Anda di halaman 1dari 639

Nuclear Physics B 602 (2001) 322

www.elsevier.nl/locate/npe

A warped supersymmetric standard model


Tony Gherghetta a , Alex Pomarol b
a IPT, University of Lausanne, CH-1015 Lausanne, Switzerland
b IFAE, Universitat Autnoma de Barcelona, E-08193 Barcelona, Spain

Received 10 January 2001; accepted 15 May 2001

Abstract
We study the breaking of supersymmetry in five-dimensional (5d) warped spaces, using the
RandallSundrum model as a prototype. In particular, we present a supersymmetry-breaking
mechanism which has a geometrical origin, and consists of imposing different boundary conditions
between the fermions and bosons living in the 5d bulk. The scale of supersymmetry breaking
is exponentially small due to the warp factor of the AdS metric. We apply this mechanism to a
supersymmetric standard model where supersymmetry breaking is transmitted through the AdS bulk
to matter fields confined on the Planck brane. This leads to a predictable superparticle mass spectrum
where the gravitino mass is 103 eV and scalar particles receive masses at the one-loop level via
bulk gauge interactions. We calculate the mass spectrum in full detail using the 5d AdS propagators.
The AdS/CFT correspondence suggests that our 5d warped model is dual to the ordinary 4d MSSM
with a strongly coupled CFT sector responsible for the breaking of supersymmetry. 2001 Elsevier
Science B.V. All rights reserved.

1. Introduction
The standard model is believed to be an effective theory valid up to some energy
scale near the electroweak scale. What lies beyond the standard model has been the
subject of active research. Among the possible candidates, there are technicolor theories,
supersymmetry, and, recently, extra dimensions [1].
Extra dimensions and supersymmetry present an additional motivation. They could
be an important ingredient in the underlying theory that includes a quantum description
of gravity, and in particular for string theory they play a crucial role. A particularly
interesting extra dimension scenario is the RandallSundrum model [2]. In this model
the extra dimension is compactified in a slice of anti-de-Sitter (AdS) space, and, as a
E-mail address: tony.gherghetta@cern.ch (T. Gherghetta).
0550-3213/01/$ see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 2 7 - 4

T. Gherghetta, A. Pomarol / Nuclear Physics B 602 (2001) 322

consequence, the electroweak scale is generated by an exponential warp factor in the


metric. This model can be supersymmetrized [35] providing an interesting alternative to
the minimal supersymmetric standard model (MSSM), and a possible connection to string
theories [6].
In this article we want to continue the study of supersymmetric extensions of the
standard model living in five dimensions where the extra dimension is compactified as
in the RandallSundrum model [3,7]. In particular, we want to study supersymmetry
breaking. A warped extra dimension allows for new ways of breaking supersymmetry. The
particular mechanism that we will consider here consists of imposing different boundary
conditions between the fermions and bosons. This supersymmetry breaking mechanism has
been previously studied in flat space but not in warped spaces. In warped space this way of
breaking supersymmetry leads to novel phenomenological consequences. For example, the
radius of compactification does not need to be large (TeV1 ) as in the case of flat space [8].
Therefore it can be consistent with a large cut-off scale that is related with the Planck
scale, MP , or grand unified theories [7,9]. As we will show, the scale of supersymmetry
breaking can be very low ( TeV) and this implies a superlight gravitino m3/2 103 eV.
Scalars are massless at tree-level and get masses at the one-loop level. We will study in
detail a warped version of the MSSM, where gravity and gauge bosons live in the fivedimensional (5d) AdS bulk, while matter fields are located on one of the boundaries, the
Planck brane. In this warped MSSM the squark and slepton masses arise at one-loop from
the gauge interactions and are therefore naturally flavor independent. One of the most
interesting properties of the model is its predictivity of the low-energy mass spectrum.
We will calculate it here in full detail. Although we present the calculation for a particular
extension of the standard model, the calculation of quantum effects in warped spaces that
we present here is much more general and can also be useful for other scenarios.
Another important motivation for the study of the MSSM in a slice of AdS arises from
the AdS/CFT correspondence [10]. This conjecture suggests that these 5d models have a
strongly-coupled 4d dual [1114]. Therefore, the study of the weakly coupled 5d gravity
theory here will be helpful in understanding supersymmetric 4d theories with a strongly
coupled sector. We will comment later on this duality.
In Section 2 we introduce the RandallSundrum compactification and its supersymmetric version. In particular we also analyze the gravitino KaluzaKlein decomposition since
it is the only field not considered in Ref. [3] (see also [15]). In Section 3, we present the
supersymmetry-breaking mechanism, which is based on imposing twisted boundary conditions for fermions in the bulk, and comment on the differences compared with the case
of a flat extra dimension. In Section 4 we introduce a version of the MSSM living in a slice
of AdS and calculate the sparticle mass spectrum, at tree-level and at one-loop level. We
will also comment on the holographic interpretation of the model. Our concluding remarks
appear in Section 5. Finally in the appendix, we present a detailed calculation of the 5d
propagators in a slice of AdS.

T. Gherghetta, A. Pomarol / Nuclear Physics B 602 (2001) 322

2. The warped supersymmetric brane-world


We will consider the scenario of Ref. [2], which is based on a nonfactorizable 5d
geometry. The fifth dimension y is compactified on an orbifold, S 1 /Z2 of radius R,
with 0  y  R. The boundary of the 5d spacetime consists of two 3-branes located
at the orbifold fixed points y = 0 and y = R. This configuration with the 5d metric
solution [2]
ds 2 = e2 dx dx + dy 2 gMN dx M dx N ,

(1)

is a slice of AdS space, where = k|y| and 1/k is the AdS curvature radius. The 5d
coordinates are labelled by capital Latin letters, M = (, 5) where = 0, . . . , 3. The
complete supergravity action for this configuration is obtained by including the gravitino
and graviphoton together with the graviton [35]. However, for the discussion of
supersymmetry breaking it will suffice to only consider the additional gravitino kinetic
and mass terms, which are given by [3]
S = S5 + S(0) + S(R) ,



S5 = d 4 x dy g



j
i MNP
i MN

DN Pi i 32

(3 )ij N ,
12 M53 R + i




S(y ) = d 4 x g4 L(y ) (y ) ,

(2)

where g4 is the induced metric on the 3-brane located at y , and M1 M2 ...Mn =


1 [M1 M2
Mn ] is the antisymmetrized product of gamma matrices. We have defined
n!

= d/dy. Supersymmetry automatically ensures the bulk/boundary conditions (0) =
(R) = /k. The action contains the 5d Planck scale M5 , the 5d Ricci scalar R, two
i (i = 1, 2), and a bulk cosmological constant . At
symplectic-Majorana gravitinos, M
y = 0 the effective 4d mass scale is of order of the Planck scale, MP2  M53 /k, and we
will refer to the brane there as the Planck brane. Similarly, at y = R the effective mass
scale is of order MP ekR , which is the TeV scale for kR  11. Consequently the 3-brane
located there will be referred to as the TeV brane. The index i labels the fundamental
representation of the SU(2)R automorphism group of the N = 1 supersymmetry algebra in
five dimensions. The gravitino supersymmetry transformation is given by [3]
i
= DM i +
M


M (3 )ij j ,
2

(3)

where the symplectic-Majorana spinor i is the 5d supersymmetry parameter.


Similarly, gauge bosons and matter can be added to the bulk [3,9,1618]. In a 5d
supersymmetric theory they form part of vector supermultiplets and hypermultiplets. The
behavior of these supermultiplets in the background of Eq. (1) was considered in Ref. [3],
where the KaluzaKlein mass spectrum was also derived. Only the analysis of the gravitino
field was not presented in Ref. [3]. For this reason, we will present below the KaluzaKlein
decomposition of the gravitino. This will also help to show how the superHiggs mechanism

T. Gherghetta, A. Pomarol / Nuclear Physics B 602 (2001) 322

operates level-by-level in the KaluzaKlein modes, and will help to better understand the
supersymmetry breaking mechanism presented in the next section.
2.1. KaluzaKlein decomposition of the gravitino and the superHiggs mechanism level
by level
Let us start by decomposing the 5d gravitino, M , and the 5d supersymmetry
parameter, , into 4d KaluzaKlein fields
L,R (x , y) =

(n)

(n)

L,R (x ) fL,R (y),

n=0

5 L,R (x , y) =

(n)

(n)

5 L,R (x )f5 L,R (y),

n=0

L,R (x , y) =

(n)

(n)

L,R (x )fL,R (y).

(4)

n=0

We have dropped the SU(2)R index i, since we need only consider the i = 1 component
(the i = 2 component is simply obtained from the symplectic-Majorana condition). We
have also defined 5 L,R = L,R . It is important to note that we have chosen the
y-dependent wavefunction of the supersymmetry parameter to be the same as that for
the KaluzaKlein gravitinos.
2.1.1. KaluzaKlein modes n = 0
The supersymmetry transformation Eq. (3) for i = 1 gives


1 5
= +
,
2
5
5 = 5 + .
2

(5)

Substituting Eq. (4) into Eq. (5) and projecting out the nth mode, 1 we find that the
supersymmetry transformation for the nth KaluzaKlein gravitino mode is given by
(n)
(n)L = L
+

(k)
ank R
,

(6)

k=0
(n)

(n)

R = R ,
where is the 4d Minkowski gamma matrix and the coefficients ank are given by

(n)
(k)
ank dy e2 fL (y)fR (y).

(7)

(8)

1 This corresponds to multiplying each side of Eq. (5) by f (n) , integrating over y, and using the gravitino
L,R

(n) (m)
orthogonality condition dy e fL,R fL,R = nm .

T. Gherghetta, A. Pomarol / Nuclear Physics B 602 (2001) 322

The coefficients ank imply that the supersymmetry transformation of (n)L at level n,
(k)
depends nontrivially on the complete tower of KaluzaKlein parameters R
. This effect
is entirely due to the fact that the bulk is AdS. Let us now impose the following relation
for the wavefunctions of 5 :

f5(n)
L,R =

 (n)
1 
,
5 + 12 fL,R
mn

(9)

where mn is the 4d mass of the gravitino KaluzaKlein mode n, which will be derived
below. The condition (9) allows us to write a simple expression for the supersymmetry
(n)
transformation of KaluzaKlein modes 5
(n)
5(n)
L = mn L ,

(10)

(n)
5 R

(11)

(n)
= mn R .

This shows that the nth KaluzaKlein mode of the 5th component of the gravitino
transforms as a Goldstino of the (n) supersymmetry transformation and that these N = 2
supersymmetries are nonlinearly realized. We can now see that the redefined gravitinos
(n)L mn (n)L 5(n)
L + mn


k=0

(k)

ank

5 R
,
mk

(n)
(n)
(n)
R mn R + 5 R ,

(12)

are invariant under supersymmetry transformations, and therefore correspond to the


physical fields. On the contrary, the fields 5(n) are gauge dependent and can be eliminated.
(n)
(n)
This is the superHiggs mechanism. The 5 are eaten by the gravitino to become
massive.
Let us now turn to the RaritaSchwinger equation for the bulk gravitino, which in the
AdS background reads
3
MNP DN P MP P = 0.
2
Using the redefined gravitino fields (12), the equation of motion (13) simplifies to
(n)
R,L = 0,
(n)
L,R mn

(13)

(14)

which represents the 4d massive RaritaSchwinger equation for the spin 3/2 field ,
and where the y-dependent KaluzaKlein wavefunctions satisfy


5 + 12 fL(n) = mn e fR(n) ,
(15)


(n)
(n)
5 52 fR = mn e fL .
(16)
(n)

(n)

One can see that the dependence on 5 has dropped out and the equation of motion
(n)
depends, as expected, only on . The solutions of Eqs. (15) and (16) are a special case
(n)
(n)
= e fL,R
one can
of the general solution appearing in Ref. [3]. In fact, defining fL,R
(n)
see that f
corresponds to the wavefunction of a hatted fermion of mass m = 3 /2
L,R

T. Gherghetta, A. Pomarol / Nuclear Physics B 602 (2001) 322

defined in Ref. [3]. Thus, using the results in Ref. [3] and the fact that L (R ) are
defined even (odd) under the Z2 -parity, we obtain the y-dependent gravitino wavefunctions





1 3
mn
mn
(n)
2
e + b(mn )Y2
e
e
J2
,
fL =
(17)
Nn
k
k





3
mn
mn
(n)
e + b(mn )Y1
e
fR =
(18)
e 2 J1
,
kNn
k
k
where J and Y are Bessel functions, Nn are normalization constants and the coefficients b(mn ) satisfy
J1 ( mkn )
,
Y1 ( mkn )


b(mn ) = b mn ekR .

b(mn ) =

(19)
(20)

The KaluzaKlein masses of the gravitinos (n) can be obtained by solving (20), and for
n > 0 they are approximately given by


mn  n + 14 kekR .

(21)

Finally, using (15) and (16), and the fact that 5 R (5 L ) are even (odd) under the Z2 -parity,
we have from the condition (9)
(n)
f5(n)
L = e fR ,
(n)

(n)

f5 R = e fL

(22)
2
mn

(n)

fR .

(23)

2.1.2. Massless sector


The y-dependence of the gravitino zero-mode wavefunction is obtained from Eq. (15),
(0)
(0)
since under the orbifold symmetry, fR is projected out. Thus for the remaining mode, fL
with m0 = 0, we obtain
1
1
(0)
fL (y) = e 2 ,
N0

(24)

where the normalization factor N0 = (1 e2kR )/k. This is consistent with the ydependence of the graviton zero-mode wavefunction, as expected from supersymmetry.
(0)
(0)
Similarly, R is projected out and L whose wavefunction is also given by Eq. (24)
parametrizes the remaining N = 1 supersymmetry of the theory. In fact one can check that
Eq. (24) satisfies the Killing spinor condition.
Similarly, for the fifth-component of the gravitino, we have that 5(0)
L is projected out
(0)
and only 5 R remains in the theory. This corresponds to the supersymmetric partner of the
radion, the radino. The 4d effective Lagrangian of this field has been recently presented
in Ref. [19].

