Supersymmetric Domain Walls from Metrics of Special Holonomy

Nuclear Physics B 623 (2002) 346
www.elsevier.com/locate/npe
Supersymmetric domain walls from metrics of

special holonomy
G.W. Gibbons a,1,5 , H. L b,2 , C.N. Pope a,c,d,1,3,4 , K.S. Stelle e,4,5
a DAMTP, Centre for Mathematical Sciences, Cambridge University, Wilberforce Road,
Cambridge CB3 OWA, UK

b Department of Physics, University of Michigan, Ann Arbor, MJ 48109, USA
c Center for Theoretical Physics, Texas A&M University, College Station, TX 77843, USA
d Laboratoire de Physique Thorique de lcole Normale Suprieure, 24 Rue Lhomond,
75231 Paris Cedex 05, France

e The Blackett Laboratory, Imperial College, Prince Consort Road, London SW7 2BZ, UK
Received 27 November 2001; accepted 12 December 2001
Abstract
Supersymmetric domain-wall spacetimes that lift to Ricci-flat solutions of M-theory admit
generalized Heisenberg (2-step nilpotent) isometry groups. These metrics may be obtained from
known cohomogeneity one metrics of special holonomy by taking a Heisenberg limit, based on
an InnWigner contraction of the isometry group. Associated with each such metric is an Einstein
metric with negative cosmological constant on a solvable group manifold. We discuss the relevance
of our metrics to the resolution of singularities in domain-wall spacetimes and some applications to
holography. The extremely simple forms of the explicit metrics suggest that they will be useful for
many other applications. We also give new but incomplete inhomogeneous metrics of holonomy
SU(3), G2 and Spin(7), which are T1 , T2 and T3 bundles respectively over hyper-Khler fourmanifolds. 2002 Published by Elsevier Science B.V.
PACS: 11.27.+d
E-mail address: k.stelle@ic.ac.uk (K.S. Stelle).

1 Research supported in part by the EC under RTN contract HPRN-CT-2000-00122.
2 Research supported in full by DOE grant DE-FG02-95ER40899.
3 Research supported in part by DOE grant DE-FG03-95ER40917.
4 Research supported in part by the EC under RTN contract HPRN-CT-2000-00131.
5 Research supported in part by PPARC SPG grant PPA/G/S/1998/00613.
0550-3213/02/$ see front matter 2002 Published by Elsevier Science B.V.

PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 4 0 - X
G.W. Gibbons et al. / Nuclear Physics B 623 (2002) 346
1. Introduction
Spaces with special holonomy are natural candidates for the extra dimensions in
string and M-theory, since they provide a simple geometrical mechanism for reducing the
number of supersymmetries. Complete non-singular examples on non-compact manifolds
have been constructed where the Ricci-flat metrics can be given explicitly. Attention has
mostly focused on cases of cohomogeneity one that are asymptotically conical (AC)
or asymptotically locally conical (ALC). The AC examples include the EguchiHanson
metric in D = 4 [1], deformed or resolved conifolds in D = 6 [2], and G2 and Spin(7)
holonomy metrics in D = 7 and 8 [3,4]. Four-dimensional ALC solutions have been
also known for some time; they are the TaubNUT [5] and AtiyahHitchin metrics [6].
Supersymmetric higher-dimensional ALC solutions have been elusive, until the recent
explicit constructions of ALC metrics with Spin(7) holonomy [7] and G2 holonomy [8].
One characteristic of those manifolds is that they all have non-Abelian isometry groups.
Another situation where special holonomies are encountered is in BPS solutions in
lower-dimensional supergravities that are supported by fields originating purely from the
gravitational sector of a higher-dimensional theory. After oxidising the solutions back to
the higher dimension, they give rise to Ricci-flat metrics. Since the BPS solutions partially
break supersymmetry, while retaining a certain number of Killing spinors, it follows that
the Ricci-flat metrics will have special holonomy. In [911], a geometrical interpretation of
these domain walls as Ricci-flat spaces with toroidal fibre bundle level surfaces was given.
Amongst the BPS solutions are a special class of domain-walls ((D 2)-branes in D
dimensions) that have the property of scale invariance. Technically, this means that they
possess homotheties, i.e., conformal Killing symmetries where the conformal scaling factor
is a constant.
The class of scale-invariant domain walls has appeared in another context, namely, the
possibility of blowing up the singularities into regular manifolds. An example, of this is
given by a singular limit of K3 that produces the transverse and internal dimensions of
the oxidation of an eight-dimensional 6-brane to D = 11 [11]. Since metrics on K3 are
not known explicitly, the discussion was necessarily a highly implicit one. For our present
purposes, however, the salient properties of the K3 degeneration for this identification with
the domain wall were the appearance of a Heisenberg symmetry in the singular limit,
as well as a characteristic rate of growth of the volume of the manifold as one recedes
from the singularity. Higher-dimensional examples with more internal directions, related
to higher-dimensional CalabiYau manifolds, were also considered in [11]. The associated
domain walls have generalised Heisenberg symmetries. Since these Heisenberg groups are
homothetically invariant, they fall into the class of scale-invariant domain walls that we are
concerned with here.
Four-dimensional supergravity domain walls arising from matter superpotentials
have been extensively studied in [1216]. Domain walls can also exist in maximal
supergravities. For example, the D8-brane of massive type IIA supergravity [17] was
discussed in [18]. Generalised ScherkSchwarz reductions give rise to lower-dimensional
massive supergravities that admit domain-wall solutions. It was shown that the D7-brane
of type IIB and the D8-brane of massive type IIB are T-dual, via a generalised Scherk
Schwarz S 1 reduction [19]. A large class of domain walls arising from ScherkSchwarz
reduction was obtained in [23]. A complete classification of such domain walls in maximal
supergravities was given in [24].
In lower-dimensional maximal supergravities, the cosmological potential associated
with the construction of supersymmetric domain walls can arise either by generalised
KaluzaKlein reduction on spheres, or by generalised toroidal reductions, where in both
cases internal fluxes are turned on. The former give cosmological potentials with at
least two exponential terms, whilst the latter can give potentials with a single (positive)
exponential. Importantly for our purposes, the latter have the feature that the potential has
no intrinsic scale, and so the associated domain walls are scale invariant.
One motivation for the present work was to study the possibility of smooth resolutions
of HoravaWitten type geometries. The idea would be to seek everywhere smooth solutions
of eleven-dimensional supergravity that resemble two domain walls at the ends of a finite
interval. This was discussed in the context of domain walls based on the ur-Heisenberg
group in Ref. [11]. In that reference, it was shown that the singularity arising from the
vanishing of the harmonic function could be resolved by replacing the four-dimensional
hyper-Khler metric M4 by a smooth complete everywhere non-singular hyper-Khler
metric Mresolved on the complement in CP2 of a smooth cubic. The smooth non-singular
metric Mresolved (called the BKTY metric in [11]) is non-compact and has a single end
(i.e., a single connected infinite region) which is given by M4 up to small terms as one
goes away from the domain-wall source. This was referred to in [11] as a single-sided
domain wall.
The question naturally arises whether two such single-sided domain walls may be
joined together by an extended neck to form a complete non-singular compact manifold
Mcompact which resembles the HoravaWitten type geometry. For these purposes, we need
not restrict ourselves to four-dimensional manifolds and shall consider any dimension less
than eleven.
To answer this question, we need to make some further assumptions about the geometry
of the neck region. In the light of the previous example, it seems reasonable to require
that the neck region be of cohomogeneity one, perhaps with the group being one of the
generalised Heisenberg or Nilpotent groups that arise in the known supersymmetric domain
walls of M-theory. We could, of course, assume a more general group or more generally
drop the assumption that the neck region is invariant under any group action. However, it
does seem reasonable to assume that the neck region is covered by a coordinate patch in
which the metric takes the form
ds 2 = dt 2 + gij (x, t) dx i dx j + ,
(1)
where t is the proper distance along the neck. In the cohomogeneity one case
gij (x, t) dx i dx j is a left-invariant metric on G/H , and the ellipsis denotes extra terms
which grow at very large |t| and which may break G-invariance, corresponding to corrections to the metric arising from the smooth resolutions at either end of the interval.
If Mcompact is Ricci-flat, or more generally if Rt t is non-negative, a simple consistency
check immediately arises. The curves x i = const, with tangent vectors T = /t , constitute
a congruence of geodesics of the metric (1). A congruence of curves in a (d + 1)dimensional manifold is a d-dimensional family of curves, one passing through every
point of the manifold. A congruence is hypersurface-orthogonal (or vorticity-free) if
the curves are orthogonal to a family of d-dimensional surfaces. The congruence we

are considering is clearly hypersurface-orthogonal, since every curve is orthogonal to the
surfaces t = constant.
Now let
V (x, t) =
det gij
(2)
and let = VV = g ij tij . Then (t, x i ) is the expansion rate of the geodesic congruence,
and is therefore subject to the Raychaudhuri equation,6 which then reads
1
d
2 2 2 ,
dt
d
(3)
g
where 2 = 12 ij ij and ij = tij d1 g rs gtrs gij with d + 1 = dim Mcompact . The

quantity 2 is a measure of the shear of the geodesic congruence given by x i = const.
It is an easy consequence of (3) that if is negative, it remains negative and moreover
tends to minus infinity in finite proper time t. This means that if the volume V of the neck
is decreasing at one value of t, it is always decreasing. This simple result, which may be
verified in our explicit examples, indicates that neck geometries in which V increases as
one goes outward to the resolved regions in both directions are excluded. They also show
that resolving periodic arrays of domain walls in such a way as to make the metric gij (x, t)
periodic in the proper distance variable t are excluded.
In this paper, we shall study the relationship between scale-invariant domain walls
with Heisenberg symmetries, and complete non-singular manifolds of special holonomy.
The simplest example, which we discuss in Section 4, involves the oxidation of the
6-brane in D = 8 to D = 11. We show that the associated four-dimensional Ricciflat space (which has SU(2) holonomy and thus is self-dual) can be obtained from
the EguchiHanson metric, by taking a rescaling limit in which the SU(2) isometry
degenerates to the Heisenberg group. (For this limit, one has to take the version of
EguchiHanson where the curvature singularity appears in the manifold. The relation to
the non-singular EguchiHanson requires an additional analytic continuation in the scale
parameter.) The other examples, corresponding to higher-dimensional Ricci-flat metrics,
are similarly obtained as Heisenberg limits of higher-dimensional metrics of special
holonomy.
It turns out that each of our scale-invariant Ricci flat metrics with nilpotent isometry
group acting on orbits of co-dimension one is closely related to a complete homogeneous
Einstein manifold with negative cosmological constant with a solvable isometry group.
The simplest case is flat space, En with metric ds 2 = dt 2 + dx dx , which is related
to the hyperbolic space H n with metric ds 2 = dt 2 + e2k t dx dx . In some cases these
metrics have been used as replacements for the hyperbolic space H n in the AdS/CFT
correspondence [25]. A striking feature is the degenerate nature of the conformal boundary.
For this reason we shall include a discussion of these metrics and some of their properties.
6 For a brief review of the Raychaudhuri equation, see Appendix A.
2. Four-dimensional manifolds with SU(2) holonomy

2.1. The basic domain-wall construction
We consider a domain wall solution in eight-dimensional maximal supergravity,
1
supported by the 0-form field strength F(0)23
coming from the dimensional reduction of
7
the KaluzaKlein vector in D = 10. The metric is given by
ds82 = H 1/6 dx dx + H 7/6 dy 2,
(4)
where H = 1 + m|y|. Oxidised back to D = 11 using the standard KK rules, we find that
2 = dx dx + ds 2 where
the eleven-dimensional metric is given by ds11
4

ds42 = H dy 2 + H 1 (dz1 + mz3 dz2 )2 + H dz22 + dz32 .
(5)
Since the CJS field F(4) is zero, ds42 must be Ricci flat. The solution preserves 1/2 of
the supersymmetry, implying that (5) has SU(2) holonomy, i.e., it is a Ricci-flat Khler
metric. The eleven-dimensional solution was obtained in [10], where domain wall charge
quantisation through topological constraints was discussed. It was used in [29] to argue
that M-theory compactified on a T 2 bundle over S 1 is dual to the massive type IIA string
theory. The solution has a singularity at y = 0. In [11], it was shown that the metric (5) is
the asymptotic form of a complete non-singular hyper-Khler metric on the complement in
CP2 of a smooth cubic curve. The metric (5) was obtained by means of a double T-duality
transformation of the D8-brane solution of the massive IIA theory in [19].
In the orthonormal basis
e0 = H 1/2 dy,
e2 = H 1/2 dz2 ,
e1 = H 1/2(dz1 + mz3 dz2 ),

e3 = H 1/2 dz3 ,
(6)
it is easily verified that the 2-form,

J e0 e1 e2 e3 = dy (dz1 + mz3 dz2 ) H dz2 dz3 ,
(7)
is closed, and in fact covariantly constant. It is a privileged Khler form amongst the
2-sphere of complex structures.
j
If we define the holomorphic and antiholomorphic projectors Pi j = 12 (i iJi j ) and
j
Qi j = 12 (i + iJi j ), then complex coordinates must satisfy the differential equations
Qi j j = 0.
(8)
The integrability of these equations is assured from the fact that our metric is alrea dy
established to be Khler.
7 In this paper, we adopt the notation of [27,28], where the lower-dimensional maximal supergravities were
obtained by consecutive S 1 reduction with the indices i = 1, 2, . . . denoting the ith coordinate in the reduction.
Let x = y + 12 my 2 , so that H dy = dx. Then (8) can be shown to reduce to just the
following pair of independent equations:

+i
= 0,
x
z1

1
i
= 0.
+i
+ mz3
z3
z2
2
x
z1
(9)
Solutions of these differential equations define the complex coordinates in terms of the
real coordinates.
Any pair of independent solutions to the above equations gives a valid choice of
complex coordinates. A convenient choice is
1 = z3 + iz2 ,
i
1
2 = x + iz1 m z22 + z32 + mz2 z3 ,
4
2
(10)
implying that the metric becomes

2
ds42 = H |d1|2 + H 1 d2 + 12 m1 d1 ,
(11)
1/2

.
H = 1 + m(2 + 2 ) + 12 m2 |1 |2
(12)
with
This agrees, up to coordinate redefinitions, with results in [11].

The metric in (11) has the characteristic Hermitean form
ds 2 = 2g d d ,
(13)
and in fact
g =
2K
,

(14)
where the Khler function K given by K = 2H 3/(3m2 ).

The metric (4) with H = 1 + m|y| physically represents a domain wall located at y = 0.
This is constructed by patching two sides, each of which is part of a smooth but incomplete
metric in which H can instead be taken to have the form H = my. The metric (5) with
H = my has a scaling symmetry generated by the dilatation operator
D=y
+ 2z1
+ z2
+ z3
.
y
z1
z2
z3
(15)
As we shall discuss in Section 2.4, this is a homothetic Killing vector. In addition, (5) is
invariant under the linear action of U (1) on (z2 , z3 ), preserving the 2-form dz3 dz2 .
2.2. Domain-wall as Heisenberg contraction of EguchiHanson

As discussed in [11], the isometry group of the metric (5) is the Heisenberg group, and
it acts tri-holomorphically. In other words, it leaves invariant all three of the 2-spheres
worth of complex structures. The Heisenberg group may be obtained as the InnWigner
contraction of SU(2). It is not unreasonable, therefore, to expect to obtain (5) as a limit
of the EguchiHanson metric, which is the only complete and non-singular hyper-Khler
4-metric admitting a tri-holomorphic SU(2) action. One could consider the larger class of
triaxial metrics admitting a tri-holomorphic SU(2) action considered in [26], but our metric
is symmetric under the interchange of z2 and z3 , and so it is only necessary to consider the
biaxial case.
The EguchiHanson metric is

Q
Q 1 2 1 2
1
dr + r 1 + 4 32 + r 2 12 + 22 ,
ds42 = 1 + 4
(16)
4
4
r
r
where the i are the left-invariant 1-forms of SU(2), satisfying
d1 = 2 3 ,
d2 = 3 1 ,
d3 = 1 2 .
(17)
If we define rescaled 1-forms i , according to

1 = 1 ,
2 = 2 ,
3 = 2 3 ,
then after taking the limit 0, we find that (17)

d 1 = 0,
d 2 = 0,
(18)
becomes8
d 3 = 1 2 .
(19)
This is the same exterior algebra as in the 1-forms appearing in the domain-wall metric (5),
as can be seen by making the associations
1 = m dz2,
2 = m dz3 ,
3 = m(dz1 + mz3 dz2 ).
(20)
To see how the EguchiHanson metric (16) limits to the domain-wall solution, we
should combine the rescaling (18) with
r = 1 r ,

Q = 6 Q,
under which (16) becomes

1 2 1 2 4 Q

1
Q
ds42 = 4 + 4
d r + r + 4 32 + r 2 12 + 22 .
r
4
r
4
(21)
(22)
= 16m4 , H = 1 m2 r 2 and take the Heisenberg limit 0, we find,

If we now define Q
4
after making the association (20), that (22) becomes precisely (5) after a further coordinate
change y = 14 mr 2 .
The metric (16) has a curvature singularity at r = 0. If Q < 0, this is not part of the
EguchiHanson manifold, which includes only r Q. Intuitively, one may consider that
8 Heisenberg limits for more general groups are discussed in Appendix B.
10
the singularity is hidden behind a bolt in this case. If Q > 0, the singularity at r = 0 is
> 0.
naked. In order to take the limit 0 in the rescaled metric (22), we must let Q
The near-singularity behaviour of the resulting metric is similar to that of (16), but with the
SU(2) orbits flattened to Heisenberg orbits.
2.3. Heisenberg limit of the superpotential
One may take the Heisenberg limit at the level of the equations of motion. Thus the
4-dimensional metric
ds42 = (abc)2 d2 + a 2 12 + b2 22 + c2 32 ,
(23)
where a, b and c are functions of , will be Ricci-flat if log a, log b and log c
satisfy the equations of motion coming from the Lagrangian
L = + + V ,
(24)
with
V=

1 4
a + b4 + c4 2b2 c2 2c2 a 2 2a 2b2 .
4
(25)
A superpotential is given by
W = a 2 + b2 + c2 21 bc 22 ca 23 ab,
(26)
for any choice of the constants i that satisfy the three equations
1 = 2 3 ,
2 = 3 1 ,
3 = 1 2 .
(27)
If 1 = 2 = 3 = 0, we get the equations of motion for hyper-Khler metrics with

tri-holomorphic SU(2) action, solved in [26]. It is possible to rescale the variables a,
b and c to obtain a superpotential giving the equations of motion for metrics admitting
a tri-holomorphic action of the Heisenberg group. One sets
a = 1 a,
b = 1 b,
c = 2 c,
(28)
, giving
together with = 4 and W = 4 W
= c2 .
W
(29)
This gives rise to the first-order equations

d
d
d
=
=
= e2 ,
d
d
d
from which one can easily rederive the domain-wall solution (5).
(30)
11
2.4. Hypersurface-orthogonal homotheties

The large-radius behaviour of the EguchiHanson metric is that of a Ricci-flat cone over
RP3 [26]. In this limit, the metric becomes scale-invariant: scaling the metric by a constant
factor 2 is equivalent to performing the diffeomorphism r r. This transformation is
generated by the Euler vector
,
r
which satisfies
Er
(31)
E = g .
The vector
satisfy
(32)
is a special kind of conformal Killing vector
K ,
which would in general
K + K = 2fg .
(33)
If f is constant, then
generates a homothety, and if K = K then K is
a gradient, and hence it is hypersurface-orthogonal. The Euler vector E in (31) is an
example of such a hypersurface-orthogonal homothetic conformal Killing vector [30].
The Heisenberg limit of the EguchiHanson metric is also scale-invariant. In other
words, the metric near the singularity is scale-free. Thus it is rescaled by a constant under
the transformation
K
z1 2 z1 ,
z2 z2 ,
z3 z3 ,
y y,
(34)
where in this section we are taking H = my. This is generated by the homothetic Killing
vector
+ 2z1
+ z2
+ z3
.
D=y
(35)
y
z1
z2
z3
In contrast to the Euler vector (31) for the cone, the homothety (35) is neither a gradient
nor is it proportional to a gradient, and so it is not hypersurface-orthogonal.
The existence of the homothety generated by (34) depends crucially on the fact that the
Heisenberg algebra, represented in the CartanMaurer form in (19), is invariant under the
scaling
1 1 ,
2 2 ,
3 2 3 .
(36)
By contrast, the original SU(2) algebra, represented in (17), is of course not invariant under
(36).
In [30], it was shown that (5) arises as the large-distance limit of a non-compact Calabi
Yau 3-fold. In the large-distance limit, the metric becomes scale-invariant.
2.5. Multi-instanton construction
The four-dimensional metric (5) can be related to the general class of multi-instantons
obtained in [31], which are constructed as follows. Let x i denote Cartesian coordinates on
12
R3 . We the write the metric

2
ds42 = V 1 d + Ai dx i + V dx i dx i ,
(37)
where V and Ai depend only on the x i . In the orthonormal basis

ei = V 1/2 dx i ,
e0 = V 1/2 d + Ai dx i ,
(38)
the spin connection is then given by

1
0i = V 3/2 i V e0 + Fij ej ,
2

1
ij = V 3/2 j V ei i V ej Fij e0 ,
(39)
2
0i of
where we have defined Fij i Aj j Ai . It is convenient to introduce a dual
0i 12 :ij k j k . It is easily seen that if the spin
the spin-connection components ij , as
connection is self-dual or antiself-dual, in the sense that
0i = 0i , then the curvature
2-forms are self-dual or antiself-dual, not only in the analogous sense 0i = 12 :ij k j k ,
but also in the normal sense ab = ab . In particular, when this condition is satisfied,
the metric is Ricci flat.
It is easy to see from (39) that the spin-connection satisfies
0i = 0i if and only if
the metric functions satisfy
=
A,

V
(40)
where the expressions here are the standard ones of three-dimensional Cartesian coordinates. In other words, i V = :ij k j Ak . Thus (40) is the condition for (37) to be Ricci
flat, and furthermore, self-dual or anti-self-dual. In particular, taking the divergence of (40)
we get 2 V = 0, so V should be harmonic, and then A can be solved for (modulo a gauge
transformation) using (40).
The multi-centre instantons [31] are obtained by taking
q
,
V =c+
(41)
|
x x |
where c and q are constants. If c = 0 one gets the multi TaubNUT metrics, while if c = 0
(conveniently one chooses c = 1), the metrics are instead multi EguchiHanson.
If we take a uniform distribution of charges spread over a two-dimensional plane of
radius R, then at a perpendicular distance y from the centre of the disc the potential V is
given by
R
V =c+q
(y 2
0

r dr
+ r 2 )1/2
= c + q R2 + y 2
1/2

|y| = c + qR q|y| + O(1/R).
(42)
Thus if we send R to infinity, while setting q = m and adjusting c such that c + qR = 1,

we obtain
V = 1 + m|y|.
(43)
13
It is easy to establish from (40) that a solution for A is then

A = (0, mz3 , 0)
(44)
(where we take the Cartesian coordinates to be x = (y, z2 , z3 )), and so we have arrived
back at our original metric (5). It can therefore be described as a continuum of Taub
NUT instantons distributed uniformly over a two-dimensional plane. (This is essentially
the construction of [19].)
A more physical picture of this limit is that there can be multi-instanton generalisations
of an AC manifold, and a uniform distribution would turn the non-Abelian isometry group
into an Abelian U (1) group.
3. Orientifold planes and the AtiyahHitchin metric

In addition to D-branes, which have positive tension, string theory admits orientifold
planes which have negative tension. Since orientifold planes are not dynamical, the
negative tension does not lead to instabilities as it would if the tension of an ordinary
D-brane were negative. This is because they are pinned in position: the inversion symmetry
employed in the orientifold projection excludes translational zero modes.
In M-theory, an orientifold plane corresponds to the product of the AtiyahHitchin
metric [6,20] with seven-dimensional Minkowski spacetime [21]. The AtiyahHitchin
metric is a smooth non-singular hyper-Khler 4-metric and hence BPS. It is invariant under
SO(3) acting on principal orbits of the form SO(3)/Z2 , where the Z2 is realised as the
group of diagonal SO(3) matrices. Near infinity, it is given approximately by the Taub
NUT metric divided by CP, taken with a negative ADM mass. The CP quotient symmetry
here takes (, x i ) to (, x i ), where is the KaluzaKlein coordinate. As in string
theories, the negative mass does not lead to instabilities because this quotient symmetry of
the asymptotic metric eliminates translational zero modes.
One might think that being BPS and having negative ADM mass would be inconsistent
because of the Positive Mass Theorem. However, the Positive Mass Theorem for ALF
spaces such as the AtiyahHitchin metric is rather subtle [22]. Suffice it to say that one
needs to solve the Dirac equation subject to boundary conditions at infinity as an essential
ingredient in the proof. If the manifold is simply connected there is a unique spin structure
but because a neighbourhood of infinity where one imposes the boundary conditions is not
simply-connected, it is not obvious that a suitable global solution exists in the unique spin
structure. In the AtiyahHitchin case it seems clear that it does not.
This example shows that in principle, gravity can resolve singularities associated with
branes of negative tension. However, to make this more precise we would need to consider
the unresolved spacetime and its relationship to the AtiyahHitchin metric. To lowest order,
one might consider this to be the flat metric on S 1 R3 with coordinates (, x). This
clearly has a singularity at (0, 0, 0, 0). However, this approximation ignores the Kaluza
Klein magnetic field generated by the orientifold plane. If we maintain spherical symmetry
we would be led at the next level of approximation to the TaubNUT metric with negative
14
mass:

2M 1
2M 2
2
dr + r 2 d 2 + sin2 d 2 .
(d + cos d) + 1 +
ds = 4 1 +
r
r
(45)
CP acts as (, , ) (, , + ) and the ADM mass M is negative. Clearly the
TaubNUT approximation breaks down at small positive r because if r < 2M, the metric
signature is rather than + + + +.
The full non-singular AtiyahHitchin metric can be written as
2
ds 2 = dt 2 + a 2 (t)12 + b2(t)22 + c2 (t)32 ,
(46)
where 1 , 2 , 3 are CartanMaurer forms for SU(2) and the allowed range of angular
coordinates is restricted by the fact that CP should act as the identity. For large t we have
a b and the metric tends to the TaubNUT metric.
The AtiyahHitchin metric and TaubNUT metric are members of a general family of
locally SU(2)-invariant hyper-Khler metrics in which the three Khler forms transform as
a triplet. They satisfy a set of first order differential equations coming from a superpotential
as described in Section 2.3. The EguchiHanson metric discussed above is a member
of another family of locally SU(2)-invariant hyper-Khler metrics in which the three
Khler forms transform as singlets.9 They satisfy a different set of first order differential
equations also given in Section 2.3. One may check that the Heisenberg limits of these
two sets of equations are identical and the solutions are precisely the metrics (5) in which
the Heisenberg group acts tri-holomorphically. Thus in the Heisenberg limit the triplet
becomes three singlets.
One may also take the Heisenberg limit directly in the asymptotic TaubNUT metric.
This also leads to the metric (5). However, in order to build in the projection under CP one
may identify (z1 , z2 , z3 ) with (z1 , z2 , z3 ). Thus by analogy with the construction of
Section 2.5 we propose that the singular unresolved metric for a stack of orientifold planes
analogous to the metric of a stack of D6-branes is (5) with this additional identification.
4. Domain-walls from pure gravity

The eight-dimensional domain-wall example of the previous section lifted to a
solution of eleven-dimensional supergravity with vanishing CJS field F(4) . Its transverse
coordinate y, together with the three toroidal coordinates (z1 , z2 , z3 ) of the reduction from
D = 11, gave the four coordinates of the cohomogeneity one Ricci-flat Heisenberg metric,
which could be viewed as a limit of the EguchiHanson metric. The principal orbits in the
Heisenberg limit were T 1 bundles over T 2 .
In this section we shall generalise this construction by considering domain-wall
solutions in maximal supergravities that lift to give purely geometrical solutions in eleven
dimensions. Thus if we begin with such a domain-wall in D-dimensional supergravity, the
9 One says that SU(2) acts triholomorphically in this case.
15
2 = dx dx + ds 2
2
metric after lifting to eleven dimensions will be ds11
12D , where ds12D

is a Ricci-flat metric of cohomogeneity one, with principal orbits that are again of the form
of torus bundles.
The cases that we shall consider here arise from domain-wall solutions in D = 7, 6
and 5. Correspondingly, these give rise to Ricci-flat Heisenberg metrics of dimensions 6,
7 and 8. Since each domain wall preserves a fraction of the supersymmetry, it follows that
the associated Ricci-flat metrics admit certain numbers of covariantly-constant spinors.
In other words, they are metrics of special holonomy. This property generalises the
special holonomy of the 4-dimensional Ricci-flat Khler metric in the previous example
in Section 2.1. Specifically, for the Ricci-flat metrics in dimension 6, 7 and 8 we shall see
that the special holonomies SU(3), G2 and Spin(7) arise.
4.1. Six-dimensional manifolds with SU(3) holonomy

4.1.1. T 1 bundle over T 4
In this first six-dimensional example, the principal orbits form a T 1 bundle over T 4 . It
arises if we take a domain wall solution in six spacetime dimensions, supported by the two
1
1
0-form field strengths F(0)23
and F(0)45
, carrying equal charges m. In what follows, we
shall adhere to the terminology charges, although in some circumstances it may be more
appropriate to think of fluxes. The domain-wall metric is
ds 2 = H 1/2 dx dx + H 5/2 dy 2,
(47)
where H = 1 + m|y|. Oxidising back to D = 11, we get the eleven-dimensional metric

2
d s2 = dx dx + H 2 dz1 + m(z3 dz2 + z5 dz4 ) + H 2 dy 2

+ H dz22 + + dz52 .
(48)
Thus we conclude that the six-dimensional metric

2

ds62 = H 2 dz1 + m(z3 dz2 + z5 dz4 ) + H 2 dy 2 + H dz22 + + dz52
(49)
is Ricci flat. Since the solution carries two charges it preserves 1/4 of the supersymmetry,
and so this 6-metric must have SU(3) holonomy. Thus it is a Ricci-flat Khler 6-metric.
Define an orthonormal basis by

e0 = H dy,
e1 = H 1 dz1 + m(z3 dz2 + z5 dz4 ) ,
e2 = H 1/2 dz2 ,
e3 = H 1/2 dz3 ,
e4 = H 1/2 dz4 ,
e5 = H 1/2 dz5 .
(50)
It can then be seen that the Khler form is given by

J = e0 e1 e2 e3 e4 e5 .
(51)
Following the same strategy as in the previous section, we can obtain the differential
equations whose solutions define complex coordinates in terms of the real coordinates.
First, define a new real coordinate x in place of y, such that H 2 dy dx, and hence
y + my 2 + 13 m2 y 3 = x. This implies that H = (1 + 3mx)1/3 . After straightforward algebra,
16
we find that a suitable choice for the definition of the complex coordinates is
1 = z3 + iz2 ,
2 = z5 + iz4 ,

1 2
3 = x + iz1 + m z2 + z42 3z32 3z52 2iz2 z3 2iz4 z5 .
4
The metric then takes the form

2

ds62 = H |d1 |2 + |d2 |2 + H 2 d3 + 12 m(1 d1 + 2 d2 ) ,
the harmonic function H is given by

1/3
H = 1 + 32 m(3 + 3 ) + 34 m2 |1 |2 + |2 |2
,
(52)
(53)
(54)
and the Khler function is K = H 4 /m2 .

The one-forms ( 1 = dz1 + m(z1 dz2 + z3 dz4 ), dz2 , dz3 , dz4 , dz5 ) may be regarded as
left-invariant one-forms on the generalized 5-dimensional Heisenberg group whose nontrivial exterior algebraic relation is
d 1 = m(dz3 dz2 + dz5 dz4 ).
(55)
In the notation of Appendix B.1 we thus have

1 1
2 F
dx dx = m(dz3 dz2 + dz5 dz4 ),
(56)
with the corresponding Lie algebra given by (B.1).

In the case that H = my, there is a scaling invariance of the metric (49) generated by
the homothetic Killing vector
D=y
3
3
3
3
+ 3z1
+ z2
+ z3
+ z4
+ z5
.
y
z1 2 z2 2 z3 2 z4 2 z5
(57)
In addition, (49) is invariant under the linear action of U (2) on (z2 , z3 , z4 , z5 ) preserving
the self-dual 2-form dz3 dz2 + dz5 dz4 .
There is a second type of 2-charge domain wall in D = 6, again supported by two
0-form field strengths coming from the KaluzaKlein reduction of the eleven-dimensional
1
2
metric. A representative example is given by using the field strengths (F(0)34
, F(0)35
). The
domain-wall metric is again given by (47), but now, upon oxidation back to D = 11, we
obtain d s2 = dx dx + ds62 with the Ricci-flat 6-metric now given by
ds62 = H 2 dy 2 + H 1 (dz1 + mz4 dz3 )2 + H 1 (dz2 + mz5 dz3 )2

+ H 2 dz32 + H dz42 + dz52 .
(58)
Defining the orthonormal basis

e0 = H dy,
e3 = H dz3 ,
e1 = H 1/2 (dz1 + mz4 dz3 ),

e4 = H 1/2 dz4 ,
e2 = H 1/2 (dz2 + mz5 dz3 ),
e5 = H 1/2 dz5 ,
(59)
17
we find that the torsion-free spin connection is given by

01 = e1 ,
02 = e2 ,
03 = 2e3 ,
04 = e4 ,
05 = e5 ,
12 = 0,
13 = e ,
14 = e ,
15 = 0,
23 = e ,
24 = 0,
25 = e3 ,
34 = e1 ,
35 = e2 ,
45 = 0,
4
5
(60)
where 12 mH 2 .
From this, it is easily established that the following 2-form is covariantly constant:
J = e0 e3 + e1 e4 + e2 e5 .
(61)
This is the Khler form. From this, using the same strategy as we used in previous sections,
we can deduce that the following are a suitable set of complex coordinates:
1 = z1 + iH z4 ,
2 = z2 + iH z5 ,
3 = y + iz3 .
(62)
The Hermitean metric tensor g can then be derived from the Khler function
K =
2

1
1
(1 1 )2 + (2 2 )2 + |3 |2 8 H 2 + H + 1 m2 |3 |2 .
4H
48
(63)
To display the Heisenberg algebra for the metric (58), we set 1 = dz1 + mz4 dz3 ,
= dz2 + mz5 dz3 and obtain the algebra (B.1) with
1 1
1 2
F = m dz4 dz3 ,
F = m dz5 dz3 .
(64)
2
2
D=y
5
3
3
+ z1
+ z2
+ z3
+ z4
+ z5
.
y 2 z1 2 z2
z3 2 z4 2 z5
(65)
In addition, (58) is invariant under the linear action of SO(2) on (z1 , z2 ) and (z4 , z5 ).
4.2. Seven-dimensional manifolds with G2 holonomy
Now consider a 4-charge domain wall in D = 5. Take the charges to be carried by the
1
1
2
2
following 0-form field strengths: (F(0)34
, F(0)56
, F(0)35
, F(0)46
). Note that here, unlike the
2-charge cases in D = 6, it matters what the relative signs of the charges are here, in order
to get a supersymmetric solution. Specifically, the Bogomolnyi matrix M is given here
by [27]
M = 1 + q1 y134 + q2 y156 + q3 y235 + q4 y246,
where
=
|qi |.
(66)
(67)
18
Having fixed a set of conventions, the signs of the first three charges (q1 , q2 , q3 ) can
be arbitrary for supersymmetry, but only for one sign of the fourth charge q4 is there
supersymmetry. With our conventions, it must be negative. Thus we may consider:
q1 = q2 = q3 = q4 = m.
(68)
The domain-wall metric in D = 5 is

ds 2 = H 4/3 dx dx + H 16/3 dy 2 .
(69)
=
Oxidising back to D = 11 in the standard way, we get
+
seven-dimensional metric is given by

2
ds72 = H 4 dy 2 + H 2 dz1 + m(z4 dz3 + z6 dz5 )
2

+ H 2 dz2 + m(z5 dz3 z6 dz4 ) + H 2 dz32 + + dz62 .
d s2
dx dx
ds72 ,
where the
(70)
Note that the minus sign in the term involving z6 dz5 is a reflection of the fact that
2
the charge associated with F(0)46
is negative for supersymmetry. On account of the
supersymmetry, we conclude that the Ricci-flat metric ds72 admits one covariantly-constant
spinor, and thus it must have G2 holonomy.
Define the orthonormal basis
e3 = H dz3 ,
e4 = H dz4 ,
e5 = H dz5 ,
e6 = H dz6 ,
e0 = H 2 dy,

e2 = H 1 dz2 + m(z5 dz3 z6 dz4 ) .
e1 = H 1 dz1 + m(z4 dz3 + z6 dz5 ) ,
(71)
In this basis, the torsion-free spin connection is given by
01 = 2e1 ,
02 = 2e2 ,
03 = 2e3 ,
04 = 2e4 ,
05 = 2e2,
06 = 2e6 ,
12 = 0,
13 = e4 ,
14 = e3 ,
15 = e6 ,
16 = e5 ,
23 = e5 ,
24 = e6 ,
25 = e3 ,
26 = e4 ,
34 = e1 ,
35 = e2 ,
45 = 0,
46 = e ,
36 = 0,
2
56 = e1 ,
(72)
where 12 mH 3 .
It can now be verified that the following 3-form is covariantly constant:
(3) e0 e1 e2 e1 e4 e6 + e1 e3 e5
e2 e5 e6 e2 e3 e4 e0 e3 6 e0 e4 e5 .
(73)
The existence of such a 3-form is characteristic of a 7-manifold with G2 holonomy. In fact

the components ij k are the structure constants of the multiplication table of the seven
imaginary unit octonions i :
i j = ij + ij k k .
(74)
19
Note that if the sign of the z6 dz4 term in (70) had been taken to be + instead of(while
keeping all other conventions unchanged), then there would not exist a covariantly-constant
3-form. This is another reflection of the fact that the occurrence of supersymmetry is
dependent up on sign of the fourth charge. (At the same time as the 8-dimensional spinor
representation of the SO(7) tangent-space group decomposes as 8 7 + 1 under G2 , the
35-dimensional antisymmetric 3-index representation decomposes as 35 27 + 7 + 1.
It is the singlet in each case that corresponds to the covariantly-constant spinor (Killing
spinor) and 3-form.)
Note that we can write the 3-form (3) as
(3) = e0 e1 e2 e0 K0 e1 K1 e2 K2 ,
(75)
where
K0 e3 e6 + e4 e5 ,
K1 e4 e6 e3 e5 ,
K2 e5 e6 + e3 e4 .
(76)
The three 2-forms K0 , K1 and K2 are self-dual with respect to the metric in the (3, 4, 5, 6)
directions. Thus in the entire construction, both of the 7-dimensional metric ds72 and the
covariantly-constant 3-form (3) , the flat 4-torus metric dz32 + + dz62 can be replaced
by any hyper-Khler 4-metric. (The potential terms that twist the fibres in the z1 and
z2 directions are now replaced by potentials for the self-dual 2-forms K2 and K1 . See
Section 6.2 below.)
To display the Heisenberg algebra for the metric (70), set 1 = dz1 +m(z4 dz3 +z6 dz5 ),
2
= dz2 + m(z5 dz3 z6 dz4 ) and obtain the algebra (B.1) with
1 1
F = m(dz4 dz3 + dz6 dz5 ),
2
1 2
F = m(dz5 dz3 dz6 dz4 ).
(77)
2
D=y
+ 4z1
+ 4z2
+ 2z3
+ 2z4
+ 2z5
+ 2z6
.
y
z1
z2
z3
z4
z5
z6
(78)
In addition, (70) is invariant under the linear action of SU(2) on (z3 , z4 , z5 , z6 ) that
preserves the two 2-forms dz4 dz3 + dz6 dz5 and dz5 dz3 dz6 dz4 . The
6-dimensional nilpotent algebra in this case is the complexification of the standard
3-dimensional ur-Heisenberg algebra.
There is an inequivalent class of domain-wall solutions in five-dimensional spacetime,
3
1
2
, F(0)46
, F(0)45
}.
for which a representative example is supported by the three fields {F(0)56
This gives the Ricci-flat 7-metric
ds72 = H 3 dy 2 + H 1 (dz1 + mz6 dz5 )2 + H 1 (dz2 mz6 dz4 )2

+ H 1 (dz3 + mz5 dz4 )2 + H 2 dz42 + dz52 + dz62 .
(79)
20
In the obvious orthonormal basis e0 = H 3/2 dy, e2 = H 1/2(dz1 + mz6 dz5 ), etc., the
covariantly-constant associative 3-form is given by
(3) = e0 e1 e4 + e0 e2 e5 + e0 e3 e6
+ e1 e2 e6 + e2 e3 e4 + e3 e1 e5 e4 e5 e6 .
(80)
To display the Heisenberg algebra for the metric (79), set 1 = dz1 + mz6 dz5 , 2 =
dz2 mz6 dz4 , 3 = dz3 + mz5 dz4 and obtain the algebra (B.1) with
1 1
F = m dz6 dz5 ,
2
1 2
F = +m dz6 dz4 ,
2
1 3
F = m dz5 dz4 .
(81)
2
D=y
3
3
3
+ 3z1
+ 3z2
+ 3z3
+ z4
+ z5
+ z6
.
y
z1
z2
z3 2 z4 2 z5 2 z6
(82)
In addition, (79) is invariant under SO(3) acting linearly on (z1 , z2 , z3 ) and (z3 , z4 , z5 , z6 ).
4.3. Eight-dimensional manifolds with Spin(7) holonomy
Now consider a 6-charge domain wall solution in D = 4, supported by the 0-form field
3
3
1
1
2
2
strengths (F(0)45
, F(0)67
, F(0)46
, F(0)57
, F(0)47
, F(0)56
). As in the previous case, the signs
of the charges must be appropriately chosen in order to have a supersymmetric solution. In
D = 4, the domain wall metric, with all charges chosen equal in magnitude, is
ds 2 = H 3 dx dx + H 9 dy 2 .
(83)
Oxidising back to D = 11 gives the eleven-dimensional metric d s2 = dx dx + ds82 ,

where the Ricci-flat 8-metric is given by
2

ds82 = H 6 dy 2 + H 2 dz1 + m(z5 dz4 + z7 dz6 )

2
+ H 2 dz2 + m(z6 dz4 z7 dz5 )

2
+ H 2 dz3 + m(z7 dz4 + z6 dz5 ) + H 3 dz42 + + dz72 .
(84)
1
Since the solution preserves 1 16
of the supersymmetry, it follows that this Ricci-flat
8-metric must have Spin(7) holonomy.
Let us choose the natural orthonormal basis,

e0 = H 3 dy,
e1 = H 1 dz1 + m(z5 dz4 + z7 dz6 ) ,

e3 = H 1 dz3 + m(z7 dz4 + z6 dz5 ) ,
e2 = H 1 dz2 + m(z6 dz4 z7 dz5 ) ,
e4 = H 3/2 dz4 ,
e5 = H 3/2 dz5 ,
e6 = H 3/2 dz6 ,
e7 = H 3/2 dz7 .
(85)
21
The spin connection is then given by ij = 12 (cij k +cikj ckj i )ek , where the non-vanishing
connection coefficients cij k = cj ik are specified by
c01 1 = c45 1 = c67 1 = c02 2 = c46 2 = c57 2 = c03 3 = c47 3 = c56 3 = 2,
c04 4 = c05 5 = c06 6 = c07 7 = 3,
(86)
and 12 mH 4 here.
From this, it is straightforward to show that the following 4-form is covariantly constant:

(4) = e0 e1 + e2 e3 e4 e5 + e6 e7

e0 e2 e1 e3 e4 e6 e5 e7

e0 e3 + e1 e2 e4 e7 + e5 e6
+ e0 e1 e2 e3 + e4 e5 e6 e7 .
(87)
The existence of this 4-form, which is self-dual, is characteristic of 8-manifolds with

Spin(7) holonomy.
To display the Heisenberg algebra for the metric (84), set 1 = dz1 +m(z5 dz4 +z7 dz6 ),
2
= dz2 + m(z6 dz4 z7 dz5 ), 3 = dz3 + m(z7 dz4 + z6 dz5 ) and obtain the algebra (B.1)
with
1 1
F = m(dz5 dz4 + dz7 dz6 ),
2
1 2
F = m(dz6 dz4 dz7 dz5 ),
2
1 3
F = m(dz7 dz4 + dz6 dz5 ).
2
(88)
5
5
+ 5z1
+ 5z2
+ 5z3
+ z4
+ z5
y
z1
z2
z3 2 z4 2 z5
5
5
+ z6
+ z7
.
2 z6 2 z7
D=y
(89)
In addition, (84) is invariant under the linear action of SU(2) on (z4 , z5 , z6 , z7 ) that
preserves the three 2-forms dz5 dz4 + dz7 dz6 , dz6 dz4 dz7 dz5 and dz7
dz4 + dz6 dz5 .
4.4. Further examples and specialisations
The various domain walls that we have obtained above are the most natural ones to
consider, since they possess the maximum number of charges in each case, and after lifting
to D = 11 they are irreducible. It is, nevertheless, of interest also to study some of the other
possible examples.
22

In seven-dimensional maximal supergravity, the largest number of allowed charges for
domain walls is 1 (see, for example, [24]). The metric is given by
ds72 = H 1/5 dx dx + H 6/5 dy 2,
(90)
2 = dx dx + ds 2 , where ds 2
where H = 1 + m|y|. After lifting to D = 11, this gives ds11
5
5
is the Ricci-flat metric

ds52 = H dy 2 + H 1 (dz1 + mz3 dz2 )2 + H dz22 + dz32 + dz42 .
(91)
This is clearly reducible, being nothing but the direct sum of the four-dimensional Ricciflat metric (5) and a circle. It is for this reason that we omitted this 5-dimensional example
in our enumeration above. It can be viewed as a T 1 bundle over T 3 , but since the bundle is
trivial over a T 1 factor in the base, it would be more accurate to describe it as T 1 times a
T 1 bundle over T 2 .
4.4.2. Ricci-flat metrics with fewer charges
There are many possibilities for obtaining other Ricci-flat metrics, by turning on only
subsets of the charges in the metrics we have alrea dy obtained. We shall illustrate this by
considering the example of the 8-dimensional metric (84). If we introduce parameters :i ,
where :i = 1 if the ith of the six charges listed above (83) is turned on, an :i = 0 if the ith
charge is turned off. After lifting the resulting domain wall from D = 4 to D = 11, we get
a Ricci-flat 8-metric given by
ds82 = H
i :i
dy 2 + H :1 :2 h21 + H :3 :4 h22 + H :5 :6 h23
+ H :1 +:3 +:5 h24 + H :1 +:4 +:6 h25 + H :2 +:3 +:6 h26 + H :2 +:4 +:5 h27 ,
h1 = dz1 + :1 z5 dz4 + :2 z7 dz6 ,
h2 = dz2 + :3 z6 dz4 :4 z7 dz5 ,
h3 = dz3 + :5 z7 dz4 + :6 z6 dz5 ,
h4 = dz4 ,
h6 = dz6 ,
(92)
h5 = dz5 ,
h7 = dz7 .
(93)
4.4.3. SU(4) holonomy in D = 8

There are other possibilities, which involve a lesser number of charges which are not
themselves a subset of the maximal set. For example, we can consider the following
8-dimensional Ricci-flat metric that comes from lifting a 3-charge four-dimensional
1
1
1
domain wall, supported by the fields F(0)23
, F(0)45
, F(0)67
. This gives

ds82 = H 3 dy 2 + H 3 (dz1 + z3 dz2 + z5 dz4 + z7 dz6 )2 + H dz22 + + dz72 .
(94)
This metric has SU(4) holonomy, and it can be viewed as a Heisenberg limit of a complex
line bundle over a six-dimensional EinsteinKhler space such as S 2 S 2 S 2 , or CP3 .
23
5. Heisenberg limits of complete metrics of special holonomy

In this section, we generalise the discussion of the Heisenberg limit of the Eguchi
Hanson metric that we gave in Section 2.2, and show how the various Ricci-flat metrics
that we obtained from domain-wall solutions in Section 4 can be viewed as arising as
Heisenberg limits of complete metrics of special holonomy.
5.1. Contractions of Ricci-flat Khler 6-metrics
5.1.1. Contractions of T 1 bundles over EinsteinKhler
The contraction to the Heisenberg limit of the EguchiHanson metric was discussed in
Section 2.2 at the level of the metric itself, and in Section 2.3 at the level of the equations of
motion and superpotential. This contraction procedure can be easily generalised to higher
dimensions. In particular, we may obtain the six-dimensional Ricci-flat Heisenberg metric
(49) as a contraction of a Ricci-flat metric on a complex line bundle over an Einstein
Khler 4-metric with positive scalar curvature, such as CP2 . If we consider the more
general case of a line bundle over CPn , the starting point will be the metric
2
ds2n+2
= dt 2 + a 2 + c2 2 ,
(95)
where the left-invariant 1-forms of SU(n + 1) are defined in Apendix B.2.4. The conditions
for Ricci-flatness for the line bundle over CPn then follow from the Lagrangian L = T V ,
where
T = 2 + (2n 1) ,
2
V = a 4n4 c4 + 2(n + 1)a 4n2c2 ,
(96)
d = dt/(a 2n c), a = e and b = e . The superpotential W is given by

n + 1 2n
a .
n
The scalings (B.33) induce the following scalings in the metric coefficients:
W = a 2n2 c2 +
a 1 a,
c 2 c.
(97)
(98)
After sending to zero, the rescaled superpotential becomes

W = a 2n2 c2 .
(99)
Solving the resulting first-order equations gives

a t 1/(n+1) ,
c t n/(n+1) .
(100)
In particular, for n = 1 we recover the 4-dimensional Heisenberg metric (5), and for n = 2
we recover the 6-dimensional metric (49).
It should be remarked that we could in fact obtain the same Heisenberg contractions
if the CPn metrics are replaced by any other (2n)-dimensional homogeneous Einstein
Khler metrics of positive scalar curvature. In Section 6.1, we shall give a version of this
construction for inhomogeneous EinsteinKhler manifolds.
24
5.1.2. Contraction of T S n+1

Starting from the left-invariant SO(n + 2) 1-forms in the notation of (B.9), the ansatz
that gives rise to the Stenzel [32] metrics on T S n+1 is [33]
ds 2 = dt 2 + a 2 i2 + b2 i2 + c2 2 .
(101)
The Ricci-flat equations can be derived from the Lagrangian L = T V with

2
1
2
T = + + n + (n 1) + ,
2

1
2n2 4
4
a + b + c4 2a 2 b2 2n a 2 + b2 c2 ,
V = (ab)
4
and V can be obtained from the superpotential [33]
(102)

1
W = (ab)n1 a 2 + b2 + c2 .
(103)
2
Solutions of the associated first-order equations give the Ricci-flat Khler Stenzel metrics
on T S n+1 [33].
After applying the scalings (B.14), which imply (a, b, c) (2 a, 1 b, 1 c), and
then sending to zero, we obtain the superpotential
1
W = a n+1 bn1 .
2
This leads to the first-order equations [33]
(104)
a
na
a2
,
b = ,
c =
,
(105)
2bc
2c
2b
and after defining a new radial variable by dt = 2bc d, we obtain the Ricci-flat Heisenberg
metric
a =
ds 2 = 2n+2 d 2 +
1 2
+ i2 + n 2 ,
i
(106)
where the left-invariant 1-forms satisfy the exterior algebra (B.15). Setting n = 2, it is
easily seen after a coordinate transformation that we reproduce the Ricci-flat metric (58).
5.2. Contractions of 7-metrics of G2 holonomy
In the present section, we shall show how the two seven-dimensional Heisenberg
metrics (70) and (79) can be obtained as contraction limits of complete G2 metrics of
cohomogeneity one. Later, in Section 6.2, we shall give a version of this construction using
inhomogeneous hyper-Khler 4-metrics.
5.2.1. Contraction of R3 bundle over S 4
The complete metric of G2 holonomy is [3,4]

1
Q
Q 1 2
dr + r 2 1 + 4 R12 + R22 + r 2 P2 ,
ds72 = 1 + 4
2
r
r
(107)
25
where R1 , R2 and are given in terms of the left-invariant 1-forms of SO(5) in

Appendix B.2.3. After implementing the rescalings given in (B.29), together with
r 2 r,
Q 12 Q,
(108)
then after sending to zero we get the Heisenberg metric

1
r4 2 Q 2
(109)
dr + 2 1 + 22 + r 2 2 ,
Q
r
2
where 1 , 2 and satisfy the contracted algebra given in (B.30). After a coordinate
transformation, this can be seen to be equivalent to the Heisenberg metric (70).
ds72 =
5.2.2. Contraction of R4 bundle over S 3

The complete metric of G2 holonomy is [3,4]

Q 1 2 1 2
1
Q 2
2
ds7 = 1 + 3
(110)
dr + r 1 + 3 i + r 2 i2 ,
r
9
r
12
in the notation of Appendix B.2.2. After taking the scaling limit (B.19), together with
r 1 r,
Q 5 Q,
(111)
and then sending to zero, we obtain the following Heisenberg limit of the metric (110):
r3 2 Q 2
1
(112)
dr + i + r 2 i2 ,
Q
9r
12
where i and i now satisfy the contracted exterior algebra given in (B.21). After a
coordinate transformation, this can be seen to be equivalent to the Heisenberg metric (79).
ds72 =
5.3. Contraction of 8-metric of Spin(7) holonomy

Here, we shall show how the eight-dimensional Heisenberg metric (84) can be obtained
as a contraction limits of a complete Spin(7) metric of cohomogeneity one. Later, in
Section 6.3, we shall give a version of this construction using inhomogeneous hyper-Khler
4-metrics.
The complete metric of Spin(7) holonomy is [3,4]

Q 1 2 9r 2
9r 2 2
Q
P ,
dr +
1 + 10/3 Ri2 +
ds82 = 1 + 10/3
(113)
r
100
r
20
where Li and P are given in terms of the left-invariant 1-forms of SO(5) in Appendix B.2.3. Implementing the scalings in (B.27), together with
r 2 r,
Q 32/3 Q,
(114)
then after sending to zero the metric (113) becomes

r 10/3 2
9Q
9r 2 2
2
(115)
dr +
,
+
i
Q
100r 4/3
20
where i and now satisfy the contracted algebra (B.28). After a coordinate transformation, this can be seen to be equivalent to the D = 8 Heisenberg metric (84).
ds82 =
26
6. More general constructions of special-holonomy manifolds in 6, 7 and 8

dimensions
It is clear from the structure of the Ricci-flat Heisenberg metrics in dimensions 6, 7 and
8 in Section 4 that in each case where the principal orbits are torus bundles over T 4 , this
4-torus can itself be replaced by an arbitrary Ricci-flat Khler 4-metric. In other words,
we can allow the 4-manifold to be any hyper-Khler space. Such a space admits a triplet
of covariantly-constant 2-forms J a , which satisfy the multiplication rules of the imaginary
unit quaternions:
Jija Jjbk = ab ij + :abc Jikc .
(116)
In this section, we shall consider this more general construction in each of the dimensions
6, 7 and 8.
6.1. 6-metric of SU(3) holonomy from T 1 bundle over hyper-Khler
6.1.1. Description in real coordinates
Let ds42 be a hyper-Khler 4-metric, and then consider the following 6-metric:
d s62 = H 2 dy 2 + H 2 (dz1 + A(1) )2 + H ds42 ,
(117)
where H = y, and dA(1) = J , a Khler form on ds42 (we take m = 1 here). In the
orthonormal frame
e0 = H dy,
e 1 = H 1 (dz1 + A(1)),
ei = H 1/2ei ,
(118)
where ei is an orthonormal frame for ds42 , we find that the spin connection is given by
01 = H 2 e1 ,
1
0i = H 2 e i ,
2
1
1i = H 2 Jij ej ,
2
1
ij = ij H 2 Jij e1 ,
2
(119)
where ij is the spin connection for ds42 .

From this, it follows that the curvature 2-forms are given by
3
01 = 3H 4 e 0 e1 + H 4 Jij ei e j ,
2
3 4 0
3 4
i
0i = H e e + H Jij e1 ej ,
4
4
3 4 1
3
1i = H e ei H 4 e 0 ej ,
4
4
1 4
ij = ij H (ik j E + Jik Jj E + Jij JkE )ek eE ,
(120)
where ij is the curvature 2-form for ds42 . From these, we can read off that the Ricci tensor
vanishes.
27
The Khler form for the 6-dimensional metric is given by

J = e0 e1 + HJ.
(121)

The Lorentz-covariant exterior derivative D acting on a spinor is given by
d + 1 AB AB
D
4

1
1
1 2
= D + H
(122)
01 Jij ij e1 H 2 0i J1j 1j ej ,
2
4
4
where D d + 14 ij ij is the Lorentz-covariant exterior derivative on ds42 (except that the
gamma matrices are the six-dimensional ones).
that a Killing spinor must satisfy the conditions
It follows from this expression for D
D = 0,
0i = Jij 1j .
(123)
6.1.2. Description in complex coordinates

The above discussion made use of real coordinates on the six-dimensional Ricci-flat
Khler manifold. The structure of the metric in the complex notation (53), and of the
Khler potential in the form (54), suggest the natural generalisation for the construction
in a complex notation. Thus we are led to the following:
Let ds 2 be a Ricci-flat Khler metric of complex dimension n, with Khler function K,
Then
and Khler form J = i K.
d s2 = H ds 2 + H n |dn+1 + A|2
(124)
is a Ricci-flat Khler metric of complex dimension (n + 1), where

1
K,
H = 1/(n+1),
A=
(125)
n+1
and we have defined
1
K.
1 + n+1 + n+1 +
(126)
n+1
(The 1 is inessential here, of course.) The Khler function for d s 2 is given by
2
2
= (n + 1) (n+2)/(n+1) = (n + 1) H n+2 ,
K
n+2
n+2
and its Khler form is given by
J = H J + iH n (dn+1 + A) (d n+1 + A).
(127)
(128)
given
The proof is as follows. First, note that calculating J from the Khler function K
above, we get

2
= i (n + 1) (n+2)/(n+1)
J = i K
n+2
1/(n+1)
= i(n + 1) 1/(n+1)
+ i n/(n+1)
= i(n + 1)
+ i n/(n+1) (dn+1 + A) (d n+1 + A)
= i 1/(n+1) K
= H J + iH n (dn+1 + A) (d n+1 + A).
(129)
28
, we
Bearing in mind that the Khler form is related to the metric by J = i
g d d
see that this does indeed agree with the metric given in (124).
This shows that the metric (124) is indeed Khler. Finally, to show that it is Ricci-flat,
we calculate the determinant:
det(g)
= H 2n H 2n det(g) = det(g).
(130)
= i log det(g),
Since the Ricci form is given by R
it follows that if the Ricci form R for
for d s2 is
the metric ds 2 is zero (which was the initial assumption), then the Ricci form R
zero also.
It is easily seen that the 6-metric metric we obtained in Section 4.1.1 is an example of
this type, since the Khler function for the flat 4-torus can be taken to be K = |1 |2 + |2 |2 .
6.2. 7-metric of G2 holonomy from T 2 bundle over hyper-Khler
Let ds42 be a hyper-Khler metric, with a triplet of Khler forms J a , with associated
1-form potentials Aa(1) :
J a = dAa(1),
J a = 0.
(131)
It turns out to be convenient to let a range over the values 0, 1, 2.

Consider the metric
d s72
= H dy + H
4
dz + A(1)
2
+ H 2 ds42 ,
(132)
=1
where H = y. (We have set m = 1.) Define vielbeins by

e0 = H 2 dy,
e i = H ei ,
e = H 1 dz + A(1) ,
(133)
where a = (0, ), i = (3, 4, 5, 6), and ei is a vielbein for the hyper-Khler 4-metric ds42 .
Then it can be verified that the following 3-form is closed:
(3) e 0 e1 e2 + H 2 e0 J 0 : e J .
(134)
In fact this 3-form is covariantly constant, as can be verified using the expressions for
the spin connection:
0 = H 3 e ,
0i = H 3 ei ,
= 0,
1
1
i = H 3 Jij e i ,
ij = ij H 3 Jij e ,
2
2
(135)
where Jija denotes the components of J a with respect to the vielbein ea for ds42 , and ij
is the spin-connection for the vielbein ei . The covariant-constancy of (3) proves that the
metric d s72 has G2 holonomy, and it is also therefore Ricci flat.
From (135) is is also straightforward to calculate the vielbein components of the
Riemann tensor for the metric d s72 . We find
00 = 4H 6 ,
R
0ij = 2H 6 Jij ,
R
0i0i = 2H 6 ij ,
R
29
0ij = H 6 Jij ,
R
= H 6 ( ),
R
ij = 1 H 6 :a Jij0 ,
R
2
ij = 5 H 6 ij + 1 H 6 : Jij0 ,
R
4
4

1
ij kE = H 2 Rij kE H 6 Jik J J Jjk + 2Jij J + 4ik j E 4iE j k ,
R
jE
iE
kE
4
(136)
where Rij kE is the Riemann tensor of the hyper-Khler metric ds42 . It is easily verified from
AB of the metric d s2 is zero.
these expressions that the Ricci tensor R
7
6.3. 8-metric of Spin(7) holonomy from T 3 bundle over hyper-Khler
In a similar fashion, we can give the general construction for 8-metrics, in terms of a
hyper-Khler base metric ds42 . This time we shall have
d s82 = H 6 dy 2 + H 2
a
2
dz + Aa(1) + H 3 ds42 .
(137)
a=1
Note that here, it is convenient to label the three Khler forms of ds42 by J a = dAa(1) with
a = 1, 2, 3. We then define the vielbeins

ei = H 3/2ei ,
ea = H 1 dza + Aa(1) ,
e0 = H 3 dy,
(138)
where here i = 4, 5, 6, 7, and ei is a vielbein for the hyper-Khler metric ds42 . It can then
be verified that the self-dual 4-form (4) given by

1
1
(4) = e0 e1 e2 e3 + H 6 J a J a + H 3 e 0 ea + :abc e b e c J a
6
2
(139)
is closed. (Note that 16 J a J a is just another way of writing the volume form of ds42 .)
In fact it can also be verified that (4) is covariantly constant, by making use of the
following results for the spin connection of the 8-metric:
3
0i = H 4 ei ,
ab = 0,
2
1
1
ai = H 4 Jija ej ,
ij = ij H 4 Jija ea .
2
2
Calculating the curvature from this, we find
0a = H 4 ea ,
0a0b = 5H 8 ab ,
R
0i0j = 15 H 8 ij ,
R
4
0aij = 5 H 8 Jija ,
R
2
5
0iaj = H 8 J a ,
R
ij
4
(140)
30
abcd = H 8 (ac bd ad bc ),
R
abij = 1 H 8 :abc J c ,
R
ij
2
aibj = 7 H 8 ab ij + 1 H 8 :abc Jijc ,

R
4
4

1
2
ij kE = H Rij kE H 8 J a J a J a J a + 2J a J a + 9ik j E 9iE j k .
R
ik j E
iE j k
ij kE
4
(141)
2

It is easily verified that the Ricci tensor RAB for the 8-dimensional metric d s8 vanishes.
7. Cosmological resolutions
There is an alternative approach to resolving the various Heisenberg metrics that we
have been discussing in this paper. This involves modifying the requirement of Ricciflatness, so that instead the metrics are now required to satisfy the Einstein condition
with a negative Ricci tensor. It turns out in all the previous examples, we can now obtain
complete and non-singular non-compact metrics. In each case, this is achieved by replacing
the various powers of H appearing as prefactors of the terms (dzi + )2 by arbitrary
functions of the radial variable, and then solving the Einstein equations.
In all the cases we consider, the homothetic conformal Killing vector D of the original
Ricci-flat metric is replaced by a true Killing vector of the associated Einstein metric. This,
together with the generators of the nilpotent Heisenberg group generate a solvable group,
which acts simply-transitively on the Einstein manifold, which may thus be taken to be a
solvable group manifold Solv. In addition, all our metrics admit some manifest compact
symmetries, which act linearly on the Heisenberg manifold. We have also identified some
non-linearly acting symmetries. In all cases, we can express the manifold as G/H = Solv,
where the non-compact group G has maximal compact subgroup H . The group H contains
the linearly-realised compact symmetries. This is quite striking because a theorem of
Alekseevskii and Kimelfeld [34] states that any homogeneous non-compact Ricci flat
Riemannian metric must be flat [35]. In fact a theorem of Dotti states that a left-invariant
Einstein metric on a unimodular solvable group must be flat, so our solvable groups cannot
be unimodular, that is the trace of the structure constants of the Lie algebra cannot vanish
[36].10
The simplest example is when the Ricci-flat manifold is flat space, and the associated
solvable group manifold is hyperbolic space. This has been encountered in studies of
the AdS/CFT correspondence and is related to the ideas of Ref. [37], in which the fifth
dimension corresponds to the Liouville mode of a non-critical string theory which thus
becomes dynamical. The idea is that the string coordinates appear in the effective action
multiplied by a function of the Liouville field. This function should vanish at large negative
values of the Liouville field, in order to enforce a zig-zag symmetry. To achieve this and
to fix the functional form, the effective Lagrangian for the string is taken to include a piece
invariant under both Poincar transformations and dilatations. As a result of imposing
10 We shall demonstrate this explicitly below for all our examples.
31
the dilatation symmetry, an exponential function of the Liouville mode multiplies the
coordinates of the string. The vanishing of this function corresponds to the horizon in
AdS spacetime. From the ten-dimensional point of view, one must take the product metric
AdS5 S 5 , where the SO(6) R-symmetry group arises from the isometry group of the S 5
factor.
Our metrics arise by replacing the usual commuting translations of the string by
non-commuting translations satisfying a Heisenberg algebra. This may be relevant when
considering strings in constant background fields. It is a striking fact that the obvious
nilpotent symmetry is, as in the standard AdS case, enhanced to a much larger group G.
A common feature of all of our Ricci-flat metrics is that the size of the toric fibres goes
to zero as a negative power of distance as we go to infinity, while the size of the base
expands as a positive power. By contrast, for our Einstein metrics both directions expand
exponentially as one goes to infinity, but the toric fibre directions expand more rapidly
than the base. In some cases the exponential expansion of some of the directions in the
base is different from that of other directions. This has the consequence that the conformal
geometry on the boundary is singular. If one were to use as conformal factor the scalesize of the fibres, then the metric on the base would tend to zero. If one used the metric
on the smallest-growing base direction, then the metric on the fibres would diverge. In
the case of a single scale-factor, the resulting metric is referred to by mathematicians as a
CarnotCarathodory metric [38]. This behaviour has been commented on in Refs. [39,40]
in the case of the four-dimensional Bergman metric on SU(2, 1)/(SU(2) U (1)). In these
references the metric was written in coordinates adapted to the maximal compact subgroup.
At constant radius the metric is a squashed 3-sphere, where the ratio of lengths on the U (1)
fibres compared with the S 2 base diverges as one approaches infinity. In fact, as we shall
illustrate below, one may also write the metric in Heisenberg-horospherical coordinates
and obtain the same behaviour. In other words, the Bergman metric is of cohomogeneity
one with respect to both SU(2) and its contraction to the ur-Heisenberg group.
In order to get a solution of type IIB theory in ten dimensions, one needs a fivedimensional rather than a four-dimensional metric. In what follows we shall present a new
complete five-dimensional Einstein metric on a solvable group manifold, which may be
used to obtain a Euclidean-signatured solution of type IIB supergravity in ten dimensions,
with a complex self-dual 5-form. (This theory is obtained by Wick rotation from the
Lorentzian IIB theory, and the components of the 5-form with a time index are purely
imaginary.) In general, these solutions need have no real Lorentzian sections. Although
the five-dimensional metric may well have appeared before in general mathematical
classification schemes, it and all of our metrics that are not symmetric spaces have not,
as far as we are aware, been previously written down explicitly, nor have they been used in
the construction of supergravity solutions.
Physically, the unusual behaviour of the boundary appears to be related to a mismatch
in dimension between the boundary theory and the dimension of the bulk theory minus
one. This behaviour presumably arises because, from the KaluzaKlein point of view,
a Heisenberg isometry gives rise to a background magnetic field. Systems in strong
magnetic fields are well known to exhibit a reduction in dimensionality. The size of the
toric fibre here is inversely proportional to the electric charge, and so as we go to infinity in
32
our metrics the charge goes to zero, at constant magnetic field. Equivalently, the magnetic
field goes to infinity, at fixed electric charge.
In some instances, we can give a more complete interpolation between a Ricci-flat
Heisenberg metric and the associated cosmological resolution; these are constructed
in Section 7.3. Specifically, for the four-dimensional metric (5), and for the two sixdimensional metrics (49) and (58), we can obtain more general solutions with both a
charge parameter and a cosmological constant. Indeed, in the case of (5) and (49) these
Heisenbergde Sitter metrics are themselves specialisations of alrea dy known metrics
with cosmological constants. Thus the four-dimensional Heisenbergde Sitter metric is a
contraction limit of the EguchiHansonde Sitter metric [41,42], and the six-dimensional
Heisenbergde Sitter generalisation of (49) is a contraction of a complete metric with
cosmological constant on the complex line bundle over CP2 . General results for such
cosmological metrics on line bundles over EinsteinKhler spaces were obtained in [43].
We expect that the Heisenbergde Sitter generalisation of the six-dimensional metric (58)
that we obtain in Section 7.3 may similarly be a contraction limit of a Stenzelde Sitter
metric.
7.1. The Einstein metrics
We shall begin by listing the results Einstein metrics for all the cases. As remarked
above, in the original Ricci-flat metrics the lengths of the KaluzaKlein fibre directions go
to zero at large y while the base space expands. By contrast, in the related Einstein metrics
both the fibre and base-space directions expand exponentially as one goes to infinity. In fact
the fibre directions now expand faster than the base. After each metric, we give its Ricci
tensor, and also the algebra of exterior derivatives of the vielbein 1-forms. We choose the
obvious basis, with ds 2 = ea ea , and e0 = dt, etc.
D = 4; T 1 bundle over T 2 :

ds42 = dt 2 + 4k 2 e4kt (dz1 + z3 dz2 )2 + e2kt dz22 + dz32 ,
Rab = 6k 2 gab ,
de0 = 0,

de1 = 2k e0 e1 e2 e3 ,
de2 = ke0 e2 ,
de3 = ke0 e3 .
(142)

ds42 = dt 2 + 22k 2 e8kt (dz1 + z3 dz2 )2 + e4kt dz22 + dz32 + e6kt dz42 ,
Rab = 22k 2gab ,
de0 = 0,

de1 = 4k e0 e1 e2 e3 ,
de2 = 2ke0 e2 ,
de3 = 2ke0 e2 ,
de4 = 3ke0 e4 .
(143)
33

ds62 = dt 2 + 4k 2 e4kt (dz1 + z3 dz2 + z5 dz4 )2 + e2kt dz22 + dz32 + dz42 + dz52 ,
Rab = 8k 2 gab ,
de0 = 0,

de1 = 2k e0 e1 e2 e3 e4 e5 ,
de2 = ke0 e2 ,
de3 = ke0 e3 ,
de4 = ke0 e4 ,
de5 = ke0 e5 .
(144)

ds62 = dt 2 + 36k 2 e10kt (dz1 + z4 dz3 )2 + (dz2 + z5 dz3 )2

+ e4kt dz32 + e6kt dz42 + dz52 ,
Rab = 18k 2gab ,

de1 = 5k e0 e1 e3 e4 ,

de2 = 5k e0 e2 e3 e5 ,
de0 = 0,
de3 = ke0 e3 ,
de4 = ke0 e4 ,
de5 = ke0 e5 .
(145)

ds72 = dt 2 + 4k 2 e4kt (dz1 + z4 dz3 + z6 dz5 )2 + (dz2 + z5 dz3 z6 dz4 )2

+ e2kt dz32 + dz42 + dz52 + dz62 ,
Rab = 12k 2gab ,

de1 = 2k e0 e1 e3 e4 e5 e6 ,

de2 = 2k e0 e2 e3 e5 e6 e4 ,
de0 = 0,
de3 = ke0 e3 ,
de4 = ke0 e4 ,
de5 = ke0 e5 ,
de6 = ke0 e6 .
(146)

ds72 = dt 2 + 6k 2 e4kt (dz1 + z6 dz5 )2 + (dz2 + z6 dz4 )2 + (dz3 + z5 dz4 )2

+ e2kt dz42 + dz52 + dz62 ,
Rab = 15k 2gab ,

de2 = 2k e0 e2 e4 e6 ,
de1 = 2k e0 e1 e5 e6 ,

de4 = ke0 e4 ,
de3 = 2k e0 e3 e4 e5 ,
de0 = 0,
de5 = ke0 e5 ,
de6 = ke0 e6 .
(147)
34

ds82 = dt 2 + 4k 2 e4kt (dz1 + z5 dz4 + z7 dz6 )2 + (dz2 + z6 dz4 z7 dz5 )2

+ (dz3 + z7 dz4 + z6 dz5 )2 + e2kt dz42 + dz52 + dz62 + dz72 ,
Rab = 16k 2gab ,

de1 = 2k e0 e1 e4 e5 e6 e7 ,

de2 = 2k e0 e2 e4 e6 + e5 e7 ,

de3 = 2k e0 e3 e4 e7 e5 e6 ,
de0 = 0,
de4 = ke0 e4 ,
de5 = ke0 e5 ,
de6 = ke0 e6 ,
de7 = ke0 e7 .
(148)
ec ,
Note that the algebras of exterior derivatives are all of the form
a
where the c bc are constants. These are in fact the structure constants of the corresponding
solvable groups. Observe that these are indeed not traceless, ca ba = 0, as is required by
Dottis theorem.
In the next subsection, we shall discuss these solvable groups as coset spaces, exploiting
the Iwasawa decomposition.
dea
= 12 ca bc eb
7.2. Coset constructions

n = Hn
7.2.1. SU(n, 1)/U (n) = CP
C
Those examples above whose principal orbits are of the form of T 1 bundles over T p
n , with p = 2n 2. These are
are in fact Bergman metrics on the non-compact forms of CP
nothing but the FubiniStudy metrics with the opposite sign for the cosmological constant.
They are obtained by starting from coordinates Z A on Cn+1 , with the constraint
B = 1,
AB Z A Z
(149)
where AB is diagonal with 00 = 1, ab = 1, where 1 a n. The Hopf fibration of

this AdS2n+1 by U (1) (taken to be timelike) then gives the Bergman metric.
We can express the Bergman metric in a horospherical form [44], by introducing real
coordinates (, , , xi , yi ), in terms of which we parametrize the Z A that satisfy (149) as
1

i
Z 0 = e 2 cosh 12 + 18 e 2 4i + xi2 + yi2 ,
1

i
Z n = e 2 sinh 12 18 e 2 4i + xi2 + yi2 ,
Z i = 12 e 2 + 2 (xi + iyi ),
i
(150)
= AB
where 1 i n 1. Substituting into the metric
find
2

2
,
d s2 = 14 d + e d + 12 (yi dxi xi dyi ) + d2n
d s2
B
dZ A d Z
on AdS2n+1 , we
(151)
where
2

2
= 14 d 2 + 14 e dxi2 + dyi2 + 14 e2 d + 12 (yi dxi xi dyi ) .
d2n
(152)
35
Thus if we fibre AdS2n+1 by the U (1) whose coordinate is the time parameter, we obtain
n . Comparing with (142) and (144), we see that these
the Bergman metric (152) on CP

2
3
correspond to CP and CP , respectively.
If n = 3, the denominator group U (3) contains the linearly-realised U (2) noted in
Section 4.1.1, and similarly if n = 2 the denominator group U (2) contains the linearlyrealised U (1) noted in Section 2.1.
In the case n = 2 one can regard this solution as a special case of the TaubNUT
de Sitter metrics, which have been applied to the AdS/CFT correspondence in [45]. (For
more general higher-dimensional metrics of this type, see [46].) The case n = 2 is of further
interest because, while the Bergman metric is not conformal to the associated Ricci-flat
metric,11 there is a metric which is conformal to our Ricci-flat metric that is distinguished
by the property that it is essentially the only non-trivial complete homogeneous hyperHermitean metric [47]. The conformally related metric is

ds 2 + z4 (d + x dy)2 + z2 dx 2 + dy 2 + dz2 .
(153)
It would be interesting to know whether our other Ricci-flat metrics are conformal to
similarly-distinguished metrics.
n = Hn
7.2.2. Sp(n, 1)/(Sp(n) Sp(1)) = HP
H
A similar construction can be given for the non-compact versions of the quaternionic
n . Now, we start from n + 1 quaternionic coordinates QA , subject to
projective spaces, HP
the constraint
B = 1,
AB QA Q
(154)
where again AB is diagonal with 00 = 1, ab = 1, where 1 a n. This restricts us to

a spacetime of anti-de Sitter type, except that now we have three timelike coordinates.
We can again introduce real horospherical coordinates. The three times appear as
the Euler angles of SU(2), and in fact we can just introduce them implicitly via the
Sp(1) = SU(2) quaternionic U . In addition, we introduce real coordinates (, , xi , yi ),
where 1 3 and 1 i n 1. We shall denote the imaginary unit quaternions by
k
i
k
= (i, j, k) (in terms of which U can be written as U = e 2 t1 e 2 t2 e 2 t3 ). We then parametrize
A
the quaternions Q that satisfy (154) as

2
1
,
Q0 = U cosh 12 + 18 e 2 4 + xi2 + yi

1
2
Qn = U sinh 12 18 e 2 4 + xi2 + yi
,

1
Qi = 12 U e 2 xi + yi .
(155)
These are closely analogous to (150) for the complex case. Substituting into the metric
B on the three-timing AdS4n+3 , we find
d s2 = AB dQA d Q

2

2
,
d s2 = U 1 dU + e d + 12 xi dyi yi dxi + d4n
(156)
11 It is impossible for two Riemannian Einstein metrics to be conformal with non-constant conformal factor.
36
where

2

2
2
= 14 d 2 + 14 e dxi2 + dyi + 14 e2 d + 12 xi dyi yi dxi .
d4n
(157)
Thus if we project orthogonally to the SU(2) timelike fibres, we obtain (157) as the metric
n . It is the coset Sp(n, 1)/(Sp(n) Sp(1)).
on HP
Comparing with (148), we see that our 8-dimensional Einstein metric is precisely
2 , which is the coset Sp(2, 1)/
the non-compact quaternionic Bergman metric on HP
(Sp(2) Sp(1)). In this case the denominator group contains the linearly-realised Sp(1)
SU(2) noted in Section 4.4.1.
In all cases, one may check that the exponential expansion of the SU(2) fibres is more
n base, as one goes to infinity.
rapid than that of the HP
7.3. Heisenbergde Sitter metrics
So far, we have considered Ricci-flat Heisenberg metrics, and also cosmological
resolutions that do not have an immediate mathematical relation to the Heisenberg
metrics. In certain cases, at least, we can find a more general solution that encompasses
both the Heisenberg metric and the cosmological metric, as certain limits. In fact, at least in
some of these examples, we know that there exist de Sitterised versions of the complete
non-singular Ricci-flat metrics, even before the Heisenberg limit is taken.
A case in point is the EguchiHansonde Sitter metric, given by

1
1
ds42 = F 1 dt 2 + r 2 F 32 + r 2 12 + 22 ,
(158)
4
4
where
Q 1
F = 1 + 4 r 2 .
(159)
6
r
It is an Einstein metric, with Rab = gab , and so is the cosmological constant.
If we now take the Heisenberg limit, as in Section 2.2, we obtain the Heisenberg
de Sitter metric

1
1
ds42 = F 1 dt 2 + r 2 F (dz1 + z3 dz2 )2 + r 2 dz22 + dz32 ,
(160)
4
4
where we are dropping the tildes used to denote the rescaled quantities in Section 2.2, and
now we have
Q 1
F = 4 r 2 .
(161)
6
r
Note that does not suffer any rescaling in the taking of this limit, so the cosmological
constant is still .
Having obtained the Heisenbergde-Sitter 4-metric, we can note that if we set to zero,
we recover (after an obvious coordinate transformation) the Heisenberg metric (5). On the
other hand, if we set Q to zero, then after an obvious coordinate transformation, (160)
gives the cosmological resolution (142). If we keep both Q and non-vanishing, we have
a more general Einstein metric that encompasses both the Heisenberg and cosmological
metrics discussed previously.
37
We can now attempt a generalisation of the above to other cases. Analogues of the
EguchiHansonde Sitter metric are known for all Ricci-flat metrics on complex line
bundles over EinsteinKhler bases [43]. Thus we can expect to be able to get generalised
Heisenbergde Sitter metrics for all the cases where the principal orbits are T 1 bundles
over T 2n . For example, in D = 6 we can get a Heisenbergde Sitter metric for the case of
T 1 bundle over T 4 , namely,

2 1
ds62 = F 1 dr 2 + r 2 F dz1 + m(z3 dz2 + z5 dz4 ) + r 2 dz22 + + dz52 , (162)
2
where
Q 1 2
(163)
r .
r6 8
This is equivalent to (49) if is set to zero. On the other hand, if Q is instead set to zero,
it is equivalent to the cosmological resolution metric (144).
A slightly more complicated example is the second of the two D = 6 Heisenberg
metrics, given in (58). Here, we find that the following is an Einstein metric, with
cosmological constant :

ds62 = F 1 dr 2 + r 2 F 2/3 (dz1 + mz4 dz3 )2 + (dz2 + mz5 dz3 )2

1
+ r 2 F 1/3 dz32 + r 2 dz42 + dz52 ,
(164)
2
where
F=
Q 1 2
(165)
r .
r6 8
This again has the appropriate limits, yielding (58) if is set to zero, and yielding (145) if
instead Q is set to zero.
Having obtained these Heisenbergde Sitter metrics, we can observe that they provide
a way to cap off the large-radius portions of the Ricci-flat Heisenberg metrics that are
transverse to the domain-wall spacetimes. In this respect, they appear to be conjugate to
the constructions of [11], which by contrast resolve the curvature singularities in the smallradius portions of the domain-wall spacetimes.
Let us illustrate this by considering the example of the 4-metric (5), transverse to the
domain wall in D = 8 supergravity. If we take the Heisenbergde Sitter metric (160), with
both Q and the cosmological constant in (161) positive, we see that the metric running
from the singularity at r = 0 reaches a natural endpoint at r = r0 , where the function F
vanishes, i.e., at
F=
6Q
.
(166)
To stu dy the behaviour of the metric near r = r0 , we introduce a new coordinate defined
by r = r0 2 . In terms of this, the metric near r = r0 takes the form

2r05
r02 6Q 2
1
2
2
2
d +
ds4
(167)
(dz1 + z3 dz2 ) + r02 dz22 + dz32 .
3Q
16 r05
4
r06 =
38
This will be regular at r = r0 provided that z1 has a period given by

Mz1 =
4r04
.
3Q
(168)
8. Conclusions
As mentioned in the introduction, one of the motivations for the present stu dy was the
possibility of resolving some of the BPS domain walls along the lines of the single-sided
domain-wall construction described in [11]. The case studied in detail in that reference
was four-dimensional, and the resolution was a certain non-compact degeneration of a K3
surface, but generalisations to higher dimensions were also indicated there which would
correspond, for example, to non-compact degenerations of CalabiYau complex 3-folds
and 4-folds. For example, the metric based on a circle bundle over a K3 surface seems
to be related to the metric one would obtain by solving the MongeAmpre equation on
the complement of a quartic surface in CP3 . In the case of CalabiYau metrics, one has
various proofs showing the existence of smooth resolved metrics, but these give very little
detailed information about the explicit forms of the metrics. The information one gets is
mainly about the asymptotic form of the metric near infinity. This is where our work may
help identify the resolutions. In the case of K3 surfaces, as well as asymptotically nilmanifolds based on generalised Heisenberg groups, one also encounters asymptotically
solv-manifolds, based on solvable groups. It may be that our work in this paper and
generalisations of it may be relevant to higher-dimensional CalabiYau spaces. A more
challenging problem would be to relate our metrics to compact CalabiYau manifolds.
For reasons explained in the introduction, it is not easy to see how to do this using
cohomogeneity one Ricci-flat metrics. One may, of course, give qualitative discussions
[48] but it is extremely hard to make quantitative progress with present-day techniques.
In the case of metrics with exceptional holonomy the situation is much less clear than in
the case of CalabiYau metrics, because existence theorems have been studied to a much
lesser extent. However, the important work of Joyce described in his recent book [49]
encourages us to believe that our results will prove applicable in that case as well.
Acknowledgements
We should like to thank Massimo Bianchi, Anna Ceresole, Mirjam Cvetic, Isabel
Dotti, Andre Lukas, Krystof Pilch, Toine van Proeyen, Simon Salamon and Paul Tod for
helpful conversations and information. In particular, we should like to thank Paul Tod
for an essential remark about conformally-related metrics. Subsets of the authors would
like to thank Texas A&M Physics Department (H.L., K.S.S.), Imperial College (C.N.P.),
CERN (G.W.G., H.L., C.N.P., K.S.S.), SISSA (C.N.P., K.S.S.), the Institut Henri Poincar
(G.W.G., C.N.P., K.S.S.), Michigan University Physics Department (C.N.P.), DAMTP
(C.N.P.), the Benasque Centre for Science (G.W.G., C.N.P., K.S.S.) and the Ecole Normale
(C.N.P., K.S.S.) for hospitality at various stages during this work.
39
Appendix A. The Raychaudhuri equation

Let T be a unit vector tangent to a congruence of curves,
T T = 1 T; T = 0.
(A.1)
Let us decompose the covariant derivative of T perpendicular and parallel to T using the
projection operator h = g T T , such that h T = 0. One has the decomposition
T; = + + T a ,
(A.2)
where = is a symmetric expansion tensor and = is an antisymmetric

vorticity tensor, both of which are orthogonal to T ,
T = T = 0.
(A.3)
is the acceleration vector, and would vanish if the congruence is

The vector a = T;
geodesic.
Since T a = 0, we have
T
= h = g .
T;
(A.4)
This is the expansion. We may then set

h
+ ,
d
is the shear, satisfying h = g = 0. Thus we have
=
where
T; T ; = .
(A.5)
(A.6)
Note that only the vorticity term is negative.

From the Ricci identity, one now has
T;;
= R T ,
T;;
(A.7)
so
d
= T; T ; R T T T; T ;
(A.8)
dt
and hence
d
= a;
R T T + .
(A.9)
dt
The assumption that the congruence is geodesic implies a = 0. The assumption that it
is hypersurface orthogonal implies T[; T] = 0 [ T] = 0 = 0. Thus,
d
= R T T ,
dt
and so
1
d
= 2 2 R T T .
dt
d
Finally, we set R = 0, and find
d
dt
< 0.
(A.10)
(A.11)
40
Appendix B. Generalised Heisenberg groups

As we discussed in the introduction, we are interested in the relation between the
holonomy spaces with non-Abelian isometry groups and those with nilpotent Heisenberg
groups. In this section, we describe Heisenberg groups as contractions of semi-simple
groups.
B.1. Definition
We may define a generalised Heisenberg group as a (nilpotent) central extension of an
Abelian group, with Lie algebra generated by e and qm , with the commutation relations
m
[e , e ] = F
qm ,
[ea , qm ] = 0,
[qm , qn ] = 0.
(B.1)
We shall suppose the original Abelian algebra to be q-dimensional, and the centre to be
p-dimensional. Thus 1 q and 1 m p. An appropriate left-invariant basis of
1-forms is ( dx , m ), where
1 m
m = dy m F
x dx ,
2
and we have
(B.2)
1 m
d m = F
(B.3)
dx dx .
2
The right-invariant Killing vectors R and Rm , which generate left translations, are given
by
1
R = + F x m ,
2
These satisfy
m
Rm ,
[R , R ] = F
Rm = m .
[R , Rm ] = 0,
(B.4)
[Rm , Rn ] = 0.
(B.5)
There is an obvious KaluzaKlein interpretation for the central coordinates y m . The

m
quantities F
are just q constant U (1) field strengths defined over Ep , and (B.2) can be
1
viewed as the quantity m = dy m Am , where Am

= 2 F x is the KaluzaKlein vector
m
potential for F
.
The BakerCampbellHausdorff formula gives
eae ebe eae ebe = eqm F
m (a,b)
(B.6)
where we have defined a e
In practice, we want to
and
consider the case where the original Abelian algebra is a torus T q , whose coordinates x
therefore live on a lattice. Because of (B.6), the coordinates y m associated to the centre
must also be identified consistently. Suppose that x and x + a are to be identified, and
that x and x + b are to be identified. The associated group elements eae and ebe can
taken to be the identity only if the group element on the right-hand side of (B.6) is also
the identity. This means that F m (a, b) must be an integer multiple of one of the periods
a e
F m (a, b)
m a b .
F
41
m
of the coordinates y m , for all lattice vectors a and b . This places conditions on the F
.
The case we are mainly interested in is when the resulting group may be thought of as a
T p bundle over T q . The simplest example is when p = 1 and the consistency conditions
reduce to the Dirac quantisation conditions for a U (1) bundle over T q . Thus in this case
F (a, b) is the magnetic flux through the cycle spanned by a and b .
B.2. Heisenberg groups as contractions

Heisenberg algebras may arise as contractions of semi-simple algebras. The simplest
example is the InnWigner contraction of SO(3) to the ur Heisenberg algebra. The
former is
[L3 , L ] = L ,
[L+ , L ] = 2L3 .
Writing e1 = L1 , e2 = L2 and
to zero, we obtain the latter:
[e1 , e2 ] = q,
q = 2 L3 ,
(B.7)
and taking the limit where the constant goes
[e1 , q] = [e2 , q] = [q, q] = 0.
(B.8)
From the KaluzaKlein point of view, SO(3) is the Dirac T 1 bundle over S 2 . Contraction
gives a magnetic field over E2 , which if we identify to make a torus, gives an T 1 bundle
over T 2 .
Since all the AC manifolds we are considering here have isometry groups of the type
SO(n) or SU(n), we shall consider the contractions of these groups.
B.2.1. Contractions of SO(n + 2)
This example extends straightforwardly to the case of SO(n + 2). It is convenient to
work with the left-invariant basis of 1-forms, which we shall denote by LAB . These satisfy
dLAB = LAC LCB . Splitting the index as A = (1, 2, i), and defining
Lij = Mij ,
L1i = i ,
L2i = i ,
L12 = ,
(B.9)
then after the scalings

Mij Mij ,
i i ,
i i ,
(B.10)
we have
di = 2 i + Mij j ,
d = i i ,
d i = 2 i + Mij j ,
dMij = Mik Mkj i j i j .
(B.11)
Taking the limit where goes to zero, we obtain the generalised Heisenberg algebra with
di = d i = 0,
d = i i ,
dMij = 0.
(B.12)
The 12 (n + 1)(n + 2)-dimensional so(n + 2) simple algebra has decomposed in the

0 limit as the direct sum of a (2n + 1)-dimensional generalised Heisenberg algebra,
spanned by (i , i , ), and a 12 n(n 1)-dimensional completely Abelian piece, spanned
by the Mij . It is consistent to identify the Heisenberg group in such as way as to obtain a
42
T 1 bundle over T 2n . One has 2n coordinates x on the base, whose differentials dx give
i and i . The real coordinates can be grouped into n complex coordinates zi , with
zi = x i + ix i+n ,
dzi = i + i i .
(B.13)
Cn .
The field strength F is proportional to the standard Khler form on

A different contraction of so(n + 2) may be obtained by instead applying the scalings
Mij Mij ,
i 2 i ,
i i ,
(B.14)
After sending to zero, we obtain

di = i ,
d i = 0,
d = 0,
dMij = 0.
(B.15)
1
2 (n + 1)(n + 2)-dimensional
The
so(n + 2) simple algebra has again decomposed in the
0 limit as the direct sum of a (2n + 1)-dimensional generalised Heisenberg algebra,
spanned by (i , i , ), and a 12 n(n 1)-dimensional completely Abelian piece, spanned by
the Mij . However now, it is consistent to identify the Heisenberg group in such as way as to
obtain a T n bundle over T n+1 . One has (n + 1) coordinates x on the base manifold T n+1
i are now all simple; F i = .
whose differentials give i and . The field strengths F
i
B.2.2. Contraction of SO(4)
The so(4) algebra is the direct sum of two so(3) algebras:
di = 12 :ij k j k ,
i = 1 :ij k
k .
j
d
2
(B.16)
Let
1 1 1 ,
1 =
2
1 = 1 ,
2 1 2 ,
2 =
2
2 = 2 ,
3 1 3 ,
3 =
2
3 = 3 .
(B.17)
These give
1
1
1
d1 = 2 3 2 3 + 3 2 + 2 3 ,
2
2
4
d1 = 2 3 , and cyclic.
and cyclic on {123},

(B.18)
We now implement the constant rescalings

(1 , 2 , 3 ) 2 (1 , 2 , 3 ),
(1 , 2 , 3 ) (1 , 2 , 3 ).
(B.19)
Now, we find that (B.18) becomes

1
1
1
d1 = 2 2 3 2 3 + 3 2 + 2 3 ,
2
2
4
d1 = 2 3 , and cyclic.
and cyclic,
(B.20)
The Heisenberg limit now corresponds to sending to zero, implying

1
1
d1 = 2 3 ,
d2 = 3 1 ,
4
4
d2 = 0,
d3 = 0.
d1 = 0,
1
d3 = 1 2 ,
4
(B.21)
43
In this limit, we may introduce coordinates (x1 , x2 , x3 ) and (y1 , y2 , y3 ) such that
1
1
1 = dy1 x3 dx2 ,
2 = dy2 x1 dx3 ,
4
4
1 = dx1 ,
2 = dx2 ,
3 = dx3 .
It is consistent to make identifications to give a T 3
1
3 = dy3 x2 dx1 ,
4
(B.22)
bundle over T 3 . The three field strengths
F i are given by
1
Fji k = :ij k .
2
(B.23)
B.2.3. Contraction of SO(5)

Our next example is for SO(5). Splitting the SO(5) index as A = (, 4), and then
splitting = (0, i) with i = 1, 2, 3, we define
1
L0i + :ij k Lj k Ri = 4 i ,
2
L0i 12 :ij k Lj k Li = 3 J i ,
L4 P = 2 ,
(B.24)
we obtain, after sending to zero,

1
d i = 0 i :ij k j k ,
(B.25)
d = 0,
dJi = 0.
2
The ten-dimensional so(5) algebra has thus been decomposed as a direct sum of a
seven-dimensional generalised Heisenberg algebra spanned by and i , and a threedimensional completely Abelian piece spanned by the J i . It is now consistent to identify
the Heisenberg group in such as way as to obtain a T 3 bundle over T 4 . There are four
i
coordinates x on the base whose differentials give the . The three field strengths F
4
give three self-dual 2-forms on T , which endow it with a hyper-Khler structure. The
left-invariant 1-forms can be written as
0 = dx0 ,
1 = dx1 ,
1 = dy1 x0 dx1 x2 dx3,
2 = dx2 ,
3 = dx3 ,
2 = dy2 x0 dx2 x3 dx1 ,
3 = dy3 x0 dx3 x1 dx2.
(B.26)
There is in fact a different contraction of so(5), in which the

is achieved by using the scalings
1
L0i + :ij k Lj k = 4 i ,
2
After sending to zero, we now find
L5 = 2 ,
Ji
act non-trivially. This
1
L0i :ij k Lj k = J i .
2
(B.27)
1
1
dJ i = :ij k J j J k ,
d i = 0 i :ij k j k ,
2
2
1 i
1 i
1
d0 = J i ,
(B.28)
di = J 0 + :ij k J j k .
2
2
2
This algebra is the semi-direct sum of the previously-obtained generalised Heisenberg
algebra with so(3), spanned by the J i .
44
A further contraction of SO(5) is possible, leading to a seven-dimensional Heisenberg

algebra in which is the direct sum of a six-dimensional Heisenberg algebra and a onedimensional summand. We obtain this by implementing a further singular scaling of 3 .
Equivalently, we can obtain the algebra directly as a limit of the so(5) algebra, by making
the rescalings
P = 2 ,
R1 = 4 1 ,
R2 = 4 2 ,
R3 = 3 3 .
(B.29)
After sending to zero, we get the contracted algebra

d1 = 0 1 2 3 ,
d3 = 0,
d2 = 0 2 3 1 ,
d = 0.
(B.30)
The left-invariant 1-forms can be written as

0 = dx0 ,
1 = dx1 ,
1 = dy1 x0 dx1 x2 dx3,
2 = dx2 ,
3 = dx3 ,
2 = dy2 x0 dx2 x3 dx1 ,
3 = dy3 .
(B.31)
B.2.4. Contraction of SU(n + 1)

The left-invariant 1-forms LA B of SU(n + 1) satisfy the MaurerCartan relations
dLA B = LA C LC B . Splitting the index as A = (0, ), we make the following definitions:
L0 0 = ,
L0 = ,
L = M .
(B.32)
After the scalings

2 ,
M M ,
(B.33)
dM = 0.
(B.34)
and sending to zero, we obtain

d = ,
d = 0,
The generalised Heisenberg algebra spanned by , and corresponds to an T 1 bundle

over T 2n . It is in fact identical to the algebra (B.12) that we obtained earlier as a contraction
of so(n + 2).
References
[1] T. Eguchi, A.J. Hanson, Asymptotically flat selfdual solutions to Euclidean gravity, Phys. Lett. B 74 (1978)
249.
[2] P. Candelas, X.C. de la Ossa, Comments on conifolds, Nucl. Phys. B 342 (1990) 246.
[3] R.L. Bryant, S. Salamon, On the construction of some complete metrics with exceptional holonomy, Duke
Math. J. 58 (1989) 829.
[4] G.W. Gibbons, D.N. Page, C.N. Pope, Einstein metrics on S 3 , R3 and R4 bundles, Commun. Math.
Phys. 127 (1990) 529.
[5] S.W. Hawking, Gravitational instantons, Phys. Lett. A 60 (1977) 81.
[6] M.F. Atiyah, N.J. Hitchin, Low-energy scattering of nonabelian monopoles, Phys. Lett. A 107 (1985) 21.
45
[7] M. Cvetic, G.W. Gibbons, H. L, C.N. Pope, New complete non-compact Spin(7) manifolds, hepth/0103155;
M. Cvetic, G.W. Gibbons, H. L, C.N. Pope, New cohomogeneity one metrics with Spin(7) holonomy,
math.DG/0105119.
[8] A. Brandhuber, J. Gomis, S.S. Gubser, S. Gukov, Gauge theory at large N and new G2 holonomy metrics,
hep-th/0106034.
[9] I.V. Lavrinenko, H. L, C.N. Pope, From topology to generalised dimensional reduction, Nucl. Phys. B 492
(1997) 278, hep-th/9611134.
[10] I.V. Lavrinenko, H. L, C.N. Pope, Fibre bundles and generalised dimensional reductions, Class. Quantum
Grav. 15 (1998) 2239, hep-th/9710243.
[11] G.W. Gibbons, P. Rychenkova, Single-sided domain walls in M-theory, J. Geom. Phys. 32 (2000) 311, hepth/9811045.
[12] M. Cvetic, S. Griffies, S. Rey, Static domain walls in N = 1 supergravity, Nucl. Phys. B 381 (1992) 301,
hep-th/9201007.
[13] M. Cvetic, S. Griffies, H.H. Soleng, Local and global gravitational aspects of domain wall spacetimes,
Phys. Rev. D 48 (1993) 2613, gr-qc/9306005.
[14] M. Cvetic, Extreme domain wallblack hole complementarity in N = 1 supergravity with a general dilaton
coupling, Phys. Lett. B 341 (1994) 160, hep-th/9402089.
[15] M. Cvetic, H.H. Soleng, Naked singularities in dilatonic domain wall spacetime, Phys. Rev. D 51 (1995)
5768, hep-th/9411170.
[16] M. Cvetic, H.H. Soleng, Supergravity domain walls, Phys. Rep. 282 (1997) 159, hep-th/9604090.
[17] L.J. Romans, Massive N = 2a supergravity in ten dimensions, Phys. Lett. B 169 (1986) 374.
[18] J. Polchinski, E. Witten, Evidence for heterotic-type I string duality, Nucl. Phys. B 460 (1996) 525, hepth/9510169.
[19] E. Bergshoeff, M. de Roo, M.B. Green, G. Papadopoulos, P.K. Townsend, Duality of type II 7-branes and
8-branes, Nucl. Phys. B 470 (1996) 113, hep-th/9601150.
[20] G.W. Gibbons, N.S. Manton, Classical and quantum dynamics of BPS monopoles, Nucl. Phys. B 274 (1986)
183224.
[21] N. Seiberg, IR dynamics on branes and spacetime geometry, hep-th/9606017;
N. Seiberg, E. Witten, Gauge dynamics and compactification to three dimensions, hep-th/9607163;
A. Sen, Strong coupling dynamics of branes from M-theory, hep-th/9708002;
A. Sen, A note on enhanced gauge symmetries in M- and string-theory, hep-th/9707123.
[22] G.W. Gibbons, C.N. Pope, Positive action theorems for ALE and ALF spaces, ICTP/81/82-20, available
from KEK;
C. LeBrun, Counterexamples to the generalised positive action conjecture, Commun. Math. Phys. 118 (1988)
591;
H. Nakajima, Self-duality of ALE Ricci-flat 4-manifolds and positive mass theorem, Adv. Stud. Pure
Math. 18-I (1990) 385;
M. Dahl, The positive mass theorem for ALE manifolds, Banach Center Publications 41, Part 1;
A. Adams, J. Polchinski, E. Silverstein, Do not panic! Closed string tachyons in ALE spacetimes, hepth/0108075.
[23] P.M. Cowdall, H. L, C.N. Pope, K.S. Stelle, P.K. Townsend, Domain walls in massive supergravities, Nucl.
Phys. B 486 (1997) 49, hep-th/9608173.
[24] H. L, C.N. Pope, T.A. Tran, K.W. Xu, Classification of p-branes, NUTs, waves and intersections, Nucl.
Phys. B 511 (1998) 98, hep-th/9708055.
[25] A. Strominger, A. Volovich, Holography for coset spaces, JHEP 9911 (1999) 013, hep-th/9905211.
[26] V.A. Belinsky, G.W. Gibbons, D.N. Page, C.N. Pope, Asymptotically Euclidean bianchi IX metrics in
quantum gravity, Phys. Lett. B 76 (1978) 433.
[27] H. L, C.N. Pope, p-brane solitons in maximal supergravities, Nucl. Phys. B 465 (1996) 127, hepth/9512012.
[28] E. Cremmer, B. Julia, H. L, C.N. Pope, Dualization of dualities. I, Nucl. Phys. B 523 (1998) 73, hepth/9710119.
[29] C.M. Hull, Massive string theories from M-theory and F-theory, JHEP 9811 (1998) 027, hep-th/9811021.
46
[30] G.W. Gibbons, P. Rychenkova, Cones, tri-sasakian structures and superconformal invariance, Phys. Lett.
B 443 (1998) 138, hep-th/9809158.
[31] G.W. Gibbons, S.W. Hawking, Gravitational multi-instantons, Phys. Lett. B 78 (1978) 430.
[32] M.B. Stenzel, Ricci-flat metrics on the complexification of a compact rank one symmetric space,
Manuscripta Math. 80 (1993) 151.
[33] M. Cvetic, G. Gibbons, H. L, C.N. Pope, Ricci-flat metrics, harmonic forms and brand resolutions, hepth/0012011.
[34] D.V. Alekseevskii, B.N. Kimelfeld, Structure of homogeneous Riemannian spaces with zero Ricci
curvature (in Russian), Funkcional. Anal. i Priloen. 9 (2) (1975) 511.
[35] For a review of non-compact homogeneous Einstein metrics, see: J. Heber, Noncompact homogeneous
Einstein spaces, Invent. Math. 133 (2) (1998) 279352.
[36] I. Dotti-Miatello, Ricci curvature of left-invariant metrics on solvable unimodular Lie groups, Math.
Z. 180 (2) (1982) 257263.
[37] 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.
[38] O. Biquard, Einstein deformations of hyperbolic metrics, in: Essays on Einstein Manifolds, Surv. Differ.
Geom., Vol. VI, Int. Press, Boston, 1999, pp. 235246.
[39] R. Britto-Pacumio, A. Strominger, A. Volovich, Holography for coset spaces, JHEP 9911 (1999) 013, hepth/9905211.
[40] M. Taylor-Robinson, Holography for degenerate boundaries, hep-th/0001177.
[41] G.W. Gibbons, C.N. Pope, The positive action conjecture and asymptotically Euclidean metrics in quantum
gravity, Commun. Math. Phys. 66 (1979) 267.
[42] H. Pedersen, EguchiHansen metrics with cosmological constant, Class. Quantum Grav. 2 (1985) 579;
H. Pedersen, Einstein metrics, spinning top motions and monopoles, Math. Ann. 274 (1986) 35.
[43] D.N. Page, C.N. Pope, Inhomogeneous Einstein metrics on complex line bundles, Class. Quantum Grav. 4
(1987) 213.
[44] C.N. Pope, A. Sadrzadeh, S.R. Scuro, Timelike Hopf duality and type IIA string solutions, Class. Quantum
Grav. 17 (2000) 623.
[45] A. Chamblin, R. Emparan, C.V. Johnson, R.C. Myers, Large N phases, gravitational instantons and the nuts
and bolts of AdS holography, Phys. Rev. D 59 (1999) 064010, hep-th/9808177.
[46] A. Awad, A. Chamblin, A bestiary of higher-dimensional TaubNUTAdS spacetime, hep-th/0012240.
[47] M.L. Barberis, Homogeneous hyperhermitian metrics which are conformally hyperkhler,
math.DG/0009035.
[48] P. Aspinwall, M-theory versus F-theory pictures of the heterotic string, hep-th/9707014.
[49] D. Joyce, Compact Manifolds With Special Holonomy, Oxford Univ. Press, 2000.

Anomaly mediated supersymmetry breaking

without R-parity
F. de Campos a , M.A. Daz b , O.J.P. boli c , M.B. Magro c ,
P.G. Mercadante c
a Departamento de Fsica e Qumica, Universidade Estadual Paulista, Av. Dr. Ariberto Pereira da Cunha 333,
Guaratinguet, SP, Brazil

b Departamento de Fsica, Universidad Catlica de Chile, Av. Vicua Mackenna 4860, Santiago, Chile
c Instituto de Fsica, Universidade de So Paulo, CP 66.318, 05389-970, So Paulo, SP, Brazil
Received 9 October 2001; accepted 4 December 2001
Abstract
We analyze the low energy features of a supersymmetric standard model where the anomalyinduced contributions to the soft parameters are dominant in a scenario with bilinear R-parity
violation. This class of models leads to mixings between the standard model particles and
supersymmetric ones which change the low energy phenomenology and searches for supersymmetry.
In addition, R-parity violation interactions give rise to small neutrino masses which we show to be
consistent with the present observations. 2002 Elsevier Science B.V. All rights reserved.
PACS: 12.60.Jv; 14.60.Pq; 14.80.Ly
1. Introduction
Supersymmetry (SUSY) is a promising candidate for physics beyond the Standard
Model (SM) and there is a large ongoing search for supersymmetric partners of the
SM particles. However, no positive signal has been observed so far. Therefore, if
supersymmetry is a symmetry of nature, it is an experimental fact that it must be
broken. The two best known classes of models for supersymmetry breaking are gravitymediated [1] and gauge-mediated [2] SUSY breaking. In gravity-mediated models, SUSY
is assumed to be broken in a hidden sector by fields which interact with the visible particles
only via gravitational interactions and not via gauge or Yukawa interactions. In gaugeE-mail address: magro@fma.if.usp.br (M.B. Magro).
0550-3213/02/$ see front matter 2002 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 1 8 - 6
48
F. de Campos et al. / Nuclear Physics B 623 (2002) 4772
mediated models, on the contrary, SUSY is broken in a hidden sector and transmitted to
the visible sector via SM gauge interactions of messenger particles.
There is a third scenario, called anomaly-mediated SUSY breaking [3], which is
based on the observation that the super-Weyl anomaly gives rise to loop contribution to
sparticle masses. The anomaly contributions are always present and in some cases they
can dominate; this is the anomaly mediated supersymmetry breaking (AMSB) scenario.
In this way, the gaugino masses are proportional to their corresponding gauge group
-functions with the lightest SUSY particle being mainly wino. Analogously, the scalar
masses and trilinear couplings are functions of gauge and Yukawa -functions. Without
further contributions the slepton squared masses turn out to be negative. This tachyonic
spectrum is usually cured by adding an universal non-anomaly mediated contribution
m20 > 0 to every scalar mass [4].
So far, most of the work on AMSB has been done assuming R-parity (RP )
conservation [57]; see [8] for an exception. R-parity violation [9] has received quite some
attention lately motivated by the Super-Kamiokande Collaboration results on neutrino
oscillations [10], which indicate neutrinos have mass [11]. One way of introducing mass
to the neutrinos is via bilinear R-parity violation (BRpV) [12], which is a simple and
predictive model for the neutrino masses and mixing angles [13,14]. In this work, we
study the phenomenology of an anomaly mediated SUSY breaking model which includes
bilinear R-parity violation (AMSB-BRpV), stressing its differences to the R-parity
conserving case.
In BRpV-MSSM [15], bilinear R-parity and lepton number violating terms are
introduced explicitly in the superpotential. These terms induce vacuum expectation values
(vevs) vi for the sneutrinos, and neutrino masses through mixing with neutralinos. At tree
level, only one neutrino acquires a mass [16], which is proportional to the sneutrino vev
in a basis where the bilinear R-parity violating terms are removed from the superpotential.
At one-loop, three neutrinos get a non-zero mass, producing a hierarchical neutrino mass
spectrum [17]. It has been shown that the atmospheric mass scale, given by the heaviest
neutrino mass, is determined by tree level physics and that the solar mass scale, given by
the second heaviest neutrino mass, is determined by one-loop corrections [14].
In our model, the presence of RP violating interactions gives rise to neutrino masses
which we show to be consistent with the present observations. Moreover, the lowenergy phenomenology is quite distinct of the conserving R-parity AMSB scenario. For
instance, the lightest supersymmetric particle (LSP) is unstable, which allows regions of
the parameter space where the stau or the tau-sneutrino is the LSP. In our scenario, decays
can proceed via the mixing between the standard model particles and supersymmetric ones.
As an example, the mixing between the lightest neutralino 10 (chargino 1 ) and ( )
allows the following decays
10 Z ,
10 W ,
1 Z ,
1 W .
49
Another effect of the mixing between the standard model and supersymmetric particles
is a sizeable change in the mass of the supersymmetric particles. For instance, the mixing
between scalar taus and the charged Higgs can lead to an increase in the splitting between
the two scalar tau mass eigenstates by a factor that can be as large as 10 with respect to the
RP conserving case.
This paper is organized as follows. We define in Section 2 our anomaly mediated SUSY
breaking model which includes bilinear R-parity violation, stating explicitly our working
hypotheses. This section also contains an overall view of the supersymmetric spectrum in
our model. We study the properties of the CP-odd, CP-even, and charged scalar particles
in Sections 3, 4, and 5, respectively, concentrating on the mixing angles that arise from
the introduction of the R-parity violating terms. Section 6 contains the analysis that shows
that our model can generate neutrino masses in agreement with the present knowledge. In
Section 7 we provide a discussion of the general phenomenological aspects of our model
while in Section 8 we draw our conclusions.
2. The AMSB-BRpV model

Our model, besides the usual RP conserving Yukawa terms in the superpotential,
includes the following bilinear terms

a H
b
a b
Wbilinear = ab H
(1)
d u + i Li Hu ,
where the second one violates RP and we take |i | ||. Analogously, the relevant soft
bilinear terms are
a
Vsoft = m2Hu Hua Hua + m2Hd HdaHda + ML2 i L a
i Li

ab BHda Hub + Bi i L ai Hub ,
(2)
where the terms proportional to Bi are the ones that violates RP . The explicit RP violating
terms induce vacuum expectation values vi , i = 1, 2, 3 for the sneutrinos, in addition to the
two Higgs doublets vevs vu and vd . In phenomenological studies where the details of the
neutrino sector are not relevant, it has been proven very useful to work in the approximation
where RP and lepton number are violated in only one generation [18]. In these cases, a
determination of the mass scale of the atmospheric neutrino anomaly within a factor of
two is usually enough, and that can be achieved in the approximation where RP is violated
only in the third generation.
In this work we assume that RP violation takes place only in the third generation, and
consequently the parameter space of our model is
m0 , m3/2, tan , sign(), 3 , and m ,
(3)
where m3/2 is the gravitino mass and m20 is the non-anomaly mediated contribution to the
soft masses needed to avoid the appearance of tachyons. We characterize the BRpV sector
by the 3 term in the superpotential and the tauneutrino mass m since it is convenient to
trade v3 by m .
50
In AMSB models, the soft terms are fixed in a non-universal way at the unification
scale which we assumed to be MGUT = 2.4 1016 GeV; see Appendix A for details. We
considered the running of the masses and couplings to the electroweak scale, assumed to be
the top mass, using the one-loop renormalization group equations (RGE) that are presented
in Appendix B. In the evaluation of the gaugino masses, we included the next-to-leading
order (NLO) corrections coming from s , the two-loop top Yukawa contributions to the
beta-functions, and threshold corrections enhanced by large logarithms; for details see [4].
The NLO corrections are especially important for M2 , leading to a change in the wino
mass by more than 20%.
One of the virtues of AMSB models is that the SU(2) U (1) symmetry is broken
radiatively by the running of the RGE from the GUT scale to the weak one. This feature
is preserved by our model since the one-loop RGE are not affected by the bilinear RP
violating interactions; see Appendix B. In our model, the electroweak symmetry is broken
by the vacuum expectation values of the two Higgs doublets Hd and Hu , and the neutral
component of the third left slepton doublet L 3 . We denote these fields as

1 0

d + vd + id0
Hu+
2

Hd =
,
Hu = 1
,
0 + vu + iu0
Hd
2 u

1 R
+ v3 + i i0
2
.
L 3 =
(4)

The above vevs vi can be obtained through the minimization conditions, or tadpole
equations, which in the AMSB-BRpV model are

1
td0 = m2Hd + 2 vd Bvu 3 v3 + g 2 + g 2 vd vd2 vu2 + v32 ,
8

2
1
0
2
2
tu = mHu + + 3 vu Bvd + B3 3 v3 g 2 + g 2 vu vd2 vu2 + v32 ,
8
2

1 2
0
2
t3 = mL3 + 3 v3 3 vd + B3 3 vu + g + g 2 v3 vd2 vu2 + v32 ,
8
(5)
at tree level. At the minimum we must impose td0 = tu0 = t30 = 0. In practice, the input
parameters are the soft masses mHd , mHu , and mL3 , the vevs vu , vd , and v3 (obtained
from mZ , tan , and m ), and 3 . We then use the tadpole equations to determine B, B3 ,
and ||.
One-loop corrections to the tadpole equations change the value of || by O(20%),
therefore, we also included the one-loop corrections due to third generation of quarks and
squarks [17]:
ti = ti0 + Ti (Q),
(6)
i (Q) are
where ti , with i = d, u, are the renormalized tadpoles, are given in (5), and T
the renormalized one-loop contributions at the scale Q. Here we neglected the one-loop
corrections for t3 since we are only interested in the value of .
Using the procedure underlined above, the whole mass spectrum can be calculated
as a function of the input parameters m0 , m3/2 , tan , sign(), 3 , and m . In Fig. 1,
we show a scatter plot of the mass spectrum as a function of the scalar mass m0 for
ti0
51
Fig. 1. Supersymmetric mass spectrum in AMSB-BRpV for m3/2 = 32 TeV, tan = 5, and < 0. The values of
3 and m were randomly varied according to 105 < 3 < 1 GeV and 106 < m < 1 eV.
m3/2 = 32 TeV, tan = 5, and < 0, varying 3 and m according to 105 < 3 < 1 GeV
and 106 < m < 1 eV. The widths of the scatter plots show that the spectrum exhibits a
very small dependence on 3 and m . Throughout this paper we use this range for 3 and
m in all figures.
We can see from this figure that, for m0 200 GeV, the LSP is the lightest neutralino
0
1 with the lightest chargino 1+ almost degenerated with it, as in RP -conserving
AMSB. Nevertheless, the LSP is the lightest stau 1+ for m0 200 GeV. This last
region of parameter space is forbidden in RP -conserving AMSB, but perfectly possible in
AMSB-BRpV since the stau is unstable, decaying into RP -violating modes with sizeable
branching ratios. Furthermore, the slepton masses have a strong dependence on m0 . We
plotted masses of the two staus, which have an appreciable splitting, the almost degenerated
smuons, and the closely degenerated tau-sneutrinos.1 The heavy Higgs bosons have also a
strong dependence on m0 and, for the chosen parameters, they are much heavier than the
sleptons. On the other hand, the gauginos show little dependence on m0 , as expected.
1 In fact, there are two tau-sneutrinos in this model, a CP-even and a CP-odd field that are almost degenerated;
see further sections for details.
52
Bounds on BRpV parameters depend in general on supersymmetric masses and

couplings, as shown in [19]. In models with BRpV in only one generation it is possible
to estimate the bound on 3 in a much simpler way: if we rotate the lepton and Higgs fields
such that the bilinear term in the superpotential is eliminated [20], a trilinear term is
generated
3ii = hdi
3
(7)
2 + 32
where hdi is the Yukawa coupling of the down quark of the ith generation. Bounds on these
couplings can be found on [9]:
311 < 0.11
mdR
100 GeV
322 < 0.52
msR
,
100 GeV
333 < 0.45,
(8)
and, considering the values of the Yukawa couplings, it is easy to see that these bounds are
satisfied for our choice 3 < 1 GeV.
3. CP-odd Higgs/sneutrino sector

In our model, the CP-odd Higgs sector mixes with the imaginary part of the tausneutrino due to the bilinear RP violating interactions. Writing the mass terms in the form
0
1 0 0 i0 2 d0
Vquadratic = d , u , M P 0 u ,
(9)
2
i0
we have
M 2P 0
m2(0) s 2 +
2(0)
v3
3 vd
2(0)
mA s c
3
3
mA s c
2(0)
mA c2 3 vvd3
c
3 s +
c2
s2
2
v32 c 2
m
2
vd s2
v3 c 2

vd s m
3 s +
v3 c 2

vd s m
m
2
(10)
with m
2 = m + 32 + 18 gZ2 v32 and gZ2 g 2 + g 2 . Here,
2(0)
m2(0)
A =
B
s c

1 2 2
2
2
and m2(0)
= ML3 + 8 gZ vd vu
(11)
are, respectively, the CP-odd Higgs and sneutrino masses in the RP conserving limit
(3 = v3 = 0). In order to write this mass matrix we have eliminated m2Hu , m2Hd , and B3
using the tadpole equations (5). The mass matrix has an explicitly vanishing eigenvalue,
which corresponds to the neutral Goldstone boson.
This matrix can be diagonalized with a rotation

0
0
d
A
G0 = R P 0 u0 ,
odd
i0
53
(12)
G0
is the massless neutral Goldstone boson. Between the other two eigenstates, the
where
one with largest i0 component is called CP-odd tau-sneutrino odd and the remaining state
is called CP-odd Higgs A0 .
As an intermediate step, it is convenient to make explicit the masslessness of the
Goldstone boson with the rotation
s
c
0
P 0 = c r
s r
vv3 c r ,
R
(13)
d
vvd3 c2 r
v3
vd s c r
where
r=
1
1+
v32 2
c
vd2
(14)
P 0 M 2 0 R
T0 which has a column and a row of zeros,
obtaining a rotated mass matrix R
P
P
0
corresponding to G . This procedure simplifies the analysis since the remaining 2 2
mass matrix for (A0 , odd ) is
2

2
2
2(0) v 2 c4
v3 c 2
v3 s c
1
2
3
mA + 2 2 m
+ 3 vd 2
3 s r
vd s m
vd s
s
2P 0 =

2
M
(15)
.
v3 c 2
1
1
2
r
m

3 s
r 2
vd s
We quantify the mixing between the tau-sneutrino and the neutral Higgs bosons through
2
2

sin2 odd = odd u0 + odd d0 .
(16)
If we consider the RP violating interactions as a perturbation, we can show that
2
v3 2 2(0)
v32 2
vd c m 3
2
sin odd 2(0)
+
c ,

2
vd2
s2 m
m2(0)
(17)
indicating that this mixing can be large when the CP-odd Higgs boson A0 and the sneutrino
are approximately degenerate.
Fig. 2(a) displays the full sneutrino-Higgs mixing (16), with no approximations, as a
function of tan for m3/2 = 32 TeV, < 0 and 100 < m0 < 300 GeV. In a large fraction
of the parameter space this mixing is small, since it is proportional to the BRpV parameters
squared divided by MSSM mass parameters squared. However, it is possible to find a
region where the mixing is sizable, e.g., for our choice of parameters this happens at
tan 15. As expected, the region of large mixing is associated to near degenerate states,
as we can see from Fig. 2(b) where we present the ratio between the CP-odd Higgs mass
mA and the CP-odd tau-sneutrino mass m odd as a function of tan .
54
Fig. 2. (a) CP-odd Higgssneutrino mixing and (b) ratio between the CP-odd Higgs mass and the sneutrino mass
as a function of tan for m3/2 = 32 TeV, < 0 and 100 < m0 < 300 GeV.
4. CP-even Higgs/sneutrino sector

The mass terms of the CP-even neutral scalar sector are
0
d

1
Vquadratic = d0 , u0 , r0 M 2S 0 u0 ,
2
r0
(18)
where the mass matrix can be separated into two pieces

+ M 2(1)
.
M 2S 0 = M 2(0)
S0
S0
(19)
The first term due to RP conserving interactions is
2(0)
mA s2 + 14 gZ2 vd2
2(0)
1 2
M S 0 = m2(0)
A s c 4 gZ vd vu
0
2(0)
mA s c 14 gZ2 vd vu
1 2 2
2
m2(0)
A c + 4 gZ vu
0 ,
m2(0)
(20)
55
while the one associated to the RP violating terms is

M 2(1)
S0
3 vvd3
0
2v v
3 + 14 gZ
d 3
2v v
3 + 14 gZ
d 3
0
2
c2
v32 c 2(0)
m 3 vvd3 2
2
2
vd s
s
c
c
2(0)
2v v
3 s vvd3 s m 14 gZ
u 3
2(0)
3 s vvd3 s m
2v v
14 gZ
u 3
2 v2
32 + 38 gZ
3
(21)
Fig. 3. (a) CP-even Higgssneutrino mixing; (b) ratio between heavy CP-even Higgs and tau-sneutrino masses
and (c) ratio between light CP-even Higgs and tau-sneutrino masses as a function of tan for m3/2 = 32 TeV,
< 0 and 100 < m0 < 300 GeV.
56
Radiative corrections can change significantly the lightest Higgs mass and, consequently,
we have also introduced the leading correction to its mass

mt1 mt2
3m4t
ln
/mu0
(22)
,
2 2 vu2 v
m2t
with
v = 1
v32
vd2 + vu2 + v32
(23)
by adding it to the element [M 2S 0 ]22 .

Analogously to the CP-odd sector, we define the mixing between the CP-even tausneutrino and the neutral Higgs bosons as

2
2 2 2
sin2 even = even d0 + even u0 = H 0 r0 + h0 r0 .
(24)
In general, this mixing is small since it is proportional to the RP breaking parameters
squared, however, it can be large provided the sneutrino is degenerate either with h0 or H 0 .
In Fig. 3(a), we present the mixing (24) as a function of tan , for the input parameters
as in Fig. 2. Similarly to the CP-odd scalar sector, this mixing can be very large, occurring
either when mH m even or mh m even . In fact, we can see from Fig. 3(b) that the
peak in Fig. 3(a) for tan 15 is mainly due to the mass degeneracy between the heavy
CP-even Higgs H 0 and the CP-even tau-sneutrino even . On the other hand, the other
scattered dots with high mixing angle values throughout Fig. 3(a) come from points in
the parameter space where the light CP-even Higgs h0 and the CP-even tau-sneutrino even
are degenerated. We see from Fig. 3(c) that this may occur for 5 < tan < 15.
It is important to notice that the enhancement of the mixing between the tau-sneutrino
and the CP-even Higgs bosons for almost degenerate states implies that large RP violating
effects are possible even for small RP violating parameters (3 1 GeV), and for neutrino
masses consistent with the solutions to the atmospheric neutrino anomaly (m 1 eV).
5. Charged Higgs/charged slepton sector

The mass terms in the charged scalar sector are
H+

Vquadratic = Hu , Hd , L , R
Hd+
M 2S
+ ,
(25)
L
R+
where it is convenient to split the mass matrix into a RP conserving part and a RP violating
one
M 2S = M 2(0)
+ M 2(1)
.
S
S
The RP conserving mass matrix has the usual MSSM form
(26)
57
2(0)
M S
2(0)
2(0)
mA s2 + 14 g 2 vu2
1 2
m2(0)
A s c + 4 g vu vd
mA s c + 14 g 2 vu vd
2(0)
mA c2 + 14 g 2 vd2
2
M
L
,
1 h (A vd vu )
2
1 h (A vd vu )
2
2
M
R
(27)
where h is the Yukawa coupling and

L2 = ML2 1 g 2 g 2 vd2 vu2 + 1 h2 vd2 ,
M
3
3
8
2

1
1
R2 = MR2 g 2 vd2 vu2 + h2 vd2 .
M
3
3
4
2
The contribution due to RP violating terms is
(28)
2(1)
M S

v3
1 2 2
1 2 2
3 vd 4 g v3 + 2 h v3
0
2
v32 c
vd2 s2
XuL
2
v c
+ 14 g 2 v32
d s2
m
2 3 v3
XdL
XdR
XuR
XuL
XuR
XdL
XdR
2 v2
32 + 18 gZ
3
0
1 h2 v 2 1 g 2 v 2
3
2 3
4
(29)
with
1
1
XuL = g 2 vd v3 3 h2 vd v3 ,
4
2
1
XuR = h (A v3 + 3 vu ),
2
c
v3 c 2
1
XdL =
m
3 + g 2 vu v3 ,
vd s
s
4
1
XdR = h (v3 + 3 vd ).
2
(30)
(31)
(32)
(33)
The complete matrix M 2S has an explicit zero eigenvalue corresponding to the charged
Goldstone boson G , and is diagonalized by a rotation matrix RS such that
H+
H+
u
+
G+
+ = R S Hd .
+

1
2+
(34)
L
R+
In analogy with the discussion on the CP-even scalar sector, we define the mixing of the
lightest (heaviest) stau 1 (2 ) with the charged Higgs bosons as

2
2
sin2 1+ = 1+ Hu+ + 1+ Hd+ ,
(35)

2
2
sin2 2+ = 2+ Hu+ + 2+ Hd+ .
(36)
58
Fig. 4(a), (b) contains the mixing between the lightest (heaviest) stau and the charged
+
as a function of tan for m3/2 = 32 TeV, < 0, and 100 <
Higgs fields sin 1(2)
m0 < 300 GeV. In this sector, the mixing can also be very large provided there is a
near degeneracy between the staus 1 , 2 and H . We can see clearly this effect in
Fig. 4(c), (d), where we show the ratio between the charged Higgs mass mH + and the
lightest (heaviest) stau mass m1(2) . In Fig. 4(a) and (b) we also notice that large light stau
charged Higgs mixing occurs at slight different value of tan compared with heavy stau
charged Higgs mixing. Large light staucharged Higgs mixing is found in Fig. 4(a) as a
peak at tan 16, as opposed to large heavy staucharged Higgs mixing, which presents a
peak at tan 15. In Fig. 4(a) we notice that the mixing angle vanishes at tan 11. This
zero occurs at the point of parameter space where the two staus are nearly degenerated, as
will be explained in Section 7.
Similarly, in the last figure, the exact value of tan at which the peak of the lightest staucharged scalar mixing occurs is somewhat larger than the analogous mixing for the CP-odd
sector sin odd . This can be appreciated in Fig. 5(a) where we show the ratio between sin 1+
and sin odd as a function of tan for m3/2 = 32 TeV, < 0 and 100 < m0 < 300 GeV.
Fig. 4. (a) Charged Higgslight stau mixing; (b) charged Higgsheavy stau mixing; (c) charged Higgslight stau
mass ratio and (d) charged Higgsheavy stau mass ratio as a function of tan for m3/2 = 32 TeV, < 0 and
100 < m0 < 300 GeV.
59
Fig. 5. (a) Ratio between the charged Higgsstau and CP-odd Higgstau-sneutrino mixing angles and (b) ratio
between the CP-odd Higgstau-sneutrino and CP-even Higgstau-sneutrino mixing angles as a function of tan
for m3/2 = 32 TeV, < 0 and 100 < m0 < 300 GeV.
The peak of the charged sector mixing is located at the peak of the ratio. On the other hand,
the peak for the neutral CP-odd sector is located at the nearby zero of the ratio. The other
zero of the ratio near tan 11 corresponds to a zero of the charged scalar sector mixing,
as shown in Fig. 4. For the sake of comparison, we display in Fig. 5(b) the ratio between
the CP-odd and CP-even mixings (sin odd/ sin even ) as a function of tan . We can see
that most of the time the ratio is equal to 1 showing that the two neutral scalar sectors have
similar behavior with tan in contrast with the charged scalar sector. The points where this
ratio is lower than 1 correspond to the case where the CP-even scalar sector mixings are
dominated by the light Higgs and tau-sneutrino degeneracy which occurs for any value of
tan lower than 16, as shown in Fig. 3(c).
6. The neutrino mass

BRpV provides a solution to the atmospheric and solar neutrino problems due to their
mixing with neutralinos, which generates neutrino masses and mixing angles. It was shown
in [14] that the atmospheric mass scale is adequately described by the tree level neutrino
60
mass
mtree
3 =
M1 g 2 + M2 g 2
2,
||
40
(37)
where 0 is the determinant of the neutralino sub-matrix and = (1 , 2 , 3 ), with

i = vi + i vd ,
(38)
where the index i refers to the lepton family. The spectrum generated is hierarchical, and
obtained typically with 1 2 3 .
As it was mentioned in the introduction, for many purposes it is enough to work with
RP violation only in the third generation. In this case, the atmospheric mass scale is well
2 2 . In Fig. 6, we plot the neutrino
described by Eq. (37) with the replacement ||
3
mass as a function of in AMSB-BRpV with the input parameters m3/2 = 32 TeV, < 0,
5 < tan < 20, 100 < m0 < 1000 GeV and 105 < 3 < 1 GeV. The quadratic dependence
of the neutrino mass on is apparent in this figure and neutrino masses smaller than 1 eV
occur for || 0.6 GeV2 . Moreover, the stars correspond to the allowed neutrino masses
when the tau-sneutrino is the LSP. In general the points with a small (large) m0 are located
in the inner (outer) regions of this scattered plot.
From Fig. 6, we can see that the attainable neutrino masses are consistent with the
global three-neutrino oscillation data analysis in the first reference of [10] that favors the
oscillation hypothesis. Although only mass squared differences are constrained
by the neutrino data, our model naturally gives a hierarchical neutrino mass spectrum,
therefore, we extract a nave constraint on the actual mass coming from the analysis of the
full atmospheric neutrino data, 0.04 m 0.09 eV [10]. In addition, we notice that it
Fig. 6. Tau neutrino mass as a function of 3 for 5 < tan < 20, 100 < m0 < 1000 GeV, m3/2 = 32 TeV and
< 0.
61
Fig. 7. Mixing between CP-even Higgses and sneutrino as a function of the tau neutrino mass.
is not possible to find an upper bound on the neutrino mass if angular dependence on the
neutrino data is not included and only the total event rates are considered.
In Fig. 7 we show the correlation between the neutrino mass and mixing of the tausneutrino and the CP-even Higgses (sin even ) for the parameters assumed in Fig. 6. As
expected, the largest mixings are associated to larger neutrino masses. Notwithstanding, it
is possible to obtain large mixings for rather small neutrino masses because the mixing is
proportional to the RP violating parameters 3 and v3 , and not directly on 3 m . In
any case, Fig. 7 suggests that large scalar mixings are still possible even imposing these
bounds on the neutrino mass. This is extremely important for the phenomenology of the
model because it indicates that non negligible RP violating branching ratios are possible
for scalars even in the case they are not the LSP.
7. Discussions
The presence of RP violating interactions in our model render the LSP unstable,
avoiding strong constraints on the possible LSP candidates. In the parameter regions
where the neutralino is not the LSP, whether the light stau or the tau-sneutrino is the
LSP depends crucially on the value of tan . This fact can be seen in Fig. 8 where we
plot the ratio between the light stau and the tau-sneutrino masses as a function of tan
for m3/2 = 32 TeV, 100 < m0 < 300 GeV, and < 0. From this figure we see that the
tau-sneutrino is the LSP for 8.5 tan 14, otherwise the stau is the LSP.2
2 Of course, m has to be small enough so that the slepton is lighter than the neutralino.
0
62
Fig. 8. Ratio between the light stau and the sneutrino masses as a function of tan for m3/2 = 32 TeV,
100 < m0 < 300 GeV and < 0.
When the stau is the LSP, it decays via RP violating interactions, i.e., its decays take
place through mixing with the charged Higgs, and consequently, they will mimic the
charged Higgs boson ones. Therefore, it is very important to be able to distinguish between
1 and H . This can be achieved either through precise studies of branching ratios, or via
the mass spectrum, or both [21].
Measurements on the mass spectrum are also important in order to distinguish AMSB
with and without conservation of RP . In Fig. 9 we present the ratio between the stau
mass splitting in AMSB-BRpV and in the AMSB, R = (m2 m1 )AMSBBRpV /(m2
m1 )AMSB , with 3 = v3 = 0 and keeping the rest of the parameters unchanged, as a
function of tan . In these figures, we took 100 < m0 < 1000 GeV, m3/2 = 32 TeV, and
(a) > 0, and (b) < 0. For > 0 (Fig. 9(a)), the stau mass splitting is always larger in
the AMSB-BRpV than in the AMSB by a factor that increases when tan decreases, and
can be as large as R 10 for tan 3! We remind the reader that, in the absence of RP
violation, the leftright stau mixing decreases with decreasing tan , thus augmenting the
importance of R-parity violating mixings. On the other hand, for < 0 (Fig. 9(b)), this
ratio can be as large as before at small tan , but in addition, the splitting can go to zero in
AMSB-BRpV near tan 11, which also constitutes a sharp difference with the AMSB.
For both signs of the ratio goes to unity at large tan because the leftright mixing in
the AMSB is proportional to tan and dominates over any RP violating contribution.
The behavior of R at tan 11 in Fig. 9(b) indicates that the two staus can be nearly
degenerated in AMSB-BRpV. In Fig. 10 we plot the ratio between the light and heavy
stau masses as a function of tan , for m3/2 = 32 TeV, 100 < m0 < 300 GeV and < 0,
observing clearly that the near degeneracy occurs at tan 11. In first approximation,
consider that the near degeneracy occurs when A vd vu 0 as inferred from Eq. (27).
In addition, the mixing XdR in Eq. (33) is also very small because it is proportional to
63
Fig. 9. Ratio (R) between the stau splitting in AMSB with and without RP violation as a function of tan , for:
m3/2 = 32 TeV, 100 < m0 < 1000 GeV and (a) > 0 or (b) < 0.
Fig. 10. Ratio between the light and heavy stau masses as a function of tan for m3/2 = 32 TeV,
100 < m0 < 300 GeV and < 0.
64
in Eq. (38), which defines the atmospheric neutrino mass, as indicated in Eq. (37). The
smallness of these two quantities implies that the mixing XuR in Eq. (31) is also small in
this particular region of parameter space, indicating that the right stau is decoupled from
the Higgs fields and thus originating the zero in the mixing angle, noted already in Figs. 4
and 5.
In order to quantify the stau mass splitting in our model, we present in Fig. 11 contours
of constant splitting between the stau masses, m2 m1 , in the plane m3/2 m0 in GeV
for < 0 and several tan . We can see in Fig. 11(a) that for small tan = 3 the stau
mass splitting in our model starts at m2 m1 30 GeV, in sharp contrast with the RP
conserving case where the biggest splittings barely goes over this value [7]. This is in
agreement with the results presented in Fig. 9(b). Furthermore, we can also see that there
is a considerable region in the m3/2 m0 plane, indicated by the grey area, where the
Fig. 11. Contours of constant splitting between the light stau and heavy stau masses in the plane m3/2 m0 in
GeV for < 0, tan = 3 (a), 15 (b) and 30 (c). The hatched area is theoretically forbidden; the grey area in (a)
and (b) is where the lightest stau is the LSP, while the small black area in (b) is where the tau-sneutrino is the
LSP.
65
lightest stau is the LSP. For intermediary values of tan 15, Fig. 11(b) shows that the
stau mass splitting goes to a minimum. This is a different behavior from the MSSM which
presents a mass splitting up to 10 times bigger as we have seen in Fig. 9(b). For this value
of tan we still have a small region where the lightest stau is the LSP (grey area) and, as a
novelty, a tiny region for small values of m3/2 and m0 where the tau-sneutrino is the LSP
(black area). For large values of tan = 30, the stau splitting mass shown in Fig. 11(c) is
similar to the MSSM one [7].
We have made below a series of three figures fixing the value tan = 15 to study the
dependence on m0 of the mass spectrum and mixings in the scalar sector. This choice of
tan is such that we find a degeneracy among the masses, and consequently we obtain
large mixings in the scalar sector. We also chose m3/2 = 32 TeV and < 0, while the RP
violating parameters were varied according to 105 < 3 < 1 GeV and 106 < m < 1 eV.
In Fig. 12(a) we plot tau-sneutrino mixing with the CP-odd neutral Higgs as a function
of m0 for the parameters indicated above. We find quite large mixings for m0 320 GeV.
In Fig. 12(b) we show the CP-odd Higgs and tau-sneutrino masses, which depend almost
linearly on m0 . Moreover, the value of m0 at which these two particles have the same mass
coincides with the point of maximum mixing.
The CP-even tau-sneutrino mixing with the CP-even Higgs is presented in Fig. 13(a) as
a function of m0 . There are two peaks of high mixing; the main one at m0 320 GeV and
Fig. 12. (a) Mixing of the CP-odd Higgs and the sneutrino and (b) the CP-odd Higgs and sneutrino masses as a
function of m0 for m3/2 = 32 TeV, < 0 and tan = 15.
66
Fig. 13. (a) Mixing between the CP-even Higgs and sneutrino and (b) the light and heavy CP-even Higgs masses
as well as the sneutrino one as a function of m0 for m3/2 = 32 TeV, < 0 and tan = 15.
a narrow one at m0 180 GeV. These two peaks have a different origin, as indicated by
Fig. 13(b), where we plot the masses of the two CP-even neutral Higgs bosons, mh and mH ,
and the mass of the CP-even tau-sneutrino m even , as a function of m0 . We observe that the
broad peak is due to a degeneracy between the tau-sneutrino and the heavy neutral Higgs
boson and the narrow peak comes from a degeneracy between the tau-sneutrino and the
light neutral Higgs boson. As expected, the H 0 and even masses grow linearly with m0 ,
contrary to the h0 mass which remains almost constant.
In Fig. 14(a) we display the light stau mixing with the charged Higgs as a function of
m0 . The maximum mixing, obtained at m0 550 GeV, is the result of a mass degeneracy
between the charged Higgs boson and the light stau. This can be observed in Fig. 14(b)
where we plot the charged Higgs mass mH and the light stau mass m1 as a function
of m0 .
In a similar way, we show the heavy stau mixing with charged Higgs as a function of
m0 in Fig. 14(c), where we observe a maximum for the mixing at m0 200 GeV. This
large mixing is due to a degeneracy between the charged Higgs boson and the heavy stau
masses, as can be seen in Fig. 14(d). One can notice that all charged scalars show an almost
linear dependency of their mass on the mass parameter m0 .
67
Fig. 14. (a) Mixing of the charged Higgs with the light stau, (b) charged Higgs and light stau masses, (c) mixing
of the charged Higgs with the heavy stau, and (d) charged Higgs and heavy stau masses as a function of m0 for
m3/2 = 32 TeV, < 0 and tan = 15.
As opposed to the scalar sector, where mixing between the Higgs bosons and sleptons
can be maximum, in the chargino and neutralino sectors the mixings with leptons are
controlled by the neutrino mass being very small. Despite this fact, the mixing in the
neutralino sector is sufficient to generate adequate masses for the neutrinos and give rise
to the neutralino decays mentioned in the introduction. Therefore, in the chargino sector
the BRpV-AMSB phenomenology changes very little with respect to the RP conserving
AMSB. One of the distinctive features of AMSB that differentiates it from other scenarios
of supersymmetry breaking in the chargino-neutralino sector is the near degeneracy
between the lightest chargino and the lightest neutralino. This feature remains in BRpVAMSB as was anticipated in Fig. 1. For m3/2 = 32 TeV, < 0, and 100 < m0 < 300 GeV,
we show in Fig. 15 the lightest chargino mass as a function of tan . The lightest chargino
mass has a small dependence on tan since its value varies only between 100 and 104 GeV.
As in RP conserving AMSB, the mass difference m + m 0 remains small.
1
68
Fig. 15. Light chargino mass as a function of tan for m3/2 = 32 TeV, < 0 and 100 < m0 < 300 GeV.
8. Conclusions
We have shown in the previous sections that our model exhibiting anomaly mediated
supersymmetry breaking and bilinear RP violation is phenomenologically viable. In
particular, the inclusion of BRpV generates neutrino masses and mixings in a natural
way. Moreover, the RP breaking terms give rise to mixing between the Higgs bosons and
the sleptons, which can be rather large despite the smallness of the parameters needed to
generate realistic neutrino masses. These large mixings occur in regions of the parameter
space where two states are nearly degenerate. Our model also alters substantially the mass
splitting between the scalar taus in a large range of tan .
The RP violating interactions render the LSP unstable since it can decay via its mixing
with the SM particles (leptons or scalars). Therefore, the constraints on the LSP are relaxed
and forbidden regions of parameter space become allowed, where scalar particles like
staus or sneutrinos are the LSP. Furthermore, the large mixing between Higgs bosons
and sleptons has the potential to change the decays of these particles. These facts have
a profound impact in the phenomenology of the model, changing drastically the signals at
colliders [22].
Acknowledgements
This work was partially supported by a joint grant from Fundacin Andes and
Fundacin Vitae, by DIPUC, by CONICYT grant No. 1010974, by Fundao de Amparo
Pesquisa do Estado de So Paulo (FAPESP), by Conselho Nacional de Desenvolvimento
Cientfico e Tecnolgico (CNPq) and by Programa de Apoio a Ncleos de Excelncia
(PRONEX).
69
Appendix A. AMSB boundary conditions

The AMSB boundary conditions at the GUT scale for the gaugino masses are
proportional to their beta functions, resulting in
33 g12
m3/2,
5 16 2
g2
M2 = 2 2 m3/2,
16
g2
M3 = 3 3 2 m3/2 ,
16
while the third generation scalar masses are given by

2
88
0
m2U = g14 + 8g34 + 2ft ft
16 2 + m20 ,
2
25
m3/2

m23/2
22
m2D = g14 + 8g34 + 2fb fb
+ m20 ,
25
(16 2 )2

m23/2
11 4 3 4
2
4
+ m20 ,
mQ = g1 g2 + 8g3 + ft ft + fb fb
50
2
(16 2 )2

m23/2
99 4 3 4
2
+ m20 ,
mL = g1 g2 + f f
50
2
(16 2)2

m23/2
198 4
2
mE =
+ m20 ,
g1 + 2f f
25
(16 2 )2

m23/2
99 4 3 4
2
+ m20 ,
mHu = g1 g2 + 3ft ft
50
2
(16 2 )2

m23/2
99 4 3 4
2
+ m20 .
mHd = g1 g2 + 3fb fb + f f
50
2
(16 2)2
M1 =
(A.1)
(A.2)
(A.3)
(A.4)
(A.5)
(A.6)
(A.7)
(A.8)
(A.9)
(A.10)
Finally, the A-parameters are given by

At =
ft m3/2
,
ft 16 2
Ab =
fb m3/2
,
fb 16 2
A =
f m3/2
,
f 16 2
(A.11)
where we have defined

16
13
ft = 16 2t = ft g12 3g22 g32 + 6ft2 + fb2 ,
15
3

16 2
7 2
2
2
2
2
2
fb = 16 b = fb g1 3g2 g3 + ft + 6fb + f ,
15
3

9 2
2
2
2
2
f = 16 = f g1 3g2 + 3fb + 4f .
5
(A.12)
(A.13)
(A.14)
70
Appendix B. The renormalization group equations

Here we present the one-loop renormalization group equations for our model, assuming
the bilinear RP breaking terms are restricted only to the third generation. First, we display
the equations for the Yukawa couplings of the trilinear terms

16 2
13 2
2 dhU
2
2
2
= hU 6hU + hD g3 3g2 g1 ,
16
(B.1)
dt
3
9

dhD
16
7
= hD 6h2D + h2U + h2 g32 3g22 g12 ,
16 2
(B.2)
dt
3
9

dh
= h 4h2 + 3h2D 3g22 3g12 .
16 2
(B.3)
dt
The corresponding RGE for cubic soft supersymmetry breaking parameters are given by
dAU
16
13
= 6h2U AU + h2D AD + g32 M3 + 3g22 M2 + g12 M1 ,
dt
3
9
16 2
7
2 dAD
2
2
2
= 6hD AD + hU AU + h A + g3 M3 + 3g22 M2 + g12 M1 ,
8
dt
3
9
2 dA
2
2
2
2
= 4h A + 3hD AD + 3g2 M2 + 3g1 M1 .
8
dt
For the soft supersymmetry breaking mass parameters we have
8 2
8 2
2
dMQ
dt
(B.4)
(B.5)
(B.6)

2
2
2
= h2U m2H2 + MQ
+ MU2 + A2U + h2D m2H1 + MQ
+ MD
+ A2D
1
1
16 2 2
g3 M3 3g22 M22 g12 M12 + g12 S,
(B.7)
3
9
6

16
dMU2
16
2
2
= 2h2U m2H2 + MQ
8 2
+ MU2 + A2U g32 M32 g12 M12 g12 S,
dt
3
9
3
(B.8)
2

dM
16
4
1
D
2
2
= 2h2D m2H1 + MQ
8 2
+ MD
+ A2D g32 M32 g12 M12 + g12 S, (B.9)
dt
3
9
3
2

dM
1
L
= h2 m2H1 + ML2 + MR2 + A2 3g22 M22 g12 M12 g12 S,
8 2
(B.10)
dt
2

dMR2
= 2h2 m2H1 + ML2 + MR2 + A2 4g12 M12 + g12 S,
8 2
(B.11)
dt
dm2H2

1
2
= 3h2U m2H2 + MQ
8 2
+ MU2 + A2U 3g22 M22 g12 M12 + g12 S, (B.12)
dt
2
dm2H1

2
2
8 2
= 3h2D m2H1 + MQ
+ MD
+ A2D + h2 m2H1 + ML2 + MR2 + A2
dt
1
(B.13)
3g22 M22 g12 M12 g12 S,
2
where
2
2
S = m2H2 m2H1 + MQ
2MU2 + MD
ML2 + MR2 .
(B.14)
For the bilinear terms in the superpotential we get

d
= 3h2U + 3h2D + h2 3g22 g12 ,
16 2
dt

2 d3
= 3 3h2U + h2 3g22 g12 ,
16
dt
and for the corresponding soft breaking terms
71
(B.15)
(B.16)
dB
= 3h2U AU + 3h2D AD + h2 A + 3g22 M2 + g12 M1 ,
(B.17)
dt
dB2
= 3h2U AU + h2 A + 3g22 M2 + g12 M1 .
8 2
(B.18)
dt
The gi are the SU(3) SU(2) U (1) gauge couplings and the Mi are the corresponding
soft breaking gaugino masses.
8 2
References
[1] A. Chamseddine, R. Arnowitt, P. Nath, Phys. Rev. Lett. 49 (1982) 970;
R. Barbieri, S. Ferrara, C. Savoy, Phys. Lett. B 119 (1982) 343;
L.J. Hall, J. Lykken, S. Weinberg, Phys. Rev. D 27 (1983) 2359.
[2] M. Dine, A. Nelson, Phys. Rev. D 48 (1993) 1277;
M. Dine, A. Nelson, Y. Shirman, Phys. Rev. D 51 (1995) 1362;
M. Dine, A. Nelson, Y. Nir, Y. Shirman, Phys. Rev. D 53 (1996) 2658.
[3] L. Randall, R. Sundrum, Nucl. Phys. B 557 (1999) 79;
G. Giudice, M. Luty, H. Murayama, R. Rattazzi, JHEP 9812 (1998) 027.
[4] T. Gherghetta, G.F. Giudice, J.D. Wells, Nucl. Phys. B 559 (1999) 27.
[5] A. Pomarol, R. Rattazzi, JHEP 9905 (1999) 013;
Z. Chacko, M.A. Luty, I. Maksymyk, E. Ponton, JHEP 0004 (2000) 001;
E. Katz, Y. Shadmi, Y. Shirman, JHEP 9908 (1999) 015;
M.A. Luty, R. Sundrum, Phys. Rev. D 62 (2000) 035008;
J.A. Bagger, T. Moroi, E. Poppitz, JHEP 0004 (2000) 009;
I. Jack, D.R.T. Jones, Phys. Lett. B 482 (2000) 167;
M. Kawasaki, T. Watari, T. Yanagida, Phys. Rev. D 63 (2001) 083510.
[6] T. Moroi, L. Randall, Nucl. Phys. B 570 (2000) 455;
G.D. Kribs, Phys. Rev. D 62 (2000) 015008;
S. Su, Nucl. Phys. B 573 (2000) 87;
R. Rattazzi, A. Strumia, J.D. Wells, Nucl. Phys. B 576 (2000) 3;
M. Carena, K. Huitu, T. Kobayashi, Nucl. Phys. B 592 (2001) 164;
H. Baer, M.A. Daz, P. Quintana, X. Tata, JHEP 0004 (2000) 016;
D.K. Ghosh, P. Roy, S. Roy, JHEP 0008 (2000) 031;
U. Chattopadhyay, D.K. Ghosh, S. Roy, Phys. Rev. D 62 (2000) 115001;
H. Baer, J.K. Mizukoshi, X. Tata, Phys. Lett. B 488 (2000) 367.
[7] J.L. Feng, T. Moroi, Phys. Rev. D 61 (2000) 095004.
[8] B.C. Allanach, A. Dedes, JHEP 0006 (2000) 017.
[9] B. Allanach et al., hep-ph/9906224;
B. Allanach et al., J. Phys. G 24 (1998) 421.
[10] M.C. Gonzalez-Garca, M. Maltoni, C. Pea-Garay, J.W.F. Valle, Phys. Rev. D 63 (2001) 033005;
J.N. Bahcall, P.I. Krastev, A.Yu. Smirnov, JHEP 0105 (2001) 015.
[11] Y. Fukuda et al., Super-Kamiokande Collaboration, Phys. Rev. Lett. 81 (1998) 1562;
Y. Fukuda et al., Super-Kamiokande Collaboration, Phys. Lett. B 433 (1998) 9;
Y. Fukuda et al., Super-Kamiokande Collaboration, Phys. Lett. B 436 (1998) 33.
72
[12] F. de Campos, O.J.P. boli, M.A. Garca-Jareo, J.W.F. Valle, Nucl. Phys. B 546 (1999) 33;
R. Kitano, K. Oda, Phys. Rev. D 61 (2000) 113001;
D.E. Kaplan, A.E. Nelson, JHEP 0001 (2000) 033;
C.-H. Chang, T.-F. Feng, Eur. Phys. J. C 12 (2000) 137;
M. Frank, Phys. Rev. D 62 (2000) 015006;
F. Takayama, M. Yamaguchi, Phys. Lett. B 476 (2000) 116;
K. Choi, E.J. Chun, K. Hwang, Phys. Lett. B 488 (2000) 145;
J.M. Mira, E. Nardi, D.A. Restrepo, J.W.F. Valle, Phys. Lett. B 492 (2000) 81.
[13] R. Hempfling, Nucl. Phys. B 478 (1996) 3.
[14] J.C. Romo, M.A. Daz, M. Hirsch, W. Porod, J.W.F. Valle, Phys. Rev. D 61 (2000) 071703;
M. Hirsch, M.A. Daz, W. Porod, J.C. Romo, J.W.F. Valle, Phys. Rev. D 62 (2000) 113008.
[15] T. Banks, Y. Grossman, E. Nardi, Y. Nir, Phys. Rev. D 52 (1995) 5319;
A.S. Joshipura, M. Nowakowski, Phys. Rev. D 51 (1995) 2421;
G. Bhattacharyya, D. Choudhury, K. Sridhar, Phys. Lett. B 349 (1995) 118;
M. Nowakowski, A. Pilaftsis, Nucl. Phys. B 461 (1996) 19;
A.Yu. Smirnov, F. Vissani, Nucl. Phys. B 460 (1996) 37;
J.C. Romo, F. de Campos, M.A. Garca-Jareo, M.B. Magro, J.W.F. Valle, Nucl. Phys. B 482 (1996) 3;
B. de Carlos, P.L. White, Phys. Rev. D 54 (1996) 3427;
B. de Carlos, P.L. White, Phys. Rev. D 55 (1997) 4222;
H. Nilles, N. Polonsky, Nucl. Phys. B 484 (1997) 33.
[16] A. Santamaria, J.W.F. Valle, Phys. Lett. B 195 (1987) 423;
A. Santamaria, J.W.F. Valle, Phys. Rev. Lett. 60 (1988) 397;
A. Santamaria, J.W.F. Valle, Phys. Rev. D 39 (1989) 1780.
[17] M.A. Daz, J.C. Romo, J.W.F. Valle, Nucl. Phys. B 524 (1998) 23.
[18] A. Akeroyd, M.A. Daz, J. Ferrandis, M.A. Garca-Jareo, J.W.F. Valle, Nucl. Phys. B 529 (1998) 3;
M.A. Daz, J. Ferrandis, J.C. Romo, J.W.F. Valle, Phys. Lett. B 453 (1999) 263;
A.G. Akeroyd, M.A. Daz, J.W.F. Valle, Phys. Lett. B 441 (1998) 224;
M.A. Daz, E. Torrente-Lujan, J.W.F. Valle, Nucl. Phys. B 551 (1999) 78;
M.A. Daz, J. Ferrandis, J.C. Romo, J.W.F. Valle, Nucl. Phys. B 590 (2000) 3;
M.A. Daz, D.A. Restrepo, J.W.F. Valle, Nucl. Phys. B 583 (2000) 182;
M.A. Daz, J. Ferrandis, J.W.F. Valle, Nucl. Phys. B 573 (2000) 75.
[19] M. Frank, K. Huitu, Phys. Rev. D 64 (2001) 095015.
[20] M.A. Daz, hep-ph/9802407.
[21] A.G. Akeroyd, C. Liu, J. Song, hep-ph/0107218.
[22] F. de Campos et al., in preparation.

