Nucl - Phys.B v.632

Nuclear Physics B 632 (2002) 350
www.elsevier.com/locate/npe
The supercharge and superconformal symmetry for

N = 1 supersymmetric quantum mechanics
T.E. Clark, S.T. Love, S.R. Nowling
a Department of Physics, Purdue University, West Lafayette, IN 47907-1396, USA
Received 5 September 2001; accepted 27 March 2002
Abstract
The superspace Lagrangian formulation of N = 1 supersymmetric quantum mechanics is
presented. The general Lagrangian constructed out of chiral and antichiral supercoordinates
containing up to two derivatives and with a canonically normalized kinetic energy term describes
the motion of a nonrelativistic spin 1/2 particle with Land g-factor 2 moving in two spatial
dimensions under the influence of a static but spatially dependent magnetic field. Noethers theorem
is derived for the general case and is used to construct superspace dependent charges whose lowest
components give the superconformal generators. The supercoordinates of charges containing an
R symmetry charge, the supersymmetry charges and the Hamiltonian are combined to form a
supercharge supercoordinate. Superconformal Ward identities for the quantum effective action are
derived from the conservation equations and the source of potential symmetry breaking terms are
identified. 2002 Elsevier Science B.V. All rights reserved.
PACS: 03.65.-w; 12.60.Jv
1. Introduction
Supersymmetric quantum mechanics [1] has proven to be a fruitful testing ground for
many of the intriguing consequences of the supersymmetry (SUSY) algebra which are
also anticipated to be applicable in supersymmetric quantum field theories. Conversely,
for those quantum mechanical models exhibiting a supersymmetry, the symmetry algebra
can be exploited to extract many of the models spectral properties [2]. For example, the
zero energy ground state wave functions for the 1 space dimension quantum mechanical
E-mail addresses: clark@physics.purdue.edu (T.E. Clark), love@physics.purdue.edu (S.T. Love),
nowling@uiuc.edu (S.R. Nowling).
0550-3213/02/$ see front matter 2002 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 2 4 5 - 6
T.E. Clark et al. / Nuclear Physics B 632 (2002) 350
systems whose dynamics is governed by the potential energy functions

1
1 dW (x)
V (x) = W 2 (x)
,
2
2 dx
where W (x) is an arbitrary real function (superpotential), are given by
0 (x) = Ne
x
x0
dx W (x )
(1.1)
(1.2)
where N is a normalization constant. This result can be derived using the underlying
N = 1 supersymmetry exhibited by models of this type. The fact that models with such
potential energy functions possess an underlying supersymmetry can be demonstrated in
a variety of ways. One approach which manifestly incorporates the N = 1 supersymmetry
from the outset, entails the construction of a superspace consisting of the ordinary time
coordinate, an anticommuting Grassmann variable and its conjugate. (See Appendix A for
a detailed construction of such a superspace and its properties.) The above class of models
then emerges from the supersymmetric invariant action

1
f () ,
0 = dV D D
(1.3)
2
where = is a real supercoordinate whose component is W (x) and after
application of the equations of motion dfdx(x) = W (x). The Hamiltonians for the two
partner potential energy functions take the factorized form
1 d2
1
1 dW (x)
+ W (x)2
2
dx 2 2
2

dx
1
1
d
d
W (x)
W (x) .
=
dx
dx
2
2
H =
(1.4)
The zero energy ground states satisfy

1
d
W (x) 0 (x) = 0

dx
2
whose solution is given above.
Besides using real supercoordinates, the N = 1 supersymmetry can also be realized
in terms of chiral and antichiral supercoordinates [35]. The purpose of this paper
is to study the consequences of such realizations. In Section 2, we present the most
general supersymmetric action constructed out of such coordinates containing up to two
derivatives. For a canonically normalized kinetic term, the model reduces to that describing
a nonrelativistic spin 1/2 particle with Land g-factor 2 moving in 2 spatial dimensions
under the influence of a static, but spatially dependent magnetic field.
The remainder of the paper is devoted to studying the spacetime symmetry structure
of the general N = 1 supersymmetric action constructed out of the chiral and antichiral
supercoordinates. The possible symmetries of the model are determined and the associated
charges are constructed. The symmetry transformations are applied to the Greens functions and their (non)conservation laws are expressed as Ward identities for the quantum
effective action. The time translation and SUSY transformations of the supercoordinates
are represented as superspace differential operators. The superspace action is shown to

be invariant under these variations. The N = 1 supersymmetry algebra and the construction of its representation on supercoordinates is presented in detail in Appendix A. There
the methodology for constructing SUSY invariant actions is prescribed. The derivation of
the superspace Noethers theorem for the construction of charges generating variations of
the coordinates is presented in Section 3. The most general spacetime symmetries of the
model are found to be those of the OSp(2, 1) superconformal group along with an additional U (1) internal symmetry phase transformation. The Noether charge for each symmetry is constructed and the conservation laws derived. In general, the U (1) and (super-)
conformal symmetries are not conserved and the symmetry violating variation of the Lagrangian is presented. The charge (non)conservation laws are used to derive the Ward identities obeyed by the generating functional for one-particle irreducible functions. The superconformal algebra is presented in Appendix B along with the representation of the symmetry transformations as superspace differential operator variations of the supercoordinates.
The charges are shown to form the components of superspace multiplets which transform
into each other under a restricted set of SUSY transformations. The multiplets are known
as quasi-supercoordinates. Appendix C recalls the quantum action principle in superspace
that underlies the derivation of the various Ward identities for the effective action, while
in Appendix D, the spinning coordinate path integral is explicitly performed by evaluating
the Grassmann determinant. So doing, the model is recast in terms of a matrix Hamiltonian
which depends only on the particle ordinary spatial coordinates. Finally in Section 4, the
multiplet that contains as its components a R-symmetry charge, the SUSY charges and the
Hamiltonian is shown to form a supercoordinate referred to as the supercharge. All of the
superconformal charges are constructed from these charges along with terms that are related to the global Weyl scaling charge for the chiral and antichiral supercoordinates. These
are just the chiral, antichiral and total number operators. The Ward identity operators for
the superconformal and U (1) symmetries also form (quasi-)supercoordinates. The symmetry violating terms are determined for all the symmetry transformations. Only the trivial
theory is both conformal and conformal SUSY invariant or simultaneously U (1) and scale
invariant. The construction of more general models that are superconformally invariant has
been delineated in [6]. The consequences of the superconformal symmetry in the operator
formalism has been studied in [7,8]. The supercharge techniques developed in the context
of the model presented here are applicable to these models but beyond the immediate scope
of the present work.
2. N = 1 supersymmetry and (anti)chiral supercoordinates

Manifestly supersymmetric quantum mechanical models are most conveniently formulated directly in superspace. The construction of this superspace can be achieved by augmenting the ordinary time variable by the introduction of an additional single complex
Grassmann variable , satisfying 2 = 0, along with its complex conjugate , also satisfy corresponds to a point in superspace. The detailed construcing 2 = 0. Together (t, , )
tion of this superspace along with the definition of SUSY covariant derivatives as well as
real, chiral and antichiral supercoordinates are presented in Appendix A. In the following,
we recall some of the salient features.
, ), have
A chiral supercoordinate (t, , ) and an antichiral supercoordinate (t,
the component expansions

(t, , ) = ei t z(t) + i 2 (t) = z(t) + i 2 (t) i z (t),

t
, ) = ei
(t,
(2.1)
z (t) i 2 (t) = z (t) i 2 (t) + i z(t).
Here the components z(t) and its complex conjugate z (t) are ordinary coordinates, while
(t), (t) are Grassmann valued (spinning) coordinates. (Throughout the paper, a time
d
derivative will be denoted by z (t) = t z(t) = dt
z(t).)
The most general SUSY invariant action through 2 derivatives may then be written in
terms of the (anti)chiral supercoordinates as

,
+ g(, )D
0 = dV L = dV K(, )
(2.2)
D
and real supercoordinate
where the real Khler prepotential supercoordinate K = K(, )
g(, ) are, in general, independent functions of the chiral and antichiral supercoordinates,
but not there SUSY covariant derivatives. The nilpotent SUSY covariant
and ,
derivatives are defined as
D=
D2 = 0
and
= + i
D
t

2

=0 .
D
= 2i . PerformThey anticommute to give the (SUSY covariant) time derivative {D, D}

t
ing the component decomposition yields the action

0 = dtL = dt 4g(z, z )zz + i Kz (z, z )z Kz (z, z )z + 2ig(z, z )

(2.3)
Kzz (z, z )[, ] i gz (z, z )z gz (z, z )z [, ] .
Throughout the paper, subscripts indicate differentiation with respect to that variable so
2K
that, for example, Kz = K
z , Kzz = z z .
We first examine the above action when the kinetic energy term is canonically
normalized which corresponds to taking g(z, z ) = 1/4 so the Lagrangian takes the form
i
+ i(Kz z Kz z) Kzz [, ].
2
Under the supersymmetry transformations of the coordinates given by
i[Q, z] = Q z = i 2 ,
i{Q, } = Q = 0,
i{Q, } = Q = 2 z ,
i[Q, z ] = Q z = 0,
L = z z +
(2.4)
(2.5)

z] = Q z = 0,
i[Q,

z ] = Q
i[Q,
z = i 2 ,

} = Q
i{Q,
= 2 z ,

} = Q
i{Q,
= 0,
(2.6)
the Lagrangian transforms as a total time derivative

i
i
d
d

z 2 Kz .
QL =
Q L =
(2.7)
z 2 Kz ,
dt
dt
2
2

Q
Consequently, the action, 0 = dt L, is
Q 0 .
SUSY invariant: 0 = 0 =
Identifying the components z(t) = m2 (x(t) + iy(t)) and z (t) = m2 (x(t) iy(t)),
where x, y are the coordinates of 2-dimensional space, and rescaling the Khler
prepotential as K qK, where q is the particle electric charge, this Lagrangian becomes

i
q
1
v 2 + q A v + +
B(x, y)[, ],
L = m
(2.8)
2
2
2m
where v = e1 x + e2 y is the particle velocity. In fact, this Lagrangian is recognized as one
describing the planar motion of a nonrelativistic spin 1/2 particle with Land g-factor
2 moving under the action of a static, but spatially dependent magnetic field oriented
perpendicular to the plane of motion: B = e3 B(x, y), with
B(x, y) = &ij i Aj (x, y) = 2 K(x, y),
(2.9)
and Ai (x, y) the components of the vector potential. Thus the general supersymmetric
quantum mechanical model in 2 spatial dimensions containing up to 2 derivatives and
with canonically normalized kinetic energy term is simply that describing the motion of a
nonrelativistic spin 1/2 particle with g-factor 2 moving under the influence of a static, but
spatially dependent magnetic field.
The more familiar form for the corresponding Hamiltonian expressed solely in terms
of the coordinates x, y and their canonical momenta can be gleaned by integrating out
the spinning variables as is explicitly performed in Appendix D. Alternatively, one could
employ a specific representation using the Pauli matrices for the spinning coordinates:
= + and = so that {, } = 1 and [, ] = 3 . Here

0 1
0 0
1 0
+ =
(2.10)
,
=
,
3 =
0 0
1 0
0 1
are the raising, lowering and third component Pauli matrices, respectively. In either case,
one secures the Lagrangian
q
1
v 2 + q A v + B(x, y)S,
L = m
2
m
(2.11)
where S = 12 3 is the spin operator. The corresponding Hamiltonian is then simply

y))2
q
(p q A(x,
B(x, y)S.
(2.12)
2m
m
Note that the Land g-factor is 2. As noted previously [912], this Hamiltonian is
supersymmetric.
H=
= (Q) , can be
The complex, nilpotent N = 1 supersymmetry charges, Q and Q
constructed as
i
Q = + ,
m
= i
Q
,
m
(2.13)
where
(px qAx ) i(py qAy ),
(px qAx ) + i(py qAy ),
(2.14)
have commutator
[,
] = 2qB(x, y).
(2.15)
nilpotent charges are seen to obey the supersymmetry algebra

These Q, Q

Q
,
{Q, Q} = 0 = Q,

H .
= 2H,
Q, Q
[Q, H ] = 0 = Q,
(2.16)
The Hamiltonians for the two different spin projections form partners of this supersymmetric quantum mechanical system. The supersymmetry algebra was exploited [9] to explicitly construct the exact zero energy ground state wave function for the system. Modulo this
ground state, the remainder of the eigenstates and eigenvalues of the two partner Hamiltonians form positive energy degenerate pairs [9].
For any supercoordinate (t, , ), the supersymmetry transformations can be represented by superspace Grassmann derivatives (see Appendix A) as

Q (t, , ) =
(t, , ),
+ i
t

Q (t, , ) =
(2.17)
i
(t, , ).
t

The SUSY variations anticommute to yield a time derivative: { Q , Q } = 2it . Since

time translations are generated by time derivatives, this gives a representation of the SUSY
algebra (2.16)
Q Q

Q Q

, = 0 = Q, Q ,
, = 2i H ,
Q H
, = 0 = Q, H ,
(2.18)
= t (t, , ).
where H (t, , )
In addition to time translation and SUSY invariance, the action of Eq. (2.3) also exhibits
a symmetry under a U (1) phase transformation of the (spinning) Grassmann coordinates
given by = ei and = ei , where is a real parameter, while the coordinates z
and z are left invariant. Moreover, if g and K are functions only of the combination ,
then the action is also invariant under rotations in the x y plane. This U (1) phase
Note that, in this
transformation takes the form = ei and = ei .
case, it is the entire multiplet, including the ordinary and Grassmann odd coordinates,
which participate in the rotation. These transformations are generated by superspace

differential operators. Denoting the generator of the U (1) transformation involving the
entire supercoordinate by J and the spinning coordinate only rotation by R0 , one finds
, )
= i (t,
, ),
J (t, , ) = i(t, , ),
J (t,

=i

R0 (t, , )
(t, , ),

, )
=i
, ).

(t,
R0 (t,
(2.19)
The Lagrangian, and hence the action, is left unchanged by R0 variations, R0 L = 0, but
not, in general, by J -transformations, except, as stated above, when g and K are functions
only of .
As discussed in Appendices A and B, the R-transformations may also be defined to
include a nonzero intrinsic R-weight, n , for the coordinates so that

= i n +
,

= i n +
(2.20)
.

As we shall see, for a consistent embedding in the full superconformal algebra, it
is required that n = 2d , where d is the scaling weight of . Note that the
R-transformation with n = 0 is R0 . Further note that since R = n J + R0 , in order
for the action to be invariant under the general R-transformations, it must be J -invariant.
Throughout the paper, the transformations will be treated as functional differential operators or as derivatives acting in superspace. For a transformation Q with superspace dif the corresponding functional differential
ferential operator representations Q and Q ,
operator is defined as

+ d
S Q .
Q dS Q
(2.21)

So defined, the Q variation obeys the same algebra as the representation of the variation
of the individual coordinates Q . The quantum action principle then relates the symmetry
variations of the quantum effective action, , to the transformations of the classical
action 0 (see Appendix C) as

Q = i Q 0 = i dV Q L ,
(2.22)
where the right-hand side is to be interpreted in terms of inserted one-particle
irreducible
(1-PI) functions. Thus, if the classical action is invariant so that Q 0 = dV Q L = 0,
then so is the quantum effective action: Q = 0. This is the Q symmetry Ward identity
for the generating functional of one-particle irreducible functions. (It should be noted that
for certain highly singular potentials, a renormalization of the Greens functions may be
required [13]. Further note that this procedure could potentially violate certain of the naive
invariances of the classical action, a well known, yet anomalous, situation in field theory.)
10
Since the classical action is SUSY and time translation invariant so is the quantum
effective action

Q = 0 = Q ,
H = 0.
(2.23)
Moreover, since the R-variations with zero R-weight are just total and derivatives,

R0 = i
,

the action is R0 -invariant. Hence

R0
R0
= i dV L = 0,
(2.24)
only, then the

for n = 0 = n . On the other hand, if K and g are not functions of
action is not J (or R) invariant but instead takes the form

J
J
= i dV L

=
dV (K K ) + (g g )D D .
(2.25)
3. Noethers theorem and superconformal transformations

A transformation of the coordinates corresponds to a good symmetry if the action
remains invariant. In the present case, the action,
0 , is given as the integral over the real
measure of thesuperspace Lagrangian as 0 = dV L. Hence, the variation of the action
is zero, 0 = dV L = 0, provided the Lagrangian transforms into the sum of chiral, F ,
2 , the
, terms: L = F + F
. Note that since {D, D}
= 2it and D 2 = 0 = D
and antichiral, F
case that the Lagrangian transforms into a total time derivative is subsummed by the above
condition. Noethers theorem is obtained by considering the variation of the Lagrangian
so that
under the transformations of the supercoordinates, + and + ,
the Lagrangian transforms as

1 0
L
1 0
L ,

+D
+ D
L =
(3.1)

D
D
D
D
where the EulerLagrange derivatives are given by

0
L D L
= D
,
D

0
L L
(3.2)
= D
D
.

D
operators in (3.1) are understood to cancel the corresponding D or D

Here the 1/D or 1/D
yields
derivatives in (3.2). Acting on Eq. (3.1) with D and D

0
L 1 0
1 0 + D D
(D )
DL =
,

D

D
D

1 0

= 0 D
D L 1 0
DL
+D
.

D
D
D
11
(3.3)
These identities lead to the superspace Noethers theorem. The Noether charge, Q,
corresponding to this transformation is defined as

L
1 0
1 0
L
1
D
+

.
D
Q
(3.4)

2
D
D
D
D
It follows that the SUSY covariant derivatives of the charge are

0 1
1 0 ,
+ DL + (D )
DQ =
2
D

1 0
= 0 1 DL
D

.
DQ

2
D
Further differentiation leads to the local form of Noethers theorem

Q = 2i Q
D, D
t

1 0
0 + (D )
=D
D

1 0
1
0

D, D
+ D
D
L.

2
D
(3.5)
(3.6)
Inserting this in the one-particle irreducible functions, applying the equations of motion
for the Greens functions (see Appendix C) and integrating over the time while ignoring
total time derivatives, the Ward identity for the variation of the effective action is secured
in the form of the quantum action principle

+ d S

dS

1 0
1 0

i d S (D )
= i dV L i dS D
,

D
D
(3.7)
with the quantum effective action (the one-particle irreducible
function
generating

functional).
In obtaining thisresult, we have used that dS = dt d = dt D

, with and Grassmann odd functions of the
and d
S = dt d = dt D
supercoordinates and their SUSY covariant derivatives.
We now focus on transformations which correspond to a reparametrization of superspace:
t t (t, , ),
(t, , ),
(t, , ).
(3.8)
Note that the variations of the superspace coordinates are general functions of superspace.
Under such transformations, the chiral and antichiral supercoordinates transform as

1 1 1
(t, , ) ei(t , , ) t 1 , 1 , 1 ,
1 , 1 , 1 ) 1 1 1
, ) ei (t
t , ,
(t,
(3.9)
,
12
where and are complex independent functions of superspace and the inverse
transformation refers to the reparametrization given by Eq. (3.8) where the primed
variables are to be replaced with unprimed ones. For infinitesimal parameters, the
reparametrization takes the form
t t + t (t, , ),
+ (t, , ),
, ),
+ (t,
(3.10)
while the chiral and antichiral supercoordinates vary as

(t, , ) (t, , ) + (t, , )
= (1 + i)(t t, , ),
, )
(t,
, ) + (t,
, )
(t,
(t
t, , ),
= (1 + i )
(3.11)
where and are infinitesimal complex independent functions of superspace.

Retaining terms through linear in the small variations gives
= (i tt )
= (i Bt D),
= (i tt )

,
= i Bt + D
(3.12)
.
where B t + i + i
Under this reparametrization, the terms in the superspace Lagrangian, L = K +
,
transform as
gD D
)
(gD D

+ i + i(D ) i D
gD D

= ig DD
+ D D

BgD D Bg D

igD D
+ D
ig D(D
)

i

g + g
D D
+ ( )
2

i
,
g g
D D
+ ( + )
2
i
i
) + ( + )(K
)
K = ( )(K
+ K
K
2
2
i i
+ D(BDK)
+ D B DK
2
2

i
i

+ DB DK + DB DK,
2
2
(3.13)
(3.14)
13
while the Noether charge corresponding to these variations has the form

1 i
i
) + ( )(K
)
Q=
( + )(K
+ K
K
2 2
2

+ D gD D
+ gDD D

K
DK DK
B K

gD D
g DD
D
iD g D
igD D

+D
gD

+ i g
DD
+ i D(
)

+ D Bg D + Bg D D + D BgD + Bg DD .
(3.15)
The charge generates the intrinsic variations of the coordinates as i[Q, ] = Q . That
is, it gives the change in the operator which results in order to insure the invariance of
their matrix elements when compared in the two frames of reference. From this it follows
that the variation of a derivative of the supercoordinate is the derivative of the variation of
the supercoordinate, e.g., Q D = D Q . We need to insure that the symmetry charges
do not change the chirality of the supercoordinates. This chirality consistency requirement
= 0, dictates
further restricts the transformation parameters. The chiral constraint, D
that
= 0,
D
= 0,
D
= 2i,
DB
(3.16)
while the antichiral constraint D = 0 mandates

D = 0,
D = 0,
DB = 2i .
(3.17)
The solutions to these equations require the transformation parameters , iX

to be chiral (antichiral) functions of superspace. Note that X and X
are
i X)
( ,
Grassmann odd. In addition, the equations for B are satisfied provided

t = A + X + X

= A X X
1
= (A + A),
(3.18)
2
are chiral (antichiral): DA
= 0 (D A = 0).
where A ( A)
Noting that the chirality of the supercoordinate variations (3.16) and (3.17) implies that
= 0 = i DB, it follows that the variation of the Khler potential simplifies
+ 2i DB
2
to

i
K + K
K K
+ i ( + )

K = ( )
2
2
i i
+ D(BDK).
+ D B DK
(3.19)
2
2
Our goal is secure the form of the various parameters characterizing the superspace
reparametrization which leads to maximum possible symmetries of the action without
restriction to the specific forms of the g and K real superfunctions. To accomplish this,
we need to further restrict the parameters so that the variation of the term containing g,
14
Eq. (3.13) has an analogous structure to the term containing K, Eq. (3.19). This is achieved
by requiring
= 0,
( + )
t
X
= 0.
+ DX + D
The resultant variation of the term containing g now becomes

gD D
i

)D D
= ( )(g
+ g
2
i

)D D
+ ( + )(g
g
2

BgD ig D(D

+D
)

igD D

.
D Bg D
(3.20)
(3.21)
Solving equations (3.20) for the transformation parameters gives

i
i
1
1
= + + t + + ,
2
2
2
2
i
i
1
1
= t + + ,
2
2
2
2

1
i
i
1
i
t + ,
X = + t +
2
2
2
2
2

= + i t + 1 + i + i t 1 ,
X
2
2
2
2
2
1
A = & t t 2 + (2 + it) + i (
+ t),
2
1
+ t),
A = & t t 2 (2 + i t) i (
(3.22)
2
where , , , , & are infinitesimal real constants and , ,
, are infinitesimal
complex constant anticommuting parameters. Recall that
1
+ X X,

= (A + A)
B = t + i + i
2
while the superspace infinitesimal transformation parameters are
1
t = (A + A)
2

1
i
i
= & t t 2 + + t + t ,
2
2
2

1
i
i
1
i
t + ,
= iX = i + t +
2
2
2
2
2

i
1
i
i
1
= i + t + + + t
.
= i X
2
2
2
2
2
(3.23)
(3.24)
15
Taken together, these are recognized as the class of infinitesimal superconformal transformations along with an additional overall UJ (1) phase transformation of the coordinates.
To obtain the finite superconformal transformations, one simply integrates these equations
yielding (with the corresponding finite parameters)
t =
e t + & +
,
1 + 1 t i + i
e
2
2i 12
1 + 12 t
i + 12 t
,
2i + 2i
e+ 2 2 + i 12 t
.
1 + 1 t i + i
i
=
(3.25)
Grouping terms with common variation parameters, the infinitesimal superconformal

and UJ (1) phase transformations of the supercoordinates take the form

1
i
1
i
t )
= + + tt + D + ( D + 2i
2
2
2
2

1
t )
+ t + i + t 2 t + t D &t + i(D + 2i
2
1
+ 2i t
t ),
i D
+ 2
t i (1 tt ) (tD
2

1
1 i
i
+ 2i
t )
= + + tt D
+ (D
2
2
2
2

1
2i t )
&t + i (
D
+ t i + t 2 t t D
2
+ i D
2i tt ).
t + i (1 tt ) 1 (t D
(3.26)
2
2
Note that these are identical to the superconformal and UJ (1) phase transformations
given in Appendix B corresponding to the case where d = d = 12 and n = 2d =
n = 1. It follows that the largest possible symmetry group is just the direct product
group OSp(2, 1) UJ (1).
is the most general (through
Since the supercoordinate Lagrangian L = K + gD D
two derivatives) SUSY and time translationally invariant object constructed from the
chiral supercoordinates, the corresponding action is, in general, not invariant under these
more restrictive superconformal transformations. Mutatis mutandis, the SUSY (and time

translation) Ward identities Q = 0 = Q are functional differential equations for the
quantum effective
action whose solution (through two derivatives) determines the classical

with arbitrary K and g. Only for
action, 0 = dV L to be given by L = K + gD D
specific forms of g and K will this larger symmetry be manifest. Isolating the UJ (1)
transformation by taking = = /2, it follows from Eqs. (3.19) and (3.21) that
and
J = 0 provided K and g satisfy Kz z Kz z = 0 = gz z gz z . As such, g(, )
are functions only of .
On the other hand, isolating the dilatation transformation
K(, )
by choosing the parameters as = = i/2, dilatation invariance, D = 0 results
provided K and g are scale independent. This follows, using Eqs. (3.19) and (3.21),
and K(, )
must
provided Kz z+Kz z = 0 = gz z+gz z which, in turn, dictates that g(, )
When the dilatation charge is conserved, the commutation relation
be are functions of /.
16
[H, D] = iH implies that eiD H eiD = e H for any real . Thus if |E is an eigenstate
of H , H |E = E|E, then eiD |E is also an energy eigenstate with energy eigenvalue
e E for any real . Consequently, the spectrum of H consists of all non-negative real
numbers with the ground state having zero energy. All excited states are doubly degenerate
as a consequence of the SUSY. Finally, we note that requiring both UJ (1) and dilatation
invariance leads to the free theory only.
The variation of the Lagrangian is given by the sum of Eqs. (3.19) and (3.21) and can
be written as

+ i ( )
(K + K
) + (g + g
)D D
L = F + F
2

i

(K K
) + (g g
)D D
+ ( + )
2
i

)D i D(
)D
)(g
+ )(g
+ D(
+ g
g
2
2
i
i
+ D( + )(g
,
) D
) D
D( )(g
+ g
g
2
2
(3.27)

where the chiral, F , and antichiral, F , combinations are defined by

i

BgD ,
+D
F = D(BDK)
i D(Dg
)
2

i
g
.

D BgD D
iD D
F = D B DK
(3.28)
2
Note that the last two terms on the right-hand side of Eq. (3.27) can be written as
Y
where
iDY i D
,
D
Y = g
= g D .
(3.29)
= 0; D Y
= 0. For the special case when D =
are chiral and antichiral, respectively: DY

and D = , where and are constant Grassmann odd superconformal

terms can be included as F and F
terms in L.
transformation parameters, the Y and Y

Since the purely chiral and antichiral terms F + F do not contribute to the variation of
the action, they may be absorbed into the definition of the Noether charge Q leading to an
improved Noether charge QI given by

1
F F ,
2
with Q defined in Eq. (3.4). Finally, introducing the combination

,
;L L F + F
QI Q +
(3.30)
(3.31)
the improved Noether charge obeys the identities

0 1
+ D;L,
2

I = 0 1 D;L,

DQ
2
DQI =
(3.32)
and the conservation relation

0
;L.
0 + 1 D, D
2i QI = D
D
t
17
(3.33)
Integrating over the time yields the quantum action principle or Ward identity

= i dV L .
(3.34)

Thus, if the classical action is invariant, dV L = 0, so is the effective action = 0.
The improved superconformal charges take the form

1 i
i
) + ( )(K
)
QI =
( + )(K
+ K
K
2 2
2

+ 4iBg i + + DX D
X
gD D

+ iBK D D
i
)D D
i ( )(g
)D D

( + )(g
+ g
g
2
2

+ 2( )g
+
+ 2( + )g
+ 2i D
gD
+ i Dg

D
2iDg D
iDg
D

D
.
(3.35)
+ 2 g + 4XgD + 4 Xg
Substituting the explicit expressions for the superspace dependent transformation parameters and isolating the coefficients of the various c-number and Grassmann odd parameters,
the superconformal charges (defined with the transformation parameters removed from
their definitions) and (non-)conservation laws can be extracted. Below we compile the various superspace dependent (improved) charges whose , independent components generate the associated superconformal symmetries. In addition, we give the SUSY covariant
derivatives of these superspace charges, their (non-)conservation laws and the form of the
explicit breaking terms.
K,
Throughout this section as well as in Section 4, we adapt the notation that H, Q, Q,

S, S, R and J represent an entire function of superspace whose , independent

component is the actual superconformal and UJ (1) generator. Elsewhere in the paper, these
same symbols are used to represent just the actual generators.
3.1. Time translations: parameter &
0

,
DH = 2i
QH
I = H 4g + K D D ,

0
0 ,
H = D
+D
t

1
H = i D 0 + D
0
D, D
,
2

H L = 0.
= 2i 0 ,
DH
(3.36)
18
3.2. Supersymmetry: parameters and

Q
Q
+ 2i H,
QI = Q 2i H,
QI = Q

Q
0
Q 0
QI = D Q
D
,
t

Q

0
0
Q
QI = D Q
D
,
t

Q
QI L = 0 = QI L,
(3.37)
where
Q 4gD ,
4g D
,
= 2iH 2iD 0 ,
DQ
= 2iH + 2i D
0 ,
DQ

0
0
Q 0
0 ,
Q = D Q
D
D
2i D
t

0

0
0
Q
0 .
D
D
+ 2i D
Q = D Q
t

(3.38)
(Note that in a slight variation from the usual notational convention Q is antichiral,
is chiral, (D
Q
= 0).)
(DQ = 0), while Q
Q,

DQ = 0 = D
3.3. UJ (1) transformations: parameter

U (1)
QI
=J

+ 4g i(g + g

)D D
2igD D
),
+ i(K + K
0 1
= 2i 0 1 D
U (1)L,
DJ
DJ = 2i
+ DU (1) L,

0
i
U (1)
0

J = D
L,
+ D, D
D

) + (g g
)D D
.
U (1)L = 2i (K K
3.4. R transformations: parameter n +

QR
I =

1
+ 2i H
,
R Q Q
2
(3.39)

R
0
R L,
R 0 1 D, D
QI = D R
+D
t
2

n U (1)
R
) + (g g
)D D
,
L=
L = in (K K
2
where
19
2i
(3.40)
R n J + R0

+ 4n g
= 2i(1 + n )gD D
)D D
+ in (K + K
).
in (g + g
(3.41)
For n = 0, the conserved supercoordinate charge R0 is obtained as

,
0 = Q,

DR0 = Q,
DR
R0 2igD D

0 ,
R0 = 2i D 0 + D
2i R0 = D, D
t

1
0 0
D
.
D, D R0 = 2iH + i D
2
(3.42)
It should be pointed out that R0 may also be obtained by directly applying Noethers
theorem to the case that the supercoordinate variations are covariant derivatives (in which
case that the transformations do not preserve the original chirality of the supercoordinates).
,
where
Considering the infinitesimal transformations = D and = D
and are independent complex anticommuting parameters, Eq. (3.4) yields the associated
Noether charge

1

Q = ( D D) gD D + K ,
(3.43)
2
while Eq. (3.1) gives the variation of L as

K.
L = D + D
(3.44)
into Q defines the improved charge QI as

Absorbing DK and DK

1

DK DK
2

i
R0 = i Q + Q
.
= D D
2
2
The derivative equations are easily obtained as

i
0 = i D Q
= H D
0 ,
DQI = D DR
2
2

0
i

I = i DDR
DQ
=
H
+
D
=
DQ
,
0
2
2
QI = Q
leading to the time derivative

0 0
R0 = D
+ D
.
t
(3.45)
(3.46)
(3.47)
20
In a similar manner, the chiral and antichiral supercoordinate counting densities can be
found by considering the number operator variations of the supercoordinates (global Weyl
scaling)

N = ,
N = 0,
N = 0,
N = .
(3.48)
Noethers theorem yields the counting charges

1
+ K 4ig ,
g D D
gD D
2

1

+ g
K
D D
4ig
QN = +gD D
2
and SUSY covariant derivatives
QN =
1
N = 0 1 D
N L,
DQN = D N L,
DQ
2
2
0 1 N

N = 1 D
N L,
DQN =
DQ
+ D L,
2
2

(3.49)
(3.50)
where
i

N L = R0 + K + g D D
2

+ (ln K) K,
= 1 + (ln g) gD D
i

+ g
D D
N L = R0 + K
2

+ (ln
g) gD D
K) K.
= 1 + (ln
(3.51)
Thus the (non-)conservation equations for the counting density are given by

QN = 2i QN
D, D
t

0
1
N
L,
= D
D, D
2

QN = 2i QN
D, D
t

N L.
0 1 D, D
=D
2

(3.52)
Combining these chiral and antichiral supercoordinate counting equations, one secures the
total supercoordinate number operator (global Weyl scaling) conservation identity:
N

U 2i
Q + QN
t
t

0
N L + N L .
0 + 1 D, D
= D
+D
(3.53)
21
This may be alternatively expressed by considering the chiral and antichiral counting
variations of the Lagrangian directly. Introducing the chiral and antichiral combinations

T = ig D
T = igD
= 0),
(DT
(D T = 0),
(3.54)
with the derivative identities

1
0
N
DT = i L i
,
D

T = i NL + i 1 0 ,
D
D

(3.55)
the total number density can be written as

+ i K K
i g g
D D
( )
t

T + i 1 0 1 0 .
= DT + D

D
U = 4g
(3.56)
The conservation equations take the form

DU = 2i
0
+ iD U L,

U L,
= 2i 0 i D
DU
from which the time derivative is obtained as usual reproducing equation (3.53)

0
0
1

U L,
D
U =D
+ [D, D]
(3.57)
(3.58)
where

U L = 2i N L + N L .
(3.59)
So defined, these are the counting operators which enter into the scaling and conformal
SUSY charges.
3.5. Dilatations: parameter

1
1
i
1
=
tH + U + Q Q ,
2
2
2
2

D
0
D L,
D 0 1 D, D
2i QI = D D
D
t
2

i
D L = U L + 2R0
4

1
.
+ g + g
D D
=
K + K
2
QD
I
(3.60)
22
3.6. Superconformal symmetry: parameters and

i
1

2iQD
+ T i D(g
)
R
QSI = t Q
+
I
2
2

i
1
= t Q + 2T iY 2iQD
I + R ,
2
2

i
1

+ 2iQD
R
QSI = tQ + T iD(g )
I
2
2

i
1
+ 2iQD
= tQ + 2T i Y
I R ,
2
2

S
1
S
S 0
S 0

L,
2i QI = D
D
D, D
t
2

1

0
S 0

S L,
2i QSI = D S
D
D, D

D
2i N L + R0
S L = 2i g

= 2iY 2i N L + R0 ,

+ 2i NL + R0 .
S L = 2ig D + 2i N L + R0 = 2i Y
(3.61)
As alluded to in the discussion following equations (3.27) and (3.28), as a consequence of

their chiral and antichiral nature, the superconformal symmetry breaking terms due to Y
can be absorbed into the charges to form new, improved charges. Defining these as
and Y

S
, the associated conservation equations are given by
QNI = QSI + iY and QSNI = QSI + i Y

1
SNI
S
S 0
S 0

D
L,
D, D
2i QNI = D

1

0
S 0

D
2i QSNI = D S
(3.62)
D, D
SNI L,
where the breaking terms are now just

SNI L = 2i N L + R0 ,

SNI L = 2i N L + R0 .
(3.63)
3.7. Conformal symmetry: parameter

i
1

tU t 2 H 2ig QSI + QSI R
2
2
i
2
= tU t H 4g i R
2

+ 4g D
+ 2g
,
D
+ tQ + 4gD + 2g D t Q
QK
I =

K
0
K L,
K 0 1 D, D
QI = D K
+D
t
2

1

K L = 2t D L S L + S L U (1)L
2
D
+ 1 U (1)L
D
= 2t L 2i g D 2i g
2
1
2i Y
+ U (1)L.
= 2t D L 2i Y
2
23
2i
(3.64)

A new, improved conformal charge can also be defined by absorbing part of the Y and Y
breaking terms into the charge. These terms in turn will become part of the new, improved
superconformal charges replacing the improved ones in the conformal charge as

i
1

(3.65)
tU t 2 H 2ig QSNI + QSNI R.
2
2
dependent terms in the conformal symmetry (non-)
However, there still remains Y and Y
conservation equation:

K
1
K
K 0
K 0

NI L
2i QNI = D
D, D
+D
Y
,
2iDY 2i D
QK
NI =
1
D
U (1) L.
K
(3.66)
NI L = 2t L +
2
By appealing to the action principle, the (local) variations of the effective action, and,
hence, the Greens functions, are given by the above equations. For general variations
the action principle yields
and ,

0
i (t, , )
= (t, , )
,
(t, , )
(t, , )

0
, )
, )
i (t,
(3.67)
= (t,
.
(t, , )
(t, , )
Introducing the Ward identity functional differential operators for each of the variations of
R, J, D, K, S,
the superconformal and U (1) transformations, Q {H, Q, Q,
S},

+ d
S Q ,
Q dS Q
(3.68)

and the local chiral and antichiral Ward identity operators

Q ,
wQ D Q
,
w Q D
(3.69)
Noethers theorem has the general form

wQ w Q = 2
i
Q
QI D, D
L.
t
2
(3.70)
24
Integrating over the time, and discarding total time differentials secures the quantum action
principle

Q
Q
=
(3.71)
dV L ,
where once again

Q = dt wQ w Q = dS Q
+ d
S Q .
(3.72)
The variations of the Greens functions under superconformal and U (1) transformations
are obtained as Ward identity functional differential equations for the effective action.
Alternatively, the operator inserted Greens functions determine the matrix elements of the
operators through the reduction formalism. The local Ward identity functional differential
operator terms vanish asymptotically, hence Noethers theorem can be viewed as an
operator relation between the time derivative of the charge and the associated variation
of the Lagrangian
2i
Q L.
QI = [D, D]
t
2
(3.73)
4. Supercharges
As discussed in the previous section, it is the and independent components of
the superspace dependent Noether charges which serve as the superconformal and UJ (1)
generators. Although these Noether charges are functions of superspace, they do not all
transform as supercoordinates. While some are supercoordinates, others involve the prod multiplying supercoordinates. Just as the
uct of explicit factors of superspace, (t, , ),
Noether charges do not form supercoordinates neither do the respective variations of the
Lagrangian. Using the superconformal algebra, however, the charges can be assembled into
multiplet structures whose components transform into each other under a restricted set of
SUSY transformations which are parametrized by the anticommuting superspace coordinates , themselves rather than by arbitrary anticommuting parameters. Such multiplet
structures are referred to as quasi-supercoordinates [14]. The restricted supersymmetry relating the Noether charges can also be used to connect the various symmetry breaking
terms. These quasi-supercoordinates of charges (called supercharges) are constructed in
Appendix B. The quasi-supercoordinates of charges are all constructed from the Hamiltonian and SUSY charge supercoordinates1 and the supercoordinate charges associated
with the chiral and antichiral global Weyl scaling of the supercoordinates (that is the chiral and antichiral number operators). The UJ (1) symmetry Noether charge forms its own
supercoordinate. Due to the quasi-supersymmetry covariance, the variations of the action
under the various symmetry transformations are also related. In fact all superconformal
1 Note that just as is the case in supersymmetric field theory, the Hamiltonian and SUSY charges can be
reconstructed from the R symmetry charge [15].
25
variations of the action are obtained from combinations of the UJ (1) symmetry variation
of the Lagrangian and the conformal SUSY variations of the Lagrangian.
Since the Noether construction is a SUSY covariant procedure, as long as the
variation of the supercoordinates under consideration is SUSY covariant, the resulting
Noether charge will be a supercoordinate. Indeed, the UJ (1) transformation of the
supercoordinates is a covariant transformation in that it commutes with the SUSY
variations of the coordinates. Thus the Noether construction of the UJ (1) charge results
in a supercoordinate. Once again, it is the , independent component which is the actual
generator, while the time derivative of the higher components simply integrate to zero. The
supercharge given in Eq. (3.39)

+ 4g i g + g
D D
J = 2igD D

+ i K + K
(4.1)
can be expanded in components as
J = J00 + 2 J10 + 2 J01 + J11 ,
(4.2)
where the , independent component is the UJ (1) symmetry generator

J00 = j = 4ig 4g(zz zz ) + i(zKz + z Kz ) 2i(zgz + z gz ) ,
(4.3)
while the higher components

0
2(zKz z Kz )z + 4i(zgz z gz )z ,

0
+ 2(zKz z Kz )z + 4i(zgz z gz )z ,
J01 = z

0
0 0
0
d
+ z
+
+
(zKz z Kz ) + 2(zgz z gz )
J11 = i z
+
z
z
dt

J10 = z
(4.4)
do not generate any symmetries.
The UJ (1) charge obeys the conservation equation

0
0
0
0
d
j = z
+ z
+
i U (1)L,
dt
z
z
(4.5)
where the UJ (1) variations of the component coordinates are

U (1) z = iz,
U (1) = i,
U (1) z = i z ,
U (1) = i .
The UJ (1) variation of the component coordinate Lagrangian L is

U (1)
L = 4i z z + 2 z z
2 (zgz z gz )
z
z

2
z z + i
(zKz z Kz )
z
z
z z
(4.6)
(4.7)
26
which gives the structure of the UJ (1) Ward identity. In superspace, this identity is obtained
by applying the action principle to Eq. (3.39) yielding

w U (1) = wU (1) w U (1)

i
U (1)
J +
=
(4.8)
L .
D, D
t
4
Integrating over time yields the variation of the effective action

i
U (1)
U (1)
U (1)
= dt w
=
L .
(4.9)
dV
2
Similarly, since the time translation of the supercoordinates is a covariant transformation, the Noether construction also yields a supercoordinate charge whose first component
is the Hamiltonian h

H = 4g + K D D
11 .
= h + 2 H10 + 2 H01 + H
(4.10)
Explicitly, the component operators are given by
0
H10 = z
,

0
0
0
0
0
,
H11 = i z
+ i z
i
+ i
H01 = z
(4.11)
.
z
z

Time translations are generated by h which satisfies the conservation law

0 0
0 0
d
h = z
+ z
+
+
.
(4.12)
dt
z
z

Equivalently, applying the quantum action principle to Eq. (3.36) yields the local Ward
identity

w H = wH w H = i H
(4.13)
t
while integrating over time yields the time translation invariance of the effective action

H = dt w H = 0.
(4.14)
h = 4g z z + 2Kzz ,
The remaining superconformal transformations do not (anti-)commute with the SUSY

charges. However, since both the R and SUSY transformations commute with H , the
quasi-supercoordinate construction will lead to a combination of charges that also forms
a supercoordinate since they act as constant supercoordinates. Specifically, the quasi is nothing but the SUSY
supercoordinate combination of variations corresponding to R
covariant U (1) transformation

= i R i Q + i Q
2 H,

i R,

Q 2
H
= R i Q i

in ,
= n U (1) =
in ,
(4.15)
27
The Noether procedure applied to the U (1) symmetry yields

where is either or .
a supercoordinate charge as given in Eqs. (3.39) and (4.1) above. Expanding the U (1)
variation of the supercoordinates, multiplying by the corresponding derivative of the action,
applying the covariant derivatives and inserting it all into one-particle irreducible functions
yields the relation

Q
H
i wR + i wQ + i w
2 w

0
Q
R 0
Q 0
= D
+ i D
+ i D

0
+ 2 D
H

0
U (1) 0
+ iD
,
= D n
(4.16)
where the last equality was obtained upon application of the action principle. A similar
expression is obtained for the antichiral supercoordinate variations yielding

i w R + i w Q + i w Q 2 w H

0
R 0 + i D
Q 0 + i D
Q
= D

H 0

+ 2 D

n U (1) 0 i D
0 .
=D

(4.17)
Taking the difference of these equations and recalling the conservation equations for
= R =
the UJ (1) and R0 charges, Eqs. (3.39)(3.42), the supercoordinate charge R
(n J + R0 ) is obtained and seen to obey the Ward identity equation

w R = wR w R + i wQ w Q + i wQ w Q 2 wH w H

i
n U (1)
L .
= R
(4.18)
+ D, D
t
2
2
A special case is obtained for n = 0 where the supercoordinate of charges R reduces to
the conserved R0 charge supercoordinate. It obeys the Ward identity
w R0 =
R0 ,
t
(4.19)
where the Ward identity operator is given by w R0 = wR0 w R0 with

wR0 = iD

w R0 = i D
(4.20)
28
beginning with Q

and Q
As is discussed in Appendix B, the quasi-supercoordinates Q
respectively, can be constructed as
and Q,

= i Q 2i H,

i Q,
H = D ,
= Q 2i

= i Q
+ 2i H,
i Q,

.
= Q + 2i H = D
(4.21)
are
Once again there are two equivalent directions along which to proceed. Since D and D

covariant derivatives, the Noether construction leads to supercoordinate charges Q and Q.
It is found that
= QQ 2i Q
H
Q
I = Q,
I
(4.22)
with the conservation equation

Q
w Q = wQ w Q + 2i wH w H = i
.
t
(4.23)
Likewise

H
= QQ

Q
I + 2i QI = Q,
(4.24)
with the corresponding conservation equation

Q

w Q = wQ w Q 2i wH w H = i
.
t
(4.25)
Equivalently, the EulerLagrange equations and hence the Ward identity operators can
Taking covariant
be used directly in order to derive the supercoordinate charges Q and Q.
derivatives of the variations (4.21) gives

0
Q 0
H 0
D
+ 2i D
= D D
,

H 0 = 2i 0 .
Q 0 + 2i D
D
(4.26)

Inserting these identities into the 1-PI functions and using the action principle, yields, after
their subtraction,

w Q = wQ w Q + 2i wH w H

0
0
= D D
+ 2i

Q
,
= i
t
(4.27)
29
where the identities in Eqs. (3.37) and (3.38) have also been exploited. In a similar fashion,
the identities lead to

0
0
0
2i D H
= 2i
,
D Q

0
Q
H 0 = D
D
0 .
D
(4.28)
2i D

Inserting these identities into the 1-PI functions and using the action principle produces

w Q = wQ w Q 2i wH w H

0
0

= D D
+ 2i

Q
,
= i
(4.29)
t
where the identities in Eqs. (3.37) and (3.38) have again been employed.
Having considered combinations of transformations that are covariant variations, either
being the supercoordinate itself or a covariant derivative of the supercoordinate, the
Noether construction led to a multiplet of charges that was a supercoordinate. Indeed,
the multiplet corresponding to the R0 transformation contains the supersymmetry charges
and the Hamiltonian among its components. Specifically expressing the supercharge R0 in
terms of its component charges gives

R0 = 2igD D

0 0
+
= r0 + i 2 q i 2 q + 2ih + i
,
(4.30)
with the component charges given in terms of the component coordinates as

q = 4g z,
r0 = 4ig ,
q = 4g z ,
h = 4g z z + 2Kzz .
The corresponding component charge conservation equations are

r0 = i
+
,
t

q = i 2i z
,
t
z

q = i + 2i z
,
t
z

h = i z + z +
+
,
t
z
z
(4.31)
(4.32)
where r0 is the R charge for phase transformations involving the Grassmann coordinates
only, while q and q are the SUSY charges and h the Hamiltonian.
30
The remainder of the superconformal charges can only be promoted to quasisupercoordinates. For example, following the procedure outlined in Appendix B, the
dilatation transformations spawn the quasi-supercoordinate of transformations

= i D 1 Q 1 Q
,
i D,
2
2
1
1 Q

= D Q +
2
2
= d + tt ,
(4.33)
which is not SUSY covariant due to the explicit time dependence. The Ward identity
operator corresponding to this transformation is obtained by taking covariant derivatives
of the above combination of variations. For the chiral supercoordinate, it takes the form

1
1

0
D
Q
D 0
Q 0
iw = D
+ D
+ D

0
0
1
0
+ tD
+ D d
= D
(4.34)
2
while for the transformation of the antichiral supercoordinate

0
Q 0 + 1 D
Q
D 0 + 1 D
i w D = D
2
2

0
0
1 0
+ tD
+ D d
.
= D
(4.35)
2

Hence, the Ward identity operator conservation equation becomes

w D = wD w D

i
D

= i D + D, D d L ,
t
2
(4.36)
is
where the dilatation quasi-supercoordinate of charges D
= tH + d U.
D
(4.37)
Here U is the global Weyl scaling charge given by the supercoordinate

U = 2i QN + QN ,
(4.38)
while the variation of the Lagrangian is given by

D d L = i
=
d U
i
L + R0
2
2
i

i

4id N L + R0 +
4id N L + R0 .
4
4
(4.39)
31
This is the same result obtained previously by combining the Noether charges to directly

form the dilatation quasi-supercoordinate D
1 Q 1 Q

= QD
D
I + QI QI
2
2
= tH + d U.
(4.40)
The conformal SUSY Noether charges can be used as the initial components of the
conformal SUSY quasi-supercoordinates of charges according to

R

Q
+ i Q
S= 2i QSI 2iQD
I QI
I
= tQ + 4gD + 2g D
+ 2i T
= tQ + 2D(g )
,
= tQ + 4i T + 2Y

S

R
Q

i Q
S = 2i QI + 2iQD
I + QI
I
+ 4g D
+ 2g
D

= tQ
+ 2D(g

+ 2iT
= tQ
)
+ 4iT + 2Y.
= tQ
(4.41)
These correspond to the variations

i
S, = i
S + (2iD + R) + i Q,

Q
= S + 2i D + R + i
= tD

i
S, = i S + (2iD + R) i Q,

Q
= S + 2i D + R i
.
= t D
(4.42)
Applying the conservation equations for the individual Noether charges gives the quasisupercoordinate conservation equations

i

i

w S = i
D, D S L +
D, D 2iD L R L
S +
t
2
2

i
S
i

S +
D, D L +
D, D 2iD L + R L
,
w S = i
t
2
2
(4.43)
where the Ward identity operators are

w S = wS w S ,
with
w S = wS w S ,

0
0
0

0
wS = D S
D 2i D
+ R
+ i D Q
,
(4.44)
32

S 0

2i D 0 + R 0
Q 0 ,
w S = D
D
+ i D

0
0
0
0
+ R
wS = D S
D 2i D
i D
Q
,

S
S 0
D 0
R 0

D 2i
+
w = D

0
Q
(4.45)
i D
.

Utilizing the identities

= D, D
+ 2D,

= D, D
+ 2D,
D, D
D, D

= D, D
1 2 D
+ 2 D
,
D, D
(4.46)
and employing the canonical R and scale weights, n = 2d = 1 = n , these take the
form

1

S
S

D
R
D, D L 2i L L
i w =
S +
t
2

D 2iD L R L

1

D D
+ 2i g
=
D, D 2i Y D 2i K
S +
t
2

=
S 2Y D 2i N L + R0
t

0
=
(4.47)
(tQ) [Q] + 2i
.
t

On the other hand, for general R and scale weights, the Ward identity becomes

0

(tQ) [Q] + i(n 2d )
i w S =
.
t

(4.48)
In analogous fashion, for canonical R and scale weights,

1
S
S
D
R
D, D L + 2i L + L
S +
i w =
t
2

2iD L + R L
+ D

1
2iK + 2ig D D

=
D, D (2iY ) + D
S +
t
2

N

2i L + R0
=
S 2Y + D
t

+ 2i 0 ,
=
(4.49)
(t Q) + [Q]
t
33
while for general R and scale weights, the Ward identity takes the form

+ i(n 2d ) 0 .
i w S =
(4.50)
(t Q) + [Q]
t
Finally, the conformal transformation yields the quasi-supercoordinate variation

= i K +

i K,
S + S + R,

S + R
= K + S
= 2td + t 2 t ,
(4.51)
along with the quasi-supercoordinate of Noether charges

S
= 2i QK
S
R
K
I QI + QI QI
= t (U tH ) 4g .
(4.52)
The conservation equation yields the conformal Ward identity

1

1
K
K
D, D L D, D
S L
i w =
K
t
2
2
1
S
1
R
+ D, D L D, D L
2
2

1
K

=
K
D, D L + S L S L + R L
t
2

S
R L ,
L + DS L + 1 D + D
D
(4.53)
where the conformal Ward identity operator is w K = wK w K with the chiral and
antichiral Ward identity operators given by

0
0
0
0
wK = D K
D S
+ D S
+ D R
,

S 0
K 0 D

S 0 + D
R 0 . (4.54)
w K = D
+ D

Substituting the expressions for the variations of the Lagrangian from Section 3 and
exploiting the identity

t = t D, D
2i D 2i D,

D, D
(4.55)
the conformal Ward identity is obtained as

i
U
Y

L + 2R0 + 2i DY + D
K + t D, D
i w K =
t
4

1 U (1)
+
L
2

i
U

K + 4g + [U ] + t D, D
L + 2R0
=
t
4

i
U L + 2R0 . (4.56)
=
tU t 2 H + [U ] + t D, D
t
4
34
In summary, the (quasi-)supercoordinates of charges for the superconformal group

are given in terms of the Hamiltonian and SUSY charge supercoordinates, which are
components of the R symmetry charge, and the charges associated with the chiral and
antichiral global Weyl scaling of the supercoordinates. In addition, the breaking of the
superconformal symmetries is given in terms of the breaking of these global Weyl scaling
symmetries and the additional U (1) J symmetry of the Lagrangian. In particular, the
symmetries and conservation equations can be summarized as the Ward identities for the
effective action (see Table 1).
Recall that in the special case that n = 0, the R charge reduces to the conserved R0
charge:
,
DR0 = Q,
R0 = 2igD D

1
R0 = 2iH + i D 0 D
0 = Q,

0 ,
D, D
DR
2

0 .
R0 = 2i D 0 + D
2i R0 = D, D
t
(4.57)
The superconformal Ward identities involve the variance of the Lagrangian under the chiral
and antichiral global Weyl scaling transformations (these are just the chiral and antichiral
number operators). The combinations
N

2i L + R0 = 2i K + g D D
Table 1
Symmetry
Charge
U (1)
Time translation
Supersymmetry
Q

Q
R transformation
Dilatation

D
Superconformal

S

S
Conformal

K
Ward identity

U (1) = 12 dV U (1) L
H = 0

Q = Q + 2i H = 0

Q = Q 2i H = 0

R = R + i Q + i Q 2 H
n
= i 2 dV U (1) L

D = D + 12 Q 12 Q = i dV D
d L

= 14 dV 4id N L + R0 + 4id N L + R0

S = S 2i D + R + i Q

= i dS 4id N L + R0

S = S 2i D + R i Q

= i d
S 4id N L + R0

K = K S + S + R

S

n U (1)
L
= i dV 2tD
d L + i dS L + 2

n
n
+ i d
S S L + 2 U (1) L i dt 2 U (1) L
and
35

+ g
D D
2i N L + R0 = 2i K
characterize the superconformal symmetry breaking. The strong requirement of superconformal symmetry implies that the theory is trivial. On the other hand, the less restrictive
dilatation symmetry breaking is characterized by the sum of the above breaking terms:

id U
L + 2R0
2

.
+ g + g
D D
= d K + K
D
d L =
As in the case of the superconformal symmetries, the conformal Ward identity is satisfied
only for a free theory.
Acknowledgements
This work was supported in part by the US Department of Energy under grant DEFG02-91ER40681 (Task B).
Appendix A. Superspace and supercoordinates

N = 1 supersymmetry (SUSY) is generated by a complex Grassmann valued (hence
and the Hamiltonian, H .
nilpotent) supersymmetry charge, Q, its complex conjugate, Q,
Together, these charges satisfy the supersymmetry algebra

= 2H,
H ,
Q, Q
[Q, H ] = 0 = Q,

Q
.
{Q, Q} = 0 = Q,
(A.1)
ei Q,
does not alter
by a real phase, Q ei Q and Q
Since multiplying Q and Q
the algebra, one can define an additional charge, R, generating this automorphism and
satisfying

= Q,

[R, Q] = Q,
R, Q
[R, H ] = 0.
(A.2)
N = 1 superspace is obtained by introducing a complex Grassmann coordinate, , and
denotes
its complex conjugate, , in addition to the time, t, so that the triplet (t, , )
2
points in superspace. The representation of the symmetry transformations as superspace
differential operators on supercoordinates is most readily obtained from the motion that
group multiplication induces on the parameter space of the SUSY graded Lie group. An
element of this group is obtained by exponentiation of the sum of the products of the
charges with the corresponding superspace point giving
)

= ei(t H +Q+Q
= ei(Q+Q) eit H .
G(t, , )
2 For the construction of superspace and supercoordinates in extended SUSY, see [5,16].
(A.3)
36
Multiplication of group elements is determined from the algebra (A.1) by exploiting the
BakerCampbellHausdorff formula. As a consequence of the SUSY algebra, the Baker
CampbellHausdorff formula truncates to
eA eB = eA+B+ 2 [A,B] .
1
(A.4)
It follows that the multiplication law for group elements is given by

G t , , G t, ,

= G t + t + i , + , + .
(A.5)
Hence, group multiplication (from the left) induces the movement in parameter space

t + t + i , + , + .
(t, , )
are
A supercoordinate, (t, , ), whose transformation properties under H, Q and Q
determined by the above shift, is defined so that
(t, , ) = G(t, , )(0,

0, 0)G1 (t, , )

= ei(Q+Q) (t, 0, 0)ei(Q+Q) ,

with (t, 0, 0) = eit H (0, 0, 0)eit H . It then follows that

G t , , t, , G1 t , ,

= t + t + i , + , + .
(A.6)
(A.7)
For infinitesimal superspace parameters (&, , ),

the group elements and shifted super Employing the notation Q for the
coordinate can be Taylor expanded about (t, , ).
superspace differential operator acting on corresponding to the operator Q so that
i[Q, ] = Q ,
(A.8)
and recalling that differentiation with respect to the Grassmann

for Q {H, Q, Q},
coordinates is given by

= 1,
(A.9)
= 1,

with all other derivatives vanishing, the representation of the SUSY transformations as
superspace differential operators is secured as
H ,
t

+ i
Q ,
i[Q, ] =
t

] = i Q
.
i[Q,
t

i[H, ] =
(A.10)
= 2H , and the identity for nested commutators,

From the charge algebra, {Q, Q}

Q
Q

[{Q, Q}, ] = { , }, it is seen that the superspace differential operator variations Q
37
obey the same algebra (up to a factor of i) as the charges:

Q Q

, = 2i H ,

Q H
, = 0 = Q, H ,
Q Q

, = 0 = Q, Q .
(A.11)
Since SUSY transformations induce a real time translation, i( ) in Eq. (A.7),

(t, , ) can be chosen to be real and denoted by . This real supercoordinate, = ,
has a component expansion

+ W
(t).
t, , = x(t) + i (t) i (t)
(A.12)
The time dependent coefficients in this decomposition consist of the even element
coordinates x(t) and W (t) and the Grassmann odd elements (t) and (t).
The reality
of dictates that x = x, W = W and = .

The R symmetry variations are determined in an analogous manner using the group
multiplication law

eiR G t, , eiR = G t, ei , ei ,
(A.13)
with a real parameter. Since R transformations form an Abelian subgroup, the variation
of is given by

eiR t, , eiR = einR t, ei , ei ,
(A.14)
where nR is the intrinsic R charge (or R-weight) of the supercoordinate. For real ,
this R-weight is required to vanish; nR = 0. On the other hand, for the case of complex ,
which we shall consider next, nR is arbitrary. For infinitesimal , the R transformations
yield the R variation of the supercoordinate,

i R, t, , = i nR +
t, ,

R t, , .
(A.15)
= Q
implies
Once again the charge algebra [R, Q] = Q and [R, Q]
R Q
R Q

, = i Q ,
, = i Q ,
(A.16)

as is verified by direct calculation using the explicit forms for R , Q , and Q .

An alternative induced motion in parameter space can be obtained by group multiplication on the right

G t, , G t , ,

= G t + t i , + , +
(A.17)
giving

t, , t + t i , + , + .
(A.18)
38
defined by
For the supercoordinate (t, , )

t, , = G1 t, , (0, 0, 0)G t, ,

= t, , ,
(A.19)
the induced motion yields

G1 t , , t, , G t , ,

= t + t i , + , + .
(A.20)
For infinitesimal superspace parameters, this right multiplication induced motion leads to
,
t

i
D,
i[Q, ] =
t

i[Q, ] =
+ i
D,
t

i[H, ] =
where the covariant Grassmann derivatives are defined as
= + i .
i ,
D
D=
t
t

Note that
Q

Q

= 0 = Q
,
, D = 0 = Q , D ,
,D
,D
(A.21)
(A.22)
(A.23)
while

= 2i ,
D, D
t

D
.
{D, D} = 0 = D,
(A.24)
(Since [/t, Q ] = 0 = [/t, Q ], it also follows that /t is a covariant derivative.)

The covariant derivatives can be used to obtain irreducible representations of the
SUSY algebra by covariantly constraining the general supercoordinate so that either
= 0, producing a chiral supercoordinate, or D = 0, which yields an antichiral
D
= 0 implies that the chiral supercoordinate is
supercoordinate. (As discussed below, D
necessarily complex and distinct from the complex antichiral supercoordinate; i.e. = .)
= 0, is given by
The solution to the chiral constraint, D

t, , = ei t (t, , 0),
(A.25)
while the antichiral constraint, D = 0, is satisfied provided

t, , = ei t t, 0, .
(A.26)
The component expansions of the chiral and antichiral supercoordinates contain complex
coordinates z, z and complex (spinning) Grassmann coordinates , so that

t, , = ei t z(t) + i 2 (t) ,

t, , = ei t z (t) i 2 (t) .
(A.27)
Alternative parametrizations of the group elements are provided by

G1 t, , = ei(t H +Q) ei Q = eiQ ei Q eit H ,

G2 t, , = ei(t H +Q) eiQ = ei Q eiQ eit H .
39
(A.28)
The relation between each parametrization is obtained using the BakerCampbell

Hausdorff formula as

, .
G t, , = G1 t i , , = G2 t + i ,
(A.29)
The parametrizations differ by purely imaginary shifts, i , in the time coordinate.
Proceeding just as before, left group multiplication with these representations also induces
a motion in parameter space which now take the forms

G1 t , , G1 t, , = G1 t + t 2i , + , + ,

G2 t , , G2 t, , = G2 t + t + 2i , + , + .
(A.30)
It follows that two different supercoordinate representations can be defined using G1
and G2 as

1,2 t, , = G1,2 t, , (0, 0, 0)G1

(A.31)
1,2 t, , .
Since the group elements are related by imaginary shifts in the time, so are the
supercoordinate representations

, .
t, , = 1 t i , , = 2 t + i ,
(A.32)
is called the real representation while 1 is known as the chiral representation and 2 as
the antichiral representation. The SUSY group transformations are given by the induced
parameter transformations

G1 t , , 1 t, , G1
1 t , ,

= 1 t + t 2i , + , + ,

G2 t , , 2 t, , G1
2 t , ,

= 2 t + t + 2i , + , + .
(A.33)
Once again, for infinitesimal superspace parameters (&, , ),
the SUSY variations in the
chiral and antichiral representations can be determined. Using the notation i[Q, 1,2]
( Q )1,2 , the variations are given by

1 ,
t
Q
1 ,
1=

Q

+ 2i
1=
1 ,
t

H
H
2 = 2 ,
t

Q
+ 2i
2=
2 ,
t
Q

2 = 2 .

In all representations the nilpotent SUSY variations obey { Q , Q } = 2i t .
(A.34)
40
Just as was the case using the real representation, the right cosets can be used to define
the covariant derivatives in the chiral and antichiral representations. From the parameter
space motion deduced from

G1 t, , G1 t , , = G1 t + t 2i , + , + ,

G2 t, , G2 t , , = G2 t + t + 2i , + , + ,
(A.35)
the supercoordinate defined as

1,2 t, , = G1
1,2 t, , (0, 0, 0)G1,2 t, , ,
(A.36)
is found to satisfy

G1
1 t , , 1 t, , G1 t , , = 1 t + t 2i , + , + ,

G1
2 t , , 2 t, , G2 t , , = 2 t + t + 2i , + , + .
(A.37)
Finally, for infinitesimal superspace parameters (&, , ),

these transformations lead to the
expressions for the covariant derivatives in the chiral and antichiral representations given
by

2i
2 ,
D2 2 =
D1 1 =
1 ,

D1 1 = 1 ,
(A.38)
2 .
D2 2 =
+ 2i
t

= 2i are SUSY covariant derivatives in all representations.
Note that t and {D, D}
t
The utility of the (anti)chiral representations follows from the observation that
when employed, the chiral and antichiral supercoordinate constraints reduce to or
independence. Thus in the chiral representation
1 1 = 0 = 1 ,
D

which implies that 1 is independent of and can thus be expanded as

1 t, , = z(t) + i 2 (t).
(A.39)
(A.40)
On the other hand, in the real representation, the chiral supercoordinate is obtained from
its representation in the chiral representation by a simple imaginary shift in time so that

,
t, , = 1 t i ,

= ei t z(t) + i 2 (t) ,
(A.41)
reproducing the result presented in Eq. (A.27). Similarly, the antichiral constraint in the
antichiral representation takes the form
D2 2 = 0 =
2 ,
(A.42)
which implies 2 is independent of and thus has the component decomposition

2 t, , = z (t) i 2 (t) .
(A.43)
41
Once again to obtain the real representation for the antichiral supercoordinate one need
only perform an imaginary time shift of the opposite sign to that used for the chiral
supercoordinate. So doing, one finds

,
t, , = 2 t + i ,

= ei t z (t) i 2 (t) ,
(A.44)
just as in Eq. (A.27). In an analogous fashion, the variations and the covariant derivatives
are also related directly through imaginary time shifts as

Q,Q = ei t Q,Q 1 ei t = ei t Q, Q 2 ei t ,
()
()
()
D = ei t D1 ei t = ei t D2 ei t .
(A.45)
The SUSY transformations of the component coordinates of each supercoordinate are

most easily determined in each supercoordinates respective representation. For a chiral
supercoordinate in the chiral representation, the SUSY variations are
Q
1 = Q z i 2 Q
1 = i 2 ,
=
Q

1 = Q z i 2 Q

=
1 = 2i z .
2i
(A.46)
t

Hence the chiral supercoordinate components SUSY transformations are
Q = 0,
Q z = i 2 ,

Q = 2 z .
Q z = 0,
(A.47)
For an antichiral supercoordinate in the antichiral representation, the SUSY variations are
Q
2 = Q z i 2 Q

+ 2i
=
2 = 2i z,
t

Q

2 = Q z i 2 Q
= 2 = i 2 .
(A.48)
Hence the antichiral supercoordinate components SUSY transformations are
Q z = 0,
Q = 2 z,

Q = 0.
Q z = i 2 ,
(A.49)
42
Finally for the real supercoordinate in the real representation, the SUSY variations are
QW
Q = Q x i Q i Q +

=
+ i
= i + W + i x + ,
t

Q
W
Q = Q x i Q i Q +

=
= i + W i x + .
i
(A.50)
Hence the real supercoordinate components SUSY transformations are

Q x = i,
,
t
Q = i(W + i x),

Q W = ,
t

Q = 0.
Q W =
Q = 0,

Q x = i ,

Q = i(W i x),
(A.51)
The last ( ) component of a real supercoordinate transforms as a total time derivative,

= t and Q W = t . Hence, assuming total differentials in time integrate to zero,

it follows that dt W is SUSY invariant; Q dt W = 0 = Q dt W . In the superspace
framework, to extract the last component of the real supercoordinate, one integrates over a
vector measure defined as

dV = dt d d = dt W.
(A.52)
Q W
In obtaining this result, one employs the convention that integration over the Grassmann
parameters is defined just as differentiation so that
d =
d =
(A.53)

Since Q dt W = Q dt W = 0, it follows that Q dV = Q dV = 0 and hence
is SUSY invariant.
For chiral, , and antichiral, , Grassmann supercoordinates, the last component of
each also transforms as a total derivative or is itself invariant. Thus additional SUSY
invariant terms can be secured as integrals of a (anti)chiral
Grassmann
supercoordinate

taken over a (anti)chiral measure defined as dS = dt d and d
S = dt d .
Appendix B. Superconformal transformations
The N = 1 superconformal group algebra can be obtained using the product rules of
the orthosymplectic group OSp(2, 1) whose algebra consists of 8 (charges) generators [8].
Three charges generate the familiar conformal subgroup SO(2, 1) of quantum mechanics.
They consist of the Hamiltonian H which generates time translations, the time dilatation
43
charge D and the conformal transformation charge K. The remaining 5 charges consist
of the generator, R, of superspace Grassmann (spinning) coordinate rotations, the
and the anticommuting special (conformal)
anticommuting SUSY charges, Q and Q,
SUSY charges, S and
S. An element of OSp(2, 1) can be represented by matrices having
the generic structure [6,17]
SO(2, 1) Q
H, D, K S .
(B.1)

Q
S R
By examination of the matrix commutators, the OSp(2, 1) algebra can be extracted as
[H, D] = iH,
[H, K] = 2iD,
[D, K] = iK,

{Q, Q} = 2H,
{S, S} = 2K,
[R, Q] = Q,
[R, S] = S,
= Q,

[R, Q]
[R,
S] =
S,
i
= i Q,

[D, Q]
[D, Q] = Q,
2
2
i
i
[D,
S] =
S,
[D, S] = S,
2
2
= iS,
[K, Q] = i
S
[K, Q]

[H, S] = i Q,
[H,
S] = iQ,

{Q, S} = 2D iR,
{Q,
S} = 2D + iR,
(B.2)
with all other (anti)commutators vanishing.

The representation of these transformations on a general supercoordinate can be
determined by the techniques of Appendix A. Employing the notation i[Q, ] = Q for
each charge Q, the variations in the real representation are given in Table 2, where n is
the R charge (weight) of the supercoordinate while d is its scaling weight (engineering
dimension). If is hermitean, then n = 0, while for chiral with R-weight n , then
the antichiral supercoordinate has R-weight n = n . Requiring chiral consistency in
with the transformations
the commutation relations of the covariant derivatives, D and D,
imply that for chiral supercoordinates n = 2d , while for antichiral supercoordinates
Table 2
Symmetry
Time translation
Supersymmetry
R transformation
Dilatation
Conformal
Superconformal
Transformation
(t, , )
H (t, , ) = t

(t, , )
Q
(t, , ) = + i t

(t, , )
Q (t, , ) = i t

(t, , )
R (t, , ) = i n +

+ 1 + 1 (t, , )
D (t, , ) = d + t t
2
2

(t, , )
+ + t 2 t
K (t, , ) = 2td + in + t

+ i (t, , )
S (t, , ) = i (2d n ) + t + it t

S (t, , ) = i (2d + n ) t it + i (t, , )
44
that n = 2d . As can be explicitly demonstrated, the variations obey the superconformal

algebra. Thus if [Q1 , Q2 ] = if Q3 , then [ Q1 , Q2 ] = f Q3 .
In terms of the component coordinates, the superconformal and U (1) transformations
are given in Table 3. Note that in order for the superconformal algebra to close, it is
necessary that d = d + 12 .
Some of the superspace dependent charges of the superconformal algebra have
explicit time dependence and thus do not transform as supercoordinates under SUSY
transformations. It is possible, however, to add explicit and terms to these charges such
that the new charge transforms under a restricted SUSY transformation as would a time
independent supercoordinate while still obeying the same superconformal algebra as the
is defined
associated charge without the explicit and . Such a quasi-supercoordinate Q
so that under a SUSY transformation parametrized with the superspace coordinates
and , it transforms as would a supercoordinate so that [14]

= + Q.

, Q
i Q + Q
(B.3)

The solution to this equation is given by
)

) = ei(Q+Q
Qei(Q+Q) .
Q(,
(B.4)
Table 3
Symmetry
Chiral components
transformation
Antichiral components
transformation
Time translation
H z = z
Supersymmetry
H =
Qz = i 2
H z = z
H =
Q = 0

Q
z=0
R0 transformation
Dilatation
Conformal
Superconformal
Q z =

Q = + 2 z
Q = 0
R0 z = 0
R0 z = 0
R0 = +i
R0 = i
D z = (d + tt )z
D = (d + tt )

K z = 2td + t 2 t z

K = 2td + t 2 t
S z = 0
D z = (d + tt )z
D = (d + tt )

K z = 2td + t 2 t z

K = 2td + t 2 t
S z = it 2

S = 2 2 d 12 + tt z

S z = it 2

S = 0
UJ (1)
Q z = 0
Q = 2 z
J z = +iz
J = +i
S = 0

S z = 0

S = + 2 2 d 12 + tt z
J z = i z
J = i
45
For each of the superconformal charges the corresponding quasi-supercoordinate is

constructed
= H,
= Q 2i H,
H
Q
2 H,
= R i Q + i Q
R
1
1
,
= D Q Q
D
2
2
= K +
K
S + S + R,

S = S + (2iD + R) i Q,

S =
S + (+2iD
+ R) + i Q.
= Q

+ 2i H,
Q
(B.5)
obey the same superconformal algebra as the Q.

The quasi-supercoordinates of charges Q
Appendix C. Quantum action principle
The generating functional, Z[, ],
for ground state expectation values of the time
ordered products of chiral and antichiral supercoordinates is given by the Feynman path
integral
Z[, ]
= 0|T ei dS+i d S |0

S )
i(0 + dS+ d
,
= [d][d ]e
(C.1)

where 0 = dV L = dt L is the classical action. For the model discussed in the body
of this paper, the action is given in Eq. (2.2) or equivalently Eq. (2.3). Here and
are chiral and antichiral Grassmann scalar supercoordinate sources having the component
decompositions

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

i t
t, , = e
(C.2)
(t) + J (t) .
2
, and d
S = dt , respectively, the
Using the chiral and antichiral measures, dS = dt
generating functional expressed in terms of the component coordinates is

i dt {L+J z+Jz + + } .
Z[, ]
= Z J, J , , = [dz][d z][d ][d ]e
(C.3)
The connected function generating functional, Z c [, ],
is defined as
Z[, ]
= eZ
c [,]
(C.4)
which is the generating functional for one-particle

The quantum effective action [, ],
irreducible (1-PI) functions is defined as the Legendre transform of Z c :

= Z c [, ]
[, ]
(C.5)
i dS i d
S ,
46
defined
where the sources for the 1-PI functions are the classical supercoordinates and ,
as
1 Z c
= ,
i
1 Z c
= .
i
(C.6)
In terms of the component coordinates, the Legendre transform is

z, z , , = Z J, J , , i dt J z + Jz + + ,
(C.7)
with the classical supercoordinates given by

1 Z c
= z,
i J
1 Z c
= ,
i
1 Z c
= z ,
i J
1 Z c
= .
i
(C.8)
It should be noted that in the supercoordinate case, the chiral supercoordinate functional
derivatives are Grassmann odd. Thus they have been used in the Grassmann odd chiral and

antichiral integrals in order to give chiral, S , and antichiral, S , superspace delta functions
that then integrate to Grassmann even functions so that

(1)
1 (1, 2) = 12 ei2 12 t1 (t1 t2 ),
= S (1, 2) = D
(2)
(1)

i
= S (1, 2) = D1 (1, 2) = 12e 12 2 t1 (t1 t2 ),
(2)
(C.9)
where the real measure delta function is defined as (1, 2) = (t1 t2 )1212 with 12 =
1 2 ; 12 = 1 2 .
The supercoordinate EulerLagrange equations for the above chiral supercoordinate
model with action (2.2) can be derived either graphically or by using a formal change
of integration variable in the path integral. So doing, one finds that, for this action, no
renormalization is needed. Consequently, the equations of motion for the Greens function
generating functional are
0
Z[, ]
= (1)Z[, ],
(1)
0
Z[, ]
= (1)Z[,
],
(1)
(C.10)
while for connected functions one finds

0 c
= (1),
Z [, ]
(1)
0 c
Z [, ]
= (1).
(1)
(C.11)
The functional derivatives of the classical action, 0 , are just the supercoordinate
generalization of the classical EulerLagrange derivatives of the Lagrangian. For the action
under consideration,

+ K dV L,
0 = dV gD D
the EulerLagrange derivatives are

0
L D L
gD D
K ,
= D
=D
(1)
D

0
L L

= D g DD
+ K .
= D
D

(1)

D
47
(C.12)
The operator inserted EulerLagrange equations can then be Legendre transformed in order
to obtain the effective action equations of motion

[, ]
0
0
[, ]
=
[, ] =
,
i
i
(C.13)
[, ]
.
(1)
(1)
(1)
(1)
In a similar fashion, using either graphical or path integral techniques, composite
operator, O, inserted EulerLagrange equations of motion can be gleaned and take the
form

0
= O [, ]
,
[, ]
iO
(C.14)
(1)
(1)
Thus the quantum action principle relates the variation of the
where is either or .
effective action to that of the classical action

+ d S
= [i0 ]
=
dS

0
0

+ i d S
= i dS
(C.15)
= i dV L ,

where and are chiral and antichiral products of the supercoordinates and their
derivatives.
Appendix D. Anticommuting Grassmann (spinning) coordinate determinant
In this appendix, we shall explicitly integrate the anticommuting Grassmann (spinning)
coordinate out of the path integral representation of the Green function generating functional resulting in a path integral involving only the particle ordinary spatial coordinates.
Expressed in terms of the component coordinates, including the anticommuting Grassmann
(spinning) coordinates, the path integral is given by

i dt {L+J z+Jz + + } ,
Z J, J , , = [dz][d z][d ][d ]e
(D.1)
where the Lagrangian with canonically normalized kinetic energy term so g = 1/4 is

i
L = z z + + i Kz z Kz z 2Kzz .
(D.2)
2
Since this Lagrangian is bilinear in the Grassmann (spinning) coordinates, they can be
integrated out as [4,18,19]

Det[SF1 ] i dt dt (t
)SF (t t )(t )
i dt [Lf + + ] =
e
,
[d ][d ]e
(D.3)
1
Det[SF 0 ]
48
where

i
2Kzz
(D.4)
2
is that part of the Lagrangian having dependence on the anticommuting (spinning)
coordinates and . Here we normalized the determinant by its noninteracting (Kzz = 0
value so that the path integral gives unity in the absence of all interactions and sources.
The inverse fermion propagator, determined from Lf , is

d
SF1 t t ; z, z = i 2Kzz + i& t t ,
(D.5)
dt
Lf =
d
while SF10 = (i dt
+ i&)(t t ). The determinant can be secured as a product of the
eigenvalues, , of SF1 defined through the eigenvalue equation

dt SF1 t t ; z, z t ; z, z = t; z, z .
(D.6)
The solution to this integral equation is given by

i(t t0)2i tt d Kzz (z,z)
0
t; z, z = t0 ; z, z e
.
(D.7)
Eventually, we will want to take the adiabatic limit, t . Since the time integral in
the exponential diverges in this limit, we must regulate the integral in the intermediate
steps. We do so by introducing a cutoff at time T . As such, we need to impose boundary
conditions on the solution. Imposing antiperiodic boundary conditions, the anticommuting
coordinate eigenfunctions are then required to satisfy
(T ) = (T ).
(D.8)
This, in turn, restricts the allowed eigenvalues to be

1
(2n + 1)
n =
2T
T
T
d Kzz ,
(D.9)
where n = 0, 1, 2, . . . is any integer.

The desired ratio of determinants is then simply given by the product of the eigenvalues
as
T

Det[SF1 ]
n= n
T d 2Kzz

=
=
1
(2n + 1)
Det[SF10 ]
n= n |Kzz =0
n=
T
= cos
d Kzz .
(D.10)
Taking the adiabatic limit, T , the Green function generating functional (ignoring the
Grassmann sources for simplicity) takes the form
Z[J, J, 0, 0]

= [dz][d z] ei dt Kzz (z,z) + ei dt Kzz (z,z) ei dt [zz +i(Kz z Kz z )+J z+J z ] .
(D.11)
Finally, letting

m
(x + iy),
z=
2

z =
49
m
(x iy),
2
and scaling K qK, with q the particle electric charge, so that

q
q
B(x, y) =
&ij i Aj (x, y),
Kzz =
2m
2m
where B(x, y) is the magnetic field and A is the vector potential, we secure the path integral
representation
Z[Jx , Jy , 0, 0]

qB(x,y)
qB(x,y)
1
2

= [dx][dy] ei dt 2m + ei dt 2m ei dt [ 2 mv +q Av+Jx x+Jy y] .
(D.12)
Here v = e1 x + e2 y is the particle velocity and we have replaced the complex sources J, J
with the real sources Jx , Jy . This path integral expression is completely equivalent (note in
this appendix, we have set h = 1) to the path integral resulting from the Lagrangian (2.11)
or the Hamiltonian (2.12) where the trace over the Pauli matrix magnetic dipole moment
interaction Hamiltonian has been taken to obtain the two magnetic field exponential terms
in the path integral.
References
[1] E. Witten, Nucl. Phys. B 185 (1981) 513.
[2] For reviews of 1-dimensional supersymmetric quantum mechanics, see, for example, F. Cooper, A. Khare,
U. Sukhatme, Phys. Rep. 251 (1995) 267;
W. Kwong, J.L. Rosner, Prog. Theor. Phys. Suppl. 86 (1986) 366, Festschrift volume in honor of Y. Nambu.
[3] V.P. Akulov, A.I. Pashnev, Theor. Math. Phys. 56 (1983) 862;
V.P. Akulov, A.I. Pashnev, Theor. Math. Phys. 65 (1985) 84.
[4] J.W. van Holten, Nucl. Phys. B 196 (1982) 509;
A.C. Davis, A.J. Macfarlane, P. Popat, J.W. van Holten, J. Phys. A 17 (1984) 2945.
[5] R. Coles, G. Papadopoulos, Class. Quantum Grav. 7 (1990) 427.
[6] J. Michelson, A. Strominger, Commun. Math. Phys. 213 (2000) 1, hep-th/9907191;
R. Britto-Pacumio, J. Michelson, A. Strominger, A. Volovich, Cargese 1999, Progress in string theory and
M-theory 235264, hep-th/9911066;
P. Claus, M. Derix, R. Kallosh, J. Kumar, P.K. Townsend, A. Van Proeyen, Phys. Rev. Lett. 81 (1998) 4553,
hep-th/9804177.
[7] S. Fubini, E. Rabinovici, Nucl. Phys. B 245 (1984) 17.
[8] E. DHoker, L. Vinet, Lett. Math. Phys. 8 (1984) 439;
E. DHoker, L. Vinet, Commun. Math. Phys. 97 (1985) 391.
[9] T.E. Clark, S.T. Love, S.R. Nowling, Mod. Phys. Lett. A 15 (2000) 2105.
[10] M. de Crombrugghe, V. Rittenberg, Ann. Phys. 151 (1983) 99;
A. Barducci, R. Casalbuoni, L. Lusanna, Nuovo Cimento 35A (1976) 377.
[11] R. Jackiw, Phys. Rev. D 29 (1984) 2375.
[12] R.J. Hughes, V.A. Kosteleck, M. M Nieto, Phys. Rev. D 34 (1986) 1100;
V.A. Kosteleck, V.I. Manko, M.M. Nieto, D.R. Truax, Phys. Rev. A 48 (1993) 951.
[13] R. Jackiw, Delta function potentials in two-dimensional and three-dimensional quantum mechanics, in:
A. Ali, P. Hoodbhoy (Eds.), Beg Memorial Volume, World Scientific, Singapore, 1991.
50
[14] T.E. Clark, O. Piguet, K. Sibold, Nucl. Phys. B 143 (1978) 445;
T.E. Clark, O. Piguet, K. Sibold, Nucl. Phys. B 169 (1980) 77;
T.E. Clark, O. Piguet, K. Sibold, Nucl. Phys. B 172 (1980) 201.
[15] T.E. Clark, S.T. Love, Int. J. Mod. Phys. A 11 (1996) 2807.
[16] C.M. Hull, The geometry of supersymmetric quantum mechanics, hep-th/9910028.
[17] C. Duval, P.A. Horvthy, J. Math. Phys. 35 (1994) 2516.
[18] E. Gildener, A. Patrascioiu, Phys. Rev. D 16 (1977) 1802.
[19] F. Cooper, B. Freedman, Ann. Phys. 146 (1983) 262.

Two-loop effective potential of O(N)-symmetric

scalar QED in 4 dimensions
H. Kleinert a , B. Van den Bossche a,b
a Institut fr Theoretische Physik, Arnimallee 14 D-14195 Berlin, Germany
b Physique Nuclaire Thorique, B5, Universit de Lige Sart-Tilman, 4000 Lige, Belgium
Received 6 April 2001; accepted 3 April 2002
Abstract
The effective potential of scalar QED is computed analytically up to two loops in the Landau
gauge. The result is given in 4 dimensions using minimal subtraction and -expansions. In three
dimensions ( = 1), our calculation is intended to help throw light on unsolved problems of the
superconducting phase transition, where critical exponents and the position of the tricritical point
have not yet found a satisfactory explanation within the renormalization group approach. 2002
Elsevier Science B.V. All rights reserved.
1. Introduction
One of the most intriguing problems in the physics of critical phenomena is a theoretical
understanding of the superconductive phase transition within the renormalization group
approach. The standard field theory investigated in this context is the GinzburgLandau
(GL) model, initially in 4 dimensions, by Halperin, Lubensky, and Ma [1] (see also [2]),
who generalized an observation by Coleman and Weinberg [3] in four-dimensional
quantum field theory to many-body physics. In a one-loop approximation, they did not
find an infrared-stable fixed point. Only by artificially allowing for an unphysically large
number of replica N/2 of the complex field , with N larger than 365, it has been possible
to stabilize the renormalization flow in 4 dimensions.
At present, and in spite of much effort, it is still unclear whether a higher-loop
renormalization group analysis would be capable of explaining the existence of a critical
point in 4 dimensions. Experimentally, this existence has been confirmed only recently
E-mail address: bvandenbossche@ulg.ac.be (B. Van den Bossche).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 2 6 0 - 2
52
H. Kleinert, B. Van den Bossche / Nuclear Physics B 632 (2002) 5168
in high-Tc superconductors. Evidence had so far come only from Monte Carlo simulations
[4] and an analogy with smecticnematic transitions in liquid crystal [5].
The confusing situation in the GL model certainly requires further investigation in
higher-loop approximations. To appreciate the problem, let us first recall what makes the
first-order result of Ref. [1] questionable: the simple argument leading to the first order
is based on a fluctuation-corrected mean-field theory obtained in three dimensions by
integrating out exactly the gauge field at fixed ||. The resulting trace-log can be evaluated
exactly, and produces a cubic term ||3 , which makes the transition of first-order.
This is in contrast to analogous mean-field analysis performed in a disorder field
theory derived via a duality transformation of the GL model by one of the authors in
Ref. [6]. The latter analysis predicted definitely a second-order
transition in the type-II
regime characterized by a Ginzburg parameter > 1/ 2. This parameter measures the

ratio between magnetic penetration depth and coherence length in the superconductor. For
large , vortex lines produce many zeros in the order field which make the mean-field
assumption of a constant || in the previous derivation of the ||3 term unreliable.
In Kleinerts field theory [6], a disorder field describes vortex loops, whose density
is ||2 . This field has a quartic self-interaction and a gauge-coupling to a massive vector
field h representing the fluctuating magnetic induction in the superconducting phase. Due
to its mass, the field h can be integrated out and the assumption of a constant mean field
|| presents no problem, leading to a Landau-like expansion of the free energy containing
interaction. Its coefficient
terms ||4 , ||6 , etc. The ||4 term represents the vortexvortex
turns out to be proportional to t , with t 0.8/ 2. For < t , vortices attract each
other and the transition is of first order, whereas for > t , they repel each other and the
transition is of second order transition. Thus, the mean-field approximation in the disorder
theory suggests a second-order transition in the type-II regime of repulsive vortex lines and
a first-order transition in the type-I regime of attractive vortex lines.
The point = t is a tricritical point. Its existence has been confirmed
by Monte Carlo
2. With better
simulations on a lattice in [7], but still at a different value t 0.38/
simulation techniques, however, Kleinerts 1982 prediction t 0.8/ 2 has recently been
verified with amazing accuracy [8].
In the mean-field theory of the original GL model [1] fluctuations the vector potential
seem to make the transition of first-order for all values of . For a satisfactory
understanding of the transition, we would certainly like to find a way of producing a
second-order transition in the type-II regime without the duality argument. For this it
will be useful to know the effective potential of the GL theory as accurately as possible.
With this motivation we calculate in this paper the two-loop effective potential of O(N)symmetric scalar QED containing N/2 complex fields in 4 dimensions which, for
N = 2, reduces to the GL model of superconductivity.
We are aided by recent progress made in evaluation techniques of Feynman diagrams in
4 dimensions. In particular, two-loop Feynman diagrams with unequal masses of the
internal lines are now available analytically [9,10]. Moreover, the full -expansion of the
so-called sunset diagram is known [11]. This is important for critical phenomena in 4
dimensions since, for instance, a three-loop calculation requires the knowledge of all terms
of order of the two-loop diagrams.
53
It must be noted that in 4 dimensions, the renormalization flow of the coupling

constants does not improve much over the one-loop calculation of [1] when going to two
loops [12]. Apparently, a direct three-dimensional approach is more promising to lead to a
second-order transition for N = 2 [13]. The absence of a charged fixed point for N = 2 in
4 dimensions seems thus to be a specific weakness of the -expansion, although some
progress has been made in [14]: there, a [1/1] PadBorel resummation of the two beta
functions (associated with the electric charge and with the self-coupling of the scalar field)
has indeed lead to a desired IR stable fixed point for N = 2. The resummation at such a
low order is, however, not very reliable and the result must be considered as fortuitous.
We hope that eventually variational perturbation theory will be able to locate a fixed
point of the GL model, thus allowing to extract physical values independently of Ref. [14].
This theory has been developed and discussed in detail in Refs. [1517] and it has
proven to be a powerful tool for determining critical exponents in three [16] as well as in
4 dimensions [17] of pure 4 -theories. Variational perturbation theory is a procedure
which allows to determine resummed quantities from a strong-coupling limit of divergent
expansions in powers of the bare coupling constant. It does not require any more Pad
or PadBorel resummation. In particular, there is no freedom in selecting different Pad
approximants which yield different results at higher orders than [1/1].
Recently, we have applied variational perturbation theory to a new range of problems:
the determination of amplitude ratios in three dimensions of the O(N)-model [18]. We
based these calculations on expansions of the critical exponents obtained by a certain
regularization method [19], in which analytic regularization is applied with a formal
minimal subtraction in 4 dimensions, but inserting at the end = 1 without invoking
further -expansions. Our results were consistent with PadBorel methods. Because of its
power, we believe that variational perturbation theory will eventually succeed in producing
the desired zeros in the beta functions to yield an IR-stable charged fixed point.
Remarkably, a similar determination of amplitude ratios in 4 dimensions with
-expansion is still missing. This is a second motivation for the present calculation of the
effective potential in 4 dimensions. In a later paper we shall use variational perturbation
theory to recast our result near the critical point into the scaling form of Widom, and hope
to be able to exhibit the tricritical point found in the disorder theory.
We shall cross-check all our results by setting the electric charge equal to zero to ensure
that we recover the results for the usual O(N)-symmetric 4 -theory.
The renormalization constants and various beta functions of scalar QED have been
calculated up to two loops in the early work on this subject [12,20]. We do not
intend to rederive these results. However, in the process of obtaining the effective
potential, i.e., of calculating various Feynman diagrams in the symmetry-broken phase,
we shall not only obtain the effective potential but also recover previously obtained
renormalization constants of mass Zm2 and 4 -coupling constant Zg through two loops.
The renormalization constants of the scalar and vector fields Z and ZA , on the other hand,
enters only with the first-loop order. The fact that the charge coupling renormalization
constant Ze is not obtained to the same order as the 4 -coupling constant one has its origin
in the fact that, due to a Ward identity, Ze = Z , see Eq. (10).
As far as related calculations are concerned, we point out that the effective potential
of the electroweak part of the standard model is available through two-loop order [9].
54
This, however, does not make our work superfluous which is not simply an abelian
U (1) subresult of this reference for two reasons: first, we work with N/2-complex scalar
fields (N = 2 in [9]); second, in view of the application to critical phenomena in three
dimensions, higher terms in the -expansion are used. At the two-loop order, for example,
the one-loop part of the effective potential is expanded up to the order . This term is, of
course, ignored in [9], where one is interested only in four dimensions.
Finally, we mention that our work can be seen as the extension to a two coupling
constants problem of the work of Brzin et al. [21] (see also [22]), which consider the
-expansion of the equation of state of the N -components 4 -theory through two-loop
order.1 Without the gauge field, the effective potential can be used to determine this
equation of state.
The paper is organized as follows. In Section 2, we specify the model and our
conventions. In Section 3, we present various intermediate steps for the obtaining of the
effective potential. The individual results are combined in Section 4, and the conclusions
are drawn in Section 5.
2. Model
The Lagrangian density to be studied contains N/2 complex scalar fields coupled to
the abelian fields A and reads, with a covariant gauge fixing,
g 4 1 2
1
|| + F +
( A )2 ,
(1)
3!
4
2
where D = ieA is the covariant derivative, F is the usual field-strength tensor and
is a gauge parameter. In principle, there are also ghost fields which, however, decouple
in the symmetric phase and remain massless. Working in dimensional regularization they
dDp
do not contribute to the energy because of Veltmans rule (2)
D p = 0 for all .
The effective potential will be obtained using the so-called background-field method of
DeWitt [24]. We shift the scalar field by an unknown constant : + . This
generates new vertices. To simplify the calculation, we shall use throughout the Landau
gauge 0. This reduces the number of Feynman diagrams and, since = 0 enforces
A 0 at the Lagrangian level, removes a possible mixing of A and A terms,
thus decoupling scalar and gauge propagators. It is further advantageous to use real fields,
defining
L = |D|2 + m2 2 +
1
= (1 + i2 ),
2
1
= (1 + i2 ).
2
(2)
Then the Lagrangian has the expansion around the background field:

1
1
T
L = L0 + GT P T + GL P L + A D T P
A
2
2
1 For the Ising model, corresponding to N = 1, the authors of [23] have succeeded going to the three-loop
order of the -expansion of the equation of state.
55

1
g 2 2
+ 4 2 + e2 A2 + e2 A2 2 + eA (2 1 1 2 ),
4!
2
(3)
where and are now N components real fields written as two-dimensional iso-vectors
= (1 , 2 ) and = (1 , 2 ). The notation is
+
1
g
L0 = m2 2 + 4 ,
2
4!
(4)
g 2
,
3!
g
GL 2 + m2L = 2 + m2 + 2 ,
2
T
2
2
2
2 2
D + m = + e ,
i j
i j
PijT = ij
,
PijL =
,
2
GT 2 + m2T = 2 + m2 +
(5)
(6)
T
P

= 2
(7)
with GT , GL being the transverse and longitudinal inverse propagators of the scalar field,
T the inverse transverse propagator of the photon field. The transversality of the
and D
latter is due to the Landau gauge. Note that there is no term eA (2 1 1 2 ) which
would mix vector and scalar propagators.
Compared to the complex-field representation, real fields have one small complication:
the gauge scalarscalar vertex involves a vector product since it mixes the real and
imaginary parts 1 and 2 . For complex fields it is diagonal in the scalar field space.
Up to and included two loops, the fastest way of determining the Feynman diagrams
with their proper weight is to use the Wick theorem. In higher loops, it might be interesting
to use a more efficient algorithm to generate the diagrams and their weight, in the same
spirit as, for instance, [25,26]. Using a similar procedure, we have derived the diagrams
of scalar QED up to the fourth order. Their calculation will form the subject of a separate
publication [27].
We shall calculate the diagrams in the following expansion, in which h counts the
number of loops

h
(N 1)
Veff. pot. = Vclassical +
2

2

2 g
+ h
N 1
4!

+
+ (D 1)

+ 2(N 1)

e2
(D 1) (N 1)
2

2
g
(N 1)
3!

+
(N 2)
+3

+3
e2
4
2
2
2

+2
(8)
56
Explicitly, this equation reads in a unit D-dimensional volume

h
1
g
V = m2 2 + 4 + (N 1) Tr ln GT + Tr ln GL + (D 1) Tr ln D T
2
4!
2

2

g
+ h 2
)
+
dp
(p,
m
)
(N 1) dp
(p,
m
T
L
4!

2

2
+ 2(N 1)
dp
dp
(p, mT ) + 2
(p, mL )

e2
dp
+ (D 1)
(p, m ) (N 1) dp
(p, mT ) + dp
(p, mL )
2

2

g
2 (N 1) dp
dq
(p, mT )(q, mT )(p + q, mL )
3!

+ 3 dp
dq
(p, mL )(q, mL)(p + q, mL)

e2
2

4
e2

2
2
T
T
(p)P
(q)(p, m )(q, m )(p + q, mL )
dp
dq
P
T
(p)q q (p, m )
dp
dq
P

(N 2)(q, mT )(p + q, mT ) + 2(q, mL)(p + q, mT )
,
(9)
where dp
d D p/(2)D , and (p, m) = 1/(p2 + m2 ).
All quantities in these expressions (fields, coupling constants, and masses) are bare
quantities. Up to the second order in the loop expansion, the divergences show up as poles
in up to the order 1/2 . They have to be removed to have a finite limit 0. They may
either be removed by adding counter-terms to the initial Lagrangian, and calculating the
associated Feynman diagrams in addition to those in (8). Alternatively, and this is how we
shall proceed, we may use (8) and (9) and include renormalization constants which are
reexpanded up to any given order in the loop parameter, i.e., up to the order h 2 in our case.
These renormalization constants are defined as

A = A r ZA ,

= r Z ,
g = gr
Zg
Z2
e = er /2
m2 = m2r
Zm2
,
Z
Ze
er
= /2,
Z ZA
ZA
(10)
where, in the last equation, we have taken into account the relation Ze = Z , which is
a consequence of a Ward identity. Intuitively, it comes from the requirement D

Z D r for the covariant derivative. In the above equations, the bare quantities are
on the left-hand side and the renormalized ones are on the right-hand side, indicated by the
subscript r.
57
We also note that the vacuum requires a special treatment [28]. For = 0, the
dimensionality requires V mD . We therefore add a term m4 h/g to the Lagrangian. With
this we define a new renormalization constant absorbing the vacuum divergencies by
m4
m4r
h=
Zv .
g
gr
(11)
In the next section, we calculate the different loop orders, based on (9).
3. Evaluation of the diagrams

In the following, the subscript indicating the renormalized quantities will be omitted,
for brevity of notation. The different renormalization constants are expanded with respect
to the loop-parameter
Zj = 1 +
L

(l)
h l Zj .
(12)
l=1
The results are then identified order by order in h . To fix the ideas, we give here to
procedure for the gauge field trace-ln: the bare diagram has to be modified according to
e2 2 e2 2 Z /ZA , and the result reads, to first order in h

Z
Tr ln D T = Tr ln 2 + e2 2
ZA

2

(1)
1
(1)
2 2
Tr ln + e2 2 + he
(13)
.
Z
Tr
A
2 + e2 2
The last term will contribute to the two-loop result, removing parts of its -poles.
3.1. Renormalized zero- and one-loop order
The renormalized zero-loop order is trivial
1
g
m4
V (l = 0) = m2 2 + 4 +
,
2
4!
g
(14)
while the renormalized one is simply a combination of the previous bare zero-order and
the trace-ln terms
g
m4 (1)
1
(1)
Z
V (l = 1) = m2 2 Zm2 + 4 Zg(1) +
2
4!
g v

2 2 2 /2 2 2 2 /2
/(1 D/2) 1
(N 1) mT
+
+ mL
(4)D/2 D
m2T
m2L

2 2 /2
+ (D 1) m2
(15)
.
m2
58

(1)
The constants Zj are determined in order to remove the -poles at D = 4 in the Euler
/(1 D/2) function. In four dimensions, it would be sufficient to make the -expansion
of (15) up to the order 0 . However, we are also interested in going to dimension D = 3.
For this reason, it will be necessary to perform the -expansion up to the order .
3.2. Renormalized two-loop order
The two-loop order contains contributions from the zero- and one-loop bare diagrams,
which will cancel the two-loop poles from the bare two-loop diagrams. These contributions
read (ct is for counter-terms)
V (l = 2)ct
1
g 4 (1)
m4 (1)
(1)
Zg +
= m2 2 Z m
Z
2 +
2
4!
g v

2 2 /2
/(1 D/2) 1
(N 1) mT
+
(4)D/2 2
m2T

(1) 1

1 2
(1)
3mT m2L Zm2 + m2L m2T Zg(1) m2T Z
2
2

2
/2
2
(1) 3 2
(1)
1 2
2
2
2 (1)
+ mL
3mT mL Zm2 + mL mT Zg mL Z
2
2
m2L

2 2 /2 2 (1)
(1)
m Z Z A
+ (D 1) m
.
m2
(16)
To simplify the evaluation of the two-loop diagrams, we introduce the functions J (m)
and I (m1 , m2 , m3 ), where we have kept the same notation as in [9]:

J (m) =
dp
(p, m),
(17)

I (m1 , m2 , m3 ) =
dp
dq
(p, m1 )(q, m2)(p + q, m3).
(18)
The function J (m) is trivial to determine, and its result has in fact been used in (16). For
I (m1 , m2 , m3 ), we use the result obtained in [9], which are in accordance with [10] and
the recent work [11]. The latter reference gives an expansion for all order in but, since
we need in the present work only the term of order 0 , we stop at this order:
J (m1 ) =
/(1 D/2) 2 D/21

(m )
,
(4)D/2
(19)
2
(4)4 I (m1 , m2 , m3 )

2 3 2

2 2
2
2
2
2
= 2 m1 + m2 + m3
m + m2 + m3 L1 (m1 , m2 , m3 )
2 1
1
L2 (m1 , m2 , m3 ) 6L1 (m1 , m2 , m3 )
2

+ m22 + m23 m21 ln m22 ln m23 + m23 + m21 m22 ln m23 ln m21

+ m21 + m22 m23 ln m21 ln m22 + (m1 , m2 , m3 )

+ m21 + m22 + m23 7 + (2) ,
where we have defined

2

m
ln m2 = ln
+ ln(4),
2

L1 (m1 , m2 , m3 ) = m21 ln m21 + m22 ln m22 + m23 ln m23 ,
2
2
2
L2 (m1 , m2 , m3 ) = m21 ln m21 + m22 ln m22 + m23 ln m23 ,

1/2
(m1 , m2 , m3 ) = 4 2m21 m22 + 2m21 m23 + 2m22 m23 m41 m42 m43

L(1 ) + L(2 ) + L(3 ) ln(2) .

2
59
(20)
(21)
(22)
(23)
(24)
In the latter expression, L(t) is the Lobachevsky function, defined as

t
L(t) =
dx ln cos x,
(25)
and the angles are given by

(m21 + m22 + m23 ) 2m2j
j = arctan
.
(2m21 m22 + 2m21 m23 + 2m22 m23 m41 m42 m43 )1/2
(26)
These expressions, as well as the function (m1 , m2 , m3 ), are valid for a positive argument
of the square root. This depends on the value of the masses. For a negative value, one has
to substitute
1/2

(m1 , m2 , m3 ) = 4 m41 + m42 + m43 2m21 m22 2m21 m23 2m22 m23

M(1 ) + M(2 ) + M(3 ) ,
(27)
t
M(t) = dx ln sinh x,
(28)
0
1
j = coth
(m21 + m22 + m23 ) 2m2j

(m41 + m42 + m43 2m21 m22 2m21 m23 2m22 m23 )1/2

.
(29)
In the following, we shall always keep the symbolic notation (m1 , m2 , m3 ). This will
avoid to check the sign of the argument of the square root. It is not necessary to do so in
60
the pure 4 -case where we know that mL > mT , and, then, where the representation (24)
is valid. In the present case, because we do not known the fixed points, it is not possible to
specify the value of the photon mass vs. the transverse or longitudinal mass. We however
mention that several simplifications arise when two masses are equal, or when one, or two
masses, are vanishing. For two equal masses, say m1 = m2 , the argument of the square
root becomes m43 (1 4m21 /m23 ). This case arises when investigating the O(N)-theory.
The ratio of the masses is the ratio 4m2T /m2L which is always smaller than unity. The
relevant equations are then (24), (25) and (26). The case of vanishing masses is a little
bit problematic for the extraction of the power 0 because of a cancellation of a diverging
part of M(t) with a corresponding lnj in (20). For this reason, we also give the following
integrals:

2
1 /(2 D/2)/(3 D)/(D/2 1)2 2 m21
m1
I (m1 , 0, 0) =
, (30)
(4)D
/(D/2)
2

2
1 /(2 D/2)/(1 D/2) 2 m21
I (m1 , m1 , 0) =
(31)
m1
,
(4)D
D3
2
2
(4)4 I (m1 , m2 , 0)

2 3 2

2 2
2
2
= 2 m1 + m2
m + m2 L1 (m1 , m2 , 0)
2 1

1
L2 (m1 , m2 , 0) 6L1 (m1 , m2 , 0) + 2m21 ln m21 ln m22
2

2
2 2

+ ln m1 m22
m1 m22 2ln m21 m22 ln m22 m21 m22

2

2 2

m22
2
2
m
+
m
7
+
(2)
+
+ 2 m21 m22 Li2
+
m
m
1
2
1
2 ,
3
m22 m21

(32)
where Li2 (z) = i=1

is the dilogarithm. It is a simple exercise to check that, using
m1 = m2 in (32), we recover the -expansion of (31), while using m2 = 0, we recover the
-expansion of (30). In the previous determination of (32), we have assumed that m1 > m2 .
The case m2 > m1 is obtained from the former by the replacement m1 m2 . This only
affects the term of order 0 . Comparing Eq. (32) with Eq. (20), we can also extract the way
(m1 , m2 , m3 ) diverges as one of the masses, say m3 , goes to zero. For m1 > m2 , we have

lim (m1 , m2 , m3 ) + m21 m22 ln m21 ln m22 ln m23
m3 0

2
2 2 2
2
2
= m1 m2 ln m1 ln m2 + ln m1 m22
zi /i 2

2ln m21 m22 ln m22 + 2Li2
m22
m22 m21

2
,
3
(33)
while the opposite case m1 < m2 is obtained by permuting m1 and m2 in this equation.
From (33) we can also determine the case of two simultaneously vanishing masses
61

2

lim (m1 , m3 , m3 ) + 2m21 ln m21 ln m23 m21 ln m23
m3 0

2 2 2
.
(34)
ln m1 +
3
With all these definitions, we are now armed to compute the two-loop diagrams. They are
given by
2
g
= (N 1)J (mT ) + J (mL )
4!

