Nucl - Phys.B v.607

Nuclear Physics B 607 (2001) 337
www.elsevier.com/locate/npe
ADE singularities and coset models

Tohru Eguchi a , Nicholas P. Warner b , Sung-Kil Yang c
a Department of Physics, University of Tokyo, Tokyo 113-0033, Japan
b Department of Physics and Astronomy and CIT-USC Center for Theoretical Physics,
University of Southern California, Los Angeles, CA 90089-0484, USA

c Institute of Physics, University of Tsukuba, Ibaraki 305-8571, Japan
Received 22 May 2001; accepted 23 May 2001
Abstract
We consider the compactification of the IIA string to (1 + 1) dimensions on non-compact 4-folds
that are ALE fibrations. Supersymmetry requires that the compactification include 4-form fluxes, and
a particular class of these models has been argued by Gukov, Vafa and Witten to give rise to a set of
perturbed superconformal coset models that also have a LandauGinzburg description. We examine
all these ADE models in detail, including the exceptional cosets. We identify which perturbations are
induced by the deformation of the singularity, and compute the LandauGinzburg potentials exactly.
We also show how the the LandauGinzburg fields and their superpotentials arise from the geometric
data of the singularity, and we find that this is most naturally described in terms of non-compact,
holomorphic 4-cycles. 2001 Published by Elsevier Science B.V.
1. Introduction
The link between the classification of singularity types and quantum effective actions of
field theories now has a fairly long and interesting history. The key to making this work is
for the field theory to have enough supersymmetry so that a non-renormalization theorem
protects the sector of the field theory that is determined by the classical singularity type.
For theories with a mass gap there are generically BPS states whose spectrum can be
computed semi-classically, and which can be used to identify the field theory. This was first
used with considerable success in N = 2 superconformal field theories, in which there is a
LandauGinzburg superpotential [1]. Most particularly the ADE classification of complex
singularities with no moduli was shown to correspond to precisely the modular invariant
N = 2 superconformal field theories with central charge c < 3. If one topologically twists
these N = 2 theories the result is topological matter [2], and the coupling to topological
gravity could be naturally incorporated into the singularity theory [3,4].
E-mail address: warner@usc.edu (N.P. Warner).
0550-3213/01/$ see front matter 2001 Published by Elsevier Science B.V.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 2 6 3 - 2
T. Eguchi et al. / Nuclear Physics B 607 (2001) 337
The advent of string dualities gave rise to many new constructions of string and field
theories, and in particular, low energy effective actions. An early offshoot of this general
program was to realize that the complete SeibergWitten quantum effective action of
N = 2 super-YangMills theories in four dimensions [5,6] could be obtained from the
period integrals of K3-fibrations that were developing an ADE singularity [7]. In this
construction, the singularity type corresponds to the gauge group of the YangMills theory.
This subject has now, in a sense, come full circle. In [8] it was shown how Landau
Ginzburg models in (1 + 1) dimensions could be constructed using compactifications of
type IIA strings on CalabiYau 4-folds. If the 4-fold in this construction is a K3-fibration,
and if the fiber develops an ADE singularity, then the (1 + 1)-dimensional field theory has
been argued to be a massive perturbation of a coset model based upon the corresponding
ADE group. The mass scale is set by the deformation of the singularity, and a conformal
field theory emerges when the fiber is singular.
Extracting field theories from singular limits of compactifications is often a subtle
process, and there are sometimes multiple limits being taken at the same time. For example,
to extract the SeibergWitten effective actions one needs to decouple gravity and yet
maintain a cut-off whose dimensions are inherited from the string tension. This is done
by scaling the size of the base of the fibration while scaling the K3 so as to isolate the
singular fiber. This results in a non-trivial role for the fibration even in the field theory
limit. A conformal field theory can occur in this context if the moduli of the singularity
are fine-tuned to an ArgyresDouglas point, but the generic theory has a scale. Things are
considerably simpler for the LandauGinzburg models that arise from singular K3 fibers in
4-folds: there is no cut-off, and the field theory is conformal when the K3 fiber is singular.
All the essential physics comes from the isolated ADE singularity in the fiber, and the base
of the fibration only plays a trivial role.
While the singularity type determines the numerator of the coset conformal field theory,
the denominator (and probably the level) of the model are determined by a background
4-form flux. Such fluxes are parameterized by weights of the Lie algebra, G, associated
with the ADE singularity. The Weyl group of the singularity acts on the flux, permuting
it around a Weyl orbit. The denominator of the coset model is defined by the subgroup
of the Weyl group that leaves a flux invariant, and the non-trivial Weyl images of each
flux represent different ground states of the same model. Deformations of the singularity
introduce masses, but in general the deformed theory does not have a mass gap: there are
still massless excitations. However, for a certain class of minimal fluxes, 1 the theory does
have a mass gap, and the associated coset model was argued in [8] to be (a perturbation of)
the N = 2, KazamaSuzuki coset model based upon the level one, A, D or E hermitian
symmetric space. This was checked in detail in [8] for the A-type models, and here we
will verify that the construction works for the D-type and the E-type singularities. It is not
immediately clear what the corresponding models are for larger (non-minimal) fluxes, but
it is tempting to try to identify the magnitude of the flux with the level of the coset model.
1 A minimal flux is one that corresponds to a miniscule weight of G.
The identification of these models, even at level one, is a non-trivial problem. If one
uses a non-compact CalabiYau 4-fold with an ADE singularity as outlined above, it
turns out that all the dynamics are frozen because the kinetic terms of the the model are
not normalizable. On the other hand, in [8] a LandauGinzburg superpotential, W , was
conjectured for compact CalabiYau 4-folds, and for non-compact CalabiYau manifolds
this yields expressions for the topological charges of solitons in terms of period integrals
on the 4-fold. The problem is that for non-compact CalabiYau manifolds, there does
not appear to be an obvious geometric characterization of the LandauGinzburg fields
themselves: one does not, a priori, know how many fields there are, let alone their U (1)
charges. Thus the singularity only yields topological data about the model, that is, one
knows only the ground states and the topological charges of the solitons. On the other
hand, knowing the coset conformal field theory determines the chiral primary fields and the
LandauGinzburg potential (if there is one). Such a theory also generically has many more
deformations of the superpotential than there are deformations of the singularity of the
4-fold. Thus there is a special, canonically deformed LandauGinzburg superpotential
whose deformations are precisely generated from the deformations of the 4-fold.
One of our purposes here is to characterize and compute all the canonically deformed
superpotentials associated with the ADE singularities with minimal fluxes. We first do
this rather abstractly by developing an algorithm for computing such superpotentials. In
so doing, we find that there are obvious choices in the procedure, and that these choices
correspond to determining the coset denominator, and hence the correct set of Landau
Ginzburg variables. We then look at the period integrals on the 4-fold, and find how
to determine the superpotentials from such calculations. A direct computation appears
to lead to only one superpotential for each A, D, or E singularity. However, we also
show that there are, once again, ambiguities in the calculation of the periods, and that
different choices lead to the full set of canonically deformed superpotentials. We will show
that the resolution of the ambiguity amounts to selecting holomorphic representatives of
the non-compact homology of the singularity. These representatives are characterized by
weights of G, and each choice is equivalent to selecting a Weyl orbit of fluxes. In doing
the calculations of the period integrals, we also find the vestigial remnants of the Landau
Ginzburg variables, and this leads to some natural conjectures as to the role of the Landau
Ginzburg variables in the geometric picture on 4-folds.
The next two sections of this paper contain a detailed review of the N = 2
superconformal coset models of Kazama and Suzuki, and the construction of associated
LandauGinzburg superpotentials. In particular, in Section 2 we will review that standard
construction of the chiral rings and LandauGinzburg potentials of the simply laced, level
one, Hermitian symmetric space (SLOHSS) models. We then generalize this to obtain
the canonically deformed superpotentials. In Section 3 we will construct some specific
examples of these deformed superpotentials, and evolve an algorithm for generating all
the superpotentials associated with the numerator Lie algebra, G, from any one such
superpotential. In Section 4 we first review the part of [8] that is relevant to the SLOHSS
models, and then use the formula of [8] to calculate topological charges of solitons in terms
of period integrals of the singularity. We then use this calculation, in combination with
the algorithm developed in Section 3, to reconstruct the LandauGinzburg superpotentials

from the geometry of the singularity. Our methods of computing the period integrals can
be more naturally cast in terms excising holomorphic surfaces, and computing intersection
numbers. We also discuss this in Section 4, and show how the selection of the holomorphic
surfaces is equivalent to choosing a flux at infinity. We also find that the Landau
Ginzburg variables appear very naturally in the parameterization of these surfaces. Finally,
in Section 5 we make some remarks about fermion numbers of solitons, and about
generalizations of the results presented here.
2. The LandauGinzburg potentials of deformed SLOHSS models

2.1. The conformal LandauGinzburg potential
The N = 2, KazamaSuzuki conformal coset models [9] have been extensively studied.
They are based upon the coset construction using:
G SO(dim(G/H ))
(2.1)
,
H
where the SO(dim(G/H )) is a level one current algebra, and represents the bosonized
fermions in a supersymmetrization of the coset model based upon G/H . In (2.1), H is
embedded diagonally into G and SO(dim(G/H )), the embedding into G is an index one
embedding, and the embedding into SO(dim(G/H )) is a conformal embedding [10] of
level gh where g and h are the dual Coxeter numbers of G and H , respectively. If G has
a level k, then H has a level k + g h. If G/H is Khler, then H = H0 U (1), with
U (1) inducing the complex structure, and the corresponding coset model has an N = 2
superconformal algebra.
The simply laced, level one, hermitian symmetric space (SLOHSS) models have been
studied even more extensively. In these models, G is represented by a level one current
algebra, and G/H is a hermitian symmetric space. These cosets are:
S=
SO(2n)
SO(2n)
SU(n + m)
,
,
,
SU(n) SU(m) U (1)
SO(2n 2) U (1)
SU(n) U (1)
E6
E7
(2.2)
,
.
SO(10) U (1)
E6 U (1)
It was shown in [11] that the chiral ring of these coset models is isomorphic to the
de Rham cohomology ring H (S, R). It has also been argued that these models have a
LandauGinzburg formulation [11]. Most of the LandauGinzburg potentials have been
determined.
The most direct way to compute the chiral ring and potential is to use the isomorphism
with the cohomology ring of S. The latter can be generated by the Chern classes of
H -bundles on S, and these can be generated from the irreducible H -representations.
The vanishing relations of the ring are characterized by the trivial H -bundles, and the
corresponding vanishing Chern classes are given by those combinations of H -representations that are actually G-representations, and hence generate trivial bundles. In more
mechanistic terms, the chiral ring is generated by the Casimir invariants of H , and to find
the vanishing relations one simply has to take all the Casimirs of G and decompose them
into the Casimirs of H .
In practice it is simplest to reduce this calculation to the Cartan subalgebra, X , of G.
(Since G and H have the same rank, this is also a Cartan subalgebra (CSA) of H .)
We then parameterize X by variables, j , j = 1, . . . , r, where r is the rank of G. The
Casimirs of G and H are then equivalent to W (G) and W (H ) invariant polynomials,
denoted Vj and xi , respectively, of the j , where W (G) and W (H ) are the Weyl groups
of G and H , respectively. The inequivalent W (H ) invariant polynomials, modulo W (G)
invariant polynomials, are given by elements of W (G)/W (H ), and thus the chiral ring has
the structure of this Weyl coset.
One should also observe that elements of the coset W (G)/W (H ), and hence the ground
states, are in one-to-one correspondence with the weights of a miniscule representation
of G. A miniscule representation, R, is defined as one in which all the weights lie in
a single Weyl orbit. The subgroup H is then defined to be the one whose semi-simple
part (not the U (1)) fixes the highest weight of R. Since all other weights lie in the
Weyl orbit of the highest weight, it follows that the representation is indeed in one-toone correspondence with W (G)/W (H ). In this manner one can generate the list (2.2). We
thus see that ground states of the SLOHSS models are naturally labeled by the weights of
a miniscule representation of G. This will be important throughout this paper.
The reduction of Casimirs procedure was used in [11,12] to construct the chiral rings.
The remarkable, and rather unexpected fact is that the vanishing relations appear to be
integrable to produce a single superpotential for each SLOHSS model. 2 In particular, the
LandauGinzburg potentials for E6 and E7 were computed in [12].
The basic procedure is, therefore, as follows: let mi and m
i , i = 1, . . . , r, be the
exponents of G and H , respectively. (We define the exponent of a U (1) to be zero.) The
degrees of the Casimirs are thus mi + 1 and m
i + 1. (We define the Casimir of U (1) by
taking a trace: it thus has degree 1.) The generators of the chiral ring are thus variables xi ,
i = 1, . . . , r, of degrees m
i + 1. The vanishing relations are polynomials Vi (xj ),
i = 1, . . . , r, of degrees mi +1, obtained by the decomposition of Casimir invariants. To get
the LandauGinzburg potential one wants to integrate the vanishing relations to obtain a
potential W (xi ) such that the set of equations Vi = 0 are equivalent to the set of equations
W/xi = 0. The only complexity in this task is that W/xj will generically be some
constant multiple of Vj plus Vk s of lower degree multiplied by polynomials in the xi s.
One has thus to find the proper combinations of Vk s before integration is possible, and as
we indicated earlier, it seems remarkable that such integration is possible.
2.2. The canonically deformed LandauGinzburg potential
The conformal LandauGinzburg potential is, of course,
and there W (G)quasi-homogeneous,

, is the degeneracy of the Rafore multi-critical. The index of the singularity, = W
(H )
SO(2n)
2 This has not yet been proven for the general coset
SU(n)U (1) , but in fact it will be implicitly established in
this paper.
mond ground states. We now wish to deform this potential in such a manner as to make a
mass gap, and yield the corresponding theories associated with the ADE singularities in [8].
The key to seeing how this must happen is to observe that the versal deformation of the
singularity preserves the W (G) symmetry of the singularity, and therefore so must the corresponding deformation of the coset model. In particular, the ground states of the deformed
model must be a W (G)-invariant family. This leads to a unique versal deformation procedure for the coset model. Instead of setting the G-Casimirs, Vi (xj ), to zero one can set
them to constant values: Vi = vi for some vi . This is manifestly W (G) invariant, and represents a set of equations that define the ground states of the canonically deformed model.
To understand the interrelationship between all the LandauGinzburg potentials for
cosets G/H with the same numerator G, but with different denominators H , it is important
to understand how these ground states, and hence the canonical deformation, is realized in
terms of the CSA variables, j . In terms of the Cartan subalgebra, X , setting Vi = vi
uniquely defines a general point, j = j(0) , in X up to the action of W (G). That is, the vj
define a general point, j(0) , in the fundamental Weyl chamber of G. Similarly, the values
of the Casimirs, xj , of H uniquely specify a point in the fundamental Weyl chamber of H .
The vacua of the deformed coset model are thus characterized by all the W (G) images of
(0)
j that lie in the fundamental Weyl chamber of H . Non-zero values of the vj generically
(0)
yield a massive theory with all vacua separated and j an interior point of the Weyl
chamber. The theory becomes multi-critical, with massless solitons when the vj assume
(0)
values at which j goes to a wall of the Weyl chamber of G.
Remarkably enough, we find that the deformed vanishing relations: Vi = vi are still
integrable to a LandauGinzburg potential W (xj ; vi ). This will be proven in the next
section. Based upon the foregoing observation we develop an algorithm for computing
the desired LandauGinzburg potential for any SLOHSS model, G/H , given one such
superpotential for any SLOHSS model with the same numerator, G. We then compute a
superpotential for each choice of G, including the exceptional cosets, and we also give
several examples of the application of the algorithm.
Finally, we note that the deformed potential is quasi-homogeneous:

W m j +1 xj ; mi +1 vi = N+1 W (xj ; vi ),
(2.3)
where N = mr is the dual Coxeter number (i.e., the degree of the highest Casimir) of G.
Once again, we stress that because we have preserved the Weyl symmetry of G, it is
these deformed potentials that must be related to those of the versal deformations of the
ADE singularities.
3. Chiral rings and superpotentials

3.1. Grassmannians
Here one has G = SU(m + n) and H = SU(m) SU(n) U (1), but it is simpler to
think of this coset as having G = U (m + n) and H = U (m) U (n). We will take m n.
The CSA of U (m + n) can be parameterized by (1 , . . . , m+n ), and the Casimirs of G are

the permutation invariants:
Vk
m+n
k = 1, . . . , m + n,
k,
=1
while those of H are:

xk
m
k,
k = 1, . . . , m,
=1
xk
m+n
k,
k = 1, . . . , n.
(3.1)
=m+1
It is convenient to introduce an equivalent set of Casimirs:

k (1)k
V
j1 j1 jk , k = 1, . . . , m + n,
1j1 <j2 <<jk m+n
along with a corresponding family of new variables:

zk (1)k
j1 j1 jk , k = 1, . . . , m.
(3.2)
1j1 <j2 <<jk m
Since xk + xk = Vk = vk , we can use this linear equation to eliminate all the xk in terms
of xk and vk . Note that for k > m one must write xk as a polynomial in the xj (j m),
and that for k > n one must write x k in terms of a polynomial in the xj (j n). Thus
the deformed LandauGinzburg potential is a function, Wm,n , that depends upon xj , j =
1, . . . , m, and vk , k = 1, . . . , m + n.
For m = 1 one can set x = x1 = 1 , and one has vn+1 = x n+1 + xn+1 . One then writes
xn+1 as a polynomial in xj , j n, and then eliminates the xj using xk = vk x k . The
result is a superpotential of the form:

1
1
x n+2 +
vk x n+2k ,
n+2
n+2k
n+2
W W1,n =
(3.3)
k=2
k given Vk = vk . The whole point is that one has

where vk are the values of V

dW
=
x (0) .
dx
n+1
(3.4)
=1
The n + 1 critical points of this superpotential are then given by the components of the
(0)
weight, , corresponding to the values, vk , of the Casimirs, Vk .
Our purpose now is to show how to generate the superpotentials of all the Grassmannians
are
from (3.3). For general, m, the chiral ring is generated by the, xk , of (3.1)

which
ground
permutation invariants of the first m generators 1 , . . . , m . Then there are m+n
m
(0)
states given by all choices of m of the . In particular, the lowest dimension chiral
(0)
(0)
primary, x1 1 + + m takes values j1 + + jm for all choices of j1 , . . . , jm .
One can thus extract the multi-variable potential from (3.3) by a rather simple algorithm.
dW
The ground states of Wm,n are characterized by subsets of m of the solutions to dx1,n = 0.
The idea is to introduce an auxiliary equation that defines such a subset of m roots, and
10
then use cross elimination between (3.3) or (3.4) and this auxiliary equation to reconstruct
the superpotential Wm,n . While the general superpotentials for the Grassmannians might
be constructed more directly, the beauty of the foregoing algorithm is that it generalizes to
other Lie algebras.
For the general Grassmannian model we need the m variables, xk , or equivalently and
more conveniently, the zk of (3.2). These characterize a subset, 1 , . . . , m , of the n + 1
dW
roots of dx1,n = 0. Moreover, from (3.2) it follows that the individual roots, 1 , . . . , m , are
related to the zk as the m roots of the auxiliary equation:
xm +
m
zk x mk =
k=1
m

(x j ) = 0.
(3.5)
j =1
Thus, given a critical point of Wm,n , one can use this formula to extract the set of m critical
points, {j(0)
, . . . , j(0)
} of (3.3), that characterize a single ground state of the Grassmannian
m
1
model.
One now reverses this procedure: the multi-variable critical points are precisely
characterized by adjusting the zk (or xk ) so that all m roots of (3.5) are critical points
of (3.3). The job is thus to properly recast (3.3) in terms of the zk using the fact that we
dW
now wish to sum (3.3) over subsets of m roots of dx1,n = 0.
dW
One can do this by using (3.5) to eliminate all powers, x k , k m, in dx1,n . The result is
a polynomial P (zj ; x) of overall degree n + 1, but of degree m 1 in x. If one makes the

expansion: P (zj ; x) = m1
=0 B (zj )x , then this will vanish for a (generic) set of m roots
dW
of dx1,n = 0 if and only if all of the B (zj ) vanish. These must therefore be the deformed
vanishing relations that characterize the chiral ring of the multi-variable model. It is these
that must be integrated to give the multi-variable superpotential.
There is, however, a simpler way to get the LandauGinzburg superpotential: perform
the same elimination procedure, using (3.5), on the superpotential (3.3). The result is once
again a function, W (zj ; x), that is a polynomial of degree at most m 1 in x. Now
replace x j in this function by m1 xj , and rewrite everything in terms of either zj or xj .
We claim that this results in the requisite multi-variable potential, Wm,n+1m (xj ; vj ). To
see why this is so, observe that this prescription for replacing x j is the same as summing
1
m W (zj ; x) over the m roots of (3.5). Thus Wm,n+1m (xj ; vj ) represents an average of the
values of (3.3) over a set of roots of
dW1,n
dx
= 0, and imposing
Wm,n+1m (xj ;vj )

xj
= 0 implies
dW1,n
dx
= 0 on all m of these roots. Also recall that the values of the superpotential
that
encodes the topological charge of solitons, and this averaging procedure reproduces the
proper topological charge in the multi-variable case.
To illustrate this procedure we consider the potentials for which m = 2. Eq. (3.5) implies:
1
1
x = z1
2
2

z12 4z2 .
(3.6)
One now substitutes this into (3.3) and sums over both roots, which amounts to dropping
all square-roots from the result. For n = 4 and 5 one obtains:

2

6
1 3
5 4
1
4
7 3
W2,3 (y, z) = 11
12 z 3 y + 4 yz + 2 a2 yz + z a3 yz + 6 z

1 3
a4 14 a22 y + z2 + a5 12 a2 a3 z + 12
a2 12 a2 a4 ,

W2,4 (z1 , z2 ) = 17 z17 z23 z1 + 2z22 z13 z2 z15 + a2 z22 z1 z2 z13 + 15 z15

+ a3 12 z22 z2 z12 + 14 z14 + a4 13 z13 z2 z1

+ a5 12 z12 z2 + a6 z1 .
11
(3.7)
In the first of the superpotentials we have made the change of variables: z = z1 , z2 =

y + 32 z2 + 12 a2 . Note, in particular that the value of the modulus (the coefficient of yz4 ,
with y and z suitably normalized) in W2,3 is consistent with the results of [1214].
3.2. The SO(2n) superpotentials
There are two infinite series of coset models with this numerator, and we begin with
the SO(2n)/(SO(2n 2) U (1)) coset model. The denominator group, SO(2n 2)
U (1), leaves an SO(2n) pair of vectors invariant, and so ground states of this model are
characterized by the weights of the vector representation of SO(2n). This model may also
be thought of as the D2n minimal model, but we will see that the canonical deformation
leads to a subset of the full set of deformations of the standard D2n potential.
The Casimirs of SO(2n) are defined by

V2k =
ai21 ai22 ai2k , k = 1, 2, . . . , n 1,
1i1 <i2 <<ik n
n = a1 a2 an ,
V
(3.8)
where the ai are the skew eigenvalues of an SO(2n) matrix in the vector representation.
Under SO(2n) SO(2n 2) U (1) the Casimirs of SO(2n) are decomposed into the
Casimirs xj of SO(2n 2) U (1),
V2 = x2 + x12 ,
V2j = x2j + x2j 2 x12 , j = 2, 3, . . . , n 2,

2
n = xn1 x1 ,
V2n2 = xn1
+ x2n4 x12 ,
V
where x1 is the degree 1 Casimir of U (1).
Set V2j = v2j with v2j being some constant and eliminate x2 , . . . , x2n4 , from the
deformed relations V2j = v2j (1 j n 1). We are then left with
2
+ (1)n x12(n1) +
xn1
n1
2(n1j )
v2j x1
= 0,
v2j = (1)n+j v2j ,
j =1
n vn = 0. Introduce the superpotential so that

and V

n1

W
1 2
2(n1j )
n 2(n1)
=
+ (1) x1
+
v2j x1
,
x
x1
2 n1
j =1
W
= xn1 x1 vn ,
xn1
12
then the desired D2n potential is obtained as

v2j
(1)n
1 2
y x+
x 2n1 +
x 2n2j 1 vn y,
2
2(2n 1)
2(2n 2j 1)
n1
W=
(3.9)
j =1
where we have set x1 = x and x n1 = y.

Note that our procedure yields only an n-dimensional subset of the 2n possible relevant
deformations of the conformal D2n superpotential: the foregoing canonical deformation is
odd under x x, y y. Any other deformations would destroy the W (Dn ) symmetry
of the ground states.
We now turn to the SO(2n)/SU(n) U (1) coset model, whose central charge is c =
3n(n1)
2(2n1) . The SU(n) denominator factor fixes a (complex) pair of SO(2n) spinors, and so
the ground states of this model are labeled by a set of spinor weights of SO(2n). The
LandauGinzburg variables are, a priori, the Casimir invariants of U (n). Let b1 , b2 , . . . , bn ,

be the parameters of the SU(n) CSA with ni=1 bi = 0. A set of SU(n) Casimirs can then
be defined by

zk =
bi1 bi2 bik , k = 2, 3, . . . , n.
1i1 <i2 <<ik n
In view of the decomposition 2n = n2 + n 2 under SO(2n)/SU(n) U (1), the SO(2n)

eigenvalues ai are expressed as ai = (bi 2z1 ), i = 1, 2, . . . , n, where z1 denotes the
degree 1 Casimir of U (1). Then one can rewrite the SO(2n) Casimirs (3.8) in terms of zj ,
j = 1, 2, . . . , n. Inspecting the Casimir decomposition we see that the variables, zj , of even
degree and the variable zn can be immediately eliminated using the vanishing relations.
The SO(2n)/SU(n) U (1) model can thus be characterized by [n/2] LandauGinzburg
variables 3 z2j 1 of degree 2j 1 with j = 1, 2, . . . , [n/2] and the superpotential is of
degree 2n 1. Evaluating the central charge we obtain the correct value
c=3
[n/2]

j =1
2(2j 1) 3n(n 1)
=
.
2n 1
2(2n 1)
For n = 2, 3, 4 the coset models are identified with the A2 , A4 and D8 minimal models,
respectively. It is easily checked that the present method produces the corresponding
LandauGinzburg superpotentials with deformations. 4
The foregoing procedure is, in general, very hard to integrate directly to obtain the
general LandauGinzburg superpotential, and so we illustrate it in the first non-trivial
example: n = 5. This model has c = 10
3 > 3 and turns out to be instructive since its shares
interesting properties with the Grassmannian as well as the E6 model. To decompose the
Casimirs Vi of SO(10) we take ai = (bi 2z1 ). 5 One finds
V2 (zj ) = 20z12 2z2 ,
3 Here [n/2] stands for the integral part of n/2.
4 In particular, the deformed superpotential for the SO(8)/SU(4) U (1) model is converted into that for the
SO(8)/SO(6) U (1) model through the SO(8) triality transformation.

5 This choice corresponds to the 16 of SO(10) as will be seen later. The other choice corresponds to the 16.
13
V4 (zj ) = 160z14 16z2 z12 + 12z3z1 + 2z4 + z22 ,

5 (zj ) = 32z5 + 8z2 z3 4z3 z12 + 2z4 z1 z5 ,
V
1
1

V6 (zj ) = 640z16 + 80z13 z3 40z4 12z22 z12 (4z2 z3 + 20z5)z1 2z2 z4 + z32 ,

V8 (zj ) = 1280z18 + 256z16z2 + 64z15z3 + 48z22 160z4 z14 + (160z5 32z2z3 )z13
+ 8z32 z12 + (4z3 z4 + 12z2z5 )z1 2z3 z5 + z42 .
(3.10)
5 ), we solve the first three equations

Once again, setting Vi to constant values, vi (v5 for V
of (3.10) for z2 , z4 and z5 . We then substitute these values of zj , j = 2, 4, 5, into equations
for V6 and V8 . With the notation z1 = x/5, z3 = y, the equation for V6 reads

6
2
4
144 3
38
33 2 2
4
V6 = 184
625 x + 25 x y + y 125 v2 x + 2v4 + 50 v2 x + 4v5 5 v2 y x
+ 12 v2 v4 18 v23 .
Similarly the equation for V8 becomes
V8 =
2 2
6
24 5
12
6
1 2 4
25 x y + 4x y 125 v2 x + 25 v4 50 v2 x

3 1
4
1 3 2
+ 56
25 v 5 + 5 v2 y x + 5 v2 v4 + 20 v2 x

1 4
+ 12 v22 2v4 y + 65 v2 v5 x + 2v5 y 18 v22 v4 + 14 v42 + 64
v2 .
36 8
625 x
In order to obtain the LandauGinzburg superpotential we set

1
W
= (V6 v6 ),
y
2

1 16 2 2
W
= (V8 v8 ) +
x v2 (V6 v6 ).
x
2 25
5
(3.11)
The second term of W/x is introduced so that the integrability condition is satisfied

W
W
=
.
x y
y x
By integrating (3.11) we obtain the deformed LandauGinzburg superpotential for the
SO(10)/SU(5) U (1) coset model
W=
7
2372 9
92 6
36 3 2
1 3
788
140625 x + 625 x y + 25 x y + 6 y 21875 v2 x

63 2
19
2
6
+ 1250
v2 25
v4 x 5 + 125
v2 y 25
v5 x 4

8

3
61 3 3
33 2
+ 75
v6 + 25
v2 v4 1500
v2 x + v4 + 100
v2 y

+ 15 v2 v5 x 2

9 2
13 4
+ 15 v2 y 2 + 2v5 y v8 40
v2 v4 + 14 v42 + 320
v2 + 15 v2 v6 x

1 3
v2 12 v6 y.
+ 14 v2 v4 16
(3.12)
This procedure does not expose any obviously generalizable structure, and so it seems
rather hard to use it to extract the superpotential for general n. Fortunately, it is possible
to generalize the algorithm developed for Grassmannians. The crucial observation is
again that the canonical deformation defines a point in the Cartan subalgebra, while
14
the ground states of the SO(2n)/(SO(2n 2) U (1)) and SO(2n)/(SU(n) U (1))

models are respectively characterized in terms of vector and spinor weights. We can
thus map the LandauGinzburg potential (3.9) onto the LandauGinzburg potential for
SO(2n)/(SU(n) U (1)) by finding the auxiliary equation that relates the vector and spinor
weights, and then cross-eliminating in the proper manner.
We start by eliminating, or integrating out, the variable y in (3.9) using W/y = 0 to
get:
(x) =
W

v2j
(1)n
v 2
x 2n1 +
x 2n2j 1 n .
2(2n 1)
2(2n 2j 1)
2x
n1
(3.13)
j =1
The 2n roots of this equation are, by construction, the 2n vector weights, x = aj . The
critical points of the SO(2n)/(SU(n) U (1)) model are determined by the 2n1 spinorial
combinations: 12 a1 12 an , with an even (or odd) number of signs. The simplest
auxiliary equation has roots x = aj , j = 1, . . . , n, and so determines the vector weights up
to a sign. One now rewrites this polynomial equation in terms of the zj : that is, we define
the polynomial
R(x) =
n
n
n

(x ai ) = (x + bi 2z1 ) = x n 2nz1 x n1 +
ck (zj )x nk .
i=1
i=1
k=2
When calculating the coefficients ck (zj ) we utilize the deformed relations V2i (zj ) = v2i to
eliminate z2i (i = 1, 2, . . .) in favor of z2j 1 with j = 1, 2, . . . , [n/2]. Then we have
ck (zj ) = ck (z1 , z3 , . . . , v2j ),
k = 2, 3, . . . , n 1,
(3.14)
n (zj ) = vn yields
whereas the relation V
cn (zj ) = (1)n vn .
(3.15)
For example, for n = 5 we find:

c2 = 50z12 12 v2 ,
c3 = 140z13 + 3v2 z1 + z3 ,
c4 = 150z14 5v2 z12 10z3 z1 18 v22 + 12 v4 ,
c5 = v5 .
The LandauGinzburg superpotential for the SO(2n)/(SU(n) U (1)) coset model is

(x) over all the roots of R(x) = 0:
then computed by summing W
W z1 , z3 , . . . , z2[n/2]1 ; v2j , vn =
n
(ai ),
W
(3.16)
i=1
where the sums of powers of the aj that appear on the right-hand side are evaluated by
making use of the auxiliary equation R(x) = 0. In particular, the sum of the last term of
(x) becomes
W
n
(1)n
vn2 1
vn cn1 (zj ),
=
2
ai
2
i=1
where (3.15) and (3.14) have been used.
15
We have confirmed that this method correctly recovers the deformed superpotentials
obtained by the procedure of linear elimination for the models with n = 2, 3, 4 and 5.
3.3. The canonically deformed E6 superpotential
There is only one SLOHSS model involving each of E6 and E7 , and so there are no
short-cuts: we need to use the Casimir decompositions. Here we present the computation
for E6 . The details for E7 are similar, but more complicated, and so they have been
included in Appendix A.
The denominator of the E6 model involves SO(10), which fixes a weight in the 27 of E6 ,
and there are thus 27 ground states. The first step to obtaining the superpotential is to
decompose the six Casimirs, Vj , of E6 into the Casimirs, xj , of SO(10) U (1). This was
done explicitly in [12], but we reproduce the result here since there was a typographical
error in the printed version:
V2 (xj ) = 12x12 + x2 ,
V5 (xj ) = 48x15 8x13 x2 + x1 x22 2x1 x4 + 4x5 ,
V6 (xj ) = 4680x16 1062x14x2
177 2 2
2 x1 x2
23 3
8 x2
15x12x4
+ 54 x2 x4 60x1x5 x6 ,
V8 (xj ) = 25830x18 + 7098x16x2 +
3027 4 2
363 2 3
171 4
4 x1 x2 + 8 x1 x2 + 128 x2
4
3
105 2
15 2
35 2
+ 555
2 x1 x4 + 4 x1 x2 x4 32 x2 x4 32 x4 + 1740x1 x5
+ 75x1x2 x5 6x12 x6 12 x2 x6 + 15
8 x8 ,
9
7
5 2
4
V9 (xj ) = 28560x1 1008x1 x2 + 42x1 x2 + 35x13 x23 + 105
16 x1 x2
2
35
2
4
924x15x4 70x13x2 x4 105
4 x1 x2 x4 + 4 x1 x4 + 840x1 x5
2
3
+ 420x12x2 x5 + 35
2 x2 x5 7x4 x5 112x1 x6 + 28x1 x2 x6 21x1 x8 ,
V12 (xj ) = 177660x112 + 97902x110x2 +
36063 8 2
1635 6 3
4 x1 x2 + 4 x1 x2
4 4
577 2 5
15 6
15705 8
+ 7569
64 x1 x2 128 x1 x2 1024 x2 + 2 x1 x4
6
7299 4 2
1741 2 3
1307 4
+ 6087
2 x1 x2 x4 16 x1 x2 x4 + 32 x1 x2 x4 + 1536 x2 x4
4 2
2
7
795 2
423 2 2
85 3
+ 7527
16 x1 x4 32 x1 x2 x4 256 x2 x4 + 128 x4 + 94104x1 x5
3 2
3
3
59
+ 13050x15x2 x5 + 1551
2 x1 x2 x5 8 x1 x2 x5 + 219x1 x4 x5
2 2
19
2
6
4
+ 243
4 x1 x2 x4 x5 + 6x1 x5 2 x2 x5 948x1 x6 + 1041x1 x2 x6
313 2 2
4 x1 x2 x6
50x1x5 x6 +
61 3
25 2
25
48 x2 x6 2 x1 x4 x6 + 24 x2 x4 x6
1 2
4257 4
561 2
97 2
3 x6 4 x1 x8 + 8 x1 x2 x8 + 64 x2 x8
45
32 x4 x8 .
We then solve the equations Vi = vi , where vi are constants, for i = 2, 5, 6, 8. These are the
equations that are linear in xi , and so we use them to directly eliminate x2 , x5 , x6 and x8 .
This leaves the equations: V9 = v9 and V12 = v12 . These are integrable, and former is
W
proportional to W/x4 , and the latter must be proportional to W
x1 + p3 (xj , vj ) x4 , where
p3 (xj , vj ) is a homogeneous polynomial of degree 3.
16
For reasons that will become apparent in Sections 3.4 and 4.2, we replace the vj with
some other constants, w according to:
v2 = 2w1 ,
v5 = 4w2 ,
v8 = 5w4 + 4w1 w3 + 15w14 ,

v12 = 5w6 +
26 2
3 w1 w4
Finally, setting x1 =
superpotential:
1
4x
v6 = w3 + 20w13 ,

v9 = 28 w5 + 2w12 w2 ,
+ 13 w32 + 4w13 w3 + 19w1w22 + w16 .
and x4 =
60
13 z
117 4
16 x
+ 5w1 x 2 + 2w12 one arrives at the
25
5 2
xz3 + 26
z w2
W = x 13 169
9
7
+ z x + x w1 + 13 x 5 w12 x 4 w2 13 x 2 w1 w2 +
(3.17)
1 3
1
12 x w3 6 xw4
247 11
13 9 2
39 8
169 7 3
13 7
165 x w1 + 15 x w1 20 x w2 + 945 x w1 + 105 x w3
26 6
13 5
13 5
91 4 2
13 4
15 x w1 w2 + 225 x w1 w3 50 x w4 180 x w1 w2 + 30 x w5
13 3 2
13 3
13 2
13 2
13
15 x w2 90 x w1 w4 120 x w2 w3 + 90 x w1 w5 270 xw6
13 4
13 2
360 w1 w2 + 90 w1 w5 .
+ 13 w5
(3.18)
This potential then has the property that:

1
W
= (V9 v9 ),
z
84

13
W
13
1
x 3 w1 x (V9 v9 ).
=
(V12 v12 ) +
x
1350
210
6
(3.19)
3.4. Single variable potentials

One can always partially eliminate variables from the LandauGinzburg potential by
using some of the equations, W/xj = 0. Indeed we have already done this for all the
quadratic variables in W : this is what is meant by using the linear vanishing relations. One
can go further and eliminate all variables except the one of lowest degree, the Casimir, x,
of degree 1. The result, W(x; vj ), is defined to be the single variable potential. To be more
explicit, let W (x, y, z, . . . , vi ) be the deformed superpotential, where x, y, z, . . . , are the
LandauGinzburg variables. One can solve the equations of motion
W
W
=
= = 0,
y
z
(3.20)
and expressing y, z, . . . , in terms of x and vi , and substitute the result back into W to
obtain W(x; vj ). This one variable potential has the property that all the ground states are
determined from the solutions of:

d
dW
=
W x, ycl (x), zcl (x), . . . , vi = 0.
dx
dx
In general one will not be able to solve (3.20) explicitly, and so the one variable potential
is generically implicit. On the other hand, for several important examples, the equations can
be explicitly solved and one obtains a polynomial or irrational potential.
17
There is also another more direct way to get the single variable potential, and this
approach is more directly related to the applications considered in [4,15]. In this approach
one takes the characteristic polynomial, P (x; vj ), of a general CSA matrix in the
representation of G that corresponds to the ground state of the SLOHSS model of interest.
One then shifts the Casimir of highest degree according to vr vr + , and solves
P (x, v1 , . . . , vv1 , vr + ) = 0 for (x; vj ). The roots of the equation (x; vj ) = 0 are
precisely the ground states of the SLOHSS model, and so the single variable potential in
terms of the U (1) Casimir, x, is given by:

W(x; vj ) = (x; vj ) dx.
(3.21)
We thus refer to (x; vj ) as the pre-potential. This function is discussed further in
Appendix A.
Given an irrational or implicit potential in a single variable, one can reconstruct the
multi-variable, polynomial potentials in a relatively straightforward manner: one replaces
every algebraically independent irrational or implicit form by a new chiral primary field.
Algebraic vanishing relations are then introduced so that their solution yields the relationships between the new variables and the original irrational or implicit forms. One then
integrates these vanishing relations (where possible) to obtain a LandauGinzburg potential. For all the SLOHSS potentials that we have studied in detail, we have found that
the irrational, single variable potentials are equivalent to the full algebraic, multi-variable
potentials.
There are several reasons why the single variable potentials are useful. First, we will
find that the results of the period integrals in the 4-fold will most directly reduce to the
single variable potential. Knowing the single variable potentials is thus valuable data
in the reconstruction process. More generally, one finds such single variable potentials
also naturally arise in the construction of SeibergWitten Riemann surfaces from period
integrals on CalabiYau 3-folds [7,15,16].
There is a further application of the single variable potentials in the coupling of
topological matter to topological gravity [3,4]. In this context, the physical operators and
their correlators are most directly expressed in terms of a single variable potential and
residue formulae.
We will therefore illustrate both approaches to computing the single variable superpotentials. We will use the first method on the E6 superpotential, and arrive at results closely
related to those of [4,16]. We will then use the second approach to rederive the result from
Section 3.2 for SO(2n)/U (n).
For E6 , one can eliminate the variable z from (3.18) by solving W/z = 0. One finds
13
(w2 p2 ) where,
that: z = 30x
p2 12x 10 + 12w1 x 8 + 4w12 x 6 12w2x 5 + w3 x 4 4w1 w2 x 3
2w4 x 2 + 4w5 x + w22 .
Substituting this into the superpotential one obtains:

13
1 3
W=
,
q0 2 p2
270
2x
(3.22)
(3.23)
18
where
q0
342
11
2 9
63
8
2
3
7
11 w1 x + 18w1 x 2 w2 x + 7 13w1 + 9w3 x

27w1 w2 x 6 + 35 (2w1 w3 9w4 )x 5 32 5w12 w2 6w5 x 4

+ 3 3w22 w1 w4 x 3 32 (w2 w3 2w1 w5 )x 2 3w1 w22 + w6 x

w2 w5 1 w23
+
.
34 w14 w2 + 2w2 w4 4w12 w5 + 3
x
2 x2
270 13
13 x
(3.24)
This is precisely the one-variable E6 potentials discussed in [4,16,17]. In particular, one

has:

1
1 3
1
d
dq0
= 3 q1 w6 ,
= 3 p1 p2 ,
p2
(3.25)
dx
dx 2x 2
x
x
where
p1 78x 10 + 60w1 x 8 + 14w12 x 6 33w2 x 5 + 2w3 x 4 5w1 w2 x 3
w4 x 2 w5 x w22 ,

q1 270x 15 + 342w1 x 13 + 162w12 x 11 252w2x 10 + 26w13 + 18w3 x 9

162w1w2 x 8 + (6w1 w3 27w4 )x 7 30w12 w2 36w5 x 6

+ 27w22 9w1 w4 x 5 (3w2 w3 6w1 w5 )x 4 3w1 w22 x 3
3w2 w5 x w23 .
(3.26)
The pre-potential:
dW
1

(3.27)
= 3 q1 p1 p2 w6
dx
x
is precisely the one that plays a crucial role in defining the SeibergWitten Riemann surface
for E6 [16].
We now start from the other end for the SO(10)/(SU(5) U (1)) model: the ground
states are classified by the 16 of SO(10), and the relevant characteristic polynomial is
therefore given by:
16
PSO(10)
(x) =
16

x 12 a1 12 a2 12 a5 ,
with an even number of signs. Expanding this we have

2

16
1
1 2
PSO(10)
q0 64
(x) = 16
q1 q2 2v8 q0 + v82

= x 16 2v2 x 14 + 74 v22 v4 x 12 12v5 x 11 + ,
5 = v5 . Here q0 , q1 and q2 are
where we have put the Casimirs to V2i = v2i and V
polynomials

5 2 4
q0 = 68x 8 20v2 x 6 8 74 v4 + 16
v2 x 24v5 x 3

3 3 3
1 4
8 32
v2 + 8 v2 v4 12 v6 x 2 + 2v5 v2 x + 64
v2 + 14 v42 18 v22 v4 ,
q1 = 48x 5 8v2 x 3 v22 x + 4v4 x 4v5 ,

q2 = 2 64x 6 16v2 x 4 + 4v22 + 16v4 x 2 x 2 32v5 x + v23 4v2 v4 + 8v6 .
Note that
dq2
16q1 = 0.
dx
19
(3.28)
16
Then we obtain the pre-potential, , by solving PSO(10)
(x; v8 + ) = 0:
16 (x) = v8 + q0 18 q1 q2 .
This expression can then be integrated with the use of (3.28) to give:

1 3
q2 .
W(x) 16 (x) dx = (v8 + q0 ) dx
192
To get to a multi-variable, rational potential one simply has to replace q2 by a new

LandauGinzburg variable. It is convenient to mix this with other degree 3 terms in x and
vj to arrive at the definition:
3
2
y = q2 72
(3.29)
25 x + 5 v2 x.
One now replaces all occurrences of q2 in W(x) using (3.29), and the result is precisely
W (x, y) of (3.12). Moreover, Eq. (3.29) is exactly the solution of W/y = 0.
4. LandauGinzburg models and the IIA theory on 4-folds
4.1. Review
The construction of [8] starts by identifying the vacuum states of the IIA theory that
correspond to the LandauGinzburg vacua. These IIA vacua are obtained by first choosing
a CalabiYau 4-fold, Y , and then choosing a set of fluxes for the 4-form field strength, G,
on Y . There is a quantization condition on the fluxes which requires that the quantity:

1 GG
+
,
N=
(4.1)
24 2
(2)2
be an integer. This means that the characteristic class [G/2] satisfies a shifted Dirac
quantization condition in that the difference between any two fluxes G and G must satisfy
H 4 (Y ; Z). The integer N is then the number of world-filling fundamental strings
in the vacuum state of the theory, and these fundamental strings are to be located at some
points, Pi , chosen in Y . For the vacuum to have a mass gap one must thus have N = 0.
Domain walls, or solitons are represented by kinks that interpolate between spatial
regions in which the G-fluxes are different. In particular, for any element S of the integral
homology H4 (Y, Z), we may choose a D4-brane that wraps this cycle and appears as a
particle in the R1,1 world. Such a brane is a source of G-flux, and indeed across this
domain wall one has = [S] where [S] H 4 (Y ; Z) is the Poincar dual of S. Thus
the model is specified by the family of G-fluxes obtained that are mapped into one another
via such solitonic D4-branes.
Having chosen a G-flux, the complex and Khler structures on Y are required to
satisfy constraints [8,18] in order to preserve the requisite supersymmetry with a zero
20
cosmological constant. Specifically, if Gp,q are the (p, q) parts of G one requires that
G0,4 = G1,3 = 0, and if K is the complexified Khler form, one requires that G K = 0.
These imply that G must be a self-dual (2, 2)-form. In [8] a superpotential was proposed
for the complex structure and Khler moduli. We will only consider the former here, and it
is given by:

1
G.
W (Ti ) =
(4.2)
2
Y
The variation of Hodge structure means that W and dW vanish in the vacuum states, and
the result is the LandauGinzburg theory at a conformal point. More generally, one can
seek vacua of massive LandauGinzburg theories, and then one only impose the condition
that dW = 0 for a vacuum state.
A LandauGinzburg soliton is a BPS state whose mass is equal to its topological charge,
and the latter is given by the change in the value of the superpotential along the soliton.
Since the solitons are represented by D4-branes wrapping integral 4-cycles, one thus
concludes that:

1
(G G ) = ,
5W =
(4.3)
2
Y
[S]

)
where [S] is (the class of) the 4-cycle dual to the class (GG
.
2
For non-compact CalabiYau manifolds, the quantization condition, (4.1), is replaced
by a boundary condition at infinity. That is, there is a new conserved quantity, the flux at
infinity:

1
GG
,
=N +
(4.4)
2
(2)2
and the value of must be given in order to specify the model. Going to non-compact
CalabiYau manifolds also affects the dynamics in other non-trivial ways. In particular,
some of the complex structure moduli will give rise to scalar fields whose kinetic terms are
non-normalizable. These moduli will thus have to have zero kinetic energy, and thus their
dynamics is frozen for the non-compact manifold. These complex structure moduli thus
become true moduli, or coupling constants of the LandauGinzburg model.
In this paper we will follow [8] and consider only non-compact CalabiYau manifolds
that are fibrations of ALE singularities over some complex 2-dimensional base. Specifically, we consider non-compact 4-folds defined by the equation P (z1 , . . . , z5 ) = 0 where
P (z1 , . . . , z5 ) = H (z1 , z2 ) z32 z42 z52 and:
H (z1 , z2 ) = z1n+1 + z22 +
An ,
H (z1 , z2 ) = z1n1 + z1 z22 +

H (z1 , z2 ) = z13 + z24 +
H (z1 , z2 ) = z13 + z1 z23 +
H (z1 , z2 ) = z13 + z25 +
Dn ,
E6 ,
E7 ,
E8 .
(4.5)
21
In these expressions + indicates the addition of all possible relevant deformations of

the singularity. The number of monomials involved in these deformations is the rank of the
ADE group, and the natural degrees of the coefficients of the monomials are the degrees
of the corresponding Casimir invariants. In [8] it was shown that for these singularities the
condition G K = 0 is always satisfied, and all the moduli of the singularity are coupling
constants, and indeed have no dynamics.
The homology, H4 (Y, Z), of a surface determined by (4.5) is naturally identified with
the root lattice, , of the corresponding group, and the intersection form is the Cartan
matrix. The Poincar duals of these cycles are the elements of the compact cohomology,
4 (Y ; Z), while the full set of non-trivial G-fluxes, , are classified by H 4 (Y ; Z), which
Hcpct
can be identified with the weight lattice, , of the group. The fluxes are thus characterized
by weights of the group, monodromies of the singularity will permute these by the action
of the Weyl group, and solitons can add or subtract root vectors.
A model is thus specified by a weight, , in . Conservation of means that adding
or subtracting a root either acts as a Weyl reflection, preserving the magnitude of the flux,
or it shortens the weight, trading some flux for a string in the ground state. As was argued
in [8], a state with N = 0 has massless excitations coming from motions of the string. Thus
a non-trivial model with a mass gap must have N = 0 in all states, and the weight, , must
therefore be miniscule. (Non-miniscule weights can be shortened by adding or subtracting
roots.) In terms of the flux at infinity, , the models with a mass gap are those with
minimum value of in each of the classes / .
Thus the models with mass gaps are classified by miniscule representations of the
underlying ADE group, and as we saw earlier, the choice of a miniscule weight determines
the denominator group of the coset model, and leads us to the hermitian symmetric space
models (2.2).
We now wish to examine how the LandauGinzburg potentials of the SLOHSS models
emerge from the period integrals of the surfaces defined by (4.5). From the foregoing
discussion there is an obvious problem that will arise: all the dynamics is frozen. The
singularity, and its period integrals encode only topological data about the Landau
Ginzburg theory, and do not contain dynamical fields. It turns out that we will still be able to
extract the LandauGinzburg potentials from this topological data, and thus implicitly find
some LandauGinzburg fields. While we cannot give definitive geometric characterizations
of these LandauGinzburg fields, we find that they emerge in the calculation in a very
interesting and natural manner. We will also find that the evaluation of the period integrals
has implicit ambiguities whose resolution corresponds to selecting the miniscule weight,
or flux at infinity, and thus determines the LandauGinzburg variables and chiral ring. We
will remark further upon this in Sections 4.4, and we will begin by direct evaluation of
periods.
4.2. Integrating over 4-cycles
The integral of the 4-form, , over a cycle, S, of the non-compact surfaces defined
by (4.5) can be written:
22

=
[S]
dz1 dz2 dz3 dz4

,
H (z1 , z2 ) z32 z42
(4.6)
where the 4-cycle is defined as follows: z3 runs around branch-cut of the square-root;
z4 runs between z4 = H (z1 , z2 ), i.e., points at which the branch cut shrinks to a point;
the variables z1 , z2 are then integrated over a 2-cycle, S2 , of the singularity H (z1 , z2 )
z52 = 0. The first two integrals are elementary, and reduce to the following (up to an overall
normalization):

dz1 dz2 H (z1 , z2 ).
(4.7)
S2
For the An singularity, one has H (z1 , z2 ) = P (z1 ; aj ) + z22 , where P (z1 ; aj ) is a
polynomial of degree n + 1. As was discussed in [8], the integral (4.7) then reduces to:

dz1 P (z1 ; aj ),
(4.8)
evaluated between the zeroes of P (z1 ; aj ). That is, we have recovered the superpotential,
W , of (3.3), with the integral (i.e., topological charge of the soliton) expressed in terms of
the values of W at its critical points.
It is almost as elementary to perform the integrals for the Dn singularity. In this instance

j
one has: H (z1 , z2 ) = P (z1 ; aj ) + z1 z22 + a0 z2 , where P (z1 ; aj ) = z1n1 + n2
j =0 an1j z1 .
One makes the change of variable z1 = x 2 , and one performs the elementary integral
over z2 to obtain:

2

a02
a2
W = dx P x ; aj 2 = P x 2 ; aj dx + 0 ,
(4.9)
4x
4x
which is to be evaluated between the zeroes of the integrand.
SO(2n)
This is the single variable potential (3.13) for the SO(2n2)U
(1) coset model. As we
remarked earlier, it is obtained from (3.9) by using W/y = 0 to eliminate y. The
superpotential, (3.9), can be recovered by removing the singular term by introducing the
variable y = a0 /2x. Also observe that (4.9) is an odd function and contains only half of
the general versal deformations of the Dn singularity.
The calculation for the E6 singularity is essentially the same as that performed in [16],
and so we will only sketch the details here.
One starts with the singularity:

1
w3 w13 z22
HE6 (zi ) = z13 + z24 + 12 w1 z1 z22 14 w2 z1 z2 + 96

1
1
w4 + 14 w1 w3 18 w14 z1 48
w5 14 w12 w2 z2
+ 96
1 6

1
3 3
3 2
3 2
+ 3456
(4.10)
16 w1 16 w1 w3 + 32 w3 4 w1 w4 + w6 ,
where we have made a convenient choice for the deformation parameters, wj . One
reparameterizes the singularity by setting
z1 = xy + (x),
z2 = y + (x),
(4.11)
and the result is a quartic in y. One now sets

(x) = 14 x 3 + 12 w1 x ,

1
1
2w1 x 2 + w12 + 24x
w2 p2 ,
(x) = 48
23
(4.12)
where p2 is given by (3.22). The functions and are chosen so that the y 3 and y 1 terms
in the quartic vanish, respectively. The integral (4.7) now takes the form:

dx dy y + (x) x (x) y 4 + A(x)y 2 + B(x),
for some functions, A(x) and B(x). By taking a contour at large y this integral reduces to:

1
2

1
1

dx 4 A(x) B(x) =
dx 3 q1 p1 p2 w6
3456
x

1
1 3
q0 2 p2
,
=
(4.13)
3456
2x
and thus we regenerate the one variable superpotential (3.23) with all the correct moduli.
As we remarked earlier, one can reconstruct (3.18) by replacing the singular irrational part
using:
13

w2 p2 z.
30x
4.3. Other superpotentials
Our integration procedure appears to have led us directly to a single superpotential for
the An and Dn models. The reason for this is that we must have implicitly chosen a flux, or
boundary condition at infinity. Indeed, the sleight-of-hand occurred when we passed from
the definite integral to the indefinite integral with a single end-point to the integration.
We know from the algorithms of Section 3 that we can get the multi-variable potentials
for any of the Grassmannian models and for the SO(2n)/U (n) series by summing the
potentials (3.3) and (3.9) over carefully selected sets of their critical points. Exactly the
same choice emerges in the period integrals. For example, in the Grassmannian we could
take the end-points of the integration to be a subset of m solutions of dW/dx = 0, with
this subset defined by (3.5). The result would be an indefinite integral that depends upon
z1 , . . . , zm , and reproduces the multi-variable potential for the Grassmannian.
We therefore see that the selection of the G-flux in H 4 (Y, Z), or the choice of the flux
at infinity, amounts to selecting a set of critical points of the single variable potential.
In the next subsection we will demonstrate how this works more explicitly by showing
that the boundaries of the integration procedure, and in particular these critical points,
correspond to special holomorphic divisors of the surfaces defined by (4.5).
It is also interesting to note that if we now consider the period integral defined by two
critical points of the Grassmannian potential, Wm,n (z1 , . . . , zm ), then it will decompose
into a difference of sums of critical values of the superpotential W1,m+n1 (x). Generically
this will represent the sum of topological charges of a multi-soliton (and thus non-BPS)
state. Thus not all pairs of ground states represent boundary conditions for BPS solitons.
24
This fact was first discovered by considering the integrable LandauGinzburg models,
where it was seen that one could not make a factorizable S-matrix involving only solitons
for certain classes of integrable model. Some boundary conditions could only give rise to
multi-soliton states [12,19]. Indeed, based on the conserved quantities of the E6 model
it was shown in [12] that for the E6 integrable model, the correct soliton spectrum was
exactly given by the prescription that two ground states labeled by weights 1 and 2
would give rise to a single, fundamental BPS soliton if and only if 1 2 was exactly
a root. It is now very satisfying to have a simpler, and far more general string theoretic
explanation of this result: a wrapped of D4-branes on the ADE singularity is a fundamental
BPS soliton if and only if it wraps a cycle that is represented by a root of the Lie
algebra.
4.4. Intersection forms and holomorphic 4-cycles
We have just seen how the LandauGinzburg potentials emerge from some kind of
semi-periods of the singularity: that is, we get the potentials by making the indefinite
integral of the single variable potential, and then parameterizing families of its critical
points. This suggests that one should try to characterize these families of critical points
and the semi-periods more geometrically. As we will see, this is indeed possible since
they represent non-compact homology cycles of the singularity. Ideally we would like to
integrate over these non-compact cycles, but such an integral will diverge. This can be
fixed by compactifying the singularity. Our approach is based upon the techniques used in
Section 17 of [6], and we will, therefore, only summarize the key steps here.
The first step is to introduce a new coordinate and make the equation of the singularity
homogeneous (see [6] for details). The resulting compact manifold will not be a Calabi
Yau manifold and so will have some kind of problem at infinity: if one wants to preserve
holomorphicity, then it will be singular. Since we wish to consider period integrals, we wish
to keep regular, but the cost is the loss of holomorphy in the patch at infinity. This will
not affect the computation of topological charges since they only involve integrals of
over compact cycles of the original singularity. This means that such a compactification
will only affect the LandauGinzburg superpotential by some additive constant.
The basic idea now is to try to write = d for some 3-form, . This is not possible
precisely because is a non-trivial element of cohomology. However, if one excises all the
non-trivial homology from the compactified singularity then it is possible to write = d.
Let Ca be a basis of the non-trivial 4-cycles. One can then write

ma [Ca ],
= d +
a
where [Ca ] denotes the Poincar dual of the cycle Ca , and is a 4-form with delta-function
support on Ca . The parameters, ma are determined from the integral of over Ca :
specifically, if Mab = I(Ca , Cb ) is the intersection form of the basis of cycles, then:

1
ma =
Mab
.
b
Cb
25
Given any other cycle, S, one then has:

=
ma I(S, Ca ),
S
where I(S, Ca ) denotes the intersection numbers of the cycle S with Ca .

The issue now is to find a good basis, Ca ; and there is a particularly nice way to
do this using holomorphic lines in the ADE singularity in two complex dimensions, or
using holomorphic planes in four complex dimensions. The important point is that =
dz1 dz2 dz3 dz4 /z5 is odd under any reflection zj zj , and so any holomorphically
defined dual homology cycle must similarly be odd.
In practice, given the singularity type:
z52 = H (z1 , z2 ) z32 z42 ,
(4.14)
one starts by looking for complex 2-planes of the form

z1 = a + b,
z2 = c + d,
z3 = iz4 ,
(4.15)
where is a complex variable, and a, b, c and d are constants. For generic a, b, c, d

(4.14) and (4.15) define an irreducible rational surface that is even under z5 z5 . At
special values of a, b, c, d one finds that H (a + b, c + d) is a perfect square and
the rational surface degenerates to two intersecting planes: the odd 4-cycles that we seek
are the differences of such pairs of 2-planes. 6 More generally, one can seek holomorphic
2-surfaces of higher degree (see, for example, [20]). Such holomorphic surfaces form Weyl
orbits whose order depends upon the degree of the surface. Here we are focussing on
planes, or surfaces of lowest degree so as to recover the miniscule orbits. One could also
go to surfaces of higher degree, and these are presumably relevant to the physics of more
general fluxes at infinity.
Going back to the earlier parts of this section, one sees that the planes described above
played a crucial role in the explicit evaluation of the integrals. The cycles were defined
by families of contour integrals and the planes defined where these contours collapse to
points, that is, the planes defined the limits of the integration. In simple terms, a 4-cycle is
essentially an S 4 and its intersection points with 2-planes define the extremities, or north
and south poles of the S 4 . Thus the difference of two lines, Ca Cb defines the compact
cycle over which we integrate, and the fact that integration is the difference in values of
the superpotential at the two extremities simply reflects the fact that the integral is given
by ma mb and that md is the value of the superpotential at the line Cd .
The Grassmannian and SO(2n)/U (n) models are obtained by considering families
of critical points of the simplest potentials, i.e., families of critical points for either
U (n)/(U (n 1) U (1)) or SO(2n)/(SO(2n) U (1)), and then summing these simple
potentials over such families. This means that the more general superpotentials must
correspond to taking periods of sums of holomorphic 2-surfaces in the singularity. The
6 Technically they will not strictly be 2-planes since z is generically going to be polynomial in the other
5
variables.
26
correspondence with the flux at infinity is now much more transparent: the Poincar duals
of these surfaces can be taken to be -functions of their volume forms. Since the surfaces
are holomorphic, their volume forms are necessarily (2, 2)-forms. We thus see a very
explicit correspondence between the non-compact (2, 2)-fluxes of the singularity, and
the ground states of the LandauGinzburg superpotential.
Having compactified the singularity, we might hope to have unfrozen the dynamics and
recover some information about the dynamical LandauGinzburg fields. In compactifying
we have made some choices about how to regularize at infinity. We expect these choices
to be related to the choice of the irrelevant D-terms, and the unrenormalized dynamical
LandauGinzburg variables should emerge in terms of properties that can be expressed
entirely in terms of the non-compact singularity. Indeed, by examining the holomorphic
2-surfaces more explicitly for the examples above, we find a much more direct role for the
LandauGinzburg variables.
For the An singularity the zeroes of P (z1 ; aj ) are precisely the points where the
singularity contains the planes:
z5 = iz2 ,
z3 = iz4 .
For the Dn singularity the solutions of P (x 2 ; aj ) = a02 /4x 2 define precisely the points
where the singularity contains the planes:

a0
z5 = x z2 + 2 ,
z3 = iz4 .
2x

Indeed, more generally, the Dn superpotential is given by W (x, y) P (x 2 ; aj ) dx
xy 2 + a0 y, and reduces to (4.9) upon eliminating y via: W/y = 0. The critical points of
the full superpotential W (x, y) precisely define the planes:
z5 = (xz2 + y),
z3 = iz4 ,
lying in the D-type singularity.

The story is similar for the E6 and E7 singularities. The details for the E7 singularity
may be found in Appendix A. For the E6 , the surfaces are quadratic in z5 , but linear in the
other zj . Once again take z3 = iz4 and introduce C with

5
1
z1 = x + 52
2w1 x 2 + w12 ,
z + 48

z2 = 18 2x 3 + w1 x .
The parameters of these surfaces are x and z. When the superpotential W (x, z) of (3.18)
has a critical point then the foregoing defines a plane in the E6 singularity (4.10) with:

2 1
3 6
5
1
15
z5 = 2 + 16
x + 6 w1 z .
x + 18 w1 x 4 + 192
w12 x 2 18 w2 x + 192
w3 + 104
Indeed, the 27 critical points of W (x, z) corresponding to the weights of the fundamental
of E6 define the celebrated 27 lines in a cubic hypersurface in P3 .
We therefore see that the LandauGinzburg variables naturally emerge in the parameterization of representatives of the non-compact homology cycles. In retrospect this is rather
natural: we know that the LandauGinzburg variables must parameterize a family of fluxes,
27
and have non-normalizable kinetic terms in the non-compact singularity. Dual to such
fluxes are non-compact homology 4-cycles, and so the LandauGinzburg dynamics can
be converted into a description of these 4-cycles: they intersect the compact cycles of the
singularity in a manner determined by the inner products of the corresponding weight and
root vectors, and the dynamics is such that these supersymmetric ground states correspond
holomorphic surfaces. It is therefore tempting to think of the ground-state flux as being
created by a non-compact D4-brane that threads the singularity along one of these noncompact cycles. The solitons then intersect these non-compact branes and then combine
with them to yield another non-compact D4-brane that threads the compact part of the
singularity with different set of intersection numbers.
5. Conclusions
The N = 2 superconformal models that arise from the non-compact ADE singularities
are determined by a flux at infinity. For the minimal, or miniscule, fluxes the versal
deformations of the singularity lead to a theory with a mass gap. It was argued in [8] that
these particular models should be perturbations of the SLOHSS models. Here we observed
that the perturbations of the singularity respect the underlying Weyl symmetry, and thus the
corresponding perturbations of the LandauGinzburg superpotential must do the same. We
then used this to either explicitly construct, or provide a precise computational algorithm
for the construction of the LandauGinzburg superpotentials of the SLOHSS models
with the most general perturbations consistent with the Weyl symmetry. We then showed
that exactly the same superpotentials and algorithms emerged from the topological data
and period integrals of corresponding ADE singularities. This provides some extremely
detailed checks on the results of [8]. There are several interesting by-products of our work
which may find application in the study of LandauGinzburg models and in topological
matter coupled to topological gravity.
There are also quite a number of interesting open questions. Our work suggests an
interpretation of the LandauGinzburg variables in terms of moduli of non-compact cycles
that thread the singularity. It would be interesting to verify these ideas more explicitly by
considering nearly singular, compact CalabiYau manifolds in more detail.
Here we have considered only the theories with a mass gap. It would be interesting to
find out exactly what perturbed conformal theories emerge for non-miniscule fluxes. It
is tempting to conjecture that the flux at infinity, , encodes the level of the numerator
current algebra of the coset model. Since versal deformation of the singularity is not
supposed to yield a mass gap, the corresponding perturbed LandauGinzburg models will
have to remain multi-critical. This is certainly possible, but there is a potentially interesting
conundrum here: we know from the results in [21] that simple perturbative attempts at
adding marginally relevant operators can create non-trivial higher order corrections, and
indeed result in a theory with a mass gap. For example, the N = 2 minimal models
perturbed by the least relevant chiral primary results in a theory with a Chebyshev
superpotential. Indeed, the such a perturbation would be a natural candidate for the A1
28
singularity with higher flux. It would thus be interesting to see if the resulting theory really
does have massless modes, or if, somehow the theory does indeed develop a Chebyshev
superpotential and hence have a mass gap. It is also quite possible that the conformal field
theories associated with the singularities with non-miniscule fluxes are not the various
coset models at higher levels and with different denominators.
One way to approach this problem that might be particularly interesting would be to
find a method of computing the elliptic genus of the conformal field theory using only the
topological data of the singularity and its flux at infinity. For LandauGinzburg theories
this is equivalent to finding the fundamental LandauGinzburg fields and determining
their U (1) charges. For more general theories, the elliptic genus gives a lot of valuable
information about the partition function.
Finally, there are some interesting properties of LandauGinzburg solitons that may have
interesting consequences for D4-branes more generally. Most notable, but probably least
relevant is the fact that some very special perturbations lead to quantum integrable models,
with factorizable S-matrices. This sort of property is not at all robust, and so it likely to be
only a property of the field theory in the near singular limit. Integrability will very likely
be spoiled in the full string theory.
More interesting is the fact that solitons have fractional fermion number [22,23]. In
particular, for a LandauGinzburg soliton running between two vacua, the fermion number
is given by [23]:
f =

1
5Im log det(i j W ) ,
2
(5.1)
where 5 indicates the change in the value of the quantity between the two critical points.
Note that this quantity is, in principle, a very complicated function of the moduli and
can presumably assume arbitrary (even irrational) values. However, for singularities with
discrete geometric symmetries (like the Coxeter resolution), this fermion number will
take simple fractional values. One might naturally wonder whether this fractional fermion
number might also be an artifact of the field theory that emerges near the singularity, and
that this fermion number would disappear in the full string theory. There are no global
conserved currents in string theory, and so one might expect that fermion number would
not persist in string theory. However, the formula (5.1) only defines the fermion number
mod 1, and indeed fermion number mod 2 is a well defined concept in string theory. The
fact that (5.1) follows from an index calculation gives some further hope that fractional
fermion number might persist in string theory. Finally, it seems particularly likely that
fractional fermion number will persist if the fractional fermion number is fixed by discrete
geometric symmetries of the singularity.
If some version of fractional fermion number persists in string theory then it would be
very interesting to find the geometric meaning of (5.1). As regards physical consequences:
there are the obvious phases that would arise in scattering, but perhaps more interesting
would be the consequences for quantization conditions under compactification, and thus
for partition functions.
29
Acknowledgements
N.P.W. would like to thank S. Gukov and E. Witten for helpful conversations. The
research of T.E. and S.K.Y. was supported in part by Grant-in-Aid for Scientific Research
on Priority Area 707 Supersymmetry and Unified Theory of Elementary Particles, Japan
Ministry of Education, Science and Culture. T.E. and S.K.Y. are grateful to Japan Society
for Promotion of Science for the exchange program between Tokyo/Tsukuba and USC.
The research of N.P.W. was supported in part by funds provided by the DOE under grant
number DE-FG03-84ER-40168 and by funds under the U.S. Japan Cooperative Science
Program: String Theory and Field Theory Dualities and Integrable Model; NSF Award
#9724831.
Appendix A. The E7 singularity and the deformed coset model
In this appendix we begin by deriving the single-variable version of the Landau
Ginzburg potential for the E7 singularity closely following [16]. We then obtain the multivariable LandauGinzburg potential for the deformed E7 /(E6 U (1)) coset model using
the method developed in the text. Finally we discuss the integral over 4-cycles in a noncompact 4-fold with a singularity of type E7 .
A.1. The single-variable E7 superpotential
The E7 singularity with versal deformations is described by
WE7 (z1 , z2 , z3 ) = 0,
(A.1)
where
WE7 (z1 , z2 , z3 ) = z13 + z1 z23 + z32 w2 z12 z2 w6 z12 w8 z1 z2 w10 z22
w12 z1 w14 z2 w18 ,
(A.2)
where wq are the deformation parameters. To obtain the single-variable potential we

determine the lines for the generic E7 singularity. The process is very similar to that of
the E6 singularity. We first make a change of variables from (z1 , z2 ) to (x, y), where:
z1 = x 2 y + (x),
z2 = y + (x),
(A.3)
where (x) and (x) will be fixed momentarily. With these substitutions, WE7 becomes a
quartic in y. We then set (x) = x 6 + w2 x 4 3(x)x 2 so as to eliminate the y 3 term in
this quartic. Eliminating the term linear in y gives rise to a cubic equation for (x). It is
(x)
+ x 4 , and then the cubic for
convenient to introduce a function Y (x), where (x) = Y4x
Y (x) reads
Y 3 + 3q(x)Y 2r(x) = 0,
where q(x) and r(x) are polynomials in x given by:
q = 28x 10
8
44
3 v2 x
83 v22 x 6 13 v6 x 4 + 29 v8 x 2 13 v10 ,
(A.4)
30

r = 148x 15 + 116v2x 13 + 36v22 x 11 + 4v23 + 83 v6 x 9 + 23 v2 v6 2v8 x 7

1 2
109
2
+ 2v10 23 v2 v8 x 5 + 81
v6 27
v12 x 3 + 8613
v2 v62 23 v14 x.
(A.5)
In writing the foregoing we have reparameterized the deformations of the E7 singularity

as follows:
1
w6 = 12
v6 ,
w8 = 16 v8 ,
w2 = v2 ,

1
1
w12 = 108
2v12 13 v62 ,
w14 = 12
2v14
w10 = 14 v10 ,

109
2
2871 v2 v6 ,
1
w18 = 36
v18 .
(A.6)
Eq. (A.1) now reduces to:

z32 = x 2 y 4 + A(x)y 2 + B(x) .
(A.7)
The lines are determined by requiring that the right-hand side of (A.7) be a perfect
square, which means that the discriminant, , vanishes:
0 = = A2 4x 2 B

1
= 16
p1 Y 2 + p2 Y + p3
2
16
9 v18 x
(A.8)
The polynomials p1 , p2 and p3 are

p1 = 1596x 10 + 88v22 x 6 + 7v6 x 4 + 660v2 x 8 2v8 x 2 v10 ,

p2 = 16872x 15 + 11368v2x 13 + 2952v22x 11 + 176v6 + 264v23 x 9
7 68
5

+ 100v8 + 100
3 v2 v6 x + 3 v2 v8 + 68v10 x

218
+ 29 v62 43 v12 x 3 + 8613
v2 v62 43 v14 x,
14

p3 = 44560x 20 + 41568v2x 18 + 16080v22x 16 + 2880v23 + 2216
3 v6 x
12

2
64
+ 312v2 v6 + 192v24 1552
3 v8 x + 32v2 v6 40v2 v10 3 v12 +

416
2 6
27776
+ 3 v14 16v22 v10 49 v6 v8 32
9 v2 v12 + 8613 v2 v6 x

3488 2 2 4 2 64
2
+ 8613 v2 v6 + 9 v8 3 v2 v14 23 v6 v10 x 4 + 43 v8 v10 x 2 + v10
.
11 2
3 v6
8
x
(A.9)
The polynomials q, r, p1 , p2 and p3 play an important role in our calculations. In particular

they obey remarkable identities
1
3
1
dq 1
dr
= q p1 ,
= r + p2 .
(A.10)
x
dx 2
6
dx 4
8
The condition = 0 yields a single-variable version of the pre-potential, E7 (x; vi ),
for E7 ,

9
E7 (x; vi ) = v18 +
(A.11)
p1 Y 2 + p2 Y + p3 ,
16x 2
where Y , a root of the cubic (A.4), is given by

Y = s+ + s , s+ + 2 s , 2 s+ + s

with = e2i/3 and s = (r q 3 + r 2 )1/3 . Using the relation between the roots and the
coefficients of the cubic equation it is easy to verify from (A.11) that
x
(E7 + v18 )3 + A2 (x)(E7 + v18 )2 + A1 (x)(E7 + v18 ) + A0 (x) = 0,
(A.12)
31
where
9
(6qp1 3p2 ),
16x 2

2
2 2

9
9q p1 6rp1 p2 12qp1 p3 + 3qp22 + 3p32 ,
A1 =
2
16x

3
2 3
9
4r p1 + 6qrp12 p2 + 9q 2 p12 p3 6rp1 p2 p3
A0 =
2
16x

6qp1 p32 + 2rp23 + 3qp22 p3 + p33 .
A2 =
The single-variable pre-potential, E7 (x; vi ), can be integrated with respect to x to yield

the single variable potential, WE7 (x; vi ). This integral has two parts:

WE7 (x; vi ) = dx E7 (x; vi ) = I1 + I2 ,
where
I1 =

dx v18 +

9
p3 ,
16x 2
9
I2 =
16

dx

1
p1 Y 2 + p2 Y .
x2
The integral I1 is easily performed. We only note that since p3 has no linear term in x
there appears no logarithm of x. To evaluate I2 , on the other hand, we first use (A.10) to
derive

4 dY
Y
9
8r
6q
d
Y + (3qY 2r)
2
I2 =
.
dx
Y2 +
16
dx
x
x
x dx
x
By virtue of (A.4) the second term is reduced to

Y
d Y4
4 dY
.
2 =
(3qY 2r)
x dx
dx x
x
Hence we find

27 2
9
dx E7 (x; vi ) =
p
qY 2rY + dx v18 +
3 .
16x
16x 2
(A.13)
Before moving on to the multi-variable potential, we wish to describe how this singlevariable potential is related to the characteristic polynomial of the 56 of E7 . Generally the
characteristic polynomial of a representation R of the Lie algebra G is defined by

PGR (x; Vi ) = det x a H
which is of degree dim R in x. Here a is an r-dimensional vector in the Cartan subspace
spanned by H and the Vi (i = 1, . . . , r = rank G) are the Casimirs built out of a among
be the
which Vr is the top Casimir whose degree equals the Coxeter number h of G. Let
weight of R, then we have

.
x a
PGR (x; Vi ) =
R
32
For the 56 of E7 we obtain

PE567 (x; vi ) = x 56 + 12v2 x 54 + 66v22 x 52 + 2v6 + 220v23 x 50

+ 10v8 + 495v24 + 20v2 v6 x 48

+ 126v10 + 792v25 + 90v22 v6 + 84v2 v8 x 46
44

2
+ 10v12 + 924v26 + 240v23 v6 + 934
3 v2 v8 + 86v2 v10 x + , (A.14)
where the Casimirs have been set to Vi = vi . Note that this has the degree 3 in the top
Casimir v18 , which is analogous to (A.12). In fact, one can show explicitly that

x2 3
2
v18 + A2 (x)v18
+ A1 (x)v18 + A0 (x) .
6
3
Hence, (A.12) for E7 may be expressed as
PE567 (x; vi ) =
PE567 (x, v2 , . . . , v14 , v18 + E7 ) = 0.

This also makes it clear that each of the lines in the E7 singularity we have constructed is
in correspondence with a weight of the 56 of E7 .
More generally, for any ADE group, the single-variable version of the LandauGinzburg
potential can be obtained from (x) dx, where (x) is obtained by solving

PGR x, v1 , . . . , vr + (x) = 0.
Thus the pre-potential, , depends upon the representation, R, and takes the form: 7
R (x; vi ) = vr + FR (x; v1 , . . . , vr1 ),
where FR is some irrational, or implicit function of x whose expansion at x = starts
with the x h term and does not carry the pole term 1/x as can be seen by degree counting.
A.2. The LandauGinzburg potential for the deformed E7 /(E6 U (1)) coset model
The calculation is also parallel to that for the E6 /(SO(10) U (1)) model. We start with
the decomposition of the seven Casimirs, Vj , of E7 into the Casimirs xj of E6 U (1).
Under E7 E6 U (1), the 56 of E7 branches as
56 = 271 + 271 + 13 + 13 .
Accordingly the characteristic polynomial for the 56 of E7 is factorized
PE567 (x) = PE276 (x + x1 ) PE276 (x x1 ) (x + 3x1 ) (x 3x1 ),
7 Although (x) depends on the representation, it turns out that the description of ADE topological matter
R
theories in terms of R is independent of R, but depends only on the singularity type G [4]. This is deeply related
with the universality of the special Prym variety in the theory of spectral curves of periodic Toda lattice, which
plays a fundamental role in formulating SeibergWitten solution of four-dimensional N = 2 YangMills theory
[15,17].
33
from which we read off the Casimir decomposition as follows:

V2 (xj ) = x2 3x12 ,
V6 (xj ) = 60x5x1 + x6 12x22x12 24x2x14 72x16,
V8 (xj ) = 18x2x5 x1 24x2 x16 + x8 12x5x13 54x18 + 3x6x12 ,
V10 (xj ) = x6 x14 4x2x5 x13 + 12x2x18 + 4x22 x16 + x52 + 12x110
12x5x15 + 4x9x1 2x8x12 ,
V12 (xj ) = 12x23x16 + 12x6x16 + 102x2x5 x15 + 36x2 x110 + 114x22x18
+ x12 + 16 x62 + 792x5x17 35x2x6 x14 + 24x24 x14 + 270x112
135x8x14 + 132x22x5 x13 + 222x52x12 144x9x13 18x2 x8 x12
18x2x9 x1 4x22 x6 x12 2x5 x6 x1 ,
V14 (xj ) =
5402 3 8
319 x2 x1
8
2616
319 x6 x1
9
164616
319 x5 x1
109 2
5742 x6 x2
x12 x12 + x2 x6 x16

+
872 4 6
319 x2 x1
4
436 2
957 x2 x6 x1
+ x5 x9
3
8720 3
319 x2 x5 x1
12
10338
319 x2 x1
2
436 3
957 x2 x6 x1
76950 14
2180
319 x1 957 x2 x5 x6 x1
5
10634 2
319 x2 x5 x1
14x2x9 x13 +
+
18x2x5 x17
3
1542
319 x5 x6 x1
11358 2 10
319 x2 x1
66357 2 4
319 x5 x1
2x5 x8 x1
36x9 x15 + 9x8 x16 +

+ x2 x8 x14
2 2
17972
319 x2 x5 x1
109 2 2
1914 x6 x1
872 5 4
319 x2 x1 ,
V18 (xj ) = x92 + 252x118 + 396x2x116 + 288x22x114 + 144x5x113 + 144x23x112

+ 288x2x5 x111 + 24x2x6 x110 + 216x22x5 x19 + 21x22 x6 x18
+ 114x2x8 x18 + 72x23 x5 x17 24x5x6 x17 + 24x2x9 x17
12x6x112 + 36x24x110 + 168x8x110 156x9x19 + 225x52x18
+ 4x62 x16 + 36x12x16 + 108x53x13 + x82 x12 + 12x5x2 x8 x13
9x52 x6 x12 2x12 x5 x1 + 2x8 x9 x1 + 12x22x8 x16 + 180x2x52 x16
+ 84x5x8 x15 + 12x22x9 x15 4x6x8 x14 + 102x5x9 x14 + 12x2x5 x6 x15
+ 36x22x52 x14 + 10x12x2 x14 4x6 x9 x13 + 12x2x5 x9 x12 .
Set Vi = vi with the vi being arbitrary parameters. Since Vj is linear in xj for j = 2, 6,
8, 12, we can express as xj = xj (x1 , x5 , x9 ; v ) by solving Vj = vj . Substituting these xj
into Vk with k = 10, 14, 18 we are left with three relations
Vk (x, y, z; v ) vk = 0,
k = 10, 14, 18,
(A.15)
where, for convenience, we have set x = 3x1 , y = x5 , z = x9 . As we found for E6 , these

relations are integrable. In order to construct the superpotential, particular linear combinations of the relations (A.15) must be chosen so that they can be integrated to a su-
34
perpotential.
These

linear combinations are determined using the integrability conditions:
W
W
=
x y
y x , etc. We then find that there is a superpotential, W (x, y, z), with:
W
= 32 (V10 v10 ),
z

W
= 3(V14 v14 ) + 43
x 4 + 4v2 x 2 (V10 v10 ),
9
y

W
= (V18 v18 ) + 26
x 4 + 43 v2 x 2 (V14 v14 )
27
x

8
172 3
16 2 4
508
6
2
2
+ 173
243 x + 9 x y v8 + 9 v2 x + 243 v2 x + 8v2 xy + 3 v6 x
(V10 v10 ).
The explicit form of the superpotential reads:
W=
1016644 19
33326 14
266 10
16850 9 2
80 5
817887699 x + 177147 x y + 6561 x z + 2187 x y + 27 x yz
4 3
2
17
3 2
753964
45392 2 15
+ 124
9 x y + xz + 2 y z + 81310473 v2 x + 1594323 v2 x

18566
96064 3 13
12
+ 2302911
v6 + 2302911
v2 x + 48826
59049 v2 x y
6532

2 10
1816 4
4640
+ 216513 v2 v6 + 72171
v2 216513
v8 x 11 + 962
729 v2 x y

8
173

239 2
178
418 3
110
v10 + 6561
v2 v6 2187
v2 v8 x 9 + 464
+ 2187
729 v6 + 729 v2 y + 729 v2 z x

2
508
508 2
4
23 2 7
+ 904
81 v2 y 1701 v2 v10 5103 v2 v8 1701 v12 + 1701 v6 x

6
145
44 2
+ 109
243 v2 v6 81 v8 y + 243 v2 z x
5

2 2
26
1736
2
2
16 2
29
+ 140
27 v2 y 135 v14 + 387585 v2 v6 405 v2 v12 45 v2 v10 405 v8 v6 x

4
43
7
+ 41
27 v2 v8 9 v10 y + 54 v6 z x

2
16
2
1 2
4
218 2 2 3
+ 11
9 v6 y + 9 v2 yz 9 v6 v10 + 9 v8 9 v2 v14 + 25839 v2 v6 x

1 2
+ 4v2 y 3 + 13 v12 4v2 v10 + 18
v6 y 13 v8 z x 2

109
+ v8 y 2 + v8 v10 v14 x + 3v14 + 1914
v2 v62 y 32 v10 z.
(A.16)
Putting vi = 0 and making a change of variables

1/4

5
x = X1 ,
y = 2 2791
X5 + 416
19
81 X1 ,

1/2

9
496 2791 1/4 4
z = 6 2791
X9 593188
X1 X5 ,
19
6561 X1 27
19
we obtain the superpotential at criticality
19
19 3/4 14
19 1/2 10
W = 100476
X1 + X1 X92 + X52 X9 + 37 2791
X1 X5 21 2791
X1 X9
19
which agrees with the result of [12].
The multi-variable superpotential W (x, y, z) does indeed reduce to the single-variable
potential, WE7 (x) of (A.13) by eliminating y and z from W with the aid of the equations
of motion. First, solving W/z = 0 we find:
1 133 10 40 5
55
22 2 6
7
zcl = 6561
x + 27 x y + 34 y 2 + 729
v2 x 8 + 243
v2 x + 108
v6 x 4
x

+ 89 v2 x 3 y 16 v8 x 2 34 v10 .
Substituting this into W/y and letting y = Y +

9 3
W
=
Y + 3qY 2r .
y z=zcl
4x
416 5
81 x
32
3
27 v2 x
35
lead to
Thus W/y = 0 is equivalent to the cubic equation (A.4). After some algebra we finally
arrive at:

W
= E7 (x; vi ).
x z=zcl , y=ycl
Therefore,

dx E7 (x; vi ) = W x, ycl (x), zcl (x); vi .
(A.17)
A.3. Integrating over 4-cycles in the E7 singularity

We consider a non-compact 4-fold defined by (4.5) of type E7 , where H (z1 , z2 ) =
WE7 (z1 , z2 , z3 = 0), and evaluate the period integral (4.6). As for E6 , after the change
of variables (A.3), H (z1 , z2 ) takes the form of (A.7), and hence the integral (4.7) becomes

dx dy 2xy + (x) x 2 (x) x 2 y 4 + A(x)y 2 + B(x).
Performing the integral over y along a contour at large |y| we obtain

2
1
1
A(x) B(x) =
dx
dx E7 (x; vi ),
4x
36
where E7 (x) is given by (A.11). The remaining integral has already been evaluated. The
result is (A.13) and (A.17).
In Section 4.4 we have seen how holomorphic 2-surfaces of the form (4.15) lying in
the An , Dn and E6 singularities are determined by the critical points of the deformed
superpotentials. Turning to E7 we may take:

y2
3
z
1787xy + 243 4 + 324 3
z1 = x 2 +
133
x
x

1 527
v10
v8
y
+
v2 x 4 + 88v22 x 2 + 63v6 + 864v2 162 2 729 4 ,
133 9
x
x
x
8
23 4 1 y
x +
v2 x 2 ,
z2 =
(A.18)
81
4 x 27
where C and (x, y, z) are parameters. The superpotential W (x, y, z) of (A.16) for
the E7 /(E6 U (1)) model has 56 critical points corresponding to the 56 of E7 . At a
critical point, (A.18) and z3 = iz4 describe the holomorphic 2-surfaces, i.e., H (z1 , z2 )
with (A.18) becomes a perfect square, yielding

211
869 9 19 4
1
1 2 5
x
x y+ z
v2 x 7
v x
z5 = x 2
26244
27
4
2916
18 2

11
1
1
1 v10
v6 x 3 v2 x 2 y +
v8 x
.
432
3
24
16 x
36
References
[1] E. Martinec, Algebraic geometry and effective Lagrangians, Phys. Lett. B 217 (1989) 431;
C. Vafa, N.P. Warner, Catastrophes and the classification of conformal theories, Phys. Lett.
B 218 (1989) 51.
[2] E. Witten, Topological quantum field theory, Commun. Math. Phys. 117 (1988) 353;
T. Eguchi, S.-K. Yang, N = 2 superconformal models as topological field theories, Mod. Phys.
Lett. A 5 (1990) 1693.
[3] R. Dijkgraaf, H. Verlinde, E. Verlinde, Topological strings in D < 1, Nucl. Phys. B 352 (1991)
59;
K. Li, Recursion relations in topological gravity with minimal matter, Nucl. Phys. B 354 (1991)
725;
T. Eguchi, Y. Yamada, S.-K. Yang, Topological field theories and the period integrals, Mod.
Phys. Lett. A 8 (1993) 1627, hep-th/9304121;
T. Eguchi, H. Kanno, Y. Yamada, S.-K. Yang, Topological strings, flat coordinates and
gravitational descendants, Phys. Lett. B 305 (1993) 235, hep-th/9302048.
[4] T. Eguchi, S.-K. Yang, A new description of the E6 singularity, Phys. Lett. B 394 (1997) 315,
hep-th/9612086.
[5] N. Seiberg, E. Witten, Electricmagnetic duality, monopole condensation, and confinement in
N = 2 supersymmetric YangMills theory, Nucl. Phys. B 426 (1994) 19, hep-th/9407087.
[6] N. Seiberg, E. Witten, Monopoles, duality and chiral symmetry breaking in N = 2 supersymmetric QCD, Nucl. Phys. B 426 (1994) 19;
N. Seiberg, E. Witten, Nucl. Phys. B 431 (1994) 484, hep-th/9408099.
[7] S. Kachru, A. Klemm, W. Lerche, P. Mayr, C. Vafa, Non-perturbative results on the point
particle limit of N = 2 heterotic string compactifications, Nucl. Phys. B 459 (1996) 537, hepth/9508155;
A. Klemm, W. Lerche, P. Mayr, C. Vafa, N.P. Warner, Self-dual strings N = 2 supersymmetric
field theory, Nucl. Phys. B 477 (1996) 746, hep-th/9604034.
[8] S. Gukov, C. Vafa, E. Witten, CFTs from CalabiYau four-folds, Nucl. Phys. B 584 (2000) 69,
hep-th/9906070.
[9] Y. Kazama, H. Suzuki, Characterization of N = 2 superconformal models generated by coset
space method, Phys. Lett. B 216 (1989) 112;
Y. Kazama, H. Suzuki, New N = 2 superconformal field theories and superstring compactification, Nucl. Phys. B 321 (1989) 232.
[10] A.N. Schellekens, N.P. Warner, Conformal subalgebras of Kac-Moody algebras, Phys. Rev.
D 34 (1986) 3092.
[11] W. Lerche, C. Vafa, N.P. Warner, Chiral rings in N = 2 superconformal theories, Nucl. Phys.
B 324 (1989) 427.
[12] W. Lerche, N.P. Warner, Polytopes and solitons in integrable, N = 2 supersymmetric Landau
Ginzburg theories, Nucl. Phys. B 358 (1991) 571.
[13] D. Nemeschansky, N.P. Warner, Refining the elliptic genus, Phys. Lett. B 329 (1994) 53, hepth/9403047.
[14] W. Lerche, A. Sevrin, On the LandauGinzburg realization of topological gravities, Nucl. Phys.
B 428 (1994) 259, hep-th/9403183.
[15] E. Martinec, N.P. Warner, Integrable systems and supersymmetric gauge theory, Nucl. Phys.
B 459 (1996) 97, hep-th/9509161.
[16] W. Lerche, N.P. Warner, Exceptional SW geometry from ALE fibrations, Phys. Lett. B 423
(1998) 79, hep-th/9608183.
[17] K. Ito, S.-K. Yang, Flat coordinates, topological LandauGinzburg models and the Seiberg
Witten period integrals, Phys. Lett. B 415 (1997) 45, hep-th/9708017;
[18]
[19]
[20]
[21]
[22]
[23]
37
K. Ito, S.-K. Yang, A-D-E singularity and prepotentials in N = 2 supersymmetric YangMills

theory, Int. J. Mod. Phys. A 13 (1998) 5373, hep-th/9712018.
K. Becker, M. Becker, M-theory on eight-manifolds, Nucl. Phys. B 477 (1996) 155, hepth/9605053.
P. Fendley, S.D. Mathur, C. Vafa, N.P. Warner, Integrable deformations and scattering matrices
for the N = 2 supersymmetric discrete series, Phys. Lett. B 243 (1990) 257;
A. LeClair, D. Nemeschansky, N.P. Warner, S-matrices for perturbed N = 2 superconformal
field theory from quantum groups, Nucl. Phys. B 390 (1993) 653, hep-th/9206041.
J.A. Minahan, D. Nemeschansky, Superconformal fixed points with En global symmetry, Nucl.
Phys. B 489 (1997) 24, hep-th/9610076;
J.A. Minahan, D. Nemeschansky, N.P. Warner, Investigating the BPS spectrum of non-critical
En strings, Nucl. Phys. B 508 (1997) 64, hep-th/9705237.
R. Dijkgraaf, E. Verlinde, H. Verlinde, Topological strings in D < 1, Nucl. Phys. B 352 (1991)
59.
R. Jackiw, C. Rebbi, Solitons with fermion number 1/2, Phys. Rev. D 13 (1976) 3398;
J. Goldstone, F. Wilczek, Fractional quantum numbers on solitons, Phys. Rev. Lett. 47 (1981)
986.
P. Fendley, K. Intriligator, Scattering and thermodynamics of fractionally-charged supersymmetric solitons, Nucl. Phys. B 372 (1992) 533, hep-th/9111014;
P. Fendley, K. Intriligator, Scattering and thermodynamics of integrable N = 2 theories, Nucl.
Phys. B 380 (1992) 265, hep-th/9202011.

Optimal charge and color breaking conditions in

the MSSM
C. Le Moul a,b
a Department of Theoretical Physics, Aristotle University of Thessaloniki, GR-54006 Thessaloniki, Greece
b Physique Mathmatique et Thorique, UMR No 5825-CNRS, Universit Montpellier II,
F-34095 Montpellier Cedex 5, France

Received 2 February 2001; accepted 4 April 2001
Abstract
In the MSSM, we make a careful tree-level study of charge and color breaking conditions in the
plane (H2 , u L , u R ), focusing on the top quark scalar case. A simple and fast procedure to compute
the VEVs of the dangerous vacuum is presented and used to derive a model-independent optimal
CCB bound on At . This bound takes into account all possible deviations of the CCB vacuum from
the D-flat directions. For large tan , it provides a CCB maximal mixing
for the stop scalar fields
t1 , t2 , which automatically rules out the Higgs maximal mixing |At | = 6 mt. As a result, strong
limits on the stop mass spectrum and a reduction, in some cases substantial, of the one-loop upper
bound on the CP-even lightest Higgs boson mass, mh , are obtained. To incorporate one-loop leading
corrections, this tree-level CCB condition should be evaluated at an appropriate renormalization scale
which proves to be the SUSY scale. 2001 Elsevier Science B.V. All rights reserved.
PACS: 12.60.Jv; 11.30.Qc; 14.80.Ly; 14.80.Cp
Keywords: Supersymmetry; Charge and color breaking; Soft terms; Higgs phenomenology
1. Introduction
Unlike the Standard Model (SM), the scalar sector of the Minimal Supersymmetric
Standard Model (MSSM) [1] is extremely large and contains many scalar fields, some
of them having non-trivial color and/or electric charges. The presence of such a large
sector is dictated by supersymmetry (SUSY) [1,2]. At the Fermi scale, global SUSY
must however be broken and soft SUSY breaking terms which enter mostly in the scalar
sector of MSSM distort the simple analytic structure of the SUSY effective potential and
are responsible of a blowing-up of the number of free parameters in the MSSM [1,2].
There are many ways to reduce to a great extent this huge number of parameters, further
E-mail address: lemouel@physics.auth.gr (C. Le Moul).
0550-3213/01/$ see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 7 2 - 9
C. Le Moul / Nuclear Physics B 607 (2001) 3876
39
improving the predictivity of the MSSM. Each one relies on some particular scenario of
SUSY breaking and mediation to the MSSM spectrum [3,4]. Whatever such a modeldependent scenario may be, phenomenological consistency at the Fermi scale requires that
spontaneous symmetry breaking of the SM gauge group should occur into the Electroweak
(EW) vacuum, not in a color and/or electric charged vacuum. This Charge and Color
Breaking (CCB) danger which does not exist in the SM was quickly realized in the MSSM
[5], and has been extensively studied ever since [510], providing useful complementary
CCB conditions on the soft parameters.
The major difficulty in obtaining reliable CCB conditions comes from the extremely
involved structure of the effective potential whose global minimum determines the vacuum
of the theory. As a consequence, CCB studies concentrated on simple directions in the
scalar field space, restricting also often to D-flat directions [59]. The last requirement
simplifies greatly the analytical study of the minima of the potential and provides already
rather strong CCB constraints which may in some cases rule out model-dependent scenarii
[7] or, at least, severely constrain them [6,7]. Another alternative is to handle the problem
in a purely numerical way [8,10], a rather blind method, time-consuming, which moreover
faces the danger of missing CCB vacua because of the complexity of the potential.
Only but a few studies considered analytically, or semi-analytically, possible deviations
of the CCB vacuum from D-flat directions. In Ref. [6], in particular, it was shown that
in the interesting field planes (H1 , H2 , tL , tR ) and (H1 , H2 , tL , tR , L ), such deviations
of the CCB vacuum typically occur, due essentially to large effects induced by the
top Yukawa coupling [6]. To take into account this important feature, a semi-analytical
procedure was proposed, and then illustrated in an mSUGRA context, giving refined CCB
conditions in terms of the universal soft parameters at the GUT scale [6]. We stress,
however, that model-dependent assumptions are implicitly present in this procedure, and
need to be relaxed to get a fully satisfactory model-independent picture of CCB conditions.
Furthermore, this procedure, somewhat, does not lend itself easily to the derivation of
analytical expressions for the CCB Vacuum Expectation Values (VEVs) of the fields, and,
hence, for the optimal conditions to avoid CCB. On the technical side, our purpose in this
paper is to overcome these difficulties, though in the restricted plane (H2 , tL , tR ). This
plane will actually provide us with a simplified framework where to present in detail an
alternative procedure to evaluate the CCB VEVs. This way we will include in our study all
possible deviations from the D-flat directions, and obtain an accurate analytical information
on them. This procedure can also be adapted to extended planes, and the present study will
be followed by a complete investigation of CCB conditions in the planes (H1 , H2 , tL , tR )
and (H1 , H2 , tL , tR , L ) [11,12].
More fundamentally, the plane (H2 , tL , tR ) is also of particular interest for the following
reasons:
(i) The CCB vacuum typically deviates largely from the SU(2)L U (1)Y D-flat
direction, as already observed in [6], but also from the SU (3)c D-flat direction. The
latter result, also shared by the potential in the extended planes (H1 , H2 , tL , tR ) and
(H1 , H2 , tL , tR , L ) [11,12], is in disagreement with the claim of [6]. As we will see,
this important feature must be incorporated in order to obtain an optimal CCB condition
40
which encompasses the requirement of avoiding a tachyonic lightest stop. We will give
simple analytic criteria for alignment of the CCB vacuum in D-flat directions and show
that alignment in the SU(3)c D-flat direction is in fact a model-dependent statement which
is approximately valid in an mSUGRA scenario [3], but not in other interesting models,
e.g., some string-inspired or anomaly mediated scenarii [4].
(ii) For large tan , the EW vacuum is located in the vicinity of the plane (H2 , tL , tR ).
Therefore, in this regime, the study of this plane is self-sufficient: the free parameters
entering the effective potential are enough to evaluate the optimal necessary and sufficient
condition on At to avoid CCB. This does not mean that CCB conditions in the plane
(H2 , tL , tR ) are useless for low tan . We will give in this paper an analytic optimal
sufficient condition to avoid CCB in this plane, and, to evaluate the complementary optimal
necessary CCB condition, we will simply need some additional information on the depth
of the EW potential, which reduces in fact to a particular choice for tan and the pseudoscalar mass mA0 .
Our purpose in this study is also to consider some physical consequences at the SUSY scale
of the CCB conditions. We will investigate in detail the benchmark scenario MSUSY =
mtL = mtR and tan = +, often considered in Higgs phenomenology [13,14]. In this
case, the stop mixing parameter equals the trilinear soft term,A t At + / tan = At . We
will show that the so-called Higgs maximal mixing |At | = 6 mt is always largely ruled
out by the optimal CCB condition. This will lead us to introduce a CCB maximal mixing,
which induces strong bounds on the stop mass spectrum. Another direct implication of
this result is a lowering of the one-loop upper bound on the CP-even lightest Higgs boson
mass mh reached for such a large tan regime [13,14]. For illustration, this point will
be considered in a simplified setting, where only top and stop contributions to mh will
be taken into account, assuming mA0 mZ 0 . This will already point out the importance
of CCB conditions in this context, but should, however, be completed, to become more
realistic, by a refined investigation including all one-loop and two loop contributions to
mh [13,14]. In these illustrations, the leading one-loop corrections to the CCB condition
will be incorporated by assuming that the tree-level CCB condition obtained are evaluated
at an appropriate renormalization scale, estimated in fact to be the SUSY scale [6,15].
This way, we expect the results presented in this paper to be robust under inclusion of such
radiative corrections. Finally, we note that these results can be shown to be also numerically
representative of the large tan regime with small enough values of the supersymmetric
term , i.e., tan 15 and || Min[mA0 , MSUSY ] [11].
(iii) As is well-known, for metastability considerations, CCB vacua associated to the
third generation of squarks are the most dangerous ones [9,10,16]. This comes from the fact
that such vacua prove to be rather close to the EW vacuum, resulting in a barrier separating
both vacua more transparent to a tunneling effect. We will see that combining experimental
data on the lower bound of the lightest stop mass, mt1 , with precise CCB conditions already
delineates large regions in the parameter space where the EW vacuum is the deepest one
and, hence, stable. Outside such regions, an optimal determination of the modified CCB
metastable conditions requires first a precise knowledge of the geometrical properties of
the effective potential, e.g., the positions of the CCB vacua and saddle-points. In this
41
light, the analytical expressions presented in this article provide an essential information
to investigate precisely metastability.
The paper is organized as follows. In Section 2, we first review the issue of CCB
conditions in the plane (H2 , u L , u R ) in the D-flat direction. We turn then to the full plane
case for the third generation of squark fields and give a first simple analytical sufficient
condition on At to avoid CCB. In Section 3, we detail our semi-analytical procedure
to obtain the VEVs of the local extrema in this plane, discuss the deviation from the
SU(3)c D-flat direction, and give an optimal sufficient bound on At to avoid CCB. We
discuss finally some geometrical features of the CCB vacuum. In Section 4, we discuss the
renormalization scale at which the tree-level CCB conditions obtained should be evaluated
in order to incorporate leading one-loop corrections. In Section 5, we summarize the
practical steps needed to evaluate numerically the optimal necessary and sufficient CCB
condition on At . Sections 67 are devoted to numerical illustrations and, for large tan ,
to phenomenological implications of the new optimal CCB condition for the stop mass
spectrum and the one-loop upper bound on the lightest Higgs boson mass. Section 8
presents our conclusions. Finally, the Appendices A and B contain some technical material
and the generalization of this study to the plane (H1 , bL , bR ), valid for a large bottom
Yukawa coupling, or equivalently for large tan .
2. CCB conditions in the plane (H2 , u L , u R )
We consider the tree-level effective potential in the plane (H2 , u L , u R ), where H2
denotes the neutral component of the corresponding Higgs scalar SU(2)L doublet, and
u L , u R are respectively the left and right up squark of the same generation. In this plane,
the tree-level effective potential reads [5]

V3 = m22 H22 + m2u L u 2L + m2u R u 2R 2Yu Au H2 u L u R + Yu2 H22 u 2L + H22 u 2R + u 2L u 2R

2 g 2
2
g12
u 2L 4u 2R 2 g22 2
2
H2 u 2L + 3 u 2L u 2R .
+
(1)
+
H2 +
8
3
3
8
6
We suppose that all fields are real and that H2 , u L are positive, which can be arranged
by a phase redefinition. The Higgs mass parameter m22 = m2H2 + 2 can have both signs,
mH2 being the soft mass of the corresponding Higgs field; mu2 L , m2u R are the squared soft
masses of the left and right up squarks and are supposed to be positive to avoid instability
of the potential at the origin of the fields; Yu and Au stand for the Yukawa coupling and
the trilinear soft coupling and are also supposed to be real and positive, which can be
arranged once again by a phase redefinition of the fields; finally g1 , g2 , g3 are respectively
the U (1)Y , SU(2)L , SU(3)c gauge couplings.
2.1. The D-flat direction
In the D-flat direction |H2 | = |u L | = |u R |, the potential V3 , Eq. (1), may develop a very
deep CCB minimum, unless the well-known condition [5,6]
42

2
2

2
2
A2u AD
u,3 3 mu L + mu R + m2
(2)
is verified. Strictly speaking, as the extremal equations easily show, the global minimum
of the potential V3 , Eq. (1), lies in the D-flat direction only for:
m2u L = m2u R = m22 .
(3)
However, in the small Yukawa coupling regime, valid for the first two generations of quarks,
the VEVs of the CCB vacuum are large, Au /3Yu . The vacuum then proves to be
located in the vicinity of this direction [5,6], even for large deviations from the mass
relations in Eq. (3). Moreover, due to the smallness of the Yukawa coupling, this CCB
2
2
minimum is very deep, V3 A2u [A2u (AD
u,3 ) ]/27Yu , and, with increasing Au , gets
rapidly 1 deeper than the realistic EW vacuum. As a result, the relation Eq. (2) turns out to
provide an accurate necessary and sufficient condition to avoid a CCB in this plane [5,6].
In the large Yukawa coupling regime, valid for the top quark case, the condition Eq. (2)
is now only approximately necessary, because in some (small) range of values for At
where it is not verified the CCB local minimum in the D-flat direction develops without
being deeper than the EW vacuum. It is however no more sufficient, the true global CCB
minimum of V3 , Eq. (1), being in general located far away from the D-flat direction [6]!
Obtaining the most accurate conditions to avoid CCB in the top quark regime needs to
explore the scalar field space outside D-flat directions, a more difficult task which is of
particular phenomenological interest, as we will see. In the following, we focus on this
interesting regime in order to get a complete model-independent picture of CCB conditions
in the plane (H2 , tL , tR ).
2.2. The full-plane case
Beyond a critical value for the trilinear soft term At , a dangerous CCB minimum, deeper
than the EW vacuum, forms and deepens with increasing values of At . In Ref. [6], it
was advocated that such a global CCB minimum is located in general far away from
the SU(2)L U (1)Y D-flat directions, but close to the SU(3)c D-flat one. This work
was performed in more extended planes with additional scalar fields, H1 and possibly
a sneutrino field L , which we will consider in separate articles [11,12]. Already in the
plane (H2 , tL , tR ), our study indeed shows a typical large deviation of the CCB vacuum
from the SU(2)L U (1)Y D-flat directions. However, in a model-independent way, we
disagree with the claim of Ref. [6] that the CCB vacuum always proves to be located in the
vicinity of the SU (3)c D-flat direction. Actually, alignment in this direction depends on
the magnitude of the soft terms At , mtL , mtR [see Section 3.2] and occurs in two different
circumstances: either (i) mtL = mtR , or (ii) At mtL , mtR . Any discrepancy between the
soft masses mtL , mtR , as happens for instance in some anomaly mediated models [4], is the
source of a possibly large departure of the CCB vacuum from the SU(3)c D-flat direction
and, ultimately, of a sizeable enhancement of the optimal condition on At to avoid CCB.
1 Within less than 1 GeV.
43
To consider this feature, we introduce a new parameter which conveniently measures the
separation of the CCB extrema from the SU(3)c D-flat direction
tR
.
(4)
tL
Alignment in the SU(3)c D-flat direction corresponds to f = 1.
We replace now tR f tL in V3 , Eq. (1), which is unambiguous provided tL = 0. By
inspection of the extremal equation associated to the field tL , it is easy to obtain a critical
bound on At below which no CCB local minimum may exist. The non-trivial solution for
the VEV tL verifies
f
A3 tL 2 + 2B3 = 0,
(5)
where A3 g12 (4 f 2 1)2 /18 + g22 /2 f 4 + 2g32 ( f 2 1)2 /3 + 4Yt2 f 2 . This term is
a sum of squared terms, therefore always positive. This implies the inequality
[(12Yt2 4g12 ) f 2 + 12Yt2 + g12 3g22 ]
12
2At Yt f H2 + m2t + f 2 m2t 0.
B3 H2 2
(6)

B3 may be considered as a polynomial in H2 . For Yt Max[ (3g22 g12 )/12, g1 / 3]

the coefficient of the quadratic term in H2 is positive, whatever f is. At the EW scale,
this relation reduces to Yt 0.3, which anyway must be verified in order to have a correct
top quark mass mt 175 GeV. Obviously, the running of the parameters Yt , g1 , g2 with
respect to the renormalization scale will not alter this result, so that we can safely conclude
that this coefficient is always positive. Actually, this is precisely the turning point where
the qualitative difference between the large and the small Yukawa coupling regimes enters
the game.
Keeping in mind this feature, Eq. (6) implies that f (and thus tR ) must be positive.
The negativity of B3 requires in addition the positivity of the discriminant of B3 considered
as a polynomial in H2 :

2
2
g12

2 2
2
2
2
4 f 1 + g2
B3 = mt + f mt 4Yt f + 1 +
L
R
3
+ 4A2t Yt2 f 2 0.
(7)
L
This inequality may be expressed as a condition on At

A2t
(m2t + f 2 m2t )[4Yt2 ( f 2 + 1) +

L
g12
2
3 (4 f
1) + g22 ]
.
(8)
4Yt2 f 2
The right-hand side of this relation considered as a function of f is bounded from below
and gives an absolute lower bound on At below which B3 cannot be positive. This lower
bound provides a first very simple sufficient condition to avoid any CCB minimum in the
plane (H2 , tL , tR ):

g12
(3g22 g12 )
(0)
At At mtL 1
(9)
+
m
1
mtL + mtR .
t
R
3Yt2
12Yt2
44
If this condition is verified, the global minimum of the potential V3 , Eq. (1), is
automatically trapped in the plane tR = tL = 0 and cannot be lower than the EW vacuum.
Such a simple relation also sets a lower bound on At above which CCB may possibly
occur and is already quite useful to secure some model-dependent scenarii. As a simple
illustration, we consider the infrared quasi-fixed point scenario for low and large tan ,
in an mSUGRA context [17,18]. For a top Yukawa coupling large enough at the GUT
scale, the soft parameters mtL , mtR and At are strongly attracted in the infrared regime to
quasi-fixed points. For Ygit |MGUT = 5 and m0 m1/2 , where gi stands for the three gauge
couplings that unify at the GUT scale, MGUT , and m0 , m1/2 are respectively the unified
scalar and gaugino masses at the GUT scale, we have at one-loop level, in the infrared
regime [18]:
m2t 0.70 M32 ,
L
m2t
L
0.58
m2t 0.48 M32 ,
for low tan ,
(10)
0.52
for large tan ,
(11)
M32 ,
m2t
R
M32 ,
(0)
giving A(0)
t 1.53 |M3 | (for low tan ) and At 1.48 |M3 | (for large tan ), where
M3 is the gluino mass. Comparing these bounds with the infrared quasi-fixed point value
|At | 0.62 |M3 | (for both low and large tan ) [18], we see that the sufficient condition
|At | A(0)
t , Eq. (9), is largely fulfilled. Therefore, we conclude that the infrared quasifixed point scenario is free of CCB danger in the restricted plane (H2 , tL , tR ).
2.3. The critical CCB condition

In order to study the CCB extrema of the potential V3 , Eq. (1), we suppose in the
following that At > A(0)
t . Consistency requires that, at any CCB extremum, f and H2
are restricted to intervals which depend essentially on the soft breaking terms At , mtL , mtR :
the positivity of B3 , Eq. (7), which may be viewed as a polynomial in f 2 , restricts f
in the interval given by the real and positive roots of B3 ; H2 must be included between
the real and positive roots of B3 , Eq. (6). The potentially dangerous region of positive tL 2
then proves to be located in a compact domain in the plane (H2 , f ), growing with At , with
maximal extension
At
6At Yt
,
0 H2
(12)
Yt
(12Y 2 + g 2 3g 2 )(3Y 2 g 2 )

0 f
A2t
m2t (1 g12 /3Yt2 ) m2t (1 (3g22

L
R

mtR 1 g12 /3Yt2
g12 )/12Yt2 )
At
.
mtR
(13)
Let us now replace in the potential V3 , Eq. (1), tL by the solution of Eq. (5), and calculate
the remaining two extremal equations. The derivative with respect to H2 provides an
equation cubic in H2 and quartic in f
3 H23 + 3 H22 + 3 H2 + 3 = 0
with the coefficients
(14)
2
2

3 = 36Yt4 f 2 + 1 + 3g32 g12 + g22 + 4g12 g22 f 2 1

+ 6Yt2 g12 4f 4 + 6f 2 1 + 18Yt2g22 2f 2 + 1 ,

3 = 9At Yt f 12Yt2 4g12 f 2 + 12Yt2 + g12 3g22 ,

3 = 72A2t f 2 Yt2 3 m2t + f 2 m2t

12Yt2 4g12 f 2 + 12Yt2 + g12 3g22
L
R
2
2

+ m22 72Yt2f 2 + g12 4f 2 1 + 9g22 + 12g32 f 2 1 ,

3 = 36At Yt f m2t + f 2 m2t .
L
45
(15)
(16)
(17)
(18)
The derivative with respect to f provides an equation quadratic in H2 and quartic in f

a3 f H22 + b3 H2 + c3 f = 0
with the coefficients

a3 = 2Yt2 18Yt2 12g32 f 2 1 + 9g22 g12 16f 2 1

+ f 2 1 4g12 g22 + 3 g12 + g22 g32 ,

b3 = At Yt 12g32 f 4 1 9g22 + g12 16f 4 1 ,

c3 = m2t 36Yt2 + 12g32 f 2 1 + 4g12 4f 2 1

L

+ m2t 36f 2 Yt2 12g32 f 2 1 + 9g22 g12 4f 2 1 .

R
(19)
(20)
(21)
(22)
A dangerous CCB minimum will have to verify the system of coupled equations Eqs. (14),
(19), with the additional constraints that H2 , f should be contained in the compact
domain where tL 2 0 [see Eqs. (12), (13)].
To determine if such a CCB vacuum induces eventually a CCB situation, we need some
additional information on the depth of the potential at a realistic EW vacuum. In the EW
direction, the tree-level potential reads:
2
(g12 + g22 ) 2
H1 H22 ,
(23)
8
where H1 denotes the neutral component of the corresponding Higgs scalar SU(2)L
doublet and is supposed to be real and positive, which can be arranged by a phase
redefinition of the fields. Without loss of generality, we may also suppose that the Higgs
mass parameters m21 , m23 are positive. The extremal equations in the EW direction read [1]:
V |EW = m21 H12 + m22 H22 2m23 H1 H2 +
m21 m22 tan2 m2Z 0

= 0,
2
tan2 1
2

m1 + m22 tan m23 1 + tan2 = 0.
(24)
(25)
For tan v2 /v1 1, where v1 , v2 are the VEVs of the EW vacuum, the minimal value
of the EW potential is given by:

2
2
m2 m21 + (m21 + m22 )2 4m43
V |EW =
(26)
.
2(g12 + g22 )
The realistic EW vacuum is furthermore subject to the phenomenological constraint v12 +
v22 = (174 GeV)2 , to reproduce correct masses for the gauge bosons Z 0 , W . Experimental
46
data also completely determine the gauge couplings g1 , g2 , g3 , and the top Yukawa
coupling Yt , the latter as a function of the physical top mass. Besides, we note that the
depth of the EW potential V |EW , Eq. (26), can be expressed with the help of the
extremal
equations (24), (25), as a function of tan and the pseudo-scalar mass mA0 = m21 + m22 ,
which have a more transparent physical meaning. Moreover, we note that in order to
incorporate leading one-loop contributions to the tree-level potential V |EW , Eq. (23), and
therefore trust the results obtained with it up to one-loop level, the parameters should be
evaluated at an appropriate renormalization scale Q QSUSY , where this SUSY scale is
an average of the typical SUSY masses at the EW vacuum [6,15]. We will come back to
this particular point in Section 4.
Comparison of the depth of the MSSM potential at the realistic EW vacuum and at the
CCB vacuum induces in addition a new non-trivial relation with the three remaining free
parameters At , mtL , mtL which enter the potential V3 , Eq. (1). As a result, a critical bound
Act,3 above which CCB occurs is identified:
CCB V3 < V |EW
At > Act,3 [mtL , mtR ; m1 , m2 , m3 , Yt , g1 , g2 , g3 ].
(27)
(28)
We anticipate again on Section 4 and stress that this comparison of the depth of the treelevel potential at both vacua, and ultimately the value of the critical bound Act,3 , also
incorporates leading one-loop contributions, provided all parameters are evaluated at the
renormalization scale Q QSUSY . To investigate this point and determine Act,3 , we need
obviously a precise knowledge on the location and geometry of the CCB vacua. Section 3
is devoted to this particular topic. For a rapid overview of the situation the interested reader
may also refer to Section 5 where we summarize some important points of this derivation
and detail the practical steps to obtain the critical bound Act,3 .
3. The CCB vacuum in the plane (H2 , tL , tR )
3.1. The algorithm to compute the CCB VEVs
To evaluate the CCB VEVs H2 and f , a numerical step is now required. A numerical
algorithm can be used for instance to solve simultaneously the two extremal equations (14),
(19). Such a method is however unable to bring any precise analytical information on the
simple geometric behaviour of the CCB extrema of the potential V3 , Eq. (1). Alternatively,
we propose a procedure which has the good numerical properties of being fast, secure and
easily implementable on a computer. Moreover, excellent analytical approximations for the
CCB VEVs (at the level of the percent), and, ultimately, for the optimal conditions on At to
avoid CCB can be obtained with it. Finally, it can be easily adapted to the extended planes
(H1 , H2 , tL , tR ) and (H1 , H2 , tL , tR , L ), first considered in [6], and this will enable us to
shed new light on vacuum stability in these directions in a fully model-independent way
[11,12].
47
This alternative procedure to evaluate the CCB VEVs may be summarized as follows:
we first insert an initial value f (0) in the extremal equation associated with H2 , Eq. (14).
This equation is then solved in H2 . A solution H2(0) , which proves to be close to the CCB
VEV H2 , is found. This solution is then inserted in the extremal equation associated with
f , (19), which is solved in f . We obtain an improved value f (1) , closer to f than f (0) .
The method is then iterated in a similar way. As a result, we obtain a set of numerical
(n)
values (H2 , f (n) )n0 which proves to converge fast toward the true CCB extremal set
( H2 , f ).
More precisely, geometrical considerations confirmed by numerical analysis show that
for a given set of free parameters At , mtL , mtR , . . . the potential V3 , Eq. (1), may have
only two non-trivial extrema with tL,R = 0: a CCB local minimum and a CCB saddlepoint. Numerical analysis thus splits into two distinct branches, each one concerning one
extremum. For simplicity, in this article we will not consider the behaviour of the CCB
saddle-point solution, which is useful essentially for metastability considerations [9,10].
Concerning the local CCB minimum, the apparent ambiguity in the implementation of the
(n)
algorithm on the correct solutions (H2 , f (n) ) to choose for each value of n 0 is easily
lifted. As will be shown in Section 3.3, when a CCB minimum develops, the extremal
equation associated with H2 , (14), has necessarily three real positive roots in H2 . The
correct solution (H2(n) )n0 to follow is always the intermediate one. There is also no real
ambiguity in the choice of (f (n) )n1 , because the extremal equation associated with f ,
(19), has always only one real positive root in f , which is our candidate. Besides, the
solutions of the extremal equation associated to H2 , (14) [which gives the set of values
(n)
(H2 )n0 ], prove to vary very slowly as a function of f . This feature tends to boost the
convergence of the procedure. Actually, starting with a clever choice for the input value
f (0) , the convergence is accelerated so that only one iteration is needed to fit the exact result
with a precision of 1% or less, providing ultimately accurate analytical approximations for
the CCB VEVs.
For completeness, let us briefly compare this method to evaluate the VEVs with the
one presented in Ref. [6]. Assuming alignment of the CCB vacuum in the SU(3)c D-flat
direction, as done in [6], i.e., f = 1, we obtain easily analytical expressions for the CCB
VEVs depending only on the free parameters At , mtL , mtR , . . . , whereas with the method
presented in [6] a numerical scan is still required: taking f (0) = f = 1, the VEV H2
is simply obtained with our method by solving analytically the cubic extremal equation
(14); the squark fields VEVs tL = tR are finally obtained by Eq. (5). In addition,
our iterative algorithm enables us to take into account, with any desired accuracy, any
deviation of the CCB vacuum from the SU(3)c D-flat direction. As noted before, in a
model-independent way, such a deviation typically occurs. This point will be investigated
more attentively in the next section, Section 3.2, by considering the extremal equation
associated with f , (19). A subsequent study of the extremal equation associated with H2 ,
(14), will also provide us with a model-independent optimal bound on At , above which a
local CCB vacuum begins to develop in the plane (H2 , tL , tR ). This point will be addressed
in Section 3.3.
48
3.2. The deviation from the SU(3)c D-flat direction

We consider now deviations of the CCB vacuum from the SU (3)c D-flat direction. In
Ref. [6], it was argued that, in a model-independent way, the CCB vacuum is located very
close to this D-flat direction. As noted before, we disagree with this statement and show
in this section that this assumption is model-dependent. Actually, such a deviation can be
quite large, in particular for large discrepancies between the soft squark masses mtL , mtR ,
and results in a substantial enhancement of the necessary and sufficient condition to avoid
CCB, At Act,3 , Eq. (27). In fact, the critical bound Act,3 can be shown to become more
restrictive and this feature is essential, on a phenomenological ground, when it comes
to relate CCB conditions with the requirement of avoiding a tachyonic lightest stop. As
is well-known, a too large trilinear soft term At can increase this danger, but also any
discrepancy mtL = mtR . The latter effect can be compensated only by taking into account
the deviation of the CCB vacuum from the SU(3)c D-flat direction.
The parameter controlling the deviation of the CCB vacuum from the SU(3)c D-flat
direction is the VEV f . In the framework of our algorithm to compute the CCB VEVs,
an educated guess for the initial value f (0) should incorporate information on the extremal
equation associated with f , Eq. (19). Numerical analysis shows that, to an excellent
accuracy, f is related to H2 by the relation

2
mt + Yt2 H2 2
.
f f ( H2 ) 2L
(29)
mt + Yt2 H2 2
R
If we neglect gauge contributions, f(H2 ) is actually the exact solution to Eq. (19).
Therefore, this numerical observation simply reflects the fact that the deviation of the
CCB vacuum from the SU(3)c D-flat direction is nearly independent of the D-terms
contributions in the potential V3 , Eq. (1).
To go further, we need to approximate H2 . We note first that f(H2 ) is a slowly varying
function of H2 , so that this approximation does not need to be very accurate. Neglecting in
the potential V3 , Eq. (1), the contributions of the gauge terms and the Higgs mass term m22 ,
and writing the potential as a function of B3 , Eq. (6), we find V3 B32 /(16Yt2f 2 ). This
simple expression shows that a rough estimate of the VEV H2 is given by the minimal
value taken by B3 , Eq. (6):
H2
f
At
.
Yt (1 + f 2 )
(30)
In fact, the exact VEV H2 is numerically typically found to be lower than this
approximate value, but the latter already contains useful enough information for our
purpose. Taking this value and solving f = f( H2 ), we obtain in turn the excellent
approximation to f

2
At + 2m2t m2t
(0)
L
R
.
f f3 2
(31)
At + 2m2t m2t
R
49
(0)
Fig. 1. f , f3 versus rk (mtL /mtR )k , for k = 1, mtR = 200 GeV (upper curves) and k = 1,
mtL = 200 GeV (lower curves). We take At = 1400 GeV, m2 = 118 GeV, Yt = 1.005, g1 = 0.356,
g2 = 0.649, g3 = 1.14.
(0)
This approximate value f3 is equal to 1 (alignment in the SU (3)c D-flat direction)

for m2t = m2t . This correctly reproduces the expected behaviour for f : when the
L
mass relation m2t = m2t holds, the potential V3 , Eq. (1), has an underlying approximate
L
R
symmetry tL tR broken by tiny O(g12 , g22 ) contributions, so that any non-trivial
extremum must be nearly aligned in the SU(3)c D-flat direction. In the large At regime,
(0)
we have also f3 1, reproducing again the expected behaviour for f . Quite similarly
to the small Yukawa coupling regime, in this limit, the VEVs of the CCB vacuum become
very large. The vacuum then moves towards the SU(3)c SU(2)L U (1)Y D-flat direction,
any splitting between the soft squark masses becoming inessential.
(0)
In Fig. 1, we illustrate the evolution of the exact VEV f and the approximation f3 ,
k
Eq. (31), as a function of the ratio of the soft squark masses rk (mtL /mtR ) for two cases
k = 1. The exact VEV f has been computed with the recursive algorithm presented
(0)
above [see also Section 5 for a practical summary], taking for initial value f (0) = f3 ,
Eq. (31). In both cases, we have taken At = 1400 GeV. This implies in particular that
the expression f3(0) , Eq. (31), is defined up to rk 7.14. However, this approximation is
appropriate only if a CCB vacuum exists. The sufficient bound given by Eq. (9) already
shows that a CCB vacuum can develop only for rk 6. Once optimized, the sufficient
bound to avoid CCB restricts even more the allowed range to rk 5.3 [see, e.g., Eq. (36)
in Section 3.3]. In Fig. 1, the evolution of the VEV is stopped at the boundary value rk 5,
because the CCB vacuum becomes dangerous and deeper than the EW vacuum only below
this value.
(0)
The first prominent feature of this illustration is that f3 is an excellent approximation:
f3(0) fits f with a precision of order 5%, or even better in the vicinity of rk 1. Moreover,
50
we note that for a large splitting of the soft squark masses, the deviation of the CCB global
minimum from the SU(3)c D-flat direction can be quite large, in particular in the vicinity
of the critical bound Act,3 , i.e., for rk 5. It is smaller for rk 1, where the CCB vacuum
is nearly aligned in the SU(3)c D-flat direction.
Finally, we may derive refined bounds on the deviation of the CCB vacuum from the
(0)
SU(3)c D-flat direction by combining the sufficient bound At , Eq. (9), with the accurate
(0)
(0)
approximation f3 , Eq. (31). Requiring At At mtL (1 + r1 ), we obtain:

mtL
r1 (3r1 + 2)
(0)
1,
1 f f3
For r1
(32)
.
mtR
2r1 + 3
For r1 1 these inequalities are reversed.
In an mSUGRA scenario [3,19], we fall typically in the regime r1 1, with furthermore
r1 perturbatively close to 1. Eq. (32) then implies 1 f 1 + 35 (r1 1), showing that
the CCB vacuum is indeed located in the vicinity of the SU(3)c D-flat direction. Such
a feature was built-in through the procedure proposed to evaluate the CCB conditions
in Ref. [6], quite consistently with the mSUGRA numerical illustration presented in this
article. However, it is important to stress that this assumption is model-dependent, and may
be badly violated in other circumstances near the critical value Act,3 , in particular in scenarii
incorporating non-universalities of the soft squark masses where the splitting parameter r1
can be rather large [4].
3.3. The optimal sufficient bound on At to avoid CCB
We consider now more attentively the extremal equation associated with H2 , Eq. (14).
This complementary equation will in fact enable us to improve the sufficient bound A(0)
t ,
Eq. (9), to avoid a dangerous CCB vacuum in the plane (H2 , tL , tR ). From a geometrical
point of view, it is reasonable to define the optimal sufficient CCB bound on At to be the
largest value below which a local CCB minimum, not necessarily global, cannot develop.
in the following, is also the critical value above
Equivalently, this bound, denoted Asuf
t
which a local CCB vacuum begins to develop in the plane (H2 , tL , tR ).
The determination of the optimal sufficient bound Asuf
t simply requires some additional
pieces of information on the extremal equation associated with H2 , Eq. (14), and on the
geometry of potential V3 , Eq. (1). In order not to surcharge the text, we will not enter here in
the details of this derivation, but rather refer the reader to the Appendix A. To summarize,
on the technical side, the essential result we obtain is that if the extremal equation, Eq. (14),
considered as a cubic polynomial in H2 , has only one real root in H2 for any given value
of f , then necessarily no local CCB minimum in the plane (H2 , tL , tR ) can develop. More
intuitively, this result merely reflects the fact that if a local CCB vacuum develops with
non trivial VEVs ( H2 , f ), then on any path connecting the local extremum at the
origin of the fields to this CCB vacuum, there will be necessarily a saddle on the top
of the barrier separating them, with tR , tL = 0. For f = f , such a point will in turn
necessarily correspond to a second real solution in H2 for the extremal equation, Eq. (14),
in contradiction with the initial assumption.
51
Considering the extremal equation (14) as a cubic polynomial in H2 , a necessary and

sufficient condition to have only one real root is
3

C3 233 93 3 3 + 2732 3 + 4 32 + 33 3 0.
(33)
The next step to evaluate the optimal sufficient bound Asuf
t is to consider this complicated
inequality in the direction of a possible CCB minimum f = f . Taking instead the
(0)
approximate value f f3 , given by Eq. (31), and taking also values for all the parameters
except the trilinear soft term At , the equation C3 = 0 may be solved numerically as a
(1)
(1)
function of At . 2 Let us denote At the largest solution of this equation. For At At , we
find numerically that we always have C3 0, showing that there can be no CCB vacuum in
this case. Moreover, numerical investigation also shows that A(1)
t has typically the desired
(0)
property of being larger than At , Eq. (9), and, therefore, improves this bound. There is
only one exception to this statement, which occurs for m22 0 and mtL , mtR mt . As will
be explained in Section 3.4, this regime actually corresponds to a rather particular situation
where no dangerous CCB vacuum deeper that the EW vacuum may develop, unless the
EW vacuum is unstable.
Taking into account this observation, numerical investigation finally shows that for At
(0)
(1)
Max[At , At ], a local CCB vacuum begins to develop in the plane (H2 , tL , tR ). Hence,
this critical value fulfills the properties required to be identified with the optimal sufficient
bound Asuf
t . In conclusion, we may write without loss of generality:
(0) (1)
At Asuf
(34)
No local CCB vacuum,
t Max At , At
where A(0)
is given by Eq. (9) and A(1)
is obtained by solving C3 = 0, Eq. (33), as
t
t
mentioned above. It is important to stress here that this optimal sufficient bound, obtained
with exact analytical expressions, incorporates all possible deviations of the CCB local
vacuum from the D-flat directions, including the SU(3)c D-flat one.
As At increases above Asuf
t , CCB local vacuum soon becomes global and deeper than the
EW vacuum. Obviously, the critical bound Act,3 , Eq. (27), is necessarily larger than Asuf
t :
Act,3 Asuf
t .
(35)
Moreover, we expect that the critical bound Act,3 should be perturbatively close to the opc
suf
timal sufficient bound Asuf
t , i.e., At,3 At , because the EW potential is not very deep,
4
2
2
V |EW mZ /(g1 + g2 ). Indeed, as will be illustrated in Section 5, the critical bound
Act,3 is typically located in a range of 5% or less above Asuf
t . This interesting feature will
c
considerably simplify the exact determination of At,3 , which will be simply obtained by
scanning a small interval in At above Asuf
t . We note finally that in the interesting phenomenological regime mtL , mtR 300 GeV, a simple empirical approximation of Asuf
t may be
(1)
suf
obtained numerically. We find on one hand At = At , with furthermore:
(1)
ap
Asuf
t = At At mtL + mtR + |m2 |.
(36)
2 Comparing with a more accurate value for f , we found that the maximal discrepancy between the results
obtained is negligible, less that 1 GeV.
52
This approximation exhibits in which amount the sufficient bound A(0)

t is improved in this
(0)
|m
|.
regime. The difference is of order |m2 |: A(1)
2
t
t
Let us come back briefly now to the implementation of the procedure to compute
the CCB VEVs. We have shown that for At Asuf
t , Eq. (34), no local CCB vacuum
may develop in the plane (H2 , tL , tR ). As noted before, in the dangerous complementary
regime, At Asuf
t , we need to evaluate the VEVs of the CCB local vacuum in order to
compare the depth of the CCB potential and the EW potential and find the necessary
(0)
and sufficient bound Act,3 , Eq. (27), to avoid CCB. For f = f3 , Eq. (31), the extremal
equation associated with H2 , Eq. (14), has three real roots, which also prove to be positive
[see Appendix A]. The intermediate root, denoted H2(0) in Section 3.1, proves to be an
excellent approximation of the VEV H2 [at a level of 1%]. 3 The analytic expression
(0)
of H2 is complicated and not particularly telling, therefore we refrain from giving it here.
This shows that, to an excellent approximation, we can obtain explicit analytic expressions
for all the CCB VEVs: ( H2 , f ) are approximated by (H2(0) , f3(0) ), where f3(0) enables
us to take into account the deviation of the CCB vacuum from the SU(3)c D-flat direction,
and the squark VEVs tL/R are subsequently obtained by Eqs. (4), (5). This way, we can
obtain in turn an accurate analytical expression for the CCB potential V3 , Eq. (1), at the
CCB vacuum. Comparison with the EW potential V |EW , Eq. (26), ultimately provides
an excellent approximation of the critical CCB bound Act,3 , Eq. (27). The accuracy of this
approximation can be improved at will by iterating the procedure to compute the CCB
VEVs, as depicted in Section 3.1. We note however that the impact on Act,3 is negligible,
O(1 GeV).
3.4. The instability condition of the EW potential
Besides the contribution of the trilinear soft term At , another negative contribution in
the potential V3 , Eq. (1), appears at the EW scale when the Higgs parameter m22 becomes
negative. At the tree-level, the sign of m22 is related in a simple way to tan by the extremal

equation (24) in the EW direction: for tan 1 + 2m2A0 /m2Z 0 , m22 is negative, whereas
it is positive in the complementary low tan regime. Numerical investigation shows that
we have only two distinct patterns for the number of local extrema of V3 , Eq. (1). They are
distinguished by the sign of m22 and, consequently, the magnitude of tan .
m22 0:
The potential V3 , Eq. (1), has one trivial local minimum, namely the origin of the fields,
and possibly a pair of non-trivial local extrema with tL,R = 0: the would-be global CCB
vacuum and a CCB saddle-point sitting on top of the barrier separating it from the origin
of the fields. In this case, the optimal sufficient bound Asuf
t , Eq. (34), always proves to
(1)
be equal to At . Besides, for soft squark masses large enough, i.e., mtL , mtR |m2 |, the
traditional CCB bound in the D-flat direction AD
t,3 , Eq. (2), typically provides an upper
3 For completeness, we note that typically the lowest solution will correspond to a directional CCB saddlepoint, whereas the largest is spurious, giving tL 2 0.
53
bound for the critical bound Act,3 , Eq. (27), so that we may write:
(1)
c
D
Asuf
t = At At,3 At,3 .
(37)
We note however that this upper bound is not very indicative of the critical value Act,3
for large values of the soft masses, mtL , mtR |m2 |, as will be illustrated in Section 5.
Actually, in this regime, the relation Eq. (3) which is the signature of an alignment in the
D-flat direction is badly violated, implying a large deviation of the CCB vacuum from the
D-flat direction.
m22 0:
Besides the origin of the fields and possibly a pair of CCB extrema (a local minimum
and a saddle-point), the potential V3 , Eq. (1), has another non-trivial extremum with
VEVs H2 2EW = 4m22 /(g12 + g22 ), tR,L EW = 0, giving V3 EW = 2m42 /(g12 + g22 ). The
origin of the fields is now unstable and the potential automatically bends down in the
direction of this non-CCB extremum. We note also that, for large tan, the EW vacuum
tends towards it as the inverse power of tan : (v1 = (174 GeV)/ 1 + tan2 , v2 )
(0, H2 EW ). [Accordingly, the negativity of m22 appears as a mark in the plane (H2 , tL , tR )
of the well-known instability condition at the origin of the fields m21 m22 m43 0, which is
the signal of EW symmetry breaking [1].]
Obviously, if this additional extremum is a saddle-point of the potential V3 , Eq. (1), then
a deeper CCB minimum is necessarily present in the plane (H2 , tL , tR ). The squared mass
matrix evaluated at the non-CCB extremum reads
g12 +g22 2
0
0
2 H2

(3g 2 g 2 )
At Yt H2
0
m2t + Yt2 H22 1 2 2 1
M2 |EW =
12Yt
L

2
g
0
At Yt H2
m2t + Yt2 H22 1 12
R
3Yt
(38)
with H2 = H2 EW .
Stability of the non-CCB vacuum is equivalent to the positivity of all the squared mass
eigenvalues of M2 |EW , Eq. (38). It is not automatic and needs
At Ainst
t ,
(39)
where

2 2
g2
(3g22 g12 )
g12 + g22 mtL mtR
1 12 + m2t 1
L
R
3Yt
12Yt2
4m22
Yt2

(3g22 g12 )
g12
4m2
1
.
2 2 2 Yt2 1
g1 + g2
12Yt2
3Yt2
2
2
(Ainst
t ) mt
(40)
Let us remark that, for tan +, the lower 2 2 matrix of M2 |EW , Eq. (38), is
simply equal to the tree-level physical squared stop mass matrix [1], so that the instability
condition, Eq. (39), is a mere rephrasing of the physical requirement of avoiding a
tachyonic lightest stop, expressed as a function of At . This statement is also valid to a
54
good accuracy when the stop mixing parameter A t = At + / tan is well approximated
by the trilinear soft term At , i.e., for || |At | tan .
To simplify the discussion, in the following we will essentially identify this non-CCB
extremum with the EW vacuum, implying in particular that the potential at both vacua are
equal, i.e., V |EW V3 EW . This assumption, accurate for tan large enough, enables us
to write the following relation on the CCB bounds
c
inst
Asuf
t At,3 At .
(41)
The first relation was actually obtained in the last section, see Eq. (35), whereas the second
means that if the non-CCB extremum is unstable, then a dangerous CCB vacuum, deeper
than the EW vacuum, 4 has developed in the plane (H2 , tL , tR ).
For m22 0, the relation (41) provides the most general upper and lower bounds on the
critical CCB bound Act,3 , Eq. (27). We note however that in the interesting regime of large
squark soft masses, i.e., mtL , mtR mt , the upper bound given by Ainst
is typically largely
t
D
improved by the traditional bound in the D-flat direction At,3 , Eq. (2). However, quite
similarly to the case m22 0, the latter bound AD
t,3 is itself typically very large compared to
the critical CCB bound Act,3 , due to a large deviation of the CCB vacuum from the D-flat
directions. A better indication of the critical CCB bound Act,3 , Eq. (27), is always given by
the optimal sufficient bound Asuf
t , Eq. (34).
Finally, let us consider more attentively the behaviour of the potential in the limit At
Ainst
t . This will enlighten the importance of taking into account any deviation of the CCB
vacuum from the SU(3)c D-flat direction in order to obtain a consistent critical CCB bound
Act,3 , Eq. (27), which encompasses the possibility of avoiding a tachyonic stop mass. Two
interesting different modes with particular geometrical features of the potential can be
considered:
(i) The CCB vacuum is located away from the non-CCB extremum. This possibility
in fact corresponds either to the case mtL mtR m2t or mtL mtR m2t , where mt is the
top quark mass. In the first case, the CCB vacuum proves to be closer to the origin of
the fields than the non-CCB extremum, whereas in the latter this hierarchy is reversed.
(1)
In both cases, the optimal sufficient bound Asuf
t , Eq. (34), is always given by At . In the
inst
limit At At , the CCB saddle-point located on top of the barrier separating the CCB
vacuum and the non-CCB extremum tends towards the non-CCB extremum and the barrier
separating both vacua eventually disappears.
(ii) The CCB vacuum interferes with the non-CCB vacuum. For At Ainst
t , this mode
corresponds to a degenerate situation where the CCB local vacuum and the CCB saddlepoint overlap and tend towards the non-CCB vacuum. This possibility appears clearly by
(0)
comparing the instability bound Ainst
t , Eq. (39), with the sufficient bound At , Eq. (9). We
4 The relation Ac Ainst is still accurate if we relax our simplifying assumption V |
EW V3 EW , because
t
t,3
the potential V3 deepens rapidly with increasing At .
55
have

2
(0)2
(3g22 g12 )
g12
1
2
2 2
(Ainst
)
A
=
m
m
H
0 (42)
1
Y
t
tL tR
t
t 2
12Yt2
3Yt2
Yt2 H22
with H2 = H2 EW . Combining the last equation with Eq. (41), we obtain:

g12
(3g22 g12 )
(0)
suf
mtL mtR = 1
1
m2t Act,3 = Ainst

t = At [= At ],
12Yt2
3Yt2
(43)
where the EW and the non-CCB vacua have been identified to write mt = Yt H2 EW .
The equalities on the right hand side of the equivalence equation (43) signal that for
this particular values of the soft squark masses, we are at the center of a critical regime
where the CCB vacuum interferes with EW vacuum. This critical regime actually extends
to a small range in mtL , mtR around this center, and is more generally characterized by
= Act,3 , meaning that no dangerous CCB vacuum, deeper than the EW
the relation Ainst
t
vacuum, may develop unless the EW vacuum is unstable. In this region, there is also
typically no room for a CCB vacuum to develop, not even a local one. This occurs already,
(0)
= Asuf
e.g., at the center of the critical regime, where we have Ainst
t
t = [At ]. As will be
illustrated in Section 5 [see Fig. 4], this critical regime includes a small domain around this
(1)
(0)
(0)
center where the typical hierarchy At At is slightly violated, giving Asuf
t = At , and
(1)
which is itself bordered by a domain where this hierarchy is respected, giving Asuf
t = At .
We come now more precisely to the relation between the critical CCB bound Act,3 , Eq. (27),
and the requirement of avoiding a tachyonic lightest stop. Obviously, this relation is crucial
in the interference regime mtL mtR m2t , corresponding to the case (ii), where we have
= Act,3 . With the help of the extremal equations, it is a straightforward exercise to
Ainst
t
c
show that for At Ainst
t [= At,3 ], the VEVs of the CCB extremum verify ( H2 , tL,R )
( H2 EW , 0), with furthermore:

12(m2t + Yt2 H2 2EW ) + (g12 3g22 ) H2 2EW
L
f
(44)
.
12(m2t + Yt2 H2 2EW ) 4g12 H2 2EW
R
This particular direction is, in fact, connected to the direction of the lightest stop eigenstate.
Let us denote (t 1 , t 2 ) the stop-like eigenstates of the 2 2 lower matrix in M2 |EW ,
Eq. (38), and the mixing angle of the rotation matrix R relating these eigenstates to
the VEVs ( tL , tR ):

t1
tL
cos
sin
(45)
=R
with R
.
tR
sin cos
t2
As noted before, if we assume that tan is large enough and || |At | tan , we may
safely identify this matrix with the physical squared stop matrix, and (t 1 , t 2 , ) with the
stop eigenstates and mixing angle (t1 , t2 , ) [1].
2
By definition, for At Ainst
t , the matrix M |EW , Eq. (38), has one zero eigenvalue and
the wall separating the CCB extremum and the non-CCB extremum lowers and eventually
disappears in the direction of the corresponding eigenstate t 1 . In the basis (tL , tR ), the
56
R
components of this eigenstate read t 1 = (t L
1 = cos , t 1 = sin ) and prove to verify
tan
tR
1
tL
1
= f ,
(46)
where f is given by the limiting value in Eq. (44). This shows on one hand that, in this
critical regime, the stop mixing angle is related in a simple way to the deviation of the
CCB vacuum from the SU(3)c D-flat direction and, on the other, that taking into account
such a deviation of the CCB vacuum to evaluate the critical CCB bound Act,3 , Eq. (27), is
crucial to avoid a tachyonic lightest stop.
4. Radiative corrections
In this section, we discuss the renormalization scale at which the tree-level necessary
and sufficient condition to avoid CCB, At Act,3 , Eq. (27), should be evaluated in order
to incorporate leading one-loop corrections. As is well-known, on a general ground,
the complete, all order effective potential V () is a renormalization group invariant.
However, this property is not shared by the tree-level approximation V (0) which typically
depends strongly on the renormalization scale Q at which it is computed [6,15]. A kind
of renormalization group-improved version of the tree-level potential which would
incorporate a resummation of all leading logarithmic contributions would certainly be more
reliable. However, one faces here the tricky problem of dealing with many mass scales. 5
A better approximation to V (), more stable with respect to the scale Q, is in fact given
by the one-level effective potential (MS scheme) [6,15]

(1)2si (2si + 1)
Mi2 () 3
4
V (1) () = V (0)() +
(47)
M
()
Log
,
i
64 2
Q2
2
i
Mi2 ()
denotes the tree-level squared mass of the eigenstate labeled i, of spin si , in

where
the scalar field direction . The scale Q enters explicitly in the one-loop correction, but
also implicitly in the running of the mass and coupling parameters.
Obviously, in the field direction (H2 , tL , tR ) studied in this paper, such a one-loop
correction will introduce very complicated field contributions which will modify the simple
tree-level geometrical picture presented here. However, we may still have locally a good
indication of the impact of these radiative corrections with the help of our tree-level
investigation. As is also well-known, around some scale Q0 which depends on the field
direction considered, the predictions obtained with the tree-level potential V (0) and the
one-loop level potential V (1) approximately coincide [6,15]. This numerical observation
was in fact intensively used, in particular in the context of CCB studies [610], precisely in
order to use the relative simplicity of the tree-level potential. This field-dependent scale Q0
is typically of the order of the most significant mass present in the field region investigated.
This roughly means that we reduce the multi-scale problem to a one-scale one, the most
5 Some attempts have be made in this direction, see [20].
57
significant mass meaning a kind of average of the field-dependent masses which provide
the leading one-loop contributions in the direction of interest [6,15].
At the EW vacuum, it has been shown that the appropriate renormalization scale QSUSY
where the one-loop corrections to the tree-level potential V |EW , Eq. (23), can be safely
neglected is an average of the typical SUSY masses [6,15]. For instance, for large MSUSY
mtL mtR mt , the tree-level potential receives important radiative corrections coming
from loops of top and stop fields. In this case, QSUSY is expected to be an average of the
top and stop masses, giving QSUSY MSUSY , whereas for low MSUSY mt , this scale
is somewhat underestimated and should be raised to a more typical SUSY mass [6,15].
In this light, we see that we may trust the results obtained with the tree-level potential
V |EW , Eq. (23), in particular the EW VEVs (v1 , v2 ) given by Eqs. (24), (25) and the depth
of the EW potential V |EW , Eq. (26), provided all parameters entering this potential are
evaluated at the appropriate scale Q QSUSY .
What is now the appropriate scale QCCB where the results obtained with the treelevel potential V3 , Eq. (1), incorporate leading one-loop corrections? At the CCB vacuum,
such corrections are expected to be induced by loops involving masses in the scalar field
direction (H2 , tL , tR ), in particular for mtL mtR mt for which these contributions are
enhanced. Accordingly,
we estimate QCCB to be a certain average of these masses, more
precisely QCCB
Tr M2 |CCB
,
3
where:

1 2 V3 2 V3 2 V3
Tr M |CCB =
+
+
2 H22
t 2L
tR2 CCB
2

t 2
= m2t + m2t + Yt2 2H22 + t 2L f 2 + 1 + At Yt f L
L
R
H2
2
2

g
g
+ 1 9H22 + 44f 2 1 t 2L + 2 H22 + 3t 2L
36
4

2g32 2
+
(48)
t 1 + f 2 .
3 L
All fields should be evaluated at the CCB vacuum. To derive the last expression, we have
used the extremal equation V3 /H2 = 0 to replace the Higgs mass parameter m2 . The
VEV tL may also be replaced with the help of the extremal equations (5), (6), giving
a complicated expression for QCCB which depends only on the soft terms At , mtL , mtR ,
the gauge and Yukawa couplings and the CCB VEVs H2 , f . For simplicity, let us
take MSUSY = mtL = mtR , which gives f = 1. Taking furthermore Yt , g3 1, neglecting
other gauge couplings and (over-)estimating the VEV H2 At /2 [see Eq. (30)], we find
Q2CCB
2
11A2t 20MSUSY
.
18
(49)
This scale is meaningful only when a CCB vacuum develops, that is for At Asuf
t [see
(0)
Eq. (34)]. For illustration, we estimate roughly this lower bound with At , Eq. (9). Taking
At 2 MSUSY , we obtain QCCB 1.33 MSUSY . Let us stress here that a refined evaluation of QCCB , with realistic values for the gauge couplings, the CCB VEV H2 and the
58
optimal sufficient bound Asuf

t , would give in fact a value for QCCB closer to MSUSY . This
simple illustration however already provides a clear indication that QCCB is typically of
order MSUSY .
For MSUSY mt and At Asuf
t , we conclude therefore that we have QCCB QSUSY .
Obviously, a similar conclusion is expected in the complementary regime MSUSY mt : in
this case, the CCB and the EW vacua prove to be close, implying a mass spectrum of the
same order at each vacuum. We note also that this estimation of QCCB is in full agreement
with the one obtained in Ref. [6] in the extended plane (H1 , H2 , tL , tR ). In this article,
the scale QCCB was estimated to be Max[QSUSY , g3 At /4Yt , At /4], which reduces for
Yt , g3 1 and At Asuf
t 2MSUSY to QCCB QSUSY .
Two important conclusions can be deducted from this result. On one hand, we see that
the optimal sufficient bound Asuf
t , Eq. (34), should be evaluated at QCCB QSUSY , in
order to minimize the one-loop radiative corrections to the tree-level potential V3 , Eq. (1).
More importantly, we see that, at this common scale QCCB QSUSY , it is also meaningful
to compare the tree-level depth of the potential at the EW vacuum, i.e., V |EW , Eq. (26),
and at the CCB vacuum, in order to determine the necessary and sufficient condition At
Act,3 , Eq. (27), to avoid CCB in the plane (H2 , tL , tR ). This point is a mere consequence
of the fact that the potential at a realistic EW vacuum is not very deep, as already noted
in Section 3.3 [see Eq. (35)], therefore giving Act,3 Asuf
t . To summarize, we expect our
tree-level refined CCB bounds to be robust under inclusion of leading one-loop corrections
to the potential, provided they are evaluated at Q QSUSY . Accordingly, stability of the
EW vacuum in the plane (H2 , tL , tR ) should be tested in model-dependent scenarii [3,4,6]
at this scale.
5. Practical guide to evaluate the CCB conditions

Let us now collect and summarize the main results we have found. As mentioned in
Section 2.3, the evaluation of the critical bound Act,3 , Eq. (27), above which there is CCB in
the plane (H2 , tL , tR ) requires the precise determination of the CCB VEVs and comparison
of the potential V3 , Eq. (1), at the CCB vacuum with the value of the potential at the EW
vacuum V |EW , Eq. (26). This comparison is meaningful and incorporates leading oneloop corrections, provided all parameters are evaluated at the appropriate renormalization
scale Q QSUSY , where QSUSY is an average of the typical SUSY masses at a realistic
EW vacuum. This assumption will be implicitly made in the following. Accordingly, the
main practical steps to evaluate Act,3 are:
Evaluation of the depth of the EW potential: take a realistic set of values for
g1 , g2 , g3 consistent with experimental data; choose in addition values for tan and the
pseudo-scalar mass m2A0 = m21 + m22 . The top mass mt = Yt v sin , with v = 174 GeV,
determines the value of the top Yukawa coupling Yt . Finally, the extremal equations in the
EW direction Eqs. (24), (25) determine the Higgs mass parameters m1 , m2 , m3 and the
depth of the potential at the EW vacuum V |EW , Eq. (26).
59
Evaluation of the CCB optimal sufficient bound: choose a set of values for
the soft mass parameters mtL , mtR and evaluate the optimal sufficient bound Asuf
t =
(0)
(1)
(0)
Max[At , At ], Eq. (34). This requires the comparison of the quantities At given by
Eq. (9), and A(1)
given by the largest solution in At of the equation C3 = 0, Eq. (33).
t
(1)
To evaluate At , the parameter f of the departure of the CCB vacuum from the SU(3)C
(0)
D-flat direction should be taken at the excellent approximated value f3 , Eq. (31) [see
suf
Section 3.2]. The value obtained At is the optimal sufficient bound to avoid CCB. This
means that for At = Asuf
t a CCB local vacuum, not necessarily global, begins to develop
in the plane (H2 , tL , tR ), and soon becomes global as At increases. This bound therefore
considerably simplifies the determination of the necessary and sufficient bound Act,3 to
avoid CCB in this plane.

(1)
For 1 tan 1 + 2m2A0 /m2Z 0 , or equivalently m22 0, we always have Asuf
t = At
[see Section 3.4]. In the complementary regime, we have m22 0, and an additional nonCCB vacuum develops in the plane (H2 , tL , tR ). For large enough tan , it essentially
coincides with the EW vacuum which is located in the vicinity of this plane [see
Section 3.4]. As a result, a new computable instability bound Ainst
t , Eqs. (39), (40), appears.
This bound merely reflects the physical requirement of avoiding a non-tachyonic lightest
stop, for large tan . Besides, an inversion of the typical hierarchy between the sufficient
(1)
(0)
suf
bounds A(0)
t At may occur, implying At = At . This inversion however takes place
only in the critical region mtL mtR mt where the CCB vacuum interferes with the nonCCB vacuum aforementioned. In this interference regime, the instability bound Ainst
is
t
c
suf
inst
quite restrictive and the relation At At,3 At , Eq. (41), is saturated on both sides,
meaning that no dangerous CCB vacuum may develop unless the EW vacuum is unstable.
Whatever the value of tan , for large enough soft masses mtL , mtR 300 GeV, a good
ap
approximation to the bound Asuf
t is given by At , Eq. (36) [see Section 3.3].
Let us remark that the parameters involved in these first two steps, basically (mA0 , tan ,
mtL , mtR ), are typical of phenomenological model-independent Higgs studies, the benchmark scenario MSUSY = mtL = mtR being often considered [13,14]. Once such a set of
values is chosen, CCB considerations induce an additional constraint on the allowed values for the trilinear soft term At .
Evaluation of the CCB critical bound Act,3 : the determination of the CCB VEVs
and comparison of the depth of the CCB potential V3 , Eq. (1), and V |EW , Eq. (26), is
needed [see Section 2.3]. This step requires a numerical scan of the region At Asuf
t ,
which is not time consuming, because typically the critical bound Act,3 proves to be just
slightly above the optimal sufficient bound Asuf
t previously determined. The computation
of the CCB VEVs may be achieved with the help of the algorithm presented in Section 3.1.
The main steps are the following:
(0)
Solve the extremal equation (14) in H2 with the initial input f = f3 given by
Eq. (31). For At Asuf
t , this cubic equation in H2 has necessarily three real positive
roots [see Section 3.3]. The intermediate solution, denoted H2(0) , which can be given an
60
explicit analytical expression, always proves to be very close to the CCB VEV H2 (the
discrepancy is less than 1%).
(0)
Solve the extremal equation (19) in f with H2 = H2 . This equation has only one
consistent (i.e., real and positive, see Section 2.2) solution f (1) , which is even closer to the
CCB VEV f than f3(0) .
The algorithm may be iterated in the same way without ambiguity. The set of values
(n)
(H2 , f (n) )n0 proves to converge very fast towards ( H2 , f ).
Once the CCB VEVs H2 , f are computed , tL is obtained by Eq. (5) and we have
tR = f tL , which completes the determination of the location of the CCB vacuum.
The final step is the comparison of the potential V3 , Eq. (1), at this dangerous vacuum with
c
V |EW . A scan for At Asuf
t then provides the critical bound At,3 .
For completeness, we have summarized the full algorithm to compute the critical bound
Act,3 . It is however important to stress that, in practice, this evaluation is considerably
simpler and more rapid. The values (H2(0), f3(0) ) obtained with the first iteration of
our algorithm already provide excellent analytic approximations of ( H2 , f ). Further
iterations will result in unimportant effects. In particular, the impact on the critical CCB
bound Act,3 is extremely tiny, O(1 GeV). Thus, to an excellent accuracy, explicit analytic
expressions for all the VEVs of the CCB vacuum and of the potential V3 , Eq. (1), at this
vacuum can be given, and the determination of the critical CCB bound Act,3 essentially
reduces to the comparison of the CCB potential V3 , Eq. (1), at the CCB vacuum with the
EW potential V |EW , Eq. (26).
6. Numerical illustration of the CCB bounds

We turn now to the numerical illustration of the various CCB bounds obtained in the
plane (H2 , tL , tR ), first for low tan [we take tan = 3], and then for large tan [we
consider the limiting case tan = +], where the additional negative contribution of the
Higgs mass parameter m22 induces new features of the potential, as shown in Section 3.4. In
order to incorporate one-loop leading corrections, the CCB bounds are implicitly supposed
to be evaluated at the SUSY scale Q QSUSY .
The low tan regime
In Figs. 2, 3, the behaviour of the CCB bounds on At , i.e., the critical bound Act,3 ,
(1)
Eq. (27), the optimal sufficient bound Asuf
in this
t , Eq. (9), which is always equal At
ap
regime, its approximation At , Eq. (36), and finally the traditional bound in the D-flat
direction AD
t,3 , Eq. (2), is illustrated as a function of the soft squark masses mtL , mtR . The
set of values chosen is consistent with a correct tree-level EW symmetry breaking with
tan = 3, mA0 = 520 GeV and a top quark mass mt = 175 GeV. As can be seen, in both
(1)
illustrations, the hierarchy Act,3 Asuf
t = At , Eq. (35), is verified, as expected.
In Fig. 2, the various CCB bounds are plotted as a function of the ratio r1 = mtL /mtR ,
taking mtR = 200 GeV. Therefore, except for r1 = 1, the CCB vacuum always deviates
from the SU(3)c D-flat direction. Comparing the critical bound Act,3 and A(1)
t , we see that
61
Fig. 2. CCB bounds versus r1 = mtL /mtR . We take mtR = 200 GeV, m1 = 400 GeV, m3 = 228 GeV
and m2 , Yt , g1 , g2 , g3 as in Fig. 1. This gives mt = 175 GeV.
(1)
they follow each other closely for all values of r1 , with Act,3 1.041.10 At , the lowest
values being reached for large r1 and the largest for r1 0. For 1 r1 5, the sufficient
ap
(1)
bound At is approximately linear in r1 and the accuracy of the approximation At is
rather good, better than 5%. Although this linear behaviour breaks down for low r1 1,
ap
At still provides a good thumbrule to evaluate A(1)
t (within 58%). Note that we can have
ap
ap
ap
(1)

A
or
A

A
,
showing
that
A
either A(1)
t
t
t is just an approximation and should be
t
t
handled with care.
(1)
c
For r1 1, we have AD
t,3 At At,3 . In this regime, the soft squark masses are
of the same order, implying that the CCB vacuum is nearly aligned in the SU(3)c Dflat direction, as noted in Section 3.4. The CCB vacuum is also located in the vicinity
of the SU(2)L U (1)Y D-flat direction, because the common value of the soft squark
masses mtL mtR 200 GeV is not so large compared to the Higgs mass parameter m2 =
118 GeV, so that the relation Eq. (3) is approximately verified [see Section 2.1]. For large
r1 5 and At Act,3 , the CCB vacuum is located far away from the SU(3)c D-flat direction
(as well as the SU(2)L U (1)Y D-flat direction). This large departure clearly appears by
comparing the traditional CCB bound in the D-flat direction, AD
t,3 1778 GeV, and the
critical bound Act,3 1383.5 GeV. The latter is about 30% below the traditional bound
AD
t,3 ! This is a typical feature of this D-flat direction condition: for large soft squark masses,
it is far from being optimal and not very indicative of the critical bound Act,3 . A better
ap
estimate is given by the optimal sufficient bound A(1)

t or even its approximation At .
Finally, we note that if we had taken r1 = mtR /mtL and mtL = 200 GeV, the curves
obtained would overlap the ones presented here. This is obviously an exact result for
62
Fig. 3. CCB bounds versus MSUSY = mtL = mtR . Same set of values as in Figs. 1, 2 for the other
parameters.
ap
AD
t,3 , At [see Eqs. (2), (36)], but it proves also to occur to a very good approximation
(1)
for Act,3 , At .
In Fig. 3, the various CCB bounds are now plotted as a function of MSUSY = mtL = mtR ,
with the same set of values as in Fig. 2 for the other parameters. The CCB vacuum is now
automatically aligned in the SU(3)c D-flat direction, but not in the SU(2)L U (1)Y D-flat
one, except for MSUSY = m2 = 118 GeV, see Eq. (3).
In this illustration, we recover the same qualitative behaviour of the CCB bounds as in
2
=
Fig. 2. Comparing this illustration with the previous one for an equal value of MSUSY
c
2
D
(mt 2 + mt )/2, we furthermore observe that the difference |At,3 At,3 | is smaller in Fig. 3
R
L
than in Fig. 2, precisely because of this alignment in the SU(3)c D-flat direction. In Fig. 2,
2
= (720 GeV)2 with r1 = 5, the critical bound Act,3 is about 22%
for instance, for MSUSY
below the traditional bound AD
t,3 , whereas for an equal value of MSUSY = 720 GeV in
c
Fig. 3, which gives the same value for AD
t,3 , the critical bound At,3 is now just about
10% below AD
t,3 . This illustrates the fact that any departure of the CCB vacuum from the
SU(3)c D-flat direction or, equivalently, any splitting between the soft squark masses, tends
to lower substantially the critical bound Act,3 below which there is no CCB danger.
The large tan regime
Fig. 4 is devoted to the large tan regime. We take MSUSY mtL = mtR , with tan =
+. This benchmark scenario is often considered in Higgs phenomenology [13,14] and
this illustration is presented to set the stage for the next section where the impact of the
CCB conditions on the stop mass spectrum and on the one-loop upper bound on the lightest
63
Fig. 4. CCB bounds versus MSUSY mtL = mtR , for tan = +. The optimal sufficient bound is
(0)
(1)
2
2
Asuf
t Max[At , At ]. We take m2 = mZ 0 /2 [see Eq. (24)] and mt = 175 GeV which implies
Yt = 1.
Higgs boson mass, mh , will be considered. As will be shown in a forthcoming article

[11], this extreme tan case also proves to be numerically representative of the large tan
regime, i.e., tan 15, with furthermore || Min[mA0 , MSUSY ].
In this benchmark scenario, the CCB vacuum is automatically aligned in the SU(3)c
D-flat direction. Obviously, any discrepancy between the soft mass terms mtL , mtR would
induce a deviation from this direction and, on the other hand, a numerical modification of
the CCB bounds illustrated here, but the qualitative behaviour of the CCB bounds would
remain the same.
For tan = +, the EW vacuum is trapped in the plane (H2 , tL , tR ) and the depth of
the EW potential is determined to be V EW = m4Z 0 /2(g12 + g22 ) [see Section 3.4]. Fig. 4
illustrates how this geometrical feature of the potential affects the various CCB bounds,
inst
including the sufficient bound A(0)
t , Eq. (9), and the instability bound At , Eqs. (39), (40).
(0)
(1)
c
suf
inst
For all MSUSY , the hierarchy At Max[At , At ] At,3 At is respected, even
in the critical region MSUSY mt . This hierarchy merely reflects the fact that, on one
c
suf
hand, no CCB vacuum may develop for At Asuf
t , implying At,3 At . On the other, if
inst
At At , the EW vacuum would be automatically unstable and would bend down in the
c
suf
direction of a deeper CCB vacuum, which implies necessarily Ainst
t At,3 , At .
For all MSUSY , the critical bound Act,3 is just above the optimal sufficient bound Asuf
t .
(1)
c
suf
suf
For instance, for MSUSY 210 GeV, we have Asuf
t At,3 1.02 At , with At = At .
suf
This shows once again how an accurate approximation At can be for the critical CCB
64

ap
bound Act,3 . We note also that At provides a good estimate of Asuf

t , at least for MSUSY
300 GeV.
The traditional bound AD
t,3 in the D-flat direction exists only for MSUSY 45.56 GeV,
and above this value it increases fast. Let us remark that for MSUSY 100 GeV, AD
t,3 is not a
c
D
necessary upper bound on At to avoid CCB, because for At [At,3, At,3 ], the CCB vacuum
is not deeper than the EW vacuum and is therefore not dangerous. The traditional bound
AD
t,3 provides a necessary condition to avoid CCB only for MSUSY 100 GeV, but in this
case it is far from being sufficient. Typically much larger than the critical CCB bound Act,3 ,
it should also be handled with care. For instance, for MSUSY 300 GeV, it allows for some
values of At above the instability bound Ainst
t , implying a tachyonic lightest stop mass!
(1)
c
inst
For 110 GeV MSUSY 210 GeV, the various CCB bounds A(0)
t , At , At,3 , At
cluster. The potential enters in the critical regime where the CCB vacuum interfere with
the EW vacuum. In this regime, the critical CCB bound Act,3 coincides with the instability
bound Ainst
t , implying that a dangerous CCB vacuum may develop only if the lightest
physical stop gets tachyonic. Included in this small region of the parameter space, more
(0)
precisely for 140 GeV MSUSY 192 GeV, the typical hierarchy A(1)
t At is violated
(0)
suf
and we have therefore At = At . We note however that the maximal discrepancy between
(0)
(1)
At and At is quite small, less than 5 GeV.
Finally, we observe that for MSUSY 300 GeV, the critical CCB bound Act,3 is much
lower than the instability bound Ainst
t . This result has an important physical consequence
in the limiting case tan = + considered here, for which the stop mixing parameter
A t coincides with the trilinear soft term At . It implies that the CCB critical bound Act,3
provides stringent restrictions on the mass spectrum of the stop quark fields. This important
point is addressed in the next section.
7. The stop CCB maximal mixing

We investigate in this section some physical implications of the critical CCB bound on
the stop mass spectrum and the one-loop upper bound on the lightest Higgs boson mass.
We consider the benchmark scenario MSUSY = m2t = m2t , with tan = + [13,14]. As
L
R
noted in the last section, this extreme tan regime is also quite representative numerically
of the large tan regime with small , i.e., tan 15 and || Min[mA0 , MSUSY ] [11].
To be optimal, the extension to the low tan regime, valid for all values of , requires
an investigation of the extended plane (H1 , H2 , tL , tR ). This case will be presented in a
separate article [11].
In the benchmark scenario considered here, the stop mixing parameter A t = At +
/ tan equals the trilinear soft term At and the squared stop mass matrix is given by
the lower 2 2 matrix of M2 |EW , Eq. (38), taking H2 = v2 . Accordingly, the squared
masses for the stop eigenstates read [1]:

(8 m2W 5 m2Z 0 )2
1
1
2
.
+ m2t m2Z 0
4m2t A2t +
m2t ,t = MSUSY
(50)
1 2
4
2
36
65
These masses depend only on two free parameters, At and the unified soft squark mass
MSUSY . However, taking into account the CCB condition, a non-trivial correlation appears
between these two parameters. To avoid CCB in the plane (H2 , tL , tR ), it is necessary and
sufficient that At Act,3 , Eq. (27), where the dependence in MSUSY of the critical bound
Act,3 is plotted in Fig. 4. In the following, all parameters are supposed to be evaluated at the
appropriate renormalization scale Q QSUSY , in order to incorporate one-loop leading
corrections to this CCB bound.
In Fig. 5, we compare the critical bound Act,3 to the so-called Higgs maximal mixing
commonly considered in Higgs phenomenology [13,14]:
A max
t
2
A max
(51)
= Amax
= 6 mt with m2t = MSUSY
+ m2t .
t
t
As is well-known, the lightest Higgs boson mass, mh , receives a large one-loop correction
arising from top and stop loops, proportional to m4t and which grows logarithmically with
MSUSY . This correction is essential to overcome the tree-level upper bound mh mZ 0 and
is maximized for the Higgs maximal mixing, Eq. (51) [13,14]. We stress however that there
is no physical reason to exclude a stop mixing larger than this one, provided the masses obtained for the lightest stop and CP-even Higgs boson are not ruled out by experimental data.
is well below unity. We have
The prominent fact in Fig. 5 is that crit Act,3 /Amax
t
8
0.35 crit 9 0.89 for MSUSY 1500 GeV, showing that the CCB critical bound is at
least 10% below the Higgs maximal mixing, Eq. (51). Thus, the Higgs maximal mixing is
always ruled out by CCB considerations! Moreover, it can be shown that this striking result
holds not only for large tan , but is also typically verified for low tan [11]. Actually, the
lower tan is and the more CCB conditions will tend to rule out such a large stop mixing.
In the light of this new result, we introduce a new quantity, the CCB maximal mixing,
defined to be the largest stop mixing A t allowed by CCB considerations. Obviously, in
the case considered here, the CCB maximal mixing coincides with the critical value Act,3 ,
plotted in Fig. 4. Such a maximal mixing has a clear physical meaning, which implies in
particular that the lightest Higgs boson mass, mh , cannot reach its maximal value [at oneloop level, for the Higgs maximal mixing, Eq. (51)], unless the EW vacuum is a metastable
vacuum.
In Fig. 6, we illustrate the bounds on the stop masses induced by the CCB maximal
mixing as a function of MSUSY , and compare them to the Higgs maximal mixing and
also to the no mixing (At = 0) cases. The lower curves correspond to the minimal values
allowed for the mass of the lightest stop t1 and the upper to the maximal values allowed
for the mass of the heaviest stop t2 .
As expected, the bounds on the stop masses induced by the CCB maximal mixing are
more restrictive than for the Higgs maximal mixing. All stop mass values compatible with
a Higgs maximal mixing are always located in the dangerous CCB region. For MSUSY
425 GeV, the Higgs maximal mixing is already ruled out, either because it gives a tachyonic
lightest stop [for MSUSY 400 GeV], or because the lightest stop is too light [we take
conservatively mt1 100 GeV] and should have been already found experimentally [21].
In this region of small MSUSY , the CCB maximal mixing enables us to avoid such a
tachyonic lightest stop mass. For MSUSY 110 GeV, the lower bound on the lightest stop
66
Fig. 5. Critical CCB parameter crit Act,3 /Amix

versus MSUSY for tan = +. Same set of
t
parameters as in Fig. 4.
mass slowly decreases with MSUSY and becomes exactly zero for 110 GeV MSUSY
210 GeV. In this critical region, the CCB vacuum interferes with the EW vacuum and
cannot be deeper than the latter, unless the lightest stop mass becomes tachyonic.
For MSUSY 310 GeV, even the CCB maximal mixing is excluded by the conservative
experimental bound mt1 100 GeV. Therefore, in this region of low MSUSY , the EW
vacuum is necessarily the deepest one and cannot be metastable. This example illustrates
how experimental limits on the lightest stop combined with a precise determination of the
CCB condition may secure the EW vacuum in a large part of the parameter space, so that
metastability considerations become completely irrelevant. In addition , we have also in
this region the upper bound mt2 490 GeV.
For MSUSY 310 GeV, the bounds on the stop spectrum increase with MSUSY . At
MSUSY = 500 GeV, the discrepancies between the CCB maximal mixing and the Higgs
maximal mixing cases are quite large for the lightest stop, mt1 75 GeV, and smaller
but still important for the heaviest stop mt2 30 GeV. They tend to decrease slowly with
MSUSY and we have, e.g., mt1 30 GeV and mt2 20 GeV, for MSUSY = 1500 GeV.
The linear behaviour of the CCB bounds on the stop masses for large MSUSY is a direct
consequence of the asymptotic behaviour of the critical parameter crit 8/9 0.89 [see
Fig. 5]. For MSUSY 500 GeV, neglecting the gauge contributions, which are unimportant
in this regime, we obtain:
67
Fig. 6. Exclusion domain for the stop masses versus MSUSY , for tan = +. The higher curves
[above the no mixing curve] provide upper bounds on the mass of the heaviest stop t2 , and the lower
[below the no mixing curve] lower bounds on the mass of the lightest stop t1 . Same set of parameters
as in Fig. 4.

128
mt mt1 mt,
mt mt
27

128
mt
mt mt2 mt mt +
(52)
27

giving indeed a linear behaviour for large MSUSY : MSUSY 128
108 mt mt1 MSUSY and

MSUSY mt2 MSUSY + 128
108 mt .
Finally, we illustrate the impact of the CCB maximal mixing on the CP-even lightest
Higgs boson mass mh . At one-loop level, this mass has an upper bound reached for tan =
+, mA0 mZ 0 , and the Higgs maximal mixing, Eq. (51), [13,14]. We have shown that
the CCB condition rules out such a large stop mixing, therefore this upper bound on mh
may be lowered. 6 For simplicity, we consider this topic in a simplified setting, somewhat
unrealistic, taking only into account leading one-loop contributions coming from top and
stops loops. This will already point out the general trend and the importance of CCB
conditions in this context. In this case, we have [13,14]
m2h
m2Z 0

m2t
3m4t
At
At
+
Log 2 + 2 1
,
4v 2
mt
mt
12m2t
v = 174 GeV.
(53)
6 We note that the CCB maximal mixing, which coincides with the critical bound Ac in this case, does not
t,3
depend on the pseudo-scalar mass mA0 . As will be shown in [11], this interesting property actually extends for
all values of tan and .
68
Fig. 7. Upper bound on the lightest Higgs boson mass versus MSUSY , for tan = +. We suppose
mt1 100 GeV. Same set of parameters as in Fig. 4.
Up to one-loop level, it is consistent to combine the optimal tree-level CCB condition

At Act,3 , Eq. (27), illustrated in Fig. 4, with this upper bound. Fig. 7 illustrates the resulting one-loop CCB bound on mh as a function of MSUSY and compares it with the
maximal value reached for the Higgs maximal mixing. As a direct consequence of the
aforementioned conservative experimental limit on the lightest stop mt1 100 GeV [21],
three regimes can be considered: (i) For 0 MSUSY 310 GeV, the lightest stop mass experimental bound is more stringent than the CCB maximal mixing. In this regime, we have
mh 131 GeV. The experimental lower bound on mh puts in turn lower bounds on MSUSY .
For instance, taking mh 115 GeV, we must have MSUSY 190 GeV. (ii) For 310 GeV
MSUSY 425 GeV, the CCB maximal mixing becomes more restrictive than the conservative stop mass experimental bound, while the Higgs maximal mixing is still irrelevant,
either because the lightest stop mass is tachyonic or too light. In this regime, Eq. (53) combined with the critical CCB condition Act,3 [see Figs. 4, 5] give 131 GeV mh 138 GeV.
(iii) For MSUSY 425 GeV, the CCB and the Higgs maximal mixing are both more restrictive than the experimental bound on the lightest stop mass. The CCB upper bound on the
lightest Higgs boson mass is lower than the one for the Higgs maximal-mixing by about
2.8 GeV for MSUSY 425 GeV, which is a rather substantial effect. This discrepancy then
decreases slowly for larger MSUSY . We have, e.g., respectively mh (2.4, 1.5, 1.2) GeV
for MSUSY = (500, 750, 1000) GeV.
A more realistic investigation of the consequences of CCB conditions on mh clearly
requires that we go beyond the simple approximation given by Eq. (53). The complete
set of one-loop contributions should be taken into account, including in particular those
arising from bottom and sbottoms loops [which are themselves constrained by strong CCB
69
conditions for large tan , see Appendix B] [14]. To leading order, we can however trust
the discrepancy on the upper bound of the lightest Higgs boson squared mass between the
Higgs and the CCB maximal mixing
2 )2 m 4
9(1 + crit
t
(54)
,
4 2 v 2
m2h is actually independent of such additional contributions and depends only of MSUSY
via the critical parameter crit .
More importantly, two-loop contributions tend to lower substantially this upper bound
on the lightest Higgs boson mass, giving typically mh 130 GeV for mA0 , MSUSY
mt , with tan 1 [14]. Two-loop non-logarithmic contributions are also responsible of
a slight displacement of the stop mixing value where this upper bound is maximized [14].
A refined study of the importance of CCB conditions in this context, to be consistent at
two-loop level, should therefore require a complete one-loop level investigation of CCB
conditions in the plane (H2 , tL , tR ), in order to take also into account sub-leading effects
induced by mass discrepancies in the loops. Such a study is clearly beyond the scope of
this article and will the subject of further investigations. We believe however that this simple illustration already clearly indicates the crucial role CCB conditions can play in this
phenomenological context.
m2h =
8. Conclusions and outlook

In this article, we have presented at the tree-level a complete model-independent study
of the CCB conditions in the plane (H2 , tL , tR ). We have proposed a new procedure
to evaluate the CCB VEVs, which moreover enables us to obtain excellent analytical
approximations (at the level of the percent) for the VEVs and, ultimately, for the optimal
necessary and sufficient conditions on At to avoid CCB. The new conditions incorporate
the effect of all possible deviations of the CCB vacuum from the D-flat directions, in
particular from the SU(3)c D-flat direction previously disregarded [59]. We have pointed
out that the CCB vacuum typically deviates from the SU(3)c D-flat direction and that
this feature must be included in a consistent study of CCB conditions to encompass the
possibility of avoiding a tachyonic lightest stop. This deviation is controlled essentially by
the discrepancy between the soft squark masses mtL , mtR . Rather small in an mSUGRA
scenario [where typically mtL mtR ], it can be very large and make substantially
more restrictive the critical CCB conditions for mtL mtR or mtL mtR . This should
constrain even more model-dependent scenarii, in particular those exhibiting such large
mass discrepancies at the SUSY scale, e.g., some anomaly mediated models [4], or,
more generally, models incorporating non-universalities for the squark soft masses of the
third generation at a high energy scale. In order to take into account one-loop leading
corrections, the tree-level CCB conditions obtained in this article should be evaluated at an
appropriate scale Q QSUSY , where QSUSY is an average of the SUSY masses.
In the benchmark scenario, MSUSY = mtL = mtR and tan = +, we have illustrated
at this scale QSUSY some physical consequences of the critical CCB condition in the plane
70
(H2 , tL , tR ). A strong bound on the stop mixing parameter A t [= At in this

case] was
obtained, ruling out by more than 10% the Higgs maximal mixing |At | = 6 mt. This
led us to introduce a CCB maximal mixing for the stop fields. We have exhibited new
strong limits on the stop mass spectrum, which simply encode the physical requirement
of avoiding CCB. Finally, we have considered the impact of the CCB maximal mixing
on the upper bound of the CP-even lightest Higgs boson mass, mh , at one-loop level,
though in a simplified and rather unrealistic setting. Taking into account only top and stop
contributions, we have shown that this upper bound can be reduced by up to 3 GeV in
comparison with the maximal value reached for the Higgs maximal mixing. We believe
that these illustrations stress the importance of a refined study of CCB conditions, such
as the one presented here, in the context of Higgs phenomenology. We note however that
a more realistic investigation of the upper bound on the lightest Higgs boson mass, mh ,
requires that we take into account all one-loop contributions to the Higgs mass, not only
the leading top and stop ones, but also two-loop contributions. As is well-known, the latter
can be large and are also responsible of a displacement of the stop mixing which maximizes
mh [13,14]. A refined analysis of this important phenomenological topic, to be consistent
at two-loop level, should therefore require a precise one-loop study of CCB conditions in
the plane (H2 , tL , tR ), in order to obtain sub-leading contributions that our renormalization
group improved tree-level CCB conditions cannot grasp. Such a tedious study will be the
subject of future investigations.
In the benchmark scenario considered in this article, we have also pointed out that
combining a precise CCB information in the plane (H2 , tL , tR ) with a conservative
experimental input on the lightest stop mass, mt1 100 GeV, already indicates that the
EW vacuum is the deepest vacuum and is therefore stable in a large part of the parameter
space, MSUSY 310 GeV. Similar regions can also be found for any value of tan [11].
Outside these regions, following the philosophy of metastability, the EW vacuum can
still be considered as safe, even in the presence of a deeper CCB vacuum, provided its
lifetime exceeds the age of the Universe [9,10]. A numerical study of the tunneling rate
into the CCB vacuum is required to evaluate the relaxed CCB metastability condition
[16]. The present study, which can be straightforwardly completed by giving accurate
analytical expressions for the CCB saddle-point, provides also some enlightening pieces
of information on the shape of the potential barrier between the vacua, and therefore give
essential tools to investigate precisely this feature.
For completeness, it is important to stress that this investigation of CCB condition in
the plane (H2 , tL , tR ) is also numerically illustrative of what can be found in the extended
plane (H1 , H2 , tL , tR ), provided tan 15 and || Min[mA0 , MSUSY ]. In particular,
the results obtained for the critical CCB bound on At and the physical implications on
the stop mixing parameter and the stop mass spectrum are not substantially modified
compared to the extreme case tan = + illustrated here. In a forthcoming paper, we
will present the extension of this study to the plane (H1 , H2 , tL , tR ), and give optimal
CCB constraints on (At , ) valid for all values of tan [11]. We have also re-analyzed in
a fully model-independent way the potentially dangerous direction (H1 , H2 , tL , tR , L ),
previously considered in [6]. Additional, though not very restrictive, CCB conditions
71
involving the sneutrino soft mass m L will be given [12].