Breaking CP and supersymmetry with orbifold

moduli dynamics
Thomas Dent
Michigan Center for Theoretical Physics, Randall Lab., University of Michigan,
Ann Arbor, MI 48109-1120, USA
Abstract
We consider the stabilization of string moduli and resulting soft supersymmetry-breaking terms
in heterotic string orbifolds. Among the results obtained are: formulae for the scalar interaction
soft terms without integrating out the hidden sector gaugino condensate, which reduce to standard
expressions in the usual truncated limit; an expression for the modular transformation of A-terms;
a study of the minima of the scalar potential in the Khler modulus direction; and a discussion of the
implications for CP violation phenomenology.
Some closely related results have appeared in a recent paper of Khalil, Lebedev and Morris,
namely, the exact modular invariance of A-terms up to unitary mixing, and the existence of certain
complex minima for string moduli. 2002 Elsevier Science B.V. All rights reserved.
PACS: 12.60.Jv; 11.30.Er; 11.25.Mj; 11.30.Ly
1. Introduction
Many phenomenologically important quantities in the low-energy effective theory of
heterotic string orbifolds, including gauge and Yukawa couplings, depend on the vacuum
expectation values (v.e.v.s) of moduli. These are scalar fields that parameterize the metric
and background fields of the compact space: their potential vanishes in perturbation theory,
so it is essential to have some mechanism for stabilizing their values. Similar remarks
apply to Wilson lines that can take a continuous range of values. Once supersymmetry
(SUSY) is broken, necessarily by nonperturbative effects, the moduli get, in general,
a nonvanishing scalar potential, or equivalently the compactification parameters are
E-mail address: tdent@umich.edu (T. Dent).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 2 9 - 0
74
T. Dent / Nuclear Physics B 623 (2002) 7396
dynamically determined [1]. If supersymmetry is broken at an energy scale far below that
of the string, one should be able to study modulus stabilization using an effective field
theory (see, e.g., [2,3]).
We will focus on the implications of the moduli dynamics for CP violation, and
other issues of phenomenology, in the resulting low-energy theory of softly-broken
supersymmetry. The resulting phenomenology is in general described by complex Yukawa
couplings yiju,d,e and soft terms [4], namely, the gaugino masses, bilinear Higgs couplings
BHU HD and trilinear scalar couplings (A-terms) Auij yiju qi u cj HU , etc. The phases of
these terms are tightly constrained by experimental limits on electric dipole moments (for
flavour-diagonal interactions) ([57], and references therein) while flavour-off-diagonal
interactions in the quark mass basis are constrained by experimental results for K and B
mesons [8].
Whilst studies have appeared parameterizing the contribution of the moduli and dilaton
(a scalar field giving the value of the string coupling) to CP violation in soft terms [9] by
various assumptions about the magnitudes and phases of F -terms, our work addresses the
question of how these terms, and the sources of CP violation in Yukawa couplings, are to be
generated from a consistent model of SUSY-breaking and modulus stabilization. Research
along similar lines has been pursued by Bailin et al. [1013].
The effective supergravity theory used here is a slight simplification relative to detailed
derivations undertaken for specific orbifold models, for example, we use the overall
modulus approximation and ignore other potential sources of CP violation in looking at
one T modulus only. Also, exact target-space modular invariance is an important part of
our analysis, but in realistic models it might be modified or explicitly broken. It is relatively
simple to go from the overall modulus case to that of 3 diagonal moduli for the complex
planes of an orbifold. The moduli may behave differently, but looking at them one by one
the dynamics are similar to the overall modulus case (although with different parameters):
see [11], where in most of the explicit orbifold models it was found that one or two moduli
(for which modular invariance was realized) could be identified as sources of CP violation
and could be described essentially in the same way as an overall modulus. We believe that
the analysis presented here gives a reasonable sample of the behaviour of T -moduli to be
expected in the many possible orbifold models.
There exist other potential sources of CP violation in heterotic string models, for
example twisted, off-diagonal, or complex structure moduli, the dilaton S, scalars charged
under an anomalous U(1) group [14], or even discrete torsion [15]. It is unlikely that all of
these could contribute simultaneously, therefore it seems reasonable to take each possibility
in turn and study it in detail, in order either to rule it out or, in the case of positive results,
to motivate attempts to embed it in a more complete model. We choose the diagonal T
moduli since their dynamics and effect on low-energy physics are the best known and do
not require excessively complicated calculation. Note that the dilaton may also contribute
to CP violation in the soft terms, but Yukawa couplings, which occur in the superpotential,
cannot have S-dependence; then the dilaton cannot alone be the source of CP violation (see
also [24]). The dynamics of S are such that the contribution to soft terms either vanishes, or
it depends on nonperturbative corrections which are currently uncalculable (see Section 4);
therefore, we leave open the possibility that this contribution is nonzero in the relevant case
(Section 2.1), but no definite statement is made on this point.
75
The effective theory is believed to be invariant under target-space modular transformations [17,18] for which the modulus T i transforms as
Ti
T i i
,
i T i +
(1)
where , , , are integers satisfying

= 1 [16]. The moduli Ti , i = 1, 2, 3,
i
are defined as iT = 2(B2i1,2i + i det Gi ), where i labels the three complex planes of
the orbifold, Bmn , Gmn are the antisymmetric tensor field and metric respectively in the
compact directions and Gi is the ith 2-by-2 submatrix of Gmn (assumed block diagonal).
We simplify to the case of a single overall modulus T i = T . The dilaton S [19] and matter
fields [20] may also have nontrivial transformation properties. This symmetry has many
implications for the behaviour of the effective theory: no observable modulus-dependent
quantity can change if the v.e.v. of T is replaced by its image under transformation.
In addition the underlying theory is exactly CP invariant, since CP is a discrete gauge
symmetry of strings [21,22], thus T and T describe formally equivalent vacua.
Then the values at which T is stabilized are important in finding the modulusdependent low-energy couplings, in particular the Yukawa couplings and the scalar
soft SUSY-breaking terms. Below, we parameterize the unknown dilaton dynamics by
assuming that the dilaton v.e.v. is constant (up to a well-defined loop correction that is
required to preserve modular invariance) and varying the contribution of the dilaton to the
potential Vdil |W S /W + K S |2 within reasonable limits. This is equivalent to assuming
that the stabilization mechanism for the T -modulus has essentially no effect on the value
of the dilaton, and thus that any variation of the dilaton v.e.v. and F -term can be neglected.
This assumption has since been tested in [23], where essentially the same range of models
for the dynamics of T was considered, but the effects of different mechanisms for dilaton
stabilization were worked out in detail. The results are striking: in most cases the dilaton
v.e.v. is indeed constant, despite some variation in the values of T obtained. Then our
previous results, and those of [11,12], remain useful, and in fact coincide with some
of the phenomenologically interesting minima found in [23]. There are, however, some
discrepancies, which may be due in part to nontrivial variation of the dilaton away from
the global minimum; but for the mechanisms which determine S uniquely we still find
minima in T that do not occur in [23].
It was shown in [24] that in orbifold models with three matter generations, CP violation
in Yukawa couplings cannot originate from a modulus v.e.v. on the boundary of the
fundamental domain, in other words satisfying either |T | = 1 or Im T = 1/2 + n, n
integer. This resulted from an interplay between the modular and CP invariances. Since
in many cases T is stabilized at precisely these values [25], this seems to rule out
many scenarios for the origin of CP violation. Recently, however, a nonzero value of
the Jarlskog parameter JCP , which derives from the Standard Model Yukawa couplings
[26] was claimed to occur for T on the unit circle [27], in a model with more than three
generations where some fields are assumed to become heavy by an unspecified mechanism.
The previous result is evaded by taking the three light generations to mix with the heavy
matter under a modular transformation. The calculated form of JCP (T ) is not modular
invariant, which appears suspicious from the point of view of an exactly invariant theory.
In fact for |T | = 1, J is sent to J under the duality T 1/T . Minima on the unit
76
circle are generic, so it is important to understand this result and investigate whether it is
consistent with an explicit mechanism for decoupling the unwanted matter. This point will
be addressed in a future publication.
2. Soft supersymmetry-breaking terms

We calculate soft supersymmetry-breaking terms explicitly for particular models
motivated by perturbative heterotic string theory, and find the dependence on the dilaton
and moduli. This will enable us to look in more detail at the scenario in which CP violating
phases appear in the scalar trilinear A-terms and bilinear B term due to a complex v.e.v. of
the T modulus. We will concentrate on the modular transformation properties of the soft
terms, and the question of whether deviations from the truncated value of the condensate
induce significant changes in the calculation of soft terms.
2.1. Calculation of soft terms
By coupling the visible sector, which has just the field content of the MSSM, to
supergravity and a hidden sector in which supersymmetry is broken, the low-energy theory
appropriate for calculating phenomenology can be found [28]. The procedure is described
in more detail in, for example, [30,31]. In principle one should find the vacuum of the full
theory by minimising the scalar potential in terms of all the fields, including visible and
hidden matter; in practice for string scale near the Planck scale and a hidden sector with a
strong interaction scale a few orders of magnitude below, the effects of the visible sector
v.e.v.s will be negligible.
To define the full theory, the superpotential and Khler potentials for the visible and
hidden sectors are simply added. The hidden sector superpotential takes the Veneziano
Yankielowicz form [32]
fg (S, T )
b0
U
U ln(kU ),
(2)
4
96 2
where b0 3c(G) is the one loop beta-function coefficient for a SYM hidden sector
without matter. The gaugino condensate is described by the composite field U
W a Wa /S03 , where S0 is the conformal compensator in the superconformal formulation
of matter-coupled supergravity [33,34]. The hidden sector gauge kinetic function fg =
S + (T )/(16 2 ) includes a modulus-dependent threshold correction (T ) [35]. On
integrating out the condensate via W/U = 0 we find the standard truncated
superpotential
Wnp =
W (tr) (S, T ) =
b0
U (tr) ,
96 2
where
U (tr) = e24
2 S/b
0 1

1
kh(T ) ,
(3)
(4)
and h(T ) is defined by h(T ) e3(T )/2b0 . For the theory to be invariant under SL(2, Z)
modular transformations acting on T we must have the form h(T ) = 618GS /b0 (T )
77
H (T )1 , where (T ) is the Dedekind eta function, GS is a numerical coefficient

associated with the cancellation of modular anomalies [19] and H (T ) is a modular
invariant function without singularity in the fundamental domain [25].
For the visible sector we take
Wv = W ij i j + yij k i j k ,
where i are the MSSM chiral matter superfields and yij k , W ij are respectively the
coefficients of Yukawa couplings, and a possible supersymmetric bilinear coupling which
is required to couple the Higgs doublets of the MSSM in order to obtain correct electroweak
symmetry-breaking (the mu-term). For this to occur, the parameter , which has the
dimensions of a mass, must be of the same order as the soft supersymmetry-breaking
masses. This seems to require a fine-tuning of the underlying theory, since dimensionful
parameters in W are naturally of the order of the fundamental scale MP . The muproblem in supersymmetric phenomenology is the question of how the hierarchy between
MP and can be generated. Various solutions have been proposed, notably a nonminimal
mixing of the two Higgs doublets in the Khler potential [36], which has an effect
analogous to the supersymmetric W term, generating a Higgsino mass term and a scalar
bilinear B term after supersymmetry-breaking. Such a term in the Khler potential
can appear naturally if the Higgs superfields are in the untwisted sector of an orbifold
compactification [37], however the resulting couplings cannot easily be written down
in a form which is manifestly modular invariant (see [23] for a recent discussion).
For simplicity we use the alternative option, that the term HU HD is present in the
superpotential at the string scale and is tuned to be of the right order of magnitude by some
unspecified mechanism. In this case we allow for modulus-dependent couplings ij (T ),
yij k (T ), which are in general required to maintain modular invariance.
If we assume that the matter fields are in the twisted sector of an orbifold compactification, the Khler potential is

n
T + T i i i
Kv =
i
to second order in i , where ni is the modular weight of i . The dilaton and modulus
T ) is taken to be of the form
Khler potential K(S,

= P (y) 3 ln T + T ,
K
(5)
where
y = S + S

1
3GS ln T + T
2
(8 )
2
= Re y. The perturbative string
is a modular-invariant combination of fields such that gstring
Khler potential for the dilaton Kpert = ln(y) has been replaced by a real function P (y)
which parameterizes stringy nonperturbative dynamics, which are hoped to contribute
to dilaton stabilization [3840]. We initially take the GreenSchwarz coefficients GS to
be zero to simplify the calculations, but when calculating the scalar potential for the
modulus we relax this assumption. The correct form of P (y) is not known, however, it
is possible to constrain it by looking for a stable minimum in the potential for the dilaton
78
and requiring P (y) > 0 to obtain the right sign kinetic term. The complete hidden sector
Khler potential is then taken to be

1/3

3 ln 1 AeK/3
,
UU
Kh (U, S, T ) = K
(6)
where A is a constant. It was shown in [33] that this expression for the Khler potential of
U has the correct dependence on S and T , as well as being modular invariant. However, it
can only be determined up to a constant factor, and may receive higher-order corrections.
The constant A cannot at present be computed, due to our incomplete knowledge of
supersymmetric gauge dynamics, but is expected to be of order unity.
The trilinear A-terms and the scalar bilinear B-term corresponding to yij k and ij ,
respectively, can be extracted from the standard formula for the scalar potential

I
V = eK WI + KI W K 1 J W J + K J W 3|W |2 ,
(7)
where the indices run over U, S, T and i (i being the scalar component of i ) and we
assume a single hidden sector gauge group. We find a general formula for soft SUSYbreaking scalar interactions:
b0 z3 eK/2
L(scalar) =
96 2 (1 A|z|2 )2

P (y)
(S )

h (T )
P
+
(T
+
T
Wv
(y)
)
P (y)
(S )
h(T )

Wv
ln(eK/2 z3 (S )h(T ))
1
h (T )
+
ni 1 + (T + T )
i
i
3
h(T )
1 A|z|2
i

1
h (T )
Wv
+
(T + T ) 1 + (T + T )
+ h.c.,
T
3
h(T )
where we include terms up to third order in the visible sector fields and to all orders in the

dilaton, modulus and hidden sector condensate z defined by z3 = eK/2 U , taking GS = 0.
2
The function (S) e24 S/b0 is defined analogously to h(T ).
If the visible sector superpotential contains a coupling y123(T )1 2 3 then the
Lagrangian will contain the trilinear interaction
b0
eK/2 z 3
A123 y1231 2 3 = y1231 2 3
96 2 (1 A|z|2 )2

P (y)
(S )

(y)
P
P (y)
(S )

3
(T )
) y (T )

(T
+
T
h
n
i
123
+ 3 + (T + T )
1+
h(T )
3
3
y123(T )
i=1

K/2
3
ln(e
z (S )h(T ))
1 .
+3
1 A|z|2
(8)
The auxiliary fields F S and F T as defined by

i

F i = eK/2 K 1 j W j + K j W
are
79
(9)

b0
(S)
1
3
2 1/2

F =
|z| 1 A|z|
P (y)
,
96 2
P (y)
(S)

b0
T
3
2 1/2 (T + T )
h (T )
|z| 1 A|z|
F =
3 + (T + T )
.
96 2
3
h(T )
S
Superscripts denote holomorphic indices, subscripts antiholomorphic ones, so FT =

(F T ) . The A-term can then be written as

3
3
(T )

T + T y123
ni
z
K/2 S
T
A123 =
e
K FS + K FT 1 +
|z|
3
3
y123 (T )
i=1

|z|3
|b0 |
ln(eK/2 z3 (S )h(T ))
1 .
+3
96 2 (1 |z|2 )1/2
1 A|z|2
Similarly, for a coupling 12 (T )1 2 in Wv a corresponding soft supersymmetrybreaking coupling B12 1 2 is generated, with
b0
eK/2 z 3
96 2 (1 A|z|2 )2

P (y)
(S )

P (y)
1 +
P (y)
(S )

2

(T + T ) 12 (T )
ni
h (T )
1+
+ 3 + (T + T )
h(T )
3
3
12 (T )
i=1

ln(eK/2 z3 (S )h(T ))
1
.
+2
1 A|z|2
B =
(10)
In these expressions the truncated approximation for z, equivalent to imposing U = U (tr)

(Eq. (4)), is implemented by neglecting A|z|2 next to 1 and setting the logarithm in the
last term equal to (1); we can also use the formula for the gravitino mass m3/2
3 |/(96 2 ) to simplify the prefactors and make contact with established
eK/2 |W | = |b0 ztr
results. Deviations from the truncated approximation are treated in detail in [41] (see also
[42]).
2.2. Phenomenological discussion
The complex phases and flavour structure of the soft breaking terms are mainly
determined by the dependence on the T modulus: the FS terms are universal and, for a
single condensate, real. Apart from the phase of z3 (which is eliminated by the redefinition
of fields in going to the softly broken globally supersymmetric theory, for which see
80
below) a complex phase can only enter through the auxiliary field F T and the modulusdependent couplings y(T ), (T ). When T is stabilized at a minimum of the scalar potential
(Section 3), F T may be zero, real or have a complex phase which is of order 0.11. If the
term involving the derivative of y123(T ) were absent, the A-terms would have a common
phase, that of F T , and their magnitudes would be determined by the modular weights ni .
Using an explicit formula for Yukawa couplings, Khalil et al. [23] recently showed that
these logarithmic derivatives were real to a good approximation in some cases of interest,
thus a common phase for A-terms may be a good approximation.
In the special case of all ni equal we would recover the minimal supergravity ansatz
for the soft terms, in which the trilinear couplings are proportional to yij k , i.e., Aij k = A
for all i, j, k. However in general the A-terms will be nonuniversal, due to the different
values of ni and the terms involving yij k /yij k (which are also essential for modular
invariance): their magnitudes and phases will be different and there will be off-diagonal
(and complex) A-terms in the super-KM basis. Similarly the B bilinear coupling may
have a phase different from that of (only the phase difference between B and is
physically observable) which will feed through into a complex mass matrix for charginos
and neutralinos at low energies. This is a phenomenologically interesting scenario, which
may result in predictions for CP-violating observables which differ significantly from the
SM. However it is severely restricted by the nonuniversality of scalar masses which would
result if the matter generations of the MSSM had different modular weights: any departure
from degeneracy of scalar masses is likely to result in contributions to flavour changing
neutral current processes in excess of the experimental limits (see, e.g., [8]).
In the light of this discussion, deviations from the truncated approximation in the
formula (8) do not have an important direct effect. New complex phases are not introduced
and the corrections arising from z = ztr are universal, i.e., flavour-independent. However
the overall magnitude of the soft terms may be slightly changed, a second-order effect. So
from now on we will use the standard formulas resulting from the truncated approximation:

3
3
(T )

T + T y123
ni
z

K/2
S
T

e
K FS + K FT 1 +
A123 =
(11)
|z|
3
3
y123(T )
i=1
and

B=
z
|z|
3
e

K/2

2

T + T 12 (T )
ni
S
T

m3/2 K FS K FT 1 +
,
3
3
12 (T )
i=1
(12)
where the auxiliary fields are now

1
(S)
S

,
F = m3/2 P
P
(S)

(T + T )
h (T )
F T = m3/2
3 + (T + T )
.
3
h(T )
(13)
The corresponding formulae in the case GS = 0, which will be needed when discussing
the effect of changing GS on the minimisation of T and on the soft terms, have been
81
derived in [10,29]. As is well known, not all complex couplings in the Lagrangian result
in CP violation. We must consider whether the phases are physical and how many can
be eliminated by redefinition of fields. In particular, the behaviour of the A-terms under
SL(2, Z) modular transformations must be found, since physical quantities should be
modular invariant and (in general) the individual couplings will not be. Little can be
deduced from the phase of a single A-term without considering the whole set of couplings.
The results (8), (11) have nontrivial modular transformations, particularly, when the
observable matter fields are mixed by modular transformations. Since we suppose only
a single -term in the superpotential, modular transformations cannot mix the bilinear B
coupling with any other and it is relatively easy to verify that this term is modular invariant.
So we will focus on the trilinear A-terms.
2.3. Modular transformations of soft terms
Recall that the superpotential is a modular form of weight 3, so under modular
transformations
U
W U
,
(i T + )3
Wv
W Wv
,
(i T + )3
where W is a phase depending only on the parameters , , , of the transformation.

The observable fields transform into one another under SL(2, Z) as
i Cim (i T + )ni m ,
where Cim is a unitary matrix and ni = nm for all i, m such that Cim = 0, thus the Yukawa
couplings are constrained to transform as
yij k (T ) (i T + )
ni 3
yij k (T ),
where
yij k Cim Cj n Ckp = W ymnp

and ni denotes the sum over the three indices, i.e. (ni + nj + nk ). This implies that

yij k (T ) = W ymnp (T )Cmi
Cnj
Cpk .
We rewrite the scalar trilinear interactions as

3
z
Aij k yij k i j k = eK/2 3 i j k
|z|

(T + T )
ni
S
T

yij k (T ) .
K FS yij k + K FT
1+
yij k (T )
3
3
T FT = b0 /(96 2 )|ztr |3 (3 + (T + T )h (T )/ h(T )) transforms as

The quantity K
2
T FT (i T + ) K
T FT ,
K
2
|i T + |
82
where we have used the fact that h(T ) must have modular weight 3. Also, under modular
transformations the expression

1 + ni
1
yij k (T + T )yij k (T )
3
3
goes to
(i T + )2

W Cmi
Cnj
Cpk (i T + )3 ni
|i T + |2

nm
(T + T )
ymnp (T )
ymnp (T )
1+
3
3
since the matrices Cij and the phase W do not depend on T . Recalling that z is defined as

(eK/2 U )1/3 , we have
3 3

z 3
z
i T +
W
3
|i T + |
|z|
|z|3
under modular transformations. It is now easy to confirm that the expression Aij k yij k i
j k is modular-invariant, since Cmi

Cij = mj and all phases and factors of (i T + )
cancel.
This also implies that Aij k yij k (no sum!) transforms inversely to i j k :
(Ay)ij k (i T + )
ni

(Ay)mnp Cmi
Cnj
Cpk .
To clarify notation here we write (Ay)ij k Aij k yij k , where no sum is implied on the RHS.
Then dividing by yij k and its modular transform we find

mnp (Ay)mnp Cmi Cnj Cpk
3 1
Aij k (i T + ) W
(14)
.
m n p ym n p Cm i Cn j Cp k
If we ignore the unitary mixings Cij by assuming Cij = eii ij then all the phases i cancel
1
by inspection and the A-terms transform with a universal factor (i T + )3 W
.
2.4. Rescaling to a global SUSY with soft breaking terms
The formulas presented so far have been in terms of fields normalised as in the effective
supergravity theory. It is usual to rescale the visible sector component fields and rotate
away the phase of the hidden sector superpotential, so that the low-energy theory is just
the MSSM with canonical kinetic terms for the chiral matter, and soft breaking terms
expressed in terms of the gravitino mass m3/2 = |b0/(96 2 )z3 | [28]. The scalar potential
1/2
is then written in terms of normalised fields i = Ki i , where Ki (T ) = (T + T )ni . The
Yukawa couplings are rescaled as
yij k =
W K/2

e (Ki Kj Kk )1/2 yij k
|W |
(15)
83
so that
yij k i j k =
W K/2

e yij k i j k ,
|W |
which is modular-invariant. We have W /|W | = (z /|z|)3 in vacuo, and since

ASUGRA
yij k i j k = ASUSY
ij k
ij k yij k i j k
we deduce that
|z|3 K/2 SUGRA
e
Aij k
z 3

n + nj + nk (T + T ) ln yij k
S FS K
T FT 1 + i
= K
3
3
T
P
(S )
= m3/2 P
P
(S )

(T + T ) yij k (T )
ni
h (T )
+ 3 + (T + T )
1+
h(T )
3
3
yij k (T )
ASUSY
=
ij k
(16)
has modular weight zero. In particular, the transformation of the factors W /|W | and eK/2
1
cancels the previously-noted factor of (i T + )3 W
, so the expression is exactly invariant
if unitary mixings are neglected.
However, the ASUSY still in general transform with the complicated mixing in terms
of the yij k and Cim as indicated in (14). These involved transformation properties are a
major obstacle to finding the implications of complex phases in the A-terms for physical
CP-violating quantities. For a particular v.e.v. of T there will be certain predictions for the
complex phases of the A-terms, but in general these predictions will not be invariant under
the modular transformation acting on T and on the matter fields. We expect that physical
quantities should depend on combinations of the couplings in L that are modular invariant,
however, in the absence of a realistic model, and in the general case where modular
eigenstates cannot be defined, it is not clear how to construct the relevant quantities.
3. Stabilizing string moduli

3.1. Introduction
The breaking of local supersymmetry leads to a nonvanishing scalar potential for the
flat directions of string theory, the dilaton and compactification moduli. This suggests
the possibility that these quantities are dynamically determined after supersymmetry is
broken. When supersymmetry-breaking is mediated by gravity the scalar potential V (S, T )
is of order |F |2 , where F denotes a supersymmetry-breaking auxiliary field, while the
values of the dilaton and moduli fields vary over the Planck scale (when the units of S
and T are restored) so the flat directions acquire masses of order m3/2 TeV. While
this scenario is not without drawbacks for cosmology, and in most models predicts a
84
large negative cosmological constant at the minimum of V , it appears more promising

for phenomenology, since the dynamics of moduli can be studied through an effective field
theory and it is possible in principle to make predictions based on specific supersymmetrybreaking mechanisms.
In the models of supersymmetry-breaking via gaugino condensation that we have
considered, the dependence of the scalar potential on the dilaton and moduli is determined
by the gauge kinetic function of the hidden sector gauge group and the Khler potential
for S and T . We consider the simplest hidden sector consisting of a single gauge group
factor without matter, however we allow the GreenSchwarz coefficient to be nonzero. For
convenience we quote the scalar potential from [41,43,44] (without assuming the truncated
approximation):

|z|4
b0 2
V=
96 2 (1 A|z|2)2

2
3

K/2
3

2
z + C2 (S, T )|z| ,
(17)
1 + ln (S)h1 (T )e
A
where

1/3
z = eK/2 U
= eP (y)/6(T + T )1/2 U 1/3
as before, and
C2 (S, T ) =

1
(S) 2
(y)
P
P (y)
(S)
2

3GS
1
h1 (T )

+
3 . (18)
+ (T + T )
3 1
GS
b
h1 (T )
3 1 + 8
2 P (y)
Recall that (S) = e24
2 S/b
, and we take the ansatz

1
h1 (T ) = 618GS /b0 (T ) H (T ) ,
0
where H (T ) is in general a modular invariant function without singularities in the

fundamental domain. This form for h1 is supposed to originate from the threshold
correction

a (T ) = (b0 3GS ) ln 4 (T )(T + T ) + (b0 /3) ln |H (T )|2 ,
with H a holomorphic function of T .
Note that modular invariant, so-called universal threshold corrections in heterotic
string theory have been calculated [45,46] which take the form

a = (b0 3GS ) ln 4 (T )(T + T ) ka Y (T ),
where Y (T ) is modular invariant but not the real part of a holomorphic function. As
with the threshold corrections involving (T ), the nonholomorphic part of the threshold
correction appears in the T -dependence of the Khler potential of the effective field theory,
so the universal threshold corrections imply a correction to both the superpotential and
85
Khler potential for the dilaton and modulus. This results in the modular transformation
properties becoming considerably more complicated; so these corrections cannot be
included in the above prescription for h1 (T ). Note that the universal corrections ka Y can
2
= Re y, which formally justifies
be absorbed by a redefinition of the string coupling gstring
neglecting them if they are small. The effect of the universal threshold corrections on
stabilizing T has been found using the truncated approximation for the gaugino condensate
[13] resulting in a T -dependent potential similar in form to V (tr) C2 |ztr |6 . We might then
anticipate that the full condensate-dependent potential would have a similar form to (17)
with a different function of S and T replacing C2 . However the mathematical complexity
of the threshold corrections prevented us from investigating further.
The invariance of the theory under target-space T-duality T 1/T will ensure that
the modulus is stabilised at a value of order unity (assuming that the extra dimensions
are indeed dynamically compactified, i.e., that V becomes large and positive for Re(T )
very large and very small). However, in heterotic string theory no such duality applies to
the dilaton; without careful model-building either the potential will have no minimum in
the S-direction, or the minimum will lead to an unrealistic value of the gauge coupling.
This is the well-known dilaton runaway problem, first noticed in the case of a single
condensing gauge group with the Khler potential for the dilaton taking its string tree-level
form. There have been various proposals for solving it: the simplest, for our purposes,
is Khler stabilisation [40] where the dilaton Khler potential is supposed to receive
large corrections from nonperturbative string effects [38]. We have assumed that this is the
case in deriving the formulas (17) for the scalar potential. However, since we are mainly
interested in finding the v.e.v. of the T modulus, the details of the dilaton stabilization will
be neglected as far as is reasonable.
3.2. Calculation procedure
We will assume that the v.e.v. of the dilaton is fixed by nonperturbative effects irrespective of the value of T , such that the unified gauge coupling takes a phenomenologically
reasonable value. The (modular invariant) quantity
1
3GS ln(T + T )
8 2
will be set equal to 4, which will fix the value of S at any given value of T . The scalar
potential also depends on the function P (y) and its derivatives at the point y = 4. We
will treat P (y) and P (y)1 |P (y) (S)/(S)|2 Vdil as independent parameters
and take P (y) = ln 4, P (y) = 1/4 (the same values as for the perturbative Khler
potential Kp = ln y) since the effects of changing these two quantities on the potential
for T are small. Specifically, changing P (y) would change the overall scale of the
condensate z but not the shape of the potential, while changing P (y) would have a small
effect on the prefactor of the second term in (18). However, Vdil , the first term of C2 ,
corresponding physically to the amount of supersymmetry-breaking originating from the
dilaton dynamics, will be treated as a free (positive semidefinite!) constant parameter. It is
proportional to |F S |2 and may have an important effect on the shape of the potential and
the cosmological constant.
y = S + S
86
To determine the shape of the potential as a function of T , a number of further

parameters have to be specified. Apart from P (y) and its derivatives, we require the values
of b0 , A, GS and the functional form of H (T ) specified by the integers m and n and the
polynomial p(J ) in the expression
H (T ) = (J 1)m/2 J n/3 p(J ),
which is the most general invariant form with no singularities at finite T [25]. Of these,
b0 and GS can be calculated in specific orbifold models for a given hidden sector gauge
group; however, we will take the phenomenological liberty of varying b0 and GS over
typical ranges of values, for the purpose of illustration. The constant A is not known
and will be set to unity. The form of H , which parameterizes unknown modular invariant
threshold corrections, is essential to finding the minimum of the potential. In the absence
of definite results from string theory, we look at a range of values for m, n and the simplest
possibilities for p(J ), in an approach similar to [10,11,25], and look at the possible
implications for CP violation when T is stabilized at a minimum of the scalar potential. In
addition, we are able to look at the effects of deviations from the truncated approximation,
which may be important if they change the position of the minimum in the complex T
plane. This was not possible in the formalism of [10,11]. We calculate the scalar potential
in the truncated approximation

b0 2
(tr)
C2 |ztr |6 ,
V =
(19)
96 2
where ztr and C2 are functions of S and T , and then find the corrections by minimising the
full scalar potential in the z-direction for each value of T , with S being fixed as described
above. Note that the only dependence on ztr is in the overall scale of the potential.
3.3. Results: no dependence on J (T )
The simplest case occurs when the T -dependent holomorphic threshold corrections do
not involve the absolute modular invariant J (T ), in other words H = constant. This is the
form that results from a direct perturbative string calculation [35], which may however
miss universal, modular invariant contributions. The T -dependence of the scalar potential
is well-known in this case [2,3], however we are able to look in more detail at the effect of
changing various parameters.
For Vdil = 0, corresponding to no supersymmetry-breaking in the dilaton sector, the
minima of the scalar potential lie along the real axis. At GS = 0 the minima are at
approximately T = 0.8, T = 1.22 (Fig. 1), as GS increases the minima approach T = 1
and merge (Fig. 2) while at negative GS the minima become more widely separated: at
GS = 10 minima occur at T 0.4, 2.5 (Fig. 3). Note that the minima are related by
the modular symmetry T 1/T ; there are also minima at the modular images under
T T + i, which are not shown. We have also checked numerically that the scalar
potential is an invariant function under the full SL(2, Z) modular group. The corrections
to the truncated approximation (at GS = 0) are shown in Fig. 4. As discussed in [41] they
have a T -dependence different from that of the truncated scalar potential, so in principle the
corrections could alter the position of the minima. In this case the effect is not significant
87
Fig. 1. Scalar potential for m = n = 0, p(J ) = 1, b0 /(16 2 ) = 0.3, Vdil = 0 = GS .
Fig. 2. Scalar potential for m = n = 0, p(J ) = 1, = 0.3, Vdil = 0, GS = 15.
for phenomenology since the minima will remain on the real axis and at T -values of order
unity.
When Vdil = 3, the cosmological constant is fine-tuned to zero at the minimum of the
scalar potential (Fig. 5). In this case, there are two degenerate minima inside F at T = 1,
(where = ei/6 ) with C2 = 0 at these points, and the positions of the minima are
unaffected by changing GS . Since the corrections to the truncated approximation vanish
at C2 = 0, in this special case the minima are also unaffected by the corrections.
We also take Vdil = 6, which corresponds to a large, positive cosmological constant.
The minima are now at T = and its images under modular transformation. As in the
88
Fig. 4. Corrections to the truncated potential for m = n = 0, p(J ) = 1, = 0.3, Vdil = 0, GS = 0.
case where Vdil = 3, the position of the minimum is not changed by taking GS = 0, and
the effect of the corrections to the truncated approximation is small. We can conclude
that when the Dedekind eta function is the only modular form arising in the threshold
corrections, either T is real at the minimum, in which case CP-violating phases vanish, or
T = , in which case F T vanishes and the moduli do not contribute to supersymmetrybreaking. While this scenario is satisfactory as a solution to the supersymmetric CP
problem (and, when F T = 0, the supersymmetric flavour problem also) it does not
make any characteristic predictions for CP-violating quantities different from those of the
standard model. If there are no other sources of CP violation which can generate a CKM
89
phase then the scenario is, of course, ruled out. Neither does it throw light on the problems
of CP violation in the SM, namely the origin of the cosmological baryon asymmetry and
/< .
the high value of <K
K
3.4. Results for threshold corrections including J (T )
3.4.1. m = 1, n = 0, p(J ) = 1
First we take the case when the function H (T ) is just proportional to (J (T ) 1)1/2 .
This is in some sense natural, since J 1 (T 1)2 near T = 1, so the square root remains
well-defined near the zero of H . However, as discussed in [25], the effective action for the
condensate may become ill-defined at some finite radius around T = 1, since some string
states are supposed to become light here; we also expect the truncated approximation to fail
badly near T = 1 since the quantity x = AC2 |ztr |2 which measures the size of deviations
from the truncated approximation becomes large [41]. We first present the form of the
scalar potential for Vdil = 0 = GS (Fig. 6). Note the dimple near T = 1 which is related
to the failure of the truncated approximation. We can look at this area of the complex
T plane in the limit where T 1 and find that while the truncated condensate value
|ztr | vanishes as the cube root of (T 1), C2 diverges as (T 1)2 , so the truncated
approximation fails at a certain radius from T = 1. The effect on the scalar potential
as a function of T is shown in Fig. 7. At the values of T , where a value for the scalar
potential is not plotted, the gaugino condensate is destabilized by the C2 |z|6 term in the
scalar potential. Here we appear to be on a branch of solutions with zero condensate
and unbroken supersymmetry [41,42] which is not connected to the rest of the surface
V (T ). If the effective action for the gaugino condensate is valid down to z = 0 then there
exists a supersymmetric, zero-energy vacuum for all T and it becomes a dynamical and
cosmological question as to how the condensate becomes nonzero [42].
90
Fig. 6. Scalar potential for m = 1, n = 0, p(J ) = 1, = 0.3, Vdil = 0 = GS .
Fig. 7. Scalar potential divided by the truncated potential near T = 1, for the same values of parameters as Fig. 6.
The minimum of the scalar potential is near T = : this point is actually a (rather flat)
maximum and the minima are close by inside the boundary of F at T = 0.8842 + 0.4844 i
and its images under modular transformations (and the complex conjugate of these values).
T F T /m3/2 = 0.0389 + 0.0136 i;
The cosmological constant is negative. In this case K
we use this quantity as a rough measure of the amount of supersymmetry-breaking and
91
CP violation that we expect to originate from a particular v.e.v. of T .1 This result may be
compared to the case where m = 0, GS = 5 and the potential is minimised at T = 1.503,
T F T /m3/2 = 1.172, a factor of 50 larger. Where T is stabilised very
which results in K
near the self-dual point (where F T vanishes) with m = 1, F T appears to be much smaller
than at a general point inside F , so its contribution to CP and flavour violation should
be small. But we must ask, small relative to what? In order to have phenomenologically
reasonable soft breaking terms we also require F S to be nonzero, so that for F T small we
are near the dilaton-dominated limit of supersymmetry-breaking. Since Vdil |F S |2 we
should consider the effect of changing Vdil on the stabilization of T , as well as varying GS
and looking at the effects of deviations from the truncated approximation.
We first take Vdil = 1.5 while keeping GS = 0 and b0 /(16 2 ) = 0.3 fixed: then
the minimum is on the unit circle near T = at T = 0.8777 + 0.4792 i and F T = 0. For
Vdil = 3, the minimum is at T = , F T = 0 and the cosmological constant is tuned to zero.
Next consider changing GS : we first take GS = 15, Vdil = 0, = 0.3, in which
case the minimum is on the boundary of F at T = 0.9058 + 0.5 i and F T = 0.1075 is
real at the minimum. Considerations of modular invariance [24] indicate that this v.e.v.
for T will not generate CP violation, however small deviations from universality may
result. For GS = 15, Vdil = 3, = 0.3 the minimum is extremely close to T =
and F T vanishes. For GS = 15, Vdil = 0, = 0.3 the minimum is on the unit circle at
T = 0.8873 + 0.4611 i and F T also vanishes; we find that this result is virtually unaffected
by changing Vdil .
Finally for this case we consider the effect of corrections to the truncated approximation
on the minimisation of the scalar potential. Since these scale with |ztr |2 efg / , the
corrections can be turned on or off by changing (although only a finite range of
values will result in a phenomenologically reasonable scale of supersymmetry-breaking).
The truncated approximation corresponds to minimising the scalar potential in the limit
|ztr | 0, i.e., 0 , while the corrections are maximised for large, negative .
We first take = 0.1 to mimic the truncated approximation: for Vdil = 0 = GS the
T F T /m3/2 = 0.0388 + 0.0134 i: differing
minimum is at T = 0.8842 + 0.4844 i and K
only very slightly from the result at = 0.3. For GS = 0, Vdil = 3 the minimum is again
at T = . Next we take the extreme large value = 0.9, keeping Vdil = 0 = GS , which
T F T /m3/2 = 0.0403 + 0.0136 i.
results in a minimum at T = 0.8844 + 0.4847 i with K
We see in this case that corrections to the truncated approximation do not much change the
position of the minimum.
3.4.2. m = n = 0, p(J ) = J 1
This ansatz, equivalent to m = 2, n = 0, p(J ) = 1 reproduces the desirable feature
of the previous case of a small or zero contribution of the T modulus to CP violation
and nonuniversality, without the hole where the gaugino condensate is destabilised.
However, due to the zero of J at T = 1, the condensate vanishes smoothly at this point,
an equally undesirable scenario for phenomenology! As for the case m = 1, n = 0,
1 For
T F T /m3/2 = 3 (T + T )(6d ln (T )/dT d ln H (T )/dT ) (see Eqs. (11)(13);
GS = 0 we have K
T F T /m3/2 = (1 3GS /b0 )(3 + 6(T + T )d ln (T )/dT ) + (T +
for nonzero GS the analogous quantity is K
T )d ln H (T )/dT (compare [10,29] and Eq. (18)).
92
the phenomenologically interesting minima are near T = . We start with the same
set of parameters, Vdil = 0 = GS and = 0.3, for which the minimum is inside the
T F T /m3/2 = 0.0199 + 0.0071 i
fundamental domain at T = 0.8754 + 0.4921 i and K
and the cosmological constant is negative. For Vdil = 3 there is a minimum on the unit
circle at T = 0.8721 + 0.4894 i with F T = 0, and T = also appears to be a minimum.
In this case corrections to the truncated approximation may be significant. As we
increase , and thus the size of the corrections, the minimum inside the fundamental
domain for Vdil = 0 moves towards the line Im(T ) = 1/2, although only slowly. We
T F T /m3/2 at the minimum: for = 0.3 it is 0.87539 +
may compare the values of K
T F T /m3/2 = 0.87547 + 0.49215 i. The effect is
0.49207 i and for = 0.9 we have K
small, which follows from the fact that the corrections tend to be smallest around the
minimum of the potential.
For GS = 15, Vdil = 0, = 0.3 the modulus is stabilised at T = 0.8857 + 0.5 i
T F T /m3/2 = 0.0538. As || is increased to 0.7 the minimum remains on the line
and K
Im(T ) = 1/2, so again the corrections do not seem to have a large effect. For GS = +15,
Vdil = 0, = 0.3 the minimum is on the unit circle at T = 0.8774 + 0.4796 i such that
F T = 0: this conclusion is also unchanged by increasing and thus the corrections to
truncation.
3.4.3. m = 0, n = 1, 2, . . . , p(J ) = 1
In this case the scalar potential is minimised at T = 1 with F T = 0, regardless of the
values of other parameters. Near T = the same phenomenon occurs as in the case of
m = 1, n = 0, p(J ) = 1 near T = 1: at some finite radius away from the fixed point the
truncated approximation fails and a nonzero value for the condensate becomes unstable.
If we set m = 0, n = 3, P (j ) = 1, which is equivalent to m = n = 0, p(J ) = J , the
supersymmetry-breaking minimum is again at T = 1 and the condensate goes smoothly
to zero as T .
3.4.4. m = 1, 2, n = 1, 2, p(J ) = 1
In this case there is a region around both fixed points T = 1, T = where the
condensate is destabilised or goes continuously to zero. For m = n = 1 and the parameter
values Vdil = 0, GS = 0, = 0.3, there are minima on the unit circle at T = 0.9713 +
0.2378 i (and at the complex conjugate value) with F T vanishing (Fig. 8). Again, the
minimisation is robust to changes in parameters, in that the minimum remains on the unit
circle with F T = 0 when Vdil , GS and are changed within reasonable limits.
3.5. Interpretation of the results
The main results from minimization in any particular case are the value of T at the
T F T /m3/2 , which measures the contribution of
minimum and the size of the quantity K
the modulus to soft supersymmetry-breaking terms. Since the phase of F T is not modular
invariant, and the phases of the soft terms receive additional contributions from the T dependence of the Yukawa couplings (which we have not explicitly calculated), we cannot
give unambiguous measures of CP violation in the soft terms. However the value of T
should allow us to diagnose whether CP is broken in the low-energy theory, and the size
93
Fig. 8. Scalar potential for m = n = 1, p(J ) = 1, = 0.3, Vdil = 0 = GS .
of F T allows us to estimate how far we are from the dilaton- or moduli-dominated limits.
The results presented seem to indicate that the most common patterns of supersymmetrybreaking are close to the dilaton-dominated limit, except when Vdil = 0 and (T ) is the
only modular form appearing in the threshold corrections: for various values of parameters,
we obtain F T vanishing or small compared to the scale of the gravitino mass, and when
F T is nonzero the v.e.v. of T is such that its contribution to supersymmetry-breaking may
be either CP-conserving or (if T lies in the interior of F ) CP-violating.
3.6. Singular points in the effective theory
We noted several times, in the case where the threshold corrections involved J (T ),
points in the T plane where the effective potential apparently became discontinuous, when
the gaugino condensate was allowed to be dynamical (see in particular, Section 3.4.1).
This was due to divergences in the threshold corrections, which can be interpreted as the
effects of new light states appearing [25]. Then, rather than a transition to another branch of
the theory with zero condensate, one should conclude that the effective theory is breaking
down. In the case of new massless matter fields, the correct effective theory at these points
may look something like supersymmetric QCD, which has a runaway vacuum and equally
unappealing phenomenology.
In the usual approximation of a nondynamical condensate, such points do not appear
pathological, and simply give extrema or zeros of the scalar potential at which the effective
superpotential vanishes. The apparent smooth behaviour is due to a conspiracy between
a vanishing gaugino condensate and a divergent factor proportional to W 1 W/T [41]
which in the dynamical condensate case would signal the breakdown of the effective theory.
Many of the minima quoted in [23] are of this type, and while this fact does not alter the
(negative) phenomenological conclusions, it is possible that other minima may have been
missed because such pathological points were not excluded from the search.
94
4. Further reflections
The existence of a viable phenomenology of CP violation in Yukawa couplings and
soft terms depends largely on the dynamics of the T modulus in these models. The main
contributions that the dilaton dynamics should make are to ensure the correct unified
gauge coupling and, in the case where F S = 0, to contribute to soft terms in such a
way as to alleviate the problems of nonuniversality and small gaugino masses in the
moduli-dominated limit. We assumed, with [11,12] that the dilaton plays essentially in
the minimization in the T -direction. The results of [23] tend to reinforce this assumption:
thus, while knowledge of the dilaton dynamics is needed to construct a full model, it is
less important if one is mainly concerned with CP violation by T . The nontrivial minima
obtained here for the cases m = 1, n = 0 and m = n = 1 are reproduced in [23], up to a
modular transformation (see their Table 2).2
However, there are some cases where apparent differences appear: namely, in the case
m = 2, n = 0 where F S = 0, and in the Khler stabilization case. For m = 2, n = 0 we
find a SUSY-preserving minimum at T = 1, which is confirmed in [23] for all dilaton
stabilization mechanisms, but we also find minima with negative vacuum energy inside F
near the self-dual point T = , as in the case m = 1, n = 0. This case (assuming a constant
dilaton) was first considered in [47] in the context of cosmology. The minima did depend
on our assumptions for the dilaton dynamics, but as discussed below, one expects that
any minimum found by our approach should be reproduced in the approach with explicit
dilaton stabilization by the racetrack or S-dual mechanisms. It is not clear what happens to
these minima in the analysis of [23].
In the case with nonperturbative dilaton Khler potential the property claimed in [40]
of giving a nonzero F S only survived in a single case, m = n = 0. This occurs because all
minima found in [23] had T at the fixed points, where W vanishes unless m = n = 0,
because the modular invariant functions chosen vanish at the minima. Then F S must
vanish, being proportional to |W |. The discussion of the relation between Vdil and F S in our
thesis (Section 3.4.1) was somewhat misleading in that we neglected the exact implications
for this point if W vanished at the minimum. Technically, |W |2 Vdil |W GS |2 |F S |2 ,
where G K + ln |W |2 . Thus, while in the racetrack and S-dual cases both GS and
F S vanish, corresponding to setting Vdil = 0 (which was our default value), for Khler
stabilization the F -term vanishes if W = 0, even though GS is nonzero. In terms of our
treatment this means that the Khler stabilization case corresponds to a nonzero Vdil , which
we did in many cases include in the analysis. The effect is indeed to change the shape of
the potential V (T ), although even with Vdil = 0 we consistently found more local minima
than those in Tables 4 and 5 of [23], including some not at fixed points, at which |W |
will likely not vanish. Specifically, for m = 1, n = 0 we found a minimum near the fixed
1/2
point T = with F T = 0 for GS = 0 and nonzero GS Vdil , and a minimum on the line
T
Im T = 1/2 with F = 0 for GS = 15 and a wide range of values of GS (Section 3.4.1).
2 It is possible that a nonzero F S with a complex phase could produce important effects in the soft terms,
however no model has yet realized this case: in the case where F S = 0 the phase vanishes. Since we are not
confident that the correct dilaton stabilization scheme has been discovered, a complex F S should still be allowed.
95
For m = 2, n = 0, GS = 0 and nonzero GS we found minima on the unit circle close to

T = (Section 3.4.2). 3 For m = n = 1 we found minima with T on the unit circle which
remained on the unit circle (avoiding fixed points) for various values of GS ( Section 3.4.4);
a similar result was obtained in [11] for slightly different values of parameters.
If the dilaton v.e.v. is insensitive to the value of T , there is little to choose between
parameterising its dynamics by two constants Vdil and P (y), or by three (the d, p, b
of [40]), so far as the effect on stabilizing T is concerned. There is currently no way
in which either set of parameters can be calculated. In fact our approach allows more
flexibility, since we can consider the effect of varying Vdil from zero to any finite value,
while the alternative is to be tied to zero or the values produced by a particular choice of
form for the dilaton Khler potential. The only case where the parameterization by Vdil and
P (y) may be invalid is if the dilaton stabilization mechanism has a nontrivial dependence
on the value of T , thus altering the shape of V (T ) still further by a nonconstant prefactor
2
|z|6 |e24 S/b0 |2 . Either this possibility, or a discrepancy in numerics, must account for
the difference of results between the two approaches.
Further investigation along the lines of [23] could easily resolve this question, either by
finding a finite variation in S|min as T is varied by hand, or by finding more minima. If
minima with nonvanishing superpotential and T away from a fixed point can be found
with the Khler stabilization mechanism it would be phenomenologically important, since
F S = 0 is crucial for realistic superpartner spectra, and conversely fixed point values of
T do not produce CP violation.
Authorial note
Much of the material in this paper, specifically Sections 23.5, appeared in the authors
D. Phil. thesis submitted at the University of Sussex in September 2000.
Acknowledgements
The author acknowledges the invaluable supervision of David Bailin at the time when
much of this work was carried out, and a helpful correspondence with Stephen Morris.
This research was also supported in part by DOE Grant DE-FG02-95ER40899 Task G.
References
[1]
[2]
[3]
[4]
K. Kikkawa, M. Yamasaki, Phys. Lett. B 149 (1984) 357.

A. Font, L.E. Ibez, D. Lst, F. Quevedo, Phys. Lett. B 245 (1990) 401.
S. Ferrara, N. Magnoli, T.R. Taylor, G. Veneziano, Phys. Lett. B 245 (1990) 409.
M. Dugan, B. Grinstein, L. Hall, Nucl. Phys. B 255 (1985) 413.
3 The vanishing of F T for T on the unit circle is due to a general property of the modular functions considered,
and does not imply that W = 0 at all such points.
96
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
S. Pokorski, J. Rosiek, C.A. Savoy, Nucl. Phys. B 570 (2000) 81, hep-ph/9906206.
T. Falk, K.A. Olive, M. Pospelov, R. Roiban, Nucl. Phys. B 560 (1999) 3, hep-ph/9904393.
S. Abel, S. Khalil, O. Lebedev, Nucl. Phys. B 606 (2001) 151, hep-ph/0103320.
F. Gabbiani, E. Gabrielli, A. Masiero, L. Silvestrini, Nucl. Phys. B 477 (1996) 321, hep-ph/9604387;
M. Misiak, S. Pokorski, J. Rosiek, Supersymmetry and FCNC effects, hep-ph/9703442.
M. Brhlik, L.L. Everett, G.L. Kane, J. Lykken, Phys. Rev. D 62 (2000) 035005, hep-ph/9908326;
E. Accomando, R. Arnowitt, B. Dutta, Phys. Rev. D 61 (2000) 075010, hep-ph/9909333;
S. Abel, S. Khalil, O. Lebedev, Phys. Rev. Lett. 86 (2001) 5850, hep-ph/0103031.
D. Bailin, G.V. Kraniotis, A. Love, Phys. Lett. B 414 (1997) 269, hep-th/9705244.
D. Bailin, G.V. Kraniotis, A. Love, Nucl. Phys. B 518 (1998) 92, hep-th/9707105.
J. Giedt, Nucl. Phys. B 595 (2001) 3, hep-ph/0007193.
C. Vafa, Nucl. Phys. B 273 (1986) 592.
R. Dijkgraaf, E. Verlinde, H. Verlinde, in: Perspectives in String Theory, Copenhagen 87 Proceedings, pp.
117137.
S. Ferrara, D. Lst, A. Shapere, S. Theisen, Phys. Lett. B 225 (1989) 363.
S. Ferrara, D. Lst, S. Theisen, Phys. Lett. B 233 (1989) 147.
J.P. Derendinger, S. Ferrara, C. Kounnas, F. Zwirner, Nucl. Phys. B 372 (1992) 145;
L.E. Ibez, D. Lst, Nucl. Phys. B 382 (1992) 305, hep-th/9202046.
J. Lauer, J. Mas, H.P. Nilles, Nucl. Phys. B 351 (1991) 353.
K. Choi, D.B. Kaplan, A.E. Nelson, Nucl. Phys. B 391 (1993) 515.
M. Dine, R.G. Leigh, D.A. MacIntire, Phys. Rev. Lett. 69 (1992) 2030.
S. Khalil, O. Lebedev, S. Morris, hep-th/0110063.
T. Dent, Phys. Rev. D 64 (2001) 056005, hep-ph/0105285.
M. Cvetic, A. Font, L.E. Ibez, D. Lst, F. Quevedo, Nucl. Phys. B 361 (1991) 194.
C. Jarlskog, Phys. Rev. Lett. 55 (1985) 1039;
C. Jarlskog, Z. Phys. C 29 (1985) 491.
O. Lebedev, hep-th/0108218.
S.K. Soni, H.A. Weldon, Phys. Lett. B 126 (1983) 215.
B. de Carlos, J.A. Casas, C. Muoz, Phys. Lett. B 299 (1993) 234, hep-ph/9211266.
A. Brignole, L.E. Ibez, C. Muoz, Nucl. Phys. B 422 (1994) 125, hep-ph/9308271, hep-ph/9707209.
V.S. Kaplunovsky, J. Louis, Phys. Lett. B 306 (1993) 269, hep-th/9303040.
G. Veneziano, S. Yankielowicz, Phys. Lett. B 113 (1982) 231.
C.P. Burgess, J.P. Derendinger, F. Quevedo, M. Quirs, Ann. Phys. 250 (1996) 193, hep-th/9505171.
S. Ferrara, L. Girardello, T. Kugo, A. Van Proeyen, Nucl. Phys. B 223 (1983) 191.
L.J. Dixon, V. Kaplunovsky, J. Louis, Nucl. Phys. B 355 (1991) 649.
G.F. Giudice, A. Masiero, Phys. Lett. B 206 (1988) 480.
I. Antoniadis, E. Gava, K.S. Narain, T.R. Taylor, Nucl. Phys. B 432 (1994) 187, hep-th/9405024.
S.H. Shenker, in: Random Surfaces and Quantum Gravity, Cargse 90 Proceedings, Cargse, France, 1990.
T. Banks, M. Dine, Phys. Rev. D 50 (1994) 7454;
T. Banks, M. Dine, Phys. Lett. B 384 (1996) 103;
P. Bintruy, M.K. Gaillard, Y.-Y. Wu, Nucl. Phys. B 481 (1996) 109;
K. Choi, H.B. Kim, H. Kim, Mod. Phys. Lett. A 14 (1999) 125.
J.A. Casas, Phys. Lett. B 384 (1996) 103, hep-th/9605180;
T. Barreiro, B. de Carlos, E.J. Copeland, Phys. Rev. D 57 (1998) 7354, hep-ph/9712443.
T. Dent, Phys. Lett. B 470 (1999) 121, hep-th/9909198.
H. Goldberg, Phys. Lett. B 357 (1995) 588.
T.R. Taylor, Phys. Lett. B 252 (1990) 59.
B. de Carlos, J.A. Casas, C. Muoz, Phys. Lett. B 263 (1991) 248.
E. Kiritsis, C. Kounnas, P.M. Petropoulos, J. Rizos, Nucl. Phys. B 483 (1997) 141, hep-th/9608034.
H.P. Nilles, S. Stieberger, Nucl. Phys. B 499 (1997) 3, hep-th/9702110.

Gravity-mediated supersymmetry breaking

in the brane world
Tony Gherghetta a , Antonio Riotto b
a Institute of Theoretical Physics, University of Lausanne, CH-1015 Lausanne, Switzerland
b INFN, Sezione di Padova, Via Marzolo 8, I-35131 Padua, Italy
Abstract
We study the transmission of supersymmetry breaking via gravitational interactions in a five-dimensional brane world compactified on S 1 /Z2 . We assume that chiral matter and gauge fields are
confined at the orbifold fixed points, where supersymmetry is spontaneously broken by effective
brane superpotentials. Using an off-shell supergravity multiplet we integrate out the auxiliary fields
and examine the couplings between the bulk supergravity fields and boundary matter fields. The
corresponding tree-level shift in the bulk gravitino mass spectrum induces one-loop radiative masses
for the boundary fields. We calculate the boundary gaugino and scalar masses for arbitrary values of
the brane superpotentials, and show that the mass spectrum reduces to the ScherkSchwarz limit for
arbitrarily large values of the brane superpotentials. 2002 Elsevier Science B.V. All rights reserved.
PACS: 12.60.-i; 12.60.Jv
1. Introduction
An important question in the supersymmetric standard model is how supersymmetry is
spontaneously broken in the low-energy world. This question has been mainly addressed
in the context of four-dimensional effective theories with limited success, but recently the
idea of extra dimensions has allowed for new possibilities [19]. The extra-dimensional
framework is particularly interesting because it provides a new geometrical perspective in
understanding some of the problems of conventional theories. In particular, the nonlocal
nature of communicating supersymmetry breaking across the compact bulk from one
boundary to another can soften the divergences of the soft mass spectrum [3,7].
The HoravaWitten scenario [2] provides the prototype model in which to study
the transmission of supersymmetry breaking via an extra dimension. In this model the
E-mail addresses: tony.gherghetta@cern.ch (T. Gherghetta), antonio.riotto@pd.infn.it (A. Riotto).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 3 7 - X
98
T. Gherghetta, A. Riotto / Nuclear Physics B 623 (2002) 97125
transmission of supersymmetry breaking can become quite involved, and to calculate

boundary soft masses one requires the bulk/boundary couplings. However, the essential
features can be captured by studying a simpler five-dimensional super-YangMills theory
coupled to chiral matter on the boundary [5]. In this toy model the couplings between
five-dimensional supermultiplets and four-dimensional boundary fields are obtained by
working with off-shell supermultiplets and including the auxiliary fields. As noticed in
Ref. [5] the dimensional reduction of bulk fields leads to new couplings between bulk
and boundary fields which are required for consistency. In particular, this allows for the
construction of realistic low-energy models with bulk gauge fields in flat [7] and warped
space [9].
In this work we study brane world supersymmetry breaking in the case where only
gravity propagates in the bulk while the chiral matter and gauge fields are confined to
the four-dimensional boundaries. We assume that due to brane dynamics supersymmetry
is spontaneously broken by effective brane superpotentials. This causes the tree-level
gravitino mass spectrum to shift by a constant amount depending on the values of the brane
superpotential [10]. At tree-level boundary chiral matter and gauge fields are massless
but due to their gravitational interactions with the bulk gravitinos they will receive a
supersymmetry breaking mass at one loop. Just like the bulk gauge field case, one can
use an off-shell formulation to study the bulk gravitational case as well.
Supersymmetric brane world scenarios from off-shell supergravity have been formulated
in Refs. [11,12]. We will predominantly use the results in Ref. [11] to study an offshell formulation of supergravity in the context of supersymmetry breaking with brane
superpotentials. In particular, we will show that after integrating out the auxiliary fields of
the off-shell supergravity multiplet there are new couplings between bulk and boundary
fields. Just like the bulk gauge field case these couplings are required in order to obtain a
consistent supersymmetric limit.
The one-loop mass spectrum will continuously depend on the brane superpotential
parameter, and due to the nonlocal nature of the supersymmetry breaking the masses will
be finite. In fact we will see that one particular limit of our mass spectrum is the familiar
ScherkSchwarz limit [3,13]. Actually depending on the size of the extra dimension our
one-loop results can be of order the anomaly-mediated contributions [6,14] which arise
from the one-loop rescaling anomalies. Thus, although we do not discuss this issue in
detail, our results could be relevant in solving the tachyonic slepton mass problem [6].
The plan of this paper is as follows: in Section 2, after briefly reviewing the bulk vector
multiplet case we consider the off-shell supergravity multiplet coupled to boundary fields.
In particular we show that after integrating out auxiliary fields there are new couplings
between boundary gauge fields and bulk supergravity fields. Supersymmetry breaking is
considered in Section 3, where we derive the unitary matrix responsible for diagonalising
the KaluzaKlein gravitino mass spectrum. This is important for determining the couplings
between the boundary and bulk fields. As a further check, the same results will also
be derived more directly using an explicit five-dimensional calculation. In Section 4
we calculate the one-loop gaugino and scalar masses for arbitrary values of the brane
superpotential. We comment on the cancellations that are required for consistency and are
99
satisfied by the new couplings. Again, for completeness we will present the calculation of
the soft mass spectrum using both the KaluzaKlein sum in four dimensions, and the direct
five-dimensional calculation. Finally, our conclusion and comments will be presented in
Section 5.
2. Off-shell bulk supergravity on S 1 /Z2
We start from a pure N = 2 five-dimensional Poincar supergravity [15], compactified
on an orbifold S 1 /Z2 . Our model will assume that only gravity propagates in the bulk
whereas chiral matter and gauge fields will be confined to the 4D boundaries. Thus all supersymmetry breaking effects will be transmitted by gravity and in particular the gravitino
mass spectrum will shift. In order to study the transmission of supersymmetry breaking
effects between the 4D boundaries it is necessary to work with an off-shell formulation of
supergravity [11,12]. In this way all bulk-boundary couplings can be derived.
A similar procedure for gauge fields and hypermultiplets in the bulk was considered in
Ref. [5]. Before launching ourselves into the more involved case of supergravity coupled
to boundary chiral and vector multiplets, we briefly summarize the procedure and results
of Ref. [5] for the case of a U (1) bulk vector multiplet which is coupled to a chiral matter
multiplet on the boundary. This will allow us to emphasize some key features which will
also be present in the supergravity case.
2.1. 5D YangMills multiplet coupled to boundary chiral matter
The five-dimensional U (1) multiplet with coupling constant g5 contains a vector field
AM , a real scalar field , and a gaugino i . The five-dimensional YangMills multiplet is
then extended to an off-shell multiplet by adding an SU(2) triplet Xa of real-valued auxiliary fields. Here capitalized indices M, N run over 0, 1, 2, 3, 5, lower-case indices m run
over 0, 1, 2, 3, and i, a are internal SU(2) spinor and vector indices, with i = 1, 2, a = 1,
2, 3.
We now compactify the theory on S 1 /Z2 and assign even Z2 -parity to the fields
1
L
, Am , 1L , X3 ,
(1)
and odd Z2 -parity to the fields

2
, A5 , , 2L , X1 , X2 ,
L
(2)
is the supersymmetry parameter of the N = 1 supersymmetry transformations

where
on the boundary at x5 = 0. A simple inspection of the supersymmetry transformations
reveals that the fields Am , 1L , and (X3 5 ) transform as the vector, gaugino, and the
auxiliary D-field of a 4D N = 1 vector multiplet [5]. It is then obvious how to couple the
five-dimensional gauge multiplet to a 4D-dimensional chiral multiplet (, L , F ) on the
boundary. One writes the Lagrangian as

S = d 5 x L5 +
(3)
x5 xi L4i ,
i
L
100
where the sum includes the walls at xi = 0, R. The bulk Lagrangian L5 is the standard
one for a 5D super-YangMills multiplet, and the boundary Lagrangian has the standard
form of a four-dimensional chiral model built from the chiral multiplet and with the gauge
fields (Am , L , D) replaced by the boundary values of the bulk fields (Am , 1L , X3 5 ).
To determine the couplings of the boundary chiral matter to the bulk vector multiplet
one has to integrate out the auxiliary fields. Integrating out the auxiliary field X3 gives a
boundary Lagrangian of the form

1 2 2
4
d x (5 ) g5 (0) ,
(4)
2
and one finds new interaction terms at the boundaries (apart from the usual ones in N = 1
4D YangMills theory coupled to chiral multiplet) involving the scalar components of the
chiral multiplet and the odd field [5].
By including the kinetic term of the field , the singular terms can be rearranged into a
perfect square

1
1 2 2 2
5
2
d x
(5 ) + (5 ) (x5 ) + g5 (x5 )
2
2g52

1
2
= 2 d 5 x 5 + g52 (x5 ) .
(5)
2g5
Varying this action with respect to , one finds that the background expectation value of
is given by

1
5 = g52 (x5 )
(6)
.
2R
Substituting this solution into the Lagrangian (5) one finds that the various singular terms
(0) cancel and one is left with the usual D-term interaction
1
S =
2
g52 2
1
=
d x
2
2
4 R
4
5

2
d 4 x g42 ,
(7)
where g42 = (g52 /R). To summarize, we have learned that starting from an off-shell
formulation of five-dimensional YangMills theory compactified on S 1 /Z2 which is
coupled to chiral matter on the boundaries, new singular interaction terms appear after
integrating out the auxiliary fields. The origin of these terms is clear: they are due to
the presence of the physical propagating odd field in the effective auxiliary D-term
on the boundary. At the level of the effective 4D theory, the singular terms disappear
after we substitute in the Lagrangian the solution of the classical equation of motion for
the odd field . It is worth emphasizing though that the singular terms play a crucial
role at the quantum level since they provide counterterms which are necessary in explicit
computations to maintain supersymmetry [5]. In particular, the role of the interaction term
proportional to (0) is to cancel the singular behaviour induced in diagrams where the
-field is exchanged.
101
2.2. The supergravity case

We now want to extend the analysis of the previous subsection to the case of supergravity. The on-shell supergravity multiplet contains the fnfbein eM A , the symplectic
Majorana gravitino M , and the graviphoton AM . The five-dimensional bulk Lagrangian
reads [15]
1
1
1
L bulk = M53 e5 R5 M5 e5 FMN F MN ! MNOP Q FMN FOP AQ
2
4
6 6

O P Q DM N i 3 1 e5 FMN
M N
+ iM5 ! MNOP Q
22

3 1 MNOP Q
O P Q + 4-fermion terms,
+i
FMN
!
24
(8)
where the five-dimensional coordinates are x M = (x m , x 5 ); M5 is the five-dimensional

Planck mass (we will set it equal to one from now on unless otherwise stated); e5 =
det eM A ; R5 is the five-dimensional scalar curvature; e4 = det em a , where the latter are
the components of the fnfbein with four-dimensional indices; finally, ! MNOP Q = e5
eA M eB N eC O eD P eE Q ! ABCDE , ! mnop = e4 ea m eb n ec o ed p ! abcd , ! 01235 = ! 0123 = +1.
The smallest five-dimensional off-shell supermultiplet contains 48 bosonic and 48
fermionic degrees of freedom. This decomposes into a minimal multiplet with (40 + 40)
components, containing the fnfbein, gravitino, graviphoton and several auxiliary fields
which include an isotriplet scalar t , an antisymmetric tensor vAB , a gauge field V M , a spinor , and a scalar C. In addition one has to introduce a compensator multiplet with (8 + 8)
degrees of freedom [11]. The compensator multiplet allows for the breaking of the gauged
SU(2)R symmetry. There exist several possibilities for the compensator multiplet and we
will assume that it is given by the tensor multiplet containing an isotriplet scalar Y , a
spinor , a scalar N and a vector WA (which can be expressed as a supercovariant field
strength of a three-form BMNP ). The SU(2)R symmetry is gauged by the auxiliary field
V M , and the gauge is fixed by requiring
Y = eu (0, 1, 0)T ,
(9)
where u is a scalar field. Thus after gauge fixing the field content of the five-dimensional
theory is given by

A
eM , M , AM , t , vAB , V M , , C, u, , N, BMNP .
(10)
Note that in flat space the choice of the tensor multiplet for the remaining (8 + 8)
components is not unique. However, it is advantageous to use the tensor multiplet because
it allows a straightforward generalisation to warped spaces. The off-shell bulk Lagrangian
is given by [11]

2
1 AB
1
)AB AB + 4C F
+ vAB v AB + 20t t 36 t 2
Lbulk = eu R(
AB F
4
6
1 A
1 1 A1 1 3 A3
i P MN
uA u VA V VA V + 8 35 t 2
DM N
P
4
4
4
2
102
35 2 AB
i
AB
12Nt 2 + 35 N
A B A B v
4
2
1
1
FAB v AB ABCDE AA FBC FDE 4 2 2i A A
3
6 3
1 2
1 1 M 3
1 3 M 1
1
A
A 2 AB DB
A u
VM +
VM +
+
2
2
2
2
1
A 2 AB B u
A 1 A t 3 + 2
A 3 A t 1
+ 2i
A A t 2 2
2
2

3i5 A
1
25 AM N BP QR 32t t
MNP QR VM
A
12
2
1 2 MN
1
BC

DM N +
2 B A v AB
2 ABC A F

2
2 3

i
1
ABCD
u
A AB B t

A
B e FCD + FCD + 1 eu
2
4 3

1
i
u
AB
vAB + FAB 3

2 t 2 + WA W A N 2
+e
4
3

i
A
A
DA + i

AW
(11)
2
2

1
4C e2u
2 A WA + L4F ,
N
2
AB = FAB + i( 3/2)

A B and L4F contains four-fermion interaction terms. We
where F
have also suppressed all the SU(2)R indices. All the definitions of the covariant derivatives
can be found in Ref. [11], and we have allowedfor generalitythe presence of a bulk
cosmological constant 5 (which will be set to zero later). Note that it will turn out that
on-shell we have u = 0, as can be seen from the variation of (11) with respect to C.
The fifth dimension is compactified on the orbifold S 1 /Z2 , obtained by the identification
x5 x5 . Under the orbifold symmetry fields can be classified as either even (P = +1) or
odd (P = 1). Distinguishing between four-dimensional and extra-dimensional indices,
into fourand decomposing the generic five-dimensional spinor and its conjugate
( 2 , 1 ), we
dimensional ones, with the convention that ( 1 , 2 )T and
A A
2i
assign even Z2 -parity to

1
, 52 , va5 , , C, Vm3 , V51 , V52 , t 1 , t 2 ,
1 , em a , e55 , A5 , m
(12)
and odd Z2 -parity to

2
2 , e5 a , em5 , Am , m
, 51 , Vm1 , Vm2 , V53 , vab , t 3 ,
(13)
where is the supersymmetry parameter (and from now on we will set e55 to unity unless
otherwise stated).
At the Z2 -fixed points half of the degrees of freedom are eliminated, reducing the
number of supercharges by one half. Thus, we can locate three-branes with N = 1
supersymmetric chiral matter content at the orbifold fixed points. Note that the orbifold
also breaks the SU(2)R symmetry at the fixed points to a residual U (1)R . Since in the
103
following we will be interested in the supersymmetric couplings between the gravitational

sector and the boundary matter fields, we notice that only the fields which are even under
the Z2 -symmetry will possess such couplings. For instance, only the KaluzaKlein tower
1 couples to boundary chiral matter.
of m
2.3. The intermediate multiplet
The crucial point in understanding how to couple chiral and vector multiplets to gravity
at the boundaries is to construct a four-dimensional gravitational multiplet involving
the even fields. One can show that all the even fields of the N = 2 minimal multiplet
form a non-minimal N = 1 supergravity multiplet in four dimensions with (16 + 16)
components [11]. If we conveniently define the four-component 4D Majorana spinors as
1
2
m
5
and
,
=
m =
(14)
5
1
m
52
then the N = 1 supergravity multiplet in four dimensions with (16 + 16) components is
given by

a
em , m , ba , am , , S, t 1 , t 2 .
(15)
This multiplet (15) is known as the intermediate multiplet and was first studied in Ref. [16].
The auxiliary fields are identified as [11]
ba = va5 ,

1
2 5
a
3
am = Vm Fm5 e5 + 4em va5 ,
2
3

1 5 3 3
S = C e5 5 t 5 + V51 t 2 V52 t 1 ,
2
(16)
(17)
(18)
where in particular the auxiliary

field am is a combination of Vm3 , va5 and the (even) field
m5 = Fm5 + i 3/2 m 5 of the propagating graviphoton field, AM .

strength F
The situation is then completely analogous to what happens in the case of an off-shell
bulk vector multiplet in 5D analyzed previously. There, the presence of the propagating odd
field in the effective D-term D = (X3 5 ) on the boundary induced new interactions
between the chiral matter fields living at the boundary and the field. This suggests that
after integrating out the supergravity auxiliary fields, one should expect new interaction
terms at the boundaries (compared to the usual ones in N = 1 4D supergravity coupled to
chiral or vector multiplets) involving the components of the chiral and vector multiplets and
the field strength Fm5 . A similar argument can be made for the even part of the gravitino
field 5 , even though we anticipate that 5 will be set to zero in the unitary gauge after
supersymmetry breaking.
2.4. Coupling to boundary matter fields
It is now quite clear how to couple boundary matter fields to the five-dimensional N = 2
gravity multiplet. We write
104

S=

d 5 x Lbulk +
x5 xi L4i ,
(19)
where Lbulk is given by (11) and the orbifold fixed points are located at x = 0 and x =
R. To obtain the explicit form of the boundary Lagrangian, L4 one simply uses the
expressions given in Ref. [16] and replaces the auxiliary fields of the intermediate multiplet
with the boundary values given by the expressions in Eq. (17).
In particular for an N = 1 vector multiplet (um , , D) on the boundary one finds
that [11]

1
1
i
a 3i
aD
a 5 ba m m 5 D
L4 = Tr DD u ab u ab + i
4
4
2

i
1
1
m ab
2 mn
2
2 mn 5
2 5
m u ab + m n
+ m n
,
4
4
4
(20)
where
u ab = uab i a b + i b a ,
m = Dm 5 am ,
D
(21)
and Dm is the usual covariant derivative for a gaugino field coupled to gravity.
Following the procedure of Ref. [5] and setting
1

= vAB F
vAB
(22)
AB ,
2 3
to canonically normalize the kinetic term of the graviphoton field strength, we can integrate
out the auxiliary fields for the case of a boundary vector multiplet. The Lagrangian for the
b is given by
auxiliary fields Vm3 and va5
a

3
1 3 2 1 2
1
2
m 5

Fm5 Vm3 (x5 ), (23)
bm +
Laux = 2(bm ) Vm + Fm5 i
4
2
2
2
which leads to the following vacuum expectation values
i

m 5 (x5 ).
bm
=
(24)
4
When inserted back into the Lagrangian (20), the conditions (24) give the usual fourdimensional interaction terms of a vector multiplet coupled to gravity plusas argued
abovenew interaction terms involving the graviphoton field strength, namely

2

3
3
5 1
Fa5 + i
(x5 )
a 5 .
a 5 i
d x
(25)
2
2
2
m 5 (x5 ),
Vm3 = i
After performing an x5 integration, we see that singular terms proportional to (0) appear
in the Lagrangian 1 , and as expected they contain the physical fields Fm5 and 5 which
appear in the boundary auxiliary fields (17). This situation, is therefore, analogous to
that described in Ref. [5] where a 5D vector multiplet is coupled to chiral matter on the
boundary, or to that occurring in E8 E8 strongly coupled heterotic string theory [2].
1 The same result can be recovered by a full on-shell procedure [17].
105
Since the singular terms can be written as a perfect

square, at least formally they can
3

be eliminated by a field redefinition Fa5 = Fa5 i 2 (x5 )
a 5 . At the level of the
effective 4D theory, the singular terms disappear after we substitute in the Lagrangian the
solution of the classical equation of motion for the odd field Aa . However, the singular
terms at the quantum level play a crucial role since they provide counterterms which are
necessary in explicit computations to maintain supersymmetry.
A similar, but much more involved procedure can also be followed to obtain the four-dimensional Lagrangian for a boundary chiral multiplet (, , F ).
2.5. The tensor multiplet at the boundaries
We have derived the couplings between the gravity fields of the intermediate multiplet
and the matter fields. There will also be couplings between the gravity fields of the
intermediate multiplet and the fields of the tensor multiplet, since in addition to the
minimal multiplet, parities can also be assigned to the tensor multiplet [11]. We assign
even Z2 -parity to
Y 1 , Y 2 , , N, Bmnp ,
(26)
and odd Z2 -parity to

Y 3 , Bmn5 .
(27)
On the boundary the even fields of the tensor multiplet form a chiral multiplet
(A, B, , F, G)
(28)
with chiral weight w = 2 at the fixed points. The precise correspondence on the boundary
is

5 Y 3 , +2W 5 + 12 t 1 Y 2 Y 1 t 2 .
(A, B, , F, G) = Y 2 , Y 1 , , 2N + D
(29)
Using the result for the F -term density of a chiral multiplet [16] this leads to the complete
off-shell action

1
5
S = d x e5 Lbulk + 3 W0 (x5 ) + W (x5 R) L4T ,
(30)
M5
where W0 and W are complex constants with dimension of (mass)3 , and the boundary
Lagrangian is given by
1
3 5 + i m m + eu a 2 ab b 12eu t 2 .
L4T = 2N + eu V51
2
Eliminating the auxiliary fields in (30) we finally obtain the on-shell action

1
5
2 ab

S = d x Lbulk +
W0 (x5 ) + W (x5 R) a b ,
2M53
(31)
(32)
bulk is the on-shell Lagrangian of bulk supergravity (8).

where we have set 5 = 0 and L
The Killing spinor equation M = 0 reduces to

5 = i W0 (x5 ) + W (x5 R) 5 2 .
(33)
106
Decomposing the 5D symplectic spinor into two component objects iT = (i+ , i ) (i =

1, 2), Eq. (33) has a non trivial solution only if W0 + W = 0. The solution is: 1+ = !
and 2 = i (x 5)!, where ! is a four-dimensional Weyl spinor which generates the
supersymmetry of the ground state. Therefore, the flat space solution is supersymmetric
provided that W0 + W = 0 [11].
3. Supersymmetry breaking
We now consider the case in which supersymmetry is broken. If W0 + W = 0
then we will see that the flat space solution spontaneously breaks supersymmetry. The
supersymmetry breaking will be transmitted to matter on branes located at the orbifold
fixed points via gravity. The brane action is assumed to be
1
Sbrane =
2

(0)
()
dx5 e4 (x5 )L4 + (x5 R)L4
+R

4
d x
R
1 mn 1
(x5 )W0 + (x5 R)W m
n
2M53

+ h.c. ,
(34)
where L(i)
4
are the boundary Lagrangians describing the interaction of matter with the bulk.
Expanding the fermions in Fourier modes consistently with their boundary conditions and
Z2 -parity assignments leads to

+ (x5 ) =
+ cos x5 ,
0+ + 2
R
R
=1

1
(x5 ) =
(35)
2
sin x5 ,
R
R
=1
2 , 1 ). The presence of the brane

where
stands for
and for (m
5
superpotential induces a mixing between the different KaluzaKlein levels. The fields
1 , 2 and 2 ( > 0) are Goldstinos, eaten up by the gravitinos via the super-Higgs
5,
5,0
5,
effect [10,18]. As we saw earlier when W0 + W = 0 it is still possible to define a Killing
spinor and the N = 1 supersymmetry is not spontaneously broken by the presence of the
brane superpotentials. This means that the amount of supersymmetry breaking is fixed by
the order parameter F = (W0 + W )/M5 .
The infinite-dimensional gravitino mass matrix can be easily diagonalized [10,19].
Defining
1 , 2)
(m
5

1 1
2
m,
= m,
m,
2
( > 0),
P =
1
2M53
(W0 W ),
(36)
1 and 2 combine to form nearly degenerate pairs of

one finds that the modes of m
m
Majorana states [10] with masses

()
M3/2 =
where
=
1
( + )
R
( = 0, 1, 2, . . .),
107
(37)

1
4P+
arctan
.
2 P2 ) + 4
2 (P
+
(38)
Note that the mass eigenvalues (37) are both positive and negative. Of course, the absolute
1 2
values give the physical masses,
R , R , R , . . . , where the physical range of the
supersymmetry breaking parameter is 0 1/2. Notice that in the supersymmetric
limit, W0 = W (P+ = 0) the gravitino mass spectrum remains unshifted as expected.
After the super-Higgs mechanism, from the four-dimensional point of view, the physical
spectrum contains one massless N = 1 gravitational multiplet with spins (2, 3/2) built up
1 ; one radion multiplet composed of the zero modes e ,
with the zero modes of em a and m
55
2
A5 and 5 and an infinite series of massive multiplets of N = 2 supergravity with spins
(2, 3/2, 3/2, 1).
Let us now consider two interesting physical limits in the instance where W0 vanishes
identically and W is nonzero. This means that P+ = P = W 3 . In this case the only
2M5
source of supersymmetry breaking appears as a constant superpotential on the hidden

brane.
(i) If the absolute value of the superpotential |W | is much smaller than M53 , |W |
M53 , the function is well approximated by P+ = W 3 . This means that the gravitino
zero-mode mass is given by
(0)
M3/2 =
W
,
M42
2M5
(39)
where we have invoked the relation M42 M53 R. This is the familiar four-dimensional
expression for N = 1 supergravity. The other massive modes are well separated from the
lowest mode by a multiple of R 1 . This means that the low-energy effective four-dimensional theory (describing the physics below the scale R 1 and consisting only of zero
modes) is simply the N = 1 supergravity theory with supersymmetry spontaneously broken
by the nonvanishing superpotential W .
(ii) If the absolute value of the superpotential |W | is much larger than M53 , |W |
3
M5 ), then the function is well approximated by 1/2. This means that the gravitino zeromode mass is given by
1
.
(40)
2R
Notice that the zero-mode gravitino mass no longer depends on the superpotential
parameter. All the massive modes are again separated from the zero mode by a multiple
of R 1 . Therefore, the gravitino mass spectrum is identical to that obtained from the
ScherkSchwarz supersymmetry breaking mechanism [13] which makes use of nontrivial (anti-periodic) boundary conditions for the five-dimensional gravitino field upon
compactification of the fifth dimension. This observation will turn out to be useful in the
(0)
M3/2 =
108
following when we will show how supersymmetry breaking is communicated to the visible
brane at one loop through the gravitational sector living in the bulk. Indeed, mass splittings
for the observable fields have been computed in the context of M-theory [3] where
supersymmetry breaking by gaugino condensation in the strongly coupled heterotic string
can be described by an analogue of ScherkSchwarz compactification on the eleventh
dimension. At the lowest order, supersymmetry is broken only in the gravitational and
moduli sector at a scale m3/2 R 1 , where R is the radius of the eleventh dimension, and
it is transmitted to the observable world by gravitational interactions. We will, therefore,
be able to reproduce the results of Ref. [3] in the limit |W | M53 .
3.1. From the interaction to the mass gravitino eigenstates
The infinite unitary matrix U which diagonalizes the infinite gravitino mass matrix,
M3/2 is defined by
UM3/2 U = MD ,
(41)
where MD is the diagonal mass matrix whose eigenvalues are given in (37). Knowledge
of the unitary matrix U is necessary in order to perform the one-loop computation of
the soft supersymmetry breaking masses of the fields living on the boundaries because
the interactions are not in the mass eigenstate basis. Correspondingly, the gravitino mass
are obtained from the relation
eigenstates
= U,
(42)
where = (01 , 1+ , 1 , 2+ , 2 , . . .) represents the infinite gravitino eigenvector for

the mass matrix M3/2 . For arbitrary values of the brane superpotentials, W0 and W , the
with mass eigenvalue M() is given by
gravitino mass eigenvector
3/2

()
()
()
()
()
()
= N() 1,
,
,
,
,
,
,
() 1 () + 1 () 2 () + 2 () 3 () + 3
(43)
where () = + , N() is a normalisation constant and
=
2P+ P
2 + P 2 + (P 2 P 2 )(1 + P tan )
P+
+
(44)
has norm equal to unity gives for the normalisation constant

Requiring that the vector
N() =
1
()
2 sin ()
1 + 2 + (1 2 ) cos ()
The matrix U can therefore be written as
(45)
N(0)
N(1)
U = N(2)
N(3)
..
.
(0) N(0)
(0) 1
(0) N(0)
(0) +1
(0) N(0)
(0) 2
(1) N(1)
(1) 1
(1) N(1)
(1) +1
(1) N(1)
(1) 2
(2) N(2)
(2) 1
(2) N(2)
(2) +1
(2) N(2)
(2) 2
(3) N(3)
(3) 1
(3) N(3)
(3) +1
(3) N(3)
(3) 2
..
.
..
.
..
.
...
109
...
... .
...
..
.
(46)
It is not difficult to check that this matrix is unitary. In addition an infinite sum over the
unitary matrix elements can be performed and leads to the result
1 + (1 ) cos (k)
(47)
,
(1)n Ukn =
2 1 + 2 + (1 2 ) cos (k)
n=

. In particular, for W = 0 and W = 0 (i.e., = 1) we
and similarly for n (1)n Ukn
0
obtain
Ukn = 1 and
n=
(1)n Ukn = cos (k) ,
(48)
n=
while for W0 = 0 and W = 0 (i.e., = 1) we have
Ukn = cos
(k)
and
n=
(1)n Ukn = 1.
(49)
n=
These relations will be useful later when we consider the gravitational interaction of
gravitinos with boundary matter.
3.2. 5D interpretation
Before closing this section, we wish to give an alternative derivation of the mass
eigenvalue formula (37), and a more transparent explanation of the relations (48) and (49).
Therefore, let us consider the equations of motion of the five-dimensional gravitino fields
after we have set W0 = 0 and chosen the unitary gauge (the opposite case in which W = 0
can be analyzed in a similar fashion). The gravitino equations of motion read
! mnpq n p q1 + 2 mn 5 n2 + 2
W
M53
! mnpq n p q2 2 mn 5 n1 = 0.
(x5 R) mn n1 = 0,
(50)
(51)
If we now write ni (xa , x5 ) = ni (xa )fi (x5 ) (i = 1, 2), and integrate Eq. (50) in the interval
(R , R + ) we obtain
n2 (xa ) =
W f1 (R) 1
n (xa ).
2M53 f2 (R)
(52)
110
The ni component of the gravitino satisfies the RaritaSchwinger equation
! mnpq n p qi = 2 mn ni ,
R
(53)
and using the relation (52) one gets from Eqs. (50) and (51) that f1 (x5 ) cos( R x5 ) and
f2 (x5 ) sin( R x5 ) together with the consistency relation
tan() =
W
2M53
(54)
The solution of this equation reproduces the KaluzaKlein mass spectrum () = +

given in Eq. (37) for P+ = P = W 3 . Note also that (52) reduces to n2 (xa ) = n1 (xa )
2M5
when we use the relation (54).

From this 5D picture we can also extract a transparent interpretation of the relations (48).
1 couples to the boundaries, the interaction between
Since only the even gravitino m
1
the gravitino mass eigenstates m (xa ) and the boundary chiral and vector multiplets are
accompanied by the five-dimensional wave-function f1 (x5 ) cos( R x5 ). Such a wavefunction is equal to unity if matter and vector multiplets live on the x5 = 0 boundary, or
alternatively equal to cos(() ) if they live on the x5 = R boundary. This is precisely
equivalent to the relation (48). We can also alternatively describe the limit in which there
is a large supersymmetry breaking on the brane at x5 = R, W M53 . In this case
the eigenvalue () tends to the value ( + 12 ) and the wavefunction of the gravitino is
suppressed at the brane where supersymmetry breaking occurs. In other words, for simple
energetic reasons, the gravitino prefers to live in the bulk as far away as possible from the
boundary where it acquires a large mass.
4. Communication of supersymmetry breaking via the bulk

As we have seen in the previous section, the introduction of a constant superpotential
W on the brane located at x5 = R induces a breaking of supersymmetry in the five-dimensional gravitational sector, while it remains unbroken in the visible sector living on
the brane located at x5 = 0. The communication of supersymmetry breaking to the visible
sector is then expected to arise radiatively via gravitational interactions. This is the issue
which we will study below.
4.1. Boundary vector multiplet
Let us consider a vector supermultiplet (um , , D) on the boundary at x5 = 0. Our goal
is to study how supersymmetry breaking on the brane at x5 = R is transmitted by gravity
to the boundary vector supermultiplet.
The coupling between the intermediate multiplet representing bulk gravity and boundary
vector multiplets is given in (20). At tree-level the gauginos are all massless, but
the interaction with gravitinos will induce a one-loop radiative mass. From the parity
1
will couple to the boundary
assignments in (35) only the even gravitino modes, m,
111
Fig. 1. The one-loop diagrams containing the gravitino m which give contributions to the gaugino
mass.
gauginos. Inspection of the off-shell Lagrangian does not reveal any new couplings
between gaugino fields and the gravitino. Instead the new couplings that do appear, such
as the (0) terms, are required in order to obtain the usual result consistent with the N = 1
supersymmetric limit.
The interaction between the gauginos and the gravitons do not give any contribution
to the gaugino mass (as can be easily understood from chirality arguments). Therefore,
there is no supersymmetric cancellation between graviton and gravitino loops; the gravitino
contributions have to sum up and give a finite result [20].
We work in the harmonic gauge m m = 0 (integrating out the auxiliary field in the
bulk Lagrangian (11) gives A A = 0 which reduces to m m = 0 for the gauge choice
5 = 0. Note that there is no dependence in L4F [11]). In N = 4 + ! dimensions, every
KaluzaKlein state of the gravitino field gives for the gravitino and gauge loops of Fig. 1,
the respective contributions

(3 + !)
k2
1
dN k
iG1a =
(55)
(4
+
!)(3
+
!)
2
(2
+
!)
,
m
n
12
(2)N
mn
k 2 + m2n

k2
(3 + !)
1
dN k
2
iG1b =
(56)
.
mn 12 + 3! ! + 2 (2 + !) 2
N
12
(2)
mn
k + m2n
We see that the leading divergent parts going like k 2 cancel exactly. This means that every
KaluzaKlein gravitino state gives a contribution to the gaugino mass m of the form

dN k
mn
!
(57)
.
(2)N k 2 + m2n
The next step is to isolate the divergent pieces in the (sum of the) integrals (57). To
do that, one can simplify the sum over the KaluzaKlein states using the contour trick
from finite temperature field theory which allows one to identify the different possible
divergences [21]. We can write the sum as

mn
k5
1
=
dk5
P(k5, ),
(58)
2 + m2
2
k
2i
k + k52
n
n=
C
where the contour C is the line from the left to the right below the real axis and another
line from the right to the left above this axis and enclosing the poles at k5 = (n + )/R of
112
the function [21]

P(k5 , ) =
1
R

2 tan R k5

.
(59)
Notice that in the limit of exact supersymmetry ( = 0) the sum (58) vanishes identically
as it should.
The expression (58) can be rewritten as
1
2i
+
dk5
k5
P(k5 , ) P(k5 , ) .
2
k 2 + k5
(60)
The dimensionally regulated integrals usually split into two pieces: a pure 5D divergent
piece and a finite piece. The constant parts 12 R of the pole functions P(k5 , ) induce
the pure 5D divergent pieces, butsince they do not depend upon they sum up to zero.
The remainders of the functions P(k5 , ) are highly convergent functions and each
gives a finite contribution to the momentum integration andthereforeno contribution
to the gaugino masses after we set ! = 0. One can understand this result using the analogy
with what happens in 4D [20]. There gaugino masses are nonvanishing at one-loop if there
is no physical cut-off in the theory. However, if a physical cut-off c is present (such as the
scale of gaugino condensation) one has to cut off the integrals (55) and (56) at k 2 = 2c and
work in exactly four dimensions (! = 0). The two contributions (55) and (56) then exactly
cancel and we are left with no contribution to gaugino masses from massive gravitinos.
In 5D the five-dimensional divergent pieces cancel and each finite loop contribution to
gaugino masses can be seen as potentially 4D divergent integrals made finite by setting a
cutoff at p R 1 . Therefore we find that the gaugino mass do not get any contribution at
the one-loop order if gravity is the mediator of supersymmetry breaking through the bulk.
This result will be confirmed in Subsection 4.3 by a full 5D computation.
Note that if the gauginos were living on the 3-brane at x5 = R then the diagrams

k
k . In this case
Ukm
in Fig. 1 would be proportional to the sum n,m (1)n+m k Unk

2
(k)

the sum would reduce to k cos ( )k k . As we learned at the end of Section 3,
the cos2 ((k) ) factor represents the wave-function suppression of the gravitino at the
boundary where supersymmetry is broken. In such case, the gravitino sum is multiplied
by the coefficient n = cos2 [(n + )]. Repeating the procedure adopted above, we again
find that gaugino masses vanish at one-loop.
4.2. Boundary chiral multiplet
We now add a generic matter chiral supermultiplet (, , F ) on the boundary at
x5 = 0. The tree-level scalar masses are zero and will again be induced at the one-loop
level. We assume that the supersymmetry breaking is on the brane at x5 = R and is
transmitted by gravity to the boundary matter supermultiplet.
The interaction terms between bulk gravity and the chiral multiplet on the boundary can
be found by coupling the supergravity intermediate multiplet (15) to the chiral multiplet.
The resulting Lagrangian is quite involved [16] and we do not report it here. Integrating out
113
all the auxiliary fields turns out to be a complicated task. Apart from the usual interaction
terms present in the N = 1 4D supergravity Lagrangian coupled to chiral matter, new
singular interaction terms appear. However, as we already explained in Section 2, these
m5 and the gravitino component 5 . In
new terms may involve only the field strength F
m
particular, the field strength and the current J = i( m m ) + m combine
to form again a perfect square

2

3
3
5 1
Ja (x5 ) .
d x
(61)
Fa5 + i
a 5 i
2
2
2
These interaction terms induce a one-loop correction to the scalar masses. In particular,
the diagram where the odd graviphoton field Am is exchanged leads to singular behaviour
which is, however, cancelled by the singular Ja J a (0) term. This can be seen by writing
(0) as
(0) =
1 k 2 (n/R)2
.
2R n= k 2 (n/R)2
(62)
We find
m2

(n/R)2
1
d 4k
3
+ (0)
M5 n= (2)4 2R k 2 (n/R)2

k2
d 4k
1
,
=
2M42 n= (2)4 k 2 (n/R)2
1

(63)
where we have used the relation M42 M53 R. This contribution to the scalar masses,
when summed up to the contribution of the whole tower of massive gravitino states
belonging to the N = 2 multiplet in 4D, gives a finite result. Similarly, the diagram where
the even field A5 propagates is cancelled, in the supersymmetric limit, by the contribution
from the 52 gravitino (they live in the same radion multiplet). In the unitary gauge the
gravitino component 5 is eaten up, and cannot induce such a cancellation. Its role is then
1 gravitino.
played by the lightest mode of the m
After these general comments we now proceed to the explicit computation of the oneloop scalar masses induced by gravity after supersymmetry breaking. It is important to
notice that the coupling between the supergravity intermediate multiplet (15) and the
boundary chiral multiplet contains a generic Khler potential (, ) of the scalar
fields . Thus, the procedure described in Ref. [16] gives rise to noncanonical kinetic
terms of the form
Skin =
1
2M53
R
4
d x
R

1
1 klmn 1
1
1
1
k l mn k l mn
dx5 e4 (x5 ) R4 !
6
2

1
1
mn mn
+ h.c. ,
2
(64)
114
where
1
1
= m n1 n m
,
mn
(65)
(66)
a (x = 0) of the intermediate
and R4 is the Ricci scalar computed using the vierbein em
5
multiplet.
At this stage one is free to perform a Weyl rescaling that renders the gravity and gravitino
kinetic terms canonical. However, we prefer to compute the one-loop scalar masses in the
unrescaled Weyl basis using the Lagrangian (64). This choice is dictated by the fact that
in the unrescaled Weyl basis both gravitons and gravitinos give rise to one-loop scalar
masses and the supersymmetric cancellations are more transparent. Furthermore, in this
basis there are no direct couplings between the radion supermultiplet containing the even
fields e55 , A5 and 52 (the radion, for instance, arises from fluctuations of g55 ; by general
covariance it can only couple to the 55-component of the matter energy-momentum tensor
which vanishes for chiral matter on the brane [22]). On the contrary, in the Weyl rescaled
basis gravitons do not give a contribution to the scalar masses if we start from vanishing
tree-level scalar masses and supersymmetric cancellations are hidden.
At the one-loop level, the diagrams that contribute to the scalar masses are those shown
in Fig. 2, where the graviton and gravitino vertices come from the kinetic terms (64). As
we already noticed, the fields from the boundary in the N = 1 chiral multiplet (, , F ),
always appear in pairs, as dictated by the Z2 invariance. Just like the gaugino case, only
the parity even gravitinos couple to the boundary chiral multiplet.
Fig. 2. The one-loop diagrams which give contributions to the scalar mass.
115
Every diagram is proportional to the following integral

1
k2
1
d 4k
d 4k 2
2
2
= 2
k
ds es(k +mn )
2
4
2
2
4
(2)
k
+
m
(2)
M4 n
M4 n
n
1
8 2 M42

n
ds sm2n
e
.
s3
(67)
As we saw for the gaugino case each diagram containing a gravitino in Fig. 2 gives rise to

k
k where the coefficient k depends on the location of the boundary
the sum k k
chiral multiplet. Thus, after subtracting the fermionic contributions from the bosonic
contributions the scalar mass-squared will be proportional to
1
IB IF =
2
8 M42

0

ds sn2 /R 2
s(n+)2 /R 2
e
e
,
n
s3 n
(68)
where we have made use of the expression (37). The infinite sum can be performed using
standard techniques as explained in Appendix A. For chiral matter on the brane x5 = 0,
the coefficient k = 1, and the sum of the diagrams in Fig. 2 gives a contribution to the
soft-breaking mass squared

2i
1 2i
1
1
Li
(5)
e
+
Li
e
,
K
m2 = K
(69)
5
5

2
8 6 R 4 M42
1
where K(, ) = 3 ln[(, )/3] is the Khler potential and the term K
K
comes from the wave-function renormalization. The supersymmetry-breaking contributions only come from the gravitino diagrams, Fig. 2(a) and (b). The fact that the sign of m2
is negative can be easily understood by working in the rescaled Weyl basis. In that basis
loops involving gravitons do not give any contribution to the scalar masses which can only
be induced by gravitino loops. The latter carry a negative sign because of their fermionic
nature.
The one-loop scalar mass-squared induced by the gravitational transmission of supersymmetry breaking is negative and diagonal in flavour space provided that the matter metric is diagonal. However, introducing other moduli fields z in the bulk with gravitational
strength coupling to the boundary fields, one can get similar contributions to (69) and
make the total scalar mass-squared positive [3]. In such case the Khler potential will be a
function of all scalar fields, K = K(, , z, z ), and different moduli dependence for the
various scalar fields may create potentially dangerous flavour-changing neutral currents.
Let us now take two different and physically interesting limits of the expression (69).
Consider first the case where the chiral multiplet is on the brane at x5 = 0. If the absolute
value of the superpotential |W | is much smaller than M53 , (|W | M53 ), then we have
seen that the function is well approximated by W 3 . Using the fact that for 0 we
have
2M5

1 2i
Li5 e
+ Li5 e2i (5) 2 2 (3)2 + O 4 ,
2
(70)
116
we find that the leading contribution from the diagrams in Fig. 2 to the soft breaking scalar
mass is proportional to

2
2
W
1
(0)
2
m
(71)
M
,
RM4 3/2
R 2 M4 M53
where we have used (39). In this limit we see that the scalar mass acquires a suppression
factor (RM4 )1 relative to the gravitino mass. When RM4 10 then the scalar mass
will be comparable to the anomaly-mediated contribution [6]. The result (71) can be
easily understood by noting that in the effective 4D supergravity there are no quadratic
divergences. The (divergent) one-loop scalar mass squared result in four dimensions would
be
m2
m23/2
M42
m23/2 2
d 4p
,
p2 + m23/2
M42
(72)
where is the ultraviolet cutoff. However, in the brane world scenario the gravitino
contribution to the scalar masses is the sum of a pure 5D-divergent piece (cancelled by
the graviton contribution) and a finite piece
m2
1/R
m23/2
M42
d 4p
p2 + m23/2
m23/2 1
,
M42 R 2
(73)
wheresince the gravitino interacting with the chiral matter at x5 = 0 has to travel
a distance at least as large as the radius of compactification to probe supersymmetry
breaking at x5 = Rthe loop integral is over gravitino momenta satisfying p < R 1
(gravitino wavelengths larger than R). Loop momenta p > R 1 are not sensitive to the
supersymmetry breaking effects on the brane at x5 = R. Thus, the effective ultraviolet
cutoff is provided by the interbrane distance and substituting = 1/R reproduces the
result (71). This situation resembles what happens in 4D theories at finite temperature
where the ultraviolet cutoff is represented by the temperature T . In the imaginary time
formalism 4D loop integrals become integrals over the three spatial momenta and a sum
over the so-called Matsubara frequencies and the one-loop contributions to the mass
squared of an interacting scalar field can be split into the usual zero temperature 4D
divergent piece plus a finite temperature dependent piece. The finiteness is due to the fact
that particles in the plasma with wavelengths smaller than T 1 (or momenta larger than T )
have exponentially suppressed abundances.
(0)
From Eq. (71) we find that for M3/2R 1 (1011GeV)2 , the soft scalar masses are of
order of the TeV scale. For instance, if W /M53 102 , we get M5 5 1016 GeV and
R 1 1012 GeV.
If the absolute value of the superpotential |W | is much larger than M53 , |W | M53 ,
then we have seen that the function is well approximated by 1/2, and the polylogarithm
functions can be expanded as

1 2
15
1 2i
Li5 e
+ Li5 e2i (5) + O
.
(74)
2
16
2
117
93
In such case we find that for large values of |W | the integral I = 32
5 (5), as defined
in Appendix A. This reproduces the value found in Ref. [3] where supersymmetry was
broken by boundary conditions via the ScherkSchwarz mechanism. This does not come
as a surprise, though. As we have already pointed out, the gravitino spectrum in the
limit |W | M53 , is exactly the same as obtained in the ScherkSchwarz supersymmetry
breaking mechanism. The soft breaking scalar mass is therefore proportional to
(0) 2
M3/2
1

,
m 2
R M4
M4
(75)
where we have used (40) and there is an RM4 suppression. Again, this result can be
understood by noting that the effective cutoff for the brane world scenario is = 1/R.
If R 1 1011 GeV, we find m of the order of TeV.
It is also instructive to discuss the alternate possibility when the chiral matter is located
on the same brane where supersymmetry breaking occurs (n = cos2 (n+) ). In this case
since loop momenta are on the same brane that has a nonzero superpotential the interbrane
distance R no longer plays any role. As happens in the four-dimensional case, scalar masses
should be ultraviolet sensitive to the cutoff . This is because gravitinos are now repelled
from the brane and the cancellation of the five-dimensional divergences no longer takes
place. This can be explicitly seen by performing the integral (68) and the details can be
found in Appendix A. If the supersymmetry breaking parameter |W | is much smaller than
M53 one finds that m M(0)
3/2 (/M4 )(R) which is the usual 4-dimensional divergences
increased by the number of KaluzaKlein states
(R) excited up to the cutoff scale. If
|W | is much larger than M53 , one finds m R (2 /M4 ), which reflects the absence
of the cancellation of the pure five-dimensional divergences.
4.3. The 5D calculation
The communication of supersymmetry breaking to the boundary matter fields can also
be obtained using the five-dimensional propagator for the gravitino field. In order to
calculate the 5D propagator it is simplest to Fourier transform the four-dimensional spatial
coordinates, while leaving the fifth spatial coordinate x5 explicit [23]. The propagator of
the massless gravitino in 5D can be written in the form

1
G (k, x5 ) = G(k, x5 ) g G(k, x5 )P ,
(76)
3
where we have omitted the longitudinal parts because they do not give any contribution
to the gaugino and scalar masses. After the addition of the boundary mass term this
reproduces the propagator of the gravitino transverse degrees of freedom. To evaluate
G(k, x5 ) in the case of supersymmetry breaking we have to solve, starting from Eqs. (50)
and (51) and their conjugates, the equations for the Greens function G1,2 . For simplicity
we assume that there is a boundary mass term, m W /M53 , at x5 = R. Following the
procedure developed in Refs. [24,25] we first solve for the gravitino propagator in infinite
118
space. It reads

2
1 (kE , x5 ) = i kE ekE |x5 | m ekE RkE |x5 R|
G
2kE
4 + m2
i
m
+
5 ekE |x5 |
ekE RkE |x5 R| ,
2kE
4 + m2
(77)
where we have analytically continued the propagator to be a function of the Euclidean fourmomentum kE . Let us consider an arbitrary point x5 in the interval [0, R]. The amplitude
1 (kE , x5 ). But since our space is compact and not
to propagate from x5 to x5 is simply G
infinite, the point x5 + 2nR where n is an integer, is equivalent to x5 and we need to
1 (kE , x5 + 2nR). This sum can be easily performed and
sum over all the contributions, G
leads to the result for a compact space with 0 x5 R

i kE
cosh kE (R x5 )
(k
,
x
)
=
G()
E 5
1
2kE sinh(kE R)

m2 cosh(kE x5 ) cosh(kE R)
4 sinh2 (kE R) + m2 cosh2 (kE R)

m cosh(kE R)

i kE sinh[kE (2R x5 )] + cos(2) sinh(kE x5 )
=
2kE
cosh(2kE R) cos(2)
cosh(kE x5 ) sin(2)
,
(78)
2[cosh(2kE R) cos(2)]
where m = 2 tan , using the relation (38) for W0 = 0. Similarly, one can follow the same
procedure to obtain for 0 x5 < R

i kE
()
G2 (kE , x5 ) =
cosh kE (R x5 )
2kE sinh(kE R)

2m sinh(kE x5 )
m2 sinh(kE x5 ) coth(kE R)
2[4 sinh2 (kE R) + m2 cosh2 (kE R)]
(79)
Notice that we have not written the 5 terms since they give no contribution to the masses.
Let us consider the case where matter is on the brane at x5 = 0. We find that the G2
propagator is particularly simple,
G()
2 (kE , 0) =
i kE
coth(kE R),
2kE
(80)
and does not depend on the supersymmetry breaking parameter, . However, the G1
propagator becomes
G()
1 (kE , 0) =
i( kE )/kE sinh(2kE R) sin(2)

.
(81)
119
In the supersymmetric limit, 0 we recover the usual 5D propagator

(0)
G1 (kE , 0) =
i kE
coth(kE R),
2kE
(82)
while in the maximally supersymmetry-breaking limit, 1/2 (or ScherkSchwarz

limit), we obtain
(1/2)
G1
(kE , 0) =
i kE
tanh(kE R).
2kE
(83)
The propagator (81) thus continuously interpolates between these two limits. Notice also
that after analytically continuing the momentum back to four-dimensional Minkowski
space the poles of the propagator (81) occur at the values ikE k4 = (n + )/R, in
complete agreement with (37). It is also interesting to consider the four-dimensional limit
kE R 1. In this limit the propagator (81) becomes
()
G1 (kE , 0) =
1 i kE /R
,
2 + 2 /R 2
2R kE
where we have also taken the limit 0.

When matter is located on the brane at x5 = R we obtain

i kE
cos2 () sinh(kE R)
()
G1 (kE , R) =
kE
cosh(2kE R) cos(2)
cosh(kE R) sin(2)
,
(84)
(85)
and we see that there is an extra cosine factor which can be thought of as being due to the
wavefunction of the gravitino at x5 = R.
It is now straightforward to obtain the one-loop contributions to the boundary matter.
Consider first the case of the gaugino mass on the boundary x5 = 0. One finds that the
contribution to the gaugino mass from the Feynman diagrams in Fig. 1(a) is simply

sin(2)
d 4k
i
3
4 2[cosh(2k R) cos(2)]
(2)
4M5
E
=
i
1
64 2 (R)4 M53
dx x 3
0
sin(2)
cosh(2x) cos(2)

1
3 2i
Li4 e
Li4 e2i ,
=
5
2
3
64 8M4 R
(86)
and similarly for Fig. 1(b)

1
3 2i
Li4 e
Li4 e2i .
5
2
64 8M4 R 3
(87)
Thus, as expected we see that the sum of the one-loop contributions to the gaugino
mass is zero. This also agrees with the result in Ref. [20], which found that there are
no gravitational radiative corrections when the theory has a cutoff. In the 5D theory the
120
effective cutoff is 1/R, and we obtain a similar result. Notice also that each separate
contribution to the gaugino mass vanishes identically for = 0, 1/2. In the KaluzaKlein
picture this can be easily seen since for = 0, 1/2 the KaluzaKlein mass spectrum is
symmetric about zero mass and leads to a vanishing sum (58). Similarly, one obtains the
same results for matter on the brane at x5 = R, since the only relevant difference is the
cosh factor in the propagator (85).
In the case of the boundary scalar fields located at x5 = 0, and using the Lagrangian (64)
the contribution to the scalar mass-squared from the gravitino loops is given by
d 4 k
1
!
Tr 5 k G (k, 0)P ,
(88)
3
4
(2)
6M5
()
() . There are also contributions from the graviton
where G (k, 0) contains G1 and G
1
and graviphoton, which in the limit of zero supersymmetry breaking must cancel the gravitino contribution (88). Thus, the total contribution to the scalar mass-squared is given by

sinh(2kE R)
d 4k
1
2 1
coth(kE R)
(8)kE
(2)4
2kE
cosh(2kE R) cos(2)
6M53

sinh(2x)
1
4
dx x coth(x)
=
cosh(2x) cos(2)
12 2 (R)5 M53
0

2i
1
1 2i
+ Li5 e
=
(89)
(5) Li5 e
.
2
8 6 R 4 M42
This agrees with the result obtained from the KaluzaKlein sum (69). Notice that we obtain
a finite result because the leading divergences cancel in the integral (89), and the remaining
part is exponentially suppressed.
When matter is located on the brane at x5 = R this cancellation no longer happens
because the gravitino propagator (85) contains a cosine factor which depends on . Only
in the supersymmetric limit ( = 0) does the cancellation in (89) occur. This is just the
5D interpretation of the result (A.13) that we obtained earlier from the KaluzaKlein
summation.
5. Conclusions and discussions

In this paper we have considered a supersymmetric five-dimensional brane world
scenario where the fifth dimension is compactified on S 1 /Z2 . In our set-up chiral matter
and gauge fields are restricted to live on the boundaries while gravity propagates in the
bulk. We have assumed that supersymmetry is broken at the orbifold fixed points and that
supersymmetry breaking is parametrized by a constant boundary superpotential. The bulk
gravitino mass spectrum is consequently shifted relative to the bosonic bulk supergravity
fields. Integrating out the supergravity auxiliary fields allows one to derive the couplings
between the boundary chiral matter or gauge fields and the bulk supergravity fields. If chiral
121
matter or gauge fields live on a brane different from the one where supersymmetry breaking
occurs, the latter is communicated to our observable world by gravitational interactions.
We have computed the contribution to the soft supersymmetry breaking scalar and gaugino masses for generic values of the brane superpotential and the radius of compactification. The one-loop computation of the mass splittings generically provides a hierarchy of
soft masses at the TeV scale with nonvanishing scalar masses and zero gaugino masses.
The nonvanishing superpotential on the boundaries and the relative shift in the bulk
gravitino mass spectrum parametrize various possible sources of supersymmetry breaking
at the fixed-points coming from FayetIliopoulos D- and F -terms inducing nonvanishing
brane superpotentials to keep the branes tensionless. In this paper we have assumed that
these terms are coupled to the bulk only through supergravity. In addition for certain values
of the radius R the anomaly-mediated contributions of scalar masses will be of roughly
the same order. It may be that a combination of brane supersymmetry breaking and the
anomaly-mediated contribution can solve the well-known slepton mass problem. On the
other hand since the one-loop gaugino masses vanish, the anomaly-mediated gaugino mass
contributions will dominate.
In deriving the results of this paper we have worked with the components of the
supermultiplets. In principle, the same results can also be more simply derived using the
N = 1 superfield calculus, such as that considered in Refs. [26] for the case of bulk vector
fields. This would require writing the N = 2 supergravity Lagrangian in terms of N = 1
superfields, and then repeating the procedure already done for the bulk gauge fields.
One should note that the supersymmetry breaking mechanism respects the tensionlessness of the branes even though we have added matter on the branes. This is because the
Goldstino is a bulk field. Of course, a brane tension may eventually be generated due to
gauge symmetry breaking (or even other forms of brane supersymmetry breaking). This
may require considering more general warped bulk backgrounds. Since we have used the
tensor multiplet it is straightforward to generalize the procedure used here for warped
geometries. In particular, one could obtain the gravitational analogue of the warped soft
mass spectrum calculated in Ref. [9].
Finally, even if the branes remain tensionless, the vacuum energy is nonzero [10] and
the radius is not stabilized. Thus, one is likely to require the addition of further fields in
the bulk in order to stabilize the radius [22,27]. These fields will arise when one embeds
the brane world setup in a more fundamental theory, such as string theory, and deserves
further study.
Acknowledgements
We wish to thank I. Antoniadis, A. Brignole, F. Feruglio, Z. Lalak, A. Pomarol,
M. Quiros, R. Rattazzi, M. Zucker and F. Zwirner for useful discussions. T.G. thanks
the University of Padova Theory Group and A.R. thanks the CERN Theory Group for
hospitality where part of this work was done. The work of T.G. is supported by the Swiss
National Science Research Fund (FNRS) contract no. 21-55560.98.
122
Appendix A
We provide the details on evaluating the infinite KaluzaKlein sum (68). The key is to
introduce the -functions defined as

2
( ) =
(A.1)
e2in ei (n+) ,
n=
which obey the Poisson resummation formula
(1/ ) = i e2i ( ).
(A.2)
Consider the expression (68) with k = 1 which can be rewritten in the form
1
IB IF =
8 2 M42
=
1
8 2 M42

ds
is
is
0
0
0
s3
R 2
R 2

2
ds R 2 1/2 0 iR 2
0 iR
0
s
s
s
s3
(A.3)
where we have used (A.2). After redefining the integration variable to be y = R 2 /s, we
obtain

1
0
IB IF =
(iy)
4 2 dy y 3/2 00 (iy)
8 R M4
0
8 R 4 M42

dy y
3/2

2 2
1 e2in e n y
n=
I,
8R 4 M42
(A.4)

1
2 2
I=
dy y 3/2
1 e2in e n y .
n=
(A.5)
When n is nonzero the exponential factor guarantees that each term in the sum has a finite integral. However, notice that the potentially dangerous n = 0 term vanishes. Thus,
simplifying the infinite sum gives

2 2
2 2
1 e2in e n y = 4
sin2 (n)e n y ,
n=
(A.6)
n=1
so that performing the y integration and summing up the finite integral pieces gives the
finite answer
3 2
1
I= 5
sin (n) 5
n
n=1
123

2i
3
1 2i
Li
e
+
Li
e
(A.7)
(5)
,
5
5
2 5
2

xn
where Lik (x) =
n=1 nk are the polylogarithm functions. Thus, substituting the value of
I back into the expression (A.4) gives the final result

2i
1 2i
1
3
(5) Li5 e
IB IF =
(A.8)
+ Li5 e
,
2
2 5 8R 4 M42
=
which leads to the result (69).

In the case where the chiral matter lives on the same brane as the supersymmetry
breaking, the evaluation of the infinite sum is only slightly more involved because of the
presence of the factor n = cos2 (n + ) . In this case (68) becomes
1
IB IF =
8 2 M42

0

ds sn2 /R 2
2
2
e
cos2 n + es(n+) /R ,
3
s n=
(A.9)
and the only complication involves the evaluation of the infinite sum with the cosine factor.
Using the expansion cos[(n + )] = [ei(n+) + ei(n+) ]/2, one can easily show that

R 2 0
2
2
(A.10)
iR 2 /s .
cos2 (n + ) es(n+) /R = cos2 ()
s
n=
Thus, after changing variables to y = R 2 /s the infinite sum (A.9) can be rewritten in the
form

1
0
iy
IB IF =
4 2 dy y 3/2 00 (iy) cos2 ()
8 R M4
0
1
4 2
8 R M4

dy y 3/2
0

2 2
1 cos2 ()e2in e n y
n=
1
I ,
8R 4 M42

1
2 2
1 cos2 ()e2in e n y .
dy y 3/2
I =
n=
(A.11)
(A.12)
Notice now that the n = 0 term in the sum no longer vanishes, and there is a divergent
piece in I . Thus, simplifying the infinite sum gives
2 2 2
1 cos2 ()e2in e n y
n=
= sin2 + 2

n=1
2 2 2
1 cos2 () cos(2n) e n y .
(A.13)
124
Only in the limit = 0 does the divergent piece from the n = 0 term vanish. This makes
sense since there is now no superpotential breaking term.
References
[1] I. Antoniadis, Phys. Lett. B 246 (1990) 377.
[2] P. Horava, E. Witten, Nucl. Phys. B 460 (1996) 506, hep-th/9510209;
P. Horava, E. Witten, Nucl. Phys. B 475 (1996) 94, hep-th/9603142;
P. Horava, Phys. Rev. D 54 (1996) 7561, hep-th/9608019.
[3] I. Antoniadis, M. Quiros, Nucl. Phys. B 505 (1997) 109, hep-th/9705037.
[4] H.P. Nilles, M. Olechowski, M. Yamaguchi, Phys. Lett. B 415 (1997) 24, hep-th/9707143;
H.P. Nilles, M. Olechowski, M. Yamaguchi, Nucl. Phys. B 530 (1998) 43, hep-ph/9801030;
H.P. Nilles, hep-ph/0004064.
[5] E.A. Mirabelli, M.E. Peskin, Phys. Rev. D 58 (1998) 065002, hep-th/9712214.
[6] L. Randall, R. Sundrum, Nucl. Phys. B 557 (1999) 79, hep-th/9810155.
[7] I. Antoniadis, S. Dimopoulos, A. Pomarol, M. Quiros, Nucl. Phys. B 544 (1999) 503, hepph/9810410;
A. Delgado, A. Pomarol, M. Quiros, Phys. Rev. D 60 (1999) 095008, hep-ph/9812489.
[8] D.E. Kaplan, G.D. Kribs, M. Schmaltz, Phys. Rev. D 62 (2000) 035010, hep-ph/9911293;
Z. Chacko, M.A. Luty, A.E. Nelson, E. Ponton, JHEP 0001 (2000) 003, hep-ph/9911323.
[9] T. Gherghetta, A. Pomarol, Nucl. Phys. B 602 (2001) 3, hep-ph/0012378.
[10] J.A. Bagger, F. Feruglio, F. Zwirner, hep-th/0107128;
J. Bagger, F. Feruglio, F. Zwirner, hep-th/0108010.
[11] M. Zucker, Nucl. Phys. B 570 (2000) 267, hep-th/9907082;
M. Zucker, JHEP 0008 (2000) 016, hep-th/9909144;
M. Zucker, Phys. Rev. D 64 (2001) 024024, hep-th/0009083.
[12] T. Kugo, K. Ohashi, Prog. Theor. Phys. 104 (2000) 835, hep-ph/0006231;
T. Kugo, K. Ohashi, Prog. Theor. Phys. 105 (2001) 323, hep-ph/0010288;
T. Fujita, T. Kugo, K. Ohashi, hep-th/0106051.
[13] J. Scherk, J.H. Schwarz, Phys. Lett. B 82 (1979) 60;
J. Scherk, J.H. Schwarz, Nucl. Phys. B 153 (1979) 61;
E. Cremmer, J. Scherk, J.H. Schwarz, Phys. Lett. B 84 (1979) 83.
[14] G.F. Giudice, M.A. Luty, H. Murayama, R. Rattazzi, JHEP 9812 (1998) 027, hep-ph/9810442.
[15] E. Cremmer, in: S.W. Hawking, M. Rocek (Eds.), Superspace and Supergravity, Cambridge
Univ. Press, 1981, pp. 267282;
A.H. Chamseddine, H. Nicolai, Phys. Lett. B 96 (1980) 89.
[16] M.F. Sohnius, P.C. West, Nucl. Phys. B 216 (1983) 100.
[17] A. Falkowski, Z. Lalak, S. Pokorski, hep-th/0102145.
[18] K.A. Meissner, H.P. Nilles, M. Olechowski, Nucl. Phys. B 561 (1999) 30, hep-th/9905139.
[19] K.R. Dienes, E. Dudas, T. Gherghetta, Phys. Rev. D 62 (2000) 105023, hep-ph/9912455.
[20] R. Barbieri, L. Girardello, A. Masiero, Phys. Lett. B 127 (1983) 429;
P. Binetruy, S. Dawson, I. Hinchliffe, Phys. Rev. D 35 (1987) 2215.
[21] S. Groot Nibbelink, hep-th/0108185.
[22] M.A. Luty, R. Sundrum, Phys. Rev. D 62 (2000) 035008, hep-th/9910202.
[23] N. Arkani-Hamed, Y. Grossman, M. Schmaltz, Phys. Rev. D 61 (2000) 115004, hepph/9909411.
[24] N. Arkani-Hamed, L.J. Hall, Y. Nomura, D.R. Smith, N. Weiner, Nucl. Phys. B 605 (2001) 81,
hep-ph/0102090.
[25] D.E. Kaplan, N. Weiner, hep-ph/0108001.
[26] N. Arkani-Hamed, T. Gregoire, J. Wacker, hep-th/0101233;

D. Marti, A. Pomarol, hep-th/0106256.
[27] E. Ponton, E. Poppitz, JHEP 0106 (2001) 019, hep-ph/0105021.
125

Unified supersymmetric model of naturally small

Dirac neutrino masses and the axionic solution
of the strong CP problem
Ernest Ma
Physics Department, University of California, Riverside, CA 92521, USA
Received 13 November 2001; accepted 7 December 2001
Abstract
Using the particle content of the fundamental 27 supermultiplet of E6 , naturally small Dirac
neutrino masses are obtained in the context of SU(3)C SU(2)L U (1)Y U (1) , where U (1)
comes from the decomposition E6 SO(10) U (1) , then SO(10) SU(5) U (1) . New
observable consequences are predicted at the TeV scale. An axionic solution of the strong CP problem
may be included at no extra cost. 2002 Elsevier Science B.V. All rights reserved.
PACS: 14.60.Pq; 12.60.Cn; 12.60.Jv; 14.80.Mz
With the present experimental evidence [13] on neutrino oscillations, the notion that
neutrinos should be massive is no longer in dispute. The next question is whether neutrino
masses are Majorana or Dirac. Experimentally, the nonobservation of neutrinoless double
beta decay at the 0.2 eV level [4] is unable to settle this issue, but there are very strong and
convincing theoretical reasons to believe that neutrino masses should be Majorana. On the
other hand, if the theoretical context is changed, naturally small Dirac neutrino masses are
possible, as shown below.
To obtain a Dirac mass, the left-handed neutrino L must be paired with a right-handed
singlet NR . Two problems arise immediately. (i) There is no symmetry to prevent NR from
acquiring a large Majorana mass. (ii) Even if such a symmetry (such as additive lepton
number) is imposed, an extremely small Yukawa coupling (less than 1011 ) is still needed
to satisfy the experimental bound m < few eV. The usual resolution of these problems
is to take advantage of (i) to make mN very large, so that the famous canonical seesaw
E-mail address: ma@phyun8.ucr.edu (E. Ma).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 2 5 - 3
E. Ma / Nuclear Physics B 623 (2002) 126132
127
mechanism [5] makes m = m2D /mN . Now (ii) is also not a problem because the Yukawa
coupling for the Dirac mass mD is no longer required to be very small.
In this paper a new scenario is proposed where (i) NR is naturally prevented from having
a Majorana mass and (ii) mD is small without having a small Yukawa coupling [6,7]. This
is possible because the theoretical framework used will be that of superstring-inspired E6
[8]. As a bonus, the axionic solution [9] of the strong CP problem may also be included.
The starting point is the gauge group E6 and its decomposition E6 SO(10) U (1) ,
then SO(10) SU(5) U (1) . It is often assumed that at TeV energies, a linear
combination of U (1) and U (1) remains [10] in addition to the standard SU(3)C
SU(2)L U (1)Y . It is usually also assumed that three complete matter supermultiplets
of the fundamental 27 representation of E6 are present at these energies, which include
the known three families of quarks and leptons as well as other new particles. Under the
subgroup SU(5) U (1) U (1) , the particle content of each supermultiplet is given by

27 = (10; 1, 1) (u, d), uc , ec + (5 ; 1, 3) d c , (e , e) + (1; 1, 5) N c

+ (5; 2, 2) h, E c , NEc + (5 ; 2, 2) hc , (E , E) + (1; 4, 0)[S],
(1)
where the U (1) charges refer to 2 6 Q and 2 10 Q . Note that the known quarks and
leptons are contained in (10; 1, 1) and (5 ; 1, 3), and the two Higgs scalar doublets
are represented by (E , E) and (E c , NEc ). Since N c and S are singlets under SU(5),
one linear combination will be trivial under the assumed low-energy gauge group, i.e.,
SU(3)C
SU(2)L U (1)Y U (1) with Q = Q cos + Q sin . For the choice
tan = 1/15, the U (1)N model [11,12] is obtained, for which N c is trivial, thus
allowing it to acquire a large Majorana mass. Combining this with the existing term
(e NEc eE c )N c , the usual seesaw Majorana neutrino mass may then be obtained.
Consider now the case sin = 1, i.e., the U (1) model. This allows S to have a large
Majorana mass, but not N c . Hence the only apparent way that e may become massive
is to pair up with N c to form a Dirac neutrino with mass proportional to the vacuum
expectation value (VEV) of the scalar component of NEc . If the latter is of the order of
the electroweak symmetry breaking scale, i.e., 102 GeV, then an extremely small Yukawa
coupling is required. This is in fact the prevailing working ansatz of all U (1) models
c has m2 > 0
except U (1)N . However, there is a very simple and natural solution. If N
E
with m large, then its VEV can be very small [6,7]. This is precisely the case in the U (1)
model, where E NEc EE c is an allowed term.
There are 11 generic terms [13] in the superpotential of such E6 models. They are:

= u N
u c E
c u c ,

c dE
(1) Q
(2)
E
c

c
d E
= u E
d E d ,
(2) Q
(3)

= e E
e E e c ,
(3) L ec E
(4)

= N
c
E
E
c

(4)
SE
(5)
S,
E E E
(5)
(6)

S h h c ,
c
= N
c
c N
,
c E

L N
e E e E
(6)
(7)
128
(7)
L h c = (u
Q
e d e )h c ,
c c
(8)
(8)
u e h,
(9)
(9)
c h,
dc N
(10)
Q
h = (u d du)
(10) Q
h,
c c c
(11) u d h .
(11)
(12)
To prevent rapid proton decay, some terms must be absent. This is usually accomplished
by the imposition of an exactly conserved discrete symmetry, such as the well-known R
parity. Here the choice is
Z3 U (1)P Q .
(13)
transform as ; u c , dc , ec as 2 ; and all other superfields

1,2 , E

Under Z3 with 3 = 1, E
1,2
h c ,
as 1. Under U (1)P Q , the only superfields with nonzero charges are h,
S1 ,
S2 ,
S3 with
charges 1/2, 1/2, 1, 2, 2, respectively. This means that the terms (7) to (11) are all
, the term (5) involves only

forbidden, the term (6) involves only E
S1 , and the term (4) is
3
forbidden, but since
S is trivial under U (1) , the soft term
= N
c

E
E
c
E
e E E
(14)
, is invariant under Z . All other

3 E

by itself is allowed. Note that only one term, i.e., E
3
3

terms will break Z3 but only softly.
EE
The superpotential of this model is then given by
(d)
(e)
+ f (u) Q

= ij E
i E
c E
c
c

W
j
j 1,2 + f(1,2)ij Qi dj E1,2 + f(1,2)ij Li ej E1,2
(1,2)ij i u
+ f (h)
jc E
c

+ fij(N) L i N
3
ij S1 hi hj + m2 S2 S3 + f2 S2 S1 S1 .
(15)
The anomalous global U (1)P Q is spontaneously broken at the intermediate scale m2 so

that an invisible axion will emerge to solve the strong CP problem. The U (1)P Q charges
of
S1,2,3 are chosen so that S1 may acquire a large VEV ( m2 109 to 1012 GeV) without breaking the supersymmetry of the entire theory at that scale. Details are contained in
Ref. [14]. Because the usual quarks and leptons here do not transform under U (1)P Q , the
axion of this model is of the KSVZ type [15], whereas that of Ref. [14] is of the DFSZ
type [16].
Note that U (1)P Q here serves the dual purpose of solving the problem of rapid proton
decay as well. Note also that the choice of U (1) as the extra gauge symmetry is the only
and the

E
one which allows that to work. It also serves the purpose of allowing the term E
to couple to N
. The Dirac mass

c , with a large mass for E
3 E

choice of Z3 allows only E
3
3
, which may

linking e to N c is proportional to the VEV of the scalar component of E
3
then be very small [6,7], as shown below.
Consider the following Higgs potential of 4 scalar doublets H1,2,3,4 representing the
,E
, respectively, [17] (assuming that E
have no
1 , E
2 and E

scalar components of E
1 3 , E
3
2
VEV):
V=

i
129

m2i Hi Hi + m213 H1 H3 + m224 H2 H4 + m212 H1 H2

+ m214 H1 H4 + m232 H3 H2 + m234 H3 H4 + h.c.