T. Gherghetta, A. Pomarol / Nuclear Physics B 602 (2001) 322

3. Supersymmetry breaking in a slice of AdS


Different mechanisms of supersymmetry breaking in brane-world scenarios have been
considered in the past. The most popular, based on the HoravaWitten model, corresponds
to breaking supersymmetry in a hidden-sector living on a brane located at a finite distance
from the observable-sector brane [20]. The moduli (e.g., the dilaton and radion) play the
role of messengers communicating the supersymmetry-breaking from the hidden to the
observable sector. These scenarios rely on gaugino condensation to occur on the hiddensector brane in order to explain the hierarchy.
Warped (AdS) spaces allow for new possibilities. First of all, since the hierarchy is now
explained by the warp factor, one does not need a gaugino condensation in a hidden sector
to be responsible for a small supersymmetry breaking. Supersymmetry can be broken at
tree-level if it occurs on the TeV brane, and therefore have a stringy origin.
The supersymmetry-breaking mechanism that we will consider here is based on
imposing different boundary conditions between fermions and bosons on the TeV brane.
This breaks supersymmetry for the bulk fields, and as we shall see, the Kaluza
Klein fermions and bosons receive TeV mass-splittings. The mechanism consists of the
following. The 5d bulk fields in the supersymmetric Z2 orbifold can be classified as either
odd or even fields under the Z2 parity. For the 5d fermions, we have two possibilities to
define the Z2 parity, namely
(y) = 5 (y).

(25)
y

= 0, since
Once a choice is made, this also defines the chirality on the 4d boundary at
(0) = 5 (0). For the supersymmetric Z2 orbifold the same chirality is chosen on the
two boundaries at y = 0 and y = R. In this way only half of the bulk supersymmetry is
broken by the boundaries, leaving an N = 1 supersymmetric theory at the massless level.
However, there also exists the possibility to separately define the chirality of fermions on
the two boundaries. For example, the choice
(0) = 5 (0),

(R) = 5 (R),

(26)

corresponds to the following y-dependence


(y) = 5 (y),

(y + R) = 5 (y + R).

(27)

Thus, under a 2 rotation around the circle S 1 , Eq. (27) leads to fermions that are
antiperiodic
(y + 2R) = (y).

(28)

As will be shown below, the boundary conditions (26) project out the massless fermion
modes arising from bulk fields. Also supersymmetry is now completely broken since no
Killing spinor can be defined. 2
The fermionic boundary conditions (26) have been considered previously in the
literature. If the space of the extra dimension is flat, imposing these boundary conditions
2 This has some similarities with finite temperature which also breaks supersymmetry [21].

10

T. Gherghetta, A. Pomarol / Nuclear Physics B 602 (2001) 322

exactly corresponds to breaking supersymmetry by the ScherkSchwarz mechanism [22].


This mechanism has been applied to the MSSM in Refs. [8,2326]. In the HoravaWitten
theory this was studied in Ref. [27]. However, for warped spaces the fermionic boundary
conditions Eq. (26) do not correspond to the ScherkSchwarz mechanism, because this
requires a smooth limit where supersymmetry is restored [22]. In the case of warped spaces,
we have not found such a smooth limit. 3
Let us now study the fermionic spectrum with the boundary conditions (26), which we
will refer to as twisted boundary conditions. The resulting KaluzaKlein mass spectrum
for L is now determined by (see appendix)
J1 ( mkn ) J ( mkn ekR )
=
,
Y1 ( mkn ) Y ( mkn ekR )

(29)

where = |c + 1/2| for a fermion of Dirac-type mass m = c and = 2 for the


gravitino. One can easily check that the equation resulting from imposing twisted boundary
conditions on R leads to an identical KaluzaKlein spectrum. The first thing to notice in
Eq. (29) is that mn = 0 is no longer a solution of the above equation and therefore no
massless fermions are present. Consequently, supersymmetry is now completely broken.
(0)
, whose wavefunction is identical to Eq. (24),
One can also see that the Killing spinor, L
is not consistent with the new boundary conditions. Thus, the only change with respect
to Section 2.1 is that Eq. (20) is now replaced by Eq. (29), and the massless sector of
Subsection 2.1.2, is no longer present in the theory. Notice that from Eq. (10) the Goldstino
(0)
of the broken N = 1 supersymmetry is 5 L .
In the limit mn  k and kR  1, the solution of Eq. (29) is given by


1
kekR .
mn  n +
(30)
2 4
Comparing with the result for untwisted boundary conditions [3], one finds that
the KaluzaKlein mass spectrum is shifted by a value that asymptotically approaches
1/2(kekR ). This is to be contrasted with the flat case where the shift in the Kaluza
Klein mass spectrum is 1/(2R).
There is an important difference when this type of supersymmetry breaking is realized in
warped spaces compared to the flat case. In flat spaces this type of supersymmetry breaking
is global. To see this, let us consider an observer living on one of the branes with the other
brane sent to infinity (R ). In this limit and in flat space, supersymmetry is restored
because the KaluzaKlein spectrum becomes continuous (the scalar-fermion mass splitting
disappears). This is related to the fact that one can locally (i.e., on either brane) define
a supersymmetric theory. Breaking supersymmetry globally (when the extra dimension
is compact, no Killing spinor can be defined in the whole space) leads to the important
3 Similarly, one can also show that the models of Refs. [24,25] with only one Higgs hypermultiplet (instead of
two) do not have this smooth limit to a supersymmetric theory. This alternative has recently been considered in
Ref. [26], where it was shown that the boundary conditions Eq. (26) can also be understood as compactifying on
a S 1 /(Z2 Z2 ) orbifold.

T. Gherghetta, A. Pomarol / Nuclear Physics B 602 (2001) 322

11

property that the vacuum energy and the one-loop scalar masses are finite and independent
of the cut-off scale [25,28].
In warped spaces the situation is different and the finiteness of the one-loop scalar
masses depends on which particular brane the observable sector lives. Consider first the
observable sector on the Planck brane where the TeV brane is sent off to infinity. In this
limit supersymmetry is restored on the Planck brane because the KaluzaKlein spectrum
becomes continuous. Therefore the one-loop scalar masses on the Planck brane will be
finite. Alternatively, suppose that the observable sector is on the TeV brane. Now, even if
we consider the limit where we send the Planck brane away (R ), the KaluzaKlein
spectrum remains discrete and supersymmetry stays broken. Therefore on the TeV brane
supersymmetry is broken and corrections to scalar masses will be sensitive to the ultraviolet
cut-off. Another way to see that supersymmetry is broken by the TeV brane (contrary to
the flat case) is that no Killing spinor can be defined if fermions have twisted boundary
conditions. Even in a noncompact space, the TeV-boundary breaks all the supersymmetries.
These expectations will be confirmed in the following sections by the explicit calculation
of the one-loop scalar masses in a warped AdS space.
Finally, as an alternative to the supersymmetry-breaking mechanism considered above,
there also exists the possibility of breaking supersymmetry by the F -term, FT of the
radion field T . This can easily be achieved by turning on a constant term, W , in the
superpotential localized on the TeV brane. In flat space this is known to generate a vacuum
expectation value (VEV) FT  W/M53 . In fact in flat space, this corresponds exactly
to the ScherkSchwarz mechanism [29] or to imposing the twisted boundary condition
for the fermion as in Eq. (26). However, in a warped space this is not the case, and a
nonzero FT  leads to a new way of breaking supersymmetry. Furthermore, in a warped
space the VEV of FT , induced by a constant term in the superpotential at the TeV brane, is
exponentially suppressed, FT  ekR W/M53 . The tree-level spectrum is easily derived.
For the gaugino we have m FT /T  TeV, while for scalars localized on either brane
their masses are zero. The scalar masses are, however, induced at the one-loop level. This
scenario leads, qualitatively, to the same mass spectrum as the one considered above, and
will not be pursued here.

4. The warped MSSM


Let us now present a candidate MSSM based on the 5d model described above. We will
assume that both gravity and gauge fields are in the bulk. Supersymmetry is spontaneously
broken by imposing twisted boundary conditions (26), on the gravitino and gaugino. The
MSSM matter fields are assumed to be completely localized on the Planck brane. At tree
level, the matter fields are massless and the dominant supersymmetry-breaking effects will
be transmitted to the matter fields on the Planck brane by the 5d gauge interactions. Thus,
the soft masses on the Planck brane will arise via radiative corrections and can be computed
using the 5d AdS propagators. On the other hand, the Higgs field can also be assumed to
be a bulk field. However, in this case we will see that radiative corrections to the Higgs

12

T. Gherghetta, A. Pomarol / Nuclear Physics B 602 (2001) 322

soft masses are sensitive to TeV-scale physics. Other alternative scenarios will also be
discussed. Since the bulk fields live in a warped space, these models will be referred to as
the warped MSSM.
Before proceeding to calculate the sparticle spectrum an important comment is in order.
Since the gauge bosons live in a warped extra dimension, the effective 4d coupling is given
by g 2 = (g52 k)/(kR), where g5 is the 5d gauge coupling [9,17]. In order to explain the
PlanckTeV scale hierarchy we need kR  10, which implies that for g52 k  1, we obtain
g 2  1/30. This is in contradiction with the experimental values of the gauge couplings
which require g 2 O(1). Therefore, in order to agree with the experimental values one
requires that g52 k  30. This inevitably means that the theory is close to the strong coupling
regime at energies E k. On the TeV brane this corresponds to energies E kekR . At
these energies the expansion parameter becomes g52 k/(16 2 ) 0.2. We will assume that
the effects from the strong coupling regime do not spoil the AdS geometry. Similarly, we
will be able to trust our low-energy predictions, provided that the energy of the processes
satisfies E  kekR .
4.1. Tree-level masses
If twisted boundary conditions are imposed on the fermions in the bulk, then all the
4d fermion modes will receive masses. In particular, the zero mode of the gravitino will
receive a mass whose magnitude can easily be obtained by solving (29) for = 2:

m3/2  8 ke2kR .
(31)
Thus, for k = MP and kekR = TeV we obtain m3/2  2.8 103 eV. This is a superlight
gravitino, as compared to the usual gravity-mediated and gauge-mediated scenarios in four
dimensions, and satisfies the usual experimental constraints from cosmology and collider
experiments [30]. In the warped case the small gravitino mass arises because the coupling
of the gravitino to the TeV brane is exponentially suppressed, and therefore it is very
insensitive to the twisting of boundary conditions on the TeV brane. The higher Kaluza
Klein gravitino modes are approximately given by


mn  n + 34 kekR .
(32)
Notice that compared to the untwisted gravitino KaluzaKlein spectrum (21), the twisted
mass spectrum has indeed shifted approximately by an amount 1/2(kekR ). This shift
is much larger than for the zero mode, because the nonzero KaluzaKlein gravitino modes
are localized near the TeV brane and therefore couple more strongly to the TeV brane as
compared to the gravitino zero mode which is localized near the Planck brane.
Similarly, the tree-level gaugino mass is obtained by solving (29) with = 1:

2
kekR .
m 
(33)
kR
Thus, for k = MP and kekR = TeV we obtain m  0.24 TeV. Notice that unlike
the gravitino zero mode, the gaugino zero mode receives a TeV-scale supersymmetry

T. Gherghetta, A. Pomarol / Nuclear Physics B 602 (2001) 322

13

breaking mass. This is because the vector superfield is not localized in the AdS space,
and therefore directly couples to the TeV brane, which is the source of the supersymmetry
breaking. Using (30), the higher KaluzaKlein modes are approximately given by mn 
(n+1/4)kekR . These masses are obtained at tree-level and we will see that interactions
of boundary fields with the bulk gauge bosons will generate boundary masses at one-loop.
Since the mediation of the supersymmetry breaking is due to gauge interactions, the flavor
problem is naturally solved. It is important to note that the theory has a U(1)R symmetry,
since the induced masses are of the Dirac-type instead of the Majorana-type. This is a
unique property of these theories, and is due to the N = 2 bulk supersymmetry.
It is also possible to add hypermultiplets in the bulk, where the fermions have twisted
boundary conditions. In particular, if c = 1/2 then the hypermultiplet is conformal, and the
resulting KaluzaKlein spectrum is identical to the vector supermultiplet case.
4.2. Radiative corrections on the Planck brane
In order to compute the radiative corrections of the matter fields completely confined on
the Planck brane, let us consider the 5d AdS propagator for the gauge boson and gaugino.
The general expression for the 5d propagator in a slice of AdS is derived in the appendix.
Using the expression for the vector field Greens function restricted to the Planck brane
(z = z = 1/k), we have


1
1
GV x, ; x ,
k
k

4
d p ip(xx ) 1 J0 (ipekR /k)Y1 (ip/k) Y0 (ipekR /k)J1 (ip/k)
. (34)
e
=
ip J0 (ipekR /k)Y0 (ip/k) Y0 (ipekR /k)J0 (ip/k)
(2)4
In the limit that p  kekR we obtain


 1
1
1
d 4 p ip(xx ) 1
GV x, ; x ,
e
,

k
k
R
(2)4
p2

(35)

which reduces to the usual massless vector field Greens function in flat space. In particular
notice that by Eq. (35), the charge screening [7,31] is absent in the slice of AdS since there
are no continuum KaluzaKlein modes. Similarly on the Planck brane, the twisted gaugino
Greens function defined in the appendix reduces to the form


1
1
GF x, ; x ,
k
k

4
d p ip(xx ) 1 J1 (ipekR /k)Y1 (ip/k) Y1 (ipekR /k)J1 (ip/k)
. (36)
=
e
(2)4
ip J1 (ipekR /k)Y0 (ip/k) Y1 (ipekR /k)J0 (ip/k)
In the limit that p  kekR and kR  1 the twisted gaugino Greens function becomes



1
1
1
1
d 4 p ip(xx )

e
,
GF x, ; x ,
(37)
2
2
k
k
R
(2)4
p kR (kekR )2

14

T. Gherghetta, A. Pomarol / Nuclear Physics B 602 (2001) 322

which reduces to a massive gaugino Greens function where the gaugino mass agrees
with (33). This difference between the gauge boson and gaugino Greens function
represents the source of supersymmetry breaking on the Planck brane.
Note that the vector supermultiplet in the bulk, is also equivalent to a conformal
hypermultiplet (c = 1/2) in the bulk, where the fermion has twisted boundary conditions.
The 5-dimensional mass-squared of the scalar is 3k 2 + 2k((y) (y R)), while the
mass of the fermion is /2 [3]. On the Planck brane (z = z = 1/k) the twisted fermion
Greens function is the same as Eq. (36), while the scalar field Greens function is identical
to Eq. (34). In particular, we will also consider the bulk Higgs fields to be conformal
hypermultiplets.
The scalar and twisted fermion Greens function on the Planck brane can be used
to calculate the one-loop contribution to the mass-squared of boundary matter fields.
The boundary matter fields couple to the vector supermultiplet in the bulk via gauge
interactions. The Feynman diagrams for the one-loop mass contributions to the boundary
scalar fields are the same as those in flat space and can be found in Ref. [25]. They give
m2i = 4g 2 C(Ri )(0),