Besides physical implications on the MSSM mass spectrum, the CCB condition on At ,
completed with the one on the term obtained in the extended plane (H1 , H2 , tL , tR ) [11],
should also have further important consequences on the phenomenology of the MSSM
Higgs bosons [22]. In particular, CCB conditions provide dramatic restrictions on physical
processes which require, to be competitive, a stop mixing parameter A t as large as the
Higgs maximal mixing, e.g., for the production of neutral Higgs bosons associated with
top squarks [23]. Further investigations are currently made in this direction in order to
delineate more precisely the potential discovery of Supersymmetry.
Acknowledgement
This work was supported by a Marie Curie Fellowship, under contract No HPMF-CT1999-00363. I would like to thank especially G.J. Gounaris and P.I. Porfyriadis for their
help and the Theory Group of the University of Thessaloniki for its hospitality. I would
like also to thank G. Moultaka for valuable discussions and support during the completion
of this work, as well as for reading the manuscript and helping me to improve it. This work
was initiated in the French GDR in Supersymmetry. I thank its participants for stimulating
discussions. Special thanks also to M. Bezouh.
Appendix A. The optimal sufficient bound

In this appendix, we give some details on the derivation of the optimal sufficient bound
Asuf
t defined in Section 3.3, see Eq. (34). Let us consider the extremal equation associated
with H2 , Eq. (14):
EH2 = 3 H23 + 3 H22 + 3 H2 + 3 = 0,
(A.1)
where the coefficients 3 , 3 , 3 , 3 are given in the text, Eqs. (15)(18). Obviously, 3 is
always positive; 3 is proportional to the positive coefficient of the quadratic term in H2
of B3 , Eq. (6), and is also positive; 3 is dominated by a large contribution in Yt4 and is
therefore negative; finally, 3 is dominated by the negative terms proportional to A2t and
mt 2 + f 2 m2t and is negative. The sign of these coefficients imply that any real root of EH2 ,
R
L
considered as a cubic polynomial in H2 , must be positive. Moreover, by simple inspection
of these roots expressed as a function of the coefficients 3 , 3 , 3 , 3 , it is straightforward
to show that the necessary and sufficient condition to have only one real root is indeed
given by Eq. (33).
Let us prove now that the potential V3 , Eq. (1) has no local CCB minimum, if the
extremal equation associated with H2 , Eq. (14), has only one real root in H2 for any
value f .
Equivalently, we can show that if the potential V3 , Eq. (1), has one local CCB minimum
( H2 , tL , f ), then for f = f the extremal equation EH2 = 0 has necessarily three
real roots in H2 .
72
Let us consider the following continuous path P in the plane (H2 , tL , tR ): for
], we take H2 = , tL = tR = 0; for [H
2 , H2 ], we take H2 = , tL =
[0, H
2
2B3 /A3 |f = f , tR = f tL , where A3 , B3 are given in Eq. (6). Here, the positive value
2 denotes the lowest solution of the equation B3 = 0 for f = f . By definition, the
H
path P goes through the origin of the fields and also through the local CCB vacuum for
= H2 [see Eq. (5)].
We can show now that, in this situation, it is absurd to have only one real root for the
extremal equation associated with H2 , Eq. (14). By definition, the solutions of this equation
provide in particular the value H2 of any directional extremum in the second part of the path
2, H2 ]. Therefore, assuming that this equation has only one real root for
P, i.e., for [H
f = f implies that on this part of the path P, there is no directional saddle-point. What
2]? Here, the potential V3 , Eq. (1), reads:
about the first part of the path P, i.e., for [0, H
V3 = H22

(g12 + g22 ) 2
2
H2 .
m2 +
8
(A.2)
For m22 0, this potential is monotonous as a function of H2 and has no non-trivial extrema. Hence, in this case, we finally conclude that we can find a continuous path P in the
plane (H2 , tL , tR ) which connects the origin of the fields and the CCB vacuum, moreover
without any directional saddle-point.
We remind the reader that we have assumed in our study positive squared soft mass
m2t , m2t to avoid an obvious CCB problem at the origin of the fields and that the CCB
L
R
extremum ( H2 , tL , f ) we consider is supposed to be a local minimum of the potential
V3 , Eq. (1). In this light, the conclusion obtained for m22 0 is absurd, because in this case
the origin of the fields is also a local minimum, so that, necessarily there must be a barrier
separating it from the CCB vacuum, and a saddle-point on any path connecting them.
The regime m22 0 is somewhat more complicated to investigate and requires that we
adjust the path P to different cases. This comes from the fact that the potential in Eq. (A.2)
has an additional non-CCB extremum, H2 2EW = 4m22 /(g12 +g22 ), as noted in Section 3.4.
Therefore, the cases H2 H2 EW , H2 H2 EW , and H2 H2 EW should be
considered separately. The path P proposed is obviously only adapted to the last case. We
will not enter into such a detailed, but straightforward, demonstration. Actually, assuming
that this additional non-CCB extremum is a local minimum, as done in this article [see
Section 3.4, Eq. (39)], it is easy to convince one-self that in all these cases a conclusion
similar to the one obtained for m22 0 is obtained: it is absurd to suppose that the extremal
equation associated with H2 , Eq. (14), has only one solution in H2 for f = f , because
on any path connecting this additional non-CCB minimum and the CCB minimum, there
should be a saddle-point which necessarily would show as an additional real solution of
this equation.
Hence, without loss of generality, we conclude that if EH2 = 0 has only one real solution
in H2 for any value of f , then the potential V3 , Eq. (1), cannot have any local CCB
minimum. In such a situation, the unique solution of EH2 = 0 found is spurious and located
outside the compact domain where tL 2 0, Eqs. (12), (13).
73
Appendix B. CCB conditions in the plane (H1 , b L , b R )

The procedure to evaluate the VEVs of the extrema of the potential and the geometrical
picture presented in this article hold also for the tree-level potential in the plane
(H1 , bL , bR ), provided the bottom Yukawa coupling Yb is large enough, or equivalently
for large tan . Here, bL and bR stand for the left and right sbottom fields of the same
generation, and H1 is the neutral component of the corresponding Higgs SU(2)L scalar
doublet. In this plane, the tree-level potential reads [1,5]:

2
2
2
2
2 2
bR
V 3 = m21 H12 + m2b bL
+ m2b bR
2Yb Ab H1 bL bR + Yb2 H12 bL
+ H12 bR
+ bL
L
R

g2 2
2b 2 2 g 2
g2
b 2
2 2
2 2
+ 1 H12 L R + 2 H12 bL
(B.1)
+ 3 bL
bR
.
8
3
3
8
6
This potential is similar to V3 , Eq. (1). In fact, redefining the fields and parameters of V 3
as follows [1,5,6]
H1 H2 ,
bR/L u R/L ,
m1 m2 ,
mbR/L mu R/L ,
Yb Yu
(B.2)
we recover V3 , Eq. (1), except for a minor difference coming from the U (1)Y D-term.
The relevant expressions to study CCB conditions in this plane are the following ones.
The VEV bL verifies:
A 3 bL 2 + 2B 3 = 0,
(B.3)
where

2
2
A 3 g12 2 fb 2 + 1 /18 + g22 /2 + 2g32 fb 2 1 /3 + 4Yb2 fb 2 ,
2
2
2 [12Yb ( fb
3 H1
B
+ 1) 3g22
g12 (2 fb 2
12
2Ab Yb fb H1 + m2b + fb 2 m2b ,
L
(B.4)
+ 1)]
(B.5)
3 tells us
with fb bL /bR . Quite similarly to what happens in the plane (H2 , tL , tR ), B
when the large
regime opens. This occurs, whatever fb is, for
bottom Yukawa coupling
2
2
Yb Max[ (g1 + 3g2 )/12, g1 / 6] 0.3 (at the EW scale). Using the tree-level relation
mt /mb = Yt /Yb tan 35 [1], with Yt 1, this requires tan 10.5.
In this large tan regime, we obtain a sufficient bound to avoid CCB in the plane
(0)
(H1 , bL , bR ), similar to At , Eq. (9). It reads

g12
(3g22 + g12 )
(0)
+
m
1
.
Ab Ab mbL 1
(B.6)
b R
6Yb2
12Yb2

can
be
quite
small
for
Y
(g12 + 3g22 )/12.

We note however that A(0)
b
b
The extremal equation associated with H1 reads
3 H13 + 3 H12 + 3 H1 + 3 = 0
(B.7)
74
with
2
2

3 = 36Yb4 fb2 + 1 + 3g32 g12 + g22 + g12 g22 fb2 1

+ 6Yb2 g12 2fb4 + 6fb2 + 1 + 18Yb2g22 2fb2 + 1 ,
(B.8)
2
2 2
2
2
3 = 9Ab Yb fb 12 fb + 1 Yb 2fb + 1 g1 3g2 ,
(B.9)

3 = 72A2b fb2 Yb2 3 m2b + fb2 m2b 12Yb2 fb2 + 1 g12 2fb2 + 1 3g22
L
R

2

2
2
2
2 2
2
+ m1 72Yb fb + g1 2fb + 1 + 9g22 + 12g32 fb2 1 ,
(B.10)

3 = 36Ab Yb fb m2 + fb2 m2 .
(B.11)
b L
b R
The extremal equation associated with fb reads

a 3 fb H12 + b3 H1 + c3 fb = 0
(B.12)
with

a 3 = 2Yb2 18Yb2 12g32 fb2 1 + 9g22 g12 4fb2 1

+ fb2 1 g12 g22 + 3 g12 + g22 g32 ,

b3 = Ab Yb 12g32 fb4 1 9g22 + g12 4fb4 1 ,

c3 = m2b 36Yb2 + 12g32 fb2 1 + 2g12 2fb2 + 1

L

+ m2b 36fb2 Yb2 12g32 fb2 1 + 9g22 + g12 2fb2 + 1 .
R
(B.13)
(B.14)
(B.15)
The VEVs ( H1 , fb ) of a consistent CCB vacuum have to verify this set of coupled
equations and must furthermore be included in the compact domain where bL 2 0.
The recursive algorithm to compute the VEVs of the CCB vacuum is identical to the
one presented in the text for the top Yukawa coupling regime. A convenient initial value to
accelerate the procedure is

A2 + 2m2 m2
b
b L
b R
(0)
.
fb,3 = 2
2
Ab + 2m m2
bR
bL
The optimal sufficient bound Asuf

b below which no CCB vacuum may develop in the plane
(H1 , bL , bR ) is always given by the largest solution in Ab of the equation
3

(0)
C3 233 9 3 3 3 + 27 32 3 + 4 32 + 3 3 3 = 0, taking f = fb,3 .
The potential V 3 , Eq. (B.1), can have at most three extrema: the origin of the fields, a CCB
local minimum and a CCB saddle-point. In particular, there is no additional non-CCB
extremum. Such a possibility would appear for m21 0, which requires tan 0. This
is obviously outside the large bottom Yukawa coupling regime and is, moreover, ruled
out by experimental data. Finally, the necessary and sufficient bound Acb,3 is obtained by
scanning the region Ab Asuf
b and comparing the potential V 3 with the EW potential
V |EW , Eq. (26). For large tan , which implies Yb 1, the appropriate renormalization
c
scale to evaluate the tree-level CCB conditions Asuf
b , Ab,3 in order to incorporate one-loop
leading corrections, is the SUSY scale QSUSY , quite similarly to the case (H2 , tL , tR ).
75
References
[1] See for instance, P. Fayet, S. Ferrara, Phys. Rep. 32 (1977) 249;
H.P. Nilles, Phys. Rep. 110 (1984) 1;
H.E. Haber, G.L. Kane, Phys. Rep. 117 (1985) 75;
A.B. Lahanas, D.V. Nanopoulos, Phys. Rep. 145 (1987) 1;
R. Arnowitt, P. Nath, Report CPT-TAMU-52-93;
M. Drees, S. Martin, hep-ph/9504324;
S. Martin, hep-ph/9709356.
[2] J. Wess, J. Bagger, Supersymmetry and Supergravity, Princeton Series in Physics.
[3] For the mSUGRA scenario, A.H. Chamseddine, R. Arnowitt, P. Nath, Phys. Lett. 49 (1982)
970;
R. Barbieri, S. Ferrara, C.A. Savoy, Phys Lett. B 119 (1982) 343;
L. Hall, J. Lykken, S. Weinberg, Phys. Rev. D 27 (1983) 2359;
For the Gauge Mediated scenario, M. Dine, W. Fischler, M. Srednicki, Nucl. Phys. B 189 (1981)
575;
S. Dimopoulos, S. Raby, Nucl. Phys. B 192 (1981) 353;
G.F. Giudice, R. Rattazzi, Phys. Rept. 322 (1999) 419, 501.
[4] For the Anomaly Mediated scenario, see, for instance, L. Randall, R. Sundrum, Nucl. Phys.
B 557 (1999) 79;
G.F. Giudice, M. Luty, H. Murayama, R. Rattazzi, JHEP 9812 (1998) 27;
See also, P. Binetruy, M.K. Gaillard, B. Nelson, hep-ph/0011081.
[5] J.M. Frere, D.R.T. Jones, S. Raby, Nucl. Phys. B 222 (1983) 11;
L. Alvarez-Gaum, J. Polchinski, M. Wise, Nucl. Phys. B 221 (1983) 495;
M. Claudson, L.J. Hall, I. Hintchliffe, Nucl. Phys. B 228 (1983) 501;
H.P. Nilles, M. Srednicki, D. Wyler, Phys. Lett. B 120 (1983) 346;
J.P. Derendiger, C.A. Savoy, Nucl. Phys. B 237 (1984) 307;
C. Kounas, A.B. Lahanas, D.V. Nanopoulos, M. Quirs, Nucl. Phys. B 236 (1984) 438;
M. Drees, M. Glck, K. Grassie, Phys. Lett. B 157 (1995) 164;
J.F. Gunion, H.E. Haber, M. Sher, Nucl. Phys. B 306 (1988) 1;
H. Komatsu, Phys. Lett. B 215 (1988) 323.
[6] J.A. Casas, A. Lleyda, C. Muoz, Nucl. Phys. B 471 (1995) 3.
[7] J.A. Casas, S. Dimopoulos, Phys. Lett. B 387 (1996) 107;
J.A. Casas, A. Lleyda, C. Muoz, Phys. Lett. B 389 (1996) 305;
H. Baer, M. Brhlik, D. Castao, Phys. Rev. D 54 (1996) 6944;
J.A. Casas, hep-ph/9707475;
S.A. Abel, C.A. Savoy, Nucl. Phys. B 532 (1998) 3;
S.A. Abel, C.A. Savoy, Phys. Lett. B 444 (1998) 119;
S.A. Abel, T. Falk, Phys. Lett. B 444 (1998) 427;
J.A. Casas, A. Ibarra, C. Muoz, Nucl. Phys. B 554 (1999) 67;
A. Dedes, A.E. Farragi, Phys. Rev. D 62 (2000) 016010;
S.A. Abel, B.C. Allanach, hep-ph/9909448.
[8] A. Bordner, hep-ph/9506409.
[9] P. Langaker, N. Polonski, Phys. Rev. D 50 (1994) 2199;
A. Riotto, E. Roulet, Phys. Lett. B 377 (1996) 60;
A. Strumia, Nucl. Phys. B 482 (1996) 24;
T. Falk, K.A. Olive, L. Roszkowski, M. Srednicki, Phys. Lett. B 367 (1996) 183;
A. Kusenko, P. Langaker, G. Segr, Phys. Rev. D 54 (1996) 5824;
T. Falk, K.A. Olive, L. Roszkowski, A. Singh, M. Srednicki, Phys. Lett. B 396 (1997) 50;
L. Dasgupta, R. Rademacher, P. Suranyi, Phys. Lett. B 447 (1999) 284.
[10] A. Kusenko, P. Langaker, G. Segr, Phys. Rev. D 54 (1996) 5824.
76
[11] C. Le Moul, in preparation.

[12] C. Le Moul, G. Moultaka, in preparation.
[13] H.E. Haber, R. Hempfling, Phys. Rev. Lett. 66 (1991) 1815;
Y. Okada, M. Yamaguchi, T. Yanagida, Prog. Theor. Phys. 85 (1991) 1;
J. Ellis, G. Ridolfi, F. Zwirner, Phys. Lett. B 257 (1991) 83;
R. Barbieri, M. Frigeni, Phys. Lett. B 258 (1991) 395.
[14] J.S. Espinosa, R.-J. Zhang, JHEP 0003 (2000) 26;
J.S. Espinosa, R.-J. Zhang, Nucl. Phys. B 586 (2000) 3;
M. Carena, H.E. Haber, S. Heinemeyer, W. Hollik, C.E.M. Wagner, G. Weiglein, Nucl. Phys.
B 580 (2000) 29.
[15] G. Gamberini, G. Ridolfi, F. Zwirner, Nucl. Phys. B 331 (1990) 331;
B. de Carlos, J.A. Casas, Phys. Lett. B 309 (1993) 320.
[16] S. Coleman, Phys. Rev. D 15 (1977) 2929;
C.G. Callan, S. Coleman, Phys. Rev. D 16 (1977) 1762;
A. Kusenko, Phys. Lett. B 377 (1996) 245;
I. Dasgupta, Phys. Lett. B 394 (1997) 16.
[17] C.T. Hill, Phys. Rev. D 24 (1981) 691;
C.T. Hill, C.N. Leung, S. Rao, Nucl. Phys. B 262 (1985) 517;
M. Carena, M. Olechowski, S. Pokorski, C.E.M. Wagner, Nucl. Phys. B 419 (1994) 217;
M. Carena, C.E.M. Wagner, Nucl. Phys. B 452 (1995) 45;
M. Lanzagorta, G. Ross, Phys. Lett. B 364 (1995) 163;
M. Carena, P. Chankowski, M. Olechowski, S. Pokorski, C.E.M. Wagner, Nucl. Phys. B 491
(1997) 103.
[18] D.I. Kazakov, G. Moultaka, Nucl. Phys. B 577 (2000) 121;
G.K. Yeghiyan, M. Jurcisin, D.I. Kazakov, Mod. Phys. Let. A 14 (1999) 601;
M. Jurcisin, D.I. Kazakov, Mod. Phys. Lett. A 14 (1999) 671.
[19] D.J. Castao, E.J. Piard, P. Ramond, Phys. Rev. D 49 (1994) 4882;
M. Carena et al., Nucl. Phys. B 426 (1994) 269;
W. De Boer, R. Ehret, D.I. Kazakov, Z. Phys. C 67 (1994) 647;
V. Barger, M.S. Berger, P. Ohmann, Phys. Rev. D 49 (1994) 4908;
M. Drees, S. Martin, hep-ph/9504324.
[20] M. Bando, T. Kugo, N. Maekawa, H. Nakano, Phys. Lett. B 301 (1993) 83;
See also, C. Ford, D.R.T. Jones, P.W. Stephenson, M.B. Einhorn, Nucl. Phys. B 395 (1993) 17;
M. Bando, T. Kugo, N. Maekawa, H. Nakano, Prog. Theor. Phys. 90 (1993) 405;
C. Ford, C. Wiesendanger, Phys. Lett. B 398 (1997) 342;
C. Ford, C. Wiesendanger, Phys. Rev. D 55 (1997) 2202;
J.A. Casas, V. Di Clemente, M. Quirs, Nucl. Phys. B 553 (1999) 511.
[21] CDF Collaboration, Abstract 652, Conference ICHEP 98, July 1998;
OPAL Collaboration, G. Abiendi et al., Phys. Lett. B 456 (1999) 95.
[22] For a recent reviews, see, for instance, A. Djouadi, R. Kinnumen, E. Richter Was, H.U. Martyn
et al., hep-ph/0002258;
M. Carena, J.S. Conway, H.E. Haber, J.D. Hobs et al., hep-ph/00010338, and references therein.
[23] A. Dedes, S. Moretti, Eur. Phys. J. C 10 (1999) 515;
G. Blanger, F. Boudjema, T. Kon, V. Lafage, Eur. Phys. J. C 9 (1999) 511;
A. Djouadi, J.L. Kneur, G. Moultaka, Nucl. Phys. B 569 (2000) 53;
A. Djouadi, J.L. Kneur, Phys. Rev. Lett. 80 (1998) 1830.

Brane mediated supersymmetry breaking

S.F. King, D.A.J. Rayner
Department of Physics and Astronomy, University of Southampton, Southampton, SO17 1BJ, UK
Received 14 December 2000; accepted 3 May 2001
Abstract
We propose a mechanism for mediating supersymmetry breaking in type I string constructions.
The basic set-up consists of a system of three D-branes: two parallel D-branes, a matter D-brane and
a source D-brane, with supersymmetry breaking communicated via a third D-brane, the mediating
D-brane, which intersects both of the parallel D-branes. We discuss an example in which the first
and second family matter fields correspond to open strings living on the intersection of the matter
D-brane and mediating D-brane, while the gauge fields, Higgs doublets and third family matter
fields correspond to open strings living on the mediating D-brane. As in gaugino mediated models,
the gauginos and Higgs doublets receive direct soft masses from the source brane, and flavourchanging neutral currents are naturally suppressed since the first and second family squarks and
sleptons receive suppressed soft masses. However, unlike the gaugino mediated model, the third
family squarks and sleptons receive unsuppressed soft masses, resulting in a very distinctive spectrum
with heavier stops, sbottoms and staus. 2001 Elsevier Science B.V. All rights reserved.
1. Introduction
The process of SUSY breaking continues to be an active area of research. Over the
years there have been various mechanisms proposed, including gauge [1] and anomaly
mediation [2]. An alternative mechanism has been put forward [3,4] called gaugino
mediated supersymmetry (SUSY) breaking which has the attractive property of solving the
flavour problem, since scalars masses effectively vanish at the GUT scale and are generated
through radiative corrections for which a GIM-like mechanism prevents flavour-changing
neutral current (FCNC) problems. This is rather like the no-scale supergravity mechanism
[5], but is implemented within a HoravaWitten [6] type set-up 1 consisting of two parallel
but spatially separated D3-branes with SUSY broken on one brane, with the SUSY matter
fields living on the other brane and the gauge sector living in the bulk and communicating
the SUSY breaking from one brane to the other. The Higgs doublets may also be in the
E-mail address: sfk@hep.phys.soton.ac.uk (S.F. King).
1 Note that in the HoravaWitten model gauge fields do not live in the bulk.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 2 2 0 - 6
78
S.F. King, D.A.J. Rayner / Nuclear Physics B 607 (2001) 7798
bulk providing a solution to the problem via the GiudiceMasiero mechanism [7]. The
advantage of this set-up is that the contact terms arising from integrating out states with
mass M are suppressed by a Yukawa factor eMr if M r, and so a modest separation
between the two branes can lead to negligible direct communication between the SUSY
breaking brane and the matter brane. This is the starting point of both the anomaly mediated
and the gaugino mediated models, and underpins the solution to the FCNC problem in both
cases.
In this paper we shall propose a mechanism for mediating SUSY breaking in type I string
models based on open strings starting and ending on D-branes. Type I string theories can
provide an attractive setting for ideas such as gaugino mediated SUSY breaking (gMSB),
and we shall explore this possibility in this paper. In place of the HoravaWitten set-up we
shall consider a type I toy model consisting of two parallel D-branes with a third D-brane
intersecting with both of the parallel D-branes. Instead of having the gauge fields in the
bulk we shall put the gauge fields onto the third mediating D-brane, which allows SUSY
breaking to be communicated between the SUSY breaking brane and the matter brane.
Thus the role of the bulk is played by the third mediating D-brane, and it is the gauge fields
which live on this brane that communicates the SUSY breaking. However in type I models
it is natural for a matter family to also live on the mediating D-brane, and this provides a
characteristic signature of the brane mediated SUSY breaking mechanism.
To illustrate these ideas we consider a toy model inspired by the work of Shiu and Tye
[8] using intersecting D5-branes, where the intersection regions are effectively parallel D3branes within a higher-dimensional spacetime. In this model two chiral families occur in
the 4d intersection region at the origin fixed point (51 52 sector), with a third family on
the D52 -brane (5252 sector). However our model differs from ShiuTye since we include
a further D51 -brane which intersects with the D52 -brane at a point located away from the
origin fixed point, and suppose that SUSY gets broken on that brane and is communicated
via the states on the D52 -brane which intersect with both D51 -branes at the two fixed
points brane mediated SUSY breaking (BMSB). In this example gauge fields, Higgs
fields and the third family all live in the mediating D-brane which plays the role of the bulk
in the original scenario. This separation of the third family 2 provides an explanation for
the large mass of the third family of quarks and leptons, without perturbing the solution to
the flavour problem since the first and second families remain almost degenerate.
The layout of the remainder of the paper is as follows. In Section 2 we review
gMSB,
and in Section 3 we introduce a type I string-inspired toy model motivated by

Shiu and Tye. Section 4 is the main section of the paper in which we present our toy
model that illustrates the BMSB mechanism, and explore its theoretical and experimental
consequences. Section 5 concludes the paper.
2 Remember that the first two families are localised within an effective 4d overlapping region, while the third
family feels two extra dimensions.
79
2. Gaugino mediated supersymmetry breaking

In this section we review the gMSB
mechanism in Refs. [3,4]. This toy model involves

D3-branes embedded in a higher-dimensional space. Two parallel D3-branes are spatially
separated along (at least) one extra dimension as shown in Fig. 1. Standard model quark
and lepton fields are localised on the matter brane as open strings, while the gauge and
(possibly) Higgs fields propagate in the bulk. 3 Supersymmetry is broken on the displaced
source D3-brane. SUSY breaking is communicated to the bulk fields by direct higherdimensional interactions, 4 and mediated to the quark/lepton fields by standard model
loops. 5
The full D-dimensional Lagrangian is split into two distinct pieces a bulk term
involving only bulk fields and terms localised on either D3-brane that allow direct bulkbrane field coupling.

D4 (y yj )Lj bulk(x, yj ), j (x) ,
LD = Lbulk(bulk (x, y)) +
(1)
j
where j runs over the branes, x are coordinates for the 4 non-compact dimensions, y are
coordinates for the D 4 compact spatial dimensions, bulk is a bulk field, and j is a
field localised on the j th brane.
A naive dimensional analysis (NDA) allows the 5d (or higher) effective theory to
be matched on to the observed 4d theory at the compactification scale. The 4 and
D-dimensional gauge couplings can be related by the size of the compact dimensions:
g42 =
2
gD
.
VD4
(2)
The D-dimensional gauge coupling gD must be smaller than its strong-coupling limit,
otherwise perturbative results become meaningless 6
2
gD
lD
,
M D4
(3)
where lD is a geometrical loop factor for D dimensions, lD = 2D D/2 (D/2), M is the

fundamental scale in the theory which acts as a regulating cutoff, and suppresses the
coupling strength. 1 corresponds to the strong coupling limit. This places a constraint,
along with FCNC suppression, that restricts the maximum size of the extra dimensions.
(See [3,4] for details.)
Following the work of Randall and Sundrum on spatially-separated D3-branes in
extra dimensions [2], contact terms between fields on opposite branes are exponentially
3 Thus feeling all 5-dimensions.
4 Higher-dimensional operators are assumed to arise from the underlying string theory, although this is not
clear at present.
5 Gauginos in the bulk couple directly to chiral fermions on the matter brane. They also couple to the hidden
sector directly through mass-insertions on the source brane.
6 Extra dimensions (and KaluzaKlein excitations) change the energy-dependence of couplings to power law
running above the compactification scale. This allows for unification at lower scales, see [9] for a review.
80
Fig. 1. An extra dimensional loop diagram that contributes to SUSY breaking scalar masses. It is
similar to a self-energy diagram, but with the virtual gaugino not confined to either 4-dimensional
brane. This figure is taken from Ref. [3].
suppressed by an amount eML , where L is the separation between D3-branes along the
extra dimension(s).
Eq. (4) is an example of an exponentially suppressed 4-point operator involving fields
from the matter and source branes that generates scalar masses:

eML
M
d 4 S S M
Lbrane
(4)
2
M
(where S , M are source and matter fields, respectively).
Compare the suppressed contact terms with the operators giving rise to gaugino masses
and Higgs SUSY breaking parameters, from Higgs superfields hu , hd and gauge field
strengths W living in the bulk. 7

lD
1
2
Lbrane
d D3 S W W + h.c.
l4
M

lD
1
4
+
hu hd + h.c.
d
l4
M D3 S

1

+ D2 S S hu hu + hd hd + (hu hd + h.c.) .
(5)
M
This leads to soft terms when we match to the D-dimensional theory 8 and using Eqs. (2),
(3) with g4 1:
m ,
S
S
lD / l4
F
1 F
,
D4
MM
VD4 l4 M
7 The scale factors M arise from the requirement of canonical normalization.

8 Notice that the B-term and Higgs mass-squared terms are enhanced by a volume factor relative to the
m -, -terms.
B, m2hu , m2hd
2
2
F
1 F
lD / l4
S
S
.
M 2 M D4 VD4 l4 M 2
81
(6)
Both papers discuss methods of generating the -term. 9 Ref. [3] suggested the inclusion
of an additional gauge singlet on the matter brane (NMSSM) with an extra superpotential
term W Nhu hd . An effective -term is produced if N acquires a non-zero vacuum
expectation value (vev). Another possibility [4] is to produce the -term on the source
brane through the GiudiceMasiero mechanism [7] (as above) L d 4 S hu hd .
3. Type I string-inspired model

Now we turn to type I string constructions and introduce a toy model motivated by
the work of Shiu and Tye [8]. The string scale ms is usually considered to be of the
order 1016 GeV, but recently the gauge unification scale was suggested to be as low as
1 TeV, which could allow the string scale at a comparable value. Shiu and Tye [8] discuss
the phenomenological possibilities within type I string theory and overlapping D5-branes.
They use the duality between the compactification of 10-dimensional type IIB string theory
on an orientifold, with type I theory on an orbifold to recover a 4-dimensional N = 1
supersymmetric chiral string model with PatiSalam-like gauge symmetry.
Tadpole cancellations and a non-zero background NSNS B-field constrain the number
and type of D-branes allowed within the model to D5- and D9-branes only [10]. In a
particular scenario they consider only one type of D5-brane (53 ) together with the D9brane, and after T-dualizing they arrive at a scenario with two intersecting branes, namely
51 - and 52 -branes which intersect at the origin fixed point. A gauge group U (4) U (2)
U (2) exists on each brane, and they discuss three scenarios where the standard model
gauge group originates from different brane combinations. Their third scenario is of
particular interest since it leads to three chiral families two families on the 51 52 overlap
and a third family on the D52 -brane as shown in Fig. 2.
We can express the allowed superpotential [11] in terms of the possible states from the
two types of D5-branes present in this model:
W = C152 C252 C352 + C352 C 51 52 C 51 52 .
(7)
We now proceed to introduce a toy model based on the above construction. 10 In order to
3 F3 h) consistent with the string selection-rules

allow the third family Yukawa couplings (F
in Eq. (7), we shall assign the Higgs hu , hd C152 or C252 . This leads to the four possible
allocations of 52 states in Table 1.
Notice that there are no free indices on the intersection states Qi , Li , UiC , DiC , EiC ,
C
Ni C 51 52 (i = 1, 2), which means that we cannot distinguish between the first two
families.
9 Ref. [3] has the Higgs fields localised on the matter D3-brane, while [4] has the Higgs fields living in the
bulk.
10 For other examples of toy models based on this construction see [12] and references therein.
82
Fig. 2. The matter fields and Higgs doublets resulting from Shiu and Tyes third scenario with
p
intersecting D5-branes, where Ci is an open-string state (matter field) starting and ending on the
pq
pth brane. C is an open-string state starting on the pth brane and ending on the qth brane.
In our type I string-inspired model, we shall assign the gauge groups and matter fields as
b
in Table 2. We ignore the custodial SU(4)51 U (1)6 symmetry. The states , , H b , H
are used to break the gauge group down to the standard model as discussed in Appendix A.1.
Gauge invariance with respect to the initial gauge group SU(4)52 SU(2)52R
SU(2)52L SU(2)51R SU(2)51L provides the mechanism to forbid both first and sec
i Fj h) and R-parity violating operators without any other
ond family Yukawa couplings (F
11
assumptions.
Note that the -term is forbidden by string selection rules which also forbid a
superpotential term involving a matter brane singlet 12 where W Nhu hd . The Giudice
Masiero mechanism offers the best opportunity of producing a -term from the soft
potential as discussed later.
4. Brane mediated supersymmetry breaking
We now augment the model in Section 3, including the states in Table 2, by including an
additional D51 -brane located at an orbifold fixed point away from the origin as shown in
Fig. 3. The idea of including the extra 51 -brane is that SUSY is broken on this brane and
communicated by the MSSM states that live on the 52 -brane which intersects it. Thus, the
gauge fields on the 52 -brane play the role of the gauge fields in the bulk in Fig. 1. Note that
there are many mass scales in this model as discussed in Appendix B.
We now consider a limiting case in which the model in Fig. 3 reduces to the gMSB
model discussed in Section 2, namely that the D52 radius of compactification is very much
larger than the D51 radius 13
11 Note that the third family right-handed neutrinos and sneutrinos receive large Majorana masses from the
3 H H resulting in a see-saw mechanism. This is discussed in Ref. [13], along with a discussion of
3 F
operators F
higher-dimensional operators suitable for first and second family fermion masses.
12 A non-renormalisable higher-dimensional 4-point superpotential term may be generated by two additional
gauge singlet fields, e.g., W N1 N2 hu hd . This can become the 3-point term when one of the singlet fields
acquire a vev.
13 Both inverse radii must be larger than the inverse string scale.
83
Table 1
Allocation of 52 states that lead to third family-only Yukawa couplings at lowest order. We use the
lower index to distinguish between doublets, singlets and Higgs fields
52 states
h hu , hd
C1 2
C1 2
C2 2
C2 2
F3 Q3 , L3
C2 2
C3 2
C3 2
C1 2
3 U C , D C , E C , N C
F
3
3
3
3
C3 2
C2 2
C1 2
C3 2
5
5
5
Table 2
SU(4)52 SU(2)52R SU(2)52L SU(2)51R SU(2)51L quantum numbers for left- and righthanded chiral fermion states and symmetry breaking Higgs fields
States
Sector
SU(4)52
SU(2)52R
SU(2)52L
SU(2)51R
SU(2)51L
Fi Qi , Li
Fi UiC , DiC , EiC , NiC ,

F3 Q3 , L3
F3 U C , D C , E C , N C ,
51 52
4
4
52
4
4
51 52
51 52
H b
52
52
4
4
b
H
h hu , hd
52
51 52
52
R52 R51 m1
s .
(8)
In this limit, the model reduces to that shown in Fig. 1, where the D3-branes correspond
to the intersection regions of the D5-branes, and the bulk corresponds to the mediating 52 brane, as shown in Fig. 4. Note that the first two families are located on the matter brane,
while the third family and Higgs doublets live on the mediating 52 -brane.
Since the gauge couplings on the branes are given by
=
g52
2
m2s v2
,
(2)3
=
g52
1
m2s v1
(2)3
(9)
we know that the coupling-squared is inversely proportional to compactification volume

(vi (2Ri )2 ), which implies that g51 g52 . This limiting case of the symmetry breaking
is discussed in Appendix A.2, but the important results are that the dominant components
with two extra

of the gauge fields live on the D52 -brane which is consistent with gMSB
bulk dimensions. After the gauge symmetry is broken down to the standard model, we
recover the relationship between gauge couplings:
84
Fig. 3. A brane-construction using overlapping D5-branes, with effective D3-branes at the

intersection points spatially separated along the D52 -brane. The first two chiral families (C 51 52 )
5
live on the first intersection region. The third family and Higgs doublets (Cj 2 ) live on the D52 -brane
in the bulk between the source and matter branes. The gauge-singlet source field in principle can
either live on the D51 -brane or be localised on the 51 52 intersection, but for definiteness we assume
the latter possibility.
Fig. 4. The intersecting D5-brane construction in the limit of small D51 compactification radius.
The D51 -branes reduce to effective D3-branes, separated in two orthogonal dimensions along the
D52 -brane (bulk). The allocation of Higgs and chiral matter fields are the same as in Fig. 3 and
Table 2.
3
(10)
g5
5 2
(where g3 g52 ). This is consistent with gauge coupling unification if g52 gGUT at the
GUT scale.
It is also interesting to note that the restrictions we place on the radii do not restrict
the radius of the third complexified dimension too strongly. This could allow a large extra
dimension felt by gravity alone (with a size of the order of 1 mm) as considered recently
[14], but we will not discuss that possibility here.
In this limiting case, we can use the results of Ref. [4] where we identify L R52 . We
can extend the analysis for the size of the extra dimensions and exponential suppression
factors. Ref. [4] considers the maximum dimension size in the strong coupling limit 1,
but for a small number of extra dimensions, the theory does not need to be strongly coupled
at the string scale, i.e., = 1.
Consider our symmetric toroidal compactification where the volume of the compact
dimensions is
gY
V52 L2 R522 .
(11)
Using Eqs. (2), (3) with D = 6, we can relate dimension size to for a 4-dimensional
gauge coupling of order 1 (as observed for SM couplings).
g522
l6
L2 ,
m2s
Lms (l6 )1/2 .
(12)
85
Note that from Eqs. (11), (12), we have

V52 m2s l6 .
(13)
We have just seen how to recover the gMSB
model, but with two extra dimensions and

the third family in the bulk. We can therefore use the gMSB
results for the operators that

lead to scalar and Higgs masses, A and B-terms and even a -term via the Giudice
Masiero mechanism. 14 However, in our model with M ms and R52 R51 > m1
s , there
are only two extra dimensions in the bulk between D3-branes. 15
We use the Eqs. (4), (5), (6) with the following identifications:

M C 5 1 5 2
Qi , Li , UiC , DiC , EiC , NiC (i = 1, 2),
W WSM ,

hu , hd Cj52

S C 51 52
hu , hd , Q3 , L3 , U3C , D3C , E3C , N3C ,

S
(14)
to generate higher-dimensional operators, subject to the full 42222 gauge invariance. We

assume that the F -component of the gauge-singlet field S, which we assume to be an open
string state on the intersection between the source brane and the mediating brane, acquires
a non-zero vev and breaks supersymmetry. We now proceed to discuss the different types
of masses in the limiting case of the BMSB model.
4.1. Gaugino masses
In the limit of R51 < R52 , the standard model gauge fields are dominated by their
we generate gaugino
components on the D52 -brane (bulk). In agreement with gMSB,
masses of the same order of magnitude from Eqs. (5), (6)
m
FS l6 / l4
1 FS
2
ms ms V 5 2
l4 ms
(15)
(where V52 is the volume of the compact dimensions inside the D52 -brane worldvolume).
4.2. First and second family scalar masses
This is the generic 4-point contact term between fields on opposite branes that leads
to exponentially suppressed first and second family squark and slepton masses, 16 using
Eq. (4).

ems R52
d 4 C 51 52 C 51 52 S S,
Lsoft
(16)
m2s
14 Remember that a superpotential -term is forbidden by string selection rules for our choice of states.
15 This allows us to use Table 3 to get restrictions on the size of R .
52
16 This operator also leads to first and second family mixing and off-diagonal mass matrix elements. There may
be another operator leading to first and third family mixing, e.g., L
4 5 5 52
d C 1 2 Cj S S.
86
Table 3
Estimates for the toroidal compactification length L and exponential suppression factor for D = 6,
where L R52

1
0.8
0.6
0.4
0.2
0.1
0.05
0.01
Vsoft
Lms
eLms /2
63
56
49
40
28
20
14
6
2 1014
6 1013
3 1011
2 109
8 107
5 105
9 104
4 102
ems R52 FS2

iC U
iC D
i
j + U
jC + D
jC + L
i Q
Lj
Q
m2s

C + N
C .
C E
C N
+E
i
j
i
j
(17)
Table 3 shows that the exponential suppression factor is strong for two extra dimensions.
Therefore, contact term contributions to the first and second family scalar masses are
negligible at high energies. Instead, they are generated by renormalization group equation
(RGE) effects.
Loop contributions to first and second family scalar masses (Fig. 1) are much larger
than contact terms and anomaly mediated contributions. So, although the first and second
family squark/slepton masses are not zero at high-energies, they are suppressed by a loop
factor relative to third family scalar masses.
4.3. Higgs mass terms and third family scalar masses
Extending Eq. (5) to include third family scalars, we have the following higherdimensional operators:

l6
1
1
4
S
h
h
+
h.c.
+
S S hu hu + hd hd + (hu hd + h.c.)
d
Lsoft
u
d
l4
m3s
m4s

C C
C C
C C
C C
+ Q3 Q3 + U3 U3 + D3 D3 + L3 L3 + E3 E3 + N3 N3 .
(18)
From Eqs. (6), (18), we obtain the -term,
FS l6 / l4
1 FS
.
2
ms ms V 5 2
l4 ms
(19)
Higgs and third family scalar masses.

B, m2hu , m2hd , m2F
3
FS2 l6 / l4
1 FS2
m2s m2s V52

l4 m2s
3 Q
3 , U
C , D
C ,
C C
(where F
3
3 L3 , E3 , N3 ).
(20)
87
4.4. Scalar mass matrix

We have generated a scalar mass matrix with an explicit third family mass hierarchy at
lowest order:
0 0 0
2
F
1 S
m2scalar
(21)
0 0 0.
l4 m2s
0 0 1
The first and second family mass matrix elements are dominated by loop corrections since
the contact term contributions are exponentially suppressed. However these contributions
are still smaller than the third family masses due to the location of the third family in the
bulk and its direct coupling to the SUSY breaking hidden sector.
4.5. Trilinear A-terms
Gauge invariant operators can be constructed for third family A-terms as follows:

l6
1
d 2 4 S hd D3C Q3 + hu U3C Q3 + hd E3C L3 + hu N3C L3 + h.c. (22)
Lsoft
l4
ms
These operators lead to trilinear A-terms:
0 0 0
0
1
FS l6 / l4
F
S
0
Aij
0
0
0
ms m3 V 3/2
ms l4 (l6 )1/2
s 52
0 0 1
0
0 0
0 0
0 1
(23)
using Eq. (13).

The first and second family A-terms are negligible in comparison to the third family
term. Instead, they will receive loop-suppressed contributions.
4.6. Yukawa textures
Using our choice of states and Eq. (7), we obtain a third family hierarchical Yukawa
texture for the quark and lepton sectors at lowest-order. This texture reflects the observation
that mt mc , mu ; mb ms , md and m > m , me .
0 0 0
Yija 0 0 0 , where a u, d, e, n.
(24)
0 0 1
Smaller NLO Yukawa couplings (and associated trilinear A-terms) are generated by
higher-dimensional operators. Notice that an interesting operator is allowed by 42222
gauge invariance, and appears to be such a small Yukawa term:
i Fi h .
L F
(25)
The fields h, and (Higgs) acquire vevs and spontaneously break the gauge
symmetry. When each field is replaced by its vev, we can generate a first and second family
mass term. This operator will be suppressed by powers of the string scale such that the first
and second family have much smaller masses relative to the third family in the bulk.
88
Table 4
Estimates for the ratio of scalar masses and third family A-terms to gaugino masses for different
and the exponential suppression factor (for masses-squared) arising from toroidal compactification

1.0
0.8
0.6
0.4
0.2
0.1
0.05
0.01
eLms /2
m2 /m2
m /m
A33 /m
2 1014
6 1013
3 1011
2 109
8 107
5 105
9 104
4 102
158
126
95
63
32
16
8
1.6
12.6
11.2
9.7
7.9
5.6
4.0
2.8
1.3
0.016
0.018
0.020
0.025
0.035
0.050
0.071
0.159
4.7. Mass ratios and FCNC constraints

Consider the ratio of Higgs and third family scalar masses B, m2hu , m2hd , m2F to
gaugino masses m2 :
m2
m2
l4 2
m V5 l4
l6 s 2
(26)
(using Eqs. (2), (3) and g4 1, where m2 B, m2hu , m2hd , m2F ).

3
Also consider the ratio of trilinear soft masses A33 to gaugino masses m using
Eqs. (15), (23):
1
A33
.
m
(l6 )1/2
(27)
Experimental constraints on FCNC 17 from mass-squared matrix elements require an

exponential suppression of 103 104 for first and second family scalar masses in
Eq. (17). Using Table 4, we get a lower limit of say 0.01. However, phenomenological
considerations restrict the ratio of m and m , and places an upper limit of say
0.1. This amount of suppression requires that the effective D3-branes are separated by
a distance of order 10/ms .
4.8. Phenomenology
As in [4] we shall consider the phenomenology based on an inverse compactification
scale (R51
in our case) close to the unification scale MGUT 2 1016 GeV. It is natural
2
to assume a high energy unification scale in the limiting case g51 g52 since in this limit
the light physical gauge fields all arise from the mediating 52 -brane, and so are all subject
to a single gauge coupling constant, g52 gGUT .
17 See [4] and references therein.
89
We have seen that in the BMSB model (at MGUT ) the trilinear and first and second
family soft masses are negligible, while the third family soft masses, and the Higgs mass
parameters are larger than the gaugino masses. In Table 5 we compare a sample spectrum
in the BMSB model to that in both the gMSB
model and the no-scale supergravity model,

where the ratio of Higgs vevs tan = 20 and a universal gaugino mass of M1/2 = 300 GeV
are chosen to give a lightest Higgs boson mass of about 115 GeV, consistent with the recent
LEP signal [15,16]. 18
In the no-scale model the only non-zero soft mass is M1/2 , which results in a very
characteristic spectrum where the right-handed slepton is very light and is in danger of
becoming lighter than the lightest neutralino. The gMSB
model differs from the no-scale

model only by the inclusion of Higgs soft masses which we have taken to be degenerate and
somewhat higher than the gaugino masses. The main effect is to reduce the parameter,
which is determined here from the electroweak symmetry breaking condition, and taken to
be positive, which results in lighter charginos and neutralinos. Also in the gMSB
model
the heavy Higgs and third family squark spectrum is also noticeably different from the
no-scale model. 19 Turning to the BMSB model, we see that the effect of having both the
Higgs and third family soft masses is to raise the parameter, and of course to significantly
increase the third family squark and slepton masses, providing an unmistakable spectrum
and a characteristic smoking gun signature of the model.
5. Conclusions
We have proposed a mechanism for mediating SUSY breaking in type I string theories
BMSB. Rather similar to the gMSB
set-up in Fig. 1 we have proposed a type I stringinspired set-up consisting of three intersecting D5-branes as shown in Fig. 3 in which the
gauge fields, Higgs doublets and third family matter fields all live on the third mediating 52 brane which plays the role of the bulk in the gMSB
scenario. The presence of the third matter family on the mediating D-brane is characteristic of type I string constructions and provides the main experimentally testable difference between the BMSB and gMSB
models.
We have considered a limiting case in which R52 R51 , and shown that in this case
the model reduces to the original gMSB
model with the role of the bulk being played by

the mediating 52 -brane. In this limiting case, the model naturally leads to approximately
universal gaugino masses and a single unified gauge coupling constant, which motivates the
identification of the string scale with the usual GUT scale. In this case the phenomenology
of the BMSB model is rather interesting, and it may be compared to the predictions of
the no-scale supergravity and the gMSB
model. As in the gMSB
model, the first two

families naturally receive very small masses at the high energy scale leading to flavour18 Note that tan = 20 is sufficiently small that we may neglect all Yukawa couplings except the top Yukawa
coupling in the RGEs.

19 As noted in [4], if we had taken non-degenerate Higgs soft masses then the lightest right-handed slepton mass
could have been significantly increased relative to the no-scale model due to the hypercharge FayetIlliopoulos
term.
90
Table 5
Comparison of spectra (in GeV) for the three models BMSB, gMSB
and no-scale supergravity. The

common parameters are tan = 20, universal gaugino mass M1/2 = 300 GeV, trilinear soft mass
A0 = 0, first and second family squark and slepton masses m2 = 0. The parameters are chosen to
F1,2
give a lightest Higgs boson mass consistent with the LEP signal [15,16]. The parameter (assumed
positive) and B are determined from the low energy electroweak symmetry breaking conditions
BMSB
gMSB
No-scale
M1/2
A0
mF
1,2
mF
3
mhu
mhd
g
300
0
0
500
500
500
830
300
0
0
0
500
500
830
300
0
0
0
0
0
830
10
124
119
124
20
30
40
1
2
239
200
237
506
258
472
517
314
485
238
195
237
518
220
546
124
515
205
540
740
783
715
628
744
787
713
871
613
832
20
115
738
738
742
500
314
220
220
124
124
205
205
740
653
715
520
744
658
713
713
492
718
20
114
596
596
602
250
486
220
220
124
124
205
205
740
676
715
577
744
681
713
713
544
745
20
115
511
511
517
467
L
E
1,2
L
E
3
R
E
1,2
R
E
3
L
N
1,2
L
N
3
L
U
1,2
L
U
3
R
U
1,2
R
U
3
L
D
1,2
L
D
3
R
D
1,2
R
D
3
t1
t2
tan
mh0
mH 0
mA
mH
(MZ )
91
changing neutral currents being naturally suppressed. The presence of third family soft
masses will not alter this conclusion very much since FCNC limits involving the third
family are much weaker. However the third family soft masses will lead to a characteristic
squark and slepton mass spectrum which may be easily distinguished from that of both
no-scale supergravity and the gMSB
model as shown in Table 5. The -problem is solved

by the GiudiceMasiero mechanism as in the original gMSB
model.
In this limiting case the BMSB model bears a close resemblance to both the no-scale
supergravity and the gMSB
models. The fact that the third family receives a non-zero soft
SUSY breaking mass is strictly not an unambiguous signal of the underlying type I string
model, since it is possible for this to happen in both the other cases also. For example in
the old heterotic models based on orbifolds, matter may be localised in the fixed points of
the orbifold (the twisted sector) or not (the untwisted sector) so it is possible to have the
third family playing a different role from that of the first two families. What is different
in the model presented here is that the gauge group is localised on two different branes,
but in the limiting case (above) the physical gauge group arises essentially from one brane,
and in this limit we return to a situation similar to that of the old heterotic string theories.
There are however three points worth noting here. Firstly, the presence of two families
at the intersection points of two branes, and one family on a single brane, seems to be
typical of type I string constructions [8]. Secondly in type I string constructions we have
the possibility of full unification of both gravity and gauge forces, precisely because gravity
exists in 10 dimensions whereas the gauge groups live in a 6-dimensional submanifold,
which is not possible in old heterotic string theories. Thirdly, the limiting case of R52
R51 would be expected to apply only approximately, and therefore in practice there will be
corrections, for example, to gauge coupling unification which may be observable. Further
comments concerning the non-limiting case are briefly discussed below.
In the more general non-limiting case, the model will have an even richer structure. In
this non-extremal radii limit (i.e., R52 > R51 = m1
s ), we must use the full gauge state expressions listed in Appendix A.1. The light gauge states are no longer dominated by their
D52 -brane components, but are instead mixtures of fields from either brane, with the exception of the gluon/gluino states that only arise from the D52 -brane. The result is that the high
energy gluino mass will be larger than the high energy wino and bino masses. In this more
general case the gauge couplings are no longer equal, so there is less motivation to identify
the string scale with the GUT scale. Generally the string scale can take any value from a
few TeV to 1016 GeV, and we have the possibility of a mm scale large extra dimension.
The toy model has other interesting features such as the fact that the gauge symmetry
forbids first and second family Yukawa couplings at lowest order, and naturally forbids
R-parity violating operators that cannot be forbidden by string selection rules alone, while
allowing the third family Yukawa coupling. Most importantly, however, the toy model
demonstrates the BMSB mechanism, which is based on having at least three branes with
two different intersection points. This minimum requirement implies that constructions
with all the branes at the origin fixed point are inadequate for our purpose. Although
there are examples in the literature of intersecting branes at different fixed points [17],
such models are generally more complicated than the simple set-up considered here.
92
Nevertheless our BMSB mechanism could provide a useful alternative starting point from
which to address the problem of SUSY breaking in more general type I string theories.
Finally note that it has been been suggested, in the context of type I theories, that singlet
twisted moduli, which appear in the tree-level gauge kinetic function, might be responsible
for generating gaugino masses if they acquire non-vanishing F -terms, and that this might
provide a brane realisation of gMSB
if the standard model gauge symmetry originates from

9-branes providing that there are in addition two sets of D5-branes located at two different
fixed points [18]. This suggestion shares some of the features with the present paper,
although model building issues were not discussed, and the characteristic possibility of
the third family on the mediating brane was not considered. Also additional contributions
from the F -terms of dilaton S and moduli fields Ti were also generically allowed, whereas
here we have implicitly assumed them to be absent.
Acknowledgements
S.K. and D.R. would like to thank PPARC for a Senior Fellowship and a Studentship.
S.K. also acknowledges very illuminating discussions with Z. Chacko and Jing Wang at
the Aspen Center for Physics and is particularly indebted to Lisa Everett and Gordy Kane
at the University of Michigan for detailed discussions concerning D-brane models and to
Lisa Everett especially for sharing her expertise with me. We are also grateful to K. Benakli
for pointing out the existence of [18] which we were ignorant of until after our paper was
submitted to the archive.
Appendix A. Spectrum of gauge bosons
A.1. General case
In this appendix we consider the effect of symmetry breaking on massless gauge
field states and gauge couplings. We begin with the gauge group SU(4)52 SU(2)51L
SU(2)52L SU(2)51R SU(2)52R . The couplings run with energy scales subject to RGEs.
Conventionally, the symmetry breaking all occurs at high energies (10151016 GeV) except
for SU(2)L U (1)Y U (1)EM which happens at the electroweak scale. In Tables 612
gauge couplings are assumed to be at high energies unless otherwise stated. Notice that i,
a and m are adjoint indices for SU(2), SU(3) and SU(4), respectively.
Table 6
The initial gauge groups, gauge couplings and states in our model
Gauge group
Coupling
States
SU(4)52
SU(2)51L
SU(2)52L
SU(2)51R
SU(2)52R
g5 2
Gm
5
g5 1
W5i L
g5 2
W5i L
g5 1
W5i R
g5 2
W5i R
93
(a) First combine the chiral SU(2) groups from either brane via diagonal symmetry
breaking to recover the PatiSalam gauge group.
SU(2)51L/R SU(2)52L/R
diagonal v ,v
SU(2)L/R
SU(4)52 SU(2)L SU(2)R GP S .
(A.1)
Spontaneous symmetry breaking (SSB) induces a change of basis, parametrised by

cos =
g52
g521 + g522
(A.2)
We can express the new massless states and gauge couplings in terms of the original
parameters. The Higgs mechanism generates massive gauge bosons with masses of the
order of the symmetry breaking scale.
Table 7
The new massless states and couplings after the original gauge symmetry is broken down to the
PatiSalam gauge group
Gauge group
SU(4)52
Coupling
g5 2
States
Gm
5
SU(2)L
SU(2)R
gL =
i
WL/R
=
g5 1 g5 2
g52 + g52
1
2
= gR

g51 W5i L/R + g52 W5i L/R
2
1
g52 + g52
1
L ) and 3 massive SU(2)R (W
R ) bosons, of mass
plus 3 massive SU(2)L (W

1
= v2 g521 + g522 .
L/R
2
(b) QCD SU(3)C is contained within SU(4)52 . The U (1)s combine to give the
hypercharge U (1) using the relationship Y = (B L) + 2IR .
2
MW
SU(4)52 SU(3)C U (1)BL ,
SU(2)R U (1)IR .
(A.3)
The PatiSalam gauge group is broken down to the standard model by giving a vev to a
Higgs field H .
vH
U (1)BL U (1)IR U (1)Y SU(3)C SU(2)L U (1)Y .

The change of basis is parametrised by

3
3(g52 + g52 )
2 g52
2
cos H =
= 21
2
3 2
2
5g
+
3g
51
52
gR + 2 g52
(A.4)
(A.5)
94
Table 8
The standard model massless states and gauge couplings expressed in terms of the original
parameters
Gauge
group
SU(3)C
Coupling
g5 2
States
Ga5
2
U (1)Y
SU(2)L
gL =
WLi =
g5 1 g5 2
g5 g5 3
gY = 1 2
5g52 + 3g52
1
2
g52 + g52
1
2
g51 W5i L + g52 W5i L

1
2
g52 + g52
1
BY =
3(g51 W53
+ g52 W53 ) +
1R

5g52 + 3g52
2R
2g51 G15
5
2
1 2 2
2
plus 6 massive SU(4)52 bosons (G952 G14
52 ), mass MG = 4 vH g52 ; 2 massive SU(2)R
1 2 2
2
bosons (WR ), mass MW
= 4 vH gR , and 1 massive SU(2)BL boson (XBL ), mass
R
2 (g 2 + 3 g 2 ).
MX2 BL = 14 vH
R
2 52
(c) Finally, we can recover the QCD and EM standard model gauge group via the familiar
low-energy Higgs mechanism, parametrised by

2
5g5 (vh ) + 3g52 (vh )
gL (vh )
2
.
= 21
cos W =
2 (v )
8g
(v
)
+
6g
2

2
h
51
52 h
gL (vh ) + gY (vh )
Electroweak symmetry breaking occurs when the Higgs field h acquires a non-zero vev.
vh
SU(2)L U (1)Y U (1)EM SU(3)C U (1)EM
(A.6)
Table 9
The massless gauge states and couplings after electroweak symmetry breaking
Gauge
group
SU(3)C
Coupling
g52 (vh )
U (1)EM
g (vh )g52 (vh ) 3
5
e = 12
8g5 (vh )+6g52 (vh )
States
Ga5
A=
plus 3 massive SU(2)L bosons (WL , ZL0 ) with masses:

1
g5 (vh )g52 (vh )vh
MW = vh gL (vh ) = 1
L
2
2 g521 (vh ) + g522 (vh )
and
3 g51 (vh )(W53 L +W53 R )+ 3 g52 (vh )(W53 L +W53 R )+ 2 g51 (vh )G15
52
2
2
1
1
8g52 (vh )+6g52 (vh )
MZ 0

= g51 (vh )g52 (vh )vh
95
4g521 (vh ) + 3g522 (vh )

2(g521 (vh ) + g522 (vh ))(5g521 (vh ) + 3g522 (vh ))
A.2. Limiting case R52 R51

In this appendix we repeat the symmetry breaking analysis for the limiting case
R52 R51 g52 g51 .
(A.7)
We find that the dominant components of the massless gauge fields live on the 52 -brane
(bulk) which is consistent with gMSB.
(a) After diagonal symmetry breaking we recover the PatiSalam gauge group
SU(4)52 SU(2)L SU(2)R
Table 10
The dominant components of massless states and couplings after symmetry has been broken down
to the PatiSalam group
Gauge group
SU(4)52
SU(2)L
SU(2)R
Coupling
g5 2
gL/R g52
States
Gm
5
i
WL/R
W5i L/R
2
L/R W i
plus 3 massive SU(2)L and 3 massive SU(2)R bosons (W
51 L/R ),
1
v2 g521 .
2
(b) We break the PatiSalam group down to the standard model. Notice the relationship
between the hypercharge gauge coupling and the other gauge couplings, which is consistent
with gauge coupling unification. This will happen if the 52 gauge coupling equals gGUT at
the GUT scale.
2
MW
L/R
Table 11
The dominant components of the massless states and couplings after the PatiSalam group is broken
down to the standard model
Gauge group
SU(3)C
SU(2)L
Coupling
g5 2
gL g5 2
States
Ga5
WLi W5i L
2
U (1)Y

gY 35 g52

BY 35 W53 + 25 G15
5
2R
1 2 2
2
plus 6 massive SU(4)52 bosons (G952 G14
52 ), MG 4 vH g52 ; 2 massive SU(2)R bosons
2 1 v 2 g 2 ; and 1 massive SU(2)
(WR 1 (W512 R iW522 R )), MW
BL boson (XBL
4 H 52
2
R
5 2 2
15
3
2
3/5 G52 2/5 W52R ), MXBL 8 g52 vH .
96
(c) Finally the Higgs mechanism induces electroweak symmetry breaking, and generates
the massive W and Z bosons.
Table 12
The dominant components of the familiar massless gauge states after electroweak symmetry
Gauge group
SU(3)C
Coupling
g52 (vh )
Ga5
States
U (1)EM

e 38 g52 (vh )

A 38 W53 L + W53 R + 12 G15
52
2
2
plus 3 massive SU(2)L bosons: (WL 1 (W512 L iW522 L )), MW 12 vh g52 (vh ) and
2
L
(ZL0 5/8 W532 L 3 W532 R 12 3/5 G15

52 ), MZ 0 2/5 vh g52 (vh ).
2 10
Appendix B. Mass scales

In this appendix we consider the different mass scales present in the model. Each time
the gauge symmetry is spontaneously broken down towards the standard model, the broken
generators have massive gauge bosons associated with them. These bosons have masses of
the same order as the symmetry breaking scale, i.e., the vevs of the breaking fields. Our
model already assumes an order for symmetry breaking, which creates a vev hierarchy
(v vH vh O(MEW )). For instance, we know that v , vH O(MEW ) since these
broken symmetry bosons have not been observed.
We must also consider the (inverse) compactification radii of the D5-branes. Their
relative sizes are arbitrary, but we choose to start with the relationship R52 > R51 or
> R51
as shown in Fig. 5.
equivalently R51
1
2
Fig. 5. At energy scales below an inverse compactification radii, the dimension appears too small to
observe. The coupling in a higher-dimension is related to the same coupling in a lower dimension
via Eq. (2).
97
Table 13
Possible ordering of symmetry breaking vevs and inverse compactification radii within the
constraints of Eq. (B.1)
vh
A
B
C
vH
vh
vH
vh
R51
2
R51
2
vh
vh
vh
vH
R51
2
R51
2
R51
2
vH
R51
2
R51
1
R51
1
vH
R51
1
vH
R51
1
R51
1
ms
ms
R51
1
ms
ms
ms
ms
Notice that we have not specified how R53 is related to the other two compactification
radii, suffice to say that a large third dimension (felt by gravity alone) is not forbidden, i.e.,
> R51
R51
. (See [14] for discussion of large extra dimensions.)
R51
1
2
3
In this work, we have adopted the standard scenario with symmetry breaking occurring
at a scale comparable to the first two compactification radii and string scale. Soft masses
are also generated at around the same scale. We have deliberately not specified these scales,
but we claim that the formalism applies for GUT/string scales of 1 TeV to 1016 GeV.
We impose the following restrictions:
> R51
,
R51
1
2
v vH vh ,
, R51
v , vH vh O(MEW ).
ms R51
1
2
(B.1)
These constraints provide six ways of ordering the inverse radii and vevs. The supersymmetry breaking scale (where soft masses are generated) also needs to be assigned, thus
giving a total of 30 possibilities.
In Table 13 we list the various possibilities for the relative ordering of mass scales,
vevs and inverse compactification radii within the constraints of Eq. (B.1). However
it is important to notice that when the inverse compactification radii are less than the
symmetry breaking vevs, the KaluzaKlein modes associated with the extra dimensions
can contribute to the running of gauge couplings, leading to a power law dependence [9].
References
[1] M. Dine, W. Fischler, M. Srednicki, Nucl. Phys. B 189 (1981) 575;
S. Dimopoulos, S. Raby, Nucl. Phys. B 192 (1981) 353;
L. Alvarez-Gaum, M. Claudson, M.B. Wise, Nucl. Phys. B 207 (1982) 96;
M. Dine, A.E. Nelson, Phys. Rev. D 48 (1993) 1277, hep-ph/9303230;
M. Dine, A.E. Nelson, Y. Shirman, Phys. Rev. D 51 (1995) 1362, hep-ph/9408384;
M. Dine, A.E. Nelson, Y. Nir, Y. Shirman, Phys. Rev. D 53 (1996) 2658, hep-ph/9507378;
H. Murayama, Phys. Rev. Lett. 79 (1997) 18, hep-ph/9705271;
98
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
S. Dimopoulos, G. Dvali, R. Rattazzi, G.F. Giudice, Nucl. Phys. B 510 (1998) 12, hepph/9705307;
M.A. Luty, Phys. Lett. B 414 (1997) 71;
For a review, see, G.F. Giudice, R. Rattazzi, Phys. Rep. 322 (1999) 419, hep-ph/9801271;
G.F. Giudice, R. Rattazzi, Phys. Rep. 322 (1999) 501.
L. Randall, R. Sundrum, Nucl. Phys. B 557 (1999) 79, hep-th/9810155;
G.F. Giudice, M.A. Luty, H. Murayama, R. Rattazzi, JHEP 9812 (1998) 027, hep-ph/9810442.
D.E. Kaplan, G. Kribs, M. Schmaltz, Phys. Rev. D 62 (2000) 035010, hep-ph/9911293.
Z. Chacko, M. Luty, A.E. Nelson, E. Pontn, JHEP 0001 (2000) 003, hep-ph/9911323.
J. Ellis, K. Enqvist, D.V. Nanopoulos, Phys. Lett. B 147 (1984) 99;
J. Ellis, C. Kounnas, D.V. Nanopoulos, Nucl. Phys. B 247 (1984) 373.
P. Horava, E. Witten, Nucl. Phys. B 460 (1996) 506, hep-th/9510209;
E. Witten, Nucl. Phys. B 471 (1996) 135, hep-th/9602070;
P. Horava, E. Witten, Nucl. Phys. B 475 (1996) 94, hep-th/9603142.
G.F. Giudice, A. Masiero, Phys. Lett. B 206 (1988) 480.
G. Shiu, S.-H. Henry Tye, Phys. Rev. D 58 (1998) 106007, hep-ph/9805157.
K.R. Dienes, E. Dudas, T. Gherghetta, Nucl. Phys. B 537 (1999) 47, hep-ph/9806292;
I. Antoniadis, Phys. Lett. B 246 (1990) 377.
Z. Kakushadze, Nucl. Phys. B 535 (1998) 311, hep-th/9806008.
L.E. Ibez, C. Muoz, S. Rigolin, Nucl. Phys. B 553 (1999) 43, hep-ph/9812397.
L. Everett, G.L. Kane, S.F. King, JHEP 008 (2000) 012, hep-ph/0005204.
S.F. King, M. Oliveira, Phys. Rev. D 63 (2001) 095004, hep-ph/0009287.
N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 429 (1998) 263;
I. Antoniadis, N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 436 (1998) 257, hepph/9804398.
The ALEPH Collaboration, Phys. Lett. B 495 (2000) 1, hep-ex/0011045;
The L3 Collaboration, Phys. Lett. B 495 (2000) 18, hep-ex/0011043.
G.L. Kane, S.F. King, L.-T. Wang, hep-ph/0010312;
J. Ellis, G. Ganis, D.V. Nanopoulos, K.A. Olive, Phys. Lett. B 502 (2001) 171, hep-ph/0009355.
M. Cvetic, M. Plumacher, J. Wang, JHEP 0004 (2000) 004, hep-th/9911021;
M. Cvetic, A.M. Uranga, J. Wang, hep-th/0010091.
K. Benakli, Phys. Lett. B 475 (2000) 77, hep-ph/9911517.

Supersymmetry and finite radiative electroweak

breaking from an extra dimension
A. Delgado, M. Quirs
Instituto de Estructura de la Materia (CSIC), Serrano 123, E-28006 Madrid, Spain
Received 8 March 2001; accepted 3 May 2001
Abstract
A five-dimensional N = 1 supersymmetric theory compactified on the orbifold S 1 /Z2 is
constructed. Gauge fields and SU(2)L singlets propagate in the bulk (U -states) while SU(2)L
doublets are localized at an orbifold fixed point brane (T -states). Zero bulk modes and localized states
constitute the MSSM and massive modes are arranged into N = 2 supermultiplets. Superpotential
interactions on the brane are of the type U T T . Supersymmetry is broken in the bulk by a Scherk
Schwarz mechanism using the U (1)R global R-symmetry. A radiative finite electroweak breaking is
triggered by the top-quark/squark multiplet T propagating in the bulk. The compactification radius R
is fixed by the minimization conditions and constrained to be 1/R 1015 TeV. It is also constrained
by precision electroweak measurements to be 1/R 4 TeV. The pattern of supersymmetric mass
spectrum is well defined. In particular, the lightest supersymmetric particle is the sneutrino and the
2.
next to lightest supersymmetric particle the charged slepton, with a squared-mass difference MZ
The theory couplings, gauge and Yukawa, remain perturbative up to scales E given, at one-loop, by
ER 3040. Finally, LEP searches on the MSSM Higgs sector imply an absolute lower bound on
the SM-like Higgs mass, around 145 GeV in the one-loop approximation. 2001 Elsevier Science
B.V. All rights reserved.
PACS: 04.50.+h; 11.10.Kk; 11.30.Pb; 12.60.Jv
1. Introduction
The Higgs boson is, for the time being, the only missing ingredient of the Standard
Model (SM) of electroweak and strong interactions and, by far, the most intriguing one.
While it is related to the origin of gauge boson and fermion masses, the mechanism of
electroweak breaking is intimately related to the so-called hierarchy problem which has
Work supported in part by CICYT, Spain, under contract AEN98-0816, and by EU under contracts HPRNCT-2000-00152 and HPRN-CT-2000-00148.
E-mail address: mariano@makoki.iem.csic.es (M. Quirs).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 2 2 1 - 8
100
A. Delgado, M. Quirs / Nuclear Physics B 607 (2001) 99116
given rise to the (minimal) supersymmetric extension (MSSM) of the SM. In particular the
radiative corrections to the squared Higgs mass in the SM have a quadratic sensitivity to the
cutoff of the theory, s , which destabilizes the Higgs mass towards the region where the
SM is no longer reliable [1]. This behaviour is softened in the MSSM where the sensitivity
to the SM cutoff is only logarithmic and can therefore be interpreted as the renormalization
group running from the scale s to the weak scale [2]. Actually, one of the great successes
of the MSSM is that the squared Higgs mass term can be driven by radiative corrections
generated by the top Yukawa coupling from positive values at the scale s to negative
values at the weak scale thus triggering radiative electroweak breaking [3]. Still the MSSM
shows some (logarithmic) sensitivity to the cutoff scale s , whose value controls the total
evolution of the squared Higgs mass.
The sensitivity of the squared Higgs mass term on the cutoff s through radiative
corrections can still be softened if the MSSM, and in particular the top/stop sector, is living
in the bulk of an extra dimensional space of size O(TeV1 ) [4]. In that case the radiative
corrections to the squared Higgs mass term are not sensitive at all to s . In fact they are
finite, controlled by the inverse radius 1/R of the compactified extra dimensions, 1 and
with a sign which depends on the spin of the bulk particle circulating in the loop [57].
This observation gave rise to proposing the top/stop (hyper)multiplet living in the bulk [7]
as the source of a finite electroweak radiative breaking, 2 while some explicit examples
along that direction have been recently proposed in the frameworks of string [8] and field
theory in higher dimensions [9,10].
Another issue which is highly related to the hierarchy problem and electroweak breaking
is that of supersymmetry breaking. The scale of supersymmetry breaking must not be
hierarchically different from the weak scale since, on the one hand, we do not want to recreate the hierarchy problem, and on the other hand, it should trigger electroweak breaking.
Because, in the finite radiative electroweak breaking, the weak scale is provided (apart from
loop factors) by the inverse radius 1/R of the compactified extra dimensions, that is the
expected order of magnitude for the scale of supersymmetry breaking. Although there are
several mechanisms in the literature which can provide the correct order of magnitude for
supersymmetry breaking, the one that naturally leads to supersymmetry breaking size of
order 1/R is the ScherkSchwarz (SS) mechanism [1114]. In fact both recent examples of
finite radiative electroweak breaking [9,10] use, among other mechanisms, a variant of the
SS-mechanism based on a discrete symmetry of the supersymmetric theory, the R-parity.
In this paper we will analyze a very simple five-dimensional (5D) model where
finite electroweak breaking is triggered by the top/stop multiplet living in the bulk
of the extra dimension and supersymmetry is broken by a SS-mechanism based on a
continuous symmetry of the 5D supersymmetric theory, SU(2)R . So, unlike Refs. [9,10]
supersymmetry breaking is controlled by a continuous parameter, and the supersymmetric
limit is continuously attainable. In this sense, and although the theoretical setup of our 5D
1 This very well known fact in ordinary field theory at finite temperature T , i.e., compactified on the circle of
inverse radius T , is at the origin of the so-called thermal (Debye) masses.

2 See footnote 11 in Ref. [7].
101
theory is rather different from those presented in Refs. [9,10], our results can be considered
in some aspects as more general than theirs.
The outline of this paper is as follows. In Section 2 we will present the model and
the mechanism of supersymmetry breaking. Finite radiative electroweak breaking will
be analyzed in Section 3 and the Higgs sector and supersymmetric spectrum will be
presented in Sections 4 anf 5, respectively. In Section 6 a discussion on unification and
non-perturbativity scales will be done and some comments concerning the relation of our
paper with Refs. [9,10] will be made. Finally in Section 7 we will present our conclusions
and comparison with recent related works.
2. The 5D MSSM and supersymmetry breaking

In this section we will describe a 5D N = 1 model whose massless modes constitute
the usual four-dimensional (4D) N = 1 MSSM, where supersymmetry breaking is a bulk
phenomenon induced by the SS-mechanism, and with finite radiative electroweak breaking
triggered by the presence of a bulk top/stop hypermultiplet.
The 5D spacetime is compactified on M4 S 1 /Z2 , where the Z2 parity is acting on the
fifth coordinate as x5 x5 . The orbifold S 1 /Z2 has two fixed points at x5 = 0, R and
the compactified space has two 3-branes located at the fixed points of the orbifold. In this
way fields in the theory can be of two types: those living in the 5D bulk, similar to untwisted
states in the heterotic string language (U -states), and those living on the branes localized
at the fixed points of the orbifold, similar to the heterotic string twisted states (T -states).
We will assume for simplicity that T -states are localized at the x5 = 0 fixed point. While
U -states feel the fifth dimension, i.e., their wave function depend on x5 , T -states do not.
Vector fields live in the bulk and they are in N = 2 vector multiplets in the adjoint
representation of the gauge group SU(3)SU(2)L U (1)Y , V = (A , 1 ; , 2 ). 3 Matter
L , L ). 4 Z2 , the
R , R ;
fields in the bulk are arranged in N = 2 hypermultiplets, H = (
parity in the fifth dimension, has an appropriate lifting to spinor and SU(2)R indices, such
R , R ), and odd, (, 2 ) and
that we can decompose V and H into even, (A , 1 ) and (
L , L ), N = 1 superfields. 5 After the Z2 projection the only surviving zero modes are
(
R , R ) is not in a real
in N = 1 vector and chiral multiplets. If the chiral multiplet (
representation of the gauge group, a multiplet of opposite chirality localized in the 4D
L , L ) can be introduced to cancel anomalies. In order to do that Z2 must have
boundary (
a further action on the boundary under which all boundary states are odd [15]. This action
can be defined as (1)i for the chiral multiplet Xi , such that i = 1 (0) for Xi living in
the brane (bulk), which creates a selection rule for superpotential interactions on the brane.
The MSSM is then made up of zero modes of fields living in the 5D bulk and chiral
N = 1 multiplets in the 4D brane, at localized points of the bulk. A superpotential
3 Where i = 1, 2 transform as SU(2) indices and the complex scalar is defined as, + iA .
R
5
4 In fact, (
R ,
L ) transforms as a doublet under SU(2)R .
5 Of course the above lifting on hypermultiplets is arbitrary and we could equally well consider (
R , R ) as
L , L ) as even.
odd and (
102
interaction can only exist at the brane of the type U T T or U U U to satisfy the
orbifold selection rule [16], although the latter, U U U , are expected to have Yukawa
couplings which are suppressed with respect to those in U T T by a factor (Rs )1 , and
corresponding to the fact that localized couplings of states propagating in the bulk must be
(volume) suppressed.
To allow for the MSSM superpotential on the brane,
W = [hU QH2 U + hD QH1 D + hE LH1 E + H2 H1 ](x5 ),
(2.1)
and to be consistent with the orbifold selection rule and with a top/stop hypermultiplet
propagating in the bulk, to trigger a finite electroweak radiative breaking, the only solution
is that the SU(2)L singlets, U, D, E propagate in the bulk. 6 In this case the SU(2)L doublets
Q, L, H2 , H1 are localized on the brane and transform as chiral N = 1 multiplets. 7
Since both Higgs fields are on the brane, the -parameter does not arise through
compactification [7] and, although it is an allowed term in the superpotential, one has to
consider it as an effective parameter and rely on its generation as a result of the integration
of the massive states of the underlying (supergravity or string) theory. 8 In this sense the
situation is no better than in the usual MSSM, except for the fact that the cutoff of the
theory, s , the scale at which the structure of the underlying theory should be considered,
is at most two orders of magnitude larger that the scale of supersymmetry breaking and
therefore a modest suppression should be sufficient for phenomenological purposes.
After compactification on S 1 /Z2 the zero mode of the vector multiplet V (0) is just the
(0) (0)
4D MSSM N = 1 vector multiplet (A , 1 ), while the massive modes are the massive
(n) (n)
(n)
(n)
(n)
4D N = 2 vector multiplets (A(n)
, 1 , 2 , ), with a mass n/R, where 1 , 2 are
(0)
Majorana spinors. Similarly the zero mode of matter hypermultiplets H is the 4D N = 1
(0) , (0) ), while the massive modes are 4D N = 2 hypermultiplets,
chiral multiplet (
R
R
(n) ,
(n) , (n) ), with a mass n/R, where (n) is a Dirac spinor with components
(
R
L
(n)
(n)
(R , L ). Finally, the chiral fields which are localized on the brane are massless, except
for the supersymmetric mass term introduced in the superpotential (2.1) that gives a
common mass to Higgs bosons and higgsinos.
Supersymmetry breaking was performed in Refs. [6,7,14] using the SS-mechanism
based on the subgroup U (1)R which survives after the orbifold action S 1 /Z2 . Using the
R-symmetry U (1)R with parameter to impose different boundary conditions for bosons
and fermions inside the 5D N = 1 multiplets, one obtains for the nth KaluzaKlein (KK)
6 Here we assume that all three generations propagate in the same way. Otherwise they can produce an acute
flavor problem and trigger strong CP violation, which translate into strong bounds on the scale 1/R [17].
7 There is technically speaking another possibility: that matter doublets Q, L are living in the bulk and matter
singlets and Higgs doublets U , D, E, H2 , H1 are localized on the brane. This possibility looks less natural and
will not be explicitly considered in this paper, although it leads to results very similar to those that will be found.
It also leads to a drawback in the supersymmetric spectrum concerning the lightest supersymmetric particle, as
we will comment later on.
8 Another possibility, that has been recently pointed out in Ref. [10], is having a singlet field S in the bulk
acquiring a vacuum expectation value (VEV) by radiative corrections induced by another field, in a similar way to
the one by which the top/stop sector makes the Higgs field to acquire a VEV in this paper. Of course this situation
requires enlarging the MSSM to the NMSSM and the Higgs sector gets mixed with the singlet states [18].
103
mode of gauge bosons and chiral fermions living in the bulk the compactification mass
n/R, while for the nth mode of gauginos, (n) , and supersymmetric partners of chiral
(n) ,
(n) the mass (n + )/R. Notice that for the particular
fermions living in the bulk,
R
L
case = 1/2 the gauginos (n) and (n+1) , n > 0, are degenerate in mass and constitute
a Dirac fermion. A detailed discussion was done in Ref. [14] for the case of gauginos. For
L ), that transform under the subgroup U (1)R we
R ,
matter scalars in hypermultiplets, (
can write the SS boundary conditions as,

R
cos x5 /R
R
sin x5 /R
(2.2)
=
,

L
L
sin x5 /R cos x5 /R
where R (L ) are even (odd) periodic functions R,L (x5 ) = R,L (x5 + 2R). Making a
(n)
we can write:
Fourier expansion along the x5 direction with coefficients R,L
1
( + n)x5 (n)
R =
,
cos
R
R
2 n=
L =

1
( + n)x5 (n)

sin
,
L
R
2 n=
(2.3)
where

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

1 (n)
(n)
(n)

= R + L(n) ,
L
R
2
n 0,
n 0,
(2.4)
are the mass eigenstates modes.

In this way the MSSM states, made out of bulk zero modes and localized states, acquire
tree-level masses: gauge bosons, right-handed fermions and left-handed chiral multiplets
localized on the brane are massless; gauginos and right-handed sfermions are massive,
with masses /R; Higgs bosons and higgsinos, localized on the brane, are massive with
a common supersymmetric mass .
Since the Higgs bosons have a positive squared mass 2 , electroweak symmetry is
preserved at the tree-level. However, we will see in the next section how this tree-level
mass pattern, along with the Yukawa couplings contained in the superpotential (2.1), and
in particular the top Yukawa coupling ht , will be able to break radiatively the electroweak
symmetry and provide a well defined spectrum for the Higgs masses and the masses of the
supersymmetric partners localized on the brane.
3. Radiative electroweak breaking

The Higgs potential along the direction of the neutral components of the fields H2 =
h2 + i2 and H1 = h1 + i1 can be written as,
V (H1 , H2 ) = m21 |H1 |2 + m22 |H2 |2 + m23 (H1 H2 + h.c.)
+
2
g2 + g 2
|H1 |2 |H2 |2 + t |H2 |4 + b |H1 |4 ,
8
(3.1)
104
Fig. 1. One-loop diagrams contributing to m23 .
where the supersymmetric tree-level relations m1 = m2 = and m23 = 0, t = b = 0 hold.

These relations are spoiled by radiative corrections which provide contributions to all the
above parameters. These corrections are driven by the SU(2)L U (1)Y gauge couplings
g and g , and by the top and bottom Yukawa couplings, defined as:

1 + t2
mt
mb

,
h
=
1 + t2 ,
ht =
(3.2)
b
v
v
t2
where
t tan v2 /v1 , vi = Hi are the vacuum expectation values of the Higgs fields,
v = v12 + v22 = 174.1 GeV, and mt and mb are the top and bottom running masses. Notice
that hb can become important only for large values of t , as those that will be found by
minimization of the one-loop effective potential. We will consider the leading radiative
corrections: in particular g 2 -corrections to the quadratic terms which are zero at the treelevel (m23 ) and h4t,b corrections to the quartic terms which are O(g 2 ) at the tree-level. For
this reason we will not consider any radiative term as (H1 H2 )2 in Eq. (3.1) and neglect
g 2 -radiative corrections in the numerical analysis.
All radiative corrections to the potential parameters in (3.1) will depend on 1/R and .
In particular the one-loop radiative corrections to any scalar localized on the brane were
computed in Ref. [7] where the corresponding diagrams were identified. 9 A simple
application to the Higgs mass terms m21 and m22 yields:
6h2 3g 2 ()
6h2t 3g 2 ()
(3.3)
,
m21 = 2 b 4
,
4
2
32
R
32
R2
where the function is defined as

() = 2 (3) Li3 (r) + Li3 (1/r) ,
(3.4)
k n
r = exp(2i), and Lin (x) = k=1 x /k is the polylogarithm function of order n.
The mass term m23 in (3.1) is generated by the one-loop diagram exchanging KK-modes
1,2 , as shown in Fig. 1.
of gauginos, (n) and localized higgsinos, H
The resulting contribution is given by,
m22 = 2
m23 =

3g 2
i Li2 (r) i Li2 (1/r) .
2
512 R
9 See Figs. 2 and 3 of Ref. [7].
(3.5)
105
Fig. 2. One-loop diagrams contributing to t .
Notice that, for the particular case = 1/2 (r = 1), m23 = 0, reflecting the fact that the
gauginos (n) are, in that case, Dirac fermions, as it was already mentioned.
The quartic couplings t,b are generated at the one-loop level by loop diagrams
(n)
(n)
exchanging KK-modes,
tR and
bR , and localized modes,
tL and
bL . The diagrams
contributing to t are shown in Fig. 2. The contribution to b being similar, just changing
H2 H1 and t b.
Notice that in Fig. 2 we are sometimes propagating the modes
tL(n) . They correspond, in
our notation, to the odd modes of the hypermultiplets (
tR(n) , tR(n) ;
tL(n) , tL(n) ) that can couple
to the brane through 5 couplings, and should not be confused with the localized multiplets
(
tL , tL ). The resulting expression is,

2
3h4t
1
r 1 log(1 r) + log(1 1/r)
t = 2 1 + log 2RMZ
8
4(r 2 1)

2
+ 1 + r Li2 (1/r) Li2 (r) ,
(3.6)
and a similar expression for b just changing ht hb .
Finally we have as free parameters, 1/R, , and t . Two of them will be fixed by the
minimization conditions Vh 1 = Vh 2 = 0 which read as

1 2
1 2 1
1 2h2t g 2 ()
2
2
2
2
M
M
=
+
+
2
v
+
,
3 1+ 2
t
A
32 4
R2
2 Z
2 Z t2
t

2 + 12 MZ2 + 2t v 2 + 2 m2A 12 MZ2 /t2
2h2t g 2
=
(3.7)
,

2h2b g 2 2 m2A 12 MZ2 + 2 12 MZ2 + 2b v 2 /t2
106
Fig. 3. Plots of 1/R, in TeV (left panel) and t (right panel), as functions of and , in TeV, as given
from the minimization conditions (3.7).
where
m2A = m23
1 + t2
t
(3.8)
is the mass of the CP-odd Higgs once the minimization conditions (3.7) have been used.
In fact we have chosen to select 1/R and t as functions of the other variables. The
corresponding plots are shown in Fig. 3.
We can see from Fig. 3 that the minimization conditions impose a solution with large
tan (t 3540) and values of the compactification scale going from a few TeV to
1015 TeV, depending on the values of and .
SM precision measurements settle bounds on electroweak observables which, for higher
dimensional models with gauge fields living in the bulk of the extra dimension, translate on
lower bounds on the compactification scale 1/R. The model we are studying was analyzed
in Ref. [17], where a very general class of models was considered. We found that, for large
values of t , the lower bound on 1/R is 4 TeV which is in the ballpark provided by
Fig. 3.
4. The Higgs mass spectrum

The neutral Higgs sector has one CP-odd and two CP-even scalar bosons. The mass
of the CP-odd Higgs boson was already given in Eq. (3.8). The squared mass matrix for
neutral CP-even scalar bosons is given by,

2 2
(m2A + MZ2 )s c
mA s + (MZ2 + 4b v 2 )c2
2
.
M0 =
(4.1)
(m2A + MZ2 )s c
m2A c2 + (MZ2 + 4t v 2 )s2
In the large t -limit the two eigenvalues are:
2
= m2A ,
MH
Mh2 = MZ2 + 4t v 2 ,
(4.2)
where h is the SM-like Higgs and H the Higgs with non-SM couplings, while the mass of
the charged Higgs H is given by
107
Fig. 4. Plots of Mh,H , in GeV (left panel) and MH , in GeV (right panel), as functions of and ,
in TeV.
2
2
2
MH
= mA + MW .
(4.3)
The masses of the neutral CP-even and charged Higgs bosons are shown in Fig. 4.
In the left panel of Fig. 4, Mh is the flat surface while MH corresponds to the steepest
one. The crossing is characteristics of the large t solution. For large values of , Mh
is the lightest (SM-like) Higgs mass. Notice that the mass Mh is not controlled by the
compatification scale 1/R, but only by the weak scale v. This is a reflection of a similar
behaviour in the MSSM where the lightest SM-like Higgs mass is not controlled by the
supersymmetry breaking scale. On the other hand, for small values of , MH corresponds
to the lightest Higgs mass. Its mass is controlled by the compactification scale.
To make contact with Refs. [9,10] we can fix = 1/2. In that case the m23 -term is not
generated, as we said previously, and mA = 0, unless we introduce an additional term like
(/s )(H1 H2 )2 in the superpotential, giving rise to an m23 -term as v 2 /s . In that
case the Higgs spectrum depends on the parameters and , with some sensitivity on the
cutoff s . We recover the results of Ref. [9] in the limit , 0, in which case we obtain
Mh 128 GeV, in agreement with the result in [9]. However, in general, for = n/2 the
PQ invariance is broken by the gaugino masses and there is no need for the -term in the
superpotential.
Finally we will comment on the constraints imposed on our model from the Higgs
searches at LEP. Preliminary results of last year run show a lower limit on the SMHiggs mass about 113 GeV. On the other hand, an excess of candidates for the process
e+ e Z Zh has been reported by the ALEPH and L3 Collaborations for center-of
mass energies s > 206 GeV, for a SM-Higgs with a mass around Mh 115 GeV [19],
Now the question is whether this mass can be
which decays predominantly into bb.
accommodated in our model. A quick glance at Fig. 4 (left panel) shows that such low
masses should be described by what we name as H eigenstate, with a mass mH mA .
In the large t region, the state H is predominantly H10 , with unconventional couplings
to gauge bosons and fermions, as opposed to the state h, the SM-like Higgs boson, with
SM-like couplings to gauge bosons and fermions, and with a mass entirely controlled by
the electroweak breaking parameter. However in this region we have Mh > MH . In this
way the coupling ZZH1 = (ZZh)SM c is strongly suppressed as 1/t and so does the
108
Fig. 5. Lower limit on from the LEP bound mA 95 GeV (left panel), and the corresponding
bound for Mh (right panel).
H direct production. In other words a SM-like Higgs with a mass 115 GeV cannot be
accommodated by our present model. 10
What are then the limits imposed to our model by the bounds on Higgs searches at LEP?
For large t and moderate values of the pseudoscalar mass mA , such that mA = MH , LEP
searches on the MSSM Higgs sector, based on the process e+ e H A, settle a lower
bound on mA as mA 95 GeV. This constraint translates into a lower limit () on
the -parameter that is shown in Fig. 5 (left panel) where the shadowed region is excluded.
In particular an absolute lower bound on around 350 GeV can be read off the plot. The
corresponding lower bound on the SM-like Higgs mass, Mh , is shown in Fig. 5 (right
panel) where again the shadowed region is the excluded one. From this plot we can see
that the absolute lower limit for the SM-Higgs mass (in the one-loop approximation) is
145 GeV. This mass is probably too heavy to be discovered at Tevatron and should await
till the LHC collider.
In summary we see that the Higgs sector of this model is very constrained and will be
probed in the next generation of colliders (LHC). It predicts, as any MSSM with large t ,
an almost degeneracy between the masses of one of the CP-even and the CP-odd states.
LEP bounds on the MSSM Higgs sector set an absolute lower bound on the SM-like Higgs
mass, around 145 GeV, and the -parameter, around 350 GeV, which provides the higgsino
masses. The model is very predictive and will be fully tested at LHC.
5. The supersymmetric spectrum

The supersymmetric mass spectrum is fully determined by the mechanism of electroweak and supersymmetry breaking that we have described in previous sections. All
soft breaking mass terms of zero modes propagating in the bulk of the fifth dimension are
equal, and given by the SS supersymmetry breaking mechanism,
M1 = M2 = M3 = mUi = mDi = mEi = Aabc = ,

(5.1)
R
10 We thank J.R. Espinosa and C. Wagner for pointing out this to us.
109
where i = 1, 2, 3, labels the generation number and Aabc denotes generically all soft
couplings corresponding to trilinear terms in the superpotential. For example, in the case
of the top quark coupling in the superpotential, ht H2 QUR , At comes from the term in the
5D Lagrangian

5 U
L (x5 ) + h.c.,
L5 = ht H2 Q
(5.2)
giving rise, after dimensional reduction and SS-breaking, to the 4D Lagrangian
L4 =
ht
n=
n + (n)
H2 QUL + h.c.,
R
(5.3)
in the notation of Eqs. (2.3) and (2.4). The term n = 0 of (5.3) gives rise to At .
Concerning now the N = 1 sector localized on the brane, it does not receive any treelevel mass, except for the higgsinos (and Higgs bosons) that get a tree-level mass .
However the scalar supersymmetric partners of left-handed quarks and leptons do receive
a radiative mass from the bulk fields, where supersymmetry is broken, mediated by gauge
and Yukawa interactions. These contributions were computed in full generality in Ref. [7]
and we will just quote here the corresponding result applied to the present model.
Squarks receive the main radiative contribution from the gluon/gluino sector proporq we can
tional to the QCD gauge coupling g3 . For the first and second generation squarks
neglect all other gauge and Yukawa couplings and write,

g32
8
1 2
2
2
m
(5.4)
M

+
.
q
9 h2t g 2 /2
2 Z
Plugging numbers in Eq. (5.4) we can see that it yields m
q 1.4 .
= (
bL ) there is, on top of the gauge
For the third generation squark doublet Q
tL ,
contribution of (5.4), a negative contribution from ht,b Yukawa couplings. Using again
the minimization conditions we can write,

3(h2t + h2b )
2
m2Q
(5.5)

1
m

q,
8g32
which gives the numerical rough estimate mQ
.
The mixing can be neglected for all states except for the third generation of up-type
squarks for which the two mass eigenvalues can be approximately written as,
2
m2Q

2
2
+ 2mt +
m2t ,
Mt2
2
R
(/R) m2Q

Mt21 m2Q
m2Q

(/R)2 m2Q
m2t .
(5.6)
= (
For the supersymmetric partners of lepton doublets L
L ,
9L ) (the three generations)
we obtain, in the same way, the soft squared mass value,
m2

L
92 2
mq ,
163
(5.7)
110
which gives the numerical estimate m

L 0.55. In this way, and neglecting the mixing,
the mass of charged sleptons can be approximated by,

2
t 1
2
2
2
2 1
2
s
,
M

m
+
m
+
M

9
Z
W
L
9L
2
t2 + 1
2
t2 1
2
2
2 2
M
(5.8)
,

+
m
+
M
s
9
Z W 2
9R
R
t + 1
while the mass of the sneutrinos
M2L
m2
2
1 2 t 1
M
.
2 Z t2 + 1
(5.9)
In this way the

L turns out to be the lightest supersymmetric particle (LSP) while the

slepton 9L is the next to lightest supersymmetric particle (NLSP). Their masses satisfy the
approximate relation
2
M2L MZ2 .
M
9
(5.10)
Since sneutrinos are neutral particles any kind of cosmological problems associated with
the existence of the LSP is automatically avoided in this model. 11
6. Unification and non-perturbativity scales

In this section we will comment on the sensitivity of the model to the sector of the theory
beyond the scale 1/R. It is a well known fact [20] that, because of the non-renormalizability
of the 5D theory, the gauge and Yukawa couplings run with a power law behaviour of the
scale. This idea was on the basis of the so-called accelerated unification proposed in
Ref. [21] and subsequently analyzed in different papers [2229]. In particular, the gauge
and Yukawa coupling one-loop renormalization for the model analyzed in this paper were
studied in Refs. [15,26] with one-loop -functions:

48 t
e 3 g13 ,
16 2 g2 = 4 et 3 g23 ,
16 2 g3 = 3 g33 ,
16 2 g1 =
5

1
8
8
8
16 2 ht = et 4h2t + h2b 3g22 g12 g32 + 2h2t g12 g32 ht ,
3
3
15
3

1
8
2
8
16 2 hb = et 4h2b + h2t 3g22 g12 g32 + 2h2b g12 g32 hb , (6.1)
3
3
15
3
where g2 g and g1 5/3 g are the SU(2)L U (1)Y gauge couplings with
hypercharge normalization k1 = 5/3.
11 The other possibility pointed out in footnote 7, where Q, L live in the bulk, and U, D, E, H
1,2 are localized
2 = 9 m2 /20 0.35, while
on the brane, would yield the charged slepton,
9R , as the LSP, with a mass m
1
3
q
the sneutrino
L is heavy, with a mass M
L /R.
9R
111
Fig. 6. Plot of the gauge couplings i , i = 1, 2, 3, (left panel), i = t, b, (right panel) as a function of
exp(t) R.
We have run the one-loop renormalization group equations (RGE) from MZ to scales
> 1/R using, for MZ MSUSY the SM beta functions for gauge and Yukawa
couplings, 12 those of the MSSM for MSUSY 1/R, and those in Eq. (6.1) for scales
> 1/R. We have chosen for the plot t 37 and 1/R 4 TeV.
The result is shown in Fig. 6 where we plot the gauge couplings i = gi2 /4 , i = 1, 2, 3,
(solid curves in the left panel) and the Yukawa couplings t,b = h2t,b /4 (right panel).
We see that all couplings, except 1 , are asymptotically free and, as a consequence, in the
region R 30 shown in the plots the theory is weakly coupled. For R > 30, 1 keeps
on growing and for R 40 the theory becomes strongly coupled.
We can compare here the above scales with those obtained in models [9,10] where all
SM particles are propagating in the bulk. There are two main differences:
The first important difference concerns the running of the QCD gauge coupling, g3 . In
our case there is no linear running in g3 because only half of the SM SU(3) triplets
(the right-handed ones) are propagating in the bulk which leads to cancellation of
the coefficient of the linear term in g3 . When also the left-handed triplets propagate
in the bulk there is an extra contribution to the SU(3) -function as 16 2 g3 =
6et g32 which makes g3 to increase with the scale and become non-perturbative at
NP R 68, depending on the value of 1/R. 13 However in our case we have seen
that only 1 is non-asymptotically free and has a linear running, which translates into
a much larger non-perturbative scale NP R 40.
The second important difference concerns the running of Yukawa couplings. As a
consequence of the superpotential structure U T T the -functions of the Yukawa
couplings are governed by the anomalous dimensions of fields in the brane. These
anomalous dimensions involve a single U -field propagating in the loop, which makes
their scale dependence linear. On the other hand, theories with all matter fields
propagating in the bulk rely on localized Yukawa couplings with a superpotential
structure of the type U U U . In that case the one-loop anomalous dimensions involve
two U -fields propagating in the loop, which makes the scale dependence Yukawa
12 To simplify the analysis we use here a common scale of supersymmetry breaking M
SUSY 1 TeV.
13 Since SU(3) is asymptotically free in the SM, the smaller 1/R the larger (1/R) and the smaller R.
NP
3
112
-functions quadratic and worsens the perturbative behaviour of Yukawa couplings.

In fact we have seen that, while in our case the top and bottom Yukawa couplings are
one-loop asymptotically free, in theories with all SM fields in the bulk they become
non-perturbative at scales NP R 36 [9,10].
Finally we can see from the left plot of Fig. 6 that the theory does not unify. This
issue was analyzed in Ref. [26] where it was shown that by adding two zero hypercharge
triplets 14 propagating in the bulk, and contributing to the -functions 16 2 g2 = 8et g23 ,
the theory unifies at GUTR 30. The running of 2 is shown in Fig. 6 (left panel) in
dashed where unification is explicit. Of course sensitivity with the scale is large and to
draw firm conclusions on unification predictions we should control all threshold effect at
the scale GUT from the underlying (string) theory.
7. Conclusions
In this paper we have presented a 5D model with the fifth dimension compactified on the
orbifold S 1 /Z2 , where the parity Z2 is defined by x5 x5 and an appropriate lifting to
spinor and SU(2)R indices. The gauge bosons are propagating in the bulk (U -states) while
matter and Higgs fields can either propagate in the bulk or be localized on the 3-branes
at the orbifold fixed points (T -states). The orbifold selection rules allow superpotential
interactions on the branes of the type U U U and U T T , where the former are suppressed as
(s R)1 with respect to the latter. We then assume a superpotential of the form U T T .
Radiative finite electroweak symmetry breaking is triggered by right-handed stops
propagating in the bulk, where supersymmetry is broken by a ScherkSchwarz mechanism
that uses the U (1)R subgroup of the SU(2)R (R-symmetry) left unbroken by the
compactification. The ScherkSchwarz parameter is considered as a free parameter in
this paper. To allow fermion masses and superpotential interactions on the brane, consistent
with the orbifold selection rules, we must assume that SU(2)L singlets propagate in the
bulk while SU(2)L doublets are localized on the 3-brane, e.g., at the fixed point x5 = 0.
In this way the 4D theory of zero modes and localized states constitute the MSSM (N = 1),
while that of massive modes possesses N = 2 supersymmetry in four dimensions.
The Higgs sector is localized on the brane and then the -parameter does not arise
through compactification, although it is an allowed term in the superpotential. One has
to consider it as an effective parameter from the underlying, supergravity or string, theory.
In this sense the situation is better than in the MSSM since the cutoff s is in the TeV range.
We have proven that electroweak breaking is induced by finite radiative corrections
triggered by the top/stop sector, and remaining as the only free parameters. In fact,
the compactification radius and tan are fixed by the minimization procedure. While
tan is large, tan 40, 1/R is in the range, 115 TeV, depending on the values of
and . The lightest Higgs mass is similar to the MSSM one [31] for large t , in particular
for large values of 1/R for which the heavy KK-modes can be integrated out. We have
14 In the spirit of Ref. [30].
113
checked that their integration produces threshold effects that increase the lightest Higgs
mass by an amount 5 GeV so we expect that once we include the genuine MSSM twoloop corrections [32] the final value predicted for the model will not differ much from
the MSSM one. On the other hand LEP searches on the MSSM Higgs sector imply (
dependent) lower bounds on the parameter and the mass of the SM-like Higgs, Mh .
They translate into the absolute bounds 350 GeV and Mh 145 GeV.
The supersymmetric mass spectrum has a well defined pattern. The heaviest states are
right-handed sfermions, the gauginos and the gravitino, with a mass /R, and higgsinos,
with a mass . The other supersymmetric partners (left-handed sfermions) receive
radiative masses from the supersymmetry breaking in the bulk. The next to heaviest states
are the left-handed squarks which receive radiative masses from the gluon/gluino sector.
We have found that the lightest supersymmetric particles are the sneutrinos, and the next
to lightest supersymmetric particles the charged sleptons, with a squared mass difference
between them MZ2 .
The gauge and Yukawa couplings run linearly with the scale (a reflection of the
non-renormalizability of the 5D theory). The theory remains perturbative for scales
R 40 while 1 and 3 unify at R 30. This is a great difference with respect
to recently proposed models, where all particles propagate in the bulk [9,10] and
that rely on superpotential interactions of the type U U U . In this case wave function
renormalization involves one-loop diagrams with two bulk states propagating and vertices
that do not conserve the KK-number. This translates into quadratic running for the Yukawa
couplings and make the theory non-perturbative for R 36. Since all properties of
the theory rely on summing over the infinite tower of KK-states, having such a low
cutoff is a drawback which can make threshold effects from heavy KK-modes nonnegligible.
Concerning the recent related works already mentioned, Ref. [9] uses a single Higgs
hypermultiplet in the bulk and all matter fields propagating in the bulk. It breaks N = 1
supersymmetry by a SS-mechanism using R-parity as the global symmetry, which amounts
to choosing = 1/2 as the SS-parameter. On the one hand, since they have a single
Higgs field, they solve the -problem by nullification. On the other hand, they have to
rely on U U U Yukawa couplings to give masses to quarks and leptons, which makes, as we
noticed above, the theory non-perturbative for low scales. The prediction on the Higgs mass
( 128 GeV) is reached by our model for the particular case = 1/2, = 0, although
this is a non-physical point for our theory. Finally the LSP in Ref. [9] is the stop, also
making a big difference with respect to our theory which predicts the sneutrino as the
LSP.
Concerning Ref. [10], its authors use two Higgs multiplets (either in the bulk or
localized on the brane) and also all matter fields propagating in the bulk. Supersymmetry
is broken by the SS-mechanism using R-parity, i.e., again = 1/2. 15 Since massive
15 Actually Ref. [10] also proposes another, non-SS-mechanism for supersymmetry breaking based on a strong
supersymmetry breaking localized on a hidden brane that has little relation with that used in this paper and that
we will not comment on here.
114
gauginos form Dirac fermions, they cannot generate radiatively the m23 H1 H2 term in the
potential and this model has to rely on non-renormalizable terms in the superpotential,
as (H1 H2 )2 /s that, along with the -term, generate the m23 -parameter, which
then gets a sensitivity to the cutoff scale, at the tree level. Also Yukawa couplings
renormalize quadratically with the scale and so the theory becomes non-perturbative at
low scales.
Let us finally conclude by emphasizing the beauty of the mechanism of electroweak
breaking triggered by a top-quark propagating in the bulk of an extra dimension. In our
opinion it constitutes a step forward in our understanding of electroweak breaking since
the generated Higgs field instability at the origin is neither imposed by hand at tree-level
nor sensitive to the Standard Model cutoff, in spite of its radiative origin. Of course the
top-quark propagation in the bulk is not without phenomenological consequences [33], that
should be worked out in future works and must be tested in future colliders. On the other
hand, the pattern of supersymmetric mass spectrum should give a hint on the particular
mechanism of supersymmetry breaking realized by the Nature.
Acknowledgements
We thank Jos Ramn Espinosa and Carlos Wagner for discussions concerning present
LEP bounds and Alex Pomarol for useful discussions and for participating in the early
stages of this paper. The work of A.D. was supported by the Spanish Education Office
(MEC) under an FPI scholarship.
References
[1] E. Gildener, S. Weinberg, Phys. Rev. D 13 (1976) 3333;
E. Gildener, Phys. Rev. D 14 (1976) 1667;
M. Veltman, Acta Phys. Pol. B 12 (1981) 437.
[2] S. Coleman, E. Weinberg, Phys. Rev. D 7 (1973) 1888.
[3] L. Ibez, G.G. Ross, Phys. Lett. B 110 (1982) 215;
K. Inoue, A. Kakuto, H. Komatsu, S. Takeshita, Prog. Theor. Phys. 67 (1982) 1889;
K. Inoue, A. Kakuto, H. Komatsu, S. Takeshita, Prog. Theor. Phys. 68 (1982) 927;
L. Alvarez-Gaum, J. Polchinski, M.B. Wise, Nucl. Phys. B 221 (1983) 495;
C. Kounnas, A.B. Lahanas, D.V. Nanopoulos, M. Quirs, Phys. Lett. B 132 (1983) 95;
L.E. Ibez, C. Lpez, Nucl. Phys. B 233 (1984) 511;
C. Kounnas, A.B. Lahanas, D.V. Nanopoulos, M. Quirs, Nucl. Phys. B 236 (1984) 438;
L.E. Ibez, C. Lpez, C. Muoz, Nucl. Phys. B 256 (1985) 218.
[4] I. Antoniadis, Phys. Lett. B 246 (1990) 377;
I. Antoniadis, C. Muoz, M. Quirs, Nucl. Phys. B 397 (1993) 515;
I. Antoniadis, K. Benakli, Phys. Lett. B 326 (1994) 69;
I. Antoniadis, K. Benakli, M. Quirs, Phys. Lett. B 331 (1994) 313;
K. Benakli, Phys. Lett. B 386 (1996) 106;
I. Antoniadis, M. Quirs, Phys. Lett. B 392 (1997) 61.
[5] E.A. Mirabelli, M. Peskin, Phys. Rev. D 58 (1998) 65002.
[6] I. Antoniadis, S. Dimopoulos, A. Pomarol, M. Quirs, Nucl. Phys. B 544 (1999) 503.
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
A. Delgado, A. Pomarol, M. Quirs, Phys. Rev. D 60 (1999) 095008.

I. Antoniadis, K. Benakli, M. Quirs, Nucl. Phys. B 583 (2000) 35.
R. Barbieri, L.J. Hall, Y. Nomura, hep-ph/0011311.
N. Arkani-Hamed, L.J. Hall, Y. Nomura, D. Smith, N. Weiner, hep-ph/0102090.
J. Scherk, J.H. Schwarz, Phys. Lett. B 82 (1979) 60;
J. Scherk, J.H. Schwarz, Nucl. Phys. B 153 (1979) 61;
P. Fayet, Phys. Lett. B 159 (1985) 121;
P. Fayet, Nucl. Phys. B 263 (1986) 649.
C. Kounnas, M. Porrati, Nucl. Phys. B 310 (1988) 355;
S. Ferrara, C. Kounnas, M. Porrati, F. Zwirner, Nucl. Phys. B 318 (1989) 75;
C. Kounnas, B. Rostand, Nucl. Phys. B 341 (1990) 641.
I. Antoniadis, M. Quirs, Nucl. Phys. B 505 (1997) 109;
I. Antoniadis, M. Quirs, Phys. Lett. B 416 (1998) 327;
E. Dudas, C. Grojean, Nucl. Phys. B 507 (1997) 553;
E. Dudas, Phys. Lett. B 416 (1998) 309.
A. Pomarol, M. Quirs, Phys. Lett. B 438 (1998) 255.
A. Delgado, M. Quirs, Phys. Lett. B 484 (2000) 355.
E. Sharpe, Nucl. Phys. B 523 (1998) 211.
A. Delgado, A. Pomarol, M. Quirs, JHEP 0001 (2000) 030.
J. Ellis, J.F. Gunion, H.E. Haber, L. Roszkowski, F. Zwirner, Phys. Rev. D 39 (1989) 844.
R. Barate et al., ALEPH Collaboration, Phys. Lett. B 495 (2000) 1, hep-ex/0011045;
M. Acciarri et al., L3 Collaboration, Phys. Lett. B 495 (2000) 18, hep-ex/0011043.
G. Veneziano, T. Taylor, Phys. Lett. B 212 (1988) 147.
K.R. Dienes, E. Dudas, T. Gherghetta, Phys. Lett. B 436 (1998) 55;
K.R. Dienes, E. Dudas, T. Gherghetta, Nucl. Phys. B 537 (1999) 47.
D. Ghilencea, G.G. Ross, Phys. Lett. B 442 (1998) 165.
C. Bachas, JHEP 9811 (1998) 023.
Z. Kakushadze, Nucl. Phys. B 548 (1999) 205;
Z. Kakushadze, Nucl. Phys. B 551 (1999) 549;
Z. Kakushadze, Nucl. Phys. B 552 (1999) 3.
C.D. Carone, Phys. Lett. B 454 (1999) 70.
A. Delgado, M. Quirs, Nucl. Phys. B 559 (1999) 235.
R.N. Mohapatra, A. Perez-Lorenzana, Nucl. Phys. B 559 (1999) 255.
T. Kobayashi, J. Kubo, M. Mondragon, G. Zoupanos, Nucl. Phys. B 550 (1999) 99.
M. Masip, Phys. Rev. D 62 (2000) 065011.
G. L Kane, C. Kolda, J. Wells, Phys. Rev. Lett. 70 (1993) 2686;
J.R. Espinosa, M. Quirs, Phys. Lett. B 302 (1993) 51;
J.R. Espinosa, M. Quirs, Phys. Rev. Lett. 81 (1998) 516;
J.R. Espinosa, M. Quirs, Nucl. Phys. B 384 (1992) 113.
Y. Okada, M. Yamaguchi, T. Yanagida, Prog. Theor. Phys. 85 (1991) 1;
Y. Okada, M. Yamaguchi, T. Yanagida, Phys. Lett. B 262 (1991) 54;
R. Barbieri, M. Frigeni, M. Caravaglios, Phys. Lett. B 258 (1991) 167;
H.E. Haber, R. Hempfling, Phys. Rev. Lett. 66 (1991) 1815;
A. Yamada, Phys. Lett. B 263 (1991) 233.
J.R. Espinosa, M. Quirs, Phys. Lett. B 266 (1991) 389;
J.A. Casas, J.R. Espinosa, M. Quirs, A. Riotto, Nucl. Phys. B 436 (1995) 3;
M. Carena, J.R. Espinosa, M. Quirs, C.E.M. Wagner, Phys. Lett. B 355 (1995) 209;
M. Carena, M. Quirs, C.E.M. Wagner, Nucl. Phys. B 461 (1996) 407;
H.E. Haber, R. Hempfling, A.H. Hoang, Z. Phys. C 75 (1997) 539;
115
116
J.R. Espinosa, R.-J. Zhang, JHEP 03 (2000) 026;

J.R. Espinosa, R.-J. Zhang, Nucl. Phys. B 586 (2000) 3;
M. Carena, H.E. Haber, S. Heinemeyer, W. Hollik, C.E.M. Wagner, G. Weiglein, Nucl. Phys.
B 580 (2000) 29.
[33] F. del Aguila, J. Santiago, Phys. Lett. B 493 (2000) 175.

Instabilities in heterotic M-theory

induced by open membrane instantons
Gregory Moore a , Grigor Peradze a,b , Natalia Saulina c
a Department of Physics, Rutgers University, Piscataway, NJ 08854, USA
b Department of Physics, Yale University, New Haven,CT 06520, USA
c Department of Physics, Jadwin Hall, Princeton University, Princeton, NJ 08544, USA
Received 1 February 2001; accepted 28 March 2001
Abstract
We study the effective low energy supergravity of the strongly coupled heterotic string
compactified on a CalabiYau 3-fold with generic E8 E8 gauge bundle. We focus on the
effective potential for the chiral scalars. The effective superpotential can be studied using the dual
11-dimensional M-theory background involving insertions of M5-branes along an interval. In such
backgrounds, in some regions of moduli space, the leading nonperturbative contributions are due to
open membrane instantons. These instantons lead to both attractive and repulsive forces between the
5-branes and the orientifold M9-branes, depending on the region of moduli space. The resulting
dynamics on moduli space include a strong coupling dual to the DineSeiberg instability, in which
the interval grows. We discuss conditions under which the 5-branes are attracted to the wall and
comment on the relevance of these results to the study of chirality-changing phase transitions in
heterotic M-theory. 2001 Published by Elsevier Science B.V.
PACS: 04.65.+e; 11.25.-w
Keywords: M-theory; Strings
1. Introduction
In the past few years there have been significant advances in the study of strongly
coupled heterotic string theory, thanks to the formulation in terms of M-theory on an
interval S 1 /Z2 [1,2]. In particular, the compactification of M-theory on a product of an
interval with a CalabiYau 3-fold (denoted hereafter by X ) leads to qualitatively different
physics from that of the weakly coupled heterotic string, as first noted in [3,4,15].
In heterotic string compactification one must choose an instanton configuration for
gauge fields along X . The so-called standard embedding identifies the gauge field
E-mail addresses: gmoore@physics.rutgers.edu (G. Moore), peradze@physics.rutgers.edu (G. Peradze),
saulina@princeton.edu (N. Saulina).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 5 5 - 9
118
G. Moore et al. / Nuclear Physics B 607 (2001) 117154
with the spin connection of the metric. Other choices of gauge instantons, the socalled nonstandard embeddings, are closely related, in the strongly coupled regime,
to backgrounds obtained by including insertions of M5-branes wrapping a product of
4-dimensional spacetime with a holomorphic curve in X . At low energies, the physics of
such backgrounds is summarized by a complicated d = 4, N = 1 supergravity theory. It has
been shown in [1620] that such backgrounds can lead to phenomenologically interesting
gauge groups, and it is therefore of interest to understand more completely the full low
energy supergravity in such backgrounds. While several aspects of the effective Lagrangian
have been worked out in [3,4,1620,49,50] (for a review see [43]) the Lagrangian is
extremely complicated, and many details remain to be understood more thoroughly. The
present paper derives some further aspects of the low energy Lagrangian. Our main result
is a formula for the potential energy for the moduli fields, valid in certain regions of moduli
space. The detailed expression is given in Eq. (102) and preceding equations, for the case
when there is a single 5-brane insertion and h1,1 (X ) = 1. Since the derivation is rather long
we explain here a few of the ingredients of this formula.
The chiral scalars in d = 4 supergravity take values in a target space which is Khler
Hodge. These fields correspond to moduli for the CalabiYau metric on X , moduli for the
instanton gauge field along X , and moduli for the positions of the M5-branes along the
interval. In addition there are chiral fields charged under the gauge group H left unbroken
by the E8 E8 instanton. These will generically be denoted by C I .
The superpotential W is a section of a line bundle on the KhlerHodge target space
for the chiral scalars. There are several sources for the superpotential in the effective
supergravity. First of all, there is a perturbative term, cubic in the scalars C I . In addition,
there are several sources of nonperturbative effects. Some of these, such as heterotic
worldsheet instantons, gluino condensation, and M5 instantons (wrapping X ) have been
studied in many previous papers [615]. The inclusion of the effects of open membrane
instantons, which have not been studied as thoroughly, is the main focus of this paper.
There are three kinds of open membrane effects we must consider, since M2-branes can
end both on M5-branes, or on the boundary M9-brane [28,29]. Membranes stretching
between the boundary M9-branes are the M-theory versions of heterotic worldsheet
instantons, and as such have been studied in the context of (0, 2) compactifications of
heterotic string backgrounds [13,14]. It is well-known that such effects often sum to
zero, e.g., in backgrounds admitting a description by a linear sigma model [2123].
The mechanism by which these contributions vanish is that a given homology class can
contain many different holomorphic curves in X . The instanton action depends only on
the homology class, but the prefactor depends on the curve, and the sum of instanton
amplitudes can, and often really does, vanish, as can already be seen in the case of the
quintic. By contrast, the M2 instantons stretching between M5 and M9, or between M5
and M5 must wrap the particular holomorphic curve already wrapped by the M5-brane.
This is obvious for the part of the membrane worldvolume ending on the 5-brane. A study
of the conditions for the supersymmetric instanton (based on [7]) reveals that the membrane
must have a direct product structure I where I is an interval and X is a
holomorphic curve. (The detailed argument is given in Section 3.) Consequently, if is a
119
rigid holomorphic curve in X there will be no sum over instantons, and no integral over the
moduli space for the curve. Moreover, if is a rational curve there will be precisely two
fermion zeromodes and the fermion 2-point function determining the superpotential will
be nonzero. (Our calculation of the induced superpotential uses the technique discussed in
[68].)
The backgrounds we study are in a regime of M-theory where we can do systematic
expansions in the long wavelength expansion. It follows from [13] that this is an
expansion in R/V 2/3 where R is the length of the interval S 1 /Z2 and V is the volume
of X in 11-dimensional Planck units. We therefore assume R/V 2/3 1. Now, gluino
condensation and 5-brane instanton effects contribute terms of order W exp[c1 V ] to
the superpotential W , where c1 is of order 1. By contrast, open membrane effects contribute
terms of order W exp[c2 RV 1/3 ] where c2 is of order 1 (or smaller). Thus, in the
backgrounds under study in this paper, open membrane instantons are the leading source
of nonperturbative effects.
Our goal is to understand the physics of the moduli in heterotic M-theory, so we need
the potential, rather than just the superpotential. The potential energy for scalars in d = 4,
N = 1 supergravity is given by the famous formula [37,38]

+ UD ,
(4 )4 U = eK K i j Di W Dj W 3W W
(1)
where 242 = 16GN is the (four-dimensional) Newton constant, K is the Khler potential,
Di W = i W + i KW is the covariant derivative, and UD are D-terms for charged scalars
a C)2 .
a (CT
The potential (1) is extremely complicated. Moreover, K is only approximately known
only in some regions of moduli space. We are therefore forced to consider perturbation
expansions in several quantities. First, we will expand in two dimensionless parameters
E eff R/V 2/3 1,
EReff V 1/6 /R 1
(2)
which are necessary for the validity of the geometrical 11-dimensional picture (more
precise formulae appear in Eq. (11)). Note that these imply that V 1 and R 1, and
that the length of the interval is much larger than the scale set by X . In addition we must
expand in powers of the charged scalars C I . The superpotential is a sum of two terms
W = Wpert + Wnonpert , where Wpert is a cubic expression in the charged scalars C I with
coefficients that are functions of the complex structure and bundle moduli. We can organize
the terms according to whether they are order 0, 1, or 2 in Wnonpert :
(4 )4 U = (U0 + U1 + U2 ).
(3)
We will now describe the leading expressions for the three terms in (3) in the case of
a CalabiYau X with h1,1 (X ) = 1 together with a single 5-brane, inserted at x, where
0 x 1 labels the position of the 5-brane along the M-theory interval. In addition to the
charged scalars C I the relevant chiral superfields are the volume superfield S = V + i ,
which determines the GUT coupling, the Khler superfield T = Ra + i , where a is the
Khler modulus for X (hence V a 3 ), and the position superfield Z = Rax + i for
the 5-brane. The fields , and are axions.
120
The first term, U0 , in (3) begins with the perturbative contribution to the potential. The
leading order expression in an expansion in the charged scalars is a positive semidefinite
quartic form:

2

1
C
I
J K
L
eff eff
U0 =
(4)
,
E
U
C
C
,
E
1
+
O
.
C
C
R
V J 2 I J KL
J
Here J := Ra. The coefficients UI JK L are functions of the complex structure and bundle
moduli. We will give precise formulae for them, but will not be very explicit about their
behavior.
The leading contribution to the second term in (3) is a one-instanton term resulting from
cross terms between the perturbative and nonperturbative superpotentials. We find that the
single instanton contribution has the form

(1 x) J x
e
Re UI J K C I C J C K ei
U1 =
2
VJ

eJ (1x) Re UI J K C I C J C K ei() + .
(5)
The coefficients UI J K are functions only of the complex structure and bundle moduli.
Finally, the third term U2 in (3) begins with a 2-instanton effect
E
U2 = 2 e2J x + e2J (1x) 2eJ cos(2 )
J

2J
4J x J
+
(1 2x)e2J (1x) +
e cos(2 ) +
3V
3V

2

C
, E eff , EReff ,
1+O
(6)
J
where E is a positive definite function that depends only on the complex structure and
bundle moduli. (We have kept some subleading terms in the second line. The reason for
this is explained in detail in Sections 5.4 and 5.5.)
A precise characterization of the region of validity of the above potential is given in
Section 5.4. The strongest constraints on the region of validity come from our ignorance
of the exact Khler potential. It is also important to bear in mind that the coefficients
of the higher order terms in the expansion in |C|2 /J , E eff , EReff are functions of the
complex structure and bundle moduli. If these coefficients become singular somewhere in
the moduli space then these higher order terms will dominate the physics. Our working
assumption is that we are at a generic smooth point in bundle and complex structure moduli
space.
Having determined the leading nonperturbative effects, and thereby the potential energy,
we investigate briefly some of the resulting classical dynamics on moduli space, at a
somewhat heuristic level. Although the M5-branes wrap all of spacetime, thanks to the
central term in the superalgebra, their positions along the M-theory interval are in fact
dynamical variables. In the regions where we can trust our answer we find two kinds
of instabilities in the compactification, depending on whether the effects of vevs of the
charged scalars C I are important or not. When the charged scalar vevs are important, the
121
leading x-dependent effect is a one-instanton effect. The axions will evolve to produce
an attractive force between the M5-brane and the nearest M9-wall. This could possibly
be interpreted as a consequence of the Witten effect: the axions evolve and continuously
change an effective brane charge in order to produce the most attractive channel, in
particular producing an attraction between the 5-brane and the boundary. It would be
interesting to understand the physics of this effect more fully.
The above discussion is valid for J x 1. As the five-brane moves towards the wall the
approximations break down. The limit x 0 is extremely interesting and is related to the
chirality-changing transitions discussed in [42,56]. In order to study this limit one needs
a multiple cover formula for the membrane instantons. This is discussed in Section 6. We
make some educated guesses and conclude that the physics depends on the (unknown)
details of the covering formula.
A second kind of instability occurs when charged scalar vevs are small or zero. In this
case the potential has a local minimum in x at x = 1/2. The value of U at such points is
small and of the form
eJ
,
JV
where is a positive function of the complex structure and bundle moduli. The M2-branes
lead to a repulsive interaction between the M5-brane and the M9-brane which induces
decompactification of both the M-theory radius and the CalabiYau, while the M5-brane
moves to the middle of the interval. Of course, in this instability new light modes appear as
the theory becomes five-dimensional, and we should describe a matching to a description
in terms of five-dimensional supergravity. (As the M5 moves to the middle of the interval
there is a balancing of forces from the two boundaries and the leading terms in U2 vanish.
This is why we must include the subleading terms.)
The second kind of instability is an 11-dimensional manifestation of the DineSeiberg
problem; it is hardly unexpected, and in the case of the standard embedding similar
instabilities have already been pointed out by Banks and Dine in [4]. Nevertheless, it
is interesting to note that in the 10-dimensional DineSeiberg instability the size of the
M-theory interval S 1 /Z2 tends to shrink. There are thus different asymptotic regions
of moduli space with qualitatively different dynamics, and hence different basins of
attraction for the classical evolution of the moduli. One consequence is that there must
be nontrivial stationary points for the potential in the middle of moduli space. The precise
nature of such stationary points is of great interest, but remains out of reach so long
as we cannot derive the Khler potential in the interior of moduli space in a controlled
approximation.
The paper is organized as follows. In Section 2 we review briefly the M-theory
geometry corresponding to strongly coupled E8 E8 heterotic strings with nonstandard
embedding. In Section 3 we study supersymmetric M2-brane instantons in X S 1 /Z2 .
In Section 4 we derive the formula for the contribution to the superpotential from M2
instantons. In Section 5 we find the potential and specify the region where we can trust
it for the simplest case of a CalabiYau with h(1,1) = 1. In Section 6 we discuss the
multiple covering formula and its relevance to chirality-changing transitions. In Section 7
U
122
we generalize the result to the case of N 5-branes on the interval. The final section contains
a discussion of some possible extensions of the present work.
We have been informed that the effects of open membranes in heterotic M-theory are
also being investigated independently by B. Ovrut, E. Lima and R. Reinbacher.
2. Review of heterotic M-theory background with M5-branes on the interval

In this section we review some of the results of [3,4,1620,43] which are needed for our
subsequent computations.
Our conventions for the Lagrangian of 11D SUGRA are set by the Lagrangian:

1
1
2
S11D = eR
211
(7)
G4 G4
C3 G4 G4 + ,
2
6
where GMNP Q = 4[M CNP Q] . 1
The Lagrangian of the boundary E8 E8 theory is given by

2
1 11 2/3
2
g tr F (1)
211 SYM =
4 4
1
4
M110
11
4
2/3

2
g tr F (2) ,
(8)
M210
where F (1,2) are the field strengths of the two E8 gauge fields, to leading order in a longwavelength expansion. In the above action and below tr means 1/30 of the trace in the
adjoint of E8 .
We begin by describing the background solution of M-theory on R 4 X S 1 /Z2 .
Our coordinates on R 4 are x , = 1, . . . , 4. Complex coordinates along X have indices
m, m
= 1, . . . , 3. The factor S 1 /Z2 in spacetime has coordinate X11 . In addition it will be
convenient to set X11 = y where y is a dimensionless coordinate 0 y 1, and is a
dimensionful constant which sets a scale.
We must now specify the metric, four-form G4 , and boundary YangMills fields. In
order to write the background metric we introduce a basis of harmonic (1, 1) forms on X ,
i , i = 1, . . . , h1,1 and denote the Khler form on X by = a i i . Then, the background
metric is a deformation of a metric of the form

2
2
ds11
(9)
= V 1 R 1 g dx dx + R 2 dX11 2imm dx m dx m .
In this formula R is dimensionless and R is the orbifold radius. Similarly, we introduce a
fiducial, dimensionful, volume v for X , and
the3 volume of X in the metric (9) defines the
1
dimensionless parameter V by V v := 3! X . We will make a convenient choice of ,
[16]
here
1 We have a different normalization of fields compared to [16]. Ghere
MNP Q = 2 GMNP Q , CMNP =
[16]
2 = (2 )8 (M )9 , and define the 11-dimensional Planck length by
6 2 CMNP . We use the convention 211
11
l11 = 1/M11 . Our signature is mostly plus.
123
6 and are independent of moduli. Because

v in Eq. (12); they will be of the order of l11 , l11
of the Weyl-rescaling in the first term in (9), g is the four-dimensional Einstein metric
and the four-dimensional Newton constant is given by
2v
1
= 2 .
2
4
11
(10)
As we have mentioned, the actual metric we will use is a deformation of Eq. (9), and is
only known to first order in a power series in two dimensionless expansion parameters
7R
7R V 1/6
eff
1,

1,
E
=
R
V 2/3
R
where we choose the constants

1/2
v 1/6
11 2/3 2 2
=
2,
7
=
7=
=
R
2/3
4
2
v
E eff =
(11)
(12)
in order to simplify the normalization of the fields in the effective Lagrangian. 2

The above inequalities (11) state, firstly, that the distortion of the background from
(9) is small, and secondly that the interval is much larger than the length scale of X .
These expansion parameters can be related to the GUT scale and the 4-dimensional
Newton constant [3,4]. In our conventions the unified coupling GUT (E eff EReff )2 1/V ,
while (MGUT 4 )2 (E eff )3 (EReff )4 1/(RV 4/3 ) determines the GUT scale in terms of
the Newton constant. The latter formula follows by computing masses of gauge bosons
and scalars associated with typical mechanisms of spontaneous symmetry breaking. 3
Unfortunately, it turns out that when we use the experimentally measured values of
GUT ,MGUT and 4 the above expansion is not necessarily a good approximation. As
discussed in [35,16,51], the experimentally measured values determine EReff E eff =
O(1). Nevertheless, our focus in this paper is on a systematic and controlled computation
of nonperturbative effects; the restriction (11) is necessary since heterotic M-theory is only
known as an effective theory to order (11 )2/3 , and for this reason we will adopt it.
To lowest order in the expansion parameter the metric for the background takes the form

B
2B 112
2
1 1
2
2iJmm dx m dx m ,
1+
ds11 = V R
g dx dx + R 1
dx
6
3

1
B = 2mm Bmm .
Jmm = mm + Bmm mm B ,
(13)
3
The deformation of the background is described by the (1, 1) form Bnm . In order to write
it explicitly we must now introduce the M5-branes.
The backgrounds we study preserve N = 1 supersymmetry. Therefore the 5-branes wrap
a product of spacetime and a holomorphic curve in X . If there are N 5-branes they will
therefore have definite locations at y = xk , k = 1, . . . , N along the interval. The kth 5-brane
wraps a curve (k) in X whose homology class may be expressed as [ (k) ] = i(k)[2i ]
2 We take v = 8 5 l 6 , = 2 1/3 l to have 7 = 2, 7 2 = 1 .
11
R
2
11
3 We thank T. Banks for very helpful discussions on this point.
124
where [2i ] is an integral basis of H2 (X, Z), and i(k) is a collection of nonnegative
integers. These integers are constrained by anomaly cancellation. Each of the M9-branes
carries an E8 vector bundle V1 , V2 , and to each bundle we associate a degree four integral
characteristic class c2 (Vi ). Identifying H2 (X ; Z) with H 4 (X ; Z) via Poincar duality we
may define

1
1
c2 (V1 ) c2 (T X) = i(0) 2i ,
c2 (V2 ) c2 (T X) = i(N+1) 2i .
2
2
The anomaly cancellation condition is then
N+1

(n)
= 0.
(14)
(15)
n=0
In terms of the above data, the formula for Bmm on the interval (xn , xn+1 ), n = 0, . . . , N is
given by
Bmm =
j =
2R
i
bi m
m
,
V
N+1

bj (y) =
(1 xk )2 j(k) ,
n

k=0
1
(k)
j (y xk ) j ,
2
(16)
k=0
where x0 = 0, xN+1 = 1 and the index i is raised with the inverse of the metric on the
moduli space of Khler structures on X :

1
1
Gij =
(17)
i (j ) = i j ln di1 i2 i3 a i1 a i2 a i3
2vV
2
X
with

di1 i2 i3 =
i1 i2 i3 .
(18)
The choice of integration constant in the solution (16) fixes V v to be equal to the volume
of X averaged along the interval (to lowest order in E eff ).
The flux of the 4-form G4 is also given in terms of B:
1
GMNP Q = 7MNP QEF 11 B EF .
(19)
2
Note that it is discontinuous across the positions of the 5-branes.
Finally, we need to specify the E8 gauge bundles V1 and V2 . For simplicity we will
follow [17] and take the bundle V2 at y = 1 to be the trivial bundle. Accordingly, there is
a hidden sector at y = 1 with unbroken E8 gauge group. The bundle V1 at y = 0 has an
instanton whose holonomy lives in a subgroup G E8 . The unbroken gauge symmetry is
the commutant H of G in E8 . It is straightforward to extend our formulae to the case when
both V1 and V2 are non-trivial bundles.
When we compactify M-theory on the above background, the physics at distances large
compared to the M-theory interval is described by an effective d = 4, N = 1 supergravity
125
theory. We now list the massless fields corresponding to small fluctuations around the
above background. In addition to the super-YangMills and supergravity multiplets there
are a number of massless chiral scalar fields. To begin with, there are chiral superfields
neutral under four dimensional gauge group H . These are:
T i = Ra i + i i ,
(20)
S = V + i,

Zn = R i(n) a i xn i An i(n) a i xn i(n) i
(21)
(22)
where
Cmm11
= i i,mm ,
i = 1, . . . , h1,1 , m, m
= 1, . . . , 3,
is a scalar dual to C11

3[ C]11 = V 2 7 ,
and Zn is a holomorphic coordinate constructed out of the position xn of the nth 5-brane
on the interval. The scalar An originates from the KK reduction of the 2-form living on the
nth 5-brane
A(2)
n = An fn ().
(23)
We have included the factor in the above formula to make An dimensionless. In

(n)
Eq. (23) fn () is the pullback of the Khler form to the cycle 2 . We denote by fn
the holomorphic embedding of the curve 2(n) in X . The pullback of each of the basis
forms fn (i ) is proportional to the pullback of the Khler form

i(n)
(n)
fn (),
fn (i ) = (n)
fn (i ) = v 1/3 i .
(j a j )
(n)
Finally, there are chiral multiplets charged under the unbroken gauge group H . Thanks
to the DonaldsonUhlenbeckYau theorem massless modes from small fluctuations of the
gauge field can be associated with holomorphic deformations of holomorphic bundles
on X . The small fluctuations are parametrized by the space H0,1(X, V ) where V is the
gauge bundle in the 248. We assume the holonomy of the instanton is in G so the gauge
bundle decomposes as V = WR VS corresponding to the decomposition of the adjoint
of E8 under the embedding H G E8 :
248 = R S.
(24)
H0,1(X, VS ).
The charged scalars will be valued in WR

KaluzaKlein reduction we decompose the gauge field as:
In order to work out the
Am =
22
u CI ,
4 I ,m
m
= 1, 2, 3.
In (25) a summation is taken over the index I which labels

I = (R, I, p),
p = 1, . . . , dim R, I = 1, . . . , dim H 1 (X, V1S ).
(25)
126
The normalization factor in (25) was chosen to make the charged scalar fields C I
dimensionless and to normalize their kinetic term conveniently.
When writing the perturbative superpotential below it will be convenient to define
uI,m = uxIm Txp ,
where x is an index for a basis for the representation S and uxIm is a basis of H 1 (X, V1S ).
The factor Txp is purely group-theoretic and corresponds to the generators of E8 in the
representation R S. The complex conjugate of these generators is denoted by T xp and
y q
the normalization is chosen such that tr(Txp T yq ) = x p .
We are not going to study four-dimensional gauge dynamics in this paper. This has been
studied, for example, in [3,4,17]. For completeness, and to fix our normalizations, we also
give the gauge kinetic term in the 4D Lagrangian
SYM =
2

=1
1
64 2
g4 Ref tr F 2 +
(26)
()
M4
where on the brane at y = 0

Ref (1) = V + Ra i
(0)
i
N

(n)
+
(1 xn )2 i

(27)
n=1
and on the brane at y = 1

Ref
(2)
= V + Ra
(N+1)
i

N

2 (n)
+
(xn ) i
.
(28)
n=1
Note, that due to the restrictions (11) on the moduli space, Ref = V + O(E eff ),
= 1, 2.
3. M2-brane instantons in X S 1 /Z2
Open M2-branes ending on an M5-brane will play a crucial role in our calculation of the
non-perturbative potential. These nonperturbative effects were first discussed in [28,29,40].
In this section we will derive the conditions for a supersymmetric open M2-brane instanton
in the background described in the previous section. We will neglect the distortion of the
background metric from a direct product metric in solving for the membrane configuration.
This is valid in our approximation scheme.
The first step in finding the supersymmetric M2 configuration is to write the constant
spinors corresponding to the supersymmetries unbroken by the background. We use a basis
for the -matrices in eleven dimensions of the form
m = 1 m,
= (RV )1/2 7 ,

= 1, . . . , 4, , = 2g ,
(29)
1
11 = 5 7 ,
R

nm
,
m, m
= 1, . . . , 3, n , m = 2gCY
127
m = 1 m ,
(30)
where (m ) = m = (m )T and is a Weyl-basis in 4D.

Four-dimensional anti-chiral (chiral) spinor indices are denoted by (),
respectively.
In this basis the surviving supersymmetry in the background X S 1 /Z2 is of the form:

= 7 71 , 7 72 ,
(31)
where 7 , 7 are constant spinors on R 4 S 1 /Z2 and 71 (72 = (71 ) ) is the chiral (antichiral) covariantly constant spinor on X , normalized as in [7]:
m 71 = 0,
nm 71 = inm 71 ,
mnp 71 = eK mnp 72 ,
71 71 = 1.
(32)
Here is the Khler form, is a holomorphic (3, 0) form on X and K = 12 (KT Kcplx )
with both Khler functions KT and Kcplx specified in Section 5.2.
The surviving supersymmetry is consistent with having 5-branes wrapped over a
holomorphic cycle X , as shown in [16]. One cannot have anti-5-branes on the interval
and preserve supersymmetry.
The presence of an M2-brane imposes an additional constraint on the supersymmetry
parameter
(2) = ,
(33)
where (see, for example, [7]),

(2) =
i

7 ij k i XM j XN k XK M
N
K
.
3! g
(34)

In formula (34) s i , i = 1, 2, 3 are coordinates on the worldvolume of the M2-brane, XM ,

= (, m, m,
M
11) are coordinates in the eleven-dimensional target space and g is the
determinant of the induced metric on the M2-brane.
Substituting (31) into (33) we find, first of all, that spinors of type 7 72 lead to

K

R ij k
ie
m
n
11
ij k
m
n
p
72 = 7 i X j X k X mn 72 +
7 i X j X k X m n p 71
g
3! g
ij k

7
+ i Xm j Xn k Xp mn p 72
g

K
ie
pq
ij k
m
n
11
+
(35)
7 i X j X k X m n p gCY q 71 .
4 g
Since the spinors 71 , 72 , m 71 , m 72 are linearly independent we get four equations
i Xm j Xn k Xp m n p = 0,
Ri Xm j Xn k X11 mn = g 7ij k ,
pq
i X j X k X m n p gCY = 0,
7 ij k i Xm j Xn k Xp mn = 0.
m
11
(36)
(37)
(38)
(39)
128
The constraints (36), (39) are automatically solved by the embedding

X11 = t,
Xm (y),
where t = s 3 is a coordinate along the orbifold interval and y, y are coordinates on a

holomorphic 2-cycle. This is our basic instanton.
We claim that if the holomorphic curve X is isolated then the above membrane
instanton is also. Moreover, we claim that the above instanton is the only instanton solution
consistent with the boundary condition of having the M2-brane ending on . Indeed, let
us consider the possibility of having M2-branes starting and ending on holomorphic cycles
inside X which differ from a direct product I . Therefore we search for t-dependent
embeddings Xm (y, t), t [x1 , x2 ] into X . In this case Eq. (39) is not satisfied automatically
and gives the constraint
[i Xm j ] Xn mn = 0.
(40)
Taking the i = y, j = y component of this equation and evaluating it at the boundary t = x1

or t = x2 shows that the volume of the holomorphic cycle must be zero.
We conclude that an open M2-brane which starts and ends on a positive volume
holomorphic curve preserves some supersymmetry iff it has the direct product form I .
One can quite analogously prove that an M2-brane which starts and ends on a
holomorphic curve should have the direct product form
X11 = t,
Xm (y)
in order to preserve the other components 7 71 of the background supersymmetry.

Note that since the M2-brane instanton must start and end on the same 2-cycle in X
(n)
(k)
there is a requirement on the 5-brane charges i = i described in Section 2 in order
for there to be an M2-instanton stretched in the interval [xn , xk ].
4. Calculation of membrane-instanton-induced superpotentials

In this section we will give the derivation of the non-perturbative four-dimensional
superpotential W induced by open membranes.
We follow the procedure outlined in [7,8]. The idea is to compute the 2-point
correlation function of four-dimensional fermions with the instanton sector included in
the supergravity path integral. An essential ingredient of this calculation is the coupling of
the four-dimensional fermions to the worldvolume degrees of freedom of the membrane
through the so-called membrane vertex operators. The computation of the superpotential
follows from a computation of a 2-point correlation function of fermions in the fourdimensional effective theory inst , where are fermionic superpartners of Z. This
in turn can be reduced to a membrane path integral with corresponding vertex operator
insertions.
129
4.1. Summary of the computation of W

Since the analysis is rather long let us summarize the computation here. Most of the work
is devoted to finding the vertex operator, but the end result is very simple. The membrane
theory has a chiral doublet of fermions transforming in the 2 of the 4-dimensional
Lorentz group. These couple to the chiral fermions in the superfield Z via the vertex
operator
i
V = .
(41)
2
Using the above coupling we can compute (1 )(2 ) in an instanton sector to be

(42)
g4 d 4 SF (1 )SF (2 )h exp(Z).
Here 1 , 2 are points in four dimensions and SF is the 4-dimensional fermion propagator
in the effective d = 4, N = 1 supergravity. This expression for the propagator is only
valid for (1 ), (1 2 ), (2 ) l11 . The integral of in Eq. (42) should be
regarded as an integral over the bosonic zero modes X = of the M2-instanton. The
integral over the 2 fermion zero-modes 1 , 2 , on the M2-brane soaks up the from the
vertex operator. There are no other zero modes because the curve is a rational curve
and hence has no extra zero-modes associated with 1-forms. The prefactor h stands
for determinants of fluctuations in 11-dimensional supergravity together with 5-branes
around the background (13), (19), together with determinants associated with the degrees
of freedom for the M2-instanton. While it is very complicated one can use holomorphy
to extract the factor heZ , which depends holomorphically on the moduli. The factor h is
a holomorphic section of a line bundle over complex structure moduli space and should
properly be regarded as the true measure for the fermion zeromodes. In this paper we will
not be very explicit about it.
We can now extract W by comparing (42) with the 2-point correlation function in the
effective 4D supergravity

1
K
2
g4 e Z Z (W )
(1 )(2 ) exp
(43)
4D
which is equal to

1
g4 d 4 SF (1 )SF (2 )e 2 K Z Z (W ),
(44)
where we drop corrections of the order O(E eff , EReff , |C|2 /Ra) to the mass term for a chiral
fermion in the 4D, N = 1 Lagrangian [37,38].
Using holomorphy of the superpotential it now follows that

1
W = h exp Z ,
= e2K.
(45)
For the M2-brane stretched between the 5-brane and the other 9-brane at y = 1 analogous
considerations give

W = h exp i T i Z .
(46)
130
4.2. Computation of the vertex operator

In this section we describe the computation of the vertex operator.
The vertex operators can be found by expanding the action of the M2-brane in the
M-theory background fields. The action of an M2-brane ending on an M5-brane was
written in [34], using the superembedding approach of [30,31]. In this approach the basic
ingredients are:
An (11|32) supermanifold M, giving the 11-dimensional supergravity background.

The supercoordinates are denoted by Z M = (XM , ), where is an index in the
irreducible spinor representation of so(1, 10). Using the torsion constraints of [32] on
the supervielbein one can expand an orthonormal frame for the cotangent space in
powers of . The expansion at low orders in has been worked out in [8,9,33].
It is convenient to introduce the notation for the vielbein:

E A (Z) = dZ M EM A = E a , E
(47)
where in the second equality we have separated bosonic and fermionic cotangent
vectors.
A (6|16) supermanifold M describing the worldvolume of the M5-brane. We denote
supercoordinates on the worldvolume by zM = (y M , ) and a cotangent frame on
M by eA (z). A decomposition of the frame analogous to (47) is given by

eA (z) = dzM eM A = ea , eq .
(48)
The index a = 0, 1, . . . , 5 is the index of the vector representation of so(1, 5), while
and q are the indices of irreducible spinor representations of so(1, 5) and so(5),
respectively.
A (3|0) manifold , to be identified with the membrane worldvolume. The
boundaries of lie inside M or in M. We denote the coordinates on by s i ,
i, . . . , 3. Coordinates on the boundary surface are denoted by r , r = 1, 2.
The relation of the pullback of the 11-dimensional supervielbein to the 5-brane to the
supervielbein of the 5-brane itself defines the embedding matrices EA A via the equation

f E A = e A EA A .
(49)
One may solve for these matrices in terms of the vielbeins
EA A = eA M M Z M EM A .
(50)
The basic superembedding condition then says that

Efermionicbosonic = Eq a = 0.
(51)
This simple equation is extremely powerful, it leads to a complete set of covariant equations
of motion for the 5-brane [30,31].
The action of an M2-brane ending on an M5-brane, in Euclidean signature, is [34,44]

SM2 = M2 d 3 s det gij + if C(3) iM2 d 2 B (2) .
(52)
Here M2 =
1
M3
(2)2 11
131
is the M2-brane tension. Also, B (2) is the super 2-form living on M
while C(3) is the super 3-form living on the target superspace M. The pullback in Eq. (52)
under the embedding f : s i Z M is
f C(3) =
1
i
j
k
i Z M j Z N k Z P C(3)
MN P ds ds ds ,
3!
while the pullback under the embedding : r zM is

1
(2)
B (2) = r zM s zN BMN d r d s .
2
We specialize the action (52) to the case of an M2-brane stretched between y = 0 and
y = x in the background described in Section 2. The membrane is a product [0, x] so it
is convenient to define coordinates on the membrane s i = (t, , ) where t is a coordinate
on the interval and is a holomorphic coordinate along the curve . The embedding
coordinates of into (11|32) superspace
M

(s), 3,11 (s) ,
Z M (s) = X3,11
(53)
have the following structure. First, the interval coordinate

i
11
11
X3,11 (s) = t + 3,11 3,11

2
(54)
has an important correction quadratic in fermions, while the coordinates

m
X3,11
(s) = Xm ( ),
X3,11
(s) = Xm ( ),
describe the holomorphic embedding. The coordinate

fermions 3,11 (s) satisfy the physical gauge condition
(55)
X3,11 (s)
(2) 3,11 (s) = 3,11 (s).
is unconstrained. The
(56)
We have omitted the coordinates describing fluctuations of the membrane within X since
we will restrict our consideration to an isolated curve and hence these degrees of
freedom will be massive.
The origin of the correction in (54) is continuity of embedding coordinates in superspace.
That is, the embedding of the membrane into 11-dimensional superspace (3|0) (11|32)
must agree, on the boundary, with the embedding of the 5-brane into superspace (6|16)
(11|32) since the membrane ends on the 5-brane in superspace. We now derive this
condition in more detail.
where y are real
We choose bosonic coordinates on the M5-brane as y M = (y , y, y)
and y is complex, and consider the static embedding of the boundary of the M2-brane into
the M5-brane
: y = , y = .
The superembedding (6|16) (11|32) is described by superfields

Z M = X M zM , zM .
(57)
132
Small fluctuations around static gauge are described by embeddings

XM = y m , Xm (y, ) ,
= , (y, ) ,
(58)

where is a fermionic coordinate in the (6|16) superspace. The superfields Xm (y, ) and
(y, ) have as their = 0 component bosonic fluctuations transverse to the worldvolume

of the M5-brane Xm (y M ) and physical fermions on the M5-brane (y M ), respectively.
m here denotes bosonic indices of coordinates transverse to the M5-brane.
As was discussed, for example, in [31], the basic superembedding condition (51)

imposes a relation on the superfields Xm (y, ) and (y, ), such that

Xm y M , = Xm y M + i m (y) + .
(59)
In particular, the superfield X11 up to linear order in is

11
X6,11
= x + i 11 + .
Recall that we introduced the factor to make x dimensionless.
In the geometry of X , the spinors and can be decomposed as

= 71 , 72 ,

= 71 , 72 .
(60)
(61)
(62)
Out of the 16-component spinors we only keep those components given by the covariantly
constant spinor along X . There are other physical degrees of freedom in the spinor , but
since we are considering a rigid curve in X only the above components lead to massless
degrees of freedom.
In Euclidean space equation (60) becomes

i
11
+ ,
X6,11 = x
(63)
R
where now chiral and anti-chiral spinors are independent from each other. (To give the
meaning to the fermionic bilinears in Euclidean space, we first define them in Minkowski
space, where

0 1
= ,
0 =
,
= ,
(64)
1 0
where the spinors indices are lowered via := . Then we continue to Euclidean
signature by dropping the reality conditions in Eq. (64).)
11 and X 11 match each other at the boundary
From Eqs. (54) and (63) we see that X3,11
6,11
of the M2-brane, i.e., at t = x, if the following boundary conditions are imposed on the
physical fermions 3,11

Y
Y

3,11
= 71Y ,
3,11
= 72Y ,
(65)
t =x
t =x
where Y (Y ) are chiral (anti-chiral) spinors indices on X .
One can identify zero modes living on the boundary of the M2-brane 71Y with
the supersymmetry broken by the M2-brane. This is compatible with our considerations in
133
Section 3. Indeed, exactly these components of background supersymmetry are broken for
the instanton described by the embedding Eq. (54).
The bosonic part of the M2-action is

A = A i a i x i i ,
SM2 = Z, Z = Ri a i x i A,
(66)
where, as in Section 2, i [2i ] is the homology class of the boundary curve.
Now, by evaluating the embedding matrices for an M5-brane up to linear order in and
solving the equation in the (6|16) superspace (see [34])
dB = H + f C,
we obtain the expression for
(67)
By(2)
y
up to linear order in

(2)
n
m
+ Cnm P P .
By(2)
y = Ay y + iy X y X nm
(68)
In solving Eq. (67) we have used constraints on the superform H. Specifically, we have
used the condition that the only non-vanishing components of the superform H , in the
basis eA , are the components Habc with all three bosonic indices. This was derived in [34]
by requiring -supersymmetry of the action of an M2-brane ending on an M5-brane.
In (68) we have dropped terms containing derivatives of the fluctuating fields such
as A, since these terms in the vertex operator will not contribute to the fermion twopoint function we wish to compute.
Now, from Eqs. (52) and (68), and using the properties (32) of covariantly constant
spinors on X , we evaluate the interaction between zero-modes of fermions living on an
M2-brane boundary and fermions to be

V = i i a i .
(69)
Note that the contribution to the interaction from the second boundary term in Eq. (68)
was cancelled by the term from the bulk, after integrating by parts, due to the presence of
the piece in the embedding (54) which was quadratic in fermions.
The last step in deriving the vertex operator for the chiral fermion superpartner of Z,
denoted , is to relate and . To achieve this we consider a supersymmetric variation
of Z with supersymmetry parameter (7 71 , 7 72 ). The result is

Z = Ri a i x + x Ri a i i A,
(70)
where

i
x = i7 (11) = 7 + 7 ,
R

i
A = i a 7 7 .
(71)
(72)
Eq. (72) is a direct consequence of Eq. (68) and the definitions (23), (66).
Denoting by i the superpartners of the bulk scalars T i , we get the desired relation

= 2 i a i + x i i
(73)
and hence the vertex operator for is
i
V = .
2
(74)
134
5. The case of one M5-brane

In this section we will discuss the scalar potential for the case of one M5-brane located
at position y = x on the M-theory interval. The general formula has been quoted above
in (1). In order to evaluate this expression we need both K and W . We will describe first
W and then K, and then put them together.
5.1. Superpotential
In the present setting the superpotential W can be written as a sum of 5 pieces
W = Wpert + W2 + W3 + W4 + W5 .
(75)
which have the following origins:

Wpert is the Yukawa superpotential for the charged chiral superfields, given in [17,45]
(4) 2
I1 I2 I3 C I1 C I2 C I3 .
Wpert =
(76)
3
The Yukawa couplings are given by

x
I1 I2 I3 = uxI11 uxI22 uI33 fx(1Rx21xR3 2 R3 )
(77)
and depend on the complex structure and bundle moduli, but are independent of T i and S.
fx(1Rx21xR3 2 R3 ) projects onto the singlet in R1 R2 R3 (if it exists), and is a choice of
nowhere zero holomorphic (3, 0) form on X .
W2 is the sum of two pieces coming from an M2-brane stretched between the M5-brane
and the boundary 9-brane at y = 0 or y = 1, respectively. We have shown in the previous
section that

W2 = h exp(Z) + exp i T i Z .
(78)
(Note that the relative sign can be changed by a shift in the imaginary part of T or Z.)
W3 is the contribution to the superpotential due to gaugino condensation studied in
[4,15,17]. It is given by

3
W3 exp
(79)
S ,
2b0
where b0 is a beta-function coefficient for the gauge group on the second 9-brane.
We are working in a region of moduli space constrained by (11). It follows that
3V 2b0Ra i i x,
3V 2b0Ra i i (1 x)
(80)
and hence the contribution of W3 to the potential is much smaller than the contribution of
W2 , and will henceforth be neglected.
W4 is the contribution from M2-brane instantons stretched between the two
M9-branes. The contribution from a single membrane wrapping a holomorphic curve
X has the form

W4 exp i T i .
(81)
135
In the case of the standard embedding (with no 5-branes) the sum over all such curves in a
fixed homology class vanishes. This happens because W4 is just the worldsheet instanton
contribution to the superpotential in the effective theory near a (2, 2) vacuum of the weakly
coupled heterotic string. Such superpotentials for moduli are known to be zero [2426].
W4 is also zero for the special cases of the non-standard embedding arising in weakly
coupled heterotic (0, 2) vacua which are related to linear sigma models. For example,
W4 = 0 for the quintic in P 4 . Nevertheless, it is expected that these corrections will be
nonzero for generic (0, 2) compactifications [13,14].
W5 is the superpotential coming from an M5-brane wrapping the whole X .

1
W5 exp(M5 Sv) = exp S .
(82)
4
W5 is of the same order as W3 and again we can neglect it relative to the effects of open
membranes.
5.2. Khler potential for bulk moduli and charged scalars
To evaluate the scalar potential in (1) one also needs the Khler potential 4
K = KS + KT + Km + Kcplx + K5 + Kbundle.
(83)
The first four pieces in this expression have already been obtained in previous papers.
We will derive a formula for K5 below. It would be interesting to learn more about Kbundle ,
but we will not do so in this paper. In this section we review the results for the first four
terms, obtained in [16,17,4648].
The first two terms in (83) are:

i
j
k

1
i
j
k
T +T
. (84)
KT = ln dij k T + T T + T
KS = ln S + S ,
6
To leading order in an expansion in C I the charged matter has a Khler potential of the
form
J + .
Km = ZIJ C I C
(85)
Here ZIJ is constructed from the metric for bundle moduli GB IJ as follows. First of all,
GB is defined by

1
mm
GB IJ =
(86)
g g uI mx uxJ m
vV
X
and depends on the Khler moduli a i , as well as on the complex structure and bundle
moduli. Next we define KB IJ := eKT /3 GB IJ . Note that the dependence on the Khler
moduli is only through the combination T i + T
i . Then we can define
ZIJ = GB IJ
eKT /3 2i i

BI J
S +
4 Here we are assuming that the moduli space is a product space, as is valid in our approximation.
(87)
136
where i was defined in (16), and

KB IJ
2 KT
,
K
=
,
T ij
BI J
T j
T i T
j

2
j
i = i T i + T
i KB IJ T i + T
i T k + T
k KT kj ,
BI J
BI J
BI J
3
ij
i = KT
(88)
ij
KT denote the inverse of the matrix KT ij .

In formulae (87) and (85) the following restrictions on the scalar fields are assumed
J 1,
ZIJ C I C
2i i
KB IJ .
S +
S B IJ
The Khler function for the complex structure moduli is

Kcplx = ln a Ga , Ga = a G, a = 1, . . . , h21 + 1
(89)
(90)
(91)
and can be expressed in terms of the periods over the A-cycles a and the prepotential G,
with complex structure moduli expressed as = / 0 , a = (0, ).
5.3. Khler potential for the M5-brane moduli
The last piece in (83) is K5 , the Khler potential giving the kinetic terms for the 5-brane
scalars x and A. As we were finishing our project we found that [39] obtained K5 in the
special case of h(1,1) = 1. Since we got our result independently and in a different way, we
will explain the derivation below.
To find K5 we start from the bosonic part of the PastiSorokinTonin action for the
M5-brane [35]

1
SM5 = M5 d 6 y det(MN + iHMN )
vL H LMN HMNP v P
4
W6

3 .
6 + 1 dA2 C
+ M5
(92)
C
2
W6
Here the tension of the M5-brane is M5 =

defined by

MN =
1
M6 .
(2)5 11
The other terms in the action are
XM XN (11)
g N ,
y M y N M
H MN = H MNP vP ,
1
7 MNP LKQ HLKQ ,
H MNP =
3!

HMNP = 3[M A(2)
NP ] (C3 )MNP ,

M1
M6
M1 M6 = X X CM
C
1 M
6 ,
y M1
y M6
M
N
P
MNP = X X X CM
C
N
P
,
y M y N y P
vN =
K K
3 .
6 is the magnetic dual of C
where is the PST scalar and C
vN v N = 1,
(93)
137
We wish to do KaluzaKlein reduction of the above action along the holomorphic

curve . We split the coordinates in the bulk as XM = ( , Xa , Xa , X11 ) where a, a =
1, . . . , 3 are indices for the complex coordinates, and are coordinates along the
noncompact R 4 . We choose to run over = 0, 1, 2, 5. 5 The coordinates along the
worldvolume W6 of the 5-brane are taken to be y M which we split as 4 real coordinates y ,
= 0, 1, 2, 5 and one complex coordinate y along the holomorphic curve . A natural
gauge choice for the PST scalar is [36]
5
,
vM = M
A5M = 0.
(94)
While the gauge choice (94) breaks six-dimensional covariance, after the KK reduction we
will obtain a covariant 4-dimensional action.
The massless fluctuations of the M5-brane are described by fields

X11 = x ,
Amn ,
A , = (m, 5), m = 0, 1, 2,
where A( ) was defined in Eq. (23). Keeping only terms of quadratic order in derivatives
we obtain the following 4-dimensional action

1 (a i i )RV y y 2
1
e(a i i )R
1/3
()2
( x) x
H
SM5 = v M5
2
V
2 eg 55
W4

mn

1
e
m A + xm i i g(3)
n A + xn i i
+ ()2 i
2
(a i )RV
5m

g
+ () 55 m A + xm i i H y y + () 5 A + x5 i i H y y
g

1
2 ex
55
i
+ () 2 5 g 5 A + x5 i
2
V

1
2 ex m
i
+ () 2 m A + xm i
2
V
(95)

where e = det g , while g 5m and g 55 are components of the inverse of the 4-dimenmn is the inverse of the 3-dimensional metric
sional metric g . One should take care that g(3)
so that
mn
g mn = g(3)
+
g 5m g 5n
.
g 55
Moreover, in (95) we have

1
H y y = 7 mnp m Anp m X11 C11np
2
(where we have used (93)) and finally we have also introduced

m = g m .
A = A i a i x i i ,
5 We have changed notation from Section 2 for this computation.
138
One can see from (95) that integrating out H y y gives

e
g 5 A + x i i .
H y y = ()
i
(i a )RV
Plugging the expression for H y y back into (95) restores 4-dimensional covariance and
results in the action

1

2

1
1 (a i i )R
2
( x) x +
4 SM5 = e
A + x i i
i
2 V
2 (a i )RV
W4

1
x
+ 2 A + x i i
.
2
V
(96)
One can now extract the Khler potential for the 5-brane moduli. The terms in the action
(96) uniquely determine K5 to be
K5 =
)2
(Z + Z
.
(S +
S )(i T i + i T
i )
(97)
A check of the supergravity kinetic terms associated with K5 shows that but there are extra
terms coming from (97) and given by:

1

e x a i i RV 2 ( x) V x 2 V 2 R a i i V
2
W4

1 2 i
3
V V + .
+ x a i RV
2
(98)
S ) after
These terms are exactly cancelled by the terms coming from KS = ln(S +
including an x-dependent correction to the definition of the chiral field S

S = V + R a i i x 2 + i.
(99)
Note, that the above correction R(a i i )x 2 is of order E eff with respect to V . There are
no x-dependent corrections to the other fields at this order.
The proper interpretation of these facts is that the complex structure on field space is
corrected at the nonlinear level by (99) and that the Khler potential KS + K5 should be
written as

)2
(Z + Z
S = log S +
S
.
K
(100)
(i T i + i T
i )
It would be interesting to learn if this expression is valid at higher order in the expansion
in Z.
5.4. Potential in the case h(1,1) = 1
In the previous sections we have given formulae valid for a generic X . We will now
specialize to the case of h(1,1) = 1 in order to simplify the analysis of the potential. As
139
we have stressed above, in this case there is no net contribution to the non-perturbative
potential from M2-branes stretched between the two boundary 9-branes, i.e., W4 = 0.
When h(1,1) = 1 the dependence on the Khler parameter a of the metric (86) for the
bundle moduli can be easily extracted by a scaling argument. We can choose a basis uxIm
that does not depend on a. Then the inverse of CalabiYau metric scales like
1 mm
mm
gCY
= (1)
,
a
where (1),nm is, say, an integral generator of H (1,1)(X ) H 2 (X , Z). Under these
conditions the Khler metric for charged scalars (88) simplifies considerably and is given
by

3
2
ZIJ =
(101)
+
HIJ , = (0) + (1 x)2 ,
T + T
S +
S
where HIJ depends only on complex structure and bundle moduli. In the case at hand
H2 (X , Z) is of rank 1 and generated by a rational curve . We take = 1, which
corresponds to wrapping a 5-brane only once around .
The perturbative potential for charged scalars was obtained in [17]. Using the
formulae from the appendix we have calculated the non-perturbative potential including
explicit leading dependence on Khler moduli and charged scalars. As mentioned in the
introduction we write the full potential in the form
(4 )4 U = (U0 + U1 + U2 ),
(102)
where, as mentioned in the introduction, we organize terms by the order in the

nonperturbative superpotential. U0 begins with the perturbative potential. U1 results from
mixing between the perturbative and nonperturbative contributions to W , while U2 is the
term of second order in the nonperturbative superpotential. The formula for the potential
contains a prefactor eK . We have used the explicit results for KS and KT , and we have
dropped Km and K5 since they contribute subleading effects to the order we are working.
We now give the leading expressions for the three terms in (102) in more detail. The
leading contributions to U0 are given by
U0 =
4 2 eKcplx +Kbundle
|CC|2 + UD .
VJ2
3d
(103)
3 , J := Ra, where d = 6d is the intersection number on X .

In the above formulae V = da
The expression
J1 C
J2 ,
|CC|2 = C I1 C I2 I1 I2 I3 H I3 J3 J1 J2 J3 C
(104)
comes from derivatives of W with respect to C I .

The D-term is given by
UD =
18 2
I
2
C HIJ T (a)C J ,
VJ2
(a)
(105)
140
where T (a) are generators of the unbroken four dimensional gauge group H , and we are
assuming there are no induced FI parameters.
To leading order U1 is given by:
eKcplx+Kbundle (1 x)
J2
2dV

J x

i1 eJ (1x) Re Wpert he
i2 ,
e
Re Wpert he
U1 =
(106)
where we define the axion fields

1 = Im Z,
2 = Im(T Z).
(107)
There will be corrections from terms higher order in the expansion in |C|2 /J , E eff , EReff .
There are also corrections from multiply-wrapped membranes.
The leading contribution to U2 is a two-instanton term
eKcplx +Kbundle 2 2J x
U2 =
|h| e
+ e2J (1x) 2eJ cos(1 2 )
2
8dJ

2J
4J
x
+
(108)
(1 2x)e2J (1x) +
eJ cos(1 2 ) + .
3V
3V
The leading piece comes from K Z Z |Z W |2 . Note that in the second line of (108) we
have kept terms which are formally higher order in our expression since they multiply
J /V E eff . We have kept these because, near x = 1/2, cos(1 2 ) = 1 the leading
piece vanishes. At these points the order J /V corrections which come from K Z Z and the
prefactor eK multiply zero and we can legitimately say that the leading term near x = 1/2,
cos(1 2 ) = 1 is given by the last term in the second line of (108). We will need these
subleading terms in Section 5.5. Of course, there are many other corrections of relative
order O((E eff )p (EReff )q , |C|2 /J ), where p 1 and q > 0.
The region of validity of our result for the potential, (102), is constrained by several
considerations.
We must assume that all sizes are much bigger than the 11D Plank length.
Rx l11 ,
R(1 x) l11 ,
a 1/2 v 1/6 l11
(109)
and from these conditions it follows, in particular, that J x 1, J (1 x) 1.

Since we are working to quadratic order in the Khler potential in a series expansion
in C we must have
J J.
|C|2 := C I HIJ C
(110)
The effective expansion parameters should be small, or, equivalently,

V J,
J2 V.
(111)
E eff
We must be able to drop

corrections to each of the 9 terms in the potential. We
count all the terms which have different structure, i.e., 5 from U2 , 2 from U1 and 2 from U0 .
U=
9

a=1

ua 1 + fa E eff ,
iff
141
9
9

E eff
fa ua
u a .

a=1
a=1
Given our ignorance regarding fa we will assume that they are O(1) and impose the
stronger condition

9
9

eff
|ua |
u a .
E
(112)

a=1
a=1
Finally, as mentioned in the introduction, we should stress that we are assuming that we
are working at a generic smooth point in the complex structure and bundle moduli space.
5.5. 5-brane dynamics
We can get some heuristic idea about the 5-brane dynamics by considering the theory
on a finite volume of 3-dimensional space and keeping only the spatially homogeneous
modes of the scalar fields. Even in this drastic approximation the resulting system is a
very complicated dynamical system described by a particle mechanics Lagrangian with
the (very) schematic form
2 2

2 (Jx + xJ
V
)2
J
(C)
vol(space) dt
(113)
+
U ,
+
+
V
J
J
VJ
where the potential energy is

1
4
3 J x
J (1x)
U=
e
(1
x)|C|
e
C
VJ2

2 2J
4J x J
+ V eJ x eJ (1x) +
(1 2x)e2J (1x)
e
.
3V
3V
(114)
We have only kept the real parts of the fields. The philosophy behind this is that by a
BornOppenheimer type approximation we expect that the axions will relax much more
rapidly than the real parts into the most attractive channel. The choice of depends on
what term is dominating, U1 or U2 .
The positive constants , and are functions of the complex structure and bundle
moduli, but these are being held fixed in this discussion.
While the dynamical system we must study is rather complicated we can get some
heuristic idea of what to expect in three distinct regions within the region of validity of
our potential.
In one region the charged scalar fields are zero, while J and V are large. The leading
contributions to the potential are positive and decrease with increasing J , V . In this region
the 5-brane leads to a repulsive interaction between the M9 walls. Setting C = 0 and
choosing a sign to set the axions in the most attractive channel in Eq. (114) we get:

2 2J
4J x J
1
(1 2x)e2J (1x) +
e
.
U 2 eJ x eJ (1x) +
(115)
J
3V
3V
142
Note that U has a minimum in x at x = 1/2. Expanding around the minimum x = 12 + y,

the resulting potential is

2 J
U =
(116)
e + 4y 2 eJ .
3V J
Now we can see the need for keeping the last two terms in the expression for U2 in
Eq. (108). Although we are neglecting the E eff -corrections to the Khler potential, such
corrections multiply the leading three terms in Eq. (108) which sum up to zero at x = 1/2.
On the other hand, the terms we have written, and which follow from the leading pieces
in K become the leading terms at the stationary points x = 1/2. Consequently, J flows
towards infinity and x moves towards the middle of the interval. We must assume V 1/2
J V to stay within the region of validity.
If the charged scalar fields are important in such a way that
|U1 | U2 ,
(117)
then the leading x-dependent term in the potential is attractive:

U = C 4 (1 x)|C|3 eJ x + eJ (1x) .
Note that in this case we have to choose the + sign in Eq. (114) if the axions are in the
most attractive channel. Thus, if x < 1/2 the 5-brane moves towards the wall at y = 0, and
if x > 1/2 the 5-brane moves towards the wall at y = 1. Note, that in each of these two
subregions U0 |U1 | as a direct consequence of (117). Since U0 is the dominant term
J and V will evolve to large values. Since the C field is simultaneously evolving a more
careful analysis of the dynamical system would be highly desirable. But we will not do
that here.
More generally, one can show that the potential (114) is non-negative at a generic
point in the bundle and complex structure moduli space, within in our region of validity
(109)(112), and thus predicts decompactification of both the CalabiYau and the orbifold
interval. The argument, which is straightforward but long can be found in Appendix B.
Note that (114) is the leading potential only under our assumption that we work at a
generic smooth point in bundle and complex structure moduli space. It would be interesting
to incorporate singularities in complex structure and bundle moduli space in the discussion.
There are potentially many new terms in the potential that must be reconsidered. It is
possible that using the known results on complex structure and bundle moduli space one
can address this problem.
5.6. Conflicting instabilities
One interesting consequence of the discussion in the previous section is that there is a
strong coupling dual of the DineSeiberg problem where the M-theory interval (and the
CalabiYau) tends to decompactify. In the case of heterotic M-theory with the standard
embedding (i.e., no 5-branes) this has already been discussed by Banks and Dine [5], who
noted that one can use holomorphy to extrapolate the weak coupling superpotential based
on gluino condensation. In the presence of an M5-brane (in the case h(1,1) = 1) the above
143
formulae show that in the region specified by (109)(112) there is a similar effect due to
open membrane instantons.
It is of some interest to compare the above result with what we expect for the weakly
coupled heterotic string, since our considerations are only valid at large heterotic string
coupling. Indeed, the heterotic coupling is related to the length of the M-theory interval by
R l11 (gs )2/3
and we require R l11 . In the regime of weak coupling and large V , the potential
has been discussed in [27]. It was shown there that the effective potential is positive and
behaves as
Ueff eV /gs .
2
(118)
This favors an evolution to weak coupling gs 0 and large volume V . One might
worry that the calculations of [27] were performed in the case of the standard embedding,
and in backgrounds with other E8 E8 gauge instantons one must take into account the
contribution of worldsheet instantons as well [13,14]. Nevertheless, as we have repeatedly
mentioned, these effects often sum to zero [21,22] so once again and we can use (118) in
the region of small R and large V .
In view of the above, we can combine our result (102) with (118), to learn that the
true potential goes to zero through positive values in both limits R 0 and R .
This suggests that there must be a stationary point somewhere in the intermediate region,
i.e., at some finite value of R. The nature of the stationary points that lie in the middle of
moduli space is unknown, and is, of course, an interesting and outstanding question.
6. Multiple covers and chirality changing transitions

It is of considerable interest to determine the nature of the low energy theory in the limit
that the M5 moves into the the boundary 10-manifold. In this section we will make some
comments on this limit. We will need to make some guesses and the results of this section
are not as rigorous as in the previous sections. For definiteness we will consider the limit
x 0.
One good reason for studying the limit x 0 is that there are strong indications that
in such limits there can be remarkable chirality-changing phase transitions in the low
energy theory. This was discovered, in the present context, by Kachru and Silverstein [56],
building on [57]. The theory of these transitions has been considerably extended to many
new examples in [42].
The main new ingredient that is needed to discuss the x 0 behavior is a multiple-cover
formula for the open membrane instantons. The fact that there must be nontrivial effects
from multiply-wrapped M2-branes (at least for those stretching between two 5-branes)
can be seen by considering the holomorphy of the full superpotential W as a function
of Zi Zj [8]. The instanton effects must be suppressed by a factor proportional to the
volume of the stretched membrane and therefore must behave like exp[(Zi Zj )] for
144
Re(Zi Zj ) > 0. This is only consistent with holomorphy if there is an infinite series
with at best a finite radius of convergence.
Multiply-wrapped worldsheet instanton corrections to d = 4, N = 2 prepotentials are
known to have a universal form f (n, ) for an n-times wrapped curve where f only
depends on the topology of [5863]. Since worldsheet instantons are special cases of
M2-brane instantons we will make the working hypothesis that there is similarly a multiple
cover factor f (n, ) for M2-brane instanton corrections to the superpotential W . Some
evidence for this can be found in [5355]. Unfortunately, the topologies studied in the
above papers do not contain our case of P 1 [0, 1]. Therefore we will take
W = h
f (n)e
nZ
+e h
i
n=1
f (n)en(T Z)
(119)
n=1
and make the rather weak assumption that the asymptotic behavior of f (n) for large n is
f (n) nm eJ x0n for some constants m and x0 . For simplicity we set x0 = 0 although there
could in principle be a shift in the location of the small instanton transition.
The constants m and above are unknown, but we can make some comments on
them. First, the relative phase ei was not important in the 1-instanton sector, where we
can change the relative phase by shifting the axion Im T . It does become a nontrivial
issue in the multi-instanton sectors. Nevertheless, for our analysis of the dynamics in the
subregion Re Z 1 the second piece in (119) is negligible, so the issue need not concern
us here.
Next, let us consider the power m in the asymptotic behavior of f (n). If we wish the
chiral fermion mass term in standard supergravity to be a single-valued function of Z in a
region surrounding Z = 0, then m cannot be a negative integer such that |m| > 2. Singlevaluedness implies that the monodromy of Z Z W around Z = 0 should be diagonalizable
thus excluding a singularity of the type Z n log Z, n 2 in W . 6
Let us now re-consider the region of validity of our expression. The infinite series for
W can be obtained reliably in the region Re Z 1 (where we can use 11-dimensional
supergravity) and then analytically continued to the region where Re Z = J x 1. To
ensure that corrections to the Khler function are small one must still require that
V 1,
V 1/2 J V ,
|C|2 J.
Since these conditions do not imply J x 1 or J (1 x) 1 we can study the physics of

the 5-brane approaching the boundary. 7
For definiteness and simplicity let us now assume that f (n) = nm for some constant m.
Then we have

Z W = h Li(m+1) eZ ,
(120)
6 There are known examples of logarithmic superpotentials for n = 0, 1 that make good physical sense. This
is usually related to some kind of pair creation phenomena. We thank K. Hori for very useful discussions on this
issue.
7 This last statement might be too naive given the presence of a strongly coupled light noncritical string.
145
where Li is a polylogarithm function

Li(m+1) (t) =
nm+1 t n .
n=1
In this case the leading order contribution to U1 is given by:

U1 =

eKcplx +Kbundle (1 x)
Re Wpert h Li(m+1) eZ
2
J
2dV
(121)
while the leading contribution to U2 is

U2 =

2
eKcplx +Kbundle 2
|h| Li(m+1) eZ .
2
8dJ
(122)
In all the cases below we will assume that the axion phases are in the maximally
attractive channel.
One interesting limit is Z 0. Here we can use the behaviour of the polylogarithm

Li(m+1) eZ Z (m+2) , Z 0
(for m = 2 we replace Z 0 by log Z) to write out (schematically) the leading potential at
Jx 1

1
|C|3
V
4
C
, m > 2,
U=
(123)
(1
x)
+
VJ2
(J x)2+m
(J x)4+2m

1
4
3
2
U=
(124)
+
|C|
(1
x)
log(J
x)
+
V
(log
J
x)
C
, m = 2,
VJ2
where , , are positive functions of complex structure and bundle moduli as above and
we have set Im Z = 0.
Therefore, for small enough J x (holding the other moduli fixed) the leading term in the
potential (123), (124) is
(log J x)2
,
m
>
2,
, m = 2
(125)
J 2 (J x)4+2m
J2
and there is a repulsive force on the 5-brane. Indeed there is an infinite energy barrier
forbidding the 5-brane from hitting the wall.
One should not conclude from the above that there will be no chirality changing
transition, since the axionic degree of freedom in Z can change the qualitative features of
the potential drastically. Unfortunately, in order to study this question in detail just knowing
the asymptotic behavior will not suffice and one needs a precise version of the formula for
f (n) in order to work out the analytic continuation from Re(Z) 1 to |eZ | = 1. For
definiteness we will consider f (n) = nm where m 2 is integral.
Let us consider first the case m 0. We take Z = J x + i , where J x 1 and expand
Z
e = 1 + J x 12 (J x)2 . We use the Taylor expansion for the polylogarithm
Li(m+1) (1 + t) = Li(m+1) (1) Li(m+2) (1)t

1
+ Li(m+3) (1) Li(m+2) (1) t 2
2
146
and the following useful relations

Lim (1) = 21+m 1 (m),
B2k
, k = 1, 2, . . . ,
2k
where is the Riemann -function and Bk are Bernoulli numbers taken in the convention:
(2k) = 0,
ey
(1 2k) =
y
1
B2 2 B4 4
=1 y +
y +
y + .
1
2
2!
4!
Substituting t = J x 12 (J x)2 in the above Taylor expansion and keeping only terms up to
(J x)2 we have

Li(m+1) (1 + t) = (1)k+1 1k 2k (J x)2 , m = 2k, k = 0, . . . ,
(126)
Li(m+1) (1 + t) = (1)k+1 22k J x,
m = 2k + 1, k = 0, . . . ,
(127)
where we define positive numbers 1k , 2k

|B2k+2 |
,
1k = 22k+2 1
2(1 + k)

|B2k+4 |
2k = 22k+4 1
.
4(2 + k)
We now analyze the potential separately for the cases of even and odd m. For even
m = 2k we have the following leading potential at J x 1

k 2

1 4
3 k
k
2
k k
2
C
.
U=
(1
x)|C|
(J
x)
(J
x)
1
2
1
1 2
VJ2
(128)
If C $= 0 then for sufficiently small x the potential is attractive. If C = 0 the potential is
repulsive.
Now, for odd m = 2k + 1 the leading potential is

k 2
1 4
k
3
2
2
(1
x)J
x|C|
+
V
2
(J
x)
C
.
U=
(129)
2
2
VJ2
Now the situation is opposite to the previous case. For C $= 0 there is a repulsive force and
if C = 0 an attractive force.
Finally, we analyze what happens for the cases m = 1 and m = 2, assuming
Im Z = and J x 1. If m = 1 the leading potential is

V
1
1
1
4
3
|C|
J
x
+
(1
J
x)
.
C
(1
x)
1
U=
(130)
VJ2
2
2
4
The force on the 5-brane is attractive only if one allows for a large vev for |C|3
|C|3 > V .
This is in principle possible since we only assume that |C|3 V 3/2 and for large V both
inequalities can be satisfied. If m = 2 the leading potential is

1
1
4
3
J
x
+
V
log
2(log
2
J
x)
|C|
(1
x)
log
2
C
U=
(131)
VJ2
2
147
and the force is attractive only if

|C|3 > 2(log 2) V .
In both cases m = 1 and m = 2, attraction is only possible for large vev of charged
scalars.
The general conclusion based on the analysis of various cases is that the physics of what
happens when the 5-brane approaches the wall depends strongly on the detailed form of
the multiple-cover formula.
Finally, let us comment on the relevance of this computation to the examples studied
by Ovrut, Pantev, and Park in [42]. One might at first conclude that in these examples the
superpotential must vanish since the five-brane wraps a high genus curve. However, the
curve wrapped by the 5-brane is not irreducible and not isolated. It can very well happen
that in the long-distance expansion of the M5 and M2 Lagrangians there are terms with
many fermions (typically multiplying factors involving curvature tensors) which can lift
the many fermion zeromodes associated with the nonisolated high genus curve. Thus, the
question of whether or not a superpotential is generated is a complicated and difficult one,
involving a discussion of the measure on the moduli space of the curve and the integral
over that moduli space. Considerations based on the global form of the moduli space for
these curves based on the results of [64] do not appear to exclude the generation of such
superpotentials.
7. The case of N M5-branes

We will now briefly consider the potential in the case that there are N M5-branes at
positions x1 < x2 < < xN . We will assume for simplicity that all the 5-branes are
wrapped over the same rational curve , so that open M2 instantons can be stretched
between any pair of 5-branes. Moreover, to simplify the analysis we assume that the
5-branes are more or less evenly separated. Finally, we restrict our consideration only to the
leading non-perturbative potential, so we do not take into account 2 instanton contributions
to W and we need only consider M2-branes between neighboring 5-branes. Similarly,
we only keep the contribution of 59 instantons coming from M2 instantons stretching
between the boundary and the nearest M5-brane. Under these conditions we will have
R(xn xn1 ) l11 ,
n = 1, . . . , N + 1.
Neglecting E eff corrections due to the distortion of the background, the Khler function
for the collection of 5-branes will be just a sum of Khler functions for each 5-brane.
The potential is again given by formula (102), with the same conditions on the region of
validity. The 2 instanton terms in the potential U2 , which dominate at C = 0, are:
U2 =
eKcplx +Kbundle 2
|h|
2
8dJ
148
N

2J (x xn )
n+1
+ e2J (xnxn1 ) 2eJ (xn+1xn1 ) cos( n ) + ,
e
n=1
(132)
where we denote

n = a(2An An+1 An1 ) + (xn+1 + xn1 2xn ) ,
n = 1, . . . , N
(133)
and x0 = 0, xN+1 = 1, A0 = AN+1 = 0. If instead we assume that

x1 xn xn1 ,
(1 xN ) xn xn1 ,
1 < n N
(134)
and choose a special subregion where cos n = 0, n, then the potential has the form of
a non-periodic Toda-chain potential. (The kinetic energies are the standard ones, in our
approximation.) As is well known, Toda theory has an exact solution, where all particles
move away from each other [41]. In heterotic M-theory this signals an instability in the time
evolution of the positions of M5-branes along the orbifold interval: they tend to run away
from each other. At the same time Ra evolves to infinity. In short, the system explodes.
Using again a BornOppenheimer type approximation we expect that the axions will
relax much more rapidly than the real parts into the most attractive channel cos n = 1, n.
This implies that the evolution with a Toda-like potential is unstable because of the axions.
When the charged vevs are nonzero we should consider instead the term U1 in the
potential arising from cross terms between perturbative and nonperturbative pieces. This is
given by

eKcplx+Kbundle |h| J x1
e
(1 x1 ) cos(1 ) eJ (1xN ) (1 xN ) cos(N )
U1 =
2V
2dJ

N1

J (xn+1 xn )
(xn+1 xn )e
cos(n ) ,
(135)
n=1
where Wpert = ei1 , h = |h|eih are decompositions into modulus and phase and
N = Im(T ZN ) + 1 h ,
1 = Im Z1 + 1 h ,
n = Im(Zn+1 Zn ) + 1 h , n = 1, . . . , N 1.
(136)
8. Possible future directions and applications

A central question in heterotic M-theory is the existence of isolated minima of the
potential for moduli. While most of our results predict runaway or unstable behavior (as
expected) we have seen some encouraging hints. We have argued that the potential must
have nontrivial stationary points in moduli space. We have also seen that a good place
to look for interesting behavior of the potential is at singular loci in complex structure
and bundle moduli space. For example, if one allows some of the coefficients , , in
Sections 5.5 and 6 to vanish it is easy to imagine scenarios where the potential predicts
compactification, rather than decompactification.
149
There are many technical issues raised by the above computations which should be
solved and which moreover can be solved with presently available technology.
One circle of questions includes finding the appropriate generalization of the multiplecover formula for worldsheet instantons. A related set of questions concerns effects
associated with membranes wrapping higher genus curves and nonisolated curves in X . As
we have seen in Section 6, results on these questions would have very interesting physical
applications.
A second circle of questions concerns the possibility of obtaining a more concrete
understanding of the dependence of the membrane determinants as functions of the
complex structure moduli. It might be possible to find classes of compactifications in which
one can give fairly explicit formulae for the dependence on gauge bundle moduli, although
this might prove to be challenging.
Beyond the extensions mentioned above, which we believe are within reach, there loom
far more difficult questions. One of the most challenging issues is to give a proper definition
of HoravaWitten theory in a regime outside the validity of the expansion in (11 )2/3 .
Another difficult, and pressing, problem is that of finding ways to make definite and
quantitative statements about the Khler potential of the effective supergravity theory in
a wider range of validity.
Nevertheless, even given the limitations of our computations, the results do have some
interesting ranges of validity. It might be quite interesting to study more thoroughly
the dynamics, both classical and quantum mechanical of the moduli in the problem. In
this paper we have limited ourselves to some very heuristic and naive pictures of the
dynamics. It might also be fruitful to see if there are any distinctive features of the modular
cosmology resulting from the above potential for moduli [52].
Acknowledgements
We would like to thank Jeff Harvey and Marcos Mario for collaboration at some stages
of this research. We would also like to thank B. Acharya, T. Banks, D.-E. Diaconescu,
M. Douglas, S. Kachru, O. Ganor, S. Gukov, M. Mario, A. Mikhailov, N. Nekrasov, and
M. Rangamani for discussions and useful correspondence. The work of G.M. is supported
by DOE grant DE-FG02-96ER40949. The work of G.P. is supported by the Rutgers DOE
grant DE-FG02-96ER40949 and by the Yale DOE grant DE-FG02-92ER-40704.
Appendix A
For the convenience of the reader we list here the leading expressions for the Khler
potential in the case h(1,1) = 1, together with formulae for the Khler metric and inverse
metric. The Khler potential is:
K = KS + KT + Km + Kcplx + K5 + Kbundle,

+ T
)3 ,
S ),
KT = ln d(T
KS = ln(S +
150
)2
(Z + Z
,
(S +
S )(i T i + i T
i )

2
3

J
+
HIJ C I C
Km =
.
T +T
S +S
K5 =
(A.1)
We now give the components of the Khler metric on the space of scalars which have
been used in Section 5.4. We keep only leading terms in each of the component, neglecting
2
corrections of the relative order O(E eff , EReff , |C|
Ra ).
KS
S=
1
,
4V 2
KS T
=
KT
I =
KIJ =
J
H C I C
IJ
2V 2
J
3HIJ C
(2Ra)2
KS Z
=
KT
=
3
H ,
2Ra I J
KS J =
KS =
x2
,
4V 2
KI =
HIJ C I
,
2V 2
x
,
2V 2
KT T
=
J
3C I HIJ C
(2Ra)2
J
3 HIJ C
,
2Ra
3
,
(2Ra)2
KZ T
=
x
,
2V Ra
KZ I =
(1 x)HIJ C J
,
V Ra
J
(1 x)C I HIJ C
(cplx)
,
K = K ,
V Ra
Now, we solve the matrix equation

2
1
eff
eff |C|
KK = 1 + O E , ER ,
.
Ra
KZ
KZ Z
=
1
.
2V Ra
The inverse metric solving this equation is
K S S = 4V 2 ,
KIJ =
2Ra IJ
H ,
3
K Z Z = 2RaV ,
KT T =
(2Ra)2
,
3
K
= Kcplx ,
K Z S = 4RaV x,
2Ra
K T J = CJ
K H I J K ,
K J = HIKC
K ZT =
(2Ra)2
x,
3
(2Ra) J
C (2 x),
3
where the components not listed above are zero in our approximation.
KJZ =
Appendix B
In Section 5.5 we asserted that, within the region of validity of our computations, the
potential is always positive. Here we give the detailed proof of that claim.
The only potentially negative term in the potential is U1 . We will show that it cannot be
larger in magnitude than both of U0 and U2 in our region of validity.
151
First, imposing
|U1 | U2
means

2
(1 x)|C|3 eJ x + eJ (1x) V eJ x + eJ (1x) .
(B.1)
It follows immediately that

1/3
V J x
|C|
+ eJ (1x)
.
e
Now, at a generic point in bundle and complex moduli space, we have

1/3

J x
4
3 V
J (1x)
+e
|C|3 (1 x) eJ x + eJ (1x)
e
C |C|
and we see that U0 |U1 |.