2
1 g12 g2
+
H1 H1 + H2 H2 H3 H3 + H4 H4
+
2 4
10
2

1 2
+ g2
Hi Hi
(16)
where ( = 1, 2, 3) are the usual SU(2) representation matrices. Let the VEVs of Hi
be vi , then the minimum of V is

m2i vi2 + 2m212 v1 v2 + 2m213 v1 v3 + 2m214 v1 v4 + 2m224 v2 v4
Vmin =
i
+ 2m232 v2 v3 + 2m234 v3 v4 +

2g2 2
2
1 2
g1 + g22 +
v1 v22 + v32 v42 , (17)
8
5
where all parameters have been assumed real for simplicity. The 4 equations of constraint
are
0 = m21 v1 + m212 v2 + m213 v3 + m214 v4

2g2

1 2
2
g1 + g2 +
v1 v12 v22 + v32 v42 ,
+
4
5
0 = m22 v2 + m212 v1 + m224 v4 + m232 v3

2g2

1 2
2
g + g2 +
v2 v12 v22 + v32 v42 ,
4 1
5
0 = m23 v3 + m213 v1 + m232 v2 + m234 v4

2g2

1 2
+
g1 + g22 +
v3 v12 v22 + v32 v42 ,
4
5
0 = m24 v4 + m224 v2 + m214 v1 + m234 v3

2g2

1 2
2
g1 + g2 +
v4 v12 v22 + v32 v42 .
4
5
(18)
(19)
(20)
(21)
Since m23 m24 233 , m213 13 33 , and m224 31 33 are the only parameters which
have contributions involving the large mass 33 , it is clear that Eqs. (20) and (21) have the
solution
v3
m213v1
m23
v4
m224 v2
m24
(22)
They may then be of order 0.1 eV if m3,4 33 1015 GeV (i.e., close to a possible grandunification mass scale), and 13 , 31 MSUSY 1 TeV. Setting v3 = v4 = 0 in Eqs. (18)
130
and (19), the usual conditions of the minimal supersymmetric standard model are obtained
except for the additional terms due to g .
Now U (1) also undergoes spontaneous symmetry breaking through the VEV of one
c )s. As a result, there appear a new massive neutral gauge
linear combination of the 3 (N

c , and the Dirac fermion which comes
boson Z , the corresponding scalar boson 2 Re N

c
c [18]. Hence only 2
from the pairing of z and N , all having the mass ( 5/2)g N
c
(N )s remain and they combine with 2 of the 3 s to form 2 light Dirac neutrinos. The
remaining gets a negligible Majorana mass from the allowed supersymmetry-breaking
soft Majorana mass of z . A satisfactory framework is thus established for describing the
oscillations of 2 light Dirac neutrinos and 1 essentially massless Majorana neutrino.
At the TeV energy scale, this model is verifiable experimentally by its many unique
predictions. First, there must be a Z gauge boson with couplings to quarks and leptons
according to Eq. (1). In particular, it will have invisible decays to neutrinos given by
c N c ) 77
+(Z + N
= .
+(Z l + l )
30
(23)
There are likely to be 4 Higgs doublets, instead of 2, and definitely not 6. There should not
be exotic quarks (i.e., h and hc ) because they are predicted to be very heavy with masses
at the axion scale. The axion itself is of course very light and very difficult to detect [19].
Its partners, the saxion and the axino, are likely to be at or below the TeV scale and may
also be components of the dark matter of the Universe. Lepton number is violated through
, this violation is highly
c , but since N
c only appears in Eq. (15) with the very heavy E

N
3
suppressed. Thus my proposed model evades the general conclusion of Ref. [12] regarding
E6 subgroups that only U (1)N [11] and the skew leftright model [20] do not have leptonnumber violating interactions at the TeV scale which would erase any preexisting lepton
or baryon asymmetry of the Universe.
In conclusion, a new unified supersymmetric model has been proposed which has the
following desirable properties.
(1) Its particle content comes from 3 complete fundamental 27 representations of E6 ,
which may be the remnant of an underlying superstring theory.
(2) Its low-energy gauge group is SU(3)C SU(2)L U (1)Y U (1) , where U (1)
comes from E6 SO(10) SU(5) U (1) .
(3) It has the additional symmetry Z3 U (1)P Q which serves many purposes, including
that of preventing rapid proton decay. Z3 is softly broken; U (1)P Q is spontaneously
broken.
term of
c E

(4) Naturally small Dirac neutrino masses [21,22] come from the L N
3

3 has a very small VEV, using the mechanism [6,7] of a large positive
Eq. (15) because E
, as shown by Eq. (22).

m2 close to a possible grand-unification mass scale for E
3

(5) The 3 singlet superfields S1,2,3 , which do not transform under U (1) , are
chosen [14] to obtain an axionic solution of the strong CP problem, such that fa MSUSY .
(6) This model predicts a definite supersymmetric particle structure associated with the
extra U (1) gauge symmetry at the TeV scale, which should be accessible in near-future
high-energy accelerators.
131
(7) It is the only model to date which incorporates naturally small Dirac neutrino
masses with the axionic solution of the strong CP problem in a comprehensive theoretical
framework of all particle interactions.
Acknowledgement
This work was supported in part by the US Department of Energy under Grant No.
DE-FG03-94ER40837.
References
[1] S. Fukuda et al., Super-Kamiokande Collaboration, Phys. Rev. Lett. 85 (2000) 3999, and references therein.
[2] S. Fukuda et al., Super-Kamiokande Collaboration, Phys. Rev. Lett. 86 (2001) 5656, and references therein;
See also: Q.R. Ahmad et al., SNO Collaboration, Phys. Rev. Lett. 87 (2001) 071301.
[3] A. Aguilar et al., hep-ex/0104049;
See also: G.B. Mills, for the LSND Collaboration, Nucl. Phys. Proc. Suppl. 91 (2001) 198, and references
therein.
[4] L. Baudis et al., Phys. Rev. Lett. 83 (1999) 41.
[5] M. Gell-Mann, P. Ramond, R. Slansky, in: P. van Nieuwenhuizen, D.Z. Freedman (Eds.), Supergravity,
North-Holland, Amsterdam, 1979, p. 315;
T. Yanagida, in: O. Sawada, A. Sugamoto (Eds.), Proceedings of the Workshop on the Unified Theory and
the Baryon Number in the Universe, KEK, Tsukuba, Japan, 1979, p. 95;
R.N. Mohapatra, G. Senjanovic, Phys. Rev. Lett. 44 (1980) 912.
[6] E. Ma, Phys. Rev. Lett. 86 (2001) 2502.
[7] E. Ma, Phys. Lett. B 516 (2001) 165.
[8] For a review see, for example, J.L. Hewett, T.G. Rizzo, Phys. Rep. 183 (1989) 193.
[9] R.D. Peccei, H.R. Quinn, Phys. Rev. Lett. 38 (1977) 1440;
S. Weinberg, Phys. Rev. Lett. 40 (1978) 223;
F. Wilczek, Phys. Rev. Lett. 40 (1978) 279.
[10] See, for example, G.C. Cho, K. Hagiwara, Y. Umeda, Nucl. Phys. B 531 (1998) 65;
G.C. Cho, K. Hagiwara, Y. Umeda, Nucl. Phys. B 555 (1999) 651, Erratum.
[11] E. Ma, Phys. Lett. B 380 (1996) 286.
[12] T. Hambye, E. Ma, M. Raidal, U. Sarkar, Phys. Lett. B 512 (2001) 373.
[13] E. Ma, Phys. Rev. Lett. 60 (1988) 1363.
[14] E. Ma, Phys. Lett. B 514 (2001) 330.
[15] J.E. Kim, Phys. Rev. Lett. 43 (1979) 103;
M.A. Shifman, A.I. Vainshtein, V.I. Zakharov, Nucl. Phys. B 166 (1980) 493.
[16] M. Dine, W. Fischler, M. Srednicki, Phys. Lett. B 104 (1981) 199;
A.R. Zhitnitsky, Sov. J. Nucl. Phys. 31 (1980) 260.
[17] E. Ma, Phys. Rev. D 64 (2001) 097302.
[18] E. Keith, E. Ma, Phys. Rev. D 54 (1996) 3587.
[19] L.J. Rosenberg, K.A. van Bibber, Phys. Rep. 325 (2000) 1.
[20] E. Ma, Phys. Rev. D 36 (1987) 274;
K.S. Babu, X.-G. He, E. Ma, Phys. Rev. D 36 (1987) 878.
[21] For other recent discussions see, for example:
P.P. Divakaran, G. Rajasekaran, Mod. Phys. Lett. A 14 (1999) 913;
P.Q. Hung, Phys. Rev. D 59 (1999) 113008;
P.Q. Hung, Phys. Rev. D 62 (2000) 053015.
132
[22] Another approach is to use extra dimensions:

N. Arkani-Hamed, S. Dimopoulos, G. Dvali, J. March-Russell, hep-ph/9811448;
K.R. Dienes, E. Dudas, T. Gherghetta, Nucl. Phys. B 557 (1999) 25;
G. Barenboim, G.C. Branco, A. de Gouvea, M.N. Rebelo, Phys. Rev. D 64 (2001) 073005.

Ricci-flat Khler manifolds from supersymmetric

gauge theories
Kiyoshi Higashijima, Tetsuji Kimura, Muneto Nitta 1
Department of Physics, Graduate School of Science, Osaka University, Toyonaka, Osaka 560-0043, Japan
Received 27 August 2001; accepted 19 November 2001
Abstract
Using techniques of supersymmetric gauge theories, we present the Ricci-flat metrics on noncompact Khler manifolds whose conical singularity is repaired by the Hermitian symmetric space.
These manifolds can be identified as the complex line bundles over the Hermitian symmetric spaces.
Each of the metrics contains a resolution parameter which controls the size of these base manifolds,
and the conical singularity appears when the parameter vanishes. 2002 Elsevier Science B.V. All
rights reserved.
PACS: 11.30.Pb; 11.15.-q; 04.65.+e
1. Introduction
N = 2 supersymmetric nonlinear sigma models in two dimensions [1] on Ricci-flat
Khler manifolds can be considered as the model of the superstring theory on curved
space [24]. Ricci-flat Khler manifolds are also important ingredient for D-branes in
curved space. In the previous letter [5], we presented the simple derivation of the O(N)
symmetric Ricci-flat metric, which actually coincides with the Stenzel metric on the
cotangent bundle over S N1 [6]. The conical singularity is resolved by S N1 with a
radius being the deformation parameter. It reduces to the EguchiHanson gravitational
instanton [7] and the six-dimensional deformed conifold [8] in the cases of N = 3 and
N = 4, respectively. In [9], a new metric for the six-dimensional conifold, in which the
conical singularity is repaired by S 2 S 2 , was found. It was generalized in our previous
letter [10] to the higher-dimensional conifold, in which the singularity is resolved by the
complex quadric surface QN2 = SO(N)/[SO(N 2) U (1)] [10]. The new manifold
E-mail addresses: higashij@phys.sci.osaka-u.ac.jp (K. Higashijima), t-kimura@het.phys.sci.osaka-u.ac.jp
(T. Kimura), nitta@het.phys.sci.osaka-u.ac.jp (M. Nitta).
1 Address after September 1: Department of Physics, Purdue University, West Lafayette, IN 47907-1396, USA.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 5 9 1 - 0
134
K. Higashijima et al. / Nuclear Physics B 623 (2002) 133149
can be regarded as the complex line bundle over QN2 , which is a Hermitian symmetric
space.
In this paper we present the new Ricci-flat metrics replacing the base manifold by other
Hermitian symmetric spaces, the Grassmann manifold GN,M = SU(N)/[SU(N M)
U (M)], SO(2N)/U (N) and Sp(N)/U (N). To do this, we apply the technique of the gauge
theory formulation of supersymmetric nonlinear sigma models on the Hermitian symmetric
spaces [1113], which was used for the study of non-perturbative effects [14]. We note that
our manifolds are natural generalizations of the Calabi metric on the complex line bundle
over CP N1 [15].
This paper is organized as follows. In Section 2, we recapitulate the construction of
compact Khler manifolds GN,M , SO(2N)/U (N) and Sp(N)/U (N) by supersymmetric
gauge theories, and extend this to non-compact Khler manifolds. In Section 3, we impose
the Ricci-flat condition on these non-compact manifolds. Symmetry plays a crucial role to
reduce partial differential equations to ordinary differential equations of one variable. In
Section 4, we present explicit expressions of Khler metrics and their Khler potentials.
It is found that these manifolds contain resolution parameter b as an integration constant,
and the conical singularity is resolved by GN,M , SO(2N)/U (N) or Sp(N)/U (N) of a
radius expressed in terms of b. These manifolds are complex line bundles over GN,M ,
SO(2N)/U (N) and Sp(N)/U (N). Section 5 is devoted to conclusion and discussions.
In Appendix A, we summarize the isomorphisms between the lower dimensional base
manifolds and the duality between the Grassmann manifolds, and show that they hold for
total spaces.
2. Construction by supersymmetric gauge theories

2.1. Compact Khler manifolds from gauge theories
In this section we recapitulate the construction of GN,M , SO(2N)/U (N) and Sp(N)
/U (N), using supersymmetric gauge theories [11]. Such a method was first found for
the projective space CP N1 [16] and the Grassmann manifold GN,M [17], and then
recognized as the symplectic or the Khler quotient [18].
2.1.1. Construction of GN,M [17,18]
Let (x, , ) be an N M matrix-valued chiral superfield. The group SU(N) U (M)
can act on it as

= gL gR 1 , (gL , gR ) SU(N), U (M) .
(2.1)
We promote the right action of U (M) to a gauge symmetry by introducing a vector
superfield V (x, , ), taking a value in the Lie algebra of U (M). The gauge transformation
is given by
= ei ,
eV eV = ei eV ei ,
(2.2)
135
where (x, , ) is a parameter chiral superfield, taking a value in the Lie algebra of
U (M). Note that the local invariance group is enlarged to the complexification of the gauge
group, U (M)C = GL(N, C), since the scalar component of (x, , ) is complex. The
Lagrangian invariant under the global SU(N) and the local U (M) symmetries is given by

L = d 4 K , , V = d 4 tr eV c tr V ,
(2.3)
where K is the Khler potential. Here c is a real positive constant, called the Fayet
Iliopoulous (FI) parameter, and c tr V is called the FI D-term.
Since V is an auxiliary field, it can be eliminated by its equation of motion 2
L
= eV c1M = 0,
V
(2.4)
where 1M is an M M unit matrix. Substituting the solution, V (, ) = log ( /c),

back into the Lagrangian (2.3), we obtain

K , , V , = c tr log = c log det

= c log det 1M + .
(2.5)
Since the gauge group is complexified, we have chosen the gauge fixing as

1M
=
,
(2.6)
where (x, , ) is an (N M) M matrix-valued chiral superfield. The constantterms

in (2.5) have been omitted, since they disappear under the superspace integral d 4 .
(2.5) is the Khler potential of GN,M = SU(N)/[SU(N M) U (M)], whose complex
dimension is M(N M). It becomes one of CP N1 = SU(N)/[SU(N 1) U (1)] if we
set M = 1, in which case the gauge group is U (1).
2.1.2. Construction of SO(2N)/U (N) and Sp(N)/U (N) [11]
Let us replace the size of the matrix , (N, M), by (2N, N), corresponding to G2N,N ,
and introduce the invariant tensor of SO(2N) or Sp(N):

0
1N
,
J=
(2.7)
1N 0
where = 1 for SO(2N), and = 1 for Sp(N). The invariant Lagrangian is given by

d 2 tr 0 T J + c.c. ,
L = d 4 tr eV c tr V +
(2.8)
where 0 (x, , ) is an auxiliary chiral superfield of an N N matrix, belonging to
(anti-)symmetric tensor representation of the gauge group U (N) for SO(2N)/U (N)
[Sp(N)/U (N)] with the suitable U (1) charge.
2 We regard eV eV as an infinitesimal parameter: L = tr[ eV (eV eV )] c tr( log eV ) =
tr[( eV c1M )X1 X], where X = eV .
136
By the integration over V , we obtain (2.5) with the same gauge fixing as (2.6). The
integration over 0 gives the constraint
T J = + T = 0,
(2.9)
which implies that the N N matrix chiral superfield is anti-symmetric or symmetric

for SO(2N)/U (N) or Sp(N)/U (N), respectively. The Khler potential (2.5) with the
constraints (2.9) is one of SO(2N)/U (N) or Sp(N)/U (N), whose complex dimension
is 12 N(N 1) or 12 N(N + 1), respectively.
Instead of the Khler potential of the Lagrangian (2.3) and (2.8), we can start from

K , , V = f tr eV c tr V ,
(2.10)
where f is an arbitrary function. We can show that we obtain the same results even if we
start from (2.10) [12]. Let us make some comments. We have used the classical equation of
motion of V to eliminate it. We can promote this to the quantum level in the path integral
formalism [12]. If we add the kinetic term for V rather than regarding V as auxiliary, our
manifolds are obtained as the classical moduli space of the gauge theories [19].
2.2. Non-compact Khler manifolds from gauge theories
Let us construct the non-compact Khler manifolds, by restricting the gauge degrees of
freedom from U (M) to SU(M). To do this, we promote the FI-parameter c in (2.10) to an
auxiliary vector superfield C(x, , ):

K0 , , V , C = f tr eV CtrV ,
(2.11)
where f is an arbitrary function. 3 Note that V (x, , ) in this Lagrangian is still taking a
value in the Lie algebra of U (M). The equations of motion of V and C read

L
= f tr eV eV C1M = 0,
(2.12a)
V
L
= tr V = 0,
(2.12b)
C
respectively, where the prime denotes the differentiation with respect to the argument of f .
The gauge group is restricted to SU(M) by (2.12b). The trace and the determinant of
(2.12a) are

f tr eV tr eV = MC,
(2.13a)
V M
M
f tr e
(2.13b)
det = C ,
since det eV = 1 for the SU(M) gauge field V . Eliminating C from these equations, the
solution of V reads
1

tr eV = M det M .
(2.14)
3 There exist independent invariants tr[( eV )2 ], . . . , tr[( eV )M ], besides tr( eV ). We can show
that, even if these are included as the arguments of the arbitrary function of (2.11), we obtain the same result
(2.15). The situation is the same for the cases of the U (M) gauge field, (2.10), for compact manifolds.
137
Substituting this back into (2.11) and taking account of (2.12b), we obtain the nonlinear
Khler potential

1
K0 , , V , = f M det M K X , ,
(2.15)
where X(, ) is a vector superfield, invariant under the global U (N) [SO(2N) or
Sp(N)] and the local SU(M) [SU(N)] symmetries, defined by

X , = log det ,
(2.16)
and K(X) is a real function of X related to f . Here the logarithm in the definition
of X is just a convention. (Note that this definition of the invariant is different from the
one in [5,10].) From the view point of the algebraic variety, X is the gauge invariant
parameterizing the moduli space of supersymmetric gauge theories [19].
Since the gauge group is complexified to SU(M)C = SL(M, C), we can choose a gauge
fixing as

1M
,
=
(2.17)
where is an (N M) M matrix-valued chiral superfield, and (x, , ) is a chiral

superfield. Comparing (2.17) with (2.6), we find that the superfield is parameterizing
a fiber, while is parameterizing a base manifold, with the total space being a complex
line bundle. Under this gauge fixing, the invariant superfield X is decomposed as
X = M log | |2 + log det(1M + ) = M log | |2 + ,
(2.18)

log det(1M + ).
(2.19)
Note that is a Khler potential of GN,M [SO(2N)/U (N) or Sp(N)/U (N)] obtained in
(2.5) [with the constraint (2.9)].
Let us introduce some notations. We denote the elements of the matrix-valued chiral
superfield by Aa , where the upper case and the lower case indices, A and a, run from
1 to N M and from 1 to M, respectively. Since the size of the matrix is N N in the
cases of Sp(N)/U (N) and SO(2N)/U (N), we denote its elements by ab . In this case,
only the components ab with b a (b > a) are considered as independent. When we
discuss the total space, we use the coordinates z (, Aa ). It should be noted that from
now on we use the same letters for chiral superfields and their complex scalar components.
We make a comment on the symmetry breaking. These non-compact manifolds can be
regarded as
R
SU(N)
G
=R
,
H
SU(N M) SU(M)
SO(2N)
,
SU(N)
Sp(N)
,
SU(N)
(2.20)
at least locally. The part of G/H is parametrized by the NambuGoldstone bosons arising
from the spontaneous breaking of the global symmetry G down to H , whereas the factor of
R is parametrized by the so-called quasi-NambuGoldstone boson (see, e.g., [11,20,21]).
138
3. Ricci-flat conditions
We would like to determine the function K(X) in (2.15), by imposing the Ricci-flat
condition on the manifold. The metric of the Khler manifold is given by g = K,
where = /z and = /z . The explicit expression of the Khler metric can be
calculated as

g
g (Bb)
g =
(3.1a)
,
g(Aa) g(Aa)(Bb)
with each block being
X X
,

X X
= K
,
Bb
g = K
g (Bb)
X X
2X
+ K
,
Aa Bb
Aa Bb
X X
g(Aa) = K
,
Aa
g(Aa)(Bb) = K
(3.1b)
(3.1c)
where the prime denotes the differentiation with respect to the argument X of the function
K(X). Here we have used equations,
2X
2X
2X
=
=
= 0 ( = 0),
Bb
Aa
which follow from (2.18).
The determinant of the metric is calculated, to yield

2X
M 2

K
det
det g =
K
,
| |2
(Aa)(Bb)
Aa Bb
(3.2)
where det(Aa)(Bb) denotes the determinant of the matrix of the tensor product, spanned
by (Aa) and (Bb) . Since the Ricci-form is given by Ric = log det g , the
Ricci-flat condition Ric = 0 implies
det g = constant |F |2 ,
(3.3)
with F being a holomorphic function.

3.1. Line bundle over GN,M
In this section let us obtain the explicit solution of the Ricci-flat metric on the line bundle
over the Grassmann manifold. Let us calculate the X differentiated by matrix fields Aa
once and twice, needed for the calculation of the Khler metric (3.1). By noting
Bb
= AB ab ,
Aa
we obtain

X
= (Aa) = (1M + )1 aA ,
Aa

X
= (Aa) = (1M + )1 Aa ,
Aa
(3.4)
(3.5a)
(3.5b)
139
2X
= (Aa) (Bb)
Aa Bb

1
= (1M + )1
ab 1(NM) (1M + ) BA .
(3.5c)
Here we have used the definition of in (2.19), and (Aa) and (Aa) represent
, respectively. Note that (3.5c) is just the Khler
differentiations with respect to Aa and Aa
metric of the Grassmann manifold GN,M . The determinant of g can be calculated as
det g =

M 2 M(NM)
K (K )
det (Aa)(Bb) .
2
| |
(Aa)(Bb)
(3.6)
To obtain the concrete expression of this determinant, we use a symmetry transformation

preserving the value of the determinant. Under the transformation of the complex isotropy
[SU(N M) SU(M)]C = SL(N M, C) SL(M, C), the coordinates transform linearly
as z z = V z . Since the transformation matrix V belongs to a subgroup of
SL(M(N M), C), the equation det V = 1 holds and the det g is invariant: det g

2
det g
= det g | det V | = det g . For an arbitrary matrix-valued chiral superfield ,
there exists a complex isotropy which permits the transformation of to the form of

0 . . .
,
= .
(3.7)
..
0
where the dots denote zero elements, and the only non-zero element is 11 0 . The
matrix of the tensor product (3.5c) is diagonalized as
(NM1)-blocks

2
2X
,
=
diag
,
,
.
.
.
,
;
;
.
.
.
;
,
1,
.
.
.
,
1
,
1,
.
.
.
,
1

Aa Bb
M1
M1
(3.8)
M1
|2 )1 .
where (1 + |0
Each block separated by the semicolons is labeled by the indices
A = B, which run from 1 to N M, and in each block the indices a = b run from 1 to M.
With noting = (1 + |0 |2 )1 = [det(1N + )]1 = | |2M eX , the determinant (3.6)
can be calculated as
det g = M 2 | |2(MN1) eNX K (K )M(NM) .
(3.9)
The Ricci-flat condition (3.3) becomes

d
(K )M(NM)+1 = a,
dX
where a is a real constant.
eNX
(3.10)
3.2. Line bundles over SO(2N)/U (N) and Sp(N)/U (N)

In this section we construct the Ricci-flat metrics on the line bundles over SO(2N)/U (N)
and Sp(N)/U (N). These cases are obtained by imposing the constraint (2.9) on the Grassmann manifold G2N,N . Under the condition (2.9), the differentiations with respect to the
140
matrix elements ab , corresponding to (3.4) for the Grassmann case, become

1
cd
= (ca db cb da ) 1 ab ,
ab
2
(3.11)
where we do not take a sum over the index a or b. Using this, the X differentiated by one
or two s can be calculated, to yield

N

1
X
(1N + )1 dc (ca db cb da ) 1 ab ,
= (ab) =
ab
2
c,d=1
(3.12a)

N

1
X
(1N + )1 cd (ca db cb da ) 1 ab ,

= (ab) =
ab
2
c,d=1
(3.12b)

1
1
2X
1 cd
= (ab) (cd) = 1 2 ab
ab cd
2

1
(1N + )bd 1N (1N + )1 ca

1
(1N + )1
bc 1N (1N + ) da

+ (a b, c d) .
(3.12c)
Here the last term in the last line implies adding the preceding two terms with the exchange
of the indices. Note again that (3.12c) is just the Khler metric of SO(2N)/U (N) or
Sp(N)/U (N). The determinant (3.2) can be calculated as
det g =

N 2 1 N(N)
K (K ) 2
det (ab) (cd) .
2
(ab)(cd)
| |
(3.13)
We again use the complex isotropy transformation of SU(N)C = SL(N, C), preserving the
determinant. We first discuss SO(2N)/U (N) followed by Sp(N)/U (N).
3.2.1. The line bundle over SO(2N)/U (N)
Using the complex isotropy transformation of SL(N, C), the arbitrary can be put
0
0 0
= 0 0 0 ,
(3.14)
0
0 0
where non-zero elements are 12 = 21 0 . The matrix of the tensor product (3.12c) is
diagonalized as

2 X

ab cd
b>a, d>c
2

= diag 2 , 2, . . . , 2 ; 2, . . . , 2 ; 2, . . . , 2; 2, . . . , 2; . . . ; 2, 2 ; 2 ,
(3.15)

N2
|2 )1 .
N2
N3
N4
Each block separated by the semicolons is labeled by the indices

where (1 + |0
a = c, which run from 1 to N , and the indices b = d run from a + 1 = c + 1 to N in
141
the ath block, by the conditions b > a and d > c. Noting = (1 + |0 |2 )1 = [det(1N +
)]1/2 = | |N eX/2 , we can calculate the determinant (3.13), given by
det g = N 2 2 2 N(N1) | |2N(N1)2 e(N1)X K (K ) 2 N(N1) .
1
(3.16)

e(N1)X
1
d
(K ) 2 N(N1)+1 = a.
dX
(3.17)
3.2.2. The line bundle over Sp(N)/U (N)

There exists an isotropy transformation which transforms an arbitrary matrix to the
form of (3.7). The matrix of the tensor product (3.12c) is diagonalized as

2 X

ab cd
ba, dc
2

= diag , 2, . . . , 2 ; 1, 2, . . . , 2; 1, 2, . . . , 2; . . . ; 1, 2; 1 ,

N1
N2
(3.18)
N3
where (1 + |0 |2 )1 . Each block separated by the semicolons is labeled by the indices

a = c, which run from 1 to N , and the indices b = d run from a = c to N in the ath block
by the conditions b a and d c. Noting = (1 + |0 |2 )1 = [det(1N + )]1 =
| |2N eX , we can calculate the determinant (3.13), given by
1
det g = N 2 2 2 N(N1) | |2N(N+1)2 e(N+1)X K (K ) 2 N(N+1) .
(3.19)

e(N+1)X
1
d
(K ) 2 N(N+1)+1 = a.
dX
(3.20)
4. Ricci-flat metrics and Khler potentials

4.1. Khler potentials
We can immediately solve (3.10), (3.17) and (3.20):

1
g M(N M) + 1,
eNX + b g ,
dK
1
(N1)X
=
+ b f , f 12 N(N 1) + 1,
e
dX
1
e(N+1)X + b h , h 12 N(N + 1) + 1,
for GN,M ,
for SO(2N)/U (N),
(4.1)
for Sp(N)/U (N),
where is a constant related to a, N and M, and b is an integration constant interpreted

as a resolution parameter of the conical singularity. Although these are sufficient to obtain
142
the Khler metrics using (3.1), we can calculate Khler potentials themselves:

1
1
1
1
g
eNX + b g + b g I b g eNX + b g ; g , for GN,M ,
1
(N1)X
1
1
1
f
+ b f + b f I b f e(N1)X + b f ; f ,
N1 e
K(X) =
for SO(2N)/U (N),
(N+1)X
1
1
1
1
e
+ b h + b h I b h e(N+1)X + b h ; h ,
N+1
for Sp(N)/U (N).
(4.2)
Here the function I (y; n) is defined by

y
I (y; n)
1
1 + (1)n
dt
=
log(y 1)
log(y + 1)
tn 1 n
2
[ n1

2 ]
2r
2r
1
2
cos
+
log y 2y cos
+1
n
n
n
r=1
cos(2r/n) y
2
2r
arctan
+
sin
.
n
n
sin(2r/n)
[ n1
2 ]
(4.3)
r=1
In the limit of b 0, these manifolds become (generalized) conifolds with their Khler
potentials,
1
N
g
gN | |2M det(1M + ) g ,
for GN,M ,
1
N1

f
K = f | |2N det(1 + ) f , T = , for SO(2N)/U (N), (4.4)
N
N1
1
h h | |2N det(1 + ) N+1

h ,
T = ,
for Sp(N)/U (N).
N
N+1
4.2. Ricci-flat Khler metrics
We can calculate the Ricci-flat Khler metrics substituting the solutions (4.1) into (3.1).
The component g is
1 1
2
Mg N eNX + b g eN | |2MN2 , for GN,M ,
1
N 2 (N1) e(N1)X + b f 1 e(N1) | |2N(N1)2 ,

f
g =
(4.5)
for SO(2N)/U (N),
1

2
1
N (N+1)
e(N+1)X + b h e(N+1) | |2N(N+1)2 ,
for Sp(N)/U (N),

where was defined in (2.19). These are singular at the surface = 0: g | =0 = 0. This
singularity is just a coordinate singularity of z = (, Aa ). To find a regular coordinate,
we perform coordinate transformations,
MN
/MN,
for GN,M ,
N(N1) /N(N 1), for SO(2N)/U (N),

(4.6)
N(N+1)
/N(N + 1), for Sp(N)/U (N),
143
with Aa (or ab ) being unchanged. The metrics in the regular coordinates z = (, Aa )

can be calculated, to give
1 1
M 2 N NX
e
+ b g eN ,
g
1 1
M 2 N 2 NX
g(Bb) =
e
+ b g eN (Bb) ,
g
1 1
M 2 N 3 NX
g(Aa)(Bb) =
+ b g eN ||2 (Aa) (Bb)
e
g
NX
1
+ e
+ b g (Aa) (Bb) ,
g =
(4.7a)
(4.7b)
(4.7c)
for GN,M ,
1 1
N 2 (N 1) (N1)X
e
+ b f e(N1) ,
f
2
1 1
N (N 1)2 (N1)X
g(cd) =
e
+ b f e(N1) (cd) ,
f
2
1 1
N (N 1)3 (N1)X
g(ab)(cd) =
e
+ b f e(N1) ||2 (ab) (cd)
f
(N1)X
1
+ e
+ b f (ab)(cd) ,
g =
(4.8a)
(4.8b)
(4.8c)
for SO(2N)/U (N), and

1 1
N 2 (N + 1) (N+1)X
e
+ b h e(N+1) ,
h
2
1 1
N (N + 1)2 (N+1)X
g(cd) =
+ b h e(N+1) (cd) ,
e
h
1 1
N 2 (N + 1)3 (N+1)X
+ b h e(N+1) ||2 (ab) (cd)
g(ab)(cd) =
e
h
(N+1)X
1
+ e
+ b h (ab)(cd) ,
g =
(4.9a)
(4.9b)
(4.9c)
for Sp(N)/U (N), where differentiated by the base coordinates Aa or ab are given in
(3.5) and (3.12).
The metrics of the submanifold defined by = 0 (d = 0) are

1
g(Aa)(Bb) =0 (, ) = b g (Aa) (Bb) , for GN,M ,
(4.10a)

1
g(ab)(cd) =0 (, ) = b f (ab) (cd) ,
(4.10b)
for SO(2N)/U (N),

1
g(ab)(cd) =0 (, ) = b h (ab)(cd) ,
(4.10c)
for Sp(N)/U (N),
where is the Khler potential of these manifolds found in (2.5), and differentiated
by two fields is given in (3.5c) or (3.12c). Therefore we find that the total spaces are the
complex line bundles over these Hermitian symmetric spaces as base manifolds, with
(or ) being a fiber. Actually, it was shown in [22] that there exists a Ricci-flat metric on
144
the complex line bundle over any KhlerEinstein manifolds. In the limit of b 0, these
base manifolds shrink and the manifolds become conifolds. The conical singularities are
resolved by GN,M , SO(2N)/U (N) and Sp(N)/U (N) of the radii b1/2g , b1/2f and b1/2h ,
respectively.
From the relation of GN,1 = CP N1 , we also have the complex line bundle over
CP N1 . The Khler potential (4.2) in the case of M = 1 (g = N ) coincides with the flat
one in the limit of b 0, but with a coordinate identification = N /N . The singular
limit is CN /ZN , and this orbifold singularity is resolved by CP N1 . This coincides with
the Calabi metric [15], so our manifolds can be considered as natural generalizations of the
Calabi metric.
5. Conclusion and discussions

We have constructed non-compact Khler manifolds, modifying the Khler quotient
construction of the Hermitian symmetric spaces of the classical groups by restricting
the gauge group U (M) to SU(M). We have presented the Ricci-flat metrics and their
Khler potentials on these manifolds. The essential point was that the partial differential
Eq. (3.3) was reduced to the ordinary differential Eqs. (3.10), (3.17) and (3.20), using the
isotropy transformation. These metrics contain the resolution parameter as an integration
constant, and the conical singularities are resolved by the Hermitian symmetric spaces with
the radii of the resolution parameter. They have been recognized as the line bundle over
the Hermitian symmetric spaces, and contain the Calabi metrics on the line bundle over
CP N1 . Our manifolds in lower dimensions are discussed in Appendix A, which are not
included in the list of [24].
Our method can be applied to the cases in which all non-compact directions can be
transformed to each other by the isotropy. Such a view point was discussed in [21] in
terms of the supersymmetric nonlinear realization. We can also construct other conifolds
with the isometry of the exceptional groups from the Hermitian symmetric spaces of the
exceptional groups [23]. We would like to discuss whether the deformation parameter
exists, as in the case of the conifold [5]. We also would like to clarify the relation between
our manifolds of the line bundle over Sp(2)/U (2) Q3 and the Spin(7) manifold in [25],
since both manifolds can be written in the form of R Sp(2)/SU(2). The investigation of
superconformal field theories corresponding to our manifolds is also an interesting task.
Acknowledgements
We would like to thank Michihiro Naka, Kazutoshi Ohta and Takashi Yokono for
valuable comments and discussions. This work was supported in part by the Grant-in-Aid
for Scientific Research (#13640283).
145
Appendix A. Isomorphism
We have sets of the isomorphism between the lower-dimensional base manifolds,
(i)
CP 1 SO(4)/U (2) Sp(1)/U (1) Q1 ,
(A.1a)
(ii)
CP SO(6)/U (3),
(A.1b)
(iii)
Sp(2)/U (2) Q ,
(A.1c)
(iv)
G4,2 Q ,
(A.1d)
in addition to the novel duality relation

(v)
GN,M GN,NM .
(A.1e)
In this appendix we show that total spaces on these base manifolds coincide. It gives a
nontrivial check for our results.
Before doing that we quote the results of the line bundle over QN2 = SO(N)
/[SO(N 2) U (1)], as a conifold [10]. The superfields and wi (i = 1, . . . , N 2)

2
constitute an N -vector as T = (1, wi , 12 N2
i=1 (wi ) ). The Khler potential differen
tiated by the invariant X = log and a coordinate transformation are
1
dK (N2)X
N2
= e
.
+ b N1 ,
=
dX
N 2
Note that the notation of X is different from that of [10].
(A.2)
A.1. EguchiHanson space

All of the lowest-dimensional manifolds of (A.1a) coincide with the EguchiHanson
gravitational instanton [7]. The matrix field I (I = 1, 2, 3, 4) and invariants XI
log det I I are

1
1 = 1
(A.3a)
,
X1 = log |1 |2 + log 1 + |1 |2 ,
1
1
0
0

1
,
2 = 2
(A.3b)
X2 = 2 log |2 |2 + 2 log 1 + |2 |2 ,
0
0
2

1
3 = 3
(A.3c)
,
X3 = log |3 |2 + log 1 + |3 |2 ,
3

1
1
2
2
4 = 4
(A.3d)
,
X4 = log |4 | + 2 log 1 + |4 | ,
4
2
12 (4 )2
for the cases of the base manifolds CP 1 , SO(4)/U (2), Sp(1)/U (1) and Q1 , respectively.
Relationsof the base and the fiber coordinates of these four manifolds are 1 = 2 =
3 = 4 / 2 and (1 )2 (2 )2 (3 )2 4 , respectively. Note that each fiber I
(I = 1, 2, 3, 4) consistently defines the same as the regular coordinate from (4.6) and
146
(A.2). The Khler potentials coincide up to a overall constant:

dK
dXI
e2X1 + b
eX2 + b
e2X3 + b
eX4 + b.
(A.4)
The orbifold singularity in C2 /Z2 is resolved by S 2 .

A.2. Complex four-dimensional Calabi metric
The matrix field I (I = 1, 2) and invariants XI log det I I are (i = 1, 2, 3)
"
!

3

1
,
X1 = log |1 |2 + log 1 +
1 = 1
(A.5a)
|wi |2 ,
wi
i=1
1
0
0
0
1
0
"
!
3

0
0
1
2
2
,
X2 = 3 log |2 | + 2 log 1 +
2 = 2
|i | ,
1 2
0
i=1

0
3
1
2 3 0
(A.5b)
3
for the cases of the base manifolds CP and SO(6)/U (3), respectively. Identifications of
the base and the fiber coordinates are wi = i and (1 )4 (2 )6 , respectively, where
each fiber I (I = 1, 2) consistently defines the same as the regular coordinate from
(4.6). The Khler potentials coincide up to a overall constant:

1
1
dK
e4X1 + b 4 e2X2 + b 4 .
(A.6)
dXI
The orbifold singularity in C4 /Z4 is resolved by CP 3 SO(6)/U (3).
A.3. Another metric with complex four dimensions
The matrix field I (I = 1, 2) and invariants XI log det I I are (i = 1, 2, 3)
1
0
0
1
1 = 1
3
1
,
2
X1 = 2 log |1 | + log 1 +
2
3

i=1
2 = 2
2
i=1 wi
X2 = log |2 | + log 1 +
2
(A.7a)
1
wi
1 3
2 "

1 2

|i | + 3 1 2 ,
2
2
3

i=1
3
2 "
1 2
|wi | +
wi ,

4
2
i=1
(A.7b)
147
for the cases of the Sp(2)/U (2) and Q3 base manifolds. Identifications of the base and the
fiber coordinates are
i
1 0 w
1
2
2
1
2 =
(A.8)
1
i
0 w2 ,
2
2
3
w
3
0
0
1
and (1 )6 (2 )3 , respectively. Again, each fiber I (I = 1, 2) consistently defines
the same as the regular coordinate from (4.6) and (A.2). The Khler potentials coincide
up to a overall constant:

1
1
dK
e3X1 + b 4 e3X2 + b 4 .
dXI
(A.9)
A.4. The line bundle over the Klein quadric (with complex five dimensions)
The embedding of G4,2 into CP 5 is known as the Plcker embedding. The matrix field
I (I = 1, 2) and invariants XI log det I I are (i = 1, 2, 3, 4)
1 0
0 1
1 = 1
1 3 ,
4
X1 = 2 log |1 | + log 1 +
2
4

"
|i | + |1 2 3 4 |
2
(A.10a)
i=1
2 = 2
1
wi
1 4
2
i=1 wi
!
X2 = log |2 | + log 1 +
2
4

i=1
2 "
4
1 2
|wi | +
wi ,

4
2
(A.10b)
i=1
for the cases of the base manifolds G4,2 and Q4 . Identifications of the base and the fiber
coordinates are
1
i
0
0

2
2
1
w1
1
i
0
0
w2
2 2 2
=
,
(A.11)
3 0
i
0
1 w3
2
2
4
w4
i
1
0
0
2
and (1 (2 , respectively, from (4.6) and (A.2). Each fiber defines the same the
regular fiber coordinate . The Khler potentials coincide up to a overall constant:
)8
)4

1
1
dK
e4X1 + b 5 e4X2 + b 5 .
dXI
(A.12)
148
A.5. Duality between the Grassmann manifolds

The matrix field I (I = 1, 2) and invariants XI log det I I are

1M
,
X1 = M log |1 |2 + log det 1M + 1 1 ,
1 = 1
(A.13a)
1

1NM
,
X2 = (N M) log |2 |2 + log det 1NM + 2 2 ,
2 = 2
2
(A.13b)
for the cases of the base manifolds GN,M and GN,NM , respectively. Here 1 and 2 are
[(N M) M]- and [M (N M)]-matrices, respectively. Identifications of the base
and the fiber coordinates are 1 = 2 T , and (1 )NM (2 )N(NM) , respectively, due
to (4.6), in which each fiber defines the same regular coordinate . The Khler potentials
coincide up to a overall constant:

1
1
dK
eNX1 + b g eNX2 + b g .
dXI
(A.14)
References
[1] B. Zumino, Phys. Lett. B 87 (1979) 203;
L. Alvarez-Gaum, D.Z. Freedman, Commun. Math. Phys. 80 (1981) 443.
[2] L. Alvarez-Gaum, Nucl. Phys. B 184 (1981) 180;
C.M. Hull, Nucl. Phys. B 260 (1985) 182;
L. Alvarez-Gaum, P. Ginsparg, Commun. Math. Phys. 102 (1985) 311.
[3] M.T. Grisaru, A.E.M. van de Ven, D. Zanon, Phys. Lett. B 173 (1986) 423;
M.T. Grisaru, A.E.M. van de Ven, D. Zanon, Nucl. Phys. B 277 (1986) 388;
M.T. Grisaru, A.E.M. van de Ven, D. Zanon, Nucl. Phys. B 277 (1986) 409.
[4] D. Nemeschansky, A. Sen, Phys. Lett. B 178 (1986) 365.
[5] K. Higashijima, T. Kimura, M. Nitta, Phys. Lett. B 515 (2001) 421, hep-th/0104184.
[6] M.B. Stenzel, Manuscripta Math. 80 (1993) 151;
M. Cvetic, G.W. Gibbons, H. L, C.N. Pope, hep-th/0012011.
[7] T. Eguchi, A.J. Hanson, Phys. Lett. B 74 (1978) 249.
[8] P. Candelas, X.C. de la Ossa, Nucl. Phys. B 342 (1990) 246;
R. Minasian, D. Tsimpis, Nucl. Phys. B 572 (2000) 499, hep-th/9911042;
K. Ohta, T. Yokono, JHEP 0002 (2000) 023, hep-th/9912266.
[9] L.A. Pando Zayas, A.A. Tseytlin, JHEP 0011 (2000) 028, hep-th/0010088;
L.A. Pando Zayas, A.A. Tseytlin, Phys. Rev. D 63 (2001) 086006, hep-th/0101043.
[10] K. Higashijima, T. Kimura, M. Nitta, Phys. Lett. B 518 (2001) 301, hep-th/0107100.
[11] K. Higashijima, M. Nitta, Prog. Theor. Phys. 103 (2000) 635, hep-th/9911139.
[12] K. Higashijima, M. Nitta, Prog. Theor. Phys. 103 (2000) 833, hep-th/9911225.
[13] K. Higashijima, M. Nitta, in: Proceedings of Confinement 2000, World Scientific, 2001, p. 279,
hep-th/0006025;
K. Higashijima, M. Nitta, in: Proceedings of ICHEP 2000, World Scientific, 2001, p. 1368,
hep-th/0008240.
[14] K. Higashijima, T. Kimura, M. Nitta, M. Tsuzuki, Prog. Theor. Phys. 105 (2001) 261, hepth/0010272.
[15] E. Calabi, Ann. Scient. Ec. Norm. Sup. 12 (1979) 269.
149
[16] A. Dadda, P. Di Vecchia, M. Lscher, Nucl. Phys. B 152 (1979) 125.

[17] S. Aoyama, Nuovo Cimento A 57 (1980) 176.
[18] U. Lindstrm, M. Rocek, Nucl. Phys. B 222 (1983) 285;
N.J. Hitchin, A. Karlhede, U. Lindstrm, M. Rocek, Commun. Math. Phys. 108 (1987) 535.
[19] M.A. Luty, W. Taylor IV, Phys. Rev. D 53 (1996) 3339, hep-th/9506098.
[20] M. Bando, T. Kuramoto, T. Maskawa, S. Uehara, Phys. Lett. B 138 (1984) 94;
M. Bando, T. Kuramoto, T. Maskawa, S. Uehara, Prog. Theor. Phys. 72 (1984) 313;
M. Bando, T. Kuramoto, T. Maskawa, S. Uehara, Prog. Theor. Phys. 72 (1984) 1207;
W. Lerche, Nucl. Phys. B 238 (1984) 582;
G.M. Shore, Nucl. Phys. B 248 (1984) 123.
[21] M. Nitta, Int. J. Mod. Phys. A 14 (1999) 2397, hep-th/9805038.
[22] D.N. Page, C.N. Pope, Class. Quantum Grav. (1987) 213.
[23] K. Higashijima, T. Kimura, M. Nitta, hep-th/0102185.
[24] M. Cvetic, G.W. Gibbons, H. L, C.N. Pope, Nucl. Phys. B 617 (2001) 151, hep-th/0102185.
[25] M. Cvetic, G.W. Gibbons, H. L, C.N. Pope, hep-th/0103155;
Y. Konishi, M. Naka, hep-th/0104208.

The covariant form of the gauge anomaly

on noncommutative R2n
C.P. Martn
Departamento de Fsica Terica I, Facultad de Ciencias Fsicas, Universidad Complutense de Madrid,
28040 Madrid, Spain
Abstract
The covariant form of the non-Abelian gauge anomaly on noncommutative R2n is computed for
U (N) groups. Its origin and properties are analyzed. Its connection with the consistent form of the
gauge anomaly is established. We show along the way that bi-fundamental U (N) U (M) chiral
matter carries no mixed anomalies, and interpret this result as a consequence of the half-dipole
structure which characterizes the charged non-commutative degrees of freedom. 2002 Elsevier
Science B.V. All rights reserved.
1. Introduction
Field theories of fermions with chiral couplings to gauge fields on commutative
manifolds play a prominent roleat least up to a few TeVin the description of
Nature. Field theories over noncommutative space-time [1,2] may turn out to be
phenomenologically relevant at the TeV scale and above [311]. It is therefore a must
to understand the properties of quantum field theories of fermions chirally coupled to
gauge fields on noncommutative manifolds. See Refs. [1215] for the mathematics of
noncommutative manifolds.
The chief feature of quantum field theories of chiral fermions interacting with gauge
fields is that they are liable to carry gauge anomaliesother types of anomalies such
the conformal anomaly [16] will not be discussed here. It is a well established fact [17
23] that if space-time is commutative, a gauge anomaly comes in either of two guises,
namely, its consistent form or its covariant form. The consistent form of the anomaly
satisfies the WessZumino consistency conditions [24], the covariant form does not. One
E-mail address: carmelo@elbereth.fis.ucm.es (C.P. Martn).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 4 2 - 3
C.P. Martn / Nuclear Physics B 623 (2002) 150164
151
can retrieve either form of the anomaly from the other by adding to the corresponding
current a polynomial of the gauge fields and its derivatives. For noncommutative R4 the
consistent form of the gauge anomaly has been obtained in a number of papers [2527]
see also Refs. [2830] for general analysis of chiral anomalies on noncommutative spaces
and Refs. [31,32] for explicit computations. As for its covariant form, there is as yet
no thorough discussion of the gauge anomaly for noncommutative space-timealthough
some results have been issued in Ref. [33]. The purpose of this paper is to remedy this
situation. First, by using path integral techniques, we shall compute explicitly the form of
the gauge anomaly on noncommutative R2n for U (N) groups. In so doing, we shall see that
the covariant form of the gauge anomaly is associated with a given definition of the path
integral. This definition being a -deformation of the ordinary one as given in Refs. [18,22,
23]. Then, we shall show that the covariant form of the gauge anomaly can be turned into
the consistent form of it, by adding to the covariant current a -polynomial of the gauge
field and the field strength; this -polynomial being a -deformation of the polynomial for
the commutative R2n case. Finally, we shall analyze the transformation properties, under
gauge transformations of the gauge field, of both the consistent and covariant currents.
We shall thus show that in the presence of the gauge anomaly the consistent current
that which can be obtained by functional differentiation of the effective actioncannot
transform covariantly; whereas, the covariant current does transform covariantly and,
hence, it cannot be the functional derivative of the effective action with respect to the gauge
field. For commutative R2n , these properties of the currents were established in Refs. [19,
20,22,23].
2. The covariant form of the gauge anomaly: fundamental matter

j
Let R denote a right handed fermion, R = P+ j , P+ = (1 + 2n+1 )/2, carrying the

fundamental
representation of the group U (N). The matrix 2n+1 is given by 2n+1 =

, where the gamma matrices , = 1, . . . , 2n, are Hermitian matrices
(i)n 2n
=1
j
which satisfy { , } = 2 . The physics of R interacting with a background U (N)

gauge field on noncommutative R2n is ruled by the classical action

ij j .
S = d 2n x i i D(A)
(1)

i j j = i(/
The operator i D(A),
which acts on the Dirac spinor j as follows i D(A)
i +
i
j
Aj P+ ), is not an Hermitian operator, but it is an elliptic operator. The Dirac
spinor j carries the fundamental representation of U (N) and the complex matrix Aij ,
with (Aij ) = Ai , is the U (N) gauge field. The indices i, j run from 1 to N . The
previous action is invariant under the following chiral gauge transformations:
j
( A )i j = i j Aik k j + i k Akj ,
( )i = i j P+ j ,
k = k k i P ,
( )
= j
(2)
where P = (1 2n+1 )/2. The complex functions j

i , i, j = 1, . . . , N , are the
infinitesimal gauge transformation parameters. The symbol denotes the Moyal product
i
152
of functions on R2n . The Moyal product is given by

d 2n p
d 2n q i(p+q) x i p q
(f g)(x) =
e
e 2
f (p)g(q).
(2)2n
(2)2n
Here, is an anti-symmetric real matrix either of magnetic type or light-like type. It is
for these choices of matrix that a unitary theory exists at the quantum level [3437].
To define the partition function,

Z[A] d d eS[A] ,
(3)
of the quantum theory with classical action given in Eq. (1), we shall follow Fujikawa [18]
and use the set of eigenvalues and the set of eigenfunctions of the Hermitian operators
i D(A)
. These sets are defined by the following equations:

and i D(A)(i
D(A))
(i D(A))

2

i D(A)
i D(A)
i D(A)
m = 2m m ,
i D(A)
m = m m ,
1

i D(A)m ,
if m = 0, and i D(A)
if m = 0,
m =
m = 0,
m

1

i D(A) m , if m = 0, and i D(A)
m =
m = 0, if m = 0,
m

(x)m (x) = mm ,

d 2n x m
(x)m (x) = mm .
d 2n x m
(4)
We take without loss of generality m 0. We next define the fermionic measure as follows

d d =
(5)
d bm dam .
m

Here, am and bm are Grassmann variables defined by the expansions = m am m and

. Notice that this definition of d d is the obvious generalization to the

= m bm m
noncommutative framework of the definition in Refs. [22,23]. Then,

d b m dam e m m bm am ;
d d eS[A,,]
m

which after Grassmann integration leads to Z[A] m m . Notice that we have taken into
account Eq. (3). Hence, we have formally defined the partition function of the theory, Z[A],
i D(A).

as the determinant of the square root of the operator (i D(A))
Now, it is not difficult to show that if g = 1 + is an infinitesimal gauge transformation,
i D(A)
g )) i D(A
g )(1 P+ ) = (i D(A))

+ 0(2 ). Hence,
one has (1 + P+ )(i D(A
g
2
m (A ) = m (A) + O( ), so that Z[A], as defined above, is formally gauge invariant
under infinitesimal gauge transformations. We can make this statement rigorous by using
PauliVillars regularization or zeta function regularization. It is thus clear that for our
definition of partition function the chiral gauge anomaly cannot be interpreted as the lack
of invariance of W[A] (W[A] = ln Z[A]) under infinitesimal gauge transformations. An
interpretation which holds true for the consistent form of the anomaly [17,19]. Let us show
next that, as in the ordinary case [18], the covariant form of the anomaly comes from
the lack of invariance under infinitesimal chiral gauge transformations of the fermionic
153
measure defined abovesee Eq. (5). Let = + and = + ,

is
where

and
given in Eq. (2). Let {am }m and {bm }m be given by the expansions = m am
m

m . Then, the identity
= m bm

d d eS[A,,] d d eS[A, , ] ,
leads to
j

d 2n x ji (x) D [A]J(cov) i (x) = J A[, A](cov) .
(6)

da
2
J , which is defined by the equation m d b m

m
m d bm dam = J + O( ), is equal
to

m
P m m
P+ m
d 2n x
m
(cov)
and the current J

is defined by the identity

i
(cov) j
j (x)( P+ ) .
J
i (x) = i
denotes the vacuum expectation value as given by

d d eS[A]
=
,
d d eS[A]
with the fermionic measure d d as defined by Eq. (5).
As it stands, the right-hand side of Eq. (6) is ill-defined; we shall obtain a well-defined
object out of it by using the Gaussian cut-off furnished by the eigenvalues 2m in Eq. (4).
This well-defined object, which we shall denote by A[, A](cov) , is the covariant form of
the gauge anomaly:

2
2
(cov)
A[, A]
= lim
em / m
P+ m m
P m
d 2n x
= lim
2n
d x

2
2
P+ em / m m

2
2
P em / m m

2

= lim
P+ e(i D(A)) i D(A)/ m m
d 2n x

= lim
d 2n x Tr

2

P ei D(A)(i D(A)) / m m
d 2n p
2 /2 ipx
/
eipx .
tr 2n+1 e(iD(A))
e
2n
(2)
The last line of the previous equation is obtained by changing to a plane-wave basis. In this
/ (A) = i(/ + A
/ ) denotes the Dirac operator, and Tr and tr stand for traces over
last line iD
the U (N) and Dirac matrices, respectively.
154
Let D(A) = + A . Taking into account that (iD

/ (A))2 = (D(A) D(A) +
ipx
and that D(A) D(A) (f e ) = ((ip + + A )2 f ) eipx , and that
the Moyal product is associative, one can show that [25]

d 2n p
2 /2 ipx
/
eipx
tr 2n+1 e(iD(A))
e
d 2n x Tr
lim
2n
(2)

1
d 2n p p2 /2
2n
lim
= d x Tr
e
m!2m
(2)2n
m=0

tr 2n+1 ( + A )2 + 2ip ( + A )

m
1
ipx
ipx
,
+ F I e e
2
1
2 F )
where I is the symbol for the unit function on R2n . Putting it all together, we conclude that

in
1 2n
Tr
d 2n x F1 2 F2n1 2n (x).
A[, A](cov) =
(4)n n!
It is clear that A[, A](cov) does not satisfy the WessZumino consistency conditions

1 A(2 , A) 2 A(1 , A) = A [1 , 2 ], A .
(con)
Hence, J (x) cannot be expressed as the derivative of effective action W[A] =

ln Z[A] with respect to the gauge field.
3. The covariant form of the gauge anomaly: bi-fundamental and adjoint chiral
matter
i = P i , i = 1, . . . , N and
We shall consider a bi-fundamental [38] chiral fermion Rj
+ j
j = 1, . . . , M coupled to a U (N) gauge field, say, A , and a U (M) gauge field, say, B .
The classical action of this theory reads

i

B) j .
S = d 2n x j i i D(A,
(7)

The elliptic operator i D(A,
B) acts on the bi-fundamental Dirac spinor i j as follows

i

j
1

j2 j1 P+ ii21 Bj2 1 P+ i2 j2 .
i D(A,
B) 1 j1 = i / i1 i2 j2 j1 + Aii
2
We are using the following notation with regard to the -product: A ) A and
B ) B . Throughout this section, the i-indices run from 1 to N and the j -indices
run from 1 to M. A and B are anti-Hermitian matrices.
The action in Eq. (7) is invariant under the following infinitesimal gauge transformations

((,) )i1 j1 = i1 i2 P+ i2 j1 P+ i1 j2 j2 j1 ,

j1 i1 = j1 i2 i2 i1 P j1 j2 j2 j1 P ,
((,) )
155
3
1
( A )i1 i2 = i1 i2 Aii
i3 i2 + i1 i3 Ai
,
3
2
( B )j1 j2 = j1 j2 Bj1 3 j3 j2 + j1 j3 Bj3 2 ,
(8)
where i1 i2 = i2 i1 , i1 , i2 = 1, . . . , N , and j1 j2 = j2 j1 , j1 , j2 = 1, . . . , M, are the

infinitesimal gauge transformation parameters.
Following the strategy developed in the previous section, we obtain

Z[A, B] d d eS[A,B] =
d b m dam e n m bm am =
m [A, B].
m

The Grassmann variables, am and bm , are given now by the expansions = m am m

and = m bm m
; m and m being the eigenvectors solving the eigenvalue problems

and satisfying the identities that one gets by replacing in Eq. (4) i D(A)
with i D(A,
B).
2m [A, B] are the eigenvalues of the problems so obtained. These eigenvalues are invariant
under the infinitesimal gauge transformations of Eq. (8). Hence, the zeta regularization
version of Z[A, B] above is gauge invariant.
Proceeding as in the previous section, one obtains the covariant form the gauge anomaly
equation for bi-fundamental chiral matter:

i
j

d 2n x ii21 D [A]J(A,cov) 2 i1 + j1 j2 D [B]J(B,cov) 2 j1
= A[, ; A, B](cov) .
Here the currents
and
(9)
(B,cov)
J
are defined, respectively, by the identities

i1
j
j1 i12 (x)( P+ ) ,
(B,cov) j1

j1
i1 (x)( P+ ) ;
J
j2 (x) = i
i1
j2
J(A,cov)
i1
(A,cov)
J
i2 (x) = i
(10)
and A[, ; A, B](cov) is given by

A[, ; A, B](cov)

2
2
= lim
em [A,B]/ m
P+ m m
P m
d 2n x
lim
d 2n x
2
2
em [A,B]/ m
P+ m m
P m .
(11)
Let iD
/ (A, B) = i(1NN 1MM / + 1MM A
/ +B
/ 1NN ) the Dirac operator
acting on the bi-fundamental Dirac spinor i j , where 1NN and 1MM denote the unit
matrices. It can be shown that the right-hand side of Eq. (11) can be written as follows

lim
d 2n x TrMNN TrMMM
d 2n p
2 /2 ipx
/
tr 2n+1 e(iD(A,B))
e

eipx
2n
(2)
156

lim
d 2n x TrMNN TrMMM

d 2n p ipx
2 /2 ipx
/

.
tr e
2n+1 e(iD(A,B))
e
2n
(2)
(12)
tr denotes the trace over the gamma matrices, and TrMNN and TrMNN stand for the trace
over N N and M M complex matrices, respectively. It is not difficult to see that

D
/ (A, B) 2 = 1NN 1MM + 1MM A B 1NN 2

1
+ 1MM F [A] F [B] 1NN .
2
In the previous equation A , B , F [A] and F [B] are to be understood as operators
acting on a appropriate matrix valued functions f and g as follows: A )f = A f ,
B )g = g B , F [A])f = F [A] f , F [B])g = g F [B]. f takes values
on N N complex matrices and g takes values on the M M complex matrices.
F [A] and F [B] are the field strengths of A and B, respectively. 1NN and 1MM
are, respectively, the identity matrices of rank N and M. Now, taking into account that
eipx f (x) eipx = f (x + p), with (p) = p , it is not difficult to show that
Eq. (12) can be cast into the form

d 2n x TrMNN TrMMM

1
d 2n p p2 /2
lim
e

2m
m!
(2)2n
m=0

2
tr 2n+1 1NN 1MM + 1MM A (x) B (x + p) 1NN

+ 2ip 1NN 1MM + 1MM A (x) B (x + p) 1NN

m
1
+ 1MM F [A](x) F [B](x + p) 1NN
I
2

eipx eipx
d 2n x TrMNN TrMMM

d 2n p p2 /2
1

e
lim
2m
m!
(2)2n
m=0

tr 2n+1 eipx eipx

2
1NN 1MM + 1MM A (x p) B (x) 1NN

+ 2ip 1NN 1MM + 1MM A (x p) B (x) 1NN
157

m
1
+ 1MM F [A](x p) F [B](x) 1NN
I . (13)
2
I is the symbol for the unit function on R2n . Let us now recall that tr(2n+1 1 k )
= 0, if k < 2n; and that for a given value of m the limit lim above vanishes, if the
number of powers of turns negative upon rescaling p to p. Keeping these two results
in mind, one can show that the expression in the previous equation is equal to

d 2n x TrMNN TrMMM

in
d 2n p p2 1 2n
lim
F1 2 [A](x) F1 2 [B](x + p)
e
2n
n!
(2)

F2n1 2n [A](x) F2n1 2n [B](x + p) I

d 2n x TrMNN TrMMM

in
d 2n p p2 1 2n
lim
F1 2 [A](x p) F1 2 [B](x)
e
2n
n!
(2)

(14)
F2n1 2n [A](x p) F2n1 2n [B](x) I .
Generally speaking, in noncommutative quantum field theory, the limits and
p 0 do not commute as a consequence of the intriguing UV/IR mixing [39]. Then, to
define the renormalized theory, one has make a choice regarding the order of these limits.
One would like to obtain the renormalized noncommutative theory at p = 0 by taking the
limit p 0 of renormalized one at p = 0. Hence, we shall take the limit first
and then take the limit p 0. Now, the gauge fields A and B satisfy the boundary
conditions F [A](y) 0 and F [B](y) 0 as |y| . It is thus plain that (14) is
equal to

in
1 2n
M
Tr
d 2n x F1 2 [A] F2n1 2n [A](x)
MNN
(4)n n!

(i)n 1 2n
Tr
N
d 2n x F1 2 [B] F2n1 2n [B](x). (15)
M
MM
(4)n n!
Putting it all together (see (9)(15)), we conclude that the covariant form of the anomaly
A[, ; A, B](cov) reads thus
A[, ; A, B](cov)

in
1 2n
=M
Tr
d 2n x F1 2 [A] F2n1 2n [A](x)
M
NN
(4)n n!

(i)n 1 2n
TrMMM d 2n x F1 2 [B] F2n1 2n [B](x).

N
(4)n n!
(16)
Notice that as in the consistent case [40,41] there are no mixed anomalies. Also notice that
if in (14) we set = 0 before sending to , i.e., we go to commutative space, the mixed
158
anomalies pop-up back; and that it is the characteristic half-dipole structure of the charged
degrees of freedom of the noncommutative field theoriessee the Fi i+1 [A](x p)
and Fi i+1 [B](x + p) terms in the mixed contributions of (14)which is responsible
for the lack of mixed anomalies in the noncommutative arena. That charged degrees of
freedom have a half-dipole structure rather than a dipole structure [42,43] was unveiled in
Ref. [44].
The consistent form of the gauge anomaly for an adjoint right-handed fermion can be
obtained by setting A = B in Eq. (16). We thus conclude that if D = 4m (D is the spacetime dimension) there is no gauge anomaly, but if D = 4m + 2, the anomaly is 2N times
the anomaly in the fundamental representation.
4. Redefinition of currents
In this section we shall show that there exists a -polynomial, X , of A and F, such that
J(con) (x) = J(cov) (x) + X (x).
(17)
Here J(con) (x) denotes the consistent gauge current for a fundamental right handed
(cov)
fermionsee Refs. [2527]and J (x) stands for the corresponding covariant gauge
current. In view of the results presented in the previous section, the generalization of the
analysis we are about to begin to bi-fundamental and/or adjoint right handed fermions is
trivial.
To compute X we shall adapt to the case at hand the techniques of Ref. [20]. To do so
we shall employ the formalism of differential forms and BRST cohomology introduced in
Ref. [26].
Let J (con) and J (cov) be the dual currents
1
1 2 2n J(con)
dx 2 dx 2n ,
1
(2n 1)!
1
1 2 2n J(cov)
dx 2 dx 2n .
J (cov) =
1
(2n 1)!
J (con) =
(18)
These currents are (2n 1)-differential forms in the sense of Ref. [26]. Let C be the
ghost zero-form introduced through the BRST transformations: sA = DC = dC + [A, C],
sc = C C, s is the BRST operator, d is the exterior derivative, and A = A dx .
s and d satisfy s 2 = d 2 = sd + ds = 0. We introduce next the two-form field-strength
F = 12 F dx dx = dA + A A. Then, J (con) and J (cov) are defined so that they satisfy,
respectively, the consistent form and the covariant form of gauge anomaly equation:

Tr DC J (cov) = A(C, A)(cov) , (19)
Tr DC J (con) = A(C, A)(con) and
where
A(C, A)(con) =
in
n
(2) (n + 1)!

Q1 2n (C, A, F)
(20)
and
A(C, A)(cov) =
in
(2)n n!
y
Tr C Fn .
159
(21)
Q12n (C, A, F), which can be obtained by solving the descent equations, reads
1
Q12n (C, A, F) = (n + 1)
dt (1 t)
n1

q

y
Tr C d Fkt A Fn1k
,
t
(22)
k=0
Ft = dAt + A2t
with
and At = tA. Fk denotes the kth power of F with respect to the Moyal
product. An expression like Tr(E1 E2 Em ) denotes the equivalence class obtained
by imposing on the space of objects of the type Tr(E1 E2 Em ) the relationship
Tr(E1 E2 Em ) (1)km (k1 ++km1 ) Tr(Em E1 Em1 ). Ei denotes a form of
degree ki . See Ref. [26] for further details.
To find X ,
X=
1
1 2 2n X1 dx 2 dx 2n ,
(2n 1)!
(23)
such that X satisfies Eq. (17), we shall first show that Q1 2n (C, A, F) in Eq. (22) is also
given by the following equation
y
q
Q12n (C, A, F) = (n + 1) Tr C Fn
|
n1
1 t

k
n1k
(n + 1) dt Tr C D
(24)
Ft At Ft
.
k=0
It can be shown that the right-hand side of Eq. (22) is equal to

1
(n + 1)
dt
Tr C Fnt
1
+ (n + 1)
dt
n1

q
y
(t 1) Tr Fkt At Fn1k
[A, C] .
t
k=0
(25)
Now, taking into account that (Dt D)C = (t 1)[A, C] and that Tr DO = dTr O for
a form, O, of even degree; one readily shows that Eq. (25) can be turned into the following
expression
1
1
n1
y
q k
q
y
n
Tr Ft At Fn1k
dt Tr C Ft d dt
C
(n + 1)
t
0
1
+
dt
0
1
k=0
n1

q k
y
Tr Ft Dt At Fn1k
C
t
k=0

n1

q k
y
n1k
Tr Ft At Ft
dt
DC
.
k=0
160
Upon employing that Dt At = tt Ft and that t Fnt =

equation can be converted into the following one
1
(n + 1)
y
q
dt Tr C Fnt d
1
dt
0
1
q
y
dt tt Tr Fnt C
+
0
n1
k
k=0 Ft
t Ft Fn1k
, the previous
t
n1

q k
y
Tr Ft At Fn1k
C
t
k=0
1

n1

q k
y
n1k
Tr Ft At Ft
dt
DC
.
k=0
Partial integration yields then

(n + 1)

q
Tr C F
1
dt
0
1
d
n1

q k
y
Tr Ft At Fn1k
DC
t
k=0

n1

q k
y
n1k
Tr Ft At Ft
dt
C
.
k=0
This equation and

y
y q
q
C
C = Tr D Fkt At Fn1k
d Tr Fkt At Fn1k
t
t
q
y
DC
Tr Fkt At Fn1k
t
finally lead to Eq. (24).
We are now ready to compute X so that Eqs. (17) and (23) hold:

(con)
Tr C DJ (cov)
Tr C DX = Tr C DJ

q
y
1
in
=
Q2n (C, A, F) (n + 1) Tr C Fn
n
(2) (n + 1)!
n1
|
1 t

i (n+2)
k
n1k
=
dt Tr C D
Ft At Ft
.
(2)n n!
0
(26)
k=0
To obtain the previous array of identities, Eqs. (19)(24) are to be taken into account. In
view of Eq. (26), we conclude that the following choice of X ,
i (n+2)
X=
(2)n n!
1
dt
0
n1

Fkt At Fn1k
,
t
(27)
k=0
would do the job. Notice that the result we have obtained is the naive -deformation of the
ordinary expression in Ref. [20] without the symmetrization carried out therein.
161
5. Currents and gauge transformations

In this section we shall study the behaviour under gauge transformations of the
consistent and covariant dual currentsJ (con) and J (cov) in Eq. (18), respectively. We
shall employ the techniques of Ref. [20] and show that, when there is an anomaly, the
following equation does not hold

sJ (con) = C, J (con) ,
but the following equation does

sJ (cov) = C, J (cov) .
(28)
The consistent current is obtained from the effective action W[A] by functional
differentiation of the latter, i.e.,

W[A] = Tr A J (con) .

The operator is given by A /A. The BRST variation of infinitesimal one-form A
is defined to be sA = [A, C]. It can be readily seen that s = s. We next introduce the
anti-derivation l as follows lA = 0, lF = A. It can be shown that ld + dl = , on the space
of polynomials of A and F with respect to the Moyal product.
The consistent form of the gauge anomaly equation runs thus in terms of W[A]:

in
sW[A] =
Q12n (C, A, F).
(2)n (n + 1)!
Q12n (C, A, F) is given in Eq. (22). Acting with on both sides of the previous equation,
one obtains

in
Q1 2n (C, A, F) = sW[A] = sW[A]
(2)n (n + 1)!

= s Tr A J (con) = s Tr A J (con)

= Tr A sJ (con) C, J (con) .
(29)
Hence, the existence of the gauge anomalyQ12n (C, A, F) does not vanishprevents
J (con) from transforming covariantly. Notice that Eq. (29) tell us that the behaviour of
J (con) under gauge transformations is given by the anomaly.
Let us finally show that the covariant current, J (cov) , obtained by subtracting X in
Eq. (27) from J (con) , deserves that name indeed, i.e., it satisfies Eq. (28). In view of
Eq. (29), we just have to show that

in
Tr A sX + [C, X ] =
Q12n , (C, A, F)
(2)n (n + 1)!

with X given in Eq. (23). Taking into account the (graded) cyclicity of Tr Ek1 Ekn ,

and that tA = lFt and that lFnt = nk=0 Fn1k
lFt Fkt , with Ft = dAt + A2t ; one shows
t
162
that
Tr A sX + [C, X ]

=s
in
Tr A X
(2)n (n + 1)!
1
sl (n + 1)
dt Tr A Fnt .
0
It is plain that
1
sl (n + 1)
dt Tr A Fnt
0

=
1
sl (n + 1)
dt
Tr A Fnt

.
Now, the descent equation formalism of Ref. [26] leads to sQ02n+1 (A, F) = dQ12n (C, A, F)
1
where Q0 2n+1 (A, F) = (n + 1) 0 dt Tr A Fnt and Q1 2n (C, A, F) is given in Eq. (22).
Putting it all together, we get

in
Tr A sX + [C, X ] =
slQ02n+1 (A, F)
(2)n (n + 1)!

in
in
0
=
(A,
F)
=
lsQ
ldQ1 2n (C, A, F)
2n+1
(2)n (n + 1)!
(2)n (n + 1)!

in
in
1
(C,
A,
F)
=
=
(dl
+
)Q
Q12n (C, A, F).
2n
(2)n (n + 1)!
(2)n (n + 1)!
6. Conclusions
In this paper we have considered the noncommutative gauge anomaly for U (N) gauge
groups. We have shown that the covariant form of gauge anomaly on noncommutative
R2n can be understood as the lack of invariance of the fermionic measure under chiral
gauge transformations of the fermion fields. This lack of invariance is given by a non-trivial
Jacobian, which when defined by using an appropriate regularization yields the covariant
form of the anomaly. By using these path integral techniques, we have finally computed
the covariant form of the gauge anomaly on R2n to show that it is given by a -polynomial
of the gauge field strength. The covariant form of the gauge anomaly on even dimensional
space is thus seen to be given by an appropriate -deformation of the ordinary expression.
We have proved that one can trade the covariant form of the gauge anomaly for the
consistent one by adding to the covariant current a -polynomial of the gauge field and
the gauge field strength. We have computed this polynomial explicitly. Gauge anomalies
are thus given by local expressions in the sense of noncommutative geometryof course,
these expressions are non-local from the ordinary quantum field theory point of view.
We have seen that the gauge transformation properties of the consistent current are given
by the consistent form of the gauge the anomaly. The existence of the gauge anomaly
prevents the consistent current from transforming covariantly, but allows the covariant
current to transform covariantly under gauge transformations of the gauge field.
163
It is worth stressing that in the course of our path integral computations we have proved
that on noncommutative R2n the covariant form of the gauge anomaly for bi-fundamental
chiral matter carries no mixed anomaly. For the noncommutative theory, the absence of
these mixed anomalies is interpretedsee Ref. [41] for an interpretation in terms of the
GreenSchwarz mechanismas a consequence of the half-dipole structure [44] which is
characteristic of the noncommutative charged fields. From the covariant form of the gauge
anomaly for bi-fundamental chiral matter, one readily obtains the covariant form of the
anomaly form adjoint chiral fermions. If D denotes the espace dimension, our resultsfor
adjoint chiral right-handed fermion in the continuumrun thus: there is no anomaly at
D = 4m; at D = 4m + 2, the anomaly is 2N times the fundamental anomaly. Interestingly
enough one can formulate such theories on the lattice in an anomaly free manner for any
even integer D [45].
It is an interesting task to try to extend the results presented here to groups other than
the U (N) groups. New techniques and ideas such us the ones introduced in Refs. [46,47]
will be unavoidably needed, if one is to succeed.
Finally, as we were writing the closing sentences in this paper, we became aware of
Ref. [48]. The results discussed above are in complete harmony with the results presented
in this last reference.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
M.R. Douglas, N.A. Nekrasov, Noncommutative field theory, hep-th/0106048.

R.J. Szabo, Quantum field theory on noncommutative spaces, hep-th/0109162.
I.F. Riad, M.M. Sheikh-Jabbari, JHEP 0008 (2000) 045, hep-th/0008132.
H. Arfaei, M.H. Yavartanoo, Phenomenological consequences of non-commutative QED, hep-th/0010244.
J.L. Hewett, F.J. Petriello, T.G. Rizzo, Phys. Rev. D 64 (2001) 075012, hep-ph/0010354.
P. Mathews, Phys. Rev. D 63 (2001) 075007, hep-ph/0011332.
S. Baek, D.K. Ghosh, X. He, W.Y. Hwang, Phys. Rev. D 64 (2001) 056001, hep-ph/0103068.
H. Grosse, Y. Liao, Anomalous C-violating three photon decay of the neutral pion in noncommutative
quantum electrodynamics, hep-ph/0104260.
H. Grosse, Y. Liao, Pair production of neutral Higgs bosons through noncommutative QED interactions at
linear colliders, hep-ph/0105090.
M. Chaichian, P. Presnajder, M.M. Sheikh-Jabbari, A. Tureanu, Noncommutative Standard Model: model
building, hep-th/0107055.
X. Wang, M. Yan, Noncommutative QED and muon anomalous magnetic moment, hep-th/0109095.
A. Connes, Noncommutative Geometry, Academic Press, 1994.
J. Madore, An Introduction to Noncommutative Geometry and its Applications, Cambridge Univ. Press,
1999.
G. Landi, An introduction to noncommutative spaces and their geometry, hep-th/9701078.
J.M. Gracia-Bondia, J.C. Varilly, H. Figueroa, Elements of noncommutative geometry, Birkhauser, Boston,
USA, 2001.
T. Nakajima, Conformal anomalies in noncommutative gauge theories, hep-th/0108158.
L. Alvarez-Gaume, P. Ginsparg, Nucl. Phys. B 243 (1984) 449.
K. Fujikawa, Phys. Rev. D 29 (1984) 285.
W.A. Bardeen, B. Zumino, Nucl. Phys. B 244 (1984) 421.
L. Alvarez-Gaume, P. Ginsparg, Ann. Phys. 161 (1985) 423;
L. Alvarez-Gaume, P. Ginsparg, Ann. Phys. 171 (1985) 233, Erratum.
K. Fujikawa, Phys. Rev. D 31 (1985) 341.
H. Banerjee, R. Banerjee, P. Mitra, Z. Phys. C 32 (1986) 445.
164
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
H. Banerjee, Chiral anomalies in field theories, hep-th/9907162.

J. Wess, B. Zumino, Phys. Lett. B 37 (1971) 95.
J.M. Gracia-Bondia, C.P. Martin, Phys. Lett. B 479 (2000) 321, hep-th/0002171.
L. Bonora, M. Schnabl, A. Tomasiello, Phys. Lett. B 485 (2000) 311, hep-th/0002210.
M.T. Grisaru, S. Penati, Phys. Lett. B 504 (2001) 89, hep-th/0010177.
E. Langmann, On anomalies and noncommutative geometry, hep-th/9507088.
D. Perrot, Lett. Math. Phys. 50 (1999) 135, math-ph/9910044.
J.A. Harvey, Topology of the gauge group in noncommutative gauge theory, hep-th/0105242.
F. Ardalan, N. Sadooghi, Int. J. Mod. Phys. A 16 (2001) 3151, hep-th/0002143.
F. Ardalan, N. Sadooghi, Anomaly and nonplanar diagrams in noncommutative gauge theories, hepth/0009233.
C.P. Martin, Mod. Phys. Lett. A 16 (2001) 311, hep-th/0102066.
J. Gomis, T. Mehen, Nucl. Phys. B 591 (2000) 265, hep-th/0005129.
O. Aharony, J. Gomis, T. Mehen, JHEP 0009 (2000) 023, hep-th/0006236.
L. Alvarez-Gaume, J.L. Barbon, R. Zwicky, JHEP 0105 (2001) 057, hep-th/0103069.
A. Bassetto, L. Griguolo, G. Nardelli, F. Vian, JHEP 0107 (2001) 008, hep-th/0105257.
S. Terashima, Phys. Lett. B 482 (2000) 276, hep-th/0002119.
S. Minwalla, M. Van Raamsdonk, N. Seiberg, JHEP 0002 (2000) 020, hep-th/9912072.
C.P. Martin, The UV and IR origin of non-Abelian chiral gauge anomalies on noncommutative Minkowski
space-time, hep-th/0008126.
K. Intriligator, J. Kumar, -wars episode I: the phantom anomaly, hep-th/0107199.
M.M. Sheikh-Jabbari, Phys. Lett. B 455 (1999) 129, hep-th/9901080.
D. Bigatti, L. Susskind, Phys. Rev. D 62 (2000) 066004, hep-th/9908056.
L. Alvarez-Gaume, J.L. Barbon, Int. J. Mod. Phys. A 16 (2001) 1123, hep-th/0006209.
J. Nishimura, M.A. Vazquez-Mozo, JHEP 0108 (2001) 033, hep-th/0107110.
L. Bonora, M. Schnabl, M.M. Sheikh-Jabbari, A. Tomasiello, Nucl. Phys. B 589 (2000) 461, hepth/0006091.
B. Jurco, L. Moller, S. Schraml, P. Schupp, J. Wess, Eur. Phys. J. C 21 (2001) 383, hep-th/0104153.
L. Bonora, A. Sorin, Chiral anomalies in noncommutative YM theories, hep-th/0109204.

Morita duality and large-N limits

L. Alvarez-Gaum, J.L.F. Barbn 1
Theory Division, CERN, CH-1211 Geneva 23, Switzerland
Abstract
We study some dynamical aspects of gauge theories on noncommutative tori. We show that Morita
duality, combined with the hypothesis of analyticity as a function of the noncommutativity parameter
, gives information about singular large-N limits of ordinary U (N) gauge theories, where the largerank limit is correlated with the shrinking of a two-torus to zero size. We study some nonperturbative
tests of the smoothness hypothesis with respect to in theories with and without supersymmetry. In
the supersymmetric case this is done by adapting Wittens index to the present situation, and in the
nonsupersymmetric case by studying the dependence of energy levels on the instanton angle. We find
that regularizations which restore supersymmetry at high energies seem to preserve -smoothness
whereas nonsupersymmetric asymptotically free theories seem to violate it. As a final application we
use Morita duality to study a recent proposal of Susskind to use a noncommutative ChernSimons
gauge theory as an effective description of the fractional Hall effect. In particular we obtain an elegant
derivation of Wens topological order. 2002 Published by Elsevier Science B.V.
PACS: 03.70.+k; 03.65.Ca; 12.38.-t
1. Introduction
Noncommutative field theories (NCFTs) provide an interesting generalization of the
framework of local quantum field theory, allowing for some degree of nonlocality while
still keeping an interesting mathematical structure [13]. From a different point of view, the
perturbative dynamics of NCFT mimics in many respects that of string theory. To be more
specific, NCFT arises as a peculiar low-energy limit of open-string theory, [4,5] so that the
string becomes a rigid, extended object, i.e., a rigid dipole [6]. This is achieved by placing
the strings in a large magnetic field. The resulting low-energy theory lives effectively on a
E-mail addresses: luis.alvarez-gaume@cern.ch (L. Alvarez-Gaum), barbon@mail.cern.ch (J.L.F. Barbn).
1 On leave from Departamento de Fsica de Partculas da Universidade de Santiago de Compostela, Spain.

PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 2 4 - 1
166
L. Alvarez-Gaum, J.L.F. Barbn / Nuclear Physics B 623 (2002) 165200
noncommutative version of Rd , where the nonlocality is parametrized by a quantum phasespace structure of spacetime: [x , x ] = i , with related to the string magnetic field
B via = B 1 . The stringy nature of NCFT is not really hidden in the short-distance
structure, but apparent at low energies, since the size of the dipoles is approximately
linear in the momentum: eff = | p |, so that high energy particles are macroscopic in
size. This is the basis for the so-called UV/IR connection, [7] that represents the main
novelty of NCFT dynamics. In particular, if time enters the noncommutativity relations,
the particles are nonlocal in time, which poses a problem for Hamiltonian methods. In
fact, these theories appear to be inconsistent when defined as field theories with a finite
number of particle degrees of freedom [8,9]. We henceforth restrict the noncommutativity
of spacetime to the purely spatial sections.
Another stringy property of NCFT is the so-called Morita duality of gauge theories
on noncommutative tori [1,4,10] a low-energy remnant of the T-duality symmetry of
the underlying string model [11] (see also [12].) It acts on the periods of the string
magnetic field B/2 by fractional linear transformations. In particular, if the matrix
of periods has rational entries, there is a Morita transformation that maps the given
noncommutative theory to an ordinary U (N) gauge theory on a smaller torus, with a larger
rank and some nonvanishing magnetic fluxes. This opens up the interesting possibility
of using the information encoded in the -dependence of physical quantities to learn
about ordinary gauge theories at finite volume, i.e., the complicated structure of confining,
oblique-confining, Higgs and Coulomb phases of gauge theories could be encoded in the
noncommutative language in terms of some interesting modular properties as a function
of . Conversely, the exotic features of perturbation theory in NCFT, notably the UV/IR
phenomenon, put into question most of our standard expectations for nonperturbative
dynamics in NCFT. In this context, Morita duality can be used to translate the standard
body of knowledge about ordinary confining gauge theories into the context of their
noncommutative cousins. In particular, it was suggested in [13] that smooth behaviour
of physical quantities in would imply very nontrivial constraints on the large-N limit of
ordinary gauge theories on tori.
The interplay between noncommutativity and the large-N limit is at the root of the
subject in its (hidden) beginnings [14], since NCFT is a formal continuum limit of
the twisted reduced EguchiKawai models [15] (see also [16,17]). In this article we
sharpen this relationship with a number of qualitative and quantitative tests. We propose
an analyticity criterion on -dependence based on the limiting behaviour of rational
approximations to generic NCFTs. We show that this criterion of -smoothness gives
information about ordinary gauge theories in a singular large-N limit. Although this
singular limit is difficult to study in the language of ordinary gauge theory, we argue that
its dynamics can be very rich.
We also add some comments on the inverse problem, i.e., we study degenerate limits
of NCFTs which are tuned to reproduce the standard confinement regime of ordinary
gauge theories in the large-N limit. Morita duality implies that all these limits necessarily
involve infinite noncommutativity , a property already hinted at by the behaviour
of perturbation theory [7,15,18].
Some exact information about the singular large-rank limits is obtained by considering
rational approximations of N = 1 theories in three and four dimensions. We define an
167
appropriate Witten index that probes aspects of confinement dynamics in the non-Abelian
sector of the models, and show that this index varies smoothly with . In some cases this
behaviour depends on subtle properties such as oblique confinement in ordinary gauge
theories.
In Section 5 we go one step further by considering the dependence of energy levels
on the instanton angle in nonsupersymmetric theories. For a specific choice of electric
and magnetic fluxes, a reliable estimate is possible within an instanton-gas approximation,
since the corresponding energy splittings receive no contribution in perturbation theory.
This is a rather sensitive test of -smoothness, because of the occurrence of levelcrossing phenomena that depend explicitly on the rank of the gauge theory. We find
that a regularization with restored N = 4 supersymmetry at high energies preserves smoothness of the low-energy physics, whereas asymptotically free nonsupersymmetric
theories seem to violate the smoothness constraints. Hence, some form of supersymmetry,
albeit softly broken seems to be necessary to maintain continuity of the physics as a
function of .
We end with an application of Morita duality to the quantum Hall effect by calculating
Wens topological order [19] for the proposed noncommutative ChernSimons effective
description of the fractional Hall effect [20].
2. Morita equivalence of gauge theories

We consider U (N) noncommutative YangMills theories (NCYM) on S1 T3 , with S1
representing a compact Euclidean time direction of length , and T3 a noncommutative
three-torus with flat metric. The microscopic parameters of the U (N) gauge theory are
the YangMills coupling at some cutoff scale, g, the instanton angle, , and the spatial
noncommutativity parameters j k . There is also a background field ij that enters the
physics as a constant U (1) shift of the field strength, so that the action reads

i
1
tr(F + ) (F + ) + ,
S = 2 tr(F + ) (F + ) +
(2.1)
2g
8 2
where the dots stand for other terms depending on extra fields, such as fermions or scalars
in supersymmetric theories, all of them in the adjoint representation of the gauge group,
and F is the noncommutative gauge field strength.
The Feynman rules of this theory are obtained from those of the ordinary U (N) theory
by the following replacement of the U (N) structure constants:

f f cos 12 ij ki kj + d sin 12 ij ki kj .
(2.2)
k, k are any two momenta entering the trilinear vertex. In particular, we see that the global
U (1) is self-coupled and coupled to the SU(N) subgroup. Indeed, the rank-one U (1)
theory shows asymptotic freedom [21].
It is useful to define a dimensionless noncommutativity parameter and background field
by
ij =
2 ij
,
Li Lj
ij =
Li Lj
ij .
2
(2.3)
168
When considering the Hamiltonian quantization on T3 , the Hilbert space splits into
sectors labelled by the integer magnetic fluxes

F
,
tr
mij =
(2.4)
2
(ij )
determined by the first Chern class of the U (N) gauge bundle, together with the usual
integer momenta

Fj k
.
pj = iLj tr
(2.5)
2 Ak
T3
There are also integrally quantized electric fluxes wj of the form (cf. [22]):

i
tr
j k pk .
wj =
Lj
Aj
(2.6)
T3
We shall frequently use the vector notation for the magnetic fluxes and the background
field:
mk = 12 $ kij mij ,
k = 12 $ kij ij .
For simplicity, we consider orthogonal three-tori of the form T3 = T2 S1e with and
of rank-two and aligned with the noncommutative two-torus T2 , which will be assumed
squared with side length L. The length of S1e is Le , so that the volume of the three-torus is
V = Le L2 . In this simple case Morita duality is represented by the SL(2, Z) action:
b a
,
s + r
,
=
|s + r|
= (s + r)2 r(s + r),

L = |s + r|L,
(2.7)
on the parameters, where a, b, s, r Z and sb + ar = 1. We have collected the gauge

coupling and instanton angle into the complex coupling 2 = + 8 2 i/g 2 . The action
on the other quantum numbers is

s
r
m
p
s
r
p
m
=
,
=
,
(2.8)
a b
N
a b
w
N
w
where m is the magnetic flux through T2 and w, p are the two-dimensional vectors of
electric fluxes and momenta along the same torus. The notation w refers to the Hodge
duality operation on the T2 , i.e., (w)i = $ij wj .
In perturbation theory, Morita equivalence follows as an algebraic identity of the
Feynman diagram expansion [15], or as T-duality of the regularized version in terms of
an underlying string model [11]. More generally, there is a very explicit proof as a formal
change of variables in a path integral [16].
Rational theories are characterized by a rational dimensionless noncommutativity
parameter. If = a/b, so that s + r = 1/b, the Morita transformation (2.7) sends the
169
theory to an ordinary one with = 0 and different parameters, involving in particular

a rescaling of the rank of the gauge group. In this case, the details of the transformation
at the level of fields are rather elementary (cf. [13,23]). Consider, for example, a U (N)valued noncommutative connection on T2 Y , which is also periodic on the two-torus T2
of length L, and has arbitrary boundary conditions on the commutative space Y :

a (y)e2ix/L.
A(x, y) =
(2.9)
Z2
The coefficient matrices a satisfy a = a and have arbitrary dependence on the

commutative coordinates y Y .
The Morita-dual is a twisted U (N ) theory with N = Nb, with a connection given by
a particular superposition of matrices in the subgroup U (N) U (b):

A (x, y) + A (x, y) =
(2.10)
a (y) V a1 U 2 a1 2 /2 e2ix/bL ,
Z2
where e2i/b and the pair U and V are the standard clock and shift matrices of SU(b)
satisfying:
U V = V U.
(2.11)
The connection A is a constant-curvature Abelian gauge field in the diagonal U (1)

subgroup of U (N ) that satisfies = dA . For our particular choice of periodic
noncommutative connection A, the effect of A is to cancel the first Chern class induced
by A :

(2.12)
tr F + dA = tr F + = 0.
T2
T2
The traceless part of the ordinary connection A furnishes a twisted bundle of

SU(N )/ZN with t Hooft magnetic flux [m ] = rN (mod N ) and periodicity conditions

A xj + L = j A (xj )j .
(2.13)
The twist matrices may be chosen as
1 = 1 N U r ,
2 = 1 N V .
(2.14)
Notice that the same matrix Fourier components a appear in (2.9) and (2.10).
Therefore, when expressed in terms of the a , both the action and the integration measure in
the path integral remain formally invariant under the Morita transformation, which acquires
the simple interpretation of a change of variables in the position-space representation. For
our example of a periodic noncommutative two-torus we have the identity

i
1
2
tr|F
|
+
tr F F
2g 2
8 2
M
1
2g 2

M

2 i
trF + +
8 2

M

tr F + F + ,
(2.15)
170
with
M = T2 (L) Y,

M = T2 L Y,
and the couplings mapping according to the rules:

r
= b,
= .
g 2 N = g 2 N,
(2.16)
b
Similar manipulations can be performed for other fields in the adjoint representation of the
gauge group.
We assume that this Morita equivalence is a true isomorphism of the physical Hilbert
space of the theory that, in particular, preserves the exact finite-volume spectrum, including
nonperturbative dynamical scales related, for example, to confinement. In this case, the
proper interpretation of (2.7) is as a mapping of bare parameters at some cutoff scale UV
that remains fixed under the duality. A convenient regularization that respects the duality
at the short-distance scales is to consider a N = 4 SYM theory with appropriate mass
terms at the scale Ms = UV . In this case, the map (2.7) applies to the N = 4 microscopic
parameters.
Thus, barring possible Morita anomalies, (2.7) should hold nonperturbatively for the
continuum theory. In N = 4 NCYM theory, the explicit computation of the spectrum of
1
1
4 and 8 BPS states by a number of groups [11,24] gives a Morita-invariant result. In
particular, the energies of Abelian electric and magnetic fluxes are Morita-invariant. Based
on the evidence provided by these examples, in the rest of this paper we assume that Morita
equivalence is free from anomalies and holds as a true quantum symmetry of the continuum
theory.
3. Rational approximations and singular large-n limits

Given an infinite convergent sequence of rational numbers determining some noncommutativity parameter: n , we may define the noncommutative theory arising as a
limit of the Morita-dual ordinary theories, provided that this limit actually exists. The existence of this limit is a necessary condition for the smoothness of the physics as a function
of . We symbolically denote the resulting limiting theory by
lim U (N, n ) U (N) .
(3.1)
Writing n = an /bn , with (an , bn ) = 1 (i.e., relatively prime)2 and bn > 0, we determine
appropriate Morita transformations to ordinary theories by defining numbers rn , sn with
the property sn bn + rn an = 1 and performing the transformation (2.7) for each value of n:

rn
sn
Tn =
(3.2)
.
an bn
For given values of an and bn , we can fix the freedom in the choice of the pair rn , sn by
picking 0 rn < bn .
2 The symbol (a, b) for any two integers a, b represents their greatest common divisor.
171
After Morita duality, we obtain a series of U (Nn ) models with rank and magnetic flux:
Nn = bn N,
mn = rn N,
(3.3)
where we have assumed that the initial magnetic flux of all U (N, n ) models vanishes, a
condition that may be enforced by an appropriate Morita transformation.
Since bn in the limit, any such rational approximation is a large-rank limit.
However, since the length of the ordinary tori also shrinks by
Ln =
L
,
bn
(3.4)
we actually have a particular singular large-n limit, in which the large-rank limit is
correlated with the scaling of a two-torus to zero size. One can say that the dynamics
of the model U (N) resolves such singular limits of ordinary gauge theories. In principle,
it is not guaranteed that the model U (N) exists in all situations, in the sense of all gaugeinvariant operators having smooth matrix elements in the limit. The evidence from BPS
sectors in N = 4 NCYM suggests that such limit makes sense at least for sufficiently
supersymmetric theories.
There is evidence that -smoothness, if true, must involve rather nontrivial dynamical
phenomena. In Refs. [25,26], Morita duality was used as a tool to provide an optimum
set of quasilocal descriptions of the physics of N = 4 NCYM theories on a torus. A
quasilocal description is defined by requiring that the elementary degrees of freedom are
well-contained in the box, taking into account both the quantum size given by the Compton
wave-length and the classical size given by the dipolar extent of noncommutative
particles: the effective size scales with energy as (E)eff = max(1/E, E ). Demanding
(E)eff < L gives the definition of a given quasilocal patch 1/L < E < 2/L.
It is found in [26] that the structure of quasilocal patches depends sensitively on .
Not only it depends on the rational or irrational character of , but even on the degree of
irrationality according to some well-defined criteria. Although some non-BPS quantities
such as the entropy in the planar approximation behave smoothly as a function of , it
is much less clear that the generic non-BPS physical quantity (particularly at the level
of nonplanar corrections) will behave smoothly given the multifractal nature of the
renormalization-group flows.
Ignoring for the moment these caveats, we can entertain a strong form of the smoothness
hypothesis (see [13] and conjecture that, in appropriate situations, the U (N) model
is actually equivalent to some other ab initio definition of the U (N) theory. This
strong form of the conjecture severely constraints the large-n limit of the ordinary gauge
theories. Unfortunately, it concerns a singular large-n limit, and it is unclear what practical
information could be extracted for the regular large-n limit of t Hooft [27].
For irrational , we can use the U (N) theory as a tentative nonperturbative definition
of the model. In this case, the best rational approximations are given by continued
fractions. For rational , the U (N) is itself Morita-dual to some finite-rank gauge theory.
Therefore, in this case, the putative limiting theory can be rigorously defined, and the
equivalence of U (N) and U (N) imposes particularly strong constraints.
Perhaps the most radical statement of -analyticity along these lines would be the one
associated to the sequence n = 1/n 0 in a rank-one model. In this case, the series of
172
Morita-dual ordinary theories consists of U (n) models with one unit of magnetic flux on a
torus of volume Le L2 /n2 . The limit would be equivalent to an ordinary U (1) theory on a
torus of volume Le L2 , i.e., a free theory. If true, such an equivalence is rather surprising in
view of the UV/IR effects discovered in [7], which tend to render the theory nonanalytic
around = 0.
In the remainder of this paper we evaluate the case for the equivalence U (N)
U (N) from different points of view. Notably, we use an appropriately defined Witten
index as a useful criterion for N = 1 supersymmetric theories, and the dependence on the
instanton angle as a (less robust) criterion for nonsupersymmetric theories.
In the remainder of this section, we collect a number of observations on the qualitative
physics of the U (N) limiting models. As a final remark, if we try to define the
noncommutative gauge theories using a lattice formulation along the lines proposed in
[16], it seems that one would end up with a prescription similar to the limiting procedure
described in these sections. According to [16] in their lattice formulation one is forced to
have rational, thus if we are interested in studying a theory with an irrational value of
, as we take the continuum limit we should also describe a rational sequence depending
on the lattice spacing converging to the value of interest. Whether this is the only way to
define noncommutative theories on the lattice is an open question.
3.1. Cutoffs and dynamical scales
We can define two variants of the singular large-n limits, depending on whether the
ultraviolet scale UV remains fixed in the large-n limit, or it is scaled appropriately. If
UV remains fixed, eventually Ln UV < 1 and the small torus shrinks below the cutoff
scale. In this case we must interpret UV as a N = 4 supersymmetry breaking scale, Ms ,
so that the short-distance definition of the U (N) model refers to the N = 4 theory.
Alternatively, we can take the continuum limit UV for each n and define
the large-n limit of the series of continuum field theories. In this case, with N = 0, 1
supersymmetry, we have asymptotically free theories for which the microscopic coupling
parameter transmutes into a dynamical scale n . Since the value of each n is adjustable,
we can specify its large-n limit as part of our specification of the limiting procedure.
One possible uniform definition is given by the localization of the one-loop infrared Landau
pole in the planar approximation:

8 2
+ ,
n = UV exp
(3.5)
0 n (UV )
where the dots stand for higher (planar) loop contributions. In this formula, 0 is the
(positive) one-loop beta function coefficient with normalization
11 2nf ns
,
(3.6)
3
for a theory with nf Majorana fermions and ns complex scalars, all in the adjoint
representation of the gauge group, and n (UV ) = gn 2 (UV )Nn is the bare t Hooft
coupling of the U (Nn ) theory. Under Morita duality, the t Hooft coupling is invariant
gn 2 Nn = g 2 N , with g 2 (UV ) the bare coupling of the U (N) theory. In the limit, n
0 =
173
converges to , the standard dynamical scale of the U () gauge theory arising in

t Hoofts planar limit.
It is important to realize that, despite the two-torus of vanishing volume Ln2 0,
there is no dimensional reduction on the series of ordinary theories, i.e., the low-energy
physics is not trivialized into a (1 + 1)-dimensional renormalization-group flow. The
reason is that all of the ordinary gauge theories in the series have nonzero magnetic flux
through the vanishing torus, so that they support light electric-flux excitations down to
energies of order 1/Ln bn = 1/L, which remains fixed in the limit, and is the true infrared
threshold for the transition to (1 + 1)-dimensional physics. The interactions of these light
delocalized modes between the high scale bn /L and the low scale 1/L are governed
by a renormalization-group flow with beta function coefficient 0 [21] (see also [28]).
Therefore, asymptotically free theories with L 1 can develop strong coupling before
the reduction to (1 + 1)-dimensional dynamics takes place.
In summary, the physics of the U (N) models can be very rich, including possible
nonperturbative phenomena at intermediate scales . It must be emphasized that our
heuristic reasoning is based on the planar one-loop approximation to the renormalizationgroup flow of the YangMills coupling. In principle,
nonplanar diagrams can yield
important contributions at energy scales of order 1/ . For example, nonplanar oneloop diagrams in the noncompact theory turn the screening behaviour of the global U (1)
modes into antiscreening,
cf. [29,30]. In this case, one should further impose the hierarchy
L L/ 1.
3.2. Decoupled photons
Each of the rational U (N, n ) theories in the approximating series n has a
free ordinary Maxwell field effectively living on a torus of volume Vn = V /bn2 . This is a
general property of any rational noncommutative theory on a finite torus, as follows from
the structure of the vertices in (2.2). Since momenta are quantized as ki = 2ni /Li and
ij = ij Li Lj /2 , we obtain for the symmetric structure constants

d sin ni ij nj ,
(3.7)
so that modes in the diagonal U (1) subgroup of U (N) and with momenta given by integral
multiples of bn decouple. In the language of the ordinary U (Nn ) Morita-dual, the free
U (1) is simply the diagonal subgroup of U (Nn ). Moreover, since bn , the excitation
gap for these free photons diverges in the limit n , so that they are irrelevant for
the U (N) theory at finite volume. The decoupled photon modes at exceptional momenta
ki = 2i bn /Li are responsible for ultraviolet divergencies in nonplanar diagrams, unlike
the case of irrational theories in perturbation theory (cf. [31]). Notice, however, that the
infinite gap that is generated in the irrational limit for these free photons should render the
perturbation theory of U (N) equivalent to the standard irrational perturbation theory.
In the Hamiltonian formalism, the decoupled U (1) contributes an energy
EU (1) = E + Eflux ,
(3.8)
174
where E is the energy in photons and Eflux denotes the contribution of electric and
magnetic fluxes, which reads, in Morita-covariant form [11,24]:
Eflux = Ee + Em
2
2
g 2 L2i ij
1 L2i
=
pj + wi + 2
2mi + 2N i ,
4N
V
V
g N
i
(3.9)
where N = N + 12 ij mij is the dimension of the noncommutative module and g 2 stands

for the bare coupling at the cutoff scale UV (the N = 4 high-energy coupling for softly
broken models). This expression is Morita-invariant and depends smoothly on . However,
for asymptotically free models we take g 2 (UV ) 0 in the continuum limit, yielding a
divergent magnetic energy for mi + N i = 0. Thus, in many of the applications, we
shall assume that the U (N) model has mi + N i = 0. Since we can set mij = 0
by an appropriate Morita transformation, we can assume with no loss of generality that
i = mi = 0. For the series of ordinary U (Nn ) theories in (3.3), we have
mn + Nn n = 0.
(3.10)
conditions: g 2 (UV ) 0,
the contribution to the energy from

Notice that, under the same
Abelian electric fluxes becomes degenerate for all values of the electric flux.
3.3. Rational constructions of theories on R3
A variant of the limit discussed here can be introduced to provide a purely rational
definition of a noncommutative theory on R3 by a blow-up of T3 . Consider a sequence of
rational theories U (N, n ) with n = 1/n 0 on RT2 , where the noncommutative tori
a series of U (nN)
have size Ln = 2n . The commutative description involves
ordinary theories defined on shrinking two-tori of size Ln = 2/n and supporting

exactly N -units of magnetic flux. In terms of the series of ordinary theories, this limit
is even more singular
than the previous one, since the perturbative gap of electric fluxes
vanishes as O(1/ n).
This rational definition of the noncompact noncommutative models may be useful in
discussing the nonperturbative physics associated to various UV/IR phenomena [7,29,30,
3235].
3.4. Large N versus large
It is somewhat disappointing that the simplest criterion of -analyticity imposes
constraints on singular large-n limits, rather than the ordinary large-n limit of t Hooft [27].
Thus, an interesting variation of the limits constructed here would involve stabilizing the
size of the shrinking commutative torus by hand. We can simply consider a combination
of the n limit in the U (N, n ) models, together with a large-volume scaling of
the noncommutative torus, Ln . Since Ln = Ln /bn , we must define the sequence of
rational noncommutative theories on tori whose size, Ln , grows at least linearly with bn .
More specifically, consider a U (1) limiting model on growing tori of size Ln =
L |bn | with > 1, and L a fixed length scale. The Morita-dual tori have sizes
175
Ln = L |bn |1 , which grow in the limit. Hence, in this case we end up with exactly the
dynamics of an ordinary U () theory on S1e R2 , with a free photon and a tower of free
glueballs at the confining scale . This rather standard large-n spectrum is encoded in
the variables of the noncommutative theory in a very nonlocal way, since the noncommutativity scale diverges as n L2n n bn2 L2 . This is a simple way of using Morita
duality to argue that the ordinary large-n limit is captured by a limit [7,14,15,36].
Hence, we find that the -analyticity criterion, when forced to reproduce a standard
large-n limit, relates it to an extremely nonlocal description of physics. As an illustration
of the arbitrariness of these limits, let us consider the marginal case of the scaling above
with = 1. The resulting noncommutative geometry is superficially identical to that of
> 1, i.e., S1e R2 . However, the Morita-dual tori have fixed volume Ln = L , so that we
still have confining phenomenology on a large box, provided L 1, except that now
we also have a photon with a discrete spectrum of gap 1/L . This is rather exotic when
viewed from the point of view of the limiting noncommutative theory, since the length
scale L is not directly associated to the limiting geometry. Thus, we learn that what could
be called the U (1) theory on R2 is not unique, having at least a one-parameter family
of theories labelled by the hidden scale ratio L .
In fact, it is likely that this marginal limit model with = 1 is not well-defined in the
absence of supersymmetry. Although the energy scale of the finite box 1/L is largely
irrelevant to the dynamics of the U () glueballs provided L 1, the (exponentially
small) finite-size effects in this regime can be quantified by t Hoofts dual confinement
criterion in terms of screening of magnetic-flux energy. Under very general assumptions,
the non-Abelian contribution to the magnetic-flux energy on a large box is given by (cf.
[37]):

Le
E mn SU(N ) = Cn 2 exp L 2 ,
n
(3.11)
2
where 1
is the effective thickness of the large-n confining string and is
the corresponding string tension. The numerical constant is

Cn = 1 cos 2mn /Nn = 1 cos(2rn /bn ).
(3.12)
This quantity is well-defined in the n limit provided the trigonometric constant Cn

has a limit. In general, this is not the case for arbitrary sequences an /bn . One can easily
find examples of approximations to irrational numbers, for which the ratio rn /bn behaves
erratically as n . In these cases, there are exponentially small quantities that are not
well-defined in this limit. This conclusion can be extended to any physical quantity whose
commutative evaluation depends explicitly on the ratio mn /Nn (see also [38]).
4. The Witten index
Having established that rational approximations of irrational theories, denoted U (N) ,
serve as a blow-up of a specific singular large-n limit of ordinary theories, we would
like to study this limit in more detail. In supersymmetric theories, a significant amount of
nonperturbative information is obtained by looking at the subset of BPS-saturated states.
176
The most primitive BPS quantity is perhaps the supersymmetric (Witten) index, Tr(1)F ,
that counts the number of Bose minus Fermi vacuum states [39]. This index gives valuable
information, not only about dynamical supersymmetry breaking [39], but also about the
dynamics of confinement and the breaking of various global symmetries [40,41]. For
example, for N = 1 pure YangMills theories in four dimensions with gauge group SU(N)
one has I = Tr(1)F = N , reflecting the spontaneous breaking of the Z2N R-symmetry
acting on the gluinos, ei/N , to the Z2 subgroup of 2 -rotations: . One can
also define refinements of the index that depend on electric and magnetic fluxes through a
torus, and probe the standard hypothesis about confinement dynamics, i.e., the confinement
of electric flux, and the screening of magnetic flux [37].
In the large-N limit, the number of confining vacua is infinite. Naively, this reflects
the restoration of the classical U (1)R symmetry. Since the classical U (1)R symmetry was
broken to Z2N by instantons, the restoration of the continuous symmetry is compatible
with the generaland sometimes naiveidea that instanton effects turn-off in the large-N
limit. Here too the situation is subtle, since domain walls separating adjacent vacua still
have divergent O(N) tension (cf. [42]).
We would like to test the hypothesis of -analyticity in minimal supersymmetric
theories by computing the Witten index of the limiting model U (N) , defined as a limit
over the rational series of N = 1 theories
IU (N) = lim IU (N,n ) ,
n
(4.1)
where IU (N,n ) is defined in terms of the index of the ordinary U (Nn ) Morita-dual.
Incidentally, although this quantity is an integer, if we do not have continuity in there
is no reason why the index should not jump in an uncontrollable way as we move along
the sequence n approximating . This makes the computation of the index meaningful
and less trivial than it might naively seem at first sight. Before applying these results to
our problem we must sort out some technical issues related to the proper treatment of the
diagonal U (1) subgroup of each ordinary U (Nn ) theory.
4.1. A supersymmetric index for U (N) theories
In U (N) gauge theories, the naive definition of the index as Tr(1)F , in terms of a
trace over the full Hilbert space of the U (N) theory, gives a trivial vanishing result from
the contribution of the ground states
in the diagonal U (1) sector, obtained by the action
the photino zero-mode operators, tr , on the vacuum. In fact, the vanishing of the index
for a theory with a U (1) factor is associated to the possibility of adding a FayetIliopoulos
term that breaks supersymmetry [39].
The usual remedy in the study of N = 1 U (1) theories is to consider instead the
refined index Tr C(1)F , where C is the charge-conjugation operator. Such an index can
be defined precisely when the FayetIliopoulos term is zero, and is nonvanishing for the
free U (1) theory. This solution is not directly applicable in our case because, although
the charge conjugation is a symmetry of the U (Nn ) theories, it acts nontrivially on the
magnetic fluxes mn mn . Thus, the corresponding index is again trivially vanishing for
a generic value of the magnetic flux mn .
177
On the other hand, the physical excitations of the decoupled diagonal U (1) multiplet
are rather uninteresting and even become infinitely massive in the n limit. Roughly
speaking, the Hilbert space of a U (N) theory factorizes between U (1) excitations and
SU(N) excitations, the latter being the interesting ones in our singular large-n limit.
Therefore, it should be possible to find a refinement of the standard index so that it does
not vanish and gives information about the degeneracy of ground-states in the interacting
(non-Abelian) sector.
The details of the factorization of the diagonal U (1) subgroup are complicated by the
global structure of the gauge group: U (N) = (U (1) SU(N))/ZN . This has two main
consequences for the structure of the canonical quantization on a torus. First, a given U (N)
bundle over T3 , with fixed integer-valued first Chern classes mij Z, can be seen as a
particular U (1) (SU(N)/ZN ) bundle with quantized magnetic fluxes m on the U (1) and
t Hooft flux [m] m(mod N) on the adjoint factor SU(N)/ZN .
A second, related subtlety is that the factorization of gauge transformations into a
diagonal U (1) part and an SU(N) part is ambiguous by elements of the center ZN of
SU(N):
U = ei U = ei z z1 U ,
(4.2)
where zN = 1 and U SU(N). Since ZN is a discrete set, it follows by continuity that

this ambiguity only affects the periodicity conditions. In particular, a strictly periodic U (N)
transformation can be factorized into nonperiodic factors satisfying:
U (x + Lj ) = zj1 j U (x)j ,
ei(x+Lj ) = zj ei(x),
(4.3)
where we have considered arbitrary twisted boundary conditions on T3 . The entangled

action of ZN means that we cannot simply factorize the Hilbert space into U (1) and
SU(N) parts, even at a fixed value of the magnetic flux. Rather, we should quantize the
theory without explicitly dividing by the gauge transformations (4.3), and then impose the
ZN -invariance at the end by an averaging procedure. This is facilitated by the fact that (4.3)
act as global symmetries on the Hilbert space obtained by satisfying the Gauss law within
the space of zj = 1 gauge transformations.
The characters of the ZN action on the Hilbert space of the SU(N)/ZN gauge theory
define the t Hooft electric fluxes: [w], a three-vector of integers modulo N . If we construct
the SU(N)/ZN Hilbert space in terms of gauge-invariant Wilson-loop operators, each
component of t Hooft electric flux corresponds to the action of a Wilson line, W ([wj ]),
wrapping the j th torus direction and carrying an irreducible representation with N -ality
equal to [wj ]. We may now add the diagonal U (1) gauge field (tr A)/N and build the
general operator:

j

1

j
W w exp iw
(4.4)
tr A .
N
j
The integer wj is interpreted as the Abelian electric flux in the j th direction. Invariance
under (4.3) imposes the expected constraint between Abelian and non-Abelian electric
178
fluxes:

wj = wj
modulo N.
(4.5)
Hence, we learn that the complete Hilbert space can be constructed by a decomposition
with respect to value of the total electric flux:

H(w, m)U (N) = H(w, m)U (1) H [w], [m] SU(N)/Z ,
(4.6)
N
which induces a general decomposition of the U (N) index:

I (m)U (N) =
I (w, m)U (1) I [w], [m] SU(N)/Z .
(4.7)
wZ
Thus, the vanishing of the index is a trivial consequence of the vanishing of all the U (1)
factors, even for vanishing electric flux. In order to include the effects of the instanton
angle in more detail, it will be instructive to rederive (4.7) using a path integral argument.
Actually, we can be slightly more general with no extra complications. Since the
covering group of U (N) is noncompact, we can define a continuous electric flux valued
on a torus, labelling representations of the covering group, R SU(N), that are not
representations of U (N). Intuitively, it corresponds to inserting unquantized U (1) probe
charges into the system, and can be incorporated by adding a topological theta-term to
the action of the form

i j
2i j
e
tr F0j =
e m0j .
Se =
(4.8)
N
N
j
(0j )
With this normalization, ej are real numbers defined modulo N . The vector of integers
m0j kj combines with the magnetic fluxes mk = 12 $ kij mij to specify the first Chern
classes of the U (N) bundle over the Euclidean four-torus. In addition, there is an integervalued second Chern class that is canonically dual to the standard instanton angle, entering
the action as

i
tr(F + ) (F + ).
S =
(4.9)
8 2
In this formula, the constant U (1) background field ij is added for notational convenience
in the applications involving Morita duality. The extra terms amount to a constant term and
a redefinition of the electric flux:
(m + N).
e e = e +
(4.10)
2
The projected index at a given value of the instanton angle can be given a path-integral
interpretation as follows.
I (e, m, )U (N) = Tr P(e,m,) (1)F eH

=N
eSe S Z(k, m, )U (N) ,
(4.11)
kZ3 Z
where N is a normalization constant and Z(k, m, )U (N) is the path integral of the U (N)
theory (with the YangMills action) on a given topological sector of T4 bundles specified
179
by magnetic fluxes mij = $ij k mk and m0i = ki , in addition to the integral Pontriagin
number Z.
On a fixed bundle topology, the partition function factorizes between U (1) and SU(N)
parts:

Z(k, m, )U (N) = Z(k, m)U (1) Z [k], [m], SU(N)/Z ,
(4.12)
N
where the non-Abelian partition function depends on k, m only through their mod N
reductions [k], [m]. The Abelian partition function can be further factorized, according to
(3.8), into the contribution of the free photons (the Maxwell term) and that of topological
fluxes:

4 (V )i
Z(k, m)U (1) = ZMaxwell eEm eke k , e diag 2
(4.13)
,
g N Li
with (V )i denoting the spatial volume orthogonal to the ith direction, and Em the
magnetic energy in (3.9).
The -dependence of the action S is conveniently factorized into Abelian and nonAbelian contributions by the decomposition of the U (N) curvature:

1
1
F = tr F + F tr F .
(4.14)
N
N
The first term yields a contribution

i
exp k (m + N) ,
N
(4.15)
to the partition function, whereas the non-Abelian contribution is

[k] [m]
exp i +
,
N
(4.16)
with Z. The fractional contribution to the instanton charge

in (4.16) cancels the
analogous term in (4.15), so that the total second Chern class tr(F F )/8 2 of the
U (N) bundle is an integer.
Let us separate explicitly the -dependence induced by integral instanton numbers by
defining the function:

ei Z [k], [m], SU(N)/Z .
Z [k], [m], SU(N)/Z =
(4.17)
N
Then, putting all factors together we have the following expression for the index:
I (e, m, )U (N)
= N ZMaxwell

k Z3
N

e[k]e [k]
2 i
N [k](e +[m] 2
[k]=1

e2ik e 2Ne [k]k eN

Z [k], [m], SU(N)/Z
2 k k
e
(4.18)
180
where we have split the sum over k as k = [k] + Nk . We find, upon Poisson resummation
in the integers k :
I (e, m, )U (N)
(e +w) 1 (e +w)

N 2 e
= ZMaxwell
e
I [w] , [m], SU(N)/Z ,
N
(4.19)
wZ3
where
.
2
[w] [w] + [m]
(4.20)
In writing down (4.19), we have adjusted the normalization constant so that N 2 = det(e ),
and we have introduced the standard index of the SU(N)/ZN theory as a function of the
t Hooft electric and magnetic fluxes:
N

1 2i[k][w]/N
I [w], [m], SU(N)/Z = 3
e
Z [k], [m], SU(N)/Z .
N
N
N
(4.21)
[k]=1
Notice the appearance of the integrally quantized electric fluxes w, whose reduction
modulo N defines the t Hooft electric flux of the SU(N) sector. The effective electric
fluxes (4.10) and (4.20) incorporate the fact, discovered in [43], that magnetic fluxes induce
an electric charge in the presence of an instanton angle.
The exponential term in (4.19) can be recognized as the contribution of the Abelian
electric fluxes to the energy, i.e., for an Abelian electric flux of quantum i we have:
E()electric =
i 2
g 2 L2i i 2
.
=
2
4N
V
N (e )ii
i
(4.22)
Notice that these energies become all degenerate in the continuum limit, which involves
g 2 N 0. However, the electric fluxes contribute a strictly positive energy in the
regularized theory. We define the index as a limit from the regularized theory, so that only
states with effective Abelian flux = 0 correspond to vacua.
A final compact expression for the U (N) index is given as a convolution of Abelian and
non-Abelian ones that generalizes the formula (4.7):

I (e + w, m)U (1) I [w] , [m], SU(N)/Z .
I (e, m, )U (N) =
(4.23)
N
wZ3
Thus, the vanishing of the index is due to the trivial vanishing of the pure U (1) terms, even
for zero effective electric flux e + w = 0. This suggests a natural definition of a purely
non-Abelian index that measures the degeneracy of ground states in the non-Abelian
sector. We just formally divide by the purely Abelian index at zero effective electric flux:
I (e, m, )U (N)
1
Tr P(e,m,) P(=0)(1)F eH ,
I ( = 0, m)U (1)
(4.24)
where = e + w. This formal operation cancels the trivial zero due to the contribution of
photino zero-modes.
181
In the case that we set m + N = 0 in the initial series of U (N, n ) models, i.e.,
we cancel the magnetic vacuum energy, we can give a more elegant definition of the
refined index by twisting the purely Abelian degrees of freedom by the charge-conjugation
operator CU (1) . We can specify the action of CU (1) exactly due to the factorization (4.6):
I (e, m, )U (N) = 14 Tr CU (1)P(e,m,) (1)F eH .
(4.25)
Since the Abelian charge-conjugation operator inverts the sign of the effective Abelian
electric flux , only the U (1) ground states with = 0 contribute, and we have a
sensible index measuring the degeneracy of non-Abelian degrees of freedom.
We are now ready to use the results of [40,41]. First, notice that for pure N = 1 SU(N)
SYM theories in four dimensions, the classical U (1)R symmetry is anomalous, so that we
can absorb the angle into a phase redefinition of the gluino field. This means that the
physics will be actually -independent. On the other hand, the symmetry + 2
is realized by a permutation of vacua. The consequence is that, in a sector with both
electric and magnetic fluxes, electric fluxes differing by a redefinition [w] [w] + [m]
give equivalent physics, since this transformation is generated by the 2 -shift in the
instanton angle. This effect, associated to t Hoofts mechanism of oblique confinement
[44], partitions the N supersymmetric vacua of the SU(N) theory into N/cm sets, each one
with degeneracy cm , where

cm max mi , N , i = 1, 2, 3.
(4.26)
Thus, the result for the SU(N) index is

I [w], [m], SU(N)/Z = cm , for [w] = 0 modulo [m]Z,
N
(4.27)
and it vanishes otherwise.

The Abelian index I (e + w, m)U (1) vanishes unless the effective electric-flux energy
is zero, which requires e = w or, equivalently, e must be an integer. Combining the
two selection rules with the convolution (4.23) we find the result for our refined U (N)
index:
I (e, m, )U (N) = cm ,
for e = 0 modulo [m]Z,
(4.28)
and it vanishes otherwise.

In particular, (4.28) implies that the non-Abelian index for a U (N) theory with periodic
boundary conditions m = 0 is concentrated at zero electric flux e = 0 and has value N .
Thus, it diverges in the large-N limit.
4.2. The supersymmetric index of the U (N) model
We are ready to apply these results to our problem of determining the number
of nontrivial supersymmetric vacua of the U (N) limiting theory. Since the U (N)
model was defined with periodic boundary conditions, a similar behaviour to that of the
commutative counterpart would suggest an infinite index at zero electric flux, given the
fact that we define the model via a large-n limit. However, this naive expectation is not
182
borne out by the explicit calculation. In our limit definition:

I (e)U (N) = lim I e, mn U (N ) ,
n
(4.29)
we use the electric flux e of the ordinary theories U (Nn ) to label the states. Note that,
for the components parallel to the noncommutative directions, this is Morita-dual to
fixing a projector over some combination of momenta and electric fluxes in the series
of noncommutative theories. In particular, a fixed value of the integer electric flux in the
ordinary U (Nn ) theories maps under the duality to the combination:
wn = bn (w + n p),
(4.30)
T2n .
where p and w are momenta and electric flux on the noncommutative

Since we find
more transparent the language of the ordinary unitary theories, we choose to parametrize
the index in terms of the U (Nn ) electric flux e. Notice also that, since Nn , this flux
is asymptotically defined as an arbitrary vector in R3 in the large-n limit.
The important point is that (mn , Nn ) = N is bounded and constant in the limit. Hence,
the result (4.28) translates into
I (e)U (N) = N,
for e = 0 modulo (0, 0, N)Z.
(4.31)
Notice that the definition of the index with the extra phase depending on the rank of the

gauge group, as in [41], would yield an extra factor of (1)Nn 1 , rendering the index
erratic in the large-n limit.
Thus, the refinement of the index by the electric flux is smooth. The index has been
fractionalized due to the physics of oblique confinement in the series of ordinary theories.
The refinement by the electric flux is essential in obtaining a smooth answer. Since the
index is computed in a BornOppenheimer approximation, the result is determined by the
structure of the space of classical ground states, which in turn depends sensitively on the
rational value of . For a periodic U (N) model with = a/b, the corresponding moduli
space has b connected components, each one of dimension N . This is clear in the Moritadual picture of the gauge bundle, where we have (for our particular choice of background
field), a commutative U (1) SU(N )/ZN bundle with N = Nb and Nr (mod N ) units
of t Hooft flux through T2 . The flat connections are characterized by holonomies in the
non-Abelian factor U1 , U2 , Ue satisfying
U1 U2 = U2 U1 e2ir/b ,
(4.32)
Ue Uj = Uj Ue ,
(4.33)
and
for j = 1, 2.
The solution has b components

Uj = (Hj 1b ) j ,
Ue = He e2iq/b 1b ,
(4.34)
where Hj , He are in the Cartan torus of SU(N) and q = 0, 1, . . . , b 1. The different

components correspond to the different sectors of t Hoofts electric flux that can be related
by tunneling via fractional instantons. This explains why the index is only smooth in the
large-b limit once it is refined at a fixed value of the electric flux. The relation between
183
the fractional instantons and the structure of the space of flat connections suggests that the
contribution of fractional instantons, if nonvanishing, tends to work against continuity in
. In the next section we present a quantitative test of this idea.
5. Instanton-induced -dependence
Our treatment of the index in U (N) theories can be formally generalized to other
physical quantities. For example, if we consider the antiperiodic spin structure for
the gauginos, we are computing the finite-temperature partition function. This partition
function admits a similar factorization on electric-flux labels:

eF (e+w)U(1) eF ([w] ,[m],)SU(N) ,
eF (e,m,)U(N) =
(5.1)
wZ3
where now we have a nontrivial -dependence due to the breakdown of supersymmetry.

