lower indices content, p + q r s l 0.
Anyhow in the more general cases, the way out to get fully local relations in only one
argument, is to use the infinitesimal holomorphic change of coordinates.
2.2. Infinitesimal holomorphic change of complex coordinates
Under infinitesimal holomorphic change of coordinates z z "(z) and following
the notation used in [2] we have:
" C (m) (z, z ) = Xm (z)C (m)(z, z ),
Xm (z) = "(z) + m"(z).
(2.16)
At the infinitesimal level, (2.3) reduces to

" X (l) (z, z ) = Xl (z)X (l) (z, z ),
(2.17)
while the conformal covariant derivative transforms as [2,8]

" , Lr = X (1+r) Lr Lr X (1r) ,
2
(2.18)
and, in particular, the case r = 1 gives [" , ] = 0. Accordingly, the aj(k) coefficients satisfy

1 k+1
(k)
(k)
" aj (z, z ) = Xj (z)aj (z, z ) +
(j 1) j +1 "(z)
2 kj

j 1

k1 kl
kl
al(k)(z, z ) j l+1 "(z) (2.19)
j l+1
j l
2
l=2
(k)
for j 2. Note that (2.19) is linear in aj and depends on its first derivative only. Turning
the holomorphic (1, 0)-differential "(z) a FaddeevPopov ghost (and accordingly Xl as
well), so that "2 = 0, one has:
1
" "(z) = X1 "(z) = "(z)"(z),
2
and using (2.18) for r = 1, one finds

" , Xl (z) = "(z) , Xl (z)
so that the operator " Xl is nilpotent for any l in turn.
(2.20)
(2.21)
681
The independence of the transformations assure that the W-transformations anticommute with the " operator:
SW "(z) = 0,
{" , SW } = 0.
(2.22)
It could be noted that the underlying algebra of " is equivalent to the W-algebra where the
chiral ghosts C (l)(z, z ), l 2, are set to be zero which is nothing but the usual conformal
algebra found in [19]. In our approach the W-algebras come out from the action of smooth
cotangent diffeomorphisms on a chain of smooth changes of local complex coordinates
(r)
(r)(z, z ) complex coordinates have
on the Riemann surface. In turn, all the Z0 (z, z ), Z
0
well defined W-transformations, and the B.R.S. algebra is kept nilpotent by performing a
doubling in order to obey the Jacobi identities. This doubling trick amounts to studying
(r)
the true structure of the Z (z, z ) coordinates. The question will be partially solved (this
will be enough for the purpose) by looking at the properties of the infinitesimal change of
charts of Z0(r)(z, z ) and their B.R.S. W-transformations as well.
(r)
The Z0 complex coordinates are scalars under change of charts:
(r)
(r)
" Z0 (z, z ) = "(z)Z0 (z, z )
(2.23)
(r)
while in their W-transformations, see (1.1), the most general form of Z can been made
explicit in terms of the chiral ghosts [65],
SW Z0(r) (z, z ) = K0(r) (z, z )Z0(r) (z, z )
+
n
(r)
Ls (z, z )C (p)(z, z )Bps (p|s|r)(z, z ).
(2.24)
s0 p=2
(r)
The anticommutativity (2.22) between the two operations on Z0 and the use of (2.18)
yield
(r)

(r)
" Xps (z) Bps (p|s|r)(z, z ) = Nps (p|s|r|a, B)(z, z ),
(2.25)
(r)
where Nps is linear in the Bs and of grading one with respect to of the form
(r)
Nps
(p|s|r|a, B)(z, z )

(r)
k "(z)
Nm+1ks (p|s|k|m; a)(z, z )Bpm
(z, z ),
=
k2
(2.26)
ms+k+1
where Nm+1ks (p|s|k|m; a) is a known polynomial in the ak(s) coefficients,

Nm+1ks (p|s|k|m; a)
m+1sk
1 m
mj
(m+1kj )
=
+p
aj(m) bm+1kj s .
m+1kj
2
(2.27)
j =0
By nilpotency, one gets

(r)
(" Xps )Nps
(p|s|r|a, B) = 0.
(2.28)
682
Owing to (2.19) this expression is still algebraic in the coefficients ak(s) and their first order
z-derivatives. Decomposing over the monomials k " " leads to a homogeneous linear
system in the Bs that can be inverted. Hence expressing algebraically the Bs in terms of
(s)
(s)
the ak coefficients and their first order z-derivatives implies that Z is expressible by
(r)
means of both Z0 and C () and their z-derivatives, see the appendix.
Turning back to the study of the behaviour of A under an infinitesimal holomorphic
(l)
change of complex coordinates. Let us now derive the properties of Ap+qrsl (r, p|s, q|l)
from (2.9) and (2.17). After some calculation we get:

(l)
" Xp+qrsl (z) Ap+qrsl (r, p|s, q|l)(z, z )
(l)
= p+qrsl (r, p|s, q|l)(z, z ),
(2.29)
(l)
where p+qrsl (r, p|s, q|l) is the following linear expression in the A(l) for p, q 0,
p, q = 2, . . . , n, l = 1, . . . , n:
(l)
p+qrsl
(r, p|s, q|l)(z, z )

k "(z)
Nm+1kr (p|r|k|m; a)(z, z )
=
k2
mr+k+1
(l)
Ap+qmsl (m, p|s, q|l)(z, z )

Nm+1ks (p|s|k|m; a)(z, z )
ms+k+1

(l)
Ap+qrml (r, p|m, q|l)(z, z )
where N has already been defined in (2.27).

Once more by nilpotency, one has
2
(l)
" Xp+qrml (z) p+qrsl
(r, p|s, q|l)(z, z ) = 0.
(2.30)
(2.31)
Solving this condition insures the proper definiteness of the whole theory over a Riemann
surface. As before decomposing over the monomials k " " leads for each sector a
homogeneous linear system in the As in finite dimension thanks to Theorem 1 and the
finite degree of W-algebra that can be inverted.
This shows that trivial solutions A() of Eq. (2.31) are algebraic expressions in the
coefficients aj(l)(z, z ) and their first order z-derivatives. Therefore they are related to the
complex coordinates Z0(r) . This drastically changes the physical scenario, namely, if we
rely on the idea that primary fields describe a conformal matter we have to change our
mind and relate them to gravitational degrees of freedom.
(r)
The closure under of the B.R.S. W-transformations of the Z0 fields secures that all
of the necessary primary fields, not involved in the breaking terms X (l) , will be also
(l)
differential polynomials (with higher order z-derivatives) in the coefficients aj . They
(r)(z,z)
coefficients and their derivatives.

accordingly turn to be also functions of the Z0
Collecting together these results yields the Statement:
683
Statement 1. All primary fields involved in a given finite Wn -algebra are local functions
(r)
of the Z0 complex coordinates and their z-derivatives.
The latter completely modifies the physical point of view of this scenario.
(l)
The identifications of the old primary fields in terms of monomials of aj (z, z ) and their
derivatives was performed in Ref. [8]: our treatment gives the exactly inverse of the proof
given there: while in this reference it was proved that the aj(l) can be expressed in terms
of the generators w of a given W-algebra. Here, we state the converse, namely, that the
primary fields of this algebra can be expressed with the aj(l) and their z-derivatives. For the
W3 instance, the currents are related to the aj(l)(z, z ) coefficients from the solutions of the
equation [8]
L3 fi (z, z ) = 0, i = 1, 2, 3,

" i,j,k 2 fi (z, z )fj (z, z )fk (z, z ) = 0
(2.32)
i,j,k
by
a2(3)(z, z )
,
2

1 1 (3)
(3)
W(z, z ) =
a (z, z ) a3 (z, z ) .
8 2 2
T (z, z ) =
(2.33)
With the help of (2.7) the cubic differential W current reads

3
1 1 3
2
W(z, z ) =
ln w(z, z ) 2 ln w(z, z ) ln w(z, z ) ln w(z, z )
24 2
9

5v(z, z )
1 v(z, z )
ln w(z, z ) ,
+
(2.34)
16 w(z, z )
3w(z, z )
and is nothing but a Laguerre invariant while a2(3)(z, z ) and the combination 9(a2(3)(z, z )
3a3(3)(z, z )) are both the so-called Painlev invariants [67] as pointed out in [71].
By the way from Eqs. (2.32), (1.8), (1.9) is easy to derive [70]:
Sfi (z, z ) = C(z, z )fi (z, z ) fi (z, z )C(z, z )
+ 2C (2) (z, z ) 2 fi (z, z ) C (2) fi (z, z )
1
8
+ fi (z, z ) 2 C (2) (z, z ) + T (z, z )fi (z, z )C (2) (z, z ).
3
3
(2.35)
The coordinates are defined [68] as the scalars obtained from the previous solutions; for
example Z0 = f2 /f1 and Z0(2) = f3 /f1 . By quotients the variations of the former turn out
to be
SZ0 (z, z ) = CZ0(z, z ) + 2C (2)(z, z ) 2 Z0 (z, z ) C (2) (z, z )Z0 (z, z )
4
C (2) (z, z )Z0 (z, z ) ln w(z, z ),
3
(2.36)
684
SZ0(2) (z, z ) = C(z, z )Z0(2) (z, z ) + 2C (2) (z, z ) 2 Z0(2)(z, z ) C (2) (z, z )Z0(2)(z, z )
4
C (2) (z, z )Z0(2) (z, z ) ln w(z, z ),
(2.37)
3
where (z, z ) has been introduced in Eq. (2.8) .
3. Conclusions
Our results show how general principles may clarify the physical landscape, in the
sense that the quantum extension of models which satisfy W-symmetry generated by
primary fields, should help the quantum description of gravitational degrees of freedom.
This extension greatly complicates this research topic, but, on an other hand, is extremely
provocative.
Also the LaguerreForsyth construction offers the possibility of expressing primary
current fields directly in terms of matter fields considered as solutions of linear
differential equations with algebraic integrals of the type Lf = 0, where L is the most
general covariant operators. This type of equations has been previously obtained in a
systematic way [70] through vanishing curvature conditions (a useful trick already used
in [12,41]). There exist few Lagrangian models giving rise to field equations containing
Bols operators [41]. Furthermore, the intimate link between these solutions and a hierarchy
(r)
of complex coordinates Z0 coming from the symplectic scenario raises the question
whether which of either coordinates or matter fields are the most important at the physical
(r)
level. Expressing Z0 in terms of the solutions of Ln f = 0 as in (A.5) generates the
breaking from W to finite Wn -algebras where n is exactly the order of the generalized
Bol operator Ln and may give a partial answer to this question. Note also that the -trick
used in [65] for introducing the chiral breaking terms X seems to be in relation with the
restriction to ker Ln .
Appendix A
The aim of this appendix is to give a coordinate description of the coefficients
aj(s)(z, z ) s = 2, . . . , n: in terms of the coordinates Z0(r)(z, z ), r = 1, . . . , n. To do this
we use the Di FrancescoItzyksonZuber [8] within our construction [63].
Consider the space V 1s of the functions fi (z, z ) with holomorphic weight d = 1s
2
2
solutions of the equation:
Ls fi (z, z ) = 0,
such that:
(s1)

f1 (z, z )

..