(35)
+ 2 (N 1)J (mT )2 + J (mL )2 ,
= m21

e2
= (D 1)J (m ) (N 1)J (mT ) + J (mL ) ,
2

2

g
=
2 (N 1)I (mT , mT , mL ) + 3I (mL , mL , mL ) ,
3!
2
2

e
1 2
2
2 2
m 2m
=4
I (m , m , mL ) (D 2) +
2
4m4 L

1
J (m )2 m2L 2m2 + 2m2 J (m )J (mL )
4
4m

2
2 2
4
+ 2 mL m I (m , mL , 0) mL I (mL , 0, 0) ,
(36)
(37)
(38)

1
(N 2) 2J (m )J (mT ) J (mT )2
4

+ I (m , mT , mT ) m2 4m2T
2
(m + m2L m2T )
+2
J (m )J (mL )
m2
= e 2
(m2 m2L + m2T )

m2
I (mT , mL , 0)
J (m )J (mT ) J (mL )J (mT )
(m2T m2L )2
m2
+ I (m , mT , mL )
(m2T m2L )2 + m2 (m2 2m2T 2m2L )

m2

,
(39)
where, compared to Eq. (8), we have included in the same diagram the different transverse
and longitudinal components, and where the vertices and corresponding multiplicities have
been once again specified.
4. Effective potential
Collecting the results from the previous section, the renormalization coupling constants
are tuned to cancel the poles in 1/2 and 1/. Working with the effective potential, as we do
62
here, logarithms are appearing. Some of them have poles in coefficients. These logarithms
with pole coefficients have been named dangerous poles by Chung and Chung [29]. The
cancellation of these dangerous poles is a nontrivial consequence of the renormalizability
of the theory. We can see it as follow: asking for the cancellation of one-loop poles, we
identify straightforwardly
(N + 2)
,
3
g 2 (N + 8) + 108e4
gZg(1) =
,
3
N
Zv(1) = g ,
2
(1)
Zm2 = g
(40)
(41)
(42)
where a factor 1/(4)2 has been absorbed in the definition of h (see the expansion (12)).
The cancellation of the poles at the two-loop order gives renormalization coefficients
(2)
(2)
Zm
which depend on ln(mL ), ln(mT ) and Z(1) and a renormalization coefficient
2 , Zv
(1)
Zg(2) which depends on ln(mL ), ln(mT ), ln(m ) and Z(1) , ZA
. We find that, asking for
(1)
(1)
the independence with respect to these dangerous logarithms, fixes Z , ZA . This is nontrivial since there are more conditions than renormalization coefficients, and this is the
mentioned consequence of the renormalization. Everything together, the cancellation of
the poles at the two-loop order gives then

(N + 2) 2
g (N + 5) 18ge2 + 54e4
2
9

1 2
(43)
g (N + 2) 8ge2 (N + 2) 6e4 (5N + 1) ,
6

1
gZg(2) = 2 g 3 (N + 8)2 18g 2 e2 (N + 8) + 108ge4 (N + 8) + 108e6(N + 18)
9
1 3
g (5N + 22) 12g 2 e2 (N + 5)
9

(44)
18ge4(5N + 13) + 18e6(7N + 90) ,

N
2
Zv(2) = g 2 18e2 + g(N + 2) + ge2 N ,
(45)
6
6
(1)
Z = e 2 ,
(46)
N
(1)
ZA = e2 ,
(47)
3
(2)
Zm2 =
where, as for the one-loop renormalization constants, a factor 1/(4)2 has also been
absorbed in the definition of h .
With these one- and two-loop renormalization constants, all the poles disappear from
the -expansion of the theory. Up to this order (two loops), and, in view of applications
of this theory to phase transitions in three dimensions, the one-loop order term has to be
developed to the order . The two-loop effective potential in the Landau gauge can then be
written as
V =V
(0)

2

h
h (1,0)
(1,)
V
+
+
+ V
V (2) ,
(4)2
(4)2
63
(48)
where we have specified that a factor (4)2 is absorbed in the definition of h , and with
1
g
m4
V (0) = m2 2 + 4 +
,
2
4!
g

(N 1)m4T 3 + 2ln m2T + m4L 3 + 2ln m2L
V (1,0) =
8

+ m4 5 + 6ln(m2 ) ,

2

(N 1)m4T 21 18ln m2T + 6 ln m2T
V (1,) =
+ 2
96

2

+ m4L 21 18ln m2L + 6 ln m2L + 2

2

+ 3m4 9 10ln m2T + 6 ln m2T
+ 2 .
(49)
(50)
(51)
The term V (2) is much longer to write. For this reason, we give its expansion on the lnbases. With the definition
V (2) =
2
2
2

(2) 2 i 2 j 2 k

Vi,j,k
ln mT
ln mL
ln m
(52)
i=0 j =0 k=0
we have (see, however, the remark after these equations)

1
(2)
2
2 2
4
2 2
4
=
18m
V0,0,0
g m (N 1)(N + 5)mT + 18mL mT (2N + 13)mL
4
432gm

2

+ ge2 42 m2L m2T m2L + m2T

6m2 (19N 31)m4T + 24m2T m2L + 7m4L

18m4 (N 1)m2T + m2L + (19N 36)m6

108e4m2 m2L m2T (N 1)m2T + 3m2L

+ 6g 2 7g 2 m4 2(N 1)m2T + (N + 8)m2L

+ 18e4 7m6L 7m4L m2 + 11m2L m4 + 54m6

+ 2 g 2 m4 2(N 1)m2T + (N + 8)m2L 3 m2T m2L + g 2

2

+ 18ge2 m2 m2L m2T m2L + m2T + 3m4 (N 1)m2T + m2L 9m6

+ 18e4 9m4 m2L m2T (N 1)m2T + 3m2L

+ g 2 2m6L m4L m2 + 5m2L m4 + 10m6

2
+ 216(4)4ge2 m2 m2L m2T I (0) (mL , mT , 0)
64

2

+ e2 2 m2 m2L I (0) (m , mL , 0)

+ 54ge2 m2 (N 2) m2 4m2T m2 (m , mT , mT )

2

+ 2 m4T 2m2T m2 + m2L + m2 m2L (m , mT , mL )

+ e2 2 m4L 4m2L m2 + 12m4 (m , m , mL )

+ 6g 3 2 m4 (N 1)(mT , mT , mL ) + 3(mL , mL , mL ) ,
(53)

m2T 2
g (N 1)m2 m2L + (N + 3)m2T + 2g 2
2
12gm

2

+ 6ge2 3 m2L m2T m2 2m2L + m2T (7N 9) + 5(N 1)m4

54e4 (N 1)m2 m2L m2T ,
(54)
(2)
=
V1,0,0
(2)
V2,0,0 =

1 2
g (N 1)m2 m2L 6m2T + g 2 + m2T 3m2T (N + 3) + 4g 2
2
72gm

2

9ge2 2 m2L m2T m2T + m2 (N 2) m4 + 6m4T

6(2N 3)m4 m2T

162e4(N 1)m2 m2L m2T m2T ,
(55)

m2L 2 4
g m (N + 5)m2L + (2N + 7)m2T + (N + 8)g 2
12gm4

2

+ 6ge2 m2 3 m2L m2T m2 5m2L + 2m2T + 5m4

6e4 m4 27 m2L m2T 16g 2 + 3g 2 m2L 2m2 m2L ,
(56)

m2L 2 4
g m 3(N + 5)m2L + 3(N + 8)m2T + (N + 17)g 2
4
72gm

2

+ 18ge2 m2 m2L m2T + 3m4

9e4 2m4 27 m2L m2T 7g 2 + g 2 m2L 4m2 3m2L ,
(57)

e2
9e2m4L 2 3m2 m4L + m4T 2m2L m2T + 4e2 2
2
6m

+ 12m4 m2L + (N 1)m2T 6e2 2 m6 (4N 9) ,
(58)
(2)
V0,1,0
=
(2)
V0,2,0 =
(2)
V0,0,1 =
(2)
V0,0,2 =

e2
18m8 + e2 2 m6L + 8m4L m2 22m2Lm4 + 40m6 ,
8m4
(59)

(2)
V1,1,0 =
(2)
V1,0,1 =
(2)
V0,1,1 =
65

1
g(N 1)m2L m2 3m2T + g 2
2
36m

2

9e2 m2L m2T m2L + m2T + 3m2 m4L + m4T

3m4 m2L + m2T + m6 ,
(60)
3

e2 2
mL m2T + 3m2 m4L m4T 3m4 m2L m2T
2
4m

+ (N 1)m6 ,
(61)

e2 2 6 2
e mL + m2 m6L m4L 3m2T + 4e2 2 + 3m2L m4T m6T
4
4m

+ m4 m2L 3m2L + 14e2 2 + 3m4T + 3m6 m2L m2T + m8 ,
(62)
where the function I (0) (m1 , m2 , 0) denotes the 0 nondiverging piece of (32), without
the coefficient ( )2 . Some part of this function may be put in the other terms of the
expansion (52), because they lead, for example, to ln(m1 )ln(m2 ). We have not proceed
in this way because, in the scalar QED case, it is not yet known if m > mT . One has
to study first the fixed points of the system. At the critical point, one knows that, in the
4 formalism with -expansion, m 0. At the critical point, one would then need to
compare e2 to g/6. If m > mT , one is allowed to use (32) with m1 = m , m2 = mT . In
the opposite case m < mT , one has to use (32) with m2 = m , m1 = mT . We also note
that I (0) (m1 , m2 , 0) generate terms containing ln(m21 m22 ). These terms can never be put
on the basis defined in (52).
To see the coherence of our effective potential, we shall look at the 4 -limit, which can
be reached using e2 0.
4.1. Limit of pure 4 -theory
In the limit e2 0, which gives the pure 4 -theory, the renormalized coupling constants
and the effective potential become simpler. The renormalization constants can be readily
read off Eqs. (40)(47) with e2 = 0. The zero- and one-loop effective potential is obtained
from Eqs. (49)(51) and (53)(62). Taking the limit e2 0, we find
1
g
m4
V (0) = m2 2 + 4 +
,
2
4!
g

V (1,0) =
(N 1)m4T 3 + 2ln m2T + m4L 3 + 2ln m2L ,
8

2

+ 2
V (1,) =
(N 1)m4T 21 18ln m2T + 6 ln m2T
96

2

+ m4L 21 18ln m2L + 6 ln m2L + 2 .
(63)
(64)
(65)
66
The two-loop contribution V (2) is simplified much more when going to e2 0. It has the
expansion
V (2) =
2
2

i 2 j
(2)
Vi,j ln m2T
ln mL ,
(66)
i=0 j =0
with the following coefficients:

g
(2)
V0,0
=
6 54m2L m2T 3(2N + 13)m4L + 7(N + 8)m2L g 2
432

+ (N 1)m2T 3(N + 5)m2T + 14g 2

+ 2(N 1)m2T + (N + 8)m2L 3 m2T m2L + g 2 2

+ 6g 2 (N 1)(mT , mT , mL ) + 3(mL , mL , mL )
g 4
=
m (N 1)(N 9) + m4L (5N + 43) + m2L m2T (7N 52)
24 T

+ m2L m2T (N 1)(mT , mT , mL ) + 3(mL , mL , mL ) ,
(2)

g
(N 1)m2T m2L + (N + 3)m2T + 2g 2
12

g
= (N 1)m2T 5m2L + (N 3)m2T ,
12
(67)
V1,0 =

g
(N 1) m2L 6m2T + g 2 m2T 3(N + 3)m2T + 4g 2
72

g
= (N 1) m4L 3m2L m2T (N 1)m4T ,
24
(68)
(2)
=
V2,0

g 2
mL (N + 5)m2L + (2N + 7)m2T + (N + 8)g 2
12

g 2
= mL (N + 17)m2T (2N + 19)m2L ,
12
(69)
(2)
V0,1
=
(2)

g 2
mL 3(N + 5)m2L + 3(N + 8)m2T + (N + 17)g 2
72

g
= m2L 4m2L 3m2T ,
8
(70)
V0,2 =
(2)
V1,1
=

g
g
(N 1)m2L 3m2T + g 2 = (N 1)m4L ,
36
12
(71)
(72)
where we have replaced g 2 by 3(m2L m2T ) in each second equation. We have checked
that Eqs. (67)(72) reproduce the effective potential given in [9]. The correct limit of the
pure 4 -theory is recovered.
Let us remark that the above effective potential of the pure 4 -theory was obtained by
taking the limit e2 0 for fixed m2 , and after this the limit m2 0. This is the fastest way
to arrive at our result, but somewhat cavalier. However, by replacing everywhere m2 by its
67
value e2 2 we have checked that we are permitted to proceed in this way: all m2 -terms in
the denominator of Eqs. (53)(62) are properly compensated, yielding the two-loop result
of Eqs. (67)(72).
5. Conclusion
We have determined the effective potential of scalar QED in 4 dimensions in an
-expansion. For zero charge, we recover the well-known result for the pure O(N)-symmetric 4 -theory. The full effective potential will be used in order to determine various
amplitude ratios in the context of a scalar field in the presence of a gauge field. This
comprises the superconducting case, for which N = 2. In this respect, it might be
interesting to work in an arbitrary gauge and to show that these amplitude ratios are gauge
independent. This could be done following Ref. [30]. The task is, however, far from being
trivial.
In a subsequent paper, we shall recast our result in the scaling form of Widom to
exhibit explicitly the critical behavior of the free energy. This will be done using variational
perturbation theory. From the result we shall extract the various critical exponents in scalar
QED, including the amplitude ratios along the lines of Ref. [18]. In 4 dimensions with
-expansion, this has not yet been done even in the absence of a gauge field.
Acknowledgements
We are grateful to Dr. F.S. Nogueira for many interesting discussions and clarifications
and to A. Pelster for several improvements of an earlier version of the manuscript. The
work of B.V.d.B. was supported by the Alexander von Humboldt foundation and the
Institut Interuniversitaire des Sciences Nuclaires de Belgique.
References
[1] B.I. Halperin, T.C. Lubensky, S.-K. Ma, Phys. Rev. Lett. 32 (1974) 292;
J.-H. Chen, T.C. Lubensky, D.R. Nelson, Phys. Rev. B 17 (1978) 4274.
[2] I.A. Lawrie, Nucl. Phys. B 200 (1982) 1.
[3] S. Coleman, E. Weinberg, Phys. Rev. D 7 (1973) 1888.
[4] C. Dasgupta, B.I. Halperin, Phys. Rev. Lett. 47 (1981) 1556.
[5] J. Als-Nielsen, J.D. Litster, R.J. Birgeneau, M. Kaplan, C.R. Safinya, A. Lindegaard-Andersen, S. Mathiesen, Phys. Rev. B 22 (1980) 312.
[6] H. Kleinert, Lett. Nuovo Cimento 35 (1982) 405, http://www.physik.fu-berlin.de/~kleinert/97;
H. Kleinert, in: Superflow and Vortex Lines, Gauge Fields in Condensed Matter, Vol. I, World Scientific,
Singapore, 1989, pp. 735742, http://www.physik.fu-berlin.de/~kleinert/b1.
[7] J. Bartholomew, Phys. Rev. B 28 (1983) 5378;
Y. Muneshisa, Phys. Lett. B 155 (1985) 159.
[8] S. Mo, J. Hove, A. Sudbo, The order of the metal to superconductor transition, cond-mat/0109260.
[9] C. Ford, I. Jack, D.R.T. Jones, Nucl. Phys. B 387 (1992) 373.
[10] A.I. Davydychev, J.B. Tausk, Nucl. Phys. B 397 (1993) 123.
[11] A.I. Davydychev, Phys. Rev. D 61 (2000) 087701.
68
[12] J.-P. Tessman, Two-loop Renormierung der skalaren Elektrodynamik, Diplomarbeit, Freie Universitt
Berlin, 1984, written under the supervision of one of the authors (H.K.). A pdf file of the thesis is available
in the internet: http://www.physik.fu-berlin.de/~kleinert/MS-Tessmann.pdf.
[13] M. Kiometzis, H. Kleinert, A.M.J. Schakel, Phys. Rev. 73 (1994) 1975;
I.F. Herbut, Z. Tesanovic, Phys. Rev. Lett. 76 (1996) 4588;
C. de Calan, A.P.C. Malbuisson, F.S. Nogueira, N.F. Svaiter, Phys. Rev. B 59 (1999) 554;
C. de Calan, F.S. Nogueira, Phys. Rev. B 60 (1999) 4255;
B. Bergerhoff, F. Freire, D.F. Litim, S. Lola, C. Wetterich, Phys. Rev. B 53 (1996) 5734.
[14] R. Folk, Yu. Holovatch, J. Phys. A 29 (1996) 3409;
R. Folk, Yu. Holovatch, Critical fluctuations in normal-to-superconducting transition, Lecture at the 1st
Winter Workshop Cooperative Phenomena in Condensed Matter, March, 1998, Pamporovo, Bulgaria,
cond-mat/9807421.
[15] H. Kleinert, Path Integrals in Quantum Mechanics, Statistics and Polymer Physics, World Scientific.
Variational perturbation theory is developed in Chapters 5 and 17.
[16] H. Kleinert, Phys. Rev. D 57 (1998) 2264;
H. Kleinert, Phys. Rev. D 58 (1998) 107702, Addendum;
H. Kleinert, Phys. Rev. D 60 (1999) 085001;
F. Jasch, H. Kleinert, J. Math. Phys. 42 (2001) 52.
[17] H. Kleinert, Phys. Lett. B 434 (1998) 74;
H. Kleinert, Phys. Lett. B 463 (1999) 69;
H. Kleinert, V. Schulte-Frohlinde, J. Phys. A 34 (2001) 1037.
[18] H. Kleinert, B. Van den Bossche, Phys. Rev. E 63 (2001) 056113.
[19] M. Strsser, L.A. Larin, V. Dohm, Nucl. Phys. B 540 (1999) 654.
[20] S. Kolnberger, R. Folk, Phys. Rev. B 41 (1992) 4083.
[21] E. Brzin, D.J. Wallace, K.G. Wilson, Phys. Rev. Let. 29 (1972) 591;
E. Brzin, D.J. Wallace, K.G. Wilson, Phys. Rev. B 7 (1973) 232.
[22] E. Brzin, J.C. Le Guillou, J. Zinn-Justin, in: C. Domb, M.S. Green (Eds.), Phase Transition and Critical
Phenomana, Vol. 6, Academic Press, New York, 1976, p. 125.
[23] D.J. Wallace, R.K.P. Zia, Phys. Lett. A 46 (1973) 261;
D.J. Wallace, R.K.P. Zia, J. Phys. C 7 (1974) 3480.
[24] B. De Witt, Phys. Rev. 162 (1967) 1195.
[25] B. Kastening, Phys. Rev. E 61 (2000) 3501.
[26] M. Bachman, A. Kleinert, A. Pelster, Phys. Rev. D 61 (2000) 085017;
H. Kleinert, A. Pelster, B. Kastening, M. Bachman, Phys. Rev. E 62 (2000) 1537.
[27] H. Kleinert, A. Pelster, B. Van den Bossche, Recursive graphical construction of Feynman diagrams and
their weights in GinzburgLandau theory, hep-ph/0107017.
[28] B. Kastening, Phys. Rev. D 54 (1996) 3965;
B. Kastening, Phys. Rev. D 57 (1998) 3567.
[29] J.M. Chung, B.K. Chung, Phys. Rev. D 56 (1997) 6508;
J.M. Chung, B.K. Chung, J. Korean Phys. Soc. 33 (1998) 643.
[30] J.S. Kang, Phys. Rev. D 10 (1974) 3455.

R 4 corrections to conifolds and G2-holonomy spaces

S. Frolov 1 , A.A. Tseytlin 2
Department of Physics, The Ohio State University, Columbus, OH 43210, USA
Received 29 November 2001; accepted 28 March 2002
Abstract
Motivated by examples that appeared in the context of string theorygauge theory duality, we
consider corrections to supergravity backgrounds induced by higher derivative (R 4 + ) terms in
superstring effective action. We argue that supersymmetric solutions that solve BPS conditions at the
leading (supergravity) order continue to satisfy a 1st order RG-type system of equations with extra
source terms encoding string (or M-theory) corrections. We illustrate this explicitly on the examples
of R 4 corrections to generalized resolved and deformed 6d conifolds and a class of non-compact 7d
spaces with G2 holonomy. Both types of backgrounds get non-trivial modifications which we study
in detail, stressing analogies between the two cases. 2002 Elsevier Science B.V. All rights reserved.
PACS: 11.25.-w; 11.25.Mj
1. Introduction
String theorygauge theory duality implies certain correspondence between perturbative expansions on the both sides of the duality. For example, in the case of the AdS/CFT
duality between type IIB string theory on AdS5 S 5 and N = 4 SYM theory, the and gs
2 N)1/2 and 1/N expansions on the gauge
expansions on the string side translate into (gYM
theory side [1]. In the absence of detailed microscopic understanding of string models in
curved RamondRamond backgrounds, one available source of non-trivial string-theoretic
information is the low-energy spacetime effective action, which, in principle, is universal,
i.e., does not depend on a particular background. The leading correction term in the type II
string effective action is of 4th power in curvature ( 3 R 4 plus additional terms depending
on other supergravity fields as required by supersymmetry).
E-mail address: frolov@mps.ohio-state.edu (S. Frolov).
1 Also at Steklov Mathematical Institute, Moscow.
2 Also at Imperial College, London and Lebedev Physics Institute, Moscow.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 2 4 1 - 9
70
S. Frolov, A.A. Tseytlin / Nuclear Physics B 632 (2002) 69100
Provided the basic supergravity background has regular (and small) curvature, the
-expansion is well-defined and should contain useful information about strong-coupling
expansion on the gauge theory side. How the effect of R 4 corrections on the string theory
side translates to the gauge-theory side was illustrated in [2] on the example of the nearextremal (finite-temperature) version of the AdS/CFT correspondence (the R 4 correction
is related to strong-coupling correction to the entropy of the N = 4 SYM theory, see also
[3] for some related work).
Another example of the important role of the R 4 term in the string theorygauge
theory duality was given in [4]. There we considered the example of duality between string
theory in the fractional D3-brane on conifold background [5] and N = 1 supersymmetric
SU(N + M) SU(N) gauge theory with bifundamental matter chiral multiplets [6]. It was
explained how the -corrections to the radial dependence of the supergravity fields should
translate into higher-order terms in the RG flow equations on the gauge theory side. In
particular, the R 4 term was related to the leading term in strong-coupling expansion of the
anomalous dimension of matter multiplets that enters the NSVZ beta-functions.3
The present work was motivated by [4]. One general question that was left open in [4]
is how the -corrected supergravity equations may be related to the RG flow equations
on the gauge theory side given that the former contain higher derivatives while the latter
only first order ones. As we shall argue in Section 2, in the supersymmetric cases where
the leading supergravity background is a solution of 1st order BPS system4
a = Gab ()
W ()
b
the solutions of the corrected effective action equations should satisfy

a = a W () + J a (, ),
a Gab ()
.
b
(1.1)
The source term J a which encodes information about string corrections should
depend only algebraically on the fields a . There are two steps involved in arriving at this
conclusion. First, one believes that, in a supersymmetric theory, if a leading-order solution
is supersymmetry-preserving, i.e., solves a 1st order system, the same should be true for
its exact -deformation. Indeed, one expects that since the -corrections to superstring
effective action (lP corrections in 11d) should preserve a deformed version of local
supersymmetry, the corrections to a globally-supersymmetric background can be found
from the deformed version of the Killing spinor equation ( + 3 RRR + = 0). The
latter starts with 1st derivative term and contains higher derivatives only in -dependent
terms,
,
. . .) + .
a = a W () + 3 B a (, ,
3 In contrast to N = 2 examples (see, e.g., [7]) where the 1-loop beta-functions are exact (and thus should
be reproduced exactly by the dual supergravity backgrounds), the beta-functions of N = 1 gauge theories with
matter receive non-trivial higher-order corrections.
4 The functions a parametrize the supergravity fields which are assumed to depend on one radial coordinate
t only. This will be the case for the examples in this paper where the metrics possess large global symmetry and
only the radial direction dependence is a non-trivial one.
71
As we shall see, one necessary condition for this to happen is that the R 4 -correction should
vanish when evaluated on the leading BPS background. Second, the fact that the leadingorder background solves the first-order equation a = a W () allows one to express, order
by order in , all derivatives of a in -correction terms B a as algebraic functions of a
only.
We shall explicitly demonstrate how this happens on the two examples: leading 3 R 4
corrections in 10d (or similar lP6 R 4 corrections in 11d) to (i) generalized 6d conifold
metrics [812] and (ii) a class of 7d metrics with G2 holonomy [1315]. Both cases
are Ricci flat metrics preserving part of global supersymmetry (8 and 4 supercharges,
respectively), and thus can be obtained as solutions of 1st order systems. These spaces
have regular curvature so that -expansion is well-defined. A priori, one would expect that
once -corrections are included, one should go back to the Einstein equations corrected
by higher-derivative terms and solve them to find the corrections to the metric (this is,
in fact, what happens in bosonic string theory). As we shall find, in agreement with the
above remarks, the situation in the superstring case is much simpler: the corrected solutions
can be found from the corresponding corrected BPS equations a = a W () + J a (),
3 . We shall solve these equations explicitly to leading order in .
While the standard singular conifold metric does not receive 3 corrections and should
be an exact string solution [4], the scale-dependent (generalized [11,12], resolved [10] and
deformed [9]) conifolds get non-trivial modifications. We shall find that in these cases the
source J a () is expressed in terms of a single function E RRR (6d Euler density).
We shall relate this to the fact that the conifolds are Khler manifolds, and thus (in an
appropriate scheme) their -deformation may be represented as a change of the Khler
potential [1618]. The -corrections to conifold metrics we find provide probably the first
explicit examples of the expected [1618] deformation of 6d CalabiYau metrics by string
corrections.
These results should guide the study of -corrections to more general backgrounds
involving p-form fluxes on conifolds (like the one in [5]) which are of interest from the
point of view of string theorygauge theory duality. There the corresponding string sigma
model will no longer be Khler, but the above remarks about the corrected 1st order form
of the effective equations will still apply. The -deformation of the metric and other
fields will then be of physical significance, being related to higher order corrections to
RG equations on the gauge theory side [4].
Below we shall also consider the deformation under R 4 corrections of some known
examples of non-compact 7d Ricci flat metrics with G2 -holonomy [1315,19,20] (these
may also have potential gauge-theory duality applications as discussed, e.g., in [15,21,
22]). In the tree-level string theory context, the corresponding NS string sigma model has
2d n = 1 world-sheet supersymmetry, and, like in the conifold (Khler, i.e., n = 2) case it is
expected [23,24] to be deformed by -corrections.5 We indeed find that these G2 metrics
are deformed by corrections, implying, in particular, a deformation of the classical
5 Our general discussion of deformation of radially dependent Ricci flat metrics applies also to simpler
examples of hyper-Khler 4d metrics like the EguchiHanson and TaubNUT (or KK monopole) which
correspond to 2d n = 4 supersymmetric (finite [25]) sigma model. Here one finds no corrections coming from
3 R 4 + terms, i.e., in these cases there are no corrections to the 1st order system. In general, one expects
72
G2 -holonomy structure which should go along with the deformation of the supersymmetry
transformation rules and the form of the Killing spinor equation. At the same time, the
number of Killing spinors, i.e., the number of corresponding global supersymmetries
should remain the same, and that should be reflected in the structure of the associated
2d CFT.6
The analysis of the -corrections to the G2 metrics has close analogy with the simpler
one in the conifold case, but here we do not have the advantage of existence of a special
scheme where corrections are parametrized by a deformation of a single functionthe
Khler potential. As in the CY case, the deformed metric should still possess the same
number (four) of global supercharges as the leading-order one, i.e., it should solve the
corresponding -deformed version of the Killing spinor equation. Again, as in the conifold
case (and, more generally, as for KhlerRicci flat spaces [31]), we shall find that there
exists a scheme where the dilaton is not changed from its constant value; also, the R 4 term
evaluated on the leading-order solution will vanish, and thus, as expected, there will be no
shift of the central charge. Considering G2 spaces as solutions of the 11d theory, we shall
find that there exists a scheme in which the corrected solution preserves the direct-product
R 1,3 M 7 form, i.e., there is no generation of a warp factor.
The paper is organized as follows. In Section 2 we shall describe the general approach to
deriving -corrected form of the 1st order system of equations for supersymmetric (BPS)
backgrounds.
In Section 3 we shall illustrate our approach on the example of conifold metrics as
solutions of 10d superstring theory. We shall first review (in Section 3.1) the structure
of R 4 string tree-level corrections to the supergravity action, and prove that for any Ricciflat leading-order solution for which R 4 invariant vanishes, there is a scheme where there is
no correction to the dilaton. This claim is based on certain identity (proved in Appendix A)
between second covariant derivatives of R 3 invariants (which is, in turn, related to the fact
that the 4-point superstring amplitude involving 3 gravitons and 1 dilaton vanishes).
In Section 3.2 we shall determine the corrected form of 1st order equations for the
generalized resolved conifold metric and explain that its structure is indeed consistent with
the expected [1618] deformation of the Khler structure. We shall explicitly compute the
corresponding correction to the Khler potential and thus the metric of this non-compact
that all 1/2 supersymmetric backgrounds of type II string theory (preserving 16 supercharges) should not receive
corrections in an appropriate scheme, while 1/4 and 1/8 supersymmetric backgrounds should get corrections.
6 It would be interesting to compare the sigma model approach with the direct conformal field theory
constructions of G2 CFTs in [2629]. The classical W -type symmetry of G2 sigma models [23] (associated
with covariantly constant 3-form on target space) was promoted to a quantum chiral algebra in [26], i.e., [26]
defined the corresponding general class of 2d CFTs, with particular examples given by particular representations
of this algebra. One expects that starting with the quantum sigma model, there should be a way to define the
corresponding quantum G2 algebra generators (with the definition involving corrections and depending on
a scheme) so that they should form the same chiral quantum algebra to all orders in , as required by the
conformal bootstrap construction [26] (we are grateful to S. Shatashvili for a clarifying discussion of this point).
The situation should be the same for the 6d CY case, where the corresponding algebra was given in [30] (see
also [28]). There the relation between the exact CFT and particular 2d sigma model may be more transparent
since the corrections preserve at least the Khler structure of the target space metric (assuming one uses
a renormalization scheme where n = 2 world-sheet supersymmetry is manifest).
73
CY space. In Section 3.3 we shall briefly discuss analogous computation for the deformed
conifold case.
In Section 4 we shall carry out similar analysis for a class of G2 holonomy spaces
viewed as solutions of 11d supergravity modified by R 4 terms (similar results are found
in 10d tree level string theory framework). In Section 4.1 we shall review the structure of
the R 4 terms in 11d theory and mention correspondence upon dimensional reduction with
the string one-loop R 4 corrections in 10 dimensions. In Section 4.3 will find explicitly the
simple corrected form of the BSGPP solution [13,14]. The analysis of the corrections to the
BGGG [15] metric will be more involved and less explicit (due to the fact that the general
solution for the homogeneous system of 1st order equations describing small perturbations
near the BGGG solution is not known in an analytic form).
2. String corrections to first-order equations

The gravity backgrounds we shall consider in this paper will have non-trivial
dependence on only one radial coordinate, and may be derived as solutions of equations
of motion following from 1-dimensional action obtained by plugging the ansatz for the
metric (and the dilaton) into the string effective action
S = S (0) + S (1) +

1
= dt Gab () a b V () + L(1) (, ,
,
. . .) + .
2
(2.1)
Here a (t) are functions of the radial coordinate that parametrize the metric and the dilaton, V is a scalar potential following from the Einstein term in the action, and L(1) stands
for the contribution of the leading higher-derivative (R 4 + ) correction term. The expansion parameter will be proportional to 3 in d = 10 or lP6 in d = 11. We will be interested
in finding the corrected form of the solutions to leading order in but the discussion that
follows should have a direct generalization to higher-order corrections. The examples of
spaces we shall consider will be non-singular, and thus perturbation theory in dimensionless ratio of and an appropriate power of the curvature scale will be well-defined.
In the cases of interest, V will be expressed in terms of a superpotential (reflecting the
fact that a fraction of global supersymmetry is preserved by the corresponding solutions)
W W
1
V = Gab a b .
2

Then the action S (0) may be rewritten, up to a total derivative, as

W
W
1
b Gbd d ,
S (0) =
dtGab a Gac c
2
and thus solutions of

W
a = Gab c
provide its extrema, and satisfy the usual zero-energy constraint T + V = 0.
(2.2)
(2.3)
(2.4)
74
The string or M-theory higher-derivative corrections to the 10d and 11d supergravity
actions should preserve a deformed version of the original local supersymmetry, so that
the corrected versions of globally-supersymmetric supergravity solutions are expected to
solve a deformed version of the Killing spinor equations, ((m + RRRm... + ) = 0,
see, e.g., [3234]), and thus a deformed version of the first-order BPS equation (2.4).
This suggests a conjecture (which will be justified in the explicit examples considered
below) that the corrected effective action (2.1) may be rewritten in the form similar to (2.3)

W
W
1
S=
dt Gab a Gac c Gac Wc(1) b Gbd d Gbd Wd(1)
2
2
+O ,
(2.5)
(1)
where Wa
by
are some functions of a s and their derivatives. Then (2.4) will be replaced
a = Gab
W
+ J a ,
b
J a Gab Wb(1) .
(2.6)
J a will play the role of sources if one solves these corrected equations (2.6) in perturbation
theory in . It is natural to expect that with higher order correction terms added to (2.1)
and thus to (2.5), one should be able to find the exact solutions to the resulting effective
equations from the system (2.6) with J a replaced by a power series in .
Comparing (2.1) to (2.5) we derive the following condition for (2.5) and (2.6) to be true
(modulo total-derivative terms which we shall always ignore)

(1)
a
ab W
Wa(1) .
L = G
(2.7)
b
One consequence of this relation is that the value of the correction L(1) evaluated on the
leading-order BPS solution of (2.4) must vanish. As we shall see, (2.7) will indeed be
satisfied in all examples we will consider. We expect that this condition should hold for
any supersymmetric solution and any correction preserving supersymmetry. In particular,
the correction to the action should vanish on the leading-order BPS solution.7
The condition (2.7) suggests the general strategy of computing the corrections J a to the
first-order equations (2.6): one should introduce the variables
Qa = a Gab
W
,
b
(2.8)
and express the correction L(1) as a function of a , Qa and derivatives of Qa . Since L(1)
should vanish for Qa = 0, expanding it in powers of Qa and its derivatives we get
(1)
L(1) = Qa Wa(1) () + Qa Qb Wab () + + total derivative terms.
(2.9)
Since we are interested in solving (2.6) to leading order in , all we need to know is the
first Wa(1) () term which we may thus identify with Wa(1) in (2.7).
7 A heuristic reason why this should be the case is that the value of the correction may be regarded as a shift
of energy which should vanish for a supersymmetric solution.
75
(1)
In each particular case discussed below the functions Wa () will be found by

a straightforward (computer-assisted) computation. It is important to note that while the
correction term L(1) in (2.1) depends, in general, not only on a but also on its derivatives,
(1)
the leading term Wa () in (2.9), and, therefore, J a , computed in this way will depend
a
only on since the leading-order equations are first-order, all higher derivatives of a
can be expressed in terms of a using (2.4) order-by-order in .
It is clear that the same should be true also at higher orders in , provided one uses
perturbation theory in . This suggests that the exact (all-order in ) form of the BPS
equations (2.6) may admit an equivalent representation where all terms in the r.h.s. are
simply algebraic functions of a , just as in the case at the leading supergravity order in
(2.4). One is tempted to conjecture that such RG equations which do not involve higherderivative terms (and resulting after a non-trivial rearrangement of the or lP expansion)
should follow from a more fundamental definition of the supersymmetry/BPS condition in
string theory which is not referring to low-energy effective action expansion.
At the leading order in perturbation theory in the corrected equations (2.6) take the
form of first-order equations. It is useful to note that to compute Wa(1) () or J a we will not
need to know the explicit form of the solution to the original equations (2.4): after fixing
a particular ansatz for the metric involving functions a and computing the corresponding
curvature invariants entering (2.1) we will be able to express them in terms of a using
only the general form of the first-order system (2.4).
3. R 4 corrections to conifold metrics in type II superstring theory

In this section we will consider type II string theory on manifolds R 1,3 M 6 , where
is either a resolved or deformed conifold. Our aim will be to determine explicitly how
the metric on these spaces is changed by the leading R 4 correction to the supergravity
action. For definiteness, we will discuss the effect of the tree-level 3 string correction
[16,3537], but all is the same for the similar 1-loop correction (the difference between the
tree-level and 1-loop R 4 invariants in type IIA theory vanishes for the 6d backgrounds we
consider).
Since the conifolds are Ricci flat Khler manifolds [8], one in general expects that (in
an appropriate scheme) the deformation of these metrics will be determined, as for all
CY metrics [1618], by a modification of the Khler potential. We shall start addressing
this problem in the 1st order equation framework of the previous section: this approach is
more universal as it applies to less supersymmetric (non-Khler) spaces like G2 holonomy
spaces discussed later in Section 4.
M6
3.1. 3 R 4 terms in type II superstring effective action

We begin by recalling the structure of 3 corrections to the relevant part of type II
effective actionthe one which depends on the graviton and the dilaton only. Using the
Einstein frame, the leading 3 corrections to the tree-level string effective action implied
by the structure of the GreenSchwarz 4-point massless string scattering amplitude can be
76
written as [35,38,39]

3
1
1
10
2
(1)

, L(1) = e 2 I4 (C),
S = 2 d x g R () + L
2
2
=C
hmnk C
pmnq C
h rsp C
q rsk + 1 C
pqmn C
h rsp C
q rsk ,
hkmn C
I4 (C)
2
(3.1)
where = 8 7/2 gs 2 , = 18 3 (3) and

ij kl = Cij kl 1 2
,
C
ij kl
4

kl
ij
= ik j l jk i l il j k + jl i k .
(3.2)
(3.3)
Here Cij kl is the Weyl tensor, and we omit terms proportional to m m in the definition
ij kl (our leading-order backgrounds will have trivial dilaton field).
of C
Due to the field redefinition ambiguity [35,40], we can assume that all the terms in (3.1)
depend only on the Weyl tensor: the Ricci tensor terms can be expressed in terms of the
dilaton terms using leading-order equations of motion, i.e., changing the scheme). For the
same reason, we shall also assume that in the scheme we use there are no terms proportional
to the dilaton equation of motion m m . We will see, however, that to preserve the
Khler structure of M 6 one will have to change the scheme, adding a certain Ricci tensor
dependent term to the effective action.
Beyond the 4-point level, the above form of the leading 3 corrections is dictated by
the sigma model considerations. As follows from [36,37], there exists a scheme in which
the metric and dilaton dependent terms in the action (reproducing the 4-loop correction to
the beta-function [16]) are given, in the string frame, by

1
S = 2 d 10 x g e2 R + 4()2 + I4 (R) ,
(3.4)
2
1
I4 (R) = R hmnk Rpmnq Rh rsp R q rsk + R hkmn Rpqmn Rh rsp R q rsk .
2
(3.5)
The action that follows from (3.4) upon the transformation gmn e/2 gmn gives the
corresponding action in the Einstein frame that differs from the one in (3.1) by a change
of the scheme (assuming one restores back the presently irrelevant ()2 terms omitted
in (3.2)).
Since the leading-order Ricci-flat backgrounds we are interested in have = 0, we need
3
to keep only the terms e 2 C 4 and C 3 2 in (3.1) as these may give corrections to the
dilaton equation. Explicitly, the relevant terms in (3.1) are found to be

3
L(1) = e 2 I4 (C) + 2E ij i j + O ()2 ,
(3.6)
where I4 (C) is given by (3.5) with Rij kl Cij kl , and Eij is defined by
1
1
Eij = Ci mkl Cjpkq Cl pmq + Ci mkl Cj mpq Ckl pq Cikj l C kmpq C l mpq .
4
2
(3.7)
77
The tensor Eij has the property

g ij Eij = E,
(3.8)
where E 6 6 RRR is the 6d Euler density, which, for a Ricci-flat space and up to
a numerical coefficient, is given by
1
E = Cj mnk C mpqn Cp j k q + Cj kmn C qmn C j k pq .
(3.9)
2
As we will show in Appendix A, for any Einstein (in particular, any Ricci-flat) manifold
the tensor Eij satisfies also the following identity
1
i j Eij = i i E.
6
Thus (3.6) may be rewritten, up to a total derivative term, as

1
(1)
32
2
2
L =e
I4 (C) + E + O () .
3
(3.10)
(3.11)
As a result, the second term can be removed by the following shift of the dilaton in (3.1)
by a local curvature invariant E
1
E.
3
(3.12)
3
Then, in such a scheme the dilaton can get corrections only from the first term e 2 I4 (C)
in (3.11). In fact, there will be no correction to the dilaton coming from the exponential
coupling of to the I4 (C) invariant in (3.6): for all supersymmetric backgrounds we will
study in this paper we will find by direct computation that I4 (C) vanishes. This invariant
should vanish for all special holonomy, supersymmetry-preserving metrics (see also [4])
I4 (C)|special holonomy Ricci flat metrics = 0.
(3.13)
Thus, in the scheme where the term E 2 in (3.11) is absent, the dilaton of these
supersymmetric backgrounds is not modified from its constant leading-order value.8
Therefore, in what follows we will set = 0.
3.2. Corrections to the resolved conifold metric
The metric on resolved conifold [8] that solves the Rmn = 0 equations is a special case
of the following metric [10]
ds62 = e10y du2 + e2y ds52 ,
(3.14)

ds52 = e8w e2 + e2w+2v e21 + e21 + e2w2v e22 + e22 ,
(3.15)
8 For 6d conifolds, these conclusions are in agreement with the previous general statements about the dilaton
shift (and the non-renormalization of central charge) for the CY spaces [31].
78
where
e = d + cos 1 d1 + cos 2 d2 ,
ei = di ,
ei = sin i di ,
and y, w, v are the functions of the radial coordinate u. Since the leading order 10d
background is the direct product R 1,3 M 6 , it is clear from the sigma model considerations
that there should exist a scheme where the -corrected string-frame metric is also a direct
product of R 1,3 and a corrected 6d metric. A priori, the metric need not remain a direct
product in the Einstein frame (as this property is scheme-dependent), so the most general
ansatz should be

2
= e2p ds42 + ds62 ,
ds10
(3.16)
where ds42 is the 4-dimensional Minkowski metric and p(u) is an additional warp factor
function.
As was already mentioned above, corrections to the metric and the dilaton can be
studied separately. Computing
curvature of(3.16), (3.14), we find, as in [10],

the scalar
the corresponding 1d action 18 d 10 x gR S (0)

1
S (0) =
du e8p 5y 2 5w 2 v 2 + 18p 2 + 20y p
2

1
+ e8y 4e2w cosh 2v e12w cosh 4v .
(3.17)
4
This action has the form (2.1) with a = (y, v, w, p) and u playing the role of the
coordinate t. It admits (cf. (2.3) the following superpotential [10]

1
W = e8p+4y e4w + e6w cosh 2v .
4
The corresponding system (2.4) of 1st order equations is then

1
y + e4y e4w + e6w cosh 2v = 0,
5

1
w e4y 2e4w 3e6w cosh 2v = 0,
10
(3.18)
(3.19)
1
p = 0.
v e4y6w sinh 2v = 0,
(3.20)
2
The general solution of these equations depends on two non-trivial integration constants
(the third one is a shift of the radial coordinate) and can be written as
e4v = 1 +
6a 2
,
2
2
e10w = ()e2v ,
3
1
e2y = () 2 e8w ,
p = 0.
9
The coordinate and the original coordinate u are related by

d
3 5y5wv
=
e
,
du
2
(3.21)
79
i.e., the metric (3.14) takes the form [1012]

ds62 = 1 () d 2

1

1
6a 2
1
+ 2 ()e2 + e21 + e21 +
1 + 2 e22 + e22 ,
9
6
6
(3.22)
where
() =
1 + 9a 2/ 2 b6 / 6
,
1 + 6a 2 / 2
0 < ,
06 + 9a 2 04 b6 = 0.
(3.23)
The case of a = b = 0 corresponds to the standard singular conifold [8]; b = 0 gives

the standard resolved conifold metric [10]; a = 0, b = 0 corresponds to the non-singular
generalized conifold [11,12] (a 6d analog of the 4d EguchiHanson metric); a = 0, b = 0
is the generalized resolved conifold metric.
The minimal value of , i.e., 0 = 0 (a, b) 0 is positive and becomes zero when
b = 0. For b = 0 we are to assume that [0, 2) to avoid the conical bolt singularity at
= 0 . The curvature invariants for this metric are regular (unless both a and b are zero
when we get back to the singular conifold metric).
3.2.1. Corrected form of 1st order equations
To find the deformation of this metric under the R 4 correction to the Einstein action
we are to determine the source terms in (2.6) as described in Section 2 (see (2.6)
(2.9)). Computing the R 4 curvature invariant in (3.1) for the metric (3.15) and then using
Eqs. (3.19), (3.20) to express the derivatives of functions in terms of functions themselves
we find that it indeed vanishes as required by (2.7). Its first variation (cf. (2.9)) gives the
sources J a = (Jy , Jv , Jw , Jp ) in (2.6)
Jy = 0,
Jv = 0,
Jp = 0,
8
Jw = e8v42w2y
5

35e2v + 73e6v + 73e10v + 35e14v 18e10w 114e4v+10w 168e8v+10w
114e12v+10w 18e16v+10w + 36e2v+20w + 111e6v+20w + 111e10v+20w

+ 36e14v+20w 18e4v+30w 32e8v+30w 18e12v+30w .
(3.24)
Since the source for the p equation vanishes, we can thus set it to zero p = 0, i.e., the 10d
space retains indeed its direct-product structure. This is a scheme-dependent property: if
we used another scheme, e.g., the one with the Riemann tensor Rij kl instead of the Weyl
tensor Cij kl in (3.6), we would get a non-trivial expression for the warp factor p.
The equation for w is thus the only one that gets corrected,

1 4y 4w
e 2e 3e6w cosh 2v = Jw .
(3.25)
10
This structure of the corrected equations is related to the fact that the resolved conifold is,
in fact, a Khler manifold (see below). Moreover, Jw can be represented as the u-derivative
of the 6d Euler density E (3.9) evaluated on the metric (3.22) (i.e., on (3.14) which solves
w
80
the system (3.19), (3.20))

4 d
E.
15 du
Computing the 6d Euler density for the metric (3.22), we get
Jw =
E=
(3.26)
864
7
14 (6a 2 + 2 )

80a 4b18 + 2592a 8b12 2 + 48a 2b18 2 + 1728a 6b12 4 + 8b18 4
+ 336a 4b12 6 + 168a 4b6 12 + 1296a 8 14 + 24a 2b6 14

+ 216a 6 16 + b6 16 + 10a 4 18 .
(3.27)
It follows that E vanishes (and thus there are no corrections) [4] for the standard (singular)
conifold which corresponds to a = b = 0, and is regular in all other cases (E(0 ) E
E() = 0).
3.2.2. Khler structure
The resulting system of linear equations for y in (3.19), v in (3.20) and (3.25) looks
rather complicated, but it can be solved explicitly if one is guided by the information
provided by the existence of the Khler structure [8] on the resolved conifold. Indeed,
the resolved conifold is a CY manifold with the following Khler potential [8,10]

K = 4 K(t) + a 2 ln 1 + ||2 ,
(3.28)
where is a function of angles and t is related to the radial coordinate r used in [10] as
r 2 = et , i.e., it is related to in (3.22) by

< t < .
e2t = 63 6 + 9a 2 4 b6 ,
(3.29)
Written in terms of K the resolved conifold metric is [10]

ds62 = K dt 2 + e2 + K e21 + e21 + K + a 2 e22 + e22 ,
dK
.
(3.30)
dt
It is clear from the metric that the radii of the spheres at t = (i.e., at = 0 (a, b))
are determined by K () which can be expressed in terms of the constants a and b.
Comparing (3.30) with the general ansatz for the metric (3.14) written in terms of the
t coordinate,

ds 2 = e2y e8w dt 2 + e2 + e2w+2v e21 + e21 + e2w2v e22 + e22 ,
K =
du
= e4(y+w) ,
dt
we get the following relations

2
e10y = K K2 K + a 2 ,
e4v = 1 +
a2
.
K
(3.31)
e20w =
K (K + a 2 )
,
K2
(3.32)
81
Substituting these relations into (3.19), (3.20), we find that the equations for v and y + w
are satisfied identically for any function K, while the equation for w reduces to

1
(3.33)
2t ln K K + a 2 K = 0.
10
This is, of course, equivalent to first integral of the only non-trivial component of
the Einstein equation for the Khler metric (3.30), Rmn = m n ln det g = 0, since
the determinant of (3.30) (defined with respect to the complex coordinates Y, U, ) is
e4t [K (K + a 2 )K ]2 .
Eqs. (3.19), (3.20) corrected by the sources (3.24), (3.25) can be written in the form
d
1
(y + w) + e10w cosh 2v = 0,
dt
2

d
3
1
2 d
w y = E,
dt
2
2
3 dt
dv 1 10w
e
sinh 2v = 0,
dt
2
y y +
4
E.
15
(3.34)
(3.35)
Note that Eq. (3.35) can be easily integrated.

The corrected 6d metric corresponding to the solution of this system may be interpreted
as follows: up to a specific conformal factor (which is related to the redefinition y y,
cf. (3.31)), it is again a Khler metric with the corresponding Khler potential satisfying
the corrected version (3.35) of (3.33):
t t0

1
4
ln K K + a 2 K = E.
2
3
(3.36)
Here t0 is an integration constant which is convenient to fix so that e2t0 = 3/2.

Indeed, the Weyl transformation gmn = e2h gmn of the metric (3.31), i.e., of (3.14), (3.15)
written in terms of the coordinate t, amounts to the redefinition: y = y + h. Choosing
4
E we may thus express y,
w, v in terms of the corrected Khler potential
h = 15
function K using the same relations as in (3.32); then (3.35) reduces to (3.36).
where det g is the determinant
Note that w 32 y is essentially the same as 18 ln det g,
of the rescaled 6d metric. Eq. (3.36) is thus a special case of the general equation for the
deformation of the Khler potential of a CY space due to the string 3 R 4 term [17,18]: the
corrected form of the beta-function equations (in the scheme where the Khler structure of
the metric is preserved) is Rmn + m n H = 0, where H is a series in of local curvature
invariants, starting with the 3 E term. That means [18] that one can prepare such a Khler
potential K( ), that after all -corrections taken into account one is left simply with
Rmn (K0 ) = m n ln det g(K0 ) = 0, where K0 corresponds to the standard CY metric,9
i.e., integrating the above equation,
16
(3.37)
.
3
We shall use the alternative interpretation: one starts with K0 and finds K = K0 + K1 +
as a solution of the corrected string (beta-function) equations.
ln det g(K0 ) = ln det g(K) + k1 E + ,
k1 =
9 The two metricsg(K ) and g(K) are, of course, related in a complicated non-local way.
0
82
As for the meaning of the Weyl rescaling of the metric, it is related to the particular
scheme choice used in (3.1), which is not the same scheme in which the metric retains its
Khler structure (see below).
3.2.3. Corrected form of the metric
Since the curvature of the metric (3.22) is regular (unless a = b = 0 but then E = 0)
and the scale of the curvature is determined by either a or b, the -corrections are small
for small enough /a 2 or /b2 (recall that 0 , see (3.23)), the -expansion is welldefined and thus it makes sense to concentrate on the leading deformation of the metric
(3.22) caused by the first non-zero R 4 correction term.
Integrating (3.36) one more time, we get (we always expand to leading order in )
3
K3 + a 2 K2 = 2
2
t

8

E(t
dt e 1 +
) + c3 ,
3
2t
(3.38)
where c is another integration constant which we choose so that

3
c3 = K3 + a 2 K2 ().
2
Since K () determines the radii of the spheres at t = in the metric (3.31), such
choice of c means that we keep the radii unchanged by the R 4 correction. For = 0
Eq. (3.38) is the equation for the Khler potential of the (generalized) resolved conifold
[1012]: the relation to the metric (3.22), (3.23) is established by setting c = 16 b2 , 2 = 6K
(see also (3.29)). Eq. (3.38) implies
K1 =
K = K0 + K1 ,
1
16
K0 + a 2 K0
9
t
dt e2t E(t ).
(3.39)
Changing from t to coordinate (3.29) and using (3.27) for E we get

K1 =

64
E() E(0 ) ,
2
2
6a +
K1
d
K1 ,
dt
(3.40)
where 0 = 0 (a, b) is defined in (3.23) and

t ()

E() =
dt e2t E(t )
+
2
3 10(6a 2

+ 2)
8b18 2 + 6804a 10 10 + 5670a 8 12 + 3b6 14

+ 9a 4 4 42b12 + 42b6 6 + 5 12 + a 2 24b18 + 63b6 12

+ 54a 6 18b12 2 + 17 14 .
(3.41)
83
This determines the form of the metric (3.30), (3.32) (which depends only on K and K ).
Note that the corrections vanish at both ends of the interval of values of : 0 < .
Eq. (3.40) simplifies in the two special cases: the generalized conifold a = 0, b = 0, where

128 11 b6
8b 18
6 18 ,
K1 = 4
(3.42)
b < ,
3
3
and the standard resolved conifold a = 0, b = 0, where
K1 =
16 2 (12a 2 + 2 )(648a 4 + 126a 2 2 + 7 4 )

3(6a 2 + 2 )
0 < .
(3.43)
These expressions give the explicit form of the corresponding metrics (3.30).
3.2.4. Scheme dependence
Let us now return to the question of the Weyl rescaling of the metric we needed to do
to make its Khler structure manifest. This is related to the issue of scheme dependence.
We have shown that the R 4 corrections (chosen in the specific scheme (3.1)) modify the
6-dimensional metric (3.22) into

4
2
ds6 = 1 E(t) ds62 (K),
(3.44)
15
where ds62 (K) is the Khler metric defined by the Khler potential satisfying (3.38). The
E-dependent conformal factor in front of the metric is scheme-dependent, i.e., it can
be removed by a redefinition of the 10d metric. Indeed, let us consider the following
redefinition of the 10d Einstein-frame metric
gmn gmn + s1 Emn ,
(3.45)
where Emn RRR is the tensor defined in (3.7). This redefinition changes the form of
the R 4 correction in (3.1) by terms of the structure R mn Emn + m n Emn + . Since our
leading-order metric has a direct-product R 1,3 M 6 form, Emn has non-zero components
in M6 directions only. Computing Emn for the resolved conifold metric (3.22), we find that
in this case
1
Eij = gij E, i, j = 1, . . . , 6.
(3.46)
6
As a result, the redefinition (3.45) does not change the 4d components of the 10d metric,
but rescales the 6d components by E. This implies that there is an explicit choice of scheme
in which the R 4 -corrected metric preserves its Khler structure.
3.3. Deformed conifold case
The relevant ansatz for the 6d metric in this case is again parametrized by 3 functions
y, w, q of a radial coordinate u [10,11] (cf. (3.14), (3.15))
ds62 = e10y du2 + e2y ds52 ,
(3.47)
84

ds52 = e8w e2 + e2w+2q g12 + g22 + e2w2q g32 + g42 ,
(3.48)
where [9]
2 + e1
1 e
2 e
,
g2 = 1 ,
g3 = 1 ,
2
2
2
1 + e1
,
g5 = e ,
g4 =
2
1 sin sin 2 d2 + cos d2 ,
2 cos sin 2 d2 sin d2 .
g1 =
(3.49)
Choosing again the 10d Einstein-frame metric as (3.16) we find that the analog of the
Einstein action (3.17) here is

1
(0)
8p
S =
du e 5y 2 5w 2 q 2 + 18p 2 + 20y p
2

1 8y 2w
12w
8w
2
+ e 4e
cosh 2q e
e sinh 2q .
(3.50)
4
This action admits the following superpotential [10]

1
W = e8p+4y e4w cosh 2q + e6w .
4
The corresponding 1st order system (2.4) is

1
y + e4y e4w cosh 2q + e6w = 0,
5

1
w e4y 2e4w cosh 2q 3e6w = 0,
10
(3.51)
(3.52)
1
q e4y+4w sinh 2q = 0,
(3.53)
p = 0.
2
The solution of this system gives the generalized [10,11] deformed conifold metric
depending on the two parameters (b, ) and having regular curvature. For b = 0 it becomes
the metric of the standard deformed conifold [8,9] while for = 0 it gives the metric of
generalized standard conifold (i.e., (3.22), (3.23) with a = 0).
The system (3.52), (3.53) is very similar to (3.19), (3.20) in the resolved conifold case
and the analysis of the R 4 correction to the deformed conifold metric is thus closely parallel
to the one in the previous section, so we will omit the details.
We find again that the R 4 invariant in (3.1) vanishes when evaluated on the solution of
(3.52), (3.53), and that the corrected form of the 1st order equations is (2.6) with the source
terms
4 d
E,
Jq = 0,
Jp = 0,
Jw =
Jy = 0,
(3.54)
15 du
where E is again the 6-dimensional Euler density (3.9).
We conclude that as in the resolved conifold case: (i) this form of the corrected equations
is consistent with the expectation that there should exists a scheme where the metric
preserves its Khler structure; (ii) in the scheme we are using the warp factor p can be
set equal to zero, i.e., the 10d metric preserves its 4 + 6 factorized form.
85
4. R 4 corrections to a class of 7d metrics with G2 holonomy

In this section we shall analyze the corrections induced by the R 4 terms in the effective
action to another class of supersymmetric leading order Ricci flat solutionsspaces
with G2 holonomy found in [13,14] and in [15]. We shall phrase our discussion in the
11d framework, i.e., look at solutions of the R + lP6 R 4 + low-energy effective action
of M-theory, but the analysis of the corresponding 10d string solutions is essentially the
same (we shall comment on this explicitly in Section 4.5). We shall see that G2 spaces
get non-trivial corrections, implying that G2 structure should be deformed (in line with the
deformation of the local supersymmetry transformation rule and thus of the form of the
Killing spinor equation).
We shall derive the corresponding corrected (inhomogeneous) form of 1st order
equations for the functions parametrizing these metrics and discuss their solutions. The
important role will be played by the analysis of solutions of the homogeneous part of the
equations, i.e., the equations for small perturbations near the leading-order solution. In
Section 4.4 we will extend (and correct) the analysis of this homogeneous system given
previously in [15].
4.1. R 4 terms in 11-dimensional theory
Let us start with recalling the structure of the leading R 4 correction terms in the
M-theory effective action (see, e.g., [34,41] and references there)

1
1
11
2
F + + S (1) ,
S = 2 d x g R
2 4! 4
2
(1)
S = b1 T2 d 11 x g (J0 2I2 ),
(4.1)
1
1
J0 = t8 t8 CCCC + E8 + , E8 11 11 CCCC,
4
3!

1
1
2
I2 = E8 + 211 C3 CCCC (CC) + .
4
4
(4.2)
Here C = (Cmnkl ) is the Weyl tensor10 and dots in the two (super)invariants J0 and I2
[41] stand for other (not completely known) terms depending on F4 = dC3 . Also, b1 =
(2)4 32 213 , and T2 = (2)2/3(2 2 )1/3 (membrane tension). The backgrounds
we will be interested in will have F4 = 0 and the direct-product structure R 1,3 M 7 ,
where at the leading order M 7 will have G2 -holonomy. That means, in particular, that
terms depending on F4 and 11 in (4.1) will not contribute, and the relevant part of the 11d
effective action may be written as (cf. (3.1), (3.4))

1
(0)
(1)
S = S + S = 2 d 11 x g R + I4 (C) ,
(4.3)
2
10 We choose the scheme where the curvature tensor is expressed in terms of the Weyl tensor and Ricci tensor
terms, with the latter replaced by the F4 -dependent terms using leading-order field equations.
86
1
I4 (C) = C hmnk Cpmnq Ch rsp C q rsk + C hkmn Cpqmn Ch rsp C q rsk ,
2
(4.4)
where = 31 25 (2)10/3 (2 2)2/3 . The direct reduction of (4.3) to 10 dimensions

should give the 1-loop R 4 + term in type IIA superstring theory. In (Appendix A we
perform a check of consistency of this reduction in the sector of CCC terms.
4.2. General ansatz for the metric
We shall consider a general class of 7d spaces of G2 holonomy studied in [1315,19].
U (1) Z2 . The general ansatz for the
They have the global symmetry SU(2) SU(2)
11d metric deformed by the quantum corrections that preserves this global symmetry is
(cf. (3.14)(3.16)) [15]

2
= e2p ds42 + ds72 ,
ds11
(4.5)
where ds42 is the metric of the 4-dimensional Minkowski space, and

ds72 = e4+4+2 +2 dt 2 + e2 (1 1 )2 + (2 2 )2 + e2 (3 3 )2

+ e2 (1 + 1 )2 + (2 + 2 )2 + e2 (3 + 3 )2 .
(4.6)
Here i and i are the basis one-forms invariant under the left action of the groups SU(2)

and SU
(2), respectively, i.e.,
1 = cos d + sin sin d,
2 = sin d + cos sin d,
3 = d + cos d,
(4.7)
, .
The functions , , , , p depend
with the analogous formulas for i in terms of ,
only on the radial coordinate t.
The warp-factor field p is introduced to account for the fact that the R 4 corrections
could, in principle, destroy the direct product structure of the original R 1,3 M 7
background (this will not happen in a particular scheme we are using but p may be nonvanishing in other schemes).
Computing the R-term in the 11d action (4.3) on this ansatz, one finds the following 1d
action, cf. (3.17) (this is the generalization of the expression derived in [15] to the case of
non-vanishing function p)

(0)
S = dt (T V ),
(4.8)

T = e9p 2 2 + 8 + 2 2 + 4 + 4 + 4 + 4 + 2

+ 36 p + 36 p + 18p + 18 p + 90p 2 ,
1
V = e9p 2e6+2+2 4e4+4+2 + 2e2+6+2 8e4+2+2+2
8

8e2+4+2+2 + 2e2+2+4+2 + e4+2+4 + e4+2+4 .
Comparing to (2.1), here
= (, , , , p),
= d a /dt.
(4.9)
87
This action admits a superpotential (in fact, two simple superpotentials related by
interchanging ). The one that leads to the system of first-order equations obtained
in [15] is

1
W = e9p 2e3++ + 2e+3+ + 2e++2+ e2++2 + e2++2 . (4.10)
2
The variables and the radial coordinate used in [15] are related to a and t as follows
A = e ,
dr = e
B = e ,
2+2+2 +
C = e ,
D = e ,
dt,
(4.11)

ds72 = C 2 (r) dr 2 + A2 (r) (1 1 )2 + (2 2 )2 + D 2 (r)(3 3 )2

+ B 2 (r) (1 + 1 )2 + (2 + 2 )2 + C 2 (r)(3 + 3 )2 .
The superpotential (4.10) expressed in terms of A, B, C, D takes the form

1 2 2
1 2 2
9p
3
3
2
W =e
A BC + AB C + ABCD A C D + B C D .
2
2
The corresponding system of first-order equations (2.4) is the one studied in [15]

1
1 A2 B 2 + D 2
1
dA 1 B 2 A2 + D 2
dB
=
+
=
,
,
dr
4
BCD
A
dr
4
ACD
B

dp
dD A2 + B 2 D 2
C
dC 1 C
=
=
,
= 0.
,
dr
4 B 2 A2
dr
2ABC
dr
(4.12)
(4.13)
(4.14)
The superpotential given in [15]11 is obtained from (4.10) (or (4.13)) by interchanging
(or A B), and setting p = 0.
There are two simple known solutions of the system (4.14): the one found in [13,14],
and another one found in [15]. Any 7-dimensional manifold M 7 with the metric (4.12)
has a particular U (1) isometry acting by the same shift on the angular coordinates
and : + , + . The field C(r) in (4.12) determines the radius of the
associated circle ( the scale of the 3 + 3 direction). The 11-dimensional manifold of
the form R 1,3 M 7 can be reduced along this U (1) direction to a 10d background of
type IIA superstring theory. The solutions found in [1315] correspond after this reduction
to a D6-brane wrapped a three-sphere S 3 of deformed conifold [21,22]. Since the type IIA
dilaton is given by e = C 3/2 , the field C determines also the value of the string coupling
(see [15,22] for details).
4.2.1. Vanishing of correction to warp factor
The contribution of the R 4 correction (4.4) to the 1d action corresponding to the ansatz
(4.5) can be obtained using the method described in Section 2. We have checked that the
combination I4 (C) given by (4.4) vanishes for any solution of the 1st order system (4.14).
Since (4.14) can be used to express any derivative of the metric and thus its curvature as an
11 Note that the factor 2 in Eq. (4.6) and
2 in Eq. (4.8) of [15] should be omitted to get the correct expression.
88
algebraic function of , , , , to check this one does not need to know the explicit form
of the solutions of (4.14).
Since I4 (C) in (4.4) depends only on the Weyl tensor, it is easy to see that then the
(1)
correction W5 in (2.9) corresponding to the warp factor p = 5 in (4.5) also vanishes.
This does not yet imply that the corresponding source component J 5 in (2.6) should also
vanish since the components G5a of the metric are non-trivial. Nevertheless, computing
(1)
all the corrections Wa and the corresponding sources J a , we have found that indeed
J 5 = 0. As a result, we conclude that the R 4 correction does not modify the equation for
p, i.e., we can set p = 0 so that the direct product structure of the background R 1,3 M 7
is preserved at the R 4 level. This is a scheme-dependent statement: if we used the action
with the Riemann tensor Rij kl instead of the Weyl tensor Cij kl in I4 in (4.4), we would get
a non-trivial correction to the warp factor p. The two schemes are, in general, related by
a redefinition similar to (3.45),
gmn gmn + s1 (RRR)mn + s2 RRRgmn + ,
which may rescale the 11d metric by a factor 1 + k1 RRR.
4.3. Corrections to BSGPP solution
We shall first study the R 4 corrections to the G2 holonomy metric found in [13,14].
The manifold has an enhanced SU(2) SU(2) SU(2) Z2 global symmetry which is
achieved by setting = , = in (4.6), i.e., D = A, C = B in (4.12). Then the system
(4.14) reduces to

dB
1
B2
1 2 ,
=
dr
4B
A
dA
1
=
,
dr
2A
D = A,
C = B.
(4.15)
The general solution of this leading-order system (A = A0 , B = B0 ) depends on two

parameters r0 and s
A0 (r; s, r0 ) =
r s,

3/2
B0 (r; s, r0 ) = 31/2 (r s)1/4 (r s)3/2 r0 .
(4.16)
In the special case r0 = 0, s = 0 the metric (4.12) takes the form:

ds72 = d 2 + 2
1 2
.
12

1
(1 1 )2 + (2 2 )2 + (3 3 )2
12

1
(1 + 1 )2 + (2 + 2 )2 + (3 + 3 )2 ,
+
36
(4.17)
89
This space is a cone which is singular at = 0. It follows from the analysis given below
that, like the standard singular conifold, this singular space is not corrected by the R 4 term
in the effective action.12
The general non-singular solution (4.16) can be obtained from the special solution with
s = 0, r0 = 1 by using the translational invariance in r and the invariance of the system
(4.15) under the following scaling transformation: r 2 r, A A, B B, which
corresponds to the rescaling of the 7d metric (4.12) by 2 . Thus, without loss of generality,
one can choose s = 0 and r0 = 1. The resulting metric (4.12) (with 1 r < ) has regular
curvature, and thus computing corrections in expansion in powers of curvature is welldefined.
The corrected analog (2.6) of the system (4.15) is found to be

1
1
dB
dA
B2
=
+ JA ,
=
(4.18)
1 2 + JB ,
dr
2A
dr
4B
A

JA = 211 3A13 A2 3B 2 133A4 414A2B 2 + 301B 4 ,
JB = A1 BJA .
(4.19)
One is supposed to solve this system to leading order in only (since we ignored higher
order in corrections to the effective action). Setting
A = e = A0 ea ,
B = e = B0 eb ,
where A0 and B0 in (4.16) solve (4.15), we find that a and b should satisfy

dA0
1
d
+
+
A0
a = JA0 ,
dr
dr
2A0

B02
dB0
1
B0
d
+
+
a = JB0 ,
B0
1+ 2 b
dr
dr
4B0
A0
2A20
(4.20)
(4.21)
where JA0 , JB0 are given by (4.19) with A, B A0 , B0 . The general solution of the
inhomogeneous system of linear equations (4.21) is given by the sum of its particular
solution and a general solution of the homogeneous system obtained by setting the sources
to zero. Since we know the general two-parameter solution (4.16) of the non-linear
equations (4.15), we can easily obtain the general solution of the corresponding linearized
system by differentiating (4.16) with respect to the parameters s and r0 :
ln A0
ln A0
cs
+ cr 0
= ,
a0 (r) = cs
s
r0
2r
3cr0
cs r 3/2 + 1/2
ln B0
ln B0
+ cr 0
.
b0 (r) = cs
(4.22)
=
3/2
3/2
s
r0
2r r 1
4(r 1)
Here cs and cr0 are arbitrary constants, and we have chosen s = 0 and r0 = 1 here and in
what follows (so that 1 r < ). Evaluating (4.19) on the solution (4.16), one finds
r 3/2 1
301 + 640r 3/2 + 256r 3
,
JB0 =
JA0 .
JA0 =
(4.23)
8
6144r
3 r 3/4
12 In contrast to the case of the singular conifold the invariant E (3.9) is non-zero for this solution.
90
Then the general solution of (4.21) is given by

cs
1
1
301
,
9/2
6
15/2
2r 84r
48r
39936r

3cr0
1
1
5
cs 1/2
1
r +
+
+
b(r) = 3/2
3
2
2r
4
r 1
63r
336r 9/2

823
2107
.
6
79872r
299520r 15/2
a(r) =
(4.24)
The corrections vanish at large r. We can fix the constant cs and cr0 by imposing the
boundary conditions that imply that the short-distance (r 1) limit of the metric is also
not changed by the R 4 correction, namely,
a(1) = 0,
b(1) = 0,
i.e., A(1) = 1,
B(1) = 0.
(4.25)
The second condition b(1) = 0 is a non-trivial one: for generic values of cs and cr0 the
function b(r) has a pole at r = 1, so that, in general, one could only require regularity
of b(r) at this point. A simple computation shows that in the present case the boundary
conditions (4.25) can indeed be satisfied, provided we choose:
cs =
3753
,
46592
cr 0 =
63
.
640
(4.26)
4.4. Corrections to BGGG solution

Next, let us analyze corrections to another class of spaces with G2 holonomy found in
[15] (see also [19]). The corresponding metric is given by a special solution to the system
(4.14) depending only on two (out of possible four) parameters s, r0 13

(r s r0 )(r s + 3r0 )
A0 (r; s, r0 ) =
,
8r0

(r s + r0 )(r s 3r0 )
B0 (r; s, r0 ) =
,
8r0

2r0 (r s 3r0 )(r s + 3r0 )
,
C0 (r; s, r0 ) =
3(r s r0 )(r s + r0 )
r s
D0 (r; s, r0 ) = .
(4.27)
6r0
As in the BSGPP case, the presence of the two free parameters s and r0 simply reflects the
invariance of (4.14) under a shift and a rescaling of the coordinate r. In what follows we
shall set s = 0 and r0 = 1, so that the range of r will be 3 r < .
To find the corrections we need to know the general solution of the homogeneous
system of linear equations describing small perturbations of A, B, C and D around the
13 The parameter r we are using differs from the one in [15] by a factor of 3/2.
0
91
solution (4.27), as well as a particular solution of the inhomogeneous system with sources
corresponding to R 4 correction expanded near the solution (4.27). Unfortunately, in the
present case it does not seem possible to find the exact general solutions of these two
systems. That makes the computation of the corrections much more complicated. Instead
of finding the solutions numerically, we will determine them in the vicinity of r = 3 by
expanding in powers of r 3, and also at large r by expanding in powers of 1/r. We
will then sew the two expansions. This will allows us to get all essential information for
determining the corrections, though the explicit analytic form of the corrected solution in
this case will not be available.
4.4.1. Linearized perturbations near the leading solution
Let us start with the system of linear equations describing small perturbations of A, B, C
and D around the solution (4.27).14 Representing the fields as in (4.20),
A = e = A0 ea ,
B = e = B0 eb ,
C = e = C0 ec ,
D = e = D0 ed ,
(4.28)
we find the system of equations for the perturbations near (4.27) with s = 0, r = 1
(cf. (4.21))
da
5r 2 12r + 3
b
c
d
=
a+
+
,
dr
2r(r 1)(r 3)
2r r + 3 r 3
db
a
5r 2 + 12r + 3
c
d
=
b
+
,
dr
2r 2r(r + 1)(r + 3)
r 3 r +3
4a
4b
dc
=
,
dr
(r 1)(r + 3) (r 1)(r + 3)
dd
2a
2b
c
5r 2 9
=
+

d.
dr
r 3 r + 3 r r(r 2 9)
(4.29)
Differentiating the solution (4.27) with respect to the parameters s and r0 (and then setting
them equal to s = 0, r0 = 1) as in (4.22), we obtain the following special two-parameter
solution to (4.29)15
r +1
r2 + 3
cr 0
,
(r 1)(r + 3)
(r 1)(r + 3)
r 1
r2 + 3
b0 (r) = cs
cr 0
,
(r + 1)(r 3)
(r + 1)(r 3)
8r
r 4 26r 2 + 9
c0 (r) = cs 2
+
c
,
r
0
(r 1)(r 2 9)
(r 2 1)(r 2 9)
cs
d0 (r) = cr0 .
r
a0 (r) = cs
14 The analysis below extends and corrects the previous discussion in [15].
15 Here a (r) c ln A0 + 2c ln A0 , etc.
s s
r0 r
0
0
(4.30)
92
The system of four linear equations (4.29) must have four independent solutions. Since we
do not know how to find the remaining two solutions in a closed form, we shall analyze
them in the vicinity of r = 3 and at large r, and then sew the solutions obtained. Let us
write the (4.29) in the form
d s
= M s q (r) q ,
dr
q = (a, b, c, d).
The matrix M(r) has the following expansion near r = 3
1 0
0
M
0 1
0
+ O(1), M =
M(r) =
0 1 0
r 3
2
0
0
(4.31)
1
0
.
0
2
(4.32)
M has the following eigenvalues q and the corresponding eigenvectors vq

1 = 3,
v1 = (1, 0, 0, 2),
2 = 1,
v2 = (0, 1, 1, 0),
3 = 0,
v3 = (1, 0, 0, 1),
4 = 1,
v4 = (0, 1, 1, 0).
(4.33)
Near r = 3 the solution corresponding to an eigenvalue behaves as (r 3) [1+O(r 3)].

The solutions corresponding to 2 = 1 and 3 = 0 can be obtained from the twoparameter solution (4.30) by choosing cs = 2, cr0 = 2/3 and cs = 6, cr0 = 1, respectively.
Note that, contrary to the claim in [15], the perturbation corresponding to 3 = 0 does not
describe a new deformation of the two-parameter solution (4.27) but corresponds simply
to a change of the parameters s and r0 .
The perturbation corresponding to 1 = 3 is singular at r 3 = 0 (i.e., it does not
represent a smooth deformation of the solution (4.27)).
The fourth linearized solution corresponding to 4 = 1 vanishes at r = 3 and (as we
shall see below) goes to a constant at r = . Since the value of the function C in
(4.12) may be interpreted as in [15] as determining the radius of the M-theory circle, this
perturbation describes a non-trivial deformation of the manifold corresponding to a finite
change in this radius at infinity [21] (cf. [15]). Near = r 3 = 0 it is given by a series
in powers of with the radius of convergence | | < 2. To match the solution to a solution
found at large r we computed it up to the order 20 . The first few terms of the series are

2 2 3 131 4
+
+ O 5 ,
5
15
1890

17 2 29 3 1141 4
+
+ O 5 ,
b4 = +
36
144
14400

13 2 227 3 103 4
+
+ O 5 ,
c4 =
36
2160
4800

4 2 8 3 244 4
+
+ O 5 .
d4 =
15
45
2835
a4 =
(4.34)
Next, let us solve the system (4.29) at large r, where M(r) in (4.31) has the form
5
1
1 1
2
2

1
5
1
M
2 1 1
+ O 2 , M = 2
M(r) =
.
0
0
0
0
r
r
2
2 1 5
93
(4.35)
The eigenvalues and eigenvectors of M are

1 = 6,
v1 = (1, 1, 0, 4),
2 = 3,
v2 = (1, 1, 0, 0),
3 = 1,
v3 = (1, 1, 0, 1),
4 = 0,
v4 = (1, 1, 1, 1).
(4.36)
Here the solution associated to an eigenvalue goes as r [1 + O( 1r )]. The solutions

corresponding to 3 = 1 and 4 = 0 are obtained from the two-parameter solution (4.30)
by imposing the relations cs = 1, cr0 = 0 and cs = 0, cr0 = 1, respectively. The solution
corresponding to 1 = 6 is given by a power series in 1/r. The series converges very
slowly for r > 3, and to match the solutions with the ones found near r = 3 we computed
the terms up to the order 1/r 30 . The first few terms are

1
4
205 754
1
a1 = 6 + 7 + 8 + 9 + O 10 ,
r
r
7r
7r
r

1
4
205 754
1
b1 = 6 7 + 8 9 + O 10 ,
r
r
7r
7r
r

2
1
c1 = 8 + O 10 ,
r
r

4
810
1
d1 = 6 8 + O 10 .
(4.37)
r
7r
r
The 1/r expansion for the solution for 2 = 3 breaks down at the order 1/r 6 where
a 1/r 6 ln r term appears. The coefficients of the ln r terms are proportional to the solution
(4.37), i.e.,

1
2
5
1
324
a1 ln r,
a2 = 3 4 5 + O 7 +
r
r
r
r
5

1
2
5
1
324
b1 ln r,
b2 = 3 4 + 5 + O 7 +
r
r
r
r
5

2
8
1
324
c1 ln r,
c2 = 4 6 + O 8 +
r
r
r
5

6
136
1
324
d1 ln r.
d2 = 4 6 + O 8 +
(4.38)
r
5r
r
5
Now we are ready to determine the large r behavior of the nontrivial solution (4.34)
vanishing at r = 3. At large r it is represented by a superposition of the independent
94
solutions (4.30), (4.37) and (4.38), i.e.,

i
i
(r) + c2 2
(r),
4i = 0i (r) + c1 1
(4.39)
i
i
where 0i (r), 1
(r) and 2
(r) are given by (4.30), (4.37) and (4.38), respectively.
Computing the l.h.s and the r.h.s. of (4.39) at different values of r = + 3, we find the
constants cr0 , cs , c1 and c2
cr0 = 2.00 0.02,

c1 = 7050 70,
cs = 9.31 0.09,
c2 = 54.3 0.5.
(4.40)
These values were obtained by matching the solutions at r = 17/4.

In the process of the analysis of small perturbations near the special solution (4.27) we
have thus shown that the system (4.14) has a three-parameter family of regular solutions,
with the third non-trivial parameter corresponding to the perturbation 4i . The existence of
a 3-parameter family of solutions of the 1st order system (4.14) was also demonstrated in
[19] by using numerical analysis of (4.14).
4.4.2. Solution of the inhomogeneous system
Now we are ready to study the R 4 corrections to the solution (4.27). Computing the
source terms J s in the corrected analog (2.6) of (4.29) from (4.4) expanded (2.9) near
the leading-order solution (4.27), we find the following inhomogeneous system of linear
equations that replaces (4.31)
d s
= Mqs (r) q + J s (r),
dr
(4.41)
where the matrix M(r) is the same as in (4.31), and the sources J s are given by16
J1 =
J2 =
J3 =
J4 =
1024(78 405r 286r 2 549r 3 58r 4 + 57r 5 26r 6 + r 7 )

8
27(r 2 1)
2
1024(78 405r + 286r 549r 3 + 58r 4 + 57r 5 + 26r 6 + r 7 )
8
27(r 2 1)
2048r(577 + 1089r 2 101r 4 + 3r 6 )
8
27(r 2 1)
2048r(233 9r 2 + 13r 4 + 5r 6 )
.
(4.42)
8
27(r 2 1)
To solve the resulting system we follow the same strategy that we used to analyze the small
perturbations near the leading solution: find approximate solutions of (4.41) near r = 3 and
at large r, and then sew the two expansions.
To solve (4.41) near r = 3 we impose the conditions

s ( ) =0 = 0, r 3.
(4.43)
16 We first found J s for an arbitrary solution of (4.14) as functions of s by using the method described in
Section 2, and then computed them on the special solution (4.27) with s = 0 and r0 = 1.
95
As follows from the analysis of small perturbations, this condition fixes the solution
modulo the solution (4.34) of the system (4.31) which also vanishes at = 0. We computed
the expansion in up to the order 20 . The series has a radius of convergence | | < 1, and
its first few terms are given by

617 2
137
+ O 3 + c4 a4 ( ),
2304
3840

5933 2
137
+
+ O 3 + c4 b4 ( ),
b( ) =
2304
82944

11213 2
+ O 3 + c4 c4 ( ),
c( ) =
82944

5189 2
137
+ O 3 + c4 d4 ( ).
d( ) =
(4.44)
2304
34560
Here c4 is an arbitrary constant which is not fixed by the boundary conditions, and
a4 , b4 , c4 , d4 is the solution (4.34) of the homogeneous system. We will later fix the
constant c4 by requiring that the corrected solution also vanishes at r = . This is a natural
requirement since corrections to the BGGG background vanish at large r.
Now let us study the large r region. Since the sources J q go to zero at large r as 1/r 9 ,
there is a solution starting with 1/r 8 . The series converges very slowly for r > 3, and we
computed it up to the order 1/r 40 . The leading terms are

7424
12800
1
+ O 10 ,
a (r) =
189r 8 189r 9
r

7424
12800
1
b (r) =
+
+ O 10 ,
189r 8 189r 9
r

256
1
c (r) = 8 + O 10 ,
9r
r

32000
1
d (r) =
(4.45)
+ O 10 .
189r 8
r
a( ) =
Now it is straightforward to find the large r behavior of the solution (4.44) vanishing
at r = 3. At large r it is given by a sum of the solution (4.45) and a solution of the
homogeneous system (4.31), i.e.,
q
q = (r) + 0 (r) + c1 1 (r) + c2 2 (r),

q
q
q
(r), 0 (r), 1 (r)
(4.46)
q
2 (r)
where
and
are given by (4.45), (4.30), (4.37) and (4.38),
respectively. Computing the l.h.s and the r.h.s. of (4.46) at different values of r = + 3, we
find that the constants cr0 , cs in (4.30) and c1 and c2 are expressed in terms of the constant
c4 as follows
cr0 = 0.10 0.01 (2.0 0.2)c4,
cs = 0.51 0.05 (9 1)c4 ,
c1 = 320 30 (7000 700)c4,
c2 = 2.5 0.3 + (54 5)c4 .
(4.47)
96
These values were obtained by sewing the solutions at = 99/100. Comparing these with
(4.40), we see that they match. We also see that if one chooses c4 0.05 then the coefficient
cr0 vanishes, and, therefore, the corrections to the metric vanish not only at r = 3 but also
at r = . The resulting corrected solution then smoothly interpolates between the same
short-distance and large-distance asymptotic values.
4.5. Corrections to G2 spaces as solutions of 10d superstring theory
It is straightforward to consider the leading R 4 corrections to the 10d backgrounds of
the form R 1,2 M 7 in type II superstring theory, in the same way as this was done for the
conifolds in Section 3. One can check that with the scheme choice corresponding to (3.1)
(i) the direct product structure of the manifold R 1,2 M 7 is again preserved, i.e., the warpfactor is zero; (ii) the metric of M 7 space gets the same corrections as in 11 dimensions;
(iii) as in the conifold case (see Section 3.2) the dilaton is again shifted by 13 E, where
E is the cubic curvature invariant in (3.9) (so that there is a scheme where the dilaton is
unchanged). The explicit expressions for E for the two spaces (4.16) and (4.27) are (s = 0,
r0 = 1)
553 + 915 r 3/2 + 480 r 3 + 320 r 9/2
,
3072 r 15/2
512 (99 + 207 r 2 + 9 r 4 + 5 r 6 )
EBGGG =
.
7
9 (r 2 1)
EBSGP P =
One may ask about the connection between the 10d and 11d results. Earlier in this section
we have shown that the direct product structure of the space R 1,3 M 7 is preserved by the
R 4 corrections to the effective action in 11 dimensions. If we reduce the 11d background
to 10 dimensions along one of the free spatial directions of R 1,3 we get a 10d type IIA
string background of the form R 1,2 M 7 with constant dilaton. Thus, the R 4 corrections
(4.4) in 11 dimensions reduced to 10 dimensions should give the (one-loop) R 4 corrections
in 10 dimensions in the scheme where the dilaton is not modified, i.e., where there is no
E 2 term in (3.11).
Acknowledgements
We are grateful to C. Nunez for a collaboration at an initial stage and many discussions.
We thank G. Papadopoulos and S. Shatashvili for useful correspondence. We also
acknowledge Yu. Obukhov for help with GRG program. The work of S.F. and A.A.T.
was supported by the DOE grant DE-FG02-91ER40690. The work of A.A.T. was also
supported in part by the PPARC SPG 00613, INTAS 99-1590 and CRDF RPI-2108 grants.
Appendix A. Identity for Eij and 11 10 dimensional reduction of R 4 terms
Here we shall first prove the identity (3.10) for the Eij tensor defined in (3.7). We shall
then comment on the role of this identity in checking the consistency of the relation by
97
dimensional reduction between the R 4 terms in the M-theory action and the superstring
effective action, which is implied by local supersymmetry.
Let us first recall the origin of Eij : it appears from the R 4 invariant (3.5) upon
performing a conformal variation of the metric: gij = gij , I4 (R) = 4E ij i j . Since
the structure of I4 (R) is strongly constrained by supersymmetry, the same should be true
for Eij . We suspect that there may be a way to understand the existence of the identity
(3.10) from the fact that the structure of Eij is constrained by the local supersymmetry.
First, it is easy to check that if (3.10) holds for any Einstein manifold, then the same
identity should hold also with the Weyl tensors in (3.7) and (3.9) replaced by the Riemann
ones. Then omitting terms that vanish for Rij = kgij we obtain
1
i j Eij i i E = I1 + I2 + I3 + I4 ,
6
where
1
I1 = i j Rimkl Rjpkq Rlpmq + i j Rimkl Rj mpq Rklpq
4
1
1
+ i i Rj mkl Rmpql Rpj kq + i i Rj mkl Rj mpq Rpqlk ,
2
4
1
I2 = Rimkl Rjpkq i j Rlpmq + Rimkl Rj mpq i j Rklpq
4
1
1
+ Rkij l i j Rkmpq Rlmpq + Rlij k i j Rkmpq Rlmpq ,
2
2
1
1
I3 = j Rimkl i (Rj mpq Rklpq ) i Rj mkl i (Rj mpq Rklpq ),
4
4
(A.1)
(A.2)
(A.3)
(A.4)
I4 = j Rimkl i (Rjpkq Rlpmq ) Rimkl i Rjpkq j Rlpmq

1
1
+ Rimkl i Rj mpq j Rklpq + (Rkij l + Rlij k )j Rkmpq i Rlmpq
4
2
1
1
+ i Rj mkl i Rmpql Rpj kq + i Rj mkl Rmpql i Rpj kq .
(A.5)
2
2
We can show that I1 = 0, I2 = 0, I3 = I4 , thus demonstrating that the r.h.s. of (A.1) is
zero. The proof consists of a repeated use of the Bianchi and cyclic identities.
As is well known, the 11d and 10d supergravities are related by dimensional reduction
along an isometric direction. Since the R 4 invariants (3.1) in 10 and (4.2) in 11 dimensions
should be consistent with the respective supersymmetries, one expects them to be also
related by dimensional reduction. While this is obviously true for the purely metricdependent terms, let us compare the simplest dilaton-dependent terms R 3 2 as they
appear upon dimensional reduction from (4.2) with the similar terms present in the string
effective action (3.1).17 As usual, we shall assume that the 11d metric can be written as

2
2
2
ds11
= e/6 ds10E
+ e4/3 dx11 + Cm dx m ,
17 The 11d R 4 term is related to string one-loop R 4 term whose form is different from the tree-level
invariant (3.1) by the 10 10 RRRR term. However, this difference is irrelevant in the present case as the term
98
2
where and Cm are the dilaton and the RR vector field, respectively, and ds10E
is the
10d metric in the Einstein frame. Since we want to consider the linear dilaton term C 3 2
that follows from the C 4 term in 11 dimensions, we can consider only the components of
the Weyl tensor that have 10-dimensional indices. Modulo field redefinition ambiguity we
can replace the Weyl tensor Cij kl by the Riemann one Rij kl . Using the general relation
between curvatures of the two conformally-equivalent spaces (gij = e2 gij )
i j kl = R i j kl ki (j l j l ) + li (j k j k )
R

gj l i k i k + gj k i l i l

m m ki gj l li gj k ,
we derive the following relation between 11d and 10d Riemann tensors

1 m
n s sm n r rn m s + sn m r ,
12 r
(A.6)
where we have omitted terms quadratic in and terms proportional to the dilaton equation
of motion (i.e., m m ). Comparing (A.6) with (3.2), we conclude that the coefficients
in front of the tensor ( 2 )m nrs differ by the factor 1/3. Therefore, if we start from
the 11-dimensional R 4 term and dimensionally reduce, we get 23 E ij i j term, while
we get 2E ij i j term if we start directly in 10 dimensions (see (3.6)). That would be
a puzzling contradiction if not for the fact that the term E ij i j is, in fact, vanishing onshell, thanks to the non-trivial identity (3.10) proved above. We suspect that there should
be several similar identities related to supersymmetry of the R + R 4 + actions which
may explicitly appear in constructing these actions in the component approach as in [33].
R (11)m nrs = R (10)m nrs +
References
[1] J. Maldacena, The large N limit of superconformal field theories and supergravity, Adv. Theor. Math. Phys. 2
(1998) 231, hep-th/9711200;
S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Gauge theory correlators from non-critical string theory, Phys.
Lett. B 428 (1998) 105, hep-th/9802109;
E. Witten, Anti-de Sitter space and holography, Adv. Theor. Math. Phys. 2 (1998) 253, hep-th/9802150.
[2] S.S. Gubser, I.R. Klebanov, A.A. Tseytlin, Coupling constant dependence in the thermodynamics of N = 4
supersymmetric YangMills theory, Nucl. Phys. B 534 (1998) 202, hep-th/9805156.
[3] J. Pawelczyk, S. Theisen, AdS(5) S(5) black hole metric at O(alpha3), JHEP 9809 (1998) 010, hepth/9808126;
H. Dorn, H.J. Otto, Q anti-Q potential from AdS-CFT relation at T 0: dependence on orientation in
internal space and higher curvature corrections, JHEP 9809 (1998) 021, hep-th/9807093;
K. Landsteiner, String corrections to the HawkingPage phase transition, Mod. Phys. Lett. A 14 (1999) 379,
hep-th/9901143;
T. Harmark, N.A. Obers, Thermodynamics of spinning branes and their dual field theories, JHEP 0001
(2000) 008, hep-th/9910036;
10 10 RRR 2 vanishes due to Bianchi identity, and thus C 3 2 term following from (3.1) coincides with the
corresponding one-loop term up to contributions that vanishing on the leading-order equations of motion.
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
99
T. Harmark, N.A. Obers, Hagedorn behaviour of little string theory from string corrections to NS5-branes,
Phys. Lett. B 485 (2000) 285, hep-th/0005021.
S. Frolov, I.R. Klebanov, A.A. Tseytlin, String corrections to the holographic RG flow of supersymmetric
SU(N ) SU(N + M) gauge theory, Nucl. Phys. B 620 (2002) 84, hep-th/0108106.
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;
I.R. Klebanov, M.J. Strassler, Supergravity and a confining gauge theory: duality cascades and SBresolution of naked singularities, JHEP 0008 (2000) 052, hep-th/0007191.
I.R. Klebanov, N.A. Nekrasov, Gravity duals of fractional branes and logarithmic RG flow, Nucl. Phys.
B 574 (2000) 263, hep-th/9911096;
I.R. Klebanov, E. Witten, Superconformal field theory on three branes at a CalabiYau singularity, Nucl.
Phys. B 536 (1998) 199, hep-th/9807080.
M. Bertolini, P. Di Vecchia, M. Frau, A. Lerda, R. Marotta, I. Pesando, Fractional D-branes and their gauge
duals, JHEP 0102 (2001) 014, hep-th/0011077;
J. Polchinski, N = 2 gauge-gravity duals, Int. J. Mod. Phys. A 16 (2001) 707, hep-th/0011193.
P. Candelas, X.C. de la Ossa, Comments on conifolds, Nucl. Phys. B 342 (1990) 246.
R. Minasian, D. Tsimpis, On the geometry of non-trivially embedded branes, Nucl. Phys. B 572 (2000) 499,
hep-th/9911042;
K. Ohta, T. Yokono, Deformation of conifold and intersecting branes, JHEP 0002 (2000) 023, hepth/9912266.
L.A. Pando Zayas, A.A. Tseytlin, 3-branes on resolved conifold, JHEP 0011 (2000) 028, hep-th/0010088.
G. Papadopoulos, A.A. Tseytlin, Complex geometry of conifolds and 5-brane wrapped on 2-sphere, Class.
Quantum Grav. 18 (2001) 1333, hep-th/0012034.
L.A. Pando Zayas, A.A. Tseytlin, 3-branes on spaces with R S 2 S 3 topology, Phys. Rev. D 63 (2001)
086006, hep-th/0101043.
R.L. Bryant, S. Salamon, 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.
A. Brandhuber, J. Gomis, S.S. Gubser, S. Gukov, Gauge theory at large N and new G2 holonomy metrics,
Nucl. Phys. B 611 (2001) 179, hep-th/0106034.
M.T. Grisaru, A.E. van de Ven, D. Zanon, Four loop beta function for the N = 1 and N = 2 supersymmetric
nonlinear sigma model in two dimensions, Phys. Lett. B 173 (1986) 423;
M.T. Grisaru, A.E. van de Ven, D. Zanon, Two-dimensional supersymmetric sigma models on Ricci flat
Khler manifolds are not finite, Nucl. Phys. B 277 (1986) 388.
M.D. Freeman, C.N. Pope, Beta functions and superstring compactifications, Phys. Lett. B 174 (1986) 48.
D. Nemeschansky, A. Sen, Conformal invariance of supersymmetric sigma models on CalabiYau
manifolds, Phys. Lett. B 178 (1986) 365.
M. Cvetic, G.W. Gibbons, H. Lu, C.N. Pope, Cohomogeneity one manifolds of Spin(7) and G2 holonomy,
hep-th/0108245.
M. Cvetic, G.W. Gibbons, J.T. Liu, H. Lu, C.N. Pope, A new fractional D2-brane, G2 holonomy and
T-duality, hep-th/0106162;
M. Cvetic, G.W. Gibbons, H. Lu, C.N. Pope, M3-branes and G2 manifolds and pseudo-supersymmetry,
hep-th/0106026.
M. Atiyah, J. Maldacena, C. Vafa, An M-theory flop as a large N duality, hep-th/0011256.
J.D. Edelstein, C. Nunez, D6 branes and M-theory geometrical transitions from gauged supergravity,
JHEP 0104 (2001) 028, hep-th/0103167.
P.S. Howe, G. Papadopoulos, A Note on holonomy groups and sigma models, Phys. Lett. B 263 (1991) 230;
P.S. Howe, G. Papadopoulos, Holonomy groups and W symmetries, Commun. Math. Phys. 151 (1993) 467,
hep-th/9202036.
G. Papadopoulos, P.K. Townsend, Compactification of D = 11 supergravity on spaces of exceptional
holonomy, Phys. Lett. B 357 (1995) 300, hep-th/9506150.
L. Alvarez-Gaum, D.Z. Freedman, Geometrical structure and ultraviolet finiteness in the supersymmetric
sigma model, Commun. Math. Phys. 80 (1981) 443;
100
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
P.S. Howe, G. Papadopoulos, Ultraviolet behavior of two-dimensional supersymmetric nonlinear sigma

models, Nucl. Phys. B 289 (1987) 264.
S.L. Shatashvili, C. Vafa, Superstrings and manifold of exceptional holonomy, hep-th/9407025.
P.S. Howe, G. Papadopoulos, P.C. West, Free fermions and extended conformal algebras, Phys. Lett. B 339
(1994) 219, hep-th/9407183.
J.M. Figueroa-OFarrill, A note on the extended superconformal algebras associated with manifolds of
exceptional holonomy, Phys. Lett. B 392 (1997) 77, hep-th/9609113.
T. Eguchi, Y. Sugawara, String theory on G2 manifolds based on Gepner construction, hep-th/0111012;
T. Eguchi, Y. Sugawara, CFT description of string theory compactified on non-compact manifolds with G2
holonomy, Phys. Lett. B 519 (2001) 149, hep-th/0108091;
K. Sugiyama, S. Yamaguchi, Cascade of special holonomy manifolds and heterotic string theory, hepth/0108219;
R. Roiban, J. Walcher, Rational conformal field theories with G2 holonomy, hep-th/0110302;
R. Blumenhagen, V. Braun, Superconformal field theories for compact G2 manifolds, hep-th/0110232.
S. Odake, Extension of N = 2 superconformal algebra and CalabiYau compactification, Mod. Phys. Lett.
A 4 (1989) 557.
A. Sen, Central charge of the Virasoro algebra for supersymmetric sigma models on CalabiYau manifolds,
Phys. Lett. B 178 (1986) 370;
I. Jack, D.R. Jones, The vanishing of the dilaton beta function for the N = 2 supersymmetric sigma model,
Phys. Lett. B 220 (1989) 176.
P. Candelas, M.D. Freeman, C.N. Pope, M.F. Sohnius, K.S. Stelle, Higher order corrections to supersymmetry and compactifications of the heterotic string, Phys. Lett. B 177 (1986) 341.
M.B. Green, S. Sethi, Supersymmetry constraints on type IIB supergravity, Phys. Rev. D 59 (1999) 046006,
hep-th/9808061.
K. Peeters, P. Vanhove, A. Westerberg, Supersymmetric higher-derivative actions in ten and eleven
dimensions, the associated superalgebras and their formulation in superspace, Class. Quantum Grav. 18
(2001) 843, hep-th/0010167.
D.J. Gross, E. Witten, Superstring modifications of Einsteins equations, Nucl. Phys. B 277 (1986) 1.
M.T. Grisaru, D. Zanon, Sigma model superstring corrections to the EinsteinHilbert action, Phys. Lett.
B 177 (1986) 347;
M.D. Freeman, C.N. Pope, M.F. Sohnius, K.S. Stelle, Higher order sigma model counterterms and the
effective action for superstrings, Phys. Lett. B 178 (1986) 199.
Q.-H. Park, D. Zanon, More on sigma model beta functions and low-energy effective actions, Phys. Rev.
D 35 (1987) 4038.
D.J. Gross, J.H. Sloan, The quartic effective action for the heterotic string, Nucl. Phys. B 291 (1987) 41.
R. Myers, Superstring gravity and black holes, Nucl. Phys. B 289 (1987) 701.
A.A. Tseytlin, Ambiguity in the effective action in string theories, Phys. Lett. B 176 (1986) 92.
A.A. Tseytlin, R 4 terms in 11 dimensions and conformal anomaly of (2, 0) theory, Nucl. Phys. B 584 (2000)
233, hep-th/0005072.

5d super-YangMills theory in 4d superspace,

superfield brane operators, and applications to
orbifold GUTs
A. Hebecker
Theory Division, CERN, CH-1211 Geneva 23, Switzerland
Received 21 December 2001; accepted 22 March 2002
Abstract
A manifestly gauge invariant formulation of 5-dimensional supersymmetric YangMills theories
in terms of 4d superfields is derived. It relies on a supersymmetry and gauge-covariant derivative
operator in the x 5 direction. This formulation allows for a systematic study of higher-derivative
operators by combining invariant 4d superfield expressions under the additional restriction of 5d
Lorentz symmetry. In cases where the 5d theory is compactified on a gauge-symmetry-breaking
orbifold, the formalism can be used for a simple discussion of possible brane operators invariant
under the restricted symmetry of the fixed points. This is particularly relevant to recently constructed
grand unified theories in higher dimensions (orbifold GUTs). Several applications, including proton
decay operators and brane-localized mass terms, are discussed. 2002 Elsevier Science B.V. All
rights reserved.
1. Introduction
The standard framework for the discussion of physics above the electroweak scale is
supersymmetric grand unification. Taking the phenomenological success of traditional
gauge coupling unification seriously, the energy range between the GUT scale and the
string (or Planck) scale is the natural domain for higher-dimensional field theories. Indeed,
starting with the proposal of Kawamura [1], a number of very simple and realistic
higher-dimensional GUT models have recently been constructed [212]. Of course,
numerous other interesting ideas that are based on supersymmetry (SUSY) in higher
E-mail address: arthur.hebecker@cern.ch (A. Hebecker).

PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 2 5 3 - 5
102
A. Hebecker / Nuclear Physics B 632 (2002) 101113
dimensions exist, for example, the extra-dimensional SUSY breaking scenarios of [13]
or the intermediate scale unification models of [14].
Both conceptually and for the discussion of low-energy phenomenology, a 4d superfield
description of the higher-dimensional SUSY is desirable. After the early work of [15],
this issue has recently been revived in [16,17]. In the present paper, we develop the
formalism of [16,17] by providing a manifestly gauge-invariant 4d superfield description
of the non-Abelian 5d theory. Our particularly simple formulation relies on the consistent
use of the supersymmetry- and gauge-covariant derivative operator in the x 5 direction.
Furthermore, we find that, starting from conventional 4d super-YangMills (SYM) theory
and introducing x 5 (together with the corresponding gauge connection) as an additional
parameter, the full 5d supersymmetric theory is unambiguously determined. This approach,
which is based on combining invariant 4d superfield expressions under the additional
restriction of 5d Lorentz symmetry, can also be used for a systematic study of higher
derivative terms in the 5d SYM theory.
The presented gauge-covariant formulation allows for a simple discussion of superfield
brane operators invariant under the restricted gauge symmetry of orbifold fixed points.
This is particularly relevant to the extra-dimensional GUT models mentioned earlier,
where the GUT symmetry is broken to the standard model (SM) gauge group by the
boundary conditions of an orbifold compactification. More specifically, SU(5) models in 5
dimensions [13,5] and SO(10) and E6 models in 6 dimensions [7,8,10,11] have recently
been constructed. At first sight, possible couplings in theories of this type are severely
restricted. On the one hand, the large gauge and supersymmetry of the bulk excludes many
couplings. On the other hand, half of the bulk fields are odd under the discrete symmetry
defining the orbifold and therefore vanish at the boundary (or brane), where more couplings
are allowed. Furthermore, certain degrees of freedom are defined as brane fields and can
therefore not participate in bulk interactions. Superfield brane operators lift many of these
restrictions since, due to the presence of the x 5 derivative operator, bulk fields that are odd
under the discrete symmetry (i.e., vanish at the boundary but have non-zero derivative)
can participate in brane-localized interactions. We briefly survey the relevance of these
operators to 5d SUSY GUTs, emphasizing, in particular, proton decay operators and branelocalized mass terms.
The paper is organized as follows. After defining the 5d SYM theory in Section 2,
we explicitly derive its 4d superfield formulation in Section 3. Here, our main result is
the simple and manifestly gauge invariant formulation based on the covariant derivative
in the x 5 direction (Eqs. (21)(24)). Section 4 describes the bulk hypermultiplet while
Section 5 outlines the classification of brane operators using the now available gauge
covariant formalism. Applications to orbifold GUTs are discussed in Section 6, followed
by the conclusions in Section 7.
2. Gauge multiplet in 5 dimensions

We begin by describing the N = 1 (8 supercharges) 5d gauge multiplet [18] (cf.
also [19]) using conventions which are as close as possible to [20]. This will make the
following transition to 4d superfields particularly simple.
103
Capitalized indices M, N, . . . run over 0, 1, 2, 3, 5; lower case indices m, n, . . . run over

0, 1, 2, 3. The metric is M,N = diag(1, 1, 1, 1, 1) and the Dirac matrices can be chosen
as

i 0
0 m
M
,
,
=
(1)
m 0
0 i
where m = (1, ) and m = (1,
). It is convenient to use symplectic Majorana spinors
i , where i = 1, 2 transforms under an SU(2) R-symmetry. The reality condition reads
i = ij C jT ,
(2)
where the 5d charge conjugation matrix C satisfies C M C 1 = ( M )T . We use the explicit

form C = diag(i 2 , i 2 ). Note that lower indices i, j, . . . transform under the 2 of SU(2)-R
and the tensor ( 12 = 21 = 1) can be used to raise or lower indices.
The 5d gauge multiplet contains a vector AM , a real scalar , and an SU(2)-R doublet of
gauginos i . Furthermore, one requires three real auxiliary fields Xa , which form a triplet
of SU(2)-R. These fields are all in the adjoint representation of the gauge group. The SUSY
parameter is a symplectic Majorana spinor i , and the transformation laws are given by1
AM = i i M i ,
(3)
= i i ,

i
i = MN FMN + M DM i + i Xa a j j ,
i

Xa = i a j M DM j + i , i ( a )ij j ,
(4)
(5)
(6)
where MN = 14 [ M , N ] and DM = M + iAM , with appropriate adjoint action of AM

implied. The 5d Lagrangian, invariant under this SUSY, reads

2

1
1
L = 2 tr(FMN )2 tr(DM )2 tr i i M DM i + tr Xa + tr i , i
.
g
2
(7)
3. Formulation in terms of 4d superfields
Orbifold compactifications of the fifth dimension break at least half of the N = 1 5d
SUSY (which corresponds to N = 2 SUSY from the 4d perspective). This is obvious since
the full set of 5d SUSY transformations generates translations in the x 5 direction, and the
latter are not a symmetry of the orbifold. To make the surviving N = 1 4d SUSY manifest,
it is convenient to consider the decomposition of a 5d symplectic Majorana spinor i into
1 The sign of the last term in Eq. (6) differs from [18] due to our opposite sign choice of A in the covariant
M
derivative DM . The signs of the last two terms in Eq. (5) appear to genuinely disagree with [18]. The consistency
of the present equations is most easily confirmed by checking the 4d part of this 5d SUSY, which is worked out
explicitly below.
104
its components (two 4d Weyl spinors L and R ) under the 4d Lorentz group. It reads

(L )
(R )
1
2
=
,
=
,
( R )
( L )

(R ) T
(L ) T
1 =
(8)
,
2 =
.
( L )
( R )
One can now easily work out the 4d Weyl spinor formulation of Eqs. (3)(7). We assume
that the surviving 4d SUSY is generated by a set of parameters i defined by the Weyl
spinor L , with R = 0. For convenience, we explicitly give the transformation rules of the
component fields under this smaller SUSY, using 4d Weyl spinors:
L Am = i L m L + iL m L ,
L A5 = L R L R ,
(9)
(10)
L = i L R iL R ,
L L =
(11)
Fmn L iD5 L + iX L ,

L R = i F5m L m Dm L + i X1 + iX2 L ,

L X1 + iX2 = 2L m Dm R 2i L D5 L + i[, 2L L ],
L X3 = L m Dm L + i L D5 R L m Dm L iL D5 R

+ i , (L R + L R ) ,
mn
(12)
(13)
(14)
(15)
where mn = 14 ( m n n m ).
Now observe [16,18] that the fields Am , L and (X3 D5 ) transform precisely as the
components of a vector superfield in WessZumino (WZ) gauge:

1
V = m Am + i 2 L i 2 L + 2 2 X3 D5 .
(16)
2
Here we use the conventions of [20] for the action of the SUSY transformation L =
on and . Furthermore, in slight deviation from the conventions of [20], we
L Q + L Q
define super-gauge transformations of vector (V ) and fundamental representation chiral
( ) superfields by
e2V e e2V e
and e ,
(17)
where is a chiral superfield depending, in general, on x 5 . For what follows, it is important

to recall that the transformation rules for the WZ-gauge component fields are obtained by
the application of L to V , followed by a super gauge transformation that takes V back to
WZ gauge. This gauge transformation is specified by

= 2 2 m L Am + 2 (2i L L )
(18)
m
m
m
in the y basis (i.e., with the component fields as functions of x 5 and
y = x + i1 ).2
The next essential observation is that the fields ( + iA5 ), (i 2 R ) and (X + iX )
transform as the components of a chiral adjoint superfield in the y basis,

= ( + iA5 ) + 2 (i 2 R ) + 2 X1 + iX2 ,
(19)
105
if, at the same time, this field is defined to transform as

e (5 + )e
(20)
under super gauge transformations. More precisely, this means that the transformation rules
of Eqs. (9)(15) are reproduced by first applying L to and then returning to WZ gauge
by a gauge transformation, Eq. (20), with specified in Eq. (18). This point is essential in
deriving the 4d SUSY transformation rules directly from the 5d SUSY.
Given the transformation rule, Eq. (20), for , it is clear that
5 5 +
(21)
represents a super gauge covariant derivative in the x 5 direction. Thus, given a superfield O, which is in some representation of the gauge group and transforms covariantly
under super gauge transformations, the superfield 5 O transforms covariantly in the same
way. For this it is essential that the Lie-algebra valued field contained in 5 acts on
O as specified by the representation under which O transforms. In particular, for the real
superfield
5 e2V = 5 e2V e2V e2V
the transformation rule is

5 e2V e 5 e2V e .
(22)
(23)
The 5d Lagrangian can now be written in 4d superfield language by combining the

two lowest-dimension invariant operators that can be built from the superfield V and the
covariant derivative2 operator 5 :

1
2 + e2V 5 e2V 2 2 2 .
W
L = 2 tr W W 2 + W
(24)
2g
Here W is the field-strength superfield constructed from V in the usual way. The above
Lagrangian reproduces Eq. (7) up to derivative terms. Note that, to achieve this agreement,
it is not necessary to integrate out the auxiliary fields.
It will prove convenient to define, by analogy to W , the Lie-algebra-valued superfield
Z = e2V 5 e2V . Now the Lagrangian takes the particularly compact form

1
L = 2 tr W W 2 + h.c. + Z 2 2 2 .
(25)
2g
The superfield Z is not Hermitian, but it satisfies the simple condition Z = e2V Ze2V .
Note that it is also possible to turn the argument around and to consider the construction
of the 5d SUSY Lagrangian on the basis of the 4d theory. To achieve this, start with a 4d
real superfield V with the usual gauge transformation properties. The gauge parameter is
the 4d chiral superfield . Now consider both superfields as functions of the additional
parameter x 5 . To be able to take derivatives in the x 5 direction, we are forced to introduce
2 Introducing this covariant derivative is crucial for the manifestly gauge invariant formulation of the nonAbelian Lagrangian, which represents a significant simplification as compared to [16,17].
106
the additional gauge connection , which is a 4d chiral superfield depending on x 5 . The

requirement that 5 = 5 + be a covariant derivative enforces the gauge transformation
property of Eq. (20). It turns out that the two lowest-dimension invariant operators that can
be built from V and 5 add up to the 5d SUSY Lagrangian, Eq. (24), where the relative
normalization of the W 2 and the 52 terms is fixed by the requirement of 5d Lorentz
covariance. As expected on the basis of the 4d N = 1 SUSY and the full 5d Lorentz
covariance, the larger 5d N = 1 SUSY emerges as an additional feature.
Pursuing this line of thinking, it is now straightforward to construct higher-derivative
operators of the 5d SYM theory in a systematic way. One simply has to write down all
4d-SUSY-invariant superfield expressions of a given (higher) dimension and constrain the
coefficients by the requirement of full 5d Lorentz symmetry.
Though motivated by the idea of orbifold compactification, the discussion of this
section was so far restricted to the 5d Lorentz invariant theory. Once 5d Lorentz
invariance is broken and a brane is introduced, the above arguments concerning the
relative normalization of the W 2 and 52 operators in Eq. (24) cease to apply. We adopt
the attitude that, in the orbifold theory, the bulk Lagrangian is nevertheless restricted by
5d Lorentz invariance, while brane localized versions of the W 2 and 52 operators with
unconstrained relative normalization become admissible (see Section 5 for more details).
Strictly speaking, one would have to appeal to supergravity to put the notion of 5d bulk
Lorentz symmetry in the presence of a brane on a firm basis.
4. The hypermultiplet
For completeness, we also present the relevant formulae for the 5d matter multiplet
(the hypermultiplet). It contains an SU(2)-R doublet of scalar fields H i , a Dirac field
and a doublet of auxiliary fields Fi . The transformation laws are (for the ungauged case
see [18])
H i = 2 ij j ,
(26)
M
i
j
i
j
i
= i 2 DM H ij 2 H ij + 2 Fi ,
(27)
M
j
k
Fi = i 2 i DM + 2 i 2i i j k H .
(28)
The off-shell 5d Lagrangian, invariant under this SUSY, reads

i
M DM + F i Fi
L = (DM H )i D M H i i
+ Hi a Xa j H j

i ij H j + h.c. .
+ Hi 2 H i + i 2
(29)
In the 4d superfield formulation, the component fields are arranged in the two chiral 4d
superfields (in the y basis):

H = H 1 + 2 L + 2 F1 + D5 H 2 H 2 ,
(30)

H c = H2 + 2 R + 2 F 2 D5 H1 H1 .
(31)
As in the pure gauge case, the L part of the transformation laws given in Eqs. (26)
(28) follows in the superfield formulation by acting with L on H and H c and then
107
gauge transforming back to WZ gauge. The two superfields gauge transform according to
H e H and H c H c e . The 4d superfield expression for the Lagrangian, Eq. (29),
reads

L = H e2V H + H c e2V H c 2 2 + H c 5 H 2 + h.c. .
(32)
5. Superfield brane operators
An obvious application of the above 4d superfield formalism is the classification of
brane operators of a 5d SYM theory compactified to 4d on an orbifold. The most general
such orbifold is R4 I , where I is an interval parameterized by x 5 = y and limited by
two orbifold fixed points. Without loss of generality, we can discuss a fixed point at y = 0
which is left invariant by a Z2 symmetry of the 5d theory corresponding to the reflection
y y of the original 5d manifold.
Consider a Z2 action on the 5d gauge multiplet given by
V (y) P V (y)P 1
and (y) P (y)P 1 .
(33)
Here, in the simplest case, P is an element of the gauge group, P G (the Z2 acts by inner
automorphism). More generally, the transformation V P V P 1 can be replaced by any
other automorphism of G = Lie(G) (outer automorphism), under the restriction that the
square of this automorphism is the identity.
Eq. (33) is a symmetry of the Lagrangian, Eq. (24), and the sign change of the superfield
is required since enters the Lagrangian in combination with 5 . The fields appearing in
Eq. (25) transform under the Z2 as W (y) P W (y)P 1 and Z(y) P Z(y)P 1 .
Given the Z2 action on V G, the Lie algebra G can be decomposed into its even and
odd components, G = H H , where H generates the subgroup H G preserved by the
orbifolding. Let T a and T a form a basis of H and H , respectively. Then the fields Wa
and Z a are even under the Z2 and can have non-zero values at the fixed point, while the
fields Wa and Z a are odd and vanish at the fixed point. Furthermore, the gauge connection
at the boundary is specified by exp(2V ), which corresponds to a restriction of the gauge
symmetry to H since only V a is non-vanishing. Thus, the lowest-dimension superfields
that can appear in brane operators are
Wa ,
(5 W )a ,
Z a ,
(5 Z)a ,
(34)
where the argument y = 0 is suppressed. Brane operators can be constructed from these
fields as in a 4d SUSY theory, given the restrictions of Lorentz invariance (W is a spinor)
and of the representation content under H .
We can write down the following quadratic operators:

a
1
W
(W )b 2 + h.c.,
O1 = cab
(35)

a
O2 = c2 W (5 W )b 2 + h.c.,
(36)
ab

a
O3 = c3 5 W (5 W )b 2 + h.c.,
(37)
a b
108

O4 = ca4 b Z a Z b 2 2 ,

5
a
b
O5 = cab
Z (5 Z) 2 2 ,

O6 = c6 (5 Z)a (5 Z)b
(38)
(39)
2 2
ab
(40)
where the ci are invariant under H . The operators O1 and O4 have structures that are
already present in the 5d Lagrangian. Nevertheless, as will be discussed in more detail in
the next section, even they give rise to distinctive new effects if they are included in the
action in this brane-localized version. The operators O2 and O5 depend on the non-trivial
condition that invariants c2 and c5 with mixed indices exist. To see that this is possible in
principle, consider a group G = U(1)U(1), broken by outer automorphism of one of the
U(1)s to H = U(1). In this case, both W and Z are singlets and all of the above operators
can be present.
Note also that, if a chiral superfield is localized at the fixed point, this field can
be gauged under the group H . In this case, the kinetic term of this field has the form
exp(2V a T a ) , with T a in the representation appropriate to .
Next, consider the case where hypermultiplets (cf. Section 4) are also present in the
bulk. For a hypermultiplet (H, H c ), a Z2 action consistent with Eq. (33) is given by
H (y) P H (y) and H c (y) H c (y)P ,
(41)
where the prefactor 1 could also be assigned to H instead of H c . Choosing a basis in

representation space where H r is even and H r is odd, the lowest-dimension superfields
that can be used for brane operators at y = 0 are
H r,
(5 H )r ,
H cr ,
r

5 H c .
(42)
They can now be combined among each other, with the fields of Eq. (34), and with chiral
brane fields to form invariant brane operators. Furthermore, chiral brane fields can be
coupled to the fields of Eq. (34). We do not attempt a complete listing of even the lowestdimensional operators but content ourselves with three examples that will be useful in the
next section:

O7 = H c 5 H 2 + h.c.,

O8 = 5 H c H 2 + h.c.,

O9 = e2V Z a T a 2 2 2 + h.c.
1
(43)
(44)
(45)
The operators O7 and O8 are brane-localized versions of an operator appearing in the bulk
hypermultiplet Lagrangian, while O9 couples the gauge fields in the broken directions to
chiral brane fields 1 and 2 in appropriate representations of the subgroup H .
The existence of brane operators of the type discussed in this section can have significant
impact on the low-energy theory emerging below the compactification scale.
109
6. Applications
Recall first the generic setup of 5d orbifold GUTs [16].3 One starts with a gauge
theory on R4 S 1 and restricts the field space by the requirement of symmetry under the
discrete group Z2 Z2 . The S 1 is parameterized by x 5 = y [0, 2R) or, equivalently,
by y = y R/2. The Z2 action is given by Eq. (33), while the Z2 action is given by
an analogous equation where y is replaced by y and P is replaced by P . Choosing the
gauge group SU(5) and representation matrices P = 15 and P = diag(1, 1, 1, 1, 1),
one finds that the surviving symmetries on the P and the P branes are SU(5) and
GSM = SU(3) SU(2) U(1). The low-energy spectrum of the gauge sector is precisely
that of the MSSM since the P reflection removes all zero modes from , and the additional
P reflection removes the zero modes corresponding to X, Y gauge bosons from V . To
solve the doublettriplet splitting problem, the Higgs multiplets have to be localized in the
bulk [1] or on the SM brane [5]. Fermions can be placed on the SU(5) brane [13], on the
SM brane [5], or in the bulk [3,5].
We now want to briefly discuss several possible implications of brane operators
localized at the two fixed points for the low-energy theory derived from the 5d orbifold
GUT.
First, consider proton decay mediated by the X, Y gauge bosons, which have GUT
scale masses because their KaluzaKlein (KK) spectrum does not contain a zero mode.
Naively one would think that this type of process is absent in models where the fermions
are localized on the SM brane [5] because the V components corresponding to the broken
direction vanish at this brane. However, operators of the type O9 in Eq. (45) may be present,
in which case all the usual couplings of SM particles to the 5d analogues of X, Y gauge
bosons may exist. Of course, the coupling strength is now not any more an unambiguous
prediction of the theory. We leave the more detailed investigation of proton decay in this
and other scenarios to a future publication [25].
Next, consider the masses of the Higgs fields. It is one of the most attractive features
of the present models that, after appropriate orbifold projections, a bulk hypermultiplet
(H, H c ) in the 5 of SU(5) gives rise to one doublet chiral superfield. More specifically,
this is realized by using Eq. (41) as it stands for the Z2 transformation and switching the
prefactor 1 from H c to H for the Z2 transformation. However, given the presence of the
operator O8 in Eq. (44) on one of the branes, one obtains a mixing between the doublet
zero mode from H and the massive KK modes of the doublet from H c .
To see this in more detail, consider explicitly the part of the full action that is quadratic in
fermionic fields and does not include derivatives in brane-parallel directions. We integrate
the Lagrangian (Eq. (29) or Eq. (32)) over y [0, R/2] (where R 1/MGUT is the
compactification scale), add cO8 at the P brane, and restrict our attention to the SU(2)
3 Of course, following the early work on symmetry breaking by compactification [21,22], orbifolds [23], and
their interrelation [24], this type of model building was extensively studied in the framework of string theory.
Nevertheless, the present, purely field-theoretic constructions are well motivated as attempts to compare the
wealth of low-energy data with the many possible GUT structures in a way that is as direct and simple as possible.
For a more detailed discussion of the structure of GUT breaking by field-theoretic orbifolding see, e.g., [6].
110
doublet components. The relevant terms read:
S=
R/2

dy 1 c(y) L 5 R + h.c. + .
d x
(46)
This action has to be varied under the constraints that R and 5 L vanish at the
boundaries. The resulting equations of motion are

5 1 c(y) L = 0,
(47)

1 c(y) 5 R = 0.
(48)
Thus, for c = 0, the usual zero mode, L = const, R = 0, is removed. Formally, one finds
a modified zero mode, L [1 c(y)]1 , R = 0. However, this is a highly singular
function which may couple strongly via various higher derivative operators. Therefore,
even though we cannot exclude the existence of a related zero-mode in the UV completion
of the theory, it does not seem to be an unambiguous prediction of the low-energy effective
theory. In fact, when integrating out the auxiliary fields to analyse the scalar part of the
action, one finds singular contributions reminiscent of the infamous (0) terms discussed
in [18]. Note also that solutions arising in the presence of different types of brane-localized
operators have been discussed, e.g., in [26].
Thus, given the limitations of our leading-order, purely field-theoretic analysis, we
conclude that the operator O8 significantly affects the Higgs zero mode. Even if a modified
zero mode should still be present, its strong suppression at the brane may cause problems
for the (necessarily brane-localized) Yukawa interactions. Fortunately, this operator is
protected from quantum corrections and may therefore safely be set to zero at a technical
level. However, one may still be concerned by the fact that, due to the presence of this
operator in the bulk action, there is no obvious symmetry argument excluding the branelocalized version.
Now we turn to the pure gauge sector. The operator O1 of Eq. (35) has already been
extensively discussed in the context of orbifold GUTs. In the case where G = SU(5) and
H = SU(3)SU(2)U(1), it contains three independent pieces which represent the (so far
incalculable) threshold corrections of the model [3]. The logarithmic running of differences
of gauge couplings above the compactification scale [27] can be understood as the running
of the coefficients of these operators [5,9] (see [28] for a recent more detailed analysis).
The operator O4 of Eq. (38) has so far not been used in the construction of orbifold
GUTs. In fact, without the present, fully gauge-covariant superfield formalism it is difficult
to even write this operator down. We now discuss an interesting and, naively, somewhat
mysterious feature of this operator. For simplicity, let c4 = ca b and focus on the quadratic
a b
term that mixes the broken-direction modes of and V :

a

O4 = 4ca b + 5 V b 2 2 + .
(49)
Now consider a situation where both P and P act non-trivially in group space and focus on
fields a and V a which correspond to a Lie algebra generator broken on both boundaries.
In this case a has a zero mode. On the one hand, Eq. (49) appears to imply that this
zero mode is lifted by mixing with the massive KK modes of V a . On the other hand, this
111
zero mode corresponds to the freedom one has in choosing the relative orientation of the
symmetry groups on the two branes as subgroups of G. This modulus can be described
by the Wilson line connecting the two boundaries.4 The latter can clearly take a nontrivial value by having a gauge potential A5 that vanishes near both branes and is nonzero only in the middle of the bulk. Thus, it should be unaffected by brane operators. This
apparent contradiction is resolved by recalling the inhomogeneous gauge transformation
property of . In fact, can always be gauged to zero at the brane. An appropriate gauge
transformation parameter is defined by
5 e = e ,
|brane = 0
(50)
(which is clearly consistent with broken gauge invariance at the brane). This explains why
Eq. (49) cannot be used to argue that a obtains a mass.5
We leave the discussion of other operators and their role in specific models to future,
more phenomenologically oriented work.
7. Conclusions
In this paper, we have given a detailed derivation of the 4d superfield formulation of a 5d
SYM theory compactified on a field-theoretic orbifold. The 4d SUSY has been explicitly
identified as the unbroken part of the larger SUSY of the original 5d theory. An essential
ingredient of our treatment is the gauge and supersymmetry covariant derivative in the
x 5 direction, 5 = 5 + . The Lie-algebra-valued chiral superfield represents the
gauge connection in the x 5 direction. Its action in field space is specified by the usual
Lie algebra action on fields in a representation of the gauge group. The recognition of
the full covariance of 5 and the resulting simplification of the 4d superfield formulation
in the non-Abelian case (cf. Eqs. (21)(24)) represents our main conceptual progress
compared to the earlier treatment of [16]. An immediate consequence is the possibility to
construct higher-derivative operators in the 5d theory by combining terms with 4d covariant
derivatives and 5 under the restriction of full 5d Lorentz invariance.
In our formulation, it is straightforward to write down brane operators localized at
orbifold fixed points where the gauge symmetry is broken to a subgroup of the original
symmetry group (cf. Eqs. (34)(45)). This has particular relevance for the phenomenology
of orbifold GUT models. It is now possible to discuss brane-localized couplings of fields
that vanish at the brane. This is achieved using the 5 derivative of those fields, which is
in general non-zero at the brane.
One implication is the possibility of proton decay mediated by X, Y gauge bosons even
in the case where fermionic matter is localized at a fixed point where the gauge symmetry
4 The vanishing tree-level potential for this degree of freedom is protected by SUSY but can receive radiative
corrections in non-SUSY theories [29].
5 A physical situation where these considerations apply arises if one attempts to construct a 5d SO(10) model
by breaking the group to SU(5)U(1) and SU(5) U(1) on the two boundaries. Although the intersection of
these groups is GSM U(1), the model is plagued by the presence of zero modes [8]. The above considerations
show that this problem is rather fundamental and cannot be overcome by brane operators.
112
is restricted to the standard model group. Another implication is the possibility of a branelocalized mass term mixing the light Higgs doublet with the heavy KaluzaKlein modes
from the 5d hypermultiplet. A further potential area of application, which has not been
discussed here but where brane operators may play an important role, is the low-energy
supersymmetry breaking in models with extra dimensions (see, e.g., [30]).
We hope that the developed framework will prove useful in the detailed phenomenological analysis of different specific orbifold GUT models. It would furthermore be important
to generalize the presented gauge-covariant treatment to SYM theories in more than 5 dimensions.
Acknowledgements
I am very grateful to J. March-Russell for numerous detailed discussions at various
stages of this project. I would also like to thank S. Ferrara, R. Rattazzi and C. Scrucca for
helpful conversations.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
Y. Kawamura, Prog. Theor. Phys. 105 (2001) 999, hep-ph/0012125.

G. Altarelli, F. Feruglio, Phys. Lett. B 511 (2001) 257, hep-ph/0102301.
L. Hall, Y. Nomura, Phys. Rev. D 64 (2001) 055003, hep-ph/0103125.
T. Kawamoto, Y. Kawamura, hep-ph/0106163.
A. Hebecker, J. March-Russell, Nucl. Phys. B 613 (2001) 3, hep-ph/0106166.
A. Hebecker, J. March-Russell, hep-ph/0107039.
T. Asaka, W. Buchmller, L. Covi, hep-ph/0108021.
L.J. Hall, Y. Nomura, T. Okui, D.R. Smith, hep-ph/0108071.
R. Contino, L. Pilo, R. Rattazzi, E. Trincherini, hep-ph/0108102.
C.S. Huang, J. Jiang, T. Li, W. Liao, hep-th/0112046.
N. Haba, T. Kondo, Y. Shimizu, hep-ph/0112132.
A.B. Kobakhidze, Phys. Lett. B 514 (2001) 131, hep-ph/0102323;
R. Barbieri, L.J. Hall, Y. Nomura, hep-th/0107004;
J.A. Bagger, F. Feruglio, F. Zwirner, hep-th/0107128;
T.j. Li, Phys. Lett. B 520 (2001) 377, hep-th/0107136;
T.j. Li, hep-ph/0108120;
T.j. Li, hep-ph/0108238;
T.j. Li, hep-th/0110065;
N. Haba, Y. Shimizu, T. Suzuki, K. Ukai, hep-ph/0107190;
A. Masiero, C.A. Scrucca, M. Serone, L. Silvestrini, hep-ph/0107201;
C.A. Scrucca, M. Serone, JHEP 0110 (2001) 017, hep-th/0107159;
L.J. Hall, H. Murayama, Y. Nomura, hep-th/0107245;
C. Csaki, G.D. Kribs, J. Terning, hep-ph/0107266;
H.C. Cheng, K.T. Matchev, J. Wang, Phys. Lett. B 521 (2001) 308, hep-ph/0107268;
N. Maru, Phys. Lett. B 522 (2001) 117, hep-ph/0108002;
N. Haba, T. Kondo, Y. Shimizu, T. Suzuki, K. Ukai, hep-ph/0108003;
L. Hall, J. March-Russell, T. Okui, D.R. Smith, hep-ph/0108161;
N. Borghini, Y. Gouverneur, M.H. Tytgat, hep-ph/0108094;
M. Chaichian, J.L. Chkareuli, A. Kobakhidze, hep-ph/0108131;
R. Dermisek, A. Mafi, hep-ph/0108139;
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
113
T. Watari, T. Yanagida, Phys. Lett. B 519 (2001) 164, hep-ph/0108152;

Y. Nomura, hep-ph/0108170;
Q. Shafi, Z. Tavartkiladze, hep-ph/0108247;
N. Haba, hep-ph/0110164;
G.A. Diamandis, B.C. Georgalas, P. Kouroumalou, A.B. Lahanas, hep-th/0111046;
H.D. Kim, J.E. Kim, H.M. Lee, hep-ph/0112094.
I. Antoniadis, Phys. Lett. B 246 (1990) 377;
G.R. Dvali, M.A. Shifman, Nucl. Phys. B 504 (1997) 127, hep-th/9611213;
A. Pomarol, M. Quiros, Phys. Lett. B 438 (1998) 255, hep-ph/9806263.
K.R. Dienes, E. Dudas, T. Gherghetta, Phys. Lett. B 436 (1998) 55, hep-ph/9803466;
K.R. Dienes, E. Dudas, T. Gherghetta, Nucl. Phys. B 537 (1999) 47, hep-ph/9806292.
N. Marcus, A. Sagnotti, W. Siegel, Nucl. Phys. B 224 (1983) 159.
N. Arkani-Hamed, T. Gregoire, J. Wacker, hep-th/0101233.
D. Marti, A. Pomarol, Phys. Rev. D 64 (2001) 105025, hep-th/0106256.
E.A. Mirabelli, M.E. Peskin, Phys. Rev. D 58 (1998) 065002, hep-th/9712214.
E. Cremmer, in: S.W. Hawking, M. Rocek (Eds.), Superspace and Supergravity, Proceedings, Cambridge
Univ. Press, 1981;
M. Gunaydin, G. Sierra, P.K. Townsend, Nucl. Phys. B 242 (1984) 244;
A. Salam, E. Sezgin, Supergravities in Diverse Dimensions, World Scientific, Singapore, 1989;
E.R. Sharpe, Nucl. Phys. B 523 (1998) 211, hep-th/9611196.
J. Wess, J. Bagger, Supersymmetry and Supergravity, Princeton Univ. Press, 1983.
J. Scherk, J.H. Schwarz, Phys. Lett. B 82 (1979) 60;
J. Scherk, J.H. Schwarz, Nucl. Phys. B 153 (1979) 61.
Y. Hosotani, Phys. Lett. B 126 (1983) 309;
Y. Hosotani, Ann. Phys. 190 (1989) 233.
L. Dixon, J.A. Harvey, C. Vafa, E. Witten, Nucl. Phys. B 261 (1985) 678;
L. Dixon, J.A. Harvey, C. Vafa, E. Witten, Nucl. Phys. B 274 (1986) 285.
L.E. Ibanez, H. Nilles, F. Quevedo, Phys. Lett. B 187 (1987) 25;
L.E. Ibanez, J. Mas, H. Nilles, F. Quevedo, Nucl. Phys. B 301 (1988) 157.
A. Hebecker, J. March-Russell, work in progress.
Z. Chacko, M.A. Luty, E. Ponton, JHEP 0007 (2000) 036, hep-ph/9909248;
N. Arkani-Hamed et al., Nucl. Phys. B 605 (2001) 81, hep-ph/0102090;
D.E. Kaplan, T.M. Tait, JHEP 0111 (2001) 051, hep-ph/0110126.
Y. Nomura, D.R. Smith, N. Weiner, Nucl. Phys. B 613 (2001) 147, hep-ph/0104041.
L.J. Hall, Y. Nomura, hep-ph/0111068.
M. Kubo, C.S. Lim, H. Yamashita, hep-ph/0111327.
L. Randall, R. Sundrum, Nucl. Phys. B 557 (1999) 79, hep-th/9810155;
D.E. Kaplan, G.D. Kribs, M. Schmaltz, Phys. Rev. D 62 (2000) 035010, hep-ph/9911293;
Z. Chacko, M.A. Luty, A.E. Nelson, E. Ponton, JHEP 0001 (2000) 003, hep-ph/9911323;
M. Luty, R. Sundrum, hep-th/0111231;
A. Anisimov, M. Dine, M. Graesser, S. Thomas, hep-th/0111235.

From super-AdS5 S5 algebra to super-pp-wave

algebra
Machiko Hatsuda a , Kiyoshi Kamimura b , Makoto Sakaguchi a
a Theory Division, High Energy Accelerator Research Organization (KEK), Tsukuba, Ibaraki, 305-0801, Japan
b Department of Physics, Toho University, Funabashi, 274-8510, Japan
Received 27 February 2002; accepted 22 March 2002
Abstract
The isometry algebras of the maximally supersymmetric solutions of IIB supergravity are derived
by the InnWigner contractions of the super-AdS5 S5 algebra. The super-AdS5 S5 algebra
allows introducing two contraction parameters; the one for the Penrose limit to the maximally
supersymmetric pp-wave algebra and the AdS5 S5 radius for the flat limit. The fact that the Jacobi
identity of three supercharges holds irrespectively of these parameters reflects the fact that the number
of supersymmetry is not affected under both contractions. 2002 Elsevier Science B.V. All rights
reserved.
PACS: 11.30.Pb; 11.17.+y; 11.25.-w
Keywords: Superalgebra; Anti-de Sitter; Group contraction
1. Introduction
Recently the supersymmetric pp-wave backgrounds, as well as the anti-de Sitter
(AdS) backgrounds, have been widely studied as supergravity vacua. A maximally
supersymmetric pp-wave solution for the eleven-dimensional supergravity theory was
found in [14], and one for the type IIB supergravity theory was found in [5]. Relations
between the pp-wave background and the AdS background are crucial for the point of
view of the AdS/CFT correspondence [68] where the pp-wave background is recognized
as an approximation of the AdS background. The plane wave space is obtained as the
E-mail addresses: mhatsuda@post.kek.jp (M. Hatsuda), kamimura@ph.sci.toho-u.ac.jp (K. Kamimura),
makoto.sakaguchi@kek.jp (M. Sakaguchi).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 2 5 8 - 4
M. Hatsuda et al. / Nuclear Physics B 632 (2002) 114120
115
limiting space from arbitrary space shown first by Penrose [9], and this idea is extended to
supergravity and superstring theories [1012].1
The maximally supersymmetric solution of the IIB supergravity consist of the pp-wave
metric and the self-dual null homogeneous 5 form flux [5]
ds 2 = 2 dx dx + 42
8
8

i 2 2
i 2
+
x
dx
dx ,
i=1
i=1

F5 = dx dx 1 dx 2 dx 3 dx 4 + dx 5 dx 6 dx 7 dx 8 ,
(1.1)
with a constant dilaton. It has 32 Killing spinors preserving the solution. Recently it was
discussed in detail that the solution (1.1) is obtained as a Penrose limit of the AdS5 S5
background [11,12]. Since the pp-wave solution (1.1) is obtained as a limiting case of the
AdS5 S5 solution by the Penrose limit [9], the isometry algebra of the former should
be also obtained from the latter by some limiting procedure. The Penrose limit is usually
discussed as a mapping between metrics in terms of local coordinates. By using canonical
symmetry generators written in terms of local coordinates, this mapping directly tells the
relation of generators, thus of algebras.
In this paper we examine an InnWigner (IW) contraction [13] of algebra which
is independent of choice of local coordinates. Then the Penrose limit maps not only the
AdS5 S5 metric into pp-wave metric but also the super-AdS5 S5 algebra into the
maximally supersymmetric pp-wave algebra2 as an IW contraction. The IW contraction
from the AdS algebra to the pp-wave algebra for the bosonic case is discussed in Section 2,
and the one for the supersymmetric case is in Section 3.
2. From AdS algebra to pp-wave algebra

The AdSd SDd algebra is given, in terms of dimensionless momenta P s and
rotations Ms, as
[Pa , Pb ] = Mab ,
[Pa , Pb ] = Ma b ,
[Pa , Mbc ] = ab Pc ac Pb ,
[Pa , Mb c ] = a b Pc a c Pb ,
[Mab , Mcd ] = bc Mad + 3 terms,
[Ma b , Mc d ] = b c Ma d + 3 terms,
(2.1)
where a = 0, 1, . . . , d 1 and = d, . . . , D 1 stand for vector indices of AdSd and

SDd , respectively. The symmetry group is isomorphic to SO(d 1, 2) SO(D d + 1)
which has a flat limit to ISO(d 1, 1) ISO(D d) by the following IW contraction [13].
1. Rescale the translation generators P s using the radii of the AdSd and SDd , R and
R , respectively, as
Pa RPa ,
Pa R Pa .
1 This limiting procedure had appeared in the context of the WessZuminoWitten model in [14].
2 We refer the isometry algebra of the pp-wave metric as the pp-wave algebra.
(2.2)
116
2. Then take R and R .

In the limit, P s become the linear momenta and Ms are the Lorentz generators. Taking
large limit of the radii, R and R , corresponds to reducing an AdS S space into a flat
space.
Besides the flat limit, as any other metrics the AdS S metric allows a limit giving
a plane wave metric (Penrose limit) [9]. The Penrose limit can be understood as an IW
contraction of the AdS S algebra into the pp-wave algebra.
1. Define the light cone components of the momenta P and boost generators P as

1
Pi
Pi M0i
Pi =
,
Pi =
P (PD1 P0 ),
(2.3)
Pi
Pi M(D1)i
2
where i = 1, . . . , D 2, i = 1, . . . , d 1, i = d, . . . , D 2.
2. Suppose the plane wave propagates with respect to x+ time. The transverse translation
and boost generators are rescaled with the dimensionless parameter as
1
1
P+ ,
Pi Pi ,
2
3. Then take 0 limit.

P+
Pi
1
P .
i
(2.4)
In the limit, P+ becomes a center, and it can be treated as a constant.

To see them explicitly the AdS S algebra (2.1) is rescaled following to (2.2) and (2.4)
2
[P+ , Pi ] =
Pi ,
2R 2
2
Pi ,
[P+ , Pi ] =
2R 2
2
[P+ , Pi ] = Pi ,
2
2
[P+ , Pi ] = Pi ,
2
2
[Pi , Pj ] = 2 Mij ,
R
[Pi , Pj ] = 2 Mij ,
1
[P , Pi ] =
Pi ,
2R 2
1
[P , Pi ] =
Pi ,
2R 2
1
[P , Pi ] = Pi ,
2
1
[P , Pi ] = Pi ,
2
2
[Pi , Pj ] = 2 Mi j ,
R
[Pi , Pj ] = 2 Mi j ,
1
[Pi , Pj ] = ij (P+ 2 P ),
2
[Pi , Mjk ] = i[
j Pk]
,
1
[Pi , Pj ] = i j (P+ + 2 P ),
2
[Pi , Mjk ] = i[
j Pk]
,
[Mij , Mk l] = k j Mil + 3 terms,
others = 0.
(2.5)
By taking 0 limit, (2.5) becomes the plane wave algebra,

1
1
P , Pi = Pi ,
[P , Pi ] = 4 2 2i Pi ,
Pi , Pj = ij P+ ,
2
2
[Pi , Mjk ] = i[
j Pk]
,
117
[Pi , Mjk ] = i[
j Pk]
,
[Mij , Mk l] = k j Mil + 3 terms,
others = 0,
(2.6)
where
i = ,
2 2R
i =
(2.7)
1
i = .
2 2R
This is the symmetry algebra of the pp-wave metric [15]. P+ is a central element in (2.6)
allowing arbitrary rescaling 3 with .
Furthermore the flat limit is taken by R, R in (2.5) or i 0 in (2.6). Although
this flat limit algebra is different from the direct flat limit from the AdS algebra, the
D-dimensional Poincare group ISO(D 1, 1) can be recovered by supplying spontaneously broken generators consistently.
3. From super-AdS5 S5 algebra to super-pp wave algebra
We extend the previous analysis to the supersymmetric case and show the maximally
supersymmetric IIB pp-wave algebra [5] is obtained by a contraction of the superAdS5 S5 algebra. In this case the Jacobi identities of the superalgebra give a restriction
on the scale parameters R and R in (2.1), as R = R . We start with the super-AdS5 S5
algebra following to the notation of Metsaev and Tseytlin [16]. In 10-dimensional IIB
theory the supersymmetry generators QA are chiral Majorana spinors and are SL(2, R)
doublet (A = 1, 2). The gamma matrices a , (a = 0, 1, . . . , 9) in 10 dimensions are
composed using those of AdS5 , a , (a = 0, 1, . . . , 4), and of S5 , a (a = 5, . . . , 9), as
a = a 1 1 ,
a = 1 a 2 ,
(3.1)
and satisfy
{a b } = 2a b ,
{a , b } = 2ab ,
{a , b } = 2a b ,
a b = ( + + + +, + + + + +),
ab = ( + + + +),
a b = (+ + + + +).
The supercharges QA s are satisfying

1 + 11
,
11 = 0123456789 = 1 1 3 .
2
The charge conjugation matrix C in 10 dimensions is taken as
QA = QA
(3.2)
C = C C i2 ,
where C and
are, respectively, the charge conjugations for AdS5 and
(3.3)
S5
spinors.
3 In [4,5] P is rescaled as 2 P , and P s and P s are e s and es, respectively. Or e s can be recognized
+
+
as 2 e corresponding to P , but this prohibits the flat limit 0 in their superalgebra.
118
The bosonic part of supersymmetric AdS5 S5 algebra is (2.1) with d = 5, D = 10. In

addition to it the odd generators QA satisfy
i
1
[QA , Pa ] = QB a BA ,
[QA , Pa ] = QB a BA ,
2
2
1
1
[QA , Ma b ] = QA a b ,
[QA , Mab ] = QA ab ,
2
2

{Q A , Q B } = AB 2iC C a Pa + 2C C a Pa

+ AB C C ab Mab C C a b Ma b ,
(3.4)
where we omit to write the chiral projection operators.

For the present purpose we rewrite this algebra in terms of 10-dimensional covariant
gamma matrices, s, rather than s. It is convenient to introduce following matrices
01234 = 1 1 i1 0 I,
56789 = 1 1 2 9 J.
(3.5)
The AdS superalgebra (3.4) are rewritten as

1
[QA , Pa ] = QB 0 I a BA ,
2
1
[QA , Ma b ] = QA a b ,
2
ab

a
a b
9 J Ma b ,
{QA
, QB
} = 2iAB C Pa + iAB C 0 I Mab + C
(3.6)
using with the useful relation in the first line

1 + 11
1 + 11
(0 I ) =
(9 J ).
(3.7)
2
2
Associating with the rescaling of P s by R as in (2.2) the dimensionless supercharge
QA is rescaled as
Pa RPa ,
QA R QA .
Pa RPa ,
(3.8)
Corresponding to the rescaling of bosonic generators (2.4) the components of the
supercharges are rescaled with proper weights. The QA is decomposed as
QA = Q+A + QA ,
QA 0,
1
(9 0 ).
2
(3.9)
The supercharges are rescaled as

Q+
1
Q+ ,
Q Q
(3.10)
in the superalgebra of the AdS5 S5 to obtain the well defined Penrose limit.
The super-AdS5 S5 algebra (3.6) after the rescaling of the P s, (2.2) with R = R and
(2.4), and those of Qs, (3.8) and (3.10), becomes
1
[Q+,A , P ] = Q+,B I BA ,
2R
2
[Q,A , P+ ] = Q,B I BA ,
2R
2
[Q+,A , Pi ] = Q,B I i BA ,
2 2R

2
Q+,A , Pi = Q,A i ,
2 2

2
Q+,A , Pi = Q,A i ,
2 2
others = 0,
119
1
[Q,A , Pi ] = Q+,B + I i BA ,
2 2R

1
Q,A , P = Q+,A + i ,
i
2 2
1
[Q,A , Mij ] = Q,A ij ,
2
(3.11)

i 2

{Q+,A , Q+,B } = 2iAB C + P+ AB C + ij I Mij C + i j J Mi j ,
2R
2i
+
Pi AB
{Q+,A , Q,B } = 2iAB C i
2
R

+ i
+ i
I Pi + C
J Pi ,
C
2
2
{Q,A , Q,B } = 2iAB C P

i

+ AB C ij I Mij + C i j J Mi j .
2R
(3.12)
Here we have used relations following from (3.7),

I
1 + 11
1 + 11
= J
.
2
2
(3.13)
It is important that negative power terms of disappear due to the presence of the chiral
and the light cone projections. Therefore, we can take the Penrose limit 0 of the
algebra to obtain
[QA , P ] = QB (I + J )BA ,

1
QA , P = QA + i ,
i
2 2
1
[QA , Mij ] = QA ij ,
2
[QA , Pi ] = QB + i I BA ,
[QA , Pi ] = QB + i J BA ,
(3.14)

{QA , QB } = 2iAB C + P+ + C P + C i Pi

+ 4 2iAB CI i Pi + CJ i Pi

+ 2iAB C I ij Mij + C J i j Mi j ,
(3.15)
where = 1/(2 2R). This is the maximally supersymmetric pp-wave algebra obtained in
[5].
Furthermore the flat limit is taken by R in (3.11) and (3.12) or 0 in (3.15).
120
4. Summary and discussions

In this paper we have derived the super-pp-wave algebra taking with an IW contraction
of the super-AdS5 S5 algebra. It is stressed that the Jacobi identities of this algebra (2.5),
(3.11) and (3.12) hold for any value of and , and the algebra is well defined even in
their zero limits. It explains naturally why the pp-wave and the flat supersymmetry are
maximally supersymmetric as the super-AdS5 S5 . The number of bosonic generators
in AdS5 S5 and pp-wave algebra are same since their algebras are connected by the
contraction.
The relation between the super-AdS background and the super-pp-wave background at
algebraic level is practical to construct mechanical actions for branes in the super-pp-wave
background. Any form field in the latter can be obtained from the former. This approach
manifests symmetries of super-pp-wave systems whose dynamics and spectrum will be
analyzed elegantly.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
J. Kowalski-Glikman, Phys. Lett. B 134 (1984) 194.

C.M. Hull, Phys. Lett. B 139 (1984) 39.
P.T. Chrusciel, J. Kowalski-Glikman, Phys. Lett. B 149 (1984) 107.
J. Figueroa-OFarrill, G. Papadopoulos, JHEP 0108 (2001) 036, hep-th/0105308.
M. Blau, J. Figueroa-OFarrill, C. Hull, G. Papadopoulos, JHEP 0201 (2002) 047, hep-th/0110242.
D. Berenstein, J. Maldacena, H. Nastase, Strings in flat space and pp waves from N = 4 super YangMills,
hep-th/0202021;
N. Itzhaki, I.R. Klebanov, S. Mukhi, PP Wave limit and enhanced supersymmetry in gauge theories, hepth/0202153;
J. Gomis, H. Ooguri, Penrose limit of N = 1 gauge theories, hep-th/0202157;
L.A. Pando Zayas, J. Sonnenschein, On Penrose limits and gauge theories, hep-th/0202186.
R.R. Metsaev, Nucl. Phys. B 625 (2002) 70, hep-th/0112044.
R.R. Metsaev, A.A. Tseytlin, Exactly solvable model of superstring in RamondRamond plane wave
background, hep-th/0202109;
J.G. Russo, A.A. Tseytlin, On solvable models of type IIB superstring in NSNS and RR plane wave
backgrounds, hep-th/0202179.
R. Penrose, Any spacetime has a plane wave as a limit, in: Cahen, Flato (Eds.), Differential Geometry and
Relativity, Reidel, Dordrecht, 1976, pp. 271275.
R. Gven, Phys. Lett. B 482 (2000) 255, hep-th/0005061.
M. Blau, J. Figueroa-OFarrill, C. Hull, G. Papadopoulos, Penrose limit and maximal supersymmetry, hepth/0201081.
M. Blau, J. Figueroa-OFarrill, G. Papadopoulos, Penrose limit, supergravity and brane dynamics, hepth/0202111.
E. Inn, E.P. Wigner, Proc. Natl. Acad. Sci. U. S. A. 39 (1953) 510;
E. Inn, Contraction of Lie group and their representations, in: F. Gursey (Ed.), Group Theoretical Concepts
and Methods in Elementary Particle Physics, Gordon & Breach, New York, 1964.
K. Sfertos, Phys. Lett. B 324 (1994) 335;
D.I. Olive, E. Rabinovici, A. Schwimmer, Phys. Lett. B 321 (1994) 361, hep-th/9311081;
K. Sfertos, A.A. Tseytlin, Nucl. Phys. B 427 (1994) 245, hep-th/9404063;
A.A. Tseytlin, Exact string solutions and duality, hep-th/9407099.
D. Kramer, H. Stephani, E. Herlt, M. MacCallum, The plane-fronted gravitational waves with parallel rays
(pp waves), in: Exact Solutions of Einsteins Field Equations, Cambridge Univ. Press, Cambridge, 1980,
pp. 233236.
R.R. Metsaev, A.A. Tseytlin, Nucl. Phys. B 533 (1998) 109, hep-th/9805028.

Inflation from tsunami-waves

D. Boyanovsky a,b , F.J. Cao b , H.J. de Vega b,a
a Department of Physics and Astronomy, University of Pittsburgh, Pittsburgh PA 15260, USA
b LPTHE, Universit Pierre et Marie Curie (Paris VI) et Denis Diderot (Paris VII), Tour 16, 1er. tage, 4,
Place Jussieu, 75252 Paris cedex 05, France

Received 30 October 2001; accepted 26 March 2002
Abstract
We investigate inflation driven by the evolution of highly excited quantum states within
the framework of out of equilibrium field dynamics. These states are characterized by a nonperturbatively large number of quanta in a band of momenta but with vanishing expectation value
of the scalar field. They represent the situation in which initially a non-perturbatively large energy
density is localized in a band of high energy quantum modes and are coined tsunami-waves. The selfconsistent evolution of this quantum state and the scale factor is studied analytically and numerically.
It is shown that the time evolution of these quantum states lead to two consecutive stages of inflation
under conditions that are the quantum analogue of slow-roll. The evolution of the scale factor during
the first stage has new features that are characteristic of the quantum state. During this initial stage
the quantum fluctuations in the highly excited band build up an effective homogeneous condensate
with a non-perturbatively large amplitude as a consequence of the large number of quanta. The
second stage of inflation is similar to the usual classical chaotic scenario but driven by this effective
condensate. The excited quantum modes are already superhorizon in the first stage and do not
affect the power spectrum of scalar perturbations. Thus, this tsunami quantum state provides a field
theoretical justification for chaotic scenarios driven by a classical homogeneous scalar field of large
amplitude. 2002 Elsevier Science B.V. All rights reserved.
1. Introduction
A wealth of observational evidence from the temperature anisotropies in the cosmic
microwave background favor inflation as the mechanism to produce the primordial density
perturbations [1,2]. Thus, inflationary cosmology emerges as the leading theoretical
framework to explain not only the long-standing shortcomings of standard big bang
E-mail address: devega@lpthe.jussieu.fr (H.J. de Vega).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 2 3 8 - 9
122
D. Boyanovsky et al. / Nuclear Physics B 632 (2002) 121154
cosmology but also to provide a testable paradigm for structure formation [36]. The recent
explosion in the quantity and quality of data on temperature anisotropies elevates inflation
to the realm of an experimentally testable scenario that leads to robust predictions that
withstand detailed scrutiny [1,2].
However, at the level of implementation of an inflationary proposal, the situation is
much less satisfactory. There are very many different models for inflation motivated
by particle physics and most if not all of them invoke one or several scalar fields,
the inflaton(s), whose dynamical evolution in a scalar potential leads to an inflationary
epoch [36]. The inflaton field is a scalar field that provides an effective description for the
fields in the grand unified theories. Furthermore, there is the tantalizing prospect of learning
some aspects of the inflationary potential (at least the part of the potential associated
with the last few e-folds) through the temperature anisotropies of the cosmic microwave
background [7].
Most treatments of inflation study the evolution of the inflaton via the classical
equations of motion in the scalar potential and the effect of quantum fluctuations is
typically neglected in the dynamics of the inflaton. Furthermore, since inflation redshifts
inhomogeneities very fast, the classical evolution is studied in terms of a homogeneous
classical scalar field. The quantum field theory interpretation is that this classical,
homogeneous field configuration is the expectation value of a quantum field operator
in a translational-invariant quantum state. While the evolution of this coherent field
configuration (the expectation value or order parameter) is studied via classical equations
of motion, quantum fluctuations of the scalar field around this expectation value are treated
perturbatively and are interpreted as the seeds for scalar density perturbations of the
metric [36].
A fairly broad catalog of inflationary models based on scalar field dynamics labels these
either as small field or large field [7]. In the small field category the scalar field begins
its evolution with an initial value very near the origin of the scalar potential and rolls down
towards larger values, an example is new inflation [3,4]. In the large field category, the
scalar field begins very high up in the potential hill and rolls down towards smaller values,
an example is chaotic inflation [3,4].
It is only recently that the influence of quantum fluctuations of the scalar fields in the
dynamical evolution of matter and geometry has been studied self-consistently, mainly
associated with the dynamics of non-equilibrium phase transitions [8,9] in models that fall,
broadly, in the small field category. The conclusion of these studies is that a treatment
of the quantum fluctuations that couple self-consistently to the dynamics of the metric
provides a solid quantum field theoretical framework that justifies microscopically the
picture based on classical inflation. At the same time these studies provide a deeper
understanding of the quantum as well as classical aspects of inflation and inflationary
perturbations. They clearly reveal the classicalization of initial quantum fluctuations [8,9],
and furnish a microscopic explanation (and derivation) of the effective, homogeneous
classical inflaton [9].
The purpose of this article is to provide a quantum treatment of models whose classical
counterpart are large field models. The classical description in these models begins with
a homogeneous inflaton scalar with very large amplitude MPl [3,4,6], i.e., very high
up in the scalar potential well. The question of how is this initial condition achieved is
123
typically answered in terms of a probabilistic distribution of initial conditions [3,4], as is

the case for chaotic inflation.
Instead, in this article we study the dynamics that results from the evolution of a
quantum state which drives the dynamics of the scale factor through the expectation value
of the energymomentum tensor.
The initial state the we consider is a squeezed state with a large number of particles
distributed in a shell in momentum. While squeezed states had been studied in quantum
optics [10] and also in cosmology [11] the state considered in this article is similar to
that invoked to model the evolution of highly excited initial quantum states to describe
the dynamics of heavy ion collisions. These states are characterized by a large population
of quanta localized in momentum shells and had been coined tsunami-waves in [1214].
The time evolution of these states leads to the formation of a non-equilibrium plasma and
features properties similar to those expected to occur in the cooling and expansion of a
quark gluon plasma after the collision [15].
The goals of this article
We here adapt the ideas and concepts in Refs. [1214] to study the self-consistent
dynamics of the metric and the evolution of a highly excited quantum state with the goal
of providing a quantum description of large field inflationary models without assuming an
expectation value for the scalar field. We introduce novel quantum states, which are the
cosmological counterpart of the tsunami-waves introduced in [1214] with the following
properties:
Pure states: the states under consideration, defined by the wave-functional of the form
(18) are pure states. Other proposals involving a mixed state density matrix as initial
state are discussed in Sections 3, 4 and Appendix A.
Vanishing expectation value of the scalar field: unlike most models of classical chaotic
inflation in which the scalar field obtains an expectation value, taken as a classical field,
the expectation value of the scalar field in the tsunami-wave states given by the wavefunctional (18) vanishes. [See Section 4 for a non-zero expectation value of the scalar
field.]
Highly excited initial modes: the tsunami-wave state described by Eq. (18) with the
covariance kernel given by Eqs. (23), (30) describes a state for which the modes inside
a band are occupied with a non-perturbatively large O(1/) number of (adiabatic)
quanta. This requires the use of non-perturbative methods as the large N approach and
makes the states considered to be very far from the vacuum. We remark that very high
energy modes, those that will become superhorizon during the last 10 or so e-folds,
hence, are of cosmological importance today, must be in the vacuum state so as not to
lead to a large amplitude of scalar density perturbations [3,6,8].
This type of quantum states is clearly a novel concept, it presents an alternative to typical
inflationary scenarios that invoke the dynamics of a classical scalar field which in most
cases ignore the quantum dynamics.
124
The quantum nature of the tsunami states is due to the coherence between different
modes and gives rise to dynamical consequences (see [14]). In the present case the system
is effectively classical only after the redshift has assembled the modes into a zero-mode
effective condensate.
The tsunami-wave states described above are the simplest states and will be the focus of
our study. These states can be generalized to describe mixed-state density matrices and to
also allow for an expectation value of the scalar field. These generalizations are described
in Sections 3.3 and 4 and in Appendix A and are found to lead qualitatively the same
features revealed by the simpler pure states.
We establish the conditions under which such quantum state leads to inflationary
dynamics and study in detail the self-consistent evolution of this quantum state and the
spacetime metric.
We emphasize that we are not proposing here yet a new model of inflation. Instead
we focus on inflation driven by the evolution of a quantum state, within the framework
of familiar models based on scalar fields with typical quartic potentials. This is in contrast
with the usual approach in which the dynamics is driven by the evolution of a homogeneous
classical field of large amplitude.
Brief summary
We find that inflation occurs under fairly general conditions that are the quantum
equivalent of slow-roll. There are two consecutive but distinct inflationary stages: the
first one is completely determined by the quantum features of the state. Even when
the expectation value of the scalar field vanishes at all times in this quantum state,
the dynamics of the first stage gives rise to the emergence of an effective classical
homogeneous condensate. The amplitude of the effective condensate is non-perturbatively
large as a consequence of the non-perturbatively large number of quanta in the band of
excited wavevectors. The second stage is similar to the familiar classical chaotic scenario,
and can be interpreted as being driven by the dynamics of the effective homogeneous
condensate. The band of excited quantum modes, if not superhorizon initially they
cross the horizon during the first stage of inflation, hence they do not modify the
power spectrum of scalar density perturbations on wavelengths that are of cosmological
relevance today. Actually, in the explicit examples worked out here, the excited modes
are initially superhorizon due to the generalized slow-roll condition. Therefore, in a
very well defined manner, tsunami quantum states provide a quantum field theoretical
justification, a microscopic basis, for chaotic inflation, explaining the classical dynamics
of the homogeneous scalar field.
In Section 2 we introduce the quantum state, obtain the renormalized equations of
motion for the self-consistent evolution of the quantum state and the scale factor. In
Section 3 we provide detailed analytic and numerical studies of the evolution and highlight
the different inflationary stages. In Section 4 we discuss generalized scenarios. The
summary of results is presented in the conclusions. We derive the equations of motion
for mixed states in Appendix A.
125
2. Initial state and equations of motion

As emphasized in the introduction, while most works on inflation treat the dynamics of
the inflaton field at the classical level, we use a quantum description of the inflaton.
We focus on the possibility of inflation through the dynamical quantum evolution of a
highly excited initial state with large energy density. Consistently with inflation at a scale
well below the Planck energy, we treat the inflaton field describing the matter as a quantum
field whereas gravity is treated semiclassically.
The dynamics of the classical spacetime metric is determined by the Einstein equations
with a source term given by the expectation value of the energymomentum tensor of the
quantum inflaton field. The quantum field evolution is calculated in the resulting metric.
Hence, we solve self-consistently the coupled evolution equations for the classical
metric and the quantum inflaton field.
We assume that the universe is homogeneous, isotropic and spatially flat, thus it is
described by the metric,
ds 2 = dt 2 a 2 (t) d x 2 .
(1)
Anticipating the need for a non-perturbative treatment of the evolution of the quantum

state, we consider an inflaton model with an N -component scalar inflaton field (x)
with quartic self-coupling. We then invoke the large N limit as a non-perturbative tool
to study the dynamics [8,9,12,13,16]. This choice is not only motivated by the necessity of
a consistent non-perturbative treatment but also because any grand unified field theory will
contain a large number of scalar fields, thus justifying a large N limit on more physical
grounds.
The matter action and Lagrangian density are given by

2
(x))

1 2
1 (
4
4
3

V (x) ,
S[] = d x Lm = d x a (t) (x)
(2)
2
2 a 2 (t)
2 2 1
m2 2
=

+ R 2 ,
V ()
(3)
+
2
8N
2
and we will consider m2 > 0, postponing the discussion of the case m2 < 0 to subsequent
work. Here R(t) stands for the scalar curvature

a 2 (t)
a(t)
+
R(t) = 6
(4)
.
a(t) a 2 (t)
The -coupling of 2 (x) to the scalar curvature R(t) has been included in the Lagrangian
since it is necessary for the renormalizability of the theory.
The discussion of the alternative inflationary mechanism that we are proposing and the
physical description of the quantum states becomes more clear in conformal time
t
dt
a(t )
in terms of which the metric is conformal to that in Minkowski spacetime

ds 2 = a 2 (T ) dT 2 dx2 .
T =
(5)
(6)
126
We introduce the conformally rescaled field

x)
(T , x) = a(t) (t,
(7)
in terms of which the matter action becomes

1

R 2
3
2
2
4
2

(T ) ( ) a (T )V
S[ ] = dT d x
+ a (T ) . (8)
2
a(T )
12
Since we are interested in describing the time evolution of an initial quantum state, we pass
on to the Hamiltonian description in the Schrdinger representation. This procedure begins
, x), and the
by obtaining the canonical momentum conjugate to the quantum field, (T
Hamiltonian density H(T , x)
(T , x) = (T , x),

R 2
1 2 1
2
4
+ ( ) + a (T ) V
H(T , x) =
a 2 (T )
,
2
2
a(T )
12

H (T ) = d 3 x H(T , x),
(9)
where the prime denotes derivative with respect to the conformal time T .
In the Schrdinger representation the canonical momentum is given by
a (T , x) = i
,
a (T , x)
a = 1, . . . , N.
(10)
The time evolution of the wave-functional [ ; T ] is obtained from the functional

Schrdinger equation

i
(11)
; [ ; T ].
[ ; T ] = H
T

The implementation of the large N limit begins by writing the field as follows

(x, T ) = (x, T ), (x,

T)

T) ,
= N (T ) + (x, T ), (x,
(12)
where we choose the 1-axis in the direction of the expectation value of the field and we
collectively denote by the N 1 perpendicular directions
(T ) = (x, T ),
(x, T ) = (x, T ) = 0,
(13)
where the expectations value above are obtained in the state represented by the wavefunctional [ ; T ] introduced above.
The leading order in the large N limit can be efficiently obtained by functional methods
(see Refs. [8,9,13,14,16] and references therein). The contributions of to the equations
of motion are subleading (of order 1/N ) in the large N limit [8,16].
It is convenient to introduce the spatial Fourier modes of the quantum field

k (T ) = d 3 x (x, T )eikx .
(14)
127
In leading order in the large N limit, the explicit form of the Hamiltonian is given by [8,9,
13,14,16]

2

H (T ) = N V hcl (T )
k k +
Hk (T ),
8N
k
a 2 (T
) 2 2
1 2
(T ) +
m (T ) + 4 (T ),
2
2
8
2
1
1
Hk (T )
+ 2 (T ) k k ,
2 k k 2 k

R(T )
2
2
2
2
k (T ) k + a (T ) M (T )
,
6
hcl (T ) =
M2 (T ) m2 + R +
2
2
+
,
2
2 a (T ) 2 N a 2 (T )
(15)
(16)
(17)
where V is the comoving volume. We assume spherically symmetric distributions in

momentum space.
That is, in the large N limit the Hamiltonian operator (9) becomes
a time dependent
c-number contribution plus a quantum mechanical contribution, k Hk (T ), given by a
collection of harmonic oscillators with time-dependent frequencies, coupled only through
the quantum fluctuations k k .
In Eqs. (15)(17) the scale factor a(T ) is determined self-consistently by the Einstein
Friedmann equations.
2.1. Tsunami initial states
To highlight the main aspects of the inflationary scenario proposed here, and to establish
a clear difference with the conventional models, we now focus our discussion on the case of
vanishing expectation value, i.e., (T ) = 0, and a pure quantum state. The most general
cases with mixed states described by density matrices and non-vanishing expectation value
of the field are discussed in detail in Sections 3.3 and 4 and in Appendix A.
For (T ) = 0 the quantum Hamiltonian (14) becomes a sum of harmonic oscillators
with time dependent frequencies that depend on the quantum fluctuations. Therefore, we
propose a Gaussian wave-functional of the form
[ ; T ] = N (T )
Ak (T )
k k
2
(18)
The functional Schrdinger equation (11) in this case leads to evolution equations for the
normalization factor N (T ) and the covariance kernel Ak (T ) whose general form is found
in Appendix A (see also [8]). The evolution of the normalization factor is determined by
that of Ak , while the equation for Ak is
iAk (T ) = A2k k2 (T ),
(19)
128
where primes refer to derivatives with respect to conformal time. As described in Appendix
A for the general case, the above equation can be linearized by defining (see Appendix A)
Ak (T ) i
k (T )
,
k (T )
(20)
where the mode functions k satisfy the equation

k + k2 (T )k = 0.
(21)
In terms of these mode functions the self-consistent expectation value

2
1
N
N
d 3k
=
k k ,
k k =
= |k |2 .
N
N
(2)3
2AR,k
2
(22)
We now must provide initial conditions on the wave functional to completely specify the
dynamics. Choosing the initial (conformal) time at T = 0 with a(T = 0) = 1, the initial
state is completely specified by furnishing the real and imaginary parts of the covariance
Ak at the initial time. We parameterize these as1
AR,k (0) = k ,
AI,k (0) = k (0)k .
(23)
Choosing the Wronskian of the mode functions k (T ) and its complex conjugate to be
k k k k = 2i
(24)
determines the following initial condition on the mode functions (see Appendix A)
1
k (0) = ,
k
k (0) = k (0)k + ik k (0),
(25)
where we also used Eqs. (20) and (23).

An important alternative interpretation of these mode functions is that they form a
basis for expanding the Heisenberg field operators (solution of the Heisenberg equations
of motion)

d 3k

ak k (T )ei kx + ak k (T )ei kx ,

(
x, T ) =
(26)
3
(2)
with ak , ak annihilation and creation operators, respectively, with canonical commutation
relations. The Wronskian condition (24) ensures that the (
x , T ) fields and their conjugate
momenta obey the canonical commutation relations at equal conformal times.
The physical interpretation of these initial states is highlighted by focusing on the
occupation number of adiabatic states as well as on the probability distribution of field
configurations.
Occupation number: it is at this point where the description in terms of conformal
time proves to be valuable. In conformal time the Hamiltonian in the large N limit is
1 In the case for which 2 (0) < 0 we choose (0) =
k
k

k 2 + |M2 (0) R(0)/6|.
129
that of a collection of harmonic oscillators with time dependent frequencies. It is then

convenient to introduce the adiabatic occupation number operator

1 Hk (T ) 1
,
n k (T ) =
(27)
N k (T ) 2
with Hk given by Eq. (15).
In particular the occupation number at the initial time is given by (see Appendix A)
nk n k (0) =
[k (0) k ]2 + k2 (0)k2
.
4k (0)k
(28)
Here, the special case with k = k (0) and k = 0 corresponds to the adiabatic
vacuum (nk = 0). Instead, we study an initial state in which a band of wave-vectors
are populated with a non-perturbatively large number of particles. More precisely, we
consider initial states for which

1
nk = O , inside the excited band,
0, outside the excited band,
(29)
where is the quartic self-coupling.
This is accomplished by choosing,

1
O
, inside the excited band,
1
k (0)
=
1
k
and k = 0, outside the excited band.

(30)
k (0)
These initial states are highly excited, the expectation value of the energymomentum
tensor in these states leads to an energy density 1/ and are, therefore, nonperturbative. We will refer to the case where the excited band is narrow as the narrow
tsunami.
It must be stressed that the particle distribution nk alone partially determines the initial
state. As we see from Eq. (25), the initial state is completely defined specifying two
functions of k: k and k .
Probability distribution: an alternative interpretation of these initial states is obtained
by focusing on the probability distribution of field configurations at the initial time. It
is given by

2
ek k k .
P[]
= [ ; T = 0] = N (0)
(31)
k
An intuitive quantum mechanical picture of the wave-functional for the modes in

the excited band is the following. At the initial time the instantaneous Hamiltonian
corresponds to a set of harmonic oscillators of frequencies k (0), while the width in
1/2
. For a mode in the
field space of theinitial Gaussian
state is determined by k
1/
(0)
and
the
typical
amplitudes
of
the field are k
vacuum
state
1/
k
k
1/ k (0) which is the typical width of the potential

well.
While
for
a mode inside the
excited band the width in field space is 1/ k 1/ k (0) [see Eq. (30)]. Thus,
130
large amplitude field configurations with kk0 1/ k (0) 1/ k (0) have a

probability of O(1), i.e., large amplitude configurations within the band of excited
wave-vectors are not suppressed. That is, the width of the probability distribution
for these modes is much larger than the typical size of the potential well and there is
a non-negligible probability for finding field configurations with large amplitudes of
O(1/).
These highly excited initial states had been previously proposed as models to describe
the initial stages of ultrarelativistic heavy ion collisions and had been coined tsunami
waves [1214]. They represent spherical shells (in momentum space) with large
occupation numbers of quanta, describing a state with a large energy density with
particles of a given momentum.
2.2. Back to comoving time: renormalized equations of motion
Having set up the initial value problem in terms of the tsunami-wave initial wavefunctionals, the dynamics is now completely determined by the set of mode equations
Eq. (21) with Eqs. (16), (17) and the initial conditions Eqs. (25). However, in order to
establish contact with more familiar results in the literature, it is convenient to rewrite the
equations of motion in comoving time. This is accomplished by the field rescaling given
by Eq. (7) which at the level of mode function results in introducing the comoving time
mode functions fk (t) related to the conformal time ones k (T ) as
fk (t) =
k (T )
.
a(t)
(32)
The equations of motion in comoving time for these mode functions are

k2
2
+
M
(t)
fk (t) = 0,
a 2 (t)
d 3k
M2 (t) = m2 + R(t) +
|fk (t)|2 ,
4
(2)3

1
fk (0) = ,
fk (0) = k (0) k + H (0) + ik fk (0).
k
fk (t) + 3 H (t) fk (t) +
(33)
(34)
(35)
The EinsteinFriedmann equation are,

H (t) =
2
a(t)
a(t)
2
=
80
2
3MPl
0 = T00 ,
(36)
where the expectation value is taken in the time evolved quantum state. It is straightforward
to see that the expectation value of the energymomentum tensor has the perfect fluid form,
as a consequence of the homogeneity and isotropy of the system [8,9].
Thus the set of Eqs. (33)(36) provide a closed set of self-consistent equation for the
dynamics of the quantum state and the spacetime metric.
131
2.2.1. Renormalized equations of motion in the large N limit

The set of equations that determine the dynamics of the quantum state and the scale
factor need to be renormalized. The field quantum fluctuations

d 3k
2
=
|fk (t)|2 ,
N
2(2)3
requires subtractions which are absorbed in a renormalization of the mass, coupling to the
Ricci scalar and coupling constant. The expectation value of the stress tensor also requires
subtractions (but not multiplicative renormalization). Since the divergence structure is
determined by the large energy, short distance behavior, the band of excited modes does
not influence the renormalization aspects. Therefore, we use the extensive work on the
renormalization program which is available in the literature referring the reader to Refs. [8,
9] for details. We here summarize the aspects that are most relevant for the present
discussion.
First, it is convenient to introduce the following dimensionless quantities,
= mt,
q =
H (t)
,
m
k
q =
,
m
h( ) =
k
,
m
k
,
m
,
g=
8 2
q=
fq ( ) =
m fk (t),
(37)
where m and stand for the renormalized mass of the inflaton and the renormalized selfcoupling, respectively [8]. In terms of these dimensionless quantities we now introduce the
dimensionless and fully renormalized expectation value of the self-consistent field as
2
(t) R ,
2m2

2
( ) = q dq |fq ( )|2
g( )

1
(q 1) M2 ( ) R( )
,
+
qa( )2
2q 3
m2
6m2
(38)
where the terms subtracted inside the integrand renormalize the mass, coupling to the Ricci
scalar and the coupling constant [8,9]. The dimensionless and renormalized expressions for
the energy density 5 and pressure p are given by
00
T R
2Nm4
g( ) [g( )]2
+
=
2
4

g
q2
+
|fq ( )|2 S2 (q, ) ,
q 2 dq |fq ( )|2 S1 (q, ) + 2
2
a ( )
ii
p( )
T R,
2Nm4

q2
|fq ( )|2 S2 (q, ) . (39)
(p + 5)( ) = g q 2 dq |fq ( )|2 S1 (q, ) + 2
3a ( )
5( )
132
Where the renormalization subtractions S1 and S2 are given by [8,9]

q
1
+
B( ) + 2a 2
a 4 ( ) 2qa 4( )
(q 1)
) + 3a( )a(
) 4a 2 ( )B( ) ,
B( )2 a( )2 B(
)B(
+ 3 4
8q a ( )

1
(q 1)
1
d
) ,
S2 (q, ) = 2
3 2 B( ) + 5 2
a( )B(
3B( )2 + a( )
qa ( ) 2q a ( )
8q a ( )
d

2
B( ) a ( ) 1 + g( ) .
(40)
S1 (q, ) =
We choose here = 0 (minimal coupling), the renormalization point = |m| and a(0) = 1.
In summary, the set of coupled, self-consistent equations of motion for the quantum
state and the scale factor are

2
q2
d
d
+ 2
+ 1 + g( ) fq ( ) = 0,
(41)
+ 3h( )
d 2
d
a ( )

1
fq (0) = ,
(42)
fq (0) = q q + h(0) + i q fq (0),
q

R(0)
2

,
q = q + 1 + g(0)
(43)
6m2
plus the EinsteinFriedmann equation of motion for the scale factor
h2 ( ) = L2 5( ),
where L2
16Nm2
,
2
3MPl
(44)
with g( ) and 5( ) given by Eqs. (38) and (39), respectively.

In order to implement the numerical analysis of the set of Eqs. (41), (42), (38) and (44)
we introduce an ultraviolet momentum cutoff . For the cases considered in this article
we choose 200 and found almost no dependence on the cutoff for larger values. As
befits a scalar inflationary model, the scalar self-coupling is constrained by the amplitude
of scalar density perturbations to be 1012 [3,5] implying that g < 1013 . Therefore,
the subtractions can be neglected because Si O(g4 ) < 104 .
The initial state is defined by specifying the q and q . We determine the range of these
parameters q and q by the excitation spectrum for the tsunami-wave initial state, as well
as the condition that lead to inflationary stage. This will be studied in detail in the next
section.
3. Tsunami inflation
As emphasized in the previous section, the scenario under consideration is very different
from the popular treatments of inflation based on the evolution of classical scalar inflaton
field [2,3,5,6]. In these scenarios all of the initial energy is assumed to be in a zero mode (or
order parameter) at the beginning of inflation and the quantum fluctuations are taken to be
133
perturbatively small with a negligible contribution to the energy density and the evolution
of the scale factor.
In contrast to this description, our proposal highlights the dynamics of the quantum
states as the driving mechanism for inflation. The initial quantum states under consideration correspond to a band of quantum modes in highly excited states, thus the name
tsunami-wave [1214]. This initial state models a cosmological initial condition in which
the energy density is non-perturbatively large, but concentrated in the quanta rather than in
a zero mode.
We now study under which general conditions such a state can lead to a period of
inflation that satisfies the cosmological constraints for solving the horizon and entropy
problems entailing the necessity for about 60 e-folds of inflation.
It is understood that inflation takes place whenever the expansion of the universe
accelerates, i.e.,
a
L2
= h2 + h = [5 + 3p] > 0,
a
2
(45)
with L given in Eq. (44) and 5 and p given by Eqs. (39).

While our full analysis rely on the numerical integration of the above set of equations,
much we learn by considering the narrow tsunami case.
3.1. Analytical study: the narrow tsunami case
Before proceeding to a full numerical study of the equations of motion, we want to
obtain an analytic estimate of the conditions under which a tsunami initial quantum state
would lead to inflation.
Our main criterion for such initial state to represent high energy excitations is that the
number of quanta in the band of excited modes is of O(1/). This criterion, as explained
above, is tantamount to requiring that field configurations with non-perturbative amplitudes
have non-negligible functional probability. Progress can be made analytically by focusing
on the case in which the band of excited field modes is narrow, i.e., its width <k is
such that <k k0 or in terms of dimensionless quantities <q/q0 1. We introduce the
following smooth distribution
q =
1+
N
g
q

2
,
exp qq0
with
2<q
<q
1,
q0
(46)
with q given by Eq. (43) and N a normalization constant that fixes the value of the total
energy.
In addition, we choose q = h(0)/q as we discuss below in Eq. (57).
This initial distribution posses the main features of the tsunami state described in the
previous section. Since g 1, we have for q q0 ,
1
1
1
q
g
nq
1
1,
g
(47)
134
corresponding to highly excited states. While for |q q0 | <q

1
1
q
q
nq 0.
(48)
Thus, these modes are in a quantum state near the conformal (adiabatic) vacuum at the
initial time, with nq the number of quanta defined by Eq. (28) in terms of dimensionless
variables. For these distributions (narrow tsunamis), the integral over mode functions for
the quantum fluctuations g( ) [given by Eq. (38)] is dominated by the narrow band of
excited states with mode amplitudes 1/ g and can be approximated by

g( ) = g<qq02|fq0 ( )|2 + O(g) + O(g <q) |q0 ( )|2 ,
where we have introduced the effective q0 mode

q0 ( ) g<q q0 fq0 ( )
(49)
(50)
we note that the initial condition (42) and the tsunami-wave condition (47) entail that
despite the presence of the coupling constant in its definition, the amplitude of the effective
q0 mode is of O(1).
The equation of motion for the effective q0 -mode takes the form
2

q
q0 ( ) + 3 h( ) q0 ( ) + 2 0 + 1 + |q0 ( )|2 q0 ( ) = 0.
(51)
a ( )
The scale factor follows from
h2 ( ) = L2 5( ),
(52)
with energy and pressure,

q02
1
1
1
5( ) = | q0 ( )|2 + |q0 ( )|2 + |q0 ( )|4 +
|q ( )|2 ,
2
2
4
2 a 2( ) 0
q02
(p + 5)( ) = | q0 ( )|2 +
|q ( )|2 ,
3 a 2( ) 0
(53)
where we have neglected terms of O(g).

We will refer to the set of evolution Eqs. (51)(53) as the one mode approximation
evolution equations.
In particular, within this one-mode approximation, the acceleration of the scale factor
obeys

|q0 ( )|2 |q0 ( )|4
a(
)
2
2
= L |q0 ( )|
.
(54)
a( )
2
4
Therefore, the condition for an inflationary epoch, a > 0, becomes
1
1
|q0 ( )|2 + |q0 ( )|4 .
2
4
A sufficient criterion that guarantees inflation is the tsunami slow roll condition
| q0 ( )|2
<
| q0 ( )| |q0 ( )|.
(55)
(56)
135
The initial conditions (42) and the condition that q0 g 1 imply that the tsunami slow
roll condition (56) at early times is guaranteed if q0 is such that
|q0 q0 + h(0)| 1.
(57)
Hence tsunami-wave initial states that satisfy the tsunami slow-roll condition (57) lead to
an inflationary stage.
Moreover, in order to have slow roll (56) at later times, the effective friction coefficient
3 h( ) should be larger than the square of the frequency in the evolution Eq. (41). That is,
q02 + 1 + g(0)
1.
3h(0)
(58)
(i.e., the q0 -mode should be deep inside the overdamped oscillatory regime). Eq. (58)
implies that h(0) 1 and this together with Eq. (57) implies that q0 must be negative.
A remarkable aspect of the narrow tsunami state is that it leads to a dynamical evolution
of the metric similar to that obtained in classical chaotic inflationary scenarios in the
slow roll approximation [3,5]. In particular the expression for the acceleration (54) and
the tsunami slow roll condition (56) are indeed similar to those obtained in classical
chaotic inflationary models driven by a homogeneous classical field (zero mode). However,
despite the striking similarity with classical chaotic models, we haste to add that both the
conditions that define a tsunami state and the tsunami slow roll condition (56) guaranteed
by the initial value (57) is of purely quantum mechanical origin in contrast with the
classical chaotic-slow-roll scenario. Furthermore, we recall that the expectation value of
the scalar field vanishes in this state.
3.1.1. Early time dynamics
Under the assumption of a tsunami wave initial state and the tsunami slow-roll condition
(56) the contribution q0 ( ) in the energy and in the pressure [see Eq. (53)] can be
neglected provided
| q0 ( )|2
q02
|q ( )|2 .
3 a 2( ) 0
(59)
We call A the time scale at which this rely no longer holds. Furthermore, we can
approximate q0 ( ) by q0 (0) if

q0 (0)
.

A
(60)
q0 (0)
This condition is fulfilled at least for A 1 due to the tsunami slow-roll condition (56).
During this interval the Friedmann equation (52) takes the form,

q02
1
a(
) 2
D
2 1
2
4
2
|q0 (0)| + |q0 (0)| +
|q (0)| = 2
+ E, (61)
=L
a( )
2
4
2 a 2( ) 0
a ( )
where we used that g(0) = |q0 (0)|2 . This equation is valid as long as the characteristic
time scale of variation of the metric is shorter than that of the mode q0 ( ).
136
The preceding equation can be integrated with solution

D
a(
)
sinh E + c ,
= E > 0,
a( ) =
E
a( )

E
)=
,
h( ) = E coth E + c ,
h(
2
sinh ( E + c)
(62)
where the constants D, E and c are given by

2
2 q0
D=L

g0 ,
E=L

g0 g02
+
,
2
4

sinh c =
E
,
D
(63)
and g0 g(0) = |q0 (0)|2 .

We see from Eq. (62) that during this interval there is an inflationary stage with an
accelerated expansion a(
)/a(
) = E > 0. We also see that h( ) decreases with time until
it reaches the constant value E that determines the onset of a quasi-de Sitter inflationary
stage.
We now estimate the range of validity of the solution in Eq. (62). The first condition in
Eq. (59) is more stringent than the second one in Eq. (60) for most of the interesting range
of parameters.
A determines the time scale at which the solution (62) ceases to be valid, and the
condition (59) is no longer fulfilled, i.e.,
| q0 (A )|2
q02
|q (A )|2 .
3 a 2(A ) 0
(64)
When this equation is valid, we see from Eq. (53) that p + 5 becomes of the order
of | q0 (A )|2 . Furthermore, the above condition together with the slow roll condition,
Eq. (56), leads to
q02
1.
a 2 (A )
(65)
Therefore, from Eq. (61) we see that h(A ) E and is slowly varying.
Since the slow roll condition guarantees that |q (A )| |q (A )|, we can now use
Eq. (49) along with the evolution Eq. (51) which setting h = E leads to the following
relation
q0 (A )
1 + g0
q0 (A ).
3 E
From Eq. (64) the scale A is therefore estimated to be given by

3q0 E
1
c
A ArgSinh
E
D (1 + g0 )

1
L 3g0 (1 + g0 /2)
= ArgSinh
c .
E
2 (1 + g0 )
(66)
(67)
137
This initial inflationary period with a decreasing Hubble parameter exists provided the
r.h.s. is here positive, i.e.,
1 + g0
.
q0 >
3E
(68)
In order to distinguish this phase from the later stages, to be described below, we refer to
this early time inflationary stage as tsunami-wave inflation because the distinct evolution
of the scale factor during this stage is consequence of the tsunami-wave properties.
At = A we have:
q0 (A ) q0 (0) =
3 E q0
a(A )
,
1 + g0
g0 ,
h(A )
q0 (A )
1 + g0
q (A ),
3 h(A ) 0
E.
(69)
For > A , q0 /a( ) 1 and the physical wavevectors in the excited band have redshifted so much that all terms containing q0 become negligible in the evolution equations.
Therefore, all modes in the excited band evolve as an effective q = 0 mode. Hence for
> A the dynamics of the scale factor is described by an effective homogeneous zero
mode and describes a different regime from the one studied above. Such regime is akin to
the classical chaotic scenario.
3.1.2. The effective classical chaotic inflationary epoch
For A , when q02 /a 2 | q |2 /|q |2 1 all the physical momenta corresponding
to the comoving wavevectors in the excited band have redshifted to become negligible in
the equations of motion. The dynamics is now determined by the following set of equations
for the effective zero mode and the scale factor,

q0 ( ) + 3 h( ) q0 ( ) + 1 + |q0 ( )|2 q0 ( ) = 0,
(70)
h2 ( ) = L2 5( ),
where the energy and pressure are given by
1
1
1
5( ) = | q0 ( )|2 + |q0 ( )|2 + |q0 ( )|4 ,
2
2
4
(p + 5)( ) = | q0 ( )|2 .
(71)
The initial conditions on q0 and q0 are determined by their values at the time A ,
while the slow-roll condition (56) determines that the imaginary parts of q0 and q0 are
negligible.
Therefore, after A the dynamic is identical to that of a classical homogeneous field
(zero mode)

eff ( ) = Re q0 ( ) ,
(72)
that satisfies the equations of motion,

2
eff + 3h eff + 1 + eff
eff = 0,
h2 ( ) = L2 5( ),
(73)
138
with energy and pressure,

1 2
1 2
1 4
2
5( ) = eff
+ eff
+ eff
,
(p + 5)( ) = eff
,
2
2
4
and initial conditions [using Eq. (69)],

eff (A ) = q0 (A ) = q0 (0) = g0 ,
1 + g0
eff (A ),
eff (A ) = q0 (A ) =
3h(A )
(74)
(75)
where the value of eff (A ) is determined by the slow roll condition ( q0 (A ) 0), the
evolution Eqs. (51) and (65) and a(A ) and h(A ) are given by Eq. (69).
When g0 1, the quadratic term in the potential dominates, and we can integrate the
previous equations to obtain
2
( A ) (for g0 1).
eff ( ) = eff (A )
(76)
3L
This evolution is similar to that of classical chaotic inflationary models [3,5]. Therefore,
for > A when the physical momenta in the excited band have redshifted so much that
their contribution in the equations of motion of the quantum modes and the energy and
pressure become negligible, the evolution of the quantum modes and the metric is akin to
a classical chaotic inflationary scenario driven by a homogeneous c-number scalar field.
This equivalence allows us to use the results obtained for classical chaotic inflation. Thus,
as the classical slow roll condition (| eff | |eff |) holds, the evolution of the effective
scalar field is overdamped and the system enters a quasi-de-Sitter inflationary epoch. This
inflationary period ends when the slowly decreasing Hubble parameter becomes of the
2 . At this stage the effective classical field exits
order of the inflaton mass, i.e., 3 h 1 + eff
the overdamped regime and starts to oscillate, the slow roll condition no longer holds and
a matter dominated epoch (| eff | |eff | p 0) follows.
However, we emphasize that while the effective zero mode eff obeys a classical
equation of motion and that the components of the energymomentum tensor Eq. (74)
are those from a classical field, the origin of this mode is purely quantum mechanical.
From the identification (75) and Eq. (49) it is clear that the effective zero mode is a
collective superposition of modes
excited band. From the initial and tsunami
in the highly
1/ q0 1/ g it follows that the amplitude of the effective

wave conditions fq0 (A )
zero mode is eff (A ) q0 <q. Restoring the dimensions and the proper powers of the
coupling that were absorbed in the constant L in the Friedmann equation we find that the
equations of motion (73) with the stress tensor components (74) are those obtained from a
(dimensionfull) classical homogeneous field eff (t) with a classical potential V (eff ) given
by
m 2
m2 2
+ 4 ,
V (eff ) =
eff (t) = eff ( ),
(77)
2 eff 8 eff
with the initial value at the time tA = A /m

m k02 <k
.
eff (tA )
(78)
m3
139
The non-perturbative amplitude of the effective zero mode is a consequence of the nonperturbative amplitude of the excited quantum modes with an O(1/) number of quanta.
3.1.3. Number of e-folds
An important cosmological quantity is the total number of e-folds during inflation. As
discussed above, there are two different inflationary stages, the first one is determined by
Eqs. (61), (62) and characterized by a rapid fall-off of the Hubble parameter approaching a
quasi-de Sitter stage. This new stage has been referred to as the tsunami-wave inflationary
stage above to emphasize that the dynamics is determined by the distinct characteristics of
the tsunami-wave initial stage.
The second stage is described by an effective zero mode and the evolution equations (73), (74) and is akin to the chaotic inflationary stage driven by a classical homogeneous scalar field. The crossover between the two regimes is determined by the time
scale A (in units of the inflaton mass) and given by Eq. (67) at which the contribution
from the term q02 /a 2 ( ) to the equations of motion becomes negligible. Therefore, there
are to distinct contributions to the total number of e-folds, which is given by

Ne q0 , h(0) = log a(A ) + Ne 0, h(A ) ,
(79)
where a(A ) is given by Eq. (75) and Ne (0, h(A )) is just the number of e-folds for classical
chaotic inflation with an initial Hubble parameter h(A ).
We can express h(A ) as a function of q0 and h(0),

2
q2
g0 (g0 )
+
= h2 (0) L2 0 g0 .
h(A ) = L
(80)
2
4
2
The number of e-folds during the first stage, is given by

3 E q0
log a(A ) log
.
1 + g0
(81)
The expression for the number of e-folds during the following, chaotic inflationary stage
simplifies when g0 1. In this case the quadratic term in the potential dominates, and
we can obtain simple analytical expressions

3L2 2
3L2
3L2 50
Ne 0, h(A ) =
eff (A ) =
g0 =
4
4
2 1 + q02
(for g0 1).
(82)
We see that the number of e-folds grow when q0 decreases at fixed initial energy 50 . That
is, we have more e-folds when the energy is concentrated at low momenta.
3.1.4. In summary
Before proceeding to a full numerical study of the evolution we summarize the main
features of the dynamics gleaned from the narrow tsunami case to compare with the
numerical results.
The conditions for tsunami-wave inflation are (i) a band of excited states centered at a
momentum k0 with a non-perturbatively large O(1/g) number of quanta in this band,
140
and (ii) the tsunami slow-roll condition Eq. (56). These conditions are guaranteed by
the initial conditions on the mode functions given by Eq. (42) with the tsunami-wave
distributions of the general form given by Eqs. (46), (57).
There are two successive inflationary periods. During the first one, described in
Section 3.1.1, the dynamics is completely characterized by the distinct features of
the tsunami-wave initial state,
the Hubble parameter falls off fast and reaches an
approximately constant value E that characterizes the quasi-de-Sitter epoch of
inflation of the second period. The second stage, described in Section 3.1.2 can be
described in terms of an effective classical zero mode and the evolution of this effective
mode and that of the Hubble parameter are akin to the standard chaotic inflationary
scenario.
The tsunami-wave initial state can be interpreted as a microscopic justification
of the classical chaotic scenario described by an effective classical zero mode of
large amplitude. The amplitude of this effective zero mode is non-perturbative as a
consequence of the non-perturbative O(1/) number of quanta in the narrow band of
excited modes. Thus the initial value of the effective, classical zero mode that describes
the second, chaotic inflationary stage, is completely determined by the quantum initial
state.
An important point from the perspective of structure formation is that the band of
excited wavevectors centered at q0 either correspond to superhorizon modes initially,
or all of the excited modes cross the horizon during the first stage of inflation, i.e.,
during the tsunami stage. This is important because the chaotic second stage of
inflation which dominates during a longer period guarantees that the band of excited
modes have become superhorizon well before the last 10 e-folds of inflation and hence
cannot affect the power spectrum of the temperature anisotropies in the CMB. The fact
that the tsunami-wave initial state is such that the very high energy modes (necessarily
trans-Planckian) that cross the horizon during the last 10 e-folds and are therefore of
cosmological importance today are in their (conformal) vacuum state leads to the usual
results from chaotic inflation for the power spectrum of scalar density perturbations.
Although these conclusions are based on the narrow tsunami case, we will see below
that a full numerical integration of the self-consistent set of equations of motion confirms
this picture. In Sections 3.3 and 4 we show how this results can be easily extended to more
general particle distributions and more general initial states.
3.2. Numerical example
To make contact with familiar models of inflation with an inflaton field with a mass near
the grand unification scale, we choose the following values of the parameters:
m
= 104 ,
MPl
= 1012 ,
N = 20,
(83)
141
Fig. 1. Tsunami inflation: isolines of constant number of e-folds obtained from Eq. (82) (valid for g0 1), for
m = 104 MPl , = 1012 and N = 20.
where the number of scalar fields N = 20 has been chosen as a generic representative of a
grand unified quantum field theory. For these values we find
L2
16Nm2
= 3.35 106 .
2
3MPl
4.
As an example we shall consider an initial energy density 0 = T00 = 102 MPl
Thus, the initial value for the Hubble parameter is H0 = 80 /3MPl = 3.53 1018
GeV(= 1.654 1052 km s1 Mpc1 ). These initial conditions in dimensionless variables
give 50 = 2.50 and h(0) = 2890.
In addition, the slow roll conditions (58) imply:
q02 + 1 + g0
1,
3 h(0)
which in this case results in
q0 95.
We choose q0 = 80.0, and initial conditions in Eq. (42) with q and q given by Eqs. (46)
and (57). These initial conditions satisfy the tsunami slow roll condition,
|q q + h(0)| 1.
(84)
Furthermore, we take <q = 0.1 and N is adjusted by fixing the value g(0) = g0
which for the values chosen for 50 and by Eqs. (39), (49) and (56) gives g0 = 7.81
104 .
Fig. 1 displays 50 vs. q0 along lines of constant number of e-folds, while Figs. 27
display the solution of the full set of Eqs. (41)(44) with (39). An important feature that
142
Fig. 2. Tsunami inflation: early time h( ). hA h(A ) is the asymptotic value for the early period that ends
4 , q = 80.0 and
at A 0.133. For m = 104 MPl , = 1012 and N = 20. Initial conditions: 0 = 102 MPl
0
<q = 0.1q0 .
Fig. 3. Tsunami inflation: h( ) for > A . The early time analytic approximation gives hA = 36.1 (also with the
one mode approx.), numerically we obtain h(A ) = 35.5. Same parameters and initial conditions as in Fig. 2.
emerges from these figures is that for the set of parameters that are typical for inflationary
scenarios and for large values of q0 = k0 /m (but well below the Planck scale) the number
of e-folds obtained is more than sufficient as shown by Fig. 6.
We also show that the dynamics of the full set of Eqs. (41)(44) with (39) is correctly
approximated by the narrow tsunami case studied in the previous subsections: the one mode
143
Fig. 4. Tsunami inflation: a(

)/a( ), it shows that there is accelerated expansion (inflation) up to times 109.
Same parameters and initial conditions as in Fig. 2.
Fig. 5. Tsunami inflation: g( ), after A 0.133, it plays the role of an effective classical field. Same
parameters and initial conditions as in Fig. 2.
approximation [Eqs. (49)(53)], the early time analytical formulae (for A ) [Eqs. (62)
and (63)], and the effective classical field (for > A ) [Eq. (74)]. The agreement between
the analytic treatment and the full numerical evolution is displayed in Figs. 27.
The early time analytic expressions predict an inflationary period during which the
Hubble parameter falls off fairly fast, that lasts up to A 0.133 [Eq. (67)] reaching
an asymptotic value of h(A ) = 36.2 [Eqs. (63) and (69)]. The one mode approximation
144
Fig. 6. ln[a( )] vs. . Same parameters and initial conditions as in Fig. 2.
Fig. 7. Tsunami inflation: p( )/5( ). It shows the onset of a matter dominated epoch after the quasi-de-Sitter
stage. Same parameters and initial conditions as in Fig. 2.
gives the same prediction h(A ) = 36.1, and numerically evolving the full set of equations
we find h(A ) = 35.5. Thus, we see from this values and from Figs. 2 and 3 that both
approximations are fairly accurate for early times.
After A , the geometry reaches a quasi-de-Sitter epoch. We have shown in the previous
subsection that after the time A the evolution equations for the one mode approximation
reduce to those of an effective classical field. The effective zero mode approximation
correctly predicts the dynamics in this epoch as can be gleaned from Figs. 36.
145
While the stage of early tsunami inflation up to A results in only 8.5 e-folds, the
following quasi-de-Sitter stage described by the effective classical scalar field lasts for
a total of 1900 e-folds. For the values of parameters chosen above, g0 1, hence, we
can estimate the number of e-folds with Eq. (82). Using Eq. (79) we obtain a total of 1970
e-folds while the one mode approximation yields 1960 e-folds. Both results agree with the
full numerical solution of the equations (see Fig. 6).
Furthermore, as stated above inflation ends when h
dominated epoch follows.
2
1+eff
3
13 , after which a matter
3.3. Other distributions and other states

The validity of the physical picture that emerges from the previous analytic and
numerical study is not restricted to pure states or narrow distributions of the form given
by (46). We have also studied more general distributions and mixed states:
Other distributions
The narrow tsunami case where a single quantum mode q0 dominates the dynamics has
been extremely useful to study the dynamics in the previous section. The generalization
to the case with continuous distributions of q-modes can be easily obtained making the
changes:

|q0 ( )|2 g q 2 dq|fq ( )|2 ,

| q0 ( )|2 g q 2 dq|fq ( )|2 ,

q02 |q0 ( )|2 g q 2 dq q 2 |fq ( )|2 .
(85)
The two stages of inflation are always present for such continuous modes distribution as
long as the following generalized slow-roll conditions is fulfilled

g q 2 dq|fq ( )|2 g q 2 dq|fq ( )|2
(86)
that imposes on q the condition |q q + h(0)| 1.
The effective zero mode in the second stage of inflation is now given by

2
eff ( ) = g q 2 dq|fq ( )|2 .
Our numerical study with general distributions reveals that the analytical picture obtained
by substituting Eq. (85) in Section 3.1 correctly reproduce the dynamics.
Other (mixed) states
Although we have focused for simplicity on tsunami pure initial states, we have also
investigated the possibility of mixed states. Mixed state density matrices and their time
146
evolution are discussed in Appendix A. The mixing can be parametrized in terms of angles
k as given in Eq. (A.4) and the number of (conformal) quanta are given by Eq. (A.18).
The only relevant changes that occur are in the integrals for ( ), ( ), p( ) in which

q
q
|fq ( )|2 |fq ( )|2 coth
(87)
,
|fq ( )|2 |fq ( )|2 coth
.
2
2
Tsunami and slow-roll conditions on the mode functions given by Eqs. (30) and (57) lead to
tsunami-wave inflation followed by chaotic inflation just as discussed above. In the narrow
tsunami case the only change is that the effective q0 -mode is rescaled by the mixing factor,
i.e.,

q0
q0
2
2
2
2
|q0 ( )| |q0 ( )| coth

(88)
,
|q0 ( )| |q0 ( )| coth
.
2
2
It is also illuminating to contrast the tsunami-wave mixed states with the more familiar
thermal mixed states. The latter are obtained by the choice
q
,
q = q ,
(89)
q = 0,
q =
T
with T some value of temperature. In this case it is straightforward to see that (quantum)
equipartition results in that the contributions of the modes and their time derivatives to the
energy and pressure are of the same order (|fq ( )|2 [h(0)2 + q2 ]|fq ( )|2 ). Hence, for
these thermal mixed states the tsunami slow-roll condition is not fulfilled. This is obviously
not surprising, such a choice of thermal mixed state leads to a FRW epoch which is not
inflationary. Hence, the tsunami-wave initial conditions along with the generalized slowroll conditions lead to two successive inflationary epochs in striking contrast to the familiar
mixed thermal states.
4. Generalized chaotic inflation

The previous analysis, confirmed by the numerical evolution of the full self-consistent
set of equations leads to one of the important conclusions of this article, that tsunami-wave
initial states provide a microscopic justification of the chaotic inflationary scenario.
We have focused our discussion on initial states with vanishing expectation value of the
scalar field (order parameter) and where the energy is concentrated in a momentum band
(tsunami initial states). This choice brings to the fore the striking contrast between this
novel quantum state and the usual classical approach to chaotic inflation. In this section
we study the dynamics in the case in which the initial state allows for a non-vanishing
expectation value of the scalar field along with some of the initial energy localized in
excited quanta. We refer to this case as generalized chaotic inflation to distinguish from
the tsunami-wave state studied above. This generalization thus includes both cases: the
classical chaotic inflation in the limit when there are no excited modes, as well as the
tsunami initial state when all of the energy is localized in a band of excited modes and the
expectation value of the field vanishes.
The relevant equations of motion in comoving time for the mode functions in this case
are given by (A.20) in Appendix A. Along with the dimensionless variables (37) it is also
147
convenient to introduce a dimensionless expectation value as

2
(90)
(t).
2m2
In this generalized case with = 0 the equations of motion for the mode functions fq ( )
(in terms of dimensionless variables) are the same as in Eq. (41) after the replacement
g( ) g( ) + 2 ( ) and the equation of motion for ( ) is given by
2 ( ) =
d( )
d 2 ( )
+ 3h( )
+ 1 + 2 ( ) + g( ) ( ) = 0,
d
d 2
(0) = 0 ,
(0)
= 0 .
(91)
(92)
The EinsteinFriedmann equation is given by (44) but with the energy density and pressure
now given by
1
2
1
1
5( ) = 2 + g + 2 + g + 2
2
2
4

g
q2
2
2
2
+
q dq |fq | S1 (q, ) + 2 |fq | S2 (q, ) ,
2
a

q2
(p + 5)( ) = 2 + g q 2 dq |fq |2 S1 (q, ) + 2 |fq |2 S2 (q, ) ,
3a
(93)
(94)
where the renormalization subtractions S1 , S2 are obtained from those given by Eq. (40)
upon the replacement g g + 2 .
From this expression we see that for fixed (large) energy density status RO there are
2
two different possibilities:
4 if the 2zero mode squared (0) is larger than the quantum
fluctuations g and g q dq|fq | , the dynamics is basically similar to that in the usual
chaotic inflationary scenarios. This corresponds to most of the initial energy density to
be in the zero mode and little energy density in the band of excited states. On the other
hand, for small 2 (0) most of the initial energy density is in the tsunami quantum state and
the initial dynamics is akin to the = 0 case. To quantify this statement and clarify the
interplay and crossover of behaviors between the = 0 and the generalized chaotic case,
we now resort again to the narrow tsunami case, which highlights the essential physics.
The relevant equations are: (i) the equations of motion for the effective q0 -mode (50),
2

q0
2
2
+ 1 + ( ) + |q0 ( )| q0 ( ) = 0
q0 ( ) + 3 h( ) q0 ( ) + 2
(95)
a ( )
and for the zero mode

(
) + 3h( )(
) + 1 + 2 ( ) + |q0 ( )|2 ( ) = 0
(96)
and the Hubble parameter given by (52) with the energy density given by
5( ) =
1

1
|q0 ( )|2 + 2 ( ) + |q0 ( )|2 + 2 ( )
2
2

q02
1
2
|q ( )|2 .
+ |q0 ( )|2 + 2 ( ) +
4
2 a 2( ) 0
(97)
148
The acceleration of the scale factor is now given by

a(
)
1
2
= L |q0 ( )|2 + 2 ( ) |q0 ( )|2 + 2 ( )
a( )
2

2
1
2
2
|q0 ( )| + ( ) ,
4
(98)
where again we have neglected the renormalization contributions and terms of O(g)
consistently in the weak coupling limit g 1. From Eq. (98) the generalized condition
for an inflationary epoch (within the narrow tsunami case) becomes
1
2
1
|q0 ( )|2 + 2 ( ) + |q0 ( )|2 + 2 ( ) ,
2
4
which is fulfilled if the following generalized slow roll condition holds
| q0 ( )|2 + 2 ( ) <
| q0 ( )|2 + 2 ( ) |q0 ( )|2 + 2 ( ).
(99)
(100)
Under these conditions and from Eqs. (95), (96) and the dynamics of the scale factor driven
by the energy density Eq. (97) we can now distinguish the following different inflationary
scenarios.
Tsunami-dominated. When (q02 +1)|q0 (0)|2 2 (0) the excited states in the tsunamiwave carry most of the initial energy density. In this case the results of the previous
section apply and the scale factor takes the form as in Eq. (61) with D given by
Eq. (63), and E given by Eq. (63) but with g0 g0 + 2 (0). There are two
consecutive inflationary stages as in the previous section. The first described by
Eq. (62), lasts up to the time scale A defined by
2 (A ) + | q0 (A )|2
q02
|q (A )|2
3 a 2(A ) 0
at which the redshift of the momentum q0 is such that q0 /a(A ) 1. The secondary
stage is a usual classical chaotic inflationary epoch determined by the dynamics of an
effective zero mode given by
2
eff
( ) = 2 ( ) + |q0 ( )|2
(101)
because for >

1 and the effective equation of motion for q0 ( ) is
the same as that for ( ).
Zero-mode-dominated. When 2 (0) (q02 + 1)|q0 (0)|2 the energy density stored
in the zero mode is much larger than that contributed by the excited states in the
tsunami-wave. In this case the energy density Eq. (97) is completely dominated by
the zero mode. The ensuing dynamics is the familiar classical chaotic scenario driven
by a classical zero mode, without an early stage in which the scale factor is given by
Eq. (62) which is the hallmark of the tsunami-wave dynamics.
A , q02 /a 2 ( )
This analysis in the narrow tsunami case does highlight the important aspects of the
dynamics in a clear manner, allowing a clean separation of the two cases described above.
149
We have carried a full numerical integration of the equations of motion that reproduce
the results described above. The criterion for the crossover between tsunami-wave and
classical chaotic inflation is determined by the relative contributions to the energy density
from the quantum fluctuations in the tsunami wave state as compared to the energy density
of the zero mode.
The previous results [Eqs. (90)(101)] can be easily generalized for generic continuous
distributions of modes and for mixed states. One has just to make the changes indicated in
Eq. (85) for generic distributions and in Eq. (87) for mixed states.
The generalized slow-roll conditions takes then the form:

q
q
2 ( ) + g q 2 dq|fq ( )|2 coth
2 ( ) + g q 2 dq|fq ( )|2 coth
2
2
during the first stage of inflation.
The dynamics is tsunami dominated provided,

q
2
2
2
g q dq 1 + q |fq (0)| coth
2 (0).
2
The effective zero mode in the second stage of inflation is now given by

q
2
.
eff
( ) = 2 ( ) + g q 2 dq|fq ( )|2 coth
2
These results have been verified by numerical integration of the full set of evolution
Eqs. (39)(44).
5. Conclusions
In this article we have studied inflation in typical scalar field theories as a consequence
of the time evolution of a novel quantum state. This quantum state is characterized by
a vanishing expectation value of the scalar field, i.e., a vanishing zero mode, but a nonperturbatively large number of quanta in a momentum band, thus its nametsunami-wave
state.
This state leads to a non-perturbatively large energy density which is localized in the
band of excited quantum modes. We find that the self-consistent equations for the evolution
of this quantum state and the scale factor lead to inflation under conditions that are the
quantum analog of slow-roll.
The self-consistent evolution was studied analytically and numerically in a wide range
of parameters for the shape and position of the distribution of excited quanta. The
numerical results confirm all the features obtained from the analytic treatment.
Under the conditions that guarantee inflation, there are two consecutive but distinct
inflationary epochs. The first stage features a rapid fall-off of the Hubble parameter and is
characterized by the quantum aspects of the state. During this first stage the large number of
quanta in the excited band are redshifted and build up an effective homogeneous classical
condensate. The amplitude of this condensate is non-perturbatively large, of O(1/), as
150
a consequence of the non-perturbatively large number of quanta in the band of excited

modes.
The second stage is similar to the classical chaotic scenario and it is driven by the
dynamics of this effective classical condensate, with vanishing expectation value of the
scalar field. Under the tsunami slow-roll conditions on the quantum state, the total number
of e-folds is more than enough to satisfy the constraints of inflationary cosmology.
The band of excited wave-vectors if not initially outside the causal horizon, becomes
superhorizon during the first inflationary stage, therefore, these excited states do not modify
the power spectrum of scalar density perturbations on wavelengths that are of cosmological
relevance today.
Therefore, these tsunami-wave quantum states provide a quantum field theoretical
justification of chaotic (or in general large field) inflationary models and yield to a
microscopic understanding of the emergence of classical homogeneous field configurations
of large amplitude as an effective collective mode built from the large number of quanta in
the excited band.
In addition, we recall that it is necessary to choose an initial state that breaks the
symmetry in classical chaotic scenarios [3,4]. This is not the case here. We
have inflation with zero expectation value of the scalar field.
For completeness we have also studied more general states and established the important
difference between tsunami (pure or mixed) quantum states leading to inflation, and
thermal mixed states which do not lead to inflation.
Acknowledgements
D.B. thanks NSF for support through grants PHY-9605186, PHY-9988720 and NSFINT-9815064. H.J.d.V. thanks the CNRS-NSF collaboration for support. F.J.C. thanks
the Ministerio de Educacin y Cultura (Spain) for financial support through the F.P.U.
Programme.
Appendix A. Equations of motion in the large N limit and initial states

In this appendix we obtain the equations of motion in conformal time for the generalized
case in which the initial state is determined by a density matrix. The evolution of the
functional density matrix is given by the Liouville equation in conformal time
i
= [H, ]
T

; T
,
; T ] = H ; H ;
,

[ ,
i

T

(A.1)
where the Hamiltonian H is given by Eq. (15) to leading order in the large N limit.
Consistently with the fact that in the large N limit the Hamiltonian describes a collection
of harmonic oscillators, we propose a Gaussian density matrix

A (T )
Ak (T )
T ] = N (T )
exp
[,
,
k k k
k k
2
2
k

Bk (T ) k k .
151
(A.2)
The hermiticity condition = for the density matrix impose that Bk must be real. In
addition, since (x,
T ) is a real field, its Fourier components must obey the hermiticity
condition k (T ) = k (T ); thus, we can assume Ak (T ) = Ak (T ) without loss of
generality.
The evolution equations for Ak (T ), N (T ) and Bk (T ) are obtained from the Liouville
Eq. (A.1) where the Hamiltonian is given by Eq. (15). We find
iBk = Bk (Ak Ak ),
iAk = A2k Bk2 a 2 (T ) k2 (T ),

T

iN
d T
Ak (T) Ak (T) ,
N (T ) = N (0) exp
2
(A.3)
where the prime denotes derivative with respect to conformal time T .

The normalization factor for mixed states N (T ) is related with the normalization factor
of pure states N (T ) by
N (T ) = N (T )N (T ) ,
where

T
N (T ) = N (0) exp i
dT

2

NV hcl (T )
k k (T )
8N

N
Ak (T )
2
Writing Ak in terms of its real and imaginary parts Ak = AR,k + iAI,k , we find that
Bk /AR,k is a conserved quantity. Thus, we can introduce without loss of generality the
variables AR,k (T ), AI,k (T ) and k defined by
AR,k (T ) AR,k (T ) coth k ,
AR,k (T )
,
Bk (T )
sinh k
AI,k (T ) AI,k (T ),
(A.4)
where k is a time-independent real function.

Introducing the complex variable
Ak = AR,k + iAI,k ,
(A.5)
152
we see that it obeys the following Ricatti equation

iAk = A2k a 2 (T )k2 (T ).
(A.6)
This equation can be linearized defining

Ak (T ) i
k (T )
.
k (T )
Then Eq. (6) implies that the mode functions k obey

R(T )
,
k2 (T ) = k 2 + a 2 (T ) M2 (T )
k + k2 (T )k = 0,
6
(A.7)
(A.8)
where R(T ) is the Ricci scalar.