It is often the case that, particularly when the instanton-gas approximation suffers from
infrared problems, the -dependence of physical quantities is apparently invariant under
+ 2N only. This is especially obvious when considering a soft breaking of N = 1
supersymmetry by the addition of a small gluino mass. This has the effect of splitting the
vacuum energies of the N vacua, that are not equivalent any more. Since + 2
shifts the vacua one by one, we need N steps to return to the same vacuum.
For SU(N)/ZN theories in finite volume, the fractional nature of the instanton number
means that, in principle, physics is only periodic in modulo 2N . However, for the case
of electric-flux energies, the 2 -periodicity of the instanton angle is restored by a nontrivial
level-crossing [45]. The fractional -dependence implicit in the redefinition of SU(N)/ZN
electric fluxes:
[w] = [w] + [m] ,

(5.2)
2
gives the appropriate spectral flow [w] [w] + [m].
Similar behaviour should be found in the U (N ) theories that appear as Morita duals
of rational noncommutative theories, provided the Abelian contribution to -dependence
cancels out. The condition for this is the same as the vanishing of the magnetic ground-state
energy: m + N = 0.
Under Morita duality, the instanton angle is fractionalized as
= b.
(5.3)
Thus, 2 -periodicity in the U (N, n ) model translates into naive 2bn -periodicity in the
U (Nn ) model. If the physics is -dependent, such as in the absence of supersymmetry,
we may have some level-crossing phenomena that recover the 2 -periodicity in n . As
a result, the U (N, n ) model would actually show a hidden 2/bn -periodicity in the
spectrum as a function of . Thus, level-crossing phenomena of order bn could be present
in rational theories with n = an /bn . In the large-n limit we have bn , so that the
-dependence of the U (N) model would be either trivial or infinitely discontinuous. In
both cases, we would violate the strongest form of the -smoothness conjecture, since the
-dependence for rational must be nontrivial and smooth.
184
We can test this scenario with an explicit example. Let us consider the nonperturbative
splittings of the electric-flux energies in the nonsupersymmetric theory, calculated using a
dilute gas of fractional instantons (see [4547]).
We consider the simplest case of a rank-one rational noncommutative theory on a
periodic torus S1 (Le ) T2 (L), with = a/b. The ordinary dual U (N ) theory lives in
S1 (Le ) T2 (L ), with N = b, L = L/b. We have m + N = 0 so that the diagonal
U (1) has effectively zero magnetic flux. The SU(N )/ZN bundle has t Hooft flux

m = 0, 0, me = (0, 0, r),
where sb + ra = 1 and 0 r < b. Then, the non-Abelian contribution to the energies
of electric fluxes of the form [w] = (0, 0, [we ]) is exactly degenerate to all orders in
perturbation theory. The leading effect lifting this degeneracy is the tunneling contribution
by fractional instantons of charge proportional to 1/N . We can estimate this splitting by a
standard computation in the dilute-gas approximation.
Let us consider the sum over all chains of k+ fractional instantons and k antiinstantons
of minimal charge along the Euclidean time direction, in the limit /L . Using
ar + bs = 1, we can write the total instanton charge in each term in the sum as
k+ k (k+ k )a[me ]
=
+ (k+ k )s.
N
N
Using the standard parametrization of the topological charge
Q=
Q=
[k ] [m ]
+ ,
N
(5.4)
(5.5)
with Z, we can identify [(k+ k )a] = [ke ] as the conjugate variable to the electric
flux [we ]. In particular, in computing the discrete Fourier transform

eE([w ],[m ], ) =
N

[k ]=1
e2i[k ][m ]/N

ei Q Z k , m , ,
(5.6)
within the instanton-gas approximation, we can replace the sum over [ke ] by a free sum
over integers (k+ k ). Assuming the usual factorization of the instanton measure in the
dilute instanton-gas approximation, the sum over k+ and k exponetiates and we find for
the instanton-induced splitting:

1

2a
w
+
E we , me , = 2K cos
(5.7)
.
e
N
Going back to the variables of the original noncommutative U (1) theory, we notice that
[we ] is Morita-invariant in our particular case, and we have

E [we ], = 2K cos 2[we ] + .
(5.8)
The numerical prefactor K is interesting. It takes the form:

4

2
2
4
Scl DetF
Scl e
K Le L N (UV )
Det + .
B
(5.9)
185
The first term in this expression gives the contribution of the four translational zero modes
of the fractional instanton. There is a degeneracy factor N 2 coming from the various
independent tunnelings for a given topological charge.3 The terms containing the classical
action
Scl =
8 2
,
g 2 (UV )N
include the usual Jacobian Scl for each collective coordinate. A factor (UV )4 comes
from the zero-modes in the regularization of the one-loop determinants, whose nonzeromode contribution for bosons and fermions is also indicated. The dots in (5.9) stand for
higher-loop contributions.
Let us consider UV = Ms as the scale of soft breaking of N = 4 supersymmetry down
to N = 0. In terms of this fixed scale, we can rewrite the numerical prefactor in U (1)
variables as

2
DetF
8 2
8 2 /g 2 (Ms )
+ .

K Le L2 (Ms )4
(5.10)
Det
g 2 (Ms )
B
Notice that the factor of (L N )2 in (5.9) combines into a single factor of L2 , as corresponds
to a single position collective coordinate for the noncommutative U (1) instanton (cf. [49])
that appears as a Morita-dual of the ordinary fractional instanton.
The important property of (5.8) and (5.10) is that the size of the instanton-induced dependence is nonperturbative in the N = 4 coupling g 2 (Ms ). This is to be compared to
the splittings induced by the energies of the Abelian fluxes (3.9):
g 2 (Ms ) Le
(5.11)
(we )2 .
4 L2
Clearly, within the conditions of applicability of the instanton expansion, the non-Abelian
contribution is completely negligible in comparison to the Abelian one. As a result, there
are no possible level-crossing phenomena induced by (5.8), and energy levels are clearly
continuous under rational approximations of any given .
The qualitative picture changes considerably if we decouple completely the N =
4 regularization. Namely, let us remove completely the UV cutoff scale by proper
renormalization in the g 2 (UV ) 0 limit. In this limit, the Abelian splittings (5.11)
vanish, and we are left with the non-Abelian contributions (5.8). Therefore, level-crossing
becomes possible.
In order to correctly renormalize (5.9) we must identify the effective infrared cutoff of
the one-loop determinants. This is given by the size of the noncommutative
U (1) instanton
2 . In the limit that L , L, this size is of

L
in the
noncommutative
box
of
volume
L
e
e
O( ), since the instantons must be a smooth deformation of the instantons at = 0. In

this case, the instanton at = 0 is point-like, and the only available scale for the resolution
is given by .
E(we )Abelian =
3 There are N 2 inequivalent instanton solutions on R T3 that tunnel between fixed pairs of vacua on T3 .
They are obtained by the action of discrete translations in the plane of twisted boundary conditions (cf. [4648]).
186
Therefore, provided the theory is still perturbative at the energy scale M = 1/ , i.e.,
g 2 (M ) 1, we can eliminate UV in favour of and the dynamical scale
= UV e8
2 /
0g
2 (
UV )
with 0 given in (3.6). We obtain

0 2
2
4
K = Le L (M )
f g (M ) ,
M
(5.12)
where we have summarized in the function f (g 2 ) the renormalized perturbative expansion

in the instanton background.
Consider now a rational approximation of some generic noncommutativity parameter:
an /bn = n . The induced splittings of electric fluxes are given by

E [we ], n = 2Kn cos 2[we ]n + .
(5.13)
In the large-n limit, Kn K with

0 2
K = Le L2 (M )4
f g (M ) ,
M
and each individual curve for fixed [we ] converges to the limiting curve

2K cos 2[we ] + .
However, the different curves obtained by varying [we ] cross one another as a function of
. Therefore, the ground-state energy as a function of is defined by the minimum:

2Kn cos 2[we ]n + .

min
E()gs = lim
(5.14)
n 0[we ]<bn
If is irrational, the minima of the cosine function are uniformly distributed in the limit.
Therefore, in this case, the ground-state energy is a constant, independent of . In general,
let us write n = + n and [we ] = xbn . In the limit, x becomes a continuous variable
in the unit interval and we have,

E()gs 2K min cos 2[we ] + + 2xbn n .
(5.15)
0x1
For rational = a/b, the direct computation of the resulting ground-state energy yields

a
E()gs min cos 2[we ] + .
(5.16)
0[we ]<b
b
Hence, the condition for (5.15) to approach (5.16) is that the combination

an
bn n = bn
0,
bn
in the large-n limit. However, this is true if and only if the limiting value is an irrational
number. We conclude that, for rational , the term 2xbn n shifts the argument of the
187
cosine function in (5.15) by an arbitrary amount in the large-n limit, so that the minimum
is given by E()gs = 2K for any .
Approximations by rationals give a constant ground-state energy, in disagreement with
the direct evaluation (5.16). This implies that the strongest possible conjecture of analyticity over the rationals fails this test. It is very interesting that the -discontinuity
that we have discussed involves a complete decoupling of the supersymmetric energy scale,
i.e., Ms faster than the large-n limit.
6. Models in 2 + 1 dimensions
Many of the previous results can be extended to the case of rational approximations of
minimal supersymmetric models in 2 + 1 dimensions. In this case we borrow the results
of Ref. [40] for the case of SU(N) gauge group. The index is a function of the topological
ChernSimons coupling k, governing the supersymmetric ChernSimons mass term

k
2
CS(A)SU(N) =
(6.1)
tr A dA + A A A + .
4
3
This quantity satisfies a topological quantization condition k Z where k k N/2, the
shift by N/2 being a consequence of the contribution of the fermionic measure. The result
of [40] for SU(N)k with k > 0 and periodic boundary conditions is
I (k, N)SU(N) =
(N + k 1)!
.
k!(N
1)!
(6.2)
We also have I (k, N) = (1)N1 I (k, N) and the index vanishes for |k| < N/2.
For the group SU(N)/ZN , i.e., allowing t Hoofts magnetic fluxes [m] ZN the
quantization condition is instead
N
N
modulo
Z,
(6.3)
2
(N, [m])
due to the fractionalization of the instanton number. In particular, for such nontrivial
gauge bundles the ChernSimons action must be defined in terms of a four-dimensional
topological action. Extending the bundle to an appropriate four-manifold X whose
boundary is the desired three-manifold M3 = X, we define the ChernSimons term as:

k
.
tr F F + d
CS(A)M3 =
(6.4)
4
k=
For example, if M3 = S1 T2 , we can take X = D T2 , where D is a two-dimensional

disk over which the bundle extends trivially. The fermionic term in (6.4) does not need to
be extended to the interior of X because the action is already well-defined on its boundary.
6.1. Morita duality
The definition (6.4) is suitable to the study of Morita duality in the noncommutative
case, since the Morita transformation of the right hand side is the same as that of the
188
YangMills action (after properly redefining the field strength F F + ). Working on

M = S1 T2 (L), with periodic boundary conditions on the noncommutative torus, under
the Morita duality (2.7) we have:
4
NCCS(A)M
k

2
=
tr A dA + A K A K A +
3
M
= (s + r)

tr F + F + + d ,
(6.5)
where X = D T2 (L ) and the notation A K A means A K A dx dx . In the

particular case of interest to us, we have s + r = 1/b and L = L/b. Thus, we learn
that the T-duality mapping of the topological ChernSimons coupling is
k k = k(s + r)1 = kb.
(6.6)
Notice that the rescaling by a factor of b agrees with the quantization condition in
the ordinary U (N ) theory, in the presence of t Hooft fluxes, since the ChernSimons
coupling of the (periodic) noncommutative theory, k, is quantized modulo integers [50].
Our normalization conventions for the ChernSimons action, together with the general
action of Morita duality for rational , gives the general Morita-covariant quantization
condition in sectors with arbitrary magnetic flux m,
k=
N
2
modulo
N
Z,
(N, m)
(6.7)
where N = N m is the dimension of the noncommutative module. This quantization

condition should extend to general irrational by analytic continuation.4 This is a
reasonable expectation since all we use to obtain (6.7) are classical properties of the
classical action which is analytic in .
In particular, for our rational sequence of T-dualities between U (N, n ) and U (Nn ),
we have
kn = bn k,
(6.8)
where k is the fixed topological coupling in the noncommutative U (N, n ) theories, which
acquires the quantization condition:
k=
N
2
modulo Z.
(6.9)
It is interesting to notice that the shift by 1/2 in the rank-one case can be understood
directly by integrating-out the noncommutative Majorana fermions [52].
4 The quantization condition (6.7) should agree with that in Ref. [51], given the proper notational conventions.
189
6.2. Level normalization and the BornOppenheimer approximation

In the microscopic evaluation of the index one considers the effective dynamics of the
gauge and fermion zero-modes [3941]. Unlike the four-dimensional case, where the index
was independent of the value of couplings in the Lagrangian, in 2 + 1 dimensions the index
depends on the ChernSimons level. Therefore, its absolute normalization is essential in
the computation.
A naive zero-mode reduction of the noncommutative ChernSimons Lagrangian on
the noncommutative torus with periodic boundary conditions yields the effective Born
Oppenheimer Lagrangian (we only consider the bosons in this discussion)

2
k
k
tr A dA + A K A K A
dt $ ij Ci t Cj ,
(6.10)
4
3
4
RT2
where

Cj tr H (a0 )j L ,
(6.11)
is defined in terms of the constant component a0 of the noncommutative gauge field in

the Fourier expansion (2.9). The N matrices H are a convenient basis of the Cartan
subalgebra of U (N), normalized as tr(H H ) = .
After the Morita transformation, in terms of the twisted ordinary U (N ) gauge theory,
the flat connections are parametrized simply by those in the non-Abelian sector.5 The
moduli space of flat connections of the twisted SU(N )/ZN bundle on a two-torus is
isomorphic to that of SU(N) flat connections on the periodic torus. It can be constructed
explicitly in terms of the embedding SU(N) SU(b) SU(bN) = SU(N ) that is realized
in formula (2.10).
The BornOppenheimer reduction to zero modes reads

ik
tr F + F +
4
DT2 (L )

ik

2

dt L $ ij (a0 )i t (a0 )j tr H 1b H 1b .
(6.12)
Using now

tr H 1b H 1b = b ,
together with L = L/b and k = kb we find exactly the same effective action

k
dt $ ij Ci t Cj ,
4
(6.13)
when expressed in terms of the angular variables Cj = (a0 )j L, just as before. Thus, we
learn that, despite the rescaling of the ChernSimons level under Morita duality k kb,
5 This is a consequence of our choice of background field and magnetic fluxes in the original model, making
the diagonal U (1) effectively flat (2.12).
190
which is required by our trace normalizations, the space of flat connections is sensitive to
the same effective ChernSimons level before and after the duality. The reason for this is
of course that the Fourier components a remain the same in both representations and, in
particular, their periodicity should not change. This is also clear from the fact that the zero
modes know little about noncommutativity.
The periodicity of the zero-mode variables Cj is enforced by gauge transformations
that are periodic on the noncommutative torus and satisfy the condition of Moyal-unitarity:
U K U = U K U = 1N . Expanding in Fourier series:

u e2ix/L,
U (x) =
(6.14)
Z2
where the coefficient matrices u are not unitary in general. It is then easy to check that the
twisted gauge transformations

u V a1 U 2 a1 2 /2 e2ix/L,
U (x) =
(6.15)
Z2
satisfying
U (x + Lj ) = j U (x)j ,
(6.16)
enforce exactly the same periodicity conditions on Cj when acting on A + A , as defined

by (2.10).
6.3. The (2 + 1)-dimensional index
The upshot of the discussion in the preceding subsection is that, when calculating
the index of (2 + 1)-dimensional gauge theories by quantization of the moduli space
of flat connections, the effective ChernSimons level is exactly given by that of the
noncommutative theory, which remains fixed. Thus, the index is -independent and can
be calculated as if the U (N) model was an ordinary gauge theory with periodic boundary
conditions on the torus.
This means that, whatever its value, the index will be only a function of k and N , and
it is obviously smooth under rational approximations of . Although this settles our main
concern, it is nevertheless interesting to pursue this matter and compute the index of the
U (N) model.
Following our general discussion (4.7), the result factorizes into Abelian and nonAbelian components for each value of the electric flux. The linking between the U (1)
and SU(N) sectors is done by the ZN quotient acting on Wilson loops in each direction,
enforcing the constraint [wi ] = wi (mod N ), i = 1, 2.
Since the infrared behaviour of the 2 + 1 models under consideration is dominated by
the ChernSimons term, the degeneracy of the ground states of the full theory is naturally
related to the dimension of the Hilbert space of the topological ChernSimons model
describing the infrared limit.
One subtlety of the description in terms of the effective topological theory is that Wilson
lines wrapping homologically inequivalent cycles of the torus are canonical conjugates of
191
one another. Thus, the states are labelled by the eigenvalues of only one set of Wilson
lines. In the Abelian case, this can be understood in elementary terms. For an Abelian
MaxwellChernSimons model with action

1
i
SU (1) = 2 |dA|2 +
(6.17)
A dA + Fermi terms,
4e
4
we can explicitly find the zero energy states on the torus with periodic boundary conditions
(see, for example, [53] and references therein). Only the holonomies of the gauge field
Ci = i A are important, with effective Lagrangian
1
ij
(6.18)
$ Ci t Cj ,
(t Ci )2 +
2
2e
4
since the fermions are massive and free, and do not contribute to the ground-state
degeneracy. This Lagrangian is equivalent to that of a nonrelativistic particle of mass 1/e2
in a torus of length 2 , threaded by units of magnetic flux. Thus, for integer , the
ground state is a degenerate Landau level of states. This generalizes for rational values
of = q/p, with (q, p) = 1, since we may allow multivaluated wave-functionals. It is
enough to focus on the purely topological term that is obtained by neglecting the Maxwell
action (the kinetic energy of the particle). Then, canonical quantization of the holonomies
gives
Leff =
ei C1 ei C2 = e2ip/q ei C2 ei C1 ,
(6.19)
which can be represented by the q-dimensional pair of clock and shift matrices. Hence,
states are labelled by the q eigenvalues of, say exp(iC1 ). Notice that the zero-modes Ci are
normalized as angular variables with period 2 .
The topological nature of the ChernSimons theory allows us to obtain this spectrum
by the general construction of [54]. A basis for the Hilbert space of the ChernSimons
1
theory on T2 can be obtained by computing path integrals on
the filled torus D S ,
with D a disk carrying an insertion of a Wilson line exp(iw1 A1 ). By the usual Chern
Simons holography, this basis is in one to one correspondence with the set of integrable
representations of the corresponding level- WZW model on the boundary of the disk.
This restricts the possible values of w1 to the set
w1 = 0, 1, . . . , 1.
The fact that only one component of the electric flux is relevant is seen here by the
topological impossibility of filling in both homologically nontrivial circles of the torus at
the same time.
A similar construction can be carried out for the non-Abelian component of the index,
in terms of the set of integrable representations of the corresponding SU(N) level-k WZW
model. Thus, the index can be computed by imposing the ZN modding on the Hilbert space
of the product theory U (1) SU(N)k .
Notice that there is no shift of the Abelian level by effects of the fermionic measure,
since there are no other fields in the theory that are charged with respect to the diagonal
U (1) subgroup. Thus, we can freely adjust it depending on the physical definition we adopt
for the Abelian level.
192
If we wish the U (N) symmetry to act classically, we define the Abelian field A = A1N .
The periodic U (1)-gauge
transformations impose an angular identification, modulo 2 , to
the holonomy variables A. Hence, the effective Abelian level is = kN . However, since
the space of ground states is equivalent to that of the effective ChernSimons model that
arises after fermion integration, it is more natural to define the U (N) symmetry referred
and the classical
to this effective bosonic Lagrangian. In this case, we would have = kN
theory would be defined with an extra Abelian counterterm with coupling = N 2 /2.
Yet another possibility would be to define the Abelian level so that the U (N) symmetry has
a simple action on expectation values of Wilson lines. Since these are naturally functions
of k + N , one would have a natural definition = N(k + N).
, the purely bosonic low-energy ChernSimons model
In the particular case = kN
is a standard U (N)k model that can be analyzed directly in terms of the U (1)N Cartan
subalgebra. The problem is exactly given by the multi-particle generalization of (6.18) to
a system of N identical bosons. The degeneracy is given by all the wave functions on N
variables that can be constructed out of k elementary single-particle orbitals. This is
(N + k 1)!
.
(6.20)
N!(k 1)!
This is the total dimension of the Hilbert space of the product theory U (1)kN
SU(N)k ,
IU (N)k =
(N + k 1)! ,
kN
(N 1)!k!
divided by N 2 , and one can think of the modding by ZN as dividing by a factor of N
for each independent cycle of the torus. The general case with asymmetric levels is more
involved. In general, one can argue that the consistency of the ZN modding procedure
(mod N 2 ). Then, a natural conjecture would be that the index is given by
requires = kN
that of the product U (1) SU(N)k theory, divided by N 2 . Notice that the selection rule
for ensures that the result is always an integer.
7. A formal digression
On general grounds, the large-n limit implied in our construction of U (N) is
intimately connected with the definition of the gauge group for noncommutative theories.
In [55,56] (see also [3]) a preliminary analysis is made of this issue. In the traditional
WeylMoyal correspondence we consider the space of functions on some space, in our case
TD , with some particular smoothness conditions, and the standard product of functions
is deformed into a K-product. Under appropriate mathematical conditions, the functions
on the noncommutative torus can be mapped into operators on an infinite-dimensional,
separable Hilbert space H. The noncommutative gauge symmetry acts as a subgroup of the
group of unitary transformations on H, U (H). Which subgroup we choose will determine
the gauge group topology and it is likely also to determine important nonperturbative
properties of the theory.
In our construction of U (N) we approximate by rationals n and then, through
a Morita transformation, we relate the noncommutative theories with an ordinary U (Nn )
193
theory with a certain amount of magnetic flux, and Nn as n , irrational. The

details of the construction imply that in this procedure we are keeping the topology of
the gauge group for each intermediate Nn . From this point of view the gauge group looks
like the inductive limit of U (N), i.e., U (), whose homotopy groups are given by Bottperiodicity. A theorem of Palais (cf. [57]) implies that
U () U1 (H) Up (H) Ucpt (H),
(where U Up (H) iff U = 1 + O, with Op being trace class, and Ucpt (H) are compact
operators) all have the same homotopy type. Hence the proposal in Ref. [55] to consider
Ucpt (H) as the appropriate gauge group arises as a natural one.
If we were to carry out the naive quantization of the theory with such a gauge group,
it seems clear that the electric and magnetic sectors of the corresponding gauge theory
would remember the electric and magnetic sectors of U (N) for each N . One could in
principle imagine bigger gauge groups (all the way to U (H), which is contractible). In this
case once Gauss law is imposed we will lose many of the electric and magnetic sectors of
the standard analysis, or perhaps the new theory would correspond to specific averages of
the standard sectors. In the case of U (1) on T2 , the classical algebra of infinitesimal U (1)gauge transformations is equivalent to the FairlieFletcherZachos algebra, a trigonometric
deformation of the algebra of area-preserving diffeomorphisms on T2 , w (T2 ) (cf. [58]).
Once again, there are many choices that could be made to define the quantum theory with
this gauge group. One of them in particular uses the U () construction above. Other
constructions might involve the different definition of W starting with w (see [59] for
details and references). It is clear that the physics depends on the choice of the gauge group,
but at this stage it is not known to what extent. Addressing this question is likely to be one
of the critical elements in the direct construction of noncommutative gauge theories for
arbitrary values of .
8. Morita duality and topological order in the quantum Hall effect

The Morita transformation of the ChernSimons action (6.5) and, in particular, the
topological coupling (6.8), has an interesting application to the model of the fractional
quantum Hall effect (FQHE) proposed in [20] (see also [60,61]). Noncommutative Chern
Simons with rank-one was proposed in [20] as a more refined low-energy description of
the FQHE than the usual U (1) ChernSimons model [62,63]. Specifically, one considers
the rank-one model

2
1
A dA + A K A K A ,
S=
(8.1)
4
3
RR2
with a noncommutativity parameter determined in terms of the electrons fluid density:

= 1/2e . The topological coupling is 1/, with the FQHE filling fraction. We restrict
for the time being to the Laughlin series = 1/q with odd q. Since noncommutative
ChernSimons theories have a very smooth ultraviolet behaviour [64], we expect the dependence of physical quantities to be analytic near = 0, so that the effects of the
194
electrons granulariry reflected in can be expanded in powers of via the Seiberg

Witten map [5].
The real impact of the proposal (8.1) on the physics of the FQHE should be evaluated on
the basis of the induced theory for edge states. Extracting the boundary dynamics induced
by the bulk action (8.1) is technically nontrivial. In fact, it is natural to expect that, because
of the nonlocality
of the Moyal products, some boundary action with nonlocality on length
scales of O( ) must be added to the bulk Lagrangian in (8.1) (see [65,66] for results
in this direction). It is also possible that the applicability of (8.1) to FQHE phenomena
is reduced to the finite matrix approximations, such as those studied in [60,61], which
show encouraging similarities with the microscopic structure of the relevant wave-function
hierarchies.
One way of testing the smooth dependence on using discrete criteria is to look at
Wens topological order [19], i.e., the universal degeneracy of Laughlins ground state
on a torus. This degeneracy is equal to the ordinary ChernSimons level 1/. If the
noncommutative model is to give a good description of the FQHE fluid, we expect that
the degeneracy on the torus is correctly accounted for.
Working on a spatial two-torus, the dimensionless noncommutativity parameter is
rational
=
1
,
ne
(8.2)
where ne is the number of electrons in the torus. Thus, we can use Morita duality to define
the noncommutative model on the torus via the ordinary U (ne ) ChernSimons theory with
one unit of magnetic flux and level k = ne /. As in the previous sections, if we choose
the background field and magnetic flux to vanish for the noncommutative U (1) Chern
Simons model, the resulting ordinary model has an effectively periodic U (1) sector. On
the other hand, the SU(ne ) sector is twisted with one unit of t Hoofts magnetic flux. This
means that the non-Abelian sector does not contribute to the degeneracy of ground states,
since there are no SU(ne )/Zne flat connections on the completely twisted torus.
Therefore, the topological order is determined by the Abelian factor with level k =
ne /. As in the previous section, we must present the Abelian level in the effective
normalization relevant to the physics of the zero modes:
1

eff
(8.3)
= .
Thus, the effective Abelian level is invariant under Morita transformations and we recover
the expected result for the topological order.
In fact, since the ordinary non-Abelian ChernSimons model is a topological theory
with a trivial Hilbert space on the twisted torus (only the vacuum remains), we can say
that the physics of (8.1) on the torus is rigorously equivalent to that of the ordinary
Abelian ChernSimons model with the same level. Possible nontrivial -dependence could
only arise when including the Maxwell term in the effective Lagrangian, i.e., the mixing
with higher Landau levels. On the other hand, if the mixing with higher Landau levels
is significant, it is unlikely that the present field-theoretical degrees of freedom (the
statistical ChernSimons gauge field) will furnish a good description of the dynamics.
195
Our results in the previous sections indicate that this situation should also generalize to
non-Abelian models that have been proposed as effective theories of the incompressible
Hall fluid with more general filling fractions [20]. Namely, according to (2.10), a
noncommutative U (N) model with = 1/ne can be traded by a twisted ordinary theory
with gauge-field momentum modes in U (N) U (ne ) and flat connections living only on
the first factor. Since the effective level, as seen by the flat connections, does not change
under Morita duality, we conclude that the ground-state degeneracy is independent of
also in this generalized situation.
For example, if a = p/q multi-layer Hall fluid is represented by a U (p) noncommutative model at level q, the results in the previous section yield a value of the topological
order:
dim H(T2 ) =
(p + q 1)!
.
p!(q 1)!
(8.4)
What is specifically noncommutative in this prediction is the particular linking of

the global U (1) group and the SU(p) non-Abelian part. If this model is regarded as a
single-layer Hall fluid with non-Abelian statistics, the predicted topological order differs
in general from other schemes. For example, an Abelian model of type [67] for the main
Jain sequences, = p/q with q = 2ps + 1 and s integer, has enhanced U (1) SU(p)1
affine symmetry, but the topological order is still given by
dim H(T2 )Jain = q = 2ps + 1.
(8.5)
The success of the standard schemes suggests that the noncommutative non-Abelian
models are unlikely to describe single-layer fluids along the main sequences. Therefore,
their possible applications would be restricted to multi-layer fluids.
8.1. A comment on level normalizations
In the conformal field theory approach to the QHE (see [68,69] for a summary) the
topological order is given by the level of the Abelian current algebra of edge excitations.
This is in turn equal to the classical coupling k of the ChernSimons Lagrangian. In the
noncommutative case, many aspects of the perturbation theory of the rank-one model (8.1)
are similar to the behaviour of non-Abelian ChernSimons theory. One such instance is the
quantum shift k k + 1 of the level, analogous to the shift k k + N in ordinary SU(N)
ChernSimons theory [70]. To a large extent, the quantum shift in ordinary ChernSimons
theory is a matter of renormalization prescription, although many observables, such as
expectation values of Wilson lines, are conveniently written in terms of the shifted level.
The important point for us is that such a shift does not affect the evaluation of the
dimension of the Hilbert space on the torus, which is still given in terms of the classical
ChernSimons coupling. This is rather clear in our computation: the topological order is
only sensitive to the diagonal U (1), for which there is no level-shift, since no fields are
charged with respect to this subgroup.
In the context of the finite-matrix models of [60], the link between the filling fraction
and the ChernSimons level does reflect the shift: k + 1 = 1/. However, these models
contain additional matter fields in the fundamental representation of the U (N) group.
196
Hence, direct comparison of the finite-matrix level and the one appearing in (8.1) may
require due attention to renormalization effects induced when integrating-out the matter
fields.
9. Conclusions
We have studied some properties of models in the dense completion of rational
noncommutative gauge theories, defined in finite volume. These models, denoted U (N) ,
are defined by n limits of rational theories U (N, n ).
Assuming that these limiting models exist, they are interesting for two reasons. First,
one could try to use them in giving a constructive definition of the generic noncommutative
gauge theory. Even if such a program should fail and the U (N) theory turned out to be
generically inequivalent to some other independent definition of the irrational U (N)
theory, the physics of the U (N) is certainly interesting in itself.
The second interesting aspect of the U (N) models is that, using Morita duality, we
may as well define them as certain large-n limits of ordinary gauge theories. Thus, to the
extent that smooth behaviour in could have consequences for the physics of ordinary
theories, one would be interested in precisely the U (N) models. There may be an
interesting space of large-n limits in ordinary gauge theories that remain to be explored,
and of which the noncommutative theories on tori are just examples.
The evidence for continuity of the physics as a function of is strong for theories with
an underlying N = 4 supersymmetry. In models with less supersymmetry, smoothness in
is expected in perturbation theory, except perhaps at = 0. There are exceptional values
of the momenta in perturbation theory for which the rational and irrational theories behave
very differently, but these modes seem to decouple in the irrational limit. At any rate, the
most interesting tests of -continuity would involve nonperturbative physics such as that
of confining N = 1, 0 models.
Using the definition of U (N) in terms of series of ordinary theories, we have used
various expectations about the dynamics of ordinary confining theories to put constraints
on the -dependence of the U (N) models. In particular, we find that the Witten
index (when properly defined, so that it gives nontrivial information) of minimal N = 1
models in three and four dimensions is smooth under rational approximations. In the fourdimensional case, this result depends on the subtle interplay between the instanton angle
and the magnetic flux, exactly as in the dynamics of oblique confinement.
A related interesting question is the dependence of the energy on the instanton angle, a
truly non-BPS quantity. An estimate of the vacuum energy as a function of can be given
by a dilute-gas fractional-instanton approximation in the series of ordinary theories. This
corresponds to the dilute-gas of ordinary noncommutative instantons after Morita duality.
We find that the nontrivial level-crossing phenomena that ensure 2 -periodicity of physics
in ordinary theories translate into a 2/bn -periodicity in the noncommutative theories. In
the bn limit, the non-Abelian contribution to the -dependence becomes either trivial
or discontinuous.
We have checked that the instanton-induced functional dependence is smooth in
provided no level-crossing phenomena takes place. Such is the case when N = 4
197
supersymmetry is restored at some high energy scale. On the other hand, if supersymmetry
is not restored, level-crossing may occur, resulting in strictly -independent energies in the
U (N) model. Such a result violates -smoothness when the rational series approximate a
rational number. Therefore, supersymmetry still acts as a custodian of continuity in . It
would be interesting to know whether there are other scaling or continuum limits that would
guarantee continuity without necessarily requiring an underlying N = 4 supersymmetric
theory in the ultraviolet. In the nonsupersymmetric cases, our treatment of the ultraviolet
cutoff is not rigorous, in the sense that we assume implicitly a commutativity of Morita
equivalence and continuum limit. It is thus possible that, when the correct physical
questions are asked, continuity of the physics in is borne out.
There are still quite a number of open questions; in particular a viable construction of
the quantum theory directly in the noncommutative setting. In this case one of the problems
to be solved is the question of the appropriate definition of the gauge group. It may turn
out that some definitions of the noncommutative theory are completely disconnected from
the naive approximations presented in this paper. A less ambitious problem that should be
tractable is to look again at the UV/IR problem at finite volume. In theories with rational
values of (whether they are gauge theories or not) the problem can be expressed in
terms of ordinary field theories whose fields are matrix valued. In this context the standard
renormalization of the theory can be carried out without problems, and in particular one
can use standard techniques to define the Wilsonian effective action (with due attention
paid to the presence of electric and magnetic fluxes). The peculiar UV/IR mixing pointed
out in [7,32] would appear as we take combinations of limits where becomes irrational,
or the volume of the torus goes to infinity or both. In any of these cases (see Section 3.4)
we end up with limits involving large noncommutativity or nonstandard large-n limits. By
looking at rational approximations to the UV/IR mixing we may gain some understanding
of this phenomenon and of its physical significance.
Finally we have shown that there are a number of interesting things that can be learned
from Morita duality when applied to the formulation of the fractional Hall effect proposed
in [20]; in particular we get an elegant derivation of Wens topological order [19]. Since
ChernSimons theory is rather soft in the ultraviolet, we do not expect any problems
concerning continuity in the noncommutative parameter. For these applications however,
this problem is not relevant since the dimensionless deformation parameter is always
rational, and given by the inverse of the number of electrons in the torus. A very relevant
question is how to extract the boundary dynamics of edge states induced by the bulk
action (8.1). It would be interesting to know how to define noncommutative spaces with
boundaries. Perhaps in this context the use of the Morita equivalence to an ordinary gauge
theory with gauge group U (ne ) should provide useful guidance.
Note added
When this paper was being finished, a paper by Z. Guralnik [71] appeared studying
noncommutative QED using also Morita equivalence and large-N limits of ordinary gauge
theories.
198
Acknowledgements
We are indebted to Pierre van Baal, Andrea Cappelli, Margarita Garca Prez, Csar
Gmez, Luca Griguolo, Zack Guralnik, Jos Labastida, Esperanza Lpez, Eliezer Rabinovici, Domenico Seminara, Albert Schwarz and Paolo Valtancoli for useful discussions.
We also thank Angel Paredes for finding an error in a previous version of Section 6.3.
L.A.-G. would like to thank the hospitality of the Humboldt University at Berlin, where
part of this work was done, and in particular Dieter Lst.
References
[1] A. Connes, Noncommutative Geometry, Academic Press, 1994.
[2] J. Madore, An Introduction to Noncommutative Geometry and its Physical Applications, 2nd edn., London
Mathematical Society Lecture Notes Series, Vol. 257, Cambridge Univ. Press, 1999;
J.M. Gracia-Bondia, J.C. Varilly, H. Figueroa, Elements of Noncommutative Geometry, Birkhuser, Boston,
2001.
[3] M.R. Douglas, N.A. Nekrasov, hep-th/0106148;
A. Konechny, A. Schwarz, hep-th/0012145;
R.J. Szabo, hep-th/0109162.
[4] A. Connes, M.R. Douglas, A. Schwarz, JHEP 02 (1998) 003, hep-th/9711162;
M.R. Douglas, C.M. Hull, JHEP 02 (1998) 008, hep-th/9711165.
[5] N. Seiberg, E. Witten, JHEP 09 (1999) 032, hep-th/9908142.
[6] C.-S. Chu, P.-M. Ho, Nucl. Phys. B 550 (1999) 151, hep-th/9812219;
M.M. Sheikh-Jabbari, Phys. Lett. B 455 (1999) 129, hep-th/9901080;
D. Bigatti, L. Susskind, Phys. Rev. D 62 (2000) 066004, hep-th/9908056.
[7] S. Minwalla, M. Van Raamsdonk, N. Seiberg, JHEP 02 (2000) 020, hep-th/9912072.
[8] N. Seiberg, L. Susskind, N. Toumbas, JHEP 06 (2000) 044, hep-th/0005015;
N. Seiberg, L. Susskind, N. Toumbas, JHEP 06 (2000) 021, hep-th/0005040;
R. Gopakumar, J. Maldacena, J. Minwalla, S. Strominger, JHEP 06 (2000) 036, hep-th/0005048;
J.L.F. Barbn, E. Rabinovici, Phys. Lett. B 486 (2000) 202, hep-th/0005073.
[9] J. Gomis, T. Mehen, Nucl. Phys. B 591 (2000) 265, hep-th/0005129;
L. Alvarez-Gaum, J.L.F. Barbn, R. Zwicky, JHEP 05 (2001) 057, hep-th/0103069.
[10] A. Schwarz, Nucl. Phys. B 534 (1998) 720, hep-th/9805034;
M.A. Rieffel, A. Schwarz, math.QA/9803057;
M.A. Rieffel, J. Diff. Geom. 31 (1998) 535, quant-ph/9712009.
[11] D. Brace, B. Morariu, B. Zumino, Nucl. Phys. B 545 (1999) 192, hep-th/9810099;
D. Brace, B. Morariu, B. Zumino, Nucl. Phys. B 549 (1999) 181, hep-th/9811213;
C. Hofman, E. Verlinde, JHEP 12 (1998) 010, hep-th/9810116;
C. Hofman, E. Verlinde, Nucl. Phys. B 547 (1999) 157, hep-th/9810219;
B. Pioline, A. Schwarz, JHEP 08 (1999) 021, hep-th/9908019.
[12] G. Landi, F. Lizzi, R.J. Szabo, Commun. Math. Phys. 206 (1999) 603, hep-th/9806099.
[13] Z. Guralnik, J. Troost, JHEP 05 (2001) 022, hep-th/0103168.
[14] A. Gonzlez-Arroyo, C.P. Korthals-Altes, Phys. Lett. B 131 (1983) 396.
[15] A. Gonzlez-Arroyo, M. Okawa, Phys. Rev. D 27 (1983) 2397;
T. Eguchi, R. Nakayama, Phys. Lett. B 122 (1983) 59.
[16] J. Ambjorn, Y.M. Makeenko, J. Nishimura, R.J. Szabo, JHEP 05 (2000) 023, hep-th/0004147.
[17] G. Landi, F. Lizzi, R.J. Szabo, Commun. Math. Phys. 217 (2001) 181, hep-th/9912130.
[18] T. Filk, Phys. Lett. B 376 (1996) 53.
[19] X.-G. Wen, Phys. Rev. B 40 (1989) 7387;
X.-G. Wen, Int. J. Mod. Phys. B 4 (1990) 239;
X.-G. Wen, Int. J. Mod. Phys. B 5 (1991) 1641.
199
[20] L. Susskind, hep-th/0101029.

[21] C.P. Mart, D. Snchez-Ruiz, Phys. Rev. Lett. 83 (1999) 476, hep-th/9903077;
M.M. Sheikh-Jabbari, JHEP 06 (1999) 015, hep-th/9903107;
T. Krajewski, R. Wulkenhaar, Int. J. Mod. Phys. A 15 (2000) 1011, hep-th/9903187.
[22] C. Hofman, E. Verlinde, Nucl. Phys. B 547 (1999) 157, hep-th/9810219.
[23] K. Saraikin, J. Exp. Theor. Phys. 91 (2000) 653, hep-th/0005138.
[24] A. Konechny, A. Schwarz, Nucl. Phys. B 550 (1999) 561, hep-th/9811159;
A. Konechny, A. Schwarz, Phys. Lett. B 453 (1999) 23, hep-th/9901077;
A. Konechny, A. Schwarz, JHEP 09 (1999) 030, hep-th/9907008.
[25] A. Hashimoto, N. Itzhaki, JHEP 12 (1999) 007, hep-th/9911057.
[26] S. Elitzur, B. Pioline, E. Rabinovici, JHEP 10 (2000) 011, hep-th/0009009.
[27] G. t Hooft, Nucl. Phys. B 75 (1974) 461.
[28] A. Armoni, Nucl. Phys. B 593 (2001) 229, hep-th/0005208.
[29] C.P. Martn, F. Ruiz-Ruiz, Nucl. Phys. B 597 (2001) 197, hep-th/0007131;
F. Ruiz-Ruiz, Phys. Lett. B 502 (2001) 274, hep-th/0012171.
[30] V.V. Khoze, G. Travaglini, JHEP 01 (2001) 026, hep-th/0011218.
[31] T. Krajewski, R. Wulkenhaar, Int. J. Mod. Phys. A 15 (2000) 1011, hep-th/9903187.
[32] M. Hayakawa, Phys. Lett. B 478 (2000) 394, hep-th/9912094;
M. Hayakawa, hep-th/9912167;
A. Matusis, L. Susskind, N. Toumbas, JHEP 12 (2000) 002, hep-th/0002075;
I.Ya. Arefeva, D.M. Belov, A.S. Koshelev, hep-th/0003176.
[33] S.S. Gubser, S.L. Sondhi, Nucl. Phys. B 605 (2001) 395, hep-th/0006119.
[34] J. Gomis, T. Mehen, M.B. Wise, JHEP 08 (2000) 029, hep-th/0006160.
[35] K. Landsteiner, E. Lpez, M.H.G. Tytgat, JHEP 09 (2000) 027, hep-th/0006210;
K. Landsteiner, E. Lpez, M.H.G. Tytgat, JHEP 06 (2001) 055, hep-th/0104133.
[36] J.L.F. Barbn, E. Rabinovici, JHEP 12 (1999) 017, hep-th/9910019.
[37] G. t Hooft, Nucl. Phys. B 138 (1978) 1;
G. t Hooft, Nucl. Phys. B 153 (1979) 141;
G. t Hooft, Nucl. Phys. B 205 (1982) 1.
[38] Z. Guralnik, hep-th/9804057.
[39] E. Witten, Nucl. Phys. B 202 (1982) 253.
[40] E. Witten, hep-th/9903005.
[41] E. Witten, hep-th/0006010.
[42] A. Kovner, M. Shifman, Phys. Rev. D 56 (1997) 2396, hep-th/9611213.
[43] E. Witten, Phys. Lett. B 86 (1979) 283.
[44] G. t Hooft, The confinement phenomenon in quantum field theory, in: M Lvy, J.-L. Basdevant (Eds.),
Cargse Summer School Lecture Notes on Fundamental Interactions, NATO Advanced Science Institutes
Series B: Physics, Vol. 85, 1981, p. 639.
[45] P. van Baal, hep-ph/0008206.
[46] A. Gonzlez-Arroyo, hep-th/9807108.
[47] J.L.F. Barbn, M. Garca Prez, in preparation.
[48] M. Garca-Prez, A. Gonzlez-Arroyo, J. Phys. A 26 (1993) 2667;
A. Montero, JHEP 0005 (2000) 022, hep-lat/0004009.
[49] N. Nekrasov, A. Schwarz, Commun. Math. Phys. 198 (1998) 689, hep-th/9802068;
A. Astashkevich, N. Nekrasov, A. Schwarz, Commun. Math. Phys. 211 (2000) 167, hep-th/9810147.
[50] V.P. Nair, A.P. Polychronakos, Phys. Rev. Lett. 87 (2001) 030403, hep-th/0102181;
D. Bak, K. Lee, J. Park, Phys. Rev. Lett. 87 (2001) 030402, hep-th/0102188;
M.M. Sheikh-Jabbari, Phys. Lett. B 510 (2001) 247, hep-th/0102092.
[51] T. Krajewski, math-ph/9810015.
[52] C.-S. Chu, Nucl. Phys. B 580 (2000) 352, hep-th/0003007.
[53] A.P. Polychronakos, Ann. Phys. 203 (1990) 231.
[54] E. Witten, Commun. Math. Phys. 121 (1989) 351.
[55] J.A. Harvey, hep-th/0105242.
[56] F. Lizzi, R.J. Szabo, A. Zampini, JHEP 08 (2001) 032, hep-th/0107115.
200
[57]
[58]
[59]
[60]
[61]
[62]
[63]
[64]
[65]
[66]
[67]
[68]
[69]
[70]
[71]
R.S. Palais, Topology 3 (1965) 271.

D.B. Fairlie, P. Fletcher, C.K. Zachos, J. Math. Phys. 31 (1990) 1088.
A. Cappelli, C.A. Trugenberger, G.R. Zemba, Nucl. Phys. B 448 (1995) 470, hep-th/9502021.
A.P. Polychronakos, JHEP 04 (2001) 011, hep-th/0103013;
A.P. Polychronakos, JHEP 06 (2001) 070, hep-th/0106011.
S. Hellerman, M. Van Raamsdonk, hep-th/0103179.
S.C. Zhang, T.H. Hansson, S. Kivelson, Phys. Rev. Lett. 62 (1988) 82.
S. Bahcall, L. Susskind, Int. J. Mod. Phys. B 5 (1991) 2735.
A. Das, M.M. Sheikh-Jabbari, JHEP 06 (2001) 028, hep-th/0103139;
C.P. Martn, Phys. Lett. B 515 (2001) 185, hep-th/0104091.
N. Grandi, G.A. Silva, Phys. Lett. B 507 (2001) 345, hep-th/0010113.
A. Pinzul, A. Stern, hep-th/0107179.
J. Frohlich, A. Zee, Nucl. Phys. B 364 (1991) 517.
A. Cappelli, C.A. Trugenberger, G.R. Zemba, hep-th/9610019.
J. Frohlich, B. Pedrini, C. Schweigert, J. Walcher, cond-mat/0002330.
G.-H. Chen, Y.-S. Wu, Nucl. Phys. B 593 (2001) 562, hep-th/0006114.
Z. Guralnik, hep-th/0109079.

A string bit Hamiltonian approach to

two-dimensional quantum gravity
B. Durhuus a,b , C.-W.H. Lee a,b
a MaPhySto, Centre of Mathematical Physics and Stochastics 1 , Denmark
b Department of Mathematics, University of Copenhagen, Universitetsparken 5,
DK-2100 Copenhagen, Denmark

Received 23 August 2001; accepted 7 December 2001
Abstract
Motivated by the formalism of string bit models, or quantum matrix models, we study a class of
simple Hamiltonian models of quantum gravity type in two spacetime dimensions. These string bit
models are special cases of a more abstract class of models defined in terms of the sl2 subalgebra
of the Virasoro algebra. They turn out to be solvable and their scaling limit coincides in special
cases with known transfer matrix models of two-dimensional quantum gravity. 2002 Published by
Elsevier Science B.V.
PACS: 04.60.Kz; 04.60.Nc
Keywords: Quantum gravity; String bit model; Large-N limit; Scaling limit; sl2 algebra
1. Introduction
So far the most productive approach to two-dimensional quantum gravity has been in
terms of path integrals. Specifically, the (Euclidean) path integral formulation of Liouville
gravity [1] together with its regularized versions involving random triangulations, or
the equivalent matrix models (see, e.g., Ref. [2]), have permitted detailed analysis of
two-dimensional quantum gravity coupled to matter fields with central charge c 1,
in particular the loop correlation functions (or HartleHawking wave functions) and, in
some cases, the fractal characteristics of spacetime. Importantly, and as a rare instance in
E-mail addresses: durhuus@math.ku.dk (B. Durhuus), lee@math.ku.dk, h11lee@math.uwaterloo.ca
(C.-W.H. Lee).
1 Funded by a grant from the Danish National Research Foundation.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 2 8 - 9
202
B. Durhuus, C.-W.H. Lee / Nuclear Physics B 623 (2002) 201219
quantum field theory, the discrete models have provided analytic insights that presently
seem out of reach of continuum methods, the most striking of which is perhaps the double
scaling limit as a procedure to incorporate spacetimes of arbitrary topology. (See Ref. [2]
and the references therein.)
One of the main motivations for studying quantum gravity in two dimensions is based
on the hope that it may serve as a testing ground for ideas and methods extendable to higher
dimensions. It is, indeed, straightforward to set up discrete models of quantum gravity in
arbitrary dimensions in terms of random triangulations, but up to now very few analytic
results have been obtained and even very basic questions are left open. A number of
numerical investigations have, however, been carried out. See, e.g., Ref. [2] and references
therein.
The question arises, naturally, if there exist alternative formulations of the regularized
models, or closely related models, that are better tailored for generalizations. A second, and
independent, purpose of such reformulations would be to make comparisons possible with
continuum approaches other than the path integral quantization, in particular canonical
quantization [3]. In this paper we address this question in two dimensions and introduce a
class of (Euclidean) Hamiltonian models of regularized two-dimensional quantum gravity.
We do not claim to resolve the above mentioned questions, but we will show that the
proposed models provide a Hamiltonian alternative to the discrete path integral (or transfer
matrix) approach to a class of models introduced recently under the name Lorentzian
gravity [4,5], which on the other hand are closely related to the full randomly triangulated
models mentioned at the beginning. More precisely, we will show that the continuum limits
of the Lorentzian models can be obtained from our Hamiltonian models. Indeed, we will
define and solve a more abstract class of models and introduce a formalism opening up the
way for even further generalizations. Detailed analysis of these generalized models as well
as generalizations to higher dimensions is, however, outside the scope of the present paper.
Here is the basic idea of this work. Recall that the primary goal of the Hamiltonian
formulation of quantum gravity is to account for the time-evolution of a spacelike universe of fixed topology.2 Restricting first to connected and compact space-like
universes implies that the two-dimensional spacetime has either the topology of a strip,
corresponding to equal-time slices that are open line segments, or a cylinder with circles as
equal-time slices. Since the only reparametrization-invariant quantity defined by a metric
on a one-dimensional (connected) manifold is its volume, a natural way to discretise the
spatial metric degree of freedom is to introduce a distance cutoff a > 0 and consider the
equal-time slices to be polygonal lines or loops, respectively, with volume n a, where the
integer n is the number of links in the slice. Keeping a fixed, we associate with each such
space-like universe of volume n a a pure quantum state |n, and these are assumed to
form an orthogonal basis for the Hilbert space T of states. The Hamiltonian acting on T is
chosen in such a way that the action couples adjacent links only.
It turns out that models of this type may conveniently be generated by a special variant
of the string-bit formalism, whose basic ingredients are annihilation and creation operators
that can be interpreted as annihilating or creating links in the equal-time slices. (We
2 Note that the models considered in this paper are all within the Euclidean framework.
203
will explain the necessary details in Sections 2 and 3.) String bit models were originally
developed as a means of regularizing string theory [6,7]. They provide suitable frameworks
for quantum chromodynamics [8] and quantum spin chain models [9], too. Our variant may
be envisaged as generalized quantum spin chain models in which the numbers of spins,
i.e., links, are variable. In this sense, the relationship between Lorentzian gravity models
and those string bit models equivalent to them is analogous to that between the six-vertex
and the XXZ model [10]; a better understanding of one class of models will spur the study
of the other.
The paper is organised as follows. In Section 2, the simplest possible Hamiltonian model
in the case when spacetime is a strip will be solved in the continuum limit. This model
turns out to coincide with the corresponding Lorentzian gravity model considered in [5];
in Section 3, we will consider a Hamiltonian model for cylindrical spacetime which is not
spatially homogeneous (or cyclically symmetric). The model will be shown to reproduce,
in the continuum limit, the Lorentzian model with a marked link in the initial space-like
slice considered in [4]; in Section 4, we will consider the cyclically symmetric version of
the previously mentioned model and show that this, as well as the model in Section 2, can
be obtained as special cases of a more general class of models expressed in terms of the
sl2 -generators of the Virasoro algebra in a certain class of highest weight representations,
the Hamiltonian being of the form
H = L0 + L1 + L1 .
(1)
In Section 5, we will solve this model. In particular, both the two-loop amplitude of the
continuum Liouville gravity model in Ref. [11] and the so-called p-seamed correlation
functions of Ref. [5] will be obtained as special cases; finally, we will discuss briefly further
generalisations and future developments in Section 6.
2. Spacetime with boundaries

In this section, we are going to consider the quantum mechanics of a spacetime with the
topology of a strip whose Hamiltonian is given by

2
1 t
1
H 1 = Tr a a + a a + a a 2 q q q q t
(2)
2
4
4
N
N
for N = . Before explicating the various terms in Eq. (2), let us briefly review what string
bit models are. We will largely follow Refs. [12,13], with a few modifications in definitions
and notations.
Consider an N N matrix of creation operators. Its matrix entry is written as a1 2 ,

where 1 and 2 are row and column indices, respectively, and can take any integer values
between 1 and N inclusive. The Hermitian conjugate of this matrix is an N N matrix of
annihilation operators whose matrix entries are written in the form a21 . The creation and
annihilation operators satisfy the canonical commutation relations

a21 , a4 3 = 41 23 .
All other commutators involving these operators vanish.
204
In addition, consider a 1 N row vector and an N 1 column vector of creation

operators. Their components are written as q and q , respectively. Their Hermitian
conjugates are an N 1 column vector and a 1 N row vector of annihilation operators.
Their components take the form q and q , respectively. These operators satisfy the
canonical commutation relations

q 1 , q2 = 21 and q 1 , q 2 = 12 .
All other commutators involving them, including those involving a1 2 or a21 as well,
vanish.
Let | be a vacuum state annihilated by all annihilation operators, and define T1/2 as
the Hilbert space spanned by all states of the form
|n 1 =
2
1
N (n+1)/2
q 1 a1 2 a2 3 an n+1 q n+1 | =
1
N (n+1)/2
n
q a q |,
where n is a positive integer and the summation convention is adopted for all row and
column indices. The choice of notation | 1/2 for vectors in T1/2 will be explained in
Section 4. The inner product | 1/2 on T1/2 is fixed uniquely by the commutation
relations and | = 1. Since [12]
lim m|n 1 = mn ,
the states |11/2 , |21/2 , . . . , and so on form an orthonormal basis of T1/2 in the large-N
limit, which is the limit we are considering. We think of |n1/2 as the quantum state of a
universe made up of n interior links and two boundary links.
There are various kinds of natural operators acting on T1/2 . We define
k =
=
1
a1 2 a2 3
(k+2)/2
N
1
N (k+2)/2
l2
ak al1 a1
a11

Tr (a )k a ,
(3)
where k and are any positive integers, and Tr denotes the trace in index space (and not
on T1/2 ). Moreover, we let
l00 = q q
(4)
t
r00 = q q = q q t ,
(5)
and
where the superscript t denotes transposition. In the large-N limit [8,12],

0
if > n,
k |n 1 = (n + 1)|n + k 1 + O 1 if n,
2
l00 |n 1
2
= |n 1 ,
2
r00 |n 1
2
= |n 1 ,
2
(6)
where O(1/N) consists of terms whose norms and whose inner products with any |n1/2
are of the order of at most 1/N . These terms are thus negligible in the large-N limit. We
205
can see from Eq. (6) that k replaces any adjacent interior links with k adjacent interior
links and annihilates |n1/2 if > n. l00 and r00 annihilate the left and right boundary links,
respectively, and then create them back. Hence, both l00 and r00 effectively act as the identity
operator on T/ , but we will nevertheless display them explicitly below.
Using Eqs. (3)(5), we can rewrite the Hamiltonian H1/2 in Eq. (2) as
1
1
H 1 = 11 + 12 + 21 l00 r00 ,
2
4
4
(7)
where is a real constant. In this formula, 11 is the interior spatial volume energy term.
Each interior link carries one unit of energy, and the volume energy of a state is proportional
to the number of links. 12 splits any interior link into two. Since |n1/2 represents n interior
links, 12 maps |n1/2 to n|n + 11/2 . On the other hand, 21 combines any two adjacent
interior links into one. Since there are only n 1 pairs of adjacent interior links in |n1/2 ,
21 maps |n1/2 to (n 1)|n1/2 . Finally, the last two terms represent boundary volume
energy, but notice that the two boundary links contribute negative energy, in total minus one
half the energy of an interior link. We stress that this value of the relative size of the volume
energy contributions is crucial for the existence of a continuum limit as will be seen. By
now, it should have been obvious that H1/2 , featuring the physics of spatial homogeneity
and locality, is among the simplest Hamiltonians one can conceive for a spacetime with
boundaries.
We now proceed to evaluate the transition amplitudes
1 (k, l; t) = l|et H1/2 |k 1
G
2
in the continuum limit. It is convenient to work with its generating function

G 1 (x, y; t) =
2
1 (k, l; t),
xkyl G
2
k,l=1
where x, y are complex variables. Introducing

n
l; n) = l|H1/2
(k,
|k 1
2
and its generating function
(x, y; n) =
l; n),
x k y l (k,
k,l=1
for non-negative integers n, we have

G 1 (x, y; t) =
2
Note, firstly, from

(t)n
n=0
n!
(x, y; n).
1
l; 1) = l|H 1 |k 1 = k
(k,
lk + kl,k+1 + (k 1)l,k1
2
2
2
(8)
206
that
(x, y; 1) =
xy
xy(1 + x + y)
.
2
2(1 xy)
(1 xy)
(9)
Secondly, since |11/2 , |21/2 , . . . , and so forth form an orthonormal basis of T1/2 , we have
[4]

dz
(x, y; n) =
(10)
x, 1z ; 1 (z, y; n 1),
2iz
where both (x, z1 ; 1) and (z, y; n 1) are treated as complex functions of z, and
the contour encircles all singularities of (x, z1 ; 1) but none of (z, y; n 1). Using
Eq. (9), we find that Eq. (10) leads to

1
2
+
(x, y; n) =
+ + x + x
(x, y; n 1).
2 x
x
By Eq. (8), this yields

1
2
+ + x + x
G 1 (x, y; t) = 0.
G 1 (x, y; t) +
+
2
t 2
2 x
x
(11)
Together with the initial condition

G 1 (x, y; 0) = (x, y; 0) =
2
xy
1 xy
(12)
this first order partial differential equation determines G1/2 (x, y; 0) uniquely.
We are presently only interested in evaluating the continuum limit of G1/2 . Singularities
appear when the coefficients in the square brackets in Eq. (11) vanish, i.e., for x = y = 1
and = 1/2. These are identical to the critical values found in Refs. [4,5], and we can
apply the same scaling procedure. Hence we set
1 2
2T
,
x = eXa ,
y = eY a , and = e 2 a ,
(13)
a
2
where a is the distance cutoff and T , X, Y , and are finite renormalized values of t, x, y,
and , respectively, and define the continuum continuum limit of G1/2 , for which we use
the same notation, by
t=
G 1 (X, Y ; T ) = lim aG 1 (x, y; t).

2
a0
(14)
Substituting Eq. (13) into Eqs. (11) and (12) yields the limiting equation

G 1 (X, Y ; T ) + X2
G 1 (X, Y ; T ) + XG 1 (X, Y ; T ) = 0,
2
2
T
X 2
with the initial condition
1
.
G 1 (X, Y ; 0) =
2
X+Y
These are identical to the equations found for a model of Lorentzian gravity with
boundaries in Ref. [14]. By inverse Laplace transformation of the solution with respect to
207
X and Y , one finds (see Ref. [4]) the continuum limit of the transition amplitude expressed
in terms of the physical lengths L = k a, L = l a of the two boundary components to be

2 LL

(L+L ) coth T

G 1 (L, L ; T ) =
(15)
,
I0
e
2
sinh(T )
sinh(T )
where I0 is the zeroth modified Bessel function. Consequently, our string bit model and
this particular model of Lorentzian gravity belong to the same universality class. We will
come back to this model again in Section 4.
3. Closed, non-homogeneous spacetime

In this section we discuss an example of a Hamiltonian model of cylindrical spacetime.
Conceptually, it is not the simplest such model, which we defer to the next section.
However, it has the virtue of being solvable by the generating function technique of the
preceding section, which is our main motivation for considering it here. The model is
spatially non-homogeneous in the sense that the equal time slices have one marked link,
that is a link created by a matrix of creation operators different from those creating the rest.
A model of Lorentzian gravity with a marked initial loop has been studied in Ref. [4], and
we will find that its continuum limit is reproduced by our model.
The Hamiltonian we consider is given by

Hc = Tr a a + b b
1
1
+ Tr a a a + a bb + b a b + a a 2 + b ba + b ab
(16)
2
2
N
for N = . As mentioned, this quantum matrix model requires a second matrix of creation
operators besides a . The entries of this matrix are written as b2 1 , whose corresponding
2
annihilation operator is b1 . They satisfy the same canonical commutation relations as the
a-operators and commute with these.
Let Tc be the Hilbert space spanned by all states of the form
|nc =
1
N n/2

Tr b (a )n1 |,
where n is an arbitrary positive integer. These states form an orthonormal basis for Tc in
the large-N limit [8,12]:
lim m|nc = mn .
We consider |nc as the state of a closed universe with n links, one of which, created by
b , is marked (cf. the string bit interpretation in Refs. [6] and [7].)
The operators acting on Tc which are relevant to us may be written either as
glk =

1
Tr (a )k a l ,
N (k+l2)/2
(17)
208
where k and l are positive integers, or as

glk11,l,k22 =
1
N (k1 +k2 +l1 +l2 )/2

Tr (a )k1 b (a )k2 a l2 ba l1 ,
(18)
where k1 , k2 , l1 , and l2 are non-negative integers. (lk and glk are the restrictions of the
same operator to the open and closed string sectors, respectively. They are elements of
different Lie algebras [12,13], a fact we will see and use in Section 4.) In the large-N limit
[8,12],

0
k
if l n,
gl |nc =
if l < n,
(n l)|n l + kc + O N1
and

glk11,l,k22 |nc =
0

|n l1 l2 + k1 + k2 c + O N1
if l1 + l2 > n 1,
if l1 + l2 < n.
Thus, glk replaces any adjacent l unmarked links in |nc with k unmarked links, and glk11,l,k22
replaces adjacent l1 + l2 + 1 links, where the (l1 + 1)-th link is marked, with k1 + k2 + 1
links, where the (k1 + 1)-th link is marked. Note that these operators always preserve the
marked link.
Using Eqs. (17) and (18), we can paraphrase Hc in Eq. (16) as

1 1,0 1 0,1
0,0
0,0
0,0
1
2
1
Hc = g1 + g0,0 + g1 + g0,0 + g0,0 + g2 + g1,0 + g0,1 ,
2
2
0,0
where is a real constant. In this formula, g11 + g0,0
is the volume energy term. The
1,0
0,1
terms g12 + 1/2g0,0
+ 1/2g0,0
implement splitting of any unmarked link into two unmarked
links or splitting the marked link into a marked and an unmarked link. Finally, the terms
0,0
0,0
g21 + g1,0
+ g0,1
combine a pair of juxtaposed unmarked links into one unmarked link or
combine the marked link and a juxtaposed unmarked link into the marked link. Notice that
the relative constants of the terms in Hc are chosen such that its action treats the marked
link in the same way as the unmarked ones. More specifically,

1
1 1,0 1 0,1
0,0
g12 + g0,0
+ g0,0 |nc = n|n + 1c ,
|nc = n|nc ,
g1 + g0,0
2
2
1
0,0
0,0
g2 + g1,0 + g0,1 |nc = n|n 1c .
Thus this form of Hc appears to represent the most natural nearest neighboring interaction
on Tc , but notice that it is non-Hermitian.
As already mentioned, it turns out that the model so defined can be solved by the same
method as that of the preceding section. Since the differences between the calculations are
only minor, we will skip the details. Using the scaling conditions (13) one finds that the
continuum limit Gc (X, Y ; T ), defined by the same procedure as in Section 2, fulfills

Gc (X, Y ; T ) + X2
Gc (X, Y ; T ) + 2XGc (X, Y ; T ) = 0
T
X
209
with the initial condition

1
.
Gc (X, Y ; 0) =
X+Y
These equations are identical to those found for a Lorentzian gravity model in Ref. [4] with
one marked link in the entrance loop. Consequently, the two continuum limits coincide, as
claimed. For later reference we note that the solution in terms of the length variables is [4]

L
2 LL

(L+L ) coth T

Gc (L, L ; T ) =
(19)
I1
.
L sinh( T )
sinh( T )
4. Closed, homogeneous spacetime and tensor product models
Marking a link in a boundary loop is a convenient technical device in triangulated
models and the relation to the same model with no marking is simple, since the marking
only gives rise to a factor equal to the length of the marked loop in the counting of
triangulations. The relation is of a different kind for Hamiltonian models but, as we will
immediately see, still quite straightforward.
In spatially homogeneous models, all links in an equal-time slice have identical status,
meaning that only one type of creation and annihilation operators is involved. The simplest
nearest neighboring Hamiltonian is then given by

2
H1 = Tr a a + a a + a a 2
(20)
N
N
with N = and a real parameter. Comparing Eqs. (16) and (20), we see that removing
the marked link restores not only cyclic symmetry but also Hermiticity.
The Hilbert space T1 on which H1 acts is spanned by all states of the form
|n1 =
1
N n/2
n
Tr a |,
where n is a positive integer. In the large-N limit [8,12],

lim m|n1 = nmn .
Hence |11 , |21 , . . . , and so on form an orthogonal basis for T1 . We think of |n1 as the
quantum state of a closed one-dimensional universe topologically equivalent to a regular
polygon with n sides (cf. the string bit interpretation in Refs. [6,7]). In terms of the
operators introduced in Eq. (17), the Hamiltonian H1 can be rewritten as
H1 = g11 + g12 + g21 .
In the large-N limit [8,12],
g11 |n1 = n|n1
if n 0,
g12 |n1 = n|n + 11 if n 0,

g21 |11 = 0 and g21 |n1 = n|n 11
if n > 0.
210
Here g11 is again a volume energy term, g12 splits any link into two and g21 combines any
pair of adjacent links into one.
It turns out to be tricky, if not impossible, to apply the previously used generating
function technique to work out the transition amplitude of this Hamiltonian in the
continuum limit. Instead, we will derive it by diagonalising H . Before doing so, however,
we will digress for a moment and make some observations about the underlying Lie
algebras of string bit models. This will lead to the introduction of a more general and
abstract class of Hamiltonian models including the one just defined as well as the model in
Section 2 and, in the continuum limit, the tensor product type models of Ref. [5] as special
examples.
It was shown in Ref. [13] that if we take T1 as the defining representation of the Lie
1 generated by all g k s, then under the identifications3
algebra C
l
g11 L0 ,
g21 L1 ,
g12 L1 ,
they satisfy the Lie brackets

[L0 , L1 ] = L1 ,

L0 , L1 = L1 ,

L1 , L1 = 2L0 ,
(21)
which the reader may verify directly and easily. This Lie algebra is nothing but the sl2
subalgebra of the Virasoro algebra. Furthermore, since
g21 |11 = 0,
g11 |11 = |11 ,
1|11 = 1,
|11 plays the role of the highest weight vector in the defining representation, and the
highest weight h = 1. It should thus be natural for us to consider the model in which the
Hamiltonian
H = L0 + L1 + L1
is an element of sl2 and acts on a certain highest weight representation. Recall that for
general h > 0 the representation space Th is the Hilbert space spanned by the vectors
1 n
L |1h ,
(22)
n! 1
where |1h is the highest weight vector and n is any non-negative integer. The actions of
L1 , L0 , and L1 are given by
|n + 1h =
L0 |n + 1h = (n + h)|n + 1h

L1 |n + 1h = (n + 1)|n + 2h
if n 0,
if n 0,
L1 |1h = 0 and L1 |n + 1h = (n 1 + 2h)|nh
if n > 0.
(23)
3 There is an important difference between the way the Virasoro generators arose in Refs. [12,13] and the way
they arise here. In those two articles, every Virasoro generator Ln was identified as a coset of certain elements of
1,1 . These cosets satisfied the Witt algebra, i.e., the classical Virasoro
1 or
1 , a subalgebra of G
the Lie algebra C
1 or
1 . Here, on the other hand, L1 , L0 and
algebra. Therefore, the Witt algebra was a quotient algebra of C
1 or a variant of
1 . They turn out to form the sl2
L1 (and nothing else) are identified with specific elements of C
1 or the variant of
1 .
subalgebra of the Virasoro algebra. Therefore, sl2 is a subalgebra of C
211
The inner product on Th is uniquely determined by the commutation relations (21),

1|1h = 1, L0 = L0 , and L1 = L1 . In particular,
n + 1|n + 1h =
1(n + 2h)
.
n!1(2h)
(24)
We note that the inner product is positive-definite if and only if h > 0.

Next, we revisit the model in Section 2 for a spacetime with boundaries. Its Hamiltonian
H1/2 was given by Eq. (7). The following observation is a slight modification of the results
1 generated by all k s in Ref. [12]: make the identifications4
concerning the Lie algebra
l
1
11 l00 L0 ,
21 L1 ,
12 L1 .
2
Then the actions of L1 , L0 , and L1 on T1/2 satisfy the Lie brackets (21). We are thus
again led to the Hamiltonian H of the form (1). Since

1 0
1
1
1
2 |1 1 = 0,
1 l0 |1 1 = |1 1 ,
1|1 1 = 1,
2
2
2
2
2 2
|11/2 plays the role of a normalised highest weight vector.
Based on these observations, we will, in the following, consider Hamiltonians of the
form (1) for arbitrary positive highest weights h. It turns out to be possible to diagonalise
H for all such h and to evaluate the continuum limit of the transition amplitude. Before we
perform this task in the next section, a few remarks on the interpretation of these models
are in order.
As is well known from the representation theory of sl2 , its h = 1 highest weight
representation is the symmetric tensor product of two copies of the h = 1/2 highest weight
representation. This fact has a clear interpretation in terms of the gravity models as follows.
Taking two copies of the Hamiltonian H1/2 , we define the symmetrised tensor product
state
|n =
n

k=1
|k 1 |n k + 1 1
2
for n 1 and regard it as representing the state of a closed polygon obtained by gluing two
polygonal lines of total length n at both ends, where each of the four boundary links of
the two polygonal lines has a negative length of 1/4. The action of H1/2 1 + 1 H1/2
on the space spanned by the vectors |n will then be easily seen to equal to that of H for
h = 1 under the identification |n = |n. Similar remarks apply to the generators L1 , L0 ,
and L1 .
Corresponding considerations of tensor products and gluing constructions for Lorentzian gravity models were discussed in Ref. [5], as were extensions to multiple tensor
products. The latter lead to the so-called p-seamed transition amplitudes. As will also
be seen from the explicit solution in the next section, these are reproduced by our algebraic
model for integer values of 2h.
4 Ibid.
212
5. Solution to sl2 gravity model

In this section we will show how to obtain the continuum limit of the models
with Hamiltonian given by Eq. (1) for arbitrary h > 0. We will first prove that H is
diagonalisable and determine the exact energy spectrum. Then we will determine the
asymptotic form of the eigenvectors close to criticality, which will enable us to extract
the continuum limit.
As already remarked, Th is spanned by the vectors |1h , |2h , . . . , and so forth defined
by Eq. (22). These vectors being orthogonal, it follows that states in Th are given by
an |nh ,
(25)
n2h1 |an |2 <
(26)
n=1
where

n=1
by Eq. (24). Clearly, H is an unbounded operator defined, e.g., on vectors for which the
sequence an is rapidly decreasing.
5.1. Diagonalisation of H
We will apply a refined version of the Frobenius method used to solve a very similar
quantum matrix model in Ref. [12]. Assume, for E 0, that
=
an |nh
(27)
n=1
is an eigenstate of H . Using Eqs. (1) and (23), we may write the eigenvalue equation
H = E
as
(n + 2h)an+2 + (n + h E)an+1 + nan = 0,
for all non-negative values of n. (The value of the new unknown a0 is immaterial because
its coefficient is 0.) Asymptotically for large n, the equation reduces to
an+2 + an+1 + an 0,
whose solutions are of the form
an pn + pn ,
where
p=
1 +
1 42
2
(28)
213
in the continuum limit. Since |p| < 1 for || < 1/2, it follows that an must asymptotically
behave as pn in order that be normalisable in accordance with Eq. (26). Hence, we set
an = bn p n ,
resulting in the equation
(n + 2h)bn+2 p2 + (n + h E)bn+1 p + nbn = 0
for all non-negative values of n.
In terms of the increments
6bn = bn+1 bn
and 62 bn = 6bn+1 6bn ,
this equation may be rewritten as

(n + 2h)p62 bn + (2p + 1)n + (4hp + h E) 6bn
+ (2hp + h E)bn = 0.
Introduce the ansatz [15]
bn =
cr n(r) ,
r=0
where the real numbers cr depend not on n but on r only, and the factorial polynomial n(r)
is defined by

n(n 1) (n r + 1) if r > 0,
n(r) =
1
if r = 0.
From
6bn =
(r1)
rcr n
and 6 bn =
r=0
r(r 1)cr n(r2) ,
r=0
we then obtain the equation

p(r + 2)(r + 1)(r + 2h)cr+2 + (3r + 4h)p + (r + h E) (r + 1)cr+1

+ (1 + 2p)(r + h) E cr = 0
(29)
as a sufficient condition for the coefficients cr to fulfill for all values of r.

For a non-negative integer R, we now set
E = ER = (1 + 2p)(R + h)
(30)
and see that we obtain a unique solution for cr , up to constant multiples, for which cr = 0
for r > R. The corresponding eigenstate R is thus of the form
R =
CR (n)pn |nh ,
n=1
where CR (n) is a polynomial of degree R in n.
214
Since H is Hermitian, the found eigenstates form an orthogonal set. Moreover, it is a

complete set, which can be seen as follows. Take any vector in Th given by (25) and assume
it is orthogonal to all eigenvectors R . Since the CR (n)s span all polynomials, this means
that
an nS pn
n=1
1(n + 2h)
= 0,
n!1(2h)
where S is an arbitrary non-negative integer and we have used Eq. (24). Multiplying this
equation by zS /S! and summing over S give

n=1
an p n
1(n + 2h) zn
e =0
n!1(2h)
for |z| < log p. Obviously, the left-hand side is an analytic function of z in the half plane
(z) < log p and hence vanishes there. Restricting z to the imaginary axis, we obtain a
vanishing Fourier series, and consequently its Fourier coefficients vanish. This proves that
an = 0 for all n 1 and the completeness of R for all R 0 follows.
Thus Eq. (30) gives the whole energy spectrum. Note that in the limit (13),
ER a(h + R) 0
as a 0, so the model is well behaved in this limit.
5.2. Asymptotic behaviour of eigenstates
In order to determine the continuum limit of the transition amplitude we will need
the asymptotic behaviour of the polynomials CR (n) under the scaling conditions given in
Eq. (13). Since n scales as a 1 we need to exhibit the leading behaviour of the coefficients
cr in CR (n) as a 0. Make the ansatz that cr+1 and acr are of the same order in the
small-a limit. The recursion relation (29) then gives
cr+1
Rr
2 acr .
(r + 1)(r + 2h)
Iterating this equation yields

(2 a)r 1(2h) R
cr c0
r
1(r + 2h)
which, owing to the scaling of n = L a 1 , yields the asymptotic form

R

(2 a)r 1(2h) R
CR (n) c0
nr ,
r
1(r + 2h)
(31)
r=0
where all summands are of order 1.

The behavior of c0 is fixed by requiring that R be normalised. Using the fact that R
is orthogonal to all vectors in Th of the form (25) with an = ns pn , for s = 0, 1, . . . , R 1,
215
we get
R |R h =
CR (n)cR mR pn+m n|mh
n=1 m=1 r=1
CR (n)cR nR p2n
n=1
1(n + 2h)
.
n!1(2h)
(32)
Substituting Eq. (31) into Eq. (32) results in

R

(2 a)R+r 1(2h) R R+r 1(n + 2h) 2n
2
p
c0
n
1(R + 2h)1(r + 2h) r
n!
r=0
n=1
r=0
n=1

R

(2 a)R+r 1(2h) R 1(n + R + r + 2h) 2n
p ,

1(R + 2h)1(r + 2h) r
n!
where, in the last step, we have used that for k real
1(n + k)
1 as n .
n!nk1
By the binomial theorem and the relation
p2 1 2 a,
we finally obtain
c02

R

(2 a)R+r 1(2h) R 1(R + r + 2h)
1(R + 2h)1(r + 2h) r (1 p2 )R+r+2h
r=0

=
(2 a)2h
(1)
+ r + 2h)1(2h)
1(R + 2h)1(r + 2h)
R+r 1(R
r=0
R
r
R!1(2h)
1
,
2h
1(R
+ 2h)
(2 a)
(33)
where, in the last step, we have made use of the identity

R

r=0
(1)
+ r + 2h)
R!1(r + 2h)
R+r 1(R
R
r

= 1,
which is a special case of the ChuVandermonde identity. (See, e.g., Ref. [16].)
5.3. The continuum limit
We are now ready to compute the continuum limit of the transition amplitude. The
unnormalized transition amplitude is defined by

u (L, L ; T ) = lim a L et H L ,
G
(34)
a
a h
a0
216
where t and are given as in Eq. (13), and the exponent is to be determined such that
the limit exists. Recall from Eq. (24) that the states | La h are not normalised. On the other
hand, the more natural amplitude defined in terms of the normalised states may simply be
u by Eq. (24); we will come back to it later.
obtained from G
Inserting the complete set of states {R }, we have
u (L, L ; T ) = lim a
G
a0
lim a
a0

R=0
et ER
e2
L L

a R h R a h
L
(R+h) L

a R h R a h .
(35)
R=0
Using Eqs. (31) and (33) as well as
pL/a e
we have

L
R a h = CR La pL/a La La h

1(R + 2h) 2h1 (2 )r R
h
1h
Lr e L .
L
a
2
R!1(2h)
1(r + 2h) r
r=0
Inserting this expression into Eq. (35) and choosing

= 2h 2,
we find that the continuum transition amplitude exists and takes the form
u (L, L ; T )
G
2h1
= (4) (LL )
h
R R

1(R + 2h) (2 )r+s Lr Ls
R!1(2h)
1(r + 2h)1(s + 2h)
R=0
r=0 s=0

R
R (L+L ) 2(h+R)T
e
.
e
r
s
(36)
A priori, it is not obvious that this series is convergent for all positive values of L ,L ,
and T . One way to see this, and simultaneously obtaining a more manageable expression
u , is to apply an integral representation of the reciprocal Gamma function (see, e.g.,
for G
Ref. [17]):
1
1
=
1(x) 2i x1
+i

i
e(z+b) dz
.
(z + b)x
(37)
In this formula, x and are arbitrary positive numbers; is real and b complex, and they
satisfy (b) > ; and the branch cut of (z + b)x lies on the negative real axis if x is not an
integer. Apply Eq. (37) to 1/ 1(r + 2h) with
x = r + 2h,
= L,
b = 1,
z = X,
217
and positive and infinitesimally small, and to 1/ 1(s + 2h) with
b = 1,
z = Y,
x = s + 2h,
= L ,
and the same in Eq. (36). The binomial theorem can then be used to perform the
summation over r and s. Apply once more the binomial theorem to the sum over R, and
keep the branch cut of every [ ]2h on the negative real axis. With this choice of the branch
cuts, (z1 z2 )2h = z12h z22h if both (z1 ) and (z2 ) are positive. Using this fact, we finally get
u (L, L ; T ) = 4
G

1h
+i

1
2i
i
2i
[Y
+i

i

L Y e 2h T
+ 1 (Y 1)e2 T ]2h
dY e
dXe
LX
X+
Y +1+(Y 1)e2T
Y +1(Y 1)e2 T
2h .
Integrals of this type have been carried out for integer values of 2h in [5], but can be
obtained for arbitrary values of 2h. Indeed, apply Eq. (37) to the integration with respect
to X and we immediately get
h e 2h T L2h1
4

u (L, L ; T ) =
G
1(2h)2i

2 T
2 T
+i

exp L Y L Y (1+e2T )+(1e2T )
Y (1e
)+(1+e
)
dY
.
2
T
2
T
2h
[Y (1 e
) + (1 + e
)]
i
Inserting the expansion

2 T
2 T
exp L Y (1+e2T )+(1e2T )
Y (1e
)+(1+e
)

1+e2T
= exp L 2 T
1e

n
1 4 Le2 T
1
,
n! 1 e2 T Y (1 e2 T ) + (1 + e2 T )
n=0
applying Eq. (37) again to the remaining integration, and recalling that
I2h1 (z) =

z2n+2h1
n!1(2h + n)
n=0
is the (2h 1)-th modified Bessel function [18], we finally obtain

)h1/2
(LL
2 LL
) coth( T )

(L+L
u (L, L ; T ) =
G
I2h1
.
1(2h) sinh( T )
sinh( T )

Define the normalized transition function G(L,
L ; T ) by normalizing | La h in Eq. (34).

It follows from Eq. (24) that we have to choose = 1 and that G(L,
L ; T ) deviates from
218
u (L, L ; T ) by a factor of (LL )h1/2 / 1(2h). Consequently,

G

2 LL

(L+L ) coth T

.
I2h1
G(L, L ; T ) =
sinh( T )
sinh( T )
(38)
We note that if we put h = 1/2

in Eq. (38), we will obtain Eq. (15); for h = 1 we obtain
Eq. (19) except for the factor L/L , which originates from the non-Hermiticity of Hc ;
if h is an integer, Eq. (38) will coincide with the propagator calculated in the proper-time
gauge of two-dimensional quantum gravity in Ref. [11] with the winding number h 1;
finally, if 2h is an integer, Eq. (38) will be exactly the (2h)-seamed correlation function in
Ref. [5].
6. Discussion
In this paper we have investigated different Hamiltonian models of two-dimensional
quantum gravity. Clearly, one open problem is the proper physical interpretation for the
sl2 gravity model with a non-integer value of 2h. Perhaps this describe the interaction
of gravity with matter, an important future problem on its own right. One could, for
instance, study one-dimensional quantum spin systems whose Hamiltonians couple spin
configurations of different sizes. Diagonalisation of such Hamiltonians seems to pose
interesting new problems. Another possibility is to extend the class of Hamiltonians defined
in this paper by exploiting the representation theory of the full Virasoro algebra instead of
the sl2 subalgebra; such models might describe the coupling between matter and gravity.
Finally, extension to higher dimensional Hamiltonian models of quantum gravity is
an ultimate goal. It is rather straightforward to produce candidate models of nearest
neighbouring type, but extracting non-trivial information from such models seems a nontrivial task. We refer to Ref. [19] for a recent work on higher dimensional Lorentzian
gravity.
Acknowledgement
We thank J. Ambjrn and H.P. Jakobsen for discussions.
References
[1] V.G. Knizhnik, A.M. Polyakov, A.A. Zamolodchikov, Mod. Phys. Lett. A 3 (1988) 819;
F. David, Mod. Phys. Lett. A 3 (1988) 1651;
J. Distler, H. Kawai, Nucl. Phys. B 321 (1989) 509.
[2] J. Ambjrn, B. Durhuus, T. Jnsson, Quantum Geometry, Cambridge Monographs on Mathematical Physics,
Cambridge Univ. Press, 1997.
[3] A. Ashtekar, R.P. Geroch, Rep. Prog. Phys. 37 (1974) 1211.
[4] J. Ambjrn, R. Loll, Nucl. Phys. B 536 (1999) 407, hep-th/9805108.
[5] P. Di Francesco, E. Guitter, C. Kristjansen, Nucl. Phys. B 567 (2000) 515, hep-th/9907084.
219
[6] R. Giles, C.B. Thorn, Phys. Rev. D 16 (1977) 366;

C.B. Thorn, in: Proceedings of Sakharov Conference on Physics, Moscow, 1991, pp. 447454, hepth/9405069.
[7] I. Klebanov, L. Susskind, Nucl. Phys. B 309 (1988) 175.
[8] C.B. Thorn, Phys. Rev. D 20 (1979) 1435.
[9] C.-W.H. Lee, S.G. Rajeev, Phys. Rev. Lett. 80 (1998) 2258, hep-th/9711052.
[10] R.J. Baxter, Exactly Solved Models in Statistical Mechanics, Academic Press, 1982.
[11] R. Nakayama, Phys. Lett. B 325 (1994) 347, hep-th/9312158.
[12] C.-W.H. Lee, S.G. Rajeev, Nucl. Phys. B 529 (1998) 656, hep-th/9712090.
[13] C.-W.H. Lee, S.G. Rajeev, J. Math. Phys. 39 (1998) 5199, hep-th/9806002.
[14] P. Di Francesco, E. Guitter, C. Kristjansen, hep-th/0010259.
[15] M.R. Spiegel, Schaums Outline of Theory and Problems of Calculus of Finite Differences and Difference
Equations, McGrawHill, 1971.
[16] M.E. Larsen, E.S. Andersen, Normat 42 (1994) 116.
[17] J.W. Dettman, Applied Complex Variables, Dover Publications, 1984.
[18] G.N. Watson, A Treatise on the Theory of Bessel Functions, Cambridge Univ. Press, 1945.
[19] J. Ambjrn, J. Jurkiewicz, R. Loll, Phys. Rev. Lett. 85 (2000) 924, hep-th/0002050.

Two-photon mediated resonance production

in e+e collisions:
cross sections and density matrices
F.A. Berends a , R. van Gulik a,b
a Instituut-Lorentz, University of Leiden, P.O. Box 9506, 2300 RA Leiden, The Netherlands
b NIKHEF, P.O. Box 41882, 1009 DB Amsterdam, The Netherlands
Received 16 October 2001; accepted 30 November 2001
Abstract
Earlier described model amplitudes are used in this paper to evaluate both cross sections and
density matrices for two-photon mediated resonance production in e+ e collisions. All 25 q q lowlying 1 S0 , 3 PJ and 1 D2 resonances can thus be treated. Two independent methods are described
to obtain the resonance production density matrices and cross sections. These density matrices
combined with a resonance decay density matrix give the detailed angular distributions of the
resonance decay products. For two particular decays, c2 , c1 J / the details are given. Several
numerical results are presented as well. 2002 Elsevier Science B.V. All rights reserved.
PACS: 12.39.Jh; 13.20.Gd; 13.40.Hq; 13.60.-r
Keywords: Two-photon; Quarkonia; Decay distributions
1. Introduction
When performing electronpositron collision experiments, there will always be data
originating from (virtual) two-photon collisions. In the latter process single resonance
production occurs. When present and future experiments collect more and more statistics
it becomes worthwhile to have theoretical predictions available for observables, which
gradually become experimentally accessible.
In a previous paper [1], hereafter called I, a specific class of such observables, namely,
form factors were studied. In this paper other observables will be focussed upon, such as
E-mail address: gulik@lorentz.leidenuniv.nl (R. van Gulik).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 1 9 - 8
F.A. Berends, R. van Gulik / Nuclear Physics B 623 (2002) 220246
221
azimuthal dependences in cross sections and resonance production density matrices. The
latter quantity combined with a resonance decay density matrix will give the distribution
of the decay products of the resonance.
In order to study the above observables one needs a model for resonance production.
In I such a model was described starting from heavy quarks in a non-relativistic bound
state. The predicted two-photon decay widths of resonances R turn out to be in reasonable
agreement with experiment. Once certain modifications are made, also the widths for
light quark mesons become reasonable. Therefore, the model of I is chosen as basis to
predict observables for the 25 low-lying 1 S0 , 3 PJ and 1 D2 resonances. In paper I analytic
amplitudes for R are presented and expressions for form factors are derived.
A number of numerical results for cross sections and Q2i dependences can also be found
there.
In the present paper numerical studies of azimuthal distributions will be given. Their
characteristic features can be understood from various analytic expressions. Whereas this
discussion is a simple extension of the results of I, the evaluation of production density
matrices requires substantial changes in calculational techniques. Moreover, rotations are
required between reference frames favoured by theory and experiment. A discussion of
two different evaluation methods will be given in this paper. Finally, in order to make a
link to the angular distribution of resonance decay products, also one specific decay mode
for two particular resonances will be discussed.
Summarizing, the purpose of the present paper is to discuss theoretical evaluation
methods for resonance production density matrices and to present results thereof. Thus
the full predictions of the resonance production model of I are in this way completed.
The paper is organized as follows. Section 2 gives the relevant amplitudes for resonance
production. The direct method of the evaluation of the observables is discussed in
Section 3, whereas Section 4 gives an extension of the BGMS [2] formalism required
for the evaluation of density matrices. After discussing a specific decay model for two
resonances in Section 5, numerical results for several quantities are presented in Section 6.
2. The starting point

In this section the basic ingredients of the evaluation are listed, such as kinematics and
matrix elements following from the model of I.
The two-photon mediated production of a resonance R in electronpositron collisions
is the reaction
e+ (p1 ) + e (p2 ) e+ (p1 ) + e (p2 ) + (k1 ) + (k2 )
e+ (p1 ) + e (p2 ) + R(pR ).
(1)
The four-momenta p1 = (E1 , p1 ) and p2 = (E2 , p2 ) correspond to the incoming positron
and electron, respectively,1 whereas p1 = (E1 , p1 ) and p2 = (E2 , p2 ) are those of the
1 We use the standard metric diag(g ) = (+1, 1, 1, 1). The totally antisymmetric Levi-Civita tensor is
defined by 0123 = +1.
222
outgoing positron and electron. The four-momenta of the intermediate photons k1 =

(1 , k1 ) and k2 = (2 , k2 ) are related to the external four-momenta by
ki = pi pi .
(2)
These two photons subsequently react to form a resonance with four-momentum pR

pR = k1 + k2 = p1 + p2 p1 p2 .
(3)
The virtuality Qi of an intermediate photon is defined by

Q2i ki2 = (pi pi )2 .
(4)
As the virtual photons are space-like, the sign in (4) makes the virtualities Qi real.
The invariant mass of the two-photon system is given by
2
= (k1 + k2 )2 .
W = pR
The total available energy
(5)
s follows from
s = (p1 + p2 )2 .
(6)
s = (p1 + p2 )2 ,
(7)
u = (p1 p2 )2 ,
u = (p2 p1 )2 ,
(8)
For a typical two photon event W is only a small fraction of s.

Some additional invariants can be defined
and as equivalents of
t = (p1 p1 )2 ,
t
(9)
Q2i
= (p2 p2 )2 .
(10)
(11)
These invariants satisfy

s + s + t + t + u + u = W 2 + 8m2e ,
(12)
where me is the electron mass.

In the lab-frame we chose the following parametrization
p1 = (Eb , 0, 0, Pb ),
p2 = (Eb , 0, 0, Pb ),

p1 = E1 , |p1 | sin 1 cos 1 , |p1 | sin 1 sin 1 , |p1 | cos 1 ,

p2 = E2 , |p2 | sin 2 cos 2 , |p2 | sin 2 sin 2 , |p 2 | cos 2 ,

pR = ER , |pR | sin R cos R , |pR | sin R sin R , |pR | cos R .
(13)
This means that the z-axis is along the direction of the incoming electron. The x- and
y-axes are chosen such that the xz-plane is the accelerator ring plane and the y-axis is
perpendicular to that plane and is pointing upwards. In these formulae Eb is the energy of
223
the incoming leptons and Pb their momentum. The choice of parameters in (13) is such
that we have defined the z-components of the positron with an explicit minus sign.
This parametrization leads to explicit expressions for the invariants
s = 4Eb2 ,

Q2i = 2 Eb Ei Pb |pi | cos i m2e
2Eb Ei (1 cos i ) = 4Eb Ei sin2
(14)

i
,
2
2m2e + 4(Eb E1 )(Eb E2 ) 2E1 E2 2|p1 ||p2 | cos 12

12
.
2 1 2 E1 E2 cos2
2
W=
(15)
(16)
The approximations are valid in the limit Eb > Ei me . In formula (16) the angle 12 is
the angle between the outgoing leptons in the lab-frame
cos 12 = cos 1 cos 2 sin 1 sin 2 cos(1 2 + ).
(17)
The propagators of the intermediate photons will appear as a product of the photon
virtualities in the denominator of the matrix elements. This is the most dominant
dependence on the photon virtualities. From (15) it follows that the photons are most likely
radiated with low energy under a small angle with respect to the incoming lepton.
When the incoming lepton scatters over a large enough angle, so that it is detected in one
of the (forward) detectors, this lepton is referred to as a tagged lepton. From the detected
lepton the four-momentum, and thus the virtuality, of the associated intermediate photon
can be reconstructed.
The two methods of calculation both start from the cross section formula
d3 p1
d3 p2
(2)4 (4) (k1 + k2 P ) 1

d,
|M|2
d =
(18)
(2)3 2E1 (2)3 2E2
4 (p1 p2 )2 m4e 4
with d the invariant phase space element for the resonance. The amplitude M has the
structure
M=
e2
j j M ,
tt 1 2
(19)
with
j1 = j1 (1 , 1 ) = v1 (p1 ) v (p1 ),
(20)
j2 = j2 (2 , 2 ) = u 2 (p2 ) u2 (p2 ),
(21)
or explicitly in the model described (cf. also [3]) in I2

c1 e 2
M 1S 0 ; 1 , 2 , 1 , 2 = k1 , k2 , j1 (1 , 1 ), j2 (2 , 2 ) ,
tt
2 In these equations we have introduced the shorthand notation (p, q, r, s) p q r s .

(22)
224

M 3P 0 ; 1 , 2 , 1 , 2
c2 e2
=
j1 (1 , 1 ) j2 (2 , 2 )k1 k2
tt

j1 (1 , 1 ) k2 j2 (2 , 2 ) k1 W 2 + k1 k2

j1 (1 , 1 ) j2 (2 , 2 )tt ,

M 3P 1 ; 1 , 2 , 1 , 2 , R
c3 e 2
= t (R ), j1 (1 , 1 ), j2 (2 , 2 ), k2
tt

+ t (R ), j2 (2 , 2 ), j1 (1 , 1 ), k1 ,

M 3P 2 ; 1 , 2 , 1 , 2 , R
c4 e 2
k1 k2 j1 (1 , 1 )j2 (2 , 2 ) + k1 k2 j1 (1 , 1 ) j2 (2 , 2 )
tt
k1 j2 (2 , 2 )j1 (1 , 1 ) k2

k2 j1 (1 , 1 )j2 (2 , 2 ) k1 (R ),

M 1D 2 ; 1 , 2 , 1 , 2 , R
(23)
(24)
c5 e2
(R )k1 k2 k1 , k2 , j1 (1 , 1 ), j2 (2 , 2 ) ,
tt
(25)
(26)
where
4g1
,
c3 = 2 6 g1 ,
c2 =
c1 = g0 ,
W
c4 = 4 3 Wg1 ,
c5 = 8 30 g2 .
Here we have introduced
gi =
16eq2 |R(i) (0)|

(s + s + u + u 8m2e )i+1
3
,
W
(27)
(28)
with R(i) (0) the (derivates of the) radial part of the wave function in the origin and eq the
fractional quark charge. The actual values are given in I.
Replacing the currents j1 , j2 by 1 and 2 , the polarization vectors of a virtual photon,
will give the matrix element of the two-photon reaction, i.e., the second step in reaction
(1). That type of matrix element can be simplified by choosing convenient axes to describe
the photon polarization. It will be the basis of the method of Section 4. In the expressions
terms of the form i ki or ji ki are omitted, since they vanish in our calculations.
3. Direct method of calculation

In this section one way of evaluating the cross section (18) and the related density matrix
will be discussed. The motivation to deal with another method in the next section is that we
want to obtain the numerical results for observables in two completely independent ways.
225
The method of this section is a 2 3 particle cross section calculation without using the
two-photon
intermediate step. For the differential cross section alone it would be sufficient
to express |M|2 in terms of the invariants (6)(11). In fact, those have been obtained
and are given elsewhere [4]. It makes a very fast numerical evaluation of the cross section
possible. It is natural to perform this calculation in the laboratory frame as specified in
Eq. (13).
For the density matrices it is more practical to use a rest system of the resonance. Every
experimental event can be rotated and boosted back to the resonance rest frame (RRF) by
successively rotating the momentum pR in the direction of the z-axis and boosting it to
rest. Thus the required Lorentz transformation is
0
LLR = LB L L =
0
cos sin
cos cos
sin
cos sin
sin sin
sin cos
cos
sin sin
cos
sin
. (29)
0
cos
In order to calculate the average density matrix in the RRF belonging to a cross section
in a certain kinematical region in the laboratory system one proceeds as follows. In the
laboratory system one generates weighted events. For this we use the event generator
GaGaRes, which is described elsewhere [4]. The event is boosted to the RRF with LLR . For
the momenta in the RRF the amplitude Ai , is calculated, where i are the lepton helicites
and the resonance helicity. From these amplitudes the normalized density matrix

i ,R
Ai , Ai ,
Ai ,R Ai ,R
(30)
is evaluated for this event. By performing the weighted sum of these individual density
matrices and dividing by the total weight one obtains the average density matrix for the
production of a resonance. Of course, when events with weight 1 are generated the above
procedure becomes simpler. Sometimes unnormalized density matrices are needed, i.e.,
just the numerator of (30).
It is clear that a method to evaluate amplitudes is needed, preferably a fast technique. For
this the Weylvan der Waerden (WvdW) spinor calculus is a good tool, as has been shown
in the literature [5]. Once Dirac spinors, momenta and polarization vectors have been
translated into WvdW spinors, the matrix elements (22)(26) become spinorial quantities.
There now are two ways to continue. The amplitudes can be expressed in terms of
spinorial inner products. Evaluating numerically these products and their combinations
one obtains the amplitudes as a complex number. Another way is to interpret the spinorial
matrix element as the trace of a string of 2 2 matrices, which can be calculated
numerically very fast thanks to efficient matrix multiplication in F ORTRAN90. The latter
method has been used in our numerical calculations, whereas the former turned out to be
useful as independent check. More about the spinorial translation and procedure can be
found in [4]. Here we only mention one particular detail, relevant for comparisons and
cross checks.
226
Usually the spin-1 polarization vectors for a particle with momentum

sin cos , |k|
sin sin , |k|
cos
k = k 0 , |k|

, |k|ss
, |k|c
,
= k 0 , |k|sc
are chosen to be
1
= (0, cc + is , cs ic , s),
2

0 |k|

k
0 =
, sc , ss , c ,
m k0
(31)
(32)
for helicities 1 and 0. The spin-2 polarization tensors then usually are taken as
(2) = (1) (1),

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

1
(0) = (+1) (1) + 2 (0) (0) + (1) (1) .

(33)
6
Of course this follows a certain convention and other choices may be made as can be seen
in the literature. In the implementation of the WvdW formalism the conventions of [6] have
been used. Unfortunately the polarization vectors and tensors in this reference are different
ei
(k) = (0, cc is , cs ic , s),

2

0 |k|

k
,
sc
,
ss
,
c
,
0 =
m k0
(34)
(2) = (1) (1),

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

1
(0) = (+1) (1) + 2 (0) (0) (1) (+1) .
(35)
6
When we label the sets of polarization vectors and tensors of Eqs. (32) and (33) with the
index A and the polarization vectors and corresponding tensors in Eqs. (34) and (35) with
an index B, the two sets are related by
B, = ei A, ,
B (1) = ei A (1),
B,0 = A,0 ,
B (0) = A (0),
B (2) = e2i A (2).
(36)
These formulae give the relations between density matrices in the two conventions. When
an event has a density matrix A , evaluated with the set A, then B can be obtained by
multiplying each matrix element with a certain phase factor. In a compact notation these
factors are summarized as follows
,B = ei6 ,A ,
(37)
with for the spin-1 density matrix elements

0
2 +
6 =
0
+ .
(2 + ) ( + )
0
For the spin-2 density matrix elements we find analogously the set of phase factors
2
3 4
0
2 3

6 = 2
0
2 .
3 2
0
4 3 2 0
227
(38)
(39)
4. Extended BGMS formalism

A long time ago in a classic paper [2] it was advocated to write the cross section of
reaction (1) as a two-step process. First virtual photons characterized by a density matrix
i are created which in a second step produce the resonance R. In this BGMS formalism
the cross section for resonance production (18) takes in first instance the form

M M d3 p1 d3 p2
2
(4)
d = 2 2 1 2 (k1 + k2 P )
(40)

d,
Q1 Q2
(p1 p2 )2 m4e 2E1 2E2
where

i

1
2
1 2

=
.
j j = 2 pi pi + pi pi Qi g
2
2Q2i
Qi
(41)
The index summation in Eq. (40) is now replaced by a summation over photon helicities
i and i with values 1, 0. Introducing the quantity

1
M ,R M1 2 ,R d,
M1 2 1 2 = (2)4 4 (k1 + k2 pR )
(42)
1 2
2
R
the cross section can be written as

d =
1 2
dp1 dp2
2
1
1
2

M
.
1 2 1 2
1
2
E1 E2
32 4 Q21 Q22 (p1 p2 )2 m4e
(43)
It should be noted that M 1 2 is evaluated in the two-photon rest system, the BGMS
1 2
frame, where the direction of the photon originating from the positron is taken as z-axis.
The polarization vectors of both photons now become very simple as can be seen from
(32). Also R is completely fixed by
R = 1 2 ,
and the summation in (42) consists of only one term.
(44)
228
The expression for the unnormalized density matrix of an event is obtained from the
cross section formula when one introduces another combination M.

1 2
=
(45)
1 1 2 2 M1 2 1 2 .
1 ,2 , ,
1 2
=1 2
=1 2
It should be noted that for a specific only specific 1 , 2 combinations can contribute.
Using the explicit form of the photon density matrices [2] where only a restricted number
of elements is independent and using
M1 ,2 ,R = R M1 ,2 ,R ,
(46)
where R = 1(1) for the normal (abnormal) J P series, one obtains expressions for
in terms of M1 2 1 2 . For every expression also new quantities are introduced,
generalizations of the quantities AB and AB of [2] with the help of
X = (k1 k2 )2 k12 k22 .
(47)
The results are

2+2+ = 22 = 1++ 2++ M++ = 4 X 1++ 2++ TBT ,
(48)
+0 +0
++ 00
00 ++
++ = = 1 2 M+0+0 + 1 2 M0+0+ 21 2 M0+0 cos()

++ 00
00 ++
= 2 X 1 2 T S + 1 2 ST

TBS ,
(49)
41+0 2+0 cos()

++
00 = 21++ 2++ M++++ + 100 200 M0000 + 21+ 2+ cos(2)M
+0 +0
00++
41 2 cos()M

++ ++ A
TT
= 2 X 41 2 T T + 100 200 SS + 21+ 2+ cos(2)
+0 +0

A .
8 cos()
(50)
1
TS
The various and definitions follow from (48)(50). In BGMS the following
combinations are used
1
T T TAT + TBT =
(M++++ + M++ ),
4 X
(51)
1
T S TAS + TBS =
(M++00 + M00+ ).
4 X
(52)
In the formulae is the angle between the two lepton scattering planes in the BGMS frame.
For the off-diagonal elements we do the same. The results are summarized below

+ = e2i 1 21+0 2+0 ei M0++0

(53)
1+ 200 M0+0 100 2+ e2i M0+0 ,
229

+0 = iei 1 1++ 2+0 ei M+++0 1+0 200 M00+0 + 1+ 2+0 M+0 ei

1+0 2+ e2i M0+ + 100 2+0 ei M000+

1+0 2++ M++0+ .
(54)
For the spin-2 resonances one has additionally

2+1+ = iei 1 1++ 2+0 ei M+0+ 1+0 2++ M0+ ,

2+0 = e2i 1 1++ 2+ e2i M+++ 1+0 2+0 ei M00+

+ 1+ 2++ M+ ,

2+1 = ie3i 1 1+ 2+0 ei M0+ + 1+0 2+ e2i M0++ ,

2+2 = e4i 1 1+ 2+ e2i M++ .
(55)
(56)
(57)
(58)
Besides the dependence, for the off-diagonal elements there is also an overall 1
dependence, 1 being the azimuthal angle of the incoming positron in the BGMS frame.
In analogy with the and the terms we can introduce
i
+0 =
M+++0 ,
2 X
i
+0 =
M00+0 ,
2 X
1
++ =
M+++ ,
2 X
i
+0 =
M+0+ ,
2 X
i
0+ =
M++0+ ,
2 X
i
0+ =
M000+ ,
2 X
1
00 =
M00+ ,
2 X
i
0+ =
M0++ .
2 X
(59)
(60)
(61)
(62)
The functions are only non-vanishing for spin-2 resonances. For all above functions
AB , . . . , AB holds: when AB M1 2 1 2 then BA M2 1 2 1 .
The off-diagonal elements can now be written as

+ = e2i 1 X 41+0 2+0 ei R TBS

(63)
1+ 200 R T S 100 2+ e2i R ST ,

+0 = ei 1 X 1++ 2+0 ei +0 1+0 200 +0 + 1+ 2+0 R ei +0

1+0 2+ e2i R 0+ + 100 2+0 ei 0+

(64)
1+0 2++ 0+ ,

2+1+ = ei 1 X 1++ 2+0 ei +0 1+0 2++ R 0+ ,
(65)

2+0 = e2i 1 1++ 2+ e2i ++ 1+0 2+0 ei 00

+ 1+ 2++ R ++ ,
(66)
230

2+1 = e3i 1 X 1+ 2+0 ei R +0 + 1+0 2+ e2i 0+ ,

2+2 = e4i 1 e2i X 1+ 2+ R TBT .
(67)
(68)
In these formulae R is the phase factor arising in Eq. (46). The functions and 00 vanish
for the 3 P1 resonance (M00 and M 00 vanish).
The other off-diagonal elements can be obtained by using that the density matrix is
Hermitian
= ,
(69)
and by using that for the polarization vectors in the BGMS formalism the following identity
holds

= (1)+ .
(70)
The unnormalized density matrix after integration over the phase space of the outgoing
leptons, denoted by int , is then given by

d3 p1 d3 p2
1
2
int

(71)
=
E1 E2
32 4 Q21 Q22 (p1 p2 )2 m4e
The trace should equal the total cross section. When we use the expressions for the diagonal
elements we indeed obtain the BGMS expression

(k1 k2 )2 k12 k22
2
d =
16 4 Q21 Q22 (p1 p2 )2 m4e

+ 2 ++ 200 ST
41++ 2++ T T + 21+ 2+ T T cos(2)
1
d3 p d3 p

1
2
TS
+ 2100 2++ T S + 100 200 SS 81+0 2+0 cos()
.
E1 E2
(72)
The explicit expressions for quantities AB , AB , AB , AB and AB follow from the

amplitudes M1 ,2 ,R as given in Table 1 of I.
In order to have some analytic results we repeat and extend form factor definitions of I
and give tables with their forms. Thus we introduce

(2J + 1) (J P )
fAB J P ,
AB = P 2 M 2 8 2
M
2

(2J + 1) (J P )
AB = P M 2 8 2
gAB J P ,
M
P

2
(2J
+
1)
(J )
kAB J P ,
AB = P M 2 8 2
M
P

2
(2J
+
1)
(J )
AB = P M 2 8 2
mAB J P ,
M
P
2

(2J
+
1)
(J )
AB = P M 2 8 2
nAB J P .
M
(73)
(74)
(75)
(76)
(77)
231
Table 1
The form factors required for the cross section fAB , gAB , (A, B = T , S)
JP
fTAT
fTBT
fT S
X2
fSS

2
X+M
3 2
Q2 Q2 2
12 2
2 2
M
2
3
X2
JP
gT T
gT S
2 X2
X+M 2
2
3 2
Q21 Q22 2
2
2
2 Q2 (Q2 +Q2 )

2Q
1 2
1
2 2
2
12
X
3
2 2
0

2
M2 Q1 Q2 (X + M 2 )
3

2
M2 Q1 Q2 ( + Q21 )( + Q22 )
0+
1+
2+
0+
1+
2+
2
0
2 2 2
6M 4 2
M 2 Q22 (Q21 )2
4 4
M 4 Q21 Q22
3 4
3 2
2
2
M 2 Q2 +Q1 2
0
M 2 2 [2Q21 Q22 (Q21 +Q22 )]2
0

M 2 Q2 Q2 2
1 2
2 Q1 Q2 2

2 2
2
2
2
2
M2
8
3 (2Q1 Q2 (Q1 + Q2 )) + ( Q1 )( Q2 )
Table 2
The additional form factors required for the off-diagonal elements. kAB , mAB and nAB (A, B = +, 0)
JP
1+
2+
2
k+0
Q2
M
2 3 ( + Q21 )(Q21 Q22 )
MQ (Q1 )
2Q21 Q22 (Q21 + Q22 )
2
4
4 3
M 3 Q Q2
1 4 2 ( Q21 )
2 3
JP
2+
2
n++
m+0
0
1 M 2 2Q2 Q2 (Q2 + Q2 )

1 2
1
2
2 6 3
n+0
n00
M 3 Q2
42
( Q21 )
3
M4Q Q
1 3 2
6
The form factors are given in Tables 1 and 2, where the quantities , and X are defined
as
M2
= ,
2 X
= k1 k2 ,
X = X =
(78)
(79)
2
k12 k22 .
(80)
It should be stressed that for every event the matrix can now be evaluated. Although
it is calculated in the two-photon rest frame, which is a frame where the resonance is at
rest, it is not the RRF of the previous section. For every event one has to relate the axes by
a rotation which gives rise to a transformation of the density matrix. The transformation
232
matrices are given in Appendix A. When one likes to compare the average density matrix
to the one of the previous section, one should also perform the change (37) for every event.
With the above procedure we extend the existing BGMS cross section Monte Carlo
program Galuga [7] to a program which can calculate the density matrix . We then
evaluate for every event generated by this program the density matrix which is then
transformed to the experimentally relevant density matrix of Section 3. In this way Galuga
can evaluate the wanted density matrix. Moreover, we thus have an independent check on
the GaGaRes evaluation.
5. Example of a decay model and density matrix

In general the produced resonance R will decay into some final state X, so that the
reaction is
e+ e e+ e R e+ e X.
The total amplitude for this process can be written as

M=
(P , M)AR DR .
(81)
(82)
AR describes the two-photon production of a resonance with helicity R . DR describes

the decay of the resonance with helicity R into the final state X. The factor (P , M)
represents the propagator of the resonance and numerical factors. The total matrix element
still depends on the external four-momenta. No implicit integrations over the outgoing
momenta have been carried out at this stage. The square of the total matrix element is
given by

2

2
|M|2 = (P , M)
(83)
AR R DR R = (P , M) Tr(AD ).
R ,R
The summation on the left-hand side represents a summation over the helicities of the
initial and final state particles. The quantities AR R and DR R are AR A and DR D ,
R
R
summed over the helicities of all particles but the resonance. These are the density matrices
for the production and decay of the resonance. For a spin-J resonance the density matrices
are formed by (2J +1)(2J +1)-matrices. For actual calculations a choice of polarization
vectors and tensors has to be made. The results of this paper are obtained in the conventions
(34) and (35).
For R we first take the spin-2 state c2 and consider the specific decay mode
c2 (pR ) (k1 )J /(k2 ).
(84)
Later on also the possibility of the J / decaying into a lepton pair will be discussed. The
choice of this example is prompted by its experimental relevance. Since we expect that a
similar c1 decay may cause a contamination of the events (84) in an actual experiment we
shall also discuss the c1 decay at the end of this section.
233
For the amplitude M of the decay of the spin-2 resonance we use the amplitude that is
given in Eq. (25)

M(, 1 , 2 ) = c4 (k1 k2 )1 (1 )2 (2 ) + 1 (1 ) 2 (2 ) k1 k2

k1 2 (2 ) 1 (1 )k2 k2 1 (1 ) 2 (2 )k1 ()
= c4 F1
where Fi
(1 )F2
(2 ) (),
(85)
is the field strength
Fi (i ) = ki i (i ) ki i (i ).
(86)
In these equations the index 1 refers to the photon and 2 refers to the massive J /. The
s refer to the helicity of the associated particles. The parameter c4 is not specified here,
but its complex square will contain the width for the decay of the spin-2 particle into these
two spin-1 particles. It will be related to c4 .
The density matrix D for the decay of the resonance is then constructed by

D =
(87)
M(, 1 , 2 )M ( , 1 , 2 ).
1 ,2
In the evaluation of this density matrix we make use of the invariant spin summations for
spin-1 particles. For the massless photon this relation reads

(88)
1, (1 )1,
(1 ) = g ,
1
whereas for the massive J / it reads

k2, k2,
2, (2 )2,
(2 ) = g +
,
M2
(89)
where M is its mass. In fact, due to the gauge invariance of the field strength in Eq. (85)
only the g term in Eq. (89) contributes. This leads to the following expression for the
density matrix for the decay of the resonance

D = |c4 |2 (k1 k2 ) ()
( ) + 2 k1 () k1 k1 ( ) k1

+ k22 + 4(k1 k2 ) k1 () ( ) k1 .
(90)
In this equation a polarization tensor in between two dots indicates that one Lorentz index
has to be contracted with the four-vector on the left side of the tensor whereas the other
Lorentz index has to be contracted with the four-vector on the right side of the tensor, i.e.,
p A q A p q .
(91)
In the derivation we have used that the final state photon is massless and we have used the
properties of the polarization tensor to replace k2 in contractions with it by k1 .
234
The density matrix for the decay of the c2 is constructed in the rest frame of the
resonance with the z-axis as polarization axis and some x- and y-axes. The density matrix
will depend on the polar and azimuthal angles of the photon in this reference system. The
specific choice of the axes in the c2 rest system is at this point arbitrary. In the applications
they should be the same as for the c2 production density matrix. The polarization tensor
looks very simple and one can use Eqs. (34) and (35) to verify that the following useful
relations hold
() ( ) = ,
()
( ) .
(92)
In this frame the four-momentum of particle 1 can be parametrized by

k1 = |k|(1,
sin cos , sin sin , cos ) = |k|(1,
s cos , s sin , c).
(93)
It can be shown that with these definitions the density matrix for the decay of the spin-2
particle can be written in a compact form as the sum of three matrices

2
4 v v + k22 + 4(k1 k2 ) |k|
R
D = |c4 |2 (k1 k2 )2 + 2|k|
2

2 2
1 2
2 MR M
+ v v + (1 + )R .
= |c4 |
(94)
2
2
In this expression is the Kronecker symbol. The vector v is given by
v = (v2 , v1 , v0 , v1 , v2 )
2

2

2
s 2i
i 1
i s 2i
e , sce , 3c 1 , sce , e
=
.
2
2
6
The tensor R is given by (the indices run from 2 to 2)
2
cs i
s2 2i
e
0
s2
2e
2 6
cs i
2
2
sc
s 2i
e i
1+c
4
4e
2e
2 6
s 2 2i
2 +1
sc
3c
i
e
e i
6
sc
R =
26 e
2 6
2 6
2
s 2 2i
e i
0
sc
1+c
4e
4
2 6
s2 2i
0
0
e
cs2 ei
2 6
(95)
s2 2i
e .
2 6
cs2 ei
2
s2
0
(96)
In the second line of Eq. (94) the expansion parameter has been introduced
=1
M2
.
MR2
(97)
Others [8] use a simplified expression for the matrix element, based on an electric dipole
transition

M = c4 (k1 k2 )1 (1 )2 (2 ) k2 1 (1 ) k1 2 (2 ) ().
(98)
This leads to the following density matrix

4
1
2
v v

D = |c4 | (k1 k2 )2 2 k22 + 2(k1 k2 ) |k|
M

(k1 k2 )2 2
|k| R
+ k22 + 2(k1 k2 )
M2

2

2 2
2
2
2 MR M
v v + 1
R .
= |c4 |

2
4(1 )
4(1 )
235
(99)
In the limit of vanishing the density matrices of Eqs. (94) and (99) become equal. For the
actual , based on M = 3.09687 GeV, MR = 3.55618 GeV [9] and, therefore, 0.242,
there are some differences, e.g., for Eq. (94)
D22 + D22 1 + 1.570 c2 + 0.0190 c4 ,
(100)
and for Eq. (99)

D22 + D22 1 + 1.028 c2 0.00956 c4 ,
(101)
whereas for = 0 one would have

D22 + D22 1 + c2 .
(102)
These distributions can be compared to the experimentally observed decay distribution [11]
of the c2 produced in pp collisions,
D22 + D22 1+1.96 c2 +0.0142 c4 ,
(103)
where the central values of the coefficients are taken and the errors are omitted. In Section 6
some numerical results for the distributions will be presented.
One could also take into account the possible decay of the massive J / into two leptons
c2 (k) (k1 )J /(k2 ) (k1 )l + (p1 )l (p2 ).
(104)
The matrix element for this reaction can be obtained from Eq. (85) by replacing the
polarization vector of the J / by the lepton current
Jrs
= u r (p2 ) vs (p1 ),
(105)
where r and s denote the spins of the final state leptons. In the following the lepton mass
ml will be neglected, which is a reasonable approximation for this decay.
In the construction of the density matrix for this decay one has to use that for the lepton
current one finds

,
Jrs
Jrs = 4 p1 p2 + p1 p2 12 k22 g = 2 k2 k2 l l k22 g .
(106)
r,s
In this equation the vector l = p1 p2 has been introduced. This has been done as the term
proportional to k2 k2 will not contribute to the density matrix. For l we use the following
parametrization
l = (l0 , lx , ly , lz )
(107)
236
in the above chosen reference frame, where (93) holds. The density matrix now reads

2 ()
( )(k1 k2 )2 k22 + 2k22 + l 2 k1 () k1 k1 ( ) k1
D = |c|

(k1 k2 ) k1 () k1 l ( ) l + l () l k1 ( ) k1

+ 2(k1 k2 ) + k22 k1 () l k1 ( ) l

2
+ k22 + 4(k1 k2 )k22 + (k1 l)2 k1 () ( ) k1

+ (k1 k2 )(k1 l) k1 () ( ) l + l () ( ) k1

+ (k1 k2 )2 l () ( ) l .
(108)
In the derivation we have used that (k2 l) = 0. This density matrix can also be written in
a more compact form

v v (k1 k2 )|k|
2 v x + x v
2 (k1 k2 )2 k22 + 2k22 + l 2 |k|
D = |c|
2

w w
+ 2(k1 k2 ) + k22 |k|

2 2
2
2 R
+ k2 + 4(k1 k2 )k2 + (k1 l)2 |k|

S + S + (k1 k2 )2 T .
+ (k1 k2 )(k1 l)|k|
(109)
In this density matrix the coefficients can again be written as functions of

2

MR M 2 2 2
2

2
D = |c|
M + v v
v x + x v
2
2
4
2M

2
1
cos2 R
+ 2 w w + 1 + +
4(1 )
M

1
cos S + S + 2 T ,
M
2M 1
(110)
where a convenient expression

MR2 M 2
cos
(111)
2
has been used. The angle is the polar angle of the outgoing l + in the rest system of
the J / where the z-axis is given by the direction of the boost from the RRF to the rest
frame of the J /, i.e., opposite to the photon direction. In the expressions some additional
vectors and tensors have been introduced
1
sl+ i 1
e , lz sei + l+ c , s(lx cos + ly sin ) 2clz ,
w=
2
2
6
1
sl
i
e
,
(112)
lz sei + l c ,
2
2

1 2
1
1 2
l+ , l+ lz , lx2 + ly2 2lz2 , l lz , l
x=
(113)
,
2
2
6
k1 l =
sl ei
S
T =
slz i
2 e
12 lsz
sl+ i
e
2 6
l
s+
2 6
16 [slxy + 4clz ]
l
cz
2 6
sl i
4 e
2
cl
2
sl i
e
2 6
12 (lx2 + ly2 )
l+ lz
2
l lz
2
2
l
2 6
2 + lz2 ]
14 [|l|
llz
2 6
2
l
4
ls+

2 6
sl i
e
2 6
sl+ i
4 e
lcz

2 6
12 lsz
sl2z ei
0
sl+ i
e
2 6
cl2+
sl2+ ei
2
l+
2 6
l+
lz
2 6
2
l+
4
2 + 3lz2 ]
61 [|l|
l l
+z
llz
2 + lz2 ]
14 [|l|
2
l+
2 6
l+2lz
l2lz
12 (lx2 + ly2 )
2 6
2
l
2 6
2 6
237
(114)
(115)
In the expressions the variables l+ and l have been introduced

l+ = lx + ily ,
l = lx ily .
(116)
In addition, we have also introduced

lxy = lx cos + ly sin ,
lcz = 2cl slz e
(117)
(118)
ls+ = 2slz e cl+ ,

(119)
sl i
e + clz .
lsz =
(120)
2
The decay density matrix is now completely specified once the lepton momenta in the c2
rest system are inserted in l . It is sometimes convenient to parametrize l in terms of
decay angles and ,
i
s = sin ,
c = cos ,
c = cos ,
s = sin ,
(121)
l+
of the in the rest system of the J /.

The components of the four-momentum l in the RRF then become (c = cos , s =
sin , the angles of Eq. (93))

MR2 M 2
c ,
lx = M s c cc + s s c sc ,
2M

R

lz = M s c s + c c .
ly = M s c cs + s c c ss ,
l0 =
(122)
Since the c2 will be predominantly produced in helicity 2 states, we have a closer

look at those density matrix elements. In the limit of vanishing the sum of density matrix
elements D22 + D22 reads
2

MR M 2 2 2
1
D22 + D22 = |c|

2
M 2 + 2 w2 w2
+ w2 w2
2
M

1
+ 2 (T22 + T22 ) + R22 + R22 .
M
(123)
238
Inserting the expressions for the vectors and the tensors gives
2

2 2

l+ l
2 MR M
2
2
D22 + D22 = |c|
(124)
M 1+c 1
.
2
2M 2
Integrating over the angles associated to the outgoing leptons yields for the sum
D22 + D22 a result that is proportional to 1 + c2 , which is in agreement with the
previously found results for the decay of a c2 into J / and , Eq. (102) in the limit
= 0. Using the expressions for l+ and l results in
D22 + D22
2

2 2

2
1
2 MR M
2
2
2
s c c c s + s s
= |c|
M 1+c
(125)
.
1
2
2
For helicity-2 predictions these results can be compared to the weight function for the
angular distribution, f||=2 , presented in [10] and [11]

f||=2 , ,

1
= A22 1 + c2 1 + 6c2 + c4 + A21 1 c2 1 c4
8

3 2
2
2
2
4
A2 A1 c 2s c s c3 + 3c
+ A0 1 + c
1 2c + c +
4
4
2

6
3
A2 A0 c s2 s 2 1 c4
A1 A0 c 2s c s 3 c.
+
(126)
4
2
A pure electric dipole transition would result in amplitudes A0 = 0.316, A1 = 0.548 and
A2 = 0.775. Experimentally [11] the values A0 = 0.21, A1 = 0.49 and A2 = 0.85 have
been found. It turns out that the compact expression (125) amounts to expression (126)
with the pure electric dipole values for Ai .
After the above detailed discussion of a c2 decay, we now briefly mention a similar
c1 decay, since it may experimentally contaminate the c2 events. Therefore, we need a
model for the decay
c1 (pR ) (k1 )J /(k2 ),
from which one can derive the density matrix. Again, we use the
one is replaced by the J /. We thus have from (24)

M(, 1 , 2 ) = c3 M 2 (), (1 ), (2 ), k1 .
(127)

amplitude where
(128)
The density matrix now becomes

D = |c3 |2 M 4

(k1 k2 )2

(k1 k2 )

k1 () k1 ( )
12
() ( ) .
M2
M2
(129)
When k1 is again parametrized as (93), one obtains

(k1 k2 ) 2
(k1 k2 )2

|
k|
u
u
+
D = |c3 |2 M 4 1 2
(130)

,
M2
M2
where
239

s
s
(131)
ei , c, ei .
2
2
Since in the production the helicity 1 states are favoured, the angular decay
distribution will now be approximately of the form
u = (u1 , u0 , u1 ) =
D11 + D11 1
2M 2 MR2
2M 2 + MR2
c2 1 0.216 c2 ,
(132)
which has a behaviour opposite to the decay distribution for the c2 , e.g., (102). When
forward or backward parts of the angular distribution J / are experimentally not
accessible this will affect the c2 decay more than the c1 decay. Although the GaGaRes
predicted two-photon mediated production of c1 J / is only 15% of that of c2
[4], the c1 contribution to the actually seen J / events may be higher than this 15%
because of the opposite decay distribution. In the experimental determination of the
production through its c2 J / decay it would be worthwhile to estimate the c1
contamination of the events.
Such a contamination would somewhat reduce the really present c2
production and, therefore, the derived c2 width. This remark is relevant since at
present the measured two-photon width is a factor 3 to 4 higher than the determination
from pp collider experiments [9]. More recently it has been argued that this factor is about
2 to 3 [12].
It is clear from these remarks that a precise measurement of the J / angular
distribution could clarify the above situation. Tools to calculate various density matrices
will then be indispensable.
One may wonder whether the above amplitudes predict the ratios of the widths 1 and
2 for the decay processes (127) and (84) correctly. This is considered in Section 6.4.
6. Numerical results
In this section numerical results for several quantities are given. They have been
obtained by using the event generator GaGaRes [4], while extended Galuga results served
as a check.
6.1. Azimuthal distributions
Before focussing on azimuthal distributions it is useful to show the shape of the cross
section d/dQ with its marked decrease with Q. This is displayed in Fig. 1, where every
curve has an arbitrary normalization.
1
3
Next, results
for the 6 = 1 2 distributions for the S0 and PJ bottomonium states
are given for s = 190 GeV. From Figs. 2(a)(d) it is seen that there are marked differences
between distributions for different resonances. The characteristics can be qualitatively
understood when the dominant features of the cross sections are considered. Similar
qualitative arguments, but without numerical results have been given in the literature [13]
240
Fig. 1. The Q spectrum of the photon radiated by the incoming positron for the generated bb states, the b
(continuous line), the b0 (dashed line), the b1 (dotted line) and the b2 (dash-dotted line). The events were
generated at s = 190 GeV without cuts on the external particles. The vertical scale is logarithmic.
in connection with Pomeron production of resonances. In the first place it is clear from
Fig. 1 that the largest contributions to the cross sections come from the region Q2i M 2 .
In those regions 6 (cf. [2]) such that a qualitative understanding of the distribution
in the BGMS cross section formula (72) is sufficient to explain the features of Figs. 2(a)
(d). The behaviour of the BGMS formula for the various resonances now follows from the
form factors fAB and gAB in Table 1 and the small Q2i approximation. One then obtains
distributions, where different
photon density matrices play a role. Using the approximate
numerical relations valid at s = 190 GeV and in the small Q2i region
++ + ,
+0 1.5 ++ ,
00 2.3 ++ ,
one arrives at the following shapes for the distributions:

b :

d
8X
2 1++ 2++ 1+ 2+ cos(2)
d
16X
2 1++ 2++ sin2 ,
b0 :

d
X + M 2 ++ ++
1 2 + 1+ 2+ cos(2)
4
2
3
d

X + M 2 ++ ++
1 2 cos2 ,
8
3 2
(133)
(134)
(135)
241
Fig. 2. The 6 spectra for the bb states b (a), b0 (b), b1 (c) and b2 (d). The events have been generated at
s = 190 GeV in the absence of additional cuts on the external particles. The vertical scale is linear.
b1 :

2

Q1 Q22 2 ++ ++
d
1 2 1+ 2+ cos(2)
4
2
d
2

M ++ 00 2
+ 4
1 2 Q2 + 100 2++ Q21 21+0 2+0 Q1 Q2 cos
2

2

2 ++ ++ 2 2
2 1 2 M Q1 + Q22 2M 2 Q1 Q2 cos + Q21 Q22 sin2 ,
(136)
242
Fig. 3. Contributions of the different diagonal density matrix elements of the c1 resonance (left plot) and
the c2 resonance (right plot) to the total cross section at s = 91.5 GeV, for different Q2i cuts. (I.e.,
Q2i > 0, 1, 5, . . . , 25 GeV2 .)
b2 :
2 2
d
M
.
2
d
(137)
It is clear that the sin2 and cos2 and constant distributions show up in Figs. 2(a), (b), (d).
A more complex structure arises in Fig. 2(c), but also here the cos and sin2 distributions
can be recognized on a constant background.
The 1 D2 b b state has not been included in the plot as it would lie exactly on top of the
plot for the 1 S0 b b state.
For completeness we note that for the cc states a similar 6 behaviour has been found.
6.2. Density matrices for resonance production
For the cc resonances c2 and c1 we have studied how the different helicities of the
resonance contribute to the cross section as a function of the Q2i cuts. The results for the
diagonal elements in the RRF are given in Fig. 3 In these plots both Q21 and Q22 have to
be greater than the value given on the horizontal axis. From the right plot in Fig. 3 one
can see that in the absence of cuts the dominant contribution comes from the helicity-2
components. This is in agreement with [14] where a similar dominance was found for
the helicity-2 component for the f2 (1525) resonance. For high Q2i cuts the helicity-0
component starts to dominate. In the BGMS-formalism [2] this contribution comes from
the SS component that is proportional to Q21 Q22 . This result also agrees with theoretical
predictions found by Close [15], in which he states that the helicity-0 contribution is of
the order O(tt M 2 ). From the left plot in Fig. 3 one can see that in the case of the
c1 resonance in the absence of cuts the helicity-1 component is dominant. It would be
interesting to verify this statement with experimental data. For higher Q2i cuts again the
helicity-0 component starts to dominate.
243
Fig. 4. The distribution of the polar angle of the outgoing photon in the RRF. The data points show the result from
the calculation using the combined matrix element both for the production and decay. The dashed line shows
the distribution (100) for the complete density matrix for the decay of the resonance where the production is
purely helicity-2. The dotted line represents the distribution (101) according to the pure dipole transition and the
dashed-dotted line shows the distribution that has been experimentally observed [11]. The vertical scale is linear.
6.3. Density matrix for the decay c2 J /

In Section 5 we have constructed the density matrix for the complete decay of a c2
resonance into a J / and a photon, where the J / subsequently decays into electrons
or muons. We can now use Eq. (83) to combine the density matrix for the two-photon
production of a c2 resonance with the density matrix for the complete decay of the
resonance to get the full matrix element squared. The average production density matrix
is given in Table 3. Again it is clear that the helicity-2 diagonal elements are dominant.
The combined production and decay matrix elements (94) have been used to generate the
distribution of the angle between the outgoing photon and the boost direction in the RRF.
The results are given as data points in Fig. 4.
In this figure also results are included which are obtained from the assumption of
pure helicity-2 production. Under this assumption the D22 + D22 distributions (100),
(101) or (103) can be chosen. Comparing the data points of the full calculation and
distribution (100) shows that the production is indeed helicity-2 dominated. Comparing
the distributions (101) and (103) with the model distribution (100) shows that the model is
closer to experiment than the simple dipole distribution.
6.4. The ratio (c1 J /)/ (c2 J /)
As the trace of the density matrices for the decay of a cJ resonance into a photon and
a J / yields the matrix element squared for this process, we can use Eqs. (94) and (130)
244
Table 3
The normalized density matrix = | |ei for c2 production at s = 190 GeV in the absence of cuts
on the external particles. In each entry the absolute value | | of the density matrix (upper value) and the phase
(lower value) are given. The open boxes are filled by Hermitian conjugation of the other elements
= +2
= +1
0.47
2.0 104
= +2
4.72
0.022
= +1
=0
= 0
= 1
= 2
0.012
6.27
5.4 104
1.53
0.012
6.0 104
0.029
0.13
6.0 104
2.41
0.012
6.27
2.0 104
1.58
0.47
= 1
5.56
4.1 103
0.023
5.4 104
4.67
0.022
= 2
to calculate the ratio of the two decay widths. We assume c3 /c4 = c3 /c4 , i.e., we assume
that the replacement of a by a J / introduces in both matrix elements (24) and (25) the
same scale factor. The mass of the cJ particle is denoted by MRJ . Using the expressions
for the traces of (94) and (130) yields for the ratio
(c1 J /)
(c2 J /)

M 2 (MR2 1 + M 2 ) MR2 4
1
,
=5
4
MR1
10 5 + 2
MR
Rth =
(138)
where has been defined like

=1
M2
MR2 1
0.225.
(139)
Inserting the numerical values gives

Rth 0.8969.
(140)
This number should be compared to the experimentally measured [9] ratio

tot(c1 ) Br(c1 J /)
= 0.89 0.15,
(141)
tot(c2 ) Br(c2 J /)
where in the calculation of the error the different contributions to this error are taken to be
independent. Using the data selection of [12] this number becomes
Rexp =
Rexp = 0.76 0.26.
(142)
These experimental numbers are compatible with the above theoretical prediction.
Acknowledgements
The authors would like to thank Dr. G. Schuler for making the program Galuga available
and providing detailed information on it.
245
Appendix A. Transforming the density matrices

In Section 4 we have constructed the complete density matrix in the BGMS frame.
However, from an experimental point of view the RRF is a more convenient reference
frame. As the elements of the density matrix depend on the chosen polarization
vectors/tensors, the density matrix is frame dependent. When one wants to rotate the
quantization axis over the angles p and p the density matrix changes into the density
matrix according to
= A A .
(A.1)
Thus a density matrix calculated in the BGMS frame can be transformed to the RRF.
In the BGMS frame the boost direction which was needed to get to the c2 rest system is
characterized by p and p .
For spin-1 resonances this transformation matrix A can be shown to be
1+c i
1c i
s
2 e
2 e
2
s ei
s e i
c
A =
(A.2)
.
2
2
1c i
2 e
1+c i
2 e
The order of the indices , is taken to be (+, 0, ). For spin-2 resonances this
transformation matrix reads (with order +2, +1, 0, 1, 2)

1c 2 2i
(1c)s i
(1+c)s i
1+c 2 2i
s2 3
e
e
e
2
2
2
2
2 e
2

(1+c)s 2i (1+c)(2c1) i
(1c)(2c+1) i
(1c)s 2i
3
e
cs
e
e
2 e
2

2
2
2
2
1 3 2 2i
3
3c 1
1 3 2 2i
i
A = 2 2 s e
.
32 scei
sce
s
e
2
2
2 2

(1c)s 2i (1c)(2c+1) i
(1+c)(2c1) i
(1+c)s 2i
3
e
cs 2
e
2 e
2
2
2 e

1c 2 2i
1+c 2 2i
(1c)s i
(1+c)s i
s2 3
e
2 e
2 e
e
2
2
2
2
(A.3)
In the expressions for the transformation matrices we have introduced the shorthand
notation
c = cos(p ),
s = sin(p ),
= p .
References
[1] G.A. Schuler, F.A. Berends, R. van Gulik, Nucl. Phys. B 523 (1998) 423.
[2] V.M. Budnev, I.F. Ginzburg, G.V. Meledin, V.G. Serbo, Phys. Rep. 15 (1974) 181.
[3] J.H. Khn, J. Kaplan, E.G. Safiani, Nucl. Phys. B 157 (1979) 125;
H. Krasemann, J.A.M. Vermaseren, Nucl. Phys. B 184 (1981) 269;
R. Gastmans, W. Troost, T.T. Wu, Nucl. Phys. B 291 (1987) 731.
[4] F.A. Berends, R. van Gulik, hep-ph/0109195.
[5] F.A. Berends, W.T. Giele, Nucl. Phys. B 294 (1987) 700;
W.T. Giele, Ph.D. Thesis, Rijksuniversiteit Leiden, 1989.
(A.4)
246
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
S. Dittmaier, Phys. Rev. D 59 (1999) 016007.

G.A. Schuler, Comput. Phys. Commun. 108 (1998) 279.
P. Cho, M.B. Wise, S.P. Trivedi, Phys. Rev. D 51 (1995) 2039.
D.E. Groom et al., Eur. Phys. J. C 15 (2000) 1.
T. van Rhee, Ph.D. Thesis, Universiteit van Utrecht, 2000.
T.A. Armstrong et al., Phys. Rev. D 48 (1993) 3037.
C. Patrignani, Phys. Rev. D 45 (2001) 034017.
F.E. Close, G.A. Schuler, Phys. Lett. B 458 (1999) 127.
M. Acciarri et al., Phys. Lett. B 501 (2001) 173.
F.E. Close, Phys. Lett. B 419 (1998) 387.

Bottom-quark fragmentation in top-quark decay

G. Corcella, A.D. Mitov
Department of Physics and Astronomy, University of Rochester, Rochester, NY 14627, USA
Abstract
We study the fragmentation of the b quark in top decay in NLO QCD, within the framework of
perturbative fragmentation, which allows one to resum large logarithms log(m2t /m2b ). We show
the b-energy distribution, which we compare with the exact O(S ) result for a massive b quark. We
use data from e+ e machines in order to describe the b-quark hadronization and make predictions
for the energy spectrum of b-flavoured hadrons in top decay. We also investigate the effect of NLL
soft-gluon resummation in the initial condition of the perturbative fragmentation function on partonand hadron-level energy distributions. 2002 Elsevier Science B.V. All rights reserved.
PACS: 12.38.Bx; 12.38.Cy; 14.65.Ha
1. Introduction
For the sake of performing accurate studies of the top-quark properties and a precise
measurement of its mass at the present Run II of the Tevatron accelerator and, in future,
at the LHC [1] and at the Linear Collider [2], a reliable description of the bottom-quark
fragmentation in top decay t bW will be essential.
As shown in [3], the b fragmentation is indeed one of the sources of uncertainty in the
measurement of the top mass at the Tevatron, as it contributes to the so-called Monte Carlo
systematics. At the LHC, recent studies [4] have suggested that final states with leptons
and J /, with the J / coming from the decay of a b-flavoured hadron and the isolated
lepton from the W decay, will be a promising channel to reconstruct the top mass. In [4],
the expected experimental error, a result of statistics and systematics, has been estimated to
be mt 1 GeV and the b fragmentation is the largest source of uncertainty, accounting
for about 0.6 GeV.
E-mail addresses: corcella@pas.rochester.edu (G. Corcella), amitov@pas.rochester.edu (A.D. Mitov).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 3 9 - 3
248
G. Corcella, A.D. Mitov / Nuclear Physics B 623 (2002) 247270
Available tools to describe the b-quark hadronization are Monte Carlo event generators,
such as HERWIG [5], PYTHIA [6] or ISAJET [7], implementing respectively cluster [8],
string [9] and independent-fragmentation [10] models. Monte Carlo programs describe
the initial- and final-state multiparton radiation in hadron collisions according to the soft
and/or collinear approximation (see, for example, [11]). HERWIG and PYTHIA parton
showers have been provided with matrix-element corrections for a few processes, such
as top decay [12], in order to allow hard and large-angle parton radiation. The analysis
of [4] was in fact performed using the PYTHIA event generator. In [13], the HERWIG
event generator was used to perform studies on the top mass reconstruction at the LHC,
relying on the b-quark fragmentation. The mB invariant mass distribution, where B is a
b-flavoured hadron and a lepton from the W decay, was exploited in order to fit the top
mass.
In this paper we analyse the b-quark fragmentation in top decay in the framework
of perturbative fragmentation at next-to-leading order (NLO) in QCD. The factorization
theorem [14] dictates that, up to power corrections O((1/Q)p ), with p 1 and Q being
a characteristic energy scale of the process, a hadron-level cross section can be written
as the convolution of a short-distance, perturbative cross section and long-distance, nonperturbative terms, corresponding to initial-state parton distribution functions and/or finalstate fragmentation functions. For heavy-quark production, the quark mass m acts as a
regulator for the collinear singularity and allows one to perform perturbative calculations.
However, fixed-order event shapes or differential distributions typically contain terms like
S log(Q2 /m2 ), where Q is, for example, the centre-of-mass energy or the heavy-quark
transverse momentum. Such terms spoil the convergence of the perturbative expansion
and make fixed-order calculations unreliable once Q is much larger than m. The method
of perturbative fragmentation functions, originally proposed in [15], allows one to resum
these large logarithms.
According to the method in [15], heavy quarks are first produced at large transverse
momentum m pT , as if they were massless, and afterwards they slow down and
fragment into a massive object. The perturbative fragmentation function D(F , m)
expresses the transition of a massless parton into a massive quark at the factorization
scale F .
The value of D(F , m) at any scale F can be obtained by solving the Dokshitzer
GribovLipatovAltarelliParisi (DGLAP) equations [16,17], once its initial value at a
scale 0F is assigned. The universality of the initial condition and, in general, of the
perturbative fragmentation function, already suggested in [15] in the framework of e+ e
annihilation, has been recently proved in a completely process-independent way [18]. As
discussed, e.g., in [19] for heavy-quark production at hadron colliders, the perturbative
fragmentation formalism yields a weaker dependence of phenomenological observables
on the renormalization/factorization scales and on the chosen set of parton distribution
functions. Furthermore, the analysis of [18] fully resums the leading (LL) and next-toleading logarithms (NLL) which are associated with the emission of soft gluons and appear
in the initial condition of the perturbative fragmentation function (process independent)
and in the parton-level differential massless cross section (process dependent) of e+ e
annihilation.
249
Finally, in order to describe the non-perturbative fragmentation of a parton into a

hadron, several phenomenological models have been proposed [20,21], besides the ones
which are implemented in Monte Carlo event generators. Non-perturbative fragmentation
functions contain parameters which need to be fitted to the experimental data. Since the
hadronization mechanism is universal and independent of the perturbative process which
produces the heavy quark, one can exploit the existing data on e+ e b b events to fit
such models and describe the b-quark non-perturbative fragmentation in other processes,
such as top decay.
Perturbative fragmentation functions and non-perturbative hadronization models have
been extensively applied to study the physics of c- and/or b-flavoured hadrons produced in
e+ e annihilation [2225], hadron collisions [19,26], Deep Inelastic Scattering [24] and
p collisions [27].
In this work, we apply this method to the b fragmentation in top decay. In Section 2
we review the method of perturbative fragmentation functions and apply it to predict
the b-quark energy spectrum in top decay. In Section 3 we analyse the non-perturbative
fragmentation of the b quark and show energy distributions of b-hadrons in top decay,
making use of fits to LEP and SLD data to parametrize the hadronization models. In
Section 4 we summarize the main results of our analysis and make comments on possible
developments of our study. In Appendices A and B we show details of our calculation and,
comparing results for massless and massive b quarks, we check that our computation is
consistent with the initial condition of heavy-quark perturbative fragmentation functions.
2. Perturbative fragmentation and parton-level results

We wish to study b-quark production in top decay within the framework of perturbative
fragmentation functions. We consider the decay of an on-shell top quark at next-to-leading
order in S 1

t (q) b(pb )W (pW ) g(pg )
(1)
and define the b-quark scaled energy fraction xE as:

xE =
2pb q
.
m2t
(2)
Neglecting the b mass, we have 0 xE 1 w, w being w = m2W /m2t . Throughout this

paper, we shall make use of the normalized b-energy fraction:
xb =
xE
,
1w
0 xb 1.
(3)
1 Our assumption B(t bW ) = 1 is consistent with recent measurements of the CDF Collaboration of the
ratio R = B(t W b)/B(t W q), where q is a d, s or b quark, and the subsequent extraction of the Cabibbo
KobayashiMaskawa matrix element Vtb [28].
250
Following [15], the differential width for the production of a massive b quark in top decay
can be expressed via the following convolution:

xb
dz d i
(z, mt , mW , , F )Di
, F , m b ,
z dz
z
i xb
(4)
where 0 is the width of the Born process t bW , d i /dz is the differential width for the
production of a massless parton i in top decay with energy fraction z, Di (xb /z, F , mb ) is
the perturbative fragmentation function for a parton i to fragment into a massive b quark. In
Eq. (4) and F are the renormalization and factorization scales respectively, the former
associated with the renormalization of the strong coupling constant. In principle, one can
use two different values for the factorization and renormalization scales; however, a choice
often made consists of setting = F and we shall adopt this convention for most of the
results which we shall show.
The approach of [15] and the factorization on the right-hand side of Eq. (4) are
rigorously valid if one can neglect terms behaving like (m/Q)p , where p 1, m is the
mass of the fragmenting heavy quark and Q is the hard scale of the process. This is indeed
our case since the scale of top decay is set by its mass and mb /mt O(102 ).
The definitions of d i /dz and Di (xb /z, F , mb ) are not unique, but they depend on the
scheme which is used to subtract the collinear singularities which appear in the massless
differential width d i /dz. In [15], the MS factorization scheme is chosen and we shall
stick to it hereinafter.
Since we have been assuming B(t bW ) = 1 and the probability to produce a b quark
via the splitting of a secondary gluon is negligible, we shall safely limit ourselves to
considering the perturbative fragmentation of a massless b into a massive b and, on the
right-hand side of Eq. (4), we shall have only the i = b contribution. We shall then need
to evaluate the differential width d b /dz for the production of a massless b quark in top
decay, in dimensional regularization, and subtract the collinear singularity according to the
MS prescription. We obtain the following MS coefficient function:
1 d
1
(xb , mt , mW , mb ) =
0 dxb
0
1 d b
0 dz
MS
with
A 1 (z) = CF
= (1 z) +

1
S ()
A1 (z),
2
m2
3
1+w
1 + z2
2 log(1 w)
+ (1 z) log 2t + 2
(1 z)+ 2
1 + 2w
F

1 + z2
1
+
4 log (1 w)z
(1 z)+
1 + 2w

w(1 w)(1 z)2
4z
1
(1 z)+
(1 + 2w)(1 (1 w)z)

1
1
2
log(1 z)
log z
+2 1+z
1z
1z
+
(5)

2 2
+ (1 z) 4Li2 (1 w) + 2 log(1 w) log w
3
1 + 8w
2w
3w
+
log(1 w)
log w +
9
1 + 2w
1w
1 + 2w
251

,
(6)
where
x
Li2 (x) =
dt
log(1 t)
t
(7)
is the Spence function. In Appendix A we shall give more details on the derivation of
Eq. (5) and present results for the differential width d /dxb once the b-quark mass is
fully taken into account.
In order to be consistent at NLO, in Eq. (5) S () is to be the strong coupling constant
at NLO as well:

1
b1 log[log(2 /2 )]
1
,
S () =
(8)
b0 log(2 /2 )
b02 log(2 /2 )
with b0 and b1 given by
33 2nf
153 19nf
,
b1 =
,
(9)
12
24 2
being the typical QCD scale and nf = 5 the number of flavours.
We note that the coefficient function (5) contains a term where the strong coupling
constant multiplies the logarithm log(m2t /2F ). For our calculation to be reliable, we shall
have to require such a logarithm not to be too large, which implies that the factorization
scale F will have to be chosen of the order of mt .
In [15], considering heavy-quark production in e+ e annihilation and comparing
the massive and massless differential cross sections in the MS factorization scheme,
the authors have obtained the NLO initial conditions at a scale 0F for heavy-quark
perturbative fragmentation functions. For the b b transition, it has been found:2
b0 =
Db (xb , 0F , mb )
= (1 xb ) +

2
S (0 )CF 1 + xb2
log 0F
2
log(1
x
)
1
,
b
2
1 xb
m2b
+
(10)
with CF = 4/3. In order to avoid large logarithms in Eq. (10), the scale 0F is to be taken
of the order of the b mass. The universality of the initial condition (10) has been lately
proved in [18] and one can therefore exploit it to predict the b fragmentation in top decay
as well.
For the sake of completeness, in Appendix A we shall show the result for the differential
width d /dxb once we keep the b mass only in contributions log(m2t /m2b ) and neglect
2 For the fragmentation of a gluon, light quark or b quark into a b quark, see [15].
252
terms proportional to powers of the ratio mb /mt . Comparing the result with the massless
rate (5), we shall be able to reproduce the initial condition of the b-quark fragmentation
function, which will be a consistency check of our calculation and, at the same time, a
confirmation of the validity of Eq. (10) in our context as well.
Assigned the initial condition (10), the value of the perturbative fragmentation function
at any other scale F can be obtained by solving the DGLAP evolution equations [16,17]:

dz
xb
Pij
, S (F ) Dj (z, F , mb ),
z
z
(11)

S (F ) (0)

S (F ) 2 (1)
Pij xb , S (F ) =
Pij (xb ) +
Pij (xb ) + O S3 .
2
2
(12)

d
Di (xb , F , mb ) =
2
d log F
j
1
xb
where
(0)
Pij (xb ) are the AltarelliParisi splitting functions [16], and the higher-order terms
(1)
Pij (xb ) can be found in [29,30].