(38)
g2

= g52 /(R)

and we have defined


where the 4d gauge coupling is given by


d 4 p  (V )
)
(0) = R
(39)
G G(F
p .
(2)4 p
The coefficient C(Ri ) is the quadratic Casimir of the representation Ri in the corresponding gauge group, and Gp is the 4d Fourier transform of the Greens function (see appendix). Similarly, the boundary matter fields can also couple to a chiral supermultiplet in
the bulk [25]. For a conformal supermultiplet (c = 1/2), where the fermions have twisted
boundary conditions we obtain
m2i = 2Y 2 (0),

(40)

where Y is the boundary-bulk Yukawa coupling. Assuming that k = MP and kekR =


TeV then we obtain kR = 34.54, and the integral in Eq. (39) can be numerically evaluated
to give
0.2525
(41)
(TeV)2  (0.0360 TeV)2 .
2 4
This result is finite and insensitive to the ultraviolet cut-off for the same reason that we
already explained in the previous section, namely that the supersymmetry breaking is
localized on the TeV brane. In fact, the integration region p  kekR = TeV already
contributes approximately 90% of the integral in Eq. (39).
Comparing the above result with the flat space case where [25] (0) = (0.0367/Rflat)2 ,
we see that the two cases are almost numerically identical for a flat-space radius of Rflat =
kekR .
(0) =

4.2.1. Superparticle spectrum


It is now straightforward to extend the above result to the case of the warped MSSM.
Assuming that the squarks and sleptons live on the Planck brane they will receive a one-

T. Gherghetta, A. Pomarol / Nuclear Physics B 602 (2001) 322

15

loop contribution from the bulk gauge and Higgs sector (if they are in the bulk). Assuming
the bulk Higgs to be a conformal supermultiplet and following Ref. [25] we obtain for
= 16(0)
(0)

4
3
1
1


m2Q
(42)
= 3 3 + 4 2 + 60 1 (0) + 2 (t + b )(0),


2
4
4

+ t (0),
mU = 3 3 + 15 1 (0)
(43)


4
1


m2D
(44)
= 3 3 + 15 1 (0) + b (0),


2
3
3
1
(0),
+
= 4 2 + 20 1 (0)
m
(45)
2
L
+ (0),

m2E = 35 1 (0)

(46)

where the bulk gauge contribution is proportional to the gauge couplings 1,2,3 and the
conformal bulk Higgs contribution is proportional to the Yukawa couplings t,b, . Thus to
obtain an experimentally allowed soft mass spectrum the scale on the TeV brane should
be at least a few TeV. Notice that the dominant corrections are proportional to the gauge
couplings. Thus, the lightest scalar field is the right-handed slepton.
4.3. Radiative corrections in the slice of AdS
Consider a conformal hypermultiplet in the bulk with twisted boundary conditions for
the fermion. The massless scalar mode in the hypermultiplet will receive a one-loop
mass contribution due to the breaking of supersymmetry from fields in the bulk with
the twisted boundary conditions. In particular the scalar can couple to the bulk vector
supermultiplet. This radiative correction can simply be calculated using the 5-dimensional
AdS scalar propagator. For an alternative method to calculate quantum effects in the AdS
slice see Ref. [32]. Since the scalar propagates in the bulk we need to integrate over the
extra dimension, and the corresponding mass correction is proportional to

(0) = R

d 4p
(2)4

R


)
(F )
dy G(V
p (z, z) Gp (z, z)
0

ke2kR

(47)
,
16 2
where is a Planck-scale cutoff. Unlike the radiative corrections of the boundary fields, it
turns out that the radiative correction (47) is not finite, as expected from the arguments of
the previous section. In fact, the bulk radiative corrections (47) are linearly divergent. This
reflects the fact that the bulk fields are propagating in five dimensions and are sensitive to
physics on the TeV brane represented by the cutoff scale ekR . This behavior is related
to the fact that the supersymmetry breaking mechanism is localized on the TeV brane and
is sensitive to the UV physics. This is different from flat space where the breaking of
supersymmetry is inherently a global effect, and consequently the nonlocality produces a
finite result. Although the Higgs soft mass depends on the UV cutoff k M5 , its value
will be of order TeV/(4) (due to the warp factor in Eq. (47)), and therefore the model
is fully viable (the hierarchy is not spoiled). However, the dependence on the UV cutoff

16

T. Gherghetta, A. Pomarol / Nuclear Physics B 602 (2001) 322

means that a precise calculation of the Higgs mass can only be done by having a complete
knowledge of the ultraviolet theory. We must also assume that the Higgs mass will be
negative in order to trigger electroweak symmetry breaking. The fact that the Higgs mass
is a factor of 4 smaller than a TeV gives a natural explanation for the weak scale being
an order of magnitude smaller than the KaluzaKlein masses (of order TeV), as required
by experimental bounds.
Let us finally comment on other possible alternatives. If the Higgs is also confined on
the Planck brane, then its mass will be generated at the one-loop level, with a magnitude
similar to that of the sleptons (without the Yukawa coupling contribution). Although this
contribution will be positive, there are sizeable two-loop effects arising from the squarks
that can make the mass-squared negative [25]. Unfortunately, the Higgsino mass cannot
be generated by radiative corrections and we will need to extend the model to include a
Higgs singlet whose VEV must induce the Higgsino mass. In the above cases we have
restricted the observable sector to the Planck brane. Nevertheless, many more possibilities
exist by placing part of the matter in the bulk or on the TeV brane. For example, consider
delocalizing the first two families off the Planck brane by changing their bulk mass
parameters [3]. In this case the corresponding squarks and sleptons of the first two families
will have masses larger that those of the third family, a scenario whose phenomenology
can have interesting consequences.
4.4. Relation to 4d strongly-coupled CFT
The AdS/CFT correspondence relates the 5d theory of gravity in AdS to a 4d strongly
coupled conformal field theory (CFT) [10]. In the case of a slice of AdS, a similar
correspondence can also be formulated [1114]. The Planck brane in AdS5 corresponds
to an ultraviolet cutoff of the 4d CFT and to the gauging of certain global symmetries.
For example, in the case we are considering where gravity and the standard model gauge
bosons live in the bulk, the corresponding CFT will have the super-Poincar group gauged
(giving rise to gravity) and also the standard model group SU(3) SU(2)L U(1)Y
(giving rise to the standard model gauge bosons and gauginos). Matter on the Planck brane
corresponds to adding new fields to the CFT which only couple to CFT states via gravity
and gauge interactions. On the other hand, the TeV brane corresponds in the dual theory
to a infrared cutoff of the CFT [13,14]. In other words, it corresponds to breaking the
conformal symmetry at the TeV scale. The KaluzaKlein states of the 5d theory correspond
to the bound states of the strongly coupled CFT.
This alternate dual description suggests that the supersymmetry-breaking mechanism
that we have discussed represents a class of strongly coupled CFTs where supersymmetry
is broken at the TeV scale. The bound states therefore do not respect supersymmetry and
give rise to a fermionboson mass splitting. The 5d warped MSSM is then simply the
ordinary 4d MSSM with a strongly coupled CFT sector responsible for the breaking of
supersymmetry. The standard model fields coupled to the CFT sector will get tree-level
masses while those coupled only via gravity or gauge interactions will receive masses
at the one-loop level. In our model the CFT sector is charged under the standard model

T. Gherghetta, A. Pomarol / Nuclear Physics B 602 (2001) 322

17

gauge group and consequently it implies that the gauginos get masses at tree-level. Notice
that as we mentioned earlier the gaugino mass is of the Dirac-type. This means that
the gaugino has married a fermion bound-state to become massive. 4 Since the gaugino
mass comes from the mixing
between the gaugino
and the CFT bound-state, the mass


will be proportional to g 2 bCFT /(8 2 ) = 1/ kR, where we have used the AdS/CFT
relation [13] g52 k = 8 2 /bCFT and g 2 = g52 /(R). This agrees with Eq. (33). Similarly, the
smallness of the gravitino mass (of order 103 eV) is also easy to understand in the CFT
picture. The gravitino coupling to the CFT sector is suppressed by 1/MP , so its mass will
be of order TeV2 /MP 103 eV.
Although the CFT picture is useful for understanding some qualitative aspects of the
theory, it is practically useless for obtaining quantitative predictions since the theory is
strongly coupled. In this sense, the 5d gravitational theory in a slice of AdS represents
a very useful tool since it allows one to calculate the particle spectrum, which would
otherwise be unknown from the CFT side.

5. Conclusion
In this paper we have presented a supersymmetric 5d theory in warped space where
supersymmetry is spontaneously broken by imposing different boundary conditions
between the fermion and bosons. While this is reminiscent of the ScherkSchwarz
mechanism in flat space, we have argued that in a warped space this is a novel way of
breaking supersymmetry. Unlike the flat-space case where the supersymmetry-breaking
mechanism is a global effect, the twisted boundary conditions in the warped space lead to
a local supersymmetry breaking effect on the TeV brane.
A particularly interesting model is the warped MSSM, where matter is confined on the
Planck brane, and gravity and gauge fields propagate in the 5d bulk. The gravitino and
gaugino receive tree-level masses from the twisted boundary conditions. In particular, the
tree-level mass of the gravitino is 103 eV and the gaugino mass TeV. The one-loop
radiative corrections to the squarks and sleptons confined to the Planck brane are finite
and insensitive to the UV cutoff. This simply reflects the fact that the supersymmetrybreaking is localized on the TeV brane, at a finite distance away from the Planck brane.
The one-loop radiative corrections from the bulk gauge fields are proportional to the gauge
couplings and thus naturally solve the flavor problem. If the Higgs sector is also included
in the bulk, then the one-loop radiative corrections also give a contribution proportional
to the Yukawa couplings. However, in this case the radiative corrections to the Higgs soft
mass are not finite. This is in contrast to the flat-space case, and is due to the fact that the
bulk Higgs directly couples to the supersymmetry breaking effects on the TeV brane.
By the AdS/CFT correspondence, the warped supersymmetric standard model can be
interpreted in terms of a strongly coupled CFT, where supersymmetry (and conformal
symmetry) are broken at the TeV scale. Thus, the warped MSSM is simply the ordinary
4 A Majorana-type mass would correspond, for example, to a breaking of supersymmetry (in the 5d dual) by a
nonzero FT .

18

T. Gherghetta, A. Pomarol / Nuclear Physics B 602 (2001) 322

4d MSSM with a strongly coupled CFT which is responsible for breaking supersymmetry.
The fact that there exists a weakly coupled 5d gravity dual, allows us to calculate the mass
spectrum. This provides a powerful tool in obtaining information about the dynamics of
this class of strongly coupled CFTs, and is worthy of further investigation.

Acknowledgements
We wish to thank Emilian Dudas and Dan Waldram for helpful discussions. One of us
(T.G.) acknowledges the Aspen Center for Physics where part of this work was done. The
work of T.G. is supported by the FNRS, contract no. 21-55560.98, while that of A.P. is
partially supported by the CICYT Research Project AEN99-0766.

Appendix A. 5D propagators in a slice of AdS


Let us consider the propagation of bulk fields in a slice of AdS. We will follow the
derivation of the Greens function presented in Refs. [12,33], except that we will extend
the previous results to the case of arbitrary bulk fields in the two-brane scenario. The result
for the bulk scalar has also recently been given in Ref. [34].
As shown in Ref. [3] the equation of motion for bulk fields = {V , , e2 L,R }, can
be conveniently written as a second-order differential equation. Thus, introducing a source
function J , one obtains



 2
e + es 5 es 5 M2 (x, y) = J (x, y),
(A.1)
where the parameter s = {2, 4, 1}, and the 5d masses are [3] M2 = {0, ak 2 + b ,
c(c 1)k 2 c }. The corresponding Greens function for (A.1) can then be defined as



(x, y) = d 4 x dy g G(x, y; x , y )e(4s)ky J (x , y ),
(A.2)
provided that J = {J , J , JR,L 5 JL,R (c 1) JL,R }, where J , J and JL,R
are the source terms for the bulk vector, scalar, and fermion, respectively. It is convenient
to introduce the variable z = eky /k. In these coordinates the Planck brane is located at
z = 1/k and the TeV brane at z = ekR /k. If we now take the 4d Fourier transform of
the Greens function

d 4 p ip(xx )
G(x, z; x , z ) =
(A.3)
e
Gp (z, z ),
(2)4
then the Fourier component Gp (z, z ) must satisfy the equation

2 
M
1s
z p2 2 Gp (z, z ) = (kz)s1 (z z ),
z2 +
z
(kz)

(A.4)

 2 = {0, ak 2, c(c 1)k 2 }. If we define Gp (z, z ) = (k 2 zz )s/2 G


p (z, z ), then
where M

Eq. (A.4) simply becomes the Bessel equation

T. Gherghetta, A. Pomarol / Nuclear Physics B 602 (2001) 322

19



1
2 
2
2
z + z p 2 Gp (z, z ) = (kz)1 (z z ),
(A.5)
z
z

2 /k 2 . The standard procedure for solving Eq. (A.5) is to use the
where = (s/2)2 + M

solution to the homogeneous equation in the regions z < z and z > z , and then impose
matching conditions at z = z . Thus writing
p (z, z ) = (z z )G
> + (z z)G
< ,
G
the solution to the homogeneous equation for z < z is given by


< (z, z ) = iA< (z ) J (ip/k)H(1)(ipz) H
(1)(ip/k)J (ipz) ,
G
and for

z > z

(A.6)

(A.7)

we obtain


 



(1) ipekR /k J (ipz) ,
> (z, z ) = iA> (z ) J ipekR /k H(1)(ipz) H
G

(A.8)

(1)

where H = J + iY is the Hankel function of the 1st kind of order , and J , Y are
> ) is even on
< (G
Bessel functions. If the boundary condition for the Greens function, G
the Planck brane (TeV brane) then [3]
J (z) = (r + s/2)J (z) + zJ (z),