Let us now assume
|U1 | U0 .
From this it follows that

(1 x) eJ x + eJ (1x) |C|
and hence

4
1 4
(1 x)4 eJ x + eJ (1x) .
(B.2)
3
Let us consider, first,the region far enough from x = 1/2. Then, for x < 1/2, we have
|U1 |
|U1 |
J 2V
1 4
(1 x)4 e4J x
J 2 V 3
and
1 2J x
e
.
J2
As a consequence, U2 |U1 |.
In the region close to x = 1/2 we have instead, for sign + in Eq. (B.2)
U2
|U1 |
1 4 2J
e
J 2 V 3
and
1 J
e
J2
and it follows immediately that
U2
U2 |U1 |.
For the sign in Eq. (B.2) the last statement is obvious.
152
References
[1] P. Horava, E. Witten, Heterotic and type I string dynamics from eleven dimensions, Nucl. Phys.
B 460 (1996) 506524, hep-th/9510209.
[2] P. Horava, E. Witten, Eleven-dimensional supergravity on a manifold with boundary, Nucl.
Phys. B 475 (1996) 94114, hep-th/9603142.
[3] E. Witten, Strong coupling expansion of CalabiYau compactification, Nucl. Phys. B 471
(1996) 135158, hep-th/9602070.
[4] T. Banks, M. Dine, Phenomenology of strongly coupled heterotic string theory, Nucl. Phys.
B 479 (1996) 173, hep-th/9609046.
[5] T. Banks, M. Dine, Couplings and scales in strongly coupled heterotic string theory, hep-th/
9605136.
[6] E. Witten, Non-perturbative superpotentials in string theory, Nucl. Phys. B 474 (1996) 343360,
hep-th/9604030.
[7] K. Becker, M. Becker, A. Strominger, Five-branes, membranes and non-perturbative string
theory, Nucl. Phys. B 456 (1995) 130152, hep-th/9507158.
[8] J. Harvey, G. Moore, Superpotentials and membrane instanons, hep-th/9907026.
[9] B. de Wit, K. Peeters, J. Plefka, Open and closed supermembranes with winding, Nucl. Phys.
Proc. Suppl. 68 (1998) 206215.
[10] E. Witten, Worldsheet corrections via D-instantons, hep-th/9907041.
[11] S. Kachru, S. Katz, A. Lawrence, J. McGreevy, Open string instantons and superpotentials,
Phys. Rev. D 62 (2000) 026001, hep-th/9912151.
[12] B.S. Acharya, M-theory, joyce orbifolds and super-YangMills, hep-th/9812205.
[13] M. Dine, N. Seiberg, X.G. Wen, E. Witten, Nonperturbative effects on the string worldsheet,
Nucl. Phys. B 278 (1986) 769789.
[14] M. Dine, N. Seiberg, X.G. Wen, E. Witten, Nonperturbative effects on the string worldsheet (II),
Nucl. Phys. B 289 (1987) 319363.
[15] P. Horava, Gluino condensation in strongly coupled heterotic string theory, Phys. Rev. D 54
(1996) 75617569, hep-th/9608019.
[16] A. Lukas, B.A. Ovrut, D. Waldram, Non-standard embedding and five-branes in heterotic
M-theory, Phys. Rev. D 59 (1999) 106005, hep-th/9808101.
[17] A. Lukas, B.A. Ovrut, D. Waldram, Five-branes and supersymmetry breaking in M-theory,
JHEP 9904 (1999) 009, hep-th/9901017.
[18] A. Lukas, B.A. Ovrut, D. Waldram, K. Stelle, Heterotic M-theory in five dimensions, hep-th/
9806051;
A. Lukas, B.A. Ovrut, D. Waldram, K. Stelle, Nucl. Phys. B 552 (1999) 246290.
[19] R. Donagi, A. Lukas, B.A. Ovrut, D. Waldram, Non-perturbative vacua and particle physics in
M-theory, JHEP 9905 (1999) 018, hep-th/9811168.
[20] R. Donagi, A. Lukas, B.A. Ovrut, D. Waldram, Holomorphic vector bundles and nonperturbative vacua in M-theory, JHEP 9906 (1999) 034, hep-th/9901009.
[21] E. Silverstein, E. Witten, Criteria for conformal invariance of (0, 2) models, Nucl. Phys. B 444
(1995) 161190, hep-th/9503212.
[22] J. Distler, S. Kachru, Singlet couplings and (0, 2) models, Nucl. Phys. B 430 (1994) 13, hep-th/
9406090.
[23] P. Berglund, P. Candelas, X. de la Ossa, E. Derrick, J. Distler, T. Hubsch, On the instanton
contributions to the masses and couplings of E6 singlets, Nucl. Phys. B 454 (1995) 127, hep-th/
9505164.
[24] P. Candelas, X. de la Ossa, Moduli space of CalabiYau manifolds, Nucl. Phys. B 355 (1991)
455481.
[25] A. Strominger, Special geometry, Commun. Math. Phys. 1333 (1990) 163.
153
[26] E. Witten, Phases of N = 2 theories in two dimensions, Nucl. Phys. B 403 (1993) 159, hep-th/
9301042.
[27] M. Dine, N. Seiberg, Is the superstring weakly coupled?, Phys. Lett. B 162 (1985) 299302;
M. Dine, R. Rohm, N. Seiberg, E. Witten, Phys. Lett. B 156 (1985) 55;
M. Dine, N. Seiberg, Phys. Rev. Lett. 55 (1985) 366.
[28] P. Townsend, Brane surgery, Nucl. Phys. B. Proc. Suppl. 58 (1997) 163175, hep-th/9609217.
[29] A. Strominger, Open p-branes, Phys. Lett. B 383 (1996) 4447, hep-th/9512059.
[30] P.S. Howe, E. Sezgin, Superbranes, Phys. Lett. B 390 (1997) 133142, hep-th/9607227;
P.S. Howe, E. Sezgin, D = 11, p = 5, Phys. Lett. B 394 (1997) 6266, hep-th/9611008;
P.S. Howe, E. Sezgin, P.C. West, Covariant field equations of the M-theory five-brane, Phys.
Lett. B 399 (1997) 4959, hep-th/9702008.
[31] D. Sorokin, Superbranes and superembeddings, Phys. Rep. 329 (2000) 1101, hep-th/9906142.
[32] E. Cremmer, S. Ferrara, Phys. Lett. B 91 (1980) 61;
L. Brink, P.S. Howe, Phys. Lett. B 91 (1980) 384.
[33] Y. Shibusa, 11-dimensional curved backgrounds for supermembrane in superspace, hep-th/
9905071.
[34] E. Sezgin, P. Sundell, Aspects of the M5-brane, hep-th/9902171.
[35] P. Pasti, D. Sorokin, M. Tonin, Phys. Lett. B 398 (1997) 41, hep-th/9701037;
I. Bandos, K. Lechner, A. Nurmagambetov, P. Pasti, D. Sorokin, M. Tonin, Phys. Rev. Lett. 78
(1997) 4332, hep-th/9701149.
[36] M. Aganagic, J. Park, C. Popescu, J. Schwarz, Worldvolume action of the M-theory five-brane,
Nucl. Phys. B 496 (1997) 191214, hep-th/9701166.
[37] J. Wess, J. Bagger, Supersymmetry and Supergravity, 2nd edn., Princeton Univ. Press, 1992.
[38] E. Cremmer, S. Ferrara, L. Girardello, A. Van Proeyen, YangMills theories with local
supersymmetry: Lagrangian, transformation laws and superhiggs effect, Nucl. Phys. B 212
(1983) 413.
[39] J.P. Derendinger, R. Sauser, A five-brane modulus in the effective N = 1 supergravity of
M-theory, hep-th/0009054.
[40] Ph. Brax, J. Mourad, Open supermembranes coupled to M-theory five-branes, Phys. Lett. B 416
(1998) 295302, hep-th/9707246.
[41] A. Perelomov, Integrable Systems of Classical Mechanics and Lie Algebras, Birkhauser Verlag,
1990.
[42] B. Ovrut, T. Pantev, J. Park, Small instanton transitions in heterotic M-theory, JHEP 0005
(2000) 045, hep-th/0001133.
[43] C. Munoz, Effective supergravity from heterotic M-theory and its phenomenological implications, hep-th/9906152;
D.G. Cerdeno, C. Munoz, Phenomenology of non-standard embedding and five-branes in
M-theory, hep-ph/9904444.
[44] E. Bergshoeff, E. Sezgin, P.K. Townsend, Supermembranes and 11-dimensional supergravity,
Phys. Lett. B 189 (1987) 75;
E. Bergshoeff, E. Sezgin, P.K. Townsend, Properties of the eleven-dimensional supermembrane
theory, Ann. Phys. 185 (1988) 330.
[45] M. Green, J.H. Schwarz, E. Witten, Superstring Theory, Cambridge Univ. Press, Cambridge,
1987.
[46] T. Li, J.L. Lopez, D.V. Nanopoulos, Phys. Rev. D 56 (1997) 2602.
[47] E. Dudas, C. Grojean, Nucl. Phys. B 507 (1997) 553.
[48] H.P. Nilles, M. Olechowski, M. Yamaguchi, Phys. Lett. B 415 (1997) 24;
H.P. Nilles, M. Olechowski, M. Yamaguchi, Nucl. Phys. B 530 (1998) 43.
[49] S. Stieberger, (0, 2) heterotic gauge couplings and their M-theory origin, Nucl. Phys. B 541
(1999) 109144.
154
[50] K. Choi, H.B. Kim, H. Kim, Moduli stabilization in heterotic M-theory, Mod. Phys. Lett. A 14
(1999) 125134, hep-th/9808122.
[51] K. Benakli, Phenomenology of low quantum gravity scale models, Phys. Rev. D 60 (1999)
104002.
[52] T. Banks, M. Berkooz, G. Moore, S.H. Shenker, P. Steinhardt, Modular cosmology, hep-th/
9503114;
T. Banks, M. Berkooz, G. Moore, S.H. Shenker, P. Steinhardt, Phys. Rev. D 52 (1995) 3548
3562.
[53] H. Ooguri, C. Vafa, Knot invariants and topological strings, Nucl. Phys. B 577 (2000) 419438.
[54] S. Kachru, S. Katz, A. Lawrence, J. McGreevy, Mirror symmetry for open strings, Phys. Rev.
D 62 (2000) 126005, hep-th/0006047.
[55] C. Vafa, Superstrings and topological strings at large N , hep-th/0008142.
[56] S. Kachru, E. Silverstein, Chirality changing phase transitions in 4d string vacua, Nucl. Phys.
B 504 (1997) 272284, hep-th/9704185.
[57] O. Ganor, A. Hanany, Small E8 instantons and tensionless noncritical strings, Nucl. Phys. B 474
(1996) 122, hep-th/9602120;
N. Seiberg, E. Witten, Comments on string dynamics in six dimensions, Nucl. Phys. B 471
(1996) 121, hep-th/9603003.
[58] P. Candelas, X.C. de la Ossa, P. Green, L. Parkes, in: S.T. Yau (Ed.), A pair of CalabiYau
manifolds as an exactly soluble superconformal theory, Essays on Mirror Manifolds.
[59] P.S. Aspinwall, D.R. Morrison, Topological field theory and rational curves, Commun. Math.
Phys. 151 (1993) 245.
[60] M. Marino, G. Moore, Counting higher genus curves in a CalabiYau manifold, Nucl. Phys.
B 543 (1999) 592614, hep-th/9808131.
[61] C. Faber, R. Pandharipande, Hodge integrals and GromovWitten theory, math.AG/9810173.
[62] R. Gopakumar, C. Vafa, M-theory and topological strings I, hep-th/9809187;
R. Gopakumar, C. Vafa, M-theory and topological strings II, hep-th/9812127.
[63] D.R. Morrison, M.R. Plesser, Summing the instantons: quantum cohomology and mirror
symmetry in toric varieties, Nucl. Phys. B 440 (1995) 279354, hep-th/9412236.
[64] R. Donagi, B.A. Ovrut, D. Waldram, Moduli spaces of fivebranes on elliptic CalabiYau
threefolds, hep-th/9904054.

AdS/CFT in the infinite momentum frame

D. Brecher a , A. Chamblin b , H.S. Reall c
a Centre for Particle Theory, Department of Mathematical Sciences, University of Durham, South Road,
Durham DH1 3LE, United Kingdom

b Center for Theoretical Physics, Massachusetts Institute of Technology, Bldg. 6-304,
Cambridge, MA 02139, USA

c Physics Department, Queen Mary College, Mile End Road, London E1 4NS, United Kingdom
Received 12 February 2001; accepted 28 March 2001
Abstract
This paper considers the spacetimes describing pp-waves propagating on extremal non-dilatonic
branes. It is shown that an observer moving along a geodesic will experience infinite curvature at
the horizon of the brane, which should therefore be regarded as singular. Taking the decoupling
limit of these brane-wave spacetimes gives a pp-wave in AdS, the simplest example being the
Kaigorodov spacetime. It has been conjectured that gravity in this spacetime is dual to a CFT in
the infinite momentum frame with constant momentum density. If correct, this implies that the CFT
must resolve the singularity of the bulk spacetime. Evidence in favour of this conjecture is presented.
The unbroken conformal symmetries determine the scalar 2-point function up to an arbitrary function
of one variable. However, an AdS/CFT calculation shows that this function is constant (to leading
order in 1/N 2 ) and the result is therefore the same as when the full conformal symmetry is unbroken.
This paper also discusses a recently discovered Virasoro symmetry of metrics describing pp-waves in
AdS and naked singularities in the RandallSundrum scenario. 2001 Published by Elsevier Science
B.V.
1. Introduction
In the standard picture [1], when one thinks of the global causal structure of nondilatonic p-branes one thinks of vacuum interpolation. Far from the brane, the fields
associated with the brane typically fall off quickly and the spacetime approaches
uncompactified, flat space. Near the brane, the geometry is usually that of a wormhole,
or EinsteinRosen bridge. As one approaches the brane, one falls down the throat of
the wormhole and approaches some vacuum which is a valid compactification of the
supergravity theory. Thus, while it is often impossible to actually locate the brane
E-mail addresses: dominic.brecher@durham.ac.uk (D. Brecher), chamblin@mit.edu (A. Chamblin),
h.s.reall@qmw.ac.uk (H.S. Reall).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 7 0 - 5
156
D. Brecher et al. / Nuclear Physics B 607 (2001) 155190
worldvolume in these effective field theory solutions which describe gravitating branes,
it is still possible to discuss the causal structure of the configurations. In particular, it is
always the case that different supersymmetric vacuum solutions of the supergravity theory
emerge in different regions of the full spacetime of a p-brane. Furthermore, when the brane
worldvolume is flat there are no curvature singularities in the near-horizon region and so
we may extend the spacetime through the horizon.
A natural question is: what happens to this causal structure when one perturbs the brane?
In this paper, we consider a large class of perturbations of the near-horizon region of
the brane spacetime, corresponding to plane fronted parallel gravitons moving tangent
to the brane directions. We find that such Weyl curvature fluctuations always lead to the
appearance of pp-curvature singularities in the near-horizon region. In other words, some
curvature components diverge in an orthonormal frame that is parallelly propagated along
a geodesic that reaches the horizon. Thus, these perturbations close off the spacetime, so
that there is no analytic extension of the manifold through the near-horizon region. Similar
results have been previously noticed for waves on strings [24]. It has also been shown that
if one attempts to extend the metric beyond the singular string horizon, then a test string
falling through the horizon would experience a divergent excitation [5]. Therefore, these
singularities should be regarded as a physical boundary to the spacetime.
The pp-wave in the spacetimes that we consider is described by a harmonic function
which can depend on both transverse and worldvolume brane coordinates. It has been
proposed that such solutions can describe fully localized intersections of a pp-wave with
a brane [6], and explicit solutions for such a localized intersection have been presented
in the near horizon limit [7]. We shall show that it is possible to derive the near horizon
solution of [7] simply by requiring that the horizon be free of pp-singularities. However, we
shall then show that the resulting solution can be related to anti-de Sitter space (AdS) by a
coordinate transformation, so this solution does not really describe a wave. The only nontrivial non-singular solutions have a harmonic function that vanishes on the brane horizon.
Such solutions describe a wave propagating in a direction parallel to the brane, but at some
distance away from the horizon, rather than a wave on the brane worldvolume. Our results,
and the results of [2,3] are evidence in favour of a generalized no-hair theorem for time
dependent perturbations of brane horizons.
Taking the decoupling limit [8] whilst keeping the momentum density of the wave
finite, one obtains solutions describing pp-waves propagating in AdS. The case in which
the harmonic function describing the wave depends only on the transverse direction was
discussed in [9], where it was shown that the Kaigorodov spacetime [10,11] is obtained
in the decoupling limit. It was also shown that if the brane is made non-extremal then
the wave can be removed by a boost. In the limit of extremality, this boost becomes
infinite. It is therefore natural to regard the extremal brane-wave solution as an infinitely
boosted version of the extremal brane solution. In the decoupling limit, this means that
the Kaigorodov spacetime can be regarded as a infinitely boosted version of AdS. This
infinite boost also induces an infinite boost on the conformal boundary of AdS. In the light
of the AdS/CFT correspondence [8], it is therefore natural to conjecture [9] that string
theory in the Kaigorodov spacetime (times a sphere of appropriate dimension) is dual
157
to some conformal field theory (CFT) in an infinitely boosted frame, i.e., the so-called
infinite momentum frame. Since the momentum density was held fixed in the decoupling
limit, there must be a non-zero background momentum density present. If this conjecture
is correct then the CFT must resolve the singularity of the bulk spacetime.
We have calculated the expectation value of the CFT energymomentum tensor
that arises from a large class of solutions describing pp-waves propagating along the
horospheres of AdS. The energymomentum tensor describes a null momentum density,
with profile fixed by the profile of the bulk wave. For the Kaigorodov spacetime, the
momentum density is indeed constant, in agreement with the conjecture of [9]. This
background momentum density breaks the conformal symmetry group of the boundary
theory down to some smaller symmetry group. We show how the isometries of the
Kaigorodov spacetime have a natural interpretation as this subgroup of the conformal
group that leaves the background momentum density invariant.
When conformal invariance is unbroken, it is well-known that the 2-point functions
of certain CFT operators are completely determined. In our case, although some of the
conformal invariance is broken, the dilatation symmetry persists (when combined with
a boost). This highly, but not entirely, constrains the form of the 2-point functions. For
a spin-0 operator, we show that the symmetry determines the 2-point function up to an
arbitrary function of one variable. It is then natural to ask whether we can calculate the 2point function using the prescription of [12,13]. This turns out to be straightforward, with
the exception that the Kaigorodov spacetime is non-static and therefore cannot be Wick
rotated to Euclidean signature. The AdS/CFT calculation must therefore be carried out on
the Lorentzian spacetime.
The 2-point function is not uniquely defined in Lorentzian signature. However, the
arbitrariness takes a form that is characteristic of the infinite momentum frame, offering
further support in favour of the conjecture of [9]. Moreover, the 2-point function turns out
to be independent of the background momentum density. We argue that this makes sense
in the large N limit.
Another topic addressed in this paper is the Virasoro symmetry group of anti-de Sitter
space. It is well-known that gravity in spacetimes asymptotic to AdS3 is equivalent to
a two-dimensional CFT [14]. It has recently been pointed out that the space of metrics
describing pp-waves propagating on the horospheres of AdS admits an action of the
Virasoro group in any dimension [15]. We explain how this symmetry arises from the
action of the Virasoro group on 3-dimensional submanifolds of these spacetimes.
2. Singular brane-waves
Our starting point is the ansatz for the metric of a non-dilatonic p-brane with a pp-wave
traveling along a worldvolume direction, which we take to be x. The line element can be
written as:

ds 2 = H (r)2/d 2dx + dx + F x + , x i , r dx +2 + dx i dx i
158

2
+ H (r)2/d dr 2 + r 2 dd+1
,
(2.1)
where d = p + 1, d = D d 2, D is the dimension of spacetime and

2 2
+ = 1,
d d
(2.2)
for the branes of interest. x i , i = 1, . . . , d 2, denote the directions tangent to the brane
and transverse to the wave, and we have written
x =
(x t)
.
2
(2.3)
d 2 is the line-element of a unit (d +1)-sphere (or, more generally, any positive Einstein
d+1
space with appropriately normalized curvature). Where necessary, we will take m, n =
1, . . . , d + 1 to denote directions in this space. The function F (x + , x i , r) depends on both
worldvolume and transverse coordinates so the wave is potentially fully localized on both
the brane and in the transverse space.
The curvature tensors for this metric are calculated in the appendix. For the M2-brane,
the four form field strength is

F[4] = d H 1 dx + dx dy,
(2.4)
where we have written x 1 = y. For the M5-brane, the dual of the four form field strength
is

F[4] = d H 1 dx + dx dx 1 dx 2 dx 3 dx 4 ,
(2.5)
and for the D3-brane, the self-dual five form field strength is
F[5] = G[5] + G[5] ,
(2.6)
where

G[5] = d H 1 dx + dx dx 1 dx 2 .
(2.7)
It is straightforward to verify that these expressions solve the Einstein equations provided
that the function H (r) is harmonic on the transverse space:
H (r) = 1 +
k
r d
(2.8)
and the function F (x + , x i , r) obeys (e.g., [6,7,16])

2F
(d + 1) F
2
+ H (r)
+
F = 0,
r 2
r
r
(2.9)
2 denotes the Laplacian with respect to the flat coordinates x i .

where
There are different types of solution to this equation, but we will not consider the
specific forms here. Suffice it to say that the usual brane-wave intersections [16] have
F = F (x + , r), the wave being localized in the transverse space only. On the other hand,
a Ricci-flat brane [1719] can be constructed by taking F = F (x + , x i ), in which case
159
the wave is now localized on the brane worldvolume, but not in the transverse space. In
the near-horizon limit, one obtains solutions describing pp-waves propagating in anti-de
Sitter space. These were discussed in four dimensions in [20]. They have also been used
to give a non-linear discussion of the RandallSundrum [21] model, with the solution F =
Hij (x + )x i x j (where Hij is symmetric and trace-free) identified as the massless graviton
[22].
There is a particular solution which will prove of interest in what follows. In the core of
the brane, as r 0, (2.9) becomes
2F
(d + 1) F
k 2
(2.10)
+
F = 0,
+
r
r
r 2
rd
which is solved by [7,25]

+ i
+
+
d 2
1
2
F x , x , r = F0 x + F1 x
(2.11)
.
|x| + k
d 2 r d2
This solution can, in fact, be extended beyond the near-horizon region. By writing F =
F0 (x + ) + F1 (x + )(|x|2 + f (r)), we find that

+ i
+
+
1
d 2 2
d 2
2
, (2.12)
F x , x , r = F0 x + F1 x
r +k
|x|
d + 2
d 2 r d2
is a solution of (2.9), valid for all r. However, the amplitude of the wave grows as one
moves away from the brane, and destroys the asymptotic flatness of the spacetime. It is
therefore unlikely that this solution is of physical relevance.
2.1. Near-horizon geometry
In the near horizon limit of the metric (2.1), the 1 in the harmonic function H can be
dropped. Defining a new coordinate by
d
z = k 1/2 r 1d/2 ,
(2.13)
d
then brings the metric to the form
ds 2 =

l2
2 dx + dx + F x + , x i , z dx +2 + dx i dx i + dz2
2
z
2 l 2 d 2 ,
+ (d/d)
d+1
(2.14)
where
1/d .
l = (d/d)k
In the new coordinate, the equation satisfied by F is

1 F
d1
2
z
F = 0.
+
z zd1 z
(2.15)
(2.16)
Throwing away the d 2 piece of the spacetime (2.14), and with Eq. (2.16) governing
d+1
the function F , we recognise the metric describing a pp-wave in AdSd+1 . The AdS horizon
is now at z = , and the conformal boundary at z = 0. The latter is, in general, no longer
Minkowski space; and we will have more to say on this below.
160
2.2. Supersymmetry considerations

It is easy enough to see that our brane-waves preserve 1/4 of the spacetime supersymmetries, as should be the case for a two-element intersection. In the AdS limit considered
above, however, one might expect a supersymmetry enhancement. It is perhaps surprising
that this does not occur. Indeed, the Kaigorodov spacetime preserves 1/4 of the supersymmetries [9] and, as we will now show, this is true of any pp-wave in AdS.
piece. The Killing spinor
Consider, then, the spacetime (2.14) without the d 2
d+1
equations for a (d + 1)-dimensional supergravity theory with cosmological constant =
(1/2l 2 )d(d 1) have the form
1
(2.17)
M = 0,
2l
where DM is the covariant derivative acting on an arbitrary spinor . It is useful to consider
the integrability condition for the existence of Killing spinors, which is
M DM = DM
1
1
HMN [DM , DN ] = RMNP Q P Q + 2 MN = 0.
(2.18)
4
2l
Substituting for the Riemann tensor, we find the non-zero (tangent space) components of
this equation to be
2

1 z2
1 F
2F
F
H+i = 2
(2.19)
ij
j
z = 0,
2l
x i x j
z z
zx i

2

F
1 z2
1 F
2F
z
H+z = 2
(2.20)
i = 0,
4l
z z
zx i
z2
where all gamma matrices are defined with respect to the tangent space metric.
The conditions (2.19), (2.20) can be satisfied in one of two ways. The first is to take
2F
1 F
1 F
2F
2F
= 2
= i j
= 0,
i
zx
z z
x x
z z
z
which is solved by

F x + , x i , z = F0 x + + F1 x + |x|2 + z2 .
(2.21)
(2.22)
This is just the solution (2.11) in the z coordinate. However, as we shall show below, it in
fact corresponds to pure AdS in an unusual coordinate system, for which the integrability
equations are satisfied identically; for bona fide waves, the function F cannot have this
form. So we are led to the second way of satisfying the integrability conditions (2.19),
(2.20), which is of course to demand that
= 0.
(2.23)
This is just the usual chirality condition for a pp-wave to be supersymmetric, and removes
1/2 of the supersymmetries. However, since this is a necessary, but not sufficient, condition
to ensure the existence of Killing spinors, we must further consider Eq. (2.17) explicitly.
161
The Killing spinor equations (2.17) reduce to the following conditions on :

1
1 F
F
F
+ = + (1 + z ) +
(2.24)
z ,
i
2z
4 x i
z
z
1
= (1 z ) ,
(2.25)
4z
1
i = i (1 + z ) ,
(2.26)
2z
1
z = z .
(2.27)
2z
Solutions of these equations must satisfy the chirality condition (2.23). This removes all
trace of the function F , leaving the pure AdS equations, for which the Killing spinors were
constructed in horospherical coordinates in [26]. In this case, there are two distinct types
of solution: one a function of all the coordinates, and one a function of z alone. Only the
latter type of solution satisfies Eq. (2.23), however. In other words, the sole solution to the
Killing spinor equations is

l
=
(2.28)
0,
z
where 0 is a constant spinor satisfying
(1 + z )
= 0,
= 0,
(2.29)
(2.30)
The pp-wave in AdS thus preserves only 1/4 of the supersymmetries. It is important to
note that this is true regardless of which coordinates F is a function of. In particular, it is
true of the Kaigorodov spacetime, to be considered below, as shown in [9]. At any rate, the
pp-wave in AdS5 should break the supersymmetry of the dual CFT from N = 4 to N = 1.
This can occur in two different ways: either the pp-wave changes the background metric of
the boundary theory, which explicitly breaks some of the supersymmetry, or the pp-wave
preserves the flat metric of the boundary theory but corresponds to a quantum state of the
boundary theory in which some of the supersymmetry is spontaneously broken. In the case
of the Kaigorodov metric, the latter possibility occurs, with the symmetry breaking arising
from a non-zero expectation value for the energymomentum tensor.
2.3. Global structure
Since the pp-wave generates Weyl curvature alone, the only potentially divergent
curvature invariant of our solutions is the square of the Riemann tensor; and it is easy
to see from the curvature components given in the appendix that this is everywhere finite.
However, it was shown in [20] that pp-waves in AdS4 generically give rise to pp-curvature
singularities at the AdS horizon, i.e., in an orthonormal frame parallelly propagated along a
causal geodesic, some curvature components will diverge at the AdS horizon. It is therefore
important that we examine the behaviour of the tidal forces experienced by observers who
move along geodesics in brane-wave spacetimes.
162
Consider, then, the geodesic equations for the spacetime (2.1). For simplicity, we take
d 2 to be the line-element on the unit (d + 1)-dimensional sphere, set all the polar
d+1
angles to /2, and take to be the azimuthal angle. We further take t M to denote the unit
tangent to a timelike geodesic, with affine parameter . Then the Killing vectors k = x
and m =
generate the conserved quantities E = k t, and L = m t, respectively. These
give the geodesic equations
dx +
= EH 2/d ,
d
d
L
=
.
d r 2 H 2/d
The x i equation is

1 dx i
E 2 2/d F
d
=
H
,
d H 2/d d
2
x i
and that governing the radial motion is

2
d
1 H 2
1
F
2/d dr
2/d dr
H
=
+H
+ E 2 H 2/d
d
d
2H d
d
2
r

2

2 H
1 L
.
+
+
2 r 2 H 2/d r
H
(2.31)
(2.32)
(2.33)
(2.34)
For the special case in which F = F (x + , x i ) and L = 0, this equation can be solved to give
1
dr
= (d2)/2d cH 2/d 1,
(2.35)
d H
for some constant c. In general, however, we are unable to solve (2.34) due to the rdependence of the function F . Using the timelike condition, the final equation for x
is

i 2
2

1
1
dx
dx
L2
2/d dr
2 2/d
1 + 2/d
=
. (2.36)
+H
+E H F +
d
2E
H
d
d
r 2 H 2/d
As we have already mentioned, we cannot solve this set of equations in general. But this
does not prevent us from analyzing the near-horizon geometry of the brane. To this end,
we set L = 0 and write the tangent vector to the geodesic as

t M = EH 2/d , x , x , r , 0, . . . , 0 ,
(2.37)
where x + = EH 2/d , a dot denotes differentiation with respect to , x = {x i } and we have
written the components in the order (t + , t , t i , t r , t 1 , . . . , t d , t ). We want to analyze the
behaviour of the components of the Riemann tensor in an orthonormal frame parallelly
propagated along a geodesic with tangent t M . We must first construct the other vectors of
this orthonormal frame. There are d 2 unit normals of the form

x i
1/d
=
H
,
n
,
0,
.
.
.
,
0
,
nM
(2.38)
0,
i
i
EH 2/d

j
163
where n i = i . The normals satisfy t ni = 0, ni nj = ij and it is easy to check that they

are parallelly propagated with respect to t:
t nM
i = 0.
(2.39)
Although they will prove to be irrelevant to our discussion, the d + 1 angular normals, nm ,
are easy to determine. They have a single non-zero component
nnm =
1
H 1/d r
1
n .
sin 1 sin d m
(2.40)
The remaining two normals of our parallelly propagated frame are somewhat harder to
determine. Denoting them by p and q, we have

1
M
2/d
p = sin () EH , x + , x , r , 0, . . . , 0
E

r 1/d
1/d
+ cos () 0, H , 0, H
(2.41)
, 0, . . . , 0 ,
E

1
q M = cos () EH 2/d , x + , x , r , 0, . . . , 0
E

r 1/d
1/d
sin () 0, H , 0, H
(2.42)
, 0, . . . , 0 ,
E
where () obeys
d
H 1/d
(2.43)
=
H
,
d
dH
which gives = constant far from the brane (r ), and = / l + constant in the
near horizon region (r 0). These normals are slight generalizations of those given by
Podolsk [20] in a study of pp-waves in AdS4 .
2.4. Waves in AdSd+1
Before analyzing the global structure of our brane-waves, it will be instructive to
consider the simpler case of waves in AdSd+1 S d+1 , the near-horizon geometry of the
metric (2.1). Note that it is not sufficient to just consider waves in AdS when studying ppsingularities at the horizon of the brane-wave spacetime. This is because in taking the near
horizon limit, one has to take the limit of the components of the curvature in a parallelly
propagated frame, e.g., of objects like t M pN t P pQ RMNP Q . However, in the AdS limit,
one takes the limits of t M , pN and RMNP Q separately and then multiplies them. As we
shall see, the product of these limits is not necessarily equal to the limit of the product.
Therefore it is necessary to examine the brane-wave spacetime and the pp-wave in AdS
separately.
The presence of the sphere is irrelevant, so we will essentially be considering waves in
AdSd+1 , and the results are again a slight generalization of those in [20]. To look at the
AdS case, as opposed to our brane-wave spacetime (2.1), we can simply set H (r) = k/r d in
the expressions for the curvature components, tangent vector and normals. In this case the
164
equation of motion (2.10) valid in the core of the brane, is exact. Analysis of the curvature
components in our parallelly propagated frame is then fairly straightforward.
There are a number of potential pp-singularities, but they are all encoded in the
quantities 1

E 2 k 14/d
1) F ,
+
(
d
(2.44)
F
2 r d
r

E 2 k 3/24/d 2 F
,
Ai =
(2.45)
2 r d
rx i

E 2 k 24/d 2 F
d r d F
,
Mij =
(2.46)
+
ij
2 r d
x i x j
d k r
which are potentially divergent near the AdS horizon (r 0). All components of the
curvature tensor in our parallelly propagated frame depend on these three quantities alone.
For example,
A+ =
1
(2.47)
A+ cos2 (/ l),
l2
R(t )(p)(t )(i) RMNP Q t M pN t P nQ
(2.48)
i = Ai cos(/ l),
ij
P Q
R(t )(i)(t )(j ) RMNP Q t M nN
(2.49)
i t nj = 2 Mij .
l
For the waves in AdS under consideration here, these expressions are exact.
The question is, how do the specific choices of the function F affect the existence,
or otherwise, of pp-singularities? It is easy to see that the partially localized solutions
with F = F (x + , r) or F = F (x + , x i ) are indeed singular 2 as r 0. However, one might
suspect that there exist fully localized solutions with F = F (x + , x i , r) that are entirely
non-singular. We will see that this is not true.
Absence of pp-singularities requires that all curvature components remain finite as
r 0. From Eq. (2.48) and the expression for Ai , this implies that r F = g0 (r, x + ) +
O(r 3d/24 ) as r 0, where we are using the notation f = O(g) to mean that there exists
some function h(x + , x i ) such that f/g h as r 0. Integrating gives F = g1 (r, x + ) +
g2 (x i , x + ) + O(r 3d/23 ), for some g1 and g2 . Substituting this into the equation of motion
for F gives

1
g1
d+1
2
g2 = O r 5d/25 ,
(2.50)
r
+ k
r r
r
R(t )(p)(t )(p) RMNP Q t M pN t P pQ =
1 The reader may be worried that these expressions appear to depend on d,

which is related to the dimension
of the sphere. However, Eq. (2.2) can be used to eliminate d.

2 The special case F (x + , x i ) = a (x + )x i +F (x + ) will be discussed below. Note that the case F = F (x + , x i )
i
0
includes the solution F = Hij (x + )x i x j , describing the RandallSundrum [21] zero mode, which was previously
thought to be non-singular [22]. In this case, we have Mij Hij (x + )/r 2d4 . Of course, in AdS3 this term
vanishes since, in this case, Hij H11 = 0 by virtue of the tracelessness of Hij . However, we do not really have
a pp-wave in this case, since there can be no propagating gravitational degrees of freedom in three dimensions.
In higher dimensions, the zero mode does, in fact, generate singularities at the AdS horizon.
165
from which it follows that

2
2
g2 = (d 2)F1 x +
k
for some function F1 . Integrating with respect to r gives
(2.51)

F1 (x + ) F2 (x + )

F x + , x i , r = F0 x + +
+
+ g2 x + , x i + O r 3d/23 . (2.52)
d2
d
r
r
Substituting this into Eq. (2.47) then yields F2 0 in order for the curvature to remain
finite as r 0. Finally, substituting into Eq. (2.49) and requiring finiteness as r 0 gives
d 2)
d(
2 g2
F1 (x+ )ij ,
=
(2.53)
x i x j
kd
which is consistent with Eq. (2.51). Solving this equation then gives the final result for F :

d 2
1
F x + , x i , r = F0 x + + F1 x + |x|2 + k
d 2 r d2
+ i
+ i
+ ai x x + x , x , r ,
(2.54)
where we have rescaled F1 . denotes a possible correction term of order r 3d/23 . Note
that F0 can be gauged away by redefining x . The ai (x + )x i term can be also be gauged
away, as we shall explain below. So, for the time being, we shall set F0 = ai = 0. We
should emphasize that Eq. (2.54) is the only solution free of pp-singularities. If is
identically zero then this is just the near-horizon solution (2.11) discussed in [7,25] (with
an additional term linear in x i ), i.e., it is an exact solution of the equation of motion (2.10)
for pp-waves in AdS. It follows that the correction must also be an exact solution of
Eq. (2.10), but this correction is negligible near the horizon. If one had just this correction
term, then one would also have a solution with no pp-singularities at the horizon. This is
easily understood: a solution of O(r 3d/23 ) vanishes at the horizon sufficiently fast that the
metric is just like pure AdS as far as geodesics and curvature tensors are concerned 3 . In
other words, such a solution would represent some perturbation of pure AdS that does not
extend as far as the AdS horizon. It seems unlikely that such a solution would correspond to
the near-horizon geometry of an intersection of a brane and a pp-wave. The only candidate
for such a solution is therefore the solution (2.11). However, it is easy to use the curvature
tensors in the appendix to see that this solution is Weyl flat, which implies that it is locally
isometric to pure AdS and therefore does not describe a pp-wave in AdS. We shall present
the coordinate transformation that gauges away this solution later. Given that this solution
is pure gauge, it seems rather doubtful that it describes the near-horizon geometry of a fully
localized intersection of a brane and pp-wave, as suggested in [7].
Now, as promised, let us re-instate the term ai (x + )x i . The coordinate transformations
that removes the first and second terms in (2.54) do not affect the form of the third term
3 This is not strictly true in general since second derivatives of would naively be expected to behave as
r 3d/25 , which vanishes at r = 0 only if d > 3, i.e., d < 6. However, it is probably possible to derive stronger
bounds on .
166
(ai is just multiplied by a function of x + ), and the change in F is the only change in
the metric. So once we have removed the first two terms in F , we can do the following
coordinate transformation:

x x + b x + + ci x + x i ,
(2.55)
xi xi + di x+ .
It is easy to see that b(x + ), ci (x + ) and di (x + ) can be chosen to cancel the ai (x + )x i term
in F .
To summarize, we have shown that there are no non-singular pp-waves in AdSd+1 with
metric of the form (2.14), except possibly solutions that vanish sufficiently fast near the
AdS horizon.
2.5. Brane-waves
We will now generalize the above analysis to the more complicated case of our branewaves, described by the metric (2.1). Concentrating on the same curvature components as
in (2.47), (2.48) and (2.49) above, in the near-horizon limit, r 0, we find

r d r 2
1
d
2
R(t )(p)(t )(p) 2 A+ cos (/ l) + (1 + d)

cos2 (/ l),
l
d
k r
R(t )(p)(t )(i) Ai cos(/ l),
d 2
ij
d
r r .
R(t )(i)(t )(j ) 2 Mij + ij (1 + d)
l
d
k r
(2.56)
(2.57)
(2.58)
Note that there are now extra terms involving r . It possible that these terms could cancel the
divergences coming from the other terms. This emphasizes the importance of examining
the full brane-wave spacetime, and not just the near-horizon geometry. We have retained
only the leading order divergences, and assumed that r diverges as r 0. (If r remains
finite then the extra contributions above are clearly finite and cancellations will not occur.)
Using Eq. (2.57), we can obtain Eq. (2.52) as above. However, there will now be an
extra O(r 2 ) correction term on the right-hand side, arising from neglecting the 1 in H in
deriving Eq. (2.10). This is subleading order for the M5-brane but leading order for the
other branes. 4
It is now possible that the term in F of order r d may be consistent with finite curvature
provided that the r terms cancel the divergences arising from this term. Eq. (2.56) shows
that r would have to be of order r 23d/2 as r 0 in order to cancel the divergence in

A+ . To see whether this is possible, we need to examine the geodesic equation (2.34). The
dominant term on the right-hand side would be the second term (involving r 2 ). Neglecting
the other terms, and integrating then gives r r 2d/2 (this could also be obtained from
Eq. (2.35)), which is a contradiction. It is therefore not possible for the r terms to cancel
the divergence arising from the r d term in F , so F2 has to be set to zero in order to obtain
finite curvature, just as above.
4 Once again, it may be possible to derive better bounds on the correction term.
167
Now consider Eq. (2.58). An identical argument shows that the r term cannot cancel
the divergence coming from Mij , so one obtains Eq. (2.53) as above. Putting everything
together, one obtains the same solution (2.54) but with the exception that the correction
term is O(r 2 ) for the M2 and D3 branes. (Note that such a term occurs in Eq. (2.12).)
Thus, we conclude that the only non-singular brane-waves must have F of the form
(2.11) near the brane. Since this form for F can be gauged away, we conclude that there
are no non-singular pp-waves on the non-dilatonic branes, of the supergravity theories that
we have been considering. This is reminiscent of the results concerning waves on strings in
five and six dimensions [24], in which similar singularities in the curvature components
at the horizon of the string were found.
3. Virasoro symmetry
Before moving on to the cases of interest to us, we will review some relevant points
concerning AdS3 and its two-dimensional CFT dual.
3.1. AdS3 , Virasoro symmetry and (2 + 1)-dimensional pp-waves
Consider AdS3 in horospherical coordinates 5
ds 2 =

l2 +
dx dx + dz2 .
2
z
(3.1)
Since AdS3 is the group manifold SL(2, R), the metric (3.1) has the isometry group
SL(2, R)L SL(2, R)R = SO(2, 2). The SL(2, R)L action is given by:
x + f (x + ),
1 f
x x z2 ,
2 f

zz f ,
(3.2)
where
ax + + b
SL(2, R),
f x+ = +
cx + d
(3.3)
and a prime denotes a derivative with respect to x + . Note that elements of SL(2, R) have a
vanishing Schwarzian derivative:

f 3 f 2

+
= 0.
f, x =
(3.4)
f
2 f
The SL(2, R)R action has the same form as (3.2), but with x + and x interchanged:
x f (x ),
5 We have rescaled x by a factor of
2.
168
1 f
x + x + z2 ,
2 f

zz f ,
(3.5)
where now
ax + b
SL(2, R),
f x =
(3.6)
cx + d
and a prime denotes a derivative with respect to x . This isometry group acts on the
boundary as the global conformal group of two-dimensional Minkowski space, generated
by two copies of the Virasoro generators L0 and L1 .
Now the AdS/CFT correspondence asserts that the AdS isometry group is identical to
the conformal group on the boundary. In two dimensions, the latter is (two copies of)
the infinite-dimensional Virasoro group, so there is a puzzle here: how can the finitedimensional isometry group be equivalent to the infinite-dimensional conformal group
on the boundary? The puzzle was resolved long ago by Brown and Henneaux [14]. The
isometries (3.2) and (3.5) leave the metric unchanged, and the corresponding generators
will annihilate the vacuum state in the quantum theory. These isometries must, however,
be supplemented by the infinite number of diffeomorphisms which change the metric, but
leave the asymptotic form of AdS3 invariant up to a conformal factor. An analysis of the
(modes of the) generators of these allowed diffeomorphisms shows that they obey (two
copies of) the Virasoro algebra. The asymptotic symmetry group of AdS3 is thus just the
conformal group in 1 + 1 dimensions, as expected.
The induced metric on the boundary is
l2
(3.7)
dx + dx ,
z2
the conformal transformations of which are x + f (x + ) and x f (x ). In the bulk,
then, the asymptotic structure-preserving diffeomorphisms are precisely those of (3.2)
and (3.5) but for arbitrary functions f (x + ) and f (x ). By arbitrary here, we mean
{f, x + }, {f, x } = 0. Under the former transformation, we find
ds 2 =
ds 2

l2 +
dx dx + F x + z2 dx +2 + dz2 ,
2
z
(3.8)
where

1
F x + = f, x + .
(3.9)
2
As promised, these general transformations are not isometries of the metric, but do leave
the boundary manifold unchanged up to a conformal factor (equal to f ).
There are two points to note about the metric (3.8). Firstly, it looks like a pp-wave
in AdS3 . However, there are no propagating degrees of freedom in (2 + 1)-dimensional
gravity, so what we mean by this is the following. The metric of a wave in AdS3 would be
of the form

l2
ds 2 = 2 dx + dx + F x + , z dx +2 + dz2 ,
(3.10)
z
169
which is a solution of Einsteins equations with negative cosmological constant if and only
if

1
z z F = 0,
(3.11)
z
implying
F (x + , z) = F0 (x + )z2 + F1 (x + ),
(3.12)
for arbitrary functions F0 and F1 . Now, F1 can be gauged away by a redefinition of x ,

leaving a metric of the form (3.8), which we know is obtained from pure AdS3 by a
coordinate transformation of the form (3.2) with arbitrary f (x + ). Therefore there are no
pp-waves in AdS3 , as expected.
The metric (3.8) is locally isometric to AdS3 . This is reminiscent of the BTZ black
hole [23,24]. In fact, when F (x + ) = constant, it is straightforward to obtain the extreme
BTZ black hole from the metric (3.8) [25]. To see this, we compactify along the direction
of propagation of the wave, taking the period to be 2 . Calling this coordinate , taking
F (x + ) = 8GMl 2 , and rewriting in terms of r = 1/z, the metric (3.8) becomes
2 2
2
2
ds 2
dr 2
2
2
2
dt
d
=
4GMl
+
r
+
4GMl
4GMl
dt
d
+
.
l2
r2
(3.13)
Rescaling t l 2 t and l gives the metric of the extreme Ml = J BTZ black hole [23,
24]:

2
ds 2 = N 2 dt 2 + 2 N dt + d +
r2
dr 2 ,
N 22
(3.14)
where
2 = r 2 + 4GMl 2 ,
N 2 = r 4 /(l)2 ,
N = 4GMl/ 2 .
So there is a nice correspondence: by periodically identifying one of the coordinates, AdS3
in Poincar coordinates becomes the M = J = 0 BTZ black hole [23,24]; and a wave in
AdS3 becomes the extreme Ml = J BTZ hole [9,25].
3.2. Virasoro symmetry in higher dimensions
The isometry group of AdS3 combines with a subgroup of general diffeomorphisms to
give the Virasoro group. In higher dimensions this is not possible. However, it has recently
been shown [15] that the space of metrics of the form

l2 +
2
dx dx + F x + , x i , z dx + + dx i dx i + dz2
2
z
does possess an infinite-dimensional symmetry group, namely

x+ f x+ ,
ds 2 =
(3.15)
170
f
1
x x |x|2 + z2 ,
2
f
i
i

x x f ,
z z f ,
(3.16)
with the function F shifting according to

1
F x + , x i , z F x + , x i , z f |x|2 + z2 f, x + ,
(3.17)
2
with the Schwarzian derivative defined in (3.4). If one starts from pure AdSd+1 (F 0)
then the change of coordinates (3.16) takes one to a pp-wave metric with

F x + , x i , z = F x + |x|2 + z2 ,
(3.18)
for some arbitrary F (x + ). Writing this in terms of the coordinate r and using the relation
(2.2) brings F to the form (2.11) without the F0 and ai x i pieces (which can be gauged away
anyway). Therefore the solution (2.11) just describes pure AdS in unusual coordinates,
which justifies our claim that this solution is pure gauge.
It was proved in [15] that the transformations (3.16) do indeed generate the Virasoro
group. It is natural to ask why this symmetry exists. To answer this question, we shall need
to discuss two different sets of coordinate on AdS. First, recall that AdS is defined as a
hyperboloid embedded in flat space with two time directions:
ds 2 = du dv + dX+ dX + dXi dXi ,
+
uv + X X + X X = l .
i
X /u
(3.19)
(3.20)
=
and
=
If one now defines
metric for AdS in horospherical coordinates:
xi
Xi /u,
and eliminates v, then one obtains the

du2
(3.21)
+ u2 dx + dx + dx i dx i .
u2
A second set of coordinate on AdS can be constructed based on the observation that
the surfaces Xi = constant have AdS3 geometry. Define new coordinates by X1 =
l sinh y cos 1 , X2 = l sinh y cos 1 sin 2 , etc., u = l cosh y u,
v = l cosh y v,
X =
l cosh y X , which gives

ds 2 = l 2 dy 2 + cosh2 y d u d v + d X + d X + sinh2 y d 2 ,
(3.22)
+
u v + X X = 1,
(3.23)
ds 2 = l 2
which makes it clear that surfaces of constant y and are AdS3 hyperboloids. Now
introduce horospherical coordinate on each of these hyperboloids: x = X /u and
eliminate v to get
2

d u
2
+
2
2
ds 2 = l 2 dy 2 + cosh2 y
(3.24)
+
u
dx
dx
y
d
+
sinh
.
u 2
It is straightforward to work out how the coordinate systems used in (3.21) and (3.24) are
related. The coordinates x are the same for the two metrics. The other coordinates are
related by u = l cosh y u and x 1 u = l sinh y cos 1 , etc.
171
Consider now a pp-wave spacetime, with metric

du2
2
+ u2 dx + dx + F x + , x i , u dx + + dx i dx i
u2

2
= ds 2 (AdS) + u2 F x + , x i , u dx + .
ds 2 = l 2
(3.25)
This is written in the horospherical coordinates. Now transform to the other coordinate
system:
ds 2 = ds 2 (AdS) + l 2 cosh2 y u 2 F dx +

2

+

d u
2
2
2
2
+2
2
2
+ sinh y d . (3.26)
= l dy + cosh y
+ u dx dx + F dx
u 2
2
It now looks like the surfaces of constant y and have the geometry of pp-waves in AdS3 ,
which were discussed above. However, we have to be slightly careful because F depends
on the coordinates y and , so these pp-waves will be off-shell in three dimensions (i.e.,
F will not satisfy Eq. (3.11)). Now consider the Virasoro symmetry (3.16) acting on the
metric (3.25) (with u = l/z). Writing this in terms of the new coordinates gives
x + f (x + ),
x x
1 f
,
2u 2 f
u
u
,
f
(3.27)
with y and unchanged and

1
(3.28)
f, x + .
2u 2
This is clearly nothing but the Virasoro symmetry acting on the three-dimensional surfaces
of constant y and . Thus the existence of the higher-dimensional Virasoro symmetry
arises from the existence of the Virasoro symmetry for three-dimensional asymptotically
AdS submanifolds. 6
Note that the higher-dimensional pp-wave (3.15) can only be asymptotic to AdS when
F 0 as z 0. More general pp-waves will not have Minkowski space as their conformal
boundary at infinity, and therefore be of less interest from the point of view of AdS/CFT.
A Virasoro transformation will introduce a term |x|2{f, x + } into F . This destroys the
preferred coordinate system on the boundary at infinity (although the boundary is, of
course, still flat space since this is only a coordinate transformation), and therefore does
not correspond to a conformal transformation on the boundary. The only transformations
that preserve the preferred flat coordinates on the boundary are those with {f, x + } = 0,
which give an SL(2, R) subgroup of the conformal group on the boundary.
F Ff
6 Actually these submanifold are only asymptotically AdS when F 0 as u , which implies F 0 as
u , which implies that the higher-dimensional spacetime is also asymptotically AdS. If this requirement is
not satisfied then the action of the Virasoro symmetry on the three-dimensional submanifolds is a generalization
of the usual action to certain spacetimes that are not asymptotic to AdS3 .
172
4. Boundary energymomentum tensor

We have shown that a pp-wave propagating on a non-dilatonic brane has pp-curvature
singularities at the horizon. The near-horizon geometry of such spacetimes is a pp-wave
in AdS space and we will now consider the AdS/CFT correspondence in such cases. If
this makes sense then the CFT provides a resolution of the singularity. The easiest thing to
compute is the CFT energymomentum tensor, making use of the boundary counterterm
method of [13,2732]. Our starting point is the metric (3.15), describing a pp-wave in
AdSd+1 . The boundary is at z = 0, and the induced metric on surfaces of constant z is
ds 2 |z=const. = gij (x) dx i dx j =

l2 +
2
dx dx + F x + , x i , z dx + + dx i dx i ,
2
z
(4.1)
where i, j = 0, . . . , d 1. The outward unit normal to such a surface is n = (l/z) dz.

Using the formulae from [32], the energymomentum tensor is given by

1
d 1
l
1
Tij =
gij +
(Kij Kgij )
Rij Rgij + ,
8Gd+1
l
d 2
2
(4.2)
where Rij and R are the Ricci tensor and Ricci scalar associated with gij and Kij denotes
the tangential components of the extrinsic curvature of the boundary, defined by
K = h h n ,
(4.3)
where
n n . For d 6 one needs further counterterms.
This calculation will yield the CFT energymomentum tensor in the conformal frame
where the metric is gij , which includes the prefactor l 2 /z2 . To get rid of this factor, one
must perform a Weyl transformation to obtain the metric g ij = (z2 / l 2 )gij . In the new
conformal frame, the energymomentum tensor is Tij = (l/z)d2 Tij [33], and the metric
is

2
ds 2 = gij dx i dx j = dx + dx + F x + , x i , 0 dx + + dx i dx i ,
(4.4)
which is manifestly flat if F (x + , x i , 0) = 0, which is what we shall usually assume. The
counterterm calculation is straightforward. The only non-vanishing component of the CFT
energymomentum tensor turns out to be
T++ =
l d1
16Gd+1
lim
z0 zd1

2 2
z
z3
F
2
F
F
+
, (4.5)
z
d 2
(d 2)2 (d 4)
where we have included the first three boundary counterterms. However, not all three are
present for low-dimensional examples. For d = 2, only the first term above should be kept;
for d = 3, 4, the first two terms are needed, one keeps the first two terms; and for d =
5, 6, one needs the third term too. Higher order terms would occur in higher-dimensional
examples.
173
Finally, we must write everything in terms of the parameters of the dual CFT. For d = 2,
the CFT has central charge c = 3l/(2G3 ) [14], so we obtain
1 F
c
lim
,
(4.6)
24 z0 z z
where we have dropped the bar on T++ since we shall always be working in the same
conformal frame. In the examples obtained from non-dilatonic branes, namely d = 3, 4, 6,
the following formulae apply [8]
l2
2 2N 3/2
,
(4.7)
=
G4
3
l3
2N 2
(4.8)
,
=
G5
l5
16N 3
(4.9)
=
,
G7
3 2
where N is the number of branes present before the decoupling limit is taken. Substituting
into T++ then gives
3/2

2N
1 F
2
lim
z F ,
T++ =
(4.10)
24 z0 z2 z

z 2
1 F
N2
T++ =
(4.11)
lim
F
,
8 2 z0 z3 z
2

z 2
1 F
N3
z 3 2 2
T++ =
(4.12)
lim
F
F
3 3 z0 z5 z
4
32
for d = 3, 4, 6, respectively.
The energymomentum tensor is obviously dependent on the form of the function F .
In particular, for the spurious wave with F as in (3.18), we have T++ = 0, which is
in agreement with the fact that our spacetime in this case is just pure AdS in unusual
coordinates. Secondly, a Ricci-flat brane, with F = F (x + , x i ), also has T++ = 0 since
2 F = 0.
the Einstein equations in this case give
In general, the function F will have z-dependence that is a linear combination of zd
and z0 as z 0. This can be seen by Fourier analyzing equation (2.9) with respect to x i to
obtain an equation in z that can be solved with Bessel functions [22]. Solutions proportional
to zd give finite T++ whereas those proportional to z0 give a divergent result in general.
This is just the distinction between normalizable and non-normalizable bulk modes [34],
with the former corresponding to a deformation of the state of the CFT and the latter to
a change in the parameters of the CFT Lagrangian, in this case the boundary metric. We
shall be interested only in the bulk solutions proportional to zd at the boundary, for which
the boundary is flat space and the CFT is in a quantum state with non-vanishing T++ ,
i.e., a null momentum density. In fact, we shall restrict ourselves to solutions of the form

F x + , x i , z = F x + zd .
(4.13)
T++ =
The boundary energymomentum tensor then describes a disturbance propagating at the

speed of light in the negative x-direction with wave profile determined by F (x + ).
174
The case F (x + ) = constant was studied in [9]. Starting from a non-extremal stack of
branes, it was shown that a finite boost produces a metric describing a pp-wave propagating
on the branes. However, in order to retain the pp-wave in the extremal limit, it is necessary
to take the boost parameter to infinity. Therefore, the metric (2.1) with F (x + , x i , z) =
Q/r d (Q = constant) can be regarded as describing an infinitely boosted stack of BPS

branes. Taking the decoupling limit keeping the momentum finite, one obtains a sphere
times an asymptotically AdS pp-wave spacetime with

F x + , x i , z = d z d ,
(4.14)
where is a constant. In four dimensions, this pp-wave spacetime was first discussed by
Kaigorodov [10,11]. Its D-dimensional generalization, denoted KD , was presented in [9].
The above discussion shows that KD can be regarded as an infinitely boosted version of
AdSD . The infinite boost in the bulk induces an infinite boost on the conformal boundary,
and it is therefore natural to conjecture that supergravity theory in the Kaigorodov
spacetime (times a sphere of appropriate dimension) is dual to a strongly coupled large
N CFT in an infinitely boosted frame with constant background momentum density [9].
The Kaigorodov spacetime is a special case of the class of pp-wave spacetimes that we
have been considering, and we have seen that all of these give a CFT energymomentum
tensor that describes a null momentum density in the x-direction. In the case of Kd+1 , the
momentum density is in fact constant: P = T++ is given by
c2
,
(4.15)
12
3/2 3
2N
,
P=
(4.16)
8
N 2 4
P=
(4.17)
,
2 2
2N 3 6
P=
(4.18)
3
for d = 2, 3, 4, 6, respectively. Note that a null momentum density is precisely what one
would expect from an infinite boost. Our results are therefore evidence in favour of the
conjecture of [9]. The classical supergravity approximation is valid in the large N limit, in
which case P .
The three-dimensional Kaigorodov spacetime, K3 , is just that described by the metric
(3.10), with F (x + , z) = 2 z2 , and it is curious that the energymomentum tensor in this
case is non-vanishing, since we know from the above discussion that this spacetime is
just pure AdS3 in unusual coordinates. The result can be understood as follows [35].
Start with pure AdS3 in horospherical coordinates, given by (3.1). This corresponds to
the SL(2, R)L SL(2, R)R vacuum of the CFT, and the corresponding energymomentum
tensor will therefore vanish. As we have seen, K3 can be obtained from this metric via the
coordinate transformation (3.2), with
P=

1
F x + = f, x + = 2 ,
2
(4.19)
a solution of which is

1 2x +
f x+ =
e
.
2
175
(4.20)

The new boundary coordinates are x + = f (x + ) and x = x . The boundary metric is

+
ds 2 = dx + dx = e2x dx + dx .
(4.21)
The AdS/CFT calculation in K3 gives the energymomentum tensor in the conformal

frame where the metric is

ds 2 = dx + dx ,
(4.22)
which can be obtained from the metric (4.21) by a Weyl transformation. Under the Weyl
transformation, the energymomentum tensor transforms as [36]: 7
c

f, x + ,
T+ + = T+ +
(4.23)
24
Since T+ + vanishes, this gives
c2
,
(4.24)
12
which is precisely what we obtained from our AdS/CFT calculation. Note that the
Kaigorodov coordinates cover only the region x + > 0 of the original spacetime. This
is very similar to what occurs for the BTZ black hole, and corresponds to adopting the
coordinates of an accelerating observer on the boundary. Such an observer would naturally
define the energymomentum tensor so that it vanished in the vacuum defined with respect
to his time coordinate. This energymomentum tensor has a finite expectation value in the
SL(2, R) SL(2, R) vacuum state and describes the thermal radiation experienced by such
an observer [35].
Having understood the AdS3 case, we now move on to discuss the higher-dimensional
generalizations of the Kaigorodov spacetime.
T+ + =
5. The Kaigorodov spacetime

5.1. The metric and its symmetries
The (d + 1)-dimensional Kaigorodov metric Kd+1 can be written

l2 +
d d
+2
i
i
2
dx
,
dx
+
z
dx
+
dx
dx
+
dz
z2
where i = 1, . . . , d 2. It will sometimes be useful to write
ds 2 =
(5.1)
x = x t.
(5.2)
7 We have inserted a factor of 1/(2 ) relative to the string theory convention [36] in order to be consistent
with the usual Lorentzian field theory definition of the energymomentum tensor.
176
The Kaigorodov spacetime is clearly asymptotic to AdSd+1 as z 0, and has a ppcurvature singularity as z . It was shown in [9] that this metric preserves 1/4
supersymmetry, and has the following Killing vector fields (we have converted the results
of [9] to our coordinate system)

2 +,
K = 2 ,
x
x
i
j
Lij = x
x
,
x j
x i
x+
Li =
2x i ,
i
x
2 x
K+ =
J =
Ki =
,
x i
(d + 2)x
(d 2)x +
xi
z
+
l z
2l
x
2l
x
l x i
(5.3)
(5.4)
(5.5)
(5.6)
The commutation relations of these Killing vector fields were presented in [9].
5.2. Matching of symmetries
The above Killing vector fields induce symmetries of the boundary theory. In this
subsection, we shall show that these symmetries are the symmetries that one would expect
for a CFT in a background with the conformal symmetry broken by a constant (null)
momentum density in the x + direction. As we saw above, such a momentum density is
described by an energymomentum tensor, the only non-vanishing component of which is
P T++ = constant. The symmetries of the boundary theory are simply those conformal
transformations that leave this energymomentum tensor invariant.
Some of the symmetries are obvious: the K generators just give boundary translations,
and the Lij correspond to transverse rotations. These transformations clearly leave the
boundary metric invariant, and preserve the form of the background energymomentum
tensor. The other generators are more complicated. Let y = x 1 and consider Ly . The finite
form of the bulk isometry generated by this Killing vector field is

2
2

t + x + y,
t = 1+
4
4
2

2
2
x t y,
x = 1
4
4
2
y = y + t + x.
(5.7)
2
2
This is a combination of a boost in the y-direction, a rotation in the xy plane and a boost
in the x-direction. Clearly the parameters of the three transformations are related. To see
how this emerges, consider the transformations separately. A boost with velocity v in the
y-direction followed by a rotation through angle in the xy plane gives
1
1

x + = (cos + + v sin )x + + (cos v sin )x (sin + v)y,
2
2
(5.8)
177
where = (1 v 2 )1/2 . The energymomentum tensor after this transformation becomes

T =
x + x +
T++ ,
x x
(5.9)
so if the direction of momentum flow is to remain unchanged then we require x + to be

independent of x and y . Inverting Eq. (5.8) (by replacing v v and ), this
requirement reduces to
v = sin ,
cos > 0.
(5.10)
The new energymomentum tensor then has only one non-vanishing component, viz.
T+ + = cos2 T++ .
(5.11)
Finally, the cos2 prefactor can be removed by boosting along the x-direction with velocity
w, which gives

1 w +
+
x ,
x =
(5.12)
1+w
so
T+ + =
1+w
cos2 T++ ,
1w
(5.13)
hence if w = sin2 /(1 + cos2 ) then the energymomentum tensor in the new frame is
the same as in the original frame. Therefore, the combined transformation is a symmetry
of the background of the boundary theory. In simple terms, it can be understood as
follows. 8 In coordinates t, x, x i , the momentum density is E(1, 1, 0, . . . , 0). (This is null
because we are working in the infinite momentum frame.) The boost in the y-direction
takes this to E (1, cos , sin , . . . , 0) for some energy E . The rotation in the xy plane
brings this to E (1, 1, 0, . . . , 0). Finally, the boost in the x-direction brings this back to to
E(1, 1, 0, . . ., 0).
The symmetries that we have discussed so far form the subgroup of the Poincar
group that is unbroken by the background momentum density. The remaining generator
J involves an unbroken conformal symmetry. The finite bulk isometry induced by J is

x + = d/2 x + ,
x = d/2 x ,
, x i = x i ,
z = z,
(5.14)
which is clearly a combination of a dilatation by factor , and the transformation

x d/2 x .
(5.15)
This latter transformation is simply a boost in the x-direction. To see why this combination
of dilatation and boost is a symmetry of the boundary theory, consider replacing the boost
by a general boost in the x-direction with velocity v. For the boundary theory the combined
8 Although this is slightly misleading, since we are dealing with a momentum density, which cannot be written
as a four vector.
178
dilatation and boost is

1v +

x ,
x =
1v
x i = x i .
(5.16)
The transformed energymomentum tensor has one non-vanishing component:

T+ + = 2
1+v
T++ .
1v
(5.17)
Clearly it is not invariant. However, neither is the flat boundary metric:

ds 2 = 2 dx + dx + dx i dx i ,
(5.18)
so to bring the metric back to the original form we must do a Weyl transformation gij
2 gij . The Weyl transformation of the energymomentum tensor is
T+ + = (d2) T+ + = d
1+v
T++ .
1v
(5.19)
The energymomentum tensor is therefore invariant under the combined dilatation and
boost provided we take (1 + v)/(1 v) = d , which gives agreement with Eq. (5.14).
Note that it is not find any combination of transformations that involves special
conformal transformations and leaves the energymomentum tensor invariant.
5.3. Scalar 2-point function
It is well-known that conformal symmetry determines the 2-point functions of operators
up to an overall constant: 9

O(x)O(x ) =
c
,
(x x )2
(5.20)
where is the scaling dimension of O. It is not surprising that AdS/CFT reproduces such
2-point functions since they are entirely determined by symmetry, and AdS does indeed
give the correct conformal symmetry group.
We are interested in the case when some of the symmetry of the boundary theory has
been broken by a background momentum density. The two point function in this case is
less constrained. To see this, write
+ 2
+ +
, x12 , x12 ,
O x1 , x1 , x1 O x2 , x2 , x2 = f x12
(5.21)
where we have used the notation x for x i , and x12 denotes x1 x2 . In writing the two
point function in this form, we have made use of the unbroken translation and transverse
rotation symmetries. Now consider the unbroken symmetries
generated by Ly . Write x =
(y, z, . . .), and perform the transformation (5.7) with = 2y/x + . Invariance of f gives

f x + , x , y 2 + z2 + = f x + , x + y 2 /x + , z2 + ,
(5.22)
9 Since we are dealing with Lorentzian theories, an i term is required, but suppressed throughout this paper.
and repeating the same procedure with Lz , etc., eventually gives

f x + , x , x2 = f x + , x + x2 /x + , 1 .
179
(5.23)
So the unbroken Poincar symmetry has reduced the 2-point function to a function of 2
variables, which can be taken to be x + and x 2 = x + x + x2 . Let us denote this function by
f (x + , x 2 ). Now consider the final unbroken symmetry, generated by J . If we assume that
O has scaling dimension then under the dilatation plus boost symmetry, O transforms
as
O(x ) = O(x).
For the 2-point function, this implies that

f (d2)/2x + , 2 x 2 = 2 f x + , x 2 ,
(5.24)
(5.25)
so choosing 2 = 1/|x 2 | gives

f x +, x 2 =

1
f x + |x 2 |(d2)/4, sign x 2 .
2
|x |
(5.26)
Hence the unbroken symmetries of the boundary theory restrict the 2-point function to take
the form
+ 2 (d2)/4
+ +
1
,
O x1 , x1 , x1 O x2 , x2 , x2 = 2 f x12
(5.27)
|x12|
|x12|
where = sign(x 2 ). The argument of f is invariant under all the unbroken symmetries.
In contrast with the conformally invariant case, the functional dependence of the 2-point
function is not completely determined by symmetry. The AdS/CFT calculation of the 2point function therefore has dynamical content for the Kaigorodov spacetime, in contrast
with the analogous calculation for pure AdS.
5.4. AdS/CFT calculation of scalar 2-point function
Having established that the scalar 2-point function is non-trivial, we shall now calculate
it using the AdS/CFT correspondence. The method is similar to that of [37] with the
exception that we can no longer Wick rotate to Euclidean signature since our the
Kaigorodov spacetime is not static. Lorentzian calculations for pure AdS were discussed
in [34,38]. We shall be careful to examine the behaviour of surface terms that arise at the
singularity.
Consider a scalar field of mass 10 m2 :

1
1
S0 = d d+1 x g ()2 m2 2 .
(5.28)
2
2
The equation of motion is
2

m2 = 0.
10 We shall use units for which l = 1.
(5.29)
180
To compute the boundary 2-point function, we have to evaluate the on-shell action of the
scalar field solution (z, x) obeying
( , x) =
0 (x),
(5.30)
where we have introduced a UV cut-off at z = . is the scaling dimension of the operator

O, related to the mass of by [12,13]
d 1
2
d + 4m2 .
= +
(5.31)
2 2
The action can be integrated by parts and the equation of motion used, giving

1
S0 =
(5.32)
d d x h n ,
2
where the sum runs over the two boundaries (z = and z ). On each boundary, we
have outward unit normal n = (1/z) dz and the determinant of the boundary metric is
(1/z)d .
We shall solve for by Fourier analyzing the x coordinates:
(z, x) = k (z)eikx ,
where k x = 12 (k + x + k x + ) + k i x i . This gives

m2 + 2 d d
1 d k
d1 d
2
z
z k = 0.
k + 2 k
dz zd1 dz
z
(5.33)
(5.34)
Note that when k + = 0, this equation is exactly the same as one would obtain in pure AdS,
with the solution [37,38]
zd/2K ( k 2 z)
d
0 (k),
k (z) =
(5.35)
d/2 K ( k 2 )
where 0 (k) denotes the Fourier transform of 0 (x). This solution is uniquely determined
since the second linearly independent solution gives rise to an infinite surface term at z =
[37,38].
Now consider the case k + = 0. For small z (i.e., near infinity), neglecting the zd term
gives solutions in terms of Bessel functions. For k 2 > 0, one obtains

d/2 I ( k 2 z) + A(k)I ( k 2 z)
z
0 (k),
k (z) = d d/2
(5.36)
I ( k 2 ) + A(k)I ( k 2 )
where A(k) is some undetermined function and
1
2
=
d + 4m2 ,
2
(5.37)
which we assume is real, corresponding to the BreitenlohnerFreedman bound [39] m2 >

d 2 /4. For k 2 < 0, one obtains

d/2 J ( k 2 z) + A(k)J ( k 2 z)
z
d
k (z) =
(5.38)
0 (k).
d/2
J ( k 2 ) + A(k)J ( k 2 )
181
For large z (i.e., near the singularity), one can neglect the k 2 and m2 /z2 terms and obtain
the solutions 11
+ d/2

2|k |
z(d+2)/2 ,
k = zd/2 J
(5.39)
d +2
where
d
.
=
(5.40)
d +2
Near the singularity, the solutions behave as

(d+2)/2
(d2)/4 sin
.
k z
(5.41)
z
cos
This implies that the surface term near the singularity is bounded but oscillatory as z .
This is exactly what happens as z for pure AdS when k 2 < 0 [38], and can be dealt
with by introducing an infra-red cut-off. The two solutions are equally acceptable near
the singularity. In contrast with the behaviour in pure AdS, it not possible to uniquely
determine the solutions with k 2 > 0 by the requirement of finiteness of the surface term at
z = .
The interpretation of the arbitrary function A(k) was explained in [38]. This function
corresponds to the freedom to choose different Lorentzian propagators in the dual CFT. For
example, one usually works with the Lorentzian Feynman propagator obtained by analytic
continuation of a unique bounded Euclidean propagator. However, there are different
propagators (for example, the retarded propagator) which one might instead choose to work
with. The choice of propagator is determined by boundary conditions, and the freedom to
choose boundary conditions is encoded in the function A(k). There may also be freedom
in the choice of vacuum state for the CFT.
In all cases, we have written the behaviour for small z as
k (z) =
K(k, , z) 0 (k),
(5.42)
where
K(k, , z) =
k (z) + A(k) k+ (z)

,
( ) + A(k) + ( )
k
(5.43)
and + and are the normalizable and non-normalizable [34,38] solutions, with the
behaviour
+ z ,
zd ,
(5.44)
as z 0. Note that A(k) is not arbitrary when k +

gives [38]
S0 =
1
2
d d k d d k d (k + k )
1d
= 0. Evaluating the surface term at z =
0 (k)z K(k, , z) 0 (k),
(5.45)
z=
11 In fact, the equation can also be solved exactly if one retains the m2 /z2 term and neglects only the k 2 term,
giving the same solution except that = d 2 + 4m2 /(d + 2).
182
from which one can read off the 2-point function [38]

O(k)O(k ) = d (k + k ) lim lim z K(k, , z).
(5.46)
0 z
At first sight, it appears that this 2-point function will be entirely determined by the part
of K coming from the non-normalizable solutions, since these terms dominate as 0.
However, in pure AdS, this part of K turns out to involve only positive integer powers of
k 2 . After Fourier transforming back to position space, this gives only contact terms (terms
that vanish at non-zero separation, e.g., derivative of delta functions). The non-trivial part
of the 2-point function arises from mixing between the normalizable and non-normalizable
solutions, which gives non-integer powers of k 2 [37,38].
For the Kaigorodov spacetime, we have to be a bit more careful since we know that
will be corrected from their pure AdS forms by subleading terms that depend on . These
subleading terms could dominate the terms arising from mixing between the normalizable
and non-normalizable solutions, and therefore change the form of the 2-point function.
We can examine this dependence by seeking series solutions for near z = 0. One
obtains: 12

k = zd a0 + a2 z2 + ,
(5.47)

+
2
k = z b0 + b2 z + ,
(5.48)
with the series agreeing with the approximate Bessel function solutions up to dth order, but
then being altered by the dependence. The deviation from the Bessel function solutions
at (d + 2)-th order is
d k +
a0 ,
2( d 1)(d + 2)
2
ad+2 =
d k +
b0 .
2( + 1)(d + 2)
2
bd+2 =
(5.49)
Since the terms a0 , . . . , ad , b0 , . . . , bd and the Bessel function parts of ad+2 and bd+2 only
involve positive integer powers of k 2 , it follows that ad+2 and bd+2 only involve positive
integer powers of k 2 and k + , which means that they will still only give contact terms in
the 2-point function. Thus the non-trivial contribution to the 2-point function will come
from the leading order terms in , which are independent of . It follows that the 2-point
function will not depend on .
The only novel feature that arises from this calculation is that the 2-point function is
uniquely determined in momentum space when k + = 0, but not when k + = 0. This is to
be contrasted with the calculation for pure AdS, which gives a unique propagator when
k 2 > 0 but not when k 2 < 0, which is what one expects in field theory. 13 The occurrence
12 If is an integer then there will be logarithmic terms in these series. We shall concentrate on the case of
non-integral for simplicity.
13 We remind the reader why this is so. In free field theory, the momentum space propagator is obtained from
the equation (k 2 + m2 )G(k; m2 ) = 1. The solution is uniquely defined except when k 2 = m2 , corresponding
to the freedom to add to G(k; m2 ) a function of the form g(k)(k 2 + m2 ). In the interacting case, one can write
the 2-point function in a KllenLehmann representation as an integral over free field propagators weighted by
some spectral function (m2 ). This spectral function vanishes for m2 < 0, guaranteeing that the arbitrariness in
the free field propagator does not affect the interacting propagator when k 2 > 0.
183
of k + in our results is characteristic of the infinite momentum frame. The freedom in the
propagator when k + = 0 and k 2 > 0 is presumably related to some subtlety of quantization
in the infinite momentum frame.
Since there is no dependence on , we are free to choose the function A(k) so that the
2-point function that we obtain from the Kaigorodov spacetime is the same as the 2-point
function that we obtain from pure AdS by Wick rotation from Euclidean signature. The
only thing we have to worry about it whether this gives the correct result when k + = 0,
which it must do since the bulk solution is the same for Kaigorodov and for pure AdS in
this case. We therefore obtain the Feynman propagator:

c
,
O(x)O(x ) =
(5.50)
((x x )2 i )
where c is given in [37].
5.5. Attempted explanation
The simplicity of the AdS/CFT result suggests that there should be some way of
understanding it in purely field-theoretic terms. The best we have come up with is the
following.
Writing the function f of (5.27) in terms of a dimensionless variable, one obtains

(d+2)/4
f P 1/2 x + x 2
(5.51)
, N, ,
where we have now included the dependence on N and explicitly. Our AdS/CFT
calculation is valid for N, , P with d P /N d/2 fixed. We would like to argue
that f must approach a constant in this limit, namely f (, , ). This would explain
our AdS/CFT result. However, this argument would break down if, at large N , f reduced
to a function of
(d2)/4
(d2)/4
d/2 x + x 2
.
N d/4 P 1/2 x + x 2
(5.52)
In order to improve on this argument, one would need to understand the explicit N dependence of the 2-point function, at least for large N . If P were kept finite in the large
N limit then perhaps it would be possible to extend the usual planar-graph arguments (see,
e.g., [40]) of large N field theory to fix the N -dependence of f . However, when P scales
with N , one is dealing with field theory in an N -dependent background, which is probably
much harder to understand.
6. Discussion
We have discussed spacetimes describing pp-waves propagating on the worldvolume of
extremal non-dilatonic branes. We have explicitly shown that any inertial observer in the
bulk of such a spacetime will encounter infinite tidal forces at the place where the horizon
would ordinarily be located. This means that these solutions cannot be extended through
the horizon region, so the spacetime terminates at a null singularity. This is evidence in
184
support of the idea [2,3] that there may be a generalized no-hair theorem that deals with
time-dependent hair.
In the near-horizon limit, these spacetimes reduce to spacetimes describing pp-waves
moving along the horospheres of AdS. We have discussed an apparent Virasoro symmetry
[15] of such spacetimes, and shown that it arises from an action of the Virasoro group on
three-dimensional asymptotically AdS submanifolds. It would be interesting to see whether
this symmetry could be used to extend the entropy calculation of [41] to extended objects
in higher-dimensional AdS, for example the BTZ black string of [42].
We have shown that pp-waves moving tangential to the horospheres of AdS will make
the AdS horizon become singular unless the wave does not extend as far as this horizon.
This result has implications for the RandallSundrum [21] scenario in which our universe
is regarded as a thin domain wall in AdS. This thin domain wall should presumably
be regarded as a mathematical idealization of a thick domain wall in AdS. In [22], it
was shown that the pp-waves propagating tangentially to such a domain wall correspond
precisely to the gravitons of the RandallSundrum scenario. We have shown that any
of these pp-waves will induce a curvature singularity at the AdS horizon. However,
this horizon is the surface at which one must impose boundary conditions for the bulk
spacetime. Thus one is faced with the unappealing prospect of having to impose boundary
conditions at a naked singularity, which presumably requires an understanding of quantum
gravity. If one were to impose the boundary condition that the past horizon remain regular
then it appears that one must exclude the possibility of there being gravitons present on the
domain wall!
We have presented evidence in favour of the conjecture [9] that gravity in the Kaigorodov
spacetime is dual to a CFT in the infinite momentum frame with constant background
momentum density. In particular, we have calculated the CFT energy momentum tensor
that arises from a large class of asymptotically AdS pp-waves, including Kd+1 . We have
discussed how the isometries of Kd+1 have a natural interpretation as the subgroup of the
conformal group that leaves the background momentum density invariant. This symmetry
group determines the 2-point function of a scalar operator only up to an arbitrary function
of one variable. Our AdS/CFT calculation of the 2-point function therefore has dynamical
content, and is found to agree with the result one obtains from pure AdS (i.e., when the
conformal symmetry group is unbroken). We have suggested that this is a large N effect.
The 2-point function exhibits some ambiguity associated with working in Lorentzian
signature, however the ambiguity does not take the expected form. This is perhaps the
only signal in our work that the boundary CFT has really been quantized in the infinite
momentum frame.
It might be possible to understand our results more clearly by doing a free field theory
calculation in the CFT. The quantum state of the CFT could be constructed by taking a
thermal density matrix (corresponding to a non-extremal stack of branes), applying a finite
boost, and then taking the limit in which the temperature tends to zero and the boost to
infinity with the energy density fixed.
It would be interesting to have a complete understanding of how string theory resolves
the Kaigorodov singularity. Certainly, if we use the criterion of Gubser [43] then this
185
singularity is good, in the sense that the slightest finite temperature deformation will
yield a non-extremal brane-wave, which we know is simply a boosted non-extremal brane
and therefore has a regular horizon [9]. On the other hand, it would be nice to know if the
singularity in the bulk could be explicitly resolved along the lines of recent work [44,45]
which deals primarily with timelike singularities.
We should also mention that our discussion of the Kaigorodov metric has been confined
to the case d > 0. However, the metric with d < 0 is still a solution of Einsteins
equations. This metric gives a CFT energymomentum tensor with negative energy, which
is unacceptable. So it appears that AdS/CFT is telling us that the d < 0 Kaigorodov
singularity cannot be resolved. Further support for this idea comes from the fact that the
metric in this case does not have a regular finite temperature deformation.
As we have seen, it is easy to write down pp-wave solutions which are asymptotic to
AdS and which induce non-constant null momentum density in the boundary theory. For
these spacetimes, we would expect a non-trivial 2-point function in the CFT. It would
be interesting to investigate such spacetimes in more detail. A particular case of interest
p
would be to choose F (x + , z) = d x + zd , with p an integer (to avoid problems when x + <
0). Such a bulk spacetime would admit a scaling symmetry analagous to the Kaigorodov
scaling symmetry.
Of course, not all gravitons in AdS will move tangential to the horospheres (and
therefore the boundary) of AdS. As discussed in [46,47], information about a graviton
emitted from deep in the bulk of AdS which heads towards the boundary of AdS must be
encoded nonlocally in Wilson loops of the boundary CFT, with the size of the Wilson
loop related to the distance of the disturbance from the boundary. These degrees of
freedom in the CFT which encode information about the graviton in the bulk were dubbed
precursors. An interesting (and open) problem is to find a full non-linear solution
describing a graviton moving in the direction transverse to the horospheres, so that we
might have a hope of explicitly computing properties of these precursors.
Acknowledgements
DB would like to thank Simon Ross and Paul Saffin for discussions, and Simon
Ross for comments on a draft of this paper. DB is supported in part by the EPSRC
grant GR/N34840/01. AC thanks Mirjam Cvetic, Dan Freedman and Andreas Karch for
useful conversations. AC is supported in part by funds provided by the U.S. Department
of Energy (D.O.E.) under cooperative research agreement DE-FC02-94ER40818. HSR
thanks Jerome Gauntlett and Prem Kumar for discussions and is funded by PPARC.
Appendix A
Our line element is

ds 2 = H (r)2/d 2dx + dx + F x + , x i , r dx +2 + dx i dx i
186

2
+ H (r)2/d dr 2 + r 2 dd+1
.
We shall compute the curvature tensors in the following basis:

1
e = H 1/d dx + F dx + ,
e+ = H 1/d dx + ,
2
er = H 1/d dr,
em = H 1/d r e m ,
(A.1)
ei = H 1/d dx i ,
(A.2)
em
where
are an orthonormal basis of 1-forms with respect to the metric gmn associated
with the line element d 2 . The metric can be written
d+1
gMN = AB e
Me N,
(A.3)
where
+ = + = rr = 1,
ij = ij ,
mn = mn
(A.4)
is the Minkowski metric. Note that we are using letters M, N, . . . as curved (coordinate)
indices and A, B, . . . as basis indices.
The connection 1-forms are defined by
deA = A B eB .
(A.5)
The non-vanishing connection 1-forms are

F
1
+i = i = H 1/d i e+ ,
2
x

1
1
H
F +
e ,
+r = r = H 1/d e + H 1/d
d
H
2
r

1
H
r = + r = H 1/d e+ ,
d
H

1
H
ir = i r = H 1/d ei ,
d
H

1 m
H
rm = r m = H 1/d
+
e ,
r
dH
mn = mn .
(A.6)
(A.7)
(A.8)
(A.9)
(A.10)
(A.11)
where mn are the connection 1-forms associated with e m . Other non-vanishing components are obtained from the ones given above by antisymmetry: AB = BA .
The curvature 2-form is defined by
AB = dAB + AC C B .
The non-vanishing curvature 2-forms are
2
1
H
+ = 2 H 2/d
e+ e ,
d
H
(A.12)
(A.13)
+i =
2F
1 2/d H F +
1
H
e ei H 2/d i j e+ ej
2d
H r
2
x x

2
1 1/d1/d F +
1 2/dt H 2
r
e e 2H
H
e ei ,
2
x i r
d
H
2
1
F
+r = H 1/d1/d i e+ ei
2
x r

2F
1
2 1 H F +
+
e er
+ H 2/d 2 +
2
d d H r
r

2
1 1
H
1
H
+ +1
e er ,
+ H 2/d
d
H
d d
H

1 F +
1
H
e em
+
+m = H 2/d
2
r r
dH

1
1 2/d H H
e em ,
+
+ H
d
H dH
r
2
1
H
i = 2 H 2/d
e+ ei ,
H
d

2
1 1
H
H
1
+ +1
e+ er ,
r = H 2/d
d
H
d d
H

H
1 +
1
H
+
m = H 2/d
e em ,
d
H dH
r
2
1
H
ij = 2 H 2/d
ei ej ,
d
H

2
1 1
1 2/d H
H
+ +1
ei er ,
ir = H
d
H
d d
H

H
1
1 i
H
im = H 2/d
+
e em ,
d
H dH
r
2

H
1 2/d H
H r
rm = H
+
e em ,
H
H
Hr
d

1 2 m
2/d H

mn = mn H
+
e en ,
r
dH
187
(A.14)
(A.15)
(A.16)
(A.17)
(A.18)
(A.19)
(A.20)
(A.21)
(A.22)
(A.23)
(A.24)
with the other non-vanishing components related to these by antisymmetry. The Riemann
curvature tensor is defined by
1
AB = RABCD eC eD .
2
The non-zero components are
2
1
H
R++ = 2 H 2/d
,
d
H
(A.25)
(A.26)
188
2F
1
1 2/d H
H
F ij ,
R+i+j = H 2/d i j +
2
x x
2d
H
2
1
F
R+i+r = H 1/d1/d
,
2
rx i
2
1
H
R+ij = 2 H 2/d
ij ,
H
d

1 2/d
2 1 H F
2F
R+r+r = H
+
,
2 +
2
d d H r
r

2
1 1
H
1
H
+ +1
,
R+rr = H 2/d
d
H
d d
H

1
1
H
+
R+m+n = H 2/d
F mn ,
2
r
dH

H
1
1
H
R+mn = H 2/d
+
mn ,
d
H dH
r

1 2/d H 2
Rij kl = 2 H
(ik j l il j k ),
d
H
2

H
1 1
1
H
+ +1
Rirj r = H 2/d
ij ,
d
H
d d
H

H
1
1
H
Rimj n = H 2/d
+
ij mn ,
d
H dH
r
2

H
1 2/d H
H
Rrmrn = H
+
mn ,
H
H
Hr
d

1 2
H
+
Rmnpq = R mnpq H 2/d
(mp nq mq np ).
r
dH
(A.27)
(A.28)
(A.29)
(A.30)
(A.31)
(A.32)
(A.33)
(A.34)
(A.35)
(A.36)
(A.37)
(A.38)
Other non-vanishing components are related to these by the symmetries of the Riemann
tensor. The Ricci tensor is
RAB = RA C C = RAB+ + RA+B + RAiBi + RArBr + RAmBm .
The non-vanishing components are

1
1 2/d 2 F
d + 1 F
2
H 2/d
R++ = H
+
F,
2
2
r
r
2
r
2 is the Laplacian with respect to the flat coordinates x i ,
where
2

H
1 2/d H
d + 1 H
R+ = H
+
,
d
H
H
r H
2

H
1 2/d H
d + 1 H
+
Rij = H
ij ,
d
H
H
r H

d H 2 d + 1 H
1
H
+
Rrr = H 2/d
+
,
H
d H
r H
d
(A.39)
(A.40)
(A.41)
(A.42)
(A.43)

2
1 2/d H
d + 1 H
H
1

Rmn = Rmn H
+
+ 2 mn
H
H
r H
r
d
2

H
1
d +1H
H
= H 2/d
+
mn ,
H
H
r H
d
189
(A.44)
where in the final expression we have used the fact that the internal space is an Einstein
space, which in our basis implies
mn = d H 2/d mn .
R
r2
(A.45)
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
G.W. Gibbons, P.K. Townsend, Phys. Rev. Lett. 71 (1993) 3754, hep-th/9302049.
N. Kaloper, R.C. Myers, H. Roussel, Phys. Rev. D 55 (1997) 7625, hep-th/9612248.
G.T. Horowitz, H. Yang, Phys. Rev. D 55 (1997) 7618, hep-th/9701077.
R.C. Myers, Gen. Rel. Grav. 29 (1997) 1217, gr-qc/9705065.
S. Ross, JHEP 9808 (1998) 003, hep-th/9710158.
H. Yang, Localized intersecting brane solutions of D = 11 supergravity, 1999, hep-th/9902128.
D. Youm, Localized intersecting BPS branes, 1999, hep-th/9902208.
J. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231, hep-th/9711200.
M. Cvetic, H. L, C.N. Pope, Nucl. Phys. B 545 (1999) 309, hep-th/9810123.
V.R. Kaigorodov, Dokl. Akad. Nauk SSSR 146 (1962) 793.
V.R. Kaigorodov, Sov. Phys. Dokl. 7 (1963) 893.
S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Phys. Lett. B 428 (1998) 105, hep-th/9802109.
E. Witten, Adv. Theor. Math. Phys. 2 (1998) 253, hep-th/9802150.
J.D. Brown, M. Henneaux, Commun. Math. Phys. 104 (1986) 207.
M. Baados, A. Chamblin, G.W. Gibbons, Phys. Rev. D 61 (2000) 081901, hep-th/9911101.
R.G. Russo, A.A. Tseytlin, Nucl. Phys. B 490 (1997) 121, hep-th/9611047.
D. Brecher, M.J. Perry, Nucl. Phys. B 566 (2000) 151, hep-th/9908018.
B. Janssen, JHEP 0001 (2000) 044, hep-th/9910077.
J.M. Figueroa-OFarrill, Phys. Lett. B 471 (1999) 128, hep-th/9910086.
J. Podolsk, Class. Quant. Grav. 15 (1998) 719, gr-qc/9801052.
L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 4690, hep-th/9906064.
A. Chamblin, G.W. Gibbons, Phys. Rev. Lett. 84 (1999) 1090, hep-th/9909130.
M. Baados, C. Teitelboim, J. Zanelli, Phys. Rev. Lett. 69 (1992) 1849, hep-th/9204099.
M. Baados, M. Henneaux, C. Teitelboim, J. Zanelli, Phys. Rev. D 48 (1993) 1506, gr-qc/
9302012.
K. Behrndt, D. Lst, JHEP 9907 (1999) 019, hep-th/9905180.
H. L, C.N. Pope, P.K. Townsend, Phys. Lett. B 391 (1997) 39, hep-th/9607164.
M. Henningson, K. Skenderis, JHEP 9807 (1998) 023, hep-th/9806087.
A. Tseytlin, H. Liu, Nucl. Phys. B 533 (1998) 88, hep-th/9804083.
M. Henningson, K. Skenderis, Fortsch. Phys. 48 (2000) 125, hep-th/9812032.
V. Balasubramanian, P. Kraus, Commun. Math. Phys. 208 (1999) 413, hep-th/9902121.
R. Emparan, C.V. Johnson, R.C. Myers, Phys. Rev. D 60 (1999) 104001, hep-th/9903238.
P. Kraus, F. Larsen, R. Siebelink, Nucl. Phys. B 563 (1999) 259, hep-th/9906127.
R.M. Wald, General Relativity, Univ. of Chicago Press, 1984.
V. Balasubramanian, P. Kraus, A. Lawrence, Phys. Rev. D 59 (1999) 046003, hep-th/9805171.
190
[35] J. Maldacena, A. Strominger, JHEP 9812 (1998) 005, hep-th/9804085.

[36] J. Polchinski, String Theory, Cambridge Univ. Press, 1998.
[37] D.Z. Freedman, S.D. Mathur, A. Matusis, L. Rastelli, Nucl. Phys. B 546 (1999) 96, hepth/9804058.
[38] V. Balasubramanian, P. Kraus, A. Lawrence, S.P. Trivedi, Phys. Rev. D 59 (1999) 104021, hepth/9808017.
[39] P. Breitenlohner, D.Z. Freedman, Ann. Phys. 144 (1982) 249.
[40] A.V. Manohar, Large N QCD, Les Houches lectures, 1997, hep-ph/9802419.
[41] A. Strominger, JHEP 9802 (1998) 009, hep-th/9712251.
[42] R. Emparan, G.T. Horowitz, R.C. Myers, JHEP 0001 (2000) 021, hep-th/9912135.
[43] S.S. Gubser, Curvature singularities: the good, the bad, and the naked, 2000, hep-th/0002160.
[44] C.V. Johnson, A.W. Peet, J. Polchinski, Phys. Rev. D 61 (2000) 086001, hep-th/9911161.
[45] J. Polchinski, M. Strassler, The string dual of a confining four-dimensional gauge theory, 2000,
hep-th/0003136.
[46] J. Polchinski, L. Susskind, N. Toumbas, Phys. Rev. D 60 (1999) 084006, hep-th/9903228.
[47] L. Susskind, N. Toumbas, Phys. Rev. D 61 (2000) 044001, hep-th/9909013.

Partial non-renormalisation of the stress-tensor

four-point function in N = 4 SYM and AdS/CFT
B. Eden a , A.C. Petkou b , C. Schubert a,c,1 , E. Sokatchev a
a Laboratoire dAnnecy-le-Vieux de Physique Thorique 2 LAPTH, Chemin de Bellevue, B.P. 110,
F-74941 Annecy-le-Vieux, France

b Department of Physics, Theoretical Physics, University of Kaiserslautern, Postfach 3049,
67663 Kaiserslautern, Germany

c California Institute for Physics and Astrophysics, 366 Cambridge Avenue, Palo Alto, CA 94306, USA
Received 22 December 2000; accepted 28 March 2001
Abstract
We show that, although the correlator of four stress-tensor multiplets in N = 4 SYM is known
to have radiative corrections, certain linear combinations of its components are protected from
perturbative renormalisation and remain at their free-field values. This result is valid for weak as
well as for strong coupling and for any gauge group. Our argument uses Intriligators insertion
formula, and includes a proof that the possible contact term contributions cannot change the form
of the amplitudes.
Combining this new non-renormalisation theorem with Maldacenas conjecture allows us to make
a prediction for the structure of the corresponding correlator in AdS supergravity. This is verified
by first considerably simplifying the strong coupling expression obtained by recent supergravity
calculations, and then showing that it does indeed exhibit the expected structure. 2001 Elsevier
Science B.V. All rights reserved.
PACS: 03.70.+k
1. Introduction
Much effort has been devoted over the years to the study of the dynamical aspects of
quantum field theory. Weak coupling expansions have been pushed to high orders providing
useful insight, however, the strong coupling regime of most theories has remained elusive.
Nevertheless, important information can be and has been obtained from the study of
E-mail address: sokatche@lapp.in2p3.fr (E. Sokatchev).
1 Address after 1. 10. 2000: Instituto de Fisica y Matematicas, Universidad Michoacana de San Nicolas de
Hidalgo, Morelia, Mexico.

2 UMR 5108 associe lUniversit de Savoie.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 5 1 - 1
192
B. Eden et al. / Nuclear Physics B 607 (2001) 191212
various quantities whose weak and strong coupling behaviour is accessible. Such quantities
are axial [1] and conformal anomalies [2,3], and also coupling constants that are not
renormalised as one goes from weak to strong coupling [46]. The emergence of the
latter quantities has been particularly noted in the context of the AdS/CFT correspondence
conjectured by Maldacena [79]. One of the most exciting features of this conjecture is
that strong coupling information for certain quantum field theories can be obtained from
tree-level supergravity calculations. Therefore, with the explicit strong coupling results
for various correlation functions in hand, one can identify non-renormalised quantities
which remain the same both in the weak and the strong coupling regimes.
So far the strongest evidence in favour of this conjecture has been gathered in the context
of N = 4 SYM theory with gauge group SU(Nc ) [4,5,1014]. This is not surprising since
the study of this theory has a long history and its dynamics is relatively well understood.
Moreover, the success of this programme has prompted more thorough investigations
of N = 4 SYM within field theory, and in particular a search for non-renormalisation
theorems for the correlators involving only short operators, i.e. a certain class of gauge
invariant composite operators which do not depend on all the Grassmann variables in
superspace. The most typical example, first introduced in [15], is the series of operators
obtained by tensoring the N = 4 SYM field strength considered as a Grassmann analytic
harmonic superfield. They were identified with short multiplets of SU(2, 2/4) and their
correspondence with the KK spectrum of IIB supergravity was established in [16]. Short
multiplets are important in the AdS/CFT correspondence because they have protected
conformal dimensions and therefore allow a reliable comparison between quantities
computed in the bulk versus quantities derived in the CFT [17].
Recently it has been found that, for correlators of short multiplets, the absence of
radiative corrections is a rather common phenomenon; by now non-renormalisation
theorems have been established not only for two- and three-point functions [5,13,1823],
but also for so-called extremal [2426] and next-to-extremal [2628] correlators;
these are n-point functions obeying certain conditions on the conformal dimensions of
the operators involved.
At the same time, some correlators of short operators are known to acquire quantum
corrections beyond tree level. The simplest example is the four-point function of N = 4
SYM supercurrents, as has been shown by explicit computations at two loops in [2931],
and at three loops in [32,33]. 3 Nevertheless, even in this case one can still formulate a
partial non-renormalisation theorem which is the main subject of this paper.
The work presented here is part of an ongoing investigation of this four-point
function, which grew out of a line of work initiated by P.S. Howe and P.C. West. Their
original aim was to study the implications of superconformal covariance for correlators
satisfying Grassmann and harmonic analyticity constraints. In Refs. [15,34] a systematic
investigation of the superconformal Ward identities and their consequences for Greens
3 In our terminology the free-field and one-loop contributions to this correlator are synonymous, the order
O(g 2 ) contribution corresponds to two loops, etc.
193
functions of N = 2 and N = 4 short operators was initiated. 4 The N = 2 operators

are gauge-invariant products of the hypermultiplet and the N = 4 operators are gaugeinvariant products of the N = 4 field strength. A number of interesting results were derived
[15,34], in particular: The N = 4 SYM field strength is a covariantly analytic scalar
superfield from which the aforementioned set of analytic gauge-invariant operators can be
built; the two- and three-point Greens functions of these operators were determined up to
constants [15,18] (see also [4,5,19,20]); the set of all non-nilpotent analytic superconformal
invariants was given [34]. Finally, in [40] it was shown how to derive certain differential
constraints on the correlator of four stress-tensor multiplets using only the superconformal
algebra and some general properties of the N = 2 [41] or N = 4 [42] harmonic superspace
formulations of N = 4 SYM. (These constraints are reproduced and their general solution
is obtained in the Appendix A to the present paper.)
However, we will see that considerably stronger restrictions can be derived for this
amplitude if not only pure symmetry-based arguments are employed, but also some
direct input from field theory making use of the explicit form of the N = 4 SYM
Lagrangian written down in terms of N = 2 superfields. The essential tool in our
analysis is Intriligators insertion (or reduction) formula, which relates the above fourpoint function to a five-point function involving the original operators and the gaugeinvariant N = 2 SYM Lagrangian. This procedure, introduced in the present context
in [13], has turned out both more efficient and more powerful than the direct approach
to four-point functions: The lowest term of the five-point function obtained in this way
in the context of N = 2 harmonic superspace is a nilpotent superconformal covariant
of mixed chiral-analytic type. Such objects are more strongly constrained than the
original non-nilpotent amplitudes. The insertion approach has been constructively used
in the harmonic superspace derivation or rederivation of some of the non-renormalisation
theorems mentioned before [21,26,28].
In the past the applicability of Intriligators insertion formula has been questioned on
the grounds that possible contact term contributions might spoil any prediction based on it
[6,43]. While contact term contributions to correlators can usually be consistently ignored,
it had been pointed out in [6] that this is not obvious if the insertion procedure is used:
It involves an integration which may promote a contact term to a regular term. Here we
resolve this issue by finding necessary conditions for the existence of a five-point contact
term with the required properties and by giving its most general allowed form. It then
becomes clear that the regular term which it produces upon integration is compatible with
our non-renormalisation statement.
The insertion formula has also proved very useful in explicit quantum calculations at
two and three loops because it allows to apply superconformal covariance arguments to
significantly reduce the complexity of Feynman diagram calculations of the correlator
discussed here in N = 2 harmonic superspace [32,43]. The results obtained show a
remarkable pattern: certain linear combinations of the amplitudes are protected from
perturbative renormalisation, and thus remain at their free-field values. As we show in
4 Similar works analysing constrained N = 1 superfields are [3539].
194
the present paper, this property can be generalised to every loop order and thus to a nonperturbative result, irrespectively of the choice of the gauge group. Combining this new
non-renormalisation theorem with Maldacenas conjecture we can make a prediction for
the structure of the strong coupling limit obtained from AdS supergravity.
AdS supergravity calculations have been initiated in [4,10,12,14,24,44,45] and most of
the relevant methodology was developed in these articles. Yet, there the focus was on
of the N = 4
correlators whose CFT counterparts involve the top components F 2 , F F
stress-tensor multiplet, since these are more readily accessible from the supergravity side.
On the contrary, on the CFT side it is easiest to investigate the lowest components (the
physical scalars) of the same multiplet. So, until recently the results on both sides were
difficult to compare. With the completion of the computation of the quartic terms in the
supergravity effective action, a strong coupling limit for the lowest component of the SYM
stress-energy tensor four-point function became available [46]. However, it is presented in a
form which still contains parameter integrals. To verify our non-renormalisation prediction,
we first bring this result into a completely explicit form, involving only logarithms and
dilogarithms of the conformal cross ratios. The non-renormalised part is then indeed
found to agree with the free-field SYM amplitude.
The organisation of the paper is as follows: In Section 2, exploiting SO(6) and conformal
covariance and point-permutation symmetry we show how to reduce the lowest component
of the correlator of four stress tensors in N = 4 SYM to that of a single N = 2
hypermultiplet correlator. Section 3 provides a minimum of information about the N = 2
harmonic superspace formulation [41] of N = 4 SYM. Section 4 is central, and contains
the proof of our partial non-renormalisation theorem based on the insertion formula.
This section also includes the necessary study of the possible contact terms. In Section 5
we verify that the AdS/CFT correspondence holds for the non-renormalised part of the
correlator. In Appendix A we state the differential constraints on this four-point function
found by the more abstract analysis of [40]. We give their general solution, from which it
is evident that these constraints are weaker than the ones obtained in Section 4.
2. The N = 4 SYM four-point stress-tensor correlator

Here we show how one can compute the leading scalar term of the N = 4 four-point
function of four SYM supercurrents (stress-tensor multiplets) from the leading scalar term
of an N = 2 hypermultiplet four-point function. This section is based on [30] but we
also derive some additional restrictions on the amplitude following from point-permutation
symmetry.
We recall that in N = 4 superspace the N = 4 field strength superfield W A , A =
1, . . . , 6, transforms under the vector representation of the R symmetry group SO(6)
SU (4). This superfield satisfies an on-shell constraint reducing it to six real scalars, four
Majorana spinors and a vector.
The N = 4 supercurrent is given by
195
1
T AB = W A W B AB W C W C
(2.1)
6
(the trace over the YangMills indices is implied). It is in the symmetric traceless 20 of
SO(6) and is conserved as a consequence of the on-shell constraints on W A .
The four-point function we are going to consider is

G(N=4) = T A1 B1 T A2 B2 T A3 B3 T A4 B4 ,
(2.2)
where the numerical subscripts indicate the point concerned. This function can be
expressed in terms of SO(6) invariant tensors multiplied by scalar factors which are
functions of the coordinates. Given the symmetry of G(N=4) , the only SO(6) invariant
tensor that can arise is the Kronecker , and there are two modes of hooking up the indices,
each of which can occur in three combinations making six independent amplitudes in all. 5
Thus, for the leading component in the expansion 6 we have

(12 )2 (34)2
(13 )2 (24 )2
+
a
(s,
t)
G(N=4) =0 = a1 (s, t)
2
4 x4
4 x4
x12
x13
34
24
(14 )2 (23 )2
13 14 23 24
+ b1 (s, t) 2 2 2 2
4 x4
x14
x
23
13 x14 x23 x24
12 14 3234
12 13 42 43
+ b2 (s, t) 2 2 2 2 + b3 (s, t) 2 2 2 2 ,
x12 x14 x32x34
x12 x13 x42 x43
+ a3 (s, t)
(2.3)
2 = (x x )2 and, for example,

where xpq
p
q
A2 B2 A4 B4
(12 )2 (34 )2 = {A
,
1 B1 } {A3 B3 }
{A {B
A }B }
13 14 23 24 = {A13B14} {A42 B23} ,
(2.4)
and where the braces denote trace free symmetrisation at each point.
When writing down (2.3) we have taken into account the conformal covariance of the
correlator. In each term we have introduced the corresponding scalar propagator structure
which has the required conformal weight of the correlator. This implies that the coefficient
functions a1,2,3 , b1,2,3 are conformal invariants, so they depend on the two conformal
cross-ratios
s=
2 x2
x12
34
2 x2
x13
24
t=
2 x2
x14
23
2 x2
x13
24
(2.5)
Next, we notice the point-permutation symmetry of the correlator (2.2). To implement it

it is sufficient to require invariance under two permutations, for instance,
1 3:
s t
and 1 2 :
s
1
s , t .
t
t
(2.6)
5 An alternative explanation is provided by a double OPE. Indeed, each pair of operators is decomposed into
six irreps of SO(6), 20 20 = 1 + 15 + 20 + 84 + 105 + 175. Then the double OPE, being diagonal, gives rise
to six structures.
6 In Section 4 we shall show that superconformal covariance allows us to restore the complete dependence
of (2.2), given its leading component.
196
This leads to the following constraints:

s 1
,
a1 (s, t) = a3 (t, s) = a1 ,
t t

s 1
b1 (s, t) = b3 (t, s) = b1 ,
,
t t

s 1
,
,
t t

s 1
b2 (s, t) = b2 (t, s) = b3 ,
.
t t

a2 (s, t) = a2 (t, s) = a3
(2.7)
So, the six coefficients in the N = 4 amplitude are in fact reduced to only two
independent ones, one of the ai and one of the bi . Now we shall show that one can
determine these two functions by studying a certain N = 2 component of the N = 4
correlator. Let us see what happens when one reduces N = 4 supersymmetry to N = 2.
The first step is to decompose the SO(6) vector W A in a complex basis as 3 + 3 under
SU(3) and further as 2 + 1 + 2 + 1 under SU(2):

.
W A W i i , W 3 W, Wi i , W3 W
(2.8)
Here W is the N = 2 SYM field-strength and i (i = 1, 2) is the N = 2 matter
hypermultiplet.
From the on-shell constraints on W A it is easy to derive that these superfields (evaluated
at 3 = 4 = 0) obey the corresponding constraints,
i
D
W = 0,
!
j
Di D W
(2.9)
= 0,
(2.10)
and
1 j k
j
i
(2.11)
D
i D
k
= 0.
2
Eq. (2.9) means that W is a chiral superfield (a kinematic constraint) while Eq. (2.10) is
the YangMills equation of motion, and Eq. (2.11) are the field equations of the hypermultiplet. 7
In N = 2 harmonic superspace (see Section 3 for a review) all SU(2) indices are
projected out by harmonic variables thus obtaining objects which carry U (1) charge but
q+
are singlets under SU(2). For instance, the harmonic superfield q + and its conjugate
are related to the ordinary hypermultiplet N = 2 superfield i by
D(i j ) = 0,
i
q + = u+
i ,

q + = u+i i ,
(2.12)
on shell.
For our purpose of identifying the coefficients in the N = 4 amplitude it will be sufficient
to restrict T AB to the following N = 2 projections involving only hypermultiplets:
Tij = i j , at points 1 and 3,
T ij = i j , at points 2 and 4.
+
Then we multiply the resulting hypermultiplet correlators by u+
i uj
(2.13)
at each point and obtain
7 Strictly speaking, Eqs. (2.9)(2.11) hold only for an Abelian gauge theory and in the non-Abelian case a
gauge connection needs to be included. However, the constraints satisfied by the gauge-invariant bilinears that we
shall consider are in fact the same in the Abelian and non-Abelian cases.
+ + + + + + + +
q
G(N=2) =
q
q q q
q q q .
197
(2.14)
Its leading component is

(12)2 (34)2
(14)2 (23)2
+
a
(s,
t)
G(N=2) =0 = a1 (s, t)
3
4 x4
4 x4
x12
x14
34
23
+ b2 (s, t)
(12)(23)(34)(41)
,
2 x2 x2 x2
x12
23 34 41
(2.15)
where, for example,

+j
(12) = u+i
1 u2 !ij .
It is now clear that if we know the coefficient functions in the N = 2 correlator (2.15),
using the symmetry properties (2.7) we can obtain all the six coefficients in the N = 4
amplitude (2.3). Recalling that the symmetry of the correlator under the exchange of points
1 3 (or 2 4) relates the coefficients a1 and a3 to each other, we conclude that the
correlator (2.15) is, in principle, determined by two a priori independent functions of the
conformal cross-ratios.
This result is, of course, completely general, and must hold perturbatively as well as nonperturbatively. In the following we will study the radiative corrections to this correlator,
and will show that they are given in terms of a single independent coefficient function; the
ratios of the quantum parts of the functions a1 , a3 , b2 are completely fixed:
q. corr.
a1
= sF (s, t),
q. corr.
a3
= tF (s, t),
q. corr.
b2
= (1 s t)F (s, t)
(see Eq. (4.18) below). To this end we have to supplement the pure symmetry arguments
given above by some knowledge about the dependence of the hypermultiplet correlator on
the Grassmann variables (G-analyticity), superconformal covariance and a new, dynamical
property, the so-called harmonic (H-)analyticity.
3. The hypermultiplet in harmonic superspace

In the preceding section we showed that all the information about the N = 4
correlator (2.2) is contained in the N = 2 one (2.14). There is another reason why we
prefer to work with N = 2 rather than N = 4 superfields: the non-renormalisation theorem
that we are going to prove in Section 4 is based on the insertion formula originating in
the standard off-shell Lagrangian formulation of field theory. The absence of an off-shell
formulation of N = 4 SYM theory makes it difficult to justify this procedure in N = 4
superspace. However, there exists an off-shell reformulation of the N = 4 theory in terms
of N = 2 harmonic superfields [41].
We start by a brief review of the formulation of the N = 2 hypermultiplet in
harmonic superspace (the reader may wish to consult [47] for the details). It can be
described as a superfield in the Grassmann (G-)analytic superspace [41] with coordinates
xA , + , + , u
i . Here ui are the harmonic variables which form a matrix of SU(2) and
198
parametrise the sphere S 2 SU (2)/U (1). A harmonic function f (p) (u ) of U (1) charge
p is a function of u
i which is invariant under the action of the group SU(2) (acting on the
index i of ui ) and homogeneous of degree p under the action of the group U (1) (acting
2
on the index of u
i ). Such functions have infinite harmonic expansions on S whose
coefficients are SU(2) tensors (multispinors). The superspace is called G-analytic since it
only involves half of the Grassmann variables, the SU(2)-covariant harmonic projections
i
+ = u+i i , + = u+
i . In a way, this is the generalisation of the familiar concept
of a left- or right-handed chiral superfield depending on a different half of the Grassmann
variables, either the left-handed ( ) or right-handed ( ) one.
In this framework the hypermultiplet is described by a G-analytic superfield of charge
q + (xA , + , + , u) where tilde means a special
+1, q + (xA , + , + , u) (and its conjugate
conjugation on S 2 preserving G-analyticity). G-analyticity can also be formulated as
differential constraints on the superfield:
+ q + = 0,
D+ q + = D
(3.1)
+
+
. Note the similarity between (3.1) and the chirality
= u+i D
where D+ = ui D i , D
i
condition (2.9). In fact, both of them are examples of what is called a short multiplet in
the AdS/CFT language [16].
It is well-known that this N = 2 supermultiplet cannot exist off shell with a finite set of
auxiliary fields [48]. This only becomes possible if an infinite number of auxiliary fields
(coming from the harmonic expansion on S 2 ) are present. On shell these auxiliary fields
are eliminated by the harmonic (H-)analyticity condition (equation of motion)
D ++ q + = 0.
(3.2)
Here D ++ is the harmonic derivative on S 2 (the raising operator of the group SU(2)
realised on the U (1) charges, D ++ u+ = 0, D ++ u = u+ ).
The reader can better understand the meaning of Eq. (3.2) by examining the general
solution to the H-analyticity condition on a (non-singular) harmonic function of charge p:

f (p) = 0
if p < 0,
D ++ f (p) u = 0
(3.3)
+ i1 ...ip
u
f
if p 0.
f (p) = u+
i1
ip
In other words, the solution only exists if the charge is non-negative and it is a polynomial
of degree p in the harmonics u+ . The coefficient f i1 ...ip forms an irrep of SU(2) of
isospin p/2. Thus, H-analyticity is just an SU(2) irreducibility condition having the form
of a differential constraint on the harmonic functions.
Now it becomes clear why the combination of the G-analyticity constraints (3.1)
with the H-analyticity one (3.2) is equivalent to the original on-shell hypermultiplet
constraints (2.11). Indeed, from (3.3) one derives (2.12) and then, by removing the arbitrary
+ , one arrives at (2.11).
harmonic commuting variables from both q + and D + , D
The crucial advantage of the harmonic superspace formulation is that the equation of
motion (3.2) can be derived from an off-shell action given by an integral over the G-analytic
superspace:
SHM = du d 4 xA d 2 + d 2 +
(3.4)
q + D ++ q + .
199
This is the starting point for quantisation of the theory in a straightforward way [49]. In
particular, one can introduce the following propagator (two-point function):
+

(12)

qa (1)qb+ (2) =
.
2 ab
4 2 x12
(3.5)
Here
x12 = xA1 xA2 +

4i + + + +
1 2 1 1 + 2 1 2 2 + 1+ 2+ + 2+ 1+ ,
(12)
+j
is a supersymmetric coordinate difference and, e.g., (1 2) = ui

1 u2 !ij . This propagator
satisfies the Greens function equation (suppressing the colour indices),
+

q (1)q + (2) = 4 (x12 ) 2 1+ 2+ 2 1+ 2+ (u1 , u2 ).
D ++
(3.6)
In deriving (3.6) one makes use of the following property of the harmonic derivative of a
singular harmonic function: D1++ (12)1 = (u1 , u2 ).
Let us now assume that the spacetime points 1 and 2 are kept apart, x1 = x2 . Then the
right-hand side of Eq. (3.6) vanishes and the two-point function (3.5) becomes H-analytic:

+
q (1)q + (2) = 0,
D1++
if x1 = x2 .
(3.7)
This property can be extended to any correlation function involving gauge-invariant

composite operators made out of hypermultiplets, O = Tr[(
q + )p (q + )q ]:
D1++ O(1) = 0,
if x1 = x2 , x3 , . . . .
(3.8)
In reality Eq. (3.8) is a SchwingerDyson equation based on the free field equation (3.2).
Hence, its right-hand side contains contact terms like in (3.6), which vanish if the space
time points are kept apart. So, H-analyticity is a dynamical property of such N = 2
correlators holding away from the coincident points. It will play an important rle in the
next section.
Finally, just a word about the other ingredient of the N = 4 theory, the N = 2 SYM
multiplet. Here we do not need to go through the details of how it is formulated in harmonic
superspace and subsequently quantised [50]. We just recall that the corresponding field
strength is a (left-handed) chiral superfield which is harmonic independent, D ++ W = 0.
The N = 2 SYM action is given by the chiral superspace integral
1
SN=2 SYM = 2 d 4 xL d 4 Tr W 2 ,
(3.9)
4
where is the (complex) gauge coupling constant. 8
4
2 . In a
d xR d 4 Tr W
topologically trivial background the coupling constant becomes real and the two forms are equivalent (up to a
total derivative).
8 In fact, there exists an alternative form given by the right-handed chiral integral
200
4. Superconformal covariance and analyticity as the origin of non-renormalisation

In this section we are going to derive the constraints on the G-analytic correlator
(2.14) following from superconformal covariance and H-analyticity. Applied to the
lowest component in its expansion (see (2.15)), H-analyticity simply means irreducibility
under SU(2), from which one easily derives the three independent SU(2) tensor structures.
Further, conformal covariance implies that their coefficients are arbitrary functions of the
conformal cross-ratios (in the particular case (2.15) point-permutation symmetry reduces
the number of independent functions to two).
The combined requirements of superconformal covariance in G-analytic superspace and
H-analyticity have further, far less obvious consequences which arise at the next level of
the expansion of the correlator. They have been derived in [40] where it was found that
one of the three coefficient functions remains unconstrained while two linear combinations
of them have to satisfy certain first-order linear differential constraints. The solution of the
latter allows for some functional freedom which cannot be fixed on general grounds. A
summary of these constraints as well as their explicit solution is given in Appendix A.
Alternatively [43], one can apply a more efficient procedure which, as it turns out,
also completely fixes the above freedom. It is based on the insertion formula [13] (for
an explanation in the context of N = 2 harmonic superspace see [26,43]). This formula
relates the derivative of a 4-point correlator of the type (2.14) with respect to the (complex)
coupling constant to a (4 + 1)-point correlator obtained by inserting the N = 2 SYM
action (3.9):
+ + + + + + + +

(N=2)
q
q
G
d 4 xL0 d 4 0 Tr W 2 (0)
q q q
q q q .
(4.1)
G(N=2)
Recall that unlike the matter superfields q + which are G-analytic and harmonic-dependent
off shell, W is chiral and harmonic-independent. The integral in the insertion formula (4.1)
goes over the chiral insertion point 0. As we shall see later on, the combination of chirality
with G-analyticity, in addition to conformal supersymmetry and H-analyticity, impose
strong constraints on this five-point function.
Let us try to find out what could be the general form of a (4 + 1)-point correlator
(0|2, 2, 2, 2) which is chiral at point 0 and G-analytic at points 1, . . . , 4 (with U (1)
charges +2), has the corresponding superconformal properties and is also H-analytic,
Dr++ (0|2, 2, 2, 2) = 0,
if xr = xs , r, s = 1, . . . , 4.
(4.2)
In particular, it should carry a certain R weight. Indeed, the expansion of the matter
superfield q + = i (x)u+
i + starts with the physical doublet of scalars of the N = 2
hypermultiplet which have no R weight. At the same time, the N = 2 SYM field strength
W = + F (x) contains the YM field strength F (R weight 0) in a term
with two left-handed s, so the R weights of W and of the Lagrangian equal 2 and 4,
respectively. From (4.1) it is clear that this weight is compensated by that of the chiral
superspace measure d 4 xL d 4 , so the correlator on the left-hand side of Eq. (4.1) has the
expected weight 0.
201
The task now is to explicitly construct superconformal covariants of R weight 4 out of

the coordinates of chiral superspace xL0 , 0i at the insertion point 0 and of G-analytic
harmonic superspace xAr , r+ , r+ , u
ri , r = 1, . . . , 4, at the matter points. To this
end we need to know the transformation properties of these coordinates under Q and S
supersymmetry (parameters ! and , correspondingly). Since we are only interested in the
leading term in the expansion, it is sufficient to examine the linearised transformations
of the Grassmann variables [51]:

i +
i
+ = u+
ui + O 2 ,
i = ! i + xL i + O 2 ,
i ! + xA
2
+ i
i
+ = u+
(4.3)
i ! ui xA + O .
We remark that Q supersymmetry acts as a simple shift, whereas the S supersymmetry
terms shown in (4.3) are shift-like. Let us assume for the moment that we stay away from
any singularities in x-space (we shall come back to this important point in a moment). Then
we can use the four left-handed parameters ! i and x i to shift away four of the six left-
handed spinors 0i and r+ . This means that our correlator effectively depends, to lowest
order, on two left-handed spinor coordinates. We can make this counting argument more
explicit by forming combinations of the s which are invariant under Q supersymmetry
and under the shift-like part of S supersymmetry. Q supersymmetry suggests to use the
differences
+
= 0i u+
0r
ri r ,
Q 0r
= 0,
r = 1, . . . , 4.
(4.4)
:
Then we can form the following two cyclic combinations of three 0r
1
1
1
(12r ) = (12) x0r
0r + (2r) x01
01 + (r1) x02
02 , r = 3, 4,
(4.5)
+j
where x0r xL0 xAr are translation-invariant and (rs) u+i

r us !ij are SU(2)-invariant
combinations of the spacetime and harmonic coordinates, correspondingly. It is now easy
to check that 12r are Q and S invariant to lowest order (i.e. shift-invariant):

Q+S 12r = O 2 .
(4.6)
Here one makes use of the harmonic cyclic identity
(rs)ti + (st)ri + (tr)si = 0.
(4.7)
As a side remark we point out that the above counting of Q and S shift-invariant
Grassmann variables can also be applied to the four-point function (2.14). It depends
on four G-analytic + s and their conjugates + . This number equals that of the Q
and S supersymmetry parameters, therefore we conclude that there exists no invariant
combination (under the assumption that we keep away from the coincident spacetime
points). In other words, there are no nilpotent superconformal G-analytic covariants having
the properties of the correlator (2.14). This means that given the lowest component (2.15)
in the expansion of (2.14) and using superconformal transformations we can uniquely
reconstruct the entire correlator (2.14). A similar argument applies to the N = 4
correlator (2.2).
202
Let us now inspect the structure of the correlator (0|2, 2, 2, 2) more closely. As we noted
earlier, it has R weight 4. In superspace the only objects carrying R weight are the odd
coordinates, so the expansion of our correlator must start with the product of four
left-handed s, i.e., the correlator should be nilpotent [21,43]. Further, superconformal
covariance requires that the shift-like transformations (4.3) do not produce structures with
less than four s, so we must use all the four available shift-invariant combinations (4.5)
(notice that they have R weight 1, even though they are right-handed spinors). Thus, we
can write down the leading term in the correlator in the following form:

2
2
(0|2, 2, 2, 2) = (12)2 123
(4.8)
124
F (x, u) + O 5 .
The coefficient function F depends on the spacetime and harmonic variables and carries
vanishing U (1) charges, due to the explicit harmonic prefactor (12)2 .
The nilpotent prefactor in (4.8) can be expanded in terms of 0 and + . In fact, what
contributes in the insertion formula (4.1) is just the purely chiral term in it:
2
2
124
= (0 )4
(12)2 123

R
+ terms containing + ,
2
2
2
2
x01x02 x03 x04
(4.9)
where
2 2
2 2
R = (12)2(34)2 x14
x23 + (14)2(23)2 x12
x34

2 2
2 2
2 2
+ (12)(23)(34)(41) x13 x24 x12 x34 x14
x23
2 2

1
2 2
2 2
x34 x13
x24 x14
x23
= (12)2 (34)2 x12
2
2 2

2 2
2 2
+ (13)2 (24)2 x13
x24 x12
x34 x14
x23
2 2

2 2
2 2
x23 x12
x34 x13
x24 .
+ (14)2 (23)2 x14
(4.10)
R
has been written in two different ways using the harmonic cyclic
Here the polynomial
identity (4.7). Note that the superconformal covariant (4.9) is completely permutation
symmetric in the points 1, . . . , 4, although this is not obvious from the form of the lefthand side.
Next, we substitute (4.9), (4.10) into (4.1) and carry out the integration over the insertion
point. The 0 integral is trivial due to the Grassmann delta function (0 )4 in (4.9). The result
is

(N=2)
(12)2 (34)2
(14)2 (23)2
G
F (s, t) s
+t
4
4
4 x4
x12 x34
x14
23

(12)(23)(34)(41)
+ (1 s t)
(4.11)
,
2 x2 x2 x2
x12
23 34 41
where
1
F (s, t) =
s
d 4 x0
4 x4 x2 x2
x12
34 14 23
2 x2 x2 x2
x01
02 03 04
F (x, u),
(4.12)
is an arbitrary conformally invariant four-point function (the factor 1/s is introduced for
convenience).
203
The reason why we have not indicated any harmonic dependence on the left-hand side
of (4.12) has to do with our last requirement, namely, the H-analyticity condition (4.2) on
the four-point amplitude (4.11):

Dr++ (12)2 (34)2F = Dr++ (14)2(23)2 F = Dr++ (12)(23)(34)(41)F = 0.
(4.13)
Since the function F has vanishing U (1) charges, from (3.3) one easily derives that it must
be harmonic independent, i.e., an SU(2) singlet:
F = F (s, t).
(4.14)
In the analysis so far we have not taken into account possible contact terms in the (4 +1)correlator (0|2, 2, 2, 2). As pointed out in [6], they may become important in the context of
the insertion formula (4.1). Concerning the G-analytic points 1, . . . , 4, we have decided
to keep away from the coincident points, and this assumption allowed us to impose Hanalyticity. Thus, we never see contact terms of the type 4 (xrs ), r, s = 1, . . . , 4. However,
there can also be contact terms involving the insertion point 0 and only one of the matter
points, i.e., singularities of the type (0 )n 4 (x0r ). Since we are supposed to integrate
over x0 in the insertion formula (4.1), such terms may result in a non-contact contribution
to the left-hand side. Therefore we must investigate them in detail. 9
In order to do this we have to adapt our construction of nilpotent superconformal
covariants. As before, we can profit from the available S-supersymmetry freedom to shift
away two of the four Q-invariant combinations 0r , but this time we should be careful
not to use singular (in spacetime) coordinate transformations. Let us suppose that we are
dealing with a contact term containing, e.g., (0 )n 4 (x04). In the vicinity of the singular
point x0 x4 the matrix x04 is not invertible anymore, so we cannot shift away 04 .
However, since the four matter points are kept apart, the matrices x01 , x02 and x03 still are
invertible. This allows us to shift away, e.g., 01 and 02 . Then, just as before, we will be
left with only two spinors, 03 and 04 , all of which must be used to construct the nilpotent
covariant of R weight 4. Thus, the latter is unique. This counting can also be done in
a manifestly Q- and S-covariant way. We can still use the shift-invariant combination 123
from (4.5) but not 124 because it is now singular. Then we replace the latter by
x04
(24)
(41)
1
1
124 = 04 +
x04 x01
x04 x02
01 +
02 ,
04 =
(12)
(12)
(12)
which is regular at x04 = 0. Thus, the contact analog of (4.8) is

(0|2, 2, 2, 2)
2 2
= (0 )n 4 (x04) 123
contact
04 Fcontact(x, u) + O 5
(4.15)
(4.16)
(note that if there are no derivatives on the delta function, we will only see the first
term in (4.15), but we need the other two in the general case). Next, using the identity
9 Contact terms of the ultra-local type and their effect on the non-renormalisation of two- and three-point
functions have been studied in [43].
204
2 = x 2 2 /(12)2 and (4.9)) in the insertion formula we obtain a contribution to

04
04 124
G(N=2) having exactly the same form as (4.11) but where now
x4 x4 x2 x2
1
Fcontact(s, t) =
(4.17)
d 4 x0 (0 )n 4 (x04 ) 122 342 142 23 Fcontact(x, u).
s
x01 x02x03
As before, H-analyticity implies that this function must be harmonic independent.

In the above construction of contact contributions we have singled out the matter point 4
but any other matter point can be obtained by cyclic permutations (it is obvious that two
such terms cannot help each other to achieve superconformal covariance). Remarking that
the harmonic polynomial R in (4.11) is invariant under such permutations, we conclude
that the possible contact terms do not alter the form of the four-point function (4.11). 10
We remark that this argument also removes a possible loophole which so far had remained in the proofs of the non-renormalisation theorems for extremal and next-toextremal four-point functions as presented in [26,28].
Thus the combination of superconformal covariance, chirality, G- and H-analyticity
within the context of the insertion formula (4.1) restricts the freedom in any four-point
correlator of the type G(N=2) to a single function of the conformal cross-ratios. We recall
that (4.1) only gives the derivative of G(N=2) with respect to the coupling. This means that
from (4.11) and (2.15) we can read off the quantum corrections to the three coefficients:
q. corr.
a1
= sF (s, t),
q. corr.
a3
= tF (s, t),
q. corr.
b2
= (1 s t)F (s, t).

(4.18)
Using the symmetry constraints (2.7) we find that F should satisfy

s 1
1
,
F (s, t) = F (t, s) = F ,
t
t t
(4.19)
and we determine the remaining coefficients a2 , b1 and b3 . Finally, we express all the six
coefficients in the N = 4 correlator (2.3) in terms of a single unconstrained function of the
cross-ratios (with the symmetry properties (4.19)):
a1 = A0 + sF (s, t),
a2 = A0 + F (s, t),
a3 = A0 + tF (s, t),
b1 = B0 + (s t 1)F (s, t),
b2 = B0 + (1 s t)F (s, t),
b3 = B0 + (t s 1)F (s, t).
(4.20)
The constants correspond to the free-field part which cannot be determined by the
insertion formula. The direct tree-level computation including disconnected diagrams
yields, respectively:
A0 =
4(Nc2 1)2
,
(2)8
B0 =
16(Nc2 1)
.
(2)8
(4.21)
The result for the connected correlator is obtained from this by setting A0 = 0.
10 It is a separate issue whether contact terms of this type can actually arise in a supergraph calculation. So far
experience at two and three loops [32,43] has shown that this is not the case. Nevertheless, the possibility remains
open.
205
Another way to state our non-renormalisation theorem is to say that in (4.20) one finds
linear combination of the coefficient functions which are non-zero at the free-field level,
but vanish for all radiative corrections.
5. The supergravity result and the AdS/CFT correspondence

According to the AdS/CFT correspondence, the connected part of the correlator of four
stress-tensor multiplets in N = 4 SYM in the limit Nc and = g 2 Nc large but fixed
should match a certain tree-level correlator in N = 8 AdS supergravity in five dimensions.
More precisely, the leading component of the CFT correlator (2.3) that we are discussing
should correspond to the correlator of four sets of supergravity scalars lying in the 20 of
SO(6). The latter has recently been computed by Arutyunov and Frolov [46]. The purpose
of this section is to first rewrite the result of [46] in a much simpler form and then compare
it to Eq. (4.20) from Section 4. We find a perfect match between the CFT prediction
and the explicit AdS result which, in our opinion, is new evidence for the validity of the
correspondence conjecture.
According to [46], the two independent coefficients of the (connected) amplitude
are, e.g.,

1 2
1 1
32Nc2
t
3
a1
x
+
D
=
D
+
D
+
(5.1)
2211
3322
2222 ,
4 x4
2
2 x34
s
s 12
2
28 10
x12
34
and
b2
32Nc2
= 8 10
2
2
2
2
x12 x34 x14x23 2

1 s 1
t
2
+ 6 D2222 + 4x12
D3322
s
s
t
t

1
1
2
2
+ 4x14 D3223 + x12D2211 2 2 2 2
x12 x34 x14x23

1
1
2
+ x14 D2112 2 2 2 2
x14 x23 x12 x34

1
1
2
,
+ x13
D2121 2 2 + 2 2
x12 x34 x14 x23

(5.2)
and the four others can be obtained from these by symmetry according to Eq. (2.7). The
scaling weights in the subscript of the D-functions refer to points 1 through 4 in the
obvious order. These functions were first introduced in [14]; they denote the basis integrals
appearing in tree-level four-point calculations in AdS supergravity. They can be written in
various forms, of which the most useful one for our purpose is the following [52]:
D1 2 3 4

1 +2 +3 +4
1
1 1
4 1
2
2
= K dt1 dt4 t1
t4
St
exp
ti tj xij .
St
0
i<j
(5.3)
206
Here St = t1 +t2 +t3 +t4 , and the prefactor is (specialising to the (4 +1)-dimensional case)
4
2 8( 1 ++
2)
2
.
(5.4)
28(1 ) 8(4 )
Using this representation for D1111 and integrating out the global scaling variable St , one
immediately obtains the Feynman parameter representation of the standard one-loop box
integral with external momenta x12 , x23 , x34 , x41 . As is well-known, this integral can
be expressed in terms of logarithms and dilogarithms [53]; following [54], we write the
result as
1
1
D1111 = 2 2 (1) (s, t),
(5.5)
K1111
x13x24
K=
where
(1) (s, t) =

2
1
t 1 + t
+ ln(s) ln(t) +
2 Li2 (s) + Li2 (t) + ln ln
,
s 1 + s
3
(5.6)
and 11
(s, t) =
(1 s t)2 4st,
(s, t) = 2(1 s t + )1 .
(5.7)
Li2 denotes the dilogarithm function.

Moreover, from the representation (5.3) it is also obvious that all the D-functions
occurring in the formulas above for a1 and b2 can be obtained from D1111 /K1111 by
appropriate differentiations with respect to the xij2 (this fact was already noted in [14]).
Applying this algorithm we rewrite a1 , b2 in term of (third-order) differential operators
acting on the basic function (1) (s, t). Next, note that

1
s +t 1
s (1) (s, t) = 2 (1) (s, t)(1 s + t) + 2 ln(s) ln(t)
(5.8)
,
s
and similarly with s t. These identities are sufficient to express inductively arbitrary
derivatives of (1) with respect to s, t by (1) itself and logarithmic terms. The final result
of this procedure is
a1 = sF (s, t),
(5.9)
16Nc2
+ (1 s t)F (s, t),
b2 =
(5.10)
(2)8
where

16Nc2 (1)
(s, t)12st (1 + s + t)2 + 10st
F (s, t) =
8
6
(2)

+ ln(s)2s 1 + t 2 s st + 10t 2 + 30st (1 + t s)

+ ln(t)2t 1 + s 2 t st + 10s 2 + 30st (1 + s t)

+ (1 + s + t)4 + 20st2 .
(5.11)
11 Here we assume 2 > 0; the case 2 < 0 requires an appropriate analytic continuation.
207
The supergravity result (5.9), (5.10) exactly matches the form of our general prediction (4.20). The first term on the right-hand side of (5.10) agrees, to leading order in Nc , 12
with the connected free-field part (4.21); the function F , which is the sum of all quantum
corrections in the appropriate limit, has the required permutation symmetry properties and
occurs with the expected simple prefactors.
6. Conclusions
The correlator of four stress-tensor multiplets in N = 4 SYM originally contains six
amplitudes which can be reduced to only two independent functions quite trivially, just by
exploiting the obvious symmetries. On grounds of superconformal invariance, Grassmann
and harmonic analyticity we show that the quantum part of the amplitudes is in fact
universal to all of them. It is given by one a priori arbitrary function of the conformal
cross-ratios which occurs in all six amplitudes with simple prefactors.
This property had been observed in weak coupling perturbation theory at two and three
loops and we verify it for strong coupling by simplifying the supergravity result of [46].
We regard this as new evidence in favour of the AdS/CFT conjecture. It provides also a
highly non-trivial check on the results of [46].
Alternatively, one can turn the argument around: The supergravity result does obey
constraints originating from harmonic superspace. This makes us believe that there exists
an appropriate harmonic superspace formulation of AdS supergravity in which all these
properties become manifest.
We remark that N = 4 supersymmetry has not played any particular rle here. What
we actually used was N = 2 superconformal invariance, a property which a large class of
N = 2 finite theories possess [56]. We would need N = 4 conformal supersymmetry if we
wish to reconstruct the correlator of four stress tensors (the highest spin in the N = 4 SYM
multiplet) from that of the scalar composites considered here. If we restrict ourselves to
finite N = 2 theories, we can still claim a non-renormalisation property of the correlator of
four hypermultiplet bilinears.
We have carried out the analysis of contact terms in the context of the reduction formula
to the extent necessary for our present purposes. By an explicit construction of the relevant
type of contact terms we have shown that, at least at the four-point level, their appearance in
the Intriligator formula would not alter the form of the result, which remains proportional
to the universal polynomial R . This also removes the only possible loophole remaining
in the proof of the non-renormalisation theorems for extremal and next-to-extremal fourpoint functions as presented in [26,28]. Those contact terms could, however, in principle
have invalidated the two- and three-loop computations presented in [32,43]. Since their
results have been confirmed by independent calculations [29,33] we conclude that, up to
the three-loop level, contact terms of the malignant type are absent.
12 As it is not clear whether the gauge group in the AdS/CFT correspondence is of unitary or special unitary
type, the supergravity calculation [46] assumed U (Nc ) for simplicity.
208
Finally, it would be of obvious interest to study whether the universality property found
here has non-trivial implications for the operator product expansion.
Acknowledgements
Burkhard Eden, Christian Schubert and Emery Sokatchev are indebted to Paul Howe
and Peter West for sharing their knowledge on correlators of constrained superfields. B.E.
and E.S. profited form stimulating discussions with Dan Freedman and Eric DHoker.
Anastasios Petkou thanks the Alexander von Humboldt Foundation for support.
Appendix A. Differential constraints on the four-point function

The main result of this paper is that the four-point correlator (2.2) is determined by
a single function of the spacetime variables, apart from simple terms reflecting the freefield part of the amplitudes. The argument we gave makes use of the insertion formula (4.1)
and therefore explicitly distinguishes free-field and quantum parts.
Alternatively [40], one can directly study the consequences of superconformal covariance and H-analyticity on the higher-level terms in the expansion of N = 2 correlators of
the type (2.14). The constraints found in this way in [40] are reproduced here and solved
explicitly. As before, we find a function F (s, t) contributing to all six amplitudes in the
same way as shown in (4.20), but there is additionally an a priori arbitrary function of one
variable, named h in the following. We express the free-field part of the amplitudes a1 , b2
in terms of these functions. The expression for h found in this way is not zero, whereby
it is clear that it cannot be omitted on general grounds. This, in our opinion, explains why
the direct approach to the four-point function inevitably yields weaker constraints.
We start by introducing the following two linear combinations of the coefficients
a1 , a3 , b2 of the N = 2 correlator (2.15): 13
1
s1
1
= a1 + 2 a3 + b2 ,
s
t
t
1s t
1
a3 b2 .
2
t
t
(A.1)
As shown in [40], the full implementation of H-analyticity combined with superconformal

covariance leads to the following constraints:
s = ss tt ,
(A.2)
t = ss + (t 1)t + .
(A.3)
These first-order coupled differential equations have an integrability condition in the form
of a second-order equation for each function:
= = 0,
13 The cross-ratios used in [40] are related to (2.5) as follows: s = t/s, t = 1/s.
(A.4)
209
where
= sss + tt t + (s + t 1)st + 2s + 2t .
(A.5)
We substitute
(s, t) =
A(s, t)
,
(s, t) =
B(s, t)
,
(A.6)
where has been defined in (5.7). Then we perform a transformation to the variables
=
2s
,
1s t +
2t
,
1s t +
(A.7)
whose inverse is 14
s=
,
(1 + )(1 + )
t=
,
(1 + )(1 + )
1
> 0.
(1 + )(1 + )
(A.8)
This brings the first equation in (A.4) to its normal form:

A = 0.
(A.9)
Its general solution obviously is

A(, ) = f ( ) + g(),
(A.10)
where f, g are arbitrary functions. So, we have found

(, ) =

1
f ( ) + g() .
(A.11)
Given , it is not difficult to solve for from the first-order system (A.2), (A.3) (the
integration constant can be absorbed into a redefinition of f, g). The result is

1 f ( )
g()
(, ) =
(A.12)
+
.
1+
1+
At this point we should recall that a1 , a3 , b2 appearing in (A.1) in fact originate from the
N = 4 correlator (2.3) and are therefore subject to the symmetry requirements (2.7). Under
the permutations (2.6) we have
1 3:
and
1 2:
1+
1
,
.
1+
(A.13)
Taking this into account leads to identifying the two functions in (A.11), (A.12):
f ( )
g( )
=
h( ).
3(1 + )
3(1 + )
(A.14)
14 When inverting (A.7) we have to remember that (5.7) is defined as a positive square root. Correspondingly,
the choice of sign in the last of Eq. (A.8), as well as the form of the first two, determines the allowed domain of
the variables , . The use of such variables has been studied in [55].
210
Further, the two independent coefficients, e.g., a1 and b2 become

1
1
a1 =
F (, ) + 2h( ) h() + h
+h
,
1
1+
1+

1 +
1
1
1 2
h( ) + h() .
b2 =
F (, ) + h
+h
1
1+
1+
1
(A.15)
Here the functions F (, ) and h( ) satisfy the following constraints:

1+
1
,
F (, ) = F (, ) = F
1+

1
1+
+h
= Cq (A0 + B0 ),
h( ) + h
(A.16)
1+
where the constants on the r.h.s. of the second equation are the free-field values from
equation (4.20), possibly plus a quantum correction Cq . The functions F, h may be split
into a free-field part

1
1
h0 ( ) =
A0 +
B0 ,
3
F0 (, ) =
A0 3 + 3 + 1 A0 3 + 3 + 1
+
+ B0 ,
3
( + 1)
3
( + 1)
(A.17)
and a quantum part hq , Fq .

In conclusion, the direct approach to the four-point correlator (2.14) leaves more
freedom than the one from Section 4. Indeed, comparing (A.15) with (4.19), (4.20) we
see that in the latter case F = Fq / and the quantum part of h vanishes, which is our nonrenormalisation result. The fact that the argument given in this appendix applies to both
the free-field and quantum part explains why it cannot be as strong as the one based on the
insertion formula.
References
[1] D. Anselmi, D.Z. Freedman, M.T. Grisaru, A.A. Johansen, Nucl. Phys. B 526 (1998) 543, hepth/9708042;
D. Anselmi, J. Erlich, D.Z. Freedman, A. Johansen, Phys. Rev. D 57 (1998) 7570, hepth/9711035.
[2] M. Henningson, K. Skenderis, JHEP 9807 (1998) 023, hep-th/9806087;
M. Henningson, K. Skenderis, Fortschr. Phys. 48 (2000) 125, hep-th/9812032.
[3] F. Bastianelli, S. Frolov, A.A. Tseytlin, Nucl. Phys. B 578 (2000) 139, hep-th/9911135.
[4] S. Lee, S. Minwalla, M. Rangamani, N. Seiberg, Adv. Theor. Math. Phys. 2 (1998) 697, hepth/9806074.
[5] E. DHoker, D.Z. Freedman, W. Skiba, Phys. Rev. D 59 (1999) 045008, hep-th/9807098.
[6] A. Petkou, K. Skenderis, Nucl. Phys. B 561 (1999) 100, hep-th/9906030.
[7] J. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231, hep-th/9711200.
[8] S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Phys. Lett. B 428 (1998) 105, hep-th/9802109.
[9] E. Witten, Adv. Theor. Math. Phys. 2 (1998) 253, hep-th/9802150.
211
[10] D.Z. Freedman, S. Mathur, A. Matusis, L. Rastelli, Nucl. Phys. B 546 (1999) 96, hepth/9804058.
[11] G. Chalmers, H. Nastase, K. Schalm, R. Siebelink, Nucl. Phys. B 540 (1999) 247, hepth/9805105.
[12] H. Liu, A.A. Tseytlin, Phys. Rev. D 59 (1999) 086002, hep-th/9807097.
[13] K. Intriligator, Nucl. Phys. B 551 (1999) 575, hep-th/9811047.
[14] E. DHoker, D.Z. Freedman, S. Mathur, A. Matusis, L. Rastelli, Nucl. Phys. B 562 (1999) 353,
hep-th/9903196.
[15] P.S. Howe, P.C. West, Int. J. Mod. Phys. A 14 (1999) 2659, hep-th/9509140;
P.S. Howe, P.C. West, Phys. Lett. B 389 (1996) 273, hep-th/9607060;
P.S. Howe, P.C. West, Is N = 4 YangMills soluble?, hep-th/9611074;
P.S. Howe, P.C. West, Phys. Lett. B 400 (1997) 307, hep-th/9611075.
[16] L. Andrianopoli, S. Ferrara, Phys. Lett. B 430 (1998) 248, hep-th/9803171.
[17] S. Ferrara, M. Porrati, A. Zaffaroni, Lett. Math. Phys. 47 (1999) 255, hep-th/9810063.
[18] P.S. Howe, E. Sokatchev, P.C. West, Phys. Lett. B 444 (1998) 341, hep-th/9808162.
[19] F. Gonzalez-Rey, B. Kulik, I.Y. Park, Phys. Lett. B 455 (1999) 164, hep-th/9903094.
[20] K. Intriligator, W. Skiba, Nucl. Phys. B 559 (1999) 165, hep-th/9905020.
[21] B. Eden, P.S. Howe, P.C. West, Phys. Lett. B 463 (1999) 19, hep-th/9905085.
[22] W. Skiba, Phys. Rev. D 60 (1999) 105038, hep-th/9907088.
[23] S. Penati, A. Santambrogio, D. Zanon, JHEP 9912 (1999) 006, hep-th/9910197.
[24] E. DHoker, D.Z. Freedman, S. Mathur, A. Matusis, L. Rastelli, UCLA/99/TEP/27, hepth/9908160.
[25] M. Bianchi, S. Kovacs, Phys. Lett. B 468 (1999) 102, hep-th/9910016.
[26] B. Eden, P.S. Howe, C. Schubert, E. Sokatchev, P.C. West, Phys. Lett. B 472 (2000) 323, hepth/9910150.
[27] J. Erdmenger, M. Perez-Victoria, Phys. Rev. D 62 (2000) 045008, hep-th/9912250.
[28] B. Eden, P.S. Howe, E. Sokatchev, P.C. West, Phys. Lett. B 494 (2000) 141, hep-th/0004102.
[29] F. Gonzalez-Rey, I. Park, K. Schalm, Phys. Lett. B 448 (1999) 37, hep-th/9811155.
[30] B. Eden, P.S. Howe, C. Schubert, E. Sokatchev, P.C. West, Nucl. Phys. B 557 (1999) 355,
hep-th/9811172.
[31] B. Eden, P.S. Howe, C. Schubert, E. Sokatchev, P.C. West, Phys. Lett. B 466 (1999) 20, hepth/9906051.
[32] B. Eden, C. Schubert, E. Sokatchev, Phys. Lett. B 482 (2000) 309, hep-th/0003096.
[33] M. Bianchi, S. Kovacs, G. Rossi, Y.S. Stanev, Nucl. Phys. B 584 (2000) 216, hep-th/0003203.
[34] P.S. Howe, P.C. West, Phys. Lett. B 400 (1997) 307, hep-th/9611075.
[35] B. Conlong, P.C. West, J. Phys. A 26 (1993).
[36] H. Osborn, Ann. Phys. 272 (1999) 243, hep-th/9808041.
[37] A.G.M. Pickering, P.C. West, Nucl. Phys. B 569 (2000) 303, hep-th/9904076.
[38] S.M. Kuzenko, S. Theisen, Class. Quant. Grav. 17 (2000) 665, hep-th/9907107.
[39] F.A. Dolan, H. Osborn, Nucl. Phys. B 593 (2001) 599, hep-th/0006098.
[40] B. Eden, P.S. Howe, A. Pickering, E. Sokatchev, P.C. West, Nucl. Phys. B 581 (2000) 523,
hep-th/0001138.
[41] A. Galperin, E. Ivanov, S. Kalitzin, V. Ogievetsky, E. Sokatchev, Class. Quant. Grav. 1 (1984)
469.
[42] G.G. Hartwell, P.S. Howe, Int. J. Mod. Phys. A 10 (1995) 39013920, hep-th/9412147;
G.G. Hartwell, P.S. Howe, Class. Quant. Grav. 12 (1995) 18231880.
[43] P.S. Howe, C. Schubert, E. Sokatchev, P.C. West, Nucl. Phys. B 571 (1999) 71, hep-th/9910011.
[44] E. DHoker, D.Z. Freedman, Nucl. Phys. B 544 (1999) 612, hep-th/9809179;
E. DHoker, D.Z. Freedman, Nucl. Phys. B 550 (1999) 261, hep-th/9811257.
[45] G. Chalmers, K. Schalm, Nucl. Phys. B 554 (1999) 215, hep-th/9810051.
[46] G. Arutyunov, S. Frolov, Phys. Rev. D 62 (2000) 064016, hep-th/0002170.
212
[47] A.S. Galperin, E.A. Ivanov, V.I. Ogievetsky, E.S. Sokatchev, Harmonic Superspace, Cambridge
Univ. Press, to appear.
[48] P.S. Howe, K. Stelle, P.C. West, Class. Quant. Grav. 2 (1985) 815;
K.S. Stelle, Santa Barbara, preprint NSF-ITP-85-001.
[49] A. Galperin, E. Ivanov, V. Ogievetsky, E. Sokatchev, Class. Quant. Grav. 2 (1985) 601;
A. Galperin, E. Ivanov, V. Ogievetsky, E. Sokatchev, Class. Quant. Grav. 2 (1985) 617.
[50] B.M. Zupnik, Theoret. Mat. Fiz. 69 (1986) 207;
B.M. Zupnik, Phys. Lett. B 183 (1987) 175.
[51] A. Galperin, E. Ivanov, V. Ogievetsky, E. Sokatchev, in: Quantum Field theory and Quantum
Statistics, Vol. 2, A. Hilger, Bristol, 1987, p. 233, preprint JINR E2-85-363 (1985).
[52] G. Arutyunov, S. Frolov, A. Petkou, Nucl. Phys. B 586 (2000) 547, hep-th/0005182.
[53] G. t Hooft, M. Veltman, Nucl. Phys. B 153 (1979) 365.
[54] N.I. Ussyukina, A.I. Davydychev, Phys. Lett. B 298 (1993) 363.
[55] A.I. Davydychev, J.B. Tausk, Nucl. Phys. B 397 (1993) 123.
[56] J.P. Derendinger, S. Ferrara, A. Masiero, Phys. Lett. B 143 (1984) 133;
P.S. Howe, K. Stelle, P. West, Phys. Lett. B 124 (1983) 55.

7-branes and higher KaluzaKlein branes

Ernesto Lozano-Tellechea a , Toms Ortn a,b
a Instituto de Fsica Terica, C-XVI, Universidad Autnoma de Madrid, E-28049 Madrid, Spain
b I.M.A.F.F., C.S.I.C., Calle de Serrano 113 bis, E-28006 Madrid, Spain
Received 29 December 2000; accepted 4 April 2001
Abstract
We present and study a new chain of 10-dimensional T duality related solutions and their
11-dimensional parents whose existence had been predicted in the literature based in U duality
requirements in 4 dimensions. The first link in this chain is the S dual of the D7-brane. The next
link has 6 spatial worldvolume dimensions, it is charged w.r.t. the RR 7-form but depends only on 2
transverse dimensions since the third has to be compactified in a circle and is isometric and hence is
similar in this respect to the KK monopole. The next link has 5 spatial worldvolume dimensions, it
is charged w.r.t. the RR 6-form but, again, depends only on 2 transverse dimensions since the third
and fourth have to be compactified in circles and are isometric and so on for the following links.
All these solutions are identical when reduced over the p spatial worldvolume dimensions and
preserve a half on the available supersymmetries. Their masses depend on the square of the radii of
the isometric directions, just as it happens for the KK monopole. We give a general map of these
branes and their duality relations and show how they must appear in the supersymmetry algebra.
2001 Elsevier Science B.V. All rights reserved.
Introduction
In the last few years there has been a lot of interest in discovering classical solutions of
effective superstring theories (supergravity theories) with such properties that one could
argue that they represent the fields produced by solitonic objects present in the superstring
spectrum. The interplay between the knowledge of the superstring spectrum and the
knowledge of classical solutions has been very fruitful since each of them has contributed
to the increase of the other. The two most important tools used in this field have been
supersymmetry and duality. Unbroken supersymmetry ensures in many cases the absence
of corrections of the classical solutions and the lack of quantum corrections to the mass
of the corresponding objects in the string theory spectrum. Hence, more effort has been
put in finding supersymmetric (i.e., admitting Killing spinors) solutions, associated to BPS
E-mail addresses: ernesto.lozano@uam.es (E. Lozano-Tellechea), tomas@leonidas.imaff.csic.es (T. Ortn).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 7 7 - 8
214
E. Lozano-Tellechea, T. Ortn / Nuclear Physics B 607 (2001) 213236
string states. Duality transformations preserve in general supersymmetry, relating different

states in dual theories. In general [1], but not always [2] duality relations between different
higher-dimensional theories manifest themselves as non-compact global symmetries of the
compactified supergravity theory that leave invariant its equations of motion so one can
use them to transform known solutions into new solutions, preserving their supersymmetry
properties.
Thus, it so happens that most classical solutions of superstring effective field theories
belong to chains or families of solutions related by duality transformations. The best known
chain of solutions is that of the Dp-branes, with p = 0, . . . , 8 in 10 dimensions. They
belong to two different theories: 10-dimensional type IIA for p even and 10-dimensional
type IIB for p odd. All of them preserve 1/2 of the supersymmetries available, represent
objects with p spatial worldvolume dimensions and 9 p transverse dimensions (Dirichlet
(p+1) whose existence was
branes), carrying charge associated to the RR (p + 1)-form C
discovered by Polchinski [3], and are related by generalized Buscher type II T duality
transformations [2,4].
Sometimes it is possible to find families of solutions that are, by themselves,
representations of the duality group in the sense that they are invariant, as families, under
the full duality group. This is the case, for instance, of the SWIP solutions of N =
4, d = 4 supergravity constructed in Refs. [5,6]. In that case one can argue that all the
solitonic objects of a given type (charged, stationary, black holes) and preserving a certain
amount of supersymmetry are described by particular solutions, with particular values of
the parameters of that general family. More interesting cases are N = 8, N = 4 with 22
vector multiplets and general N = 2 theories, all in d = 4, but fully general solutions
in their duality-invariant form are not available. A great deal is, however, known of the
solitonic spectrum of the 4-dimensional theories due to our knowledge of their duality
groups (the so-called U duality group in the N = 8 case). All these theories can be
obtained from 10-dimensional theories by compactification (toroidal or more general) and
the compactification of the solitonic 10-dimensional objects gives rise to 4-dimensional
solitonic objects of different kinds, depending on how the 10-dimensional objects are
wrapped in the internal dimensions and one can study if these objects fill 4-dimensional
duality multiplets. It has been realized that this is not the case if one considers only the
standard 10-dimensional solitons: the Dp-branes, KK monopole, gravitational wave (W),
fundamental string (F1) and solitonic 5-brane (S5) [710]. More 10-dimensional solitons
are needed to give rise to all the 4-dimensional solitons predicted by duality and some of
the properties they should exhibit, in particular the dependence of the mass in the radii of
the internal dimensions and the coupling constant, have been deduced.
In this paper we present candidates for some of the missing 10-dimensional solitons
and study them. The key to their construction is the realization that there are 4dimensional duality symmetries which are neither present in 10 dimensions nor are a
simple consequence of reparametrization invariance in the internal coordinates. These are,
in general, S duality (i.e., electricmagnetic) transformations which only exist in certain
dimensions and that enable us to use the mechanism reduction-S dualization-oxidation to
generate new solutions in higher dimensions.
215
Let us consider a familiar example: 5-dimensional gravity compactified in a circle. The

4-dimensional theory has electricmagnetic duality and one expects an S duality symmetric
spectrum. However, if we only considered the 5-dimensional plane wave solution we would
only find electrically charged 4-dimensional solitons. To find the magnetically charged
ones we S dualize these and, oxidizing the solutions to 5 dimensions we find the Kaluza
Klein (KK) monopole [11,12]. In principle, this is a solution one would not expect in 5
dimensions since it has one dimension necessarily compactified in a circle.
At this point it is useful to remember why the KK monopole has necessarily one compact
dimension. First of all, it is known that only if the special isometric direction is compact
with the right periodicity the solution is free from string singularities. Secondly, the mass
(tension per unit (world)volume) of the KK monopole depends quadratically on the radius
of that direction. The masses of standard branes are always proportional to the volume
of the internal manifold in which they are compactified and tend to infinity when those
volumes tend to infinity. Thus, when considering uncompactified standard branes (all Dbranes, the fundamental string and the solitonic 5-brane) the quantity of interest is not
the mass, but the brane tension which is finite. In the KK monopole case, though, due to
the quadratic dependence of the mass on the radius of the special isometric direction, the
tension is proportional to the radius of that direction and diverges in the decompactification
limit.
The solutions we present here can also be generated by a mechanism similar to the
one we have explained for the KK monopole, exploiting S dualities present in dimensions
lower than 10 and 11 and also have properties similar to those of the KK monopole: there
are dimensions that cannot be decompactified because the masses of these objects depend
quadratically on the radii of those dimensions. Somehow this is consistent with the fact that
they are generated using dualities that only exist if some of the dimensions are compact.
One of the problems raised by the need to consider new 10- and 11-dimensional
solutions was that fact that the 10- and 11-dimensional supersymmetry algebras did
not contain central charges associated to those possible new objects. In our opinion the
predictive power of the supersymmetry algebras has been overestimated and we will
propose a way to include in them these new objects.
The rest of this paper is organized as follows: in Section 1 we present our family of
T duality-related solutions whose construction via the reduction-S dualization-oxidation
mechanism is explained in Section 2. In Section 3 we find other duality-related solutions
in 10 and 11 dimensions. In Section 4 we calculate the dependence of the masses of these
objects on compactification radii and coupling constants and in Section 5 we calculate the
Killing spinors of all the solutions we have presented. Our conclusions are in Section 6.
In Appendix B we derive the SL(2, R)/SO(2) sigma model from toroidal compactification
and explain how SL(2, R) is broken to SL(2, Z) and in Appendix A we briefly review
holomorphic (d 3)-brane solutions of the SL(2, R)/SO(2) sigma model to clarify certain
points.
216
1. The basic family of solutions

The basic family of solutions are solutions of the type II supergravity theories in d =
10 and are a sort of deformation of the family of Dp-brane solutions for 0 p 7. As
such, they have p + 1 worldvolume coordinates t, yp = (y 1 , . . . , y p ) and 9 p transverse
coordinates. We combine two of them into the complex coordinate and the remaining
7 p we denote by x7p = (x 1 , . . . , x 7p ). The solutions can collectively be written in
the string-frame metric in the form 1

H 1/2 2
H 1/2 2
2=
2
1/2

d
s
d x7p ,
dt d yp (H HH) d d

HH
HH

(p+1) ty 1 y p = (1)[(p+1)/2] H
C

HH
(1.1)
(7p)
C
,
x 1 x 7p =

HH

H 3p/4
e =
,

HH
where we function H = H() is a complex, holomorphic, (multivalued) function of ,
1
log around = 0, where we assume the object
i.e., H = 0 with the behavior H 2i
2
is placed. Its real and imaginary parts are
H = A + iH.
(1.2)
These solutions have the same form as the standard Dp-brane solutions if we delete
but they are clearly different. In particular we can
everywhere the combination HH,
understand them as having 7 p extra isometric directions that should be considered
compact. 3 Our goal will be to understand how they arise, their M theoretic origin and
their supersymmetry properties and explore the implications of it all. We will also find
other solutions related by dualities with them or belonging to the same class. Since we will
find that all these solutions preserve a half of the symmetries, we are going to argue that
they describe the long range fields of elementary, non-perturbative objects of string theory
and we will calculate their masses.
1 For convenience, we give the form of the potential to which the p-brane naturally couples C
(p+1) and the
(7p)

dual one C
. In the p = 3 case, these are the two non-vanishing sets of components of the 4-form potential
with self-dual field strength. (Our conventions are those of Ref. [4] whose type II T duality rules, generalizing
those of Ref. [2], we use.) Since the solutions we will be dealing with are not asymptotically flat, we do not write
explicitly the asymptotic values of the scalars (for example, 0 for the dilaton).
2 The analytic extension of H to the whole space (for which the above expression is clearly not valid) is
a non-trivial problem that depends, among other things, on the topology assumed for the space. In general, it
requires the introduction of other singularities around which H is also multivalued so that one gets a consistent
monodromy. This problem was first considered in Ref. [21].
3 It seems difficult (it is perhaps impossible) to extend the dependence of the function H to those coordinates.
Furthermore, the construction procedure reduction-S dualization-oxidation and the dependence of the masses on
the radii of those dimensions that we are going to calculate later on suggest that those coordinates should be
compactified on a torus.
217
2. Construction of the solutions

The solutions (1.1) can be obtained by successive T duality transformations in
worldvolume directions of the p = 7 solution. The p = 7 solution is nothing but the
type IIB solitonic 7-brane (S7) that was obtained by S duality from the D7-brane and
called Q7-brane in Ref. [4]. The worldvolume directions are transformed into transverse
isometric directions that should be considered compact. 4 Thus, we obtain a chain of T dual
solutions of both type II theories.
There is an alternative way of constructing these solutions that also helps to understand
them. Let us consider a piece of the 10-dimensional type II supergravity theories in which
(8p) of the RR (7 p)-form
we only keep the metric, the dilaton and the field strength G
(7p) . The action is
C

(1)7p (8p) 2

2 +
.
S = d 10 x |g|
(2.1)
e2 R
4( )
G
2 (8 p)!
Now, let us compactify it over a (7 p)-torus using a simplified KaluzaKlein ansatz
that only takes into account the volume modulus of the internal torus, the dilaton (both
rewritten in terms of two convenient scalars and ), the internal volume mode of the RR
(7 p)-form, a and the (3 + p)-dimensional Einstein metric g :

1 7p
1 p+1
1
1
2 ,
d s = exp +
g dx dx exp +
d x7p
2
2 p+1
2
2 7p
(7p) x 1 x 7p = a,
(2.2)
C

7p p+1
p3
+
.
e = exp
4
4
7p
After some straightforward calculations one obtains, in all cases, the reduced action

1
1
2
()
S = d p+3 x |g| R +
(2.3)
,
+
2 (m )2 2
where
= a + ie ,
(2.4)
i.e., gravity coupled to an SL(2, R)/SO(2) sigma model parametrized in the standard form
by the complex scalar (sometimes known as axidilaton although here this name could be
misleading since in some cases (p = 3) the string dilaton simply does not contribute to
it) and another scalar, , decoupled from . In the p = 7 case (d = 10) this is the wellknown piece of the type IIB supergravity action. In lower dimensions, it is integrated in
much bigger sigma models associated to much bigger U duality groups 5 but it is a most
interesting part of it.
4 This is somewhat analogous to what happens in the well-known duality between the solitonic five-brane S5
and the KK monopole in which a transverse direction of the S5 is T dualized into an isometric, compact, direction
of the KK monopole.
5 In d = 6 dimensions, this model was studied in Ref. [13] and in d = 8 it was studied in Ref. [14].
218
There is a very general solution of this model

2
2
2
ds = dt d yp H d d ,
(2.5)
= H,
= 0,
with H = 0. In d = 10 (p = 7) this is just the general D7-brane solution. Choosing H
log we get the single D7-brane solution. In lower dimensions, these solutions are just
compactifications of the standard general Dp-brane solution in which we have assumed
that the harmonic function only depends on two transverse directions () and we have
dualized the RR (p + 1)-potential, giving rise to the real part of H. Thus, this is a wellknown solution.
We can now perform an SL(2, R) duality rotation of this solution 6 1/ , since
this is a symmetry of the dimensionally reduced action 7 that leaves the Einstein metric
invariant. This is not a symmetry of the 10-dimensional action and one really needs extra
compact dimensions to establish it. The resulting solutions 8
2
2
2
ds = dt d yp H d d ,
(2.6)
= 1/H,
= 0,
are nothing but the solutions Eqs. (1.1) reduced according to the above KK ansatz.
What we are doing here is similar to what one does in standard KK theory: reducing
to 4 dimensions the 5-dimensional pp wave one obtains the electric, extreme KK black
hole. Since the d = 4 theory has electricmagnetic duality as a symmetry, one can find
the magnetic, extreme KK black hole and then uplift it to d = 5 to find the KK monopole
[11,12] that has a special isometric direction that cannot be decompactified. The symmetry
between the pp wave and the KK monopole cannot be established without assuming one
compact direction. It is only natural, by analogy, to consider here that the dimensions that
we have compactified cannot be decompactified after the duality transformations. We will
support this assumption not by geometrical arguments but calculating the masses of these
objects and finding its dependence on the radii of those dimensions.
3. Duality-related solutions and M theoretic origin

Since we are dealing with many new solutions, we first propose to denote them by Dpi
where p + 1 is the worldvolume and i is the number of isometric directions. According
6 Continuous duality symmetries are usually broken to their discrete subgroups, for instance SL(2, R) is usually
broken to SL(2, Z). This can be clearly seen in the case in which the SL(2, R)/SO(2) sigma model originates in
a toroidal compactification and is explained in Appendix B. In other cases one has to study the quantization of
charges to arrive to the same conclusion. We will loosely use the continuous of the discrete form of the duality
group in the understanding that in some contexts only the discrete one is really a symmetry of the theory.
7 In general, it is only a symmetry of the equations of motion of the complete, untruncated, type II theory.
8 In Appendix A we discuss these general solutions and in which sense they are new. We stress that we are
considering only the choice holomorphic function H 21 i log .
219
to this notation, the solutions described by Eq. (1.1) are in the p = 7 case D70 (the type IIB
S dual of the D7-brane, called Q7 in Ref. [4]), D61 for p = 6, and D52 , D43 , D34 , D25 ,
D16 , D07 for the remaining cases.
For all the type IIB solutions in the class (1.1) we can find an S dual using the 10dimensional type IIB S duality symmetry. While in the p = 7 case the S dual solution
is just the well-known D7-brane, and in the p = 3 case the solution is self-dual, in the
p = 5, 1 cases we find genuinely new solutions. For D52 we get a solution which is a
deformation of the solitonic five-brane, and we call S52
d ss2 = dt 2 d y52 H d d
d x 2 ,
2
H
H
Bx 1 x 2 = H H

H 1

Bty 1 y 5 =
,

HH

H 1/2
e =
,

HH
S52
(3.1)
and for D16 , we get a sort of deformation of the fundamental string solution that we call
F16

H 1 2
d ss =
dt dy 2 H d d d x62 ,
HH

H 1
,
B
ty

HH
F16

x 1 x 6 = A ,
HH

1/2
e =
.

HH
(3.2)
These two solutions only have non-trivial common sector NSNS fields and therefore they
are also solutions of the heterotic string effective field theory. We can also understand
these solutions by appealing to the existence in both cases of a reduced action of the form
Eq. (2.3) that arises from the 10-dimensional actions

1 2

2+
S = d 10 x |g|
(3.3)
e2 R
H ,
4( )
2 3!
and

S=
d 10 x

2 +
4( )
|g|
e2 R

1 2
2 ,
e H
2 7!
(3.4)

is the dual 7-formfield
is the NSNS 3-form field strength and H
= e2 H
where H
strength. Reducing the first action to 8 dimensions with the ansatz
220
1
2
d x
3 g
22 ,
dx dx e
d s = e
x 1 x 2 = a,
(3.5)
B

3
1
e = e 2 2 ,
and the second action down to 4 dimensions with the ansatz
1
2
2
d s = e g dx dx e 3 d x6 ,
1 6 = a,

(3.6)
B
x x

3
1
e = e 2 + 2 ,
we get in both cases Eq. (2.3) in 8 and 4 dimensions.
As for the M theoretic origin of the type IIA solutions, they can be derived from the
following 11-dimensional solutions through compactification of the 11th dimension (z): a
pp wave with 7 extra isometries
WM7
d s2 = 2 dt dz
H
d d d x72 ,
dz2 HH

HH
a deformation of the M2-brane

H 2/3 2
d s =
dt d y22
HH

1/3
2/3 d d H
H 1/3HH
d x62 ,

HH
M26

H 1

,
C ty 1 y 2 = HH

x 1 x 6 = A

C

HH
a deformation of the M5-brane

H 1/3 2
d
s
=
dt d y52
HH

H 2/3 2
2/3
1/3

H (HH) d d
d x3 ,

HH
M53

H 1

,
C ty 1 y 5 =
HH
x 1 x 2 x 3 = A ,
C

HH
and the KK monopole (with no dependence on the 11th dimension)

2
KK7M d s2 = dt 2 d y62 H d d + dz2 H 1 dy 7 A dz .
(3.7)
(3.8)
(3.9)
(3.10)
In these four cases we can also trace the origin of the solution to the existence of a sector
like that in Eq. (2.3) in the reduced action of 11-dimensional supergravity. In the purely
gravitational cases, the action Eq. (2.3) can be derived from the dimensional reduction of
221
the Einstein term alone as shown in detail in Appendix B. In the second and third cases,
one needs the 6-form or the 3-form dual potential, respectively.
In some cases the dimensional reduction of these 11-dimensional solutions in isometric
directions different from z produce new 10-dimensional solutions. In particular, we get two
purely gravitational solutions
H
d d d x 2 ,
dz2 HH
6

HH
and the KaluzaKlein monopole with no dependence in z

2
KK6 d s2 = dt 2 d y52 H d d + dz2 H 1 dy 7 A dz .
W6
d s2 = 2 dt dz
(3.11)
(3.12)
In all cases (see Fig. 1) we see that whenever we reduce the same 11-dimensional
solution over 2 directions to 9 dimensions and we do it in different order, we get a pair
of 9-dimensional solutions that form an SL(2, R) (SL(2, Z)) doublet and also originate
from a type IIB SL(2, R) (SL(2, Z)) doublet as it must [2].
4. Masses
The mass of the Dpi solutions can be calculated using S and T duality rules from the
standard D7-brane and can be written in a general formula:
MDpi =
R3 Rp+2 (Rp+3 R9 )2
p+2i+1
g 3 $s
(4.1)
The masses of the NSNS solutions found by S duality from the D52 and the D16 are
MS52 =
R3 R7 (R8 R9 )2
,
g 2 $10
s
MF16 =
R3 (R4 R9 )2
.
g 4 $14
s
(4.2)
The masses of the 11-dimensional objects from which the type IIA objects can be
derived can be calculated using the relations between the 11-dimensional Planck length
9
$(11)
Planck and the radius of the 11th dimension R10 and the type IIA string coupling constant
1/3
(11)
gA and the string length $s $Planck = 2$s gA and R10 = $s gA :
MM26 =
MM53 =
R3 R4 (R5 R10 )2
(11)
(
$Planck )15
R3 R6 (R7 R8 R9 )2 R10
(11)
(
$Planck )12
(4.3)
$Planck = $Planck /2 .
where $Planck is the reduced 11-dimensional Planck length
(11)
9R
(11)
(11)
m
11 is the conventional name in the literature. Here we use Rm for the radius of the coordinate x .
222
Fig. 1. Duality relations between KK branes. The numbers in parenthesis represent the worldvolume dimension, isometric and transverse directions. The
arrows indicate dimensional reduction in the corresponding kind of direction. In the upper row we represent M-theory KK branes, below 10-dimensional
type IIA branes and below them 9-dimensional branes. Type IIB KK branes are in the bottom row. Pairs of branes in boxes are S duality doublets. They are
always related to reductions from 11 to 9 dimensions of the same object in two different orders. Sometimes there is an third object with the same numbers
as those in a doublet, but transforming as a singlet and we denote it with (s).
223
These expressions should be compared with the well-known expression of the mass of
the 11-dimensional KK monopole KK7M when the special isometric direction is x 10
MKK7M =
2
R4 R9 R10
9
(
$(11)
Planck )
(4.4)
or the 10-dimensional KK monopole KK6 (A or B) when the special isometric direction

is x 9
MKK6 =
R4 R8 R92
.
g 2 $8s
(4.5)
In both cases the mass is not simply proportional to the volume of the brane which is
assumed wrapped on a torus but depends quadratically on the radius of the special isometric
direction. The same happens to the masses of all the Dpi branes: they depend quadratically
on the radii of the directions that we have argued are isometric, which supports our
assumption.
Apart from the dependence on the radii we see that in general these objects are highly
non-perturbative since their masses are proportional to g 3 and g 4 except for S52 , whose
mass goes like g 2 , as for any standard solitonic object.
The momentum of the WM7 solution is
MWM7 =
3
(R3 R9 )2 R10
(11)
(
$Planck )18
(4.6)
5. Killing spinors and unbroken supersymmetries

It is important to find the amount of supersymmetry preserved by our solutions since,
if they preserve less than one half of the total supersymmetry available, one could
argue that they correspond to composite objects. Since all these solutions are related by
S and T duality transformations to the D7-brane, which preserves exactly 1/2 of the
supersymmetries, it is to be expected that they will do so as well. Nevertheless, a direct
calculation of the Killing spinors should always be performed since it will confirm our
expectations and it will also provide us with projectors that will help us to associate the
solutions to central charges in the supersymmetry algebra and therefore to identify them
with supersymmetric states in the string spectrum.
We first calculate the Killing spinors of the Dpi family of solutions with the obvious
choice for the vielbein basis 10
1/4
1/4
HH
HH
ei i =
,
em m =
,
H
H
1/4
HH
8
9
e8 = e9 =
(5.1)
H 1/2.
H
10 Underlined indices are world indices and non-underlined indices are tangent space indices. They take values
in the ranges i = 0, 1, . . . , p , m = p + 1, . . . , 7.
224
For the type IIA solutions we use the supersymmetry transformation rules for the
gravitino and dilatino which, in the purely bosonic background we are considering, take
the form 11

8p
1
1
i
(8p)
2
/ +
e G
/
' ,
(11 )
' = 4
8 (8 p)!
(5.2)

8p
i p 3 (8p)
' = /

e G
+
/
(11 ) 2 ' .
4 (8 p)!
Imposing the vanishing of dilatino transformation rule we obtain the following constraint
in the Killing spinor:

8p
1 i p+1 8 9 (11) 2 ' = 0,
(5.3)
or, equivalently

10p
1 (1)[p/2] i 0 p (11) 2 ' = 0.
(5.4)
zero the worldvolume (t, y i ) and transverse, isometric
This constraint automatically sets to

(x m ) components of the supersymmetry variation of the gravitino. The remaining
transverse components (x 8 , x 9 ) give in all cases, the following coupled partial differential
equations

1 8 9
1
H
log(H
H)
+
log
' = 0,
8
9
8
' 8

4
8
HH
(5.5)

+ 1 9 log H
' 9 = 9 9 8 8 log(HH)
' = 0.