As shown in [15], solving the DGLAP equations for the evolution from a scale 0F
to F allows one to resum potentially-large logarithms (F ) log(2F /20F ). Assuming
0F mb and F mt , and considering the splitting functions (12) at O(S ), one resums
the leading logarithms Sn (mt ) logn (m2t /m2b ). Accounting for O(S2 ) terms in Eq. (12)
leads to the inclusion of next-to-leading logarithms Sn+1 (mt ) logn (m2t /m2b ) as well. In
this paper, we shall always assume that the b-quark perturbative fragmentation function
evolves with NLL accuracy.
The DGLAP equations get highly simplified in Mellin space; the solution reads [15]:
Di,N (F , mb ) = Di,N (0F , mb )

(1) 2b1 (0)
1
(0)
(
)
(
)
P
exp PN t +
,
P
S
0F
S
F
N
4 2 b0
b0 N
(13)
with Di,N (F , mb ) being the Mellin transform of the x-space perturbative fragmentation
function
1
dx xbN1 Di (xb , F , mb )
Di,N (F , mb ) =
(14)
and the variable t defined as

S (0F )
1
.
log
t=
(15)
2b0
S (F )
Throughout our analysis, i = b in Eqs. (11)(14) and Db,N (0F , mb ) is the N -space
transform of Eq. (10). Expressions for Db,N (0F , mb ) and for the Mellin transforms of
the NLO splitting functions in Eq. (12) PN(0) and PN(1) can be found in [15]. We shall have
to compute the N -space transform of the MS coefficient function (5)3 and multiply it by
3 In Appendix B we shall present results for the N -space counterpart of Eq. (5).
253
Eq. (13) in order to get the Mellin transform of Eq. (4):

N (mt , mW , mb ) = N (mt , mW , , F )Db,N (F , mb ),
(16)
with
1
N (mt , mW , mb ) =
0
1
dxb xbN1
d
(xb , mt , mW , mb ).
dxb
(17)
The b-quark energy distribution in x-space will finally be obtained by inverting the N space result (16) numerically.
Furthermore, in Eq. (10) the coefficient multiplying the strong coupling constant
contains terms behaving 1/(1 xb )+ or [log(1 xb )/(1 xb )]+ once xb 1,
which corresponds to behaviours log N or log2 N in moment space, for large N .
The limit xb 1 (N ) corresponds to soft-gluon radiation in top decay. Soft LL
Sn (0 ) logn+1 N and NLL Sn (0 ) logn N contributions in the initial condition of
the perturbative fragmentation function have been resummed in [18]. Due to the process
independence of the heavy-quark perturbative fragmentation function, we can exploit the
result of [18], which we do not report here for the sake of brevity, to resum LL and NLL
terms in Eq. (13) in top decay as well.
We wish to present results for the normalized b-quark energy distribution in top decay
using the technique just described. We normalize our plots to the total NLO width ,
obtained neglecting powers (mb /mt )p , whose expression can be found in [31]. In
fact, the factorization on the right-hand side of Eq. (4), the DGLAP evolution and the
resummation of soft logarithms in the initial condition of the perturbative fragmentation
function with NLL accuracy do not affect the total NLO normalization.4
It can be observed that, as long as one neglects interference between top production and
decay, one has:
1 d
1 d
=
,
dxb dxb
(18)
where (1/ ) d/dxb is the normalized differential cross section for the production of a b
quark with energy fraction xb via top quarks, independently of the production mechanism.
Our results will then be applicable to pp (Tevatron), pp (LHC) or e+ e (Linear Collider)
collisions. For our numerical study, we shall assume mt = 175 GeV, mW = 80 GeV,
mb = 5 GeV and = 200 MeV.
We show our results for the xb spectrum in Fig. 1. We plot the xb distribution
according to the perturbative fragmentation approach, with and without NLL soft-gluon
resummation in the initial condition of the perturbative fragmentation function. For the
sake of comparison, we also show the exact O(S ) result for a massive b quark, whose
analytical expression will be given in Appendix A. We set = F = mt and 0 = 0F
= mb .
4 As far as soft-gluon resummation is concerned, matching the resummed initial condition to the exact O( )
S
result (10) (see Eq. (76) in Ref. [18]), guarantees that the total NLO width is left unchanged.
254
Fig. 1. b-quark energy distribution in top decay according to the perturbative fragmentation approach, with (solid
line) and without (dashes) NLL soft-gluon resummation in the initial condition of the perturbative fragmentation
function, and according to the exact NLO calculation, with (dot-dashes) and without (dots) inclusion of powers
of mb /mt . In the inset figure, we show the same curves on a logarithmic scale.
We note that the use of perturbative fragmentation functions has a strong impact on the
xb distribution. The fixed-order result lies well below the perturbative fragmentation results
for about 0.1 xb 0.9 and diverges once xb 1, due to a behaviour 1/(1 xb )+ .
Moreover, the full inclusion of powers of mb /mt has a negligible effect on the xb spectrum;
the dot-dashed and dotted lines in Fig. 1 are in fact almost indistinguishable. As for the
perturbative fragmentation results, the distribution with no soft-gluon resummation shows
a very sharp peak, though finite, once xb approaches unity. This behaviour is smoothed out
after we resum the soft NLL logarithms appearing in the initial condition of the perturbative
fragmentation function, as the b-energy spectrum gets softer and shows the so-called
Sudakov peak. Both perturbative fragmentation distributions become negative for xb 0
and xb 1, which is a known result, already found for heavy-quark production in e+ e
annihilation [18,25]. For xb 0, the coefficient function (5) contains large logarithms
S () log xb which have not been resummed yet. Likewise, in the soft limit xb 1,
Eq. (6) contains contributions S ()/(1 xb )+ and S ()[log(1 xb )/(1 xb )]+ .
Since S (mb ) 2s (mt ), for = mt and 0 = mb such terms are smaller than the similar
ones which appear in the initial condition of the perturbative fragmentation function (10),
but nonetheless they would need to be resummed. As stated in [18], once xb gets closer
to unity, non-perturbative contributions also become important and should be taken into
account. The region of reliability of the perturbative calculation at large xb may be related
to the Landau pole in the expression for the strong coupling constant (8), and estimated to
be xb 1 /mb .
255
Fig. 2. (a) xb spectrum for F = mt and 0F = mb /2 (dots), 0F = mb (solid) and 0F = 2mb (dashes);
(b) 0F = mb and F = mt /2 (dots), F = mt (solid) and F = 2mt (dashes). The renormalization scales are
kept at = mt and 0 = mb . All curves include NLL soft-gluon resummation in the initial condition of the
perturbative fragmentation function.
Figs. 2 and 3 show the dependence of the perturbative fragmentation xb distributions

on the factorization scales 0F and F , with (Fig. 2) or without (Fig. 3) soft-gluon
resummation. We observe in Fig. 2(a) that the dependence on the initial scale 0F is
small when we resum soft logarithms in the initial condition of the b-quark perturbative
fragmentation function. Little impact is only visible in the neighbourhood of the Sudakov
peak and, in particular, the distributions obtained for 0F = mb and 0F = 2mb are
indistinguishable from each other. On the contrary, Fig. 3(a) shows that the xb spectrum
has a remarkable dependence on 0F once soft logarithms are not resummed. Comparing
Figs. 2(b) and 3(b) we note a quite similar dependence on the choice of F . In fact,
0F rather than F is the scale which enters the expression of the initial condition
Db (xb , 0F , mb ). For F approaching 0F , the distribution with no soft resummation gets
closer to the fixed-order, unevolved one shown in Fig. 1. We expect that once one includes
soft resummation in the MS coefficient function (5), which contains the factorization scale
256
Fig. 3. As in Fig. 2, but with no soft resummation. Though not visible, all distributions show a finite, sharp peak
once xb is close to 1.
F , the dependence of the b-energy spectrum on F will become weaker as well, as found
in [18] for the purpose of e+ e annihilation. As a whole, we can state that resumming
soft logarithms yields a reduction of the theoretical uncertainty, as the dependence on
factorization scales is indeed an estimation of effects of higher-order contributions which
we have been neglecting.
3. Non-perturbative fragmentation and hadron-level results

In this section we shall present results for the energy distribution of b-flavoured hadrons
in top decay. We consider the transition b B, where B is either a meson or a baryon
containing a b quark and define the normalized b-hadron energy fraction xB similarly to
the parton-level one in Eq. (3).
257
The energy distribution of a hadron B can be expressed as the convolution of the partonB (x ):
level spectrum with the non-perturbative fragmentation function Dnp
B
1 d
1
(xB , mt , mW , mb ) =
0 dxB
0
1

dz d
B xB
(z, mt , mW , mb )Dnp
.
z dz
z
(19)
xB
In Eq. (19), (1/0 ) d /dz is the parton-level differential width (4) for xb = z and
B (x /z) is the non-perturbative fragmentation function describing the hadronization
Dnp
B
b B, which is process independent. We can therefore exploit data from e+ e b b
processes to predict the b-quark hadronization in top decay.
Several models have been proposed to describe the non-perturbative transition from
a quark- to a hadron-state. One of the most-commonly used [22,24,26,27] consists of a
simple power functional form:
Dnp (x; , ) =
1
(1 x) x ,
B( + 1, + 1)
(20)
B(x, y) being the Euler Beta function.

The model of Kartvelishvili et al. [20] is still a power law, but with just one free
parameter :
Dnp (x; ) =
1
(1 x)x .
(1 + )(2 + )
(21)
We expect that if we use the non-perturbative fragmentation function in (20), which has
two fittable parameters, we shall be able to get better agreement to the data than when
using (21). However, in our analysis we shall try to tune the model (21) as well, in order to
investigate how good it is at reproducing the data and how it compares to the other models
considered.
Finally, the Peterson model [21] describes the transition of a heavy quark into a heavy
hadron according to the following non-perturbative fragmentation function:
Dnp (x; 3) =
A
.
x[1 1/x 3/(1 x)]2
(22)
For an explicit expression of the normalization factor A and of the Mellin transform of
Eq. (22), see [32]. The N -space transforms of Eqs. (20) and (21) are straightforward. The
parameters and in Eq. (20), in Eq. (21) and 3 in Eq. (22) will have to be obtained
from fits to experimental data.
To predict the b-quark hadronization properties, we shall use LEP data from the ALEPH
Collaboration [33] and SLD data [34] for e+ e collisions at the Z pole, i.e., s =
91.2 GeV. Both data sets refer to weakly-decaying b-hadrons,5 however, while the ALEPH
xB -data just accounts for B mesons, the SLD set also considers b-flavoured baryons,
5 By weakly-decaying b-hadrons we mean hadrons containing a b quark which decay according to a weak
transition. For example, in the decay chain B B (D () ) , the B meson rather than the B is the
weakly-decaying b-hadron whose energy fraction is experimentally measurable.
258
mainly the b . In [33], it is stated that the mean values xB of ALEPH and SLD are
consistent with each other, within the range of systematic and statistical errors. However,
no complete statistical analysis aiming at checking the consistency of the xB distributions
has been done yet; therefore, some difference is to be expected when comparing fits of
non-perturbative models to data actually referring to different b-hadron samples, as we do
have.
In order for the results of our fits to be applicable to the b-hadronization in top decay, we
in the same framework as

shall have to describe the perturbative process e+ e b b(g)
we did for t bW (g) when we do the fits. In fact, the factorization on the right-hand side
of Eq. (19) and the splitting between perturbative and non-perturbative part is somewhat
arbitrary and the parametrization of the non-perturbative model indeed depends on the
approach which is used to describe the perturbative, parton-level process and on the values
which are chosen for quantities like , mb and for the renormalization and factorization
scales. We shall, therefore, use MS coefficient functions [23] for e+ e annihilation into
massless quarks and convolute them with the perturbative b-quark fragmentation function
evolved to NLL accuracy according to the DGLAP equations, possibly including softgluon NLL resummation in the initial condition of the perturbative fragmentation function.
We shall then obtain a parton-level differential cross section (1/0 )(d/dxb ), equivalent to
Eq. (4), which we shall have to convolute with the non-perturbative fragmentation function
of the hadronization model considered.
The experimental analyses [33,34] use the PYTHIA Monte Carlo event generator [6]
to simulate e+ e b b processes and the subsequent parton showering; as a result, their
framework is pretty different from the one which we have been using for top decay. We
cannot therefore simply use the parametrizations of the hadronization models as they are
reported in [33,34].
An extensive phenomenological study of fragmentation in e+ e processes, with more
details on fits to LEP and SLD data is currently under way [35]. For the purposes of our
paper, we just tune the hadronization models to the data sets for a particular choice of the
quantities which enter the perturbative calculation.
We point out that, although we convolute our perturbative result with a non-perturbative,
smooth function, problems are still to be expected once xB 0, 1. The region xB 0 will
not be reliably described since the perturbative calculation itself includes unresummed
log xb terms. For xB 1 one should in principle correctly account for all the missing,
non-perturbative power corrections. Resumming a class of perturbative soft logarithms
and using a specific functional form with few fittable parameters to describe the nonperturbative fragmentation is not sufficient to be able to include all such terms.
In order to perform trustworthy fits to the e+ e data and acceptable predictions for the
b-hadron spectrum in top decay, we shall limit ourselves to analysing data not too close to
the critical points xB = 0, 1. In particular, we shall consider ALEPH data within the range
0.18 xB 0.94 and SLD data for 0.18 xB 0.90, which implies that our predictions
for top decay will have to be considered in the same xB ranges. When doing the fits, we
account for both statistical and systematic errors on the data.
In Tables 1 and 2 we report the parameters which correspond to our best fits to the data,
along with the 2 per degree of freedom. We also investigate the impact of NLL soft-gluon
259
Table 1
Results of fits to e+ e bb data, using NLO coefficient functions, NLL DGLAP evolution and NLL softgluon resummation in the initial condition of the perturbative fragmentation function. We set = 200 MeV,
0F = 0 = mb = 5 GeV and F = = s = 91.2 GeV. and are the parameters in the power law (20),
refers to (21), 3 to (22)
2 (, )/dof
2 ()/dof
3
2 (3)/dof
ALEPH
SLD
0.31 0.15
13.21 1.62
2.62/14
20.39 0.77
17.27/15
(1.12 0.16) 103
22.96/15
1.88 0.42
27.04 4.02
11.12/16
18.80 0.60
17.46/17
(1.17 0.10) 103
130.80/17
Table 2
As in Table 1, but without resumming soft logarithms
2 (, )/dof
2 ()/dof
3
2 (3)/dof
ALEPH
SLD
0.66 0.13
12.39 1.04
7.12/14
14.97 0.44
13.30/15
(2.87 0.21) 103
52.76/15
2.05 0.28
22.10 2.13
40.23/16
14.57 0.37
58.63/17
(2.33 0.19) 103
275.69/17
resummation in the initial condition of the perturbative fragmentation function. Standard

deviations for best-fit parameters are included as well.
From Table 1, we learn that the use of the power law (20) which has two tunable
parameters leads to excellent fits to both ALEPH and SLD data, but the values of and
which minimize the 2 are affected by fairly large errors. The model of Kartvelishvili
et al. is good at fitting in with the ALEPH and SLD data as well. The Peterson model
is marginally consistent with the ALEPH data and unable to reproduce the SLD data.
Although we are comparing data samples with different b-hadron contents, we observe
that the best-fit values of 3 and obtained for ALEPH and SLD are compatible within one
and two standard deviations respectively. A bigger difference between ALEPH and SLD is
found once we try to fit the power law (20).
Comparing Tables 1 and 2 we observe that the implementation of soft-gluon resummation in the perturbative fragmentation function results in statistically-different values of
the best-fit parameters. With no soft resummation, using power laws with one or two fittable parameters still yields very-good fits to the ALEPH data, while none of the models
considered is able to describe consistently the SLD data and the Peterson non-perturbative
fragmentation fails in reproducing the ALEPH data as well.
We wish now to predict the b-hadron spectra in top decay, using models and parameters
which give reliable fits to the e+ e data. In order to account for the uncertainties on the
parameters in the non-perturbative fragmentation functions, as shown in Tables 1 and 2,
260
Fig. 4. xB spectrum in top decay, with the hadronization modelled according to a power law (solid lines), the
Kartvelishvili et al. (dashes) and the Peterson (dots) model, with the relevant parameters fitted to the ALEPH
data. The plotted curves are the edges of bands at one-standard-deviation confidence level. NLL soft-gluon
resummation in the initial condition of the perturbative fragmentation function is included. We set F = = mt
and 0F = 0 = mb .
Fig. 5. As in Fig. 4, but fitting the hadronization-model parameters to the SLD data.
for each hadronization model we shall plot two curves which delimit a set of predictions at
one-standard-deviation confidence level.6
6 We point out that the correlations between the errors on and of the non-perturbative fragmentation
function (20) are correctly taken into account throughout our analysis and in the plots which we show.
261
Fig. 6. As in Fig. 4, using the model of Eq. (20) and fitting its parameters to ALEPH (solid lines) and SLD
(dashes), including NLL soft resummation in the initial condition of the perturbative fragmentation function. We
also show results from the fit to ALEPH, but with no soft resummation (dots).
In Figs. 4 and 5 we show the xB distribution using all three considered models fitted
to the ALEPH data (Fig. 4) and the power forms (20) and (21) fitted to SLD (Fig. 5),
with all curves including NLL soft resummation in the initial condition of the perturbative
fragmentation function. We note that different hadronization models yield statisticallydifferent predictions for b-hadron spectra in top decay, within the accuracy of one standard
deviation. However, one can show that the xB distributions according to the models (20)
and (21) fitted to the SLD data are consistent within two standard deviations. This result
can be related to the use of similar functional forms and to the large errors on and .
The prediction obtained fitting the Peterson non-perturbative fragmentation function to the
ALEPH data looks pretty different from the others, especially at small and middle values
of xB , where the predicted errors are pretty small for all the considered models. Moreover,
the Peterson distributions are peaked at slightly-larger values of xB .
In Fig. 6 we compare the ALEPH and SLD predictions according to the power law of
Eq. (20) since, as shown in Table 1, this is the only hadronization model where the fitted
parameters are statistically different. In fact, the overall shapes of the ALEPH- and SLDbased distributions lead to different predictions within one-standard-deviation accuracy,
especially for xB 0.7, with the SLD ones being peaked at smaller xB values. This result
can be checked to be true even at higher confidence level and, as anticipated, can be
associated with the different b-hadron samples which the two experiments reconstructed.
For the sake of comparison, we also show the ALEPH-based prediction without resumming
soft logarithms, as Table 2 reports a small 2 even in this case. We find that, although
the perturbative predictions look rather different according to whether such contributions
are resummed or not (see Fig. 1), the hadron-level results agree for xB 0.5. In fact, the
convolution with the non-perturbative fragmentation function smears the sharp peak shown
262
Fig. 7. Comparison of hadron- (j = B, solid) and parton-level (j = b, dashes) results using the hadronization
model of Eq. (20), with and being the central values in the fit to the ALEPH data, and including NLL soft
resummation in the initial condition of the perturbative fragmentation function.
by the unresummed result and the parameters and are accordingly set by the fit to the
e+ e data in such a way that the two xB -predictions do not differ too much from each
other. Statistically-significant differences are nonetheless found for xB 0.5, where the
predicted bands get narrower.
It is finally interesting to gauge the overall impact of the non-perturbative fragmentation
by comparing our parton-level result with one of our hadron-level predictions, in particular
the one obtained using the power law with two parameters fitted to the ALEPH data, as
it corresponds to our smallest 2 . From Fig. 7, we learn that the hadronization effects are
remarkable and the xB distribution is considerably softened with respect to the xb one. This
means that, even when one uses a refined perturbative approach, with DGLAP evolution
and soft resummation in the initial condition of the perturbative fragmentation function up
to NLL accuracy, the role played by the non-perturbative input and hence by the e+ e
experimental data will still be crucial to perform any prediction on b-fragmentation in top
decay.
4. Conclusions
We discussed the b-quark fragmentation in top decay in NLO QCD using the method of
perturbative fragmentation, which resums large logarithms log(m2t /m2b ) which multiply
the strong coupling constant in the fixed-order massive calculation.
We computed the NLO differential width of top decay with respect to the b-quark energy
fraction xb for a massless b, fully including b-mass effects and for a massive b quark, but
neglecting contributions proportional to powers of the ratio mb /mt . We determined the
MS-subtracted coefficient function and checked that our result is consistent with the known
263
expression of the initial condition for heavy-quark perturbative fragmentation functions.

We convoluted our MS coefficient function with the process-independent, perturbative
fragmentation function for a massive b quark, evolved up to NLL accuracy using the
DGLAP equations. We showed results for the b energy-fraction distribution in top decay,
which we compared to the fixed-order results for a massive b quark. We found that the
use of the perturbative fragmentation approach has a remarkable effect on the parton-level
distribution which is smoothed out with respect to the O(S ) one, which gets arbitrarily
large once xb approaches unity. We also investigated the impact of the resummation of
process-independent next-to-leading logarithms, which appear in the initial condition of
the perturbative fragmentation function Db (xb , 0F , mb ) and are associated with softgluon radiation. We found that it softens the xb distribution and makes the dependence
on the scale 0F weaker.
We then studied the energy distribution of b-hadrons in top decay and fitted some
hadronization models to e+ e data. We used ALEPH data on b-flavoured mesons and
SLD data on b-flavoured baryons and mesons. In order to perform such fits we described
as we did for b-quark production in top decay.

the perturbative process e+ e b b(g)
Throughout our analysis, we discarded data points where the hadron-level energy fraction
is close to xB 0, 1, since our approach is unreliable in the neighbourhood of these points.
We found that, within our perturbative framework, models which describe the nonperturbative fragmentation according to power laws with one and, in particular, with two
fittable parameters lead to good descriptions of ALEPH and SLD data. The Peterson model
is marginally consistent with the ALEPH results and unable to describe the SLD data. We
also found that, although the ALEPH and SLD data refer to different b-hadron samples, the
best-fit parameters obtained using the Kartvelishvili or the Peterson model are statisticallyconsistent within the error ranges. The implementation of NLL soft-gluon resummation in
the initial condition of the perturbative fragmentation function yields pretty different fits
to the e+ e data. An unresummed perturbative calculation still leads to good fits to the
ALEPH data once we use power laws to model the hadronization, while it is unable to
reliably describe the SLD data.
We then showed distributions of the energy fraction xB of b-flavoured hadrons in top
decay, using only models which give reliable descriptions of the e+ e data. We found that
the models fitted to the ALEPH data yield statistically-distinguishable predictions for the
b-meson spectrum in top decay. In particular, the Peterson-model distributions lie quite far
from the others and are peaked at slightly-larger xB values. If we model the hadronization
using power laws with one or two parameters, fitted to the SLD data sample, the predictions
for the xB spectrum in top decay are compatible within two standard deviations, a result
which is due to the large uncertainties on the best-fit values of and . We also compared
results obtained using the power law with two parameters, but fitted to the ALEPH or SLD
data, and obtained predictions which are statistically different.
We investigated the impact of NLL soft-gluon resummation in the initial condition of
the perturbative fragmentation function on hadron-level distributions using ALEPH-based
fits. We found a significant impact on the xB spectrum only at relatively-small xB , while
predictions with or without soft resummation are indistinguishable at large xB values. This
result stresses the importance of the non-perturbative input and consequently of the use
of the e+ e results to perform reliable predictions in top decay. This can be learned from
264
direct comparison between parton- and hadron-level distributions shown throughout the
paper.
It will be now very interesting to use the present approach to perform predictions of
other observables relying on the b-fragmentation in top decay, such as the invariant-mass
distributions used in [4,13] to fit the top mass value. It is clearly mandatory to compare the
results obtained in the framework of perturbative fragmentation functions with the ones
of Monte Carlo event generators, taking particular care about the induced uncertainty on
the top mass measurement. However, for such a comparison to be reliable, Monte Carlo
programs will have to be tuned to fit with the experimental data. This is in progress.
Finally, we plan to extend the method developed in [18] in the framework of e+ e
processes to resum with NLL accuracy also the process-dependent soft logarithms
Sn () logn+1 N and Sn () logn N appearing in the Mellin transform of the MS
coefficient function (5). This further step will allow one to include in the perturbative
calculation all soft logarithms, besides the process-independent ones which are present
in the initial condition of the perturbative fragmentation function and have been correctly
accounted for throughout this paper. It will be interesting to analyse the impact of softgluon resummation in the MS coefficient function on the energy spectrum of b quarks and
b-flavoured hadrons. This is in progress as well.
Acknowledgements
We are especially indebted to M. Cacciari for useful correspondence and for providing
us with the computing code to perform numerical inversions of Mellin-space formulas
and fits of hadronization models to e+ e data. We acknowledge T. Boccali, D. Dong,
R.H. Hemingway, L.H. Orr, C. Macesanu, K.S. McFarland, C. Oleari, M.H. Seymour and
D. Wackeroth for discussions on these and related topics. The work of G.C. was supported
by grant number DE-FG02-91ER40685 from the U.S. Department of Energy.
Appendix A. Details of the calculation

We wish to give some details on the calculation of the top-decay differential width at
O(S ) for massless and massive b quarks. In particular, the comparison of the two results
will allow us to obtain the initial condition (10) of the perturbative fragmentation function.
We adopt the on-shell mass-renormalization scheme and use dimensional regularization
to regulate the ultraviolet and soft singularities and, in the massless case, the collinear
divergence as well. We define the parameter 3, which is related to the number d of
dimensions via d = 4 23.
In the massless case, the differential width will contain a pole 1/3, due to the collinear
singularity, which disappears only in the total NLO width. This requires that, in order to
get the correct finite term in the normalized (1/0) d b /dxb , with xb defined in Eq. (3), the
Born width 0 will have to be evaluated in dimensional regularization at O(3). We find:
0 =
mt |Vt b |2 (1 w)2 (1 + 2w)

w
16 sin2 W

1+w
1 + 3 E + log 4 2 log(1 w) + 2
,
1 + 2w
265
(A.1)
where is the electromagnetic coupling constant, E = 0.577216 . . . is the Euler constant,

W is the Weinberg angle and w is the ratio w = m2W /m2t , already introduced in Section 2.
We thus obtain:

1 + xb2
1 d b
S ()
3
= (1 xb ) +
+ (1 xb )
CF
0 dxb
2
(1 xb )+ 2

1
(A.2)
+ E log 4 + A 1 (xb ) ,
3
with A 1 (xb ) defined in Eq. (6). We note that A 1 (xb ) depends on the scale F , remnant of
the regularization procedure, which disappears only in the total width. We also checked that
the integral of Eq. (A.2) agrees with the result of [31], where the O(S ) corrections to the
top width have been evaluated in the approximation of a massless b quark. In order to get
the MS coefficient function (5), by definiteness, we shall have to subtract from Eq. (A.2)
the O(S ) term multiplying the characteristic MS constant (1/3 + E log 4).
We wish to derive the NLO differential width with the full inclusion of the b mass mb .
For this purpose, it is convenient to define the following quantities:
b=
m2b
,
m2t
1
s = (1 + b w),
2
b
,
=
s

Q = s 1 2,

(1 b)2
1
,
G0 = 1 + b 2w +
2
w

xb2 2
2
2
(xb ) = s
xb arcth
,
xb
(A.3)
(A.4)
(A.5)
(A.6)
(A.7)
(A.8)
where xb is the normalized b-quark energy fraction

xb =
xE
xE
,
=
xE,max
2s
xb 1,
with xE defined in Eq. (2). We find:

1 d
= (1 xb )
0 dxb
(A.9)
266

2s Li2

2Q
2Q
Li2
1s+Q
s b+Q

1s +Q
s b+Q
log(s + Q) log
+ log
2s(1 )
w

sb+Q
1
+ log(b) log
2
2s(1 )
2

Q
1s+Q
s+Q
+ 3
+ (1 b) log
+ s b log
G0
w
b

(w b)(s b)
2s(1 )
log(b)
+Q 6
2 log
1
2 (1 xb )
wG0
4
w

1
s
1+b
+
2(xb )
1+
(1 xb ) 1
(1 xb )+ G0
2w

2

s
s
1 xb
1+b
2
2
+ 2s xb 2
+
(1 xb ) 1 .
1+
G0 1 2sxb + b
G0
2w
(A.10)
CF S ()
+
Q
In Eqs. (A.2) and (A.10) we note the presence of the so-called plus prescription
1/(1 xb )+ . Such a term arises after one integrates over the phase space for real-gluon
radiation. For a massive b quark, where xb,min = 0, one makes use of the following
expansion:
(xb xb,min)3
1
1
= (1 xb ) +
+ O(3),
1+3
3
(1 xb )+
(1 xb )
(A.11)
with the plus prescription defined as:

1
xb,min

g(xb ) + h(xb ) dxb =
1

g(xb ) h(xb ) h(1) dxb .
(A.12)
xb,min
We observe that in Eq. (A.10) the quark mass regulates the collinear divergence,
therefore, unlike the result with a massless b quark (A.2), Eq. (A.10) does not contain
any dependence on the dimensional-regularization quantities 3 or F .
For the sake of checking the initial condition of the b-quark perturbative fragmentation
function (10), we need to rewrite Eq. (A.10) neglecting powers of mb /mt .
We find:
S ()
1 d
A1 (xb ),
= (1 xb ) +
0 dxb
2
(A.13)
267
with

A1 (xb ) = CF

1 + xb2
1 + xb2
3
m2
+ (1 xb ) log 2t + 2
log (1 w)xb
(1 xb )+ 2
(1 xb )+
mb
4xb
4w(1 w) xb (1 xb )
+
(1 xb )+
1 + 2w 1 (1 w)xb

2 2
+ (1 xb ) 4Li2 (1 w) + 2 log w log(1 w)
3

2w
2(1 w)
log(1 w)
log w 4 . (A.14)
1 + 2w
1w
We note that in Eq. (A.14) the strong coupling constant multiplies the large logarithm
log(m2t /m2b ), which spoils the reliability of the fixed-order calculation and makes the
approach of perturbative fragmentation mandatory. Such large logarithms are absent in
the total NLO width, which can be checked to agree with the result in the literature, once
one accounts for the mass of the b quark [36].
If F is of the order of mb , one can express the perturbative fragmentation function
D(xb , F , mb ) as a power expansion in S :
D(xb , F , mb ) = d (0) (xb ) +

S () (1)
d (xb , F , mb ) + O S2 .
2
(A.15)
Inserting Eqs. (5), (A.13) and (A.15) in Eq. (4), and solving for d (0) and d (1) , one finds:
d (0) (xb ) = (1 xb ),
(A.16)
d (1) (xb , F , mb ) = A1 (xb ) A 1 (xb ).
(A.17)
Comparing then A1 (xb ) and A 1 (xb ), it is straightforward getting Eq. (10). It should be
noted that, although A1 (xb ) and A 1 (xb ) separately depend on mW via the ratio w, their
difference and hence the initial condition for the perturbative fragmentation function do
not, which is essential for it to be process independent.
Appendix B. Coefficient function in Mellin space

For the sake of completeness, we wish to present the result for the Mellin transform of
the MS coefficient function (5):
N (mt , mW , , F ) =
1
dz z
N1
1 d b
0 dz
MS
(z, mt , mW , , F ).
(B.1)
Given (x) the Euler Gamma function, we define the poligamma functions:
k (x) =
d k log (x)
dx k
(B.2)
268
and the combinations

S1 (N) = 0 (N + 1) 0 (1),
(B.3)
S2 (N) = 1 (N + 1) + 1 (1).
(B.4)
The basic, non-trivial N -space transforms of the terms which appear in Eq. (5) are given
by:
1
dz
zN1
= S1 (N 1),
(1 z)+
(B.5)
dz
log z N1
z
= 1 (N),
(1 z)+
(B.6)
1
0
1
0
1
log(1 z)
dz
1z
1
dz
0
zN1 =

1 2
S (N 1) + S2 (N 1) ,
2 1
(B.7)
z(1 z)
2 F1 (1, N + 1, N + 2; 1 w)
zN1 =
1 (1 w)z
N +1
2 F1 (1, N
+ 2, N + 3; 1 w)
.
N +2
(B.8)
We shall then get:

N (mt , mW , , F )

3
m2t
S ()CF
1
log 2
2S1 (N) +
=1+
2
2
F N(N + 1)

1
21 (N) 21 (N + 2)
+ 1 + 2 log(1 w)
N(N + 1)

4w(1 w) 2 F1 (1, N + 1, N + 2; 1 w) 2 F1 (1, N + 2, N + 3; 1 w)
+
1 + 2w
N +1
N +2
+ S12 (N 1) + S12 (N + 1) + S2 (N 1) + S2 (N + 1)

+ 2 1 2 log(1 w) S1 (N) + 2 log w log(1 w)

2 2
2w
1w
log(1 w)
log w + 4 Li2 (1 w) 6
.
2
1 + 2w
1w
3
(B.9)
One can show that, for N :

S1 (N) 0 (N) log N.
(B.10)
Similar large-N behaviour is also shown by the Mellin transform of the initial condition
of the perturbative fragmentation function (10), whose most-singular term at large xb has
a N -space counterpart analogous to Eq. (B.7).
269
References
[1] M. Beneke, I. Efthymiopoulos, M.L. Mangano, J. Womersley et al., in: G. Altarelli, M.L. Mangano (Eds.),
Proceedings of 1999 CERN Workshop on Standard Model Physics (and more) at the LHC, CERN, 2000,
p. 419, CERN 2000-004, hep-ph/0003033.
[2] J.A. Aguilar-Saavedra et al., ECFA/DESY LC Physics Working group, hep-ph/0106315;
T. Abe et al., American Linear Collider Working Group, Studies of exotic and Standard Model physics,
Part 3, hep-ex/0106057;
K. Abe et al., ACFA Linear Collider Working Group, hep-ph/0109166.
[3] D Collaboration, B. Abbott et al., Phys. Rev. D 58 (1998) 052001;
CDF Collaboration, T. Affolder et al., Phys. Rev. D 63 (2001) 032003.
[4] A. Kharchilava, Phys. Lett. B 476 (2000) 73.
[5] G. Corcella, I.G. Knowles, G. Marchesini, S. Moretti, K. Odagiri, P. Richardson, M.H. Seymour,
B.R. Webber, JHEP 0101 (2001) 010.
[6] T. Sjstrand, Comput. Phys. Commun. 82 (1994) 74;
T. Sjstrand, P. Edn, C. Friberg, L. Lnnblad, G. Miu, S. Mrenna, E. Norrbin, Comput. Phys. Commun. 135
(2001) 238.
[7] H. Baer, F.E. Paige, S.D. Protopopescu, X. Tata, hep-ph/0001086.
[8] B.R. Webber, Nucl. Phys. B 238 (1984) 492.
[9] B. Andersson, G. Gustafson, G. Ingelman, T. Sjstrand, Phys. Rep. 97 (1983) 31.
[10] R.D. Field, R.P. Feynman, Nucl. Phys. B 136 (1978) 1.
[11] G. Marchesini, B.R. Webber, Nucl. Phys. B 310 (1988) 461.
[12] G. Corcella, M.H. Seymour, Phys. Lett. B 442 (1998) 417;
E. Norrbin, T. Sjstrand, Nucl. Phys. B 603 (2001) 297.
[13] G. Corcella, J. Phys. G 26 (2000) 634;
G. Corcella, M.L. Mangano, M.H. Seymour, JHEP 0007 (2000) 004.
[14] J.C. Collins, D.E. Soper, Annu. Rev. Nucl. Part. Sci. 37 (1987) 383.
[15] B. Mele, P. Nason, Nucl. Phys. B 361 (1991) 626.
[16] G. Altarelli, G. Parisi, Nucl. Phys. B 126 (1977) 298.
[17] L.N. Lipatov, Sov. J. Nucl. Phys. 20 (1975) 95;
V.N. Gribov, L.N. Lipatov, Sov. J. Nucl. Phys. 15 (1972) 438;
Yu.L. Dokshitzer, Sov. Phys. 46 (1977) 641.
[18] M. Cacciari, S. Catani, Nucl. Phys. B 617 (2001) 253.
[19] M. Cacciari, M. Greco, Nucl. Phys. B 421 (1994) 530.
[20] V.G. Kartvelishvili, A.K. Likehoded, V.A. Petrov, Phys. Lett. B 78 (1978) 615.
[21] C. Peterson, D. Schlatter, I. Schmitt, P.M. Zerwas, Phys. Rev. D 27 (1983) 105.
[22] G. Colangelo, P. Nason, Phys. Lett. B 285 (1992) 167.
[23] P. Nason, B.R. Webber, Nucl. Phys. B 421 (1994) 473;
P. Nason, B.R. Webber, Nucl. Phys. B 480 (1996) 755, Erratum.
[24] M. Cacciari, M. Greco, Phys. Rev. D 55 (1997) 7134.
[25] P. Nason, C. Oleari, Nucl. Phys. B 565 (2000) 245.
[26] M. Cacciari, M. Greco, S. Rolli, A. Tanzini, Phys. Rev. D 55 (1997) 2736.
[27] M. Cacciari, M. Greco, Z. Phys. C 69 (1996) 459.
[28] CDF Collaboration, T. Affolder et al., Phys. Rev. Lett. 86 (2001) 3233.
[29] G. Curci, W. Furmanski, R. Petronzio, Nucl. Phys. B 175 (1980) 27;
W. Furmanski, R. Petronzio, Phys. Lett. B 97 (1980) 437.
[30] E.G. Floratos, D.A. Ross, C.T. Sachrajda, Nucl. Phys. B 129 (1977) 66;
E.G. Floratos, D.A. Ross, C.T. Sachrajda, Nucl. Phys. B 139 (1978) 545, Erratum;
E.G. Floratos, D.A. Ross, C.T. Sachrajda, Nucl. Phys. B 152 (1979) 493;
A. Gonzales-Arroyo, C. Lopez, F.J. Yndurain, Nucl. Phys. B 153 (1979) 161;
E.G. Floratos, R. Lacaze, C. Kounnas, Nucl. Phys. B 192 (1981) 417.
[31] C.S. Li, R.J. Oakes, T.C. Yuan, Phys. Rev. D 43 (1991) 855.
[32] B.A. Kniehl, G. Kramer, M. Spira, Z. Phys. C 76 (1997) 689.
[33] ALEPH Collaboration, A. Heister et al., Phys. Lett. B 512 (2001) 30.
270
[34] SLD Collaboration, K. Abe et al., Phys. Rev Lett. 84 (2000) 4300;
D. Dong, private communication.
[35] M. Cacciari, private communication.
[36] A. Czarnecki, Phys. Lett. B 252 (1990) 467.

Heavy quark effective theory at one-loop order:

an explicit example
Alpha Collaboration
Martin Kurth a, , Rainer Sommer b
a Department of Physics and Astronomy, University of Southampton, Southampton, SO17 1BJ, UK
b Deutsches Elektronen-Synchrotron, DESY, Platanenallee 6, D-15738 Zeuthen, Germany
Received 16 August 2001; accepted 12 December 2001
Abstract
We consider correlation functions containing the axial current of one light and one heavy quark
in the static approximation as well as in full QCD, using the lattice regularization. Up to one-loop
order of perturbation theory, we study the difference between the full and the effective theory in
the continuum limit. In the full theory we find a term nonanalytic in 1/m, revealing the asymptotic
character of the 1/m-expansion. In general, deviations from the m limit turn out to be small
and are well described by the first nontrivial terms when m is a factor 23 above the external scale.
We also investigate the mass dependence of discretization errors, and find that the behaviour of the
correlation functions at finite lattice spacing differs significantly from that in the continuum limit
when the quark mass is large. 2002 Elsevier Science B.V. All rights reserved.
PACS: 11.15.Ha; 12.38.Bx; 12.38.Gc; 12.39.Hg
Keywords: Heavy quark effective theory; Matching; Lattice QCD
1. Introduction
Heavy quark effective theory (HQET) is an important tool for heavy quark phenomenology. As a means of decoupling the light degrees of freedom from the mass and spin
of heavy quarks (with masses that are much larger than QCD ), it is a powerful method to
calculate, e.g., the heavy hadron spectrum and weak decay form factors. For a review of
* Corresponding author.
E-mail address: mkurth@hep.phys.soton.ac.uk (M. Kurth).

PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 3 4 - 4
272
M. Kurth, R. Sommer / Nuclear Physics B 623 (2002) 271286
HQET see [1]. An interesting object to study is the axial current of one light and one heavy
quark, as the B-meson decay constant fB is defined by a matrix element of such a current.
QCD matrix elements have to be evaluated nonperturbatively, which means that the
lattice regularization has to be used. As only finite volumes can be treated numerically,
and the volume must be large enough to accommodate the physical system such that finite
size effects are negligible, current computing power enforces a lower bound on accessible
lattice spacings, i.e., an upper bound on the cutoff. As, on todays computers, this cutoff
cannot be chosen much larger than the b quark mass, the simulation of bottom flavoured
particles is not possible in full QCD, and the use of effective theories like HQET becomes
a necessity.
In a previous paper [2], we presented the outline of a method for the nonperturbative
renormalization of the heavy-light axial current in the static approximation, which
corresponds to the leading order in the heavy quark expansion, where we used Schrdinger
functional techniques. In this framework, one considers QCD in a finite volume of linear
size L and suitable boundary conditions. The length scale L takes over the rle played by
external momenta p in transition amplitudes in infinite volume. HQET emerges in the
large mass limit at fixed L. When no other scale is present, as it is the case in perturbation
theory, one has to consider the z limit with z = mL and some definition of the heavy
quark mass, m. The predictive power of HQET comes from the fact that the matching
of QCD and HQET does not depend on the external states (momenta). It is therefore
also independent of the boundary conditions imposed (and L). For this reason we may
use the Schrdinger functional to study the relation between the static approximation
and full QCD. The advantage is that on the one hand gauge invariant, infrared finite
correlation functions may be computed in perturbation theory (rather easily) and on the
other hand its lattice-regulated version is very well suited for nonperturbative Monte Carlo
simulations [3].
In this paper, we investigate the difference between the static approximation and full
QCD in the continuum limit, for the specific correlation functions defined in [2], both at
tree level and at one-loop order of perturbation theory. For the first time (to our knowledge),
it is explicitly demonstrated that gauge invariant, infrared finite correlation functions are
described by HQET in the limit m . Indeed, we observe that the 1/m-expansion is
only an asymptotic expansion; small nonanalytic terms exp(z) are present in the full
theory.
A second point we look at in this paper is the mass dependence of discretization errors,
after applying Symanziks improvement programme [4] to actions, boundary fields, and
the axial current.
We use the notation of Ref. [2]. This paper is organized as follows. In Section 2, the
basic definitions of the static quark action, both in the continuum and on the lattice, are
given. Schrdinger functional boundary conditions and the correlation functions we use
are also explained in that section. In Section 3, the renormalization schemes we use in
the static approximation and in full QCD are introduced. The investigation of finite mass
effects, i.e., the difference between the full and effective theories, is carried out in Section 4.
Section 5 is devoted to the study of discretization errors and their dependence on the heavy
quark mass. Our results are summarized in Section 6. Appendix A contains the relevant
one-loop Feynman diagrams.
273
2. Correlation functions
2.1. Static quarks
Static quarks can be treated as two-component objects. To simplify the notation, they
will be described as four-component spinors here, satisfying the constraint
P+ h = h ,
h P+ = h ,
P+ = 12 (1 + 0 ).
(2.1)
The static quark action is

Sh = d4 x h (x)D0 h (x).
(2.2)
Following [5], we discretize the static quark action as

Sh = a 4
h (x)0 h (x).
(2.3)
2.2. The Schrdinger functional

The Schrdinger functional consists of a finite spacetime volume T L L L, with
specifically chosen boundary conditions. Explicitly, the gauge field is chosen to be periodic
in space and to satisfy Dirichlet boundary conditions in time,

Ak (x)|x0 =T = PC
(x),
Ak (x)|x0 =0 = C(x),
(2.4)
where C(x) and C
(x) are fixed boundary fields, and P projects onto the gauge invariant
content of C
(x) (see [6]). Here we choose C = C
= 0, leading to the boundary conditions
U (x, k)|x0 =0 = 1,
U (x, k)|x0=T = 1
(2.5)
for the lattice gauge field.

The light quark field is chosen to be periodic up to a phase in the three space directions,

x + Lk = (x)ei ,
x + Lk = ei (x),
(2.6)
where is kept as a free parameter.
As for the gauge field, Dirichlet boundary conditions are imposed on the light quark
field,
P+ l (x)|x0 =0 = l (x),
P l (x)|x0 =T = l
(x),
(2.7)
l (x)P |x0 =0 = l (x),
l (x)P+ |x0 =T =
l (x),
(2.8)
and
as well as on the static quark field,

h (x)|x0 =0 = h (x),
h (x)|x0 =T =
h (x).
(2.9)
Note that no projectors are necessary in the heavy quark case because of Eq. (2.1). For
the same reason, we have P h (x) = 0. Spatial boundary conditions do not need to
274
be discussed for the static quarks since they do not propagate in space. The quark field
boundary conditions given above can be applied both in the continuum and on the lattice.
The Schrdinger functional action, including boundary and O(a) improvement terms,
can be found in [2,6,7]. In this paper, it is understood that all quantities are calculated using
the O(a) improved action.
2.3. Correlation functions
In the following, we study two different correlation functions. The function fA contains
the time component of a static-light axial current
Astat
0 (x) = 1 (x)0 5 h (x)
(2.10)
or a light-light axial current

A0 (x) = 1 (x)0 5 2 (x),
(2.11)
where the indices 1 and 2 label two flavours of relativistic quarks. It is correlated with
a suitable combination of boundary quark fields to yield the diagram illustrated in the
left of Fig. 1 upon Wick-contraction. In addition a boundary-to-boundary correlation f1 ,
shown on the right of the figure is needed.
With the boundary quark fields , (see [2] for their proper definition), the correlation
functions

1
stat
fA (x0 ) =
(2.12)
d3 y d3 z Astat
0 (x) h (y)5 1 (z)
2
and

1
f1stat = 6 d3 u d3 v d3y d3 z
1 (u)5 h
(v) h (y)5 1 (z)
(2.13)
2L
can be introduced as in [2]. In addition, the corresponding functions in full QCD,

1
fA (x0 ) =
(2.14)
d3 y d3z A0 (x) 2 (y)51 (z) ,
2
Fig. 1. Schematic drawing of the correlation functions fA (left) and f1 (right). The dot in the middle of the left
diagram symbolizes the axial current at x0 , the other dots represent the boundary quark fields.
and
f1 =
1
2L6

d3 u d3 v d3y d3z
1 (u)5 2
(v) 2 (y)5 1 (z)
275
(2.15)
are used. They are discretized as in Refs. [8,9], including O(a) improvement terms

stat
stat
1
acA
(2.16)
Astat
0 = acA 1 2 j + j j 5 h
and

acA A0 = acA 12 0 + 0 1 5 2 ,
(2.17)
with the improvement coefficients

stat(1)
stat
cA
= g02 cA

+ O g04 ,
stat(1)
cA
1
1.00(1)
4
(2.18)
and

(1)
cA = g02 cA + O g04 ,
(1)
cA = 0.00567(1)CF,
(2.19)
stat(1)
has been calculated in [10] using NRQCD (a calculation using

where the coefficient cA
(1)
a different method can be found in [11]), and the coefficient cA
has been taken from [12].
With these correlation functions, the ratios
fAstat,I(T /2)
XI (g0 , L/a) =
f1stat
(2.20)
and
YI (g0 , z, L/a) =
fAI (T /2)
f1
(2.21)
are defined, where the index I means that the O(a) improved axial currents are used, and z
parametrizes the mass dependence as defined in Section 3.1 below. In the ratios XI and YI ,
renormalization constants for the light and static quark boundary fields cancel as well as a
1/a divergence due to a heavy quark mass counterterm in the static approximation (see [2]
for details).
For our perturbative analysis, we expand the ratios to one-loop order,

XI (g0 , L/a) = XI(0) (L/a) + XI(1) (L/a)g02 + O g04 ,
(2.22)

YI (g0 , z, L/a) = YI(0) (z, L/a) + YI(1) (z, L/a)g02 + O g04 .
(2.23)
3. Renormalization and matching

When taking the continuum limit, renormalization of quark masses, couplings,
and composite operators is required. Bare lattice quark masses m0 need an additive
276
renormalization
mq = m0 mc
(3.1)
due to the chiral symmetry breaking term in the Wilson fermion action, defining a reduced
quark mass mq . The critical mass mc is determined by the requirement that in the case of
two mass-degenerate quarks the PCAC mass is zero at mq = 0. In perturbation theory, the
critical mass can be expanded as
4
mc = g02 m(1)
(3.2)
c + O g0 .
(1)
Depending on the precise definition of the PCAC mass, mc (and therefore also mc )
depend on the lattice spacing. However, in perturbation theory one may take the limit
where external momentum scales are small compared to the cutoff 1/a, and extrapolate
(1)
the coefficient mc to its continuum limit. This gives [12]
am(1)
c = 0.2025565(1) CF ,
(3.3)
a value which will be used also in Section 5 at finite values of a. Throughout this paper,
we use mq = 0 for the light quark field 1 .
3.1. Renormalization schemes
Unlike the case of full QCD, which is known to be renormalizable at all orders of lattice
perturbation theory [13], renormalizability of the static theory has not yet been proven.
However, a large number of calculations has been performed in the static approximation,
and none of these has given a reason to doubt renormalizability. Throughout this paper, we
assume that the static theory is renormalizable as usual by adding local counterterms to
action and composite fields.
At one-loop level, the ratio X is divergent in the continuum limit, i.e., a scale
dependent renormalization is required. The one-loop coefficient X(1) is expected to diverge
logarithmically, and a renormalization scheme, the lattice MS scheme can be defined by
subtracting this logarithm,
stat
Xlat () = ZA,lat
()XI (a/L),
(3.4)
stat
2
ZA,lat
() = 1 0 ln(a)glat
,
(3.5)
where

2
glat
= g02 + O g04 ,
(3.6)
and is the renormalization scale. The one-loop anomalous dimension of the static axial
current
1
4 2
does not depend on the renormalization scheme and is known from [14,15].
0 =
(3.7)
277
The ratio YI (z, a/L) remains finite in the continuum limit as long as the coupling and
the heavy quark mass are made finite by renormalization. The renormalization condition
used here is that, in the continuum limit, they equal the MS coupling and quark masses
respectively, at the renormalization scale 1/L. At one-loop level, this means that the
renormalized quantities

2
= g02 + O g04
gMS
(3.8)
and
m2,MS = Zm,MS (1 + abm mq,2 )mq,2
(3.9)
are used, where mq,2 is the reduced quark mass of the second quark flavour (remember that
the mass of the first flavour is chosen to be zero),

4
1
2
Zm,MS = 1
(3.10)
ln(a/L) 0.122282CF gMS
+ O gMS
2
2
is known from [16], and the improvement coefficient

1
bm = 0.07217(2)CFg02 + O g04
(3.11)
2
has been calculated in [9].
Taking the continuum limit means to send the lattice spacing to zero while keeping the
physical box size fixed. One must thus take the limit a/L 0, while gMS and z = Lm2,MS
are kept at fixed values.
A finite renormalization of the light-light axial current is possible, and we define a
renormalized current (ACA )0 by requiring that it satisfies the current algebra relations in
the case z = 0 (see [8,17]). This renormalization condition, together with the MS definition
of the coupling and the quark mass, gives a renormalized ratio

YCA (z, a/L) = ZA 1 + 12 abAmq,2 YI ,
(3.12)
with the perturbative renormalization constant [16]

(1)
(1)
ZA = 1 ZA CF g02 + O g04 ,
ZA = 0.0873435,
(3.13)
and the improvement coefficient [9]

bA = 1 + 0.11414(4)CFg02 + O g04 .
(3.14)
In principle, YCA should be considered also as a function of g 2 . Since at 1-loop order the
MS
dependence on the coupling is rather trivial, we have not indicated it explicitly.
3.2. Matching of full and effective theories
By construction, the effective theory is expected to describe the physics of full QCD in
the limit m , when this limit is taken appropriately [18]. Beyond the classical level this
physics boundary condition translates into a matching condition for the two theories, fixing
a renormalization scheme (match) for the effective theory. The details of the matching
278
conditions should of course be irrelevant, as long as they fix the renormalization constants.
To match the axial current, we choose the condition

Xmatch m2,MS = YCA (z, a/L) + O(1/z) + O (a/L)2 ,
(3.15)
which means that the effective theory and full QCD are matched at the scale = m2,MS .
The renormalized static axial current in this scheme is related to the current in the lattice
MS scheme by finite renormalization, i.e.,

Xmatch () = Astat Xlat () + O (a/L)2 ,
(3.16)
with

4

2
+ O glat
Astat = 1 + BAstat glat
,
(3.17)
where BAstat is a finite constant. Using

(1)
(1)
Xlat m2,MS = Xlat (z/L)

(1)
= Xlat (1/L) 0 ln(z)X(0) (a/L) + O (a/L)2 ,
(3.18)
this constant can be calculated from the matching condition Eq. (3.15),
(1)
(1)
BAstat X(0) (a/L) = YCA (z, a/L) Xlat (1/L) + 0 ln(z)X(0) (a/L)

+ O(1/z) + O (a/L)2 .
(3.19)
Here it is understood that all quantities are extrapolated to the continuum limit first, and
then the extrapolation to 1/z = 0 is performed to determine BAstat . In practical terms, we
stat (z) as the continuum limit of
define a quantity B
A
stat (z, a/L)
B
A
= 0 ln(z) +
1
X(0) (a/L)
(1)
(1) (0)
(1)
Y (z, a/L) Xlat
(a/L) . (3.20)
YI (z, a/L) + ZA
The continuum limit of X(1) and Y (1) is calculated separately by fitting the numerical data
with a function of the expected form [19]

1
1
1
1
+ h2 2 ln(1/ l) + h3 3 + h4 3 ln(1/ l) + O 1/ l 4 , l = L/a,
(3.21)
2
l
l
l
l
using the fit procedure described in Ref. [20]. This reference also explains a method to
estimate the errors of the fit parameters, which we used in our calculations. For = 0.5,
we used Eq. (3.21) without the O((a/L)3 ) terms for z > 5 and included those terms in all
stat values resulting from our numerical data and the
other fits. For some parameters, the B
A
fits are shown in Fig. 2.
It is expected that
h0 + h1

stat (z) = B stat + f1 1 + f2 1 ln(1/z) + O 1/z2 ,
B
A
A
z
z
(3.22)
where O(1/z2 ) means terms of the form 1/z2 and ln(z)/z2 , and the constants f1 and f2
can be determined from a fit to the numerical data. Fig. 3 shows the data for a set of values
279
stat (z, a/L) at = 0.5, for z = 8 (squares) and z = 12 (triangles). The bigger symbols are the data for
Fig. 2. B
A
L/a = 32.
of z between 1 and 12 at = 0.0 and between 1 and 16 at = 0.5, as well as fits of the
form (3.22), where only the filled symbols have been included in the fits. The results for
the fit constants are
f1
f2
0.05(2)
0.011(5)
0.04(2)
0.054(5)
BAstat
0.0 0.136(3)
0.5 0.137(1)
which means that our final result for BAstat is

BAstat = 0.137(1).
(3.23)
This constant is already known from [21], where it has been calculated from matrix
elements between quark states, using a gluon mass as infrared regulator. Extracting the
relevant terms from that calculation, one gets
BAstat = 0.14,
(3.24)
in agreement with our result.

It is more illuminating to turn this logic around: If we take the matching constant BAstat
from matrix elements between quark states, the correlation functions calculated from the
280
stat by extrapolation to 1/z = 0 for = 0.0 (triangles) and

Fig. 3. Calculation of the matching coefficient BA
= 0.5 (squares). Where not indicated the errors are smaller than the symbol sizes.
static effective theory are the infinite mass limits of the corresponding correlation functions
in full QCD, and the mass-dependence of the QCD correlation functions at one-loop level
has the form expected from HQET. To our knowledge this is the first explicit demonstration
of this fact for gauge invariant, infrared finite correlation functions.
4. Heavy quark expansion as an asymptotic seriesfinite mass effects

As Xmatch is defined by matching to YCA , the static theory and QCD with relativistic
quarks can be compared directly, when these two renormalization schemes are used. To
calculate the actual size of the finite mass effects, i.e., the difference between the ratio YCA
and the static limit, we define
(z) =
YCA (z) Xmatch (m2,MS )

Xmatch (m2,MS )
(4.1)
and expand
4
2
(z) = (0) (z) + (1) (z)gMS
+ O gMS
.
(4.2)
At tree level, the continuum limits of X(0) and Y (0) can be calculated analytically,

31
X(0) (a/L) =
1 + e1 + O (a/L)2 ,
2R1

1
(z + 2 ) + (z 2 )e2 X(0) (a/L)
Y (0) (z, a/L) =
22 R2

3 3 2
+
1 e1 1 e2 + O (a/L)2 ,
2 1 2 R1 R2
281
(4.3)
(4.4)
where
=
T
,
L
1 =
3 ,

R1 = 1 1 + e21 ,
2 =
z2 + 3 2 ,
R2 = z + 2 (z 2 )e22 .
(4.5)
(4.6)
Eq. (4.4) shows that the continuum limit of Y (0) is nonanalytic in 1/z, which means that an
expansion in 1/z in the sense of a convergent Taylor series is not possible, the expansion is
at most asymptotic. However, the nonanalytic terms in Eq. (4.4) are suppressed by terms of
Fig. 4. Y (z) at tree level (solid curve), for T = L and two values of , with the expansion up to order 1/z (long
dashes), order (1/z)2 (short dashes), and order (1/z)3 (dotted curve).
282
the form e2 T , and in the static limit z , the ratio Y (0) equals the ratio X(0) calculated
from the static quark action Eq. (2.2). While Y (0)(z) = X(0) for = 0, there are finite mass
effects for other values of . For = 0.5 and = 1.0 the resulting curves for Y (0) as a
function of 1/z are shown in Fig. 4, together with the expansion up to order (1/z)3 . The
1/z 0 limit of these plots gives the respective value of X(0) . It is clearly visible that the
heavy quark expansion breaks down for z < 2, which corresponds to m2,MS < 2/L.
From Eqs. (4.3) and (4.4), we get the tree level quantity (0) . The result at T = L is
that the finite mass effects for z 1 are in the range of 014 per cent for = 0.5 and 040
per cent for = 1.0. Details can be read off from Fig. 4.
At one-loop level, the correlation functions are evaluated numerically as described
above, and (1)(z) is calculated from the numerical data shown in Fig. 3. For the set of
parameters used in Fig. 3, (1) lies between 0 and 5 per cent for = 0.0, and between 0
and 2 per cent for = 0.5.
5. Cutoff effects
As we use an O(a) improved action, improved boundary fields, and improved operators
in our lattice calculation, we expect discretization errors of O(a 2 ). These are expected
Fig. 5. Discretization errors in the ratio Y (z) at tree level. The circles are z = 3 data, the triangles z = 4, the
squares z = 8, and the pentagons z = 16.
283
Fig. 6. Discretization errors in the ratio Y (z) at one-loop level. The circles are z = 3 data, the triangles z = 4, the
squares z = 8, and the pentagons z = 16.
to depend on the heavy quark mass. To shed some light on the mass dependence of the
discretization errors, we define
YCA (z, a/L) YCA (z, 0)

z, am2,MS =
,
YCA (z, 0)
(5.1)
and expand again,

4
2
= (0) + (1) gMS
+ O gMS
.
(5.2)
We calculate (0) and (1) for a range of different values of z and a/L, and print them
as a function of (am2,MS )2 in Figs. 5 and 6. The figures show that (for our choice of
parameters) the discretization errors are roughly proportional to (am2,MS )2 . This is a
significant problem in heavy quark extrapolations at finite lattice spacing, as due to the
mass dependence of the discretization errors, there is a certain risk that the slope being
used in the extrapolation contains a sizeable contribution from discretization errors. To
stat introduced in Section 3, both in
illustrate this a bit further, Fig. 7 shows the quantity B
A
the continuum with the fit described above, and at finite lattice spacing L/a = 32. Here
again it can be seen that the discretization errors increase with the value of the heavy quark
mass, and for z > 8 (which means am2,MS > 1/4)) the behaviour of the curve at finite
lattice spacing changes dramatically. That this is really due to a cutoff effect can be seen
284
stat in the continuum limit (squares) and at L/a = 32 (triangles) for = 0.5.
Fig. 7. The matching coefficient B
A
The dotted line is the fit explained in Section 3.
from Fig. 2, where the data points for L/a = 32 have been marked by slightly bigger
symbols. This shows that O(a) improvement breaks down for too large values of am2,MS ,
and heavy quark extrapolations should be performed in the continuum limit rather than
at finite lattice spacing. A detailed explanation of the improvement breakdown for heavy
quarks can be found in Section 4 of Ref. [22], where this effect is discussed in the context
of the Schrdinger functional renormalization of the QCD gauge coupling.
A comparison between the values of (z, am2,MS ) for am2,MS < 1/2 and (z) for
z 2, both at tree and one-loop level, shows that the discretization errors and the finite
mass effects are of comparable size, i.e., of the order of a few per cent.
6. Conclusions
The main point of this paper is a comparison between the axial current in QCD with
finite quark masses, and in the static approximation at one-loop order of perturbation
theory. To achieve this, we have defined gauge invariant, infrared finite, finite-volume
correlation functions with Schrdinger functional boundary conditions, and constructed
ratios of these correlation functions to cancel renormalization constants except for that of
the axial current. Furthermore, the divergence due to the residual mass term in the static
theory cancels in our ratios.
285
Using these ratios, and assuming renormalizability of the static theory, we have defined
a renormalization scheme for the static-light axial current by matching to the full theory,
such that the renormalized static ratio equals the heavy quark limit of the corresponding
ratio calculated from full QCD. This renormalization scheme allows a comparison between
the full and the static theory. Taking the matching constant BAstat from one specific
matching condition, we have verified explicitly that the heavy quark limit of an independent
correlation function is described by HQET. For quark masses down to 1/L, where L is the
extent of the spacetime box, the finite-mass effects in our ratio are in the range of a few
per cent, both at tree level and at one-loop order of perturbation theory.
As we performed our calculations in the lattice regularization, we were also able to
investigate discretization errors in our ratio of correlation functions, both at tree and oneloop level. We have shown that (for our choice of parameters), the tree level and one-loop
coefficients approach the continuum limit at a rate roughly proportional to (am)2 , if O(a)
improvement is applied to the action, the boundary fields, and to composite operators. We
observe a significant difference between continuum limit data and results at finite lattice
spacing, with a qualitative difference in the behaviour of the two curves for am > 1/4.
This demonstrates that extrapolations in the heavy quark mass should be carried out in the
continuum limit rather than at finite lattice spacings.
Acknowledgement
We thank F. Jegerlehner for useful comments on the manuscript.
Appendix A. Feynman diagrams
The expansion of the correlation functions fA and f1 is described in detail in Ref. [12].
Following that method, we get the one-loop Feynman diagrams shown in Figs. A.1 and A.2.
Fig. A.1. One loop diagrams contributing to fA (T /2).
286
Fig. A.2. One loop diagrams contributing to f1 . The dotted lines denote the link from T a to T .
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
M. Neubert, Phys. Rep. 245 (1994) 259, hep-ph/9306320, and references therein.
ALPHA, M. Kurth, R. Sommer, Nucl. Phys. B 597 (2001) 488, hep-lat/0007002.
K. Jansen et al., Phys. Lett. B 372 (1996) 275, hep-lat/9512009.
K. Symanzik, Nucl. Phys. B 226 (1983) 187.
E. Eichten, B. Hill, Phys. Lett. B 234 (1990) 511.
M. Lscher, R. Narayanan, P. Weisz, U. Wolff, Nucl. Phys. B 384 (1992) 168, hep-lat/9207009.
S. Sint, Nucl. Phys. B 421 (1994) 135, hep-lat/9312079.
M. Lscher, S. Sint, R. Sommer, H. Wittig, Nucl. Phys. B 491 (1997) 344, hep-lat/9611015.
S. Sint, P. Weisz, Nucl. Phys. B 502 (1997) 251, hep-lat/9704001.
C. Morningstar, J. Shigemitsu, Phys. Rev. D 57 (1998), hep-lat/9712015.
K.I. Ishikawa, T. Onogi, N. Yamada, Nucl. Phys. Proc. Suppl. 83 (2000) 301, hep-lat/9909159.
M. Lscher, P. Weisz, Nucl. Phys. B 479 (1996) 429, hep-lat/9606016.
T. Reisz, Nucl. Phys. B 318 (1989) 417.
M.A. Shifman, M.B. Voloshin, Sov. J. Nucl. Phys. 45 (1987) 292.
H.D. Politzer, M.B. Wise, Phys. Lett. B 206 (1988) 681.
E. Gabrielli, G. Martinelli, C. Pittori, G. Heatlie, C.T. Sachrajda, Nucl. Phys. B 362 (1991) 475.
M. Bochicchio, L. Maiani, G. Martinelli, G.C. Rossi, M. Testa, Nucl. Phys. B 262 (1985) 331.
T. Mannel, W. Roberts, Z. Ryzak, Nucl. Phys. B 368 (1992) 204.
M. Lscher, P. Weisz, Nucl. Phys. B 266 (1986) 309.
ALPHA, A. Bode, P. Weisz, U. Wolff, Nucl. Phys. B 576 (2000) 517, hep-lat/9911018.
A. Borrelli, C. Pittori, Nucl. Phys. B 385 (1992) 502.
S. Sint, R. Sommer, Nucl. Phys. B 465 (1996) 71, hep-lat/9508012.

A lattice evaluation of four-quark operators

in the nucleon
M. Gckeler a , R. Horsley b,c , B. Klaus b,d , D. Pleiter b ,
P.E.L. Rakow a , S. Schaefer a , A. Schfer a , G. Schierholz b,e
a Institut fr Theoretische Physik, Universitt Regensburg, D-93040 Regensburg, Germany
b John von Neumann-Institut fr Computing NIC, D-15735 Zeuthen, Germany
c Institut fr Physik, Humboldt-Universitt zu Berlin, D-10115 Berlin, Germany
d Institut fr Theoretische Physik, Freie Universitt Berlin, D-14195 Berlin, Germany
e Deutsches Elektronen-Synchrotron DESY, D-22603 Hamburg, Germany
Received 10 April 2001; accepted 12 December 2001
Abstract
Nucleon matrix elements of various four-quark operators are evaluated in quenched lattice
QCD using Wilson fermions. Some of these operators give rise to twist-four contributions to
nucleon structure functions. Furthermore, they bear valuable information about the diquark structure
of the nucleon. Mixing with lower-dimensional operators is avoided by considering appropriate
representations of the flavour group. We find that for a certain flavour combination of baryon structure
functions, twist-four contributions are very small. 2002 Elsevier Science B.V. All rights reserved.
PACS: 12.38.Gc; 13.60.Hb
1. Introduction
The knowledge of higher quark and gluon correlators in hadrons is of fundamental
interest in order to understand the structure of baryons and mesons on the basis of
QCD. Matrix elements of four-quark operators contain information on the quark and
diquark structure of the nucleon. Within the operator product expansion (OPE), four-quark
operators give rise to higher-twist contributions (cats-ear diagram). While this has been
known for many years [16], the size of these contributions is still uncertain. Because the
structure function F2 (x, Q2 ) of the proton is one of the best measured hadronic quantities,
E-mail address: stefan.schaefer@physik.uni-regensburg.de (S. Schaefer).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 3 1 - 9
288
M. Gckeler et al. / Nuclear Physics B 623 (2002) 287300
the natural idea would be to extract the higher-twist contribution from the Q2 dependence
of F2 . This has, however, proven to be a difficult task (for a recent attempt see Ref. [7]).
A computation of higher-twist effects from first principles is possible by means of
Monte Carlo simulations of lattice QCD, and first estimates using this method in the case
of the pion became available recently [8]. In this paper we shall extend our previous work
to the case of the nucleon.
By means of the OPE F2 (x, Q2 ) can be expressed through forward nucleon matrix elements of local operators. In the deep inelastic limit Q2 it is dominated by the leading
twist-two contributions. These have been the subject of intensive studies in the past. The
next-to-leading contributions have twist four and are suppressed by one power of 1/Q2 .
1
More precisely, the OPE relates (Nachtmann) moments [9] 0 dx x n2 F2 (x, Q2 )|Nachtmann ,
which take into account the effects of the finite proton mass mp , to the product of Wilson
coefficients and hadronic matrix elements. Schematically one finds for n = 2, 4, 6, . . .
1

dx x n2 F2 x, Q2 Nachtmann

4m2p x 2 1/2
2
dx n+1
2
F
+
2n
+
3
+
3(n
+
1)
1
+
x,
Q
n
2
x3
Q2
0

n(n + 2)4m2p x 2
1
+
2
Q
(n + 2)(n + 3)

(4)
2
2

cn (Q / , g()) (4)
1
= cn(2) Q2 /2 , g() A(2)
()
+
A
()
+
O
,
n
n
Q2
Q4

with = 2x/ 1 + 1 + 4m2p x 2 /Q2 .
1
(1)
)
The reduced matrix elements A(t
n of twist t and spin n depend on the renormalisation
(t )
scale . The mass dimension of An is t 2. The dimensionless Wilson coefficients cn(t )
can be calculated in perturbation theory. In the flavour-nonsinglet channel, the twist-two
operators are two-quark operators,
1 D 2 D n ,
(2)
symmetrised in all indices and with trace terms subtracted.

The four-quark operators we are interested in have twist four and higher. In particular,
the twist-four, spin-two matrix element A(4)
2 is given by (indices in {. . .} are symmetrised)
1
(4)
P , S|Ac{} trace|P , S P |Ac{} trace|P = 2A2 (P P trace) (3)
2
S
with the four-quark operator

a
a
Ac = G
5 t G
5 t
(4)
(using the nomenclature introduced in Ref. [8]). The quark field carries flavour, colour,
and Dirac indices, the matrices t a are the usual generators of colour SU(3)c , and for two
289
flavours the flavour matrix G reads

G = diag(eu , ed ) = diag(2/3, 1/3)
(5)
in terms of the quark charges eq . The proton states with momentum P and spin vector S
are normalised such that
P , S|P , S = (2)3 2EP (P P )SS .
(6)
The Wilson coefficient reads [4,5] c2(4) = g 2 (1 + O(g 2 )).

These expressions are to be compared with their twist-two counterparts:
(2)
P |O{} trace|P = 2A2 (P P trace)
(7)
with the operator

i 2
O = G
D
2
(8)
and the Wilson coefficient c2(2) = 1 + O(g 2 ).

The operators (4) and (8) transform identically under Lorentz transformations, but
(4) has dimension six, whereas (8) has only dimension four: four-quark operators will
in general mix with two-quark operators of lower dimension. This fact complicates
the investigation of four-quark operators, because the mixing with lower-dimensional
operators cannot be calculated reliably within perturbation theory. A nonperturbative
computation in lattice QCD could proceed along the same lines as in the case of the twistthree matrix element d2 [10]. For the time being, we do not attempt such a nonperturbative
calculation of the renormalisation and mixing coefficients of four-quark operators. Instead
we restrict ourselves to cases where mixing with lower-dimensional operators is prohibited
by flavour symmetry. As will be explained in the following section, this constrains us to
consider the flavour group SU(3). Then one can find four-quark operators whose flavour
transformation properties forbid any mixing with two-quark operators. However, mixing
among different four-quark operators is still possible. We do not have to take into account
four-quark operators with additional covariant derivatives as these are of higher dimension.
For the renormalisation of operator (4), we thus only have to consider four-quark operators
of the same type, but with different Dirac and colour structures chosen such that they have
the same transformation properties under Lorentz transformations as operator (4).
We shall present results obtained in the quenched approximation of lattice QCD with
Wilson fermions. A preliminary account of some of these results has already been given in
Ref. [11]. Since the lattice formulation of gauge theories requires an analytic continuation
from Minkowski space to Euclidean space, we now switch to the Euclidean formulation. In
particular, all operators will be written down in Euclidean space, unless otherwise noted.
2. The general framework

In our previous publication [8] we have studied four-quark operators in the pion. In this
case we could avoid mixing with lower-dimensional operators by working with operators
290
which carry isospin I = 2. Obviously, operators with I = 2 vanish in the proton. Therefore,
we enlarge the flavour symmetry group from SU(2)F to SU(3)F assuming three quarks of
the same mass. Correspondingly, we then have
G = diag(eu , ed , es ) = diag(2/3, 1/3, 1/3)
(9)
and the flavour structure of the operator in the OPE is now

+ es s s)(eu uu
+ es s s).
O = (eu uu
+ ed dd
+ ed dd
(10)
While two-quark operators transform under SU(3)F according to 3 3 = 1 8, we have

for four-quark operators: ( 3 3) ( 3 3) = 2 1 4 8 10 10 27. Four-quark
operators with I = 0, 1, I3 = 0, and hypercharge Y = 0 belonging to the multiplets 10,
10, 27 do not mix with two-quark operators and do not automatically vanish in a proton
expectation value. The operators belonging to the 27 multiplet are (giving only the flavour
structure)
I =1
=
O27

1 2
eu ed2 2eu es + 2ed es
10

dd)
(uu)(
uu)
(dd)(
s d) + (s d)(ds)
(us)(
s u) (s u)(us)
+ (ds)(

+ (dd)(
s s) ,
(s s)(uu)
(uu)(
s s) + (s s)(dd)
I =0
O27
=
(11)

1 2
eu + ed2 + eu ed 3eu es 3ed es + 3es2
60

dd)
+ (dd)(
uu)
ud)
2(uu)(
uu)
+ 2(dd)(
+ (du)(
+ (ud)(
du)
3(us)(
s d) 3(s d)(ds)
+ (uu)(
dd)
s u) 3(s u)(us)
3(ds)(

3(dd)(
s s) + 6(s s)(s s) . (12)
3(s s)(uu)
3(uu)(
s s) 3(s s)(dd)
Inserting the values of the quark charges one finds

eu2 ed2 2eu es + 2ed es = eu2 + ed2 + eu ed 3eu es 3ed es + 3es2 = 1.
(13)
As the operators belong to the same multiplet, the WignerEckart theorem tells us that the
proton matrix elements of these two operators are proportional to each other:
1
I =0
I =1
P |O27
(14)
|P = P |O27
|P .
2
Furthermore, the WignerEckart theorem relates proton matrix elements to neutron matrix
elements. Thus our results can easily be rephrased in terms of neutron expectation values.
However, unless otherwise stated, we shall only present the proton results.
The operators belonging to the multiplets 10 and 10 read
I =1
uu)
ud)
(uu)(
O10
= (dd)(
(du)(
+ (ud)(
du)
dd)
s d) (dd)(
s s) (us)(
+ (uu)(
s s) + (ds)(
s u)
+ (s u)(us)
(s s)(uu),
(s d)(ds)
+ (s s)(dd)
(15)
291
I =1 for the operator V c trace (see Eq. (20)), divided by m4 ,

Fig. 1. Plateau for the bare matrix element of O27
p
44
with = 0.1515.
I =1
uu)
ud)
(uu)(
O10
= (dd)(
+ (du)(
(ud)(
du)
dd)
s d) (dd)(
s s) + (us)(
+ (uu)(
s s) (ds)(
s u)
+ (s d)(ds) (s u)(us)
+ (s s)(dd) (s s)(uu).
(16)
Being antisymmetric with respect to the interchange of the two quark-antiquark pairs they
do not appear in the flavour decomposition of the OPE operator (10).
In the following we present Monte Carlo data from quenched simulations at = 6.0
with Wilson fermions on a 163 32 lattice. We have performed simulations at three
different values of the hopping parameter = 0.1515, 0.1530 and 0.1550 and we have
collected about 300 configurations. The statistical errors have been determined by the
jackknife method.
The proton matrix elements are computed in the standard fashion from ratios of
(0), to two-point functions, B(t)B
(0), with the
three-point functions, B(t)O( )B
of Ref. [12]. For 0 t this ratio should be independent
interpolating fields B and B
of t:
R=
B(t)O( )B(0)
1
=
P |O|P + .
B(t)B(0)
(2)2 2mp
(17)
If we vary t keeping fixed we should therefore find a region where R is constant, i.e.
shows a plateau. An example of such a plateau is shown in Fig. 1. We have always chosen
= 5 and the spatial components of the proton momentum have been set to zero. The ratio
R equals the matrix element in the lattice normalisation of states and fields. In order to
obtain
the fields in the continuum normalisation we have to multiply each quark field by
2. To normalise the states according to Eq. (6) we must multiply R by an additional

factor of 2mp . We determine the matrix elements by fitting the ratio R to a constant in the
region 11 t 17.
292
I =1 for the operator V c trace (see Eq. (20)),

Fig. 2. Chiral extrapolation of the bare proton matrix element of O27
44
divided by m4p .
(0) consists
For a general four-quark operator the three-point function B(t)O( )B
of three types of contributions, which can be represented pictorially by the following
diagrams:
It is precisely through contributions of the form of the first two diagrams that the mixing
with lower-dimensional operators occurs. Therefore these contributions cancel in the
operators which we consider, and we are left with the contributions of the last type only.
dd)
For proton matrix elements only some terms of the operators contribute, e.g., the (dd)(
terms vanish as those containing s quarks. Therefore the expectation values of the operators
(11), (15), (16) reduce to
27
1
P |(uu)(
uu)|P
,
10
10
I =1
uu)
ud)
(uu)(
P , S|O10 |P , S = P , S|(dd)(
(du)(
+ (ud)(
du)
dd)|P
, S ,
10
I =1
uu)
ud)
(uu)(
P , S|O10
|P , S = P , S|(dd)(
+ (du)(
(ud)(
du)
dd)|P
, S .
(18)
I =1
P |O27
|P =
At each value the matrix elements are made dimensionless by dividing them by the
corresponding value of m4p . As the bare quark mass is given by amq = 1/(2) 1/(2c )
we perform the chiral limit by linear extrapolation in 1/ to 1/ = 1/c = 6.3642. As an
example of our results we show in Fig. 2 the chiral extrapolation of the bare proton matrix
293
c
element of V44
trace (I = 1 component in the 27 representation of SU(3)F ) divided by
4
c see Eq. (20).)
mp . (For the definition of V
3. Operators from the 27 multiplet

The twist-four contribution in the F2 structure function comes from the four-quark
operator Ac , see Eq. (4). In order to access the flavour-27 component experimentally
one has to combine the structure functions of several baryons (p, n, , , ) in such a
way as to project out the desired flavour combination, e.g.,

I =1
|p = + |O| + 2 0 O 0 + |O|
p|O27

= + |O| + |O| 6 |O| + 2 0 O 0
+ 2 |O| + 2 p|O|p + 2 n|O|n.
(19)
Unfortunately, most of these terms will not be measured in the foreseeable future. So a
direct comparison with data is out of question. On the other hand, they can be used as a
testing ground for models of hadrons, taking the role of experimental data. Note that the 27
contribution can also be isolated by studying combinations of electromagnetic and weak
structure functions [13].
Of course, we need to know the renormalised operators. Although due to our choice of
the flavour-27 component mixing with two-quark operators is absent, different four-quark
operators may still mix under renormalisation. Therefore, we have computed the matrix
elements of the following operators (using the nomenclature introduced in Ref. [8]):
c
a
t a G
V
= G
t ,
a
a
Ac = G
5 t G
5 t ,
c
a
a
T
= G
t G
t ,
,
V = G
G
5 G
A = G
5 ,
T = G
G
.
(20)
The bare expectation values divided by m4p and extrapolated to the chiral limit are
given in Table 1 for the spin-two components, while the traces are given in Table 2.
E.g., the number shown for the operator Ac trace in Table 1 is what we obtain for
1
(u
4 5 t a u)(u
4 5 t a u) trace|P /m4p in the chiral limit. We have checked that these
P | 10
operators fulfill their Fierz identities.
The renormalisation constants have been calculated in one-loop perturbation theory [8].
The renormalised spin-two piece of the operator Ac reads

ren
Ac ()

g02
3 ln(a) + 46.072285 Ac
2
16

8
5
c
+ ln(a) + 0.083982 V + ln(a) + 0.157467 V
9
3

c
,
1.071448T 2.008965 T
(21)
= Ac
294
where g0 is the bare coupling constant ( 6/g02 ).The renormalisation scale will be
identified with the inverse lattice spacing 1/a. In our simulations this has a value of
1/a 2.12 GeV (using r0 to set the scale). In terms of the renormalised operator the
(4)
reduced matrix element A2 is then given by
1
(u
4 5 t a u)(u
4 5 t a u) trace|P
1 (4)27,I=1 2 P | 10
A
=
,
m2p 2
3
m4p
(22)
and we obtain for the lowest moment of F2 in our special flavour channel
1
m2ps (Q2 )

27,I=1

dx F2 x, Q2 Nachtmann = 0.0005(5)
+ O s2 .
2
Q
(23)
The analogous result for the neutron differs from the above only by the sign.
In the proton the corresponding twist-two contribution is about 0.14 at Q2 = 5 GeV2 .
As in the pion, the twist-four correction is tiny. Our result may be compared with bag model
estimates. In this model the scale for the prefactor in Eq. (23) is set by B/m4p 0.0006,
where B (145 MeV)4 is the bag constant. The factor B/m4p is, however, multiplied by a
relatively large (and negative) number [3].
It is rather difficult to determine the first moment of the higher-twist contribution to
F2 (x) experimentally. Phenomenological fits to the available data give a positive value
of about 0.005(4) GeV2 /Q2 [7,14]. Our matrix element, which is due to its flavour
Table 1
Matrix elements of the spin-two operators from
the 27 multiplet, divided by m4p and extrapolated
to the chiral limit
Operator
Ac{} trace
c
V{}
trace
c
T{} trace
A{} trace
V{} trace
T{} trace
(0.9 0.8)104
(4.5 1.0)104
(4.9 0.8)104
(0.0 1.3)104
(8.4 1.4)104
(4.6 1.9)104
Table 2
Matrix elements of the spin-zero operators from the 27 multiplet, divided by m4p
and extrapolated to the chiral limit
Dirac structure
ta ta
11
5 5
11
5 5

(2.80.7)104
(7.11.2)104
(17.22.0)104
(18.22.2)104
(4 7) 104
(13.4 2.6) 104

(17.8 2.9) 104
(12.1 3.9) 104
(7.4 4.0) 104
(17.6 9.6) 104
295
structure only one contribution to the full moment, is considerably smaller than this
phenomenological number.
4. Operators from the 10 and 10 multiplets

Having found rather small matrix elements for our four-quark operators from the 27 one
may ask if operators from the 10 or 10 of SU(3)F (although not contributing to F2 in the
OPE) would have larger matrix elements. With the two possible colour structures that can
form colour singlet operators, these operators are linear combinations of terms of the form
t a ) and (G
)( G
), respectively, where and are
t a )( G
(G
Dirac matrices. We have chosen the flavour matrices G, G such that we get the following
flavour structures:
uu)
(dd)(
(uu)(
dd)
(24)
ud)
(du)(
(ud)(
du).
(25)
and
These can be combined to yield the 10 and 10 structures in Eq. (18).

Discrete symmetries impose restrictions on the matrix elements of these operators. We
have
P , S|O|P , S = P , S|T POP 1 T 1 |P , S = P , S|O |P , S
(26)
where P is the parity and T is the time inversion operator. For the Dirac matrices used in
our computations we define sign factors s1 , s1 , s2 and s2 by
4 4 = s1 ,
4 4 = s1 ,
4 5 C C 1 5 4 = s2 ,
4 5 C C 1 5 4 = s2 .
Here C is the charge conjugation matrix with

determined by
CT C 1
(27)
= . One more sign 5O is
P , S|O|P , S = 5O P , S|O|P , S.
(28)
From Eq. (26) we now get for the flavour structure (24)
P , S|O|P , S = 5O s2 s2 P , S|O|P , S = s1 s1 P , S|O|P , S
(29)
and for the flavour structure (25)

P , S|O|P , S = 5O s2 s2 P , S|O|P , S = s1 s1 P , S|O|P , S.
(30)
real if 5O s2 s2 = 1 and purely imaginary if

5O s2 s2 = s1 s1 for the flavour structure (24)
Thus, the matrix element P , S|O|P , S is

5O s2 s2 = 1; the matrix element vanishes if
or 5O s2 s2 = s1 s1 for the flavour structure (25). We have checked that these restrictions are
satisfied by our results within statistical errors. We restrict ourselves in the following to the
matrix elements which are not forced to be zero by the above relations. Note that for a given
296
Table 3
Expectation values of operators with the flavour structures (24) and (25) in an unpolarised proton, divided by m4p
and extrapolated to the chiral limit
Dirac structure
1 4
54 5
5 4 5
4
Flavour
(24)
(24)
(25)
(25)
ta ta
(0.4 0.6) 103
(5.62.8)103 i
(0.6 0.5) 103
(3.11.2)103 i
11
(3.0 1.6) 103
(5.93.6)103 i
(3.3 1.8) 103
(0.11.9)103 i
Table 4
Expectation values of operators with the flavour structures (24) and (25) in a polarised proton (S = e3 ), divided
by m4p and extrapolated to the chiral limit
Dirac structure
1 3 5
1 21
5 43
4 3 5 3 4 5
53
5 3
2 1
2 5 1 5
5 3
2 1
Flavour
(24)
(24)
(24)
(24)
(24)
(25)
(25)
(25)
(25)
(25)
ta ta
(3.3 0.6) 103 i
(4.1 0.6) 103
(2.2 0.5) 103
(9.1 0.9) 103 i
(42 4) 103
(4.9 0.5) 103 i
(2.4 0.5) 103 i
(6.1 0.8) 103 i
(21.5 1.8) 103
(8.9 1.5) 103 i
11
(15.9 1.6) 103 i
(10.7 1.8) 103
(4.4 1.3) 103
(13.0 2.1) 103 i
(51 5) 103
(15.6 1.4) 103 i
(5.6 0.9) 103 i
(7.6 1.1) 103 i
(19.2 2.6) 103
(8.1 1.8) 103 i
Dirac structure at most one of the flavour structures (24) and (25) yields a non-vanishing
result.
The definite Lorentz transformation properties of our operators could be used to define
reduced matrix elements, e.g., in Minkowski space one gets
5 d)(u
5 d)|P , S
P , S|(d
5 u) (u
5 u)(d

= A5 P S S P .
(31)
Thus in this case the matrix element with = 1, = 2 and S = 3 is equal to the one
with = 2, = 3 and S = 1 . This holds only on average, so in order to increase the
statistics we averaged over these matrix elements to reduce the statistical error. The bare
expectation values divided by m4p are given together with their statistical errors in Tables 3
and 4.
The order of magnitude of the results does not differ greatly from those found for the
operators in the 27. The renormalisation constants for the 10 and 10 operators are not
known, but we do not expect that the renormalised operators have much larger matrix
elements than the bare ones.
297
5. Diquarks
The four-quark operators can be rewritten to look like a diquark density. We have
computed matrix elements of operators of the following form:

1
u a 5 C u Tb uTa C 1 5 ub ab ba aa bb ,
10
(32)

1
u a 5 C u Tb uTa C 1 5 ub ab ba + aa bb ,
10
(33)
where a, b, a and b are the colour indices. These are the two possibilities to form a colour
singlet. In Eq. (32) the diquark is in a 3 of colour and thus anti-symmetric in colour. In
Eq. (33) it is in a 6 and symmetric in colour. Because of the Pauli principle it has to be
symmetric (anti-symmetric) in the other indices. The flavour structure being symmetric,
the Dirac structure has therefore to be symmetric (anti-symmetric). Thus a given Dirac
structure will appear only for one of the two possible colour structures. The expectation
values of operators of the form (32) and (33) can be computed from those of the fourquark operators studied in Section 3. But in order to get the correct errors we have redone
the analysis. The results (again divided by m4p ) are presented in Tables 5 and 6 for the
spin-zero and the spin-two contributions, respectively.
Strictly speaking, we are again studying operators from the 27 representation of SU(3)F ,
whose u uuu
component is given by Eqs. (32) and (33), respectively. At least within

the quenched approximation it seems however reasonable to consider (32) and (33) as
Table 5
Matrix elements of the spin-zero operators from the 27 multiplet in the diquark
picture, divided by m4p and extrapolated to the chiral limit
Operator
Colour
1 C u T )(uT C 1 u)
5 5
5 5
10 (u
(83 8) 104
1
T
T 1
5 C u )(u C 5 u)
10 (u
(42 22) 104
1 C u T )(uT C 1 u)
5
5
10 (u
(45 9) 104
1 C u T )(uT C 1 u)
5 5
5 5
10 (u
(0.82.6)104
1 C u T )(uT C 1 u)
5
5
10 (u
(4.96.7)104
Table 6
Matrix elements of the spin-two operators from the 27 multiplet in the diquark picture,
divided by m4p and extrapolated to the chiral limit
Operator
Colour
1 C u T )(uT C 1 u) trace
4 5 5
5 4 5
10 (u
(12.9 2.2) 104
1
T
T 1
4 5 C u )(u C 5 4 u) trace
10 (u
(16.8 4.3) 104
1 C u T )(u T C 1 u) trace
4 5
5 4
10 (u
(3.9 2.9) 104
298
Table 7
Expectation values of operators from the 27 multiplet which correspond to diquarks of spin zero and spin one,
divided by m4p and extrapolated to the chiral limit. For the operator O the spin-zero contribution is O44 , the

spin-one contribution is 3i=1 Oii
Colour
Spin 0
1 C u T )(uT C 1 u)
5 5
5 5
10 (u
Operator
(6.8 2.4) 104
Spin 1
1
T
T 1
5 C u )(u C 5 u)
10 (u
(5.7 6.1) 104
(48 18) 104
1 C u T )(u T C 1 u)
5
5
10 (u
(15.0 3.6) 104
(30 7) 104
(76 7) 104
representing valence diquark densities. In the same spirit, one could also investigate ud
diquarks. But due to flavour symmetry (see Eq. (14)) the corresponding matrix elements
are proportional to those of the uu diquarks (32) and (33). Writing down only the flavour
structure one finds for the expectation values in the proton

1
P | u d T d T u |P = P | u u T uT u |P .
4
(34)
In order to interpret our results we have combined operators from Tables 5 and 6 such
that they correspond to diquarks of spin zero and spin one. Specifically, for an operator
O with two space-time indices we take the expectation value of O44 to represent a spin
zero diquark and the expectation value of 3i=1 Oii to correspond to a spin-one diquark.
The results (once again completely reanalysed) are given in Table 7.
For the operators in the 3 of colour the absolute values for the spin-one diquarks are
considerably larger than for the spin-zero diquarks. For the single operator in the 6 of
colour the difference is less pronounced. This pattern can tentatively be understood in a
non-relativistic quark picture. When the diquark is in the 3 of SU(3)c it is anti-symmetric
in the colour indices, and therefore the symmetric (in the spin indices) spin-one state is
favoured over the anti-symmetric spin-one state. On the other hand, when the diquark is
in the (symmetric) 6 of colour one might at first sight expect the anti-symmetric spin-zero
state to dominate over the symmetric spin-one state. Although the spin-zero contribution
is indeed less suppressed than in the 3 case it is not really dominating. This is probably
related to the fact that a diquark in the 6 of SU(3)c must be accompanied by (at least) one
gluon if it is to form a colour singlet with the remaining quark. The coupling to the gluon,
mixing large and small components of the quark spinors, would invalidate the above
arguments which worked reasonably well for diquarks in the 3 of colour.
Of course, the operators from the 10 and 10 multiplets can also rewritten in diquark
form. They then appear as linear combinations of operators in which the diquark is either
in a 3 of SU(3)c or in a 6. The fact that the matrix elements in Table 4 for the colour
structures ta ta and 1 1 have opposite signs translates into a suppression of the 6
diquarks relative to the 3 diquarks. This is in accord with our observations made above in
the case of the operators from the 27 multiplet.
299
6. Summary
In this paper we have computed the expectation values of a variety of four-quark
operators in the proton by means of quenched Monte Carlo simulations. Since it is
rather difficult to treat the mixing with lower-dimensional operators correctly, we have
restricted ourselves to operators whose flavour structure forbids this kind of mixing. The
additional requirement that the operators should not automatically vanish in the nucleon
led us to consider the flavour group SU(3)F and to choose operators from the 27, 10,
and 10 representations. One of the operators from the 27 is responsible for the twist-four
contribution to the lowest moment of F2 in a somewhat exotic flavour channel. We find
a rather small value for this contribution. This can be regarded as a first evidence of the
smallness of the higher twist effects in nucleon structure functions generated by four-quark
operators. But we cannot exclude, of course, that our result is due to strong cancellations
between different flavour contributions.
Thus we arrive for the nucleon at a conclusion which is similar to what we observed in
the pion [8]. For a more detailed comparison we plot in Fig. 3 the renormalised pion matrix
elements [8] with the flavour structure
dd)
(uu)(
uu)
+ (dd)(
dd)
+ |(uu)(
+
uu)
(du)(
ud)|
/m2
(dd)(
(ud)(
du)
(35)
I =1
together with the corresponding renormalised matrix elements for the proton p|10 O27
2
|p/mp (in lattice units). We display the results for the spin-two components setting =
= 4 (with the trace term subtracted). The normalisation of the operators is chosen such
that the flavour structure (uu)(
uu)
appears with the factor 1 in both cases. (Alternatively, it
may be remarked that SU(3)F makes the above pion matrix element equal to the expectation
I =1 in the meson-octet analogue of the proton, the K + .) It is no great surprise
value of 10 O27
Fig. 3. Renormalised four-quark matrix elements in the pion and in the proton (in lattice units).
300
that the numbers do not show many similaritiesafter all, the pion and the proton are very
different particles.
Four-quark operators in the 10 and 10 representations do not lead to much larger matrix
elements than the 27 operators, although quite a few of those which we studied give clean
signals. In the 27 sector, a rewriting of our operators in terms of diquarks reveals a structure
which lends itself to an interpretation with the help of quark model ideas: diquarks in the 3
of SU(3)c have preferably spin one. Diquarks in the 6 of SU(3)c , on the other hand, do not
fit so well into a nonrelativistic picture.
Higher-twist effects will challenge lattice QCD for a few more years. Our investigations
show that four-quark operators can give reasonable signals in present quenched Monte
Carlo simulations. But the study of physically more interesting flavour channels and
further twist-four operators remains an open problem whose solution requires progress
in nonperturbative renormalisation, especially in the treatment of mixing with operators of
lower dimension.
Acknowledgements
This work is supported by the DFG (Schwerpunkt Elektromagnetische Sonden) and
by BMBF. The numerical calculations were performed on the Quadrics computers at DESY
Zeuthen. We wish to thank the operating staff for their support.
References
[1] S.P. Luttrell, S. Wada, B.R. Webber, Nucl. Phys. B 188 (1981) 219;
S.P. Luttrell, S. Wada, Nucl. Phys. B 197 (1982) 290;
S.P. Luttrell, S. Wada, Nucl. Phys. B 206 (1982) 497 (E).
[2] R.K. Ellis, W. Furmanski, R. Petronzio, Nucl. Phys. B 207 (1982) 1;
R.K. Ellis, W. Furmanski, R. Petronzio, Nucl. Phys. B 212 (1983) 29.
[3] R.L. Jaffe, M. Soldate, Phys. Lett. B 105 (1981) 467.
[4] R.L. Jaffe, M. Soldate, Phys. Rev. D 26 (1982) 49.
[5] E.V. Shuryak, A.I. Vainshtein, Nucl. Phys. B 199 (1982) 451.
[6] J.L. Miramontes, J. Snchez Guilln, Z. Phys. C 41 (1988) 247.
[7] S.I. Alekhin, hep-ph/0011002, and private communication.
[8] S. Capitani, M. Gckeler, R. Horsley, B. Klaus, V. Linke, P.E.L. Rakow, A. Schfer, G. Schierholz, Nucl.
Phys. B 570 (2000) 393.
[9] O. Nachtmann, Nucl. Phys. B 63 (1973) 237;
S. Wandzura, Nucl. Phys. B 122 (1977) 412.
[10] M. Gckeler, R. Horsley, W. Krzinger, H. Oelrich, D. Pleiter, P.E.L. Rakow, A. Schfer, G. Schierholz,
Phys. Rev. D 63 (2001) 074506.
[11] S. Capitani, M. Gckeler, R. Horsley, B. Klaus, W. Krzinger, D. Petters, D. Pleiter, P.E.L. Rakow,
S. Schaefer, A. Schfer, G. Schierholz, Nucl. Phys. B (Proc. Suppl.) 94 (2001) 299.
[12] M. Gckeler, R. Horsley, E.M. Ilgenfritz, H. Perlt, P. Rakow, G. Schierholz, A. Schiller, Phys. Rev. D 53
(1996) 2317.
[13] S. Gottlieb, Nucl. Phys. B 139 (1978) 125.
[14] S. Choi, T. Hatsuda, Y. Koike, S.H. Lee, Phys. Lett. B 312 (1993) 351.

Towards OPE based local quarkhadron duality:

light-quark channels
Ralf Hofmann
Max-Planck-Institut fr Physik, Werner-Heisenberg-Institut, Fhringer Ring 6, 80805 Mnchen, Germany
Received 3 September 2001; accepted 4 December 2001
Abstract
Various light-quark channel currentcurrent correlators are subjected to the concept of a nonperturbative component of coarse graining in operator product expansions introduced in a parallel
work. This procedure allows for low-energy structure of the OPE-derived spectral function. With
naive vacuum saturation for 4-quark operators and using lattice data for the gauge invariant scalar
quark correlator the results are far off the experimentally measured behavior. However, using the
correlation length of the gauge invariant vector quark correlator, which is about 10 times smaller
than the scalar one, the qualitative results are rather realistic. Namely, the input of information on the
mass of the lowest resonance in one channel yields the corresponding masses within acceptable errors
in other channels. Still, the shapes of the calculated spectral functions are considerably deformed as
compared to experiment. This may be a consequence of vacuum saturation and the truncation at a
mass dimension which is below the critical dimension from which on the asymptotic expansion does
not approximate anymore. To improve on this high-resolution lattice information on gauge invariant
n > 2 point correlators would be needed. Motivated by the small effective correlation length in the
4-quark contributions the relevance of the approach for heavy quark physics, in particular in the
calculation of non-leptonic, inclusive , is discussed. 2002 Elsevier Science B.V. All rights
reserved.
PACS: 12.38.-t
1. Introduction
The basic assumption made in QCD sum rules (QSR) in particular [1] and in
applications of the operator product expansion (OPE) to hadronic physics in general [2] is
quarkhadron duality, namely, the possibility to express the low-energy dispersive part of
E-mail address: ralfh@mppmu.mpg.de (R. Hofmann).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 2 2 - 8
302
R. Hofmann / Nuclear Physics B 623 (2002) 301315
a correlator of QCD currents in terms of hadronic cross sections. For external momentum
q with q 2 = Q2 < 0 correlators are believed to be (asymptotically) approximated by
the OPE which is a power series in Q2 . A first estimate for the critical dimension
dc 12 from which on the entire coefficients, being products of Wilson coefficients and
averages over local, gauge invariant operators, are dominated by short distance effects
was performed in Ref. [1] for pure gluodynamics. Thereby, the vacuum was described by a
dilute instanton gas. Powers with d > dc have no potential to reflect the hadronic properties
of the respective channel which relate to the operator averages via dispersion relations in
QSR (global duality) and via the optical theorem and analytical continuation in applications
such as calculations of non-leptonic decay widths of heavy mesons (local duality).
Apart from some puzzles1 the tremendous success of the sum rule method seems to
support the global version of quarkhadron duality.
The observation that an analytical continuation of the OPE to time-like, external
momenta does not yield the phenomenological resonance structure in the imaginary part
has lead to the notion of local duality violation [2,4,5]. The up-to-date claim is that the
asymptotic nature of the expansion does not allow for additive, exponential-like terms to
appear. These terms are to be responsible for a wiggling of the low-energy part of the
spectral function [2]. This, however, seems to contradict the QSR philosophy [1]: if only
the first few terms of the OPE (up to the critical dimension dc ) are needed to describe the
lowest resonances in a given channel via a dispersion relation then operators of dimension
d dc should also be forgotten in an OPE-based construction of the spectral function.
Therefore, we propose an alternative in [6]. The observation is that the non-perturbative
behavior of the QCD vacuum, when probed with low Euclidean momenta, is not
only characterized by non-vanishing vacuum expectation values (VEVs) of local, gauge
invariant operators but also by finite correlation lengths of the corresponding gauge
invariant correlators [7,8]. As a consequence the fundamental field operators must loose
their relevance with decreasing resolution (see Section 2). From the knowledge of
gauge invariant correlation functions of fundamental operators at high resolution Q0
evolution equations for the VEVs A(Q) of the relevant local, effective operators can be
derived for Q < Q0 . For d = 4, where 2-point functions are needed, we have A(Q)
exp[4/5(Q)1 ] with being the correlation length. Compared with the conventional
power correction the effect becomes noticeable if Q 1 . For gluon and quark
condensates the lattice implies A(Q0 )s at Q0 2 GeV which are compatible with the
phenomenological values obtained from conventional QSRs at 1 GeV. Since the
inverse correlation lengths are below the mass of the -resonance (assuming naive vacuum
saturation at dimension 6) light-quark channel QCS sum rules are hardly touched by the
exponential decrease of the operator VEVs [6]. However, calculating the spectral function
from the truncated OPE these large correlation lengths do not reflect the resonance physics
in the respective channels. A much smaller, effective correlation length is needed at
dimension 6 to yield more realistic spectra. To investigate the properties of OPE-based
spectral functions in various light-quark channels is the main purpose of this paper.
1 Large channel-to-channel variation of spectral continuum thresholds, see Ref. [3], large error in vacuum
averages of local operators, unclarified status of such approximations as vacuum saturation in vacuum averages
over 4-quark operators, very large non-perturbative effects in the scalar and pseudoscalar channels.
303
The presentation is set up as follows: in Section 2 we briefly review the idea of nonperturbative coarse graining of local operator VEVs as it is developed in [6]. Thereby, the
focus is on correlation functions which are represented by so-called connected diagrams.
Section 3 investigates the vacuum saturation hypothesis for 4-quark operators [1] in the
light of non-perturbative coarse graining. The low-energy parts of the spectral functions in
the , a1 , and channels are calculated in Section 4 by analytical continuation of the
OPE to time-like external momenta. Implications for the calculation of the difference of
non-leptonic inclusive decay widths in neutral B-meson systems are discussed in Section 5.
In Section 6 summarizes the results.
2. Euclidean exponentials and Minkowskian oscillations

In this section we briefly review the work of [6] on non-perturbative coarse graining for
VEVs of local, gauge invariant operators.
At a large Euclidean momentum Q Q2 , where we expect the description of the
dynamics in terms of the continuum action and local operators made of fundamental fields
to be sufficiently accurate, we start by expanding the currentcurrent correlator into a
conventional OPE. The evolution to lower momenta is obtained by running the Wilson
coefficients perturbatively via the running coupling s (Q2 ) and the anomalous operator
dimensions [1]. According to [6] the non-perturbative running of an operator average is
governed by the non-perturbative part of the corresponding gauge invariant correlator in
Euclidean position space. This correspondence can be expressed as

np
np
np
F1 (0) Fn (xn ) Q .
F (0) Q F1 (0) Fn (0) Q 1/N
(1)
In Eq. (1) parallel transporters
x
S(0, x) P exp ig
dz A

(2)
are appropriately contained in the non-local expression to define gauge invariant correlations. The sum runs over all relevant,2 piecewise straight [7] trajectories of parallel
transport, and the symbol P demands path ordering. With the normalization factor 1/N ,
which depends on n and the numbers of fields transforming under the fundamental and
adjoint representation, the correlation function reduces to the condensate in the limit
x1 , . . . , xn 0. Making the convention that an arrow pointing from xi towards xj stands
for the parallel transport S(xi , xj ), we have a way to depict correlators. Note that points
with fields transforming under the fundamental (adjoint) representation are connected to
one (two) lines of parallel transport. There are disconnected and connected diagrams. In
this section we only consider the latter.
To coarse grain from resolution Q to resolution Q dQ we average the nonperturbative part of the correlator corresponding to a connected diagrams over a
2 This will be specified later.
304
(a)
(b)
(c)
Fig. 1. Diagrammatic representation of the gauge invariant correlators corresponding to the local operators
(a) q(0)q(0),
(b) tr F (0)F (0)F (0), and (c) tr F (0)F (0)F (0)F (0).
(Euclidean) ball of radius dRQ with

1
1
1
1
dQ
=
dRQ =
2.
Q dQ Q Q 1 dQ/Q 1
Q
Using the short-hand notation of Eq. (1), this is written as

np
F (0) QdQ

np
1
=
d 4 x1 d 4 xn F1 (0) Fn (xn ) Q ,
(n1)
(V (dRQ ))
(3)
(4)
|x1 |,...,|xn |dRQ
where

1 2
1 2 dQ 4
4
V (dRQ ) = (dRQ ) =
.
2
2
Q2
(5)
In Fig. 1 there are 3 examples for connected diagrams.

Let us now focus on 2-point functions as they are relevant for dimension 3 and 4 quark
and gluon operators, respectively. Only the gauge invariant bilocal quark correlator [8]

tr q(x)S(x,
0)q(0)
(6)
and the gluonic field strength correlator [7,8]

tr F (x)S(x, 0)F S (x, 0)
(7)
have been measured on the lattice. The results imply that there exists an additive
decomposition into a perturbative, power-like in |x|, and a non-perturbative, exponential
in |x| piece [8]. We are interested in the non-perturbative part, which in short-hand reads
np

F1 (0)F2 (x) Q = A(Q) exp(|x|/).
(8)
As explained in [6] is expected to have direct phenomenological meaning, and so it
cannot depend on the resolution Q. Hence, coarse graining may only affect the preexponential factor A(Q). An evolution equation for A(Q) can be derived if we combine
Eqs. (4) and (8):

2 1
V dQ/Q
d 4 x A(Q)e|x|/ = A(Q dQ)e0/ = A(Q dQ).
|x|dQ/Q2
(9)
305
Note that for A is invariant under coarse graining. Therefore, the conventional
treatment of non-perturbative corrections in the framework of the OPE [1] is recovered,
and condensates do not depend on the resolution in this limit. In particular, they could be
determined at a large resolution, where fundamental fields make sense. In the real world,
however, correlation lengths are finite [8]. We will see later how this is reflected in the
existence of hadronic resonances.
Expanding the l.h.s. and r.h.s. of Eq. (9) in dQ/Q2 and comparing coefficients of the
linear terms, we obtain the following equation:
4
d
A(Q) = Q2 A(Q),
dQ
5
(10)
with solution

1
4 1
.
A(Q) = A(Q0 ) exp
5 Q Q0
(11)
At dimension 4 there are no anomalous operator dimensions and being content with a
determination of Wilson coefficients at the lowest possible order in s the generic form of
a non-perturbative correction is

A4 (Q0 )
1
1
4
(12)
.
exp
54 Q Q0
(Q2 )2
Eq. (11) implies that compared to conventional dimension 4 power corrections there is a
noticable suppression if Q is of the order of 1 or less.
Lattice measurements with NF = 4 staggered fermions of mass a mq = 0.01 and lattice
resolution Q0 = a 1 2 GeV suggest that scalar fermionic and gluonic correlation lengths
are sq 3.2 GeV1 and g 1.7 GeV1 , respectively [8]. For the vector fermionic
correlation length vq 1/10sq was found [8]. So for dimension 4 and, assuming naive
vacuum saturation, also for dimension 6 this does not seem to pose a problem for the sum
rule analysis of light-quark correlators (at Q0 = a 1 2 GeV Aq (Q0 ) as well as Ag (Q0 )
are compatible with their QSR determined values at 1 GeV [8]). However, as we will
see later, naive vacuum saturation gives light-quark channel spectra which are completely
off the experimentally measured behavior. Qualitatively more realistic spectral functions
can be obtained using the much smaller vq .
The contribution of dimension 4 to the spectral function (s) is (up
to a normalization!)
obtained by analytically continuing to Q2 = (s + i) or Q = i s (s > 0), and taking
the imaginary part. A term like the one in Eq. (12) corresponds to a term

A4 (Q0 )
4
4 1
(13)
sin
exp
54 Q0
s2
54 s
in (s). The oscillatory behavior manifests itself in a quite different way than it was
1
suggested in Ref. [5]: sin[ s ] instead of sin[ s ] or sin[s]. There, oscillations are
present everywhere though power suppressed at high s. Here, oscillations only start if
1
s is larger than the correlation length of the corresponding 2-point function.
306
3. Vacuum saturation in the context of gauge invariant correlations

After the treatment of connected diagrams in the previous section we turn to
disconnected diagrams in this section. A sufficient condition for factorized coarse graining
is the factorization of the non-perturbative part in the corresponding correlation function.
It is unlikely that such a factorization occurs for connected diagrams.
Let us focus on the case of 4-quark operators since generalizations to higher
dimensions are straightforward. Disconnected diagrams are associated with 4-quark
operators q(0)
q(0)q(0)
q(0) composed of color singlet currents (Fig. 2). Thereby,

q is a single flavor quark field, and denotes one of 1, 5 , , . . . or a product of
them. Diagrams with permutated arguments are irrelevant because of the integrations in
Eq. (4) and the fact that we may assume translational, P and T invariance [9]. In practical
applications one encounters color octet currents in 4-quark operators. These structures lead
to special forms of disconnected diagrams which are due to the fact that gauge invariant
correlation functions can only be defined if 2 of the 4 arguments coincide, hence yielding
3 point functions. For example, an operator q(0)
t a q(0)q(0)
t a q(0) demands non-local

contributions of the form

(i)
q(0)
S(0, x1 )t a q(x1)q(x
2 )S(x2 , 0) t a q(0) ,

1 ) t a S(x1 , x2 )q(x2) .
(ii) q(x
1 ) t a S(x1 , 0)q(0)q(x
(14)
Thereby, the color matrices are normalized as tr t a t b = 2 ab . The corresponding diagrams
are depicted in Fig. 3. Lines do not meet at the points 0 (for (a)) and x1 (for (b)) because
they connect to different fields.
Since gauge invariant n > 2 point functions have yet not been measured on the lattice we
do have to think of approximations involving only 2-point functions. The (quite intuitive)
hypothesis of vacuum saturation can be implemented in two ways: (1) vacuum saturation
on the level of local operators according to the formula of Ref. [1] with separate coarse
graining for each factor and (2) vacuum saturation on the level of the correlation function,
that is, first delocalization of the operator and then vacuum insertion. There is one obvious
reason why the two prescriptions lead to different results. In case (1) we essentially square
the result for the quark condensate, which implies correlations between points of maximal
separation dQ/Q2 , whereas in case (2) one factor may contain correlations between points
of maximal separation 2dQ/Q2 (Fig. 2 or case (b) in Fig. 3). Another, not-so-obvious
reason becomes apparent if we insert the vacuum state into the correlator for singlet
Fig. 2. Diagrammatic presentation of the gauge invariant correlator corresponding to the local operator
q(0)
q(0)q(0)
q(0).
(a)
307
(b)
Fig. 3. Diagrammatic presentation of the contributions of Eq. (14) to the gauge invariant correlator associated
t a q(0).
with the local operator q(0)
t a q(0)q(0)
currents

q(0)
S(0, x1 )q(x1 )q(x

2 ) S(x2 , x3 )q(x3 ) .
(15)
Then information of the 2-point function

q(0)
S(0, x)q(x)
(16)
is needed. From Ref. [9] we know that on a lattice the correlation length v of the
vector correlator ( = ) is about 10 times smaller than the correlation length s of
the scalar correlator ( = 1)! This result has been obtained for a quark mass m with
a m = 0.01 and a 1 2 GeV in the NF = 4 theory with staggered fermions. In the octet
case vacuum insertion inbetween the currents factorizes the correlator into gauge variant
2 point functions. The gauge invariant product of them has not yet been measured on the
lattice so we cannot compare the corresponding correlation length with the singlet case.
Let us keep all these points in mind in the next section, where for simplicity we apply
prescription (1). With the necessary lattice information available we hope to investigate
case (2) in a separate publication.
4. Light-quark channels
After writing down the coarse grained OPEs in the , a1 , , and channels, in this
section we calculate the respective spectral functions by analytical continuation from
Euclidean to time-like momenta. We restrict ourselves to light-quark channels because
here the two needed correlation functions for corrections up to dimension 6 (assuming
vacuum saturation in the sense of (1)) have been measured on the lattice. We investigate
with two sets of parameters. Set (A) takes the NF = 4 results of Ref. [8] for the lowest
quark mass (a m = 0.01) literally, that is
(A):
q = 3.1 GeV1 ,
g = 1.7 GeV1 ,

Aq a 1 = Q0 2 GeV = (0.212 GeV)3 ,

Ag a 1 = Q0 2 GeV = 0.015 (GeV)4 .
(17)
308
Due to the uncertainty in the way dimension 6 contributions are treated and motivated by
the 10 times smaller vector correlation length v we use q = v in (B)
(B):
q = 0.3 GeV1 ,
g = 1.7 GeV1 ,

Aq a 1 = Q0 2 GeV = (0.212 GeV)3 ,

Ag a 1 = Q0 2 GeV = 0.015 (GeV)4 .
(18)
4.1. -correlator
d) which at
u d
Here, we investigate the correlator of currents j = 1/2(u
Q Q0 and up to dimension 6 has the following conventional OPE [1]

i d 4 eiqx T j (x)j (0)

= q q g q 2

Q2
s (Q2 )
1
log 2
2 1+
8
Q0

1 1
1 s a a
+ 4
+ md dd
+
F
mu uu
F
Q0
24 Q0
Q 2

s (Q2 ) 1
5 t a d 2
u
5 t a u d
Q0
6
2
Q

1
t a d
(19)
+
q
t a q
.
u
t a u + d
9
q=u,d,s
Q0
After application of the vacuum saturation hypothesis [1], going to the SU(2) chiral
limit mu = md = 0, and implementing the non-perturbative coarse graining of operator
averages3 the restriction Q Q0 can be dropped, and the scalar part of the correlator
(curly brackets on the r.h.s. of Eq. (19)) reduces to

1
s (Q2 )
Q2
Q2 = 2 1 +
log 2
8
Q0

1
1
4
1
s a a
F F
+
exp
24Q4
5g Q Q0
Q0
3 At dimension 6 there is logarithmic running of the Wilson coefficients due to the non-vanishing anomalous
dimensions of the corresponding operators. Together with the running coupling s (Q2 ) these dependences almost
cancel [1], and hence we will omit them throughout what follows. Operators of dimension 4 are perturbatively
invariant under the renormalization group. We take the value s (Q20 = (2GeV)2 ) 0.2. This is motivated by
Ref. [10] where a non-perturbative evolution of s has been obtained in two flavor lattice QCD using the
Schrdinger functional scheme. With the conversion MS = 2.382 and taking MS = 0.4 GeV, one obtains
s (2 GeV)2 0.19. We have varied the coupling up to s = 0.4 but no drastic changes occur.
309

s (Q20 ) 112
8
1
1
exp
qq
2Q0 .
81
5q Q Q0
Q6
(20)
4.2. a1 -correlator
The vacuum averaged OPE for the correlator of the current ja1 = 1/2(u
5 u
5 d) at Q Q0 and up to dimension 6 reads in the SU(2) chiral limit [1]
d

i

d 4 eiqx T ja1 (x)ja1 (0)
= q q g q 2

1
1
s (Q2 )
s a a
Q2
F F
2 1+
log 2 +
8
Q0 24Q4
Q0

2

s (Q )
a 2
a 2
u
t u u
5 t u + (u d) Q .
0
2Q6
(21)
After vacuum saturation and implementation of non-perturbative operator running the

restriction Q Q0 can be forgotten, and we have
a1

Q2
s (Q2 )
1
log 2
= 2 1+
8
Q0

1
1
4
1
s a a
F F
+
exp
24Q4
5g Q Q0
Q0

s (Q20 ) 176
8
1
1
exp
qq
2Q0 .
+
81
5q Q Q0
Q6
(22)
4.3. -correlator
5 d), and in the SU(2) chiral limit its
The pion current is given as j = 1/2i(u
5 u d
correlator can be expanded as [1]


1
Q2
s (Q2 )
s a a
1
2
log 2
F F
1+
= 3Q
16 2
Q0 48Q4
Q0

s (Q2 ) 1
5 t a d 2
u
5 t a u d
Q0
6
12
Q

1
a
a
a
+
q
t q
. (23)
u
t u + d t d
18
q=u,d,s
Q0
310
After vacuum saturation and with non-perturbative operator running (no restriction Q
Q0 anymore) the piece in curly brackets becomes

1
s (Q2 )
Q2
Q2 =
1
+
log
16 2
Q20

1
1
4
1
s a a
F F
exp
48Q4
5g Q Q0
Q0

s (Q20 ) 56
8
1
1
exp
qq
2Q0 .
(24)
5q Q Q0
Q6 81
4.4. -correlator
The -meson current is defined as j = 1/3s s. If we keep the s-quark mass finite,
let mu = md = 0, assume that the s-quark condensate at Q = Q0 does not deviate from the
ones of u or d quarks, then at Q Q0 the OPE of the -correlator becomes [11]


= q q g q 2

1
1
1 s a a
2
s (Q2 )
Q2
2 1+
F F
Q0 +
log 2 + 4 ms qq

9
8
24 Q0
Q0 Q

s (Q2 )
2
a 2
a
a
.
(25)
q
t q
s 5 t s Q + s t s
0
9
Q6
q=u,d,s
Q0
Building in vacuum saturation and non-perturbative running of the operator VEVs, the
restriction Q Q0 does not apply anymore, and the scalar piece in curly brackets becomes

1
Q2
s (Q2 )
2
2
2 1+
log 2
Q =
9
8
Q0

1
4
1
1
ms qq
+ 4 exp
Q0
5q Q Q0
Q

4
1
1
s a a
1
exp
F F
+
24
5g Q Q0
Q
0

2
s (Q0 ) 112
8
1
1
exp
(26)
qq
2Q0 .
Q6
81
5q Q Q0
4.5. Spectral functions
Here, we calculate the spectral functions of the respective channels from the OPE
by
2 ) at analytical continued Q2 = s i or Q = Q2 =
taking
the
imaginary
part
of
(Q
i s with real s and s > 0. As in the conventional sum rule approach we expect to obtain
(A.1)
(A.2)
(A.3)
(A.4)
(B.1)
(B.2)
(B.3)
311
(B.4)
Fig. 4. The spectral functions Im (Q2 s i), (s > 0) for sets (A) and (B), where (1), (2), (3), and (4)
correspond to the , a1 , , and channels, respectively.
312
information on the lowest resonances (but now without the a priori assumption of quark
hadron duality). Fig. 4 shows the results of the calculations for s (Q20 = (2 GeV)2 ) = 0.2
we use ms = 120 MeV. Exact vacuum saturation in the sense of (1) is assumed with the
commonly introduced correction factor k 1 set equal to unity. Due to the truncation
of the OPE at dimension 6 and other errors (large quark mass and NF = 4 in the lattice
calculation, vacuum saturation, chiral SU(2) limit, and a relatively low fundamentality
scale Q0 ) we have to cutoff the spectra at some lower bounds. Since spectral functions
ought to be positive definite natural

candidates s0 for these cutoffs are the first zeros
encountered when decreasing s.

Looking at the spectra belonging to set (A) above these cutoffs, there is no resemblance
of the measured behavior. In contrast, set (B) seems to give a more realistic behavior: there
is higher spectral strength at lower energy in the than there are in the a1 or channels.
In the latter we have varied the strange quark mass: for ms below 50 MeV the channel
behaves -like, that is, there is a large peak at the lower bound of the spectrum. For ms >
50 MeV the -spectrum is a1 -like, that is, the spectral strength is steadily decreasing from
its perturbative value down to its first zero with decreasing energy. Similarily, the channel
exhibits a large concentration of spectral strength at low energies.
Let us process the spectral information contained in Fig. 4 to more quantitative
statements. One may ask where the center-of-mass
s1
M=

(d s ) s Im(s)
s0
s1
(d s ) Im(s)
s0
(27)
of a given spectrum within a low-energy domain s0 s s1 is. This is motivated

by the experimental fact that the lowest resonance dominates the spectrum within a large
range of energies. Note that using a narrow resonance model, the residue of this single
resonance cancels in the definition of M (ratio of moments). We may now define the
universal upper bound s1 of this low-energy domain by demanding

M to coincide with

the mass m = 770 MeV of the lowest -resonance. With s0 = 420 MeV Eq. (27) then
yields s1 = 1.2 GeV. Using this, we calculate M for the other channels with s0 always
being the first zero of the respective spectral function. The result can be compared with the
measured mass of the lowest resonance. We have

Ma1
855
a
s0 1 = 540 MeV,
Ma1 = 855 MeV,
=
0.68,
ma 1
1260

680
M
=
0.94,
s0 = 420 MeV,
M = 680 MeV,
1/2[m + m (1300)] 720

M
960
0.94.
(28)
s0 = 680 MeV,
M = 960 MeV,
=
m
1020
Due to the distinguished role of the -meson as a Goldstone-boson we have taken along
the next pion resonance ((1300)) assuming their residues to be equal. The differences
between the measured (still using the narrow resonance model) and the computed ratios of
moments is about 30% for the a1 channel and less than 10% level for and channels.
313
We do emphasize at this point that conventional OPEs would have given the same value for
M in all channels (apart from small perturbative corrections to the Wilson coefficients).
Why do the shapes deviate considerably from the experimental ones (as a general
feature, they seem to be shifted to lower energies)? Although the use of the small
correlation length in set (B) was motivated by ambiguities concerning vacuum saturation
we do believe that the truncation of the OPE at dimension 6 is the major source of deviation.
It is quite plausible that higher operator dimensions introduce shorter and shorter effective
correlation lengths and therefore higher mass scales to govern the operator VEVs at low
resolution [6]. Viewed in this context, the use of a small correlation length at dimension
6 may be an effective way to simulate higher mass dimensions. Apart from this there are,
of course, the unresolved problems linked to vacuum saturation as such and the way it is
being implemented as they were discussed in the previous section.
5. Implications for the calculation of B

Motivated by the occurrence of a large (effective) mass scale in the previous section we
discuss in this section how the notion of non-perturbative coarse graining of local operator
averages may influence the calculation of non-leptonic, inclusive width differences in
neutral B-meson systems.
s system as it was treated in Ref. [12].
To be specific, we take the example of the Bs B
On the level of an effective weak Hamiltonian, which is obtained by integrating out the
heavy bosons Z 0 , W of the fundamental theory (also including pQCD corrections) and
which, omitting Cabibbo suppressed contributions, is of the form
6

GF
ci O i + c8 O 8 ,
Heff = Vcb Vcs
(29)
2
i=1
the width difference Bs between the mass eigenstates |BH/L can be calculated by virtue
of the optical theorem as (relativistic normalization of states)

1
Bs =
(30)
Bs | Im i d 4 x T Heff (x)Heff |Bs .
2MBs
The first six B = 1 operators of Eq. (29) are of the four quark type, for example,
a ta s .
O1 = (b c )V A (c s )V A , and O8 = g/(8 2 )mb b (1 5 )F

To be able to handle Eq. (30) one expands in position space the T -product into an OPE
which contains B = 2 operators, for example,
Q = (b s )V A (b s )V A .
(31)
In view of the subsequent x integration the OPE only makes sense if the momentum q
of the external states is in the Euclidean (or space-like) domain. So for a meaningful
calculation formally an analytical continuation of the time-like momentum squared q 2
m2b of the heavy mesons at rest to q 2 = Q2 < 0 must be performed. After a determination
of the Wilson coefficients in the Euclidean the result is analytically continued back to
q 2 m2q . Thereby, one has to deal with two renormalization scales 1,2 . The former is
314
due to the expansion of the fundamental, electroweak (together with QCD corrections)
interaction into the effective Hamiltonian, whereas the latter comes in via the separate scale
dependence of the B = 2 operators and Wilson coefficients. It is claimed in Ref. [12] that
in the B = 2 OPE the dependence on 1 is almost cancelled on the level of the Wilson
coefficients at the same order in s . Since the MS-scheme was used in [12] the MS matrix
elements had to be matched to the matrix elements obtained in lattice renormalization at
the low lattice renormalization point 2 GeV. Subsequently, they were run up to mb .
Keeping the scale of the matrix element fixed in a purely perturbative renormalization
group evolution, the transcendental dependences on the external momentum scale are
powers of logarithms due to their resummation. Violating local duality, we have seen
in the previous sections that these logarithmic dependences do not introduce resonance
structure in the spectral functions of light-quark channel vacuum correlators. If the scale
of the process is comparable to the inverse (effective) correlation length then there is a
much stronger dependence of the product of Wilson coefficient and matrix element of a
local operator than the logarithmic one. Along the lines of Section 3 it would, therefore, be
important to measure the 4-quark correlators corresponding to the B = 2 matrix elements
of Eq. (30) in order to decide whether mass scales (inverse correlation lengths) occur which
are dangerous for the heavy quark expansion. This should also be done for contributions
of higher-dimensional operators formally corresponding to higher powers in 1/mb . After
all, the inverse, effective correlation length, used in the OPEs of vacuum correlators in
/
light-quark channels (set (B) in Section 4), is not too small compared to mb (10/3 GeV
4.5 GeV).
6. Summary
In this paper we investigated the consequences of non-perturbative coarse graining of
operator VEVs as it was proposed in Ref. [6]. The focus was on light-quark correlators.
After a brief review of OPE coarse graining, based on the knowledge of nonperturbatively calculated, gauge invariant n-point functions, we addressed the issue of
vacuum saturation in the case of so-called disconnected diagrams. In particular, it was
pointed out that there are ambiguities in the way approximations to the general prescription
are implemented for coarse graining VEVs of local 4-quark operators. Using the simplest
option for vacuum saturation, the machinery was applied to the light-quark correlators
in the , a1 , and channels. The spectral functions of the respective channels were
calculated from the OPE by analytical continuation to time-like external momenta. Using
lattice data on the gauge invariant field strength and scalar quark correlators, the spectra
were found to be far off their experimentally measured behavior. With a 10 times smaller
correlation length at dimension 6 the basic phenomenological features turned out to be
contained in the spectral functions. However, in general the spectra are shifted to lower
energies and resonance information is vague in the a1 and channels. This may be a
consequence of excluded higher mass dimensions and the treatment of the VEVs of 4quark operators. Despite the obvious shortcomings a calculation of the ratios of the first
and zeroth moments of the spectral distributions in the a1 , and channels gave quite
realistic results after fixing this ratio in the channel. Finally, we discussed the potential
315
impact non-perturbative coarse graining may have on the theoretical determination of the
inclusive, non-leptonic in Bs -meson decays.
Acknowledgement
The author would like to thank Uli Nierste for a stimulating conversation. Financial
support from CERNs theory group during a research stay in June are gratefully
acknowledged. The author is indebted to V.I. Zakharov for numerous useful discussions
and valuable comments.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
M. Shifman, A. Vainshtein, V. Zakharov, Nucl. Phys. B 147 (1979) 385.

I. Bigi, N. Uraltsev, hep-ph/0106346.
K.G. Chetyrkin, S. Narison, V.I. Zakharov, Nucl. Phys. B 550 (1999) 353.
M. Shifman, Theory of preasymptotic effects in weak inclusive decays, in: A. Smilga (Ed.), Proc. Workshop
on Continuous Advances in QCD, World Scientific, Singapore, 1994.
M. Shifman, in: J. Bagger et al. (Eds.), Particles, Strings and Cosmology, World Scientific, Singapore, 1996,
hep-ph/9505289;
B. Chibisov, R.D. Dikeman, M. Shifman, N. Uraltsev, Int. J. Mod. Phys. A 12 (1997) 2075, hep-ph/9605465;
B. Blok, M. Shifman, D. Zhang, Phys. Rev. D 57 (1998) 2691;
B. Blok, M. Shifman, D. Zhang, Phys. Rev. D 59 (1999) 019901, Erratum.
R. Hofmann, Phys. Lett. B 520 (2001) 257.
H.G. Dosch, Phys. Lett. B 190 (1987) 177;
H.G. Dosch, Yu.A. Simonov, Phys. Lett. B 205 (1988) 339.
M. DElia, A. Di Giacomo, E. Meggiolaro, Phys. Lett. B 408 (1997) 315;
A. Di Giacomo, E. Meggiolaro, H. Panagopoulos, Nucl. Phys. B 483 (1997) 371;
M. DElia, A. Di Giacomo, E. Meggiolaro, Phys. Rev. D 59 (1999) 054503.
M. DElia, A. Di Giacomo, E. Meggiolaro, Phys. Rev. D 59 (1999) 054503.
A. Bode et al., hep-lat/0105003.
M. Shifman, A. Vainshtein, V. Zakharov, Nucl. Phys. B 147 (1979) 448.
M. Beneke, G. Buchalla, C. Greub, A. Lenz, U. Nierste, Phys. Lett. B 459 (1999) 631;
U. Nierste, hep-ph/0009203.

Charged fermion masses with Yukawa coupling

strength universality
M.D. Tonasse
Instituto Tecnolgico de Aeronutica, Centro Tcnico Aeroespacial, Praa Marechal do Ar Eduardo Gomes 50,
12228-901 So Jos dos Campos, SP, Brazil
Received 14 June 2001; accepted 7 December 2001
Abstract
We consider the problem of the ordinary charged fermion mass patterns in the framework of a
version of the 3-3-1 electroweak gauge model which includes charged heavy leptons. The masses
of the top and the bottom quarks are given at the tree level. All the other charged fermion get
their masses at the one loop level. The Yukawa coupling strength can be universal. The model
is compatible with a mechanism for neutrino mass generation proposed by other authors. 2002
Elsevier Science B.V. All rights reserved.
PACS: 12.10.Kt; 12.15.Ff; 12.60.-i
1. Introduction
The pattern of the fermion masses and mixing angles is one of the major problems of
the standard electroweak model. The standard model is based on SU(3)C SU(2)L U(1)Y
symmetry group and contains only one Higgs scalar doublet of which only one neutral
scalar field remains afterthe symmetry breakdown. In this model, the fermion masses are
given by mf = gf vW / 2, where gf is the Yukawa coupling constant associated with
the fermion f and vW is the vacuum expectation value (VEV) of the Higgs field in the
theory. Therefore, the correct values of the fermion masses in the standard model require
appropriate choice of the Yukawa parameters, making gf to run in a range of O(106 )
(for the electron) to O(1) (for the top quark). All the fermion masses, except to the quark
top one, lie very below of the dynamic scale vW = 246 GeV. Such problems and other
undesired features support the belief that the standard model is only an effective low
E-mail address: tonasse@fis.ita.cta.br (M.D. Tonasse).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 2 6 - 5
M.D. Tonasse / Nuclear Physics B 623 (2002) 316324
317
energy electroweak theory and give motivation for developing extensions and alternative
models.
There are essentially two attempts to understand the fermion mass and mixing patterns.
One of them consists in searching the answer in more fundamental theories such as
grand unified theories [1], supersymmetric leftright [2] and composite models [3], etc.
The alternative road is to construct phenomenological mass matrices, by employing some
specific ansatz, attempting to fit low energy data [4].
In the present paper, we address the problem of the charged fermion masses in the
context of the SU(3)C SU(3)L U(1)N (3-3-1 for short) model in which the weak
and electromagnetic interactions are described through a gauge theory based on the
SU(3)L U(1)N semi simple symmetry group [5]. The most interesting feature of the
model is that the anomaly cancellations occur only when the three fermion generations are
considered together and not family by family as in the standard model. This implies that
the number of families must be a multiple of the color number and, consequently, the 3-3-1
model suggests a route towards the response of the flavor question [5]. The model has also
a great phenomenological interest since the related new physics can occur in a scale near
of the Fermi one. Therefore, since in this model individual lepton number conservation can
be violated, typical 3-3-1 processes which are free of the standard model background can
be studied in the next generation of colliders [6].
There are several versions of the 3-3-1 model and all of them preserve the essential
features discussed above [5,7,8]. In this work we are interested in a version which includes
heavy leptons [8]. It is interesting to notice that these heavy leptons do not belong to any
of the four types of heavy leptons usually considered in the literature, i.e., (i) sequential
leptons, (ii) paraleptons, (iii) ortholeptons, or (iv) long-lived penetrating particles. Hence,
the experimental limits already existing do not apply directly to them [8,9].
Fermion masses and the CabibboKobayashiMaskawa mixing matrix problem have
already been considered in another version of the 3-3-1 model in Ref. [10] where an SU(2)
horizontal symmetry was imposed and all masses of the first family vanish.
2. Basic facts about 3-3-1 heavy lepton model

Let us summarize the most relevant points of the original 3-3-1 heavy lepton model.
The left-handed lepton and quark fields form SU(3)L triplets, i.e.,

a
u1
J

,
Q1L = d1
,
QL = u ,
aL = la
(1)

Pa L
J1 L
d L
which transform as (3, 0), (3, 2/3) and (3 , 1/3), respectively, where la = e, , , and 0,
2/3 and 1/3 are the U(1)N charges [8]. Throughout this work we are using the convention
that the Latin indices run over 1, 2, and 3, while the Greek ones are 2 and 3. Each charged
left-handed fermion field has its right-handed counterpart transforming as a singlet of
the SU(3)L group. Right-handed neutrinos are optional in the model. In order to avoid
anomalies one of the quark families, no matter which, must transform in a different way
with respect to the other two. The J1 exotic quark carries 5/3 units of elementary electric
318
charge, while J2 and J3 carry 4/3 each. The primed fields are the interaction eigenstates,
which are related with the respective mass eigenstates by
aL = Nab bL ,

laL,R
= LL,R
ab lbL,R ,
(2a)

PaL,R
= PL,R
ab PbL,R ,

UaL,R
= UL,R
ab UbL,R ,
(2b)

DaL,R

JL,R
(2c)
= DL,R
ab DbL,R ,
= JL,R
JL,R ,
where N, LL,R , PL,R , UL,R , DL,R and JL,R are unitary mixing matrix. We define the
physical quark fields as

u
d
J1
,
DL,R = s
,
JL,R = J2
UL,R = c
J3 L,R
t L,R
b L,R
with their analogues for the interaction eigenstates.
In the gauge sector the model predicts, in addition to the standard W and Z 0 , the extras
V , U and the Z 0 gauge bosons.
The fermion and gauge boson masses are generated in the model through the three
Higgs scalar triplets
0

+
0
,
= ,
= 1 ,
(3)
=
++
0
+
which transform under the SU(3) group as (3, 0), (3, 1) and (3, 1), respectively. The
neutral scalar fields develop the VEVs 0
= v, 0
= u, and 0
= w, with
2
v 2 + u2 = vW
.
(4)

The pattern of symmetry breaking is SU(3)L U(1)N SU(2)L U(1)Y U(1)em and
so we can expect w v, u. In the original heavy lepton model from Ref. [8] the and
give masses to the ordinary quarks and the charged lepton masses come from the Higgs
multiplet. The masses of all the exotic fermions rise through the triplet. The neutrinos
can gain their masses at the tree level by the Higgs triplet. In the original 3-3-1 heavy
lepton model the full Yukawa Lagrangians, which must be considered are

1
(l)
(P )

+
ij k ()
C
* Gab aiL bj L k + Gab aL lbR Gab aL PbR + H.c.,
Ll =
2
ab
(5a)

(U )

(J )
LQ = Q1L
G1b UbR
+ G(D)
1b DbR + G J1R
b

(J )
(U )
(D)

QL Fb UbR
+ Fb DbR
+
F JR
+ H.c.,
()
(l)
(5b)
(P )
(U )
(U )
(D)
(D)
(J )
where Gab , Gab , Gba , G1b , Fb , G1b , Fb , G(J ) and F are the Yukawa coupling
constants. SU(3)C indices have been suppressed and , and denote the , and
antiparticle fields, respectively.
319
The physical massive eigenstates H1+ , H2+ and H ++ of the charged Higgs bosons of
the model are defined by
+

+
1 v u
G1
1
(6a)
=
,
+
u
v
H1
vW

2+
+
++
++
=
v 2 + w2
v
w
1
u2 + w 2
w
v
u
w
w
u

G+
2
,
H2

G++
,
H ++
(6b)
(6c)
+
++ are the massless charged Goldstone bosons [11].
where G+
1 , G2 and G
We notice that from Eq. (5b) quarks of the same charge are coupled through different
Higgs multiplets in Eq. (3). This leads to the undesired flavor changing neutral currents
(FCNCs) at the tree level with the couplings of the quarks through the Higgs and the extra
neutral gauge boson Z 0 , but not through those to the standard Z 0 . They appear also in
the Yukawa interactions involving exotic quarks. This problem was extensively studied
in several papers where is shown that the GlashowIliopoulosMaiani mechanism can be
implemented and the FCNCs are naturally suppressed in the model [7,12].
3. Charged fermion masses

Let us now study the consequences of the model to the quark and lepton masses. In
this paper we concentrate in the Yukawa sector of the model and we implement a scheme
to generating charged ordinary mass terms with Yukawa universality. Thus, in order to
describe the mass patterns we see from Eq. (5b) that some of them must be generated at
loop level. It is possible since we allow all the exotic fermions to gain their masses at tree
level via the Higgs triplet.
We analyze firstly the quark sector. We note that if the S3 permutation symmetry
operates within each quark family in Eq. (5b) before the symmetry breaking we have
not Yukawa hierarchy in each charge sector of the quarks at the tree level. In addition,
if we impose also S3L S3R permutation symmetry among the quark families and the S3
symmetry among the Higgs multiplets, suplemented by the set of discrete symmetry
,
UR UR ,
DR DR ,
QaL QaL ,
JR JR ,
(7a)
(7b)
the Yukawa coupling strenght universality is implemented in all quark sector. Therefore,
the mass matrices for the 2/3 and the 1/3-charge sector can be written as

v v v
u u u
U = G u u u ,
(8)
D = G v
v
v ,
u u u
v
v
v
320
Fig. 1. One-loop diagram which contributes to the mass matrix of the quarks of 2/3 charge.
respectively. Thus, only two quarks, which we identify as the top and the bottom one, gain
masses at the tree level. Therefore, from Eq. (8) we have
2mt mb
mt + 2mb
(9)
,
u=
.
5G
5G
In the D mass matrix of the Eq. (8) we made a convenient choice of the signs in order to
obtain positive values for the VEVs v and u in Eq. (9). We should noticed that this result
can also be obtained by an appropriated transformation of the fermion fields via the 5
matrix. From Eqs. (4) and (9) we have

1 m2t + m2b
G=
(10)
.
vW
5
v=
The masses of the other quarks can be induced by one loop radiative corrections scheme.
From Eqs. (2b), (2c), (5b) and (6) we constructed the diagrams of the Figs. 1 and 2. It is
easy to see that the contributions of the Fig. 1 to the mass matrix elements of the 2/3charge sector are

v
wG2 R
u
)
m(U
=
U
1c UL
c UL
ab
c
1b I (mJ1 , mH2 ) + 2
8 2 cb v 2 + w2
u + w2

R

L R
JL
J
I
(m
,
m
)
+
J
J
I
(m
,
m
)
(11)
,
J2
H
J3
H
2 2
3 3
where mH2 and mH are the masses of one of the single-charged and of the double-charged
Higgs boson, respectively [see Eq. (6)], ab = 1 and
I (x, y) =
x3
x
ln .
2
2
x y
y
For the 1/3-charge sector the contributions from Fig. 2 are similar, i.e.,

wG2 R
v
u
(D)
mab =
D
1c DL
c DL
cb
c
1a I (mJ1 , mH ) + 2
2
2
2
8
u +w
v + w2

R

L R
JL
J
I
(m
,
m
)
+
J
J
I
(m
,
m
)
,
J2
H2
J3
H2
2 2
3 3
(12)
(13)
Let us discuss now the case of the charged lepton masses. We can see from Eq. (5a) that
in the original 3-3-1 heavy lepton model the ordinary charged leptons can get their masses
from the Higgs triplet in Eq. (3) at the tree level. However, from Eq. (9) the VEV u
321
Fig. 2. One-loop diagram which contributes to the mass matrix of the quarks of 1/3 charge.
Fig. 3. One-loop diagram which contributes to the mass matrix of the charged leptons.
is 2vW / 5 and from Eq. (10) G 0.3, with mt 174 GeV mb [9]. Therefore, the
tree level charged lepton masses would be out of the experimental ranges in our scheme.
However, we can introduce a set of discrete symmetries in the leptonic sector, in addition to
the ones in Eq. (7) for the quarks and scalar fields, which forbids tree level charged lepton
masses, i.e.,
1L i1L ,
L L ,

lR
lR
,

PaR
PaR
.

l1R
l1R
,
(14a)
(14b)
In order to implement Yukawa universality in the leptonic sector, we assume also the S3
symmetries within each lepton family and S3L S3R symmetry among the families. Now,
from the Eq. (5a) one realizes that the charged lepton masses at one loop level can be read
from Fig. 3. The mass matrix elements for the charged leptons are given by
m(l)
ab =
G2 uw
L R
R
I (mPc , mH )1 PL
c L1b d La Pdc .
8 2 u2 + w2 c
(15)
For sake of simplicity we are assuming equal strenght of the Yukawa coupling for the quark
and the lepton sectors.
Let us give a simple numerical example in order to demonstrate the model reliability
to reproduce the correct charged fermion mass patterns. For simplicity we assume all
the parameters to be real. For the standard model parameters we put vW = 246 GeV,
mb = 4.3 GeV and mt = 174 GeV [9]. In addition, we assume for the 3-3-1 model
w = 2000 GeV, mJ = Gw. Considering experimental bounds from other models we
impose mH2 = mH = 50 GeV as reasonable values [9]. In the charge-2/3 quark sector
322
we take the mixing matrix of the right-handed fields [see Eq. (11)] as

0.98
0.0010 0.0020
R
U = 0.0015
0.97
0.12 .
0.00030 0.014
0.90
(16)
L
L
For the left-handed fields we take UL
11 = 0.74, U12 = 0.020, U13 = 0.68. The other elements
of the UL mixing matrix are not necessary in these calculations. In order to obtain the
standard model experimental values of the ordinary charged quark masses we assume that
the mixing in the second and third terms of the right-handed side in Eq. (11) contribute as
L L R
L L R
R
UR
cb c Uc J2 J 2 = Ucb c Uc J3 J 3 = 2.64.
(17)
With these values for the parameters we obtain mu 3 103 GeV and mc 1.5 GeV,
respectively, for the up and charm quark masses. The values of the free parameters were
choosing in a way that one loop level contribution to the quark top mass vanishes. In the
L
L
1/3-charge sector we take DR = UR , DL
11 = 0.34, D12 = 0.77, D13 = 0.55 and
L L R
L L R
R
DR
cb c Dc J2 J 2 = Dcb c Dc J3 J 3 = 0.17.
(18)
These parameters give for the masses of the down and strange quarks md 5 103 GeV
and ms 0.1 GeV, respectively. Radiative contribution to the bottom quark mass is zero.
Taking the limit in which the mixing term in Eq. (15) is a diagonal matrix we can also
reproduce the standard model experimental values for the ordinary charged lepton masses
(l)
(l)
provided the relations m(l)
11 = 0.46 me , m22 = 0.46 m and m11 = 0.46 m are valid. For
simplicity we considering here equal contribution of the three terms in Eq. (15).
Certainly there are several reasonable combination of the free parameters which lead to
the correct phenomenology. The numerical example above is one of them which we are
employing to illustrate the viability of the model.
4. Neutrino masses
Finally, let us comment on how our scheme for charged fermion masses generation can
be connected with a preexisting mechanism for generation of neutrino masses. Models
for neutrino masses generation in the 3-3-1 model are discussed in several papers [13,14].
We show that our scheme can be compatible with the one of the Ref. [14], where the
neutrino mass terms can receive three types of contributions. They come from diagrams
with ordinary charged leptons, with heavy leptons and from diagrams allowed by mixing
between ordinary charged leptons and the heavy ones. If we consider only the contribution
of the heavy leptons we have
2
vu
2
I (mH1 , mH2 ),
m()
(19)
ab = ab mPb mPa 2
w
where mH1 and mH2 are masses of the single charged Higgs bosons and is a parameter
of the Higgs potential. Here we are taking lepton mixings into account in the ab term,
which is not present in the model of the Ref. [14]. Yukawa coupling constants are also
323
included in this matrix. However, mixing from the scalar sector is not contained in it.
Thus, the constant make the role of the scalar mixing. The VEVs v and u are given by
the Eq. (9). For guarantee heavy lepton dominance we take |m2Pb m2Pa | = 10 GeV [14].
Assuming Yukawa universality, with coupling constant given by Eq. (10) we can write,
for simplicity, ab = ab . The other numerical parameters are the ones of the charged
fermion discussion in Section 3. Therefore, from Eq. (19) we have, for example, m()
12
()
m13 0.01 GeV for mP2 mP3 . However, to get bi-maximal neutrino mixing we
()
()
()
2
2
2
3 eV2
require m()
12 m13 0.01 eV and m13 /m23 <matm /<msol , with <matm 10
2
5
2
10
2
and <msol 10
eV or 10
eV [9]. Therefore, from Eq. (19) we must have
()
107 and m23 105 102 eV. These values for the neutrino masses are in
agreement with the recent data on solar and atmospheric neutrinos oscillations [15].
5. Conclusion
In conclusion, we implemented a model for generation of ordinary charged fermion
masses in the context of the 3-3-1 heavy lepton model [8]. The scheme is able to reproduces
the observed hierarchy of the ordinary charged fermion masses without hierarchy in the
Yukawa coupling strength. In the model the spontaneous symmetry breaking occurs in
two steps, SU(3) U(1) SU(2) U(1), governed by the VEV w, and SU(2) U(1)
U(1), governed by v and u. The VEV w is responsible for the masses of the exotic
fermions and v and u for the ordinary charged fermion masses. It is required in this work
that v and u are of the same order of magnitude [see Eq. (9)]. Only the top and the bottom
quarks get masses at the tree level. The difference among the masses in each charge sector
is controlled by the mixing parameters. It is possible to show, by a reasonable choice of
the free parameters, that some masses can be very smaller that the electroweak scale. Our
scheme employs S3 permutation symmetries and the sets of discrete symmetries in Eqs. (7)
and (14). We remember that permutation symmetry was already employed for studies of
the fermion mass and mixing problems in context of other models [16]. In our mechanism,
compared with the standard model and other extended models, it is easier to understand
why the top and the bottom quarks are the heavier known fermions and why the top quark
mass is the only one near to the Fermi scale. We show also that the scheme is compatible
with a mechanism for neutrino mass generation at one loop level proposed by other authors
[14].
Acknowledgements
I would like to thank the Instituto de Fsica Terica, UNESP, for the use of its facilities,
the Fundao de Amparo Pesquisa no Estado de So Paulo (Processo No. 99/07956-3)
for full financial support and the Dr. M.M. Leite for a critical reading of the manuscript.
324
References
[1] S. Dimopoulos, L.J. Hall, S. Rabi, Phys. Rev. Lett. 68 (1992) 1984;
S. Dimopoulos, L.J. Hall, S. Rabi, Phys. Rev. D 45 (1992) 4145;
G.F. Giudice, Mod. Phys. Lett. A 7 (1992) 2429;
H. Arason, D.J. Castano, P. Ramond, E.J. Piard, Phys. Rev. D 47 (1993) 232;
C.H. Albright, S.M. Barr, Phys. Lett. B 452 (1999) 287.
[2] R.N. Mohapatra, J.C. Pati, Phys. Rev. D 11 (1975) 566;
G. Sejanovic, R. Mohapatra, Phys. Rev. 12 (1975) 1502;
H. Fritzsch, P. Minkowski, Nucl. Phys. B 103 (1976) 61;
M. Beg, R. Budny, R. Mohapatra, A. Sirlin, Phys. Rev. Lett. 38 (1977) 1252.
[3] K.S. Babu, J.C. Pati, H. Strenitzer, Phys. Lett. B 256 (1991) 206;
K.S. Babu, J.C. Pati, H. Strenitzer, Phys. Rev. Lett. 67 (1991) 1688.
[4] K. Kang, S.K. Kang, C.S. Kim, S.M. Kim, Neutrino oscillations and lepton flavor mixing, hep-ph/9808419;
P.S. Gill, M. Gupta, Mod. Phys. Lett. A 13 (1998) 2445;
T. Ito, N. Okamura, M. Tanimoto, Phys. Rev. D 58 (1998) 077301.
[5] F. Pisano, V. Pleitez, Phys. Rev. D 46 (1992) 410;
P.H. Frampton, Phys. Rev. Lett. 69 (1992) 2889;
R. Foot, O.F. Hernandez, F. Pisano, V. Pleitez, Phys. Rev. D 47 (1993) 4158.
[6] F. Cuypers, S. Davidson, Eur. Phys. J. C 2 (1998) 503;
Y.A. Coutinho, P.P. Queirz Filho, M.D. Tonasse, Phys. Rev. D 60 (1999) 115001.
[7] J.C. Montero, F. Pisano, V. Pleitez, Phys. Rev. D 47 (1993) 2918.
[8] V. Pleitez, M.D. Tonasse, Phys. Rev. D 48 (1993) 2353.
[9] D.E. Groom et al., Particle Data Group, Eur. Phys. J. C 15 (2000) 1.
[10] M.B. Tully, G.C. Joshi, Mod. Phys. Lett. A 13 (1998) 2065.
[11] M.D. Tonasse, Phys. Lett. B 381 (1996) 191.
[12] See also: M. zer, Phys. Rev. D 54 (1996) 4561.
[13] J.C. Montero, C.A.S. Pires, V. Pleitez, Phys. Lett. B 502 (2001) 167;
J.C. Montero, C.A.S. Pires, V. Pleitez, Seesaw tau and calculable neutrino masses in a 3-3-1 model, hepph/0103096;
P.H. Frampton, P.I. Krastev, J.T. Liu, Mod. Phys. Lett. A 9 (1994) 761;
T. Kitabayashi, M. Yasu, Nearly bimaximal neutrino mixing in an SU (3)L U (1)N gauge model with
radiative neutrino masses, hep-ph/0006040;
T. Kitabayashi, M. Yasu, Radiatively induced neutrino masses and oscillations in an SU (3)L U (1)N
gauge model, hep-ph/0010087.
[14] Y. Okamoto, M. Yasu, Phys. Lett. B 466 (1999) 267.
[15] C. Athanassopoulos et al., LNSD Collaboration, Phys. Rev. Lett. 81 (1998) 1774;
C. Athanassopoulos et al., LNSD Collaboration, Phys. Rev. Lett. 77 (1996) 3082;
K.S. Hirata et al., Kamiokande Collaboration, Phys. Rev. Lett. 77 (1996) 1683;
Abdurashitov et al., SAGE Collaboration, Phys. Rev. C 60 (1999) 055801;
Abdurashitov et al., SAGE Collaboration, Phys. Rev. Lett. 77 (1996) 4708;
B.T. Cleveland et al., Homestake Collaboration, Astrophys. J. 496 (1998) 505;
Y. Fukuda et al., Super Kamiokande Collaboration, Phys. Rev. Lett. 82 (1999) 2644;
Y. Fukuda et al., Super Kamiokande Collaboration, Phys. Rev. Lett. 81 (1998) 1562;
W. Hampel et al., GALLEX Collaboration, Phys. Lett. B 477 (1999) 127.
[16] Some recent and early papers are:
S. Pakvasa, H. Sugawara, GALLEX Collaboration, Phys. Rev. Lett. B 73 (1978) 61;
E. Derman, Phys. Rev. D 19 (1979) 317;
K.S. Babu, R.N. Mohapatra, Phys. Rev. Lett. 64 (1990) 2747;
H. Fritzsch, D. Holtmannsptter, Phys. Lett. B 338 (1994) 290;
G. Cvetic, Phys. Rev. D 51 (1995) 201;
J.I. Silva-Marcos, Breaking democracy with non renormalizabe mass terms, hep-ph/0102079;
G.C. Branco, J.I. Silva-Marcos, The symmetry behind extended flavor democracy and large leptonic mixing,
hep-ph/0106125.