.

f1 (z, z )

f (z, z )
1
i = 1, . . . , s
..
.

(s1)f(s) (z, z )

..

.
= 1.
f(s) (z, z )
f (z, z )
(s)
(A.1)
(A.2)
685
So for a whatever other function f (z, z ) V 1s the action Ls f (z, z ) can be defined by:
2
(s)
f (z, z )
(s1)

f (z, z )

..
Ls f (z, z ) =
.

f (z, z )

f (z, z )
(s) f1 (z, z )
(s1)f1 (z, z )
..
.
..
.
f1 (z, z )
f1 (z, z )

(s)f(s) (z, z )

(s1)
f(s) (z, z )

..
,
.

f(s) (z, z )
f (z, z )
(A.3)
(s)
(s)
) can be identified with minors
which is a quantity with weight d = 1s
2 . So the aj (z, z
of this determinant:
(s)

(s) f1 (z, z )
(s)f(s) (z, z )
f (z, z )

..
..
..
..

.

.
.
.
(j +1)

(j
+1)
(j
+1)

f (z, z )
f1 (z, z )
f(s) (z, z )

aj(s)(z, z ) = (j 1) f (z, z ) (j 1) f1 (z, z ) (j 1) f(s) (z, z ) .
(A.4)

.
.
.
.
..
..
..
..

f (z, z )
f(s) (z, z )
f1 (z, z )

f(s) (z, z )
f (z, z )
f1 (z, z )
From the fi (z, z ), i = 1, . . . , s functions we can define s 1 scalars:

Z (j 1) (z, z ) =
fj (z, z )
,
f1 (z, z )
j = 2, . . . , s,
(A.5)
which provide a local (s 1)-dimensional system of coordinates. Assuming that this

system coincides with the coordinates previously introduced we invert the procedure to
construct a system of fi (z, z ) functions.
First we define the quantity
(s1) (1)
Z (z, z )

..

(z, z ) =
.

(1)
Z (z, z )
is an object with weight
weight d = 1s
2 [68]
s(s1)
2 ,
..
.

(s1)Z (s1)(z, z )

..

.

Z (s1)(z, z )
(A.6)
so we can build the s linearly independent functions with
f1 (z, z ) = s (z, z ),
f2 (z, z ) = Z (1) (z, z ) s (z, z ),
..
.
1
fs (z, z ) = Z (s1)(z, z ) s (z, z ),

1
(A.7)
which provide a system of solutions for the Eq. (A.1), with the coefficients aj(s)(z, z ) given
in Eq. (A.4).
686
References
[1] A.B. Zamolodchikov, Infinite additional symmetries in two-dimensional conformal quantum
field theory, Theor. Math. Phys. 65 (1985) 1205.
[2] F. Gieres, Conformally covariant operators on Riemann surfaces (with applications to
conformal and integrable models), Int. J. Mod. Phys. A 8 (1993) 1.
[3] B. Gustafsson, J. Peetre, Nagoya Math. J. 116 (1989) 63.
[4] L. Bonora, M. Matone, KdV equation on Riemann surfaces, Nucl. Phys. B 327 (1989) 415.
[5] G. Bol, Abh. Math. Sem. Hamburger Uniiv. 16 (1949) 1.
[6] W. Scherer, Covariant differential operators for densities and the KdV equation as a flow on
diffeomorphism groups, Lett. Math. Phys. 17 (1989) 45.
[7] Q. Wang, P.K. Panigrahi, U. Sukhatme, W.Y. Keung, Covariant operators and higher spin
conformal algebras, Nucl. Phys. B 344 (1990) 196.
[8] P. Di Francesco, C. Itzykson, J.-B. Zuber, Classical W algebras, Commun. Math. Phys. 140
(1991) 543568;
M. Bauer, P. Di Francesco, C. Itzykson, J.-B. Zuber, Covariant differential equations and
singular vectors in Virasoro representations, Nucl. Phys. B 362 (1991) 515562.
[9] N.S. Hawley, M. Schiffer, Acta Math. 107 (1962) 175.
[10] R.C. Gunning, Math. Ann. 170 (1967) 67;
R.C. Gunning, Lectures on Riemann Surfaces, Princeton University Press, 1966.
[11] I.M. Krichever, S.P. Novikov, Funct. Anal. Appl. 21 (1987) 126;
I.M. Krichever, S.P. Novikov, Funct. Anal. Appl. 21 (1987) 294.
[12] A. Bilal, V.V. Fock, I.I. Kogan, On the origin of W algebras, Nucl. Phys. B 359 (1991) 635.
[13] V.G. Drinfeld, V.V. Sokolov, Lie algebras and equations of Korteweg de Vries type, J. Sov.
Math. 30 (1984) 1975.
[14] B.A. Kupershmidt, G. Wilson, Invent. Math. 62 (1981) 403.
[15] A.A. Kirillov, Funct. Anal. Appl. 15 (1981) 135.
[16] A. Jaffe, S. Klimek, A. Lesniewski, Representations of the Heisenberg algebra on a Riemann
surface, Commun. Math. Phys. 126 (1989) 421.
[17] R. Dick, KricheverNovikov-like bases on punctures Riemann surfaces, Lett. Math. Phys. 18
(1989) 255.
[18] M. Schlichenmaier, KricheverNovikov algebras for more than two points, Lett. Math. Phys. 19
(1990) 151.
[19] C. Becchi, On the covariant quantization of the free string: the conformal structure, Nucl. Phys.
B 304 (1988) 513.
[20] J. Gervais, A. Neveu, Dual string spectrum in Polyakovs quantization. 2. Mode separation,
Nucl. Phys. B 209 (1982) 125;
J. Gervais, Phys. Lett. B 160 (1985) 277.
[21] S.D. Mathur, S. Mukhi, A. Sen, Differential equations for correlators and characters in arbitrary
rational conformal field theories, Nucl. Phys. B 312 (1989) 15.
[22] D.P. Jatkar, SU(2) WZW theory and KdV equation, Phys. Lett. B 255 (1991) 61.
[23] C. Teleman, Comment. Math. Helv. 33 (1959) 206.
[24] V.E. Zakharov, B.A. Shabat, Funct. Anal. Appl. 8 (1974) 226.
[25] H.C. Morris, J. Math. Phys. 17 (1981) 1867.
[26] A.P. Fordy, J. Gibbons, Factorization of operators. 2, J. Math. Phys. 22 (1981) 1170.
[27] P. Mathieu, Extended classical conformal algebras and the second Hamiltonian structure of Lax
equations, Phys. Lett. B 208 (1988) 101.
[28] I. Bakas, Higher spin fields and the GelfandDickey algebra, Commun. Math. Phys. 123 (1989)
627.
[29] A. Das, S. Roy, Integrable models and spin algebras, Int. J. Mod. Phys. A 6 (1991) 1429.
687
[30] A. Das, S. Roy, Covariant Lax operators and KacMoody algebras, J. Math. Phys. 32 (1991)
869.
[31] D. Friedan, E. Martinec, S. Shenker, Conformal invariance, supersymmetry and string theory,
Nucl. Phys. B 271 (1986) 93.
[32] L. Crane, J.M. Rabin, Superriemann surfaces: uniformization and Teichmuller theory, Commun. Math. Phys. 113 (1988) 601.
[33] M. Batchelor, P. Bryant, Commun. Math. Phys. 114 (1988) 243.
[34] A.M. Baranov, A.S. Schwarz, Y.I. Manin, I.V. Frolov, A superanalog of the Selberg trace
formula and multiloop contributions for fermionic strings, Commun. Math. Phys. 111 (1987)
373.
[35] F. Delduc, F. Gieres, Beltrami differentials, conformal models and their supersymmetric
generalizations, Class. Quantum Grav. 7 (1990) 1907.
[36] A.O. Radul, JETP Lett. 50 (1998) 371.
[37] D.J. Smit, Commun. Math. Phys. 128 (1990) 137.
[38] K. Yamagishi, The Kp hierarchy and extended Virasoro algebras, Phys. Lett. B 205 (1988) 466.
[39] I.M. Gelfand, L.A. Dikii, Asymptotic behavior of the resolvent of SturmLiouville equations
and the algebra of the Korteweg de Vries equations, Russ. Math. Surveys 30 (1975) 77;
I.M. Gelfand, L.A. Dikii, Funct. Anal. Appl. 10 (1976) 4;
I.M. Gelfand, L.A. Dikii, Funct. Anal. Appl. 11 (1977) 93.
[40] B.L. Feigin, D.B. Fuchs, Funct. Anal. Appl. 17 (1982) 114.
[41] R. Zucchini, Light cone W (N) geometry and its symmetries and projective field theory, Class.
Quant. Grav. 10 (1993) 253278, hep-th/9205102.
[42] R. Zucchini, The DrinfeldSokolov holomorphic bundle and classical W algebras on Riemann
surfaces, J. Geom. Phys. 16 (1995) 237274, hep-th/9403036.
[43] R. Zucchini, A KricheverNovikov formulation of classical W algebras on Riemann surfaces,
Phys. Lett. B 323 (1994) 322329, hep-th/9310061.
[44] E. Aldrovandi, L. Bonora, Liouville and Toda field theories on Riemann surfaces, J. Geom.
Phys. 14 (1994) 65109, hep-th/9303064.
[45] M. Matone, Conformal field theories in higher genus, Ph.D. Thesis, RX-1274 SISSA,Trieste,
October 1989.
[46] M. Adler, On a trace functional for formal pseudo differential operators and the symplectic
structure of the Korteweg de Vries equation, Invent. Math. 50 (1979) 219.
[47] P. Bryant, Graded Riemann surfaces and KricheverNovikov algebras, Lett. Math. Phys. 19
(1990) 97.
[48] L. Bonora, M. Martellini, M. Rinaldi, J. Russo, Phys. Lett. B 206 (1988) 444.
[49] A.M. Polyakov, Phys. Lett. B 103 (1981) 211.
[50] F.A. Bais, P. Bouwknegt, M. Surridge, K. Schoutens, Extensions of the Virasoro algebra
constructed from KacMoody algebras using higher order Casimir invariants, Nucl. Phys. B 304
(1988) 348.
[51] F.A. Bais, P. Bouwknegt, M. Surridge, K. Schoutens, Coset construction for extended Virasoro
algebras, Nucl. Phys. B 304 (1988) 371.
[52] V.A. Fateev, A.B. Zamolodchikov, Conformal quantum field theory models in two dimensions
having Z(3) symmetry, Nucl. Phys. B 280 (1987) 644.
[53] J.L. Gervais, Infinite family of polynomial functions of the Virasoro generators with vanishing
Poisson brackets, Phys. Lett. B 160 (1985) 277.
[54] J.L. Gervais, A. Neveu, Dimension shifting operators and null states in two-dimensional
conformally invariant field theories, Nucl. Phys. B 264 (1986) 557.
[55] O. Babelon, Extended conformal algebra and YangBaxter equation, Phys. Lett. B 215 (1988)
523.
[56] I. Bakas, Conformal invariance, the KdV equation and coadjoint orbits of the Virasoro algebra,
Nucl. Phys. B 302 (1988) 189.
688
[57] I. Bakas, The Hamiltonian structure of the spin-4 operator algebra, Phys. Lett. B 213 (1988)
313.
[58] A.A. Belavin, Stud. Appl. Math. 19 (1989) 117.
[59] A. Bilal, J. Gervais, Systematic approach to conformal systems with extended Virasoro
symmetries, Phys. Lett. B 206 (1988) 412.
[60] A. Bilal, J. Gervais, Extended C = Infinity conformal systems from classical Toda field theories,
Nucl. Phys. B 314 (1989) 646.
[61] V.A. Fateev, S.L. Lukyakhov, The models of two-dimensional conformal quantum field theory
with Z(N) symmetry, Int. J. Mod. Phys. A 3 (1988) 507.
[62] S.L. Lykyanov, Funct. Anal. Appl. 22 (1988) 1.
[63] G. Bandelloni, S. Lazzarini, W algebras from symplectomorphisms, CPT-99-P-3923, J. Math.
Phys. 41 (2000) 22332250, hep-th/9912202.
[64] G. Bandelloni, S. Lazzarini, The role of complex structures in W symmetry, Nucl. Phys. B 577
(2000) 471499, hep-th/0003027.
[65] G. Bandelloni, S. Lazzarini, The geometry of w(3) algebra: a twofold way for the rebirth of a
symmetry, CPT-00-P-4078, Nucl. Phys. B 594 (2001) 477500, hep-th/0011208.
[66] G. Bandelloni, Gauge theories on loop groups, J. Math. Phys. 32 (1991) 15991606;
G. Bandelloni, Renormalization of gauge coset models, J. Math. Phys. 32 (1991) 34913499.
[67] A.R. Forsyth, Theory of Differential Equations, Vol. 4, Dover, 1959.
[68] L. Guieu, V.Yu. Ovsienko, Structure symplectiques sul les espaces de courbes projectives et
affines, J. Geom. Phys. 16 (1995) 120148.
[69] C.M. Hull, Geometry and W-gravity, Talk at Pathways to Fundamental Interactions, 16th John
Hopkins Workshop on Current Problems in Particle Theory, Goteborg, 1992, hep-th/9301074;
C.M. Hull, Classical and quantum W-gravity, hep-th/9201057;
C.M. Hull, W-geometry, Commun. Math. Phys. 156 (1973) 245275, hep-th/9211113.
[70] D. Garajeu, S. Lazzarini, R. Grimm, W gauge structure and their anomalies: an algebraic
approach, J. Math. Phys. 36 (1995) 70437072.
[71] S. Govindarajan, Higher dimensional uniformisation and W -geometry, Nucl. Phys. B 457
(1995) 357374, hep-th/9412078.