4
8
HH
Now, using the CauchyRiemann equations for the holomorphic function H, i.e.,
8 A = +9 H,
9 A = 8 H,
and 9 log(HH)
in the following way:
we can express 8 log(HH)
= 28 (arg H),
9 log(HH)
= +29 (arg H),

8 log(HH)
and the Killing spinor equations are easily seen to be solved by

10p
[p/2] i
0 p (11) 2 '0 = 0,
1 (1)
1/8
HH
1
' = e 2 arg(H) 8 9
'0 .
H
'0 being any constant spinor satisfying the above constraint.
In the type IIB cases we use the relevant supersymmetry transformation laws 12

1
1
1
(8p)
e G
P 9p ,
/ +
/
= 4
2
8 (8 p)!

1 3 p (8p)
= /
e G
+
/
P 9p ,
2
4 (8 p)!
(5.6)
(5.7)
(5.8)
(5.9)
11 Our type IIA spinors are full 32-component Majorana spinors.

12 Our type IIB spinors are pairs (whose indices 1, 2 are not explicitly shown of 32-component, positive chirality,
MajoranaWeyl spinors. Pauli matrices act on the indices not shown.
where Pn is

1,
Pn
i 2 ,
225
n even,
n odd.
Proceeding as in the type IIA case, we find the Killing spinors

1 + (1)[p/2] 0 p P p+1 0 = 0,
2
1/8
8 9 HH
12 arg(H)
0 ,
= e
H
(5.10)
where, now, 0 is any pair of constant positive-chirality MajoranaWeyl spinors satisfying

the above constraint.
The Killing spinors of the S52 and the F16 can be found in a similar fashion and are,
respectively

1 6 7 8 9 3 0 = 0,
8 9
= e 2 arg(H)
1
and
0 ,

0 1 3
1 + 0 = 0,
1/4
1
8 9 HH
= e 2 arg(H)
0 .
H
Before discussing these results it is worth finding the Killing spinors of the 11dimensional solutions. The only relevant supersymmetry transformation rule is that of the
gravitino, which with our conventions is:

1
i
G

'.
/ +
' = 2
(5.11)

2
144
In the obvious vielbein basis we find, for the W M7 solution

0 10 ' = 0,
1/4
1
9 HH
8
' = e 2 arg(H)
' 0
H
for the M26 solution

0 1 2 ' = 0,
1
+
i
1/6
1
9 HH
8
' = e 2 arg(H)
' 0 ,
H
for the M53 solution

0 4 10 ' = 0,
1/12
1
9 HH
8
' = e 2 arg(H)
' 0 ,
H
(5.12)
(5.13)
(5.14)
226
and for the KK7M solution, as it is well known, the Killing spinor is any constant spinor
' 0 satisfying the constraint

1 + i 0 6 ' 0 = 0.
(5.15)
In all cases one can see that these solutions preserve one half of the supersymmetries.
6. Conclusions
In this paper we have presented new 10-dimensional solutions of the type IIB theories
that can be thought of as having a certain number of isometric, compact, dimensions,
that cannot be decompactified (one could say that these are really solutions of lowerdimensional theories) and which we have referred generically to as KK-branes. We have
described how they can be obtained via the reduction-S dualization-oxidation which could
explain why some of the directions have to be compactified in circles since S duality only
exists in the compactified theory. Furthermore, we have computed the masses of these
solutions and we have found that they depend on the square of the radii of the directions
that we have identified as compact, just as it happens in the KK monopole case, which
is consistent with our identification. The mass formula are also coincident with what is
needed to complete the U duality invariant spectrum of N = 8, d = 4 supergravity [810].
It has also been recently argued that the presence of certain KK-branes is necessary to
explain from the M-theory point of view the existence of some massive/gauged type II
supergravities in lower dimensions [15].
Perhaps the only element that does not seem to fit in the picture we are putting forward
is the supersymmetry algebra since there seems to be no place in it for the new objects.
For the sake of concreteness we will focus in the 11-dimensional supersymmetry algebra
(M-algebra) but the problems and the solutions we propose can be applied in the obvious
way to other cases.
The M-algebra is usually written, up to convention-dependent numerical factors c, cn ,
in the form 13

c2 a1 a2 1 (2)
Q , Q = c a C 1 Pa +
C
Za1 a2
(6.1)
2
c5 a1 a5 1 (5)
+
C
Za1 a5 .
5!
A lightlike component of the momentum is then associated to the gravitational
waves moving in that direction, the spatial components of Z (2) and Z (5) are associated
respectively to M2- and M5-branes wrapped in those directions. The timelike components
have more complicated interpretations: in the Z (5) case, they are associated to the KK
monopole in a complicated way and in the Z (2) case they are associated to an object
that we would call the KK9-brane of which we only know that it should give the D8brane upon dimensional reduction. All these objects break (preserve) a half of the available
13 See, e.g., [16].
227
supersymmetries and strict relations between their masses and charges can be derived from
the algebra.
Clearly the M-algebra contains a good deal of information about the solitons of the
theory that realizes it (11-dimensional supergravity or M-theory). However, it is clear that
it does not contain all the information about them. To start with, it does not tell us why
some branes are fundamental and some are solitonic, it does not tell us why some objects
exist in the uncompactified theory (the wave, M2 and M5) while other objects only exist
when one dimension is compactified in a circle (the KK monopole and the KK9-brane).
Furthermore, all solitonic objects should be associated to spacelike components of central
charges: that is the result we would always get if we performed the calculation. All this
is not so surprising: the M-algebra is not derived from the theory and their solutions but
just by imposing consistency of the possible central charges. If we were able to derive the
algebra from M-theory and its solitonic solutions, the central charges would be associated
to specific objects and we would know whether they have compact dimensions or not.
Since we do know many things about the solitonic solutions, we can try to reflect what we
know in a form of the M-algebra mathematically consistent and then we can check if the
results are consistent with dualities.
To start with, we consider the M-algebra with the most general central extensions
allowed:

cn a1 an 1 (n)
C
Za1 an .
Q , Q = c a C 1 Pa +
(6.2)
n!
n=2,5,6,9,10
We know the wave is associated to P , the M2-brane to Z (2) and the M5-brane to Z (5) .
We also know [17] that the KK monopole is a sort of 6-brane with one of the 4 possible
transverse dimensions wrapped in a circle. We are going to reflect this fact by writing,
instead of just the Z (6) term as above, the term
c6 a1 a6 1 (7)
(6.3)
C
Za1 a6 a7 k a7 ,
6!
where k a is a vector pointing in the compact direction.
We also know that the KK9-brane (or M9-brane) [18] has 9 spacelike worldvolume
dimensions one of which is always wrapped on a circle. We reflect this fact by writing,
instead of just the Z (9) term as above, the term
c9 a1 a9 1 (8)
(6.4)
C
Za1 a8 la9 ,
9!
where la is a vector pointing in the direction around which the KK9-brane is wrapped.
We do not know of any brane associated to Z (10) and so we will not consider it in the
M algebra, which takes the form

c2 a1 a2 1 (2)
c5 a1 a5 1 (5)
Q , Q = c a C 1 Pa +
C
Za1 a2 +
C
Za1 a5
2
5!

c9 a1 a9 1 (8)
c6 a1 a6 1 (7)
C
Za1 a6 a7 k a7 +
C
Za1 a8 la9 .
+
6!
9!
(6.5)
228
We could certainly write more general central charges by allowing more vectors to be
present in the algebra, meaning allowing objects with more isometric directions such as
the M26 or the M53 branes presented in this paper. However, considering objects with just
one special isometry will be enough to present our ideas.
Let us now reduce this algebra in one dimension. From each of the standard central
charges we get two central charges in one dimension less, namely P , Z (0) from P , Z (1) ,
Z (2) from Z (2) and Z (4), Z (5) from Z (5) , corresponding to the known reductions of Mtheory solitons: wave and D0-brane from the wave, F1- and D2-brane from the M2brane and D4- and S5-brane from the M5-brane. From each of the new charges we have
introduced we get instead three lower dimensional central charges: from the contraction
Z (7)k associated to the KK monopole we get a Z (6) associated to the D6-brane when
k points in the direction we are reducing, we get a contraction Z (6)k associated to the
type IIA KK monopole (KK6A) if we reduce on the KK monopole worldvolume and we
get a Z (7) k associated to the D61 (called KK7A in Ref. [4], also studied in Ref. [19]) if
we reduce in a transverse direction. From the product Z (8)l we get a Z (8) , associated to
the D8-brane when we reduce the KK9-brane in the isometric direction l points to, we get
a product Z (7)l associated to an object with the same features of the M-theory KK9-brane
but in one dimension less and a product Z (8)l associated to a type IIA spacetime filling
KK9-brane referred to as (NS 9A)-brane in Ref. [20]. The result is the following form
of the type IIA supersymmetry algebra:
cn

a1 an 11 C 1 Za(n)
Q , Q = c a C 1 Pa +
1 an
n!
n=0,1,4,8
cn

a1 an C 1 Za(n)
+
1 an
n!
n=2,5,6

c5 a1 a5
c6 a1 a6 1 (7)
+
11 C 1 Za(6)
k a6 +
C
Za1 a6 a7 l a7
1 a5 a6
5!
6!
c9 a1 a9 1 (8)
c8 a1 a8 1 (7)
C
Za1 a7 ma8 +
C
Za1 a8 na9 .
+
8!
9!
(6.6)
Every known solitonic solution of the type IIA supergravity theory has an associated
charge in this algebra. If we now reduce again to nine dimensions we will get the algebra
of the massive 9-dimensional theories presented in Ref. [4] with SL(2, Z) covariance. This
is possible only because we have allowed for charges corresponding to KK-branes in 11
dimensions. To get the same algebra from the type IIB side a charge has to be introduced
for the S7-brane which, even though it does not carry any SO(2) R-symmetry indices, is
not invariant but is interchanged with the D7-brane charge under S duality. We will present
these results elsewhere.
Acknowledgements
T.O. would like to thank Eric Bergshoeff, Joaquim Gomis, Yolanda Lozano and Patrick
Meessen for most useful conversations, the C.E.R.N. Theory Division for its hospitality in
229
the first stages and the Groningen Institute for Theoretical Physics for its warm hospitality
in the last stages of this work and M. Fernndez for her support.
The work of E.L.-T. is supported by a U.A.M. grant for postgraduate studies. The
work of T.O. is supported by the European Union TMR program FMRX-CT96-0012
Integrability, Non-perturbative Effects, and Symmetry in Quantum Field Theory and by
the Spanish grant AEN96-1655.
Appendix A. Holomorphic (d 3)-branes
In this appendix we briefly discuss holomorphic (d 3)-brane solutions of the ddimensional SL(2, R)/SO(2) sigma model

1
,
S = d d x |g| R +
(A.1)
2 (m )2
where lives in the complex upper half plane and is defined up to modular PSL(2, Z)
transformations, so multivalued solutions are allowed if the value of changes by a
modular transformation.
(d 3)-brane-type solutions of this model were first considered in Ref. [21] in d = 4.
In these dimensions (d 3)-branes are strings. In that reference, the following general
solution of the above model was found 14

2
H d d ,
ds 2 = dt 2 d y(d3)
(A.2)
= H,
where H is, in principle, any complex holomorphic or antiholomorphic function of the
complex variable (i.e., either H = 0 or H = 0) and H = m(H). H is, therefore, a
real harmonic function of the 2-dimensional Euclidean spacetime transverse to the (d 2)dimensional worldvolume directions. Only functions with H 0 are admissible.
A few remarks are in order here: although g = H is in this solution equal to the
imaginary part of , it does not transform under PSL(2, Z). Modular invariance of the
metric is, therefore, not an issue. We could have wrongly concluded that in this solution,
the metric is not modular invariant because g = m( ) but, by definition, it is, since
the metric does not transform under PSL(2, Z). Then, the l.h.s. if that equation does not
transform, and the r.h.s. does, and we get a new solution (denoted by primes) with
a () + b aH + b
=
H ,
c () + d
cH + d
m( )
m(H )
g = g = m( ) =
=
.
| c + a|2 | cH + a|2
() =
(A.3)
We could remove if we wished the extra factor by a conformal reparametrization:

d =
d
,
cH () + a
14 Here we write the obvious generalization to any dimension d (see also Refs. [4,22]).
(A.4)
230
and we then could write again the new solution in a form similar to that of the original one
Eq. (A.2) but with a new holomorphic function H [( )]. Thus, as in Ref. [21] we could
have written from the beginning the general solution in the form

2
H |f ()|2 d d ,
ds 2 = dt 2 d y(d3)
(A.5)
= H,
where f () is any holomorphic function of , but this function can always be reabsorbed
into a holomorphic coordinate change = F (), dF /d = f and ( ) = [F 1 ( )].
All this said, it must be acknowledged that, even though modular invariance of the
metric is not an issue, its single-valuedness is. Since H will in general be a multivalued
function with monodromies in G, its imaginary part will also be multivalued and it might
be necessary to multiply it by |f ()|2 , with f () multivalued to make g single valued.
A second remark we can make here is that there exists another form of the general
solution which is manifestly SL(2, R) invariant without having to invoke coordinate
changes to show it:
2
2
2
2U d d ,
ds = dt d yp e
(A.6)
= H1 /H2 ,
2U
2 ),
e
= m(H1 H
where H1,2 are two arbitrary complex functions of the complex variable transforming as
a doublet under SL(2, R), i.e.,

H1
a b
H1
(A.7)
=
,
H2
H2
c d
both in and in the metric (but e2U is invariant, as it must). The structure of this family is
similar to that of the duality-invariant families of black-hole solutions of pure N = 4, d =
4 supergravity presented in Refs. [5,6,23], closely related to special geometry objects as
discovered in [24]. We can relate this general solution either to the solution Eq. (A.2) as
the particular case H1 = H, H2 = 1 or to the solution Eq. (A.5) as the particular case
2 ) = |H2 |2 m(H1 /H2 ).
H1 /H2 = H, f = H2 since m(H1 H
All this means that we cannot generate new solutions not in this classes via SL(2, R)
transformations.
Since all these solutions are equivalent, up to coordinate transformations, we take now
Eq. (A.2) and now consider the choice of function H. First, we have to choose between
holomorphic and anti-holomorphic H. This choice is related to the choice between (d 3)branes and anti-(d 3)-branes with opposite charge with respect to the (d 2)-form
potential dual to a. The impossibility of having H depending on both and is due
to the impossibility of having objects with opposite charges in equilibrium. We opt for
holomorphy.
Which holomorphic function should one choose? As usual, the choice has to be based
on local and global conditions. Local conditions are essentially related to the existence of
extended sources (with (d 3) spatial dimensions) at given points in transverse () space
manifold. Global conditions are essentially related to the choice of global transverse space.
231
Not all local conditions are possible for a given choice of transverse space. For instance,
there is no holomorphic function for a single (d 3)-brane in the Riemann sphere. 15
To clarify these issues, let us consider the simplest solution in this class: let us couple the
action Eq. (A.1) to a charged (d 3)-brane source. We first have to dualize the pseudoscalar
a into a (d 2)-form potential A(d2) with field strength F(d1) = (d 1)A(d2): a =
e2 F(d1) . The bulk plus brane action is

1
1
1
d
2
2
S=
()
F
x
|g|
R
+
+
d
2
2 (d 1)! (d1)
16G(d)
N
2

T
d d2 | | e (d2) ij gij (d 4)
2
(A.8)
d d2 A(d2) i1 i(d2) ' i1 i(d2) ,
(d 2)!
where gij and A(d2) i1 i(d2) are the pullbacks through the embedding coordinates X ( )
of the metric and (d 2)-form potential. T is the tension (in principle, a positive number)
and = 1 gives the sign of the charge (which is evidently proportional to the tension).
The coupling to is the only one that allows for solutions of the form we want.
A solution is provided by
2
2
ds = dt 2 d y(d3)
H d x22 ,
e = H,
(A.9)
A(d2) ty 1 y (d3) = H 1 ,
2 = 0,
X
Y i = i,
where H satisfies the equation
(d)
2 H = 16GN T (2) (
x2 ),
(A.10)
i.e., it is a harmonic function with a pole at x2 = 0, where the brane is placed. The above
equation is solved by a function H that behaves near x2 = 0
(d)
H 8GN T log |
x2 |.
(A.11)
It is clear that this solution cannot be globally correct as H becomes negative for |
x2 | > 1,
but the local behavior of the global solution has to be the same. Any solution behaving in
this way at any given point will describe a (d 3)-brane placed there.
Let us now compute the charge. This is defined by

p = e2 F(d1) = da,
(A.12)
where is a closed loop around the origin. a is given by

n a = 'nm m H,
(A.13)
15 Of course, one meets the same situation for other branes. However, for smaller branes one can always
find harmonic functions with a single pole (describing a single brane) that lead to spaces asymptotically flat
in transverse directions. This is not true for higher ((d 3)- and (d 2)-branes).
232
i.e., combining x 1 + ix 2

= 0, = +1,
(A.14)
= 0, = 1,
that is: a is the real part of a holomorphic or antiholomorphic function of , whose
1
imaginary part is the above function H . We find a = 8G(d)
(d) T .
N T Arg() and p =
16GN
1
(d) T
16GN
The choice = +1 then, corresponds to a single (d 3)-brane with charge p = +
placed at the origin and corresponds to a holomorphic function = H() that close to the
origin is given by
(d)
H 8GN T i log .
(A.15)
Observe that the charge is given by the multivaluedness of around the source, which
(d)
goes from to + 16GN T which should be identified with . The charge is usually
quantized due to quantum-mechanical reasons in multiples of the unit of charge (e, say)
which implies the identification + ne and the breaking of SL(2, R). If e = 1
(i.e., 16G(d)
N T = 1 which we can always get by rescaling ) then SL(2, Z) is the unbroken
symmetry of the theory and the above (d 3)-branes are associated to the modular group
element T . 16
We see that in this context solutions (and charges) can be characterized by the non-trivial
monodromies around singular points which, by hypothesis, are elements of the modular
group.
We can clearly generate via modular (duality) transformations of this solution with
T monodromy other solutions with different monodromies. it is easy to see that if we
perform a transformation M( ) M PSL(2Z) on the above solution, the monodromy
of the new solution around the origin will be MT M 1 . The most interesting modular
transformation is S( ) = 1/ which in other contexts relates electric and magnetic
(S dual) objects. Then, the S dual of the above solution will have monodromy STS around
2i
the origin and will be given either by H = log
using the general solution in the form
1
log and the form (2.6) of the solution. This is the form
of Eq. (A.2) or with H = 2i
we have used in the main text to stress that we are dealing with a solution different from
the one with monodromy T , the difference being in the choice of holomorphic function
since, as we have stressed at the beginning of this appendix all homomorphic solutions can
always be written in the form (A.2), no matter if the monodromy is T or STS.
Appendix B. The KK origin of the SL(2, R)/SO(2) model

We are going to see how the modular group PSL(2, Z) = SL(2, Z)/{I22 } and the
SL(2, Z)/SO(2) sigma model arise in standard KaluzaKlein compactification on a 2-torus
T 2.
16 For 10-dimensional type IIB D7-branes 16 G(10) = (2 )7 $8 g 2 and T = (2 )7 $8 g, and, thus, H
s
s
N
(8)
gi
log . On the other hand, C 1 7 = g 1 H 1 ( = +1) and we get p = 1 in a most natural way.
2
ty y
233
B.1. The modular group

As usual in KK compactifications, we use two periodic coordinates x m , m = 1, 2, whose
periodicity is fixed to 2$ where $ is some fundamental length. This means that we make
the identifications
1
x
, n Z2 .
x x + 2$
n,
x =
(B.1)
x2
The information on relative sizes and angles of the periods and the size of the torus is
codified in the internal metric Gmn ,
2
dsint
= d x T G d x ,
(B.2)
which is, by hypothesis, independent on the torus coordinates x (but may depend on the
remaining coordinates).
The KK ansatz is invariant under global diffeomorphisms in the internal manifold. These
are, generically, of the form
x = R 1 x + a ,
R GL(2, R)
a R2 .
(B.3)
a simply shifts the coordinate origin and does not affect the metric. R acts on the internal
metric according to

G = R T GR Gmn = R p m Gpq R q n .
(B.4)
We want to separate the volume part of the metric from the rest. 17 Thus, we define 18
K |detGmn |,
Gmn K 1/2 Mmn .
(B.5)
M has determinant +1 and, therefore, it is a symmetric SL(2, R) matrix and, in fact, it

can be understood as an element of the coset SL(2, R)/SO(2) with only two independent
entries. If we factor out the determinant of the GL(2, R) transformations too,

s = sign detR m n ,
R m n sR 1/2 S m n ,
R detR m n ,
(B.6)
then the volume element K and the matrix M transform according to
M = S T MS,
K = RK.
(B.7)
|K| is an element of the multiplicative group R+ and S is an element of SL(2, R). This
decomposition reflects the decomposition GL(2, R) = SL(2, R) R+ Z2 . s does not act
neither on K nor on M.
We have not yet taken into account the periodic boundary conditions of the coordinates,
that have to be preserved by the diffeomorphisms in the KK setting. Clearly the rescalings
R do not respect the torus boundary conditions, but they rescale $. The rotations S respect
17 This is necessary, for instance, when we are interested in conformal classes of equivalence of metrics, as in
string path integrals, but convenient in general.

18 Remember that G has signature ().
234
the boundary conditions only if S 1 n Z2 the matrix entries are integer, i.e., S SL(2, Z).
Up to a reflection S = I22 , these diffeomorphisms are known as Dehn twists and are
not connected with the identity (in fact, they constitute the mapping class group of torus
diffeomorphisms) and they constitute the modular group PSL(2, Z) = SL(2, Z)/{I22 }.
This is the group that acts on M.
We are going to write the modular group matrices in the slightly unconventional form

S=
,
(B.8)

to get the conventional form of the transformation of the modular parameter Eq. (B.15).
B.2. The modular parameter
We can define a complex modular-invariant coordinate on T 2 by
1 T
x ,
(B.9)
= C2 ,
2$
where, under modular transformations, we assume that the complex vector
transforms
according to
=
= S T .
(B.10)
The periodicity of is
+
T n ,
n Z2 .
(B.11)
What we have done is to transfer the information contained in the metric (more precisely,
in M) into the complex periods .
The relation between these two is

2
e(1 2 )
|1 |
1
.
M=
(B.12)
m(1 2 ) e(1 2 )
|2 |2
We can check that the transformation rules for the complex periods Eq. (B.10) and for
the matrix M Eq. (B.7) are perfectly compatible.
In terms of the modular-invariant complex coordinate, the torus metric element takes the
form
1
2
= K 1/2
d d .
dsint
(B.13)
m 1 2
Observe that m(1 2 ) is modular-invariant (and a quite important one).
It should be clear that not all pairs of complex periods characterize different tori. Recall
that M only has 2 independent entries while
contains 4 real independent quantities. In
particular, we can see that multiplying
by any complex number leaves the matrix M
invariant. It is customary to multiply by 21 both the coordinate and define
= /2 ,
= 1 /2 ,
(B.14)
that can always be chosen to belong to the upper half complex plane H m( ) 0 (1
defines the same torus as 1 ).
235
Under a modular transformation with S given by Eq. (B.8), the modular parameter
undergoes a fractional-linear transformation
=
+
+
(B.15)
and the torus coordinate transforms

=
.
(c + d)
Finally, in terms of , the matrix M reads

| |2 e( )
1
M=
.
m( ) e( )
1
(B.16)
(B.17)
B.3. The SL(2, R)/SO(2) sigma model

In pure KK theory (with no higher-dimensional fields apart from the metric), the toroidal
compactification of the EinsteinHilbert action from d to d dimensions with the KK ansatz
a

em i A m
e
(e a ) =
(B.18)
0
em i
where the internal metric
Gmn = em i enj = em i en j ij .
(B.19)
gives, upon the rescaling

2
gE = K (d2) g ,
S=
d d x
d x
(B.20)

|g|
R

2
(d 2)(d d)
1
( log K)2 + Tr MM1
|gE | RE +
4(d 2)
4

d2)
1 ((d2)
K
Mmn F m F n .
4
(B.21)
The kinetic term for the scalar matrix M is manifestly invariant under SL(2, R)
transformations (the action we started from is diffeomorphism-invariant). Using the
parametrization Eq. (B.17), it takes the standard form
1
.
2 (m( ))2
(B.22)
236
References
[1] C.M. Hull, P.K. Townsend, Unity of superstring dualities, Nucl. Phys. B 438 (1995) 109137.
[2] E. Bergshoeff, C.M. Hull, T. Ortn, Duality in the type II superstring effective action, Nucl.
Phys. B 451 (1995) 547578.
[3] J. Polchinski, Dirichlet branes and RamondRamond charges, Phys. Rev. Lett. 75 (1995) 4724
4727.
[4] P. Meessen, T. Ortn, An Sl(2, Z) Multiplet of nine-dimensional type II supergravity theories,
Nucl. Phys. B 541 (1999) 195245.
[5] E. Bergshoeff, R. Kallosh, T. Ortn, Stationary axion/dilaton solutions and supersymmetry,
Nucl. Phys. B 478 (1996) 156180.
[6] E. Lozano-Tellechea, T. Ortn, The general, duality-invariant family of non-BPS black hole
solutions of N = 4, d = 4 supergravity, Nucl. Phys. B 569 (2000) 435450.
[7] C.M. Hull, Gravitational duality, branes and charges, Nucl. Phys. B 509 (1998) 216251.
[8] C.M. Hull, U-duality and BPS spectrum of super YangMills theory and M-theory, JHEP 9807
(1998) 018.
[9] M. Blau, M. OLoughlin, Aspects of U-duality in matrix theory, Nucl. Phys. B 525 (1998)
182214.
[10] N.A. Obers, B. Pioline, E. Rabinovici, M-theory and U-duality on T d with gauge backgrounds,
Nucl. Phys. B 525 (1998) 163181.
[11] R.D. Sorkin, KaluzaKlein monopole, Phys. Rev. Lett. 51 (1983) 87.
[12] D.J. Gross, M.J. Perry, Magnetic monopoles in KaluzaKlein theories, Nucl. Phys. B 226
(1983) 29.
[13] E. Bergshoeff, H.-J. Boonstra, T. Ortn, S duality and dyonic p-brane solutions in type II string
theory, Phys. Rev. D 53 (1996) 72067212.
[14] J.M. Izquierdo, N.D. Lambert, G. Papadopoulos, P.K. Townsend, Dyonic membranes, Nucl.
Phys. B 460 (1996) 560578.
[15] N. Alonso-Alberca, P. Meessen, T. Ortn, An SL(3, Z) multiplet of 8-dimensional type II
supergravity theories and the gauged supergravity inside, Report IFT-UAM/CSIC-00-38, KULTF-2000/28, hep-th/0012032.
[16] P.K. Townsend, M-theory from its superalgebra, Lectures delivered at Cargse, 1997, hepth/9712004.
[17] E. Bergshoeff, B. Janssen, T. Ortn, KaluzaKlein monopoles and gauged sigma-models, Phys.
Lett. B 410 (1997) 131141.
[18] E. Bergshoeff, J.P. van der Schaar, On M-9-branes, Class. Quant. Grav. 16 (1999) 23.
[19] E. Eyras, Y. Lozano, Exotic branes and nonperturbative seven-branes, Nucl. Phys. B 573 (2000)
735767.
[20] E. Bergshoeff, E. Eyras, R. Halbersma, J.P. van der Schaar, C.M. Hull, Y. Lozano, Spacetime
filling branes and strings with sixteen supercharges, Nucl. Phys. B 564 (2000) 2959.
[21] B.R. Greene, A. Shapere, C. Vafa, S.-T. Yau, Stringy cosmic strings and noncompact Calabi
Yau manifolds, Nucl. Phys. B 337 (1990) 1.
[22] G.W. Gibbons, M.B. Green, M.J. Perry, Instantons and seven-branes in type IIB superstring
theory, Phys. Lett. B 370 (1996) 3744.
[23] R. Kallosh, T. Ortn, Charge quantization of axiondilaton black holes, Phys. Rev. D 48 (1993)
742747.
[24] S. Ferrara, R. Kallosh, A. Strominger, N = 2 extremal black holes, Phys. Rev. D 52 (1995)
54125416.

D-branes in B fields
Jian-Ge Zhou
Institut fr Theoretische Physik, Technische Universitt Wien, Wiedner Hauptstrasse 8-10/136,
A-1040 Wien, Austria
Abstract
The RR Page charges for the D2-, D4-, D6-brane in B fields are constructed explicitly from the
equations of motion and the nonvanishing (modified) Bianchi identities by exploiting their properties
conserved and localized. It is found that the RR Page charges are independent of the background
B fields, which provides further evidence that the RR Page charge should be quantized. In our
construction, it is highly nontrivial that the terms like B B B, B B F , B F F from
different sources are exactly cancelled with each other. 2001 Elsevier Science B.V. All rights
reserved.
PACS: 11.25.-w; 11.27.+d; 11.40.-q
1. Introduction
In IIA string theory, a D2p-brane is coupled to the RR gauge fields [1], the coupling of
the RR gauge fields to the pullback of the spacetime
B field and U(1) gauge field strength F

on D2p-brane seems to demand the integral (B + F ) to be quantized. However, in [2],
Bachas, Douglas and Schweigert found that in WZW model the RR D0-brane
charge on

the stable
spherical D2-brane takes irrational value due to the integral B, and suggested

only F can be interpreted as properly defined RR D0-brane charge.
Up to now, three different approaches have been proposed to resolve this puzzle. In [3],
Taylor argued that the D0-brane charge arising from the integral over the D2-brane of the
pullback of the B field is cancelled by the bulk contributions, but in his calculation it was
implicitly assumed that the gauge field C (1) is constant. Also his analysis is hard to be
generalized to D2p-branes (p > 1). 1 In [4], Alekseev, Mironov and Morozov observed
that if one rewrites the WZ action in terms of the gauge-invariant field strength G(2p+2) =
dC (2p+1) C (2p1) H , the resulting RR brane charges are independent of B fields.
E-mail address: jgzhou@hep.itp.tuwien.ac.at (J.-G. Zhou).
1 Here we mean that when p = 2, 3, we have D4-, D6-brane.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 2 4 0 - 1
238
J.-G. Zhou / Nuclear Physics B 607 (2001) 237246
However, in their approach when the RR gauge fields C (2p+1) are not constant they have to
deal with some nonlocal transformation, and treat RR gauge fields C (2p+1) as independent
fields even though in IIA string theory the RR gauge field C (2p+1) and C (72p) are dual
to each other. In [5], Marolf suggested that due to the presence of ChernSimons term
in IIA supergravity, there are three notions of charge: (1) brane source charge is gaugeinvariant and localized, but not conserved; (2) Maxwell charge is conserved and gaugeinvariant, but not localized; (3) Page charge is conserved, localized and invariant under
small gauge transformation. Since brane source charge and Page charge were defined
separately in [5], it is unclear how to relate one to other which is crucial to the questions
whether it is possible for us to consistently define the general RR Page charges for a D2pbrane from brane source charges and whether Page charges are independent of B fields or
not. 2
In the present paper, we study how to consistently construct B-independent RR charges
of D2p-branes with B fields in our new frame. Since D-branes are sources of RR gauge
fields, we consider IIA supergravity plus D-brane sources, from which we derive the
equations of motion and the nonvanishing (modified) Bianchi identities that define the
duals of brane source currents for D2p-branes. Exploiting the relation dAp Bq = d(Ap
Bq ) (1)p Ap dBq , we insert the equations for the duals of brane source currents for
D(2p + 2n)-branes (n > 1) into that for the D2p-brane iteratively. The new feature is that
the resulting equations can be recast into the form whose left sides of equations are exterior
derivative and the right sides are localized objects, thus the right side localized objects can
be identified as Page charges because of their conservation and locality. The results also
show that the Page charges for D2p-branes can be consistently defined from the brane
source charges. Inserting the brane source charges into the expression of the Page charges,
we find that all the Page charges are independent of the background B fields. In our explicit
construction, it is highly nontrivial that the B-dependent terms like B B B, B B F ,
B F F from different sources are exactly cancelled with each other.
The layout of the paper is as follows. In Section 2, we consider a D2-brane in Bfields in order to illustrate our method. The brane source currents for D2-, D0-branes are
derived from the equations of motion. The explicit expression for RR Page charges is
obtained by exploiting their properties conserved and localized, the result is consistent
with those in [3] and [4]. In Section 3, we study a D4-brane in B fields, which is bound
state of D4/D2/D0-branes. With additional D4-brane source defined by the nonvanishing
of modified Bianchi identity, the RR Page charges for D2-, D0-branes living on the D4brane are constructed by combining the equations of motion and the nonvanishing of the
modified Bianchi identity. Late we discuss a D6-brane in B fields, where we find that it is
highly nontrivial that the RR Page charges are independent of B fields. In Section 4, we
present our summary and discussion.
2 In general, the definition of RR Page charge should be compatible with T-duality, and quantization of the
Page charge for a D2p-brane follows from T-duality. T-duality implies Page charge quantization in the systems
with sufficient translational symmetry [5].
239
2. D2-brane in B fields
The IIA supergravity in ten dimensions can be obtained from N = 1, D = 11
supergravity by dimensional reduction, the action for IIA supergravity is
1
SIIA = 2
d 10 x G
210

2 (4) 2
1
1

e2 R + 4 |H |2 F (2) + F
2
2

1
2
(1)
B F (4) F (4) ,
410
2 is the gravitational coupling in ten dimensions, and the field strengths
where 10
(2)
(4) are defined as
H, F , F (4) , and F
H = dB,
F (2) = dC (1) ,
(4) = dC (3) C (1) H.
F
F (4) = dC (3) ,
The above action can be rewritten as

SIIA = d 10 x G e2 R + 4

1
+
e2 dB dB + dC (1) dC (1)
2

+ dC (3) C (1) dB dC (3) C (1) dB

B dC (3) dC (3) ,
(2)
(3)
2 = 1, that is, the kinetic term is canonical.

where for simplicity we have chosen 210
Consider a D2p-brane in the background of B fields, the D2p-brane world-volume
action is
SD-brane = SBI + SWZ .

The BornInfeld part of the action is

SBI = T2p e det(Gab + Bab + Fab ),
(4)
(5)
where Gab , , Bab are the pullback of spacetime metric, dilaton and NeveuSchwarz two
form B to the D2p-brane world-volume, Fab is the field strength of U(1) gauge field living
on the brane.
The WessZumino (WZ) terms couple the brane to the spacetime RR gauge field through

SWZ =
(6)
C (i) eB+F
i
and in the D2-brane case, it is reduced to

(3)

D2
=
C + C (1) (B + F ) ,
SWZ
which means there are D0-branes living on D2-brane.
(7)
240
Since D-branes are sources of RR gauge field, we consider IIA supergravity + the
coupled D2-brane, and see how D2-, D0-branes induce C (3) , C (1) gauge field, which can
be described by the equations of motion. As brane source charge arises from varying the
brane action with respect to the gauge field, from the total action ST = SIIA + SD2-brane we
get the equations of motion for C (1) and C (3) 3
bs
(4) = jD0
d F (2) H F
,
(4)
(4)
bs
H F
=j ,
d F
D2
(8)
where we have replaced the second term H F (4) in the left side of the second equation
(4) due to the fact H H = 0. The brane source currents for D2-, D0-branes are
by H F
given by

bs

jD2
(9)
(x) = dX ( ) dX ( ) dX ( ) 10 x X( ) ,

bs
jD0
(x) =
1
2

d 3 $ abc a X ( ) B X( ) b X ( )c X ( ) + Fab ( )

10 x X( ) ,
(10)
where x are ten-dimensional coordinates, X ( ) are the embedding coordinates of D2brane in ten dimensions, and a are three-dimensional brane world-volume coordinates.
The brane source charge for D0-brane is

bs
Qbs
(11)
=
j
=
(B + F ),
D0
D0
V2
where V2 is two-dimensional space volume of D2-brane.

Since we only consider D2-, D0-brane without other brane sources, we have dF (2) = 0,
bs
bs
= 0, d jD0
= 0,
dF (4) = 0, dH = 0, H H = 0. Under these conditions, we see d jD2
bs
bs
that is, QD2 is conserved while QD0 nonconserved, however, when H = dB = 0, Qbs
D0 is
still conserved.
(4) )
(4) in the first equation in (8) as d(B F
By rewriting the second term H F
(4) , we arrive at
(4) and then inserting the second equation of (8) into B d F
BdF
(4) = j bs B j bs ,
(4) 1 B B F
d F (2) + B F
(12)
D0
D2
2
which shows that we can indeed recast two equations in (8) into the new equation whose
left side is written as exterior derivative and the right side as localized object. In the above
derivation, we have assumed that the D2-brane stays in the region where the relation H =
dB holds. According to the properties of the RR Page current, Eq. (12) indicates that we
can define consistently
Page
bs
bs
B jD2
,
jD0 = jD0
(13)
3 Since we choose different sign for the WZ term, there is minus sign difference between our (8) and that in [5],
similarly for the definition of the nonvanishing Bianchi identities.
241
Page
and jD0 is indeed conserved and localized. The RR Page charge takes

Page
Page
QD0 = jD0 = F,
(14)
V2
which is consistent with [3] and [4]. In [2] and [5],

it was argued that the RR Page charge
for D0-branes in spherical D2-brane defined by S2 F should be quantized. In the context
of WZW model, it has been shown that the integral S2 F is quantized indeed [6]. Other
related discussion on this topic can also be found in [79].
If we do T-duality along other extra four dimensions, the D2/D0 bound state turns into
D6/D4 one, and Eq. (13) becomes
Page
bs
bs
B jD6
,
jD4 = jD4
(15)
which means we define the RR Page current for D4-branes living on the D6-brane in the
way of compatible with T-duality. 4
3. D4-, D6-brane in B fields

At first, we consider a D4-brane in B fields, which is D4/D2/D0 bound state. The
WZ term for D4-brane in the background of B fields is

1
D4
=
C (5) + C (3) (B + F ) + C (1) (B + F ) (B + F ) .
SWZ
(16)
2
Besides the equations of motion (8), there is the nonvanishing of modified Bianchi identity
which describes the hodge dual of the D4-brane source current [5]
(4) + F (2) H = j bs .
dF
D4
(17)
If we rewrite IIA supergravity in terms of the field C (5) dual to C (3) , the modified Bianchi
identity for C (3) becomes an equation of motion for C (5) , and the brane source current can
be derived by variation of the WZ term for D4-brane with respect to C (5) . From Eq. (16),
we know the brane source charges for D2-, D0-branes in the D4-brane are

bs
bs
QD2 = jD2 = (B + F ),

1
bs
(B + F ) (B + F ).
Qbs
(18)
=
j
=
D0
D0
2
In the presence of D4-brane, dF (4) is no longer equal to zero, then Qbs
D2 is nonconserved.
bs
(2)
Without D6-brane, we have dF = 0, so QD4 is still conserved. Combining Eq. (8) and
Eq. (17), we have
(4) + 1 B B F (2) = j bs B j bs ,
(4) + B F
d F
(19)
D2
D4
2
4 Here we should mention that (15) is obtained just by the use of T-duality as suggested by Marolf.
242
(4) 1 B B B F (2)
(4) 1 B B F
d F (2) + B F
2
6
1
bs
bs
bs
B jD2
+ B B jD4
,
= jD0
(20)
2
where we use the similar way as in the derivation of Eq. (12) to get Eq. (19). To derive (20),
(4) )
(4) in the first equation in (8) as d(B F
we first rewrite the second term H F
(4)
(4)

and insert the second equation of (8) into B d F , then besides the
BdF
(4) which can rewrite
exterior derivative and localized terms we have the term B H F
1
1
(4)
(4)
) B B dF
. Inserting the nonvanishing modified Bianchi
as d( 2 B B F
2
1
(4) and exploiting dF (2) = 0, we get Eq. (20).
identity (17) into the term 2 B B d F
Eqs. (19) and (20) show that we can consistently define the hodge duals of Page currents
Page
bs
bs
jD2 = jD2
B jD4
,
1
Page
bs
bs
bs
B jD2
+ B B jD4
,
jD0 = jD0
2
Page
(21)
Page
from the brane source currents, the resulting jD2 and jD0 are conserved and localized
on D4-brane, and the RR Page charges are

Page
Page
QD2 = jD2 = F,

1
1
Page
Page
(B + F ) (B + F ) B (B + F ) + B B
QD0 = jD0 =
2
2

1
=
(22)
F F,
2
where we have seen the B-dependent terms B B, B F have been cancelled with each
other.
Now we turn to a D6-brane in B fields, the WZ terms for D6-brane is

1
D6
=
C (7) + C (5) (B + F ) + C (3) (B + F ) (B + F )
SWZ
2
1
+ C (1) (B + F ) (B + F ) (B + F ) ,
(23)
6
which describes D6/D4/D2/D0 bound state. In the presence of D6-brane, besides
the equations of motion (8) and the modified Bianchi identity (17), we have another
nonvanishing Bianchi identity 5
bs
dF (2) = jD6
.
(24)
5 In IIA supergravity, the RR gauge field C (2p+1) is dual to C (72p) . Because of A

2p = A2p in
10D Minkowski spacetime, there are two ways to performing the electromagnetic duality. For example, consider
D6-brane RR gauge field C (7) (for illustration, we assume B fields vanish), we can define either dC (7) = dC (1) ,
i.e., dC (1) = dC (7) , or dC (1) = dC (7) , i.e., dC (7) = dC (1) . In case 1, the D6-brane source current is
bs , and in case 2, it is dF (2) = j bs . The WZ term does not contain any information
described by dF (2) = jD6
D6
about which definition one should use [10]. In order to make the definition for RR Page current compatible with
bs
(2)
T-duality, we have to choose dF = jD6 .
Combining two nonvanishing Bianchi identities (17) and (24), we have

(4)
+ B F (2) = j bs B j bs ,
d F
D4
D6
243
(25)
which shows that we can define the hodge dual of RR Page current for D4-branes living in
D6-brane as 6
Page
bs
bs
jD4 = jD4
B jD6
,
(26)
which is conserved and localized. Here we should point out that Eq. (26) is derived from
the nonvanishing modified Bianchi identity (17) for D4-brane and the other nonvanishing
Bianchi identity (24) for D6-brane. On the other hand, Eq. (15) is obtained from T-duality,
which means the ambiguity in the definition of C (7) can be fixed by T-duality.
In [5], the hodge dual of RR Page current for D4-brane was defined as 7
(4)

+ C (1) H = j Page .
d F
(27)
D4
Comparing the Eq. (25) with Eq. (27), we find there is a difference in the definition of the
RR Page current, which can be interpreted as gauge dependence of the RR Page charge
associated with choosing a surface in N = 1 eleven-dimensional supergravity that projects
onto the chosen surface in IIA supergravity.
By making use of the equations of motion and two nonvanishing (modified) Bianchi
identities, in the way similar to deriving Eqs. (19) and (20) we get the relations

(4) + 1 B B F (2)
(4) + B F
d F
2
1
bs
bs
bs
= jD2
(28)
B jD4
+ B B jD6
,
2

1
1
(2)
(4)
(4)
(2)

d F + B F B B F B B B F
2
6
1
1
bs
bs
bs
bs
B jD2
+ B B jD4
B B B jD6
.
= jD0
(29)
2
6
Eqs. (28) and (29) indicate that the hodge dual of the RR Page currents for D2-, D0-branes
can be defined as
1
Page
bs
bs
bs
B jD4
+ B B jD6
,
jD2 = jD2
2
1
1
Page
bs
bs
bs
bs
B jD2
+ B B jD4
B B B jD6
.
jD0 = jD0
2
6
(30)
(31)
6 If we choose dF (2) = j bs , then we would have j Page = j bs + B j bs , which is incompatible with

D4
D6
D4
D6
T-duality.
7 In [5], the brane source charge and Page charge were defined separately, they did not study how to define
Page charge based on the definition for brane source charge, and (27) was given just by definition, but here (26)
is derived from the consistent consideration.
244
Recall that for D6/D4/D2/D0 bound state, the brane sources can be read off from the
WZ term for D6-brane (23), the corresponding brane source charges are

bs
Qbs
=
j
=
(B + F ),
D4
D4

1
bs
Qbs
(B + F ) (B + F ),
=
j
=
D2
D2
2

1
bs
Qbs
(32)
jD0
=
(B + F ) (B + F ) (B + F ).
D0 =
6
Putting Eqs. (26), (30), (31) and (32) together, we obtain for D4-, D2-branes

Page
Page
QD4 = jD4 = F,
Page
QD2

1
bs
bs
bs
=
jD2 B jD4 + B B jD6
2

1
1
(B + F ) (B + F ) B (B + F ) + B B
=
2
2

1
=
F F,
2
and for D0-branes

1
1
Page
bs
bs
bs
bs
QD0 =
jD0
B jD2
+ B B jD4
B B B jD6
2
6

1
1
(B + F ) (B + F ) (B + F ) B (B + F ) (B + F )
=
6
2
1
1
+ B B (B + F ) B B B
2
6

1
=
F F F.
6
(33)
(34)
(35)
Eq. (35) shows that the terms B B B, B B F , B F F are cancelled with each
other in the RR Page charge of D0-branes living in D6-brane, which was expected in [35].
4. Summary and discussion

So far we have developed new approach to consistently construct B-independent
RR charges of D2p-branes in B fields. As D-branes are sources of RR gauge fields, we
have induced the equations of motion and the nonvanishing (modified) Bianchi identities
from total action for IIA supergravity plus D-brane sources, with which we have defined the
duals of brane source currents for D2p-branes. By making use of the relation dAp Bq =
d(Ap Bq ) (1)p Ap dBq , we have plugged the equations for the duals of brane
source currents for D(2p + 2n)-branes into that for the D2p-brane iteratively, and found
that the resulting equations can be recast into that the left sides of equations are exterior
245
derivative and the right sides are localized objects. This particular feature indicates that the
right side localized objects can be identified as Page charges because of their conservation
and locality, which shows that the Page charges for D2p-branes can be consistently defined
from the brane source charges. Inserting the brane source charges into the expression of the
Page charges, we have shown that all the Page charges are independent of the background
B fields which provides further evidence that the RR Page charge should be quantized. In
our explicit construction, it is highly nontrivial that the B-dependent terms like B B B,
B B F , B F F from different sources are exactly cancelled with each other. Thus
we conclude that the properly defined RR charges of D-branes should be B-independent,
and we can identify them as Page charges which are conserved and localized. 8
The IIA supergravity only contains 2-form F (2) and 4-form F (4) field strength, until
now we only considered D2-, D4-, D6-brane in B fields. It would be interesting to study
D8-brane in B fields and construct the corresponding RR Page charges for D0-, D2-, D4-,
D6-branes which couple to D8-brane via WZ term. In this case, we should turn to massive
type IIA supergravity. In the above construction, we have assumed H = dB, however,
when it does not hold in whole region, for instance, in the near-horizon region of the NS5brane background where the geometry is R 1,5 R S 3 , we cover S 3 with two open
sets U = {u }, the RR Page charge defined on each open set with B is conserved. It
is interesting to see how two RR Page charges defined in each open set are connected.
It is probably related to brane creation phenomenom [9]. We hope to discuss the above
questions in near future.
Acknowledgement
I would like to thank B. Harms, M. Kreuzer, D. Marolf, H.J. Schnitzer and P.K. Townsend
for valuable discussion. This work is supported by the Austrian Research Funds FWF under grant No. M597-TPH.
References
[1] J. Polchinski, TASI lectures on D-branes, hep-th/9611050.
[2] C. Bachas, M. Douglas, C. Schweigert, Flux stabilization of D-branes, J. High Energy Phys. 05
(2000) 048, hep-th/0003037.
[3] W. Taylor, D2-branes in B fields, hep-th/0004141.
[4] A. Alekseev, A. Mironov, A. Morozov, On B-independence of RR charges, hep-th/0005244.
[5] D. Marolf, ChernSimons terms and three notions of charge, hep-th/0006117.
[6] A. Kling, M. Kreuzer, J.-G. Zhou, SU(2) WZW D-branes and quantized worldvolume U(1) flux
on S 2 , hep-th/0005148.
[7] C. Stanciu, A note on D-branes in group manifolds: flux quantization and D0-charge, hepth/0006145.
8 When B fields are constant, all the brane source charges are conserved, so in that case there is no necessity
to construct RR Page charge.
246
[8] J.M. Figueroa-QFarrill, S. Stanciu, D-brane charge, flux quantization and relative
(co)homology, hep-th/0008038.
[9] T. Kubota, J.-G. Zhou, RR charges of D2-branes in group manifold and HananyWitten effect,
J. High Energy Phys. 12 (2000) 030, hep-th/0010170.
[10] M. Li, Boundary states of D-branes and Dy-strings, Nucl. Phys. B 460 (1996) 351, hepth/9510161;
M. Douglas, Branes within branes, in: Cargese 97: String, Branes and Dualities, p. 267, hepth/9512077.

Causality and CPT violation from an Abelian

ChernSimons-like term
C. Adam, F.R. Klinkhamer
Institut fr Theoretische Physik, Universitt Karlsruhe, 76128 Karlsruhe, Germany
Received 5 January 2001; accepted 4 April 2001
Abstract
We study a class of generalized Abelian gauge field theories where CPT symmetry is violated by a
ChernSimons-like term which selects a preferred direction in spacetime. Such ChernSimons-like
terms may either emerge as part of the low-energy effective action of a more fundamental theory
or be produced by chiral anomalies over a nonsimply connected spacetime manifold. Specifically,
we investigate the issues of unitarity and causality. We find that the behaviour of these gauge field
theories depends on whether the preferred direction is spacelike or timelike. For a purely spacelike
preferred direction, a well-behaved Feynman propagator exists and microcausality holds, which
indicates the possibility of a consistent quantization of the theory. For timelike preferred directions,
unitarity or causality is violated and a consistent quantization does not seem to be possible. 2001
Elsevier Science B.V. All rights reserved.
PACS: 11.15.-q; 11.15.Bt; 11.30.Cp; 11.30.Er
Keywords: Gauge invariance; Causality; Lorentz noninvariance; CPT violation
1. Introduction
Lorentz and CPT invariance are two of the cornerstones of modern quantum field
theory. Both invariances are respected by the Standard Model of known elementary
particles (quarks and leptons) and their interactions. Possible signals of Lorentz and
CPT violation could, therefore, be indicative of new physics, e.g., quantum gravity [1,2]
or superstrings [3]. But even within local quantum field theory an anomalous breaking
of Lorentz and CPT symmetry might occur, at least for a nontrivial global spacetime
structure [4,5].
Consequently, a considerable amount of attention has been devoted over the last
years to the possible occurrence of Lorentz and CPT noninvariance. Phenomenological
E-mail addresses: adam@particle.uni-karlsruhe.de (C. Adam), frans.klinkhamer@physik.uni-karlsruhe.de
(F.R. Klinkhamer).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 6 1 - 4
248
C. Adam, F.R. Klinkhamer / Nuclear Physics B 607 (2001) 247267
consequences of breaking Lorentz and CPT symmetry in electromagnetism were studied

in Ref. [6]. It was shown that the symmetry breaking would result in, for example, optical
activity (birefringence) of the vacuum, that is, a direction-dependent rotation of the linear
polarization of an electromagnetic plane wave. Reference [7], in turn, investigated CPTand Lorentz-noninvariant extensions of the Standard Model (interpreted as low-energy
limits of more fundamental theories). Furthermore, there have been extensive discussions
in the literature on the possibility of CPT- and Lorentz-symmetry breaking in the gauge
field sector induced by radiative corrections of an explicitly symmetry-breaking matter
sector, see Refs. [710] and references therein.
At this point, the question arises whether or not a quantum field theory with Lorentzand CPT-violating terms can be consistent at all, cf. Refs. [1113]. Also, in each of the
papers quoted in the previous paragraph, the CPT- and Lorentz-noninvariant terms in the
gauge field sector were of the ChernSimons type [14]. In this paper, therefore, we intend
to study possible implications of a ChernSimons-like term for the quantization of Abelian
gauge fields, focusing on the issues of unitarity and causality.
We start from the following Lagrangian density:

2
1
1
L(x) = F (x)F (x) 1 n A (x) + LCS-like (x),
4
2
with the ChernSimons-like term
(1.1)
1
LCS-like (x) = mk A (x)F (x),
(1.2)
4
in terms of the Abelian gauge potential A (x) and field strength tensor F (x)
A (x) A (x). The spacetime metric is taken to have Lorentzian signature
(, +, +, +) and is the completely antisymmetric Levi-Civita symbol, normalized
to 0123 = +1. (Our conventions, with h = c = 1, will be given in more detail later on.)
The Abelian ChernSimons-like term (1.2) is characterized by a real mass parameter
m and a real symmetry-breaking vector k of unit length, which may be spacelike
(k 2 = +1) or timelike (k 2 = 1) but is fixed once and for all (hence, the quotation marks
around the word vector). Strictly speaking, k can also be lightlike (k 2 = 0), but the
present paper considers only the extreme cases, spacelike or timelike k . As long as k and
m = 0 are fixed external parameters (coupling constants), both Lorentz and CPT invariance
are broken, but translation invariance still holds. Note that the Lagrangian term (1.2) is
called ChernSimons-like, because a genuine topological ChernSimons term exists only
in an odd number of dimensions [14].
For later convenience, we have added a gauge-fixing term to the Lagrangian (1.1), where
n determines the axial gauge condition and is a gauge parameter. Choosing an axial
gauge, which selects a particular direction n , seems natural because the vector k
already selects a preferred direction. In other words, there is no compelling reason to prefer
Lorentz-covariant gauge choices over noncovariant ones, cf. Ref. [15].
The Lagrangian (1.1) is Abelian and, therefore, describes a photon-like gauge field. But
Eq. (1.1) may as well be interpreted as one component of the quadratic part of a nonAbelian Lagrangian. The discussion that follows is, in principle, also relevant for Lorentz-
249
and CPT-symmetry breaking in a non-Abelian context. Still, the issue of locality may be
more subtle for the non-Abelian case due to gauge invariance, as discussed in Section 4 of
Ref. [4].
Let us now give in more detail the reasons for studying the MaxwellChernSimons theory (1.1), with broken Lorentz and CPT symmetry. First, a nonzero mass scale
m may be introduced by hand as a symmetry-breaking parameter. Possible physical
consequences and experimental bounds on the value of m may be studied, as was done in
Ref. [6], under the assumption that the Lagrangian (1.1) describes the photon. Recently, it
has also been claimed [16] that certain astronomical observations indicate a nonzero value
of m for the case of a spacelike ChernSimons parameter k , but this claim is apparently
not substantiated by more accurate data (see Ref. [17] and references therein).
Second, the symmetry-breaking term in the Lagrangian (1.1) may be thought of as being
part of the effective action which results from integrating out the fermionic matter fields.
Here, the source of the symmetry breaking might be an explicit symmetry-breaking term
in the fermionic matter sector [710]. Alternatively, the symmetry-breaking term in the
effective action might be traced to a quantum anomaly which occurs when Weyl fermions
in suitable representations are quantized on a nonsimply connected spacetime manifold
(e.g., R3 S 1 ). This CPT anomaly was discovered and described in Ref. [4], where the
precise conditions for its occurrence can be found. In this case, the experimentally required
smallness of m for photons is naturally accounted for, because the mass scale m is inversely
proportional to the linear extension (L) of the universe in the compact direction,
1
mCPT anomaly h(Lc)
,
(1.3)
with the fine-structure constant and the dependence on h and c made explicit. For
L 1.5 1010 lightyears, this mass scale corresponds to 1035 eV, which might be within
reach of future astronomical observations (this point will be discussed further in Section 6).
Third, the Lagrangian (1.1) may be interpreted as the quadratic gauge field part of a lowenergy effective action of a truly fundamental theory, which could, for example, replace
point-particles by superstrings, cf. Ref. [3].
Our paper is organized as follows. In Section 2, we focus on the classical aspects of
the MaxwellChernSimons theory (1.1) and discuss the resulting dispersion relations and
causality behaviour. This turns out to be rather different for spacelike and timelike Chern
Simons parameter k . In Section 3, the Feynman propagator for the Lagrangian (1.1) is
calculated both for Minkowskian and Euclidean spacetime. Again, the cases of spacelike
and timelike k have to be discussed separately. In Section 4, we address the related issue
of reflection positivity for the Euclidean theory corresponding to Eq. (1.1), which also
depends on the type of parameter k . In Section 5, we determine the field commutators of
the quantum field theory based on Eq. (1.1), first for a purely spacelike k . We find that
the usual microcausality holds for this case, which is perhaps the most important result
of this paper. (Some details of our calculation are relegated to Appendix A.) On the other
hand, unitarity and microcausality cannot be maintained simultaneously for a timelike k .
In Section 6, finally, we summarize our results and briefly discuss possible applications and
250
open questions. The present paper is, by necessity, quite technical and the general reader
may wish to concentrate on Sections 2 and 6.
2. Dispersion relations
As a first step we discuss the dispersion relations which result from the Lagrangian (1.1)
without the gauge fixing term and investigate the implications for the causal behaviour of
the classical theory. Throughout this section, we take the spacetime manifold M = R4
and Minkowskian spacetime metric g = diag(1, 1, 1, 1), with indices running over
0, 1, 2, 3.
The Lagrangian (1.1) then leads to the following dispersion relation for the gauge
fields [6]:

p4 + m2 k 2 p2 (k p)2 = 0,
(2.1)
for momentum p = (p0 , p1 , p2 , p3 ) and c = 1. 1 Due to the breaking of Lorentz
invariance, there exist preferred coordinate systems. A particular preferred coordinate
system for spacelike ChernSimons parameter k is one in which k is purely spacelike
(k0 = 0), which we shall choose in the sequel.
Let us discuss this last point in somewhat more detail (see also Ref. [7]). As mentioned
in the Introduction, the ChernSimons parameters k are considered to be fixed coupling
constants (four real numbers) belonging to a particular coordinate system. For localized
gauge fields (that is, A (x) = 0 for |x| R), one can nevertheless make a Lorentz
transformation x x = x + a , so that the ChernSimons-like term (1.2)
changes into
1
1
mk A (x )F (x ) mk A (x )F (x ).
(2.2)
4
4
The new reference frame (with coordinates x ) thus has its own ChernSimons
parameters k , determined by the old k and the Lorentz parameters for the change of
frame. It is, however, not at all obvious that this change of k parameters is unitarily
implementable for the quantum theory. The quantization of the MaxwellChernSimons theory (1.1) is, therefore, considered rather explicitly in the following sections. For
now, we continue our discussion of the classical dispersion relation.
with |k|
2 = 1, Eq. (2.1) is
For a purely spacelike ChernSimons parameter k = (0, k)
2
a quadratic equation in p0 , with the following solutions:

1 2 1
2
2
2.
p0 = |p|
(2.3)
+ m m m2 + 4(p k)
2
2
Apparently, there are two very different degrees of freedom, especially towards the infrared
(|p|
m). (The identification of these two degrees of freedom with circular polarization
1 For the moment, the velocity parameter c is only used to define the Minkowski spacetime coordinates (x 0
ct, x 1 , x 2 , x 3 ). As will become clear later, c corresponds to the front velocity of light propagation in vacuo for
the electromagnetic theory based on the Lagrangian (1.1) with a spacelike parameter k .
251
Fig. 1. The dispersion relation (2.3) in the (p0 , p3 ) plane for a purely spacelike ChernSimons parameter k = (0, 0, 0, 1) and mass scale m, with broken (solid) curves corresponding to the plus
(minus) sign in Eq. (2.3).
cf. Eq. (26) of Ref. [6].) For directions p perpenmodes depends on the sign of p k,
Eq. (2.3) effectively describes one massive degree of freedom with mass
dicular to k,
m, corresponding to the plus sign, and one massless degree of freedom, corresponding
both the massive and the massless
to the minus sign. In the p direction parallel to k,
dispersion relations get distorted. However, both degrees of freedom may still be separated
into positive and negative frequency parts, as is obvious from Eq. (2.3). In Fig. 1 we plot
the dispersion relation (2.3) restricted to the (p0 , p3 ) plane, where k is assumed to point
into the x 3 -direction as well. The separation into positive and negative frequency parts is
clearly seen in Fig. 1.
Without loss of generality we now assume that k points into the x 3 -direction, i.e.,
k = (0, 0, 0, 1). The dispersion relations for the two degrees of freedom then read
2
2
p02 =
p12 + p22 +
,
(2.4)
with

1
4p32 + m2 m .
(2.5)
2
Consider, for simplicity, the case of p1 = p2 = 0 and p3 0. Then, the dispersion relations
2 lead to the phase velocities
p02 =

4p32 + m2 m

vph =
(2.6)
=
p3
2p3

and group velocity
252
Fig. 2. The dispersion relation (2.8) in the (Rep0 , |p|)

halfplane for a purely timelike ChernSimons
parameter k = (1, 0, 0, 0) and mass scale m, with broken (solid) curves corresponding to the plus
(minus) sign in Eq. (2.8). For the minus sign in Eq. (2.8), the energy p0 becomes imaginary at low
momenta |p|
< m.
vg =
d
2p3
=
1.
dp3
4p32 + m2
(2.7)
For the case considered and m = 0, both velocities approach 1 in the limit p3 .
|
vph |, which is relevant for signal
More generally, the front velocity vf lim|p|

propagation [18], has the same value 1 in all directions (recall that c = 1 in our units).
This classical reasoning already indicates that the causal structure of the theory remains
unaffected by the additional CPT-violating term in Eq. (1.1), at least for the case
k = (0, 0, 0, 1). In Section 5.1, we shall find further evidence for this statement by
calculating the commutators of the quantized fields.
Before closing this section, we want to contrast the discussion above with that for the
case of a timelike ChernSimons parameter k , which has already been studied in detail
by the authors of Ref. [6]. Here, a particular preferred coordinate system is one where k
is purely timelike, k = (1, 0, 0, 0), which we assume in the following. Again, Eq. (2.1)
leads to a quadratic equation in p02 , with the solutions
2
p02 =
|p|
2 m|p|.

(2.8)
(These two degrees of freedom correspond to circular polarization modes, cf. Eq. (26)
of Ref. [6].) The dispersion relation (2.8) is plotted in Fig. 2. It is obvious that there is
no separation into positive and negative frequency parts. 2 Worse, the energy becomes
imaginary at low momenta |p|
< m for the minus sign in Eq. (2.8). In addition, the group
2 This fundamental difference of the MaxwellChernSimons theory (1.1) for the case of, say, k = (1, 0, 0, 0)
and k = (0, 0, 0, 1) traces back to our fixed choice of the time coordinate, namely x 0 ct.
253
velocities of both degrees of freedom may become arbitrarily large. For the minus sign in
Eq. (2.8), one has, for example,
2|p|
m
d
=
,
d|p|

2 |p|
2 m|p|

(2.9)
which is singular at |p|

= m. These results indicate that the case of timelike Chern
Simons parameter k is rather different from the case of spacelike k and does not allow
for quantization. In the next section, we shall find further evidence for this statement by
investigating the Feynman propagator.
3. Feynman propagator in a general axial gauge

We now consider the Feynman propagator which may be formally derived from the
Lagrangian (1.1), and investigate what may be inferred for the possible quantization of the
theory.
The action corresponding to Eq. (1.1) can be re-expressed as follows:

1
d4 x A g (1/ )n n m k A ,
S=
(3.1)
2
R4
so that the inverse propagator in momentum space becomes

1
G
(p) = p2 g + p p (1/ )n n im k p .
(3.2)
(G1 ) G = ,
The corresponding propagator, which obeys

reads

p2 + n2 + (m2 /p2 )(k n)2 + m2 k 2 (m2 /p2 )(k p)2
G (p) = g +
p p
(p n)2
1
m2
m2 (k n)
(p n + n p ) + 2 k k 2
(p k + k p )
(p n)
p
p (p n)

1
(p k)

+ im
p
k
n
n
K,
p2 (p n)
(p n)
(3.3)
with
K
p2

.
p4 + m2 k 2 p2 (k p)2
(3.4)
(Note that the equivalent propagator in a covariant gauge has already been computed in
Ref. [19].)
Up till now, the calculation of the propagator was formal and purely algebraic. We
did not discuss the pole structure, nor even define whether we are in Minkowski or
Euclidean spacetime. 3 A systematic treatment can be given for the spurious singularities
3 The classical Euclidean theory is derived from Eq. (1.1) by rotating both x 0 and k 0 to the imaginary axis.
(See Ref. [20] for a general discussion of Euclidean field theories.) An alternative Euclidean theory could perhaps
be defined by keeping k 0 absolutely fixed, but this is not what has been discussed in, for example, Refs. [4,7].
254
related to the axial gauge vector n (see, for example, Ref. [15]), and we ignore
these singularities in the following. Instead, we focus on the pole structure of the
propagator function K as given by Eq. (3.4). For clarity, we relabel our previous
ChernSimons parameter k in Minkowski spacetime as kM and use kE in Euclidean
space.
First, let us discuss the case of a purely spacelike kM = (0, 0, 0, 1), with index
running over 0, 1, 2, 3. For Minkowski spacetime with metric signature (, +, +, +), we
get
K=
p02 |p|
2
m2
2
m
2
2
p02 + |p|

m2 + 4p32 + i p02 |p|
2
m2
2
m
2
m2 + 4p32 + i
,
(3.5)
where both poles are displaced with the help of the usual Feynman i prescription
( = 0+ ), cf. Refs. [21,22]. For Euclidean space with metric signature (+, +, +, +) and
indices running over 4, 1, 2, 3, we find instead
K=
2+
p42 + |p|
m2
2
m
2
2
p42 + |p|

m2 + 4p32 p42 + |p|
2+
m2
2
m
2
m2 + 4p32
.
(3.6)
The poles from both factors in the denominator are placed on the positive and negative
imaginary axis of the complex p4 plane. A Wick rotation [21,22] to Minkowski spacetime
can be performed and leads to the i prescription (3.5) for the Feynman propagator (3.3)
of both degrees of freedom. Hence, the propagator is well-behaved, at least for the case of
a purely spacelike ChernSimons parameter.
For a purely timelike kM = (1, 0, 0, 0) in Minkowski spacetime, with index running
over 0, 1, 2, 3, we obtain
K =
=
p02 |p|
2
2
(p02 |p|
2 + m|p|)(p

2 m|p|)

0 |p|
1
2(p02 |p|
2 + m|p|)

1
2(p02 |p|
2 m|p|)

(3.7)
For low momenta |p|

< m, the poles in the first term are placed on the imaginary
p0 axis, which means that the energy becomes imaginary. This, in turn, implies that
unitarity is violated already at tree level, i.e., for the free theory (1.1). The region
|p|
< m has, therefore, to be excluded for this degree of freedom. (The situation is
2 + m2 = 0, where
similar to the case of a tachyon field with dispersion relation p02 |p|
the region |p|
< m has to be excluded in order to maintain unitarity of the quantum
field theory at tree level. See, for example, the discussion in Refs. [23,24].) But we
shall find in Section 5.2 that exclusion of the region |p|
< m leads to a violation of
microcausality.
If we now assume a purely timelike kE = (1, 0, 0, 0) in Euclidean space, with index
running over 4, 1, 2, 3, the function K becomes
K=
255
2
p42 + |p|
2
(p42 + |p|
2 + im|p|)(p

2 im|p|)

4 + |p|
1
1
+
.
=
2(p42 + |p|
2 + im|p|)

2(p42 + |p|
2 im|p|)

(3.8)
Here, the poles of the first (second) term are placed in the second and fourth (first and third)
quadrants of the complex p4 plane. In order to determine the behaviour of the propagator
(3.8) under a Wick rotation to Minkowski space, we have to remember that according to
our prescription we have to rotate k4 as well. This makes that the poles of Eq. (3.8) move
under Wick rotation. For sufficiently small |p|,
two poles will, in fact, move to the real
axis and, therefore, cross the Wick-rotated p4 -axis. In short, the analytic behaviour of the
propagator is problematic for the case of a purely timelike ChernSimons parameter.
For spacelike k in Minkowski spacetime, we have used up till now a special coordinate
system in which k is purely spacelike, that is, k0 = 0 exactly. Let us, finally, relax this
condition and investigate what happens if we allow for k0 = 0. In general, the four roots
of the denominator of Eq. (3.4) are rather complicated. We shall, therefore, make some
simplifying assumptions. By choosing k1 = k2 = 0, we can restrict ourselves to the plane
p1 = p2 = 0. Also, we choose units of energy and momentum such that k mk =
(k0 , 0, 0, 1). With these assumptions, we still find four real roots p0 = ri , i = 1, . . . , 4, as
long as |k0 | < 1.
With the same simplifications, we find in Euclidean space the following four roots
p4 = qi , i = 1, . . . , 4, for the denominator of Eq. (3.4):

i
q1 = 1 1 + 4i k4 p3 + 4p32 ,
q2 = q1 ,
2

i
q4 = q3 .
q3 = 1 + 1 + 4i k 4p3 + 4p32 ,
(3.9)
2
For k4 = 0, q1 and q3 have nonzero real parts of opposite sign (Re q1 = Re q3 ) and the
four poles of Eq. (3.4) are placed in all four quadrants of the complex p4 plane. Under
a Wick rotation, with k4 rotated as well, all four poles (3.9) move towards the imaginary
axis together with p4 (as long as k is spacelike, |k4 | < 1), and a Wick rotation may
be performed without crossing poles in the complex p4 plane. Hence, the propagator is
well-behaved, provided the ChernSimons parameter is spacelike.
This completes our elementary discussion of the Feynman propagator for the Maxwell
ChernSimons theory (1.1). In the next section, we will study the Euclidean propagator in
somewhat more detail.
4. Reflection positivity
An important condition for the quantization of a field theory in the Euclidean
formulation is reflection positivity [20,25]. This condition is essential for establishing the
existence of a positive semi-definite self-adjoint Hamiltonian H in Minkowski spacetime,
with the corresponding unitary time evolution operator exp(iH t).
256
The reflection positivity condition for an Euclidean two-point function is simply

x 4 , x x 4 , x 0,
(4.1)
where x 4 is the Euclidean time coordinate, (x 4 , x ) a scalar field of the theory, and
: (x 4 , x ) (x 4, x ) the reflection operation. Reflection positivity then gives the

following inequality for the scalar Euclidean propagator function G(p4 , p):

3
d p
dp4 e
ip4 x 4

G p4 , p

d3 p G x 4 , p 0,
(4.2)
for arbitrary values of x 4 . By choosing suitable smearing functions, it is even possible to

0, but for our purpose the condition (4.2) suffices.
derive the stronger condition G(x 4 , p)
For the gauge-invariant degrees of freedom of the MaxwellChernSimons theory (1.1),
it turns out to be sufficient to check the issue of reflection positivity for the Euclidean
propagator function K (as introduced in Eq. (3.3) above), thereby effectively reducing the
problem to the investigation of a scalar two-point function. Concretely, we then have to

verify whether or not the inequality (4.2) holds for our propagator function K(p4 , p).
For the case of purely spacelike kE = (0, 0, 0, 1), with index running over 4, 1, 2, 3,
is given by Eq. (3.6) and we get
the function K(p4 , p)

K x 4 , p =
+
4
dp4 eip4 x
p42 + |p|
2
2 )(p2 + 2 )
(p42 + +
+ e+ |x | e |x |
|p|
2 + e |x | e+ |x
+
2 2
2 2
+
+
+
2 ) e |x | + (2 |p|
2 ) e+ |x
(|p|
2
+
+
4
2 2 )
+ (+
4|
4|
(4.3)
where Eqs. (3.728.1) and (3.728.3) of Ref. [26] have been used to evaluate the integral
and the frequencies are defined in Eq. (2.4). This expression is manifestly positive
+ , and reflection positivity (4.2) holds.
semi-definite, since |p|
is given by
For the case of purely timelike kE = (1, 0, 0, 0), the function K(p4 , p)
Eq. (3.8) and we get

K x 4 , p =
+
4
dp4 eip4 x
p42 + |p|
2
p44 + 2p42 |p|
2 + |p|
4 + m2 |p|
2
+

=
dp4 2 cos p4 x 4
0
with
cos 2a |p|
2 /b2 ,
p42 + |p|
2
p44 + 2p42 b2 cos 2a + b4

b2 |p|
|p|
2 + m2 .
(4.4)
257
The integration over p4 can be performed explicitly and we obtain, using Eqs. (3.733.1)
and (3.733.3) of Ref. [26],

sin[a |x 4 |b sin a] |p|
4
4
2 sin[a + |x 4 |b sin a]

K x , p = exp x b cos a
+ 2
b sin 2a
b
b sin 2a
4

4
= exp x b cos a cos a + x b sin a b.
(4.5)
Clearly, this expression is not positive semi-definite (as long as m = 0) and numerical
integration over p shows reflection positivity (4.2) to be violated for large enough values
of m|x 4 |. The different behaviour of, respectively, Eqs. (4.3) and (4.5) is caused by the
different pole structure in Eqs. (3.6) and (3.8), which was also the crux of the previous
section.
5. Microcausality
Having dealt with unitarity, we continue our investigation of the hypothetical quantum
field theory based on the Lagrangian (1.1) and focus on the issue of causality. Minkowskian
conventions, with metric signature (, +, +, +) and indices running over 0, 1, 2, 3, are
assumed throughout this section and the units are such that c = h = 1.
5.1. Purely spacelike ChernSimons parameter
We prefer
Let us, again, start with the case of a purely spacelike vector k = (0, k).
to use a physical gauge condition, in order to avoid the problem of constructing the
subspace of physical states. Furthermore, we will try to connect to the well-known results
of Quantum Electrodynamics, i.e., the Lagrangian (1.1) for m = 0. We, therefore, switch
from the general axial gauge to the Coulomb gauge, A = 0, cf. Refs. [22,27].
0 , x ) is then given by
The resulting commutator for the gauge field A(x

Ai (x), Aj (0) = iTij i0 , i D(x),
(5.1)
with the commutator function

4
D(x) = (2)
dp0 d3 p
C
eip0 x
0 +i p
x
2 (p k)
2)
(p2 )2 + m2 (p2 |k|
(5.2)
for an integration contour C that encircles all four poles of the integrand in the
counterclockwise direction, cf. Appendix A1 of Ref. [28]. The denominator of Eq. (5.2)
is given by the dispersion relation (2.1) for purely spacelike ChernSimons parameter k .
The tensor Tij on the right-hand side of the commutation relation (5.1) is found to be
given by

Tij (p0 , p ) = p02 |p|
(5.3)
2 ij m2 sij + imp0aij ,
with
ij ij
pi pj
,
|p|
2
(5.4)
258

p k
p k
sij ki
pi kj
pj ,
|p|
2
|p|
2
aij ij a ka +
pj
pi
p k

p
k

p
k
=
ij l pl .
j
ab
a
b
iab
a
b
|p|
2
|p|
2
|p|
2
(5.5)
(5.6)
One immediately verifies that the commutator (5.1) respects the Coulomb gauge, since
pi Tij = 0 and pj Tij = 0. Further details on the derivation of this commutation relation can
be found in Appendix A.
Microcausality (i.e., the commutativity of local observables with spacelike separations,
cf. Refs. [12,21,28]) holds, provided that:
x |, and
1. The commutator function D(x) vanishes for spacelike separations |x 0 | < |
2. The poles of the type |p|
2 which occur in the tensor Tij are absent in the
commutators of physical, gauge-invariant fields (i.e., the electric and magnetic fields).
0 and magnetic field
In our case, the commutators of the electric field E 0 A A

B A are found to be the following (see Appendix A for details):

Ei (x), Ej (0) = 02 2 ij 02 i j + m2 02 ki kj

(5.7)
m03 ij l kl m0 (i j ab a kb j iab a kb ) iD(x),

Ei (x), Bj (0) = 02 2 ij l l 0 + m2 0 ki j ab a kb

+ m02 k ij m k i j m 02 2 i kj iD(x), (5.8)

Bi (x), Bj (0) = 02 2 ij 2 i j

2 k 2 |k|
2 i j ki kj 2
+ m2 ij 2 |k|

+ i kj + ki j k m0 k ij l l iD(x).
(5.9)
Remark that poles of the type |p|

2 , which could spoil causality, are indeed absent in
these commutators of physical field operators. In addition, we recover the JordanPauli
commutators of Quantum Electrodynamics [27] in the limit m 0 (remember that our
D(x) for m 0 obeys the massless KleinGordon equation squared, D = 0).
We still have to discuss the commutator function (5.2). Henceforth, we assume that k
points into the 3-direction, k = (0, 0, 1). We first observe that (5.2) vanishes for equal
times (x 0 = 0), because of the symmetry in p0 of the integrand (5.2), which results in
a cancellation of the residues (compare Eq. (5.10) for x 0 = 0 below). The commutator
function is also zero for (x 0 )2 < (x 1 )2 + (x 2 )2 , because the integrand can be made to
be symmetric in a new variable p0 , which is related to p0 by a conventional Lorentz
boost involving only p0 , p1 , and p2 . We will now show that the commutator function
x |. The reader who is not
D(x) vanishes, in fact, over the whole spacelike region |x 0 | < |
interested in the details may skip the rest of this subsection.
For (x 0 )2 (x 1 )2 + (x 2 )2 , it is still useful to perform a Lorentz transformation
involving only p0 , p1 and p2 , because there exists a transformation which allows us to
259
rewrite Eq. (5.2) as

D x 0 , x 3 = (2)4
= (2)

dp0
with

x
0
d3 p
dp0
eip0 x
0 +ip
3x
2 (p k)
2)
(p2 )2 + m2 (p2 |k|
eip0 x +ip3 x
d p
,
(p0 + )(p0 + + )(p0 )(p0 + )
(5.10)
0
(x 0 )2 (x 1 )2 (x 2 )2 0
x .
(x 0 )2
(5.11)
(For (x 0 )2 < (x 1 )2 + (x 2 )2 , we can effectively set x 0 = 0.) The contour integral is readily
performed,

0 3
sin + x 0
sin x 0
3
3
ip3 x 3
D x , x = (2)
(5.12)
d pe
,
2 2 )
2 2 )
(+
+ (+
with the roots explicitly given by Eq. (2.4). The integral (5.12) obviously vanishes for
x 0 = 0 (as long as x 3 = 0), and we are interested in determining its behavior for other
values of x 0 . We will start by demonstrating that D(x 0 , x 3 ) at the time slice x 0 = 0 is
ultra-local in x 3 , i.e., xn 0 D(x 0 , x 3 )|x 0 =0 is a sum of derivatives of the delta function (x 3 ).
If n is even, then xn 0 D(x 0 , x 3 )|x 0 =0 is obviously zero. If n = 2l + 1 is odd, then one has

x2l+1
D x 0 , x 3 x 0 =0 = (1)l+1(2)3
0
d3 p eip3 x
2 )l (2 )l
(+
2 2
+
(5.13)
2
Two remarks are in order. First, the fraction in the integrand is, in fact, a polynomial in +
2
and , namely,
2 )l (2 )l
(+
2
+
2 l2 2 l1
2 l1 2 l2
2
= +
+ +
+ + +
+
P2l2 .
(5.14)
2 as
Second, if we temporarily re-express
2
= a b,
(5.15)
where a is a polynomial in the momenta p and b is the square-root of a polynomial, then

the above polynomial (5.14) only depends on even powers of b, P2l2 = P2l2 (a, b2). The
last observation follows from the simple fact that P2l2 is invariant under the interchange
2 2 . These two remarks make clear that P
+
2l2 is a polynomial in the momenta p1 ,
p2 , and, especially, p3 , which implies ultra-locality.

The finite domain of vanishing D may be determined by a direct evaluation of the
integral (5.12). The end result of a straightforward calculation is that

D x 0 , x 3 = 0, for x 0 < x 3 ,
(5.16)
260
which corresponds to the usual spacelike region (x 0 )2 < (x 1 )2 + (x 2 )2 + (x 3 )2 , see

Eq. (5.11) above. The calculation proceeds in three steps.
2 2 ) in the denominators of (5.12) are independent
First, one notes that the factors (+
of p1 and p2 , so that these integrals can be readily performed for x 0 = 0,

D x , x
0
1
= 2 0
4 x

dp3 eip3 x
cos + x 0 cos x 0
2
2
+
(5.17)
where the |p2 +p2 =0 are given by Eq. (2.5). (To arrive at Eq. (5.17) we have
1
dropped the contribution at p12 + p22 = , which corresponds to a rapidly oscillating

function of x 0 that vanishes upon integration.) Note that the Taylor expansion of the
integrand of (5.17) in powers of x 0 has precisely the polynomials (5.14) as coefficients,
but now in terms of .
Second, we replace the variable p3 by , which is defined as follows

1
1
1
p3 m sinh ,
(5.18)
p32 + m2 m cosh .
2
4
2
2
2 ) in (5.17), so that only
This change of variables eliminates the denominator ( +
exponentials remain in the integrand. (The same procedure is followed in Section 15.1 of
Ref. [29] for the standard commutator function of massive scalars.) The result is then
D x , x
0
1
= 2 0
8 mx

3
d eip3 x cos + x 0 cos x 0 ,
(5.19)
with p3 and defined in terms of .

Third, the integral over can be evaluated, taking care of the relative signs and
magnitudes of x 0 and x3 . For the case of 0 < x 0 < x 3 , we write

x 0 x 2 sinh 0 ,
(5.20)
x 3 x 2 cosh 0 .
Defining 12 m x 2 , the integral in (5.19) becomes after a simple manipulation

d i sin[ sinh 0 ] ei sinh(+0 ) ei sinh(0 ) .
(5.21)
The first factor in brackets is a constant which can be taken out of the integral. But the
remaining integral of the second factor in brackets vanishes trivially (in the second term
shift + 20 ). Since the integral (5.19) is even in both x 0 and x 3 , and the
original commutator function D as given by Eq. (5.12) manifestly vanishes for x 0 = 0,
this establishes the result (5.16) announced above.
It may also be instructive to see what happens for the case of, say, 0 < x 3 < x 0 . Defining

x 0 x 2 cosh 0 ,
(5.22)
x 3 x 2 sinh 0 ,
261
one now gets for the integral in Eq. (5.19)

d i sin[ cosh 0 ] ei cosh(+0 ) ei cosh(0 ) ,
(5.23)
with 12 m x 2 . Taking out the constant factor and making a change of variables
+ 20 for the second term, one obtains

d ei cosh(+0 ) ei cosh(+0 ) ,
(5.24)
which need not vanish. Using Eq. (8.421) of Ref. [26], the integral (5.24) gives 2iJ0 (),
where the Bessel function J0 () is, in general, nonzero. All together, one has

1 0 sin( 12 mx 0 )
1 2
D x 0 , x 3 =
(5.25)
J
x
x
m
, for x 0 > x 3 ,
0
1
0
8
2
2 mx
where the antisymmetry in x 0 has been made explicit. 4 For fixed timelike separation x
and ChernSimons mass scale m 0, the commutator function approaches a constant
value (8)1 . (Remark that derivatives operating on D will result in further singularities on the null-cone x 2 = 0.) This completes our discussion of the gauge field commutator (5.1) for the case of a purely spacelike ChernSimons parameter k = (0, 0, 0, 1), with
microcausality established.
5.2. Purely timelike ChernSimons parameter
Let us, briefly, discuss the commutator function for the case of a purely timelike
vector k . In this case there is no invariant separation of the dispersion relation into
positive and negative frequency parts (see Fig. 2). However, as Lorentz invariance is broken
anyway, we may simply choose to quantize in the particular coordinate frame where
k = (1, 0, 0, 0). Specifically, we want to study the degree of freedom with dispersion
2 m|p|.
For this degree of freedom, the region |p|
< m has to be
relation p02 = |p|
excluded in order to maintain unitarity, as was mentioned a few lines below Eq. (3.7).
The relevant commutator [29] is then

(0), (x) = i D(x),
(5.26)
with (x) the quantum field corresponding to this particular degree of freedom of the
gauge field (recall that h = c = 1) and the commutator function

= 1
D(x)
2 + m|p|
(|p|
m) (p0 )eipx
d4 p p02 |p|
3
i(2)

1
(|p|
m)
3

=
p
|p|
2 m|p|
x 0 ei x p .
(5.27)
d
sin
3
(2)
|p|
2 m|p|

4 The structure of our commutator function (5.25) closely resembles the one obtained for a CPT-violating
massive Dirac fermion, as given in Appendix E of the first paper in Ref. [7]. Note, however, that in our case m
sets the scale of the CPT violation for the photons.
262
Here, (x) x/|x| and is the usual step function, (x) = 0 for x < 0 and (x) = 1
for x > 0. We will now demonstrate that microcausality is violated for this commutator
= 0 somewhere in the spacelike region |x 0 | < |
function, i.e., D(x)
x |.

x |, this would imply that D(x 0 , x 3 ) dx 1 dx 2 D(x)
If D(x) were to vanish for |x 0 | < |
0
3
0
3
= 0 for |x | < |x |. So, let us show that the latter relation is violated. If D(x , x ), in turn,
were to vanish for |x 0 | < |x 3|, this would imply that 0 D(x 0 , x 3 )|x 0 =0 had to be an ultralocal expression in x 3 , but it can be easily checked that this is not the case. Indeed, we
calculate

0 D x 0 , x 3 x 0 =0

dp3 (|p3 | m)
0 ix 3 p3
2

sin |p3 | m|p3 | x e
= 0
0
2
2 |p3 | m|p3 |
x =0

=
sin mx 3
dp3
3
(|p3 | m)eix p3 = x 3
,
2
x 3
(5.28)
which can be nonzero for finite x 3 , as long as m = 0. (Recall that for purely spacelike
k = (0, 0, 0, 1) we have shown the ultra-locality of Eq. (5.13) above.) For the case
of a purely timelike ChernSimons parameter k , the commutator (5.26) thus violates
microcausality.
6. Summary and discussion

Lorentz- and CPT-violating field theories might emerge as low-energy effective theories
of a more fundamental theory, where Lorentz and CPT symmetry are broken spontaneously
or dynamically [7]. Alternatively, these symmetry-breaking theories might result from
a quantum anomaly within the realm of quantum field theory itself [4]. In both instances,
the question arises whether or not these Lorentz- and CPT-violating theories are valid
quantum field theories, that is, whether or not a consistent quantization is possible. For
theories which contain an Abelian ChernSimons-like term (1.2), issues like powercounting renormalizability and conservation of energymomentum have already been
discussed in Ref. [7], where it was demonstrated that these features continue to hold.
In this paper, we have focused on the issues of unitarity and causality for theories
containing an Abelian ChernSimons-like term (1.2). The results found strongly depend
on whether the preferred direction k of the Lagrangian term (1.2) is spacelike or timelike.
our results are certainly
For a purely spacelike ChernSimons parameter k = (0, k),
encouraging for the issue of quantization. By investigating the dispersion relations we have
found in Section 2 that a universal, direction-independent signal propagation speed c can
still be defined (in this paper, we have chosen units such that c = 1). In addition, the group
velocity is less or equal to c. This suggests that the CPT-violating term in Eq. (1.1) does not
change the causal structure of spacetime, but rather acts like a medium with a directiondependent dispersion for the field excitations (e.g., the photons).
263
In fact, the anisotropic propagation of the circular polarization modes makes clear
is T-odd (and
that the Abelian ChernSimons-like term for purely spacelike k = (0, k)
P- and C-even). According to the dispersion relation (2.4) for ChernSimons parameter
k = (0, 0, 1), left-handed wave packets propagating in different directions along, say,
the x 2 -axis (with an infinitesimal p3 -component added) have unequal group velocities,
vgL (0, p2 , p3 )|, as long as m = 0. The physics is thus non-invariant
|
vgL (0, p2 , p3 )| = |
under time reversal, which flips the momentum, p p,
and preserves the helicity,
L L, cf. Ref. [22].
We have also found from the dispersion relations that a separation into positive and
negative frequency modes is still possible for purely spacelike k . Therefore, particles and
antiparticles may be defined and the field may be quantized in the usual fashion [2729].
As shown in Section 5.1, the resulting commutators of the electric and magnetic fields
vanish for spacelike separations, which demonstrates that microcausality holds for the
potential quantum field theory based on the Lagrangian (1.1). This result is not quite
trivial. Recall, for example, that the CPT-violating theories of Ref. [13], with self-conjugate
bosons of odd-half-integer isospin, fail precisely in this respect, because certain fields
do not commute outside the light-cone. The well-known result [11] that microcausality,
Lorentz invariance and the existence of a unique vacuum state imply CPT invariance does
not contradict our result, because in the Lagrangian (1.1) Lorentz invariance is broken as
well (k is fixed once and for all).
In addition, we have demonstrated in Section 3 that a Feynman propagator for the
relevant Lagrangian (1.1) can be defined in Minkowski spacetime with the usual i
prescription and that this propagator can be Wick-rotated to Euclidean space. As shown
in Section 4, reflection positivity holds for the Euclidean Feynman propagator. Again, this
requires the choice of a (purely) spacelike k .
For both a classical and a quantum treatment, the causal structure of the Maxwell
ChernSimons theory (1.1) thus remains unaltered by the inclusion of the CPT-violating
term, provided that the ChernSimons parameter k in Eq. (1.2) is spacelike. This suggests
that a CPT- and Lorentz-symmetry-violating theory like (1.1) may lead to a consistent local
quantum field theory. If so, the particular chiral gauge field theories discussed in Ref. [4],
which display the remarkable phenomenon of a CPT anomaly, could perhaps be realized
in nature (see also the remarks below).
On the other hand, a consistent quantization for a timelike ChernSimons parameter
k does not seem to be possible. As noted in Section 3, the presence of imaginary
energies at low momenta requires the exclusion of these momenta if unitarity is to be
maintained. 5 But we have seen in Section 5.2 that this exclusion leads to a violation of
microcausality. It is, therefore, not possible to maintain both unitarity and causality. In
fact, these results are just the quantum analogs of the results of Ref. [6]. The authors
of that paper have pointed out that the Greens functions of the classical equations of
motion resulting from the Lagrangian (1.1) either are causal but with exponential growth
in time, or without exponential growth but noncausal. For timelike k , there are no
5 Moreover, tachyon pair production would destabilize the perturbative vacuum state, see Section 4 of Ref. [19].
264
Greens functions that are both causal (i.e., propagating signals only into the future) and
without exponential growth. As shown in Section 4 of this paper, reflection positivity in
the Euclidean formulation is indeed violated for the Feynman propagator with Chern
Simons parameter k = (1, 0, 0, 0).
Hence, the Abelian ChernSimons-like term (1.2) for timelike k , which contains a Podd (and C- and T-even) part, does not allow for a consistent quantization. It does appear,
however, that a ChernSimons-like term could play a role for T-violation (leaving C- and Pinvariance intact), provided the parameter k is (purely) spacelike. Such a ChernSimonslike term would, in fact, provide a fundamental arrow-of-time, cf. Ref. [1]. (Recall the
Gedankenexperiment with the left-handed wave packet presented at the beginning of this
section.) This problem is currently under investigation.
As briefly mentioned in Ref. [4], the birefringence of a photonic ChernSimons-like
term (1.2) with purely spacelike parameter k could also affect the polarization of the
cosmic microwave background. The expected polarization pattern [30] around temperature
hot- and cold-spots would be modified, due to the action of the ChernSimons-like term on
the photons traveling between the last-scattering surface (redshift z 103 ) and the detector
(z = 0). Future satellite experiments such as NASAs Microwave Anisotropy Probe and
ESAs Planck Surveyor could look for this effect, see Ref. [31] for further details.
Let us end this section by discussing three somewhat more theoretical issues. The first
issue concerns stability, which has been discussed recently for a theory of a massive Dirac
fermion with (spontaneous) Lorentz and CPT breaking [32]. For the MaxwellChern
Simons theory (1.1), the photon is, of course, stable. This holds, in particular, for the
case of purely spacelike ChernSimons parameter k . But even in the context of the CPT
anomaly [4], with all chiral fermions integrated out, the effective action can be expected
to have additional quartic (and higher-order) interaction terms for the photons. A rough
estimate suggests an extremely small, but nonzero, effect for the decay of the photon,
cf. Eq. (1.3). (The photons of the cosmic microwave background would not be affected
significantly.) Whether or not photon decay is physically acceptable remains an open
question, though.
The second issue concerns the case of a lightlike ChernSimons parameter k , which
has not been discussed so far. No imaginary energies appear for lightlike k . There is,
therefore, no obvious obstacle against quantization, as was the case for timelike Chern
Simons parameters. But explicit calculations of the type performed in Sections 4 and 5 are
hampered by the complicated pole structure for the lightlike case. For the CPT anomaly,
there may still be a problem with the lightlike theory, as will become clear shortly.
The third, and final, issue concerns the possible implications of our microcausality results for the case of an anomalous origin of the Lagrangian (1.1) considered here. As mentioned several times by now, it has been shown in Ref. [4] that certain chiral gauge field
theories defined over the Euclidean spacetime manifold M = R3 S 1 , for example, could
give rise to a CPT anomaly of the form of a ChernSimons-like term (1.2), with an additional factor i for the Euclidean signature of the metric. (In the context of the CPT anomaly,
the exponential of the integrated Euclidean ChernSimons-like term appears as the phase
factor of the chiral determinant, whereas the absolute value of the chiral determinant is CPT
265
invariant.) In that case, the specific parameters of Eq. (1.2) would be m = (2n + 1)/L and
k = (0, 0, 0, 1), with the integer n defining the theory in the ultraviolet and the nonzero
entry of k corresponding to the single compact dimension of length L.
The microcausality results of the present paper then imply that this compact dimension
should correspond to a spatial direction after the Wick rotation to the Lorentzian signature
of the metric (x 3 S 1 becoming a spatial coordinate and x 4 R, say, the time coordinate).
If, on the other hand, the compact dimension were to correspond to the time direction
(x 4 S 1 ), our results would lead us to expect problems with unitarity or causality. Indeed,
we would then have started from a spacetime manifold with closed timelike curves, which
has a built-in violation of what might be called macrocausality, cf. Section 8.2 of
Ref. [33]. 6 Still, the proper fundamental theory (gauge fields and chiral fermions over
a spacetime manifold with a separable compact dimension that is spacelike) and the
corresponding effective theory (gauge fields with a ChernSimons-like term for purely
spacelike parameter k ) appear to be consistent as far as causality is concerned.
Acknowledgements
It is a pleasure to thank the anonymous referee for useful comments.
Appendix A. Commutators for the Coulomb gauge

In this appendix, we first discuss the equations of motion of the Lagrangian (1.1) for
but with the general axial gauge
purely spacelike ChernSimons parameter k = (0, k),

replaced by the Coulomb gauge A = 0. From these equations of motion, we then
determine the tensor structure Tij of the gauge field commutator (5.1). Finally, we
calculate the commutators of the electric and magnetic fields.
The equations of motion in momentum space are (with the conventions g =
diag(1, 1, 1, 1), 0123 = 123 = 1, and using p A = 0)

p02 |p|
2 g A p p0 A0 + im l A p kl = 0.
(A.1)
This leads to a nondynamical equation for A0 ,

A0 = i
m
abc Aa pb kc ,
|p|
2
(A.2)
and dynamical equations for Ai ,

6 Another spacetime manifold with closed timelike curves would be, for example, M = S 1 R2 S 1 , with
periodic coordinates x 4 [0, L4 ] and x 3 [0, L3 ] and noncompact coordinates x 1 , x 2 R. The corresponding
1
effective action could then contain a ChernSimons-like term (1.2) with parameter k (L1
4 , 0, 0, L3 ). For
L4 = L3 exactly, the ChernSimons parameter k would be lightlike. Note also that the transformation of k
as given by Eq. (2.2) may not be applicable in general for this manifold (compare with Eq. (13) of Ref. [5]).
266

Mij Aj
= 0.
2
(p k)
|p|
m
ij
|p|
2

k
pi
2p
2
imp0
j ab pa kb + ij a ka m
pi kj + m ki kj Aj
|p|
2
|p|
2
p02

2 |p|
2 |k|
2
(A.3)
The gauge field commutator (5.1) will obey the equations of motion (A.3) provided that
Mij Tj k (p) = ik D 1 (p),
(A.4)
2

2 p k 2 .
D 1 (p) = p2 + m2 p2 |k|
(A.5)
with
(The tensor structure on the right-hand side of Eq. (A.4) could, in principle, be more
complicated, but turns out to be just ik .) With some effort, it can be verified that (A.4)
holds for the Tij as given by Eq. (5.3).
The magnetic fields then have the following commutator:
[Bi , Bj ](p) = i 2 iab pa j cd (pc )Tbd iD(p)

= p02 |p|
2 ij |p|
2 pi pj + imp0 p k ij l pl
2 2

p k 2 pi pj |k|
2 ki kj |p|
|k|
m2 ij |p|
2

+ (pi kj + ki pj ) p k iD(p).
(A.6)
The electric fields are more involved,

E(x)
= 0 A(x)
m d 3 z 1 x z z abc ka bz Ac x 0 , z ,
(A.7)
and one finds after some algebra the following commutators:

2 ij l pl p0 m2 p0 ki j ab pa kb imp02 p k ij
[Ei , Bj ](p) = p02 |p|

+ im p k pi pj + im p02 |p|
2 pi kj iD(p),
(A.8)
[Ei , Ej ](p) =

p02 |p|
2 ij p02 p02 |p|
2 pi pj m2 p02 ki kj

+ imp03 ij l kl + imp0 pi j ab pa kb pj iab pa kb

2

2

pi pj 2
2 2
2
2 2

p
+
m
k
|
p|

|
p|

|
k|
iD(p).
0
0
|p|
2
(A.9)
Apparently, the electric field commutator contains a term with a |p|

2 pole. But this term
1
is multiplied by precisely the function D (p), which cancels the over-all factor D(p).
Therefore, this term does not contribute to the contour integral (5.2), and we reproduce the
commutators (5.7)(5.9) as given in the main text.
267
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
R.M. Wald, Phys. Rev. D 21 (1980) 2742.

S.W. Hawking, Phys. Rev. D 32 (1985) 2489.
V.A. Kosteleck, R. Potting, Nucl. Phys. B 359 (1991) 545.
F.R. Klinkhamer, Nucl. Phys. B 578 (2000) 277.
F.R. Klinkhamer, J. Nishimura, Phys. Rev. D 63 (2001) 097701.
S.M. Carroll, G.B. Field, R. Jackiw, Phys. Rev. D 41 (1990) 1231.
D. Colladay, V.A. Kosteleck, Phys. Rev. D 55 (1997) 6760;
D. Colladay, V.A. Kosteleck, Phys. Rev. D 58 (1998) 116002.
S. Coleman, S.L. Glashow, Phys. Rev. D 59 (1999) 116008.
R. Jackiw, V.A. Kosteleck, Phys. Rev. Lett. 82 (1999) 3572.
M. Chaichian, W.F. Chen, R. Gonzales Felipe, Phys. Lett. B 503 (2001) 215.
R. Jost, Helv. Phys. Acta 30 (1957) 409.
R. Streater, A. Wightman, PCT, Spin and Statistics, and All That, Benjamin, New York, 1964.
P. Carruthers, Phys. Rev. Lett. 18 (1967) 353;
P. Carruthers, Phys. Lett. B 26 (1968) 158.
S. Chern, J. Simons, Ann. Math. 99 (1974) 48.
G. Leibbrandt, Noncovariant Gauges, World Scientific, Singapore, 1994.
B. Nodland, J.P. Ralston, Phys. Rev. Lett. 78 (1997) 3043.
J.F.L. Wardle, R.A. Perley, M.H. Cohen, Phys. Rev. Lett. 79 (1997) 1801.
L. Brillouin, Wave Propagation and Group Velocity, Academic Press, New York, 1960.
A.A. Andrianov, R. Soldati, L. Sorbo, Phys. Rev. D 59 (1999) 025002.
I. Montvay, G. Munster, Quantum Fields on a Lattice, Cambridge Univ. Press, Cambridge,
1994.
M. Veltman, Diagrammatica The Path to Feynman Rules, Cambridge Univ. Press,
Cambridge, 1994.
S. Weinberg, The Quantum Theory of Fields I, Cambridge Univ. Press, Cambridge, 1996.
J. Dhar, E. Sudarshan, Phys. Rev. 174 (1968) 1808.
T. Jacobson, N.C. Tsamis, R.P. Woodard, Phys. Rev. D 38 (1988) 1823.
K. Osterwalder, R. Schrader, Commun. Math. Phys. 31 (1973) 83;
K. Osterwalder, R. Schrader, Commun. Math. Phys. 42 (1975) 281.
I.S. Gradshteyn, I.M. Ryzhik, Table of Integrals, Series and Products, Academic Press, New
York, 1980.
W. Heitler, The Quantum Theory of Radiation, 3rd edn., Oxford Univ. Press, London, 1954.
J.M. Jauch, F. Rohrlich, The Theory of Photons and Electrons, 2nd edn., Springer, New York,
1976.
N.N. Bogoliubov, D.V. Shirkov, Introduction to the Theory of Quantized Fields, Wiley, New
York, 1959.
D. Coulson, R. Crittenden, N. Turok, Phys. Rev. Lett. 73 (1994) 2390.
N.F. Lepora, gr-qc/9812077;
M. Kamionkowski, A. Kosowsky, Ann. Rev. Nucl. Part. Sci. 49 (1999) 77.
V.A. Kosteleck, R. Lehnert, Phys. Rev. D 63 (2001) 065008.
R.M. Wald, General Relativity, Chicago Univ. Press, Chicago, 1984.

Dynamical CP violation and flavour-changing

processes
G.C. Branco a,b , D. Delpine c , R. Gonzlez Felipe b,
a CERN Theory Division, Geneva, Switzerland
b Centro de Fsica das Interaces Fundamentais, Departamento de Fsica, Instituto Superior Tcnico,
1049-001 Lisboa, Portugal

c INFN, Laboratori Nazionali del Gran Sasso, I-67010 Assergi (AQ), Italy
Received 2 February 2001; accepted 14 May 2001
Abstract
We investigate the phenomenological constraints on a model where, besides the standard model
Higgs sector, there is an effective new strong interaction acting on the third generation of quarks
and characterized by a -like term. This term induces electroweak symmetry breaking and leads
to dynamical spontaneous CP violation. We show that the constraints coming from K physics and
the electric dipole moment of the neutron impose that the new physics scale should be of the order
of 35 TeV. Contrary to naive expectations, the predictions of the model for B physics are very close
to the standard model ones. The main differences appear in processes involving the up quarks such
0 mixing and in the electric dipole moment of the neutron, which should be close to the
as D 0 D
experimental limit. Possible deviations from the standard model predictions for CP asymmetries
in B decays are also considered. 2001 Elsevier Science B.V. All rights reserved.
PACS: 12.60.Rc; 12.15.Ff; 12.15.Hh
1. Introduction
The fact that the top quark is much heavier than the other quarks, mt = 174.35.1 GeV
[1], is suggestive of a new dynamics at the electroweak scale, where the third generation
may be playing a special role. In particular, effective four-fermion interactions [2] can
lead to the formation of quarkantiquark bound states which in turn can dynamically
trigger the breaking of the electroweak symmetry [3,4]. This is the basic idea of top-quark
condensation as well as of technicolor models, i.e., the Higgs sector of the standard model
* Corresponding author.
E-mail addresses: gbranco@cfif.ist.utl.pt (G.C. Branco), david.delepine@lngs.infn.it (D. Delpine),

gonzalez@gtae3.ist.utl.pt (R. Gonzlez Felipe).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 2 3 8 - 3
G.C. Branco et al. / Nuclear Physics B 607 (2001) 268292
269
is just an effective GinzburgLandau-type description of low-energy physics represented

by a composite isodoublet scalar field (or fields) [5].
In the above framework, a particularly interesting scenario is provided by models
where the top quark mass arises mostly from a t t condensate, generated by a new strong
dynamics, plus a small fundamental component, generated by an extended technicolor
or Higgs sector [610]. Such a structure for the top quark mass avoids the problems
usually found in pure (minimal) top-quark condensation scenarios, which assume that the
t t condensate is fully responsible for the electroweak symmetry breaking [4], thus leading
to a too large mt value (mt 220 GeV) and a very large scale for the new dynamics
( 1015 GeV) with significant fine tuning.
Along this line, a dynamical scheme was proposed in Ref. [11], where it is assumed that
the third generation of quarks does indeed experience new forces, symmetric in t and b,
and that these new forces also generate a strong CP phase . It is then possible to show
that, in such a scenario, the term triggers the breaking of the symmetry between t and
b and induces a large CP-violating phase in the CabibboKobayashiMaskawa (CKM)
matrix, due to the smallness of the mb /mt mass ratio [11]. In this model one expects to
have a richer low-energy phenomenology when compared to the standard model (SM),
which could lead to potentially interesting effects, specially in K and B physics.
The purpose of this paper is to study the low-energy phenomenological implications
of the model proposed in [11] and, in particular, its implications for K and B physics.
We will show that new observable effects arise due to the fact that the third generation
of quarks experiences new strong forces which in turn lead to scalar flavour-changing
neutral current (FCNC) interactions at tree level. These FCNC interactions result from
the fact that both the up and down quark mass matrices receive contributions not only from
Yukawa interactions with the standard Higgs but also from interactions involving the third
generation quarkantiquark bound states.
The present and near future experiments at B factories and the large hadron collider
(LHC) will certainly improve the bounds on many of the CP-violating and flavourchanging processes, which are forbidden or strongly suppressed in the SM. Therefore it is
particularly interesting to determine possible experimental signatures in models involving
new FCNC physics.
2. The model
In this section we shall briefly present the main features and physical consequences of
the model in question. A more complete and detailed analysis can be found in Ref. [11].
We consider a standard model Higgs sector in combination with an effective new strong
interaction acting on the third generation of quarks and characterized by a term. We
require that this new strong interaction conserves the isospin symmetry between t and b
quarks. Moreover, if one assumes that the electroweak symmetry breaking is induced
by radiative corrections due to top-quark (and possibly, bottom-quark) loops, the quartic
self-interactions of the Higgs field may be neglected. In this case, the relevant classical
Lagrangian for the fundamental scalar field H is given by
270

+ h.c. ,
LH = D H D H m2H H H + ht L tR H + hb L bR H
where
H=

H0
,
H
=
H
H+
H 0

,
L =
tL
bL
(1)

;
ht and hb are the Yukawa couplings and D is the usual covariant derivative of the SM.
Next, one assumes that the interactions acting on the members of the third generation of
quarks are strong enough to form quarkantiquark bound states at the electroweak scale.
The latter can be described in terms of two complex doublet scalar fields
0
+
t
b
t
bR L ,
t =
(2)
R L
b
t
b0
and the corresponding effective Lagrangian then reads:
L = D t D t + D b D b

b + h.c. .
m2 t t + b b + g L tR t + L bR
(3)
The effects of a new strong CP phase can, in principle, be described through an

arbitrary function of det U , where

0

t
b
tL tR tL bR
U
(4)
=
.
t+ b0
bL tR bL bR
In analogy with QCD [12] we shall assume the Lagrangian form 1

2

L = i Tr ln U ln U + 2 ,
4
which typically arises as a leading term in a 1/N -expansion.
The total effective Lagrangian of the model is then given by
L = LH + L + L ,
(5)
(6)
with LH , L and L defined by Eqs. (1), (3) and (5), respectively. Notice that if ht = hb the
Lagrangian (6) conserves an isospin symmetry. However, as shown in [11], the angle
provides a dynamical origin for both CP violation and isospin breaking, once the neutral
components of the three doublets H , b and t acquire nonzero vacuum expectation
values (VEVs).
Denoting the VEVs of the neutral components of the fields by
0
0
0
v
t
b
b = eib ,
H = ,
(7)
t = eit ,
2
2
2
the effective potential reads

v 2 m2 2
+
t + b2 2t + 2b + 4t + 4b
2
2
+ ( t + b )2 ,
V = m2H
1 Another simple choice is given by the t Hooft determinant, i.e., L = ei det U + h.c.
(8)
271
where

1 2 2
(9)
h v + g 2 i2 + 2hi vgi cos i , i = t, b;
2 i
and are some effective quadratic and quartic couplings, respectively. All couplings and
parameters in the potential are assumed to be real and positive.
The minimization of the potential implies the following system of equations:
2i =
AH v = ght It t cos t + ghb Ib b cos b ,

At t = ght It v cos t ,
Ab b = ghb Ib v cos b ,
ght It vt sin t = ghb Ib vb sin b = 2( t + b ),
(10)
where
AH = m2H h2t It h2b Ib ,
Ai = m2 g 2 Ii ,
Ii = 22i .
(11)
The mass parameters mH and m are chosen such that the quantities AH , At and Ab defined
in Eqs. (11) are always positive.
If the parameter is large, m2t , then the last equation in (10) implies the constraint
t b .
(12)
Furthermore, if = 0, it is easy to show that t = b = 0 is the only solution of the

equations and therefore CP is conserved.
A simple analytical solution can be given for the isospin symmetric case ht = hb = 0
and assuming 2m2t . In this case It Ib , At Ab and therefore sin 2t sin 2b .
Clearly the large splitting between the physical values of the bottom and top masses (mb
mt ) requires b t and thus t 0, b /2, which in turn demands that the CPviolating phase be close to /2. In other words, the presence of a phase close to /2
induces both isospin breaking and CP violation with
b
b
v = 0,
t ,
b + .
b t = 0,
(13)
t
2
t
The actual values of the VEVs can be determined from the physical values of the masses
mb , mt and mW . For small values of v, i.e., v b , t , one has the simple expressions
mb v0
mt v0
mb
,
t
,
tan t
,
b
(14)
mt
m2 + m2
m2 + m2
t
where v0 = v 2 + t2 + b2 = ( 2 GF )1/2 246 GeV, GF is the Fermi coupling

constant.
The mass spectra of the neutral and charged (pseudo) scalars are easily found. In the
neutral sector, it is straightforward to find the linear combination corresponding to the
Goldstone boson eaten up by the Z 0 gauge boson. For very large, one of the eigenvalues
of the mass matrix will be proportional to and therefore the corresponding linear
combination of the fields will decouple from the theory. The remaining 4 4 mass matrix
272
can be easily
diagonalized. One finds that the standard Higgs scalarh has a mass given by
mh 2g mt , two of the remaining masses are proportional to and thus are quite
5 b bound state is very sensitive
large. Finally, the mass which corresponds mainly to a b
to the difference ht hb , but as soon as ht and hb differ (as expected from higher order
corrections) it will also get a contribution proportional to . In the charged sector one
of the eigenstates is eaten up by the W gauge boson through the usual Higgs mechanism.
For the isospin symmetric case hb = ht , we find that one of the charged Higgs masses is
very small, i.e., a new pseudo-Goldstone boson appears as it happens in the neutral sector.
Nevertheless, radiative corrections yield hb = ht and therefore this mass will get a large
contribution proportional to .
To conclude this section let us comment on the origin of CP violation in the present
model. As shown in Ref. [11] the new interaction characterized by a = 0 term induces
a CP-violating effect which filters down to the SM only if mb /mt = 0. Moreover, this
new source of CP violation can be in principle responsible for what is observed in the
0 system, since it leads indeed to a sizeable CP-violating phase in the CKM matrix,
K 0 K
KM (t + b ) /2.
3. The structure of flavour-changing interactions
In general, the presence of more than one Higgs doublet in the SM leads to FCNC
interactions at the tree level, which are mediated by the physical neutral scalars. Such
interactions are severely constrained by the smallness not only of the CP-violating
0 and B 0 B
0 mixing. The model we are considering is
parameter K but also of the K 0 K
effectively equivalent to a three Higgs doublet model with a specific structure for Yukawa
couplings. It is therefore straightforward to determine the form of the induced FCNC
interactions by generalizing the results obtained in the two Higgs doublet case [13].
Let us consider 3 Higgs doublets j and make the decomposition:

j+
j = eij
(15)
, j = 1, 2, 3,
1 (vj + Rj + iIj )
2
where Rj , Ij are real fields and vj eij denote the VEVs of the Higgs fields. The Yukawa
couplings of the Higgs fields to the quark weak eigenstates are given by
LY = (u L dL )1 g1d dR (u L dL )2 g2d dR
1 g1u uR (u L dL )
3 g3u uR ,
(u L dL )
(16)
i2 and g u,d (i = 1, 2, 3) are the Yukawa coupling matrices. The quark

where
i
mass matrices are easily obtained,
1
1
Mu = v1 g1u + ei3 v3 g3u ,
(17)
2
2
1
1
Md = v1 g1d + ei2 v2 g2d ,
(18)
2
2
where the phase 1 has been put equal to zero by an appropriate redefinition of the fields.
273
To single out the pseudo-Goldstone boson G0 we introduce the new fields 0 , R, R ,

G0 , I and I defined through the transformation

0

0
R1
I1
G
R2 = O R ,
I2 = O I ,
(19)
R
I
R3
I3
with
v1 /v0
O = v2 /v0
v3 /v0
v2 /v
v1 /v
0
v1 v3 /v0 v
v2 v3 /v0 v
v /v0
(20)
and v02 = v12 + v22 + v32 , v 2 = v12 + v22 . In terms of the new fields, the scalar couplings to
the down quarks can be written as

v2
v1
R + iI
1
LdY = dL Md dR 0 + iG0 dL g1d g2d ei2 dR
v0
v
v
2
v3
dL Md dR (R + iI ) + h.c.
(21)
v0 v
We notice that the couplings to the fields 0 , G0 , R and I are flavour-conserving while
the couplings to R and I are flavour-violating. Similarly, for the couplings to the up quarks
one obtains

0

1
R iI
u
0
u v2
LY = u L Mu uR iG u L g1 uR
v0
v
2

iI
v
v
v
R
1 3
+ h.c.,
g3u ei3 uR
u L g1u
(22)
v0 v
v0
2
and thus the couplings of 0 , G0 conserve flavour while the couplings of R, R , I, I do
violate flavour.
It is useful to obtain the scalar-quark couplings in terms of the quark mass eigenstates.
In the down quark sector we find

1
R + iI
dL Dd dR 0 + iG0 dL Nd dR
v0
2
v3
dL Dd dR (R + iI ) + h.c.,
v0 v
LdY =
Md UdR = diag(md , ms , mb ) and

where Dd = UdL

2 v2
v
d v2
d v1 i2
UdR = Dd ei2 Gd2 ,
Nd = UdL g1 g2 e
v
v
v v1
v1
d
Gd2 UdL
g2 UdR .
(23)
(24)
(25)
In the up quark sector the couplings to the scalars in the quark mass eigenstate basis are
given by
274

1
u L Du uR 0 iG0
v0
R iI
R iI
u L Nu uR
u L Nu uR
+ h.c.,
2
2
LuY =
(26)
Mu UuR = diag(mu , mc , mt ) and

where Du = UuL
v2 u
U g UuR ,
v uL 1

v1 v3 u v u i3
Nu = UuL
UuR ,
g
g
e
v0 v 1 v0 3
Nu =
(27)
(28)
which can be rewritten as

v2
v2 v3
Nu = 2 2Du 2 ei3 Gu3 ,
v1
v1

2
v3
v3
v i3 u

e
Nu =
2Du
+
G3 ,
v0 v
v0 v v0
g3u UuR .
Gu3 UuL
The Yukawa
namely,
0
g2d = 0
0
(30)
(31)
coupling matrices
0 0
0 0 ,
0 gb
(29)
g2d , g3u
have a very simple form in the present model,
0 0
g3u = 0 0
0 0
0
0 ,
gt
and therefore the matrices Gd2 , Gu3 defined in Eqs. (25) and (31) are given by

d
UdR 3j ,
G2 ij = gb UdL
3i

u
UuR 3j .
G3 ij = gt UuL
3i
(32)
(33)
(34)
These matrices completely determine the structure of tree-level FCNC interactions in the
model. Without further assumptions we cannot predict the size of such interactions. We
shall assume that the quark mass matrices Mu,d are hermitian 2 and that the CKM mixing
matrix V UuL
UdL is dominated by UdL , i.e., UuL 1, as favoured phenomenologically.
Under the above reasonable assumptions, the off-diagonal elements of Nd in Eq. (24) are
entirely predicted in terms of V since from Eq. (33) we obtain
d
G2 ij = gb V3i V3j .
(35)
Finally, new contributions to flavour-changing processes will be also induced by the
couplings of the heavy charged Higgs fields to the quarks. Such contributions correspond to
Feynman box diagrams with W -boson and charged Higgs particle exchanges. To determine
2 According to the polar decomposition theorem, the mass matrices M
u,d can always be written as a product of
a hermitian matrix and a unitary matrix. The latter can be rotated away by a redefinition of the right quark fields.
Notice however that the form of the coupling matrices g2d and g3u given in Eq. (32) is in general not invariant
under such a transformation. Here we shall assume that the quark mass matrices are hermitian in the basis where
the couplings have the special form (32). Our analysis can be easily extended to a more general case.
275
the magnitude of these couplings, let us introduce the new charged fields G+ , H1+ , H2+
through the decomposition
+
+
1
G
+
+
(36)
2 = O H 1 ,
3+
H2+
where the matrix O is given by Eq. (20) and G+ corresponds to the pseudo-Goldstone
boson. Going to the physical basis for the charged Higgs fields, the couplings to the dR
and uR quarks are given by
L Md dR G+ u L Ad1 dR H1+ u L Ad2 dR H2+

L+
Y = 2u
2 dL Mu uR G dL Au1 uR H1 dL Au2 uR H2 + h.c.,

(37)
where we have introduced the coupling matrices

Aui = ei2 g1u O1(i+1) + g3u ei3 O3(i+1) ,

Adi = ei3 g1d O1(i+1) + g2d ei2 O2(i+1) , i = 1, 2.
After performing a rotation to the quark mass eigenbasis we obtain

2
v3
u
i2
Ai = V e
Du O1(i+1) + Gu3 ei3 O3(i+1) O1(i+1) ,
v1
v1

2
v2
d
i3
d i2
O2(i+1) O1(i+1) ,
Ai = V e
Dd O1(i+1) + G2 e
v1
v1
(38)
(39)
(40)
(41)
with Gd2 , Gu3 defined in Eqs. (25) and (31), respectively.

It is clear that in order to analyze the charged Higgs contributions to the relevant flavourchanging processes we need to know the structure of the unitary matrices UuL , UuR . To be
able to predict the size of such contributions, we shall assume that the up quark mass
matrix Mu is approximately given by the texture zero structure [14],
0 a 0
Mu = a b c .
(42)
0 c d
In this case
UuL UuR mu /mc
7mu /mt
mu /mc
1
7mc /mt
7mu m2c /m3t
7mc /mt ,
1
(43)
where
7
d mt
.
mc
(44)
Such a choice is of course in agreement with our previous assumption of the matrices UuL
and UuR being close to the identity matrix.
276
Under the above conditions, the coupling matrix Gu3 defined in Eq. (34) takes the simple
form
7 mu mc /mt
7mu /mt
7mu /mt
Gu3 = gt 7 mu mc /mt
(45)
7m /m
7mc /mt .
c t
7mu /mt
7mc /mt
1
In the context of our model, where the hierarchy v b t is expected among the
VEVs, Eqs. (20), (40) and (41), together with (45) and (35), yield
0
0
0
2 ib
V e
Au1
(46)
0 mc (1 7 et ) 7mc mt eit ,
v
i
i
t
t
0 7mc mt e
mt (1 e
)
3
d

2 it
e
Vik (Dd )kj mb V3k
V3j eib ,
A1 ij
(47)
v
k=1
Au2
Au1 ,
Ad2
Ad1 ,
(48)
after the corresponding identification v1 = v, v2 = b , v3 = t and 2 = b , 3 = t . In

particular, this implies that the contributions to flavour-changing processes coming from
the charged Higgs H2+ will be strongly suppressed, provided that the Higgs mass mH +
2
mH + . In what follows we assume that the latter condition is satisfied. Moreover, we shall
1
discuss two limiting cases: 7 mu /mc 0, which corresponds to b mc in Eq. (42), and
7 1, i.e., b mu 0.
4. New physics and K , mBd , mBs , mD

Within the SM, the CKM matrix is constrained by unitarity and experimental data. These
constraints are usually expressed in terms of the Wolfenstein parameters A, and [15],
and presented as a unitarity triangle in the complex plane (,
)
(see Fig. 1 below) [1,16].
They can be summarized as follows [17,18]:
From semileptonic K and B decays we have 3
|Vus | = = 0.2205 0.0018,
|Vub | = (3.56 0.56) 103 ,
|Vcb | = 0.040 0.002,

|Vcb |
A = 2 = 0.826 0.041,
which implies

1 Vub
2
2
= 0.39 0.07,
Rb + =
Vcb
(49)
(50)
with = (1 2 /2), = (1 2 /2). The above results are extracted from tree
level decays with large branching ratios and therefore their determination is essentially
independent of physics beyond the SM.
3 All our input parameters are taken from [1,17].
277
Fig. 1. Constraints on the plane (, ) coming from the measurements of |Vub /Vcb | (dashed), K
(solid), ;mBd (dot-dashed) and ;mBs (dotted) within the Standard Model. The dot-filled area
corresponds to the presently allowed region.
Next, for the CP violating parameter K (and assuming K ),

0
K 0 |Heff (;S = 2)|K
,
(51)
2mK
0 mixing in SM gives
the calculation of the box diagrams describing the K 0 K

SM
K Im(t ) Re(c ) 1 S0 (xc ) 3 S0 (xc , xt ) Re(t )2 S0 (xt ) ,
= ei/4 C B
K
K)
ei/4 Im(M12
,
K
2 ;mK
C =
K
M12
=
G2F m2W fK2 mK
= 3.84 104,
6 2;mK
i = Vis Vid ,
xi =
m2i
m2W
(52)
Comparing this result with the experimental value |K | = (2.280 0.013) 103 , one
obtains a constraint in the form of the hyperbola

K = 0.226.
(1 )A
2 2 S0 (xt ) + Pc (7) A2 B
(53)
In the above formulas, Pc (7) = 0.3 0.05 summarizes the charmcharm and charmtop
K = 0.80 0.15 is a nonperturbative parameter, the correction
contributions in the SM, B
factors 1 = 1.38 0.20, 2 = 0.57 0.01, 3 = 0.47 0.04 describe the short-distance
QCD effects, fK = 160 MeV is the kaon decay constant, mK = 497.672 0.031 MeV
is the kaon mass and ;mK = (3.489 0.008) 1012 MeV is the mass difference in
the K system. The gauge independent functions S0 which govern the FCNC processes are
approximately given by

1.52
mt
S0 (xt ) = 2.46
,
S0 (xc ) = xc ,
170 GeV

3xt2 ln xt
3xt
xt
S0 (xc , xt ) = xc ln
(54)
,
xc 4(1 xt ) 4(1 xt )2
where mc = 1.30 0.05 GeV and mt = 165 5 GeV correspond to the running quark
pole
masses defined as mq mq (mq ).
278
Substituting the numerical values for the parameters in Eq. (53) the K constraint reads

+ 0.16
+ 0.15
(1 )
0.91
(55)
0.14 + (0.31 0.05) = 0.41 0.10.
0
0 systems is given in the
B
Next, the amplitude for the ;B = 2 transition in the Bd,s
d,s
SM by
0
0
Bq Heff (;B = 2)
Bq

2
SM
= q Vt b Vtq , q = d, s,
M12 (Bq ) =
2mBq
G2F 2
Bq S0 (xt ),
m B mBq fB2q B
12 2 W
so that the mass differences are

SM

(Bq ) = 2q (Vt b Vtq )2 .
;mBq = 2M12
q =
(56)
(57)
Here B = 0.55 0.01 is a QCD correction coefficient, mBd = 5.28 GeV and mBs =
1/2 measures the hadronic
5.37 GeV are the B-meson masses and the factor fBq B
Bq
1/2

uncertainties. Recent lattice QCD estimates give fBd B
= 200 40 MeV and s
Bd
1/2 )/(fBd B
1/2 ) = 1.14 0.08.
(fBs B
Bs
Bd
Combining the experimental value ;mBd = 0.471 0.016 ps1 with Eq. (57) we can
determine the parameter

|Vt d |
1 Vt d
0.040
+ 0.37
2
2
=
+ =
Rt (1 )
(58)
= 0.98
0.22.
Vcb
8.8 103
|Vcb |
On the other hand, the measurement ;mBs > 12.4 ps1 allows us to determine Rt in a
different way, namely,

1 Vt d s mBs ;mBd
+ 0.15
< 1.03
Rt =
(59)
=
0.14 .
Vt s mBd ;mBs
Fig. 1 summarizes the constraints given by Eqs. (50), (55), (58) and (59) in the plane (,
)
within the SM. The dot-filled area corresponds to the presently allowed region if no new
physics beyond the SM is invoked.
The model of interest to us and presented in Section 2 contains new physical fields when
compared to the SM. As the masses of such fields are expected to be much larger than
mW , their contributions to charged current tree level decays should be negligible. They
can however significantly contribute to quantities such as K and ;mBd,s , thus playing
an important role in the determination of the unitarity triangle. To establish their impact,
first we shall compute the new contributions to K , ;mBd and ;mBs coming from the
FCNC processes induced by the heavy neutral Higgs field. Then we shall compare these
contributions with the ones induced by the new heavy charged Higgs fields. As we shall
see, if the mass scale for the heavy charged Higgs (mH + ) is of the same order than the
scale for the heavy neutral Higgs (mH 0 ), new physics contributions are always dominated
by tree-level FCNC effects. Finally, new contributions to the ;mD mass difference are
also expected and they are discussed at the end of this section.
279
4.1. FCNC contributions

Since the couplings of down quarks to the neutral scalar fields R and I are flavourviolating (cf. Eqs. (23) and (24)), they will induce a tree-level FCNC contribution to K 0
0 mixing. In the framework of our model such couplings are determined by Eqs. (24),
K
(35) and given by
v 2 + b2
d
2 mb ib
d
ib
e
e V3i V3j ,
G2 ij
ij =
(60)
v
v
where we have used the fact that v b , mb gb b / 2. To estimate the hadronic matrix
elements it is customary to use the so-called vacuum insertion approximation [16]. In this
approximation, the new physics contribution to the matrix element M12 of the transition
0 will be given by
K 0 K
new 0
K 0 |Heff
|K
2mK

2
K m2
2 1
f 2 mK B
mK
b
+
V
= e2ib K
V
,
t
s
td
6
md + ms
4v 2 m2H 0
new
M12
(K) =
(61)
new
is the effective ;S = 2 Hamiltonian induced by the neutral Higgs exchange.
where Heff
It is now straightforward to compute the FCNC contribution to the CP-violating
parameter K defined in Eq. (51). We obtain

m2
H0
K
= C(0) |Vus |2 |Vcb |4 (1 )2 2 sin 2b + 2(1 ) cos 2b 2b , (62)
mH 0
where
C(0)

K 1 mK 2
mK fK2 B
GeV 2
+
=
,
6.8 1012
md + ms
v
4 2 ;mK v 2 6
(63)
with ms (mc ) = 130 MeV, md (mc ) = 8 MeV. Substituting the central values |Vus | =
0.2205, |Vcb | = 0.040, mb = 4.25 GeV in Eq. (62) and assuming b /2, we find

GeV 2 TeV 2
H0
30.6(1 )
.
K
(64)
v
mH 0
A lower bound on the scale of the heavy neutral Higgs can be then obtained by requiring
H0| <
the new physics contribution (64) to be smaller than the SM contribution, i.e., |K
SM |. Since for the central values of the parameters, the contribution to in the SM (cf.
|K
K
Eq. (52)) is approximately given by
SM
5.2 103 (1.34 ),
(65)
K
we find for 0 0.3,

GeV
mH 0 65 TeV
.
v
(66)
280
In particular, for v =
2 mc 1.84 GeV we obtain
mH 0 35 TeV.
(67)
New FCNC contributions induced by the heavy neutral Higgs field H 0 in the mass
differences ;mBq (q = d, s) are easily obtained from the previous results on K . For the
new (B ) we have
matrix elements M12
q

2 m2b
new
M12
(Bq ) = e2ib q(0) Vtq Vt b
,
m2H 0
2
Bq mBq 1 m
fB2q B
Bq
(0)
q =
+
,
4v 2
6
mq + mb
(68)
with

(0)
d 0.09 GeV
GeV
v
2
,
s(0) 0.12 GeV
GeV
v
2
.
(69)
This implies
new

2
0
M (Bd ) = 2 (0)|Vus |2 |Vcb |2 (1 )2 + 2 mb
;mH
=
2
Bd
12
d
m2H 0

GeV 2 TeV 2
3.84 102 ps1 (1 )2 + 2
,
v
mH 0
2
new

0
(0)
2 mb

;mH
Bs = 2 M12 (Bs ) = 2s |Vcb |
2
mH 0

2
TeV 2
4 1 GeV
1.02 10 ps
.
v
mH 0
(70)
(71)
These FCNC contributions to ;mBq are to be compared with the SM contributions given
by Eq. (57) and which can be approximately written as

1
(1 )2 + 2 ,
;mSM
Bd 0.48 ps
1
;mSM
Bs 13.04 ps .
(72)

q(0) m2b
GeV 2 65 TeV 2
=
0.19
0.19,
q m2 0
v
mH 0
(73)
We have then
0
wq
;mH
Bq
;mSM
Bq
if the lower bound given in Eq. (66) for mH 0 is satisfied. We see that the contributions to
;mBq coming from the neutral Higgs are much smaller than the SM ones. In Fig. 2 we
illustrate our results for mH 0 = 35 TeV and v = 2 mc . We notice that while K is quite
0 mixing practically do not
sensitive to new physics, the constraints coming from B 0 B
change when compared to the SM results.
281
Fig. 2. Constraints on the plane (, ) after including the new FCNC contributions
induced by
the heavy neutral Higgs H 0 . The curves are given for mH 0 = 35 TeV, v = 2 mc and we
assume mH + mH + mH 0 . The dot-filled area corresponds to the region allowed by the present
1
2
experimental bounds.
4.2. Charged-current contributions

The main contributions to flavour-changing processes induced by the charged Higgs
field H1+ are described by box diagrams where H1+ and the W gauge boson are circulating
inside the box. 4 Their computation is also straightforward.
0 system we find
For the new physics contribution to the amplitude M12 in the K 0 K
K
2 GF m2W mK fK2 B
new
M12 (K) =
24 2 v 2
m2

t t t ft + c cc fc + (ct ct + t c t c )fct 2t ,
mH +

3 ln xt
xt
3
ft =
+
,
fc = 2xc (1 + ln xc ),
1 + ln xt
2
xt 1 (xt 1)2

1 x +
2xc ln xc 3 ln xt
+ ln H ,
+
fct = xc xt
xt
2 xt 1 2
xt
ij = Vis Vj d , i = c, t,
ij = gid gj s ,
(74)
with i and xi defined in Eq. (52). Moreover, according to Eq. (46),

it 7mc
gt = Vc e
+ Vt 1 eit ,
mt

mc
7mc
1 7 eit
Vt eit
, = d, s, b.
gc = Vc
mt
mt
(75)
4 As discussed at the end of Section 3, the contributions coming from the charged Higgs field H + can be
2
neglected if mH + mH + (cf. Eq. (48)).

2
1
282
Therefore, the charged Higgs contribution to K reads

+
H
K
= C(+) [At ft + Ac fc + Act fct ]
m2t
,
m2H +
(76)
where

At = Im t t t ,

Ac = Im c cc ,
Act = Im t c t c + ct ct ,
2

K
GF m2W mK fK2 B
9 GeV

0.93
10
.
C(+) =
v
24 2 v 2 ;mK
(77)
When the texture parameter 7 = 0, the coefficients Ai in Eq. (76) are approximately given
by
Ac 0,
At 4(1 cos t )|Vcb |2 (1 )J,
mc
Act 2 (1 cos t )J,
J = |Vus |2 |Vcb |2 ,
mt
while for 7 = 1 they are
mc
At Ac J
,
mt
Act 2J (1 )
mc
.
mt
(78)
(79)
We can now give a numerical estimate of the above contributions. Using the fact that
cos t 1 m2b /(2m2t ) we obtain

GeV 2 TeV 2
H+
0.016(1 )
,
K
(80)
7=0
v
mH +

GeV 2 TeV 2
H+
K
(81)

2.55(23.6
+
)
.

7=1
v
mH +
H + SM
< , where SM is given in Eq. (65), then we can find the
If we require that K
K
K
following lower bounds on the charged Higgs mass for 0 0.3,

GeV
+
for 7 = 0,
mH 1.5 TeV
(82)
v

GeV
for 7 = 1.
mH + 101 TeV
(83)
v
This in particular implies for v 2 mc ,

mH + 800 GeV if 7 = 0,
mH + 55 TeV if 7 = 1.
(84)
0 B
0 mixing. Their
Let us now consider the charged Higgs contributions to Bd,s
d,s
0
0
system. We obtain
computation is analogous to the one in the K K
(q)
m2
(q)
(q)
(q)
new
M12
(Bq ) = q(+) At t ft + Acc fc + Act + At c fct 2t ,
mH +
2
2
Bq
2 GF mW mBq fBq B
(q)
,
Aij = Vib Vjq giq gjb , q = d, s,
q(+) =
24 2 v 2
(85)
with

GeV 2
d(+) 0.95 104 GeV
,
v
s(+) 1.26 104 GeV
283
GeV
v
2
,
(86)
fi and gij defined in Eqs. (74) and (75), respectively.

For 7 = 0 the amplitude (85) is dominated by the top-quark contributions described by
(q)
the coefficient At t . We have
2

2 m2b
(q)
.
At t 2 Vt b Vtq (1 cos t ) Vtq
m2t
(87)
(q)
The topcharm terms proportional to At c will, however, give an important contribution to

the amplitude in the case of 7 = 1. In the latter case we find for the relevant coefficients:

mc
(q)
2
mb mc
Vt b 1 eit
iVtq Vcq
,
At t Vtq Vcq
mt
mt mt
2 2 mc 2 mc
(q)
At c Vcq
(88)
Vt b
Vcq
.
mt
mt
Therefore, the new contributions to the mass differences ;mBq will be given by

+
;mH
Bd

m2
(+)
= 2d |Vus |2 |Vcb |2 (1 )2 + 2 |ft | 2b
7=0
mH +
2

GeV
TeV 2
1.65 ps1 (1 )2 + 2
,
v
mH +
2

m2b
TeV 2
H+
(+)
2
1 GeV
;mBs
= 2s |Vcb | |ft | 2 22.4 ps
.
7=0
v
mH +
mH +
(89)
(90)
Comparing these values with the SM result given in Eqs. (72) we arrive at
+
;mH
Bq
;mSM
Bq
GeV
1.55
v
2
1.5 TeV
mH +
2
.
(91)
We see that for mH + close to the lower bound (82) the charged Higgs contributions to
0
0 mixing are of the same order of magnitude than the SM ones. A slightly higher
B
Bd,s
d,s
+
SM
value of mH + is required if we impose ;mH
Bq < ;mBq . From Eq. (91) it follows then
mH + 1 TeV for 7 = 0,
(92)
with v 2 mc .
In the case of 7 = 1 we have

mb mt
mc
mc
(+)
H+
2
;mBd
= 2d |Vus | iVcb (1 + i)
ft +
fct 2
7=1
mt
mb
mH +

GeV 2 TeV 2

mH +
1

+ i(1 )
,
140 ps 1.8 + 0.7 ln
TeV
v
mH +
(93)
284
Fig. 3. Constraints on the plane (, ) after including the new flavour-changing charged contributions
induced by the heavy charged Higgs H1+ . The curves are given for mH + = 2 TeV, v = 2 mc and
1
the parameter 7 = 0. We assume mH + mH + mH 0 .
1
Fig. 4. As in Fig. 3, but taking the parameter 7 = 1 and mH + = 80 TeV.

1

+
;mH
Bs

mb mt
mc
mc
= 2s(+)iVcb
ft +
fct 2
7=1
mt
mb
mH +

GeV 2 TeV 2
mH +
3.8 103 ps1 1.8 + 0.7 ln
.
+ i
TeV
v
mH +
(94)
Comparing these contributions with the SM result (72) we find:

+
;mH
Bq
;mSM
Bq
0.15,
for mH + 55 TeV and v
(95)
2 mc as given by the bound (83).
285
In Fig. 3 we illustrate the constraints on the (, )-plane for mH + = 2 TeV, v =

2 mc and the parameter 7 = 0. We assume mH 0 mH + and thus, only the contribution
coming from the flavour-changing charged current is taken into account. The dot-filled area
corresponds to the allowed region. A similar plot is given in Fig. 4 for the case 7 = 1 and
mH + = 80 TeV.
4.3. FCNC and ;mD
0 mixing is perhaps one of the most interesting processes
In the up quark sector, D 0 D
15 GeV. To estimate
given that this process is highly suppressed in the SM: ;mSM
D < 10
the size of FCNC contributions to the mass difference ;mD we use the expression (29)
and the approximate form (43) for the matrices UuL , UuR . The relevant couplings are then
given by
iju =
b t it u
2mb mt it u
G3 ij
G3 ij ,
e
e
2
v
v2
(96)
where Gu3 is given in Eq. (45) and we have used mb gb b / 2, mt gt t / 2, gb,t 1,

to approximate the right-hand side of Eq. (96).
Keeping the dominant term we obtain
new
M12
(D) = 7 2 e2it u(0)
m2b
,
m2H 0
D mD 1 mD 2
4mu mc fD2 B
u(0) =
+
,
6
mu + mc
v4

GeV 4
3
(97)
5.4 10 GeV
,
v

D 225 MeV [19], mD = 1.86 GeV and we take mu 5 MeV. Therefore,
where fD B

m2
;mD = 2M12 (D) = 27 2 u(0) 2b
mH 0

GeV 4 65 TeV 2
2
11
4.6 7 10
GeV
.
v
mH 0
(98)
Using the bound on mH 0 coming from K physics (cf. Eq. (67)), one obtains the upper limit
;mD 1.357 2 1011 GeV.
(99)
We notice that except when 7 1, our model predicts a value for ;mD much larger
than in the SM. This is a clear signature of the model.
Comparing the above value with the experimental limit
(;mD )exp < 5 1014 GeV,
(100)
we find an upper limit on the parameter 7,

7 0.06,
(101)
286
which in turn translates into constraints on the texture form (42) assumed here for the
up-quark mass matrix Mu .
At this point it is worth recalling that flavour-changing contributions induced by new
physics crucially depend on the specific patterns of the fermion mass matrices. In our
analysis we have assumed a simple but quite generic ansatz for the up-quark mass matrix,
expressed in terms of quark mass ratios and a free parameter 7. This parameter is therefore
expected to be constrained in order to avoid dangerous large contributions to low-energy
observable effects. It is clear that more specific mass matrix textures (e.g., triangular-type
textures in the up-quark sector) could lead to a natural suppression of these contributions
and the above constraints thus be avoided.
5. Electric dipole moment of the neutron

The electric dipole moment (EDM) of a fermion is defined as
iFD (k 2 ) (p1 )5 (p2 )F ,
where F is the electromagnetic tensor and k = p2 p1 . In the present model, we
can expect a large contribution to the EDM of the neutron induced by the spontaneous CP
violation if the phases t,b = 0, . In the latter case, the exchange of the heavy neutral and
the heavy charged Higgs fields are expected to contribute to the EDM of the neutron. For
the down quark contribution with exchange of a heavy neutral Higgs the dominant term is
given by (f = d, u)

FD (0)
f f xf (xf2 1 2xf ln xf )
|
|Q
f

,
(102)
e 16 2 Im 31 13
(xf 1)3 mH 0
f
where xd,u m2b,t /m2H 0 , Qu = 2/3, Qd = 1/3, ijd and iju are defined in Eqs. (60)
and (96), respectively. Using the fact that mb gb b / 2, mt gt t / 2 and assuming

as before gb,t 1, we have
2 mb ib
2 mb ib
d
d
e Vt d ,
e Vt d ,
31
(103)
13
v
v
2mb 7mu mt it
u
u
= 13

e
.
31
(104)
v2
Since xf 1 and sin 2b,t 2mb /mt , Eq. (102) are approximately given by

FD (0)
m4b
GeV 2 65 TeV 2
30
1 |Vt d |2

8 10
cm
,
(105)
e
v
mH 0
12 2
v 2 mt m2H 0
d

4
3
FD (0)
65 TeV 2
7 mu mb mt 7 1023 cm GeV

(106)
,
e
3 2 v 4 m2 0
v
mH 0
u
H
after substituting the values mu 5 MeV, mb 4.25 GeV, mt 165 GeV and |Vt d |
0.01.
287
Thus, using the lower bound on mH 0 given by Eq. (67), we obtain the following upper
bounds on the electric dipole moments of the quarks:

FD (0)
FD (0)
24
8 1030 cm,

(107)
cm.
e
e 37 10
d
u
Taking into account that mu /mc 103 7 1 and assuming

FD (0)

FD (0) + FD (0) ,

e
e
e
n
d
u
(108)
we conclude from Eqs. (107) that the EDM will be always dominated by the up-quark
contribution. Moreover,

FD (0)
27
1024 cm

cm
10
(109)
e n
in the allowed range of the parameter 7. Of course, the above conclusions hold for our
specific choice of the up-quark mass matrix texture (42). If such is the case, then the
predicted EDM is expected to be very close to the present experimental limit [1]

FD (0)
26

cm.
(110)
e < 10
exp
The comparison of Eqs. (107) with the experimental bound (110) allows us to further
constrain the parameter 7. We find
103 7 3 103 ,
(111)
0 mixing (see
which is more restrictive than the upper bound previously found from D 0 D
Eq. (101)). Notice, however, that the constraint (111) was obtained assuming the lower
bound on the Higgs mass given by Eq. (67) and implied by K . We could of course relax
this constraint by pushing the heavy neutral Higgs mass to a higher scale, but then the
raison dtre of our model would be lost and its predictions would be close to the SM
ones. From a phenomenological point of view we find more plausible to fix the Higgs
scale from the constraints coming from K and B physics and, in particular, from the CPviolating parameter K . Other constraints, such as the EDM of the neutron, will then give
us a hint on what kind of quark mass matrix textures are favoured in the theory.
Let us now consider the charged Higgs contributions. In this case the dominant
contribution is coming from the diagrams with a top quark circulating inside the loop.
The photon line can be attached to the Higgs line or to the quark line. Therefore, we have

FD (0)
d u
1

e + 16 2 Im A1 31 A1 13
H

xu (xu2 1 2xu ln xu )
QH
(xu 1)3

xu (xu2 4xu + 3 + 2 ln xu )
1
(112)
+ Qu
,
3
(xu 1)
mH +
288
with xu now defined as xu m2t /m2H + , QH = 1; Au1 and Ad1 given by Eqs. (46) and
(47), respectively. Since
u

i 2 ib
2 ib
it
A1 13
(113)
e V t d mt 1 e
e V t d mb ,

v
v
d
2 i(t b )
A1 31
(114)
e
V t d mb ,
v
we obtain

2
2
FD (0)
TeV 2
3 mb mt |Vt d |2 2 1022 cm GeV

.
(115)
e + 8 2 v 2 m2
v
mH +
H
H+
Imposing the experimental constraint given in Eq. (110), we get a stronger constraint on
mH + than the lower bound (82), namely,

GeV
+
.
mH 140 TeV
(116)
v
In particular, for v 2 mc we obtain

mH + 75 TeV.
(117)
We also remark that in leading order this bound is independent of the texture parameter 7.
6. FCNC in top decays: the example t q
The FCNC decays t q and t qZ are strongly suppressed in the SM at the level
of 1012 . Observation of any of these events would be an indication of physics beyond the
SM [20]. It is of particular interest to study the order of magnitude that our model predicts
for such processes. The amplitude of t q can be parametrized as

q
t (p2 )A ,
1 ) i(A + B5 )
M q(p
(118)
mt
where A is the photon field and q p2 p1 . The decay width of this process is given
by

mt 2
|A| + |B|2 ,
(t q ) =
(119)
8
and is dominated by the process t bW ,

m4W
m2W
m2W
GF
2 3
(t bW ) = |Vt b | mt 1 2
(120)
1+ 2 2 4 .
mt
mt
mt
8 2
Once the coefficients A and B are known, it is straightforward to compute the branching
ratio corresponding to the process t q . Recently, an experimental limit on this process
has been reported [21]
B(t q ) < 0.032,
(121)
289
which can be translated into a limit on the parameters A and B,

|A|2 + |B|2 < 6.5 103 .
(122)
This limit should improve with the future LHC, which is expected to decrease the above
bound by two orders of magnitude [22]
B(t q ) < 104 ,
(123)
|A|2 + |B|2 < 2 105 .
(124)
i.e.,
In our model one expects that a such process will be induced by one loop diagrams as is
the case for the EDM. Moreover, both contributions, with a heavy neutral Higgs exchange
and with a charged Higgs exchange, should be taken into account. One expects then using
the bounds given by Eqs. (67) and (117),
AB
m2b
m2H 0
108
(125)
for the neutral Higgs exchange, while

AB
m2t
5 106
m2H +
(126)
for the charged Higgs exchange.

Since the above bounds are further multiplied by additional suppression factors coming
from the CKM matrix, we can conclude that the prediction of the present model for the
process t q is out of the reach of the next future colliders.
7. Implications on CP asymmetries
As discussed in Section 4, if mH + mH 0 then the flavour-changing charged Higgs
contributions to ;mBd,s are negligible. In order to study the implications of the model on
CP asymmetries, we shall assume that new physics appears only through tree-level FCNC
effects.
q0 mixing is by
An easy way to parametrize the effects of new physics on Bq0 B
2
introducing the parameters rq and the phases 2q through the relation
SM
new
SM
M12 (Bq ) = M12
+ M12
rq2 e2iq M12
(Bq ).
On the other hand, from Eqs. (56) and (68) we have

SM

M12 (Bq ) = 1 wq e2ib M12
(Bq ),
(127)
(128)
with wq defined in Eq. (73). Comparing Eqs. (127) and (128) we find the relations
wq sin 2b
rq2 = 1 2wq cos 2b + wq2 ,
(129)
tan 2q =
.
1 wq cos 2b
290
0 ) KS decays is related to the

Within the SM, the CP asymmetry aKs in Bd0 (B
d
angle of the unitarity triangle,

Vcd Vcb
aKs = sin 2, = arg
(130)
.
Vt d Vtb
While global analyses of the CKM unitarity triangle yield the values
0.75 0.06 [23],

(sin 2)SM = 0.73 0.20 [24],
0.63 0.12 [25],
(131)
the recent experimental measurements of the above time dependent CP asymmetry give
0.34 0.20 0.05 (BABAR) [26],

(sin 2)KS = 0.58 0.32 0.09 (BELLE) [27],
(132)
0.79 0.42
(CDF)
[28].
The above experimental values imply the average
(sin 2)KS = 0.46 0.16.
(133)
Although the SM estimates are consistent with the present experimental results, the small
values of sin 2 found by BABAR and BELLE collaborations might indicate the presence
of new physics contributions.
If the new physics modifies the phase of the mixing amplitude, then the asymmetry will
also get a contribution from the d phase,
aKs = sin 2( + d ).
(134)
Moreover, if we assume that the term in Lagrangian (6) is the only source of CP
violation in our model, then b /2 + mb /mt . In this case Eq. (129) imply
rd
1 + wd ,
tan 2d
2wd mb
.
1 + wd mt
(135)
We see that the new phase d is suppressed by the ratio mb /mt . Using the upper bound
wd 0.19 given by Eq. (73) we find
rd 1.1,
rs rd ,
tan 2d 0.008.
(136)
Although these predictions are consistent with the global average (133), we notice that
the deviations from the SM predictions are very small in this case and, consequently, it
is not possible to achieve consistency [18] with the small values reported by the BABAR
collaboration [26].
In order to illustrate the dependence of our result on the strong CP phase , let us assume
that CP violation in the CKM matrix is independent of the value of . In other words,
let us assume that besides the angle , there exist other sources of CP violation and we
consider as an arbitrary parameter. Of course, when is different from /2, the isospin
symmetry between the top and bottom quarks has to be explicitly broken in the effective
Lagrangian (3). From the minimization of the potential (cf. Eq. (10)) and taking mb mt ,
291
one easily finds

mb
mb
,
t
.
mt
mt
At first order in mb /mt we obtain in this case:
rd = 1 2wd cos 2 + wd2 ,

tan 2d =
b +
(137)
wd sin 2
.
(138)
1 wd cos 2
It is interesting to study how the angle d varies as a function of . The extrema values
for d are obtained when cos 2 = wd and this implies
wd
tan 2d =
.
rd = 1 wd2 ,
(139)
1 wd2
With wd 0.19 as given by Eq. (73) we have then
rd 0.98,
| tan 2d | 0.19.
(140)
Thus, if the strong CP phase is assumed to be a free parameter of the model, the
constraint (136) on the new physics contribution to the CP asymmetries in B decays
is relaxed and we can reach a rather sizeable phase d 6 . This in turn would allow
to accommodate [18] the present experimental measurements, including the small values
obtained by BABAR collaboration.
8. Conclusion
In this paper we have studied the phenomenological constraints on a model where CP
violation is dynamically induced by a strong CP phase [11]. The most promising tests
for the model are given by the new experimental prospects to measure ;MD or to improve
the experimental limit on the electric dipole moment of the neutron.
Contrary to naive expectations, the fact that the new force responsible for the
electroweak symmetry breaking and for the spontaneous CP violation is only sensitive
to the third generation of quarks does not imply that the most stringent constraints come
from processes involving the heavy flavours (t and b). We have shown that the stringent
constraints on the scale of new physics come from K physics and from the electric dipole
moment of the neutron. This means that even if FCNC processes are naturally suppressed
in the model by the CKM matrix elements, this suppression is not strong enough to allow
for a mass scale of the heavy Higgs to be of the order of few TeV.
Acknowledgements
We are grateful to L.T. Handoko, F. Krger and A. Teixeira for useful discussions and
comments. One of us (R.G.F.) would like to thank the Fundao para a Cincia e a
Tecnologia for financial support under the grant SFRH/BPD/1549/2000. One of us (D.D.)
would like to thank the DESY theory group where part of this work was completed.
292
References
[1] D.E. Groom et al., Review of Particle Physics, Eur. Phys. J. C 15 (2000) 1.
[2] Y. Nambu, G. Jona-Lasinio, Phys. Rev. 122 (1961) 345.
[3] A. Miransky, M. Tanabashi, K. Yamawaki, Mod. Phys. Lett. A 4 (1989) 1043;
A. Miransky, M. Tanabashi, K. Yamawaki, Phys. Lett. B 221 (1989) 177.
[4] W.A. Bardeen, C.T. Hill, M. Lindner, Phys. Rev. D 41 (1990) 1647.
[5] For a recent review on top-quark condensation and an extended list of references see: G. Cvetic,
Rev. Mod. Phys. 71 (1999) 513.
[6] C.T. Hill, Phys. Lett. B 345 (1995) 483.
[7] D. Kominis, Phys. Lett. B 358 (1995) 312.
[8] G. Buchalla, G. Burdman, C.T. Hill, D. Kominis, Phys. Rev. D 53 (1996) 5185.
[9] D.E. Clague, G.G. Ross, Nucl. Phys. B 364 (1991) 43.
[10] D. Delpine, J.-M. Grard, R. Gonzlez Felipe, Phys. Lett. B 372 (1996) 271.
[11] D. Delpine, J.-M. Grard, R. Gonzlez Felipe, J. Weyers, Phys. Lett. B 411 (1997) 167.
[12] E. Witten, Ann. Phys. 128 (1980) 363.
[13] G.C. Branco, W. Grimus, L. Lavoura, Phys. Lett. B 380 (1996) 119.
[14] G.C. Branco, D. Emmanuel-Costa, R. Gonzlez Felipe, Phys. Lett. B 477 (2000) 147.
[15] L. Wolfenstein, Phys. Rev. Lett. 51 (1983) 1945.
[16] G.C. Branco, L. Lavoura, J.P. Silva, CP Violation, International Series of Monographs on
Physics, Vol. 103, Clarendon Press, Oxford, 1999.
[17] A.J. Buras, Lectures given at the 14th Lake Louise Winter Institute, February 1420, 1999,
hep-ph/9905437.
[18] G. Eyal, Y. Nir, G. Perez, J. High Energy Phys. 08 (2000) 028.
[19] L. Lellouch, C.-J. David Lin, UKQCD Collaboration, hep-ph/0011086.
[20] J.A. Aguilar, G.C. Branco, Phys. Lett. B 495 (2000) 347.
[21] F. Abe et al., CDF Collaboration, Phys. Rev. Lett. 80 (1998) 2525.
[22] Beneke et al., in: G. Altarelli, M.L. Mangano (Eds.), Proceedings of the 1999 CERN Workshop
on Standard Model Physics (and more) at the LHC, CERN, Geneva, Switzerland, May 2526,
1999, hep-ph/0003033.
[23] F. Caravaglios, F. Parodi, P. Roudeau, A. Stocchi, Talk given at the Third International
Conference on B physics and CP violation, National Taiwan University, Taipei, December 37,
1999, hep-ph/0002171.
[24] A. Ali, D. London, Eur. Phys. J. C 9 (1999) 687;
A. Ali, D. London, in: Proceedings of the Workshop: Physics and Detectors for DAPHNE,
Frascati, November 1619, 1999, hep-ph/0002167.
[25] S. Plaszczynski, M.-H. Schune, Talk given at Heavy Flavours 8, Southampton, UK, 1999, hepph/9911280.
[26] B. Aubert et al., BABAR Collaboration, Phys. Rev. Lett. 86 (2001) 2515.
[27] A. Abashian et al., BELLE Collaboration, Phys. Rev. Lett. 86 (2001) 2509.
[28] T. Affolder et al., CDF Collaboration, Phys. Rev. D 61 (2000) 072005.

Dualised -models at the two-loop order

Guy Bonneau, Pierre-Yves Casteill
Laboratoire de Physique Thorique et des Hautes Energies, Unit associe au CNRS UMR 7589,
Universit Paris 7, 2 Place Jussieu, 75251 Paris Cedex 05, France
Received 2 April 2001; accepted 3 May 2001
Abstract
We adress ourselves the question of the quantum equivalence of non-abelian dualised -models
on the simple example of the T-dualised SU(2) -model. This theory is classically canonically
equivalent to the standard chiral SU(2) -model. It is known that the equivalence also holds at the
first order in perturbations with the same functions. However, this model has been claimed to be
non-renormalisable at the two-loop order. The aim of the present work is the proof that it is at
least up to this order still possible to define a correct quantum theory. Its target space metric being
only modified in a finite manner, all divergences are reabsorbed into coupling and fields (infinite)
renormalisations. 2001 Elsevier Science B.V. All rights reserved.
PACS: 11.10.Gh; 11.10.Kk; 11.10.Lm
Keywords: Sigma models; T-duality; Renormalisation
1. Introduction
The subject of classical versus quantum equivalence of T-dualised -models has been
strongly studied in recent years, and extensive reviews covering abelian, non-abelian
dualities and their applications to string theory and statistical physics are available [13].
More recent developments on the geometrical aspects of duality can be found in [4].
The interpretation of T-duality as a canonical transformation, for constant backgrounds,
was first given by [5,6]. Its more general formulation [7] was applied to the non-abelian
case in [8,9].
After the settling of the classical equivalence, the most interesting problem was its study
at the quantum level. This was done mostly for dualisations of Lie groups, with emphasis
put on SU(2). For this model the one-loop equivalence was established in [10,11]. This
one-loop quantum equivalence was recently settled for the general class of models built on
GL GR /GD , with an arbitrary breaking of GR [12]. An interesting intermediary result
E-mail addresses: bonneau@lpthe.jussieu.fr (G. Bonneau), casteill@lpthe.jussieu.fr (P.-Y. Casteill).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 2 1 6 - 4
294
G. Bonneau, P.-Y. Casteill / Nuclear Physics B 607 (2001) 293304
is an expression for the Ricci tensor of the dualised geometry (with torsion) exhibiting its
dependence with respect to the geometrical quantities of the original model. In the same
work, the two-loop renormalisability problem was tackled and the need for extra (nonminimal) one-loop order finite counter-terms was emphasized. Some years ago, it was
noted that in the minimal dimensional scheme, two-loop renormalisability does not hold
for the SU(2) T-dualised model [13].
The aim of the present work is a more precise analysis of this two-loop (in)equivalence
for the non-abelian T-duality, still on the simple example of the original SU(2) T-dualised
model.
The main remark is that, part of the isometries being somehow lost, the T-dualised
models are not as they should be if one wants to give an all-order analysis defined
by a sufficient system of Ward identities. For example, in our simple case there is, a priori
only a linear SU(2) [or O(3)] invariance, and any O(3) invariant action is allowed (let us
remind the reader that in higher-loop corrections to a classical action, all the terms which
are not prohibited by some reason such as power counting, isometries or conservation
laws . . . , would appear). To our present knowledge, the extra constraints coming from the
origin of the model (dualisation of an (SU(2)L SU(2)R )/SU(2)D chiral model) are not
understood. 1 As it is highly probable that they are linked with the spacetime dimension,
it is not surprising that a minimal dimensional renormalisation scheme fails: as is well
known, when the regularization method does not respect all the properties that define the
theory, extra finite counter-terms are needed [15].
The content of this article is the following: in Section 2 we recall the expression of
the classical action of the dualised theory and set the notations. In Section 3, we start
from the corresponding a priori quantum bare action and obtain through h expansion the
possible counter-terms that may be added to the classical action in order to reabsorb the
divergences. Then in Section 4 we give the 2-loop divergences and in Section 5 we discuss
how they match with the candidates in Section 3. Our result is that coupling constant and
field renormalisations (infinite and finite ones) are not sufficient to ensure the two-loop
existence of the T-dualised theory but the metric itself has to be deformed (in a finite way).
Some concluding remarks are offered in Section 6.
2. The classical action

At the classical level and in light-cone co-ordinates, the dual action can be writen
[10,12]:

1
Gij + i j ,
S=
1 In [14] the quantisation of a U(1)-invariant non-linear model, the so-called Complex sine-Gordon model,
was performed by imposing as extra constraints its classical property of factorisation and non-production; there it
was shown that definite extra finite one-loop counter-terms are needed to enforce this property to one-loop order
and then they also restore the two-loop renormalisability.
295
where gij = G(ij ) is the target space metric and hij = G[ij ] is the torsion potential. The
torsion Tij k is defined by Tij k = 32 [i hj k] . The connections with torsion jik and without
torsion jik respectively write:
1
jik = g is (j Gks + k Gsj s Gkj ) = jik + Tjik ,
2
1 is
i
j k = g (j gsk + k gsj s gj k ),
2
and the corresponding covariant derivatives are:
Di kj = i kj ijs ks = i kj Tijs vs ,
j
Di k j = i k j + is k s = i k j + Tis v s .
The Riemann tensor without torsion will be noted Rij,kl whereas we will denote the one
with torsion as R ij,kl .
The expression of the dualised target space metric Gij as a function of the original
one is well known and in [12] the various geometrical quantities (Ricci tensor, . . .) were
also related. In the special case considered here, where the original model is the SU(2)
SU(2)/SU(2) non-linear model, the metric writes:
=
Gij []
1
1 + 2

ij + i j + ij k k ,
(1)
where is a SU(2) (real) vector representation and the i , i = 1, 2, 3, are the co-ordinates
on the dualised manifold. Then 2 is a SO(3) invariant and the symmetry is linearly
realised. Torsion breaks parity, but the model is invariant under the simultaneous change
and ij k ij k . Let us emphasize that no other local symmetry exists for that
model.
3. The two-loop order bare action

In order to analyse the two-loop renormalisability of the dualised SU(2) -model, we
first examine all the possible ways to reabsorb the divergences through local counter-terms.
As usual, we allow for finite and infinite renormalisations of both fields and coupling. But,
as we shall see later on, this appears as insufficient to reabsorb the various divergences.
Thus, we also allow for a finite deformation of the classical metric and torsion potential
gij + hij = Gij to describe its quantum extension: of course, this la Friedan [16]
extension of the notion of renormalisability involves a priori an infinite number of new
parameters. Let us emphasize that we shall consider only finite deformations.
Even if by doing so we obviously introduce too many parameters, we first let them all
independent in order to show the announced need for such intrinsic metric deformation.
Let us first write the bare action:

1
S o = o Goij + oi oj ,
(2)
296
where:

h
1
h 1
1
2
c
+
b
+
+
d
+
,
=
+
1
+
o
2
2
2

h
w
2 ()
v 2 ()
v 1 ()
+ , (3)
+ h
+w
1 ()
+
x

(
)
o = +
+
2
2
ij + .
ij + h G
Goij = Gij + G
2
2
To express (2) we shall need the Lie derivative L and a second order Lie derivative
k
L(2) . Indeed, for any tensor Sij defined on a manifold with co-ordinates j , in a change of
k
co-ordinates:
+ i j ,
Sij0 ( 0 )+ oi oj = Sij ()
and if 0 = + k (note that k is not a vector field on the manifold):

1

= Sij0 ()
L Sij0 ()
+ 2 L(2) Sij0 ()
+ O 3 .
Sij ()
2
k
k
(4)
We remind the reader that

s
s
s
L(Sij ) = k s Sij + Ssj i k + Sis j k .
k
(5)
One can show that

L(2) (Sij ) = L L(Sij ) L (Sij ).
k
k k
k s s k
With i gj k = 0, we rewrite Eqs. (5), (6) for Sij Gij as:
i = 2k l Gli ,
L(Gij ) = 2Dj ki + [i j ] ,
(2)
si,j u + 2Di k s Dj ks 4Tius k u Dj k s
L (Gij ) = 2k s k u R

k
+ L (Gij ) + [i j ] .
(6)
(7)
(k s k u su )
i is some quantity whose computation is useless as, in the same manner as i , it gives a
vanishing contribution to the action or, the torsion potential being always defined up to a
gauge transformation, such term can always be put into hij (moreover, in our particular
situation, the O(3) symmetry implies that such [i j ] terms vanish). Then, we shall not
write them anymore.
Then, expending (2) with the help of (3), (4), one gets the possible counter-terms at
lowest orders:
h
:
0 order in
2
1
Gij + i j ;
first order in
1
h
:
2
1
+ b Gij +

i
j

(G
)
+
G
L
ij
ij + ;
297
(8)
v 1
1
+w
h
:
2

1 1
1
ij ) + 1 L (Gij ) + b L (Gij )
( ) +
Gij + L (G
2

v 1
w 1
v 1

2
1 (2)
Gij + L (Gij ) + L (Gij ) + ( ) + i j ,
+
1
w
2
2 v 1 +w 1
at second order in
(9)
where Q| 1 means that we only take the term in
in the expression Q.
As we dont consider the 3-loop order, in expression (9) we only need the coefficient of
h 2
1
(the double poles 2 are not new quantities as they are directly related to first order
simple poles and it has already been proved that the dualised SU(2) -model is one-loop
renormalisable [10]).
Using the following identity between Lie derivatives:
LLLL = L
Y
X

Y X
with
Z i = Xj j Y i Y j j Xi ,
the term with the second order Lie derivative may be re-expressed:

(2)
L (Gij ) = L L (Gij ) + L L (Gij )
(Gij )
L
v 1
1
k w
k v
v 1 w
1
w
1 v 1
(v

+w

)
+
w

k
k
1
1
1
1
1

= 2 L L (Gij ) L (Gij ) .
v 1 w
1
v1k k w
1
So, the O(h ) term (8) may be rewritten as:

1 1

1 Gij + L(Gij ) + L (Gij ) + bGij + Gij + i j ,

v 1
w
1
and the O(h )2 term (9) as:

1
ij + (2 b1)Gij
1 L (Gij ) + bGij + G
w
1

ij +
+ L L (Gij ) + bGij + G
(Gij ) .
L
v 1 w
1
(w
2 v1k k w
1)
As a consequence, as expected, any term L (Gij ) + bGij may be reabsorbed into the finite
w
1
ij (and vice-versa) to the expense of a change in the O(h )2 parameters:

deformation G

ij + L (Gij ) + bGij G
ij 2 2 b1 , w 2 w 2 v k k w 1 .
G
(10)
1
w
1
298
h 2
Finally, for the term in 1 ( 2
) in the bare action, one has the following expression:

1
ij ) + 2 Gij + L (Gij ) + Hij (v1 , w1 ) ,
1 Gij + L(G
v 1
2
W
where

2 = w
W
2 + b v 1 + 1 w
1 + v1s w1u su ,
is,uj + Di v s Dj w1s 2Tius v u Dj ws + (
v1 w
1 ).
Hij (v1 , w1 ) = v1s w1u R
1
1
1
(11)
(12)
4. The two-loop order divergences

We use the expression of the covariant divergences given by Hull and Townsend [17], 2
in the background field method and in the minimal dimensional scheme, up to the two-loop
order:
Div1 = h Ric ,
ij
ij
2
(13)

2
h klm
k T lmn R
lmkj + 2Tmn
kij l .
Rklmj 12 R
Div2ij = 8
2 Ri
In order to ensure the renormalizability of the theory, these divergences should match
with the candidate counter-terms given by (8) and (11):

h
(Gij ) ,
CTij = 2 1 Gij + vL
1
(14)
2

h
2 =
CT
G
+
(
G
)
+
G
+
(G
)
+
H
(v
,
w
)
.
L ij
ij
2 ij
ij 1
1
ij 4 2 1 ij vL
1
2
W
It has been previously proven [12] that the dualised metric is quasi-Einstein as soon as the
original metric is Einstein. In our special case, we get:

1
1 1 2
Ric ij = Gij + 2Dj vi , = 1 = , v = v 1 =
(15)
.
2
2 1 + 2
The addition to the effective action of a h finite deformation of the metric and of some
finite renormalisations for the coupling and fields (non-minimal scheme) modifies the h 2
divergences. The additional term is easily obtained as

h
h
kl Ricij (Gkl )
Ricij Gkl +
L (Gkl ) + bGkl + G
2
2 w 1

h 2
2ij + O h 3 .
2
4
ij + L (Gij ) + bGij appears. Then, we could decide

Here also, only the combination G
w
1
ij , but, as announced at the beginning of Section 3, in

to reabsorb bGij and L (Gij ) into G
w
1
order to see if they would be sufficient by themselves, we keep them apart in a first step.
2 We checked for our example that the two other calculations in [18,19] give the same result.
299
Finally, the dualised SU2 -model will be renormalisable at two loops if and only if we
ij [],
b, w
2 , W
2 []}
such that:
can find {G
1 [];
h 2
2ij
4 2

h 2

+
+
(
G
)
+
G
+
(G
)
+
H
(v
,
w
)
= 0.
G
L
L
1 ij
ij
2 ij
ij
ij 1
1
4 2
v 1
2
W
Div2ij
(16)
5. Results
According to the linearly realised symmetry of the T-dualised SU(2) -model, the finite
2 () respectively write:
ij and the vectors w
1 () and W
deformation of the metric G
ij = ( )ij + ( ) i j + ij k ( ) k ,
G

w
1 = w1 ( ),
2 = W2 ( ),

W
where = 2 . Moreover, the symmetry also implies that terms of the form [i kj ] or of the
t are equal to zero. It is then possible to re-express (16) as a set of three linear
form k s K u su
differential equations :
W2 ( ) +
=
(1 + )2 45 + 68 18 2 12 3 3 4
1
+
w1 ( )
3
2
(1 + )2
16(1 + )
4 + 5 + 6 2 + 3
3 + 10 + 5 2 + 2 3
( ) +
( )
4(1 + )
4(1 + )
3(1 + )(3 + )
4 + 11 + 5 2 3
( )
( ) + ( )
2
2
2

2
(1 + )(3 + ) ( ) (1 + ) ( ),
32
3(5 + 60
+ 10 2
+ 12 3
(17)
+ 3 4 )
8(1 + )4
(7 + 10 )
(12 + 5 )
=
( ) +
( ) 3(11 + 5 ) ( )
2
2
+ (17 22 + 9 2 ) ( ) + (5 + 4 + 2 ) ( ) 2(5 + 2 )(3 + 5 ) ( )
+ 2(5 19 12 2 + 3 ) ( ) + 2 ( ) 4 (1 + )(3 + ) ( )
4 (1 + )2 (3) ( ),
2 +
3(1 )(13 + 6
(18)
+ 2)
8(1 + )3
(6 + )
(5 + 2 )
( ) +
( ) (17 + 3 ) ( )
=
2
2
+ (7 + )(3 + ) ( ) 2(5 + 6 + 2 ) ( )
2 (3 + ) ( ) + 4 ( ).
(19)
300
ij immediately appears: setting both ( ), ( ) and

The need for a true deformation G
( ) to zero, Eqs. (18) and (19) cannot be satisfied, even if 3 we allowed for some finite
2 ()
both hidden into the vector W

renormalisations of the coupling (b) and field (w
1 ()),
(see Eq. (12)). Then, as first proven in [13], we have checked that:
In a purely dimensional scheme (even with non minimal subtractions), the dualised
SU(2) model is not renormalisable at the two-loop order.
So, from the discussion in the previous sections, and without restricting the generality
as vanishing quantities.
of our analysis, one can take b and w
1 ()
Remarks.
As 2 is not a function, but a constant, differentiating Eqs. (18) and (19) will relate
ij , the finite one loop renormalisation, has
( ), ( ) and ( ). Then, as soon as G
2 ()

been definitely set, Eq. (17) will give the infinite two-loop renormalisations W
and 2 .
ij will be fixed up to some b Gij +
From the previous discussions, we know that G
L (Gij ); it is then natural to use this freedom, for example to reabsorb ( ), and to

W ()
ij such that:
redefine G
ij ,

ij = b Gij + L ()(G
G
ij ) + G

W

W ()
= W ( )
with
+ )
b(1
(1 + )2
ij = (
) i j + ij k ( ) k
( )
G
W ( ) =
2
2
with
(3 + )W
b
.
1+
(1 + )2
(20)
) and ( ), Eqs. (17), (18), (19)
We know that, when expressed as functions of (
remain unchanged, up to the substitutions discussed in Section 3 (Eq. (10)):
=
2(2 + )W
b
4W
1+
(1 + )2
and =
b
2 2 = 2 + ,
2
1
1
W 2 ( ) = W2 ( ) + b
+ W ( )
W2 ()
2(1 + ) 2

1
+
W ( ) + 2 W ( ) .
2(1 + )
(21)
3 One notices also that the parameters b and w

1 do not appear in (18) and (19). So, the existence of some
solution to this set of differential equations is independent of the finite renormalisations of both coupling and
fields, as is usual in perturbation theory. This freedom corresponds to a change of renormalisation scheme. This
absence is only true if we take the very vector v 1 (15) that reabsorbs the divergences at the one-loop order:
would appear in (18) and (19). This is a check of a correct renormalisation at the one-loop
otherwise, w
1 ()
order.
301
) as a function of ( ) which itself satisfies a nonEqs. (18) and (19) give (

homogeneous linear fourth order differential equation:
b + 22 3(1 )(13 + 6 + 2 ) 2(17 + 3 )
=
( )
+
6+
6+
4(6 + )(1 + )3
4(5 6 2 )
8
(22a)
( )
( ),
6+
6+
(6 )(7 + ) (3)
1260 276 91 2 + 3 3 + 4
(4) ( ) +
( ) +
( )
(6 + )
4 2 (6 + )2

120 + 254 + 57 2 + 3 3
138 + 25 + 2
b + 22
+
(
)
(
)
2
8 2 (6 + )2
8 2 (6 + )2
)
(
3(6402 8681 5856 2 22 3 + 390 4 + 39 5 )

.
64 2 (1 + )5 (6 + )2
Note that under the change

b + 22
,
( ) = ( )
2

) 3(b + 22 ) ,
B( ) = (
(22b)
(23)
the parameter b and the constant 2 disappear from the set (22). Then, 2 being an
unknown constant, the general solution of the differential equation (22b) will be
( ) = ( ) + c,
where c is an arbitrary constant,
and the two-loop coupling constant renormalisation 2 will be:

b
2 = c .
2
The model will be renormalisable up to two loops iff equation (22b), where ( ) has
been replaced by ( ) according to (23), has a solution which is analytic near =
0. In order to reach such a conclusion, we use the method of Frobenius for linear
differential equations [20]. = 0 is a regular singularity (notice that we are only interested
in 0). The indicial equation of the linear differential equation (22b) around the
singular point 0 has four different solutions: = 32 , 12 , 0, 1. For each one, we can

n
find convergent series
n=0 cn that are independent solutions of the homogeneous
equation associated to (22b). We give here the first terms of such series (it happens that for
= 32 we have an exact solution):

35 2
1
1
1
11
,
1 ( ) = 1 +
+ ,
3 ( ) = 3 +
2
2
20
6
108
20
1 2
23 2
1
+ ,
+ .
0 ( ) = 1 +
(24)
1 ( ) = 1 +
840
42
324
Then, we use the method of variation of parameters to find 3 ( ), 1 ( ), 0 ( ) and
2
2
1 ( ) such that
( ) = 3 ( ) 3 ( ) + 1 ( ) 1 ( ) + 0 ( )0 ( ) + 1 ( )1 ( )
2
302
is the general solution of the inhomogeneous equation (22b) where ( ) has been replaced
by ( ) according to (23).
The first terms in the expansion of these functions are:

7
1067 13691
+ ,
3 ( ) = o 3 + 2
2
1680
3780
2

5
1067 2543509
o
2
1 ( ) = 1 +
+
+ ,
2
240
100800
2
1067 2 9805 3

+ ,
0 ( ) = o0 +
192
288
27887 2
1067
+
+ .
1 ( ) = o1
(25)
480
5760
The analyticity requirement near = 0 enforces the choice o
32
= o
12
= 0; ( ) is
). The final
then expressed as a convergent series in , and the same will be true for (
o
o

expression for the deformation Gij depends on 3 constants [c, 0 and 1 ] and an arbitrary
function [W ( )] and is given by the three functions:
2W
b
+
,
1+
(1 + )2
2(2 + )W
b
+
+ 4W
( ) = 6c +
1+
(1 + )2
( ) =
4(5 6 2 )
3(1 )(13 + 6 + 2 ) 2(17 + 3 )
(
)
( )
+
4(6 + )(1 + )3
6+
6+
8
( ),
6+
(3 + )W
b
+
+ ( ).
( ) = c +
(26)
1+
(1 + )2
+
We now use the up to now free parameter b to reabsorb the parameter c. Let us define
b = b 2c,
( ) = W ( ) + c(1 + ),
W
we get
ij = G
ij + bG
ij + L Gij
G

W
=W

( ) and
with W

ij
ij = G
G
W ( )0 .
Eq. (26) for c=b=
The dualised SU(2) -model is therefore renormalisable at the two-loop order if and
only if we add a finite h deformation of the classical metric, depending on two new
parameters o0 and o1 .
303
6. Concluding remarks
We have been able to exhibit some set of counter-terms that ensures the two-loop
renormalisability of the T-dualised chiral non-linear model. The one-loop effective
metric is defined up to two constants (o0 and o1 ), and some finite arbitrary field
and coupling renormalisations. As is well known (e.g., in [21]), the two-loop Callan
Symanzik function (related to 2 ) 4 depends on these finite counterterms.
We emphasize that, contrarily to D. Friedans approach to models quantisation,
where the classical metric receives infinite perturbative deformations, our candidate for the
deformation of the classical metric is a finite one, depending on only two parameters (plus
the usual infinite, and finite, renormalisations of the fields and of the coupling constant):
our ansatz is that a proper understanding of the dualisation process will precisely offer
the extra constraints that uniquely define the quantum extension of the classical theory,
order by order in perturbation theory, in the same spirit as Ward identities determine what
otherwise would appear as new parameters (see also footnote 1).
Note added in proof

For completeness, let us mention that for abelian T-duality similar works were achieved
in Refs. [22,23].
Acknowledgements
It is a pleasure to thank Galliano Valent whose interest in that subject was really
stimulating for us.
References
[1] E. Alvarez, L. Alvarez-Gaum, J.L.F. Barbn, Y. Lozano, Nucl. Phys. B 415 (1994) 71, hepth/9309039.
[2] E. Alvarez, L. Alvarez-Gaum, Y. Lozano, Nucl. Phys. Proc. Suppl. 41 (1995) 1, hepth/9410237.
[3] A. Giveon, M. Porrati, E. Rabinovici, Phys. Rep. 244 (1994) 77, hep-th/9401139.
[4] O. Alvarez, Nucl. Phys. B 584 (2000) 659;
O. Alvarez, Nucl. Phys. B 584 (2000) 682, hep-th/0003177/0003178.
[5] A. Giveon, E. Rabinovici, G. Veneziano, Nucl. Phys. B 322 (1989) 167.
[6] T. Curtright, C. Zachos, Phys. Rev. D 49 (1994) 5408, hep-th/9401006.
4 The two loops quantities and W
2 are fixed as:
2
2 =
b
,
2
2 obtained through (21).

W
Notice that the normalisation condition b = 0 (no h extra finite coupling constant renormalisation) enforces
2 = 0.
304
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
E. Alvarez, L. Alvarez-Gaum, Y. Lozano, Phys. Lett. B 336 (1994) 183, hep-th/9406206.

Y. Lozano, Phys. Lett. B 355 (1995) 165, hep-th/9503045.
K. Sfetsos, Phys. Rev. D 54 (1996) 1682, hep-th/9602179.
B.E. Fridling, A. Jevicki, Phys. Lett. B 134 (1984) 70.
E.S. Fradkin, A.A. Tseytlin, Ann. Phys. 162 (1985) 31.
P.Y. Casteill, G. Valent, Nucl. Phys. B 591 (2000) 491, hep-th/0006186.
J. Balog, P. Forgcs, Z. Horvth, L. Palla, Nucl. Phys. Proc. Suppl. 49 (1996) 16, hepth/9601091;
A. Subbotin, I.V. Tyutin, Int. J. Mod. Phys. A 11 (1996) 1315, hep-th/9506132.
G. Bonneau, F. Delduc, Nucl. Phys. B 250 (1985) 581.
G. Bonneau, Int. J. Mod. Phys. A 5 (1990) 3831.
D. Friedan, Ann. Phys. 163 (1985) 1257.
C.M. Hull, P.K. Townsend, Phys. Lett. B 191 (1987) 115.
R.R. Metsaev, A.A. Tseytlin, Phys. Lett. B 191 (1987) 354.
D. Zanon, Phys. Lett. B 191 (1987) 363.
E.L. Ince, Ordinary Differential Equations, Dover Publications, New York, 1956, Chapter 16.
C.M. Hull, Lectures on non-linear sigma models and strings, 1986 Workshop Super Field
Theories, Vancouver, Canada; published in Vancouver Theory Workshop 1986:77, Cambridge
Preprint-87-0480.
J. Balog et al., Phys. Lett. B 388 (1996) 121, hep-th/9606187.
N. Kaloper, K. Meissner, Phys. Rev. D 56 (1997) 7940, hep-th/9705193.