Pairs of charged heavy-leptons from an

SU(3)L U(1)N model at CERN LHC
J.E. Cieza Montalvo a , M.D. Tonasse b
a Instituto de Fsica, Universidade do Estado do Rio de Janeiro, Rua So Francisco Xavier 524,
20559-900 Rio de Janeiro, RJ, Brazil

b Instituto Tecnolgico de Aeronutica, Centro Tcnico Aeroespacial,
Praa Marechal do Ar Eduardo Gomes 50, 12228-901 So Jos dos Campos, SP, Brazil
Abstract
One of the versions of the SU(3)L U(1)N electroweak model predicts charged heavy-leptons
which do not belong to any class of heavy-leptons proposed up to now. We investigate the production
and signatures of pairs of these heavy-leptons via the DrellYan process and the gluongluon fusion
at the CERN Large Hadron Collider (LHC). As an example we study the decay of the exotic leptons
into another ones. We see that the lifetime of the exotic leptons can be short. 2002 Elsevier Science
B.V. All rights reserved.
PACS: 12.60.-i; 12.60.Cn; 13.10.+q; 14.60.Hi; 14.80.-j
1. Introduction
Measurements of the total width of the Z neutral gauge boson of the standard model at
CERN and SLAC colliders provide an undeniable evidence for only three neutrino flavors
[1]. CP-violation mechanism and big-bang nucleosynthesis, in the framework of the standard model, reinforce the idea that in the Nature there are only three families of fermions.
However, since the standard model must be only a low energy effective electroweak
theory, there are several motivations to consider the possibility of additional quarks and
leptons. For example, grand unified theories such as SO(10) and E(6) incorporate new
fermions naturally [2]. In some extensions of the standard model, anomalies are canceled
Submitted to NPB on 11 October 2001.

E-mail address: tonasse@fis.ita.cta.br (M.D. Tonasse).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 4 3 - 5
326
J.E. Cieza Montalvo, M.D. Tonasse / Nuclear Physics B 623 (2002) 325341
by introducing additional fermions [3] and this new particles can play a role in CP-violation
schemes [4]. Heavy-leptons phenomenology are widely studied in many extended electroweak models, such as supersymmetric [5], grand unified theories [6], technicolor [7],
superstring-inspired models [8], mirror fermions [9], etc. All these models predict the existence of new particles with masses around of the scale of 1 TeV and they consider the
possible existence of new generations of fermions.
The standard electroweak model provides a very satisfactory description of most
elementary particle phenomena up to the presently available energies. However, there are
some unsatisfactory features as the family number and their complex pattern of masses and
mixing angles, which are not predicted by the model. In addition, the recently emerged
experimental data on the muon anomalous magnetic moment [10], solar and atmospheric
neutrinos [11] strongly suggest physics beyond the standard model. Searches on new
fermions can reveals some directions for the solution of these problems.
In this paper we addresses to a new class of heavy-fermions, i.e., the heavy-leptons
which can be produced by the strong and electroweak processes which emerge from a
model based on the SU(3)C SU(3)L U(1)N (331 for short) semi simple symmetry
group [12]. In this model the new leptons do not require new generations, as occur in
the most of the heavy-lepton models [13]. This ones is a chiral electroweak model whose
left-handed charged heavy-leptons, which we denote by Pa = E, M and T (a = 1, 2, 3),
together with the associated ordinary charged leptons and its respective neutrinos, are
accommodate in SU(3)L triplets. We study the production of these charged heavy-lepton
pairs in hadronic collisions at the CERN Large Hadron Collider (LHC). This process will
be studied through the well known DrellYan mechanism, i.e., the quarkantiquark fusion
q q P P + (Fig. 1(a)) and the gluongluon fusion gg P P + (Fig. 1(b)).
We remember at first that the heavy-leptons appearing in the literature up to now can
be classified in four kinds [1]: (a) sequential leptons, in that the new leptons are associated
with new neutrinos, forming new SU(2)L doublets. This new leptons and its neutrinos
have the same leptonic numbers which are conserved in all interactions; (b) paraleptons,
Fig. 1. Feynman diagrams for production of charged heavy-lepton pairs via (a) DrellYan process and
(b) gluongluon fusion.
327
in that the new leptons have the same leptonic numbers of the associated ordinary charged
leptons of opposite charge; (c) ortholeptons, in that the new leptons have the same leptonic
numbers of the associated ordinary charged leptons of the same charge and (d) long-lived
penetrating particles. As we mentioned above there are very motivations to search heavyleptons in electroweak extensions of the standard model. However, in 331 model we
have in addition two particular motivations. Firstly, as we will see in the next section, the
331 heavy-leptons do not belong to any of these types of new leptons. Consequently, the
existing experimental bounds on heavy-lepton parameters do not apply to them. Secondly,
there is nothing on the 331 heavy-lepton phenomenology up to now.
The outline of this paper is the following. In Section 2 we present the relevant features
of the model. The production of a pair of 331 heavy-leptons in pp colliders are discussed
in Sections 3 and 4 we summarize our results and conclusions.
2. The 331 heavy-lepton model

In 331 model the strong and electroweak interactions are described by a gauge
theory based on the SU(3)C SU(3)L U(1)N semi simple symmetry group. In its
original version new leptons are not required, since the lepton representation content of
each SU(3)L triplet consists of one charged lepton, its charge conjugated counterpart
and the associated neutrino field [14,15]. However, the version of the model which we
work here differs from the original one, in that the charge conjugated lepton fields of
the original model are replaced by heavy-leptons in the SU(3)L triplets [12]. The most
interesting feature of this class of models is related with the anomaly cancellations, which
is implemented only when the three fermion families are considered together and not
family by family as in the standard model. This implies that the number of families must be
a multiple of the color number and, consequently, the 331 model suggests a route towards
the response of the flavor question [14]. The model has also a great phenomenological
interest since the related new physics can be expected in a scale near of the Fermi one [16].
Let us summarize the most relevant points of the model (for details see Refs. [12,14,
15]). In its heavy-lepton version the left-handed leptons and quarks transform under the
SU(3)L gauge group as the triplets

u1
a
2
,
(3, 0),
Q1L = d1 3,
aL = a
3
Pa L
J1 L

J
1
QL = u
(1a)
3 , ,
3
d L
where Pa = E , M , T are the new leptons, a = e , , and = 2, 3. The J1 exotic
quark carries 5/3 units of elementary electric charge while J2 and J3 carry 4/3 each. In
Eq. (1) the numbers 0, 2/3 and 1/3 are the U(1)N charges. Except the neutrino fields,
which we are considering massless here, each left-handed fermion has its right-handed
counterpart transforming as a singlet in the presence of the SU(3)L group, i.e.,
R (1, 1),
PR (1, 1),
UR (1, 2/3),
(1b)
328
DR (1, 1/3),
J1R (1, 5/3),

J2,3R
(1, 4/3).
(1c)
We are defining U = u, c, t and D = d, s, b. In order to avoid anomalies, one of the quark

families must transforms in a different way with respect to the two others. In Eq. (1) all
the primed fields are linear combinations of the mass eigenstates. The charge operator is
defined by
Q 1
= 3 3 8 + N,
(2)
e
2
where the s are the usual Gell-Mann matrices. We notice, however, that since QL in
Eq. (1a) are in antitriplet representation of SU(3)L , the antitriplet representation of the
Gell-Mann matrices must be used in Eq. (2) in order to get the correct electric charge for
the quarks of the second and third generations.
The three Higgs scalar triplets
0
+
= 1 (3, 0),
= 0 (3, 1),
++
2+

= (3, 1),
(3)
0
generate the fermion and gauge boson masses in the model. The neutral scalar fields
develop the vacuum expectation values (VEVs) 0 = v , 0 = v and 0 = v , with
2 = (246 GeV)2 . The pattern of symmetry breaking is
v2 + v2 = vW

SU(3)L U(1)N SU(2)L U(1)Y U(1)em

and so, we can expect v v , v . The and scalar triplets give masses to the ordinary
fermions and gauge bosons, while the scalar triplet gives masses to the new fermions and
new gauge bosons. The most general, gauge invariant and renormalizable Higgs potential is

2
V (, , ) = 21 + 22 + 23 + 1 + 2 + 3

+ 4 + 5 + 6 + 7

+ 8 + 9 + 10
1
+ f - ij k i j k + H.c. .
(4)
2
Here f is a constant with dimension of mass and the i (i = 1, . . . , 10) are adimensional
constants with 3 < 0 from the positivity of the scalar masses. The term proportional to
10 violates lepto-barionic number and so, it was not considered in the analysis of the
Ref. [17] (another analysis of the 331 scalar sector are given in Ref. [18] and references
cited therein). We can notice that this term contributes to the mass matrices of the charged
scalar fields, but not to the neutral ones. However, can be checked that in the approximation
v v , v we can still work with the masses and eigenstates given in Ref. [17]. Here this
term is important to the decay of the lightest exotic fermion. Therefore, we are keeping it
in the Higgs potential.
329
Symmetry breaking is initiated when the scalar neutral fields are shifted as =
v + + i , with = 0 , 0 , 0 . Thus, the physical neutral scalar eigenstates H10 ,
H20 , H30 and h0 are related to the shifted fields as

0

H1
1 v v
,
H30 ,
(5a)
h0 ,
vW v v
H20
and in the charged scalar sector we have
1+
v +
H ,
vW 1
v +
H ,
vW 2
++
v ++
H ,
v
(5b)
with the condition that v v , v [17]. From Eqs. (1) to (5) we can notice that the
ordinary quarks couple only through H10 and H20 and the heavy-leptons and quarks couple
only through H30 and h0 . So, it is easy to see that in this regime there will be no contribution
of the Higgs bosons to the diagrams analogous to one of the Fig. 1. Due the transformation
properties of the fermion and Higgs fields under SU(3)L [see Eqs. (1) and (3)] the Yukawa
interactions in the model are

PbR
+ H.c.,
LY = Gab aL bR Gab aL
(6a)

1a DaR
a DaR
L Fa UaR
1L G1a UaR
LYq =
+G
+
+ F
Q
Q
a
J

F
QL JR
+ G Q 1L J1R + H.c.
J
(6b)
F s and F
s are Yukawa coupling constants with a, b = 1, 2, 3 and
The Gs, Gs,
, = 2, 3. The interaction eigenstates which appear in Eq. (6) can be transformed in
the corresponding physical eigenstates by appropriated rotations. However, since the cross
section calculation imply summation on flavors (see Section 3) and the rotation matrix
must be unitary, the mixing parameters have not essential effect for our purpose here. So,
thereafter we suppress the primes notation for the interactions eigenstates.
The gauge bosons consist of an octet Wi (i = 1, . . . , 8) associated with SU(3)L and a
singlet B associated with U(1)N . The covariant derivatives are
D a = a + i
g
b
b + ig N a B ,
W .
a
2
(7)
where = , , . The model predicts single charged (V ), double charged (U ) vector

bileptons and a new neutral gauge boson (Z ) in adition to the charged standard gauge
bosons W and the neutral standard Z. We take the physical eigenstates of the neutral
gauge bosons from Ref. [14],

3 3
W 3 W8 tW + B ,
A =
(8a)
f (tW )

3
3
1
3
2
8
2
Z
(8b)
1 + 3tW W +
t
W

B
,
2 W
f (tW )
1 + 3tW
2
1 + 3tW
330
1
W + 3 tW B .
Z
2
1 + 3tW
(8c)
Here the Z and Z eigenstates are valid in the approximation v v , v . We can see that
in this approximation Z does not interacts with the standard W gauge bosons. In Eq. (8)
2
tW
=
sin2 W
(9)
1 4 sin2 W
2 ). We observe that the t parameter in Eq. (9) has a Landau pole

and f 2 (tW ) = 3(1 + 4tW
W
which imposes sin2 W < 1/4. It is another good feature of the model, since evolving the
Weinberg angle sin2 W to high values it is possible to find an upper bound to the masses
of the new charged gauge bosons [15,19]. It does not occur in the popular extensions of
the standard model. Another consequence of this pole is an enhancement of the couplings
of the Z to quarks and to the leptons.
The trilinear interactions of the Z with the V and U bileptons has strength
g
= i
2
3
2
1 + 3tW
(k1 k2 ) g + (k2 k3 ) g + (k3 k1 ) g
(10)
with the quadrimoments defined in Fig. 2. The relevant neutral vector current interactions
are
LAP = ePa Pa A ,
(11a)
LZP = g sin W tan W Pa Pa Z ,

2

g tan W 2
Pa 3tW 1 + 3tW
+ 1 5 Pa Z ,
LZ P =
2 3 tW

g
qa v a + a a 5 qa Z ,
LZq =
4 cos W a

g
LZ q =
qa v a + a a 5 qa Z ,
4 cos W a
= V , U ).
Fig. 2. Trilinear interaction among Z and the exotic charged gauge bosons (X
(11b)
(11c)
(11d)
(11e)
331
where W is the Weinberg mixing angle and qa is any quark [12,14]. The coefficients in
Eqs. (11d) and (11e) are
vU =
2
3 + 4tW
2 )
3(1 + 4tW
vD =
2
3 + 8tW
2 )
3(1 + 4tW
a U = a D = 1,
2
1 + 8tW
,
f (tW )
2
2
1 2tW
1 + 2tW
f (tW )
,
v d =
,
v s = v b =
v c = v t =
,
f (tW )
f (tW )
(3)
1
a u =
,
f (tW )
2
1 + 6tW
a c = a t =
,
a d = a c ,
a s = a b = a u ,
f (tW )
2 )
2(1 7tW
,
v J1 =
f (tW )
2 )
2 )
2(1 5tW
2(1 + 3tW
,
a J2 = a J2 = a J2 = a J1 =
.
v J2 = v J3 =
f (tW )
f (tW )
v u =
(12a)
(12b)
(12c)
(12d)
As we comment in Section 1 and by inspection of Eqs. (1), (11b) and (11c), we conclude
that the heavy-leptons Pa belong to another class of exotic particles differently of the
heavy-lepton classes usually considered in the literature. Thus, the present experimental
limits do not apply directly to them [1] (see also Ref. [12]). Therefore, the 331 heavyleptons phenomenology deserves more detailed studies.
The charged lepton current interactions are
g

aL aL W + P aL bL V+ + aL PbL
U + H.c.
LCC
=
2 a
(13a)
For the first and second quark generations we have
g

+
LCC
(13b)
+ H.c.,
Q1 = u L dL W + J1L uL V + d L J1L U
2
g

+

LCC
(13c)
+ H.c.,
Q2 = c L dL W s L J2L V + c L J2L U
2
while the charged current interactions for the quarks of the third generation are obtained
from those of the second generation replacing c t, s b and J2 J3 .
Also, the main decay modes of the new leptons are among the exotic leptons themselves
such as T + E + e , considering that MT > ME . The heavy-leptons can decay also
via single or double charged bileptons in a standard lepton, a standard quark and an
exotic quark (see Fig. 3). Detailed analysis of the 331 heavy-lepton decays will be
given elsewhere [20]. The lightest exotic fermion of the model can decay in the ordinary
particles via the Higgs sector. Looking for the term proportional to 10 in the Higgs
potential (4), we see that after the symmetry breaking this term leads to the couplings
10 1+ (v + + v 1+ ), with the charged scalar fields given by Eq. (5). From Eq. (6a)
332
Fig. 3. Decay channels for the 331 heavy-leptons (a) via V + and (b) via U ++ .
Fig. 4. Example of how an exotic 331 particle (the heavy-leptons Pa ) can decay in standard particles via Higgs
sector. Here U = u, c, t.
it is easy to see that the scalar couples the exotic leptons Pa to ordinary ones a
and from Eq. (6b) the 1+ couples ordinary quarks of 1/3 charge to ones of 2/3 charge.
Therefore, the decay of the exotic leptons Pa can proceeds as in the diagram of the Fig. 4.
Inspection of Eq. (6b) tell us also that the exotic quarks can decay in a similar form as if
in the Fig. 4, where the ordinary leptons a would be replaced by the appropriate standard
quarks. Therefore, the lightest exotic fermion can be not stable.
3. Cross section production

To calculate the cross-section, we begin suppressing the family indices of the Section 2,
i.e., we assume P = E, M, T ; = e, , and J = J1 , J2 , J3 . We will study the mechanism
333
of the DrellYan production of pair of heavy-leptons, that is, we analyze the process
pp q q P P + (Fig. 1(a)). This process take place through the exchange of the
bosons Z, Z and in the s channel. Using the interaction Lagrangians of the Eq. (11) and
the parameters of the Eq. (12), we evaluate the differential cross section for the subprocess
q q P P + obtaining

d
d cos P + P

2
2 2
2
2 eq2
2
2sM
=
+
M
t
+
M
u
P
P
P
Nc s 2 s
eq
2
2
2
2 sin W cos W (s MZ,Z
+ iMZ,Z Z,Z )

2
q
q
2sMP2 gVP P gV + gVP P gV MP2 t + MP2 u

2
P P q
gA MP2 u MP2 t
+ gA
2
2
2
16Nc
W sin W s(s MZ,Z + iMZ,Z Z,Z )

2 P P
2 q
2 q
2
gV + gA
gVP P + gA
cos4
2
MP2 u + MP2 t

2 P P
2 q
2 q
2
gV + gA
+ 2sMP2 gVP P gA

2
2
P P q q
gV gA MP2 u MP2 t
+ 4gVP P gA
8Nc sin4 W cos4 W s(s MZ2 + iMZ Z )(s MZ2 + iMZ Z )

q
q

P P P P
2sMP2 gV + gA gVP P gVP P gA
gA
2 q
2 q
2
P P P P
P P P P
gV gV + gA
+ MP2 t
gA
gV + gA

q q
q q P P P P
P P
2gV gA gA
gV
2gV gA gVP P gA
2 q
2 q
2
P P P P

P P P P
gV + gA
gV gV + gA
gA
+ MP2 u

q q
q q P P P P
P P
+ 2gV gA gVP P gA
+ 2gV gA gA
gV
,
(14)
where
a aR
aL aR
,
gV,P,AP = L
.
(15)
2
2
Here the primes () are forthe case when we take a boson Z , Z,Z is the total width of
P
gVP ,A
=
the boson Z and Z , =
1 4MP2 /s is the velocity of the heavy-lepton in the c.m. of
334
q
the process, eq is the electric charges of the quark q, Nc is the number of colors, gV ,A are
the standard quark coupling constants, MZ is the mass of the Z boson, s is the center of
mass energy of the q q system, t = MP2 (1 cos )s/2 and u = MP2 (1 + cos )s/2,
where is the angle between the heavy-lepton and the incident quark in the c.m. frame.
For Z boson we take MZ = (0.5 3) TeV, since MZ is proportional to VEV v
[14,15]. For the standard model parameters we assume PDG values, i.e., MZ = 91.19 GeV,
sin2 W = 0.2315 and MW = 80.33 GeV [1].
The total width of the Z boson into exotic leptons, standard leptons, neutrinos, standard
and exotic quarks and new vector bosons are
(Z all) = Z P P + + Z + + Z + Z q q(J
J) + 2Z X X + ,
(16)
where X = V or U and we have for everyone
Z P P + =
Z + =

1 4MP2 /s
12MZ sin2 W cos2 W

2
P P
2

2
P P
2
2MP2 gVP P 4MP2 gA
+ MZ2 gVP P + MZ2 gA
,

MZ
2
12 sin W cos2 W
(17a)
2
2
gV + gA
,
MZ
Z =
,
18h sin2 W cos2 W

1 4Mq2 /s
Z q q(J
J) =
16MZ sin2 W cos2 W

qq
2
qq
2
qq
2
qq
2
2Mq2 gVi 4Mq2 gAi + MZ2 gV
+ MZ2 gA
,
Z Y Y + =
1
(17b)
(17c)

MZ2 (MY1 + MY2 )2 MZ2 (MY1 MY2 )2
8MZ3 sin2 W (1 + 3t 2 )
MZ6
4MY21 MY22
+
MY61
4MZ2 MY22
+ 4MY21 +
2MZ4
MY21
MY41
MZ2
2MY42
MY21
MZ4
MY22
3MZ2 MY21
2MY22
3MY21 MY22
2MZ2

5MY22 ,
9MZ2 MY22
2MY21
MY62
4MZ2 MY21
5MZ2
MY42
MZ2
MY41
MY22
(17d)
335
with h = 1 + 4 tan2 W . In Eq. (17d) Y1 and Y2 are any vector bosons and we take for our
case MY1 = MY2 = MX . Therefore we will have

6

1 4MX2 /s
MZ
MZ4
2
Z X X+ =
(18)
+ 2 8MZ .
MX
8MZ sin2 W (1 + 3t 2 ) 4MX4
The total cross section for the process pp qq P P + is related to the subprocess
qq P P + total cross section , through
1
ln min
d dy q
y 2
y 2
e , Q q e , Q (, s),
(19)
min ln min
where = (min = 4MP2 /s)s /s and q(x, Q2 ) is the quark structure function.
Another form to produce a pair of heavy-leptons is via the gluongluon fusion, namely,
through the reaction of the type pp gg P P + . Since the final state is neutral, the s
channel involves the exchange of the two neutral Higgs boson H3 (h) and Z . The exchange
of a photon is not allowed by C conservation (Furrys theorem), which also indicates that
only the axial-vector couplings of the neutral gauge boson Z (Z ) contribute to this process.
Moreover the axial-vector coupling of the Z to the exotic lepton is equal to zero. Therefore,
there is not contribution of this particle to the total cross section. The interference of the Z
with the H3 and h vanishes, since the first one is antisymmetric in the gluon polarization,
while the latter are symmetric. Therefore, the differential cross section for this reaction can
be expressed as
d
1
(20)
=
|MZ |2 + |MH3 |2 + |Mh |2 + 2 Re MH3 Mh .
2
d
64 s
Writing separately the cross-section for the subprocesses, we will have first for the Z

2
P P )2 2 2

(gA
MP2
q
q
s
Z =

gA (1 + 2q Iq ) ,
(21)
4
4
4
512 sin W cos W MZ mq =mu ,md ,mJ
where the summations run over all generations and q = m2q /s . For the H3 we have
2

s2 s (4s 5MP2 ) MP2
q
H3 =
(22)
3 (s )
2q + q (4q 1)Iq ,
3
4
32768
v
q=J
where the summations run only over J1 , J2 and J3 . We also define

i (s ) =
2
MH
i
1
+ iMHi Hi
with Hi being the Higgs-boson total width. The H3 J J decay widths are

2 4M 2 M 2
MH
J
J
3
2
2MH
5MJ2 .
(H3 J J) =
3
2
16MH3 v2
(23)
(24)
336
On the other hand the H3 P P decay widths are

2 4M 2 M 2
MH
P
P
3
2

(H3 P P ) =
4MH
5MP2 .
3
2
2
64MH3 v
(25)
q
We notice here that the cross section h and the decays h J J and h P P give
contributions analogous to (22), (24) and (25), respectively, with MH3 replaced to Mh .
The loop integrals involved in the evaluation of the elementary cross section can be
expressed in terms of the function Ii (i ) Ii which is defined through

(1 x)x
dx
ln 1
x
i
1
Ii =
0

=
2

,
2 sin1 21
i > 1/4,
(26)
ln (r+ /r ) + i ln(r+ /r
i < 1/4,
with r = 1 1 4i . Here, i = q stands for the particle (quark) running in the loop.
The diagram with a Z in the s channel does not exhibit the resonance for s = MZ2 ,
since the on shell production of a massive spin-one particle on shell from two massless
spin-one particle is forbidden (Yangs theorem) [21]. The total cross section for the process
pp gg P P + is related to the subprocess gg P P + total cross section through
i
1
2
1
ln min
=
min
) 2 /2,
d dy G
y 2
y 2
e , Q G e , Q (, s),
(27)
ln min
where G(x, Q2 ) is the gluon structure function.

As an example for the decay of the heavy exotic leptons we have T E e , i.e.,
(T E e ) =
G2 MT5 MW
1 8 + 8 2 4 12 2 ln ,
4
3
192 MV
(28)
where = ME2 /MT2 and MV is the mass of the new single charged gauge boson which can
run from 200 GeV to 3000 GeV [19].
4. Results and conclusions

The process for the heavy-lepton production in hadronic colliders was well studied in
the literature and was shown that the dominant contribution are the well known DrellYan
process (Fig. 1(a)) and gluongluon fusion (Fig. 1(b)) [5,22].
We present the cross section for the process pp P P + , involving the DrellYan
mechanism and the gluongluon fusion, to produce the 331 heavy-leptons. In Fig. 5
we show the different cross sections for different masses of neutral boson Z , H3 and h.
We see that the DrellYan and the gluongluon fusion production are suppressed for
337
Fig. 5. Total cross section for the process pp P P + as a function of MP at s = 14 TeV using the mechanism
of the DrellYan and of the gluongluon fusion. For the DrellYan case we take the mass of the boson Z equal
to 0.6 TeV (solid line) and 2 TeV (dot dot dashed line) and for the gluongluon fusion we have for the mass of
the boson Z equal to 0.6 TeV (med dashed line) and 2 TeV (dotted line).
2MP > MZ (MH3 , Mh ) where the Z , the H3 and h later must produce the heavy-lepton
pairs. This figure still exhibit the resonance peaks associated with the boson Z and the two
neutral Higgs bosons. Considering that the expected integrated luminosity for the LHC will
be of order of 105 pb1/yr and taking the mass of the heavy-lepton equal to 300 GeV, it
could be seen in the LHC, that the DrellYan production dominates over gluongluon
fusion. So for the DrellYan mechanism we have a total of 8.8 104 and 1.7 104
heavy-leptons produced per year if we take the mass of the boson Z equal to 0.5 TeV
and 2 TeV. For both figure was taken, MJ1 = 200 GeV, MJ2 = 300 GeV, MJ3 = 500 GeV
and for the mass boson MV = 800 GeV. For the gluongluon fusion we will have a total
of 2 103 and 1.8 103 heavy-leptons produced per year if we take the masses of the
Higgs H3 (h) equal to 600 (600) GeV and 400 (800) GeV, respectively. Here we take for
the mass of the boson Z and the quarks the same values as for the DrellYan case. In
addition we take for v = 700 GeV.
From Eq. (21) it is to see that the cross-section via gluongluon fusion is very sensitive
to the vacuum expectation value, so if we take v = 500 GeV and the same values as was
taken above for the mass of the boson MZ = 2000 GeV, we will have a total of 3.5 103
338
Fig. 6. Total cross section for the process pp P P + vs. s in according with the mechanism of the DrellYan
(solid line) and the gluongluon fusion (dashed line). Here was taken for the mass of the heavy exotic leptons
equal to 600 GeV.
heavy leptons produced per year, that is nearly the value of 1.7 104 , that we have received
for the case of DrellYan mechanism.
The total cross sections for exotic heavy-leptons
produced by the mechanisms of Drell
Yan and of gluongluon fusion vs. s are showed in Fig. 6. Here was taken for the
mass of the heavy exotic leptons MP = 300 GeV, for the Higgs masses MH3 = 400 GeV,
Mh = 800 GeV (for the other parameters was taken the same values as in Fig. 5). From this
figure it is to see that for very high energy we will have for every heavy lepton produced
via gluongluon fusion, five or six leptons produced via DrellYan.
The Fig. 7 shows the partial width vs. the mass of the decaying particle for the process
T E e , considering that the mass of the boson V runs from 500 GeV to 1200 GeV.
The Fig. 8 shows the T -lepton lifetime vs. MT . This figure exhibits that the heavy-lepton
can be short-lived for the masses of the boson V here considered.
1 (uJ1 ), could comes
The main background for the signal, q q P P + uJ

+

from the process q q W W (Z Z) q q q q (q qq
q),
but this background can be
eliminated by measuring the undetected neutrino contribution pT , since all hadrons with
appreciable pT are detected. Then the overall pT imbalance for detected particles gives
this undetected pT . We could have another backgrounds such as q q W W + (Z Z)
but this backgrounds also can be eliminated, then the number of jets
eq q (e+ e q q),
are different. The very striking signal comes from the reaction Pa+ + Pb+ since it
consists of three charged particles.
In summary, we showed in this work that in the context of the 331 model the signatures for heavy-leptons can be significant in LHC collider. Our study indicates the possibility of obtaining a clear signal of these new particles with a satisfactory number of events.
339
Fig. 7. The partial decay widths of the heavy-lepton T vs. MT for several masses of the new single charged gauge
boson V : for MV = 0.5 TeV (dashed line), MV = 0.8 TeV (dot dot dashed line) and MV = 1.2 TeV (dotted line).
Fig. 8. Lifetime of the heavy-lepton T vs. the MT for several masses of the new gauge boson V : MV = 0.5 TeV
(dashed line), MV = 0.8 TeV (dot dashed line) and MV = 1.2 TeV (dotted line).
340
Acknowledgements
One of us (M.D.T.) would like to thank the Instituto de Fsica Terica of the UNESP,
for the use of its facilities and the Fundao de Amparo Pesquisa no Estado de So Paulo
(Processo No. 99/07956-3) for full financial support. We are grateful to Prof. O.J.P. boli
for calling the attention to some points.
References
[1] D.E. Groom et al., Particle Data Group, Eur. Phys. J. C 15 (2000) 1.
[2] For details of grand unified theories see: R.N. Mohapatra, Unification and Supersymmetry, Springer-Verlag,
New York, 1992;
G.G. Ross, Grand Unified Theories, BenjaminCummings, USA, 1985.
[3] P.H. Frampton, S.L. Glashow, Phys. Lett. B 190 (1987) 157;
P.H. Frampton, B.H. Lee, Phys. Rev. Lett. 64 (1990) 619.
[4] P.H. Frampton, T.W. Kephart, Phys. Lett. B 66 (1991) 1666;
P.H. Frampton, D. Ng, Phys. Rev. D 43 (1991) 3034.
[5] J.E. Cieza Montalvo, O.J.P. boli, S.F. Novaes, Phys. Rev. D 46 (1992) 181.
[6] See, e.g., P. Langacker, Phys. Rep. 72 (1981) 185.
[7] S. Dimopoulos, Nucl. Phys. B 168 (1981) 69;
E. Farhi, L. Susskind, Phys. Rev. D 20 (1979) 3404;
J. Ellis et al., Nucl. Phys. B 182 (1981) 529.
[8] J.L. Hewett, T.G. Rizzo, Phys. Rep. 183 (1989) 193.
[9] J. Maalampi, K. Mursula, M. Roos, Nucl. Phys. B 207 (1982) 233.
[10] H.N. Brown et al., Muon (g 2) Collaboration, hep-ex/0102017.
[11] Abdurashitov et al., SAGE Collaboration, Phys. Rev. C 60 (1999) 055801;
Abdurashitov et al., Phys. Rev. Lett. 77 (1996) 4708;
W. Hampel et al., GALLEX Collaboration, Phys. Lett. B 477 (1999) 127;
Y. Fukuda et al., Super-Kamiokande Collaboration, Phys. Rev. Lett. 82 (1999) 2644;
Y. Fukuda et al., Super-Kamiokande Collaboration, Phys. Rev. Lett. 81 (1998) 1562;
B.T. Cleveland et al., Homestake Collaboration, Astrophys. J. 496 (1998) 505;
K.S. Hirata et al., Super-Kamiokande Collaboration, Phys. Rev. Lett. 77 (1996) 1683.
[12] V. Pleitez, M.D. Tonasse, Phys. Rev. D 48 (1993) 2353.
[13] A review on the other types of heavy-leptons is given by: P.H. Frampton, P.Q. Hung, M. Sher, Phys. Rep. 330
(2000) 263.
[14] F. Pisano, V. Pleitez, Phys. Rev. D 46 (1992) 410;
R. Foot, O.F. Hernandez, F. Pisano, V. Pleitez, Phys. Rev. D 47 (1993) 4158.
[15] P.H. Frampton, Phys. Rev. Lett. 69 (1992) 2889.
[16] For recent references on the phenomenology of the 331 model see: Y.A. Coutinho, P.P. Queirz Filho,
J. S Borges, M.D. Tonasse, Three hadronic jet production in the 331 model at CERN LHC energies,
submitted to Phys. Rev. D;
Y.A. Coutinho, P.P. Queirz Filho, M.D. Tonasse, Phys. Rev. D 60 (1999) 115001;
B. Dion, T. Grgoire, D. London, L. Marleau, H. Nadeau, Phys. Rev. D 59 (1999) 075006;
F. Cuypers, S. Davison, Eur. Phys. J. C 2 (1998) 503, and papers cited therein.
[17] M.D. Tonasse, Phys. Lett. B 381 (1996) 191.
[18] N.T. Anh, N.A. Ky, H.N. Long, The Higgs sector in the minimal 331 model with the most general leptonnumber conserving potential, hep-ph/0011201.
[19] P. Jain, S.D. Joglekar, Phys. Lett. B 407 (1997) 151;
D. Ng, Phys. Rev. D 49 (1994) 4805.
341
[20] J.E. Cieza Montalvo, M.D. Tonasse, work in progress.

[21] C.N. Yang, Phys. Rev. 77 (1950) 242.
[22] J.E. Cieza Montalvo, P.P. de Queirz Filho, Production of a single and a pair of heavy exotic lepton through
the pp collisions, preprint DFNAE-IF-UERJ-98-05, hep-ph/9811521;
D.A. Dicus, P. Roy, Phys. Rev. D 44 (1991) 1593;
P.H. Frampton, D. Ng, M. Sher, Y. Yuan, Phys. Rev. D 48 (1993) 3128;
A. Datta, M. Guchait, A. Pilaftsis, Phys. Rev. D 50 (1994) 3195;
M.M. Boyce, M.A. Doncheski, H. Knig, Phys. Rev. D 55 (1997) 68.

AdS/CFT duality and the black hole

information paradox
Oleg Lunin, Samir D. Mathur
Department of Physics, The Ohio State University, Columbus, OH 43210, USA
Abstract
Near-extremal black holes are obtained by exciting the Ramond sector of the D1-D5 CFT, where
the ground state is highly degenerate. We find that the dual geometries for these ground states have
throats that end in a way that is characterized by the CFT state. Below the black hole threshold we
find a detailed agreement between propagation in the throat and excitations of the CFT. We study the
breakdown of the semiclassical approximation and relate the results to the proposal of gr-qc/0007011
for resolving the information paradox: semiclassical evolution breaks down if hypersurfaces stretch
too much during an evolution. We find that a volume V stretches to a maximum throat depth of
V/2G. 2002 Elsevier Science B.V. All rights reserved.
PACS: 04.50.+h; 04.70.Bw
1. Introduction
One of the most puzzling paradoxes in physics is the black hole information paradox.
When black holes form and evaporate, the Hawking radiation appears to be correctly given
by a semiclassical calculation, but the radiation so computed destroys unitarity and thus
violates the principles of quantum mechanics. String theory has made substantial progress
in understanding the quantum physics of black holes, and its results suggest very strongly
that the evaporation process maintains unitarity (see, for example, [1,2] and references
therein). But there is no clear understanding of the mechanism by which the information
manages to leak out of the black hole via the radiation.
The essential strength of the information paradox lies in the fact that Hawkings
semiclassical calculation of the radiation is independent of any details of quantum gravity
E-mail address: lunin@pacific.mps.ohio-state.edu (O. Lunin).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 2 0 - 4
O. Lunin, S.D. Mathur / Nuclear Physics B 623 (2002) 342394
343
at small scales [3]. Thus even though we find in string theory many perturbative and
nonperturbative quantum gravity effects at the Planck scale and string scale, it is unclear
how to use them to effect the information transfer. It is plausible that the effects that correct
the semiclassical computation are nonlocal, and finding them will involve fundamental
changes in our understanding of when semiclassical gravity is valid.
We will work with the D1-D5 system. For this case the microscopic entropy and low
energy radiation rates agree with the Bekenstein entropy and Hawking radiation of the
corresponding black hole, including numerical factors [47]. Such agreements contributed
to Maldacenas remarkable conjecture [8] that the CFT describing the brane system
is dual to the near horizon geometry produced by the branes. Though this AdS/CFT
correspondence has been extensively studied, most of the analysis has pertained to the
NeveuSchwarz (NS) sector of the CFT, while the black holes in asymptotically flat
spacetime arise in the Ramond (R) sector [9,10]. (We will discuss this further below.)
The ground state in the NS sector is unique, but in the R sector the ground state has a
large degeneracy, so that a large number of microstates appear to correspond to the same
geometry. This fact is not just a technical complication but an essential issue, since similar
degeneracies yield the Bekenstein entropy of black holes.
In this paper we will analyze the AdS/CFT correspondence in the R sector. We find
several agreements between microscopic quantities computed in the CFT (which we take to
be a sigma model at the orbifold point) and the corresponding supergravity computations.
The length scales emerging in these relations are quite different from those that are relevant
for the analysis in the NS sector. The results indicate the way in which the semiclassical
approximation may break down in string theory to allow information leakage in Hawking
radiation.
1.1. Review of some basic issues
Consider first the classical geometry of D1- and D5-branes with no momentum charge
and no angular momentum. At r the geometry is flat, while at small r the geometry
is locally AdS3 S 3 M where the 4-manifold M is T 4 or K3. We have a horizon at
r = 0, and the part of AdS3 covered by the geometry is one Poincare patch. The direction
y along the D1-branes is compactified on a circle of radius R when we wish to make a
black hole in 5 dimensions, and this creates a periodic identification of points in the AdS
space.
When we attempt to apply the AdS/CFT correspondence conjecture to this D1-D5
geometry we encounter several important questions.
(a) The Poincare patch of the AdS geometry smoothly extends past the horizon r = 0
to the entire global AdS spacetime. The complete geometry thus obtained has more than
one asymptotically flat region [11]. But it appears strange if new spatial infinities could be
created just by bringing together a (large but finite) number of branes in a normal spacetime
with one asymptotic infinity.
One may argue that the semiclassical Poincare patch geometry stops short of the horizon
r = 0 because the identification y y + 2R relates points that are closer and closer
together as r 0. But this leads to the question: what is the effective value of r where
344
we should end the geometry, and what happens when an incident wave reaches this value
of r? One may naively think that the critical value of r would correspond to the length of
the y circle becoming order string length or Planck length. It was argued however in [12]
that the geometry ends much further down the throat, and that the truncation of the throat
is to be traced to the existence of nonzero angular momentum in the CFT state. We will
discuss this issue in more detail in this paper.
(b)
The D1-D5 bound state has a large degeneracy; the number of ground states is

e2 2 n1 n5 , where n1 and n5 give the number of D1- and D5-branes. The metric of the
Poincare patch appears to be the same for all these states. This is of course a version of
the standard black holes have no hair problem. One could try to argue that the branes
sit at r 0 and thus carry information about the microstate. But this would contradict our
expectation from the AdS/CFT duality: the duality suggests that the entire AdS region is
already a dual description of the state of the branes, and we should not find the branes
to be present at the end of the throat. (We will in fact argue below that no branes need
be included at the end of the throat; rather the way that the throat ends characterizes the
microstate.)
1.1.1. The difference between NS and R sectors.
The initial proposal by Maldacena of the AdS/CFT correspondence [8] was motivated
partly by results on black holes, and pertained to the near horizon geometry produced by
branes in asymptotically flat spacetime. When we wrap D1, D5 branes on a circle y in
spacetime, then the behavior of the fermions in the CFT is induced from the behavior
of fermionic fields in the bulk. Since the supergravity in the bulk has fermions periodic
around y, the D1-D5 CFT also has fermions that are periodic around the y circle, so the
CFT is in the R sector. (If we put the fermions in the bulk supergravity to be antiperiodic
around y, then the vacuum energy is nonzero, and flat spacetime ceases to be a solution.
Since we want asymptotically flat spacetime, we will not consider this situation.)
The proposal in [8] did not, however, give an explicit way to relate correlators via the
duality. The map of chiral operators and correlators was carried out by Gubser, Klebanov,
Polyakov [13] and by Witten [14]. Let us analyze these and other approaches to AdS
correlators, keeping in mind that we are interested in the following essential physics. A
wave traveling towards r = 0 becomes one of shorter and shorter wavelength, due to the
redshift between infinity and the small r region. The wave does not reflect back to larger
r unless we have some explicit modification to the physics of the throat. This monotonic
infall of an incident quantum to the horizon is an essential aspect of the black hole problem;
the particle itself does not appear to return but the deformation it creates in the geometry
causes Hawking radiation to be emitted, and this radiation appears to carry no information
about the infalling quantum.
In the analysis of Witten [14] the AdS space was rotated to Euclidean signature. This
causes the Poincare patch to become a smooth space with no singularity at r = 0. However
at the same time we lose the physics of the horizon; we cannot have traveling waves
in Euclidean signature, nor do we have the accumulation of wavefronts near r = 0 that
signals the infall of the particle to the horizon. The situation is similar to rotating a black
hole geometry to Euclidean signature: we get only the space outside the horizon, and the
345
geometry is smooth at the location r = rhorizon . But with such a rotated geometry we cannot
address the question of what happens to quanta that fall into the hole.
In [13] correlation functions were computed in the Poincare patch arising from the
near horizon geometry of branes. The correlators were however computed for spacelike
momenta p2 > 0. The wavefunction in this case has a growing part and a decaying part
at r = 0, and the growing part was set to zero to solve for the Greens function. But real
infalling particles have timelike or null momenta. The waveform is then oscillatory near
r = 0, and we have no natural way to relate the ingoing and outgoing parts of the wave. (If
we just drop the outgoing part, then we have particles being swallowed into the horizon,
and we cannot make a Greens function at the AdS boundary by using the propagator.)
Correlation functions have also been computed in the AdS geometry with Lorentzian
signature and with arbitrary momenta, but in these calculations the spacetime was extended
to the global AdS space past r = 0 [15]. To relate computations with this global AdS to
the black hole question that we started with, we would have to connect the global AdS to
asymptotically flat spacetime. It is unclear how this is to be done, especially if we wish
to avoid the appearance of new asymptotically flat regions. As was mentioned above, it
would be strange if matter falling onto the branes were to move smoothly through r = 0
and emerge in a new asymptotically flat region, since we could make the brane geometry by
gathering together branes in a spacetime which started with only one asymptotic infinity.
Thus we see that we cannot directly use any of the usual ways of computing correlators
in the AdS/CFT correspondence to understand how information may return (as Hawking
radiation) after we throw some energy into the throat of the D1-D5 brane geometry. As
mentioned before the difficulty is not just a technical issue, but rather the fact that if we
could set up AdS correlators in a supergravity geometry that pertains to a black hole then
we would directly observe information emerging in Hawking radiation. But extensive work
with supergravity solutions have shown that we cannot get such a simple resolution of the
black hole information problem; further the emerging Hawking radiation is a complicated
multi-particle state even for a single high energy incident particle, so it makes sense that
we are unable to compute simple 2-point functions in AdS that correspond to returning a
particle from r = 0.
To address the black hole information problem we need to work in the R sector of
the CFT, with Lorentzian signature, an asymptotically flat infinity, and with quanta that
have timelike or null momenta. The global AdS geometry corresponds to the NS sector of
the CFT; we must instead use the Poincare patch and try to resolve the physics at r = 0.
Changing any of these conditions bypasses the information problem, even though we may
get interesting AdS/CFT correspondence theorems.
1.2. Results
(A) We start by listing the R ground states of the D1-D5 CFT. These states may be
represented pictorially in terms of the effective string, which has total winding number
n1 n5 N around the y circle. In a generic state the effective string is broken up into a
number m of component strings;
each component string is wrapped ni times around y

before closing on itself, and i ni = N .
346
Naively, the geometry of the D1-D5 supergravity solution appears to have a uniform
throat (in the sense that that angular S 3 asymptotes to a constant radius), all the way down
to a horizon at r = 0. We find however that the throat ends at some point before r = 0.
There is a singularity at the end of the throat characterized by a curve in 4-d space. The
location and shape of this curve mirrors the CFT microstate. In the special case that the
curve is a circle we recover the metrics of [16].
(B) If the throat extended all the way down to r = 0 then throwing even an infinitesimal
energy into the throat would lead to horizon formation at some value of r. Since the throat
is actually finite, there is an energy scale Ethreshold below which we can study matter
quanta moving in the throat without formation of a horizon (this is the hot tube studied
in [12]).
Below the threshold of black hole formation Ethreshold we construct a detailed map
between the throat geometry and the dual CFT effective string. A quantum placed in the
throat bounces several times up and down the throat before escaping to infinity; let the
time for each bounce be tSUGRA . In the dual CFT this quantum is represented by a set of
left and a set of right movers on the effective string; these vibration modes travel around
the string and meet each other at intervals tCFT . For the special metrics of [16] we can
separate the wave equation and obtain tSUGRA precisely [12]. We find
(i) tCFT = tSUGRA .
(ii) The probability per unit time for the quantum to escape from the supergravity throat
exactly equals the probability per unit time for the vibration modes on the effective string
to collide and emerge as a graviton.
We then proceed to argue that even though we cannot solve the wave equation for a
general microstate, we still get tCFT tSUGRA for all microstates.
Note that the scale tSUGRA is much larger than the radius of the AdS3 which appears
in AdS/CFT computations of correlation functions performed in the NS sector [13,14,17];
thus we are probing a different set of quantities than are usually studied with the duality.
Also note that in comparing emission rates we are using explicitly the part of spacetime
that joins the near horizon geometry to flat space.
(C) In setting up the map between quanta in the supergravity throat and vibration modes
of the CFT component strings we find the following: the naive map breaks down if we are
forced to put more than one set of vibration modes on the same component string. In
general the CFT state has several component strings, and we can thus describe several
quanta in the supergravity throat. Interestingly, when we put enough quanta that we would
run out of component strings in the CFT, we find on the supergravity side that we reach the
threshold Ethreshold for black hole formation.
At this point we note a close similarity of these results with a conjecture made in [18].
In [18] it was argued that to resolve the black hole information paradox we need spacetime
to have the following property: semiclassical propagation on the spacetime breaks down
if during the course of evolution an initial slice in a foliation is stretched beyond a certain
length. Thus spacelike slices need to be endowed not only with their intrinsic geometry but
347
also with a density of degrees of freedom; if the stretching dilutes these degrees too much
then nonlocal effects spoil the usual evolution equation.
We find that the number of component strings in the CFT state act like the density of
degrees of freedom for the dual throat geometrywhen space is stretched to give longer
throats we also have fewer component strings. We characterize the stretch of a volume V
by the time t to travel down and back up the throat. We then find the relation
tmax =
V
2G(5)
N
(1.1)
2. R ground states in the CFT and dual metrics for a special family
2.1. The states in the CFT
We consider the bound states of n1 D1 branes and n5 D5 branes in IIB string theory. We
set
N = n1 n5 .
(2.1)
The D5 branes are wrapped on a 4-manifold M, and thus appear as effective strings in the
remaining 6 spacetime dimensions. M can be T 4 or K3. The D1 branes and the effective
strings from the D5 branes extend along a common spatial direction x5 y, and y is
compactified on a circle of length 2R. The low energy dynamics of this system is a
N = (4, 4) supersymmetric (1 + 1)-dimensional conformal field theory (CFT). The CFT
has an internal R-symmetry SU(2)L SU(2)R SO(4). This symmetry arises from the
rotational symmetry of the brane configuration in the noncompact spatial directions x1 , x2 ,
x3 , x4 . The group SU(2)L is carried by the left movers in the CFT and the group SU(2)R
is carried by the right movers.
Consider this CFT at the orbifold point [1923]. Then the CFT is a (1 + 1)dimensional sigma model where the target space is the orbifold M N /SN , the symmetric
product of N copies of the 4-manifold M. We are interested in the R sector ground states
of this system (these will be states with no left- or right-moving excitations).
These R ground states can be obtained by first finding all the chiral primary fields in
the NS sector, which are states with h = j3 , h = j3 . Spectral flow then maps these chiral
primaries to ground states of the R sector. Spectral flow acts independently on the left and
right movers, in the following way
hR = hNS j3NS +
c
,
24
j3R = j3NS
c
.
12
(2.2)
The R ground states all have h = h = c/24.

2.1.1. Chiral primaries in the NS sector
The M N /S N orbifold CFT and its states can be understood in the following way. We
take N copies of the supersymmetric c = 6 CFT which arises from the sigma model with
target space M. The vacuum of the theory is just the product of the vacuum in each copy of
348
the CFT. In the orbifold theory we find twist operators n [24,25]. The copies 1, 2, , n
of the CFT permute cyclically into each other 1 2 n 1 as we circle the
point of insertion of n . (The other copies are not touched, and we ignore them for the
moment.) In this given twist sector there are operators with various values of j3 , but we
are interested in those that are chiral operators. The chiral operator in this twist sector with
lowest dimension and charge is termed n [26] and has
h = j3 =
n1
,
2
n1
h = j3 =
.
2
(2.3)
Each copy of the CFT has the SU(2) currents J (i)a , J(i)a , where the index i labels the
copies. Define
J =
a
n
(i)a
Ja =
i=1
n
J(i)a .
(2.4)
i=1
Then we can make three additional chiral primaries from :

+
n ,
n+ = J1
n+1
,
2
n1
,
h = j3 =
2
n+1
,
h = j3 =
2
h = j3 =
+
n+ = J1
n ,
+ +
n++ = J1
J1 n ,
n1
h = j3 =
,
2
n+1
h = j3 =
,
2
n+1
h = j3 =
.
2
(2.5)
The chiral primaries n , n+ , n+ , n++ correspond respectively to the (0, 0), (2, 0),
(0, 2), (2, 2) forms from the cohomology of M. Both T 4 and K3 have one form of each
of these degrees. We will not consider the chiral primaries arising from the other forms on
M; this will not affect the nature of the arguments that we wish to present.
The operator 1 is just the identity operator in one copy of the c = 6 CFT. Thus for
the complete CFT made from N copies we can write the above chiral operators as
n [1 ]Nn .
(2.6)
It is understood here that we must symmetrize the above expression among all permutations
of the N copies of the CFT; we will not explicitly mention this symmetrization in what
follows.
More generally, we can make the chiral operators
k

s ,s mi
nii i ,
k
i=1
i=1
ni mi = N,
(2.7)
where si , si can be +, . This gives the complete set of chiral primaries that result if we
restrict ourselves to the above mentioned cohomology of M.
2.1.2. Ground states in the R sector
In the state (2.7) the N copies of the c = 6 CFT are naturally grouped into subsets of
size ni . The operation of spectral flow proceeds independently in each such subset. Thus
349
consider a subset corresponding to the n copies linked by a twist ns,s . For this subset we
have c = 6n in the spectral flow relations (2.2). After the flow we get the charges
1
1
n j3R = , j3R = ,
2
2
1
1
n+ j3R = ,
j3R = ,
2
2
1
1
n+ j3R = , j3R = ,
2
2
1
1
++
R
R
j3 = .
n j3 = ,
(2.8)
2
2
In fact, these 4 operators, after flowing to the R sector, join up to form a representation
+
of SU(2)L SU(2)R with (j, j) = ( 12 , 12 ). This can be seen by noting that J1
in the NS
+
sector flows to J0 in the R sector.
This treatment of each separate ns,s would complete our treatment of the R ground state
except for the fact that when we have more than one copy of the same operator then we
must take only the symmetric combination of the different copies. Thus consider the set of
chiral primaries (2.7) where the twist sector n occurs a total of m times, and concentrate
attention on the nm copies of the c = 6 CFT that are involved in these operators.
In the absence of symmetrization we would have, after flowing to the R sector, spin
values (j, j) = ( 12 , 12 ) from each set (2.8), and we would add the spins according to the
rules for SU(2) independently in the left and right SU(2)s. The symmetrization has the
effect of giving, in the R sector, only the states with equal values of jL and jR :
m m
(j, j 3 ; j, j3), j = , 1, . . . , 0; m even,
2 2
1
m m
j = , 1, . . . , ; m odd.
(2.9)
2 2
2
(The values of j 3 , j3 range independently from j to j .)
2.1.3. Pictorial description of R ground states
The twist operator Ns,s joins all N copies of the c = 6 CFT into one copy of the c = 6
CFT (on an N times longer spatial circle). We depict the corresponding R ground state
pictorially in Fig. 1(a); we have a multiwound string wrapped N = n1 n5 times around the
y circle. A generic state of the form (2.7) is pictured in Fig. 1(b); there are k component
strings, with the ith string wrapped ni times around the y circle.
Each component string carries spin (jL , jR ) = ( 12 , 12 ), which we have depicted by the
arrows on the component strings. We get a particularly simple set of geometries if all the
spins are aligned (these geometries will be reviewed in the subsection below), but we will
need to consider general spin orientations to address the physics of the generic microstate.
2.2. A special family of metrics for the D1-D5 system
Let us recall the physics of a special family of metrics for the D1-D5 system; these
metrics were studied in [12,16]. We will see in the next section that these special metrics
350
Fig. 1. (a) The state with one component string wrapped n1 n5 times around the direction y; (b) A generic state
with several component strings. The arrows indicate the spins of the component strings.
m
correspond to CFT states arising from chiral primaries [N/m
] ; i.e., we have many
component strings of equal length, and their spins are all aligned. The D1 and D5 charges
are Q1 , Q5 respectively, and the angular momentum is specified by a parameter a. The
10-D metric and other supergravity fields are given in Appendix B, but for the most part
we will need only the 6-D Einstein metric obtained by dimensional reduction on the 4manifold M:

1 2
dr 2
2
2
2
dsE = dt dy + hf d + 2
h
r + a2

2a Q1 Q5 2
cos dy d + sin2 dt d
hf

a 2 Q1 Q5 cos2
cos2 d 2
+ h r2 +
h2 f 2

a 2 Q1 Q5 sin2
2
2
+ r 2 + a2
d
(2.10)
sin
,
h2 f 2
where
f = r + a cos ,
2

Q1
Q5 1/2
h= 1+
.
1+
f
f
(2.11)
This geometry is flat at spatial infinity. The direction y is assumed to be compactified on

a circle of radius R. The size of the S 3 in the variables , , settles down to a constant for
a r (Q1 Q5 )1/4 , so we term the region r < (Q1 Q5 )1/4 as a throat. Note, however,
that since the y circle shrinks as r decreases, we will not in fact get a uniform throat if we
dimensionally reduce along y and look at the 5-D Einstein metric. Thus we do not have
a throat in the same sense as we get for the D1-D5-momentum system, but we continue
to use the term throat since it conveys the constancy of the S 3 radius. The start of the
throat is at r (Q1 Q5 )1/4 , but it is important that the throat is capped off at r a: the
geometry ends smoothly except for a singularity on the curve r = 0, = /2.
351
Consider the massless scalar wave equation in the above metric. We can write the anzatz

).
(t, r, , , , y) = exp it + im + in + iy (r,
(2.12)
It turns out however that there is an additional hidden symmetry in the metric, and there is
a further separation between r and [27].
In [12] the following process was studied. We start with a quantum of this scalar field
at spatial infinity, sent in towards r = 0. Let the wavelength 1 be much larger than the
length scale (Q1 Q5 )1/4 . Then most of the waveform reflects back to infinity from the start
of the throat, but there is a small probability P for the quantum to enter the throat [12,28]

2
Q1 Q5 4 l+1
1
,
P = 4 2
(2.13)
16
(l + 1)!l!
where l specifies the spherical harmonic of the wave.
The part of the wave which enters the throat travels to r = 0, where it reflects back;
we find naturally reflecting boundary conditions at the singularity. When the wave reaches
back to the start of the throat we again have the same probability P that the quantum will
escape to infinity, while the probability is 1 P 1 that it will reflect back into the throat.
Thus the quantum travels several times in the throat before escaping. We can find the time
for traveling once up and down the throat by looking at the phase shift between successive
wavepackets emerging from the throat. This time is
Q1 Q5
.
t =
(2.14)
a
The parameter a is related to the angular momentum of the geometry as follows. The
rotation group on the noncompact spatial directions is SO(4) SU(2) SU(2). The value
of j in each SU(2) factor is the same, and it can be an integer or half integer. We have
2j
Q1 Q5
.
a=
(2.15)
n1 n5
R
Following [16] we set
2j
.
n1 n5
The maximum value of j is n1 n5 /2, so the maximum value of is unity.
Substituting (2.15) in (2.14) we get
t =
R
.
(2.16)
(2.17)
3. CFT states and their dual supergravity throats

3.1. Microscopic model used for black hole absorption computations
We have seen in the past that absorption cross sections for D1-D5 black holes
can be reproduced, including numerical factors, by the computation of absorption into
352
the corresponding microstate. Let us briefly recall the nature of these microscopic
computations.
The microstate was described by an effective string wound N = n1 n5 times around
a circle; this circle is in the direction x5 y and has length 2R. Among the various
supergravity quanta that can be considered, the simplest are the s-wave (l = 0) modes
of minimally coupled scalars. Let the compact 4-manifold M be T 4 , spanned by the
coordinates x6 , x7 , x8 , x9 . Then the fluctuation h67 is an example of such a minimally
coupled scalar, and we will use it for illustration in what follows.
When an incident quantum h67 encounters the effective string, its energy can get
converted to that of vibrations of the string [5,6,40]. At low energies, the dominant process
is the absorption of the s-wave, and this creates one left-moving vibration quantum and one
right-moving vibration quantum on the string. The incident quantum has an energy , and
its momentum lies only in the noncompact directions x1 , x2 , x3 , x4 . The vibrations of the
string have momenta along x5 , and thus the energymomentum vectors are, respectively,
for the left-moving quantum and the right-moving quantum
(p0 = /2, p5 = /2)
and (p0 = /2, p5 = /2).
(3.1)
Further, one of these vibrations is polarized in the direction x6 and the other is polarized in
the direction x7 . The graviton in fact corresponds to the symmetric combination

1
h67 |x6 L |x7 R + |x7 L |x6R .
(3.2)
2
(The antisymmetric combination corresponds to absorption of the RamondRamond 2RR .)
form B67
The amplitude for this absorption was computed in [6,7] from the action for the effective
string coupled to gravity. It was found that such a calculation with the effective string gave
the same absorption cross section as that for absorption of low energy s-wave minimally
coupled scalars into the D1-D5-momentum black hole. In [28] it was noted that this
agreement persisted even if we do not have a large momentum charge, and in fact was
also true to higher orders in the energy of the scalar as long as the D1 and D5 charges were
kept large. This indicates that the simplest system that captures the effective physics would
be the D1-D5 system with no momentum excitations; any momentum charge would be
considered one of the possible excitations of our starting configuration. We will thus work
with states of the D1-D5 system with no net momentum charge in this paper.
3.2. Time delay and its microscopic interpretation
Let us first consider the state in the R sector that results from spectral flow of the chiral
primaries
Ns,s .
(3.3)
In this case the twist operator joins together all N = n1 n5 strands of the effective string
into one single multiply wound strand, which was drawn schematically in Fig. 1(a).
This state corresponds to the string configuration used in [6,7,28,29] in the study of
black hole entropy and absorption. In our present considerations we will be also interested
353
Fig. 2. (a) A graviton is incident on the component string; (b) The graviton is converted to a pair of vibration
modes; (c) The vibration modes meet again at B.
in the spin of this effective string. As discussed in the previous section, the variables s, s
take the values 12 , and in the R sector the effective string forms a multiplet with
1
jL = jR = .
(3.4)
2
To study the picture of absorption by this effective string we have redrawn in Fig. 2 the
string opened up into a large circle; the length of this circle is 2Rn1 n5 . The incident
graviton h67 breaks up into a pair of vibration modes at the point marked A. Since the
graviton had no momentum along the direction x5 y, the amplitude for this process is
the same for any position of the point A along the string. The wavefunction of the two
vibration quanta is thus of the form
(y1 , y2 ) = (y1 y2 ).
(3.5)
Here y1 , y2 are the coordinates of the left-moving and right-moving vibration modes on
the effective string; thus 0 yi < 2Rn1 n5 in the present example. (We can think of these
modes as massless open string states traveling along the effective string, and then y1 , y2
are the locations of the these open strings.) The modes travel in opposite directions at the
speed of light v = c = 1, and encounter each other again at the point B which is halfway
along the effective string. The time interval between the start at A and the meeting at B is
2Rn1 n5
= Rn1 n5 .
(3.6)
2
As we will see below, there is only a small chance that the modes interact and re-emerge
as a graviton at the point B. If they do not interact to leave the effective string, then they
travel around again and re-encounter each other at the point A, after a further time (3.6).
As mentioned already there is nothing special about the points A, B on the string, since
the center of mass of the two vibration modes is uniformly smeared over the string. The
only physical quantity therefore is the time between successive encounters of the modes
(not the place where they interact), and this time is given by (3.6).
Now let us look at the supergravity background corresponding to this R sector state. As
summarized in the last section, the D1-D5 geometry with angular momentum jL = jR = j
has a throat terminating after a certain distance. If a quantum of a minimally coupled scalar
is incident from infinity, then there is a small probability that it enters the throat. If it does
enter the throat, then it travels to the end where it reflects back and travels again to the start
of the throat. Since the probability to enter the throat was small, the probability to leave
tCFT =
354
the throat is also small, since these two probabilities are equal. Thus the quantum travels
several times up and down the throat before exiting the throat back to spatial infinity. The
time for traveling once down the throat and back was computed in [12] and is given by
(Eq. (2.17))
tSUGRA =
R Rn1 n5
=
= n1 n5 R,
2j
(3.7)
where we have used the fact that we have j = 12 .

This time tSUGRA exactly equals the time tCFT found in the microscopic computation
in (3.6) above.
The essential idea of the AdS/CFT correspondence then yields the following picture:
(i) The effective string is dual to the throat region of the supergravity solution. It is
important to note here that the throat is not infinitely long; otherwise we could not have
found the above relation between tSUGRA and tCFT .
(ii) A graviton outside the throat in the supergravity solution is described by the
graviton being present outside the effective string in the CFT picture. Note that we are
working with quanta of wavelength much larger than the scale (Q1 Q5 )1/4 , so this
quantum outside the throat effectively travels in a flat metric in the supergravity solution.
(iii) The process where the quantum enters the throat in the supergravity solution maps
to the process in the CFT where the incoming graviton converts its energy to that of the two
vibration modes of the effective string. Similarly, the process where the graviton manages
to leave the throat and escape to infinity maps to the process in the CFT where the vibration
modes collide and leave the effective string as a single graviton.
(iv) From (iii) above it is logical to identify the supergravity state where the quantum
is near the start of the throat with the CFT configuration where the two vibration modes
are close to each other, as at points A, B in Fig. 2. The quantum at the end of the throat
(near r = 0) maps to the CFT configuration where the vibration modes are separated by
the maximal possible distance along the effective string. This is of course just a version of
the UV/IR correspondence [30].
(v) The supergravity quantum travels several times up and down the throat, with a
small probability to escape each time it reaches the start of the throat. Correspondingly,
the vibration modes travel around the string, with a small probability to collide and leave
the string as a graviton each time they meet.
In the following we will make this correspondence more precise, and provide additional
evidence for the proposed relation between CFT excitations and their description in the
supergravity dual.
3.3. General values of j
As we will see later in the section on black hole formation, for the case jL = jR = 12
considered above the supergravity solution is not trustworthy all the way to the end of the
throat. More precisely, the backreaction of the quantum placed in the throat deforms the
throat in a significant way, at least near the end of the throat. To avoid this we now look
355
m
Fig. 3. Component strings for the state [N/m
] . The spins are all aligned to give j = m/2.
at larger values of angular momentum, where it will turn out that the backreaction of the
quantum can indeed be ignored.
Let us look at a state of the CFT with
jL = jR = j = m/2.
(3.8)
The simplest way to make such a state is to take the chiral primary in the NS sector
m
N/m .
(3.9)
Upon spectral flow to the R sector, each component N/m

gives a multiwound string that
is wound
n1 n5
N
=
,
n=
(3.10)
m
m
times around the circle x5 y. There are m such component strings. We sketch this
configuration in Fig. 3. The spin of each multiwound component is jL = jR = 12 , and
in the state that we have taken all these spins are aligned to produce the spin values (3.8).
An incident graviton can be absorbed into any of these m component strings. But now
the travel time for the vibration modes around the string, before they meet each other again,
is
n1 n5 R
.
tCFT =
(3.11)
m
Now we look at the corresponding supergravity solution. When the angular momenta
are given by (3.8) then the time for a quantum to travel once down and back up the throat
is (Eq. (2.17))
R Rn1 n5 n1 n5 R
=
=
.
2j
m
Thus again we find
tSUGRA =
(3.12)
tCFT = tSUGRA .
(3.13)
We have used the s-wave quanta as an illustration but we see immediately that we get
the relation (3.13) for higher partial waves as well. The supergravity travel time tSUGRA
356
was found in [12] to be independent of the harmonic order l. In the CFT the absorption of
a higher partial wave creates set of massless left movers X . . . and a set of massless
. . . [3133]. Each set of excitations travels at the speed of light around
right movers X
the component string, so that tCFT is independent of l as well.
3.4. The rate of radiation
Suppose that we place an s-wave scalar quantum in the supergravity throat; the dual
CFT state has a pair of vibration modes on the effective string. We wish to compute the
probability per unit time for the scalar quantum to escape from the throat, and compare this
to the probability per unit time for the vibration modes on the effective string to collide and
leave the string.
First we look at the CFT computation. Let there be one left-moving vibration and
one right-moving vibration on the string, with the momentum vectors (3.1) and in the
wavefunction corresponding to a graviton (3.2). The interaction that leads to the emission
of the graviton is given by using the DBI action for the effective string. The computation is
essentially the same as that done for absorption into the black hole microstate, but since we
are here looking at the emission rate we reproduce the relevant calculation in Appendix A.
m
Consider the R ground state arising from the chiral primary [N/m
] . This state has m
component strings, all in the same state, and the spins are aligned to give jl = jR = m2 .
Then the probability per unit time for the vibrations to collide and emit a graviton is found
to be
RCFT = m
2 4 g 2
.
2V (2)R
(3.14)
Here g is the string coupling, (2)4 V is the volume of the T 4 and we have set = 1.
Now we look at the supergravity solution. The throat is assumed to be long compared
to the scale (Q1 Q5 )1/4 of the geometry near the start of the throat. A quantum incident
from infinity has a probability P to enter the throat, after which it propagates down the
throat till it reaches the end where it reflects. The long length of the throat implies that the
probability P depends only on the geometry near the start of the throat, which is the same
as the geometry in the nonrotating case j = 0. Thus we can read this probability off from
[12,28], and we find (using Eqs. (2.13), (B.4))
P=
2
2 4 n1 n5 g 2
Q1 Q5 4 =
.
4
4
V
(3.15)
The number of times the quantum tries to exit the throat per unit time is
1
tSUGRA
(3.16)
Thus the probability of emission per unit time is (using (3.12), (3.15))
RSUGRA =
P
tSUGRA
=m
2 4 g2 1
.
2
V 2R
(3.17)
357
We see that
RCFT = RSUGRA
(3.18)
so that the rates of emission agree exactly between the supergravity and CFT pictures.
Let us compare the above calculation with the computations [6,7,28] that show
agreement between microscopic radiation rates and Hawking radiation rates from the
corresponding black holes. The CFT computation is the same in each case. But on the
supergravity side in the present case we do not have a horizon, and the graviton does not
represent Hawking radiation. We are in fact below the black hole formation threshold, and
so instead of finding a horizon somewhere down the throat we find a point from which we
can reflect back. The travel time tSUGRA has no analogue if we have a horizon. Thus the
agreement (3.18), while similar to that in the black hole case, has a somewhat different
interpretation.
4. Generic microstates and the dual throat geometries

So far we have used specific examples of D1-D5 microstates to argue that tCFT =
tSUGRA . For this to be a general principle, however, we must examine all possible R
ground states of the D1-D5 system.
4.1. Unbound states: a potential difficulty and its resolution
A simple way to violate tCFT = tSUGRA would appear to be the following. Take two
D1-D5 bound states, with angular momenta

jL = jL3 = m/2; jR = jR3 = m/2 and jL = jL3 = m/2; jR = jR3 = m/2 .
(4.1)
Since such D1-D5 states are mutually BPS, we can construct a supergravity solution with
arbitrary locations for the centers of the two parts. Let the two centers be coincident. The
angular momentum of the two bound states cancel each other, so that we get jL = jR = 0.
Naively, looking at Eq. (2.14), (2.15) one may think that this geometry will have an infinite
throat and infinite tSUGRA . But looking at the lengths of the component strings in the
CFT we still find tCFT = n1mn5 R , so that we have a potential contradiction.
But a closer look at the supergravity solution reveals the following. The rotation
parameter a in the solution (2.10) multiplies dt d, but it also appears in the function
f = r 2 + a 2 cos2 . Suppose we take a solution (2.10) with rotation parameter a and
superpose another solution with the same charges and rotation parameter a = a, and
the same center. Then the coefficient of dt d disappears, but f remains unchanged since
it does not depend on the sign of a. We have developed in Appendix C an extension of
the chiral null models to yield directly general solutions for the D1-D5 system. The 6-D
Einstein metric for this superposition, worked out in Appendix D, is

f0 2
dr 2
2
dt dy 2 + f1 f5 2
+
d
dsE2 =
r + a2
f1 f5
358

f1 f5 2
r cos2 d 2 + r 2 + a 2 sin2 d 2 ,
f0
(4.2)
where
f0 = r 2 + a 2 cos2 ,
f1 = f0 + Q1 ,
f5 = f0 + Q5 .
(4.3)
We cannot solve the wave equation in this metric since the variables do not separate. But
we can estimate the time delay by looking at geodesics in the throat. A generic geodesic
goes down the throat only down to r a; the travel time is then1
tSUGRA tCFT .
(4.4)
We can superpose several D1-D5 bound states, each having rotation but such that the
combined angular momentum is zero. It is easy to generalize the above construction to see
that in each case we still get tSUGRA tCFT .
4.2. Arbitrary j configurations of the bound state
Let us now return to our principal considerationthe analysis of a single D1-D5 bound
state. Here again we face the following potential problem. Consider the R ground state
s,s
. If the
that results from a chiral primary made of m components, each of the form N/m
m
spins s, s are all aligned then we get jL = jR = 2 ; this is the maximum value of j in (2.9)
and gives the configuration we studied in the Section 3. But from (2.9) we see that we
can also combine the spins s, s to obtain states with lower j , all the way down to j 0.
Does the effective length of the throat depend on the value of m or on the value of j ? If
it depends on j then we have a problem; by choosing m 1 we make the component
strings of the CFT small (length 2RN/m each) but the supergravity throat could be
made long by letting j 0. We will see however that the supergravity throat terminates at
a distance that depends on m rather than j . This fact will be crucial to our conjecture that
tCFT = tSUGRA .
4.2.1. The duality map from D1-D5 to FP
To see the nature of supergravity solutions corresponding to general values of j in (2.9)
we will map the D1-D5 system by a chain of dualities to the FP system: F represents
fundamental string winding (along x5 y) and P represents momentum (also along y).
The reason for starting the analysis with the FP system is the following. The M N /S N
orbifold CFT describes a single D1-D5 bound state. To study the supergravity dual we
need the metric for a single bound state system, and not for a superposition of many such
bound states. In the FP language a single bound state is easy to characterize: it arises from
a single multiwound string carrying left moving vibrations. Duality then maps such FP
solutions to metrics of single D1-D5 bound states.
1 There exist an exceptional set of geodesics that head directly into the singularity at r = 0, = /2,
which do not return back in a finite time t. As we will note below (and this is explained in more detail in
Appendix I) this situation is not generic; in a generic solution the geodesics (and waves) reflect back in a finite
time from the singularity. We also argue in Appendix I that quantum fluctuations of the metric will result in
texceptional tCFT log j even for such nongeneric geometries.
359
The sequence of dualities needed to map the D1-D5 system to the FP system is easy to
write, and can be found for example in [34]. The following is the relation of quantities and
states in the two systems:
(i) The number of D5-branes in the D1-D5 system becomes the winding number nw
of the fundamental string (i.e., nw = n5 ). The number of D1-branes becomes the number
of units of momentum (np = n1 ). The bound state of the D1-D5 system maps to a bound
state of the winding and momentum modes of the FP system. Let the radius of the y circle
after dualities be R . The string of the FP system then has a total length LT = 2R nw .
Momentum modes on this string come in fractional units
2n
n
(4.5)
=
LT
nw R
with the constraint that the total momentum has the form
np
nw np
PT = =
(4.6)
R
nw R
with np an integer [35]. Thus on the fundamental string of total length LT the total level
number of excitations is
p=
N = np nw = n1 n5 .
(4.7)
(ii) Let the fundamental string carry a quantum of vibration in a transverse direction
i, i = 1, 2, 3, 4 with wavelength = LT /n. This excitation is generated by the oscillator
i . If the string has n = N/n units of momentum then the total state of the string is of
n
p
w
the form
i1 m1 ik mk
n1
(4.8)
nk
|0,
with
mj nj = N.
(4.9)
The state (4.8) corresponds in the D1-D5 system to the chiral primary
s ,s m1 s ,s mk
n11 1
nkk k
.
(4.10)
Thus in the R ground state of the D1-D5 system if we have a component string that is
wound n times around y then this component string maps to a momentum mode n , which
has wavelength LT /n around the fundamental string in the FP system. The component
string in the D1-D5 system has four polarization states (s, s ) = (1/2, 1/2) which gives
the spins under SU(2)L SU(2)R = SO(4), and these map to the four polarizations i of
the momentum mode.2
2 The M N /S N orbifold does not include the center of mass U (1) factor of the D1-D5 system, so we
see only one vacuum |0 in (4.8) rather than all the ground states of a single fundamental string without
momentum. Further we are concentrating on only those chiral primaries in the D1-D5 system that arise from the
(0, 0), (0, 2), (2, 0), (2, 2) forms on the 4-manifold M, so we write only the polarizations i for the momentum
modes rather than all possible polarizations.
360
4.2.2. Mapping D1-D5 states to FP configurations

Let us study the FP representation of some selected R ground states of the D1-D5
system.
(i) Take the chiral primary
[n ]N/n
(4.11)
of the NS sector of the D1-D5 system and consider its corresponding R ground state. There
are N/n component strings of winding number n each. Fig. 4(a) depicts this state for n = 2.
Fig. 4. (a) Component strings for the D1-D5 microstate [2 ]N/2 . (b) The fundamental string in the dual FP
system, oscillating in the harmonic n = 2, shown opened up to its full length 2 nw R . (c) The string of (b) as it
actually appears due to the identification y y + 2 R . (d) The cross section of the singularity created by the
strands in (c) in the classical limit.
361
In the FP system we have N/n quanta of the nth harmonic around the fundamental string.
Further, since the spins of the factors in (4.11) are all aligned, the angular momentum is
N
jL = jR = 2n
. The fundamental string rotates in a circle; the metric of this state and the
duality maps were constructed in [34]. Let the rotation generators of the two SU(2) factors
and the SO(4) be related such that jL3 = 12 (M12 + M34 ), jR3 = 12 (M12 M34 ); then the
choice (jL3 , jR3 ) = ( 12 , 12 ) for each n in (4.11) implies that the rotation of the string is
in the x1 -x2 plane.
This F string is pictured in Fig. 4(b), opened up to its total length 2R nw = 2R n5 . Due
to the identification y y + 2R the string actually covers a cylindrical surface in a
helical fashion; this is pictured in Fig. 4(c). We can characterize this cylinder by its cross
section, which is a circle, depicted in Fig. 4(d).
Mapping by dualities back to the D1-D5 system we get the metrics (2.10) with jL = jR =
N
2n . If we take n = 1 then we get the maximal possible angular momentum (jL , jR ) =
( N2 , N2 ); it was shown in [16] that the corresponding D1-D5 supergravity solution was a
smooth geometry.
(ii) The states described in (i) correspond, in the list (2.9), to j taking its maximal
N
possible value 2n
. If all the Nn spins are not aligned, then from the same set of component
strings we can get smaller values of j . In the dual FP system these latter states are just
obtained by taking the same harmonic on the fundamental string, but making the string
swing around an ellipse rather than a circle:

n
n

x1 = A cos
(4.12)
y +t ,
y +t .
x2 = B sin
R nw
R nw
The string still forms a helix but covers a surface with elliptical cross section as depicted
in Fig. 5. In particular letting the string vibrate only in one direction (e.g. B = 0 in (4.12))
would give jL = jR = 0. The supergravity solution for this case is written down explicitly
in Appendix I.
(iii) Consider a chiral primary that has twist operators of two different orders, for
example,
m2
] ,
[n ]m1 [2n
nm1 + 2nm2 = N.
(4.13)
In the FP dual we have harmonics of order n and 2n on the string, so the waveform looks
like

n
n

+
A
,
x1 = A1 cos
+
t
cos
2
+
t
y
y
2
R nw
R nw

n
n

x2 = B1 sin
(4.14)
y
y
+
t
sin
2
+
t
+
B
.
2
R nw
R nw
The surface covered by the string now has a more complicated cross section, depicted in
Fig. 6. For a range of parameters Ai , Bi the singularity curve exhibits a self intersection.
Such self-intersections are not generically present however, since the singular curve lies in
a 4-dimensional space x .
(iv) More generally, the singularity can have the shape of a general curve. Let the string

(Here x is a 4 component vector
in the FP solution be described by the profile x = G(v).
362
Fig. 5. (a) Component strings with spins not all aligned; (b) The shape of the singularity.
Fig. 6. The singular curve when the FP string carries two different harmonics: Eq. (4.14) parameter values (a)
A = 1, B = 0.4; (b) A = 1, B = 0.6.
{x1 , x2 , x3 , x4 }.) For a generic state the typical wavelength involved in the oscillations is
much larger than the compactification radius R of the y coordinate. Then for any fixed
y = y0 the string passes through a set of points x that are closely spaced along a smooth
curve. This curve is given parametrically by

x = G(),
0 < 2nw R .
(4.15)
(By construction this curve is independent of the choice of y0 .) In the classical limit the
spacing of points along the curve goes to zero. Thus the classical geometry that corresponds

to this microscopic FP solution has a singularity along the curve x = G(),
with the

singularity extending over all 0 y < 2R and for all t. We will not explicitly mention
in what follows the extent in the y , t directions, and characterize the singularity by the
shape of the singular curve (4.15).
It is important that in the classical limit the large value of the winding nw gives a singularity
structure that is invariant under translations in y . This invariance in y allows us to apply
T-duality along y , which is needed to obtain the corresponding D1-D5 solution. The latter
solution will also have a singularity that is a curve in the space x , and which extends
uniformly in the y, t directions. The singular curve for the D1-D5 geometry is
g
x = F (), F = G.
R V
363
(4.16)
Given that the curve (4.16) is a 1-D subspace in 4 dimensions, the following situation is
generic:
(i) x = F () is a simple closed curve with no self-intersections.
(ii) |F ()| = 0 at all points along the curve.
Note that in the FP system if we take any y = y0 and look for the points in the x space
through which the string passes, then the number of these points per unit length along the
curve (4.15) is

G()
1 1 .
(4.17)
2R
The singularity is characterized by both its shape and the density (4.17).
Let us look at the D1-D5 system and examine the geometry for a general singular curve
F . By translational invariance we can set
L
dv Fi (v) = 0.
(4.18)
We will also assume that singularity in confined in the region with a typical size a, which
means that
Fi (v)Fi (v) a 2
for all v.
(4.19)
The throat geometry then has the following two properties:

(i) For x 2 a 2 we get the metric (Appendix E)
ds 2

1 2
dt dy 2 + h d x d x ,
h
(4.20)
where

Q5
Q1 1/2
h= 1+ 2
.
1+ 2
x
x
(4.21)
If we compute the time of flight for a quantum from the start of the throat to r = a and
back then we get
Q1 Q5
.
tSUGRA =
(4.22)
a
(ii) The geometry ends smoothly at r a except for the singular curve. We show in
Appendix F that waves incident on the singular curve in fact reflect from the curve, as
long as we have the genericity conditions (i) and (ii) mentioned above for the curve. In
364
Appendix G we estimate the maximum time tsing that a quantum can spend in the vicinity
of the singularity even if it heads radially into the singularity from r a. We find that
tsing
2
tSUGRA ,
n
(4.23)
where n is the average winding number of the component strings, and so n 1.3
Putting together properties (i) and (ii) we find that
tCFT tSUGRA .
(4.24)
We in fact expect an exact match between the CFT and supergravity, but if all the
component strings in the CFT are not of equal length then tCFT can only be defined
upto some spread. Similarly the waveform in the supergravity throat returns with some
distortion and thus tSUGRA has a spread as well. For this reason the above relation appears
as an approximate one, but in the cases where these time scales could be exactly defined
and computed we obtained the exact equality (3.13).
4.3. Hair in the D1-D5 geometry
For geometries that have a black hole horizon there exists the well known no hair
conjecture, which states that the microstates of the hole will not be visible in its geometry.
While this conjecture may indeed not be valid as a general theorem, it is nevertheless true
that we do not know how to see the entropy of black holes as a count of different spacetime
geometries. Thus for the D1-D5 momentum black hole in 4 + 1 dimensions, the metric
lifted to 6-D is

QP
2

1
dsE2 = dt 2 dy 2 + 2 dt + dy + h dr 2 + r 2 d32 .
(4.25)
h
hr
Classically this metric represents all the e2 n1 n5 np microstates of the hole. If we set the
momentum charge to zero, we still find e2 2 n1 n5 microstates. The geometry (4.25)

reduces to

1
dsE2 = dt 2 dy 2 + h dr 2 + r 2 d32 .
(4.26)
h
The classical horizon area is now zero, but the classical geometry (4.26) still seems to
exhibit a version of no hair, since it is the same for all microstates. How do we reconcile
this situation with the fact that we have found different throat geometries for different
microstates of the D1-D5 system?
It turns out that the geometry produced by generic microstates all behave as (4.26) upto
a classical distance down the throat (so that we see no hair) (Fig. 7(a)), but start to differ
at a distance which is 1/h down the throat, where we find the end of the throat and the
3 Note that for the metrics (2.10) we have performed the exact computation for the travel time t
SUGRA
(Eq. (3.7)); this time includes all effects of approaching the singularity and returning back, so that we do not need
to consider separately the time tsing spent near the singularity.
365
Fig. 7. (a) A leading order classical analysis suggests an infinite throat; (b) The throat actually ends at a distance
that diverges as h 0.
above discussed singularities (Fig. 7(b)). More precisely assume fixed the parameters of
the string background, and take the limit n1 , n5 this is the classical limit of the
solutions. The radius of the throat is (Q1 Q5 )1/4 (n1 n5 )1/4 . Choose any dimensionless
number , and look at distances that are reached by a particle falling down the throat for a
time t (Q1 Q5 )1/4 . As n1 , n5 , the metrics for generic microstates will all become
the same in this domain in the throat. The end of the throat comes for t (n1 n5 )1/2 , and
in the classical limit this will appear to be infinitely far down the throat. Thus we recover a
no hair classical limit, but see different geometries for different microstates near the end
of the throat.
4.4. A flat space computation to understand the size of the singularity
An essential part of the above results can be understood by considering the string of
the FP system in flat space. It was crucial in obtaining tCFT tSUGRA that the size of
the singularity, for fixed winding and momentum charges, depended on the wavelength of
the oscillations and not on the angular momentum they carried. Let the F string have total
length LT = 2nw R . Let it carry np units of momentum which implies the energy of
vibration
np
E= .
(4.27)
R
Further, let the wavelength of the vibrations be LT /n. The Hamiltonian for small
vibrations is
T
H=
2
z=L
T
dz
d x
dt

+
d x
dz
2
.
(4.28)
z=0
By the virial theorem

2
2 2
d x
n .
H = T
LT T x 2
dz
nw R
Setting H = E we get

2 1/2
N 1/2 1
.
x
2T
n
(4.29)
(4.30)
366
Thus we see that (for fixed N ) the rough size of the singularity depends on n, which
determines the wavelength in the FP solution, and maps to the winding number of
individual component strings in the dual D1-D5 system. In the supergravity solution
the size of the singularity determines the approximate end of the throat, and we see
from the above analysis that for the D1-D5 system the effective length of the throat is
determined by the number of component strings rather than the angular momentum of the
configuration.
For large N , the generic ground

state of the FP system has N quanta of vibration
modes with wavelength LT / N . In the dual D1-D5 system we find that for the
geometries (2.10) we have 1/ N , which is zero in the classical limit.
In Appendix H we compute the length of the singular curve for the D1-D5 bound state in
flat space. Multiplying this length with the height 2R of the singularity in the y direction
we find for the area of this classical singularity

Area 4 2 NG(6) ,
(4.31)
where equality is attained only for the cases where all component strings have the same
length.
4.5. Nonsymmetric excitations of component strings
Let the CFT state have m component strings. Note that the calculation of Appendix A
for the amplitude of absorption/emission from the component string did not depend on the
length of the component string. Thus an incident graviton will create an excited state of the
form

1
|excited = (string 1 excited) + (string 2 excited) + + (string m excited) .
m
(4.32)
If all the m component strings were identical, then (4.32) is the only state allowed by
Bose symmetry; this is the state that we used in the computation of Appendix A. But in
general the m component strings will not be in the same statethey may have different
lengths or different spin orientations. Then instead of (4.32) we can get the more general
state
|excited = 1 (string 1 excited) + 2 (string 2 excited)
+ + m (string m excited).
(4.33)
m/2 ++ m/2
] [N/m ]
and consider it excited
As a simple example, take the R ground state [N/m
to the nonsymmetrical combination

1
|excited = ( string excited) ( ++ string excited) .
2
(4.34)
A naive calculation would now suggest that the amplitude to emit a graviton from this state
is zerothe amplitude for emission from the two parts of the state give a total amplitude
1 [R R] = 0 (R is the amplitude of emission from either type of component string). It
2
367
would thus seem that there are CFT excitations that are decoupled from the supergravity
modes at infinity.
But a closer look reveals that this naive computation requires the following modification. We have seen above that the singularity of the D1-D5 system is not a point
in the space x but rather an extended curve. Even in the limit of weak coupling when
we have the branes in flat space, the location of the branes will be along this curve
of nonzero extent (see Appendix H for a computation of the length of the singular
curve). In the emission calculation if the component strings were all at one point then
the graviton emission amplitude would indeed vanish for the state (4.34). But the finite extent of the singularity implies that even if the leading order s-wave emission
is cancelled by a judicious choice of phase as in (4.34), the excitation will decay
(though more slowly) through the emission of p-wave and higher partial wave gravitons.
We can see the presence of states analogous to (4.33) in the dual supergravity. A simpler
example can be found using an unbound state rather than a single bound state. Thus
consider two identical D1-D5 bound states, each corresponding to the CFT state N .
Construct a superposed state where the centers of the two components are placed a distance
b apart, with b (Q1 Q5 )1/4 . Then the supergravity solution exhibits a single combined
throat for a certain distance, and then the throat branches into two smaller throats, which
we call throat 1 and throat 2.4
An s-wave scalar traveling down the combined throat enters throat 1 and throat 2
with equal amplitude. But we can also construct a wavefunction for the scalar which has
opposite amplitudes in throat 1 and throat 2. How will such a wavefunction emerge into
the combined throat and thus out to infinity?
While the above wavefunction does not emerge into the combined throat as an s-wave, it
does have a nonzero amplitude to emerge as a p-wave. The p-wave has a smaller amplitude
to exit the combined throat, and thus we see a suppression in the radiation rate for such a
wavefunction.
In a general state of the D1-D5 system the end of the throat exhibits no particular
symmetry, and so an s-wave going down the throat can get converted to a mixture of
harmonics after reflection from the end of the throat. The geometries (2.10) are special
however, since the different harmonics separate exactly and an s-wave returns as an s-wave.
The corresponding CFT states are special too, since their excitation takes the symmetrical
form (4.32) and not the more general form (4.33).
5. Breakdown of the semiclassical approximation and the threshold of formation of

black holes
5.1. Several vibration modes on the same component string
Consider an s-wave minimally coupled scalar, with energy , traveling up and down the
throat of the D1-D5 geometry. We have seen that this maps in the CFT to a left-moving and
4 Branching throats were considered in a slightly different context in [36].
368
a right-moving vibration mode on the effective string, each with energy /2. In particular
we had noted (Eq. (3.2)) the maps

1
h67 |x6 L |x7 R + |x7 L |x6R ,
2

1
RR
|x6 L |x7 R |x7 L |x6 R .
B67
2
(5.1)
We will find it convenient to work with the linear combinations

1
+
RR
|x6 L |x7 R ,
h67 + B67
S67
2

1
RR
|x7 L |x6 R .
S67
h67 B67
2
5.1.1. Throat corresponding to the CFT state with one component string
Consider the R ground state that arises from the chiral primary N . This state has
one component string wound n1 n5 times around the circle of radius R (Fig. 1(a)), and
thus has the longest possible throat (tCFT = tSUGRA = n1 n5 R). Place in this throat a
+
quantum of S67
with energy
=
2
0 .
n1 n5 R
(5.2)
This gives left and right vibrations on the effective string of energy 1/(n1 n5 R) each, which
is the lowest energy excitation possible.
+
Now imagine that we also place in the throat a quantum of S89
with energy 0 . In the
CFT we get the state
+
+
(0 )S89
(0 ) |x6L |x8 L |x7 R |x9 R .
S67
(5.3)
+
+
(0 ), S78
(0 ):
But now look at the CFT state corresponding to the supergravity scalars S69
+
+
S69
(0 )S87
(0 ) |x6L |x8 L |x7 R |x9 R .
(5.4)
The supergravity states in (5.3), (5.4) are different (the scalars involved are not the same in
the two cases) but they seem to map to the same state in the CFT. It would appear that we
have found a contradiction with the proposed duality map.
But note that these quanta in the throat have an energy of order 1/n1 n5 R or greater.
It was shown in [12] that for this throat geometry the threshold of horizon formation is
Ethreshold
1
.
n1 n5 R
(5.5)
Thus just when we seemed to be getting a contradiction between the CFT effective string
and the throat geometry, we find that the physics changes because of black hole formation.
Note that for this particular state of the CFT the supergravity approximation will not be
good even to describe the propagation of one quantum to the end of the throat, since the
backreaction on the geometry will deform the geometry by order unity.
369
5.1.2. Throats corresponding to CFT states with m component strings.

Let us see how the above discussion extends to states of the CFT that have m > 1
component strings. Let the CFT state be composed of m component strings with winding
number N/m each. Then in the supergravity solution we see that (Eq. (2.15)
1
m
Q1 Q5 .
a
(5.6)
n1 n5
R
The throat ends at r a while for r a it has the geometry (4.26) of the D1-D5 system
with no rotation. In the geometry (4.26) if we add a nonextremal energy E then we get a
horizon at
2

2g E 1/2
rH =
(5.7)
RV
(we have set = 1). Thus rH a then we get horizon formation, while if rH a then
the matter quanta just move back and forth in the throat, giving the hot tube [12]. The
threshold energy for black hole formation Ethreshold is then obtained by setting rH a,
which we write in a suggestive fashion as

m2
1
=m N
Ethreshold
(5.8)
.
n1 n5 R
mR
We see that a black hole forms if we have enough energy to excite one quantum of the
lowest allowed vibration mode on each component string.
The minimum wavelength that fits in the throat is (m0 )1 . Note that we can place
1 m1 m quanta of this wavelength in the throat without any significant deformation
of the throat geometry. Thus supergravity analysis performed in Subsection 3.3 using the
test particle approximation is seen to be valid, as long as we do not take too high an energy
(E > m2 0 ) for this test particle.
To summarize, we have seen that if we place more than one pair of vibration modes on
each component string, then the naive count of supergravity states fails to agree with the
count in the CFT (Eqs. (5.3), (5.4)). But the energy required to excite the lowest vibration
on each component string turns out to equal Ethreshold, the energy to form a black hole;
thus the physics changes at this point, and a contradiction is averted.5
5.2. Black hole formation
Thus far we have ignored all interactions when dealing with the component strings
in the CFT; we just considered left and right moving vibration modes on individual
component strings. But the supergravity solution does not in general map to the CFT at
the orbifold point M N /SN ; we have a deformation of the orbifold by blowup modes [21,
2 : the chiral
24]. Such deformations can be generated by (1, 1) operators of the form
5 In [37] it was found that in the c = 1 matrix model (D = 2 string theory) the process of fold formation
distributed an initial energy m2 as 1 + 2 + + m; i.e., we get one quantum each of frequencies 1, 2, . . . , m.
It may be that the interactions in the present problem distribute energies in this fashion rather than m2
m + m + + m as was suggested by (5.10).
370
primary 2 has dimensions ( 12 , 12 ) and , are dimension- 21 fermions from the left and
right sectors. This operator cuts or joins component strings, at the same time creating a
left- and a right-moving fermionic excitation on the resulting component strings [33].
To see what this interaction might do, let us look at the location of the horizon rH
(Eq. (5.7)) when we throw in an energy E Ethreshold. (We still assume that E is
low enough that the horizon forms in the throat rather than outside.) From our map between
the supergravity throat and the CFT we find that when a quantum travels from the start of
the throat to r = rH then on the CFT string the two vibration quanta separate by a distance
equal to the time of flight to rH :
(Q1Q5 )1/4
lH = (Q1 Q5 )
1/2
rH
dr
(Q1 Q5 )1/2
.
2
r
rH
(5.9)
This requires the CFT string to have length at least 2lH . When a black hole forms we must
clearly expect some change in the description on the CFT side. Using (5.7) we find that

4 2n1 n5 R
.
E
(5.10)
2lH
2lH
Thus we see that if the interaction were to change the CFT state to one where all the
component strings had length 2lH , then the energy E would excite the lowest allowed
vibration energy 2l4H on each of the resulting 2n2l1Hn5 R component strings.
Even though we are not investigating the interaction in detail here, there are two features
of the interaction that deserve comment. First, below the black hole formation threshold
we have obtained good results by ignoring the interaction altogether.6 Secondly, when a
horizon does form (and thus interactions presumably become relevant) then we seem to
need enough energy to excite all the component strings. We speculate now on how this all
or none feature of the interactions might arise.
In the CFT the normalized interaction operator has the form
2
N
1
ij .

N
(5.11)
i,j =1
Let us start with a CFT state which has n1 n5 singly wound component strings, and let the
first string carry a pair of vibrations. The above interaction can join another string to the
first one, giving a state of the form

1
(12) + (13) + + (1N) .
|f
(5.12)
N 1
From (5.11), (5.12) we see that the interaction generates a change
1
|f |i .
(5.13)
N
6 This may no longer be true when we have a bound state with component strings of different lengths. The
shape of the singularity is obtained by mapping to the dual FP system where the F string carries harmonics of
two different orders. The resulting singularity shape depends in a complicated way on all the harmonics present,
so we expect some interaction effects between the component strings in the CFT.
371
This small parameter 1/ N is also the coupling constant of supergravity in AdS3 , so it

appears that ignoring interactions in the CFT was equivalent to ignoring gravity corrections
at leading order in the dual supergravity.
If we use several such interactions, then we get further suppression by powers of 1/ N .

But now suppose that we convert all the initial component strings to component strings of
length 2, each carrying, say, the lowest vibration mode. Then because all these component
strings will be in the same state, we get a Bose enhancement factorif there are already
N units of a given component string then creating the next one gives a factor N .
Thus while the amplitude may be small to create a few interactions, it may be order unity
if it affects almost all the component strings, which is what we observed in the process of
horizon formation.7
5.3. Validity of the supergravity approximation
We have noted after Eq. (5.8) that if the number of component strings is m 1 then the
backreaction of a quantum with energy E m2 0 will not make a significant distortion to
the geometry (m2 0 = Ethreshold, the threshold for black hole formation). The geometry
of the throat is (except near the end) locally AdS3 S 3 M; since this is an exact string
solution it is not modified by stringy corrections. As we go down the throat the sizes of
S 3 and M do not change, but the length of the y circle shrinks. One might wonder if this
leads to a large effective coupling constant for quanta deep in the throat, thus possibly
invalidating the linear wave equation we have used for the propagation of the scalar.
But in [12] it was shown that if the scalar has for instance an interaction 3 then this
interaction becomes relevant only after a distance down the throat corresponding to a time
of flight n1 n5 R. But the end of the throat is reached after the quantum travels for a time
n1 n5 R/m. Thus again we find that if m 1 the interactions can be ignored, and we can
use the linear wave equation to explore the end of the throat at r a.
6. A proposal to resolve the information paradox

In [18] it was argued that the information paradox is resolved if we assume that spatial
slices in a foliation cannot be stretched too much. We review this argument, and then relate
it to the computations of the present paper.
6.1. The argument
The information paradox arises because we can find a smooth region in the black hole
spacetime which can be given a nice foliation by spacelike hypersurfaces; the spatial
slices satisfy all the smoothness conditions that we might wish to impose. On the initial
slice we have a regular distribution of matter, while late time slices capture both the initial
matter and the corresponding outgoing Hawking radiation. (The slices do not approach the
7 The issue of a phase transition at the threshold of black hole formation was discussed in [38].
372
singularity.) Since information cannot be duplicated, we must somehow have bleached

the information out of the matter and transferred it to the radiation, in the course of
the evolution. But before we can postulate the existence of some nonlocal process to
accomplish this goal we must identify the trigger for the process: what distinguishes the
smooth foliation of this region of the black hole spacetime from the foliation of a smooth
manifold without black holes? In the absence of such a trigger criterion, the postulated
nonlocal process would invalidate normal quantum evolution even in the absence of black
holes.
Since the foliation is smooth, the trigger cannot involve short distance quantum gravity
effects. Two features are however observed in the smooth black hole foliations. The first
is that one part of the spacelike slice evolves much slower than the other. This is the
redshift effect familiar in black hole geometries. But if we consider a scalar field evolving
along the foliation, we see no reason why this differential evolution should invalidate
normal quantum evolution. String theory is more complicated, but here again there is no
convincing reason why the differential evolution rates would lead to nonlocal information
transport (see, however, [39]). These are the standard difficulties that one runs into when
trying to resolve the information problem.
The second feature of the slices is that they stretch exponentially in the course of the
evolution. More precisely, if we want the slices to be smooth (thus they should not go
near the singularity) then the price that we must pay is that a region of the initial slice
(with radius Rs , the Schwarzschild radius) will get stretched to a final length that is
O(1/h ) and thus nonclassical.
It was argued in [18] that spatial slices should be attributed a density of degrees of
freedom. Stretching dilutes these degrees. If we place more matter quanta in a region of the
slice than there are available degrees of freedom then semiclassical evolution breaks down
and nonlocal information transport can be triggered (Fig. 8). This new way of violating
the semiclassical approximation allows information leakage in Hawking radiation while
preserving locality of quantum mechanics in the absence of black holes, and thus the
information paradox would be resolved. The length of this critically stretched slice was
expressed in [18] in terms of the number of degrees of freedom eS (S is the black hole
entropy), but noting that the separation of these degrees along the slice was order the
Schwarzschild radius Rs , we can write the critically stretched length of the slice as
smax
V
,
GN
(6.1)
where V is the volume enclosed in flat space in a ball of radius Rs .
Fig. 8. (a) An initial hypersurface with a high density of degrees of freedom; (b) Stretching dilutes these degrees;
(c) The degrees of freedom are so sparse that usual local physics breaks down.
373
The above arguments can be made on very general grounds, reflecting the universal
nature of the information problem. Let us now see what aspects of this picture emerge in
our D1-D5 system.
6.2. Stretching of slices in the D1-D5 system
(i) In the absence of the D1, D5 branes the spacetime geometry is flat (Fig. 9(a)). The
branes deform the geometry to a long but finite throat. Let us characterize the stretching
of the geometry by the time delay tSUGRA caused by propagation down and back up the
throat. We have found that the maximum value of this time delay is
tmax = Rn1 n5 .
(6.2)
We wish to relate this length scale (6.2) to the size of the region before stretching. The
radius r of the ball drawn in Fig. 9(b) is (Q1 Q5 )1/4 . The radius of the S 3 stabilizes to
a constant value in the throat (until we near the end), and for concreteness we set r to this
radius
r = (Q1 Q5 )1/4.
(6.3)
For thus value of r the enclosed volume in flat space (i.e., the space before deformation) is
2 4 2
(r ) =
Q1 Q5 .
2
2
We now note that
V=
tmax =
V
2G(5)
N
G
(6.4)
(6.5)
(10)
(10)
6 2 4
N
where G(5)
N = (2)5 V R and GN = 8 g . We observe that the relation (6.5) has the
same form as (6.1).
(a)
(b)
Fig. 9. (a) Flat spacetime in the absence of the branes; (b) The presence of branes stretches the region marked in
(a) to a long throat.
374
It is noteworthy that the individual quantities in (6.5) depend on g, , V , R, n1 , n5 but

these parameters appear only in certain physically intuitive combinations in the final
relation. This fact leads one to believe that the relation (6.5) has a validity beyond the
specific D1-D5 system being studied here.
(ii) Let the state in the D1-D5 CFT be given by m component strings of winding
number N/m each. The time delay in the supergravity throat is then tSUGRA
tmax /m. We have seen in Section 5 that if more than m supergravity quanta are
placed in this throat geometry then we get evolution that departs from naive semiclassical
expectations. This suggests that a throat stretched to have a time delay tmax /m has just
enough degrees of freedom to support m quanta while maintaining the supergravity
approximation. This indicates a dilution of the degrees of freedom when the throat is
stretched to longer lengths.
(iii) The argument of [18] requires that is we try to put more data on a slice than there
are degrees of freedom then we will trigger nonlocal effects. Let us examine a similar
feature that we find in our present study. Let the supergravity throat contain the two quanta
+
+
, S89
. Instead of putting these quanta in the lowest energy states in the throat let them
S67
have reasonably localized positions r1 = r2 . Let the CFT state have only a few component
strings. Then from the form of the wavefunctions (4.32) we see that there is a significant
amplitude to have both pairs of vibrations on the same component string. But this CFT
state can be regarded, a seen above in Section 5, as describing with some amplitude the
+
+
. If we try to extract the wavefunction for S69
however, we find
supergravity quantum S69
that this supergravity quantum is not localized near either x1 or x2 .
To summarize, while we have not analyzed in the present paper the same spatial
slices as those considered in [18], we have seen analogous phenomena of stretching and
consequent dilution of degrees of freedom; further, the relations expressing the maximum
stretch of spacelike slices (6.1), (6.5) appear to be similar.
7. Discussion
Let us summarize our results and note their relation to earlier work.
In [8,13] the Poincare patch arising from the near horizon geometry was considered.
It was noted that geometry stopped being AdS at large r (r > (Q1 Q5 )1/4 ). The naive
geometry Eq. (4.26) for the D1-D5 system (without rotation) extends as a uniform throat all
the way down to a horizon at r = 0. What we have noted is that the throat ends at a certain
point, and the location and nature of this end of the throat mirrors the D1-D5 microstate of
the dual CFT. In the process we found that the metrics of [16] were only a small family out
of a general family of metrics for the D1-D5 system, and we studied these more general
metrics using chiral null models.
Having this end to the throat was crucial to all our considerationsif we took instead
the naive geometry (4.26) then all quanta falling down the throat would reach the horizon
at r = 0 and create a black hole in the process. The actual ending of the throat implies a
minimum threshold Ethreshold for black hole formation, and yielded an energy domain
below this threshold where we could set up a detailed duality map to the CFT. The finite
375
length of the throat is essential to obtaining the relation tCFT = tSUGRA . In the analysis
of [41] a quantum falling down the throat of the 3-brane geometry was represented by a
spreading wavepacket in the CFT, but there was no analogue of the relation (3.13) since
the infalling quantum did not turn back without horizon formation.
The finite length of the throat arises both from the truncation at small r (r a) and
at large r (r (Q1 Q5 )1/4 ). We used explicitly the geometry at r (Q1 Q5 )1/4 (where
the throat joins flat spacetime) in computing the radiation rate RSUGRA from the throat,
and found that this equaled the radiation rate RCFT from the CFT state. Note that our
computations were somewhat different from earlier studies [44] where the join of the AdS
region to flat space was made at an arbitrary location in the AdS; the operator breaking
conformal symmetry in the CFT then depended on this location.
It is remarkable that the black hole threshold for different microstates depends only on
the energy scale 1/R and the winding numbers of the component strings in the CFT. We
noted that a breakdown of the semiclassical approximation occurred when there was more
than one pair of vibration modes on the same component string. Such a proposal would
appear to be closely related to the stringy exclusion principle, but there are some important
differences which we now discuss.
The microscopic D1-D5 CFT is manifestly unitary and the degrees of freedom are
limited. The information paradox asks for the implication of this fact in supergravity. It
has often been argued that the stringy exclusion principle [10] implies that there is a cut
off in the order of spherical harmonics on the S 3 . While this is probably true, it has not
yielded any direct implications for the information problem, in part because the Hawking
radiation is mostly confined to the first few spherical harmonics. What we have suggested
here is that the cutoff of degrees of freedom in the CFT implies a maximal radial stretching
of slices in the geometry, and the arguments of [18] then relate this to a resolution of the
information paradox.
A further essential difference is that we are looking at non-BPS excitations, rather than
the BPS quantities studied in [10,20]. If we consider a component string wound n times on
a circle of radius R, then left and right vibrations occur in fractional units but the total units
2
of momentum must be an integer. Thus the lowest non-BPS excitation has energy nR
, while
1
the lowest BPS excitation has (for large n) a much higher energy R . Correspondingly, we
find have found very low black hole thresholds for CFT states which have just one long
component string. By contrast the black hole threshold discussed in [20] pertains to the
NS sector of the CFT, and is a single, high energy threshold rather than a collection of
thresholds depending on the particular CFT microstate.
Perhaps most interesting is the fact that we have found some support for the proposal
of [18] that if we stretch a spatial hypersurface too much then we will encounter a
breakdown of normal semiclassical evolution, even though we would not see any of the
usual factors that violate the semiclassical approximation like planckian curvature or
planckian energies. It is interesting that the order of magnitude relation (6.1) emerging
from [18] has the same form as the more precise relation (6.5), though we must note that
the hypersurfaces involved in (6.1) and (6.5) are somewhat different. If the relation (6.5)
turns out to be a general result, as we have conjectured, then we would find a profound
change in our understanding of the ultimate structure of spacetime.
376
Acknowledgements
We are grateful to T. Banks, I. Klebanov, F. Larsen and A. Strominger for useful
discussions. This work was supported in part by DOE grant DE-FG02-91ER40690.
Appendix A. Rate of radiation from the effective string

Consider a string wrapped around the direction x5 y. There is one right-moving mode
and one left-moving mode on the string, with energy-momentum vectors given by (3.1).
The polarizations of the vibration modes are given by the wavefunction (3.2). We wish to
find the probability per unit time for these modes to interact and emerge from the string as
a graviton.
The vertex that couples a left vibration mode, a right vibration mode, and a graviton
is found by using the DBI action for the string [6,7]. We first place the entire system in
a large box. The volume of the noncompact directions x1 , x2 , x3 , x4 is Vnc , the volume of
the compact T 4 (in the directions x6 , x7 , x8 , x9 ) is (2)4 V , and the length of the x5 circle
is 2R. Thus the volume of the 9-dimensional space is V9 = (2)4 V (2)RVnc . We set
= 1. The 10-d Newtons constant is GN = 8 6 g 2 , and 8GN = 2 .
First consider the case where we have just one component string wrapped n1 n5 times
around y. For such a string in [7] we computed the amplitude per unit time for the state
|6L |7R to decay into any given Fourier mode of the graviton h67 . This quantity is
h =
R
1
1
2 |p5 | ,
2 V9
(A.1)
where p5 is the momentum of any one of the vibration modes along x5 , and is the energy
of the graviton. For the
state (3.1) of the vibration modes the amplitude per unit time will
be higher by a factor 2. Using that |p5 | = /2 we get for the amplitude of decay per unit
time
1
1
Rh = 2|p5 | .
2 V9
(A.2)
Using the Fermi golden rule we find the probability of emission per unit time
2|Rh |2
,
(A.3)
E
where E is the spacing between levels for the graviton. The graviton has momentum only
in the noncompact directions x1 , . . . x4 . Thus
R=
E =
(2)4 1
.
Vnc 2 2 3
(A.4)
Putting all this together we finally get for the probability of emission per unit time
R=
2 4 g 2
.
2V (2)R
(A.5)
377
m
Now consider the R ground state arising from the chiral primary [N/m
] . There are m
component strings, each in the same state. Because of Bose symmetry between identical
component strings, the wavefunction for the excited system must have the form

1
| = (string 1 excited) + (string 2 excited) + + (string m excited) . (A.6)
m
The amplitude Rh is now
1
1
Rh = m 2|p5|
(A.7)
2 V9
(there are m contributions each weighted by 1/ m). The probability per unit time for
decay is then
R=m
2 4 g 2
.
2V (2)R
(A.8)
Appendix B. Rotating D1-D5 system

In this appendix we review some properties of the geometries (2.10). These metrics
preserve a U (1) U (1) subgroup out of the SO(4) rotational symmetry of the S 3 . As we
have seen in this paper, these metrics represent only a small subclass of all the ground
states of the D1-D5 system, but they are useful since the scalar wave equation factorizes in
these backgrounds.
In supergravity approximation the system is parameterized by two charges Q1 , Q5 and
the rotation parameter a. The 10-D metric in the string frame is given by:8

4

f1
f0 2
dr 2
2
2
2
dt dy + f1 f5 2
ds =
+
d
dzi dzi
+
r + a2
f5
f1 f5
i=1

2
2
a Q1 Q5 cos
f1 f5
+
r2 +
cos2 d 2
f0
f1 f5

a 2 Q1 Q5 sin2
2
2
2
2
+ r +a
sin d
f1 f5

2Q1 Q5 a 2

(B.1)
sin dt d + cos2 dy d .
f1 f5
(2)
This system also has a nonzero value of dilaton field and the RR two-form C :
Q1 Q5 a cos2
f1
Q1
(2)
Cty
=
,
Ct(2)
=
,
e2 = ,
f5
f1
f1
Q1 Q5 a sin2
Q5 a 2 sin2 cos2
(2)
(2)
Cy
=
,
C
= Q5 cos2 +
.
f1
f1
(B.2)
8 In [34] we wrote the metric for rotating F1-NS5 system, but (B.1) can be obtained from it by applying the S
duality transformation.
378
Here we have introduced three convenient functions:

f0 = r 2 + a 2 cos2 ,
f1 = f0 + Q1 ,
f5 = f0 + Q5 .
(B.3)
The coordinates z1 , . . . , z4 form a torus with volume (2)4 V , and y is compactified on a

circle with circumference 2R.
From the microscopic point of view the system (B.1) describes a configuration of n1 D1
and n5 D5 branes with angular momentum J :

V
Q1 V
1
Q5
,
n
,
n1 =
e
H
=
=
H=
5
2
2
g
g
4 g
4
S3
S3
aV R
j=
Q1 Q5 .
2g 2
(B.4)
Here g is the string coupling constant and we always put = 1.

In Appendix C we will also need the relation between the solution (B.1) and the solution
for the rotating fundamental string carrying momentum. This relation was studied in detail
in [34], here we just mention some facts which will later be used. Using the chain of string
dualities one can map (B.1) into the solution describing the fundamental string which
is wrapped n5 times around the y circle, it also carries n1 units of momentum and J
units of angular momentum. The metric and matter fields of the solution describing the
fundamental string can be written in terms of the chiral null model. To do this one should
first go to the coordinates r , instead of r, :
r =
cos =
r 2 + a 2 sin2 ,
r cos
r 2 + a 2 sin2
(B.5)
and then introduce the Cartesian coordinates in the r , , , space:

x1 = r sin cos ,
x3 = r cos cos ,
x2 = r sin sin ,
x4 = r cos sin ,
(B.6)
as well as two null coordinates:

u = t + y,
v = t y.
(B.7)
In the coordinates (B.6), (B.7) the geometry of the rotating string reads (see [34] for
details):

2
Q
ds 2 = e2 du dv + 1 e2 1 dv
Q5

x1 dx2 dv x2 dx1 dv
Q1
2
e
4a
1
Q5
x x + a 2 + f0
+ d x d x + dz dz ,
Buv = Guv ,
Bvi = Gvi ,
(B.8)
e
= 1 + Q5 /f0 .
Here rescaled coordinates and charges are defined by:
R V
R V
z1 V

,
u =u
,
z1 =
xi = xi
,
g
g
gR1
Q1 R 2 V
Q5 R 2 V
aR V

Q1 =
,
Q
=
,
a =
,
5
2
2
g
g
g
V
V
VR

.
V =
,
g =
,
R1 =
R = V,
2
gR
g
gR1
1
379
(B.9)
(B.10)
(B.11)
Here f0 = f0 R g V (not the derivative of f0 !), and in terms of the new coordinates it is
given by

2 2
2
2
2
4 1/2
x x )2 + 2a x 3 + x 4 x 1 x 2 + a
.
f0 = (
(B.12)
The geometry (B.8) is an example of the chiral null model, we will discuss some properties
of such models as well as their relations to the D1-D5 systems in the next appendix.
Appendix C. A chiral null model approach to D1-D5 solutions

In Appendix B we have seen that under a chain of string dualities the rotating D1D5
system (B.1) is mapped to the rotating fundamental string carrying momentum charge.
The resulting solution (B.8) belongs to the class of the chiral null models [42]. Since
chiral null models have been used in the past as a powerful tool for generating new
solutions in supergravity, we will devote this appendix to studying such models and their
transformations under the chain of dualities relating the fundamental string (B.8) and the
D1-D5 system (B.1).
Let us first recall some of the properties of chiral null models [42]. The metric and
matter fields for such models are given by:

2
ds 2 = H (
x , v ) du dv + K (
x , v ) dv + 2Ai (
x , v ) dxi dv
+ d x d x + dz dz ,
1
Buv = Guv = H (
x , v ),
2
e2 = H 1 (
x , v ).
Bvi = Gvi = H (
x , v )Ai (
x , v ),
(C.1)
Regarding Ai as a gauge field we can construct the field strength Fij = Aj,i Ai,j . The
functions in the chiral null model are required satisfy the equations
2 H 1 = 0,
2 K = 0,
i F ij = 0.
(C.2)
Here 2 is the Laplacian in the xi coordinates. Note that the indices i, j span the subspace
{xi } where the metric is just ij , and thus these indices are raised and lowered by this flat
metric.
380
Note that the geometry (B.8) has a form of the chiral null model with
H 1 = 1 +
A1 =
Q5
,
f0
K =

2 Q1 Q5 a x2
f0 (f0 + x x + a 2 )
Q1
,
f0
A2 =

2 Q1 Q5 a x1
f0 (f0 + x x + a 2 )
A3 = A4 = 0.
(C.3)
We now wish to develop an analogue of the chiral null models directly for the D1D5 system; i.e., we wish to write supergravity solutions to the D1-D5 system in terms of
functions that can be linearly superposed to generate new solutions. We can obtain such
a formalism by applying dualities to the chiral null modes describing the FP system. But
at some stage in these dualities we will have to perform a T-duality along the direction y ,
so we look at FP solutions that are independent of y to start with. This means that the
functions H , K and Ai do not depend on v , but are only functions of x .
First we make an S duality transformation to make the original string into the D1
brane carrying momentum. Then by applying T dualities along all directions of the torus
(z1 , z2 , z3 , z4 ) we produce the D5 brane carrying momentum:

ds 2 = H 1/2 dt 2 dy 2 + KH 1/2(dt dy)2 + 2H 1/2Ai dx i (dt dy)
+ H 1/2 d x d x + H 1/2 dz dz,
(C.4)
e2 = H,
Ct(6)
i6789
(C.5)
(6)
= Ciy6789
= H Ai ,
(6)
Cty6789
= H 1.
(C.6)
Since we will not use this metric later on, we are not writing a rescaling of coordinates
which should be done to go from (C.1) to (C.4).
At this stage it is convenient to describe the RR fields not in terms of the six form
(C.6), but in terms of the dual two form C (2) . To construct this form we apply the electric
magnetic duality to (C.6). The field strength corresponding to (C.6) has following nonzero
components:
(7)
(7)
Gt ij 6789 = Gijy6789 = i (H Aj ) j (H Ai ),
(7)
Gt iy6789 = i H.
(C.7)
By applying the Hodge duality in ten dimensions we get a magnetically dual field strength:
ij kl
G(3)
l H 1 ,
ij k = H
(C.8)
G(3)
t ij
(C.9)
= G(3)
ijy
= H ij kl k Al .
(2)
This field strength corresponds to the two-form RR field C :

(3)
(2)
G = 3[ C] .
(C.10)
So far we were only looking at configuration with no dependence upon t, y, zi , so it is

(2)
natural to require that C has the same property. This requirement is equivalent to fixing
(2)
the gauge in C .
381
From the structure of G(3) we conclude that the only nontrivial components of C (2) are
(2)
Cij Cij ,
(2)
(2)
Ct i = Ciy Bi
(C.11)
and in terms of these new fields equations (C.9) become:

dC = dH 1 ,
dB = dA.
(C.12)
x 1, x 2, x 3, x 4
with
Here Hodge dual is taken with respect to the four-dimensional space
flat metric. Note that due to the equations of motion for the null chiral model (C.2):
d dH 1 = 0,
d dA = 0
(C.13)
the equations (C.12) can be integrated to give the forms C and B.

To summarize, the chiral null model dualized to the D5-P solution has metric (C.4) and
dilaton field (C.5), but the RR fields can be described in two alternative ways. We either
have a RR 6-form (C.6) or the RR 2-form
Cij(2) = Cij ,
(2)
Ct(2)
i = Ciy = Bi ,
(C.14)
which is related to (C.6) by the electric magnetic duality (C.9).

We can now apply S duality followed by T dualities along y and z1 to transform the
solution (C.4), (C.5), (C.14) into the F1-NS5 system. Application of another S duality
gives the D1-D5 solution:

2
2
H
ds 2 =
dt Ai dx i + dy + Bi dx i
1+K

1+K
d x d x + H (1 + K) dz dz,
(C.15)
+
H
e2 = H (1 + K),
(2)
Ciy
=
Ai
,
1+K
Bi
K
(2)
,
Cty
,
=
1+K
1+K
Ai Bj Aj Bi
Cij(2) = Cij +
.
1+K
Ct(2)
i =
(C.16)
The functions H , K and Ai appearing in this solution have the same values as H , K and
Ai :
x ),
H (
x ) = H (
K(
x ) = K (
x ),
Ai (
x ) = Ai (
x ),
(C.17)
and the forms Bi and Cij are defined by (C.12).

The functions H 1 , K, Ai , Bi can be linearly superposed to give different solutions of
the D1-D5 system. These functions depend only on x ; further Bi is determined by Ai as
described above.
Appendix D. Unbound solution

We wish to construct a solution of the D1-D5 system which represents two D1-D5
bound states, with opposite angular momenta so that the total angular momentum is zero.
382
We can either construct the solution from two strings in the FP system and perform dualities
to the D1-D5 system, or work directly with the chiral null models derived for the D1-D5
system in Appendix B and superpose two solutions. We will do the latter in this appendix.
Let us look at the simplest such superposition: we take a solution with Q1 , Q5 , a and
add it to the solution with Q1 , Q5 , a (actually we will take an average to preserve the
asymptotic value of H ). The resulting system is described by
H 1 = 1 +
Q5
,
f0
K=
Q1
,
f0
Ai = 0.
(D.1)
We can rewrite the result of superposition in terms of the original spherical coordinates
r, , , (see (B.5), (B.6)):

4

f0 2
dr 2
f1
2
2
2
dt dy + f1 f5 2
+
d
dzi dzi
+
ds =
r + a2
f5
f1 f5
i=1

Q5 2
f1
1+
r cos2 d 2 + (r 2 + a 2 ) sin2 d 2 ,
+
(D.2)
f5
f0
e2 =
f5
,
f1
(2)
Cty =
Q1
,
f1
(2)
C =
Q5 (r 2 + a 2 ) cos2
.
f0
(D.3)
To compare this solution with metrics for infinite throat and for the rotating D1-D5
branes we reduce the system (D.2) to six dimensions (t, r, , , , y) and look at the
resulting metric in the Einstein frame:

4
dr 2
f0 2
62
2
2
2
2
ds6 =
+ d
dt dy + f1 f5 2
dsE = e
r + a2
f1 f5

f1 f5 2
r cos2 d 2 + r 2 + a 2 sin2 d 2 .
+
(D.4)
f0
We see that this metric has the form of a uniform throat only for r a; the geometry
ends at r a. Thus the resulting throat has a length that reflects the value of |a| of each
component, rather than the total angular momentum (which is zero). We will examine the
nature of the singularity at r = 0, = /2 later on in Appendix E.
Appendix E. Throat geometry for generic D1-D5 bound states

We wish to understand
the supergravity solution corresponding to a general state out

of the set of e2 2 n1 n5 Ramond ground states of the D1-D5 system. While we have
seen in Appendix C how to write a metric using the idea of chiral null models directly for
the D1-D5 system, we will proceed by first writing the metric for the FP system and then
taking its dual. The reason for this is that we wish to make solutions that correspond to
a single bound state of the D1-D5 system, rather than to a multicenter solution obtained
from a collection of D1-D5 bound states. In the FP language a single bound state is given
by the oscillations of a single string, and we start with these solutions.
383
).
Consider a single fundamental string carrying a momentum wave with profile G(v
Then the solution (C.1) with
H
Q
,
)|2
|
x G(v
i (v )
Q G
(
x , v ) = 1 +
Ai (
x , v ) =
K (
x , v ) =
)|2
Q |G(v
,
)|2
|
x G(v
(E.1)
)|2
|
x G(v
).
describes a fundamental string located at x = G(v
As in [34] we construct a superposition of such solutions by smearing over the
coordinate y (or equivalently, by smearing over v). This smearing arises because there
are many closely spaced strands of the string in the space x , and in the classical limit their
smoothed out effect is just obtained by superposing the continuous distribution given by
the average over v :

1
(
x ) =
L
dv 1
H (
x , v ),
L
K (
x ) =
Ai (
x ) =
L
L
dv
K(
x , v ),
L
dv
A (
x , v ).
L i
(E.2)
):
Note that we obtain the solution (B.8) if we perform this smearing on the profile G(v
G1 (v ) = a cos(v + ),
G2 (v ) = a sin(v + ),
G3 (v ) = G4 (v ) = 0,
(E.3)
After duality transformations we get the D1-D5 solution (C.15) with coefficient
functions

Q
(
x) = 1 +
L
L

0
K(
x) =
Q
L
(E.4)
L

0
Q
Ai (
x) =
L
dv
,
(xi Fi (v))2
Fi Fi dv
,
(xi Fi (v))2
L

0
Fi dv
.
(xj Fj (v))2
(E.5)
(E.6)
(We will not need the explicit form of Bi and Cij .) From these expressions one can
see that the functions H 1 , K, Ai are regular everywhere, except at points where
by
x = F (v) for some 0 v < L. Note that F is related to G
g
F = G.
(E.7)
R V
384
After dimensional reduction on the compact 4-manifold M we get the following 6-D
Einstein metric for the D1-D5 system

H
1+K
2
i 2
i 2
d x d x
dsE =
(E.8)
+
dt Ai dx + dy + Bi dx
1+K
H
with parameters given by (E.4)(E.6). By translational invariance we can set
L
dv Fi (v) = 0.
(E.9)
We will also assume that singularity in confined in the region with a typical size a, which
means that
Fi (v)Fi (v) a 2
Then for
x 2
a2
for all v.
(E.10)
we get:

Q
x) 1 + 2 ,
H 1 (
x
L
Q 1 2
K(
x) 2
F dv,
x L
0

3/2
Ai (
x ) = O x 2
,

(E.11)
(E.12)
x) = O
Bi (

3/2
x 2
.
(E.13)
This leads to the leading approximation for the metric (E.8):

dsE2

1 2
dt dy 2 + h d x d x ,
h
(E.14)
where

1/2
Q
Q
h= 1+ 2
1+ 2
,
x
x
=Q1
Q
L
L
2
F dv.
(E.15)
(E.16)
This solution can be trusted in the region x 2 a 2 . We observe that in this approximation

we get a geometry of the D1-D5 system with Q5 = Q and Q1 = Q.
We now wish to define a parameter a which reduced to the parameter a for the special
geometries (B.1) but for a general throat measures the effective size of the singularity (and
thus determines the effective length of the throat). Thus we set

1
a
L
L
1/2
|F | dv
2
1/2

|F |2
.
(E.17)
385
Taking into account the relation between charges (E.16), we can rewrite this expression as

Q1 |F |2 1/2
a =
(E.18)
.
Q5 |F |2
Similarly, in the dual FP system we have

2 1/2
QP |G|

a =
,
Qw |G|
2
are related as in (E.7) and
where F , G
QP
V
aR
Q1

a =
,
=
.
g
Qw
Q5
(E.19)
(E.20)
In the FP system the string closes after nw = n5 windings around the circle y of length
2R , so we may write

nv
(n)
Ci cos
+ n,i .
Gi (v) =
(E.21)
n5 R
n=1
This leads us to the following averages:

(n) 2
2 =1
|G|
,
Ci
2
(E.22)
n=1 i
2
=
|G|

(n) 2
1
n2 Ci
.
2(n5 R )2
(E.23)
n=1 i
If we define an average harmonic n by

1/2
2
|G|
n n5 R
,

2
|G|
then equation (E.19) becomes:

QP n5 R

a =
.
Qw n
Using the relations (E.20) and (B.4) we find for the D1-D5 system
n1 n5 g
a =
.
n R V
(E.24)
(E.25)
(E.26)
We also define the generalization of the dimensionless parameter (Eq. (2.16)) for the
D1-D5 system:

aR
,
Q1 Q5
(E.27)
386
which yields (using (B.4))

=
1
.
n
(E.28)
Appendix F. Wave equation near the singularity

Let us look at the massless KleinGordon equation in the metric (E.8); the coefficient
functions in the metric are given by (E.4)(E.6). We will assume that the scalar field
does not depend on y coordinate, and it depends on t only through the factor exp(it):
(xi ).
(xi , t, y) = exp(it)
Then the wave equation becomes:

1+K
+ Ai Aj ij ii Ai 2iAi i + i i = 0.
2
H
(F.1)
(F.2)
x ), K(
x ),
From the expressions (E.4)(E.6) one can see that the functions H 1 (
x ) are regular everywhere, except at points where x = F (v) for some 0 v < L.
Ai (
Consider the wave equation in the neighborhood of the point x0 = F (v0 ) on the
singularity. We assume that (i) the singularity curve has no self-intersections, so that the
direction of the curve is well defined at x and (ii) |F (v0 )| = 0. Consider the 3-dimensional
plane normal to the singular curve at x0 , and parameterize the points in this plane by y :
xi = Fi (v0 ) + yi ,
Fi (v0 )yi = 0,
(F.3)
and we will look only at the displacements orthogonal to the singularity:

Fi (v0 )yi = 0.
(F.4)
Since y is a compact direction, all functions Fi (v) are periodic (Fi (v + L) = Fi (v)).
We then get
i Ai = 0.
(F.5)
Consider the potential term in (F.2)

1+K
2
Ai Aj ij .
H
(F.6)
Each of the coefficient functions H 1 , K, Ai diverge as y 2 near the singular curve. It

will be important for us that the combination occurring in the potential has a much softer
divergence ( y 1 ) rather than the naively suggested y 4 . We work this out below, and
give an explanation of this softening in Appendix I.
We have

1 + K H 1 Ai Ai

=
Q
L
2 L L
0 0
(F (v) F (v ))2

2
(
x F (v))2 (
x F (v ))2
dv dv
Q
+
L
L
0
(1 + Fi Fi ) dv
+ 1.
(
x F (v))2
387
(F.7)
We will consider only the curves without self intersection, then the main contribution to
this integral comes from the vicinity of the point v0 . Let us evaluate the contribution from
such vicinity to the first integral in (F.7):

Q
I1
L
2 L L
0 0
Q
2
L
2 L
0

2

2
Q
L
Q
L
(F (v) F (v ))2

2
(
x F (v))2 (
x F (v ))2
dv dv
L
dv (F (v) F (v0 ))2
dv
y 2 + (F (v) F (v0 ))2
y 2 + [F (v0 )]2 (v v0 )2
0
2 L
i
L
dv (Fi (v) Fi (v0 ))
dv (Fi (v ) Fi (v0 ))
y 2 + (F (v) F (v0 ))2
y 2 + (F (v) F (v0 ))2

|F (v0 )| y 2

Q
2
L
L
0

F (v0 )2
dv (F (v) F (v0 ))2

(F (v) F (v0 ))2
L
0
dv (v v0 )
2
y + (F (v) F (v0 ))2
2
.
(F.8)
The integral in the square brackets behaves as log(

y 2 ), and thus the leading asymptotics
for I1 is
1
C1 (v0 )
+ o |
y| ,
I1 =
|
y|
Q
C1 (v0 ) = 2
L
|F (v0 )|
L
0
dv (F (v) F (v0 ))2

.
(F (v) F (v0 ))2
(F.9)
In the same fashion we will get the asymptotics of the second integral in (F.7):
Q
I2
L
L
0
C2 (v0 ) =
1
(1 + Fi Fi ) dv C2 (v0 )
+ o |
y| ,
=
2

|
y
|
(
x F (v))
1
Q
1 + |F (v0 )|2
.
L
|F (v0 )|
(F.10)
(F.11)
Thus we find that the expression (F.7) behaves near the singularity as

C(v0 )
,
1 + K H 1 Ai Ai
|
y|
with C(v0 ) > 0.
(F.12)
388
To analyze Eq. (F.2) we will also need the asymptotic behavior of Ai , which can be
easily extracted from (E.6) and (F.10):
Ai
C3 (v0 )Fi (v0 )

,
|
y|
C3 (v0 ) =
Q 1
.
L |F (v0 )|
(F.13)
In the four-dimensional space x1 , x2 , x3 , x4 we have a vector y which parameterizes

a three dimensional subspace orthogonal to F (v0 ) and coordinate z which measures the
distance along the curve. The flat metric can be rewritten in terms of these four coordinates:
2
2

d x d x = d y d y + F (z) dz2 = dr 2 + r 2 d 2 + sin2 d 2 + F (z) dz2 . (F.14)
Here we have introduced the spherical coordinates r, , in the three-dimensional space
of y.
Then
substituting (F.12) and (F.13) into the Eq. (F.2), we get in the leading order in
r = y y and z:

2iC3 (z)
C(z)2
1
1
z +
z
z
r
r
|F (z)|
|F (z)|

1
+ r 2 r r 2 r + 2 , = 0.
(F.15)
r
The variables and in this equation separate and we will look for the solution in the
form:
(t, r, z, , ) = exp(it)Ylm (, )(r, z),
(F.16)
Let us first consider the case l > 0. In this case in the vicinity of r = 0 we get an
approximate equation:

l(l + 1)
r 2 r r 2 r
(F.17)
= 0.
r2
This is a Schroedinger equation with repulsive potential and the particle is reflected from
the barrier.
In the case of the S wave (l = m = 0) it is convenient to introduce a new coordinate
= r 1/2 . Then Eq. (F.15) becomes

1
z + 3 3 = 0.
C(z)2 2iC3 (z)z + 2 z
(F.18)
|F (z)|
Assuming that z derivatives of are bounded:
2
B2 ||,
|z | B1 ||,
z
(F.19)
we get the asymptotic equation near = 0:

3 3 + = 0.
(F.20)
This is just the equation for the three dimensional spherical wave which moves in the
constant potential. Such a wave reflects from = 0 after a finite number of oscillations
389
and in a finite time. In the next appendix we estimate the time spent near the singularity
and show that it is less than (or at most comparable) to the time spent reaching r a from
the start of the throat.
Appendix G. Null geodesics near the singularity

In this appendix we will look at null geodesics near the singularity, and estimate the time
which a particle traveling along such geodesics spends in the vicinity of the singularity.
To analyze the geodesics we consider the HamiltonJacobi equation
S S
(G.1)
= 0,
x x
in the background (E.8). We will assume that the particle does not move in the y direction:
g
S
= 0.
y
(G.2)
Then the HamiltonJacobi equation (G.1) becomes:
2
S S
1+K
S S
S
ij
+ Ai Aj
+ 2Ai
+ i i = 0.
H
t
t x i
x x
(G.3)
As in the Appendix F we will go from coordinates x i to the coordinates z, r, , using

(F.14). We will only study the geodesics which approach the singularity in the radial
direction. For such geodesics the Eq. (G.3) becomes:9
2 2
1+K
S
S
+ Ai Aj ij
+
= 0.
(G.4)
H
t
r
Since coefficients in this equation do not depend on t, the solution has the form:
S(t, r) = t +
S(r).
(G.5)
Near the singularity we can use the approximation (F.12), which allows us to rewrite (G.4)
as
C1 (z) + C2 (z)

S
(G.6)
,
r
r
where C1 (z) and C2 (z) are given by (F.9) and (F.11). For the generic singularity we have:

Q
C1 (z) 2
L
|F (z)|
|F |2 2Q5 1 Q1
=
.
L |F (z)| a 2
|F |2
(G.7)
9 Note that A diverges at r = 0 so there may be a residual contribution from the middle term in (G.3) even
i
though the geodesic is radial to leading order as r 0. We do not expect this to qualitatively change the analysis,
and so ignore such a potential contribution to (G.3).
390

Here we have used (E.16), (E.17) and the definitions of Q1 and Q5 : Q5 = Q, Q1 = Q.
From Eq. (E.16) we can see that if Q1 Q5 , then |F (z)| 1, and comparing (F.11) with
1/2
(G.7), we observe that C1 (z) C2 (z) for a Q1 .
Eq. (G.6) becomes

2 1 Q1 Q5 1/2

S
.
(G.8)
r
r L |F (z)| a 2
Integration of this equation gives the expression for the action (G.5)

2 1 Q1 Q5 1/2
.
S(t, r) = t + 2 r
L |F (z)| a 2
(G.9)
To get the travel time from r = r0 to r = 0 we differentiate this action with respect to :

1/2
2 r0 Q1 Q5 1/2 2 2 r0
=
tSUGRA ,
tsing = 2
(G.10)
L |F (z)| a 2
L |F (z)|
where
Q1 Q5
a
is the travel time from the start of the throat to r = a and back.
To determine the value of L we take the ratio of (E.18) and (E.19):

1/2

2 1/2 L
|F |2
|G|
a
=
= .
2
a
L
|G|
|F |2
tSUGRA =
(G.11)
(G.12)
are related by a simple

At the last step we used the fact that the profiles F and G

rescaling. Using the expression for a from (E.20) and the effective length for the FP
system: L = 2n5 R , we get:
L=
2R gn5 2gn5 2Q5

=
.
=
R
R
R V
Substituting this value in (G.10) and replacing |F (z)| by the average value

2 1/2
Q1
|F |
=
,
Q5
(G.13)
(G.14)
we get
tsing
2
=
tSUGRA
r0 R
(Q1 Q5 )1/2
1/2
.
(G.15)
Taking r0 a and using (E.27) and (E.28), we get:

tsing
2
2
= .
tSUGRA
n
(G.16)
391
Since n 1, we see that the time spent near the singularity is of the order or smaller
than the estimate tSUGRA . Note that for the metrics (2.10) we have performed the exact
computation for the travel time tSUGRA (Eq. (3.7)); this time includes all effects of
approaching the singularity and returning back.
Appendix H. Length of the singularity

Consider the FP system, and look at the length of the singular curve in the space x .
We assume that the coupling is weak, so that the metric is flat; the location of the string
will give a localized singularity representing the physical location of the string. From the
energy of oscillations we get:

2
nP
G
dy .
(H.1)
=
T

R
Then the length of the string is

1/2

nP L
1/2

2

2
L
dy
D = dy |G|
= 2 N ,
=
|G|

RT
(H.2)
1
where T = 2
. Note that equality is attained only if all the vibrations of the F string are
in the same harmonic.
Let us map this to the D1-D5 system, and consider this system also at weak coupling so
that spacetime is flat everywhere except the singularity. Then we get for the length of the
singular curve
3/2
g 3/2
NG(6)
g

.
D = D 2 N = 2 2
(H.3)
R
R V
R V
We can then write for the area occupied by the singularity in the x , y space

Area = 2RD 4 2 NG(6) ,
(H.4)
where we again note that equality is attained only if all the component strings in the
microstate are of equal length (this is equivalent to all vibrations being in the same
harmonic for the dual FP system).
Appendix I. Singularity curves having self-intersections

Consider the potential term (F.6) that we found in the scalar wave equation near the
singular curve of the D1-D5 geometry. We had found that this term was much less
singular than would appear from the behavior of its individual factors. This softening of
the singularity was essential for the fact that the wave reflects back in a finite time from
the singularity. Let us look at the FP system, and investigate the essential reason for this
softening as well as the situations where the wave equation may become more singular.
392
Suppose we have a single strand of a fundamental string oscillating with some profile
). Then we find, using (E.1), that
G(v

P K H 1 1 Ai Aj ij = 0.
(I.1)
If, however, we have two strands of the string (whether joined into one string or arising
from two separate strings) then
P=
Q1 Q2
1 (v )|2 |
2 (v )|2
|
x G
x G

G
2 (v )2 ,
1 (v ) G
(I.2)
2 (v ) are the same.