(A.9)

where the parameter r = {0, b, c}, while if the boundary condition is odd then [3]
J (z) = J (z),

(A.10)

(1) .
H

Note that in the presence of the boundary mass terms parametrized


and similarly for
by r, the even boundary condition is equivalent to imposing the modified Neumann
condition, (z r )Gp (z, z )|z=z = 0, while the odd boundary condition is equivalent
to imposing the Dirichlet condition, Gp (z, z )|z=z = 0.
The unknown functions A< (z ) and A> (z ) are determined by imposing matching
p at z = z leads to the condition
conditions at z = z . Continuity of G


>  = G
<  ,
G
(A.11)
z=z
z=z
p gives the condition
while the discontinuity in z G

> z G
<  = 1 .
z G
z=z
kz
This leads to the solutions


(A.12)

A< (z ) =

(1)
(1)(ipekR /k)J (ipz )
J (ipekR /k)H (ipz ) H
,
2k J (ipekR /k)H
(1) (ip/k) H
(1)(ipekR /k)J (ip/k)

(A.13)

A> (z ) =

(1)(ip/k)J (ipz )

J (ip/k)H(1)(ipz ) H
.
2k J (ipekR /k)H
(1) (ip/k) H
(1)(ipekR /k)J (ip/k)

(A.14)

< gives the expression


> and G
Finally substituting these functions into the equations for G
for the Greens function in a slice of AdS

Gp (z, z ) = i k s1 (zz )s/2


2

20

T. Gherghetta, A. Pomarol / Nuclear Physics B 602 (2001) 322


(1)(ipekR /k)J (ipz> )
J (ipekR /k)H(1)(ipz> ) H
(1) (ip/k) H
(1)(ipekR /k)J (ip/k)
J (ipekR /k)H


(1)(ip/k)J (ipz< ) ,
J (ip/k)H(1)(ipz< ) H

(A.15)

where we have defined z> (z< ) to be the greater (lesser) of z and z . The Greens
function (A.15) is the general expression for arbitrary bulk fields in a slice of AdS.
The KaluzaKlein mass spectrum can be obtained from the pole condition of the Greens
function, namely


 (1)

(ip/k) H
(1) ipekR /k J (ip/k) = 0.
J ipekR /k H
(A.16)
This leads to the condition
J ( mk ) J ( mk ekR )
=
,
( m ) Y
( m ekR )
Y
k
k

(A.17)

where the four-momentum p2 = m2 . The solutions of this equation for the various
combinations of boundary conditions on the Planck and TeV branes, reproduces all the
KaluzaKlein mass spectrum results from [3].
For the twisted boundary condition, (0) = (0) and (R) = (R), we must
impose on the corresponding Greens function the even boundary condition on the Planck
brane, Eq. (A.9), and the odd boundary condition on the TeV brane, Eq. (A.10). The mass
spectrum is obtained by solving (A.17) which now becomes
(r + s/2)J ( mk ) +
(r + s/2)Y ( mk ) +

m m
k J ( k )
m m
k Y ( k )

J ( mk ekR )
,
Y ( mk ekR )

(A.18)

and for the fermion and gravitino the equation simplifies further to Eq. (29).

References
[1] N. Arkani-Hamed, S. Dimopoulos, G. Dvali, The hierarchy problem and new dimensions at a
millimeter, Phys. Lett. B 429 (1998) 263, hep-ph/9803315.
[2] L. Randall, R. Sundrum, A large mass hierarchy from a small extra dimension, Phys. Rev.
Lett. 83 (1999) 3370, hep-ph/9905221.
[3] T. Gherghetta, A. Pomarol, Bulk fields and supersymmetry in a slice of AdS, Nucl. Phys. B 586
(2000) 141, hep-ph/0003129.
[4] R. Altendorfer, J. Bagger, D. Nemeschansky, Supersymmetric RandallSundrum scenario, hepth/0003117.
[5] N. Alonso-Alberca, P. Meessen, T. Ortin, Supersymmetric brane-worlds, Phys. Lett. B 482
(2000) 400, hep-th/0003248;
A. Falkowski, Z. Lalak, S. Pokorski, Supersymmetrizing branes with bulk in five-dimensional
supergravity, Phys. Lett. B 491 (2000) 172, hep-th/0004093;
E. Bergshoeff, R. Kallosh, A. Van Proeyen, Supersymmetry in singular spaces, JHEP 0010
(2000) 033, hep-th/0007044.
[6] See for example, H. Verlinde, Holography and compactification, Nucl. Phys. B 580 (2000) 264,
hep-th/9906182;
M.J. Duff, J.T. Liu, K.S. Stelle, A supersymmetric type IIB RandallSundrum realization, hepth/0007120;

T. Gherghetta, A. Pomarol / Nuclear Physics B 602 (2001) 322

[7]
[8]
[9]
[10]
[11]

[12]
[13]
[14]
[15]

[16]
[17]
[18]

[19]

[20]
[21]
[22]

[23]

[24]
[25]

21

M. Cvetic, M.J. Duff, J.T. Liu, H. Lu, C.N. Pope, K.S. Stelle, RandallSundrum brane tensions,
hep-th/0011167.
A. Pomarol, Grand unified theories without the desert, Phys. Rev. Lett. 85 (2000) 4004, hepph/0005293.
I. Antoniadis, A possible new dimension at a few TeV, Phys. Lett. B 246 (1990) 377.
A. Pomarol, Gauge bosons in a five-dimensional theory with localized gravity, Phys. Lett. B 486
(2000) 153, hep-ph/9911294.
J. Maldacena, The large N limit of superconformal field theories and supergravity, Adv. Theor.
Math. Phys. 2 (1998) 231, hep-th/9711200.
S.S. Gubser, AdS/CFT and gravity, hep-th/9912001;
M.J. Duff, J.T. Liu, On the equivalence of the Maldacena and RandallSundrum pictures, Phys.
Rev. Lett. 85 (2000) 2052, hep-th/0003237.
S.B. Giddings, E. Katz, L. Randall, Linearized gravity in brane backgrounds, JHEP 0003 (2000)
023, hep-th/0002091.
N. Arkani-Hamed, M. Porrati, L. Randall, Holography and phenomenology, hep-th/0012148.
R. Rattazzi, A. Zaffaroni, Comments on the holographic picture of the RandallSundrum model,
hep-th/0012248.
I. Oda, Localization of gravitino on a brane, hep-th/0008134;
A. Falkowski, Z. Lalak, S. Pokorski, Five-dimensional gauged supergravities with universal
hypermultiplet and warped brane worlds, hep-th/0009167.
W.D. Goldberger, M.B. Wise, Bulk fields in the RandallSundrum compactification scenario,
Phys. Rev. D 60 (1999) 107505, hep-ph/9907218.
H. Davoudiasl, J.L. Hewett, T.G. Rizzo, Bulk gauge fields in the RandallSundrum model,
Phys. Lett. B 473 (2000) 43, hep-ph/9911262.
B. Bajc, G. Gabadadze, Localization of matter and cosmological constant on a brane in anti de
Sitter space, Phys. Lett. B 474 (2000) 282, hep-th/9912232;
Y. Grossman, M. Neubert, Neutrino masses and mixings in non-factorizable geometry, Phys.
Lett. B 474 (2000) 361, hep-ph/9912408;
S. Chang, J. Hisano, H. Nakano, N. Okada, M. Yamaguchi, Bulk standard model in the Randall
Sundrum background, Phys. Rev. D 62 (2000) 084025, hep-ph/9912498.
M.A. Luty, R. Sundrum, Hierarchy stabilization in warped supersymmetry, hep-th/0012158;
J. Bagger, D. Nemeschansky, R. Zhang, Supersymmetric radion in the RandallSundrum
scenario, hep-th/0012163.
For a review, see H.P. Nilles, On the low-energy limit of string and M-theory, hep-ph/0004064.
I. Brevik, K.A. Milsimilarton, S. Nojiri, S.D. Odintsov, Quantum (in)stability of a brane-world
AdS(5) universe at nonzero temperature, hep-th/0010205.
J. Scherk, J.H. Schwarz, Spontaneous breaking of supersymmetry through dimensional
reduction, Phys. Lett. B 82 (1979) 60;
J. Scherk, J.H. Schwarz, How to get masses from extra dimensions, Nucl. Phys. B 153 (1979)
61.
I. Antoniadis, C. Munoz, M. Quiros, Dynamical supersymmetry breaking with a large internal
dimension, Nucl. Phys. B 397 (1993) 515, hep-ph/9211309;
I. Antoniadis, K. Benakli, Limits on extra dimensions in orbifold compactifications of superstrings, Phys. Lett. B 326 (1994) 69, hep-th/9310151;
K. Benakli, Perturbative supersymmetry breaking in orbifolds with Wilson line backgrounds,
Phys. Lett. B 386 (1996) 106, hep-th/9509115.
A. Pomarol, M. Quiros, The standard model from extra dimensions, Phys. Lett. B 438 (1998)
255, hep-ph/9806263.
I. Antoniadis, S. Dimopoulos, A. Pomarol, M. Quiros, Soft masses in theories with supersymmetry breaking by TeV-compactification, Nucl. Phys. B 544 (1999) 503, hep-ph/9810410;

22

[26]
[27]

[28]
[29]
[30]

[31]
[32]

[33]
[34]

T. Gherghetta, A. Pomarol / Nuclear Physics B 602 (2001) 322

A. Delgado, A. Pomarol, M. Quiros, Supersymmetry and electroweak breaking from extra


dimensions at the TeV scale, Phys. Rev. D 60 (1999) 095008, hep-ph/9812489.
R. Barbieri, L.J. Hall, Y. Nomura, A constrained standard model from a compact extra
dimension, hep-ph/0011311.
E. Dudas, C. Grojean, Four-dimensional M-theory and supersymmetry breaking, Nucl. Phys.
B 507 (1997) 553, hep-th/9704177;
I. Antoniadis, M. Quiros, Supersymmetry breaking in M-theory and gaugino condensation,
Nucl. Phys. B 505 (1997) 109, hep-th/9705037;
A. Lukas, B.A. Ovrut, D. Waldram, Gaugino condensation in M-theory on S 1/Z(2), Phys.
Rev. D 57 (1998) 7529, hep-th/9711197;
M. Fabinger, P. Horava, Casimir effect between world-branes in heterotic M-theory, Nucl. Phys.
B 580 (2000) 243, hep-th/0002073.
I. Antoniadis, S. Dimopoulos, G. Dvali, Millimeter range forces in superstring theories with
weak-scale compactification, Nucl. Phys. B 516 (1998) 70, hep-ph/9710204.
S. Ferrara, C. Kounnas, F. Zwirner, Mass formulae and natural hierarchy in string effective
supergravities, Nucl. Phys. B 429 (1994) 589, hep-th/9405188.
T. Gherghetta, The role of scalar and pseudoscalar fields in determining nucleosynthesis bounds
on the scale of supersymmetry breaking, Phys. Lett. B 423 (1998) 311, hep-ph/9712343;
A. Brignole, F. Feruglio, F. Zwirner, Signals of a superlight gravitino at e+ e colliders when
the other superparticles are heavy, Nucl. Phys. B 516 (1998) 13, hep-ph/9711516.
N. Kaloper, E. Silverstein, L. Susskind, Gauge symmetry and localized gravity in M theory,
hep-th/0006192.
J. Garriga, O. Pujolas, T. Tanaka, Radion effective potential in the brane-world, hep-th/0004109;
S. Nojiri, S.D. Odintsov, S. Zerbini, Bulk versus boundary (gravitational Casimir) effects in
quantum creation of inflationary brane world universe, Class. Quantum Grav. 17 (2000) 4855,
hep-th/0006115.
J. Garriga, T. Tanaka, Gravity in the brane-world, Phys. Rev. Lett. 84 (2000) 2778, hepth/9911055.
B. Grinstein, D.R. Nolte, W. Skiba, On a covariant determination of mass scales in warped
backgrounds, hep-th/0012074.

Nuclear Physics B 602 (2001) 2338


www.elsevier.nl/locate/npe

Supersymmetric triplet Higgs model of neutrino


masses and leptogenesis
Thomas Hambye a,b , Ernest Ma c , Utpal Sarkar d
a Centre de Physique Thorique, CNRS Luminy, 13288 Marseille, France
b INFN Laboratori Nazionali di Frascati, 00044 Frascati, Italy
c Department of Physics, University of California, Riverside, CA 92521, USA
d Physical Research Laboratory, Ahmedabad 380 009, India

Received 16 November 2000; accepted 8 March 2001

Abstract
We construct a supersymmetric version of the triplet Higgs model for neutrino masses, which can
generate a baryon asymmetry of the Universe through lepton-number violation and is consistent with
the gravitino constraints. 2001 Published by Elsevier Science B.V.
PACS: 12.60.Fr; 12.60.Jv; 14.60.Pq; 98.80.Cq

1. Introduction
The first definite evidence for physics beyond the standard model came from the
recent evidence for the mass of the neutrinos. The atmospheric neutrino anomaly [1], as
observed by the SuperKamiokande experiment, has established that there is a mass-squared
difference between the muon neutrino and the tau neutrino. On the other hand, the solar
neutrino problem [2] implies a mass-squared difference between the electron neutrino and
the other two active neutrinos. Hence it has now been established that at least two neutrinos
are massive.
The mass-squared differences between the different generations of neutrinos have to
be very small, but the mixing angles large, to explain the atmospheric and solar neutrino
anomalies. The required masses for the neutrinos are several orders of magnitude smaller
than those of other fermions, which are all Dirac particles. The smallness of the neutrino
mass is naturally explained if the neutrinos are Majorana particles [3], hence lepton number
is not conserved and that should be due to some physics beyond the standard model.
There are several motivations for lepton-number violation in Nature [4]. In addition, the
E-mail address: hambye@cpt.univ-mrs.fr (T. Hambye).
0550-3213/01/$ see front matter 2001 Published by Elsevier Science B.V.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 0 9 - 2