General analysis of lepton polarizations in rare

B K + decay beyond the standard model
T.M. Aliev, M.K. akmak, M. Savc
Physics Department, Middle East Technical University, 06531 Ankara, Turkey
Received 12 September 2000; accepted 4 April 2001
Abstract
The general analysis of lepton polarization asymmetries in rare B K + decay is
investigated. Using the most general, model independent effective Hamiltonian, the general
expressions of the longitudinal, normal and transversal polarization asymmetries for and + and
combinations of them are presented. The dependence of lepton polarizations and their combinations
on new Wilson coefficients are studied in detail. Our analysis shows that the lepton polarization
asymmetries are very sensitive to the scalar and tensor type interactions, which will be very useful
in looking for new physics beyond the standard model. 2001 Elsevier Science B.V. All rights
reserved.
PACS: 12.60.-i; 13.20.-v; 13.25.He
1. Introduction
Rare B meson decays, induced by flavorchanging neutral current (FCNC) b s(d)
transitions, are quite promising for establishing new physics beyond the standard model
(SM). In particular the flavor changing inclusive b s(d)+ decay, which takes place
in the SM at loop level, is very sensitive to the gauge structure of the SM. Moreover
b s(d)+ decay is known to be very sensitive to the various extensions of the SM.
New physics effects manifest themselves in rare B meson decays in two different ways,
either through new contributions to the Wilson coefficients existing in the SM or through
the new structures in the effective Hamiltonian which are absent in the SM. Note that
b s(d)+ transition has been extensively studied in framework of the SM and its
various extensions [115]. One of the efficient ways in establishing new physics beyond
the SM is the measurement of the lepton polarization [1524]. All previous studies for
the lepton polarization have been limited to SM and its minimal extensions, except the
E-mail addresses: taliev@metu.edu.tr (T.M. Aliev), savci@metu.edu.tr (M. Savc).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 8 0 - 8
306
T.M. Aliev et al. / Nuclear Physics B 607 (2001) 305325
work [19]. In [19] the analysis of the lepton polarization for the inclusive b s +
decay was presented in a model independent way. In the same work, it was also shown that
the investigation of the lepton polarization can give unambiguous information about the
existence of the scalar and tensor type interactions.
It is well known that the theoretical study of the inclusive decays is rather easy but their
experimental investigation is difficult. However for the exclusive decays the situation is
contrary to the inclusive case, i.e., their experimental detection is very easy but theoretical
investigation has its own drawbacks. This is due to the fact that for description of the
exclusive decay form factors, i.e., the matrix elements of the effective Hamiltonian between
initial and final meson states, are needed. This problem is related to the nonperturbative
sector of the QCD and and it can only be solved in framework of the nonperturbative
approaches.
These matrix elements have been studied in framework of different approaches, such as
chiral theory [25], three-point QCD sum rules [26] and light cone QCD sum rules [27,28].
In this work we will use the weak decay form factors calculated using light cone QCD sum
rules method [27,28].
The aim of the present work is to present a rigorous study of the lepton polarizations
in the exclusive B K + decay for a general form of the effective Hamiltonian
including tensor type interactions as well. Note that we concentrate only on the
and leptonic modes because of the following two reasons. Firstly, the electron
polarization is hard to measure experimentally, and secondly, it is well known that in
the SM the normal PN and transversal PT polarizations are both proportional to the
lepton mass and hence their measurements might be possible especially in the +
channel. In the present work we extend results of previous studies on lepton polarization
[2023] and then perform a general analysis (in a model independent way in the sense
without forcing concrete values for the Wilson coefficients corresponding to any specific
model) including all possible form of interactions. Our analysis shows that the socalled tensor type interactions give dominant contribution to the lepton polarization
asymmetries.
The paper is organized as follows. In Section 2, using a general form of four-Fermi
interaction we derive the model independent expressions for the longitudinal, transversal and normal polarizations of leptons. In Section 3 we investigate the dependence
of the above-mentioned polarizations on the four-Fermi interactions. We also present
the combined analysis of the and + asymmetries and our results in this section.
2. Lepton polarizations
We start this section by computing the lepton polarization asymmetries, using the most
general, model independent four-Fermi interactions. Following [19], we write the effective
Hamiltonian for the b s+ transition in terms of twelve model independent fourFermi interactions.
307
G
q
+ CBR si q Rb

Heff = Vt s Vtb CSL s i 2 Lb
q
q2
2
tot
tot
+ CLL
sL bL L L + CLR
sL bL R R + CRL sR bR L L
+ CRR sR bR R R + CLRLR sL bR L R + CRLLR sR bL L R

+ CLRRL sL bR R L + CRLRL sR bL R L + CT s b

,
+ iCT E s b
(1)
where the chiral projection operators L and R in (1) are defined as

1 + 5
1 5
,
R=
,
2
2
and CX are the coefficients of the four-Fermi interactions. The first two of these
coefficients, CSL and CBR , are the nonlocal Fermi interactions which correspond to
2ms C7eff and 2mb C7eff in the SM, respectively. The following four terms in this
expression are the vector type interactions with coefficients CLL , CLR , CRL and CRR .
tot
tot
and CLR
do already exist in the SM in
Two of these vector interactions containing CLL
eff
eff
combinations of the form (C9 C10 ) and (C9 + C10 ). Therefore by writing
L=
tot
CLL
= C9eff C10 + CLL ,
tot
CLR
= C9eff + C10 + CLR ,
tot and C tot describe the sum of the contributions from SM and the
one concludes that CLL
LR
new physics. The terms with coefficients CLRLR , CRLLR , CLRRL and CRLRL describe the
scalar type interactions. The remaining two terms leaded by the coefficients CT and CT E ,
obviously, describe the tensor type interactions.
Having the general form of four-Fermi interaction for the b s+ transition, our
next problem is calculation of the matrix element for the B K + decay. In other
words, we need the matrix elements
K |s (1 5 )b|B,
K |s i q (1 5 )b|B,
K |s (1 5 )b|B,
K |s b|B,
in order to calculate the decay amplitude for the B K + decay. These matrix
elements are defined as follows:
K (pK , )|s (1 5 )b|B(pB )

2V (q 2 )
i (mB + mK )A1 q 2
mB + mK

A2 (q 2 )
2mK
i(pB + pK ) ( q)
iq 2 ( q) A3 q 2 A0 q 2 , (2)
mB + mK
q
= pK
q
308
K (pK , )|s i q (1 5 )b|B(pB )

2

2i m2B m2K (pB + pK ) ( q) T2 q 2

= 4 pK
q T1 q

q2
2i( q) q (pB + pK ) 2
(3)
T3 q 2 ,
2
mB mK
K (pK , )|s b|B(pB )

2
= i 2T1 q 2 (pB + pK ) + 2 m2B m2K T1 q 2 T2 q 2 q
q

2
2
2
2
4
q
T
2 1 q T2 q 2
T3 q ( q)pK q ,
(4)
q
mB m2K
where q = pB pK is the momentum transfer and is the polarization vector of K
meson. In order to ensure finiteness of (2) at q 2 = 0, we assume that A3 (q 2 = 0) =
A0 (q 2 = 0) and T1 (q 2 = 0) = T2 (q 2 = 0). The matrix element K |s (1 5 )b|B can
be calculated from Eq. (2) by contracting both sides of Eq. (2) with q and using equation
of motion. Neglecting the mass of the strange quark we get

1
2imK ( q)A0 q 2 .
mb
In deriving Eq. (5) we have used the relation (see [15,26])

2mK A3 q 2 = (mB + mK )A1 q 2 (mB mK )A2 q 2 .
K (pK , )|s (1 5 )b|B(pB ) =
(5)
Taking into account Eqs. (1)(5), the matrix element of the B K + decay can be
written as
M(B K + )

2A pK
= Vt b Vts
q iB + iC( q)(pB + pK )
4 2

5 2E pK
+ iD( q)q +
q iF

5 [iN( q)]
+ iG( q)(pB + pK ) + iH ( q)q + [iQ(

q)] +

(iCT ) 2T1 (pB + pK ) + B6 q B7 ( q)pK q
+ 4

2T1 (pB + pK ) + B6 q B7 ( q)pK q , (6)
+ 16CT E
where
tot

V
T1
tot
A = CLL
+ CLR
+ CRL + CRR
4(CBR + CSL ) 2 ,
mB + mK
q

tot
tot
B = CLL
+ CLR
CRL CRR (mB + mK )A1
T2

4(CBR CSL ) m2B m2K 2 ,
q
tot

A2
tot
C = CLL + CLR
CRL CRR
mB + mK

1
q2
4(CBR CSL ) 2 T2 + 2
T3 ,
q
mB m2K
tot

A3 A0
T3
tot
D = 2 CLL
+ CLR
CRL CRR mK
+ 4(CBR CSL ) 2 ,
2
q
q
V
tot
tot
E = (CLR
+ CRR CLL
CRL )
,
mB + mK

tot
tot
+ CRL (mB + mK )A1 ,
F = CLR CRR CLL
tot

A2
tot
G = CLR CRR CLL
+ CRL
,
mB + mK
tot

A3 A0
tot
H = 2 CLR
CRR CLL
+ CRL mK
,
q2
mK
A0 ,
Q = 2(CLRRL CRLRL + CLRLR CRLLR )
mb
mK
N = 2(CLRLR CRLLR CLRRL + CRLRL )
A0 ,
mb

T1 T2
,
B6 = 2 m2B m2K
q2

4
q2
B7 = 2 T1 T2 2
T3 .
q
mB m2K
309
(7)
The form of Eq. (6) reflects the fact that its difference from the SM case is due to the last
four terms only, namely, scalar and tensor type interactions. The next task to be considered
is calculation of the final lepton polarizations with the help of the matrix element for the
B K + decay. For this purpose we define the following orthogonal unit vectors,
+
SL in the rest frame of and SL in the rest frame of + , for the polarization of the
leptons along the longitudinal (L), transversal (T ) and normal (N ) directions:

p

,
SL 0, eL = 0,
|p |

p p

SN 0, eN = 0,
,
|p p |

ST 0, eT = 0, eN eL ,

p+
+
SL 0, eL+ = 0,
,
|p+ |

p p+
+
,
SN 0, eN+ = 0,
|p p+ |

+
ST 0, eT+ = 0, eN+ eL+ ,
(8)
where p and p are the three momenta of and K meson in the center of mass (CM)
frame of the + system, respectively. The longitudinal unit vectors SL and SL+ are
boosted to CM frame of + by Lorentz transformation,

|p | E p
,
SL,CM =
,
m m |p |
310

+
SL,CM =

|p |
E p
,
,
m
m |p |
(9)
while vectors of perpendicular directions are not changed by boost.

The differential decay rate of the B K + decay for any spin direction n () of the
()
, where n () is the unit vector in the () rest frame, can be written in the following
form

() ()

d (
n() ) 1 d
() ()
() ()
(10)
1 + PL eL + PN eN + PT eT
n () ,
=
2
2
dq
2 dq 0
where the superscripts and + correspond to and + cases, (d /dq 2)0 corresponds
to the unpolarized differential decay rate, and PL , PN and PT represent the longitudinal,
normal and transversal polarizations, respectively. The unpolarized decay rate in Eq. (10)
can be written as

G2 2
d
=
|Vt b Vts |2 1/2 (1, r, s)v,
(11)
dq 2 0 214 5 mB
2 2ab 2bc 2ac is the usual triangle function, s =
where (a, b, c) = a 2 + b2 + c
q 2 /m2B , r = m2K /m2B and v = 1 4m2 /q 2 is the lepton velocity. The explicit form of
is presented in the appendix.
The polarizations PL , PN and PT are defined as:
()
()
()
()
d
d
n

n

=
e

=

e

2
2
i
i
dq
dq
() 2
q =
Pi

,
()
d
d
() = e

n

+
n() = ei()
i
dq 2
dq 2
where P () represents the charged lepton () polarization asymmetry for i = L, N, T ,
i.e., PL and PT are the longitudinal and transversal asymmetries in the decay plane,
respectively, and PN is the normal component to both of them. With respect to the direction
of the lepton polarization, PL and PT are P -odd, T -even, while PN is P -even, T -odd and
CP-odd. After lengthy calculations for the longitudinal polarization of the () , we get

8
4
m
()
PL = m2B v
Re F + m2B (1 r)G + m2B sH Q m4B s Re(AE )
r
3

1
Re B m2B (1 r s)G ( + 12rs)F
3r

+ C m2B (1 r s)F 2 m4B G
1
256 2
m2B s Re(QN )
mB m Re([CT E CT E A ]T1 )
2r
3

4
+ m2B m Re m2B B7 + 2(1 r s)B6 4(1 + 3r s)T1
3r

4
[CT C 4CT E G ] + m Re m2B (1 r s)B7 2( + 12rs)B6
3r

+ 4[ + 12r(1 r)]T1 [CT B 4CT E F ]
+
311
16 2
mB Re(CT CT E ) 2 m4B s|B7 |2 4s( + 12rs)|B6 |2
3r
+ 4m2B s(1 r s) Re(B6 B7 ) 8m2B s(1 + 3r s) Re(B7 T1 )

2
2
+ 16s[ + 12r(1 r)] Re(B6 T1 ) 16 (4r + s) + 12r(1 r) |T1 |
.
+
(12)
Further calculations lead the following expressions for the transverse polarization PT() :

mB s
()
8m2B m Re(AB )
PT =

1 2
m m (1 r) (1 r s) Re(BG ) m2B Re(CG )
rs B

1
1
m2B m Re(F C ) m2B Re B(1 r s) m2B C N
rs
2r

1 4
1 2 2
mB m Re(CH ) + mB v Re F (1 r s) m2B G Q
r
2r
2

1
m (1 r s) Re mB sH F B
rs

8
m Re 2m F m2B sH 2m2B m (1 r)G m2B sN
rs

m2B (B7 CT E ) 2(1 r s)(B6 CT E )

16m4B v 2 Re [sCT B6 + 2(1 r)CT T1 ]E

2 2 v 2 Re [sCT E B6 2(1 r)CT E T1 ]A

32
m2B m (1 + 3r s) Re [2m (1 r)G + 2m sH + sN](CT E T1 )
rs

16
m2B 4r 1 v 2 (1 + 7r s) Re(F CT E T1 )
r

+ 32m2B 2 v 2 Re(BCT T1 )

2
2

2
+ 2048mB m Re(CT T1 B6 CT E ) (1 r)|T1 | Re(CT CT E ) .

(13)
s
Finally for normal asymmetries we get

()
3
PN = mB v s 4 Im m BE + m AF 16BCT E T1 + 8CT T1 F

128m 4|CT E |2 |CT |2 B6 T1

1
m2B Im 2m H G GN + CQ + 16m B7 CT E Q
2r
1
+ (1 + 3r s)m Im(GF + 32CT E T1 Q )
r

1
+ (1 r s) Im m H F 12 NF + 12 QB + 16m QB6 CT E
r
312

+ 16m2B s Im AB6 CT + 2B6 CT E E

+ 32m2B (1 r) Im A CT T1 + 2ECT E T1 .
()
()
(14)
()
Concerning expressions PL , PT and PN few remarks are in order. The difference

between PL and PL+ results from the scalar and tensor type interactions. Similar situation
()
takes place for the normal polarization PN of leptons and antileptons. In the m 0
+
limit, the difference between PT and PT is due to again existence of new physics, i.e.,
scalar and tensor type interactions. For these reasons the experimental study of PL() , PN()
()
and PT can give essential information about new physics. Note that similar situation
takes place for the inclusive channel b s+ (see [19]).
Combined analysis of the lepton and antilepton polarizations can also give very useful
hints in search of new physics, since in the SM PL + PL+ = 0, PN + PN+ = 0 and PT
PT+ 0, which follows from Eqs. (12)(14). Therefore a measurement of the nonzero value
of the above-mentioned combined lepton asymmetries in experiments can be considered
as an unambiguous indication of the discovery of the new physics beyond the SM.
3. Numerical analysis
The input parameters we used in our analysis are: |Vt b Vts | = 0.0385, 1 = 129, GF =
1.17 105 GeV2 , B = 4.22 1013 GeV, C9eff = 4.344, C10 = 4.669. It should
be noted here that the above-value of the Wilson coefficient C9eff we have used in our
numerical calculations corresponds only to short distance contribution. In addition to
the short distance contribution C9eff also receives long distance contributions associated
with the real cc
intermediate states, i.e., with the J / family. In this work we restricted
ourselves only to short distance contributions. As far as C7eff is concerned, experimental
results fixes only the modulo of it. For this reason throughout our analysis we have
considered both possibilities, i.e., C7eff = 0.313, where the upper sign corresponds to
the SM prediction. The values of the input parameters which are summarized above, have
been fixed by their central values.
For the values of the form factors, we have used the results of [28], where the radiative
corrections to the leading twist contribution and SU(3) breaking effects are also taken
into account. The q 2 dependence of the form factors can be represented in terms of three
parameters as

F q2 =
F (0)
2
1 aF q 2
mB
+ bF
q 2 2 ,
m2B
where the values of parameters F (0), aF and bF for the B K decay are listed in
Table 1. Note that in the present analysis the final state Coulomb interactions of the leptons
with the other charged particles are neglected since this effect is known to be much smaller
than the averaged values of the SM (see [24]). Furthermore the final state interaction of the
313
Table 1
B meson decay form factors in a three-parameter fit, where the radiative corrections to the leading
twist contribution and SU(3) breaking effects are taken into account
ABK
1
ABK
2
V BK
T1BK
T2BK
T3BK
F (0)
aF
bF
0.34 0.05
0.60
0.023
0.28 0.04
1.18
0.281
0.46 0.07
0.19 0.03
1.55
1.59
0.575
0.615
0.19 0.03
0.49
0.241
0.13 0.02
1.20
0.098
lepton polarization for the KL + or K + + + decays is estimated to be

of the order of (m /mK ) 103 [29]. For this reason the final state interaction effect is
neglected as well.
We observe from the explicit form of the expressions of the lepton polarizations that
they all depend on q 2 and the new Wilson coefficients. Therefore it may be experimentally
difficult to study the dependence of the the polarizations of each lepton on all + center
of mass energies and on new Wilson coefficients. So we eliminate the dependence of the
lepton polarizations on one of the variables, namely q 2 , by performing integration over
q 2 in the allowed kinematical region, so that the lepton polarizations are averaged. The
averaged lepton polarizations are defined as
(mb mK )2
4m2
dB
2
Pi dq
2 dq
Pi =
.
(mb mK )2 d B
2
dq
2
2
dq
4m
(15)
We present our analysis in a series figures. Figs. 1 and 2 depict the dependence of
the averaged longitudinal polarization PL of and the combination PL + PL+ on
new Wilson coefficients, at C7eff = 0.313 for B K + decay. From these figures
we observe that PL is more sensitive to the existence of the tensor interaction, while
the combined average PL + PL+ is to both scalar and tensor type interactions. As has
already been noted, this is an expected result since vector type interactions are canceled
when the combined longitudinal polarization asymmetry of the lepton and antilepton is
considered. From Fig. 2 we see that PL + PL+ = 0 at CX = 0, which confirms the SM
result as expected. For the other choice of C7eff , i.e., C7eff = 0.313 the situation is similar
to the previous case, but the magnitude of PL + PL+ is smaller. Figs. 3 and 4 are the
same as Figs. 1 and 2 but for the B K + decay. Similar to the muon longitudinal
polarization, PL is strongly dependent on the tensor interaction coefficients CT and CT E .
It is very interesting to observe that when CT 1.5 and CT E 0.5, PL is positive
while for all other cases PL is negative.
From Fig. 4 we see that the dependence of PL + PL+ on CT is stronger. Furthermore
if the values of the new Wilson coefficients CLRRL , CLRLR are negative (positive) and CT
314
Fig. 1. The dependence of the average longitudinal polarization asymmetry PL of on the new
Wilson coefficients at C7eff = 0.313 for the B K + decay.
Fig. 2. The dependence of the combined average longitudinal polarization asymmetry PL + PL+
of and + on the new Wilson coefficients at C7eff = 0.313 for the B K + decay.
is positive (negative), so is PL negative (positive). The situation is to the contrary for the
coefficients CRLRL , CRLLR , i.e., PL +PL+ is positive (negative) when the corresponding
Wilson coefficients are negative (positive). Absolutely similar situation takes place for
C7eff > 0. For these reasons determination of the sign and of course magnitude of PL
315
Fig. 3. The same as in Fig. 1, but for the B K + decay.
and PL + PL+ can give promising information about new physics.
In Figs. 5 and 6 the dependence of the average transversal polarization PT and
the combination PT PT+ on the new Wilson coefficients, respectively, for the B
K + decay and at C7eff = 0.313. We observe from Fig. 5 that the average transversal
polarization is strongly dependent on CT , CT E , CLRRL and CRLRL and quite weakly to
remaining Wilson coefficients. It is also interesting to note that for the negative (positive)
values of the coefficients CT E and CLRRL , PT is negative (positive) while it follows
316
Fig. 5. The same as in Fig. 1, but for the average transversal polarization asymmetry PT of .
Fig. 6. The same as in Fig. 2, but for the transversal polarization asymmetry PT PT+ .
the opposite path for the coefficients CT and CRLRL . For the PT PT+ case, there
appears strong dependence on the tensor interactions CT and CT E , as well as all four scalar
interactions with coefficients CLRRL , CRLLR , CLRLR , CRLRL . The behavior of this
combined average is identical for the coefficients CLRLR , CRLRL and CLRRL , CRLLR in
pairs, so that four lines responsible for these interactions appear only to be two. Moreover
PT PT+ is negative (positive) for the negative (positive) values of the new Wilson
coefficients CT E , CLRRL and CRLLR . The situation is the other way around for the
317
coefficients CT , CLRLR and CRLRL . Remembering that in SM, in massless lepton case
PT 0 and PT PT+ 0, determination of the signs of the PT and PT PT+
can give quite a useful information about the existence of new physics. For the choice of
C7eff = 0.313, apart from the minor differences in their magnitudes, the behaviors of PT
and PT PT+ are similar as in the previous case.
As is obvious from Figs. 7 and 8, PT and PT PT+ show stronger dependence on
CT and CT E at C7eff = 0.313 for the B K + decay. Again change in signs of PT
318
Fig. 9. The dependence of the average normal asymmetry PN of on the new Wilson coefficients
at C7eff = 0.313 for the B K + decay.
and PT PT+ are observed depending on the change in the tensor interaction coefficients
CT and CT E . As has already been noted, determination of the sign and magnitude of PT
and PT PT+ are useful tools in looking for new physics.
Note that for simplicity all new Wilson coefficients in this work are assumed to be real.
Under this condition PN and PN + PN+ have non-vanishing values coming from the
imaginary part of SM, i.e., from C9eff . From Fig. 9 we see that PN is strongly dependent
on all tensor and scalar type interactions. On the other hand Fig. 10 depicts that the behavior
PN + PN+ is determined mainly by the tensor interactions only, for B K + decay.
Similar behavior takes place for the B K + decay as well, as can easily be seen
in Figs. 11 and 12. The change in sign and magnitude of both PN and PN + PN+ that
are observed in these figures is an indication of the fact that an experimental verification of
them can give unambiguous information about new physics.
In this work the sensitivity of the lepton polarizations on the new Wilson coefficients is
analyzed and their efficiency for establishing new physics beyond SM is discussed. The
branching ratio of the B K + decay depends also on the new Wilson coefficients
and it might give useful information about new physics. In this connection there comes
into mind the following question: suppose that lepton polarizations and branching ratio
of the B K + decay are measured in experiments. Can one find sizable regions
in the CX parameter space in which the decay rate would agree with the SM prediction
while the polarizations would not? A possible existence of such a region is an indication
of the fact that the new physics can be established only by the measurement of the lepton
polarizations. In order to answer this question, we present in Figs. 1317 the correlations
between branching ratio and combined average lepton polarizations. In Fig. 13 we present
the flows in the (B, PL ) plane by varying the coefficients of the new scalar, vector and
319
Fig. 10. The dependence of the combined average normal polarization asymmetry PN + PN+ of
and + on the new Wilson coefficients at C7eff = 0.313 for the B K + decay.
tensor type interactions. Fig. 14 depicts the flows in (B, PL + PL+ ) plane by varying
the coefficients of the tensor and scalar type interactions. Similarly, parametric plot of
the average transversal polarization PT of the lepton and the combined average
polarizations PT + PT+ of the and + leptons and of the branching ratio are presented
in Figs. 15 and 16, respectively. In Figs. 1317, the branching ratio is restricted to have
values in the range 107 B(B K + ) 5 107 .
320
Fig. 13. Parametric plot of the correlation between the integrated branching ratio B (in units of 107 )
and the average longitudinal lepton polarization asymmetry PL at C7eff = 0.313 as function of
the new Wilson coefficients as indicated in the figure, for the B K + decay.
The range of variation of the new Wilson coefficients are determined by assuming that
the value of the branching ratio of the B K + decay coincides with the SM prediction. Under this assumption it follows from Figs. 3 and 13 that the new Wilson coefficients
CT and CT E lie in the region 1 CT 0 and 0 CT E 1, respectively, and CRR varies
in the region 2.5 CRR 0 while rest of the remaining Wilson coefficients vary in the
321
Fig. 14. The same as in the Fig. 13, but for the combined average longitudinal lepton polarization
asymmetry PL + PL+ .
Fig. 15. The same as in the Fig. 13, but for the average transversal lepton polarization asymmetry
PT .
region between 4 and 4. The same result is obtained from an analysis of the combined
average longitudinal lepton polarizations (see Figs. 4 and 14). An analogous analysis of the
dependence of the branching ratio on the averaged transversal polarization of lepton and
combined average transversal polarization of and of + leptons lead to the same regions
of the variation of the new Wilson coefficients, as in the longitudinal polarization case.
322
Fig. 16. The same as in the Fig. 15, but for the combined average transversal lepton polarization
asymmetry PT + PT+ .
Fig. 17. The same as in Fig. 14, but for the combined average normal lepton polarization asymmetry
PN + PN+ .
Fig. 17 depicts the flows in the (B, PN + PN+ ) plane by changing the coefficients of
the tensor type interactions only, from which we conclude that the regions of variation of
the tensor interaction coefficients CT and CT E are the same as the ones obtained from from
the analysis of the longitudinal and transversal polarizations.
323
Finally we would like to discuss briefly the detectibility of the lepton polarization
asymmetries. To give the reader an idea about this possibility, we choose the averaged
values of the longitudinal, transversal and normal polarizations of the lepton, which
are approximately close to the SM prediction, i.e., PL 0.5, PT 0.6 and PN
0.02. Experimentally, to be able to measure an asymmetry Pi of a decay with the
branching ratio B at the n , the required number of events are N = n2 /(BPi 2 ). It follows
from this expression that to observe the lepton polarizations PL , PT and PN in B
K + decay at 1 level, the expected number of events are N = (1.8; 1.3; 1.2 103)
pairs that is expected to be
107 , respectively. On the other hand, The number of B B
produced at B factories is about N 5 108 . So a comparison of these two numbers
allows us to conclude that while measurement of the normal polarization of the -lepton
is impossible, measurements of its longitudinal and transversal polarizations could be
accessible at B factories.
4. Summary and conclusions

In this work we present the most general analysis of the lepton polarization asymmetries
in the rare B K + ( = , ) using the general, model independent form of
the effective Hamiltonian. The dependence of the longitudinal, transversal and normal
polarization asymmetries of + and and their combined asymmetries on the new
Wilson coefficients are studied. It is found that the lepton polarization asymmetries are
very sensitive to the existence tensor and scalar type interactions. Moreover, PT and
PN change their signs for the B K + and PL and PT change their signs for
the B K + decays, respectively, as the new Wilson coefficients vary in the region
of interest. This conclusion is valid also for the combined average polarization effects
PL + PL+ , PT PT+ and PN + PN+ for the same decay channel. It is well known
that in the SM, PL +PL+ = PT PT+ = PN +PN+ 0 in the limit m 0. Therefore
any deviation from this relation and determination of the sign of polarization is decisive
and effective tool in looking for new physics beyond SM.
Appendix A.
In this appendix we present the explicit expression for 7 which is present in the
expressions for the unpolarized differential decay rate and polarization asymmetries of
the leptons.

4
= m2B m Re F + m2B (1 r)G + m2B sH N
r

8
+ m6B s 3 v 2 |A|2 + 2v 2 |E|2
3

1
2
1
+ m4B sv 2 |Q|2 + m2B 3 v 2 |C|2 (1 r s) 3 v 2 Re(BC )
r
3
3
324

2
(1 r s) 3 v 2 + 3s 1 v 2 Re(F G ) 2s 1 v 2 Re(F H )
3

2

2
2 2
2
2
2
+ s|N| + mB s 1 v |H | + 2mB s(1 r) 1 v Re(GH )

1 2
mB ( + 12rs) 3 v 2 |B|2
3r

+ m4B 3 v 2 3s(s 2r 2) 1 v 2 |G|2 + 3 v 2 + 24rsv 2 |F |2
+

64 2
mB m Re B6 CT E ( + 12rs)B m2B (1 r s)C
r

32 4
+ mB m Re B7 CT E m2B C (1 r s)B
r

16 4 2 4
+ mB s mB |B7 |2 + 4( + 12rs)|B6 |2 4m2B (1 r s) Re B6 B7
3r

16[ + 12r(1 r)] Re B6 T1 + 8m2B (1 + 3r s) Re B7 T1

v 2 |CT |2 + 4 3 2v 2 |CT E |2

256 4
+
mB |T1 |2 |CT |2 sv 2 [ 12r(s 2r 2)] + 8r 3 v 2
3r

1024 4
+
m |T1 |2 |CT E |2 3 2v 2 (s 8r) + 12r(1 r)2 + 12r
3r B

128 2
mB m [ + 12r(1 r)] Re CT E T1 B
r

m2B (1 + 3r s) Re CT E T1 C 4m2B r Re CT T1 A .
(A.1)
+
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
W.-S. Hou, R.S. Willey, A. Soni, Phys. Rev. Lett. 58 (1987) 1608.
N.G. Deshpande, J. Trampetic, Phys. Rev. Lett. 60 (1988) 2583.
C.S. Lim, T. Morozumi, A.I. Sanda, Phys. Lett. B 218 (1989) 343.
B. Grinstein, M.J. Savage, M.B. Wise, Nucl. Phys. B 319 (1989) 271.
C. Dominguez, N. Paver, Riazuddin, Phys. Lett. B 214 (1988) 459.
N.G. Deshpande, J. Trampetic, K. Ponose, Phys. Rev. D 39 (1989) 1461.
W. Jaus, D. Wyler, Phys. Rev. D 41 (1990) 3405.
P.J. ODonnell, H.K. Tung, Phys. Rev. D 43 (1991) 2067.
N. Paver, Riazuddin, Phys. Rev. D 45 (1992) 978.
A. Ali, T. Mannel, T. Morozumi, Phys. Lett. B 273 (1991) 505.
A. Ali, G.F. Giudice, T. Mannel, Z. Phys. C 67 (1995) 417.
C. Greub, A. Ioannissian, D. Wyler, Phys. Lett. B 346 (1995) 145;
D. Liu, Phys. Lett. B 346 (1995) 355;
G. Burdman, Phys. Rev. D 52 (1995) 6400;
Y. Okada, Y. Shimizu, M. Tanaka, Phys. Lett. B 405 (1997) 297.
[13] A.J. Buras, M. Mnz, Phys. Rev. D 52 (1995) 186.
[14] N.G. Deshpande, X.-G. He, J. Trampetic, Phys. Lett. B 367 (1996) 362.
[15] T.M. Aliev, M. Savc, Phys. Lett. B 452 (1999) 318;
T.M. Aliev, A. zpineci, H. Koru, M. Savc, Phys. Lett. B 410 (1997) 216.
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
325
Y.G. Kim, P. Ko, J.S. Lee, Nucl. Phys. B 544 (1999) 64.
J.L. Hewett, Phys. Rev. D 53 (1996) 4964.
F. Krger, L.M. Sehgal, Phys. Lett. B 380 (1996) 199.
S. Fukae, C.S. Kim, T. Yoshikawa, Phys. Rev. D 61 (2000) 074015.
T.M. Aliev, M. Savc, Phys. Lett. B 481 (2000) 275.
T.M. Aliev, D.A. Demir, M. Savc, Phys. Rev. D 62 (2000) 074016.
F. Krger, E. Lunghi, Phys. Rev. D 63 (2001) 014013.
Q.-S. Yan, C.-S. Huang, L. Wei, S.-H. Zhu, Phys. Rev. D 62 (2000) 094023.
D. Guetta, E. Nardi, Phys. Rev. D 58 (1998) 012001.
R. Casalbuoni, A. Deandra, N. Di Bartolemo, R. Gatto, G. Nardulli, Phys. Lett. B 312 (1993)
315.
P. Colangelo, F. De Fazio, P. Santorelli, E. Scrimieri, Phys. Rev. D 53 (1996) 3672;
P. Colangelo, F. De Fazio, P. Santorelli, E. Scrimieri, Phys. Rev. D 57 (1998) 3186, Erratum.
T.M. Aliev, A. zpineci, M. Savc, Phys. Rev. D 56 (1997) 4260.
P. Ball, V.M. Braun, Phys. Rev. D 58 (1998) 094016.
L.B. Okun, I.B. Khriplovich, Sov. J. Nucl. Phys. 6 (1968) 598;
P. Agrawal, G. Belanger, C.Q. Geng, J.N. Ng, Phys. Rev. D 45 (1992) 2383.

A combined treatment of neutrino decay

and neutrino oscillations
Manfred Lindner a , Tommy Ohlsson a,b , Walter Winter a
a Institut fr Theoretische Physik, Physik Department, Technische Universitt Mnchen, James-Franck-Strae,
D-85748 Garching bei Mnchen, Germany

b Division of Mathematical Physics, Theoretical Physics, Department of Physics,
Royal Institute of Technology (KTH), SE-100 44 Stockholm, Sweden

Abstract
Neutrino decay in vacuum has often been considered as an alternative to neutrino oscillations.
Because nonzero neutrino masses imply the possibility of both neutrino decay and neutrino
oscillations, we present a model-independent formal treatment of these combined scenarios. For that,
we show for the example of Majoron decay that in many cases decay products are observable and
may even oscillate. Furthermore, we construct a minimal scenario in which we study the physical
implications of neutrino oscillations with intermediate decays. 2001 Elsevier Science B.V. All
rights reserved.
PACS: 13.35.Hb; 13.15.+g; 14.60.Pq; 14.80.Mz
Keywords: Neutrino decay; Neutrino oscillations; Majoron models
1. Introduction
The most favorable alternative for fast neutrino decay is [1,2] to introduce an effective
decay Lagrangian which couples the neutrino fields to a massless boson carrying lepton
number, i.e., a complex scalar field or a Majoron field [36]. One possibility for such a
Lagrangian for the case of Majoron decay is 1

c J,
Lint =
gij j,L
(1)
i,L
i
j
i=j
E-mail addresses: lindner@physik.tu-muenchen.de (M. Lindner), tohlsson@physik.tu-muenchen.de,

tommy@theophys.kth.se (T. Ohlsson), wwinter@physik.tu-muenchen.de (W. Winter).
1 We are not interested in transitions described by g = 0, since they are forbidden for kinematical
i
i
ii
reasons. Furthermore, because the gij s are usually assumed to be small, only the first order processes in the
Lagrangian Lint will be relevant to our discussion.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 2 3 7 - 1
M. Lindner et al. / Nuclear Physics B 607 (2001) 326354
327
where the s are Majorana mass eigenfields and J is a Majoron field.

Neutrino decay has been studied as an alternative to neutrino oscillations, especially
for atmospheric [710] and solar [7,11,12] neutrinos. So far, either decay only or some
sequential combination of decay and oscillations has been considered (e.g., MSWmediated solar neutrino decay [12,13]). The values of individual gij s or combinations
were restricted, for instance, from double beta decay, pion and kaon decays, or supernovae
[1417]. For pure neutrino decay without oscillations the constraints on the coupling
constants have been used to derive conditions for other parameters. For example, for
atmospheric neutrinos m2 0.73 eV2 was found in Ref. [8]. Since this is incompatible
with the known (or usually assumed) mass hierarchy of active neutrinos, decay into a
sterile neutrino was suggested. However, massive neutrinos allow neutrino decay as well
as neutrino oscillations. In this paper, we present a combined model-independent treatment
of neutrino decay and neutrino oscillations. We refer to two different cases:
Invisible decay: Decay into neutrinos or other particles which are not observable (e.g.,
sterile decoupled neutrinos 2 ).
Visible decay: Decay into neutrinos which are, in principle, observable (e.g., active
neutrinos).
Our paper is organized as follows: in Section 2, we will introduce a Majoron decay
model for illustration, in order to show two important properties: first, as already mentioned
before (e.g., in Ref. [7]), we may be able to detect active decay products. Second, we will
show that in such a model, decay products can, in principle, oscillate. In Section 3, we
will develop a model-independent operator framework for the calculation of transition
probabilities and describe its properties. Furthermore, we will show that the phase
relationships given by the S-matrix approach of quantum field theory are satisfied. The
transition probability for invisible decay as a generalization of the neutrino oscillation
formula will be derived in Section 4. As an example, we will apply this to atmospheric
neutrinos. In the following section, Section 5, this formalism will be used for visible decay
in the approximation of almost stable decay products. Again, we will make an application
to atmospheric neutrinos. In Section 6, we will, for illustration, construct a minimal model
in which decay products oscillate. Finally, in Section 7, we will present a summary as well
as our conclusions.
2 In this paper, sterile decoupled neutrinos refers to neutrinos which can be described by a block diagonal
mixing matrix such as

U=
U33
0
0
UnS nS

,
(2)
where nS is the number of sterile neutrinos, U33 the active neutrino mixing matrix, and UnS nS the sterile
neutrino mixing matrix.
328
2. Majoron models
Neutrino decay through emission of a massless NambuGoldstone boson is currently
among the most favorable alternatives for neutrino decay scenarios. Therefore, we will use
it as an introductory example for our framework.
2.1. Dynamics of Majoron decay
For quantitative estimates we will investigate decay by Majoron emission described by
an interaction Lagrangian with Yukawa (scalar or pseudoscalar) couplings (cf. Ref. [18]),
Lint =
g1
g2
1 2 J + 1 i5 2 J.
2
2
(3)
Here 1 and 2 are assumed to be Majorana neutrinos. Since the weak interaction couples
only to left-handed neutrinos and right-handed antineutrinos, and Majorana particles are
identical to their anti-particles, Majorana neutrinos and Majorana antineutrinos can only
be distinguished by their spin. The matrix elements in the observers rest frame are given
by [18]
2
2

M(2 1 )2 = g1 (A + 2) + g2 (A 2),
4
4

2 g12 + g22 m21 + m22
g12 + g22 1
M(2 1 ) =
+xA ,
A =
4
m1 m2
4
x
(4)
(5)
where we have with the assumed mass hierarchy m2 > m1 and

1 E2
E1
m1 E 2 m2 E 1
+
=
+x ,
m2 E 1 m1 E 2 x E 1
E2
m2
> 1.
x
m1
The differential decay rate for the decay 2 1 is

2
d21 m1 m2 1
=
M(E1 ) ,
dE1
4 E2 |p2 |
with the kinematics in spherical coordinates (z-axis in the direction of propagation)

E2 E1 = |p2 p1 | = p21 + p22 2|p1 ||p2 | cos .
The condition cos 1 implies for the allowed energy range of E1 that

E2
E2
|p2 |
|p2 |
1
1
1
1
1+ 2
1 2 E1
1+ 2 +
1 2 .
2
x
2
x
2
x
2
x
(6)
(7)
(8)
For relativistic neutrinos, i.e., m2

E2 , this reduces to
E2
E1 E2 .
x2
(9)
329
We immediately see that for x 1, i.e., m2

m2 , the energy shift by decay is very
small. For an energy resolution

1
Eres E2 1 2
(10)
x
the decay products are detected in the same energy bin as the original particles.
Now, we can integrate the differential decay rate given in Eq. (6) over the range
determined by Eq. (8). Using Eqs. (4), (5), and (6) as well as assuming relativistic
neutrinos, i.e., m2
E2 , we obtain for the total decay rates [18]

2
2
m1 m2 2 x
1
+ 2 + ln x 2 3
(2 1 ) =
g1
16E2
2
x
x
2x

(11)
2
2
1
x
2 + ln x + 2 3
+ g22
2
x
x
2x
and

x 2
1
m1 m2 2
2
ln x 3 .
g + g2
(2 1 ) =
16E2 1
2 x
2x
(12)
2.2. Visibility of decay products

Solving Eq. (7) for cos shows that cos as a function of E1 always has a minimum at
E1,min =
2E2 m21
m21 +m22
(cf. Fig. 1). We can use this to constrain . For relativistic neutrinos, i.e.,
m2
E2 , we obtain

m1 2
max =
x 1 .
2E2
(13)
Let us assume m1 and m2 to be comparable by orders of magnitude, i.e., x 1, which is

plausible for active neutrinos for the currently assumed mass squared differences. For d,
the distance to the detector, the area F covered by a neutrino beam at the detector due to a
beam-spread by decay is approximately given by

2

2
2
dm1 2
2
2 m1
x 1 =O
F (max d) = d
(14)
2E2
E2
for x = O(1). Table 1 shows typical values for several cases, where neutrino beams and
not point sources are assumed. For a beam-spread at the detector (area F ) much smaller
than a usual detection area (D), F
D, everything will be visible, i.e., in the cases of
atmospheric, accelerator, and reactor neutrinos.
Let us now introduce a function which describes the fraction of the decay products that
will arrive at the detector by geometry, i.e., does not escape from detection by changing
direction. We will first of all consider the case of a neutrino beam, and discuss the case
of a point source later. We introduce this function in a quite general form applicable to
any decay model such as a model involving more than two decay products. Thus, is a
function of intrinsic decay properties such as energy and mass of the original particle Ei ,
mi , and of the decay products Ej , mj , as well as geometric properties such as length of
330
Fig. 1. The function cos = cos (E1 ) plotted for the (nonrelativistic) sample data m1 = 1 eV,
m2 = 2 eV, and E2 = 10 eV. Since cos 1, this function is only defined below the horizontal
line.
Table 1
Typical energies (E2 ), distances (d), and areas (F ) covered by decay beam-spread for different types
of neutrinos. Here it is assumed that m = O(eV) and x = O(1)
Neutrino type
E2 (MeV)
Solar
Atmospheric
Accelerator
Reactor
Supernova
0.1 10
102 103
103 104
1 10
10
d (m)
1011
104 107
102 103
10 102
1021
F (m2 )
1010 1011
1012 102
1016 1012
1012 108
1016
Fig. 2. A neutrino beam pointing from a source S to the detection area D with decay at X after
traveling a distance l along the baseline L. The decay products of the beam cover the area F at the
surface of the detector by a beam-spread of the angle 2 .
331
the baseline L, decay position l, and detection area perpendicular to the beam direction D
(see Fig. 2), i.e.,
= (Ei , mi , Ej , mj , L, l, D) ij (L, l, D).
(15)
Note that the intrinsic properties of the beam and the detector can be independently folded
with the transition probability, since they are not directly affected by decay.
Since we assume that we cannot detect particles other than the neutrino decay product,
we have to integrate over all momenta and energies except for the angle (cf. Fig. 2) and
the energy of the decay product E . In general, these two parameters are to be integrated
over the relevant detection region and the energy bin:

Emax 1

d 2 ij
1

ij (L, l, D) = tot
(16)
d cos dE dE d cos .
ij
Emin (cos )D
Here [Emin , Emax ] is the energy bin range and (cos )D is the cosine of the maximal angle,
such that the decay product will still hit the detector. Since we only have one independent
d 2
d
parameter in Majoron decay, the double-differential decay rate d cos ijdE will reduce to dE

times a -distribution determined by Eq. (7). Assuming that the energy bin width Ebin
and the angle D are very small, i.e., the differential decay rate is approximately constant
within the target region and the energy bin range, we obtain for

1 d 2 ij
Ebin (cos )D ,
ij (L, l, D) tot
(17)
ij d cos dE cos =1, E =Ebin
bin is the mean energy of the target bin considered, Ebin the energy bin width
where E
of that bin, and (cos )D the cosine range, i.e., (cos )D = 1 (cos )D , of the angle
covered by the detector. From geometry (Fig. 2) we can determine (cos )D for the decay
position l and a spherical detection area D:
2(L l)2 (cos )D D.
(18)
Table 1 indicates that for solar neutrinos the spread of a beam at the detector is normally
larger than the detection area. Fig. 3 shows a numerical evaluation of 21 for solar neutrino
sample data and a very large detection area for illustration. One can see that in all cases
the 1/(L l)2 -dependence dominates in the range of about 6 1010 m to 9 1010 m over
the effects of the couplings or helicities, so that we obtain the same qualitative behavior.
However, the Sun is a radially symmetric (4 ) source. It is not a point source, because
the production area is large compared to the distance to the Earth. Eq. (13) indicates that
for decay the maximal angle of deviation of the direction of flight ( 107 ) is much
smaller than the angle under which the Suns core, where the neutrinos are produced, is
observed from the Earth ( 103 ). Hence, equally distributed small deviations of the
direction of flight will already be washed out by the flux profile. Besides that, after decay,
the profile will still remain radially symmetric. Thus, for the case of the Sun, the function
21 describing a beam can only serve as an example to show the qualitative behavior, and
we conclude that Sun 1.
332
Fig. 3. The fraction 21 = 21 (l) for decay products of a solar neutrino beam hitting the detector
as a function of the traveling distance from the surface of the Sun l. The sample data used are
E2 = 10 MeV, m1 = 0.02 eV, m2 = 0.07 eV (cf. Ref. [7]), L = 1011 m, and D = 105 m2 . Note that
Sun 1 for the Sun, because of the flux profile.
21
Supernova neutrinos are different from solar neutrinos, since they come from a fardistant point source, and hence from a well-defined direction. One might argue that it
is impossible to distinguish neutrinos produced by decay in the direction of flight or by
decay in slightly different paths due to the radially symmetric flux profile, since decay
scatters the neutrinos back into the direction of the detector. However, these two cases may
be distinguishable for certain cases or sets of parameters because of the time resolution
(e.g., different traveling times or different oscillation lengths due to different path lengths).
Hence, supernova neutrinos might be the only nontrivial example of partly visible Majoron
decay, i.e., < 1. In general, any far-distant beam source will show similar effects for
Majoron decay.
Eventually, we conclude that Majoron decay products are in most cases visible to a
good approximation, i.e., ij (L, l, D) 1. In our calculations, we will generalize this
to ij (L, l, D) ij (L, D). Hence, will be assumed not to depend on the distance l.
Moreover, for supernova neutrinos it should be a good approximation to use a step function
for ij (L, l, D) according to Fig. 3.
2.3. Neutrino oscillations of decay products
In this subsection, we demonstrate that neutrino decay products in Majoron decay may
oscillate. Since neutrino oscillations are a consequence of phase coherence, several issues
connected to this subject need to be addressed for a combined treatment of neutrino decay
and neutrino oscillations. We will do that in detail in Subsection 3.2.
In a charged current weak interaction, it is normally assumed that the phase coherence
within a superposition of states as an in state is destroyed. This means that such a process
produces a superposition of states with random relative phase shifts, i.e., the states within
the superposition cannot interfere. Hence, cross sections referring to the individual states
333
have to be summed incoherently, i.e., by summation of probabilities and not amplitudes (cf.
Property 4 in Subsection 3.2). This comes from kinematical features related to the mass
differences of accompanying leptons [19], i.e., the coherence length is relatively small due
to the large mass differences.
In a neutral current weak interaction, the out state can be a coherent superposition of
states. One example is Z 0 decay, producing a coherent superposition of different neutrino
and antineutrino mass eigenstates [19]

1
|Z |1 |1 + |2 |2 + |3 |3
3
(19)
by an interaction Lagrangian

3
g
iL iL Z0 .
Lint = i
2
(20)
i=1
Majoron decay resembles a neutral current weak interaction in the sense of a Lagrangian
similar to Eq. (20). It couples a superposition of different mass eigenstates to a single
particle, so that all eigenstates and their wave packets, respectively, experience the same
energy shift E transferred to the massless boson. From the wave packet treatment of
neutrino oscillations we know that within a certain coherence length, the mass differences
of neutrinos do not destroy the coherence of the mass eigenstates, i.e., neutrino oscillations.
Hence, since in Majoron decay the energies of the mass eigenstates are all shifted by the
same amount E in decay, the mass differences will not be able to destroy the coherence
of out states in this case either. 3 Thus, the decay products can oscillate.
Using a Majoron-like decay interaction Lagrangian, we know that to first order of the
S-matrix expansion we can write

2

4
d (in out) out|
d x
gij j i J |in .

(21)
i=j
This expression implies that the interaction Lagrangian destroys the incoming superposition completely, and creates a superposition of outgoing mass eigenstates with coefficients
determined by the coupling constants and the coefficients of the incoming superposition.
The outgoing states have a fixed relative initial phase such as the outgoing mass eigenstates
in Z 0 decay [19]. In order to observe neutrino oscillations of decay products, we need to
have at least two different mass eigenstates in the out state superposition. We immediately see from Eq. (21) that this implies that at least two of the gij s, say gi1 j1 and gi2 j2 ,
have to be nonzero for the corresponding states |i1 and |i2 with non-vanishing in state
amplitudes i1 and i2 .
3 Later in this paper, a more detailed proof of this analogy applied to our framework will be given in Property 9
in Subsection 3.2.
334
3. Operators and properties

In this section, we will introduce effective decay operators for a model-independent
combined treatment of neutrino decay and neutrino oscillations. Decay can be described by
decay rates ij gij2 , which are calculated from the corresponding Feynman diagrams. We
will use such an approach to effectively integrate neutrino decay into neutrino oscillation
scenarios. For this we do not need details of the specific model as long as neutrino decay
can be described by a generic effective Lagrangian destroying one neutrino mass eigenstate
and creating another one, i.e., Lint gij j i . This also includes multi-particle decay into
decay products other than neutrinos, where all momenta and energies of the undetected
decay products are to be integrated over in the decay rates. 4 We will show that our operator
framework satisfies the usual properties of the S-matrix approach and implements the
correct phase relations. In certain cases, we will demonstrate this for Majoron decay.
3.1. Operators
Let us introduce decay and propagation operators in terms of creation and annihilation
respectively, as well as calculation rules for the transition probabilities.
operators, a and a,
The symbols introduced in the operators will be explained thereafter. In the next subsection,
we will show how the operators satisfy the expected properties. We define three operators:
Definition 1 (Disappearance operator). D is the transition operator generally known as
the decay operator. Effectively, it returns the amplitude for an undecayed state remaining
undecayed after traveling a distance l along the baseline L:

i l
a a i .
exp
D (l) =
(22)
2Ei i
i
Definition 2 (Appearance operator). D+ is the differential transition operator, which

destroys an in state and creates an out state in [l, l + dl] along the baseline L, i.e., a new
state appears:
ij
D+ (l, L) =
(23)
ij (L, l, D) a j a i ei .
Ei
i
i=j
The phase is a random phase taking into account the phase shift by additional
(not measured) particles produced in the decay, such as Majorons (cf. Property 7 in
Subsection 3.2). The probability density, which is the square of the amplitude, will have to
be integrated over l.
4 For decay models involving more than one outgoing neutrino we could use a similar framework with slightly
different properties. For Majoron-like models higher-order processes are assumed to be suppressed, because the
coupling constants are considered to be small.
335
Definition 3 (Propagation operator). E is the operator propagating a state a distance l along

the baseline L: 5

E(l) =
exp(iEi l)a i a i .
(24)
i
In these definitions, ij (L, l, D) is the geometrical function introduced in Eq. (15) and
defined in Eq. (16). It describes the fraction of decay products which will still arrive at
the detector, i.e., is not redirected from the detector by decay. The decay rate is defined
as ij mi /ij0 for i j decay, and ij0 refers to the rest frame lifetime for that decay
channel. Time dilation by the factor i = Ei /mi implies that
ijobserver = ij0 i1 =
ij
1 mi
=
.
0 E
Ei
ij i
(25)

The rate i j ij is the overall decay rate of the state i, so that B ij /i is the
branching ratio for i j decay. The factor of two in the exponent of D and the square
root in D+ comes from the fact that, besides some phase factors, amplitudes effectively
behave like square roots of particle probabilities. However, as we will see in the next
subsection, the above operators together with the definitions of the transition probabilities
also implement the correct phase relations. Let us now define how to calculate the transition
probabilities. This is to be done by successive application of decay and propagation
operators:
Definition 4 (Calculation of transition probabilities). The transition probability P (
n
by exactly n intermediate decays is for n = 0 given by
)n = P

2
0
P
(26)
= |E(L)D (L)| ,
and for n > 0 given by
L
n
P
=
l1 =0
L
|E(L l)D (L l)

ln =ln1
2 n
n

dli .
D+ (li , L)E(li )D (li ) |

i=1
(27)
i=1

Here l ni=1 li . The order of D and E is arbitrary, since these operators commute as
we will see below, in Eq. (32).
In many cases, the total transition probability can be calculated as
P =
N
max
n
P
,
(28)
n=0
5 Here we are using the non-relativistic operator, which is different from the relativistic one only by an overall
phase factor. This phase factor will cancel in the calculation of transition probabilities. For justification of this
approximation see Ref. [20].
336
where Nmax denotes the maximal number of decays possible, depending on the decay
model. The individual terms in Eq. (28) add up only if the particles produced by a different
number of decays cannot be distinguished, such as by energy resolution, energy threshold
or signature (flavor or spin). Depending on the problem, the terms may have to be split
appropriately in general. 6
Finally, we can introduce two simplified definitions for invisible as well as stable or
almost stable visible decay products:
Definition 5 (Invisible decay). For invisible decay products we can, in principle, only
observe the undecayed particles themselves. Hence, we do not admit appearance operators,
i.e.,
invisible
0
= P
.
P
(29)
Definition 6 (Approximations for visible decay). Considering maximal one decay or

lifetimes longer than Lm/E, we can neglect repeated decay terms (n > 1) and approximate
the total transition probability in Eq. (28) by
appearance
0
1
invisible
+ P
= P
+ P
P P
(30)
appearance
P
where
describes the appearance of new particles by decay. From Eq. (27) we
1 for visible decay in addition to P invisible :
can read off the first correction term P
L
1
P

|E(L l)D (L l)D+ (l, L)E(l)D (l)| 2 dl.
(31)
l=0
3.2. Properties
In this subsection, we want to make the most important properties of our operator
framework more transparent. Since we deal with decay of states instead of decay of
particles, we need to show that this describes particle decay correctly while preserving
interference properties for states.
Property 1 (Creation and annihilation operators). The effective creation and annihilation
respectively, are consistent with the S-matrix approach.
operators, a and a,
i a a i in Eqs. (22) and (24) is as usual the occupation number operator,
The operator N
i
whereas the operator Tij a j a i in Eq. (23) is an effective transition operator coming from
inserting the fermionic neutrino field expansions into an equation like Eq. (21). Here i
6 For instance, we may even have two different P 2 coming from and described
i
j
k
i
j
k
i may have to be split into

by the appropriate decay rates. Another example is energy binning, where one P
fractions of different energy ranges to be calculated from the differential decay rate, such as Eq. (6) for Majoron
decay.
337
and j refer to all quantum numbers, such as spin and momentum. Since we assume that
conceptual issues, such as detection in a different energy bin or decay into an antiparticle,
are dealt with by splitting Eq. (28) appropriately (cf. Definition 4), we do not need to take
care of different quantum numbers here.
Property 2 (Commutation relations). The operators D and E satisfy the commutation
relation
[D , E] = 0,
(32)
whereas in general
[D , D+ ] = 0,
[D+ , E] = 0.
(33)
These relations can be shown by direct calculation of the commutators, using the fermionic
anticommutation relations {a i , a j } = ij .
One can also derive Eq. (32) by introducing the notion of a complex mass square, i.e.,
m
2i m2i ii .

=
i , where E
i p + m
For E(l)
2i /(2Ei ), we can read off from Eqs. (22)
i exp(i Ei l)a i a
and (24) that
= E(l)D (l).
D (l)E(l) = E(l)
This means that the disappearance and propagation operators can be combined by
the introduction of complex mass squares. Therefore, the commutator reduces to the
commutator of two scalars, i.e., the exponentials of the real and imaginary parts of the
mass square.
Property 3 (Decay into different channels). Our description for the decay of states
corresponds to a description for the decay of particles, which is given by the differential
equation system

d
Ni =
ij Ni +
j i Nj .
dt
j, j =i
(34)
j, j =i
Here ij is the rest frame decay rate for the particle decay i j and Ni is the number of
particles i. From this we obtain for the disappearance of a particle Ni,0 = 1, Nj,0 = 0 for

all j = i, and i j,j =i ij
Ni (t) = ei t ,
(35)
and for the appearance of a stable decay product Ni,0 = 0, Nj,0 = 1 for all j = i, and
i = 0
j i

1 ej t .
Ni (t) =
(36)
j
j, j =i
338
Neglecting propagation operators, 7 we obtain from Eq. (26) for the disappearance
probability of a mass eigenstate

2
iL
Pii = i |D (l)|i = e Ei ,
(37)
which is identical to the disappearance probability in Eq. (35) if i = i /Ei , which we

have in the observers rest frame.
Neglecting propagation operators, we obtain from Eq. (31) for the appearance of a stable
decay product (j = 0) for exactly one decay
L
Pij =
i

j |D+ (l, L)D (l)|i 2 dl = ij 1 e Ei L ,
i
(38)
l=0
which is identical to the appearance probability in Eq. (36) if = 1 (everything detected)

and i = i /Ei , which we again have in the observers rest frame.
In general, Eqs. (27) and (34) obey the same structure, since transitions among states are
described by differential rates as in a Markov chain model.
Property 4 (Coherent and incoherent summation of amplitudes). In quantum field theory,
coherent and incoherent summation are often distinguished in the calculation of Feynman
diagrams [21]. The notion coherent refers to the summation of amplitudes

2
Ai
d
i
and incoherent to the summation of squares of amplitudes, i.e., summation of probabilities

|Ai |2 .
d
i
Basically, coherent summation applies to processes which are, in principle, indistinguishable and incoherent summation to processes which are, in principle, distinguishable. Our
formalism implements the proper application of coherent or incoherent summation.
To show this, we split the statement into more specific parts in what follows.
Property 5 (Incoherent summation and repeated decay). Repeated decay can be described
by the incoherent summation in Eq. (28).
The differential decay rate in terms of the S-matrix is

2
d (in out) out|S|in ,
(39)
7 In this case, we are not interested in neutrino oscillation probabilities and phase factors will cancel anyway
when squaring the probability amplitudes.
339
where S can be expanded as usually as [22]

S=
n

i
n=0
n!

d 4 x1 d 4 xn T Lint (x1 ) Lint (xn ) .
(40)
Thus, the in and out states select the corresponding S-matrix expansion terms, which
will evaluate to nonzero results. For a different number of decays we obtain different
decay products characterized by different quantum numbers. Therefore, different S-matrix
expansion terms will be selected in Eq (39). This means that Feynman diagrams of different
orders cannot interfere, even if we do not distinguish them by the measured decay products
(often only the neutrinos but no other particles, such as Majorons, are measured). Hence,
decay implies the principal possibility to measure the decay products and, thus, in terms
of quantum mechanics acts as a measurement. Since repeated decay corresponds to higher
order Feynman diagrams in quantum field theory, incoherent summation properly describes
it.
Property 6 (On-shell propagation in between decays). It is shown in Ref. [23] that
Feynman propagators for virtual particles reduce to on-shell propagators for macroscopic
distances. 8 Thus, individual decays in repeated decay can be treated independently
without interference. More precisely, Feynman diagrams of the same order do not produce
interference terms, because they can (in principle) be distinguished due to the large
separation of vertices.
Eq. (28) together with Eq. (27) imply that decays are treated separately with decay rates
calculated as for independent processes.
Property 7 (Phase coherence between in and out neutrinos). Since at least a third
unmeasured particle, e.g., the Majoron, is involved, the overall phase coherence between
in and out neutrinos is destroyed.
0 and
The random phase in Eq. (23) takes care of this. Since it does not appear in P
1
cancels in P , it is only relevant for repeated decays for n > 1. It also ensures that terms
from different individual decays do not produce interference terms, i.e., Feynman diagrams
of the same order are summed incoherently in this case (cf. Property 6).
Property 8 (Decay model and causality). The decay model used in Eqs. (26) and (27)
properly implements causality.
The operator D gives the probability amplitude for a state remaining undecayed until
the position l. The operator D+ gives the conditional probability amplitude for decay in
the interval [l, l + dl] for an undecayed state. Successive alternating applications of D
and D+ give the (conditional) probability amplitude for the decision path. This probability
8 In this reference, this was explicitly shown for the creation and detection processes.
340
needs to be integrated to obtain the average over all possible decay positions l in a causal
order, which is taken into account by the integration limits, i.e., a particle cannot decay
before it even exists. We are integrating probability amplitudes for different decay positions
incoherently, taking into account that undecayed particles cannot interfere with decay
products.
Property 9 (Neutrino oscillations of decay products). The appearance operator D+ gives
the proper relative phase in a superposition of mass eigenstates as an out state. Kinematics
induced by different masses does not result in destruction of coherence among the
superimposed states such as in charged current weak interactions.
The decay rate is to first order of the S-matrix expansion as usually defined as [22]

2

1
4
(in out) = out| d x Lint |in .
(41)
T
Inserting a Majoron-like interaction Lagrangian (e.g., Eq. (1)) with the corresponding field

|i into another one
expansions yields for the decay rate from one flavor | = i Ui

| = j Uj |j
2

2

1
1
4

= | d x Lint | =
Ui Uj Fij ,
T
T
i
(42)
and from one mass eigenstate |i into another one |j

2

1
1
4
ij = j | d x Lint |i = |Fij |2 .
T
T
(43)
Here the Fij s refer to transition functions depending on the properties of the states i and
j as well as the Lagrangian and the used field expansions. Without loss of generality, for
Dirac field expansions of fermions and a KleinGordon field expansion of the boson, we
obtain for Lagrangians of the Majoron decay type

Fij = j | d 4 x Lint |i

= Nij d 4 x eipJ x A uj (p2 )eip2 x gij B ui (p1 )eip1 x ,
(44)
where Nij are normalization factors depending on energy, mass, and volume and A and B
are some operators depending on the Lagrangian, e.g., combinations of charge conjugation
and chiral operators. By introducing ij as the phase factor of Fij = |Fij |eiij determined
by Eq. (44), we can read off the following relation from Eq. (43):

1
1
ij eiij = |Fij |eiij = Fij .
T
T
(45)
As we will show below, for relativistic neutrinos ij , i.e., ij is independent of the

indices i and j .
341
Without loss of generality, in our framework the rest frame transition probability
between two flavor eigenstates for ij = 1 (i.e., everything is detected) is given by

2

2

P = |D+ | T = |
(46)
ij a j a i | T .

i
This can be obtained from Eq. (27) for an infinitesimally short baseline (dl = T ) by using
D E 1 to zeroth order in the expansion of the exponentials as well as ij = ij /Ei .
Applying Eq. (45) and the expansions of the flavor eigenstates, we finally obtain from
Eq. (46)

2
2

P
1
1

=
Ui Uj |Fij | =
Ui Uj |Fij | .
=
(47)

T
T
T
i
This is equivalent to Eq. (42) if ij = is independent of the indices i and j . In this case,
the appearance operator is identical to the expression derived from the S-matrix expansion
and the different masses do not destroy phase coherence in a superposition of states.
It remains to be shown that ij is independent of the indices i and j for relativistic
neutrinos. Since ij = arg Fij , any real factor in Fij such as, for instance, a normalization
factor, does not affect ij . We conclude from Eq. (44) that for real coupling constants
the i and j dependence of ij can only be influenced by the exponentials and the spinors.
By spatial integration, the exponentials are transformed into -distributions implying
momentum conservation and, hence, cannot affect the phase. Without loss of generality,
the spinors can be written in the Dirac representation as [22]
1
0
0
1
,
,
u (p) = N1
u (p) = N1
(48)
3
1
2
N2 p
N2 (p ip )
N2 (p1 + ip2 )
N2 p3
where N1 and N2 are (real) normalization factors. For relativistic neutrinos we can assume
the three momentum p to be identical for all mass eigenstates and absorb the mass
dependence of the four-momentum in a correction of the energy Ei p + m2i /2p. Then,
they do not affect ij in this approximation, because the phases of these spinors only
depend on the common three-momentum. Hence, ij is independent of the indices i and j .
4. Invisible decay
In this section, we will treat invisible decay with our operator framework, i.e., decay into
particles, such as sterile decoupled neutrinos, which are in principle unobservable. In fact,
this is the most often assumed neutrino decay scenario.
4.1. Transition probability for invisible decay
The transition probability for invisible decay is calculated according to Eqs. (26) and
(29). Since the operator D only gives real factors, the calculation for relativistic neutrinos
342
is quite straightforward and similar to the one for the general neutrino oscillation formula.

We use | = i Ui
|i and define the following quantities:
Ui
Uj ,
Jij Ui Uj
ij
m2ij L
(49)
(m2i m2j )L
,
4E
4E

mj L
mj L
mi L
mi
L
ij
(i + j )
.
+

+
20,i Ei
20,j Ej
0,i
0,j 2E
2E
(50)
(51)
Here Ei E has been used in the decay terms, because for decay the zeroth order terms in
Ei p + m2i /(2p) do not cancel, and therefore, the higher order terms can be neglected. 9
For more details on this kind of approximations see Ref. [20].
After some calculations, not to be shown here, we finally obtain

invisible
P
=
!Jij eij 4
!Jij sin2 ij eij

Ppure

2

j
i>j
decay

i

PCP
conserving
"Jij sin 2ij eij .
(52)
j
i>j

PCP
violating
In our definition, pure decay admits neutrino flavor mixing, but no neutrino oscillations
(ij = 0 for all indices i and j ).
4.2. Example: application to atmospheric neutrinos
In Ref. [9], neutrino decay was considered as an alternative to atmospheric neutrino
oscillations. Assuming 10 m221 = 0, 12 = 13 = 0, and the CP-phase = 0, we obtain
from Eq. (52) for decay of 2 into a sterile decoupled neutrino with the rate 2 the
same transition probabilities as in Ref. [9]: for pure decay ( m232 = 0) we have

2 L
2 2
P = c23
(53)
e 2E + s23
,
and for decay and oscillations
4
P = s23
4 L
+ c23
e E
2 2
+ 2c23
s23 cos

m232 L L
e 2E .
2E
(54)
9 For neutrino oscillations the zeroth order terms cancel due to multiplication with their complex conjugates in
the calculation of the transition probability.

10 In this paper, we use the standard parameterization for the active neutrino mixing matrix [24]
s12 c13
s13 ei
c12 c13
U33 = s12 c23 c12 s23 s13 ei
c12 c23 s12 s23 s13 ei
s23 c13 ,
s12 s23 c12 c23 s13 ei
c12 s23 s12 c23 s13 ei c23 c13
where sij sin ij and cij cos ij and where ij are the vacuum mixing angles.
343
Fig. 4. The survival probability P as a function of and L/E for cos2 23 = 0.30 and
m232 = 3.3 103 eV2 .
The authors of Ref. [9] fitted Super-Kamiokande data in the case of decay and oscillations
1 GeV
2
2
with negligible m232 finding = 63
km and cos cos 23 = 0.30. Pure decay
without neutrino flavor mixing, i.e., cos2 = 1, cannot describe the atmospheric neutrino
data [10,25]. Fig. 4 shows the survival probability P as a function of the decay rate
and the sensitivity L/E. One can easily see that the transition from pure oscillation to pure
decay is, in principle, determined by the curve L/E = O(1).
5. Visible decay
In this section, we will assume that the neutrino decay products are, in principle,
observable. We know that we may either detect them separately from the undecayed
neutrinos by different signatures or energies, or indistinguishably to the undecayed
particles with the same signature in the same energy bin. 11 We postulated in Definition 4
that these conceptual differences manifest themselves in a problem-dependent splitting of
Eq. (28) into appropriate parts.
11 This strongly depends on the detector properties. For example, for solar neutrino decay into antineutrinos
with detection in Borexino or Super-Kamiokande it was discussed in Ref. [7].
344
5.1. Transition probability for visible decay

appearance
We use Eqs. (30) and (31) to first approximation for P

, i.e., the case of maximal
one decay or sufficiently small decay rates. For the calculation of the transition probability
we make the following assumptions:
Stable decay products or i
Ei /L for all i.
The fraction ij (L, l, D) of decayed particles not escaping detection by a change of
direction does not depend on the decay position l, i.e., ij (L, l, D) = ij (L, D) ij
is a (real) constant in l. In fact, we observed that ij (L, l, D) 1 in most cases of
Majoron decay (cf. Subsection 2.2).
The masses and decay rates are non-degenerate. 12
For ij and ij we use the definitions in Eqs. (50) and (51). Applying Eqs. (30) and (31)
and assuming that we cannot distinguish decay products from undecayed particles, i.e., we
0 and P 1 , we have for P
add P
appearance
0
1
invisible
P P
+ P
= P
+ P
L
0
= P

|E(L l)D (L l)D+ (l, L)E(l)D (l)| 2 dl.
(55)
l=0
invisible is given by Eq. (52), i.e.,
The invisible decay term. P
invisible
=
!Jij eij
P
i

i
!Jij sin2 ij eij 2
j
i>j
The appearance term.
appearance

i
"Jij sin 2ij eij .
(56)
j
i>j
evaluates after some algebra (cf. Appendix A) to
appearance
P

ij kl
L
=
ij kl
E (j l ik )2 + 4(ij + lk )2
i
i=j
k=l

! Kij kl (j l ik ) eik cos(2ki ) ejl cos(2lj )

!
2(ij + lk ) eik sin(2ki ) ejl sin(2lj )

" Kij kl (j l ik ) eik sin(2 ki ) ejl sin(2 lj )

!"
+ 2(ij + lk ) eik cos(2ki ) ejl cos(2lj ) ,
(57)
U U U is a generalization of J .
where Kij kl Ui
j k l
ij
12 The calculation of integrals changes for degenerate values of the s and m s (see calculation in
i
i
Appendix A). Nevertheless, we will calculate limits with degenerate values later on.
345
5.2. Limiting cases

Let us now consider some special cases for the general expression for the appearance
term in Eq. (57).
No oscillations. If we ignore neutrino oscillations, i.e., for degenerate masses or
extremely small m2 s, we will obtain in the limit m2 0

2 ij kl
appearance
P
ij kl
! Kij kl
j + l i k
i
k
l
k=l
i=j

eik ejl .
(58)
No interference. For random phases within superpositions of mass eigenstates, i.e.,

| = a|1 + bei1 |2 + cei2 |3 ,
(59)
with a, b, and c some coefficients and 1 and 2 random phases, the interference terms
in Eq. (57) vanish (e.g., for incoherent propagation). We can include this by introducing
ik j l in the sum of Eq. (57) and we finally obtain

ij
appearance
ij
! Kij ij ei L/E ej L/E .
P
(60)
j i
i
i=j
Only one possible decay channel. Assume that there is only one possible decay channel,
i.e., ij i = 0 and all other kl = 0. Then, we obtain the simplified formula

appearance
ij ! Kij ij 1 eL/E .
P
(61)
5.3. Example: application to atmospheric neutrinos
Here we will extend the example from Subsection 4.2, where Eq. (54) gives the
oscillation formula for decay of 2 into an invisible decay product such as a sterile neutrino.
Let us now assume that 2 decays into the active neutrino 3 . According to Eq. (55),
appearance
invisible , which can be calculated from
we are now looking for P
in addition to P
Eq. (61) in this limit. If we assume that both undecayed and decayed neutrinos are detected
indistinguishably, we will obtain
appearance
invisible
P = P
+ P
invisible in Eq. (54) and

with P

appearance
2 2
= 23 s23
c23 1 eL/E .
P
(62)
(63)
Here is the decay rate of 2 3 decay and E is the energy of the undecayed neutrino.
Fig. 5 shows the different parts of the survival probability for the parameter values given in
the figure caption. One can easily see the effect of the increase of the survival probability
for L/E 1/.
346
Fig. 5. The different parts of the survival probability P plotted as functions of the sensitivity L/E
2 = 0.3, = 0.001 GeV , = 1, and m2 = 3.3 103 eV2 [9].
for the parameter values c23
23
km
32
6. An example for neutrino oscillations of decay products

In this subsection, we will construct a minimal example for observing neutrino
oscillations among decay products.
6.1. The scenario
Since we want to create a superposition of out states in decay, we have to assume that
at least two of the gij s, i.e., ij s, in the interaction Lagrangian are nonzero. We define a
mass hierarchy m3 > m2 > m1 and only allow the decay chain 3 2 1 , i.e., 32
3 = 0, 21 2 = 0, and all the other ij s are equal to zero. In order to demonstrate the
physical effects we choose suitable parameters. Hence, we define the m2 s quite close to
each other, i.e., m221 = 5 104 eV2 , m232 = 3.3 103 eV2 , maximal mixing, i.e.,
1 GeV
|e = 1 |1 + 1 |2 + 1 |3 , and the s close to each other, i.e., 2 = 1000
km and
3 =
3
1 GeV
1100 km .
Furthermore, we assume that ij 1 for all i, j and the CP-phase = 0.
6.2. Transition probabilities

Let us now look at the disappearance probability of electron neutrinos Pee . From Eq. (52)
we obtain the invisible decay disappearance probability as

2
L
L
L
invisible
ee
ee 2 E
ee 3 E
ee 2 E
2 m21 L
= J11 + J22 e
+ J33 e
+ 2J12 e
1 2 sin
Pee
4 E

2
m31 L
L
ee 3 E
1 2 sin2
e
+ 2J13
4 E

+ L
m232 L
ee 2 2 3 E
+ 2J23
(64)
e
1 2 sin2
.
4 E
347
The appearance term is obtained from Eq. (57). Since in this case Kijeekl = Uei Uej Uek Uel =
1/9 and "K = 0, the terms in the sum of Eq. (57) are equal for simultaneous i k
and j l index exchanges. In order to make the result physically more transparent, we
recognize that only a limited number of the terms in the sum are non-vanishing and split
the formula into three parts:
appearance
Pee
app,1
= Pee
app,2
+ Pee
int
+ Pee
.
(65)
app,1
Here Pee
refers to the production of new 1 s by decay,

L
app,1
ee
Pee = K2121
1 e2 E ,
(66)
app,2
Pee
to the production of new 2 s by decay,

L
3 3 L
app,2
ee
E e 2 E ,
e
Pee = K3232
2 3
(67)
int to an interference term describing neutrino oscillations among the decay products
and Pee
ee
ee
1 and 2 (here K3221
= K2132
is used),
3 2
int
ee
Pee
= 4K3221
2
+ ( m221 m232)2
# 3

2 +3 L
2 L
m232 L
m221 L
e 2 E cos
3 e 2 E cos
2 E
2 E

2
2 L

m21 L
+ m221 m232 e 2 E sin
2 E

$
+ L
m232 L
22 3 E
e
sin
(68)
.
2 E
6.3. Results and analysis

appearance
for the scenario data. The

Fig. 6 shows the different parts of the probability Pee
app,2
appearance of 2 as decay product is described by Pee . It grows in the beginning,
because it is dominated by 3 2 decay. Since 2 > 3 , the 2 is decaying faster than
app,2
produced. Therefore, Pee falls again for large L/E. The appearance of 1 is determined
app,1
app,1
grows until it reaches its asymptotic
by Pee . Since 1 is defined to be stable, Pee
app,1
ee
value limx Pee = K2121 = 1/9.
Describing neutrino oscillations of decay products, i.e., 1 and 2 , the interference term
int basically oscillates as a beat with the two frequencies determined by m2 m2 .
Pee
32
21
The m232 -dependency comes from the phase propagation before decay and the m221 dependency from the phase propagation after decay.
Fig. 7 shows the different parts of the total survival probability Pee . The invisible decay
probability (disappearance) is the ordinary oscillation probability damped by decay. Note
appearance
invisible can be only sensibly added if the decay products cannot be
and Pee
that Pee
distinguished from the original ones in the detection process. In this case, the detection
348
appearance
Fig. 6. The different parts of the probability Pee

data.
plotted as functions of L/E for the scenario
Fig. 7. The different parts of the probability Pee plotted as functions of L/E for the scenario data.
rates will be enhanced by a fixed amount at large L/E. Neutrino oscillations of undecayed
particles will also vanish in this region. For small L/E there is a correction due to 2
abundance by decay and in addition due to the interference term.
Neutrino oscillations of decay products, induced by the oscillations of the interference
term, can become unobservable due to suppression by the mixing angles or the decay
rates different by some orders of magnitude. The first point is quite obvious, since the
interference term is in general proportional to
Uj Uk Ul
,
Kij kl = Ui
where i = j, k = l, (i, j ) = (k, l).
(69)
This requires that there are coefficients of at least three different states involved as a
product, i.e., all of them need to be quite large. The second point is physically obvious,
but harder to see in our equations. Since the interference describing neutrino oscillations of
349
int | for different decay rates and .

Fig. 8. Numerical evaluation of the absolute maximum of |Pee
3
2
decay products can only be observed if two states are produced simultaneously in decay, the
couplings to both of them need to be quite large. Large couplings imply large decay rates
for both transitions. If one of them is too large, the corresponding state will have decayed
before oscillations can be observed. Hence, the decay rates need to be large enough and
of the same order of magnitude in order to obtain reasonable results. A measure for the
int |. In Fig. 8, we show
magnitude of the interference effects is the absolute maximum of |Pee
the result of a numerical evaluation of this function for different values of the decay rates
3 and 2 . One can easily verify that for too small or too different decay rates, this function
gives small values, i.e., we do not see any interference effects.
6.4. Application to limiting cases
In Subsection 5.2, we studied the appearance probability in certain limits. Let us now
apply these to our scenario to see what happens if certain terms are assumed to be
negligible.
No oscillations. In this limit, all m2 s are equal to zero (cf. Eq. (58)). Here we can
simply use

appearance
app,1
app,2
int
= Pee + Pee + Pee
.
Pee
(70)
m2 0
ij
No interference. This limit is described by Eq. (60), which is equivalent to neglecting the
interference term. We obtain
appearance
Pee
app,1
= Pee
app,2
+ Pee
(71)
350
appearance
Fig. 9. The probability Pee

model data.
for different limiting cases plotted as functions of L/E for the
Fig. 9 shows the different limits of the appearance term compared to the original one. In
all cases, we obtain similar values for small and large L/E, since for L/E
( m2 )1 the
oscillation terms are ineffective and for L/E 1 they are suppressed by decay. In the
limit of no oscillations, there is no oscillating behavior of the curve. Nevertheless, there are
interference terms coming from the coherent coupling of the Lagrangian to different mass
eigenstates. Hence, in this case we obtain an enhancement in the appearance probability.
In the limit of no interference, we obtain a curve similar to the original one, but without
oscillations, which are described by the interference term.
7. Summary and conclusions

In this paper, we have introduced a combined treatment of neutrino decay and neutrino
oscillations in a formal model-independent operator framework. We have shown that it can
be easily applied to special cases such as atmospheric neutrinos. We have also seen that
for Majoron decay active neutrinos as decay products have to be taken into account in the
detection rates and may even be indistinguishable from the original undecayed particles
by the detector. In addition, in Majoron-like decay models neutrino decay products may
oscillate.
We conclude that the visibility of decay products in Majoron decay models may crucially
affect the detection rates depending on the properties of the detector. For other decay
models only a certain fraction of the decay products may arrive at the detector due to
kinematics. Moreover, neutrino oscillations of decay products are observable for relatively
large neutrino flavor mixing only.
One can show that for the combination of neutrino decay and neutrino oscillations, the
dimension of the parameter space is equal to 32 n(n 1), where 12 n(n 1) parameters
come from decay and n(n 1) from oscillations. Here n denotes the number of neutrino
flavors. Hence, the number of parameters is increased when neutrino decay is included.
351
Fig. 10. Possible decay scenario for four neutrinos.
Thus, especially in multi-neutrino scenarios, effects not consistent with the conventional
neutrino oscillation framework could be described by a combined neutrino decay and
neutrino oscillation approach.
As a realistic scenario, we could, for example, use a four-neutrino framework with a
decoupled sterile neutrino S (cf. Eq. (2)) and a mass hierarchy m3 > m2 > m1 > mS (cf.
Fig. 10). In such a scenario, we have up to six decay rates in addition to the usual mixing
parameters. From other conditions such as the observability of supernova neutrinos, we
then would need to constrain the lifetimes for individual i s (e.g., 1 almost stable). Since
the maximal number of possible decays is Nmax = 3 in this scenario, the exact transition
0
1 , P 2 , and
, P
probability P without lifetime constraints would be determined by P
3 in Eq. (28).
P
Since nonzero neutrino masses imply both neutrino decay and neutrino oscillations,
neutrino decay may become interesting not only as an alternative to neutrino oscillations
but also as a part of the whole neutrino scenario. Even small gij s may affect detection rates
of long traveling neutrinos such as supernova or primordial ones. Flux measurements (e.g.,
of supernova neutrinos) cannot always be used to extract neutrino mass limits reliably,
because of different path lengths of undecayed and decayed neutrinos. Thus, dispersion
may be mimicked by decay.
As a final conclusion, our operator formalism as a generalization for a combined
treatment of neutrino decay and neutrino oscillations could be of interest to those, who
are interested in analyzing neutrino data in a more general way. In addition, it may also be
applied in other oscillation scenarios outside of neutrino physics, where coherence is not
destroyed in decay.
Acknowledgements
We would like to thank Evgeny Akhmedov for useful discussions and Jrn Kersten for
proof-reading the manuscript.
This work was supported by the Swedish Foundation for International Cooperation in
Research and Higher Education (STINT), the Wenner-Gren Foundations, the Studienstiftung des deutschen Volkes (German National Merit Foundation), and the Sonderforschungsbereich 375 fr Astro-Teilchenphysik der Deutschen Forschungsgemeinschaft.
352
Appendix A. Calculation of the appearance term

We are using Eq. (31) as an approximation for the appearance term:
2
L

appearance
P
|E(L l)D (L l)D+ (l, L)E(l)D (l)| dl

l=0
L
|A |2 dl,
(A.1)
l=0
where A is the (differential) transition amplitude. Applying Eqs. (22), (23), (24), and

|i , we obtain for the transition amplitude 13
| = i Ui
(Ll)
l

ij
j
i
A =
Ui
Uj eiEj (Ll) e 2Ej
ij eiEi l e 2Ei .
(A.2)
Ei
i
i=j
In this equation, we already have evaluated the creation and annihilation operators for
readability. Now we use the approximation for relativistic neutrinos Ei p + m2i /(2p)
p + m2i /(2E). For neutrino oscillations the three-momentum p only gives a common
phase factor, which will cancel. For neutrino decay, the exponentials ei l/(2Ei ) and
ej (Ll)/(2Ej ) are real, and thus Ei p E in the lowest order nontrivial approximation.
Using the relativistic approximations as well as the definition m2ab m2a m2b then yields

2
2

ij j L i mj L (j i )l i mji l
2E
2E
2E
2E
A =
(A.3)
e
Ui Uj ij
e
e
e
.
E
i
j

l -dependent
i=j
l -independent
For the appearance probability we obtain
appearance
P
L
=
|A |2 dl
l=0
L
=
l=0
A A dl =

i
k
l
k=l
i=j
ij kl
L
Ui
Uj Uk Ul

ij
E
j +l i k
2E
Kijkl
2
kl j +l L i mjl L
e 2E e 2E
E
l i
2
m2
ji mlk
2E
dl,
(A.4)
l=0
13 Here we already neglect the random phase factor ei from the operator D , since it will cancel anyway by
+
taking its absolute value squared in the transition probability.
353
where the ij s are assumed to be real. For further evaluation we use the integral (a, b real)
L
e(a+bi)l dl =
0

1 (a+bi)L
a bi (a+bi)L
e
e
1 = 2
1 ,
2
a + bi
a +b
for a + bi = 0. Comparison with Eq. (A.4) shows that the latter condition implies

j + l i k + i m2j m2i m2l + m2k = 0.
(A.5)
(A.6)
Since i = j k = l by the summation rules, this can only happen for (i = k) (j =

l) (i = j ) for non-degenerate ms. Here we assume the s to be non-degenerate for
simplicity, so that we can apply Eq. (A.5). Using the same abbreviations as for invisible
decay in Eqs. (50) and (51), we can continue evaluating Eq. (A.4) with Eq. (A.5) by
identifying a = (j + l i k )/(2E) = j l ik and b = ( m2j i m2lk )/(2E) =
2(j i + kl )
appearance

i
i=j
k=l
Kij kl
m
jl
L j +l
ij kl ij kl e 2E L ei 2E L
E

L
j l ik 2i(j i + kl ) (j +l i k +i( m2ji m2lk )) 2E
1
e
2
2
(j l ik ) + 4(j i + kl )

L j l ik 2i(j i + kl )
=
Kij kl ij kl ij kl
E (j l ik )2 + 4(j i + kl )2
i=j
k=l

eik e2iki ejl e2ilj .
appearance
appearance
is real, we have P
Because P
complex, it can be shown that

! zei = !(z) cos "(z) sin ,

! iyei = "(y) cos !(y) sin .
(A.7)
appearance
= !P
. For real, y and z

(A.8)
(A.9)
We can use this to find

! K(c + di)e2i
= !(K)c cos 2 "(K)d sin 2 "(K)c cos 2 !(K)d sin 2,
(A.10)
for c, d real. In comparison with Eq. (A.7), we identify c = j l ik and d = 2(ij +lk ).
Applying this formula and regrouping in !K and "K yields the final result in Eq. (57).
References
[1] J.N. Bahcall, N. Cabibbo, A. Yahil, Phys. Rev. Lett. 28 (1972) 316.
[2] S. Pakvasa, K. Tennakone, Phys. Rev. Lett. 28 (1972) 1415.
354
[3] G.T. Zatsepin, A.Y. Smirnov, Yad. Fiz. 28 (1978) 1569;

G.T. Zatsepin, A.Y. Smirnov, Sov. J. Nucl. Phys. 28 (1978) 807.
[4] Y. Chikashige, R.N. Mohapatra, R.D. Peccei, Phys. Rev. Lett. 45 (1980) 1926.
[5] G.B. Gelmini, M. Roncadelli, Phys. Lett. B 99 (1981) 411.
[6] S. Pakvasa, hep-ph/0004077.
[7] A. Acker, A. Joshipura, S. Pakvasa, Phys. Lett. B 285 (1992) 371.
[8] V. Barger et al., Phys. Rev. Lett. 82 (1999) 2640, astro-ph/9810121.
[9] V. Barger et al., Phys. Lett. B 462 (1999) 109, hep-ph/9907421.
[10] G.L. Fogli et al., Phys. Rev. D 59 (1999) 117303, hep-ph/9902267.
[11] S. Choubey, S. Goswami, D. Majumdar, Phys. Lett. B 484 (2000) 73, hep-ph/0004193.
[12] A. Bandyopadhyay, S. Choubey, S. Goswami, hep-ph/0101273.
[13] R.S. Raghavan, X.G. He, S. Pakvasa, Phys. Rev. D 38 (1988) 1317.
[14] V. Barger, W.Y. Keung, S. Pakvasa, Phys. Rev. D 25 (1982) 907.
[15] A. Acker, S. Pakvasa, J. Pantaleone, Phys. Rev. D 45 (1992) 1.
[16] M. Kachelrie, R. Toms, J.W.F. Valle, Phys. Rev. D 62 (2000) 023004, hep-ph/0001039.
[17] R. Toms, H. Ps, J.W.F. Valle, hep-ph/0103017.
[18] C.W. Kim, W.P. Lam, Mod. Phys. Lett. A 5 (1990) 297.
[19] A.Y. Smirnov, G.T. Zatsepin, Mod. Phys. Lett. A 7 (1992) 1272.
[20] C. Giunti, C.W. Kim, hep-ph/0011074.
[21] W. Grimus, P. Stockinger, S. Mohanty, Phys. Rev. D 59 (1999) 013011, hep-ph/9807442.
[22] F. Mandl, G. Shaw, Quantum Field Theory, Wiley, New York, 1993.
[23] W. Grimus, S. Mohanty, P. Stockinger, hep-ph/9909341.
[24] Particle Data Group, D.E. Groom et al., Eur. Phys. J. C 15 (2000) 1.
[25] P. Lipari, M. Lusignoli, Phys. Rev. D 60 (1999) 013003, hep-ph/9901350.

Sterile neutrinos in tau lepton decays

Vladimir Gribanov 1 , Sergey Kovalenko 1 , Ivan Schmidt
Departamento de Fsica, Universidad Tcnica Federico Santa Mara, Casilla 110-V, Valparaso, Chile
Received 15 February 2001; accepted 4 April 2001
Abstract
We study possible contributions of heavy sterile neutrinos h to the decays e ( ) .
From the experimental upper bounds on their rates we derive new constraints on the h mixing in
the mass region 140.5 MeV mh 1637 MeV. We discuss cosmological and astrophysical status
of h in this mass region and compare our constraints with those recently derived by the NOMAD
Collaboration. 2001 Elsevier Science B.V. All rights reserved.
PACS: 13.35.Dx; 13.35.Hb; 14.60.Pq; 14.60.St
Keywords: Sterile neutrino; Lepton flavor violation; Neutrino decay
1. Introduction
Hypothetical sterile neutrinos s , blind to the electroweak interactions, were invoked
into particle physics in various contexts as possible means to resolve observed anomalies.
The most prominent case deals with neutrino anomalies: solar neutrino deficit, atmospheric
neutrino anomaly and results of the LSND neutrino experiment. Simultaneous explanation
of all these three anomalies in terms of neutrino oscillations requires at least one sterile
neutrino s [1,2]. In this case s together with e,, allows one to introduce three
independent mass square differences associated with these three anomalies. Although the
recent Super-Kamiokande global analysis [3] disfavours active-sterile neutrino oscillations
as a dominant channel this hypothesis is not ruled out. Along this line one may reasonably
admit the existence of more than one sterile neutrino, say, one per generation, as it is
suggested by various extensions of the SM. In this situation among the neutrino mass
eigenstates, which are superpositions of the active e,, and sterile s(i) weak eigenstates,
one can encounter not only light neutrinos, but also heavy states h . The question of
viability of this scenario is the subject of experimental searches as well as cosmological
and astrophysical constrains.
E-mail address: kovalenko@fis.utfsm.cl (S. Kovalenko).
1 On leave from the Joint Institute for Nuclear Research, Dubna, Russia.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 6 9 - 9
356
V. Gribanov et al. / Nuclear Physics B 607 (2001) 355368
Recently some phenomenological, cosmological and astrophysical issues of the intermediate mass neutrinos h in the MeV mass region have been addressed in the literature [4].
This was stimulated by the attempts of explanation of the KARMEN anomaly [5] in terms
of a neutrino state with a mass of 33.9 MeV mixed with [6]. Although recent data of
this collaboration [7] have not confirmed this anomaly, the question of existence of heavy
sterile neutrinos h remains open.
These sterile neutrinos can be searched for as peaks in differential rates of various
processes and by direct production of h followed by their decays in the detector (for
summary see Ref. [8], p. 361). The h can also give rise to significant enhancement of
the total rate of certain processes if their masses happen to be located in an appropriate
region [9]. This effect would be especially pronounced in reactions that are forbidden in
the SM. The lepton number/flavor violating (LNV/LFV) processes belong to this category.
Many of them are stringently restricted by experiment and allow one to derive stringent
limits on the h contribution.
In the present paper we study the h contribution to the LFV and LNV -decays:
e ( ) + and e+ (+ ) . The first (LFV) process can receive
contribution both from Dirac and Majorana neutrinos while to the second (LNV) process
only Majorana neutrinos can contribute. These processes are capable to provide us
with unique information on the h mixing matrix element U h . In the mass region
140.5 MeV mh 1637 MeV, which we are going to study, the h contribution to
the considered -decays gains resonant enhancement [9]. This effect makes discussed decays very sensitive to presence of h with the masses mh in the resonant region. Under
certain assumptions we extract from the experimental data [10] on these -decays new
constraints on the U h matrix element in the resonant region for both Majorana and Dirac
heavy sterile neutrinos h . Up to recently there were no credible laboratory limits on this
quantity in the MeV mass region [8]. Only recent data of NOMAD Collaboration [11]
allowed establishing constraints on U h in the MeV mass region. In that part of the resonant
region, which overlaps with the region probed by NOMAD, our constraints are more
stringent by around two orders of magnitude.
2. Neutrino mass matrix and interactions

Consider an extension of the SM with the three left-handed weak doublet neutrinos
=
right-handed neutrinos Ri
can be written as
= ( , , ) and n species of the SM singlet

Li
Le L L
, . . . , ). The general mass term for this set of fields
(R1
Rn

ML
1
1
M() c + H.c. = L , Rc
MTD
2
2
MD
MR
1
mi c i i + H.c.
2
Lc
R

+ H.c.
3+n
i=1
(1)
357
Here ML , MR are 3 3 and n n symmetric Majorana mass matrices, MD is 3 n

Dirac type matrix. Rotating the neutrino mass matrix by the unitary transformation to the
diagonal form
U T M() U = Diag{mi }
(2)
with the masses m . In special

we end up with n + 3 Majorana neutrinos i = Uki
i
k
cases there may appear among them pairs with masses degenerate in absolute values. Each
of these pairs can be collected into a Dirac neutrino field. This situation corresponds to
conservation of certain lepton numbers assigned to these Dirac fields.
The considered generic model must contain at least three observable light neutrinos
while the other states may be of arbitrary mass. In particular, they may include hundred
MeV neutrinos h , which we are going to consider in the next section. Presence or absence
of these neutrino states is a question for experimental searches.
In this scenario neutrino mass eigenstates have in general non-diagonal LFV couplings
to the Z-boson
PL = Z
n+3

Ui Uj
j PL i
=e,, i,j =1
=e,,
n+3

ij j PL i ,
(3)
i,j =1
where the last two expressions are written in the mass eigenstate basis and PL = (1 5)/2.
For the case of only three massive neutrinos one has ij = ij as a consequence of unitarity
of Ui . In general ij is not a diagonal matrix and flavor changing neutral currents in the
neutrino sector become possible at tree level.
Z is an efficient counter of the number of
As is known the invisible Z-boson width inv
neutrinos. In the introduced scenario, despite possible presence of an arbitrary number of
Z is always consistent with its experimental
light neutrinos, the predicted value of the inv
value. Indeed, if all the neutrinos are light with masses mi MZ their contribution to the
invisible Z-boson width can be written as
Z
=
inv
n+3

i,j =1
|ij |2 SM = SM
= 3SM .
(4)
,=e,,
With the SM prediction for the partial Z-decay width to a pair of light neutrinos SM =
Z =
167.24 0.08 MeV this formula always reproduces the experimental value of inv
498.8 1.5 MeV. The chain of equalities in Eq. (4) follows again from the unitarity of
Un . Thus, independently of the number of light neutrinos with masses m MZ the
factor 3 in the last step counts the number of weak doublet neutrinos. This conclusion is
changed in the presence of heavy neutrinos N with masses MN > MZ /2 which do not
contribute to inv . In this case the unitarity condition is no longer valid and the factor 3 is
changed to a smaller value.
358
Fig. 1. The lowest order diagrams contributing to l + + decay.
Having these arguments in mind we introduce in the next section neutrino states h with
masses in the hundred MeV region. These states can be composed of sterile and active
neutrino flavors as described in the present section.
3. e ( ) decay rates
Neutrinos contribute to the LNV -decay l + according to the lowest
order diagrams shown in Fig. 1. The -contribution to the LFV -decay l +
is determined by the tree-level diagram similar to the diagram in Fig. 1(a). The loopdiagram analogous to the diagram in Fig. 1(b) is absent in the latter case. The diagram
in Fig. 1(b) requires knowledge of the -meson wave function at long distances. As yet it
is poorly known and would introduce into the calculations uncontrollable uncertainties. In
the present paper we concentrate on the resonant mass domain determined in (16), where
the diagram in Fig. 1(a) absolutely dominates over that in Fig. 1(b). For this reason we
neglect this diagram latter on. The hadronic structure parameters necessary for calculation
of the tree-level contribution Fig. 1(a) involves only one parameter of hadronic structure,
the pion decay constant f , accurately known from experiment. For the l
decays with l = e, we derive the following decay rate formula

l+
s +
=c
sl

Ulk U k mk 2 l s

ds
G
m2
s m2k
k
c
+ 2 2 Re
m
s +
ds

Ulk U k mk
s m2k
sl
+
vl
vl
dv

Uln U n mn
n
v m2n

H
s
v
, 2
2
m m

,
(5)

l + = c
s +
sl

Ulk U k 2 l s
ds
sG
.
2
m2
s
m
k
k
359
(6)
The unitary mixing matrix Uij relates i = Uij h weak and mass neutrino eigenstates.
The numerical constant is c = (G4F /32)()3 f4 m5 |Vud |4 , where f = 93 MeV and m =
1777 MeV is the -lepton mass. We introduced the functions

2

Gl (z) = l (z)z2 z xl2 x2 z + xl2 (z 1)2 x2 (z + 1) ,

H l (z1 , z2 ) = 1 + xl2 z1 z2 xl2 z1 z2

+ x2 1 + xl2 z1 + z2 x2 2 z1 z2 + xl2 ,
(7)
where xi = mi /m , (x, y, z) = x 2 + y 2 + z2 2xy 2yz 2xz, l (z) = 1/2 (1, z, x2 )
1/2 (z, x2 , xl2 ). The integration limits in Eqs. (5), (6) are
sl = m2 (x + xl )2 ,
vl =
s + = m2 (1 x )2 ,

m2
2
x xl2 x2 1 + y 1 + xl2 + 2x2 y 2 l (y)
2y
(8)
with y = s/m2 .
Assuming that there exist only light neutrinos with masses mi m we can rewrite
Eqs. (5), (6) approximately in the form

|m l |2 (l)
A+ ,
l+ = c
m2

|m2 l |2 (l)
l + = c
A ,
m4
(9)
where
s +
(l)
A+ = m2
sl
A(l)
s +
= m4
sl
and
m l =

k
ds l s
G
s2
m2
s +

+2
sl
ds
s
vl
vl

dv l s
v
H
,
,
v
m2 m2

ds l s
G
s3
m2
Ulk U k mk ,
(10)
m2
Ulk U k m2k .
(11)
From atmospheric and solar neutrino oscillation data, combined with the tritium beta decay
endpoint, one can derive upper bounds on masses of all the three neutrinos [12] me,,
3 eV. Then we have conservatively [13]

2

m < 27 eV2 .
m l < 9 eV,
(12)
l
360
With these upper bounds we obtain from Eqs. (9) a rough estimate
( e+ (+ ) )
3.9(2.6) 1031,
( All)
( e ( ) + )
3.0(1.2) 1047.
Re ( )
( All)
Re+ (+ )
(13)
(14)
This is to be compared with the present experimental bounds on these branching ratios [8]
Re+ (+ ) 1.9(3.4) 106 ,
exp
Re ( ) 2.2(8.2) 106 ,
exp
(15)
The comparison clearly shows that the light neutrino contributions to the processes
l are far from being detected in the near future. On the other hand experimental
observation of l decays at larger rates would indicate some new physics
beyond the SM, or the presence of an extra neutrino state h with the mass in the hundred
MeV domain.
Assume that there exist neutrinos h with masses mh in the interval

(16)
se 140.5 MeV mh s + 1637 MeV.
These neutrinos would give resonant contributions to the processes l since
the first term in Eq. (5) and the expression in Eq. (6) have non-integrable singularities at
s = m2h . Apparently, in the resonant region (16) one has to take into account the finite width
h of the neutrino h . This is accomplished by the substitution mh mh (i/2)h in
Eqs. (5), (6). As we will show in the next sections the total decay width of the neutrino
is small h mh . Therefore, in the resonant domain (16) the neutrino propagator in
Eqs. (5), (6) has a very sharp maximum at s = m2h . The second term in Eq. (5), being finite
in the limit h = 0, can be neglected in this case. This allows us to rewrite Eqs. (5), (6) in
the resonant mass region with good precision in a simple form

mh |U h |2 |Ulh |2
res l cGl (z0 )
,
h
(17)
with z0 = (mh /m )2 .
4. Heavy sterile neutrino decays

Heavy sterile neutrinos h can decay in both charged (CC) and neutral (NC) current
i M(q q),
where l = e, and M are
channels h l1 l2 l2 ; lM(q1q2 ) and h i l l;
mesons specified below. These decays are induced by the Lagrangian terms
g2
g2
PL i W +
Uj Ui
i PL j Z ,
L = Uli l
2 cos W
2
(18)
where l = e, , and U is the neutrino mixing matrix defined in Eq. (2). Despite the
fact that the NC term is of second order in the mixing matrix the NC decay channels of the
heavy sterile neutrinos h , as will be seen, are as important as the CC one.
Calculating the total width h we divide the interval 16 into two parts

se 140.5 MeV mh m = 958 MeV, domain (I),
m = 958 MeV < mh s + 1637 MeV, domain (II),
361
(19)
where m = 958 MeV is the mass of the isoscalar pseudo-scalar meson (958).
In the domain (I) we obtain h directly, by calculating all the h decay channels to the
leptons l and mesons M one by one, using the following meson matrix elements
5 u|0 = i 2f p ,
K (p)|s 5 u|0 = i 2fK p ,
(p)|d
5 u|0 = 2m f p 9 ,
(p)|d
K (p)|s 5 u|0 = mK fK p 9 ,
0
0
5 d|0 = if p ,
5 u|0 = (p)d
(p)u
f
5 d|0 = 1 (p)|s 5 s|0 = i
p ,
(p)|u
5 u|0 = (p)|d
2
3
0
0
d|0 = m f 9 ,
u|0 = (p)d
(p)u
d|0 = m f 9 ,
(p)|u
u|0 = (p)|d
(20)
where 9 is a vector meson polarization 4-vector; meson decay constants and masses are
f = 93 MeV, fK = 113 MeV, f = 153 MeV, fK = 224 MeV, f = 138 MeV, m =
139.6 MeV, mK = 494 MeV, m = 547 MeV, m = 770 MeV, m = 782 MeV, mK =
892 MeV.
This channel-by-channel approach cannot be extended beyond the threshold of (985)
meson production since its decay constant f is unknown. Above the (985) threshold,
in the domain (II), we approximate the decay rate of h l()M by the decay rate
of the inclusive process h l()q1 q2 without perturbative and nonperturbative QCD
corrections. This leading-order approximation should work for mh 200 MeV. At
lower masses a more viable approach would be to relate the semileptonic h decay rate
by the dispersion relations to the imaginary parts of the W and Z self-energies (s) in
analogy to the approach applied in the literature for the + hadrons inclusive decay
[14]. However for our rough estimations we do not need this more sophisticated treatment
and will use the above mentioned leading-order approximation.
The decay channels of the Dirac neutrino h in the domain (I) are
+
e e e , e + , e+ e , + ,
(CC) : h e + , + , e K + , K + ,
(21)
+ + + +
e , ,e K , K ,

i e e+ , i + , i 0 ,
(NC) : h
(22)
i , i 0 , i ,
where i = 1 , 2 , 3 are the three conventional light neutrino mass eigenstates. Latter on
we neglect their masses since mi mh .
Partial decay widths for these channels are readily calculated using the hadronic matrix
elements in Eq. (20). The results can be summarized as
362
(h l1 l2 ) = |Ul1 h |2
G2F
(l l )
m5h H (yl1 , yl2 ) |Ul1 h |2 3 1 2 ,
3
192
G2F 2 3
(lP )
f m FP (yl , yP ) |Ulh |2 2 ,
4 P h
G2
(lV )
(h lV ) = |Ulh |2 V F fV2 m3h FV (yl , yV ) |Ulh |2 2 ,
8
m5 G2
(l)
(l)
(l)
(h i l l ) = |i |2 h F3 aH1 + bH2 |i |2 3 ,
192

2

m 3 G2
0
(P 0 )
,
h i P 0 = |i |2 h F f2 P 1 yP2 0 |i |2 2
64

2

m 3 G2
h i V 0 = |i |2 h F fV2 0 1 yV2 0 1 + 2yV2 0
16
(h lP ) = |Ulh |2
(V 0 )
|i |2 2
where
i =
(23)
Uh Ui
.
(24)
=e,,
In Eq. (23) we denoted P = { + , K + }, V = { + , K + }, P 0 = { 0 , }, V 0 = { 0 , },

0
yi = mi /mh , = 1, = 1/3, = 2, K = 1, a = 2 sin4 W sin2 W + 1/4, b =

sin2 W (sin2 W 1/2). The kinematical functions are
z2
H (x, y) = 12

dz
z y 2 1 + x 2 z 1/2 1, z, x 2 1/2 0, y 2, z ,
z
z1
H1(l) = 6
1
dz
(2z wl )(1 z)2 z wl ,
z
wl
H2(l)
1
= 6wl
dz
(1 z)2 z wl ,
z
wl

FP (x, y) = 1/2 1, x 2 , y 2 1 + x 2 1 + x 2 y 2 r 4x 2 ,

2

FV (x, y) = 1/2 1, x 2, y 2 1 x 2 + 1 + x 2 y 2 2y 4 ,
where the integration limits are z1 = y 2 , z2 = (1 x)2 and w = 4yl2 .
In the mass domain (II) of Eq. (19) we approximate the CC and NC decay rates by the
rates of the inclusive reactions
(CC) :
h l q1 q2 ,
(NC) :
h i q q,
(25)
where l = {e, } and q = {u, d, s}. The corresponding leading order decay rate formulas
are
(h l q1 q2 ) = |Ulh |2

G2F 5
m |Vud |2 + |Vus |2 |Ulh |2 (lX) ,
64 3 h

G2F
m5h 9 16 sin2 W 1 sin2 W
3
3072
|i |2 (X).
363
= |i |2
(h i q q)
(26)
d c contributions.
Here we neglected small cd,
NC is in order. Summation over all
The following comment on the total NC decay rate h
the NC channels gives
NC
= h
h
3

|i |2 = h
i=1
= h
Uh Uh
,=e,,
3

Ui
Ui
i=1
Uh Uh
( Uh
Uh ) h
|Uh |2 .
(27)
=e,,
,=e,,
Here we assumed that there are only four neutrino mass eigenstates 1,2,3 , h and used
unitarity of the mixing matrix U . At the last step we neglected the subdominant term
4 . Thus, both CC and NC decay rates are proportional to |U |2 and, as we
Uh
h
mentioned at the beginning of this section, a priori are equally important.
Taking into account this fact we collect the partial decay rates into the total h -decay
width
(e)
()
()
()
()
h = |Ueh |2 h + h + |Uh |2 h + h + |U h |2 h
(28)
with
(lM)
(e)

(l)
= 3 + 3(ll) + (m mh )
2
+ (mh m ) (lX) ,
h
(29)
(M 0 )
(e)
()
()
+ (m mh )
2
+ (mh m ) (X)
h = 3 + 3
(30)
M0
where the summations run over M = + , K + , + , K + and M 0 = 0 , , 0 , .

If neutrinos h are Majorana particles, i.e., h hc , then both h l X(CL = 0) and
h l + Xc (CL = 2) decay channels are open. This results in multiplication of the righthand side of Eq. (29), (30) by factor 2.
5. Bounds on heavy sterile neutrino masses and mixings

Substituting the total h -decay width from Eq. (28) into the resonant formula (17), we
can derive, from the experimental bounds (15), constraints on the h neutrino mass mh and
the mixing matrix elements Uh . In general these constraints represent a hardly readable 4dimensional exclusion plot. However, under certain simplifying assumptions one can infer
more valuable information on the individual size of the mixing matrix elements. In this
paper we are interested in the U h matrix element which is not constrained in the literature
in the h mass range (16) (see [8]). Recently the NOMAD Collaboration [11] obtained
constraints for mh < 190 MeV which overlap with a small part of this range. Therefore, it
would be interesting to infer individual constraints on this mixing matrix element, at least
364
Fig. 2. Exclusion plots in the plane |U h |2 mh . Here mh and U h are the heavy neutrino h mass
and its mixing matrix element to , respectively. The shaded regions are excluded by the big-bang
nucleosynthesis, the duration of the SN 1987A neutrino burst [4], by the NOMAD Collaboration [11]
and by e ( ) decay [present result]. The crosses () mark the doubtful points where
our limits have to be taken with caution (see the text).
roughly. A reasonable simplifying assumption would be to take |U h | |Uh | |Ueh |.

Then from the experimental bounds (15) we derive a 2-dimensional mh |U h |2 exclusion
plot given in Fig. 2 for the case of Dirac h . Multiplying the Dirac case limiting curve
in Fig. 2 by the factor 2 one obtains the exclusion plot for the case of Majorana h . For
comparison we also present in Fig. 2 the bounds from big-bang nucleosynthesis, SN1987A
[4] and the NOMAD bounds [11]. The cosmological bounds are shortly discussed in
Section 6.
Note that at the points mh = 200, 300, 400 MeV the limits in Fig. 2 look incompatible
with our assumption that |U h | |Uh | |Ueh | since other experiments [8] claim the
upper limits on |Ueh |2 , |Uh |2 to be around 108 . In this respect we stress the following.
The latter limits refer to rather narrow vicinities of the given points on the mh axis and
have no impact on the main part of our exclusion plot. However, in the vicinities of these
points no limits on |U h |2 can be established from the l decays at present
experimental sensitivity (15). Indeed, the above limits |Ueh |2 , |Uh |2 108 and Fig. 2
imply |U h | |Uh | |Ueh | rather than |U h | |Uh | |Ueh | which we used in our
analysis. In this case, as seen from Eqs. (17), (28), the matrix element |U h |2 drops out
from the analysis unconstrained. We marked these doubtful points on our exclusion plot
in Fig. 2 by crosses (). Let us stress, however, that the complications around these points
can be ignored at present. This is because the above cited limits on |Ueh |2 , |Uh |2 [8] were
extracted from the h decay searches neglecting the neutral current (NC) decay channels
given in Eq. (22). However we have shown that these channels are as important as the
365
charged current (CC) ones. It is crucial that the NC channels of h decay introduce in
the analysis all the mixing matrix elements |Ueh |2 , |Uh |2 , |U h |2 . Thus, for extracting
individual constraints on |Uih |2 one needs certain assumptions on their relative size like
those we used in our extraction procedure. As a result the actual upper bounds on |Ueh |2 ,
|Uh |2 may differ from the presently quoted values 108 [8].
Let us finally note that all known individual constraints on |Ulh | [8] have been obtained
under certain simplifying assumptions of this type.
However, there is a special case when limits on |U h | can be extracted without any
simplifying assumptions. This case is realized when there are experimental upper bounds
on the rates of three processes involving , and e lepton pairs. These could be, for
instance, decays
Bc+ + + , + + , + e+ .
(31)
Intersection of their resonant regions is 1917 MeV mh 4620 MeV. In this domain the
decay rates can be written schematically as
=
e =
|2
|U h |4
,
+ b|Uh |2 + c|U h |2
|2
|U h
eh
.
+ b|Uh |2 + c|U h |2
a|Ueh
|2 |U
a|Ueh
a|Ueh
|2
|2
|U h |2 |Uh |2
,
+ b|Uh |2 + c|U h |2
(32)
From these equations we find

|U h |2 = a e + b + c .
(33)
Notice that all the decay rates appear with positive degree. Only in such a case we can set
upper bound on |U h |2 having upper experimental bounds l Expl . This would not be
the case, for instance, in the system , e , ee .
At present the described approach can not be realized on practice due to absence of
necessary experimental data. However the required upper limits on the semileptonic Bc decays can be, probably, obtained in future experiments at the B-factories. Then, derivation
of limits on |U h | free of simplifying ad hoc assumptions will become possible.
6. Heavy sterile neutrinos in astrophysics and cosmology

It is well known that massive neutrinos may have important cosmological and
astrophysical implications. They are expected to contribute to the mass density of the
universe, participate in cosmic structure formation, big-bang nucleosynthesis, supernova
explosions, imprint themselves in the cosmic microwave background, etc. (for a review see,
for instance, Ref. [15]). This implies certain constrains on the neutrino masses and mixings.
Currently, for massive neutrinos in the mass region (16), the only available cosmological
constraints arise from the mass density of the universe and cosmic structure formation.
The contribution of stable massive neutrinos to the mass density of the universe is
described by the Lee-Weinberg h2 m curve. From the requirement that the universe
366
is not overclosed this leads to the two well know solutions m 40 eV and m 10 GeV
which seem to exclude the domain Eq. (16). However for unstable neutrinos the situation
is different. They may decay early to light particles and, therefore, their total energy can be
significantly redshifted down to the overclosing limit. Constraints on the neutrino life
times h and masses mh in this scenario are found in Ref. [16]. In the mass region (16) we
have an order of magnitude estimate
h < 1014 s,
mass density limit.
(34)
Decaying massive neutrinos may also have specific impact on the cosmic structure
formation introducing new stages in the evolution of the universe. After they decay
into light relativistic particles the universe returns for a while from the matter to the
radiation domination phase. This may change the resulting density fluctuation spectrum
since the primordial fluctuations grow due to gravitation instability during the matter
dominated stages. Comparison with observations leads to an upper bound on the neutrino
life time [17]. In the mass region (16) one finds roughly
h < 107 s,
structure formation limit.
(35)
On the other hand, from our formula (28) we estimate

1012 s < h ,
theoretical limit.
(36)
Thus massive neutrinos with masses in the interval (16) are not yet excluded by the known
cosmological constraints (34), (35).
Big-bang nucleosynthesis (BBN) and the SN 1987A neutrino signal may presumably
lead to much more restrictive constraints [4]. The impact of heavy sterile neutrinos h
on the BBN is twofold: via their contribution to the cosmic energy density and via the
secondary light neutrinos e , , produced in h -decays. The first issue results in faster
expansion which enhances the neutron-to-proton ration n/p. The secondary light neutrinos
can produce also an opposite effect. The increased number of e maintains neutron
proton thermal equilibrium longer diminishing the n/p-ratio at the freeze-out point. There
are also effects related to the e energy spectrum distortion and energy injection to the
electromagnetic radiation from the e+ e pairs produced in h -decays [4]. The observed
duration of the neutrino signal from the SN 1987A stringently constraints the energy
emitted from the core in the form of penetrating particles like neutrinos. The sterile
neutrinos could affect this energy-loss limits and, therefore, their mixings with the active
neutrinos have to obey certain constraints. Analysis of the BBN and SN 1987A constraints
on h parameters has been made in Ref. [4]. We present the corresponding exclusion plots
in Fig. 2 for comparison with our limits. Unfortunately, the analysis of Ref. [4] is restricted
to the mass range mh < 200 MeV for the BBN and mh < 100 MeV for the SN 1987A case.
7. Conclusion
We discussed a generic model with sterile neutrinos which can contain heavy i mass
eigenstates. If among them there are h states with masses 140.5 MeV mh 1637 MeV
367
they can resonantly contribute to e ( ) decays.

We derived decay rates for these processes valid both outside and inside of the
resonant mh region (16). In the latter case knowledge of the total h -decay width h is
required. We calculated the h in the whole resonant mh region (16) taking into account
both charged and neutral current decay channels.
The effect of resonant enhancement of the heavy sterile neutrino h contribution allowed
us to extract from the experimental upper bounds on the -decay rates rather stringent
constraints on h mixing matrix element |U h | in the whole resonant region (16). The
corresponding exclusion plot is presented in Fig. 2 together with the recent constraints from
NOMAD Collaboration [11] and cosmological constraints [4]. The -decay constraints
were derived under certain simplifying assumptions about h -mixing with the other
flavors. These or similar assumptions are always required when h -mixing with one
selected flavor is extracted from those processes which have been studied in the
literature. We propose the set of processes Bc , , e, , ee which would
allow one to avoid such ad hoc assumptions and extract individual constraints on |U h |,
|Uh |, |Ueh |. Finally, we discussed cosmological implications of the heavy sterile neutrinos
h and pointed out that extension of the known big bang nucleosynthesis and the SN 1987A
constraints [4] to the resonant region 16 could significantly strengthen our constraints on
|U h | by combining them with these cosmological constraints.
Acknowledgements
We thank Ya. Burdanov, G. Cvetic, C. Dib and O. Espinosa for fruitful discussions.
We are grateful to S. Gninenko for helpful comments on the recent NOMAD results [11].
This work was supported in part by Fondecyt (Chile) under grant 8000017, by a Ctedra
Presidencial (Chile) and by RFBR (Russia) under grant 00-02-17587.
References
[1] S.M. Bilenky, G. Giunti, W. Grimus, Eur. Phys. J. C 1 (1998) 247;
S.M. Bilenky et al., Phys. Rev. D 60 (1999) 073007;
V. Barger, T.J. Weiler, K. Whisnant, Phys. Lett. B 427 (1998) 97;
V. Barger, S. Pakvasa, T.J. Weiler, K. Whisnant, Phys. Rev. D 58 (1998) 093016.
[2] J. Conrad, Nucl. Phys. A 663664 (2000) 210;
J. Ellis, Nucl. Phys. A 663664 (2000) 210;
K. Zuber, Phys. Rep. 305 (1998) 295, and references therein.
[3] Super-Kamiokande Collaboration, S. Fukuda et al., Phys. Rev. Lett. 85 (2000) 3999.
[4] A.D. Dolgov, S.H. Hansen, G. Raffelt, D.V. Semikoz, Nucl. Phys. B 590 (2000) 562;
A.D. Dolgov, S.H. Hansen, G. Raffelt, D.V. Semikoz, Nucl. Phys. B 580 (2000) 331, and
references therein.
[5] KARMEN Collaboration, B. Armbruster et al., Phys. Lett. B 348 (1995) 19;
KARMEN Collaboration, K. Eitel et al., Nucl. Phys. Proc. Suppl. 77 (1999) 212.
[6] V. Barger, R.J.N. Phillips, S. Sarkar, Phys. Lett. B 352 (1995) 365;
V. Barger, R.J.N. Phillips, S. Sarkar, Phys. Lett. B 356 (1995) 617, Erratum;
368
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
J. Govaerts, J. Deutsch, P.M. Van Hove, Phys. Lett. B 389 (1996) 700;
D.V. Ahluwalia, T. Goldman, Phys. Rev. D 56 (1997) 1698.
K. Eitel, Proceedings of Neutrino2000, hep-ex/0008002.
D.E. Groom et al., Eur. Phys. J. C 15 (2000) 1.
C. Dib, V. Gribanov, S. Kovalenko, I. Schmidt, Phys. Lett. B 493 (2000) 82.
CLEO Collaboration, D. Bliss et al., Phys. Rev. D 57 (1998) 59035907.
NOMAD Collaboration, P. Astier et al., hep-ex/0101041.
F. Vissani, hep-ph/9708483;
V. Barger, T.J. Weiler, K. Whisnant, Phys. Lett. B 442 (1998) 255.
C. Dib, V. Gribanov, S. Kovalenko, I. Schmidt, to appear in the Proceedings of NANPino
Conference, Dubna, Russia, July 1922, 2000, hep-ph/0011213.
E. Braaten, S. Narison, A. Pich, Nucl. Phys. B 373 (1992) 581;
G. Cvetic, T. Lee, hep-ph/0101297, and references therein.
G.G. Raffelt, Stars as Laboratories for Fundamental Physics, University of Chicago Press, 1996;
G.G. Raffelt, Annu. Rev. Nucl. Part. Sci. 49 (1999).
D.A. Dicus, E.W. Kolb, V.L. Teplitz, Phys. Rev. Lett. 39 (1977) 169.
G. Steigman, M.S. Turner, Nucl. Phys. B 253 (1985) 253;
J. Bardeen, J. Bond, G. Efstathiou, Astrophys. J. 321 (1987) 28;
L.M. Kraus, Phys. Lett. B 263 (1991) 441;
J. Bond, G. Efstathiou, Phys. Lett. B 265 (1991) 245;
S. Dodelson, G. Gyuk, M.S. Turner, Phys. Rev. Lett. 72 (1994) 3754;
M. White, G. Gelmini, J. Silk, Phys. Rev. D 51 (1995) 2669.

A strategy for the analysis of semi-inclusive deep

inelastic scattering
Ekaterina Christova a,1 , Elliot Leader b,2
a Theory Division, CERN, CH 1211 Geneva 23, Switzerland
b Vrije Universiteit, Amsterdam, The Netherlands
Received 3 August 2000; accepted 14 May 2001
Abstract
We present a strategy for the systematic extraction of a vast amount of detailed information on
polarized parton densities and fragmentation functions from semi-inclusive deep inelastic scattering
l + N l + h + X, in both LO and NLO QCD. A method is suggested for estimating the errors
involved in the much simpler, and therefore much more attractive, LO analysis. The approach is
based upon a novel interplay with data from inclusive DIS and from e+ e hX, and leads to a
much simplified form of the NLO expressions. No assumptions are made about the equality of any
parton densities and the only symmetries utilised are charge conjugation invariance and isotopic spin
invariance of strong interactions. 2001 Elsevier Science B.V. All rights reserved.
1. Introduction
There are two major problems in the QCD analysis of polarized inclusive deep inelastic
scattering (DIS): (i) the absence of neutrino data makes it impossible, in principle, to
Q2 ), (ii) the
determine the non-strange polarized sea-quark densities u(x,
Q2 ) and d(x,
2
separate determination of the polarized strange quark density s(x, Q ) and the polarized
gluon density G(x, Q2 ) relies heavily on the QCD evolution in Q2 and use of the flavour
SU(3)F -invariance relation
1
dx [u + u + d + d 2s 2s ] = 3F D.
(1)
The absence of a long lever arm in Q2 , in the polarized case, and doubt concerning the
reliability of SU(3)F -invariance for hyperon -decay means that s and G are still
E-mail addresses: echristo@inrne.bas.bg (E. Christova), e.leader@ic.ac.uk (E. Leader).
1 Permanent address: Institute of Nuclear Research and Nuclear Energy, Sofia, Bulgaria.
2 Permanent address: High Energy Physics Group, Imperial College, London, UK.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 2 4 4 - 9
370
E. Christova, E. Leader / Nuclear Physics B 607 (2001) 369390
rather poorly known [1], despite the dramatic improvement in the quality of the data in the
past few years.
The direct resolution of these problems must await a series of new machine development
projects, based on very high intensity neutrino beams, which are most unlikely to come into
operation before the year 2015.
In the meantime there is currently a major experimental effort at CERN [2], HERA [3]
and Jefferson Lab. to study semi-inclusive polarized DIS reactions
e + N e + h + X,
(2)
involving the detection of the produced hadron h.

The theoretical structure for the analysis of such reactions, in both leading order QCD
(LO) and next-to-leading order (NLO), exists [4]. However we are critical of the type of
LO analysis carried out thus far by the experimental groups.
The analysis of semi-inclusive DIS [2,3] involves both parton densities and fragmentation functions. In the LO treatments referred to above, the fragmentation functions (FFs)
are treated as known quantities from inclusive e+ e hadrons, and are used in constructing an auxiliary quantity called purity. However, it is well known that in e+ e
hadrons, both in LO and NLO, first, only the combinations Dqh + Dqh can be determined
while for the analyses of semi-inclusive DIS both Dqh and Dqh are needed separately, and
second, the existing analysis are rather ambiguous: a detailed study in [5] makes a 31 parameter fit to the data, and no errors are quoted, and, in a more recent study [6], the FFs differ
significantly from those in [5], by 40% or more in some regions of z. Under these circumstances it is unreasonable to pretend to have an absolute knowledge of the fragmentation
functions.
Two NLO analyses based upon a global analysis of the inclusive and semi-inclusive data
have been attempted [7,8]. In the more recent analysis [8] the authors relax the equality of
u(x)
and d(x)
and find a preference for a positive u(x),
but effectively no constraint
on the sign of d(x).

However, in both these analyses it is again assumed that the FFs are
known exactly: those of [5] being used in [7], and those of [6] in [8].
In addition, in the above mentioned analysis, some simplifying assumptions are made
about relations between various polarized parton densities. In the following, except where
expressly indicated, we make no assumptions at all concerning the polarized or unpolarized
parton densities. Indeed there are persuasive arguments from the large-Nc limit of QCD
>
that a significant difference should exist between u(x)
and d(x)
[9] with |u d|
and it has been argued that such a situation is compatible with all present day
|u d|,
data [10]. Further, bearing in mind recent arguments [11] that s(x)
= s (x) and s(x)
=
s (x), we even refrain from the very common assumption of the equality of these
densities.
However, from experience gained in the analysis of inclusive polarized DIS, it appears
that the parameter space is sufficiently complicated to be able to produce biases in the
2 analysis, which can lead to unphysical results. We thus believe it to be dangerous in
either LO or NLO QCD, to put together all inclusive and semi-inclusive data in one global
371
analysis. Rather, what is required, is a working strategy, making optimal use of selected
parts of the data.
The aim of this paper is precisely to provide a strategy, in both LO and NLO QCD, for the
analysis of the semi-inclusive data. We proceed as follows: information about the FFs is
obtained from unpolarized semi-inclusive data. Then this information is used to determine
the polarized parton densities from polarized semi-inclusive DIS. Appropriate use of the
information from DIS and e+ e hadrons is made as well. We suggest, for example, that
one should use as input not just a knowledge of the unpolarized parton densities and their
errors, but rather the polarized isotopic combination

(d + d)
q3 x, Q2 = (u + u)
(3)
which is by now very well determined from polarized inclusive data, and which is free
from any influence of s and G. Inclusive e+ e hadrons and DIS are used also to
considerably simplify the NLO expressions for the cross sections of (2).
Further, in stead of dealing with each parton density and FF separately, as is often done
we work with their singlet and non-singlet combinations. As both the parton densities and
the FFs enter the semi-inclusive cross sections, we single out such observables that are
singlet (non-singlet) combinations on both the parton densities and the FFs.
This leads to the fact that we often consider linear combinations of experimentally
measured quantities, and it may be objected that thereby we are dealing with experimental
observables with possibly large errors. It has to be understood that that is a reflection of the
true situation and not an artifact of our approach. To take an absolutely trivial example,
suppose E1 and E2 are two experimentally measured functions used in an attempt to
determine the theoretical functions T1 and T2 , where E1 = T1 + T2 and E2 = T1 T2 .
Now, if it happens that E1 E2 and if we write T1,2 = (1/2)(E1 E2 ) then T2 will be
very poorly determined. This is unavoidable. It does not help to do a best fit to E1 and
E2 with some parametrisations of T1 and T2 . Inherently the result for T2 will have a large
relative error.
Thus we believe that any relatively large errors occurring in our manipulation
of the experimental quantities reflects a genuine and unavoidable imprecision in the
determination of certain theoretical quantities.
In order to minimize systematic errors experimentalists prefer to consider asymmetries
or ratios of cross sections where the detection efficiencies should roughly cancel out, e.g.,
ratios of polarized to unpolarized DIS or ratios of polarized to unpolarized semi-inclusive
for a given detected hadron. We appreciate that this is a fact of experimental life. However
we wish to stress that a large amount of information is lost in restricting oneself to only
these ratios. It is vitally important to gain control of the systematic errors in detection
efficiencies, and although we try as far as possible to deal with the favoured kind of ratios
we will be forced also to consider other types of cross-section ratios.
Throughout the rest of this paper we assume that a kinematic separation is possible
between hadrons produced in the current fragmentation region and those produced from
the target remnants. We consider only the current fragmentation region so that our formulae
apply only to this region and fracture functions [13] play no role in our discussion.
372
It has to be stressed that there is a huge difference in complexity between the LO and
NLO treatments. Thus it makes sense to utilize the LO approach, provided appropriate
checks, (which we suggest) are carried out.
Our analysis proceeds in a step-by-step fashion. Firstly we describe a generic test for the
reliability of a LO treatment. Assuming this to be successful we present the analysis in LO.
In LO the information on the polarized valence quark densities uV and dV and on
the breaking of SU(2) invariance of the polarized sea is obtained without any knowledge
of the FFs. However, it should be clear that any information on the sea-quark densities
in LO cannot be reliable, since they are expected to be small and thus comparable to the
NLO corrections. We consider also the experimentally difficult case of production, which
seems to be the best way to get an accurate determination of s + s in LO. We discuss
also how, in principle, one can test s(x) = s (x) and/or s(x) = s (x).
In the NLO treatment we show in a new way how information on e+ e h + X
and inclusive DIS can be incorporated directly so as to considerably simplify the NLO
expressions for semi-inclusive cross sections. This key result is presented in Eqs. (69)
+
and (70). Next we discuss the fragmentation combination Du Du and use it to evaluate
+
+
and for
uV and dV in NLO. Then we obtain expressions for Du + Du , for DG
+
in NLO. Finally we have a set of

Ds + Ds . With these we are able to obtain (u d)
(s + s ) and G.
2 equations involving 3 unknown functions, (u + u + d + d),
An accurate determination of all three functions would require data over a presently
impossibly wide range of Q2 . We thus suggest using here for G its determination from
charm production [12]. It should then finally be possible to get an accurate assessment of
s + s in NLO. Lastly we consider the evaluation of s(x) s(x) and s(x) s (x).
2. A strategy for semi-inclusive DIS
In NLO QCD, the expressions for semi-inclusive cross sections involve convolutions of
parton densities and fragmentation functions with (known) Wilson coefficients. Our lack of
knowledge of the errors on the FFs will make it difficult to assess the accuracy of the parton
densities which we are trying to determine. In the LO QCD approximation, on the other
hand there are no convolutions, but simple products only (independent fragmentation), and
it becomes possible to construct measurable combinations of cross sections in which the
FFs completely cancel out [15,16]. However it is not clear how reliable the LO is.
We believe it is quite safe when determining the large u and d densities, but could
be quite misleading for u,
d and s. In any event it is absolutely essential to test
independent fragmentation in order to have a feeling for the errors on parton densities
obtained via the LO formalism.
In LO, the structure of the expressions is generally of the form

parton density q x, Q2 = experimental observable E x, z, Q2
(4)
or

fragmentation function D z, Q2 = experimental observable E x, z, Q2 .
(5)
373
In both cases the characteristic feature of the LO treatment is that the RH sides, which can
in principle depend on (x, z, Q2 ), should only depend on two of these, either (x, Q2 ) or
(z, Q2 ), respectively, so that there is an independence of the third variable, which we shall
call the passive variable.
Every expression of the form (4) or (5) should be tested for dependence on the passive
variable. If a significant dependence is found it does not mean that the LO analysis must be
abandoned, but it suggests that the variation with the passive variable be used as an estimate
of the theoretical errors, TH [q(x, Q2 )], TH [D(z, Q2 )] on the sought for quantities.
3. Parton densities in LO QCD

It is useful to introduce the following notation for semi-inclusive processes:

3 h
2y 2
d
x(P + l)2
h

4 2
1 + (1 y)2 dx dy dz
and

3 h

h
d ++
d 3 +
x(P + l)2
y

,
4 2
2y
dx dy dz dx dy dz
h
(6)
(7)
where P and l are the nucleon and lepton four momenta, and refer to a lepton of
helicity and a nucleon of helicity . The variables x, y, z are the usual DIS kinematic
variables.
Then in LO the cross sections for the semi-inclusive production of a hadron h have the
simple y-independent form

2

eq qi x, Q2 Dih z, Q2 ,
h x, z, Q2 =
(8)
x, z, Q
q,q

eq2 qi x, Q2 Dih z, Q2 ,
(9)
q,q
where the sum is over quarks and antiquarks, and where Dih is the fragmentation function
for quark or antiquark i to produce h.
We consider sum and difference cross sections for producing h and its charge conjugate
h on both protons and neutrons, and define
h
Ah
p,n
h
h
hh
p,n
p,n

p,n
x, z, Q2 =
,
h
h
hh
p,n
p,n
p,n

h
2
Ah
pn x, z, Q =
phh nhh
(10)
phh nhh
For inclusive unpolarized and polarized DIS cross sections we use the notation:

2 DIS
x(P + l)2
2y 2
d
DIS
2
2
4
1 + (1 y)
dx dy
(11)
(12)
374
and
DIS

3 DIS
DIS
d ++
d 3 +
x(P + l)2
y
.
4 2
2y
dx dy
dx dy
(13)
Then in LO we have the y-independent expressions:

eq2 qi x, Q2 ,
DIS x, Q2 = 2F1N x, Q2 LO =
(14)
q,q

DIS x, Q2 = 2g1N x, Q2 LO =
eq2 qi x, Q2 .
(15)
q,q
In addition to (10) and (11) we consider the ratios of sum and difference hadron yields
for the unpolarized semi-inclusive and inclusive processes:
h
h
p,n

p,n
hh
x, z, Q2 =
,
Rp,n
DIS
p,n
phh nhh
hh
x, z, Q2 = DIS
Rpn
.
p nDIS
(16)
(It is equally good to use a sum over any set of hadrons h and their charge conjugate h.)
3.1. Testing LO QCD
Using only charge conjugation invariance it is easy to show that [16]
p
pDIS nDIS

(g1 g1n )|LO
h
2
2
Ah+
x,
z,
Q
=
x,
Q
=
x, Q2 .
p
pn
n
DIS
DIS
p n
(F1 F1 )|LO
(17)
This is a key relation for testing the reliability of the LO. The RHS is completely known
from inclusive DIS and is independent of z. The LHS, in principle, depends also upon z and
upon the hadron h. Only in LO (or in the simple parton model) should it be independent of
z and of h. It is thus crucial to test this feature.
To help with statistics it is also possible to formulate an integrated version of (17). This
is given in [16]. For the rest of this section we assume that the test (17) has been successful
and proceed with the analysis in LO.
3.2. The valence quark densities in LO
The polarized valence quark densities can be obtained from production, assuming
only isospin invariance:
1
+

4(4uV dV )Ap + (4dV uV )An ,
15
1
+

dV =
(18)
4(4dV uV )An + (4uV dV )Ap .
15
For the case of K or , production, if one makes also the conventional assumption
uV =
that s = s and s = s one has in addition:

+ K
+ K
1
,
uV = (uV + dV )AK
+ (uV dV )AK
p+n
pn
2
dV =
+ K
+ K
1
(uV + dV )AK
.
(uV dV )AK
p+n
pn
2
+
375
(19)
Note that we have not assumed DdK = DdK , although that is suggested by the absence
of a d quark in the leading Fock state of K . Indeed the above equality can be tested (see
Section 5.2).
For isoscalar hadrons like , again assuming s = s and s = s , one finds

1
4(4uV + dV )A
,
(4dV + uV )A
p
n
15

1
4(4dV + uV )A
.
(4uV + dV )A
dV =
(20)
n
p
15
We shall comment in Section 5.5 on the situation if one does not assume s = s and
s = s , where we suggest a method for estimating if the failure of these equalities is
serious or not. In any event, the safe way to obtain uV and dV is via production,
Eq. (18).
Once again the reliability of these LO equations can be tested by checking that the RH
sides of (18), (19) and (20) do not depend on z. If it is found that for a given x-bin the RH
sides vary with z by some amount TH [uV ], TH [dV ], then these could be regarded as
an estimate of the theoretical error at this x value.
uV =
4. Use of q3 (x, Q2 ) from polarized inclusive DIS

At NLO the spin dependent structure functions for protons and neutrons are given by:

1 2
s (Q2 )
s (Q2 )
p
Cq +
G CG
1+
eq (q + q)
g1 x, Q2 =
2
2
2
q=u,d,s
(21)
involving a convolution of polarized parton densities with known Wilson coefficients. For
the neutron, g1n is obtained by the replacement
u d
(22)
in g1 .
Now it is clear that one can obtain information only on the combinations:
u + u,
d + d,
s + s ,
G
(23)
and that it is impossible to obtain separate information on the valence and non-strange
sea-quark polarizations from inclusive, neutral current, polarized DIS [14]. 3
In our semi-inclusive analysis we shall use as a known quantity only the non-singlet
combination of the polarized parton densities q3 , Eq. (3). Now that there is such an
3 However, sometimes it is convenient to parametrize separately the valence and sea-quarks and to assume,
for example, u = d = s, where is a free parameter. It should be obvious then, that any claim that the
2 analysis favors some particular value of must be fictitious and a consequence of some hidden bias in the
minimization procedure. Yet such claims have been made.
376
improvement in the quality of the neutron data we believe this quantity is very well
constrained directly by the inclusive data. For one has, from (21)

1
s (Q2 )
p
2
n
2
g1 x, Q g1 x, Q = q3 1 +
(24)
Cq
6
2
and q3(x, Q2 ) is determined without any influence from the less well known quantities
s and G, either in LO or in NLO. Of course if the semi-inclusive analysis is done in
or
LO one must use q3(x, Q2 )|LO . Note that we do not use information on (u + u)
from inclusive DIS, since these are subject to the less known strange quark and
(d + d)
gluon effects.
4.1. SU(2) symmetry of the sea-quark densities in LO
One has

1
u x, Q2 d x, Q2 LO = q3 x, Q2 + dV x, Q2 uV x, Q2 LO .
2
(25)
Eq. (25) determines the SU(2) symmetry breaking of the polarized sea without requiring
any knowledge of the unknown q and G. A possible test for SU(2) breaking for the
polarized sea densities that does not require any knowledge of the polarized densities was
given in [16].
In order to determine the polarized sea-quark densities separately we need one more
relation, namely the value of

q+ x, Q2 u + u + d + d.
(26)
For then
LO = 1 (q+ uV dV )LO ,
(u + d)
2
which combined with (25) yields u and d separately.
(27)
Although each term on the RHS of (25) and (27) should be well determined in LO, the
corresponding linear combinations are expected to be small and may thus be very sensitive
to NLO corrections. An indication of the sensitivity may be inferred from the fact that in
inclusive polarized DIS, s(x, Q2 ) changes by roughly a factor of 2 in going from LO to
NLO or when one changes factorisation schemes from MS to AB or JET.
uV ] is unreliable since
Note that determining, say, u via u = 12 [(u + u)
determination of (u + u)
from inclusive DIS requires a knowledge of s.
In order to determine q+ (x, Q2 ) it will first be necessary to extract some information
on the FFs.
4.2. Fragmentation functions in LO
h+h
of the semi-inclusive to inclusive DIS cross
(1) From measurements of the ratios Rpn
sections on protons and neutrons for any given hadron h, it is feasible in LO to learn a
377
great deal about the FFs Dqh + Dqh Dqh+h . This combination is measured also in e+ e
hadrons.
Analogous to the polarized case, we define

q3 x, Q2 = u x, Q2 + u x, Q2 d x, Q2 d x, Q2 ,
(28)

2
2
2
2
2
q+ x, Q = u x, Q + u x, Q + d x, Q + d x, Q
(29)
which are well determined from inclusive DIS data and which thus can be taken as known
quantities in the semi-inclusive analysis.
Using data on unpolarized semi-inclusive DIS we have, in LO,

2 h+h

4Du
z, Q2 Ddh+h z, Q2 .
(30)
3
For the case of pions, kaons and , when SU(2) invariance can be used this simplifies
to

+

+

+ +
= 2Du + z, Q2 = 2Dd + z, Q2 .
Rpn
(31)
h+h
Rpn
=
For K mesons and hyperons,

K,+
Rpn
= 2DuK,+ z, Q2 = 2DdK,+ z, Q2 ,
(32)
where the superscript K stands for the sum over all produced kaons:
0.
K K+ + K + K0 + K
(33)
It will also be of great interest to compare these FFs with those obtained from e+ e
hadrons [5,6] and those used in Monte Carlo models.
(2) Given that uV (x, Q2 ) and dV (x, Q2 ) are well determined from inclusive DIS one
can proceed further to obtain the other combinations of FFs Dqh Dqh Dqhh .
One finds for
+
Du
=
=
9 p
4uV dV
+
9 n
4dV uV
18(F1 )LO Rp
4uV dV
+
18(F1n )LO Rn
.
4dV uV
+
(34)
Combined with (31) we have expressions for Du and Du separately.

For K one obtains
+
+

9 pK K nK K
K + K
K + K
4Du
Dd
=
uV d V
p
+
+

18 (F1 )LO RpK K (F1n )LO RnK K
=
.
uV d V
(35)
It is usually assumed, and this is presumably a very good approximation, that

+
DdK = DdK
(36)
378

+
in which case (35) can be read as an expression for DuK K . It is not possible to test
relation (36) without taking s = s and/or s = s , but such an approach is hard to
justify given that any failure of (36) is presumably very small.
For isoscalar hadrons like ,
p
+
+

6 (F1 r)LO Rp (F1n )LO Rn
Du
(37)
= Dd
=
.
uV d V
Given that uV and dV are determined via (18) we can write analogous expression for
the above Dqh functions using the polarized data.
Of course all the expressions (30), (31), (32), (34), (35) and (37), being LO results, must
be tested by demonstrating that the RH sides are essentially independent of the passive
variable x.
+
Now that we have determined Du + we can determine Ds + in LO via

+
Ds +
+
9(F1 + F1n )LO Rp+n
5q+ Du
+ +
2(s + s )
(38)
Similarly, since Du+ is determined via (32), we can find Ds+ from
p
+
9 F1 + F1n LO Rp+n
5q+ Du+
+
.
Ds
=
2(s + s)
(39)
4.3. The non-strange sea-quark densities revisited, in LO

+
Now that we have determined Du + and Ds + in LO we can, in principle,

determine q+ (x, Q2 ) and s(x, Q2 ) + s (x, Q2 ) from the semi-inclusive and inclusive
relations, in LO,
p
g1 + g1n =
+
Ap+n+ =
5q+ + 2(s + s )
,
18
+
5q+ Du + + 2(s + s )Ds +

5q+ Du
+ +
+ 2 (s + s )Ds
+ +
(40)
.
(41)
Such a determination of q+ in LO is likely to be reliable, but s + s and the

individual u and d obtained in LO via (27) may be subject to significant uncertainty.
We note that there is an alternative way to determine q+ , but it requires the ability to
detect K 0 . In that case one has, in LO,
q+ = q+
pK
pK
+ +K (K 0 +K
0)
+ +K (K 0 +K
0)
+ nK
+ nK
+ +K (K 0 +K
0)
+ +K (K 0 +K
0)
(42)
4.4. The strange quark density s + s in LO

The approach to s + s in Section 5.3 and the approach discussed in [16] are unlikely
to be reliable. The problem is that for production of pions the strange quark contribution
is doubly small, since, e.g., one must compare uDu with sDs in which both
379
|s + s | |u| and |Ds | |Du |. For kaons and hyperons it is somewhat better
in that |DuK | |DsK | and |Du+ | = |Dd+ | |Ds+ |, but this is similar to the situation
in inclusive DIS where we know that the LO determination of s + s is quite unreliable.
The only possibility we can see for a reasonable determination of s in LO is via
production. For in this case one has, |s + s | |u| but presumably |Ds | |Du | so
that the strange and non-strange quarks are on equal footing.
One has, by charge conjugation invariance Ds = Ds , and it should be quite safe to take
Du = Du = Dd = Dd . One then obtains in LO

q+
s + s 3 p+n 5 q3 pn
=
.

s + s
3 p+n 5 qq+3 pn
Moreover one has expressions for the fragmentation functions as well:

3
q+
3 p+n 5
pn ,
Ds =
2(s + s )
q3
(43)
(44)
Du
3 pn
q3
(45)
As always expressions (43)(45) must be tested for non-dependence on the relevant passive
variable.
Finally we note that the same equations (43)(45) hold for K, + and -production
if the superscript is replaced by K, + and + + , respectively. And though
the production rates are higher, the sensitivity to the strange quarks in the K, + and
-productions is lower.
4.5. Concerning s = s and s = s in LO 4
In the analysis of DIS it is conventional to assume s = s and s = s . However there
are models and arguments [11] which suggest that these equalities might not hold.
Within the limitations of the LO we can test these relationships via (K + , K ) and
+
production. One has for the unpolarized case, assuming D K K = 0 several

(, )
d
different possibilities:
p
+
(s s )DsK K = 18 F1 LO RpK K 4uV DuK K

+
= 18 F1n LO RnK K 4dV DuK K
+

9 uV nK K dV pK K
=
.
(46)
uV d V
Then for the polarized case, given that uV and dV are known from (18) and
+
(s s)DsK K is determined, we can proceed to determine (s s )DsK K from

4 We are grateful to M. Anselmino, M. Boglione and U. DAlesio for drawing our attention to this issue.
380
AK
p
+ K
AK
p
AK
n
+
or AK
n
+ K
+ K
=
=
+ K
4uV DuK
+ K
4uV DuK
4dV DuK
+ K
+ K
4dV DuK
+ (s s )DsK
+ (s s )DsK
+ (s s)DsK
+ K
+ (s s )DsK
+ K
+ K
+ K
+ K
(47)
DuK K is assumed to be known through (35).

For isoscalar hadrons like , we have the possibilities
p
(s s )Ds = 18 F1 LO Rp (4uV + dV )Du

= 18 F1n LO Rn (4dV + uV )Du

(4uV + dV ) n (4dV + uV ) p ,
=
uV d V
(48)
where Du is assumed to be determined in (37). Here and in Eq. (46) it is

+
clear that NK K and N can be replaced by the corresponding experimentally

+
K K
measured quantities 2(F1N )LO RN

or 2(F1N )LO RN
. Then (s s )Ds can
be determined via Ap

or An :
A
=
p
A
=
n
(4uV + dV )Du + (s s )Ds
(4uV + dV )Du + (s s)Ds
(4dV + uV )Du + (s s )Ds
(49)
.
(4dV + uV )Du + (s s)Ds

Although (46), (47), (48) and (49), being LO expressions, cannot be expected to yield
accurate values for s s and s s , they should nonetheless enable one to say whether
they are compatible with zero since Ds should be relatively large. Note that s s and/or
s s different from zero would break the independence of the RH sides of (47) and
(49) on the passive variable z. Of particular interest is the speculation that s s but
s s , the consistency of which could perhaps be tested from (46), (47), (48) and (49).
5. Semi-inclusive analysis in NLO QCD
The situation in NLO [4,17] is much more complicated than in LO, since factorization
is replaced by convolution, and it is also more complicated than inclusive DIS in NLO
since here one has to contend with double convolutions of the form q C D and
q C D for the unpolarized and polarized cases, respectively, where C and C
are Wilson coefficients first derived in [17] and [4].
The double convolution is defined as

dx
x
z
dz

q C D =
(50)
q C(x , z )D ,
x
z
x
z
where the range of integration is given as follows:
I1 : The range is
x
x 1
x + (1 x)z
with z z 1.
381
(51)
I2 : In addition to (51) there is the range

x x
x
x + (1 x)z
with
x(1 x )
z 1.
x (1 x)
(52)
Note, that contrary to the case of the usual DIS convolution, the double convolution
q C D is not commutative.
We shall frequently encounter expressions of the form
s
qCD
qD +
(53)
2
corresponding to the LO plus NLO corrections. The flavour structure of the results becomes
much more transparent if we adopt the following symbolic notation:

s
s
q C D = q 1 + C D.
qD +
(54)
2
2
Then the semi-inclusive polarized cross section ph defined in (7) is given by

s
s
h
2
h
2
h
Cqg DG
p =
ei qi 1 + Cqq Dqi +
ei qi
2
2
i
i

s
Cgq
+ G
ei2 Dqhi ,
(55)
2
i
where the sum is over quarks and antiquarks of flavour i and parton densities and
fragmentation functions are to be taken in NLO.
For the unpolarized semi-inclusive cross section in NLO it is not possible to completely
factor out the y-dependence. Consequently h defined in (6) will depend upon y in NLO,
in contrast to the LO situation in (9).
In the notation of the seminal paper of Graudenz [17] the cross section is a sum of
what he refers to as metric (M) and longitudinal (L) terms, with corresponding Wilson
coefficients C M and C L . A further complication is that the Wilson coefficients are different
in the two regions of integration I1 and I2 . Thus we have coefficients C j M , C j L with
j = 1, 2.
We then define the y-dependent combinations of Wilson coefficients:

jL
j
jM
Cqq = Cqq + 1 + 4 (y) Cqq ,
(56)

jL
j
jM
Cqg = Cqg + 1 + 4 (y) Cqg ,
(57)

jL
j
jM
Cgq = Cgq + 1 + 4 (y) Cgq ,
(58)
where
(y) =
1y
.
1 + (1 y)2
(59)
382
Then the unpolarized semi-inclusive cross section can be written in a form analogous to
the polarized one:

s
s
h
2
h
2
h
Cqg DG
p =
ei qi 1 + Cqq Dqi +
ei q i
2
2
i
i

s
Cgq
+G
ei2 Dqhi .
(60)
2
i
In (60) we have used the symbolic notation:

s
s 1
s 2
h
h
Cqq Dqi = qi
Cqq Dqi + qi
C Dqhi ,
qi
2
2
2 qq
I1
(61)
I2
and analogously for Cqg and Cgq .

Note that in NLO the unpolarized inclusive cross section, DIS (12) is given by

DIS = 2F1 1 + 2 (y)R ,
(62)
where R is the usual DIS ratio of longitudinal to transverse cross sections. Note that strictly
speaking here and throughout the rest of this paper R should be replaced by

4M 2 x 2
4M 2 x 2
R R
1+
,
Q2
Q2
since the correction terms may be important for low values of Q2 .
As in the LO discussion we doubt the reliability of a global NLO analysis of inclusive
and semi-inclusive data, and we suggest that one should feed into the semi-inclusive
analysis as much reliable information as one can from other sources.
As we shall see there is the oft found opposition between what is simple theoretically and
what is simple experimentally. However, if the systematic errors in detection efficiencies
can be brought under control then we can make remarkable theoretical simplifications and
we can then extract a vast amount of information from the semi-inclusive data. This is an
important experimental challenge as will be seen from the power of the results given below.
5.1. Simplification of the semi-inclusive NLO results
p
Bearing in mind the NLO result (21) for g1 for polarized DIS and the analogous result
for F1 in unpolarized DIS, we see that in the second term of (60) we may make the
replacement, correct to NLO,

p
(63)
ei2 qi 2F1
i
and similarly, in the analogous equation for p in (55):

p
ei2 qi 2g1 .
(64)
i
p,n
Note that here and throughout the rest of this paper F1

measured structure functions.
p,n
and g1
are the experimentally
383
Further in the reaction e+ (p+ ) + e (p ) h + X in the kinematic region when we can

neglect Z 0 -exchange effects, we have

3
3

h z, cos , Q2 = 1 + cos2 Th z, Q2 + 1 cos2 Lh z, Q2 ,
8
4
where Q2 = (p+ + p )2 .
In NLO QCD one has

s T
s T
h
2
2 h
2 h
T z, Q = 30
ei Dqi 1 +
ei DG
+
C
C ,
2 q
2 G
i
(65)
(66)
where the sum is over quarks and antiquarks, the C T s are Wilson coefficients, and
0 =
4 2
.
3Q2
(67)
Then, correct to the required NLO accuracy, in the third term in (60), and in its polarized
analogue (55), we may make the replacement

ei2 Dqhi =
Th (z, Q2 )
,
30
(68)
where Th is the experimentally measured cross section.

Hence, for the unpolarized and polarized semi-inclusive cross sections, in NLO
accuracy, we have

s
s
p
h
2
h
Cqg DG
p =
ei qi 1 + Cqq Dqhi + 2F1
2
2
i
1
s
Cgq Th ,
G
30
2

s
s
p
h
Cqg DG
ei2 qi 1 + Cqq Dqhi + 2g1
ph =
2
2
+
(69)
1
s
Cgq Th .
G
(70)
30
2
Given that the unpolarized gluon density is reasonably well known, the last term in (69)
can be considered as a known quantity. In the following we take as known quantities the
NLO values for q+ (x, Q2 ), q3 (x, Q2 ), uV (x, Q2 ), dV (x, Q2 ), G(x, Q2 ) and q3 (x, Q2 ).
+
5.2. The polarized valence densities in NLO

Using charge conjugation invariance one obtains, for semi-inclusive pion production
+
Rp
[4uV dV ][1 + (s /2)Cqq ]Du

=
p
18F1 [1 + 2 (y)R p ]
Rn
[4dV uV ][1 + (s /2)Cqq ]Du

=
p
18F1 [1 + 2 (y)R p ]
,
.
(71)
384

+
The only unknown function in these expressions is Du (z, Q2 ), which evolves as a

non-singlet and does not mix with other FFs. A 2 analysis of either or both of Eqs. (71)
+
should thus determine Du in NLO without serious ambiguity.

+
Assuming Du is now known, one can then determine uV and dV in NLO via
the equations
+
Ap
An
=
=
(4uV dV )[1 + (s /2)Cqq ]Du

(4uV dV )[1 + (s /2)Cqq ]Du
(4dV uV )[1 + (s /2)Cqq ]Du

(4dV uV )[1 + (s /2)Cqq ]Du
(72)
(73)
where, of course, uV and dV evolve as non-singlets and do not mix with other densities.
Eqs. (72) and (73) determine the densities uV and dV in NLO without any assumptions
about the less known polarized gluon and sea densities.
5.3. SU(2) symmetry of the sea-quark densities in NLO
Once uV and dV are known in NLO we can calculate

u x, Q2 d x, Q2 NLO

1
= q3 x, Q2 + dV x, Q2 uV x, Q2 NLO ,
2
(74)
Eq. (74) determines the breaking of SU(2) symmetry for the polarized sea densities in
NLO without requiring any knowledge of q and G. It will be interesting to compare
the values obtained from (74) with information on u(x)
and d(x)
which will emerge
from DrellYan and W production experiments at RHIC [18].

The separate determination of u and d requires knowledge of q+ , defined in (26),
in NLO.
uV ] is unreliable, since the
Note that determining, say, u via u = 12 [(u + u)
determination of (u + u)
from inclusive DIS involves knowledge of gluon and strange
quark densities.
As in the LO case we can only determine q+ after obtaining some information about
the fragmentation functions.
5.4. Fragmentation functions in NLO
We consider the sum for the unpolarized production of + and . We have

+
s
s
+ +
q3 1 + 2
Cqq Du + + 2
Cqg DG
p

=
,
6 F1 (1 + 2 (y)R p ) F1n (1 + 2 (y)R n )

+
s
s
+ +
Cqq Du + + 2
Cqg DG
q3 1 + 2
+ +
.
Apn =
+

s
s
+ +
q3 1 + 2
Cqq Du + + 2
Cqg DG
+ +
Rpn
(75)
(76)

+
385
+
The only unknown functions in these relations are Du + and DG
, which will mix
with each other under evolution. Thus it should be possible to obtain them from a 2
analysis of (75) and (76).
+
Once we know Du + and utilise Du from Section 5.2 we clearly have access
+
to both Du and Du .
K , where K =
Note that if K 0 can be detected one can obtain information on DuK and DG
+
0
0
+

K + K + K + K . One simply replaces the labels + by K everywhere in (75)
if is detected.
and (76). Analogous equations hold also if + + is replaced by + ,
+ +
allows the determination of the
Returning to the case of + + , the ratio Rp+n
+
only unknown function Ds + . We have

s
s
+
+ +
5q+ 1 + Cqq Du + + 2(s + s) 1 + Cqq Ds +
Rp+n =
2
2

p
s
6
s
+ +
+ +
n
+ 18 F1 + F1
Cgq DG
Cgq T
+ G
2
0
2

p
6 F1 1 + 2 (y)R p + F1n 1 + 2 (y)R n .
(77)
+
+
Under evolution Du + and Ds + mix with DG
, but this is not a problem since
the latter is supposed to be known. It would be interesting to compare these results with
those from e+ e hadrons, obtained recently in [6].
Again analogous sets of equations holds for kaon and , production. One simply
replaces + + by K or + and this allows the determination of the only unknown
function DsK or Ds+ , respectively.
5.5. The sea-quark densities in NLO

+
+
we are now in a position
With the NLO knowledge of Du + , Ds + and DG
2
2
2
to try to determine q+ (x, Q ), s(x, Q ) + s (x, Q ) and G(x, Q2 ). We have, in
NLO,

1
s (Q2 )
p
5q+ + 2(s + s ) 1 +
Cq
g1 + g1n =
18
2
2
s (Q )
G CG ,
+
(78)
2
and
+
Ap+n+ =
p
+ +
+ n
+ +
+ n
where
p
+ +
+ n
+ +
+ +
+ +

s
+
= 5q+ 1 + Cqq Du +
2

s
+
+ 2(s + s ) 1 + Cqq Ds +
2
(79)
386
p
s (Q2 )
+ +
Cqg DG
+ 18 g1 + g1n
2
+
6
s (Q2 )
Cgq T +
+ G
0
2
(80)
and
p
+ +
+ n
+ +

s
+
= 5q+ 1 + Cqq Du +
2

s
+ +
+ 2(s + s ) 1 + Cqq Ds
2
s (Q2 )
p
+ +
Cqg DG
+ 18 F1 + F1n
2
2
+
6
s (Q )
Cgq T + .
+ G
0
2
(81)
Note that an analogous set of equations hold for + + replaced by K or + .

Eqs. (78) to (81) contain three unknown functions q+ , s + s and G, which
nonetheless can all be determined in principle because of their different evolution in Q2 .
However, to be at all efficacious such a determination would require a huge range of Q2 ,
far larger than is available in present day polarised DIS.
On the other hand there is a direct and superior method for obtaining G, namely via
cc production. This is one of the major goals of the COMPASS experiment at CERN. We
shall thus assume that G has been determined, so that the last term on the RHS of (70)
may be taken to be known.
It should be then straightforward to determine q+ and s + s in NLO from a 2
analysis of (78) to (81) in which the evolution of q+ and s + s would involve mixing
with the supposed known G.
An independent determination of q+ and s + s could be obtained by combining
+
(78) with AK

p+n for kaons and Ap+n for , production.
Once q+ is known in NLO, we can obtain the individual u and d from (74).
5.6. Concerning s = s and s = s in NLO
It is possible to get some information on s s and s s in NLO.
+
We have, assuming DdK = DdK ,

+
RpK K
RnK
+ K

+

+
s
s
4uV 1 + 2
Cqq DuK K + (s s ) 1 + 2
Cqq DsK K
p
18F1 [1 + 2 (y)R p ]
4dV
(82)

+
s
s
1 + 2
Cqq DuK K + (s s ) 1 + 2
Cqq DsK K
.
18F1n [1 + 2 (y)R n ]
(83)
These two equations, taken together with those for AK

p
+ K
and AK
n
387
+ K
+ K
AK
p
AK
n
+ K

s
+
= 4uV 1 + Cqq DuK K

2

K + K
s
+ (s s ) 1 + Cqq Ds
2

s
+
4uV 1 + Cqq DuK K

2

s
K + K
+ (s s ) 1 + Cqq Ds
,
2

s
+
= 4dV 1 + Cqq DuK K

2

s
+
+ (s s ) 1 + Cqq DsK K

2

s
+
4dV 1 + Cqq DuK K

2

s
K + K
+ (s s ) 1 + Cqq Ds
2
(84)
(85)
+
provide four equations for the three unknown functions (s s ) DsK K , (s s )
+
DsK K and DuK K , so that, in principle, all can be determined via a 2 analysis.
In addition one has

s
s
(4uV + dV ) 1 + 2
Cqq Du + (s s ) 1 + 2
Cqq Ds
Rp
=
p
18F1 [1 + 2 (y)R p ]
(86)
and
Rn
s
s
(4dV + uV ) 1 + 2
Cqq Du + (s s ) 1 + 2
Cqq Ds
=
18F1n [1 + 2 (y)R n ]
(87)
and A :
together with A
p
n
A
p

s
= (4uV + dV ) 1 + Cqq Du

2

s
+ (s s ) 1 + Cqq Ds
2
388

s
(4uV + dV ) 1 + Cqq Du
2

s
+ (s s ) 1 + Cqq Ds ,
2

s
C
=
(4d
+
d
)
1
+
Du
A
V
V
qq
n
2

s
+ (s s ) 1 + Cqq Ds
2

s
(4dV + dV ) 1 + Cqq Du
2

s
+ (s s ) 1 + Cqq Ds .
2
(88)
(89)
These provide four more equations but only two new unknown functions Du and
Ds . The system of eight equations (82)(89) therefore over-constrains the unknown

functions and might, hopefully, allow a reasonable determination of the relation between
s s and s s and whether or not s(x) = s(x) and/or s(x) = s (x). The actual
+
determination of s s or s s requires knowledge of either Ds or DsK K .

These could be taken from the study of e+ e hX.
This completes the determination of all the polarized densities in NLO.
6. Conclusions
We have argued that the present LO QCD method of analysing polarized semi-inclusive
DIS, using the concept of purity, is quite unjustified. We have also argued against attempts
at a global analysis, either in LO or in NLO QCD, based on the combined data on polarized
inclusive and semi-inclusive DIS and taking as known exactly the relevant FFs.
Instead, we have presented a strategy for a step by step evaluation of the polarized parton
densities and fragmentation functions from semi-inclusive data using selectively chosen
information from inclusive DIS reactions.
In our approach the usually made simplifying assumptions about relations between u
and between the strange and non-strange polarized sea densities are unnecessary
and d,
and we have even considered the possibility that s(x)
= s (x) and s(x)
= s (x).
Given the simplicity of the LO QCD analysis, we discuss where and when it is likely to
be reliable, and stress the need to test the consistency of the LO treatment at each step. In
this connection we have introduced the concept of a passive variable in the experimentally
measured observables. We have also suggested how one might estimate the errors induced
in doing the LO analysis.
389
In the NLO treatment we have shown how the expressions for the experimental
observables can be much simplified by incorporating information from the reaction
e+ e hX in a novel way. Determination of the polarized valence quark densities uV
However
and dV is shown to be relatively straight forward, as is the difference u d.
s + s and G separately, from semiwe argue that the determination of u,
d,
inclusive DIS involving production of , K, is unlikely to be successful, because of the
limited range of Q2 available now and in the foreseeable future. It is suggested that the
independent determination of G from charm production is thus an essential element if
s + s are to be determined accurately. Finally, motivated by the arguments
u,
d,
that possibly s(x)
= s (x) and s(x)
= s (x), we have demonstrated how, in principle,
one can learn about s(x) s(x) and s(x) s (x).
The procedure we have advocated poses a real challenge to the experimentalists, since it
requires a control over the systematic errors involved in hadron detection efficiencies. The
price paid in the current practice of considering certain ratios of cross sections in order
to limit systematic errors, is enormous, and vast amounts of interesting and theoretically
valuable information are thereby lost. We hope this paper will encourage efforts to proceed
further.
Acknowledgements
We thank M. Anselmino, H. Blok, M. Boglione, S.J. Brodsky, U. DAlesio, D. de
Florian, A. Kotzinjan, S. Kretzer, R. Sassot, X. Song, D. Stamenov and C. Weiss for
helpful discussions. We are grateful to the UK Royal Society for a Collaborative Grant.
E.C. is grateful to the Theory Division of CERN for its hospitality where this work was
partly done and finished, her work was also supported by the Bulgarian National Science
Foundation. E.L. is grateful to P.J. Mulders for the hospitality of the Department of Physics
and Astronomy, the Vrije Universiteit, Amsterdam, where part of this work was carried out
supported by the Foundation for Fundamental Research on Matter (FOM) and the Dutch
Organisation for Scientific Research (NWO).
References
[1] G. Altarelli, R.D. Ball, S. Forte, G. Ridolfi, Nucl. Phys. B 496 (1997) 337;
G. Altarelli, R.D. Ball, S. Forte, G. Ridolfi, Acta Phys. Polon. B 29 (1998) 1145;
E. Leader, A. Sidorov, D. Stamenov, Phys. Lett. B 462 (1999) 189 and references therein;
C. Bourrely, F. Buccella, O. Pisanti, P. Santorelli, J. Soffer, Prog. Theor. Phys. 99 (1998) 1017.
[2] B. Adeva et al. (SMC), Phys. Lett. B 369 (1996) 93;
B. Adeva et al., Phys. Lett. B 420 (1998) 180;
B. Adeva et al., Phys. Lett. B 435 (1998) 420.
[3] K. Ackerstaff et al., HERMES Collaboration, hep-ex/9906035.
[4] D. de Florian, C.A. Garcia Canal, R. Sassot, Nucl. Phys. B 470 (1996) 195.
[5] J. Binnewies, B.A. Kniehl, G. Kramer, Phys. Rev. D 52 (1995) 4947.
[6] S. Kretzer, Phys. Rev. D 62 (2000) 054001.
390
[7] D. de Florian, L.N. Epele, H. Fanchiotti, C.A. Garcia Canal, S. Joffily, R. Sassot, Phys.
Lett. B 389 (1996) 358;
D. de Florian, O.A. Sampayo, R. Sassot, Phys. Rev. D 57 (1998) 5803.
[8] D. de Florian, R. Sassot, hep-ph/0007068.
[9] A.E. Dorohov, N. Kochelov, Phys. Lett. B 304 (1993) 167;
D.I. Diakonov, V.Yu. Petrov, P.V. Pobylitsa, M.V. Polyakov, C. Weiss, Nucl. Phys. B 480 (1996)
341;
D.I. Diakonov, V.Yu. Petrov, P.V. Pobylitsa, M.V. Polyakov, C. Weiss, Phys. Rev. D 56 (1997)
4069;
A. Efremov, K. Goeke, P.V. Pobilitsa, hep-ph/0004196.
[10] B. Dressler, K. Goeke, M.V. Polyakov, C. Weiss, Eur. Phys. J. C 14 (2000) 147.
[11] X. Song, Phys. Rev. D 57 (1998) 4114;
B.-Q. Ma, S.J. Brodsky, Phys. Lett. B 381 (1996) 317.
[12] M. Glck, E. Reya, Z. Phys. C 39 (1988) 569;
A.D. Watson, Z. Phys. C 12 (1982) 123.
[13] L. Trentadue, G. Veneziano, Phys. Lett. B 323 (1993) 201.
[14] E. Leader, A. Sidorov, D. Stamenov, Phys. Rev. D 58 (1998) 114028.
[15] L. Frankfurt et al., Phys. Lett. B 230 (1989) 141.
[16] E. Christova, E. Leader, Phys. Lett. B 468 (1999) 299.
[17] D. Graudenz, Nucl. Phys. B 432 (1994) 351.
[18] C. Bourelly, J. Soffer, Nucl. Phys. B 423 (1994) 329.

Curved BPS domain wall solutions in

four-dimensional N = 2 supergravity
Klaus Behrndt, Gabriel Lopes Cardoso, Dieter Lst
Institut fr Physik, Humboldt University, Invalidenstrae 110, 10115 Berlin, Germany
Received 12 March 2001; accepted 12 April 2001
Abstract
We construct four-dimensional domain wall solutions of N = 2 gauged supergravity coupled to
vector and to hypermultiplets. The gauged supergravity theories that we consider are obtained by
performing two types of Abelian gauging. In both cases we find that the behaviour of the scalar
fields belonging to the vector multiplets is governed by the so-called attractor equations known
from the study of BPS black hole solutions in ungauged N = 2 supergravity theories. The scalar
fields belonging to the hypermultiplets, on the other hand, are either constant or exhibit a run-away
behaviour. These domain wall solutions preserve 1/2 of supersymmetry and they are, in general,
curved. We briefly comment on the amount of supersymmetry preserved by domain wall solutions in
gauged supergravity theories obtained by more general gaugings. 2001 Elsevier Science B.V. All
rights reserved.
PACS: 11.27.+d; 11.30.Pb; 04.65.+e
Keywords: BPS domain walls; Supergravity
1. Introduction
Recently there has been a lot of interest in constructing domain wall solutions in
gauged supergravity. As consequence of the AdS/CFT correspondence [13], domain wall
solutions can provide the supergravity duals of field theories near a fixed point and may,
in addition, give a holographic picture of the renormalization group (RG) flow. In order to
understand this in more detail it is important to have explicit solutions on the supergravity
side.
The BPS domain wall solutions of five-dimensional gauged supergravity theories
describe the RG flow between superconformal field theories in four dimensions (for an
example see [4]). Here, the gravitational AdS5 backgrounds arise as the near horizon
E-mail addresses: behrndt@physik.hu-berlin.de (K. Behrndt), gcardoso@physik.hu-berlin.de
(G.L. Cardoso), luest@physik.hu-berlin.de (D. Lst).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 9 3 - 6
392
K. Behrndt et al. / Nuclear Physics B 607 (2001) 391405
geometry of type IIB D3-branes located at certain transversal six-dimensional spaces.

In the same way, four-dimensional AdS4 supergravities emerge, for example, from Mtheory membranes located at certain transversal eight-dimensional spaces. These are
then holographically related to three-dimensional conformal field theories [5], which are
however not very well understood at the present time. In this paper we will focus on BPS
domain wall solutions of gauged supergravity theories with eight supercharges (gauged
N = 2 supergravity) in four dimensions. Domain wall solutions in theories with less
supersymmetries were previously considered in [6,7].
The four-dimensional N = 2 gauged supergravity theories we will consider are obtained
by performing a gauging either of a U (1) subgroup of the SU(2) R-symmetry or of a
particular Abelian Killing symmetry of the universal hypermultiplet moduli space. The
former results in models with a potential term for the vector scalars only, whereas the
latter results in a potential term for the vector scalars and for the dilaton field entering the
universal hypermultiplet. In the context of string theory, the latter models can be obtained
by CalabiYau threefold compactifications of type II string theory in the presence of
internal H-fluxes 1 [1115] The vacuum structure of these type II CalabiYau threefold
compactifications was discussed in some detail in [1115]. It was shown that N = 2
supersymmetric ground states with flat four-dimensional Minkowski spacetime are only
possible in certain degeneration limits of the underlying CalabiYau spaces (otherwise
N = 2 supersymmetry gets completely broken at generic points in the CalabiYau
moduli space). Thus, in general, non-vanishing H-fluxes do allow for supersymmetric
anti-de Sitter groundstates with non-vanishing cosmological constant 4 . This implies
that the associated four-dimensional field equations possess solutions describing non-flat
gravitational backgrounds, such as domain walls.
In this paper we will construct four-dimensional domain wall solutions of the gauged
supergravity theories described above. We will show that (as already indicated in [15,16])
the behaviour of the scalar fields belonging to the vector multiplets is governed by a set
of equations already known from the study of BPS black hole solutions in ungauged N =
2 supergravity theories, namely the so-called attractor equations [1726]. We will also
show that in general, the four-dimensional BPS equations are solved by curved walls, i.e.,
there is in general also a three-dimensional cosmological constant 3 on the domain wall.
In fact, the curvature of the wall must cancel contributions coming from an expression
closely related to the U (1)-Khler connection (as we will see, this effect already occurs
in the context of pure N = 2 supergravity). This is in contrast with what happens in five
dimensions in analogous models. Since such a connection is not present in five dimensions
(and whenever the pullback of the SU(2) connection is trivial), the five-dimensional BPS
equations only allow for flat domain wall solutions [27,28].
We can then give the following dictionary between the quantities which play a role
in black hole solutions of ungauged N = 2 supergravity and those of the corresponding
domain wall solutions of gauged N = 2 supergravity:
1 It is conceivable [8,9] that string CalabiYau compactifications with internal H-fluxes in the field theory limit
are T-dual to D3-branes in the presence of transversal RR-fluxes [10].
393
In gauged N = 2 supergravity we will use the following symplectic invariant

superpotential:

W = es I LI I MI , s = 0, 2,
(1.1)
where the symplectic vector (LI , MI ) (I = 0, . . . , NV ) denotes the sections which

depend on the vector scalar fields, and is the dilaton field. In the case of the gauging
of a particular Abelian isometry of the universal hypermultiplet moduli space, the
entries of the constant symplectic vector (I , I ) correspond to the electric/magnetic
U (1)NV +1 charges of the universal hypermultiplet and emerge in type IIA/B string
compactifications as internal H-fluxes on CalabiYau threefolds with h1,1 (h2,1 ) =
NV for type IIA (type IIB) [12,13]. The dependence of this superpotential on the
vector scalars is, on the other hand, the direct analogue of the central charge Z of
the N = 2 supersymmetry algebra that plays a role in the context of BPS black holes
in ungauged N = 2 supergravity. Here, the (I , I ) are just the electric/magnetic
U (1)NV +1 charges of the black holes.
The attractor equations (see Eq. (3.29)) determine the running of the vector scalar
fields from the domain wall to the supersymmetric extrema, DW = 0, reached at
spatial infinity. In fact, the vector scalar fields are stabilized in the sense that their
values at spatial infinity only depend on the constants (I , I ). Therefore the attractor
equations are relevant for the RG-flow in the corresponding field theories. In the
context of N = 2 black holes the same stabilization equations determine the evolution
of the vector scalars from spatial infinity towards the horizon, where the central charge
is extremized, DZ = 0, and the scalars are entirely expressed in terms of the (I , I ).
For N = 2 domain walls, the extremum of the potential with respect to the vector
scalar fields, Vextr |W |2 , is again a function only of the (I , I ) and corresponds
to a four-dimensional cosmological constant 4 at spatial infinity. The analogous
quantity, |Z|2extr , is just the entropy S of the N = 2 black holes, i.e., the area of the
horizon.
We will show that the quantity (see Section 3)

1
I d FI F
I ,
AY = e2U Y I Y
(1.2)
2
provides the three-dimensional cosmological constant 3 on the domain wall. This
means that a non-vanishing one-form AY leads to curved anti-de Sitter like domain
wall solutions. On the black hole side, the same object corresponds to the angular
momentum of stationary but in general non-static solutions [23,25,26].
Finally, closely related brane configurations play a role in the comparison of N =
2 domain wall and black hole solutions. In type IIA superstring compactified on
a CalabiYau threefold M, D4-branes, being A -times wrapped around 4-cycles
of M, together with 0 D0-branes, lead to black holes with magnetic charges pA
and one electric charge 0 . The corresponding entropy is determined by the triple
intersection form CABC . Increasing the dimensionality of the branes by two, i.e.,
considering boundstates of wrapped D6-branes together with 0 D2-branes, leads
to membranes in four dimensions, which represent the source for the supergravity
394
domain wall solutions in four uncompactified dimensions. Similarly, in type IIB

theory compactified on a CalabiYau threefold, D3-branes wrapped around 3-cycles
provide charged black hole solutions, whereas the same configurations of wrapped
D5-branes correspond to domain wall solutions.
The paper is organized as follows. In the next section we will recall certain aspects of
gauged N = 2 supergravity in four dimensions. In Section 3 we will solve the BPS
equations for domain wall solutions. Section 4 contains a discussion of our solutions as
well as various concluding remarks.
2. N = 2 gauged supergravity
Domain walls are codimension one solutions that separate the spacetime into regions
corresponding to different vacua. In the simplest case, a domain wall is supported by a
gauge potential that couples to its world volume. The field strength of this gauge potential
is dual to a cosmological constant. In a more general setting with non-trivial couplings to
scalar fields, this cosmological constant appears as an extremum of the potential term in
the Lagrangian. The resulting solution then describes a flow towards an extremum, and if
the potential possesses several extrema, the solution may interpolate between them. In the
following we will be interested in constructing domain wall solutions in N = 2 gauged
supergravity theories in four dimensions.
The N = 2 gauged supergravity theories that we consider are based on Abelian vector
multiplets (labelled by an index I = 0, . . . , NV ) and hypermultiplets coupled to the N = 2
supergravity fields. Potentials that are allowed by N = 2 supersymmetry are then obtained
by performing a gauging of (some of) the various global symmetries. There are two
different types of gaugings, namely (i) one can either gauge some of isometries of the
moduli space of ungauged N = 2 supergravity or (ii) one can gauge (part of) the SU(2)
R-symmetry, which only acts on the fermions. In N = 2 supergravity, at the two-derivative
level, the moduli space is a direct product M = MV MH , where MV and MH are
parameterized by the scalars belonging to the vector multiplets and to the hypermultiplets,
respectively. In the following, we will only consider Abelian gaugings, either of (some of)
the isometries of the hyper scalar manifold MH or of the SU(2) R-symmetry of N = 2
supergravity. We refer to [2931] for a detailed description of the gauging.
The hyper scalar manifold MH is a quaternionic space, and hence it possesses three
complex structures J x as well as a triplet of Khler two-forms K x (here x = 1, 2, 3, denotes
an SU(2) index). The holonomy group is SU(2) Sp(nH ) and the Khler forms have to
be covariantly constant with respect to the SU(2) connection. The isometries of MH are
generated by a set of Killing vectors kIu ,
q u q u + kIu $ I ,
(2.1)
and therefore the gauging of (some of) the Abelian isometries results in the introduction of
gauge covariant derivatives via the replacement dq u dq u + kIu AI . In order to maintain
supersymmetry, the gauging of the isometries has to preserve the quaternionic structure,
395
which implies that the Killing vectors have to be tri-holomorphic. This is the case whenever
it is possible to express the Killing vectors in terms of a triplet of real Killing prepotentials
PIx , as follows:
y
x v
Kuv
kI = u PIx u PIx $ xyz u PIz .
(2.2)
x =
Here u are the SU(2) connections, which are related to the Khler forms by Kuv
x
x
[u v] . By using the Pauli matrices one can also revert to matrix notation and write
(PI )ij =
3

PIx x i k j k ,
(2.3)
x=1
where j k denotes the two-dimensional antisymmetric -symbol.

In this paper we will, for concreteness, only consider the quaternionic space associated
with the so-called universal hypermultiplet. Classically, it is given by the coset space
SU(2, 1)/U (2). Its Khler potential is, in a certain parameterization, given by

2 .
K = log S +
S 2 C+C
(2.4)
This coset space has two Abelian isometries which are generated by the Killing vectors
associated to shifts in the imaginary parts of S and C. The gauging of these two Abelian
isometries has been discussed in [12] (we refer to [27,3234] for a discussion of the
gauging of (some of) the other isometries of the universal hypermultiplet). In this paper,
however, we will only consider the case when the shift in the imaginary part of S is gauged.
The associated Killing vector is given by [12]
kI = kIu
= I i(S
S ),
q u
(2.5)
where the I are constant and real. This is the four-dimensional analogue of the gauging of
five-dimensional supergravity discussed in [27]. The case where one independently gauges
both the Abelian isometries results in a potential which makes the construction of explicit
domain wall solutions rather difficult, and will not be discussed here.
The Killing prepotential associated with the Killing vector (2.5) is, in matrix notation
(2.3), given by [12]

0 1
ij1 =
,
PI ij = e2 I ij1 ,
(2.6)
1 0 ij
2.
where the dilaton is given by e2 = S +
S 2(C + C)
It can be shown [29] that the gauging of a U (1) subgroup of the SU(2) R-symmetry
results in a Killing prepotential of the form PI = I 1 , where the I are again constant
and real. Thus, the Abelian gaugings we will consider in this paper are characterised by a
Killing prepotential of the form
PI ij = es I ij1 ,
s = 0, 2.
(2.7)
The class of spacetime metrics that we will consider in the following is given by
ds 2 = e2U (z)gmn dx m dx n + e2pU (z) dz2 ,
(2.8)
396
with gmn = g mn (x m ). Here we denote spacetime indices by x = (x m , z), x m = (t, x, y),

and the corresponding tangent space indices by a = (0, 1, 2, 3). The constant p is
introduced for later convenience. We assume Lorentz invariance in the three-dimensional
subspace a = (0, 1, 2). The metric gmn is thus a three-dimensional constant curvature
metric. We take the solutions to be uncharged, that is we set the gauge fields to zero.
In general, however, as a consequence of the gauging of (some of) the isometries of
MH , some of the hyper scalar fields are charged and therefore they act as sources for
the gauge fields. The associated currents are given by huv kIu (dq v + kJv AJ ). Then, in order
for a vanishing gauge field to be a consistent solution of its equation of motion, either the
associated current has to vanish or the charged hyper scalars have to satisfy the equation
huv kIu dq v = 0. This implies that the hyper scalar fields that are not constant have to be
neutral. This will indeed turn out to be the case for the gauging (2.5), where the only
non-trivial hyper scalar field is the neutral dilaton field .
In the absence of gauge fields, the supersymmetry transformation laws for the gravitini
i , the gaugini A
i (A = 1, . . . , NV ) and the hyperini read [29,35]
1
i = D $i PI ij L I $ j ,
2
I j
A
A
i = z $i + PI ij g AB D
L $ ,
i
= Vu
q u ij $ j
B
i u I
2Vu kI L $i .
(2.9)
Here we have denoted the scalar fields residing in the Abelian vector multiplets by LI .
For later convenience we now also introduce a non-holomorphic section which we write
I ) = 1. The
as V T = (LI , MI ). It is subject to the symplectic constraint i(L I MI LI M
assignment of chiral weights c is as follows. The non-holomorphic section V has c = 1,
whereas the supersymmetry parameters $i and $ i have c = 12 and c = 12 , respectively.
Inspection of (2.9) shows that the following quantity will play a central role when
solving the equations resulting from the vanishing of the supersymmetry transformation
laws (2.9):
W = es I LI ,
s = 0, 2.
(2.10)
Supersymmetric domain wall solutions will, in general, only preserve part of N = 2

supersymmetry. A condition on the supersymmetry parameters that is consistent with local
Lorentz invariance in the subspace a = (0, 1, 2) (and, hence, with (2.8)) is the following,
$i = Aij 3 $ j ,
(2.11)
where Aij denotes an SU(2) matrix. Since the spacetime metric (2.8) only has a non-trivial
dependence on the coordinate z, we have Aij = Aij (z). Taking the complex conjugate of
(2.11) and recalling that ($i )< = $ i (as well as using a convention where 3 is real) then
yields that (A A)ij $ j = $ i . Let us now assume that (2.11) is the only restriction on the
supersymmetry parameters, so that (A A)ij = ji . On the other hand, inserting (2.11) into
the gravitino variation mi and demanding its vanishing yields Aij PI ij L I . Now, since
for general gaugings the condition (A A)ij = ji is in general not satisfied, we conclude
397
that for general gaugings (2.11) cannot be the only restriction on the supersymmetry
parameters, but that there are further constraints leading to an additional (or possibly
complete) breaking of supersymmetry. For the Abelian gaugings (2.7) considered in this
paper, however, (A A)ij = ji is satisfied, so that (2.11) is the only condition on the
supersymmetry parameters, thus leading to domain wall solutions which preserve 1/2 of
N = 2 supersymmetry.
3. Solving the BPS equations
In this section we will construct supersymmetric domain wall solutions (2.8) for the
Abelian gaugings specified by (2.5) and (2.7).
The symplectic extension of (2.10) is given by

W = es I LI I MI , s = 0, 2.
(3.1)
Here the I denote real constants associated with the dual Killing vector k I and the
dual Killing prepotential PijI . The quantity (3.1) is symplectically invariant, provided that
(I , I ) transforms as a vector under symplectic transformations. In the context of string
theory, a constant vector (I , I ) does indeed arise when turning on H-fluxes on Calabi
Yau threefolds [11,12]. Moreover, in the presence of these fluxes, it is then possible to
engineer supersymmetric domain wall solutions by wrapping D-branes around appropriate
supersymmetric cycles. We take this to be an indication that, in general, supersymmetric
domain wall solutions do depend on both I and I .
We will therefore consider the following symplectic extension of the supersymmetry
transformation rules (2.9) (which are valid in the absence of gauge fields),
1 1
i = D $i W
ij $ j ,
2
1 j
A
AB
DBW ij $ ,
A
i = z $i + g

i
u
j
= Vu q ij $ 2 kIu LI k I u MI $i ,
(3.2)
with W given by (3.1). The BPS solutions we will construct in this section are obtained by
demanding the vanishing of the supersymmetry variations (3.2), subject to a condition of
the form (2.11). It should be pointed out, however, that we are not aware of any construction
of gauged supergravity theories giving rise to supersymmetry transformation rules of the
type (3.2) involving not only PI and kIu but also their duals. Our justification for starting
with (3.1) and (3.2) is twofold. On the one hand we will be able to solve the system
of equations resulting from the vanishing of (3.2). On the other hand, as stated above,
there are examples of supersymmetric domain wall solutions in string theory which are
characterised by both the parameters I and I appearing in (3.1), and we would like to
able to reproduce them.
We now impose the following condition on the supersymmetry parameters, in accordance with the discussion given below (2.11),
ij1 3 $ j ,
$i = h
(3.3)
398
where h = h(z) denotes a phase factor with chiral weight c = 1.