The relation (A.7) defines the mode functions k (T ) up to an arbitrary multiplicative
constant that we choose such that the Wronskian takes the value,
k k k k = 2i.
(A.9)
For this choice of the Wronskian the definition (A.7) becomes

1
i d
ln |k |2 .
(A.10)
|k |2 2 dT
The mass term in Eq. (A.8) given by Eq. (17) requires the self-consistent expectation value

d 3k
2
=
k k ,
N
(2)3

1
1
k
=
coth
k k = Tr k k =
2 [AR,k + Bk ] 2AR,k
2

k
1
= |k |2 coth
(A.11)
.
2
2
Ak =
Thus, the evolution equations in terms of the mode functions are given by Eq. (A.8) with

2
k
d 3 k |k |2
+
coth
.
M2 (T ) = m2 + R +
(A.12)
2
3
2
2 a
4
(2) a
2
The evolution equation of the mode functions k is the same as the Heisenberg equations
of motion for the fields, hence we can write the Heisenberg field operators as
d 3k
ak k (T )ei kx + ak k (T ) ei kx .
(x,
T)=
(A.13)
3
2(2)
Thus, the definition (A.7) gives the relation between Schrdinger and Heisenberg pictures,
since the functional density matrix (A.2) is in the Schrdinger picture.
The expectation value (T ) [see Eq. (13)] in conformal time obeys the following
equation of motion [8]

R(T )

2
2
(T ) = 0,
(T ) + a (T ) M (T )
(A.14)
6
(0) = 0 ,
(A.15)
(0) = 0 .
153
Hence, the evolution equations are given by (A.8), (A.12) and (A.14) with (A.15).
The initial density matrix in the Schrdinguer picture is determined by specifying the
initial values of AR,k , AI,k and k . We will take a(0) = 1 and parameterize the initial
value of Ak as follows,
AR,k (0) = k ,
AI,k (0) = k (0)k .
(A.16)
The corresponding initial conditions for the mode functions are obtained from Eq. (A.16)
using Eq. (A.7) and the Wronskian constraint Eq. (A.9). These are given by
1
k (0) = ,
k

k (0) = k (0) k + ik k (0).
(A.17)
Defining the number of particles in terms of the adiabatic eigenstates of the Hamiltonian
(14) as in Eq. (27), it is straightforward to find that the initial occupation numbers are given
by

k2 + k2 (0) + k2 (0) k2
k
1
coth
nk (0) = n k (0)(0) =
(A.18)
.
4 k k
2
2
For any mixing parameter k = 0 the density matrix represents a mixed state since Bk = 0,
a pure initial state is obtained by taking k = , in which case Bk 0 and the density
matrix becomes a product of a wave functional times its complex conjugate.
It is convenient to pass to comoving time, this is achieved by the rescaling of the fields

T (t) = (t)a(t),
k T (t) = fk (t)a(t)
(A.19)
in terms of which the equations of motion are
+ 3 H (t) (t)
+ M2 (t)(t) = 0,
(t)
2

k
2
fk (t) + 3 H (t) fk (t) + 2 + M (t) fk (t) = 0,

a (t)

k
d 3k
2
M2 (t) = m2 + R(t) + 2 (t) +
|f
(t)|
coth
,
k
3
2
4
(2)
2
(A.20)
where the dots denote derivative with respect to the comoving time t. The initial conditions
for the order parameter are its initial value (0), and its initial derivative (0).
For
a(0) = 1, the initial conditions for the fluctuations are given by k and
1
fk (0) = ,
k

fk (0) = k (0) k + H (0) + ik fk (0).
(A.21)
[Those are the transformed of the initial conditions in conformal time Eq. (A.17).]
References
[1] M.S. Turner, J.A. Tyson, Rev. Mod. Phys. 71 (1999) S145.
[2] See, for example, M. Kamionkowski, A. Kosowsky, Annu. Rev. Nucl. Part. Sci. 49 (1999) 77;
A.H. Jaffe et al., Cosmology from Maxima-1, Boomerang and COBE/DMR CMB observations, astroph/0007333.
154
[3] E.W. Kolb, M.S. Turner, The Early Universe, AddisonWesley, Redwood, CA, 1990;
A.H. Guth, Eternal inflation, astro-ph/0101507.
[4] P. Coles, F. Lucchin, Cosmology, Wiley, Chichester, 1995.
[5] A.R. Liddle, D.H. Lyth, Cosmological Inflation and Large Scale Structure, Cambridge Univ. Press, 2000.
[6] A.R. Liddle, The early universe, in: D. Valls-Gabaud, M.A. Hendry, P. Molaro, K. Chamcham (Eds.), From
Quantum Fluctuations to Cosmological Structures, Astronomical Society of the Pacific Conference Series,
Vol. 126, 1997, p. 31.
[7] J. Lidsey, A. Liddle, E. Kolb, E. Copeland, T. Barreiro, M. Abney, Rev. Mod. Phys. 69 (1997) 373.
[8] D. Boyanovsky, H.J. de Vega, in: N. Snchez (Ed.), Proceedings of the VIIth Erice Chalonge School on
Astrofundamental Physics, Kluwer Academic, 2000, astro-ph/0006446;
D. Boyanovsky, H.J. de Vega, R. Holman, Phys. Rev. D 49 (1994) 2769;
D. Boyanovsky, D. Cormier, H.J. de Vega, R. Holman, A. Singh, M. Srednicki, Phys. Rev. D 56 (1997)
1939;
D. Boyanovsky, D. Cormier, H.J. de Vega, R. Holman, Phys. Rev. D 55 (1997) 3373.
[9] D. Boyanovsky, D. Cormier, H.J. de Vega, R. Holman, S.P. Kumar, Phys. Rev. D 57 (1998) 2166.
[10] M.O. Scully, M.S. Zubairy, Quantum Optics, Cambridge Univ. Press, 1997;
D.F. Walls, G.J. Milburn, Quantum Optics, Springer-Verlag, New York, 1994.
[11] L.P. Grishchuk, Y.V. Sidorov, Phys. Rev. D 42 (1990) 3413;
L.P. Grishchuk, Y.V. Sidorov, Class. Quantum Grav. 6 (1989) L161;
A. Albrecht, P. Ferreira, M. Joyce, T. Prokopec, Phys. Rev. D 50 (1994) 4807;
C. Kiefer, D. Polarski, A.A. Starobinsky, Int. J. Mod. Phys. D 7 (1998) 455;
C. Kiefer, J. Lesgourgues, D. Polarski, A.A. Starobinsky, Class. Quantum Grav. 15 (1998) L67.
[12] R.D. Pisarski, Nonabelian Debye screening, tsunami waves, and worldline fermions, in: N. Snchez,
A. Zichichi (Eds.), Proceedings of the International School of Astrophysics D. Chalonge, Erice, Italy,
September 415, 1997, Kluwer Academic, Dordrecht, 1998, p. 195.
[13] D. Boyanovsky, H.J. de Vega, R. Holman, S. Prem Kumar, R.D. Pisarski, Phys. Rev. D 57 (1998) 3653.
[14] F.J. Cao, H.J. de Vega, Phys. Rev. D 63 (2001) 045021.
[15] See, for example, J.W. Harris, B. Muller, Annu. Rev. Nucl. Part. Sci. 46 (1996) 71, and references therein.
[16] F. Cooper, S. Habib, Y. Kluger, E. Mottola, J.P. Paz, P.R. Anderson, Phys. Rev. D 50 (1994) 2848;
F. Cooper, Y. Kluger, E. Mottola, J.P. Paz, Phys. Rev. D 51 (1995) 2377;
F. Cooper, E. Mottola, Mod. Phys. Lett. A 2 (1987) 635;
F. Cooper, E. Mottola, Phys. Rev. D 36 (1987) 3114;
F. Cooper, S.-Y. Pi, P.N. Stancioff, Phys. Rev. D 34 (1986) 3831.

Color-flavor transformation for the special

unitary group
B. Schlittgen a , T. Wettig a,b
a Department of Physics, Yale University, New Haven, CT 06520-8120, USA
b RIKEN-BNL Research Center, Upton, NY, 11973-5000, USA
Received 4 December 2001; accepted 15 March 2002
Abstract
We extend Zirnbauers color-flavor transformation in the fermionic sector to the case of the
special unitary group. The transformation allows a certain integral over SU(Nc ) color matrices
to be transformed into an integral over flavor matrices which parameterize the coset space
U(2Nf )/U(Nf ) U(Nf ). Integrals of the type considered appear, for example, in the partition
function of lattice gauge theory. 2002 Elsevier Science B.V. All rights reserved.
1. Introduction
In the context of models describing disordered systems in condensed matter physics,
Zirnbauer recently [1] developed a generalized supersymmetric HubbardStratonovich
transformation which transforms a certain integral over U(Nc ) matrices into an integral
over matrices in the coset space U(n+ + n |n+ + n )/U(n+ |n+ ) U(n |n ). This
transformation reads

i
j
j ij i
dU exp +a
U ij +a + b U
b
U(Nc )

i
j
j
i
exp +a
DNc Z, Z
Zab b
+ b Z
ba +a .
(1)
Here, the -fields are Z2 -graded tensors containing bosonic and fermionic variables.
Integrals of this type (with only fermionic degrees of freedom) appear in lattice gauge
theory, where U represents a link variable. It is therefore natural to think of the indices
E-mail address: tilo.wettig@yale.edu (T. Wettig).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 2 1 5 - 8
156
B. Schlittgen, T. Wettig / Nuclear Physics B 632 (2002) 155172
i, j = 1, . . . , Nc as color, and of the indices a = (, ) and b = (, ) as flavor, where

= 1, . . . , n+ , = 1, . . . , n , and bosonic and fermionic components are denoted by
= B, F . The bar denotes complex conjugation.
The transformation (1) trades the integral over the gauge fields U , which couple the
which
color degrees of freedom of , for an integral over color-singlet fields Z and Z,
on the right-hand
couple the flavor degrees of freedom of . The matrices Z and Z
side are supermatrices whose bosonboson (fermionfermion) blocks satisfy the relation
BB = Z (Z
FF = Z ). The integration measure is given by
Z
BB
FF

= D Z, Z
Sdet 1 ZZ
Nc ,
dNc Z, Z
(2)
is the flat Berezin measure and Sdet denotes the super determinant.
where D(Z, Z)
BB ZBB
Furthermore, integration is constrained to the region in which all eigenvalues of Z
are less than unity.
The color-flavor transformation has been applied to derive a field theory of the random
flux model [2] and, in the context of lattice gauge theories, to derive chiral Lagrangians in
the strong-coupling and large-Nc limits [35]. The extension of the transformation to the
cases of gauge group Sp(2Nc ) and O(Nc ) is relatively straightforward. It was worked out in
Refs. [1,6] and applied to lattice gauge theories in Ref. [4]. However, we would also like to
be able to apply the transformation to the lattice version of quantum chromodynamics
(QCD) where the gauge group is SU(Nc ) (with Nc = 3) rather than U(Nc ). This case
turns out to be more difficult. Since our main interest is in lattice gauge theory with
fermionic matter fields, we restrict ourselves for now to the fermionic sector, in which
the tensor contains only anticommuting components. A first step towards extending the
transformation to the special unitary group was taken in Ref. [5] where it was realized that
the right-hand side of Eq. (1) becomes a sum over disconnected sectors, each characterized
by a different U(1) charge Q, where Q = Nf , . . . , Nf . However, in that paper, only the
terms corresponding to Q = 0 and 1 were given, without proof. Moreover, these terms
involved unspecified constants.
In the present paper, we derive the complete color-flavor transformation for SU(Nc ).
Applications to lattice gauge theory are deferred to future work. It should be noted that
the integral over the gauge group in Eq. (1) does not contain the gauge action but only
the terms coupling quarks to gluons. For applications of the transformation beyond the
strong-coupling limit it will be essential to find a way of including the gauge action in the
integral.
This paper is organized as follows. In Section 2, we state our result for the convenience
of the reader who is less interested in the technical details. Section 3 collects a number
of mathematical preliminaries that are needed for the proof of our result in Section 4.
Conclusions are drawn and an outlook on possible applications of the color-flavor
transformation in lattice gauge theory and generalizations to the case involving both
bosonic and fermionic matter fields is given in Section 5. Finally, the appendix contains
some simple examples to illustrate the transformation.
157
2. Statement of the result

Our result for the color-flavor transformation involving the special unitary group is

j
j
dU exp ai U ij a + ai U ij a
SU(Nc )
f

i
D(Z, Z )
a Zab bi ai Z bi
exp
Q ,
ab
det(1 + ZZ )Nc
= C0
where

Q>0 = CQ det(M)Q + det(N )Q .
0 = 1,
(3)
Q=0
N N
C f f
(4)
Here, and are tensors containing only fermionic variables, with color indices i, j =
1, . . . , Nc , and flavor indices a, b = 1, . . . , Nf . The entries of the Nc Nc matrices M
j
j
and N are given by Mij = ai (1 + ZZ )ab b and N ij = ai (1 + Z Z)ab b . The overall
normalization constant C0 is consistent with vol(SU(Nc )) = 1, see Eqs. (27) and (50). The
constants CQ are given by
CQ =
Q1
(Nc + n)!(Nf + n)!
1
.
N
Q
(Q!) c (Nc !)
n!(Nc + Nf + n)!
(5)
n=0
The matrix Z is a complex matrix of dimension Nf , and the measure of integration in

Eq. (3) reads

D Z, Z = C
dZ dZ =
dZ dZ
det(1 + ZZ )2Nf
Nf

with
d Re Zab d Im Zab
(6)
a,b=1

and a constant C chosen so that D(Z, Z ) = 1, see Eq. (25).
The next two sections are dedicated to the proof of Eq. (3). To a large extent, we shall
proceed along the lines of Zirnbauers original proof in Ref. [1], but several extensions are
needed to adapt it to the case of SU(Nc ). We shall give a complete account of all relevant
details to make the discussion self-contained.
Let us note that the above result can easily be generalized to the case where the tensors
and contain unequal numbers of flavors, i.e., the flavor index on () runs from 1
to Nf1 (Nf2 ). In this case, the matrix Z has dimension Nf1 Nf2 , and the sum over Q in
Eq. (3) is replaced by
N1
1+
N2
f

1
CQ
det(M)Q
Q=1
with
1
CQ
2)
(CQ
f

2
CQ
det(N )Q ,
Q=1
given by the CQ from Eq. (5) with Nf replaced by Nf2 (Nf1 ).
(7)
158
3. Mathematical preliminaries
In this section, we introduce the mathematical framework needed to prove Eq. (3) and
quote some of the relevant intermediate results. This will allow us to give the proof in
Section 4 in a relatively straightforward manner.
3.1. Generalized coherent states and projection onto the SU(Nc ) neutral sector
We introduce two sets of fermionic creation and annihilation operators, cai , cai and dai ,
where in each case, a runs from 1 to Nf , and i runs from 1 to Nc . These operators
may be interpreted as quantized versions of the classical Grassmann variables ai and ai ,
respectively. For simplicity, we also introduce an index A = 1, . . . , 2Nf , such that

for A Nf ,
ci
i
cA
(8)
= Ai
dANf for A > Nf
dai ,
i
and similarly for cA
. In general, we shall use lower case letters a, b etc. to denote indices
ranging from 1 to Nf , and upper case letters A, B etc. to denote indices ranging from 1 to
2Nf . The creation and annihilation operators satisfy canonical anticommutation relations,
i j
i j
ca , cb = da , db = ij ab .
(9)
We also define a Fock vacuum |0 which is annihilated by both cai and dai , i.e., cai |0 =
dai |0 = 0. The Fock space is generated by acting on |0 with all possible combinations of
creation operators cai and dai . In the dual space, we correspondingly have a vacuum state
0| with the property

0|cai =
0|dai = 0.
The set of all bilinear products of one creation operator with one annihilation operator
forms a set of generators of a gl(2Nf Nc ) algebra,
1
ij
j
Eab cai cb ab ij ,
2
ij
Ea,Nf +b cai db ,
1
ij
j
(10)
ENf +a,Nf +b dai db ab ij ,
2
where a, b = 1, . . . , Nf and i, j = 1, . . . , Nc . For convenience, we have chosen these
operators not to be traceless. They satisfy the usual commutation relations,

ij
kj
k"
i"
= CB kj EAD
EAB , ECD
(11)
AD i" ECB .
ij
j
ENf +a,b dai cb ,
This algebra has two subalgebras that are important for our purposes, namely, gl(2Nf ),
Nc ii
which is generated by
i=1 EAB , and sl(Nc ), which is generated by

2Nf
ij
ij
E
EAA ; i = j
A=1
and

Nc Nc
ii
EAA EAA ; i = 1, . . . , Nc 1 .
H
2Nf
A=1
159
These two subalgebras commute.

An essential part of Zirnbauers proof is the projection of states in the Fock space onto
the color-neutral sector. This is the subset of Fock space whose elements are invariant under
(in our case) SU(Nc ) rotations, i.e., for a color-neutral state |N , we have
E ij |N = 0
and Hi |N = 0
(12)
for all generators in the sl(Nc ) subalgebra. The basic idea of the proof is to derive two
different implementations of a projection operator onto the color-neutral sector. One of
these is obtained by integrating over the gauge group, corresponding to the left-hand side of
Eq. (3) (see Section 3.3). The other one, corresponding to the right-hand side of Eq. (3), is
obtained using some properties of generalized coherent states and integrating over a certain
coset space G/H (see this and the next subsection). Identification of the two projection
operations then establishes Eq. (3).
Eq. (12) may equivalently be written as
Nf

i j
j
ca ca + dai da |N = (Nf + Q) ij |N .
(13)
a=1
Here, Q can take on integer values between Nf and Nf . These bounds on Q are due to
the fact that the Fock space contains only fermions, and the operator on the left-hand side
of Eq. (13) simply counts the number of fermions in the state |N . (Note also that this is
the place where the differences to the original case of U(Nc ) start to show up. In the latter
case, the color-neutral sector is invariant under U(Nc ), and there is an additional diagonal
generator HNc which annihilates color-neutral states. This implies that Q = 0 in Eq. (13)
so that there is no sum over Q in Eq. (3).)
In order to derive the projector onto the color-neutral sector, we consider the action
of the group U(2Nf ) on the Fock space. (The restriction of Gl(2Nf ) to the submanifold
U(2Nf ) leads to a projection onto the SU(Nc ), rather than Sl(Nc ), invariant sector,
and it leads to a Riemannian integration domain on the right-hand side of Eq. (3).) In
the following, we show how to associate to each value of Q a particular irreducible
representation of U(2Nf ).
From Eq. (13), we see that the color-neutral sector splits into subsectors labeled by Q,
with fermion occupation number Nc (Nf + Q), respectively. Hence, the Young diagram
of a representation of a group acting irreducibly in such a subsector of Fock space
necessarily has Nc (Nf + Q) boxes. Since the fermionic operators obey anticommutation
relations, permutations of the fermions furnish a totally antisymmetric representation of
the permutation group SNc (Nf +Q) , with the Young diagram shown in Fig. 1(a).
The action of the group SU(Nc ) on a state in the subsector labeled by Q leaves that state
invariant. Therefore, the representation of SU(Nc ) in this sector is the trivial one. Again, the
corresponding Young diagram is made up of Nc (Nf + Q) boxes. It is shown in Fig. 1(b).
Viewed as a representation of SNc (Nf +Q) , it is totally symmetric under permutations of
flavor indices and totally antisymmetric under permutations of color indices.
We now seek to determine the representation of U(2Nf ) acting in the subsector labeled
by Q. We will conclude below that this representation is irreducible, but for the moment
we leave open the possibility that it may be reducible, in which case we decompose
160
Fig. 1. Irreducible representations defined by the different Q sectors.
it into irreducible parts labeled by their corresponding Young diagrams. These Young
diagrams also define representations of SNc (Nf +Q) . The subsectors labeled by Q are
invariant subspaces under the action of SU(Nc ) U(2Nf ) (recall that the generators of
these two groups commute). Therefore, we obtain a representation of SNc (Nf +Q) by taking
the direct product of the representation of SNc (Nf +Q) shown in Fig. 1(b) and the direct sum
of the representations of SNc (Nf +Q) given by the Young diagrams corresponding to the
irreducible representations of U(2Nf ) contained in the decomposition mentioned above.
The representation of SNc (Nf +Q) thus obtained must contain the totally antisymmetric
representation given by the Young diagram in Fig. 1(a).
We now use the fact that the totally antisymmetric representation of the permutation
group is contained only in the direct product of a given irreducible representation with its
associate, which has the transposed Young diagram. Therefore, we conclude that the direct
sum representation of U(2Nf ) mentioned above must contain the irreducible representation
shown in Fig. 1(c). In fact, it cannot contain any other irreducible representation, since the
direct product of the representation given by Fig. 1(b) with such a representation does not
lead to a totally antisymmetric representation of SNc (Nf +Q) . Also, the representation of
Fig. 1(c) cannot occur more than once, since the direct product with Fig. 1(b) would then
lead to more than one totally antisymmetric representations of SNc (Nf +Q) . This possibility
can be excluded because SNc (Nf +Q) acts irreducibly in the subsector labeled by Q. Thus,
Fig. 1(c) defines an irreducible representation of U(2Nf ) in this subsector. When referring
to this representation, we shall simply denote it by Q.
The corresponding highest-weight vector is obtained by acting on the Fock vacuum with
a product of Nc (Nf + Q) creation operators [7],

2 2
Nc Nc

Nc
1
2
|Q c11 c21 cN
c1 c2 cN
c1 c2 cN
|0 .
f +Q
f +Q
f +Q
(14)
161
The action of the group U(2Nf ) on the Fock space is now defined by the following
mapping of group elements g U(2Nf ) to operators Tg ,
i

Tg = exp cA
(log g)AB cBi .
(15)
Zirnbauer showed in Ref. [1] that this is indeed a well-defined map, and furthermore,
that it is a homomorphism, i.e., Tg1 Tg2 = Tg1 g2 . Thus, the mapping g Tg defines
a (reducible) representation of U(2Nf ). We choose the color-neutral sector to be the carrier
space for this representation. It decomposes into invariant subspaces, the carrier spaces of
i
obey the important transformation property
the (irreducible) representations Q. The cA
i
i
i T 1 = g c i .
1
Tg cA Tg = cB gBA , and similarly we have Tg cA
g
AB B
Next, we set up a system of generalized coherent states for each representation Q
defined above. A comprehensive introduction to generalized coherent states is given in
Ref. [8]. They form an overcomplete set of states, which have a number of nice properties.
Most importantly for our purposes, they allow a resolution of the identity operator.
For a general Lie group G and an irreducible unitary representation Tg of G, such a set
of states is obtained as follows. Take any state in the carrier space of Tg , say |T , and
act with all elements Tg on this state, resulting in the set {Tg |T }. In general, not all of
the resulting states will be distinct. Let H be the maximal subgroup of G such that for
all h H, Th |T |T . The subgroup H is called the isotropy subgroup of |T , and
the set of coherent states may be parameterized without overcounting by the elements of
the coset space G/H . A particularly convenient state to pick as the starting vector is the
highest- (or lowest-) weight vector of the representation Tg . In this case, the corresponding
isotropy subgroup will contain the Cartan subgroup of G. (Note that it is also possible
to work with a subgroup H that is not maximal. In this case a parameterization by the
elements of G/H would not eliminate all the redundant states. This is of no concern as
long as for integrals over G/H the measure of integration is suitably normalized.)
Now consider the representation Q of U(2Nf ). The highest-weight vector is given in
Eq. (14), and hence the corresponding set of coherent states is {Tg |Q }. The resolution of
the identity takes the form [8]

1Q = CQ
d(g)Tg |Q
Q |Tg1 ,
(16)
U(2Nf )
where d(g) is the invariant Haar measure for the group U(2Nf ) and CQ is a normalization constant which will be determined explicitly in Eq. (49). We normalize the measure
so that the group volume is unity. The proof that 1Q must be proportional to the identity
uses the fact that it commutes with all group elements, by virtue of the invariance of the
measure, and an invocation of Schurs lemma. Note that 1Q is actually a projector onto the
part of Fock space that forms the carrier space of Q since it not only is the identity operator
in this sector but also annihilates all states outside this sector. (This is easy to see, because
it clearly annihilates states which are not color neutral, as well as states with a different occupation number, i.e., a different value of Q.) The projection P onto the SU(Nc ) invariant
162
sector of Fock space can therefore be written as a sum over Q of the 1Q s,

P=
Nf

1Q .
(17)
Q=Nf
In the next section, we show how to remove the redundancy in the coherent states and
how to parameterize them.
3.2. Parameterization of coherent states in terms of coset spaces
As mentioned in the previous section, there is some redundancy in the set {Tg |Q }.
Let us first focus on the case of Q = 0 and determine the coset space U(2Nf )/H needed
to label the corresponding coherent states. The highest-weight vector of the representation
Q = 0 is
2 2
Nc Nc

Nc
1
2
|0 c11 c21 cN
(18)
c1 c2 cN
c1 c2 cN
|0 ,
f
f
f
j
and thus, this state has the property that cai |0 = db |0 = 0. It immediately follows that
for elements of U(2Nf ) of the form h = diag(h+ , h ), where h are unitary Nf Nf
matrices, the corresponding Fock operators Th leave the state |0 unchanged,

Th |0 = exp cai (log h+ )ab cbi + dai (log h )ab dbi |0

= exp Nc Tr log(h+ ) |0 = |0 det(h+ )Nc .
(19)
Hence, we can parameterize the coherent states by elements of G/H , where G = U(2Nf )
and H = U(Nf ) U(Nf ). To do so, we follow Ref. [1] and define : G G/H to be
the canonical projection which assigns to each group element g G the corresponding
equivalence class gH . Choosing a representative group element s((g)) from each such
equivalence class, we decompose an arbitrary group element g into a product g =
s((g))h(g), where s((g)) takes values in G and h(g) takes values in H . The map
s : G/H G is called a local section of the bundle : G G/H .
In order to arrive at an expression for s((g)), write the (2Nf 2Nf )-dimensional
matrix g in terms of (Nf Nf )-dimensional blocks,

a b
g=
.
(20)
c d
Multiplying g by h = diag(h+ , h ) on the right leaves the products Z = ca 1 and
= bd 1 invariant. From the unitarity condition for g, it is also immediate that Z
= Z .
Z
A particular s((g)) is then given by

s (g) s Z, Z

(1 + Z Z)1/2 Z (1 + ZZ )1/2
=
Z(1 + Z Z)1/2
(1 + ZZ )1/2

1 0
(1 + Z Z)1/2
1 Z
0
=
(21)
.
Z 1
0
(1 + ZZ )+1/2
0
1
163
(Note that this parameterization differs slightly from Zirnbauers notation, because we have
chosen the state |0 so that it is annihilated by cai and dai , which reverses the roles of cai
and dai with respect to Ref. [1].) We can thus label the coherent states for Q = 0 by Z, and
obtain
|Z Ts(Z,Z ) |0

1
1
= exp dai Zab cbi exp cai log 1 + Z Z ab cbi + dai log 1 + ZZ ab dbi
2
2
i i
exp ca Zab db |0

Nc /2

= exp dai Zab cbi |0 det 1 + Z Z
(22)
.
Similarly, we have

i
1
Nc /2
Z|
0 |Ts(Z,Z
0 | exp cai Zab

db .
) = det 1 + ZZ
(23)
Regarding integration over the coset space U(2Nf )/U(Nf ) U(Nf ), a U(2Nf ) invariant
measure exists so that

d(g)f (g) =
D Z, Z
d(h)f s (g) h(g) .
(24)
U(2Nf )
U(2Nf )/H
In fact, this measure is given by (cf. [1,5])

D Z, Z = C
dZ dZ =
dZ dZ
det(1 + ZZ )2Nf
Nf

with
d Re Zab d Im Zab .
a,b=1
As mentioned in Section 2, the constant C is chosen such that

expression for C is
D(Z, Z ) = 1. An explicit
Nf 1
C=
(6)
Nf2
(Nf + n)!
,
n!
(25)
n=0
as computed, among many other results, in a very interesting article [9] which appeared
after the original submission of the present paper.
The resolution of unity in the Q = 0 sector thus becomes

10 = C0
d(g)Tg |0
0 |Tg1
U(2Nf )
= C0

1
D Z, Z Ts(Z,Z ) |0
0 |Ts(Z,Z
)
U(2Nf )/H
= C0

D Z, Z |Z
Z|,
(26)
164
where we have set vol(H ) = 1. As mentioned in Section 2, our choice of normalization

implies that

C0
D(Z, Z )
= 1,
det(1 + ZZ )Nc
(27)
consistent with vol(SU(Nc )) = 1. The constant C0 is given in Eq. (50).

This essentially completes the discussion of the Q = 0 sector which was necessary to
derive the color-flavor transformation for U(Nc ). For the gauge group SU(Nc ) we now
have to consider the nonzero values of Q in Eq. (17). The isotropy subgroup for the
highest-weight state belonging to representations with Q = 0 is not U(Nf ) U(Nf ) but
rather HQ = U(Nf + Q) U(Nf Q). Parameterizing the coherent states in terms of
the coset spaces U(2Nf )/HQ would result in a sum over Q of integrals with different
parameterizations and different integration domains for each Q. This would lead to
a correct result whose usefulness, however, is rather questionable. Instead, we seek to
express the projectors 1Q in terms of the variables Z, Z just derived and integrate over
U(2Nf )/HQ=0 . This can be achieved by relating the highest-weight vector for Q = 0 to
|0 by the action of an appropriate combination of creation or annihilation operators on
|0 . Consider for instance the case of Q > 0,

1Q = CQ
d(g)Tg |Q
Q |Tg1
U(2Nf )
= CQ
U(2Nf )
1

Nc
d(g)Tg dQ
dQ
d11 d1Nc |0
Nc
1

1
Tg .
dQ

0 | d1Nc d11 dQ
(28)
and each pair of ds. We also insert

We can now insert Tg1 Tg between each pair of ds
N
such a term between d1 c and |0 , and between
0 | and d1Nc . Now, make use of the fact
that Tg dai Tg1 = ci i gAi ,N +a . Recall that Aia is a label which runs over all values from 1
Aa
to 2Nf . Similarly, Tg dai Tg1 = gN
i
f +a,Ba
cBi i . Next, we perform a coset decomposition of

a
the group elements g, g = s(Z, Z )h(g). Note that Tg |0

0 |Tg1 = |Z
Z|. Thus, we
Fig. 2. The (reducible) product of fundamental representations of U(Nf ) contains an irreducible representation
of U(Nf ) with symmetric color indices and antisymmetric flavor indices. Here Q > 0.
165
arrive at the following expression for the projector 1Q in the case of Q > 0,

1Q = CQ D Z, Z

1
1
N
N
N
N
aQ aQc a11 a1 c bQ
bQc b11 b1 c
d(h )(QQ)(11)
(QQ)(11)

U(Nf )

sA1 ,Nf +a 1 sANc ,N +a Nc sA1 ,Nf +a 1 sANc ,N +a Nc
f
f
Q
Q
1
1
Q
Q
1
1

s
s
Nc
Nc sN +b1 ,B 1
Nc
Nc sN +b1 ,B 1
Nf +bQ ,BQ
Nf +b1 ,B1
f
f
Q Q
1 1

N
N
N
1
1
cA
Nc c cA
Nc c |Z
Z| c Nc c cB1 1 cNNc c cB1 1 ,
1 c
1 c
AQ
A1
B1
BQ
(29)

where = (h ) denotes the representation matrix of the tensor product of Nc Q
fundamental representations of U(Nf ) (cf. Fig. 2),
1

Nc 1
aQ aQ
a1 a1Nc
Nc 1
Nc
1
bQ bQ b1 b1

(h )a 1 , b1 (h )a Nc , bNc (h )a 1 , b1 (h )a Nc , bNc .
Q
(30)
The case of Q < 0 is very similar, the analog of Eq. (28) being

1
1
Nc
Nc
1Q = CQ
cNf cN
|0
d(g)Tg cN
cN
f +Q+1
f +Q+1
f
U(2Nf )
Nc
1
Nc
1
1
cNf +Q+1 cN
Tg .
cN

0 | cN
f
f +Q+1
f
(31)
3.3. Fermi coherent states

We shall also need Fermi coherent states. These are defined as

exp ai cai + dai ai |0 ,
(32)
and they span the entire Fock space. In the same way that averaging over rotations of
a vector in R3 around the z-axis projects out the z-component of that vector, i.e., the part
that is invariant under such a rotation, we can project out the color-neutral part of a Fermi
coherent state by averaging over SU(Nc ) rotations,

P exp ai cai + dai ai |0

j
j
dU exp ai U ij ca + dai U ij a |0 .
=
(33)
SU(Nc )
The identification of the two projection operators in Eqs. (17) and (33) now enables us to
complete the proof of the color-flavor transformation.
166
4. Proof of the result

Starting from the left-hand side of Eq. (3), we obtain

j
j
dU exp ai U ij a + ai U ij a
I=
SU(Nc )

j
j
dU
0 | exp ai dai cai ai exp dai U ij a + ai U ij ca |0
=
SU(Nc )

=
0 | exp ai dai cai ai P exp dai ai + ai cai |0
Nf

0 | exp ai dai cai ai 1Q exp dai ai + ai cai |0 .
(34)
Q=Nf
The second line is obtained from the first one using the anticommutation relations and
the action of the creation and annihilation operators on the state |0 . To obtain the third
line, we have used Eq. (33). The implementation of the projection operator onto the colorneutral sector given by Eq. (17) then yields the fourth line.
The next step is to evaluate each term in the sum over Q. As before, consider the case
of Q > 0. From Eqs. (29) and (34), we need to consider matrix elements of the form

1

1
0 | exp ai dai cai ai cA

NNc c cA
NNc c exp dai Zab cbi |0
1 c
1 c
AQ
A1

1

NNc
NNc ,
11
1
= exp ai Zab bi
c
c
A
A
Q
AQ
A1
(35)
where we have again used the properties of the creation and annihilation operators. We
have also introduced the shorthand notation

i = i Z, i
(36)
ai = i Zba and
i
i
for
b
Nf +a = a . Similarly, we find that

i Nc
db c Nc cB1 1 cNNc c cB1 1 exp dai ai + ai cai |0
0 | exp cai Zab

B1
BQ

bi NNcc B11 NNcc B11 ,

= exp ai Zab
B1
where
i =
Z i
i
BQ
(37)

.
(38)
We also need the matrix-vector products

i s Z, Z = 0, i
and s Z, Z i =
0
i

,
(39)
167
where i = i (1 + ZZ )1/2 and i = (1 + ZZ )1/2 i . Putting everything together and

using Eq. (29), we find that

I = C0
Nf
i

D(Z, Z )
i
i
i
exp a Zab b a Zab b

Q ,
det(1 + ZZ )Nc
(40)
Q=Nf
where for Q > 0, we have

1
1
N
N
N
N
aQ aQc a11 a1 c bQ
bQc b11 b1 c
CQ
Q =
d(h )(QQ)(11)
(QQ)(11)
C0
U(Nf )

1a 1 NNc c 1a 1 NNc c
aQ
a1

NNcc b11 NNcc b11 .
b1
bQ
(41)
From the definition of in Eq. (30) it is clear that the expression in square brackets is

totally symmetric under the exchange of aai with aai , and of bai with bai . Furthermore, the
are
product of the s is totally antisymmetric under exchange of aai with aai , since the s
Grassmann variables. Similarly, the product of the s is antisymmetric with respect to the
interchange of bai with bai . In each term, we now pick out only those pieces of that have
the correct symmetry properties, i.e., symmetric in color and antisymmetric in flavor. To
this end, we decompose the (reducible) product of fundamental representations of U(Nf ),
represented by , into irreducible representations. Of these, only the one with the correct
symmetry properties, shown in Fig. 2, survives the contraction of indices. We can thus
replace by the representation matrix corresponding to this irreducible representation,
which we denote by , and obtain

1
CQ
Nc
Nc
1
1
Q =
a 1 Nc a 1 Nc
Q Q
1 1
Q a Q
1 a 1
C0 (Nc !)Q
Q
1

1
sgn 1 sgn Nc
N
(Q!) c 1 N

c
c
c
c1 1

c1 1
cNN
cNN
c (Q)
c (1)
(Q)
(1)
Q
1
1
Q

cQ cQ c1 c1
1

a a Nc a 1 a Nc b1 bNc b1 bNc
d(h ) 1Q NQc 1 1 N1c 1Q NQc 11 N1 c
dQ dQ d1 d1
U(Nf )
1
(Q!)Nc

1 Nc

sgn 1 sgn Nc

d 11
d 11
d NNc c
d NNc c
(Q)
(1)
(Q)
(1)
1
Q
Q
1
168

N
1
Nc
1
1
c
Nc b1 Nc b1
.
Q Q
1 b1
1 1
Q bQ
(Nc !)Q
(42)
Here, and denote permutations, where the upper (lower) indices correspond
to permutations of the color (flavor) indices. As the hypermatrix is simply the
representation matrix corresponding to the irreducible representation of U(Nf ) shown in
Fig. 2, we may use the group theoretic result that for irreducible unitary representations

k and k of the compact Lie group G,

1
d(g) k (g)ij k (g)i j = ii jj kk d(g).
(43)
k
G
Here, k is the dimension of the representation

of G. We normalize the group volume
of U(Nf ) to unity.
For convenience, we define N
n to be the dimension of a representation of U(N ) which
has a rectangular Young diagram with n rows and Nc columns. It is given by Weyls
formula,

1i<j N (i j i + j )
N
n =
(44)
,
(N 1)!(N 2)! 0!
where i and j enumerate the rows of the Young diagram, and i is the length of the ith row.
Thus, i = Nc if 1 i n and i = 0 otherwise. For the present application of Eq. (43),
N
the k are simply given by Qf .
It is now a matter of unraveling the expression for Q in Eq. (42). An important part of
this expression is

1
1 1 Nc Nc 1 1 Nc Nc
Q a Q
1 a 1
Q a Q
1 a 1
(Nc !)2Q 1 Q
1 Q
N
a1
a 1
N
a c
a c
b1Q NQc b11 N1 c
bQ

Nc
Nc
1
1 b11
1 b1
b1
Nc
Nc
1
1
Q bQ
Q bQ

1
sgn(1 ) sgn(Q ) sgn(1 ) sgn(Q )
=
2Q
(Nc !) 1 Q
1 Q
1 (1)

1 (1)
1 (Nc )
1 (Nc )
11
Q1
QNc
1Nc
aQ
aQ
a1 c
a1
a1
a1
1 (Nc )
1 (N )
1 (1)
1 (1)
c
NQ
N1
11
1Q

aQc
Nc

1
sgn(
)
sgn()
a (i) a(i)
(Nc !)2Q ,
i=1
aQ
Q
(45)
169
The term in parentheses in the last line of Eq. (45) is the Laplace expansion for
a determinant (more precisely, Nc ! copies thereof ). Thus, we obtain

CQ
det(M) Q
Q = N
(46)
,
Nc !
Qf C0 (Q!)Nc
where the matrix M has entries Mij = ai (1 + ZZ )ab b .
For Q < 0, the argument is analogous, and we find

C|Q|
det(N ) |Q|
.
Q = N
Nc !
f C0 (|Q|!)Nc
j
(47)
|Q|
The matrix N has entries N ij = ai (1 + Z Z)ab b .

Finally, let us find the constants CQ . To this end, we evaluate the matrix element

1 =
Q |1Q |Q = CQ
(48)
d(g)
Q |Tg |Q
Q |Tg1 |Q .
U(2Nf )
Note that
Q |Tg |Q is simply a diagonal element of the representation matrix Q (g)
of the representation Q of the group U(2Nf ) in Fig. 1(c). Similarly,
Q |Tg1 |Q is the
complex conjugate of this matrix element. Hence, we can once again use Eq. (43) together
with Weyls formula (44) for the dimension of the representation Q of U(2Nf ) shown in
Fig. 1(c) to obtain
2N
CQ = Nf f+Q .
(49)
In particular, the overall normalization constant in Eq. (3) is

Nf 1
C0 =
n!(Nc + Nf + n)!
.
(Nc + n)!(Nf + n)!
(50)
n=0
Note that CQ = CQ . Thus, the constants CQ which appear in Eq. (4) are given by
2N
Nf f+Q
1
CQ =
.
(Q!)Nc (Nc !)Q Nf 2Nf
Q
(51)
Nf
Using Eq. (44), a number of cancellations then leads to the result quoted in Eq. (5). This
completes the proof.
5. Conclusions and outlook

We have generalized Zirnbauers color-flavor transformation in the fermionic sector to
the case of gauge group SU(Nc ). This transformation is expected to have a number of
interesting applications in lattice gauge theories with gauge group SU(Nc ). In fact, a first
application has already appeared in Ref. [5] where a low-energy effective theory, valid in
170
the strong-coupling limit, was derived in the vacuum (Q = 0) and one-baryon (Q = 1)

sectors.
There are a number of obvious open problems. The most difficult one appears to be
the inclusion of the gauge action beyond the strong-coupling limit. Another issue is the
problem of chiral symmetry on the lattice which has recently been solved in a variety of
ways that can all be traced back to the GinspargWilson relation [10]. From the point of
view of the color-flavor transformation, is seems easiest to address this problem in the
domain-wall fermion approach [11].
Finally, it would be interesting to extend the color-flavor transformation with gauge
group SU(Nc ) to the bosonic sector and to the supersymmetric case where the tensors
and contain both bosonic and fermionic variables. This results in some complications
since the sum over Q in Eq. (17) then extends from to . Work in this direction is in
progress.
Acknowledgements
This work was supported in part by the US Department of Energy under contract
No. DE-FG02-91ER40608. We thank S. Shatashvili for interesting conversations.
Appendix A
In this appendix we give some simple examples to illustrate the color-flavor transformation.
Example 1. Nc = 1, Nf = 1.
This is the simplest case, and it was already mentioned in Ref. [5]. Integration over
SU(1) simply amounts to the evaluation of the integrand at unity, and so we have

+ U

dU exp U
SU(1)
+ )
+
.
= exp(
= 1 +
+
(A.1)
On the flavor side of the transformation, we have the integral over the complex number
z = x + iy,

2
dx dy
z
(A.2)
exp z
(0 + 1 ).
(1 + zz )
+ ).
It is straightforFrom Eq. (4), we find that 0 = 1 and 1 = (1/2)(1 + zz )(

ward to expand the exponential and evaluate the integral in Eq. (A.2) and to show that the
resulting expression is equal to the right-hand side of Eq. (A.1).
171
The integral over SU(2) color matrices can be evaluated by parameterizing them as
i

e cos ei sin
U = i
(A.3)
with 0 ; 0 , 2.
e sin ei cos
2
The corresponding Haar measure is dU = (1/2 2 ) sin cos d d d. We thus obtain

dU exp i U ij j + i U ij j
SU(2)
= 1 + 1 1 2 2 + 1 1 2 2

1
+ 1 1 1 1 + 1 2 2 1 + 2 1 1 2 + 2 2 2 2 .
2
For the flavor integral, we have (again with z = x + iy)

3
dx dy
exp i z i i z i (0 + 1 ).
(1 + zz )
(A.4)
(A.5)
Here, 0 = 1 and 1 = (1/3)(1 + zz )2 ( 1 1 2 2 + 1 1 2 2 ). Again, it is straightforward to expand the exponential and evaluate the integral. The result is equal to the
right-hand side of Eq. (A.4).
In this case, as in Example 1, integration over SU(1) is trivial, and we have

dU exp a U a + a U a
SU(1)
= exp( a a + a a )
= 1 + a a + a a + a a b b + 1 1 2 2 + 1 1 2 2
+ a a 1 1 2 2 + 1 1 2 2 a a + 1 1 2 2 1 1 2 2 .
Note that repeated indices are summed over. The corresponding flavor integral is

72
dZ dZ
exp a Zab b a Zab

b (0 + 1 + 2 ),
4
det(1 + ZZ )5
where
0 = 1,
(A.6)
(A.7)

1 = (1/3) a 1 + ZZ ab b + a 1 + Z Z ab b ,
and

2 = (1/6) det 1 + ZZ 1 1 2 2 + 1 1 2 2 .
This integral can be evaluated using the singular value decomposition of Z given by
Z = U V . Here, U U(2), V U(2)/[U(1) U(1)], and = diag(1 , 2 ), where
172
1 , 2 0. In these coordinates,

det 1 + ZZ = det 1 + 2 = 1 + 21 1 + 22 .
The Jacobian of the coordinate transformation is J = 1 2 (21 22 )2 . The measure of
integration is thus obtained using

2
dZ dZ = dU dV d1 d2 1 2 21 22 .
(A.8)

4
From our normalization of the measure it follows that dU dV = 2 .
We now expand the exponential in Eq. (A.7) and drop all terms for which the number
of elements of U is not equal to those of U , since
integrated
such terms vanish when
over U . We also make repeated use of the identity dU Uij Ukl = (1/2)il j k dU . Next,
we perform the integrals over the a and use the unitarity of U and V . This eliminates all
dependence on U and V from the integrand. After performing the trivial integration over
U and V , one obtains the same result as in Eq. (A.6).
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
M.R. Zirnbauer, J. Phys. A 29 (1996) 19.

A. Altland, B.D. Simons, Nucl. Phys. B 562 (1999) 445.
J. Budczies, Y. Shnir, AIP Conf. Proc. 508 (2000) 172.
T. Nagao, S.M. Nishigaki, Phys. Rev. D 64 (2001) 014507.
J. Budczies, Y. Shnir, hep-lat/0101016.
M.R. Zirnbauer, chao-dyn/9810016.
N. Read, S. Sachdev, Nucl. Phys. B 316 (1989) 609.
A. Perelomov, Generalized Coherent States and Their Applications, Springer, Berlin, 1986.
J. Budczies, S. Nonnenmacher, Y. Shnir, M.R. Zirnbauer, hep-lat/0112018.
P.H. Ginsparg, K.G. Wilson, Phys. Rev. D 25 (1982) 2649.
D.B. Kaplan, Phys. Lett. B 288 (1992) 342;
V. Furman, Y. Shamir, Nucl. Phys. B 429 (1995) 54.

Effective action for bubble nucleation rates

Ian G. Moss, Wade Naylor 1
Department of Physics, University of Newcastle Upon Tyne, NE1 7RU UK
Received 11 July 2001; accepted 28 March 2002
Abstract
We develop a method to calculate the prefactor in the expression for the bubble nucleation
rate. A fermion with Yukawa coupling is considered where a step potential can be used as a
good approximation in the thin wall limit. Corrections due to thicker walls are investigated by
perturbing about the thin wall case. We derive the thermal one loop effective action, calculating
it numerically, and find that the prefactor in the nucleation rate can both suppress and enhance, for a
given temperature, when the usual renormalisation conditions are applied to the effective potential.
2002 Elsevier Science B.V. All rights reserved.
PACS: 03.70.+k; 98.80.Cq
1. Introduction
Bubble nucleation can occur in a first order phase transition from the false to the true
vacuum. The bubble nucleation rate per unit volume per unit time, written in the language
of Coleman [1], is /V = AeB , where B is the classical Euclidean action of the bubble
and A is the one-loop contribution including zero and negative modes.
The purpose of this paper is to present a method that enables one to calculate the
nucleation rate in the thin wall limit. Techniques such as the derivative expansion break
down for this type of background [7]. The thin wall limit is a possible scenario in
electroweak theory with multiple Higgs fields [2]. It is also useful as a possible explanation
for the generation of the baryon asymmetry we observe today [3]. In the interests of brevity
we focus on just the fermion fields. However, the general discussion includes scalar and
E-mail address: wnaylor@vega.ess.sci.osaka-u.ac.jp (W. Naylor).

1 Current address: Department of Earth and Space Science, Osaka University, Osaka 560-0043, Japan.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 2 4 6 - 8
174
I.G. Moss, W. Naylor / Nuclear Physics B 632 (2002) 173188
spinor fields for completeness. The calculation of other fields, using the method presented,
is currently under way [4].
Previous work on nucleation rates includes an analysis of the prefactor for scalar fields
coupled to fermions by Gleiser, Marques and Ramos [5], who computed the determinant
using the derivative expansion. Issues regarding which loop corrections should be included
in the bounce solution were investigated. Garriga [6] calculated the determinantal prefactor
analytically, assuming the free energy can be approximated by that of a massless field
living on the surface of a membrane. This was for a scalar field in the thin wall
approximation (corrections due to thicker walls were also studied) at finite temperature,
where enhancement of the nucleation rate was found. Kripfganz, Laser and Schmidt [7]
computed the prefactor for the one loop Higgs fluctuations at the electroweak phase
transition, using the derivative expansion as an approximation.
A particularly useful numerical method uses a theorem on functional determinants that
can be found in Colemans work [8]. Baacke and Kiselev [9] develop an exact numerical
scheme to work out the one loop corrections for a scalar field at finite temperature,
using Colemans theorem. Thick and thin walls were considered but an infinitely thin
wall ( = 1 in their notation) was not. Hence, there is no direct method of comparison
with [6]. However, they also found an enhancement of the nucleation rate (depending
on the choice of renormalisation scheme). Baacke [10] then studied vector bosons in the
t HooftFeynman gauge, generalising Colemans theorem to the matrix case. The results
were compared to those in [7], where there was good agreement for small (thick wall)
bubbles and strong deviations between the results for large (thin wall) bubbles, as expected.
The nucleation rate was suppressed for this case. Also, Baacke and Srig [11] calculated
the fermionic fluctuation determinant using Colemans theorem and a gradient expansion
for comparison. For an optimal choice of renormalisation, they found that the rate was
enhanced at the electroweak temperature.
Brahm and Lee [12] used a different numerical procedure based on the phase shift
method and similar to the one we shall employ (using some old results due to Schwinger
[13]). They computed the prefactor for scalar fields at finite temperature for the thin wall
limit, based on the assumption that the surface free energy is equal to that of a domain
wall (thick walls were also considered). The WKB approximation was used for the high
energy modes to improve convergence of the exact result. The results were compared with
the effective potential and derivative expansion approximations (see references therein).
Further work has been done by Mnster and Rotsch [14], who calculated the prefactor
in the nucleation rate for a scalar field using a PschlTeller potential and heat kernel
coefficients to remove the ultraviolet divergences of the theory. Mnster et al. [15] have
compared this work to an entirely different method (not relying on saddle point evaluation)
that uses coarse graining and renormalisation group techniques. They find good agreement
between the two in the region of validity. Further work has been done using this coarse
graining method and we refer the reader to [15] and the references therein.
In our method we shall use phase shifts and relate the prefactor to the heat kernel. The
theory is regulated by subtracting off the relevant heat kernel coefficients. We consider
fermions with a Yukawa coupling and a step function profile to begin with. Using the
step function gives an exact analytic expression for the phase shift, making the calculation
more manageable. For the fermion case, the thin wall limit introduces problems of its own
175
because the derivative of the mass term leads to a delta function in the potential. As far
as we are aware, this is the first time this specific case has been investigated, although the
method used in [10] or [12] could be applied. Results are at finite temperature (not only
in the high temperature limit) and the method can be extended to the zero temperature
case. Also, the technique involves a simple regularisation step, unlike methods based
on Colemans theorem which require evaluating uniform asymptotic expansions of the
relevant field equations.
The eigenvalues in the determinant are found using partial wave analysis and phase
shifts, with the eigenmodes discretised by putting them in a sphere of large radius that
we let tend to infinity (not to be confused with the bubble wall radius at R). Thus, we must
impose boundary conditions on the fields. In the case of fermions, the correct eigenvalue
problem requires mixed boundary conditions (see Appendix A).
The paper is organised as follows. In Section 2.1 we relate the phase shift to the heat
kernel and zeta function. In Section 2.2 we calculate the thermal effective action leading to
an expression for the prefactor in terms of the phase shift. In Section 3 we discuss how to
consider corrections from thicker walls. In Section 4 results are presented and in Section 5
we draw conclusions. In Appendices AC we give details on the calculation of the fermion
phase shift, numerical zeta function regularisation and renormalisation, respectively.
2. Heat kernels and phase shifts

2.1. Nucleation and regularisation
We begin with the nucleation rates for the decay of a false vacuum at finite temperature
due to an instanton bubble. Any field that acquires a mass on the instanton background can
contribute to the prefactor. In three dimensions [16] (see also comments in [5,12]),
B
A=T
2

3/2
det [ 2 + m2 (bubble)] 1/2

.
det[ 2 + m2 ( )]
sym
(1)
fields
Three zero eigenvalues, arising from breaking the Poincare symmetry, each contribute
(B/2)1/2 to the total and these are omitted from the scalar determinant, as indicated
by the prime.
In the thin wall limit we assume that bubble has a bubble wall at some radius R, such
that takes the false vacuum value at radii r > R and the true vacuum value at radii
r < R, with a narrow transition region near r = R. For scalar bosons and fermions (and if
calculated, the vector bosons), the relevant mass terms vanish in the false vacuum and are
nonzero in the true vacuum, leading to a step function profile,

m, r < R,
m(r) =
(2)
0, r > R.
In the same limit, the scalar Higgs field masses differ little for large and small radii.
This suggests that the Higgs contribution to the prefactor is smaller than the fermion
176
contribution (and also any other fields). We will consider the accuracy of the thin-wall
approximation later.
The eigenvalues in the determinant can be found by using a partial wave analysis and
phase shifts [17]. We first discretise the eigenmodes by putting them in a sphere of large
radius . After separating the eigenmodes into radial functions and spherical harmonics,
the radial parts asymptotically approach trigonometric functions of kr + , where is a
constant phase depending on k and the angular momentum l. On the boundary,
kn n .
(3)
(0)
In the false vacuum, the potential is zero and we label the free eigenvalues kn
kn(0) n (0) .
(4)
On letting (continuum limit), the above Eqs. (3) and (4) imply a relationship
(0)
between the phase shift l (k) = (0) , the density of states gl (k) and gl (k), [13],
gl (k) = gl(0) (k) +
1 dl (k)
,
dk
(5)
for the radial modes.

The difference between the heat kernels for the instanton and the true vacuum will be
2
(0)2
K(t) =
(6)
ekn t ekn t .
n
Using the density of states factor gl (k) and the degeneracy factor l ,

K(t) =
dk ek
2t

l gl (k) gl(0) (k) .
(7)
Substituting (5) into the above equation and integrating by parts we obtain
2
K(t) =
dk ek t kt
l l (k),
(8)
where the degeneracy factor l = (2l + 1) in three dimensions.

The heat kernel can now be used to regularise the determinants appearing in the
prefactor A in the nucleation rate. We define the generalised function [18] by
1
(s) =
(s)

t s1 tr K(t) dt.
(9)
The analytic continuation of (s) then gives

()W
log A =
fields
(10)
177
where we take for scalar and spinor fields, respectively, and

1
1
W = (0) + (0) log 2
(11)
2
2
where is the renormalisation scale.
For numerical work, the analytic continuation can best be performed by subtracting
terms from the heat kernel. As t 0, the heat kernel in d + 1 dimensions has the
asymptotic expansion [19]

Bn t n .
K(t) t (d+1)/2
(12)
n=0
The leading terms, which cause the poles in the function, can be removed by replacing
the sum over phase shifts in Eq. (8) by

l l =
l l
B1 k d1
( d+1
2 )
B3/2 k d2
( d2 )
B2 k d3
( d1
2 )
(13)
where the B0 coefficient cancels because it is equal to the free heat kernel K (0) . An
infrared cutoff MIR must also be included, noting (see Appendix B) that the dependence
on MIR is illusory given that changing MIR does not affect the value of W . The B3/2
coefficient only occurs for fermions because squaring the Dirac equation leads to a delta
function in the potential (for a step profile). Heat kernel coefficients have been calculated
for distributional backgrounds [20] and we only quote the result below.
For the step potential, standard expressions for the heat kernel coefficients give [19,20]:
B1 =
B3/2 =
B2 =
m2 R d+1
2d (d + 1) ( d+1
2 )
m2 R d
2d ( d+1
2 )
m4 R d+1
2d+1(d + 1) ( d+1
2 )
(14)
(15)
(16)
For example, the phase shift [17] for a scalar boson field is
tan l =
kJl1/2 (kR)Jl+1/2 (k R) k Jl1/2 (k R)Jl+1/2 (kR)

,
kNl1/2 (kR)Jl+1/2 (k R) k Jl1/2 (k R)Nl+1/2 (kR)
(17)
where k = k 2 m2 and l = 0, 1, . . . in three dimensions. The phase shift for fermions is

a rather lengthy calculation that is left until Appendix A.
2.2. Thermal effective action
Using the techniques of the last section we are now ready to calculate what is essentially
the difference in the effective action for the true and false vacua W . We refer the reader
to [21] for a detailed discussion of heat kernel methods at finite temperature for scalar
and spinor fields. The thermal heat kernel K can be expressed as an infinite sum of zero
178
temperature heat kernels
K (t | , x; , x ) =
()n K(t | , x; + n, x )
(18)
n=
(where is for scalar and spinor fields, respectively) and for an ultrastatic spacetime, the
heat kernel can be factorised into temporal and spacial parts giving
( )2
1
e 4t K (3)(t | x, x ).
K(t | , x; , x ) =
4t
(19)
It is then possible to show using the above relations that,

n2 2
K (t) =
e 4t K (3) (t).
4t n=
(20)
W is related to the thermal heat kernel K (t) for scalar fields by

1
W =
2
dt
tr K (t).
t
(21)
Substituting (20) into the above equation, we obtain for scalars,
W =
2

dt 2 n2 1 2
k 2 t
4t
e
dk e
kt
(2l + 1)l (k).
t n=
4t
(22)
l=0
Then, using the fact that

12 4t n2 k 2 t
nk
e
dt =
,
k
(23)
we get
W =
2

dk

l=0
(2l + 1)l (k)

dk

(2l + 1)l (k)
enk .
l=0
(24)
n=1
The sum over n is standard, leading to the result
W =
2

dk
0

l=0
(2l + 1)l (k)

dk
0

l (k)
.
(2l + 1) k
e 1
(25)
l=0
The first term is the zero point energy, that contains the ultraviolet divergences of
the theory and hence l (k) (see Eqs. (20) and (13)). The second term is the temperature
dependent part. The above expression can be derived using the density of states method
[12]. Thus, upon regulating the nonthermal part, using zeta function regularization and
introducing a mass that we let tend to zero at the end of the calculation (see Appendix B),
179
we have
W
=
2
B2
dk
(2l + 1)l (k) 2 B1 k
2 )
(k 2 + MIR
l=0
B2
2
+ log MIR
,
2 4

l (k)
T
W =
,
dk
(2l + 1) k
e 1
(26)
(27)
l=0
= W N
where W
+ W T is the thermal effective action and we are working explicitly
2 and for the scalar boson, B
in three dimensions. (Note that W is independent of MIR
3/2
is zero.)
For fermions the spinor effective action W(1/2) is related to the heat kernel K(1/2)(t)
by (twice the scalar result for massive fermions and also a colour factor of three if the top
quark is considered)

W(1/2) =
dt
tr K(1/2)(t).
t
(28)
Thus,

W(1/2) = 4
0
2 2
dt
(1)n e 4t n
t n=
4t
1
dk ek t kt
2
2(2j + 1)f (k),
(29)
j =1/2
where f = + + (see Appendix A) and j = 1/2, 3/2, . . . in three dimensions (the factor
of four comes from the trace over spinor indices). Applying (23) gives
4
W(1/2) =

dk
2(2j + 1)f (k)
j =1/2

dk
0
2(2j + 1)f (k)
j =1/2
Then, summing over n we have

4
W(1/2) =

dk
0

j =1/2
2(2j + 1)f (k)

n=1
(1)n enk .
(30)
180

dk
2(2j + 1)
j =1/2
f (k)
.
ek + 1
(31)
Of course, we could have guessed this result from looking at (25), taking into account
the properties of spinors. It is fairly simple to derive the above equation using the density
of states in the same way as in [12], using the spinor phase shift. Thus,

4
N
dk
2(2j + 1)f (k) 2 B1 k
W(1/2) =
j =1/2
0

2B2
B2
2
B3/2
(32)
,
log MIR
2
2
(k + M )
IR
T
=
W(1/2)

dk
0

j =1/2
2(2j + 1)
f (k)
.
ek + 1
(33)
Note the opposite sign in the zero point energy (nonthermal) contribution to the one loop
effective action as compared to the scalar case Eq. (26).
3. Thicker walls
The phase shift method works equally well for any bubble profile, although a differential
equation must be solved numerically to find the phase shift. In the general case, one could
consider the difference between the phase shift and the thin wall phase shift and then add
this correction onto the effective action numerically. Alternatively, it is possible to estimate
the corrections due to the nonzero thickness of the wall by perturbing about the thin wall
case. For example, consider the scalar boson, for which we must solve

l(l + 1)
r 2 r 2 u + m2 u +
(34)
u k 2 u = V u,
r2
where
m is given by Eq. (2), V is the correction due to a thicker wall and we define R such
that 0 r 2 V (r) dr = 0. The Greens function is

1

G(r, r ) = kA1 u1 (r)u2 (r ), r < r ,
(35)
kA u2 (r)u1 (r ), r > r ,
and

u1 (r) =

r < R,
jl (k r),
Ajl (kr) Bnl (kr), r > R,
Cjl (kr ) + Dnl (kr ), r < R,

(36)
nl (kr),
r > R,
where k = k 2 M 2 and jl (z) = /2 z(1d)/2Jl+(d1)/2(z) in d + 1 dimensions.

u2 (r) =
181
One then imposes u jl as r 0 and u Ajl B nl as r . Then, the solution is

u = u1
G(r, r )V (r )u(r )r 2 dr .
(37)
Therefore, as r
u u1 + kA

u2
V (r )u21 (r )r 2 dr .
(38)
From this it is possible to show that the correction to the phase shift is
(1)
k
= 2
A + B2
V (r)u21 (r)r 2 dr.
(39)
Then, assuming that the Bessel functions change little as r varies over the bubble wall, they
can be Taylor expanded about R, giving
(1)
2kk
2
jl (k R)jl (k R)
A + B2
R
V (r)r 3 dr,
(40)
where the continuity of u1 at the bubble wall has been used. For the scalar case,

B = kR 2 kjl1 (kR)jl (k R) k jl (kR)jl1 (k R)
(41)
and

A = kR 2 knl1 (kR)jl (k R) k nl (kR)jl1 (k R) .
(42)
All the terms outside the integral in Eq. (40) are numerical factors. For reasons discussed
in the conclusion we only add the correction onto the thermal part of the effective action,
giving
W
(1)
= 3
R
R
V (r)r 3 dr,
(43)
where is given by
1
=

dz
0

(2l + 1)
2zz
jl (z )jl (z ).
ez/R 1 A2 + B 2
(44)
l=0
If one assumes a tanh-like potential m2 (r) for the thicker wall, then V = m2 (r) m2
m2 /2 at the bubble wall radius R. Therefore, in terms of the width w of the bubble
wall one obtains the approximate result W (1) m2 w/2. We have calculated
numerically and found that it is of order 1 for a range of values of /R.
182
4. Results
The thermal one loop effective action was calculated numerically for fermions. The
phase shift is substituted into Eqs. (32) and (33), where it is convenient to change variables
k z = kR for the step potential. Then the nonthermal part has only one parameter
= m2f R 2 , where mf is the mass of the fermion. The thermal part has parameters and
mf (given that /R = mf / ) as independent variables.

We work with Eqs. (32) and (33) using a numerical package. For each value of z the
function is summed over l (or j ), with l increasing up to a given L until the value of the
function at z converges. The thermal part of the integral converges due to the exponential
damping terms. When considering the nonthermal part of the function, we must check that
the integrand has the correct k dependence after making the subtraction of the divergent
quantities from the sum over the phase shift. This is a good check verifying that the heat
kernel coefficients are correct.
We integrate up to Z chosen to obtain the required accuracy (all results are accurate to
1%). For large values of (and small mf ), larger values of Z and L are needed to give
convergence. The arctan function (from the phase shift) has problems with branches for
large values of (whenever hits ), requiring numerical glueing of the phase shift.
The nonthermal part of the fermion effective action can be written mf F ().
Numerically, F () fits well to a power law dependence on (see Fig. 1), giving, in the
original variables,
N
W(1/2)
= 1.51 m3f R 2 + 0.32 m4f R 3 ,
(45)
where we have set the renormalisation scale = MIR = mf as explained in Appendix C.

The nonthermal part is plotted in Fig. 2. The full one loop effective action for fermions
plotted against is in Fig. 3 for various values of the parameter mf .
N , plotted
Fig. 1. Fit of the nonthermal part of the one loop contribution to the fermion effective action W(1/2)
against .
183
T
Fig. 2. The thermal part of the one loop contribution to the fermion effective action W(1/2)
, plotted against
for values of mf (from bottom to top); 0.5, 1.0 and 5.0.
Fig. 3. The full one loop contribution to the fermion effective action W(1/2) , plotted against for values of
mf (from bottom to top); 0.5, 1.0 and 5.0.
5. Conclusion
We have presented a simple method to compute the prefactor in the expression for the
bubble nucleation rate, applying this to infinitely thin walls. Analytic corrections due to
thicker walls were considered by perturbing about the thin wall case. It is easy to see that
these corrections do not effect the B1 heat kernel coefficient. Using the arguments from
Appendix C, where it was shown that we only need to consider the renormalisation of the
thin wall bubble, one can ignore the correction from B2 . Thus, only adding corrections to
the thermal part of the effective action should be a good approximation.
In the context of the electroweak phase transition, we would also like to examine the
vector bosons, using the method presented. This requires vector spherical harmonics and
the relevant boundary conditions to calculate the phase shift. Then a full treatment of all the
particle species at the phase transition can be worked out in the thin wall approximation,
including analytic expressions for corrections due to thicker walls.
184
The fermion contribution generally enhances the rate, but for large mf (when = mf )
it does not and becomes suppressive. In fact, at mf = 5.0 (see Fig. 3), the sign of the
effective action changes for various values of = mf R (for a fixed bubble wall radius).
Choosing = mt and mf = mt (the mass of the top quark) as in [11], then the log
term cancels for the fermion determinant (see Eq. (45)). Baacke and Srig [11] found a
negative contribution (enhance), whereas we find it can also be positive (suppress) for a
temperature ( 1 = 35 GeV with mt = 5.0 and mt 175 GeV) lower than the that at the
electroweak phase transition. However, there is no direct method of comparison between
the two methods because in [11] the thin wall limit was not considered.
The renormalisation scale can be set by imposing conditions on the effective potential,
calculated for constant background fields [12]. In Appendix C it was shown that these
conditions on the effective potential (that include the classical and regularised terms) lead
to the choice = MIR = mf .
In the case of bubble nucleation, the bubble wall radius is found by extremising an
action which includes the effective potential. The three-dimensional action for the bounce
solution is
B = 4R 2
4
R 3 ,
3
(46)
where and 7 are corrected surface and volume energy densities, respectively. The
nonthermal part of the effective action (Eq. (45)) has a similar form and can be absorbed
into a redefinition of and 7. The value of R should be presumably chosen to coincide
with the extrema of the new action. Then, the nucleation rate is determined as a function
of temperature by W T .
The method described can also be used to calculate the one loop effective action in
general, but in this case the B0 heat kernel coefficient must be renormalised away in the
vacuum energy. The numerical procedure is simple and efficient.
Acknowledgements
W.N. would like to thank JSPS for Postdoctoral Fellowship for Foreign Researchers No.
P01773, where this work was revised and completed.
Appendix A. Fermion phase shift

Here we present the calculation of the spinor phase shift. One must use the Dirac
equation separated into radial and angular coordinates and also be careful in the way one
works out the eigenvalues of the problem. Starting with
(i m)+ = ,
(A.1)
(i + m) = + ,
(A.2)
185
so that upon squaring the Dirac equation we get the KleinGordon equation. The radial
components are then [17]

3 f
(E m)g + f + j +
= g ,
2 r

1 g

= f ,
(E m)f g
+ j
2 r
(A.3)
(A.4)
where

= =
ig Y
f r Y

,
r =
0
i
i
0

,
(A.5)
for j = l + 1/2. The solutions are spherical Bessel functions f = A jj +1/2 and g =
C jj 1/2 (this can be seen easily by substituting (A.3) into (A.4)) where the constants are
as yet undetermined. On substituting the power series for the Bessel functions into (A.3)
and (A.4) one finds the relations
(E m)C + A k C = 0,
(A.6)
(E m)A + C k A = 0.
(A.7)
The correct eigenvalue problem requires mixed boundary conditions [19] (also for a
self-adjoint action) leading to f+ +g+ = f g = 0 as r . It is also possible to show
that there is a symmetry among the solutions such that f+ f and g+ g . For the
calculation we set E = 0 and impose the boundary conditions with the above symmetries,
noting that m 0 as r . Using this information, it is possible to show that there are
two solutions
tan =
B+
,
A+
(A.8)
for A+ = C+ and B+ = D+ , where B and C are the corresponding constants for

nj 1 , the irregular solutions, that are spherical Neumann functions.
2
Then, matching the wavefunction at the junction and performing some algebra, we get
the phase shifts,
tan =
(k m)Jj (kR)Jj +1 (k R) k Jj (k R)Jj +1 (kR)

,
(k m)Nj (kR)Jj +1 (k R) k Jj (k R)Nj +1 (kR)
(A.9)
where all symbols have their usual meanings as in the rest of the paper.
The j = l 1/2 modes have symmetry such that f g changes the boundary
conditions and equations into the j = l + 1/2 case. Thus we have the above phase shifts
with a total degeneracy 2(2j + 1), where j = 1/2, 3/2, . . . . All results can be generalised
to the four-dimensional case.
186
Appendix B. Numerical zeta function regularisation

Consider the finite temperature case. The nonthermal part of the heat kernel is
2
K0 (t) =
(4t)
dk ke(k
2 +M 2 )t
(k)t,
(B.1)
where all symbols have the same conventions as in the rest of the paper and we have
introduced a mass term M. The zeta function is then
2 (s + 12 )
0 (s) =
(4) (s)
s 1

2
dk k k 2 + M 2
(k).
(B.2)
The integral converges for s > 1. In order to apply the zeta function to numerical work, we
subtract terms from the asymptotic expansion of the heat kernel and define

s 1

2 (s + 12 )
2
P (s) =
dk k k 2 + M 2
(4) (s)
0

B2
.
2 B1 k B3/2
k
(B.3)
Performing the integration over k, and adding back the subtracted terms, one finds
B1
0 (s) = P (s) + M 22s

s1
2
(s 12 ) 12s
+
M
B3/2 + M 2s B2 .
2 (s)
2
(B.4)
This is a regular function at s = 0. Taking the derivative with respect to s (for s = 0) and
the limit M 0, we obtain

B2
B2

2
0 (0) =
log MIR
dk
2 B1 k B3/2
,
2
4
(k 2 + M )
IR
(B.5)
where MIR is an infrared cutoff. It is important to note that if one subtracts the difference
between 0 (0) for different values of infrared cutoff (MA and MB say) and using the fact
that

1
1
MB
(B.6)
dk

,
= log
MA
k2 + M 2
k2 + M 2
0
it is clear that this difference is zero. Thus, Eq. (B.5) is independent of MIR .
187
Appendix C. Renormalisation
The renormalisation scale represents an arbitrary parameter that can be expressed
in terms of other constants by imposing conditions on the effective potential (see [22]
for example). We shall show that the usual renormalisation conditions for the effective
potential,
V (vac ) = 0,
V (vac ) = mh ,
(C.1)
where vac is the classical vacuum expectation value and mh is the mass of the Higgs field,
give the optimal choice, removing the logarithmic term.
The first two terms in the zero point energy (45) come from the finite, subtracted, part
of the effective action (see Eq. (B.3)). A similar approach was used in [11], where the
divergent effective action was regularised by subtracting terms proportional to 1st and
2nd powers of the background potential, leaving a finite contribution to the fluctuation
determinant. The two divergent terms were then regulated by using a momentum cutoff
and renormalised by applying conditions on the propagators. This procedure is equivalent
to the subtraction of the heat kernel coefficients up to and including B2 (see Eq. (B.3)).
However, zeta function regularisation is unusual, as compared with most regularisation
schemes, in that it does not need infinite counter terms. Thus, in Eq. (B.5) only the last
term, containing the logarithm, requires renormalisation.
The effective potential V = V0 + V1 will be defined

1 2
1
3
2 2
3
4
V = d x
m + mh (g + g) + ( + )
2 h
4
M2
2B2
log IR
,
2
(C.2)
where the terms in the square brackets make up the classical potential (including
counterterms), the last term is the -regularised 1-loop contribution with

1
B2 = 0 (0) =
(C.3)
x ),
d 3 x m4f 4 (
3/2
(4)
2
and we have included the renormalisation scale (see Eq. (11)).
In our case the renormalisation must be applied to a step function field configuration,
but the ultraviolet divergences are independent of the background profile. This is intuitively
obvious because k equates to the short distance behaviour of the theory. Thus, the
x )) need only be considered for (
x ) = constant, i.e., the
B2 heat kernel coefficient ( 4 (
effective potential.
By applying the renormalisation conditions (C.1) to the effective potential, it is simple
to show that the counter terms are given by m2h = g = 0 and
m4f
M2
log IR
.
=
2
2 2
(C.4)
On substituting the above counterterms into the effective potential it is clear that
the logarithmic term will cancel. Naturally, for our numerical procedure we can choose
188
MIR = mf because it was shown in Appendix B that the infrared cutoff was independent
of the result (Eq. (B.5)).
With our choice of renormalisation we can compare with a similar work [12], where
the scalar Higgs field was investigated. In the case of the thin wall limit they only
considered the surface free energy. However, comparing with our zero point surface energy
contribution, for a scalar boson [23], we find it is of a similar order of magnitude and also
suppresses the nucleation rate. This contribution is considerably smaller than that made by
a fermion with a Yukawa coupling which, generally, enhances the nucleation rate.
References
[1] S. Coleman, Phys. Rev. D 15 (1977) 2929;
C.G. Callan, S. Coleman, Phys. Rev. D 16 (1977) 1762.
[2] V. Zarikas, Phys. Lett. B 384 (1996) 180.
[3] G.W. Anderson, L.J. Hall, Phys. Rev. D 45 (1992) 2685.
[4] W. Naylor, in progress.
[5] M. Gleiser, G.C. Marques, R.O. Ramos, Phys. Rev. D 48 (1995) 1571.
[6] J. Garriga, Phys. Rev. D 49 (1994) 5497.
[7] J. Kripfganz, A. Laser, M.G. Schmidt, Nucl. Phys. B 433 (1995) 467.
[8] S. Coleman, Aspects of Symmetry, Cambridge Univ. Press, Cambridge, UK, 1985.
[9] J. Baacke, V.G. Kiselev, Phys. Rev. D 48 (1993) 5648.
[10] J. Baacke, Phys. Rev. D 52 (1995) 6760.
[11] J. Baacke, A. Srig, Phys. Rev. D 53 (1996) 4499.
[12] C.L.Y. Lee, Phys. Rev. D 49 (1994) 4101;
D.E. Brahm, C.L.Y. Lee, Phys. Rev. D 49 (1994) 4094.
[13] J. Schwinger, Phys. Rev. 94 (1954) 1362.
[14] G. Mnster, S. Rotsch, Eur. Phys. J. C 12 (2000) 161.
[15] G. Mnster, A. Strumia, N. Tetradis, Phys. Lett. A 271 (2000) 80.
[16] A.D. Linde, Phys. Lett. B 70 (1977) 306;
A.D. Linde, Phys. Lett. B 100 (1981) 37.
[17] L.I. Schiff, Quantum Mechanics, McGraw-Hill, Singapore, 1968.
[18] J.S. Dowker, J. Critchley, Phys. Rev. D 13 (1976) 3224.
[19] I.G. Moss, Quantum Theory, Black Holes and Inflation, Wiley, 1996.
[20] M. Bordag, D.V. Vassilevich, J. Phys. A 32 (1999) 8247;
I.G. Moss, Phys. Lett. B 491 (2000) 203.
[21] J.S. Dowker, J.P. Schofield, Nucl. Phys. B 327 (1989) 267;
Y.V. Gusev, A.I. Zelnikov, Phys. Rev. D 59 (1999) 024002.
[22] A. Berkin, Phys. Rev. D 46 (1992) 1551.
[23] W. Naylor, Ph.D. Thesis, University of Newcastle Upon Tyne, 2001.

Electroweak two-loop corrections to the MWMZ

mass correlation in the Standard Model
A. Freitas a , W. Hollik b,c , W. Walter b , G. Weiglein d
a DESY Theorie, Notkestr. 85, D-22603 Hamburg, Germany
b Institut fr Theoretische Physik, Universitt Karlsruhe, D-76128 Karlsruhe, Germany
c Max-Planck-Institut fr Physik, Fhringer Ring 6, D-80805 Mnchen, Germany
d IPPP, University of Durham, Durham DH1 3LE, United Kingdom
Abstract
Recently exact results for the complete fermionic two-loop contributions to the prediction for the
W-boson mass from muon decay in the electroweak Standard Model have been published [Phys. Lett.
B 495 (2000) 338; Nucl. Phys. (Proc. Suppl.) 89 (2000) 82]. This paper illustrates the techniques
that have been applied for this calculation, in particular, the renormalisation procedure and the
treatment of IR-divergent QED contributions. Numerical results are presented in terms of simple
parametrisation formulae and compared in detail with a previous result of an expansion up to
next-to-leading order in the top-quark mass. An estimate of the remaining theoretical uncertainties
of the MW -prediction from unknown higher-order corrections is given. For the bosonic two-loop
corrections a partial result is presented, yielding the Higgs-mass dependence of these contributions.
2002 Elsevier Science B.V. All rights reserved.
1. Introduction
One of the most important quantities for testing the Standard Model (SM) or its
extensions is the relation between the massive gauge boson masses, MW and MZ , in
terms of the Fermi constant, G , and the fine structure constant, . This relation can be
derived from muon decay, where the Fermi constant enters the muon lifetime, , via the
E-mail addresses: ayres.freitas@desy.de (A. Freitas), georg.weiglein@durham.ac.uk (G. Weiglein).

PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 2 4 3 - 2
190
A. Freitas et al. / Nuclear Physics B 632 (2002) 189218
expression
1
G2 m5
m2e
=
F
192 3
m2

3 m2
1+
(1 + q),
2
5 MW
(1)
with F (x) = 1 8x 12x 2 ln x + 8x 3 x 4 . By convention, this defining equation is

supplemented with the QED corrections within the Fermi Model, q. Results for q
have been available for a long time at the one-loop [2] and, more recently, at the twoloop level [3]. Commonly, tree-level W propagator effects giving rise to the (numerically
2 ) in Eq. (1) are also included in the definition of G ,
insignificant) term 3m2 /(5MW
although they are not part of the Fermi Model prediction.

Comparing the prediction for the muon lifetime within the SM with Eq. (1) yields the
relation

M2
2
(1 + r),
MW
(2)
1 W
=
2
MZ
2 G
where the radiative corrections are summarised in the quantity r [4]. This relation allows
a prediction of MW , to be tested against the experimental result for MW . The current
exp
accuracy of the measurement of the W-boson mass, MW = 80.451 0.033 GeV [5], will
be further improved in the final LEP analysis and Tevatron Run II [6], each with an error
of MW 30 MeV. At the LHC, an error of MW 15 MeV can be expected [7], while
a high-luminosity linear collider running in a low-energy mode at the W+ W threshold
could reach a reduction of the experimental error down to MW 6 MeV [8]. This offers
the prospect for highly sensitive tests of the electroweak theory [9], provided that the
accuracy of the theoretical prediction matches the experimental precision.
The quantum correction r has been under extensive theoretical study over the last
two decades. The one-loop result [4] involves large fermionic contributions from the shift
in the fine structure constant due to light fermions, log mf , and from the leading
contribution to the parameter, , which is quadratically dependent on the top-quark
mass mt , resulting from the topbottom mass splitting [10],
r () =
2
cW
2
sW
+ rrem (MH ),
(3)
2 = 1 M 2 /M 2 . The remainder part r

with sW
rem contains in particular the dependence
W
Z
on the Higgs-boson mass, MH .
Beyond the one-loop order, resummations of the leading one-loop contributions and
have been derived [11]. They correctly take into account the terms of the form ()2 ,
(), and (rrem ) at the two-loop level and ()n to all orders.
Beyond the two-loop order, complete results for the pure fermion-loop corrections
(i.e., contributions containing n fermion loops at n-loop order) are known up to four-loop
order [12]. These results also include the contributions arising from resummation of
and . Recently, the leading three-loop contributions to the parameter of O(G3 m6t )
and O(G2 s m4t ) have been computed in the limit of vanishing Higgs boson mass [13], but
were found to have small impact on the prediction of the W mass.
191
Fig. 1. Examples for types of fermionic two-loop diagrams contributing to muon decay.
Higher order QCD corrections to r have been calculated at O(s ) [14] and for the
topbottom contributions at O(s2 ) [15]. The O(s2 ) contributions with light quarks
in the loops can be derived from the formulae (29)(31) in [16] and turn out to be
completely negligible. First results for the electroweak two-loop contributions have been
obtained using asymptotic expansions for large Higgs [17] and top-quark masses [1820].
Concerning the expansion in mt , the formally leading term of O(G2 m4t ) [18,19] and the
next-to-leading term of O(G2 m2t MZ2 ) [20] were found to be numerically significant for the
prediction of the W mass. Since both contributions turned out to be of similar magnitude
and of same sign, a more complete calculation of electroweak two-loop corrections to r
without using expansions is desirable.
As a first step in this direction, exact results have been obtained for the Higgs-mass
dependence (e.g., the quantity MW,subtr(MH ) MW (MH ) MW (MH = 65 GeV)) of the
fermionic two-loop corrections to the precision observables [21]. They were shown to
agree well with the previous results of the top-quark mass expansion [22].
For the bosonic two-loop corrections to r, the complete result is not available up to
now. However, in Ref. [23] the effect of the bosonic terms up to O( 2 ) on the relation
between the MS and on-shell definition of the gauge boson masses has been studied. For
this purpose the corresponding two-loop self-energies have been evaluated in the MSscheme using large-mass expansions.
This paper discusses the exact computation of all fermionic two-loop corrections to r
which has been presented recently [1]. These include all two-loop diagrams contributing
to the muon decay amplitude and containing at least one closed fermion loop (except the
pure QED corrections already contained in the Fermi Model result, see Eq. (1)). Some
typical examples are shown in Fig. 1. No expansion in the top-quark mass or the Higgs
boson mass is made, so that the full dependence on mt and MH as well as the complete
light-fermion contributions at two-loop order are contained. Previously, corrections from
light fermions have only been taken into account via resummations of the one-loop lightfermion contribution (the two-loop light-fermion contributions have been calculated within
the MS-scheme in Ref. [24]).
The result of [1] has been included in the Standard Model fits and the indirect derivation
of constraints on the Higgs boson mass performed by the LEP Electroweak Working
Group [5].
As a further step towards a complete two-loop result for r, a partial result is presented
for the purely bosonic electroweak two-loop corrections which yields the Higgs-mass
dependence of these terms.
192
The paper is organised as follows. Sections 24 enlarge on the methods which were
employed for the calculation of the fermionic two-loop corrections. While Section 2
presents an overview over the techniques, the renormalisation procedure is explained in
Section 3 and the extraction of the QED corrections, which are already contained in the
Fermi Model, is described in Section 4. A discussion of the numerical results and remaining
theoretical uncertainties due to unknown higher orders can be found in Sections 5 and 6,
respectively. In Section 7 the Higgs-mass dependence of the bosonic two-loop corrections
is studied. Before concluding, an outlook to the situation for future colliders is given in
Section 8.
2. Outline of the two-loop calculation

This section presents an overview over the main features of the calculation.
Since the definition of the Fermi coupling constant according to Eq. (1) contains
QED corrections in the Fermi Model summarised in the quantity q, the corresponding
contributions have to be identified and extracted from the Standard Model computation of
muon decay in order to arrive at the quantity r. As shown in Section 4, all IR-divergent
loop contributions in our calculation are already contained in q, contributing to r are
IR-finite. As a consequence, after extraction of the Fermi Model contributions, it is possible
to neglect the masses and momenta of the external particles, thereby reducing the generic
diagrams contributing to the muon-decay amplitude to vacuum diagrams.
For the renormalisation the on-shell scheme is used throughout. It entails, in addition,
the evaluation of two-loop two-point functions with non-zero external momentum, which
is technically more involved. However, it should be noted that the evaluation of this type of
integrals is generally necessary in all renormalisation schemes if the result shall be related
to the physical gauge boson masses. The details of the renormalisation procedure are given
in Section 3.
Since the calculation involves the computation of more than thousand diagrams,
it is convenient to employ computer-algebra tools. The generation of diagrams and
Feynman amplitudes, including the counterterm contributions, was performed with the
package F EYNA RTS [25]. The program T WO C ALC [26] was applied for the algebraic
evaluation of these amplitudes, which were reduced, by means of two-loop tensor-integral
decompositions, to a set of standard scalar integrals. Throughout the calculation, a general
R gauge was used, and the gauge-parameter independence of the final result was checked
algebraically. For the evaluation of the scalar one-loop integrals and the two-loop vacuum
integrals we have used analytical results as given in Refs. [27,28], while the two-loop twopoint integrals with non-vanishing external momentum have been evaluated numerically
using one-dimensional integral representations with elementary functions [29]. These
allow a fast and stable calculation of the integrals for general mass configurations.
Since we use Dimensional Regularisation [30,31] in our calculation, it is necessary
to investigate the treatment of the Dirac algebra involving 5 . It is known that a naively
anticommuting 5 respects all Ward identities of the Standard Model [32]. However, while
it can safely be applied for all two-loop two-point contributions (for a discussion, see,
e.g., Ref. [18]) and most of the two-loop vertex- and box-type diagrams, it would yield
193
Fig. 2. Two-loop vertex diagrams containing a triangle subgraph, which require a careful treatment of 5 in D
dimensions.
an incorrect result for vertex diagrams containing a triangle subgraph (see Fig. 2). This
originates from an inconsistent treatment of the trace of 5 together with four Dirac
matrices, which in four dimensions is given by Tr{5 } = 4i , while this
trace would vanish when using the naively anticommuting 5 in D dimensions.
A mathematically consistent definition of 5 in D dimensions [31,33] would require
the introduction of additional counterterms to restore the Ward identities, which is a
very tedious procedure at the two-loop level. For recent discussions on this topic, see
Refs. [34,35].
In order to calculate the class of diagrams in Fig. 2, we have first evaluated the triangle
subgraph with a consistent 5 according to Refs. [31,33] (here we made use of the package
T RACER [36] for checking). After adding appropriate counterterms to restore the Ward
identities, the result differs from the result obtained using a naively anticommuting 5
only in terms proportional to the totally antisymmetric tensor , which are finite for
D 4. Inserting this difference term into the two-loop diagrams, it turns out that the
second loop only yields a finite contribution, so that it can be evaluated in four dimensions
without further complications. After contraction with the external fermion line in the vertex
diagrams, a non-zero contribution to the result for r is obtained from this term.
3. On-shell renormalisation
3.1. One-loop renormalisation
In the on-shell renormalisation scheme the mass parameters and coupling constants are
related to physical observables. The gauge bosons of the U(1) and SU(2)L group, B ,
W1,2,3 are conveniently expressed in terms of their mass eigenstates,

1
W = W1 iW2 ,
2

3
Z
cW sW
W
=
,
A
sW cW
B
(4)
(5)
where W , Z denote the fields of the massive vector bosons W, Z with masses MW ,
MZ , A represents the massless photon field, and the weak mixing angle enters in the
combination

2 /M 2 .
cW = cos W = MW /MZ ,
(6)
sW = sin W = 1 MW
Z
194
The eigenstates of the Higgs doublet are the physical Higgs field H with mass MH and
the neutral and charged Goldstone bosons , . We neglect mixing between the fermion
generations throughout.
In the following the conventions of [37] are adopted. In our approach all physical fields
and masses as well as the electromagnetic coupling e are renormalised:
e0 = Ze e,
2
2
2
= MW
+ MW
,
MW0
2
MZ0
= MZ2 + MZ2 ,
mf 0 = mf + mf ,

1/2
W0 = Z W
W ,
1/2

1
Z
A0 = 2 Z Z + Z
A,

1/2
f0L = Z f L
f L (f = e, e , . . .),
2
MH0
= MH2 + MH2 ,

1/2
Z0 = Z ZZ
Z + 12 Z Z A,
H 1/2
H0 = Z
H,

1/2
f0R = Z f R
f R.
(7)
Here the index 0 indicates the bare quantities. The renormalisation constants for the
masses, MX , mf , the fields, Z X = 1 + Z X , and the charge, Ze = 1 + Ze , are fixed
by on-shell renormalisation conditions.
The physical squared masses MX2 are defined as the real part of the poles M2X of
the propagators D X . The field renormalisation constants are determined by demanding
unity residues of the poles. This ensures that all Greens functions are finite and external
wave function corrections need not be taken into account. With the index T denoting the
transverse part of the vector boson propagators, the on-shell renormalisation conditions
read

W 1 2
W 1 2
= 1,
MW = 0,
Re i 2 DT
k
DT
k
k 2 =M2W

ZZ 1 2
ZZ 1 2
= 1,
DT
MZ = 0,
Re i 2 DT
k
k
k 2 =M2Z

Z 1
Z 1 2
1 2
MZ = 0,
DT
k
DT
(0) = 0,
Re i 2 DT
= 1,
k
k 2 =0

H 1 2
H 1 2
= 1,
D
MH = 0,
Re i 2 D
k
k
k 2 =M2H

f 1
f 1
D
D
(p)p2 =M2 = 0,
Re i
(p)
= 1.
f
/
p
p 2 =M2
f
(8)
The propagators are related to the one-particle-irreducible two-point functions by
1
1

D H = H ,
DTW = TW ,

Z
Z 1
DTZZ DT
TZZ T
=
,
Z
D
D

T
(9)
1

f
f
f 1
D f = f
= p
/ L + p
/ + R + mf S
.
(10)
195
Here L,R,S represent the left-/right-handed and scalar component of the fermion two-point
functions, respectively. The two-point functions can be separated into a Born contribution
and the self-energies,

Tab k 2 ,
Tab (k 2 ) = i k 2 Ma2 ab +
(11)
2

2
2

2
f
f
f
f

p .
,
S p = i 1 +
L,R p = i 1 +
(12)
L,R p
S
f
= + counterterms.
The hat indicates renormalised quantities, i.e.,
In addition to the aforementioned renormalisation conditions, one has the freedom to
renormalise the Higgs tadpole t. Here the condition
t + t = 0
(13)
is applied, which requires that all tadpole contributions are exactly cancelled by the
counterterms t, so that no tadpoles need to be taken into account in the actual calculation.
Using the renormalisation conditions Eq. (8) one obtains for the renormalisation
constants in Eq. (7) at the one-loop level
2

2
MW
,
MZ2 = Re TZZ MZ2 ,
MH2 = Re H MH2 ,
= Re TW MW

2
,
Z ZZ = Re TZZ MZ2 ,
Z W = Re TW MW
Z 2
Z
T (MZ )
T (0)
Z
Z
Z = 2 Re
=
2
,
,
Z
MZ2
MZ2

Z = T (0),
(14)
Z H = Re H MH2 ,
f

mf
f
f
Re L m2f + R m2f + 2R m2f ,
2

f
f
f
f
fL
Z = Re L m2f m2f Re L m2f + R m2f + 2S m2f ,

f
f
f
f
Z f R = Re R m2f m2f Re L m2f + R m2f + 2S m2f ,
mf =
(15)
with (k 2 ) indicating the derivative of the self-energy with respect to k 2 .

For the charge renormalisation an additional condition is required. Usually it is fixed
by demanding that the electric charge e coincides with the coupling of the electromagnetic
vertex in the Thomson limit,

u(p)
ee (p, p)u(p)p2 =m2 = ieu(p)
(16)
u(p).
e
Employing the U(1) Ward identity this yields at one-loop order

1
sW
Ze = Z
(17)
Z Z .
2
2cW
The weak mixing angle is a derived quantity, expressed in terms of the gauge boson masses,
see Eq. (6). Thus the renormalisation of MW , MZ also determines the counterterm sW for
sW [4]. At one-loop order one obtains
2
2
cW
MZ2 MW
sW
= 2
(18)
.
2
sW
2sW MZ2
MW
196
From two-loop order on, a one-loop sub-renormalisation is necessary for the Faddeev
Popov ghost sector, which is associated with the gauge-fixing part. It is possible to keep the
gauge-fixing part invariant under renormalisation. For technical convenience, we arrange
for this by a renormalisation of the gauge parameters in such a way that it precisely cancels
the renormalisation of the parameters and fields in the gauge-fixing Lagrangian.1 To this
end, we start with the following, rather general form of the bare gauge-fixing term:

1 2 Z 2
F
+ F
+ F +F + F F + ,
Lgf =
(19)
2
1/2
Z
Z ,
F = 1
(20)
A +
2
1/2
1/2
Z
F Z = 1Z
(21)
Z +
MZ ,
A 2Z
2
1/2
1/2
F = 1W
(22)
W i 2W
MW ,
allowing two different bare gauge parameters for both W and Z, 1W,Z and 2W,Z , and also
mixing gauge parameters, Z and Z . The renormalised parameters shall comply with
the R gauge, providing one free gauge parameter for each gauge boson, , Z , W . With
the following renormalisation prescription

1/2
1/2
1/2
1 Z
1
0
Z Z
1
Z
2
2
Z 1/2
ZZ 1/2 ,
Z 1/2 =
1 Z
1
Z
1
Z
0
2
2 Z
(23)
2 + M 2
M
Z = 2Z Z 2 Z Z ,
(24)
MZ
W =
1 W
,
ZW 1
W = 2W
2 + M 2
MW
W
2
MW
(25)
no counterterm contributions arise from the gauge-fixing sector. Here we have allowed for
field renormalisation constants Z , Z of the unphysical scalars , as well. Starting
at the two-loop level, counterterm contributions from the ghost sector have to be taken into
account in the calculation of physical amplitudes. They follow from the variation of the
gauge-fixing terms F a under infinitesimal gauge transformations, b , b = , Z, ,

F a
F a (x) b
LFP =
(26)
u (y).
u a b ub = d 4 y
u a (x) b
(y)
a,b= ,Z,
a,b
These contributions can be derived from the action of the gauge transformations on the
gauge and Goldstone fields as follows,

A A + + ie W+ W + ,
1 In another approach to avoid counterterms emerging from the gauge-fixing sector, the gauge-fixing part
could alternatively be added to the Lagrangian only after renormalisation. In this case the counterterms to the
ghost sector arise only from the renormalisation of the gauge transformations and not from the gauge-fixing
functions F V .
197

cW +
W W + ,
sW

cW Z
cW
W A +
Z ,
W W + ie W
sW
sW

e
e
MZ +
+ + + ,
H Z +
2sW cW
2sW

2 c2
sW
ie
e
W Z

ie ie
iMW +
H
.
2cW sW
2sW
2sW
Z Z + Z ie
(27)
We have derived all the counterterms arising from the ghost sector (extending the results of
Refs. [38,39] to a general R gauge) and implemented them into the program F EYNA RTS.
In this way we could verify the finiteness of individual (gauge-parameter-dependent)
building blocks (e.g., the W- and the Z-boson self-energy) as a further check of the
calculation. The explicit Feynman rules of the ghost sector including counterterms can
be found in the appendix.
3.2. Two-loop counterterms
In the O( 2 ) calculation of the muon decay, two-loop counterterms arise for the
transverse W propagator and the charged current vertex:
2

W
2
2
W
2
k MW
MW,(2)
= Z(2)
Z(1)
MW,(1)
,
(28)

e
sW 1 eL
W
L
=i
Ze(2)
+ Z(2) + Z(2)
+ Z(2)
sW
2
2sW

(29)
+ (1-loop renormalisation constants) .
The numbers in parentheses indicate the loop order. Throughout this paper, the two-loop
contributions always include the subloop renormalisation.
Concerning the mass renormalisation of unstable particles, from two-loop order on it
makes a difference whether the mass is defined according to the real part of the complex
pole of the S matrix,
2 i M
,
M2 = M
(30)
or according to the pole of the real part of the propagator. In Eq. (30) M denotes the

complex pole of the S matrix as specified by the renormalisation conditions in Eq. (8). M,

are then interpreted as the corresponding mass and width of the unstable particle. For
It is determined by
the real pole, on the other hand, we use the symbol M.
2

= 0.
Re (DT )1 M
(31)
In the context of the present calculation, these considerations are relevant for the
renormalisation of the gauge-boson masses, MW and MZ . The two-loop mass counterterms
198
according to the definition of the mass as the real part of the complex pole are obtained
from Eq. (8),
W 2
2
W
2
M
W,(2) = Re T,(2) MW MW,(1) Z(1)
W 2 W 2
MW Im T,(1) MW ,
+ Im T,(1)
(32)

M2
2 = Re ZZ MZ2 M 2 Z ZZ + Z Z Z 2
M
Z,(2)
T,(2)
Z,(1)
(1)
(1)
4
Z
(Im{T,(1)(MZ2 )})2
MZ2
ZZ 2 ZZ 2
MZ Im T,(1) MZ .
+ Im T,(1)
(33)
2
When compared with the mass counterterms according to the real-pole definition, M
W,(2)
2 , there remains a finite difference,
and M
Z,(2)
W 2 W 2
2
2
M
W,(2) = MW,(2) + Im T,(1) MW Im T,(1) MW ,
ZZ 2 ZZ 2
2
2
Z,(2)
Z,(2)
M
MZ Im T,(1) MZ .
= M
+ Im T,(1)
(34)
(35)
It can easily be checked by direct computation that the difference terms in Eqs. (34), (35)
are gauge-parameter-dependent, thus showing that at least one of the two prescriptions
leads to a gauge-dependent mass definition. The problem of a proper definition of unstable
particles in gauge theories has already been addressed several times in the literature [40].
However, the present work, for the first time, involves an explicit calculation of a
physical process which is sensitive to the gauge-parameter-dependent difference between
the two mass renormalisation methods. In the previous results for MW , incorporating
terms up to O(G2 m2t MZ2 ) [20] and MH -dependent fermionic terms [21], the contribution

(M 2 )} Im{T,(1)(M 2 )} was zero, making thus a strict distinction between the
Im{T,(1)
two mass definitions unnecessary at the considered order.
Using a general R gauge, we can test the two mass renormalisation prescriptions in
our result by regarding the two-loop counterterms to physical observables, which should
be gauge-parameter-independent. In particular, we only find an invariant result for the
counterterm to the weak mixing angle, sW,(2) , with the definition of the gauge-boson
masses according to the complex pole.
2
In order to verify the gauge-parameter independence of the mass counterterms MW,(2)
2
and MZ,(2)
one needs an appropriate treatment of the Higgs tadpole diagrams, which
do not contribute to physical observables. Alternatively to Eq. (13), one can include all
Higgs tadpole diagrams in the calculation by demanding the tadpole counterterm to be
zero, t = 0. Technically, this corresponds to the inclusion of all tadpole diagrams not only
in the two-loop self-energies, but also in the subloop renormalisation. In this case, also
the mass counterterms themselves are gauge-parameter-independent when using the mass
definition via the complex pole.
These results confirm the expectations from S-matrix theory that the complex pole is a
gauge-invariant quantity [40,41].
199
We have thus adopted the complex-pole definition as given in Eqs. (32) and (33). Using
this mass definition and expanding the gauge boson propagator around its pole
DT
1

q 2 = (DT )1 M2 + q 2 M2
(DT )1 k 2 k 2 =M2
2
k

2
2 2
+O q M
(36)
one obtains with the renormalisation conditions Eq. (8)

DT q 2 =
q2
i const
+ non-resonant terms,
2 + iM

M
(37)
which corresponds to a BreitWigner parametrisation of the resonance line shape with a

constant decay width.
Experimentally the gauge-boson masses are determined using a BreitWigner function
with a running (energy-dependent) width,

DT q 2
q2
M2
i
.
+ iq 2 /M
(38)
As a consequence of the different BreitWigner parametrisations, there is a numerical

difference between the experimental mass parameters (denoted as MW , MZ henceforth)
W , M
Z . The shift between these parameters
and the mass parameters in our calculation, M
2

is given by [42] MW,Z = MW,Z + W,Z /(2MW,Z ). Since MW and MZ enter on a different
footing in our computationMZ is an experimental input parameter, while MW is
calculatedin order to evaluate the mass shifts we use the experimental value for the Zboson width, Z = 2.944 0.0024 GeV [5],
and the theoretical value for the W-boson
3
width, which is given by W = 3G MW
/(2 2 )(1 + 2s /(3)) in sufficiently good
Z + 34.1 MeV and in the mass shifts MW
approximation. This results in MZ M

MW + 27.4 MeV and MW MW + 27.0 MeV for MW = 80.4 GeV and MW = 80.2 GeV,
respectively.2
For an extension of the renormalisation formalism for unstable particles to higher loop
orders and to the field renormalisation of unstable particles, see Refs. [43]. However, in a
physical process with particles in the initial and final state whose mass can be neglected,
a treatment of complex poles is only necessary for internal particles. Since the field
renormalisation of internal particles does not contribute to the physical result, it is not
necessary to examine this issue for our purposes.
The two-loop charge renormalisation constant follows from the condition Eq. (16). With
the help of the U(1) Ward identity the electromagnetic vertex can be related to photonic
two-point functions, thus resulting in the following relation between the renormalisation
2 The difference in according to the way it is calculated, through the tree-level result parametrised with ,
W
or the improved Born result parametrised with G , or the improved Born result including QCD corrections (which
is the one we used), is formally of higher order (i.e., beyond O( 2 )) in the calculation of MW . Its numerical effect
is nevertheless not completely negligible; it changes the shift in MW by about 2.9 MeV if the tree-level result
for W parametrised with is used and by about 1.4 MeV if the G parametrisation of the Born width (without
QCD corrections) is employed.
200
constants, which is valid in all orders of perturbation theory [39]:

1/2 sW + sW Z Z
= 1.
Ze Z
+
cW + cW 2
(39)
Expansion up to O( 2 ) yields

2 1 2
1
sW
1
Z
Z
Ze (2) = Z(2)
Z + Ze (1) + Z(1) 3 Z(1) sW (1).
2
2cW (2)
8
2cW
(40)
Furthermore, the two-loop field renormalisation constants for the external leptons are
needed in Eq. (29). These can be easily obtained in the limit of vanishing masses and
momenta of the external fermions, yielding

fL
f L
Z(2) = (2) m2f = 0 .
(41)
4. Extraction of fermi model QED contributions

In the evaluation of r, the IR-divergent QED corrections that are already contained in
the Fermi Model QED factor have to be extracted. For the two-loop calculation presented
here, the corresponding Fermi Model contributions consist of virtual and real photonic
corrections of order O() and of order O( 2 ) with one closed fermion loop, see Fig. 3,
where it is understood that all lepton and quark flavours can appear in the loop. Denoting
the virtual corrections to the Fermi Model by qV and the real corrections by qR , this
reads
|MFermi |2 = |MBorn |2 (1 + q)

()
()
( 2 )
( 2 )
= |MBorn |2 1 + qV + qR + qV + qR .
(42)
The calculation of the virtual corrections to muon decay in the full Standard Model
involves box-type diagrams with IR divergences. In the following, all Standard Model
(a)
(b)
Fig. 3. Virtual (a) and real (b) QED corrections to muon decay in the Fermi Model.
(a)
(b)
201
(c)
Fig. 4. Examples for virtual IR-divergent diagrams contributing to muon decay in the Standard Model. Besides the
one-loop diagram (a), at two-loop order, there are diagrams with four (b) and two (c) electromagnetic couplings.
contributions involving photons in the loop are encompassed by the quantity . The
one-loop QED corrections V() originate from the diagram given in Fig. 4(a). At twoloop order one can distinguish between corrections with only electromagnetic couplings in
2
( )
addition to the tree-level couplings, V,em
, see Fig. 4(b), and corrections with additional
2
( )
QED and non-QED couplings, V,em
/weak , see Fig. 4(c). Here it is always understood that
the two-loop corrections involve one closed fermion loop. By performing a tensor integral
decomposition of these classes of diagrams one observes that they can be expressed in
terms of the virtual corrections in the Fermi Model and an IR-finite remainder rfr ,
()
()
()
V = qV + 2rfr ,
(43)
( 2 )
V,em
(44)
( 2 )
= qV
( 2 )
+ 2 rfr,1 ,
()
()
()
( 2 )
V,em/weak = 2 qV + rfr rferm + 2 rfr,2 .
( 2 )
(45)
Note that the factor 2 in these formulae arises due to fact that q enters linearly into
the muon decay width, see Eq. (1), while there is a quadratic dependence on r, see
Eq. (2). The finite remainders are then combined with all remaining virtual Standard Model
2
()
contributions into r () , r ( ) . In Eq. (45) rferm
corresponds to the non-QED one-loop
corrections with a closed fermion loop.
Besides box-type diagrams, IR divergences are also present in the field renormalisation
of the external leptons. Here the correspondence to the Fermi Model contributions is trivial.
Similar to the virtual diagrams, the real bremsstrahlung corrections to muon decay
in the Standard Model can be divided into the one-loop contribution R() , two-loop
2
( )
corrections with additional electromagnetic couplings only, R,em
, and mixed QED/non( 2 )
QED couplings, R,em/weak , see Fig. 5. Exploiting the fact that the momentum transfer
q 2 through the W boson propagator of these diagrams is much smaller than the W mass,
2 , in the limit of zero momentum transfer the real contributions can be reduced to
q 2 MW
the real Fermi Model contributions,
()
()
R = qR ,
( 2 )
( 2 )
R,em = qR ,
( 2 )
()
R,em/weak = 2qR()rferm
.
(46)
(47)
(48)
202
(a)
(b)
(c)
Fig. 5. Examples for real bremsstrahlung diagrams in the two-loop calculation of muon decay in the Standard
Model, involving one-loop diagrams (a) and two-loop diagrams with four (b) and two (c) QED couplings.
Fig. 6. In order to extract the QED corrections already present in the Fermi Model from the Standard Model
computation of r, differences between QED loop diagrams in the Standard Model and Fermi Model of the
same order have to be evaluated, here exemplified for the one-loop case.
In total, the contributions to the two-loop Standard Model matrix element amount to

2 2
()
( 2 )
()
()
|MSM |2 = |MBorn |2 1 + r () + r ( ) + qV + qV + 2qV rferm
()
( 2 )
()
()
+ qR + qR + 2qR rferm

= |MBorn |2 (1 + q)(1 + r)2 + O 3 ,
(49)
which can be written in the factorised form at least up to two-loop order including one
closed fermion loop. Contributions with two closed fermion loops at O( 2 ) are not present
in the Fermi Model and do not contain any IR divergences. Comparing Eq. (49) with
Eq. (42) one obtains
|MSM |2 = |MFermi |2 (1 + r)2 ,
(50)
showing that a factorisation of electromagnetic corrections to the Fermi Model and the
remaining electroweak corrections in the Standard Model according to Eq. (1), (2) is
possible at least up to the given order.
The calculation of the remaining terms rfr , in which the QED Standard Model and
Fermi Model contributions in Eqs. (43)(45) differ, requires the subtraction of Fermi
Model diagrams from Standard Model graphs, as shown in Fig. 6. These terms are IRfinite but UV-divergent and therefore require regularisation. In our approach, dimensional
regularisation turns out to be problematic for this purpose since for the computation of the
fermion lines in diagrams like those in Fig. 6 we use the Chisholm identity
= i 5 + g + g g .
(51)
203
This identity, however, is only valid in 4 dimensions. In order to circumvent this

problem we employ PauliVillars regularisation for the QED corrections to the Fermi
vertex. The combination of these vertex corrections, Fig. 3(a), with the QED part of the
field renormalisation of the external leptons forms an UV-finite quantity. It is therefore
possible to evaluate this combination using PauliVillars regularisation (PaVi) and employ
dimensional regularisation (DReg) for the rest. This can formally be written as

rfr = QED box graphs in SM DReg QED vertex corr. to FM PaVi

1 L 1 eL
1 L 1 eL
Zem + Zem
Zem + Zem
+
.
(52)
2
2
2
2
DReg
PaVi
L
eL indicates that only the

The index em at the field renormalisation constants Zem , Zem
QED-like diagrams of the lepton self-energies are taken into account for the calculation
of these constants. Since the Standard Model box diagrams are UV finite, they can also
be computed with a PauliVillars regulator. Thus the cancellation of the IR divergences
between the two terms in the first line of Eq. (52) proceeds in a straightforward manner
and no IR regulator is required.
A similar cancellation of IR divergences takes place for the terms in the second line of
Eq. (52). This can be made explicit by introducing the PauliVillars regulator in photon
propagators according to the replacement
1
1
1
2 2
(53)
k2
k
k 2
with k being the photon momentum. In the difference in the second line of Eq. (52) this
corresponds to the replacement of the photon propagator in the lepton self-energies by

1
1
1
1
1
(54)
= 2
2 2 2
,
2
2
k
k
k
k
k 2
or, for the case of two photon propagators in the loop,

1
1
1
1
1
1
1
1
2 2 2 2
k2 k2
k
k
k
k 2
k 2 k 2 2
1
1
1
1
=2 2 2
(55)
.
k
k 2 k 2 2 k 2 2
It can be seen that this replacement effectively leads to the introduction of massive photons
with mass so that no IR divergences are present anymore.
5. Numerical results
We shall now discuss the numerical evaluation of our result for r. It should be noted
that our definition of r according to Eq. (2) is based on the expanded form (1 + r)
2
with r = r () + r ( ) + rather than on the resummed form 1/(1 r). The
terms obtained at two-loop order from a resummation of leading one-loop contributions
are directly contained in our two-loop contribution to r. The following contributions to
204
Fig. 7. Two-loop contributions with one closed fermion loop to r as a function of the Higgs mass. The plot
2
2
shows the full result, r (Nf ) , the contributions from the topbottom doublet, r (Ntb ) , and the light-fermion
2
doublets, r (Nlf ) , as well as the light-fermionic contribution with subtracted resummed terms proportional to
(N 2 )
lf
, rsub,1
-loop , see Eq. (58).
r are taken into account