24

T. Hambye et al. / Nuclear Physics B 602 (2001) 2338

associated lepton-number violation may have the added virtue of accounting for the present
observed baryon asymmetry of the Universe. One model of neutrino mass having this virtue
is the triplet Higgs model [5]. In the nonsupersymmetric case, this model has been studied
in detail and found to share all the interesting features of other models of neutrino mass.
Moreover, in theories with large extra dimensions [6], this mechanism happens to be the
only one which gives Majorana (rather than Dirac) masses to the neutrinos [7]. In this
article we will study the supersymmetric version of this model.
In the supersymmetric version of the triplet Higgs model, there are several new aspects.
Similar to the requirement of two Higgs doublets in the supersymmetric extension of the
standard model, we now have two Higgs triplets. Only one of them couples to the leptons,
but it can acquire a vacuum expectation value (vev) only if the other Higgs triplet is present.
This is related to the fact that a mass term in the superpotential requires two triplet Higgs
superfields, which are of course also necessary for anomaly cancellation. This mass term
connecting the two triplet superfields in the superpotential also allows a trilinear coupling
to exist between two scalar doublets and the scalar triplet which couples to leptons, which
is necessary for neutrino mass as well as leptogenesis [5]. In the present supersymmetric
version of the triplet Higgs model, we must consider the decays of both heavy triplets. Note
that supersymmetry is not yet broken at this energy scale. There are now also several new
diagrams which contribute to the CP violation.
Another important feature of the supersymmetric model comes from the constraints of
nucleosynthesis. In supersymmetric models there is a strong bound on the scale of inflation
from nucleosynthesis due to the gravitino problem [8]. This means that baryogenesis has to
occur at temperatures below about 1011 GeV. On the other hand, in the triplet Higgs model,
the gauge interactions of the triplet Higgs scalars and fermions bring their number densities
to equilibrium at temperatures below 1012 GeV. This naive order-of-magnitude estimate
thus implies that the supersymmetric triplet Higgs model of leptogenesis is probably not
consistent with the gravitino constraints [9]. However, detailed calculations give several
possible ways out of this problem. In the following we will consider the cases where this
potential problem is first ignored and then taken into account. We point out here that the
supersymmetric triplet Higgs model can evade this problem of gravitinos when the masses
of the triplet Higgs superfields are moderately degenerate.
In Section 2 we introduce the model and describe its consequences for neutrino masses.
Then in Section 3 we calculate the amount of CP violation in the decays of the triplet Higgs
scalars and fermions which can generate a lepton asymmetry of the Universe. In Section 4
we solve the Boltzmann equations to calculate the evolution of the lepton asymmetry and
present our results. In Section 5 the gravitino problem is discussed. Finally in Section 6 we
summarize and conclude.

2. The model
The Majorana masses of the neutrinos can be generated by extending the standard model
to include a triplet Higgs scalar [10], which acquires a small vev and couples to two leptons.

T. Hambye et al. / Nuclear Physics B 602 (2001) 2338

25

If lepton number was spontaneously broken by this vev [11], the so-called triplet Majoron
(i.e., the resulting massless Goldstone boson) coupling to the Z boson would be predicted.
This scenario is now ruled out by the known invisible Z width [12]. Moreover, such models
do not explain the present observed baryon asymmetry of the Universe. A new scenario
was then proposed in which lepton number is broken explicitly at a very high energy
scale [5]. The triplet Higgs scalar would then be extremely heavy. However, it acquires a
very tiny vev through its lepton-number violating trilinear coupling to the standard-model
Higgs doublet, which can then give a small Majorana mass to the neutrinos. The decays
of the triplet Higgs scalars also generate a lepton asymmetry of the Universe, which gets
converted to a baryon asymmetry of the Universe before the electroweak phase transition.
To implement the triplet Higgs mechanism in a supersymmetric model, we need to
extend the supersymmetric standard model to include two triplet Higgs superfields. Since
we want these fields to be very heavy, supersymmetry should be unbroken at that stage and
the generation of the lepton asymmetry will not depend on the supersymmetry-breaking
mechanism. We also assume that R-parity is not violated, so that there is no other source of
lepton-number violation except for the Yukawa couplings of the triplet Higgs superfields.
We introduce one triplet 1 ([1++ , 1+ , 10 ] [1, 3, 1] under SU(3)c SU(2)L U (1)Y ) and
another triplet 2 ([20 , 2 , 2 ] [1, 3, 1]) so that a mass term M 1 2 may appear in the
superpotential. However, CP violation is not possible with just these two Higgs triplets. For
that, we need two of each type of the above Higgs triplets. So, if heavy triplet superfields
are used to generate neutrino masses as well as a lepton asymmetry of the Universe,
there should be at least four: 1a ([1a++ , 1a+ , 1a0 ] [1, 3, 1]) and 2a ([2a0 , 2a , 2a ]
[1, 3, 1]), where a = 1, 2 corresponds to the two scalar superfields, whose mixing gives
CP violation for generating the lepton asymmetry of the Universe.
The essential part of the superpotential for the interactions of these scalar superfields

) [1, 2, 1/2] and the standard Higgs doublets


with the lepton superfields L i (Li eLi

0
2 ([ + , 0 ] [1, 2, 1/2]) is given by
1 ([ , ] [1, 2, 1/2]) and H
H
2
1 1
2
1 H
1 a + ha H
  a
 
W = Mab 1a 2b + fija L i L j 1a + ha1 H
1
2 2 H2 2 + H1 H2 + ,

(1)

where i = 1, 2, 3 is the generation index. The first term gives masses to the triplets. The
condition for leptogenesis and neutrino masses would determine this scale M. The next
term gives the Yukawa couplings of the triplet Higgs scalar superfield with the left-handed
lepton chiral superfields of the three generations. When the scalars 1a acquire vacuum
expectation value (vevs), this term gives Majorana masses to the neutrinos. The next two
terms give small vevs to the triplet Higgs scalars.
1 , to two scalars H2 and to
The scalars 1a couple to two leptons, to two higgsinos H
a

a H1 H2 pair. The scalars 2 couple to two higgsinos H2 , to two sleptons, to two scalars
H1 and to a H1 H2 pair. This simultaneous decay of the triplets to products with different
lepton numbers breaks lepton number explicitly. Thus the scale of lepton-number violation
is the same as the mass of the triplet Higgs scalars, which is very heavy, say of the order
O(109 1014) GeV. However, since SU (2)L is unbroken at this scale, these fields do not
acquire any vev. Only after the electroweak symmetry breaking is there an induced tiny
vev for these scalars and the neutrinos would acquire mass.

26

T. Hambye et al. / Nuclear Physics B 602 (2001) 2338

The vevs of the triplet Higgs scalars are obtained from the vanishing of the F -terms,
which corresponds to the minima of the potential. From the conditions F1a = F2a = 0, and
assuming that R-parity is conserved (so that the sneutrinos do not acquire any vev), we get
F1a = Mab 2b + fija L i L j + ha1 H1 H1 = 0
1 a
1 a 2
2b  = ub2 = Mba
h1 H1 2 = Mba
h1 v1 ,

F2a = Mab 1b + ha2 H2 H2 = 0


1 a
1 a 2
1b  = ub1 = Mba
h2 H2 2 = Mba
h2 v2 .

(2)

Since the masses of the triplet scalar fields are several orders of magnitude higher than
the electroweak symmetry breaking scale v, the effective vev of the triplet Higgs fields
are several orders of magnitude smaller than v = 246 GeV. 1 Since these vevs give masses
to the neutrinos, the smallness of the neutrino mass is now directly related to the large
lepton-number violating scale.
The vevs of the triplet scalars will give a mass to the neutrinos given by


1 b 2
2fija ua1 =
2fija Mab
h2 v2 .
(m )ij =
(3)
a

a,b

Since the leptons do not couple with the other triplet scalar 2 , there is no contribution to
the neutrino mass from ua2 . Since the lepton number is now broken at a very large scale
explicitly, there is no Majoron in this scenario. There is one would-be Majoron, which
becomes too heavy to affect any low-energy phenomenology. This makes it consistent with
the measured invisible Z width from LEP (Large Electron Positron Collider) at CERN.
The decay of these scalars to two leptons or two higgsinos can be read off from the
F -terms in the superpotential. The decays of these scalars into two sleptons and the
standard-model Higgs doublets can be read off from the relevant part of the scalar potential,
V = |Mab 2b + fija L i L j + ha1 H1 H1 |2 + |Mab 1b + ha2 H2 H2 |2
+ |2ha1 H1 1a + H2 + |2 + |2ha2 H2 2a + H1 + |2 + .

(4)

The various decay modes of the scalar and fermionic components of the triplet scalar
superfields are listed below and shown in Figs. 1 and 2. The decay modes of the 1a (i.e.,
the scalars 1a++ and the fermions 1a++ ) are
+ +
L L
(L = 2),

i j
a ++
+
+
1
(5)
H2 H2 (L = 0),

+ +

 H
(L = 0)
H
1

and

+ +

Li Lj
++
a
+ H +
1
H

2+ 2+

H H
1

(L = 2),
(L = 0),

(6)

(L = 0)

1 The smallness of these vevs makes this triplet model perfectly consistent with the usual constraints on
additional triplets coming from the measurement of the parameter at LEP.

T. Hambye et al. / Nuclear Physics B 602 (2001) 2338

while the decay modes of 2a are


+ +

(L = 2),

Li Lj
a ++
+ (L = 0),
+ H
2
H

2+ 2+
H1 H1 (L = 0)
and

+ +

Li Lj
++
+ H +
2a
H

2+ 2+

H H
1

27

(7)

(L = 2),
(L = 0),

(8)

(L = 0).

The couplings entering in the various decay modes can be read off from the superpotential.
1 H
2 pair) which are
Note that we do not consider the decays proportional to 2 (i.e., to a H
negligible. If there is CP violation and the decays satisfy the out-of-equilibrium condition,
then these decays can generate a lepton asymmetry of the Universe [13,14]. This lepton
asymmetry can then get converted to a baryon asymmetry of the Universe [15].
When this lepton asymmetry is generated, the B + L violating (but B L conserving)
sphaleron transitions are taking place at a very fast rate [16]. In fact, during the period
1012 GeV > T > 102 GeV
the anomalous B + L violating sphaleron processes remain in equilibrium. During
this period, any lepton asymmetry of the Universe would be equivalent to the B L
asymmetry. The sphaleron interactions would then convert this lepton asymmetry to a
baryon asymmetry of the Universe within this period [17].

3. CP asymmetry in triplet Higgs decay


The various decay modes of 1a and 1a are given in Fig. 1 and the decay modes of
and 2a are given in Fig. 2. The simultaneous decay of the triplet Higgs scalars or the
triplet higgsinos into states with lepton number 0 (two scalar Higgs doublets or higgsinos)
and with lepton number 2 (two leptons or sleptons) implies lepton-number violation. For
CP violation, the tree-level diagrams by themselves are not enough. Even if the couplings
are complex, the probability will be positive definite and hence there will not be any CP
violation. However, if there are one-loop diagrams, which interfere with these tree-level
diagrams, then the interference may be complex, which gives the CP violation.
In the present case there are one-loop diagrams which are given in Fig. 3 (for 1a and
a decays) and in Fig. 4 (for a and a decays). As in the nonsupersymmetric case,
1
2
2
although some of the tree-level diagrams appear similar to the right-handed neutrino decay
diagrams [1820], there are no one-loop diagrams which are similar to the vertex diagrams
of the right-handed neutrino decays. From this point of view, leptogenesis with the triplet
Higgs scalars have this unique feature that CP violation comes only from the self-energy
diagrams, which has the interpretation of oscillations of the scalars before they decay [20].
Moreover, it was pointed out that the CP violation coming from the self-energy diagrams

2a

28

T. Hambye et al. / Nuclear Physics B 602 (2001) 2338

Fig. 1. Tree level diagrams for the decay of 1 .

has an interesting feature of resonant oscillation. Thus the amount of lepton asymmetry can
get highly enhanced when the masses of the triplet Higgs superfields are almost degenerate
[20].
In none of the loop diagrams of Figs. 3, 4 is there any interference between 1a and
a
2 . So, with one each of 1a and 2a , there cannot be any CP violation. In this case, the
relative phases between various couplings can be chosen to be real. Only when there are
at least two 1a or 2a , there can be CP violation. In this case, decays of both 1a and 2a
will contribute to the amount of CP violation. The relative phases between the couplings
of the 1a to the leptons of different generations cannot generate a lepton asymmetry of the
Universe, because they all correspond to final states of the same lepton number. Among the
loop diagrams, figures (c) and (d) are supersymmetric counterparts of figures (a) and (b), so
supersymmetry ensures that the contributions from the first two diagrams are the same as
that of the last two diagrams. In the following we will consider explicitly only the decays

T. Hambye et al. / Nuclear Physics B 602 (2001) 2338

29

Fig. 2. Tree level diagrams for the decay of 2 .

of the scalar triplets keeping in mind that the decays of their fermionic superpartners give
the same lepton asymmetry.
We shall now calculate the amount of CP violation generated from the interference of the
tree-level processes and the one-loop diagrams. In the mass-matrix formalism, it is possible
to give a physical interpretation to this CP violation. A triplet scalar superfield oscillating
into another type before it decays, has a different decay rate compared to its conjugate
states. Although the total decay rates are equal by CPT, the partial decay rates now differ,
which give rise to CP violation. This CP violation will then lead to a lepton asymmetry due
to the fact that (1) the partial decay products do not all have the same lepton number and
(2) the interaction rate is not much faster than the expansion rate of the Universe.
Without loss of generality, we shall assume that the mass matrix for the triplet Higgs
scalars starts out as real and diagonal,
Mab = Ma ab ,

30

T. Hambye et al. / Nuclear Physics B 602 (2001) 2338

Fig. 3. One loop diagrams contributing to CP violation in decays of 1 .

with Ma real. However, in the presence of interactions, they will no longer remain real.
Including the interactions, the mass matrix for the left and right chiral superfields gets
different contributions from the interference of the tree and loop diagrams. The physical
states of the left and right chiral superfields will evolve in a different way and their decays
into leptons and antileptons would generate the lepton asymmetry of the Universe. In
m
m
and 2+
with m = 1, 2 the physical states which are
the following we denote by 1+
m and m the
combinations of the left chiral superfields 1a and 2a , respectively, and by 1
2
physical states which are combinations of the conjugates of these superfields 1a and 2a
respectively (which are the right chiral superfields).
The effective scalar triplet mass matrix we obtain at one loop is given by




1a M21 ab 1b + 2a M22 ab 2b ,
(9)

T. Hambye et al. / Nuclear Physics B 602 (2001) 2338

31

Fig. 4. One loop diagrams contributing to CP violation in decays of 2 .

where, for a given value of the squared momentum p2 of the incoming or outgoing particle:

M2k =

kM
M12 i11
1
k
i21M1

kM
i12
2
kM
2
M2 i22
2

(10)

k
k
Mb = (ba
) Ma and
with ab
1
Mb
ab

and

1
=
8


i,j


fija fijb p2

b 2
+ ha
1 h1 p

+ Ma Mb ha2 hb
2

32

T. Hambye et al. / Nuclear Physics B 602 (2001) 2338

2
ab
Mb




1
a b
a b 2
a b
=
fij fij + h2 h2 p + Ma Mb h1 h1 .
Ma Mb
8
i,j

k
a of the tree scalars in the triplet a
a
The decay widths k
k are given by k = aa
ka . Neglecting terms of order [ij Mj /(M12 M22 )]2 the two mass matrices have the
a = Ma and the eigenvectors are
eigenvalues Mk
1
k+
= k1 i
2
k+
=i

kM
12
2

k M
12
2

M12 M22

1
= k1 i
k
2
k
=i

k2 ,

(11)

k1 + k2 ,

(12)

M12 M22

k M
12
2

k2 ,

(13)

k1 + k2 .