Let us first consider the variation mi = 0. For a spacetime metric of the form (2.8),
we find that

b
m ab =
(3.4)
m ab + e(p+1)U z U 3a m
(a b) ,
where
m ab denotes the spin connection associated to the three-dimensional metric g mn =
a
b
em em ab , where a = 0, 1, 2. The metric gmn is a constant curvature metric. In the anti-de
Sitter case, we may write the constant curvature condition as

mn ab = 4 em a en b (m n) ,
R
(3.5)
2
l
where l denotes a real constant, which is related to the three-dimensional cosmological
= 243 . The condition
constant 3 by 3 = l 2 . The associated curvature scalar is then R
(3.5) can be viewed [36] as the integrability condition associated with

m h1/2 $i = i em a a 3 h1/2 $i .
D
(3.6)
l
The case of zero curvature is obtained from the above by sending l , whereas the
de Sitter case is obtained by replacing l il. The latter does not lead to supersymmetric
domain wall solutions. We also note that the imposition of (3.6) does not lead to a further
restriction of the residual supersymmetry preserved by (2.8).
Using (3.3) as well as (3.6) we then obtain from mi = 0,
2i U
e .
l
Next, let us consider the variation zi = 0. It yields
= epU z U +
hW
1
ij1 3 h$
j.
Dz $i = epU hW
2
Inserting (3.3) on the r.h.s. of (3.8) yields
1
$i .
Dz $i = epU hW
2
Hermitian conjugation then gives
1
$i .
Dz $ i = epU hW
2
On the other hand, inserting (3.3) on the l.h.s. of (3.8) yields
(3.7)
(3.8)
(3.9)
(3.10)

1
$ i hDz h $ i .
Dz $ i = epU hW
(3.11)
2
Comparison of (3.10) with (3.11) then gives

2i (1+p)U
1
hW
= e
.
hDz h = epU hW
(3.12)
2
l
Also, using zab = 0 and that the pullback of the SU(2) connection is trivial (as we will see
later), we find that (3.9) is solved by
1
h1/2 $i = i e 2 U ,
(3.13)
399
where i denotes a spinor satisfying (3.6).

Next, let us consider the vanishing of the gaugini variation, A
i = 0. Using the special
geometry relations [29]
L J ,
gAB = 2DA LI Im NI J D
B
I J DA LJ ,
DA MI = N
V = 0,
D
A
(3.14)
as well as (3.3) we obtain from A

i = 0,

W
M

L I = epU hD

gABz zA = i Dz LI D
B I Dz MI DB
B ,
(3.15)
and, hence,

spU I
M

z LI + i I espU D
DBL .
hD
B I = hDz MI + iI e
(3.16)
(L I MI LI M
I ) = 0 we may rewrite this as
Using D
B

I
spU I
M

+ i I espU D
DBL ,
z hL
B I = z hMI + iI e
(3.17)
= 0, where denotes the sympletic metric and where and

or equivalently as T U
B

UB denote two symplectic vectors, as follows,
I
I

z hL + i I espU
DBL
0 1

=

,
UB =
,
=
.
I + iI espU
M

D
z hM
1 0
B I
(3.18)
We note that has chiral weight c = 0.
= 0 constructed out of the symplectic
The general solution to the equation T U
B

vectors V , V , UA and UA is given by

+ DhV
+ Eh z z A U
,
= C hV
A
(3.19)
= 0 as well as V
T U
= 0,
with parameters C, D and E. This is so because V T U
B
B
T
= ig = 0. We now determine C and D by considering the symplectic
whereas UA U
B
AB
T . We then find that
and hV
, i.e., hV
T and hV
product of (3.19) with hV
C = z U,
D = z U hDz h
(3.20)
by virtue of (3.7) and (3.12).

A , i.e.,
The parameter E, on the other hand, can be determined by contracting with hU
T
T

from hUA . Using UA UB = igAB as well as (3.15) then yields
E = 1.
Thus, it follows from (3.19) that
I
I

I ) + z U hL
I
z ( hL
L
DB
B
spU
= ie
+ hz z

I
DBM
z ( hMI ) + z U hMI
I
I
I
hL
hL
+ z U
hDz h
I .

hM
hMI
(3.21)
(3.22)
400
This we rewrite as

I
I
z Y I
L
DB
B
s+(1p)U
U
= ie
+ e hz z

I
DBM
z FI
I
I
I
Y
Y
hDz h
,
+ z U
I
I
F
F
(3.23)
where we introduced the chiral invariant variables

I,
Y I = eU hL
I.
FI (Y ) = eU hM
I ) = 1 yields
Inserting (3.24) into i(L I MI LI M

I
.
I Y
FI (Y ) Y I F
e2U = i Y
Using
I
I
DBL
L
U
= e hDz
e hz z

I
DBMI
M
I
I
I
Y
Y
Y
zh
= z
z U
hD
,
I
I
I
F
F
F
U
(3.24)
(3.25)

B
it follows from (3.23) that

I I
I
Y Y
z
= ies+(1p)U
.
I
FI F
I
(3.26)
(3.27)
If we now set
es = e(p1)U ,
then we may integrate (3.27),

I
I I
I
H
Y Y
h I z
=i
=i
,
I
FI F
HI
hI I z
(3.28)
(3.29)
where the (H I , HI ) denote harmonic functions. These equations, which determine the
behaviour of the physical vector scalar fields zA = Y A /Y 0 , are also known from the study
of BPS black hole solutions in ungauged N = 2 supergravity theories, where they are
called attractor equations [17,18].
Let us now return to (3.28). For the case s = 0 (corresponding to an Abelian gauging
of the SU(2) R-symmetry), (3.28) holds for the convenient choice p = 1. For the case
s = 2 (corresponding to the Abelian gauging of a particular isometry of the universal
hyper moduli space) we will show below that (3.28) solves the equation resulting from
the vanishing of the hyperino variation, provided that p = 3.
Next, consider inserting (3.29) into (3.12). This yields

1
l 1 = es I hI I hI .
(3.30)
4
Since l has to be constant, we conclude that it is not possible to have a non-vanishing
three-dimensional cosmological constant 3 for the case s = 2. That is, for s = 2 the
401
domain wall solutions have to be flat (I hI I hI = 0), whereas for s = 0 they may be
curved. This implies that in the latter case there is no restriction on the integration constants
(hI , hI ), whereas in the former case there are such restrictions.
Before turning to the hyperini variation = 0, let us perform two consistency checks
on the solution given above. Let us first return to (3.7). Using (3.27) as well as (3.28) the
+ hW
= 2epU z U , may be rewritten as
real part of (3.7), hW

I

I I FI + F
I = iz Y
FI Y I F
I ,
z e2U = I Y I + Y
(3.31)
which is in perfect agreement with (3.25). Next, let us consider hDz h = hz h iAz , where
h), AYz = 1 e2U (Y I Y

I ) z (FI F
I ). Using (3.29) it follows
Az = AYz 2i z log(h/
2
i
Y
2U
I
I
that hDz h = iAz = 2 e
(I h hI ) which, upon using (3.30), precisely yields
(3.12).
Finally , let us turn to the variation = 0. For the case s = 0 we have kIu = 0, k I u = 0,
so that = 0 is solved by constant hyper scalars, q u = const. On the other hand, for

s = 2 we have kISS = 2 I and k I SS = 2 I . We solve the hyperino variation by
setting S
S = const and C = const, so that also in this case the pullback of the SU(2)
S) = e2 , it follows that
connection is trivial. Then, using (S +

i
i
= VS+
(3.32)
LI I MI $i .
e2 ij $ j + 4VS

S,
S, I
Using

0 e2 i
0 e2 i
1
1
i
(3.33)
,
VS
=
,
S
0
0
2 2 e2
2 2 e2
as well as (3.3) and 12 = 1, we obtain

epU z e2 = 4h I LI I MI = 4e2 hW.

(3.34)
1

It follows that hW = hW , and hence we obtain from (3.7) that l = . Using (3.7) it
follows that (3.34) is solved by
i
VS+
=

S
e2(0 ) = e4U .
(3.35)
This is consistent with (3.28) provided that p = 3.

Let us briefly summarize some of the properties of the domain wall solutions we have
constructed. For both cases (s = 0, 2) we find that the behaviour of the vector scalar fields
is determined by a set of attractor equations given in (3.29). For the case s = 0, the hyper
scalars are all constant and the domain wall solutions are, in general, curved, with the
three-dimensional cosmological constant 3 = l 2 determined by (3.30). The associated
spacetime metric is given by (2.8) with p = 1. For the case s = 2, on the other hand, we
find that the dilaton field exhibits the run-away behaviour (3.35). The domain wall solutions
must be flat, and the associated spacetime metric is given by (2.8) with p = 3.
4. Discussion
The domain wall solutions specified by Eqs. (2.8), (3.25), (3.29) and (3.30) solve the
Killing spinor equations for any prepotential function describing the couplings of Abelian
402
vector multiplets to supergravity. In the case that only the R-symmetry is gauged (s = 0),
the hyper scalars are constant and the solutions possess an asymptotic (z ) AdS
vacuum. In the case s = 2 (corresponding to the gauging of an axionic shift symmetry),
on the other hand, the dilaton is not constant but instead given by (3.35). Thus, it is
not stabilized by the potential. In either case, the behaviour of the scalars in the vector
multiplets is determined by the same set of attractor equations (3.29) known from the study
of BPS black holes in ungauged supergravity. Moreover, the domain wall may possess
constant AdS-like curvature according to (3.30). This feature also has a counterpart in
single-center black hole solutions, namely in the angular momentum carried by these
stationary solutions. There is the following one-to-one correspondence between quantities
determining black hole solutions and domain wall solutions with non-trivial vector scalars:
Black hole
Central charge Z
Entropy S
Angular momentum J
Domain wall
Superpotential W
Cosmological constant Vextr
Constant wall curvature 3
This analogy is related to the fact that both types of solutions can be reduced to equivalent
one-dimensional systems [16,25].
The fact that all the regular critical points are reached at the boundary of the AdS space
(e2U ) excludes the possibility of a regular RG-flow and/or of trapping of gravity
near the wall [37,38]. From the RG-flow point of view the AdS vacuum corresponds to
an UV fixed point, and when moving towards negative z the warp factor e2U decreases
monotonically (in accordance with a c-theorem). If we do not add a source, then e2U will
have to pass through a zero at some finite value of z, which describes a singular end-point
of the RG flow. As in any other cases with vector multiplets only, the absence of an IR
fixed point forces the solution to run into a singularity.
On the other hand, in the context of string theory, it is possible to engineer domain
wall solutions with a non-trivial dilaton field in terms of D-branes wrapped around
supersymmetric cycles. It is thus reasonable to put appropriate sources at some place where
the warp factor is still positive, say at z = 0. These source terms will then appear on the
right-hand side of the harmonic equations
2 HI I (z),
2 H I I (z),
(4.1)
where (I , I ) are the charges in a basis of the wrapped cycles. For dilatonic walls that are
flat (which correspond to the case s = 2 considered in this paper), there are two ways in
which one can solve (4.1). The first possibility consists in continuing the solution through
the source in a symmetric way, which implies the replacement HI hI + I |z|, H I
hI + I |z|, and which is equivalent to a sign change in the flux vector (I , I ) while
passing through the source at z = 0. In this case the scalar fields and the warp factor e2U ,
which are determined from (3.29), are Z2 -symmetric. One may, however, also consider
the case where the flux jumps from zero to a finite value, i.e., on the side behind the
source we can set I = 0, I = 0, which is equivalent to performing the replacement
403
HI hI + I 21 (z + |z|), H I hI + I 12 (z + |z|), so that for negative z the HI and H I

are constant and the spacetime is flat. Thus, by adding sources, we cut off the (singular)
part of the spacetime and instead glue on either an identical piece (yielding a Z2 -symmetric
solution) or flat spacetime (corresponding to vanishing flux on one side of the wall).
Let us discuss a few concrete examples of dilatonic domain walls. Consider the case
where the only non-vanishing components of (I , I ) are given by 0 and A (A =
1, 2, 3). In type IIA theory, the corresponding domain wall solution can then be obtained
by a torus compactification of the D-brane intersection 2 6 6 6, where the 6branes wrap a 4-cycle in the internal space and where the common 2-brane represents
the domain wall in the four non-compact directions. The dual configuration with nonvanishing A (A = 1, 2, 3) and 0 is described by the intersection 4 4 4 8, where
the 4-branes wrap a 2-cycle and where the 8-brane wraps the whole internal space. These
solutions can of course also be mapped onto the type IIB side, where they are given by
the brane intersections 5 5 5 5 and 3 5 5 7, respectively. This picture in
terms of brane intersections is the one appropriate for torus compactifications. For generic
CalabiYau threefold compactifications, a domain wall solution with non-vanishing 0
and A is obtained on the type IIA side by wrapping a single 2- and a single 6-brane
around a holomorphic 0- and a holomorphic 4-cycle of the CalabiYau manifold. Similarly,
on the type IIB side, a domain wall solution is obtained by wrapping a single 5-brane
around a supersymmetric 3-cycle. In either case the symplectic flux vector (I , I ) is the
decomposition of the corresponding brane charges in a basis of 2-, 3- or 4-forms.
Let us describe the type IIA solution with non-vanishing 0 and A in some more
detail. In the limit where the volume of the CalabiYau threefold is taken to be large,
the associated homogeneous function F (Y ) is given by
1
DABC Y A Y B Y C
,
DABC = CABC ,
(4.2)
0
6
Y
where the coefficients CABC denote the intersection numbers of the 4-cycles of the
threefold (A = 1, . . . , NV ). Solving the attractor equations (3.29) in terms of the harmonic
functions H0 and H A , and using (3.25), yields
F (Y ) =
0,
Y0 = Y
i
Y A = H A,
2
Y0
2
DABC H A H B H C
,
4H0
e2U = 4Y 0 H0 .
(4.3)
= hW
. Using
Since the curvature on the wall (3.30) is zero, it follows from (3.7) that hW
5U
0
A
= |W | = e
(0 Y FA (Y )), and hence
(3.24) as well as (3.35), we then obtain hW

2

0 A B C
1 10U 0 2
2
A B C
DABC
Y
|W | = e
(4.4)
H H H + 3 H H .

16
H0
There are some issues that we did not address in this paper and which may be worth
studying in the future. First, when evaluating the pullback of the SU(2) connection of the
quaternionic space for the s = 2 solution we find that it is trivial. In general, however, we
expect that it will also contribute to the curvature of the domain wall. This effect, by the
404
way, may also happen for BPS walls in five-dimensional gauged supergravity. Second, the
geometry of the curved wall is AdS3 and it would be interesting to understand whether a
global identification can give rise to a BTZ black hole. Third, in the case of dilatonic walls,
the curvature on the wall had to be zero (I hI I hI = 0). It may, however, happen that
this restriction on the curvature is circumvented by allowing the dilaton also to depend
on the worldvolume coordinates. And finally, it would be interesting to investigate more
general potentials for the dilaton field. These can be obtained either by gauging some of
the non-compact isometries of the universal hypermultiplet moduli space or by turning
on some of the NS 3-form fluxes in type IIB CalabiYau compactifications. In particular,
SL(2, Z) covariant superpotentials may eliminate the run-away behavior of the dilaton field
and hence provide a vacuum which is stable with respect to the dilaton field.
Acknowledgements
We would like to thank G. Curio, G. DallAgata and B. de Wit for fruitful discussions.
The work of K.B. is supported by a Heisenberg grant of the DFG. This work is also
supported by the European Commission RTN Programme HPRN-CT-2000-00131.
References
[1] J. Maldacena, The large N limit of superconformal field theories and supergravity, Adv. Theor.
Math. Phys. 2 (1998) 231252, hep-th/9711200.
[2] E. Witten, Anti-de Sitter space and holography, Adv. Theor. Math. Phys. 2 (1998) 253291,
hep-th/9802150.
[3] 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.
[4] D.Z. Freedman, S.S. Gubser, K. Pilch, N.P. Warner, Renormalization group flows from
holography supersymmetry and a c-theorem, hep-th/9904017.
[5] D. Fabbri et al., 3d superconformal theories from Sasakian seven-manifolds: New nontrivial
evidences for AdS(4)/CFT(3), Nucl. Phys. B 577 (2000) 547608, hep-th/9907219.
[6] M. Cvetic, F. Quevedo, S.-J. Rey, Stringy domain walls and target space modular invariance,
Phys. Rev. Lett. 67 (1991) 18361839.
[7] M. Cvetic, S. Griffies, S. Rey, Static domain walls in N = 1 supergravity, Nucl. Phys. B 381
(1992) 301, hep-th/9201007.
[8] A. Karch, D. Lst, D. Smith, Equivalence of geometric engineering and HananyWitten via
fractional branes, Nucl. Phys. B 533 (1998) 348, hep-th/9803232.
[9] G. Curio, D. Lst, work in progress.
[10] J. Polchinski, M.J. Strassler, The string dual of a confining four-dimensional gauge theory,
hep-th/0003136.
[11] J. Polchinski, A. Strominger, New vacua for type II string theory, Phys. Lett. B 388 (1996)
736742, hep-th/9510227.
[12] J. Michelson, Compactifications of type IIB strings to four dimensions with non-trivial classical
potential, Nucl. Phys. B 495 (1997) 127, hep-th/9610151.
[13] T.R. Taylor, C. Vafa, RR flux on CalabiYau and partial supersymmetry breaking, Phys. Lett.
B 474 (2000) 130, hep-th/9912152.
405
[14] P. Mayr, On supersymmetry breaking in string theory and its realization in brane worlds, Nucl.
Phys. B 593 (2001) 99126, hep-th/0003198.
[15] G. Curio, A. Klemm, D. Lst, S. Theisen, On the vacuum structure of type II string
compactifications on CalabiYau spaces with H-fluxes, hep-th/0012213.
[16] K. Behrndt, S. Gukov, M. Shmakova, Domain walls, black holes, and supersymmetric quantum
mechanics, hep-th/0101119.
[17] S. Ferrara, R. Kallosh, Supersymmetry and attractors, Phys. Rev. D 54 (1996) 15141524, hepth/9602136.
[18] S. Ferrara, R. Kallosh, Universality of supersymmetric attractors, Phys. Rev. D 54 (1996) 1525
1534, hep-th/9603090.
[19] S. Ferrara, G.W. Gibbons, R. Kallosh, Black holes and critical points in moduli space, Nucl.
Phys. B 500 (1997) 7593, hep-th/9702103.
[20] K. Behrndt, G.L. Cardoso, B. de Wit, R. Kallosh, D. Lst, T. Mohaupt, Classical and quantum
N = 2 supersymmetric black holes, Nucl. Phys. B 488 (1997) 236260, hep-th/9610105.
[21] W.A. Sabra, General static N = 2 black holes, Mod. Phys. Lett. A 12 (1997) 2585, hepth/9703101.
[22] W.A. Sabra, Black holes in N = 2 supergravity theories and harmonic functions, Nucl. Phys.
B 510 (1998) 247, hep-th/9704147.
[23] K. Behrndt, D. Lst, W.A. Sabra, Stationary solutions of N = 2 supergravity, Nucl. Phys. B 510
(1998) 264, hep-th/9705169.
[24] G. Moore, Arithmetic and attractors, hep-th/9807087.
[25] F. Denef, Supergravity flows and D-brane stability, JHEP 08 (2000) 050, hep-th/0005049.
[26] G.L. Cardoso, B. de Wit, J. Kppeli, T. Mohaupt, Stationary BPS solutions in N = 2
supergravity with R 2 interactions, JHEP 0012 (2000) 019, hep-th/0009234.
[27] A. Lukas, B.A. Ovrut, K.S. Stelle, D. Waldram, Heterotic M-theory in five dimensions, Nucl.
Phys. B 552 (1999) 246, hep-th/9806051.
[28] K. Behrndt, S. Gukov, Domain walls and superpotentials from M-theory on CalabiYau threefolds, Nucl. Phys. B 580 (2000) 225, hep-th/0001082.
[29] L. Andrianopoli et al., N = 2 supergravity and N = 2 super YangMills theory on general
scalar manifolds: Symplectic covariance, gaugings and the momentum map, J. Geom. Phys. 23
(1997) 111, hep-th/9605032.
[30] B. de Wit, B. Kleijn, S. Vandoren, Superconformal hypermultiplets, Nucl. Phys. B 568 (2000)
475502, hep-th/9909228.
[31] B. de Wit, M. Rocek, S. Vandoren, Hypermultiplets, hyperkahler cones and quaternion-Kahler
geometry, hep-th/0101161.
[32] A. Ceresole, G. DallAgata, General matter coupled N = 2, D = 5 gauged supergravity, Nucl.
Phys. B 585 (2000) 143170, hep-th/0004111.
[33] K. Behrndt, C. Herrmann, J. Louis, S. Thomas, Domain walls in five-dimensional supergravity
with non-trivial hypermultiplets, JHEP 01 (2001) 011, hep-th/0008112.
[34] K. Behrndt, M. Cvetic, Gauging of N = 2 supergravity hypermultiplet and novel renormalization group flows, hep-th/0101007.
[35] B. de Wit, P.G. Lauwers, A.V. Proeyen, Lagrangians of N = 2 supergravity matter systems,
Nucl. Phys. B 255 (1985) 569.
[36] O. Coussaert, M. Henneaux, Supersymmetry of the (2 + 1) black holes, Phys. Rev. Lett. 72
(1994) 183186, hep-th/9310194.
[37] R. Kallosh, A. Linde, Supersymmetry and the brane world, JHEP 02 (2000) 005, hepth/0001071.
[38] K. Behrndt, M. Cvetic, Anti-de Sitter vacua of gauged supergravities with 8 supercharges, Phys.
Rev. D 61 (2000) 101901, hep-th/0001159.

Gravitational waves in non-singular string

cosmologies
C. Cartier a , E.J. Copeland a , M. Gasperini b
a Centre for Theoretical Physics, University of Sussex, Falmer, Brighton BN1 9QJ, UK
b Dipartimento di Fisica, Universit di Bari, Via G. Amendola 173, 70126 Bari, Italy
Abstract
We study the evolution of tensor metric fluctuations in a class of non-singular models based on
the string effective action, by including in the perturbation equation the higher-derivative and loop
corrections needed to regularise the background solutions. We discuss the effects of such higher-order
corrections on the final graviton spectrum, and we compare the results of analytical and numerical
computations. 2001 Elsevier Science B.V. All rights reserved.
1. Introduction
Recent studies, including strong coupling corrections to the string effective action, have
provided the pre-Big Bang scenario [1,2] with a number of promising models for the
transition from the growing to the decreasing curvature regime [36]. However, a crucial
question remains: is it possible to test this cosmological scenario and, in particular, to
distinguish it from other, more conventional descriptions of the very early Universe?
The answer to this essential question is in principle positive, since the pre-Big Bang
scenario incorporates a dilaton field which modifies the standard inflationary kinematics
and which, through its non-minimal coupling, also directly modifies the evolution of
perturbations (of the metric, and of the other background fields) [7]. It is well known,
on the other hand, that the transition from an inflationary period to the usual Friedman
RobertsonWalker (FRW) phase amplifies tensor metric perturbations, and is associated
with the production of a stochastic background of relic gravitational waves [8,9].
Such a primordial background decouples from matter immediately below the Planck
scale, unlike the electromagnetic radiation which underwent a complicated history until
recombination, and has been transmitted almost unperturbed down to our present epoch. As
E-mail addresses: c.cartier@sussex.ac.uk (C. Cartier), e.j.copeland@sussex.ac.uk (E.J. Copeland),
gasperini@ba.infn.it (M. Gasperini).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 8 2 - 1
C. Cartier et al. / Nuclear Physics B 607 (2001) 406428
407
a consequence, its present spectrum should be a faithful portrait of the very early universe
[10], thus opening a window for the observation of processes occurring near to the Planck
scale, and for discriminating among various models of primordial evolution.
In the context of string cosmology [11], several mechanisms may contribute to
the generation of gravitational waves from the initial vacuum state. One contribution
comes from the process of dynamical dimensional reduction [1214], when the internal
dimensions shrink down to a final compactification scale. Another contribution is due to
the time-dependence of the dilaton field and thus of the effective gravitational coupling
constant during the pre-Big Bang phase [15,16]. Finally, there is the usual contribution
arising from the accelerated expansion of the external three-dimensional space. Both the
dilaton and the full, higher-dimensional metric thus contribute to the external pump
field which is responsible for the parametric amplification of tensor metric fluctuations,
normalised to an initial vacuum fluctuation spectrum.
With respect to the standard inflationary scenario, the amplification of tensor perturbations in the pre-Big Bang scenario is strongly enhanced for large comoving wavenumber k
[15,17], in such a way that the produced background of gravitational waves could in principle be detected in the near future by various planned experiments [2,17,18]. Arguments
in favour of this conclusion arise from the particular kinematics of string cosmology. First,
since the Hubble parameter is increasing in the pre-Big Bang phase, higher frequency
modes will cross the horizon at higher values of |H |, which implies that the amplitude of
the spectrum increases with frequency. In this context, the comoving amplitude of perturbations may grow even outside the horizon [1921], instead of being frozen as happens
in standard inflation. Second, the peak amplitude of the spectrum (naturally located in the
high frequency regime, near the end point of the spectrum) is normalised so as to match the
string scale [18,22], i.e., nearly eight orders of magnitude above the high frequency normalisation of the standard scenario [23]. Hence, the pre-Big Bang phase could lead to a very
efficient production of gravitons in the sensitivity range of present gravitational antenna.
In fact such a relic background could in principle be detected by the second (planned)
generation of interferometric detectors [24,25].
To date, extended studies of tensor perturbations have been mostly performed in the
context of the tree-level effective action, with and without the possible contribution of
a high curvature stringy phase. In all cases the background cosmologies that have been
investigated have been singular, i.e., the curvature and dilaton have become singular at
some fixed proper time. There have been a number of suggested ways to tackle the
singularity problem, considering also anisotropic [26,27] and inhomogeneous [28,29]
backgrounds, or the presence of a non-local dilaton potential [2]. However, if we confine
our attention to homogeneous and isotropic metrics, and to a local potential, one of the
most promising approaches to the graceful exit problem [30] suggests that the curvature
singularities may be cured by adding higher-order corrections to the string effective action
(see for instance [3,4,6,3133]). These include both higher-derivative corrections, due to
finite-size effects, and quantum corrections, due to the loop expansion in the string coupling
parameter. In the context of cosmological perturbation theory, it is known [34] that such
corrections may induce an additional amplification when the curvature becomes large in
408
string units. The aim of this paper, therefore, is to discuss the evolution and the spectrum of
tensor perturbations in the context of a class of non-singular cosmological backgrounds, by
taking into account (in the perturbation equation) the full contribution of those corrections
that are responsible for the regularisation of the background evolution. In this way it will be
possible (for the first time, to the best of our knowledge) to follow the complete evolution
of perturbations, from the initial vacuum down to the present time, through a continuous
numerical integration of the linearised perturbation equations.
The paper is organised as follows. In Section 2 we recall the general form of the
first order corrections arising in the context of the massless bosonic sector of the
low-energy heterotic string action. We extend them by including possible one- and twoloop quantum corrections, and we write down the corresponding background equations.
In Section 3 we compute the linearised tensor perturbation equation, including such
higher-order corrections. In Section 4 we present a general discussion of the evolution
of perturbations in the presence of higher-order curvature corrections. We then check the
main points of our discussion through a numerical integration of the perturbation equations
for regular backgrounds, by using an appropriate dilaton potential, or particle production
effects, to stabilise the dilaton in the post-Big Bang era. We find that the general effect
of the higher-order corrections is to flatten the slope of the high-frequency part of the
spectrum, but the effect is probably too small in the class of models discussed in this
paper to make the graviton background visible to the planned advanced detectors. The
main results of this paper are finally summarised in Section 5.
2. Effective action and background equations

In the context of the pre-Big Bang scenario, our present FRW Universe is expected
to emerge from the string perturbative vacuum as a consequence of some process for
instance, plane wave collisions [35] triggering the gravitational collapse of a sufficiently
large portion of spacetime [36]. The initial evolution is driven by the kinetic energy of the
dilaton, and can be described, in the string frame, as inflationary expansion with growing
coupling and curvature. After reaching a maximal scale controlled by the fundamental
string length S , |H | O(1
S ), it is then expected that the Universe is smoothly connected
to the FRW regime, with a constant dilaton field.
In the early stage of the dilaton-driven inflationary (DDI) period, strings are propagating
in a background of small curvature, and the fields are weakly coupled, so that the
cosmological evolution is consistently described by the lowest order string effective action
which, by assuming static internal dimensions, can be written as:

1
= d 4 x g e R + ( )2 + Lc .
(1)
Here = 2S , we have adopted the convention (, +, +, +), R = + , R =
and set our units such that h = c = 16G = 1. Following [3], we have introduced
R
an additional Lagrangian Lc allowing for the inclusion of higher-order corrections to the
409
tree-level effective action, a non-perturbative potential yet to be determined and/or

the backreaction of the produced radiation. Considering this additional Lagrangian as
an independent source we shall assume that the associated energymomentum tensor is
diagonal, homogeneous and isotropic, and can be written in the perfect fluid form as T =
diag(c , pc , pc , pc ).
The resulting equations of motion for the background are obtained from the variation of
Eq. (1) with respect to the metric and the dilaton field:
6H 2 6H + 2 = e c ,
4H 6H 2 2 + 2 = e pc ,
4H
= e ,
6H + 12H 2 + 2 2 6H
c + 3H (c + pc ) = .
(2)
(3)
(4)
(5)
A dot represents the differentiation with respect to proper time t and H = a/a
is the Hubble
parameter. arises from the variation of the additional Lagrangian with respect to .
Finally, Eq. (5) is the conservation equation, a direct consequence of Eqs. (2)(4).
To lowest order, the homogeneous and isotropic version of the pre-Big Bang scenario,
without non-local dilaton potential, inevitably faces a curvature and dilaton singularity at
the end of the DDI phase. However, a singular behaviour is often a late manifestation
of the breakdown of the description, extended beyond its domain of validity. Hence, the
low-energy dynamics should be supplemented by corrections, in order to give a reliable
description of the high curvature regime, expected to be crossed before reaching the FRW
universe. Such corrections are generally of two types: tree-level corrections, referring to
the finite size of the string, and quantum corrections, resulting from a more conventional
loop expansion in the string coupling parameter.
It has been shown in [33] that the higher curvature terms, arising from the expansion
around the point-particle limit, may eventually stabilise the evolution of the Universe into
a de Sitter like regime of constant curvature H O(1
S ) and linearly growing (in cosmic
time) dilaton. In this paper we will consider the most general tree-level corrections to the
gravi-dilaton string action up to fourth-order in derivatives (see [6] for a detailed analysis),
which can be written in the form [37]:

2
+ c2 G + c3 ( )2 + c4 ( )4 ,
L = 0 e c1 RGB
(6)
c1 = 1,
c2 + 2(c3 + c4 ) = 2c1 .
(7)
The constraint Eq. (7) on the coefficients is required so that the action reproduces the usual
string scattering amplitudes [37]. The parameter 0 allows us to move between different
string models, and we will later use 0 = 1/4 to agree with previous studies of the
2 =R
4R R + R 2 is the GaussBonnet
heterotic string [33]. Finally, RGB
R
combination, guaranteeing the absence of derivatives higher than two in the equations of
motion.
Another conventional expansion of the string effective action is controlled by the string
2
2
= e . Initially, gstring
1. However, the growth of the dilaton
coupling parameter, gstring
410
during the DDI era implies that the late evolution of the background could be dominated by
quantum loop corrections, which will eventually force the universe to escape from the fixed
point determined by the corrections [33], for a smooth connection to the post-Big Bang
regime. Unfortunately, there is as yet no definitive calculation of the full loop expansion in
string theory, so we are left to speculate on plausible terms that will eventually make up
the loop contribution. Multiplying each term of the tree-level correction by a suitable
power of the string coupling is the approach which has already met with some success
[3,5,6], and which we shall adopt also in this paper. Since the quantum corrections are not
formally derived from an explicit computation, we shall allow for different coefficients di
(ei ) at one-loop (two-loop) in the string coupling, replacing the coefficients ci of the treelevel action, where di and ei are not necessarily subject to the constraint Eq. (7). With such
assumptions, the effective Lagrangian density Lc of Eq. (1), which is in general the sum
of the tree-level and loop corrections, Lc = L + Lg , in the present case will take the
form
Lc = L (ci ) + Ae L (di ) + Be2 L (ei ),
(8)
where L (ci ) is given by Eq. (6), with the constant parameters A and B actually
controlling the onset of the loop corrections.
The very similarity between the first order and loop corrections enables a simple form
for their contributions to the background equations of motion. From now on, we introduce
an additional parameter n in the exponential of the correction Lc , such that corrections
correspond to n = 0, whereas one- or two-loop corrections correspond to n = 1 and n = 2,
respectively. For instance, using the notation { }1 to represent the one-loop contribution
(n = 1 inside the brackets), we easily get the source terms for the system Eqs. (2)(4):
c = {c }0 + A{c }1 + B{c }2 ,
(9)
pc = {pc }0 + A{pc }1 + B{pc }2 ,
(10)
= { }0 + A{ }1 + B{ }2 ,
(11)
and the terms in brackets are given, respectively, by:

(n1) 24c1 (n 1)H 3 + 9c2 H
2
{c }n = 0 e

+ 6c3 2 H c3 (n 1) 3c4 3 ,

{pc }n = 0 e(n1) 8c1 (n 1)2 3c2 2 H 2 2c2 (n 1) 3 H 2c3 2

2 + 16c1 (n 1)H
H + H 2
+ 8c1 (n 1)H

2c2 2 H + [c4 c3 (n 1)] 4 ,
4c2 H

{ }n = 0 e(n1) 24c1 (n 1)H 2 H + H 2

+ (n 1) 2
3c4 2 4 + 4H

+ c3 (n 1)2 3 + 4(n 1) 6 H + 3H 2 12H

2 + 6H
3 + 4H
H + (n 1) 2 H 2 .
3c2 2H
411
Here, of course, for n = 1 the coefficients ci are replaced by di , while for n = 2 the ci are
replaced by ei . Note that for n = 0, 0 = 1/4, c4 = c1 = 1, one easily recovers the
corrections used in [33,34].
In general, the combination of tree-level and quantum loop corrections does not lead
to a fixed value for the dilaton in a finite amount of time. Although a non-perturbative
(supersymmetry breaking) potential [38] is expected to stabilise the dilaton, here we shall
follow [3] by assuming the dilaton is frozen out by radiation, after the transition to the
post-Big Bang regime. To this aim we will introduce by hand the presence of radiation,
coupled to the dilaton and satisfying the conservation equation
1
+ 4H 2 = 0,
(12)
2
where represents the decay width of the dilaton.
The solutions of the system of equations Eqs. (2)(4), including the radiation and the
quantum corrections, provide the cosmological gravi-dilaton background in which we
will study the propagation of tensor perturbations. We will consider a spatially flat FRW
manifold, and we will perturb the above equations keeping the dilaton and all sources
fixed, = 0, T = 0. To lowest order, the tensor fluctuations will obey the usual string
frame perturbation equation [15], including the non-minimal coupling to a time-dependent
dilaton. On the other hand, the inclusion of higher-order corrections, needed to regularise
the background, will inevitably lead to a modification of such a perturbation equation
(see [3942] for initial studies taking into account higher curvature contributions to the
evolution of perturbations).
3. Tensor perturbations
It is well known that gravitational waves, arising from linearised tensor perturbations,
do not couple to pressure and energy density, and hence do not contribute to classical
gravitational instabilities. Nevertheless, they are of interest as a specific signature of preBig Bang cosmologies since, as stressed in a number of papers [2,15,17], the production
of high-frequency gravitons is strongly enhanced, in string cosmology, with respect to the
standard inflationary scenario. In this study we will restrict our attention to tensor metric
perturbations around a d = 3 spatially flat background, parameterised by the transverse,
trace-free variable h ,
g g + h
with h = 0 = h
,
(13)
where denotes covariant differentiation with respect to the background metric. Note
also that the indices of h are raised or lowered with the unperturbed metric, h =
g h .
3.1. Linearised equations
The linear evolution equation for the metric fluctuations can be obtained by perturbing
the action Eq. (1) up to terms quadratic in the tensor variable h :
412
(2) =

d 4 x e (2)( g R) + (2) g

+ 0 d 4 x (2) L 0 + A (2) L 1 + B (2) L 2 ,
1
(14)
where the integrand of the n-correction is given by:

(2)

2
L n = e(n1) (2) c1 g RGB
+ (2) c2 g R 12 g R

(15)
+ (2) c3 g ( )2 + (2) c4 g ( )4 .
Following [34], it is convenient to use the synchronous gauge where
g00 = 1,
h00 = 0,
g0i = 0,
h0i = 0,
gij = a 2 ij ,
g ij hij = 0 and j hj i = 0.
(16)
This enables us to write the perturbed action as a quadratic form depending on the first
and second derivatives of the symmetric, trace-free matrix h = hi j , with time-dependent
coefficients fixed by the background fields a(t), (t). Introducing the matrix notation
Tr h2 hi j hj i , and using 2 = ij i j for the Laplace operator in a flat space, we have

1
(2)
= d 4 x a 3 e

1 2 3

3 2 1 2
2 2
Tr 4 2 H 3H h hh 4H hh h + h 2 h
4
4 a

+ 0 d 4 x Tr {}0 + A{}1 + B{}2 ,
(17)
where the integrand of the n-correction is given by:

{}n = a 3 e(n1) 6c1H 2 H + H 2

1
+ c4 2 h2
2 3c2 H 2 + c3 + 3c3 H
4

c1 H + 7c1 H 2 + 18 c2 2 h 2 4c1 H 2 hh 2c1 H h h

2
8c1 H H + 2H 2 + c2 2 H + 12 c3 3 hh + 2c1 h 2 h
a

2
2

2
2
1
+ c1 H + c1 H + 8 c2 h 2 h + 4c1 H h 2 h + 2c1 h 2 h .
a
a
a
Here, and in what follows, the coefficients ci are to be replaced by di and ei for n = 1
and n = 2, respectively. Again, for n = 0, 0 = 1/4, c4 = c1 = 1, one recovers the first
order corrections discussed in [34].
We can now integrate by parts all the terms with more than two partial derivatives acting
This will drop all terms with more than two
on h, as well as the terms in hh and hh.
derivatives thanks to the GaussBonnet combination, leading to
(2) =
1
413
d 4 x a 3 e

1
2 1 2 1 2
1 2
3 2
Tr 2 4 H 2 H + H h + h + h 2 h
4
4 a

+ 0 d 4 x Tr {}0 + A{}1 + B{}2 ,
(18)
where we have
1

{}n = a 3 e(n1) h2 8c1 (n 1)2 3c2 2 H 2 2c2 (n 1) 3 H
4
2
+ [c4 c3 (n 1)] 4 2c3 2 + 8c1 (n 1)H

H + H 2 4c2 H
2c2 2 H
+ 16c1(n 1)H

2
h c1 (n 1)2 + 18 c2 2 + c1 (n 1)
a2

1 c2 2 .
+ h 2 c1 (n 1)H
8
+h
We stress that all tree-level and quantum loop corrections disappear in the limit of
a constant dilaton field, since in that case the contribution of the GaussBonnet term
corresponds a topological invariant. This ensures that the resulting perturbation equation
reduces to the KleinGordon form, typical of a minimally coupled scalar, as we enter the
frozen dilaton, post-Big Bang regime.
The rather complicated form of Eq. (18) can be simplified as the coefficient of the h2
term vanishes identically, thanks to the background equation Eq. (3). By decomposing the
matrix hij into the two physical polarisation modes of tensor perturbations, h+ and h ,

j
Tr h2 hij hi = 2 h2+ + h2 ,
(19)
we can finally write the action, for each polarisation mode h(x, t), as
(2) h =
1
2

d 4 x a 3 e h 2 1 + {}0 + A{}1 + B{}2
+h

2
h
1
+
{}
+
A{}
+
B{}
,
0
1
2
a2
(20)
with

1 c2 2 ,
{}n = 0 en 4c1 (n 1)H
2

{}n = 0 en 4c1 (n 1)2 2 + 12 c2 2 + 4c1 (n 1) ,
where h is now a scalar variable standing for either one of the two polarisation amplitudes
h+ , h . The variation of the action with respect to h gives then the modified perturbation
equation:
414

0 = 1 + {}0 + A{}1 + B{}2 h

+ 3H + {}0 + A{}1 + B{}2 h
2

1 + {}0 + A{}1 + B{}2 2 h,
a
with
(21)

1 c2 2 ,
{}n = 0 en 4c1 (n 1)H
2

+ H ) c2 1 c2 (n 1) 3
{}n = 0 en 4c1 (n 1)(H
2

2 + 4c1 (n 1)2 3 c2 2 H ,
+ 12c1(n 1)H
2

n
2 2
1
2
{}n = 0 e 4c1 (n 1) + 2 c2 + 4c1 (n 1) .
Eq. (21) controls the time evolution of the Fourier components hk of the two polarisation
modes, and is valid even during the high curvature regime connecting the pre-Big Bang
branch to our present FRW universe, since it incorporates the contributions of the higherorder corrections needed to regularise the background evolution.
3.2. Quantisation in conformal time
We shall now focus on the generation of gravitational waves in the context of the preBig Bang scenario. To normalise the graviton spectrum to the quantum fluctuations of
the vacuum we need the canonical variable that diagonalises the perturbed action, and
that represents in this case the normal modes of tensor oscillations of our gravi-dilaton
background. To this purpose we note that introducing the conformal time coordinate ,
defined by a = dt/d, the action Eq. (20) can be re-written

1
2
(2) h = d 3 x d z2 h + y 2 h 2 h ,
(22)
2
where a prime denotes differentiation with respect to the conformal time, and

y 2 () = e a 2 + {}0 + A{}1 + B{}2 ,
(23)

2
2
a + {}0 + A{}1 + B{}2 ,
z () = e
(24)
with

1
a
2
2
,
{}n = 0 en 4c1 (n 1)2 + c2 + 4c1 (n 1)
2
a

a 1
2
{}n = 0 en 4c1 (n 1)
c2 .
a
2
Introducing the canonical variable = zh enables us to diagonalise the kinetic part of the
action:

1
z 2 y 2
(2)
3
2
2
h = d x d + + 2 .
(25)
2
z
z
For each Fourier mode, k = zhk , we can now obtain from the perturbed action a linearised
wave equation in terms of the eigenstates of the Laplace operator, 2 k = k 2 k which
takes the explicit form:

2

2y
k + k 2 V () k = 0,
z
V () =
z
.
z
415
(26)
This linear perturbation equation is the major result of the first part of this paper,
since it encodes (through Eqs. (23)(24)) the full contribution of those higher-order
corrections which can be used to regularise the background evolution. We note that the
above equation defines an effective time-dependent mode, ky/z, whereas the effective
potential is determined by the evolution of the generalised pump field z(). These new
features of our generalised perturbation equation will be discussed in detail in the next
section.
The graviton spectrum of pre-Big Bang models generated in the context of the lowestorder perturbation equation, with and without the contribution of a high curvature string
phase, has been discussed extensively in the literature (see for example [7,18,21,4345]).
In the remaining part of this section we shall briefly recall some general results of these
studies, so as to highlight the modifications induced by the higher-order corrections in the
perturbation equation, and in the spectrum of the relic pre-Big Bang gravitons.
In the extremely weak coupling and low curvature regime, the tree-level solutions of the
pre-Big Bang scenario are fully adequate to describe, asymptotically, the initial evolution
of the quantum fluctuations of the gravi-dilaton background. In normalising the fluctuations
in the asymptotic past ( ), we shall thus neglect any correction to the low-energy
action, considering the standard perturbation equation for the Fourier mode k ():

( 1)
k + k 2 1
(27)
k = 0.
(k)2
Such an equation describes the evolution of a classical harmonic oscillator undergoing a
parametric amplification driven by the effective potential V = / , with () . By
recalling the (conformal time) expression of the tree-level solutions [7],
a() = () ,
(28)
= (1 3 )/2,
() = ln(),
(29)
3,
ae/2
=
= 1/2. Hence, the power-law
we obtain the external pump field =
type of the effective potential leads to two distinct regimes for the evolution equation (27).
Sub-horizon modes (|k| 1) are freely oscillating, and no quantum particle production
occurs. However, the non-adiabatic behaviour of modes having left the Hubble radius
(|k| 1) in the dilaton-driven phase, are described asymptotically by the general solution
[46]

hk () = Ak + Bk
d
,
2 ( )
() ,
(30)
exit
where exit = k 1 , and hk = k 1 is the comoving amplitude of tensor perturbation for

the given mode (Ak and Bk indicate the coefficients of the growing and decaying solutions,
416
respectively). The amplitude of super-horizon modes is thus frozen, modulo logarithmic

corrections. The overlapping of these two extreme behaviours is given by the exact solution
to the Bessel-type equation (27), which can be expressed in a linear combination of the
first- and second-kind Hankel functions:

k () = (k)1/2 1 H(1)(k) + 2 H(2)(k) ,
(31)
where = |1 2 |/2 = 0 encompasses the background dynamics.
In the asymptotic past ( ) any given mode k is well inside the Hubble radius (as
H 1 blows up as we go back in time in pre-Big Bang models), and the two types of Hankel
functions H(1,2)(k||) are oscillating, corresponding to negative and positive frequency
modes, respectively. It is therefore possible to normalise tensor perturbations to a spectrum
of quantum fluctuations of the initial background. Indeed, the annihilation and creation
operators resulting from the expansion of the h field, with (k, ) = ak + ak ,
obey canonical commutation relations [ak , ak ] = [ak , ak ] = 0 and [ak , ak ] = kk . For
the normalisation to an initial vacuum state,
to restrict to positive energy states,
we choose
1
i 4 (1+2)
ik
e
, hence 1 = 0 and 2 = 2 e
.
k
2k
The effective potential of the perturbation equation is expected to grow monotonically
during the initial dilaton-driven regime, up to a maximal value reached during the string
phase, while it is expected to decrease rapidly to zero at the onset of the standard, radiation
dominated phase. Emerging from the asymptotic past, a typical mode k will thus first
oscillate inside the horizon and then progressively feel the potential barrier. The process of
quantum particle production will take place and carry on until the background enters the
standard FRW-regime. After that, the effective potential identically vanishes, and Eq. (27)
yields the free oscillating solution:

1
c+ (k)eik + c (k)eik , +,
k =
(32)
2k
with incoming and outgoing waves. The mean number of produced gravitons is obtained
from the square modulus of the Bogoliubov coefficient, |c (k)|2 , using the matching of
the tensor perturbation variable h and h at the onset of the FRW phase.
The low-energy perturbation equation may allow for estimating a lower bound on the
spectrum of the produced gravitons [47]. For the tree-level solution with = 1/2, by
defining the spectral energy density per logarithmic interval of frequency, gw (k), one
obtains in particular [7]:

k14 k 3
dgw
k4
2
=
gw (k)
(33)
|c (k)| 2
,
d ln k 2
k1
where k1 |1 |1 refers to the maximal height of the effective potential barrier,
|V (1 )|1/2 |1 |1 . We recall that the cubic slope of the spectrum is a direct consequence
of the evolution of the background during the DDI branch. In [18,43,44], the authors argued
that the contribution of a high curvature string phase would lead in general to a reduction
of the slope of the spectrum of relic gravitational waves. This was explicitly confirmed
for a model based only on the tree-level corrections [34]. However, we stress that the
417
investigation of the evolution of high frequency modes, hitting the potential barrier during
the high curvature regime, needs to be performed using the full perturbation equation
Eq. (26), including also the loop corrections.
4. Results
In this section we highlight the modifications to the tensor perturbation equations
induced by the corrections adopted to regularise the background evolution. To this purpose,
we shall first restrict the discussion to the tree-level corrections; the reason for this
choice is twofold.
First, corrections are derived unambiguously since we require them to reproduce the
string scattering amplitude, as opposed to our lack of precise knowledge for quantum
loop corrections, which forces us to speculate on plausible terms. Second, if any clear
imprint of a string phase like the one introduced in [18,43,44] is to be found in
the spectrum of relic gravitons, it should be a direct consequence of a long enough string
phase 1 characterised by a constant Hubble parameter. As suggested in [33], this can be
achieved by assuming that the evolution of the universe is driven by corrections into a
H ) space.
fixed point in the (,
In the final part of this section we will comment on the impact of the quantum loop
corrections, and confront our predictions with the results obtained through exact numerical
integrations of the perturbation equation.
4.1. Tree-level corrections
The lowest-order effective action of string theory provides an adequate description of
the dilaton-driven branch of the pre-Big Bang scenario. In this weakly coupled and low
curvature regime, no correction is required either to the effective potential nor to the wavelength of a given perturbation, and the evolution equation of each Fourier mode reduces to
its usual form Eq. (27). However, it is well known that the kinetic energy of the dilaton necessarily drives an isotropic and homogeneous background towards a curvature singularity,
imposing the need to supplement the tree-level action with corrections in order to establish
a reliable description of the high curvature regime. The very presence of the tree-level
corrections Eq. (6), subject to the constraint Eq. (7), may lead in general to a regime of
constant Hubble parameter and a linearly growing dilaton, H = const, > 0, which seems
to represent an exact conformal solution of string theory also to all orders in [33].
However, at any given finite order, we recall that not all sets of coefficients of the
H ) plane (see [6] for a detailed
truncated action lead to such a good fixed point in the (,
analysis). We shall thus restrict the analysis to the region of parameters which in principle
may allow for a graceful exit, when loop corrections are included. In the (c2 , c4 ) parameter
space, such a region is delimited on the one hand by c2 38/9, referring to a change of
sign of the time derivative of the shifted dilaton, = 3H [48]. On the other hand,
1 By long string phase, we mean N 1, where N = ln(a) is the number of e-folds.
418
H ) space of the fixed points

Fig. 1. The figure on the left, from Ref. [6], shows the location in the (,
. The curve
for different choices of the parameters c2 and c4 , and the corresponding ratio /H
which links points (1) and (2) shows the c2 -dependence of the fixed points for a constant c4 = 1,
while the curve which links points (1) and (3) shows the c4 -dependence for a constant c2 = 0.
The point labeled by (1) corresponds to c2 = 0, c4 = 1, with (1) 0.72. The point labeled by
(2) corresponds to c2 = 40, c4 = 1, with (2) 0.56. The point labeled by (3) corresponds to
c2 = 0, c4 = 15, with (3) 0.13. On the right, we show the tree-level asymptotic shift c given
by Eq. (35) as a function of the parameter c2 and of the fixed point distribution. The line shows the
evolution of the shift for a constant c4 = 1.
tree-level corrections, truncated to first order, are not expected to violate the null energy
condition, and thus cannot lead in principle to fixed points located after the Einstein bounce
[48] (i.e., the bounce of the scale factor in the Einstein frame). This implies c4 c2 /2 for
< c2 < 4, to a first approximation. The location of such fixed points, satisfying the
above constraints, can be seen in Fig. 1.
4.2. Effective frequency shift
To describe the evolution of the high frequency modes, which are expected to leave
the Hubble radius during the high curvature regime, we must use the perturbation
equation (26), including those corrections required to regularise the background evolution.
As mentioned in Section 3.2, the incorporation of the higher-order corrections in the
perturbation equation implies a time-dependence of the effective wavenumber k. Indeed,
Eq. (26) can be conveniently rewritten

k + k 2 [1 + c()] V () k = 0,
(34)
where we have introduced the effective shift in the comoving frequency c (y 2 z2 )/z2
which, at next-to-leading order in the string tension expansion, reads

2 (4 c2 ) 2 4( 2 a a )
.
c=
(35)

8a 2 8 aa c2
In the weak coupling regime, and for small curvature ( 0), the shift in the
frequency is zero, hence no modification to the DDI cubic branch of the relic gravitational
419
spectrum is expected. However, this frequency shift will be no longer negligible when
H O( ). In general, we find that c() tends to grow, although a short decrease (which
still satisfies c() > 1) may happen at the onset of the string phase. We also observe
that the modification arising in the asymptotic regime, where the curvature is saturated by
a linearly growing dilaton, corresponds to a constant shift of the comoving frequency, as
, we can
first suggested in [34]. In that case, by setting = /a and H = a /a 2 H
rewrite the asymptotic version of Eq. (35) as
c=
}
2 {(4
c2 ) + 4H
.
2 8H
}
8 + {c
(36)
During such an intermediate string phase the external potential reduces to its treelevel version, namely V () = /, = ae/2 , and the enhancement in frequency
associated to Eq. (35) is the same for all modes, since the shift c does not depend on
the wavenumber k.
In Fig. 1 we have represented the effective shift as a function of the parameter c2 ,
highlighting the c2 -dependence by drawing the curve for c4 = 1. Although the asymptotic
shift cannot be expressed as a function of an unique variable, it is however possible to
extract its minimal value for the type of correction we are considering. The numerical
simulations suggest that the lowest shifts correspond to fixed points located near the branch
change line, 3H = 0. The bound c2 = 38/9 attached to the sign change of
the effective asymptotic shift is always
implies that, in the limit of large values of ,
positive, and is given by c 10/7.
Does the frequency shift have any implication on the spectral distribution? In principle
the answer is yes, since higher frequency modes tend to leave the Hubble radius at larger
values of H in the pre-Big Bang scenario. As a consequence, a given comoving mode
will spend less time under the potential barrier, resulting in a smaller amplification of the
frequency attached to this mode. Although it is difficult to quantify such an effect, we
believe the higher-curvature corrections amount to an overall rescaling, by a numerical
factor of order unity, of the total energy density of the background, as already discussed in
[34].
Finally, the form of the effective shift Eq. (35) suggests that a background undergoing
+ c2 2 = 0 will automatically imply
a non-singular evolution in the vicinity of 8 8H
a rapid growth (if not a divergence) of the amplitude of the tensor fluctuations. Hence, we
can understood the denominator of Eq. (35) as a source of quantum gravitational instability,
of the type already discussed in [49,50].
4.3. Full corrections and non-singular evolution
In recent studies of the graceful exit problem, in the context of the pre-Big Bang
scenario, the regularisation of the background curvature and the transition to the post-Big
Bang regime are obtained by including quantum loop corrections, according to Eq. (8).
In this context, we may note that the tree-level shift Eq. (35) provides an interesting
constraint on the choice of parameters we can use in order to implement such a graceful
420
exit, since in general the divergence curve for the high curvature shift is very close to the
H ) plane. In [6], it has been shown that for a constant
condition for the fixed point in the (,
c4 the fixed point value of the Hubble parameter tends to be reduced when c2 is decreasing.
H ) plane, so
Simultaneously the location of the singularity curve also decreases in the (,
that the range of coefficients leading to a non-singular evolution of the background, and
avoiding the instability of quantum fluctuations, is strongly restricted to values very close
2 ()4 } (as also discussed in [33]).
to the simplest case of correction, L {RGB
This is an unexpected, although fortunate constraint on the loop corrections, which allows
restricting the range of values for the parameters di and ei .
In our attempt to characterise the power spectrum of metric perturbations generated in
such a class of non-singular models, we will thus consider the perturbation equation (34)
with a full shift determined by

2 2 [4 c2 Ad2e (4e1 + e2 )Be2 ] 4( 2 a a )[1 + 4Be1 e2 ]
.
c=

8a 2 8 aa [1 + Be1 e2 ] [c2 + Ad2 e + Be2 e2 ]
(37)
Here the coefficients of the high curvature, one- and two-loop corrections are, respectively,
ci , di and ei , with c1 = 1 from the constraint Eq. (7).
As already stressed, this frequency shift reduces to a constant in the asymptotic stage of
the string phase. The loop corrections re-establish the time-dependence during the graceful
exit, but its contribution soon becomes negligible at the end of the phase of accelerated
evolution. Indeed, Eqs. (23)(24) rapidly reduce to their tree-level expressions y z
ae/2 , hence we expect keff k before we enter the radiation-dominated epoch, where
the frequency shift vanishes identically.
4.4. Observable implications
We are interested in the generic features of the primordial gravitational waves, generated
by quantum fluctuations of the background metric during the pre-Big Bang phase, which
could be detected by both ground-based detectors such as LIGO [51] and VIRGO [52],
and space-based experiments such as LISA [53] (see also [54] for a recent review and
references therein). For such a purpose, we use the dimensionless spectral energy density
of perturbations, gw (), to describe the background of gravitational radiation. In critical
units, the energy density is defined by
gw (, t)
1 dgw
,
c d ln
(38)
where dgw is the present energy density in the stochastic gravitational waves per
logarithmic interval of frequency, (t) = k/a(t) is the physical (angular) frequency of
the wave, and c (t) = 3Mp2 H 2 (t)/8 is the critical energy density required to close the
universe.
Following the standard procedure for the computation of the energy spectrum [46], we
=
must determine the expectation number of gravitons per cell of the phase space, n(
x , k)
nf , which only depends on the frequency for an isotropic and stochastic background.
421
Neglecting and loop corrections, both in the background and perturbation equations,
the lowest order result Eq. (33) leads to [7]
3
8 14
gw (, t)
(39)
, < 1
2
2
3 Mp H 1
where 1 = k1 /a = H1 a1 /a is the maximal amplified proper frequency, associated with
the value H1 of the Hubble expansion parameter at the end of the string phase.
Including higher-order corrections, the relation between the ultraviolet cutoff of the
spectrum and the time of horizon crossing is in general more complicated, and a precise
definition of the maximal amplified frequency is out of reach without a complete analytic
solution for the background dynamics. However, a reliable estimate of the slope of the
intermediate string branch of the spectrum can be obtained by using the leading features of
the string phase.
First, we shall neglect (during the string phase) the effective frequency shift in the
perturbation equation, whose role is simply expected to induce an overall rescaling of
the amplitude [34], as already pointed out in Section 4.2. Second, we will assume that
the duration of the exit phase is small compared to the intermediate string phase,
and that no dramatic physics take place during the transition to the standard radiationdominated epoch (indeed, in some cases where the exit is catalysed by the loop corrections,
the process is almost instantaneous). This is a big assumption, of course, which could
possibly underestimate pre-heating and re-heating effects. However, the accuracy of
these assumptions can be confirmed with numerical simulations.
In summary, we suppose that the evolution of the background consists mainly of three
phases: an initial, low energy dilaton-driven branch, the intermediate string phase with a
) and a linearly growing dilaton ( = = const), and the
constant Hubble parameter (H
final radiation phase with decreasing Hubble parameter and frozen dilaton. It follows from
our assumptions that the spectral graviton distribution can be correctly estimated through
the low-energy perturbation equation, even for the high frequency branch of the spectrum.
In that case, it is already well known [18] that the slope of the high-frequency modes,
crossing the horizon in the high curvature, stringy regime, and re-entering in the radiation
H
. Indeed, during such a string phase,
era, is fully determined by the fixed point values ,
the scale factor undergoes the usual de Sitter exponential expansion, while the logarithmic
, i.e., ()
H
evolution of the dilaton, in conformal time, is weighted by the ratio /

,
(/H ) log()+ const [55]. By introducing the convenient shifted variable 3H

and referring the spectrum to a fixed point allowing a subsequent (loop catalysed) exit, i.e.,
< 0, one easily finds [18]

gw
H
3|/
|
s 1 ,
(40)
where s is the limit frequency marking the transition to the high curvature regime.
In our model of the background evolution, on the other hand, the allowed fixed points
H ) plane between the branch change line, < 0, and
are located in the (,
H
,
the Einstein bounce, < H , see Fig. 1 and [6]. It follows that, in the context of
422
our approximations, the spectrum of tensor fluctuations, at high frequency, has a slope
constrained by

< 3 (branch change).
H
(Einstein bounce) 2 < 3 /
(41)
In order to check this important analytical result, we analysed numerical solutions for
different choices of coefficients of the tree-level corrections.
4.5. Numerical results
The spectral distribution of relic gravitational waves can be obtained by numerical
integration of the perturbation equation (26), in the background Eqs. (2)(5). The initial
conditions are chosen in the low curvature and weakly coupled regime and are thus based
on the tree-level solutions, which provide an adequate description of this phase. We then
evolve the system until a given mode re-enters the Hubble radius, in the late time FRW
radiation-dominated epoch, where we extract the Bogoliubov coefficients by comparing
the result of the simulation and the free-oscillating solution Eq. (32).
To highlight the impact of the time-dependence of the frequency shift, we first present
the spectral distribution in a case where we do not consider the and loop corrections
in the perturbation equation (26). Fig. 2 illustrates the results of such a simulation, for the
particular model of background corresponding to c1 = d1 = e1 = 1, c4 = d4 = e4 = 1
(all the other ci , di , ei = 0), A = 1, B = 2 103 and = 5.63 104 . We show, in
and the evolution of the quantity aH ,
particular, the non-singular evolution of H and ,
which enables us to determine if a given mode leaves the Hubble radius during the DDI
phase or during the intermediate string phase. Finally, we present the graviton distribution
Fig. 2. The choice of coefficients for the tree-level corrections is c1 = 1, c2 = c3 = 0 and

c4 = 1 (and the same for di and ei ), corresponding to the case (1) in Fig. 1. The left figure shows a
non-singular evolution for the Hubble parameter H = a/a

and for /3,
as a function of the number
of e-folds, N = ln a. The middle figure shows the evolution of aH /Max(aH ) as function of N .
The low-energy, dilaton-driven phase takes place approximatively for < N 3. After a short
transition, this initial period is followed by a string phase with nearly constant Hubble parameter
and linearly growing dilaton, for 2 N 9. After a successful exit triggered by loop corrections,
the background evolution enters the FRW radiation-dominated phase at N 16. Comoving modes
leaving the Hubble radius during the string phase (106 k/Max(aH ) 102 ) are characterised
by a slope smaller than 3 in the spectral distribution, as illustrated in the right figure which shows
dk /d ln(k) in units of k/kmax . The upper dashed line corresponds to the cubic slope typical of the
low frequency part of the spectrum, emerging from the dilaton-driven epoch.
423
Fig. 3. Using the same choice of coefficients

as in Fig. 2, we compare the evolution of aH /Max(aH )
(solid curve) and aH /Max(aH / 1 + c ) (dashed curve) as a function of N . When higher-order

corrections are included in the perturbation equation, the high frequency cutoff and the duration of
the super-horizon epoch are found to be reduced. The right hand figure is a magnified image of
the end point of the string branch of the spectral distribution, which highlights the impact due to a
non-trivial frequency shift: the significant new feature is the reduction of the high frequency peak
and a further reduction of the slope, which may also become smaller than two, for comoving modes
leaving the Hubble radius during the exit phase.
for such a background, computed from Eq. (26), in units k/kmax , where kmax = Max(aH )
is a maximal amplified frequency.
The effects of the higher-order corrections are expected to arise when the curvature scale
becomes large in string units. As a consequence, the low frequency branch of the spectrum
is unaffected by such corrections, and remains characterised by a cubic slope. However,
modes leaving the Hubble radius during the intermediate string phase are strongly affected
by the background dynamics. The resulting slope of the spectrum is found to be reduced,
but remains confined between 2 and 3, as expected from Eq. (41).
Does the inclusion of the time-dependence of the comoving wavenumber lead to
any modification? We stress that when such a shift in the frequency is considered, we
have k 1 + c = aH at horizon crossing. As a consequence, the shortest scale to leave
the Hubble radius is given by kmax = Max(aH / 1 + c), instead of Max(aH ). This is
illustrated in Fig. 3, where we compare the results of the numerical integration with and
without corrections in the perturbation equation, for identical non-singular background
evolutions (the same model of Fig. 2). No modification appears during the low energy
epoch, where the frequency shift is negligible. However, higher-order corrections in the
perturbation equation, which result in the time-dependent shift, lead to a reduction of the
maximal amplified frequency and of the time spent on super-horizon scales.
The impact of the time-dependence of the comoving wavenumber k is non-trivial for
very high-frequency modes, leaving the Hubble radius during the graceful exit to the FRWradiation dominated epoch. In that case, the result Eq. (40) no longer applies, and the slope
can be smaller than two, as shown in Fig. 3. However, the impact remains negligible for
modes well inside the string branch of the spectrum, i.e., for modes crossing the horizon
during the asymptotic, fixed point regime. In that case, our analytical approximations
424
Table 1
Choices of coefficients for the tree-level corrections with c1 = 1 and the remaining coefficient
satisfying c3 = (c1 + c4 + c2 /2), according to Eq. (7). We have compared the predicted slope
determined by the fixed point, according to Eq. (40), with the measured slope well inside the string
branch of the spectrum, obtained by numerical integration of Eq. (26), including all corrections.
Coefficients at O( )
Case (1)
Case (2)
Case (3)
c2
c4
0
40
0
1
1
15
H
|
Predicted slope 3 |/
Measured slope (numerical)
2.44
2.28
2.87
2.45
2.3
2.85
for the slope of the spectrum are confirmed by the numerical simulations. In Table 1,
we present the predicted and measured slopes well inside the string phase, for the three
different choices of coefficients illustrated in Fig. 1. The agreement is striking.
It is well known [45] that the spectrum of relic gravitons, in the context of the pre-Big
Bang scenario, easily evades constraints arising from both the CMB anisotropy at COBE
scale [56] and pulsar timing data [57], because of its rapid growth at low frequency. Hence,
the peak amplitude of the spectrum is only constrained by the primordial nucleosynthesis
bound [58], as well as by primordial black hole production [59].
The relic graviton background is in general compatible with these constraints if it is normalised so as to match the string scale at the high-frequency end point of the spectrum. In
that case one finds [22] that the high frequency peak is typically of order gw () 106 ,
for a maximal amplified frequency 1011 Hz. Saturating this high-frequency end point,
and using the prediction Eq. (41) for the slope of the string branch of the spectrum, the
energy density of relic gravitons from a pre-Big Bang phase could be at most of O(1015 )
at 102 Hz, regardless of the duration of the string phase with constant Hubble parameter, i.e., far below the sensitivity of the second (planned) generation of interferometric
gravity wave detectors [54]. It seems thus difficult to detect the relic gravitons from a high
curvature string phase, associated to a fixed point of the truncated effective action at
least within the model of background considered in this paper unless the exit phase is
able to affect a sufficiently wide band of the high frequency spectrum.
Finally, the numerical integration of Eq. (26) enables us to follow carefully the time
evolution of modes, from the DDI epoch up to the FRW-radiation dominated era. We have
explicitly checked that the amplitude of tensor perturbations is rapidly frozen in on superhorizon scale, regardless of the particular higher-order dynamics of the background during
the string phase and the exit epoch.
5. Conclusion
In this paper we have discussed the evolution of tensor perturbations in a class of nonsingular cosmological models based on the gravi-dilaton string effective action, expanded
425
up to first order in , and including one-loop and two-loop corrections in the string
coupling parameter.
The coefficients of the loop expansion have been chosen in such a way as to avoid
the curvature singularities in the background, and the quantum instability of metric
fluctuations. In addition, we have included an appropriate radiation backreaction to
stabilise the dilaton in the final, post-Big Bang regime. We have been able to obtain, in this
way, a class of regular cosmological solutions in which the initial Universe, emerging from
the string perturbative vacuum, is first trapped in a fixed point of high, constant curvature,
and then spontaneously undergoes a smooth transition to the standard FRW regime. The
fixed point is determined by the corrections, while the instability and the decay of the
high curvature regime is triggered by the quantum loop corrections.
We have determined, in this class of backgrounds, the linearised equation governing the
evolution of tensor metric fluctuations, taking into account all the and loop corrections
contributing to the background regularisation. We have discussed, on this ground, the
amplification of tensor perturbations normalised to the quantum fluctuations of the vacuum,
and we have estimated their final spectral energy distribution through analytical and
numerical methods.
Our results confirm previous expectations, that the low frequency modes, crossing the
horizon in the low-curvature regime, are unaffected by higher-order corrections. Also,
we have explicitly checked (through numerical simulations) that the so-called string
branch of the spectrum, associated with the asymptotic, fixed point regime, can be reliably
estimated by means of the low-energy perturbation equation, that its slope is determined
and that it is flatter than at low frequency.
by the asymptotic constant values of H and ,
However, when the background regularisation crosses a fixed point, and when the fixed
point is determined by the expansion truncated to first order (like in the class of models
of this paper), it turns out that the slope of the string branch is at least quadratic, and
thus probably too steep to be compatible with future observations by planned advanced
detectors. In the transition regime dominated by the quantum loops, on the other hand, the
influence of the higher-order corrections on tensor perturbations is more dramatic, and the
corresponding slope may be much flatter than the one of the preceding string phase.
In summary, the results of this paper agree with the general expectation that the shape
of the spectrum of the relic graviton background, obtained in the context of the pre-Big
Bang scenario, is strongly model-dependent. However, it turns out that the formulation of
a complete and consistent regular scenario, associated to a visible graviton spectrum,
may be more difficult than previously expected at least within a truncated perturbative
approach, like the one adopted in this paper.
Acknowledgements
C.C. is supported by the Swiss NSF, grant No. 83EU-054774 and ORS/1999041014.
E.J.C. was partially supported by PPARC. We thank R. Durrer, M. Giovannini, S. Leach,
L. Mendes, C. Ungarelli, G. Veneziano, F. Vernizzi and D. Wands for helpful and
stimulating discussions.
426
References
[1] G. Veneziano, Scale factor duality for classical and quantum strings, Phys. Lett. B 265 (1991)
287294, CERN-TH. 6077/91.
[2] M. Gasperini, G. Veneziano, Pre-Big Bang in string cosmology, Astropart. Phys. 1 (1993)
317339, hep-th/9211021, an updated collection of papers and references on the pre-Big Bang
scenario is available at the URL, http://www.to.infn.it/gasperin/.
[3] R. Brustein, R. Madden, A model of graceful exit in string cosmology, Phys. Rev. D 57 (1998)
712724, hep-th/9708046.
[4] S. Foffa, M. Maggiore, R. Sturani, Loop corrections and graceful exit in string cosmology,
Nucl. Phys. B 552 (1999) 395, hep-th/9903008.
[5] R. Brustein, R. Madden, Classical corrections in string cosmology, JHEP 07 (1999) 006, hepth/9901044.
[6] C. Cartier, E.J. Copeland, R. Madden, The graceful exit in string cosmology, JHEP 01 (2000)
035, hep-th/9910169.
[7] M. Gasperini, G. Veneziano, Dilaton production in string cosmology, Phys. Rev. D 50 (1994)
25192540, gr-qc/9403031.
[8] L.P. Grishchuk, Amplification of gravitational waves in an isotropic universe, Sov. Phys.
JETP 40 (1975) 409415.
[9] A.A. Starobinsky, Spectrum of relict gravitational radiation and the early state of the universe,
JETP Lett. 30 (1979) 682685.
[10] L.P. Grishchuk, M. Solokhin, Spectra of relic gravitons and the early history of the Hubble
parameter, Phys. Rev. D 43 (1991) 25662571.
[11] J.E. Lidsey, D. Wands, E.J. Copeland, Superstring cosmology, Phys. Rep. 337 (2000) 343, hepth/990906.
[12] J. Garriga, E. Verdaguer, Particle creation due to cosmological contraction of extra dimensions,
Phys. Rev. D 39 (1989) 1072.
[13] M. Demianski, Dynamics of multidimensional KaluzaKlein cosmological models, in: Proc.
General relativity and gravitational physics, Capri, 1990, pp. 1941.
[14] M. Gasperini, M. Giovannini, Gravity waves from primordial dimensional reduction, Class.
Quant. Grav. 9 (1992) L137.
[15] M. Gasperini, M. Giovannini, Dilaton contributions to the cosmic gravitational wave background, Phys. Rev. D 47 (1993) 15191528, gr-qc/9211021.
[16] J.D. Barrow, J.P. Mimoso, M.R. de Garcia Maia, Amplification of gravitational waves in
scalar tensor theories of gravity, Phys. Rev. D 48 (1993) 36303640.
[17] M. Gasperini, M. Giovannini, Constraints on inflation at the Planck scale from the relic graviton
spectrum, Phys. Lett. B 282 (1992) 3643.
[18] R. Brustein, M. Gasperini, M. Giovannini, G. Veneziano, Relic gravitational waves from string
cosmology, Phys. Lett. B 361 (1995) 4551, hep-th/9507017.
[19] R.B. Abbott, B. Bednarz, S.D. Ellis, Cosmological perturbations in KaluzaKlein models:
Appendix, Phys. Rev. D 33 (1986) 2147.
[20] M. Gasperini, G. Veneziano, Inflation, deflation, and frame independence in string cosmology,
Mod. Phys. Lett. A 8 (1993) 37013714, hep-th/9309023.
[21] R. Brustein, M. Gasperini, M. Giovannini, V.F. Mukhanov, G. Veneziano, Metric perturbations
in dilaton driven inflation, Phys. Rev. D 51 (1995) 67446756, hep-th/9501066.
[22] R. Brustein, M. Gasperini, G. Veneziano, Peak and endpoint of the relic graviton background
in string cosmology, Phys. Rev. D 55 (1997) 38823885, hep-th/9604084.
[23] D. Polarski, A.A. Starobinsky, Semiclassicality and decoherence of cosmological perturbations,
Class. Quant. Grav. 13 (1996) 377392, gr-qc/9504030.
[24] K.S. Thorne, Gravitational radiation: a new window onto the universe, gr-qc/9704042.
427
[25] B. Allen, J.D. Romano, Detecting a stochastic background of gravitational radiation: signal
processing strategies and sensitivities, Phys. Rev. D 59 (1999) 102001, gr-qc/9710117.
[26] M. Gasperini, J. Maharana, G. Veneziano, From trivial to nontrivial conformal string
backgrounds via 0(d,d) transformations, Phys. Lett. B 272 (1991) 277284.
[27] M. Gasperini, Looking back in time beyond the Big Bang, Mod. Phys. Lett. A 14 (1999) 1059,
gr-qc/9905062.
[28] M. Gasperini, J. Maharana, G. Veneziano, Boosting away singularities from conformal string
backgrounds, Phys. Lett. B 296 (1992) 5157, hep-th/9209052.
[29] M. Giovannini, Regular cosmological examples of tree-level dilaton-driven models, Phys. Rev.
D 57 (1998) 72237234, hep-th/9712122.
[30] R. Brustein, G. Veneziano, The graceful exit problem in string cosmology, Phys. Lett. B 329
(1994) 429434, hep-th/9403060.
[31] I. Antoniadis, J. Rizos, K. Tamvakis, Singularity free cosmological solutions of the
superstring effective action, Nucl. Phys. B 415 (1994) 497514, hep-th/9305025.
[32] S.J. Rey, Back reaction and graceful exit in string inflationary cosmology, Phys. Rev. Lett. 77
(1996) 19291932, hep-th/9605176.
[33] M. Gasperini, M. Maggiore, G. Veneziano, Towards a nonsingular pre-Big Bang cosmology,
Nucl. Phys. B 494 (1997) 315330, hep-th/9611039.
[34] M. Gasperini, Tensor perturbations in high curvature string backgrounds, Phys. Rev. D 56
(1997) 48154823, gr-qc/9704045.
[35] A. Feinstein, K.E. Kunze, M.A. Vazquez-Mozo, Initial conditions and the structure of the
singularity in pre-Big Bang cosmology, Class. Quant. Grav. 17 (2000) 35993616, hepth/0002070.
[36] A. Buonanno, T. Damour, G. Veneziano, Pre-Big Bang bubbles from the gravitational instability
of generic string vacua, Nucl. Phys. B 543 (1999) 275, hep-th/9806230.
[37] R.R. Metsaev, A.A. Tseytlin, Order (two-loop) equivalence of the string equations of
motion and the sigma model Weyl invariance conditions: dependence on the dilaton and the
antisymmetric tensor, Nucl. Phys. B 293 (1987) 385.
[38] N. Kaloper, K.A. Olive, Dilatons in string cosmology, Astropart. Phys. 1 (1993) 185194.
[39] A.A. Starobinsky, Evolution of small excitation of isotropic cosmological models with one loop
quantum gravitation corrections, Zh. Eksp. Teor. Fiz. 34 (1981) 460, (In Russian).
[40] V.F. Mukhanov, L.A. Kofman, D.Y. Pogosyan, Cosmological perturbations in the inflationary
universe, Phys. Lett. B 193 (1987) 427432.
[41] L. Amendola, M. Litterio, F. Occhionero, Very large angular scales and very high-energy
physics, Phys. Lett. B 231 (1989) 4348.
[42] H. Noh, J.C. Hwang, Cosmological gravitational wave in a gravity with quadratic order
curvature couplings, Phys. Rev. D 55 (1997) 52225225, gr-qc/9610059.
[43] A. Buonanno, M. Maggiore, C. Ungarelli, Spectrum of relic gravitational waves in string
cosmology, Phys. Rev. D 55 (1997) 33303336, gr-qc/9605072.
[44] A. Buonanno, K.A. Meissner, C. Ungarelli, G. Veneziano, Quantum inhomogeneities in string
cosmology, JHEP 01 (1998) 004, hep-th/9710188.
[45] M. Gasperini, Elementary introduction to pre-Big Bang cosmology to the relic graviton
background, in: I. Ciufolini et al. (Eds.), Proc. of the Second SIGRAV School on Gravitational
Waves, Centre A. Volta, Como, April 1999, IOP, Bristol, 2001, p. 280, hep-th/9907067.
[46] V.F. Mukhanov, H.A. Feldman, R.H. Brandenberger, Theory of cosmological perturbations,
Phys. Rept. 215 (1992) 203333.
[47] R. Brustein, M. Gasperini, G. Veneziano, Duality in cosmological perturbation theory, Phys.
Lett. B 431 (1998) 277285, hep-th/9803018.
[48] R. Brustein, R. Madden, Graceful exit and energy conditions in string cosmology, Phys. Lett.
B 410 (1997) 110, hep-th/9702043.
428
[49] S. Kawai, M. Sakagami, J. Soda, Perturbative analysis of non-singular cosmological model,

gr-qc/9901065.
[50] S. Kawai, J. Soda, Evolution of fluctuations during graceful exit in string cosmology, Phys.
Lett. B 460 (1999) 41, gr-qc/9903017.
[51] B. Allen, R. Brustein, Detecting relic gravitational radiation from string cosmology with LIGO,
Phys. Rev. D 55 (1997) 32603264, gr-qc/9609013.
[52] D. Babusci, M. Giovannini, Sensitivity of a VIRGO pair to stochastic gravitational waves
backgrounds, Class. Quant. Grav. 17 (2000) 26212633, gr-qc/0008041.
[53] C. Ungarelli, A. Vecchio, Are pre-big-bang models falsifiable by gravitational wave experiments?, gr-qc/9911104.
[54] M. Maggiore, Gravitational wave experiments and early universe cosmology, Phys. Rep. 331
(2000) 283, gr-qc/9909001.
[55] M. Gasperini, M. Giovannini, G. Veneziano, Electromagnetic origin of the cosmic microwave
backgrounds anisotropy in string cosmology, Phys. Rev. D 52 (1995) 66516655, astroph/9505041.
[56] C.L. Bennett et al., Astrophys. J. 430 (1994) 423.
[57] V.M. Kaspi, J. Taylor, M. Ryba, Evolution of fluctuations during graceful exit in string
cosmology, Astrophys. J. 428 (1994) 1519.
[58] T.P. Walker, G. Steigman, D.N. Schramm, K.A. Olive, H.S. Kang, Primordial nucleosynthesis
redux, Astrophys. J. 376 (1991) 5169.
[59] E.J. Copeland, A.R. Liddle, J. E Lidsey, D. Wands, Black holes and gravitational waves in string
cosmology, Phys. Rev. D 58 (1998) 063508, gr-qc/9803070.

Erratum
Erratum to: k-Factorization and small-x

anomalous dimensions
[Nucl. Phys. B 496 (1997) 305]
G. Camici, M. Ciafaloni
Dipartimento di Fisica, Universit di Firenze, and INFN, Sezione di Firenze, Italy
We apologize to the reader for several typographical errors present in this paper,
especially in Section 3 and Appendix A, which deal with the calculation of the moments
production squared matrix element. After correction, the results agree
of the off-shell QQ
with those of Ball and Ellis [1], who have recently redone the calculation by the same
method.
Eq. (3.10), which organizes the squared matrix element as a sum of three terms should
carry a () sign in front of Eq. (3.10a) and of the first term in Eq. (3.10b). It should read:

1
(B + C)2
,
D1 = (2CF CA )
(3.10a)
k2 k 2
(M 2 t )(M 2 u)

1
1
B2 + C2
1
D2 = CA
x
)
(1
1
2
s M 2 t M 2 u
k2 k 2

2(B C) 2
2

(3.10b)
k (1 x2 ) + k (1 x1 ) + k k ,
+ 2 2
k k s

2 (k2 (1 x2 ) + k 2 (1 x1 ) + k k )2
2
2 2
.
D3 = CA
(3.10c)
s k k
s2
Eq. (3.14b) should carry a factor of 1/2 in front of the first term in round brackets and
an overall factor of (1 1 2 ). It should read:
H (1) = (2CF CA )H ab (1 , 2 ),
s
H (2) = CA B(1 , 1 1 )B(2 , 1 2 )(1 1 2 )

(3 1 2 )
(2 1 2 )
2(4 21 22) (6 21 22 )
PII of original article: S0550-3213(97)00261-7.

E-mail address: ciafaloni@fi.infn.it (M. Ciafaloni).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 2 5 9 - 0
(3.14a)
(3.14b)
432
G. Camici, M. Ciafaloni / Nuclear Physics B 607 (2001) 431432
H (3) =
B(1 , 1 1 2 )B(2 , 1 1 2 )B(2 1 2 , 2 1 2 )

s
CA
2(1 1 2 )

1
.
1
(3.14c)
5 21 22
The quantity H ab defined in Eq. (2.25) and quoted in Eq. (3.14a) should carry a ()
sign in front of the second term. It should read:
s
H ab (1 , 2 ) = (1 1 2 )(1 )(2 )

B(1 1 , 1 1 )B(1 2 , 1 2 )
4

B(1 1 , 2 1 )B(1 2 , 2 2 )
1 + (1 1 )(1 2 ) .
(3 21 )(3 22 )
(2.25)
While the expression of the q q contribution to the kernel eigenvalue function in
Eq. (4.6) is correct, and consistent with the formulas corrected above, its k-space
expression in Eq. (4.4) should be corrected for typos in Eq. (3.17), in which a factor
of 2 is misplaced in the third term in square brackets, an overall factor of 1/2 is missing in
front of the square brackets, and CA should be replaced by Nf in the last term. It should
read:
q q
Kreg
1
= 2
k
1
2 +i
2
d q q
k
reg ( ) 2
2i
k
1
2 i

1
1 1
2CF CA Nf s
1 +2
+1
log
=
CA

128k2>

1
1
+ 2L2 () + log L1 () 22
Nf s
k2
1
log
.
6 k2 k 2
k 2
References
[1] R.D. Ball, R.K. Ellis, hep-ph/0101199.
(3.17)

Erratum
Erratum to: Higher twist distribution

amplitudes of the nucleon in QCD
[Nucl. Phys. B 589 (2000) 381]
V. Braun, R.J. Fries, N. Mahnke, E. Stein
Due to a mistake in a M ATHEMATICA program, unfortunately three expansion
parameters in Eqs. (3.19) and (3.20) are incorrectly quoted. The correct coefficients read

5
5+ = 1 1 + 2f1d + 2f1u +fN 5 + 2Au1 2V1d ,
6

5
+
5 = 2 2 9f2d ,
36

1
6+ = 1 1 2f1d +fN 4V1d 1 .
(1)
2
Accordingly the corresponding numerical values in Table 3 are changed to
5+ = 0.99 102 GeV2 ,
5+ = 0.46 102 GeV2 ,
6+ = 0.25 102 GeV2 .
PII of original article: S0550-3213(00)00516-2.

E-mail address: eckart.stein@physik.uni-regensburg.de (E. Stein).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 2 5 4 - 1
(2)
Nuclear Physics B 607 [FS] (2001) 437455

Form factors of soliton-creating operators

in the sine-Gordon model
Sergei Lukyanov a,b , Alexander Zamolodchikov a,b
a Department of Physics and Astronomy, Rutgers University, Piscataway, NJ 08855-0849, USA
b L.D. Landau Institute for Theoretical Physics, Chernogolovka 142432, Russia
Received 23 May 2001; accepted 29 May 2001
Abstract
We propose explicit expressions for the form factors, including their normalization constants, of
topologically charged (or soliton-creating) operators in the sine-Gordon model. The normalization
constants, which constitute the main content of our proposal, allow one to find exact relations
between the short- and long-distance asymptotics of the correlation functions. We make predictions
concerning asymptotics of fermion correlation functions in the massive Thirring model, SU(2)
Thirring model with anisotropy, and in the half-filled Hubbard chain. 2001 Elsevier Science B.V.
All rights reserved.
PACS: 11.10.Kk
Keywords: Integrable quantum field theory; Sine-Gordon model; Form factor; Correlation function
1. Introduction
Exact form factors are of significant interest in integrable quantum field theories in two
spacetime dimensions, and much progress has been made during the last 25 years in
computing these quantities in various models, and in their applications to the analysis
of correlation functions. By n-particle form factors we understand as usual the matrix
elements of any local (or semi-local, see below) field operator O(x) between the vacuum
and the states containing n particles. One of the reasons for the interest lies in the fact
that exact form factors allow one to generate large-distance expansions for the correlation
functions by inserting a complete set of states of asymptotic particles. If the factorizable
S-matrix of these particles is known, the form factors are determined by solving the form
factor bootstrap equations. (See [1] for coherent exposition of the bootstrap program.
Ref. [2] includes some early works on the subject.)
E-mail address: sergei@physics.rutgers.edu (S. Lukyanov).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 2 6 2 - 0
438
S. Lukyanov, Al. Zamolodchikov / Nuclear Physics B 607 [FS] (2001) 437455
It is important to note that these form factor bootstrap equations constitute a linear
system, i.e., if a certain collection of form factors solves the equations, the renormalized
collection (with all form factors associated with the same operator O multiplied by the
same constant) also does; this of course represents the freedom in normalization of the
field operator O. On the other hand it is usually convenient to fix the normalizations of
the field operators in terms of the short-distance behavior of their correlation functions.
If the short-distance behaviour is controlled by associated CFT, the two-point correlation
function of a spinless field O(x) has the asymptotic form

O(x)O (0)
1
.
|x|4O
(1.1)
Conventional choice of the coefficient 1 in the power law (1.1) fixes (up to a phase, which
usually can be fixed through other correlation functions) the normalization of O. The
problem arises of finding the specific normalization of the form factors which corresponds
to the CFT normalization (1.1).
In principle, this problem can be solved by analyzing the short-distance behavior of the
form factor expansion. However, in practice it is usually possible to compute only the first
few terms of this expansion. While yielding in many cases excellent numerical data even
for rather small distances (see, e.g., [3]), this truncated series does not provide exact analytic information about the coefficient in the short-distance asymptotic (1.1). Therefore
the problem of determining the CFT normalizations of the form factors for a generic
operator O remains largely open (notable exceptions being the cases when O is a component of some conserved current, e.g., energy-momentum tensor, and the normalizations
of its form factors can be fixed through the Ward identities, see, e.g., [1]). Some progress
was made in [4] where the expectation values of topologically neutral primary operators
in the sine-Gordon model (and in some other integrable QFT [57]) were determined. This
result makes it straightforward to fix the normalizations of all higher form factors of these
operators using Smirnovs annihilation pole relation [1] of the form factor bootstrap [8].
In this paper we extend the above result to a class of topologically charged fields in
the sine-Gordon model. Obviously, the vacuum expectation values of fields with nonzero
topological charge vanish. The simplest form factor of an operator carrying topological
charge n is its matrix element between the vacuum and an n-soliton (or (n)-antisoliton,
if n is negative) state. In Section 3 we propose an explicit expression for these simplest
form factors of CFT-normalized primary operators. Our proposal, Eq. (3.1) below,
actually concerns the corresponding normalization factor Zn in (2.12) (without regard
of normalization, some form factors with n = 0 were considered before in Ref. [9]). In
Section 4 we carry out various checks of this conjectured expression.
Let us mention here that there is a one-parameter family of primary fields of given
topological charge n (see Section 2 for details), and for generic values of the parameter a
these fields (denoted Oan henceforth) are not local (they are semi-local in the terminology
of [10]). Nonetheless, there are several important reasons to be interested in form factors
and correlation functions of such nonlocal operators. Let us mention a few. First, specific
discrete values of the parameter a provide a set of local topologically charged fields which
439
in fact coincide with the fundamental Fermi fields and their composites in the massive
Thirring model. For other discrete values of this parameter these fields become components
of conserved nonlocal currents generating the affine quantum group symmetry of the sineGordon model. These relations will be used in Section 4. Second, analytic properties in
the parameter a are likely to be illuminating, as was demonstrated in [5] for expectation
values of neutral fields. We will say some words on this subject in Section 6. Finally,
in many cases it has proved to be useful to factorize local fields into nonlocal ones. For
example, the SU(2) Thirring model (and its anisotropic deformation) can be bosonized in
terms of a sine-Gordon field plus a free massless boson, the correlation functions of basic
Fermi fields in this model being expressed through the correlators of the nonlocal fields of
the sine-Gordon theory. In Section 5 we use this relation to make predictions about exact
asymptotics of fermion correlation functions in the SU(2) Thirring and Hubbard models.
2. Topologically charged fields in the sine-Gordon model

The sine-Gordon model, which is described by the Euclidean action 1

1
2
2
( ) 2 cos() ,
AsG = d x
16
(2.1)
is invariant w.r.t. the field translations (x) (x)2/. As a well known consequence,
the model admits a class of fields creating nonzero topological (or soliton) charge. The
simplest of these fields can be described as the dual exponentials
n
O0n (x) = e 4
Cx
dx
(2.2)
with integer n. Here the integration goes along some contour Cx (Dirac string) starting at
infinity and ending at x, the precise shape of this contour being very much arbitrary as the
correlation functions involving these fields depend only on its homotopy properties. The
field (2.2) creates a discontinuity of the field along Cx , so that the values of right across
this contour differ by a constant equal 2n/. In operator formalism, the field operator
associated with (2.2) creates the topological charge n. The fields (2.2) are mutually local
and they have zero spin. Since the exponential in the r.h.s. of (2.2) requires regularization
this expression defines the field O0n only up to normalization. We fix this ambiguity by
assuming the CFT normalization (1.1), specifically
n2

lim |x| 8 2 O0n (x)O0n (0) = 1.
|x|0
(2.3)
Although the fields (2.2) are themselves local, they obviously are not local with respect
to the field (x). Nevertheless one can define a class of useful nonlocal fields Oan (x)
by fusing the fields (2.2) with the exponentials of , i.e., by taking the limits
limx x O0n (x) exp(ia(x )). In view of the above nonlocality this limit has a phase
1 Here and below we use the notations and conventions adopted in [4]; in particular, the language of Euclidean
field theory is used by default.
440
ambiguity, which can be fixed, for instance, by specifying the direction, relative to the
contour Cx , from which x approaches x. To eliminate all ambiguities one can choose
some Cartesian coordinates x = (x, y), and define
Oan (x) =

x

n
lim exp
y (x, y) dx exp ia(x + , y) .
+0
4
(2.4)
In fact, rigid specification of the contour Cx in (2.4) results in inconveniently discontinuous

correlation functions involving these objects. To avoid this inessential complication, in
what follows by a correlation function
n

OaNN (xN , yN ) Oan11 (x1 , y1 ) ,
(2.5)
we always understand the analytic continuation of this correlator from the domain
y1 < y2 < < yN . With this definition the correlation function (2.5) becomes a
multivalued function of the coordinates x1 , . . . , xN which acquires a certain phase factor
exp(i(aj nk + ak nj )/) when the point xj is brought around xk counterclockwise. In
particular, the two-point correlation function can be written as

Oan (x)Oan
(0) =
s(a ,n)s(a,n)

2
s(a,n)+s(a ,n) (n)
i z
2
e
ei
Ga,a (r),
z
(2.6)
where
an
(2.7)
,
z=
x + iy, z = x iy, and the function Ga,a is real and depends only on the distance
r = zz. In the terminology of [10] the fields Oan are semi-local. If a = 0 the field Oan
carries nonzero spin (2.7). It becomes a local, bosonic or fermionic, field when s(a, n) takes
integer or half-integer values. Note that the definition (2.4) fixes also the normalization
of the field Oan , since the normalizations of O0n (x) and Oa0 (x) eia are already fixed
through (2.3) and (1.1), respectively. The same normalization can be specified by the
condition

1
n2
(n)
Ga,a (r)r0 2d(a,n) , with d(a, n) = 2a 2 + 2 .
(2.8)
8
r
s(a, n) =
The aim of this work is to describe the form factors of the operators Oan with n-soliton
states. We will assume the Hamiltonian picture with coordinate y taken as Euclidean time.
Note that in this picture the operators defined by (2.4) obey the simple Hermiticity relation
n
n
,
Oa = Oa
(2.9)
which allows us to restrict our attention to the form factors of Oan with generic a but n 0.
We will use the notation A and A+ for the soliton and antisoliton; these particles carry
negative and positive units of the topological charge,
H=
2

x (x, y) dx,
(2.10)
441
respectively. Conservation of the topological charge implies that nonvanishing form factors
of the operator Oan are of the form

vac|Oan (0)A (1 ) A (n+N )A+ 1 A+ N ,
(2.11)
where i and j denote rapidities of solitons and antisolitons (for simplicity, here and below
we ignore the possible presence of breathers). Up to overall normalization, all these form
factors can be written down in closed form, as certain N -fold integrals [8]. In the simplest
case, N = 0, an explicit formula exists,
n
n

i na a
vac|Oan (0)A (1 ) A (n ) in = Zn (a) e 2

e m
G(m j ).
m=1
m<j
(2.12)
Here

G( ) = iC1 sinh(/2) exp

dt sinh2 t (1 i/) sinh(t ( 1))
,
t
sinh(2t) sinh(t ) cosh(t)
(2.13)
with

dt sinh2 (t/2) sinh(t ( 1))
C1 = exp
,
t sinh(2t) sinh(t ) cosh(t)
(2.14)
and we have assumed that the rapidities are arranged so that 1 < 2 < < n . Here and
below we use the notation
=
2
.
1 2
(2.15)
The only unknown component in (2.12) is the real normalization constant Zn (a). This
constant controls the long-distance asymptotics of the two-point function (2.6) (with the
normalization already fixed by (2.8)) because it is dominated by n-soliton intermediate
states, namely
(n)
Ga,a (r) =
Zn (a)Zn (a )
n!
+

n

(aa )
Mr cosh m dm
G(m j )2
+ ,
e m
2
m<j
(2.16)
m=1
where M is the soliton mass, and the dots stand for the subleading terms, which are of the
order of e(n+2)Mr . Let us also stress here that once the constant Zn (a) in (2.12) is fixed,
the normalizations of all higher form factors (2.11) are also fixed by the annihilation pole
condition of the form factor bootstrap (see [1] for details). For example, using the approach
of [8], one can derive the following expression for the (n + 2)-particle form factor (2.11),
442

vac|Oan (0)A (1 ) A+ (k ) A (n+2 ) in
n+2
i C2 Zn (a) i2an a m k
=
e
e
e
G(m j )
4C1
m<j
m=1

k
n+2
i

d ( 2a + 1 )
e
W (p )
W ( p )
e 2 2
2
C+
i
2 2
p=1
p=k+1

n+2
k1

d ( 2a + 1 )
W (p )
W ( p ) .
e
2
p=1
(2.17)
p=k
Here the function W and the constant C2 are

dt sinh2 t (1 i/) sinh(t ( 1))
2
exp 2
,
W ( ) =
cosh( )
t
sinh(2t) sinh(t )
0

dt sinh2 (t/2) sinh(t ( 1))
C2 = exp 4
.
t
sinh(2t) sinh(t )
(2.18)
The integration contours C+ and C in (2.17) are described as follows. The contour C+
starts from along the real axis of the complex plane, and winds around the poles
of its integrand located in the strip 2 0 < m < 2 + 0, going first above the poles
i
at p + i
2 , p = 1, . . . , k, and then below the poles at p 2 , p = k + 1, . . . , n + 2, and
finally extends to + along the real axis, as illustrated in Fig. 1a. Similarly, the contour
i
C goes above p + i
2 , p = 1, . . . , k 1, and below p 2 , p = k, . . . , n + 2, of its
integrand, see Fig. 1b.
Under the assumed normalization of the fields Oan , the leading short-distance asymptotic
of the two-point correlation function (2.6) is

(n)
Ga,a (r)r0
ei(a+a )

r d(a,n)+d(a ,n)d(a+a ,0)
with the coefficient Ga+a ei(a+a ) which can be read off from [4]. To determine
the coefficient in its leading large-distance asymptotic (2.16) one needs to know the
normalization constant Zn (a) in (2.12).
(a)
(b)
Fig. 1.
443
3. Conjecture
We propose the following explicit formula for the constant Zn (a) in (2.12):

2

M'( 32 + 2 ) d(a,n)
C2 n/4 C2 n /8
Zn (a) =
16
2C12
'( 2 )

dt
cosh(4 at/) e(1+ )nt 1
exp
t 4 sinh( t) sinh((1 + )t) cosh(t)
0

n
2t
d(a, n)e
+
.
4 sinh(t )
(3.1)
Here the constants C1 and C2 are given by (2.14) and (2.18), and d(a, n) is as in (2.8). We
would like to stress here that this formula should be used only with n 0. For example,
n
the
form factor vac|Oa (0)|A+ (1 ) A+ (n )in should be written with Zn (a) (not
Zn (a) ) for its normalization constant.
In the next section some calculations supporting the conjecture (3.1) are presented (and
some are just mentioned).
4. Supporting calculations
4.1. The case n = 0
In this case the field (2.4) reduces to the exponential field exp(ia(x)), and the form
factor (2.12) becomes its vacuumvacuum matrix element. Correspondingly, for n = 0
(3.1) reduces to the expectation value of this exponential field (see [4]).
4.2. Free fermion point
At 2 = 1/2 the sine-Gordon model reduces to free Dirac fermions. In this case the
form factors (2.12), and in particular the normalization factors Zn (a), can be calculated
directly [11]. For this value of 2 Eq. (3.1) coincides with the result of [11].
4.3. Massive Thirring model perturbation theory
As is well known [12,13] the sine-Gordon theory is equivalent to the massive Thirring
model,

g
2

AMTM = d x + M + J J ,
(4.1)
2
is its (nonanomalous) vector current, and the
where is a Dirac Fermi field, J =
coupling constant g is related to in (2.1) by
1
g
= 2 1.
(4.2)
444
The action (4.1) requires field- and mass-term renormalizations, and the precise relation
are
between M and in (2.1) depends on the renormalization scheme. The fields ,
1
related to O/2 from Section 2 as follows [13]
1
O/2
1
(x) =
,
1
2Z O/2
1
1
1
(x) =
O/2 , O/2
,
2Z
(4.3)
where we have chosen a chiral representation of the (Euclidean) Dirac matrices, i.e., x =
2 , y = 1 . The constant Z depends on the normalization condition for the field
. In what follows we assume the most common renormalization scheme, in which the
renormalized M coincides with the soliton mass M, and the normalization of is fixed
as follows,
(i/
p + M)S(p) 1,
as p2 + M 2 0.
Here S(p) is the momentum-space propagator,

d 2p
(0) = S(p)eipx
(x)
,
(2)2
and p
/ stands for p . Comparing this equation with the n = 1 and a = a = /2 case
of (2.16), one finds that this scheme corresponds to the choice Z = Z1 (/2) in (4.3).
1
1
O/2

On the other hand, the short-distance behavior of the correlation functions O/2
is given by (2.6), (2.8). This leads to the following prediction for the constant factor in the
large-momentum asymptotic of the fermion propagator in this scheme,
S(p) i/
p
M 212d '(3/2 d ) 2 d 32
,
p
Z
'(1/2 + d )
as p2 .
(4.4)
Here
d =
1
2
,
+
8 2
2
and

Z =
C2
2C12
1/2
C2
16
1/4

M'( 32 + 2 ) 2d
'( 2 )

cosh(2 t)e(1+ )t 1
dt
exp
t 2 sinh( t) sinh((1 + )t) cosh(t)
0

1
2t
2d e
.
+
2 sinh(t )
(4.5)
This prediction can be verified in standard renormalized perturbation theory in (4.1).

Straightforward evaluation of the diagram in Fig. 2 yields

2

g
p+M +
2 (p) + O g 3 ,
S 1 (p) = i/
2
445
Fig. 2.
where
2

p
2
6 + log
(4.6)
, as p2 .
3
M2
Using a power expansion of (4.5) in the parameter (4.2),

2

g
2
Z = M 1 +
(4.7)
log M 2 + 2E + 6 2 log 2
+ O g3 ,
2
3
where E = 0.577216 . . ., is Eulers constant, one can check that (4.4) agrees with (4.6) to
order g 2 .
2 (p) i/
p
4.4. Nonlocal integrals of motion

As was mentioned in the introduction, the fields Oa2 with appropriately chosen values
of a coincide with the components of the nonlocal currents found in [14]. Namely, the
fields
2
J (x) = O(2)
1 (x),
2
H (x) = O
(x),
2
J (x) = O(2)
1 (x),
2
(x) = O
H
(x),
(4.8)
with = (2)1 , satisfy the continuity equations

(x) = H ,
J
where =
2 (x iy ),
Q =
1
ZQ
= 1
Q
ZQ
,
J (x) = H
=
(4.9)
1
2 (x + iy ). They give rise to four nonlocal integrals of motion

J (x, y) + H (x, y) dx,

(x, y) dx,
J (x, y) + H
(4.10)
of level zero,
which generate (as was found in [14]) the affine quantum group Uq (sl(2))
with
2
q = ei/ .
More precisely, under a special choice of the (- and -dependent) constant ZQ , the
operators (4.10) obey the commutation relations
+ q 2 Q
+ Q =
Q Q
1 q 2H
,
1 q 2
q 2 Q
Q+ =
Q+ Q
1 q 2H
,
1 q 2
(4.11)
446
where H is the topological charge (2.10). The constant ZQ can be expressed through the
expectation value of the fields ei(1/) ,

i(1/)
M'( 32 + 2 ) 2/
1
2
ZQ = (1 + ) sin(/ ) e
= 2
,
' (1 + 1/ )
2'(1 + 2 )
(4.12)
where again M denotes the soliton mass.
The vacuum |vac is annihilated by all the generators (4.10), while the soliton
antisoliton pair A forms two-dimensional representation of the algebra (4.11),

A ( ) = 0,
Q A ( ) = Q

A ( ) = e A ( ) ,
Q
Q A ( ) = e A ( ) ,
(4.13)
and H |A ( ) = |A ( ). The action of these operators on any multisoliton state is
coproduct, i.e.,
described by a standard Uq (sl(2))

1 + q H Q
= Q
.
Q
(Q ) = Q 1 + q H Q ,
(4.14)
It is not difficult to see that consistency of the form factor formula (2.12) with the above
continuity equation (4.9) and with (4.13) requires that the coefficient Zn (a) in (2.12) satisfy
the following identities
MZQ 2
1
1
2
= ( ) Z2
=
,
Z2
2
2
2C1
which both are easily verified for (3.1).
More interesting consistency conditions come from the following observation. For all
integer n and m the fields Oann,m with
an,m =
m
n
+
,
4
2
n
defined in (4.8) (a similar set Oa
are local with respect the currents J and H
exists
n,m
for the currents J and H ). As a result, commutators (more precisely, q-commutators)
+ and Q
are expressed through the fields Oan2 (with
of Oann,m with the generators Q
the parameter a somewhat shifted) or their descendants. In particular, one can derive the
following commutation relation
+ Oan (x) q n Oan (x)Q

+ = 2i q n/2 Oan+2 (x).
Q
n,1
n,1
n2,1
ZQ
(4.15)
This operator equation of course leads to certain relations between the form factors.
For instance, sandwiching (4.15) between the vacuum and the state containing n + 2
antisolitons and using (4.14) and (4.13) to transform the matrix element appearing on the
left-hand side, one obtains
n+2

k

q kn2 e vac|Oann,1 (0)A (1 ) A+ (k ) A (n+2 ) in
k=1

2i n/2
q
vac|Oan+2
(0)A (1 ) A (n+2 ) in .
n2,1
ZQ
(4.16)
447
Now, substituting (2.17) for the form factors in the left-hand side of (4.16), one observes
that all but two integrals in the sum cancel each other, which results in the relation

2 Zn+2 (an2,1 )
Zq
Zn (an,1 )

n+2
n+2

n
d (n+2)
iC2 2k2
q 2 +1
e 2 2
=
e
W (k )
4C1
2
k=1
k=1
C+
n2 1

n+2

d (n+2)
e 2 2
W ( k ) ,
2
(4.17)
k=1
where W is the function (2.18). Thanks to the identity W ( i) = W ( i), the sum
of the integrals in the bracket {. . .} on the r.h.s. of (4.17) in fact reduces to the following
limit
+i
{. . .} = lim
+
i

n+2
n+2

d (n+2)
C2 2 1 2k2
2
2
e
W (k i) = i
e
.
2
16
n
k=1
k=1
In turn, the limit can be evaluated using the asymptotic behavior

C2 1/2 2 2
W ( i)
e
, as e +,
16
and therefore (4.15) leads to the following relation between the normalization factors
in (2.12):

2ZQ 2 C2 n
Zn+2 (an2,1 )
=
.
(4.18)
Zn (an,1 )
C1
16
It is possible to check that (3.1) indeed obeys (4.18).
4.5. Scattering in the lattice XYZ model
The sine-Gordon field theory describes the scaling limit of the integrable XY Z spin
chain [15],
1
y y
z
x
+ Jy k k+1 + Jz kz k+1
Jx kx k+1
.
2
N
HXY Z =
(4.19)
k=1
As is known, at Jy = Jx |Jz | the spin chain (4.19) is critical, i.e., the gap in the spectrum
of (4.19) vanishes, and its correlation length Rc becomes infinite in units of the lattice
spacing [16]. In the scaling limit Jx Jy 0 the spin correlations in (4.19) are described
by the field theory (2.1) with [16,17]

cos 2 = Jz /Jx ,
M
4
22 2

|Jx Jy |
8 sin2 ( 2 )Jx
0,
448
provided all relevant distances are Rc , i.e., infinitely greater than the lattice spacing .
If, however, Jx Jy is small but finite, there are corrections due to finite lattice size. These
corrections can be taken into account by adding certain irrelevant perturbations to (2.1),
i.e., by using the effective action
2
2
d x 4
O0 (x) + O04 (x)
Aeff = AsG +
2 2

(4.20)
+ T
T + T2 +
T 2 2 + ,
where we have written down explicitly only the most significant of those irrelevant
terms. Here T and
T are components of the sine-Gordon stress-energy tensor, in standard
4
notations, and O0 (x) are the fields (2.2). In (4.20) the explicit dependence on the lattice
spacing is exhibited to show the relative smallness of various terms; the omitted terms are
of higher order in . Fortunately, the dimensionless coupling constants and in (4.20)
are known exactly [18], in particular
2/

'(1 + 2 )
4'(1 + 1/ )
.
=
'(1/ )
2 '( 32 + 2 )
This makes (4.20) a working tool for determining the leading lattice corrections to the
scaling limit in the XY Z model.
Since the fields O04 carry nonzero topological charge, the -terms in (4.20) generate
processes violating conservation of the topological charge, which is now conserved only
modulo 4. In particular, two solitons can be produced in antisolitonantisoliton scattering,
with nonzero amplitude. In the Born approximation this amplitude is
++
(2 1 ) =
S
(i)2/
4M 2 sinh(2 1 )

out A+ (1 )A+ (2 )O04 (0)A (1 )A (2 ) in ,
where 1 and 2 (1 < 2 ) are rapidities describing the kinematics of the scattering. The
matrix element here can be obtained by crossing transformation from the form factor (3.1)
with n = 4 and a = 0. Simple calculation yields
++
( ) = iK sinh(/ )S( ),
S
(4.21)
where
S( ) = G( )/G( ),
is the sine-Gordon S-matrix element associated with elastic process A A A A , and
C12 C2 2/
(4.22)
Z4 (0).
16M 2
The appearance of this amplitude is of course expected. The scattering of excitations in the
lattice model (4.19) is described by Baxters elliptic S-matrix. It depends on the elliptic
nome p,
whose precise relation to the parameters in (4.19) is rather transcendental (one
K =
449
can consult [19,20] for details), but here we only need to know that when |Jx Jy | 0
this nome also goes to zero as

M 2/
p
.
4
++
If Jx Jy = 0 this elliptic S-matrix has a nonzero element S
( ), presented explicitly
in [19,20]. One can check that when p gets small this amplitude assumes precisely the
form (4.21) with
K = 4p sin(/ ).
(4.23)
This agrees with (4.22) provided

M 2 16 2 M '( 32 + 2 ) 2/
Z4 (0) = 2
.
C1 C2 ' 2 (1/ )
2'(1 + )
(4.24)
It is not difficult to verify that for given values of the parameters (3.1) indeed reduces
to (4.24).
5. Deformed SU(2)Thirring model

The normalization factor (3.1) leads to certain predictions about the asymptotic behavior
of fermion correlation functions in the so-called deformed (or anisotropic) SU(2)Thirring
model. The latter is described by the action (we use again Euclidean notations)

g0
g 3 3
2
+

ADTM = d x
(5.1)
+ J J + J J + 2g J J ,
2
2
=,
where , is a doublet of Dirac Fermi fields, and

,
J =
A ,
JA =
(5.2)
are their vector currents. The Pauli matrices A = ( 3 , + , ) in (5.2) act on the flavor
indices = , . The model (5.1) is renormalizable, and its coupling constants g , g
should be understood as running ones (the singlet coupling g0 does not renormalize).
The corresponding RG flow pattern is known as KosterlitzThouless flow (see, e.g., [21]).
In particular, in the domain g |g | all RG trajectories originate from the line g = 0
of UV stable fixed points, and (5.1) indeed defines a quantum field theory. In this domain
(which is the only one that we discuss here) each RG trajectory is uniquely characterized
by the limiting value
g = lim g (L),
L
of the running coupling g (L) at extremely short distances (L stands for the logarithm of
the length scale), i.e., the theory (5.1) depends on two dimensionless parameters g and
g0 , besides the mass scale appearing through dimensional transmutation. The model (5.1)
450
attracts much attention in condensed matter physics, e.g., because of its relation to the
scaling limit of the Hubbard model (see, e.g., [22]).
As is well known (see, e.g., [23]), the model (5.1) can be bosonized in terms of the
sine-Gordon field (x), with in (2.1) related to g by
2g
1
=1+
,
2
(5.3)
and a free massless boson (x). For the latter we assume conventional CFT normalization

(x)(0) = 2 log zz .
The chiral components of the Fermi fields

1
R (x)
(x) =
,
2 L (x)
are expressed in terms of these bosons as
1 1
O /4 ,
L = O/4
1 1
O /4 ,
R = O/4
1 1
O /4 ,
L = O/4
1
1 ,
O
R = O/4
/4
(5.4)
where = are Klein factors (2 = 2 = 1, = ). In (5.4) Oa1 are nonlocal

a1 are similar expressions
fields (2.4) of the sine-Gordon part of the bosonic theory, and O
in terms of the free field , with replaced by , where
1
2g0
.
=1+
2
These fields can also be written as

an (x) = exp i a n R (z) i a + n L z ,
O
4
4
(5.5)
through the right and left moving chiral parts of (x). Note that each of the factors O and
in (5.4) is nonlocal (they each have spin 1 ), while the products appearing there are local
O
4
Fermi fields.
As a result of the factorization (5.4) (known as spin-charge separation in condensed
matter applications) the correlation functions of the fermions also factorize, for instance
i 1 ( 1 )2 /4 (1)
r2
G/4,/4 (r),
R (x) R (0) =
(5.6)
z
where again z = x + iy, r = zz. In writing (5.6) we have used (2.6), along with the well
known form of the two-point correlation function of the free-field exponentials (5.5),

an
2

n
4a 2 4n 2
n
i z
a
(x)O
(0)
=
e
r
a a .
O
a
z
The short-distance asymptotic form of this correlation function follows from (2.8),

z
lim r 2d R (x) R (0) = i ,
r0 r
451
with
1 ( 1 )2 ( 1 )2
+
+
.
2
8
8
This of course is nothing but our convention concerning the normalization of which
we implicitly assumed in writing (5.4). On the other hand, using (2.16) with n = 1 one can
obtain its large-distance asymptotic,
d =
i Z1 (/4) ( 1 )2 /4 Mr

e
+ O e3Mr ,
R (x) R (0) =
(5.7)
r
z
2M
where M is the sine-Gordon soliton mass. While the functional form of (5.7) is well
known [24], the coefficient Z1 (/4) constitutes a nontrivial prediction which follows
from (3.1). Let us mention here that according to (5.7), in the case = = 1 (which
corresponds to the symmetric model (5.1) with g0 = 0 and g = g > 0) one has

1,
as r 0,

i
R (x) R (0)
(5.8)
Mr
z
Ce
, as r ,
where the constant C is expressed in terms of Z1 (/4) at 2 = 1. In this case the integral
in (3.1) can be evaluated explicitly, yielding
C = 2 6 e 4 A3G = 0.921862 . . . ,
5
(5.9)
where AG is the Glaisher constant. The numerical coefficients in the short- and longdistance asymptotics in (5.8) are remarkably close; this suggests that in this symmetric case
the leading term in the long-distance asymptotic (5.8) may be rather good approximation
for the correlation function at all scales.
Eq. (5.8) can be translated to asymptotic formulae for the fermion correlation functions
in the half-filled Hubbard chain
+

HHub = t
j = =,
+U
cj,
cj +1, + cj+1, cj,
+

1
1
nj,
.
nj,
2
2
(5.10)
j =
c,j , c,j
In (5.10)
are the Fermi operators,

cj, , cj , = jj ,
cj, . As is known [2529], in the symmetric case (g0 = 0 and g = g > 0),
and nj, = cj,
the field theory (5.1) describes the scaling limit of this chain. If one sends U +0, the
correlation length

t 2 t
eU ,
Rc =
2 U
diverges, and the correlation functions of (5.10) at large lattice separations assume certain
scaling forms. In particular, if |j j | 1 the equal-time fermion correlator can be written
452
as

sin( 2 (j j ))

cj , cj,

F |j j |/Rc ,

(j j )
(5.11)
where F (0) = 1. The scaling function F ( ) here is related directly to the field-theoretic
correlation function (5.6) with = = 1. Therefore Eq. (5.8) leads to the following
prediction for the large-distance asymptotic of the scaling function F in (5.11),

F ( ) = Ce + O e3 , as ,
where C is the constant (5.9).
6. Discussion
Conjecture (3.1) for the factor Zn (a) in (2.12) is the main statement of this paper. In
certain sense it extends the proposal of [4] about vacuum expectation values of Oa0 (x)
exp(ia(x)) to the case of fields Oan (x) carrying nonzero topological charge n. In this
connection we would like to add the following remark.
As was shown in [5], the vacuum expectation values

vac|eia |vac = Z0 (a),
obey the interesting analytic relation

Z0 (a) = R0 (a) Z0 (Q a),
where Q =
(6.1)
, and the amplitude
R0 (a) = ()2Q(2a+Q)
'(1 +
'(1
2a
2a
+Q
)'(1 2a Q)
,
Q
)'(1 + 2a + Q)
M'( 32 + 2 ) 1 +
,
=
2'(1 + )
(6.2)
can be related to the Liouville reflection amplitude [30] by analytic continuation

ib. Although our understanding of the reflection relation (6.1) is still far from being
satisfactory, it has proved to be very useful in determining vacuum expectation values of
other fields in the sine-Gordon [31] and in other integrable models [57]. It is possible to
check that Zn (a) in Eq. (3.1) satisfies a similar relation

Zn (a) = Rn (a) Zn (Q a),
(6.3)
with
Rn (a) = ()
2Q(2a+Q)
'(1 +
'(1
2a
2a
+Q
+
Q
+
n
)'(1 2a
2 2
n
)'(1 + 2a
2 2
Q + n/2)
+ Q + n/2)
(6.4)
453
At the moment we are not in possession of any clear interpretation of the reflection
amplitude (6.4), neither in terms of Liouville nor any other CFT, nor even as a natural
analytic interpolation of the normalization factors of [7] (note that (6.4) does not
have n n symmetry, which the Coulomb-gas integrals defining the corresponding
normalization factors clearly do, thus indicating a significant ambiguity in such analytic
interpolation for n = 0). Nonetheless, Eq. (6.3) could be helpful in the search for
generalizations of (3.1).
Acknowledgements
The authors acknowledge helpful discussions with Alexei Tsvelik. They are also
indebted to Chris Hooley for careful reading of the manuscript. S.L. heartily thanks the
organizers of the research program New Theoretical Approaches to Strongly Correlated
Systems, at the Isaac Newton Institute for Mathematical Sciences, Cambridge, where
parts of this work were done, for their kind hospitality. He has especially benefited from
discussions with Fabian Essler.
This research is supported in part by DOE grant #DE-FG02-96ER10919.
References
[1] F.A. Smirnov, Form-factors in Completely Integrable Models of Quantum Field Theory, World
Scientific, Singapore, 1992.
[2] S. Vergeles, V. Gryanik, Two-dimensional quantum field theories having exact solutions, Yad.
Fiz. 23 (1976) 13241334 (in Russian);
M. Karowski, P. Weisz, Exact form factors in (1 + 1)-dimensional field theoretic models with
solution behavior, Nucl. Phys. B 139 (1978) 455476;
A.B. Zamolodchickov, Quantum sine-Gordon model, The soliton form-factors, Preprint, ITEP45-1977, 1977, p. 12.
[3] Al.B. Zamolodchikov, Two-point correlation function in scaling LeeYang model, Nucl. Phys.
B 348 (1991) 619641;
C. Acerbi, G. Mussardo, A. Valleriani, Form-factors and correlation functions of the stress(1)
(1)
(2)
energy tensor in massive deformation of the minimal models E(N) E(N) /E(N) , Int. J. Mod.
Phys. A 11 (1996) 53275364;
J. Balog, M. Niedermaier, Off-shell dynamics of the O(3) NLS model beyond Monte Carlo and
perturbation theory, Nucl. Phys. B 500 (1997) 421461.
[4] S. Lukyanov, A. Zamolodchikov, Exact expectation values of local fields in the quantum sineGordon model, Nucl. Phys. B 493 (1997) 571587.
[5] V. Fateev, S. Lukyanov, A. Zamolodchikov, Al. Zamolodchikov, Expectation values of
boundary fields in the boundary sine-Gordon model, Phys. Lett. B 406 (1997) 8388;
V. Fateev, S. Lukyanov, A. Zamolodchikov, Al. Zamolodchikov, Expectation values of local
fields in the BulloughDodd model and integrable perturbed conformal field theories, Nucl.
Phys. B 516 (1998) 652674.
[6] P. Baseilhac, V.A. Fateev, Expectation values of local fields for a two-parameter family of
integrable models and related perturbed conformal field theories, Nucl. Phys. B 532 (1998)
567587.
454
[7] V.A. Fateev, Normalization factors in conformal field theory and their applications, Mod. Phys.
Lett. A 15 (2000) 259270.
[8] S. Lukyanov, Free field representation for massive integrable models, Commun. Math.
Phys. 167 (1) (1995) 183226;
S. Lukyanov, Form-factors of exponential fields in the sine-Gordon model, Mod. Phys. Lett.
A 12 (1997) 25432550.
[9] H. Babujian, A. Fring, M. Karowski, A. Zapletal, Exact form-factors in integrable quantum
field theories: the sine-Gordon model, Nucl. Phys. B 538 (1999) 535586;
G. Delfino, Off critical correlations in the AshkinTeller model, Phys. Lett. B 450 (1999) 196
201.
[10] V.A. Fateev, A.B. Zamolodchikov, Parafermionic currents in the two-dimensional conformal
quantum field theory and self-dual critical points in ZN invariant statistical systems, Zh. Eksp.
Teor. Fiz. 89 (1985) 380399.
[11] S. Lukyanov, A. Zamolodchikov, in preparation.
[12] S. Coleman, The quantum sine-Gordon equation as the massive Thirring model, Phys. Rev.
D 11 (1975) 20882097.
[13] S. Mandelstam, Soliton operators for the quantized sine-Gordon equation, Phys. Rev. D 11
(1975) 30263030.
[14] D. Bernard, A. LeClair, Residual quantum symmetries of the restricted sine-Gordon theories,
Nucl. Phys. B 340 (1990) 721751.
[15] R.J. Baxter, Exactly Solved Models in Statistical Mechanics, Academic Press, London, 1982.
[16] A. Luther, Eigenvalue spectrum of interacting massive fermions in one dimension, Phys. Rev.
B 14 (1976) 21532159.
[17] J.D. Johnson, S. Krinsky, B.M. McCoy, Vertical-arrow correlation length in the eight-vertex
model and the low-lying excitation of the X-Y-Z Hamiltonian, Phys. Rev. A 8 (1973) 2526
2547.
[18] S. Lukyanov, Low energy effective Hamiltonian for the XXZ spin chain, Nucl. Phys. B 522
(1998) 533549.
[19] O. Foda, K. Iohara, M. Jimbo, R. Kedem, T. Miwa, H. Yan, An elliptic quantum algebra for
SL(2), Lett. Math. Phys. 32 (1994) 259268.
[20] S. Lukyanov, Ya. Pugai, Bosonization of ZF algebras: direction toward deformed Virasoro
algebra, J. Experiment. Theoret. Phys. 82 (1996) 10211045.
[21] C. Itzykson, J.-M. Drouffe, Statistical Field Theory, Vol. 1, Cambridge Univ. Press, 1989.
[22] D. Controzzi, F.H.L. Essler, A.M. Tsvelik, Dynamical properties of one-dimensional Mott
insulators, in: Proceedings of Newton Institute EuroConference on Strongly Correlated Electron
Systems: Novel Physics and New Materials, Cambridge, England, 2630 June, 2000, in press,
cond-mat/0011439.
[23] A.O. Gogolin, A.A. Nersesyan, A.M. Tsvelik, Bosonization and Strongly Correlated Systems,
Cambridge Univ. Press, 1999.
[24] F.H.L. Essler, A.M. Tsvelik, private communication.
[25] V.M. Filev, Spectrum of two-dimensional relativistic model, Teoret. Mat. Fiz. 33 (1977) 119
124.
[26] V.J. Emery, A. Luther, I. Peschel, Solution of the one-dimensional electron gas on a lattice,
Phys. Rev. B 13 (1976) 12721276.
[27] E. Melzer, On the scaling limit of the 1-d Hubbard model at half filling, Nucl. Phys. B [FS] 443
(1995) 553564.
[28] F. Woynarovich, P. Forgacs, Scaling limit of the one-dimensional attractive Hubbard model: the
half-filled band case, Nucl. Phys. B 498 (1997) 65603.
[29] V.E. Korepin, F.H.L. Essler, SU(2) SU(2) invariant scattering matrix of the Hubbard model,
Nucl. Phys. B 426 (1994) 505533.
455
[30] A.B. Zamolodchikov, Al.B. Zamolodchikov, Structure constants and conformal bootstrap in
Liouville field theory, Nucl. Phys. B 477 (1996) 577605.
[31] V. Fateev, D. Fradkin, S. Lukyanov, A. Zamolodchikov, Al. Zamolodchikov, Expectation values
of descendent fields in the sine-Gordon model, Nucl. Phys. B 540 (1999) 587609.

Cost of the generalised hybrid Monte Carlo

algorithm for free field theory
A.D. Kennedy, Brian Pendleton
Department of Physics and Astronomy, The University of Edinburgh, The Kings Buildings, Edinburgh,
EH9 3JZ, Scotland, UK
Received 22 August 2000; accepted 12 March 2001
Abstract
We study analytically the computational cost of the generalised hybrid Monte Carlo (GHMC)
algorithm for free field theory. We calculate the Metropolis acceptance probability for leapfrog and
higher-order discretisations of the molecular dynamics (MD) equations of motion. We show how to
calculate autocorrelation functions of arbitrary polynomial operators, and use these to optimise the
GHMC momentum mixing angle, the trajectory length, and the integration stepsize for the special
cases of linear and quadratic operators. We show that long trajectories are optimal for GHMC, and
that standard HMC is more efficient than algorithms based on second order Langevin Monte Carlo
(L2MC), sometimes known as Kramers equation. We show that contrary to naive expectations HMC
and L2MC have the same volume dependence, but their dynamical critical exponents are z = 1 and
z = 3/2, respectively. 2001 Elsevier Science B.V. All rights reserved.
PACS: 12.38.Gc; 11.15.Ha; 02.70.Lq
Keywords: Hybrid Monte Carlo (HMC); GHMC; Molecular dynamics; Field theory; Lattice field theory
1. Introduction
Hybrid Monte Carlo (HMC) [1] remains the most popular algorithm for simulation
of quantum chromodynamics (QCD) including the dynamical effects of fermions. It is
therefore imperative that we have a detailed understanding of how the computational cost
of HMC varies as we change simulation parameters such as the lattice volume and the
fermion mass. In this paper we present detailed analytical results for the computational
cost of the generalised HMC algorithm for free field theory. We expect many of the results
obtained to also be applicable to more general field theories where most of the modes
are weakly interacting, for example, the high momentum modes of asymptotically free
theories such as QCD.
E-mail addresses: adk@ph.ed.ac.uk (A.D. Kennedy), bjp@ph.ed.ac.uk (B. Pendleton).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 2 9 - 8
A.D. Kennedy, B. Pendleton / Nuclear Physics B 607 [FS] (2001) 456510
457
We shall introduce some powerful techniques for analysing a wide class of algorithms
for free field theory. Whereas inventing algorithms which are efficacious for free field
theory but useless for interesting ones is a popular but fruitless exercise, the analysis of
generic algorithms for free fields is apparently very informative. The reason that this is so
is because our algorithms are sufficiently dumb that they spent most of their time in dealing
with the trivial almost free modes of the system.
In this paper we give detailed descriptions of the techniques we have developed to enable
us to calculate fictitious-time evolution operators, Metropolis acceptance probabilities, and
autocorrelation functions for arbitrary polynomial operators in GHMC simulations of free
field theory. This allows us to minimise the computational cost of such simulations, and
to compare the efficiency of various popular limits of GHMC. This paper brings together
and extends many results that have been summarised previously [28]. The techniques
developed for this paper have also been used in several other papers [912].
The structure of this paper is as follows: in Section 2 we describe the generalised hybrid
Monte Carlo (GHMC) algorithm and discuss various limiting cases. Section 3 demonstrates that GHMC computations for free field theory are equivalent to GHMC computations for a set of uncoupled harmonic operators with judiciously chosen frequencies.
In Section 4 we describe the generalised leapfrog discretisation schemes for the classical equations of motion in fictitious time, and introduce specific parameterisations of the
evolution operators in order to facilitate the calculation of the Metropolis acceptance probability in Section 5. In Section 6 we introduce general techniques for calculating Laplace
transforms of autocorrelation functions of polynomial operators in free field theory, and
present explicit results for arbitrary linear and quadratic operators for several of the limiting
cases of GHMC, both for fixed- and exponentially-distributed GHMC trajectory lengths.
Section 7 presents a detailed analysis of the cost of HMC simulations in the approximation that the Metropolis acceptance rate is unity, whilst Section 8 analyses the relative
cost of GHMC, HMC, and the second order Langevin algorithm (L2MC) for non-trivial
acceptance rates under some very weak assumptions. Finally Section 9 discusses the somewhat related topic of how to optimise relative frequencies of updates and measurements in
general Monte Carlo simulations. Some concluding remarks are contained in Section 10.
Several detailed technical results are relegated to appendices.
2. Generalised hybrid Monte Carlo algorithm

For simplicity, we shall describe the generalised hybrid Monte Carlo (GHMC) algorithm
for a theory of scalar fields, denoted generically by , with action S().
In order to generate the distribution
1 S()
e
[d]
Z
we introduce a set of fictitious momenta and generate the fields and according to
the distribution
458
1 H (,)
e
[d][d],
Z
(1)
where
H (, ) =
1 2
+ S(),
2 x x
and ignore .
We begin by recalling [13] that a Markov process will converge to some distribution
of configurations if it is constructed out of update steps each of which has the desired
distribution as a fixed point, and which taken together are ergodic. The generalised HMC
algorithm [5,6,14] is constructed out of two such steps.
2.1. Molecular dynamics Monte Carlo
This in turn consists of three parts:
(i) Molecular dynamics (MD): an approximate integration of Hamiltons equations on
phase space which is exactly area-preserving and reversible; this is a map U ( ):
(, ) ( , ) with det[( , )/(, )] = 1 and U ( ) = U 1 ( ).
(ii) A momentum flip F : .
(iii) Monte Carlo (MC): a Metropolis accept/reject test

F U (, ), with probability min(1, eH ),

( , ) =
(, ),
otherwise.
This may be implemented by generating a uniformly distributed random number
y [0, 1], so that
1

dy eH y = min 1, eH .
The composition of these gives the MDMC update step

H

H
= F U ( ) e
y + I y e
.

(2)
This satisfies detailed balance because (F U )2 = I.

2.2. Partial momentum refreshment
This mixes the Gaussian-distributed momenta with Gaussian noise

cos
sin
=
F
.
sin cos

If and are Gaussian distributed, PG ()[d] e

PG ( ) =

d
2 /2
[d], then so is PG ( ):
d PG ( )PG () ( + cos sin ).
(3)
459
The extra momentum flip F is included so that the trajectory is reversed upon an MC
rejection instead of on an acceptance.
We may combine the MDMC update of Eq. (2) with the partial momentum refreshment
of Eq. (3) to obtain the following expression for their combined effect

1
0
0

= 0 cos
sin U ( ) eH y + F y eH .

0 sin cos
(4)
2.3. Special cases of GHMC
Several well-known algorithms are special cases of GHMC:
The usual hybrid Monte Carlo (HMC) algorithm is the special case where = /2,
i.e., the fictitious momenta p are replaced by the Gaussian distributed , the old
momenta being discarded completely. The momentum flips may be ignored in this
case as long as MCMD and momentum refreshment steps are strictly alternated.
= 0 corresponds to pure MDMC, which is an exact version of the MD or
microcanonical algorithm. It is in general non-ergodic.
The second order Langevin Monte Carlo (L2MC) algorithm of Horowitz [14,15]
corresponds to choosing arbitrary but MDMC trajectories of a single leapfrog
integration step, = . This is an exact version of the second order Langevin
algorithm [1619].
The Langevin Monte Carlo (LMC) [20] algorithm has = /2 and = . This is
an exact version of the Langevin algorithm.
2.4. Tunable parameters
The GHMC algorithm has three free parameters, the trajectory length the momentum
mixing angle , and the integration step size . These may be chosen arbitrarily without
affecting the validity of the method, except for some special values for which the algorithm
ceases to be ergodic. We may adjust these parameters to minimise the cost of a Monte Carlo
computation, and the main goal of this work is to carry out this optimisation procedure for
free field theory.
Horowitz [14] pointed out that the L2MC algorithm has the advantage of having a higher
acceptance rate than HMC for a given step size, but he did not take in to account that it also
requires a higher acceptance rate to get the same autocorrelations because the trajectory is
reversed at each MC rejection. It is not obvious a priori which of these effects dominates.
The parameters and may be chosen independently from some distributions PR ( )
and PM () for each trajectory (but of course they cannot be chosen in a way which is
correlated with the starting point in phase space). In the following we shall consider various
choices for the momentum refreshment distribution PM , but we shall always take a fixed
value for , PM () = ( 0 ). The generalisation of our results to the case of other
distributions of values is trivial, but it is not immediately obvious that it is useful. We
460
choose each trajectory length independently from some distribution PR ( ), as this avoids
the lack of ergodicity caused by choosing a fixed trajectory length which is a rational
multiple of the period of any mode of the system [21]. This is a disease of free field theory
which in interacting models is removed to some extent by mode coupling.
3. Free field theory

3.1. Complex fields on a finite lattice
Consider a d-dimensional free scalar field theory on a V Ld lattice for which the
expectation of an arbitrary operator () is defined by the functional integral

1

[d] eS()(),
Z
where Z is chosen such that 1 = 1. For a complex field : ZdL C the action is

1 2
S() =
x + m2 x ,
2
d
xZL
where
is the lattice Laplacian
2 x
d
(x+ 2x + x ).
=1
For the generalised hybrid Monte Carlo (GHMC) algorithm discussed in Section 2 we
introduce a set of fictitious momentum fields , and the corresponding Hamiltonian
1
H (, )
x x + S(),
2
d
xZL
and we evaluate the functional integral

1
[d][d] eH (,)().
Z
For theoretical analysis it is useful to diagonalise the Hamiltonian by Fourier transformation to real (as opposed to fictitious) momentum space. From the identity of
Example B.1.1 we have

e2ipx/L x
p (F )p Ld/2
xZdL
(5)

x = F 1 x = Ld/2
e2ipx/L p
pZdL
and similarly for the fictitious momenta. We obtain

(,
) = 1
H (, ) = H
p p + p2 p p ,
2
d
pZL
(6)
461
where the frequencies are

d
p2 m2 + 4
sin2
=1
p
.
L
(7)
The phase space configuration (, ) is generated by GHMC with probability density

proportional to

(,
)
H (,)
H
]
e
[d][d] = e
det
det
[d ][d
],
eH (, ) [d ][d
where the last relation follows because the Jacobian is a (real) constant. We thus see that
free field theory corresponds to a set of harmonic oscillators with a specific choice of
frequency spectrum.
3.2. Infinite lattices
S1d
For an infinite lattice Zd (instead of ZdL ) the corresponding momentum space is the torus
(instead of ZdL ), and Eqs. (5), (6) and (7) get replaced by
p (F )p (2)d/2
eipx x
xZd

1
x = F x = (2)d/2 dp eipx p
d
(,
) = 1
H (, ) = H
2
S1

dp p p + p2 p p ,
S1d
and
(p)2 m2 + 4
d

=1
sin2
p
,
2
where we made use of Examples B.1.2 and B.1.3.

3.3. Real fields
For real (as opposed to complex) fields : ZdL R we define
p(e) Re p ,
p(o) Im p .
(e)
(o)
Since p
= p(e) and p
= p(o) these two fields are independent degrees of freedom
only on a subset of the momentum space p ZdL . We may choose to define the real
momentum field : Zd R as
L
462
p =

2
p(e) , if p p,
(o)
p , if p < p,
where the ordering is arbitrary (e.g., lexicographic). The Hamiltonian is then

(e) + i (o)2 + 2 (e) + i (o)2
(,
) = 1
H
p
p
p p
p
2
d
pZL

1 (e)2
p + p(o)2 + p2 p(e)2 + p(o)2
2
d
pZL

1 2
p + p2 p2
2
d
pZL
2 = 2 .
because p
p
3.4. Spectral averages

Results obtained for a general set of harmonic oscillators will be expressed in terms
of spectral averages, which may be evaluated explicitly for free field theory. We use
the notation (f ) to denote the spectral average of some function f : S1d R, (f )

Ld pZd f (2p/L).
L
The finite lattice results may further be expanded about the infinite lattice results to
obtain a large volume expansion by means of the Poisson resummation formula (B.10) of
Section B.2,

d 2

1
dp cos(p kL) f (p).
(f ) =
2
=1
kZ
Example 3.4.1. Consider the massless spectral average of the th power of the square of
the frequency in one dimension (where the volume V Ld = L)

2 1
p
2p 2
1 2 4
=

=
p =
sin2
V
V
V
V
V
pZV
2
pZV
2
pZV
p 4
p
+
dp sin2 cos pkV
2
2
k=1
0
0

1 2
4
= B 12 , + 12 =
,

2
where all the higher order terms vanish for large V because 0 dp sin2 p2 cos pkV = 0
for kV > upon integration by parts. This result is to be compared with the exact answer
given in Eq. (B.1) of Appendix B.
=
4
2
dp sin2
463
4. Leapfrog evolution
4.1. Lowest order leapfrog
We wish to consider a system of V uncoupled harmonic oscillators {p } for p ZV . The
hybrid Monte Carlo algorithm requires us to introduce a corresponding set of fictitious
momenta {p }, and the dynamics on this fictitious phase space is described by the
Hamiltonian

1 2
p + p2 p2 .
2
V
H=
(8)
p=1
The classical trajectory through phase space must obey Hamiltons equations (using the

symplectic 2-form N
i=1 dp dp )
H
p =
= p ,
p
p =
H
= p2 p .
p
The leapfrog equations are the simplest discretisation of these which are exactly reversible
and area-preserving,

p 12 = p (0) + p (0) 12 = p (0) p2 p (0) 12 ,

p ( ) = p (0) + p 12 = p (0) + p 12 ,

p ( ) = p 12 + p ( ) 12 = p 12 p2 p ( ) 12 .
This leapfrog integration scheme is a linear mapping on phase space 1

( + )
( )
= U0 ( )
,
( + )
( )
where the matrix

1 12 2
U0 ( )
+ 14 3
1 12 2

(9)
satisfies det U0 = 1, as it must because the mapping is area-preserving. This lowest-order

leapfrog integration agrees with the exact Hamiltonian evolution up to errors of O( 3 ),

U0 ( ) = eH I + U0,1 3 + O 5 ,
(10)
where

H
H
0 1

H
1 0
(11)
is the generator of a translation through fictitious time, and U0,1 is some operator on phase
space. The error must be an odd function of because leapfrog integration is reversible
to all orders in .
1 For the rest of this section we shall consider only a single oscillator with = 1, as everything will be
p
diagonal in p and the frequency dependence can be recovered by dimensional analysis.
464
4.2. Higher order leapfrog

Following Campostrini et al. [22,23] we can easily construct a higher-order leapfrog
integration schemes with errors of O( 5 ) by defining a wiggle

U0
U0
.
U1 ( ) U0
1
1
1
Clearly this is area-preserving and reversible because U0 is. Using Eq. (10), we find that

3

2 1
U1 ( ) = exp
H I + U0,1 2 13
+ O 5 ,
1
1
3
if we choose 1 = 2 then the coefficient of 3 vanishes, and we can arrange the step
size to equal that of the lowest-order leapfrog scheme by taking 1 = 2 1 . The explicit
form for the first-order wiggle U1 ( ) is

6+5 3 2+4 3 4
4+4 3 2+3 3 4
1
1
1
2
4
6
3
5
1 2 + 24 +
288
3
3
4
25+20 3 2+16 3 4 7
5
+ 16 3 + 2+
144
1728
144
3
3
1 4 + 6+5 2+4 4 6
1 12 2 + 24
288
This construction can be iterated by defining

Un ( ) Un1
Un1
Un1
,
n
n
n
(12)
which again clearly is area-preserving and reversible. This may be shown to have errors of
the form

Un ( ) = eH I + Un,1 2n+3 + O 2n+5
(13)
by induction on n: assume Eq. (12) holds n < N , then from Eqs. (12) and (13) we find
that
2N+1

2N+3

2 N
2N+1
+ O
,
H I + UN1,1 2 N
UN ( ) = exp
N
N
which gives us Eq. (13) for n = N upon setting N = 2N+1 2 and N = 2 N .

4.3. Parameterisation of leapfrog evolution operators
In order to calculate the explicit form for a molecular dynamics trajectory consisting
of / leapfrog steps it is useful to parameterise the leapfrog matrices Un ( ) in the
following way. Reversibility requires that for any leapfrog matrix Un ( ) = Un ( )1
and area-preservation requires that det Un ( ) = 1, hence

Un2,2 ( ) Un1,2 ( )
Un1,1 ( ) Un1,2 ( )
Un ( ) =
=
.
Un2,1 ( ) Un2,2 ( )
Un2,1 ( ) Un1,1 ( )
For the lowest-order leapfrog matrix of Eq. (9) we observe that the diagonal elements are
even functions of and the off-diagonal elements are odd functions of ; this property
also holds for all the higher-order leapfrog matrices defined by Eq. (12). We may therefore
parameterise Un in terms of two even functions

n ( ) = 1 + n,1 2n+2 + O 2n+4 ,

n ( ) = 1 + n,1 2n+2 + O 2n+4 ,
as

Un ( ) =
cos[n ( ) ]
n ( ) sin[n ( ) ]
sin[n ( ) ]
n ( )
.
cos[n ( ) ]
465
(14)
(15)
We may easily compute the leading terms in the Taylor expansions of these functions. For
the lowest-order leapfrog scheme we obtain

1
3
2 +
4 + O 6 = 1 + 0.0416 2 + O 4 ,
24
640

1 2
1
4 + O 6 = 1 0.125 2 + O 4 ,
0 = 1
8
128
for the Campostrini wiggle

(32 + 25 3 2 + 20 3 4) 4
+ O 6 1 0.06614 4 + O 6 ,
1 = 1
1440
3

(4 + 3 2 + 2 3 4) 4
+ O 6 1 + 0.03804 4 + O 6 ,
1 = 1 +
288
and for the second-order wiggle
!

1
3
3
1321294 + 1039220 2 + 831376 4
2 = 1 +
881798400

3
3
5
+ 1146320 + 901600 2 + 721280 4 2

5
3
3
+ 958174 + 753620 2 + 602896 4 4

3
3
5
+ 811769 + 638470 2 + 510776 4 8
"

3
3
5
+ 726418 + 571340 2 + 457072 4 16 6 + O 8

1 + 0.02169 6 + O 8 ,
!

1
3
3
391582 + 312244 2 + 247752 4
2 = 1
125971200

3
3
5
+ 348752 + 277664 2 + 219456 4 2

3
3
5
+ 293134 + 233308 2 + 184248 4 4

3
3
5
+ 243521 + 194042 2 + 153684 4 8
"

3
3
5
+ 211570 + 168880 2 + 134352 4 16 6 + O 8

1 0.04573 6 + O 8 .
0 = 1 +
Approximate values for n,1 and n,1 are listed in Table 1. We note in passing that the
magnitude of these leading non-trivial coefficients do not grow rapidly with increasing n.
466
Table 1
The coefficients n,1 and n,1 of Eq. (14) for n up to 8. The limiting
values of log10 H n /x for with m = 0 are also given for the
values of n appearing in Fig. 1
n
n,1
n,1
log10 H n /x
0
1
2
3
4
5
6
7
8
0.0417
0.0661
0.0217
0.0204
0.0258
0.0437
0.0137
0.0528
0.1545
0.1250
0.0380
0.0457
0.0038
0.0118
0.0483
0.0371
0.1178
0.0813
1.0280
0.9945
0.2861
0.7422
1.4128
4.4. Time evolution operators

The parameterisation given in Eq. (15) facilitates the computation of the time evolution
operator Un (, ) Un ( )/ for trajectories of length where, in general, .
We assume as an induction hypothesis that

sin[n ( )k ]
cos[n ( )k ]
k
n ( )
Un (k , ) = Un ( ) =
,
n ( ) sin[n ( )k ] cos[n ( )k ]
from this it immediately follows that Un ( )j Un ( )k = Un ( )j +k using simple
trigonometric identities. Expressing the result in terms of the trajectory length rather than
the number of integration steps, k , we obtain

sin[n ( ) ]
cos[n ( ) ]
n ( )
.
Un (, ) =
(16)
n ( ) sin[n ( ) ] cos[n ( ) ]
This time evolution matrix may be expanded about the exact Hamiltonian evolution as a
Taylor series in ,

Un (, ) = eH I + Un,1 ( ) 2n+2 + O 2n+4 ,
(17)
where from Eq. (11)

cos

eH =
sin
sin
cos
and

Un,1 ( ) = n,1 sin

,
sin
cos
(18)
cos
sin
It is instructive to compare Eqs. (13) and (17).

+ n,1
0
1

1
.
0
467
5. Acceptance rates
In this section we describe the calculation of the acceptance rate for MDMC (and thus
for GHMC) for a system of V uncoupled harmonic oscillators [24,25]. The method of
calculation is the same for both Langevin Monte Carlo and hybrid Monte Carlo, and is
independent of whether one uses lowest order leapfrog or a higher order discretisation
scheme; the various algorithms differ only in the explicit form of the time evolution matrix
Un (, ) given in Eq. (16).
The Hamiltonian of Eq. (8) is a quadratic form

V
1 p p T p p
H
,
p
p
2
p=1
so the change in fictitious energy over a trajectory is

V

1 p p T
p p
,
Mn
Hn H ( ), ( ) H (, ) =
p
p
2
p=1
where Mn
I, and we have abbreviated p (0) and p (0) to p and p .

Inserting the Taylor expansion of Eq. (17) and noting that (eH )T eH = 1, corresponding
to exact energy conservation for = 0, we find that

Mn Mn,1 2n+2 + Mn,2 2n+4 + O 2n+6
T

I + Un,1 2n+2 + O 2n+4 I
= I + Un,1 2n+2 + O 2n+4
T
2n+2
2n+4
= Un,1 + Un,1
.
+ O
UnT Un
The probability of the change in energy H n having the value when averaged over the
equilibrium distribution of starting points on phase space is

1
PH n ( ) =
[d][d] eH ( H n ),
Z
where as usual
the partition function Z is just the normalisation constant required to
ensure that d PH n ( ) = 1. It is convenient to choose an integral representation for the
-function as a superposition of plane waves, as then
+i0

d i( H n )
1
H
e
PH n ( ) =
[d][d] e
Z
2
+i0

d i 1
e
=
[d][d] eH iH n
2
Z

d i 1
e
=
[d][d]
2
Z

V
1 p p T
q q
exp
{I + i Mn }p,q
p
q
2
p,q=1

ei
d
d i 1 tr ln(I+i Mn )
e e 2
=
.
=
2 det(I + iMn )
2
(19)
468
It will prove useful to observe that the exact area-preservation property of the time
evolution operator ensures that det Un = 1, and thus that
exp tr ln(I + Mn ) = det UnT Un = det UnT det Un = 1,
and hence tr ln(I + Mn ) = 0. This implies that the quantity tr ln(I + iMn ) must vanish
not only for i = 0 but also for i = 1. Expanding the logarithm, and making use of this
fact, we find that

1
tr ln(I + iMn ) = i(1 i) tr M2n,1 4n+4 + O 4n+6 .
(20)
2
We may perform an asymptotic expansion of the integral over using Laplaces method.
First we recast Eq. (20) into a form where the V dependence is explicit

1
1, 3
2
,
tr ln(I + iMn ) = i(1 i) Mn,1 x + O 2n+2
(21)
2
V
where we have introduced the variable x V 4n+4 , and the spectral average (f ) =
1
V tr f , which has a finite limit as V . In Eq. (21) the correction terms are
negligible if
2n+2
3
V . Laplaces
the only important contributions come from the regions where
method shows that such contributions are exponentially suppressed. Using Eq. (21) in
Eq. (19) we obtain

d i 1 i(1i) ( M2n,1 )x
e e 4
.
PH n ( )
2
Completing the square gives
(M2n,1 )x
$2
exp
PH n ( ) #
4
(M2n,1 )x
(M2n,1 )x

and as H n = d PH n ( ) = 14 (M2n,1 )x this may also be written as [25]

( H )2
1
PH ( )
exp
,
4H
4H
where we have suppressed the index n.
The average Metropolis acceptance rate is now easily found,
&
%

Pacc min 1, eH =

d PH ( ) min 1, e
0
=

d PH ( ) +
2
=

d e
1
2
H
d PH ( )e
(

1'
1 % 2&
H
.
= erfc
H = erfc
2
8
(22)
The explicit form for

1
2
(sin )2 4n+4
H n = M2n,1 x = 2xn,1
4
V
1
2
(sin p )2 p4n+4 ,
= 2xn,1
V
469
(23)
p=1
where
1
= 0.03125
32
for the lowest-order leapfrog integration scheme, and
3
3
(40 + 32 2 + 25 4)
2
21,1 =
0.002894,
41472
1
2
22,1
=
793437161472000
!

3
3
221124374726 + 175505620552 2 + 139299694184 4

3
3
5
+ 192732972517 + 152971330676 2 + 121414378108 4 2
5

3
3
+ 167855113064 + 133226114824 2 + 105742133216 4 4

3
3
5
+ 145990198426 + 115872242816 2 + 91967883784 4 8
"

3
3
5
+ 126961318646 + 100768983952 2 + 79980474344 4 16
2
20,1
=
0.004183
for higher-order wiggles.
5.1. One-dimensional free field theory
For the case of free field theory we can compute the spectral averages using the explicit
form for the frequency spectrum. Using the methods of Appendix B we find that the
spectral average
1
(2)
(sin p )2 p4
m2
V
pZV

= 3 3J0 (4 ) + 4J2 (4 ) J4 (4 )

+ 2 2J0 (4 ) + 4tJ1 (4 ) + J2 (4 ) m2

1
+ 1 + 2t 2 1 J0 (4 ) + 3tJ1 (4 ) m4 + O m6
(24)
2
up to terms which vanish as V . The finite volume corrections are given explicitly in
Appendix B. For the higher order integration schemes the corresponding spectral averages
are also given in Appendix B.
The values for the logarithm of H /x are shown in Fig. 1. The limiting values for
for the massless case (m = 0) are given in Table 1; for m > 0 the corresponding
quantities diverge.
470
Fig. 1. The logarithm of H n /x, where x V 4n+4 , for one-dimensional free field theory in an
infinite volume and with a mass m = 0.01 is shown for various orders of Campostrini wiggles (n = 0
corresponds to leapfrog integration). See Table 1 for the limiting values of the curves as with
m = 0.
6. Autocorrelation functions
6.1. Simple Markov processes
Let (0 , 2 , . . . , N1 ) be a sequence of field configurations generated by an equilibrated ergodic Markov process, and let () denote the expectation value of some operator for distributed according to the fixed point distribution of this Markov process.
For simplicity we shall assume that = 0 in Sections 6.1 and 6.2. We may define an
unbiased estimator over the finite sequence of configurations by
N
1
(j ),
N
j =1
so =
%
1
N
N
j =1 (j ) = = 0.
The variance of this estimator is
N1 N1
& % &
1
(j )(j )
( )2 = 2 = 2
N
j =0 j =0
)
N1
N2
N1

&
1 %
= 2
(j )2 + 2
(j )(j )
N

j =0
j =0 j =j +1
471

)
N1
&
2
1%
2
() 1 +
=
(N 6)C (6) ,
N
N
6=1
where
C (6)
(6 )(0 )
()2
(25)
is the autocorrelation function for . If the Markov process is ergodic, then for large 6,
|C (6)| 6max e6/Nexp ,
(26)
where max is the second-largest eigenvalue of the Markov matrix and Nexp is the
exponential autocorrelation time. If N Nexp then
)

% 2&
Nexp
()2
1+O
= 1+2
C (6)
N
N
6=1

Nexp
()2
{1 + 2A }
(27)
1+O
,
N
N

where A
6=1 C (6) is the integrated autocorrelation function for the operator .
This result tells us that on average 1 + 2A correlated measurements are needed to
reduce the variance by the same amount as a single truly independent measurement.
6.2. Hybrid stochastic processes
Suppose that a sequence of configurations ((t0 ), (t1 ), (t2 ), . . . , (tN )) is generated
from (t0 ) as follows: the configuration (tj ) is generated from (tj 1 ) by choosing
momenta (tj 1 ) as described in Section 2.2, and integrating Hamiltons equations for
a time interval j tj tj 1 , where each trajectory length j is chosen randomly from the
distribution PR (j ). The autocorrelation function C defined by Eq. (25) may be expressed
in terms of the autocorrelation function
((t6 ))((t0 ))
C (1 , . . . , 6 )
2
by averaging it over the refresh distribution

C (6) =
d1 d6 PR (1 ) PR (6 )C (1 , . . . , 6 ).
(28)
The integrated autocorrelation function then becomes

A =
d1 d6 PR (1 ) PR (6 )C (1 , . . . , 6 ).
6=1 0
If we wish to determine autocorrelations in terms of the total elapsed fictitious time

t = t6 t0 = 6k=1 k of the sequence of trajectories we may introduce yet another
autocorrelation function by
472

C (t)
d1 d6 PR (1 ) PR (6 ) t
6
j C (1 , . . . , 6 ),
(29)
j =0
6=1 0
and in terms of this we find that

A =
dt C (t).
(30)
The function C (t) has the advantage of giving the autocorrelations as a function of MD
time which is approximately proportional to computer time.
If we make the reasonable assumption that the cost of the computation is proportional to
the total fictitious (MD) time for which we have to integrate Hamiltons equations, and to
the volume V of the lattice, 2 then the cost C per independent configuration is proportional
to (1 + 2A )V / with denoting the average length of a trajectory. The meaning of
independent configuration was discussed in Section 6.1, and depends on the particular
operator under consideration. The optimal trajectory length is obtained by minimising
the cost, that is by choosing so as to satisfy
dC
=0
d
dC
=0
d
1 + 2A + 2
dA
= 0.
d
dA
= 0,
d
(31)
6.3. Autocorrelation functions for polynomial operators

In order to carry out these calculations we make a simplifying assumption:
Assumption 6.3.1. The acceptance probability min(1, eH ) for each trajectory may
be replaced by its value averaged over phase space Pacc min(1, eH ); we neglect
correlations between successive trajectories. Including such correlations leads to seemingly
intractable complications. It is not obvious that our assumption corresponds to any
systematic approximation except, of course, that it is valid when H = 0.
The action of a generalised HMC update on the fields , their conjugate fictitious
momenta , and the Gaussian noise for the trajectory under consideration from Eq. (4)
is
= A(, , y, H ) ,

where the matrix A depends on the trajectory length , the noise rotation angle (which
can be chosen independently for each trajectory if we so wished), the uniform random
number y used in the Metropolis accept/reject test, and the value of H .
2 For free field theory the volume V is just the number of uncoupled harmonic oscillators.
473
We may ignore the corrections of non-leading order in to the MD evolution operator

because for any given value of Pacc of O(1) there is a corresponding value of x which is
also O(1), and thus is of order V 1/(4n+4) . These corrections therefore only contribute
to the autocorrelations through the acceptance rate itself at leading order in the large
volume expansion.
From the leading order contribution (18) we obtain
1
0
0
cos
sin
0
A = sin cos cos cos
sin + + 0 cos sin
0
sin
cos
sin sin
cos sin cos
with [(eH y)], and being the appropriate frequency for each mode.
6.3.1. Linear operators
We are interested in calculating the autocorrelation function for a general linear operator

p p
=
(32)
p
for a set of uncoupled harmonic oscillators {p }. Such an operator is of course connected,

meaning that = p p p = 0. Its autocorrelation function may be expressed in
terms of the autocorrelation functions for the individual harmonic oscillators themselves,
for
p (t)p (0)
(t)(0)
=
,
p p
C =
2

2

p,p
and as the oscillators are uncoupled

p (t)p (0) = p,p p (t)p (0) + (1 p,p )p p = p,p p (t)p (0),
thus

C (t) =
2
2
p p Cp (t)p

,
2 2
p p p
where
Cp (t) =
p (t)p (0)
.
p2
Let us proceed to calculate these single mode autocorrelation functions; while doing so we
can drop the subscript p for notational simplicity.
The average over the Gaussian distribution of gives = 0, so we can drop the last
column of A, and since a new is taken from a heatbath at the start of each trajectory we
may drop the last row also. This leaves us with the basis (1) (, ), which is updated

by (1) = A(1) (1) where the matrix

cos
sin
1
0
(1)
A
(33)
+ +
.
sin cos cos cos
0 cos
474
6.3.2. Quadratic operators

Let be a generic quadratic operator for the set of uncoupled harmonic oscillators {p },

pq p q
(34)
pq
whose connected part is c = . The autocorrelation function for c is 3

Cc (t) =
c (t)c (0)
.
c2
If we define the elementary autocorrelation functions for linear and quadratic modes to be
Cp (t) and C(p2 )c (t), then we may express Cc (t) in terms of them:

* +

1
2
2
Cc (t) =
C
(t)
p2 c
2
c2 p pp (p )c
)

% 2 &% 2 &
2
pq Cp (t)Cq (t) p c q c .
+2
p>q
Higher degree polynomial operators may be treated in the same way. As before we
calculate the single mode autocorrelation functions and again drop the subscript p while
doing so.
We cannot directly average the GHMC update matrix A over as we did in the linear
case, but we can linearise the problem by considering a homogeneous quadratic operator
as a linear combination of the quadratic monomials (()2 , , 2 , , , 2 ).
These quadratic monomials are updated by the symmetrised tensor product of the update
matrix A,
( )2
()2

2
= A s A ,

2
2
where the update matrix A s A is explicitly
2
2
2
A
2A A
2A A
2A A
A2
2A A
A2
2A A
2A A
A2
A A A A + A A A A A A + A A A A + A A A A
A2
2A A
A2
2A A
2A A
A2
As the update is now linear we can average over the Gaussian distribution of as before.
The basis monomials become (()2 , , 2 , 0, 0, 2 = 1), and this leads to three
simplifications:
3 The integrated autocorrelation function for a disconnected operator diverges in general.
475
The fourth and fifth columns are multiplied by zero, and can thus be dropped.
The fourth and fifth rows are not of interest and may be dropped too, as is chosen
from a heatbath before the next trajectory, and we already know how the linear
monomials (, ) are updated.
The last row may be replaced by (0, 0, 0, 0, 0, 1) as we know that is Gaussian
distributed and thus 2 = 1.
We are thus led to consider the update of the inhomogeneous monomials of even degree

in and alone, namely (2) (()2 , , 2 , 1), which is updated by (2) =
A(2) (2) where
A2
2A A
A2
A2
A A
A(2)
A2
A A + A A
2A A
0
cos2
cos sin cos
=
sin2 cos2
0
1
0
0 cos
+
0
0
0
0
A A
A2
0
2 cos sin
(cos2 sin2 ) cos
2 cos sin cos2
0
0
cos2
0
A A
A2
1
sin2
cos sin cos
cos2 cos2
0
0
0
sin2 +
1
0
0
.
sin2
1
(35)
6.4. Laplace transforms of autocorrelation functions

We can extract more information about the autocorrelation function C (t) than just the
integrated autocorrelation function A . We shall discuss this further in Section 7. In order
to do this it is convenient to compute the Laplace transform of the autocorrelation function

F () LC (t)
dt C (t)et

d1 d6 PR (1 ) PR (6 )C (1 , . . . , 6 )e1 e6 .
(36)
6=1 0
We may generalise the results of Sections 6.3.1 and 6.3.2 and observe that the update
step for the vector of inhomogeneous monomials in and , (d) , is of the form 4
(d) ( ) = A(d)( ) (d) (0).
The matrix A(d) is block upper triangular with with first block corresponding to the d + 1
homogeneous monomials of degree d, the next block to the d 1 homogeneous monomials
of degree d 2, and so forth. The dimension of the matrix, which is the number of
4 After averaging over the distributions of and y which are chosen independently for each trajectory.
476
even or odd inhomogeneous monomials of degree d or less, is 14 (d + 2)2 for d even and
1
4 (d + 3)(d + 1) for d odd.
The connected operator ( d )c d d = v T (d) where v = (1/d , 0, . . . , 0, d ),
and the 6-trajectory autocorrelation function is thus
&
% T (d)
v [A (1 ) A(d)(6 ) (d) (0)]v T (d)(0)
.
C( d )c (1 , . . . , 6 ) =
(v T (d) )2
By virtue of Assumption 6.3.1 we may replace each matrix A(d) by its phase space average,
C( d )c (1 , . . . , 6 ) =
vT
.6
j =1 A
(d) (d) T v
j )
,
(d) (d) T
(d) (
v T
v
and the Laplace transform of the degree d single mode autocorrelation function is therefore

(d) 6 (d) (d) T v
vT
6=1 A
F( d )c () = LC( d )c (t) =
,
T
v T (d) (d) v
where

A
(d)
%
&
d PR ( )e A(d)( ) .
(37)
Summing the geometric series we obtain a simple expression for the Laplace transformed
autocorrelation function,
T
1 + F( d )c () =
v T [I A(d)]1 (d) (d) v

T
v T (d) (d) v
(38)
In order to evaluate the expectation values, we need the Gaussian averages for the
equilibrium probability distribution of Eq. (1) for the Hamiltonian given in Eq. (8):
% 2k+1 & % 2k+1 &
)
p
= p
=0
&
%
&

%

(p p )2k = p2k = 1 2k k + 12
and
%
& % &% &% &% &
p q r s = p q r s
when p %= q and r %= s.
For linear operators the matrix of expectation values in Eq. (38) is
% (1) (1) T & 1 0
=
.

0 1
The explicit form of the Laplace transformed update matrix is obtained from Eq. (33) by
first averaging over phase space, which replaces + by Pacc ( ) and by 1 Pacc ( ),
and then Eq. (37) gives

+
G1,0 + G
0,0
A(1) =
G+
cos
0,1
G+
0,1
(G+
G
1,0
0,0 ) cos
477

,
(39)
where
G+
j,k
d PR ( )e Pacc ( )(cos )j (sin )k
and
G
j,k

d PR ( )e 1 Pacc ( ) (cos )j (sin )k .
The Laplace transform of the linear single mode autocorrelation function is
+
G+
1,0 + G0,0 (G1,0 G0,0 + G0,1 ) cos
2
F () =
+
1 G+
1,0 G0,0 + (G1,0 + G0,0 + G1,0 G0,0 + G0,1 ) cos
2
, (40)
which for the special case of = /2 (HMC) simplifies to

F (; = /2) =
G+
1,0 + G0,0
1 G+
1,0 G0,0

For quadratic operators the matrix of expectation values is
3 0 1 1
% (2) (2) T & 0 1 0 0
=

1 0 3 1,
1 0 1 1
and from Eqs. (35) and (37)
+
G2,0 + G0,0
G+ cos
1,1
A(2) =
+
G0,2
2G+
1,1
(G+
2,0 G0,2 G0,0 ) cos
2
2G+
1,1 cos
cos2
G+
0,2
G+
1,1 cos
2
(G+
2,0 + G0,0 ) cos
sin2
G0,0
G0,0
where
Gj,k G+
j,k
+ G
j,k

=
d PR ( )e (cos )j (sin )k .
(41)
The Laplace transform of the quadratic single mode autocorrelation function F( 2 )c ()

is
+ 2
+ 2
+
+
+
2
G+
2,0 + G0,0 + G2,0 2G1,1 + G0,2 G0,0 + G0,2 G2,0 + G0,0 cos

+

+
+
2
G+
2,0 G0,2 + G0,0 G0,0 + G2,0 + G0,2 (cos )
478
"

+
+ + 2
+
+
+ 2
+ 2
2
3
+ G
0,0 + G2,0 + G0,2 G2,0 2G0,2 G2,0 + G0,2 + 4G1,1 G0,0 (cos )
!
+
+
+ 2
2
+
+
+
1 G
0,0 G2,0 G0,2 G0,0 2G1,1 + G0,0 + G0,2 G2,0 G0,0 + G2,0
2
+
G+
2,0 G0,2 cos

2
+ 2
2
+
2
G+
2,0 + G0,0 2G2,0 G0,0 + G0,2 G0,0 + G2,0 (cos )
+ 2
3
+ 2
+
+ 2
+
+
+ 2G1,1 G+
0,2 4G1,1 G0,0 + G0,2 G2,0 G0,2 G2,0
+
+
+
+
+ G+
0,2 G2,0 G0,2 G0,0 + G2,0 G0,0 G2,0 + G2,0
2
+
+
+
+
+
G
0,0 4G2,0 G1,1 G2,0 G0,0 4G1,1 G0,2
2
"1
2
+ 2
3
+
+

3
+ G+
,
0,2 G0,0 G0,2 G0,0 + G0,0 + 2G0,2 G2,0 G0,0 (cos )
(42)
which for the special case of = /2 (HMC) simplifies to

F( 2 )c (; = /2) =
G+
2,0 + G0,0
1 G+
2,0 G0,0
6.5. Exponentially distributed trajectory lengths

To proceed further we need to choose a specific form for the momentum refresh
distribution. In this section we will analyse the case of exponentially distributed trajectory
lengths, PR ( ) = rer where the parameter is just the inverse average trajectory length
r = 1/ . In Section 6.6 we will consider fixed length trajectories.
We make the approximation that H and thus Pacc are independent of (cf., Fig. 1), so
G+
j,k Pacc Gj,k ,
G
j,k (1 Pacc )Gj,k ,
where Pacc Pacc ( ). This approximation is only made in order to avoid unpleasant
integrals which cannot be evaluated in closed form. The integral in Eq. (41) may be
evaluated, and we find
1
j
k

( 12 )j +k (1)+' 2 k( B
j
k
exp
,
Gj,k =
(43)
( + 1)2 + 2 (j + k 2 2)2
=0 =0
where we have introduced the dimensionless quantity and

+ 1,
if 12 k N,
B=
(j + k 2 2), if 12 k
/ N.
The Laplace transform of the linear single mode autocorrelation function for exponentially distributed trajectories is obtained by substituting the explicit form for G into
Eq. (40), and we obtain

exp
F () = r 3 + (1 2Pacc ) cos + 3 r 2
479

+ 2(1 2Pacc ) cos + (1 Pacc ) 2 + 3 r 2

2 2
2
+ (1 Pacc ) + 1 2Pacc cos + (1 Pacc ) + 1 r 3

4 + (1 2Pacc ) cos + 3 r 3 + 2(1 2Pacc ) cos + 3 + 2 r 2 2

+ (1 Pacc ) 2 + 1 2Pacc cos + (1 + Pacc ) 2 + 1 r 3

1

2
2 4
.
+ Pacc (1 Pacc ) cos + Pacc r
Evaluating this at = 0 gives the integrated autocorrelation function
exp
A =
((1 Pacc )2 2 + 1 2Pacc ) cos + (1 Pacc ) 2 + 1

.
Pacc 2 ((1 Pacc ) cos + 1)
For HMC where = /2 we have

exp
F (; = /2) =
( 2 + 2r + ((1 Pacc ) 2 + 1)r 2 )r

,
3 + 2r 2 + ( 2 + 1)r 2 + Pacc r 3 2
with the corresponding integrated autocorrelation function being

exp
A ( = /2) =
1 Pacc
1
+
.
Pacc
Pacc 2
In the limit of unit acceptance rate, Pacc = 1, we obtain

exp
F (; Pacc = 1) =
( + (1 cos )r)r
+ (1 cos )r + r 2 2
and
exp
A (Pacc = 1) =
1 cos
.
2
Finally, for HMC in the limit of unit acceptance rate

exp
F (; = /2, Pacc = 1) =
(r + )r
,
+ r + r 2 2
(44)
and
exp
A ( = /2, Pacc = 1) =
1
.
2

The Laplace transform of the quadratic single mode autocorrelation function for
exponentially distributed trajectories is obtained by substituting the explicit form for G
into Eq. (42), and we obtain

exp
F( 2 ) () = r 4 + (cos )2 + (1 2Pacc ) cos + 4 r 3
c

+ (1 + 2Pacc )(cos )3 3(cos )2
480

+ (3 6Pacc ) cos + (4 2Pacc ) 2 + 6 r 2 2

+ (4Pacc 2)(cos )3 + (4Pacc 4) 2 3 (cos )2

+ 2(1 + Pacc )(Pacc 2) 2 + 3 6Pacc cos

+ (8 4Pacc ) 2 + 4 r 3

+ 4(1 + Pacc )2 2 1 + 2Pacc (cos )3

+ (4Pacc 4) 2 1 (cos )2

+ 2(1 + Pacc )(Pacc 2) 2 + 1 2Pacc cos

2
+ (4 2Pacc ) + 1 r 4

5 + (cos )2 + (1 2Pacc ) cos + 4 r 4

+ (1 + 2Pacc )(cos )3 3(cos )2

+ (3 6Pacc ) cos + 6 + 4 2 r 2 3

+ (4Pacc 2)(cos )3 + (2Pacc 4) 2 3 (cos )2

+ (4 4Pacc ) 2 + 3 6Pacc cos

+ (8 + 2Pacc ) 2 + 4 r 3 2

+ 2(1 + Pacc )(Pacc 2) 2 1 + 2Pacc (cos )3
+ (4 2 1)(cos )2

+ 2(Pacc + 2)(1 + Pacc ) 2 + 1 2Pacc cos

+ (4Pacc + 4) 2 + 1 r 4

+ 2Pacc (1 + Pacc ) 2 (cos )3 2(cos )2 Pacc 2

1
2
2 5
2Pacc (1 + Pacc ) cos + 2Pacc r
.
!

exp
A( 2 ) = 4(1 Pacc )2 2 + 1 2Pacc (cos )3 + 4(1 Pacc ) 2 + 1 (cos )2
c
"

+ 2(1 Pacc )(Pacc 2) 2 1 + 2Pacc cos + 2(Pacc 2) 2 1
!

"1
.
2Pacc 2 (cos 1)(cos + 1) (1 Pacc ) cos + 1
For HMC where = /2 these equations simplify to
exp
F( 2 ) (; = /2) =
c
( 2 + 2r + (2(2 Pacc ) 2 + 1)r 2 )r

3 + 2r 2 + (4 2 + 1)r 2 + 2Pacc r 3 2
481
and
exp
A( 2 ) ( = /2) =
c
2 Pacc
1
+
.
2Pacc 2
Pacc

!

exp
F( 2 ) (; Pacc = 1) = 2 + cos + 2 (cos )2 r
c

"
+ (cos )3 (cos )2 cos + 2 2 + 1 r 2 r
!

3 + cos + 2 (cos )2 r 2

+ (cos )3 cos + 1 + 4 2 (cos )2 r 2

"1
+ 2(cos )2 2 + 2 2 r 3
and
exp
A( 2 ) (Pacc = 1) =
c
(cos )3 (cos )2 cos + 2 2 + 1

.
2 2 ((cos )2 1)

exp
F( 2 ) (; = /2, Pacc = 1) =
c
( 2 + 2r + (1 + 2 2 )r 2 )r
,
3 + 2r 2 + (4 2 + 1)r 2 + 2r 3 2
(45)
and
exp
A( 2 ) ( = /2, Pacc = 1) = 1 +
c
1
.
2 2
6.6. Fixed length trajectories

In this section we consider the case of fixed length trajectories, PR ( ) = ( ). In
this case we find that
G+
j,k = Pacc Gj,k
and G
j,k = (1 Pacc )Gj,k ,
without making any approximations, and from Eq. (41) we obtain

Gfix
(cos )j (sin )k .
j,k = e

The Laplace transform of the linear single mode autocorrelation function for fixed length
trajectories Ffix () is obtained by substituting the explicit form for G into Eq. (40),

(1 + 2Pacc ) cos e2 (Pacc cos + 1 Pacc )e
!
(1 2Pacc ) cos e2
"1

+ (Pacc + Pacc cos 1) cos + Pacc cos + 1 Pacc e 1 .
482

Afix
=
(1 2Pacc ) cos + Pacc cos + 1 Pacc

.
Pacc (cos + 1)(cos 1)
For HMC where = /2 we have

Ffix (; = /2) =
(Pacc cos + 1 Pacc )e

,
1 (Pacc cos + 1 Pacc )e
with the corresponding integrated autocorrelation function being

Afix
( = /2) =
Pacc cos + 1 Pacc

.
Pacc (1 cos )

Ffix (; Pacc = 1) =
cos e cos e2
,
(cos cos + cos )e cos e2 1
and
Afix
(Pacc = 1) =
cos cos
.
cos cos + cos cos 1

Ffix (; = /2, Pacc = 1) =
cos e
,
1 cos e
(46)
and
Afix
( = /2, Pacc = 1) =
cos
.
1 cos

The Laplace transform of the quadratic single mode autocorrelation function for fixed
length trajectories is obtained by substituting the explicit form for G into Eq. (42), and
we obtain
!

fix
2
e3 (1 + 2Pacc )(cos )3
F(
2 ) () = Pacc (cos ) 1 + Pacc e
c

+ e2 2Pacc + 1 + 2Pacc (cos )2 (cos )2
"

+ e2 Pacc + Pacc (cos )2 1 cos

Pacc (cos )2 1 + Pacc (cos )2

+ 2Pacc (cos )2 + 1 cos Pacc (cos )2 1 + Pacc e

+ e2 + e2 Pacc + e2 Pacc (cos )2

+ e3 2e3 Pacc (cos )3

+ e2 2Pacc + 1 + 2Pacc (cos )2 (cos )2
+e

Pacc + Pacc (cos )2 1 cos + 1
483
1
.

(1 2Pacc )(cos )2 + 2 cos Pacc (cos )2 Pacc (cos )2 1 + Pacc
.
c
(cos 1)(cos + 1)(cos 1)(cos + 1)Pacc
For HMC where = /2 these equations simplify to
=
Afix
( 2 )
fix
F(
2 ) (; = /2) =
c
(Pacc (cos )2 + 1 Pacc )e

,
1 Pacc (cos )2 + 1 Pacc e
and
Afix
( = /2) =
( 2 )
c
Pacc (cos )2 + 1 Pacc

.
Pacc (1 cos )(1 + cos )

fix
F(
2 ) (; Pacc = 1)
c
!

= (cos )2 e + (cos )3 e3 e2 1 + 2(cos )2 (cos )2
"
e2 (cos )2 cos

(cos )2 (cos )2 + 1 + 2(cos )2 cos + (cos )2 e

+ e2 (cos )2 + e3 (cos )3 e2 1 + 2(cos )2 (cos )2
1
2
2
e
(cos ) cos 1
,
and
(cos )2 2 cos (cos )2 + (cos )2
.
c
(cos )2 (cos )2 (cos )2 (cos )2 + 1
Afix
(Pacc = 1) =
( 2 )
fix
F(
2 ) (; = /2, Pacc = 1) =
c
(cos )2 e
,
1 (cos )2 e
(47)
and
Afix
( = /2, Pacc = 1) =
( 2 )
c
(cos )2
.
1 (cos )2
7. Autocorrelations and costs for HMC with Pacc 1

In this section we calculate autocorrelations for the special case = 0 (HMC) and in
the approximation where the acceptance rate is unity, Pacc = 1. We shall consider more
general cases in the following section.
We begin with some general observations:
484
Integrated autocorrelation time. It is immediately obvious that the integrated

autocorrelation time may be obtained from the Laplace transform (36) by evaluating
it at = 0.
Exponential autocorrelation time. For an ergodic Markov process we can write the
typical asymptotic form of the autocorrelation function as
C (t) constant et /texp ,
for t .
This exponential autocorrelation time is a different quantity from Nexp of Section 6.1,
which was defined in terms of Markov steps. texp can be extracted from F () by
considering its analytic structure in the complex plane. Since texp governs the most
slowly decaying exponential, it follows from the definition of the Laplace transform
(36) that F () will have its rightmost pole at Re = 1/texp .
Dynamical critical exponent. One of the most relevant measures of the effectiveness
of an algorithm for studying continuum physics is the exponent z relating the cost C
to the correlation length of the system, C z .
Optimal choice for . For the GHMC algorithm we should minimise the cost
by varying both and . For the case of quadratic operators with exponentially
distributed trajectory lengths, for instance, the optimal choice of parameters when
Pacc = 1 is to take 0 and = 12 2 0. However, this ignores the fact that
the cost does not decrease when we take smaller than the required to obtain
a reasonable Metropolis acceptance rate. If we choose opt = (L2MC) and the
corresponding value for opt we find that the cost is less than for the HMC case, but
only by a constant factor. As the cost is only defined up to a implementation dependent
constant factor anyhow we may conclude that generalised HMC does not appear to
promise great improvements over HMC. The situation is more complex in the real
world where Pacc %= 1, which is explored in Section 8.
7.1. Exponentially distributed trajectory lengths
The least uninteresting linear operator in free scalar field theory is the magnetisation

M = x (x). From Eqs. (5) and (7) this is simply the zero-momentum mode in Fourier
space, = m, and is thus expected to evolve most slowly in fictitious time.
Example 7.1.1. From Eq. (32), with pq = p0 , and Eq. (44), the Laplace transform of
the autocorrelation function for the magnetisation is
FM () =
r(r + )
.
(r + ) + m2
As explained previously, the exponential autocorrelation time texp = 1/ Re exp

corresponds to the rightmost pole in FM , and this occurs when
/
2
r
r
exp =
m2 ,
2
2
and therefore

texp =
2
,
1(2m )2
2 ,
if
if
485
1
2m ,
1
2m ,
where we have used the fact that r = 1/ . We observe that the exponential autocorrelation
time is minimised when the average trajectory length is chosen to be = 1/2m. Note that
M only couples to the zero momentum mode, and its relaxation rate determines texp in this
case.
The integrated autocorrelation function AM is given by

1 2
,
AM = FM (0) =
m
and we can minimise the cost of computingthe magnetisation by making use of Eq. (31).
The optimal trajectory length is opt = 2/m, which corresponds to the minimum
integrated autocorrelation function value AM (opt ) = 1/2. Note that the optimal trajectory
length
does not minimise the exponential autocorrelation time they differ by a factor of
2 2.
The correlation length for this system is 1/m, and our result indicates that the
optimal trajectory length is opt z with a dynamical critical exponent z = 1. Indeed,
if we were to choose an average trajectory length then we would find that the cost
per effectively independent configuration would grow as

,
for 1, ,
C
2 , for 1, .
Keeping the average trajectory length fixed as the correlation length increases, i.e.,
choosing = 0, thus leads to a dynamical critical exponent z = 2.
We can calculate the autocorrelation function CM (t) in closed form by inverting the
Laplace transform. If we expand FM in partial fractions
FM () =
where
'
1
1+
,
(1 + 2 ) (1 + + 2 )
1 (2m )2 , then it is easy to verify that the autocorrelation function is

1
1 0
CM (t) =
(1 + )e(1)t /2 (1 )e(1+)t /2
2
$

1
t
t
1
+ sinh
= et /2 cosh
when 0 < < 1. When 2 < 0 we have

$

1 t /2
t
t
1
CM (t) = e
cos
+ sin
,
2
'
where i = (2m )2 1. We observe that there are oscillations in the autocorrelation function for long trajectories where > 1/2m. For the critical case = 0 we have

1
t
CM (t) =
1+
et /2 .
486
Finally, the autocorrelation function expressed in terms of Markov steps is CM (n) =

en/T0 , where the exponential autocorrelation time Nexp = T0 = 1/ ln[1 + (m )2 ].
Example 7.1.2. We obtain the Laplace transform of the connected autocorrelation
function 5 for M 2 by setting pq = p0 q0 in Eq. (34). From Eq. (45)
FMc2 () =
r 3 + 2r 2 + r 2 + 2m2 r
.
3 + 2r 2 + (r 2 + 4m2 ) + 2m2 r
(48)
The integrated autocorrelation function is thus

1 1 2
AMc2 = FMc2 (0) = 1 +
.
2 m
Minimising the cost by means of Eq. (31), we obtain opt = 1/( 3 m) and thence
AM 2 (opt ) = 5/2. Again, the dynamical critical exponent is z = 1. The optimum trajectory
length is different from that for M, which is to be expected.
For exponentially distributed trajectory lengths the Laplace transform of the autocorrelation function is a rational function in with the numerator of lower degree than the
denominator (see, e.g., Eqs. (44) and (45)), which implies that the autocorrelation function
is a sum of exponentials. In this case the exponents are the roots of the cubic denominator,
and they are either all real, or one is real and the other two are complex conjugates depending on the value of the mean trajectory length 1/r. This is shown explicitly in Appendix A.
Example 7.1.3. Following Eq. (45), it is easy to write down the the Laplace transform of

the connected autocorrelation function for the energy E = p 12 p2 p2
FE () =
N
r 3 + 2r 2 + r 2 + 2p2 r
1
.
N
3 + 2r 2 + (r 2 + 4p2 ) + 2p2 r
(49)
p=1
The integrated autocorrelation function for the energy is

N
1 1
1 1 (1)
1 + 2 m2 ,
AE = FE (0) = 1 + 2
2
p2
2 N
p=1
(50)
#
(1)
and the optimal trajectory length is opt = m2 /3, leading to an integrated autocorrelation function value of AE (opt ) = 5/2.
(1)
For one-dimensional free field theory it remains to evaluate the spectral sum m2 ,
details of which are discussed in Appendix C. In the physically
interesting limit of small
m,
and
thus
1/
6m.
Hence
the dynamical critical
m and large V , we find m(1)
opt
2
2
exponent for the energy is 1/2.
For two-dimensional free field theory we get z = 0 up to logarithmic corrections (see
Appendix D).
5 This is a synonym for the autocorrelation function for M 2 , the connected part of M 2 .
c
487
The critical exponent is smaller in higher dimensions where most modes correspond to
large momenta 6 and the spectral sum in Eq. (50) is dominated by the higher frequencies.
From Section 6.6 the Laplace transform of the autocorrelation function for the general
linear operator of Eq. (32) is easily expressed in terms of Eq. (46). The exponential
autocorrelation time texp is related to the rightmost pole of the Laplace transform of the
autocorrelation function,
texp = 1/ Re exp .
Eq. (46) has poles in the complex plane for
=
1
ln cos p ,
and hence
1
ln | cos p |,
which leads to
Re =
texp = max
p
.
| ln | cos p ||
This diverges when p / Z for any p, which just reflects the fact that the hybrid Monte
Carlo algorithm is not ergodic for these cases, as was first pointed out by Mackenzie [21].
Perhaps this is most simply understood by considering the trajectory of the harmonic
oscillator with frequency p in the (p p , p ) phase space. The molecular dynamics
trajectories are circular arcs subtending an angle of p about the origin, and the
momentum refreshment corresponds to a change of the p coordinate leaving the p p
coordinate unchanged. If p is an integer multiple of the value of p can at most
change sign, and thus this mode certainly cannot explore the whole of its phase space.
Example 7.2.1. For the magnetisation the integrated autocorrelation function is
AM = FM (0) =
cos m
.
1 cos m
If we minimise the cost we find that the optimal trajectory length corresponds to m being
an odd multiple of , and the worst case occurs when m is an even multiple of . Both
cases just reflect the non-ergodic nature of the updating scheme discussed above: the fact
that the optimal ergodic update occurs when m is taken very close to an odd multiple
of is just an accidental consequence of the fact that M = 0.
6 This holds for two or more dimensions because the density of modes goes like p d1 , at least in the
continuum.
488

Example 7.2.2. In the case of the quadratic operator Mc2 we find from Eq. (47) that
FMc2 () =
cos2 m
,
cos2 m
and thus AMc2 = FMc2 (0) = cot2 m for fixed length trajectories. The optimal trajectory
length mopt 1.3866. As is to be expected, the non-ergodicity at m / Z manifests
itself as a divergence in AMc2 .
Example 7.2.3. For the energy of one-dimensional free field theory we obtain
FE () =
cos2 p
1
,
N p e cos2 p
and therefore AE = FE (0) =

any p.
1
N
2
p cot p
which diverges whenever p / Z for
8. Comparison of costs for Pacc = 1

We wish to compare the performance of the HMC, L2MC and GHMC algorithms for
one-dimensional free field theory. To do this we shall compare the cost of generating
a statistically independent measurement of the magnetisation M and the magnetic
susceptibility Mc2 , choosing the optimal values for the angle and the average trajectory
length .
Following Eq. (31) we can minimise the cost with respect to without having to specify
the form of the refresh distribution.
The next step is to minimise the cost with respect to the average trajectory length
= m . Strictly speaking we should note that the acceptance probability Pacc is a
function of , but to a good approximation we may assume that Pacc depends only
upon the integration step size except in the case of very short trajectories, such as
Langevin-type algorithms (see Fig. 1). The exact solution of the problem would clearly
be highly transcendental. For this reason we shall find that although L2MC is a special
case of GHMC and thus can never be cheaper, our optimal solution 7 for very short
trajectories (acceptance probabilities very close to unity) will in fact cost more than
L2MC. Fortunately, this is irrelevant because the minimum cost for L2MC is far greater
than the minimum for GHMC, the latter occurring for long trajectory lengths where our
approximations are very good.
Another implicit approximation we make is that we treat and as independent
parameters, although the trajectory length must be an integer multiple of the step size;
again this is a very good approximation except for near to the Langevin limit. Of course
7 I.e., the solution obtained by neglecting the dependence of P
acc on when optimising the parameters
and .
489
we are also ignoring subtleties such as that a multistep leapfrog integration is cheaper
than a series of single steps, that acceptance tests are a significant fraction of the cost for
very short trajectories, and that HMC requires less memory than GHMC because the old
momenta need not be saved. All of these facts would disfavour L2MC, but we shall see
that it is not competitive even without these factors being taken into account.
8.1. Linear operators
Applying Eq. (31) to Eq. (40) we find that the cost is minimised by choosing 8 opt = 0,
and at this value the Laplace transform of the single mode autocorrelation function
becomes
+
+
G+
1,0 + G0,0 G0,1 + G1,0 + G0,0
2
FM (; opt) =
+
1 + 2G+
1,0 G1,0 + G0,0 G0,1
2
8.1.1. Exponentially distributed trajectory lengths

The integrated autocorrelation function for exponentially distributed trajectory lengths is
obtained by substituting the results (43) for the integrals of Eq. (41) into FM (; = opt )
and setting = 0. The cost becomes

V
4(Pacc 1)
(Pacc 2)
exp
CM ( , opt , ) =
+
.
m
Pacc
Pacc (Pacc 2)
Minimising this with respect to we find that
'
1 Pacc
opt =
1 12 Pacc
and that the cost at the optimal parameters is
exp
CM ( , opt , opt) =
4V (1 Pacc )
'
m Pacc 1 Pacc
which corresponds to a dynamical critical exponent z = 1.