2
2 2
r = r () + r (s ) + r (s ) + r (Nf ) + r (Nf ) ,
(s2 )
(56)
are the two-loop [14] and

where r () is the one-loop result, Eq. (3), r (s ) and r
2
three-loop [15] QCD corrections, while r (Nf ) is the new electroweak two-loop result.
The symbolic notation (Nf 2 ) encompasses the contribution of all diagrams containing one
fermion loop, i.e., both the top/bottom and light-fermion contributions. Correspondingly,
2 2
the term r (Nf ) contains the pure fermion-loop contributions in two-loop order.
The pure fermion-loop contributions in three- and four-loop order turn out to be
numerically small, as a consequence of accidental numerical cancellations, with a net
effect of only about 1 MeV in MW (using the real-pole definition of the gauge-boson
masses) [12]. Furthermore, also the leading three-loop contributions for a large top-quark
mass in the limit of zero Higgs mass, proportional to 3 m6t and 2 s m4t , have very little
impact on the prediction of MW [13]. Therefore these two corrections have not been
included in this analysis.
In Fig. 7 and Table 1 numerical values for different two-loop contributions with one
closed fermion are given as a function of the Higgs boson mass MH , using MW =
80.451 [5], mt = 174.3 [44] and = 0.05911 [45]. The contributions with two closed
fermion loops are not given in the figure and table since they are independent of MH .
2 2
Numerically, they yield a contribution of r (Nf ) = 16.3 104 . The Higgs-mass
205
Table 1
Two-loop contributions with one closed fermion loop to r for different values of the Higgs mass. Besides the full
2
2
result, r (Nf ) , also the contributions from the topbottom doublet, r (Ntb ) , and the light-fermion doublets,
2
r (Nlf ) , are given. From the latter the term proportional to originating from the resummation prescription
Eq. (57) is subtracted
MH /GeV
2
r (Nf ) /104
2
r (Ntb ) /104
2
()
r (Nlf ) 2rbos /104
19.4
20.1
22.6
24.2
23.8
15.0
13.9
10.9
8.2
4.4
2.0
2.2
3.1
3.9
4.6
65
100
300
600
1000
dependence of the two-loop result for r agrees perfectly with the result previously
obtained in Ref. [21].
2
It can be seen that both corrections with a top/bottom loop, r (Ntb ) , and with a light2
fermion loop, r (Nlf ) yield important contributions. At first glance it looks surprising
that the light-fermion contributions even dominate over the top/bottom contributions for
large Higgs masses (MH 300 GeV), which seems to endanger the validity of the largemt expansions [1820]. However, in previous analyses, resummation prescriptions [11]
have been derived in order to obtain partial terms of the two-loop result. With the
replacement
1 + r
1
1 r
(57)
()
the term 2rbos , generated from the charge renormalisation in bosonic one-loop terms,
2
is correctly predicted [11], as we have checked by comparing with the full result, r (Nlf ) .
Therefore, in order to demonstrate the effect of the new two-loop contribution, the difference
2
(Nlf )
()
(Nlf )
2rbos
rsub,1
-loop = r
(58)
is shown in Fig. 7 and Table 1. This expression does not exceed the top/bottom contributions for any value of the Higgs mass below 1 TeV. The new contribution to r from
diagrams with a light-fermion loop amounts up to 3.3 104 which corresponds to a shift
in MW of > 5 MeV.
The prediction for MW is obtained from Eq. (2) by means of an iterative procedure,
since r itself depends on MW ,

1
1
2
+
= MZ2
MW
(59)
1 + r(MW , MZ , MH , mt , . . .) .
2
4
2 G MZ2
In Fig. 8 the prediction for MW based on the results of Eq. (56) is shown as a function of
MH for mt = 174.3 5.1 GeV [44] and = 0.05911 0.00036 [45]. For comparison,
exp
the present experimental value, MW = 80.451 0.033 GeV [5], and the experimental
95% C.L. lower bound on MH (MH = 114.1 GeV [46]) from the direct search are also
indicated. The plot exhibits the well-known preference for a light Higgs boson within
206
Fig. 8. The SM prediction for MW as a function of MH for mt = 174.3 5.1 GeV is compared with the
exp
current experimental value, MW = 80.451 0.033 GeV [5], and the experimental 95% C.L. lower bound on the
Higgs-boson mass, MH = 114.1 GeV [46].
Table 2
The two-loop result for MW based on Eq. (56) is compared with the results of an
expa
expansion in mt up to O(G2 m2t MZ2 ) [20,47], MW . The last column indicates
the difference between the two results
MH /GeV
65
100
300
600
1000
expa
MW /GeV
MW /GeV
MW /MeV
80.3997
80.3771
80.3051
80.2521
80.2134
80.4039
80.3805
80.3061
80.2521
80.2129
4.2
3.4
1.0
0.0
0.5
the SM. In particular, the theoretical prediction (including the band from a variation
of mt , which at present dominates the uncertainty of the prediction, see below, and
exp
within 1 ) cannot be matched with the 1 region of MW and the 95% C.L. exclusion
limit on MH .
We have compared our results with those of an expansion for asymptotically large values
of mt up to O(G2 m2t MZ2 ) [20,47]. The results are shown in Table 2 for different values of
MH . The values for the input parameters are taken from Ref. [20], i.e., mt = 175 GeV,
MZ = 91.1863 GeV, = 0.0594, s (MZ ) = 0.118. One observes a relatively good
agreement, with maximal deviations in MW of about 4 MeV.
It should be noted that the deviations in the last column of Table 2 cannot be attributed
solely to differences in the two-loop fermionic contributions, because the results also
207
Table 3
Investigation of different sources of deviations between the two-loop
expa
result for MW based on Eq. (56) and the results of [20,47], MW . In the

second column (MW ), differences in the QCD implementation and in
the resummation of bosonic one-loop terms have been eliminated. In
), in addition the two-loop contributions from
the third column (MW
light fermions are excluded
MH /GeV
/MeV
MW
/MeV
MW
4.2
3.7
2.8
3.1
3.8
1.1
0.4
1.2
1.7
1.7
65
100
300
600
1000
differ by a slightly different treatment of higher-order terms that are not yet under
control.
In a further analysis, we have aimed at reducing the latter deviations as far as
possible in order to focus on the effects from the two-loop top-quark and light-fermion
contributions (see also the discussion in Ref. [22]). While the result of Ref. [20]
() 2
contains a term (rbos
) generated from the purely bosonic one-loop contributions by
the resummation Eq. (57), no such term is included in our result. A second possible source
of deviation could be caused by different implementations of the QCD corrections. We
have therefore performed a comparison in which the QCD corrections were removed from
both results. With these modifications the maximum deviation between the results is not
decreased, see second column in Table 3, but the maximal difference in the Higgs-mass
dependence MW (MH ) MW (MH = 65 GeV) is reduced from 4.7 MeV in Table 2 to
1.4 MeV.
It is also interesting to separately examine the effects of the topbottom contributions
that are not contained in Ref. [20] and of the two-loop terms from the remaining lightfermionic flavours. In the third column in Table 3 we have therefore excluded all lightfermionic O( 2 ) contributions from the comparison. This is achieved by subtracting the
2
()
()
()
()
expression r (Nlf ) 2rlf rbos , where the second term 2rlf rbos accounts for
the light-fermionic terms that were included in Ref. [20] by means of the resummation
prescription Eq. (57). The remaining deviations between the results, which now only
contain top/bottom contributions at the two-loop level, are somewhat smaller, while there
are larger differences in the Higgs mass dependence of up to 2.8 MeV.
In Ref. [1] a simple formula was given which parametrises our full result for MW ,
0
MW = MW
c1 dH c5 dH2 + c6 dH4 c2 d + c3 dt c7 dH dt c4 ds ,
(60)
where

MH
,
dH = ln
100 GeV

1,
d =
0.05924
mt
dt =
174.3 GeV
s (MZ )
ds =
1,
0.119
2
1,
(61)
208
and MZ = 91.1875 GeV [5] has been used. By a least square fit we have obtained for the
coefficients c1 , . . . , c7
0
MW
= 80.3767 GeV,
c4 = 0.0763 GeV,
c1 = 0.0561 GeV,
c5 = 0.00936 GeV,
c2 = 1.081 GeV,
c6 = 0.000546 GeV,
c3 = 0.5235 GeV,
c7 = 0.00573 GeV.
(62)
The parametrisation of Eq. (60) approximates our full result for MW within 0.4 MeV for
65 GeV MH 1 TeV and the other input values within their 1 experimental bounds.
Since this region of validity is in general not sufficient for global fits of the Standard
Model, here we supply a more elaborate parametrisation, including the dependence on the
Z-boson mass,
0
MW = MW
d1 dH d2 dH2 + d3 dH4 d4 d + d5 dt d6 dt2
d7 dH dt d8 ds + d9 dZ,
(63)
with dZ = MZ /(91.1875 GeV) 1 and the other symbols as given in Eq. (61). With the
following values for the coefficients d1 , . . . , d9 ,
0
= 80.3768 GeV,
MW
d5 = 0.5236 GeV,
d1 = 0.05619 GeV,
d6 = 0.0727 GeV,
d2 = 0.009305 GeV,
d7 = 0.00544 GeV,
d3 = 0.0005365 GeV,
d8 = 0.0765 GeV,
d4 = 1.078 GeV,
d9 = 115.0 GeV,
(64)
the full result for MW is approximated by Eq. (63) better than 0.3 MeV for 65 GeV
MH 1 TeV and 2 variations of all other experimental input values.
6. Remaining theoretical uncertainties

Presently, the prediction of the W mass from r is mainly affected by the experimental
error in the top mass determination, mt = 174.3 5.1 [44]. This induces an error of
30 MeV in the predicted W mass. It is expected that the LHC can reduce the error
on the top mass down to about 1.5 GeV [48] and a high-luminosity linear collider even to
below 200 MeV [8], resulting in an error in the MW -prediction from the mt -uncertainty
of 10 MeV and 1.2 MeV, respectively. Another important source of uncertainty is the
experimental error in the determination of , which in a recent analysis was quoted to be
36 105 [45], inducing an error of 6.5 MeV in the predicted W mass. It is expected
that this uncertainty will be further reduced significantly in the future [49]. On the other
hand, the experimental error of the direct measurement of the W mass, currently 33 MeV,
is expected to reduce to 15 MeV for the LHC [7] and 6 MeV for a linear collider running
at the W pair threshold [8].
209
Concerning the theoretical prediction, there are three main sources for uncertainties
induced by unknown higher orders: the missing purely bosonic two-loop contributions,
three-loop electroweak contributions and the lowest missing QCD corrections of order
O( 2 s ) and O(s3 ).
For the three-loop O( 3 ) electroweak corrections, partial results are available. In
particular the contribution from purely fermionic loops [12] is known, which amounts
to a net effect of about 1 MeV in MW . Recently, the leading terms for large top masses
proportional to m6t [13], which enter via the quantity , have been calculated, having an
effect of much less than 1 MeV on MW . However, for the case of the O( 2 ) correction
it turned out that the formally leading term m4t [19] in the limit MH = 0 is suppressed
relative to the next term m2t [20], which suggests that the full O( 3 ) corrections could
be of O(1 MeV). Alternatively, one could try to estimate their size from residual scheme
dependencies of the O( 2 ) result. For example, the W width is needed at tree-level in order
to translate between the different BreitWigner parametrisations mentioned in Section 3.2.
Whether W is parametrised with or G is formally of order O( 3 ) and also results in a
shift in MW of O(1 MeV).
For the QCD correction of order O( 2 s ), the leading contribution m4t in an
expansion for large mt has been calculated in the limit of vanishing Higgs mass [13]. It
turned out to result in a W mass shift of only less than 0.5 MeV. However, as explained
above, the formally leading term m4t can be suppressed relative to the sub-leading terms,
so that the total O( 2 s ) contribution could be considerably larger.
Higher order QCD contributions may be estimated from the renormalisation scale
dependence of the available results. For this purpose running MS values at different scales
are used for the top-mass mt or strong coupling s . With this variation in the O( 2 )
result we obtain a shift in MW of 3.8 MeV, which gives an estimate of the O( 2 s )
contributions. Accordingly, from the scale dependence of the O(s2 ) result, we estimate
the effect of the O(s3 ) term to yield 0.7 MeV.
A second method for estimating the missing QCD corrections relies on the assumption
that the ratios of consecutive coefficients in the pertubative series do not change very
much. In this case the ratio between the O( 2 s ) and O( 2 ) contributions to r should
be of the same size as the ratio between r (s ) and r () .3 From this we deduce
that the effect of the O( 2 s ) contribution on MW is about 3.5 MeV, which is in nice
agreement with the previous estimate. In a similar manner, we can compare the ratios
2
3
2
r (s ) /r (s ) and r (s ) /r (s ) and obtain 0.7 MeV for the impact of the O(s3 )
contribution.
The third important higher order contribution, the bosonic O( 2 ) correction, is more
difficult to estimate. In a simple approach the one-loop bosonic contribution is resummed
() 2
according to the prescription Eq. (57) which results in a term (rbos
) . This term shifts
the W mass by less than 0.5 MeV for MH = 100 GeV, but more than 2.5 MeV for
MH = 1 TeV.
We obtain the total theory uncertainty by linearly adding up all sources for theoretical
errors. This results in an uncertainty for MW of 6 MeV for light Higgs masses and about
3 This ratio amounts to 12% in accordance with (M ).
s
Z
210
Fig. 9. Theoretical error band of the MW -prediction due to unknown higher order contributions.
8 MeV for MH 1 TeV, which is similar to the value given in Ref. [50]. The error band
due to these theoretical uncertainties is shown in Fig. 9.
Recently the new two-loop results for the prediction of the W-boson mass have been
implemented into the Standard Model fits with Z FITTER [51] and are subject to the latest
LEP electroweak analyses [5]. While the effect on the predicted value for MW is relatively
small compared to the experimental error, it induces a significant shift in the prediction of
lept
the effective leptonic weak mixing angle sin2 eff according to

2

MW
2 lept
2
sin eff = 1 2 MW
(65)
,
MZ
where incorporates the contributions from radiative corrections. The effect of inserting
the new result for MW in Eq. (65) instead of the previous result obtained from an expansion
in powers of mt [20] amounts to an upward shift of about 8 105 , which is about half the
experimental error of 17 105 [5]. Since the corresponding complete fermionic two-loop
corrections for are not yet known, this shift has been treated as a theoretical uncertainty
and is represented as a rather wide band in the well-known blue-band plot [5].
7. Higgs-mass dependence of bosonic two-loop result

As a first step towards a full O( 2 ) result for r we have calculated the dependence
of the bosonic two-loop corrections on the Higgs-boson mass. This includes the evaluation
of all diagrams without closed fermion loops which contain internal Higgs bosons or MH dependent scalar couplings. Some typical examples are given in Fig. 10. This subset of
211
Fig. 10. Examples for types of bosonic two-loop diagrams with internal Higgs bosons contributing to muon decay.
Table 4
Shift in the predicted W mass caused by varying Higgs-boson mass MH when including
ferm ) and all two-loop corrections (M ferm+bos ).
fermionic two-loop corrections (MW,sub
W,sub
MW,sub gives the difference between these two contributions, i.e., the effect of the
bosonic loops
MH /GeV
100
200
400
600
1000
ferm /MeV
MW,sub
ferm+bos
MW,sub
/MeV
0
43.1
93.7
124.8
163.4
0
42.6
92.9
123.7
161.4
MW,sub /MeV
0
0.5
0.8
1.1
2.0
the complete bosonic two-loop corrections can be evaluated with the methods described in
Sections 24. In particular, the factorisation of IR-divergent QED corrections as in Eq. (49)
also applies for the bosonic MH -dependent contributions.
In order to study the Higgs-mass dependence, the subtracted quantity

( 2 )
( 2 )
( 2 ) 0
MH , MH0 = rbos
MH
(MH ) rbos
rbos,sub
(66)
is considered, using a fixed offset value for the Higgs-boson mass, MH0 = 100 GeV.
( 2 )
The contribution of the MH -dependent diagrams to rbos,sub(MH , MH0 ) forms a finite

and gauge-parameter independent quantity, as we have explicitly checked. This analysis
is in analogy to Ref. [21], were the corresponding quantity for the fermionic two-loop
contributions was studied.
In Table 4 the variation of the prediction for the W-mass MW as a function of the Higgs
mass MH is shown without and with the bosonic two-loop terms, using the input values of
Table 1. As before the values are given in terms of the subtracted quantity

MW,sub (MH ) = MW (MH ) MW MH0 = 100 GeV .
(67)
Fig. 11 shows how the slope of the Higgs-mass dependence is modified due to the
inclusion of the bosonic two-loop terms. The maximum change amounts to 2 MeV in the
region 100 GeV < MH < 1 TeV. As a consequence, from the MH -dependence we get no
indications for any particularly large effects in the full bosonic two-loop corrections to r.
In this context we would like to point out the observation that the Higgs-mass dependence
of the fermionic two-loop corrections [21] provides a rough assessment of the effect of the
full two-loop corrections [1]. This supports the estimation in Section 6 that the expected
size of the purely bosonic O( 2 ) contributions is relatively small.
212
Fig. 11. Variation of the Higgs-mass dependence of the MW -prediction due to the inclusion of bosonic two-loop
corrections.
8. Prospects for future colliders

In the following we illustrate the accuracy that can be reached with future colliders
concerning tests of electroweak physics. Taking the current central values of the
experimental input values, Fig. 12 shows the situation that can be obtained with the LHC
with expected errors for MW and mt of MW = 15 MeV and mt = 1.5 GeV, respectively.
Even more impressive results could be achieved by a high-luminosity linear collider
running at low energies, where errors of MW = 6 MeV and mt = 200 MeV may be
obtained [8], see Fig. 13. In both figures we furthermore assumed that the error in the shift
of the electromagnetic fine structure constant, , will be cut to half of the present value.
9. Conclusion
In this paper, the evaluation of the complete fermionic two-loop contributions to the
MW MZ mass correlation was described, elucidating the applied techniques and the
implications of the new result.
The renormalisation within the on-shell scheme was described in detail. In particular,
the definition of the gauge-boson masses via the complex pole of the S matrix was
studied, ensuring in particular gauge-parameter independence of the renormalised weak
mixing angle and the gauge boson masses. The latter requires to take tadpole contributions
into account. It was shown how the radiative corrections in the Standard Model can be
factorised from the QED corrections within the Fermi Model.
213
Fig. 12. Comparison of prediction and measurement for MW using expected experimental uncertainties at the
LHC and the current central values.
Fig. 13. Comparison of prediction and measurement for MW for expected experimental errors at a high-luminosity
linear collider, assuming current central values.
214
The result for MW was expressed in terms of an accurate numerical parametrisation

valid for all values of the Higgs mass up to 1 TeV. A detailed comparison with a previous
result obtained by an expansion in powers of mt up to next-to-leading order was performed.
Here the effects of the top/bottom and light-fermion contributions were studied separately
and found to yield a contribution of a few MeV to the prediction of MW each.
Furthermore, the remaining theoretical uncertainties due to unknown higher orders were
discussed and an overall uncertainty of the W-boson mass prediction of 6 MeV was
estimated for light Higgs-boson masses. A careful treatment of the theoretical uncertainties
proved to be important for precision tests of the Standard Model. The situation for present
experimental uncertainties was contrasted to the capabilities of aspired future colliders.
As an additional result, the Higgs-mass dependence of the purely bosonic electroweak
two-loop contributions was computed, so that the only yet missing piece of the
complete two-loop calculation for r, i.e., muon decay, is a constant (MH -independent)
contribution. The numerical impact of the bosonic two-loop corrections on the Higgs-mass
dependence of the MW -prediction is relatively small, in accordance with our estimates for
theoretical uncertainties.
Acknowledgements
We thank D. Bardin, P. Gambino, M. Grnewald, S. Heinemeyer, T. Hurth and G. Quast
for useful discussions and communications. We also thank D. Bardin for cooperation in
implementing our result in Z FITTER. This work was supported in part by the European
Communitys Human Potential Programme under contract HPRN-CT-2000-00149 Physics
at Colliders.
Appendix A
In the appendix the explicit Feynman rules for the ghost interactions in the Standard
Model are listed. In the vertices all particles are considered as incoming. The following
expressions comply with a non-renormalisation of the gauge-fixing sector and the use of a
linear R gauge, introducing the (renormalised) gauge parameters , Z , W according to
Eqs. (23)(25). Counterterms in the one-loop approximation are included. The Feynman
rules for the other sectors of the theory can be found, e.g., in Ref. [37].
=
Ghost propagator
u u :
i( )1/2
,
k2
u Z uZ :
i( Z )1/2
,
k 2 Z MZ2
counterterm
GG
u u :

1/2
1
1 Z ,
C1 =
2
i( G )1/2
k 2 G MG2
u u :
i( W )1/2
,
2
k 2 W MW

= i C1 k 2 C2
C2 = 0,
(68)
u uZ :
u Z u :
u Z uZ :
u u :

1/2
1
C1 =
Z Z ,
2

Z 1/2
1 Z
C1 =
Z
,
2

1/2
1
1 Z ZZ ,
C1 = Z
2

W 1/2
1
C1 =
1 Z W ,
2
215
C2 = 0,
C2 = 0,

1/2
1
1
C2 = Z
MZ2 1 Z + MZ2 ,
2
2

W 1/2
1
1
2
2
C2 =
MW
1 Z + MW
.
2
2
(69)
1 G2 V coupling
G
= iek C

1/2
1
1
cW
1 + Ze + Z Z W
C = W
Z Z ,
2
2
2sW

cW W 1/2
1 ZZ 1 W
sW
sW

Z
u u Z:
C =
Z
1 + Ze + Z Z
,
2
sW
2
2
2cW
sW cW

1/2
1
1
cW 1/2 Z
Z ,
1 + Ze + Z W Z
u u W : C =
2
2
2sW

cW Z 1/2
1
1
sW
u Z u W : C =
1 + Ze + Z W Z ZZ
2
sW
2
2
sW cW
1 1/2 Z
Z
Z ,
2
1/2
u u W : C = W
(1 + Ze ),

cW W 1/2
sW
u uZ W : C =
(70)
1 + Ze
.
2
sW
sW cW
u u :
1 G2 S coupling
G
u u H :
Z Z
u u H :
= ieC

2 c2
Z 1/2
sW
1
1 H 1
W sW
1 + Ze +
C=
MZ
+ Z Z ,
2
2sW cW
sW
2
2
cW

1/2
1
sW 1 H 1
1 + Ze
C=
MW W
+ Z Z ,
2sW
sW
2
2
216
u u :
u Z u :
u u :
u uZ :

1/2
1
sW 1 1
MW W
+ Z Z ,
1 + Ze
2sW
sW
2
2

Z 1/2
1
sW 1 1
C=
MZ
+ Z Z ,
1 + Ze
2sW
sW
2
2
W 1/2
C = MW
(1 + Ze ),

2
1/2
s 2 cW
sW
1
1 + Ze + 2
.
C= W
MW W
2 )c 2 s
2sW cW
(sW cW
W W
C = i
(71)
References
[1] A. Freitas, W. Hollik, W. Walter, G. Weiglein, Phys. Lett. B 495 (2000) 338, hep-ph/0007091;
A. Freitas, S. Heinemeyer, W. Hollik, W. Walter, G. Weiglein, Nucl. Phys. (Proc. Suppl.) 89 (2000) 82,
hep-ph/0007129;
A. Freitas, S. Heinemeyer, W. Hollik, W. Walter, G. Weiglein, in: Proc. of the 5th International Symposium
on Radiative Corrections, RADCOR 2000, Carmel, CA, September 2000, hep-ph/0101260.
[2] R.E. Behrends, R.J. Finkelstein, A. Sirlin, Phys. Rev. 101 (1956) 866;
S.M. Berman, Phys. Rev. 112 (1958) 267;
T. Kinoshita, A. Sirlin, Phys. Rev. 113 (1959) 1652.
[3] T. van Ritbergen, R.G. Stuart, Phys. Rev. Lett. 82 (1999) 488, hep-ph/9808283;
T. van Ritbergen, R.G. Stuart, Nucl. Phys. B 564 (2000) 343, hep-ph/9904240;
M. Steinhauser, T. Seidensticker, Phys. Lett. B 467 (1999) 271, hep-ph/9909436.
[4] A. Sirlin, Phys. Rev. D 22 (1980) 971;
W.J. Marciano, A. Sirlin, Phys. Rev. D 22 (1980) 2695;
W.J. Marciano, A. Sirlin, Phys. Rev. D 31 (1980) 213, Erratum.
[5] D. Charlton, in: Proc. of the Int. Europhysics Conference on High-Energy Physics, HEP 2001, Budapest,
July 2001, hep-ex/0110086;
J. Drees, in: Proc. of the XX Int. Symposium on Lepton and Photon Interactions at High Energies, Rome,
July 2001, hep-ex/0110077.
[6] D. Amidei et al., Future electroweak physics at the fermilab tevatron, in: Report of the tev_2000 Study
Group, 1996, FERMILAB-PUB-96/082.
[7] ATLAS Collaboration, Detector and Physics Performance Technical Design Report, CERN/LHCC/99-15
(1999);
CMS Collaboration, Technical Design Reports, CMS TDR 15 (1997/98);
S. Haywood et al., Electroweak physics, in: G. Altarelli, M.L. Mangano (Eds.), Proc. of the Workshop on
Standard Model Physics (and more) at the LHC, Report CERN 2000004, hep-ph/0003275.
[8] Tesla Technical Design Report, Part III, R. Heuer, D.J. Miller, F. Richard, P.M. Zerwas (Eds.), DESY-200111C, hep-ph/0106315;
T. Abe et al., American Linear Collider Working Group Collaboration, in: R. Davidson, C. Quigg (Eds.),
Proc. of the APS/DPF/DPB Summer Study on the Future of Particle Physics, Snowmass, 2001, hepex/0106057.
[9] S. Heinemeyer, T. Mannel, G. Weiglein, DESY 99-117, in: Proc. of the International Workshop on Linear
Colliders, Sitges, April/May 1999, hep-ph/9909538;
J. Erler, S. Heinemeyer, W. Hollik, G. Weiglein, P.M. Zerwas, Phys. Lett. B 486 (2000) 125, hepph/0005024.
[10] M.J. Veltman, Nucl. Phys. B 123 (1977) 89.
[11] W.J. Marciano, Phys. Rev. D 20 (1979) 274;
A. Sirlin, Phys. Rev. D 29 (1984) 89;
M. Consoli, W. Hollik, F. Jegerlehner, Phys. Lett. B 227 (1989) 167.
217
[12] G. Weiglein, Acta Phys. Polon. B 29 (1998) 2735, hep-ph/9807222;

A. Stremplat, Diploma thesis, Univ. of Karlsruhe, 1998.
[13] J.J. van der Bij, K.G. Chetyrkin, M. Faisst, G. Jikia, T. Seidensticker, Phys. Lett. B 498 (2001) 156, hepph/0011373.
[14] A. Djouadi, C. Verzegnassi, Phys. Lett. B 195 (1987) 265;
A. Djouadi, Nuovo Cimento A 100 (1988) 357;
B.A. Kniehl, Nucl. Phys. B 347 (1990) 86;
F. Halzen, B.A. Kniehl, Nucl. Phys. B 353 (1991) 567;
B.A. Kniehl, A. Sirlin, Nucl. Phys. B 371 (1992) 141;
B.A. Kniehl, A. Sirlin, Phys. Rev. D 47 (1993) 883;
A. Djouadi, P. Gambino, Phys. Rev. D 49 (1994) 3499;
A. Djouadi, P. Gambino, Phys. Rev. D 53 (1994) 4111, hep-ph/9309298, Erratum.
[15] L. Avdeev, J. Fleischer, S. Mikhailov, O. Tarasov, Phys. Lett. B 336 (1994) 560;
L. Avdeev, J. Fleischer, S. Mikhailov, O. Tarasov, Phys. Lett. B 349 (1994) 597, hep-ph/9406363, Erratum;
K.G. Chetyrkin, J.H. Khn, M. Steinhauser, Phys. Lett. B 351 (1995) 331, hep-ph/9502291;
K.G. Chetyrkin, J.H. Khn, M. Steinhauser, Phys. Rev. Lett. 75 (1995) 3394, hep-ph/9504413.
[16] K.G. Chetyrkin, J.H. Khn, M. Steinhauser, Nucl. Phys. B 482 (1996) 213, hep-ph/9606230.
[17] J. van der Bij, M.J. Veltman, Nucl. Phys. B 231 (1984) 205.
[18] J.J. van der Bij, F. Hoogeveen, Nucl. Phys. B 283 (1987) 477.
[19] R. Barbieri, M. Beccaria, P. Ciafaloni, G. Curci, A. Vicere, Phys. Lett. B 288 (1992) 95;
R. Barbieri, M. Beccaria, P. Ciafaloni, G. Curci, A. Vicere, Phys. Lett. B 312 (1992) 511, hep-ph/9205238,
Erratum;
R. Barbieri, M. Beccaria, P. Ciafaloni, G. Curci, A. Vicere, Nucl. Phys. B 409 (1993) 105;
J. Fleischer, O.V. Tarasov, F. Jegerlehner, Phys. Lett. B 319 (1993) 249;
J. Fleischer, O.V. Tarasov, F. Jegerlehner, Phys. Rev. D 51 (1995) 3820.
[20] G. Degrassi, P. Gambino, A. Vicini, Phys. Lett. B 383 (1996) 219, hep-ph/9603374;
G. Degrassi, P. Gambino, A. Sirlin, Phys. Lett. B 394 (1997) 188, hep-ph/9611363.
[21] S. Bauberger, G. Weiglein, Phys. Lett. B 419 (1998) 333, hep-ph/9707510;
G. Weiglein, in: J. Sol (Ed.), Proc. of the IVth International Symposium on Radiative Corrections, World
Scientific, Singapore, 1999, p. 410, hep-ph/9901317.
[22] P. Gambino, A. Sirlin, G. Weiglein, JHEP 9904 (1999) 025, hep-ph/9903249.
[23] F. Jegerlehner, M.Y. Kalmykov, O. Veretin, hep-ph/0105304.
[24] P. Malde, R.G. Stuart, Nucl. Phys. B 552 (1999) 41, hep-ph/9903403.
[25] J. Kblbeck, M. Bhm, A. Denner, Comput. Phys. Commun. 60 (1990) 165;
H. Eck, J. Kblbeck, Guide to FeynArts 1.0, Univ. of Wrzburg, 1992;
T. Hahn, Nucl. Phys. (Proc. Suppl.) 89 (2000) 231, hep-ph/0005029;
T. Hahn, FeynArts 2.2 Users Guide, Univ. of Karlsruhe, 2000.
[26] G. Weiglein, R. Scharf, M. Bhm, Nucl. Phys. B 416 (1994) 606, hep-ph/9310358;
G. Weiglein, R. Mertig, R. Scharf, M. Bhm, in: D. Perret-Gallix (Ed.), New Computing Techniques in
Physics Research 2, World Scientific, Singapore, 1992, p. 617.
[27] G. t Hooft, M.J. Veltman, Nucl. Phys. B 153 (1979) 365.
[28] A.I. Davydychev, J.B. Tausk, Nucl. Phys. B 397 (1993) 123.
[29] S. Bauberger, F.A. Berends, M. Bhm, M. Buza, Nucl. Phys. B 434 (1995) 383, hep-ph/9409388;
S. Bauberger, F.A. Berends, M. Bhm, M. Buza, G. Weiglein, Nucl. Phys. (Proc. Suppl.) B 37 (1994) 95,
hep-ph/9406404;
S. Bauberger, M. Bhm, Nucl. Phys. B 445 (1995) 25, hep-ph/9501201.
[30] C.G. Bollini, J.J. Giambiagi, Nuovo Cimento B 12 (1972) 20;
J.F. Ashmore, Lett. Nuovo Cimento 4 (1972) 289.
[31] G. t Hooft, M.J. Veltman, Nucl. Phys. B 44 (1972) 189.
[32] W.A. Bardeen, R. Gastmans, B. Lautrup, Nucl. Phys. B 46 (1972) 319.
[33] P. Breitenlohner, D. Maison, Commun. Math. Phys. 52 (1977) 11.
[34] F. Jegerlehner, Eur. Phys. J. C 18 (2001) 673, hep-th/0005255.
[35] P.A. Grassi, T. Hurth, M. Steinhauser, Ann. Phys. 288 (2001) 197, hep-ph/9907426;
218
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
[51]
P.A. Grassi, T. Hurth, in: Proc. of the 5th International Symposium on Radiative Corrections, RADCOR
2000, Carmel, CA, September 2000, hep-ph/0101183.
M. Jamin, M.E. Lautenbacher, Comput. Phys. Commun. 74 (1993) 265.
A. Denner, Fortsch. Phys. 41 (1993) 307.
S. Bauberger, G. Weiglein, Nucl. Instrum. Methods A 389 (1997) 318, hep-ph/9611445.
S. Bauberger, Doctoral thesis, Univ. of Wrzburg, 1997.
S. Willenbrock, G. Valencia, Phys. Lett. B 259 (1991) 373;
A. Sirlin, Phys. Rev. Lett. 67 (1991) 2127;
R.G. Stuart, Phys. Lett. B 262 (1991) 113;
H. Veltman, Z. Phys. C 62 (1994) 35;
M. Passera, A. Sirlin, Phys. Rev. D 58 (1998) 113010, hep-ph/9804309;
P. Gambino, P.A. Grassi, Phys. Rev. D 62 (2000) 076002, hep-ph/9907254;
A.R. Bohm, N.L. Harshman, Nucl. Phys. B 581 (2000) 91, hep-ph/0001206.
A. Denner, G. Weiglein, S. Dittmaier, Nucl. Phys. B 440 (1995) 95, hep-ph/9410338.
D.Y. Bardin, A. Leike, T. Riemann, M. Sachwitz, Phys. Lett. B 206 (1988) 539.
M.L. Nekrasov, Talk presented at the XVth International Workshop High Energy Physics and Quantum
Field Theory, Tver, September 2000, hep-ph/0102284;
B.A. Kniehl, A. Sirlin, hep-ph/0110296.
D.E. Groom et al., Particle Data Group Collaboration, Eur. Phys. J. C 15 (2000) 1.
H. Burkhardt and B. Pietrzyk, LAPP-EXP-2001-03.
The LEP working group for Higgs boson searches, LHWG-NOTE-2001-03, hep-ex/0107029.
P. Gambino, private communication.
M. Beneke et al., Top quark physics, in: G. Altarelli, M.L. Mangano (Eds.), Proc. of the Workshop on
Standard Model Physics (and more) at the LHC, Report CERN 2000-004.
F. Jegerlehner, LC Note LC-TH-2001-035, hep-ph/0105283.
U. Baur et al., in: R. Davidson, C. Quigg (Eds.), Proc. of the APS/DPF/DPB Summer Study on the Future
of Particle Physics, Snowmass, 2001, hep-ph/0202001.
D.Y. Bardin, P. Christova, M. Jack, L. Kalinovskaya, A. Olchevski, S. Riemann, T. Riemann, Comput. Phys.
Commun. 133 (2001) 229, hep-ph/9908433.

Hermitian analyticity, IR/UV mixing and unitarity of

noncommutative field theories
Chong-Sun Chu a, , Jerzy Lukierski b,1 , Wojtek J. Zakrzewski a
a Centre for Particle Theory, Department of Mathematical Sciences, University of Durham,
Durham, DH1 3LE, UK

b Institute for Theoretical Physics, University of Wroclaw, pl. M. Borna 9, 50-205 Wroclaw, Poland
Abstract
The IR/UV mixing and the violation of unitarity are two of the most intriguing aspects of
noncommutative quantum field theories. In this paper the relation between these two phenomena
is explained and established in an explicit form. We start out by showing that the S-matrix of
noncommutative field theories is hermitian analytic. As a consequence, a noncommutative field
theory is unitary if the discontinuities of its Feynman diagram amplitudes agree with the expressions
calculated using the Cutkosky formulae. These unitarity constraints relate the discontinuities of
amplitudes with physical intermediate states and allow us to see how the IR/UV mixing may lead to a
breakdown of unitarity. Specifically, we show that the IR/UV singularity does not lead to the violation
of unitarity in the spacespace noncommutative case, but it does lead to its violation in a spacetime
noncommutative field theory. As a corollary, noncommutative field theory without IR/UV mixing
will be unitary in both the spacespace and spacetime noncommutative case. To illustrate this, we
introduce and analyse the noncommutative Lee modelan exactly solvable quantum field theory.
We show that the model is free from the IR/UV mixing in both the spacespace and spacetime
noncommutative cases. Our analysis is exact. Due to absence of the IR/UV mixing one can expect
that the theory is unitary. We present some checks supporting this claim. Our analysis provides a
counter example to the generally held belief that field theories with spacetime noncommutativity
are nonunitary. 2002 Elsevier Science B.V. All rights reserved.
Keywords: Noncommutative geometry; Unitarity; Analyticity; S-matrix
* Corresponding author.
E-mail address: chong-sun.chu@durham.ac.uk (C.-S. Chu).

1 Supported by KBN grant 5P03B05620.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 2 1 6 - X
220
C.-S. Chu et al. / Nuclear Physics B 632 (2002) 219239
1. Introduction
Recently there has been a lot of activities in constructing and understanding field
theories on noncommutative spacetime (see, e.g., [1,2]). There are many reasons why such
approaches are of interest, most of them related to the desire to take into consideration the
quantum gravity effects and to understand the nature of spacetime at very short distances
(see, e.g., [3,4]). Some of the most recently considered noncommutative geometries are the
2 [1012], and the
noncommutative Minkowski space RD1,1 [59], the fuzzy sphere SN
-Minkowski spacetime [1315]. The algebra of functions on noncommutative RD1,1
is generated by noncommutative spacetime coordinates x obeying the commutation
relations (, = 0, 1, . . . , D 1)
[x , x ] = i ,
(1.1)
2 is generated by
where is an antisymmetric constant matrix. The fuzzy sphere SN
hermitian operators x = (x1 , x2 , x3 ) satisfying the defining relations (i, j, k = 1, 2, 3)
[xi , xj ] = iN ij k xk ,
x 12 + x22 + x32 = R 2 .
(1.2)
Here the noncommutativity parameter N has the dimension of length and should be taken
positive. The radius R of the fuzzy sphere is quantized, in units of N , by

R
N N
(1.3)
+ 1 , N = 1, 2, . . . .
=
N
2 2
The -Minkowski spacetime is defined by the basic relations between the three commuting
space coordinates ([xi , x j ] = 0) and a noncommutative quantum time variable t(x0 = ct ):
[x0 , xi ] =
i
xi .
(1.4)
In this paper we consider the case of noncommutative RD1,1 . This topic has been
studied extensively (for a recent review, see, e.g., [1,2] and references therein). Field theory
on this noncommutative space can be obtained by the replacement of standard products of
fields by the Moyal -product induced by the relation (1.1),2

i
A B(x) A B(x) = e 2 z
z
A(x + z
)B(x + z
)

.
(1.5)
z =z =0
In the momentum basis, the result of such an operation is the appearance of an additional
Moyal phase factor V (k 1 , . . . , k N )

i
1
2
N
eik x eik x eik x = V k 1 , . . . , k N ei i k x ,

i
1

N
i j
V k , . . . , k := exp
(1.6)
k k .
2
ij
2 We denote x = (x, t), z

= (z
,
), z
= (z
) and use the notation a b a b.
221
Due to this phase factor one has to fix a definite cyclic ordering (say, anticlockwise) of the
momenta that enter any vertex of a given Feynman diagram.
An intriguing phenomenon for the quantum field theory on noncommutative RD1,1
is the existence of an infrared/ultraviolet (IR/UV) mixing [16] in the quantum effective
action. Due to this mixing, IR singularities arise from integrating out the UV degrees
of freedom. This threatens the renormalizability and even the consistency of a QFT on
noncommutative RD1,1 . Hence a better understanding (beyond the technical level) of
the mechanism of IR/UV mixing and possible ways to resolve it are certainly highly
desirable. We recall that so far in the literature, field theory on noncommutative RD1,1
has been quantized by following the standard perturbative procedures: namely, the action
is expanded around the free action and the corresponding Feynman rules are then written
down. This is justified in the commutative case; however, since the introduction of
necessarily breaks the Lorentz symmetry from SO(D 1, 1) to a smaller group that is left
unbroken by the commutation relations (1.1), it is actually quite unnatural to employ the
standard perturbative vacuum, i.e., the one defined by the free action and so respecting the
full Lorentz symmetry. This leads one to suspect that the IR/UV mixing may be reflecting
only the properties of the perturbation theory, and may be altered or disappear completely
in the full nonperturbative regime (see, for example, [17]). An exactly solvable field theory
would be a good ground for testing this idea.3 This leads us to introduce and study the
noncommutative Lee model.
Another intriguing phenomenon for any quantum field theory on noncommutative
spacetime is that unitarity could be violated. It is commonly believed that noncommutative
field theory with spacespace noncommutativity is unitary, while theory with spacetime
noncommutativity is not. This is consistent with the fact that spacespace noncommutative
field theory can be embedded in string theory [9,1822], while field theory with space
time noncommutativity cannot [2325]. In [26] it was found that the unitarity constraints
(see (2.11)) are satisfied for noncommutative theories with space noncommutativity but are
violated for theories with a noncommuting time (see also [2730] for recent discussions).
However, these constraints are, in general, actually a stronger statement than the unitarity
itself. The constraints presume a symmetric condition (see (2.12)) which is not generally
valid. Without making any additional assumptions, in this paper, we examine directly the
analyticity and unitarity of the S-matrix of a general noncommutative field theory. We show
that Feynman amplitudes of a noncommutative theory are hermitian analytic (see (2.6)), a
useful characterization of the S-matrix as introduced and proven by Olive [31]. As a result,
the statement that the S-matrix is unitary leads to the boundary-analytic form (2.7); and
that the discontinuity of a Feynman diagram amplitude can be computed according to the
Cutkosky formulae [32].
Although these two phenomena have received a lot of attention and have been throughly
discussed in the literature, as far as we know, the relation between them has not been
identified explicitly and explained before. One of the main aims of this paper is to identify
and explain such a relation between IR/UV singularity and the possible violation of
3 This idea in analysing the IR/UV mixing using an exactly solvable theory was also considered by Alvarez-
Gaum and his collaborators. In particular, the noncommutative Schwinger model has been considered. But the
model becomes untractable in the noncommutative case. We thank L. Alvarez-Gaum for discussions on this.
222
unitarity in a noncommutative field theory. This relation will be established through the
boundary-analytic unitarity constraints (2.7). The basic idea is that the unitarity constraints
allow one to relate the discontinuity of a scattering amplitude in a physical region with
the appearance of intermediate states that can be put on-shell in this region. However,
in a noncommutative theory, IR singularities can also be generated due to the IR/UV
mixing. These new singularities do not correspond to any physical intermediate degrees
of freedom. So, generally, one can expect that the unitarity constraints could be violated.
In this paper we show, that in the case of spacespace noncommutativity, the new IR
singularities are safe in the sense that they do not generate any discontinuities in the
scattering amplitudes. However, the IR singularities do generate such discontinuities in
the spacetime noncommutative case. This is the basic field theoretic mechanism for the
violation of unitarity in a noncommutative theory. We stress that this violation of unitarity
occurs only if time is noncommuting and in the presence of singularities due to the IR/UV
mixing.
To illustrate the above ideas, we introduce and analyse the noncommutative Lee
model. Lee model [33] is an exactly solvable, nonrelativistic model. The noncommutative
Lee model can be defined by using the deformed product of fields (1.5). The model
remains exactly solvable. We show that the noncommutative Lee model is free from
the IR/UV mixing both at the perturbative level, and in the full exact answer. Thus the
noncommutative Lee model does not provide a resolution of the IR/UV mixing issue. This
may appear to be disappointing from the point of view of looking for a nonperturbative
resolution of the IR/UV mixing issue. Nevertheless, the absence of an IR/UV singularity
in a noncommutative field theory is nontrivial. This is one of the main results of this
paper. Moreover, due to the absence of the IR/UV mixing, one can expect, from the above
mentioned general arguments, that the Lee model with spacetime noncommutativity is
unitary. We provide some further arguments to support this claim.
The plan of our presentation is as follows. In Section 2.1, we review some basic facts
about the S-matrix of commutative field theory. In Section 2.2, we prove that Feynman
diagram amplitudes in a noncommutative field theory are hermitian analytic and we
investigate the consequences of this statement on the unitarity of the theory. We show that
the usual form of the unitarity constraints used by many people is not correct in general.
We derive the correct form of the unitarity constraints and show how they can be used to
check the unitarity of a given noncommutative theory. In Section 2.3, we explain how a
IR/UV singularity may lead to a breakdown of unitarity in spacetime noncommutative
field theory. In Section 3, we study the issue of the IR/UV mixing and unitarity in the
noncommutative Lee model. In Section 3.1 we describe the commutative Lee model. We
show that this model is renormalizable with the renormalization constants easily computed
in a closed form. It is well known that the original Lee model in 4-dimensional spacetime
has a ghost state and is not unitary [3335]. We discuss improved versions of the original
Lee model that do not have these problems; and restrict ourselves to these models when we
introduce noncommutativity and address the issue of the unitarity of the noncommutative
model. This we do in Section 3.2 where we introduce the spacespace noncommutative
and the spacetime noncommutative Lee model via the substitutions (3.33) and (3.34). We
show that there is no IR/UV mixing in either case and one can expect that the theory is
unitary. We present some arguments supporting this claim.
223
2. Unitarity and hermitian analyticity

In this section, we discuss some useful properties of the S-matrix. We refer the reader
to [36] and to the excellent monograph [37] for further details on this subject. We follow
the notations and nomenclature of [37].
2.1. S-matrix in the commutative case
First we consider the commutative case. Unitarity of a quantum field theory follows
from the existence of a hermitian Hamiltonian. In terms of the onshell S-matrix, unitarity
is the statement that
SS = S S = 1.
(2.1)
Due to the cluster decomposition property of the S-matrix, it is meaningful to decompose

S into two parts
S = 1 + iT ,
(2.2)
where T is the transition matrix. Written in terms of Tab := a|T |b, we have

Tab Tba
=i
Tna
Tnb = i
Tan Tbn
,
n
(2.3)
where the sum is over all intermediate states associated with putting particles onshell.
The S-matrix and the transition matrix T are defined for external particles with real
momenta. Since both are invariant under proper Lorentz transformations their matrix
elements (transition amplitudes) must be functions of Lorentz scalars which can be formed
out of the momenta. We call a combination of external lines of the amplitude for a given
physical process a channel, and two channels whose lines are disjoint and exhaustive a
reaction. For an amplitude with n external lines, there are 2n1 n 1 different reactions
provided that we exclude reactions with single-particle channels and do not distinguish the
direction of the reaction. The channel invariant variable is the square of the energy in the
given channel C,

2

pi
s = sC =
(2.4)
iC
where pi are the momenta of incoming and outgoing lines, respectively. sC s are
generalizations of the Mandelstam s, t, u variables for 2 2 scattering. It is convenient
to discuss the singularity structure of a scattering amplitude in terms of the space of these
2n1 n 1 different channel invariants. For more details see [36].
The transition amplitudes typically have singularities. In perturbation theory, the
transition amplitude Tab is given by the sum of a number of Feynman diagrams Mab ,
each corresponding to a different channel. The Feynman integral is typically of the form
IG (p) =

L
l
d D kl
I

i
qi2
i
B,
m2i
(2.5)
224
where B is a real normalization factor that contains the couplings and factors of , i, etc.,
and ps are the external momenta. As we have said before, the integral can be written
in terms of the ss. If one extends s to the complex plane, then the singularities are
typically branch points in the complex s-plane.4 Extending s to the complex domain, one
can think of Tab (or Mab ) as the boundary value of an analytic function defined on the
complex s-plane. The resulting analytic function has singularities on the real s-axis that
correspond to physically accessible momenta. These singularities are called the physical
region singularities. In addition, this analytic function may have additional singularities
that correspond to external momenta that are not physically accessible. The analysis of
these additional singularities is more complicated and is not usually performed.
The existence of singularities in the amplitude is a consequence of unitarity [31]. The
reasoning is that as the channel invariant increases past a certain threshold (in the physical
region of the considered amplitude) that corresponds to a new possible intermediate
state, a new term enters the unitarity equation and this gives rise to a singularity in that
channel. Such singularities are called normal thresholds. The physical region is divided
into segments by the normal thresholds singularities. It can shown, within perturbation
theory, that the amplitudes in these segments can be continued consistently into the
complex plane and be related analytically if one adopts in the Feynman integrals the +i
prescription by replacing m2 m2 i, > 0. This corresponds to associating an +i
with a channel invariant when it is close to a normal threshold. The +i prescription in the
correct invariant is appropriate for all physical region normal thresholds in all amplitudes
[37]. Furthermore, it can be shown that the Feynman amplitudes (and hence also T ) are
hermitian analytic [31], i.e., they satisfy

Mab (s) = Mba s .
(2.6)
As a consequence of the hermitian analyticity (2.6), the unitarity relation (2.3) can be
put in a more elegant form
Disc Tab = i
() (+)
Tan
Tnb = i
(+) ()
Tan
Tnb .
(2.7)
Here f () denotes the boundary values, on the real axis, respectively from above and below
the cut, of a complex function f ,
f () (s) := lim f (s i),
0+
s R,
(2.8)
and Disc f is the discontinuity across this cut

Disc f := f (+) f () .
(2.9)
4 The locations of the singularities are determined by the Landau equations, see, for example, [37]. We remark
that the Landau equations are entirely fixed in terms of the singularity manifold T of the integrand of the
Feynman integral, and since noncommutativity modifies the integrand by a phase factor, the Landau equations
are unmodified by noncommutativity.
225
The relation (2.7) is actually somewhat stronger. Indeed, as a result of unitarity and
hermitian analyticity, it holds for each individual Feynman diagram [32]

() (+)
(+) ()
Disc Mab = i
(2.10)
Man
Mnb = i
Man
Mnb .
n
In (2.7) and (2.10) the discontinuities in a given channel of the amplitude are associated
with normal thresholds.
()
In terms of Feynman diagrams, the matrix elements Mab are given, respectively, in
2
2
terms of the i prescription: m m i. The RHS of (2.10) can be computed using
the cutting rules of Cutkosky [32]: first cut the diagram in all possible ways such that
the cut propagators can go on shell simultaneously (for a given set of ss), then, for
each cut, replace the propagators by 2i(p2 m2 ) in the relativistic case, and by
2i(p0 E(p, m)) in the nonrelativistic case. Finally sum the contributions of all
possible cuts.
Before we embark on the noncommutative case, let us remark that Eq. (2.3) is
sometimes written in the form [38] (or for M),

Tna
Tnb .
2 Im Tab =
(2.11)
n
To arrive at this form, the following symmetric relation

Tab = Tba
(2.12)
has been assumed. This relation holds, for example, when the theory is T -invariant and
rotationally invariant, and the basis vectors |a are chosen to be eigenstates of the total
angular momentum [39]. However, we would like to stress that this relation is not true in
general. Failure of (2.11) can be due to either the symmetry condition (2.12) or the unitarity
of the theory (2.3) not being satisfied or if the amplitude possesses singularities which are
not due to the possible intermediate states. Therefore, generically, (2.11) is not a conclusive
check of whether a given theory is unitary or not. In the next subsection we show that the
hermitian analyticity remains valid in the noncommutative case and, therefore, that (2.7)
and (2.10) can be used to check unitary of a noncommutative theory.
2.2. S-matrix in the noncommutative case
In a noncommutative quantum field theory the propagators take the same form as in
the commutative case while the vertices are modified by the Moyal phase factor (1.6) that
arises from the noncommutative multiplication. For example, in the noncommutative 3
model, the modification of the (real) coupling is a multiplication by a real factor

1
g g cos p k ,
(2.13)
2
where k and p are the momenta entering the vertex. However, it is easy to see that when the
theory involves more fields, the modification of the vertex is, generally, a phase factor. For
example, this is the case for the noncommutative Lee model to be introduced in the next
226
section. The phase factor (1.6) is cyclically symmetric but not permutation symmetric.
Therefore, the symmetric relation is, in general, not valid.
Since Lorentz invariance is broken, in addition to the channel invariants we have
introduced above, the S-matrix of a noncommutative field theory generally depends also
on the variables

2

pi ,
pi := pi .
sC =
(2.14)
iC
A novelty in noncommutative theory is the possible existence of the IR/UV mixing

[16], which states that the amplitudes in a noncommutative theory become singular in
the s = 0 limit as one removes the cutoff, i.e., . These singularities occur in
the physical region of momenta but do not correspond to normal thresholds since the
IR/UV singularities are not related to any new degrees of freedom. One may extend the
amplitude analytically to above the cut associated with these singularities by adding +i
to s . This corresponds to extending the i prescription for the Feynman diagram to the
cutoff: 2 2 + i since the combination 1/2 s often appears together [16].
Hermitian analyticity
Next we examine the hermitian analyticity of a noncommutative Feynman diagram. We
show that the Feynman amplitudes for noncommutative theories are hermitian analytic. To
see this, we note that under the complex conjugation, the Moyal phase factor (1.6) becomes

V k1, k2, . . . , kN = V kN , . . . , k2, k1 ,
(2.15)
i.e., it reverses the cyclic ordering of the momenta entering the vertex. We can interpret the
RHS as the Moyal phase factor of a vertex which is the mirror image of the original one,
see Fig. 1. In the operator language the RHS of (2.15) corresponds to a Wick contraction
in the reverse order. For example,
Mab 0|a1 a2 (1 2 3 )a3 |0 V (k1 , k2 , k3 ),
Mba 0|a3 (3 2 1 )a1 a2 |0 V (k3 , k2 , k1 ) = V (k1 , k2 , k3 ) ,
where |a = a1 (k1 )a2 (k2 )|0, |b = a3 (k3 )|0 in this example. In general, let

V G :=
Vv
(2.16)
(2.17)
vG
be the product of the Moyal phase factors associated with the vertices v of a Feynman
diagram G. We have
G

V
(2.18)
= V G,
is the mirror diagram of G.
where G
In a noncommutative theory, the Feynman amplitude for a diagram G takes the form
d D kl B
G
G
Mab (s, s ) =
(2.19)
+ V .
D
i
l,i
227
They have the opposite Moyal phase factors.

Fig. 1. A Feynman diagram G and its mirror diagram G.
Here 1/Di is the propagator of the ith internal line and the mass square has a small
imaginary part and B is a real normalization factor that contains the couplings5 and factors
of , i, etc. Complex conjugating, one has

G
(s, s )
Mab
d D kl B

G
G
(V ) = Mba s , s ,
Di
l,i
(2.20)
where we have used in the last step the observation that a change of sign in the imaginary
part of the mass (or cutoff) corresponds to the change of sign in the imaginary part of s
(or s ). In the discussion given above, for the clarity of the argument, we have been careful to
is to be drawn for the Feynman amplitude to be computed.
indicate which diagram (G or G)
However, this is not really necessary as which diagram has to be drawn is already clear once
the process to be considered (a b or b a) is specified. Therefore, can simply write

Mab (s, s ) = Mba s , s .
(2.21)
Thus we have shown that the Feynman diagrams (and hence the S-matrix) of a
noncommutative theory are hermitian analytic. We stress that our result is general and
does not depend on the detailed form of the propagators or vertices. For example, it applies
to the noncommutative Lee model to be introduced in Section 3.
2.3. Unitarity constraints and their relation to the IR/UV singularities
Note that the symmetric condition (2.12) is, in general, not valid and so the condition
(2.11) may not hold even if a theory is unitary. However, since Feynman amplitudes satisfy
5 We emphasis that the couplings (bare as well as the renormalized one) have to be real. As we discuss at the
end of Section 3.1, the original Lee model (defined in 4-dimensional spacetime and with the dispersion relations
(3.2)) has an imaginary bare coupling [33] and hermitian analyticity does not hold, in both the commutative and
noncommutative cases. However, the improved Lee models have real couplings and so have hermitian analytic
S-matrix.
228
hermitian analyticity, (2.7) and (2.10) hold if the S-matrix is unitary. Therefore, we propose
to use (2.7) or (2.10) instead of (2.11)6 as a check of unitarity.
Before we consider a specific model, let us discuss how the IR/UV singularities may
lead to a breakdown of unitarity in general. Generally, a new IR/UV singularity in a
scattering amplitude can be a pole or a branch point in s = 0, for some s . Note that

s = (E )2 p02 p12 + (B )2 p22 + p32 ,
(2.22)
where we have chosen, for example, 01 = E , 23 = B with all other components

vanishing. Therefore, for space noncommutativity, s is positive definite and so there is
no new contribution to the discontinuity of the amplitude from this singularity. However,
in the case of spacetime noncommutativity, s is not of definite sign in the physical region
[26]. Therefore if s is a branch point singularity, there will now be a new contribution to
the LHS of (2.10). Since the IR/UV singularities do not correspond to any intermediate
degrees of freedom that can go on shell, these new contributions will not be accounted
for by the onshell sum and (2.10) will be violated. This is the basic mechanism how
unitarity is violated by the IR/UV singularities when time is noncommuting. Both the
IR/UV singularity and the noncommuting time must be present in order to violate unitarity.
Finally, we would like to add, as was shown in [27], that even when one tries to add new
degrees of freedom to satisfy the cutting rules in a formal sense, these new degrees of
freedom have to be tachyonic and so the theory is inconsistent.
3. An application: the noncommutative Lee model

In this section, we consider the Lee model in D spacetime dimensions and its
noncommutative generalization. In particular, we consider the issues of the IR/UV
mixing and unitarity for the noncommutative Lee model. We will find that due to the
presence of the Moyal phase factors the symmetric condition is not satisfied. Therefore,
one should check unitarity using (2.10). We show that (and this result is exact) the
noncommutative Lee model is free from any IR/UV singularity. As a result, one can expect
that the noncommutative Lee model is unitary for both the spacespace and spacetime
noncommutative case. We give further arguments supporting this claim.
Another model which is free from the IR/UV mixing is the noncommutative Chern
Simons model. This model is finite and, as shown by [40], free from the IR/UV mixing
at the one loop level. However, it is actually a free theory, at least in the axial gauge [41].
Thus this model is not suitable for our purposes.
6 In [26], the 1 1 propagator diagram in the noncommutative 3 and the 2 2 scattering diagram in
the noncommutative 4 were considered. It is easy to see that the symmetric condition (2.12) is satisfied for
these processes, and so checking of (2.11) constitutes a valid test of unitarity for the noncommutative theories
considered there.
229
3.1. Commutative case

The Lee model was originally introduced by Lee in [33] where it was shown that the
model is renormalizable with its mass, wavefunction and charge renormalizations easily
performed in an exact manner. In the following, we follow the presentation of [39]. The
(0)
model has two fermions V and N with masses m(0)
V , mN , respectively, and a real scalar
(0)
with mass m := 0 . The Hamiltonian for the free fields is

H0 = d D1 p EV (p) V (p)V (p) + EN (p)N (p)N(p) + E (p) (p)(p) ,
(3.1)
where EV (p), EN (p), E (p) are the dispersion relations for the free V , N and
particles, N(p), V (p) and (p) are the annihilation operators of the N, V and particles,
respectively. In the original Lee model [33], D = 4 and the fermions are taken to be very
heavy while is assumed to be relativistic. In this case, the dispersion relations are given by
1/2

(0)
(0)
E N = mV ,
E (k) = k2 + 20
:= k .
E V = mV ,
(3.2)
The Galilei-invariant form [44]
EA (p) =
p2
2m(0)
A
A = V , N, ,
as well as the relativistic choice [45]

(0) 2 1/2
EA (p) = p2 + mA
(3.3)
(3.4)
were also studied in the literature. The interacting Hamiltonian of the model is taken to be
given by

d D1 k
d D1 p V (p)N(p k)(k)f (k)
Hint = g0
D1
(2)
2k

+ (k)N (p k)V (p)f (k) ,
(3.5)
where f (k) is a form factor7 introduced to smooth out the interaction to avoid the divergences connected with a point interaction. In fact f can be taken to be f = 1 and the
divergences can be absorbed by renormalization. This is the case of interest to us. However, as we will see, the introduction of noncommutativity to the Lee model amounts to a
modification of f by a phase factor. Therefore, we will keep f explicitly in the presentation below, with the understanding that it will be set to 1 (or to the Moyal phase factor for
the noncommutative case) in the final answer.
7 Note that, in principle, one can also use a more general form factor f that depends on the momentum of the
N pair. It is easy to see that this amounts to a simple replacement
f (k) f (k, p)
in the analysis below.
(3.6)
230
We note that the interaction Hint is nonlocal in space even in the limit f = 1. To see
this, it is convenient to introduce the negative and positive frequency parts of :
(x) = a(x) + a (x),

a(x) =

a (x) =
(3.7)
d D1 k

(k)eikx ,
(2)D1 2k

d D1 k
(2)D1 2k
(k)eikx .
(3.8)
In terms of a and a , Hint can be written in the coordinate space as

Hint = g0 d D1 x d D1 y V (x, t)N(x, t)f(x y)a(y, t)

+ N (x, t)V (x, t)f (x y)a (y, t) ,
(3.9)
where f is the Fourier transform of the Lee model form factor and f (x) in the limit
f 1. It is now clear that the coupling term is nonlocal in space since the operation of
taking the positive frequency part involves the integration over all space. However, the
model is local in time.
Since the theory is local in time, it can be described equivalently in the Lagrangian
formulation by performing the Legendre transformation. The Lagrangian density of the
model is given by
L = L0 + Lint ,
where L0 is the free part

L0 = V i + EV (i) V + N i + EN (i) N
t
t

+ a i + E (i) a,
t
and the interaction is described by

Lint = g0 d D1 y V (x, t)N(x, t)f(x y)a(y, t) + H.C.
(3.10)
(3.11)
(3.12)
The Lagrangian formulation will be useful when we introduce an electric deformation of

the model.
The Lee model can be solved by considering directly the Schrdinger equation with
the Hamiltonian H = H0 + Hint , where H0 is given by (3.1) with the choice (3.2) and
Hint is given by (3.5). Due to the structure of the interaction (3.12), the only elementary
interaction of the theory involves the process
V N + .
(3.13)
In a standard relativistic model, the antiparticle would appear and the crossed reaction
V + N
(3.14)
231
would be possible, but this is not allowed in the Lee model due to the particular form of
the interaction Hamiltonian (3.5). The system possesses two simple conservation laws
nV + nN = constant,
nV + n = constant,
(3.15)
where nV , nN , n are the total numbers of V , N, particles, respectively. Due to the

conservation laws (3.15), the eigenfunctions of H contain only a finite number of particles
and, consequently, the theory is exactly solvable [33].
Renormalization
The quantization of the theory is straightforward. Locality in time allows us to perform
the standard canonical quantization of the theory. The nontrivial commutation relations of
the field operators are

(k), (k
) = (k k
),
N(p), N (p
) + = (p p
),

V (p), V (p
) + = (p p
),
(3.16)
with the rest equal to zero. The vacuum of the theory |0 is defined by
N(p)|0 = V (p)|0 = (p)|0 = 0.
(3.17)
It is easy to verify that

Hint (k)|0 = 0,
Hint N (p)|0 = 0;
(3.18)
thus we can take the and N -quanta as the physical particles (of masses and mN ,
respectively) and identify = 0 , mN0 = mN , and there is only the renormalization of the
mass of V to be considered.
Without any loss of generality we consider the dispersion relations (3.2) in order to
study the renormalization of the theory. Consider the sector of the theory associated with
(p). Due to the conservation
one physical V -particle. Denote the physical V -particle as |V
law (3.15), we have

V (p) = ZV V (p) |0 + d D1 k (k)N (p k) (k)|0
(3.19)
(p) is an eigenstate of H
with the wavefunction (k) still to be determined. Here |V

H
V (p) = mV
(3.20)
V (p) .
(p) yields
The normalization of |V

2

D1

k (k) .
1 = ZV 1 + d
Contracting (3.20) with 0|V (p
), one obtains
D1
k
g0
d
f (k)(k) = mV .
mV 0 +
(D1)/2
(2)
2k
(3.21)
(3.22)
232
On the other hand, contracting (3.20) with 0|N(q)(k), one obtains

(mV mN k )(k) =
g0
(2)(D1)/2
f (k)
,
2k
(3.23)
which gives
f (k)
, for mV < mN + ,
(2)
2k (mV mN k )
g0
f (k)
, for mV > mN + .
(k) =
P
(2)(D1)/2
2k (mV mN k )
g0
(k) =
(D1)/2
(3.24)
(3.25)
Note that Eq. (3.24) corresponds to the case when the V particle is stable; i.e., it cannot
spontaneously decay into an N and particle. The decay of the V particle is allowed in the
case of Eq. (3.25). The renormalized coupling can be obtained by requiring the scattering
process
N + N +
(3.26)
to be nonzero in the limit f 1.

As a result, we obtain the following renormalization constants
ZV1 = 1 +
g02
(2)D1
mV = mV 0 +
g 2 = g02 ZV .
g02
(2)D1
|f (k)|2
d D1 k
,
2k (mV mN k )2
D1
k
|f (k)|2
d
,
2k (mV mN k )
(3.27)
(3.28)
(3.29)
The integrals in (3.27) and (3.28) are generally divergent in the limit f 1. As usual, all
the scattering amplitudes (N N, V V , V N, N N, etc.) become
finite after we have performed the renormalization (3.27), (3.28) and (3.29).8
We would like to add a couple of comments.
(i) One can perform a path integral quantization of the theory and one obtains the
Feynman rules given in Fig. 2. Using these Feynman rules, it is straightforward to show
that the above results for the renormalization can also be obtained in the Lagrangian
framework and are exact in perturbation theory. Later we will use these Feynman rules
to study the noncommutative Lee model, particularly, in the time-noncommuting case.
(ii) In the original Lee model [33], D = 4 and the mass renormalization constant is linearly
divergent while the wavefunction renormalization is logarithmically divergent. It has
been shown that the different choices (3.3) (Galilean kinematics) and (3.4) (relativistic
kinematics) of dispersion relations lead to finite renormalizations when f 1.
8 For example, in the sector V N , the renormalized scattering amplitudes V V , V N and
N N were studied in [42] and [43].
233
Fig. 2. Feynman rules for the Lee model.
Unitarity and the ghost state

The relation (3.29) between the renormalized coupling g and the bare coupling g0 can
be rewritten as (with f set to 1)

d D1 k
g2
1
,
where
I
> 0.
g02 =
(3.30)
2
D1
1g I
(2)
2k (mV mN k )2
For D = 4, I is logarithmically divergent. If g is to remain fixed and nonvanishing, the
bare coupling has to be imaginary
g0 = i1
(3.31)
and the wavefunction renormalization

ZV = 1 g 2 I .
(3.32)
This contradicts the interpretation of ZV as the probability of finding a bare V quantum

in the physical V -particle state. Such negative probabilities imply that the S-matrix is not
unitary. In fact one can show that [34] ZV < 0 corresponds to a new state in the theory. This
state |G has a negative norm and is referred to as the ghost state by Kallen and Pauli.
As a result, the S-matrix is explicitly nonunitary. In fact, the nonunitarity of the theory is
related to the original Hamiltonian being nonhermitian due to the presence of an imaginary
bare coupling.
Two improvements of the original Lee model are possible. One is to consider other
dispersion relations, e.g., (3.3) and (3.4). This leads to 4-dimensional theory with finite
renormalizations and without a ghost [44,45]. Another possibility is to consider the Lee
234
model in lower dimensions [46]. In D = 3, the integral

I in (3.30) is finite and so the
model is ghost free for physical coupling 0 < g < 1/ I . The improved Lee model is still
exactly solvable in both cases. To minimize the number of new formulae, we consider the
second class of models when we generalize to the noncommutative case.
3.2. The noncommutative Lee model
The noncommutative framework is generated by using the -product (1.5). As
mentioned in the introduction, the noncommutative deformation can be introduced either
in the Hamiltonian or the Lagrangian formulation in the magnetic case ( i0 = 0). The
replacement (1.5) amounts to the following substitution in the formula (3.5)
i
f (k) f (k, p) := f (k) e 2 pi
ij k
j
(3.33)
In the electric case with nonvanishing components i0 = 0,9 the substitution takes the form
i 0i
(p
f (k) f (k, p) := f (k) e 2
0 ki pi k0 )
(3.34)
Obviously the -product involves an infinite number of time derivatives. The nonlocalities
in time destroy not just the usefulness of the Hamiltonian formulation, but also the
standard way of relating the Lagrangian and the Hamiltonian description.10 We are thus
left only with the Lagrangian framework. For example, when there is only the nonvanishing
component 01 = = 0, one obtains the modification of the product of V and fields

i

V (x, t) a(y, t) = e 2 t y1 x1 t
V (x, t)a(y, t
)
t =t

i
i
a y, t +
= V x, t
(3.35)
2 y1
2 x1
in the interaction Lagrangian (3.12). Note that due to the associativity of the Lagrangian
and the integration over spacetime, the -product of the three fields in (3.12) can be
represented by a modification of the product for any pair of fields (V as in (3.35), V N or
N).
Note also that the phase factor in (3.33) and (3.34) does not lead to a real factor as in the
noncommutative scalar 3 case. Thus the noncommutative modification in the Lee model
involves a complex factor. This, in particular, implies that the symmetric condition (2.12)
is not satisfied.
Quantization of the magnetically deformed theory can be achieved by using either
the canonical quantization, or equivalently a path integral quantization. In the electric
case, canonical quantization fails due to the nonlocality in time. Nevertheless, formally,
the theory can be quantized using the path integral method. In the following, we will
use the path integral method to analyze both the magnetic and the electric Lee models.
9 Besides magnetic and electric cases one can also consider lightlike deformations [47], corresponding to the
case = 0.
10 For recent efforts at introducing a Hamiltonian framework for Lagrangian densities nonlocal in time see
[4850]. We have not been able to employ here these results in a constructive way.
235
The Feynman rules are those of Fig. 1 with f (k, p) given by (3.33) and (3.34) and work for
general D. To be specific, below we consider the noncommutative Lee model in (D 4)dimensional spacetime and with the standard dispersion relations (3.2).
Renormalization and (no) IR/UV mixing
Since the effect of noncommutativity is a modification (3.33) or (3.34) of f by
a phase factor, it is clear that the mass, wavefunction and coupling renormalization
(depending on |f |2 ) are not affected. Thus we conclude that the renormalization constants
of the noncommutative Lee model are exactly computable and are independent of the
noncommutativity parameter .
Moreover, one can easily convince oneself that the UV-divergences of the theory
reside in planar diagrams that simply do not have nonplanar counterparts. Thus, the UVdivergences of the noncommutative Lee model remain untouched in the limit when the
cutoff is removed. This is quite different from the other noncommutative field theories
which display an intriguing mixing of IR/UV [16]. In these models, the introduction of a
nonzero noncommutativity improves the UV convergence of nonplanar diagrams but also
leads to new IR singularities for these diagrams. In the present case of the noncommutative
Lee model, there simply are no UV-divergences in the nonplanar diagrams, and hence
there are also no new IR singularities that could be generated. We conclude that the
noncommutative Lee model is free from IR/UV mixing. This result is exact.
Unitarity
First we consider the unitarity constraints at the one loop level. Due to the structure of
the vertices (Fig. 2) in the theory, it is easy to convince oneself that only planar diagrams
can be drawn at the one loop level. Therefore, the one loop Feynman amplitudes take the
form
(0) iab
Mab = Mab
e ,
(3.36)
(0)
where Mab
are the corresponding amplitudes in the commutative case, and eiab is the
Moyal phase factor associated with the planar diagram. As a result, Eq. (2.10) is satisfied
since

(0)
(0) (0)
Disc Mab = eiab Disc Mab
(3.37)
= ieiab
Man
Mnb = i
Man Mnb .
n
In the second step, we have used the fact that the constraint (2.10) is satisfied for
the commutative Lee model since this model is unitary (or one can verify this in a
straightforward manner since the Ms that appear in the sum are tree level ones). In the
last step we have used the fact that the planar Moyal phase factor of the 1-loop diagram
decomposes simply into the product of factors of the tree level ones:
eiab = eian einb .
(3.38)
Note that due to the form of the modification for the one-loop amplitude (3.36), checking
the imaginary part (2.11) would lead to the incorrect conclusion that the noncommutative
Lee model is not unitary at a one loop level. Note also that the above argument is general
236
Fig. 3. A nonplanar diagram.
and does not depends on whether is spacelike or timelike. Therefore, we conclude that
the noncommutative Lee model is unitary at a one loop level for general . This result is
valid to all orders in .
At a higher loop level, one can have nonplanar diagrams, for example, the one in Fig. 3.
The phase factor associated with this diagram is
i
e 2 (p1 p2 p1 p3 p2 p3 ) eik(p2 p3 ) .
(3.39)
The second phase factor depends on the loop momentum and is a characterization of a
nonplanar diagram. As one can check easily, this amplitude is regular in the variable s (and
hence ). Generally, due to the absence of the IR/UV singularity, a nonplanar amplitude
will be regular in the variable s and so there is no new discontinuity in the LHS of the
unitarity equation (2.10). Since both the LHS and RHS are regular in , the unitarity
constraint will be satisfied at the zeroth order in . Although we believe this to be the
case, it may not be easy to verify the unitarity relations to all orders in as one would
have to exploit various nontrivial relations among special functions and integrals. The fact
that unitarity constraints are satisfied at a one loop level; and also (at the zeroth order
in ) for any higher loop amplitude, is already a nontrivial property of the noncommutative
Lee model. Without any other source of violation of unitarity in sight, we expect that the
noncommutative Lee model is unitary for any .
4. Discussion
In this paper, we have discussed and examined two basic aspects of noncommutative
field theories: the IR/UV mixing and unitarity. We have showed that the S-matrix of a
noncommutative field theory is hermitian analytic. This implies that unitarity provides
a direct evaluation of the discontinuities associated with the cuts of normal thresholds.
We have also explained how the IR/UV singularities can lead to a violation of unitarity
for field theories with spacetime noncommutativities. As a corollary, we have argued
237
that a noncommutative field theory without any IR/UV mixing will be unitary in both the
spacespace and spacetime noncommutative cases.
As an illustration of the general discussion, we have introduced and analysed the
noncommutative Lee model. We have found that the model is entirely free from the IR/UV
mixing. This result is exact. Our general arguments show that the noncommutative Lee
model is unitary in both the spacespace and spacetime noncommutative cases. Simple
explicit checks are consistent with this claim. Thus we provide a counter example to the
general belief that field theories with spacetime noncommutativity have to be nonunitary.
A consistent quantum field theory on a noncommutative spacetime should be unitary. It
should also be free from the problems related to the IR/UV mixing. One can broadly divide
the IR/UV mixing phenomena in noncommutative field theories into those that could be
called good ones and bad ones. For example, the IR/UV singularities which appear in a
purely bosonic noncommutative gauge theory or in a noncommutative QED are bad ones
[51]. However, IR/UV singularities are milder and may be absent [52] in the presence of
supersymmetry. The milder form of the IR/UV mixing in supersymmetric noncommutative
gauge theories leads to a decoupling of the U (1) degrees of freedom in the IR [53,54]. Not
only the U (1) degrees of freedom become free in the IR [54], they also trigger spontaneous
supersymmetry breaking [55] in the presence of an appropriate FayetIliopoulos D-term
and play the role of the hidden sector. This we refer to as good IR/UV mixing effects.
More details are provided in [56]. With unitarity better understood and (some) IR/UV
mixing turned to be our advantage, it seems not unreasonable to contemplate that nature
could indeed be noncommutative (at least at some level of explanation of its phenomena).
Acknowledgements
C.S.C. and J.L. would like to thank Luis Alvarez-Gaum for useful discussions and
comments and the theory group at CERN, where this work was started, for its hospitality.
We would also like to thank David Fairlie, Pei-Ming Ho, Valya Khoze, Rodolfo Russo,
Lenny Susskind, Richard Szabo and Gabriele Travaglini for helpful discussions.
References
[1] M.R. Douglas, N.A. Nekrasov, Noncommutative field theory, hep-th/0106048.
[2] R.J. Szabo, Quantum field theory on noncommutative spaces, hep-th/0109162.
[3] S. Doplicher, K. Fredenhagen, J.E. Roberts, The Quantum structure of spacetime at the Planck scale and
quantum fields, Commun. Math. Phys. 172 (1995) 187.
[4] L.J. Garay, Quantum gravity and minimum length, Int. J. Mod. Phys. A 10 (1995) 145, gr-qc/9403008.
[5] A. Connes, M.R. Douglas, A. Schwarz, Noncommutative geometry and matrix theory: compactification on
tori, JHEP 9802 (1998) 003, hep-th/9711162.
[6] M.R. Douglas, C.M. Hull, D-branes and the noncommutative torus, JHEP 9802 (1998) 008, hep-th/9711165.
[7] C.S. Chu, P.M. Ho, Noncommutative open string and D-brane, Nucl. Phys. B 550 (1999) 151, hepth/9812219;
C.S. Chu, P.M. Ho, Constrained quantization of open string in background B field and noncommutative
D-brane, Nucl. Phys. B 568 (2000) 447, hep-th/9906192.
[8] V. Schomerus, D-branes and deformation quantization, JHEP 9906 (1999) 030, hep-th/9903205.
238
[9] N. Seiberg, E. Witten, String theory and noncommutative geometry, JHEP 9909 (1999) 032, hep-th/9908142.
[10] J. Madore, The fuzzy sphere, Class. Quantum Grav. 9 (1992) 69.
[11] H. Grosse, C. Klimcik, P. Presnajder, Towards finite quantum field theory in noncommutative geometry, Int.
J. Theor. Phys. 35 (1996) 231, hep-th/9505175.
[12] C.S. Chu, J. Madore, H. Steinacker, Scaling limits of the fuzzy sphere at one loop, JHEP 0108 (2001) 038,
hep-th/0106205.
[13] S. Zakrzewski, J. Phys. A 27 (1994) 2075.
[14] S. Majid, H. Ruegg, Bicrossproduct structure of kappa Poincar group and noncommutative geometry, Phys.
Lett. B 334 (1994) 348, hep-th/9405107.
[15] J. Lukierski, H. Ruegg, W.J. Zakrzewski, Classical quantum mechanics of free kappa relativistic systems,
Ann. Phys. 243 (1995) 90, hep-th/9312153.
[16] S. Minwalla, M. Van Raamsdonk, N. Seiberg, Noncommutative perturbative dynamics, JHEP 0002 (2000)
020, hep-th/9912072.
[17] L. Griguolo, M. Pietroni, Wilsonian renormalization group and the noncommutative IR/UV connection,
JHEP 0105 (2001) 032, hep-th/0104217.
[18] A. Bilal, C.S. Chu, R. Russo, String theory and noncommutative field theories at one loop, Nucl. Phys. B 582
(2000) 65, hep-th/0003180;
C.S. Chu, R. Russo, S. Sciuto, Multiloop string amplitudes with B-field and noncommutative QFT, Nucl.
Phys. B 585 (2000) 193, hep-th/0004183.
[19] O. Andreev, H. Dorn, Diagrams of noncommutative Phi**3 theory from string theory, Nucl. Phys. B 583
(2000) 145, hep-th/0003113.
[20] Y. Kiem, S.M. Lee, Nucl. Phys. B 586 (2000) 303, hep-th/0003145.
[21] J. Gomis, M. Kleban, T. Mehen, M. Rangamani, S.H. Shenker, JHEP 0008 (2000) 011, hep-th/0003215.
[22] H. Liu, J. Michelson, Phys. Rev. D 62 (2000) 066003, hep-th/0004013.
[23] 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.
[24] R. Gopakumar, J. Maldacena, S. Minwalla, A. Strominger, S-duality and noncommutative gauge theory,
JHEP 0006 (2000) 036, hep-th/0005048.
[25] J.L. Barbon, E. Rabinovici, Stringy fuzziness as the custodian of timespace noncommutativity, Phys. Lett.
B 486 (2000) 202, hep-th/0005073.
[26] J. Gomis, T. Mehen, Spacetime noncommutative field theories and unitarity, Nucl. Phys. B 591 (2000) 265,
hep-th/0005129.
[27] L. Alvarez-Gaum, J.L. Barbon, R. Zwicky, Remarks on timespace noncommutative field theories,
JHEP 0105 (2001) 057, hep-th/0103069.
[28] T. Mateos, A. Moreno, A note on unitarity of non-relativistic non-commutative theories, Phys. Rev. D 64
(2001) 047703, hep-th/0104167.
[29] A. Bassetto, L. Griguolo, G. Nardelli, F. Vian, On the unitarity of quantum gauge theories on noncommutative spaces, JHEP 0107 (2001) 008, hep-th/0105257.
[30] T.C. Cheng, P.M. Ho, M.C. Yeh, Perturbative approach to higher derivative and nonlocal theories, hepth/0111160.
[31] D.I. Olive, Unitarity and evaluation of discontinuities, Nuovo Cimento 26 (1962) 73.
[32] R.E. Cutkosky, Singularities and discontinuities of Feynman amplitudes, J. Math. Phys. 1 (1960) 429.
[33] T.D. Lee, Some special examples in renormalizable field theory, Phys. Rev. 95 (1954) 1329.
[34] G. Kallen, W. Pauli, Kgl. Danske Videnskab. Selskab, Mat.-fys. Medd. 30 (7) (1955).
[35] K.W. Ford, Problems of ghost states in field theories, Phys. Rev. 105 (1957) 320.
[36] D.I. Olive, Exploration of S-matrix theory, Phys. Rev. B 135 (1964) 745.
[37] R.J. Eden, P.V. Landshoff, D.I. Olive, J.C. Polkinghorne, The Analytic S-Matrix, Cambridge Univ. Press,
1966.
[38] G.F. Chew, The Analytic S Matrix, Benjamin, 1966.
[39] S.S. Schweber, Relativistic Quantum Field Theory, Harper & Row, 1961.
[40] A.A. Bichl, J.M. Grimstrup, V. Putz, M. Schweda, Perturbative ChernSimons theory on noncommutative
R**3, JHEP 0007 (2000) 046, hep-th/0004071.
[41] A.K. Das, M.M. Sheikh-Jabbari, Absence of higher order corrections to noncommutative ChernSimons
coupling, JHEP 0106 (2001) 028, hep-th/0103139.
239
[42] D. Amado, Phys. Rev. 122 (1961) 696.

[43] M.G. Fuda, Phys. Rev. C 25 (1982) 1972;
M.G. Fuda, Phys. Rev. C 26 (1982) 204;
M.G. Fuda, Phys. Rev. C 29 (1984) 1222.
[44] J.M. Levy-Leblond, Galilean quantum field theories and a ghostless Lee model, Commun. Math. Phys. 4
(1967) 157.
[45] F.J. Yndurain, S-matrix formalism, charge renormalization, and the definition of the Hamiltonian in a simple
field-theoretic model, J. Math. Phys. 7 (1966) 1133.
[46] C.M. Bender, C. Nash, Asymptotic freedom and the Lee model, Phys. Rev. D 10 (1974) 1753.
[47] O. Aharony, J. Gomis, T. Mehen, On theories with light-like noncommutativity, JHEP 0009 (2000) 023,
hep-th/0006236.
[48] J. Llosa, J. Vives, J. Math. Phys. 35 (1994) 2856.
[49] R.P. Woodard, A canonical formalism for Lagrangians with nonlocality of finite extent, Phys. Rev. A 62
(2000) 052105, hep-th/0006207.
[50] J. Gomis, K. Kamimura, J. Llosa, Hamiltonian formalism for spacetime non-commutative theories, Phys.
Rev. D 63 (2001) 045003, hep-th/0006235.
[51] A. Matusis, L. Susskind, N. Toumbas, The IR/UV connection in the non-commutative gauge theories,
JHEP 0012 (2000) 002, hep-th/0002075.
[52] V.V. Khoze, G. Travaglini, Wilsonian effective actions and the IR/UV mixing in noncommutative gauge
theories, JHEP 0101 (2001) 026, hep-th/0011218.
[53] A. Armoni, R. Minasian, S. Theisen, On non-commutative N = 2 super-YangMills, Phys. Lett. B 513
(2001) 406, hep-th/0102007.
[54] T.J. Hollowood, V.V. Khoze, G. Travaglini, Exact results in noncommutative N = 2 supersymmetric gauge
[55] C.S. Chu, V.V. Khoze, G. Travaglini, Dynamical breaking of supersymmetry in noncommutative gauge
theories, Phys. Lett. B 513 (2001) 200, hep-th/0105187.
[56] C.S. Chu, V.V. Khoze, G. Travaglini, Noncommutativity and model building, hep-th/0112139.

UV/IR mixing via closed strings and tachyonic

instabilities
Adi Armoni a , Esperanza Lopez b
a Theory Division, CERN, CH-1211 Geneva 23, Switzerland
b Max-Planck-Institut fr Gravitationsphysik, Albert-Einstein-Institut, Am Mhlenberg 1,
D-14476 Golm, Germany

Received 23 October 2001; accepted 12 April 2002
Abstract
We discuss UV/IR mixing effects in non-supersymmetric non-commutative U (N) gauge theories.
We show that the singular (non-planar) terms in the 2- and 3-point functions, namely the poles and
the logarithms, can be obtained from a manifestly gauge invariant effective action. The action, which
involves open Wilson line operators, can be derived from closed strings exchange between two stacks
of D-branes. Our concrete example is type 0B string theory and the field theory that lives on a
collection of N electric D3-branes. We show that one of the closed string modes that couple to
the field theory operator which is responsible for the infrared poles, is the type 0 tachyon. 2002
Elsevier Science B.V. All rights reserved.
1. Introduction
Non-commutative gauge theories attracted recently a lot of attention, mainly due to the
discovery of their relation to string/M theory [1,2]. The perturbative dynamics of these
theories is very interesting: planar graphs of non-commutative theories are exactly the
same as the planar graphs of ordinary theories apart from global phases which depend
on external momenta [3]. Non-planar graphs, on the other hand, are regulated by the
non-commutativity parameter and they are therefore UV-finite. This regularization is,
however, only effective when there is a non-zero momentum inflow into the graph. In
particular, as a result of this, the non-planar contribution to the propagator contains, usually,
a pole 1/(p)2 . This pole, which originates from the high momentum region of the integral
E-mail addresses: adi.armoni@cern.ch (A. Armoni), lopez@aei-potsdam.mpg.de (E. Lopez).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 2 9 0 - 0
A. Armoni, E. Lopez / Nuclear Physics B 632 (2002) 240256
241
(UV) seems to affect the large distance dynamics. This unusual phenomenon is called
UV/IR mixing [4]. In supersymmetric theories this pole cancels and a softer version of
UV/IR mixing exists due to a logarithmic contribution [5]. Aspects of the UV/IR mixing
phenomena in scalar theories [611] as well as gauge theories [1222] were studied by
many authors over the past two years. See [23,24] for comprehensive reviews.
In this work we would like to focus on non-supersymmetric non-commutative gauge
theories in 4 dimensions. The non-planar pole modifies the dispersion relation of the photon
as follows [5,20,21]
2
adj
1
adj g
,
E 2 = p2 NB NF
2
(p)2
adj
(1)
adj
where NB and NF are the numbers of bosons and fermions in the adjoint representation,
adj
adj
respectively. In the case of pure YangMills theory, or in general when NB > NF , the low
momentum end of the spectrum acquires imaginary energy. Namely, the one loop analysis
suggests that the theory suffers from an instability. In [20] it was shown that the quadratic
pole-like infrared divergence is gauge-fixing-independent. In [21] non-commutative N = 4
YangMills at finite temperature was considered. This theory presents a regularized version
of UV/IR mixing, where the temperature acts as a UV cut-off. Although perturbation theory
seems to be under better control in this case, tachyonicexcitations at long wavelength
appear for T bigger than a critical temperature Tc 1/ g . All this suggests that the
associated instability might not be an artefact of perturbation theory, but instead its source
should be searched in an expansion around the wrong vacuum. Of course, the existence of
a stable vacuum is an open question.
As already mentioned, apart from the tachyonic poles in the non-planar graphs, these
graphs contain also subleading logarithmic contributions. These contributions do not
change the photon dispersion relation and are not expected to cause any instabilities, but
they are relevant in the IR. In particular, they affect the effective coupling [18].
The purpose of this note is to gain a better understanding of the tachyonic poles and
the logarithmic effects by using string theory. We summarize briefly our main results. We
consider a field theory that lives on D3 electric branes of the non-supersymmetric type 0B
string theory. We show that all singular amplitudes involving pole-like infrared divergences
can be encoded in a rather simple gauge-invariant effective action

adj
d 4p
m2
adj
I
Seff
(2)
NB NF
tr W (p) tr W (p) 2 K2 (mp),
4
(2)
p
where W (p) denotes the open Wilson loop operator [25,26]. This effective action is
structurally analogous to a closed string exchange between two stacks of D-branes. This
coincidence is more than formal since we will see that the Bessel function kernel in (2)
can be directly related to a closed string propagator in type 0 string theory. Although (2)
has the form of a closed string exchange, all the tower of closed string modes contribute to
it, similarly to the proposal by [6]. Among the closed string modes that couple to the open
Wilson line operator is the type 0 closed string tachyon. This is in contrast to the ordinary
commutative case, where the tachyon just contributes to the vacuum energy of the field
theory. The fact that the closed string tachyon couples to a non-trivial operator in the field
242
theory, which in addition is responsible of the pole-like infrared divergences, suggests that
there could be a relation between string and non-commutative instabilities.
Although at a more intuitive level, our analysis can be extended to the logarithmic
infrared-divergent terms. They admit the following gauge-invariant completion

d 4p
II
tr F W (p) tr F W (p)
Seff 0
(2)4

+ tr D i W (p) tr D i W (p) K0 (mp).
(3)
II is presented in Section 4.
This expression is only schematic; a precise definition of Seff
From a string point of view, we can interpret (3) as due to the exchange of massive 2-form
closed strings. Note that this sector of the closed string does not contain tachyons. Indeed,
as already mentioned, there is no tachyonic instability associated with the logarithmic part
of the action.
The organization of this article is as follows: in Section 2 we describe our model and we
calculate the various singular amplitudes. Sections 3 is devoted to a derivation of the full
gauge invariant effective action related to the infrared pole-like terms in the case of pure
U (1) theory. In Section 4 we suggest a derivation of the effective action via closed strings
exchange between D3-branes of type 0 string theory. We discuss our results in Section 5.
We use the following notations and conventions. The field theory under consideration
is a 4d one with spacespace non-commutativity [x 1 , x 2 ] = i (or a tensor with nonvanishing components in the 1, 2 directions). We also use the notation p = p . The
U (N) generators are normalized such that tr t A t B = 12 AB and, in particular, the U (1)
generator is t 0 = 1/ 2N .
2. Field theory calculationsvarious non-planar amplitudes

In this section we describe the UV/IR mixing effects in a concrete model. The 4d field
theory under consideration is the theory that lives on a stack of N coincident electric
D-branes of type 0B string theory. It is obtained by dimensional reduction of pure (nonsupersymmetric) 10d non-commutative YangMills theory. The model contains a vector
and 6 adjoint scalars and it is described by the following action

1
S = tr d 4 x 2 F ! F + D i ! D i ,
(4)
2g
where
and
F = A A i(A ! A A ! A )
(5)

D i = i i A ! i i ! A
(6)
for i = 1, . . . , 6. The model is invariant under the following non-commutative U (N) gauge
transformation
A = i(A ! ! A )
(7)

i = i i ! ! i .
243
(8)
Let us focus on the one-loop structure of the theory. The planar sector is well
understood. Apart from global phases associated to external legs, the various amplitudes
are the same as in the commutative cousin of the theory [3]. In particular the theory is
one-loop renormalizable with the same counterterms as those of the commutative theory
[14].
The non-planar sector of the theory exhibits an interesting pattern. In this case the Moyal
phases associated with the vertices do not cancel and lead to a UV-finite result [4]. Consider
first the propagator of the gluon. The only non-vanishing non-planar graph exists when the
external gluons are in the U (1) [4,14]. The sum of the various contribution, due to gluons,
ghosts and scalars running in the loops yields [5,1214]

d 4 q (2q q g q 2 )
64g 2 N p p
exp(2i pq)
=
.
A(11) = 8g 2 N
(9)
4
4
(2)
q
(4)2 p 4
A similar calculation yields a similar result for the scalar propagator

d 4q 1
32g 2 N 1
A(11) = 8g 2 N
exp(2i pq)
=
.
4
2
(2) q
(4)2 p 2
(10)
Note that there is a relative factor of 2 between (9) and (10).

The poles in (9) and (10) signal the interesting UV/IR mixing that is typical of noncommutative theories. The origin of these contributions is the UV regime and they seem to
affect the IR of the theory. These poles create a potential problem in the renormalization
process, since when the non-planar graphs are inserted in higher loop diagrams they create
new divergences. It was suggested [4] that, in certain cases, the sum of the geometric series
of these contributions can shift the pole such that these new divergences are avoided. This
procedure, however, cannot be implemented in the present case, due to the positive sign in
front of (9) and (10) [20,21]. In general, the coefficient in front of (9) and (10) is determined
by the number of bosons in the adjoint representation minus the number of fermions in the
adjoint representation [21]. In cases where there are more bosons than fermions in the
adjoint (such as the present case), a resummation of the series is impossible, since the
series does not converge. At present, there is no known procedure to cure this pathology.
In particular, the pure non-commutative YangMills theory seems to be sick.
Let us now proceed to the non-planar corrections to the 3-point vertices. The pattern
is similar: whereas planar graphs take the same form as in the commutative theory and
they are divergent, non-planar graphs are UV-finite, but singular at 0. Non-vanishing
diagrams involve U (1) factor in at least one of the external legs [14]. The amplitude in the
case of 3 external gluons, when all gluons are in the U (1) is

p 3 p 3 p3
p2 p2 p 2
i64g 3 N/2 p 1 p 1 p1
.
+
+
A(111) =
(11)
(4)2
p14
p 24
p34
We have ignored in this expression a factor cos p 1 p2 /2, which appears in previous
calculations of the leading IR contribution to the 3-point function. The reason is that,
in the approximation used to obtain (11), i.e., pi pj 1, we cannot distinguish between
cos p1 p2 /2 and 1. We keep this convention in the following. When one gluon is in the
244
U (1) and the two other gluons are in the SU(N) the amplitude takes the form
i64g 3 N/2 p 1 p 1 p1
A(1NN) =
,
(4)2
p14
(12)
where p 1 is the momentum of the U (1) field. Similarly, the amplitude for two external
scalars and one gluon, all in the U (1), is

p 3
p 2
i32g 3 N/2 p 1
A(111) =
(13)
+ 2 + 2 .
(4)2
p 12
p 2
p 3
In the case of two scalars and one gluon transforming in U (1) and SU(N) the amplitude is
the following
i32g 3 N/2 p 1
,
A(1NN) =
(14)
(4)2
p 12
where, again, p 1 is the momentum of the U (1) field.
The information about the various non-planar diagrams can be summarized in the
following effective action

2
p p
1
SI = g 2 d 4 p 2 4 tr A (p) tr A (p) + 2 tr i (p) tr i (p)
2
p
p

3
ig
+
d 4 p1 d 4 p2 d 4 p3 (p1 + p2 + p3 )
(2)4

p p p
2 1 14 1 tr A (p1 ) tr A (p2 )A (p3 )
p1
p
+ 12 tr A (p1 ) tr i (p2 ) i (p3 )
p1

i
i
+ 2 tr (p1 ) tr (p2 )A (p3 ) .