Discrete torsion and branes in M-theory

from mathematical viewpoint
Shigenori Seki
Graduate School of Human and Environmental Studies, Kyoto University, Kyoto 606-8501, Japan
Abstract
We study orbifold group actions on locally defined fields upon M-theory branes in a threeform C-fields background. We derive some constraints from the consistency of the orbifold group
actions. We show the possibility of the existence of M-theory discrete torsion for the fields on the
worldvolume and discuss its features. 2001 Elsevier Science B.V. All rights reserved.
1. Introduction
NSNS B-fields are studied in various contexts, particularly noncommutative geometry
and discrete torsion. The relation between the noncommutative geometry and the B-fields
has been mentioned in [1] recently. When non-trivial B-fields exist along a D-brane,
a gauge theory on the D-brane has noncommutativity. On the other hand discrete torsion
was originally pointed out in [2] and the orbifold group action on a worldvolume gauge
theory on D-branes at the orbifold singularity has been studied recently in [3,4] . In this
direction a lot of works [59] have been done, and this paper also has a great interest in
this subject.
Since orbifold theories have twisted sectors, constraints for modular invariance on a oneloop partition function of a closed string on an orbifold M/ , which is a quotient space
of a manifold M by an orbifold group , induce a degree of freedom, which is called
discrete torsion (g, h) U (1) for g, h [2]. Higher loop modular invariance requires
the following constraints
(g1 g2 , g3 ) = (g1 , g3 )(g2 , g3 ),
(g, h) = (h, g)1 ,
(g, g) = 1,
E-mail address: seki@phys.h.kyoto-u.ac.jp (S. Seki).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 2 4 5 - 0
(1.1)
690
S. Seki / Nuclear Physics B 606 [PM] (2001) 689698
where (g, h) is classified by a second group cohomology H 2 (, U (1)) .

Discrete torsion for open strings has been shown in [3,4]. For the orbifold M/ ,
supersymmetric YangMills fields in the worldvolume gauge theory on D-branes at the
orbifold singularities are projected by as
(g)1 (g) = r(g),
g ,
where (g) is a representation of in the gauge group and r(g) is a spacetime action
of . Then (g) is in a projective representation
(g) (h) = (g, h) (gh).
(1.2)
The moduli space of the gauge theory has the same structure as the expectations based on
[10] . More studies on non-abelian orbifolds with discrete torsion have been done in [5,6]
in terms of Schur multipliers.
String theories have physical and mathematical aspects. Mathematical understanding
of discrete torsion has been proposed in [1116], which define the B-fields on each local
patch in terms of gerbes and state that discrete torsion is the choice of orbifold group action
on the B-fields. Since there exist gauge symmetries, it is in fact not sufficient to define the
orbifold action only on the base space. It is necessary to choose the action on fields as well.
These remarks include more applications. For example, the choice of orbifold group action
on vector fields gives rise to degrees of freedom which are known as orbifold Wilson lines.
In Section 2 we briefly review these mathematical aspects of discrete torsion and branes
in the string theories. Section 3 is devoted for the calculations of orbifold group actions
on fields in M-theory. In Section 4 we present summaries and conclusions with some
discussions on problems left for further study.
2. Review of discrete torsion and branes in string theories

In this section we review the mathematical aspects of discrete torsion and branes
proposed in [1113]. When we define the NSNS B-fields, we need a two-form field B
on a patch U , a one-form p on U ( U U ) and a U (1)-valued function q on
U ( U U U ) which satisfy the following equations
B B = dp ,
p
+p
+p
= d log q ,
(q ) = 1.
(2.1)
(2.2)
(2.3)
From Eq. (2.3), in order to preserve Cech

cocycle q by an orbifold group action g ,
up to coboundaries, we require that the pullback of q becomes
g q = q g ,
g
(2.4)
where are some Cech

cochains for each g. From Eqs. (2.2) and (2.4), we obtain the
orbifold group action on p as
g p = p + d log + (g) (g) ,

g
(2.5)
691
for some one-forms (g) . Using Eqs. (2.1) and (2.5), the orbifold group action on B
becomes
g B = B + d(g) .
(2.6)
Expanding (g1 g2 ) q , 1 2 3 , (g1 g2 ) p and (g1 g2 ) B , respectively, in two different

ways, we obtain some constraints
g g g
(g1 g2 ) = (g2 ) + g2 (g1 ) d log hg1 ,g2 ,

g ,g 1
g g
g
g
,
1 2 = 2 g2 1 hg1 ,g2 h1 2
g ,g g g ,g g ,g g g ,g
h1 2 3 h2 3 = g3 h1 2 h1 2 3 ,
g ,g
where h1 2 are some Cech

cochains.

Orbifold Wilson surfaces exp( B) which appear in the one-loop partition function, are
the analogues of orbifold Wilson loops and give rise to phases. Now let us take two kinds
g,h
g,h
of definitions of the orbifold group actions. We describe h in one definition and h in
the other. In order to consider the difference between these two definitions we use
hg,h
.
h g,h
Then the phases lead to
g,h h,g 1
g,h =
(2.7)
g,h
g,h
Note that we assume the B-fields are completely trivial so that h and are globally
defined. Eq. (2.7) stands for the phases from the contribution of twisted sectors to the
partition function, in other words, it corresponds to discrete torsion introduced in [2]. In
fact the phases (2.7) satisfy the conditions (1.1) for the modular invariance.
Next we consider the orbifold group action on N coincident D-branes. There are U (N)
gauge fields, which come from the ChanPaton factors of open strings ending on the
D-branes. Since gauge transformations associate the gauge fields A with the B-fields, the
gauge fields also should be defined on local patches. From [17] we can link A to B by
using the following equations
1
1
d log g
= p I,
A g A g
g g g = q I,
(2.8)
(2.9)
where g is a N N matrix and I is a unit matrix. g is a transition function for the

gauge bundle A when the B-fields are completely trivial.
From Eqs. (2.4), (2.5), (2.8) and (2.9) the orbifold actions on the gauge fields become
1 g g 1
+ d
+ I (g) ,
g A = g A g
(2.10)

g
g 1
g
,
g g = g
(2.11)
for some N N matrices . Expanding (g1 g2 ) g in two different ways, we obtain a
constraint

g g
g2 1 2 = hg1 ,g2 g1 g2 .
(2.12)
g
692
Let us consider the completely trivial B-fields given by B = 0, p = 0 and q = 1.