(14)

M12 M22

kM
12
2

M12 M22

Similarly we have
1
k1 = k+
+i

k2 = i

M12 M22

k
M2
12

M12 M22

1
k1 = k
+i

k2 = i

k
M2
12

M12 M22

M12 M22

(15)

1
2
k+
+ k+
,

k
M2
12

k
M2
12

2
k+
,

(16)

2
k
,

(17)

1
2
k
+ k
.

(18)

i
i
Note that, due to CP violation, the k
are not Hermitian conjugates of the k+
but the
j

i
i
|k  = k
|k+  = ij between the in and out states are
orthonormality relations k+
satisfied (as they should be) when diagonalizing a non-Hermitian mass matrix (see, e.g.,
Refs. [21,22]). The resulting lepton asymmetries km induced by the decay of the scalar
a
triplet k
are given by

1a = 2
2a = 2

a ll) ( a l c l c )
(1
1+
a + a
1
1+

a
a
ll) (2
lclc )
(2+
a + a
2+
2

(19)

(20)

Putting Eqs. (15)(18) in Eqs. (1) and (4) we obtain




  2  2 1 1 2 
2 1 1 2
1
i,j Ma Im h1 h1 fij fij + Im M2 M1 h2 h2 fij fij
a

1 
(21)
a 2
a 2
a 2
2(M12 M22 )
i,j |fij | + |h1 | + |h2 |

T. Hambye et al. / Nuclear Physics B 602 (2001) 2338

33

and similarly we have


2a 

M1 M2
1
2
M1 ) Ma2


  2  2 1 1 2 
2 1 1 2
i,j Ma Im h2 h2 fij fij + Im M1 M2 h1 h1 fij fij


.
a 2
a 2
a 2
i,j |fij | + |h2 | + |h1 |

2(M22

(22)

As expected the asymmetries come from the interference of the leptonic sector (through
the fija s) and the non-leptonic sector (through the hak s). Such asymmetries are obtained
from the decay of each one of the tree states in each scalar triplet. Equal asymmetries are
also obtained from the decay of the tree fermionic partners of the scalar triplets. Note that
for M1 close to M2 , 1a 2a .
When the mass difference between the two Higgs scalars is very small and is comparable
to the decay width, there is a resonance in the amount of CP asymmetry, hence in the
amount of lepton asymmetry [20]. Our present method fails in the limit when the decay
width is larger than the mass differences. In this case it is necessary to apply a resummation
of the self-energies as it has been done in Ref. [19]. Note that at the resonance (M1 M2
k
/2) and with maximal CP violating phase the asymmetries of Eqs. (21) and (22) can
12
be as large as of order one. However as we did already in Eqs. (11)(18) we will restrict
ourselves to a region where the mass squared difference can be small but still much larger
than the decay widths, so that the formalism we consider can be used safely.

4. Boltzmann equations
We shall now check if the out-of-equilibrium condition is satisfied in this scenario
and can generate the required amount of baryon asymmetry of the Universe. The naive
consideration for the out-of-equilibrium condition that the decay rates of the triplet Higgs
scalars to be less than the expansion rate of the universe is satisfied for a wide range of
parameters. This out-of-equilibrium condition reads,
Kka =

ka
H (Ma )

<1

where the Hubble constant H (T ) at the temperature T is given by



4 3 g T 2
H (T ) =
,
45 MP

(23)

(24)

with g 100 the number of massless degrees of freedom and MP 1019 GeV is
the Planck scale. Given any particular temperature, the out-of-equilibrium condition
constrains the various coupling constants. If this condition is satisfied and if the various
damping terms due to scatterings are negligible, the total amount of lepton asymmetry per
comoving volume XL nL /s = (nl nl)/s that will be generated through the decays of


the four triplet Higgs superfields will be given by k 6(1k + 2k )n /(2s) = k 6(1k +
2k )45/(2g 4 ) where the entropy s and the photon number density n are given by

34

T. Hambye et al. / Nuclear Physics B 602 (2001) 2338

2 2 3
T ,
(25)
45
2T 3
n = 2 .
(26)

a
For the out-of-equilibrium condition of 1,2
to be satisfied, we get a bound on the
parameters


a 2
a 2
a 2


4 3 g 8
i,j |fij | + |h1 | + |h2 |
<
4 1017 GeV1 .
(27)
Ma
45 MP
s = g

It is interesting to compare this condition with the condition that a neutrino mass of order
103 eV is generated from Eq. (3),

 fija ha2
a

Ma

(m )ij
2v22



1017 GeV1 ,

(28)

where v2 has to be of order v = 246 GeV. A neutrino mass of order 103 eV can
therefore be obtained while the out-of-equilibrium condition is satisfied for any value of
M1 and M2 provided the couplings fija and ha1,2 have the appropriate values. 2 This is in
general achieved if ha2 together with at least one of the fija for a = 1 or 2 are of order
[(1017 GeV1 )M1,2 ]1/2 (with all other couplings taking smaller values). For M1,2
1014 GeV this requires fija ha2 102 101 while for M1,2 109 GeV this requires
fija ha2 103 104 . Assuming a maximal CP violating phase, the lepton asymmetry
obtained from Eqs. (21), (22) is then typically of order XL 105 106 in the former
case and XL 1010 1011 in the latter case. A smaller asymmetry can be generated if
for example this CP violating phase is not maximal or if in general larger values of the f s
and the hs are taken in such a way that Eq. (28) is satisfied but not Eq. (27). In the latter
case the damping term of the inverse decay process will suppress the asymmetry. A larger
asymmetry can be obtained if M1 and M2 are more degenerate. For M1,2 < 109 GeV a
certain degree of degeneracy is needed in order to obtain a baryon asymmetry of the order
of the one required, i.e., XL 1010 .
The above estimate has however not taken into account possible scattering damping
terms. There are for example lepton-number violating scattering processes, which can
deplete the generated lepton asymmetry of the Universe. For example, H1 + H1 L + L
and H1 + H1 2 L + L scattering (which are absent in the nonsupersymmetric case)
come from renormalizable terms and may not be suppressed. However, it can be shown that
these processes are not really relevant because they go out of equilibrium once we require
the decays of the triplets to be slow enough to be away from thermal equilibrium. There
is also one lepton-number conserving process which is more of a problem, i.e., the gauge
interactions of the triplet Higgs superfields. These induce the very fast a + a G1 + G2
scattering process, where a and a are two scalar triplets, and G1 and G2 are two SU(2)L
2 For small values of M
1,2 this would require however very small values of the f s and the hs whose
naturalness could be questioned.

T. Hambye et al. / Nuclear Physics B 602 (2001) 2338

35

or U (1)Y gauge bosons, as obtained from the kinetic term of the scalar triplets. This gives
a suppression in the generation of the lepton asymmetry of the Universe and implies that
the mass of the triplets cannot be too small (except if the two triplets are almost degenerate
as shown below). The presence of this damping term requires the explicit calculation of
the evolution of the asymmetry using the Boltzmann equations.
Defining the variable z M1 /T and the various number densities per comoving volume
Xi ni /s, the Boltzmann equations are:


X2 a
1
K1 (z) Mka 2
eq
k
a
= zKka
1 eq2
Xka X a + z
scatt.
,
k
dz
K2 (z) M1
sH (M1 )
X a

dXka

dXL
=
dz


a,k

K1 (z) Mka
zKka
K2 (z) M1

eq
2 


1 Xka
eq
a
a
XL .
k Xk X a
k
2 X

(29)
(30)

In Eqs. (29)(30) the equilibrium distributions of the number densities are given by the
MaxwellBoltzmann statistics:
nka = gka
where gka

M2a

k
T K2 (Mka /T ),
(31)
2 2
= 1 are the numbers of degrees of freedom of the ka and K1,2 are the usual

modified Bessel functions. The reaction density for the scattering process a + a
G1 + G2 is given by
a
scatt.

T
=
64 4

ds a (s) s K1 ( s/T ),

(32)

4Ma2

where is the reduced cross section which is given by 2(s 4Ma2 )a (s). Note that a
precise result would require an explicit calculation of all scattering processes involving
gauge interactions in all channels. 3 However it can be checked that the dependence of the
generated lepton asymmetry on the magnitude of the scattering is much slower than linear.
Therefore, considering also the fact that the model allows some freedom in the range of
parameters used, this explicit calculation will not add much to our understanding in any
case. We will thus make the following estimate:
a =

1
1
g4 ,

s s 4Ma2

(33)

where g is the SU(2)L coupling (which at tree level is given by the relation m2W = g 2 v 2 /4).
Putting Eq. (33) in Eqs. (32) and (29), it turns out that the scattering term has a small
effect on the evolution of the lepton asymmetry for values of M1,2 above 1011 1012 GeV.
For smaller values of M1,2 the suppression can be very strong due to the fact that the
3 There are more than 20 different physical processes of the type +  G + G . There is also scattering
a
1
2
a
of the type a + a l + l with an intermediate gauge boson which is of the same order.

36

T. Hambye et al. / Nuclear Physics B 602 (2001) 2338

last term of Eq. (29) increases when M1 decreases and T M1 . This will suppress the
asymmetry which at some point becomes much smaller than 1010 except if M1 and
M2 are sufficiently degenerate. Note that the suppression due to these scattering processes
is the most effective when the triplet starts decaying. At lower temperatures, the scattering
effect is suppressed by the Boltzmann factor due to the higher threshold in Eq. (32). Taking
a , f a , f a and f a (for both a = 1 and 2) equal to the same value f with all
for example f22
23 32
33
a
other fij equal to zero, i.e., assuming negligible all the fija with i = 1 and/or j = 1 (which
constitutes one of the possible structures leading to a maximal mixing between the second
and third generation of neutrinos) and taking all hak couplings equal to the same value h,
four typical sets of parameters which give an asymmetry of order 1010 together with a
neutrino mass of order 103 102 eV are shown below:
M1 = 1013 GeV,

M2 = 3.0 1013 GeV,

h = 1 103 ,

f = 3 102 , (34)

M1 = 1012 GeV,

M2 = 3.0 1012 GeV,

h = 1 102 ,

f = 5 104 , (35)

M1 = 1011 GeV,

M2 = 2.0 1011 GeV,

h = 8 104 ,

f = 2 103 , (36)

M1 = 1010 GeV,

M2 = 1.1 1010 GeV,

h = 1 103 ,

f = 5 104 . (37)

A maximal CP-violating phase has been assumed. Note that the degree of degeneracy
which is required for M1,2 1010 GeV is relatively small. Note also that smaller values
of M1,2 are possible if they are even more degenerate. As M1,2 decreases, the degree
of degeneracy required becomes however very high, due to the damping effects of the
scattering processes. 4

5. Gravitino problem
So far we have not taken into account the gravitino problem. The main constraint
comes from the fact that the lepton asymmetry has to be generated after inflation, which
is very important in supersymmetric models [8,23]. The thermal production of massive
gravitinos restricts the beginning of the radiation-dominated era following inflation. The
reheating temperature after inflation is constrained by requiring gravitino production to
be suppressed so that it will not overpopulate the Universe. Since the gravitinos interact
very weakly, they decay very late and modify the abundances of light elements which may
become inconsistent with nucleosynthesis. On the other hand, if they are stable, then they
overclose the universe. The upper bound on the reheating temperature from the gravitino
constraint is [23]


m3/2
1 TeV 2
10
,

TRH  10 GeV
(38)
100 GeV
mg ()
4 Note also that, as the degree of degeneracy increases, contributions from thermal gauge effects to the triplet
scalar mass matrix become important and cannot be neglected anymore. This feature is not relevant for the
leptogenesis mechanism based on heavy right-handed Majorana neutrinos [15], due to the fact that heavy righthanded neutrinos have no SU(2)L or U (1)Y charges.

T. Hambye et al. / Nuclear Physics B 602 (2001) 2338

37

where m3/2 is the gravitino mass and mg is the running mass of the gluino. This gravitino
constraint is satisfied if the lepton asymmetry is generated at temperatures below the
reheating temperature T < TRH . From this result we can assume that the masses of the
triplet Higgs scalars should be around TRH 1010 1011 GeV, so that leptogenesis occurs
at a temperature T < 1010 1011 GeV. As shown above, a lepton asymmetry and neutrino
masses of the size required can be generated with this value of the mass, but it requires
some (moderate) degree of degeneracy 5 between M1 and M2 [see Eq. (37)].

6. Summary and conclusion


We have shown that the supersymmetric triplet Higgs model developed in this article
constitutes an interesting and simple alternative for generating neutrino masses and
baryogenesis. This requires typically triplet superfields with mass of order 109 1014 GeV.
We have shown that this mechanism has the interesting property of possible resonant
behavior. For masses of order 1010 GeV which are consistent with the gravitino problem,
this feature of resonant CP violation is necessary (i.e., masses of the triplets need to be
moderately degenerate).

Acknowledgement
This work was supported in part by the U.S. Department of Energy under Grant
No. DE-FG03-94ER40837 and by the TMR, EC-contract No. ERBFMRX-CT980169
(EuroDane). U.S. and T.H. thank the Physics Department, University of California at
Riverside for hospitality.