(15)
Apart from the terms which are summarized in the effective action (15), there are
other contributions which are less singular when 0. In contrast to the poles, these
terms (which as we shall see in a moment are log-like terms) do not cancel even in the
supersymmetric case, apart from the N = 4 SYM case [5]. These terms have a different
Lorentz structure than the poles and they are all proportional to the one-loop beta function
coefficient. Most of our analysis of this part is based on [16] and [18].
The gluon propagator (for the U (1) degrees of freedom) contains the following nonplanar contribution
M(11) =

26g 2 N 2
p g p p log m2 p 2 ,
3(4)2
(16)
where m2 is an IR cut-off. We can think about it as a mass term for the scalars (and vectors),
given via a Higgs mechanism. Similarly to the gluon, the correction to the scalar propagator
245
is
M(11) =
26g 2 N 2
p log m2 p 2 .
3(4)2
(17)
The subleading corrections to the 3-point vertices are the following: for 3 gluons, all in
the U (1), we have

i26g 3 N/2
1
M(111) =
(18)
sin p 1 p2 log m2 p 12 g p1 + perm. ,
3(4)2
2
where perm. means permutations of the three momenta and Lorentz indices due to the
symmetry of the amplitude. Similarly for the case of 1 gluon in the U (1) and 2 gluons are
in the SU(N)

i26g 3 N/2
1
M(1NN) =
(19)
sin p 1 p2 log m2 p12 g p1 + perm. ,
2
3(4)
2
where now p1 is the momentum of the U (1) gluon, and the permutations are with respect
to the 2 gluons in the SU(N).
In the case of amplitudes where there are two scalars and one gluon we have

2 2

i26g 3 N/2
1
sin
p
p
M(111) =
(20)
1 2 log m p 1 p1 + perm. ,
2
2
3(4)
and the same expression for the SU(N)SU(N)U (1) amplitude.
The log-like amplitudes can be summarized by the following effective action

24 2
2
SII =g
d 4 p p2 g p p log m2 p 2 tr A (p) tr A (p)

13

+ p2 log m2 p 2 tr i (p) tr i (p)

1
ig 3
4
4
4
d p1 d p2 d p3 (p1 + p2 + p3 ) sin p 1 p2 log m2 p 12
+
4
(2)
2

p1 tr A (p1 ) tr A (p2 )A (p3 )

(21)
+ tr i (p1 ) tr A (p2 ) i (p3 ) .
The actions (15), (21) are not gauge invariant. In order to have a (non-commutative)
gauge invariant action, higher order terms in A should be added. In the following sections
we will derive a manifestly gauge-invariant action which includes (15) and (21) as part of it.
3. The effective actionfield theory derivation

3.1. The poles
We will start by considering the pole-like IR-divergent contributions to the 2- and
3-point functions with only gluons as external legs. We observe that in both cases each
vector field A (pi ) is contracted with p = p , where p is the total momentum
246
flowing on each trace operator. This suggests that these terms are related to the simplest
gauge-invariant operators carrying non-zero momentum, the straight open Wilson line
defined by [25,26]

1
ipx
)
e .
W (p) = tr d 4 x P eig 0 d p A (x+p
(22)
Indeed, the gluon 2- and 3-point functions (15) can be obtained as the first terms in the
expansion of the following gauge-invariant expression

2 + Ns
I
=
(p),
d 4 pW (p)f (p)W
Seff
(23)
2 2
with f (p)
a function that tends in the IR to 1/p 4 ; Ns is the number of scalars in the adjoint
representation (Ns = 6 in the type 0 case). We denote by W (p) the Wilson loop operator
(22) once the O(g 0 ) term has been subtracted
W (p) = ig p tr A (p) g
lp
d 4 l sin 2
p p tr A (p l)A (l) + .
(2)4 lp
(24)
By inserting (24) in (23), we immediately recover the gluon 2-point function. The
expressions (11), (12) for the gluon 3-point function are valid in the limit p i pj 1. In
lp
that limit sin lp
2 / 2 1 and thus also the 3-point function is correctly obtained from (23).
This was to be expected since the IR divergent contribution to the 3-point function satisfies
the Ward identity [16].
For the pure U (1) non-commutative theory, (23) can be obtained from a direct
calculation of the 1-loop N -point functions. This will allow us to determine the function f
in (23). Due to the structure of the argument in the exponential of the Wilson loop, (23)
contributes to the N -point function with terms proportional to p 1 p N . The N -point
functions will have in general a complicated Lorentz index structure. However, it is easy
to isolate the terms of the mentioned form. They can only come from diagrams with
3-point vertices. Diagrams with 4-point vertices will give rise to a tensor structure
containing g i j , and therefore are not of the desired form. We will like to point out
that diagrams with 4-point vertices can produce as strong an IR divergence as those with
only 3-point vertices. Indeed, the tadpole induces a quadratic pole-like contribution to
the 2-point function of the form g /p 2 . However, the role of this term is to cancel a similar
contribution coming from diagrams with 3-point vertices, and which would otherwise
violate the Ward identity [5]. The same applies to the 3-point function. We will thus ignore
diagrams with 4-point vertices when analyzing the leading UV/IR mixing effects.
We will use the background field method in the following analysis; for the associated
Feynman rules see [16]. The diagrams we are interested in are those depicted in Fig. 1. We
have

N
d 4 l (2l + 2p1 + + 2pi1 + pi )i
N
(a) + (b) = (2ig)
(2)4
(l + p1 + + pi1 )2
i=1
sin
p i (l + p1 + + pi1 )
.
2
(25)
(a)
247
(b)
Fig. 1. Amplitudes containing terms p 1 p N . Wavy lines refer to gauge bosons and dotted lines to ghost.
The end points of the external lines are background vector fields B .
As explained, we will disregard those parts of (25) whose tensor structure is such that
they cannot contribute to (23). This allows us to discard all the terms in the numerator
proportional to external momenta, and keep only 2li for each i. Expression (25) then
reduces to

2i j<k p j pk k

l1 lN ei pl
d 4l
N
n
()
,
(2ig)
(26)
(2)4 l 2 (l + p1 )2 (l + p1 + + pN1 )2
where the summation

1, . . . , N , comes from expanding the sine. We
i =
on i ,ii = 1 for
1i
and
n
=
have defined p = i pi 1
i
2
2 . We can interpret the N vertices as twisted
or untwisted depending if = 1 or 1, respectively. Thus n is the number of twisted
vertices and p the total momentum flowing in the twisted vertices. The li in the numerator
can now be substituted by derivatives with respect to p acting on the exponential. In order
to simplify the analysis will we consider
p asi an independent variable, and only in the end
we will set p = p with p = i pi 1
2 . This allows us to bring the derivatives out
of the integral, and rewrite (26) as
(2ig) 1 N
N

i

()
2 j<k p j pk k
ei pl
d 4l
. (27)
(2)4 l 2 (l + p1 )2 (l + p1 + + pN1 )2
i
The integral appearing in this expression coincides with that of the N -point function of
a non-commutative 3 theory and has been calculated in [28,29] (see also [30] for a recent
two-loops analysis). There, a small mass m for the field was introduced as an ordinary
infrared regulator. The evaluation of the previous integral gives1
N2
p
Jn (p)
KN2 (mp)J
Nn (p),
(28)
m
where KN2 are modified Bessel functions. We have denoted by Jn (p) the kernel of
the n -product defined in [31], i.e., Jn (p) J (p
r(1), . . . , pr(n) ) where pr(j ) are
the n momenta entering the twisted vertices and p = pr(j ) . A comment is now in order.
1 This result is not affected by considering p and p as independent variables.
248
Expression (28) is not the complete answer, but the leading term in the infrared. Subleading
terms are suppressed by powers of p 2 pi pj and therefore they do not give rise to infrared
divergences for any N .2 In the following, we will keep in the evaluation of the N -point
functions only the infrared-leading term. Then, (27) reduces to
N2

1
p
N
n
(ig)
() Jn (p) 1 N
KN2 (mp)
JNn (p). (29)
m
2 2
Using the properties of the modified Bessel functions, it is easy to see that the term in
square brackets gives rise to a contribution of the form
m2
K2 (mp).
p 2
Adding up all such contributions to the effective action we get

N1

()n
1
I
N
4
Seff
=
(ig)
p
d
2
n!(N n)!
2
()N p 1 p N
N=2
m2
p 2
(30)
n=1

K2 (mp)
p 1 p N A1 An n (p) An+1 AN Nn (p).
(31)
This expression reproduces (23) by setting 2f (p)

=
m2
p 2
K2 (mp).
Although the n also
appear in the effective action of the non-commutative 3 theory, they only combine to
form the scalar analog of Wilson loop operators in the limit of large non-commutative
parameter [29]. On the contrary, the invariance of the effective action with respect to
gauge transformations of the background field suggests that, in gauge theories, Wilson loop
operators will play an important role for any value of . As a first example, a Wilson loop
completion of the non-planar contributions to the F 4 terms in N = 4 gauge theory has been
proposed in [32]. We have just seen that the puzzling pole-like divergent terms originating
from UV/IR mixing are part of the simplest gauge-invariant double-trace operator that
can appear in the effective action. We will leave for the next section the extension of the
previous considerations to gauge theories with adjoint matter.
3.2. The logarithms
We would like to comment on the IR logarithmic-divergent terms arising from
UV/IR mixing. As already mentioned, these subleading contributions occur also in the
supersymmetric case. We suggest here a gauge invariant completion of the IR logarithmic
divergent terms. This suggestion is not as rigorous as the derivation in the previous
subsection, but our result is fixed by the requirement of gauge invariance.
It was shown in [16] that the logarithmic singularities of the 2-, 3- and 4-point function
of pure NC U (1), in the limit |pi | |p i + p j | IR 0, combine into the following
2 It is interesting that the subleading terms do not seem to have such a simple expression in terms of
n
products as (28).
contribution to the effective action:

1
II
= 0 log( IR )2 d 4 x F F ,
Seff
4
249
(32)
with IR an infrared cut-off and 0 the coefficient of the 1-loop beta function. It is tempting
to propose the following gauge-invariant completion of (32), which generalizes to the
U (N) case

1
II
Seff = 0 d 4 p O (p)K0 (mp)O
(33)
(p),
4
where the operator O is defined by

1

ipx
)
e .
O (p) = tr d 4 x L F (x)eig 0 d p A (x+p
(34)
Following the notation of [32], L denotes integration of F along the open Wilson line
together with path ordering with respect of the -product of all terms inside the parenthesis.
The action (33) reproduces the pure gluonic log-like N -point functions (21) in the small m
limit.
4. The effective action via closed strings exchange

The recent interest in the study of non-commutative field theories has been mainly
motivated by their connection to string theory. The world-volume coordinates of Dbranes in the presence of a constant B-field background turn out to satisfy the relation
[x , x ] = i , with 1/B . As a consequence, the low energy theory on the brane
is a non-commutative gauge theory. In this section we would like to analyze (23) and (33)
(or (15) and (21)) from a string-inspired point of view. In a series of recent papers it has
been shown that closed string modes couple to non-commutative D-branes through open
Wilson line operators [3235]. This result was obtained by evaluating the disk amplitude
between a closed string and open string modes.
Fig. 2. The annulus amplitude.
250
Let us consider the annulus diagram with boundaries on non-commutative D-branes as

in Fig. 2. It can be seen as a loop of open strings or a tree level exchange of closed strings.
In the limit of a large cylinder the closed string channel picture is more adequate since the
annulus diagram factorizes to closed string insertions on a disk connected by a closed string
propagator [36]. In the opposite limit of a small cylinder, the exchange of the lowest open
string modes dominates. This provides the field theory limit, and the annulus amplitude
reproduces the 1-loop field theory effective action. Thus in general we could expect in
the field theory effective action more complicated contributions than (23) and (33), which
structurally are reminiscent of a closed string exchange. Notice that a similar structure was
proposed as the gauge-invariant completion of the non-planar F 4 terms in the effective
action of N = 4 non-commutative YangMills [31,32]. In that case, the function f had the
interpretation of a closed string propagator in type II string theory. This, however, comes
as no surprise since the F 4 terms in the maximally supersymmetric case are protected
by non-renormalization theorems [36]. In contrast, it is remarkable that (23) emerges in
a non-supersymmetric theory. We will show below that the function f appearing in the
IR-divergent terms can also be directly related to a closed string propagator.
In the rest of this section we will consider type 0 string theory. This theory can be
obtained as a world-sheet orbifold of type II, which projects out spacetime fermions. It
contains a closed string tachyon arising from the twisted sectors. There are, however, no
open string tachyons on D-branes in type 0 theory. This makes it especially adequate for
our considerations. We will work with the gauge theory on N electric D3-branes. It is given
by the dimensional reduction of pure YangMills in 10 dimensions, i.e., gauge fields plus
6 scalars in the adjoint representation.
We start by analysing which closed string modes couple to the open Wilson line operator
(22). The first candidate is the type 0 tachyon. In the absence of B-field and at leading order
in , it couples to the brane tension as [37,38]
N
(35)
.
4(2 )2
Following the same analysis done for bosonic and type II string theory [32,34], it is easy
to see that the trivial field theory operator (35) gets promoted to an open Wilson loop in
the presence of a B-field. The coupling of the type 0 tachyon to the D-brane field theory at
leading order is described by

10
d 10 P
detG T (P )O(P ),
SI = 2
(36)
(2)10
gYM
where G is the open string metric and

1
O(P ) =
(37)
tr
d 4 x W (x, C) eipx .
4(2 )2
We denote by PM the 10-dimensional momentum, p the momentum along the
4-dimensional world-volume of the D3-brane and p i the momentum in the transverse
directions. In the previous expressions W (x, C) is a generalization of (22) which involves
the transverse scalars
1
)+yi i (x+p
)
,
W (x, C) = P eig 0 d p A (x+p
(38)
251
where we have defined yi = 2 p i and i = Xi /2 for i = 1, . . . , 6, which provides

the correct normalization for the field theory scalar fields.
The on-shell condition for the type 0 tachyon is PM g MN PN = 2/ , with g the closed
string metric. Closed and open string metrics are related by g 1 = G1 G/(2 )2
[2]. In the SeibergWitten limit, i.e., 0 keeping G and fixed, the on-shell condition
becomes [32,34]
p 2 + y 2 = 8 2 .
(39)
The closed string mass is a subleading effect with respect to the momentum in the noncommutative directions in the SeibergWitten limit. In spite of that, it will be crucial in the
following to keep its contribution to the mass-shell condition. We want to analyse how the
tachyon exchange contributes to the annulus amplitude. For two D3-branes separated by
a distance r we obtain

2
eip r
d 10 P det G
I
O(P
)O(P
)
= 410
.
Seff
(40)
M2
2
(2)10 det g
gYM
+
p

2
(2 )
The quantity M 2 /(2 )2 is the effective mass of the closed string tachyon propagating in
the six transverse dimensions; from (39) M 2 = p 2 8 2 .
In order to make contact with the previous section we will first consider the dependence
1
of O on the gauge fields only. Then O = (4
)2 W (p), with W (p) given by (22). Using
(see, for example, [32])
2
10
det g
4
(41)
=
(2
)
,
4
gYM
det G
and defining m = r/2 , the previous expression can be rewritten as
d 4p
I
Seff =
det G tr W (p) tr W (p)G(p),
(4 )4
(2)4
where
G(p) =
d 6y
eiym
1 M2
=
K2 (mM).
6
2
2
(2) M + y
(2)3 m2
(42)
(43)
G(p) represents the closed string propagator in the transverse dimensions, rescaled
appropriately to the field theory limit. Indeed, it is finite in the limit 0. However,
(42) diverges in this limit due to the O( 2 ) dependence of the brane tension to which the
tachyon couples. We can define a finite contribution to (42) by expanding G(p) to O( 4 ),
using the explicit dependence of M 2 on . We then obtain a contribution to the field theory
effective action of the form (23), with

m2
f (p) G(p) 4 = c 2 K2 (mp),

p
(44)
with c = 4 5 /3. This agrees with the result derived from the field theory calculation, up to
a global coefficient. We will comment on this below.
252
Notice that in order to obtain the IR divergent terms from the string exchange, it was
essential that the field theory operator that couples to the tachyon carries negative powers
of . The reason for this is that at O( 0 ), G 1/m4 as p 0. Such a term is related
to the ordinary infrared problems of a field theory with massless degrees of freedom.
However, remarkably, G(p) contains information about the new divergences due to UV/IR
mixing effects in non-commutative field theories when expanded to higher orders in .
We have analyzed above the coupling of the type 0 tachyon to the D-brane field theory at
leading order in . At O( 0 ) it couples to the field theory operators tr F 2 and (D i )2
[37,38]. For the reasons just exposed, the coupling of the tachyon to these operators would
contribute non-singular terms in the effective action and therefore we will not consider
them.
Eq. (42) differs from (23), as we subtracted the 1 from the open Wilson line in (23).
The 1 in coordinate space is in fact (p), as we work in Fourier space, and therefore
this difference affects only the p = 0 component of W . We would like to stress that the
expansion of G(p) in powers requires that p is non-zero. At p = 0 and in the limit
0, the string propagator is G 1/m4 . The associated contribution to the effective
action is proportional to

1
(4)
4
(45)
(0)
d 4 x,
( m)4
where 1/( m) can be interpreted as a field theory scale. Therefore, the difference
between (42) and (23) reflects the vacuum energy of the gauge theory, which is taken into
account in the string theory calculation. Once this infinity is subtracted, the string exchange
just reproduces the field theory result (23).
We will now show that (40) can also reproduce the pole-like divergent terms associated
with the adjoint scalars. Expanding O to linear order in the fields, we obtain the following
contribution involving the adjoint scalars

d 4p
I
det G tr i (p) tr j (p)fij (p),

Seff
(46)
(2)4
where

d 6 y eiym yi yj
fij =
(2)6 M 2 + y 2 4

m
= mi mj G(p) = c ij K1 (mp)
+ mi mj K0 (mp)
.
(47)
p
The first term in (47) leads to the action

d 4p
m
I
Seff
det G tr i (p) tr i (p) K1 (mp),
p
(2)4
(48)
which corresponds, in the mp 0 limit, to the pole-like contribution in the effective action
of the scalars (15). The second term in (47) yields a m2 log contribution which vanishes
when m 0. Notice that while f in (44) tends to 2c/p 4 in the infrared limit, fij tends
to c/p 2 . This reproduces the relative factor of two between the pole-like contributions to
the propagator of the gauge field and adjoint scalars, Eqs. (9), (10). The same applies to
253
the linear poles of the 3-point functions. Therefore, the gauge invariant expression (42),
defined such that we only keep the finite terms in the 0 limit, accounts for all polelike divergent terms of the field theory up to a global coefficient.
The discrepancy in the global coefficient can be related to the fact that not only
the tachyon, but also massive scalar closed strings couple to the brane tension. In the
SeibergWitten limit these contributions are of the same form as that of the tachyon, since
momentum in the non-commutative directions dominates over the oscillator mass. Thus
they will renormalize the overall coefficient in front of the effective action. To summarize,
we have seen that the gauge invariant effective action containing the infrared poles can be
directly related to a closed string exchange between D-branes. It is tempting to think of
this as the exchange of an effective closed string mode. Remarkably, among the original
closed string modes that contribute to this effect is the tachyon mode.
We will briefly address the log-like contributions which appear also in the supersymmetric field theory (21). Consider a two-form (denoted by MMN ) closed string which couples
to the operator OMN (separated into 4d and 6d indices):

10
d 10 P
S II = 2
(49)
det G M (P )O (P ) + Mi (P )Oi (P ) ,
10
(2)
gYM
with

1
tr d 4 x L F W (x, C) eipx ,
O (P ) =

2

1
i
tr
d 4 x L D i W (x, C) eipx .
O (P ) =
(50)

2
Repeating the same steps as for the tachyon field we can write the effective action due to
an exchange of a massive 2-form as
4 6
d pd y
eiym
II
det G OMN (p, y)OMN (p, y) 2

,
Seff
(51)
10
(2)
M + y2
with M 2 = p 2 +8 2 l and l some positive integer number which corresponds to the string
excitation number. For simplicity let us set the adjoint scalar fields to zero in W (x, C),
which does not affect gauge invariance. We get then

d 4p
II
O (p)O (p) + Oi (p)Oi (p) G(p),
Seff
(52)
4
(2)
but now we should simply keep the terms in G(p) which are O( 2 ). This yields

G(p) 2 K0 (mp).
(53)
which reproduces the action (33) and in addition the log-like pieces of the scalars (21).
Thus, we have shown that the logarithmic like (K0 , in fact) contribution to effective
action can be understood from massive 2-form closed strings exchange. Note that the
massless NSNS 2-form does not contribute here. Only massive modes. Another comment
is that we could not reproduce the overall factor in front of the effective action, 0 . The
understanding of the overall factor, from the string theory point of view, is equivalent to
the understanding of the weight of each individual massive string in the coupling to the
operator F MN on the brane. We will not address this problem here.
254
5. Discussion
In this work we have discussed UV/IR effects in a non-supersymmetric gauge theory.
Our main results are the effective actions (2) and (3). These actions incorporate the two
kinds of non-planar singular amplitudes: the poles and the logs.
The log-like contributions exist also in the supersymmetric theory, apart from the N = 4
SYM theory. The picture that emerges from our work is that one can understand these
effects as due to an exchange of massive two-form closed strings which couple to the
operator tr F .
The more interesting contributions are the poles. These poles cancel in the supersymmetric gauge theory. Our picture here is that these terms can be understood as due to an
exchange of a tachyon and massive scalar closed strings that couple to the brane tension.
In the superstring theory there is no tachyon. Moreover, the contributions from the NSNS
sector cancel the contributions from the RR sector and this is our explanation of why we
do not see such effects in the (super-)gauge theory side.
The (partial) contribution of the closed string tachyon to the tachyonic instabilities of the
non-commutative theory suggests that the two phenomena are related. Indeed, it is true that
the poles are also due to massive closed strings, since in the SeibergWitten limit all the
massive tower contributes similarly to the exchange between the D-branes [6]. Therefore,
we do not argue that the tachyon in the field theory has a one to one correspondence with
the closed string tachyon. The relation is more indirect. We have not found any example
of a non-supersymmetric string theory with a NSNS two-form which does not contain
a closed string tachyon (or a tree level open string tachyon). It is possible to construct
a non-tachyonic non-supersymmetric string theory [39,40] by using a special orientifold,
but the orientifold removes the NSNS two-form from the spectrum. In addition, in nonsupersymmetric string theories, such as strings on orbifold singularities, there is always
a tachyon in the twisted sector. In these cases the non-commutative field theory on the
brane is tachyonic. Namely, in all these constructions there are more bosons in the adjoint
representation than fermions.3 Finally, the effective action (3) is not tachyonic and indeed
it can be understood as due to massive 2-forms exchange (no closed string tachyon is
involved in this case). These observations support our claim about a relation between
the closed string tachyon and the generated tachyon on the brane. We suggest that the
closed string tachyon that couples non-trivially to the brane (in contrast to the commutative
theory), is behind the instabilities in the field theory. This point of view is somewhat similar
in spirit to [41], where it is was argued that a field theory which is holographically dual to
a tachyonic string theory should suffer from instabilities.
In the light of our picture, we would like to address the problem of the stability of the
non-commutative non-supersymmetric YangMills theory. Since this theory is tachyonic,
similarly to type 0 string theory, we suggest that the consistency issue is related to the
fate of the closed string tachyon. Type 0 string theory might be consistent if tachyon
condensation occurs (for concrete suggestions see [4244]). In particular, the true vacuum
of the type 0 string might be supersymmetric! It is tempting to suggest that if this is
3 A.A. would like to thank Rodolfo Russo for discussions on this issue.
255
the case, there will be examples of non-commutative non-supersymmetric gauge theories

which are consistent and that (1) is a consequence of the expansion around the perturbative
vacuum.
Note added
As we finished our work, paper [27] appeared. The author of this paper arrived to the
result (2) and discussed it from the matrix theory perspective.
Acknowledgements
We would like to thank Luis Alvarez-Gaum, Carlo Angelantonj, Jose Barbon, ChongSun Chu, Yaron Oz and Rodolfo Russo for discussions. Part of the work of A.A. was done
at the Theoretical Physics Centre of the Ecole Polytechnique. A.A. wishes to take this
opportunity to thank all members of the String Theory Group there.
References
[1] A. Connes, M.R. Douglas, A. Schwarz, Noncommutative geometry and matrix theory: compactification on
tori, JHEP 9802 (1998) 003, hep-th/9711162.
[3] T. Filk, Divergencies in a field theory on quantum space, Phys. Lett. B 376 (1996) 53.
[4] S. Minwalla, M. Van Raamsdonk, N. Seiberg, Noncommutative perturbative dynamics, JHEP 0002 (2000)
020, hep-th/9912072.
[5] A. Matusis, L. Susskind, N. Toumbas, The IR/UV connection in the non-commutative gauge theories,
JHEP 0012 (2000) 002, hep-th/0002075.
[6] G. Arcioni, J.L. Barbon, J. Gomis, M.A. Vazquez-Mozo, On the stringy nature of winding modes in
noncommutative thermal field theories, JHEP 0006 (2000) 038, hep-th/0004080.
[7] I.Y. Arefeva, D.M. Belov, A.S. Koshelev, O.A. Rychkov, Renormalizability and UV/IR mixing in
noncommutative theories with scalar fields, Phys. Lett. B 487 (2000) 357.
[8] A. Micu, M.M. Sheikh-Jabbari, Noncommutative 4 theory at two loops, JHEP 0101 (2001) 025.
[9] I. Chepelev, R. Roiban, Convergence theorem for non-commutative Feynman graphs and renormalization,
JHEP 0103 (2001) 001, hep-th/0008090.
[10] L. Griguolo, M. Pietroni, Wilsonian renormalization group and the non-commutative IR/UV connection,
JHEP 0105 (2001) 032, hep-th/0104217.
[11] Y. Kinar, G. Lifschytz, J. Sonnenschein, UV/IR connection: a matrix perspective, JHEP 0108 (2001) 001,
hep-th/0105089.
[12] M. Hayakawa, Perturbative analysis on infrared and ultraviolet aspects of noncommutative QED on R 4 ,
hep-th/9912167.
[13] A. Bilal, C. Chu, R. Russo, String theory and noncommutative field theories at one loop, Nucl. Phys. B 582
(2000) 65, hep-th/0003180.
[14] A. Armoni, Comments on perturbative dynamics of non-commutative YangMills theory, Nucl. Phys. B 593
(2001) 229, hep-th/0005208.
[15] K. Landsteiner, E. Lopez, M.H. Tytgat, Excitations in hot non-commutative theories, JHEP 0009 (2000)
027, hep-th/0006210.
[16] C.P. Martin, F. Ruiz Ruiz, Paramagnetic dominance, the sign of the beta function and UV/IR mixing in
non-commutative U (1), Nucl. Phys. B 597 (2001) 197, hep-th/0007131.
256
[17] M. Pernici, A. Santambrogio, D. Zanon, The one-loop effective action of noncommutative N = 4 super
YangMills is gauge invariant, Phys. Lett. B 504 (2001) 131, hep-th/0011140.
[18] V.V. Khoze, G. Travaglini, Wilsonian effective actions and the IR/UV mixing in noncommutative gauge
[19] D. Zanon, Noncommutative N = 1, 2 super-U (N ) YangMills: UV/IR mixing and effective action results
at one loop, Phys. Lett. B 502 (2001) 265, hep-th/0012009.
[20] F.R. Ruiz, Gauge-fixing independence of IR divergences in non-commutative U (1), perturbative tachyonic
instabilities and supersymmetry, Phys. Lett. B 502 (2001) 274, hep-th/0012171.
[21] K. Landsteiner, E. Lopez, M.H. Tytgat, Instability of non-commutative SYM theories at finite temperature,
JHEP 0106 (2001) 055, hep-th/0104133.
[22] A. Armoni, R. Russo, Non-commutative gauge theories and the cosmological constant, hep-th/0106189.
[23] M.R. Douglas, N.A. Nekrasov, Noncommutative field theory, hep-th/0106048.
[24] R.J. Szabo, Quantum field theory on noncommutative spaces, hep-th/0109162.
[25] N. Ishibashi, S. Iso, H. Kawai, Y. Kitazawa, Nucl. Phys. B 573 (2000) 573, hep-th/9910004.
[26] D.J. Gross, A. Hashimoto, N. Itzhaki, Observables of non-commutative gauge theories, hep-th/0008075.
[27] M. Van Raamsdonk, The meaning of infrared singularities in noncommutative gauge theories, hepth/0110093.
[28] Y. Kiem, S.J. Rey, H.T. Sato, J.T. Yee, Open Wilson lines and generalized star product in nocommutative
scalar field theories, hep-th/0106121.
[29] Y. Kiem, S. Rey, H. Sato, J. Yee, Anatomy of one-loop effective action in noncommutative scalar field
theories, hep-th/0107106.
[30] Y. Kiem, S.S. Kim, S.J. Rey, H.T. Sato, Anatomy of two-loop effective action in noncommutative field
theories, hep-th/0110066.
[31] H. Liu, J. Michelson, -Trek: the one-loop N = 4 noncommutative SYM action, hep-th/0008205.
[32] H. Liu, -Trek II: n operations, open Wilson lines and the SeibergWitten map, hep-th/0011125.
[33] S.R. Das, S.P. Trivedi, Supergravity couplings to noncommutative branes, open Wilson lines and generalized
star products, JHEP 0102 (2001) 046, hep-th/0011131.
[34] Y. Okawa, H. Ooguri, How noncommutative gauge theories couple to gravity, Nucl. Phys. B 599 (2001) 55,
hep-th/0012218.
[35] H. Liu, J. Michelson, Supergravity couplings of noncommutative D-branes, hep-th/0101016.
[36] M.R. Douglas, D. Kabat, P. Pouliot, S.H. Shenker, D-branes and short distances in string theory, Nucl. Phys.
B 485 (1997) 85, hep-th/9608024.
[37] I.R. Klebanov, A.A. Tseytlin, Asymptotic freedom and infrared behavior in the type 0 string approach to
gauge theory, Nucl. Phys. B 547 (1999) 143, hep-th/9812089.
[38] M.R. Garousi, String scattering from D-branes in type 0 theories, Nucl. Phys. B 550 (1999) 225, hepth/9901085.
[39] A. Sagnotti, Some properties of open string theories, hep-th/9509080.
[40] A. Sagnotti, Surprises in open-string perturbation theory, Nucl. Phys. (Proc. Suppl.) B 56 (1997) 332, hepth/9702093.
[41] I.R. Klebanov, Tachyon stabilization in the AdS/CFT correspondence, Phys. Lett. B 466 (1999) 166, hepth/9906220.
[42] O. Bergman, M.R. Gaberdiel, Dualities of type 0 strings, JHEP 9907 (1999) 022, hep-th/9906055.
[43] M.S. Costa, M. Gutperle, The KaluzaKlein Melvin solution in M-theory, JHEP 0103 (2001) 027, hepth/0012072.
[44] A. Adams, J. Polchinski, E. Silverstein, Dont panic! Closed string tachyons in ALE spacetimes, hepth/0108075.

Non-extremal fractional branes

M. Bertolini a , T. Harmark b , N.A. Obers c,d,e , A. Westerberg f
a NORDITA, Blegdamsvej 17, DK-2100 Copenhagen , Denmark
b Jefferson Physical Laboratory, Harvard University, Cambridge, MA 02138, USA
c Spinoza Institute and Institute for Theoretical Physics, Utrecht University, Leuvenlaan 4,
3584 CE Utrecht, The Netherlands

d Institute for Theoretical Physics, University of Amsterdam, Valckenierstraat 65,
1018 XE Amsterdam, The Netherlands

e The Niels Bohr Institute, Blegdamsvej 17, DK-2100 Copenhagen , Denmark
f CERN, TH Division, CH-1211 Geneva 23, Switzerland
Received 20 March 2002; accepted 26 March 2002
Abstract
We construct non-extremal fractional D-brane solutions of type-II string theory at the Z2 orbifold
point of K3. These solutions generalize known extremal fractional-brane solutions and provide
further insights into N = 2 supersymmetric gauge theories and dual descriptions thereof. In
particular, we find that for these solutions the horizon radius cannot exceed the non-extremal
enhanon radius. As a consequence, we conclude that a system of non-extremal fractional branes
cannot develop into a black brane. This conclusion is in agreement with known dual descriptions of
the system. 2002 Published by Elsevier Science B.V.
1. Introduction
Fractional D-branes [13] have proved an interesting and rich subject in string theory,
generalizing the ordinary D-branes of type-II string theory. They differ from the latter
not only because they carry fractional charges but also in that their world-volume
gauge theories are in general non-conformal. In particular, fractional D-branes with eight
supercharges can be obtained from type-II string theory on an orbifold limit of K3, the
Work supported in part by the European Communitys Human Potential Programme under contract HPRNCT-2000-00131 Quantum Spacetime.
E-mail addresses: teobert@nordita.dk (M. Bertolini), harmark@bose.harvard.edu (T. Harmark),
n.obers@phys.uu.nl (N.A. Obers), anders.westerberg@cern.ch (A. Westerberg).
0550-3213/02/$ see front matter 2002 Published by Elsevier Science B.V.

PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 2 3 7 - 7
258
M. Bertolini et al. / Nuclear Physics B 632 (2002) 257282
simplest one being T 4/Z2 , or on the orbifold limit of an ALE space, which is C2/ , where
corresponds to the ADE-classified finite symmetry group of the ALE space [4,5].
In the present paper we consider the type-II string theories on the T 4/Z2 orbifold limit of
K3. For these cases, fractional D-branes have half the charge of the usual regular D-branes.
Their world-volume theories contain only a vector multiplet, while the regular branes carry
in addition hypermultiplets. The extremal supergravity solutions for this class of fractional
branes were found in Ref. [6]. Here we study the non-extremal generalizations of these
solutions.
The three main reasons for studying this problem are as follows. The first is that it is
interesting to determine whether fractional branes in string theory are dynamical objects in
the same sense as ordinary D-branes. Making the branes non-extremal tests whether it is
possible to consider thermally excited fractional D-branes.
The second reason is related to the fact that the gauge theories living on the fractional
branes that we consider are (3 + 1)-dimensional pure N = 2 super-YangMills (SYM)
theory or dimensional reductions thereof. These are non-conformal theories and it is of
principal interest to extend the Maldacena conjecture [79] to such settings. To make the
fractional branes non-extremal would in principle enable us to obtain a dual description of
finite-temperature pure SYM theories with eight supercharges.
The study of a supergravity/gauge-theory duality for pure SYM with eight supercharges
was initiated in Ref. [10] by considering D-branes wrapping K3, a setup which is T-dual
to the one with fractional D-branes [1012]. The upshot of the analysis in Ref. [10] is
that the supergravity solution dual to pure SYM with eight supercharges has a repulson
singularity [1315] near the center. However, this repulson can be excised from the
solution by noticing that at a certain distance from the center, the so-called enhanon
radius, an abelian field in the effective theory becomes non-abelian, the gauge symmetry
being enhanced from U(1) to SU(2). This means that the low-energy effective theory
contains additional massless fields which have to be taken into account. Moreover, a probe
computation shows that D-brane probes become massless as one reaches the enhanon
radius. The interpretation of this phenomenon is that there is a spherical distribution of
D-branes at the enhanon radius with flat space inside the sphere [10]. Similar results have
subsequently also been obtained for other configurations of wrapped branes [1618] as
well as for fractional brane configurations both on non-compact orbifolds like C2 / [19
24] and on the compact T 4 /Z2 orbifold limit of K3 [6].
The third main reason for our interest in non-extremal fractional branes is the interplay
between the non-extremality features and the enhanon mechanism; since extremal
fractional branes on T 4 /Z2 have an enhanon radius [6], it is of interest to see whether
a horizon covering the enhanon can develop and thus provide another kind of excision
mechanism for the fractional branes. For systems of D6-branes wrapped on K3, this
question has been addressed in Refs. [10,25] where it was found that a horizon can
indeed be formed. However, in the supergravity/gauge-theory duality the horizon radius
corresponds to energies that are beyond reach in the pure SYM theory. This observation
has been regarded as evidence for the conjecture that the dual of pure SYM with eight
supercharges is a non-gravitational theory [10,26]. As we are going to show, fractional
branes provide further corroboration of this conjecture.
259
For N = 1 SYM in 3 + 1 dimensions the thermal version of the KlebanovStrassler

setup [27] that builds on the paper [28] has been explored in Refs. [2931]. Here it has
proved very hard to find an exact non-extremal solution. For the N = 1 SYM theory
progress has been made in Refs. [3235], building on the setup of Ref. [36] describing zerotemperature N = 1 SYM. In the latter case, decoupling the scale of confinement from
the string scale in the NS5-brane world-volume theory has turned out to be problematic,
making it difficult to do computations. However, the problems encountered in N = 1 and
N = 1 SYM seem rather unrelated to the problems of finding a non-extremal dual to
N = 2 SYM.
The summary and organization of the paper are as follows. In Section 2 we present
the supergravity solutions describing non-extremal fractional branes at low energy. More
precisely, we consider a bound state of M fractional Dp-branes of type-II string theories
on the orbifold T 4 /Z2 . Our solutions thus generalize the fractional-brane solutions of
Ref. [6]. Their structure turns out to be somewhat simpler than that of those found in
analogous investigations of the N = 1 theory in Refs. [2931]. In Section 3 we discuss
the physical properties of the non-extremal solutions, focusing on the interplay between
the non-extremal version of the enhanon and the black-hole horizon r0 . In particular, we
find that the horizon radius cannot exceed the enhanon radius. As a consequence, we
conclude that these systems of non-extremal fractional branes cannot develop into black
branes. The consequences of the latter conclusion and further interpretation of the results
are our primary concerns in Section 4. Finally, some computational details are given in
an appendix. There we also discuss another solution branch, with well-defined black-hole
thermodynamics but the physical interpretation of which is presently not clear.
2. The non-extremal solutions

In this section we present the non-extremal low-energy supergravity solutions for
fractional Dp-branes on K3 in the T 4 /Z2 orbifold limit. For the type-IIA case, the relevant
truncated six-dimensional supergravity action was obtained in Ref. [6] by compactification
of type-IIA supergravity on T 4 /Z2 , and used to construct extremal solutions for fractional
D0- and D2-branes. After deriving the corresponding model on the type-IIB side, we
will consider the non-extremal generalizations of these solutions for the full range p =
0, 1, 2, 3.
2.1. Fractional branes
Fractional branes [13] are certain types of BPS D-branes that one encounters when
considering string theory in singular backgrounds. They can be defined in many different
(but equivalent) ways. Probably, the most intuitive way to understand their properties is
through their description as higher-dimensional D-branes wrapped on a vanishing cycle
of the singular manifold [37,38]. This geometric picture makes manifest the characteristic
feature of fractional branes as being stuck at the singularity while free to move in the
flat transverse directions only. This viewpoint also shows why they lack the world-volume
degrees of freedom associated to fluctuations in the orbifold directions which correspond,
260
in general, to hypermultiplet excitations. Moreover, as will become clear below, this also
automatically accounts for the coupling of fractional branes with the twisted sector of
string theory on the orbifold. We refer to Ref. [39] for a recent review on the properties of
fractional branes on orbifolds for theories with eight supercharges.
Let us now focus on our main case of interest. The K3 manifold has dim(H2 (K3)) = 22
two-cycles, on a space of signature (19,3). At the Z2 orbifold point, there are three selfdual and three anti-selfdual cycles from the six two-cycles in H2 (T 4 ), which are invariant
under the Z2 involution. In addition, there are 16 anti-selfdual cycles that come from the
collapsed spheres at the 16 orbifold singularities. A fractional Dp-brane is then a D(p + 2)brane wrapped on one of the these cycles, C, in the Z2 orbifold limit of K3. The presence
of a background NSNS two-form flux through the shrinking cycle makes the D-brane
tension non-vanishing. The background value of this flux is dictated to be

1 2
b0 = B(2) = 2
(2.1)
2
C
by the requirement of conformal invariance of the string world-sheet in the orbifold

background [40]. The effective theory describing the dynamics of such a brane at low
energy is a (p + 1)-dimensional pure SYM theory with eight supercharges. From the closed
string theory point of view, this translates into the fact that fractional branes couple to both
the untwisted and the twisted sector of string theory on the orbifold.
At low energy, in the infinite-volume limit of T 4 /Z2 , a given fractional Dp-brane
simply couples to the metric, the dilaton and the RR (p + 1)-form potential in the untwisted
sector, and to a scalar field and a (p + 1)-form potential in the twisted sector. The two latter
correspond, respectively, to the zero mode of the NSNS B(2) field and of the RR (p + 3)form potential when dimensionally reduced on the shrinking two-cycle, and belong to a
non-gravitational multiplet of the effective supergravity theory [37,38]. Denoting by (2)
the closed differential two-form Poincar dual to the vanishing two-cycle C on which the
branes are wrapped, the relations between higher-dimensional forms and twisted fields read
C(p+3) = 2V A(p+1) (2) .

B(2) = b(2),
(2.2)
In the compact case (anticipated here by the introduction of the dimensionful constant V
to be defined shortly) the effective six-dimensional theory is augmented by the zero modes
of massless fields on the internal manifold. However, these come only from the untwisted
sector since, by construction, the twisted fields have no dynamics on the internal space. For
(2) we adopt the following conventions:

1
(2) + (2) = 0.
(2) = 1,
(2) (2) = ,
(2.3)
2
C
C2 /Z2
2.2. The actions

Let us start by fixing our conventions and presenting the consistently truncated sixdimensional actions describing the dynamics of the supergravity fields to which the
fractional Dp-branes couple. As just discussed, the gauge fields that enter the solution
261
for a given p are the two (p + 1)-form potentials C(p+1) and A(p+1) . From the twisted
scalar field arising from the NSNS two-form potential we separate out a fluctuating part
b according to
b=
1
V + 2V b,
2
4
V = 2 ,
(2.4)
where V is the volume of the compact space T 4 /Z2 and V is introduced as a shorthand
notation. Furthermore, the KaluzaKlein reduction of the metric gives rise to four scalar
these
fields, e2a , a = 6, . . . , 9. Assuming homogeneity in the compactified directions

are equal and can be replaced by a single scalar field defined as = 9a=6 a . The
dimensionally reduced dilaton = |d=10 12 completes the scalar field content.
For the case p = 0 the truncated bosonic d = 6 action governing the dynamics of these
fields was derived in Ref. [6], together with the electro-magnetically dual two-brane case.
From these results the corresponding action for the fractional D3-brane on the type-IIB
side can readily be inferred. These three actions take the form1

1
V
1
(p=0,2,3)
Sbulk
= 2
d 6 x g R + d d + d d + e d b d b
4
2
2

1
1
(1p)

e G(p+2) G(p+2) + F(p+2) F(p+2) ,
+e
2
2
(2.5)
where = 8 7/2 gs 2 is the ten-dimensional gravitational coupling and
G(p+2) = dC(p+1) ,
(p+2) = dA(p+1) + d b C(p+1)

F
(2.6)
are the gauge-invariant field strengths of the untwisted- and twisted-sector potentials
C(p+1) and A(p+1) , respectively.
(3)
For the remaining case, p = 1, some subtleties arise because the field strength F
is self-dual, a property inherited from its ten-dimensional type-IIB five-form parent. As
usual, the self-duality condition has to be imposed on shell. With this proviso in mind we
can nevertheless write a truncated d = 6 gravity action also for the fractional D-string:2

1
V
1
(p=1)
Sbulk = 2
d 6 x g R + d d + d d + e d b d b
2
4
2
1
1
1

A
+ e G(3) G(3) + F
d
b
.
(3)
(3)
(2)
(3)
2
4
2
(2.7)

(3) ,
The C(2) equation of motion is compatible with the self-duality constraint F(3) = F
a fact which relies crucially on the presence of the ChernSimons term.
1
6
1 Our conventions are
n
012345 = +1.
n+1 6 and $
(6n) = n!g dx 1 dx $1 n n+1
(n) + H
(3) C
(n3) .
(n+1) = d C
The RR field strengths in ten dimensions were defined as G
2 This action results from taking a type-IIB action in ten dimensions compatible with an anti-selfdual five-form
(5) , and using the conventions (2.3) in the KaluzaKlein reduction on T 4 /Z2 .
field strength G
262
In solving the equations of motion obtained from the actions (2.5) and (2.7) one should
also specify the boundary conditions at infinity that the various fields should satisfy (i.e.,
the mass and the charges of the soliton). We are interested in describing a general bound
state of non-extremal fractional Dp-branes whose corresponding supergravity solution
should match the extremal one at infinity. The boundary conditions at infinity for the latter
are encoded in the bosonic world-volume action describing the low-energy dynamics of an
extremal fractional brane. For the case at hand this action has been derived in Ref. [6] and
reads

Tp
2V
p+1
/2 (1p)/2
Swv =
b
g e
e
1+2
d
2
V

Tp
2V
2V
+
(2.8)
A(p+1) ,
b +2
C(p+1) 1 + 2
2
V
V
where Tp = (2 )3p . The world-volume action for a stack of M coincident

fractional Dp-branes is obtained simply by multiplying the above action by M (this will
be implicitly assumed in what follows).
2.3. Anstze and solutions
Below we present the solutions of the equations of motion, given in Appendices A.1 and
A.2, referring to Appendix A.3 for an outline of their derivation. We will wherever possible
treat the cases p = 0, 1, 2, 3 in parallel. As far as the starting-pointi.e., the ansatz for
the metricis concerned, the three-brane (being of codimension two in six dimensions)
however needs some special consideration. The solutions, nevertheless, share a common
structure for all four cases as will become clear below.
Let us first discuss the lower-dimensional cases, p = 0, 1, 2, for which the standard
non-extremal p-brane ansatz applies:

p

i 2

2
(p3)/4
2
2
ds = H
dx
. (2.9)
f dt +
+ H (p+1)/4 f 1 dr 2 + r 2 d4p
i=1
The non-extremality is introduced by the function f which, like H , depends on the

transverse radial coordinate only. It is constrained by the equations of motion to satisfy
the harmonic equation
f +
4p
f = 0.
r
(2.10)
Requiring the non-extremal solution to approach the extremal one at infinity, the solution
to Eq. (2.10) takes the form

f =1
r0
r
3p
,
(2.11)
263
with the horizon radius r0 governing the degree of non-extremality.3

The upshot of the analysis of Appendix A.3 is that the non-extremal solutions can be
expressed entirely in terms of f and two additional harmonic functions, h1 and h2 , which
in turn depend on three parameters: the horizon radius r0 and two charges, q1 and q2 ,
depending linearly on the number of branes which act as sources for the solution. More
specifically, the scalar fields are given by

H
q2 h2
(1p)/4
,
e = 2,
b=
1 ,
e =H
(2.12)
q1 h1
h1
while the gauge potentials take the form
C0p =
q2
3p
r
h3
,
H
A0p =
q1
3p
r
1
,
h1
(2.13)
the associated field strengths (defined in Eq. (2.6) above) being

(3 p)q2 h1 h2
r0p = (3 p)q1 1 .
,
F
r 4p
H2
r 4p H
The functions H and h3 entering the solution read

1
1 q22 2 1 q22 2
h1
h2 ,
h3 = (h1 + h2 ),
H = 1+
2
2
2 q1
2 q1
2
Gr0p =
where the two basic harmonic functions are

3p
ri
, i = 1, 2.
hi = 1
r
3p
r2
where
1 3p
= r0 +
2
(2.15)
(2.16)
The radial parameters of these functions are given by

2(3p)
2q14 + (q12 + q22 )r0
2q12
1 3p
3p

r1 = r0 +
,
2
2 2q 2 + q 2

(2.14)
(2.17)
2(3p)
2q14 + (q12 + q22 )r0

2 q22
+ 2q12
(2.18)

2(3p) 1 4(3p)
q14 + q12 + q22 r0
+ r0
.
(2.19)
4
As already noticed, the solution we have found is completely fixed by extracting the
relation between the free-supergravity values of the charges q1 and q2 and those dictated by
the world-volume action (2.8) for the M fractional branes. Using Eqs. (A.71) and (A.72)
3 Although the metric (2.9) develops a horizon at r we shall see in the next section that the supergravity
0
approximation ceases to be valid at a radius re strictly larger than r0 . Keeping this in mind, we shall nevertheless
refer to r0 as the horizon radius.
264
and equating the corresponding charges with the coupling to the twisted and untwisted
potentials A0p and C0p in the WZ action one gets

2V
M
q1 =
(2.20)
Qp M,
q2 = Qp ,
V
2
where
Qp =
1
2 Tp
V 3p
=
2
gs ,
kp 4p V
kp 4p V
(2.21)
with 4p denoting the volume of the unit (4p)-sphere surrounding the p-branes and
kp = 3 p for p < 3 while k3 = 1.4
As a consistency check let us also take the extremal limit, r0 = 0. Assuming the charges
to be positive, as we will from now on, we obtain
q12 1
q12
q2
(2.22)
,
H
=
1
+
,
extr
q2 r 3p
r 3p 2r 2(3p)
so that the solution of Ref. [6] is recovered:
q1
extr
1
eextr = Hextr ,
(2.23)
bextr = 3p = Aextr
C0p
= Hextr
1.
0p ,
r
Note that the structure of the function H that determines the non-extremal metric (together
with f ) and the dilaton is the same as for the extremal case; while the coefficients of
course differ, introducing a non-extremality parameter gives no terms beyond the r 2(3p)
correction, the latter being the usual fractional brane modification to the harmonic function
governing the regular-brane solution. For the other fields the non-extremal modifications
are somewhat more intricate. Nevertheless, for our class of non-extremal fractional branes,
these modifications are entirely due to the non-triviality of the harmonic function h1
for r0 > 0. In particular, we note that, contrary to the extremal case, the non-extremal
twisted fields do get corrections with respect to their harmonic asymptotic behavior. The
absence of such corrections was taken as input for the ansatz relevant to extremal fractional
to be a manifestation of the
branes [6], and was argued (for the NSNS twisted scalar b)
fact that N = 2 SUSY only allows for one-loop perturbative corrections. The fact that
this property ceases to hold for our non-extremal generalization suggests that the nonextremality parameter r0 indeed does switch on a temperature in the system.
Turning to the case p = 3, the appropriate non-extremal ansatz for the metric reads [41]
2
3

i 2

r
dx +
ds 2 = f dt 2 +
(2.24)
H f 1 dr 2 + r 2 d 2 ,
r
hextr
1 = 1,
hextr
2 =1
i=1
where r is an (as yet) undetermined radial parameter. Using this ansatz, the analysis in
Appendix A.3 can be done in parallel with the lower-dimensional cases. As a consequence,
4 The solution describing a composite bound state of M fractional and N regular Dp-branes is identical
to the one we have discussed so far, the only difference being that the untwisted charge q2 will now be
q2 = Qp ( M
2 + N ). By taking M = 0 one gets q1 = 0 and q2 = Qp N , giving back the usual regular brane
solution, with no coupling to twisted fields.
265
the expressions listed above for the metric, the scalars and the gauge field strengths (upon
substituting (3 p) 1) in terms of the functions h1 and h2 are valid also for p = 3.
Although the reason is slightly less obvious, by using the mapping r (3p) log(r /r)
the same turns out to be the case for the gauge potentials in (2.13), giving
r
q1
r
h3
log ,
A0123 = log .
(2.25)
H
r
h1
r
Here r can be interpreted as a long-distance cut-off in the transverse radial direction of the
three-brane world volume, corresponding to a UV cut-off on the dual gauge-theory side.
This mapping originates from the harmonic functions which in two dimensions involve a
logarithm. The function f , for instance, again satisfies the harmonic equation (2.10) but
the solution now reads
r
f = 1 a0 log .
(2.26)
r
The dimensionless non-negative parameter a0 governs the degree of non-extremality. As
will become evident below, a0 can be viewed as the direct formal analogue of the parameter
r0 in the solutions for the lower-dimensional branes. The horizon radius 5 for the threebrane is r0 = r e1/a0 and the extremal limit is obtained by letting a0 0 with r fixed.
Moreover, it is convenient to identify the parameter r in (2.24) with r since the latter sets
the length scale of the transverse geometry.
Hence, the conditions that h1 and h2 be harmonic now imply
r
r
h1 = 1 a1 log ,
(2.27)
h2 = 1 a2 log .
r
r
In an exact analogy with the results (2.17) and (2.18) for p < 3 the parameters a1 and a2
are given by

2q14 + (q12 + q22 )a02 2q12
1

a1 = a0 +
(2.28)
,
2
2 2q 2 + q 2
C0123 = q2
1
a2 = a0 +
2
with
2q14 + (q12 + q22 )a02 + 2q12

,
2 q22
(2.29)

1
(2.30)
q14 + q12 + q22 a02 + a04 .
4
The reason for these close formal similarities between p = 3 and the lower-dimensional
cases is explained in Appendix A.3.
Taking the extremal limit, the gauge potentials (2.25) simplify to
r
extr
1
,
= Hextr
1,
Aextr
C0123
(2.31)
0123 = q1 log
r
=
5 We write the term within quotes since the transverse space is two-dimensional and black holes therefore can
never develop for p = 3.
266
where Hextr = 1 + q2 log rr 12 q12 (log rr )2 . In addition we have bextr = q1 log rr , with
the remaining fields formally identical to their lower-dimensional counterparts. Note that
in the extremal limit of the three-brane solution we have = 12 . From the definition
of this relation immediately implies that the ten-dimensional dilaton is constant, in
agreement with the fractional D3-brane solution of Refs. [19,20] for the non-compact
orbifold spacetime R1,5 C2 /Z2 .
3. Enhanon versus horizon

In this section we first review the enhanon mechanism and describe its manifestation
in the case at hand. Then we examine the interplay between the event horizon and the
enhanon shell.
3.1. Review of the enhanon
Supergravity solutions of brane configurations which have pure SYM with eight
supercharges as their low-energy world-volume theories are in general plagued by naked
singularities. This is in particular true for fractional branes and, more generally, for Dbranes wrapped on topologically non-trivial cycles. The naked singularities one encounters
are of repulson type [1315] and can be excised by the so-called enhanon mechanism [10].
The logic of this mechanism is as follows. Far away from the source we have a perfectly
valid supergravity solution. However, when approaching the source there is a certain radius
re , dubbed the enhanon radius, at which the effective supergravity description needs
to be augmented by additional massless degrees of freedom, the appearance of which
leads to an enhancement of the gauge symmetry from U(1) to SU(2).6 A brane-probe
calculation shows that the tension of the probe vanishes precisely at the enhanon radius.
The interpretation of this phenomenon is that the branes are not located at the origin but
instead smeared over the spherical shell r = re . The geometry for r < re is consequently
completely different, and, in particular, the singularity is excised [10,25].
For fractional branes, the symmetry enhancement occurs when the NSNS two-form
flux through the vanishing cycle on which the brane is wrapped flows from the value 1/2
at infinity to 0 at the enhanon radius; when the flux vanishes all parameters of the cycle
are zero, which corresponds to a point of enhanced symmetry in the moduli space. At
this point, the fractional branes become tensionless. In particular, the massless degrees of
freedom responsible for the enhanced gauge symmetry correspond to tensionless fractional
D-particles for type IIA and to tensionless fractional D-strings for type IIB [43].
For the extremal fractional D-branes on T 4 /Z2 the enhanon mechanism has been
examined in Ref. [6]. As we are going to discuss in the next section, the presence of
the enhanon has crucial consequences for the decoupling limit and for the nature of the
gauge-theory/gravity duality for this kind of systems. It is therefore important to investigate
6 While on the type-IIA side the symmetry enhancement occurs for an ordinary vector field, the enhanced A
1
gauge symmetry in the type-IIB case is carried by a self-dual two-form potential [42].
267
if and how the enhanon is modified in the non-extremal case we are discussing, and more
specifically what the relation between the event horizon and the enhanon is.
As discussed in the previous section, fractional Dp-branes are a particular kind of
wrapped D(p + 2)-branes arising in orbifold compactifications of string theory. While
the geometric volume of the compact two-cycle characterizing the orbifold is identically
zero, there is a non-trivial NSNS two-form background flux on it, displayed in (2.1),
which makes the effective stringy volume asymptotically non-vanishing. In fact, as already
explained in the previous section, this asymptotic value is modified by the presence of the
fractional branes (either extremal or non-extremal). Let us recall the relation between the
fields entering the solution discussed in Section 2 and the NSNS two-form flux:

b=
B(2) =
C
1
V + 2V b(r).
2
(3.1)
Here we recall that V is the volume of the compact orbifold and V = (2 )4 as defined
in (2.4). From the above considerations it follows that the enhanon is located at the radius
re determined by

1+2
2V
b(re ) = 0.
V
(3.2)
This equation gives the position of the enhanon in the general case. In the extremal limit
it reduces to the result found in Ref. [6].
Keeping in mind that it is Eq. (3.2) that defines the enhanon shell, we can also check
that fractional Dp-brane probes indeed become tensionless there. One simply takes the
DBI action for the probe evaluated in the background generated by the source branes,
chooses static gauge with the transverse coordinates depending on time only, and finally
expands the action up to terms quadratic in the velocities. The appropriate DBI action is
Tp
SDBI =
2

d
p+1
g e
/2 (1p)/2

2V
b .
1+2
V
(3.3)
By proceeding as just outlined one easily sees that the probe brane becomes tensionless
when Eq. (3.2) is satisfied. More precisely, for slow radial motion (r 2 f 2 /H ) the kinetic
part of its effective Lagrangian is

Tp Vp h1 (r)
2V
T (r, r ) =
1+2
b(r) r 2 ,
4 f 3/2 (r)
V
(3.4)
showing that the effective tension of the probe is r-dependent and that, as promised, the
brane becomes tensionless at the distance re given by Eq. (3.2).7
7 This argument rests on r being larger than both r and r , which, as we shall see, is always the case.
e
0
1
268
3.2. Application to non-extremal fractional branes

For our class of non-extremal fractional branes, we can obtain a very simple expression
for the enhanon. By substituting the solution discussed in the previous section, in
particular Eqs. (2.12) and (2.16), in Eq. (3.2) we find

2V q2 3p
3p
3p
3p
3p
r2 r1
+ r1 = r2 ,
re = 2
(3.5)
V q 1
where in the last step we have used the relations (2.20). A completely analogous expression
holds for ae , and thus for re , in the three-brane case (recall that re = r e1/ae ).8
We now turn to examining the position of the enhanon shell relative to that of the event
horizon. Using Eqs. (2.18) and (2.19), it is easy to see from Eq. (3.5) that the enhanon
always lies outside the horizon, no matter the value of r0 . This means that in the region
of validity of the supergravity approximation, the bound state never develops into a black
brane. This might seem puzzling, since one would think that for large enough mass the
system would indeed develop into a black brane, while the above equations show that the
enhanon increases with r0 faster than r0 itself. However, one has to remember that the
energy density of this configuration is not concentrated in the center of the solution, but
rather spread out on the enhanon shell. Indeed, the fact that we cannot arrange for the
horizon to lie outside the enhanon shell is nicely consistent with the fact that, while the
mass is not bounded, the density of mass is. To see this, note first that the density of mass is
the total mass Mp divided by the volume Vtot that we can fit the system into. The mass Mp
3p
5p
goes like r0 while the volume Vtot goes like re , and hence Mp /Vtot actually decreases
as we increase r0 , since r0 is smaller than re .
One can also try to extend the solution to the interior of the enhanon shell. Fractional
branes cannot get inside the enhanon since their tension vanishes there and their energy
(and charge) is believed to be distributed on the enhanon shell. Therefore, the extremal
solution is flat in the interior and the energy density vanishes there [10,44].9 However, by
making the solution non-extremal one could imagine creating a neutral black hole on the
inside, characterized by an internal horizon radius r0 . One could then try to increase r0
enough to make it larger than the enhanon, thus allowing the system to develop into a
black brane.10
If we demand that such an interior black hole be in equilibrium with the branes at the
enhanon shell, we can in principle find r0 in terms of r0 and M. This raises the interesting
question whether the interior horizon can reach the enhanon radius. Unfortunately,
we cannot answer this question in a precise way since we do not understand nonextremal fractional branes sufficiently well to determine when we have thermodynamical
8 As already mentioned, black holes cannot appear in 2 + 1 dimensions so we exclude the three-brane case
from the following discussion.
9 By studying the SeibergWitten curve it has been explicitly shown that the supergravity fields should be
constant in the interior [45]. On the other hand, to discuss the system at the enhanon scale one should include
the extra massless fields into the analysis.
10 Obviously, we would then have to jump to another branch of the solution. In terms of the four branches of
the solution summarized in Table 1 in Appendix A.4 we should jump from branch I to branch IV.
269
equilibrium. However, the fact that the density M/Vtot , as defined above, decreases for
increasing r0 makes it seem unlikely that by increasing r0 and thereby decreasing the
density of mass M/Vtot , one could make the system collapse into a black hole. In the next
section we interpret the fact that the non-extremal fractional-brane system on the orbifold
under study never collapses into a black hole from the perspective of the SYM theory living
on the brane and its supergravity dual.
At this point we would like to alert the reader to an apparent puzzle.11 The discussion
so far can readily be extended to the more general bound states where N regular branes
are also present. As noticed in the previous section, the only modification of the solution
is that we now have q2 = Q2 (M/2 + N). Of course, probing the geometry with regular
branes does not give any information on the enhanon locus since regular branes do not
couple to the B(2) -flux and are insensitive to the enhanon. On the other hand, revisiting
the fractional-brane probe computation, the enhanon radius is now found to be
N
(3.6)
(r2 r1 ) .
M
Surprisingly, when examining the relative positions of the enhanon and the horizon, the
conclusions do not change with respect to the pure fractional-brane case. Indeed, this
follows from a simple inspection of Eqs. (2.17) and (2.18) which gives at hand that r2 r1
is always positive for our solution. Thus, r0 is always smaller than re , also in the presence
of regular branes and regardless of the relative value of N with respect to M. This contrasts
with the result found in Refs. [10,25] for branes wrapped on K3. However, the two systems
are different. In our case, as already noticed, regular branes do not feel the enhanon and
a regular D-brane probe can thus go all the way to the center r = 0. Indeed, one finds
that re M at extremality not only in Eq. (3.5) but also in Eq. (3.6). The extremal N regular/M-fractional brane configuration can therefore be thought of as composed of N
regular branes at the origin and M fractional branes smeared on the enhanon shell. In
contrast, for the case discussed in Refs. [10,25] the unwrapped branes (which correspond to
regular branes here) do indeed influence the enhanon at extremality, causing it to decrease
in size. When the number of unwrapped branes exceeds the number of wrapped ones, the
enhanon is small enough that, for a sufficiently large non-extremality parameter r0 , the
system can be turned into a black hole with r0 larger than re . In our case the situation
is different. Nevertheless, one would expect a limit N M in which the enhanon at
extremality would be quantitatively irrelevant and thus should not sensibly affect the
regular-brane thermodynamics. In this respect, it would be very interesting to find the
precise relation between r0 and the external parameters when imposing thermodynamical
equilibrium between the internal black hole with horizon radius r0 and the enhanon shell.
Having extensively discussed the possibility of our system developing a horizon, we
should also mention that there is an additional characteristic radius whose relative position
with respect to the enhanon is of interest. We are referring here to the radius where the
potential for a massive particle probe feeling only the string-frame geometry becomes
repulsive. This radius (denoted rd in Ref. [25]) is readily obtained as the stationary point
re = r2 + 2
11 For simplicity of notation we give expressions appropriate for the two-brane case although the discussion
applies generally.
270
of the effective potential for a radially moving probe with world-line action given by the

string-frame NambuGoto action. The resulting condition is (gts.f.
t ) (rd ) = 0, which for our
non-extremal case translates to (f H 1/2) (rd ) = 0. The solution reads (still for p = 2)
rd =
(2q12 + q22 )(r0 r1 )r1 + q22 (r2 r0 )r2

2q12 (r0 r1 ) + q22 (r2 r1 )
(3.7)
It is straightforward to verify that this radius is always smaller than or (at extremality)
equal to re , and that the enhanon hence excises the unphysical region.
Finally, let us recall that Dp-brane probes (fractional or regular) moving in a background generated by the same objects also feel non-trivial potentials out of extremality,
where supersymmetry is broken. Although these potentials are not due solely to gravity, we
have nevertheless observed that the regions where they would have become repulsive are
excised by the enhanon. In fact, it is amusing to notice that for a regular Dp-brane probing a non-extremal fractional-brane background, the radius at which the potential changes
sign actually coincides with the enhanon radius.
4. Discussion and conclusions

The upshot of the previous section is that a system of non-extremal fractional branes on
the orbifold T 4 /Z2 cannot collapse into a black hole. We discuss in the following a possible
interpretation of this observation in terms of the pure SYM theory living on the fractional
D-branes and the theories dual to them in the sense of the AdS/CFT correspondence [79].
It is well known [11] that a transverse T-duality on a fractional brane gives a Hanany
Witten setup [46]. In particular, a fractional Dp-brane on the orbifold T 4 /Z2 is T-dual
to a D(p + 1)-brane stretched between two NS5-branes, the distance between them being
proportional to the flux b of the NSNS two-form of our scenario. Moreover, from the
NS5-brane setup one obtains the wrapped brane setting of Refs. [10,18] by a transverse
T-duality. All of these brane setups describe at low energies a pure SYM with eight
supercharges. It was argued in Refs. [10,26] that the dual of pure SYM with eight
supercharges is a non-gravitational theory. This means that the gravitational multiplet
decouples for a fractional D-brane solution in type-II string theory on T 4 /Z2 in the
decoupling limit of the pure SYM on the brane. In the T-dual HananyWitten setup
this is the same as saying that gravity decouples from the NS5-branes so that the dual
theory is described by the non-gravitational theory living on the NS5-brane [4749]. In the
fractional-brane setup, the (5 + 1)-dimensional fields living on the NS5-branes correspond
to the fields of the twisted sector, which indeed have (5 + 1)-dimensional dynamics.
In fact, these are the only fields entering all the relevant gauge-theory quantities in the
correspondence while the contribution from the gravitational multiplet always cancels as
shown for instance in Refs. [19,2123,5052].
2 M and
Now, in the decoupling limit r/re is fixed (this because r and re gYM
2
we have gYM M fixed in the limit). Hence, if there had been a horizon r0 > re , it would have
remained after taking the decoupling limit, and we would thus have had a black hole in the
dual theory. This would have been in contradiction with the conjecture that the dual is a
non-gravitational theory. Hence, our analysis can be seen as a further piece of evidence that
271
the dual theory is indeed non-gravitational. In Ref. [10] a similar consideration was made
for D-branes wrapping K3. For that case it was found that the solution can collapse into a
black hole. However, it was subsequently shown that the energies needed for the solution
to collapse to a black hole correspond via the gauge-theory/gravity duality to energies that
are beyond reach in the pure SYM theory. We intend to return to the decoupling-limit issue
for this kind of pure SYM theories [53].
Acknowledgements
We are grateful to M. Bill, J. de Boer, P. Di Vecchia, E. Imeroni, E. Kiritsis, A. Lerda,
E. Lozano-Tellechea, R. Russo and M. Wijnholt for useful discussions, and to P. Bain
for some interesting remarks. T.H. would like to thank the Niels Bohr Institute and the
Institute for Theoretical Physics of Amsterdam University, N.O. and A.W. the Niels Bohr
Institute and Nordita, and M.B. the Department for Theoretical Physics of Turin University
for hospitality while part of this work was carried out. M.B. is supported by a European
Commission Marie Curie Postdoctoral Fellowship under contract number HPMF-CT2000-00847. N.O. is supported by the Stichting FOM.
Appendix A. Details on the solutions and their derivation

In this appendix we present the derivation of the non-extremal solutions discussed in
the main text. We begin by giving the equations that they solve.
A.1. Equations of motion
The equations of motion for the scalar fields encoded in the actions (2.5) for p = 0, 2, 3
and (2.7) for p = 1 turn out to be identical in form:
1 p (1p)

1
(p+2) 2 ,
e
e (G(p+2))2 + F
g =
g
4

1
b,
g = e(1p) e (G(p+2))2 e b
g

1
(p+2) .
g e b = e(1p) G(p+2) F
g
(A.1)
(A.2)
(A.3)
Here we have introduced the notation (G(n) )2 G(n) G(n) n!1 G1 n G1 n .

For the gauge fields A(p+1) and C(p+1) the case p = 1 differs slightly from the others;
while for p = 0, 2, 3, the equations of motion all take the common form

1 p+2 = 0,
g e(1p) F

1 p+2 = 0,
1 g e(1p) e G
1
(A.4)
(A.5)
272
we obtain instead for p = 1 (in form notation)

(3) + bG
(3) = 0,
d F
(3) F
(3) = 0,
F

1 2

d e G(3) + b G(3) = 0.
2
(A.6)
(A.7)
(A.8)
(3) , as this constraint on the solution has

Here we included the self-duality condition for F
12
the status of an equation of motion. For convenience we have introduced the notation
(p+2) = G(p+2) e b F
(p+2) .
G
(A.9)
To the above equations should be added the Einstein equations which we shall give
once we have introduced the spherically symmetric ansatz that we will employ. Let us
first, however, mention that under such an ansatz the self-dual fractional D1-brane case
(3) and
can alternatively be solved by taking a standard electric field-strength ansatz for F
using the equations of motion obtained for p = 1 from the naive action (2.5). The
(3) (as well as the full potential A(2) ), if required, can
dual, magnetic, components of F
be obtained by imposing the self-duality condition at the very end. We will adopt this
(standard) effective procedure below, allowing us to discuss the case p = 1 in parallel with
the others.
A.2. The spherical p-brane ansatz
A general ansatz for a metric possessing the symmetries of the non-extremal solution is
ds 2 = B 2 dt 2 + C 2
p

dx i
2
2
+ F 2 dr 2 + G2 r 2 d4p
,
(A.10)
i=1
where B, C, F and G are functions of the transverse radial coordinate r only. The equations
of motion (A.4) and (A.5) for the two gauge fields A(p+1) and C(p+1) can trivially be
integrated to give13
q1
,
(A.11)
(Gr)4p
q2
r0p = e(1p) e BC p F
G
(A.12)
,
(Gr)4p
where we used the result g = g4p r 4p BC p F G4p . The two hatted charges, q1
and q2 , introduced as a shorthand notation for this appendix, each absorb a p-dependent
factor according to

3 p, p = 0, 1, 2,
q1 = kp q1 , q2 = kp q2 , kp =
(A.13)
1,
p = 3,
r0p = e(1p) BC p F
F
12 Eqs (A.6) and (A.7) are not independent; the self-duality condition is stronger and implies (A.6).
13 In form notation this corresponds to F
(p+2) = (1)p q1 e(1p) d4p , so that d[e(1p) F
(p+2) ] =

2
(1p)
(p+2) = q1 4p .
q1 d 0 and S 4p e
F
273
with q1 and q2 being the charges which appear in the solutions and which are natural from
the physical perspective. Defining
q1
b = 1 + b,
q2
(A.14)
we have
Gr0p = e(1p) e BC p F
q2 b
,
(Gr)4p
(A.15)
so that
e(1p) e (G(p+2))2 = e(1p) e b 2

(p+2) 2 = e(1p)
e(1p) F
q22
,
(Gr)2(4p)
q12
.
(Gr)2(4p)
(A.16)
(A.17)
Introducing the notation

L BC p F 1 (Gr)4p ,
(1p)

(p+2) 2 = e
q12 + q22 e b 2 ,
Y e(1p) e (G(p+2) )2 + F
2(4p)
(Gr)
(A.18)
(A.19)
the scalar-field equations of motion (A.1)(A.3) can be written compactly as

+ (log L) =
1p 2
F Y,
4
(A.20)
+ (log L) = F 2 e(1p) e b 2

L1 Le b = F 2 e(1p) e
q22 2
q22
e b ,
(Gr)2(4p) q12
q12
b.
(Gr)2(4p)
(A.21)
(A.22)
Finally, the Einstein equations read (no summation over indices)

p3
Y,
8
p3
Y,
Ri i =
8

2 1 2 1 q22 2
p3
r
2
Y,
+
R r =F
e b
+ +
2
4
2 q1
8
Rt t =
R =
p+1
Y.
8
(A.23)
(A.24)
(A.25)
(A.26)
Here (for p > 0) i = 1, . . . , p are the spatial world-volume directions while runs over
the 4 p transverse angular coordinates. The Ricci-tensor components for a metric of the
274
form (A.10) are

1
(log B) (log B) (log L) ,
2
F

1
i
R i = 2 (log C) (log C) (log L) ,
F

2
1
R r r = 2 (log B) (log B) + (log F ) (log B)
F
2

+ p (log C) (log C) + (log F ) (log C)
2

+ (4 p) (log Gr) (log Gr) + (log F ) (log Gr) ,

1
F2
R = 2 (log Gr) (log Gr) (log L) + (3 p)
.
F
(Gr)2
Rt t =
(A.27)
(A.28)
(A.29)
(A.30)
A.3. Solving the equations of motion

In order to find the non-extremal versions of the supergravity solutions for fractional
D0- and D2-branes on T 4 /Z2 of Ref. [6] that reduce to Schwarzschild black-brane metrics
for vanishing charges, the natural ansatz to employ is

p

i 2

2
(p3)/4
2
2
ds = H
dx
. (A.31)
f dt +
+ H (p+1)/4 f 1 dr 2 + r 2 d4p
i=1
This ansatz applies equally well for p = 1, while for the three-brane, as mentioned in
Section 2, it is preferable to use the adapted ansatz
2
3

i 2

r
dx +
ds 2 = f dt 2 +
(A.32)
H f 1 dr 2 + r 2 d 2 .
r
i=1
From these metrics we can read off expressions for B, C, F and G in terms of the radial
functions H and f , and plug them in the equations of motion listed above. Solving the so
obtained equations is the objective of the present section.
We start by observing that the harmonic equation (2.10) for f follows from the Einstein
equations (A.23) and (A.24) (for p = 0, Eq. (A.24) is not present and other equations need
to be used instead). Equipped with this result, we combine the dilaton equation (A.20) with
the angular Einstein equation (A.26) to get the equation
+ (log L) = 0,
= log(e H (1p)/4)
(A.33)
L = f r 4p .
and where now

Requiring that
where we have defined
the dilaton behave in a non-singular manner at the horizon we find that must be constant.
Asking furthermore that at infinity we recover flat Minkowski space with vanishing dilaton,
we obtain
e = H (1p)/4 .
(A.34)
For the case p = 1, Eq. (A.33) simply gives that the dilaton is constant, as appropriate for
the string in six dimensions.
275
It remains to solve the scalar-field equations (A.21) and (A.22) together with the radial
Einstein equation (A.25), using also either (A.20) or (A.23). With
F 2Y =
Hf r 2(4p)

q12 + q22 e b 2 ,
(A.35)
the two latter amount to the single equation

e b 2 = ,
(A.36)
where we have introduced

1
4p
4p H
.
fr
2 2Hr
H
q2
q2
q12
(A.37)
From (A.21) and (A.22) we, on the other hand, obtain the equation

q2
L + 22 Le b b = 0,
q1
which can trivially be integrated to

1 q22 2
e +
b
= q3 L1 e ,
2 q12
(A.38)
(A.39)
q3 being a constant of integration. Since L = f r 4p we see that the right-hand side

becomes singular at the horizon unless q3 = 0. Requiring the scalars and b to be nonsingular at the horizon we are hence led to set q3 to zero. Imposing, furthermore, that these
scalars vanish at infinity we arrive at the relation
e = 1 +
1 q22 1 q22 2
b ,
2 q12 2 q12
(A.40)
which when combined with the dilaton equation written in the form (A.36) finally yields
the expressions

1 q22
1 q22 1
e = 1 +
(A.41)
,
1
+
2 q12
2 q12

1 q22 1/2
1 q22 1/2
b = 1 +
(A.42)
1
+
.
2 q12
2 q12
At this point, it remains only to determine the function H by solving Eqs. (A.21)
and (A.25) with the above results as input. To this end, we define the functions h1 and
h2 by

h22
h21
1 q22
1 q22
1+
,
= ,
(A.43)
1
+
=
H
2 q12
2 q12
H
so that

1 q22 2 1 q22 2
H = 1+
h ,
h
2 q12 1 2 q12 2
e =
H
,
h21
h2
b = .
h1
(A.44)
276
Using these expressions the gauge field strengths take the form
kp q2 h1 h2
k q h2
k q
r0p = p 2 1 ,
r0p = p 1 1 .
,
G
F
(A.45)
r 4p H 2
r 4p H 2
r 4p H
Having expressed all fields in terms of h1 and h2 , we rewrite also the remaining equations
of motion which determine these functions. Eq. (A.25) thus reads

1 q22
4p
4p
1 q22

1+
(A.46)
h1 =
h2 ,
h2 h2 +
h1 h1 +
2 q12
r
2 q12
r
Gr0p =
while (A.21) may be written as

q12
H
h2
0 = 2(4p)
+ Hf log
log
f
h1
r

1
1
4p
4p

h1
h2 .
h +
h +
+ Hf
h1 1
r
h2 2
r
(A.47)
Clearly, the simplest non-trivial way to satisfy Eq. (A.46) is by taking both h1 and h2 to be
harmonic, like f , and we will do so here. Taking the boundary conditions at infinity into
account we thus have for p = 0, 1, 2
3p
3p
3p
r0
r
r
,
h1 = 1 13p ,
h2 = 1 23p .
(A.48)
3p
r
r
r
For p = 3 the three harmonic functions governing the non-extremal solution instead take
the form
r
r
r
f = 1 a0 log ,
(A.49)
h1 = 1 a1 log ,
h2 = 1 a2 log ,
r
r
r
where r is a large-radius cut-off. In both cases, we are left with two undetermined
parametersr1,2 and a1,2 , respectivelyand in both cases these parameters are fixed by
Eq. (A.47). However, before analysing this equation, let us derive the solutions for the
gauge potentials C(p+1) and A(p+1) .
Starting with the untwisted sector, we get an integral for C0p from Eq. (A.45) by

recalling that Gr0p = C0p
. For p = 0, 1, 2, this integral can readily be evaluated with
the result
f =1
q2 h3
1
, h3 (h1 + h2 ).
(A.50)
2
r 3p H
To obtain this result we used the fact that h1 and h2 being harmonic implies the identity
C0p =
h3 H +

r
h3 H h3 H = h1 h2 .
3p
With C0...p in hand, we similarly obtain A0p by integrating the equation

q1 1
h2
r q22

A0p = Fr0p b C0p = (3 p) 4p
h3
.
1+
r
H
3 p q12
h1
(A.51)
(A.52)
The solution is found to be

q1 1
A0p = 3p .
r
h1
277
(A.53)
Here we used the identity

H = h21 +

r q22
h3 h1 h2 h1 h2 ,
2
3 p q1
(A.54)
which, like (A.51), follows directly from the form of H and h3 in terms of h1 and h2 .
For the three-brane the situation is somewhat more subtle since there are no analogues
of the first-order differential equations (A.51) and (A.54). Instead, led by the form of the
harmonic functions, we apply the mapping r (3p) log(r /r) to the potentials in (A.50)
and (A.53) above. It is then straightforward to check that the so obtained expressions give
the correct gauge potentials:
C0123 =
r
q2 h3
log ,
H
r
A0123 =
q1
r
log .
h1
r
(A.55)
Let us then, finally, address Eq. (A.47), which for harmonic h1 and h2 immediately
simplifies to

(kp q1 )2
H
h2
(A.56)
+
Hf
log
= 0.
log
f
h1
r 2(4p)
By performing the p-dependent substitutions of variables and parameters
(3p)
3p
r
p = 0, 1, 2,
r0,1,2 p = 0, 1, 2,
=
=
0,1,2
p = 3,
log rr
a0,1,2 p = 3,
this equation reduces to the single, p-independent, equation

H
h 2
2

q1 + H f log
= 0,
log
f
h 1
where
f = 1 0 ,
h 1,2 = 1 1,2 ,

1 q22 2 1 q22 2

H = 1+
h ,
h
2 q12 1 2 q12 2
(A.57)
(A.58)
(A.59)
and the prime now denotes differentiation with respect to . Note that only the unhatted
charges q1 and q2 enter in Eq. (A.58), showing that these are the charges that appear in the
harmonic functions. Inserting the expressions (A.59) we find that the two undetermined
parameters 1 and 2 are required by (A.58) to satisfy
u0
2 =
,
1 = u,
(A.60)
2u 0
where u is a solution of the quartic equation

4 2q12 + q22 u4 8 2q12 + q22 0 u3

+ 2 2q22 02 + 5q1202 2q14 u2 + 2 2q14 0 q12 03 u q14 02 = 0.
(A.61)
278
This equation has the four solutions

2q14 + (q12 + q22 )02 $2 2q12
1

,
u = 0 + $1
2
2 2q 2 + q 2
1
(A.62)
where $1 = 1, $2 = 1 and

1
= q14 + q12 + q22 02 + 04 .
(A.63)
4
Consequently, the solutions to the equations of motion have four branches given by

2q14 + (q12 + q22 )02 $2 2q12
1

1 = 0 + $1
(A.64)
,
2
2 2q 2 + q 2

1
2 = 0 + $1
2
2q14 + (q12 + q22 )02 + $2 2q12

,
2 q22
(A.65)
where, again, $1 = 1 and $2 = 1. There are thus four solutions, which may be
summarized by the expressions given in section 2.3, with 1,2 as in (A.64), (A.65). By
taking the limit 0 0 in the these expressions, one can easily see that the branch which
correctly reduces to the extremal fractional-brane solutioni.e., the branch which satisfies
the boundary conditions imposed by the action (2.8)and which thus represents the nonextremal fractional brane solution, is the branch with $1 = +1 and $2 = +1. This choice
corresponds to (2.17) and (2.18) (p = 0, 1, 2) and (2.28) and (2.29) (p = 3) in the text.
A.4. Some properties of the four branches
To understand the properties of these four branches it is useful to consider the sign of
H at the horizon. (Since, as pointed out in Section 2.3, the three-brane case is special, we
restrict the discussion to p < 3.) Thus, setting r = r0 in Eq. (A.56) we find that

h2 h1 1
q12
H (r0 ) = (3 p) 72p
(A.66)
.

h2 h1
r0
r=r0
Using Eq. (A.48) we then obtain

sign H (r0 ) = sign(r2 r1 ) sign(r0 r1 ) sign(r0 r2 ).
(A.67)
Hence, H (r0 ) is positive if either r0 < r1 < r2 or r1 < r2 < r0 or r2 < r0 < r1 , while it is
negative if either r1 < r0 < r2 or r0 < r2 < r1 or r2 < r1 < r0 .
One can now check that the sign of H (r0 ) cannot be changed within a particular branch.
From Eq. (A.67) we see that this precisely means that r0 , r1 and r2 cannot cross within a
particular branch. In Table 1 we have listed the restrictions on the ranges for r0 , r1 and
r2 , along with the sign of H (r0 ) for the four branches. The solution discussed in the text
corresponds to branch I.
279
Table 1
Properties of the four branches
Branch
$1
$2
Restrictions
sign(H (r0 ))
I
II
III
IV
+1
1
+1
1
+1
+1
1
1
r1 < r0 < r2
r2 < r1 < r0
r2 < r0 < r1
r1 < r2 < r0
+
+
In order to further examine the physics of the four branches obtained above we compute
the ADM mass (per unit p-volume)
Mp =

4p
3p
(4 p)r0 + (3 p) ,
16G6
16G6 =
2 2
V
(A.68)
where 4p denotes the volume of the unit (4 p)-sphere and

=

1 2
2 3p
2 3p
2q
r
+
q
+
q
r
1
2
2
1
2
q12
(A.69)
is the coefficient of the leading 1/r 3p term in the function H in Eq. (A.44). Focusing first
on the four branches at r0 = 0, it is not difficult to see that branches II and III are unphysical
since they both have negative mass, and hence should be discarded. On the other hand, for
branches I and IV we find
(I)
q2 ,
=
(A.70)
2q 2 + q 2 , (IV)
1
2
so that branch I at r0 = 0 has the lowest mass. To obtain (A.70)
we have used (2.22) for
2

2
branch I while for branch IV at r0 = 0 we have h1 = 1 + q1
2q1 + q22 r 3p , h2 = 1.
The charges are
1 =
Q
1
16G6
(p+2) = (3 p)q1
e(1p) F
4p
,
16G6
(A.71)
S 4p
2 =
Q
1
16G6
(p+2) = (3 p)q2
e(1p) e G
4p
16G6
(A.72)
S 4p
2 irrespective of the branch,

(recall that 16G6 = 2 2 /V ). Since the untwisted charge is Q

we conclude that branch I at extremality is BPS (Mp = Q2 ), while branch IV apparently
describes a system that is non-BPS even in the limit r0 = 0.
Although it is presently not clear what the precise physical meaning, if any, of
the supergravity solution of branch IV is, we note that this solution has well-defined
black-brane thermodynamics. Using standard methods of black-hole thermodynamics, we
280
compute the temperature and entropy:14

T=
1
3p
,
4 r0 H (r0 )
S=
4p 4p
r
H (r0 ).
4G6 0
(A.73)
Using the WessZumino term of the world-volume action (2.8) the corresponding chemical
potentials, dual to the charges in (A.71) and (A.72), are

01p )
1 = (A01p + bC
(A.74)
,
2 = C01p
.
r=r0
r=r0
More explicitly, using (A.44), (A.50) and (A.53) the chemical potentials read

q2 h2 (r0 )
q2 h3 (r0 )
1
q1
1 = 3p
+
1 3p
,
r0 h1 (r0 ) q1 h1 (r0 )
r0 H (r0 )
q2 h3 (r0 )
2 = 3p
,
H (r0 )
r
(A.75)
in terms of the harmonic functions hi , i = 1, 2, 3 given in Eqs. (A.48) and (A.50).

i can be
As a check, we note that the first law of thermodynamics dMp = T dS + i d Q
integrated to yield Smarrs formula

1 + 2 Q
2 .
(3 p)Mp = (4 p)T S + (3 p) 1 Q
(A.76)
One may verify explicitly that this law holds since
= 1 q 1 + 2 q 2 ,
where is defined in (A.69). To prove (A.77) one uses the non-trivial identity

1
1
2
2 h2 (r0 )h3 (r0 )
q1 + q2
= ,
3p
H (r0 )
r0 h1 (r0 )
(A.77)
(A.78)
which holds for all branches. To verify this statement one uses the form of H in (A.44)
along with h3 in terms of h1,2 , and subsequently employs the relation (A.60) to eliminate
r2 . The identity then reduces exactly to the quartic equation (A.61) satisfied by r1 .
References
[1] M.R. Douglas, Enhanced gauge symmetry in M(atrix) theory, JHEP 07 (1997) 004, hep-th/9612126.
[2] M.R. Douglas, B.R. Greene, D.R. Morrison, Orbifold resolution by D-branes, Nucl. Phys. B 506 (1997)
84106, hep-th/9704151.
[3] J. Polchinski, Tensors from K3 orientifolds, Phys. Rev. D 55 (1997) 64236428, hep-th/9606165.
[4] M.R. Douglas, G.W. Moore, D-branes, Quivers, and ALE Instantons, hep-th/9603167.
14 We note that the following expressions are algebraically valid for all branches. However, we have already
excluded branches II and III on the grounds of positivity of the mass, while for branch I it is seen from Table 1
that H (r0 ) < 0 and hence the expressions below are not physically meaningful. Alternatively, we have already
argued extensively in the text that branch I does not develop into a black brane.
281
[5] C.V. Johnson, R.C. Myers, Aspects of type IIB theory on ALE spaces, Phys. Rev. D 55 (1997) 63826393,
hep-th/9610140.
[6] M. Frau, A. Liccardo, R. Musto, The geometry of fractional branes, Nucl. Phys. B 602 (2001) 3960, hepth/0012035.
[7] J. Maldacena, The Large N limit of superconformal field theories and supergravity, Adv. Theor. Math.
Phys. 2 (1998) 231252, hep-th/9711200.
[8] S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Gauge theory correlators from noncritical string theory, Phys.
Lett. B 428 (1998) 105, hep-th/9802109.
[9] E. Witten, Anti-de Sitter space and holography, Adv. Theor. Math. Phys. 2 (1998) 253, hep-th/9802150.
[10] 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.
[11] A. Karch, D. Lst, D. Smith, Equivalence of geometric engineering and HananyWitten via fractional
branes, Nucl. Phys. B 533 (1998) 348372, hep-th/9803232.
[12] K. Dasgupta, S. Mukhi, Brane constructions, fractional branes and anti-de-Sitter domain walls, JHEP 07
(1999) 008, hep-th/9904131.
[13] K. Behrndt, About a class of exact string backgrounds, Nucl. Phys. B 455 (1995) 188210, hep-th/9506106.
[14] R. Kallosh, A.D. Linde, Exact supersymmetric massive and massless white holes, Phys. Rev. D 52 (1995)
71377145, hep-th/9507022.
[15] M. Cvetic, D. Youm, Singular BPS saturated states and enhanced symmetries of four-dimensional N = 4
supersymmetric string vacua, Phys. Lett. B 359 (1995) 8792, hep-th/9507160.
[16] J.P. Gauntlett, N. Kim, D. Martelli, D. Waldram, Wrapped fivebranes and N = 2 super YangMills theory,
Phys. Rev. D 64 (2001) 106008, hep-th/0106117.
[17] F. Bigazzi, A.L. Cotrone, A. Zaffaroni, N = 2 gauge theories from wrapped five-branes, Phys. Lett. B 519
(2001) 269276, hep-th/0106160.
[18] P. Di Vecchia, H. Enger, E. Imeroni, E. Lozano-Tellechea, Gauge theories from wrapped and fractional
branes, hep-th/0112126.
[19] M. Bertolini, P. Di Vecchia, M. Frau, A. Lerda, I. Marotta, R. Pesando, Fractional D-branes and their gauge
duals, JHEP 02 (2001) 014, hep-th/0011077.
[20] J. Polchinski, N = 2 gauge-gravity duals, Int. J. Mod. Phys. A 16 (2001) 707718, hep-th/0011193.
[21] M. Grana, J. Polchinski, Gauge/gravity duals with holomorphic dilaton, hep-th/0106014.
[22] M. Bertolini, P. Di Vecchia, M. Frau, A. Lerda, R. Marotta, N = 2 gauge theories on systems of fractional
D3/D7 branes, hep-th/0107057.
[23] M. Bill, L. Gallot, A. Liccardo, Classical geometry and gauge duals for fractional branes on ALE orbifolds,
Nucl. Phys. B 614 (2001) 254278, hep-th/0105258.
[24] P. Bain, Fractional D3-branes in diverse backgrounds, hep-th/0112145.
[25] C.V. Johnson, R.C. Myers, A.W. Peet, S.F. Ross, The enhanon and the consistency of excision, Phys. Rev.
D 64 (2001) 106001, hep-th/0105077.
[26] M. Wijnholt, S. Zhukov, Inside an enhanon: Monopoles and dual YangMills theory, hep-th/0110109.
[27] I.R. Klebanov, M.J. Strassler, Supergravity and a confining gauge theory: Duality cascades and SBresolution of naked singularities, JHEP 08 (2000) 052, hep-th/0007191.
[28] I.R. Klebanov, A.A. Tseytlin, Entropy of near extremal black p-branes, Nucl. Phys. B 475 (1996) 164178,
hep-th/9604089.
[29] A. Buchel, Finite temperature resolution of the KlebanovTseytlin singularity, Nucl. Phys. B 600 (2001)
219234, hep-th/0011146.
[30] A. Buchel, C.P. Herzog, I.R. Klebanov, L. Pando Zayas, A.A. Tseytlin, Non-extremal gravity duals for
fractional D3-branes on the conifold, JHEP 04 (2001) 033, hep-th/0102105.
[31] S.S. Gubser, C.P. Herzog, I.R. Klebanov, A.A. Tseytlin, Restoration of chiral symmetry: A supergravity
perspective, JHEP 05 (2001) 028, hep-th/0102172.
[32] D.Z. Freedman, J.A. Minahan, Finite temperature effects in the supergravity dual of the N = 1: gauge
theory, JHEP 01 (2001) 036, hep-th/0007250.
[33] J.M. Maldacena, C. Nunez, Towards the large N limit of pure N = 1 super YangMills, Phys. Rev. Lett. 86
(2001) 588591, hep-th/0008001.
[34] S.S. Gubser, A.A. Tseytlin, M.S. Volkov, Non-Abelian 4-d black holes, wrapped 5-branes, and their dual
descriptions, JHEP 09 (2001) 017, hep-th/0108205.
282
[35] A. Buchel, On the thermodynamic instability of LST, hep-th/0107102.

[36] J. Polchinski, M.J. Strassler, The string dual of a confining four-dimensional gauge theory, hep-th/0003136.
[37] D.-E. Diaconescu, M.R. Douglas, J. Gomis, Fractional branes and wrapped branes, JHEP 02 (1998) 013,
hep-th/9712230.
[38] M. Bill, B. Craps, F. Roose, Orbifold boundary states from Cardys condition, JHEP 01 (2001) 038, hepth/0011060.
[39] M. Bertolini, P. Di Vecchia, R. Marotta, N = 2 four-dimensional gauge theories from fractional branes,
hep-th/0112195.
[40] P.S. Aspinwall, K3 surfaces and string duality, hep-th/9611137.
[41] H. L, C.N. Pope, K.W. Xu, Black p-branes and their vertical dimensional reduction, Nucl. Phys. B 489
(1997) 264278, hep-th/9609126.
[42] E. Witten, Some comments on string dynamics, hep-th/9507121.
[43] C.V. Johnson, D-brane primer, hep-th/0007170.
[44] P. Merlatti, The enhanon mechanism for fractional branes, hep-th/0108016.
[45] M. Petrini, R. Russo, A. Zaffaroni, N = 2 gauge theories and systems with fractional branes, Nucl. Phys.
B 608 (2001) 145161, hep-th/0104026.
[46] A. Hanany, E. Witten, Type IIB superstrings, BPS monopoles, and three-dimensional gauge dynamics, Nucl.
Phys. B 492 (1997) 152190, hep-th/9611230.
[47] N. Seiberg, New theories in six dimensions and matrix description of M theory on T 5 and T 5 /Z2 , Phys.
Lett. B 408 (1997) 98104, hep-th/9705221.
[48] R. Dijkgraaf, E. Verlinde, H. Verlinde, Notes on matrix and micro strings, Nucl. Phys. Proc. Suppl. 62 (1998)
348, hep-th/9709107.
[49] O. Aharony, M. Berkooz, D. Kutasov, N. Seiberg, Linear dilatons, NS5-branes and holography, JHEP 10
(1998) 004, hep-th/9808149.
[50] I.R. Klebanov, N.A. Nekrasov, Gravity duals of fractional branes and logarithmic RG flow, Nucl. Phys.
B 574 (2000) 263274, hep-th/9911096.
[51] I.R. Klebanov, P. Ouyang, E. Witten, A gravity dual of the chiral anomaly, hep-th/0202056.
[52] M. Bertolini, P.D. Vecchia, M. Frau, A. Lerda, R. Marotta, More anomalies from fractional branes, hepth/0202195.
[53] Work in progress.

Boundary states for AdS2 branes in AdS3

Peter Lee, Hirosi Ooguri, Jongwon Park
California Institute of Technology 452-48, Pasadena, CA 91125, USA
Abstract
We construct boundary states for the AdS2 D-branes in AdS3 . We show that, in the semi-classical
limit, the boundary states correctly reproduce geometric configurations of these branes. We use the
boundary states to compute the one loop free energy of open string stretched between the branes. The
result agrees precisely with the open string computation in hep-th/0106129. 2002 Elsevier Science
B.V. All rights reserved.
1. Introduction
Recently much progress has been made in understanding string theory in AdS3 , threedimensional anti-de Sitter space. With a background of NSNS B-field, the worldsheet is
described by the SL(2, R) WZW model. The spectrum of the WZW model was proposed
in [1], and it was verified by exact computation of the one loop free energy in [2]. This
has lead a proof of the no ghost theorem for string in AdS3 , which had been an outstanding
problem for more than 10 years.1 The unitary string spectrum emerged from these results
agrees with expectations from the AdS/CFT correspondence [3,4] applied to the case of
AdS3 [57]. Correlation functions of vertex operators on the string worldsheet were derived
in [810]. In [11], they are used to compute target space correlation functions of the string
theory and the results were compared with what we expect for the conformal field theory
on the boundary of AdS3 . The correlation functions turned out to have various singularities,
and physical interpretations were given for all of them. It was also shown that the four point
functions of the target space conformal field theory obey the factorization rules when they
should.
E-mail address: ooguri@theory.caltech.edu (H. Ooguri).
1 For a list of historical references, see the bibliography in [1].
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 2 3 9 - 0
284
P. Lee et al. / Nuclear Physics B 632 (2002) 283302
In this paper, we study D-branes whose worldvolume fill AdS2 subspaces of AdS3 [12,
13]. We call such D-branes as AdS2 branes. From the point of view of the CFT dual,
these configurations are supposed to describe conformal field theories in two dimensions
separated by domain walls, each of which preserves at least one Virasoro algebra [14]. The
spectrum of open strings ending on the AdS2 brane was studied in [15].2 It was shown that,
when the brane carries no fundamental string charge, the open string spectrum is exactly
equal to the holomorphic square root of the spectrum of closed strings in AdS3 derived in
[1,2]. It contains short and long strings, and is invariant under spectral flow.
A boundary state [17,18] is a useful concept in studying D-branes. Although the
construction in [18] assumes rationality of conformal field theory, it was shown in [19
22] that the idea can be applied to the non-rational case of the Liouville field theory. More
recently the technique developed in the Liouville theory is applied to the AdS2 branes [23
26]. In this paper, we will closely follow the construction in [23,24]. We will, however,
make a different ansatz about one point functions of closed string operators in a disk
worldsheet, which leads to different expressions for the boundary states. We show that, in
the semi-classical limit, these new boundary states reproduce the geometric configurations
of the AdS2 branes.
Given a pair of boundary states, it is straightforward to compute a partition function on
an annulus worldsheet. This gives the spectrum of the conformal field theory on a strip
where the boundary conditions on the two sides of the strip are specified by the choice of
the boundary states. We can also compute an annulus partition function when the target
space Euclidean time is periodically identified. An integral of this partition function over
the moduli space of worldsheet then gives the one loop free energy at finite temperature.
From this, we can read off the physical spectrum of the open string on the AdS2 brane. We
find that the result agrees with the exact open string computation in [15].
This paper is organized as follows. In Section 2, we will compute one point functions of
closed string operators on a disk with the boundary conditions corresponding to the AdS2
branes. Following [23,24] closely but using a different ansatz, we will derive functional
relations for the one point functions. The general solution to the relations will be given. In
Section 3, we will choose particular solutions which agree with the semi-classical limit of
the AdS2 branes. We will use them to compute annulus amplitudes and derive spectral
densities of open string states. We will also compute one loop free energy to identify
physical states of open strings on the AdS2 branes. We will close this paper with some
discussions on future directions. In Appendix A, we perform some integrals used in this
paper. Various coordinate systems for AdS3 are summarized in Appendix B. We also list
useful formulae of the hypergeometric function and the gamma function in Appendix C.
2 A related work was also done in [16].
285
2. One point functions on a disk

The boundary state can be found by computing the one point functions of closed string
operators on a disk worldsheet. Here we will derive them for the AdS2 brane, following the
approach in [23,24].
2.1. Review of closed string in AdS3
In this paper, we will mostly work in Euclidean AdS3 , which is the three-dimensional
hyperbolic space H3+ given by one of the two branches of
X0
2
2 2 3 2
X1 X2 XE
= R2 ,
embedded in R1,3 with the metric

2
2
2 3 2
ds 2 = dX0 + dX1 + dX2 + dXE
,
(1)
(2)
where R is the curvature radius of H3+ . It is related to AdS3 with the Lorentzian signature
3 = iX 3 . The Euclidean AdS can also be
metric by the analytic continuation X3 XE
3
realized as a right-coset space SL(2, C)/SU(2). Accordingly, the conformal field theory is
the SL(2, C)/SU(2) coset model. In the semi-classical approximation, which is applicable
when the level k R 2 / of the SL(2, C) current algebra is large, states in the theory
are given by normalizable functions on the target space. The space of such states is
decomposed into a sum of principal continuous representations of SL(2, C) with j = 12 +is
(s R+ ). It was shown in [27] that the exact Hilbert space of the coset model at finite value
of k consists of the standard representations of SL(2, C) current algebra whose lowest
energy states are given by principal continuous representations.
Let us introduce a convenient coordinate system (, , ) on H3+ . They are related to
3 ) in (1) in the following way:
the embedding coordinates (X0 , X1 , X2 , XE

3
/R,
= log X0 + XE
=
X2 + iX1
3
X0 + XE
X2 iX1
3
X0 + XE
A point in the coset SL(2, C)/SU(2) has its representative as a Hermitian matrix:

e + e e
,
e
e
and the metric can be written as

ds 2 = R 2 d 2 + e2 d d .
(3)
(4)
(5)
In this theory, there exists an important set of primary fields defined by

z, z ) =
j (x, x;
2j
1 2j
e + | x|2 e
.
(6)
286
The labels x, x are introduced to keep track of the SL(2, C) quantum numbers.3 The
SL(2, C) currents act on it as
Da
w, w),
a = , 3,
j (x, x;
zw
where D a are differential operators with respect to x defined as
w, w)

J a (z)j (x, x;
(7)
,
D3 = x
+ j,
D = x 2
+ 2j x.
(8)
x
x
x
The energymomentum tensor is given by the Sugawara construction, and the worldsheet
conformal weights of this operator is
D+ =
j =
j (j 1)
.
k2
(9)
When j = 12 + is with s R, the operators (6) correspond to the normalizable states in the
coset model. The operators with the SL(2, C) spin j and (1 j ) are not independent but
are related to each other by the following reflection symmetry relation,

4j

2j 1
z, z ) = R(j )
d 2 x x x 1j x , x ; z, z ,
j (x, x;
(10)
where

1
1 2j
k2
12j
R(j ) =
(11)
.

1
1 + 2j
k2
The two and three point functions of these operators have been computed in [810]. The
two point function has the form,

j (x1 , x1 ; z1 , z 1 )j (x2 , x2 ; z2 , z 2 )

B(j )
1
2

(x1 x2 ) j + j 1 +
j j .
=
(12)
|x12 |4j
|z12 |4j
The coefficient B(j ) is given by
B(j ) =
k 2 12j
,

2j 1
(13)
(x)
.
(1 x)
(14)
k2
where
(x) =
The choice of will not play an important role in the discussion of this paper, except for
its behavior in the semi-classical limit
1
(k ),
(15)
3 In the string theory interpretation discussed in Section 3, (x, x)

is identified as the location of the operator
in the dual CFT on S 2 on the boundary of H3+ .
which can be deduced by comparing (10) with its classical counterpart [28].
The three point function is expressed as

j1 (x1 , x1 ; z1 , z 1 )j2 (x2 , x2 ; z2 , z 2 )j3 (x3 , x3 ; z3 , z 3 )
1
= C(j1 , j2 , j3 )
2(
+
)
2(
+3 1 ) |z |2(3 +1 2 )
1
2
3
2
|z12 |
|z23 |
31
1
,
|x12|2(j1 +j2 j3 ) |x23|2(j2 +j3 j1 ) |x31|2(j3 +j1 j2 )
287
(16)
with the coefficient C(j1 , j2 , j3 ) given by

C(j1 , j2 , j3 )
G(1 j1 j2 j3 )G(j3 j1 j2 )G(j2 j3 j1 )G(j1 j2 j3 )
=
,

2 2 j1 +j2 +j3 1 k1
k2 G(1)G(1 2j1 )G(1 2j2 )G(1 2j3 )
(17)
where
G(j ) = (k 2)
j (k1j)
2(k2)
2 (j |1, k 2)2 (k 1 + j |1, k 2),
and 2 (x|1, ) is the Barnes double Gamma function defined by

log 2 (x|1, )

%
%
= lim
(x + n + m)
(n + m)
.
%0 %
n,m=0
n,m=0
(18)
(19)
(n,m)=(0,0)
2.2. Constraints on one point functions

In this subsection, we will derive functional equations satisfied by one point functions
of closed string operators on a disk with boundary conditions corresponding to AdS2
branes. We will follow the discussion in [23,24] closely, except that we use a more general
ansatz for the one point function as we will explain in the next paragraph. Given the one
point functions, we can find boundary states |B for the AdS2 branes. According to [13],
AdS2 branes preserve one half of the current algebra symmetry because of the boundary
condition on the currents,
J a (z) = Ja (z),
at z = z ,
a = 3, +, .
(20)
In the closed string channel, it is translated into the condition on the boundary states as
a

a
Jn + Jn
(21)
|B = 0, a = 3, +, , n Z.
We start with the ansatz that the one point function is of the form

U + (j )
|x x|
2j |z z |2j
U (j )
=
|x x|
2j |z z |2j

j (x, x;
z, z ) =
for Im x > 0,
for Im x < 0.
(22)
288
The z dependence is determined from conformal invariance on the worldsheet and the x
dependence is fixed by conformal invariance on the target space. The parameters (x, x)
can be regarded as coordinates on the boundary of Euclidean AdS3 , which is S 2 . The

AdS2 brane divides S 2 into half, and the upper half plane covers one patch and the lower
half the other. From the point of view of the conformal field theory on the boundary, the
AdS2 introduces an one-dimensional defect on S 2 , across which the two different CFTs are
glued together [14]. Therefore the one point function j may have a discontinuity across
Im x = 0. The expression (22) allows such a discontinuity since the coefficients U + and
U can be different.
The reflection symmetry (10) implies a relation between U + and U . Taking the
expectation value of (10) on both sides, we get

2j 1
|x x |4j
U (j )
= R(j )
d 2 x U + (1 j )
2j
|x x|
|x x |2(1j )
Im x>0

|x x |4j
2
+
d x U (1 j )
.
(23)
|x x |2(1j )
Im x<0
On the left-hand side, we choose U + for Im x > 0 and U for Im x < 0. Setting x =
x1 + ix2 , we can rewrite the above x -integration as

= U (1 j )
dx1
dx2
+ U (1 j )
dx1
0
(x1 2 + (x2 x2 )2 )2j

|2x2 |2(1j )
dx2
(x1 2 + (x2 x2 )2 )2j

.
|2x2 |2(1j )
(24)
By using the Euler integral

1
dx x a1 (1 x)b1 =
0
(a) (b)
,
(a + b)
(25)
we see that for the case x2 > 0, the first term vanishes and only the second term contributes.
On the other hand, when x2 < 0, the second term vanishes and only the first term
contributes. We can summarize the result as
U (j ) = R(j )U (1 j ).
(26)
Introducing f (j )
by

2j 1 1 j
U (j ) = 1
2 f (j ),
k2
(27)
the reflection relation (26) becomes

f (j ) = f (1 j ).
(28)
289
To determine f + (j ), it is useful to consider the bulk two point function of j with the
degenerate field with j = 1/2, which satisfies
x2 1/2 = 0.
(29)
It implies that there are only two terms in its operator product expansion with j as
1 j C (j )j 1 + C+ (j )j + 1 + ,
2
(30)
Here the denote the current algebra descendants. To simplify the equations, in this
subsection, we are suppressing the dependence on x and z. The coefficients C (h) have
been derived in the earlier literature, but let us compute them here for completeness. Let
us take a correlation function of j with both sides of (30). The left-hand side gives

1
1/2 j j = C , j, j .
(31)
2
Using the expression for the three point function and the identity

k1
G(j1 j2 + %)G(j2 j1 + %)
lim
= 2(k 2)
(j1 j2 ),
%0
G(1)G(1 + 2%)
k2
(32)
we find that (31) contains delta functions at j = j 1/2. On the other hand, the right-hand
side of (30) gives

1
1

C (j )B j
1/2 j j j j +
2
2

1
1

+ j j
(33)
C+ (j )B j +
+ .
2
2
Comparing the coefficients of the delta functions, we find
1
1 2j 2
k2
k2
k2
.
C (j ) = 2 2j
C+ (j ) = 2 ,
1
k2
k2 k2
(34)
Given the coefficient C (j ) for the operator product expansion of 1/2 j , one can
deduce a functional relation for U + (j ).4 Consider the following bulk two point function
involving a degenerate field 1/2

1/2 (x; z)j x ; z .
(35)
Using the operator product expansion derived in the above paragraph, we find

1/2 (x; z)j x ; z
3u
|x x |12j |z z | 2 2j
=
|x x |2
|z z |3u
4 Here we take x to be on the upper half plane. A similar relation for U can be derived by considering x
in the lower half plane. Since U are related to each other by the reflection relation, it is sufficient to determine
conditions on U + .
290

1
1
C+ (j )U + j +
F+ x, x ; z, z + C (j )U + j
F x, x ; z, z .
2
2
(36)
Here F are four-point conformal blocks. According to [8], we have

F+ x, x ; z, z

u(1j )
u/2
(1 )
=
F u, 1 2uj, 1 u(2j 1);

u
F 1 u, 1 2uj, 2 u(2j 1); ,
+
1 + u(1 2j )
F x, x ; z, z
(37)

= uj (1 )u/2

1
F 2u(j 1), 1 u, u(2j 1) + 1;
2j 1

+ F 1 u, u, u(2j 1); ,
(38)
where F (a, b, c; z) are the hypergeometric functions 2 F 1 (a, b, c; z),

u=
1
,
k2
(39)
and and are cross ratios of the target space and the worldsheet coordinates,
=
|x x |2
,
|x x |2
|z z |2
.
|z z |2
(40)
It was pointed out in [23] that, when the operator 1/2 approaches the boundary of the
worldsheet, it overlaps only with the identity operator and the operator with j = 1. This
is analogous to the situation in the Liouville theory discussed in [10]. Using this, the two
point function 1/2 j can be evaluated as

1/2 (x; z)j x ; z
3u
|x x |12j |z z | 2 2j
=
|x x |2
|z z |3u

B+ (j )G+ x, x ; z, z + B (j )G x, x ; z, z ,
(41)
where

G+ x, x ; z, z

= uj (1 )3u/2 (1 )F (1 + u, 2uj, 1 + 2u; 1 )
(1 )

2uj
F 1 + u, 1 + 2uj, 2(1 + u); 1 ,
1 + 2u
(42)

G x, x ; z, z
= uj (1 )u/2

1
F 2u(j 1), 1 u, 1 2u; 1
2

+ F 2u(j 1), u, 2u; 1 ,
291
(43)
and u is given by (39). The conformal blocks in the two expressions, (36) and (41), are
related to each other as
(u(2j 1)) (1 + 2u)
(u(2j 1)) (2u)
G +
G+ ,
F =
(44)
(2uj ) (1 + u)
(2u(j 1)) (u)
(1 + u(1 2j )) (1 + 2u)
(1 + u(1 2j )) (2u)
G
G+ .
F+ =
(45)
(1 + 2u(1 j )) (1 + u)
(1 2uj ) (u)
(See Appendix C for some useful identities involving the hypergeometric function.)
According to [19,23,24], B+ is given by
B+ (j ) = A0 U + (j ),
(46)
where A0 is a constant that depends on the boundary condition and can be interpreted as
the fusion coefficient of 1/2 with the boundary unit operator. Likewise, B (j ) term can
be interpreted as the contribution coming from the fusion with j = 1 boundary operator.
Comparing the terms dependent on G+ in (36) and (41), we derive the following functional
equation for U + (j ):

1 (u(2j 1))
(2u)
+
+
A0 U (j ) =
C (j )U j
(u)
2 (2u(j 1))

1 (1 + u(1 2j ))
+
(47)
C+ (j )U j +
.
2
(1 2uj )
Substituting the expression for C (j ) given by (34), this reduces to

1
1
+
1
1
k2
1/2
+
+
+
A0
f j +
.

f (j ) = f j
2
2
2
1 + k2
Given f + (j ) satisfying (48), we get f (j ) by using (28).
It is convenient to introduce a parameter by

1
1 + k2
1/2
A0

= 2 sinh .
2
1 + k2
(48)
(49)
We can regard as parametrizing the boundary condition as A0 depends on it. In fact, the
semi-classical analysis in the next section shows that specifies the location of the AdS2
brane. A general solution to (48) and (28) is then a linear combination of

f (j ) = exp ( + i2n)(2j 1) ,

exp i(2n + 1) (2j 1) ,
(50)
where n Z.
292
3. Boundary states for AdS2 branes

Among the solutions (50), we claim that the one which correctly represents the
boundary condition for a single AdS2 brane is
f (j ) = Ce(2j 1) ,
(51)
where C is a constant independent of j but may depend on and k. From now on, we
neglect this constant and it will not affect the rest of our discussion. In this section, we will
provide evidences for this claim.
Given the f (j ) solution (51), and therefore the one point function U (j ), the
boundary state is expressed as

|B =

d 2x
dj
1
+
2 +iR
Im x>0

+
Im x<0
U+ (1 j )
|j, x, x
I
|x x|
2(1j )

U (1 j )
d x
|j, x, x
I ,
|x x|
2(1j )
2
(52)
where |j, x, x

I is an Ishibashi state built on the primary state |j, x, x.
The coefficients
are chosen so that the one point functions are correctly reproduced as

2j
j, x, x|B
j (x, x;
z, z ) .
= lim |z z |
z
(53)
It can be verified using the two point function (12) and the reflection relation (26) that the
boundary state given by (52) satisfies this condition. In the following, we will examine
aspects of |B .
3.1. Semi-classical analysis
Let us first study the semi-classical limit of the boundary state for the solution (51).
We use the method of [24,29] to identify the D-brane configuration corresponding to the
boundary state. In the semi-classical limit (k ), we can consider a closed string state
|g localized at g H3+ . Namely, it is defined so that
g|j, x, x
= j (x, x|g).
(54)
The overlap of the localized state |g with the boundary state |B can then tell us about
the configuration of the brane. (An analogous idea has been used in [30] to characterize
boundary states for D-branes wrapping on cycles in CalabiYau manifolds.) In the limit
k , the overlap g|B is simplified as
lim g|B

d 2x
dj
1
+
2 +iR
Im x>0

+
Im x<0
293
f+ (1 j )
j (x, x|g)
|x x|
2(1j )
(1
j
)
f
d 2x
j (x, x|g)
,
|x x|
2(1j )
where we used (15). First let us focus on the integral on the upper half plane,

2j 1
dj
1
+
2 +iR

d 2x
x2 >0
f+ (1 j )
(2x2 )2(1j )

1
.
[e (x12 + x22 ( + )x1 + i( )x2 + ) + e ]2j
(55)
After integrating over x1 and changing integration variable x2 = e x2 , we get

2j 1 (2j 12 ) +
f (1 j )
dj 22j
(2j )
2
1
+
2 +iR
2j 2
dx2 x2

2
1 2j
2
i
x2 + ( )e + 1
.
2
(56)
In terms of AdS2 coordinates defined in Appendix B, we have

i
( )e = sinh .
2
The x2 -integral is performed in Appendix A, and the result is

dj
e
f+ (1 j )
(2j 1)
cosh
1
+
2 +iR

=
ds
(57)
e2i()s
,
cosh
(58)
where j = 1/2 + is. Likewise, we can perform the integral over the lower half plane.
Combining the results together, we find that the overlap is given by

lim g|B =
e2i()s
+
ds
cosh

0
e2i()s
=
ds
cosh
ds
e2i()s
cosh
1
1
( ) =
(sinh sinh ).
=
4 cosh
4
(59)
(60)
294
Thus we found that g|B has a support in the two-dimensional subspace at = in

AdS3 . Namely the parameter of the boundary state specifies the location of the AdS2
brane.
In this formalism, the insertion of the identity operator should reproduce the Born
Infeld action in the semi-classical regime. Using the reflection symmetry and taking the
large k limit in (23), we see that

2 .
1 cosh d 2 x|x x|
(61)
The BornInfeld action of AdS2 branes has been computed by independent methods in [13]
to be

SBI cosh d dt cosh .
(62)

2 as the volume of AdS2 branes, the two expressions agree
If we interpret d 2 x|x x|
with the identification = .
The one point function is given by a linear combination of the solutions (50) to the
functional equations discussed in the last section. Since we found that f e(2j 1)
reproduces the correct semi-classical geometry of the AdS2 brane, the coefficients for
all other solutions in (50) should vanish in the semi-classical limit k . In fact we
can make a stronger statement. If we assume the state |g satisfying (54) exists at finite
value of k, then the j -integral to compute overlap g|B is finite only for the particular
solution, f (j ). For all other solutions in (50), the integral is divergent for k > 3. This
suggests that the coefficients in front of the other solutions in (50) should vanish identically
even for finite k.
3.2. Annulus amplitudes
Given the boundary states, the partition function on the annulus worldsheet with the
boundary conditions 1 and 2 on the two boundary of the annulus is computed as
exchanges of closed string states between |B1 and |B2 . If we view this in the open
string channel, one should be able to express it as a sum over states of open string stretched
between the two AdS2 branes at 1 and 2 . These open string states are normalizable.
For the Euclidean AdS2 branes in Euclidean AdS3 , they belong to principal continuous
representations.
We want to compute the following amplitude between two boundary states:
1 B|
c
1/2 L0 +L 0 12
qc

d 2x
dj
1
+
2 +iR
|B2
Im x>0

+
Im x<0
U + 1 (j )U + 2 (1 j )
|x x|
2
s2
U 1 (j )U 2 (1 j ) qck2
d x
|x x|
2
(qc )3
2

=
d x|x x|

0
295
s2
cos 2(1 2 )s 2
qck2
ds
.
s
2
k 2 (qc )3
sinh k2 s
Using the fact

s2
s 2

k2
2 2
4
qck2
q
o
ss s
=
ds sin
,
s
(qc )3
k2
(qo )3
k2
(63)
where
qc = e2ic ,
qo = e2io ,
o =
1
,
c
(64)
we can go to the open string channel:

4 2
= d 2 x |x x|
2
(k 2)3/2
s 2
4

k2
ss
cos(212 s) sin k2

qo
ds
ds
s
2
(qo )3
sinh k2
s
0
0

= d 2 x |x x|
2
2 k2

s 2
2s sinh 2s
qok2
ds
,
cosh((k 2)12 ) + cosh(2s ) (qo )3
where 12 = 1 2 . Obviously, the overall factor involving the x-integral is divergent.