1 (v ) and G
which is nonzero unless the profiles G
The quantity P differs from the potential (F.6) only by terms that are comparatively
nonsingular. Now consider the multiwound string. If the singular curve it produces in the
space x has no self-intersections, then neighboring strands along this curve have almost
). In that case the quantity (I.2) is comparatively regular, and thus
the same profile G(v
we see that the softening of the potential in the wave equation can be traced back to the
vanishing of P for a single strand.
But if the singular curve has self-intersections, then at the same point in the space x
we have strands with quite different profiles G1 (v ), G2 (v ). The denominators in (I.2)
now cause P to be large, and the potential in the wave equation is correspondingly more
singular.
Since the singular curve is a 1-D hypersurface in the 4-D space x , the generic
singular curve has no self-intersections. But since simple examples may actually have such
intersections, we examine two such cases in this appendix.
(i) The unbound state solution (D.2) was constructed by superposing two bound
states, each of which had a singularity on the same circle of radius a. Let us look at the dual
FP system. Since the rotation was oppositely directed in the two components, we find at
each point along the singular curve a pair of strands with different profiles G1 (v ), G2 (v ).
Thus each point of the singular curve gives a more singular potential in the wave equation
than the generic 1/|y|
(Eq. (F.15)).
We now find that geodesics that go radially into any point with r = 0, = /2 do not return
back in a finite time t. The divergence in this time of flight is, however, only logarithmic:
if we separate the two components of the D1-D5 solution by a distance r then the time
of flight to the singularity again becomes finite and of order tCFT log r
a . But now note
that each bound state in the solution had some mass M, and thus can be localized in the
space x only to an accuracy | x| 1/M. If we set Q1 Q5 for the moment, and work out
the effect of this fluctuation, then we find that
tSUGRA tCFT log j,
(I.3)
where j is the angular momentum of each of the bound states in the solution. We will not
explore the meaning of this logarithm further, but just note that such logs have appeared
before in relating CFT and supergravity quantities [43].
(ii) As a second example we write down an explicit solution for the single bound state
where the singularity is a straight line: this will be the limiting case where the ellipse in
Fig. 5 degenerates and the angular momentum becomes zero.
393
Consider the string of the FP system oscillating in the direction x1 :

G1 (v ) = a cos(v + ),
G2 (v ) = G3 (v ) = G4 (v ) = 0.
(I.4)
Averaging over in the usual manner we get the classical solution (C.15) with coefficient
functions

Q5 2(r 2 z2 + a 2 ) + 2 (r 2 z2 + a 2 )2 + 4z2 r 2 1/2
1
,
H = 1+
(I.5)
2r
(r 2 z2 + a 2 )2 + 4z2 r 2

1/2

Q1
2
K = 2 2r + 2 r 2 z2 + a 2 + 2 r 2 z2 + a 2 + 4z2 r 2
(I.6)
,
ra
Ai = 0.
(I.7)
At each point of the singularity we have two differently moving strands of the string
arising from the two sides of the degenerating ellipse. Correspondingly, we find that there
is a logarithmic divergence in the time of flight to each point on the singularity; in addition
)| vanishes as well,
there is a stronger singularity at the endpoints x1 = a since here |G(v
and so the density of points along the singular curve (4.17) diverges .
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]
O. Aharony, S.S. Gubser, J. Maldacena, H. Ooguri, Y. Oz, Phys. Rep. 323 (2000) 183, hep-th/9905111.
S.R. Das, S.D. Mathur, Annu. Rev. Nucl. Part. Sci. 50 (2000) 153, gr-qc/0105063.
S.W. Hawking, Commun. Math. Phys. 43 (1975) 199.
A. Strominger, C. Vafa, Phys. Lett. B 379 (1996) 99, hep-th/9601029.
C.G. Callan, J.M. Maldacena, Nucl. Phys. B 472 (1996) 591, hep-th/9602043.
S.R. Das, S.D. Mathur, Nucl. Phys. B 478 (1996) 561, hep-th/9606185.
S.R. Das, S.D. Mathur, Nucl. Phys. B 482 (1996) 153, hep-th/9607149.
J. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231;
J. Maldacena, Int. J. Theor. Phys. 38 (1998) 1113, hep-th/9711200.
O. Coussaert, M. Henneaux, Phys. Rev. Lett. 72 (1994) 183, hep-th/9310194.
J. Maldacena, A. Strominger, JHEP 9812 (1998) 005, hep-th/9804085.
G.T. Horowitz, H. Ooguri, Phys. Rev. Lett. 80 (1998) 4116, hep-th/9802116.
O. Lunin, S.D. Mathur, hep-th/0107113.
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.
V. Balasubramanian, P. Kraus, A.E. Lawrence, Phys. Rev. D 59 (1999) 046003, hep-th/9805171.
V. Balasubramanian, J. de Boer, E. Keski-Vakkuri, S.F. Ross, Phys. Rev. D 64 (2001) 064011, hepth/0011217;
J. Maldacena, L. Maoz, hep-th/0012025.
D.Z. Freedman, S.D. Mathur, A. Matusis, L. Rastelli, Nucl. Phys. B 546 (1999) 96, hep-th/9804058;
E. DHoker, S.D. Mathur, A. Matusis, L. Rastelli, Nucl. Phys. B 589 (2000) 38, hep-th/9911222.
S.D. Mathur, Int. J. Mod. Phys. A 15 (2000) 4877, gr-qc/0007011.
N. Seiberg, E. Witten, JHEP 9904 (1999) 017, hep-th/9903224.
J. de Boer, JHEP 9905 (1999) 017, hep-th/9812240.
R. Dijkgraaf, Nucl. Phys. B 543 (1999) 545, hep-th/9810210.
F. Larsen, E.J. Martinec, JHEP 9906 (1999) 019, hep-th/9905064.
J.R. David, G. Mandal, S.R. Wadia, Nucl. Phys. B 564 (2000) 103, hep-th/9907075.
R. Dijkgraaf, E. Verlinde, H. Verlinde, Nucl. Phys. B 500 (1997) 43, hep-th/9703030.
394
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
A. Jevicki, M. Mihailescu, S. Ramgoolam, Nucl. Phys. B 577 (2000) 47, hep-th/9907144.

O. Lunin, S.D. Mathur, hep-th/0103169.
M. Cvetic, F. Larsen, Phys. Rev. D 56 (1997) 4994, hep-th/9705192.
J. Maldacena, A. Strominger, Phys. Rev. D 55 (1997) 861, hep-th/9609026.
J.M. Maldacena, L. Susskind, Nucl. Phys. B 475 (1996) 679, hep-th/9604042.
A.W. Peet, J. Polchinski, Phys. Rev. D 59 (1999) 065011, hep-th/9809022.
J. Maldacena, A. Strominger, Phys. Rev. D 56 (1997) 4975, hep-th/9702015.
S.S. Gubser, Phys. Rev. D 56 (1997) 4984, hep-th/9704195.
S.D. Mathur, Nucl. Phys. B 514 (1998) 204, hep-th/9704156.
O. Lunin, S.D. Mathur, Nucl. Phys. B 610 (2001) 49, hep-th/0105136.
S.R. Das, S.D. Mathur, Phys. Lett. B 375 (1996) 103, hep-th/9601152.
J. Michelson, A. Strominger, JHEP 9909 (1999) 005, hep-th/9908044;
J. Gutowski, G. Papadopoulos, Phys. Rev. D 62 (2000) 064023, hep-th/0002242.
S.R. Das, S.D. Mathur, Phys. Lett. B 365 (1996) 79, hep-th/9507141.
S.D. Mathur, Nucl. Phys. B 529 (1998) 295, hep-th/9706151.
D.A. Lowe, J. Polchinski, L. Susskind, L. Thorlacius, J. Uglum, Phys. Rev. D 52 (1995) 6997, hepth/9506138.
A. Dhar, G. Mandal, S.R. Wadia, Phys. Lett. B 388 (1996) 51, hep-th/9605234.
T. Banks, M.R. Douglas, G.T. Horowitz, E.J. Martinec, hep-th/9808016.
G.T. Horowitz, A.A. Tseytlin, Phys. Rev. D 51 (1995) 2896, hep-th/9409021;
A.A. Tseytlin, Phys. Lett. B 381 (1996) 73, hep-th/9603099.
I.R. Klebanov, S.D. Mathur, Nucl. Phys. B 500 (1997) 115, hep-th/9701187.
S.R. Das, S.P. Trivedi, Phys. Lett. B 445 (1998) 142, hep-th/9804149;
S.S. Gubser, A. Hashimoto, I.R. Klebanov, M. Krasnitz, Nucl. Phys. B 526 (1998) 393, hep-th/9803023.

Phenomenological implications of neutrinos

in extra dimensions
Andr de Gouva, Gian Francesco Giudice, Alessandro Strumia 1 ,
Kazuhiro Tobe
Theoretical Physics Division, CERN, CH-1211, Genve 23, Switzerland
Received 25 July 2001; accepted 4 December 2001
Abstract
Standard Model singlet neutrinos propagating in extra dimensions induce small Dirac neutrino
masses. While it seems rather unlikely that their KaluzaKlein excitations directly participate in
the observed neutrino oscillations, their virtual exchange may lead to detectable signatures in
future neutrino experiments and in rare charged lepton processes. We show how these effects can
be described by specific dimension-six effective operators and discuss their experimental signals.
2002 Elsevier Science B.V. All rights reserved.
PACS: 14.60.St; 14.60.Pq; 14.60.Ef; 13.15.+g; 13.35.Bv
1. Introduction
The hypothesis that Standard Model (SM) singlet fields propagate in extra dimensions
leads to striking results. When applied to the graviton, it allows to lower the quantum
gravity scale down to few TeV [1,2], suggesting a new scenario for addressing the Higgs
mass hierarchy problem. It is also natural to consider the case of right-handed neutrinos
(i.e., fermions without SM gauge interactions) propagating in extra dimensions. The
smallness of the neutrino masses, of the Dirac type, could in fact be a manifestation of
this hypothesis [36].
If the radius of the compactified dimensions is very large, R eV1 , Kaluza
Klein (KK) modes of right-handed neutrinos would significantly participate in neutrino
E-mail address: astrumia@difi.unipi.it (A. Strumia).
1 On leave from Dipartimento di Fisica dellUniversit di Pisa and INFN, Italy.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 2 1 - 6
396
A. de Gouva et al. / Nuclear Physics B 623 (2002) 395420
oscillations. However, KK interpretations of the atmospheric and solar neutrino puzzles

are disfavoured by the following arguments:
A KK tower of sterile neutrinos gives rise to active/sterile oscillations at a small
m2 1/R 2 only in the case of a single large extra dimension. In the case of more
large extra dimensions, the active/sterile mixing is not dominated by the lightest KK
modes, and the mixing with the heaviest KK modes does not lead to oscillations. One
could argue that right-handed neutrinos in a single very large extra dimension can be
effectively obtained if one of the extra dimensions happens to be much larger than the
other(s). However, the fact that infrared effects are dominant in this case destabilises
the hierarchy [7]: the Newton or Coulomb potential grows with the size R of the largest
dimension.
Even if one forgets about the hierarchy problem, there are severe bounds from supernov observations [8]. One needs to prevent resonant neutrino conversion in supernov by choosing a small radius R 1/ MeV or by adding an ad-hoc 5-dimensional mass term m MeV for the right-handed neutrino(s) responsible for atmospheric oscillations. The latter still allows to build models for the solar anomaly
roughly compatible with supernov bounds [9].
Finally, the recent results from the SuperKamiokande [10,11] and SNO [12] experiments provide strong indications for appearance in atmospheric oscillations and for
, appearance in solar oscillations. In both cases, a sterile interpretation is now
strongly disfavoured by experiments.
In this paper we explore the phenomenology of more promising models with > 2
extra dimensions and radii which are not larger than what is required to reproduce
the gauge/gravitation hierarchy. The neutrino puzzles are solved by normal (active)
oscillations, but the presence of the heaviest KK neutrinos can still lead to small but
detectable effects in neutrino flavour transitions. After integrating out the heavy KK modes,
we obtain an effective Lagrangian that contains massive Dirac neutrino states and a specific
set of nonrenormalizable operators. Since the dominant effects come from the heaviest KK
states, the coefficients of these operators can only be estimated by introducing an arbitrary
ultraviolet cut-off.
At tree level one only obtains the dimension-six operators

Ltree =
ij 2 2 GF H L i i/
(1.1)
(H Lj ),
where Li are the lepton left-handed doublets, H is the Higgs doublet, and
ij =
ji are
dimensionless couplings. The presence of these effective operators leads, for example,
to potentially large flavour transitions P (i j ) |
ij2 | at O(L0 ) (i.e., at very short
baselines L
E /m2 ), and CP-violating effects at O(L1 ) (rather than at O(L2 ) and
O(L3 ) as in ordinary oscillations). Since this peculiar tree level operator only affects
neutrinos, potentially detectable effects, especially at a neutrino factory, are not already
excluded by bounds from rare charged leptons processes, like , e , eee,
etc. We will show, however, that Eq. (1.1) cannot explain the LSND anomaly [13], due to
the present constraints from searches for e .
397
At one-loop level, other operators are generated, giving rise to rare muon and tau
processes that violate lepton flavour. Again, the coefficients of these operators are cut-off
dependent and can only be estimated. However, in minimal models, their flavour structure
is directly related to the physical neutrino mass matrix, giving predictions for rare muon
processes in terms of neutrino oscillation parameters. This is in contrast with other cases of
physics beyond the SM. In supersymmetry, for instance, the rates for rare muon processes
are perturbatively calculable, but their relations with neutrino oscillations parameters are
strongly model-dependent and can vary by many orders of magnitude.
Many of the effects studied here have already been considered in previous analyses
[14,15] as due to mixing between ordinary neutrinos and the whole tower of KK states. The
equivalent language of effective operators we employ allows the study of different models
(e.g., large and warped extra dimensions) in a unified framework, the discrimination of
what is really computable from what can only be estimated, and a more transparent
identification of all potentially interesting experimental signatures.
The paper is organized as follows. In Section 2, we describe the models under
consideration. In Section 3 we discuss the effects of the tree level operator Eq. (1.1) in
neutrino physics. In Section 4 we discuss the effects of the operators induced at one-loop
in charged lepton processes. In Section 5 we summarize our results.
2. Right-handed neutrinos in extra dimensions

We will study models with large flat extra dimensions [1,35] and models with one
warped extra dimension [2,6].
2.1. Large flat extra dimensions
We consider (4 + )-dimensional massless fermions i (x , y) which, inside their
components, contain the degrees of freedom of the right-handed neutrinos Ri (i =
1, 2, 3 is the generation index). The fermions i interact in our brane, through their
components Ri , with the standard left-handed lepton doublet Li in a way that conserves
total lepton number. The relevant part of the action is

4

S = d x d y i iD
(2.1)
/ i + d 4 x L i i/Li L i ij Rj H + h.c. ,
where iD
/ is a (4 + )-dimensional Dirac operator, H is the SM Higgs doublet in fourdimensions and is a matrix of Yukawa couplings with dimensions (mass)/2 . As is
manifest from Eq. (2.1), can be made diagonal without loss of generality at the price of
introducing the usual unitary matrix U , which describes flavour-changing charged current
neutrino interactions.
Using the KK decomposition

1
i n y
,
i (x, y ) =
(2.2)
n i (x) exp
R
V
n
398
where V (2R) is the compactified volume, and performing the d y integration,

Eq. (2.1) yields the four-dimensional neutrino Lagrangian

ij
n

n i Li R n j H + h.c. ,
n i i/
Li +
L = Li i/
(2.3)
R
V
n
where are the extra-dimensional Dirac matrices.After electroweak symmetry breaking,

the neutrinos obtain a Dirac mass matrix m = v/ V , where v = 174 GeV is the vacuum
expectation value of the Higgs boson. The heavy modes give corrections suppressed by
Rm , that are negligible in the cases of interest.
In order to compute the tree-level effects due to the presence of the tower of KK
states, we construct an effective Lagrangian by substituting in Eq. (2.3) the solutions of
the equations of motion for the heavy fields,

n 1 ij
Lj H.
n i = i/
(2.4)
R
V
With an abuse of notation, we have identified Lj in Eq. (2.4) with a higher-dimensional
spinor, in which Lj fills the components corresponding to the right-handed neutrinos, while
all other components are zero. Ignoring higher derivative terms and summing over all KK
states up to an ultraviolet cut-off , we obtain the effective operator in Eq. (1.1) with
coefficients
ij = &
( )ij 2 v 2
,
2
&
S
,
(2)
(2.5)
where S = 2 /2 / ((/2) is the surface of a unit-radius -dimensional sphere, is the

dimensionful matrix of Yukawa couplings and parameterizes some ultraviolet cut-off.
We take to be the KK mass at which we cut off the summation. The operator in Eq. (1.1)
is gauge invariant, in spite of containing an ordinary derivative (rather than a covariant
one), because H L is a SM gauge singlet. The coefficients
ij of this operator have the
same flavour structure as the physical neutrino masses m m , although its overall
factor is model-dependent. The order of magnitude of this single free parameter is fixed
ifmotivated by the hierarchy problemwe assume that all extra-dimensional physics
is at the TeV scale, 2/ TeV. In this case the coefficients
ij in Eq. (2.5) are of
order & (v/ TeV)2 , where & is a typical loop factor in dimensions. It is useful to write the
coefficient of Eq. (2.5) as
ij
(m m )ij /m2atm , because
is the only free parameter
of the model (at tree level).2 Explicitly,
& m2atm
V .
(2.6)
2 2
We now recall how this model can be linked with the Higgs mass hierarchy problem. If
gravity propagates in d extra dimensions (d , where all extra dimensions have the same
4 = 2.4 1018 GeV is
radius R and the topology of a torus), the reduced Planck mass M
2 Throughout the paper, we will assume that the neutrino masses obey a normal hierarchy, i.e., m2
m2
1
2
m23 , so that m2atm m23 and m2sun m22 .
399
D for different values of /d, assuming = M

D . The
Fig. 1. m / (left) and
(right) as a function of M
horizontal band in the left panel shows the mass range selected by atmospheric neutrino data for = 1.
D in D = 4 + d dimensions by
related to the reduced Planck mass M
2+d .
2 = (2R)d M
M
4
D
(2.7)
It is useful to rewrite the dimensionful Yukawa coupling in terms of a dimensionless

Yukawa parameter as
.
M
/
D
/2
With this definition, the neutrino mass becomes

/d
MD
.
m = v
4
M
(2.8)
(2.9)
D in order to obtain neutrino masses

The simplest case = d requires very large or M
that satisfy the atmospheric neutrino data (see Fig. 1). On the other hand, for = 5 and
d = 6,
5/6
MD
m =
(2.10)
3 102 eV,
TeV
D TeV, while the solar
and the atmospheric mass scale is obtained with 1 and M
neutrino puzzle can be solved for values of slightly smaller than one. Fig. 1 (left)
400
D , for different values of /d. Supernov

depicts values of m / as a function of M
bounds [8] force all KK neutrino states to be heavier than the typical supernova temperature
T 30 MeV. In the present model, these bounds imply d 3. The reduced Planck mass
D is related to the phenomenological parameter MD used to study graviton effects at
M
D is M
D 200 GeV
D . The present collider bound on M
colliders as MD = (2)d/(2+d)M
(for d 6) while the LHC can improve it by a factor 10 [16]. Fig. 1 (right) depicts the
D for different
D in the case = M
values of
, defined in Eq. (2.6), as a function of M
values of and d, and for m2atm = 3 103 eV2 . It is important to keep in mind that
D ) ). For a fixed ratio /M

D ,
D (it scales like (/M

is strongly enhanced if > M
2
decouples like 1/ , as the new-physics scale increases. This behavior is clearly visible in
Fig. 1.
One may consider a few variations to the minimal model described above:
(A) The right-handed neutrinos could have some extra dimensional mass term m m [4].
The mass term m does not affect the dimension-six operators, but affects neutrino
massesnow related to m and by a higher-dimensional see-saw relation, m
2 /m. In these models, the
ij coefficients are not directly related to neutrino masses,
and therefore contain additional mixing angles and CP violating phases beyond the
ones in the neutrino mixing matrix.3
(B) Different massless right-handed neutrinos could have a different UV cut-off, or live
in a different number of extra dimensions. An interesting case is obtained with 2
right-handed neutrinos living, respectively, in 5 and 6 extra dimensions with equal
radii. In this case one can reproduce the smallness of the solar m2 with respect to
the atmospheric m2 using comparable Yukawa couplings (see Fig. 1). This model
contains 4 mixing angles and 2 CP-violating phases (rather than the 3 mixing angles
and 1 CP phase of the minimal model). The reason is that, unlike in the minimal
model, it is not possible to perform flavour rotations of the right-handed neutrinos.
Qualitatively, the main new feature of nonminimal models is that the tree level exchange of
bulk neutrinos again yields the operator Eq. (1.1), but the overall coefficient is now different
for the atmospheric and solar contributions. Assuming that all Yukawa couplings are
of order of the fundamental mass scale, in case (B) the solar contribution becomes
comparable to the atmospheric contribution, i.e., all
ij in Eq. (1.1) are expected to be of
the same order, contrary to the minimal model where
/
e m2atm /m2sun , assuming
that |Ue3 | is negligible (this will be discussed in detail in the next section). Warped models
generically give effects qualitatively similar to case (B), as we describe below.
2.2. Warped extra dimensions
A strong suppression of gravity could be generated by an extra-dimensional red-shift
factor [2,17]. We will concentrate on the simplest scenario [2], containing one extra
3 If m m , more than three mass eigenstates participate in oscillations. Explaining the origin of a higher
dimensional mass term comparable to the neutrino masses is, perhaps, the biggest challenge for this type of
scenario.
401
dimension with topology S1 /Z2 (i.e., a segment), parameterized by a coordinate y. The

SM fields live on a brane fixed at one of its borders (y = R), and another brane lives
at the opposite border (y = 0). After an appropriate fine-tuning of the cosmological
constant in the extra dimension and the tension of the two branes, the background
metric has the form ds 2 = e2ky dx dx + dy 2 , where the warping constant k is
determined by the five-dimensional cosmological constant and gravitational scale M
as k 2 = /24M 3 [2]. By rescaling the metrics at y = 0 to its canonical form, one finds
that all mass scales in the four-dimensional SM Lagrangian get red-shifted by a factor ekR
(including the ones that should suppress unwanted nonrenormalizable operators), and that
2 = (1 e 2kR )M 3 /k. One can try to explain the
the four-dimensional Planck mass is MPl
gauge/gravitation hierarchy by stabilizing the size of the extra dimensions R [18] such that
ekR TeV/MPl .4
It is again interesting to consider five-dimensional right-handed neutrinos with
Yukawa couplings to the SM fields living on a brane. In this case it is necessary to give
some five-dimensional Dirac mass terms m to the extra-dimensional neutrinos in order to
naturally explain the smallness of the neutrino masses [6]. Despite this higher-dimensional
mass term, there is still one very light KK mode, and its effective Yukawa coupling with
the active neutrinos is strongly suppressed if m > k/2. Then, the Dirac neutrino masses
depend very strongly on m, m vemR , and neutrino masses which span many orders
of magnitude are easily obtained even if all the higher-dimensional Yukawa couplings are
comparable.
The other KK states have TeV-scale masses and unsuppressed four-dimensional Yukawa
couplings to SM fields. They generate the operator Eq. (1.1) with coefficient
e2kR v 2
ij .
(2.11)
k
Because this is a (4 + 1)-dimensional theory, the infinite sum over all KK modes is finite,
and the introduction of an arbitrary ultraviolet cut-off is unnecessary. On the other hand,
it is not useful to perform a precise computation because, similarly to the nonminimal
flat models, there is no direct connection between
ij and the neutrino masses. The main
conclusion is that all
ij have comparable values,
ij 10few , if , k, M have comparable
mass scales. Smaller
ij can be obtained for smaller Yukawa couplings .
We do not consider nonminimal warped models, because they do not seem to lead to
new interesting effects in neutrino physics.5
ij
4 It has been conjectured that this model is equivalent to walking composite technicolour [19]. This dual
version is not as stetichally appealing, but its problems with experimental data are better known.
5 In more general extra-dimensional black-hole backgrounds, the space warping factor can be different from
the time warping factor [20]. This could lead to right-handed neutrinos that travel with a velocity c(1 + c)
different from light. This model has a conjectured holographic dual [19] where Lorentz invariance is broken in a
more obvious and old way [21]: normal right-handed neutrinos have an index of refraction n due to interactions
with an ther (composed of some hot conformal matter). At tree level and up to negligible m/E corrections,
these effects do not affect neutrino oscillations as long as there are only Dirac mass terms [22], or more generically
as long as the energy eigenstates do not mix left with right-handed neutrinos. Even in models where this is not
the case, neutrino effects caused by a nonuniversality of the speed of the light do not seem phenomenologically
interesting because they have an energy dependence (different from the observed one [23]) that allows to derive
402
3. Tree level effects in neutrinos

The tree-level operator Eq. (1.1) only affects neutrinos.6 For this reason, Eq. (1.1) can
give rise to detectable effects in neutrinos which are compatible with bounds from chargedlepton processes, affected only at the one loop order. After electroweak symmetry breaking,
the operator in Eq. (1.1) contributes to the kinetic term of the neutrinos. The effective
Lagrangian for the 3 left-handed neutrinos and for the zero mode right-handed neutrinos
R becomes
L = i i/
(ij +
ij )j + R i/
R ( m R + h.c.)

g
g
/ + h.c. +
Z
/ ,
+ &W
(3.1)
2 cos W
2
where & = (eL , L , L ) are the left-handed charged leptons in the mass eigenbasis, m is
the neutrino mass matrix, and
ij are adimensional numbers defined in Eq. (2.5), expected
to be of order & (v/ TeV)2 . The kinetic and mass terms can be simultaneously diagonalized
by the following nonunitary field redefinitions
R UR R ,
L UL KUL L .
(3.2)
Here U are unitary matrices

(such that UL diagonalizes
ij ) and K is a diagonal matrix
whose elements are 1/ 1 +

i , where
i are the eigenvalues of
ij . In this new basis the
neutrino kinetic and mass terms are flavour-diagonal, but the neutrino interactions with the
gauge bosons are modified
R (n mn Rn + h.c.)
L = i/
+ R i/

g
g
+ &VW W
(3.3)
/ + h.c. +
VZZ
/ ,
2 cos W
2
where mn are the mass eigenvalues given by mn / 1 +

n , VW UL KUL and VZ
UL K 2 UL . It is now easy to compute oscillation probabilities. The transition probability
for a neutrino both produced and detected via a charged-current W interaction is given by

W W iE t 2
m2nm
n

.
Vin Vf n e
En Em =
P (i f ) =
(3.4)
,
2E
n
Differently from the standard case, this expression cannot be simplified to the usual form,
because V W is not a unitary matrix.
The minimal model described in detail in the previous section predicts
ij (m m )ij ,
such that UL = 1. The transition probability reduces to the more transparent form
2

Uin Ufn

iEn t

e
P (i f ) =
(3.5)
,
2
2
n 1 +
mn /matm
very strong constrains. Upward through going atmospheric muons in SuperKamiokande (with L 104 km and
E TeV) put the bound c 1/LE 10(2425) (for large mixing between flavour and velocity eigenstates),
preventing detectable effects in planned new neutrino experiments.
6 The operator in Eq. (1.1) does not contribute to the h decay, due to conservation of angular momentum.
Only the emission of KK states lighter than mH yields h n corrections to Higgs decays [5,24].
403
where U is
the usual unitary neutrino mixing matrix. Notice that the summed survival
probability f =e,, P (i f ) 1 2
is not equal to 1, because of the nonunitary
effects. In the effective theory description one detects less neutrinos than in the SM because
their gauge interactions with the W and Z bosons are suppressed with respect to the SM,
due to the presence of the effective operator (1.1). In the fundamental theory this same
effect is due to transitions of active neutrinos into sterile KK states.
Next, we discuss a few possible experimental signals, concentrating on searches for
neutrino oscillations.
3.1. Flavour transitions at very short baselines
In terms of the
ij parameters in Eq. (1.1), the transition probability at a very short
baseline (L 0) derived from Eq. (3.4), for small
ij , is given by
P (i j ) |
ij |2
(i = j ),
P (i i ) 1 2
ii .
(3.6)
The present bounds on the flavour violating

ij =
ji , coming from neutrino experiments,
are
|
| < 0.013,
|
e | < 0.09,
|
e | < 0.05.
(3.7)
The dominant bounds on

and
e are due to the N OMAD experiment7 [25], P (
) < 1.6 104 and P (e ) < 7.4 103 at 90% CL. The bound on
e comes from
K ARMEN [26] and N OMAD [25]. The LSND anomaly [13], which may be interpreted as
evidence for P ( e ) 2.5 103 could tentatively be solved if |
e | 0.05. This
is in slight conflict with the bound from K ARMEN, and in strong contradiction with the
bounds from e , which we will discuss in the next section.
The diagonal elements
ii =
ii are constrained by C HOOZ [27] and Bugey [28],
yielding
ee < 0.025 and, because of the modified neutrino couplings to the Z boson, by
the invisible Z width measured at LEP [29], which leads to |
ee +
+
| < 0.013 at
90% CL.8 The &i W couplings are a factor 1
ii /2 smaller than in the SM, so that lepton
universality tests [30] in and decays give
ee
= 0.0027 0.0024,
ee
= 0.0011 0.0045.

= 0.0003 0.0042,
(3.8)
A slightly more stringent bound,

ee +
< 0.002, can be obtained by comparing a global
fit to the precision LEP data with the lifetime. The bounds on the flavour violating
ij from Z, , , decays are not competitive with the ones from neutrino oscillation
searches.
7 The CERN neutrino experiments, N OMAD and C HORUS , were motivated by theoretical prejudices for small
mixing angles and for warm dark matter. Today they are among the most significant probes of extra-dimensional
neutrinos.
8 Since the present measurement of neutrino counting at LEP agrees with the SM only at the 2- level, and
a nonzero
reduces the effective number of neutrinos to N = 3 2 Tr
ij , this measurement gives a possible
indication of a positive effect,
ee +
+
= 0.008 0.004.
404
In the flat minimal model, the

ij parameters are determined in terms of the single
unknown parameter
, defined in Eq. (2.6), and of the measurable (and already partially
measured) neutrino oscillation parameters as

m2sun
2
ee |Ue3 | +

,
2
2m2atm

|Ue3 | ei m2sun
|Ue3 | ei m2sun
e +
(3.9)
e
2
2 2m2atm
2
2 2m2atm
where we assumed maximal solar and atmospheric mixing. Here Ue3 is the element of the
neutrino mixing matrix which is currently constrained to be small by the C HOOZ reactor
neutrino data [27] and is the CP-violating phase. The bounds previously discussed can
be turned into constraints on
, in the case of the minimal model. It turns out that the most
stringent bounds are
0.004 (from , , decays and precision LEP data) and
0.03
(from neutrino experiments).
The most sensitive future experimental search seems to be or wrong signed
appearance at a future near-detector of a -factory. Such a detector, located at 1 km
from the neutrino source, has already been studied in detail as a tool for neutrino deep
inelastic scattering experiments [31]. It expects to observe 108 charged current SM events
per year. It seems possible [31] to probe
and
e values down to the level of 104 ,
by looking for appearance, and
e down to similar values by looking for wrong
sign muons, therefore improving the present bounds by two orders of magnitude. More
dedicated studies, however, are still required. The sensitivity on the flavour diagonal
from searches in the disappearance channel seem to be slightly worse, being limited by the
uncertainty on the incoming neutrino flux. The sensitivity on
ee should be weaker than the
one to
, while
effects cannot be probed.
3.2. CP violating effects
In order to understand qualitatively what the general exact formula (3.4) means in the
various possible models, it is useful to specialize it to the short-baseline case ij
1, where
ij m2ij L/2E . Furthermore, it is useful to count how many CP-violating phases are
contained in the
ij =
ji parameters.
1. In processes where neutrino masses can be neglected, the
ij contain (Ng 1)(Ng
2)/2 CP-violating phases (where Ng = 3 is the number of generations) which do not
give rise to any observable effects.
2. If the
ij can be neglected (normal oscillations), the neutrino mixing matrix contains
the usual (Ng 1)(Ng 2)/2 CP phases. For Ng = 3 there is one CP-violating phase,
, and the CP-asymmetric part of the oscillation probability is the same in all flavour
transitions i j = i , and, for small L, is proportional to L3 :
P (i j ) P (i j )
2
13
23
12
sin
sin
JCP 12 13 23 ,
= 8JCP sin
2
2
2
PCP
(3.10)
405
where JCP 12 Ue3 sin (assuming maximal mixing in atmospheric and solar
oscillations, and |Ue3 |
1).
3. In a generic process affected by neutrino masses and by the
ij , the neutrino mixing
matrix contains the usual (Ng 1)(Ng 2)/2 CP phases, and the
ij parameters
contain other Ng (Ng 1)/2 CP phases. Therefore, unlike the standard case, CP
violation can be present when only the dominant two-generation atmospheric
oscillations is turned-on.
For example, ignoring subleading sun effects and keeping terms up to second order in atm

iatm 2

P ( ) =
sin 2atm
,
2
m2atm
L 20 GeV
atm
= 104
.
2
3
2
3 10 eV km E
(3.11)
The fact that these two effects could be comparable offers an opportunity to measure
CP-violating effects present, if
is complex, as a difference in P ( ) from
P ( ) proportional to L. The CP asymmetry can be maximal with a suitable choice
of the pathlength and neutrino energy, if
has a large CP-violating phase [32].
In the minimal model presented in the previous section, the only CP-odd phase is the
one present in the neutrino mass matrix (recall that
ij (m m )ij ), and new CP-violating
effects (proportional to L) are suppressed by JCP and by two powers of
:

2
PCP
(3.12)
= JCP 212
+ 12 13 23 .
Given the current constraint on
0.01, such effects are negligible.
For specific nonminimal models, it is necessary to check whether the (indirect) relation
of
ij and the neutrino mass matrix allows to rotate away potentially physical phases
in
ij . This is the case if a single right-handed neutrino gives a dominant contribution
to
ij , as happens in the minimal model. Furthermore, in order to obtain observable effects, it is necessary to generate a large
ij with a large phase, while keeping |
e | small,
due to the severe constraints already imposed by searches for e (see next section).
In spite of all the difficulties, however, it is possible to build specific models which yield
significant CP-violating effects.
3.3. Matter effects
Nonflavour diagonal interactions with the Z, W bosons give rise to nonstandard MSW
corrections. The corresponding matter potential, however, is too small to solve the solar
neutrino puzzle or affect significantly the SM effect due to charged current e e scattering.
On the other hand, the SM matter effects in oscillations are suppressed by
(m /MW )2 [33], so that non-SM effects could be dominant and perhaps detectable.
In the minimal flat model, oscillations in neutral matter are modified to

m2atm

2 GF Ne L + O
2 , (3.13)
P ( ) = 1 sin2 (2atm ) sin2
4E
8
406
and could result in a narrow resonance in long baseline oscillations of high energy antineutrinos, E 30 GeV/
. Similar effects have been discussed in [34].
Matter effects are also known to substantially change the expected neutrino fluxes from
supernov [35]. The new matter effects due to the KK right-handed neutrinos produce a
new MSW resonance in supernov which seems, however, not to perturb the observable
e and e spectra in a significant way.
4. One-loop effects in charged lepton processes
The effects of any high energy theory can be described at much lower energies by
higher-dimensional operators. The relevant set of operators and their coefficients depend
on the ultraviolet cut-off which is used to make computations in the low energy effective
theory. Using dimensional regularization (so that no power divergences arise in loop
computations), the low energy effects of bulk neutrinos are described by a few dimensionsix operators, including &i &j magnetic moment operators, &i &j Z vertices, and four
fermion interactions.
These effects are partly due to computable SM loop corrections to the operator in
Eq. (1.1), generated at tree level. However, comparable contributions come from high
energy effects in the full theory containing the KK modes of the higher-dimensional
fermions. As before, it is important to emphasize that the coefficients of the effective
operators cannot be computed, since we do not have a renormalizable high-energy theory.
The coefficient of each operator is a free parameter that cannot be calculated in terms
of the nonrenormalizable Yukawa couplings of the extra-dimensional neutrino. At best,
they can be estimated by introducing some explicit cut-off . It proves particularly useful
to rewrite the dimensionful Yukawa coupling as
,
&
/2
(4.1)
because we can interpret the dimensionless couplings as parameters for understanding

how strongly coupled the right-handed neutrinos are.9 As defined in Eq. (2.5), & is a
-dimensional loop factor (for example &4 1/(4)2 ). All virtual effects are comparable
to the effects of a four-dimensional heavy right-handed neutrino with a Yukawa coupling
(so that 4 corresponds to strong coupling). The effective Lagrangian at energies
smaller than the cut-off is

&4

2
/LH + 2 e 2 [ e ] + 2 2 + g 2 gv 2 [Ze] + [ 3e]
L 2 LH

+ 2 g 2 [q eq] .
(4.2)
We have used a shorthand notation to indicate the various SU(2)L U(1)Y -invariant
L). While the
operators. For example, gv 2 [Ze] indicates operators like (H D H )(L
9 When in Section 2 we discussed the connection of neutrino masses with gravity, it was convenient to
parameterize the dimensionful fundamental Yukawa coupling in terms of another dimensionless coupling ,
defined using the reduced Planck mass as the unit of mass.
407
magnetic operator [ e ] is generated at order 2 , other dimension-six operators also

receive contributions at order 4 . In the case of the minimal model presented in Section 2,
the loop effects are explicitly calculated (making use of a hard ultraviolet cut-off) in
Appendix A.
Note that all dimension-six operators decouple as 1/2 when . The correct
decoupling is also found in the expressions presented in Ref. [15], once the definition of
the mixing parameters is taken into account.
We remark that we are only considering the effects due to the extra-dimensional righthanded neutrinos: a full quantum gravity theory at TeV energies is expected to give
additional effects. In particular, it is important to note that while some unknown mechanism
could be responsible for suppressing effects which lead, e.g., to proton decay, it is hard
to believe that there are no extra contributions to lepton flavour violating processes,
given that the neutrino data indicates that individual lepton flavours are strongly broken
symmetries (see [36] for a discussion of these effects and possible ways of addressing
this issue). Nonetheless, we will ignore here such contributions, which are impossible to
estimate.
In spite of all the intrinsic uncertainties, useful results can be obtained. Both the tree
level and the more relevant one-loop operators are suppressed by the same order of
/.
This implies that one-loop effects are only suppressed by a factor of order 1/16 2
with respect to the tree level term.10 Some operators, like the four-fermion operators
contributing to 3e and e conversion in nuclei, may in fact be enhanced
with respect to e by 2 /g 2 if > g. Moreover, the coefficients of the operators
included in [Ze] and [ 3e] receive a ln 2 /m2W enhancement. Extra-dimensional
models predict (up to order one factors) relations between flavour violating effects in the
charged and neutral lepton sectors. In the simplest flat model we discussed in Section 2,
predictions for many charged lepton flavour violating processes will be strictly related to
the observed neutrino oscillation parameters (m2 , and neutrino mixing angles), such that,
e.g., it is possible to predict BR( )/BR( e ).
Therefore, minimal extra-dimensional models are more predictive than other beyondthe-SM sources of lepton flavour violating phenomena related to right-handed neutrinos.
In the constrained MSSM, for example, the presence of very heavy right-handed neutrinos
(that generate small neutrino masses via the see-saw mechanism) yields potentially large
flavour violating effects in the muon sector. While all branching ratios are precisely
calculable in terms of the parameters of the theory (the MSSM is a perturbative gauge
field theory), it is impossible to establish a connection between the observed flavourviolating neutrino masses with predictions for e , etc., due to the presence of too
many unknown flavour-mixing parameters in the right-handed neutrino sector.
In this section, we concentrate on charged lepton flavour violating phenomena, and
also comment on the anomalous magnetic moment of the muon. Note, however, that the
effective Lagrangian (4.2) also allows e+ e &+

i &j (i, j = e, , ) processes [15]. Such
effects could be studied at the Z resonance with a next-generation e+ e collider. However,
10 This nice property is not shared by virtual effects mediated by higher-dimensional gravitons, that generate
dimension-6 operators at loop level but only dimension-8 operators at tree level.
408
searches for processes with i = j do not seem very promising given the current bounds on
ij [37], and the effects on processes with i = j seem less sensitive than corrections to Z
observables, which are generated at tree level.
Explicit expressions for the processes of interest are derived and listed in Appendix A.
They agree, apart from minor differences, with the corresponding expressions in [15] when
the same model and the same arbitrary cut-off is chosen. Our numerical results are, instead,
different. In particular, the experimental bounds we obtain are much weaker and do not,
in general, require fundamental scales above 100 TeV or 10 TeV. We discuss some of our
results in what follows.
4.1. Flat extra dimensions
We will concentrate on the minimal model outlined in Section 2. This model contains
only two new free parameters,
and . Up to order one factors, the rates for the different
charged lepton process depend only on these two parameters, on the neutrino masses, and
on the elements of the standard neutrino mixing matrix Ui . The largest Yukawa coupling
is then determined by the relation

=
( 2).
v
(4.3)
As mentioned before, we will assume that the neutrino masses are hierarchical (m21
m22
m23 ) such that all information regarding neutrino parameters can be obtained from
neutrino oscillation experiments. We later comment on nonminimal flat models and warped
models.
4.2. e and
Using the expressions derived in Appendix A, the branching ratio for e can be
written as follows:

2
2
m2j 2 3 2

3 2
msun

BR( e ) =
Ue2 U2
Uej Uj
=
+ Ue3 U3 .

2
2
8
8
matm
matm
j
(4.4)
Similarly, the branching ratio for is given by
BR( ) = 0.174

m2j 2
3 2

Uj Uj
8
m2atm
j

2

m2sun
3

= 0.174
2 U2 U2
+
U
U
3 3 .
2
8
m
(4.5)
atm
The branching ratio for e depends on the solution to the solar neutrino puzzle
(i.e., Ue2 and m2sun ) and on the unknown Ue3 element. On the other hand, BR( )
409
depends dominantly on the well measured atmospheric parameters:11 |U3 U 3 | 0.5 and
m2atm 3 103 eV2 , such that
2
3
BR( ) 0.174
2 U3 U3 = 4 105
2 .
(4.6)
8
The bound Eq. (3.7) from neutrino experiments on

/2 implies, in a modelindependent way, that a possible effect in is at least two orders of magnitude
below the present limit BR( ) < 1.1 106 .
The dependence on the free parameter
cancels out in the ratio between BR( e )
and BR( ), that is expressed only in terms of neutrino oscillation parameters,

2
2 2
BR( e ) Ue2 U2 msun + Ue3 U3 matm
=
.
(4.7)
2

BR( )
0.174 U3 U 3 m2atm
The numerical value depends on the still unmeasured Ue3 and on which is the true solution
of the solar neutrino puzzle. The dominant contribution to e could be related
to the atmospheric or solar neutrino mass splitting. If the solar anomaly is due to
LMA oscillations, and if |Ue3 | is such that m2sun Ue3 m2atm , the solar contribution
dominates, and we predict

2
m2sun
,
BR( e ) 1 108
2
(4.8)
3 105 eV2
2

0.5
BR( e )
4 |Ue2 U2 |
3 10
BR( )
0.35 |U3 U 3 |

3 103 eV2 2
m2sun
(4.9)
.
3 105 eV2
m2atm
When compared to the current experimental bound BR( e ) < 1.2 1011, we obtain
the limit

m2sun
3 105 eV2
|
< 3 102
(4.10)
if
|U
e3
m2sun
m2atm
In the near future the experimental sensitivity to BR( e ) is expected to reach
1014 [38].
On the other hand, if the solar parameters fall in the LOW or SMA regions,12 and |Ue3 |
is large enough, the atmospheric contribution to BR( e ) dominates, so that

BR( e ) Ue3 /0.3 2

. (4.11)
BR( e ) 4 105
2 |Ue3 /0.3|2,
BR( )
U 3 /0.7
In this case the bound on
becomes
< 5 104 |0.3/Ue3 | if
|Ue3 | m2sun/m2atm .
(4.12)
11 The e branching ratio is suppressed by the unknown reactor angle U , so that it is smaller and more
e3
uncertain than the branching ratio.
12 After SNO [12], the SMA solution is strongly disfavoured by the solar neutrino data [39].
410
Fig. 2. The most interesting leptonic observables as a function of |Ue3 |, for = 5,

= 0.003 (top) or 0.0003
(bottom), and = 1 TeV (left) or 10 TeV (right). We assume m2sun /m2atm = 102 , |U3 /U 3 |2 = 1,
|Ue2 /Ue1 |2 = 2/3 (i.e., maximal mixing in the atmospheric sector and the LMA solution to the solar neutrino
puzzle), no CP-violation in the neutrino mixing matrix, and hierarchical neutrino masses (m21
m22
m23 ). Note
that is the largest Yukawa coupling, P (i j ) is the neutrino conversion probability at very short baselines,
and the e conversion rate is computed for 27Al.
If Ue3 is close to its current experimental upper bound, |Ue3 | 0.3, the bound on
is so
strong that experimental signals in the neutrino sector can only be very small.
Fig. 2 shows the values of BR( e ) and BR( ) as a function of |Ue3 | for
m2sun/m2atm = 102 , |U3 /U 3 |2 = 1, |Ue2 /Ue1 |2 = 2/3 (as suggested by atmospheric
and LMA solar oscillations), and no CP violation in the neutrino mixing matrix, for
different values of
(and ). If the solar and atmospheric contributions to e
are comparable (which happens at Ue3 10(32) for LMA solar oscillations), the CP-violating phase in the neutrino mixing matrix may either enhance or suppress the branching
ratio, depending on how the two terms interfere.
411
4.3. eee and e conversion in nuclei

Unlike the &i &j decays, these rare leptonic processes are more dependent on the
unknown ultraviolet details of the models, but still a few interesting results can be obtained.
In eee process, several operators may contribute significantly (see Appendix A for
detailed expressions). First of all, the e magnetic penguin operator contribution to
eee is enhanced by ln(m /me ), which is a consequence of a collinear divergence
of the electronpositron pair in the me 0 limit. Second, the Z and photon penguin
diagrams are enhanced by an ultraviolet divergence ln(/mW ). When is significantly
larger than mW , this log-enhancement is important. Finally, as discussed after Eq. (4.2), the
Z-penguin and box contributions contain 4 -terms. If the higher-dimensional KK neutrinos
are coupled strongly enough such that > g, then the 4 -terms may dominate over the
other contributions to BR( eee).
In light of this discussion, the ratio of branching ratios BR( eee)/BR( e ), in
which the overall
2 dependence in both processes cancels out, can be written as
BR( eee)
Ce + Cln + C .
BR( e )
Here, Ce is the contribution from the e magnetic penguin operator

m2 11
Ce =
ln 2
= 6 103 .
3
me
4
(4.13)
(4.14)
The second term Cln contains the ln /mW contributions from the Z and photon penguin
diagrams of eee. For example, the Z-penguin diagram leads to

5 2
3
1
2
2
2
Cln
+
ln 2
3 2
8
2 3
mW
sin W
2 sin2 W

5 2
2
2
103 ln 2
.
(4.15)
2 3
mW
We find that Cln = 0.03 (0.2) for = 5 and = 1 TeV (10 TeV), when all ln terms
present in the Z and photon penguin diagrams are included.
The last term in Eq. (4.13), C , contains the contributions from 4 terms in the
Z-penguin and box diagrams:

4
(m2 /m2 )2 + U U 2
Ue2 U2
e3 3
sun
atm
C
(4.16)

.
g
Ue2 U (m2 /m2atm ) + Ue3 U 2
sun
2
3
When |Ue3 | is small, C is small (suppressed by the small ratio of neutrino mass-squared
differences squared). On the other hand, if |Ue3 | is sufficiently large, C may in fact be
the dominant contribution to eee due to the potentially large ( /g)4 enhancement.
Numerically, C = 0.6 for = 10 TeV, |Ue3 | = 0.1, and
= 0.003, which corresponds to
= 5.4.
Because of the contributions Cln and C , the ratio BR( eee)/BR( e ) can
be significantly larger than 101 if 1 TeV or > g. This is very different from
412
predictions for lepton flavour violating processes from SUSY models with slepton flavour
mixing, in which the contribution Ce is almost always dominant [40] and therefore
BR( eee)/BR( e ) 6 103 . Large contributions to BR( eee) may be
obtained in SUSY models with R-parity violation [41].
The e conversion rate in nuclei behaves similarly to the eee branching ratio:
for mW , the log-enhancement in Z and photon penguin diagrams is important and
perhaps dominant, while if > g, the 4 term in the Z-penguin and box contributions
can be significant. In both eee and e conversion in nuclei, the Z-penguin
contribution tends to dominate because of the ln and terms. When this is the case,
the ratio R( e)/BR( eee) does not depend on the unknown ultraviolet details of
the models. We verified numerically that the ratio is almost constant, varying in the range
1013 in the case of e conversion in 27Al, and 2025 in 48 Ti, in a large region of the
parameter space. This feature is a definite prediction of the models under consideration.
Again, the situation is different from SUSY models with slepton flavor mixing, where one
expects R( e)/BR( eee) 1.
Fig. 2 shows the branching ratio of eee and the rate for e conversion in
27Al as a function of |U | for different values of and
(we also fix = 5). All of
e3
the features we have discussed can be readily observed. First, one can clearly see that
R( e in 27Al)/BR( eee) is roughly constant ( 10) for all the depicted values of
and . Second, at small values of |Ue3 | (where the 4 terms are negligible), the effect of
the ln(/mW ) enhancement is visible: at = 1 TeV, BR( e ) > R( e in 27 Al),
while at = 10 TeV the situation is reversed. Finally, the 4 enhancement is also clear if
one compares R( e in 27Al)/BR( e ) and BR( eee)/BR( e ) at small
and large values of |Ue3 |, for large. This behavior is more easily observed for
= 0.003
and = 10 TeV. Indeed, we have verified that at even larger values of , BR( eee)
exceeds BR( e ). Note also that, for all values of
and depicted in Fig. 2, the rate
for e conversion in 27Al is larger than the proposed sensitivity reach of the MECO
experiment [42], even in the limit of small |Ue3 | (assuming the LMA solution to the solar
neutrino puzzle).
Finally, we point out that, in general, the atmospheric and solar contributions to
eee have different CP-violating phases. Furthermore, their relative weight in the
magnetic penguin operator is different from their relative weight in the four-fermion
operators in Eq. (4.2), unless
g. If the atmospheric and solar contributions are
comparable, the interference between them produces observable CP-violating effects in
polarized eee decays [43].
4.4. g 2 of the muon
As derived in Appendix A (see also [44]), the contribution of the KK tower of neutrinos
to the muon anomalous magnetic moment is
a =
m2j
g 2 m2
2
|U
|
109
.
j
32 2 m2W
m2atm
(4.17)
Here we have assumed maximal mixing in the solar sector and neglected Ue3 and
m2sun/m2atm corrections. In the minimal model
> 0, and in general
> 0 (unless
413
Table 1
Bounds and future sensitivity reaches for the
ij coefficients defined by Eq. (1.1)
Present bounds from experiments with
neutrinos
|
e |
|
e |
|
|
ee
< 0.05
< 0.09
< 0.013
< 0.025
Future sensitivity from experiments with
charged leptons
(K ARMEN)
(NOMAD)
(NOMAD)
(reactors)
104
101
101
103
103
102
( e )
( e )
( )
(Z data)
(Z data)
(Z )
neutrinos
104
104
104
103
103
(
(
(
(
(
factory)
factory)
factory)
factory)
factory)
charged leptons
1056
1012
1012
1034
1034
103
( decays)
( e )
( )
(Z data)
(Z data)
(Z )
additional effects are generated by unknown physics around the cut-off scale). Therefore,
the sign of the effect is negative, in contrast to the BNL experimental result [45], that
claims a discrepancy a = +(4.3 1.6) 109 with respect to SM predictions (afflicted,
however, by significant hadronic uncertainties). On the other hand, the present bounds
on the
ij parameters already require in a model-independent way that the right-handed
neutrino correction to g 2 is smaller than the theoretical uncertainty in the hadronic
SM contributions. If, in the future, more precise experimental and theoretical results
establish the presence of a non-SM correction to g 2, extra-dimensional models could
a
a DL)W
/
still account for it by invoking ad-hoc dimension-six operators like 12 (L
with (3 4) TeV. There is no direct contradiction between such nonrenormalizable

operators and bounds from the more sensitive precision data.
4.5. Nonminimal flat models and warped models
Nonminimal flat models and models with warped extra dimensions do not allow one
to relate the coefficients of the dimension-six effective operators to the parameters in
the neutrino Dirac mass matrix. For this reason, it is not possible to make interesting
predictions. In nonminimal flat models (B) and in warped models the naive expectation
is that all
ij are comparable, but is easy to avoid this conclusion. It is useful to estimate
the &i &j decays in terms of the
ij parameters defined in Eq. (1.1). Up to order one
uncomputable factors, &i &j experiments give the following bounds
|
e | 104 ,
|
e |, |
| 101 .
(4.18)
Such bounds prohibit the interpretation of the LSND anomaly [13] as due to
e , as
discussed in the previous section. In fact, the present bound on
e is so strong that it
will be hard to observe its effects in e transitions, even with a neutrino factory.
5. Conclusions
We have shown that the various minimal and nonminimal models with right-handed
neutrinos in flat or warped extra dimension are described at low energy by neutrino
masses plus a specific set of dimension six operators. Up to order one factors (that are
anyhow uncomputable since these models are not renormalizable), flavour conserving
414
effects are described by three positive numbers

ee ,
,
, and flavour-violating effects
are described by three complex numbers
e ,
e ,
, defined in Eq. (1.1). This gives rise
to several relations between non-SM effects in neutrino observables and in lepton flavourviolating processes. Effects in charged leptons are suppressed by a one loop factor with
respect to effects in neutrinos:

P (i j ; L 0)
ij2 ,
(5.1)
BR(&i &j ) |e
ij /4|2 (i = j ).
Present bounds from neutrino experiments and from charged lepton processes are
summarized in Table 1. Due to the loop factor, detectable neutrino effects are compatible
with lepton flavour violating bounds, unlike what is obtained with a generic larger set
of SU(2)L -invariant dimension-six operators, where neutrino and charged lepton effects
both arise at tree level. Values of
ij 104 (including possible CP-violating phases)
will be probed by future neutrino experiments. The importance of eee and e
conversion relative to e can be enhanced, with respect to the usual magnetic-penguin
dominance approximation, if the right-handed neutrino is strongly coupled or if the cut-off
of the theory is significantly larger than the W mass.
All these effects can be generated in four-dimensions by adding right-handed
neutrinos with TeV-scale masses and order one Yukawa couplings, but such choice of
parameters is not motivated by neutrino masses. On the contrary, extra-dimensional models
that try to address the hierarchy problem and to generate neutrino masses give an order-ofmagnitude expectation for the
ij parameters of
ij & (v/ TeV)2 10few , where & is a
loop factor in dimensions. Therefore, these models are compatible with a natural value
for the fundamental scale.
In the flat minimal model, the six
ij parameters are related and can be expressed in
terms of a single unknown
, see Eq. (3.9). The value of
is at present constrained by
LEP and neutrino experiments (
0.004) and by the e decay, see Eqs. (4.10)
and (4.12). Improvements in the sensitivity on rare muon processes and measurements at
a future neutrino factory will significantly extend the probe on the hypothesis of an extradimensional origin of neutrino masses.
Note added
The NuTeV Collaboration [48] has recently reported a 3- anomaly in nucleon
scattering, that can interpreted as due to a ratio between the Z and the W
couplings (0.58 0.21)% lower than in the SM. Since LEP found that the Z couplings
to other fermions agree with SM predictions at the per-mille level, the NuTeV anomaly
could be due to physics beyond the SM that mainly affects the neutrinos. This is what
happens in the extra-dimensional models considered here, where the Z coupling is
1
smaller than in the SM, while the W coupling is smaller by only 1
/2
(see Section 3). The NuTeV anomaly could be fitted by
= 0.0116 0.0042. This value
gives a reduction of the Z width compatible with the LEP measurement, 2- lower
than its SM prediction (see footnote 8). However, the fact that the charged current is also
modified imposes severe constraints (the strongest bound,
+
ee 0.002, comes from
415
a comparison of the muon lifetime with precision electroweak data), which prevent a clean
explanation the NuTeV anomaly.
Acknowledgements
We thank Paolo Criminelli, Riccardo Rattazzi, Serguey Petcov and Alexei Smirnov for
useful discussions.
Appendix A. Integrating out right-handed neutrinos

In this Appendix, we present in detail the expressions for the effective operators that
mediate rare muon processes and the muon anomalous magnetic moment when the entire
tower of KK right-handed neutrinos (up to some arbitrary cut-off) is integrated out. The
computation is done in the minimal flat model, but can be easily generalized to other cases
of interest.
A.1. Magnetic moment type operator for e
The rare decay e takes place at the one-loop level. The following magnetic
moment operator is generated when all one-loop diagrams involving KK neutrinos are
added:
L = m e
(AL PL + AR PR )F + h.c.,
(A.1)
where AL = 0,
AR =
eg 2 1 j j (j ) 2 2
Ve n V n G mn /mW ,
64 2 m2W
(A.2)
n ,j
G (x) =
x(1 6x + 3x 2 + 2x 3 6x 2 ln x)
.
4(1 x)4
(A.3)
Here the summation of all KK modes, which are labeled by the generation j and the
(j )
integer vector n in the -extra dimension, are taken into account. mn is the mass of a
j
j n
KK neutrino labeled by j and n , and Vi n is defined as Vi n Uij VL , where Uij is the

j n
standard active neutrino mixing matrix and VL is the active KK-mixing matrix. When
(j )
(j )2
j n
(j )
mn mj , mn n 2 /R 2 and VL mj /mn . The branching ratio for e is
given by

BR( e )
48 2 |AR |2
3 j j (j ) 2 2 2
(A.4)
=
=
Ve n V n G mn /mW .
BR( ee )
2
G2F
n ,j
The same expressions can be used for l after replacing m m , V n V n ,

j
Ve n Vl n , and BR( ee ) BR( ll ), for l = e or .
416
Next, we sum over all KK modes. In models with large extra dimensions, the sum over
the KK states can be accurately replaced by an integral
|
n
|<R
n
n 2
f
R2

S R

dE E 1 f E 2 ,
(A.5)
where f is any function and S = 2 /2 / ((/2). The cut off is expected to be of order
D . Therefore,
of the fundamental scale M
j j (j ) 2

Ve n V n G mn /m2W
n ,j
Uej Uj
S R

mj 2 2 2
G E /m
dE E 1
W
E
Uej Uj
m2j S R

dE E

1

2/

1 S
M
1
1
4
=
Uej Uj
m2j .
D
D
2 2 M
2 (2)
M
(A.6)
We only consider > 2. Note that we have used the definition of the reduced Planck mass,
Eq. (2.7). The term in the square brackets equals
/m2atm (as defined in Section 2).
A.2. 4-fermion operators for eee and e conversion in nuclei
In addition to the magnetic moment type operator Eq. (A.2), the following 4-fermion
operator contributes to the eee process:

L = e
PL e A + gLe AZ + AB PL + A + gRe AZ PR e + h.c.
(A.7)
f
Here gL(R) = T3L(R) Qf sin2 W (gLe = 1/2 + sin2 W and gRe = sin2 W ). The
coefficients A , AZ , and AB correspond to contributions from photon penguin, Z-penguin,
and box diagrams, respectively. Explicitly,
e2 g 2 1 j j (j ) 2 2
Ve n V n F mn /mW ,
32 2 m2W
n ,j

g 4 1 j j (j ) 2 2
AZ =
Ve n V n FZ mn /mW
64 2 m2W
n ,j
A =
n ,m,j

AB =
(A.8)
(j ) 2

j j n j m

j
(j ) 2
Ve n VL VL Vm GZ mn /m2W , mm /m2W

g 4 1 i i j j (i)2 2
(j ) 2
Ve n V n Vem Vem GB mn /mW , mm /m2W ,
64 2 m2W
n ,m,i,j

(A.9)
(A.10)
x{12 11x 8x 2 + 7x 3 + 2x(x 2 10x + 12) ln x}

,
12(1 x)4
5x(1 x + x ln x)
FZ (x) =
,
(1 x)2
2

1
x (1 y) ln x y 2 (1 x) ln y
GZ (x, y) =
,
xy
1x
1y

xy
1
x 2 ln x
y 2 ln y
1
1
1+
+
GB (x, y) =
xy
4
1 x (1 x)2 1 y (1 y)2

x ln x
y ln y
1
1
+
2xy
.
1 x (1 x)2 1 y (1 y)2
F (x) =
The branching ratio for eee process is
2
2

1
A + gRe AZ + 2 A + gLe AZ + AB
BR( eee) =
2
8GF

m2 11
+ 32|eAR |2 ln 2
me
4

e

+ 4eAR 3A + gR + 2gLe AZ + 2AB + h.c. .
417
(A.11)
(A.12)
(A.13)
(A.14)
(A.15)
In addition to the magnetic moment type operator Eq. (A.2), the following 4-fermion
operator contributes to e conversion in nuclei:

q
q

g + gR
q
L=
(A.16)
AZ + BB PL ,
q
q e
Qq A + L
2
q=u,d
BBd =
2 (j ) 2

g4
1 j j
Ve n V n Vui d GB mn /m2W , m2ui /m2W ,
2
2
128 mW
(A.17)
g 4 1 j j (j ) 2 2
Ve n V n GB mn /mW , 0 .
32 2 m2W
(A.18)
n ,i,j
BBu =
n ,j
The e conversion rate is

R( e) =
4 |F (m2 )|2 m5
3 Zeff
4 2 Z
-capt

u

gL + gRu
u

AZ + BB
(2Z + N)
2
2
d

g + gRd
AZ + BBd Z(2eAR + A ) ,
+ (Z + 2N) L
2
(A.19)
where -capt is the muon capture rate in nuclei of interest [46], Z and N are the
proton and neutron numbers, respectively, F (m2 ) is the nuclear form factor and Zeff
is the nuclear effective charge [47]. Numerically, these nuclear parameters are -capt =
418
1.7 1018 GeV (4.6 1019 GeV), Zeff = 17.6 (11.6), and |F (m2 )| = 0.535 (0.64)
27
for 48
22 Ti (13 Al) [46,47].
Summing over the KK states following the steps previously outlined, we obtain

m2j
e2 g 2 1
2
7 1
2
+
A
Uej Uj
ln 2
,
64 2 m2W
2
m2atm 6 3
mW
(A.20)
AZ
m2j
g4 1
U
U
ej j
64 2 m2W
m2atm
j

m2j
m2j 2
2
2
5
,
ln

J
+
3
2
m2atm m2W
m2atm
m2W
m2j
g4 1
U
U
ej
j
64 2 m2W
m2atm
j

2
2
2

2 mi
2 mi

J
|Uei |
+
|Uei |
1 ,
m2atm m2W
m2atm
i
i

m2j |Vt d |2 m2top
2
g4 1
2
d
,
ln 2
Uej Uj
BB =
64 2 m2W
2
m2atm 8 m2W
mW
j
(A.21)
AB
BBu small,
eAR
(A.22)
(A.23)
(A.24)
e2 g 2 1
64 2 m2W 2

m2j
Uej Uj
,
m2atm
j
(A.25)
where

1 1
2
ln(z/w)
dz dw,
(zw)/21
J 1
2
zw
0 0
which is the number of order 1. Numerically, J3 = 0.23, J4 = 0.43, J5 = 0.55, and

J6 = 0.63.
Note that, unlike the e case, most of the amplitudes depend not only on the
neutrino mass-squared difference (which is directly measured by neutrino oscillation
experiments), but also on the magnitude of the neutrino mass-squared.
A.3. Magnetic moment operator for muon g 2
The anomalous magnetic moment (a ) of the muon is defined as the coefficient of the
effective operator
L=
e
a
F .
4m
(A.26)
419
The contribution to muon anomalous magnetic moment induced by massive KK modes is

given by

g 2 m2 j 2 (j ) 2 2 5
V
f
m
/m
a =
(A.27)
,
W

n
n
16 2 m2W
6
n ,j
f (x) =
10 43x + 78x 2 49x 3 + 4x 4 + 18x 3 ln x

.
12(1 x)4
(A.28)
Note that we have subtracted the SM (W, )-loop in order to calculate the new physics
contribution. It is useful to separate the sum into the massless part (we neglect the effect
j
of the small active neutrino masses) and the massive part. Using the unitarity of VM
and the fact that f (0) = 5/6, we can rewrite
a =
g 2 m2 j 2 (j ) 2 2
V n G mn /mW .
16 2 m2W
(A.29)
n ,j
Here, the function G (x) f (x) 5/6 is defined by Eq. (A.3). Summing over the KK
states (see Eq. (A.6)) we obtain Eq. (4.17).
References
[1] I. Antoniadis, Phys. Lett. B 246 (1990) 377;
J.D. Lykken, Phys. Rev. D 54 (1996) 3693, hep-th/9603133;
N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 429 (1998) 263, hep-ph/9803315.
[2] L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 3370, hep-ph/9905221.
[3] S. Dimopoulos, talk given at the SUSY 1998 Conference.
[4] K.R. Dienes, E. Dudas, T. Gherghetta, Nucl. Phys. B 557 (1999) 25, hep-ph/9811428.
[5] N. Arkani-Hamed et al., hep-ph/9811448.
[6] Y. Grossman, M. Neubert, Phys. Lett. B 474 (2000) 361, hep-ph/9912408.
[7] I. Antoniadis, C. Bachas, Phys. Lett. B 450 (1999) 83, hep-th/9812093.
[8] R. Barbieri, P. Creminelli, A. Strumia, Nucl. Phys. B 585 (2000) 28, hep-ph/0002199.
[9] A. Lukas, P. Ramond, A. Romanino, G.G. Ross, JHEP 104 (2001) 10, hep-ph/0011295;
For earlier extra-dimensional models of the solar anomaly see, e.g.:
G. Dvali, A.Yu. Smirnov, Nucl. Phys. B 563 (1999) 63, hep-ph/9904211;
R.N. Mohapatra, A. Perez-Lorenzana, S.J. Yellin, Phys. Rev. Lett. 87 (2001) 041601, hep-ph/0010353, and
references therein.
[10] SuperKamiokande Collaboration, hep-ex/0105023.
[11] SuperKamiokande Collaboration, Phys. Rev. Lett. 86 (2001) 5651, hep-ex/0103032.
[12] SNO Collaboration, nucl-ex/0106015.
[13] LSND Collaboration, hep-ex/0104049.
[14] A.E. Faraggi, M. Pospelov, Phys. Lett. B 458 (1999) 237, hep-ph/9901299;
R. Kitano, Phys. Lett. B 481 (2000) 39, hep-ph/0002279;
T.P. Cheng, L. Li, Phys. Lett. B 502 (2001) 152, hep-ph/0101068.
[15] A. Ioannisian, A. Pilaftsis, Phys. Rev. D 62 (2000) 066001, hep-ph/9907522.
[16] G.F. Giudice, R. Rattazzi, J.D. Wells, Nucl. Phys. B 544 (1999) 3, hep-ph/9811291;
E.A. Mirabelli, M. Perelstein, M.E. Peskin, Phys. Rev. Lett. 82 (1999) 2236, hep-ph/9811337;
T. Han, J.D. Lykken, R. Zhang, Phys. Rev. D 59 (1999) 105006, hep-ph/9811350.
[17] M. Gogberashvili, hep-ph/9812296.
[18] W.D. Goldberger, M.B. Wise, Phys. Rev. D 60 (1999) 107505, hep-ph/9907218.
420
[19] J. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231, hep-th/9711200;

E. Witten, Talk at ITP Conference New Dimensions in Field Theory and String Theory, www.
itp.ucsb.edu/online/susy_c99/discussion.
[20] C. Csaki, J. Erlich, C. Grojean, Nucl. Phys. B 604 (2001) 312, hep-th/0012143.
[21] J. Maxwell, A Treatise on Electricity and Magnetism, 1873.
[22] L. Wolfenstein, Phys. Rev. D 17 (1978) 2369;
S.P. Mikheyev, A. Yu Smirnov, Sov. J. Nucl. Phys. 42 (1985) 913;
P. Langacker, S.T. Petcov, G. Steigman, S. Toshev, Nucl. Phys. B 282 (1987) 589.
[23] G.L. Fogli, E. Lisi, A. Marrone, G. Scioscia, Phys. Rev. D 60 (1999) 53006, hep-ph/9904248.
[24] K. Agashe, N.G. Deshpande, G.H. Wu, Phys. Lett. B 489 (2000) 367, hep-ph/0006122.
[25] N OMAD Collaboration, hep-ex/0106102;
See also: C HORUS Collaboration, Phys. Lett. B 497 (2001) 8.
[26] K ARMEN Collaboration, Nucl. Phys. Proc. Suppl. 91 (2000) 191, hep-ex/0008002.
[27] C HOOZ Collaboration, Phys. Lett. B 466 (1999) 415, hep-ex/9907037;
See also: PALO V ERDE Collaboration, Phys. Rev. Lett. 84 (2000) 3764, hep-ex/9912050.
[28] Y. Declais et al., Nucl. Phys. B 434 (1995) 503.
[29] LEP and SLD Collaborations, hep-ex/0103048.
[30] For a review on decays see, e.g., A. Pich, Nucl. Phys. Proc. Suppl. 98 (2001) 385, hep-ph/0012297. See
also [15].
[31] M. Mangano et al., nu-DIS working group of the ECFA-CERN neutrino factory study group, hepph/0105155.
[32] Y. Grossman, Phys. Lett. B 359 (1995) 141, hep-ph/9507344;
M.C. Gonzalez-Garcia, Y. Grossman, A. Gusso, Y. Nir, hep-ph/0105159, and references therein.
[33] F.J. Botella, C.S. Lim, W.J. Marciano, Phys. Rev. D 35 (1987) 896.
[34] A.M. Gago, M.M. Guzzo, H. Nunokawa, W.J. Teves, R.Z. Funchal, hep-ph/0105196.
[35] C. Lunardini, A. Yu Smirnov, Phys. Rev. D 63 (2001) 073009, hep-ph/0009356.
[36] G. Barenboim, G.C. Branco, A. de Gouva, M.N. Rebelo, hep-ph/0104312.
[37] F. Vissani, D. Delepine, hep-ph/01006287.
[38] L. Barkov et al., Research proposal to PSI, meg.icepp.s-tokyo.ac.jp.
[39] For global post-SNO analyses of the solar neutrino data see:
G.L. Fogli, E. Lisi, D. Montanino, A. Palazzo, hep-ph/0106247;
A. Bandyopadhyay, S. Choubey, S. Goswami, K. Kar, hep-ph/0106264;
J.N. Bahcall, M.C. Gonzalez-Garcia, C. Pea-Garay, hep-ph/0106258;
The addendum to: P. Creminelli, G. Signorelli, A. Strumia, hep-ph/0102234.
[40] Y. Kuno, Y. Okada, Rev. Mod. Phys. 73 (2001) 151, hep-ph/9909265, and references therein.
[41] A. de Gouva, S. Lola, K. Tobe, Phys. Rev. D 63 (2001) 035004, hep-ph/0008085.
[42] MECO Collaboration, BNL proposal AGS P940, meco.ps.uci.edu.
[43] Y. Okada, K. Okumura, Y. Shimizu, Phys. Rev. D 58 (1998) 51901, hep-ph/9708446.
[44] G.C. McLaughlin, J.N. Ng, Phys. Lett. B 493 (2000) 88, hep-ph/0008209.
[45] Muon g 2 Collaboration, Phys. Rev. Lett. 86 (2001) 2227, hep-ex/0102017.
[46] T. Suzuki, D.F. Measday, J.P. Roalsvig, Phys. Rev. C 35 (1987) 2212.
[47] H.C. Chiang et al., Nucl. Phys. A 559 (1993) 526;
T.S. Kosmas et al., Phys. Rev. C 56 (1997) 526, nucl-th/9704021.
[48] NuTeV Collaboration, hep-ex/0110059;
See also the slides of the 26 October 2001 seminar at FNAL, www.pas.rochester.edu/~ksmcf/NuTeV.

Loitering phase in brane gas cosmology

Robert Brandenberger a,1 , Damien A. Easson a,b , Dagny Kimberly a
a Department of Physics, Brown University, Providence, RI 02912, USA
b Physics Department, McGill University, Montral, QC, Canada
Abstract
Brane gas cosmology (BGC) is an approach to M-theory cosmology in which the initial state of the
Universe is taken to be small, dense and hot, with all fundamental degrees of freedom near thermal
equilibrium. Such a starting point is in close analogy with the standard big bang (SBB) model.
The topology of the Universe is assumed to be toroidal in all nine spatial dimensions and is filled
with a gas of p-branes. The dynamics of winding modes allow, at most, three spatial dimensions to
become large, thus explaining the origin of our macroscopic (3 + 1)-dimensional Universe. Here we
conduct a detailed analysis of the loitering phase of BGC. We do so by including into the equations
of motion that describe the dilaton gravity background some new equations which determine the
annihilation of string winding modes into string loops. Specific solutions are found within the model
that exhibit loitering, i.e., the Universe experiences a short phase of slow contraction during which
the Hubble radius grows larger than the physical extent of the Universe. As a result the brane problem
(generalized domain wall problem) in BGC is solved. The initial singularity and horizon problems of
the SBB scenario are solved without relying on an inflationary phase. 2002 Elsevier Science B.V.
All rights reserved.
PACS: 04.50.+h; 98.80.Bp; 98.80.Cq
1. Motivation and introduction

The necessity to search for alternatives to the standard big bang (SBB) scenario is
driven by a significant number of problems within the theory such as the horizon, flatness,
structure formation and cosmological constant problems. Although inflationary models
E-mail addresses: rhb@het.brown.edu (R. Brandenberger), easson@het.brown.edu (D.A. Easson).
1 Address from 15 September 2001 to 15 March 2002: Theory Division, CERN, CH-1211 Geneva,
Switzerland.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 3 6 - 8
422
R. Brandenberger et al. / Nuclear Physics B 623 (2002) 421436
have managed to address many of these issues, all current formulations remain incomplete.
In particular, the present models of inflation suffer from the fluctuation, super-Planck
scale physics, initial singularity and cosmological constant problems, as discussed in [1].
Furthermore, several theorems have appeared which show that de Sitter spacetime, the
simplest example of an inflationary Universe, cannot be a classical solution to supergravity
theories [2,3]. As supergravity is the low energy limit of M-theory, it does not seem clear
that a pure de Sitter model of inflation will arise naturally within the theory.
M-theory is currently our best candidate for a quantum theory of gravity. As such, the
theory should provide the correct description of physics in regions of space with high
energies and large curvature scales similar to those found in the initial conditions of the
Universe. Therefore, it is only natural to incorporate string and M-theory into models of
cosmology.2 It is our hope that M-theory will provide answers to the pending questions
of cosmology while preserving the triumphs of the SBB model ultimately leading to a
complete description of the Universe.
Brane gas cosmology (BGC) which is presented in [6] and based on the earlier
work of [7] is one example of a cosmological scenario motivated by M-theory. The BGC
model is relatively simple and starts out in close analogy with the standard big bang
scenario. According to [6], the initial state of the Universe is small, dense, hot and with all
fundamental degrees of freedom in approximate thermal equilibrium.3 For simplicity, the
background spatial geometry is assumed to be toroidal, and the Universe is filled with a
hot gas of p-branes, the fundamental objects appearing in string theories.
These branes may wrap around the cycles of the torus (winding modes), they can have a
center-of-mass motion along the cycles (momentum modes) or they may simply fluctuate
in the bulk space (oscillatory modes). By symmetry, we assume equal numbers of winding
and anti-winding modes. As the Universe tries to expand, the winding modes become
heavy and halt the expansion [8]. Spatial dimensions can only dynamically decompactify
if the winding modes can disappear, and this is only possible (for string winding modes)
in 3 + 1 dimensions [7]. Thus, BGC may provide an explanation for the observed number
of large spatial dimensions. However, by causality at least one winding mode per Hubble
volume will be left behind, leading to the brane problem for BGC [6], a problem analogous
to the domain wall problem of standard cosmology.
The present paper provides a simple solution to the brane problem: the winding modes
will halt the expansion of the spatial sections, and lead to a phase of slight contraction
(loitering [9,10]) during which the Hubble radius becomes larger than the spatial sections
and hence all remaining winding modes can annihilate in the large 3 + 1 dimensions. We
supplement the equations for the dilaton gravity background of BGC [8,11] by equations
which describe the annihilation of string winding modes into string loops. A study of these
equations demonstrates that solutions exist in which the winding modes force the Universe
to contract for a short time and enter a loitering phase. During the phase of contraction
2 For a review of some string theory motivated cosmological models see [4]. A recent proposal, not covered
in the review, is the Ekpyrotic scenario introduced in [5].
3 Note that mathematically it is not possible for the Universe to be in thermal equilibrium as the FRW
cosmological model does not possess a time-like Killing vector field. However, it is true that the Universe has
been very nearly in thermal equilibrium. Obviously, the departures from equilibrium make things interesting!
423
the number density of the remaining winding modes increases and the winding and antiwinding modes begin to annihilate. The winding branes appear as solitons (analogous to
cosmic strings) in the bulk space. The annihilation of winding and anti-winding modes
(analogous to cosmic string intersections) leads to the production of string loops which
have the same equation of state as cold matter.
Brane gas cosmology is a simple non-singular model which addresses some of the
problems of the SBB scenario, simultaneously providing a dynamical resolution to the
dimensionality problem of string theory. In this respect it is very different from other
attempts to incorporate M-theory into cosmology, such as brane world scenarios in
which our Universe is located on the world volume of a 3-brane. In our opinion, brane
world models suffer from a lack of cosmological motivation. All current models require
that the extra dimensions be compactified by hand. Although this is a serious problem from
a cosmological viewpoint it is rarely, if ever, discussed.
The organization of this paper is as follows. We begin with a brief review of the brane
gas model in Section 2. Our concrete starting point is presented, followed by a derivation
of the equation of state for a gas of branes and an analysis of the background dynamics.
This is followed (in Section 3) by a discussion of the dilaton gravity equations in the
presence of a brane gas, of the benefits of loitering, and of attractor solutions. In Section 4,
we supplement the system of equations with equations which describe the annihilation of
string winding modes into string loops, and based on this we provide a detailed analysis of
a loitering solution. We use both numerical and analytical methods to study the solutions.
We conclude, in Section 5, with a brief summary and a few conjectures concerning
supersymmetry breaking, the effective breaking of T-duality, and dilaton mass generation
in the late Universe.
2. Brane gases
The brane gas scenario of [6] is formulated within the context of eleven-dimensional
M-theory compactified on S 1 . This leads to ten-dimensional, type-IIA string theory, whose
low-energy effective action is that of supersymmetrized dilaton gravity.
Since M-theory admits the graviton, 2-branes and 5-branes as fundamental degrees
of freedom, the compactification on S 1 leads to 0-branes, strings (1-branes), 2-branes,
4-branes, 5-branes, 6-branes and 8-branes in the ten-dimensional Universe. The remaining
nine spatial dimensions are assumed to be toroidal (of radius R). The Universe starts out
hot, dense and near thermal equilibrium. It is filled with a gas of all the branes (wrapped
and unwrapped) which appear in the spectrum of the theory.
2.1. Brane gas equation of state
In this section we derive the equation of state of the brane gas described above. The gas
consists of all the branes of spatial dimension p. Contributions of the winding, momentum
and oscillatory modes are treated separately below.
424
The total action for the above model is the sum of the bulk effective action and the
action of all the branes in the gas. The low-energy bulk effective action is

1
1
10
2
Sbulk = 2 d x G e
(1)
R + 4G H H
,
12
2
where G is the determinant of the background metric G , is the dilaton, H denotes
the field strength corresponding to the bulk antisymmetric tensor field B , and is
determined by the ten-dimensional Newton constant.
Fluctuations of each of the p-branes are described by the DiracBornInfeld (DBI)
action [13] and are coupled to the ten-dimensional action via delta function sources. The
DBI action is

Sp = Tp d p+1 e det gmn + bmn + 2 Fmn ,
(2)
where Tp is the tension of the brane, gmn is the induced metric on the brane, bmn is the
induced antisymmetric tensor field, and Fmn the field strength tensor of gauge fields Am
living on the brane. The constant lst2 is given by the string length scale lst .
The induced metric on the brane gmn (with indices m, n, . . . , denoting spacetime
dimensions parallel to the brane), is determined by the background metric G and by
scalar fields i (not to be confused with the dilaton ) living on the brane. The indices
i, j, . . . , denote dimensions transverse to the brane and the i describe the fluctuations
of the brane in the transverse directions. In the string frame, the tension of a p-brane,
appearing in front of the action (2), is given by
2 (p+1)/2
4
Tp =
(3)
,
gs
where gs is the string coupling constant. We assume small string coupling so that the
fluctuations of the branes will be small. We choose to work with conformal time in a
background metric of the form
G = a()2 diag(1, 1, . . ., 1),
(4)
where a() is the cosmological scale factor.

Assuming that the transverse fluctuations of the brane and the gauge fields on the brane
are small, it is possible to expand the brane action as

1
2

Sp = Tp d p+1 a()p+1e e 2 tr log(1+m i n i +a() 2 Fmn )

1
= Tp d p+1 a()p+1e 1 + (m i )2 2 2 a 4 Fmn F mn .
(5)
2
The first term inside the parentheses in the last line represents the brane winding modes, the
second term corresponds to the transverse fluctuations, and the third term relates to brane
matter. In the low-energy limit, the transverse fluctuations of the brane are described by a
free scalar field action, and the longitudinal fluctuations are given by a YangMills theory.
The induced equation of state which describes the second and third terms has pressure
p 0.
425
We are now ready to compute the equation of state for the brane gases for various p.
There are three types of modes that we will need to consider. First, there are the winding
modes. The background space is T 9 , and hence a p-brane can wrap around any set of
p toroidal directions. These modes are related by T-duality to the momentum modes
corresponding to the center of mass motion of the branes. Finally, the modes corresponding
to fluctuations of the branes in the transverse directions are (in the low-energy limit)
described by the scalar fields on the brane, i . There are also bulk matter fields and brane
matter fields.
Let us begin by considering the winding modes. From Eq. (5), one can compute the
equation of state for a winding p-brane:
p
(6)
with wp = ,
d
where d is the number of spatial dimensions, and p and represent the pressure and energy
density, respectively.
Both fluctuations of the branes and brane matter are described by free scalar and gauge
fields living on the brane. These can be viewed as particles in the directions transverse to
the brane and thus the equation of state is just that of ordinary matter:
p = wp ,
p = w,
with 0 w 1.
(7)
From the action (5) we see that the energy in the winding modes will be
Ep (a) Tp a()p ,
(8)
where the constant of proportionality is dependent on the number of branes. Note that the
energy in the winding modes increases with the expansion of the Universe in contrast to
the energy of the brane fluctuations and brane matter.
2.2. Background equations of motion
The background equations of motion are calculated from the variation of (1) with the
metric (4) and were derived (in the absence of H ) in [8] (see also [11]). It is convenient to
define

(t) = log a(t) ,
(9)
and to use the shifted dilaton
= 2 d,
(10)
which absorbs the space volume factor. Assuming for simplicity the isotropic case (all
ai = a and therefore i = ) the background equations are
d 2 + 2 = e E,
1
= e P ,
2
1
d 2 = e E,
2
(11)
(12)
(13)
426
where E and P denote the total energy and pressure, respectively. Both sources E and P
are made up from contributions of all the branes in the gas. We have already calculated
these contributions in the previous section:

E=
(14)
Epw + E nw ,
P=
wp Epw + wE nw ,
p
where the superscripts w and nw stand for the winding modes and the non-winding modes,
respectively. Here the contributions of the non-winding modes of all branes are combined
into one term. The constants wp and w are given by (6) and (7). Each Epw is the sum of the
energies of all of the winding branes with dimension p.
2.3. Brane gas cosmology
We now turn our attention to the model of brane gas cosmology proposed in [6]. The
first important aspect of BGC is that it is free of the initial cosmological singularity of the
SBB model. This can be explained by the T-duality symmetry of string theory as was first
demonstrated in [7]. As the Universe contracts, the T-duality self-dual fixed point (R = 1)
is reached at some temperature T less than the limiting Hagedorn temperature. As the
background continues to contract, the temperature decreases according to the T-duality
equation for temperature,

1
T
(15)
= T (R).
R
Thus, there is no physical singularity as R approaches 0.
The T-duality symmetry of the spectrum of states was analyzed in [7] in the absence
of branes. With only fundamental string states, T-duality interchanges the winding and
momentum quantum numbers of the same object. The action of T-duality on branes is more
complicated [12]. A T-duality parallel to a p-brane produces a (p 1)-brane, while a Tduality perpendicular to a p-brane yields a (p + 1)-brane. Hence, after T-dualizing in all d
spatial dimensions, a p-brane becomes a (d p)-brane. The p winding modes of the initial
p-brane are heavy, and the (d p) transverse momentum modes are light (considering as
starting point R 1). After T-dualizing, the p transverse momentum modes of the (d p)brane are heavy, and the (d p) winding modes are light (since now the radius of the torus
is R 1). We conclude that the spectrum of states respects T-duality in the presence of
branes.4
Let us now examine the de-compactification mechanism which leads to three large
spatial dimensions [6,7]. Recall, that our starting point is a Universe with nine spatial
dimensions all equal and near the self-dual point, R = 1. The Universe is in thermal
equilibrium, and by symmetry we argue that there are equal numbers of winding and antiwinding modes.
As the Universe begins to expand (symmetrically in all directions), increases and the
total energy in the winding modes increases according to Eq. (8). Note that the energy in
4 We thank S. Alexander and F. Quevedo for key discussions on this point.
427
the winding modes of the branes with the largest value of p increases the fastest. Therefore
branes with the largest value of p will have an important effect first. As we have already
explained, the winding modes will prevent further expansion until they have annihilated.
A simple counting argument demonstrates that the world-volumes of two p-branes
will probably intersect in at most 2p + 1 spatial dimensions.5 Clearly, in d = 9 spatial
dimensions the winding branes with p = 8, 6, 5 and 4 will have no problems interacting
and self-annihilating. However, branes with smaller values of p will allow a hierarchy of
dimensions to expand. Since the energy of the branes with the largest value of p is greatest,
the 2-branes will first allow a 5-dimensional torus to expand. Within this T 5 , the strings
(1-branes) will allow a T 3 subspace to become large. Hence, this mechanism provides a
solution to the dimensionality problem of string theory and may explain the origin of our
large (3 + 1)-dimensional Universe.6
An important unsolved issue in brane gas cosmology is the stability of the radius
of the compact dimensions to inhomogeneities as a function of the three coordinates
xi ; i = 1, 2, 3 corresponding to the large spatial dimensions. The separation in xi between
the branes wrapping the small tori is increasing, and there appears to be no mechanism
to keep the internal dimensions from expanding (inhomogeneously in xi ) between the
branes. A simple but unsatisfactory solution, is to invoke a non-perturbative effect similar
to what is needed to stabilize the dilaton at late times, namely to postulate a potential
which will stabilize the moduli uniformly in xi (after the winding modes around the
xi directions have disappeared). Work on a possible solution within the context of the
framework presented here is in progress.7
3. A loitering Universe
Despite the ability of winding modes to self-annihilate in a distinguished number of
dimensions, causality demands that there is at least one winding mode per Hubble volume
remaining [14]. In our macroscopic four-dimensional spacetime, the wrapped branes with
p 2 will appear as domain walls. The presence of such topological defects would result
in a cosmological disaster, as even one wall per Hubble volume today will overclose the
Universe if the tension of the brane is larger than the electroweak scale.8 This domain
wall problem [15] is common in cosmological scenarios based on quantum field theories
which admit domain wall solutions.
As a solution of this brane problem in BGC the authors of [6] suggested a phase of
cosmological loitering. If at some stage in the evolution of the Universe the Hubble radius
5 For example, consider two particles (0-branes) moving through a space of dimension d. These particles will
definitely interact (assuming the space is periodic) if d = 1, whereas they probably will not find each other in a
space with d > 1.
6 By large we mean large compared to the string scale. Without inflation we are not able to solve the entropy
problem of standard cosmology, namely to produce a Universe large enough to contain our present observed
Universe.
7 We thank D. Lowe for discussions on this issue.
8 Even strings have large tensions at low temperatures and will therefore contribute too much to the energy
density of the Universe.
428
becomes larger than the spatial extent of the Universe, there is no causal obstruction for all
winding modes to annihilate. Loitering allows for previously causally disconnected regions
of the Universe to communicate, and therefore provides a solution to the horizon problem
of the SBB model.
Within the context of dilaton gravity, cosmological solutions which exhibit a loitering
phase appear rather naturally due to the presence of winding modes. To see this, we will
examine the phase space of solutions to the background Eqs. (11)(13). The phase space
of solutions for general p was discussed in [8], and a numerical plot of the full phase space
is given below.
By defining l = and f = ,
the background EOM equations (11)(13) simplify to two
first order differential equations:
pf 2
pl 2
+ lf
,
l =
(16)
2
2d
dl 2 f 2
+
.
f =
(17)
2
2
Notice that for positive energy density E, Eq. (11) implies that will never change sign.
We are interested in studying the initial conditions with < 0. If 0 the boosting effect
of the dilaton on will invalidate the adiabatic approximation used in the derivation of
the EOM. Furthermore, growing together with expanding implies the growth of the
effective coupling exp() in contradiction with a weak coupling assumption [8].
We will therefore consider solutions to the background EOM with initial conditions
corresponding to an expanding Universe > 0 with < 0. Note that positivity of E
imposes another restriction (see Eq. (11)):
1
|l| < |f |.
(18)
d
Solutions with initial conditions described above are driven towards the l = 0 line in the
f vs. l phase space at a finite value of f (see Fig. 1).
There are three special lines in the phase space which correspond to straight line
trajectories and pass through the origin:
1 p
l
l
= = , .
f f
d d
(19)
Notice that the line l/f = 1/ d is a repeller and solutions which start out in the
energetically allowed region are pushed away from the line, cross the f -axis and then
approach the attractor line l/f = p/d to the origin. Near the l = 0 line, Eqs. (16) and (17)
may be approximated by
f2
p
.
f
l f 2 ,
(20)
2d
2
When a solution crosses the f -axis, l changes sign. This means that the Universe begins to
contract. Since f 0 only as t the solution never crosses the l-axis.
Note that this analysis assumes that the winding modes are not decaying into loops. The
above provides an accurate description of the early Universe, before winding modes have
429
Fig. 1. Phase space trajectories of the solutions of the background equations (11)(13) for the values p = 2 and
d = 9. The energetically allowed
region lies near the l = 0 axis between the special lines a and c, which are
the lines given by l/f = 1/ d. The trajectory followed in the scenario investigated in this paper starts out in
the upper left quadrant close to the special line c (corresponding to an expanding background), crosses the l = 0
axis at some finite value of f (at this point entering a contracting phase), and then approaches the loitering point
(l, f ) = (0, 0) along the phase space line b which corresponds to l/f = p/d.
self-annihilated. We will examine the case of loop production and the late time evolution
of the Universe in Section 4.
4. Unwinding and loop production

We now wish to extend the analysis presented in Section 3 in order to study the late time
behavior of the Universe, i.e., to include the effects of winding mode annihilation and loop
production.
Recall that after the winding modes have annihilated, a three-dimensional subspace will
grow large. In what follows we will therefore take d = 3. The strings in the theory are the
last branes to unwind which implies at late times that we should consider the case of p = 1.
When the winding strings self-annihilate they create loops in the (3 + 1)-dimensional
Universe.
We now set up the equations describing the unwinding and corresponding loop
production. They are analogous to the corresponding equations for cosmic strings in an
expanding Universe. First, note that the energy density w in winding strings can be
expressed in terms of the string tension and the number (t) of winding modes per
430
Hubble volume t 3 as
w (t) = (t)t 2 .
(21)
Since loops are produced by the intersection of two winding strings, the rate of loop
production is proportional to 2 :
dn(t)
= c (t)2 t 4 ,
(22)
dt
where n(t) is the number density of loops and c is a proportionality constant expected
to be of the order 1. The energy density in the winding modes decreases both due to the
expansion of space and due to the decay into loops:
dw (t)
dn(t)
(23)
+ 2lw (t) = c t
= cc (t)2 t 3 ,
dt
dt
where c is a constant which relates the mean radius R = c t of a string loop to its
length. Without loop production (c = 0), the energy density w redshifts corresponding
to the equation of state p = 13 . This explains the coefficient of the Hubble damping
term in (23).9 Inserting Eq. (21) into the energy conservation equation (23), we obtain an
equation for (t):

d (t)
= 2 t 1 l cc t 1 2 .
(24)
dt
In addition to w (t), we will also require information about the energy density in loops,
l (t). The energy density in loops obeys the conservation equation
dl (t)
+ 3ll (t) = cc (t)2 t 3 .
(25)
dt
As in Eq. (23), the second term on the left hand side of the equation represents the decrease
in the density due to Hubble expansion, with the coefficient reflecting the equation of state
p = 0 of a gas of static loops, and the term on the right hand side representing the energy
transfer from winding modes to loops. Without loop production, l (t) would scale as
l (t) = g(t)e3((t )0) ,
(26)
with g(t) constant. Here 0 = (t0 ), where t0 is some initial time, and g(t) is a function
which obeys the equation
dg(t)
= cc t 3 2 e3((t )0) .
(27)
dt
Using the expressions for w (t) and l (t) as sources for the energy density E and
pressure P in Eqs. (11)(13) we can obtain background equations analogous to Eqs. (16)
and (17):
1
1
1
l = lf + l 2 f 2 + ge+30 ,
(28)
2
6
6
3
1
f = f 2 + l 2 .
(29)
2
2
9 For more on strings in an expanding Universe see, e.g., [16].
431
(Recall that p = 1 and d = 3 in the above equations.)

Eqs. (24), (27), (28) and (29), along with the equations l = and f = provide six firstorder differential equations which fully describe the Universe during the process of loop
production. These provide us with the initial conditions required for a numerical analysis.
Note that the c = 0 case corresponds to no loop production and the background equations
reduce to the previous Eqs. (16) and (17).
Fig. 2 demonstrates the behavior of a typical numerical solution to the EOM having
initial conditions in the energetically allowed region of the phase space, taking into account
the effects of loop production. Recall that when no loops are produced (see Fig. 1) the
solutions of interest cross the l = 0 line only once and approach the origin of the phase
space as t . When the decay of the winding modes is taken into account, the solutions
are pushed back over the l = 0 line as in Fig. 2.
In more detail, the dynamics of our loitering solution is depicted in Figs. 35 and 6.
Fig. 3 shows the time evolution of the Hubble expansion rate H = l. Note that since we
Fig. 2. A solution of the background equations (28) and (29) including the effects of loop production. This depicts
a typical solution which starts in the energetically allowed region of the phase space. The solution crosses the l = 0
axis at some finite value of f (at which point the Universe enters a contracting phase), and then crosses the l = 0
line a second time when the winding modes have fully annihilated. At this point the Universe begins to expand,
is matter dominated and the dilaton is assumed to become massive.
Fig. 3. The time evolution of H = l. The loitering phase begins when l(t) crosses the l = 0 line for the first time
and ends when l(t) crosses back over the l = 0 line.
432
Fig. 4. The time evolution of the scale factor a. By comparing this plot with Fig. 3 we see that the loitering phase
lasts long enough to allow all winding modes to self-annihilate in the large three-dimensional Universe.
Fig. 5. Time evolution of .

Initially, increases as the Universe contracts. The winding modes begin to
self-annihilate ( decreases) and eventually vanish ( 0).
have set Newtons constant G = 1 in our background equations, time is measured in Planck
time units. By comparing with the value of a(t) from Fig. 4, we see that
H 1 (t) a(t),
(30)
during the loitering phase. Keeping in mind that the initial spatial size of the tori is Planck
scale, it follows immediately from Eq. (30) and from the time duration of the loitering
phase that loitering lasts sufficiently long to allow causal communication over the entire
spatial section. This is reflected in Fig. 5 which shows that the winding modes completely
annihilate by the end of the loitering phase, after which g(t) tends to a constant (Fig. 6).
Our modified picture is as follows: the Universe begins to expand until the winding
modes become too massive and force the expansion to stop and contraction to begin. This
corresponds to the solution in Fig. 2 crossing over the l = 0 axis and signals the beginning
of the loitering phase. The loitering phase ends when the solution is pushed back over the
l = 0 line due to the decay of the winding modes. From this point on the Universe begins
to expand again.
433
Fig. 6. Time evolution of g. Note that g goes to a constant when goes to zero.
Soon after the solutions cross the l = 0 line for the second time, our equations become
singular. Soon after the winding modes vanish ( (t) 0), all the fields l, f , and
in our equations of motion blow up. This singularity does not concern us however
since it can be eliminated by the introduction of a simple potential V () used to freeze
the dilaton at the moment loop production is exhausted. The massless dilaton does not
appear in nature and therefore such a potential is required in any string theory-motivated
cosmological model at late times. The precise mechanism responsible for dilaton mass
generation is unknown, although it is often suspected that this mechanism will coincide
with the breaking of supersymmetry. From this analysis we are lead to the conjecture that
dilaton mass generation may coincide with the elimination of winding modes. We will
comment more on this below.
For the time being, let us assume that the dilaton has frozen at the value C = (tfreeze )
and therefore = = 0. We will also assume that this occurs when the winding modes
have vanished, (t) 0 and hence the number of loops has reached a constant so that
g(t) Cg . Now the EOM simplify greatly. By fixing the dilaton in Eqs. (28) and (29) we
can derive an equation for the scale factor (after shifting back to the true dilaton ):
a C a 1/2 = 0,
where C is a constant given by
Cg C + 3 0
2
e
C =
.
12
The most general solution to Eq. (31) is

(1/3)
3C (2/3) 2
a(t) =
t 2Ct + C 2
,
2
(31)
(32)
(33)
where C is an integration constant. For the value C = 0 or for large values of t (late times),
the scale factor grows as
a(t) t 2/3 ,
which is exactly the behavior for a matter dominated Universe.
(34)
434
Our interpretation is that the winding modes look like solitons in the (3 + 1)-dimensional Universe. The self-annihilation of these winding modes corresponds to the creation
of matter in the Universe and the scale factor evolves appropriately. In our equations,
the loops are modelled as static. In reality, the loops will oscillate and decay by emitting
(mostly) gravitational radiation, thus producing a radiation dominated Universe.
Let us return to the issue of SUSY breaking and dilaton mass generation. One thing
which appears inevitable within the context of this model is the spontaneous breaking of
T-duality in the large four-dimensional Universe. This is most easily understood once all of
the winding modes have self-annihilated since it is impossible to create new ones. It would
cost too much energy for a brane to wrap around the large dimensions. Thus, the state of the
system is not symmetric under T-duality, and in the absence of string winding modes and
for fixed dilaton, our background equations reduce to those of Einsteins general relativity
which do not exhibit the R 1/R symmetry of string theory.
It is also interesting to note that there seems to be a relation between the amount
of supersymmetry in a theory and the presence of T-duality. Using a specific example
in [17], Aspinwall and Plesser show that T-duality can be broken by non-perturbative
effects in string coupling. Furthermore, a holonomy argument is given to show that Tdualities should only be expected when large amounts of supersymmetry are present. It
seems very likely that the dynamics in the BGC scenario will cause SUSY to break. This
result seems to be in agreement with the possibility of dilaton mediated SUSY breaking
occurring simultaneously with the breaking of T-duality.
Considering the above evidence we are inclined to hypothesize about the possible
relations between supersymmetry breaking and the breaking of T-duality, as well as dilaton
mass generation and the vanishing of winding modes.
5. Conclusions and speculations

In this paper we have conducted a detailed analysis of a loitering phase in the model
of brane gas cosmology presented in [6]. The concrete starting point of BGC is Mtheory compactified on S 1 which gives ten-dimensional, type-IIA string theory. We assume
toroidal topology in all nine spatial dimensions. The initial conditions for the Universe
include a small, hot, dense gas of the p-branes in the theory. These fundamental degrees
of freedom are assumed to be in thermal equilibrium. T-duality ensures that the initial
singularity of the SBB model is not present in this scenario.
We compute the equation of state for the brane gas system and the background equations
of motion. We study the solutions which initiate in the energetically allowed region of the
phase space. The Universe tries to expand until winding modes force the expansion to stop
and a phase of slow contraction (loitering) to begin. Loitering provides a solution to the
brane problem first discussed in [6]. It also provides a solution to the horizon problem of
the SBB model without requiring an inflationary phase.10
10 Here we do not address the structure formation problem or the flatness problem of the SBB model. Both
of these are solved by an inflationary phase. It is likely that the brane gas scenario will require something like
inflation in order to produce a Universe which is large enough to contain the known Hubble radius.
435
The counting argument of [6] demonstrates that winding modes will allow a hierarchy
of dimensions in the T 9 to grow large. When the string winding modes self-annihilate we
are left with a large T 3 subspace, simultaneously explaining the origin of our (3 + 1)-dimensional Universe and solving the dimensionality problem of string theory.
Branes wrapped around the cycles of the torus appear as solitons in the early Universe.
They are topological defects (domain walls for p 2). When the winding modes and antiwinding modes self-annihilate, matter is produced and the Universe begins to expand again.
We hypothesize that winding mode annihilation corresponds to dilaton mass generation.
We also believe there may be a relation between SUSY breaking and the breaking of
T-duality, although we cannot provide any direct evidence for this. Once winding states
have vanished, we cannot map momentum modes into winding modes via T-duality. The
breaking of T-duality requires further study.
Particle phenomenology demands compactification on manifold with non-trivial
holonomy such as CalabiYau three-folds if the four-dimensional low energy effective
theory is to have N = 1 supersymmetry. The initial steps towards generalizing the brane
gas model to manifolds of non-trivial homology are discussed in [18]. However, in the
context of early Universe cosmology it is not reasonable to require N = 1 supersymmetry. Toroidal backgrounds, such as the one considered here, are compatible with maximal
supersymmetry.
Brane gas cosmology provides a method of incorporating string and M-theory into
cosmology which is an alternative to popular brane world scenarios. In our opinion,
BGC has the advantage over brane world scenarios in that its foundations are analogous
to those of the standard big bang model. In the BGC model the Universe starts out small,
hot and dense, with no initial singularity. All the current versions of brane world scenarios
embedded in string theories rely on the compactification of the extra dimensions by hand.
In our opinion this is a considerable problem which is often overlooked. A dynamical
mechanism in BGC leads naturally to four large spacetime dimensions.
Acknowledgements
We are thankful to S. Alexander, D. Lowe and F. Quevedo for important discussions.
R.B. was supported in part by the US Department of Energy under Contract DE-FG0291ER40688, TASK A. D.E. was supported in part by the US Department of Education
under the GAANN program.
References
[1]
[2]
[3]
[4]
R.H. Brandenberger, Inflationary cosmology: progress and problems, hep-ph/9910410.

K. Bautier, S. Deser, M. Henneaux, D. Seminara, Phys. Lett. B 406 (1997) 49, hep-th/9704131.
J. Maldacena, C. Nunez, Int. J. Mod. Phys. A 16 (2001) 822, hep-th/0007018.
D.A. Easson, The interface of cosmology with string and M(illennium) theory, Int. J. Mod. Phys. A, in press,
hep-th/0003086.
[5] J. Khoury, B.A. Ovrut, P.J. Steinhardt, N. Turok, The ekpyrotic universe: colliding branes and the origin of
the hot big bang, hep-th/0103239.
436
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
S. Alexander, R. Brandenberger, D. Easson, Phys. Rev. D 62 (2000) 103509, hep-th/0005212.

R. Brandenberger, C. Vafa, Nucl. Phys. B 316 (1989) 391.
A.A. Tseytlin, C. Vafa, Nucl. Phys. B 372 (1992) 443, hep-th/9109048.
V. Sahni, H. Feldman, A. Stebbins, Astrophys. J. 385 (1992) 1.
H.A. Feldman, A.E. Evrard, Int. J. Mod. Phys. D 2 (1993) 113, astro-ph/9212002.
G. Veneziano, Phys. Lett. B 265 (1991) 287.
J. Polchinski, Cambridge Univ. Press, 1998, p. 531.
J. Polchinski, TASI lectures on D-branes, hep-th/9611050.
T.W. Kibble, J. Phys. A 9 (1976) 1387.
Y.B. Zeldovich, I.Y. Kobzarev, L.B. Okun, Zh. Eksp. Teor. Fiz. 67 (1974) 3, Sov. Phys. JETP 40 (1974) 1
(in English).
[16] A. Vilenkin, E.P.S. Shellard, Cosmic Strings and other Topological Defects, Cambridge Univ. Press, 1994.
[17] P.S. Aspinwall, M.R. Plesser, JHEP 9908 (1999) 001, hep-th/9905036.
[18] D.A. Easson, Brane gases on K3 and CalabiYau manifolds, hep-th/0110225.
Nuclear Physics B 623 [FS] (2002) 439473

Dimers and the Critical Ising Model on lattices

of genus > 1
Ruben Costa-Santos , Barry M. McCoy
C.N. Yang Institute for Theoretical Physics, SUNY Stony Brook, NY 11794-3840, USA
Abstract
We study the partition function of both Close-Packed Dimers and the Critical Ising Model on
a square lattice embedded on a genus two surface. Using numerical and analytical methods we
show that the determinants of the Kasteleyn adjacency matrices have a dependence on the boundary
conditions that, for large lattice size, can be expressed in terms of genus two theta functions.
The period matrix characterizing the continuum limit of the lattice is computed using a discrete
holomorphic structure. These results relate in a direct way the lattice combinatorics with conformal
field theory, providing new insight to the lattice regularization of conformal field theories on higher
genus Riemann surfaces. 2002 Elsevier Science B.V. All rights reserved.
PACS: 05.50.+q; 02.40.-k; 11.25.Hf
1. Introduction
When Kaufman [1] first evaluated the finite size partition function for the Ising model on
a toroidal square lattice, she found that the trace of the transfer matrix could be expressed
in a very compact form as the sum of four terms. In later solutions of the same problem,
both in the combinatorial approaches [2,3] or in the closely related Grassmann variable
approach [4], the sum of four terms appears as a natural way of expressing the torus
partition function. For the Close-Pack Dimer problem, on the same lattice, Kasteleyn [5]
showed that the partition function is given also by the sum of four terms, one of them found
to vanish after explicit calculation.
Kasteleyn [3,5,6] developed a combinatorial method of solution [79] that can be applied
to both the Ising model and the Dimer problem. In this method the four terms are the
* Corresponding author.
E-mail addresses: rcostas@insti.physics.sunysb.edu (R. Costa-Santos), mccoy@insti.physics.sunysb.edu

(B.M. McCoy).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 1 1 - 3
440
R. Costa-Santos, B.M. McCoy / Nuclear Physics B 623 [FS] (2002) 439473
Pfaffians of four adjacency matrices corresponding to different lattice edge orientations

1
Pf(A1 ) + Pf(A2 ) + Pf(A3 ) + Pf(A4 ) .
(1.1)
2
The Pfaffian of an antisymmetric matrix is the square root of the corresponding
determinant. These determinants can be explicitly evaluated for a square lattice with M
rows and N columns. For the Ising model at criticality and for the Dimer problem with any
dimer weights, one of the Pfaffians vanishes and the three remaining behave asymptotically
[1012], in the N = MN limit with a fixed ratio M/N , as a common bulk term times
a Riemann theta function [13] k (0, ) of even characteristic. In this limit, the partition
function for both models can be written as

1 2 (0| ) d 3 (0| ) d 4 (0| ) d
Zg=1 (Tc ) =
+
+
2
( )
( )
( )

exp(fd N ) 1 + O (log N )3 /N
(1.2)
Zg=1 (T ) =
with d = 1, 2 for Ising and Dimers respectively, fd being the free energy of the model. The
modular parameter in the theta functions given by

iM cosh 2Khc /N cosh 2Kvc for Ising,
=
(1.3)
iMzh /Nzv
for Dimers,
with Kvc / and Khc / being the vertical and horizontal coupling constants of the Ising
model at the critical point and zv and zh the vertical and horizontal Dimer weights.
Eq. (1.2) provides a bridge between lattice combinatorics and conformal field theory.
For the Ising model the term in square brackets is the modular invariant partition function
of the c = 1/2 conformal field theory on a torus, while for Dimers it correspond to a c = 1
conformal field theory [17]. The theta function dependence of the determinants of the
different adjacency matrices reproduce the dependence of the determinant of the Dirac
operator over the different boundary conditions (or spin structures) of the conformal field
theory [1417]. This kind of relation between combinatorics and analysis is well know
in the mathematics literature in the context of the RaySinger theorem [18,19] and it has
been mentioned in the context of the Ising model on the torus by Nash and OConnor
[12,20].
In this paper we study the extension of (1.2) to higher genus lattices. For a lattice
embedded in a genus g surface, the partition function of both Ising and Dimers can be
expressed as the sum over the Pfaffians of 4g adjacency matrices [3,8,9,24]. The number
of different Kasteleyn orientations 4g being precisely the number of different fermion
boundary conditions in the corresponding genus g Riemann surface. We expect the higher
genus lattice determinants to be related to the functional determinants of the conformal
field theory in a way similar to the toroidal case.
While expression (1.2) has been known for the torus for over thirty years, the corresponding higher genus case has not been studied. Simple nonorientable topologies have
been considered, such as the Mbius strip [21,22] and the Klein bottle [23], and lattices
with special symmetries [24] including the genus three Klein group lattice L(2, 7) [25], a
lattice with 168 vertices where the thermodynamic limit cannot be considered. Three-dim-
441
ensional lattices have also been studied with the Kasteleyn formalism by embedding the
three-dimensional lattice on a two-dimensional lattice of genus of the order of the lattice
size [26]. While the Kasteleyn formalism is well understood for any lattice embedded on
an orientable [8] or a nonorientable [9] surface, explicit computations of the adjacency
matrix determinants are not known for higher genus lattices.
We investigate these problems by studying Dimers and the critical Ising model on the
genus two lattice shown in Fig. 1. We show numerically that the corresponding sixteen
Pfaffians of adjacency matrices converge to genus two theta functions and that in the large
N limit the partition functions of both models can be written as

16

d
[i](0|) A exp(fd N )
Zg=2 (Tc )
(1.4)
i=1
where d = 1, 2 for Ising and Dimers respectively. The [i](z|) are the sixteen genus
two theta functions [13] with half-integer characteristics, defined in Section 3. The factor
Fig. 1. The genus two lattice with the boundary identifications given by the integers from 1 to 22. The
ai , bi cycles form a basis of the first homology group and should be seen as drawn over the lattice
edges. All lattice faces are squared except for two octagons whose edges are marked with the doted
line.
442
A includes normalization factors and nonbulk terms and is not obtainable by our method
that consists in the numerical evaluation of ratios of determinants of adjacency matrices
for large lattice sizes. The term in square brackets reproduces the classical winding part of
the corresponding conformal field theory partition function [17,27].
While on the torus, the modular parameter (1.3) can be understood as a weighted lattice
aspect ratio, on higher genus lattices the dependence of the period matrix of Eq. (1.4)
on the lattice properties is more elusive. We will study this dependence by introducing a
theory of discrete differentials on the lattice and extract the period matrix from solutions
of the finite difference Poisson equation on that lattice. The mathematical motivation
for this procedure is the idea of approximating smooth objects defined on a surface by
combinatorial objects defined on a triangulation of that surface [18,19,29,30].
The period matrix characterizing a surface is evaluated in two steps, the first step is the
determination of the space of harmonic differentials on that surface, the second step being
the decomposition of this space into the holomorphic and antiholomorphic subspaces.
Harmonic differentials are determined by Hodge decomposition [31,32] starting from
an inner product defined on differential forms. On a lattice embedded on a surface the role
of p-forms is played by the p-cochains, which are linear functionals defined on formal
sums of the lattice p-elements: vertices, edges and faces. We will call them respectively
the lattice functions, lattice differentials and lattice volume forms. Eckmann [28] showed
that any choice of inner product on the p-cochains gives rise to a combinatorial Hodge
theory and to a definition of harmonic p-cochains. Later Dodziuk [29], and Dodziuk
and Patodi [30] proved that for a suitable choice of this inner product on p-cochains, of
a triangulation of a surface, the resulting combinatorial Hodge theory converges, in the
continuum limit, to the surface Hodge theory.
While our definitions are based on the same general idea, replacing p-forms on a surface
by p-cochains on a lattice embedded on that surface, we depart from Dodziuk and Patodi
work on several points. The main difference is that their definition of the inner product
on p-cochains depends on a mapping of these cochains into forms on the embedding
surface [33] while our definition is done in term of lattice quantities. Dodziuk and Patodi
discrete theory is defined referring to the surface properties while we are interested in the
inverse procedure: to built the continuum theory from the combinatorics of the discrete
lattice.
To determine the period matrix we need the additional concept of discrete holomorphic
differentials. These problems have been considered in the study of finite element
electrodynamics [3438] where it is well known that the concept of holomorphic
differentials defy a straightforward discretization [39]. In the context of the Ising
model, discrete holomorphy can be traced back to the finite difference equations on the
correlation functions found by McCoy, Perk and Wu [40] and has been recently discussed
by Mercat [41,42]. In Mercats work the difficulty of defining discrete holomorphy for
a finite size lattice is patent from the fact that the various spaces of differentials have
dimensions double of the continuum analogues.
In this paper we give a construction procedure for holomorphic differentials, based
exclusively on the direct lattice, which becomes meaningful for large lattice size. To the
443
numerical precision we were able to test it, this procedure has a well defined continuum
limit and reproduces the period matrices observed in Eq. (1.4). This is the first time that an
argument is given to evaluate modular parameters for higher genus lattices.
The structure of the paper is as follows:
In Section 2 we review the Kasteleyn formalism and specify the adjacency matrices for
both the Ising model and the Dimer problem. In Section 3 we establish (1.4) by showing
numerically that ratios of determinants of the adjacency matrices converge to ratios of theta
functions. In Section 4 we define discrete harmonic differentials on the lattice and give
a procedure for numerical evaluation of the corresponding basis. In Section 5 we discuss
discrete holomorphic differentials and the computation of approximations to the period
matrix. In Section 6 these evaluations of lattice period matrices are compared with the
period matrices of (1.4), obtained by direct fit of determinant ratios. Our conclusions are
presented in Section 7 and the technical details of the Kasteleyn formalism are discussed
in Appendix A.
2. Determinants of adjacency matrices

Consider the square lattice with the boundary conditions shown in Fig. 1. This lattice,
G, is characterized by five integers sizes (M1 , M2 , K, N1 , N2 ) and two coupling constants
or dimer weights (zh , zv ) distinguishing the vertical and the horizontal directions. The
boundary conditions are such that the lattice can be drawn without superposition of edges
only on a surface of genus two or higher. The cycles of lattice edges ai , bi , with i = 1, 2,
represent a canonical basis of the first homology group of the embedding surface. The same
basis drawn over the dual lattice edges will be denoted by a i , bi (see Fig. 2).
We start by considering the Close-Packed Dimer problem on this lattice and for
simplicity we assume that all integer sizes (M1 , M2 , K, N1 , N2 ) are even integers. A closepacked dimer configuration in G is a selection of edges such that every vertex is included
once and only once as a boundary of an edge, see Fig. 3 for examples. If a weight zh is
assigned to the horizontal edges and a weight zv to the vertical edges, the dimer partition
function is defined to be
Z Dimer =
zhnh zvnv
(2.1)
dimer config.
where nh and nv are the number of horizontal and vertical edges in a given dimer
configuration and the sum runs over all possible dimer configurations on G.
While the Close-Packed Dimers problem is an interesting statistical mechanics model on
its own it also provides a combinatorial approach to the Ising model. There is a well known
correspondence between polygon configurations of the Ising model high temperature
expansion and dimer configurations on a decorated lattice [3]. The Ising model partition
function on the lattice G can then be expressed as
444
Fig. 2. The A(0000) clockwise odd orientation of the original lattice. The a i , bi cycles form a basis of
the first homology group drawn over the dual lattice edges. The remaining clockwise odd orientations
can be obtained from this one by inverting the orientations of the edges crossed by a given choice of
the a i , bi cycles.

Ising
= (2 cosh Kv cosh Kh )
wvnv whnh
(2.2)
dimer config.
with wi = tanh Ki , where Kv and Kh are the vertical and horizontal coupling constants
and N is the number of vertices in the lattice G. The sum runs over all the dimer
configurations of the decorated lattice G represented in Fig. 3.
Kasteleyn developed a combinatorial formalism [3,59] that allows the expression of
these dimer partition functions as a sum over the Pfaffians of 4g adjacency matrices,
where g is the genus of the simplest surface where that lattice can be drawn without
superposition of edges. A detailed discussion of the higher genus Kasteleyn formalism
is given in Appendix A, here we will state only the main results.
The Kasteleyn adjacency matrices are defined in the following way: label the lattice
vertices with an integer from 1 to N and choose a lattice edge orientation by assigning to
each edge a direction represented by an arrow (see Figs. 2 and 4 for two examples). The
signed adjacency matrix corresponding to this edge orientation is the N N matrix Aij
445
Fig. 3. The Ising decorated lattice, the dimer weights of each edge type are shown in the box.
with entries
z
Aij =
z
0
if there is an arrow from vertex i to vertex j of weight z,

if there is an arrow from vertex j to vertex i of weight z,
otherwise.
(2.3)
The Pfaffian of such an N N antisymmetric matrix, with N even, is defined as
1
$p Ap1 p2 Ap3 p4 ApN 1 pN
Pf(A) = N /2
2
(N /2)! p
(2.4)
where the sum goes over all the permutations p of the integers from 1 to N and $p = 1
for even and odd permutations respectively.
From the definition of the adjacency matrix it is clear that each nonzero term in the
Pfaffian expansion equals, in absolute value, a term in the corresponding dimer partition
function. The relative sign between different terms depends on the choice of the edge
orientation. We would like to choose an orientation such that all the terms in the Pfaffian
have the same relative sign. Then the Pfaffian of the adjacency matrix would equal the
partition function modulo an overall sign.
Kasteleyn showed in a beautiful tour de force of combinatorics [5] that edge orientations
with this property exist. These are the edge orientations such that every lattice face was an
446
Fig. 4. The A(0000) clockwise odd orientation of the Ising decorated lattice. The remaining
clockwise odd orientations can be obtained from this one by inverting the orientations of the edges
crossed by a given choice of the a i , bi cycles.
odd number of clockwise oriented edges. In a genus g lattice there will be 4g different
such orientations and the dimer partition function is given as a linear combination of their
Pfaffians [6,8,9].
For the genus two lattices G and G , that we are considering in this paper, the
relevant sixteen edge orientations can be labeled as A(na 1 , nb1 , na 2 , nb2 ) with nx = 0, 1
for x = a 1 ,a 2 , b1 , b2 . The starting orientations A(0000) are shown in Figs. 2 and 4 for the
lattices G and G respectively. An orientation with a certain nx = 1 is obtained from the

corresponding orientation with nx = 0 by introducing a disorder loop along the nontrivial
cycle x, this is, by reversing the orientation of all the edges crossed by the cycle x.
To make connection with the theta functions characteristics and allow for more compact
equations we will also use the alternative notations

n
nb2
A(na 1 , nb1 , na 2 , nb2 ) = A b1
(2.5)
= Ai
na 1 na 2
447
with the integer label given by i = 16 8na 1 4nb1 2na 2 nb2 .

We show in Appendix A that, in terms of these orientations, the Dimer and the Ising
model partition functions on G are given by
1
Z = (P1 P2 P3 P4 P5 + P6 + P7 + P8
4
P9 + P10 + P11 + P12 P13 y P14 + P15 + P16 )
(2.6)
where Pi = Pf Ai , and the adjacency matrices are defined on the lattice G for Dimers and
on the lattice G for the Ising model.
For translational invariant lattices, as the torus square lattice, the Pfaffians of the
adjacency matrices can be evaluated in a closed form and the theta function dependence
can be extracted by a careful asymptotic analysis [1012]. For the genus two lattices G
and G such an analytic treatment is not possible and we are forced to resort to numerical
evaluations of the Pfaffians.
3. Ratios of determinants and theta functions

Pfaffians of adjacency
matrices can be numerically evaluated for large lattice sizes using
the fact that Pf A = det A for an antisymmetric matrix A. For a given lattice aspect
ratio (m1 , m2 , k, n1 , n2 ) we evaluate the ratios of determinants of the adjacency matrices
for a sequence of lattices characterized by the integers sizes (M1 , M2 , K, N1 , N2 ) =
(m1 , m2 , k, n1 , n2 ) L with increasing L. The Ising model coupling constants were chosen
to satisfy the criticality condition for a square lattice on the torus
sinh(2Kh ) sinh(2Kv ) = 1,
(3.1)
while no restrictions were imposed on the vertical and horizontal weights of the Dimer
model. Numerical examples for the critical Ising model and Dimers are shown in Tables 1
and 2.
The fifteen ratios of determinants, obtained is this way, are found to converge smoothly
with the lattice size N , see Figs. 5, 6 and 7. The solid line in Figs. 5 and 6 is a fit with a
quadratic polynomial on 1/N to the values obtained for different lattice sizes.
Unlike the toroidal case, none of the sixteen determinants vanishes at finite size for the
(3.1) choice of the Ising coupling constants or for any choice of vertical and horizontal
dimer weights of the Dimer problem. There are choices of Ising coupling constants with
small deviations from the criticality condition (3.1) that will make some of the determinants
vanish at finite size but not all of them simultaneously. However six of the determinant
ratios converge to small values for large N and can be associated with theta functions of
odd characteristic. The convergence of these six ratios is shown in Fig. 7 where the values
corresponding to the Ising model ratios are squared for comparison.
The genus two theta functions are defined by the quickly converging series
expansion [13]

(z, ) =
(3.2)
exp i(n + )T (n + ) + 2i(n + )T (z + )
2
nZ
448
Table 1
Convergence with lattice size N of the ratios of determinants of the adjacency matrices Ri =
2 for the critical Ising model in a lattice with w = 0.66403677026784896368, w =
Pi2 /P16
v
h
0.20189651799465540849 and aspect ratio (m1 , m2 , k, n1 , n2 ) = (4, 2, 2, 2, 4). See Section 3 for
discussion
L
448
1008
1792
2800
4032
5488
extrapolated
R12 0.001419
0.001746
0.001867
0.001923
0.001953
0.001972
0.002023
R22 0.000007
0.000005
0.000003
0.000002
0.000002
0.000001
107
R32 0.000000
0.000000
0.000000
0.000000
0.000000
0.000000
108
R42 0.000004
0.000003
0.000002
0.000001
0.000001
0.000001
107
R52
R62
R72
R82
R92
2
R10
2
R11
2
R12
2
R13
2
R14
2
R15
2
R16
0.000003
0.000002
0.000001
0.000001
0.000001
0.000001
108
0.004802
0.005766
0.006120
0.006284
0.006375
0.006432
0.006584
0.000001
0.000001
0.000001
0.000001
0.000001
0.000001
0.000001
0.003146
0.003899
0.004182
0.004316
0.004390
0.004437
0.004562
0.000213
0.000117
0.000072
0.000048
0.000034
0.000026
0.000001
0.631982
0.657856
0.666786
0.670990
0.673343
0.674812
0.678782
0.311574
0.315561
0.316467
0.316709
0.316744
0.316704
0.316655
0.991780
0.992740
0.993053
0.993190
0.993260
0.993301
0.993414
0.000212
0.000116
0.000072
0.000048
0.000034
0.000026
0.000001
0.637340
0.662754
0.671528
0.675663
0.677981
0.679430
0.683345
0.309128
0.313357
0.314346
0.314625
0.314679
0.314650
0.314633
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
where , , z and n are 2-vectors half-integers, complex numbers and integers, respectively. The 2 2 period matrix is a symmetric complex matrix with positive definite
imaginary part.
We found that the extrapolated values of the determinants ratios, in the L limit,
can be expressed in terms of ratios of theta functions as
c /2 c /2
d

det A cc13 cc24
c13 /2 c24 /2 (0, )
(3.3)

=

det A 00 00 T
00 00 (0, )
c
for the 16 combinations of ci = 0, 1 with d = 2 for critical Ising and d = 4 for Dimers.
The period matrix being determined from the determinant ratios by a suitable numerical
fitting procedure. See Tables 3 and 4 for examples. Each table displaying the results
for three different lattices, the first column shows the L extrapolated ratios of
determinants and the second column shows the theta ratios

c /2 c2 /2
0 0
(168d1 4c1 2d2 c2 ) () = 1
(3.4)
(0, )/
(0, )
d1 /2 d2 /2
0 0
449
Table 2
Convergence with lattice size N of the ratios of determinants of the adjacency matrices for the Dimer
model in a lattice with zh = 0.780, zv = 0.560 and aspect ratio (m1 , m2 , k, n1 , n2 ) = (4, 2, 2, 6, 2).
Additional discussion is given in Section 3
L
10
11
640
1000
1440
1960
2560
3240
4000
4840
extrapolated
R1
R2
R3
R4
R5
R6
R7
R8
R9
R10
R11
R12
R13
R14
R15
R16
0.209606
0.000024
0.000340
0.000010
0.000017
0.805524
0.000024
0.555230
0.000457
0.123728
0.262676
0.183361
0.000116
0.686459
0.050147
1.000000
0.213826
0.000017
0.000243
0.000007
0.000012
0.808824
0.000024
0.569831
0.000325
0.127266
0.266688
0.184107
0.000082
0.701963
0.050986
1.000000
0.215781
0.000012
0.000181
0.000005
0.000008
0.810645
0.000024
0.577729
0.000240
0.129148
0.268415
0.184460
0.000060
0.710275
0.051336
1.000000
0.216790
0.000009
0.000139
0.000004
0.000006
0.811765
0.000023
0.582534
0.000184
0.130271
0.269229
0.184642
0.000046
0.715308
0.051492
1.000000
0.217348
0.000007
0.000110
0.000003
0.000005
0.812510
0.000023
0.585711
0.000145
0.130998
0.269627
0.184739
0.000036
0.718626
0.051559
1.000000
0.217669
0.000006
0.000089
0.000003
0.000004
0.813033
0.000023
0.587939
0.000117
0.131498
0.269816
0.184791
0.000029
0.720950
0.051584
1.000000
0.217855
0.000005
0.000073
0.000002
0.000003
0.813417
0.000023
0.589574
0.000097
0.131857
0.269893
0.184818
0.000024
0.722654
0.051586
1.000000
0.217961
0.000004
0.000061
0.000002
0.000003
0.813709
0.000023
0.590815
0.000081
0.132125
0.269907
0.184830
0.000020
0.723949
0.051576
1.000000
0.218541
107
0.000004
107
107
0.815064
0.000022
0.596530
0.000004
0.133375
0.270077
0.184916
0.000001
0.729883
0.051561
1.000000
for ci , di = 0, 1, with a period matrix obtained by numerical fit. The precision to which the
two sets of numbers agree is remarkable, typically a precision from 104 to 106 .
For the Ising model, way from the criticality condition (3.1), the sixteen determinants
are found to converge rapidly to the same value, while for the Dimer problem the theta
function expression (3.3) is valid for all values of the dimer weights.
The period matrices found in (3.3) are purely imaginary for all lattice aspect ratios and
coupling constants. For both Ising and Dimers we have

11 12
.
=i
(3.5)
12 22
For such a period matrix the theta functions at zero argument are real. This seems to be a
property of locally square lattices observed on the toroidal square lattice (1.3) but not on the
corresponding triangular lattice [12] where the modular parameter is in general complex.
4. Harmonic differentials on the lattice

In this section we will consider quantities defined on the lattice p-elements: vertices,
edges and faces. We will call them respectively the lattice functions, lattice differentials
and lattice volume forms.
450
Fig. 5. The continuum limit for Dimers: the nine nonvanishing ratios of determinants, defined as
Ri = det Ai / det A16 , and the corresponding ratios of theta functions are shown as a function of the
number of lattice vertices N , for the lattice with aspect ratio (m1 , m2 , k, n1 , n2 ) = (4, 2, 2, 6, 2) and
zh = 0.780, zv = 0.560. The solid curve is the plot of the determinant ratios and the dashed curves
are the plots of ratios of theta functions, with period matrices evaluated for the A and B choices of
the homology basis (see Section 6 for discussion). The curves are fits with a polynomial on 1/N to
the values obtained for different lattice sizes.
A lattice function f is defined by its value at each lattice vertex and can be represented
by a N -vector f [n]: n = 1, . . . , N , after some integer labeling of the N vertices of the
lattice is chosen.
A lattice differential w is defined by its value on the oriented lattice edges and will be
represented by a N 2-matrix w[n|p]: n = 1, . . . , N ; p = 1, 2, where [n|1] stands for the
edge immediately right of the vertex n and [n|2] for the edge immediately above, referring
to the lattice drawing of Fig. 1. We define the horizontal edges to be oriented from left
to right and vertical edges from bottom to top. The line integral of a lattice differential w
along a path C of lattice edges is the sum of the values of w on all the edges included in
the path

w=
w[n|p];
(4.1)
C
[n|p]C
the minus sign applying to edges with opposite orientation to the one of the path.
A lattice volume form is defined by its value on each lattice face. The lattice in Fig. 1
when drawn on a genus two surface has N 2 faces, two of which are octagons the
451
Fig. 6. The continuum limit for Ising: the nine nonvanishing ratios of determinants and ratios of theta
functions are shown, as a function of the number of lattice vertices N , for the lattice with aspect ratio
(m1 , m2 , k, n1 , n2 ) = (2, 2, 2, 2, 2) and wh = 0.537, wv = 0.301. Some of the ratios are equal due
to the high symmetry of the lattice.
Fig. 7. The six vanishing ratios of determinants for four different lattices. For comparison Ising
determinant ratios are shown squared. Coupling constants and aspect ratios for the lattices I1, I3, P1
and P3 are given in Tables 3 and 4.
452
Table 3
Comparison of ratios of determinants of adjacency matrices with ratios of theta functions for the
critical Ising model in three different lattices: I1, I2 and I3. Each lattice is characterized by the
couplings and aspect ratio [wh , wv ] (m1 , m2 , k, n1 , n2 ). The L ratios are compared with
theta function ratios for a fitted period matrix (fit) and the period matrix (eval) evaluated by
the procedure of Section 4 (modular choice A). The first two sets of values illustrate the fact that rescaling of the couplings can be compensated by a re-scaling of the overall vertical/horizontal aspect
ratio
I1: [0.537, 0.301] (2, 2, 2, 2, 2)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
I2: [0.335, 0.498] (4, 4, 2, 2, 2)
I3: [0.664, 0.202] (4, 2, 2, 2, 4)
Ri2
i4 (fit)
i4 (eval)
Ri2
i4 (fit)
i4 (eval)
Ri2
i4 (fit)
i4 (eval)
0.214707
0.000001
0.000001
0.000001
0.000000
0.664619
0.000339
0.450251
0.000001
0.214707
0.335047
0.335047
0.000000
0.664619
0.120683
1.000000
0.214705
0.000000
0.000000
0.000000
0.000000
0.664616
0.000338
0.450250
0.000000
0.214705
0.335045
0.335045
0.000000
0.664616
0.120679
1.000000
0.214872
0.000000
0.000000
0.000000
0.000000
0.663636
0.000350
0.449115
0.000000
0.214880
0.336005
0.336014
0.000000
0.663645
0.121484
1.000000
0.214748
0.000002
0.000001
0.000001
0.000001
0.664364
0.000341
0.449951
0.000002
0.214748
0.335287
0.335287
0.000001
0.664364
0.120883
1.000000
0.214751
0.000000
0.000000
0.000000
0.000000
0.664366
0.000341
0.449957
0.000000
0.214751
0.335292
0.335292
0.000000
0.664366
0.120883
1.000000
0.214917
0.000000
0.000000
0.000000
0.000000
0.663404
0.000353
0.448840
0.000000
0.214917
0.336243
0.336243
0.000000
0.663404
0.121679
1.000000
0.002023
0.000000
0.000000
0.000000
0.000000
0.006584
0.000001
0.004562
0.000001
0.678782
0.316655
0.993414
0.000001
0.683345
0.314633
1.000000
0.002023
0.000000
0.000000
0.000000
0.000000
0.006585
0.000001
0.004563
0.000000
0.678782
0.316655
0.993414
0.000000
0.683344
0.314633
1.000000
0.002015
0.000000
0.000000
0.000000
0.000000
0.006544
0.000001
0.004529
0.000000
0.677686
0.317785
0.993456
0.000000
0.682214
0.315771
1.000000
11
12
22
11
12
22
11
12
22
(fit) 1.174907 1.024878 2.049756 1.174646 1.024365 2.048730 0.403512 0.368642 1.526888
(eval) 1.17389 1.02281 2.04563 1.17365 1.02232 2.04464 0.40320 0.367695 1.52404
remaining being squares. The volume form will be represented by a (N 2)-vector

[q]: q = 1, . . . , N 2, where q is an integer labeling the N 2 faces of the lattice. The
integral of a volume form over a given area A on the lattice is the sum of the values that
takes on all the lattice faces contained in that area

=
[q].
(4.2)
A
[q]A
It is convenient to relate the labeling of vertices with the labeling of faces. For that
purpose we introduce the following notation: if q labels a squared face then q1 stands for
the label of its lower left vertex; if q labels an octagonal face then q1 and q2 stand for
the labels of the two vertices that can be seen as the octagons lower left vertices in Fig. 1.
Reciprocally if n labels a lattice vertex then n is the label of the lattice face of which n can
be seen as the lower left vertex. We then have the relations
qi = q,
(4.3)
453
Table 4
Comparison of ratios of determinants of adjacency matrices Ri with ratios of theta functions for the
Dimer model in three different lattices: P1, P2 and P3. Each lattice is characterized by the dimer
weights and aspect ratio [zh , zv ] (m1 , m2 , k, n1 , n2 )
D1: [1.100, 0.900] (2, 2, 2, 2, 2) D2: [0.275, 0.450] (4, 4, 2, 2, 2) D3: [0.780, 0.560] (4, 2, 2, 6, 2)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
Ri
i4 (fit)
i4 (eval)
Ri
i4 (fit)
i4 (eval)
Ri
i4 (fit)
i4 (eval)
0.136210
0.000001
0.000001
0.000001
0.000000
0.828760
0.000385
0.692938
0.000001
0.136210
0.170862
0.170862
0.000000
0.828760
0.035039
1.000000
0.136207
0.000000
0.000000
0.000000
0.000000
0.828757
0.000384
0.692934
0.000000
0.136207
0.170859
0.170859
0.000000
0.828757
0.035036
1.000000
0.136552
0.000000
0.000000
0.000000
0.000000
0.828055
0.000397
0.691901
0.000000
0.136552
0.171547
0.171547
0.000000
0.828055
0.035393
1.000000
0.136168
0.000001
0.000001
0.000001
0.000000
0.828291
0.000387
0.693182
0.000001
0.13628
0.170903
0.171024
0.000000
0.828514
0.035084
1.000000
0.136061
0.000000
0.000000
0.000000
0.000000
0.828547
0.000405
0.692891
0.000000
0.136213
0.170896
0.171048
0.000000
0.828699
0.03524
1.000000
0.136671
0.000000
0.000000
0.000000
0.000000
0.827822
0.000402
0.691553
0.000000
0.136666
0.171781
0.171776
0.000000
0.827818
0.035512
1.000000
0.218541
0.000000
0.000004
0.000000
0.000000
0.815064
0.000022
0.596530
0.000004
0.133375
0.270077
0.184916
0.000001
0.729883
0.051561
1.000000
0.218546
0.000000
0.000000
0.000000
0.000000
0.815061
0.000022
0.596536
0.000000
0.133377
0.270087
0.184917
0.000000
0.729892
0.051562
1.000000
0.219812
0.000000
0.000000
0.000000
0.000000
0.814718
0.000023
0.594928
0.000000
0.133252
0.271820
0.185259
0.000000
0.728157
0.052031
1.000000
11
12
22
11
12
22
11
12
22
(fit) 1.422286 1.208265 2.416530 1.422067 1.205355 2.411027 1.389893 1.297942 2.457896
(eval) 1.42099 1.20562 2.41124 1.42056 1.20477 2.40953 1.38931 1.29614 2.45252

n n 1 , n 2 .
(4.4)
An inner product on lattice functions, lattice differentials and lattice volume forms can
be defined in the following way

f, f =
f [n]f [n],

w, w =

, =

(h/v)w[n|1]w [n|1] + (h/v)1 w[n|2]w [n|2] ,
[q] [q],
(4.5)
where h/v is a positive parameter providing different weighting of the horizontal and
vertical components of a differential.
The lattice exterior derivative d is a linear operator that takes lattice functions into lattice
differentials and lattice differentials into lattice volume forms. For our labeling and choice
of orientations d is given by

(d f )[n|1] = f right(n) f [n],

(d f )[n|2] = f up(n) f [n],
454
(d w)[q] =

w qi |1 + w right(qi )|2 w up(qi )|1 w qi |2 ,
(4.6)
with i = 1, 2 for octagonal faces and i = 1 for squared faces. The functions right(n), left(n),
up(n) and down(n) give the label of the vertex immediately right, left, above and below of
vertex n, referring to the lattice drawing of Fig. 1.
The exterior derivative defined in this way satisfies a discrete version of Stokes theorem.
Let C(n, n ) be a path of lattice edges from vertex n to vertex n and A an area on the
lattice, then we have that

d f = f [n ] f [n],
C(n,n )

dw=
A
w,
(4.7)
where A is the path along the boundary of the area A with an anticlockwise orientation.
The co-derivative is defined as the adjoint operator of the exterior derivative
(w, d f ) = ( w, f ),
(, d w) = ( , w),
(4.8)
and can be expressed as an operator that takes lattice differentials into lattice functions and
lattice volume forms into lattice differentials

v

h
( w)[n] = w left(n)|1 w[n|1] + w down(n)|2 w[n|2] ,
(4.9)
v
h

v
(n) ,
( )[n|1] = n down
h

h
( )[n|2] = l
(4.10)
eft(n) n ,
v
(n) stands for the face given by the tilde of the vertex left(n) and similar for
where left
(n).
down
Both the exterior derivative and the co-derivative satisfy dd = = 0. Explicitly in terms
of the formulas above we have that

(dd f )[q] =
(4.11)
f up right qi f right up qi
= 0,

left(n) = 0.
( )[n] = l
eft down(n) down
(4.12)
There are four vertices on the octagonal faces where right(up(n)) = up(right(n)), at these
faces dd vanishes because of cancellation between the q1 and the q2 terms. There are also
(down(n)) = down
(left(n)) is true
four vertices where left(down(n)) = down(left(n)) but left
for all vertices.
In exact analogy with the continuum definitions, a lattice differential w is said to be
closed if d w = 0, exact if w = d f , co-closed if w = 0 and co-exact if w = . We are
mainly interested in the harmonic differentials, a lattice differential is said to be harmonic
if it is both closed and co-closed

d w[q] = 0, q = 1, . . . , N 2,
w is harmonic
w[n] = 0, n = 1, . . . , N .
455
(4.13)
In general, on a genus g lattice, the lattice harmonic differentials form a vector space
of dimension 2g. For the genus two lattice G this can be seen directly from the system of
Eqs. (4.13). It is a system of 2N 2 equations on 2N unknowns and there are at least
two independent solutions. Two additional solutions are provided by constant differentials
with independent vertical and horizontal components and the system has at least four
independent solutions.
It is remarkable that, on the lattice, the dimension of the space of harmonic differentials
is a linear algebra problem determined by the Euler characteristic. Consider a genus g
lattice with the property that its edges can be distinguished into two classes: vertical and
horizontal. For such a lattice the dimension of the space of harmonic differentials, that is,
the number of independent solutions of the system (4.13) on that lattice, is at least
#(edges) #(vertices) #(faces) + 2 = + 2 = 2g,
(4.14)
where is the Euler characteristic of the lattice.