For the lowest order leapfrog integration scheme where we must scale V 1/4 to
exp
keep Pacc constant we thus find CM V 5/4 as expected.
8.2. Quadratic operators
We can make the minimisation problem for quadratic operators manifestly algebraic by
writing c cos in Eq. (42). The condition for the cost to be a minimum is that copt is a
8 Setting = 0 corresponds to never refreshing the momenta, and thus to a non-ergodic algorithm. This is just
a peculiarity of linear operators in free field theory, and we can instead consider choosing to be very small but
non-zero to circumvent this difficulty.
490
root of the polynomial

+ 2
2
+
+ 2 4
+
+
+ 2
+
2 3
G+
1,1 G0,0 + G0,2 + G2,0 copt 2 G1,1 G0,2 G2,0 + G0,2 + G0,2 G0,0 copt

2
+ 3
+ 2
+
+ 2
+ 2 G+
0,2 G0,0 2 G0,2 2 G1,1 G0,0 2 G2,0 G1,1

2 +
+ 2 +
+ 2
+ 2
2
+ 2 G+
G
6
G
G
(51)
0,2
2,0
1,1
0,2 copt G0,2 copt + G1,1
lying in the interval [1, 1].
8.2.1. Exponentially distributed trajectory lengths
Just as in Eqs. (31) the extrema of the cost occur on the ideal defined by the polynomials
dC/dc and dC/d. Finding the Grbner basis with respect to the purely lexicographical
ordering with c < we find the point (copt, opt ) at which the cost is minimal is defined by
the equations
4
3
0 = (4 + 3Pacc )copt
+ 4(1 2Pacc )(1 Pacc )copt
2
+ 16(1 Pacc )copt
+ 4copt + Pacc 4,
2
2
2
(2 Pacc )(4 Pacc ) Pacc
6Pacc + 6 opt
0 = 4Pacc
3
5
2
+ (4 + 3Pacc ) Pacc
+ 2Pacc
10Pacc + 8 copt
4
4
3
2
+ 2(1 + Pacc ) 4Pacc
+ 6Pacc
47Pacc
+ 68Pacc 32 copt

3
4
3
2
+ 38Pacc
+ 108Pacc
16Pacc
152Pacc + 96 copt

2
5
4
3
2
+ 16Pacc
+ 56Pacc
20Pacc
168Pacc
+ 248Pacc 96 copt

4
3
2
+ 35Pacc
110Pacc
+ 54Pacc
+ 88Pacc 64 copt
5
4
3
2
+ 8Pacc
60Pacc
+ 126Pacc
62Pacc
48Pacc + 32.
(52)
(53)
Using Sturm sequences we may easily show that Eq. (52) has exactly one real root in [0, 1]
and none in [1, 0], and obviously Eq. (53) has exactly one positive solution for opt . The
cost at the point (copt, opt ) is given by
!
2 3
exp
2
CM 2 ( , opt , ) = V 7Pacc 3Pacc
4 opt
copt + (2Pacc + 1)copt + 1
c
3
2
2
copt
+ (Pacc + 4)opt
+ (2Pacc 1)copt
"
2

2
2
2
+ Pacc
5Pacc + 4 coptopt
+ (4 + 3Pacc )copt
opt
!
3
Pacc m(Pacc 1)copt
opt + optPacc m
"1
2
opt Pacc mcopt
Pacc m(1 + Pacc )coptopt
.
This solution is a function of and Pacc which are not independent variables, and using
the results (22), (23) and (24) we can compute the cost as a function of Pacc as shown in
Fig. 2.
491
Fig. 2. Cost as a function of average Metropolis acceptance rate for the GHMC algorithm compared
to HMC and L2MC for free field theory using the lowest order leapfrog integration scheme. The
operator under consideration is the magnetic susceptibility, i.e., the connected quadratic operator
depending only on the lowest frequency mode. The corresponding parameters, the momentum
mixing angle opt and the average trajectory length measured as a fraction of the correlation length
opt = opt / are also shown, all as a function of the acceptance rate Pacc . The inset graph shows
the region where the acceptance rate is very close to unity which is where the L2MC algorithm has
its minimum cost.
8.2.2. HMC
The hybrid Monte Carlo algorithm corresponds to setting
= /2, and thus we find that
'
the optimal trajectory length in this case is opt = 1/ 4 Pacc , corresponding to a cost
'
2V 4 Pacc
Copt =
.
Pacc m
This is also shown in Fig. 2.
For fixed length trajectories we shall only analyse the case of L2MC for which the
trajectory length = m . In this case copt satisfies
492
2
2
2
Pacc copt
Pacc + Pacc (cos opt )2 Pacc 2copt
(cos opt)2 + 3copt
"!
"1
2
2
4copt Pacc (cos opt )2 + 2 Pacc copt
(cos opt)2 copt
(cos opt)2 + 1
= 0.
This, together with the corresponding cost, is also plotted in Fig. 2. From this figure it is
clear that the minimum cost occurs for Pacc very close to unity, where the scaling variable
x = V 6 is very small. We may then express copt as
1/6
1/3
1/2
x
1
11
x
x
+ m2
m3
+ ,
copt = 1 m
V
2
V
24
V
and likewise
Pacc = 1

9m2 39m4
5
1+
+
+ O m6
x + O x 3/2 .
8
20
800
From these relations we find that the minimum cost for L2MC is
1/4
10
Copt = 2
V 5/4 m3/2 .
This result tells us that not only does the tuned L2MC algorithm have a dynamical critical
exponent z = 3/2, but also it has a volume dependence of exactly the same form as HMC.
We may understand why this behaviour occurs rather than the naive [14] V 7/6 m1 by the
following simple argument.
If Pacc < 1 then the system will carry out a random walk backwards and forwards along a
trajectory because the momentum, and thus the direction of travel, must be reversed upon a
Metropolis rejection. A simple minded analysis is that the average time between rejections
must be O(1/m) in order to achieve z = 1. This time is approximately
1
Pacc
= .
m
1 Pacc
n=0
For small we have 1 Pacc = erf kV 6 V 6 where k is a constant, and

hence we must scale so as to keep V 4 /m2 fixed. Since the L2MC algorithm has
a naive dynamical critical exponent z = 1, this means that the cost should vary as C
V (V 4 m2 )1/4/m = V 5/4 m3/2 .
n
Pacc
(1 Pacc )n =
9. Autocorrelations and frequency of measurement

In this final section we perform an elementary analysis of the general problem of
determining the optimal frequency for making measurements of observables on a Markov
chain.
Suppose we wish to measure the expectation value of an operator by means
of a Markov process (not necessarily HMC). If the cost 9 of making one Markov step
9 I.e., the cost measured in units of computer time.
493
is CS , and the cost of making one measurement of is CM , how often should we make
measurements in order to minimise the cost of the calculation? Due to the presence of
correlations between successive measurements, the answer is not quite trivial.
Consider a sequence of Nu uncorrelated measurements of . The sample variance Vu
is related to the intrinsic variance V in the distribution of in the usual way
& ( )2 Vu
%
Vu ( )2 =
(54)
.
Nu
Nu
For the general case of Nc successive correlated measurements, from Eq. (27), the sample
variance V is
&
%
( )2
V
= (1 + 2A )
Vu ( )2 = (1 + 2A )
(55)
Nc
Nc
so that on average 1 + 2A correlated measurements are needed to reduce the variance by
the same amount as a single truly independent measurement. If the cost of measuring is
small (large), then it should be beneficial to make more (less) than one measurement per
decorrelation time.
Let T be the total number of Markov steps required to generate one independent sample,
and let K be the number of Markov steps between each measurement of , so that the
number of measurements performed per independent sample is T /K. If we wish to make
sufficient measurements to generate N independent samples, from Eqs. (54) and (55) we
have
1 + 2AK
= 1,
(56)
T /K
where AK is the integrated autocorrelation function for measurements separated by K
Markov steps. The total cost C of the entire process (steps plus measurements) is clearly

T
C = N CS T + CM
(57)
.
K
Eliminating T from Eqs. (56) and (57) we obtain

CM
C = N CS +
(1 + 2AK )K.
K
The integrated autocorrelation function AK can be written in terms of the autocorrelation
function C (Kl) of Section 6.1. Assuming Eq. (26) we have
AK =
C (Kl) =
l=1
and so
eKl/Nexp =
l=1
1
eK/Nexp

eK/Nexp + 1
C = (KCS + CM ) K/N
.
exp 1
e
Minimising with respect to K gives
sinh x = x + ,
(58)
494
where
x
K
Nexp
and
CM
.
KCS
For small where measurements are cheap we find

6CM 1/3
Kopt = Nexp
,
CS Nexp
while for large where they are expensive, we obtain

2CM
Kopt = Nexp ln
,
CS Nexp
the slow logarithmic increase of Kopt being due to the exponential decay in the
autocorrelation function C (l). The crossover point, K = 1, occurs when 0.1752.
10. Conclusions
We have introduced some powerful techniques for analysing a wide class of algorithms
for free field theory. As evidence of the more general applicability of our analysis we
point to the excellent agreement with empirical Monte Carlo data of the erfc form
for the Metropolis acceptance probability as a function of integration step size [25,26].
Furthermore, results for the O(4) and O(2) spin models in two dimensions [5,27] indicate
that their autocorrelations have at least the same form as expected from our free field
theory results. Indeed, the dynamical critical exponent for the cost in the O(2) model is
consistent with z = 1 [27]. Results for the renormalised coupling in quenched QCD using
the Schrdinger functional on a series of small lattices indicate a value of z = O(1.5) [28].
For full QCD near the chiral limit, one might expect the optimal trajectory length to scale as
the inverse mass of the lightest particle, namely opt 1/m , but we know of no numerical
investigations of this hypothesis.
Perhaps the most surprising result is the failure of the L2MC (Kramers) algorithm to
have superior performance to the HMC algorithm. Despite the fact that the noise is put
into the Markov process more smoothly the desirable properties of the L2MC algorithm
are upset by its Zitterbewegung caused by its rare Metropolis rejections.
It is also somewhat unexpected that the HMC algorithm performs nearly as well as the
full GHMC algorithm, of which it is a special case, when the parameters of both are chosen
optimally. The broad minimum of the costs of these algorithms as a function of acceptance
rate has the pleasant consequence that no very fine tuning of parameters is required.
The result that the optimal HMC trajectory length is about the same as the correlation
length of the underlying field theory is significant, in that it indicates that the temptation to
use shorter, and hence cheaper, trajectories for systems near criticality should be avoided.
The results pertaining to higher order (Campostrini) integrators [22,23] are of theoretical
interest, but in practice they do not seem to be very useful because the trajectories
for interacting field theories are chaotic [29,30] and apparently limited by the intrinsic
instability of the integrators [12].
495
Acknowledgements
We gratefully acknowledge financial support from PPARC under grant number
GR/L22744. This research was also supported in part by the U.S. Department of Energy
through Contract Nos. DE-FG05-92ER40742 and DE-FC05-85ER250000.
We would like to thank David Daniel, Robert Edwards, Philippe de Forcrand, Alan
Horowitz, Ivan Horvth, Alan Irving, Blint Jo, Julius Kuti, Steffen Meyer, Hidetoshi
Mino, Stephen Pickles, Jim Sexton, Stefan Sint, Alan Sokal, and Zbyszek Sroczynski for
useful discussions, comments, and hospitality during the extremely prolonged gestation
period of this paper.
Appendix A. Inverse Laplace transforms for autocorrelation functions

A.1. Exponentially-distributed trajectory lengths
In this case the Laplace transform of the autocorrelation function, F () = LC(t),
is a rational function with the numerator of lower degree than the denominator. If the
denominator is square free then we have the partial fraction expansion
F=

k
Ak
.
Bk
Since
Le
dt et et =
1
+
(Re + > 0),
we have
L1 F =
A k e Bk t .
The roots Bk of the denominator of F come in complex conjugate pairs because the
coefficients in F are real. The autocorrelation function can therefore always be written
as a sum of real exponentials (corresponding to the real roots) and of real exponentials
multiplied by cosines (corresponding to pairs of complex conjugate roots).
If the denominator of F is not square free then the repeated roots give terms of the form
A/( B)n in the partial fraction expansion. Consider then
Lt
n1 t
dt et t n1 et = ( + )n (n)
(Re + > 0).
This serves to give the inverse Laplace transform in the general case.
496
A.2. Fixed length trajectories

Here F is a rational function of e with the numerator being of lower or
equal degree than the denominator. For simplicity we first consider the case where the
denominator is square free, whence by partial fractions
Ak
.
F =C+
Bk
k
Observe that

L(t)
et (t) = 1,
and, more generally,

L
j (t j ) =
j =0
j ej =
j =0
1
1 e
(|| < 1),
so we have
L1 F = C(t)
Ak
k
= C(t)

k
Bk
Ak
Bk (t j )
j =0
e(j +1) ln Bk (t j ).
j =0
All the the roots must satisfy |Bk | > 1 for the geometric series to converge at = 0 ( = 1),
which is necessary for the integrated autocorrelation function F (0) to be finite.
In the general case where the denominator of F has multiple roots the general form of
the inverse Laplace transform may be obtained from the identity

n+j 1 j
(t j ) = (1 )n (| | < 1).
L
j
j =0
A.3. Example of computation of autocorrelation function

To illustrate the computation of autocorrelation functions we shall explicitly evaluate
the connected autocorrelation function for M 2 for the HMC algorithm with exponentially
distributed trajectory lengths. The Laplace transform of the autocorrelation function is
F () =
r 3 + 2r 2 + r 2 + 2m2 r
.
3 + 2r 2 + (r 2 + 4m2 ) + 2m2 r
If we write this in terms of two distinct roots 1 and 2 of the denominator

F () =
r(r 2 + 2r + 2 + 2m2 )
( 1 )( 2 )(2r + 1 + 2 + )
497
Fig. 3. The autocorrelation function and the cumulative cost are shown as functions of MD time for
the optimal trajectory length and those costing 50% more.
we can expand this in terms of partial fractions to give

F () = r 2r 4 + 4r 3 1 + 212 m2 r 2 + 1 rm2 + 32m4 + 412m2

"1

( 1 ) 2r 4 13m2r 2 + 64m4
!

+ r 4r 3 1 + 212 13m2 22 1 r 2
"

+ (72 81 )m2 r + 16m4 + 412 42 1 m2
!
"1

( 2 ) 2r 4 13m2r 2 + 64m4

2 2
2
4
2
+ r 22 1 + m r + (72 + 71 )m r + 16m + 42 1 m
!

"1
.
(2r + 1 + 2 + ) 2r 4 13m2r 2 + 64m4
The inverse Laplace transform of this is immediately obvious

r 2r 4 + 4r 3 1 + 212 m2 r 2 + 1 rm2 + 32m4 + 412 m2 e1 t
1

2r 4 13m2 r 2 + 64m4
!

+ r 4r 3 1 + 212 13m2 22 1 r 2 + (72 81 )m2 r + 16m4
"

1

+ 412 42 1 m2 e2 t 2r 4 13m2 r 2 + 64m4
498
"
!

+ r 22 1 + m2 r 2 + (72 + 71 )m2 r + 16m4 + 42 1 m2 e(2r+1+2 )t
1

2r 4 13m2 r 2 + 64m4 .
In Fig. 3 we show this function for m = 102and the

optimal
value
r
=
3m, together
opt
with the two neighbouring values of r = (3 3 15)m/2 for which the cost is 50%
greater.
Appendix B. Evaluation of spectral averages

In this section we show how to evaluate the following spectral sums for one-dimensional
free field theory:

1
()
2 p
sin
A
(B.1)
V
V
pZV
1
= 2
2
()
m2
'/V (

)

2
2
2
kV
+
+
,
()
kV
+ kV
(B.2)
k=1
1 2
p
V
(B.3)
pZV

)
'/V

(
2
2
2()
kV
=
()
+2
m
,
kV
=0
m2
(B.4)
k=1
1
cos(2 p )
V
(B.5)
pZV

(2m2 )
=0
m()
2
1

d +1
=0
1 2 2
p sin ( p )
V
J2kV (4 ),
(B.6)
kZ
(B.7)
pZV

1 () 1
1 d2
= m2
m2 ,
2
2
4 d 2
where
p2

m + 4 sin
2

p
.
V
(B.8)
(B.9)
B.1. Completeness relations

Let G be a group with a family of D-dimensional representations q : G (CD
i.e., q (g) : CD CD , labelled by some parameter q; thus q (gh) = q (g)q (h)
for g, h G. Summing over group elements (or integrating with respect to Haar measure
CD ),
499
in the continuous case) gives

q d q (g),
G
and multiplying by q (h) for any h G we find

q (h)q = q (h) d q (g) = d q (h)q (g)
G

=
d q (hg) =
G
d q (g ) = q ,
where we have made use of the (left) invariance of Haar measure d. Hence either
q (h) = 1 h G or q = 0. If we let q = 0 label the identity representation of G, and
all other q be non-identity representations, we have
q = (q).
Example B.1.1. Let G = ZV , q ZV , and q (p) = e2ipq/V :
1 2ipq/V
e
= q,0 .
V
pZV
Example B.1.2. Let G = Z, q S1 = [0, 2], and q (p) = eipq :

1 ipq
e = (q).
2
pZ
Example B.1.3. Let G = S1 , q Z, and q (p) = eipq :

1
2
2
dp eipq = q,0 .
0
B.2. A Poisson resummation formula

Let f : S1 S1 be a periodic function. As it is periodic it obviously wants to be
expanded in eigenfunctions of S1 :
2
f (p) =
dp (p p)f (p )
[p S1 ]
2
=
0
dp
1 iq(p p)
e
f (p )
2
qZ
500
iqp
qZ
1
2
2
dp eiqp f (p ),
hence the spectral average

1
2p
(f ) =
f
V
V
pZV
2
1 2ipq/V 1

=
e
dp eiqp f (p )
V
2
pZV qZ
1
=
e2ipq/V
V
qZ
pZV

1
2
kZ
1
2
2
iqp
dp e
f (p )
)
q,kV
qZ kZ
)
1
2
2
)
dp e
iqp
f (p )
2
dp eip kV f (p ).
As f is real valued we may take the real part of this identity to obtain

2
1
1
2p
(f )
f
dp cos(p kV )f (p ).
=
V
V
2
pZV
kZ
(B.10)
B.3. Multiple angle expansion

The following simple identity is often useful

2
1 i
e ei
( N)
2i
2
2
1
2
=
()q e2i(q)
2i
q

sin2 =
q=0
)

1
2
1 2 2
q
=
() +
() 2 cos[2( q) ]
q
2i
q=0

)

1
2
2
q

= 2
+
() 2 cos(2q ) .
2
q

q =1
(B.11)
B.4. Free-field spectral sums

Let us first consider Eq. (B.1). From Eq. (B.10) we have

2
1
2pq
1
cos
dp cos(p kV ) cos(p q )
=
V
V
2
pZV
kZ
2
1 10
1
=
cos[p (kV + q )] + cos[p (kV q )]
2
2
kZ
1
kZ
{kV +q ,0 + kV q ,0 }.
Using the multiple-angle expansion (B.11) we obtain

p
1
A()
sin2
V
V
pZV

)

2
2
1 1
2pq
q
+
() 2 cos
=
V
V
q
22
pZV
q =1

)

2pq
1
2
2
q 2
cos
= 2
+
()
q
2
V
V
pZV
q =1
)

2
2
1
q
(kV +q ,0 + kV q ,0 )
()
= 2
+
2
q

q =1
kZ
'/V (

)

1
2
2
2
kV
= 2
()
+
+
.
2
kV
+ kV
k=1
()
From this result we may obtain an expression for the quantity m2 defined in Eq. (B.3)
()
m2

1 2
1
p
p =
m2 + 4 sin2
V
V
V
pZV
=0
pZV

1 2() 2 p
m
4 sin
V
V

pZV

()
=
m2()4 A
=0

)
'/V

(
2
2
2()
kV
+2
m
.
=
()
kV
=0
k=1
501
502
Example B.4.1.
(2)
m2 =

1 4
p = 6 + 4m2 + m4 + V ,1 6 4m2 + V ,2 2.
V
pZV
Example B.4.2.
(3)
m2 =

1 6 6
p = m + 6m4 + 18m2 + 20 + V ,1 6m4 18m2 20
V
pZV

+ V ,2 6m2 + 12 V ,3 2.
In order to evaluate Eq. (B.5) we first investigate the simpler case where m = 0
1
0
cos(2 p )
V
pZV
2

1
p
dp cos(p kV ) cos 4 sin
2
2
kZ
1
d cos(2kV ) cos[4 sin ]
kZ
0

=
J2kV (4 ).
kZ
For small m we Taylor expand m2

m2 d m2
m2 =
.
! d(m2 ) m2 =0
=0
Consider the quantity

1 d d m2
1 d 1 d 1
=
cos(2 p )
2 dm2 d
2 d dm2 V
pZV

dp
d 1 1
sin(2 p )
=
d V
dm2
pZV

1
d 1
sin(2 p )
=
d V
2p
pZV
1
cos(2 p ) = m2
=
V
pZV
upon integrating with respect to we obtain

d m2
= 2
dm2
d m2 ,
503
and hence
1

1
d m2
= (2) d1 1 d2 1
d 0 .
d(m2 ) m2 =0
By this means we have obtained the desired result

1

(2m2)
d +1 0 .
m2 =
!
=0
=0
Eq. (B.7) may be found by differentiating m2 with respect to ,

1 2 2
()
p sin ( p )
m2
V
pZV
1 2
p {1 cos(2 p )}
=
2V
pZV

1 () 1
1 d2
= m2
m2 .
2
2
4 d 2
Example B.4.3.
(2)
m2 =
1 4 2
p sin ( p )
V
pZV

1 d 2 2 d m2
1 (2) 1 m2
= m2
2
2
!
4 d 2 d(m2 ) m2 =0
=0
1

1 (2) (2m2)
1 d2 2
= m2
d +1 J2kV (4 ). (B.12)
2
2!
4 d 2
=0
kZ =0
The = 0 term in the sum is

1
1 d2 2
J2kV (4 ) = 8
J
2kV (4 ).
2
2
4 d
kZ
kZ
From the recurrence relation

1
Jn = (Jn1 Jn+1 )
2
we find that
k
1 k
d k Jn (z)
() Jnk+2 (z),
= k
dzk
2
=0
which we may prove by induction:

k
dJnk+2 (z)
1 k
d k+1 Jn (z)
()
= k
2
dz
dzk+1
=0
504
k

1 k
() Jnk+2 1 (z) Jnk+2 +1 (z)
k+1
2
=0
k

1 k
= k+1
() Jn(k+1)+2 (z) Jn(k+1)+2( +1)(z)
2
=0
k

k
1
k
= k+1
+
() Jn(k+1)+2 (z)
1
2
=1
)
+ Jn(k+1) (z) Jn+(k+1) (z)

=
using
k
k
1

k+1
1 k+1
() Jn(k+1)+2 (z),
2k+1
=0
k+1
. We therefore find that
4

4
() J2kV 4+2 (4 )
kZ =0
0
J2kV 4 (4 ) 4J2kV 2 (4 ) + 6J2kV (4 )
= 2
J
2kV (4 ) = 2
kZ
kZ
4J2kV +2 (4 ) + J2kV +4 (4 )
= 3J0 (4 ) + 4J2 (4 ) J4 (4 )

0
J2kV 4 (4 ) 4J2kV 2 (4 ) + 6J2kV (4 )
2
k=1
1
4J2kV +2 (4 ) + J2kV +4 (4 ) .
The = 1 term in the sum (B.12) is

1
1 d2 2
2
2m
d1 J2kV (41 )
2
4 d 2
kZ

d4
d J2kV ( )
d 4
= m2
(where 4 )
kZ

)
d3

=m
d J2kV ( ) + J2kV ( )
d 3
kZ
0
1
2
4J2kV ( ) + J
=m
2kV ( )
2
kZ
2
)
3

2
3
() J2kV 2+2 ( ) +
() J2kV 3+2 ( )
= m2
8
kZ =0
=0
505
J2kV 2 ( ) 2J2kV ( ) + J2kV +2 ( ) + J2kV 3 ( )

8
kZ
$
3
3
J2kV 1 ( ) + J2kV +1 ( ) J2kV +3 ( )
8
8
8

= m2
= m2 2J0 (4 ) + 2J2 (4 ) + 3 J1 (4 ) J3 (4 )

+
2J2kV 2 (4 ) 4J2kV (4 ) + 2J2kV +2 (4 ) + J2kV 3 (4 )
k=1
)

3 J2kV 1 (4 ) + 3 J2kV +1 (4 ) J2kV +3 (4 ) .
B.5. Results for Campostrini

m(4)
2 = 35 35J0 (4 ) + 56J2 (4 ) 28J4 (4 ) + 8J6 (4 ) J8 (4 )

+ 40 40J0 (4 ) + 64tJ1 (4 ) + 31J2 (4 ) 8J4 (4 ) + J6 (4 ) m2
!
"

+ 18 + 16t 2 18 J0 (4 ) + 48tJ1 (4 ) + 8J2 (4 ) J4 (4 ) m4 + O m6 ,

(6)
m2 = 462 462J0(4 ) + 792J2(4 ) 495J4(4 ) + 220J6(4 )

66J8 (4 ) + 12J10 (4 ) J12 (4 )

+ 756 756J0(4 ) + 1024tJ1(4 ) + 698J2(4 ) 256J4 (4 )

+ 69J6 (4 ) 12J8 (4 ) + J10 (4 ) m2
!

+ 525 + 256t 2 525 J0 (4 ) + 1152tJ1(4 ) + 316J2(4 )
"

74J4 (4 ) + 12J6 (4 ) J8 (4 ) m4 + O m6 ,

m(8)
2 = 6435 6435J0 (4 ) + 11440J2(4 ) 8008J4 (4 ) + 4368J6 (4 )

1820J8(4 ) + 560J10(4 ) 120J12(4 ) + 16J14 (4 ) J16 (4 )

+ 13728 13728J0(4 ) + 16384tJ1(4 ) + 14075J2(4 ) 6128J4(4 )

+ 2185J6(4 ) 608J8 (4 ) + 123J10(4 ) 16J12 (4 ) + J14 (4 ) m2
!

+ 12936 + 4096t 2 12936 J0 (4 ) + 24576tJ1(4 ) + 9280J2(4 )
"
2807J4(4 ) + 688J6 (4 ) 128J8(4 ) + 16J10(4 ) J12 (4 ) m4

+ O m6 ,

(10)
m2 = 92378 92378J0(4 ) + 167960J2(4 ) 125970J4(4 ) + 77520J6(4 )
38760J8(4 ) + 15504J10(4 ) 4845J12(4 ) + 1140J14(4 )

190J16(4 ) + 20J18 (4 ) J20 (4 )

+ 243100 243100J0(4 ) + 262144tJ1(4 ) + 267814J2(4 )
506
129872J4(4 ) + 54252J6(4 ) 19024J8(4 ) + 5420J10(4 )

1200J12(4 ) + 193J14(4 ) 20J16 (4 ) + J18 (4 ) m2
!

+ 289575 + 65536t 2 289575 J0 (4 ) + 491520tJ1(4 )
+ 233612J2(4 ) 82714J4(4 ) + 25164J6(4 ) 6392J8(4 )
"

+ 1300J10(4 ) 198J12(4 ) + 20J14 (4 ) J16 (4 ) m4 + O m6 .
Appendix C. Spectral sum for one-dimensional free field theory

We wish to evaluate the spectral sum
1 2
m(1)
p
2
V
pZV
for large V . Using the Poisson resummation formula we obtain

m(1)
2
2
2
1
1
cos(p kV )
2

=
dp cos(p kV )V p /2 =
dp
.
2
2
m2 + 4 sin2 ( 12 p )
kZ
kZ
0
ip
Upon substituting z e we get

i 2
zkV + zkV
(1)
dz 2
m2 =
4
z (m2 + 2)z + 1
kZ
|z|=1

i 2
1 + V
1
z|k|V
=
dz
,
=
2
(z + )(z ) m m2 + 4 1 V
kZ
|z|=1
where 1 +
1 2
1
2m 2m
m2 + 4. For m 1 we have = 1 m + O(m2), and thus
coth 12 mV
m m2 + 4
which is correct to all orders in the large V asymptotic expansion.
(1)
m2
Appendix D. Spectral sum for two-dimensional free field theory

We want to evaluate the spectral sum

1
p2 ,
Lx Ly
px ZLx
py ZLy
where
p2

2 px
2 py
m + 4 sin
+ sin
.
Lx
Ly
2
507
Using Poisson resummation we find that for Lx , Ly

1
=
(2)2
2

1

py
p
dpx dpy m2 + 4 sin2 x + sin2
,
2
2
0

so upon substituting z eipx we obtain

i
=
4 2
2
|z|=1
i
=
4 2
2
dpy
dpy
|z|=1
z2
dz
m2

+ 4 sin2 ( 12 py ) z + 1
dz
,
(z )(z 1/)
where
py
m2
+ 2 sin2
+1
2
2
py
m2
+ 2 sin2
+1
2
2
2
1.
Evaluating the contour integral gives

1
=
2
2
dpy

2

2
1/2
2
py
1 1
m
1
+ 2 sin2
+1 1
=
dpy
.
4
2
2
0
ipy
z
We now make the substitution e , which leads to

2
i
dz
'
=
2
[(m2 + 4)z z 2 1]2 4z 2
|z |=1
2
dz
i
'
=
2
[z 2 (m2 + 2)z + 1][z 2 (m2 + 6)z + 1]
|z |=1
2
i
dz
'
=
,
2
(z a)(z 1/a)(z b)(z 1/b)
|z |=1
where
m2
+3
a
2
m2
+1
b
2
/
/

m2
+3
2
m2
+1
2
2
1
1 = (3 2 2) (3 2 4)m2 + ,
8
2
1 = 1 m + 12 m2 + .
Since 1/a > 1/b > 1 > b > a > 0 the contour integral may be shrunk to be the integral
along the branch cut from a to b,
=
b
a
dz
.
'
(z a)(b z )(1/a z )(1/b z )
508
If we change variables to
x
(a + 1)(z 1)
(a 1)(z + 1)
we obtain
1/ k

dx
2 ab
'
,
=
2
(a + 1)(b 1)
(x 1)(1 k 2 x 2 )
1
where k
1/ k
(a1)(b+1)
(a+1)(b1) .
With the substitution
'
1
dx
(x 2
1)(1 k 2 x 2 )
1
'
=
0
1
k
1
x2
d
(1 2 )(1 k 2 2 )
we find that
= K(k),
where K is the complete elliptic integral and k 2 + k 2 = 1; thus
2 ab
K(k).
=
(a + 1)(b 1)
This may be further simplified using Landens transformation

2
K
= (1 + )K()
1+
with
ab
1ab ,
which leads to

ba
2 ab
K
=
,
(1 ab)
1 ab
using the fact that K is an even function. Expanding in powers of m we obtain

2
32
1
1
1 + O(m) K 1 2 m + O m =
ln
+ O(m ln m),
=
4
m2
because
K(k) = K
'

4
1 k 2 ln
k
as k 0.
References
[1] S. Duane, A.D. Kennedy, B.J. Pendleton, D. Roweth, Hybrid Monte Carlo, Phys. Lett. B 195 (2)
(1987) 216222.
[2] A.D. Kennedy, The theory of hybrid stochastic algorithms, in: Workshop on Probabilistic
Methods in Quantum Field Theory and Quantum Gravity, Cargse, August 1989, Plenum, New
York, 1990, pp. 209223.
[3] A.D. Kennedy, B.J. Pendleton, Acceptances and autocorrelations in hybrid Monte Carlo, Nucl.
Phys. Proc. Suppl. B 20 (1991) 118121.
509
[4] A.D. Kennedy, Hybrid Monte Carlo algorithm on the connection machine, Int. J. Mod. Phys.
C 3 (1992) 126.
[5] A.D. Kennedy, R.G. Edwards, H. Mino, B.J. Pendleton, Tuning the generalized hybrid Monte
Carlo algorithm, Nucl. Phys. Proc. Suppl. B 47 (1995) 781784, hep-lat/9509043.
[6] A.D. Kennedy, B.J. Pendleton, Cost of generalised HMC algorithms for free field theory, Nucl.
Phys. Proc. Suppl. B 83, 84 (1999) 816818, hep-lat/0001031.
[7] A.D. Kennedy, The hybrid Monte Carlo algorithm on parallel computers, Parallel Computing 25 (1999) 13111339.
[8] A.D. Kennedy, Monte Carlo methods for quantum field theory, Chin. J. Phys. 38 (2000) 707
720.
[9] R.G. Edwards, I. Horvth, A.D. Kennedy, Non-reversibility of molecular dynamics trajectories,
in: C. Bernard, M. Golterman, M. Ogilvie (Eds.), Nucl. Phys. Proc. Suppl. B 53 (1997) 971
973, hep-lat/9608020.
[10] R.G. Edwards, I. Horvth, A.D. Kennedy, Instabilities and non-reversibility of molecular
dynamics trajectories, Nucl. Phys. B 484 (1997) 375399, hep-lat/9606004.
[11] I. Horvth, A.D. Kennedy, The local hybrid Monte Carlo algorithm for free field theory:
reexamining overrelaxation, Nucl. Phys. B 510 (1998) 367400, hep-lat/9708024.
[12] B. Jo, B.J. Pendleton, A.D. Kennedy, A.C. Irving, J.C. Sexton, S.M. Pickles, S.P. Booth,
Instability in the molecular dynamics step of hybrid Monte Carlo in dynamical fermion lattice
QCD simulations, Phys. Rev. D 62 (2000) 114501, hep-lat/0005023.
[13] A.D. Kennedy, J. Kuti, S. Meyer, B.J. Pendleton, Program for efficient Monte Carlo
computations of quenched SU(3) lattice gauge theory using the quasi-heatbath method on a
CDC CYBER 205 computer, J. Comput. Phys. 64 (1986) 133160.
[14] A.M. Horowitz, A generalized guided Monte Carlo algorithm, Phys. Lett. B 268 (1991) 247
252.
[15] M. Beccaria, G. Curci, The Kramers equation simulation algorithm 1. Operator analysis, Phys.
Rev. D 49 (1994) 25782589, hep-lat/9307007.
[16] O. Klein, Arkiv. Mat. Astr. Fys. 16 (5) (1922).
[17] J. Kuti, Lattice field theories and dynamical fermions, in: R.D. Kenway, G.S. Pawley (Eds.),
Computational Physics, Scottish Universities Summer School in Physics, 1987, pp. 311378.
[18] A. Horowitz, Stochastic quantization in phase space, Phys. Lett. B 156 (1985) 89.
[19] A. Horowitz, The second order Langevin equation and numerical simulations, Nucl. Phys.
B 280 [FS18] (3) (1987) 510522.
[20] R.T. Scalettar, D.J. Scalapino, R.L. Sugar, New algorithm for the numerical simulation of
fermions, Phys. Rev. B 34 (1986) 7911.
[21] P.B. Mackenzie, An improved hybrid Monte Carlo method, Phys. Lett. B 226 (1989) 369.
[22] M. Campostrini, P. Rossi, A comparison of numerical algorithms for dynamical fermions, Nucl.
Phys. B 329 (1990) 753.
[23] M. Creutz, A. Gocksch, Higher order hybrid Monte Carlo algorithms, Phys. Rev. Lett. 63
(1989) 9.
[24] H. Gausterer, M. Salmhofer, Remarks on global Monte Carlo algorithms, Phys. Rev. D 40 (8)
(1989) 27232726.
[25] S. Gupta, A. Irbck, F. Karsch, B. Petersson, The acceptance probability in the hybrid Monte
Carlo method, Phys. Lett. B 242 (1990) 437443.
[26] B. Joo et al., Parallel tempering in lattice QCD with O(a)-improved Wilson fermions, Phys.
Rev. D 59 (1999) 114501, hep-lat/9810032.
[27] S. Gupta, Dynamical critical properties of the hybrid Monte Carlo algorithm, Nucl. Phys. B 370
(1992) 741761.
[28] B. Gehrmann, U. Wolff, Efficiencies and optimization of HMC algorithms in pure gauge theory,
Nucl. Phys. Proc. Suppl. 83 (2000) 801803, hep-lat/9908003.
510
[29] K. Jansen, C. Liu, Study of Liapunov exponents and the reversibility of molecular dynamics
algorithms, in: C. Bernard, M. Golterman, M. Ogilvie (Eds.), Nucl. Phys. Proc. Suppl. B 53
(1997) 974976, hep-lat/9607057.
[30] K.M. Bitar, R.G. Edwards, U.M. Heller, A.D. Kennedy, QCD function with two flavours of
dynamical Wilson fermions, Phys. Rev. D 54 (1996) 35463550, hep-lat/9602010.

Spectral and transport properties of quantum wires

with bond disorder
Alexander Altland a , Rainer Merkt b
a Theoretische Physik III, Ruhr-Universitt-Bochum, 44780 Bochum, Germany
b Institut fr theoretische Physik, Zlpicher Str. 77, 50937 Kln, Germany
Dedicated to E. Mller-Hartmann on the occasion of his 60th birthday
Abstract
Systems with bond disorder are defined through lattice Hamiltonians that are of pure nearest
neighbour hopping type, i.e., do not contain on-site contributions. They stand representative for
the general family of disordered systems with chiral symmetries. Application of the Dorokhov
MelloPereyraKumar transfer matrix technique P.W. Brouwer et al. [Phys. Rev. Lett 81 (1998)
862; Phys. Rev. Lett. 84 (2000) 1913] has shown that both spectral and transport properties
of quasi one-dimensional systems belonging to this category are highly unusual. Most notably,
regimes with absence of exponential Anderson localization are observed, the single particle density
of states exhibits singular structure in the vicinity of the band centre, and the manifestation of
these phenomena depends in an apparently topological manner on the even- or oddness of the
channel number. In this paper we re-consider the problem from the complementary perspective of
the nonlinear -model. Relying on the standard analogy between one-dimensional statistical field
theories and zero-dimensional quantum mechanics, we will relate the problem to the behaviour of a
quantum point particle subject to an AharonovBohm flux. We will build on this analogy to re-derive
earlier DMPK results, identify a new class of even/odd staggering phenomena (now dependent on the
total number of sites in the system) and trace back the anomalous behaviour of the bond disordered
system to a simple physical mechanism, viz. the flux periodicity of the quantum AharonovBohm
system. We will also touch upon connections to the low energy physics of other lattice systems,
notably disordered chiral systems in 0 and 2 dimensions and antiferromagnetic spin chains. 2001
Elsevier Science B.V. All rights reserved.
PACS: 72.15.Rn; 72.10.Bg; 11.30.Rd
E-mail address: alexal@thp.uni-koeln.de (A. Altland).

PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 2 0 9 - 7
512
A. Altland, R. Merkt / Nuclear Physics B 607 [FS] (2001) 511548
1. Introduction
In 1979, Wegner and Oppermann [1] introduced a variant of the disordered lattice
Anderson model they termed sublattice system. The sublattice system differs from
the generic Anderson model in that its Hamiltonian does not contain on-site matrix
elements, i.e., is of pure hopping type. For a long time this species of disordered electronic
systems went largely unnoticed. The status rapidly changed [213] after two aspects
became generally appreciated: first, models with sublattice structure occur in a number
of physical applications. The random flux model, lattice QCD models [14], random
antiferromagnets [15], models of gapless semiconductors [16] and effective models of
transport in manganese oxides [17] are of sublattice type or at least acquire sublattice
structure in limiting cases. Second, and contrary to naive expectations based on universality
and insensitivity to details of the microscopic implementation of disorder, the low energy
properties of the sublattice system differ drastically from those of the generic Anderson
model:
In contrast to the Anderson model, average spectral and transport properties of the
sublattice system sensitively depend on the value of the Fermi energy, EF . For Fermi
energies far away from the centre of the tight binding band, EF = 0, the sublattice
system falls into the universality class of standard disordered electron systems (which
simply follows from the fact that a Fermi energy EF = 0 can be interpreted as a
constant non-vanishing on-site contribution to the Hamiltonian). However, in the
vicinity of zero energy drastic deviations from standard behaviour occur.
In dimensions d 2, the average density of states (DoS), (E), exhibits singular
behaviour upon E approaching the band centre.
Perturbative one-loop RG calculations [2] as well as the analysis of Ref. [9] indicate
that right in the middle of the band the 2d system is metallic, i.e., does not drive
towards an exponentially localized phase.
Several phenomena of apparent topological origin are observed. E.g., the DoS profile
of sublattice quantum dots (systems in the zero-dimensional limit) sensitively
depends on the total number of sites of the host lattice being even or odd. Similarly,
the properties of quasi one-dimensional sublattice systems depend on the number of
channels, N , being even or odd. For N even the transport behaviour is conventional
conductance exponentially decreasing on the scale of a certain localization length
whereas the energy dependent DoS vanishes in a close to linear fashion. In contrast,
for N odd, the wire at E = 0 is metallic, i.e., the length-dependent conductance decays
only algebraically. At the same time, the DoS diverges upon approaching the band
centre.
A schematic summary of the band centre phenomenology of sublattice models is displayed
in Table 1.
All these phenomena root in the fact that the sublattice Hamiltonian possesses a discrete
]+ = 0, where [ , ]+ is the antisymmetry not present in the Anderson model: [3 , H
commutator and 3 a site-diagonal operator that takes values +1/1 on alternating sites.
513
Table 1
Schematic overview over phenomenology of sublattice systems. N is number of channels of a quasi
one-dimensional quantum wire, L is number of sites of a sublattice quantum dot
0d
Staggering
L even
Conductance ( = 0) Vanishing
Density of states
Spectral gap
1d
2d
L odd
N even
N odd
Non-vanishing
Zero energy states
Localization
Spectral gap
Deloc.
Divergence
Deloc.
Divergence
The presence of this chiral symmetry implies that sublattice systems fall into symmetry
classes different from the three standard WignerDyson classes unitary, orthogonal
or symplectic. To be specific, let us focus on the simplest case of a sublattice system
with broken time reversal invariance but good spin rotation symmetry (the analogue of the
Anderson model of unitary symmetry). Chiral systems fulfilling these two extra symmetry
criteria belong to a symmetry class denoted AIII in the terminology of Ref. [18]. An
alternative denotation, coined in the paper [20], is ChGUE for Chiral GUE.
As with conventional disordered electronic systems of WignerDyson symmetry,
universal transport and thermodynamic properties of systems of class AIII can be
described in terms of effective low energy theories. E.g., the results for sublattice
quantum wires summarized above have been obtained within a symmetry adapted variant
of the DorokhovMelloKumarPereyra (DMPK) transfer matrix approach [4,10,11].
This theory differs from the standard cases of unitary symmetry in that the transfer
matrices describing the propagation through the system take values in a different target
space. Similarly, the general field theory approach to disordered electronic systems, the
nonlinear -model, has been extended to systems of class AIII [2,5,6,13], too. Like
its conventional relatives, the AIII variant of the -model is a matrix field theory
whose internal structure depends on the specific implementation (boson replicas, fermion
replicas or supersymmetry). Previous studies of these models have focused on the twodimensional case [2,9], where information on long range behaviour can be obtained from
renormalization group calculations, or on the zero-dimensional case [21] where the model
can be evaluated rigorously by full integration over the zero-mode manifold.
It is the purpose of the present paper to analyse the intermediate, quasi one-dimensional
variant of the field theory. As mentioned above key aspects of the phenomenology of
quasi one-dimensional sublattice quantum wires have been discussed previously within
the framework of the DMPK approach, and it is near at hand to ask what motivates
revisiting the problem. Referring for a more substantial discussion to Section 1.1 below,
we here merely mention a lose collection of points. The field theory approach enables one
to approach the problem from a comparatively broad perspective. Specifically, the onedimensional variant of the model is but a representative of a larger family of chiral models. This makes possible to relate the behaviour of the one-dimensional system to the
extensively studied zero- and two-dimensional cases. Further, the Green function oriented
-model formalism enables one to microscopically implement coupling operators
514
connecting the wire to external leads. (Within previous DMPK formulations, the coupling
has been treated somewhat implicitly, see however [22].) Unexpectedly, we will observe
strong, albeit universal sensitivity to the modelling of the coupling and yet another class
of staggering phenomena. The origin of these effects, and their connection to the channel
number staggering mentioned before will be discussed below. Finally, the -model of the
time reversal noninvariant sublattice is the by far most simple of all ten [18] nonlinear models of disordered systems: it has only four degrees of freedom, two Grassmann and
two ordinary integration variables, the minimal set needed to construct a supersymmetric
matrix model. This makes it an ideal tutorial system on which generic properties of
the field theory approach to disordered quantum wires can be studied. We have tried to
pedagogically expose several of these aspects, particularly the analogy one-dimensional
field theories vs. point-particle quantum mechanics which plays an all important role in
the present context. Nonetheless, the analysis below will be at times technical. To make
its results and various qualitative connections generally accessible, the following section
contains a synopsis of the paper.
1.1. Summary of results and qualitative discussion
Consider the system depicted in Fig. 1: a sequence of L sites (alternatingly designated
by crosses and circles) each of which supports N electronic states, or orbitals (represented
by the vertical stacks of ovals). Nearest neighbour hopping is controlled by a regular tight
binding contribution, diagonal in the orbital index, (the horizontal line segments) plus some
bond randomness that connects different orbitals (the hatched areas). As in Ref. [4], we
allow for some staggering in the tight binding amplitudes, i.e., the hopping amplitudes
regularly alternate in strength (the alternating bond length). At either end, a number of sites
is coupled through some tunnelling barriers (horizontal hatched areas) to leads.
In this paper, three different regimes will be considered: (i) the quantum dot case,
defined through the criterion that the time to diffusively propagate through the system is
shorter than the Golden rule escape time into the leads, (ii) a diffusive regime, where the
order of these time scales is inverted but localization effects do not yet play a role, and (iii)
the regime of long systems with pronounced Anderson localization.
Beginning with the quantum dot case (i), we find that both transport and spectral
properties sensitively depend on the number of sites L being even or odd. Specifically,
the DoS exhibits a singular spike in the band centre if the number of sites is odd (and the
coupling to the leads switched off). Away from zero energy, () is strongly suppressed
up to values N, where is the mean level spacing. The conductance equals N/2,
where is a measure for the strength of the lead coupling, as for conventional quantum
dots. In contrast, for an even number of sites, the DoS shows the spectral microgap of
width characteristic for finite random systems with chiral symmetries [14]. Curiously,
the conductance vanishes provided that only one site at either end is connected to the leads.
The strong sensitivity to the total number of sites in the system disappears if more than one
site in the terminal regions is coupled to leads (cf. Fig. 1). Summarizing, the tendency
of sublattice systems to exhibit staggering phenomena pertains to the zero-dimensional
515
Fig. 1. Quasi one-dimensional sublattice system. The system is realized in terms of a sequence of
sites, alternatingly designated by crosses and circles, each of which carries N states (the ovals).
Hopping between the sites is mediated by a regular Hamiltonian (the lines connecting the ovals) plus
a weak random contribution (hatched areas). At either end, a number of sites is coupled to continuum
states (the wavy lines) propagating in leads.
limit. However, unlike in the localized regime, the relevant integer control parameter is the
number of sites L and not the number of channels N . A qualitative interpretation of these
phenomena will be given below.
In this paper, only limited attention will be payed to the intermediate diffusive
regime (ii). It is likely that the staggering phenomena observed in the zero-dimensional
regime will have interesting, albeit nonuniversal extensions into the diffusive regime. We
here avoid the confrontation with these effects by connecting several sites to the leads. This
leads to equilibrated behaviour with Ohmic conductance, as for ordinary wires. As for the
density of states, we have explored the influence of spatially fluctuating diffusion modes
on the spectral microgaps discussed above. (Semiclassically, the spectral gaps observed in
chiral or superconductor systems can be interpreted as due to an accumulation of diffusion
modes (see Ref. [5] for a detailed discussion of this point). For large energies , these
modes can be treated perturbatively by standard diagrammatic methods.) Surprisingly, no
corrections are found up to and including three loop order. This is a speciality of the
one-dimensional case. For a two-dimensional system diffusive modes would lead to a
modulation of the spectrum on the scale of the Thouless energy.
In the localized regime we reproduce the results found earlier within the DMPK
approach: for an even number of channels, the conductance decays exponentially on the
scale of a localization length Nl, where l is the mean free path. In contrast, for an
odd number of channels algebraic scaling, g L1/2 , supported by one delocalized mode
is observed [4]. A comparably strong even/odd dependence is observed in the behaviour
of the DoS. For N even, the DoS vanishes at zero energy as () |s| ln(|s|), where
s = / , and the characteristic scale of the gap, , is the level spacing of a system
of length L . This behaviour has a simple interpretation: roughly speaking, a system
with L can be viewed as a sequence of L/ decoupled localization volumes. On
516
small energy scales < , dynamics within each of these is approximately ergodic.
One would then expect the DoS to be gapped, as for the sublattice quantum dot, with
as the characteristic level spacing. Generalizing an earlier result [23,24] for the specific
case N = 1, it has been found that for an odd number of channels the DoS diverges as
() 1/(|s| ln3 |s|) [11]. This accumulation of spectral weight can be interpreted in two
different directions. First, it is near at hand to view the algebraically decaying conductance
observed in the odd case as a resonant tunnelling phenomenon: the principal tendency to
localize is outweighed by the high density of states in the vicinity of zero energy. (This
picture was first suggested by V. Kravtsov.) Second, it is tempting to interpret the zero
energy peak as an generalization of the singular spike found in the L-odd zero-dimensional
case. Unfortunately, we are not aware of a qualitative picture explaining this analogy.
At least technically, however, both phenomena can be traced back to a common origin.
Finally, a periodic modulation, or staggering, of the hopping amplitudes can be employed
to continuously interpolate between the N -even and N -odd case, respectively.
We next briefly outline the field theory route to exploring the above phenomenology.
Within the fieldtheoretical approach, long range properties of the system are described by
a functional integral with action
L
S=
0

N +f

s

1
1
1
str T T
i str T + T
dr str T T
8
2
2
+ SGade + ST .
(1)
Here we have introduced a continuum variable r [0, L] replacing the formerly discrete
site index, f is a parameter related to the staggering strength, s = , where is the
bulk metallic density of states. Further, T is a matrix field taking values in GL(1|1), i.e.,
the group of two-dimensional invertible supermatrices, and str is the supertrace. Finally,
SGade [T ] C str2 (T T 1 ), where C is some small constant and ST is a contribution
describing the coupling to the leads.
The action S[T ] defines a one-dimensional representative of the general family of
chiral nonlinear -models. In contrast to its well investigated zero- and two-dimensional
relatives much of the results summarized in Table 1 have been derived within these
models the 1d variant has not been explored so far.
Much of our analysis of this model will rely on the standard analogy between onedimensional statistical field theory and zero-dimensional quantum mechanics: S[T ] can
be interpreted as the quantummechanical action of a point particle propagating on the
supersymmetric target manifold, in our case GL(1|1). The first term of the action represents
a kinetic energy, the third term a potential, and the second term, linear in the velocity,
coupling to a constant vector potential of strength (N + f )/2. Quantum analogies of this
type have been proven a powerful technical tool in previous analyses of the standard models [25]. However, the present case is special in that the target manifold is so simple
that an intuitive interpretation of the quantum picture becomes straightforward. Indeed, the

fields T have the explicit matrix representation T = u v , where , (u, v) are Grassmann
(commuting) variables, to be compared with the four- or eight-dimensional matrices
517
entering the standard -models. Later on we will see that convergence criteria constrain
the component u to be positive real, while v = exp(i) is a pure phase. Thus, temporarily
leaving the Grassmann variables aside, our model describes quantum propagation on a
(half)line and on a circle, respectively. Notice that the latter component is topologically
nontrivial.
At this stage, the role of the vector potential contribution to the action becomes evident.
While inessential in the noncompact sector of the theory, in the compact, circular sector,
it describes the presence of an AharonovBohm type magnetic flux. This analogy explains
the presence of phenomena periodic in the channel number. An even number of channels
translates to an integer number of flux quanta through the ring, which has no effect.
However, for N odd or, alternatively, a finite staggering parameter f , a fractional flux
pierces the system and this influences both, spectrum and dynamics of the quantum system.
To develop the picture somewhat further, notice that the conductance, essentially the
transition probability through the system, maps onto the Green function of the quantum
system evaluated at imaginary time L. Imagining the latter represented through a spectral
decomposition, the large L behaviour depends on the low lying portions of the quantum
spectrum, in particular the discrete, and flux periodic level structure of the compact sector
of the theory. Later on we will see that for half integer flux (i.e., N odd) there is a zeroenergy level ( absence of localization) while for all other magnetic configurations the
spectrum is gapped ( exponential localization). This mechanism and its relevance for the
localization behaviour of the system was first analysed by Martin Zirnbauer [26]. Similar
but slightly more elaborate arguments can be used to understand the profile of the DoS.
We finally mention some intriguing parallels to the physics of the antiferromagnetic spin
chain. According to Haldanes conjecture, a chain of spins with half-integer (integer) S is
in a long range ordered (disordered) phase [27]. It has also been found (see Ref. [28] for
review) that for a chain with staggered hopping amplitude j , the strict integer/half-integer
pattern is violated, e.g., an integer chain can be fine-tuned to an ordered phase. Technically,
the system is described by a -model with a topological term not dissimilar to the one
above. The differences are that in the spin case (a) the base manifold is (1 + 1)-dimensional
(a quantum chain), (b) the target manifold is the two-sphere SU(2)/U(1), and (c) the
topological term classifies field configurations according to the number of coverings of the
sphere (instead of winding numbers around the circle, as in our case). In the spin case,
the coupling constant of the topological term is given by S + j , i.e., spin replaces channel
number and staggering plays a similar role as in our case. (In fact, the linear coupling of
the topological term to the staggering amplitude follows, independently of the model, from
parity-conservation arguments to be discussed below.) Beyond these apparent technical
parallels, the connection between the quenched disordered sublattice and the spin chain,
respectively, is not understood.
The rest of the paper is organized as follows. In Section 2 we quantify the definition
of the model. Its field theory representation is introduced in Section 3. In Section 4 the
-model transfer matrix approach, i.e., the representation of observables in terms of the
quantum Green function is discussed. This formulation is then applied to the calculation of
conductance (Section 5) and density of states (Section 6). We conclude in Section 7.
518
2. Definition of the model

We begin this section by upgrading the above qualitative introduction of the sublattice
system to a more quantitative formulation. Quantum transport through the bulk of the
system is described by the Hamiltonian

=
cl, tll + Rll cl ,
H
(2)
ll
where cl
creates a spinless electron at site l in state = 1, . . . , N , the sum extends over
nearest neighbour sites, and Rll are N N Gaussian distributed random hopping matrices
with moments

Rll = 0,

Rll Rll
2
,
N
1.
Apart from the Hermiticity condition Rll = Rl l matrices on different links are statistically
independent. These random matrices compete with the regular contribution to the hopping
matrix elements, tll . To be specific, we set tll = 1 + ()a if the smaller of the two
neighbouring site indices l and l is even (odd). The real parameter a is a measure for the
staggering strength. Notice that tll = O(1) implies that we are dealing with a weakly
disordered system.
At both ends, a number of sites is coupled to leads (see Fig. 1). Quantum propagation
within these leads is assumed to be generic, i.e., governed by a Hamiltonian without
sublattice symmetry. To describe the coupling, we add to our bulk Hamiltonian a tunnelling
contribution
TL + H
TR ,
T = H
H

L L
TL =
c Wa
da + h.c. ,
H
l=1,2,...
TR =
H

R R
da + h.c. ,
c Wa
l=L,L1,...
L(R)
where da
creates an electron propagating in the left (right) lead in a certain state a =
1, . . . , M 1 and we have introduced a composite index = (l, ) comprising site and
L/R
are subject to the
orbital index of the bulk system. The coupling matrix elements W
orthogonality relation [29]

L
Wa,l
WlL ,b = f (l)ab ll ,
R
R
Wa,(Ll)
W(Ll
),b = f (l)ab ll ,
(3)
where f (x) is some envelope function decaying on a scale of O(1) and normalized through
l f (l) = 1. The function f and parameter describe profile and strength of the
519
coupling, respectively. Why did we introduce the multi-site coupling operators (3) instead
of connecting just the two terminal sites l = 1, L to the lead continuum? Modelling the
coupling in a more general way is motivated by the presence of the site number staggering
phenomena mentioned in the introduction. The above implementation of the coupling
operator is sufficiently flexible to selectively probe these effects (Sections 5.1 and 6.1)
or to average over any boundary oscillatory structures (Sections 5.2 and 6.2).
To conclude the definition of the problem, let us introduce the Green function,
1

C
CT
W

+ i(sgn Im z)
,
W
G(z) = z H
(4)
C=L,R
C W
CT = { a W C W C } is an operator describing the escape of electrons
where W
l,a a, l
from the bulk system into the leads [29,30]. Expressed in terms of these objects, the
Landauer conductance of the system assumes the form
g = (2)
M

+

L
R
R
L
G 0 .
Wa
Wb
Wb
W a G 0
(5)
ab=1
Our second quantity of interest, the density of states per site, is given by the standard
expression

1
() =
(6)
Im
G + .
L
Finally, to prepare the field theory formulation, let us consider the symmetries of the
problem. Expressed in the notation introduced above, the chiral symmetry assumes the
denotes the bulk part of the Hamiltonian.
, 3 ]+ = 0, where 3 = {()l ll } and H
form [H
(Coupling to a non-sublattice continuum breaks chirality.) The presence of this symmetry
implies invariance under the two parameter family of transformations
cl cl ez1 ,
cl ez2 cl ,
l even,
cl cl e+z2 ,
cl e+z1 cl ,
l odd,
(7)
where z1,2 are complex numbers. Depending on the choice of these parameters, (7) expresses the standard U(1)-invariance of a model with conserved charge (z1 = z2 ), or the
axial symmetry characteristic for chiral systems (z1 = z2 ). (For the time being we treat the
transformation as purely formal, i.e., ignore the fact that for a general choice z1,2 the transformed operators are no longer mutually adjoint.) We will come back to discussing the role
of these symmetries after the effective field theory of the system has been introduced.
3. Field theory
The construction of the low energy effective field theory of the sublattice wire follows
the standard route of deriving nonlinear -models of disordered fermion systems [31],
there are no conceptually new elements involved. Referring to Appendices A and B for
520
technical details of this derivation, we here discuss structure and key features of the
resulting model.
As shown in the appendices, the field theory representation of conductance and DoS is
given by

M 2
T T 1 12 (0) T T 1 21 (L) ,
g=
(8)
2
and
0
() =
Re
2L
L

1
dr T11 (r) + T11
(r) ,
(9)
0
N
, T (r) is a field taking values in the supergroup
respectively. Here, the bulk DoS, 0 = 2
GL(1|1) and the continuum variable r [0, L] replaces the site index. The angular brackets
stand for functional averaging

. . . D T eS[T ] (. . .)
over a functional integral with action S
L
0
drL and effective Lagrangian
L Lfl + Ltop + LT + Lz + LGade .
(10)
The individual contributions are given by

N +f
str T r T 1 ,
2

Lfl = str r T r T 1 ,
16

z0
str T + T 1 ,
Lz = i
2

M
str T (r) + T 1 (r) (r) + (L r) ,
LT =
2
2

LGade C str T r T 1 ,
Ltop =
(11)
where C is a coupling constant that need not be specified other than that it is small,
C = O(1) (N, M) and the parameter f 2Na
measures the degree of staggering.
2
(Notice our f is identical to the control parameter f defined in Ref. [4].) Finally, we
have introduced a parameter
N
22
which will later identify as the localization length of the system.
To prepare the further discussion of the functional expectation values, let us briefly
discuss the internal structure of the field theory. We first note that save for the values of
the coupling constants, the structure of the action (10) can be anticipated from inspection
of Eq. (7) and its supersymmetric extension: the field theory approach starts out from
521
a promotion of the fermionic operator representation (2) to a supersymmetric formulation

in terms of Bose and Fermi fields. Within the latter representation (cf., e.g., Eq. (A.1)),
the space of permissible invariance transformations is enlarged. The formerly complex
parameters ezi are replaced by two-dimensional matrices Ti acting on the bosonic and
fermionic components of the theory. Any sensible subsequent manipulation done on the
functional integral must respect this invariance property. On the level of the effective theory
described by S[T ], the transformation acts as T T1 T T2 and indeed one verifies that the
contributions Ltop , Lfl and L of (10) are invariant under this operation. Further, the two
building blocks str(T T 1 ) and str(T T 1 ) are the only operators with 2 derivatives
compatible with the global GL(1|1) symmetry. In other words, the gross structure of the
bulk action readily follows from the invariance criterion. For finite energies or coupling
to the leads chirality is broken and global U(1) remains the only symmetry of the
Hamiltonian. Within the supersymmetry formulation, the set of allowed transformations
is then reduced down to configurations with T1 = T21 (the super-generalization of U(1)).
The operator str(T + T 1 ) is the minimal positive (see below) choice compatible with the
restricted symmetry. Summarizing, the three terms str(T T 1 ), str(T T 1 ), and str(T +
T 1 ) exhaust the set of operators compatible with the global transformation behaviour of
the model.
As for the coupling constants of the theory not specified by symmetry arguments
but all derived in Appendix B notice that contrary to the standard -models of systems
with WD symmetry two, instead of just one second derivative operators appear in the
action. Mathematically, the reason for the appearance of two contributions is that the
target manifold of the theory, GL(1|1) is a reducible symmetric space; it decomposes into
two irreducible factors, a point discussed in detail in Ref. [18]. Each of these factors can
be endowed with its own metric which implies the existence of two independent second
derivative operators in the theory. Physically, the presence of the nonstandard operator
str2 (T T 1 ) has profound consequences for the behaviour of the 2d-version of the
field theory [2]: the RG analysis of the model shows that the coupling constant of this
operator grows under renormalization while driving the coupling of the energy operator
str(T + T 1 ) to large values. At the same time the coupling constant of the standard
gradient operator str(T T 1 ), essentially the conductance, remains un-renormalized.
In one dimension, the situation is different. Contrary to what one might expect, the
contribution LGade is not remotely as important as in two dimensions, and it is another
operator that drives the localization behaviour of the model. In parentheses we remark
that the target space of the transfer matrix approach to the problem, GL(M)/U(M)
SL(M)/SU(M) R+ factors into two components, too [10]. Accordingly, the Fokker
Planck equation governing the Brownian motion on that space is controlled by two
independent coupling constants, both determined by the microscopic definition of the
model. The second of these contributions, essentially the analogue of our Lagrangian L ,
leads to nonuniversal corrections to the overall picture which perish in the limit of a large
number of channels.
A second aspect discriminating the present model from its WD relatives is the
appearance of a first order gradient operator in the action. In fact, the presence of this
522
contribution might raise suspicion: although allowed by symmetry, str(T r T 1 ) is not

invariant under space reflection r r, in contrast to the microscopic
parent model (for

a = 0). The resolution of this puzzle lies in the fact that Stop L is of topological origin
and, although not manifestly so, does respect the space inversion property. We will discuss
this point momentarily after the internal structure of the field manifold and the integration
measure have been specified.
Functional integrals can only be defined on manifolds that are Riemannian, i.e.,
endowed with a positive metric. The supergroup GL(1|1) (like all other supersymmetric
spaces that appear in the context of field theories of disordered Fermi systems) does not
fulfil this criterion, a point discussed in detail in Ref. [18]. However, it does contain a
maximally Riemannian submanifold M, defined as follows: the bosonboson block of M,
a one-dimensional manifold by itself, is isomorphic to the noncompact symmetric space
GL(1)/U(1) R+ , i.e., the positive real numbers. (This space is trivially Riemannian.)
The fermionfermion block is isomorphic to the compact symmetric space U(1). No
limitations in the Grassmann valued bosonfermion sectors of the theory are needed since
the whole issue of convergence does not arise here. Summarizing, M = R+ U(1),
where the notation is symbolic, specifying the bosonboson and fermionfermion sector,
respectively.
A convenient field representation respecting these convergence criteria is given by

x
k = exp
,
a = exp
, x, y R.
T = kak 1 ,
(12)
iy

In this parameterization, the group integration dT extends over the degrees of freedom
x, y, , , without further constraints. The invariant group measure associated to the
representation (12) are defined in Eqs. (C.1) and (C.2).
We are now in a position to discuss the role of the contribution Stop . First notice that for
sufficiently strong lead coupling the boundary action LT projects the fields onto the group
origin T (x, y) = 1, i.e., enforces x(0) = x(L) = 0 and y(0) = 2k, y(L) = 2k , where
k, k are integer. As a consequence, the first derivative operator can be written as

N +f

N +f
Stop [T ] =
dr str T r T 1 =
dr r str ln T
2
2

N +f
str ln T (L) ln T (0) = i(N + f )(k k ).
=
(13)
2
This makes the topological nature of the term manifest: it counts winding numbers in the
fermionic sector of the theory. The integer k k is a topological invariant characterizing
each field configuration T (r). Further notice that for f = 0,

eStop [T ] = e+iN(kk ) = eiN(kk ) .

Since it is the exponentiated action and not the action itself that matters, the last identity
tells us that the global sign of the first gradient operator is irrelevant (all for f = 0). This
settles the above raised issue of the behaviour of the model under space reflection: although
the first order derivative is not invariant under r r the exponentiated action is.
523
4. Transfer matrix approach

Principal aspects of the system properties we are going to analyse are nonperturbative,
i.e., cannot be obtained as power series in the coupling constants of the action. Progress
with such type of problems can be made by applying the -model transfer-matrix
technique [25], an approach conceptually similar to the DMPK formalism.
We begin by defining the two functions

r
YL (T1 , T2 , r)
D T e 0 dr L[T ] ,
T (0)=T1
T (r)=T2
D T e
YR (T1 , T2 , r)
L
r
dr L[T ]
T (r)=T1
T (L)=T2
Expressed in terms of these objects, the DoS assumes the form

0
() =
Re
2L
L
dr

1
YR (T , 1, L r),
dT YL (1, T , r) T11 + T11
(14)
where we have used that for sufficiently strong coupling to the leads, the boundary
configurations T are close to unity. (Throughout much of this paper we will consider the
DoS of coupled systems. For sufficiently large systems, the choice of boundary conditions
is inessential, a point to be verified below.) Similarly, the conductance obtains as

M
M 2
1
g=
dT dT e 2 str(T +T ) T T 1 12 YL (T , T , L)
2

M

1
T T 1 21 e 2 str(T +T ) .
(15)
From Eqs. (14) and (15) it is clear that the functions YL,R encode the essential system
properties we are interested in.
As a first step towards the computation of these objects let us explore how the
symmetries of the action translate to symmetries of YL,R . We first consider the case
z = 0, relevant to the analysis of the conductance, where the action is fully invariant
under GL(1|1)-transformations. The invariance L[T ] = L[T1 T T2 ], T1,2 = const. then
directly implies YL,R (T , T , r) = YL,R ((T1 T T2 ), (T1 T T2 ), r). From this identity one
readily verifies that

YR [T , T , r] = YR T T 1 , 1, r = YR T 1 T , 1, r ,

YL [T , T , r] = YL 1, T T 1 , r = YL 1, T 1 T , r (z = 0).
In other words, for z = 0 the arguments of the heat kernels enter in an invariant product
type form and it suffices to consider the reduced functions
YR (T , r) YR (T , 1, r),
YL (T , r) YL (1, T , r),
(16)
depending on a single argument only. The same invariance property (now evaluated for
T2 = T11 ) implies that YL,R (T , r) = YL,R (T1 T T11 , r). Imagining the argument T to be
524
represented in the polar decomposition (12) and setting T1 = k 1 the argument can be
reduced to the diagonal matrix of eigenvalues a:

YL,R kak 1 , r = YL,R (a, r).
(17)
For z = 0, the invariance of the theory collapses down to transformations with T2 = T11 .
However, the representation of the DoS above implies that we are solely interested in functions of the type (16), with second argument set to unity, anyway. Since transformations
T T1 T T11 are still permissible, these objects depend on the eigenvalues of the argument matrix only, as before for the case z = 0. Summarizing, in the analysis of both
conductance and DoS, it is sufficient to consider functions YL,R depending on a single
argument with invariance property (17).
Following the basic philosophy of the transfer matrix approach, we will compute the
functions YL,R (T , r) iteratively, by asking how much they change under infinitesimal
variation of the arguments r r + . Considering the function YL for definiteness, we
first notice that, by definition,

r+

YL (T , r + ) = D T e r dr L[T ] YL T (r), r .
For sufficiently small , the action can be expanded and we obtain
YL (T , r + ) YL (T , r)

N+f
z
1
1
1
1
= dT e 16 str(T T +T T ) e 2 str(T T )+i 2 str(T +T )

YL (T , r) YL (T , r) ,

r+
where we have used that, due to supersymmetry, D T exp( r L[T ]) 1 = 1. We
have also temporarily set the coupling constant of the operator LGade to zero. As mentioned
above, this term does not have much relevance in the present context. We will briefly
discuss its role in Section 5.5.
From hereon, the derivation of an evolution equation for YL is conceptually straightforward: the exponential weight factor in the first line of the equation enforces that T is close
to T , symbolically,
T T 1 = 1 + O(). We should thus expand T around T and evaluate

the integral dT perturbatively in . This expansion is most conveniently done in the polar coordinates introduced above (because the heat kernel depends on the radial degrees of
freedom, only). As a result of a calculation similar but much more simple than the one for
the standard -models with their larger matrix fields [31] one obtains the Schrdinger type
equation

t 4(D A)2 + V (x, y) Y L (a, r) = 0,
(18)
R
where V (x, y) = is(cosh(x) cos(y)) and we have introduced the dimensionless

parameters
r
s = .
t ,
Physically, t is the length coordinate measured in units of the localization length and s
the energy measured in units of the single particle level spacing of a system of
525
length . Further, the symbol D = (Dx , Dy )T denotes a vector differential operator defined
through

x iy
1
Dx = x coth
,
2
2

x iy
i
,
Dy = y + coth
(19)
2
2
where the constant vector
N +f
A=
(20)
(1, i)T .
2
Finally, evaluation of (18) for asymptotically small times t 0 leads to the initial condition
1
lim YL,R (x, y, t) = (x, y) lim e 8t (x
t 0
t 0
2 +y 2 )
(21)
It is very instructive to interpret the structure of the evolution equation (18) in the light
of the analogy between field theory and point particle quantum mechanics on GL(1|1)
discussed in the introduction. Within this picture, the functions YL [T , r] acquire the
meaning of Green functions, i.e., transition amplitudes for the propagation between the
origin of the manifold T = 1 at time 0 and a final configuration T at time t. The
evolution equations (18) becomes a a time-dependent Schrdinger equation with quantum
Hamiltonian, H = 2(D A)2 + V (x, y). While the term V (x, y) = is(cosh(x)
cos(y)) = is2 str(T + T 1 ) simply represents the potential inherited from the Lagrangian,
the kinetic operator is more interesting. The covariant structure (D A)2 describes
minimal coupling to the constant vector potential where the unfamiliarly looking structure
of the derivative operator D is a consequence of the fact that our particle lives on a curved
manifold. Indeed, it is straightforward to verify that for A = 0

J 1 i J i ,
DD=
i=x,y
where J (x, y), specified in Eq. (C.2), is the square root of the determinant of the metric
tensor on GL(1|1). This structure identifies D D as the radial part of the Laplace operator
on GL(1|1). (Radial, because the two coordinates x and y, spanning a maximally
commutative sub-algebra of the Lie algebra of GL(1|1).)
Summarizing, we have identified H as the Hamiltonian of a charged particle on the
group manifold GL(1|1). Our next task will be to compute its Green functions YL/R . We
begin by considering the case of a free particle, V = 0 = 0. The solution of this
problem will contain the information needed to compute the conductance.
5. Conductance
In this section the general formalism developed above is employed to compute the
conductance g of the system. Our main objective will be to understand the impact of
topology on the localization behaviour of the system. However, before embarking on
526
this analysis it is tempting to digress for a moment and to briefly consider the transport
behaviour of short systems, specifically the aforementioned anomalous sensitivity to the
coupling to the leads. Being not directly related to the mainstream of the paper, the
technicalities of this discussion have been deferred to Appendix D and we here restrict
ourselves to a summary of results.
5.1. Digression: conductance of short wires
In the present context, the phrase short means that systems in the quantum dot regime
L < /(M ) are considered: the conductance is not so much determined by the bulk
transport properties of the system but rather by the strength of the coupling to the leads.
Moreover, and this is a speciality of the sublattice system, the parity of the coupling, i.e.,
the even- or oddness of the connecting sites, turns out to play a crucial role. More precisely
we find that (a) for systems where at both ends a number of sites of alternating parity are
connected the setup considered in much of this paper the short system conductance
is given by
M
,
2
in accord with the behaviour of conventional quantum dots. The same result obtains (b) for
systems where only one site at either end is coupled and these sites have opposite parity
(even/odd or odd/even). However, (c) for single site coupling with even/even or odd/odd
connectors, the conductance vanishes. To heuristically understand the phenomenon, it is
instructive to consider the toy-model case of a strictly one-dimensional clean sublattice
system. For zero energy, the wave length of current carrying excitations is commensurable
with the lattice spacing. This means that the relevant quantum wave functions have nodes
at alternating sites. E.g., for an eveneven configuration a state entering from the right
has zero quantum amplitude at the exit site on the right. This implies a total blockade
of electric current. A less obvious fact is that this phenomenon survives generalization to
many channels and disorder.
We repeat that all these results are obtained for short systems; field fluctuations,
describing the propagation of spatially non-uniform diffusive excitations, are neglected.
An interesting question, not considered in this paper, is how such modes would affect the
conductance as the system size is increased.
g=
5.2. Reduction of the problem to (0-dimensional) quantum mechanics on GL(1|1)

We next turn back to the discussion of large systems (equilibrated coupling of type (a)
understood). Inspection of the basic expression (15) shows that the problem factorizes
into doing the boundary integrals and analysing the bulk behaviour of the heat kernel,
respectively.
We begin by discussing the boundary regions. Following a line of arguments developed
in Ref. [32], we first notice that due to the exponential weights exp( M
2 str(T +
T 1 )), the integrands are confined to the immediate vicinity of the group origin. This
527
suggests to write T = exp W, and do the integrals over generators W in a Gaussian
approximation. Setting W = u iv and expanding in coefficients we arrive at integrals
of the structure,

M
2
2
g = const. du dv d d e 4 (u +v +2) F T (u, v, , ) ,
where the symbol F (T ), represents the functional dependence of the heat kernel on the
boundary field. Evaluation of the Gaussian superintegral leads to
g = const.

1
F (T )T =1 .
M
Doing the same procedure for the second boundary integral and fixing factors we obtain

g = YL (T , T , L)T =T =1 = YL T T 1 , L T =T =1 ,
where the second expression contains the one-argument heat kernel introduced in Eq. (16).
According to this expression, the conductance is obtained by second order expansion of
the heat kernel around the origin. We next note that due to the invariance property (17) the
expansion starts as

2

2

= eW
+ .
) + c2 str W
+ c3 str W
, L = 1 + c1 str(W
YL T
(22)
Now, in our case,

= ln T T 1 = ln exp
W

exp

0

=

1
2

.
Substitution of this expression into (22) and differentiation leads to

g = 2c2 ,
(23)
i.e., the problem has been reduced to fixing the coefficients of the series expansion of
YL . Notice that this series representation is totally determined by the features of the bulk
system; all aspects related to the coupling to the leads have disappeared from the problem.
We now have to face up to the principal task, the calculation of the heat kernel. Its
interpretation as the Green function of the problem suggests to begin by representing YL
through a formal spectral decomposition: consider Eq. (18) at zero potential, (t 4(D
A)2 )YL (x, y, t) = 0, and suppose we had managed to find a set of eigenfunctions,
4(D A)2 n(r) = n n(r) .
(24)
Assuming completeness, we can then span the heat kernel as

cn n(r) (x, y)en t ,
YL (x, y, t) =
n
where the expansion coefficients are determined through the initial condition (21).
Provided the spectrum is suitably structured (positive and gapped against some low lying
levels), and keeping in mind that we are interested in asymptotically large values of t, the
sum may be restricted to a limited set of ns. The problem thus reduces to (a) exploring the
low-lying spectral content of the operator D A, and (b) fixing the expansion coefficients.
528
5.3. Spectrum and eigenfunctions of the kinetic energy operator on GL(1|1)

In spite of the unfamiliarly looking coordinate representation of (D A)2 analytic
progress with the problem is straightforward, the reason being that this operator is nothing
but a plane wave operator in disguise. To make the hidden simplicity of the problem
manifest, let us first remove the dependence of the differential operator on the (pure
N+f
n (x, y) brings the
gauge) potential A: transformation n (x, y) e 2 (xiy)n (x, y)
eigenvalue equation (24) into the form
n = n
n ,
4
where = DD is the radial part of the Laplacian on GL(1|1). Notice that the structure
of this equation does not imply that the vector potential has disappeared from the
problem. It has merely been transferred from the differential operator to the boundary
conditions attached to the differential equation. While irrelevant in the noncompact sector
of the theory, changes of the boundary conditions in the compact sector generally cause
qualitative effects. To appreciate this point, notice that the un-gauged Hilbert space of the
problem has periodic boundary conditions, n (x, y) = n (x, y + 2). Yet, after the gauge
n (x, y + 2)()N+f , i.e., for N odd or nonzero staggering
n (x, y) =
transformation,
a transmutation to twisted boundary conditions has taken place. Needless to say that this
change bears consequences for the spectral structure of the problem and, therefore, for the
transport behaviour of the system.
To make further progress, we subject the eigenvalue problem to the similarity
transformation

,
J 1/2
= x2 + y2 .
J 1/2J 1/2
This change of representation entails an enormous simplification of the problem. The
uninvitingly looking operator has become a flat two-dimensional Laplace operator [33].
However, notice that the transformation by J 1/2 effects yet another change in the boundary
conditions: due to J 1/2(x, 0) = J 1/2(x, 2), the transformed states are subject to the
condition
n (x, y + 2)()N+f +1 .
n (x, y) =
At this point, the solution of the eigenvalue problem has become a triviality. The equation

n (x, y),
n (x, y) = n
4 x2 + y2
k,l (x, y) = eipl x+ipk y , where (k, l) n are two quantum
is solved by the exponentials
numbers, pk,l the associated momenta and the eigenvalues kl = 4(pk2 + pl2 ). From these
states we obtain our un-gauged and un-transformed original wave functions as

x iy (ipl N+f )x+i(pk + N+f )y
2
2
.
kl (x, y) = sinh
(25)
e
2
To give this set of solutions some meaning, we need to specify the range of permissible
ks and ls. In the compact sector the situation is clear the circular boundary condition
529
specified above enforces pk = k N+f

2 , with half integer k. The conditions to be imposed
in the noncompact sector are tightly linked to the integrability properties of our wave
functions:
The space of radial functions on GL(1|1) is endowed with a natural scalar product, viz.

f, g
2
dx
dy J (x, y)f (x, y)g(x, y).
(26)
We demand that the eigenfunctions contributing to the spectral decomposition of the heat
kernel be square integrable w.r.t. , . Inspection of Eq. (25) shows that this requirement
enforces pl = l i N+f
2 , where l may be arbitrary and real.
Finally, we need to specify a set of functions sufficiently complete to generate an
expansion of the heat kernel. The present problem does not come with a natural Hermitian
or symmetric structure (i.e., for finite A the kinetic energy neither symmetric nor
kl = k,l it is
Hermitian). However, defining a fake complex conjugation through
straightforward to show that
kl , k l = (2)2 (l l )kk .

(27)
We may thus attempt to represent sufficiently well behaved (for the cautious formulation,
see below) functions as

kl , g.
gkl = (2)2
dl gkl kl (x, y),
g(x, y) =
(28)
k
Before applying this procedure to the heat kernel, let us summarize our main findings for
clarity: the radial Laplacian on GL(1|1) is diagonalized by the set of functions

x iy i(lx+ky)
kl (x, y) = sinh
(29)
,
e
2
where k Z + 1/2, l real and the eigenvalues are given by

N +f 2
N +f 2
kl = 4 k
+ l i
.
2
2
(30)
(Notice that the appearance of an imaginary part proportional to the strength of the vector
potential is due to the fact that for finite A, the Hamiltonian of the theory is neither
symmetric nor Hermitian.) The spectral decomposition of radial functions is defined
through Eqs. (26), (27) and (28).
5.4. Computation of the conductance
We now apply the machinery developed in the last section to the analysis of the heat
kernel. In principle, the strategy seems to be prescribed by what was said above. We should
, , from
determine the Fourier coefficients of the initial configuration (21),
=
kl lkt
2
kl (x, y).
where the heat kernel would follow as YL (x, y, t) = (2)
kl
k dl kl e
530
There is a problem with this procedure, viz. the expansion coefficients of the distribution (x, y) vanish. Indeed one verifies that in the limit t 0, the support of the
Gaussian in (21) shrinks to zero while its maximum remains limited by one. This readily
implies , = 0. The reason for this pathological behaviour is that our radial theory
memorizes that it derived from a supersymmetric parent theory. In a sense, supersymmetry
can be interpreted as a theory on a zero-dimensional background, i.e., there is no singular
volume factor compensating for the vanishing support of the -function, as would be the
case in spaces with positive dimension.
The problem can be circumvented by a cute trick [32]. Instead of Fourier expanding YL ,
we consider the function YL 1. Since unity by itself solves the heat equation, no harm has
been done and all that has changed is the boundary condition: limt 0 (YL (x, y, t) 1) =
(x, y) 1, a function that equals minus unity almost everywhere save for the origin where
it vanishes.
We thus represent the heat kernel as

2
YL (x, y, r) = 1 (2)
(31)
dl 1kl kl (x, y)ent ,
k
where
kl , 1 =
1kl =

dx
dy
4i
ei(lx+ky)

=
.
l ik
sinh xiy
2
(To obtain the last equality, it is convenient to first do the x-integral. Closure of the
integration contour in the upper/lower complex half plane for positive/negative l yields
a semi-infinite sum over residues of the sinh-function, along with a y-integral that is of
simple plane wave type. Doing sum and integral one obtains the result.)
Substitution of this result into the expansion of the heat kernel now yields

x iy i(lx+ky) kl t
i
1
sinh
e
e
dl
YL (x, y, t) 1 =
l ik
2
k

t
2
l + ik
O (x,y)2 1
2
2
(x iy) + x + y
dl
e kl
4
l ik
k

2 t
2
1
l + ik
e kl ,
=
dl
str(W ) + str W
4
l ik
k
where in the last line we have switched back to a coordinate invariant representation.
Comparison with Eqs. (22) and (23) finally leads to the identification
1
g=
2

dl e
2
2
4L
k N+f
+ li N+f
2
2
kZ +1/2
=
2 4L
1/2

kZ +1/2
N+f 2
4L
k 2
531
where we have inserted the explicit form of the eigenvalues. Notice that all manipulations
leading from the original -model representation to the above Gaussian integral representation have been exact.
We next evaluate this result in the two limiting cases of physical interest, L (Ohmic
regime) and L (localized regime). Beginning with the Ohmic case, we first notice
that for L many terms contribute to the k summation implying that the sum can be
approximated by an integral. Thus,

1/2
Nf 2
4L
L 1
.
=
dk e k 2
g
2 4L
16L
As for any ordinary conductor, g is inversely proportional to the system size; the parity of
the channel number does not play a role.
In the opposite case, L , only those discrete indices that minimize the exponent
contribute to the sum. Specifically, for an even channel number and no staggering,

1/2
L
g
eL/ , N even, f = 0.
4L
In contrast, for N odd and f still zero, the exponent vanishes for the half integer k =
and

1/2
L 1
g
, N odd, f = 0
2 4L
N
2
depends algebraically on the system size. Finally, it is clear that for non-vanishing
staggering, f = 0, intermediate types of behaviour are realized. E.g., an N even chain
with staggering f = 1 behaves like a symmetric N odd chain, etc.
5.5. The role of the Gade term
Before leaving this section let us briefly discuss the role of the, so far neglected, Gade
operator SGade [T ]. The inclusion of this term in the derivation of the heat equation is
= x2 + y2 gets replaced by
straightforward. As a result, the planar Laplacian
= x2 + y2 + (
x iy )2 ,

1 1. This operator is still diagonalized by the plane waves
where 16C
N
discussed above. The eigenvalues change to

N +f 2
N +f 2
.
k
+ (1 + )
li
kl = 4 (1 )
2
2
Recapitulating the computation of the conductance, one finds that the small dilatation
introduced by finite values of does not affect the long range transport behaviour of the
system.
All this is compatible with the structure of the DMPK transfer matrix approach to the
problem. As mentioned above, the DMPK evolution equation is controlled by two coupling
constants. One of these, in Ref. [10] denoted by , is small, O(M 1 ) and becomes
532
inessential in the limit of a large number of channels. The analysis above suggests that
the coupling constant of the Gade operator, ,
and the of the DMPK approach play
the same role. This analogy is supported by the above discussed geometric structure of the
target spaces of the two theories.
6. Density of states
We now turn our attention to the low energy density of states of the sublattice system. As
with the conductance, we will first consider the behaviour of short wires, and then discuss
the localized regime.
6.1. Density of states of short wires
As in Section 5.1 we consider a short sublattice wire of length L < /(M ). The
spectrum of such systems exhibits structure on the scale of the mean level spacing. In
order not to blur these fine structures, the coupling to the external leads will be switched
off throughout this section. Following
the logics of Section 5.1, one would then conclude
that the action reduces to S[T ] = dr Lz [T ] = i 2s str(T + T 1 ), where the matrix T
1 is the level spacing of
parameterizes a zero mode configuration, s =
and = (L)
the isolated system. (Temporarily deviating from the convention of the rest of the paper,
in this section s measures the energy in units of the total level spacing and not the level
spacing of a localization volume.) This presumption is almost but not quite correct. The
point is that our so far discussion of the low energy action implicitly assumed that the
number of sites of the system is even. In the opposite case, an extra contribution Stop,2 [T ] =
N str ln(T (L)), derived and discussed in Section 3, appears. The structure of this term
reflects the fact that due to the presence of one uncompensated site the global GL(1|1)invariance of the model is lost. (Stop,2 [T ] is not invariant under T TL T TR even for
constant TL,R .) In the majority of cases this extra term is of little interest. However for the
spectral properties of a short isolated system, the presence of Stop,2 bears crucial effects.
Indeed we will see that this term is responsible for the formation of staggering phenomena
akin, and probably related to the effects discussed earlier in Section 5.1.
Adding the two contributions to the action and using Eq. 9) we obtain

1
1 i 2s (T +T 1 )N str ln(T )
,
e
dT T11 + T11
() =
2
for the -model representation of the zero-dimensional DoS. The task thus is to integrate
over a single copy of the target manifold GL(1|1). For sufficiently small energies ,
the action is not large enough to confine the integrand to the origin of the group manifold
implying that the integral has to be done nonperturbatively. Referring to Ref. [5] and
Appendix D for technical details we here merely display the final result of this integration
procedure,

s 2
J (s) + J12 (s) , L even,
(s) =
2 0
533
Fig. 2. Density of states of the short sublattice system. Solid line: DoS of a system with L even.
Dashed line: DoS of a system with L odd and N = 3. (The singular zero energy spike causing the
spectral depletion up to values s 3 is not shown.)

s 2
JN (s) JN+1 (s)JN1 (s) , L odd.
(32)
2
The structure of these DoS profiles is shown in Fig. 2 for the example N = 3.
Eqs. (32) have been obtained earlier within pure random matrix theory [19], and its
supersymmetry implementation [21,35]. Studies of chiral random matrix ensembles were
largely motivated by the relevance of the spectral structure of effectively zero-dimensional
chiral systems in finite size lattice QCD. More generally, microgaps of the type shown in
Fig. 1 are an omnipresent side effect seen in the spectrum of generic chiral systems with
finite mean level spacing .
Qualitatively, origin and structure of these gaps can easily be understood. First, the chiral
symmetry [H, 3 ]+ = 0 entails that for any nonvanishing energy level n its negative n
is an eigenvalue, too. Disorder generated level repulsion prevents these states from coming
close to each other (on the scale of the mean level spacing) which explains the presence
of the spectral gap in the L-even case. For L odd, this picture has to be modified. To
understand what is happening, let us imagine the Hamiltonian as a block off-diagonal
matrix in sublattice space:

Z
odd,
H=
even.
Z
(s) = N() +
Since L is odd, the number of odd sites exceeds the number of even sides by one, i.e.,
the blocks Z are rectangular with N(L 1)/2 rows and N(L + 1)/2 columns. Now, any
block-off diagonal matrix with k rows and l columns has |k l| eigenvalues 0. Applied
to our system, this means that the L odd system has N zero modes for any realization
of disorder (the -function contribution in (32)). Other levels repel from this concentrated
accumulation of spectral weight which explains the bathtub type suppression of the DoS
up to energies N.
Summarizing we have seen that the L even/odd staggering behaviour observed earlier in
connection with the conductance pertains to spectral properties. Above we had argued that
534
the vanishing of the conductance in the L even case was due to the peculiar spatial profile
of zero energy wave functions. In view of (32) it is tempting to relate the same effect to the
vanishing of spectral weight at zero energy, although we have not analysed this picture any
further.
Keeping in mind the tendency towards buildup of zero energy spectral weight in systems
with mis-matched sublattice structure we next turn back to the analysis of large systems.
6.2. Heat kernel for finite energies
The aim of this section is to compute the DoS of a large system with L . To
facilitate comparison with the behaviour of the conductance we will drop the assumption of
isolatedness and again couple the system to leads. Notice that this stands complementary
to the analysis of Ref. [11], where a closed system was discussed. That we will obtain
identical scaling of the DoS is proof of the (in view of the existence of topological zero
modes not entirely obvious) assertion [11] that boundary conditions do not affect the bulk
spectrum.
We will not be able to compute the DoS () for arbitrary . Instead, an asymptotic
expression valid for low energies will be derived. Remembering the behaviour found
for small systems we anticipate nonanalytic behaviour, () lnn (||)||, where the
logarithmic factor is due to potentially existent localization corrections to the zero mode
behaviour. Our objective is to identify the most singular contribution of this type.
Starting point of the analysis is the transfer matrix representation (14) of (). To
evaluate this expression we need to compute the functions Y L/R , now for finite potential
V . We will do this following a procedure developed in Ref. [36]. Starting out from a
spectral decomposition of the type (31), we first notice that only zero energy eigenfunctions
contribute to long distance behaviour of YL/R (r). This means that our time dependent
Schrdinger equation (18) can effectively be replaced by the stationary form

t :
(D A)2 + cosh(x) cos(y) YL/R (x, y) = 0,
(33)
where we have substituted the explicit form of the potential and omitted the spatial
argument of YL/R for simplicity. For later convenience we have also analytically continued
from real to imaginary energy arguments, is + 4 > 0. Substitution of these functions
into Eq. (14) yields the reduced representation

cosh(x) cos(y)
1
(E) = 0 1 + Re
(34)
(x,
y)Y
(x,
y)
.
dx dy
Y

L
R
2
sinh2 xiy
2
To obtain this equation we have expanded the pre-exponential sources in Grassmann
variables and integrated over these. The constant contribution 0 appears as a consequence
of the EfetovWegner theorem. (The representation of the pre-exponential term in polar
coordinates contains a purely non-Grassmann contribution. Integration over a term of this
type obtains the integrand at the origin [31] which, in our case, equals unity.)
We now turn to the actual computation of the functions YL/R . The first step is the
derivation of a set of matching, or boundary conditions relating the heat kernel to its 0
535
asymptotics. To this end we evaluate the = 0 spectral decomposition (31) in the limit
t . Neglecting contributions that decay exponentially in t/ , it is straightforward to
obtain the asymptotic expressions:
0
Y L (x, y) 1,
N even,
0 1
Y L (x, y) (e(xiy) + 1), N odd,
2
R
where we have temporarily neglected the staggering parameter f . The structure of these
functions will be motivated shortly. Turning to the case = 0, we next gauge and transform
Eq. (33) as in Section 5.3. Transformation
Y L (x, y) J 1/2 (x, y)e

R
1PN
2
(xiy)
L (x, y),
Y L (x, y) Y
R
where PN
brings the equation into the form
2

L/R (x, y) = 0,
x + y2 cosh(x) cos(y) Y
(35)
while the transformed 0 asymptotics read

xiy

|x| 1
0
L/R (x, y)
Y
sinh1 xiy
2 sgn(x)e sgn x 2 , N even,
2

|x| 1
0
L/R (x, y)
Y
coth xiy
sgn(x),
N odd.
2
(36)
(1 + ()N )/2
Here we have anticipated that the dominant contribution to the above double integral
representation of the DoS will come from large values of the noncompact variable x.
(x, y) =
Eqs. (35) and (36) have the nice feature of full separability. Writing Y
N Y1 (x)Y2 (y), where N is a normalization factor and the subscript L/R has been dropped
for simplicity, Eq. (35) can be traded for the set of decoupled ordinary differential equations

PN
Y1 (x) = 0 + O(),
x2 ex
2
4

PN
y2 +
Y2 (y) = 0 + O().
4
y
The second equation is trivially solved by Y2 (y) ei PN 2 . The first line is a Bessel
equation. Its two solutions are given by IPN ((2)1/2e|x|/2 ) and KPN ((2)1/2e|x|/2).
Discarding the exponentially divergent solutions I and using that for small arguments,
K0 (z) ln(z) and K1 (z) z1 + 2z ln( 2z ), the normalization constants N can now be
fixed by matching to the zero energy asymptotics in the limit of small . This obtains the
approximate solution

|x| 1
y
L/R (x, y) (8)1/2K1 (2)1/2e|x|/2 ei sgn x 2 , N even,
Y

|x| 1
L/R (x, y) 2 K0 (2)1/2e|x|/2 ,
N odd.
Y
ln
It is now a straightforward matter to substitute this expression back into the above integral
representation for and to integrate over coordinates. The y-integration, extending over a
536
purely harmonic integrand, is trivially done. (Notice that the sinh2 -factor in (34) cancels
against the factor (J 1/2)2 from the similarity transformation.) As for the x-integration,
we note that due to the exponentially decaying asymptotics K (z) exp(z/2) for
|z| 1, the integral can be cut off at (2)1/2e|x|/2 1 |x| ln(2). Within the
domain of integration, the Bessel functions can be replaced by the small argument
asymptotics specified above. Substituting these expressions it is then straightforward to
obtain () 0 Re ln2 (even N ) and () 0 Re( ln2 )1 (odd N ) for the small
-asymptotics of the DoS. Analytic continuation back to real energies finally leads to the
result
(s) 0 |s| ln |s|, N even,
0
, N odd,
(s)
|s| ln3 |s|
(37)
for the low energy behaviour of the DoS. Eqs. (37) agree with the results found earlier in
Ref. [11].
We finally discuss the extension of the above results to non-zero staggering. For
nonvanishing f and even N , the large argument asymptotics (36) generalize to
L/R (x, y) 0,x 1
2e 2 (f +1)(xiy).
Y
1
(38)
One can now follow the same steps as in the nonstaggered cases above to obtain the result

(|f |) 22(1|f |)
= 20
(39)
cos
1 |f | |s|1|f | .
(2 + |f |)
|f |
2
For nonzero f , the DoS vanishes in a more singular manner as in the nonstaggered case.
This behaviour provokes the question, how matching with the diverging profile in the case
N odd, f = 0 might be obtained. The principal structure of the theory entails that (N
even/f = 1) should be equivalent to (N odd/f = 0). On the other hand, the f " 1 version
of Eq. (39) certainly does not agree with the divergent result (37).
To resolve this paradox, it is helpful to re-interpret the asymptotic expressions (36) and
(38) within the quantum mechanical picture of the theory. Focusing on the compact sector
and temporarily ignoring the factor J 1/2 from the transformation to a flat Laplacian, these
functions acquire the meaning of ground state wave functions 0 of a one-dimensional ring
subject to a gauge flux N+f
2 . For N even and zero f , an integer number of flux quanta
pierce the ring, and the ground state wave function carries zero momentum, 0 (y) 1
which, after multiplication with J 1/2 leads to the first line of Eq. (36). In contrast, for
N odd and f still zero, a half integer flux pierces the ring. This is a special situation
in the sense that the ground state wave function is two-fold degenerate, i.e., 0 (y) =
c+ eiy/2 + c eiy/2 . Our earlier analysis has fixed the a priori un-determined constants
c to a symmetric configuration. Inclusion of the noncompact variable and multiplication
with J 1/2 then leads to the second line of (36).
We can now understand what happens as f is turned on for an N even configuration:
a flux f/2 is sent through the ring and the ground state wave function remains unique
(cf. Eq. (38)) until f comes close to the critical value 1. In the immediate vicinity of the
537
degeneracy point, the fact must no longer be neglected that our one-dimensional system
is subject to a weak potential cos(y). For values of f such that the level splitting
1 f between the two nearly degenerate levels becomes comparable with the characteristic
strength of the potential , the true ground state configuration is given by the symmetric
superposition of the two levels, as in the N odd case. For these values of f the ground
state configuration is given by the second line of Eq. (36) and the DoS follows the N odd
asymptotics.
This qualitative argument predicts that for asymptotically small energies, the DoS scales
as in Eq. (39). However, for larger values of the energy, |1 f |, a crossover to the
characteristics of the N odd f = 0 DoS profile takes place.
7. Summary
In this paper, transport and spectral properties of weakly disordered quantum sublattice
wires have been explored from a fieldtheoretical perspective. We re-derived results
obtained previously within the DMPK transfer matrix formalism, observed a surprisingly
strong sensitivity of system properties to the realization of the lead/device coupling, and
found that conductance and DoS, at least of short systems, exhibit drastic dependence
on the parity of the total site number in the sublattice chain. It is likely that both this
phenomenon and the dependence of system properties on the parity of the channel number
root in the same origin, i.e., the existence of zero energy states for effectively block offdiagonal Hamiltonians with rectangular (nonquadratic) block structure. Although we are
not aware of an intuitive explanation for the channel number parity effects, this belief is
supported by the observation that all staggering phenomena are controlled by the same
topological term Stop in the action of the -model. From these findings one might expect
that the delocalized band-centre behaviour exhibited by the two-dimensional sublattice
model, is driven by the 2d-analogue of this operator. Curiously, this is not so. In the
two-dimensional field theory, the so-called Gade term SGade , a two-gradient operator
contribution with small and nonuniversal coupling constant, drives the system towards delocalization. The operator Stop does have a generalization to two dimensions [5], but its role
has not been investigated so far. Summarizing we find that comparable phenomenology
(metallic behaviour and diverging DoS) in one and two dimensions is described by different
operators in the -model action. This indicates that some deeper physical principle, not
understood at present, lies beyond the visible structure of the field theory.
Acknowledgement
Discussions with P. Brouwer, J. Chalker, V. Kravtsov, A. Ludwig, C. Mudry, B. Simons,
and J. Verbaarschot are gratefully acknowledged. Special thanks to M. Zirnbauer for
discussions, numerous supportive hints, and for drawing our attention to the most important
aspect of the field theory, the topological term. This work was partly supported by
Sonderforschungbereich 237 of the Deutsche Forschungsgemeinschaft.
538
Appendix A. Field integral formulation

-model representations of zero- and two-dimensional systems with AIII-symmetry
have been constructed in different contexts before [2,5,9,21]. That the following two
sections discuss the construction of the field theory in some detail is motivated by
nongeneric features particular to the 1d-system, most notably the coupling operators
and the existence of topological structures. In the present section, we will derive a
representation of the model and its correlations functions in terms of a supersymmetric
field integral. The projection of this, a priori exact representation onto its low energy sector
will be the subject of the subsequent Appendix B.
Consider the two correlation functions
+

(1)
C
,
DoS,
G E

(2)
+
C
G 0 , conductance,
G 0
relevant for the computation of DoS and conductance, respectively. To compute these
objects, we follow the now standard supersymmetry scheme for disordered electronic
systems [31] and represent the Green functions as

G (z) = b
Z[J ] = f
Z[J],
J J=0
J J=0
where
Z[J] =
)ei (zH +i
D(,
W C W CT J)
(A.1)
and we have assumed that sgn Im z > 0. (In the opposite case, the sign of both the total
action and the coupling operator change.) Here = {(S , )T } and = {(S , )} are
two-component superfields where S is the complex conjugate of S , while and are
independent Grassmann variables. The source field
b

0
J
J =
,
0 Jf
b,f
where Jb,f = {J
} are ordinary matrices in site and orbital space.
(1)
From this representation, the one-point correlation function obtains as C
=
+
J b |J=0 Z[J ] for z = E . However, at first sight it looks like two Gaussian field in
tegrals (A.1) are needed to compute the two-particle correlator C (2) , one for the Green
function G(0+ ) the other for G(0 ). Fortunately, this is not so, a direct consequence of the
]+ = 0 implies that
chiral symmetry of the Hamiltonian: the relation [3 , H
= 3 G(z)3
G(z)
and thus G (0+ ) = ()l+k+1 G (0 ), where l and k are the site indices carried by the
composite variables and , respectively. In other words, the retarded and the advanced
Green function are not independent and we can obtain the correlation function C (2) as

2

(2)

C = ()l +k
Z[J]
b f
J J J=0
539
from the comparatively simple generating functional of the one-point function (evaluated
for z = 0+ ).
Appendix B. Derivation of the field theory
In this appendix we derive the effective Lagrangian (10) from the basic representation (A.1). To keep the discussion simple, we will suppress the source-field dependence
of the partition function in much of what follows. The final results for the correlation functions C (1,2) , obtained by straightforward expansion of the action to first and second order
in J are displayed in the final Eq. (B.1).
The derivation of the effective action essentially follows the standard [31] construction
route of field theories of disordered electronic systems. We begin by averaging the partition
function over the Gaussian distribution of the random hopping matrices R:
)ei (zH0 +i C W C W CT ) N l str(l l (l+1) (l+1) ) ,

Z[0] = D(,
0 = {tll } is the clean part of the Hamiltonian. Next, the quartic contribution is
where H
decoupled by means of two auxiliary fields:

2
2
)ei (zH0 +i C W C W CT )
Z[0] = D(Q, P )eN l str(Ql,l+1 +Pl,l+1 ) D(,
ei
+
+
l,even l [Ql,l+1 +Ql,l1 ]l i

l,odd l [Ql,l+1 +Ql,l1 ]l
where Q = Q iP , and Q and P are two-component supermatrix fields (reflecting the

two-component matrix structure of the dyadic products ) living on the non-directed
links of our system. Both, internal structure and symmetry properties of these fields will
be discussed momentarily. At this stage, we merely anticipate that the field configurations
relevant to the long range behaviour of the system will be smooth. The structure of the
1
action then suggests to define Q
l 2 (Ql,l+1 + Ql,l1 ) as a new field variable, which
now again lives on the sites of our system.
To account for the staggered even/odd structure of the theory we next double the unit
cell and define a two component field

2j +1
1,j
j =
.
2j
2,j
Here we have introduced a new counting index j = 0, . . . , L/2 enumerating the doubled
unit cells of our system. To avoid confusion, we will systematically designate the index
0, . . . , L of the primitive sites by l, l , . . . and the new index by j, j , . . . . Expressed in
terms of the functional integral assumes the form

2N j str(Q2j +Pj2 )
, )
Z[0] = D(Q, P )e
D(

z + 2(Q + iP 3 )
exp i

+ i
W C W CT + H012 H021 .
C=L,R
540
Here, = (1 i2 )/2 where i , i = 1, 2, 3 are Pauli matrices acting in the space defined
through the two-component structure of . In lack of better terminology, we will refer
to this space as the chiral space. The second line of the expression above, purely off
diagonal in the chiral space, contains the clean Hamiltonian. The explicit lattice structure
of the blocks H021 = (H012 ) is given by
12
H0,ll
= ll (1 + a) + ll 1 (1 a),
21
H0,ll
= ll (1 + a) + ll +1 (1 a),
where a is the staggering parameter introduced in (2). Finally, notice that the chiral matrix
structure of the coupling operator is given by

C
CT
W2j +1 W2j
C CT
+1
.
Wj Wj =
C W CT
W2j
2j
At this stage, the superfield can be integrated out and we arrive at
2
2
Z[0] = D(Q, P )e2N j str(Qj +Pj )

exp N str ln z + i
W C W CT + 2(Q + iP 3 )
C
+ H012 H021

.
To make further progress, we subject the functional integral to a saddle point analysis.
and P that
Following the standard scheme [31], we seek for saddle point configurations Q
are matrix-diagonal and spatially uniform. Further, we temporarily set a = W = Re z = 0
and neglect boundary effects due to the finite extent of the system. Making the ansatz
= iq 1, P = p 1, where q and p are complex numbers and 1 is the two-dimensional
Q
and P then generates the set of
unit matrix in superspace, a variation of the action w.r.t. Q
equations
1
i 1
tr G0 + 2i(q + p3 ) jj ,
2
1
i 1
p = tr G0 + 2i(q + p3 ) jj 3 ,
2
q=
where G0 (i + + H012 + H021)1 |a=0 and the trace extends over the two chiral
components of the operators (but not over superspace). To evaluate the trace, we Fourier
transform to momentum space. Defining the transform through
1/2
2
ei2kj fj ,
f (k) =
L
j
1/2
2
ei2kj f (k)
fj =
L
k
(where the factor of two in the exponent serves as a mnemonic indicating that we have
doubled the unit cell of our system) one finds that the Green function G0 is diagonal
541
in momentum space with G0 (k) = (i + (1 + e2ik )+ + (1 + e2ik ) )1 . It is now

straightforward to verify that our equations are solved by p = 0 and q determined through
the self consistency equation
1 = 22
2
1
.
L
(2q)2 + 2(1 cos(2k))
k
L
Replacing the sum by an integral, k 2
0 dk, one finds that the equation is solved
by q = /2.
= i 1 is but a
The presence of the chiral symmetry implies that the configuration Q
2
particular representative of a whole manifold of solutions. To explore the morphology
of that manifold, we notice that our restricted (z = W = 0) action is invariant under the
transformation
1 T 1 ,
1
2
2 T2 ,
1 T21 1 ,
2 T 1 2 ,
Q + iP T1 (Q + iP )T2 ,
Q iP T21 (Q iP )T11 ,
where T1 , T2 GL(1|1). This symmetry, the super-generalization of the fermion symmetry (7), states that the model has GL(1|1) GL(1|1) as a global invariance group. Applying
the transformation to the diagonal stationary phase solution discussed above we find that
i
i
1
2
2
T
T 1
where T = T1 T2 . Arguing in reverse, we conclude that any matrix T defines a solution

of the mean field equations, i.e., GL(1|1) is the Goldstone mode manifold of the model.
This is the celebrated mechanism of chiral symmetry breaking (see Ref. [14] for review):
field theory implementations of models with a discrete chiral symmetry on the microscopic
level, possess continuous factor groups G G as symmetry manifolds. (In our case G =
GL(1|1).) This symmetry is spontaneously broken by the saddle point configurations of
the model. What remains is a Goldstone mode isomorphic to a single factor G.
Combining these results, we parameterize our field manifold as

Q + iP
Q iP
i
2
UT
T 1 U
where both T , U GL(1|1). Here, the matrices T span the Goldstone model manifold
whereas the U s, incompatible with the global symmetry of the model, represent massive
modes. The next logical step in the construction of the field theory is to substitute these
configurations back into the action and to expand in (i) energy arguments z and matrix
elements W , (ii) long-ranged spatial fluctuations of the Goldstone modes, and (iii) massive
modes.
542
B.1. Goldstone mode fluctuations

We begin with the second element of the program formulated above, the expansion of
the action in long ranged spatial fluctuations of the Goldstone mode. Temporarily setting
z = W = 0, U = 1, we re-organize the str ln of the action according to
2

i T
H012
H012
i2
X N str ln
= N str ln
,
H021 i2 T 1
T H021T 1 i2
where T is a slowly fluctuating field of Goldstone modes. Writing T H021 T 1 = H021 +
T [H021, T 1 ] and noticing that, due to the slow fluctuation of T , the commutator is small,
we expand as
12 21 1 N 12 21 1 12 21 1
T H ,T
T H ,T
T H ,T
str G
G
+ ,
X = N str G
0
0
0
2
0 = (G0 + i)1 and the ellipses stand for infrared irrelevant
where we have defined G
higher order commutator terms. One next Fourier transforms these expressions, substitutes
the explicit momentum representation G0 (k) and uses that the characteristic momentum q
carried by the transforms T (q) is small. The subsequent integral over the fast momentum
k is most economically done by noticing that full integration over k amounts to integrating
the characteristic phases exp(i2k) once over the complex unit circle. This integral has
a simple pole inside the integration contour, whose residues depend on the small
momentum q. Expanding the residues to lowest nonvanishing order in q and transforming
back to coordinate space one obtains X = Stop + Sfl , where the two contributions are
L
displayed in Eq. (11) and a continuum limit j 12 0 dr has been taken.

Notice that the above discussion implicitly assumed that the number of sites of our
system is even: the operator X had a structure where each T (living at an odd site) came
with a partner T 1 at the neighbouring even site. For a system with an odd number of
sites L, however, there remains one uncompensated degree of freedom T ((L 1)/2) at
the terminating site. Neglecting gradients, the action due to this extra contribution reads
Stop,2 = N str ln T (L). In principle, this contribution is as relevant as the bulk contribution
to the topological action: it is local in space but, on the other hand, does not contain
derivatives. Indeed, the structure of Stop,2 is closely related to that of Stop as can be
seen by representing the latter as a boundary action (cf. the third line of (13)). Yet in the
majority of cases, the extra contribution Stop,2 does not play much of a role, wherefore we
have largely ignored it in main analysis. E.g., for a system coupled to the outside world,
Stop,2 = N str ln(T (L)) = N2i, evaluates to a phase of no physical effect (cf. the related
discussion around Eq. (13)). There is, however, one exception to that rule, viz. the physics
of isolated short systems with an odd number of sites discussed in Section 6.1, where the
presence of the extra contribution is of key relevance.
B.2. Finite energies and coupling to the leads
To obtain the action associated to finite z and W W T , we organize the action as,
543

z + iW W T + i2 T
H012
H021
z + iW W T + i2 T 1

2

i
(z + iW W T )T 1
H012
+
str ln
(z + iW W T )T
H 21 i2

0
1
G

1 11

jj + z + i W W T
22
T G
Tl G
str z + i W W T
jj
str ln
2j +1
= iz0

j
2j

str T + T 1 +
str W W T 2j +1 Tl1 + W W T 2j Tl ,
2
j
N 22
N
11
where we have used that (from the saddle point equation) = i N
Gjj = i Gjj = and
is the DoS per site evaluated for energies far away from the middle of the band. We next
assume that in the coupling region to the leads, i.e., the region where the envelope function
f defined through (3) is nonvanishing, fluctuations of the Goldstone modes are negligible.
Application of the orthogonality relation (3) then directly leads to

M
str Tj + Tj1
2
j =0,L/2
for the contribution of the coupling term. Combining terms and taking the continuum limit,
we finally obtain the two expressions Sz and ST of Eq. (11) for the contribution of energy
and coupling operators to the effective action, respectively.
B.3. Integration over massive modes
We finally turn to the discussion of the role played by the massive modes U . First, notice
that due to the presence of a weight term exp( N2

str(Q2 + P 2 )) in the action and

N2 2
str U
N str Q2 + P 2 = N str (Q + iP )(Q iP ) =
8
fluctuations of these fields are strongly inhibited. Starting out from an ansatz U = exp iW ,
where W is some generator, we may thus perturbatively expand the action around W = 0.
The actual realization of this program is cumbersome (cf. [5] for a concrete example).
Since the result, an extra Goldstone mode operator weakly coupled to the action, will not
play much of a role in the present analysis we restrict ourselves to a schematic outline of
the calculation.
Perturbative expansion of the action in powers of W obtains an expression like

(1)
(2)
Z[0] = DT eS[T ] eS [T ,W ]+S [T ,W ]+ W

1
(2)
(1)
2
DT eS[T ] eS [T ,W ]W + 2 (S [T ,W ]) W ,

N2
2
where . . .W DW e 8 str(W ) and S (n) [T , W ] denotes the expansion of the action to
n-th order in W . The ellipses stand for contributions of higher order in W which can safely
544
be neglected (due to the large overall factor N 1). It is straightforward to verify that
the explicit evaluation of the operators S (n) [W, T ] obtains contributions of the structure
c1 N str(W ) and c2 N str(W W ), where T T 1 and the coupling constants c1,2
are proportional to powers of . Substituting these expressions back into the action and
performing the Gaussian integration over W , we arrive at

2

(1)
str S [T , W ] W Nc12 str 2 = Nc12 str T T 1 ,
(2)

2

2
str S [T , W ] W c2 str() = c2 str T T 1 .
What can be said about structure and relevance of these operators? First, the two
expressions str( 2 ) and str2 () are the only invariant Goldstone mode operators with
two derivatives. The first of these is already contained in the action (cf. Eq. (11)) with a
coupling constant parametrically larger than the constant Nc12 obtained above. Thus, the
first of the two contributions coming from the massive mode integration is irrelevant. In
contrast, the second contribution must be taken seriously and, after integration over spatial
coordinates, gives the term SGade of Eq. (11).
This completes our derivation of the effective action of the model. Finally, to compute
the correlation functions C (1,2) defined in the text, we have to add the source field J to
the partition function, expand to first or second order and differentiate. The structure of
the resulting expressions depends on the index configuration of the correlation functions.
For the correlators relevant to the computation of conductance and DoS, respectively, we
obtain

T l ,11 ,
l even,
i
2
(1)
C =
1
T l1 , l odd,
2
,11
2
T
T l ,
l even, l even,
l
2 ,12 2 ,21
T 2l ,12 T (l 1) ,21 , l even, l odd,

1
(2)
2

1
C =
(B.1)
T l ,21 , l odd, l even,
T (l1)
,12
1 2

T (l1)
T (l 1) ,21 , l odd, l odd,
2
,12
where the angular brackets stand for the functional average.
Appendix C. Geometry of GL(1|1)

The canonical metric on the supergroup GL(1|1) derives from the differential two-form
str(dT dT 1 ). is not a positive two-form, but its restriction to the sub-manifold
M GL(1|1) defined in the text is; M is the maximally Riemannian subset of GL(1|1).
Parameterizing the group manifold in terms of some coordinates, T = T (x1 , . . . , x4 ) (half
of which are anti-commuting), the metric two-form assumes the form

gij dxi dxj ,
=
ij
545
where {gij } defines the metric tensor (represented in the basis {xi }). As with nonsuper
Riemannian manifolds, the metric tensor determines the geometry of the manifold.
Specifically, the invariant group integral is defined through

dT =
dx1 J (x1 , . . . , x4 ),
i
where the Jacobian J sdetg. Similarly, the Laplacian has the standard structure
1
=
x g
g 1/2 i
ij 1/2
xj ,
where g ij is the inverse of g, gij g j k = ik .

We next wish to represent the metric tensor, the invariant measure, and the Laplacian in
the polar decomposition, T = kak 1 defined in Eq. (12). Due to the manifest invariance of
under transformation T k0 T k01 , k0 a fixed rotation matrix, it is sufficient to evaluate
the metric tensor at k = 1. Substitution of the decomposition T = kak 1 into the two form
then yields the covariant structure

= str d kak 1 d ka 1 k 1 k=1

= str [dk, a] + da dk, a 1 + da 1

= str [dk, a] dk, a 1 + da da 1 .
Using that
dk =

da =
ex dx
ieiy dy
it is straightforward to verify by elementary matrix manipulation that the metric bilinear

form assumes the form
1 0
0 1
g=
2 xiy ,
0
4 sinh ( 2 )
0
+4 sinh2 ( xiy
2 )
where g = {gij }, a vectorial structure dx = (dx, dy, d, d) is understood and empty
blocks are filled with zeros. The associated superdeterminant,
g=
1
2
(4 sinh2 ( xiy
2 ))
From g derives the unit-normalized group integral

dT f (T ) =
dx
with Jacobian
J (x, y) = sinh2
2
dy
d d J (x, y)f (x, y, , ),
(C.1)

x iy
.
2
(C.2)
546
Further, the radial part of the Laplacian reads as

g 1/2 i g 1/2 i .
=
i=x,y
Appendix D. Conductance of short systems

Consider a short sublattice system of length L < /(M ). This is the quantum dot
regime, where the conductance is not Ohmic but rather determined by the coupling of the
system to the leads. For a normal, nonsublattice system, g M , reflecting that each of the
M channels contributes to transport with an efficiency set by the coupling. We here wish
to explore to which extent this behaviour generalizes to the sublattice system. Specifically,
three different cases will be discussed: (a) smoothened coupling where a set of sites at
each end is connected the type of coupling considered in the text, (b) coupling through
a single site on either end where the two terminal sites are of the same parity, odd/odd, say,
(c) one of the two terminating sites is even, the other odd,
That the system is short means that the functional integral is controlled by spatially
uniform configurations T = const.; the stiffness introduced by the gradient term is too
strong to allow for significant fluctuations. (More precisely, fluctuating field configurations
lead to relative corrections of O(LM / ) which we are not going to consider.)
Let us begin by considering case (a). Evaluation of the functional expectation value (8)
for a spatially uniform field configuration leads to the expression

M 2
1
dT T T 1 12 T T 1 21 eM str(T +T ) .
g (a) =
2
We next need to do the group integral. Inspection of the exponent shows that the integration
is dominated by configurations close to the group origin, T = 1. This suggests to represent
our T s as T = exp W and to integrate over the generators W in a Gaussian approximation:

2
(a)
2
g (M )
dW W12 W21 eM str W ,
where we have used that close to the group origin the integration measure is flat. Doing the
integral over the components of W one finds
M
2
in agreement with the behaviour of a nonsublattice system. The situation in case (b) is not
much different. Noticing that fields T sit at the odd sites of our system, while T 1 s are
attached to even sites, we find that the conductance is expressed as

1
(b)
2
g = (M )
dT T12 T 1 21 eM str(T +T ) .
g (a) =
The disappearance of the combination T T 1 in the pre-exponent reflects the fact that
only single sites of different parity are coupled to the continuum. Again the integral is
547
dominated by the group origin and similar reasoning as above leads to

M
M 2
M
2
(b)
dW W12 W21 e 4 str W =
g
.
2
2
In case (c), however, the situation is different. Owing to the identical parity of the
terminating sites, the integral representation of the conductance now assumes the form

2
g (c) = (M )2 dT T12 T21 eM str(T ) .
This is an unpleasant expression: the exponential weight no longer projects onto the group
origin, in fact does not even have a stable saddle point. To compute the conductance we
therefore have to integrate over the full group manifold, a task that is most efficiently done
in polar coordinates. Using Eqs. (12) and (C.1) and integrating out Grassmann components
it is straightforward to verify that the radial part of the integral assumes the form

g
(c)
= (M )
2
dx
0
(ex eiy )2
sinh2 ( xiy
2 )
e(e
x eiy )
= 0.
& dz

1
To understand the vanishing of the integral, notice that dy = 2i
z can be transformed
into the integral of the complex variable z = eiy over the unit circle. One verifies that,
regardless of the value of x, the integrand is analytic and void of singularities inside the
integration contour. Cauchies theorem then implies vanishing of the integral.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
R. Oppermann, F. Wegner, Z. Phys. B 34 (1979) 327.

R. Gade, Nucl. Phys. B 398 (1993) 499.
J. Miller, J. Wang, Phys. Rev. Lett. 76 (1996) 1461.
P.W. Brouwer, C. Mudry, B.D. Simons, A. Altland, Phys. Rev. Lett. 81 (1998) 862.
A. Altland, B.D. Simons, J. Phys. A 32 (1999);
A. Altland, B.D. Simons, Nucl. Phys. B 562 (1999) 445.
T. Fukui, Nucl. Phys. B 562 (1999) 477.
A. Furusaki, Phys. Rev. Lett. 82 (1999) 604.
P. Biswas, P. Cain, R. Rmer, M. Schreiber, cond-mat/0001315.
S. Guruswamy, A. LeClair, A.W.W. Ludwig, Nucl. Phys. B 583 (2000) 475.
P.W. Brouwer, C. Mudry, A. Furusaki, Nucl. Phys. B 565 (2000) 653.
P.W. Brouwer, C. Mudry, A. Furusaki, Phys. Rev. Lett. 84 (2000) 2913.
P.W. Brouwer, C. Mudry, A. Furusaki, Physica E 9 (2000) 333.
M. Favrizio, C. Castellani, cond-mat/0002328.
See J.J.M. Verbaarschot, T. Wettig, hep-ph/0003017, for review.
D.S. Fisher, Phys. Rev. B 50 (1994) 3799.
A.A. Ovchinikov, N. Erikhman, Sov. Phys. JETP 46 (1977) 340.
E. Mller-Hartmann, E. Dagotto, Phys. Rev. B (Rapid Comm.) 54 (1996) 13158.
M.R. Zirnbauer, J. Math. Phys. 37 (1996) 4986.
J.J.M. Verbaarschot, I. Zahed, Phys. Rev. Lett. 70 (1993) 3852.
J.J.M. Verbaarschot, Phys. Rev. Lett. 72 (1994) 2531.
548
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
J.J.M. Verbaarschot, Phys. Lett. B 368 (1996) 137.

P.W. Brouwer, C. Mudry, A. Furusaki, to be published.
G. Theodorou, M. Cohen, Phys. Rev. B 13 (1976) 4597.
T. Eggarter, R. Riedinger, Phys. Rev. B 18 (1978) 569.
K.B. Efetov, Adv. Phys. 32 (1983) 53.
M. Zirnbauer, private communication.
F.D.M. Haldane, Phys. Rev. Lett. 50 (1983) 1153.
I. Affleck, in: E. Brezin, J. Zinn-Justin (Eds.), Fields, Strings, and Critical Phenomena, Les
Houches Lecture Notes, North-Holland, Amsterdam, 1988.
S. Iida, H.A. Weidenmller, J. Zuk, Ann. Phys. (USA) 200 (1990) 219.
C. Mahaux, H.A. Weidenmller, Shell Modell Approach to Nuclear Reactions, North Holland,
Amsterdam, 1969.
K.B. Efetov, Sypersymmetry in Disorder and Chaos, Cambridge Univ. Press, Cambridge, 1997.
A.D. Mirlin, A. Mller-Groeling, M.R. Zirnbauer, Ann. Phys. (NY) 236 (1994) 325.
Note that the simplicity inherent to the radial sector of the heat equation is not specific to the
manifold GL(1|1). In fact, it is a generic feature of all group valued supermanifolds that the
radial components of their Laplacians can be reduced to the form + C, where is flat and C
some constant [34]. In our case C = 0 which is special for GL(n|n).
F.A. Berezin, Introduction to Superanalysis, MPAM, Vol. 9, Reidel, Dordrecht, 1987.
A.V. Andreev, B.D. Simons, N. Taniguchi, Nucl. Phys. B 432 [FS] (1994) 487.
A. Altland, D. Fuchs, Phys. Rev. Lett. 74 (1995) 4269.

Non-abelian spin-singlet quantum Hall states:

wave functions and quasihole state counting
E. Ardonne a , N. Read b , E. Rezayi c , K. Schoutens a
a Institute for Theoretical Physics, Valckenierstraat 65, 1018 XE Amsterdam, The Netherlands
b Department of Physics, Yale University, P.O. Box 208120, New Haven, CT 06520-8120, USA
c Department of Physics, CSU-Los Angeles, Los Angeles, CA 90032, USA
Abstract
We investigate a class of non-abelian spin-singlet (NASS) quantum Hall phases, proposed
previously. The trial ground and quasihole excited states are exact eigenstates of certain (k + 1)body interaction Hamiltonians. The k = 1 cases are the familiar Halperin abelian spin-singlet states.
We present closed-form expressions for the many-body wave functions of the ground states, which
for k > 1 were previously defined only in terms of correlators in specific conformal field theories. The
states contain clusters of k electrons, each cluster having either all spins up, or all spins down. The
ground states are non-degenerate, while the quasihole excitations over these states show characteristic
degeneracies, which give rise to non-abelian braid statistics. Using conformal field theory methods,
we derive counting rules that determine the degeneracies in a spherical geometry. The results are
checked against explicit numerical diagonalization studies for small numbers of particles on the
sphere. 2001 Elsevier Science B.V. All rights reserved.
PACS: 73.43.-f; 71.10.-w; 71.10.Pm
1. Introduction and summary

The observation [13] of a quantum Hall (QH) state at an even-denominator filling
factor, = 5/2, stimulated the development of trial wave functions outside the usual
hierarchy (or later, composite fermion) approach, which generates only odd-denominator
fractions. The 5/2 state is interpreted as half-filling of the first excited Landau level (LL),
the lowest one being filled with electrons of both spins, and can be mapped to half-filling
of the lowest LL, with a suitable Hamiltonian. There are now strong indications that this
state is spin-polarized [4,5], and described [4] by the paired pfaffian state of Moore
and Read (MR) [6], which has filling factor 1/2. This state was originally proposed as
E-mail address: ardonne@science.uva.nl (E. Ardonne).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 2 2 4 - 3
550
E. Ardonne et al. / Nuclear Physics B 607 [FS] (2001) 549576
an example that manifests non-abelian braid statistics of its quasiparticle excitations [6].
Generalizations exist in which the particles are clustered in k-plets (k = 1, 2, 3, . . . ),
but still spin-polarized [7]. In these states, the non-abelian statistics are associated with
parafermion conformal field theories (CFTs).
It is well known that, despite the presence of a strong magnetic field, spin-singlet QH
states are sometimes favored over their spin-polarized counterparts. The possibility to
manipulate the Zeeman splitting by the application hydrostatic pressure and by tilting the
magnetic field has opened the possibility of systematic studies of transitions between spinpolarized and non-polarized QH states (see, e.g., [8]). In this light, states analogous to
the spin-polarized clustered QH states of [7], but with a spin-singlet structure, have been
constructed [9]. In [9], trial wave functions for these non-abelian spin-singlet (NASS) states
have been written in terms of correlators in a CFT describing parafermions associated to
the Lie algebra SU(3). We remark that one may consider an alternative series of NASS
states, whose algebraic structure is related to SO(5) rather than SU(3). The simplest state
of this type is a paired spin-singlet state that exhibits a separation of spin and charge in the
quasihole excitation spectrum [10]. The SO(5)-based NASS states will not be discussed in
this paper.
In the present paper, we study in detail some of the properties of the NASS states,
paying special attention to the case k = 2. We give explicit closed form expressions for
the ground state wave functions, and study the degeneracies of their quasihole excitations.
The degeneracies of states with fixed spins and fixed well-separated positions of the
quasiparticles are the origin of the non-abelian braid statistics.
We first strengthen the case for the existence of the incompressible phases of matter
with the universal properties of the states of Ardonne and Schoutens (AS) [9], by showing
that their trial wave functions for the ground state and for states with quasiholes are exact
zero-energy eigenstates of certain (k + 1)-body interaction Hamiltonians for particles in
a single LL, in a similar way as the spin-polarized cases [7]. The explicit closed-form wave
functions for the ground states are obtained. In the study of the quasihole degeneracies, we
then follow two complementary approaches. The first is an analytical path, which relies
heavily on the formal structure of the associated parafermion CFT, and on the analogy
with earlier studies for spin-polarized non-abelian QH states [1113]. While at present
we lack explicit expressions for the many-body wave functions describing the quasiholes,
we have enough control to derive explicit counting formulas for the degeneracies, for
k = 2, for particles on a sphere. The second approach is a numerical study of the (k + 1)body Hamiltonian for the case k = 2, on the sphere. The numbers of zero-energy states
for each number of electrons N and of quasiholes n considered are in exact agreement
with the analytical derivation. In addition, we study the excitation spectrum of the same
Hamiltonian, and compare the ground state with that of electrons interacting via the lowest
LL Coulomb interaction.
A highlight of the analytical approach in this paper comes in the derivation of the total
degeneracies of quasihole states. In the CFT set up (which will be described in more
detail in Section 2) the QH states are described as conformal blocks of particle and
quasihole operators. The particle operator factorizes as a product of a vertex operator
551
and a parafermion field, and the quasihole operator is the product of a vertex operator
and a so-called spin field of the parafermion CFT. The nontrivial fusion rules of these
spin fields cause a degeneracy of the ground states in the presence of quasiholes at fixed
positions and spins. There is also further degeneracy associated with the positions and spins
of the quasiholes, which is finite on the sphere. The two contributions need to be combined
in the right way. It turns out, as in earlier cases, that the various states which stem from
the non-trivial fusion rules have a different spatial degeneracy. Therefore, we can not just
multiply the two degeneracy factors, but we need to break up the degeneracies due to the
fusion rules. To accomplish this task, and arrive at the final counting formula, we analyze
truncated chiral spectra in the SU(3) parafermion CFT, using the methods of [14].
This paper is organized as follows. In Section 2, we explain in which way CFT is used to
describe QH states, and review the NASS states as correlators. In Section 3, we introduce
the (k + 1)-body interaction, and show that the AS correlators give zero-energy ground
states. In Section 4 we give the explicit ground state wave functions for the NASS states,
and discuss their spin-singlet properties. Section 5 describes the correlators which give the
states with quasiholes present. The derivation of the counting formula is done in Sections 6
to 9, using the method which is outlined above, with Eq. (49) as the final outcome. Explicit
results of this formula for the degeneracy of the ground states in the presence of quasiholes
are given in the same section for several N (the number of electrons) and n (the number
of quasiholes). In Section 10 we present numerical diagonalization studies on a sphere,
finding full agreement with the analytical expression obtained in Section 9, and compare
the states with the ground state of the Coulomb interaction.
2. QHECFT correspondence
In the QHE, following Laughlin [15], trial wave functions have long been used as
paradigms that represent an entire phase of incompressible behavior. This notion was
reviewed in Ref. [7], so in this paper we will concentrate on the properties of trial states
and their position-space wave functions. As explained by MR [6], many QH trial wave
functions of the (2 + 1)-dimensional system can be obtained as conformal blocks (i.e.,
chiral parts of correlation functions or correlators) in a suitable chiral CFT in 2 euclidean
dimensions, as we will briefly explain.
For particles with complex coordinates zj = xj + iyj , j = 1, . . . , N , for N particles, we
(z), and neglect spin temporarily. For the lowest Landau
will use reduced wave functions
level (LLL), the reduced wave function must be holomorphic in the zi s. For particles in the

(z) by exp ( i |zi |2 /4);
plane, the full wave function (z) is recovered by multiplying
we have set the magnetic length to 1. For particles on the sphere [16], the coordinate zi
refers to stereographic projection, and the full wave function is recovered by multiplying by

2
2 (1+N /2) , where N is the total number of magnetic flux quanta through
i (1 +|zi | /4R )
(z) must be a polynomial
the sphere [7,12]. In the latter case, the reduced wave function
of degree no higher than N in each zi , so that the z component of angular momentum
of each particle lies between N /2 and N /2. Note that we use the term particles for
552
the underlying particles, which are either charged bosons or charged fermions (electrons),
because it is convenient to consider cases of either statistics together.
The simplest example of a state with a uniform density (a state of zero total angular
momentum on the sphere [16]) is the Laughlin wave function [15]:

LM (z1 , . . . , zN ) = (zi zj )M ,
(1)
i<j
for a fixed integer M. The filling factor can be defined for a sequence of states as =
limN N/N , where N is identified with the largest power of any zi in the state. For the
Laughlin state it is = 1/M. Note that this function is antisymmetric (describes fermions)
for M odd, and symmetric (describes bosons) for M even.
The Laughlin wave function can be obtained as

LM (z1 , . . . , zN ) = lim (z )a Ve (z1 ) Ve (zN )ei M N (z ) ,
(2)
z
with Ve (z) = exp(i M ) a chiral vertex operator in the c = 1 chiralCFT describing a

single scalar field compactified on a radius R 2 = M. The operator ei M N (z ) brings
in a positive background charge, which guarantees the overall neutrality of the system. The
constant a must be chosen in such a way that the effect of the background charge does not
go to zero in the limit z ; in Eq. (2) we need a = MN 2 . This procedure is simpler
for our purposes than the uniform background charge used in MR [6], though the latter has
the additional feature of reproducing the factors in the unreduced wave function.
Other, more complicated, QH states can be constructed by invoking more complicated
CFTs. The CFT framework guarantees that a number of consistency requirements for
such states are met [6]. The trial wave functions become more meaningful, and the
corresponding phase really exists, when there is a (local) Hamiltonian for which the trial
state is the nondegenerate ground state, and the excitation spectrum (for the same N as
the ground state) has a gap in the thermodynamic limit. Short range 2-body interaction
Hamiltonians with these properties for the Laughlin state were found by Haldane [16], and
3-body interactions for which the MR state is an exact zero energy eigenstate were found
beginning from the work of Ref. [17]. Read and Rezayi (RR) discovered [7] that these
constructions are the first two cases in an infinite sequence, and found the parafermion
states as the exact zero energy eigenstates of (k + 1)-body interactions for all k. Here we
will show similarly that, in the case of particles with spin, the NASS states of Ref. [9]
are exact eigenstates of zero energy for (k + 1)-body spin-independent interactions, with
the Halperin state and the corresponding 2-body interaction [16] as the only case known
previously. First, we recall the construction in Ref. [9], then in the following section we
establish that the wave function defined by a correlator here is a zero-energy eigenstate of
a (k + 1)-body interaction Hamiltonian.
The NASS states proposed in [9] can be viewed as non-abelian generalizations of
the abelian spin-singlet Halperin states labeled as (m + 1, m + 1, m) (see Eq. (6)), or
alternatively, as spin-singlet analogs of the spin-polarized clustered or parafermion
2k
, with M an integer
states of RR [7]. The filling fraction of the NASS states is = 2kM+3
(M is odd when the particles are fermions, even when they are bosons). The wave functions
553
of these states are constructed as conformal blocks in basically the same way as was done
above for the Laughlin state. In the basic case M = 0, the component of the ground state
with any set of N/2 of the particles having spin up, and the remainder spin down, is defined
(up to phases that may be needed when reconstructing the full state from these components)
as the correlator [9]

N
k,M=0
3N 2 /(2k)

NASS = lim (z )
exp i (2 1 ) (z )
z
2 k

B1 z1 B1 zN/2 B2 z1 B2 zN/2 .
(3)
In this equation the particle operators (to avoid confusion, we emphasize that this means
the operators that correspond to the particles in the CFT, not the operators that create actual
particles in the (2 + 1)-dimensional system) are current operators B (z) in an SU(3)k
(i.e., level k) WessZuminoWitten CFT. These currents can be written in terms of two
bosons = (1 , 2 ) and a Gepner parafermion associated to SU(3)k /[U(1)]2 [18].
The currents are labeled by the corresponding roots of SU(3)

B (z) = exp i /
(4)
k (z).
The roots are given by 1 = ( 2, 0), 2 = ( 2/2, 6/2). We note that these two
roots form an SU(2) doublet under an SU(2) subalgebra of SU(3); the embedding of the
subalgebra is isomorphic to that given in terms of 3 3 Hermitian matrices [generators of
SU(3)] by the 2 2 Hermitian blocks at the upper left corner. For such an embedding, there
is also a U(1) subalgebra [generated by hypercharge diag(1, 1, 2); the particles carry
hypercharge 1] that commutes with the SU(2) and corresponds to the particle number. Note
that a (hyper-)charge at infinity is again needed for neutrality.
Working out the vertex-operator part of this correlator, we arrive at the following
factorized form of the NASS state (after multiplication by an additional Laughlin factor
to obtain general M 1 )

k,M z , . . . , z ; z , . . . , z
N/2 1
N/2
NASS 1

= 1 z1 1 zN/2 2 z1 2 zN/2
(2,2,1)
1/ k M

L z ; z .

(5)

zi ; zj
H
i
j
(2,2,1) is one of the Halperin wave functions [19]
The wave function
H

(m ,m ,m) z , . . . , z ; z , . . . , z
H
1
N/2 1
N/2

m
m
m
=
zi zj
zi zj
zi zj .
i<j
i<j
(6)
i,j
The latter give rise to spin-singlet states whenever m = m + 1 [16]. The wave function
Eq. (5) contains a term which is a correlator of parafermions, the explicit form of which
will be found below.
1 We can also obtain the Laughlin factor by using the particle operators (19) and (20) in the correlator, together
with a suitably adjusted background charge.
554
We also mention here that the CFT construction implies that the number of sectors
of edge states (representations of the chiral algebra), and the number of ground states
(conformal blocks with N particle operators inserted) on the torus for N divisible by 2k,
are both given by (k +1)(k +2)(2kM +3)/6, which is an integer. For M = 0, this coincides
with the numbers for SU(3)k current algebra. The filling factor is P /Q = 2k/(2kM + 3),
so if P and Q are defined as being coprime, then the denominator Q = 2kM + 3 unless
2k and 3 have a common factor, that is unless k is divisible by 3, in which case we have
Q = (2kM + 3)/3. The number of ground states on the torus is then always divisible by the
denominator Q of the filling factor, as it should be. We also expect that for some k values
there are ground states on the torus for other N values, as for the MR states [20] and RR
states.
3. Solution of (k + 1)-body Hamiltonian

It is known [16] that the abelian Halperin spin-singlet state is the unique (on the sphere)
exact zero-energy eigenstate of a two-body interaction Hamiltonian. Other than the trivial
case (1, 1, 0), which is two filled Landau levels, where the Hamiltonian in question is zero,
the simplest case is m = 1 in Eq. (6), the (2, 2, 1) state, which corresponds to M = 0 in
Eq. (3) or (5). In the latter case this Hamiltonian is simply a repulsive -function interaction
between any two particles. As in Refs. [7,12], it is simplest to start by generalizing this
M = 0 case. Because higher M values are obtained by multiplying by additional Laughlin
factors, the Hamiltonians for M = 0 can be straightforwardly extended to M > 0 by
extending the range of the (k + 1)-body part, and adding 2-body interactions as needed,
which have the effect of requiring zero-energy states to contain the Laughlin factors. We
will not detail this here, however, see Section 10.
The natural choice of Hamiltonian for M = 0 is to consider the (k + 1)-body -function
as in Ref. [7], but here for particles with spin. The Hamiltonian (acting within the LLL) is

2 (zi1 zi2 ) 2 (zi2 zi3 ) 2 (zik zik+1 ),
H =V
(7)
i1 <i2 <<ik+1
with V > 0. Note that here we have reverted to labeling the particles independently of their
spin. For this Hamiltonian, a state is a zero-energy eigenstate if it vanishes whenever any
k + 1 particles coincide; for this to be satisfied for some nontrivial states, the particles must
be bosons.
We will now show that the correlator as in Eq. (3) is such a zero-energy eigenstate. The
argument we give here makes direct use of the current algebra satisfied by the currents,
and also sheds new light on the previous spinless case of RR, where a slightly different
argument was used. Without loss of generality, we can consider letting the first k + 1
k+1
all
particles, of either spin, come to the same point, say z = 0, that is z11 , z22 , . . . , zk+1
0. In the standard radial quantization scheme for CFT, we can consider the current
operators as acting in a Hilbert space that is built starting from a highest weight state
that in the present case is simply the vacuum |0 of radial quantization about the origin
555
z = 0. As zii tend to 0 one by one, the resulting operator product expansion (ope) contains
no singular terms. This follows from the standard current-algebra opes of the currents,
together with the fact that the roots 1 and 2 do not sum to either 0 or another root (for
simplicity, let us replace these two roots by the natural notation = and , respectively).
Indeed the only nonvanishing term as the zi i tend to zero sequentially is the operator at 0
that corresponds to the state
B1 ,1 B2 ,1 Bk+1 ,1 |0
(8)
in the highest weight representation, and we need to show that this vanishes for all choices
of 1 , . . . , k+1 . Here we have used the modes of the currents,

zn1 Ba,n ,
Ba (z) =
(9)
n
which holds for all generators a of SU(3), not only roots. In fact, the commutation relations
of the affine Lie algebra for these modes also imply that B,1 and B,1 commute, so we
need only to prove
(B,1 )m (B,1 )k+1m |0 = 0,
(10)
for all m in the range 0 m k + 1.

Let us begin by choosing m = 0. Then we need to show that
(B,1 )k+1 |0 = 0.
(11)
But this is simply the pure-current-algebra null-vector equation, which first entered the
physics literature in Refs. [21,22]. Thus this is satisfied in the irreducible, unitary vacuum
highest weight module of the SU(3) affine Lie algebra at level k, and there are similar
equations for all current algebras, and for each integrable highest weight representation.
This already suffices to rederive the RR states, which are related to SU(2) current
algebra [7]. The RR states are in fact the same as those in Eq. (3) above, but with all
spins , so the only root that appears lies in an SU(2) subalgebra (the spin singlet property
of the above states is of course lost when this is done, and the charge at infinity should be
adjusted). Hence, we have very quickly rederived the fact that the RR states for M = 0 are
the zero-energy states of the (k + 1)-body -function Hamiltonian.
It is straightforward to complete the proof for the NASS states. Essentially, we use the
SU(2) symmetry under which B,1 and B,1 form a doublet, and notice that the set of
states labeled by m forms a highest weight multiplet under this algebra, of SU(2) spin k/2.
(This is clear from the 2 + 1 point of view, where we are looking at states of k + 1 spin
1/2 bosons all at the same point.) Then since the highest weight vanishes, all the others do
also, which completes the proof.
As in RR, a similar argument also establishes that quasihole states, written as similar
correlators but with spin fields (primary fields of the SU(3) current algebra) inserted at the
quasihole positions [9], are exact zero-energy eigenstates. These are considered explicitly
in Section 5. Similar arguments also imply that the zero-energy ground states of H on the
torus are given by correlators for some number N of the above fields inserted, with the
number of such ground states (for N divisible by 2k) already given in Section 2.
556
4. Ground state wave function

Based on the results of the previous section, we expect the structure of the trial wave
functions that is, of the chiral correlators (5) of the NASS states to be similar to that
of the RR states, and also to generalize the Halperin (2, 2, 1) state. The RR wave functions
were constructed by dividing the (same-spin) particles into clusters of k, writing down
a product of factors for each pair of clusters, and finally symmetrizing over all ways of
dividing the particles into clusters. Hence in the case with spin, we guess that we should
divide the up particles into groups of k, the downs into groups of k and then multiply
together factors that connect up with up, down with down, or up with down clusters, and
finally ensure that the function is of the correct permutational symmetry type to yield
a spin-singlet state (in particular, it should be symmetric in the coordinates of the up
particles, and also in those of the downs). We expect that the upup and downdown parts
of this should closely resemble the RR wave functions, before the symmetrization; it was
shown in Ref. [7] that the functions found there vanish when k + 1 particles come to the
same point, even inside the sum over permutations that symmetrizes the final function.
These considerations guided the following construction.
Due to the spin-singlet nature of the state, the wave function will be nonzero only if
the number of spin-up and spin-down particles is the same. Furthermore, there must be an
integer number of clusters, so the total number of particles N must be divisible by 2k, and
will be written as N = 2kp, where p N. One example was already given in [9], namely
the wave function for the case k = 2, M = 0 with the number of particles equal to 4 (i.e.,
p = 1),

k=2,M=0 z , z ; z , z = z z z z + z z z z .
(12)
1 2 1 2
1
1
2
2
1
2
2
1
NASS
This is part of the two-dimensional irreducible representation of the permutation group on
4 objects, S4 , as can easily be seen. This is the correct symmetry type to obtain a spinsinglet state, as we discuss further below.
We will now describe the different factors that enter the NASS wave functions. Because
the only effect of M being nonzero is to give an overall Laughlin factor, we will assume at
first that M = 0. First we give the factors that involve particles of the same spin, say spin
up. They are the same as in RR [7]. We will divide the particles into clusters of k in the
simplest way,

z1 , . . . , zk , zk+1 , . . . , z2k , . . . , z(p1)k+1, . . . , zpk ,
(13)
and the same for the z s. (In a more precise treatment, we would say that the first N/2
particles are spin up, the remainder spin down.) We write down factors that connect the ath
with the bth cluster:
z
= z(a1)k+1 z(b1)k+1 z(a1)k+1 z(b1)k+2 z(a1)k+2 z(b1)k+2
a,b
z(a1)k+2 z(b1)k+3 zak zbk zak z(b1)k+1 .

(14)
z
= (za zb )2 . The factors that connect up with down spins
For k = 1, we would write a,b
are simpler:

z ,z
a,b
= z(a1)k+1 z(b1)k+1 z(a1)k+2 z(b1)k+2 zak zbk .
z ,z
557
(15)
For k = 1, the factor would be a,b = (za zb ). We multiply all these factors for all
pairs of clusters, upup, downdown, or updown:
p

a<b
z
a,b
p

c,d
z ,z
c,d
p
z
e,f
.
(16)
e<f
Notice that for k = 1, we do obtain the Halperin (2, 2, 1) wave function.

To obtain a spin-singlet state when the spatial function is combined with the spin state
(which lies in the tensor product of N spins 1/2), some symmetry properties must be
satisfied. For the M = 0 case, the particles are bosons, hence the full wave function must
be invariant under permutations of spins and coordinates of any two particles. This can be
used to obtain the correct form of the function from that component in which, say the first
N/2 are spin up, the rest spin down, as above, so knowledge of that component is sufficient.
The requirement that the full wave function be a spin-singlet can be shown to reduce to the
Fock conditions: the component just defined must be symmetric under permutations of
the coordinates of the up particles, and also of the down particles, and must also obey the
Fock cyclic condition, as given in Ref. [23] (modified in an obvious way for the boson
case). These three conditions can be shown to imply that the spatial wave function is of
a definite permutational symmetry type (belongs to a certain irreducible representation
of the permutation group), that corresponds to the Young diagram with two rows of
N/2 boxes each. In general, given a function of arbitrary symmetry, a Young operator
can be constructed that projects it onto a member of the correct representation (though
the result may vanish); this construction generalizes the familiar symmetrization and
antisymmetrization operations. For the present case, the Young operator is the following
operation, equivalent to summing over the function with various permutations of its
arguments, and some sign changes: first, antisymmetize in z1 , zN/2+1 ; then in z2 , zN/2+2 ;
. . . , zN/2 , zN ; then symmetrize in z1 , . . . , zN/2 ; then finally symmetrize in zN/2+1 , . . . , zN .
This clearly satisfies the first two requirements of Fock, and can be proved to satisfy also
the cyclic condition. It remains to check that it is nonzero, we believe it is. Incidentally,
the application of the Young operator is the analog of symmetrizing over the down spins in
the spatial wave function of the permanent state (see, e.g., Ref. [12]), to which it reduces
for the case of BCS paired wave functions of spin 1/2 bosons (there are similar statements
in the more familiar case of spin-singlet pairing of spin-1/2 fermions). However, based
on the example of the Halperin (k = 1) case, we also considered the function defined as
in Eq. (16), and then simply symmetrized over all the ups and over all the downs. For
the Halperin function [which in fact is already symmetric in Eq. (16)], this satisfies the
cyclic condition, as can be seen using the fact that the (1, 1, 0) state is a Landau level filled
with both spins, plus the Pauli exclusion principle for fermions. For k = 2, 3, we verified
the cyclic condition numerically for several moderate sizes. Hence, we expect that this
simpler form actually works for all k (as well as for all N divisible by k). Apparently, this
procedure and the application of the Young operator give the same function in the end (up
to a normalization).
558
For M = 0, our wave function is then:

k,0 = Sym
NASS
p

a<b
z
a,b
p

c,d
z ,z
c,d
p
z
e,f
,
(17)
e<f
where Sym stands for the symmetrization over the ups and also over the downs. This
function is nonzero, as may be seen by letting the up coordinates coincide in clusters of k
each, and also the downs, all clusters at different locations, and making use of the result in
RR [7] that only one term in the symmetrization is nonzero in the limit. This term is the
Halperin (2k, 2k, k) function for 2p particles. To obtain the wave function for general M,
M .
we multiply by an overall Laughlin factor,
L
We can give a simple proof that our wave function (for M = 0) vanishes if any k + 1
particles, each of either spin, come to the same point. This works also for the RR wave
functions, and is simpler, though less informative, than the proof in RR [7]. It works
term by term, inside the sum over permutations in the symmetrizer. Thus, without loss of
generality, we may use the simple clustering considered above. We note that on the clock
face formed by the labels 1, . . . , k within each cluster, there is always a factor connecting
any two particles at the same position, regardless of their spin. This factor vanishes when
the particles coincide. Since there are only k distinct positions, when k + 1 particles come
to the same point, the clock positions must coincide in at least two cases, so that the wave
function vanishes, which completes the proof.
We do not have a direct general proof of the equality of these explicit wave functions and
the formal expressions Eq. (5), but we have performed a number of consistency checks.
First, the wave functions are polynomials of the correct degree. From Eq. (5), we can
infer what the total degree should be. The parafermions of the correlator contribute with
(see [18]) 1 2kp (1 k1 ). The factors of the (2, 2, 1) part are 2 2k 12 kp(kp 1) and
1 1k (kp)2 . Adding these gives, for M = 0, pk(3p 2). We need to check whether Eq. (16)
gives the same degree. For the ith up particle, the degree of zi in the product of upup
factors is 2(p 1), and in the updown factors is p. Thus the net degree in zi is N =
3p 2 = 3N/2k 2, or for general M, N = 3p +M(N 1)2 = (M +3/2k)N 2 M.
This gives the filling factor = 2k/(2kM + 3) [9], which reduces to that for the Halperin
states for k = 1, and also the shift, defined as N = N/ S, which here is S = M + 2 on
the sphere (for more on the shift, see Ref. [24]). Finally, the total degree is N/2 times that
in zi , namely kp(3p 2) for M = 0, the same as for the correlator. Also, the numerical
work described in Section 10 confirms that the ground state of the appropriate Hamiltonian
on the sphere for k = 2, M = 1 at the given number of flux does have a unique spin-zero
ground state at zero energy, so that the correlator and the wave function constructed above
must coincide. This also implies that the wave functions above must be spin-singlet.
5. Correlators corresponding to states with quasiholes

To obtain wave functions for NASS states with quasiholes, one inserts corresponding
operators into the correlator that corresponds to the ground state wave function. Here we
559
will give the form of the correlators, using standard CFT techniques, but not the wave
functions. The operators we insert have the form of a spin field times a vertex operator,
similar to the RR states [7] (note that the term spin field is traditional, and has no
relation to the SU(2) spin symmetry). In the exponent of the vertex operators, the
fundamental
bosons
are multiplied
by (fundamental) weights of the Lie algebra SU(3):

)1 = ( 2/2, 6/6), )2 = (0, 6/3). We take the quasiholes from the triplet 3 of SU(3).
The corresponding operators are

k w ,
C)1 w = )1 exp i)1 /

C)2 w = )2 exp i)2 /

(18)
k w .
In order to find the wave function for general M, we note the following. The two bosons
and spin boson (c and
s , respec = (1 , 2 ) can be written in terms of a (hyper-)charge
3
1
1
3
tively) by means of a simple rotation: 1 = 2 c + 2 s and 2 = 2 c + 2 s . The Mdependence is then brought in via a rescaling of the scale associated with the charge boson
c . The particle and quasihole operators for general M become in terms of these bosons

i
B 1 = 1 exp
(19)
z ,
2kM + 3 c + s
2k

i

z ,
B
(20)
=
exp
2kM
+
3
2
c
s
2
2k

i
1

C)
(21)
=
exp
w ,
c
s
1
2kM + 3
2k

i
1

C)
(22)
=
exp
w ,
c
s
2
2kM + 3
2k
where we have written 1 = 1 , 2 = 2 , )1 = and )2 = for simplicity.
The most basic spin fields , transform as a doublet of the SU(2) subalgebra we identify with the spin of the particles. Also, the hypercharge of the quasihole operators has the
same sign as that of the particle operators. This implies that these are indeed quasiholes,
as in earlier cases [6,7]; the wave functions are nonsingular as any particle coordinate zi
approaches any quasihole coordinate wj .
Note that when these operators are used in the CFT correlator (together with a suitably
chosen background charge), the extra Laughlin factor is automatically generated. The
correlator for the component of the wave function with N, spin-up and -down particles
and n, spin-up and -down quasiholes is given by

n + n
i
k,M
a

NASS,qh = lim (z ) exp
2kM + 3 (N + N ) +
c
z
2kM + 3
2k

+ [N N + n n ]s (z )

w1 C)
wn C)2 w1 C)
wn
C)
1
1
2

B1 z1 B1 zN B2 z1 B2 zN .
(23)
560
The value of a will be given momentarily. In the wave function (23) we inserted the
most general background charge required for neutrality in the Cartan subalgebra of SU(3).
However, the background charge can involve only the charge boson c , which corresponds
to the spin-independent background magnetic field in the QH problem. Thus we find the
condition
N + n = N + n ,
(24)
which is part of the requirement of SU(2) symmetry for the correlator. The correlator is a
spin-singlet, which means that the wave function for the particles is a nonzero spin state
of the particles, with spin determined by the quasiholes. Effectively, the quasiholes carry
spin 1/2 which is added to the spin-singlet ground state. Note that a quasihole labeled
up carries a spin down from the latter point of view, by Eq. (24), just as it carries negative
charge (hence the term quasihole), since there is a deficiency of particles at its location. For
N = N +N sufficiently large, N n in fact, the spin (up or down) for each quasihole can
be chosen freely, as we will see in some examples. Using condition (24), we can calculate
that a must be

2
2kM + 3
n
,
a=
(25)
N+
2k
2kM + 3
where n = n + n , in order that the limit z exists and is nonzero.
By working out the vertex-operator part, we arrive at the following form
k,M
NASS,qh z1 , . . . , zN ; z1 , . . . , zN ; w1 , . . . , wn ; w1 , . . . , wn

= w1 wn w1 wn

1 z1 1 zN 2 z1 2 zN

(2,2,1)
1/ k M

1/ k
1/ k
L z ; z
zi ; zj
zi wj
zi wj

H
i
j
wi
1 ( 2 +M)
wj 2kM+3 k
i<j
i,j
wi
i,j
1 ( 1 +M)
wj 2kM+3 k
i,j
wi wj
1
2
2kM+3 ( k +M)
(26)
i<j
Note that the correlator is nonzero only if the parafermion and spin fields can be fused to
yield the identity operator.
The number of magnetic flux N seen by any particle is found to be
1
2kM + 3
N + n (M + 2),
(27)
2k
2k
where we used Eq. (24). Since N must be an integer (so that the wave function is
a polynomial in the zi s), this gives another condition, that (3N + n)/2 [which is an integer
by Eq. (24)] must be divisible by k. [For the RR states, there is an analogous condition,
2N +n must be divisible by k. For k even, this means that n is even, as in the k = 2 case (the
N =
561
MR state) [6]. In Ref. [7], only the case n and N both divisible by k was considered.] From
Eq. (27), we can deduce that the quasiparticle charge is 1/(2kM + 3). This corresponds
to a fractional flux, 1/2k of the usual flux quantum. In effect, the flux quantum has been
reduced by 1/k by the formation of clusters, as in paired states and in spin-polarized RR
states [7]. The factor of 1/2 is present already in the Halperin k = 1 case. So if k is not
divisible by 3, the quasihole charge is 1/Q (Q is the denominator of the filling factor,
defined in Section 2), as in many other cases, but if k is divisible by 3, the quasihole charge
is 1/3Q. This is similar to what happens in the MR and RR states, where the quasihole
charge is further fractionalized (smaller than 1/Q) when k is divisible by 2 [7].
The conditions (24) and that N be integer are clearly necessary, but in fact are also
sufficient, to ensure that the quasihole wave functions are nonzero polynomials in the zi s,
except in the special case n = 1 where the function vanishes. To see this, one must examine
the fusion rules for the parafermion system, and check that the fields can be fused to the
identity operator under the stated conditions. This will be considered in the next section.
For completeness, we give the conformal dimensions of the particle and quasihole
(w). To obtain these, we need the dimensions of the parafermionic
operators B (z) and C)
k1
, respectively [18]. Using these,
and spin fields, which are - = 1 1/k and - = k(k+3)
(5k1)M+8
we find (see also [9]) -part = (M + 2)/2 and -qh = 2(k+3)(2kM+3)
.
We can show that the quasihole states we have obtained are zero-energy eigenstates for
the (k + 1)-body Hamiltonian above, as follows. We again concentrate on the case M = 0.
The argument using the opes of the currents B, (z) again applies [7], as long as the
k + 1 zi s are brought to a point that does not coincide with a quasihole coordinate wj . To
complete the argument, we must also consider the case where the latter condition does not
hold. There are two ways to do this. One is to note first that, as a function of the zi s for
fixed wj s, the correlator is a polynomial, as it must be to be a valid QHE wave function. (It
is not a polynomial in the quasihole coordinates wj .) This is because we chose to examine
quasiholes rather than the opposite charge objects. Then the fact that it vanishes when k + 1
zi s coincide away from a quasihole coordinate wj also implies that it vanishes when they
are at a wj , by continuity, which holds because the function is a polynomial in the zi s.
A second argument is also instructive. We may generalize the argument using the current
algebra null vectors to directly address the limit of Bs approaching a wj . There is a
generalization of the central equation for this case,
(B,1 )k | = 0.
(28)
Here the state | = C)1 (0)|0 is the state in radial quantization corresponding to the
quasihole operator at 0. There are similar equations, with successively lower exponents,
for the higher-order quasiholes (with multiples of the charge of the basic one) obtained by
successively fusing quasiholes together. These are the null vector equations for the highest
weights in distinct representations of the affine Lie algebra (or for the distinct primary
fields) [21,22]. We want to emphasize that the equation says that for certain choices of
the spins, the wave function vanishes when only k particles come to the same point (or
fewer for the higher-order quasiholes). This is a generalization of the fact that the Laughlin

quasihole is defined as the factor i (zi w) which vanishes when any one particle
562
approaches w. (This generalization applies already to the spinless RR states.) It is also

a generalization of the Halperin case, where a quasihole is a factor i (zi w ) which
vanishes when a single up particle goes to w , but not when a single down particle does.
When k basic quasiholes are fused at w (taking the leading term at each fusion), the null
vector equations state that the function does vanish when a single particle of appropriate
spin approaches w, so we have a Laughlin- (or rather Halperin-) type quasihole in that case,
as was already known for the MR (k = 2 RR) state, for example. Note that the above spin
k/2 state is the highest weight in a multiplet, so there is a set of k + 1 such null vectors in
total. This does not include all possible spin choices, as we have pointed out. To complete
the proof that the function vanishes when any k + 1 zi s (i.e., k + 1 particles of any spin)
come to wj , we must show that, for all m,
(B,1 )m (B,1 )k+1m | = 0,
(B,1 )m (B,1 )k+1m | = 0.
(29)
This can be done by an elementary argument, applying the SU(2) lowering operator to
Eq. (28), then another B,1 , using the same equation, and then lowering further, and so on.
6. Fusion rules
From now on, we focus mainly on the case k = 2, which is a spin-singlet analogue of the
MR state. The numerical studies reported in Section 10 were all performed for this special
case. Analytical results for k 2 will be presented elsewhere [27].
As pointed out in the introduction, the non-trivial fusion rules play a crucial role in the
ground state degeneracies. In fact, the correlator in Eq. (26) does not represent a single
wave function, because in general there is more than one way in which the spin fields and
parafermion fields can be fused to the identity. To show how this works, we will give the fusion rules, and explain that they can be written in terms of a Bratteli diagram. By using the
correspondence between the fields of the parafermion theory and fields of the corresponding WessZuminoWitten models (see [18]), one finds the fusion rules listed in Table 1.
Table 1
Fusion rules of the parafermion and spin fields associated to the parafermion theory SU(3)2 /[U(1)]2
introduced by Gepner [18]
12
1
2
12
1+
12 + 3
1 +
2 +
3
1+
2 +
1 +
1+
12 + 3
1+
1
12
2
1
1
563
Fig. 1. Bratteli diagram for the spin fields of SU(3)2 /[U(1)]2 .
An examination of the fusion rules shows that there are different cases, according to the
parity, even or odd, of n , n , N , N . In the case where all four numbers are even, the
spin fields and the parafermions can be fused to the identity separately. In the case where
all four are odd, the spin fields and the parafermions can be fused to 12 separately, and
these two 12 s can then be fused to the identity. Because the quasiholes only involve the
, fields, we in fact only need the first two columns of Table 1. With this restriction, the
fusion rules can be written in terms of a Bratteli diagram, see Fig. 1.
Each arrow stands for either a or field. The arrow starts at a certain field which
can only be one of the fields on the left of the diagram at the same height. This last field
is fused with the one corresponding to the arrow, while the arrow points at a field present
in this fusion. As an example, the arrows starting at the are encoding the fusion rules
() = 2(1) + () and 3 () = 1(2) + () . One checks that the diagram is
in accordance with the first two columns of Table 1. The symbol 1 at the right-hand side of
Fig. 1 indicates that in the end we have fused the fields to the identity. This is possible only
when n and n are both even; in the case where both numbers are odd, 12 is obtained at
that position in the diagram. In the remaining two cases, where n = n + n is odd, we can
draw a similar diagram with the last point at the top, representing 1 or 2 (except when
n = 1, to which we return in a moment). In these cases, N = N + N must also be odd,
in order that the fusion of the parafermions 1 and 2 for the particles can produce the
appropriate field which can fuse with the result of the spin field fusions to finally give the
identity. In the case n = 1, it is not possible to fuse the spin fields to obtain a parafermion,
and the correlator vanishes.
For the counting formula we need to know the number of ways in which a given number
of spin fields can be fused to give a field in the parafermion sector, that is 1, 1 , 2 ,
or 12 . This number equals the number of paths (of length the number of spin fields) on the
Bratteli diagram leading to the corresponding point on the diagram. One finds that when
the numbers of spin-up and -down quasiholes are n and n , respectively, this number,
the number of fusion channels, is dn ,n = fn2 , where the Fibonacci number is defined
by fm = fm1 + fm2 , with the initial conditions f0 = 1 and f1 = 1. This is valid for
n 2 in all four cases of n , n even or odd, while for n = 1, dn ,n = 0, and for n = 0,
d0,0 = 1. The result is obtained by examining the Bratelli diagram and seeing that the
number of paths obeys the recurrence relation that defines the Fibonacci number. We note
that this result, the Fibonacci number fn2 , is the same as for n quasiholes in the k = 3 RR
states [7]. This is a manifestation of level-rank duality, here between SU(3)2 and SU(2)3 .
The final fusion of the spin fields with the parafermions from the particles must produce
the identity, in order that the correlator be nonzero. When dn ,n is nonzero, necessary and
564
sufficient conditions for this are N + n N + n (mod 2) and N n 0 (mod 2).

The first condition is the mod 2 version of the condition (24), while the second is equivalent
to the condition (3N + n)/2 0 (mod 2), on using Eq. (24). Note that, as can be inferred
from the fusion rules for the fields (see Table 1), the fusion of the fields from the
particle operators does not increase the degeneracy. The abelian properties of the fields
correspond to the abelian statistics (Bose or Fermi) of the underlying particles.
These results give the degeneracy dn ,n of quasihole states for fixed positions and
spins of the quasiholes, which is the basis for non-abelian statistics properties. We see
that the result does not depend on how many quasiholes are spin up or spin down, and the
sum over all choices of spins gives a further factor of 2n spin degeneracy for sufficiently
large N , when only the positions are fixed. In the following sections, we examine the
total degeneracies of quasihole states when both their positions and spins are unrestricted.
These are more suitable for numerical checks, and are finite numbers on the sphere (for a
disk in the plane, they are infinite, and contain information about edge excitations as well
as bulk quasiholes). As in cases studied earlier [12,7], the total degeneracies are not just
the numbers found above times a factor for the spatial degeneracy contribution, but involve
partitioning the Fibonacci numbers above into a sum of positive integers. We also note here
that when a generic Hamiltonian has a ground state in the NASS phase, the degeneracies
will not be exact, but will be as given in this section when all quasiholes are asymptotically
far separated. This will not be considered further in this paper.
7. Spatial degeneracies
Techniques for calculating degeneracies for a spherical geometry are described in full
detail in [12]. On the sphere, the relation between the number of particles and the number of
flux quanta for the ground state is given by N = 1 N S. By increasing the number of
flux quanta at fixed N , quasiholes are created. Moreover, when flux have been added, there
may be zero-energy states with N not satisfying the conditions required in the ground state,
for example that N be even in the MR state [12]. In general, we would define the number
of flux added as -N = N 1 N + S. This is defined relative to a ground state at
the same number of particles, even though such a zero-energy state for N not divisible by
2k (or k for the RR states) would require a non-integer number of flux, and does not exist.
Consequently, while our N is always an integer, as discussed in Section 5, -N does not
have to be an integer. The number of quasiholes n can then be defined as n = k-N (for
the RR states), or n = 2k-N for the NASS states considered here, in agreement with our
formula for N in Eq. (27).
To explain the spatial degeneracies, we use the Laughlin case as an example, and follow
the discussion of [12]. The Laughlin wave function for N particles in the presence of n
quasiholes can be written as [15]

M
L,qh
(z1 , . . . , zN ; w1 , . . . , wn ) = (zi zj )M
(zi wk ).
(30)
i<j
i,k
565
For this state, n = -N (adding one quantum of flux creates one quasihole), so we have
N = M(N 1) + n, where we used that S = M for the Laughlin state. To continue, we
calculate the degeneracy due to the presence of the quasiholes, by expanding the factor

i,k (zi wk ) in sums of products of the elementary symmetric polynomials

wi1 wi2 wim .
em =
(31)
i1 <i2 <<im
Viewing the coordinates wi as coordinates of bosons, we find that n bosons are to be placed
in N +1 orbitals. The dimension of the space of available states (linearly-independent wave
functions) equals the number of ways in which one can put n bosons in N + 1 orbitals,
which is

N +n
.
(32)
n
This is the spatial degeneracy we are after, although for the simple case of the Laughlin
states.
The situation for the MR state is discussed in detail in [12]. For the MR state, there is
an additional complication, namely quasihole states in which there are unpaired fermions
are possible; this is the origin of the degeneracies for fixed quasihole positions, already
discussed. We will denote the number of unpaired particles by F , with the requirement
that N F be even, so that the number of unbroken pairs (N F )/2 is an integer. For
n sufficiently large, N need not be even in the zero-energy states. It was found that the
spatial degeneracy depends on the number of unbroken pairs; in fact, for the MR state, it
was given by [12]

NF
2 +n .
(33)
n
For the clustered state of [7] (in which the particles form clusters of order k rather than
pairs), the spatial degeneracy is given similarly by [7,13]

NF
+
n
k
,
(34)
n
where N F must be divisible by k (there appears to be no known general analytic proof
of this formula for k > 2).
Based on these earlier results, and because the NASS wave functions involve clusters of
up particles and separately of downs, we expect that the spatial degeneracy for the NASS
states is just a product of two binomial coefficients, involving the F1 , F2 unclustered
particles (or parafermions) for spins up and down, respectively:
N F
N F

1
2
+
n
+
n
k
k
(35)
.
n
n
Again, N F1 and N F2 must be divisible by k. Notice that these numbers never
depend on M.
566
The case k = 1, F1 = F2 = 0 gives the spatial degeneracies for the general Halperin
(m , m , m) abelian states (as was mentioned briefly in Ref. [12] for a particular case). For
the spin-singlet cases (m + 1, m + 1, m) (m = M + 1) of interest here, the conditions that
correspond to zero-energy states with only the particle and quasihole numbers, N and n,
fixed are that N + n = N + n . For N sufficiently large, these allow any choice of spin,
up or down, for each quasihole, giving a factor of 2n spin degeneracy for fixed positions
(this holds for all k). Such degeneracy does not contribute to the degeneracy on which
non-abelian statistics depends, and the statistics is abelian in the present k = 1 case (as of
course was expected). The full degeneracy in this case k = 1 is obtained by summing (35)
over N , N , n , n , subject to the conditions just stated, with F1 = F2 = 0. These imply
that the summation is over only the possible values of Sz = (N N )/2, and the result is

N +n
(36)
.
n
Note that this number includes the spin degeneracy. If we take the ratio to the number

(N + n)/2
(37)
n
of quasihole states with, say, n = 0, the result tends to 2n as N , which is again
the spin degeneracy for fixed positions. In the remainder of this paper, we concentrate
exclusively on k = 2.
With the orbital degeneracies in hand, we need to know how to break up the degeneracies
stemming from the fusion rules of the spin fields. So in fact we need to know the number of
unpaired particles of either spin, F1 , F2 , for each possible path on the Bratteli diagram. The
next section will treat this problem, by partitioning the numbers dn ,n in the following way
n n
.
dn ,n = fn +n 2 =
(38)
F1 F2 2
F1 ,F2

The symbol nF nF is interpreted as the number of zero-energy states containing n , n
1 2 2
quasiholes at fixed positions and F1 , F2 unpaired parafermions. The symbol vanishes if the
conditions that F1 n , F2 n be even are not satisfied; these conditions are equivalent
to the two mod 2 conditions discussed at the end of Section 6. The subscript 2 indicates
that we are dealing with the case k = 2.
8. Counting SU(3)2 /[U(1)]2 parafermion states
It was explained in [12] that the state counting for the MR state involves the systematics
of Majorana fermions, which act as BCS quasiparticles (unpaired fermions) occupying
zero-energy states [20]. For the more general RR (spin-polarized) states with order k
clustering [7], the Majorana fermion is replaced by an SU(2)k /U(1) parafermion [7,13].
We recall that for the NASS state at level k = 2, the role of the Majorana fermion is taken
over by the parafermions that are associated to SU(3)2 /[U(1)]2. The (spin-up or -down)
quasiholes correspond to (two different) spin fields , of this parafermion theory (see
567
Section 5). In Section 6, we calculated the number of different quantum states (conformal
blocks for the correlators) that can result from introducing n and n spin fields
(and also varying the number of particles N , N ) with the result dn ,n = fn +n 2 . The
degeneracy results from the presence of varying numbers of particles that are not members
of clusters, which in the correlators can be identified with the fundamental parafermions
1 , 2 of the parafermion theory. F1 (F2 ) is the number of 1 (2 ) excitations. These
numbers are subject to the conditions that F1 n (mod 2), F2 n (mod 2), otherwise
the number of zero-energy states is zero; these conditions come from those discussed in
Section 6. In the previous section, we found that the orbital degeneracy depends on the
numbers F1 and F2 , see Eq. (35). We now turn to the calculation of the symbols

n n
,
(39)
F1 F2 2
which partition the degeneracies dn ,n in the correct way. We will also keep track of the
angular momentum (L) multiplet structure associated with these parafermion states. To do
this, we have to go through a series of steps.
First of all, we consider the infinite (chiral) character corresponding to the full
parafermionic CFT (see [25])
q (F12 +F22 F1 F2 )/2 F F
ch(x1 , x2 ; q) =
x1 1 x2 2 ,
(q)F1 (q)F2
(40)
F1 ,F2

where (q)a = aj=1 (1 q j ) for integer a. Here F1 and F2 are unrestricted non-negative
integers, and x1 , x2 , q are indeterminates.
What is needed for our purposes here is the truncation of this expression to (a sum of)
polynomials Yn ,n (x1 , x2 ; q) that describes the states that occur when n , n spin fields
(quasiholes) are present. The approach is described in Refs. [14,26], see also Ref. [13],
and details for the present case will be given in Ref. [27]. We find that these polynomials
satisfy the following recursion relations
Y(n ,n ) = Y(n 2,n ) + x1 q (n 1)/2 Y(n 2,n +1) ,
Y(n ,n ) = Y(n ,n 2) + x2 q (n 1)/2 Y(n +1,n 2)
(41)
with initial conditions

Y(1,0) = Y(0,1) = 0,
Y(0,0) = Y(2,0) = Y(0,2) = 1,
Y(1,1) = q 1/2x1 x2 .
(42)
Recursion relations similar to the above (but lacking the x1,2 dependence), have been
considered in the mathematical literature on special polynomials associated to SU(3)2 ,
see for instance [28]. The coefficient of x1F1 x2F2 in the polynomial Y(n ,n ) is a polynomial
in q with the sum of the coefficients equal to the symbol (39), that is
F F n n
Y(n ,n ) (x1 , x2 , 1) =
(43)
x1 1 x2 2
.
F1 F2 2
F1 ,F2
568
We notice that the recursion relations preserve the conditions on the parities of F1 , F2 (the
exponents of x1 , x2 ) that are part of the definition of the symbol (39). In the limit where
(n , n ) (, ), the sum of these polynomials over the four choices, n and n each
either even or odd, approaches the expression ch(x1 , x2 ; q) given above.
The coefficient of x1F1 x2F2 in the polynomial Yn ,n has a special form, which allows us to
extract information on the L quantum numbers of the parafermion states: after multiplying
with a factor q (n F1 +n F2 )/4 , we recognize a sum of terms of the form q lz , which together
form a collection of angular momentum (L) multiplets with quantum numbers Lz = lz [13].
To illustrate the above, we present the polynomials for the case of two added flux quanta,
giving eight quasiholes. The polynomials are

Y(8,0) = 1 + q 2 + q 3 + 2q 4 + q 5 + q 6 x12 + q 8 x14

+ q 6 + q 7 + q 8 + q 9 + q 10 x14 x22 ,
7
3
5
7
9
11
13
15
Y(7,1) = q 2 + q 2 + q 2 + q 2 x1 x2 + q 2 + 2q 2 + 2q 2 + 2q 2 + q 2 x13 x2
19
+ q 2 x15 x23 ,

Y(6,2) = 1 + q 2 + q 3 + q 4 x12 + q 2 + q 3 + 2q 4 + q 5 + q 6 x12 x22

+ q 6 + q 7 + q 8 x14 x22 ,
7
3
5
7
9
11
Y(5,3) = q 2 + 2q 2 + 2q 2 + q 2 x1 x2 + q 2 + q 2 + q 2 x13 x2
9
11
13
15
+ q 2 + q 2 + q 2 + q 2 x13 x23 ,

Y(4,4) = 1 + q 2 x12 + q 2 x22 + q 2 + 2q 3 + 3q 4 + 2q 5 + q 6 x12 x22 + q 8 x14 x24 ,

etc.
(44)
From the polynomial Y(5,3) (as an example), we read off the following nonzero values of
the symbols

1
3
5 3
,
=6
L = ,L =
1 1 2
2
2

5 3
= 3 (L = 1),
3 1 2

3
5 3
.
=4
L=
(45)
3 3 2
2
In fact, it is possible to write the polynomials Y(n ,n ) in a closed form [29],
Y(n ,n ) (x1 , x2 ; q) =

F1 ,F2
2
2
q (F1 +F2 F1 F2 )/2x1F1 x2F2
n
+F2
n
+F1
F1
F2

,
(46)

a
where ab is the q-deformed binomial (q-binomial), defined as ab = (q)b(q)
(q)ab , and
theprime on the sum denotes the restriction on F1 , F2 values. Using the property that in
the limit q 1 the q-binomials become ordinary binomials, we find the following explicit
formula for { }2 (under the same conditions on F1 , F2 , otherwise it vanishes):

n +F2 n +F1

n n
2
2
=
.
F1 F2 2
F1
F2
569
(47)
Note that if we take the sum over all F1 and F2 , we indeed find the correct value, namely
a Fibonacci number
n +F2 n +F1
2
2
= fn +n 2 .
(48)
F1
F2
F1 ,F2
While Eq. (47) gives us just
a the number, we can also obtain the angular momentum

content easily. The binomial f is interpreted as the number of possible ways of putting
f fermions in a boxes which have quantum numbers Lz = (a 1)/2, . . . , (a 1)/2
assigned to them. In this way, an angular momentum multiple structure is assigned to the
binomials (see [12]). The angular momentum content of the symbols {}2 is obtained by
adding the angular momenta associated to the binomials in the usual way.
9. Final counting formula

We are now in the position to write down the formula for the total degeneracy of zeroenergy quasihole states of the k = 2 non-abelian spin-singlet states. Recall that there
are two conditions on the numbers of quasiholes (see Section 5). The first condition is
N + n = N + n , which implies that the correlator is a spin-singlet, or that the wave
functions have total spin determined by the spin-1/2 quasiholes added. The other condition
was that (3N + n)/2 be even, to ensure that N is an integer, where N = N + N , and
n = n + n = 4-N , which relates the number of excess flux quanta and the number of
quasiholes added. These imply that N n = N n must be even.
The result of the previous few sections is now that the total number, summed over all
spin components, of zero-energy states as a function of the number of particles and added
flux quanta is
n +F2 n +F1 N F1
N F2

+ n
+ n
2
2
2
2
,
#(N, -N ) =
F1
F2
n
n
N, ;n, ;F1,2
(49)
where the prime on the sum indicates that it is restricted to values obeying all the conditions
just mentioned, and to N F1 and N F2 even as discussed in Section 7. Note again
that these conditions imply that n + F2 and n + F1 are even.
In addition, the orbital angular momentum decomposition of the states can be obtained,
by combining the angular momenta found in the orbital and parafermion factors in the
preceding two sections. The spin quantum number of any given state is simply Sz = (N
N )/2 and one readily recognizes the multiplet structure for the SU(2) spin symmetry. (We
remark that the parafermion theory by itself does not have a proper SU(2) spin symmetry.)
In Table 2, we present counting results for N = 4, 8, 12 and -N = 1, 2, 3, 4. We
specify the number of states as a function of the L and S quantum numbers. In Table 3
570
Table 2
Counting results for the NASS states at k = 2. N is the number of electrons; -N is the number of
excess flux quanta. The results are given as a function of the L (angular momentum) and S (total
spin) quantum numbers. The total number of states is also indicated
-N = 1
-N = 2
-N = 3
N = 4 # = 20 S = 0
L=0
1
L=1
0
L=2
1
1
0
1
0
2
1
0
0
# = 104 S = 0
L=0
1
L=1
0
L=2
2
L=3
0
L=4
1
1
0
2
1
1
0
2
1
0
1
0
0
N = 8 # = 105 S = 0
L=0
2
L=1
0
L=2
2
L=3
0
L=4
1
1
0
2
1
1
0
2
1
0
1
0
0
1
1
7
7
9
6
5
2
1
0
2
3
2
6
3
4
1
1
0
0
N = 12 # = 336 S = 0
L=0
3
L=1
0
L=2
3
L=3
1
L=4
2
L=5
0
L=6
1
# = 1719 S = 0
L=0
4
L=1
1
L=2
7
L=3
3
L=4
6
L=5
2
L=6
3
L=7
0
L=8
1
1
0
3
3
3
1
1
0
2
1
0
2
0
1
0
0
# = 321 S = 0
L=0
2
L=1
0
L=2
2
L=3
1
L=4
2
L=5
0
L=6
1
3
0
1
1
1
0
0
0
0
0
-N = 4
1
0
2
2
3
1
1
0
2
1
0
2
0
2
0
0
# = 755 S = 0
L=0
2
L=1
0
L=2
3
L=3
1
L=4
3
L=5
1
L=6
2
L=7
0
L=8
1
1
0
3
2
4
3
3
1
1
0
2
1
0
2
1
2
0
1
0
0
4
1
0
0
0
0
0
0
0
0
we give some results for N not a multiple of four. Notice that for n = 1, there are no zeroenergy states, as expected from Section 6. The results listed in Tables 2 and 3 are for the
cases we checked numerically, as we discuss in the next section, and are in full agreement
with those results.
10. Numerical methods and results

We next turn to some numerical studies of the NASS states. We consider only cases
where the particles are fermions, to represent electrons. We have studied the k = 2, M = 1
( = 4/7) case in both the toroidal (PBC) and spherical geometries. We first present the
results for the sphere. As discussed before the flux-charge relation for this state is N =
571
Table 3
Counting results for the NASS states at k = 2 with fractional -N (symbols as in Table 2)
-N = 1/2
S =0 1
0 1
N =2
#=3
L=0
N =6
# = 10
L=0
L=1
N =5
#=0
N =3
-N = 3/4
#=4
S = 1/2
L = 1/2
1
-N = 3/2
S=0 1
1 0
0 1
S=0 1 2 3
0 2 0 1
2 1 1 0
0 3 1 0
2 1 0 0
0 1 0 0
-N = 5/4
# = 48
S = 1/2 3/2
L = 1/2
1 1
L = 3/2
1 1
L = 5/2
1 0
-N = 7/4
# = 28
S = 1/2 3/2
L=0
0 0
L=1
1 1
L=2
1 0
# = 10
L=0
L=1
# = 175
L=0
L=1
L=2
L=3
L=4
S =0 1
1 0
0 1
-N = 1/4
7N/4 3. The number of single-particle orbitals (the lowest LL degeneracy) is N + 1.