We can identify it as the volume divergence coming from the fact that the AdS2 brane
is non-compact. This divergence also reflects the fact that normalizable states in the
open string sector belong to principal continuous representations which are infinite
dimensional.5
If we define (s) as the density of states in open string Hilbert space belonging
to the current algebra representation whose ground states are in principal continuous
representation with j = 12 + is, the above result shows that it is given by
(s)
s sinh 2s
.
cosh((k 2)12 ) + cosh(2s)
(65)
The spectral density is real and non-negative as it should be. For the case 1 = 2 ,
1 dependence completely disappears. Note, however, we have neglected the overall
5 Here we have focused on the part of the annulus amplitude that scales as the volume of the AdS brane. After
2
the completion of the manuscript, we were informed by B. Ponsot, V. Schomerus and J. Teschner that, under a
certain regularization of the volume divergence, one finds a finite additive term to the annulus partition function
with nontrivial dependence on s and . This would add corrections to the spectral density (65) that do not scale
as the volume of the AdS2 brane.
296
constant C in the one point function (51), which can be -dependent but is independent
of s.
3.3. One loop free energy at finite temperature
In this subsection, we consider AdS2 branes in finite-temperature AdS3 and compute the
partition function by using the boundary states we have constructed. For the boundary state
with = 0, we can directly compare the result with that of Appendix A in [15].
Finite-temperature AdS3 is given by identifying the Euclidean time tE = it in the target
space as,
tE tE + .
(66)
It induces identification of boundary coordinates as well,

|x| |x|e .
(67)
The thermal identification introduces new sectors of closed strings winding around the
compact time direction [25]. As shown in [1], the winding sectors are generated by the
spectral flow
i
g e 2 wR x
ge 2 wL x
(68)
where 1 , 2 , and 3 are the Pauli matrices. If we choose wR = wL = iw/2 with

w Z, the action generates the following change in the string coordinates in Euclidean
AdS3 ,
w
w
,

,
.
(69)
2
2
The string worldsheet remains periodic in + 2 , modulo the identification (66).
This induces the following transformation on the Virasoro generator,
tE tE +
2
3
J0 + kw2
,
2
16 2
2
L 0 L 0 iw J03 + kw2
.
2
16 2
Correspondingly the boundary states include all the winding sectors:

|B; =
|B; ,w .
L0 L0 + iw
(70)
(71)
Here |B; ,w=0 is the boundary state given by (52), except that the x integral is restricted
in the range
e |x| 1,
(72)
which is the fundamental domain of the identification (67). The other states |B; ,w are
given by performing the spectral flow (70). The amplitude we want to compute then is
1/2L0 +L 0 c
12 |B;
1 B; | qc
2
1 ,w B; |
c
1/2 L0 +L 0 12
qc
|B; 2 ,w .
297
(73)
w=
The overlap in the w winding sector can be expressed as

1/2L0 +L 0 c
12 |B;
1 ,w B; | qc
2 ,w
k 2
w2
16 2
= qc
1 ,w=0 B; |
1/2L0 +L 0 c c w(J 3 +J3 )

12 e 2
0
0 |B; ,w=0 ,
qc
2
(74)
where c = itc and 0 = tc w/2. Substituting (52) into this, we find that the right-hand
side is expressed as an integral over x in the range (72). The integration domain is divided
into four regions, depending on the signs of Im x and Im(ei0 x). Combining them together,
we find

k 2

w2
2(1j
)
2
2qc16
dj
d 2 x|x x|
2j xei0 xe
i0
1
+
2 +iR
Im x>0, Im(xei0 )>0,

e <|x|<1

U+1 (j )U+2 (1 j ) + U+1 (j )U2 (1 j ) + U1 (j )U+2 (1 j )

+ U1 (j )U2 (1 j ) sin 0
s 2 /(k2)
qc
11 ( 0 |itc )
The x-integral is performed in Appendix A,
(x-integral) =
(s).
| sin 0 |
Thus the overlap in the winding number w sector is given by
1 ,w B, |
c
1/2 L0 +L 0 12
qc
|B, 2 ,w
ktc 2
8
(75)
w2

.
11 ( tc w |itc )
(76)
Altogether we have,
1 B; |
c
1/2 L0 +L 0 12
qc
|B; 2

w=
ktc 2
8
w2

.
11 ( tc w |itc )
(77)
Note that the dependence on 1 and 2 has disappeared after the j -integral. In fact,
the expression (77) agrees precisely with the result of [15] in the case of 1 = 2 = 0,
i.e., when the branes carry no fundamental string charge. There the annulus amplitude
is evaluated exactly using the iterative Gaussian integral method developed in [2,27]. To
compute the one loop free energy, we need to multiply the partition functions of the ghost
sector and the internal CFT and to integrate the result over the moduli space of the annulus
worldsheet. The method has been explained in detail in the appendix of [15], and we do
not repeat it here.
We have found that the boundary state |B; precisely reproduces the one loop free
energy computed in [15] when 1 = 2 = 0. It turned out that the partition function
does not depend on 1 or 2 at all. The reason for this is unclear and deserves a closer
inspection.
298
4. Conclusion
In this paper, we constructed the boundary states for AdS2 branes in AdS3 , following [23,
24] but using the different ansatz. The boundary states are expressed as linear combinations
of the Ishibashi states as in (52), where the one point functions U (j ) are given by,

2j 1 1 j (2j 1)
2 e
,
U (j ) = 1
(78)
k2
modulo factors independent of j . In the semi-classical approximation, the location of the
brane is given by = in the AdS2 coordinates defined in Appendix B.
From the point of view of the boundary conformal field theories, the AdS2 branes
create defects which connect different conformal field theories while preserving at least
one Virasoro algebra. Since the boundary states allow study of the AdS2 branes beyond
the supergravity approximation, it would be interesting to use them to explore the
correspondence further.
Note added
Toward the completion of this manuscript, we were informed of a related work by B.
Ponsot, V. Schomerus and J. Teschner. Their result of the one point functions on the AdS2
branes agrees with ours shown in (78). We thank them for communicating their results
prior to the publication.
Acknowledgements
We would like to thank C. Bachas, J. de Boer and R. Dijkgraaf for useful discussion on
the holographic interpretation of the AdS2 branes. This research was supported in part by
DOE grant DE-AC03-76SF00098.
Appendix A. Some useful integrals

A.1. Integral used in the semi-classical approximation
We start with the following integral formula,

dx
0
xn
n!
=
,

2
n+3/2
(ax + 2bx + c)
(2n + 1)!! c ( ac + b)n+1
(A.1)
where a 0, c > 0, b > ac. We want to integrate

0
2j 2 2
1 2j
x2 2 sinh x2 + cosh2 2
dx2 x2
.
(A.2)
299
Using n = 2j 2, a = 1, c = cosh2 and b = sinh , we see the

restrictions are clearly
satisfied for all values of R. Using the fact that (1/2 + n) = 2n (2n 1)!!, we
have

0
1 2j
2j 2 2
dx2 x2
x2 2 sinh x2 + cosh2 2
222j (2j 1) e(2j 1)

.
=

2j 12 cosh
(A.3)
A.2. Integral used for finite temperature free energy

The integral we want to compute is:

2(1j )
d 2 x |x x|
2j xei0 xe
i0
.
(A.4)
Im x>0, Im(xei0 )>0,

e <|x|<1
Use polar coordinates such that x = rei . Then we get,

1
=
e
dr
r
2 sin 0

0

1
sin( + 0 )

d sin sin( + 0 ) exp 2is ln

sin
0
d s eis s
| sin 0 |
(s),
=
| sin 0 |
where, in the second line, we changed the variable to s defined by

sin( + 0 )
.

s = 2 ln

sin
(A.5)
(A.6)
(A.7)
Appendix B. Coordinate systems for AdS3

The space AdS3 is defined as the hyperboloid
0 2 1 2 2 2 3 2
X X X + X
= R2 ,
embedded in R2,2 . The metric
2
2
2
2

ds 2 = dX0 + dX1 + dX2 dX3
(B.1)
(B.2)
on R2,2 induces a metric of constant negative curvature on AdS3 . The quantity R that
appears in (B.1) is the anti-de Sitter radius; for convenience, we set R = 1. In addition, to
300
avoid closed time-like curves, we work not with the hyperboloid (B.1) itself, but with its
universal cover.
The two coordinate systems we use most extensively are global coordinates and
AdS2 coordinates. The global coordinates (, , ) are defined by
X0 + iX3 = cosh eit ,
X1 + iX2 = sinh ei .
(B.3)
The range of the radial coordinate is 0 < ; the angular coordinate ranges over
0 < 2 ; and the global time coordinate t may be any real number. The AdS3 metric in
global coordinates is
ds 2 = cosh2 dt 2 + d 2 + sinh2 d 2 .
(B.4)
The AdS2 coordinate system is convenient to use when considering AdS2 branes in
AdS3 . They are defined by
X1 = cosh sinh ,
X2 = sinh ,
X0 + iX3 = cosh cosh eit .
(B.5)
All three AdS2 coordinates range over the entire real line. In this parametrization, the fixed
slices have the geometry of AdS2 . The AdS3 metric in AdS2 coordinates takes the form

ds 2 = d 2 + cosh2 cosh2 dt 2 + d2 ,
(B.6)
the quantity in parentheses is the metric of the AdS2 subspace at fixed . The
transformation between global and AdS2 coordinates is
sinh = sin sinh ,
cosh sinh = cos sinh .
(B.7)
The global time t is the same in both coordinate systems.

The space AdS3 is the group manifold of the group SL(2, R). A point in AdS3 is given
by the SL(2, R) matrix

0
X + X1 X2 + X3
.
g=
(B.8)
X2 X3 X0 X1
In the global coordinate system,

cos t cosh cos sinh
g=
sin t cosh + sin sinh
sin t cosh + sin sinh

cos t cosh + cos sinh

.
(B.9)
In this paper, we work mostly in the Euclidean rotation of this geometry, which is given
by t tE = it in the above.
Appendix C. Some useful formulae
In this appendix, we list some useful formulae involving the hypergeometric function
and the gamma function.
The hypergeometric function, 2 F 1 (, , ; z), enjoys following useful identities:
2 F1 (, , ; z) = (1 z)
2 F 1 (
, , ; z).
(C.1)
301
The Gauss recursion formulae are given by

2 F 1 (, , ; z) + ( ) 2 F 1 ( + 1, , + 1; z)
(1 z) 2 F 1 ( + 1, + 1, + 1; z) = 0,
(C.2)
2 F 1 (, , ; z) 2 F 1 ( + 1, , ; z) + z 2 F 1 ( + 1, + 1, + 1; z)
= 0,
(C.3)
2 F 1 (, , ; z) ( ) 2 F 1 (, , + 1; z) 2 F 1 (, + 1, + 1; z)
= 0.
(C.4)
Under z 1 z, we have
2 F1 (, , ; 1 z)
( ) ( )
2 F 1 (, , 1 + + ; z)
( ) ( )
( ) ( + )
+ z
2 F 1 ( , , 1 + ; z).
() ()
(C.5)
Lastly, under z 1/z, we get

2 F1 (, , ; z)

( ) ( )
1
1
=
2 F 1 , 1 + , 1 + ;
() ( )
z
z

1
1
( ) ( )
.
,
1
+
,
1
+
;
+
F
2 1
() ( )
z
z
(C.6)
We list some useful identities involving the gamma, (z), function:

(1 + z) = z (z),

1
= ,
2
(1 z) (z) =
(C.7)
(C.8)
,
sin(z)
x
(1 + ix) (1 ix) =
,
sinh(x)

22z1
1
(2z) = (z) z +
.
2
(C.9)
x R
(C.10)
(C.11)
References
[1] J. Maldacena, H. Ooguri, Strings in AdS3 and SL(2, R) WZW model. Part 1: The spectrum, J. Math. Phys. 42
(2001) 29292960, hep-th/0001053.
[2] J. Maldacena, H. Ooguri, J. Son, Strings in AdS3 and the SL(2, R) WZW model. Part 2: Euclidean black
hole, J. Math. Phys. 42 (2001) 29612977, hep-th/0005183.
302
[3] J. Maldacena, The large N limit of superconformal field theories and supergravity, Adv. Theor. Math. Phys. 2
(1998) 231252, hep-th/9711200.
[4] O. Aharony, S.S. Gubser, J. Maldacena, H. Ooguri, Y. Oz, Large N field theories, string theory and gravity,
Phys. Rep. 323 (2000) 183386, hep-th/9905111.
[5] J. Maldacena, A. Strominger, AdS3 black holes and a stringy exclusion principle, JHEP 12 (1998) 005,
hep-th/9804085.
[6] A. Giveon, D. Kutasov, N. Seiberg, Comments on string theory on AdS3 , Adv. Theor. Math. Phys. 2 (1998)
733780, hep-th/9806194.
[7] N. Seiberg, E. Witten, The D1/D5 system and singular CFT, JHEP 04 (1999) 017, hep-th/9903224.
[8] J. Teschner, On structure constants and fusion rules in the SL(2, C)/SU(2) WZNW model, Nucl. Phys.
B 546 (1999) 390422, hep-th/9712256.
[9] J. Teschner, Operator product expansion and factorization in the H3+ WZNW model, Nucl. Phys. B 571
(2000) 555582, hep-th/9906215.
[10] V. Fateev, A. Zamolodchikov, A. Zamolodchikov, unpublished notes.
[11] J. Maldacena, H. Ooguri, Strings in AdS3 and the SL(2, R) WZW model. Part 3: Correlation functions,
hep-th/0111180.
[12] S. Stanciu, D-branes in an AdS3 background, JHEP 09 (1999) 028, hep-th/9901122.
[13] C. Bachas, M. Petropoulos, Anti-de Sitter D-branes, JHEP 02 (2001) 025, hep-th/0012234.
[14] C. Bachas, J. d. Boer, R. Dijkgraaf, H. Ooguri, Permeable conformal walls and holography, hep-th/0111210.
[15] P. Lee, H. Ooguri, J. w. Park, J. Tannenhauser, Open strings on AdS2 branes, Nucl. Phys. B 610 (2001) 348,
hep-th/0106129.
[16] P.M. Petropoulos, S. Ribault, Some remarks on anti-de Sitter D-branes, JHEP 07 (2001) 036, hepth/0105252.
[17] N. Ishibashi, The boundary and crosscap states in conformal field theories, Mod. Phys. Lett. A 4 (1989) 251.
[18] J.L. Cardy, Boundary conditions, fusion rules and the Verlinde formula, Nucl. Phys. B 324 (1989) 581.
[19] V. Fateev, A.B. Zamolodchikov, A.B. Zamolodchikov, Boundary Liouville field theory. I: Boundary state
and boundary two-point function, hep-th/0001012.
[20] A.B. Zamolodchikov, A.B. Zamolodchikov, Liouville field theory on a pseudosphere, hep-th/0101152.
[21] J. Teschner, Remarks on Liouville theory with boundary, hep-th/0009138.
[22] B. Ponsot, J. Teschner, Boundary Liouville field theory: Boundary three point function, hep-th/0110244.
[23] A. Giveon, D. Kutasov, A. Schwimmer, Comments on D-branes in AdS3 , Nucl. Phys. B 615 (2001) 133168,
hep-th/0106005.
[24] A. Parnachev, D.A. Sahakyan, Some remarks on D-branes in AdS3 , JHEP 10 (2001) 022, hep-th/0109150.
[25] A. Rajaraman, M. Rozali, Boundary states for D-branes in AdS3 , hep-th/0108001.
[26] Y. Hikida, Y. Sugawara, Boundary states of D-branes in AdS3 based on discrete series, hep-th/0107189.
[27] K. Gawedzki, Noncompact WZW conformal field theories, hep-th/9110076.
[28] J. Teschner, The mini-superspace limit of the SL(2, C)/SU(2) WZNW model, Nucl. Phys. B 546 (1999)
369389, hep-th/9712258.
[29] J. Maldacena, G.W. Moore, N. Seiberg, Geometric interpretation of D-branes in gauged WZW models,
JHEP 07 (2001) 046, hep-th/0105038.
[30] H. Ooguri, Y. Oz, Z. Yin, D-branes on CalabiYau spaces and their mirrors, Nucl. Phys. B 477 (1996)
407430, hep-th/9606112.

A note on noncommutative D-brane actions

Calin Ciocarlie, Peter Lee , Jongwon Park
California Institute of Technology 452-48, Pasadena, CA 91125, USA
Abstract
We use the nonabelian action of N coincident D(1) branes in constant background fields, in
the N limit, to construct noncommutative D-brane actions in an arbitrary noncommutative
description and comment on tachyon condensation from this perspective. 2002 Elsevier Science
B.V. All rights reserved.
1. Introduction and review

It is a well known result that from the nonabelian BornInfeld action of infinitely many
D(1) instantons one can construct the background independent, = B, description
of noncommutative D-branes. Similarly, in [13], the same type of equivalence was
shown for the ChernSimons terms. In [4], it has been remarked that by placing D(1)
instantons in a constant B-field one can construct noncommutative D-branes with arbitrary
noncommutativity. In this note, we clarify this point by starting from the action of
N coincident D(1) instantons in a constant B-field as given by [57]. We show
that such actions lead us to construct D-brane actions in an arbitrary noncommutative
description. The map relating the BornInfeld terms is seen to be consistent with the map
relating the ChernSimons terms. We use this result to study tachyon condensation on a
noncommutative D-brane with arbitrary .
We will now review some relevant results of [5,6] and [8]. For concreteness, we will
assume Euclidean spacetime and maximal rank constant B-field along the directions of
a Dp-brane. We use the convention where 2 = 1. Then the world-volume Dp-brane
* Corresponding author.
E-mail addresses: calin@theory.caltech.edu (C. Ciocarlie), peter@theory.caltech.edu (P. Lee),

jongwon@theory.caltech.edu (J. Park).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 2 3 6 - 5
304
C. Ciocarlie et al. / Nuclear Physics B 632 (2002) 303310
action can be described in noncommutative variables, i.e., [x i , x j ] = i ij , as

(2)(1p)/2

+ ),
SBI =
d p+1 x det(G + F
Gs
(1)
where the product is implicit in the above equation. For abelian and constant F , the
are given by
SeibergWitten transformations relating F to F
1
1
=
F.
(2)
,
F

1 + F
1 F
For every closed string background characterized by the NSNS 2-form B, the closed
string metric g, and the closed string coupling constant gs , there is a continuum of
descriptions given by a choice of . The open string metric G, the open string coupling
constant Gs and the noncommutativity parameter can be expressed in terms of closed
string variables as follows:

F =F
1
1
+ =
,
G+
g+B

det(G + ) 1/2
Gs = gs
.
det(g + B)
(3)
Finally, let us review the main results of [5,6]. The nonabelian BornInfeld action
describing N (Euclidean) coincident Dp-branes in a closed string background defined by
, B and g is
SBI =
(2)(1p)/2
g
s

d p+1 Str e det P [Eab + Eai (M 1 )ij Ej b ] + Fab det Mji ,
(4)
where E g + B and is the bulk dilaton. Furthermore, i, j are indices for the transverse
coordinates, a, b are indices for the coordinates parallel to the D-brane, and the Xs are
N N matrices representing the transverse displacements expressed in the static gauge.
We also defined1

Mji ji i Xi , Xk Ekj .
(5)
For the nonabelian ChernSimons action, we have

C (n) eB
eF ,
SCS = p Str P ei(iX iX )
(6)
where p is the RR charge of a Dp-brane. In the aforementioned actions, the bulk fields
should be considered functionals of the N N matrices X, and the trace should be
symmetrized between all expressions of the form Fab , Da Xi , [Xi , Xj ], and Xk . However,
1 Unlike in [5], we used the convention F = A A i[A , A ] in order to be consistent with the
a b
a
ab
b a
b
in [8].
definition of F
305
since we are only going to consider D(1) instantons in constant background fields, these
details are irrelevant for our purposes.
More precisely, in the next two sections we consider an infinite number of D(1)
instantons with = 0 and where g and B are constants. The presence of the B field
will allow us to construct D-brane actions in an arbitrary noncommutative description.
In Section 2, we show that the BornInfeld action of D(1) instantons in a constant B
field naturally leads to NC BornInfeld action, where the B field is identified as
B = B + 1 for arbitrary noncommutativity parameter . Having shown this, the
nonabelian generalization of the ChernSimons action for an infinite number of D(1)
instantons should correspond to the NC ChernSimons action in the same noncommutative
description as the BI action. This fact is confirmed in Section 3. In Section 4, we comment
on tachyon condensation using the connection to the matrix model. We find that in the
presence of a constant B-field, the vacuum becomes noncommutative.
2. BornInfeld action
In this section, we follow the line of thought in [9] and derive the equivalence of the
nonabelian BI action of an infinite number of D(1) instantons and the BI action of a
noncommutative Dp-brane in a general noncommutative description. First consider the
nonabelian BI action of N D(1) branes (N ) in a constant B -field

2
SBI =
(7)
Str detij i j i(g + B )ik Xk , Xj .
gs
We are interested in a particular classical configuration given by
i j
x , x = i ij .
(8)
The degrees of freedom on the noncommutative Dp-brane arise by expanding the matrix
variable Xi around this classical configuration as follows:
j .
Xi = x i + ij A
Then, we have

ij ,
i Xi , Xj = F
(9)
(10)
where

ij = i 1 x k , A
i , A
j + i 1 x k , A
i i A
j .
F
ik
jk
(11)
We can reexpress Tr over the Hilbert space as an integral over the volume of noncommutative space by replacing

1
Tr
(12)
d p+1 x,
(2)(p+1)/2 Pf
where Pf is the Pfaffian. We write the action in terms of new variables,
p+1

(2)(1p)/2
d
x
)
SBI =
det 1 (g + B )( F

gs
Pf
(13)
306
(2)(1p)/2
gs
(2)(1p)/2
=
gs

d p+1 x

d p+1 x

1)
det 1 (g + B )( F
(14)

.
det g + B + 1 (g + B ) F
(15)
We would like to compare this with the BI action of a noncommutative Dp-brane in

a description with the same noncommutativity parameter which appears in the above
action. The NC BI action for a Dp-brane is

(2)(1p)/2
+ ).
d p+1 x det(G + F
SNC BI =
(16)
Gs
Reexpressing it in terms of closed string variables by using the relations (3) gives us

(2)(1p)/2 det(g + B)
)
SNC BI =
(17)
d p+1 x det(G + + F
gs
det(G + )

1
(2)(1p)/2
p+1
F
x det g + B + (g + B)
d
=
(18)
gs
G+

(2)(1p)/2
.
=
(19)
d p+1 x det g + B + 1 (g + B) F
gs
We observe that (7) agrees with (16) once we make the following identifications:
= ,
= F
,
F
B = B + 1 .
(20)
Notice that here is a free parameter, not fixed to be B 1 as in [9]. By identifying B in

the nonabelian action for N D(1) instantons (N ) with B 1 , we can go to the
noncommutative description of Dp-brane with arbitrary noncommutativity parameter . It
is interesting to note that takes the following form in matrix-model-like variables:

1
,
= 1 1 + (g + B )1
(21)
A
where A denotes antisymmetrization.
3. ChernSimons action
If the nonabelian BI action for an infinite number of D(1) instantons in a constant B
field gives rise to the NC BI action with B = B + 1 and noncommutativity parameter ,
then we should expect the same identification relates the ChernSimons term of the
nonabelian action with that of the NC theory. This is precisely what occurs, and the Chern
Simons action for a Dp-brane with a constant B field and noncommutativity can be
expressed as the nonabelian CS action for an infinite number of D(1) branes in a constant
B field given by [5]

2

SCS =
(22)
Str ei(iX iX )
C (n) eB ,
B = B 1 .
gs
n
307
Here iX acts on an n-form (n) as

iX (n) =
1
X1 (n)
dx 2 dx n .
1 2 ...n
(n 1)!
(23)
This provides a natural explanation of the rather surprising result recently derived by [2],
where they express an arbitrary NC CS action in terms of matrix-model like variables,
which turns out to be identical to (22). For simplicity, we follow the proof of [10] to show
that the nonabelian action gives rise to the NC action for D9-branes, where we can ignore
transverse scalar fields. In that case, the NC CS action is given by [2,10]

1
)
SNC CS = 9
(24)
det(1 F
C (n) eB+F (1 F ) ,
x
,
where 9 = (2)4 /gs is the RR charge of a BPS D9-brane. In terms of Q = + F
(24) can be expressed as

1
)
det(1 F
C (n) eB eQ .
SNC CS = 9
(25)
x
The nonabelian CS action for an infinite number of D(1) instantons (22) naturally leads
to the NC CS action for Dp-branes (24). Expanding the action (22) and using the fact that
i[X, X] = Q give terms of the form

2
(10 2r)!
Qi2r+1 i2r+2 Qi9 i10
Tr 5r
gs
2 (s r)!(5 r)!2sr (10 2s)!

,
B[i 2r+1 i2r+2 Bi2s1 i2s Ci(102s)
(26)
2s+1 ...i10 ]
where [ ] denotes antisymmetrization and 5 s > r 0. Employing the identity (12),
one gets

(10 2r)!
Qi2r+1 i2r+2 Qi9 i10
9 d 10 x 5r
2 (5 r)!(s r)!2sr (10 2s)! Pf
B[i 2r+1 i2r+2 Bi2s1 i2s Ci2s+1 ...i10 ] .
(102s)
(27)
Finally, the above expression can be simplified to

Pf Q(1)r
1
, i1 ...i10 Q1
9 d 10 x
i1 i2 Qi2r1 i2r
Pf 2s r!(s r)!(10 2s)!
B[i 2r+1 i2r+2 Bi2s1 i2s Ci2s+1 ...i10 ] .
(102s)
(28)
One can immediately see that (28) are the terms coming from the expansion of (24). We
have shown that our claim holds for the special case p = 9. The general case has been
already considered in [2].
Up to now, we have restricted the RamondRamond fields to be constants, but we can
generalize our procedure to the case where the RamondRamond fields are varying by
308
writing the fields as Fourier transforms2 such that

2

SCS =
C (n) (q)eB eiqX ,
d 10 q Str ei(iX iX )
gs
n
B = B 1 .
(29)
To conclude, motivated by the identification relating the nonabelian BI action of D(1)

instantons to the BI action of Dp-branes in the last section, we have proposed and
verified that the NC CS action of a Dp-brane with arbitrary noncommutativity and varying
RamondRamond fields can be derived from considering the nonabelian CS action for an
infinite number of D(1) branes after identifying B = B 1 .
4. Tachyon condensation
In this section, we study tachyon condensation [12,13] in open string theory via the
matrix model connection as in [9,14]. Using the results of the previous sections, we
study this from an arbitrary noncommutative description. The effective action on a single
unstable noncommutative D-brane is

(2)(1p)/2
p+1
+ ) + G f (T )Gij Di T Dj T + .
x V (T ) det(G + F
d
Gs
(30)
In terms of matrix-model-like variables, is given by (21), while the open string metric is
1 where S denotes symmetrization. Thus the effective action can
G = 1 (g + B )1
S
be written as

j

2
Tr V (T ) detij i i(g + B )ik Xk , Xj
gs

i
j
g
1
f (T )
(31)
g B g B ij X , T X , T .
g B
Following [9], we assume that V (T ) has a unique minimum at T = Tc (Tc proportional
to the unit matrix). The end-point of tachyon condensation obtained by minimizing the
BornInfeld term3 is characterized by X = Xc satisfying

ij
i j
Xc , Xc = i (g + B )1 A .
(32)
We expect this minimum to be exact, in the sense that even if there are corrections to the
symmetrized-trace proposal for nonabelian BornInfeld action, these corrections are of the
type [F, F ] and DF , so they are irrelevant for our solution. For B = 0, (32) implies that
the vacuum is commutative. In the presence of a nonzero constant NSNS field it changes
to a noncommutative state given by (32).
2 See [11] for how to relate the currents expressed in matrix model language to those in noncommutative
gauge theory.
3 One can write det(1 i(g + B )[X, X]) = det(g + B ) det((g + B )1 + (g + B )1 i[X, X]).
S
A
309
5. Discussion
It is interesting to analyze the SeibergWitten limit in this context. In this limit, using
the conventions of [8], we have
g ,,
, 1/2 ,

B = B (0) + ,B (1) + O , 2 ,
1 = B (0) + O(,).
(33)
Keeping track of , we see that 2 B = 2 (B 1 ) scales as O(, 3/2 ) and [X,X]

2
goes like O(, 1/2 ). Hence, in the SeibergWitten limit, the BornInfeld action reduces to

i j
k l

1
iBij Tr Xi , Xj
g
g
Tr
X
,
X
,
X
SBI =
X
ik
j
l
gs
2(2 )2
1/2
+ constant,
+O ,
(34)
where the second term is the usual potential of the matrix model. Furthermore, the
nonabelian ChernSimons action takes the standard matrix model form

2
2i (iX iX )
(n)
SCS =
(35)
Str e
C
,
gs
n
where we assumed appropriate scaling of RR potentials, C (n) , such that the limit is well
defined.
Finally, lets remark that since B = B + 1 , the freedom of description of NC
Dp-branes translates in the matrix model like variables into how one separates the B-field
into the external part B and the internal part 1 . The internal part, 1 , is generated by
the configuration of D(1) instantons and B corresponds to the external field imposed on
them.
Acknowledgements
We have greatly benefitted from discussions with Iosif Bena, Iouri Cheplev, Sangmin
Lee, John Schwarz and especially Yuji Okawa.
References
[1] S. Mukhi, N.V. Suryanarayana, Gauge-invariant couplings of noncommutative branes to RamondRamond
backgrounds, JHEP 0105 (2001) 023, hep-th/0104045.
[2] H. Liu, J. Michelson, -Trek III: the search for RamondRamond couplings, Nucl. Phys. B 614 (2001)
330366, hep-th/0107172.
[3] H. Liu, J. Michelson, RamondRamond couplings of noncommutative D-branes, Phys. Lett. B 518 (2001)
143152, hep-th/0104139.
[4] N. Ishibashi, S. Iso, H. Kawai, Y. Kitazawa, Wilson loops in noncommutative YangMills, Nucl. Phys. B 573
(2000) 573593, hep-th/9910004.
[5] R.C. Myers, Dielectric branes, JHEP 9912 (1999) 022, hep-th/9910053.
310
[6] A.A. Tseytlin, On non-abelian generalisation of BornInfeld action in string theory, Nucl. Phys. B 501
(1997) 4152, hep-th/9701125.
[7] W. Taylor, M.V. Raamsdonk, Multiple Dp-branes in weak background fields, Nucl. Phys. B 573 (2000)
703734, hep-th/9910052.
[9] N. Seiberg, A note on background independence in noncommutative gauge theories, matrix model and
tachyon condensation, JHEP 0009 (2000) 003, hep-th/0008013.
[10] S. Mukhi, N.V. Suryanarayana, RamondRamond couplings of noncommutative branes, hep-th/0107087.
[11] Y. Okawa, H. Ooguri, An exact solution to SeibergWitten equation of noncommutative gauge theory, Phys.
Rev. D 64 (2001) 046009, hep-th/0104036.
[12] A. Sen, Descent relations among bosonic D-branes, Int. J. Mod. Phys. A 14 (1999) 40614078, hepth/9902105.
[13] A. Sen, Universality of the tachyon potential, JHEP 9912 (1999) 027, hep-th/9911116.
[14] R. Gopakumar, S. Minwalla, A. Strominger, Symmetry restoration and tachyon condensation in open string
theory, JHEP 0104 (2001) 018, hep-th/0007226.

Violation of sum rules for twist-3 parton

distributions in QCD
Matthias Burkardt a , Yuji Koike b
a Department of Physics, New Mexico State University, Las Cruces, NM 88003, USA
b Department of Physics, Niigata University, Niigata 950-2181, Japan
Received 3 December 2001; accepted 5 April 2002
Abstract
Sum rules for twist-3 distributions are reexamined. Integral relations between twist-3 and twist-2
parton distributions suggest the possibility for a -function at x = 0. We confirm and clarify this
result by constructing hL and h3L (quarkgluon interaction dependent part of hL ) explicitly from
their moments for a one-loop dressed massive quark. The physics of these results is illustrated by
calculating hL (x, Q2 ) using light-front time-ordered pQCD to O(S ) on a quark target. A (x) term
is also found in e(x, Q2 ), but not in gT (x, Q2 ), to this order in O(S ). 2002 Elsevier Science B.V.
All rights reserved.
PACS: 11.55.Hx; 12.38.-t
1. Introduction
Ongoing experiments with polarized beams and/or targets conducted at RHIC, HERMES and COMPASS, etc., are providing us with important information on the spin distribution carried by quarks and gluons in the nucleon. They are also enabling us to extract
information on the higher twist distributions which represent the effect of quarkgluon
correlations. In particular, the twist-3 distributions gT (x, Q2 ) and hL (x, Q2 ) are unique in
that they appear as a leading contribution in some spin asymmetries: for example, gT can
be measured in the transversely polarized leptonnucleon deep inelastic scattering and hL
appears in the longitudinal-transverse spin asymmetry in the polarized nucleonnucleon
DrellYan process [1]. The purpose of this paper is to reexamine the validity of the sum
rules for these twist-3 distributions.
E-mail address: burkardt@nmsu.edu (M. Burkardt).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 2 6 3 - 8
312
M. Burkardt, Y. Koike / Nuclear Physics B 632 (2002) 311329
The complete set of the twist-3 quark distributions in our interest are given as the lightcone corelation functions in a parent hadron with momentum P , spin S and mass M:

d ix
e P S| (0) 5 (n)Q2 |P S
2

= 2 g1 x, Q2 p (S n) + gT x, Q2 S + M 2 g3 x, Q2 n (S n) ,
(1.1)

d ix
e P S| (0) i5 (n)Q2 |P S
2

= 2 h1 x, Q2 S p S
p /M + hL x, Q2 M p n p n (S n)

+ h3 x, Q2 M S n S
(1.2)
n ,

d ix
(1.3)
e P S| (0)(n)Q2 |P S = 2Me x, Q2 ,
2
where two light-like vectors p and n are introduced by the relation p2 = n2 = 0, n+ =
2
p = 0, P = p + M2 n and S is decomposed as S = (S n)p + (S p)n + S .

The variable x represents the partons light-cone momentum fraction and each function
has a support for x on [1, 1]. The antiquark distributions g1,T ,3 (x, Q2 ), h 1,L,3(x, Q2 )
are obtained by the replacement of into its charge conjugation field C T in (1.1)
(1.3) and are related to the quark distributions as g1,T ,3 (x, Q2 ) = g1,T ,3 (x, Q2 ) and
h 1,T ,3 (x, Q2 ) = h1,T ,3(x, Q2 ). The sum rules in our interest are obtained by taking
the first moment of the above relations. For example, from (1.1), one obtains

P S| (0) 5 (0)Q2 |P S
1
=2

dx g1 x, Q2 p (S n) + gT x, Q2 S + M 2 g3 x, Q2 n (S n) . (1.4)
From rotational invariance, it follows that the left-hand side of (1.4) is proportional to the
spin vector S and thus g1,T ,3 (x, Q2 ) must satisfy
1
dx g1 x, Q
1
1
=

dx gT x, Q2 ,
(1.5)
dx g1 x, Q
1
=2

dx g3 x, Q2 .
(1.6)
The same argument for (1.2) leads to the sum rule relations for h1,L,3 (x, Q2 ):
1
dx h1 x, Q
1
1
=
1

dx hL x, Q2 ,
(1.7)
1
dx h1 x, Q
1
1
=2

dx h3 x, Q2 .
313
(1.8)
The sum rule (1.5) is known as BurkhardtCottingham sum rule [2] and (1.7) was first
derived in Refs. [35]. Since the twist-4 distributions g3 , h3 are unlikely to be measured
experimentally, the sum rules involving those functions (1.6) and (1.8) are practically
useless and will not be addressed in the subsequent discussions. Since the left-hand side
of (1.5) and (1.7) are, respectively, the axial charge and the tensor charge, the integral
itself in the right-hand side of (1.5)(1.8) are finite. As is clear from the above derivation,
these sum rules are mere consequences of the rotational invariance and there is no doubt
in its validity in mathematical sense. However, if one try to confirm those sum rules by
experiment, a great care is required to perform the integral including x = 0. For example,
in DIS, x is identified as the Bjorkens variable xB = Q2 /2P q and x = 0 corresponds
to P q and this limit can never be achieved in the rigorous sense. Accordingly, if
hL (x, Q2 ) has a contribution proportional to (x) and h1 (x, Q2 ) does not, experimental
measurement can never confirm the sum rule (1.7) but would rather claim the violation of
the sum rule. Such a contribution has already been suggested in Refs. [3,5]. In Refs. [68]
it was also argued that the integral of hL (x, Q2 ) h1 (x, Q2 ) can be related to the value of
certain surface terms that appear in formal manipulations involving integrations by parts.
The purpose of this paper is to reexamine the sum rules involving the first moment of
the twist-3 distributions. In particular, we will argue that the twist-3 distribution hL (x, Q2 )
has a potential -singularity at x = 0, assuming that the twist-2 distributions g1 (x, Q2 ) and
h1 (x, Q2 ) do not have such singularity. The paper is organized as follows: in Section 2,
we examine the sum rule for hL . Starting from the general decomposition of hL based
on the QCD equation of motion, we will show that it contains a function hm
L which has
(x)-singularity (Section 2.1). In Section 2.2, we construct hL for a massive quark from
the moment of h3L in the one-loop level and show that the h3L also has an (x)-singularity,
which together with the singularity in hm
L will give rise to a (x) singularity in hL itself. In
Section 2.3, we will perform an explicit light-cone calculation of hL in the one-loop level
to confirm the result in the previous sections. Sections 3 and 4 are, respectively, devoted to
similar examination for the sum rules for e(x, Q2 ) and gT (x, Q2 ).
2. hL (x, Q2 ) for a massive quark

2.1. (x)-functions in hL (x, Q2 )
The OPE analysis of the correlation function (1.2) allows us to decompose hL (x, Q2 )
into the contribution expressed in terms of twist-2 distributions and the rest which we call
h3L (x, Q2 ). Since the scale dependence of each distribution is inessential in the following
discussion, we shall omit it in this subsection for simplicity. Introducing the notation for
314
the moments on [1, 1],

1
Mn [hL ]
dx x n hL (x),
(2.1)
this decomposition is given in terms of the moment relation [1]:

2
n mq
Mn [h1 ] +
Mn1 [g1 ] + Mn h3L (n 1),
n+2
n+2 M
M0 [hL ] = M0 [h1 ],
Mn [hL ] =
with the condition

M0 h3L = 0,

M1 h3L = 0.
(2.2)
(2.3)
(2.4)
(2.5)
By inverting the moment relation, one finds

W
m
3
hL (x) = hW
L (x) + hL (x) + hL (x)

1
1
m
g
(y)
(x)
(y)
h
g
q
1
1
1
2x dy 2 +
+ h3L (x)
2x dy 3
M
x
y
y
x
(x
>
0),
=

x
x
h1 (y) mq g1 (x)
g1 (y)
+ h3L (x)
+ 2x dy 3
2x dy y 2 + M
x
y
1
1
(x < 0),
(2.6)
(2.7)
where the first and second terms in Eq. (2.6) denote the corresponding terms in (2.7). In
this notation the sum rule (2.3) and the condition (2.4) implies1

M0 hm
(2.8)
L = 0.
If one naively integrates (2.7) over x at x > 0 and x < 0, while dropping all surface terms
one arrives at
1
1
dx hL (x) =
1
dx h1 (x) +
0
dx h3L (x)
0
and likewise for 1 dx hL (x). This result, in combination with (2.4), implies that
1
1
1 dx hL = 1 dx h1 . In the following we will argue that this procedure may be wrong
due to the potentially very singular behavior of the functions involved near x = 0.
Investigating this issue in detail will be the main purpose of this paper.
1 More precisely, the original OPE tells us M [h3 + hm ] = 0. But as long as g (0 ) is finite, which we will
0 L
1
L
assume, this is equivalent to stronger relations (2.4) and (2.8).
315
We first address the potential singularity at x = 0 in the integral expression for hm

L (x)
in (2.7). In order to regulate the region near x = 0, we first multiply hm

L (x) by x , integrate
from 0 to 1 and let 0. This yields
1
dx hm
L (x) =
mq
lim
2M 0
0+
1
dx x 1 g1 (y) =
mq
g1 (0+),
2M
(2.9)
while multiplying Eq. (2.7) by |x| and integration from 1 to 0 yields

0
0
mq
mq
m
lim dx|x|1g1 (y) =
g1 (0),
dx hL (x) =
2M 0
2M
(2.10)
where we have assumed that g1 (0+) and g1 (0) are finite. Adding these results we have
0
1

mq
m
dx hL (x) + dx hm
g1 (0+) g1 (0) .
L (x) =
2M
(2.11)
0+
Since we have the sum rule (2.8) and, in general, limx0 g1 (x) g1(x) = 0,2 we are lead
to conclude

mq
m
hm
(2.12)
g1 (0+) g1 (0) (x),
L (x) = hL (x)reg
2M
where hm
L (x)reg stands for the part which is defined by the integral in (2.7) at x > 0 and
W
x < 0 and is regular at x = 0. Since it is unlikely that hW
L (x) contains a -function at the
origin, the relation (2.12) indicates that hL has a (x) term unless h3L (x) has a (x) term
and it cancels the above singularity in hm
L (x).
Eq. (2.12) clearly demonstrate that the functions constituting hL (x) are more singular
near x = 0 than previously assumed and that great care needs to be taken when replacing
integrals over nonzero values of x by integrals that involve the origin. In particular, if it
turns out that hL (x) itself contains a (x) term, then (2.3) implies
1

hL (x) h1 (x) +
0

0,
hL (x) h1 (x) =
(2.13)
0+
since h1 (x) is free from a (x) singularity at x = 0:

0
1
1
dx h1 (x) + dxh1 (x) = dx h1 (x).
1
0+
(2.14)
Accordingly an attempt to verify the hL -sum rule [3] would obviously fail.
However, as we mentioned earlier, in order to see whether the (x) identified in (2.12)
eventually survives or not in hL (x), we have to investigate the behavior of h3L (x) at x = 0.
To this end we will explicitly construct hL (x) for a massive quark to O(S ).
2 For example, dressing a quark perturbatively at O( ) yields g (0+) = 0 and g (0) g (0+) = 0.
S
1
1
1
316
2.2. hL (x, Q2 ) from the moment relations

In this subsection we will construct hL (x, Q2 ) for a massive quark to O(S ) from the
one-loop calculation of Mn [h3L ].
One-loop calculation for a massive quark (mass mq ) gives hL (x, Q2 ) in the following
form:

S
Q2
hL x, Q2 = h(0)
CF ln 2 h(1)
(x),
L (x) +
2
mq L
(2.15)
where the scale Q2 is introduced as an ultraviolet cutoff and the CF = 4/3 is the color
W W,3,m(0,1)
(0)
(x) are defined similarly. g1(0) (x) = h(0)
factor. hL
1 (x) = (1 x) gives hL (x) =
(1 x), as it should. One loop calculation for g1 (x) and h1 (x) for a quark yields the
well-known splitting functions [9,10]:
1 + x2
3
+ (1 x),
[1 x]+ 2
2x
3
(1)
+ (1 x).
h1 (x) =
[1 x]+ 2
(1)
g1 (x) =
(2.16)
(2.17)
Inserting these equations into the defining equation in (2.7), one obtains
1x
(2.18)
,
x

3
3
2
1x
m(1)
hL
3 + (1 x) + 3x (1 x)
(x) =
4x ln
(1 x)+
x
2
2
1
(x)
2
1
2
1x
3 + 3x (x).
=
(2.19)
4x ln
(1 x)+
x
2
W W (1)
hL
(x) = 3x + 4x ln
In the first line of (2.19), the term (3x 32 (1 x)) comes from the self-energy correction,
i.e., from expanding

3 Q2
S
CF ln 2
M = mq 1 +
2
2 mq
m(1)
in Eq. (2.6), and 12 (x) = 12 g1 (0+)(x) in hL (x) accounts for the second term on
the right-hand side of Eq. (2.12). This term is necessary to reproduce the original moment
W W (1)
relation (2.8). We also note that hL
does not have any singularity at x = 0 and satisfies
1
0
W W (1)
dx hL
(x) =
1
dx h(1)
1 (x)
as it should.
(1)
3(1)
hL (x) can be constructed if we know the purely twist-3 part hL (x) in the one-loop
level. h3L (x, 2 ) can be written in terms of the quarkgluon light-cone correlation function
317
as ( 2 = 0, + = 0) [1113]
h3L
x, Q
1
u
d 2ixP +
e
u du t dt
2
u
0

P S | u i5 gF t u Q2 |P S .
iP +
=
M
(2.20)
Starting from this expression, one can, in principle, obtain h3(1)

L (x). Alternatively, and
this is the approach that we will use, one can construct it from already existing one-loop
calculations of its moments. Taking the nth moment of (2.20) or from the OPE analysis of
hL [1], we have
1

dx x n h3L x, Q2
[(n+1)/2]

k=2

2k
1
P S |Rn,k Q2 |P S
n + 2 2M
(2.21)
with

1
n i5 (in D)nk igF n (in D)k2 (0)Q2

Rn,k Q2 = (0)
2
(k n k + 2).
(2.22)
One-loop renormalization of hL was completed in [14] and the mixing matrix for the local
operators contributing to the moments of hL (x, Q2 ) was presented. In particular, matrix
(1)
elements of Rn,k for a massive quark is given in Eq. (3.18) of [14]. Since hL (x) for the
antiquark is zero, we obtain for the moment of the quark distribution
1
dx x n h3(1)
L (x) = 2
[(n+1)/2]

l=2
2l
n+2

1
1
[l 1]3 [n l + 1]3
3
6
1
(2.23)
+ ,
n+1 n+2 2
for n 2 with [k]3 k(k + 1)(k + 2). (The prefactor in (2.23) is determined by comparison
with the anomalous dimension for h1 and by noting that the operator basis for the quark
mass operator in [14] has a sign opposite to those for g1 .) From this result and the defining
3(1)
relation for the lowest two moments of h3L , (2.4) and (2.5), we can construct hL (x) as
=
1
1
3(1)
hL (x) = 3 6x + (1 x) (x).
(2.24)
2
2
We emphasize that the 1/2(x) in (2.24) is necessary to reproduce the n = 0 moment of
3(1)
hL (x). From (2.18), (2.19) and (2.24), one obtains

S Q2
1
2
2
hL x, Q = (1 x) +
(2.25)
ln
CF
+ (1 x) (x) .
2 m2q
[1 x]+ 2
318
We remark that the above calculation indicates that the (x) term appears not only in hm
L but
also in h3L . Furthermore they do not cancel but add up to give rise to (x) in hL (x, Q2 )
itself.
In the next subsection we will confirm Eq. (2.25) through a direct calculation of
hL (x, Q2 ) for a quark.
2.3. Light-cone calculation of hL (x, Q2 )
In order to illustrate the physical origin of the (x) terms in hL (x), and in order to
develop a more general and convenient procedure for calculating such terms, we shall
now evaluate hL (x) using time-ordered light-front (LF) perturbation theory. The general
method has been outlined in Ref. [16] and we will restrict ourselves here to the essential
steps only.
There are two equivalent ways to perform time-ordered LF perturbation theory: one
can either work with the LF Hamiltonian for QCD and perform old-fashioned perturbation
theory (the method employed in Ref. [16]), or one can start from Feynman perturbation
theory and integrate over the LF-energy k first. We found the second approach more
convenient for the one-loop calculation of hL (x) and this is what we will use in the
following.
In LF gauge, A+ = 0, parton distributions can be expressed in terms of LF momentum
densities (k + -densities). Therefore, one finds for a parton distribution, characterized by the
Dirac matrix at O(S ) and for 0 < x < 1
f (x)u(p)
u(p)

d 4k
k+
= ig 2 u(p)
p+
(2)4
1
1
u(p)D (p k),
/k mq + i /k mq + i
where
D (q) =

1
q n + n q
q 2 + i
qn
(2.26)
(2.27)
is the gauge field

propagator in LF gauge, and n is a light-like vector such that nA =
+
0
3
A (A + A )/ 2 for any four vector A .
The k integrals in expressions like Eq. (2.26) are performed using Cauchys theorem,
yielding for 0 < k + < p+

1
1
dk
i
2 (k 2 m2q + i)2 (p k)2 + i
=
1
1
+
2
+
(2k ) 2(p k + )
1 1x
,
2p+ k4
1
p
m2q +k2
2k +
(p k )2 2
2(p + k + )
(2.28)
319
where we used k + = xp+ . In order to integrate all terms in Eq. (2.26) over k , Cauchys
theorem is used to replace any factors of k in the numerator of Eq. (2.26) containing k
by their on-shell value at the pole of the gluon propagator
(p k )2
.
k k p
2(p+ k + )
(2.29)
In the following we will focus on the UV divergent contributions to the parton distribution
only. This helps to keep the necessary algebra at a reasonable level because we restrict
ourselves to the leading behavior in k , which arises from those terms in the numerator
of Eq. (2.26) that are ratio in k , and therefore give rise to a logarithmically divergent k
integral. The transverse momentum cutoff can be replaced by Q2 in these expressions. The
rest of the calculation is just tedious algebra and we omit the intermediate steps here.
We find for 0 < x < 1 to O(S )

S
Q2
2
hL x, Q2 =
CF ln 2
,
2
mq [1 x]+
(2.30)
1
1
1
applies at x = 1, i.e., [1x]
= 1x
for x < 1
where the usual +-prescription for [1x]
+
+
1
1
and 0 dx [1x]+ = 0. Furthermore, hL (x) = 0 for x < 0, since antiquarks do not occur in
the O(S ) dressing of a quark. In addition to Eq. (2.30), there is also an explicit (x 1)
contribution at x = 1. These are familiar from twist-2 distributions, where they reflect the
fact that the probability to find the quark as a bare quark is less than one due to the dressing
with gluons. For higher-twist distributions, the wave function renormalization contributes
is
S
Q2 3
CF ln 2 (x 1).
2
mq 2
The same wave function renormalization also contributes at twist-3. However, for all
higher twist distributions there is an additional source for (x 1) terms which has,
in parton language, more the appearance of a vertex correction, but which arises in fact
from the gauge-piece of self-energies connected to the vertex by an instantaneous fermion
+
2
propagator 2p
+ . For gT (x, Q ) these have been calculated in Ref. [16] where they give
an additional contribution
S
Q2
CF ln 2 (x 1),
2
mq
i.e., the total contribution at x = 1 for gT (x, Q2 ) was found to be

Q2 1
S
CF ln 2 (x 1).
2
mq 2
320
We found the same (x 1) terms also for hL (x, Q2 ).3 Combining the (x 1) piece with
Eq. (2.30) we thus find

S
Q2
1
2
CF ln 2
+ (x 1)
hL x, Q2 = (x 1) +
2
mq [1 x]+ 2
for x > 0.
Comparing this result with the well-known result for h1 [10]

2x
3
S Q2
2
ln
CF
+ (x 1) ,
h1 x, Q = (x 1) +
2 m2q
[1 x]+ 2
(2.31)
(2.32)
one realizes that

1
lim

S Q2
ln
dx hL x, Q2 h1 x, Q2 =
CF = 0,
2 m2q
(2.33)
i.e., if one excludes the possibly problematic region x = 0, then the hL -sum rule [3] is
violated already for a quark dressed with gluons at order O(S ).
In the above calculation, we carefully avoided the point x = 0. For most values of k + ,
the denominator in Eq. (2.26) contains three powers of k when k . However, when
k + = 0, k 2 m2q becomes independent of k and the denominator in Eq. (2.26) contains
only one power of k . Therefore, for those terms in the numerator which are linear in k ,4
the k -integral diverges linearly. Although this happens only for a point of measure zero
(namely, at k + = 0), a linear divergence is indicative of a singularity of hL (x, Q2 ) at that
point. 5 To investigate the k + 0 singularity in these terms further, we consider

k
1
f k + , k dk 2
2
2
(k mq + i) (p k)2 + i

k + (k k )
= dk 2
(k m2q + i)2 [(p k)2 + i]
+

= fcan. k , k + fsin. k + , k ,
(2.34)
where the canonical piece fcan. is obtained by substituting for k its on energy-shell
value k (2.29) (the value at the pole at (p k)2 = 0)
(p k )2
k p
.
2(p+ k + )
(2.35)
3 Physically, the reason why the wave function renormalization contribution depends on the twist is that in LF
gauge, different components of the fermion field acquire different wave function renormalization. However, since
all twist-3 parton distributions involve one LC-good and one LC-bad component, it is natural to find the same
wave function renormalization for all three twist-3 distributions.
4 This is the highest power of k that can appear in the numerator for twist three distributions.
5 There is another point, k + = p + , where a similar divergence occurs, but the latter is only logarithmic.
321
For k + = xp+ = 0, it is only this canonical piece which contributes. To see this, we note
that
k k =
(p k)2
,
2(p+ k + )
and therefore
k k
1
2
2
mq + i) (p k)2 + i

1
1
=
.
dk 2
2(p+ k + )
(k m2q + i)2

fsin k + , k =
dk
(2.36)
(k 2
Obviously [17]

1
=0
dk
2
+
(2k k k m2q + i)2
(2.37)
(2.38)
for k + = 0 because then one can always avoid enclosing the pole at
k =
m2q + k2 i
2k +
by closing the contour in the appropriate half-plane of the complex k -plane. However, on
the other hand

1
dk + dk
2
+
(2k k k m2q + i)2

1
i
= d 2 kL 2
= 2
(2.39)
2
2
2
(kL k mq + i)
k + m2q
and therefore

1 i(k + )
fsin k + , k = + 2
.
2p k + m2q
(2.40)
Upon collecting all terms k in the numerator of Eq. (2.26), and applying Eq. (2.40) to
those terms we find after some algebra those terms in hL (x, Q2 ) that are singular at x = 0

S Q2
ln
CF (x).
hL,sin x, Q2 =
2 m2q
(2.41)
Together with Eq. (2.31), this gives our final result for hL , up to O(S ), valid also for x = 0

2
Q2
1
S
2
CF ln 2 (x) +
+ (x 1) .
hL x, Q = (x 1) +
(2.42)
2
[1 x]+ 2
mq
As expected, hL from Eq. (2.42) does now satisfy the hL -sum rule, provided of course the
origin is included in the integration.
This result is important for several reasons. First of all it confirms our result for
hL (x, Q2 ) as determined from the moment relations. Secondly, it provides us with a
322
method for calculating these (x) terms and thus enabling us to address the issue of validity
of the naive sum rules more systematically. And finally, it shows that there is a close
relationship between these (x) terms and the infamous zero-modes in LF field theory
[18].
While we were completing the manuscript for this paper, we learned of Ref. [15], where
canonical Hamiltonian light-cone perturbation theory is used to calculate hL (x). For x = 0
the result obtained in Ref. [15] agrees with ours which provides an independent check
of the formalism and the algebra. However, the canonical light-cone perturbation theory
used in Ref. [15] is not adequate for studying the point x = 0. From the smooth behaviour
of hL (x) near x = 0 the authors of Ref. [15] conclude that the sum rule for the parton
distribution hL (x) is violated to O(S ). Our explicit calculation for hL (x) not only proves
that the sum rule for hL (x) is not violated to this order if the point x = 0 is properly
included, but also shows that it is incorrect to draw conclusions from smooth behaviour
near x = 0 about the behaviour at x = 0.
3. e(x, Q2)
The other chiral-odd twist-3 distribution e(x, Q2 ) is also expected to satisfy a simple
operator sum rule
1
1

1
P | (0)(0)Q2 |P ,
dx e x, Q2 =
2M
(3.1)
which follows trivially by integrating (1.3) over x. Because of our results from above for
hL (x, Q2 ), we are now of course more cautious and address in the following the issue
whether Eq. (3.1) is also valid if the origin is excluded from the region of integration.
Again, we will consider a massive quark to O(S ).
3.1. Constructing e(x, Q2 ) from its moments
The OPE analysis of (1.3) decomposes e(x) into the twist-2 contribution and the purely
twist-3 piece e3 (x) as
mq
Mn1 [f1 ] (n 1)
Mn [e] = Mn e3 +
M
with the relation (3.1). This moment relation defines the decomposition
(3.2)

mq

f1 x, Q2 ,
xe x, Q2 = xe3 x, Q2 +
(3.3)
M
where we multiplied x to regularize any possible (x) contributions, which will be
specified later. The one loop calculation of e(x, Q2 ) for a massive quark yields

S
Q2
CF ln 2 e(1)(x).
e x, Q2 = (1 x) +
2
mq
(3.4)
323
The lowest moment of e(1)(x) can be obtained directly from the -term relation (3.1)
M0 [e] =
1
M
S
3 Q2
1
M2 =
=1+
P | |P =
CF ln 2 ,
2M
2M mq
mq
2
2 mq
(3.5)
i.e., the lowest moment of e(1) reads

3
M0 e(1) = .
2
Corresponding to (3.2), we have

1
Mn e(1) = Mn e3(1) + Mn1 (mq f1 )(1)
M
(3.6)
(n 1).
(3.7)
Our problem is to construct e(1)(x) for a massive quark from (3.6) and (3.7). In principle
we can obtain the purely twist-3 piece e3(1) from the one-loop calculation starting from the
correlation function [1113]:

P+
d 2ixP
e
e3 x, Q2 =
M
2
1 u

du dtP | (u ) gF (t ) (u )Q2 |P .
(3.8)
0
However, we again make a short cut to get it from the moment. The moment of e3(1)(x)
for a massive quark is given in [19] as a part of the mixing matrix in the context of the
renormalization. From Eqs. (4.2) and (3.18) of [19] we have for the nth moment of e3(1)
(n 1):
1
dx x n e3(1)(x)
1
n/2

1
2
1
2
+
+
[l
1]
[n
l
+
1]
[n/2]
3
3
3
l=2
=

(n+1)/2

1
1
2
+
[l 1]3 [n l + 1]3
l=2
1
1 1
+
2 n n+1
which gives
=
1
xe3(1)(x) = x (1 x) 1 + x.
2
Together with the twist-2 contribution
1 + x2
1
(mq f1 )(1) (x) =
,
M
[1 x]+
(n: even),
(n: odd)
(3.9)
(3.10)
(3.11)
324
one obtains
e(1)(x) =
2
1
+ (1 x)
[1 x]+ 2
for x > 0.
(3.12)
Note that 1/x singularity in e3(1)(x) and f1(1) /x cancel each other in e(1) , and e(1) (x) itself
is integrable at x = 0. In order to satisfy the -term sum rule (3.6) one needs to introduce
a (x) term. Accordingly the final result, which satisfies the moment relations both for
n = 0 and n = 0, reads
e(1)(x) =
2
1
+ (1 x) + (x).
[1 x]+ 2
(3.13)
This result clearly indicates that the sum rule (3.1) is satisfied only by including the point
x = 0 in the integration.
3.2. Light-cone calculation
We start from Eq. (2.26) with = 1
e(x)u(p)
u(p)

2
= ig u(p)

d 4k
k+
x +
(2)4
p
1
1
u(p)D (p k)
/k mq + i /k mq + i
the numerator algebra in Eq. (3.14) yields

(p k) n + (p k) n
2
(
/
k
+
m
)
uu(p)
g
u(p)
q
n
p+ k +

k 2 p+ m2q k +
= 2u(p)
k 2 + m2q + 2
2m
/
k
u(p).
q
p+ k +
(3.14)
(3.15)
In order to determine the canonical part of e(x, Q2 ) we use again contour integration,
k2
picking up the pole at k = k for 0 < k + < p+ , which allows us to replace k 2 1x
k2
and p+ k 2(1x)
, where we kept only the leading terms in k . After some algebra
this yields for 0 < x < 1
2
(3.16)
1x
in agreement with the result obtained from the moments. For the wave function
renormalization (the coefficient in front of (1 x)) the same coefficient is obtained as
for hL (x, Q2 ) and the details will be omitted here.
Finally, we focus on possibly singular terms near x = 0. The only numerator term
/
k u(p).
involving k in Eq. (3.15), which is not multiplied by k + appears in 4mq u(p)
Upon repeating the same steps as in Section 3, i.e., isolating the singular piece by adding
and subtracting k , canceling k k against the gluon propagator, and making use of
e(1)(x) =
325
Eq. (2.40) one finds

e(1)(x)sin = (x).
(3.17)
Collecting all terms we thus find

e(1)(x) =
2
1
+ (1 x) + (x),
[1 x]+ 2
(3.18)
which completely agrees with the result from above (3.13).
4. gT (x, Q2 )
4.1. Possibility of (x) in gT
From a practical point of view, gT (x, Q2 ) is the most important among the twist-3
distributions, because it is the least difficult to measure experimentally. For this reason,
gT (x, Q2 ) has been the subject of many studies in the literature.
We again start with the decomposition of gT (x) into the WandzuraWilcek term
gTW W (x) [20], the term proportional to the quark mass gTm (x), and the purely twist-3 part
gT3 (x):6
gT (x) = gTW W (x) + gTm (x) + gT3 (x)
(4.1)

1
1
g1 (y) mq h1 (x)
h1 (y)
+
dy 2
dy
+ gT3 (x)
y
M
x
y
x
x
(x > 0),
=

x
x
g1 (y) mq h1 (x)
h1 (y)
+ gT3 (x)
+
+ dy 2
dy
y
M
x
y
1
(x < 0).
(4.2)
This decomposition is obtained by the analysis of the correlation function using QCD
equation of motions. The corresponding moment relations are

1
n mq
Mn1 [g1 ] +
Mn1 [h1 ] + Mn gT3
n+1
n+1 M
M0 [gT ] = M0 [g1 ],
Mn [gT ] =
(n 1),
(4.3)
(4.4)
with

M0 gT3 = 0,
(4.5)
6 Here and below in this subsection we omit the scale dependence of the distribution functions for simplicity.
326
where we assumed that gT3 (x) itself is integrable at x = 0. The relation (4.4) is known as
the BurkhardtCottingham (BC) sum rule [2]. We note (4.4) and (4.5) implies

M0 gTm = 0.
(4.6)
In the above discussion we again assumed separate relations (4.5) and (4.6) instead of
M0 [gTm + gT3 ] = 0. This is justified as long as h1 (0) is finite (see below).
Following the same argument leading to (2.12) for hm
L (x), we have
0
1

mq
m
h1 (0+) h1 (0) .
dx gT (x) + dx gTm (x) =
M
(4.7)
0+
This relation together with (4.6) implies gTm (x) has a singularity at x = 0 as

mq
gTm (x) = gTm (x)reg
h1 (0+) h1 (0) (x),
M
(4.8)
where gTm (x)|reg is a part obtained from the integral (4.2) at x > 0 and x < 0 and is regular
at x = 0. This relation shows that if h1 (0) h 1 (0+) = h1 (0+) then the Burkhardt
Cottingham relation would be violated if data is taken only for nonzero xunless of
course there is another (x) contribution to gT3 (x) which happens to cancel exactly the
one in gTm . Perturbation theory predicts at small x that h1 (x) x in the leading order as
is seen from (2.17) and h1 (x) h1 (x) const at the next-to-leading order [21]. BFKL
approach gives h1 (x) const at small x [22]. Therefore, there is a possibility that the (x)
term in (4.8) survives and hence a seeming violation of BC sum rule.
In the next subsections, we will calculate gT (x, Q2 ) for a massive quark to O(S ) in
order to look for (possibly) another origin of the (x) contribution.
4.2. Constructing gT (x, Q2 ) from its moments to O(S )
One loop calculation of gT (x) for a massive quark takes the form of

Q2
S
CF ln 2 gT(1) (x).
gT x, Q2 = (1 x) +
2
mq
(4.9)
Corresponding to the decomposition (4.1) we write

gT(1) (x) = gTW W (1) (x) + gTm(1) (x) + gT3(1) (x).
(1)
(4.10)
(1)
Inserting the expression (2.16) for g1 and (2.17) for h1 into (4.2), and also taking into
account the self-energy correction, we have
W W (1)
gT
m(1)
(x) + gT
(x) = ln x +
1
2
.
+x+
2
[1 x]+
(4.11)
This result has no (x) term, which is simply because h1 (0+ ) = h1 (0 ) = 0 in the one-loop
calculation (see (4.8)).
327
3(1)
As in the case of h3L (x) and e3 (x), we construct gT from its moment for a massive
quark. From Eqs. (5) and (18) of [23], we have for the nth moment of gT3(1) (x) as
1
dx x n gT3(1) (x) =
n2

(n l)
l=1
2
(n + 1)l(l + 1)(l + 2)
3
1
1
+
.
2 2(n + 1) (n + 1)2
(4.12)
This moment relation is originally given for n 1, but it also gives zero for n = 0, which
is consistent with (4.5). Accordingly no (x) term is necessary to obtain
3
1
(x) = (1 x) ln x.
2
2
From (4.11) and (4.13), we get
3(1)
gT
(4.13)

S
Q2 1 + 2x x 2 1
2
gT x, Q = (1 x) +
CF ln 2
+ (1 x) .
2
mq
[1 x]+
2
(4.14)
4.3. LC calculation
We calculated gT (x, Q2 ), using the same LF pQCD techniques that we used to study
hL (x, Q2 ) and e(x, Q2 ). The algebraic steps involved are rather lengthy and we therefore
present only the final result here, which reads

1 + x2
Q2
1
S
CF ln 2 2x +
+ (1 x)
gT x, Q2 = (1 x) +
(4.15)
2
[1 x]+ 2
mq
without any (x) term. This result completely agrees with the findings from Ref. [16],
which confirms our formalism.
Even though the numerator for = 5 in Eq. (2.26) contains k , those terms turn
out to be multiplied by at least one power of k + . Since x(x) = 0, there are no (x)
terms in gT (x, Q2 ) to O(S ). However, we do not have a simple explanation (other than
working out the numerator algebra) as to why factors of k in the numerator algebra are
always accompanied by at least one power of k + for gT (x, Q2 ) and not for hL (x, Q2 ) and
e(x, Q2 ) in this one loop calculation.
5. Summary
We have investigated sum-rules for twist-3 distributions in QCD, and found examples
where these sum-rules are violated if the point x = 0 is not properly included.
For a massive quark, to O(S ) we found

S
Q2
1
1 + x2
2
CF ln 2 2x +
+ (x 1) ,
gT x, Q = (x 1) +
2
mq
[1 x]+ 2
328

2
Q2
1
S
CF ln 2 (x) +
hL x, Q2 = (x 1) +
+ (x 1) ,
2
[1 x]+ 2
mq

S
Q2
1
2
e x, Q2 = (x 1) +
CF ln 2 (x) +
+ (x 1) .
2
mq
[1 x]+ 2
(5.1)
At O(S ) neither hL (x, Q2 ) nor e(x, Q2 ) satisfy their respective sum rule if one excludes
the origin from the region of integration (which normally happens in experimental attempts
to verify a sum rule). gT (x, Q2 ) is the only exception and its sum-rule is satisfied even
when the origin is not included.
Of course, QCD is a strongly interacting theory and parton distribution functions in
QCD are nonperturbative observables. Nevertheless, if one can show that a sum rule fails
already in perturbation theory, then this is usually a very strong indication that the sum rule
also fails nonperturbatively (while the converse is often not the case!).
From the QCD equations of motion, we were able to show nonperturbatively that7 the
difference between hL (x, Q2 ) and h3L (x, Q2 ) contains a function at x = 0

hL x, Q2 h3L x, Q2
singular

mq
2
g1 0+, Q g1 0, Q2 (x).
=
(5.2)
2M
Since

g1 0+, Q2 g1 0, Q2 lim g1 x, Q2 g1 x, Q2
x0
seems to be nonzero (it may even diverge8), one can thus conclude that either hL (x, Q2 )
or h3L (x, Q2 ) or both do contain such a singular term.
We checked the validity of this relation to O(S ) and found that, to this order, both h3L
and hL contain a term (x). We also verified that even though the sum rule for hL (x)
and e(x) are violated if x = 0 is not included, the sum rules for all three twist-3 parton
distributions are still satisfied to O(S ) if the contribution from x = 0 (the (x) term) is
included.
Acknowledgements
M.B. was supported by a grant from the DOE (FG03-95ER40965), through Jefferson
Lab by contract DE-AC05-84ER40150 under which the Southeastern Universities Research Association (SURA) operates the Thomas Jefferson National Accelerator Facility.
Y.K. is supported by the Grant-in-Aid for Scientific Research (No. 12640260) of the Ministry of Education, Culture, Sports, Science and Technology (Japan). We are also grateful to
JSPS for the Invitation Fellowship for Research in Japan (S-00209) which made it possible
to materialize this work.
7 The only assumption that we made is that twist-2 distributions do not contain a -function at x = 0.
8 In the next-to-leading order QCD for a quark, lim
x0 g1 (x) g 1 (x) is logarithmically divergent [24].
329
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]
R.L. Jaffe, X. Ji, Nucl. Phys. B 375 (1992) 527.

H. Burkhardt, W.N. Cottingham, Ann. Phys. 56 (1970) 453.
M. Burkardt, in: Hasegawa et al. (Eds.), Proc. to Spin 92, Universal Academy Press, Tokyo, 1993.
R.D. Tangerman, P.J. Mulders, hep-ph/9408305.
M. Burkardt, Phys. Rev. D 52 (1995) 3841.
R. Jakob, P.J. Mulders, J. Rodrigues, Nucl. Phys. A 626 (1997) 937.
D. Boer, Ph.D. Thesis, Vrije U., Amsterdam, 1998.
M. Burkardt, in: C. Carlson, A. Radyushkin (Eds.), Exclusive and Semi-Exclusive Processes at High
Momentum Transfer, Newport News, VA, 1999, hep-ph/9908479.
V.N. Gribov, L.N. Lipatov, Sov. J. Nucl. Phys. 15 (1972) 438;
Yu.L. Dokshitser, Sov. Phys. JETP 46 (1977) 641;
G. Altarelli, G. Parisi, Nucl. Phys. B 126 (1977) 298.
X. Artru, M. Mekhfi, Z. Phys. C 45 (1990) 669.
I.I. Balitsky, V.M. Braun, Y. Koike, K. Tanaka, Phys. Rev. Lett. 77 (1996) 3078.
A.V. Belitsky, D. Mller, Nucl. Phys. B 503 (1997) 279.
J. Kodaira, K. Tanaka, Prog. Theor. Phys. 101 (1999) 191.
Y. Koike, K. Tanaka, Phys. Rev. D 51 (1995) 6125.
R. Kundu, A. Metz, hep-ph/0107073.
A. Harindranath, W.-M. Zhang, Phys. Lett. B 408 (1997) 347.
S.-J. Chang, T.-M. Yan, Phys. Rev. D 7 (1972) 1147.
M. Burkardt, Adv. Nucl. Phys. 23 (1996) 1.
Y. Koike, N. Nishiyama, Phys. Rev. D 55 (1997) 3068.
S. Wandzura, F. Wilczek, Phys. Lett. B 72 (1977) 195.
A. Hayashigaki, Y. Kanazawa, Y. Koike, Phys. Rev. D 56 (1997) 7350;
W. Vogelsang, Phys. Rev. D 57 (1998) 1886.
R. Kirschner, L. Mankiewicz, A. Schfer, L. Szymanowski, Z. Phys. C 74 (1997) 501.
J. Kodaira, Y. Yasui, K. Tanaka, T. Uematsu, Phys. Lett. B 387 (1996) 855.
W. Vogelsang, Phys. Rev. D 54 (1996) 2023;
W. Vogelsang, Nucl. Phys. B 457 (1996) 47;
R. Mertig, W.L. van Neerven, Z. Phys. C 70 (1996) 637.

Scalar and pseudoscalar meson pole terms in

the hadronic light-by-light contributions to ahad
E. Barto a , A.-Z. Dubnickov a , S. Dubnicka b , E.A. Kuraev c ,
E. Zemlyanaya d
a Department of Theoretical Physics, Comenius University, Bratislava, Slovak Republic
b Institut of Physics, Slovak Academy of Sciences, Bratislava, Slovak Republic
c Bogoliubov Laboratory of Theoretical Physics, JINR Dubna, 141980 Dubna, Russia
d Laboratory of Information Technologies, JINR Dubna, 141980 Dubna, Russia
Received 22 June 2001; accepted 28 March 2002
Abstract
LBL (M) to the anomalous
Third QED order hadronic light-by-light (LBL) contributions a
had from the pole terms of scalar , a (980) and pseudoscalar
magnetic moment of the muon a
0
0

, , mesons (M) in the framework of the linearized extended NambuJonaLasinio model
are evaluated. The off-shell structure of the photonphotonmeson vertices is taken into account
by means of constituent quark triangle loops. The mass of the quark is taken to be mu = md =
mq = (280 20) MeV. The unknown strong coupling constants of 0 , , and a0 mesons with
quarks are evaluated in a comparison of the corresponding theoretical two-photon widths calculated
in the framework of our approach with experimental ones. The -meson coupling constant is
taken to be equal to 0 -meson coupling constant as it follows from the linearized NambuJona
LBL ( ) = (81.83 16.50) 1011 , a LBL () =
Lasinio model Lagrangian. Then one obtains a
0
LBL ( ) = (8.00 1.74) 1011 , a LBL ( ) = (11.67 2.38) 1011 and

(5.62 1.25) 1011 , a
LBL (a ) = (0.62 0.24) 1011 . The total contribution of meson poles in LBL is a LBL (M) =
a
0
(107.74 16.81) 1011 . 2002 Published by Elsevier Science B.V.
PACS: 13.40.Em; 12.39.Fe; 14.60.Ef

Keywords: Meson; Light-by-light; Anomalous magnetic moment
E-mail addresses: bartos@thsun1.jinr.ru (E. Barto), dubnicka@sophia.dtp.fmph.uniba.sk

(A.-Z. Dubnickov), dubnicka@nic.savba.sk (S. Dubnicka), kuraev@thsun1.jinr.ru (E.A. Kuraev),
elena@cv.jinr.ru (E. Zemlyanaya).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 2 4 2 - 0
E. Barto et al. / Nuclear Physics B 632 (2002) 330342
331
1. Introduction
The muon is described by the Dirac equation and its magnetic moment is related to the
spin by means of the expression

e
=g
(1)
s
2m
where the value of the gyromagnetic ratio g is predicted (in the absence of the Pauli term)
to be exactly 2.
In fact, however, the interactions existing in nature modify g to be exceeding the
value 2 because of the emission and absorption of virtual photons (electromagnetic
effects), intermediate vector and Higgs bosons (weak interaction effects) and the vacuum
polarization into virtual hadronic states (strong interaction effects).
In order to describe this modification of g theoretically, the magnetic anomaly was
introduced by the relation
(g 2)
2

3
(2)QED
2
(1)
(2)had
(2)weak
+ a
+ a
+ a
+O
= a
(2)
where = 1/137.03599976(50) [1] is the fine structure constant.

Here we would like to make a note that from all three charged leptons (e , , ) the
muon magnetic anomaly is the most interesting object for theoretical investigations due to
the following reasons:
(i) it is one of the best measured quantities in physics [2]
a = (116592020 160) 1011;
exp
(3)
(ii) its accurate theoretical evaluation provides an extremely clean test of Electroweak
theory and may give hints on possible deviations from Standard Model (SM);
(iii) the new measurement [3] in BNL is expected to be performed with a definitive
accuracy
a = 40 1011 ,
exp
(4)
i.e., it is aimed at obtaining a factor 4 in a precision above that of the last E-821
measurements (3).
At the aimed level of the precision (4) a sensibility will already exist to contributions
[4,5]
a(2,3)weak = (152 4) 1011 ,
(5)
arising from single- and two-loop weak interaction diagrams.

However, as we have mentioned above, the muon magnetic anomaly may also contain
contributions from a new physics, which can be revealed, we hope, in a comparison
332

exp
of a with an accurate theoretical evaluation of ath . While QED and weak interaction
contributions to ath seem to be estimated very reliably, there is still opened a door for
improvements in hadronic contributions.
As the most critical from all hadronic contributions are the light-by-light (LBL) meson
pole terms [6], we recalculate the third-order hadronic LBL contributions to the anomalous
magnetic moment of the muon ahad from the pole terms of the scalar , a0 and pseudoscalar
0 , , mesons (M) in the framework of the linearized extended NambuJonaLasinio
model. The reason for the latter are predictions of series of recent papers
aLBL = (+52 18) 1011
[7]
aLBL = (+92 32) 1011
[8]
aLBL = (+79.2 15.4) 1011
[9]
aLBL = (+83 12) 1011
[10]
aLBL (0 ) = (+58 10) 1011
[10, 11]
(6)
which differ in the value. Moreover, in these papers only the pseudoscalar pole
contributions were considered. In our paper we include the scalar meson (, a0 ) pole
contributions as well.
The crucial point is description of transition form factor M . Current methods
used for this aim are the ChPT and the vectormesondominance (VMD) model.
In this paper the corresponding transition form factors by the constituent quark triangle
loops with colourless and flavourless quarks with charge equal to the electron one are
represented. (An application of a similar modified constituent quark triangle loop model for
a prediction of the pion electromagnetic form factor behavior can be found in [12], where
also a comparison with the naive VMD model prediction is carried out.) The mass of the
quark in the triangle loop is always taken to be mu = md = mq = (280 20) MeV [13],
which was determined in the framework of the chiral quark model of the NambuJona
Lasinio type by exploiting the experimental values of the pion decay constant, the -meson
decay into two-pions constant, the masses of pion and kaon and the mass difference of
and mesons. The unknown strong coupling constants of 0 , , and a0 mesons with
quarks are evaluated in a comparison of the corresponding theoretical two-photon widths
with experimental ones. The -meson coupling constant is taken to be equal to 0 -meson
coupling constant as it follows from the corresponding Lagrangian. The -meson mass
is taken to be m = (496 47) MeV as an average of the values recently obtained
experimentally from the decay D + + + [14] and excited decay [15] processes.
As a result we present explicit formulas for aLBL (M) (M = 0 , , , , a0 ) in terms of
Feynman parametric integrals of 10-dimensional order, which subsequently are calculated
by MIKOR method [16].
The paper is organized as follows. In the next section all definitions and a derivation of
basic relations is presented. More detail can be found in Appendix A and in Appendix B.
The last section is devoted to numerical results and discussion.
333
2. Meson pole terms in light-by-light contributions to ahad

The third QED order hadronic LBL scattering contributions to ahad are generally
represented by the diagram in Fig. 1A, which contains a class of pseudoscalar mesons
0 ) square loop diagrams (Fig. 1B), a class of quark square loop
( , K and also KS,L
diagrams (Fig. 1C) and scalar , a0 (980) and pseudoscalar 0 , , meson pole diagrams
(Fig. 2), where the off-shell structure of the photonphotonmeson vertices is taken into
account by means of flavourless and colourless constituent quark triangle loops. Here the
interaction of mesons (M) with quarks is described by the linearized NambuJonaLasinio
type Lagrangian

(x) + i(x)5 q(x),

Lq qM
(7)
= gM q(x)
had (A) and class of pseudoscalar meson

Fig. 1. Third order hadronic light-by-light scattering contribution to a
square loop diagrams (B) and quark square loop diagrams (C) contributing to (A).
had .
Fig. 2. Meson (M) pole diagrams in the third order hadronic light-by-light scattering contributions to a
334
with unknown strong coupling constant gM . The latter for 0 , , and a0 mesons are
evaluated by a comparison of the corresponding theoretical two-photon widths with the
experimentally determined values [1].
For the on-mass shell scalar (S) and pseudoscalar (P ) mesons decays one can write the
following matrix elements

M S(p) (k1 ) + (k2 )
igS
=
(8)
K(mS /mq )(k1 k2 g k1 k2 )e1 (k1 )e2 (k2 ),
mq
gP

J (mP /mq )(k1 k2 e1 e2 ),
M P (p) (k1 ) + (k2 ) =
(9)
mq
with
1
K(z) = 2
1x

dx
0
1 4xy
dy,
1 xyz2
2
J (z) = 2
z
1

dx
ln 1 z2 x(1 x)
x
(10)
and
p2 = m2M ,
k12 = k22 = 0,
(abcd) = ( a b c d .
They lead to the theoretical two-photon widths (for more detail see Appendix A)
=
2 m3S gS2 2
K (mS /mq ),
64 3 m2q
(11)
P =
2 m3P gP2 2
J (mP /mq ),
64 3 m2q
(12)
dependent on the scalar gS and pseudoscalar gP meson coupling constants with quarks,
respectively.
Taking the meson masses with errors and the following world averaged values of the
two-photon widths from the last Review of Particle Physics [1], 0 = (0.008 0) keV,
= (0.464 0.044) keV, = (4.282 0.339) keV and a0 = (0.24 0.08) keV,
the quark mass value mq = (280 20) MeV, one finds the meson coupling constants with
quarks as follows
g2 0 = 9.120 1.305,
g2 = 2.220 0.378,
g2 = 6.708 1.096,
ga20 = 0.757 0.277.
(13)
Since there is no experimental result on the -meson decay into two-photons up to now,
we identify g2 with g2 0 as it follows from the Lagrangian (7).
Now we are ready to evaluate LBL meson pole terms contributions to ahad .
Usually the fermionphoton vertex function is written in the form
(q)u(p1 ),
2 )V
= u(p

(q) = F1 q 2 + 1 q,
V
F2 q 2 ,
4m
q = p2 p1 ,
(14)
335
with on mass-shell muons, i.e., p12 = p22 = m2 (m is the muon mass). Then the expression
for the muon anomalous magnetic moment is
(g 2)
= F2 (0)
2

1
3
(q) p1 + m P (q),
lim
Sp
p
+
m
V
=
2
16m2 q 2 0 q 2
a =
(15)
with a projection operator

m
P (q) = q 2 +
q,
.
3
The gauge invariant set of Feynman diagrams containing the LBL scattering block with
meson pole intermediate states is drawn in Fig. 2.
Their contributions to ahad are
aLBL (S) + aLBL (P ) =

S+P
1
3
(p1 + m)P (q),
lim 2 Sp p2 + m V
2
2
16m q 0 q
(16)
with
S+P =
V
26 3
(P )
(S)
T
d 4 k1 d 4 k2 T
1
O
+
,
(i 2 )2
k12 m2P
k12 m2S k12 k22 (k1 k2 )2
(17)
O = 2O1 + O2 ,
p 2 k2 + m
p 1 k1 + m
,
2
2
(p2 k2 ) m
(p1 k1 )2 m2
p 2 k2 + m
p 1 + k1 k2 + m
=
,
2
2
(p2 k2 ) m
(p1 + k1 k2 )2 m2
O1 =
O2
where
(P )
T =

4 2 gP2
(k1 q)( k2 k1 )f1 k12 f2 (k1 , k2 ),
2
2
mq
(18)
with

f1 k12 =
1
0
m2q
1
dx,
2
m2q x(1 x)k1
f2 (k1 , k2 ) = 2
1x
1
dx1
0
m2q
d
dx2 ,
(19)
d = m2q k22 x3 (1 x3 ) k12 x2 (1 x2 ) + 2k1 k2 x2 x3 ,

x3 = 1 x1 x2 .
Here we would like to note that the functions f1 (k12 ), f2 (k1 , k2 ) at the limit of an infinite
value of the quark mass, mq , tend to one.
The expression for T (S) has a similar form
(S)
=
T

4 2 gS2
(g k1 q k1 q )f1 k12 f (k1 , k2 ),
2
2
mq
(20)
336

f1 k12 =
1
m2q (1 2x(1 x)) dx

m2q x(1 x)k12
1
f k1 , k2 = 2
1x
1
dx1
0
m2q

(1 4x1x2 ) g k2 (k1 k2 ) k2(k1 k2 )
d

+ 2(k1 k2 )2 x1 (2x1 1)g dx2 .
Usual procedure of performing the loop-momenta integration using the Feynman

parameters method (see Appendix B) leads to
aLBL (M) = RM FM (mM , mq )
(21)
where
RM =
2
3m2 2 gM
64 5 m2q
(22)
and
FP (mP , mq ) = 2Ma + Mb ,
FS (mS , mq ) = 2Na + Nb .
The Mi , Ni are given by the expressions

1 APa
P
Ma = da 2
+ Ba ,
zD1 D1
P

Ab
1
P
Mb = db
+ Bb ,
2 D22 D2
1 ASa
S
Na = da 2
+ Ba ,
zD1 D1
S

Ab
1
S
Nb = db
+ Bb ,
2 D22 D2

(23)
with da,b the 10-dimensional integrals on Feynman parameters

1

da =
4
0
1
db =
4
0
dx
x x
dx
x x
x1
1
dx1
0
1
x1
dx1
dx3
x3 x3
dx3
x3 x3
z 1
1
dz1
1
2 32 d 3 dy 2 d,
dz2
1
z1
1
dz1
0
2 32 43 d 4 2 d,
dz2
0
mq
,
=
x = 1 x,
z1 = 1 z1 ,
m
d 3 = d1 d2 d3 ,
d 4 = d1 d2 d3 d4 .
The explicit expressions for polynomials A, B are

8y 2
2a1 + 2r1 a1 (a1 + b1 ) r12 a1 b1 8r13 ,
3z
y
BaP = 4 + 12r1 + 4a1 b1 ,
z
APa =
(24)
337
16
8 3
8
a2 r22 a2 b2 (b2 1) r22 a2 + r23 (1 2b2),
3
3
3
8
4
P
a2 b2 (b2 1),
Ba = 4a2 + r2 (8b2 4) +
3

y
1
1
S
3
8a1 16r1 a1 a1 b1 a1 + b1
Aa =
z
3
3

2
2
2
8r12 a1 b1 a1 + b1
b1 d1 (x1 , x2 )
3
3
3

y 16 2
32
+
a (a1 + 1) + r1 (b1 1)(a1 + 1)a1
z 3 1
3
16
2
2
+ r1 (a1 + 1)(b1 1) d2 (x1 ),
3

16
8y
BaS = 8a1 +
d2 (x1 ) (4 + 12a1)d1 (x1 , x2 )
(a1 + 1)(b1 1)2 d2 (x1 )
3
z
y
(12a1b1 8a1 + 8b1 8)b1d1 (x1 , x2 ),
+
z

16 2
32
a2 (a2 1) + r2 a2 (a2 1)(b2 1)
ASb =
3
3
16
+ r2 (a2 1)(b2 1)2 d2 (x1 )
3

16
16
+ a22 + r22 b2 (b2 1) d1 (x1 , x2 ) ,
3
3

8
8
BbS =
b2 (b2 1) d1 (x1 , x2 )
3
16 8
(25)
(a2 1)(b2 1)2 d2 (x1 )

+ 8a2
3
APb =
with
a1 = 3 2 ,
b1 = 2 3 1 ,
b2 = 4 (3 1 + 3 ),
d1 (x1 , x2 ) = 1 4x1x2 ,
a2 = 4 (3 2 + 3 ),
r1 = (y c1 y),
r2 = c2 ,
d2 (x1 ) = 2x1 (2x1 1).
(26)
Quantities c1,2 , 1 , z, D1 , D2 and are given in Appendix B.
3. Results and discussion

The result for the meson pole contributions of LBL type has the form
aLBL (M) = aLBL( 0 ) + aLBL () + aLBL( ) + aLBL ( ) + aLBL(a0 )
where

aLBL 0 = R 0 F 0 (m 0 , mq ),
(27)
338
aLBL () = R F (m , mq ),
aLBL ( ) = R F (m , mq ),
aLBL ( ) = R F (m , mq ),
aLBL (a0 ) = Ra0 Fa0 (ma0 , mq ).
(28)
2 determined in Section 2, the mass of the quark

Now, taking the values (13) of gM
mq = (280 20) MeV [13], the mass of the -meson m = (496 47) MeV [14,15],
the mass of the muon and the corresponding scalar and pseudoscalar mesons from the
Review of Particle Physics [1] one finds

aLBL 0 = (81.83 16.50) 1011,
aLBL () = (5.62 1.25) 1011,

aLBL ( ) = (8.00 1.74) 1011 ,
aLBL ( ) = (11.67 2.38) 1011,
aLBL (a0 ) = (0.62 0.24) 1011 .
(29)
Here we would like to note that our value for the 0 is in the framework of error bars
consistent with [10,11].
So, the total contribution of meson poles in LBL is
aLBL (M) = (107.74 16.81) 1011 ,
(30)
where the resultant error is the addition in quadrature of all partial errors of (30).
0
Together with the contributions of the pseudoscalar meson ( , K , KS,L
) square
loops and constituent quark square loops [7,8] it gives
aLBL (total) = (111.20 16.81) 1011.
(31)
The others 3-loop hadronic contributions derived from the hadronic vacuum polarizations (VP) were most recently evaluated by Krause [17]
a(3)VP = (101 6) 1011 .
(32)
Then the total 3-loop hadronic correction from (32) and (33) is
a(3)had = aLBL(total) + a(3)VP = (10.20 17.28) 1011
(33)
where the errors have been again added in quadratures.

If we take into account the most recent evaluation [18] of the lowest-order hadronic
vacuum-polarization contribution to the anomalous magnetic moment of the muon
a(2)had = (7021 76) 1011
(34)
(it is interesting to notice that this result almost coincides in the central value and in the
error as well with the result [19] of one of the authors (S.D.) of this paper obtained 12 years
ago) the pure QED contribution up to 8th order [20]
aQED = (116 584 705.7 2.9) 1011
(35)
339
and the single- and two-loop weak interaction contribution [5], finally one gets the SM
theoretical prediction of the muon anomalous magnetic moment value to be
ath = (116 591 888.9 78.1) 1011.
(36)
Comparing this theoretical result with experimental one (3) one finds
a ath = (131 178) 1011
exp
(37)
which implies very good consistency of the SM prediction for the anomalous magnetic
moment of the muon with experiment.
Acknowledgements
We are grateful to participants of BLTP Seminar for discussions. E.A.K. is grateful to
RFBR 01-02-17437 as well as NNC, JINR for financial support via INTAS Grant No. 9730494 and to SR grant 2000.
The work was in part also supported by Slovak Grant Agency for Sciences, Grant
No. 2/5085/20 (S.D.) and Grant No. 1/7068/20 (A.Z.D.).
Appendix A
We put here the details of calculations of the two-photon widths in the framework of the
linearized NambuJonaLasinio model.
The decay matrix element of a pseudoscalar meson P into two photons has a form (we
take into account two directions of fermion line in the triangle fermion loop)
4

2igP mq

d k S1
,
M P (p) e1 (k1 ) e2 (k2 ) =
(A.1)
i 2 a1 a2 a3
with

1
S1 =
Tr k k2 + mq e2 k + mq e1 k + k1 + mq 5 = i(k2 k1 e1 e2 ),
4mq
a1 = (k + k1 )2 m2q ,
a2 = (k1 k2 )2 m2q ,
a3 = k 2 m2q ,
(A.2)
where we use the definitions 5 = i 0 1 2 3 , 14 Sp = i , 0123 = +1.

Joining the denominators and performing the loop momentum integration one obtains

x 1 m2q (k2 k1 e1 e2 )
d 3 x
2gP

MP
(A.3)
=
.
mq
m2q x1 x2 p2 x1 x3 k22 x2 x3 k12
This expression is used for the construction of f2 (k1 , k2 ) in (19).
For the case of real particles we have
gP
(k1 k2 e1 e2 )J (mP /mq )

MP =
mq
with J (z) given by (10).
(A.4)
340
Note that for the case of -meson the matrix element becomes complex. Numerically
one obtains

J (m /mq )2 = 2.12. (A.5)
J 2 (m 0 /mq ) = 1.04,
J 2 (m /mq ) = 3.73,
For the matrix element of a decay of the scalar meson S into two photons one can write

M S(p) (k1 , e1 ) (k2 , e2 ) = e1 e2 M ,
(A.6)
where
M
4
2igS mq
d k T
=
,
i 2 a1 a2 a3

1
=
Sp k k2 + mq k + mq k + k2 + mq .
4mq
(A.7)
This quantity suffers from the ultraviolet divergences. It is the gauge invariance requirement which provides the regularization. General form of M (k1 , k2 ), which satisfies the
current conservation condition
M (k1 , k2 )[k1 ; k2 ] = 0,
(A.8)
has the form

M (k1 , k2 ) = A g k1 k2 k1 k2 + B k12 k2 k1 k2 k1 k22 k1 k1 k2 k2 .
(A.9)
Usual procedure of joining the denominators, the calculation of the trace, a performation
of the loop momentum integration and an application of the gauge conditions to the muon
block, leads to a final result.
For the case of real particles the structure B disappears and we have
(1)igS
K(mS /mq ),
M (S ) = e1 (k1 )e2 (k2 ) g k1 k2 k1 k2
mq
(A.10)
which leads to the expression (11) for the width with K(z) given by (10). Numerically

K(m 0 /mq )2 = 0.96.
a
Appendix B
For a performance of the loop momenta integration we use Feynman trick of joining
the denominators
(n + m 1)!
1
=
a m bn (n 1)!(m 1)!
1
0
x m1 (1 x)n1 dx
(ax + b(1 x))m+n
and the relation

4
(m + 1)!(n m 3)!
d k (k 2 )m
= (1)m+n
,
i 2 (k 2 d)n
(n 1)!d nm2
(B.1)
n > m + 2.
(B.2)
For this aim we present the denominators containing k2 in the form

1 1
1
1
;
,
x3 x3 {1}{2}{3}{4} {5} {6}
341
(B.3)
where
{1} = k22 ,
{5} = k12
{2} = k22 2p2 k2 ,
2p1 k1 ,
{3} = (k2 k1 )2 ,
{6} = (p1 + k1 k2 ) m ,
2
{4} = d/(x3 x3 ),
x = 1 x.
Using Feynman parameter 1 to join the denominators {3}, {4} and the parameter 2 for
joining the result with {2} we have

2 1 {4} + 1 {3} + 2 {2} = k22 2k2 a + 2 1 ,
a = 2 1 k1 + 2 p2 ,
1 = 2 1
(B.4)
x2 x2 k12 m2q
1 = 1 + 1
x3 x3
x2
,
x3
+ 2 2 k12 ,
a = 1 = a.
For the diagram in Fig. 2A we use the parameter 3 for the joining of the denominator {1} with the above result
(k2 b)2 zd1 ,
b = 2 a,
(B.5)

1 x2 x2
,
z = (2 3 1 ) 2 3 1 +
x3 x3
2
d1 = k12 + 2c1 k1 p1 + 2 ,
c1 =
2 32 2 1
,
z
2 =

1 2 3
1 2 2 2
.
m 3 2 + m2q
z
x3 x3
For the diagram in Fig. 2B we use the parameter 3 for the joining of the denominator (B.3) with {6} and parameter 4 to join {1} with the result
(k2 b2 )2 d2 ,
(B.6)
where

x2 x2
+ 3 ) 4 3 2 3 4 1 + 1
,
x3 x3

b2 = 4 ( 3 + 1 1 )k1 + ( 3 + 3 2 )p1 ,
= 42 (3 1
d2 = k12 + 2c2 p1 k1 + ,

1
c2 = 4 4 (3 1 + 3 )(3 2 + 3 ) 3 ,

1
1 2 3 4 2
=
m2 42 (3 2 + 3 )2 +
mq .
x3 x3
342
To perform the k1 integration we use the parameters z1 , z2 to join the denominators

containing k12
1
2
1
=
x x k12 (k12 M 2 )(k12 M12 ) x x
z1
1
dz1
0
dz2
d33
(B.7)
with
d3 = k12 z1 M12 z2 M 2 ,
M12 =
m2q
.
x x
For the diagram in Fig. 2B we use parameter to join the denominators d2 , d3
d2 + d
3 = (k1 q2 )2 m2 D2 ,
q2 = c2 p1 ,

m2 D2 = m2 (c2 )2 + z1 M12 + z2 M 2 .
(B.8)
For the diagram in Fig. 2A we use parameter y to join d1 with {5}

yd1 + y(5)
= k12 2p1 k1 (y c1 y) + y1
(B.9)
and parameter to join the result with d3 . As a result one gets

(k1 q1 )2 D1 m2 ,
q1 = (y c1 y)p1 ,

m2 D1 = m2 2 (y yc1 )2 + z1 M12 + z2 M 2 y1 .
(B.10)
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
Review of Particle Phys., D.E. Groom et al., Eur. Phys. J. C 15 (2000) 1.

Muon (g-2) Collaboration, H.N. Brown et al., Phys. Rev. Lett. 86 (2001) 2227.
Muon (g-2) Collaboration, R.M. Carey et al., Phys. Rev. Lett. 82 (1999) 1632.
A. Czarnecki, B. Krause, W.J. Marciano, Phys. Rev. Lett. 76 (1996) 3267.
G. Degrassi, G.F. Giudice, Phys. Rev. D 58 (1998) 53007.
K. Melnikov, hep-ph/0105267.
M. Hayakawa, T. Kinoshita, A.I. Sanda, Phys. Rev. D 54 (1996) 3137;
M. Hayakawa, T. Kinoshita, hep-ph/0112102.
J. Bijnens, E. Pallante, J. Prades, Nucl. Phys. B 474 (1996) 379, hep-ph/0112255.
M. Hayakawa, T. Kinoshita, Phys. Rev. D 57 (1998) 465, hep-ph/0112102.
M. Knecht, A. Nyffeler, hep-ph/0111058.
I. Blokland, A. Czarnecki, K. Melnikov, hep-ph/0112117.
S. Dubnicka, G. Georgios, V.A. Meshcheryakov, Can. J. Phys. 63 (1985) 1357.
M. Nagy, M.K. Volkov, V.L. Yudichev, in: A.-Z. Dubnickov, S. Dubnicka, P. Strenec (Eds.), Proc. of Int.
Conf. Hadron Structure 2000, Stara Lesna, Slovak Republic, 2.7.10. 2000, Comenius Univ., Bratislava,
2001, p. 188.
E.M. Aitala et al., Phys. Rev. Lett. 86 (2001) 770.
T. Komada, M. Ishida, S. Ishida, Phys. Lett. B 508 (2001) 31.
I.M. Korobov, Number Theory Method in Approximate Analysis, Fizmatgiz, Moscow, 1963 (in Russian).
B. Krause, Phys. Lett. B 390 (1997) 392.
S. Narison, Phys. Lett. B 513 (2001) 53.
L. Martinovic, S. Dubnicka, Phys. Rev. D 42 (1990) 884.
V.W. Hughes, T. Kinoshita, Rev. Mod. Phys. 71 (2) (1999) S133.