It still remains to be proven that there are not more than 2g linear independent harmonic
lattice differentials. Note that 2g is the dimension of the lattice first homology group.
We will prove, in a way similar to the continuum, that a harmonic lattice differential is
completely determined by its integrals, or periods, around the lattice nontrivial loops and
therefore there can only be 2g independent harmonic differentials.
Harmonic differentials are by definition closed and its integrals depend only on the
homology class of the path considered. If two differentials w and w have exactly the
same periods along the 2g classes of loops then their difference w w has zero integral
along any closed path on the lattice and the function
n
f [n] =

w w
(4.15)
is well defined for any lattice path going from a fixed vertex o to the vertex n. This function
is a harmonic function, in the sense that it satisfies

1f [n] (d f )[n] = w w [n] = 0
(4.16)
with the Laplacian operator given explicitly by
(1f )[n] = 2(h/v + v/ h)f [n]

h/v f right(n) + f left(n)

v/ h f up(n) + f down(n) .
(4.17)
It is easily seen that a function harmonic at n cannot have a local extremum at n and that
the only functions harmonic everywhere on the lattice are constants. From this it follows
that the function defined in (4.15) vanishes at all lattice vertices. In this way, we proved that
456
if two harmonic differentials w and w have the same periods then they are equal, w = w .
The space of harmonic differentials has therefore the same dimension as the first homology
group, 2g.
For a given lattice a basis of the space of harmonic differentials can be evaluated by
direct solution of the linear system of Eqs. (4.13). Such a solution can always be found,
at least numerically, although the evaluation of the kernel of a large linear system is
computationally demanding. For practical purposes there is a much more efficient method
based on a discrete version of the Hodge decomposition theorem.
The discrete Hodge decomposition theorem states that the space of lattice differentials
has an orthogonal decomposition in terms of exact, co-exact and harmonic differentials
and that any lattice differential w can be written in an unique way as
w = df + h + ,
(4.18)
where h is a harmonic differential. The orthogonality of the different kinds of differentials

follows directly from (4.13) and the property dd = = 0.
This result allows the determination of harmonic differentials by orthogonal projections.
Our objective is to obtain a basis {Ak , Bk : k = 1, 2} of the space of harmonic differentials
satisfying the normalization conditions

Ak = kj ,
Ak = 0,
aj
Bk = 0,
aj
bj
Bk = kj ,
(4.19)
bj
with the aj , bj being any choice of closed paths on the lattice representing a basis of
the first homology group. We can proceed in the following way: start with closed but not
harmonic differentials A k , B k with the periods required by (4.19). A possible choice for
these differentials is shown in Fig. 8 where we take A k (B k ) to be zero on all edges except
the ones crossed by the dual lattice cycles bk (respectively a k ).
Since closed differentials are orthogonal to co-exact differentials it follows from (4.18)
that closed differentials can be written as the sum of a harmonic differential with the same
periods and an exact differential
A k = Ak + dfkA ,
B k = Bk + dfkB ,
(4.20)
applying on the these equations and using (4.13) and (4.16) we obtain
1fkA = A k ,
1fkB = B k .
(4.21)
The Laplacian operator on functions (4.17) is a well defined N N matrix of rank

(N 1) acting on the functions N -vector and (4.21) can be solved numerically for
reasonably large N by fixing the value of the functions at a lattice vertex. The solutions
457
Fig. 8. The closed differentials A k , B k . The differentials are zero on all edges except the ones crossed
by the respective bk , a k loop. These edges are shown in bold on the figure, together with the value
that the differential takes at that edge. For convenience we use a a 2 loop different but homologically
equivalent to the one shown in Fig. 1.
can then be differentiated and subtracted to the A k , B k to obtain a normalized basis of

the space of harmonic differentials {Ak , Bk : k = 1, 2}. This method is numerically more
efficient than the direct solution of the linear system (4.13).
5. Holomorphic differentials and the period matrix

To evaluate a discrete approximation to the period matrix we need to introduce
the concept of lattice holomorphic differentials and decompose the space of harmonic
differentials {Ak , Bk : k = 1, 2} obtained in the previous section into the holomorphic and
antiholomorphic subspaces.
In the continuum theory this is accomplished by the projection operators
P = (1 + i)/2 and P = (1 i)/2,
(5.1)
where is the (continuum) Hodge operator [31,32]. This operator is an endomorphism on

the space of harmonic differentials and satisfies 2 = 1. Its action on differentials is given
by

fx (x, y) dx + fy (x, y) dy = fy (x, y) dx + fx (x, y) dy
(5.2)
and a differential is said to be holomorphic if it is of the form P w with w harmonic.
458
We will proceed in a similar way on the discrete theory and define lattice holomorphic
differentials by a projection with a discrete Hodge operator 3. It proves very difficult to
define a discrete Hodge operator acting on lattice differentials, on the finite size lattice,
with the same properties of the continuum one. We propose the following definition

(3 w)[n|1] = w down(n)|2 (h/v)1 ,

(3 w)[n|2] = w left(n)|1 (h/v).
(5.3)
The action of the discrete Hodge star on the lattice differentials is shown in Fig. 9. In this
figure an arrow connecting two edges represents that the value of 3 w at the second edge
is obtained from the value of differential w at the first edge. Besides the definition above
there are three related definitions of the Hodge star, corresponding to /2 rotations, that
are also shown in Fig. 9. Due to the symmetries of the lattice, the four different definitions
produce the same numerical results for the period matrix.
The principal merit of these definitions is the way they relate the exterior derivative d
with the co-derivative . From the definition above (4.6) and (4.9) it follows that

w qi ,
(d 3 w)[q] =
(5.4)

( 3 w)[n] =
(n)
if n {n1 , n2 , n3 , n4 },

(left(n)) otherwise,
dw down
(5.5)
where in the first equation the sum goes over i = 1, 2 for octagonal faces and is i = 1 for
squared faces. The nk are the four lattice vertices where left(down(nk )) = down(left(nk )).
Let n1 and n2 be on the same octagon and the two remaining points on the other octagon
then the (nk ) satisfy

left(n1 ) ,
(n1 ) + (n2 ) = dw down

left(n3 ) .
(n3 ) + (n4 ) = dw down
(5.6)
The appearance of the terms in (5.5) prevents the discrete Hodge operator from being
an endomorphism on the harmonic lattice differentials. These terms originate from the fact
Fig. 9. Graphical description of the action of the discrete Hodge operator on the edge values of lattice
differentials. Four different definitions, related by /2 rotations, are possible. The definition given
in the text (5.3) corresponds to the drawing A. The four choices have similar properties and generate
the same period matrix.
459
that the lattice has more vertices than faces: the requirement that a differential w is closed is
not sufficient to ensure that the corresponding 3 w is co-closed. For the dual lattice, where
the number of faces is larger than the number of vertices, the opposite happens and the
terms appear in (5.4) instead of (5.5).
Unlike the continuum Hodge star, our lattice definition does not satisfy 32 = 1 but
rather

32 w[n|1] = w left down(n) |1 ,

32 w[n|2] = w down left(n) |2 ;
(5.7)
there is a displacement of the differential and mixing at the octagons.
Notice that for a toroidal square lattice, where all faces are squares and left(down(n)) =
down(left(n)) for all vertices n, there are no (n) terms on the equivalent of Eqs. (5.4)
and (5.5). The lattice Hodge star (5.3) on a torus is an endomorphism on the space of
harmonic differentials.
Lets proceed to evaluate the torus modular parameter and compare with the know
result (1.3). Consider a toroidal square lattice with M rows and N columns. Let {a, b} be a
basis of the first homology group with a being a horizontal loop and b a vertical loop. This
lattice has two independent normalized harmonic differentials A and B with components
A[n][1] = 1/N,
A[n][2] = 0,
B[n][1] = 0,
B[n][2] = 1/M,
(5.8)
for all vertices n. The only holomorphic differential on the lattice is = A + i 3 A, in

components,
h 1
1
,
[n][2] = i
,
(5.9)
N
vN
for all vertices n. The modular parameter is the integral of this differential along a loop of
the homology class of b and we obtain

hM
= (A + i 3 A) = i
(5.10)
,
vN
[n][1] =
that reproduces the well known result (1.3) if we take for the h/v ratio

cosh 2Khc / cosh 2Kvc for the critical Ising Model,
h/v =
for the Close-Packed Dimers.
zh /zv
(5.11)
For the Ising model this ratio corresponds to the ratio of the diverging correlation lengths
along the two lattice directions [43], the relevant geometrical quantity at criticality. We
will use the same expression for the genus two case, however in this case the lattice Hodge
operator (5.3) is not an exact endomorphism on the harmonic lattice differentials, due to
the terms that appear on the four nk vertices.
We will take (5.3) as the definition of our lattice Hodge operator and proceed to evaluate
a basis for the space of differentials of the form {k = (1 + i3)Hk : k = 1, . . . , g} where the
460
Hk are harmonic differentials, satisfying the normalization conditions

(Hl + i 3 Hl ) = kl ,
(5.12)
ak
(Hl + i 3 Hl ) = kl .
(5.13)
bk
Eq. (5.12) is a set of 2g real constraints on Hk from which we can determine the
coefficients of Hk in the basis {Ak , Bk } and Eq. (5.13) gives our approximation to the
lattice period matrix.
Since {k = (1 + i3)Hk : k = 1, . . . , g}, seen as a complex vector space, is not invariant
under P = (1 + i3)/2 we are not allowed to call these lattice differentials holomorphic
in a strict sense. However a detailed numerical study, discussed on the next section, will
show that the period matrix obtained by this procedure reproduces to excellent precision
the period matrices of Eq. (3.3).
6. Period matrices and determinant ratios

Following the procedure described in the two previous sections we can evaluate
numerically the period matrix for lattices with different aspect ratios (m1 , m2 , k, n1 , n2 )
and coupling constants. Table 5 shows the period matrix for a particular lattice aspect ratio
(m1 , m2 , k, n1 , n2 ) and various lattice sizes (M1 , M2 , K, N1 , N2 ) = (m1 , m2 , k, n1 , n2 )L,
with increasing L. The period matrix elements converge in a smooth way with the number
of lattice vertices N and the resulting matrix is a purely imaginary, symmetric and positive
definite matrix for all lattice sizes.
The three different set of values shown in Table 5 correspond to three different choices
of the first homology group basis. The first set of values, labeled as A, corresponds to the
Table 5
Convergence with lattice size N of the period matrix elements for a lattice with h = 0.780, v = 0.560
and aspect ratio (m1 , m2 , k, n1 , n2 ) = (4, 2, 2, 6, 2) for three different choices of the homology basis:
A, B and C. See Section 6 for discussion
L
B: (4, 2, 2, 6, 2) with ai bi
A: (4, 2, 2, 6, 2)
C: (2, 4, 2, 2, 6)
11
12
22
11
12
22
11
12
22
5
6
7
8
9
10
11
1000
1440
1960
2560
3240
4000
4840
1.383663
1.384849
1.385681
1.386311
1.386801
1.387192
1.387513
1.279193
1.282726
1.285248
1.287139
1.288608
1.289782
1.290743
2.401681
2.412281
2.419848
2.425518
2.429926
2.433450
2.436332
1.414286
1.415022
1.415555
1.415958
1.416274
1.416528
1.416737
0.7392862
0.7407580
0.7418236
0.7426305
0.7432626
0.7437711
0.7441889
0.7821439
0.7850874
0.7872187
0.7888325
0.7900967
0.7911136
0.7919492
1.226958
1.231669
1.235032
1.237552
1.239511
1.241077
1.242358
1.122488
1.129555
1.134599
1.138380
1.141318
1.143667
1.145589
2.401681
2.412281
2.419848
2.425518
2.429926
2.433450
2.436332
1.38931
1.29614
2.45252
1.41792
0.746553
0.796677
1.24955
1.15638
2.45252
461
choice of homology basis ai , bi shown in Fig. 1; the set labeled as B corresponds to a choice
where the ai and bi loops are interchanged from Fig. 1 and the set C corresponds to the
choice where the top and bottom of Fig. 1 are interchanged. The different period matrices
obtained for each choice of the homology basis are related by a modular transformation.
In Table 6 the theta function ratios corresponding to each choice are shown to be related
by permutations. While the A and C sets of values agree at finite size, the period matrix
corresponding to B gives numerical values different from the other two. This fact is related
with the terms in Eqs. (5.4) and (5.5) that keep the lattice Hodge operator from being an
endomorphism on lattice harmonic differentials.
While at finite size, modular invariance is slightly broken this does not seem to be the
case in the continuum limit. In Table 6 we see that the theta function ratios extrapolated
to the N limit, agree within a 102 error. Figs. 5 and 6 show the theta function
ratios corresponding to the two different modular choices A and B, plotted as a function
of number of lattice vertices N , together with the determinant ratios. In these figures we
observe that theta ratios for the period matrices A and B seem to converge to a common
value and that, for large enough lattice size, the values of the nine nonvanishing determinant
ratios are bounded by two sets of the theta ratios. Qualitatively the pattern observed in these
plots leads us to believe that the procedure described on the previous section to evaluate
holomorphic differentials and the period matrix has a well defined continuum limit.
A quantitative analysis is given in Tables 3 and 4 where the period matrices evaluated
by the procedure above (using the choice A for the homology basis) are compared with the
Table 6
The sixteen theta function ratios for the three period matrices calculations of Table 5, both the largest
size and the extrapolated value are given. The different ratios are rearranged to facilitate comparison
i
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
A: (4, 2, 2, 6, 2)
i (4840)
i ()
0.223711
0.000000
0.000000
0.000000
0.000000
0.813656
0.000027
0.589972
0.000000
0.132858
0.277170
0.186317
0.000000
0.722803
0.053486
1.000000
0.219812
0.000000
0.000000
0.000000
0.000000
0.814718
0.000023
0.594928
0.000000
0.133252
0.271820
0.185259
0.000000
0.728157
0.052031
1.000000
1
3
2
4
9
11
10
12
5
7
6
8
13
15
14
16
B: (4, 2, 2, 6, 2) with ai bi
i (4840)
i ()
0.210138
0.000000
0.000000
0.000000
0.000000
0.817352
0.000014
0.607228
0.000000
0.134120
0.258651
0.182634
0.000000
0.741334
0.048528
1.000000
0.213846
0.000000
0.000000
0.000000
0.000000
0.816342
0.000017
0.602514
0.000000
0.133805
0.263681
0.183640
0.000000
0.736302
0.049852
1.000000
10
9
4
3
13
14
7
8
2
1
12
11
5
6
15
16
C: (2, 4, 2, 2, 6)
i (4840)
i ()
0.223711
0.000000
0.000000
0.000000
0.000000
0.813656
0.000027
0.589972
0.000000
0.132858
0.277170
0.186317
0.000000
0.722803
0.053486
1.000000
0.219812
0.000000
0.000000
0.000000
0.000000
0.814718
0.000023
0.594928
0.000000
0.133252
0.271820
0.185259
0.000000
0.728157
0.052031
1.000000
462
period matrices of Eq. (3.3), obtained from the ratios of adjacency matrices determinants
by numerical fit. There is a remarkable agreement between the two sets of values. The theta
function ratios corresponding to the two period matrices agree with a precision of 103 or
better.
The small numerical differences are associated with the difficulty in extrapolating the
finite size values of the period matrix to the continuum limit. The extrapolated values
given in Tables 3, 4, 5 and 6 are obtained by fitting the size dependence of the period
matrix entries with a quadratic polynomial in N .
As in the toroidal case, the vertical and horizontal coupling constants provide different
weighting to the vertical and horizontal directions. The period matrix for a lattice characterized by the aspect ratio (m1 , m2 , k, n1 , n2 ) and parameter h/v is equivalent to the
period matrix of a lattice characterized by the aspect ratio (m1 , m2 , k, n1 , n2 ) and
a re-scaled parameter h/v. This property is exemplified in the first two sets of values in
Tables 3 and 4 for both the critical Ising model and Dimers. The parameter h/v can therefore be absorbed into a rescaling of the overall vertical/horizontal aspect ratio of the lattice.
7. Conclusions
In this paper we used the Kasteleyn formalism to study Dimers and the critical Ising
models on a genus two lattice. It is the first time that such a study has been done in
higher genus lattices. We found that the determinants of the Kasteleyn adjacency matrices
converge (3.3) to a common term times theta functions of half-integer characteristic. This
result generalizes a thirty year old observation on the asymptotics of Pfaffians of adjacency
matrices on the torus [1012] and gives a new meaning to this convergence.
For the critical Ising model, the dependence of the determinants of adjacency matrices
on the Kasteleyn orientations is exactly the same as the dependence of the determinant of
the Dirac operator, of the corresponding conformal field theory, on the spin structures of
the Riemann surface; both are given in terms of theta functions of half integer characteristics. There is therefore an one to one correspondence between the 4g Kasteleyn clockwise
odd orientations of the lattice and the 4g spin structures of the continuum limit Riemann
surface.
These observations elevate the Kasteleyn formalism from a combinatorial trick to obtain the partition function to a discrete formulation of some of the model conformal field
theory properties. The relation between the lattice determinants and the zeta regulated determinants of the conformal field theory is reminiscent of the RaySinger theorem [18,19]
that relates determinants of the analytical Laplacians on a surface with the determinants of
combinatorial Laplacians on a triangulation of that surface.
In the context of conformal field theory the period matrix is usually seen as a free parameter and, except on the torus, is not related to the thermodynamic limit of a lattice or a set
of particular lattices. We have shown that the period matrix, characterizing the continuum
limit of a lattice, can be understood in terms of a lattice discrete holomorphic structure.
463
An important difference between the genus two case (1.4) and the torus (1.2) is that the
determinants corresponding to the odd characteristic theta functions do not vanish at finite
size and only converge to zero on the N limit. It interesting to notice that on the
toroidal square lattice a completely satisfactory definition of discrete holomorphy at finite
size can be given, a fact that is probably related with the finite size vanishing of the odd
characteristic determinant.
The results we obtained for the genus two case can be readily generalized for arbitrary
genus. Higher genus lattices can be constructed by pasting two or more of the lattices of
Fig. 1, as shown in Fig. 10. The construction of the adjacency matrices and period matrices
can also be done in a similar way to the one described in this paper for genus two.
There are many ways of choosing the boundary conditions of a square lattice to obtain
a higher genus lattice. An important requirement to reproduce the results we obtained is
that the boundary conditions do not destroy the distinction between vertical and horizontal
edges. With the boundary conditions suggested in Fig. 10, the genus g lattice has a total of
3g 1 integer sizes, respectively (M1 , . . . , Mg , K1 , . . . , Kg1 , N1 , . . . , Ng ). These account
for 3g 2 independent aspect ratios, the h/v parameter been absorbed into a combination
of these ratios. These 3g 2 rational parameters cannot cover, or even approximate, all
the complex structures on a genus g Riemann surface, that are parameterized by 3g 3
complex parameters. This may be possible by generalizing the genus two lattice of Fig. 1
from locally square to locally triangular and allowing for different couplings constants in
each sublattice.
Fig. 10. The integer sizes characterizing pair of pants decompositions of higher genus lattices for
genus = 2, 3 and 4.
464
Acknowledgements
This work is partially supported by the Fundao para a Cincia e Tecnologia (Portugal)
under Grant BD 11503 97 and by the National Science Foundation (USA) under Grant
No. DMR-0073058. Both authors profited from useful discussions with P. Pearce and
C. Mercat. B.M.M. is pleased to thank M. Kashiwara and T. Miwa for hospitality at
the Research Institute of Mathematical Sciences where part of this work was completed.
R.C.-S. thanks Rui Sousa for his help with the numerical computations.
Appendix A. Kasteleyn combinatorics

In this appendix we derive (2.6), the Pfaffian expansion of the partition function. We will
show that from the Pfaffians of the sixteen Kasteleyn adjacency matrices Pf(Ai ) we can
obtain the partition functions of Ising and Dimers for sixteen different choices of boundary
conditions on the lattice of Fig. 1. Let Z(0000) denote the partition function of Dimers or
Ising on the lattice of Fig. 1, then
Zi = Z(na 1 , nb1 , na 2 , nb2 )
(A.1)
with i = 16 8na 1 4nb1 2na 2 nb2 and nx = 0, 1, for x = na 1 , nb1 , na 2 , nb2 , represent
the partition functions of the same model with coupling constants or dimer weights
multiplied by 1 along a choice of the nx = 1 representatives of the first homology group.
This is a labeling similar to the one introduced for adjacency matrices in (2.5).
We will show that there is a matrix bij such that
g
Zi =
j =1
bij Pf(Aj )
and
Pf(Ai ) =
bij Zj .
(A.2)
j =1
The property that b2 = 1 was first noticed for the torus by Fradkin and Shteingradt [44].
This places the partition functions and the Pfaffians of the adjacency matrices on equal
footing. For both models the partition function Z(na 1 , nb1 , na 2 , nb2 ) is independent of the
choice of representatives of the cycles ak , bk . In the Ising model it is well known that
a deformation of a disorder loop does not alter the partition function since a disorder
loop that crosses a vertex corresponds to an interchange of the up and down spin at that
vertex. For the Dimer problem the invariance of the partition function Z(na 1 , nb1 , na 2 , nb2 )
follows from (A.2) and from the invariance of the determinants of adjacency matrices under
disorder loop deformation.
To keep the paper self contained we start by giving a short overview of the Kasteleyn
formalism in higher genus lattices following a combination of references [69]. By a
genus g lattice we mean a graph, not necessarily regular, that can be drawn without
superposition of edges only in a surface of genus g or higher. The Ising model decorated
lattice, Fig. 3, has crossing edges and cannot be considered a genus two lattice. To apply
the results of this appendix to the Ising model it is preferable to consider the Fisher six
vertices decoration [45] instead of the Kasteleyn four vertices decoration, see Fig. 11. The
465
Fig. 11. The Kasteleyn four vertices decoration and the Fisher six vertices decoration for the Ising
model with four near neighbors.
Fisher decorated version of our lattice can be embedded on a genus two surface without
superposition of edges. Determinants of adjacency matrices of the two lattices are equal [7].
We will use the Fisher lattice on the formal discussion and the Kasteleyn lattice for actual
numerical calculations.
Recall the definition of the Pfaffian of antisymmetric matrix, Eq. (2.4)

1
1
2 N
Ap1 p2 Ap3 p4 ApN1 pN , (A.3)
$
Pf(A) = N/2
p1 p2 pN
2 (N/2)! p
where p is a permutation of the integers from 1 to N and $p = 1 for even and odd
permutations respectively. The normalization factor 2N/2 (N/2)! accounts for the fact that
permutations differing by the order of the indices in a Api pi+1 factor or by the interchange
of two such factors give equal contributions to the Pfaffian. We will call two permutations
related in such a way equivalent permutations.
A nonzero Pfaffian term Ap1 p2 Ap3 p4 ApN1 pN corresponds to a choice of edges Cp =
{(p1 , p2 ), . . . , (pN1 , pN )}, where no edges share a common vertex and all vertices are
included; (i, j ) representing the edge between vertices i and j . This choice of edges
is a dimer configuration with weight equal, in absolute value, to the Pfaffian term.
Dimer configurations are therefore in one to one correspondence with sets of equivalent
permutations with nonzero Pfaffian term.
To express the dimer partition function in terms of Pfaffians of adjacency matrices we
need to choose edge orientations such that all the terms in the Pfaffian have the same
relative sign or a linear combinations of Pfaffians with this property. The relative sign
between Pfaffian terms corresponding to different permutations p and p can studied by
considering their product

1
2 N
1
2 N
$
$

p1 p2 pN
p1 p2 pN
Ap1 p2 ApN1 pN A p p A p
1 2

N1 pN
(A.4)
where A and A represent the same adjacency matrix, the prime being introduced for
convenience. This product has a simple graphical interpretation: it corresponds to the
superposition diagram of the two dimer configurations Cp and Cp . The superposition
466
diagram is the set of double edges and even length cycles obtained by drawing both dimer
configurations on the lattice, see Fig. 12 for an example. The even length cycles have
alternating edges belonging to each one of the two dimer configurations and are called
transition cycles.
The product (A.4) can always be rewritten in a form that resembles the superposition
diagram [6]

i i iq iq+1 iq+2
Ai1 i2 A i2 i3 Ai3 i4 A i4 i5 A iq i1
$ 1 2
i2 i3 i1 iq+2 iq+1

Aiq+1 iq+2 A iq+2 iq+1 ( ) ,
where we used the fact that the sign of a product of permutations is the product of the sign
of each permutation and we decomposed the overall permutation into cyclic permutations
of even length. We also chose equivalent permutations to p and p and rearranged the
matrix elements in such a way that the terms corresponding to the double edges and the
even length cycles of the superposition diagram are singled out. The equation above is a
possible example where one cycle of 2p length and a double edge are shown.
From this formula we can read out the relative sign of two Pfaffian terms. The sign of the
overall permutation is (1)s where s is the total number of even length cycles, including
Fig. 12. Two dimer configurations: the standard dimer configuration C0 and an arbitrary dimer
configuration C. The overlap of the two configurations with double edges removed classifies C to be
of type T (0102) = T (eoee).
467
double edges, in the superposition diagram. From the double edges Aij A j i we get an
additional minus sign since A is antisymmetric, while each transition cycle contributes
with a sign given by (1)p where p is the number of edges in that cycle oriented, say, in
the clockwise direction.
Following the notation of reference [8] let C C denote the set of transition cycles
resulting from the overlap of the dimer configurations C and C with the double edges
removed. Then the above discussion motivates the following result due to Kasteleyn [6]
Theorem 1. Let sign(C) be the sign of the terms in the Pfaffian corresponding to a given
dimer configuration C. Then for any two dimer configurations C and C we have that

sign(C) sign(C ) = (1)p( )+1
(A.5)
where the product runs over all transition cycles in C C and p( ) is the number of
clockwise oriented edges in the cycle .
We will call the overlap operator. If dimer configurations are seen as sets of edges
then
C1 C2 = (C1 C2 )\(C1 C2 )
is the symmetric difference of the two sets. Since superpositions of configurations will
play a major role in the following discussion it is worth to collect some properties of the
overlap operator . The following properties follow directly from the definition, the Ci
being dimer configurations or more generally sets of edges
C1 C2 = C2 C1 ,
C1 C1 = ,
C1 = C1 ,
C1 (C2 C3 ) = (C1 C2 ) C3 .
(A.6)
We can use this overlap operator to classify the dimer configurations, and the
corresponding Pfaffian terms, into equivalence classes or types. We start by choosing a
standard dimer configuration C0 , the standard configuration for our lattice being shown in
Fig. 12. Let ai , bi with i = 1, . . . , g, be a canonical basis of the first homology group on the
lattice then we say that C is of type T (nb1 , na1 , . . . , nbg , nag ) if there are nx topologically
nontrivial cycles of kind x = ai , bi , present in the overlap C0 C. An example for our
genus two lattice is show in Fig. 12 where a dimer configuration C is seen to be of type
T (0102).
We are then classifying C by the homology class of the overlap C0 C, two dimer
configurations C and C being of the same type if (C0 C) = (C0 C ) X where X is
a boundary, that is, a set of closed cycles with trivial homology. Since the overlap operator
eliminates double edges it can only distinguish homology classes modulo two. For
instance in Fig. 13 we show how a T (2001) superposition can be obtained from a T (0001)
468
Fig. 13. An overlap of type T (2001) can be built by overlapping a type T (0001) with a boundary X
and therefore both belong to the same type T (eeeo).
superposition by overlapping a homologically trivial boundary X. We will therefore take

nx = e, o, where were e stands for even and o stands for odd.
The relevance of these equivalence classes is that the relative sign of a configuration
C to the standard configuration C0 for a clockwise odd edge orientation depends only on
the equivalence class of C. Clockwise odd edge orientations were defined in Section 2
to be orientations such that every lattice face has an odd number of clockwise oriented
edges along its boundary (the clockwise orientation is a convention, anticlockwise odd
edge orientations would work equally well). These edge orientations have the following
property, found by Kasteleyn [5].
Theorem 2. For an edge orientation such that all lattice faces have an odd number
of clockwise oriented edges we have that sign(C) sign(C ) = 1 for any pair of dimer
configurations C and C such that C C has a trivial homology.
A complete proof of this theorem can be found in Refs. [6,7] but the general idea is very
simple: if C C has a trivial homology its transition cycles can be built by the successively
overlap of elementary lattice faces with the operator. It can then be shown that any such
transition cycle will have clockwise odd parity and contribute with a minus sign in the
Pfaffian term that cancels the minus sign of the corresponding cyclic permutation.
469
We can now state the main result of the higher genus Kasteleyn formalism, whose proof
is so simple that we present it here.
Theorem 3. The relative sign of a dimer configuration C to the standard configuration C0
in a clockwise odd orientation depends only on the homology class modulo two of the set
of transition cycles C0 C.
Proof. Let C and C be two dimer configurations such that C0 C and C0 C belong
to the same homology class. Then by definition we must have (C0 C) = (C0 C ) X
were X is a set of homologically trivial cycles. Using (A.6) we find that X = C C
and it follows from Theorem 2 that, in a clockwise odd orientation, we have sign(C)
sign(C ) = 1.
In a genus g lattice there are 4g classes of dimer configurations, the number of
elements of the homology group H1 (G, Z2 ), that in general will have different signs in
a given clockwise odd orientation. Conversely there are also 4g inequivalent Kasteleyn
orientations, the number of elements of the cohomology group H 1 (G, Z2 ) and it is possible
to find a linear combination of the corresponding 4g Pfaffians of adjacency matrices such
that all dimer configurations have the same overall sign.
More precisely: two edge orientations are said to be equivalent if they are related by
a sequence of local transformations in which we reverse the edge orientation of all the
edges coming to a given vertex. It is clear that each local transformation is equivalent to
the multiplication of the ith column and row of A by 1 and will not affect the parity p( )
of the transition cycles or the absolute value of det A. We then have the following result
Theorem 4. In a genus g lattice there are 4g inequivalent Kasteleyn edge orientations.
For our genus two lattice it is easy to see that there are only sixteen inequivalent
Kasteleyn orientations. Consider the lattice as drawn in Fig. 14. We can choose the edge
orientation on all the edges of a spanning tree of the lattice (heavy line) using the freedom
given by the local equivalence transformations. Then the condition that the elementary
faces must be clockwise odd will fix the orientation on most edges (the dashed line edges)
but there will be 4 edges (rectangular edges in the figure) in which the orientation is
undetermined. The sixteen different choices of orientations in these edges will correspond
to sixteen inequivalent Kasteleyn orientations. Using a similar reasoning the reader can
convince her/himself that the theorem is true for any lattice.
In Section 2 we constructed the sixteen edge orientations for our genus two lattice by
starting from an initial orientation and introducing disorder loops along the a i , bi cycles.
This procedures makes it clear the connection between the edge orientations and the the
cohomology group H 1 (G, Z2 ).
In order to obtain the linear combination of Pfaffians that yields the partition function
we still need to determine the relative sign that the different dimer configuration types,
T (nb1 na1 nb2 na2 ) with nx = e, o, take in the various clockwise odd configurations,
A(ma 1 mb1 ma 2 mb2 ) with mx = 0, 1.
470
Fig. 14. Graphical proof of the existence of 16 independent clockwise odd orientations of the genus
two lattice. The lattice is redrawn with oriented edge identifications given by the letters a to d. The
full line gives a spanning tree on the lattice edges were, using the local equivalence transformation,
we can choose the orientation. On the dashed edges the orientation is determined by the clockwise
odd condition and on the four rectangular edges the orientation is not determined. This gives a total
of 24 independent clockwise odd orientations.
Dimer configurations of type T (eeee) will have the same sign on all clockwise
odd orientations since its overlap with the standard configuration C0 is topologically
trivial. The sign of a general configuration of type T (nb1 na1 nb2 na2 ) under the edge
orientation A(0000) can be evaluated by inspection on Figs. 2 and 4 after choosing a
representative overlap diagram of each type, remembering that we are considering all the
sizes (M1 , M2 , K, N1 , N2 ) to be even integers. This is the last column of Table 7. Once
these signs are know the remaining signs on the Table 7 can be evaluated in a simple way:
a relative sign of a configuration changes with the introduction of a disorder loop if that
471
Table 7
The relative signs for each dimer configuration type in the various clockwise odd edge orientations.
The linear combination of the orientations that gives the same sign to all configurations is shown in
the last row
A1 A2 A3 A4 A5 A6 A7 A8 A9 A10 A11 A12 A13 A14 A15 A16
(eeee)
(eeeo)
(eeoe)
(eeoo)
(eoee)
(eoeo)
(eooe)
(eooo)
(oeee)
(oeeo)
(oeoe)
(oeoo)
(ooee)
(ooeo)
(oooe)
(oooo)
sign
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
Table 8
The b matrix relating the sixteen partition functions with the Pfaffians of the sixteen clockwise odd
orientations. Each Zi is given by 14 of the linear combination of the Pj in the corresponding row
Z1
Z2
Z3
Z4
Z5
Z6
Z7
Z8
Z9
Z10
Z11
Z12
Z13
Z14
Z15
Z16
P1
P2
P3
P4
P5
P6
P7
P8
P9
P10
P11
P12
P13
P14
P15
P16
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
472
configuration has a nontrivial transition cycle intersecting that disorder loop.

Once all the relative signs are known it is simple to find the linear combination of
Pfaffians that will give the same overall sign to all dimer configurations and we obtain
Eq. (2.6) and the last row of the matrix bij in Eq. (A.2). Eq. (2.6) can be rewritten using
the notation of (2.5) as

1
nb na1 nb na2 Pf A(nb1 na 1 nb2 na 2 )

Z(0000) =
(A.7)
1
2
4
nX =0,1
with some choice of i = 1. The Pfaffian expansion of the remaining partition functions
Zi can be obtained form this equation by successive inversions 0 1, 1 0 on the
same entry nx on both the Z() and the A() terms and noticing that this correspond to a
permutation of the i . In this way we can construct the full matrix bij that for our lattice,
and labeling of the orientations, is given in Table 8.
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]
B. Kaufman, Phys. Rev. 76 (1949) 1232.

R.B. Potts, J.C. Ward, Prog. Theor. Phys. 13 (1955) 38.
P.W. Kasteleyn, J. Math. Phys. 4 (1963) 287.
V.N. Plechko, Sov. Phys. Dokl. 30 (1985) 271.
P.W. Kasteleyn, Physica 27 (1961) 1209.
P.W. Kasteleyn, in: F. Harary (Ed.), Graph Theory and Theoretical Physics, Vol. 43, Academic
Press, New York, 1967.
B.M. McCoy, T.T. Wu, The Two-Dimensional Ising Model, Harvard University Press,
Cambridge, MA, 1973.
N.P. Dolbilin, A.S. Mishchenko, M.A. Shtanko, M.I. Shtogrin, Y.M. Zinoviev, Func. Anal. and
Applic. 30 (1996) 163.
G. Tesler, J. Comb. Theor. B 78 (2000) 198.
A.E. Ferdinand, J. Math. Phys. 8 (1967) 2332.
A.E. Ferdinand, M.E. Fisher, Phys. Rev. 185 (1969) 832.
C. Nash, D. OConnor, Phys. Rev. Lett. 76 (1996) 1196.
D. Mumford, Tata Lectures on Theta, Birkhuser, Boston, 1983.
L. Alvarez-Gaum, G. Moore, C. Vafa, Commun. Math. Phys. 106 (1986) 1.
L. Alvarez-Gaum, J.B. Bost, G. Moore, P. Nelson, C. Vafa, Commun. Math. Phys. 112 (1987)
503.
E. Verlinde, H. Verlinde, Nucl. Phys. B 288 (1987) 357.
R. Dijkgraaf, E. Verlinde, H. Verlinde, Commun. Math. Phys. 115 (1988) 649.
D.B. Ray, I.M. Singer, Adv. in Math. 7 (1971) 145.
W. Muller, Adv. in Math. 28 (1978) 233.
C. Nash, D. OConnor, Ann. Phys. 273 (1999) 72.
J.G. Brankov, V.B. Priezzhev, Nucl. Phys. B 400 [FS] (1993) 633.
W.T. Lu, F.Y. Wu, Phys. Lett. A 259 (1999) 108.
W.T. Lu, F.Y. Wu, Phys. Rev. E 63 (2001) 26107.
F. Lund, M. Rasetti, T. Regge, Commun. Math. Phys. 51 (1976) 15.
T. Regge, R. Zecchina, J. Math. Phys. 37 (1996) 2796.
T. Regge, R. Zecchina, J. Phys. A 33 (2000) 741.
N. Behera, R.P. Malik, R.K. Kaul, Phys. Rev. D 40 (1989) 1993.
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
473
B. Eckmann, Commen. Math. Helvetici 17 (1944) 240.

J. Dodziuk, Am. J. of Math. 98 (1976) 79.
J. Dodziuk, V.K. Patodi, J. Indian Math. Soc. 40 (1976) 1.
H.M. Farkas, I. Kra, Riemann Surfaces, Springer, New York, 1992.
C. von Westtenholz, Differential Forms in Mathematical Physics, North Holland, Amsterdam,
1981.
H. Whitney, Geometric Integration Theory, Princeton University Press, Princeton, 1957.
J. Jin, The Finite Element Method in Electrodynamics, Wiley, New York, 1993.
P.R. Kotiuga, Hodge decompositions and computational electromagnetics, Ph.D. Thesis,
McGill University, Montreal, Canada, 1984.
A. de la Bourdonnay, S. Lala, in: Proceedings of ECCOMAS96, Numerical Methods in
Engineering, Vol. 557, Wiley, Paris, 1996.
T. Tarhasaari, L. Kettunen, A. Bossavit, IEEE Trans. Magn. 35 (1999) 1494.
F.L. Teixeira, W.C. Chew, J. Math. Phys. 40 (1999) 169.
R. Hiptmair, Report No. 126, SFB 382, Universitt Tbingen, 1999.
B.M. McCoy, J.H.H. Perk, T.T. Wu, Phys. Rev. Lett. 46 (1981) 757.
C. Mercat, Holomorphie discrte et modle dIsing, Ph.D. Thesis, Univ. Louis Pasteur,
Strasbourg, France, 1998.
C. Mercat, Commun. Math. Phys. 218 (2001) 177.
M.N. Barber, I. Peschel, P.A. Pearce, J. of Stat. Phys. 37 (1984) 497.
E.S. Fradkin, D.M. Shteingradt, Nuovo Cimento 47 (1978) 115.
M.E. Fisher, J. Math. Phys. 7 (1966) 1776.

Bound states and glueballs in three-dimensional

Ising systems
M. Caselle a , M. Hasenbusch b,1 , P. Provero c,a , K. Zarembo d,e
a Dipartimento di Fisica Teorica dellUniversit di Torino and Istituto Nazionale di Fisica Nucleare,
Sezione di Torino, via P. Giuria 1, I-10125 Torino, Italy

b Humboldt Universitt zu Berlin, Institut fr Physik, Invalidenstr. 110, D-10115 Berlin, Germany
c Dipartimento di Scienze e Tecnologie Avanzate, Universit del Piemonte Orientale, Alessandria, Italy
d Department of Physics and Astronomy, University of British Columbia, Vancouver, BC V6T 1Z1, Canada
e Institute of Theoretical and Experimental Physics, B. Cheremushkinskaya 25, 117259 Moscow, Russia
Received 23 March 2001; accepted 13 December 2001
Abstract
We study the spectrum of massive excitations in three-dimensional models belonging to the Ising
universality class. By solving the BetheSalpeter equation for 3D 4 theory in the broken symmetry
phase we show that recently found non-perturbative states can be interpreted as bound states of the
fundamental excitation. We show that duality predicts an exact correspondence between the spectra
of the Ising model in the broken symmetry phase and of the Z2 gauge theory in the confining phase.
The interpretation of the glueball states of the gauge theory as bound states of the dual spin system
allows us to explain the qualitative features of the glueball spectrum, in particular, its peculiar angular
momentum dependence. 2002 Published by Elsevier Science B.V.
PACS: 05.50.+q; 11.10.Kk; 75.10.Hk
1. Introduction
The concept of universality is one of the fundamental principles in the theory of
critical phenomena. According to the universality hypothesis, the long-range behavior
of a statistical system near a phase transition is governed by fluctuations of order
parameters and is essentially determined by symmetries, no matter how complicated the
E-mail addresses: caselle@to.infn.it (M. Caselle), martin.hasenbusch@desy.de (M. Hasenbusch),
provero@to.infn.it (P. Provero), zarembo@physics.ubc.ca (K. Zarembo).
1 Current address: NIC/DESY Zeuthen, Platanenallee 6, D-15738 Zeuthen, Germany.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 4 4 - 7
M. Caselle et al. / Nuclear Physics B 623 [FS] (2002) 474492
475
underlying microscopic dynamics is. For systems with Z2 symmetry universality leads to
the description in terms of the one-component 4 field theory (see, e.g., Refs. [1,2]).
It is known that an order parameter in three-dimensional systems does not fluctuate
strongly and the mean-field approximation (Landau theory) works reasonably well. The
systematic methods that start from the mean-field approximation and perturbatively take
into account fluctuative corrections give very accurate predictions for universal quantities
such as critical exponents. However, a detailed numerical study of correlation functions
in the scaling regime of various systems with Z2 symmetry revealed deviations from the
mean-field approximation which are not too large numerically, but at the same time have no
qualitative explanation within perturbation theory. These deviations arise in the hierarchy
of length scales in the phase with broken Z2 symmetry.
The only macroscopic scale in the problem, of course, is the correlation length .
A generic correlator decreases as er/ at infinity, but there are some correlation
functions that decrease more rapidly. Composite operators in the Ising model whose
correlation functions decay faster than er/ were constructed in [3,4] and include, as
typical examples, operators of non-zero angular momentum. What are the length scales
associated with these operators? The answer is simple in the Gaussian approximation,
when correlators factorize: a correlation function of composite operators that factorizes
on n irreducible correlators behaves as (er/ )n . The characteristic decay rate of any
correlation function is therefore an integer multiple of the inverse correlation length.
The fact that fluctuations are not exactly Gaussian should not change this conclusion.
The ratio of the correlation length to any other length scale is an integer to any order
in perturbation theory! This can be easily understood from the field theory perspective.
The behavior of a correlation function at infinity is determined by the singularities of its
Fourier transform in the complex momentum plane. These singularities are associated with
particles described by the quantum field. The only singularities that arise in perturbative
4 theory are the single-particle pole associated with the elementary quantum and the
multiparticle cuts. Correlation functions of operators that couple to the single-particle
state decay as er , where is the inverse correlation length, or the particle mass.
Correlation functions of operators that have zero overlap with the one-particle state
(because of the angular momentum conservation, for instance) do not have poles in the
complex momentum plane. Their only singularities are multiparticle branch cuts that start
at p2 = n2 2 . Consequently, such correlators decay as enr with some integer n.
The numerical simulations clearly show that the above picture does not hold in the phase
with broken Z2 symmetry. The firmest evidence comes from the existence of a spin zero
state with mass just below the two-particle threshold: M 1.8, strongly supported by
numerical results [35]. This and other non-perturbative states give visible contributions
to the quantities which are sensitive to distances of order of the correlation length. For
instance, the spinspin correlator in the Ising model near the critical point cannot be fitted
by a simple pole-plus-cut structure suggested by perturbation theory and definitely contains
an extra pole below the cut [3].
The simulations show that dimensionless ratios like M/ become constant when
approaching the critical temperature. Moreover, they have the same value within statistical
errors in two different realizations of the universality class: the Ising model itself and the
lattice regularized 4 theory. These two facts allow us to be confident about the universal
476
character of these effects. Specifically, we expect the same spectrum to appear in all
realizations of the universality class, as long as one moves away from the critical point
by perturbing with a Z2 -symmetric operator towards the region where the symmetry is
spontaneously broken.
To resolve the aforementioned contradiction, we proposed to interpret the nonperturbative states in the broken symmetry phase of 4 theory as bound states of
elementary quanta of the scalar field [4]. We shall develop this idea further in this paper.
Let us briefly recollect some facts about bound states. Two-particle bound states
show up as poles in the Fourier transform of correlation functions below the branch cut
associated with the two-particle threshold. In three dimensions, the threshold singularity is
logarithmic, and, though the two-particle cut comes from one-loop Feynman diagrams and
thus is accompanied by a coupling constant, the smallness of the coupling is compensated
by the large logarithm. In the vicinity of the threshold, perturbation theory breaks
down. Fortunately, the large logarithms can be resummed. This resummation substantially
changes an analytic structure of correlation functions near the threshold and produces a
pole that is invisible in perturbation theory. This pole is associated with a two-particle
bound state. Its position is determined by the BetheSalpeter (BS) equation, which can be
solved order by order in perturbation theory.
In this work we analyze the BS equation for 3D 4 theory in the broken symmetry
phase. The analysis shows that a bound state of the elementary quanta indeed exists. We
compute its binding energy at the next-to-leading order. While the value we obtain at the
leading order is in very good agreement with the Monte Carlo data, the series appear to be
badly behaved, so that a reliable numerical estimate of the subleading corrections to the
binding energy would require the knowledge of a large number of terms in the series.
The 3D Z2 gauge theory is related to the Ising model by an exact duality transformation.
Since the duality transformation maps the confining phase of the gauge theory into the
broken symmetry phase of the spin model, all of our considerations apply to the glueball
spectrum of the gauge theory as well. Indeed, we shall prove that duality implies the exact
equality between the spectra of the spin model in the broken symmetry phase and the gauge
theory in the confining phase for all values of the coupling.
This correspondence suggests a novel interpretation of the glueballs of the gauge
theory as bound states of the dual spin model, which in turn allows one to understand
the qualitative features of the glueball spectrum, and in particular its peculiar angular
momentum dependence. Bound states with different values of the angular momentum
are easily seen to appear in the low-temperature expansion of the 3D Ising model.
Simple group-theoretic arguments determine how many elementary quanta are necessary
to construct a bound state of any given angular momentum and parity. Assuming that the
binding energy is always much smaller than the mass of the elementary components, and
that this description of the spectrum in terms of bound states is conserved when one moves
from the deep low-temperature region to the scaling region, one ends up with a prediction
for the qualitative features of the glueball spectrum in Z2 gauge theory, and in particular
for the angular momentum dependence of the various masses. This prediction is in very
good agreement with the numerical data for the glueball spectrum, thus lending support to
the interpretation of glueballs as bound states of the dual spin model.
477
The rest of the paper is organized as follows. In Section 2 we will present the detailed
computations for the bound state energy in the framework of the BS equation for 3D 4
field theory; in Section 3 we study the dependence of the binding energy on the magnetic
field. Section 4 contains a proof of the exact correspondence between the spectrum of the
3D Ising model in the broken symmetry phase and the Z2 gauge theory in the confining
phase. In Section 5 we describe in detail the argument, already presented in [4], that allows
us to explain the angular momentum dependence of the glueball spectrum of the Z2 gauge
theory using the dual bound state picture. In Section 6 we discuss obtained results and
outline possible future developments.
2. Bound states in 3D 4 : BS equation

We consider 3D 4 theory in the broken symmetry phase. The Euclidean action is

1
g 2
3
2 2
S = d x +
(1)
.
v
2
4!
The field acquires an expectation value and the perturbative expansion is performed
around the stable vacuum = v. Extracting the fluctuating part,
= v + ,
(2)
and setting
gv 2
3
we write the action as

1
1
1
1
S = d 3 x + m2 2 +
3gm2 3 + g 4 .
2
2
3!
4!
m2 =
(3)
(4)
The quartic coupling g has a dimension of mass, so the dimensionless expansion parameter
is the ratio g/m.
If the coupling is weak, the bound state lies very close to the threshold. We shall see
that the binding energy is exponentially small in g/m. While the exponent can be easily
found [4], computing the pre-exponential factor requires sufficiently complicated next-toleading order calculations. The calculations are to certain extent similar to those of Ref. [6]
where the BS equation was solved to the first few orders in perturbation theory for (1 + 1)dimensional scalar field theory with the cosine interaction.
2.1. Leading order solution
The two-particle bound state manifests itself as a pole in the four-point function. The
pole arises after summation of the ladder diagrams (Fig. 1(a)). Each ladder diagram
develops a kinematic singularity when the momentum flowing through the diagram is near
the two-particle threshold: p2 4m2 , since the momentum integration in each loop then
contains a region where two of the intermediate states simultaneously go on-shell and
478
(a)
(b)
Fig. 1. (a) Ladder diagrams. (b) BS equation.
Fig. 2. Tree-level BS kernel.
the propagators in the loop integral simultaneously have poles. This singularity produces
large logarithms that compensate for powers of the coupling, so near the threshold all
ladder diagrams are of the same order. The vertices in Fig. 1(a) correspond to the sum of
the two-particle irreducible (2PI) diagrams, that is, the diagrams that cannot be split into
disconnected pieces by cutting two internal lines whose total momentum is p. 2PI diagrams
are analytic at the two-particle threshold in the s-channel.
The position of the pole is determined by the BS equation, graphically represented in
Fig. 1(b). The BS equation is the homogeneous part of the Dyson equation for ladder
diagrams [7]:

d 3q
(q, p q)G(q)K(q, p q; p1, p2 )G(p q),
(p1 , p2 ) =
(2)3
p = p1 + p2 .
(5)
Here, G is the propagator and K is the BS kernel, the aforementioned sum of 2PI diagrams.
The BS equation is a linear eigenvalue problem for the wave function and the bound state
mass:
M 2 p2 .
(6)
We take:
M = (2 )m,
(7)
where is small.
To the leading order in g/m, the BS kernel is given by the sum of three diagrams in
Fig. 2. The diagram a contributes g/2 to the kernel:
g
Ka = .
2
The diagram b gives
Kb =
1
3gm2
g
= + O().
2 m2 M 2
2
479
The contribution of the diagrams c and d depends on the momentum transfer in the t and u
channels. However, the momentum transfer is very small when external momenta and both
of the propagators in (5) are on shell: simple kinematic arguments show that the typical
momentum in the t or u channel propagators is of order m. To the first approximation,
we can just set it to zero:
Kc =
1
3gm2
3g
,
2 (p1 q)2 + m2
2
Kd =
1
3gm2
3g
.
2 (p2 q)2 + m2
2
Collecting all three terms together, we get:

K = 2g.
(8)
The kernel corresponds to a local four-point interaction in this approximation. The effective
quartic coupling appears to be negative. This happens because the exchange diagram c
dominates over the pure four-point vertex a and the annihilation graph b, so the interaction
is attractive. No matter how weak the interaction is, it will bind elementary quanta into
bound states.
Since the binding energy is small, the non-relativistic approximation should be valid
at small g/m. The non-relativistic limit of the BS equation with the positive local kernel
is the two-body Schrdinger equation with attractive -function potential. This quantummechanical problem has a bound state at any value of the coupling with exponentially
small binding energy [8]. An approach based on the non-relativistic approximation was
elaborated in [4]. This approach is simple and physically transparent, but it only allows us
to find the binding energy up to a numerical factor. Here we follow a more systematic route
based on the full BS equation.
The BS equation with the local kernel Eq. (8) corresponds to summing an infinite series
of bubble diagrams, and can be written as

d 3q
1
1 = 2g
(9)
.
(2)3 (q 2 + m2 )[(p q)2 + m2 ]
Evaluating the integral,

p2

arcsin
4m2 +p 2
ln 2m+M
d 3q
1
2mM
=
=
8M
(2)3 (q 2 + m2 )[(p q)2 + m2 ]
4 p2
ln(4/)
=
+ O( ln ),
16m
we obtain:
ln(4/)
.
16m
This equation predicts an exponentially small binding energy:

8m
= const exp
,
g
1 = 2g
(10)
(11)
(12)
480
in accord with the result obtained in the non-relativistic approximation [4]2.

The normalization of the binding energy cannot be deduced from Eq. (11). We cannot
keep the constant term in (10), because it has the same magnitude as logarithmic terms in
the next order of perturbation theory, which are proportional to g 2 /m2 ln g/m. There
are other sources of corrections of order g/m, as well. We need to compute all of them to
fix the normalization of the binding energy.
2.2. Corrections to propagator
One-loop corrections to the propagator are given by three diagrams in Fig. 3. They shift
the pole of the propagator to
3gm
ln 3 g,
(13)
16
where by we denote the linearly divergent integral

d 3q
1
(14)
.
(2)3 q 2 + m2
The mass renormalization given by Eq. (13) is sufficient to make all one-loop
Feynman diagrams finite. Therefore we can use the renormalization scheme in which
correlation functions are expressed in terms of the renormalized (physical) mass and
the bare coupling g. In Subsection 2.6 we will translate the final result into a different
renormalization scheme, more suitable for comparison with numerical data.
Expanding the third diagram in Fig. 3 near the mass shell, we get:

2
2
9
g
1
2
+ O p 2 + 2
G (k) = p + 1 + 3 ln 3
(15)
4
24
and

g
9
1
+ O(1).
ln
3
1
G(k) = 2
(16)
p + 2
4
24
In evaluating the integral (10), the corrections to the propagator will result in the
replacement of the bare mass m by the renormalized (physical) mass and in the
multiplication of the whole integrand by the wave function renormalization factor.
Expressing everything in terms of the physical mass gap (13), we set:
2 = m 2
M = (2 ).
(17)
This definition affects the BS kernel, since the contribution of the diagram b in Fig. 2
depends on how exactly we define M:

1
m2
1
3gm2
3g
3g
ln 3 + g 2
=
1
Kb =
2 m2 M 2
2 2 M 2
16
M2
2
2

g
2g
g
ln 3
=
(18)
+ O() + O g 3 .
2
2 8
3
2 The coupling used in [4] differs from g by a factor of 24.
481
Fig. 3. One-loop corrections to the propagator.
This differs from the result based on our previous definition by an amount
Kb =
g 2 ln 3 2g 2
.
8
32
(19)
2.3. One-loop corrections to the kernel

The one-loop corrections to the kernel of the BS equations are given by thirteen
diagrams in Fig. 4. Since we are interested only in the logarithmically enhanced terms
in the BS equation, all these diagrams can be evaluated on shell and at zero momentum
transfer in the t and u channels.
The linearly divergent diagrams 13 sum up to zero:
K13 = 0.
(20)
The diagrams 47 cancel the divergence in the diagram b, the second term in Eq. (19):
K47 =
2g 2
.
32
(21)
All linear divergences eventually have canceled. It is satisfying to see that the net one-loop
contribution to the kernel is finite, as it should be.
The remaining diagrams give:
K813 = (9 6 + 1 + 12 4 + 2)
g2
g2
=7
.
16
8
(22)
Collecting together the contributions (19)(22) we get the one-loop correction to the BS
kernel:
K = (7 ln 3)
g2
.
8
(23)
2.4. Momentum dependence

Finally, we consider the most subtle corrections that come from the momentum
dependence of the kernel and of the wave function. Those were neglected in the leading
order calculation because the logarithmic enhancement of the bubble diagram near
threshold comes from the region of integration in (5) where all momenta are almost on
shell. This allowed us to forget about the momentum dependence of the wave function and
to set it to one. In fact, we can always require that the wave function is equal to the one on
482
Fig. 4. One-loop two-particle-irreducible diagrams.
shell. This can be regarded as a normalization condition. But to compute non-logarithmic

terms we need to know the complete off-shell wave function.
Once the on-shell condition on the external momenta p1 , p2 is relaxed, we can no longer
neglect the momentum transfer in the diagrams c and d in Fig. 2. The BS equation then
gives:

1
d 3q
3gm2
(p1 , p2 ) =
g
(2)3 (q 2 + m2 )[(p q)2 + m2 ] (q p1 )2 + m2

1
g
1
3m2
ln
=
8m
4m2 + p12 + p22 2

1
g
3m2
+
O
.
=
(24)
m
4m2 + p12 + p22 2
2.5. Next-to-leading order BS equation
With all corrections taken into account, the BS equation reads:

2

9
g
32
1
d 3q
1 = 1 3 ln 3
4
24
(2)3 42 + p2 + (p q)2 2

1
g2
3g2
2
g + (7 ln 3)
,
(q + 2 )[(p q)2 + 2 ] (q p1 )2 + 2
8
(25)
2 = 2 . We need to retain terms of

where external momenta p1 and p2 are on shell: p1,2
order g/ and (g 2 /2 ) ln on the right-hand side. For this reason, O(g/m) corrections to
the wave function can be neglected, since they vanish on shell and therefore do not lead to
the logarithmic enhancement. The loop integrals encountered in (25) are listed in Appendix
A and in (10). Using these results, we get after a little of algebra:

1
g
8
+ 2 log 2 log 3 + O
log =
(26)
g
2
483
that is3

8
4
= exp
.
g
3
(27)
2.6. Comparison with numerical results

To compare Eq. (26) with numerical results one has to know a value for the
dimensionless coupling u g/. The safest thing to do is to relate u to the renormalized
coupling uR defined, e.g., as in Ref. [18], whose critical value is precisely known from
Monte Carlo simulations [13]
uR = 14.3(1)
(28)
in terms of uR we have, using results from [3,18]

3 log 3
15
+
u = uR 1 + uR
128
32
(29)
so that
log =

15 log 3
8
+
+ 2 log 2 + O(uR ).
+
uR
16
4
(30)
Therefore, the leading-order result is

log =
8
= 1.76(2)
uR
(31)
corresponding to
= 0.172(3).
(32)
This result is compatible with the assumption of weak binding energy that underlies the
BetheSalpeter formalism, and is in a very good agreement with the numerical result,
MC = 0.17(3).
(33)
However, the parameter of expansion in perturbation theory is comparatively large, and the
NLO correction is not small: the O(1) contribution to log is 2.6, that is actually greater
than the LO contribution. Therefore, a reliable estimate of the subleading corrections to the
binding energy seems to require the computation of a large number of terms in perturbative
series, so as to make a resummation possible, which is typical for application of 3D 4
theory to critical phenomena.
3 Note that the pre-exponential factor quoted in the published version of Ref. [4] is not correct.
484
3. Dependence on the magnetic field

It is possible to study the dependence of the bound state masses on the magnetic field.
Because the magnetic field couples to the order parameter, we know how to introduce it in
the effective description in terms of 4 theory. We can take the magnetic field into account
by adding a linear term to the action:

2
1
g 2
S = d 3 x +
(34)
v 2 h .
2
4!
The magnetic field shifts the minimum of the potential, which now is determined by the
equation

g 2
0 0 v 2 = h.
(35)
6
Expanding the order parameter around 0 ,
= 0 + ,
(36)
we get for the action:

1
1
1
1
S = d 3 x + m2h 2 + g3 3 + g 4 + S0 ,
2
2
3!
4!
where the mass and the trilinear coupling are

g 2
m2h =
30 v 2 ,
6
g3 = g0 .
(37)
(38)
(39)
The sum of the diagrams in Fig. 2 gives the tree-level BS kernel:

Kh =
g32
m2h
702 + v 2
g32
1
g
=
g
.
2 m2h 4m2h 2
602 2v 2
(40)
Repeating the same steps as we used to calculate the binding energy at zero magnetic field,
we get

16mh
= const exp
(41)
,
Kh
where is the dimensionless binding energy: Mh = (2 )mh .
It is useful to introduce the dimensionless variable
0
,
v
which satisfies the equation
=
3 =
6
2
h= 3
3
gv
m
(42)

g
h.
3
(43)
485
The binding energy is expressed in terms of and m, the mass at zero magnetic field, as

8m (6 2 2)3/2
= const exp
(44)
.
g
7 2 + 1
It is easy to see that is a decreasing function of h. Thus the magnetic field loosens the
binding of composite states and shifts their masses closer to the threshold.
4. Duality and the spectrum of Ising systems

The 3D Ising model and Z2 gauge theory are related by an exact duality transformation.
The broken symmetry phase of the spin model is mapped into the confining phase of
the gauge theory. The purpose of this section is to show that, in this phase, the duality
relationship implies an exact coincidence of the spectra of the two theories. The proof
relies on the existence of a non-zero interface tension and therefore applies exclusively to
the broken symmetry phase of the spin model.
There are several books and reviews which discuss duality in spin systems and in
particular in Ising-type models (see, e.g., Ref. [9]). However, the duality transformation
is usually treated in the thermodynamic limit only, while to study the spectrum of the
transfer matrix it is necessary to extend the analysis to finite lattices. This is possible as
long as one carefully takes into account all possible boundary conditions.
The novel feature of our approach, which greatly simplifies the whole analysis, is the
use of the Transfer Matrix (TM) formalism. Let us first recall the definitions of the lattice
models we are concerned with, and the notion of duality in the thermodynamic limit.
4.1. The spin Ising model
The Ising model is defined by the action

sn sm ,
Sspin = spin
(45)
n,m
where the field variable sn takes the values 1 and +1; n (n0 , n1 , n2 ) labels the sites of
a simple cubic lattice of size L0 , L1 and L2 in the three directions. The notation n, m in
Eq. (45) indicates that the sum is taken over pairs of nearest neighbor sites only.
4.2. The gauge Ising model
The building blocks of the Z2 gauge model are the link variables gn; {1, 1}, which
play the role of gauge fields. Denoting by the direction of the link, the action is

Sgauge =
(46)
gn; ,
n,<
where gn; are the plaquette variables, defined by

gn; = gn; gn+; gn+; gn; .
(47)
486
This action is invariant under local Z2 gauge transformations defined as follows: one
chooses arbitrarily a subset of the sites of the lattice and changes signs of all variables
defined on the links which end on these sites (if two neighboring sites belong to the chosen
subset the link that joins them is changed twice, that is not changed). It is immediately clear
that the plaquette values, like any other product of links along a closed path, are invariant
under this gauge transformation. For more details about this model, see, e.g., Ref. [10].
4.3. Duality
There is an exact duality transformation which relates the Ising model and the Z2 gauge
model. This transformation is known as KramersWannier duality. It relates the partition
functions of the two models evaluated at two different values of the coupling constants:
Zgauge() Zspin(),

1
= log tanh() ,
2
(48)
where will be denoted as the dual coupling in what follows.

It is easy to see that low values of are mapped into high values of and vice versa.
Thus the confining region of the gauge theory is mapped into the broken symmetry phase
of the spin model. In particular the end points of these two phases, the deconfinement
transition and the magnetization transition, are mapped into each other.
An important feature of the dual transformation on a lattice of finite size is that it does
not conserve the boundary conditions (BC). In the thermodynamic limit this fact becomes
irrelevant, but on lattices of finite extent it cannot be neglected. In particular, the Z2 gauge
model with periodic BC in all directions is mapped by duality into the Ising spin model
with fluctuating BC, so that the partition function is given by the sum of the partition
functions with all possible choices of periodic (p) or antiperiodic (a) BC:

1
+ Zspin,ppa ()
+ + Zspin,aaa ()
,
Zgauge,ppp () = cV Zspin,ppp ()
(49)
2
where V L0 L1 L2 is the volume of the lattice, and c is a constant which can be easily
evaluated, but is irrelevant for our purposes.
This result is discussed in full generality (for a generic lattice geometry and symmetry
group) in [11]. In the particular case of the Ising model on a cubic lattice it can be easily
obtained by a direct implementation of the duality transformation.
Let us now consider the duality transformation in the framework of the transfer matrix
approach. The 0 direction will be our time. If we choose periodic BC in the 0 direction
without specifying the BC in the 1 and 2 directions we obtain:
Zspin,pxy = tr(Txy )L0 ,
(50)
where Txy denotes the transfer matrix of the model in which x, y {a, p} BC are chosen
in the 1 and 2 directions, respectively.
The antiperiodic BC in the 0 direction can be obtained by acting with a spin-flip operator
P, which changes the sign of all spins in a given time slice. Thus we may write
Zspin,axy = tr P(Txy )L0 .
(51)
487
Since the operators Txy and P commute, they have a common set of eigenfunctions. Let
us denote the eigenvalues of Txy by xy,i and those of P by pi . The possible values of pi
are 1 and 1. States that are symmetric in the magnetization have pi = 1 and those that
are antisymmetric have pi = 1.
Thus we can write
L
0
xy,i
,
Zspin,pxy + Zspin,axy = tr(1 + P)(Txy )L0 = 2
(52)
i,sym
where i,sym means that the sum is restricted to the states that are symmetric in the
magnetization (see Ref. [12] for further details on this type of construction).
Using this result we can write the duality relation in terms of the transfer matrix
eigenvalues:

L
L
L
L
L
L0 L1 L2
0
0
0
0
0
j = c
pp,i +
pa,i +
ap,i +
aa,i ,
(53)
j
i,sym
i,sym
i,sym
i,sym
where j are the eigenvalues of the transfer matrix of the gauge system.
It is instructive to show explicitly that the sums on the two sides of Eq. (53) have the
same number of terms. On the spin side (right hand side of Eq. (53)) we have 2L1 L2 1
terms for each sector. In fact, the transfer matrix is a 2L1 L2 2L1 L2 matrix, but only half of
the eigenvalues fulfill the symmetry requirement. Thus, taking together all the four sectors
we end up with a sum of 2L1 L2 +1 terms. On the gauge side we have 2L1 L2 link variables,
but we may fix the gauge. The maximum number of gauge variables that we may fix, in
order to preserve the periodic BC, is exactly L1 L2 1. This is easy to see: we may fix,
say, all the links in the 1 direction except those in the last row, where we may fix all the
links in the 2 direction, except the last one. Thus, we end up with only L1 L2 + 1 degrees
of freedom left. Hence, the corresponding transfer matrix has 2L1 L2 +1 eigenvalues, that is
exactly the same number as for the spin model.
Since for finite values of L1 and L2 the two sums contain a finite number of terms and
since Eq. (53) holds for any integer L0 , each j has to have an exact (up to an overall factor
cL1 L2 ) counterpart in the left-hand side of the equation. The overall factor cancels when
one takes the ratios of all the eigenvalues to the lowest one to obtain the physical spectrum,
and so the latter coincides in the two models.
However, our goal is to compare the spectra of the gauge theory and the spin model
both with periodic BC in all directions, while Eq. (53) involves a sum over many different
choices of the spin model BC. Therefore, we must find a relationship between the pp,i
eigenvalues and those belonging to the other three sectors. This is easily done by noticing
that in the low temperature phase of the Ising model with antiperiodic boundary conditions
at least one interface has to be created in the system. Therefore, at leading order (neglecting
the cases in which more than one interface appear):
Zppa exp( L0 L1 )Zppp ,
(54)
where is the interface tension.

For the transfer matrix eigenvalues this gives
pp,max exp( L1 )pa,max ,
(55)
488
where, with the notation pp,max we denote the low lying states of the spectrum.4
So we can conclude that for sufficiently large L1 and L2 , the low lying spectrum of the
Z2 gauge theory in the confining phase with periodic boundary conditions coincides with
the symmetric sector of the Ising spin model spectrum in the low temperature phase with
periodic boundary conditions only. (This last argument was already presented in [13].)
5. The dual bound state picture of glueballs in Z2 gauge theory

In this section we show that the interpretation of the spectrum of three-dimensional
models in the universality class of the 3D Ising model as bound states of the elementary
quanta provides an analytical tool to explain the qualitative features of the glueball
spectrum of Z2 gauge theory, and especially its angular momentum dependence.
Consider first the spin Ising model in the low-temperature region. Here the existence of
bound states can be immediately inferred from the diagrammatics of the low-temperature
expansion. This was shown in [14] for the 4D case, where however all bound states are
expected to disappear in the continuum limit due to triviality.
The mechanism is very simple, and is best explained in the transfer matrix formalism.
As discussed in [15,16], to the leading order in the low-temperature expansion of the
transfer matrix, all time slices are forced to have the same configuration of spins, and
the eigenvectors are given by all the possible spin configurations in a single time slice.
The eigenvalues of the Hamiltonian (that is minus the logarithm of the transfer matrix) are
given by the energy of each configuration.
The ground state then corresponds to the configuration in which all the spins on any time
slice are parallel; the first excited state corresponds to one flipped spin, which requires four
bonds to be frustrated. To proceed, one has to flip two spins: if these are chosen in nonneighboring sites, the energy is just twice the one of the first excitation, since eight bonds
will be frustrated. However, one can flip two neighboring spins at the cost of frustrating
six bonds only: the corresponding state is a bound state of the fundamental excitation, with
mass just below the two-particle threshold.
One can also construct states of given angular momentum and parity by choosing linear
combinations of time slice configurations with nontrivial transformation properties under
spatial rotations and parity reflections. This can be done systematically by using standard
group theory results, namely, the theory of representations of the dihedral group D 4 , which
is the relevant group on a square lattice. This analysis is performed in Ref. [17], to which
we refer for details.
The important point is that any given angular momentum requires a certain minimum
number nc of spin flips. This means that a bound state of such angular momentum will be
composed of at least nc elementary quanta. In Table 1 we have reported the values of nc
for the various values of the angular momentum. For example, Fig. 5 shows the simplest
lattice operator corresponding to angular momentum J = 2 and parity P = 1, which
4 Notice that this last argument does not hold in the 2D case, where the anti-periodic eigenvalues are only
suppressed by a constant, independent of the system size.
489
Fig. 5. Three elementary quanta are required to construct a state with quantum numbers 2 . Dots represent flipped
spins, solid lines are frustrated links, dashed lines satisfied links.
Fig. 6. Bound states of nc = 4 elementary quanta.
requires nc = 3 elementary quanta, while Fig. 6 shows how states with quantum numbers
J = 0, 2 and P = +, can be constructed with four elementary quanta. If we assume
that the binding energy is always much smaller that the common mass of the elementary
constituents of the bound states, we end up with a prediction for the angular momentum
dependence of the glueball spectrum of the Z2 gauge theory in the strong coupling regime.
It is not at all guaranteed, of course, that the spectrum will retain these same features
when going from the strong coupling regime to the scaling region: however, the data from
Monte Carlo simulations of the gauge theory show that this is actually the case. The
third column of Table 1 contains the Monte Carlo results for the glueball masses (from
Ref. [17]), normalized to the mass of the lowest glueball. This shows that our dual bound
state picture indeed explains the peculiar qualitative features of the glueball spectrum.
In particular, some characteristic approximate degeneracies in the spectrum find a
natural explanation in this picture: for example the states 0+, and 2 are nearly
degenerate because they are both bound states of nc = 3 elementary constituents. Bound
states of nc = 4 constituents give rise to the 0 and 2, states, again explaining their near
490
Table 1
The bound states that can be constructed with nc elementary quanta are classified
in terms of their angular momentum and parity. The last column shows the mass
of the corresponding glueball state in Z2 gauge theory
nc
1
2
3
4
5
JP
0+
m(J P )/m(0+ )
0+
1
1.88(2)
0+
2
2
0
0
1
2.59(4)
2.59(4)
3.23(7)
3.24(16)
4.48(20)
4.12(17)
degeneracy, and the same applies to 0, and 1 when interpreted as nc = 5 bound states.5
We do not expect these degeneracies to be exact, since there is no reason to expect the
binding energies to be exactly the same for bound states of different angular momentum.
Once the minimum number nc of constituents for a given value of the angular
momentum has been reached, it is easy to see that states with the same angular momentum
can be constructed out of any number n > nc of constituents. This suggests that the
approximate degeneracies just discussed appears between pairs of states only because, in
general, only two states can be detected with numerical methods in each channel. It is very
likely that the degeneracies actually group together larger and larger sets of states as the
number of constituents is increased. For example, we might conjecture that the states 0,
and 1 (nc = 5) are nearly degenerate with the (as yet undetected) states 0+, , 2, .
This degeneracy pattern is a prediction of the dual bound state picture of glueballs.
6. Summary and further developments

The broken symmetry phase of 3D models with Z2 symmetry shows a rich spectrum
of massive excitations, contrary to naive expectations based on the perturbative fieldtheoretical treatment. The non-perturbative states in the spectrum can be interpreted as
bound composites of the elementary particle excitation. When applied to 3D Z2 gauge
theory, this approach gives a nice explanation of the angular momentum dependence of the
glueball spectrum. In particular, an observed rather peculiar degeneracy pattern naturally
arises in the bound-state interpretation.
It would be interesting to consider the contribution of the bound states to universal
amplitude ratios. The obvious way to do that is to use the ladder approximation for
correlation functions, which incorporates all diagrams that blow up near the two-particle
threshold. It would be also interesting to understand if there are bound states in the systems
with N > 1 order parameters. The physics is completely different in this case because
5 The 2+ state could be realized with n = 2, but a general theorem (see Ref. [17]) forces all states with J = 0
c
to be degenerate in parity.
491
the dominating forces are long-range due to the Goldstone bosons of the broken O(N)
symmetry.
Acknowledgements
K.Z. would like to thank P. Simon for discussions. The work of K.Z. was supported by
NSERC of Canada, by Pacific Institute for the Mathematical Sciences and in part by RFBR
grant 98-01-00327 and RFBR grant 00-15-96557 for the promotion of scientific schools.
Appendix A

d 3q
1
1
4
=
ln
.
(2)3 [42 + q 2 + (p q)2 ](q 2 + m2 )[(p q)2 + m2 ] 323
9
(A.1)

1
4
1
d 3q
.
ln
=
9
(2)3 (q 2 + m2 )[(p q)2 + m2 ][(q p1 )2 + m2 ] 163
(A.2)
1
d 3q
(2)3 [42 + q 2 + (p q)2 ](q 2 + m2 )[(p q)2 + m2 ][(q p1 )2 + m2 ]

2
4
1
.
ln
=
(A.3)
9
3
325
2 = 2 , and near the two-particle threshold:

All integrals are calculated on shell: p1,2
p2 (p1 + p2 )2 = M 2 = (2 )2 2 .
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
C. Itzykson, J.M. Drouffe, Statistical Field Theory, Cambridge Univ. Press, Cambridge, 1989.
J. Zinn-Justin, Quantum Field Theory and Critical Phenomena, Clarendon, Oxford, 1989.
M. Caselle, M. Hasenbusch, P. Provero, Nucl. Phys. B 556 (1999) 575, hep-lat/9903011.
M. Caselle, M. Hasenbusch, P. Provero, K. Zarembo, Phys. Rev. D 62 (2000) 017901, hep-th/0001181.
D. Lee, N. Salwen, M. Windoloski, hep-lat/0010039.
R.F. Dashen, B. Hasslacher, A. Neveu, Phys. Rev. D 11 (1975) 3424.
V.B. Berestetsky, E.M. Lifshits, L.P. Pitaevsky, Quantum Electrodynamics, Pergamon, Oxford, 1982.
C. Thorn, Phys. Rev. D 19 (1979) 639.
R. Savit, Rev. Mod. Phys. 52 (1980) 453.
J. Drouffe, J. Zuber, Phys. Rep. 102 (1983) 1.
C. Gruber, A. Hintermann, D. Merlini, Group Analysis of Classical Lattice Systems, Springer-Verlag, Berlin,
1977.
[12] M. Caselle, R. Fiore, F. Gliozzi, M. Hasenbusch, K. Pinn, S. Vinti, Nucl. Phys. B 432 (1994) 590, heplat/9407002.
[13] M. Caselle, M. Hasenbusch, Nucl. Phys. B 470 (1996) 435, hep-lat/9511015.
492
[14]
[15]
[16]
[17]
[18]
M. Lscher, P. Weisz, Nucl. Phys. B 295 (1988) 65.

M.E. Fisher, W.J. Camp, Phys. Rev. Lett. 26 (1971) 565.
W.J. Camp, Phys. Rev. B 7 (1973) 3187.
V. Agostini, G. Carlino, M. Caselle, M. Hasenbusch, Nucl. Phys. B 484 (1997) 331, hep-lat/9607029.
G. Mnster, J. Heitger, Nucl. Phys. B 424 (1994) 582, hep-lat/9402017.

Cluster percolation and first order phase

transitions in the Potts model
Santo Fortunato, Helmut Satz
Fakultt fr Physik, Universitt Bielefeld, D-33501 Bielefeld, Germany
Abstract
The q-state Potts model can be formulated in geometric terms, with FortuinKasteleyn (FK)
clusters as fundamental objects. For vanishing external field, the phase transition of the model can
be equivalently described as a percolation transition of FK clusters. In this work, we investigate
numerically the percolation behaviour along the line of first-order phase transitions of the 3d
3-state Potts model in a non-vanishing external field and find that the percolation strength exhibits a
discontinuity along the entire line. The endpoint is also a percolation point for the FK clusters, but the
corresponding critical exponents are neither in the Ising nor in the random percolation universality
class. 2002 Published by Elsevier Science B.V.
PACS: 04.60.Nc
1. Introduction
It is quite well established today that thermal features of physical systems can in many
cases be described through the structural properties of connected geometric objects, or
clusters. This mapping between thermal aspects of the given model and the geometrical
properties of the clusters turns out to be particularly fruitful for the study of the critical
behaviour of such models. The increase of the correlation length near the critical point
is paralleled by the increase of the average cluster radius, and its divergence to the
formation of an infinite cluster. Percolation theory [1,2] is the natural framework to study
the properties of cluster-like structures of a system.
One of the most basic results in this field [3] shows that the q-state Potts model without
external field can be mapped onto a geometric model: spin configurations become FK
cluster configurations by connecting nearest-neighbouring spins of the same orientation
with a bond probability pB = 1 exp(J /kT ), where J is the Potts spinspin coupling.
E-mail address: satz@physic.uni-bielefeld.de (H. Satz).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 0 4 - 6
494
S. Fortunato, H. Satz / Nuclear Physics B 623 [FS] (2002) 493502
The mapping is one to one, so that any statement about thermal properties of the model can
be equivalently expressed in terms of cluster quantities; in particular, the magnetization
transition is equivalent to the percolation transition of FK clusters.
The FK transformation is independent of the number q of different spin states of the
model. In particular, it remains valid as well in those cases in which the model for vanishing
external field exhibits a first-order phase transition. It is thus natural to ask whether one can
establish a relation between the behaviour of FK clusters and the thermal properties of the
system also for discontinuous phase changes when the external field H does not vanish. In
this case, a first-order transition persists as long as H remains smaller than some critical
value Hc . So for 0 H < Hc , there is a whole line of first-order phase transitions in the
phase diagram of the model. For H = Hc , the transition becomes second-order and the
critical exponents are conjectured to belong to the Ising universality class; for H > Hc , the
partition function becomes analytic and one has at most a rapid crossover.
We note that the equivalence of the thermal and the percolation description provided by
the FK transformation has been established only for the case H = 0. If one does not insist
on spatial connectivity of clusters, the presence of an external field H = 0 can be taken into
account by introducing a ghost spin connected to all the normal spins of the system [4];
the resulting clusters then are no longer purely spatial and thus differ from those defined
in [3]. We are here specifically interested in spatial connectivity aspects for H = 0 and
want to study whether in this case the percolation of FK clusters has a relation to thermal
first-order phase transitions. In fact, it is known that the usual FK clusters show a number of
non-trivial features also in the presence of an external field [57]. A question of particular
interest is whether for such clusters the line of percolation temperatures Tp (H ) (Kertsz
line) coincides with the line of first-order phase transitions.
Here we want to study by means of Monte Carlo simulations the behaviour of the FK
clusters near the line of first-order phase transitions of the 3-dimensional 3-state Potts
model. This model has been investigated quite extensively, since its phase transition is
closely related to the deconfinement transition of finite temperature QCD [8]. It exhibits
a weak first-order phase transition for H = 0, which disappears already for quite small
values of the external field. The line of first-order phase transitions of this model was
recently studied in detail [9], and the position of the endpoint was determined with great
precision. In what follows we shall exploit the results of this investigation.
2. The FortuinKasteleyn transformation and percolation variables

The ferromagnetic q-state Potts model is defined by the Hamiltonian

(1 i j ) H
i h ,
H=J
ij
(1)
where J > 0 is the spinspin coupling and H the external field; the s represent the spin
variables of the model and can take on q different values. The direction of the external field
is specified by the spin variable h . The partition function Z is given by
Z(T , H ) =

H( )
exp
,
kT
495
(2)
with the sum over all spin configurations. By distributing randomly bonds with probability
pB between all pairs of nearest neighbour sites in the same spin state, one can rewrite
Z(T , H = 0) in the form

c(n)
.
pB
(1 pB )q
Z=
(3)
n
ij ,nij =1
ij ,nij =0
Here the sum runs over all bond configurations {n} ({nij = 1}: active bond, {nij = 0}: no
bond), and c(n) is the number of FK clusters of the configuration. We stress that in Eq. (3)
the spins of the system do not appear; the partition function is given entirely in terms of
bond configurations.
In this work we are mainly interested in the percolation transition of the FK clusters.
We therefore first recall the relevant variables. The order parameter is the percolation
strength P , defined as the probability that a randomly chosen site belongs to an infinite
cluster. The analogue of the magnetic susceptibility is the average cluster size S,

ns s 2
,
S s
(4)
s ns s
where ns is the number of clusters of size s; the sums exclude the percolating cluster.
For thermal second order phase transitions it is found that near the critical temperature Tc ,
P (Tc T ) ,
S |T Tc | ,
T Tc ,
(5)
(6)
where and are the critical exponents for the magnetization and the susceptibility,
respectively.
To study the percolation transition it is helpful to define also the percolation cumulant.
It is the probability of reaching percolation at a given temperature and lattice size, i.e., the
fraction of percolating configurations. This variable has two remarkable properties:
if one plots it as a function of T , all curves corresponding to different lattice sizes
cross at the same point, which marks the threshold of the percolation transition;
the percolation cumulants for different values of the lattice size L coincide, if
considered as functions of [(T Tc )/Tc ]L1/ .
These two features allow a rather precise determination of the critical point already from
simulations of the system for two different lattice sizes. In general, it is preferable to utilize
several lattices in order to evaluate finite size effects and eventual corrections to scaling.
Moreover, by using lattices of very different sizes, the scaling of the curves leads to a more
precise estimate of the critical exponent .
496
3. Results
We have performed Monte Carlo simulations of the 3d 3-state Potts model for several
lattice sizes (403 , 503 , 603 , 703 ) and for different values of the parameters = J /kT and
h = H /kT , using the Wolff cluster update extended to the case of a non-vanishing external
field [10]. To identify the clusters of the different configurations, we used the algorithm
of Hoshen and Kopelman [11]. We have always adopted free boundary conditions and
assumed that a cluster percolates if it connects each pair of opposite faces of the lattice.
At each iteration, we measured the energy E of the system, the magnetization M, the
percolation strength P , the average cluster size S and the size of the largest cluster. We
recall that the magnetization is the fraction of spins pointing in the direction of the external
field. In the case of vanishing field, the majority spin state of the configuration defines
the magnetization: M is given by the fraction of the sites in such spin state. We have
also measured the size of the largest cluster because it allows us to calculate the fractal
dimension D of the percolating cluster at the critical point. Although the Wolff algorithm
is in general very efficient, we were forced to perform many updates between consecutive
measurements to reduce appreciably the correlations of the corresponding configurations.
In order to get independent configurations for the percolation variables, we have taken up
to 1000 updates for the 703 lattice, which made the data production rather slow.
For the analysis near the threshold it turns out to be useful to plot the time history of M
and P . For a first-order phase transition, because of the finite size of the lattice, the system
tunnels from one phase to the other, which is the lattice realization of the coexistence of
the two phases of the system. This can allow a visual check of the order of the percolation
transition.
In general, we expect that a step in the magnetization is accompanied by a step in the
average size of the magnetic domains of the system, and consequently by a step of the FK
cluster size as well. It is thus natural to expect discontinuous variations of the percolation
variables at some threshold. However, the relation between the thermal and the geometric
thresholds is not a priori evident. There are, in principle, three possible scenarios:
The configurations of FK clusters percolate at a temperature Tp above Tc (p < c ),
and the percolation transition is continuous with critical exponents; unrelated to this
transition, both P and S then make a jump at Tc (Fig. 1a).
The configurations of FK clusters percolate at a temperature Tp below Tc (p > c ),
and the percolation transition is continuous with critical exponents; unrelated to this
transition, only S makes a jump at Tc , since P = 0 there (Fig. 1b).
The configurations of FK clusters percolate at Tc ; the percolation transition is
discontinuous and both P and S make a jump at Tc . In particular, P jumps at Tc
from zero to a non-zero value and is still an order parameter (Fig. 1c).
For h = 0, the equivalence between percolation and thermal critical behaviour is
established (see Section 1). What happens in the presence of an external field is so far
not known. We will therefore first consider briefly the case h = 0, as a test present for the
methods to be used in our numerical investigation, and then consider in detail the situation
for h = 0.
497
Fig. 1. Scenarios for the relation between percolation and a first-order thermal transition; M denotes
the magnetization, P the percolation strength.
Fig. 2. Time history at the transition point for the magnetization M and the percolation strength P of
FK clusters in the 3d 3-state Potts model without external field. The lattice size is 703 .
3.1. The case h = 0

For the model without an external field, the magnetization M is the order parameter of
the thermal transition. By studying the time history of M at Tc , we observe a tunneling
between zero and a non-zero value.
In Fig. 2 we show the magnetization M and the percolation strength P as a function
of the number of iterations for a 703 lattice at c . Here we have taken c = 0.550565, as
determined in [12]. We see that P follows the variations of M; the step of P is somewhat
larger and the two geometric phases are clearly visible. In particular, we notice that the
system passes from a non-percolating phase to a percolating one, as expected. Since the
definition of percolation is sharp (either there is a percolating cluster or there is none), we
498
Fig. 3. Time history at the transition point for the magnetization M and the percolation strength P of
FK clusters for the 2d 5-state Potts model without external field. The lattice size is 2002 .
obtain almost always P = 0 when the system is in the non-percolating phase. In contrast,
the magnetization varies smoothly in the lattice average and hence shows significant
fluctuations also in the paramagnetic phase. Hence P can resolve eventual discontinuities
due to different phases better than M can.
To show that such behaviour of the percolation strength remains clearly evident also in
the simulation of other systems, we repeat our analysis for the 2d 5-state Potts model. In
two dimensions the critical temperature is given by the exact formula c (q) = log(1 + q),
which for q = 5 yields c (5) = 1.174359. Fig. 3 shows the time history of M and P at
c (5): the result is the same as before.
3.2. The case 0 < h < hc
In the presence of an external field, the magnetization is different from zero at
any temperature, so that it is no longer a genuine order parameter for a thermal
transition. Nevertheless, for 0 < h < hc , M shows discontinuous behaviour at some critical
temperature Tc (h), and hence it remains interesting to compare again the behaviour of M
and P for h = 0 along the line of discontinuity Tc (h). This line ends at a critical value of the
field, at which the thermal transition becomes continuous; for the 3d 3-state Potts model,
hc = 0.000775(10) (see [9]). In [9], several points of the line of first order phase transitions
of the model were determined as well. We choose two of these points, corresponding to
the values 0.0005 and 0.0006 of the reduced field h, and there determine the time history
of M and P ; it is shown in Figs. 4 and 5, respectively.
From these figures we see that there still are two phases, represented by the two bands of
the magnetization values, although now M is never zero. The step between the two bands
499
Fig. 4. Time history at the transition point and with field h = 0.0005 for the magnetization M and
the percolation strength P of FK clusters. The lattice size is 703 .
Fig. 5. Time history at the transition point and with field h = 0.0006 for the magnetization M and
the percolation strength P of FK clusters. The lattice size is 703 .
is narrower for h = 0.0006, as it should be, since we are approaching the critical value
hc of the field, at which there would be a continuous variation of M. In both cases, the
percolation strength also makes a jump from zero to a non-zero value, as happens for h = 0.
We notice that the number of intermediate P values between the two bands increases
500
when one passes from h = 0.0005 to h = 0.0006, which indicates that the transition from
one geometrical phase to the other is getting smoother: we are therefore approaching a
continuous percolation transition as well.
3.3. The case h = hc
The line of first-order phase transitions terminates with a continuous phase transition
at Tc (hc ). All thermal variables vary continuously or diverge at this critical point. In
particular, the magnetization varies continuously here, and the critical exponents at Tc (hc )
put the transition into the universality class of the 3d Ising model. The continuous
thermal transition suggests that also the percolation transition of the FK clusters becomes
continuous at h = hc . Nevertheless, there are in principle three possible scenarios for the
percolation transition:
The FK clusters percolate at a temperature Tp = Tc (hc ); in this case, due to the finite
correlation length of the thermal system at any T = Tc (hc ), the critical percolation
exponents will belong to the universality class of random percolation in three
dimensions.
The configurations of FK clusters percolate at Tc (hc ), but the exponents do not
coincide with the ones of the 3d Ising model, which govern the thermal transition
at this point.
The configurations of FK clusters percolate at Tc (hc ), and the exponents are the 3d
Ising exponents.
We have seen so far that the threshold of the (first-order) percolation transition coincides
with the (first-order) thermal threshold from h = 0 up to h = 0.0006: that suggests that also
for h = hc = 0.000775, the two critical points coincide.
To check if this is indeed correct, we plot in Fig. 6 the percolation cumulant defined in
Section 2 as a function of for different lattice sizes. We see that within errors the lines
cross at the same point. The vertical dashed lines of the figure mark the thermal threshold
determined in [9] within one standard deviation. The agreement between percolation and
thermal critical temperatures is very good, particularly if we take into account that we
performed our simulations for hc = 0.000775, even though the value of the critical field hc
also contains some uncertainty.
From Fig. 6 we also obtain first indications of the critical exponents of the percolation
transition. The height of the crossing point is a universal number, i.e., it identifies a
universality class. In our figure, the horizontal lines shown correspond to the universality
classes of the 3d Ising model and 3D random percolation (see [13]). The crossing point
does not fall on either of the two lines, which means that the percolation exponents of our
transition coincide neither with the 3d Ising exponents nor with the exponents of random
percolation.
Using standard finite size scaling techniques, we obtain the following values for the
critical percolation exponents: / = 0.32(3), / = 2.32(2), = 0.45(3). From the
scaling of the size of the largest cluster at Tc we find that the fractal dimension D of
the percolating cluster is D = 2.66(3). It is easy to check the our values satisfy the scaling
501
Fig. 6. Percolation cumulant at the transition endpoint h = hc for FK clusters as a function of for
four lattice sizes.
relations
+ 2 = d,
+ = D,
(7)
within the errors we have determined; here d is the space dimension of the lattice.
However, the critical indices we have found differ from the Ising (/ = 0.5187(14),
/ = 1.963(7), = 0.0.6294(10)) and from the random percolation exponents (/ =
0.477(2), / = 2.045(10), = 0.0.8765(17)), as expected. The fact that our exponents
do not coincide with the random percolation exponents is in fact a further proof that
the geometrical transition takes place exactly at the thermal threshold, because only the
presence of an infinite correlation length can shift the values of the critical indices out of
the random percolation universality class.
4. Conclusions
We have shown that FK clusters reveal interesting complementary critical features also
for spin models undergoing a first-order phase transition. In particular, we find that for the
3d 3-state Potts model, the percolation strength jumps from zero to a non-zero value for
any value of the external field up to the endpoint. Therefore, the line of thermal first-order
phase transitions is also a line of first-order percolation transitions for the FK clusters,
and the percolation strength constitutes a genuine order parameter for any 0 h < hc , in
contrast to the magnetization. For h = hc , the percolation transition becomes continuous,
and its threshold coincides with the thermal critical point. However, the critical indices of
the geometrical transition are not in the 3d Ising universality class.
The value of D we have found is bigger than the fractal dimension of the percolating
FK cluster in the 3d Ising model (DIsing = 2.48(2)); hence the FK clusters seem too large
502
to define correctly the thermal critical behaviour of the model at the endpoint. This is
reminiscent of the situation encountered for the pure site clusters of the 2d Ising model,
which reproduce the right critical temperature [14] but not the exponents [15]. In that case
the introduction of the bond probability pB reduces the size of the clusters and restores
the correct critical behaviour. We thus conclude that also in our case one has to define a
correct bond probability pB (J, H, T ) to obtain a coincidence of thermal and percolation
transitions.
In this work, we have limited ourselves to the 3d 3-state Potts model, mainly because
here the endpoint is known precisely. As mentioned in Section 3, much computer time
is needed to obtain uncorrelated configurations, and hence it would have been very timeconsuming to investigate further systems. Nevertheless, it would be of interest to check
whether the results we have found are valid for any Potts model which undergoes a firstorder phase transition for h = 0.
Acknowledgements
It is a pleasure to thank F. Karsch and S. Stickan for helpful discussions. We would
gratefully acknowledge the financial support of the TMR network ERBFMRX-CT-970122
and the DFG Forschergruppe FOR 339/1-2.
References
[1] D. Stauffer, A. Aharony, Introduction to Percolation Theory, Taylor & Francis, London, 1994.
[2] G.R. Grimmett, Percolation, Springer-Verlag, 1999.
[3] P.W. Kasteleyn, C.M. Fortuin, J. Phys. Soc. Japan (Suppl.) 26 (1969) 11;
C.M. Fortuin, P.W. Kasteleyn, Physica 57 (1972) 536;
C.M. Fortuin, Physica 58 (1972) 393;
C.M. Fortuin, Physica 59 (1972) 545.
[4] R.H. Swendsen, J.S. Wang, Physica A 167 (1990) 565.
[5] J. Kertsz, Physica A 161 (1989) 58.
[6] J. Adler, D. Stauffer, Physica A 175 (1991) 222.
[7] S. Fortunato, H. Satz, Phys. Lett. B 509 (2001) 189.
[8] B. Svetitsky, L.G. Yaffe, Phys. Rev. D 26 (1982) 963.
[9] F. Karsch, S. Stickan, Phys. Lett. B 488 (2000) 319.
[10] I. Dimitrovic et al., Nucl. Phys. B 350 (1991) 893.
[11] J. Hoshen, R. Kopelman, Phys. Rev. B 14 (1976) 3438.
[12] W. Janke, R. Villanova, Nucl. Phys. B 489 (1997) 679, and references therein.
[13] S. Fortunato, Ph.D. thesis, Bielefeld University, hep-lat/0012006.
[14] A. Coniglio et al., J. Phys. A 10 (1977) 205218.
[15] M.F. Sykes, D.S. Gaunt, J. Phys. A 9 (1976) 21312137.

The fermion boson interaction within the linear

sigma model at finite temperature
H.C.G. Caldas
Departamento de Cincias Naturais, DCNAT Fundao de Ensino Superior de So Joo del Rei, FUNREI,
Praa Dom Helvcio, 74, CEP:36300-000, So Joo del Rei, MG, Brazil
Received 19 June 2001; accepted 12 December 2001
Abstract
We reinvestigate the interaction of massless fermions with massless bosons at finite temperature.
Specifically, we calculate the self-energy of massless fermions due the interaction with massless
bosons at high temperature, which is the region where thermal effects are maximal. The calculations
are concentrated in the limit of vanishing fermion three momentum and after considering the effective
fermion and boson dressed masses, we obtain the damping rate of the fermion up to order g 3 . It is
shown that in the limit k0 T the fermion acquire a thermal mass of order gT and the leading term
of the fermion damping rate is of order g 2 T + g 3 T . 2002 Published by Elsevier Science B.V.
PACS: 12.39.Fe; 12.38.Lq
1. Introduction
The fermion dispersion relation to one-loop order at high temperature has been studied
several times in the literature [13]. The dispersion relation of a fermion, which gives its
energy as it propagate through the medium as a function of its momentum k, is very
important in different kinds of physical situations [46] and has been focused on various
approximations and limits [7,8]. During the last few years, there has been some controversy
if the damping rate is gauge dependent or not and other problems [9]. It has been proposed
that a proper resummation cures such problems found in most one-loop calculations [10].
In fact, if one wants consistent results of damping rates at positive temperatures, it is
crucial the use of methods like the well-known hard thermal loop (HTL) resummation of
Braaten and Pisarski [11] which employs resumed propagators and vertices and keep the
E-mail address: hcaldas@funrei.br (H.C.G. Caldas).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 6 3 3 - 2
504
H.C.G. Caldas / Nuclear Physics B 623 [FS] (2002) 503512
correct order of the coupling constant. With this method, Braaten and Pisarski has given in
Ref. [12] a formal proof that resummation produces gauge-invariant results for the damping
rates of quarks and gluons. In this article we reinvestigate the self-energy of massless
fermions interacting with massless bosons at high temperature in the framework of the
linear sigma model. This is the simpler, but instructive situation of fermions interacting
with scalar and pseudoscalar fields. These are similar to Yukawa couplings which has
been discussed by M. Thoma in Ref. [13]. Using the HTL, Thoma has found a kinematic
restriction (not observed in the case of quarks or gluons) depending on the effective fermion
and boson masses involved in the calculation of the damping rate of the fermion. More
recently, in Ref. [7], a fermion-scalar plasma has been considered in the context of real-time
formulation and it was found that the Yukawa fermions acquire a width through the induced
decay of the scalar in the medium. Here, as in [13], we use imaginary time formalism and
calculate the fermion dispersion relation for interacting massless bosons and fermions in
the limit k0 , k T and compute the fermion damping rate at rest, k = 0, for the case
where the effective dressed (induced by the thermal medium) fermion and boson masses
are consistently considered.
In a recent paper [14] we proposed a modified self-consistent resummation (MSCR)
which resums higher-order terms in a non-perturbative way in order to cure the problem
of breakdown of the perturbative expansion at finite temperature up to one-loop order in
the perturbative expansion. We have shown that the MSCR, when applied to the study of
the chiral fermion meson model, has the essential features which lead to the satisfaction
of Goldstones theorem and renormalization of the UV divergences, in the low and high
temperature regions. We have explicitly shown that the scheme breaks down around Tc ,
i.e., in the region of intermediate temperatures, since quantum fluctuations are known to
play a major role there. In this region higher-order terms in the perturbative expansion are
required.
It is well known that at high temperature the perturbative expansion can also be broken
in theories with spontaneous symmetry breaking (SSB) or in massless field theories (like
QCD) because powers of the temperature can compensate for powers of the coupling
constant, even if the strength of the coupling is small [15,16]. Infrared divergence appears.
When a set of infrared-divergent diagrams is summed up one gets an infrared-finite result.
This is implemented in the MSCR by the recalculation of the self-energy. A comparison
between the MSCR and the HTL resummation methods showing its similarities and
differences, which is out of the scope of this paper, is under preparation and will be reported
elsewhere [17]. So, a motivation to study the fermion dispersion relation and damping rate
at high temperature is the fact that the MSCR showed itself an efficient method to execute
the resummation in a divergence-free way in this region. Thus, in the computation of the
damping rate of the fermion, we shall use at the starting point the dressed by interaction
boson mass obtained in [14] rather than the zero mass parameter of the Lagrangian. In
this way, this calculation is interesting since it provides another simple example for the
application of the MSCR method.
This paper is organized as follows. In Section 2 we study the fermion self-energy and
obtain the dispersion relation of the fermion with some of its interesting limits. In Section 3
we compute the fermion damping rate at rest. Section 4 is devoted to conclusions.
505
2. The fermion self-energy

We describe the fermion-bosons vertex by the interaction Lagrangian extracted from the
linear sigma model [18]

Lint. = g + i 5 ,
(1)
where , , and represent the quark, sigma and pion fields, respectively, and g is a
non-dimensional positive coupling constant.
The fermion self-energy is defined by
D(n , k)1 = D0 (n , k)1 + (n , k),
(2)
where D0 (n , p) is the tree-level fermion propagator, expressed as

D0 (n , k)1 = k m ,
(3)
and
with s , 0 and being the contributions proportional to the unit,
0 and matrices, respectively.
To one-loop order the fermion self-energy expression is given [20] by

2-loop
d 3p
ln ZI
=
(k0 , k)
= g 2 T
D0 (n+l , p + k)D0 (n , p)
D0
(2)3
1PI
n
d 3p
3g 2 T
D0 (n+l , p + k)D0 (n , p),
(2)3
n
(4)
since the logarithm of the two-loop interaction partition function is found to be [14]:
= s + ,
2-loop
ln ZI
1
= g2
2

d1 d2
d x1 d x2
[d]eS0 [( )2 + (i 5 )2 ]

. (5)
[d]eS0
In Eq. (4) n are the Matsubara frequencies, defined as n = 2nT for bosons and
n = (2n + 1)T for fermions and D0, (n , p) is the tree-level boson propagator,
expressed as
D0, (n , k)1 = n2 + k2 + m2, .
(6)
An evaluation of Eq. (4) at zero three momentum gives (k0 , |k| = 0) = ( 0 0 +

s ) + 3( 0 0 (m m ) + s (m m )) with the 0 and scalar- parts of the
sigma contribution given, respectively, by
0
g2
= k0
2

0
g2
+ k0
2
2
k02 + 2 +
dp p2 n
2 [k02 ( )2 ][k02 ( + )2 ]

0
2
2
dp p2 n
,
2 [k02 ( )2 ][k02 ( + )2 ]
(7)
506
Fig. 1. The one-loop fermion self-energy. The pseudoscalar-boson contribution (a) and the scalar-boson
contribution (b).
g2
=
2

0
g2
+
2
2)
m (k02 2 +
dp p2 n
2 [k02 ( )2 ][k02 ( + )2 ]

0
2
,
where
in Fig. 1.
p2
2)
m (k02 2 +
dp p2 n
,
2 [k02 ( )2 ][k02 ( + )2 ]
(8)
2 p2 + m2 . The one-loop fermion self-energy is shown

+ m2, and
2.1. The dispersion relation of massless fermions interacting with massless bosons
As a first approximation, in this subsection by considering the interaction of massless
fermions with massless bosons we get an effective thermal fermion mass and also calculate
the dispersion relation of the fermions.
The poles of the massless fermion propagator (m = s = 0) gives the dispersion
relation which occurs at the positive-energy root of

2

2 ,
0 , k)
k0 (k) + 0 (k0 , k) = k + (k
(9)

where k |k|.
For our intentions, it is sufficient to evaluate Eq. (4) in the limit k0 , k T , and consider
the interaction of massless bosons and fermions in order to obtain an effective fermion
thermal mass and dispersion relation. Thus,

1 2 T 2 k0 + k
ln
0 (k0 , k) = g
(10)
,
8
k k0 k

1 T 2 k0 k0 + k

k,
0 , k) = g 2
1
ln
k
(k
(11)
4 k 2 2k k0 k

1 2 T 2 k0 k0 + k

1
.
g 2
ln
4 k 2k
k k
0
Some limits of expressions (10) and (11) are [21]:

0 (k0 = 0, k) = 0,
0 (k0 , k = 0) =
g2 T 2
,
4k0
(12)
507
g2 T 2
(13)
k.
4k
Defining the fermion mass as the location of the pole in the limit k = 0, we have
k0 + 0 (k0 , k 0) = 0, which implies
0 , k = 0) = 0,
(k
0 = 0, k) =
(k
g2 T 2
.
(14)
4

From (9), we see that the fermion dispersion relation is given by k0 + 0 = k(1 + ),
that is

2
M2 k0 k0 + k
1 M k0 + k

1 ,
=
k
ln
ln
k0
(15)
2 k
k0 k
k 2k k0 k
M2 =
which has the following well-known form in the low momentum expansion [1,21]
k2
1
k0 (k) k0 (k)+ = M + k +
.
3
3M
(16)
This dispersion relation, k0 (k)+ M + 13 k for k M , represents an ordinary

fermion whose chirality is equal to its helicity. There is another dispersion relation,
k0 (k) M 13 k, termed a plasmino [22], which describes a quasiparticle with chirality
opposite to its helicity [3].
3. The fermion damping rate

Let us now proceed with the computation of the damping rate at rest (k = 0). These
calculations will be done considering in the first step m = 0 and the dressed (effective)
boson mass in the internal lines of the fermion self-energy. The effective fermion mass
will be taken into account in the second step of the recalculation of the self-energy as the
MSCR dictates [14]. In the high temperature region, the bosons dressed masses (given by
the MSCR) to be used in internal lines of the fermion self-energy read
g2 2
T .
3
So, Eq. (7) may be written as
m2 = m2 MB2 =
2g 2 k0
0 (k0 , |k| = 0) = 2
(17)

dp p2
0
4g 2 k0
+
2
nB k02 MB2 2p2

B [k02 MB2 ]2 4k02 p2

dp p4
0
n
1
,
[k02 MB2 ]2 4k02 p2
(18)
where nB and n are the usual distribution functions for bosons and fermions given,
respectively, by
nB (B ; T ) =
1
,
eB 1
(19)
508
n ( ; T ) =
(20)
+1

with B p2 + MB2 and |p|.
It is worth to note that for k0 MB in Eq. (18), it is easy to see that 0 (k0 , |k| = 0)
reduces to the second line of Eq. (12) which is a stable state without singularities and we
have that there is no decay.
An explicit evaluation of Eq. (18) in the pole of the corrected propagator (where the
square of the matrices is equal the unit matrix) furnishes
g2
0 = 2
4

dx
0
3g 2
8 2

0

B )
nB (
1
1+

B
4 x 2
x + 2
1
1
dx n (
)
+
x2
x + 2
3g 2 T 2
2

dx x
0
B )
2 nB (
B

3g 2 T 2
2

dx x 2
0
n (
)
,

(21)

M2
with the definitions k0 k0B ,
B 2 p2 + 2 MB2 ,
|p| and p x. The
interesting physics happens when k0 > MB . Otherwise (if k0 < MB ) one would get
imaginary (forbidden) frequency.
The expression for 0 in (21) has singularities, and now we adopt the prescription
= i , since in general k0 is complex, where is the real frequency and is a real
constant. With this assumption for , Eq. (21) is expressed as
g2
0 = 2
4

0
g2
+ i 2 2
8
3g 2
4 2

2x
2x
B )
nB (

+
1+
dx

2x
2
2
2x
2
2

B
2
+
+ +

B )
nB (
dx
+
2
2
2x
2x
2
2
B

+
+ +

dx n (
)
2x
3g 2
i 2
4
3g 2 T 2
2

dx n (
)
0

dx x
0
2x
+
+
2x
2
2
2
+2
+ +
2x
B )
2 nB (
B

2x
+
2
2
2x
2
2
+
+ +
3g 2 T 2
2

dx x 2
0
n (
)
.

(22)
509
Now making use of the definition of the delta function

lim
(y) = & 0
&
1
,
2
y + &2
(23)
2x
,
2x
2
2
2
+2
+
(24)
2x
+
+
,
2x
2
2
2
+2
+
(25)
and the definitions

2x
F (x, )
2x
and
2x
G(x, )
2x
we get

g2
nB (
B )
1
2

dx
1 + F (x, ) + i

B
2
8
2
2 42 +MB2
+
M
1
0
B e
4

||
3g 2
3g 2
dx
n
(
)G(x,
)
+
i
4 e||/2 + 1
4 2
g2
0 = 2
4
3g 2 T 2
2

dx x
B )
2 nB (
3g 2 T 2
2

dx x 2
0
n (
)
.
(26)
Here we use some results from high temperature expansion of one-loop integrals derived
by Dolan and Jackiw in [16]:

0

2
MB
MB
3
B )
nB (
T
1
+O
,

dx
=
+ ln

B
2MB
2
4T
2g
T2
where in Eq. (27) we have used from (17) that MB =
g 3
8
gT
.
3
(27)
This means that the first term in
the r.h.s. of Eq. (26) is

1. On the other hand, for the last two terms in the r.h.s. of
this same expression, we have
T2
2

dx x
0
B )
2 nB (

B
T2
+ 2

dx x 2
0
n (
)

2

T 2 MB T
T2
T
+ O g2 +
.
6
2
12
4
(28)
The parts involving F (x, ) and G(x, ) are less important contributions in comparison
to the dominant term that is proportional to g 2 T 2 /, mainly for small . So, in a first
glance one can neglect them. Putting these results in Eq. (26) and assuming ( ), the
510
leading term of the real frequency can be written as

9g 2 T 2
2
M,1
(29)
.
4
The next step is the recalculation of the self-energy of the fermion to get the second
order corrected fermion mass from the pole location
12 =
k0,2 M,1 + (MB , M,1 , k = 0) = 0,
(30)
2 and Im k
where from this equation we will identify (Re k0,2 )2 = 22 M,2
0,2 = . With
this resummation procedure it is possible to take into account the induced by the thermal
medium fermion mass. At each recalculation we take into account more infinity subsets of
graphs which result is shown in the mass. Now we are able to find out the damping rate of
the fermion, which is defined by the imaginary part of the self-energy on-shell [3].
From Eq. (4) the self-energy (MB , M,1 , k = 0) = s (MB , M,1 , k = 0) + 0 (MB ,
M,1 , k = 0) in Eq. (30) to be evaluated now is

= 2k0g
2
0
2 + 2p2 )
k02 (MB2 + M,1
dp p2 nB
2 )]2 4k 2 p2 4M 2 M 2
2 B [k02 (MB2 + M,1
B ,1
0

+ 4k0 g
2
0

+ 2g
2
0

+ 2g 2
0
2 + p2
M,1
dp p2 n,1
2 )]2 4k 2 p2 4M 2 M 2
2 ,1 [k02 (MB2 + M,1
0
B ,1
2 )
M,1 (k02 MB2 + M,1
dp p2 nB
2 )]2 4k 2 p2 4M 2 M 2
2 B [k02 (MB2 + M,1
B ,1
0
2 )
M,1 (k02 MB2 + M,1
dp p2 n,1
,
2 )]2 4k 2 p2 4M 2 M 2
2 ,1 [k02 (MB2 + M,1
0
B ,1
(31)
2 ) in the denominators in the r.h.s. of Eq. (31) will be

where the last term (4MB2 M,1
neglected, since it is of O(g 4 ).
Doing again the calculations necessary to get the real and imaginary parts of the poles,

M 2 +M 2
2 , one
where now k0 B k0 ,1 ,
B 2 p2 + 2 MB2 and
,1 2 p2 + 2 M,1
finds

2g
2
2
2
M,2
,
2 = M,1 1
(32)
3

22
22
9g 2
3g 2
1
1

+
.
=
2

2
2
2
8
8
2
22 /4+M,1
2 2 /4+MB
22
2
+
M
e
1
e
+
1
B
4
4 + M,1
(33)
It is important to note that Eq. (32) allows us to define a critical value of g [19], that
is gcr 2.7. This value agrees with the one obtained in Ref. [3]. As pointed out by M.
511
Le Bellac in Ref. [19], collective excitations should disappear at the critical temperature,
although we do not expect perturbation theory is still valid for such large values of g.
Eqs. (32) and (33) has the following interpretation: the leading order of the real frequency
2 is of order gT , in concordance with (14). The damping rate is proportional to the
probability n ( 12 ) of having a fermion with energy 12 and a probability nB ( 12 ) of
having a boson with energy 12 . These probabilities are weighted by numerical factors
and the available phase space [20]. The distribution functions can be expanded for low
energies, nB ( 12 ) T / and n ( 12 ) 1/2 and the damping rate (up to order g 3 ) reduces
to

9g 2 T
27g 3 T
+
.
16
32
(34)
4. Concluding remarks
In this paper, we have considered the fermion boson interaction at finite temperature.
First, we have calculated the self-energy of the fermions due the interaction with scalarbosons and pseudoscalar-bosons in the framework of the linear sigma model.
Next, we have calculated the fermion dispersion relation in the limit k0 , k T of
massless fermions interacting with massless bosons and some of its limits. Also, we have
obtained the thermal fermion mass which is of order gT .
Finally, we have computed the frequency and the damping rate of the fermion at rest,
considering the dressed fermion and boson masses in the internal lines of the fermion selfenergy rather than the zero mass parameter of the Lagrangian. The damping rate of the
fermion was found to be of order g 2 T from the boson internal line of the fermion selfenergy plus a part which is of order g 3 T from the fermion internal line of the self-energy.
This is a remarkable result since it shows the signature of the alternative MSCR method
(up to second order of the non-perturbative correction) in the desired effective mass and
damping rate. One important feature of this calculation (up to this order) is that we have
gotten this result algebraically in a clear manner differently from the cases where the results
are reached numerically.
The calculation of the fermion damping rate at rest constitutes another simple but
instructive application of the MSCR method.
Acknowledgements
The author thanks the hospitality given by the Nuclear Theory group during his visit
at the University of Minnesota were this work was initiated. He is gratefully indebted to
Professors Joe Kapusta and Paul Ellis for various helpful discussions. Also, the author
would like to thank Professor Paul Ellis for a critical reading of the manuscript.
512
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
H.A. Weldon, Phys. Rev. D 26 (1982) 2789.

G. Gatoff, J. Kapusta, Phys. Rev. D 41 (1990) 611.
E. Braaten, R.D. Pisarski, Phys. Rev. D 46 (1992) 1829.
J. Letessier, J. Rafelski, A. Tomsi, Phys. Lett. B 323 (1990) 393.
E. Braaten, R.D. Pisarski, T.C. Yuan, Phys. Rev. Lett. 66 (1991) 2183.
M.H. Thoma, C.T. Traxler, Phys. Rev. D 56 (1997) 198.
D. Boyanovsky, H.J. de Veja, D.-S. Lee, Y.J. Ng, S.-Y. Wang, Phys. Rev. D 59 (1999) 105001, hepph/9810393.
H.A. Weldon, Phys. Rev. D 61 (2000) 036003.
A. Rebhan, Phys. Rev. D 46 (1992) 4779.
R. Kobes, G. Kunstatter, A. Rebhan, Phys. Rev. Lett. 64 (1990) 2992;
R. Kobes, G. Kunstatter, A. Rebhan, Nucl. Phys. B 355 (1991) 1.
E. Braaten, R.D. Pisarski, Nucl. Phys. B 337 (1990) 569.
E. Braaten, R.D. Pisarski, Nucl. Phys. B 337 (1990) 369;
E. Braaten, R.D. Pisarski, Phys. Rev. Lett. 64 (1990) 1338.
M.H. Thoma, Z. Phys. C 66 (1995) 491.
H.C.G. Caldas, A.L. Mota, M.C. Nemes, Phys. Rev. D 63 (2001) 56011, hep-ph/0005180.
S. Weinberg, Phys. Rev. D 9 (1974) 3357.
L. Dolan, R. Jackiw, Phys. Rev. D 9 (1974) 3320.
H.C.G. Caldas, work in progress.
M. Gell-Mann, M. Levy, Nuovo Cimento 16 (1960) 705.
Michel Le Bellac, Thermal Field Theory, Cambridge Univ. Press, Cambridge, 1996.
J. Kapusta, Finite-Temperature Field Theory, Cambridge Univ. Press, Cambridge, 1989.
J. Kapusta, CERN preprint, 1982, unpublished.
E. Braaten, Astrophys. J. 392 (1992) 70.

CUMULATIVE AUTHOR INDEX B621B623
Aliev, T.M.
Alpha Collaboration
Altarelli, G.
Alvarez-Gaum, L.
Asorey, M.
Astier, P.
Autiero, D.
B621 (2002) 479

B623 (2002) 271
B621 (2002) 359
B623 (2002) 165
B622 (2002) 593
B621 (2002) 3
B621 (2002) 3
Bajnok, Z.
Bajnok, Z.
Bak, D.
Baldisseri, A.
Baldo-Ceolin, M.
Ball, R.D.
Banner, M.
Barbn, J.L.F.
Bassett, B.A.
Bassompierre, G.
Bazhanov, V.V.
Benslama, K.
Berends, F.A.
Bertolini, M.
Besson, N.
Bird, I.
Blumenfeld, B.
Bobisut, F.
Bosch, S.W.
Bouchez, J.
Boyd, S.
Brandenberger, R.
Buchalla, G.
Buchbinder, I.L.
Bueno, A.
Bunyatov, S.
Caldas, H.C.G.
Camilleri, L.
Cardini, A.
Caselle, M.
B622 (2002) 548

B622 (2002) 565
B622 (2002) 95
B621 (2002) 3
B621 (2002) 3
B621 (2002) 359
B621 (2002) 3
B623 (2002) 165
B622 (2002) 393
B621 (2002) 3
B622 (2002) 475
B621 (2002) 3
B623 (2002) 220
B621 (2002) 157
B621 (2002) 3
B621 (2002) 3
B621 (2002) 3
B621 (2002) 3
B621 (2002) 459
B621 (2002) 3
B621 (2002) 3
B623 (2002) 421
B621 (2002) 459
B621 (2002) 179
B621 (2002) 3
B621 (2002) 3
Cattaneo, P.W.
Cavasinni, V.
Cervera-Villanueva, A.
Challis, R.
Chapovsky, A.P.
Chen, G.-H.
Cho, Y.M.
Chu, C.-S.
Chukanov, A.
Cieza Montalvo, J.E.
Collazuol, G.
Conforto, G.
Conta, C.
Contalbrigo, M.
Contino, R.
Corcella, G.
Costa-Santos, R.
Cousins, R.
B621 (2002) 3
B621 (2002) 3
B621 (2002) 3
B621 (2002) 3
B621 (2002) 257
B622 (2002) 189
B621 (2002) 388
B621 (2002) 101
B621 (2002) 3
B623 (2002) 325
B621 (2002) 3
B621 (2002) 3
B621 (2002) 3
B621 (2002) 3
B622 (2002) 227
B623 (2002) 247
B623 (2002) 439
B621 (2002) 3
Daniels, D.
De Campos, F.
Degaudenzi, H.
De Gouva, A.
Del Prete, T.
Dent, T.
De Santo, A.
Daz, M.A.
Dignan, T.
Di Lella, L.
Di Vecchia, P.
Do Couto e Silva, E.
Dudas, E.
Dumarchez, J.
Durhuus, B.
B621 (2002) 3
B623 (2002) 47
B621 (2002) 3
B623 (2002) 395
B621 (2002) 3
B623 (2002) 73
B621 (2002) 3
B623 (2002) 47
B621 (2002) 3
B621 (2002) 3
B621 (2002) 157
B621 (2002) 3
B622 (2002) 46
B621 (2002) 3
B623 (2002) 201
B623 (2002) 503

B621 (2002) 3
B621 (2002) 3
B623 (2002) 474
Easson, D.A.
boli, O.J.P.
Ellis, J.
Ellis, M.
B623 (2002) 421

B623 (2002) 47
B621 (2002) 208
B621 (2002) 3
0550-3213/2002 Published by Elsevier Science B.V.

PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 0 1 1 - 1
514
Falceto, F.
Fazio, T.
Fehr, L.
Feldman, G.J.
Ferrari, R.
Ferrre, D.
Flaminio, V.
Forger, M.
Forte, S.
Fortunato, S.
Fraternali, M.
Frau, M.
Fujikawa, K.
Fyodorov, Y.V.
B622 (2002) 593

B621 (2002) 3
B621 (2002) 622
B621 (2002) 3
B621 (2002) 3
B621 (2002) 3
B621 (2002) 3
B621 (2002) 523
B621 (2002) 359
B623 (2002) 493
B621 (2002) 3
B621 (2002) 157
B622 (2002) 115
B621 (2002) 643
Gaillard, J.-M.
Gangler, E.
Gardi, E.
Gattnar, J.
Geiser, A.
Geppert, D.
Gherghetta, T.
Ghilencea, D.M.
Gibbons, G.W.
Gibin, D.
Giudice, G.F.
Giveon, A.
Gninenko, S.
Gckeler, M.
Godley, A.
Gomez-Cadenas, J.-J.
Gosset, J.
Gling, C.
Gouanre, M.
Grant, A.
Graziani, G.
Guglielmi, A.
B621 (2002) 3
B621 (2002) 3
B622 (2002) 365
B621 (2002) 131
B621 (2002) 3
B621 (2002) 3
B623 (2002) 97
B622 (2002) 215
B623 (2002) 3
B621 (2002) 3
B623 (2002) 395
B621 (2002) 303
B621 (2002) 3
B623 (2002) 287
B621 (2002) 3
B621 (2002) 3
B621 (2002) 3
B621 (2002) 3
B621 (2002) 3
B621 (2002) 3
B621 (2002) 3
B621 (2002) 3
Hagner, C.
Hasenbusch, M.
Hernando, J.
Hibberd, A.N.
Higashijima, K.
Hisano, J.
Hofmann, R.
Horsley, R.
Hubbard, D.
Hurst, P.
Hyett, N.
B621 (2002) 3
B623 (2002) 474
B621 (2002) 3
B622 (2002) 475
B623 (2002) 133
B621 (2002) 208
B623 (2002) 301
B623 (2002) 287
B621 (2002) 3
B621 (2002) 3
B621 (2002) 3
Iacopini, E.
Ishibashi, M.
B621 (2002) 3
B622 (2002) 115
Jamin, M.
Joseph, C.
Juget, F.
B622 (2002) 279

B621 (2002) 3
B621 (2002) 3
Kazakov, V.
Kent, N.
Khoroshkin, S.M.
Khoze, V.A.
Khoze, V.V.
Kim, S.-W.
Kimberly, D.
Kimura, T.
Kirsanov, M.
Klaus, B.
Klimov, O.
Kniehl, B.A.
Kofinas, G.
Kokkonen, J.
Kostov, I.K.
Kovzelev, A.
Krasnoperov, A.
Kurth, M.
Kustov, D.
Kutasov, D.
Kutasov, D.
Kuznetsov, V.
B622 (2002) 141

B621 (2002) 3
B622 (2002) 475
B621 (2002) 257
B621 (2002) 101
B622 (2002) 95
B623 (2002) 421
B623 (2002) 133
B621 (2002) 3
B623 (2002) 287
B621 (2002) 3
B621 (2002) 337
B622 (2002) 347
B621 (2002) 3
B622 (2002) 141
B621 (2002) 3
B621 (2002) 3
B623 (2002) 271
B621 (2002) 3
B621 (2002) 303
B622 (2002) 141
B621 (2002) 3
Lacaprara, S.
Lachaud, C.
Lakic, B.
Langfeld, K.
Lanza, A.
La Rotonda, L.
Lashkevich, M.
Laveder, M.
Lee, C.-W.H.
Lerche, W.
Lerda, A.
Letessier-Selvon, A.
Levy, J.-M.
Lindner, M.
Linssen, L.
Ljubicic, A.
Lola, S.
Long, J.
L, H.
Lunin, O.
Lupi, A.
Ltken, C.A.
B621 (2002) 3
B621 (2002) 3
B621 (2002) 3
B621 (2002) 131
B621 (2002) 3
B621 (2002) 3
B621 (2002) 587
B621 (2002) 3
B623 (2002) 201
B622 (2002) 269
B621 (2002) 157
B621 (2002) 3
B621 (2002) 3
B622 (2002) 429
B621 (2002) 3
B621 (2002) 3
B621 (2002) 208
B621 (2002) 3
B623 (2002) 3
B623 (2002) 342
B621 (2002) 3
B622 (2002) 269
Ma, E.
Ma, J.P.
Macfarlane, A.J.
Magro, M.B.
Maes, J.L.
B623 (2002) 126

B622 (2002) 416
B621 (2002) 712
B623 (2002) 47
B621 (2002) 37
Marchionni, A.
Marotta, R.
Martelli, F.
Martn, C.P.
Martin, I.
Mathur, S.D.
McCoy, B.M.
Mchain, X.
Mendiburu, J.-P.
Mercadante, P.G.
Meyer, J.-P.
Mezzetto, M.
Mielke, E.W.
Mishra, S.R.
Mitov, A.D.
Moch, S.
Moorhead, G.F.
Mosconi, P.
Mourad, J.
Mussardo, G.
B621 (2002) 3
B621 (2002) 157
B621 (2002) 3
B623 (2002) 150
B622 (2002) 240
B623 (2002) 342
B623 (2002) 439
B621 (2002) 3
B621 (2002) 3
B623 (2002) 47
B621 (2002) 3
B621 (2002) 3
B622 (2002) 457
B621 (2002) 3
B623 (2002) 247
B621 (2002) 413
B621 (2002) 3
B621 (2002) 571
B622 (2002) 46
B621 (2002) 571
Naumov, D.
Necco, S.
Ndlec, P.
Nefedov, Yu.
Nepomechie, R.I.
Neupane, I.P.
Nguyen-Mau, C.
Nitta, M.
NOMAD Collaboration
B621 (2002) 3
B622 (2002) 328
B621 (2002) 3
B621 (2002) 3
B622 (2002) 615
B621 (2002) 388
B621 (2002) 3
B623 (2002) 133
B621 (2002) 3
Ohlsson, T.
Oller, J.A.
Orestano, D.
zpineci, A.
B622 (2002) 429

B622 (2002) 279
B621 (2002) 3
B621 (2002) 479
Palla, L.
Palla, L.
Pando Zayas, L.A.
Park, J.-S.
Pastore, F.
Peak, L.S.
Peloso, M.
Pennacchio, E.
Pessard, H.
Petrov, A.Yu.
Petti, R.
Pich, A.
Pilo, L.
Placci, A.
Pleiter, D.
Polesello, G.
Pollmann, D.
Polyarush, A.
B622 (2002) 548

B622 (2002) 565
B622 (2002) 257
B621 (2002) 689
B621 (2002) 3
B621 (2002) 3
B622 (2002) 393
B621 (2002) 3
B621 (2002) 3
B621 (2002) 179
B621 (2002) 3
B622 (2002) 279
B622 (2002) 227
B621 (2002) 3
B623 (2002) 287
B621 (2002) 3
B621 (2002) 3
B621 (2002) 3
515
Ponsot, B.
Pope, C.N.
Popov, B.
Poulsen, C.
Provero, P.
Pusztai, B.G.
B622 (2002) 309

B623 (2002) 3
B621 (2002) 3
B621 (2002) 3
B623 (2002) 474
B621 (2002) 622
Raidal, M.
Rakow, P.E.L.
Rattazzi, R.
Reinhardt, H.
Restuccia, A.
Rico, J.
Riemann, P.
Riotto, A.
Riva, V.
Roda, C.
Romanino, A.
Rubbia, A.
B621 (2002) 208

B623 (2002) 287
B622 (2002) 227
B621 (2002) 131
B622 (2002) 240
B621 (2002) 3
B621 (2002) 3
B623 (2002) 97
B621 (2002) 571
B621 (2002) 3
B622 (2002) 73
B621 (2002) 3
Sahakian, V.
Salvatore, F.
Sarkar, S.
Satz, H.
Savc, M.
Schaefer, S.
Schfer, A.
Schahmaneche, K.
Schierholz, G.
Schmidt, B.
Schmidt, T.
Schweigert, C.
Sevior, M.
Sierra, G.
Signer, A.
Sillou, D.
Soler, F.J.P.
Sommer, R.
Sommer, R.
Sorbo, L.
Sozzi, G.
Steele, D.
Stelle, K.S.
Stiegler, U.
Stipcevic, M.
Stirling, W.J.
Stolarczyk, Th.
Strumia, A.
Strumia, A.
Sugiyama, K.
Suyama, T.
B621 (2002) 62
B621 (2002) 3
B621 (2002) 495
B623 (2002) 493
B621 (2002) 479
B623 (2002) 287
B623 (2002) 287
B621 (2002) 3
B623 (2002) 287
B621 (2002) 3
B621 (2002) 3
B622 (2002) 269
B621 (2002) 3
B622 (2002) 593
B621 (2002) 257
B621 (2002) 3
B621 (2002) 3
B622 (2002) 328
B623 (2002) 271
B622 (2002) 393
B621 (2002) 3
B621 (2002) 3
B623 (2002) 3
B621 (2002) 3
B621 (2002) 3
B621 (2002) 257
B621 (2002) 3
B622 (2002) 73
B623 (2002) 395
B622 (2002) 3
B621 (2002) 235
Takcs, G.
Takcs, G.
B622 (2002) 548

B622 (2002) 565
516
Tareb-Reyes, M.
Taylor, G.N.
Tereshchenko, V.
Teschner, J.
Tobe, K.
Toldr, R.
Tonasse, M.D.
Tonasse, M.D.
Toropin, A.
Tth, G.Zs.
Touchard, A.-M.
Tovey, S.N.
Tran, M.-T.
Travaglini, G.
Trincherini, E.
Tsesmelis, E.
Tseytlin, A.A.
Tsujikawa, S.
B621 (2002) 3
B621 (2002) 3
B621 (2002) 3
B622 (2002) 309
B623 (2002) 395
B621 (2002) 495
B623 (2002) 316
B623 (2002) 325
B621 (2002) 3
B622 (2002) 548
B621 (2002) 3
B621 (2002) 3
B621 (2002) 3
B621 (2002) 101
B622 (2002) 227
B621 (2002) 3
B621 (2002) 179
B622 (2002) 393
Ulrichs, J.
B621 (2002)
Vacavant, L.
Valdata-Nappi, M.
Valuev, V.
Van Gulik, R.
Vannucci, F.
Varvell, K.E.
B621 (2002) 3
B621 (2002) 3
B621 (2002) 3
B623 (2002) 220
B621 (2002) 3
B621 (2002) 3
Veltri, M.
Vercesi, V.
Vermaseren, J.A.M.
Vidal-Sitjes, G.
Vieira, J.-M.
Vinogradova, T.
Vogt, A.
Voituriez, R.
B621 (2002) 3
B621 (2002) 3
B621 (2002) 413
B621 (2002) 3
B621 (2002) 3
B621 (2002) 3
B621 (2002) 413
B621 (2002) 675
Wang, Y.-S.
Weber, F.V.
Weisse, T.
Wesson, P.S.
Wilson, F.F.
Winter, W.
Winterhalder, A.
Winton, L.J.
Wu, Y.-S.
B622 (2002) 633

B621 (2002) 3
B621 (2002) 3
B621 (2002) 388
B621 (2002) 3
B622 (2002) 429
B621 (2002) 523
B621 (2002) 3
B622 (2002) 189
Yabsley, B.D.
Yamaguchi, S.
B621 (2002)
B622 (2002)
Zaccone, H.
Zarembo, K.
Zuber, K.
Zuccon, P.
Zwirner, L.
B621 (2002) 3
B623 (2002) 474
B621 (2002) 3
B621 (2002) 3
B621 (2002) 337
3
3

Supersymmetric Domain Walls from Metrics of Special Holonomy

Caricato da

Informazioni sul documento

Descrizione originale:

Titolo originale

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Supersymmetric Domain Walls from Metrics of Special Holonomy

Caricato da

Copyright:

Formati disponibili

Nuclear Physics B 623 (2002) 346

Supersymmetric domain walls from metrics of

Cambridge CB3 OWA, UK

75231 Paris Cedex 05, France

Received 27 November 2001; accepted 12 December 2001

E-mail address: k.stelle@ic.ac.uk (K.S. Stelle).

0550-3213/02/$ see front matter 2002 Published by Elsevier Science B.V.

G.W. Gibbons et al. / Nuclear Physics B 623 (2002) 346

G.W. Gibbons et al. / Nuclear Physics B 623 (2002) 346

G.W. Gibbons et al. / Nuclear Physics B 623 (2002) 346

the curves are orthogonal to a family of d-dimensional surfaces. The congruence we

where 2 = 12 ij ij and ij = tij d1 g rs gtrs gij with d + 1 = dim Mcompact . The

G.W. Gibbons et al. / Nuclear Physics B 623 (2002) 346

2. Four-dimensional manifolds with SU(2) holonomy

e1 = H 1/2(dz1 + mz3 dz2 ),

it is easily verified that the 2-form,

G.W. Gibbons et al. / Nuclear Physics B 623 (2002) 346

implying that the metric becomes

This agrees, up to coordinate redefinitions, with results in [11].

where the Khler function K given by K = 2H 3/(3m2 ).

G.W. Gibbons et al. / Nuclear Physics B 623 (2002) 346

2.2. Domain-wall as Heisenberg contraction of EguchiHanson

If we define rescaled 1-forms i , according to

then after taking the limit 0, we find that (17)

3 = m(dz1 + mz3 dz2 ).

under which (16) becomes

= 16m4 , H = 1 m2 r 2 and take the Heisenberg limit 0, we find,

G.W. Gibbons et al. / Nuclear Physics B 623 (2002) 346

If 1 = 2 = 3 = 0, we get the equations of motion for hyper-Khler metrics with

This gives rise to the first-order equations

G.W. Gibbons et al. / Nuclear Physics B 623 (2002) 346

2.4. Hypersurface-orthogonal homotheties

is a special kind of conformal Killing vector

which would in general

G.W. Gibbons et al. / Nuclear Physics B 623 (2002) 346

R3 . We the write the metric

where V and Ai depend only on the x i . In the orthonormal basis

the spin connection is then given by

Thus if we send R to infinity, while setting q = m and adjusting c such that c + qR = 1,

G.W. Gibbons et al. / Nuclear Physics B 623 (2002) 346

It is easy to establish from (40) that a solution for A is then

3. Orientifold planes and the AtiyahHitchin metric

G.W. Gibbons et al. / Nuclear Physics B 623 (2002) 346

ds 2 = dt 2 + a 2 (t)12 + b2(t)22 + c2 (t)32 ,

4. Domain-walls from pure gravity

G.W. Gibbons et al. / Nuclear Physics B 623 (2002) 346

12D , where ds12D

4.1. Six-dimensional manifolds with SU(3) holonomy

where H = 1 + m|y|. Oxidising back to D = 11, we get the eleven-dimensional metric

It can then be seen that the Khler form is given by

G.W. Gibbons et al. / Nuclear Physics B 623 (2002) 346

and the Khler function is K = H 4 /m2 .

In the notation of Appendix B.1 we thus have

dx dx = m(dz3 dz2 + dz5 dz4 ),

with the corresponding Lie algebra given by (B.1).

Defining the orthonormal basis

e1 = H 1/2 (dz1 + mz4 dz3 ),

e2 = H 1/2 (dz2 + mz5 dz3 ),

G.W. Gibbons et al. / Nuclear Physics B 623 (2002) 346

we find that the torsion-free spin connection is given by

G.W. Gibbons et al. / Nuclear Physics B 623 (2002) 346

The domain-wall metric in D = 5 is

It is easy to establish from (40) that a solution for A is then