Then it is meaningful that we set the bundle on the D-branes to be topologically trivial. We
can replace the locally defined gauge field A with a globally defined U (N) gauge field A
and set g = 1. If furthermore the gauge field A is constant, we can assume that g and
hg1 ,g2 are constant. From Eq. (2.12) we obtain
g g

1 2 = hg1 ,g2 g1 g2 .
g has a projective representation. And hg1 ,g2 is classified by H 2 (, U (1)). These results
are in good agreement with [3,4].
3. Discrete torsion and branes in M-theory

In M-theory there exist membranes and M5-branes, and the membranes are considered
as three-form C-fields in an eleven-dimensional supergravity. Since the C-fields compactified on S 1 lead to the B-fields in the string theories, membrane twisted sector would derive
discrete torsion in M-theory. In the string theories discrete torsion,
which is classified by

H 2 (, U (1)), appears as the phases derived from the term exp( B), while in M-theory the

term exp( C) leads to some phases. In [14] the phases have been calculated from the contribution of membrane twisted sectors on T 3 in terms of 2-gerbes and have been classified
by a third group cohomology H 3 (, U (1)).
In order to define the C-fields on each patch, we need a three-form C-field C on a
patch U , a two-form u on U , a one-form v on U and a function h on
U ( U U U U ). The forms and the function are related by the following
equations
C C = du ,
u
v
+u
+v
+u
= dv
(3.1)
,
(3.2)
= d log h ,
(3.3)
h = 1.
(3.4)
The actions of g for the C-fields described in [14] are

g C = C + d(2)(g) ,

g u
=u
=v
+ d (g)
(1)
(3.5)
+ (g) (g) ,
(2)
(2)
+ (g) + (g) + (g)

g g 1 g g 1

,
g h = h
g v
(1)
(1)
(1)
(3.6)
g
+ d log ,
(3.7)
(3.8)
cochains,
where, for each element g in the orbifold group , are some Cech
(1)
(2)
(g) are some local one-forms and (g) are some local two-forms. And we
obtain the constraints [14],
(2) (g1 g2 ) = (2) (g2 ) + g2 (2) (g1 ) + d(3) (g1 , g2 ) ,
(3.9)
693
(1) (g1 g2 ) = (1) (g2 ) + g2 (1) (g1 ) (3) (g1 , g2 )

g ,g
+ (3) (g1 , g2 ) d log 1 2 ,
(3.10)
(3) (g2 , g3 ) + (3) (g1 , g2 g3 ) = g3 (3) (g1 , g2 ) + (3) (g1 g2 , g3 )

+ d log g1 ,g2 ,g3 ,

g1 g2
g2
g1
g ,g
g ,g
g2
1 2 1 2 g1,g2 ,
=
g1 g2 ,g3 g1 ,g2 g1 ,g2 g3 g2 ,g3 g ,g ,g g1 ,g2 ,g3 1
g3
=
1 2 3
,
g ,g ,g g g g ,g ,g g ,g g ,g g ,g ,g g ,g ,g
1 2 3 4 1 2 3 4 = 1 2 3 4 2 3 4 g4 1 2 3 .
(3.11)
(3.12)
(3.13)
(3.14)
g ,g ,g
In the similar way to Section 2 the difference of orbifold group actions 1 1 3 =

g ,g ,g
g ,g ,g
1 2 3 / 1 2 3 leads to the membrane twisted sector phase on T 3 and is classified by
H 3 (, U (1)) for the completely trivial C-fields [14].
Now let us consider the orbifold group actions on fields in a worldvolume theory on
branes. When a membrane ends on the branes as a string, the end line is assumed as a
two-form field B in the worldvolume theory. Since the three-form field C is mapped to
C + dB by gauge transformations, on the analogy of the string theories the B-fields are
also to be defined on each local patch in terms of a two-form B , a one-form p and a
function q . From [17] we are able to associate the B-fields with the C-fields as
B B + dp = u ,
p
+p
+p
(q )(q )
+ d log q = v
(q )(q )
(3.15)
,
= h .
(3.16)
(3.17)
We calculate the orbifold group actions on the B-fields. Firstly we suppose the action on
q as
g g

g q = q g ,
g
so that q is preserved up to coboundaries. are some Cech

cochains for each g. Then
the pullback of the left-hand side of Eq. (3.17) becomes

g (q )(q )1 (q )(q )1 = (q )(q )1 (q )(q )1 = h ,
while from Eq. (3.8) the pullback of the right-hand side of Eq. (3.17) has additional
factors g . So we should instead define the orbifold action on q as
g g

g
g q = q g .
(3.18)
From Eqs. (3.7), (3.16) and (3.18) we obtain
g p = p d log + (1) (g) + (1) (g) (1) (g) ,
g
(3.19)
for some local one-forms (1) (g) . Using Eqs. (3.6), (3.15) and (3.19), we calculate the
orbifold group action on B as
g B = B d(1) (g) + (2) (g) .
g
and
(1) (g)
determine the structure of orbifold group action on the B-fields.
(3.20)
694
If a membrane extends in a two-dimensional subspace transverse to the branes and ends

on the branes as a point, we can consider the end point as a one-form field A in the
worldvolume gauge theory on the branes. From [17] we write down the relations between
the B-fields and the A-fields,
1
d log g = p ,
A g A g
g g g = q .
(3.21)
(3.22)
A and g are the data of A-fields described on local patches in the same way as Section 2.
We suppose the orbifold group action on g as
g 1
,
g g = g g
(3.23)
where g are some functions. Since from the left-hand side of Eq. (3.22) we obtain
g [g g g ] = g g g = q ,
we need additional factors in Eq. (3.23) in order for the above equation to be consistent
g
with Eq. (3.18). When is equal to one, we are able to define the orbifold action on
g as
g g 1
.
g g = g g
(3.24)
If (1) (g) vanishes, we can obtain the action of g on A as
1
+ d log g + (1)(g) .
g A = g A g
(3.25)
These two requirements,

= 1 and (1) (g) = 0, are realized, for example, when the
C-fields are topologically trivial.
Next we derive some constraints. We calculate (g1 g2 ) q for g1 , g2 in two
different ways and they become

g2

g1 g2 g1 g2 g1 g2 g1
g2 g1 q =
q g2
g2 g2 g2 ,

g g g g

g
g
g ,g g ,g
2
1
q g2
1 2 1 2 g1,g2 1 2 1 2 g1g2 .
(g1 g2 ) q =
Comparing these equations, we read the following constraint
g2 g1 g1 ,g2 g1 g2 g ,g 1 g1 ,g2
1 2
,
g2 =
(3.26)
g ,g
where 1 2 are some functions. We also compute the pullbacks of p by g1 g2 ,

g
g2 g1 p = p d log 2 + (1) (g2 ) + (1) (g2 ) (1) (g2 )

g
g2 d log 1 + g2 (1) (g1 ) + g2 (1) (g1 ) g2 (1) (g1 ) ,
(g1 g2 ) p = p d log 1 2 + (1) (g1 g2 ) + (1) (g1 g2 ) (1)(g1 g2 )

g ,g
g
g
= p d log g1 ,g2 + d log 1 2 d log 2 g2 d log 1
g g
+ (1) (g2 ) + g2 (1) (g1 ) (3) (g1 , g2 ) + (3)(g1 , g2 )

+ (1) (g1 g2 ) (1) (g1 g2 ) .
695
Since g2 (g1 p ) is equal to (g1 g2 ) p , we obtain the constraint

(1) (g2 ) + g2 (1) (g1 ) = (1) (g1 g2 ) (3) (g1 , g2 ) d log g1 ,g2 .
(3.27)
For g we calculate the following pullbacks as

g 1 g1 1
g g
g2 g1 g = g2 1 2 g2 g1 g2 g 2
g2
g1 ,g2 g1 g2 g ,g 1 g1 ,g2 g g g2 1 g1 1
1 2
g2 1 2 g
g2
=
,

g g
g g 1
(g1 g2 ) g = 1 2 g1 g2 g 1 2
.
g ,g
From these equations, when 1 2 becomes one, we can obtain

g g g ,g g g
g2 1 2 = 1 2 1 2 .
We compare
(g g )g
1 2 3
with
g (g g )
1 2 3 .
(3.28)
These two terms become
g
g
g g g ,g 1 g1 ,g2 1
= 3 g3 2 (g2 g3 ) 1 1 2 3
g3
g g ,g g ,g g1 g2 ,g3 1 g1 ,g2 1
g3
1 2 3 g3 1 2
,

g (g g )
g
g
g
g ,g g 1 g2 ,g3 1
1 2 3 = 3 g3 2 (g2 g3 ) 1 1 2 3
g ,g g g ,g g1 ,g2 g3 1 g2 ,g3 1
1 2 3 2 3
.
(g g2 )g3
From these equations and Eq. (3.13) we obtain the constraint

g g ,g g ,g

1 2 3 g3 1 2 = g1 ,g2 ,g3 g1 ,g2 g3 g2 ,g3 .
Note that we have required
with the constraint (3.12).
= 1 and
g ,g
1 2
(3.29)
= 1 and these conditions are consistent
4. Conclusions and discussion

A lot of works on orbifolds and discrete torsion have been done in the string theories,
but we do not know precisely these subjects in M-theory. So we considered orbifold and
discrete torsion in M-theory on the analogy of the string theories.
We used the results shown in [14], where the three-form C-fields with connections were
presented in terms of the three-forms C , the two-forms u , the one-forms v and the
functions h defined on local patches. In [14] the two-forms (2) (g) , the one-forms
g
cochains were also introduced as the structures describing