Multijet matrix elements and shower evolution in

hadronic collisions: W bb + n-jets as a case study
Michelangelo L. Mangano a , Mauro Moretti b , Roberto Pittau c
a CERN, Theoretical Physics Division, CH 1211 Geneva 23, Switzerland
b Dipartimento di Fisica, Universit di Ferrara, and INFN, Ferrara, Italy
c Dipartimento di Fisica, Universit di Torino, and INFN, Turin, Italy
Received 12 February 2002; accepted 2 April 2002
Abstract
We study in this paper the production, in hadronic collisions, of final states with W gauge bosons,
heavy quark pairs and n extra jets (with n up to 4). The complete partonic tree-level QCD matrix
elements are evaluated using the ALPHA algorithm, and the events generated at the parton level are
then evolved through the QCD shower and eventually hadronised using the coherent shower evolution
provided by the HERWIG Monte Carlo. We discuss the details of our Monte Carlo implementation,
and present results of phenomenological interest for the Tevatron Collider and for the LHC. We also
production, when only
comment on the impact of our calculation on the backgrounds to W (H bb)
one b-jet is tagged. 2002 Elsevier Science B.V. All rights reserved.
1. Introduction
Multijet final states are characteristic of a large class of important phenomena present
in high-energy collisions. QCD interactions generate multijet final states via radiative
processes at high orders of perturbation theory. Heavy particles in the Standard Model
(SM), such as W and Z bosons or the top quark, t, decay to multiquark configurations
(eventually leading to jets) via electroweak (EW) interactions. In addition to the above
SM sources, particles possibly present in theories beyond the SM (BSM) are expected to
decay to multiparton final states, and therefore to lead to multijets. Typical examples are
This work was supported in part by the EU Fourth Framework Programme Training and Mobility of
Researchers, Network Quantum Chromodynamics and the Deep Structure of Elementary Particles, contract
FMRX-CT98-0194 (DG 12-MIHT).
E-mail address: michelangelo.mangano@cern.ch (M.L. Mangano).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 2 4 9 - 3
344
M.L. Mangano et al. / Nuclear Physics B 632 (2002) 343362
the cascade decays to quarks and gluons of supersymmetric strongly interacting particles,
such as squarks and gluinos.
Whether our interest is in accurate measurements of top quarks or in the search for more
exotic states, multijet final states always provide an important observable, and the study of
the backgrounds due to QCD is an essential part of any experimental analysis.
In addition to fully hadronic multijet final states, a special interest exists in final states
where the jets are accompanied by gauge bosons. For example, the associated production
of several jets with a Z boson decaying to neutrinos gives rise to multijet + missing
/ T ) final states. These provide an important background to the search
transverse energy (E
for supersymmetric particles. Likewise, the associated production of 4-jets and a W boson
decaying leptonically provides the leading source of backgrounds to the identification and
study of top quark pairs in hadronic collisions.
Several examples of calculation of multijet cross-sections in hadronic collisions exist
in the literature. Some of them are included in parton-level Monte Carlo (MC) event
generators, where final states consisting of hard and well isolated partons are generated.
Among the most used and best documented examples are PAPAGENO [1] (a compilation
of several partonic processes), VECBOS [2] (for production of W/Z bosons in association
with up to 4-jets), NJETS [3] (for production of up to 6-jets). In addition to these ready-touse codes, programs for the automatic generation of user-specified parton-level processes
exist, and have been used for the calculation of many important processes in hadronic
collisions: MADGRAPH [4], CompHEP [5] and GRACE [6].
Studies of the partonic final states can be performed by assuming that each hard
parton can be identified with a jet, and that the jets momenta are equal to those of the
parent parton. This simplification is extremely useful to get rough estimates of production
rates, but cannot be used in the context of realistic detector simulations, for which a
representation of the full structure of the final state (in terms of hadrons) is required. This
full description can be obtained by merging the partonic final states with so-called shower
MC programs (such as HERWIG [7], PYTHIA [8] or ISAJET [9]), where partons are
perturbatively evolved through emission of gluons, and subsequently hadronized. As we
shall discuss in the following, however, this merging is not always possible, since common
parton-level MCs sum and average over flavours and colours, and do not usually provide
sufficient information on the flavour and colour content of the events.
The goal of the present work is twofold. First we review a general strategy proposed
in [10] for the construction of event generators for multijet final states, based on the
exact leading-order (LO) evaluation of the matrix elements for assigned flavour and
colour configurations, and the subsequent shower development and transition into a fully
hadronized final state. In view of the complexity of the problem, we shall argue in favour
of the use of the algorithm ALPHA [11] as the best possible tool to carry out the relative
matrix-element computations. Applications of ALPHA to the case of hadronic collisions
have already been shown successful in the case of multijet production in [10,12].
+ n-jet
Then we apply the technique to the specific example of production of W QQ
final states (with Q being a massive quark, and n 4). The matrix-element calculation
and the generation of parton-level events is completed with the shower evolution and
hadronization generated by the HERWIG MC [7]. We present the results, in the case of
Q = b, of the parton-level calculations for several production rates and distributions of
345
interest at the Tevatron and at the LHC. We discuss some interesting features of the final
states, exploring in particular the relative size of processes where the b and b give rise
to either 1 or 2 taggable jets. This comparison will underscore the importance of full
matrix-element calculations taking fully into account the heavy-quark mass effects. We
then present some results relative to final states reconstructed after the shower evolution.
We compare parton-level to jet-level distributions, and study the ability of HERWIG to
approximate the emission rate of extra jets via the shower evolution. Independent work
on the merging of parton-level calculations with shower MCs (for the specific case of
PYTHIA) has been pursued by the CompHEP group [13] and by the GRACE group [14],
limited however to final states with at most 4 partons.
We conclude this presentation by listing future projects which could be realized within
the framework of the approach presented here.
2. Matrix-element evaluation
The emission of soft gluon radiation in state-of-the-art shower-evolution programs
accounts for quantum coherence, which is implemented via the prescription of angular
ordering in the parton cascade [15]. Angular ordering dictates that the radiation emitted
by a colour dipole be confined within the cone defined by the directions of the two
colour charges defining the dipole itself. The set of colour connections among the partons
which defines the set of dipoles for a given event will be called a colour flow, or colour
configuration.
The comparison with existing hadron-collider data [16], confirms that the constraint
of angular ordering is essential to properly describe the particle multiplicity and the
momentum distribution inside jets, as well as to describe the correlations between primary
jets and softer jets emitted during the shower evolution. In order to reliably evolve a
multiparton state into a multijet configuration, it is therefore necessary to associate a
specific colour-flow pattern to each generated parton-level event. This requires an adhoc approach to the evaluation of the matrix elements. A specific proposal was presented
in [10], and will be shortly summarized here.
2.1. Reconstruction of colour flows
We discuss for simplicity the case of multigluon processes [17], as the extensions to
cases with quarks and electroweak particles [18,19] follow the same pattern. The scattering
amplitude for n gluons with momenta pi , helicities i and colours ai (with i = 1, . . . , n),
can be written as [17]:

M {pi }, {i }, {ai }

tr ai1 ai2 ain A {pi1 }, {i1 }, . . . , {pin }, {in } .
=
(1)
P (2,3,...,n)
The sum extends over all permutations Pi of (2, 3, . . . , n), and the functions A({Pi })
(known as dual or colour-ordered amplitudes) are gauge-invariant, cyclically-symmetric
346
functions of the gluons momenta and helicities. Each dual amplitude A({Pi }) corresponds
to a set of diagrams where colour flows from one gluon to the next, according to
the ordering specified by the permutation of indices. When summing over colours the
amplitude squared, different orderings of dual amplitudes are orthogonal at the leading
order in 1/N 2 :

M {pi }, {i }, {ai } 2
cols

A {Pi } 2 + 1 (interf.) .
= N n2 N 2 1
(2)
N2
Pi
At the leading order in 1/N 2 , therefore, the square of each dual amplitude is proportional
to the relative probability of the corresponding colour flow. Each flow defines, in a gauge
invariant way, the set of colour currents which are necessary and sufficient to implement
the colour ordering prescription necessary for the coherent evolution of the gluon shower.
Because of the incoherence of different colour flows, each event can be assigned a specific
colour configuration by comparing the relative size of |A({Pi })|2 for all possible flows.
When working at the physical value of Nc = 3, the interferences among different flows
cannot be neglected in the evaluation of the square of the matrix element, Eq. (1). As a
result, the basis of colour flows does not provide an orthogonal set of colour states, and a
MC selection of colour flows is not possible. In [10] we proposed to use the Gell-Mann
basis of SU(3) matrices as an orthogonal basis to represent the colour state of a given set
of partons:

0 1 0
0 0 1
1
1
1
2
=
=
0 0 0 ,
0 0 0 ,
2 0 0 0
2 0 0 0

0 0 0
0 0 0
1
1
3
5
=
=
1 0 0 ,
0 0 1 ,
2 0 0 0
2 0 0 0

0 0 0
0 0 0
1
1
6
7
=
0 0 0 ,
=
0 0 0 ,
2 1 0 0
2 0 1 0

1 0 0
1 1 0 0
1
4
8
0 1 0 .
0 1 0 ,
=
=
2 0 0 0
12 0 0 2
In this basis, only a fraction of all possible 8n colour assignments gives rise to a nonzero amplitude. For each event, we randomly select a non-vanishing colour state for
the external gluons, and evaluate the amplitude M. We then list all dual amplitudes
contributing to the chosen colour configuration according to Eq. (1) and, among these
dual amplitudes, we randomly select a colour flow on the basis of their relative weight.
In a 6-gluon amplitude, for example, a possible non-zero colour assignment is given by
(2, 7, 5, 6, 1, 3). Up to cyclic permutations, only three orderings of the colour indices give
rise to non-vanishing traces: tr(2 7 5 6 1 3 ), tr(2 6 1 5 7 3 ) and tr(2 7 3 1 5 6 ).
Therefore, only three dual amplitudes contribute to the full amplitude: A(2, 7, 5, 6, 1, 3),
347
A(2, 6, 1, 5, 7, 3) and A(2, 7, 3, 1, 5, 6). The colour ordering to be specified for the
coherent parton-shower evolution can be selected by comparing the size of the squares
of tr(2 i2 i6 )A(2, i2, . . . , i6 ) for the three contributing permutations (i2 , . . . , i6 ) of the
colour indices.
In the limit of a large number of generated events, and in the case of processes with
only gluons or with gluons and one quarkantiquark pair, this algorithm is equivalent to
the colour-flow extraction algorithm proposed in [20] and employed in HERWIG. There,
the full sum over all colours is performed for each event. For each event one then calculates
the individual subamplitudes M(f ) corresponding to all possible colour flows f . The event
is then assigned the colour flow f with a probability:
|M(f)|2
P (f) =
.
2
f |M(f )|
(3)
In the case of two or more q q pairs our prescription and that of [20] differ at order
1/Nc2 , since by generating colour states we produce configurations whose matrix element
has no leading colour contribution. One such example is the process qi qi qj qj , where
i, j = 1, . . . , Nc , are fixed colours. This transition is mediated by the gluons corresponding
to the diagonal Gell-Mann matrices. The colour coefficient is trivially given by

1
1
(4)
aik ajl =
ik j l ,
il j k
2
Nc
a
which, when k = i and l = j as in the proposed example, is of order 1/Nc . The
corresponding colour flow links the quark and antiquark of the initial state, and those of
the final state. No such colour flows can appear in the algorithm by Odagiri, where only
colour configurations of leading order in Nc are generated.
Since the colour-flow information is only relevant for events which will be evolved
through a shower MC, and since one usually does this only for unweighted events, it is
sufficient to evaluate the possibly large set of dual amplitudes corresponding to a given
colour state only for the small set of unweighted events. The extraction of a colour
flow, therefore, does not increase significantly the computing time necessary to generate
unweighted events. The full sum over colours is obtained by averaging over a large sample
of events. The cross section thus obtained is correct to all orders in 1/Nc . Relative to the
prescription of [20], this approach has the advantage that the random selection of colours
can in principle be optimized by exploiting the strong colour-dependence of the matrix
elements (see an example along these lines in [21]).
While the algorithm proposed above can in principle be adopted in the context of any of
the existing calculational techniques, including those based on the automatic generation of
matrix-elements such as CompHEP, MADGRAPH or GRACE, the flexibility to calculate
amplitudes relative to fixed colour states, as well as those relative to dual amplitudes, and
the ability to efficiently calculate matrix elements for processes with large numbers of
final-state partons, single out in our view the ALPHA algorithm as the best suitable tool.
The ability to tackle computations of this complexity in the case of hadronic collisions was
proved in [10,12].
348
+ n-jets production
3. W QQ
The associated production of W bosons and heavy quarks (Q = b, t) is one of the most
important background processes to several searches for new physics, as well as to the
detection of top quarks. The final state W bb is also the leading irreducible background
()
The relative
to the production of SM Higgs bosons via the process p p W (H b b).
matrix elements have been known for long time [22]. The NLO corrections to the final
states where both b and b are hard and can be treated as massless have recently become
()
has been studied for massive b at LO in [25],
available [24]. The process p p W bbj
while the processes with up to two extra jets, but again in the limit of massless b, are
included in VECBOS [2].
The extension to larger jet multiplicities, and the inclusion of mass effects, are
necessary, among other things, to allow more accurate studies of the backgrounds to t t
production. In this case, final states have typically 2-jets in addition to the bb pair, but
initial and final state radiation can lead to the presence of extra jets. As we shall show later,
the increase in the number of light jets in the final state which the background calculation
should be able to cope with is also mandated by the need to account for background
configurations where only one jet can be reconstructed from the bb system.
As an application of our techniques, we then carried out the calculation of the process
()
+ n-jets, covering final states with up to n = 4 jets in addition
p p (W ) + QQ
to the heavy-quark pair. The full spin correlations between the leptons from the W decay
and the other particles are included in the matrix element, as well as the finite width of the
W , described by a BreitWigner. For simplicity, in the following we shall however always
refer to the W , instead of the lepton pair which is used in the actual computations.
The classes of processes necessary to describe these final states, and with up to 2 lightquark pairs, are listed here:
PROC = 1:
PROC = 2:
PROC = 3:
PROC = 4:
PROC = 5:
PROC = 6:
q q W b b,
qg W bb q ,
(5)
(6)
q ,
gg W bbq
(7)

q q W b bq q + q q W b bq q + q q W b bq q

q + qq W bbqq

q + q q W bbq
+ q q W b bq

+ qq W b bqq
(8)
,

q q + qg W bbq
q q + qg W bbqq
q , (9)
qg W bbq

q q q .
q q q + gg W bbq
gg W bbq
(10)
We did not indicate possible additional final-state gluons, and did not explicitly list trivial
permutations of the initial state partons and charge-conjugated processes. This list of
processes fully covers all those possible in the case of up to 3-jets in addition to the b
In the case of 4 extra jets, we did not include processes with 3 light-quark pairs.
and b.
Within the uncertainties of the LO approximation, these can be safely neglected [2]. The
349
matrix elements are obtained using ALPHA, which calculates the Green function generator
via an exact numerical iterative algorithm, as explained in [11].
Note that PROC = 6 only contributes to 4 light-jet production; PROC = 5 only
contributes to 3 light-jet production; PROC = 3 and 4 only contribute to 2 light-jet
production; PROC = 2 only contributes to 1 light-jet production; PROC = 1 contributes
to all W bb + X final states. Every time the jet multiplicity is increased, new classes of
processes appear. These new processes cannot be simply obtained by adding one extra
gluon to the lower-order ones. As a result, their rate cannot be estimated in a shower MC
approach, where lower-order processes are allowed to evolve and emit more jets due to
gluon radiation. This fact stresses once more the importance of a complete parton-level
calculation of all relevant matrix elements.
In the cases where comparisons with existing results were possible, we verified that
our code reproduces previous calculations. We carried out these tests at the level of total
cross sections for the massive W bb case (comparing with the results of [22]), and for W bb
plus up to 2 extra jets, for massless b (comparing with the results obtained running the
VECBOS code1 ).
All calculations are performed using MC techniques, and the resulting code can be used
as an event generator. For each generated event the code provides the full kinematics and
flavours of the external partons, selected according to the relative probabilities, as well
as the colour flow. The code includes an interface to HERWIG, allowing the events to
be fully evolved through the coherent parton shower, and to be hadronized. A universal
interface to shower MCs, following the criteria discussed in Ref. [23], is being prepared.
The full code, as well as more detailed documentation, are available from the URL:
http://home.cern.ch/mlm/wbb/wbb.html.
4. Study of the partonic results

In this section we present some numerical results for parton-level cross-sections and
distributions, using experimental
configurations corresponding to the upgraded Tevatron
Collider (pp collisions at s = 2 TeV) and to the LHC (pp collisions at s =

14 TeV). As default parameters for our calculations we shall use: mW = 80.23 GeV,
W = 2.03 GeV, mb = 4.75 GeV, PDF set CTEQ5M [26], with 2-loop s (Q2 ), and
renormalization/factorization scales 2R = 2F = m2W + pT2 , where pT2 is the average
pT2 of all outgoing jets.
We shall define here as light jets those formed by the light quarks and gluons. They
are required to be separated from each other, and from the b quarks, by *Rij > 0.4, with
*Rij = [(i j )2 +(i j )2 ]1/2 for each pair of partons i and j . The cut at *Rij > 0.4,
rather than at the value of *Rij > 0.7 used as standard in jet physics at the Tevatron, is
motivated by the choice of jet definition used in most top-quark studies at the Tevatron [27].
We shall analyse later the impact of this choice on the comparison between jet rates at the
parton level and at the fully-showered level. Finally, all jets are also required to be hard
1 http://www-theory.fnal.gov/people/giele/vecbos.html.
350
and central:
pti > 20 GeV,
|i | < 2.5.
(11)
We shall not set any cut on the charged lepton and on the neutrino (giving rise to missing
transverse energy, E
/ T ) from the W decay, and we shall present rates including only one
possible leptonic flavour in the W decay.
In the following, we shall use the symbol NJ to indicate the total number of jets,
including the jets generated by the b or b quarks.
4.1. Total rates
We start by considering final states where both b and b are sufficiently hard and
well separated to form independent jets, and apply to them the same cuts defining light
jets, namely, *Rbb > 0.4 and Eq. (11). We shall call these 2-b-jet events. Table 1
gives the rates for final states with 2-b-jets plus extra light jets at the Tevatron, with the
contributions from the different classes of processes listed separately. The Table shows
that, for multiplicities up to 4-jets, the qualitatively new processes appearing every time the
multiplicity is increased by one are of the same order, or larger, as the processes obtained
from the lower-order channels by radiating one extra gluon.
The results for the LHC are given in Table 2. Note here that the effect of the new
processes appearing for larger jet multiplicities is even more important. This is due to the
Table 1
3 jN rates (fb), for the Tevatron at
Contributions from different initial states to the p p (W )bbj
J
S = 2 TeV, with the cuts given in Eq. (11). The different processes are defined in Eqs. (5)(10). The indicates
that the process is not available for the given jet multiplicity. The PDF set is CTEQ5M, and only one lepton
flavour is considered. The uncertainties (quoted in parentheses as errors on the last significant figure) reflect the
statistical accuracy of our MC evaluation
Process
NJ = 2
1
2
3+4
5
6
360(1)
NJ = 3
68.6(4)
37.6(2)
NJ = 4
10.4(1)
12.1(1)
4.3(1)
NJ = 5
1.46(1)
2.63(3)
1.66(3)
0.085(2)
NJ = 6
0.20(1)
0.47(1)
0.41(1)
0.036(1)
0.00038(2)
Total
360(1)
106.4(4)
26.8(2)
5.84(4)
1.11(2)
Table 2
Same as Table 1, for the LHC. Rates in pb
Process
NJ = 2
NJ = 3
NJ = 4
NJ = 5
NJ = 6
1
2
3+4
5
6
2.60(1)
0.63(1)
2.97(1)
0.144(3)
2.11(2)
0.288(1)
0.036(2)
1.08(2)
0.24(1)
0.030(1)
0.008(1)
0.47(2)
0.13(2)
0.031(4)
0.0010(3)
Total
2.60(1)
3.60(1)
2.54(2)
1.38(2)
0.64(3)
351
fact that the lowest-order process involving initial state antiquarks is strongly suppressed
by the sea-quark density relative to higher-order final states, which are enhanced by the
large initial-state gluon contribution. This result is consistent with the numbers quoted in
Table 7 of [2] in the case of massless b.
it
Since experimentally one does not always require the identification of both b and b,
is important to consider, in addition to the case of 2-b-jet events, cases where the event has
only one taggable b-jet. These events receive contributions from final states where either
one of the b and b is too soft or outside the rapidity range for the jet definition, or where
the b and b are close enough as to merge into a single jet. If we treated the b as a massless
particle, both these limiting cases would lead to infinite rates at LO, due to soft or collinear
divergences. Only the inclusion of NLO virtual corrections, with an infrared and collinear
safe jet definition, would restore a finite and physical answer. Since we treat the b quark
as massive, the LO calculation is however meaningful and finite (expressing the fact that
the b-ness of a jet is in principle measurable via its decays regardless of how small its
energy is, and regardless of how collinear the bb pair is).
We can therefore define and predict the rates for final states where only one jet is
taggable. We call these 1-b-jet events. More specifically, 1-b-jet events are those which
fail the 2-b-jet definition, but fall in one of these classes:
(1) One of the b and the b satisfies Eq. (11).
(2) Both b and b fail Eq. (11), but

pt b + b > 20 GeV,
*Rbb < 0.4,

b + b < 2.5.
(12)
For these events, production of a (W + NJ )-jet final state requires an O(sNJ +1 ) process,
as one extra light-parton jet is needed to give the required jet multiplicity. In spite of the
presence of the extra power of s , the contribution of these events to the 1-tag final states
can be very large, because of the presence of potentially large soft and collinear logarithms.
This is confirmed by Table 3, which shows that the contribution to single-tag events coming
from the higher-order processes is as large as that of the leading-order ones.2 In the case of
the LHC, Table 4, the higher-order contributions are significantly larger than the leadingorder ones, consistently with the results shown for the multiplicity-dependence of the total
jet rates in Table 2. This is particularly true for the process most relevant to the study
of associated W H production, where the O(s3 ) process is almost 5 times larger than the
O(s2 ) one. The requirement of double-b tagging is very important in this case to efficiently
reduce this extra background.
Before concluding this section we stress that all rates shown in the tables are affected
by a large overall uncertainty due to the choice of renormalization and factorization scales.
Since the matrix-element calculations only include the LO contributions, and given the
large powers of s which multiply the rates (the processes with NJ final state partons scale
2 The actual relative contribution of 2-b-jet events and 1-b-jet events to the single-tagged rate depends on
the experimental value of the tagging efficiency, which will depend on the structure of the b-jet. We expect,
for example, that tagging efficiencies for semileptonic tags will be approximately a factor of two larger for jets
containing two b than for single-b-jets.
352
Table 3
Contributions to W + NJ -jet rates (fb) at the Tevatron, from final states with 2 (upper row) or 1 (lower row) b-jets
Perturbative order
N
O(s J )
N +1
O(s J )
NJ = 1
NJ = 2
NJ = 3
NJ = 4
NJ = 5
360(1)
106.4(4)
26.8(2)
5.84(4)
1316(3)
371(2)
94.0(5)
20.5(4)
3.8(1)
Table 4
Contributions to W + NJ -jet rates (fb) at the LHC, from final states with 2 (upper row) or 1 (lower row) b-jets
Perturbative order
N
O(s J )
N +1
O(s J )
NJ = 1
9.38(5)
NJ = 2
2.60(1)
12.3(1)
NJ = 3
NJ = 4
NJ = 5
3.60(1)
2.54(2)
1.38(2)
7.4(1)
3.71(5)
1.7(1)
like sNJ ), one should assume overall systematic uncertainties of the order of a factor of 2
4, depending on the jet multiplicity. In principle, this uncertainty can be reduced by using
the information coming from data on W +NJ jets (without final-state b-jets) or Z +bb +njets, which are not affected by the contamination of the t t signal. The calculation of the
Z + b b + n-jet processes will be done in the future.
4.2. Jet distributions
We present in this subsection some interesting distributions for parton-level multijet
events. Here and in the following sections we shall limit our studies to the case of the
Tevatron. Similar qualitative results hold for the LHC as well. Fig. 1 shows the inclusive
pT distribution of b-jets, for final states with jet multiplicities NJ = 25. We compare
the distributions of b-jets in 2-b-jet and 1-b-jet events, where in the first case each event
will contribute two entries to the histograms. The shapes of the two distributions are rather
similar through most of the energy range. The curve relative to 2-b-jet events has a larger
normalization because of the double probability of finding a b-jet in the event.
The large contribution coming from events where only one b-jet is reconstructed is
clearly visible. To show the impact that such events have on the background to the
associated production of a W boson and a resonance decaying to a b pair (e.g., a Higgs
boson), we present in Fig. 2 the invariant mass spectrum of the dijet pair in dijet events. The
solid line corresponds to the O(s2 ) process, where both b quarks give rise to independent
jets. The dashed line corresponds to the 1-b-jet events obtained from the O(s3 ) process.
This figure shows that the requirement of having two independently b-tagged jets is crucial
for the background suppression. A detailed study of the tagging efficiencies for 1-b-jet
events is necessary to ensure that the residual contribution from fake tags on the second,
non-b, jet is small.
To explore more in detail the structure of the b-jet in 1-b-jet events, we plot in Fig. 3 the
showing the curves for different ranges of pT of
*R separation between the b and the b,
the b-jet. Events where *R > 0.4 correspond to cases where one of the two bs is either too
soft or is outside the rapidity range. The figures show that at large pT we are dominated
by 1-b-jets with the b and b merged inside the jet (*R < 0.4). The tagging efficiency
353
Fig. 1. Inclusive pT distributions of b-jets. The solid lines refer to 2-b-jet events (in which case both jets are
entered in the histograms); the dashed lines refer to 1-b-jet events.
Fig. 2. Dijet invariant mass distribution, in W plus 2-jet events, where at least one of the two jets has a b in it.
The solid line refers to 2-b-jet events; the dashed line refers to 1-b-jet events.
354
Fig. 3. bb angular correlations in events with 2-b-jets (plots) and with 1-b-jet (histograms). The case of 1-b-jet is
divided into subsamples defined by the pT of the b-jet in the ranges 2040, 40100 and > 100 GeV.
for these jets is presumably larger than that for jets made of a single b, in particular if
semileptonic tags are considered. It is an important experimental issue to evaluate what the
actual tagging efficiency is for these jets, as a function of pT .
5. Study of the fully showered final states

In this section we describe the results obtained after evolution through the parton
shower. The goal of our calculations is to be able to generate as accurately as possible
the full jet structure of the partons generated at the matrix-element level. We first generate
a sample of parton-level events with a fixed multiplicity (NJ ), and then process these
events through HERWIG for the shower evolution. When applied to an event with NJ
hard final-state partons, we expect that the parton-shower evolution will generate an event
with an NJ -jet inclusive final state. Initial or final state radiation may change the overall
jet multiplicity (e.g., by splitting a jet in two, or radiating a new one from the initial state),
but these effects are of order s relative to the inclusive NJ -jet properties of the event.
In principle one could use the shower evolution to predict the rates for configurations
with jet multiplicities larger than NJ [28,29]. We shall give examples later on of how
well the MC succeeds in this goal. As is well known, however, the shower MCs tend
to underestimate the fraction of events with extra jets, unless explicit matrix-element
355
corrections for the emission of hard and non-collinear radiation are included. Algorithms
exist for implementing these matrix-element corrections in the case of low jet multiplicity
processes (corrections to DY production [30], top decays [31], W Z-pair production [32]),
but their extension to the case of high-jet multiplicities we are interested in is currently
severely limited by the complexity of these processes.
In the following two subsections we present some tests of our scheme. In the first one
we discuss some sanity checks of our approach, showing that the inclusive properties of
NJ -jet final states are preserved by the shower evolution. In the second subsection we
discuss the ability of the shower MC evolution to predict the emission rates for extra jets.
5.1. Sanity checks
To start with, we compare the inclusive jet rates before and after evolution. While the
evolution of the partonic events through HERWIG can generate fully hadronized final
states, we chose here to stop the evolution after its perturbative part, in order to be able
to compare as closely as possible the effects of higher-order perturbative corrections to
the evolution of the final state. For the jet clustering after the shower evolution we use
the standard cone algorithm, implemented in the routine GETJET [33] provided with the
HERWIG code. To match the cuts used at the parton level, we use Rjet = *R = 0.4, and a
jet pT threshold of 20 GeV.
Fig. 4 compares the pT spectra at parton level with those after the shower, for events
generated at the parton level with 2, 3, 4 and 5 jets. The pair of curves correspond to the
series of jets, from the hardest to the softest one. The curves show that the spectra after
radiation do not exactly coincide with those at the parton level. The difference is consistent
with an average energy-loss of 34 GeV outside the jet cone. One should therefore expect
that a better matching of the partonic and fully showered jet spectra will be achieved by
using wider jet cones, since these will more efficiently contain the energy radiated by the
partons during their evolution. That this is the case, is shown in Fig. 5. Here we compare
the spectra obtained using a cut of *R > 0.7 on the partons with those relative to fully
showered jets defined by a cone of Rjet = 0.7. To generate this sample of Rjet = 0.7 jets we
used the sample of parton-level events defined by the *R > 0.4 cut, in order to cover the
cases where two partons merge into a single jet, and extra jets are produced by the radiative
processes. The agreement between the spectra is now perfect, consistently with the results
obtained in the 2-jet case in [28]. The message here is that when preparing parton-level
samples to be used for the QCD evolution by a shower MC algorithm, it is important to set
generation cuts looser than those used in the definition of the full jets, in order to account
for downward fluctuations in the jet energy and for jet merging induced by the jet clustering
algorithms. In principle these problems should be avoided by using jet algorithms based
on kT clustering [34].
Fig. 6 shows the matching between the directions of the partons before the shower,
and the direction of the fully evolved jets. We note a smearing of the direction, which is
more enhanced in the case of the soft jets, as should be expected. The matching of the jet
momenta before and after the shower is shown in Fig. 7. The distributions show on average
a momentum loss, mostly due to radiation outside the jet cone, but have a width which is
only of the order of 510% of the parton momentum.
356
Fig. 4. Inclusive pT distributions of jets at the parton level (solid curves) and after shower evolution (dashed
curves). The curves are ordered going from the hardest to the softest jet in the event. Parton level jets are defined
by an isolation cut *R > 0.4, and showered jets by a cone Rjet = 0.4.
Fig. 5. Inclusive pT distributions of jets at the parton level, with separation cut *R > 0.7 (solid curves), and
of fully-showered Rjet = 0.7 jets (dashed curves). These last were obtained starting from the full sample of
*R > 0.4 partonic events.
357
Fig. 6. Parton-jet alignment. We plot the distance *R between the direction of the parton before radiation, and
that of its daughter jet. Jets are labeled in decreasing order of pT .
Fig. 7. Jets momentum matching. We plot the difference between the momentum of the parton before radiation
and that of the jet, normalized to the parton momentum. Jets are labeled in decreasing order of pT .
358
Fig. 8. Inclusive spectrum of b-jets: the result for 2-b-jet parton-level events (solid) is compared to the spectrum
obtained for the same events after shower evolution (dashed line); the plotted points correspond to the result for
events which have still 2-b-jets after the shower evolution.
In Fig. 8 we display the comparison between the inclusive spectra of b-jets, before and
after shower evolution. At the parton level we use events with 2-b-jets. Only a fraction of
them will survive as 2-b-jet events after the shower, once again due to some energy loss
outside the cone, and to artifacts of the jet merging during the clustering.
The outcome of all these studies is that the shower evolution preserves the inclusive
properties of parton-level events, up to corrections induced by energy losses outside the
small, Rjet = 0.4, jet cones.
5.2. Jet radiation in HERWIG
We address in this section the issue of the ability of the shower MC to correctly predict
the rate for hard radiation leading to extra final-state jets. The plots in Fig. 9 show the
spectrum of the (NJ + 1)th jet in events obtained by evolving through HERWIG partonlevel events with NJ jets. These spectra are compared with what obtained by using directly
the (NJ + 1)-jet parton-level matrix elements, before (solid lines) and after (dashed lines)
the HERWIG evolution. As we pointed out in a previous section, large contributions arise
when higher-order parton level processes are considered. These contributions are due to the
appearance of new processes, which cannot be generated via radiation processes in the MC
(e.g., new initial states not present at the lower orders). In order to make a fair comparison,
359
Fig. 9. Comparison of jet rates evaluated with matrix elements with those obtained from hard radiation during the
shower evolution of lower-order parton-level processes.
we therefore included in the parton-level estimate of the (NJ + 1)th jet spectrum only those
N
processes which are present at order s J . In the case of 3 final-state jets, we find a good
agreement in the range pT 45 GeV between the rates obtained by evolving 3-jet parton
level events through HERWIG (plotted points) and the rates obtained via radiation during
the shower evolution of 2-jet parton-level events (dashed line). A similar result is obtained
in the case of 4-jets. In this case we also studied the prediction of HERWIG based on 2-jet
parton-level events, probing therefore the ability to predict the rate for the emission of 2
extra jets (diamond symbols). The deficit in 4-jets predicted using the 2-jet matrix elements
is in part due to the lack of qg-initiated processes, which account for 35% of the 3-jet rate.
Even including this correction, however, we see that not enough hard radiation is emitted to
correctly predict the emission of 2 extra jets. The situation is improved when considering
5 and 6 jets, as shown in the last frames of Fig. 9, presumably because the overall amount
of energy present in the events in this case improves the validity of the soft-gluon emission
approximation even in the case of radiation leading to emission of 2 extra jets.
6. Conclusions and outlook

In this work we presented a general framework for the evaluation of complex
multiparton matrix elements, including an algorithm for the consistent merging with a
coherent shower evolution, leading to fully hadronized multijet final states. We applied
360
these ideas to the specific case of W bb + n-jet production, completing the calculation for
We presented
processes with up to a total of 6 final-state partons (including the b and b).
results both at the parton level, and after the shower evolution, which was done using the
HERWIG MC. In addition to providing a proof of feasibility for these calculations and to
showing the consistency of the merging with the coherent shower evolution, we pointed
out several important consequences of the higher-order calculations and of the ability to
maintain a non-zero mass for the b quarks. On one side we stressed the fact that jet rates
do not obey the naive s scaling with multiplicity, due to the appearance at higher orders
of new, large contributions. This is especially true in the case of the lowest-order W bb
and W bb + j processes, and it is even more remarkable in the case of pp collisions at the
LHC. On the other side, we pointed out that measurements done by requiring the presence
of only one tagged b-jet are very sensitive to the contribution of 1-b-jet events, namely,
events where the b-jet contains both the b and the b quarks, or where one of the two b does
not reconstruct a taggable jet. The impact of these events on the background subtraction
needed to determine the top cross section from the current measurements performed at the
Tevatron remains to be assessed.
The formalism we have outlined, and the key use of ALPHA to enable the calculation
of the matrix elements and the extraction of the colour information needed for the coherent
shower evolution, are readily portable to the study of other complex multiparton processes.
The removal of the bb pair from the calculation will immediately lead to the evaluation
of (W + n)-jet rates, with n up to 4. While these processes have already been studied in
the literature, and have been encoded in the VECBOS MC, our approach will add to the
existing tools the ability to consistently evolve the final states through the shower MCs, and
to generate events suitable for realistic detector simulations. The work on the (W + n)-jet
program is in progress. If one can neglect contributions from subprocesses with 6 or more
light quarks, an extension to n up to 6 will be readily available using the ALPHA code.
Inclusion of the processes with 6 or more light quarks presents no conceptual problems,
but simply requires a more involved bookkeeping of all possible flavour channels.
()
The calculation of several other hard processes (including p p t tb b + n jets,
()
b + n jets (with n up to 2), Zbb + n jets, t t + n jets, production of multiple
p p b bb
W and Z bosons, and associated production of single and double Higgs bosons) is being
completed,3 and will soon be reported on.
Acknowledgements
M.M. and R.P. thank the CERN Theory Division for hospitality during various stages
of this work.
3 M.L. Mangano, M. Moretti, F. Piccinini, R. Pittau and A. Polosa, work in progress.
361
References
[1] I. Hinchliffe, LBL-34372 Submitted to Workshop on Physics at Current Accelerators and the Supercollider,
Argonne, IL, 25 June 1993.
[2] F.A. Berends, H. Kuijf, B. Tausk, W.T. Giele, Nucl. Phys. B 357 (1991) 32.
[3] F.A. Berends, W.T. Giele, H. Kuijf, Phys. Lett. B 232 (1989) 266.
[4] T. Stelzer, W.F. Long, Comput. Phys. Commun. 81 (1994) 357, hep-ph/9401258.
[5] A. Pukhov et al., hep-ph/9908288.
[6] T. Ishikawa, T. Kaneko, K. Kato, S. Kawabata, Y. Shimizu, H. Tanaka, MINAMI-TATEYA group
Collaboration, KEK-92-19.
[7] G. Marchesini, B.R. Webber, Nucl. Phys. B 310 (1988) 461;
G. Marchesini, B.R. Webber, G. Abbiendi, I.G. Knowles, M.H. Seymour, L. Stanco, Comput. Phys.
Commun. 67 (1992) 465;
G. Corcella et al., JHEP 0101 (2001) 010, hep-ph/0011363.
[8] T. Sjostrand, Comput. Phys. Commun. 82 (1994) 74;
T. Sjostrand, P. Eden, C. Friberg, L. Lonnblad, G. Miu, S. Mrenna, E. Norrbin, Comput. Phys. Commun. 135
(2001) 238, hep-ph/0010017.
[9] F.E. Paige, S.D. Protopopescu, H. Baer, X. Tata, hep-ph/9810440.
[10] F. Caravaglios, M.L. Mangano, M. Moretti, R. Pittau, Nucl. Phys. B 539 (1999) 215, hep-ph/9807570.
[11] F. Caravaglios, M. Moretti, Phys. Lett. B 358 (1995) 332, hep-ph/9507237.
[12] P. Draggiotis, R.H. Kleiss, C.G. Papadopoulos, Phys. Lett. B 439 (1998) 157, hep-ph/9807207.
[13] A.S. Belyaev et al., hep-ph/0101232.
[14] K. Sato, S. Tsuno, J. Fujimoto, T. Ishikawa, Y. Kurihara, S. Odaka, hep-ph/0104237.
[15] A. Bassetto, M. Ciafaloni, G. Marchesini, Phys. Rep. 100 (1983) 201;
L.V. Gribov, E.M. Levin, M.G. Ryskin, Phys. Rep. 100 (1983) 1.
[16] F. Abe et al., CDF Collaboration, Phys. Rev. D 50 (1994) 5562;
B. Abbott et al., D0 Collaboration, Phys. Lett. B 464 (1999) 145, hep-ex/9908017;
B. Abbott et al., D0 Collaboration, Phys. Lett. B 414 (1997) 419, hep-ex/9706012.
[17] F.A. Berends, W. Giele, Nucl. Phys. B 294 (1987) 700;
M. Mangano, S. Parke, Z. Xu, Nucl. Phys. B 298 (1988) 653.
[18] M. Mangano, S.J. Parke, Nucl. Phys. B 299 (1988) 673;
F.A. Berends, W.T. Giele, Nucl. Phys. B 306 (1988) 759.
[19] M.L. Mangano, S.J. Parke, Phys. Rep. 200 (1991) 301.
[20] K. Odagiri, JHEP 9810 (1998) 006, hep-ph/9806531.
[21] P.D. Draggiotis, A. van Hameren, R. Kleiss, Phys. Lett. B 483 (2000) 124, hep-ph/0004047;
A. van Hameren, R. Kleiss, Eur. Phys. J. C 17 (2000) 611, hep-ph/0008068.
[22] Z. Kunszt, Nucl. Phys. B 247 (1984) 339;
M.L. Mangano, Nucl. Phys. B 405 (1993) 536.
[23] E. Boos et al., hep-ph/0109068.
[24] R.K. Ellis, S. Veseli, Phys. Rev. D 60 (1999) 011501, hep-ph/9810489.
[25] A.S. Belyaev, E.E. Boos, L.V. Dudko, Phys. Rev. D 59 (1999) 075001, hep-ph/9806332.
[26] H.L. Lai et al., CTEQ Collaboration, Eur. Phys. J. C 12 (2000) 375, hep-ph/9903282.
[27] F. Abe et al., CDF Collaboration, Phys. Rev. D 50 (1994) 2966;
F. Abe et al., CDF Collaboration, Phys. Rev. Lett. 74 (1995) 2626, hep-ex/9503002;
S. Abachi et al., D0 Collaboration, Phys. Rev. Lett. 74 (1995) 2632, hep-ex/9503003.
[28] W.T. Giele, T. Matsuura, M.H. Seymour, B.R. Webber, FERMILAB-CONF-90-228-T Contribution to Proc.
of Summer Study on High Energy Physics: Research Directions for the Decade, Snowmass, CO, June 25
July 13, 1990. Published in Snowmass Summer Study, 1990, pp. 0137147.
[29] J.M. Benlloch, A. Caner, M.L. Mangano, T. Rodrigo, CDF/DOC/MONTECARLO/1823, October 1992;
J.M. Benlloch, CDF Collaboration, Published in the Proceedings of DPF 92, Vol. 2, pp. 10911093.
[30] M.H. Seymour, Comput. Phys. Commun. 90 (1995) 95, hep-ph/9410414;
G. Corcella, M.H. Seymour, Nucl. Phys. B 565 (2000) 227, hep-ph/9908388;
G. Miu, T. Sjostrand, Phys. Lett. B 449 (1999) 313, hep-ph/9812455.
362
[31] G. Corcella, M.H. Seymour, Phys. Lett. B 442 (1998) 417, hep-ph/9809451;
G. Corcella, M.L. Mangano, M.H. Seymour, JHEP 0007 (2000) 004, hep-ph/0004179.
[32] M. Dobbs, Phys. Rev. D 64 (2001) 034016, hep-ph/0103174.
[33] GETJET, F. Paige and M. Seymour, private communication.
[34] S. Catani, Y.L. Dokshitzer, B.R. Webber, Phys. Lett. B 285 (1992) 291;
S.D. Ellis, D.E. Soper, Phys. Rev. D 48 (1993) 3160, hep-ph/9305266;
M.H. Seymour, Nucl. Phys. B 513 (1998) 269, hep-ph/9707338.

Cosmological bounds on neutrino degeneracy

improved by flavor oscillations
A.D. Dolgov a,1 , S.H. Hansen b , S. Pastor c , S.T. Petcov d,2 ,
G.G. Raffelt c , D.V. Semikoz c,3
a INFN sezione di Ferrara, Via del Paradiso 12, 44100 Ferrara, Italy
b NAPL, University of Oxford, Keble road, OX1 3RH, Oxford, UK
c Max-Planck-Institut fr Physik, Fhringer Ring 6, 80805 Mnchen, Germany
d Scuola Internazionale Superiore di Studi Avanzati, and INFN sezione di Trieste,
Via Beirut 2-4, 34014 Trieste, Italy

Received 5 February 2002; accepted 9 April 2002
Abstract
We study three-flavor neutrino oscillations in the early universe in the presence of neutrino
chemical potentials. We take into account all effects from the background medium, i.e., collisional
damping, the refractive effects from charged leptons, and in particular neutrino self-interactions that
synchronize the neutrino oscillations. We find that effective flavor equilibrium between all active
neutrino species is established well before the big-bang nucleosynthesis (BBN) epoch if the neutrino
oscillation parameters are in the range indicated by the atmospheric neutrino data and by the large
mixing angle (LMA) MSW solution of the solar neutrino problem. For the other solutions of the
solar neutrino problem, partial flavor equilibrium may be achieved if the angle 13 is close to the
experimental limit tan2 13 0.065. In the LMA case, the BBN limit on the e degeneracy parameter,
| | 0.07, now applies to all flavors. Therefore, a putative extra cosmic radiation contribution from
degenerate neutrinos is limited to such low values that it is neither observable in the large-scale
structure of the universe nor in the anisotropies of the cosmic microwave background radiation.
Existing limits and possible future measurements, for example in KATRIN, of the absolute neutrino
mass scale will provide unambiguous information on the cosmic neutrino mass density, essentially
E-mail address: pastor@mppmu.mpg.de (S. Pastor).

1 Also at: ITEP, Bol. Cheremushkinskaya 25, Moscow 117259, Russia.
2 Also at: Institute of Nuclear Research and Nuclear Energy, Bulgarian Academy of Sciences, 1784 Sofia,
Bulgaria.
3 Also at: Institute for Nuclear Research of the Academy of Sciences of Russia, 60th October Anniversary
Prospect 7a, Moscow 117312, Russia.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 2 7 4 - 2
364
A.D. Dolgov et al. / Nuclear Physics B 632 (2002) 363382
free of the uncertainty of the neutrino chemical potentials. 2002 Elsevier Science B.V. All rights
reserved.
PACS: 14.60.Pq; 26.35.+c
Keywords: Physics of the early universe; Neutrino physics
1. Introduction
The cosmic matter and radiation inventory is known with ever increasing precision,
but many important questions remain open. The cosmic neutrino background is a generic
feature of the standard hot big-bang model, and its presence is indirectly established
by the accurate agreement between the calculated and observed primordial light-element
abundances. However, the exact neutrino number density is not known as it depends on the
unknown chemical potentials for the three flavors. (In addition, there could be a population
of sterile neutrinos, a hypothesis that we will not discuss here.) The standard assumption is
that the asymmetry between neutrinos and antineutrinos is of order the baryon asymmetry
(nB nB)/n 6 1010 . This would be the case, for example, if B L = 0 where
B and L are the cosmic baryon and lepton asymmetries, respectively. While B L = 0 is
motivated by scenarios where the baryon asymmetry is obtained via leptogenesis [1], there
are models for producing large L and small B [25].
In order to quantify a putative neutrino asymmetry we assume that well before thermal
neutrino decoupling a given flavor is characterized by a FermiDirac distribution with
a chemical potential , f (p, T ) = [exp(Ep /T ) + 1]1 , where /T is the
degeneracy parameter and Ep p since we may neglect small neutrino mass effects on
the distribution function. For antineutrinos the distribution function is given by = .
A neutrino chemical potential modifies the outcome of primordial nucleosynthesis in
two different ways [6]. The first effect appears only in the electron sector because electron
neutrinos participate in the beta processes which determine the primordial neutron-toproton ratio so that n/p exp(e ).4 Therefore, a positive e decreases Yp , the primordial
4 He mass fraction, while a negative increases it, leading to an allowed range
e
0.01 < e < 0.07,
(1)
compatible with e = 0 (see Refs. [710] and Section 5). A second effect is an increase of
the neutrino energy density for any non-zero which in turn increases the expansion rate
of the universe, thus enhancing Yp . This applies to all flavors so that the effect of chemical
potentials in the or sector can be compensated by a positive e . Altogether the bigbang nucleosynthesis (BBN) limits on the neutrino chemical potentials are thus not very
restrictive.
Another consequence of the extra radiation density in degenerate neutrinos is that it
postpones the epoch of matter-radiation equality. In the cosmic microwave background
4 We use the notation etc. to avoid double subscripts. We never discuss charged-lepton chemical
e
e
potentials so that there should be no confusion.
365
radiation (CMBR) it boosts the amplitude of the first acoustic peak of the angular power
spectrum and shifts all peaks to smaller scales. Moreover, the power spectrum of density
fluctuations on small scales is suppressed [11,12], leading to observable effects in the
cosmic large-scale structure (LSS).
A recent analysis of the combined effect of a non-zero neutrino asymmetry on BBN and
CMBR/LSS yields the allowed regions [13]
0.01 < e < 0.22,
|, | < 2.6,
(2)
in agreement with similar bounds in [14,15]. These limits allow for a very significant radiation contribution of degenerate neutrinos, leading many authors to discuss the implications
of a large neutrino asymmetry in different physical situations. These include the explanation of the former discrepancy between the BBN and CMBR results on the baryon asymmetry [16] or the origin of the cosmic rays with energies in excess of the GreisenZatsepin
Kuzmin cutoff [17]. In addition, if the present relic neutrino background is strongly degenerate, it would enhance the contribution of massive neutrinos to the total energy density [18,
19] and affect the flavor oscillations of the high-energy neutrinos [20] which are thought
to be produced in the astrophysical accelerators of high-energy cosmic rays.
The limits in Eq. (2) ignore neutrino flavor oscillations, an assumption which is no
longer justified in view of the experimental signatures for neutrino oscillations by solar and
atmospheric neutrinos. For zero initial neutrino chemical potentials, the flavor neutrinos
have the same spectra so that oscillations produce no effect. This is true up to a small
spectral distortion caused by the heating of neutrinos from e+ e annihilations, an effect
which is different for electron and muon/tau neutrinos and which causes a small relative
change in the final production of 4 He of order 103 [21]. This relative change is slightly
enhanced by neutrino flavor oscillations [22,23]. In the presence of neutrino asymmetries,
flavor oscillations equalize the neutrino chemical potentials if there is enough time for this
relaxation process to be effective [24]. If flavor equilibrium is reached before BBN, then
the restrictive limits on e in Eq. (1) will apply to all flavors, in turn implying that the
cosmic neutrino radiation density is close to its standard value. As a consequence, it is no
longer necessary to use the neutrino radiation density as a fit parameter for CMBR/LSS
analyses, unless one considers exotic models with decaying massive particles.
The effects of flavor oscillations on possible neutrino degeneracies have been considered
in [20], where it was concluded that flavor equilibrium was achieved before the BBN epoch
if the solar neutrino problem was explained by the large-mixing angle (LMA) solution. The
LMA solution is favored by the current solar neutrino data [25]. Thus, it was concluded
that in the LMA case a large cosmic neutrino degeneracy was no longer allowed.
We revisit this problem because the flavor evolution of the neutrino ensemble is more
subtle than previously envisaged if medium effects are systematically included. Contrary
to the treatment of Ref. [20], the refractive effect of charged leptons cannot be ignored, and
actually is one of the dominant effects. While the background neutrinos produce an even
larger refractive term, its effect is to synchronize the neutrino oscillations [26,27] which
remain sensitive to the charged-lepton contribution. Still, equilibrium is essentially, but not
completely, achieved in the LMA case so that our final conclusion is qualitatively similar
to that of Ref. [20].
366
One counter-intuitive subtlety is that the neutrino self-potential actually can suppress
oscillations in a situation where the excess of neutrinos in one flavor is exactly matched
by an excess of antineutrinos in another flavor. In this case the synchronized oscillation
frequency is zero so that oscillations begin only once the cosmic expansion has diluted
the self-term. Therefore, one can construct cases where flavor equilibrium is not achieved
before BBN even in the LMA case. However, this is only possible for specially chosen
initial conditions where | | is equal for all flavors, but the absolute signs may be different.
For the purpose of deriving limits on | |, however, this case is equivalent to the one where
equilibrium is achieved, the only important point being that | | is approximately equal for
all flavors at the BBN epoch.
Another subtlety appears in a full three-flavor analysis. We show that achieving
equilibrium in the LMA case does not depend on the value of the mixing angle 13 , to
which strict limits from reactor experiments apply. Moreover, for the non-LMA solutions
of the solar neutrino problem, partial flavor equilibrium may be reached if the angle 13 is
small but non-zero.
In Section 2 we set up the formalism to study primordial flavor oscillations. We then
turn in Section 3 to the primordial flavor evolution of a simplified system where e mixes
maximally with one other flavor. This two-flavor case will illustrate many of the important
subtleties of our problem. Then we turn in Section 4 to realistic three-flavor situations
which involve yet further complications. In Section 5 we finally derive new limits on the
degeneracy parameters and summarize our findings.
2. Neutrino flavor oscillations in the early universe

In order to study neutrino oscillations in the early universe we characterize the neutrino
ensemble in the usual way by generalized occupation numbers, i.e., by 3 3 density
matrices for neutrinos and antineutrinos as described in [28,29]. The form of the density
matrices for a mode with momentum p is

ee e e
ee e e
(p, t) = e ,
(3)
(p,
t) = e ,
e
e
where overbarred quantities refer to antineutrinos. The diagonal elements of the density
matrices correspond to the usual occupation numbers of the different flavors. The definition
of the density matrix for antineutrinos is transposed relative to that for neutrinos,
allowing one to write the equations of motion in a compact form [28].
The equations of motion for the density matrices relevant for our situation of interest
are [28,30]
2

M
8p
it p = +
, p + 2 GF 2 E + , p + C[p ],
2p
3mW
2

M
8p
, p + 2 GF + 2 E + , p + C[p ],
it p =
(4)
2p
3mW
367
where we use the notation p = (p, t) and [ , ] is the usual commutator. Further, M 2 is
the mass-squared matrix in the flavor basis; in the mass basis it would be diag(m21 , m22 , m23 ).
The diagonal matrix E represents the energy densities of charged leptons. For example,
Eee is the energy density of electrons plus that of positrons. The density matrix is the
integrated neutrino density matrix so that, for example, ee is the e number density while
is non-linear in the neutrino
ee is the e number density. The term proportional to ( )
density matrix and represents self-interactions. Finally, C[] is the collision term which
is proportional to G2F . We have neglected a refractive term proportional to the neutrino
energy density which is much smaller than the ( )
term in our present situation where
the neutrino asymmetries are assumed to be large. On the other hand, we have neglected the
usual term which is proportional to the charged-lepton asymmetries. This asymmetric term
is always negligible; at early times (high temperatures) it is negligible compared to the E
term, while at temperatures near n/p freeze out (T 1 MeV) it is negligible compared to
the vacuum term M 2 /2p if the mass-squared differences coincide with those characterizing
the atmospheric neutrino and the solar neutrino LMA oscillations.
In the expanding universe we need to substitute t t Hpp with H the cosmic
expansion parameter. We have taken this into account rewriting the equations in terms
of comoving variables, as listed in Appendix A. We solve these equations numerically,
calculating the evolution of and on a grid of neutrino momenta.
3. Two-flavor oscillations
3.1. Equations of motion for the polarization vectors
In order to develop a first understanding of flavor oscillation in the early universe it
will be useful to study first a two-flavor situation involving e and . The usual relation
between flavor and mass eigenstates is
e = cos 1 + sin 2 ,
= sin 1 + cos 2 .
The 2 2 density matrices are

ee e
(p, t) =
=
e

ee e
(p,
t) =
=
e
(5)

1
P0 (p, t) + P(p, t) ,
2

1
P(p, t) .
P0 (p, t) +
2
(6)
Here, i are the Pauli matrices while P(p, t) and

P(p, t) are the usual polarization vectors
for the neutrino and antineutrino modes p, respectively. We normalize and to a Fermi
Dirac distribution with zero chemical potential, f (p) = [exp(p/T ) + 1]1 , so that for
instance fa (p, t) = aa (p, t)f (p). For the polarization vectors, the equations of motion
are given by the usual spin-precession formula

8 2 GF p
#m2
P
B
E
+
2 GF (P
P ) Pp + C[Pp ],
t Pp = +
ee
p
2p
3m2W
368

8 2 GF p
#m2

P
B
t Pp =
E
+
2 GF (P
P)
Pp + C[
Pp ],
ee
p
2p
3m2W
(7)
where #m2 = m22 m21 and B = (sin 2, 0, cos 2 ) with the vacuum mixing angle.
Further, z is a unit vector in the z-direction, Eee the electronpositron energy density, and
P and
P are the integrated polarization vectors.
We will see that a detailed treatment of the collision terms is not crucial. Therefore, we
approximate them with a simple damping prescription of the form
C[Pp ] = Dp PTp ,
(8)
where the transverse part PTp consists of the (xy)-projection of Pp . We use the damping
functions as, for instance, in [29,31], but with slightly modified coefficients. We neglect
a small dependence on , i.e., we use the same damping functions for neutrinos and
antineutrinos. We have checked that our results are insensitive to the inclusion of the
as given, for instance, in [29,31]. Moreover, we
repopulation function, i.e., P0 and P
0
did not include the neutrino heating from e+ e annihilations, since it is small even in the
presence of degeneracies [32].
3.2. Evolution of simple flavor asymmetry
As a first case we consider a simple situation where initially electron neutrinos do not
have a chemical potential (e = 0) while muon neutrinos are asymmetric ( = 0.05).
We use a rather small asymmetry so that the anticipated equilibrium state e = will
be close to the BBN limit on e in Eq. (2). The quantitative evolution with a large is the
same. Moreover, we use maximal mixing and #m2 = 4.5 105 eV2 as suggested by the
LMA solution of the solar neutrino problem.
In order to illustrate the relative importance of the various contributions to the equation
of motion we first calculate the primordial evolution of the integrated polarization vectors
P and
P for pure vacuum oscillations, including only the #m2 /2p terms in Eq. (7), without
any medium effects whatsoever. Our initial conditions e = 0 and = 0.05 imply that
there are more e than , i.e., P points initially in the +z direction, while there are
more than e , i.e.
P points initially in the z direction. In the upper panel of Fig. 1
we show the evolution of Pz as a function of cosmic temperature; the evolution of Pz is
similar, except with a negative initial value. The oscillations begin when the expansion rate
has become slow enough at about T 30 MeV. The oscillation frequencies are different
for different modes, leading to quick decoherence and thus to an incoherent equal mix
of both flavors. Evidently flavor equilibrium is reached long before n/p freeze-out at
T 1 MeV.
As a next step we include the neutrino self-potential proportional to P
P. The evolution
is shown as a dotted line in the upper panel of Fig. 1. It is now dominated by the effect
of synchronized oscillations, i.e., the self-potential forces all neutrino modes to follow
the same oscillation [27]. Because P and
P point in opposite directions, the common or
synchronized oscillation frequency synch is very much smaller than a typical #m2 /2p.
369
Fig. 1. Evolution of the z-component of the integrated polarization vector P for different situations as explained
in the text.
Explicitly we have [27]

synch =
1
|I|
#m2
#m2
d 3p

f
(p)
+
P
)
,
I
(P
p
p
(2)3
2p
2peff
(9)
where I P
P and I is a unit vector in the direction of I. Numerically our initial
conditions imply peff 132 T .
Next, we add to the vacuum term the effective potential caused by the e background
[see Eq. (4)], without including the self-term. The e potential always points in the
z-direction, thereby suppressing flavor oscillations in the usual way. This effect is shown
in the lower panel of Fig. 1 (solid line). P stays frozen at its initial value up to T 3 MeV
where the medium potential becomes unimportant compared to the vacuum term. At these
temperatures collisions are also about to become unimportant. Therefore it is not surprising
that including collisions (dotted line) does not dramatically change the evolution of Pz .
However, this impression is misleading as can be seen from Fig. 2 where we show the
evolution of Px for the same cases as in the lower panel of Fig. 1. Without collisions (solid
line) the evolution of P is a simple turning from the z- to the x-direction. The e potential
adiabatically disappears, leading to the usual MSW-type evolution. In this case the final
result corresponds to equal numbers of both flavors, yet a coherent flavor superposition.
With collisional damping (dotted line) the x-component is damped, leaving little flavor
coherence of the final state. Therefore, with or without collisional damping we reach
flavor equilibrium in the sense of equal densities of both flavors, but only collisions
ensure the damping of the otherwise large transverse part of the final P.
370
Fig. 2. Evolution of the x-component of the integrated polarization vector P for the same cases as in the lower
panel of Fig. 1.
Fig. 3. Evolution of the z-component of the integrated polarization vector P in the presence of the vacuum term,
the e+ e potential, collisional damping, and varying strength of the neutrino self-interaction.
Finally, we study how oscillations evolve in the presence of all effects: vacuum, e
background, neutrino self-potential and collisions. The results are shown in Fig. 3 for
different strengths of the self-potential. In one run it was switched off (No Self), it was
taken at full strength (Self), or it was suppressed by a factor 104 or 105 . When the selfinteraction term is close to its real value (solid line), again all modes oscillate in synch
so that we see a combination of the synchronized oscillations and the MSW-like effect of
the background medium. However, collisions ensure that no large transverse component
survivessee the corresponding evolution of Px and Py in Fig. 4. Almost perfect flavor
equilibrium is achieved with or without self-interactions.
This conclusion is not to be taken for granted. When the self-interaction term is
artificially adjusted to be of comparable strength to the other terms (vacuum or e
background), the evolution shows complicated features (Fig. 3). After some initial
conversion, Pz remains constant for a long time, only to reach equilibrium at a much
later time. This behavior is not easy to understand as it depends on all the ingredients
of our calculation, i.e., the background potential, the neutrino self-potential and damping
by collisions. We believe that the long phase where the polarization vector essentially
stands still is caused by the synchronized oscillation frequency becoming very small.
This is strictly the case in a situation where the chemical potentials for two flavors are
equal but oppositesee Section 3.3. We believe that in the present case a combination
371
Fig. 4. Evolution of the x and y components of the integrated polarization vector P in the presence of all effects
and full-strength neutrino self-interactions, corresponding to the solid line in Fig. 3.
of decoherence of some of the modes and collisions drives the system into a state where
the synchronized oscillations frequency vanishes so that oscillations stop until the cosmic
expansion has diluted the neutrino self-term enough to eliminate this effect.
3.3. Equal but opposite asymmetries
Our discussion thus far suggests that for the parameters of interest it is generic to achieve
flavor equilibrium before the BBN epoch, even though the loss of coherence may not
always be complete. However, this conclusion depends on the simple initial conditions that
we have used thus far. The different neutrino flavors may show large asymmetries, yet the
total cosmic lepton asymmetry could still be small if initially e = , which corresponds
P
to conservation of the lepton number Le L . These initial conditions imply that P =
Pp ) = 0 [27].
so that the self-potential term (P
P ) is large while for each mode (Pp +
This implies, in turn, that the synchronized oscillation frequency synch = 0. Therefore, in
this case the synchronization effect caused by the self-potential prevents flavor oscillations
entirely, at least until the self-term becomes weak, long after the BBN epoch. While flavor
equilibrium is not achieved in this case, from the start we have |e | = | | so that the BBN
limits on |e | would apply to both flavors.
4. Three-flavor oscillations
4.1. Oscillation parameters
We now turn to the more interesting case of three-flavor neutrino oscillations. The
neutrino flavor eigenstates e , , and are related to the mass eigenstates via the mixing
matrix

s12 c13
s13
c12 c13
s12 c23 c12 s23 s13 c12 c23 s12 s23 s13 s23 c13 .
(10)
s12 s23 c12 c23 s13 c12 s23 s12 c23 s13 c23 c13
Here cij = cos ij and sij = sin ij for ij = 12, 23, or 13, and we have assumed
CP conservation. The set of oscillation parameters is now five-dimensional (see, for
372
instance, [33]),
#m2sun #m221 = m22 m21 ,
#m2atm #m232 = m23 m22 ,
sun 12 ,
atm 23 ,
13 .
(11)
We do not perform a global analysis of all possible values of these parameters, but fix
them to be in the regions that solve the atmospheric and solar neutrino problems [33,34].
In particular, we take #m2atm = 3 103 eV2 and maximal mixing for atm from the
former, while from the solar analyses we consider the following values for #m2sun in eV2 :
4.5 105 , 7 106 , 1 107 , 8 1011 for the Large Mixing Angle (LMA), Small
Mixing Angle (SMA), LOW and Vacuum regions, respectively. For the angle sun we take
the approximation of maximal mixing for all cases except SMA where we use sun = 1.5 .
4.2. Simple three flavor case
We begin with the simplified situation where initially only the muon neutrinos are asymmetric ( = 0.1). We perform a full three-flavor calculation, but for now set 13 = 0. The
evolution of the neutrino asymmetries is shown in Figs. 5 and 6 for the LMA and LOW
cases, respectively, both with and without the neutrino self-interactions. For this choice of
oscillation parameters the three-flavor oscillations effectively separate as two two-flavor
problems for the atmospheric and solar parameters, respectively. The oscillations caused
by the largest #m2 are effective at T 20 MeV, as soon as the background disappears
Fig. 5. Evolution of the neutrino degeneracy parameters for the LMA case, 13 = 0, and the initial values
e = = 0 and = 0.1.
373
Fig. 6. Evolution of the neutrino degeneracy parameters for the LOW case, 13 = 0, and the initial values
e = = 0 and = 0.1.
completely. The presence of the self-term causes only a slight delay in the equilibration of
and .
The oscillations due to #m2sun and sun are effective only when the vacuum term overcomes the e potential. In the LMA case, the conversions takes place above T 1 MeV,
leading to nearly complete flavor equilibrium before the onset of BBN. For the LOW parameters the synchronized oscillations just start at that epoch. The presence of the neutrino self-potential does not significantly change the picture in the LMA case while for
the LOW case one clearly observes the phenomenon of synchronized oscillations. For the
SMA and Vacuum regions primordial oscillations involving e are not effective before
BBN if 13 = 0.
4.3. Non-zero 13 mixing
The angle 13 is restricted to the approximate region tan2 13 0.065 (see, for instance,
[35]) from a combined analysis of solar, atmospheric and reactor (CHOOZ) data. However,
a small but non-zero 13 does modify the oscillation behavior. This effect is shown in Fig. 7
for the LMA case and different values of 13 . Even small values of 13 lead to conversion
to the electronic flavor at larger temperatures and enhance flavor equilibration if sun is in
the LMA region.
The effect of a non-zero 13 is more important if the solar neutrino problem is solved by
oscillations with parameters in a region other than LMA. This can be seen in Fig. 8, where
we have calculated the evolution in the limit #m2sun 0 and sun 0, thus corresponding
approximately to the SMA, LOW and Vacuum regions. For values of tan2 13 close to
the limit discussed at the beginning of this section, neutrino oscillations lead to almost
complete flavor equilibration, while for smaller 13 angles the conversion is only partial.
374
Fig. 7. Evolution of the neutrino degeneracies for different values of tan2 13 in the LMA case for the initial
conditions e = = 0 and = 0.1.
Fig. 8. Evolution of the neutrino degeneracies for different values of tan2 13 in the limit #m2sun 0 and sun 0
for the initial conditions e = = 0 and = 0.1.
4.4. Can the self-term prevent flavor equilibrium?

In the two-flavor case of Section 3.3 we saw that for equal but opposite asymmetries
of the two flavor distributions the synchronized oscillation frequency was strictly zero,
375
Fig. 9. Evolution of the neutrino degeneracies if initially e = 0 and = = 0.1, with or without neutrino
self-interactions.
suppressing oscillations entirely before the BBN epoch. In the three-flavor case the
situation is more complicated so that again we need to raise the question if there is a
special configuration of initial neutrino distributions which suppresses flavor oscillations
by the neutrino self-potential.
In analogy to Section 3.3 we first consider situations where the asymmetries of two
flavors are equal but opposite, while the third asymmetry is strictly zero. In this case
one expects that oscillations between the asymmetric flavors are suppressed by the selfpotential while oscillations into the third flavor are unimpeded. We show the numerical
results for two specific cases. In Fig. 9 we take = = 0.1 and e = 0. The
atmospheric oscillations between and are indeed blocked in the presence of selfinteractions while the solar oscillations, here the LMA case and tan2 13 = 0, proceed
at the appropriate temperature. In Fig. 10 we take = 0 and e = = 0.1. Both the
atmospheric and solar oscillations take place and equilibrate the flavors essentially as in
the previous cases with a simple initial condition.
The question remains if one could devise special initial conditions such that | | and
| | are large while |e | is small, and that this system avoids flavor oscillations by reducing
the synchronized oscillation frequency to a very small value. In this case our new limits
would not apply.
We do not believe that such a case can be constructed. To argue in favor of this
claim we first note that the most general initial condition consists of thermal equilibrium
distributions which are diagonal in flavor space, i.e., in the weak-interaction basis the
initial 33 density matrices for neutrinos and antineutrinos are diagonal and characterized
by FermiDirac distributions. All off-diagonal elements would be quickly damped by
reactions which involve muons and electrons and thus measure the flavor content of
376
Fig. 10. Evolution of the neutrino degeneracies for the LMA case with 13 = 0 with initial conditions = 0 and
e = = 0.1, with or without neutrino self-interactions.
the participating neutrinos. Therefore, the most natural initial condition for our problem is
indeed characterized by e , and .
The general equations of motion Eq. (4) reveal that the quantity which develops in
a synchronized fashion is ,
in agreement with the discussion of 2 2 synchronized
oscillations in Ref. [27] where the picture of polarization vectors and the associated
internal magnetic fields of the system was used. Therefore, the synchronized equation of
motion is obtained by subtracting the two lines of Eq. (4) and integrating over all modes,

8p
d 3p M 2
d 3p
,
(
+
2
G
E,
(
+
)
p
p
F
p
p
(2)3 2p
(2)3 3m2W

( )
,
+ 2 GF ( ),
(12)

it ( )
=
where we have dropped the collision term. The last term, of course, is identically zero.
The second term vanishes also as long as the density matrices are diagonal in the flavor
basis since the matrix E is diagonal in that basis. Only the first term, involving the mass
matrix, represents a force such as to move away from being diagonal in the flavor
basis, i.e., which causes synchronized flavor oscillations. This force is identically zero if
the matrix

d 3 p p + p
(13)
(2)3
2p
is proportional to the unit matrix. As the matrices p and p are initially diagonal and given
by FermiDirac distributions, the integral can be solved explicitly, leading to an expression
proportional to
2

e
1
3
1
+ 2
377
(14)
If all are initially equal, then the neutrinos are already in flavor equilibrium and indeed
oscillations trivially do not operate.
The only non-trivial configuration where synchronized oscillations proceed with a
vanishing frequency is when one of the chemical potentials is opposite in sign to the others.
We have checked numerically that indeed no flavor conversion arises in this case as long
as the self-potential is large, i.e., as long as we are in the synchronized regime. In the
two-flavor case this corresponds to the situation discussed in Section 3.3.
As in the two-flavor case we conclude that it is possible to avoid flavor equilibrium
by specially chosen initial conditions, but these conditions require |e | = | | = | |.
Therefore, the strict BBN limits on |e | apply to all flavors.
5. New limits on neutrino degeneracy

We conclude that in the LMA region the neutrino flavors essentially equilibrate long
before n/p freeze out, even when 13 is vanishingly small. For the other cases the outcome
depends on the magnitude of 13 . In the LMA case it is thus justified to derive new limits
on the cosmic neutrino degeneracy parameters under the assumption that all three neutrino
flavors are characterized by a single degeneracy parameter, independently of the primordial
initial conditions. We do not derive the corresponding limits for the other solar neutrino
solutions, since they would strongly depend on the value of a non-zero 13 . However, if that
angle is close to the experimental limit, the bounds that we describe would approximately
apply.
We first note that the energy density in one species of neutrinos and antineutrinos with
degeneracy parameter is

7 2
30 2 15 4
+
.
1+
= T4
(15)
120
7
7
It is clear that the BBN limit will imply 1 for all flavors so that the modified energy
density and the resulting change of the primordial helium abundance Yp will be negligibly
small. If there are additional relativistic species, such as sterile neutrinos or majorons, then
Eq. (2) will simply apply to all the active neutrinos
| | < 0.22.
(16)
Therefore, the only remaining BBN effect is the shift of the beta equilibrium by e . We
recall that Yp is essentially given by n/p at the weak-interaction freeze-out, and that
n/p exp(e ) 1 e where the latter expansion applies for |e | 1. Therefore,
#Yp Yp (1 Yp /2)e 0.21e . Modifications of Yp by new physics are frequently
expressed in terms of the equivalent number of neutrino flavors #N which would
cause the same modification due to the changed expansion rate at BBN. If the radiation
378
density at BBN is expressed in terms of N , the helium yield can be expressed by the
empirical formula #Yp = 0.012#N [36]. Therefore, the effect of a small e on the helium
abundance is equivalent to #N 18e . A conservative standard limits holds that BBN
implies |#N | < 1 which thus translates into |e | 0.057.
A more detailed recent analysis reveals that the measured primordial helium abundance
implies a 95% CL range N = 2.5 0.8 or #N = 0.5 0.8 [8,13]. We conclude that
the BBN-favored range for the electron neutrino degeneracy parameter is at 95% CL
e = 0.03 0.04.
(17)
If all degeneracy parameters are the same, then this range applies to all flavors.
It should be noted that the actual limit we obtain on the neutrino degeneracy depends on
the adopted BBN analysis. For instance #N could be as high as 1.2 when the primordial
abundance of lithium is used instead of that of deuterium [37]. At any rate, a limit of
|e | 0.1 seems rather conservative and does not modify our conclusions.
Using | | < 0.07 as a limit on the one degeneracy parameter for all flavors, the extra
radiation density is limited by (# )/ < 3 0.0021 = 0.0064, i.e., #N < 0.0064. If
the same radiation density were to be produced by the asymmetry of one single species,
this would correspond to | | < 0.12.
For comparison with the future satellite experiments MAP and PLANCK that will measure the CMBR anisotropies, it was calculated that they optimistically will be sensitive to a
single-species above 0.5 and 0.25, respectively [38]. However, with proper consideration
of the degeneracy with the matter density, M , and the spectral index, n, a more realistic
sensitivity is 2.4 and 0.73, respectively [39]. Turning this around we conclude that our
new limits are so restrictive that the CMBR is certain to remain unaffected by neutrino
degeneracy effects so that | | can be safely neglected as a fit parameter in future analyses.
If our new limits apply the number density of relic neutrinos is very close to its standard
value. Therefore, existing limits and possible future measurements of the absolute neutrino
mass scale, for example in the forthcoming tritium decay experiment KATRIN [40], will
provide unambiguous information on the cosmic mass density in neutrinos, free of the
uncertainty of neutrino chemical potentials.
Acknowledgements
We thank Alexei Smirnov, Gary Steigman, Karsten Jedamzik and Gianpiero Mangano
for useful discussions and comments. In Munich, this work was partly supported by
the Deutsche Forschungsgemeinschaft under grant No. SFB 375 and the ESF network
Neutrino Astrophysics. S.H. Hansen and S. Pastor are supported by Marie Curie
fellowships of the European Commission under contracts HPMFCT-2000-00607 and
HPMFCT-2000-00445.
Appendix A. Evolution equations
In this appendix we list in detail the evolution equations for the neutrino and antineutrino
density matrices in the two-flavor case as in Eq. (5), while the generalization to the threeflavor case that we consider in Section 4 is straightforward.
379
In our treatment of primordial neutrino oscillations we use the following dimensionless

expansion rate and momenta
x mR,
y pR,
(A.1)
where R is the universe scale factor and m an arbitrary mass scale that we choose to be
1 MeV. The neutrino and antineutrino density matrices are

1 P0 + Pz Px iPy
1
(x, y) = P0 (x, y) + P(x, y) =
(A.2)
,
2
2 Px + iPy P0 Pz

1 P0 + Pz Px i Py
1
(x,
y) = P0 (x, y) +
(A.3)
P(x, y) =
y P0 Pz
2
2 Px + i P
with the derivatives
d
d
(x, y) =
(t, p),
dx
dt
d
d
H x (x, y) =
(A.4)
(t, p).
dx
dt
The initial conditions are for large temperatures as follows. For density matrices
normalized to feq (y) = (ey + 1)1 and initial degeneracies = and = (for
flavor neutrinos and )
Hx
feq (y, ) + feq (y, )

feq (y, ) feq (y, )
,
Pz (y) =
,
feq (y)
feq (y)
feq (y, ) + feq (y, )
feq (y, ) feq (y, )
P0 (y) =
,
Pz (y) =
,
feq (y)
feq (y)
(A.5)
where feq (y, ) = 1/[exp(y ) + 1]. Finally,
P0 (y) =
Px (y) = Py (y) = Px (y) = Py (y) = 0.

The components of the neutrino polarization vectors evolve as
dP0
(x, y) = R (x, y) + R (x, y),
dx

#m2
Vl
D
dPx
(x, y) =
cos 2
Px (x, y),
Py (x, y)
dx
2pH x
Hx
Hx

dPy
#m2
Vl
D
(x, y) =
cos 2
Py (x, y)
Px (x, y)
dx
2pH x
Hx
Hx
#m2
sin 2 Pz (x, y),
2pH x
dPz
#m2
(x, y) =
sin 2 Py (x, y) + R (x, y) R (x, y),
dx
2pH x
d P0
(x, y) + R
(x, y),
(x, y) = R
dx
(A.6)
380

#m2
Vl
d Px
D
(x, y) =
cos 2
Py (x, y)
Px (x, y),
dx
2pH x
Hx
Hx

d Py
#m2
Vl
D
(x, y) =
cos 2
Px (x, y)
Py (x, y)
dx
2pH x
Hx
Hx
+
#m2
z (x, y),
sin 2 P
2pH x
d Pz
#m2
(x, y) R
(x, y),
(x, y) =
sin 2 Py (x, y) + R
dx
2pH x
(A.7)
where

D feq (y, ) 1
P0 (x, y) + Pz (x, y) ,
R (x, y) = 2
Hx
feq (y)
2

D feq (y, ) 1
P0 (x, y) Pz (x, y) ,
R (x, y) = 2
Hx
feq (y)
2
(A.8)
, given by the same expressions with . However, our numerical results do

with R
not change significantly if the total neutrino and antineutrino number densities are taken to
be constant (dP0 /dx = d P0 /dx = 0).
The different terms in Eqs. (A.7) are given as follows. The vacuum oscillation terms are
proportional to

#m2
1010MP #m2 1 x 2
=
,
2pH x 2 8/3 eV2

y
(A.9)
where MP 1.221, = (x/m)4 tot , and tot is the total energy density of the universe.
The l + l background with l = , e the charged leptons (for - and e -, oscillations,
respectively) is described by
Vl
8 2 105MP GF 1 y
(A.10)
=
4 (l + + l ),
Hx
x
3 8/3 m2W
where GF 1.1664 and mW 80.42. Collisions are described by damping terms where,
according to our calculations,

D 2 4 sin4 W 2 sin2 W + 2 F0 ,

D 2 2 sin4 W + 6 F0 ,
(A.11)
for and e , oscillations, respectively, where all collision terms are proportional
to
MP G2F (3)ymed 1 y
F0
=
4
Hx
3 3 8/3
x
and (3) 1.20206 and ymed 3.15137.
(A.12)
381
Finally one has to add the contribution from the neutrinoantineutrino background [28
30]

Vsym

Vasym
J(x)
J(x)
U(x) +
U(x) P(x, y),
Hx
Hx

Vasym
Vsym
(A.13)
J(x)
J(x) +
U(x) +
U(x)
P(x, y),
Hx
Hx
for dP/dx and d
P/dx, respectively. Here we have defined

1
1
2
U(x) =
du u P(x, u),
du u3 P(x, u),
J(x) =
2 2
2 2
and the same for
J and
U. Moreover,
Vasym
2 1011MP GF 1
=
2,
Hx
8/3
x
5
Vsym 8 2 10 MP GF
1 y
=
cos2 W 4 .
2
Hx
x
3 8/3 mW
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
W. Buchmller, M. Plmacher, Int. J. Mod. Phys. A 15 (2000) 5047.

J.A. Harvey, E.W. Kolb, Phys. Rev. D 24 (1981) 2090.
A.D. Dolgov, D.P. Kirilova, J. Moscow Phys. Soc. 1 (1991) 217.
A. Casas, W.Y. Cheng, G. Gelmini, Nucl. Phys. B 538 (1999) 297.
A.D. Dolgov, Phys. Rep. 222 (1992) 309.
S. Sarkar, Rep. Prog. Phys. 59 (1996) 1493.
H. Kang, G. Steigman, Nucl. Phys. B 372 (1992) 494.
S. Esposito, G. Mangano, G. Miele, O. Pisanti, JHEP 0009 (2000) 038.
S. Esposito, G. Mangano, A. Melchiorri, G. Miele, O. Pisanti, Phys. Rev. D 63 (2001) 043004.
P. Di Bari, Phys. Rev. D 65 (2002) 043509.
J. Lesgourgues, S. Pastor, Phys. Rev. D 60 (1999) 103521.
W. Hu, D.J. Eisenstein, M. Tegmark, M.J. White, Phys. Rev. D 59 (1999) 023512.
S.H. Hansen, G. Mangano, A. Melchiorri, G. Miele, O. Pisanti, Phys. Rev. D 65 (2002) 023511.
S. Hannestad, Phys. Rev. D 64 (2001) 083002.
J.P. Kneller, R.J. Scherrer, G. Steigman, T.P. Walker, Phys. Rev. D 64 (2001) 123506.
J. Lesgourgues, M. Peloso, Phys. Rev. D 62 (2000) 081301.
G. Gelmini, A. Kusenko, Phys. Rev. Lett. 82 (1999) 5202.
P.B. Pal, K. Kar, Phys. Lett. B 451 (1999) 136.
J. Lesgourgues, S. Pastor, S. Prunet, Phys. Rev. D 62 (2000) 023001.
C. Lunardini, A.Yu. Smirnov, Phys. Rev. D 64 (2001) 073006.
A.D. Dolgov, S.H. Hansen, D.V. Semikoz, Nucl. Phys. B 503 (1997) 426.
P. Langacker, S.T. Petcov, G. Steigman, S. Toshev, Nucl. Phys. B 282 (1987) 589.
S. Hannestad, Phys. Rev. D 65 (2002) 083006.
M.J. Savage, R.A. Malaney, G.M. Fuller, Astrophys. J. 368 (1991) 1.
Q.R. Ahmad et al., SNO Collaboration, nucl-ex/0204009.
S. Samuel, Phys. Rev. D 48 (1993) 1462.
S. Pastor, G.G. Raffelt, D.V. Semikoz, Phys. Rev. D 65 (2002) 053011.
G. Sigl, G. Raffelt, Nucl. Phys. B 406 (1993) 423.
(A.14)
(A.15)
382
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
B.H. McKellar, M.J. Thomson, Phys. Rev. D 49 (1994) 2710.

J. Pantaleone, Phys. Lett. B 287 (1992) 128.
N.F. Bell, R.R. Volkas, Y.Y. Wong, Phys. Rev. D 59 (1999) 113001.
S. Esposito, G. Miele, S. Pastor, M. Peloso, O. Pisanti, Nucl. Phys. B 590 (2000) 539.
M.C. Gonzlez-Garca, M. Maltoni, C. Pea-Garay, J.W.F. Valle, Phys. Rev. D 63 (2001) 033005.
G.L. Fogli, E. Lisi, A. Marrone, D. Montanino, A. Palazzo, hep-ph/0104221.
A. Bandyopadhyay, S. Choubey, S. Goswami, K. Kar, Phys. Rev. D 65 (2002) 073031.
R.A. Malaney, G.J. Mathews, Phys. Rep. 229 (1993) 145.
K.A. Olive, astro-ph/0202486.
W.H. Kinney, A. Riotto, Phys. Rev. Lett. 83 (1999) 3366.
R. Bowen, S.H. Hansen, A. Melchiorri, J. Silk, R. Trotta, astro-ph/0110636, to be published in Mon. Not.
R. Astron. Soc.
[40] A. Osipowicz et al., KATRIN Collaboration, hep-ex/0109033.

The CardyVerlinde formula for 2D gravity

M. Cadoni a,c , P. Carta a,c , S. Mignemi b,c
a Universit di Cagliari, Dipartimento di Fisica, Cittadella Universitaria, 09042 Monserrato, Italy
b Universit di Cagliari, Dipartimento di Matematica, Viale Merello 92, 09123 Cagliari, Italy
c INFN, Sezione di Cagliari, Cagliari, Italy
Abstract
We discuss the different bounds on entropy in the context of two-dimensional cosmology. We show
that in order to obtain well definite bounds one has to use the scale symmetry of the gravitational
theory. We then extend the recently found relation between the Friedmann equation and the Cardy
formula to the case of two dimensions. In particular, we find that in two dimensions this relation
requires that the central charge c of the conformal field theory is given in terms of the Newton
constant G of the gravitational theory by c = 6/G. 2002 Elsevier Science B.V. All rights reserved.
1. Introduction
In a recent paper [1], Verlinde has discussed the cosmological bounds on entropy for
spacetimes of dimension d > 2. These are based on the holographic principle [2], which
states that the entropy contained into a given region of space should be bounded by the
area of the spacelike surface that encloses it. Another important result of [1] is that the
Friedmann equation of cosmology for a radiation-dominated universe can be shown to be
equivalent to the Cardy formula for the entropy of a conformal field theory describing the
radiation. This observation has of course important implications, which have not been fully
clarified yet.
The context of the original proposal of Verlinde was d > 2 cosmology. Although
the CardyVerlinde formula has been generalized to describe other gravitational systems
[3,4], in particular black holes, a discussion of the d = 2 case is still lacking. The
two-dimensional (2D) limit of the CardyVerlinde proposal is interesting for various
E-mail addresses: mariano.cadoni@ca.infn.it (M. Cadoni), paolo.carta@ca.infn.it (P. Carta),
salvatore.mignemi@ca.infn.it (S. Mignemi).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 2 5 9 - 6
384
M. Cadoni et al. / Nuclear Physics B 632 (2002) 383393
reasons. From investigations of the anti-de-Sitter (AdS)/Conformal Field Theory (CFT)

correspondence, we know that there are 2D gravitational systems that admit 2D CFTs
as duals [5,6]. In this case one can make direct use of the original Cardy formula [7]
to compute the entropy [5,6]. A comparison of these results with a 2D generalization
of the CardyVerlinde formula could be very useful in particular for the understanding
of the puzzling features of the AdS/CFT correspondence in two dimensions [8]. Another
point of interest in extending the CardyVerlinde formula to d = 2 is the clarification of
the meaning of the holographic principle for 2D spacetimes. The boundaries of spacelike
regions of 2D spacetimes are points, so that even the notion of holographic bound is far
from trivial.
A generalization of the work of Verlinde to two spacetime dimensions presents several
difficulties, essentially for dimensional reasons. First of all, in two dimensions one
cannot establish a area law, since black hole horizons are isolated points. Moreover, the
spatial coordinate is not a radial coordinate and hence one cannot impose a natural
normalization on it. As we shall see later on this paper in detail, this fact is connected,
at least for the 2D gravity model we consider here, to a scale symmetry of 2D gravity [9].
Related to this symmetry is also the fact that the 2D gravitational coupling constant G
is dimensionless, and hence one cannot even define a Planck length. Finally, if one
works, as we do in this paper, in the context of scalartensor theories of gravity, the 2D
cosmological equations are quite different from their FriedmannRobertsonWalker d > 2
counterparts.
Some of these problems may of course be solved if one considers gravity in two
dimensions as a dimensionally reduced theory. However, if one wants to keep a purely
two-dimensional point of view, one has to deal with the particular features of 2D gravity.
In this paper, we wish to extend Verlindes results to a radiation-dominated 1 + 1
universe, in which the gravitational interaction is governed by a JackiwTeitelboim (JT)
model [10,11]. We shall see that this goal can be achieved, provided that some free
parameters appearing in the solutions are fixed using the scale symmetry of the theory.
The paper is organized as follows. In Section 2 we discuss the cosmological model
derived from the JT model. In Section 3 we discuss how the standard cosmological bounds
on the entropy can be generalized to our case. In Section 4 we use the scale symmetry of
the gravitational theory to fix the free dimensionless parameters appearing in the bounds.
In Section 5 we investigate the relations between the cosmological equations and the Cardy
formula. The cosmological bounds on the temperature of a radiation-dominated universe
are discussed in Section 6. Finally, in Section 7, we present our conclusions.
2. Two-dimensional cosmology
Let us consider the action for JT gravity minimally coupled to matter,

R 22
+ LM ,
I = d 2 x g
16G
(1)
where is a scalar field (the dilaton), G is the dimensionless 2D Newton constant, which
could be absorbed in a redefinition of the dilaton, 2 is a cosmological constant and LM is
385
the matter Lagrangian. We want to discuss a radiation-dominated 1 + 1 universe, in which

case LM describes free (or weak interacting) massless particles. In general, LM can be
given in terms of a 2D CFT. Because the matter Lagrangian is that of a perfect fluid, it can
be taken proportional to its density, LM = [12]. A constrained variation of (1) gives
the field equations [11]

R = 22 ,
(2)
g 2 + 2 g = 8T ,
where T is the standard energymomentum tensor of a perfect fluid, T = pM g +
( + pM )u u , with pM the pressure of the fluid. The field equation (2) tell us that,
independently of the matter, the spacetime has constant, positive curvature. It is, therefore,
given by a 2D de Sitter (dS) spacetime.
We make the ansatz ds 2 = dt 2 + R 2 (t) dx 2 , = (t), with 0 x 2 . Since we
take x periodic, we are considering a closed 1 + 1 universe. However, our considerations
can be easily extended to a open universe. The field equations then take the form
R 2 R = 0,
R 2 R = 8GR,
2 = 8GpM .
(3)
Combining the field equations, one obtains the energymomentum conservation in the
For a perfect fluid, pM = , and this relation can be integrated

form = (pM + )R/R.
to yield R 1+ = const = M/2 . We are considering the case in which the matter is
constituted of pure radiation, for which = 1, so that we have,
M
.
2R 2
The general solution of the first of Eqs. (3) is
=
(4)
t .
R(t) = ae
t + be
(5)
and with a suitable

Depending on the relative sign of the integration constants a and b,
choice of the origin of time, the solution can assume three qualitatively different forms:
a
(I) R = et ,
(6)
a
(II) R = sinh t,
(7)
a
(III) R = cosh t,
(8)
where a is a dimensionless parameter (we choose this normalization, in order to give to R

the physical dimension of a length).
The solutions for the scalar field are, respectively [11],
(I)
(II)
(III)
4GM 2t
e
,
3a 2

t
4GM
,
= 0 cosh t +
1 + cosh t log tanh
a2
2
4GM
= 0 sinh t
(1 + sinh t arctan sinh t),
a2
= 0 et
(9)
386
with 0 an integration constant. All solutions are of course locally isomorphic to de

Sitter spacetime, but with different parametrization, covering different regions of the 2D
manifold. In particular, I and II possess a horizon at t = and t = 0, respectively.
Moreover, for all solutions I, II and III the scalar field has a zero at a finite value t0 of the
cosmological time t. Since we are dealing with a BransDicke-like theory of gravity, 1
represents a time-dependent effective Newton constant. The instant t = t0 may, therefore,
be interpreted as an initial singularity and we will restrict ourselves to consider only times
t t0 , when 0.
In two dimensions there is no direct analog of the d-dimensional Friedmann equation
H 2 = 2 +
1
16G
2,
(d 1)(d 2)
R
(10)
where H R/R
is the Hubble parameter (notice that the Friedmann equation (10) is
singular for d = 2). However, an equation for H 2 can be obtained by integrating the first
equation in (3):
H 2 = 2 2 .
(11)
R
The constant of integration for the solutions IIII is respectively, 0, a 2 , a 2 . For a
radiation-dominated universe, one can use Eq. (4) into Eq. (11), to obtain an expression
formally similar to (10),
,
(12)
R2
where = + 4GM. Notice that we are using the arbitrariness of the integration
constant to make the metric of the spacetime dependent on the matter. Consistently with
the field equations (3) the effect of the matter on the metric can be only encoded in the
choice of integration constants.
Our gravity model (1) has a cosmological constant different from zero. Therefore, one
can assign to the vacuum an energy E = 2 R/4G and a pressure p = E . This permits
to write Eq. (12) in terms of the total energy E = (E + EM ), where EM = M/R is the
energy of the matter,
H 2 = 2 + 8G
H2 =
4GE
2.
R
R
(13)
3. Entropy bounds
In d > 2 a bound on the entropy of a macroscopic system, S SB , is believed to hold,
where the Bekenstein entropy SB is defined as [13]
2
ER,
(14)
d 1
with E the total energy and R the linear size of the system. This bound is verified for
standard gases, but the numerical factor in front of ER is fixed by the assumption that the
bound is saturated by black holes.
SB =
387
A generalization of this bound to two dimensions is not straightforward. Consider, for

example, the 2D anti-de-Sitter black hole [14], which is a solution of the gravity model (1)
with negative 2 ,

1 2
dx ,
= 0 x.
ds 2 = 2 x 2 m2 dt 2 + 2 x 2 m2
(15)
A horizon occurs at x0 = m/, and one can associate to it the temperature T = m/2
and the entropy S = 20 m. Moreover, by standard methods, one can assign to the black
hole the ADM mass M = 0 m2 /2. Thus one gets the relation
4Mx0
(16)
.
m2
If one identifies the energy E of the black hole with its ADM mass and its size R with
the length x0 , one finds S ER, but the ratio S/ER grows without limit for small m (in the
limit m 0, however, all quantities vanish). The same situation occurs in more general
two-dimensional models. The problem is connected to the fact that in two dimensions
there is no radial coordinate and hence the coordinate x cannot be properly normalized,
or equivalently to the scale symmetry of the model, which will be discussed in the next
section.
Thus, although one can envisage a Bekenstein bound of the form
S=
S SB = 2#ER,
(17)
which can also be deduced from the thermodynamics of a one-dimensional gas, the
coefficient # is not clearly determined. Notice that in the radiation-dominated cosmological
model of the previous section SB = 2M is a conserved quantity, proportional to the matter
density.
The Bekenstein bound is believed to hold when the gravitational energy of the system
is small with respect to its total energy, i.e., in a weak-gravitating regime. For stronggravitating systems, i.e., systems for which H R > 1, a different bound must be introduced.
In order to establish in which regime a given system is, it is useful to define a Bekenstein
Hawking entropy SBH as the Bekenstein entropy of a system with H R = 1 [1]. From the
Friedmann equation (13) one obtains
V
(1 + )
=
(1 + ),
(18)
4GR
2G
where V = 2R is the spatial volume. Recalling the definition of ,
Eq. (18) can also be
written as
SBH =
(19)
(1 + ) + 2M.
2G
This is the sum of two constant contributions: the first depends only on the geometry, while
the second, which is proportional to SB , depends only on the matter content and is not
present in d > 2. In absence of matter, the second contribution vanishes. The appearance
of a factor proportional to a 2 is again a consequence of the scale-invariance of the theory,
which does not fix the scale of the spatial coordinate in the solutions.
Notice that the bound S SBH is a truly holographic bound for a 2D spacetime. The
boundary of a spatial section of a 2D universe are two points: thus the holographic principle
SBH =
388
states that the entropy can only depend on the Newton constant G and on a dimensionless
parameter.
In a cosmological contest and when H R > 1 the Bekenstein bound must be replaced
by a holographic bound. However, it has been argued that the BekensteinHawking bound
S SBH is not the right choice. A suitable bound is given by the Hubble entropy SH ,
defined as the entropy of a universe filled with black holes of the size of a particle
horizon [15]. Later, a weaker definition of SH was proposed, in which the maximal size
of the black holes is the Hubble radius H 1 [16]. In our 2D context, SH can be calculated
as follows [17]. From Eq. (18), a black hole of radius H 1 has entropy (1 + )H
VH /4G.
Since the universe can contain NH = V /VH black holes, one obtains
SH =
RH
VH
(1 + )
=
(1 + ).
4G
2G
(20)
Notice that, while the solution I and II have a particle horizon of the size of the Hubble
radius, in case III the size of the particle horizon grows exponentially with time.
To conclude, although in the 2D case one may be able to obtain different entropy
bounds, they do not seem to be universal. The entropy bounds S SBH and S SH
depend on the arbitrary, dimensionless parameters and #. They appear to be defined
up to arbitrary scales. In the next section we will show that this fact is a consequence
of a scale symmetry of a our 2D gravity model. This scale symmetry is a peculiarity of
two-dimensional gravity and is related to the impossibility of defining an area law for the
entropy.
4. Scale symmetry and entropy bounds

It is well known that 2D AdS space has SL(2, R) as isometry group (see, for instance,
Ref. [5]). The spacetime metric is therefore invariant under the subgroup of SL(2, R)
describing dilatations, which for the ground state m = 0 in Eq. (15) is realized as
x x,
t t/.
(21)
Under this scale transformation the dilaton is not invariant, , but the scale factor
can be absorbed in a different definition of the integration constant 0 appearing in Eq. (15).
It is evident that this scale transformation is a classical symmetry of the theory because
under the action for pure gravity changes just for an overall constant factor.
This scale symmetry is also a classical invariance of our (de Sitter) action (1) in absence
of matter. In the matter-coupled case the scale transformation just changes by a constant
factor the Newton constant G.
It is not difficult to realize that the presence of the integration constants 0 , m (in the
AdS solution) and of 0 , a (in the dS solution) is a consequence of the scale symmetry. The
transformation (21) maps one solution of the fields equations characterized by 0 , m into
an other solution with different values of the integration constants, 0 , m . It is, therefore,
evident that we can use the scale symmetry (21) to write the entropy bounds in a form that
is independent of the dimensionless parameters ,
#.
389
Instead of directly working on the cosmological solution, it is more instructive

to fix these parameters by considering the two-dimensional black hole. Since the
2D cosmological solutions are the analytical continuation i of the black hole
solutions (15), one can fix ,
# using the latter as the maximum entropy configuration.
The problem reduces then to fix the dependence on m of the thermodynamical
parameters E = M, T , S of the AdS2 black hole (15). Introducing the length scale
L = 1/ and the central charge c = 120 of the thermal CFT arising in the AdS2 /CFT
correspondence one has [5,9]
T=
1 m
,
2 L
S = 2
c
m,
12
E=
c m2
.
24 L
(22)
For a generic black hole solution the behavior under the scale transformations has
been given in Ref. [9]. The AdS2 black hole metric (15) is invariant under the scale
transformations
x x,
t
t ,
M 2 M.
(23)
The dilaton transforms as

,
(24)
whereas, T , S, E scale as
T T ,
S S,
E 2 E.
(25)
The physical meaning of this scale invariance of the theory can be easily understood.
It is a general feature of all metric theories of gravity that lengths (or masses) can be
only measured with reference to an (asymptotic) reference frame. For asymptotically
Minkowskian solutions this frame is given by Minkowski space with the usual normalization (ds 2 = dt 2 + dx 2 ). Owing to the dilatation isometry of AdS2 , for solutions which
are asymptotically AdS there is no such preferred reference frame. We are free to change
the length of our rule using the scale transformations (23), without changing the physics.
We have a sort of gauge symmetry, stating that black hole solutions connected by the scale
transformations (23) are physically equivalent.1 However, the energy E and entropy S
change under the scale transformation, they are not gauge-invariant quantities. For this
reason, although we cannot find an absolute upper bound for S, every m-dependent bound
of the form S(m) SH (m) has a gauge-invariant meaning. Thus, fixing the gauge we can
remove from the entropy bounds the dependence on the dimensionless parameters , #, m.
This can be easily done by using Eqs. (25) with = 1/m into Eqs. (22) to remove the
m-dependence of T , S, E,
T=
1 1
,
2 L
S=
c
2,
12
E=
c 1
,
24 L
(26)
1 Also the geodesic motion is unaffected by a rescaling of m, which simply shifts the definition of the energy
of a test particle.
390
and choosing the value of m to fix, by means of Eq. (24), 0 (hence the central charge c)
in terms of the Newton constant G,
c
1
=
.
12 2G
With this choice, one has for the two-dimensional black hole
0 =
(27)
SB = 2ER,
(28)
fixing # = 1 in the Bekenstein bound (17). Moreover, if one identifies SBH with the entropy
of the AdS2 black hole given by Eq. (26), one finds = 1. It follows immediately
,
G
SBH =
(29)
and hence
RH
.
G
With these definitions, one has the relation
SH =
(30)
SH2 = SBH (2SB SBH ).
(31)
Once the value of has been fixed, = 1 the cosmological equation (13) takes the
Friedmann form
H2 =
1
4GE
2.
R
R
(32)
5. CardyVerlinde formula for 2D cosmology

Recalling the definition of the BekensteinHawking entropy, one can define a BekensteinHawking energy EBH , as the energy corresponding to the condition SB = SBH , at
which the gravitational system becomes strong-coupled:
1
.
(33)
2RG
Using Eqs. (30), (33) and (32) one gets a first form of the 2D CardyVerlinde formula

SH = 2R EBH (2E EBH ).
(34)
EBH =
Following Verlinde [1] it is now useful to split the total energy E in extensive EE and
non-extensive (Casimir) part EC /22
E = EE +
EC
,
2
(35)
2 The Casimir energy is of course E /2. We adopt this rather odd notation in order to conform to Ref. [1] and
C
to simplify some of the following formulas.
391
where EC is defined as
EC E + pV T S,
(36)
with p = pM + p .
The scaling behavior of EC follows from general arguments (for instance, the form of
the Casimir energy of a CFT on the cylinder), EC (S, V ) = 1 EC (S, V ). Moreover,
conformal invariance implies that ER is independent of the volume V = 2R. Combined
with the scaling behavior of EE , which by definition is EE (S, V ) = EE (S, V ), this
gives
b 2
d
S ,
,
EC =
4R
2R
where b, d are arbitrary constants. Using this equation and (35) one gets
EE =
(37)
2R
S=
EC (2E EC ).
bd
(38)
Eq. (38) becomes the Cardy formula if we take bd = 1 and consider a 2D CFT on the
cylinder. In fact, in this case the total and Casimir energy are given by
l0
EC
c
(39)
,
=
,
R
2
24R
where c and l0 are the central charge and the eigenvalue of the Virasoro operator L0 of the
CFT and R is the radius of the cylinder. Inserting the previous equations into Eq. (38) one
finds the Cardy formula

c
c
l0
.
S = 2
(40)
6
24
E=
Comparing Eq. (34) with Eq. (38) (bd = 1) we find that they agree if we take S = SH
and EBH = EC . We are therefore led to a cosmological bound for the Casimir energy
EC EBH ,
(41)
analogous to the one proposed for higher-dimensional cosmology [1]. Using the Friedmann
equation (32), one easily finds that for H R 1, E EBH . Hence, for strongly gravitating
systems we have
EC EBH E.
(42)
This bound shares some nice features with its higher-dimensional counterpart.
(a) It is always valid and does not break down for H R 1.
(b) Its physical meaning is that the cosmological energy by itself is not sufficient to
produce a black hole of the size of the entire universe. In fact, the AdS2 black hole
saturates the bound.
(c) For H R > 1 it is equivalent to the Hubble bound S < SH .
392
(d) When the bound is saturated EC = EBH and the cosmological equation (32) becomes
the Cardy formula (40). The translation table between 2D cosmology and 2D CFT is
given by
1
c
,
12
2G
l0 ER,
RH
.
G
(43)
6. Limiting temperature
The cosmological equation (3) can be used, in conjunction with Eq. (32), to give a lower
bound for temperature in a radiation-dominated universe. From (3) and (32) follows
1
4GE
1 4GM
H = 2
+ 2=
.
R
R
R2
Defining the Hubble temperature
1 H
,
2 H
and using the definition of SH and EBH , Eq. (44) becomes
(44)
TH =
(45)
EBH = TH SH + E + pV .
(46)
Comparing (46) with (36) and using the bounds S SH , EC EBH one obtains a lower
bound for T ,
T TH .
(47)
7. Conclusions
We have shown that the analysis of Verlinde on cosmological entropy bounds and their
relations with CFT can be extended to a two-dimensional model, in spite of the difficulties
related to the definition of a holographic principle in two dimensions. The identification of
the Friedmann equation with the Cardy formula requires the use of the scaling invariance
of the theory in order to fix some dimensionless parameters.
The most striking feature of the mapping between 2D cosmology and 2D CFT is the
identification of the Newton constant in terms of the central charge of the CFT. The
correspondence between cosmological equations and the Cardy formula requires c = 6/G.
This relation has an obvious holographic nature. In higher dimensions the holographic
principle requires c V /GR. Extended to the 2D case where V = 2R this relation
reproduces our result. Further support to the holographic origin of this relation, comes from
the fact that it can also be deduced using the AdS/CFT correspondence in two dimensions
[5,6,18]. In this context, it has been found that the central charge of the CFT living on
the boundary of AdS2 is given by c = 120 , which is our present result, because 0 , is
proportional to the inverse of the 2D Newton constant. Owing to the dilatation symmetry
of our model, under which 0 scales as in Eq. (24), the coefficient of proportionality between 0 and G1 depends on the dilatation-gauge we choose.
393
References
[1] E. Verlinde, On the holographic principle in a radiation dominated universe, hep-th/0008140.
[2] G. t Hooft, Dimensional reduction in quantum gravity, gr-qc/9310026;
L. Susskind, J. Math. Phys. 36 (1989) 6377.
[3] R. Brustein, S. Foffa, G. Veneziano, Phys. Lett. B 507 (2001) 270, hep-th/0101083;
I. Savonije, E. Verlinde, Phys. Lett. B 507 (2001) 305, hep-th/0102042;
S. Nojiri, S.D. Odintsov, Int. J. Mod. Phys. A 16 (2002) 3273, hep-th/0011115;
I. Brevik, S.D. Odintsov, On the CardyVerlinde entropy formula in viscous cosmology, gr-qc/0110105;
A.J.M. Medved, AdS holography and strings on the horizon, hep-th/0201215;
A.J.M. Medved, Asymptotically flat holography and strings on the horizon, hep-th/0202083;
D. Youm, A note on the CardyVerlinde formula, hep-th/0201268.
[4] R. Cai, Phys. Rev. D 63 (2001) 124018, hep-th/0102113;
R. Cai, Phys. Lett. B 525 (2002) 331;
D. Birmingham, S. Mokhtari, Phys. Lett. B 508 (2001) 365, hep-th/0103108;
D. Klemm, A.C. Petkou, G. Siopsis, Nucl. Phys. B 601 (2001) 380, hep-th/0101076;
D. Klemm, A.C. Petkou, G. Siopsis, D. Zanon, Nucl. Phys. B 620 (2002) 519, hep-th/0104141;
J. Jing, CardyVerlinde formula and asymptotically flat rotating charged black holes, hep-th/0202052.
[5] M. Cadoni, S. Mignemi, Phys. Rev. D 59 (1999) 081501, hep-th/9810251;
M. Cadoni, S. Mignemi, Nucl. Phys. B 557 (1999) 165, hep-th/9902040.
[6] M. Cadoni, M. Cavagli, Phys. Rev. D 63 (2001) 084024, hep-th/0008084;
M. Cadoni, M. Cavagli, Phys. Lett. B 499 (2001) 315, hep-th/0005179.
[7] J.A. Cardy, Nucl. Phys. B 270 (1986) 186.
[8] A. Strominger, J. High Energy Phys. 01 (1999) 007, hep-th/9809027;
S. Cacciatori, D. Klemm, D. Zanon, Class. Quantum Grav. 17 (2000) 1731, hep-th/9910065.
[9] M. Cadoni, P. Carta, Phys. Lett. B 522 (2001) 126, hep-th/0107234.
[10] R. Jackiw, in: S.M. Christensen (Ed.), Quantum Theory of Gravity, Adam Hilger, Bristol, 1984;
C. Teitelboim, in: S.M. Christensen (Ed.), Quantum Theory of Gravity, Adam Hilger, Bristol, 1984.
[11] M. Cadoni, S. Mignemi, Cosmology of the JackiwTeitelboim model, gr-qc/0202066.
[12] See, for example, F. de Felice, C.J.S. Clark, Relativity on Curved Manifolds, Cambridge Univ. Press,
Cambridge, 1990.
[13] J.D. Bekenstein, Phys. Rev. D 23 (1981) 287.
[14] M. Cadoni, S. Mignemi, Phys. Rev. D 51 (1995) 4319, hep-th/9410041.
[15] W. Fischler, L. Susskind, Holography and cosmology, hep-th/9806039.
[16] R. Easther, D. Lowe, Phys. Rev. Lett. 82 (1999) 4967, hep-th/9902088.
[17] G. Veneziano, Phys. Lett. B 454 (1999) 22, hep-th/9902126.
[18] M. Cadoni, P. Carta, D. Klemm, S. Mignemi, Phys. Rev. D 63 (2001) 125021, hep-th/0009185.

Erratum
Erratum to: The KM phase in semi-realistic

heterotic orbifold models
[Nucl. Phys. B 595 (2001) 332]
Joel Giedt
1 Cyclotron Road, 50A-5101, LBNL, Berkeley, CA 94720, USA
Received 15 January 2002; accepted 27 March 2002
Fig. 1 on p. 25 should not be replica of Fig. 3 on p. 27. Rather, the attached figure should
have appeared as Fig. 1; the caption is correct and is reproduced here.
i . The T -moduli are stabilized at

Fig. 1. Vub (MZ ) for complex mixing matrices X1,2 and Xiggs vevs Y1,2,3
TcI = 1. For comparison, the experimentally preferred region [43] is outlined.
PII of original article: S0550-3213(00)00620-9.

E-mail address: jtgiedt@lbl.gov (J. Giedt).

PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 2 4 4 - 4
398
J. Giedt / Nuclear Physics B 632 (2002) 397398
The matrix Z = diag(ei/3 , ei/3 , 1) which occurs in Section 2 is not in the center of
U (3). Thus the entire mu1 = md1 counterexample on p. 10 fails. All but the first sentence of
the main paragraph on p. 10 should be ignored.

CUMULATIVE AUTHOR INDEX B631B632
Akemann, G.
Antoniadis, I.
Antoniadis, I.
Apreda, R.
Argeri, M.
Armoni, A.
B631 (2002) 471

B631 (2002) 3
B631 (2002) 66
B631 (2002) 342
B631 (2002) 388
B632 (2002) 240
Barto, E.
Benakli, K.
Bertolini, M.
Bianchi, M.
Boyanovsky, D.
Brignole, A.
Bueno, A.
Buras, A.J.
Burkardt, M.
B632 (2002) 330

B631 (2002) 3
B632 (2002) 257
B631 (2002) 159
B632 (2002) 121
B631 (2002) 195
B631 (2002) 239
B631 (2002) 219
B632 (2002) 311
Cadoni, M.
Campanelli, M.
Cao, F.J.
Carta, P.
Chu, C.-S.
Ciocarlie, C.
Clark, T.E.
Czarnecki, A.
B632 (2002) 383

B631 (2002) 239
B632 (2002) 121
B632 (2002) 383
B632 (2002) 219
B632 (2002) 303
B632 (2002) 3
B631 (2002) 219
Degrassi, G.
Delduc, F.
Deser, S.
De Vega, H.J.
Di Vecchia, P.
Dolgov, A.D.
Dotsenko, V.S.
Dubnicka, S.
Dubnickov, A.-Z.
B631 (2002) 195

B631 (2002) 403
B631 (2002) 369
B632 (2002) 121
B631 (2002) 95
B632 (2002) 363
B631 (2002) 426
B632 (2002) 330
B632 (2002) 330
Enger, H.
B631 (2002) 95
Freedman, D.Z.
Freitas, A.
B631 (2002) 159

B632 (2002) 189
0550-3213/2002 Published by Elsevier Science B.V.

PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 3 6 9 - 3
Froggatt, C.D.
Frolov, S.
B631 (2002) 285

B632 (2002) 69
Giedt, J.
B632 (2002) 397
Hansen, S.H.
Harmark, T.
Hartnoll, S.A.
Hatsuda, M.
Hebecker, A.
Hollik, W.
B632 (2002) 363

B632 (2002) 257
B631 (2002) 325
B632 (2002) 114
B632 (2002) 101
B632 (2002) 189
Imeroni, E.
B631 (2002) 95
Jacobsen, J.L.
B631 (2002) 426
Kamimura, K.
Khlebnikov, S.
Kleinert, H.
Koike, Y.
Kuraev, E.A.
B632 (2002) 114

B631 (2002) 307
B632 (2002) 51
B632 (2002) 311
B632 (2002) 330
Laugier, A.
Lee, P.
Lee, P.
Litim, D.F.
Lopez, E.
Love, S.T.
Lozano-Tellechea, E.
Lukierski, J.
B631 (2002) 3
B632 (2002) 283
B632 (2002) 303
B631 (2002) 128
B632 (2002) 240
B632 (2002) 3
B631 (2002) 95
B632 (2002) 219
Maggiore, M.
Malbouisson, A.P.C.
Malbouisson, J.M.C.
Mangano, M.L.
Mastrolia, P.
Mignemi, S.
Minasian, R.
Misiak, M.
B631 (2002) 342

B631 (2002) 83
B631 (2002) 83
B632 (2002) 343
B631 (2002) 388
B632 (2002) 383
B631 (2002) 43
B631 (2002) 219
400
Moretti, M.
Moss, I.G.
Moss, I.G.
B632 (2002) 343

B631 (2002) 500
B632 (2002) 173
Navas-Concha, S.
Naylor, W.
Nepomechie, R.I.
Nguyen, X.S.
Nicolis, A.
Nielsen, H.B.
Nowling, S.R.
B631 (2002) 239

B632 (2002) 173
B631 (2002) 519
B631 (2002) 426
B631 (2002) 342
B631 (2002) 285
B632 (2002) 3
Obers, N.A.
Ooguri, H.
B632 (2002) 257

B632 (2002) 283
Park, J.
Park, J.
Pastor, S.
Pearce, P.A.
Petcov, S.T.
Pittau, R.
B632 (2002) 283

B632 (2002) 303
B632 (2002) 363
B631 (2002) 447
B632 (2002) 363
B632 (2002) 343
Raffelt, G.G.
Remiddi, E.
Richard, C.
Riotto, A.
Rubbia, A.
B632 (2002) 363

B631 (2002) 388
B631 (2002) 447
B631 (2002) 342
B631 (2002) 239
Sakaguchi, M.
Santachiara, R.
Santana, A.E.
Schlittgen, B.
Semikoz, D.V.
Skenderis, K.
Slavich, P.
Sorin, A.S.
Sturani, R.
B632 (2002) 114

B631 (2002) 426
B631 (2002) 83
B632 (2002) 155
B632 (2002) 363
B631 (2002) 159
B631 (2002) 195
B631 (2002) 403
B631 (2002) 66
Takanishi, Y.
Tomasiello, A.
Tseytlin, A.A.
B631 (2002) 285

B631 (2002) 43
B632 (2002) 69
Urban, J.
B631 (2002) 219
Van den Bossche, B.

Vernizzi, G.
B632 (2002) 51
B631 (2002) 471
Waldron, A.
Walter, W.
Weiglein, G.
Westerberg, A.
Wettig, T.
B631 (2002) 369

B632 (2002) 189
B632 (2002) 189
B632 (2002) 257
B632 (2002) 155
Zakrzewski, W.J.
Zemlyanaya, E.
Zwirner, F.
B632 (2002) 219

B632 (2002) 330
B631 (2002) 195

Nucl - Phys.B v.632

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.632

Uploaded by

Copyright:

Available Formats

Nuclear Physics B 632 (2002) 350

The supercharge and superconformal symmetry for

Received 5 September 2001; accepted 27 March 2002

T.E. Clark et al. / Nuclear Physics B 632 (2002) 350

systems whose dynamics is governed by the potential energy functions

The zero energy ground states satisfy

T.E. Clark et al. / Nuclear Physics B 632 (2002) 350

are represented as superspace differential operators. The superspace action is shown to

2. N = 1 supersymmetry and (anti)chiral supercoordinates

T.E. Clark et al. / Nuclear Physics B 632 (2002) 350

 = 2i . PerformThey anticommute to give the (SUSY covariant) time derivative {D, D}

T.E. Clark et al. / Nuclear Physics B 632 (2002) 350

the Lagrangian transforms as a total time derivative

where S = 12 3 is the spin operator. The corresponding Hamiltonian is then simply

T.E. Clark et al. / Nuclear Physics B 632 (2002) 350

 nilpotent charges are seen to obey the supersymmetry algebra

The SUSY variations anticommute to yield a time derivative: { Q , Q } = 2it . Since

T.E. Clark et al. / Nuclear Physics B 632 (2002) 350

which participate in the rotation. These transformations are generated by superspace

T.E. Clark et al. / Nuclear Physics B 632 (2002) 350

only, then the

3. Noethers theorem and superconformal transformations

T.E. Clark et al. / Nuclear Physics B 632 (2002) 350

T.E. Clark et al. / Nuclear Physics B 632 (2002) 350

while the chiral and antichiral supercoordinates vary as

where and are infinitesimal complex independent functions of superspace.

T.E. Clark et al. / Nuclear Physics B 632 (2002) 350

while the antichiral constraint D = 0 mandates

The solutions to these equations require the transformation parameters , iX

T.E. Clark et al. / Nuclear Physics B 632 (2002) 350

Solving equations (3.20) for the transformation parameters gives

T.E. Clark et al. / Nuclear Physics B 632 (2002) 350

Grouping terms with common variation parameters, the infinitesimal superconformal

T.E. Clark et al. / Nuclear Physics B 632 (2002) 350

and D = , where and are constant Grassmann odd superconformal

the improved Noether charge obeys the identities

T.E. Clark et al. / Nuclear Physics B 632 (2002) 350

and the conservation relation

S, S, R and J represent an entire function of superspace whose , independent

T.E. Clark et al. / Nuclear Physics B 632 (2002) 350

3.2. Supersymmetry: parameters and

3.3. UJ (1) transformations: parameter

3.4. R transformations: parameter n +

T.E. Clark et al. / Nuclear Physics B 632 (2002) 350

For n = 0, the conserved supercoordinate charge R0 is obtained as

 into Q defines the improved charge QI as

leading to the time derivative

T.E. Clark et al. / Nuclear Physics B 632 (2002) 350

Noethers theorem yields the counting charges

T.E. Clark et al. / Nuclear Physics B 632 (2002) 350

with the derivative identities

the total number density can be written as

The conservation equations take the form

T.E. Clark et al. / Nuclear Physics B 632 (2002) 350

3.6. Superconformal symmetry: parameters and

As alluded to in the discussion following equations (3.27) and (3.28), as a consequence of

where the breaking terms are now just

3.7. Conformal symmetry: parameter

T.E. Clark et al. / Nuclear Physics B 632 (2002) 350

Noethers theorem has the general form

T.E. Clark et al. / Nuclear Physics B 632 (2002) 350

T.E. Clark et al. / Nuclear Physics B 632 (2002) 350

can be expanded in components as

= 2i . PerformThey anticommute to give the (SUSY covariant) time derivative {D, D}

nilpotent charges are seen to obey the supersymmetry algebra

into Q defines the improved charge QI as

and recalling that differentiation with respect to the Grassmann

= 2H , and the identity for nested commutators,

obey the same superconformal algebra as the Q.