References
[1] Y. Fukuda et al., Phys. Lett. B 433 (1998) 9;
Y. Fukuda et al., Phys. Lett. B 436 (1998) 33;
Y. Fukuda et al., Phys. Rev. Lett. 81 (1998) 1562;
Y. Fukuda et al., Phys. Rev. Lett. 82 (1999) 2644.
[2] R. Davis, Prog. Part. Nucl. Phys. 32 (1994) 13;
P. Anselmann et al., Phys. Lett. B 357 (1995) 237;
P. Anselmann et al., Phys. Lett. B 361 (1996) 235;
J.N. Abdurashitov et al., Phys. Lett. B 328 (1994) 234;
J.N. Abdurashitov et al., Phys. Rev. Lett. 83 (1999) 4686;
Y. Fukuda et al., Phys. Rev. Lett. 77 (1996) 1683;
Y. Fukuda et al., Phys. Rev. Lett. 81 (1998) 1158;
Y. Fukuda et al., Phys. Rev. Lett. 82 (1999) 1810;
Y. Fukuda et al., Phys. Rev. Lett. 82 (1999) 2430.
5 Note that the fact that the resonance behaviour of the self-energy contribution can be used in order to avoid
the gravitino constraint was also noticed in Ref. [24].

38

T. Hambye et al. / Nuclear Physics B 602 (2001) 2338

[3] S. Weinberg, Phys. Rev. Lett. 43 (1979) 1566;


E. Ma, Phys. Rev. Lett. 81 (1998) 1171.
[4] E. Witten, hep-ph/0006332.
[5] E. Ma, U. Sarkar, Phys. Rev. Lett. 80 (1998) 5716.
[6] N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 429 (1998) 263;
I. Antoniadis, N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 436 (1998) 257;
N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Rev. D 59 (1999) 086004.
[7] E. Ma, M. Raidal, U. Sarkar, Phys. Rev. Lett. 85 (2000) 3769;
E. Ma, G. Rajasekaran, U. Sarkar, Phys. Lett. B 495 (2000) 363.
[8] M.Yu. Khlopov, A.D. Linde, Phys. Lett. B 138 (1984) 265;
J. Ellis, J. Kim, D.V. Nanopoulos, Phys. Lett. B 145 (1984) 181;
J. Ellis, G.B. Gelmini, J.L. Lopez, D.V. Nanopoulos, S. Sarkar, Nucl. Phys. B 373 (1992) 399;
M. Kawasaki, T. Moroi, Progr. Theor. Phys. 93 (1995) 879;
T. Moroi, Ph.D. Thesis, hep-ph/9503210.
[9] D. Delepine, U. Sarkar, Phys. Rev. D 60 (1999) 055005.
[10] J. Schechter, J.W.F. Valle, Phys. Rev. D 22 (1980) 2227.
[11] G.B. Gelmini, M. Roncadelli, Phys. Lett. B 99 (1981) 411;
C. Wetterich, Nucl. Phys. B 187 (1981) 343;
G. Lazarides, Q. Shafi, C. Wetterich, Nucl. Phys. B 181 (1981) 287;
R.N. Mohapatra, G. Senjanovic, Phys. Rev. D 23 (1981) 165;
R. Holman, G. Lazarides, Q. Shafi, Phys. Rev. D 27 (1983) 995;
G. Lazarides, Q. Shafi, Phys. Rev. D 58 (1998) 071702.
[12] See, for example, P.B. Renton, Int. J. Mod. Phys. A 12 (1997) 4109.
[13] A.D. Sakharov, Pisma Zh. Eksp. Teor. Fiz. 5 (1967) 32.
[14] E.W. Kolb, M.S. Turner, The Early Universe, AddisonWesley, Reading, MA, 1990.
[15] M. Fukugita, T. Yanagida, Phys. Lett. B 174 (1986) 45.
[16] V.A. Kuzmin, V.A. Rubakov, M.E. Shaposhnikov, Phys. Lett. B 155 (1985) 36.
[17] S.Yu. Khlebnikov, M.E. Shaposhnikov, Nucl. Phys. B 308 (1988) 885;
J.A. Harvey, M.S. Turner, Phys. Rev. D 42 (1990) 3344.
[18] P. Langacker, R. Peccei, T. Yanagida, Mod. Phys. Lett. A 1 (1986) 541;
M. Luty, Phys. Rev. D 45 (1992) 445;
A. Acker, H. Kikuchi, E. Ma, U. Sarkar, Phys. Rev. D 48 (1993) 5006;
L. Covi, E. Roulet, F. Vissani, Phys. Lett. B 384 (1996) 169;
W. Buchmuller, M. Plmacher, Phys. Lett. B 389 (1996) 73;
W. Buchmuller, M. Plmacher, Phys. Lett. B 431 (1998) 354.
[19] A. Pilaftsis, Phys. Rev. D 56 (1997) 5431;
A. Pilaftsis, Nucl. Phys. B 504 (1997) 61.
[20] M. Flanz, E.A. Paschos, U. Sarkar, Phys. Lett. B 345 (1995) 248;
M. Flanz, E.A. Paschos, U. Sarkar, J. Weiss, Phys. Lett. B 389 (1996) 693.
[21] R.G. Sachs, Ann. Phys. 22 (1963) 239.
[22] M. Beuthe, R. Gonzalez-Felipe, G. Lopez-Castro, J. Pestieau, Nucl. Phys. B 498 (1997) 55.
[23] M. Bolz, W. Buchmller, M. Plumacher, Phys. Lett. B 443 (1998) 209.
[24] A. Pilaftsis, Int. J. Mod. Phys. A 14 (1999) 1811.

Nuclear Physics B 602 (2001) 3960


www.elsevier.nl/locate/npe

The geometry of fractional branes


M. Frau a , A. Liccardo a , R. Musto b,1
a Dipartimento di Fisica Teorica, Universit di Torino, and INFN, Sezione di Torino, Via P. Giuria 1,

I-10125 Torino, Italy


b Dipartimento di Scienze Fisiche, Universit di Napoli, and INFN, Sezione di Napoli, Monte S. Angelo,

I-80126 Napoli, Italy


Received 27 December 2000; accepted 8 March 2001

Abstract
By looking at fractional Dp-branes of type IIA on T4 /Z2 as wrapped branes and by using
boundary state techniques we construct the effective low-energy action for the fields generated by
fractional branes, build their worldvolume action and find the corresponding classical geometry. The
explicit form of the classical background is consistent only outside an enhanon sphere of radius re ,
which encloses a naked singularity of repulson-type. The perturbative running of the gauge coupling
constant, dictated by the NSNS twisted field that keeps its one-loop expression at any distance, also
fails at re . 2001 Elsevier Science B.V. All rights reserved.
PACS: 11.25.Hf

1. Introduction
The duality relation between string and gauge theories [1] has been studied in detail in
the AdS/CFT case, but recently more effort has been devoted to study the gauge/gravity
correspondence in the case of nonconformal gauge theories and YangMills theories with
reduced supersymmetry [216]. Fractional branes [17], that arise for example in orbifold
models, provide a natural set-up to probe the Maldacena conjecture when supersymmetry
is partially broken with respect to the N = 4 case and conformal invariance is lost. The
full knowledge of spacetime geometry of fractional branes is then not only an interesting
problem relative to singular geometries and their resolution but also a physical issue
providing a new insight for understanding the stringy aspects of gauge theory in more
realistic models.

Work partially supported by the European Commission RTN programme HPRN-CT-2000-00131 in which
R.M. is associated with Frascati-LNF, and by MURST.
E-mail address: frau@to.infn.it (M. Frau).
1 R.M. dedicates this work to the memory of Lochlain ORaifeartaigh.

0550-3213/01/$ see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 0 7 - 9

40

M. Frau et al. / Nuclear Physics B 602 (2001) 3960

In this paper we will discuss the case of fractional Dp-branes of type IIA string theory
compactified on T4 /Z2 , where Z2 is generated by the parity operator on the four compact
spatial coordinates. While the techniques we use to reconstruct the spacetime geometry
associated to fractional branes are rather general, for the sake of simplicity we will mainly
refer to the case of a fractional D0-brane of type IIA theory on T4 /Z2 . As it is well known,
this is an orbifold limit of the type IIA string compactified on the smooth hyperKhler
manifold K3. Within this frame a fractional Dp-brane can be interpreted as a D(p + 2)brane wrapped on one of the supersymmetric two-cycles of K3 that vanish in the orbifold
limit. This point of view provides a natural ansatz to rewrite the relevant massless fields of
the ten-dimensional IIA orbifold theory in terms of fields coupled to the brane. Then one
obtains the truncation of the low energy action of IIA on T4 /Z2 that provides the dynamics
of fields coupled to fractional Dp-branes. In order to reconstruct the full geometry of
a fractional brane one also needs the expression of relevant source terms, namely of
a boundary action. This information can be obtained by analyzing the structure of the
boundary states [18]. The use of the boundary state technique [19] is rather natural in
this context as it translates open string boundary conditions in terms of closed string,
introducing automatically into the game gravity together with all other fields interacting
with the fractional brane. The analysis of the couplings between the massless fields and
the boundary state allows us to infer the structure of the complete worldvolume action of
fractional branes on the compact orbifold, which turns out to be also an essential tool for
analyzing the motion of a probe brane in the classical background.
One expects that the spacetime geometry of fractional branes shares a common feature
with other classical backgrounds which are dual to non conformal gauge theories, namely
the presence of naked singularities of repulson type. This feature has been found in the
study of fractional branes on singular spaces [5,6,8,12,20], of stable non-BPS branes [10]
and of type IIB fractional branes on R1,5 R4 /Z2 [13,15]. Our detailed investigation on
the D0-brane solution and its generalization to the D2-brane, shows that this is the case also
for fractional branes of type IIA on T4 /Z2 . However, also in our case, like in [2022] this
singularity is actually unphysical as it is outside the region where the simple supergravity
approximation is reliable. In fact, at a distance re greater than the one where the singularity
is located, a probe fractional brane becomes tensionless, and at r = re there is a geometric
locus, called enhanon, where extra massless degrees of freedom become relevant (for a
review on the probe technique see, e.g., Ref. [23]). In this case, as discussed in Ref. [20],
the fractional branes building up the classical background are forced to cover uniformly
the hypersphere at r = re rather than pile-up at r = 0. Then at a distance shorter than the
enhanon radius the supergravity description is not valid any more and one has to modify
the effective theory.
In this paper we do not study in detail the connection between the fractional brane
classical solution and the worldvolume gauge theory and we leave it to a future work.
We just mention that in general the running of the coupling constant of the dual gauge
theory turns out to be dictated by the behavior of the twisted fields and in particular of the
NSNS twisted field. A remarkable property of our solution, is that twisted fields keep their
harmonic asymptotic form at any distance. This appears as a general feature for fractional

M. Frau et al. / Nuclear Physics B 602 (2001) 3960

41

branes, apparently translating in geometrical classical terms the N = 2 SUSY gauge theory
property of allowing only one-loop perturbative corrections. This is in fact the case for type
IIB fractional D3-brane [13,15], where the appropriate perturbative logarithmic behavior
of YangMills coupling constant in terms of a stringy description has been obtained.
The paper is organized as follows: Section 2 is devoted to the reduction of the low
energy effective action for type IIA theory in ten dimension to the six-dimensional action
describing the dynamics of the fields coupled to a fractional Dp-brane on T4 /Z2 . The
boundary state technique, which is used to obtain a boundary action and the asymptotic
behavior of relevant fields, is described in some detail in Section 3. In Section 4 we give
the solution of the equations of motion and discuss their physical implications. The three
appendices are devoted, respectively, to an alternative derivation of the low energy action
based on the S-duality between IIA on T4 /Z2 and heterotic on T4 (Appendix A), to the
analysis of the massless spectrum of the Z2 orbifold theory (Appendix B) and to the explicit
derivation of the equations of motion and their solution (Appendix C).

2. The low energy effective action


One of the ingredients necessary to determine the geometry of a fractional brane is the
low energy action expressing the dynamics of the relevant fields in the bulk. In this section
we start with type IIA supergravity in ten dimensions with a Z2 orbifold projection acting
as a reflection on four coordinates (which we choose to be x6 , x7 , x8 , x9 ). Introducing a
specific ansatz for the fields, we obtain the peculiar truncation of the low energy action
which is appropriate to describe the fractional brane in R 1,5 R 4 /Z2 . Then, by performing
a KaluzaKlein reduction to six dimensions, we get the low energy action for fractional
branes in the orbifold R 1,5 T4 /Z2 . In Appendix A we obtain the same action exploiting
the S-duality relation between our theory and heterotic theory compactified on T4 .
The ten dimensional type IIA supergravity action on the orbifold R 1,5 R 4 /Z2 in the
string frame can be written as





1
10
2
R(G) + e2 4 d d 12 H3 H3
d x G e
S= 2
2


1



(2.1)
+ F2 F2 + F4 F4 B2 F4 F4 ,
2
where
H3 = dB2 ,

1 ,
F2 = d C

F4 = dC3

(2.2)

are, respectively, the field strengths of the KalbRamond field, of the RR vector field and
of the RR 3-form potential and
1 H3 .
4 = F4 C
F

(2.3)

Moreover is the 10-dimensional dilaton fluctuation and = (2)7/2 2 gS is the


gravitational coupling constant appropriate to our orbifold background.