(1) (g) and the Cech
the actions of orbifold group on the C-fields. For the constraints we introduced
g ,g
g ,g ,g
(3) (g1 , g2 ) , 1 2 and 1 2 3 , and the difference of orbifold actions g1 ,g2 ,g3 was
classified by H 3 (, U (1)).
Firstly we studied the two-form fields B in the worldvolume gauge theory on the branes.
The two-form B-fields are linked to the three-form C-fields, because C are transformed
into C + dB by the gauge transformations. So we also described the B-fields and their
connections on local patches by the two-forms B , the one-forms p and the functions
696
q . We wrote down the relation between the C-fields and the B-fields in Eqs. (3.15),
(3.16) and (3.17). From these equations and the orbifold group actions (3.5), (3.6), (3.7)
and (3.8) on the C-fields we obtained the actions on the B-fields, which are presented as
g
Eqs. (3.18), (3.19) and (3.20), and then we introduced some Cech
cochains and some
local one-forms (1) (g) . These factors are the structures constructing the orbifold group
actions on the B-fields. Calculating the pullbacks for the actions of g1 g2 on q and p
in two different ways, we derived the constraints (3.26) and (3.27), especially in Eq. (3.27)
g ,g
g ,g
we added some functions 1 2 . In Eq. (3.28) we are able to find that 1 2 plays a role
similar to discrete torsion in the string theories.
Next we considered the one-form fields A. We also defined the A-fields on local patches
in terms of the one-forms A and the matrices g . On the analogy of the string theories we
described the relationships between the B-fields and the A-fields as Eqs. (3.21) and (3.22).
In order to define the orbifold group actions on A and g , we needed some conditions
g
for and (1)(g) , which are the data for the orbifold group actions on the C-fields.
g
The conditions are that becomes one and that (1) (g) vanishes. They are satisfied
when the C-fields are topologically trivial, that is, when h = 1 and v = 0. Then
we obtained the orbifold group actions (3.24) on g from Eqs. (3.18) and (3.22), where
g
we introduced some functions , and the actions (3.25) on A from Eqs. (3.19), (3.21)
and (3.24).
Let us consider the specialized situation so that the C-fields are completely trivial, that is,
g
C is constant, u and v vanish and h is equal to one. Since we can take = 1
g ,g
g ,g ,g
from Eq. (3.8), and 1 2 = 1 from Eq. (3.12), we replace 1 2 3 with a globally defined
constant g1 ,g2 ,g3 . We also require that the B-fields are completely trivial, in other words,
g
B becomes globally constant, p vanishes and q is equal to one. We can take = 1
g ,g
from Eq. (3.18) and 1 2 are replaced with globally defined g1 ,g2 from Eq. (3.26). And
we regard the A-fields as the topologically trivial fields. Since g becomes one, from
g
g
Eq. (3.24) we obtain = . After all Eq. (3.28) leads to
g h g,h gh
=
, g, h .
This equation implies that the representation of orbifold groups g is projective. This result
is similar to Eq. (1.2) shown in [3,4].
g g g
We were able to have more interesting features in M-theory. Calculating 1 2 3 in two
different ways, we obtained Eq. (3.29). In the situation mentioned above we can define
g ,g
g ,g ,g
1 2 and 1 2 3 as global constants, and from Eq. (3.29) we obtain
g g ,g g ,g

1 2 3 1 2 = g1 ,g2 ,g3 g1 ,g2 g3 g2 ,g3 , g1 , g2 , g3 .
We should recall that g1 ,g2 ,g3 (= g1 ,g2 ,g3 / g1 ,g2 ,g3 ) are classified by H 3 (, U (1)). So
we assume g1 ,g2 ,g3 as discrete torsion for the branes in M-theory.
In M-theory there exist membranes and M5-branes. Open membranes have end points
and end lines on the M5-branes, and in a gauge theory on the worldvolume of N coincident
M5-branes we assume the end lines and the end points as two-form tensor fields and
one-form gauge fields, respectively. So we considered the two-form B-fields and the oneform A-fields in the three-form C-fields background. When the transverse space for the
697
M5-branes is R C2 , the gauge theory on the worldvolume includes two complex scalars
and one real scalars as fluctuations of the M5-branes to the transverse directions. When
the M5-branes are located at the singularity of orbifold R C2 / , where C2 / is ALE
space, the gauge theory becomes a six-dimensional N = 1 supersymmetric YangMills
theory for large N [18,19]. Then a tensor multiplet, a hypermultiplet and a vector multiplet
consist of the two-form field and the real scalar field, of the complex scalars and of the
one-form field, respectively. The moduli space of the YangMills theory on the M5-branes
at the orbifold singularity with discrete torsion may have good correspondences [20] to the
geometric structure of orbifold on the analogy of [7,8].
Though we know that the end points of open strings have ChanPaton factors in the
string theories, the analogues of ChanPaton factors as the end lines of open membranes
are not clear in M-theory. So we do not know precisely what type of values g1 ,g2 and
g1 ,g2 ,g3 are. But at least we are able to suggest that the representation g of the orbifold
group is projective and that the phase g1 ,g2 has the structure which are determined by the
M-theory discrete torsion g1 ,g2 ,g3 .
We will need to make more precise mathematical analyses, for example, quotient stacks
[21,22] and K-theory. Dp-branes and M-theory branes are studied in the contexts of
K-theory [23] and twisted K-theory [24,25]. Since the p-branes are realized as (p + 1)form fields in low energy effective actions, there may exist some analogues of discrete
torsion for the p-branes and the open p-branes.
Acknowledgement
I am grateful to S. Matsuda for encouragement. This work is supported in part by JSPS
Research Fellowships for Young Scientists (#4783).
References
[1] N. Seiberg, E. Witten, String theory and noncommutative geometry, JHEP 9909 (1999) 032,
hep-th/9908142.
[2] C. Vafa, Modular invariance and discrete torsion on orbifolds, Nucl. Phys. B 273 (1986) 592.
[3] M. Douglas, D-branes and discrete torsion, hep-th/9807235.
[4] M. Douglas, B. Fiol, D-branes and discrete torsion II, hep-th/9903031.
[5] B. Feng, A. Hanany, Y. He, N. Prezas, Discrete torsion, non-abelian orbifolds and the Schur
multiplier, JHEP 0101 (2001) 033, hep-th/0010023.
[6] B. Feng, A. Hanany, Y. He, N. Prezas, Discrete torsion, covering groups and quiver diagrams,
hep-th/0011192.
[7] D. Berenstein, R. Leigh, Discrete torsion, AdS/CFT and duality, JHEP 0001 (2000) 038, hep-th/
0001055.
[8] D. Berenstein, R. Leigh, Non-commutative CalabiYau manifolds, Phys. Lett. B 499 (2001)
207, hep-th/0009209.
[9] K. Dasgupta, S. Hyun, K. Oh, R. Tatar, Conifolds with discrete torsion and noncommutativity,
JHEP 0009 (2000) 043, hep-th/0008091.
[10] C. Vafa, E. Witten, On orbifold with discrete torsion, J. Geom. Phys. 15 (1995) 189, hep-th/
9409188.
698
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
E. Sharpe, Discrete torsion and gerbes I, hep-th/9909108.

E. Sharpe, Discrete torsion and gerbes II, hep-th/9909120.
E. Sharpe, Discrete torsion, hep-th/0008154.
E. Sharpe, Analogues of discrete torsion for the M-theory three-form, hep-th/0008170.
E. Sharpe, Discrete torsion in perturbative heterotic string theory, hep-th/0008184.
E. Sharpe, Recent developments in discrete torsion, Phys. Lett. B 498 (2001) 104, hep-th/
0008191.
D. Freed, E. Witten, Anomalies in string theory with D-branes, hep-th/9907189.
S. Ferrara, A. Kehagias, H. Partouche, A. Zaffaroni, Membranes and fivebranes with lower
supersymmetry and their AdS supergravity duals, Phys. Lett. B 431 (1998) 42, hep-th/9803109.
C. Ahn, K. Oh, R. Tatar, Orbifolds AdS7 S 4 and six-dimensional (0, 1) SCFT, Phys.
Lett. B 442 (1998) 109, hep-th/9804093.
S. Seki, work in progress.
E. Sharpe, String orbifolds and quotient stacks, hep-th/0102211.
E. Sharpe, Quotient stacks and string orbifolds, hep-th/0103008.
E. Witten, D-branes and K-theory, JHEP 9812 (1998) 019, hep-th/9810188.
A. Kapustin, D-branes in a topologically nontrivial B-field, Adv. Theor. Math. Phys. 4 (2001)
127, hep-th/9909089.
D. Diaconescu, G. Moore, E. Witten, E8 gauge theory, and a derivation of K-theory from
M-theory, hep-th/0005090.

CUMULATIVE AUTHOR INDEX B601B606
Abel, S.
Ahn, C.
Akemann, G.
Alford, M.
Ali, D.B.
Aloisio, R.
Alonso-Alberca, N.
ALPHA Collaboration
lvarez, E.
Amoros, G.
Anastasiou, C.
Anastasiou, C.
Anastasiou, C.
Arkani-Hamed, N.
Arnowitt, R.
Arutyunov, G.
Astier, P.
Astier, P.
Autiero, D.
Autiero, D.
Azcoiti, V.
B606 (2001) 151

B601 (2001) 539
B601 (2001) 77
B602 (2001) 61
B605 (2001) 337
B606 (2001) 322
B602 (2001) 329
B603 (2001) 180
B603 (2001) 286
B602 (2001) 87
B601 (2001) 318
B601 (2001) 341
B605 (2001) 486
B605 (2001) 81
B606 (2001) 59
B602 (2001) 238
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B606 (2001) 322
Bagrov, V.G.
Bajnok, Z.
Baldisseri, A.
Baldisseri, A.
Baldo-Ceolin, M.
Baldo-Ceolin, M.
Bandelloni, G.
Banner, M.
Banner, M.
Bassompierre, G.
Bassompierre, G.
Behrndt, K.
Bellini, M.
Bellucci, S.
Benatti, F.
Beneke, M.
Benslama, K.
Besson, N.
Besson, N.
Bhaseen, M.J.
B605 (2001) 425

B601 (2001) 503
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B606 (2001) 673
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B601 (2001) 49
B604 (2001) 441
B606 (2001) 119
B602 (2001) 541
B606 (2001) 245
B601 (2001) 3
B601 (2001) 3
B605 (2001) 3
B604 (2001) 537
0550-3213/2001 Published by Elsevier Science B.V.

PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 2 9 0 - 5
Bialas, A.
Bialas, P.
Bijnens, J.
Bintruy, P.
Bird, I.
Bird, I.
Blumenfeld, B.
Blumenfeld, B.
Bobisut, F.
Bobisut, F.
Boer, D.
Bolz, M.
Bonini, M.
Bouchez, J.
Bouchez, J.
Boyd, S.
Boyd, S.
Brandenburg, A.
Braun, V.M.
Buchalla, G.
Buchmller, W.
Bueno, A.
Bueno, A.
Bunyatov, S.
Bunyatov, S.
Buras, A.J.
Burda, Z.
Bytsko, A.G.
B603 (2001) 218

B603 (2001) 369
B602 (2001) 87
B604 (2001) 32
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B603 (2001) 195
B606 (2001) 518
B606 (2001) 231
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B606 (2001) 518
B603 (2001) 69
B606 (2001) 245
B606 (2001) 518
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B605 (2001) 600
B602 (2001) 399
B604 (2001) 455
Cachazo, F.
Cai, R.-G.
Camilleri, L.
Camilleri, L.
Campbell, B.A.
Cardini, A.
Cardini, A.
Carter, B.
Castro-Alvaredo, O.A.
Cattaneo, P.W.
Cattaneo, P.W.
Cavasinni, V.
Cavasinni, V.
B603 (2001) 3
B606 (2001) 137
B601 (2001) 3
B605 (2001) 3
B606 (2001) 613
B601 (2001) 3
B605 (2001) 3
B606 (2001) 45
B604 (2001) 367
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
700
Cerdeo, D.G.
Cervera-Villanueva, A.
Cervera-Villanueva, A.
Chandrasekharan, S.
Chim, L.
Chukanov, A.
Cognola, G.
Colangelo, G.
Collazuol, G.
Collazuol, G.
Conforto, G.
Conforto, G.
Conta, C.
Conta, C.
Contalbrigo, M.
Contalbrigo, M.
Cousins, R.
Cousins, R.
Cox, J.
Cski, C.
Curio, G.
Cvetic, M.
Cvetic, M.
B603 (2001) 231

B601 (2001) 3
B605 (2001) 3
B602 (2001) 61
B601 (2001) 539
B605 (2001) 3
B602 (2001) 383
B603 (2001) 125
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B602 (2001) 61
B604 (2001) 312
B602 (2001) 172
B605 (2001) 141
B606 (2001) 18
Dalitz, R.H.
Dalmazi, D.
Damgaard, P.H.
Daniels, D.
Daniels, D.
Dasgupta, A.
Davydychev, A.I.
De Azcrraga, J.A.
De Foss, L.
Degaudenzi, H.
Degaudenzi, H.
Deger, N.S.
Del Prete, T.
Del Prete, T.
De Santo, A.
De Santo, A.
Desrosiers, P.
Di Carlo, G.
Diehl, M.
Dignan, T.
Dignan, T.
Di Lella, L.
Di Lella, L.
Do Couto e Silva, E.
Do Couto e Silva, E.
Dolgushev, V.A.
Domenech-Garret, J.L.
Dorca, M.
Dorey, P.
Dorey, P.
Duff, M.J.
Duff, M.J.
B606 (2001) 483

B601 (2001) 77
B601 (2001) 77
B601 (2001) 3
B605 (2001) 3
B606 (2001) 357
B605 (2001) 266
B604 (2001) 75
B603 (2001) 413
B601 (2001) 3
B605 (2001) 3
B604 (2001) 343
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B606 (2001) 547
B606 (2001) 322
B605 (2001) 647
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B606 (2001) 647
B601 (2001) 395
B605 (2001) 215
B603 (2001) 581
B603 (2001) 582
B605 (2001) 141
B605 (2001) 234
Dumarchez, J.
Dumarchez, J.
Dutta, B.
B601 (2001) 3
B605 (2001) 3
B606 (2001) 59
Ellis, M.
Ellis, M.
Emig, T.
Erlich, J.
Evslin, J.
B601 (2001) 3
B605 (2001) 3
B604 (2001) 479
B604 (2001) 312
B602 (2001) 486
Fabris, J.C.
Fazio, T.
Fazio, T.
Feldman, G.J.
Feldman, G.J.
Feldmann, Th.
Feng, J.L.
Ferrari, R.
Ferrari, R.
Ferrre, D.
Ferrre, D.
Flaminio, V.
Flaminio, V.
Floreanini, R.
Font, A.
Forte, S.
Fradkin, E.
Fraternali, M.
Fraternali, M.
Frau, M.
Fring, A.
Frizzo, A.
Frolov, S.
Fujikawa, K.
B602 (2001) 644

B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B605 (2001) 647
B602 (2001) 307
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B602 (2001) 541
B605 (2001) 319
B602 (2001) 585
B601 (2001) 591
B601 (2001) 3
B605 (2001) 3
B602 (2001) 39
B604 (2001) 367
B604 (2001) 92
B602 (2001) 238
B605 (2001) 365
Gabrielli, E.
Gaillard, J.-M.
Gaillard, J.-M.
Gaillard, M.K.
Galajinsky, A.
Galante, A.
Gangler, E.
Gangler, E.
Garbarino, G.
Garousi, M.R.
Garriga, J.
Gasser, J.
Gava, E.
Gehrmann, T.
Gehrmann, T.
Geiser, A.
Geiser, A.
Geppert, D.
Geppert, D.
Gherghetta, T.
Ghilencea, D.M.
B603 (2001) 231

B601 (2001) 3
B605 (2001) 3
B604 (2001) 32
B606 (2001) 119
B606 (2001) 322
B601 (2001) 3
B605 (2001) 3
B606 (2001) 483
B602 (2001) 527
B605 (2001) 192
B603 (2001) 125
B605 (2001) 17
B601 (2001) 248
B601 (2001) 287
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B602 (2001) 3
B606 (2001) 101

Gibbons, G.W.
Gibin, D.
Gibin, D.
Gilkey, P.B.
Gitman, D.M.
Glover, E.W.N.
Glover, E.W.N.
Glover, E.W.N.
Glover, E.W.N.
Gninenko, S.
Gninenko, S.
Godley, A.
Godley, A.
Gomes, J.F.
Gmez, C.
Gomez-Cadenas, J.-J.
Gomez-Cadenas, J.-J.
Gomis, J.
Gorbunov, D.S.
Gosset, J.
Gosset, J.
Gling, C.
Gling, C.
Gouanre, M.
Gouanre, M.
Gracey, J.A.
Grant, A.
Grant, A.
Grassi, P.A.
Graziani, G.
Graziani, G.
Greene, B.R.
Grillo, A.F.
Grojean, C.
Gubser, S.S.
Gueuvoghlanian, E.P.
Guglielmi, A.
Guglielmi, A.
Gukov, S.
B606 (2001) 18
B601 (2001) 3
B605 (2001) 3
B601 (2001) 125
B605 (2001) 425
B601 (2001) 318
B601 (2001) 341
B605 (2001) 467
B605 (2001) 486
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B606 (2001) 441
B603 (2001) 286
B601 (2001) 3
B605 (2001) 3
B606 (2001) 3
B602 (2001) 213
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B605 (2001) 337
B601 (2001) 3
B605 (2001) 3
B606 (2001) 380
B601 (2001) 3
B605 (2001) 3
B604 (2001) 181
B606 (2001) 322
B604 (2001) 312
B605 (2001) 395
B606 (2001) 441
B601 (2001) 3
B605 (2001) 3
B601 (2001) 49
Hagner, C.
Hagner, C.
Hjcek, P.
Hjcek, P.
Hall, L.
Hambye, T.
Hammou, A.B.
Hara, T.
Hautmann, F.
Hernndez, L.
Hernando, J.
Hernando, J.
Hikami, K.
Hubbard, D.
Hubbard, D.
Huber, S.J.
B601 (2001) 3
B605 (2001) 3
B603 (2001) 531
B603 (2001) 555
B605 (2001) 81
B602 (2001) 23
B605 (2001) 17
B602 (2001) 499
B604 (2001) 391
B603 (2001) 286
B601 (2001) 3
B605 (2001) 3
B604 (2001) 580
B601 (2001) 3
B605 (2001) 3
B606 (2001) 183
701
Huerta, M.
Hurst, P.
Hurst, P.
Hyett, N.
Hyett, N.
B601 (2001) 591

B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
Iacopini, E.
Iacopini, E.
Ilha, A.
Intriligator, K.
Ishibashi, M.
Iso, S.
Iucci, A.
Izquierdo, J.M.
B601 (2001) 3
B605 (2001) 3
B604 (2001) 426
B603 (2001) 3
B605 (2001) 365
B604 (2001) 121
B601 (2001) 607
B604 (2001) 75
Jger, S.
Jakob, R.
Joseph, C.
Joseph, C.
Juget, F.
Juget, F.
Jurco, B.
B605 (2001) 600

B605 (2001) 647
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B604 (2001) 148
Kalmykov, M.Yu.
Kamani, D.
Kaminsky, K.
Kardar, M.
Karsch, F.
Kaya, A.
Ketov, S.V.
Khalil, S.
Khalil, S.
Khodjamirian, A.
Kiefer, C.
Kiem, Y.
Kim, J.E.
Kim, Y.
Kimura, Y.
Kirsanov, M.
Kirsanov, M.
Kirsten, K.
Klein, M.
Klemm, D.
Klimov, O.
Klimov, O.
Klishevich, S.M.
Koerber, P.
Kokkonen, J.
Kokkonen, J.
Korchemsky, G.P.
Kovzelev, A.
Kovzelev, A.
Krasnoperov, A.
Krasnoperov, A.
Krause, A.
Kroll, P.
B605 (2001) 266