In order to make contact with the results on more conventional geometries the radius R of
the sphere has to be chosen so that the number of flux is N = 2R 2 (where the magnetic

field strength B is fixed, such that the magnetic length is 1 in our units), so R = N /2
where n = N/(4R 2 ) is the particle number
[16]. The filling factor is = N/N = 2 n,
density.
For numerical purposes, it is best to re-express the interaction Hamiltonian in terms
of projection operators onto different values of the total angular momentum for different
groups of particles [16]. For the M = 1, k = 2 case of the NASS states, the required
Hamiltonian can be written as

Pij k (3N /2 3, 3/2) + V
Pij (N , 0),
H =U
(50)
i<j <k
U, V
i<j
> 0. Here Pij k (L, S) (Pij (L, S)) are projection operators for the three
with
(respectively, two) particles specified onto the given values of total angular momentum
L and spin S for the three (respectively, two) particles. Each projection is normalized to
P 2 = P . To see that this is the required Hamiltonian, that corresponds to the short range
-function interaction for M = 0, and gives the same numbers of zero-energy states found
above, note the following. First, the maximal angular momentum for several particles
corresponds to the closest approach of those particles [16]. In particular, the two-body
term is a contact interaction, and V = V0 , the zeroth Haldane pseudo-potential [16]. The
two-body term implies that any zero-energy states must have no component with total
572
Fig. 2. The spectrum of the NASS model ground state for N = 8 and 4/7 filling. The last panel shows
all S values combined. The insets are the low lying levels.
angular momentum N and total spin zero, which, since we are dealing with spin 1/2
fermions, means the wave function must vanish when any two particles coincide. The wave
1 ; multiplication by this factor defines a one-one
function must therefore contain a factor
L
mapping of the full space of states of spin 1/2 bosons in the lowest LL, with N reduced by
N 1, onto the subspace of states of the fermions that is annihilated by the two-body term
in H . Under this mapping, the three-body Hamiltonian for the M = 0 case corresponds
to the three-body term in H , and selects the corresponding states as zero-energy states. In
particular, the total spin of the three bosons when they coincide (and hence of the fermions)
must be 3/2. Hence the zero-energy eigenstates of the present Hamiltonian are given by
the results derived earlier. Note also that H can be rewritten in terms of -functions and
their derivatives. The zero-energy eigenstates of this Hamiltonian were found for various
N and N values, and analyzed in terms of L and S. The results are shown in Tables 2
and 3, and agree with the counting formulas presented above.
Next we discuss the full spectra of the Hamiltonian. In Fig. 2 we show the excitation
spectrum of an N = 8, N = 11 system classified by the total spin S = 0 to S = 4.
573
Fig. 3. The spin upup and spin updown pair correlation functions, together with their sum (solid
line) and difference (dashed line), versus the chord distance, calculated in the ground state for N = 8,
N = 11 ( = 4/7).
Whenever necessary we have shown the low-lying spectrum in an inset. The frame in the
lower right hand corner shows the entire spectrum irrespective of the total spin quantum
number of the state. The choice of U and V is immaterial to the ground state, which is
always the unique zero-energy eigenstate of H . Obviously, the excitation spectrum will
depend on the choice of these coefficients. In producing Fig. 2 we chose U = V = 1 so
that H is a sum of projection operators. There appears to be a well-defined gap, suggesting
that the system is incompressible (for spin as well as charge) in the thermodynamic limit, as
assumed in the preceding analysis. In this connection, we may point out that, as well as the
quantized Hall conductivity for charge, our system then has a quantized Hall conductivity
for spin, given by k/4 in natural (h = 1) units, which is associated with the SU(2)k
subalgebra of the chiral algebra (see, e.g., Ref. [20] and references therein). Collective
modes with S = 0 (L = 2, 3, 4) and S = 1 (L = 1, 2, 3) below the continuum can be
tentatively identified in the spectra (see insets in Fig. 2). That is, these may be finite-size
dispersion curves of single neutral excitations in plane-wave (spherical harmonics on the
sphere) wave functions, which would be charge and spin modes, respectively. We shall not
address the precise nature of these neutral modes here.
In Fig. 3 we show the various pair correlation functions of interest, g (r), g (r), as
well as gTotal = g (r) + g (r) and g (r) g (r). The widely-different correlations
between like and opposite spins is no doubt magnified by finite-size effects.
The 8-electron system is the first non-trivial size and is probably too small for any meaningful comparisons of the overlap with the 2-body Coulomb potential problem. Nonetheless we found a nontrivial overlap-squared (about 55%) with the ground state of the
Coulomb potential for particles in the lowest LL (again, with no Zeeman term), at the
same N , N . By modifying the value of the lowest pseudo-potential for the Coulomb interaction this overlap-squared can be improved to 93% (and probably beyond) without any
intervening phase transition (an energy gap with the ground state is maintained at all times
while the pseudo-potentials are varied). However, in the lowest LL we do not expect to
produce a better trial wave function than one constructed by the composite fermion (CF)
method [30], in which two flux quanta per particle are attached (in the opposite direction
574
Fig. 4. The ground state energy of the pure Coulomb potential in the lowest Landau level versus N ,
at N = 8. The numbers of flux N for the NASS and the spin-singlet CF states are marked.
to the background magnetic field) and the resulting CFs fill completely the first two LLs of
CFs (with both spins). By construction this is a uniform (L = 0) spin-singlet state. We have
not constructed this state, as it occurs at a different flux for a given N (N = 7N/4), making
a direct comparison with our NASS state difficult. We note, however, that for N = 8 the CF
state corresponds to a spin-singlet Fermi-liquid-like state, as at = 1/2. That is because the
net effective field of the CFs is zero for this size (states that lie in sequences for different filling factors can coincide at finite size on the sphere, because the shifts S may be different
see, e.g., [31]). We expect that, as usual, this CF state will have a very large overlap with
the exact ground state of the Coulomb potential. However, our numerical data for N = 8
shows a much stronger cusp at the N of the NASS state than at the N of the state obtained
by the hierarchy/CF construction, where in fact no cusp can be discerned. See Fig. 4.
We have also studied the N = 8 size on the torus. Unfortunately, as on the sphere this
size is too small for any meaningful comparison (e.g., there are only four distinct manybody k vectors for this size; one is at the zone center, the other three at the zone boundary).
We would just like to point out that, for the analog of H on the torus, the degeneracy for the
4/7 state is 2 (excluding the 7-fold center of mass degeneracy), in agreement with the count
in Section 2, since the number of particles is divisible by 4. These are two k = 0 states.
For the pure Coulomb potential in the lowest LL the state at 4/7 is in fact compressible:
The total spin is S = 1 and its k vector varies with the geometry of the PBC unit cell.
However, one obtains an incompressible state by increasing V0 or V1 and we obtained
overlap-squared as large as 50% when we compared the lowest two states (which happen
to be both S = 0, k = 0 states) to the model NASS states. Note that the shift is zero on
the torus, and there can be no interference from = 1/2 here. We have not performed
any further or systematic studies of such issues because, as in the case of the sphere, we
suspect that the CF-based state will be closer to that of the Coulomb potential. That is, we
expect that the system with Coulomb interaction in the lowest LL at = 4/7 is in fact in
the hierarchy/composite-fermion phase (whether spin-singlet or not), not the NASS phase
considered in this paper. We will return to more comprehensive studies of larger sizes in
the future.
575
Acknowledgements
We thank P. Bouwknegt, J. Schliemann and S.O. Warnaar for discussions. The research
of E.A. and K.S. is supported in part by the Netherlands Organisation for Scientific
Research (NWO) and by the foundation FOM of the Netherlands. N.R. was supported by
NSF Grant No. DMR-98-18259. E.R. was supported by NSF Grant No. DMR-0086191.
References
[1] R.L. Willett, J.P. Eisenstein, H.L. Strmer, D.C. Tsui, A.C. Gossard, J.H. English, Phys. Rev.
Lett. 59 (1987) 1776.
[2] W. Pan, J.-S. Xia, V. Shvarts, D.E. Adams, H.L. Strmer, D.C. Tsui, L.N. Pfeiffer, K.W.
Baldwin, K.W. West, Phys. Rev. Lett. 83 (1999) 3530.
[3] For a short review, see N. Read, cond-mat/0011338.
[4] R. Morf, Phys. Rev. Lett. 80 (1998) 1505;
E.H. Rezayi, F.D.M. Haldane, Phys. Rev. Lett. 84 (2000) 4685.
[5] W. Pan, H.L. Stormer, D.C. Tsui, L.N. Pfeiffer, K.W. Baldwin, K.W. West, cond-mat/0103144.
[6] G. Moore, N. Read, Nucl. Phys. B 360 (1991) 362.
[7] N. Read, E. Rezayi, Phys. Rev. B 59 (1999) 8084.
[8] H. Cho, J.B. Young, W. Kang, K.L. Chapman, A.C. Gossard, M. Bichler, W. Wegschneider,
Phys. Rev. Lett. 81 (1998) 2522.
[9] E. Ardonne, K. Schoutens, Phys. Rev. Lett. 82 (1999) 5096.
[10] E. Ardonne, F.J. M van Lankvelt, A.W.W. Ludwig, K. Schoutens, Separation of spin and charge
in paired spin-singlet quantum Hall states, cond-mat/0102072.
[11] C. Nayak, F. Wilczek, Nucl. Phys. B 479 (1996) 529.
[12] N. Read, E. Rezayi, Phys. Rev. B 54 (1996) 16864.
[13] V. Gurarie, E. Rezayi, Phys. Rev. B 61 (2000) 5473.
[14] K. Schoutens, Phys. Rev. Lett. 79 (1997) 2608.
[15] R.B. Laughlin, Phys. Rev. Lett. 50 (1983) 1395.
[16] F.D.M. Haldane, Phys. Rev. Lett. 51 (1983) 605;
F.D.M. Haldane, in: R.E. Prange, S.M. Girvin (Eds.), The Quantum Hall Effect, 2nd edn.,
Springer-Verlag, New York, 1990.
[17] M. Greiter, X.-G. Wen, F. Wilczek, Phys. Rev. Lett. 66 (1991) 3205;
M. Greiter, X.-G. Wen, F. Wilczek, Nucl. Phys. B 374 (1992) 567.
[18] D. Gepner, Nucl. Phys. B 290 (1987) 10.
[19] B. Halperin, Helv. Phys. Acta 56 (1983) 75.
[20] N. Read, D. Green, Phys. Rev. B 61 (2000) 10267.
[21] D. Gepner, E. Witten, Nucl. Phys. B 278 (1986) 493.
[22] A.B. Zamolodchikov, V.A. Fateev, Yad. Phys. 43 (1986) 1031;
A.B. Zamolodchikov, V.A. Fateev, Sov. J. Nucl. Phys. 43 (1986) 657.
[23] M. Hamermesh, Group Theory and Its Application to Physical Problems, Dover, New York,
1989, Chapter 7.
[24] X.-G. Wen, Adv. Phys. 44 (1995) 405.
[25] A. Kuniba, T. Nakanishi, J. Suzuki, Mod. Phys. Lett. A 8 (1993) 1649;
G. Georgiev, Combinatorial constructions of modules for infinite-dimensional Lie algebras, II.
Parafermionic space, q-alg/9504024.
[26] P. Bouwknegt, K. Schoutens, Nucl. Phys. B 547 (1999) 501.
[27] E. Ardonne, K. Schoutens, to be published.
[28] A. Schilling, S.O. Warnaar, Commun. Math. Phys. 202 (1999) 359.
576
[29] P. Bouwknegt, Multipartitions, generalized Durfee squares and affine Lie algebra characters,
math.CO/0002223;
E. Ardonne, P. Bouwknegt, K. Schoutens, J. Stat. Phys. 102 (2001) 421.
[30] J.K. Jain, Phys. Rev. Lett. 63 (1989) 199.
[31] E. Rezayi, N. Read, Phys. Rev. Lett. 72 (1994) 900;
E. Rezayi, N. Read, Phys. Rev. Lett. 73 (1994) 1052 (C).

Partial masslessness of higher spins in (A)dS

S. Deser, A. Waldron 1
Physics Department, Brandeis University, Waltham, MA 02454, USA
Abstract
Massive spin s 3/2 fields can become partially massless in cosmological backgrounds. In the
plane spanned by m2 and , there are lines where new gauge invariances permit intermediate
sets of higher helicities, rather than the usual flat space extremes of all 2s + 1 massive or just 2
massless helicities. These gauge lines divide the (m2 , ) plane into unitarily allowed or forbidden
intermediate regions where all 2s + 1 massive helicities propagate but lower helicity states can
have negative norms. We derive these consequences for s = 3/2, 2 by studying both their canonical
(anti)commutators and the transmutation of massive constraints to partially massless Bianchi
identities. For s = 2, a Hamiltonian analysis exhibits the absence of zero helicity modes in the
partially massless sector. For s = 5/2, 3 we derive Bianchi identities and their accompanying gauge
invariances for the various partially massless theories with propagating helicities (5/2, 3/2) and
(3, 2), (3, 2, 1), respectively. Of these, only the s = 3 models are unitary. To these ends,
we also provide the half integer generalization of the integer spin wave operators of Lichnerowicz.
Partial masslessness applies to all higher spins in (A)dS as seen by their degree of freedom counts.
Finally a derivation of massive d = 4 constraints by dimensional reduction from their d = 5 massless
Bianchi identity ancestors is given. 2001 Elsevier Science B.V. All rights reserved.
PACS: 03.65.Pm; 03.70.+k; 04.62.+v; 04.65.+e
1. Introduction
The (m2 , ) mass-cosmological constant plane for spin s = 2 is unexpectedly rich
[13]: for generic (m2 , ), the theory describes 5 massive degrees of freedom. However,
the field equations, having two open indices, are susceptible to both single and double
divergence Bianchi identities each signaling the onset of gauge invariances. This is indeed
realized when m2 = 0 and m2 = 2/3, respectively. The former corresponds to linearized
cosmological Einstein gravity and the latter to the partially massless de Sitter (dS) theory
of [1]. A peculiarity of (Anti) de Sitter ((A)dS) spaces is that gauge invariance does
E-mail addresses: deser@brandeis.edu (S. Deser), wally@brandeis.edu (A. Waldron).
1 Address after September 2001: Department of Mathematics, UC Davis, CA, USA.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 2 1 2 - 7
578
S. Deser, A. Waldron / Nuclear Physics B 607 [FS] (2001) 577604
not necessarily imply null propagation, which in fact occurs only for the m2 = 2/3
theory [1]. Furthermore, these two gauge invariant theories lie on the boundary of two
physically distinct regions (see Fig. 1). As demonstrated in [2], in the dS region with
masses bounded by 0 < m2 < 2/3, the massive s = 2 spectrum is not unitary. This is an
example of a generic difficulty encountered by higher spin theories in backgrounds; it was
first observed in a study of canonical anticommutators for charged massive s = 3/2 fields
in electromagnetic backgrounds [4] and later shown to imply a causal propagation [5].
The above s = 2 behaviors are common to spins s 3/2 in (A)dS, as seen by studying
possible Bianchi identities. The number of massive higher spin covariant field components
exceeds the 2s + 1 propagating degrees of freedom (PDoF) count. The correct count
is imposed by constraints constructed from divergences of each open index of the field
equations, D 1 D n G1 ...n n+1 ... + = 0. In flat space, all these constraints hold
identically when the mass m vanishes and are encapsulated by a single Bianchi identity
1 G1 2 ... + = 0 (for s > 2 the terms + ensure (gamma-)tracelessness of the
remaining open indices) reflecting the gauge invariance of the massless theory [6]. Once a
cosmological constant is present, this degeneracy is broken and for special tunings of m2
to , intermediate situations are possible where some of the lower divergence constraints
remain and all the higher ones are replaced by a single Bianchi identity [3]. This new
identity entails a gauge invariance of the theory. As a consequence, higher spins in (A)dS
spaces can be partially massless: not only are there strictly massive theories with all 2s + 1
helicities and strictly massless ones with only helicities s, but also intermediate theories
of helicities (s, (s 1), . . . , (s t)) (t < s). 2
The new Bianchi identities appear along lines in the (m2 , ) plane and are also
manifested in the canonical (anti)commutators of the fields. These follow directly from the
covariant transverse-traceless decomposition of the on-shell fields [3]. 3 The gauge lines in
the (m2 , ) plane correspond to poles in the lower helicity terms of (anti)commutators, the
usual indication that a theory enjoys a gauge invariance at special values of parameters.
Therefore, lower helicity (anti)commutators flip sign across these gauge lines. The
intermediate regions correspond to (unitarily allowed or forbidden) massive theories;
the one including the massive Minkowski theory is always unitary. The norms of those
helicities that are removed by a gauge invariance flip sign across the corresponding line.
This implies that the partially massless theories of the remaining helicities are unitary
only when, starting from the unitary Minkowski region, the line in the (m2 , ) halfplane removing the lowest helicity state(s) can be traversed first and subsequent lines
are encountered in order, ending on the strictly massless helicity s gauge line. The
latter is guaranteed to be unitary since it removes all dangerous lower helicities. We show
explicitly that this condition is fulfilled for s = 2, 3 but not for the novel s = 5/2 partially
massless gauge line. We will also demonstrate, by a simple counting argument, that partial
masslessness is enjoyed by all higher spins in (A)dS.
2 An additional subtlety for s 5/2 is that auxiliary fields are necessary (over and above the field content of
the strictly massless models), but at least in the massless limit, these can be shown to decouple.
3 Their restriction to equal time commutators agrees with the detailed s = 2 Dirac analysis of [2].
579
Fig. 1. Respective phase diagrams for spins s = 3/2, 2, 5/2, 3 showing partially massless gauge lines
and unitarily allowed and forbidden regions.
In Section 2 we introduce massive spins s 2 and their gauge invariances. We also

summarize the higher integer spin wave operators of Lichnerowicz [7] and provide their
generalizations to half integer spins. In Section 3, we study canonical (anti)commutators
for s 2 and show how nonunitary regions arise. In Section 4 we examine the partially
massless m2 = 2/3, s = 2 theory, demonstrating by Hamiltonian analysis that it describes
propagating helicities (2, 1) only. Section 5 deals with all spins: We first present a
counting argument indicating that partial masslessness occurs for all higher spins in (A)dS.
Then we display explicit new Bianchi identities and partially massless gauge invariances
for s = 5/2, 3. Appendix A extends the dimensional reduction derivation of massive higher
spin field equations in flat space from their massless dimension d = 5 predecessors [8], to
obtain massive constraints from the d = 5 Bianchi identity for s = 5/2, 3 examples. Some
future directions and open problems are discussed in the Conclusion. A brief version of
some of our results was given in [3].
2. Field equations and identities in constant curvature spaces

The Riemann tensor in constant curvature spaces is
R =
2
g[ g ] ;
3
(1)
580
the cosmological constant is positive in dS and negative in AdS. The actions of

commutators of covariant derivatives are summarized by the vector-spinor example 4
2
g[ ] + .
(2)
3
6
The actions and field equations for massive spins 0 s 2 in constant curvature
backgrounds 5 are
[D , D ] =
L(0) =
L(1) =
1
g G,
2
1
g G ,
2
L(1/2) = g R,
L(3/2) = g R ,
1
g G ,
2

G = (0) m2 ,

R = D(1/2) + m ,

G = (1) m2 D D.,

R D = D(3/2) D
/ m ,

G (2) m2 + g +
L(2) =
+ D( D) 2D( D.) + g D.D..
(3)
(4)
(5)
(6)
(7)
(8)
The operators (n) are the wave operators of [7]
,
(3) D 2 5 + 2g( )

8
14 g ,
(2) D 2
3

(1) D 2 ,
D ,
(0)
(9)
(10)
(11)
(12)
whose introduction is justified by the following identities,

(3) D( ) = D( (2) ) ,
D (3) = (2)D. ,
(3) g( ) = g( (1) ) ,
g (3) = (1) ,
(2) D( ) = D( (1) ) ,
D (2) = (1) D. ,
(2) g = g (0) ,
g (2) = (0) ,
4 Our metric is mostly plus, Dirac matrices are mostly hermitean and the Dirac conjugate is
i 0 .
We denote (anti)symmetrization with unit weight by round (resp. square) brackets. Antisymmetrized products of
Dirac matrices are given by 1 ...n [1 n ] .
5 For s = 2 in generic gravitational backgrounds, the minimally coupled PauliFierz action does not yield field
equations with the correct 2s + 1 = 5 PDoF [9]. Recently [10], s = 2 actions have been constructed with the
correct PDoF count in background Einstein spaces (R = g ).
(1) D = D (0) ,
D (1) = (0) D..
581
(13)
Here ... is a symmetric tensor. The half integer spin generalizations D(n/2) of these
operators are given by [11]
D(5/2) 3D
/ 2D( .) + 2((D
/ .) D.) ),
D
(3/2)
(14)
2D
/ D . + (D
/ . D.) = D + D
/ ,
(1/2)
D,
/
(15)
(16)
with a symmetric spinor-tensor, and satisfy analogous identities

D(5/2)D( ) = D( D(3/2)) ,
D D(5/2) = D(3/2) D. ,
D(5/2)( ) = ( D(3/2) ) ,
D(5/2) = D(3/2) . ,
D(3/2) D = D D(1/2) ,
D D(3/2) = D(1/2) D.,
D(3/2) = D(1/2),
D(3/2) = D(1/2) ..
For completeness we include the familiar Weitzenbock identity,

(1/2)2

D
= D 2 .
(17)
(18)
Notice that we have written the massive RaritaSchwinger field equation (7) in terms of
the s = 3/2 operator D(3/2) with explicit mass term as well as in a more compact form
involving the operator

m
[D , D ] = [D , D ] + m2 /2 ,
D D + ,
(19)
2
encountered in cosmological supergravity [12].
For s 1 there are more relativistic field components than PDoF. As usual, the correct
PDoF count is obtained by studying the constraints implied by divergences and (gamma-)
traces of the field equations. Less usual, for special lines in the (m2 , ) plane, these
constraints are satisfied identically and become Bianchi identities associated with gauge
invariances. Explicitly, the divergence of the field equations (6)(8) read
D.G = m2 D.,
D.R =
D.G = m2 (D. D ),

1 2
3m + .,
2
(20)
and are constraints for generic values of the parameters m2 and . For s = 1, 2, the value
m2 = 0 yields Bianchi identities and their associated (electromagnetic and general
coordinate) gauge invariances
= D
and = D( ) ,
(21)
respectively. Both m2 = 0 theories are strictly massless: they propagate with two physical
helicity states. For s = 3/2, the sole gauge invariance
1
/3 ,
= D = D +
(22)
2
582
is inherited from cosmological supergravity and occurs for m2 = /3 in AdS. This

(rather than the m = 0 model) is the strictly massless helicity 3/2 theory.
For s = 2 there is a new effect: the field equation has two open indices and thereby also
admits a double divergence Bianchi identity. The double divergence constraint

1
1
D.D.G + m2 G = m2 3m2 2 ,
2
2
(23)
becomes a Bianchi identity not only along the strictly massless line m2 = 0, but also along
the dS gauge line m2 = 2/3 corresponding to the Weyl-like theory of [1]. The relevant
gauge invariance

= D( D) + g ,
(24)
3
may be employed to show that this partially massless model propagates with helicities
(2, 1) on the null cone in dS. We discuss this theory further in Section 4.
To summarize, the (m2 , ) half-plane offers no surprises for spins s 1, the usual null
propagating massless theories inhabit the line m2 = 0 and all other points m2 0 describe
2s + 1 massive DoF. For s = 3/2, the massless line is m2 = /3 and bisects the (m2 , )
half-plane as depicted in Fig. 1. For s = 2, the massless theory again lies on the axis m2 = 0
and a new gauge invariance emerges at m2 = 2/3 which also divides the half-plane into
two distinct physical regions.
3. (Anti)commutators and nonunitary regions

Our analysis is rather simple and harks back to the original inconsistency of the
local field theory of charged spin 3/2 particles [4]. Generically,
given (anti)commutation
(with a = (x +i x)/
2) and a vacuum 6 |0 =

relations {, } = = [a, a ] = i[x, x]
0 = a|0, positivity of norms requires > 0. For quantum field theories, exactly the same
criterion can be applied, but now in a distributional sense.
The local canonical (anti)commutators for quantum fields with s 2 in cosmological
spaces were presented long ago [7]. Let us summarize those results: For s = 0, 1/2 one has

(x), (x ) = iD (0) x, x ; m2 ,
(25)

) = iS (1/2)(x, x ; m).
(x), (x
(26)
The distributions on the right hand sides of (25) and (26) are general coordinate/local
Lorentz biscalars and bivectors [14], respectively, and share the symmetry properties of
the (anti)commutators on the left hand sides

S (1/2)(x, x ) 0 = 0 S (1/2) (x , x) .
D (0) (x, x ) = D (0) (x, x ),
(27)
6 It has recently been suggested that the non-perturbative definition of quantum gravity in dS be reexamined
[13]; in particular the definition of the vacuum requires careful consideration. Our local quantum field theoretic
computation ignores such subtleties.
583
We often drop the the labels m2 and m used to indicate that these distributions satisfy
the field equations of the onshell quantum fields (x) and (x)

(0)

(0)
2
x, x ; m2 ,
x m2 D (0) x, x ; m2 = 0 = (0)
(28)
x m D
(1/2)
(1/2)

(1/2)
D
(29)
+m S
(x, x ; m) = 0 = S (1/2) (x, x ; m) D x + m .
Here the backwards arrow over an operator O is defined by
O O.
(30)
The boundary conditions for these distributions are
1
d
3 (
D (0) (x, x )
x 0 =x 0 =
x x ),
0
dx
g
0 3
S (1/2) (x, x )
x 0 =x 0 =
(
x x ),
g
D (0) (x, x )
x 0 =x 0 = 0,
(31)
for any choice of timelike coordinate x 0 .

Note that in general, the (anti)commutator distributions can be written as the difference
of advanced and retarded propagators [7]. In the Minkowski limit, the scalar distribution
D (0) (x, x ; m2) is the usual PauliJordan commutator function.
For s 1, away from the gauge invariant boundaries, the constraints (20) and (23) imply
the (gamma-)tracelesstransverse conditions
D. = 0,
D. = 0 = .,
D. = 0 = ,
which allow the field equations (6)(8) to be rewritten as

(1)
m2 = 0,

1 (3/2)
D
+ m = 0,
2

(2)
2
m + 2 = 0.
(32)
(33)
We must now write (anti)commutators for fields satisfying both (32) and (33). The latter
requirement is just the analog of the s = 0, 1/2 solutions given above. The former is easily
imposed using the (gamma-)tracelesstransverse decompositions
T = D
1
D.,
D2
D. T = 0;
1
1
TT = . + D
/ . 4D.),
(D
2
4
3D +
D. TT = 0 = . TT ;
(34)
(35)
584
2
1
(D.) )T g
+
4

4
D.D. 14 D 2 ,
D{ D} 2
D (3D 2 + 4)
TT
= D(
D2
D.TT = 0 = TT ;
(36)
1
where a tilde over an index denotes its gamma-traceless part, i.e., X
X 4 .X
and { } denotes the symmetric-traceless part of any symmetric tensor, i.e., X{}
X() 14 g X .
Therefore the (anti)commutators for spins 1 s 2 are given by

i

(1)
x, x ; m2 + 2 Dx Dx D (0) x, x ; m2 ,
(x), (x ) = iD
(37)
m

i

(3/2)
(x), (x ) = iS
(x, x ; 2m) x S (1/2)(x, x ; 2m)x
4
i

(38)
Dx S (1/2)(x, x ; 2m)Dx ,
+
3m2 +

2i

(2)
(1)
(x), (x ) = iD,
x, x ; m2 2 + 2 Dx Dx D
x, x ; m2 2
m

x x
x x x x
i
+ 2
g
2D D D D + m2 m2 g
2
m (3m 2)

x
x
+ m2 Dx Dx g
+ m2 g
Dx Dx D (0) x, x ; m2 2 . (39)
For brevity, we have suppressed obvious symmetrizations () and ( ) on the right-hand

side of (39). The field equations (33) have been used throughout to eliminate any factors D 2
appearing in the (gamma-)tracelesstransverse decompositions (34)(36), since the higher
distributions are also onshell
(n)

+ m2 D(n)
(40)
x, x ; m2 0,
1 ...n ,1 ...n
(n+1/2)
(n+1/2)

+ m S1 ...n ,1 ...n (x, x ; m) 0.
D
(41)
x (1/2) (x, x ; 2m) = D x S (1/2) (x, x ; 2m) is also useful. The (anti)commuThe identity D
S
tators (37)(39) are the difference between advanced and retarded propagators [7]. The
higher distributions D (1) , D (2) and S (3/2) above satisfy

(1)
Dx D
x, x , m2 = Dx D (0) x, x ; m2 ,
(3/2)
(x, x ; m)Dx = Dx S (1/2)(x, x ; m),
S
(3/2)
(x, x ; m)x = x S (1/2) (x, x ; m),
S

(2)
(1)
x
Dx D,
x, x , m2 = D(
D) x, x ; m2 ,

(2)
x
gx D,
D (0) x, x , m2 ,
x, x , m2 = g
with boundary conditions
g 3
d
(1)

(
D (x, x )
=
x x ),
0
dx
g
x 0 =x 0
(42)
(1)
D
(x, x )
x 0 =x 0 = 0,
d
(2)

D
(x,
x
)
,
0 0
0
dx
x =x
D (2) (x, x )
0 0 = 0.
,
585
g 0 3
(3/2)
S
(x, x )
x 0 =x 0 =
x x ),
(
g
g(( g )) 3
(
=
x x ),
g
x =x
(43)
Equipped with the above tools, one can easily uncover the nonunitary regions. Starting
with the (anti)commutators (37)(39), the aim is to determine whether the distributions on
their right-hand sides have definite sign in the dangerous lower helicity sectors.
For concreteness we work in the simple synchronous dS metric

M /3,
ds 2 = dt 2 + e2Mt d x 2 ,
(44)
and concentrate on the equal time (anti)commutators of the time components of the fields
(and their time derivatives). While the metric (44) does not cover the entire dS space, nor
is it real when continued to negative AdS values of , these disadvantages are outweighed
by its simplicity (we will consider more general coordinate systems later). Selecting the
lowest helicity components by looking at time components of fields, a simple computation
reveals that

2 i
3 (
x x ),
0 (t, x ), 0 (t, x ) = 2
g
m

2
1
3 (
x x ),
0 (t, x), 0 (t, x ) = 2
3m + g

2 4
i
3 (
00 (t, x), 00 (t, x ) = 2
x x ),
m (3m2 2) g

(45)
(46)
(47)
where 2 g ij i j = e2Mt 2 is a negative operator. The final equation (47) agrees with
the detailed massive s = 2 Dirac analysis presented in [2]. Our derivation only requires
writing out the Laplace and Dirac operators explicitly in the metric (44) and using the
field equations (33) to maximally eliminate time derivatives, after which the boundary
conditions (31) and (43) may be applied.
The interpretation of Eqs. (45)(47) is as follows: Positivity of the distributions on the
right-hand sides is completely determined by the respective denominators m2 , 3m2 +
and 3m2 2, precisely the factors appearing in the Bianchi identities of the previous
section. For the various spins, we learn:
Spin 1: The model is unitary in the entire (m2 , ) half-plane. A line of gauge invariant
models emerges at m2 = 0 (the same value as in flat space).
Spin 3/2: The model is unitary in the region m2 > /3 which includes the
Minkowski background. Strictly massless, gauge invariant unitary models are found
along the AdS line m2 = /3. The region m2 < /3 is nonunitary. In contrast
to flat space, the m2 = 0 theories are massive when = 0 and even nonunitary for
negative (AdS) values of .
586
Spin 2: Models with m2 > 2/3 are unitary. There are now two lines of gauge
invariant theories; the usual linearized cosmological Einstein theory at m2 = 0 and
a partially massless theory [1] at m2 = 2/3. Both are unitary but the region m2 <
2/3 is not.
These conclusions are depicted in Fig. 1.
Finally, as promised, we address the concern that, strictly the metric (44) applies only
to dS. On the one hand, given that (i) the final results are a function of the real variable
only and (ii) the picture presented here is backed up by the emergence of Bianchi identities,
there can be no doubt of its correctness. However, for complete certainty, we repeat, as an
example, the s = 3/2 computation in the metric (M /3 )

2
1
2
2
2
2
2
2
2
.
ds = dt + cosh (Mt) dr + 2 sin (Mr) d + sin d
(48)
M
Upon rescaling r /M, the three-metric d 2 = d 2 + sin2 (d 2 + sin2 d 2 ) is seen
to describe a unit three-sphere. We prefer the initial parametrization, however, since for
pure imaginary values of M, the cosmological constant is negative and the metric
continues to AdS. Performing a calculation analogous to the one above we find

0 (t, r, , ), 0 (t, r , , )
=
cosh2 (Mt)((3)D 2 /4) 1

(r r )( )( ).
g
3m2 +
(49)
The operator (3)D 2 is the square of the intrinsic 3-dimensional covariant derivative
(LaplaceBeltrami operator) acting on a spinor. In dS the operator (3)D 2 /4 is not
manifestly positive. However (in our parametrization) the eigenvalues of (3)D 2 acting on
spinors are (/3)(l(l + 2) + 1/2) with l 1/2 (see, e.g., [2]), and the highest eigenvalue
is precisely /4. Hence the operator (3)D 2 /4 is indeed positive and in dS we may
draw precisely the conclusions given above. Now, continuing the metric (48) to AdS space
the same result holds for the local anticommutator except the 3-space is a hyperboloid.
Nonetheless (assuming we can neglect spatial boundary terms), both (3)D 2 and /4
are now separately positive, and unitarity is determined by the sign of the denominator
3 + m2 . This concludes our derivation of the unitarily forbidden regions for spins s 2.
We emphasize that once one knows the gauge lines and their corresponding Bianchi
identities, our unitarity results in fact follow by inspection: whenever a coefficient in
a massive constraint vanishes and then becomes negative, all the corresponding lower
helicity modes are first excised by the accompanying gauge invariance and thereafter
reemerge with opposite norms. Therefore, starting from the unitary Minkowski region,
it is easy to map out the unitarily allowed and forbidden regions, as shown in Fig. 1.
Furthermore, for higher spin partially massless theories to be unitary, the ordering criterion
for the gauge lines, discussed in the introduction, must hold. A simple example is provided
by the s = 2 strictly massless (m2 = 0, linearized graviton) theory for > 0: to reach it
starting from the unitary Minkowski region, one must pass through the unitarily forbidden
region 0 < m2 < 2/3. Nonetheless, the theory is unitary, since the highest helicities 2
587
are left untouched by the unitarity flip of the helicity 0 mode across the m2 = 2/3 gauge
line.
4. Partially massless spin 2: canonical analysis

At the partially massless dS boundary, m2 = 2/3 = 2M 2 , we showed that the scalar
constraint (23) is a Bianchi identity; as such it removes 2 DoF, leaving 4 physical DoF
corresponding to helicities 2, 1. In this section we prove this claim via an explicit
canonical analysis. 7 Our method is similar to that originally used to prove the stability of
massless cosmological gravity [17].
A possible starting point is the second order massive s = 2 action in Eq. (3). Equivalently
(and much simpler) one can begin with the first order ADM form of cosmological Einstein
gravity,

(3)
S+E = d 4 x ij g ij + N g R 2 + 2Ni Dj ij

N
1
+ ij lm gij glm gil gj m ,
(50)
g
2
then linearize around a dS background and add by hand an explicit mass term. Here g is the
determinant of the 3-metric gij and N (g 00 )1/2 , N0i g0i . We take the synchronous
dS metric (44), denoted ds 2 = dt 2 + gij dx i dx j in this section, reserving gij for the
dynamical 3-metric which we linearize as
gij gij + ij ,
gij f 2 (t)ij ,
f (t) eMt .
(51)
N 1 + n.
(52)
The remaining fields are linearized as

ij ij + P ij ,
ij 2Mf ij ,
(The background metric is block diagonal so no expansion is needed for Ni .) In terms of

these deviations, the mass term is

2

m2
Sm =
(53)
d 4 x g g g g ,
4
here 0i Ni , g det gij = det g and 00 g00 + g00 = (1 + N 2 ). The final
action is the sum S = S+E + Sm , discarding any terms of higher than quadratic order
in (dynamical) fields.
Notice that the only explicit time dependence of the integrand of (53) is through f 1 .
Indeed it proves useful to make the field redefinition
ij f 1/2 hij ,
Ni f 1/2 ni ,
P ij f 1/2 pij ,
n = f 3/2 n.
(54)
7 A detailed canonical analysis of massive s = 2 for general m2 is given in [15]; an early attempt can be found
in [16].
588
The cost is an extra contribution generated by the symplectic term of (50), P ij ij

pij h ij + (M/2)pij hij . A dividend is that the only explicit time dependence in what follows
will be through 2 g ij i j f 2 02 . Index contractions are just with ij , all quantities
are now in (3 + 1) form.
Next examine the mass term

m2 2
m2 2
4
2
2
h hii +
N + m nhii .
Sm = d x
(55)
4 ij
2 i
Were it not for the term proportional to Ni2 , the field Ni would be a Lagrange multiplier for
3 constraints (as is the case for the m = 0 strictly massless theory). Instead, when m = 0
we must integrate out Ni via its algebraic equation of motion. The field n, however, only
appears linearly and remains a Lagrange multiplier for the constraint

1
(3)
(56)
g R 2 + m2 hii + ij lm 12 gij glm gil gj m
= 0.
g
linearized
For generic values of (m2 , ), this constraint eliminates one degree of freedom from the
6 pairs (pij , hij ) leaving 5 physical helicities (2, 1, 0). Our aim now is to show that a
(single) further constraint emerges at the gauge invariant value m2 = 2M 2 .
We decompose the fields hij and pij according to their helicity and drop (for now) the
helicity 2 traceless-transverse (htijt , pijt t ) and helicity 1 transverse (hti , pit ) modes since
they manifestly decouple from the 2 remaining helicity 0 modes (to the quadratic order
used here). The latter are defined by the projection

i j
i j
1
hij =
(57)
ij 2 hT + 2 hL ,
2
0
0

and similarly for pij . (Note that under the integral Aij Bij = 12 AT BT + AL BL .) In
terms of these variables the linearized constraint is

C 2 hT + 2M(pT + pL ) m2 2M 2 (hT + hL ) = 0.
(58)
Note that the leading term comes from the linearized 3-dimensional curvature scalar and
that both the cosmological 2 g and momentum squared terms in (56) contribute to
the final term in (58), which vanishes on the critical gauge line m2 = 2M 2 . Henceforth we
concentrate on the critical case and eliminate m2 via this relation.
Next we write out the quadratic action S = SE+ + Sm remembering the constraint (58)
which we solve as
2
hT .
2M
Observe that the symplectic terms become
pT = pL
(59)
1
1
1
pij h ij = pT h T + pL h L = pL q hT 2 hT ,
(60)
q hL hT
2
4
2
(suppressing integrations throughout). Upon eliminating hL = q + hT /2 in favor of q and
hT , the action depends only on the 3 variables (pL , q, hT ) and its most general form is
1
S(pL , q, hT ) pL q = Ah2T + BhT + C,
2
(61)
589
where A is constant, B linear and C quadratic in (pL , q). If A = 0, we have an additional

constraint B(pL , q) = 0 and no zero helicity PDoF remain, whereas for non-zero A, hL can
be removed via its algebraic field equation leaving behind one zero helicity DoF. In fact,
A does vanish on the critical line m2 = 2/3 so this model describes helicities (2, 1)
only. Indeed, a lengthy calculation yields

1 pL
pL
2
q hT +
q
S(pL , q, hT ) pL q =
2 M
M
2

2
3
2
(62)
pL m q .
m2 2
As claimed, A = 0 and even the zero helicity Hamiltonian vanishes once the Lagrange
multiplier hT is integrated out.
Let us now examine the remaining helicities (2, 1). A series of canonical transformations yields a simple action

1
M2
S(2,1) =
(63)
p q p2 + q 2
q .
2
4
=(2,1)
Notice again, all time dependence is through 2 in the Hamiltonian. The field equations
are

d2
M2
2 + 2
p = q ,
(64)
q = 0.
dt
4
The covariant field equation (33) evaluated at m2 = 2M 2 is (D 2 4M 2 ) = 0. Consider,
for example, helicities 2, for which i ij = 0 = i i . In this frame the transverse-traceless
part of the covariant field equation reads

d2
d
2 + M + 2 ij = 0.
(65)
dt
dt
The action (63) was obtained by the same rescaling as in (54), namely, ij = f 1/2 qij . The
factor f 1/2 is precisely the integrating factor which removes the single time derivative from
Eq. (65) at the cost of a term M 2 /4, i.e., Eqs. (65) and (64) are identical (helicities 1
agree via a similar calculation).
Stability of the partially massless theory requires that it possess a conserved, positive,
energy function. The latter can be obtained by an argument similar to that given in [17] for
the strictly massless s = 2 theory: the Hamiltonian in (63) is not conserved because of the
explicit time dependence of 2 . However, inside the intrinsic dS horizon at (f Mx i )2 = 1,
the background metric (44) possesses a timelike Killing vector

2

= 1, Mx i 2 = 1 + f Mx i .
(66)
Therefore, the energy associated with time evolution in this Killing direction

1
E = T 0 = H Mx i p i q i (p q )
2
(67)
590
satisfies E = 0 (H is the Hamiltonian in (63) and we have suppressed the sum over
helicities ). Furthermore, writing out E explicitly and relabeling the variable p
p + (3M/2)q gives
E=
2

1

1 i 2 1 1
x p + f i q f M|x| x i p f 1 i q + 2M 2 q2 ,
2
2
2
(68)
with x i |x|x i . The last (mass) term is manifestly positive and the first three terms are
positive by the triangle equality whenever
f M|x| < 1,
(69)
that is, inside the physically accessible region.

A final interesting feature of the partially massless s = 2 theory is null propagation. The
dS metric is conformally flat and it can be shown that the m2 = 2/3 theory propagates
on its null cone [1]. An interesting open question is whether any of the s > 2 partially
massless theories which we discuss next, share this behavior.
5. Higher spins
Having seen that the s = 2 field equation G yields both single divergence, D.G , and
double divergence, D.D.G, Bianchi identities we are led to inquire whether even higher
divergence Bianchi identities occur for s > 2. The answer is yes. This implies that, in
addition to the usual massive and strictly massless possibilities, a spin s field in (A)dS can
be partially massless with propagating helicities (s, (s 1), . . . , (s t)) (t < s). We
first demonstrate this claim by simple counting arguments and then write explicit gauge
invariances and Bianchi identities for s = 5/2, 3 examples.
5.1. Higher spin bosons
Define the number of components ( s ) of a symmetric s-index tensor 1 ...s
(s+1)(s+2)(s+3)
, s > 0,
3!
(s )
0,
otherwise.
(70)
Then a traceless symmetric tensor has ( s )T components

( s )T = ( s ) ( s 2 ) = (s + 1)2 , s 0 ,
(71)
while a doubly traceless (0 = ... ) one has

( s )TT = ( s ) ( s 4 ) = 2 s 2 + 1 , s 1 .
(72)
Strictly massless bosons

First consider massless bosons. The field content is a doubly traceless s index symmetric
tensor enjoying gauge invariances with an s 1 index traceless gauge parameter:
Fields
Gauge
591
( s )TT
2.( s 1 )T
Since 2( s 1 )T DoF can be gauged away, the PDoF for s 1 are ( s )TT 2( s 1 )T = 2,
corresponding to helicities s.
Massive bosons
For massive s > 2 theories, the massless field content must be augmented by a set
of traceless symmetric auxiliary fields, (, . . . , 1 ...s3 ). Each divergence of the sindex symmetric field equations is a constraint whenever the remaining open indices are
traceless:
Fields
+ Auxiliaries
( s )TT
( 0 )T + + ( s 3 )T
Constraints
( 0 )T + + ( s 3 )T + ( s 2 )T + ( s 1 )T
and the PDoF are ( s )TT ( s 2 )T ( s 1 )T = 2s + 1, the sum of all helicities.

Partially massless bosons
For partially massless higher spin theories the field content is the same as the massive
case but new Bianchi identities appear. There are as many of these as possible divergences
of the s-index symmetric field equations. On a gauge line where a constraint with t
divergences becomes a Bianchi identity we have:
Fields
+ Auxiliaries
( s )TT
( 0 )T + + ( s t )T + ( s t + 1 )T + + ( s 3 )T
Constraints
Gauge
( s t + 1 )T + + ( s 3 )T + ( s 2 )T + ( s 1 )T
2.( s t )T
The t-divergence Bianchi identity replaces the set of constraints with t, t + 1, . . . , s

divergences and the corresponding gauge invariance removes 2.( s t )T DoF. Therefore
we find 2t PDoF along with a set (( 0 )T + + ( s 2 t )T ) of leftover auxiliary fields.
When the new Bianchi identity is the scalar one with the maximal number of divergences
t = s, there are 2s PDoF and only the massive helicity 0 mode is removed. For the vector
Bianchi with t = s 1 there are 2s 2 PDoF because helicities (0, 1) are excised. For
even lower values of t, the sum over leftover auxiliaries ( 0 )T + + ( s 2 t )T is
non-empty: In the strictly massless case t = 1, there 2 PDoF (helicities s) and all the
auxiliaries ( 0 )T + + ( s 3 )T remain. However, at least in this case, it is obvious
that they decouple. Less obvious is whether they decouple for the partially massless s
592
4 theories. If not, we must then ask how many PDoF they represent and whether they
are unitary excitations, questions clearly inaccessible via the simple counting arguments
presented here.
5.2. Higher spin fermions
For brevity define s 1/2. The (real) components of a -index, symmetric, gammatraceless (Majorana) field are denoted

4.( ) = 4 ( ) ( 1 ) = 2( + 1)( + 2), 0 ,
(73)
and a traceless gamma-tracelessness one ( . 2 ... = 0) has

4.( ) T = 4 ( ) T ( 3 ) T = 2 3s 2 + 3s + 2 , 0
(74)
components.
Strictly massless fermions
For massless half integer s = + 1/2 the field content is a traceless gamma-traceless,
symmetric -index spinor. The correct number of PDoF is ensured by gauge invariances
with a gamma-traceless, symmetric ( 1)-index spinor parameter:
1/2
Fields
4.( ) T
1/2
Gauge
4.( 1 )
Gauge
4.( 1 )
To explain: we begin with 4.( ) T fields, of which gauge invariance removes 4.( 1 )
components. However, before excising further DoF via residual invariances we must
impose the projector field equations (recall that the Dirac equation divides the DoF of a
spinor by 2). Thereafter residual gauge invariances remove 4.( 1 ) further DoF. Hence
the PDoF count is 2[( ) T 3( 1 ) ] = 2, corresponding to helicities s.
Massive fermions
In addition to the same fields as for massless fermions, there are traceless auxiliary
spinor fields (, . . . , 1 ... 2 ) when s 3/2. All possible divergences on vector
indices of the field equations yield constraints when the leftover indices are gammatraceless:
1/2
Fields
4.( ) T
1/2
Auxiliaries
4.( 0 )T + + 4.( 2 )T
Constraints
4.( 0 ) + + 4.( 1 )
Adding these up yields 2 + 2 = 2s + 1 PDoF and all helicities (s, . . . , 1/2) propagate.
593
Partially massless fermions

The field content is identical to massive case but now suppose that t divergences of the
field equations yield a Bianchi identity. We have:
Fields
4.( ) T
1/2
1/2
Auxiliaries
Gauge
4.( 0 )T + + 4.( 2 )T
4.( t )
Constraints
Gauge
4.( t + 1 ) + + 4.( 1 )
4.( t )
1/2
The sum is 2t + 2[( 0 )T + + ( t 1 )T ]. Again we find 2t propagating helicity states

(s, . . . , (s t)) along with leftover auxiliaries which decouple on the strictly massless
(t = 1) gauge line. (Just as in the bosonic case, for partially massless spins s 7/2, little
is known about the rle of these auxiliaries.)
We next present the explicit s = 5/2, 3 examples. For s = 5/2 we will exhibit one new
gauge line resulting from the fermionic generalization of the s = 2 double divergence
Bianchi identity (23). The resulting helicity (5/2, 3/2) partially massless theory lives
in an AdS region where helicities 3/2 have negative norms and therefore fails to
be unitary. In contrast, for s = 3 we find both double and triple divergence Bianchi
identities corresponding to unitary, partially massless, dS theories of helicities (3, 2)
and (3, 2, 1).
5.3. Spin 5/2
The s = 5/2 spinorial field equation has two open indices, so as for s = 2, there are
two possible Bianchi identities; they appear along the AdS gauge lines m2 = 4/3 and
m2 = /3 (see Fig. 1). The former is the strictly massless theory with helicities 5/2
whereas the novel gauge invariance of the latter removes only the lowest 1/2 leaving
helicities (5/2, 3/2). Since the massless gauge lines all lie in AdS (just as for their s =
3/2 counterpart), the (m2 , ) half-plane is divided into 3 regions. Only the m2 > 4/3
one including Minkowski space, is unitary.
The s = 5/2 action and field equations are
5,
L = g R g R
(75)

/ + g .D. + D( ) 12 gD
/
R = D(5/2) 2D

5
+ m 2( .) 12 g g = 0,
(76)
12
5
/ 3m) = 0.
R5 = (D
(77)
12
They can be derived by minimally coupling the flat space equations of Appendix A.1 to the
cosmological background. The former were obtained by KK descent from their massless
594
counterparts in d = 5. Minimal coupling alone does not provide equations of motion

describing the 6 = 2s + 1 massive PDoF: an additional nonminimal coupling contained
by the term
is necessary. In fact, to achieve a proper set of constraints requires
fixing the auxiliary coupling to

12 2
m + 4/3 .
(78)
5
Here we encounter a new subtlety. Implicitly we have so far assumed that physical models
live in the half plane m2 > 0, since for fermions negative m2 implies a nonhermitean mass
term, and for bosons, one that is unbounded below. While there are regions with both m2 <
0 and the correct sign for anticommmutators, the dynamics is nonunitary there. Therefore
we continue to require m2 > 0 and examine the relation (78). Since hermiticity of the
action (75) demands 2 > 0, we find two regions: (i) m2 > 4/3 > 0, (ii) 0 < m2 <
4/3. Up until now, was a free parameter which we could set to 1. In the region (i),
m2 + 4/3 > 0 so we must take = +1. In region (ii), hermiticity of the action can be
maintained at the cost of changing the sign of to = 1 (the actions are then different
in each region). In either case, we will find that region (ii) is unitarily forbidden.
Before continuing, it is interesting to compare these difficulties to s = 3/2 and the
problem of constructing dS supergravities [18]. As we have shown, s = 3/2 is unitary
for m2 /3 0 and the boundary is the strictly massless AdS theory corresponding
to cosmological supergravity. As one follows the massive theory into dS, the canonical
anticommutators all have the correct sign for unitary representations. In fact, keeping >
0, there is no obstruction at the level of anticommutators to continuing to m2 < 0 until the
branch of the line m2 = /3 with m2 < 0 < is reached. The theory there is formally
supersymmetric (i.e., strictly massless) but the action is no longer hermitean, which is
a example of the general statement that dS supergravities do not exist [18]. One might
speculate that this clash is generic to higher spin fermions.
Returning to the s = 5/2 field equations, for generic (m2 , ) the constraints
2 =
1
C D.R + m .R
4

5 2
5
= m + 4/3 . D ,
4
12

5
5 2
m + 4/3 R (D
/ + 3m)R5
C D.C +
16
16

10
= m2 + /3 ,
3
ensure that the model describes 6 = 2s + 1 massive PDoF. Along the AdS lines
m2 = 4/3,
m2 = /3
(79)
(80)
(81)
(82)
(83)
the constraints (80) and (82) transmute to Bianchi identities with respective (distinct) gauge
invariances

1
( ) ,
= 0,
= D( ) +
(84)
2
3
= D( D ) +
5
g ,
16
1
15 (D
/ + 3 ).
8
595
(85)
The vector-spinor Bianchi identity (79) at m2 = 4/3 implies strict masslessness

(propagating helicities 5/2) since its invariance removes helicities (3/2, 1/2). Notice
also that = 0 on the strictly massless line so, as claimed above, the spinor auxiliary
decouples there. The novel spinor Bianchi identity (82) at m2 = /3 and invariance (85)
removes helicities 1/2 leaving a partially massless theory of helicities (5/2, 3/2).
Once again, the coefficients (m2 + 4/3) and (m2 + /3) appearing in the constraints
(79) and (82) control the positivity of equal time anticommutators. Therefore, since the
gauge lines all lie in AdS, the (m2 , )-plane is divided into 3 regions; only the one
including Minkowski space m2 > 4/3 is unitary. Although the strictly massless, AdS,
m2 = 4/3 theory is unitary, the partially massless one is not, as it fails the line ordering
requirement: starting from the unitary Minkowski region where all norms are positive,
one would like first to traverse the line m2 = /3, but that is only possible in dS with
negative m2 (imaginary values of m violate hermiticity of the action and unitary evolution).
Crossing the AdS strictly massless line m2 = 4/3 first flips the norm of both lower
helicities (3/2, 1/2) so the partially massless AdS theory cannot be unitary. Ironically,
were negative values of m2 not prohibited, we could traverse the lines in the correct order
in dS. This observation lends weight to the speculation that the unitarity difficulties of
partially massless theories are peculiar to half integer spins. Indeed, in the next section we
exhibit the bosonic s = 3 example, which enjoys two partially massless unitary dS lines.
5.4. Spin 3
Spin 3 is the first example of a system with two new Bianchi identities over and above
the usual one at m2 = 0. The action and field equations (with flat space limits derived by
KK descent in Appendix A.2) are
3
1
g G
g G5 ,
(86)
2
8

G = (3) m2 + 16/3 3D( D.) + 3D( D )

3g( (1) m2 + 11/3 ) D.D.) + 12 D) D.
3m
(87)
g( D) = 0,
+
4

3
G5 = (0) 4m2 + 8 + mD. = 0.
(88)
2
The field is a symmetric 3-tensor and the auxiliary field decouples at m = 0
(the strictly massless theory). Fixing the ordering of covariant derivatives as shown and
requiring constraints to remove all but the physical 7 = 2s +1 degrees of freedom, uniquely
specifies all terms with an explicit -dependence. Indeed, we find the following constraints
L=

1
B{} D.G{} = m D{ D} + 2mD.{} 4mD{ } ,
2
(89)
596

m
m4
5
D G5 +
G = m 3m2 4 D + 23 m ,
4
4
8

5
5
B D.B m 3m2 4 G5 = m 3m2 4 m2 2 .
12
2
B D.B
(90)
(91)
The explicit tensorial structures on the right-hand sides of (89)(91) are the covariantizations of the flat space ones in (A.33)(A.35). The new phenomenon is the splitting of
the prefactors to m, (3m2 4) and (m2 2), thanks to the additional parameter .
Therefore, in addition to the usual massless theory at m = 0 there are new gauge invariant
systems at m2 = 4/3 and m2 = 2, since whenever these prefactors vanish, the corresponding constraints in (89)(91) become Bianchi identities with accompanying, respective, gauge invariances
= D( {}) ,
= D( D{ }) +
= 0;
g( ) ,
3
= D( D{ D}) +
g( D) ,
2
D. ;
3

2
10
2
D +
.
=
3 2
3
2
3
(92)
(93)
(94)
The new gauge invariant lines bound regions in the (m2 , ) half-plane whose unitarity
properties are determined by the signs of the prefactors (3m2 4) and (m2 2).
To analyze these new properties, decompose the 7 = 2s + 1 physical DoF into helicities
(3, 2, 1, 0). We find the following phase structure of the (m2 , ) half-plane:
m2 > 2 > 0: This region includes Minkowski space and is clearly unitary. All
helicities (3, 2, 1, 0) propagate with positive norm.
m2 = 2: A partially massless theory appears since the scalar constraint B = 0 is
now a Bianchi identity whose associated gauge invariance removes the scalar helicity
0 excitation. The remaining 6 DoF (3, 2, 1), propagate with positive norm since
they are unaffected by the scalar gauge invariance.
4/3 < m2 < 2: Although all 7 DoF are now again propagating, the scalar helicity
0 mode reemerges from the gauge boundary m2 = 2 with negative norm (since
the factor (3m2 4)(m2 2) appears in canonical commutators as a negative
denominator). This is a unitarily forbidden region.
m2 = 4/3: This partially massless theory has Bianchi identities B = 0 = B whose
gauge invariances excise the helicities (1, 0). The remaining 4 PDoF, helicities
(3, 2) propagate with positive norm.
0 < m2 < 4/3: Again all 7 DoF propagate but now the scalar helicity has again
positive norm since its denominator (3m2 4)(m2 2) is again positive. However,
the region is still unitarily forbidden because now the vector helicities 1 suffer a
negative denominator (3m2 4).
m2 = 0: This is the unitary strictly massless model with tensor Bianchi identity
B{} = 0 = B = B. Only the uppermost helicities 3 remain. An added subtlety
is the remnant decoupled auxiliary field .
597
Notice how the uppermost helicity 3 always emerges unscathed as a pair of positive
norm states but, unlike Minkowski space, splittings into theories with intermediate lower
helicities, rather than only the full complement of 2s + 1 states of the massive theory, are
possible. Furthermore, the ordering of gauge lines is the same as for the s = 5/2 example.
But, unlike for s = 5/2, the lines can be traversed in the order required for unitarity of the
partially massless dS theories without recourse to unphysical, negative, values of m2 .
6. Conclusions
Life in cosmological backgrounds [19] promises to be less degenerate than its
Minkowski limit. In this paper we have examined the notion of mass for higher spins
in (A)dS geometries and found a far richer structure than in flat space, where higher spin
fields are either strictly massless or massive. Once a cosmological constant is present,
the Minkowski gauge invariances implying masslessness split into a set of partial ones
allowing intermediate, partially massless theories along lines in the (m2 , ) plane. Their
unitarity is easily analyzed by examining the line ordering. Although interactions of
relativistic higher spin fields are fraught with many difficulties 8 even in flat backgrounds,
and the observed cosmological scales are far removed from those encountered in (for
example) nuclear physics, this new phenomenon is rather fascinating. However, many open
questions remain, some of which are following.
Perhaps the most pressing is an explicit generalization of our results, beyond simple
counting arguments, to generic values of s. It is clear that partial masslessness applies
to higher values of s when m2 is appropriately tuned to , thereby converting massive
constraints to Bianchi identities. Yet a simple formalism for deducing the pertinent tunings
of of m2 would be extremely useful since the unitarity properties of these partially massless
theories can then be deduced by inspection by studying the ordering of the gauge lines.
Although this problem may appear technically challenging, an interesting approach is
suggested by the flat space KK descent methods presented in Appendix A.
Detailed knowledge of models with s > 3 would reveal whether the failure of the s =
5/2 partially massless theories to be unitary is a universal fermionic characteristic. This
question is also related to the lack of unitary supergravity theories in dS spaces [18]. One
would also like to know whether auxiliary fields of the massless theories decouple, not
only in the strictly massless limit, but also partially massless ones.
Another interesting issue is the higher spin version of the s = 2 van DamVeltman
Zhakarov discontinuity for massless limits of massive s = 2 exchange processes [21].
Clearly, if higher spins are allowed to run in loops, there will generically be discontinuities
thanks to the jump in the virtual DoF over which one functionally integrates [22]. A much
simpler physical problem, whose eikonal limit is purely classical, is that of exchange
processes mediated by massive higher spins. For s = 2, the flat space discontinuity of the
massless limit can be cured by taking the limit in a = 0 background [23]. The same flat
8 For massless fields, it has been suggested that these problems can be alleviated in AdS backgrounds [20].
598
space difficulty extends to s = 3/2 [24] but is again removed in a cosmological background
[11]. 9 The same question can be posed for higher s both in strictly and partially massless
limits.
An interesting feature of the m2 = 2/3 partially massless s = 2 theory is that the
remaining helicity (2, 1) excitations propagate on the null cone of the conformally flat
dS metric [1]. Do partially massless s > 2 theories also exhibit this conformal behavior?
Note added in proof

It has recently been established that all partially massless spins s 3 propagate on the
null cone [26]. Assuming this result also for s > 3 allows a prediction of all the higher
spin partially massless lines. Consequently, the unitary region for massive higher spins is
squeezed around = 0 which provides a novel resolution to the cosmological constant
problem.
Acknowledgements
This work was supported by the National Science Foundation under grant PHY9973935.
Appendix A. Massive constraints from dimensional reduction of bianchi identities

The (A)dS higher spin wave equations presented in this paper were obtained by starting
with their flat space antecedents, coupling minimally to gravity and then adding the
(unique) nonminimal terms required for constraints to produce the correct PDoF count.
Although both massless [6,27] and massive [28] higher spin Minkowski field equations are
available, for our purposes a derivation simultaneously yielding the massive constraints
is desirable. Fortunately, dimensional reduction provides the desired derivation of the
massive field equations from their simpler, d = 5 massless ancestors [8]. The key idea
is that a massless field has the same DoF count as a massive one in one dimension
lower. In this appendix, we summarize the dimensional reduction method and show, in
flat space, that it allows the massive constraints to be derived from the Bianchi identities
of the massless theories. [It would be most interesting to rederive the (A)dS results of our
paper by dimensional reduction in a curved background, a problem we reserve for future
investigations.] The methods reported here for s = 5/2 and s = 3, of course, apply equally
well to other values of s.
9 Curiously the limit m 0, to a nonunitary massive theory, in AdS is also continuous [25].
599
A.1. Free massive spin 5/2

The massless s = 5/2 action and field equations are
1
L = MN RMN ,
2
MN 2(M .N) 2(M .N) + 2(M .

RMN =
N)

R
1
+ MN .. + (M N) 2 MN R = 0,
(A.1)
(A.2)
where, in d = 5, M = (, 5) and Dirac matrices M are defined analogously to our d = 4
.X. The action enjoys the gauge invariance
conventions and X
MN = (M N)
(A.3)
subject to
. = 0,
(A.4)
thanks to the Bianchi identity

1
.RN = N ..R.
(A.5)
5
Before proceeding, observe that the massless field equations (A.2) in d = 5 can be rewritten
in the simple Christoffel form [6,29]
1
2
MN 2(M .N) = 0.
CMN RMN MN RR R (M .RN) =
(A.6)
5
5
We now dimensionally reduce: First, we would like to gauge away components carrying
the index 5, so examine the gauge transformations
55 = 5 5 ,
1
5 = (5 + 5 ).
2
Assuming that 5 is invertible, we replace
5 im,
(A.7)
(A.8)
(A.9)
where m will be the d = 4 mass. Hence, 55 may be gauged away. However, the gammatrace condition on the gauge parameter . = 0 prevents us from doing the same to all
components of 5 . (This is the reason [8] auxiliary fields cannot be avoided beyond s =
2.) Instead, we choose a gauge in which only the (d = 4) gamma-traceless components are
removed 10
1
5 .5 5 = 0,
55 = 0,
(A.10)
4
10 Conversely, it may be tempting to gauge away
5 and retain 55 as an auxiliary field. This gauge choice
does not completely fix the gauge, and the resulting system of equations is invariant under residual gauge
+ m) = 0.
transformations 55 = 2m and = (1/2m) where the gauge parameter satisfies (/
One obtains the correct DoF upon fixing this remaining freedom. In fact there are a wide range of (residual)
gauge invariant equations describing a massive s = 5/2 field depending on the details of the gauge choice made.
The choice (A.10) has the advantage of completely fixing the gauge freedom algebraically.
600
where the d = 4 Dirac matrices are

i 5 ,
5 5 .
(A.11)
In terms of the auxiliary Dirac spinor

i
i 5 5 = ,
4
the equations of motion (A.6) read

i5 C = (/
+ m) 2( .) + 14 ) = 0,

i5 C5 = i m . + 12 + 12 / = 0,
i5 C55 = 2m = 0;
(A.12)
(A.13)
(A.14)
(A.15)
they imply
(/
+ m) = 0,
. = 0 . = 0 = .
(A.16)
These are the usual on-shell conditions for a massive d = 4, s = 5/2 field. In particular
observe that we can identify the constraints as C5 and C55 . We stress that the field
equations in their simplest form (A.13)(A.15) do not directly follow from the variation
of an action. However the massive d = 4 action can be obtained from its d = 5 massless
counterpart by imposing the gauge condition (A.10) and performing the KK descent:
i
1
R5 .
L = R
2
4
(A.17)
Here we have relabeled i5 RMN RMN and i 0 is the d = 4 Dirac

conjugate (remember that 0 = i 0 5 ). The field equations are
R = / 2( . 2( .) + 2(/ .)

+ .. + ( ) 12 / + 12

+ m 2( .) 12 + m = 0,
1
1
1
i
R5 = (/
2m) + . . (/
+ 4m) = 0.
4
2
4
8
After the field redefinition
(A.18)
(A.19)
1
,
(A.20)
8
the auxiliary field couples only via m
and the field equations are then the flat limit
of the ones presented in (76) and (77).
Finally, we show how to deduce the constraints C5 and C55 from the Bianchi identity
(A.5): We need the remaining components R5 and R55 of RMN , since the complete set
RMN inserted in (A.6) provides the equations CMN . The Bianchi identity (A.5) rewritten
as
1
imR5N = RN + N ..R,
5
(A.21)
601
can be used to reconstitute the missing equations, in particular

R5 =
i
R ,
m
(A.22)
and in turn

1
1
R +
(/
m) R + im R5 .
(A.23)
2
2
m
4m
Clearly, since (A.21) is an identity, the system of equations (A.18) and (A.19) along with
(A.22) and (A.23) are guaranteed to be equivalent to (A.13)(A.15) when inserted in the
definition (A.6). It is easy to verify that this calculation combined with the field redefinition
(A.20) yields the flat limit of the constraints (79) and (81). Incidentally, this proves that the
action (A.17) describes the correct 2s + 1 = 6 PDoF.
R55 =
A.2. Free massive spin 3

The s = 3, d = 5 massless Lagrangian and field equations are
1
L = MNR G MNR ,
2
GMNR = 5 MNR 3(M .NR) + 3(M N R)S S

3(MN 5 R)S S ..R)S S + 12 R) .S S = 0.
(A.24)
(A.25)
The action (A.24) has the gauge invariance

MNR = (M NR) ,
(A.26)
with the restriction

N N = 0,
(A.27)
corresponding to the Bianchi identity

1
.GMN = MN .GR R .
5
Again, there is a simpler Christoffel form for the field equations [6]
3
BMNR GMNR (MN GR)S S
5
= 5 MNR 3(M .NR) + 3(M N R)S S = 0.
(A.28)
(A.29)
The massive field content is obtained by algebraically gauging away the components
1
(A.30)
5{} 5 5 = 0.
4
There are no residual gauge invariances; the remaining fields are and the scalar
auxiliary
555 = 0,
55 = 0,
i
i5 5 = .
4
(A.31)
602
Upon KK descent, 5 im the field equations (A.29) become

3m
B = m2 3( .) + 3( ) +
( ) ,
4

i
i
B5 = + im . 2( ) ,
4
2
3
B55 = m m2 ,
2
B555 = 3im2,
from which the usual massive s = 3 on-shell conditions follow immediately,

= 0 = = . .
m2 = 0,
(A.32)
(A.33)
(A.34)
(A.35)
(A.36)
Clearly B{}5 , B55 and B555 are constraints. The d = 4 massive action principle,
guaranteed to yield (A.32)(A.35) by virtue of the Bianchi identity, is
3i
1
L = G + G5 .
2
8
The d = 4 massive field equations read
(A.37)

3m
( )
G = m2 3( .) + 3( ) +

4

1
(A.38)
3( m2 ) ..) + . ,
2

3i
9
3m
G5 = 4m2 +
..
(A.39)
4
8
4
These agree with the flat limit of (87) and (88). The Bianchi identity (A.28) can be rewritten
as
1
imG5MN = GMN + MN .GR R
5
and yields the missing field equations
(A.40)
i
G{} ,
(A.41)
m

1
im
G5 ,
G55 = 2 G {}
(A.42)
4
m

i
1
i
2
G .
m
G5
G555 = 3 G {}
(A.43)
m
4m2
4m
Substituting these in the constraints B{}5 , B55 and B555 following from (A.29) yields
the flat limit of the ones given in (90)(92).
G{}5 =
References
[1] S. Deser, R.I. Nepomechie, Phys. Lett. B 132 (1983) 321;
S. Deser, R.I. Nepomechie, Ann. Phys. 154 (1984) 396.
603
[2] A. Higuchi, Nucl. Phys. B 282 (1987) 397;

A. Higuchi, Nucl. Phys. B 325 (1989) 745;
A. Higuchi, J. Math. Phys. 28 (1987) 1553.
[3] S. Deser, A. Waldron, Gauge invariances and phases of massive higher spins in (A)dS, hepth/0102166.
[4] K. Johnson, E.C. Sudarshan, Ann. Phys. 13 (1961) 126.
[5] G. Velo, D. Zwanziger, Phys. Rev. 186 (1969) 1337;
A. Shamaly, A.Z. Capri, Ann. Phys. 74 (1972) 503;
S. Deser, V. Pascalutsa, A. Waldron, Phys. Rev. D 62 (2000) 105031, hep-th/0003011.
[6] B. de Wit, D.Z. Freedman, Phys. Rev. D 21 (1980) 358.
[7] A. Lichnerowicz, Institut des Hautes tudes Scientifiques 10 (1961) 293;
A. Lichnerowicz, Bull. Soc. Math. France 92 (1964) 11.
[8] C. Aragone, S. Deser, Z. Yang, Ann. Phys. 179 (1987) 76.
[9] D.G. Boulware, S. Deser, Phys. Rev. D 6 (1972) 3368.
[10] I.L. Buchbinder, D.M. Gitman, V.A. Krykhtin, V.D. Pershin, Nucl. Phys. B 584 (2000) 615,
hep-th/9910188;
I.L. Buchbinder, D.M. Gitman, V.D. Pershin, Phys. Lett. B 492 (2000) 161, hep-th/0006144.
[11] S. Deser, A. Waldron, Phys. Lett. B 501 (2001) 134, hep-th/0012014.
[12] P.K. Townsend, Phys. Rev. D 15 (1977) 2802;
S. Deser, B. Zumino, Phys. Rev. Lett. 38 (1977) 1433.
[13] E. Witten, Talk presented Strings 2001, Tata Institute, Mumbai, 2001.
[14] B.S. DeWitt, R.W. Brehme, Ann. Phys. 9 (1960) 220;
B.S. DeWitt, R.W. Brehme, Phys. Rev. 4 (1960) 317.
[15] S. Deser, A. Waldron, Stability of massive cosmological gravitons, hep-th/0103255.
[16] I. Bengtsson, J. Math. Phys. 36 (1995) 5805, gr-qc/9411057.
[17] L.F. Abbott, S. Deser, Nucl. Phys. B 195 (1982) 76.
[18] K. Pilch, P. van Nieuwenhuizen, M.F. Sohnius, Commun. Math. Phys. 98 (1985) 105.
[19] P. Riess et al., Astron. J. 116 (1998) 1009, astro-ph/9805021;
P.M. Garnavich et al., Astrophys. J. 509 (1998) 74, astro-ph/9806391;
S. Perlmutter et al., Nature 391 (1998) 51, astro-ph/9712212.
[20] M.A. Vasiliev, Higher spin symmetries, star-product and relativistic equations in AdS space,
hep-th/0002183;
L. Brink, R.R. Metsaev, M.A. Vasiliev, Nucl. Phys. B 586 (2000) 183, hep-th/0005136.
[21] H. van Dam, M. Veltman, Nucl. Phys. B 22 (1970) 397;
V.I. Zakharov, JETP Lett. 12 (1970) 312;
L.D. Faddeev, A.A. Slavnov, Theor. Math. Phys. 3 (1970) 18;
S.K. Wong, Phys. Rev. D 3 (1971) 945.
[22] F.A. Dilkes, M.J. Duff, J.T. Liu, H. Sati, Quantum M 0 discontinuity for massive gravity
with a Lambda term, hep-th/0102093.
[23] I.I. Kogan, S. Mouslopoulos, A. Papazoglou, The m 0 limit for massive graviton in
dS(4) and AdS(4): how to circumvent the van DamVeltmanZakharov discontinuity, hepth/0011138;
M. Porrati, Phys. Lett. B 498 (2001) 92, hep-th/0011152.
[24] S. Deser, J.H. Kay, K.S. Stelle, Phys. Rev. D 16 (1977) 2448.
[25] P.A. Grassi, P. van Nieuwenhuizen, Phys. Lett. B 499 (2001) 174, hep-th/0011278.
[26] S. Deser, A. Waldron, Null propagation of partially massless higher spins in (A)dS and
cosmological constant speculations, hep-th/0105181.
[27] C. Fronsdal, Phys. Rev. D 18 (1978) 3624;
J. Fang, C. Fronsdal, Phys. Rev. D 18 (1978) 3630;
T. Curtright, Phys. Lett. B 85 (1979) 219.
604
[28] L.P. Singh, C.R. Hagen, Phys. Rev. D 9 (1974) 910;

L.P. Singh, C.R. Hagen, Phys. Rev. D 9 (1974) 898;
S.-J. Chang, Phys. Rev. 161 (1967) 1308;
F.A. Berends, J.W. van Holten, P. van Nieuwenhuizen, B. de Wit, Nucl. Phys. B 154 (1979)
261.
[29] T. Damour, S. Deser, Ann. Poincare Phys. Theor. 47 (1987) 277.

Large-n critical behavior of O(n) O(m)

spin models
Andrea Pelissetto a , Paolo Rossi b , Ettore Vicari b
a Dipartimento di Fisica dellUniversit di Roma I and I.N.F.N., I-00185, Roma, Italy
b Dipartimento di Fisica dellUniversit di Pisa and I.N.F.N., I-56127, Pisa, Italy
Abstract
We consider the LandauGinzburgWilson Hamiltonian with O(n) O(m) symmetry and
compute the critical exponents at all fixed points to O(n2 ) and to O( 3 ) in a = 4 d expansion.
We also consider the corresponding non-linear -model and determine the fixed points and the
critical exponents to O( 2 ) in the = d 2 expansion. Using these results, we draw quite general
conclusions on the fixed-point structure of models with O(n) O(m) symmetry for n large and all
2 d 4. 2001 Elsevier Science B.V. All rights reserved.
PACS: 05.70.Jk; 64.60.Fr; 75.10.Hk; 11.10.Kk; 11.15.Pg
Keywords: Critical phenomena; Frustrated models; O(m) O(n)-symmetric models; Field theory; 1/n
expansion; -expansion
1. Introduction
The critical behavior of frustrated XY and Heisenberg spin systems with noncollinear
order has been the subject of many recent theoretical studies, where the standard tools
of renormalization-group (RG) theory have been applied to field theories which were
conjectured to be appropriate for the description of the systems under investigation (see,
e.g., Refs. [1,2] for reviews on this issue).
The critical behavior of these systems is rather controversial. Indeed, while experimentally there is good evidence of a second-order phase transition 1 belonging to a new (chiral)
E-mail addresses: pelissetto@roma1.infn.it (A. Pelissetto), rossi@df.unipi.it (P. Rossi), vicari@df.unipi.it
(E. Vicari).
1 The experimental results are reviewed in, e.g., Refs. [1,3]. Essentially, experiments with hexagonal
perovskites find a clear second-order phase transition except for CsCuCl3 . The results for helimagnetic rare
earths are instead less clear. We also mention Ref. [4] where it is shown experimentally that chiral order and spin
order occur simultaneously, thereby supporting Kawamuras [5,6] conjecture that chiral transitions are different
from the standard O(n) transitions.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 2 2 3 - 1
606
A. Pelissetto et al. / Nuclear Physics B 607 [FS] (2001) 605634
universality class, theoretically the issue is still debated. On one side, field-theoretical studies based in approximate solutions of the RG equations (ERG) do not find any stable fixed
point and favor a first-order phase transition [79]. On the other hand, perturbative field
theory gives the opposite answer: a stable fixed point is identified with exponents in agreement with the experiments [10]. Monte Carlo simulations [1113] do not help clarifying
the issue. While simulations of the antiferromagnetic O(n) model on a stacked triangular
lattice find a second-order phase transition with exponents reasonably near to the experimental ones, modified spin systems which supposedly belong to the same universality
class apparently favor a first-order transition [13]. 2 Note that the existence of a new chiral
universality class does not exclude the possibility that some systems undergo a first-order
transition. Indeed, they may lie outside the attraction domain of the stable fixed point and
thus belong to runaway RG trajectories. In this case, a first-order transition is expected.
The field-theoretical studies have been focusing either on the so-called Landau
GinzburgWilson (LGW) Hamiltonian with O(n) O(2) symmetry or on the corresponding nonlinear sigma (NL ) model. In this paper we will study a generalization of these
theories, by considering general O(n) O(m) Hamiltonians and we will try to understand
the nature of the fixed points of the theory. In particular, we will relate the LGW and the
NL descriptions showing explicitly that the stable fixed points of the two models are exactly the same, as conjectured in Refs. [14,15,32], for values of m and n consistent with
the existence of a second-order phase transition. Moreover, we will clarify the nature of
the unstable fixed points. However, this analysis will only be valid in the large-n region,
where, by using the large-n expansion, we will be able to identify nonperturbatively all
fixed points of the different Hamiltonians.
Since the large-n expansion plays a major role in our discussion, our results will only
be valid for n > n(m,
d), i.e., in the region of large-n analyticity. Such a function is

conjectured to be identified with the line n+ (m, d) on which the LGW chiral and antichiral
fixed points merge. The function n+ (m, d) has been the object of extensive studies that
tried to understand whether, in the physical case d = 3 and m = 2, n(m,
d) was smaller or
larger than three. In this case, studies using various approaches gave n(2,
3) 4 (Refs. [8,
16,17]), 5 (Ref. [9]), 6 (Ref. [10]). Here, we will provide another determination, together
with generalizations for other values of m, that substantially confirms previous findings,
i.e., n(2,
3) 5. Since the results that we will present are essentially adiabatic moving from
the large-n and small 4 d region, they are not expected to provide the (essentially
nonperturbative) features of the models in the region below n(m,
d). Therefore, the fact that

n(2,
3) > 3 does not necessarily imply an inconsistency with the field-theoretical results
of Ref. [10], where a rather robust evidence for stable chiral fixed points was found for
O(n) O(2) models with n = 2, 3 in fixed dimension d = 3. Such fixed points are not
analytically connected with the large-n and small- criticalities discussed above.
2 Note that not all modified models show a first-order phase transitions. Some of them have a behavior that is
compatible with a second-order phase transition. However, the measured exponents do not satisfy the condition
0, which must be satisfied in unitary (reflection-positive) models as these are, so that, the measured exponents
can only be effective ones. This is interpreted as a signal of a first-order phase transition. However, there is also
the possibility that the results are strongly biased by corrections to scaling induced by the constraint.
607
We will also show that for m > 2 the identification n(m,
d) = n+ (m, d) may not

be correct for d near two dimensions. Indeed, in this case a new critical line appears
which corresponds to the merging of the chiral fixed point with the NL antichiral fixed
point.
In order to obtain quantitative predictions for all m and n, we have extended the
4 d expansion of the LGW theory to order 3 and the d 2 expansion of
the NL model to order 2 for all n and m. Also, we present O(1/n2 ) results for the
LGW theory.
The paper is organized as follows. In Section 2 we define the general class of models
with O(n) O(m) symmetry that will be considered in the paper and find a general
representation that is the starting point of the large-n expansion. In Section 3 we compute
the O( 3 ) contributions to the critical exponents and 1 within the = 4 d expansion
of the LGW Hamiltonian. In Section 4 we analyze in detail the 1/n expansion of the
LGW Hamiltonian with O(n) O(m) symmetry to O(1/n2), thereby extending the results
of Ref. [14]. Interestingly enough, we can determine the large-n expansion of the exponents
at all fixed points and show explicitly their different physical nature: at the stable fixed
point both tensor and scalar excitations propagate, while at each unstable fixed point
one of the degrees of freedom is suppressed. At the Heisenberg fixed point there are
only scalar excitations, while at the antichiral one, there are only tensor excitations. In
Section 5 we discuss the 1/n expansion of a more general theory in which the coupling
to a (gauge) vector field is included, extending the results of Ref. [18]. In Section 6
we extend to arbitrary values of m and to O( 2 ) the -expansion of NL models,
evaluating the unstable fixed point and the coalescence value of n under which the two
fixed points actually disappear. We also identify the gauge criticality of the models.
In Section 7 we draw some general conclusions and present a new determination of the
function n(m,
d).
2. Models
We will consider a non-Abelian gauge model coupled to a scalar field with gauge
symmetry O(m) and global symmetry O(n). In particular, we consider a set of m
n-dimensional vectors = {a }, = 1, . . . , m, a = 1, . . . , n, a vector field A

antisymmetric in and , and the Hamiltonian density
H=
2
2 1 2
1
1
r
u
+ g0 A
+
2
0
0

2
2
4!
1 2
1
t0
v0
( )2 2 2 + F
+
A A ,
4!
4
2
(2.1)
where F is the non-Abelian field strength associated with the fields A . This
Hamiltonian is gauge-invariant (with local O(m) invariance) for t0 = 0, and in this case
608
it has already been studied 3 in Ref. [18]. For t0 = 0 and A = 0 we obtain a generic
LGW Hamiltonian density with global O(m) O(n) invariance:
2
1 2
1
1
( )2 + r0
+ u0
2
H=
2
2
4!

1
( )2 2 2 .
+ v0
(2.2)
4!
Assuming n > m, stability requires u0 > 0 and w0 u0 + (1 N )v0 /N > 0, where N =

min(m, n).
Other particular cases of the Hamiltonian (2.1) are interesting. If we set u0 = v0 and
r0 = v0 1 /6 and take the limit v0 + keeping 1 fixed, we obtain an O(n) O(m)
-model coupled to an O(m) vector field. The Hamiltonian density is given by
2 1 2
t0
1
+ g0 A
A A ,
H=
(2.3)
+ F +
2
4
2
where the fields satisfy the constraint

= 1 .
(2.4)
This limit is well defined only if n m, otherwise the constraint (2.4) cannot be satisfied.
In the absence of the kinetic term for the vector field, the Hamiltonian density depends
quadratically on the vector field that can then be eliminated by integration. We obtain a
new Hamiltonian of the form

2
2
g02
1

.
H=
(2.5)
2
2
8(g0 1 + t0 )
This is the -model studied in Refs. [14,15,32]. In order to recover the notations 4 of
Ref. [15], we set = 1 e and 2 2t0 1 /(g02 1 + t0 ). Then

m
m

1
= 1 1
2 1
H
(2.6)
e e +
(e e )2 ,
2
2
=1
>
where the fields e are m n-component vectors (or equivalently mn matrices) with n m
subject to the nonlinear constraint
e e = .
(2.7)
In order to study the large-n behavior of these models we rewrite the general Hamiltonian (2.1) in such a way that the dependence on the field is quadratic. This is obtained
3 Hikamis couplings [18], labelled by the subscript H , are related to ours by the correspondence: =
H
(u0 v0 )/3, vH = v0 /6, H H + 2vH = u0 /3.
4 Notice that our are consistent with the couplings employed in Section 4 of Ref. [15], but they are twice
i
as big as the couplings defined in Appendix B of Ref. [15], due to a slight inconsistency in the notation adopted
by these authors. The couplings i are related to those of Ref. [14] by 1 1/T + 1/T , 2 2/T . For easy
comparison, the reader should keep in mind that, due to the constraints, e e = e e .
609
by introducing two auxiliary fields: a scalar field S and a symmetric and traceless tensor
field T , i.e., such that T = T , T = 0. By means of these auxiliary fields we can
rewrite the Hamiltonian (2.1) as
H = Heff
3v0 2 3w0 2 t0 2 1 2
T
S + A + F ,
2
2
2
4
(2.8)
where
Heff =
1
X ,
2
(2.9)
and
X = + r0 2g0 A
+ w0 S + g02 A
+ v0 T
A .
(2.10)
Note that the effective action for the fields is the most general one which is O(m)
covariant. Therefore, the analysis of this class of models provides the critical behavior
of the most general O(m) O(n) theory.
3. -expansion for the LandauGinzburgWilson model

In this section we study the LGW Hamiltonian (2.2) and report our results for the critical
exponents and the -function to order 3 , thereby extending the results of Ref. [19] to three
loops. We consider the massless theory and renormalize it using the MS scheme. We set

1/2
= Z (u, v)
(3.1)
R ,
u0 = Zu (u, v)Nd1 ,
(3.2)
v0 = Zv (u, v)Nd1 ,
(3.3)
where the renormalization constants are normalized so that Z (u, v) 1, Zu (u, v) u,

and Zv (u, v) v at tree level. Here Nd is a d-dependent constant given by Nd1 =
2d1 d/2 (d/2). Moreover, we introduce a mass renormalization constant Zt (u, v) by
requiring Zt (1,2) to be finite when expressed in terms of u and v. Here (1,2) is the twopoint function with an insertion of 2 . Once the renormalization constants are determined,
we compute the functions from

u
v
u (u, v) =
(3.4)
,
v (u, v) =
,
u0 ,v0
u0 ,v0
and the critical exponents from

log Z
=
,
log u0 ,v0

log Zt
t =
,
log
u0 ,v0
= (2 t )1 .
(3.5)
(3.6)
(3.7)
610
For the -functions we obtain: 5

v
3mn + 14 3
mn + 8 2 (m 1)(n 1)
u
v u
u
u = u +
6
3
2
12

5
11 2 13
u uv + v 2
+ (m 1)(n 1)v
18
24
36

u4
33m2n2 + 922mn + 2960 + (3)(480mn + 2112)
1728
v
(m 1)(n 1)
+
3456

4 79mn + 1318 + 768(3) u3

+ 555mn 460(m + n) + 6836 + 4032(3) u2 v

2 213mn 358(m + n) + 1933 + 960(3) uv 2

+ 121mn 309(m + n) + 817 + 216(3) v 3 ,
(3.8)
m + n 8 2 5mn + 82 2
v
u v
6
36
5mn 11(m + n) + 53 2 13mn 35(m + n) + 99 3
uv
v
+
18
72

v4
52m2n2 57mn(m + n) 2206mn 111 m2 + n2

+
3456
+ 4291(m + n) 8084

+ 1416mn + 3216(m + n) 7392 (3)
v = v + 2uv +

v3 u
39m2 n2 + 35mn(m + n) + 1302mn + 36 m2 + n2

864
2401(m + n) + 5725

+ 768mn 1824(m + n) + 4896 (3)
u2 v 2
78m2n2 35mn(m + n) 2114mn + 3182(m + n) 12520

1728

+ 1152mn + 2304(m + n) 10368 (3)

u3 v
13m2 n2 + 368mn + 3284 + (192mn + 2688) (3) .
864
(3.9)
5 We mention that for the particular case n = m = 2 one may derive the corresponding four-loop -expansion
from the series reported in Ref. [20] for the so-called tetragonal model. Indeed an exact mapping [16] exists
bringing from the LGW Hamiltonian (2.2) with m = n = 2 to the tetragonal model considered in Ref. [20]. Note
that our renormalized couplings differ from those defined for m = 2 in Ref. [6]. Kawamuras couplings uK , vK
are related to ours by u = 12Nd uK , v = 6Nd vK .
611
For the critical exponents we obtain:

u
(mn + 2)(mn + 8) 3
v
mn + 2 2
u + (m 1)(n 1)v
u
72
48 36
1728
(m 1)(n 1) 2
v v (2mn 5m 5n + 26)
+
3456

6uv(mn m n + 10) + 6u2 (mn + 8) ,
(3.10)
1
mn + 2
(m 1)(n 1)
5(mn + 2) 2
=2
u+
v+
u
6
6
72

u
v
(mn + 2)(5mn + 37) 3
u
+ 5(m 1)(n 1)v
48 36
288
(m 1)(n 1) 2
+
v 3v (7mn 16m 16n + 79)
1152

uv(61mn 58m 58n + 550) + 12u2(5mn + 37) .
(3.11)
As discussed at length in Refs. [1,6,19], the critical behavior of these systems depends on
the values of n and m. In general, the -functions admit four solutions: the Gaussian fixed
point (u = v = 0), the O(mn) Heisenberg fixed point (v = 0) and two new fixed points
with nontrivial values of u and v , the chiral and the antichiral fixed points. These two
additional fixed points do not exist for all n and m, but, at fixed m, only for n n+ (m) and
n n (m). The functions n (m) will be computed below. The critical behavior depends
on the stability of the fixed points. At fixed m, for the physically relevant case m 1, the
-expansion predicts four regimes:
=
(1) For n > n+ (m), there are four fixed points, and the chiral one is the stable one.
(2) For n (m) < n < n+ (m), only the Gaussian and the Heisenberg O(n m)symmetric fixed points are present, and none of them is stable.
(3) For nH (m) < n < n (m), there are again four fixed points, and the chiral one is
the stable one. For small , the chiral fixed point has v < 0 for m < 7 and v > 0
for m > 7.
(4) For n < nH (m), there are again four fixed points, and the Heisenberg O(m n)symmetric one is the stable one.
The antichiral fixed point is Gaussian for m 1 and m = 2 (or, equivalently for n 1
and n = 2). Indeed, for m 1, u 0 for the antichiral fixed point, so that, from
Eqs. (3.10) and (3.11), we obtain = 0 and = 1/2 at order 3 . For m = 2 we obtain
u = 3v /2 and again = 0 and = 1/2 at order 3 . We conjecture that this holds to
all orders in , and in Section 4 we will provide a large-n interpretation of these results.
The general behavior for n and m is better understood from Fig. 1. In particular, the two
functions n (m) are nothing but the two different branches of the curve that separates the
region in which no fixed point is stable from the region in which the chiral fixed point is
the stable one. Note that the boundaries of the different regions are symmetric under the
exchange (n, m). Because of this symmetry it is more natural to consider the behavior in
the variables
= m + n,
= m n.
(3.12)
612
Fig. 1. The fixed-point structure in the (m, n) plane for d = 4. The solid line represents the
curves n (m). The dashed line shows nH (m) and the symmetric curve obtained interchanging n
and m.
At fixed there are then three regions:

(1) For > + only the Gaussian and the Heisenberg O(m n)-symmetric fixed
points are present and none is stable.
(2) For H < + there are four fixed points and the chiral one is stable.
(3) For H , there are four fixed points and the O(m n)-symmetric one is the
stable one.
Using the above results, we can compute the expansion of n (m) and nH (m). For
we expand
3
2
n (m) = n
(3.13)
0 + n1 + n2 + O .
n (m)
Then, by requiring

u u , v ; n = 0,
and

v u , v ; n = 0,

(u , v )
u , v ; n = 0,
det
(u, v)
(3.14)
(3.15)
we obtain
n
0 = 5m + 2 2s,

1
25m2 + 22m 32 ,
n
1 = 5m 2
2s
1
1
1 R3
1 sR4
R
R2
1
n
2 = 24 Q 32 sQ + 8 Q (3) 48 Q (3).
1
1
2
2
Here
(3.16)
(3.17)
(3.18)
s=
613
6(m 1)(m + 2),
R1 = 33024 + 18880m + 45444m2 + 9288m3 1883m4 417m5

+ 21m6 + 4m7 ,
R2 = 253952 160256m + 176192m2 + 139240m3 + 7756m4
5854m5 389m6 + 58m7 + 5m8 ,
R3 = 1632 + 1184m 1376m2 426m3 + 31m4 + 8m5 ,
R4 = 6176 2960m 1230m2 + 73m3 + 20m4,
Q1 = (m + 8)2 (m + 2)(m 1)(m 7)2 ,
Q2 = (m + 8)(m + 2)(m 1)(m 7).
(3.19)
For m = 2 this expression is in agreement with that given in Ref. [17]. The expression for
+
n
2 are singular for m = 7. However, this is not the case for n2 , and indeed, by taking the
limit we obtain
23871617 5487
n+
(3.20)
2 = 9331200 + 320 (3).
For nH (m) we have

1
5
6(3) 1 2 + O 3 ,
nH (m) =
(3.21)
4 2 +
m
12
which is a trivial generalization of the result of Ref. [17]. The calculation of the functions
+ () and H () follows the same lines. In particular,

2
1
5 24 2s + O 2 ,
+ () = 1 + s +
(3.22)
2
8s

where s = 6(2 + 18).
From our calculation of the RG functions we can derive the fixed points of the theory.
We expand

v = v1 + v2 2 + v3 3 + O 4 .
u = u1 + u2 2 + u3 3 + O 4 ,
(3.23)
Following Ref. [19] we define
1
Bmn
(mn + 8)(m + n 8)2 + 24(m 1)(n 1)(m + n 2),
Dmn mn(m + n) 10(m + n) + 4mn 4,

Rmn (m + n 8)2 12(m 1)(n 1).
(3.24)
We then easily find:

1 1
1/2
(m + n 8) Bmn Dmn 6Rmn ,
2 2

1/2
v1 = 6Bmn Dmn 6Rmn ,
u
1 =
(3.25)
where we indicate by (+) the stable chiral fixed point and by () the unstable antichiral
one. In order to compute u

2 and v2 , it is convenient to define two additional auxiliary
functions:
614

(3mn + 14) 3
5
11 2 13
u1 + (m 1)(n 1)v1
u1 u1 v1 + v12 ,
12
18
24
36
(5mn + 82) 2
5mn 11(m + n) + 53
u1 v1 +
u1 v12
S2
36
18
13mn 35(m + n) + 99 3
v1 .
72
S1
Then, the O( 2 ) coefficients of the fixed points are given by

m+n8
v1 S1 (m1)(n1)
(u
1 2u
1
1 v1 )S2
3
3
u2 = 6
,
1/2
v1 Rmn

(m1)(n1)
v1 S2
2v1 S1 + 1 mn+8
3 u1 +
3
v2 = 6
.
1/2
v1 Rmn
(3.26)
(3.27)
The expressions for u

3 , v3 are particularly cumbersome and they will not be reported
here.
Once the fixed points are determined, the critical exponents are computed by expanding
in power of the exponent series computed at the fixed point. Such a computation gives us
the exponents only for

n > 5m + 2 + 2 6(m + 2)(m 1),
(3.28)
or

n < 5m + 2 2 6(m + 2)(m 1).
(3.29)
Indeed, if these bounds are not satisfied the fixed points are complex and therefore also the
exponent series. In order to obtain series for the exponents in all the relevant domain we can
perform the following trick. For n > n+ , which is the case of physical interest, we set n =
n+ (m, ) + 5n and reexpand all series in powers of keeping 5n fixed. In particular, for
5n = 0 we obtain the critical exponents for n = n+ . In such a case, for m = 2 we obtain:
1 2
5 3
+
,
48
288

1
1
1
6
2
=2 +
2
50
50

37 6
37
6
397
3
+
(3) +
(3) .
+
15000 15000 1000
250
=
(3.30)
4. The LandauGinzburgWilson theory in the large-n limit

In this section, we study the large n-behavior of the LGW theory (2.2) at fixed m. The
starting point is the general Hamiltonian (2.8) with A = 0. In the high-temperature phase
the symmetry is unbroken and thus the relevant saddle point is given by
S = ,
T = 0.
(4.1)
Correspondingly, we obtain the gap equation:

1
dd p
6
,
=
(2)d p2 + M 2 nm
where M 2 = r0 + w0 . For the -model (2.3) we obtain analogously 6

1
dd p
1
= .
d
2
2
(2) p + M
n
615
(4.2)
(4.3)
From the gap equation we can obtain the scaling of the mass and thus the exponent . For
w0 = 0, proceeding as in the case of the ordinary O(n) model, we obtain for 2 < d < 4
=
1
.
d 2
(4.4)
However, for w0 = 0, we obtain simply M 2 = r0 , indicating

1
= ,
(4.5)
2
for all values of the dimension d.
Within the large-n limit we can recover the critical behavior of the theory at all fixed
points. For generic v0 and u0 , satisfying w > 0, v0 > 0 we obtain the critical chiral
behavior. The standard Heisenberg behavior is obtained by setting v0 = 0, while the
antichiral critical behavior is obtained at the stability boundary, i.e., by setting w0 = 0. It
is easy to see the different types of excitations that appear in these cases: at the chiral fixed
point both the scalar and the tensor degrees of freedom propagate, while at the Heisenberg
and antichiral fixed points one observes only the scalar and the tensor degrees of freedom
respectively. Note that, as a consequence of Eqs. (4.4) and (4.5), the Heisenberg and the
chiral point have the same exponents for n = , and that they differ from those of the
antichiral point which shows mean-field behavior in all dimensions.
In order to perform the calculation we heavily relied upon the results obtained by
Vasilev et al. [21,22], who studied the models corresponding to m = 1 with a method
which lends itself to a reasonably simple extension, appropriate to the case we are
investigating. In order to make our presentation self-contained, we must briefly review
the essentials of the method.
One first considers the second Legendre transform with respect to the field and the
,
a,b
(p), DS (p), and DT
(p) the dressed
two-point function [23]. We indicate by D
a
propagators of the field and of the auxiliary fields S and T . Here , , , and go
from 1 to m, while a and b go from 1 to n. It is useful to factorize the group dependence
and to introduce scalar propagators
(p),
(p) = ab D

1
2
,

DT
+
(p) =
DT (p).
2
m
a,b
(4.6)
(4.7)
6 This result can be obtained by using Eq. (4.2) and by taking the limit considered before Eq. (2.3). Notice that,
in order to keep M 2 finite in the limit, must converge to m1 as v0 .
616
(1)
(2)
(3)
(4)
Fig. 2. The graphs appearing in the second Legendre transform. The continuous line represents
the dressed propagator of the field, while the dashed line indicates the dressed propagator of an
auxiliary field.
Also we can reabsorb the coupling v0 and w0 in the fields. Using the same notations of
Ref. [21], we have for the second Legendre transform

1
1
n
1
Tr log D + Tr log DS + Tr log DT + 1 (S) + 1 (T )
2
2
2
2
1
1
1
+ nm2 (S) + n(m 1)(m + 2)2(T ) + nm3 (SS)
4
8
8
1
n
(m 1)(m2 4)3 (T T )
+ n(m 1)(m + 2)3 (T S) +
8
32m

1
1
+ n2 m2 4 (SSS) + n2 (m 1)(m + 2) 4 (SST ) + 4 (ST T )
12
8
n2
+
(4.8)
(m 1)(m2 4)(m + 4) 4 (T T T ) + .
96m
In this equation 1 , . . . , 4 , are the graphs reported in Fig. 2 and the letters in parentheses
indicate which auxiliary fields are propagating in the graph. In these graphs one should
T while each vertex is trivially one. In the
and D
use the scalar dressed propagators D
equation we have of course reported only those graphs that are relevant for the computation
of the critical indices at order 1/n2 . The group-theoretical factors in Eq. (4.8) have been
obtained by using Eqs. (4.6) and (4.7) and keeping into account that the vertex T has
the form:

1
2
a b T ab + .
(4.9)
2
m
=
Eq. (4.8) is completely general and can be used in the computation of the critical indices
for all fixed points: while we should keep into account all terms for the chiral fixed
point, we should set T = 0 and S = 0 for the Heisenberg and the antichiral fixed points,
respectively.
From Eq. (4.8) we can derive the skeleton Dyson equations for the dressed propagators.
It is enough to compute the variation of with respect to the dressed propagators. We
obtain for the field the equation
1 + u + g1 (S) + 1 (m 1)(m + 2)g1 (T ) + g2 (SS)
D
2m

1
1
+ (m 1)(m + 2) g2 (ST ) +
(m 1) m2 4 g2 (T T ) + mng3 (SS)
m
4m2
(1)
(2)
(3)
(4)
(5)
(6)
617
Fig. 3. The graphs appearing in the Dyson equations. The continuous line represents the dressed
propagator of the field, while the dashed line indicates the dressed propagator of an auxiliary field.

n
(m 1)(m + 2) g3 (ST S) + 2g3 (T SS) + g3 (T ST ) + 2g3 (T T S)
2m

n
+
(4.10)
(m 1) m2 4 (m + 4) g3 (T T T ) + = 0,
2
8m
while for the auxiliary fields we have
nm
n
g4 + nmg5 (S) + (m + 2)(m 1)g5 (T )
DS1 + cS +
2
2
n2 m2
n2
+
(4.11)
g6 (SS) + (m + 2)(m 1)g6 (T T ) + = 0,
2
2
1 + cT + n g4 + n g5 (S) + n (m 2)g5 (T )
D
T
2
2
2m
2
n
(m + 2)(m 4)g6 (T T ) + = 0.
+ n2 g6 (ST ) +
(4.12)
8m
Here cS and cT are two constants, u a momentum-independent contribution due to the
tadpoles, = p2 + M 2 , and gi are the graphs reported in Fig. 3. As before, in parentheses
we report which auxiliary fields are propagating. Each line is associated to a scalar dressed
propagator, while the vertices are one.
The critical exponents are determined following closely the method of Ref. [21]. As in
Ref. [21] we introduce two auxiliary functions:

a x d2
d2 x
,
p(x) d
,
a(x)
(4.13)
(x)
a(x)

a(x y) a x + y d2
q(x, y)
(4.14)
.

a(x)a x d2
+
Note the trivial symmetry of the function q(x, y) which will play a role below:

q(x, y) = q x, d2 y .
(4.15)
The calculation of the 1/n correction starts from assuming for x 0 the following
behavior of the dressed propagators:

AX
DX (x) = 2 1 + BX x 2 + ,
(4.16)
x X
618
where X is any of the fields. Here X is related to the dimension of the field. For X =
we have = d/2 1 + /2. The correction term we report is the analytic one in the
temperature and therefore = 1/(2). From Eq. (4.16) we obtain for the inverse functions:
1
(x) =
DX

p(X )
1 BX q(X , )x 2 + .
AX x 2d2X
(4.17)
Plugging these expressions in the skeleton equations and equating the corresponding terms
we obtain six equations for the amplitudes. Such equations have nontrivial solutions only
if
S = T = 2 ,
(4.18)
and the following consistency equations are satisfied:

M
p( ) = 2 p(),
n

q( , ) + 1 q(, ) = 2,
where M is a group-theoretical factor that depends on the fixed point:
M + (m + 1)
(chiral f.p.),
1
M = M
(m 1)(m + 2) (antichiral f.p.),
2m
MH 1
(Heisenberg f.p.).
m
(4.19)
(4.20)
(4.21)
Note that M = 0 for m = 1, 2, a result which follows from the fact that only a
symmetric traceless tensor propagates. Eq. (4.19) allows the determination of the first
large-n coefficient appearing in the expansion of the exponent ,
1 2 3
+ 2 + 3 + .
=
(4.22)
n
n
n
For 1 we obtain
1 = M11 ,
(4.23)
where
4 (d 2)
11 =

,

2 d2 d2 1 d2 2 d2 + 1
(4.24)
and the dependence on the fixed point is encoded in the factor M. The quantity 11 is the
well-known result for the m = 1 model; among its most important properties we wish to
mention that 11 0 both in the d 4 and in the d 2 limit.
Eq. (4.20) should allow the determination of the exponent . However, there is a subtle
point that has been overlooked in the previous analyses. It is convenient for our discussion
to introduce the auxiliary function

d
+ 1, + 1 q(2 , ),
r(, ) q
(4.25)
2 2
619
which corresponds to the left-hand side of Eq. (4.20). Because of Eq. (4.15), the function
r(, ) has the symmetry

r(, ) r , d2 .
(4.26)
As a consequence, once is fixed by solving Eq. (4.19), we still have the possibility
of finding two different solutions for 1/2. It is convenient to parametrize the two
solutions by

d
d

(4.27)
1+ .
4
4
Using Eq. (4.20) and the fact that is of order 1/n, one can show that also is of order
1/n and therefore has an expansion of the form
1 2
=
(4.28)
+ 2 + .
n
n
The coefficient 1 is computed from Eq. (4.20) using the fact that for n large

q d2 1 + 12 , d2 1 + = q d2 1 + 12 , 1

4d
21
=
(4.29)
+ O n1 .
1
d
1
We then obtain
1 = M11 ,
(4.30)
where
(d 1)(d 2)
(4.31)
11 .
4d
As we observed the consistency equations are satisfied for two independent choices of .
In order to associate to correct one to each fixed point we use the large-n estimates of the
exponent given in Eqs. (4.4) and (4.5). Then we have
d
1
1 + M + 11 chiral f.p.,
n
2
1
= 1 M 11
(4.32)
antichiral f.p.,
d
1
1 + M H 11 Heisenberg f.p.
2
n
11 =
The corrections of order 1/n2 can be obtained by generalizing the arguments of Refs. [21,
22]. One must only pay attention to insert proper group-theoretical factors in front of
the corresponding m = 1 contributions: they can be obtained from Eqs. (4.10), (4.11),
and (4.12).
We introduce the following definitions:

d 2 + 2d 4
HF
2
21 = (11 ) +
(4.33)
,
2d(d 2)

32 + 8d 30d 2 + 7d 3
d 1
HF
2
+
= (11 )2
21
(d
2)
4
+
2d
d
, (4.34)
(d 4)2
d(d 4)
620
where
d
(d 2) + 2
2
Also:

d
2 (2).
2
(4.35)

d(6 d)
d
+
(4.36)
,
4d
2(4 d)2

d(d 2)
(b)
2 d(d 3)
21 = (11 )
+
,
4d
4d

d(d 2)
3
(a)
21
=
(11 )2 2(d 1) + d(d 3)R1 + 2d 8
2(4 d)2
2

6
4
12
+
+
,
4 d d 2 (d 2)2
d(d 2)
(b)
=
(11 )2
21
2(4 d)2

3
4(d 3)
2d(d 3)2
+ d(3d 8)R1 +
6R1 R2 R32 + d 2
(d 2)
2
(4 d)

12(4 d)
16
+ d + 20
(4.37)
,
(d 2)2
4d
(a)
21
= (11 )2
where

d
1 (1),
2

d
d
1 + (1),
R2 (d 3) 2

2
2

d
d
1 (1).
R3 (d 3) + 2
2
2
R1
(4.38)
Recalling the above considerations about the choice of in conjunction with the choice
of the critical point, we can now write down our final results for the 1/n expansion of the
critical exponents in the O(n) O(m) models, both for the chiral (stable) critical point
and for the antichiral (unstable) one:

m+1
1 (m + 1)2 H F m + 3 (a) m2 + 3m + 4 (b)
11 + 2
21 +
21 +
21
+ =
2n
4
4
8
n

1
+O 3 ,
(4.39)
n
(m 1)(m + 2)
11
=
2mn

1 (m 1)2 (m + 2)2 H F (m 1)(m2 4) (a)
21 +
21
+ 2
n
4m2
4m2

(m 1)(m2 4)(m + 4) (b)
1
+
(4.40)
+
O
,
21
2
8m
n3
+ =
621
2
m+1
1
11
2
d 2 (d 2) 2n

1 (m + 1)2 H F m + 3 (a) m2 + 3m + 4 (b)
2
21 +
21 +
21
(d 2)2 n2
4
4
8

2 (m + 1)2 2
1
11 + O 3 ,
d 2
4
n
1 1 (m 1)(m + 2)
+
11
2 2
2mn

1 1 (m 1)2 (m + 2)2 H F (m 1)(m2 4) (a)
+
21 +
21
2 n2
4m2
4m2
+

1
+O 3 .
n
(m 1)(m2 4)(m + 4) (b) (m 1)2 (m + 2)2 2

21 +
11
8m2
4m2
(4.41)
(4.42)
The expressions for the stable fixed point at order 1/n coincide with those of Ref. [14].
Note that = 0, = 1/2 for m = 2, 1, in agreement with our -expansion
results.
It is possible to expand the above large-n results in powers of = 4 d. The resulting
expressions can be compared with the the -expansion results for the LGW Hamiltonian
presented in the previous section. We find full agreement both for the stable and the
unstable fixed point for all m, thus confirming our identification of the large-n fixed points
with the perturbative ones.
For d = 3 the large-n expansions simplify to:

4(m + 1) 16(m2 7m 26)
1
+
+O 3 ,
=
2
2
4
3n
27n
n
+

4(m 1)(m + 2) 16(m 1)(m + 2)(m2 8m 2)
1
+
+O 3 ,
=
3mn 2
27m2 n2 4
n

16(m + 1)
1 4(m2 + 3m + 4) 64(5m2 + 19m + 32)
+
2
=1
3n 2
n
2
27 4

1
+O 3 ,
n
(4.43)
(4.44)
(4.45)
1 4(m + 2)(m 1)
+
2
3mn 2

(m 1)(m + 2)
16 13m2 + 4m + 28 + 27(m + 4)(m 2) 2
2
2
4
27m n

1
+O 3 .
n
(4.46)
622
5. The 1/n-expansion in the presence of a vector field

It is quite instructive to extend the discussion of the previous paragraph to the more
general case in which gauge-invariant vector degrees of freedom are allowed. This
corresponds to studying the general Lagrangian (2.1) with t0 = 0.
In the large-n limit one starts from Eqs. (2.8) and (2.9) with t0 = 0. As discussed by
Hikami [18], the gauge kinetic term is irrelevant for d < 4, as well as the T 2 and S 2 terms
in Eq. (2.8). Thus, the large-n limit can be studied by keeping only into account Heff . The
discussion of the fixed points is identical to that presented in the previous Section. If g0 = 0
a new set of fixed points appear: for generic v0 > 0, w0 > 0 we have the chiral-gauge fixed
point in which all excitations (S, T , A ) propagate; for w0 = 0 we have the antichiral
gauge fixed point (T , A ), for v0 = 0 the Heisenberg-gauge fixed point (S, A ), and
for u0 = v0 = w0 = 0 the pure gauge fixed point (A ).

Here, we want to compute the critical behavior for n , keeping only the leading
correction. Since the model is gauge-invariant, the large-n propagator of the field A
is not uniquely defined. Indeed, by integration over the field we obtain a coupling
1 ,
, where in momentum space
2 A A M
,
(p) =
M

2
1
p ,
p p p2 M
2
(5.1)
which is not invertible. A propagator for the field A is obtained by adding a gaugefixing term, that introduces a longitudinal term, makes the matrix invertible, but does not
contribute to physical quantities [24].
The calculation is completely analogous to that performed in the previous section. For
the second Legendre transform, we obtain to order O(1/n)
n
nm(m 1)
1
Tr log DA + 1 (A) +
2 (A) + ,
(5.2)
2
2
8
where 1 and 2 correspond to the graphs reported in Fig. 2, all group-theoretical factors
have been explicitly singled out, and (A = 0) is the expression reported in Eq. (4.8).
Generalizing the results of Refs. [18,24] we obtain then

d2 1
(m 1) ,
1 = 11 M +
(5.3)
2
= (A = 0) +
where M is a group-theoretical factor defined in Eq. (4.21) (for the pure-gauge fixed point
M = 0) and 1 is the 1/n contribution to the exponent defined in Eq. (4.27). Note that
the result (5.3) does not depend on the gauge fixing used to define the propagator of the
field A. In principle, one could also compute a gauge-fixing dependent exponent but its
significance is not so clear, since, because of the gauge invariance, the field does not
have a well-defined anomalous dimension.
As a check we can compare our results with those obtained in perturbation theory for
the gauge Hamiltonian (2.1) with t0 = 0. Hikami [18] determined the following one-loop
functions in the MS scheme:

mn + 8 2 (m 1)(n 1) 1 2
u +
v uv
6
6
2

3
9
(m 1)u + (m 1) 2 ,
2
8

9
m+n8 2
3
v + 2uv (m 1)v + (m 2) 2 ,
v = v +
6
2
4

11
n
(m 2) 2 ,
= +
12
3
623
u = u +
(5.4)
where = Nd g 2 . These expressions generalize the results presented in Section 3.

Choosing the = 0 solution of the fixed-point equations we obtain the four critical points
already discussed. However, if we choose the solution
1

11
n
(m 2)
+ O 2 ,
=
(5.5)
12
3
we find another set of four critical points, corresponding to the distinct roots of a quartic
algebraic equation. This equation cannot be solved in closed form, but it is easy to find its
roots in the form of a series in the powers of 1/n. The relevant terms in the expansion of
the roots are:
Chiral-gauge fixed point:
u
1 2 10m
= +
+ O(),
6 n
n2
Antichiral-gauge fixed point:
1 32 10m
v
= +
+ O().
6 n
n2
u m 1 (m 1)(16 + 88m 10m2)

=
+
+ O(),
6
mn
m2 n2
1 12 + 32m 10m2
v
= +
+ O().
6 n
mn2
Heisenberg-gauge fixed point:
u
1
27m3 117m2 + 90m 8
=
+
+ O(),
6 mn
m2 n2
v 27(m 2)
=
+ O().
6
n2
Pure-gauge fixed point:
(5.6)
(5.7)
(5.8)
u 27(m 1)
27(m 2)
v
(5.9)
=
=
+
O(),
+ O().
6
n2
6
n2
Thus, the gauge model has in general 8 fixed points and, at least for large n, the chiral-gauge
fixed point is the stable one. 7 Substituting these expressions into the relationship [18]
1 = 2

mn + 2 (m 1)(n 1) 3(m 1)
u +
v +
+ O 2 ,
6
6
4
(5.10)
7 This is not true for generic n and m. In order to obtain the general fixed-point structure, one should generalize
the analysis performed in Section 3.
624
we find a 1/n expanded form of the O() contribution to the critical exponent for each
of the four solutions:
2 + (48m 42) + O( 2 , n2 )

chiral-gauge f.p.,
6(m 1)(8m + 1)
2
+ O( 2 , n2 )
antichiral-gauge f.p.,
1
m
n
(5.11)
=

45m2 45m + 6
2
2
2 +
+ O( , n ) Heisenberg-gauge f.p.,
m
n
2 45(m 1) + O( 2 , n2 )
pure-gauge f.p.
n
It is then a matter of trivial algebra to verify that these expressions are in full agreement
with the -expansion of the four solutions discussed above in the context of the 1/n
expansion, which explains the names we have given to each fixed point. Again, we think
it is important to notice that the conformal bootstrap approach can naturally accommodate
for the expansion of all solutions, not only the stable ones.
6. The -expansion of O(n) O(m) nonlinear models

The LGW Hamiltonian is the natural tool for the study of the critical behavior of systems
near the upper critical dimension d = 4. If one is interested in the critical behavior near
the lower critical dimension, one can still use perturbation theory, applied however to the
nonlinear models (NL ). The degrees of freedom of the NL models should correspond
to the interacting Goldstone modes of the system, while the effect of the massive modes is
only taken into account in the form of constraints for the massless fields. In this context, it
is possible to perform an expansion in powers of d 2. In the present paper we extend
the results of Refs. [14,15,32,33] to a general O(n) O(m) symmetry group and to O( 2 ).
Comparing with our previous 1/n expansion results we will be able to identify the nature
of the fixed points of the NL . In particular, we will show explicitly that the stable fixed
point of the generic model can be identified with the stable fixed point of the LGW theory.
We consider the Hamiltonian (2.6). This Hamiltonian is geometric in nature, and its
variables are best understood as generalized coordinates spanning a manifold. The cases
we shall be interested in correspond to manifolds that are coset spaces. More specifically,
we must study the coset space (remember that n m)
O(n) O(m)
,
O(n m) O(m)
(6.1)
which is topologically equivalent to

O(n)
.
O(n m)
(6.2)
Associating fields I with the Goldstone modes that correspond to the broken generators

Lie O(n) Lie O(n m) ,
(6.3)
625
the Hamiltonian may be formulated in purely geometric terms, i.e.,

= 1 gI J () I J .
H
(6.4)
2
The couplings Ti 1/i are related to the independent entries of the tangent-space
metric I J .
A number of important RG properties of the NL models have been derived in the
general case by Friedan [25] and specialized to the models of interest in Refs. [15,32,33].
If RI J KL and RI J are, respectively, the Riemann and Ricci tensor for the metric gI J , the
RG functions of the model can be written to two-loop order as
gI J
1
P QR
+ .
= I J + RI J + RI P QR RJ
(6.5)
s
2
The number of algebraically independent -functions i coincides with the number of
independent couplings Ti . Therefore, we should consider two -functions associated with
T1 and T2 . The fixed points are determined from the equations

i T1 , T2 = 0,
(6.6)
I J s
that can be perturbatively solved in powers of .
The evaluation of the two-loop functions for arbitrary m and n requires no special
skills, but it takes some time and effort in view of the many computational steps involved.
Without belaboring on the intermediate steps, we report here our final results:
1 s
T1
s

m1
X T12 + A + BX + CX2 T13 + O Ti4 ,
= T1 + n 2
2
T2
2 s
s

m2 nm 2 2
+
X T2 + DX2 + EX3 + F X4 + G T23
= T2 +
2
2
4
+ O Ti ,
where X is shorthand for the ratio T1 /T2 and we have defined the coefficients:
3
A(n, m) 2m(n m) n + (m 1)(m 2),
8

3
3
B(n, m) (m 1) (n m) + (m 2) ,
2
8

3
m
C(n, m) (m 1) (n m) +
,
8
8
3
D(n, m) (n m)(m 2),
4
3
E(n, m) (n m)(m 2),
4
(6.7)
626
1
F (n, m) (n m)(2m 3),
8
1
G(n, m) (m 2)2 .
(6.8)
8
Our results were submitted to a number of basic consistency checks:
(i) when m = 1 there is no 2 coupling, and 1 (T1 ) reduces to the well-known function
for the vector -model defined on the coset space O(n)/O(n 1);
(ii) when m = 2 our expressions reduce to those of Ref. [15];
(iii) when m = n there is no 1 coupling, 8 and the model reduces to a standard O(n)
O(n) principal chiral model. One may verify that 2 (T2 ) is directly related to the known
function of these models [26].
One may also consider the gauge limit 2 = 0, which was studied by Hikami [18]: the
identification with Hikamis coupling is 1 1/t. One must however recognize that the
gauge
(t) is not obtainable from our
limit is singular, and as a consequence the function 1
expressions by setting X = 0 (with the notable exception of m = 2 models). If we assume
2 = 0 from the very beginning of our calculation, the result is

gauge
1
(6.9)
(t) = t + (n 2)t 2 + [2m(n m) n]t 3 + O t 4 ,
consistent with that reported in Ref. [18].
The -functions (6.7) are the starting point for the perturbative evaluation of the critical
points and exponents to O( 2 ). A consistent ansatz for the simultaneous solutions of the
equations i (T1 , T2 ) = 0 is the following:

T1 = t1 + t2 2 + O 3 ,
(6.10)
2
X = X0 + X1 + O .
(6.11)
It is straightforward to obtain the following algebraic equations for t1 and X0 :
1
m1
X0 = 0,
+ (n 2)
t1
2
m2 nm 2
1
+
X0 = 0.
X0 +
t1
2
2
They are trivially solved by

n 2 (n 2)2 (n 1)(m 2)
,
X0 =
n1
1
m1
X0 .
=n2
2
t1
(6.12)
(6.13)
(6.14)
Iterating the procedure we may also obtain

X1 =
(D A)(X0 )2 + (E B)(X0 )3 + (F C)(X0 )4 + G

,

X0 (n 2) (n 1)X0 n 2 m1
2 X0
(6.15)
8
m the vectors e are an orthogonal basis in R m and therefore satisfy the completeness relation
If n =
(v e )(w e ) = v w for all vectors v, w. Then, it is a simple matter to show that the second term in
Eq. (2.6) is one half of the first one.
t2
627

m 1 A + BX0 + C(X0 )2
1
=
.
X1
m1
2
2
n 2 2 X0
n 2 m1
2 X0
(6.16)
This analysis shows the existence of a couple of nontrivial fixed points of the RG equations.
However, it is evident from Eq. (6.14) that such a pair of solutions does not exist for all
m and n: for some values X0 is indeed complex. Repeating the analysis we performed in
Section 3 for the -expansion, we see that these two fixed points exist only for n >
n+ and
n < n , where

m + 2 m2 4 1 m + 2
m2 + 4 m m2 4
n =

2
2 m 2 (1 m2 4 )(m m2 4 )

+ O 2 .
(6.17)
Note that for small, we have m n + < m + 1 and n < m. Thus, since n m, all models
with integer n m + 1, have a a pair of nontrivial fixed points, at least for small. Beside
these two fixed points, there is also a fixed point for T1 = 0 and T2 = T2 that belongs to the
universality class of the O(m) O(n) principal chiral model. Such a fixed point always
exists perturbatively for m > 2 and in particular is the only present for n = m.
As one may easily notice, the expansion (6.17) is singular when m = 2. This is related
to the following peculiar feature of m = 2 models: for any value of n the (unstable) fixed
point corresponds to the solution X = 0, and as a consequence we observe its coalescence
with the gauge fixed point obtained by setting 2 = 0. This phenomenon does not happen
for m > 2. In this case, the gauge fixed point and the antichiral fixed point are distinct.
The exponent is easily computed. To two-loop order we have

= + (n m)T1 + (m 1)T2 + O 3 .
(6.18)
Substituting the expression of the fixed point, we obtain

t1
t2
t1 X1
= (n m) t1 + t2 2 + (m 1)
+
2 2 + O 3 . (6.19)
X0
X0
X0
We can expand at the stable critical point in powers of 1/n, obtaining

m + 1 3m2 + 7m + 6
1
+
=
+ O 3
2
2n
8n
n

2

m + 1 3(m + 1)
1
+
+ O 3 2 + O 3 .
2
2n
4n
n
(6.20)
If we compare such expression with the -expansion of + as obtained from the

large-n expansion of Section 4, Eq. (4.40), we find complete agreement, confirming the
identification of the two fixed points.
In NL models the evaluation of stability goes together with the evaluation of the critical
exponent , since both are related to the eigenvalues of the derivative matrix
i
(6.21)
T ,T .
Tj 1 2
More precisely, stability requires that the above matrix possesses only one positive
eigenvalue + = 1 . The presence of two positive eigenvalues signals the instability of
628
the fixed point. It is possible to evaluate the above-mentioned eigenvalues in the context of
the expansion, obtaining:

+ = 2 (n, m) 2 + O 3 ,
(6.22)
3
2
= 1 (n, m) + 2 (n, m) + O ,
(6.23)
where
a112 a221 + a111a222 a122a211 a121 a212
,
a111 + a221
2 (n, m) = 2 (n, m) a112 a222.
1 (n, m) = 1 a111 a221,
2 (n, m) =
(6.24)
Here we defined
m1
t1 X0 = a121X0 ,
2
a211 (n m)t1 X02 = a221X0 ,

m1
(t1 X1 + t2 X0 ) t12 A + 2BX0 + 3CX02 ,
a112
2
m1
t2 + t12 (B + 2CX0 ),
a122
2

a212 (n m) 2t1 X0 X1 + t2 X02 t12 2DX0 + 3EX02 + 4F X03 ,

a222 (n m)(t1 X1 + t2 X0 ) + t12 D + 2EX0 + 3F X02 G/X02 .
a111
(6.25)
The 1/n expansion of evaluated at the stable fixed point coincides with the expansion
of + obtained in Section 4. The result of the expansion is:

m + 1 1 (m + 1)2 1
1
+
+O 3 .
2 =
(6.26)
2 n
2
n2
n
Notice that the coalescence value n can be easily determined within the expansion by
imposing the condition

n , m = 0.
(6.27)
While the stable fixed point is identified with the chiral fixed point of the O(n) O(m)
LGW model, the unstable one is unrelated to those of the LGW model. In order to
understand its nature, it is again useful to consider the large-n limit. From Eq. (6.14) one
observes that X0 1/n as n . Therefore, the fixed point survives in the large-n
limit if we scale the coupling constants as T1 = O(1/n), T2 = O(1). Then, for large n the
functions decouple. Moreover, while 1 (T1 ) gets no contributions beyond one loop (as
usual in vector models), 2 (T2 ) turns into the function of an O(m) O(m) principal
chiral model [26]. Therefore, in this case the pattern of spontaneous symmetry breaking is
highly nontrivial, even in the strict n limit. This can be understood from Eqs. (2.8)
and (2.9). In the large-n limit the relevant Hamiltonian is
H = Heff +
t0 2
A .
2
(6.28)
629
Now, the field couples only to the gauge-invariant degrees of freedom, and thus at the
saddle point the field A is a pure gauge transformation, i.e.,

1

A
(6.29)
,
= O O
where O is an O(m) matrix. Thus, for n , the Hamiltonian can be rewritten as the
sum of two terms:

t0
1
+ M 2 + Tr O 1 O,
H=
(6.30)
2
2
where, as in Section 4, M 2 = r0 + w0 . Thus, the unstable fixed point is directly related to
the nontrivial fixed point of the principal O(m) O(m) chiral model.
Finally, we consider the gauge limit 1 = 1/t, 2 = 0. We find a nontrivial fixed point:
t =

n 2m(n m) 2
1
+
+ O 3 .
3
n2
(n 2)
(6.31)
Correspondingly we obtain:
1 (t ) = +

2m(n m) n 2
+ O 3 .
2
(n 2)
(6.32)
The 1/n expansion of the above result gives

2m 1 2
+ O 3 ,
(6.33)
n
and one may easily check that it agrees with the expansion of the results found in
Section 5 for the stable fixed point of the gauge model, see Eq. (5.3), with M = M + .
The behavior of NL models in the case m = n 1 is worth a special discussion [33].
In this case one may naturally define a new n-component field en such that en en = 1 and
e en = 0, and one can show that

n1
1
1
1

1 2 en en .
H = 2
(6.34)
e e +
4
2
4
1 = +
=1
When 1 = 2 , this is the Hamiltonian of an O(n) O(n) principal chiral model. The
stable fixed point is characterized by the property that X = 1 and one finds:
2
1
,
t2 =
.
(6.35)
n2
n2
Direct substitution shows that

1
1 = + 2 + O 3 ,
(6.36)
2

n
n1 2
=
(6.37)

+ O 3 .
n2
n2
As one may easily check, these exponents coincide with those obtained in the case m = n.
Therefore, the symmetry of the O(n) O(n 1) model is dynamically promoted to
O(n) O(n) at the stable fixed point, thus generalizing the results of Ref. [15] concerning
t1 =
630
the case O(3) O(2). This property is certainly true for sufficiently small d > 2, but at
this level of analysis it is impossible to establish the maximum dimension d for which a
stable critical point possessing the enlarged symmetry can be found.
7. Conclusions
At this stage of our analysis, we can draw quite general conclusions on the general
fixed-point structure of the models with O(n) O(m) for all dimensions 2 d 4.
By comparing the -expansion results near two dimensions, the -expansion results near
four dimensions and the large-n results, we have been able to identify the nature of all
(stable and unstable) fixed points of these models. In particular, the LGW stable fixed
point coincides with the stable one of the NL model. We thus quantitatively confirm one
of the conclusions of Refs. [15,32]: above n(2,
d) in the (m, d) plane a second-order phase
transition occurs, and with varying d, for n large enough, the critical exponents smoothly
interpolate between NL and GLW model values.
The unstable fixed points, which give rise to different types of tricritical behavior and
crossover phenomena, are instead unrelated and correspond to systems with completely
different types of excitations.
The correspondence we have found holds only for sufficiently large values of n, i.e., for
n n(m,
d), which is the region of analyticity of the large-n expansion. Since the 1/n
expansion commutes with the = 4 d expansion of the LGW Hamiltonian, one may
expect n(m,
d) to coincide with n+ (m, d) in a neighborhood of d = 4; n+ (m, d) might in

turn be evaluated within the 1/n expansion by solving the coalescence equation:
+ (m, n+ , d) = (m, n+ , d).
The estimate obtained from the lowest order approximation for is:

1 11 (d)
n+ (m, d) 2 m + 1
.
m 4d
(7.1)
(7.2)
This expression has been obtained by solving exactly the equation 1/ + (m, n+ , d) =
1/ (m, n+ , d), where 1/ + (m, n+ , d) and 1/ (m, n+ , d) are expanded to order 1/n.
This expression shows the correct qualitative behavior for all 2 d 4 and a rough
quantitative agreement. It is possible to improve the approximation by including the 1/n2
correction in Eq. (7.1). For m = 2 and d = 3, it predicts n+ (2, 3) 5.3, in substantial
agreement with the results obtained by using the ERG approach [8,9], the perturbative
expansion in fixed dimension [10], and, as we shall show below, the -expansion.
We want now to understand the behavior of n(m,
d) near two dimensions. Near two

dimensions, using the NL model results we know that the (LGW and NL ) stable fixed
d) =
n+ (m, d). Thus, for
point exists only for n >
n+ (m, d), so that in this case n(m,
generic values of d we conjecture
+
n (m, d) for dc (m) d 4,
n(m,
d) =

n+ (m, d) for 2 d dc (m),
631
Fig. 4. Sketch of the coalescence line as a function of the dimension d.
where dc (m) is a critical dimension that we cannot determine with our means. Of course,
this expression is valid for n m. The symmetry under exchange of n and m, implies the
existence of a similar boundary curve in the region n m, obtained by interchanging n
and m. A sketch of n(m,
d) is reported in Fig. 4.
Now, let us discuss the behavior of the LGW fixed points for d 2. Since the LGW
stable fixed point is equivalent to that of the NL model, and, for all n 2, m 2 except
n = 2, m = 2, the NL model is asymptotically free, we expect + = , a conclusion
that is confirmed by the large-n expression (4.41). On the other hand, for d = 2, Eq. (4.42)
predicts = 1/2 without 1/n and 1/n2 corrections. It is thus natural to conjecture that
= 1/2 for all n 2 and m 2, i.e., that the LGW antichiral fixed point is a Gaussian
fixed point. The case m = 2, n = 2 needs a special discussion. Using the fact that the
O(2) O(2) LGW model is equivalent to the so-called mn model 9 with m = n = 2
[16,27] one can show that in the v < 0 region a stable fixed point exists for all values
of d [1,2]. Finally, for d 2, using the -model results of the previous section, one finds
that it becomes Gaussian.
We want now to use the knowledge of n(m,
2) in order to obtain some informations

on n(m,
3). For this purpose we will make two hypotheses: first we will assume dc (m) < 3,
d) to be sufficiently smooth in
so that n(m,
3) = n+ (m, 3); second, we will assume n(m,

d at m fixed, so that we can use the interpolation method of Ref. [29]. Such a method has
provided very precise estimates of critical quantities (see, e.g., Refs. [29,30]).
9 The mn model with m = 2 describes n XY models coupled by an O(n)-symmetric interaction. Using
essentially nonperturbative arguments (see, e.g., Ref. [27]) related to the specific-heat exponent of the XY
universality class, one can argue that for d < dc with dc > 3 there is a stable fixed point belonging to the XY
universality class, while for d > dc there is a stable fixed point with the tetragonal symmetry. This fact has been
recently confirmed by high-order field-theoretical calculations in three dimensions [2,28].
632
Let us first consider the case m = 2. We start from [17]

14
6
n(2,
4 ) = 12 + 4 6 12 +
3

91
137
13 47
+
+
+
6+
6 (3) 2 + O 3
150 300
5
60
3
2
= 21.80 23.43 + 7.09 + O .
Following Ref. [29], we rewrite this equation in the following form

n(2,
4 ) = 2 + (2 ) 9.90 6.67 + 0.16 2 + O 3 .
(7.3)
(7.4)
Note that the new perturbative series is much better behaved than the original one,
the coefficients of the series decreasing rapidly. Setting = 1, we obtain an estimate
for n(2,
3):
n(2,
3) 5.3(2),
(7.5)
where the error indicates how the estimate varies from two loops to three loops. It should
not be taken seriously; it should only provide an order of magnitude for the precision of
the results. The estimate (7.5) is in good agreement with the determinations of Ref. [9,10]:
n(2,
3) 5 (Ref. [9]), 6 (Ref. [10]). We can try to estimate the exponents for n = n(2,
3),
by using Eq. (3.30). The coefficients decrease steadily with and thus we can simply set
= 1, obtaining
0.038,
0.63.
(7.6)
It is also interesting to compute the exponents for n = 6, m = 2, and d = 3, in order to

make a numerical comparison with the results of Ref. [31] who found = 0.700(11), =
1.383(36), and the ERG results of Ref. [9] who found 0.707, 1.377. If we use our
O(n2 ) expansions for the critical exponents, we obtain 1.22 and 0.63. We can
also use the -expansion, by using the method explained at the end of Section 3. In this
case, we must fix 5n = 6 n+ (2, 3). Conservatively, we have 0 5n 1. Then, from
the perturbative series we estimate 0.630.64, 1.241.26, which is rather close to
the large-n result, and somewhat lower than the numerical results of Refs. [9,31]. It is also
worth mentioning that for n = 6 the fixed-dimension field-theoretical approach does not
find fixed points that are sufficiently stable with respect to the order of the expansion up to
six loops [10]. We believe that these apparent discrepancies among the various approaches
deserve further investigation.
Our expressions may also be employed in order to establish an upper bound on the
critical dimensionality dc (n, m) for the existence of a stable fixed point analytically
connected with the critical point found in the 1/n expansion. This bound can be obtained
by forcing the condition n(m,
dc ) = n. In particular, one may determine the dimension dc

such that, for d < dc the O(3) O(2) has a nontrivial fixed point with symmetry
O(4) [15]. This corresponds to solving the equation n(2,
dc ) = 3. If we use Eq. (7.4),
we find dc 2.71. We may compare our result to those obtained in the ERG approach:
dc = 2.83 (Ref. [7]), 2.87 (Ref. [8]). They are in substantial agreement with our result,
633
when allowing for the systematic errors of both approaches. It should also be noticed that
our interpolation (7.5) is also in very good agreement with the ERG results of Refs. [7,8]
for all [34].
These analyses can be repeated for larger values of m. Since only n+ (m, 3) seems to be
rather precisely determined, we only report the results for this quantity. For m = 3 and 4
we have the constrained estimates
n(3,
3) 9.1(9),
(7.7)
n(4,
3) 12(1).
(7.8)
For large m we have

n(m,
4 ) = m 9.90 10.10 + 2.66 2 + O 3 , m1

= m + (2 )m 4.45 2.83 0.084 2 + O 3 , m1 ,
(7.9)
where, as already observed, the coefficients of the constrained series are smaller than
the original ones. Setting = 1, we obtain n(m,
3) 2.5m. Note that the large-m

approximation is already good at m = 4.
It is very important to notice that, since all extrapolation techniques are adiabatic in
their parameters, it is not possible to catch the (essentially nonperturbative) features of
the models in the region below n.
As a consequence there is no inconsistency between
the present statements and our results [10] concerning O(n) O(2) models for n = 2, 3 in
fixed dimension d = 3. The fixed points we found for n = 2, 3 are certainly not analytically
connected with the large-n and small- 4 d criticalities discussed in this paper.
Finally, we recall that an enlarged parameter space for the O(n) O(m) symmetric
models with critical dimension d = 4 leads to the appearance of several new, generally
unstable, fixed points, that physically correspond to tricritical transitions and give rise to
crossover phenomena. It is important to recognize that the conformal bootstrap approach
to the 1/n expansion allows a consistent treatment of all these criticalities. Systems with
O(n) O(m) symmetry may also possess a gauge criticality, which can be described
by the appropriate 1/n expansion as well as within the expansion of the Hamiltonian
for scalar chromodynamics and within the expansion of a class of gauge-invariant
NL models.
Acknowledgements
We thank B. Delamotte, D. Mouhanna, and M. Tissier for useful correspondence.
References
[1] H. Kawamura, J. Phys. Condens. Matter 10 (1998) 4707.
[2] A. Pelissetto, E. Vicari, Critical phenomena and renormalization-group theory, condmat/0012164.
[3] M.F. Collins, O.A. Petrenko, Can. J. Phys. 75 (1997) 605.
634
[4] V.P. Plakhty, J. Kulda, D. Visser, E.V. Moskvin, J. Wosnitza, Phys. Rev. Lett. 85 (2000) 3942.
[5] H. Kawamura, J. Phys. Soc. Japan 54 (1985) 3220;
H. Kawamura, J. Appl. Phys. 61 (1987) 3590.
[6] H. Kawamura, Phys. Rev. B 38 (1988) 4916;
H. Kawamura, Phys. Rev. B 42 (1990) 2610, Erratum.
[7] M. Tissier, D. Mouhanna, B. Delamotte, Phys. Rev. B 61 (2000) 15327.
[8] M. Tissier, B. Delamotte, D. Mouhanna, Phys. Rev. Lett. 84 (2000) 5208.
[9] M. Tissier, B. Delamotte, D. Mouhanna, Int. J. Mod. Phys. A 16 (2001) 2131.
[10] A. Pelissetto, P. Rossi, E. Vicari, Phys. Rev. B 63 (2001) 140414(R).
[11] H. Kawamura, J. Phys. Soc. Japan 61 (1992) 1299.
[12] T. Bhattacharya, A. Billoire, R. Lacaze, Th. Jolicoeur, J. Physique I (Paris) 4 (1994) 181;
A. Mailhot, M.L. Plumer, A. Caill, Phys. Rev. B 50 (1994) 6854;
M.L. Plumer, A. Mailhot, Phys. Rev. B 50 (1994) 16113;
D. Loison, H.T. Diep, Phys. Rev. B 50 (1994) 16453;
E.H. Boubcheur, D. Loison, H.T. Diep, Phys. Rev. B 54 (1996) 4165.
[13] D. Loison, K.D. Schotte, Eur. Phys. J. B 5 (1998) 735;
D. Loison, K.D. Schotte, Eur. Phys. J. B 14 (2000) 125;
D. Loison, Physica A 275 (2000) 207.
[15] P. Azaria, B. Delamotte, F. Delduc, T. Jolicoeur, Nucl. Phys. B 408 (1993) 485.
[16] S.A. Antonenko, A.I. Sokolov, Phys. Rev. B 49 (1994) 15901.
[17] S.A. Antonenko, A.I. Sokolov, V.B. Varnashev, Phys. Lett. A 208 (1995) 161.
[18] S. Hikami, Prog. Theor. Phys. 64 (1980) 1425.
[20] A.I. Mudrov, K.B. Varnashev, Critical thermodynamics and the three-dimensional MN component field model with cubic anisotropy from higher-loop RG expansions, condmat/0011167 (2000).
[21] A.N. Vasilev, Yu.M. Pismak, Yu.R. Khonkonen, Theor. Math. Phys. 46 (1981) 104.
[22] A.N. Vasilev, Yu.M. Pismak, Yu.R. Khonkonen, Theor. Math. Phys. 47 (1981) 291.
[23] J.M. Cornwall, R. Jackiw, E. Tomboulis, Phys. Rev. D 10 (1974) 2428.
[24] A.N. Vasilev, N.Yu. Nalimov, Yu.R. Khonkonen, Theor. Math. Phys. 58 (1984) 111.
[25] D.H. Friedan, Ann. Phys. 163 (1985) 318.
[26] S. Hikami, Phys. Lett. B 98 (1981) 208.
[27] A. Aharony, in: C. Domb, J. Lebowitz (Eds.), Phase Transitions and Critical Phenomena, Vol. 6,
Academic Press, New York, 1976, p. 357.
[28] A. Pelissetto, E. Vicari, Phys. Rev. B 62 (2000) 6393.
[29] J.C. Le Guillou, J. Zinn-Justin, J. Physique 48 (1987) 19.
[30] A. Pelissetto, E. Vicari, Nucl. Phys. B 519 (1998) 626;
A. Pelissetto, E. Vicari, Nucl. Phys. B 522 (1998) 605;
A. Pelissetto, E. Vicari, Nucl. Phys. B 575 (2000) 579.
[31] D. Loison, A.I. Sokolov, B. Delamotte, S.A. Antonenko, K.D. Schotte, H.T. Diep, Pisma v
ZhETF 72 (2000) 487, [JETP Lett. 72 (2000) 337].
[32] P. Azaria, B. Delamotte, T. Jolicoeur, Phys. Rev. Lett. 64 (1990) 3175.
[33] F. David, Th. Jolicoeur, Phys. Rev. Lett. 76 (1996) 3148.
[34] B. Delamotte, private communication.

CUMULATIVE AUTHOR INDEX B601B607
Abel, S.
Adam, C.
Ahn, C.
Akemann, G.
Alford, M.
Ali, D.B.
Aliev, T.M.
Aloisio, R.
Alonso-Alberca, N.
ALPHA Collaboration
Altland, A.
lvarez, E.
Amoros, G.
Anastasiou, C.
Anastasiou, C.
Anastasiou, C.
Ardonne, E.
Arkani-Hamed, N.
Arnowitt, R.
Arutyunov, G.
Astier, P.
Astier, P.
Autiero, D.
Autiero, D.
Azcoiti, V.
B606 (2001) 151

B607 (2001) 247
B601 (2001) 539
B601 (2001) 77
B602 (2001) 61
B605 (2001) 337
B607 (2001) 305
B606 (2001) 322
B602 (2001) 329
B603 (2001) 180
B607 (2001) 511
B603 (2001) 286
B602 (2001) 87
B601 (2001) 318
B601 (2001) 341
B605 (2001) 486
B607 (2001) 549
B605 (2001) 81
B606 (2001) 59
B602 (2001) 238
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B606 (2001) 322
Bagrov, V.G.
Bajnok, Z.
Baldisseri, A.
Baldisseri, A.
Baldo-Ceolin, M.
Baldo-Ceolin, M.
Bandelloni, G.
Banner, M.
Banner, M.
Bassompierre, G.
Bassompierre, G.
Behrndt, K.
Behrndt, K.
Bellini, M.
Bellucci, S.
Benatti, F.
B605 (2001) 425

B601 (2001) 503
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B606 (2001) 673
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B601 (2001) 49
B607 (2001) 391
B604 (2001) 441
B606 (2001) 119
B602 (2001) 541
0550-3213/2001 Published by Elsevier Science B.V.

PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 3 0 1 - 7
Beneke, M.
Benslama, K.
Besson, N.
Besson, N.
Bhaseen, M.J.
Bialas, A.
Bialas, P.
Bijnens, J.
Bintruy, P.
Bird, I.
Bird, I.
Blumenfeld, B.
Blumenfeld, B.
Bobisut, F.
Bobisut, F.
Boer, D.
Bolz, M.
Bonini, M.
Bonneau, G.
Bouchez, J.
Bouchez, J.
Boyd, S.
Boyd, S.
Branco, G.C.
Brandenburg, A.
Braun, V.
Braun, V.M.
Brecher, D.
Buchalla, G.
Buchmller, W.
Bueno, A.
Bueno, A.
Bunyatov, S.
Bunyatov, S.
Buras, A.J.
Burda, Z.
Bytsko, A.G.
B606 (2001) 245

B601 (2001) 3
B601 (2001) 3
B605 (2001) 3
B604 (2001) 537
B603 (2001) 218
B603 (2001) 369
B602 (2001) 87
B604 (2001) 32
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B603 (2001) 195
B606 (2001) 518
B606 (2001) 231
B607 (2001) 293
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B607 (2001) 268
B606 (2001) 518
B607 (2001) 433
B603 (2001) 69
B607 (2001) 155
B606 (2001) 245
B606 (2001) 518
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B605 (2001) 600
B602 (2001) 399
B604 (2001) 455
Cachazo, F.
Cai, R.-G.
akmak, M.K.
Camici, G.
B603 (2001) 3
B606 (2001) 137
B607 (2001) 305
B607 (2001) 431
636
Camilleri, L.
Camilleri, L.
Campbell, B.A.
Cardini, A.
Cardini, A.
Cardoso, G.L.
Carter, B.
Cartier, C.
Casteill, P.-Y.
Castro-Alvaredo, O.A.
Cattaneo, P.W.
Cattaneo, P.W.
Cavasinni, V.
Cavasinni, V.
Cerdeo, D.G.
Cervera-Villanueva, A.
Cervera-Villanueva, A.
Chamblin, A.
Chandrasekharan, S.
Chim, L.
Christova, E.
Chukanov, A.
Ciafaloni, M.
Cognola, G.
Colangelo, G.
Collazuol, G.
Collazuol, G.
Conforto, G.
Conforto, G.
Conta, C.
Conta, C.
Contalbrigo, M.
Contalbrigo, M.
Copeland, E.J.
Cousins, R.
Cousins, R.
Cox, J.
Cski, C.
Curio, G.
Cvetic, M.
Cvetic, M.
B601 (2001) 3
B605 (2001) 3
B606 (2001) 613
B601 (2001) 3
B605 (2001) 3
B607 (2001) 391
B606 (2001) 45
B607 (2001) 406
B607 (2001) 293
B604 (2001) 367
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B603 (2001) 231
B601 (2001) 3
B605 (2001) 3
B607 (2001) 155
B602 (2001) 61
B601 (2001) 539
B607 (2001) 369
B605 (2001) 3
B607 (2001) 431
B602 (2001) 383
B603 (2001) 125
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B607 (2001) 406
B601 (2001) 3
B605 (2001) 3
B602 (2001) 61
B604 (2001) 312
B602 (2001) 172
B605 (2001) 141
B606 (2001) 18
Dalitz, R.H.
Dalmazi, D.
Damgaard, P.H.
Daniels, D.
Daniels, D.
Dasgupta, A.
Davydychev, A.I.
De Azcrraga, J.A.
De Foss, L.
Degaudenzi, H.
Degaudenzi, H.
Deger, N.S.
Delpine, D.
Delgado, A.
B606 (2001) 483

B601 (2001) 77
B601 (2001) 77
B601 (2001) 3
B605 (2001) 3
B606 (2001) 357
B605 (2001) 266
B604 (2001) 75
B603 (2001) 413
B601 (2001) 3
B605 (2001) 3
B604 (2001) 343
B607 (2001) 268
B607 (2001) 99
Del Prete, T.
Del Prete, T.
De Santo, A.
De Santo, A.
Deser, S.
Desrosiers, P.
Di Carlo, G.
Diehl, M.
Dignan, T.
Dignan, T.
Di Lella, L.
Di Lella, L.
Do Couto e Silva, E.
Do Couto e Silva, E.
Dolgushev, V.A.
Domenech-Garret, J.L.
Dorca, M.
Dorey, P.
Dorey, P.
Duff, M.J.
Duff, M.J.
Dumarchez, J.
Dumarchez, J.
Dutta, B.
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B607 (2001) 577
B606 (2001) 547
B606 (2001) 322
B605 (2001) 647
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B606 (2001) 647
B601 (2001) 395
B605 (2001) 215
B603 (2001) 581
B603 (2001) 582
B605 (2001) 141
B605 (2001) 234
B601 (2001) 3
B605 (2001) 3
B606 (2001) 59
Eden, B.
Eguchi, T.
Ellis, M.
Ellis, M.
Emig, T.
Erlich, J.
Evslin, J.
B607 (2001) 191

B607 (2001) 3
B601 (2001) 3
B605 (2001) 3
B604 (2001) 479
B604 (2001) 312
B602 (2001) 486
Fabris, J.C.
Fazio, T.
Fazio, T.
Feldman, G.J.
Feldman, G.J.
Feldmann, Th.
Feng, J.L.
Ferrari, R.
Ferrari, R.
Ferrre, D.
Ferrre, D.
Flaminio, V.
Flaminio, V.
Floreanini, R.
Font, A.
Forte, S.
Fradkin, E.
Fraternali, M.
Fraternali, M.
Frau, M.
Fries, R.J.
Fring, A.
Frizzo, A.
B602 (2001) 644

B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B605 (2001) 647
B602 (2001) 307
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B602 (2001) 541
B605 (2001) 319
B602 (2001) 585
B601 (2001) 591
B601 (2001) 3
B605 (2001) 3
B602 (2001) 39
B607 (2001) 433
B604 (2001) 367
B604 (2001) 92

Frolov, S.
Fujikawa, K.
B602 (2001) 238

B605 (2001) 365
Gabrielli, E.
Gaillard, J.-M.
Gaillard, J.-M.
Gaillard, M.K.
Galajinsky, A.
Galante, A.
Gangler, E.
Gangler, E.
Garbarino, G.
Garousi, M.R.
Garriga, J.
Gasperini, M.
Gasser, J.
Gava, E.
Gehrmann, T.
Gehrmann, T.
Geiser, A.
Geiser, A.
Geppert, D.
Geppert, D.
Gherghetta, T.
Ghilencea, D.M.
Gibbons, G.W.
Gibin, D.
Gibin, D.
Gilkey, P.B.
Gitman, D.M.
Glover, E.W.N.
Glover, E.W.N.
Glover, E.W.N.
Glover, E.W.N.
Gninenko, S.
Gninenko, S.
Godley, A.
Godley, A.
Gomes, J.F.
Gmez, C.
Gomez-Cadenas, J.-J.
Gomez-Cadenas, J.-J.
Gomis, J.
Gonzlez Felipe, R.
Gorbunov, D.S.
Gosset, J.
Gosset, J.
Gling, C.
Gling, C.
Gouanre, M.
Gouanre, M.
Gracey, J.A.
Grant, A.
Grant, A.
Grassi, P.A.
Graziani, G.
B603 (2001) 231

B601 (2001) 3
B605 (2001) 3
B604 (2001) 32
B606 (2001) 119
B606 (2001) 322
B601 (2001) 3
B605 (2001) 3
B606 (2001) 483
B602 (2001) 527
B605 (2001) 192
B607 (2001) 406
B603 (2001) 125
B605 (2001) 17
B601 (2001) 248
B601 (2001) 287
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B602 (2001) 3
B606 (2001) 101
B606 (2001) 18
B601 (2001) 3
B605 (2001) 3
B601 (2001) 125
B605 (2001) 425
B601 (2001) 318
B601 (2001) 341
B605 (2001) 467
B605 (2001) 486
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B606 (2001) 441
B603 (2001) 286
B601 (2001) 3
B605 (2001) 3
B606 (2001) 3
B607 (2001) 268
B602 (2001) 213
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B605 (2001) 337
B601 (2001) 3
B605 (2001) 3
B606 (2001) 380
B601 (2001) 3
637
Graziani, G.
Greene, B.R.
Gribanov, V.
Grillo, A.F.
Grojean, C.
Gubser, S.S.
Gueuvoghlanian, E.P.
Guglielmi, A.
Guglielmi, A.
Gukov, S.
B605 (2001) 3
B604 (2001) 181
B607 (2001) 355
B606 (2001) 322
B604 (2001) 312
B605 (2001) 395
B606 (2001) 441
B601 (2001) 3
B605 (2001) 3
B601 (2001) 49
Hagner, C.
Hagner, C.
Hjcek, P.
Hjcek, P.
Hall, L.
Hambye, T.
Hammou, A.B.
Hara, T.
Hautmann, F.
Hernndez, L.
Hernando, J.
Hernando, J.
Hikami, K.
Hubbard, D.
Hubbard, D.
Huber, S.J.
Huerta, M.
Hurst, P.
Hurst, P.
Hyett, N.
Hyett, N.
B601 (2001) 3
B605 (2001) 3
B603 (2001) 531
B603 (2001) 555
B605 (2001) 81
B602 (2001) 23
B605 (2001) 17
B602 (2001) 499
B604 (2001) 391
B603 (2001) 286
B601 (2001) 3
B605 (2001) 3
B604 (2001) 580
B601 (2001) 3
B605 (2001) 3
B606 (2001) 183
B601 (2001) 591
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
Iacopini, E.
Iacopini, E.
Ilha, A.
Intriligator, K.
Ishibashi, M.
Iso, S.
Iucci, A.
Izquierdo, J.M.
B601 (2001) 3
B605 (2001) 3
B604 (2001) 426
B603 (2001) 3
B605 (2001) 365
B604 (2001) 121
B601 (2001) 607
B604 (2001) 75
Jger, S.
Jakob, R.
Joseph, C.
Joseph, C.
Juget, F.
Juget, F.
Jurco, B.
B605 (2001) 600

B605 (2001) 647
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B604 (2001) 148
Kalmykov, M.Yu.
Kamani, D.
Kaminsky, K.
Kardar, M.
Karsch, F.
Kaya, A.
B605 (2001) 266

B601 (2001) 149
B606 (2001) 613
B604 (2001) 479
B605 (2001) 579
B604 (2001) 343
638
Kennedy, A.D.
Ketov, S.V.
Khalil, S.
Khalil, S.
Khodjamirian, A.
Kiefer, C.
Kiem, Y.
Kim, J.E.
Kim, Y.
Kimura, Y.
King, S.F.
Kirsanov, M.
Kirsanov, M.
Kirsten, K.
Klein, M.
Klemm, D.
Klimov, O.
Klimov, O.
Klinkhamer, F.R.
Klishevich, S.M.
Koerber, P.
Kokkonen, J.
Kokkonen, J.
Korchemsky, G.P.
Kovalenko, S.
Kovzelev, A.
Kovzelev, A.
Krasnoperov, A.
Krasnoperov, A.
Krause, A.
Kroll, P.
Kustov, D.
Kuznetsov, V.
Kuznetsov, V.
B607 (2001) 456

B604 (2001) 256
B603 (2001) 231
B606 (2001) 151
B605 (2001) 558
B603 (2001) 531
B601 (2001) 27
B602 (2001) 346
B602 (2001) 467
B604 (2001) 121
B607 (2001) 77
B601 (2001) 3
B605 (2001) 3
B601 (2001) 125
B605 (2001) 319
B601 (2001) 380
B601 (2001) 3
B605 (2001) 3
B607 (2001) 247
B606 (2001) 583
B603 (2001) 413
B601 (2001) 3
B605 (2001) 3
B603 (2001) 69
B607 (2001) 355
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B602 (2001) 172
B605 (2001) 647
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
Lacaprara, S.
Lacaprara, S.
Lachaud, C.
Lachaud, C.
Laermann, E.
Lakic, B.
Lakic, B.
Lanza, A.
Lanza, A.
Lapointe, L.
LaRotonda, L.
La Rotonda, L.
Laveder, M.
Laveder, M.
Lazaroiu, C.I.
Lazaroiu, C.I.
Lazaroiu, C.I.
Lazzarini, S.
Leader, E.
Leo, C.R.
Lebedev, O.
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B605 (2001) 579
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B606 (2001) 547
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B603 (2001) 497
B604 (2001) 181
B605 (2001) 159
B606 (2001) 673
B607 (2001) 369
B602 (2001) 514
B606 (2001) 151
Lee, H.M.
Le Moul, C.
Letessier-Selvon, A.
Letessier-Selvon, A.
Leutwyler, H.
Levin, A.
Levy, J.-M.
Levy, J.-M.
Li, K.
Li, M.
Liccardo, A.
Lindner, M.
Ling, Y.
Linssen, L.
Linssen, L.
Lipan, O.
Liu, J.T.
Liu, J.T.
Liu, J.T.
Ljubicic, A.
Ljubicic, A.
Loginov, E.K.
Loll, R.
Long, H.N.
Long, J.
Long, J.
Lozano-Tellechea, E.
L, H.
L, H.
Lu, J.X.
Lukyanov, S.
Lupi, A.
Lupi, A.
Lst, D.
Lyakhovich, S.L.
B602 (2001) 346

B607 (2001) 38
B601 (2001) 3
B605 (2001) 3
B603 (2001) 125
B605 (2001) 425
B601 (2001) 3
B605 (2001) 3
B601 (2001) 607
B602 (2001) 201
B602 (2001) 39
B607 (2001) 326
B601 (2001) 191
B601 (2001) 3
B605 (2001) 3
B604 (2001) 603
B605 (2001) 116
B605 (2001) 141
B605 (2001) 234
B601 (2001) 3
B605 (2001) 3
B606 (2001) 636
B606 (2001) 357
B601 (2001) 361
B601 (2001) 3
B605 (2001) 3
B607 (2001) 213
B605 (2001) 141
B606 (2001) 18
B606 (2001) 137
B607 (2001) 437
B601 (2001) 3
B605 (2001) 3
B607 (2001) 391
B606 (2001) 647
Ma, E.
Ma, J.P.
Ma, J.P.
Macfarlane, A.J.
Magnea, L.
Mahnke, N.
Manashov, A.N.
Mangano, M.L.
March-Russell, J.
Marchionni, A.
Marchionni, A.
Martelli, F.
Martelli, F.
Martn-Delgado, M.A.
Mathieu, P.
McInnes, B.
Mchain, X.
Mchain, X.
Meessen, P.
Meggiolaro, E.
B602 (2001) 23
B602 (2001) 572
B605 (2001) 625
B604 (2001) 75
B604 (2001) 92
B607 (2001) 433
B603 (2001) 69
B602 (2001) 585
B602 (2001) 307
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B601 (2001) 569
B606 (2001) 547
B602 (2001) 132
B601 (2001) 3
B605 (2001) 3
B602 (2001) 329
B602 (2001) 261

Meggiolaro, E.
Mendiburu, J.-P.
Mendiburu, J.-P.
Merkt, R.
Merlatti, P.
Meyer, J.-P.
Meyer, J.-P.
Mezzetto, M.
Mezzetto, M.
Miao, Y.-G.
Mishra, S.R.
Mishra, S.R.
Molke, H.
Moon, S.-H.
Moore, G.
Moorhead, G.F.
Moorhead, G.F.
Morales, J.F.
Moreau, G.
Morel, A.
Mueller, A.H.
Mller-Kirsten, H.J.W.
Munier, S.
Muoz, C.
Musto, R.
B606 (2001) 337

B601 (2001) 3
B605 (2001) 3
B607 (2001) 511
B602 (2001) 453
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B606 (2001) 84
B601 (2001) 3
B605 (2001) 3
B603 (2001) 180
B602 (2001) 467
B607 (2001) 117
B601 (2001) 3
B605 (2001) 3
B605 (2001) 17
B604 (2001) 3
B603 (2001) 369
B603 (2001) 427
B606 (2001) 84
B603 (2001) 427
B603 (2001) 231
B602 (2001) 39
Nagao, T.
Nan, C.M.
Narain, K.S.
Naumov, D.
Naumov, D.
Navelet, H.
Ndlec, P.
Ndlec, P.
Nefedov, Yu.
Nefedov, Yu.
Nelson, B.D.
Neubert, M.
Nguyen-Mau, C.
Nguyen-Mau, C.
Nielsen, H.B.
NOMAD Collaboration
NOMAD Collaboration
Nomura, Y.
Norrbin, E.
B602 (2001) 622

B601 (2001) 607
B605 (2001) 17
B601 (2001) 3
B605 (2001) 3
B603 (2001) 218
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B604 (2001) 32
B606 (2001) 245
B601 (2001) 3
B605 (2001) 3
B604 (2001) 405
B601 (2001) 3
B605 (2001) 3
B605 (2001) 81
B603 (2001) 297
Ohlsson, T.
Oleari, C.
Oleari, C.
Oleari, C.
Oleari, C.
Oller, J.A.
Orestano, D.
Orestano, D.
Orland, P.
Ortn, T.
B607 (2001) 326

B601 (2001) 318
B601 (2001) 341
B605 (2001) 467
B605 (2001) 486
B602 (2001) 641
B601 (2001) 3
B605 (2001) 3
B605 (2001) 64
B602 (2001) 329
639
Ortn, T.
B607 (2001) 213
Palla, L.
Park, D.H.
Park, D.K.
Pastore, F.
Pastore, F.
Peak, L.S.
Peak, L.S.
Pearce, P.A.
Peikert, A.
Pelinson, A.M.
Pelissetto, A.
Pendleton, B.
Pennacchio, E.
Pennacchio, E.
Peradze, G.
Perez, E.
Peschanski, R.
Pessard, H.
Pessard, H.
Petersson, B.
Petersson, B.
Petkou, A.C.
Petkou, A.C.
Petkou, A.C.
Petkova, V.B.
Petriello, F.J.
Petrov, K.
Petti, R.
Petti, R.
Placci, A.
Placci, A.
Plyushchay, M.S.
Polesello, G.
Polesello, G.
Polesello, G.
Policastro, G.
Pollmann, D.
Pollmann, D.
Polyarush, A.
Polyarush, A.
Pomarol, A.
Pope, C.N.
Pope, C.N.
Popov, B.
Popov, B.
Porrati, M.
Poulsen, C.
Poulsen, C.
Pujols, O.
B601 (2001) 503

B601 (2001) 27
B606 (2001) 84
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B601 (2001) 539
B605 (2001) 579
B602 (2001) 644
B607 (2001) 605
B607 (2001) 456
B601 (2001) 3
B605 (2001) 3
B607 (2001) 117
B604 (2001) 3
B603 (2001) 218
B601 (2001) 3
B605 (2001) 3
B602 (2001) 399
B603 (2001) 369
B601 (2001) 380
B602 (2001) 238
B607 (2001) 191
B603 (2001) 449
B601 (2001) 169
B603 (2001) 369
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B606 (2001) 583
B601 (2001) 3
B604 (2001) 3
B605 (2001) 3
B606 (2001) 380
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B602 (2001) 3
B605 (2001) 141
B606 (2001) 18
B601 (2001) 3
B605 (2001) 3
B606 (2001) 380
B601 (2001) 3
B605 (2001) 3
B605 (2001) 192
Quevedo, F.
Quirs, M.
B605 (2001) 319

B607 (2001) 99
Rathouit, P.
B601 (2001)
640
Ravindran, V.
Rayner, D.A.J.
Read, N.
Reall, H.S.
Reisz, T.
Remiddi, E.
Remiddi, E.
Resco, P.
Rtey, A.
Rey, S.-J.
Rezayi, E.
Riccioni, F.
Rico, J.
Rico, J.
Ridolfi, G.
Riemann, P.
Riva, V.
Rivelles, V.O.
Roda, C.
Roda, C.
Rodriguez-Laguna, J.
Romano, R.
Ross, G.G
Rossi, P.
Rubbia, A.
Rubbia, A.
Russo, J.G.
Russo, R.
B605 (2001) 517

B607 (2001) 77
B607 (2001) 549
B607 (2001) 155
B603 (2001) 369
B601 (2001) 248
B601 (2001) 287
B603 (2001) 286
B604 (2001) 281
B602 (2001) 467
B607 (2001) 549
B605 (2001) 245
B601 (2001) 3
B605 (2001) 3
B602 (2001) 585
B605 (2001) 3
B604 (2001) 511
B602 (2001) 514
B601 (2001) 3
B605 (2001) 3
B601 (2001) 569
B602 (2001) 541
B606 (2001) 101
B607 (2001) 605
B601 (2001) 3
B605 (2001) 3
B602 (2001) 109
B604 (2001) 92
Sabella, G.
Sabra, W.A.
Sachrajda, C.T.
Sahlmann, H.
Sakai, N.
Salvatore, F.
Salvatore, F.
Sanchis-Lozano, M.A.
Santoso, Y.
Sarkar, U.
Sati, H.
Sato, H.-T.
Saulina, N.
Savc, M.
Schahmaneche, K.
Schahmaneche, K.
Schmidt, B.
Schmidt, B.
Schmidt, I.
Schmidt, M.G.
Schmidt, T.
Schoutens, K.
Schubert, C.
Schupp, P.
Seki, S.
Sekino, Y.
Sethi, S.
B602 (2001) 453

B605 (2001) 234
B606 (2001) 245
B606 (2001) 401
B602 (2001) 413
B601 (2001) 3
B605 (2001) 3
B601 (2001) 395
B606 (2001) 59
B602 (2001) 23
B605 (2001) 116
B601 (2001) 27
B607 (2001) 117
B607 (2001) 305
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B607 (2001) 355
B606 (2001) 183
B605 (2001) 3
B607 (2001) 549
B607 (2001) 191
B604 (2001) 148
B606 (2001) 689
B602 (2001) 147
B602 (2001) 307
Sevior, M.
Sevior, M.
Sevrin, A.
Sevrin, A.
Sezgin, E.
Shapiro, I.L.
Sharapov, A.A.
Shmakova, M.
Shore, G.M.
Sierra, G.
Sillou, D.
Sillou, D.
Siopsis, G.
Sjstrand, T.
Smith, D.
Smolin, L.
Smolin, L.
Soa, D.V.
Sokal, A.D.
Sokatchev, E.
Soler, F.J.P.
Soler, F.J.P.
Sondhi, S.L.
Sotkov, G.M.
Sozzi, G.
Sozzi, G.
Starinets, A.O.
Stasto, A.M.
Steele, D.
Steele, D.
Stein, E.
Stelle, K.S.
Stiegler, U.
Stiegler, U.
Stipcevic, M.
Stipcevic, M.
Stolarczyk, Th.
Stolarczyk, Th.
Sundell, P.
B601 (2001) 3
B605 (2001) 3
B603 (2001) 389
B603 (2001) 413
B604 (2001) 343
B602 (2001) 644
B606 (2001) 647
B601 (2001) 49
B605 (2001) 455
B601 (2001) 569
B601 (2001) 3
B605 (2001) 3
B601 (2001) 380
B603 (2001) 297
B605 (2001) 81
B601 (2001) 191
B601 (2001) 209
B601 (2001) 361
B601 (2001) 425
B607 (2001) 191
B601 (2001) 3
B605 (2001) 3
B605 (2001) 395
B606 (2001) 441
B601 (2001) 3
B605 (2001) 3
B601 (2001) 425
B603 (2001) 427
B601 (2001) 3
B605 (2001) 3
B607 (2001) 433
B605 (2001) 141
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B604 (2001) 343
Tabaczek, J.
Takcs, G.
Takanishi, Y.
Takayanagi, T.
Talavera, P.
Tamaryan, S.
Tanaka, K.
Tanaka, T.
Tani, T.
Tanii, Y.
Tareb-Reyes, M.
Tareb-Reyes, M.
Tateo, R.
Tateo, R.
Taylor, G.N.
Taylor, G.N.
B602 (2001) 399

B601 (2001) 503
B604 (2001) 405
B603 (2001) 259
B602 (2001) 87
B606 (2001) 84
B604 (2001) 121
B605 (2001) 192
B602 (2001) 434
B604 (2001) 343
B601 (2001) 3
B605 (2001) 3
B603 (2001) 581
B603 (2001) 582
B601 (2001) 3
B605 (2001) 3

Tejeda-Yeomans, M.E.
Tereshchenko, V.
Tereshchenko, V.
Theis, U.
Thiemann, T.
Tlyachev, V.B.
Tomizawa, S.
Toropin, A.
Toropin, A.
Torrente-Lujan, E.
Toublan, D.
Touchard, A.-M.
Touchard, A.-M.
Tovey, S.N.
Tovey, S.N.
Tran, M.-T.
Tran, M.-T.
Tricarico, E.
Troost, J.
Troost, W.
Tsesmelis, E.
Tsesmelis, E.
B601 (2001) 318

B601 (2001) 341
B605 (2001) 467
B605 (2001) 486
B601 (2001) 3
B605 (2001) 3
B602 (2001) 367
B606 (2001) 401
B605 (2001) 425
B602 (2001) 413
B601 (2001) 3
B605 (2001) 3
B603 (2001) 231
B603 (2001) 343
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B606 (2001) 231
B603 (2001) 389
B603 (2001) 389
B601 (2001) 3
B605 (2001) 3
Ulrichs, J.
Ulrichs, J.
Urban, J.
Uzan, J.-P.
B601 (2001) 3
B605 (2001) 3
B605 (2001) 600
B606 (2001) 45
Vacavant, L.
Vacavant, L.
Vafa, C.
Valdata-Nappi, M.
Valdata-Nappi, M.
Valuev, V.
Valuev, V.
Van de Bruck, C.
Van Neerven, W.L.
Van Neerven, W.L.
Vannucci, F.
Vannucci, F.
Varadarajan, U.
Varvell, K.E.
Varvell, K.E.
Vassilevich, D.V.
Veltri, M.
Veltri, M.
Verbaarschot, J.J.M.
Verbaarschot, J.J.M.
Vercesi, V.
Vercesi, V.
Vermaseren, J.A.M.
Vicari, E.
B601 (2001) 3
B605 (2001) 3
B603 (2001) 3
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B605 (2001) 215
B603 (2001) 42
B605 (2001) 517
B601 (2001) 3
B605 (2001) 3
B602 (2001) 486
B601 (2001) 3
B605 (2001) 3
B601 (2001) 125
B601 (2001) 3
B605 (2001) 3
B601 (2001) 77
B603 (2001) 343
B601 (2001) 3
B605 (2001) 3
B604 (2001) 281
B607 (2001) 605
641
Vidal-Sitjes, G.
Vidal-Sitjes, G.
Vieira, J.-M.
Vieira, J.-M.
Vinogradova, T.
Vinogradova, T.
Vogt, A.
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3
B603 (2001) 42
Wgner, F.
Wakatsuki, K.
Waldron, A.
Wang, J.E.
Warner, N.P.
Weber, F.V.
Weber, F.V.
Weiner, N.
Weisse, T.
Weisse, T.
Wess, J.
Wetterich, C.
Wiese, U.-J.
Wilczek, F.
Wilson, F.F.
Wilson, F.F.
Winkler, O.
Winter, W.
Winton, L.J.
Winton, L.J.
Wolff, U.
Wotzasek, C.
Wu, Y.-S.
Wu, Y.-S.
B601 (2001) 503

B604 (2001) 121
B607 (2001) 577
B602 (2001) 486
B607 (2001) 3
B601 (2001) 3
B605 (2001) 3
B605 (2001) 81
B601 (2001) 3
B605 (2001) 3
B604 (2001) 148
B606 (2001) 337
B602 (2001) 61
B602 (2001) 307
B601 (2001) 3
B605 (2001) 3
B606 (2001) 401
B607 (2001) 326
B601 (2001) 3
B605 (2001) 3
B603 (2001) 180
B604 (2001) 426
B604 (2001) 551
B606 (2001) 137
Xiong, Z.
B602 (2001) 289
Yabsley, B.D.
Yabsley, B.D.
Yang, H.-X.
Yang, J.M.
Yang, S.-K.
Yoneya, T.
Yu, Y.
B601 (2001) 3
B605 (2001) 3
B604 (2001) 551
B602 (2001) 289
B607 (2001) 3
B602 (2001) 499
B604 (2001) 551
Zaccone, H.
Zaccone, H.
Zamolodchikov, Al.
Zemba, G.R.
Zerbini, S.
Zhou, J.-G.
Zimerman, A.H.
Zuber, J.-B.
Zuber, K.
Zuber, K.
Zuccon, P.
Zuccon, P.
B601 (2001) 3
B605 (2001) 3
B607 (2001) 437
B601 (2001) 591
B602 (2001) 383
B607 (2001) 237
B606 (2001) 441
B603 (2001) 449
B601 (2001) 3
B605 (2001) 3
B601 (2001) 3
B605 (2001) 3

Nucl - Phys.B v.607

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.607

Uploaded by

Copyright:

Available Formats

Nuclear Physics B 607 (2001) 337

ADE singularities and coset models

University of Southern California, Los Angeles, CA 90089-0484, USA

Received 22 May 2001; accepted 23 May 2001

T. Eguchi et al. / Nuclear Physics B 607 (2001) 337

T. Eguchi et al. / Nuclear Physics B 607 (2001) 337

T. Eguchi et al. / Nuclear Physics B 607 (2001) 337

the algorithm developed in Section 3, to reconstruct the LandauGinzburg superpotentials

2. The LandauGinzburg potentials of deformed SLOHSS models

T. Eguchi et al. / Nuclear Physics B 607 (2001) 337

T. Eguchi et al. / Nuclear Physics B 607 (2001) 337

3. Chiral rings and superpotentials

T. Eguchi et al. / Nuclear Physics B 607 (2001) 337

The CSA of U (m + n) can be parameterized by (1 , . . . , m+n ), and the Casimirs of G are

while those of H are:

It is convenient to introduce an equivalent set of Casimirs:

along with a corresponding family of new variables:

1j1 <j2 <<jk m

k given Vk = vk . The whole point is that one has

T. Eguchi et al. / Nuclear Physics B 607 (2001) 337

Wm,n+1m (xj ;vj )

T. Eguchi et al. / Nuclear Physics B 607 (2001) 337

all square-roots from the result. For n = 4 and 5 one obtains:

In the first of the superpotentials we have made the change of variables: z = z1 , z2 =

V2j = x2j + x2j 2 x12 , j = 2, 3, . . . , n 2,

v2j = (1)n+j v2j ,

n vn = 0. Introduce the superpotential so that

T. Eguchi et al. / Nuclear Physics B 607 (2001) 337

then the desired D2n potential is obtained as

where we have set x1 = x and x n1 = y.

In view of the decomposition 2n = n2 + n 2 under SO(2n)/SU(n) U (1), the SO(2n)

SO(8)/SO(6) U (1) model through the SO(8) triality transformation.

T. Eguchi et al. / Nuclear Physics B 607 (2001) 337

V4 (zj ) = 160z14 16z2 z12 + 12z3z1 + 2z4 + z22 ,

5 ), we solve the first three equations

In order to obtain the LandauGinzburg superpotential we set

T. Eguchi et al. / Nuclear Physics B 607 (2001) 337

the ground states of the SO(2n)/(SO(2n 2) U (1)) and SO(2n)/(SU(n) U (1))

For example, for n = 5 we find:

c4 = 150z14 5v2 z12 10z3 z1 18 v22 + 12 v4 ,

The LandauGinzburg superpotential for the SO(2n)/(SU(n) U (1)) coset model is

where (3.15) and (3.14) have been used.

T. Eguchi et al. / Nuclear Physics B 607 (2001) 337

V12 (xj ) = 177660x112 + 97902x110x2 +

T. Eguchi et al. / Nuclear Physics B 607 (2001) 337

v8 = 5w4 + 4w1 w3 + 15w14 ,

+ 13 w32 + 4w13 w3 + 19w1w22 + w16 .

+ 5w1 x 2 + 2w12 one arrives at the

This potential then has the property that:

3.4. Single variable potentials

T. Eguchi et al. / Nuclear Physics B 607 (2001) 337

T. Eguchi et al. / Nuclear Physics B 607 (2001) 337

This is precisely the one-variable E6 potentials discussed in [4,16,17]. In particular, one

with an even number of signs. Expanding this we have

T. Eguchi et al. / Nuclear Physics B 607 (2001) 337

To get to a multi-variable, rational potential one simply has to replace q2 by a new

T. Eguchi et al. / Nuclear Physics B 607 (2001) 337

H (z1 , z2 ) = z1n1 + z1 z22 +

T. Eguchi et al. / Nuclear Physics B 607 (2001) 337

In these expressions + indicates the addition of all possible relevant deformations of

T. Eguchi et al. / Nuclear Physics B 607 (2001) 337

dz1 dz2 dz3 dz4

T. Eguchi et al. / Nuclear Physics B 607 (2001) 337

k given Vk = vk . The whole point is that one has

B3 may be considered as a polynomial in H2 . For Yt Max[ (3g22 g12 )/12, g1 / 3]

(m2t + f 2 m2t )[4Yt2 ( f 2 + 1) +