B601 (2001) 149
B606 (2001) 613
B604 (2001) 479
B605 (2001) 579
B604 (2001) 343
B604 (2001) 256
B603 (2001) 231
B606 (2001) 151
B605 (2001) 558
B603 (2001) 531
B601 (2001) 27
B602 (2001) 346
B602 (2001) 467
B604 (2001) 121
B601 (2001) 3
B605 (2001) 3
B601 (2001) 125
B605 (2001) 319
B601 (2001) 380
B601 (2001) 3
B605 (2001) 3
B606 (2001) 583
B603 (2001) 413
B601 (2001) 3
B605 (2001) 3
B603 (2001) 69
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B602 (2001) 172
B605 (2001) 647
702
Kustov, D.
Kuznetsov, V.
Kuznetsov, V.
B605 (2001)
B601 (2001)
B605 (2001)
3
3
3
Lacaprara, S.
Lacaprara, S.
Lachaud, C.
Lachaud, C.
Laermann, E.
Lakic, B.
Lakic, B.
Lanza, A.
Lanza, A.
Lapointe, L.
LaRotonda, L.
La Rotonda, L.
Laveder, M.
Laveder, M.
Lazaroiu, C.I.
Lazaroiu, C.I.
Lazaroiu, C.I.
Lazzarini, S.
Leo, C.R.
Lebedev, O.
Lee, H.M.
Letessier-Selvon, A.
Letessier-Selvon, A.
Leutwyler, H.
Levin, A.
Levy, J.-M.
Levy, J.-M.
Li, K.
Li, M.
Liccardo, A.
Ling, Y.
Linssen, L.
Linssen, L.
Lipan, O.
Liu, J.T.
Liu, J.T.
Liu, J.T.
Ljubicic, A.
Ljubicic, A.
Loginov, E.K.
Loll, R.
Long, H.N.
Long, J.
Long, J.
L, H.
L, H.
Lu, J.X.
Lupi, A.
Lupi, A.
Lyakhovich, S.L.
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B605 (2001) 579
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B606 (2001) 547
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B603 (2001) 497
B604 (2001) 181
B605 (2001) 159
B606 (2001) 673
B602 (2001) 514
B606 (2001) 151
B602 (2001) 346
B601 (2001) 3
B605 (2001) 3
B603 (2001) 125
B605 (2001) 425
B601 (2001) 3
B605 (2001) 3
B601 (2001) 607
B602 (2001) 201
B602 (2001) 39
B601 (2001) 191
B601 (2001) 3
B605 (2001) 3
B604 (2001) 603
B605 (2001) 116
B605 (2001) 141
B605 (2001) 234
B601 (2001) 3
B605 (2001) 3
B606 (2001) 636
B606 (2001) 357
B601 (2001) 361
B601 (2001) 3
B605 (2001) 3
B605 (2001) 141
B606 (2001) 18
B606 (2001) 137
B601 (2001) 3
B605 (2001) 3
B606 (2001) 647
Ma, E.
Ma, J.P.
B602 (2001) 23
B602 (2001) 572
Ma, J.P.
Macfarlane, A.J.
Magnea, L.
Manashov, A.N.
Mangano, M.L.
March-Russell, J.
Marchionni, A.
Marchionni, A.
Martelli, F.
Martelli, F.
Martn-Delgado, M.A.
Mathieu, P.
McInnes, B.
Mchain, X.
Mchain, X.
Meessen, P.
Meggiolaro, E.
Meggiolaro, E.
Mendiburu, J.-P.
Mendiburu, J.-P.
Merlatti, P.
Meyer, J.-P.
Meyer, J.-P.
Mezzetto, M.
Mezzetto, M.
Miao, Y.-G.
Mishra, S.R.
Mishra, S.R.
Molke, H.
Moon, S.-H.
Moorhead, G.F.
Moorhead, G.F.
Morales, J.F.
Moreau, G.
Morel, A.
Mueller, A.H.
Mller-Kirsten, H.J.W.
Munier, S.
Muoz, C.
Musto, R.
B605 (2001) 625

B604 (2001) 75
B604 (2001) 92
B603 (2001) 69
B602 (2001) 585
B602 (2001) 307
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B601 (2001) 569
B606 (2001) 547
B602 (2001) 132
B601 (2001) 3
B605 (2001) 3
B602 (2001) 329
B602 (2001) 261
B606 (2001) 337
B601 (2001) 3
B605 (2001) 3
B602 (2001) 453
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B606 (2001) 84
B601 (2001) 3
B605 (2001) 3
B603 (2001) 180
B602 (2001) 467
B601 (2001) 3
B605 (2001) 3
B605 (2001) 17
B604 (2001) 3
B603 (2001) 369
B603 (2001) 427
B606 (2001) 84
B603 (2001) 427
B603 (2001) 231
B602 (2001) 39
Nagao, T.
Nan, C.M.
Narain, K.S.
Naumov, D.
Naumov, D.
Navelet, H.
Ndlec, P.
Ndlec, P.
Nefedov, Yu.
Nefedov, Yu.
Nelson, B.D.
Neubert, M.
Nguyen-Mau, C.
Nguyen-Mau, C.
B602 (2001) 622

B601 (2001) 607
B605 (2001) 17
B601 (2001) 3
B605 (2001) 3
B603 (2001) 218
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B604 (2001) 32
B606 (2001) 245
B601 (2001) 3
B605 (2001) 3
703
Nielsen, H.B.
NOMAD Collaboration
NOMAD Collaboration
Nomura, Y.
Norrbin, E.
B604 (2001) 405

B601 (2001) 3
B605 (2001) 3
B605 (2001) 81
B603 (2001) 297
Popov, B.
Porrati, M.
Poulsen, C.
Poulsen, C.
Pujols, O.
B605 (2001) 3
B606 (2001) 380
B601 (2001) 3
B605 (2001) 3
B605 (2001) 192
Oleari, C.
Oleari, C.
Oleari, C.
Oleari, C.
Oller, J.A.
Orestano, D.
Orestano, D.
Orland, P.
Ortn, T.
B601 (2001) 318

B601 (2001) 341
B605 (2001) 467
B605 (2001) 486
B602 (2001) 641
B601 (2001) 3
B605 (2001) 3
B605 (2001) 64
B602 (2001) 329
Quevedo, F.
B605 (2001) 319
Palla, L.
Park, D.H.
Park, D.K.
Pastore, F.
Pastore, F.
Peak, L.S.
Peak, L.S.
Pearce, P.A.
Peikert, A.
Pelinson, A.M.
Pennacchio, E.
Pennacchio, E.
Perez, E.
Peschanski, R.
Pessard, H.
Pessard, H.
Petersson, B.
Petersson, B.
Petkou, A.C.
Petkou, A.C.
Petkova, V.B.
Petriello, F.J.
Petrov, K.
Petti, R.
Petti, R.
Placci, A.
Placci, A.
Plyushchay, M.S.
Polesello, G.
Polesello, G.
Polesello, G.
Policastro, G.
Pollmann, D.
Pollmann, D.
Polyarush, A.
Polyarush, A.
Pomarol, A.
Pope, C.N.
Pope, C.N.
Popov, B.
B601 (2001) 503

B601 (2001) 27
B606 (2001) 84
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B601 (2001) 539
B605 (2001) 579
B602 (2001) 644
B601 (2001) 3
B605 (2001) 3
B604 (2001) 3
B603 (2001) 218
B601 (2001) 3
B605 (2001) 3
B602 (2001) 399
B603 (2001) 369
B601 (2001) 380
B602 (2001) 238
B603 (2001) 449
B601 (2001) 169
B603 (2001) 369
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B606 (2001) 583
B601 (2001) 3
B604 (2001) 3
B605 (2001) 3
B606 (2001) 380
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B602 (2001) 3
B605 (2001) 141
B606 (2001) 18
B601 (2001) 3
Rathouit, P.
Ravindran, V.
Reisz, T.
Remiddi, E.
Remiddi, E.
Resco, P.
Rtey, A.
Rey, S.-J.
Riccioni, F.
Rico, J.
Rico, J.
Ridolfi, G.
Riemann, P.
Riva, V.
Rivelles, V.O.
Roda, C.
Roda, C.
Rodriguez-Laguna, J.
Romano, R.
Ross, G.G
Rubbia, A.
Rubbia, A.
Russo, J.G.
Russo, R.
B601 (2001) 3
B605 (2001) 517
B603 (2001) 369
B601 (2001) 248
B601 (2001) 287
B603 (2001) 286
B604 (2001) 281
B602 (2001) 467
B605 (2001) 245
B601 (2001) 3
B605 (2001) 3
B602 (2001) 585
B605 (2001) 3
B604 (2001) 511
B602 (2001) 514
B601 (2001) 3
B605 (2001) 3
B601 (2001) 569
B602 (2001) 541
B606 (2001) 101
B601 (2001) 3
B605 (2001) 3
B602 (2001) 109
B604 (2001) 92
Sabella, G.
Sabra, W.A.
Sachrajda, C.T.
Sahlmann, H.
Sakai, N.
Salvatore, F.
Salvatore, F.
Sanchis-Lozano, M.A.
Santoso, Y.
Sarkar, U.
Sati, H.
Sato, H.-T.
Schahmaneche, K.
Schahmaneche, K.
Schmidt, B.
Schmidt, B.
Schmidt, M.G.
Schmidt, T.
Schupp, P.
Seki, S.
Sekino, Y.
Sethi, S.
Sevior, M.
B602 (2001) 453

B605 (2001) 234
B606 (2001) 245
B606 (2001) 401
B602 (2001) 413
B601 (2001) 3
B605 (2001) 3
B601 (2001) 395
B606 (2001) 59
B602 (2001) 23
B605 (2001) 116
B601 (2001) 27
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B606 (2001) 183
B605 (2001) 3
B604 (2001) 148
B606 (2001) 689
B602 (2001) 147
B602 (2001) 307
B601 (2001) 3
704
Sevior, M.
Sevrin, A.
Sevrin, A.
Sezgin, E.
Shapiro, I.L.
Sharapov, A.A.
Shmakova, M.
Shore, G.M.
Sierra, G.
Sillou, D.
Sillou, D.
Siopsis, G.
Sjstrand, T.
Smith, D.
Smolin, L.
Smolin, L.
Soa, D.V.
Sokal, A.D.
Soler, F.J.P.
Soler, F.J.P.
Sondhi, S.L.
Sotkov, G.M.
Sozzi, G.
Sozzi, G.
Starinets, A.O.
Stasto, A.M.
Steele, D.
Steele, D.
Stelle, K.S.
Stiegler, U.
Stiegler, U.
Stipcevic, M.
Stipcevic, M.
Stolarczyk, Th.
Stolarczyk, Th.
Sundell, P.
B605 (2001) 3
B603 (2001) 389
B603 (2001) 413
B604 (2001) 343
B602 (2001) 644
B606 (2001) 647
B601 (2001) 49
B605 (2001) 455
B601 (2001) 569
B601 (2001) 3
B605 (2001) 3
B601 (2001) 380
B603 (2001) 297
B605 (2001) 81
B601 (2001) 191
B601 (2001) 209
B601 (2001) 361
B601 (2001) 425
B601 (2001) 3
B605 (2001) 3
B605 (2001) 395
B606 (2001) 441
B601 (2001) 3
B605 (2001) 3
B601 (2001) 425
B603 (2001) 427
B601 (2001) 3
B605 (2001) 3
B605 (2001) 141
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B604 (2001) 343
Tabaczek, J.
Takcs, G.
Takanishi, Y.
Takayanagi, T.
Talavera, P.
Tamaryan, S.
Tanaka, K.
Tanaka, T.
Tani, T.
Tanii, Y.
Tareb-Reyes, M.
Tareb-Reyes, M.
Tateo, R.
Tateo, R.
Taylor, G.N.
Taylor, G.N.
Tejeda-Yeomans, M.E.
B602 (2001) 399

B601 (2001) 503
B604 (2001) 405
B603 (2001) 259
B602 (2001) 87
B606 (2001) 84
B604 (2001) 121
B605 (2001) 192
B602 (2001) 434
B604 (2001) 343
B601 (2001) 3
B605 (2001) 3
B603 (2001) 581
B603 (2001) 582
B601 (2001) 3
B605 (2001) 3
B601 (2001) 318
B601 (2001) 341
B605 (2001) 467
Tereshchenko, V.
Tereshchenko, V.
Theis, U.
Thiemann, T.
Tlyachev, V.B.
Tomizawa, S.
Toropin, A.
Toropin, A.
Torrente-Lujan, E.
Toublan, D.
Touchard, A.-M.
Touchard, A.-M.
Tovey, S.N.
Tovey, S.N.
Tran, M.-T.
Tran, M.-T.
Tricarico, E.
Troost, J.
Troost, W.
Tsesmelis, E.
Tsesmelis, E.
B605 (2001) 486

B601 (2001) 3
B605 (2001) 3
B602 (2001) 367
B606 (2001) 401
B605 (2001) 425
B602 (2001) 413
B601 (2001) 3
B605 (2001) 3
B603 (2001) 231
B603 (2001) 343
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B606 (2001) 231
B603 (2001) 389
B603 (2001) 389
B601 (2001) 3
B605 (2001) 3
Ulrichs, J.
Ulrichs, J.
Urban, J.
Uzan, J.-P.
B601 (2001) 3
B605 (2001) 3
B605 (2001) 600
B606 (2001) 45
Vacavant, L.
Vacavant, L.
Vafa, C.
Valdata-Nappi, M.
Valdata-Nappi, M.
Valuev, V.
Valuev, V.
Van de Bruck, C.
Van Neerven, W.L.
Van Neerven, W.L.
Vannucci, F.
Vannucci, F.
Varadarajan, U.
Varvell, K.E.
Varvell, K.E.
Vassilevich, D.V.
Veltri, M.
Veltri, M.
Verbaarschot, J.J.M.
Verbaarschot, J.J.M.
Vercesi, V.
Vercesi, V.
Vermaseren, J.A.M.
Vidal-Sitjes, G.
Vidal-Sitjes, G.
Vieira, J.-M.
Vieira, J.-M.
Vinogradova, T.
B601 (2001) 3
B605 (2001) 3
B603 (2001) 3
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B605 (2001) 215
B603 (2001) 42
B605 (2001) 517
B601 (2001) 3
B605 (2001) 3
B602 (2001) 486
B601 (2001) 3
B605 (2001) 3
B601 (2001) 125
B601 (2001) 3
B605 (2001) 3
B601 (2001) 77
B603 (2001) 343
B601 (2001) 3
B605 (2001) 3
B604 (2001) 281
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
705
Vinogradova, T.
Vogt, A.
B605 (2001) 3
B603 (2001) 42
Wu, Y.-S.
Wu, Y.-S.
B604 (2001) 551

B606 (2001) 137
Wgner, F.
Wakatsuki, K.
Wang, J.E.
Weber, F.V.
Weber, F.V.
Weiner, N.
Weisse, T.
Weisse, T.
Wess, J.
Wetterich, C.
Wiese, U.-J.
Wilczek, F.
Wilson, F.F.
Wilson, F.F.
Winkler, O.
Winton, L.J.
Winton, L.J.
Wolff, U.
Wotzasek, C.
B601 (2001) 503

B604 (2001) 121
B602 (2001) 486
B601 (2001) 3
B605 (2001) 3
B605 (2001) 81
B601 (2001) 3
B605 (2001) 3
B604 (2001) 148
B606 (2001) 337
B602 (2001) 61
B602 (2001) 307
B601 (2001) 3
B605 (2001) 3
B606 (2001) 401
B601 (2001) 3
B605 (2001) 3
B603 (2001) 180
B604 (2001) 426
Xiong, Z.
B602 (2001) 289
Yabsley, B.D.
Yabsley, B.D.
Yang, H.-X.
Yang, J.M.
Yoneya, T.
Yu, Y.
B601 (2001) 3
B605 (2001) 3
B604 (2001) 551
B602 (2001) 289
B602 (2001) 499
B604 (2001) 551
Zaccone, H.
Zaccone, H.
Zemba, G.R.
Zerbini, S.
Zimerman, A.H.
Zuber, J.-B.
Zuber, K.
Zuber, K.
Zuccon, P.
Zuccon, P.
B601 (2001) 3
B605 (2001) 3
B601 (2001) 591
B602 (2001) 383
B606 (2001) 441
B603 (2001) 449
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3

Nucl - Phys.B v.606 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.606 PDF

Uploaded by

Copyright:

Available Formats

Nuclear Physics B 606 (2001) 317

D-branes, holonomy and M-theory

Received 18 May 2001; accepted 18 May 2001

J. Gomis / Nuclear Physics B 606 (2001) 317

G = Spin(7), 3-dim. N = 1 SUSY,

The number of unbroken supercharges is given by the number of singlets in the

J. Gomis / Nuclear Physics B 606 (2001) 317

J. Gomis / Nuclear Physics B 606 (2001) 317

2. The geometry of D6-branes and the O6-plane

2 is the metric of the Euclidean multi-centered Taub-NUT space [17]

which is a hyper-Kahler metric on a U (1) bundle over R3 . The x 11 coordinate isperiodic

J. Gomis / Nuclear Physics B 606 (2001) 317

which carries 2 units of D6-brane charge 5 (the O6 plane). N coincident D6-branes on

J. Gomis / Nuclear Physics B 606 (2001) 317

can be described by M-theory on a manifold with irreducible holonomy are severely

the transverse coordinate.

J. Gomis / Nuclear Physics B 606 (2001) 317

4. From SU(2) holonomy to SU(3) holonomy via D6-branes

J. Gomis / Nuclear Physics B 606 (2001) 317

and the following hypermultiplet content

J. Gomis / Nuclear Physics B 606 (2001) 317

5. From SU(3) holonomy to G2 holonomy via D6-branes

J. Gomis / Nuclear Physics B 606 (2001) 317

6. From SU(3) holonomy to SU(4) holonomy via D6-branes

|z1 |2 + |z2 |2 + |z3 |2 3|z4 |2 = r.

J. Gomis / Nuclear Physics B 606 (2001) 317

J. Gomis / Nuclear Physics B 606 (2001) 317

7. From G2 holonomy to Spin(7) holonomy via D6-branes

J. Gomis / Nuclear Physics B 606 (2001) 317

The eleven-dimensional description of this Type IIA background is somewhat subtle.

J. Gomis / Nuclear Physics B 606 (2001) 317

J. Gomis / Nuclear Physics B 606 (2001) 317

H. Niemer, Y. Oz, Supergravity and D-branes wrapping supersymmetric 3-cycles, hepth/0011288;

Nuclear Physics B 606 (2001) 1844

Supersymmetric non-singular fractional D2-branes

Cambridge CB3 OWA, UK

Received 6 April 2001; accepted 14 May 2001

M. Cvetic et al. / Nuclear Physics B 606 (2001) 1844

symmetry, an extensive programme of studying D3-branes in conifold-type backgrounds

M. Cvetic et al. / Nuclear Physics B 606 (2001) 1844

M. Cvetic et al. / Nuclear Physics B 606 (2001) 1844

in particular showing that it is supersymmetric, with 1/16 of the original supersymmetry

M. Cvetic et al. / Nuclear Physics B 606 (2001) 1844

and z is the eleventh coordinate. The 4-form in D = 11 is given by

where is a covariantly-constant spinor in the Ricci-flat 8-metric. Thus, in order to check

M. Cvetic et al. / Nuclear Physics B 606 (2001) 1844

M. Cvetic et al. / Nuclear Physics B 606 (2001) 1844

The field strengths J i dAi +

M. Cvetic et al. / Nuclear Physics B 606 (2001) 1844

G(3) = f1 dr X(2) + f2 dr J(2) + f3 X(3) ,

M. Cvetic et al. / Nuclear Physics B 606 (2001) 1844

The coordinate transformation x = r 4 puts the equations into the form

M. Cvetic et al. / Nuclear Physics B 606 (2001) 1844

Note that the functions ui satisfy u1 + 2u2 + 4u3 = 0.

M. Cvetic et al. / Nuclear Physics B 606 (2001) 1844

G156 = G134 = G264 = G235 = u3 .

(4) , obtained as described above, into (2.7), we find after

M. Cvetic et al. / Nuclear Physics B 606 (2001) 1844

M. Cvetic et al. / Nuclear Physics B 606 (2001) 1844

The metric represents a smoothing out of a cone over S 3 S 3 . However, it should be

G(3) = v1 e0 ei ei + v2 -ij k ei ej ek + v3 -ij k ei ej ek ,

It should be noted that these satisfy the relation

This harmonic 3-form is square-integrable at short distance, but it gives a linearly

and hence the unit normalisation condition g x x = 1

In order for the required transformations dt = t d + t d and dr = r d + r d to preserve

1. Type I scenarios, characterised by 1: in these scenarios the contribution e{4} F

2. Type II scenarios, characterised by 1: in these scenarios the contribution e{4} F