42

M. Frau et al. / Nuclear Physics B 602 (2001) 3960

A fractional Dp-brane can be interpreted as a D(p + 2)-brane wrapped on the vanishing


2-cycle of the orbifold [24]. According to this interpretation we make the following ansatz
on the KalbRamond and RR 3-form potential
B = b2 ,

C3 = A1 2 ,

(2.4)

where 2 is the closed 2-form dual to the vanishing 2-cycle which we normalize in such a
way that

2 2 = 1,
(2.5)
and the scalar field b is
1  2
b = 2  + D.
(2.6)
2
Using the previous equations in Eq. (2.1) we obtain the following expression for the
action



 2

1
10
2
d x G e
R(G) +
4e
d d + 12 F2 F2
S= 2
2



1  2



e
dD dD (dA1 C1 dD) (dA1 C1 dD) , (2.7)

2
6

where the index 6 means that the fields are integrated over the 6-dimensional spacetime
which is unaffected by the orbifold projection. Notice that under the ansatz (2.4) the
ChernSimon term appearing in Eq. (2.1) vanishes, showing that it does not contribute
to the dynamics of fractional branes. The action appropriate to the case of the compact
orbifold T4 /Z2 can now be obtained from (2.7), by making a simple KaluzaKlein
reduction of the ten-dimensional part




V
6
2
S= 2
R(g) + e2 4 d d d a d a
d x g e
2
6

  2

1
1
e
+
e a 2
e a dD dD
F2 F2
2
2
2
a
a
6


1
1 dD) + C
1 dD (C
1 dD) ,
F2A F2A 2F2A (C
(2.8)
2
where g is the six-dimensional metric, while the 6-dimensional dilaton field and the
scalar fields a are defined as follows

1
Gaa , Gaa = e2 a
= ln
(2.9)
4
a
with a = 6, 7, 8, 9. Moreover we have introduced F2A = dA1 and denoted the volume of T4
by

V = (2Ra ).
(2.10)
a

M. Frau et al. / Nuclear Physics B 602 (2001) 3960

43

Defining orb = /V 1/2 and going to the Einstein frame, we get



 
1
6
S= 2
d x g R(g) +
d d d a d a
2orb
6

1
a
 d D

+ e
e F2 F2
e a d D
2
a
a


A

e  A A







F F2 2 2 F2 (C1 d D) + 2C1 d D (C1 d D)
+
,
2 2
(2.11)
where we have made the following rescaling
A1
1 =
,
A
V

 = D .
D
2V

(2.12)

The action (2.11) describes the fields that couple to an electric fractional D0-brane, or
equivalently to a magnetic fractional D2-brane. In next section, using the boundary state
formalism, we will instead obtain the asymptotic behavior of the classical fields generated
by a generic electric fractional Dp-brane. In solving the classical equations of motion
derived from (2.11) with the boundary conditions dictated by the boundary state, it is then
more natural to address the case of the D0 fractional brane. This is what we will actually
do in detail in the following. The explicit solution found in that case can be however easily
generalized also to case of D2 fractional brane, as we will see in Section 4.

3. The boundary state description


The boundary state formalism allows one to connect microscopic and supergravity
description of D-branes by using the language of closed string state to evaluate both the
couplings of supergravity massless fields to the brane and their asymptotic behavior. The
boundary state description of fractional Dp-branes in type II theories on orbifold has been
already discussed in Refs. [2527] in the case of ZN projection, and for general discrete
groups in Ref. [28]. Here we briefly review the structure of the boundary state in the simple
case of Z2 , in order to derive the relevant physical information about fractional branes.
The boundary state is a closed string state which embodies the presence of a Dp-brane in
spacetime and is defined by the overlap equations, which, in the notation and conventions
of Ref. [19], read as

X  =0 |B = 0, = 0, . . . , p,
(3.1)

i
i
X
(3.2)
|B = y |B , i = p + 1, . . . , d 1.
=0

and


(, 0) |B , = 0,
(, 0) iS +

where S = ( , ij ), and = 1.

(3.3)

44

M. Frau et al. / Nuclear Physics B 602 (2001) 3960

As it is well known, the spectrum of type II strings on orbifolds consists not only of states
of the original theories which are invariant under the orbifold projection, the untwisted
sector, but also of states which belong to the so-called twisted sectors living at the orbifold
fixed hyperplanes. Therefore in an orbifold theory the complete boundary state is a linear
combination of all the Ishibashi states coming from the resolution of Eqs. (3.1)(3.3) in
each sector.
In the specific case of the orbifold T4 /Z2 , there are 16 distinct twisted sectors, one
for each fixed plane. However, when the fractional brane does not wrap in the compact
directions, there are only two different coherent states which solve the overlap equations,
one for the untwisted and the other for the twisted sector associated to the particular fixed
plane on which the brane lives.
In this particular case we can therefore write


 T

U
T
|B NS + .2 |B TR ,
|B = N U |B U
(3.4)
NS + .1 |B R + N
where N U and N T are two normalization constant to be fixed, U and T stand for untwisted
and twisted, respectively, and .1 ,.2 are the two RR charges in the untwisted and twisted
sectors. The states |B U are the usual boundary states of a bulk-brane on a compact space
with the standard GSO-projection [29], while the states |B T will be constructed in the
following subsection.
Notice that the boundary state in (3.4) is the one associated to the trivial representation
of the Z2 group on ChanPaton factors of open strings attached to fractional branes [28].
3.1. Construction of the boundary state
The Ishibashi states of the twisted sector, satisfying the overlap conditions (3.1)(3.3),
can be written as
|B, TNS = |BX T |B , TNS

(3.5)

in the NSNS twisted sector and similarly


|B, TR = |BX T |B , TR

(3.6)

in the RR twisted sector, 2 where




1 S
1 a a
n
|BX T = 5p q i y i
e n n
e r r r |0 |0 |p = 0 ,

|B , TNS =

n=1

eit S t

t =1/2

|B , TR =

t =1

eit S t

(3.7)

r=1/2
a

eit t |B , NS ,
a

(0)T

(3.8)

t =1
a

eit t |B , (0)T
R ,
a

(3.9)

t =1/2

2 In (3.5) and (3.6) we omit the ghost and superghost contribution which is not affected by the orbifold
projection.

M. Frau et al. / Nuclear Physics B 602 (2001) 3960

45

and , {0, . . . , 5} and a {6, . . . , 9}. The zero modes part of the boundary state has a
non trivial structure in both sectors; in the NSNS case it is given by


1 + i5
(0)T

|B , NS = C
(3.10)
|l |
m ,
1 + i lm
 the charge conjugation matrix of SO(4), 5 =
where i are the gamma matrices and C
m are spinors of SO(4), while in the RR case we have
6 9 and finally |l , |



 0 p 1 + i7
=
|a |b ,
|B , (0)T
(3.11)
C
R
1 + i ab
 the charge conjugation matrix of
where analogously i are the gamma matrices and C

SO(1, 5), 7 = 0 5 and finally |a , |b are spinors of SO(1, 5).


The GSO-projection selects both in the NSNS and RR twisted sectors the following
combination [26]
|B T =


1
|B, + + |B, .
2

(3.12)

The normalization constants N U and N T appearing in (3.4) can be fixed by evaluating


the interaction amplitude B|D|B between two fractional Dp-branes due to the tree level
exchange of closed strings and comparing the result with the corresponding open string
channel calculation by means of the worldsheet duality transformation. The closed string
amplitude can be written as the sum of two contributions, one for each sector
B|D|B =

B|D|B U + T B|D|B T .

(3.13)

An explicit computation easily shows that the untwisted sector contribution vanishes due
to the abstruse identity, as in the case of usual bulk brane, while in the twisted sector the
NSNS and RR contributions cancel each other. Therefore the amplitude Eq. (3.13) is
vanishing and this is consistent with the fact that fractional branes, being BPS objects,
satisfy a no-force condition.
The 1-loop vacuum amplitude which represents the interaction between two fractional
Dp-brane in the open string channel is also expressed as the sum of two terms, one for
each element of the orbifold group Z2 = {1, g}. Under worldsheet duality they transform,
respectively, in the two terms of Eq. (3.13) and by comparison one can fix the normalization
constants as follows
Tp
Tp
NT =
.
NU = ,
(3.14)
2 2
2 2 2 
3.2. Brane couplings to bulk fields and their asymptotic behavior
In order to get information about the geometry of a fractional brane we need to determine
its couplings with all the massless fields of the theory. This is achieved by projecting the
boundary state onto the massless string states corresponding to the fields of the theory [18].
We need therefore a table of correspondence between the classical fields and the string
states belonging to the massless spectrum.

46

M. Frau et al. / Nuclear Physics B 602 (2001) 3960

As already remarked, the action (2.11) is a consistent truncation of the full action of
type IIA supergravity in six dimensions, and describes the dynamics of the graviton h ,
four KaluzaKlein scalars a originating from the ten dimensional metric, a 1-form gauge
field C1 , plus a 1-form gauge field A1 and a scalar D, which originates, respectively,
from a three form gauge field and the KalbRamond field with two components along
the supersymmetric vanishing cycle of the orbifold. From the string analysis carried out in
Appendix B, it is easy to realize that the graviton, the KaluzaKlein scalars and the 1-form
C1 are represented by the usual massless states of the untwisted sectors, while the 1-form
A1 and the scalar D are described by massless states belonging, respectively, to the twisted
RR and NSNS sector. In the case of the fractional D2-branes the twisted and untwisted
1-forms are obviously replaced by two 3-forms (C3 and A3 ).
By projecting the boundary state on the massless states corresponding to these fields
(whose explicit expressions are given in Appendix B, see also Ref. [30]) we find the
following couplings
p

Tp
h
Jh = Vp+1
2V
=0

(3.15)

for the graviton,


Tp
J = Vp+1 (1 p),
2 2V

Tp
Ja = Vp+1 a
2 2V

(3.16)

for the dilaton and the KaluzaKlein scalars,


Tp
JC = Vp+1 C0p U Vp+1 C0p
V

(3.17)

for the RR untwisted field, and finally


JD =

Tp
Vp+1 D,
2 2 

JAp+1 =

Tp
Vp+1 A0p T Vp+1 A0p
2 2 

(3.18)

for the NSNS twisted scalar and the RR twisted field. 3


The previous couplings allow us to infer the structure of the BornInfeld action for a
fractional brane which turns out to be





a
Tp
p+1
orb (1p)

2
Swv =
|g | e
e orb 2 2 orb Cp+1
d
2V orb
a



a
Tp
p+1
orb (1p)
2

|g
|
e
eorb 2 D
d

2

2
a






Ap+1 + 2 orb D Cp+1 .


(3.19)

3 Eqs. (3.15)(3.18) can be extended to the case of the non-compact orbifold C / by dropping the factor V
2
in all equations.

M. Frau et al. / Nuclear Physics B 602 (2001) 3960

47

The presence of the last term is due to the requirement of gauge invariance of the previous
action under the gauge transformation
Ap+1 = p dD,

Cp+1 =

(3.20)
dp
2 orb
which for p = 0 is the gauge transformation that leaves invariant the bulk action (2.11).
The structure of the action (3.19) is confirmed also by explicit calculations of closed string
scattering amplitudes on a disk with appropriate boundary conditions [31].
From the couplings in Eqs. (3.15)(3.18) one can also determine the contributions to the
interaction between two fractional branes due to the exchange of each massless state. In
particular, looking first at the untwisted states one finds the contribution of graviton
 2
Tp Vp+1 (p + 1)(3 p)
Uh =
(3.21)
,
2
2
V
2k
of the dilaton and KaluzaKlein scalars
 2
 2
Tp Vp+1 (1 p)2
Tp Vp+1 1
U =
,
U
=
,
a
2
2
2
V
2
V 2k
2k
and of the RR untwisted field
Vp+1 1
UC = Tp 2
.
2
V k

(3.22)

(3.23)

In the twisted sector instead, we find that the NSNS scalar and the RR gauge field
contribute, respectively, as
2
2


Tp
Tp
1
1
UD =
(3.24)
Vp+1 2 ,
UA(p+1) =
Vp+1 2 .
2

2

2
2
k
k
The potential energies due to the exchange of the various fields satisfy the conditions

Uh + U +
(3.25)
Ua + UC = 0,
UA(p+1) + UD = 0,
a

which are consistent with the fact that the no-force condition is due to the separate
cancellation of the untwisted and twisted contributions to the interaction energy.
The knowledge of the contribution to the interaction of each massless state is also useful
to determine the mass of the fractional brane. Indeed in order to evaluate the mass we have
to sum up all attractive contributions to the potential energy and compare it with Newtons
law in six dimensions. By making a Fourier transformation to configuration space, we
obtain that the attractive energy is


Tp2
1
1
+
Uattr. =
(3.26)
,
(3 p)r 3p 4p V (2 2  )2
where q = 2 (q+1)/2/ ((q + 1)/2) is the area of a unit q-dimensional sphere.
Comparing Eq. (3.26) with Newtons law
UNewt =

2 M2
2orb
p

(3 p)r 3p 4p

(3.27)

48

M. Frau et al. / Nuclear Physics B 602 (2001) 3960

we get

2
M2p = Tp 2
2orb


1
1
+
= (U )2 + (T )2 ,
V (2 2  )2

(3.28)

which is the usual relation between mass and charges of a BPS object charged with respect
to two different gauge fields.
Another important information provided by the boundary state analysis is the behavior
of all classical fields generated by a fractional brane at large distance. This can be obtained
by saturating the boundary state with the various states of the theory after inserting a closed
string propagator. The asymptotic behavior for the various fields in our case is


Tp
1
(p 3), ij (p + 1)
h =
3p
4p
4 2V (3 p)r

(3.29)

for the graviton,


Tp
1p
,
=
2 2V (3 p)r 3p 4p

Tp
1
a =
2 2V (3 p)r 3p 4p

(3.30)

for the dilaton and the scalars,


Tp
1

C0p
=
V (3 p)r 3p 4p

(3.31)

for the untwisted RR field, and finally


D =

Tp
1
,
2

2 (3 p)r 3p 4p

A
0p =

Tp
1
2

2 (3 p)r 3p 4p

(3.32)
for the twisted NSNS scalar and the twisted RR form. We remind that these asymptotic
fields obtained with boundary state techniques have canonical normalization; therefore
they do not coincide with the corresponding fields appearing in the bulk action given in
Eq. (2.11). The relations between the two set of fields are

,
orb

D
D=
,
orb

a =

a
,
orb

(p+1)
C
C(p+1) =
,
2 orb

(p+1)
A
A(p+1) =
.
2 orb

Using these new fields, the worldvolume action of Eq. (3.19) becomes



a 

Tp
p+1
(1p)
2

2
d
|g | e
e
Cp+1
Swv =
2V orb
a


(1p)

a
Tp

d p+1 2 |g | e 2
e 2 D

2

2 2 orb
a





C
p+1 .
p+1 + 2 D

(3.33)

(3.34)

This action will generate the source terms in the equations of motions for the bulk fields.

M. Frau et al. / Nuclear Physics B 602 (2001) 3960

49

4. Classical solution of fractional D-branes


In order to determine the classical solution of a fractional Dp-brane we can either solve
the inhomogeneous field equations obtained by varying the total action, i.e., the sum of
Eqs. (2.11) and (3.34), or solve the homogeneous field equations supplemented by the
asymptotic behavior of fields, determined by the boundary state analysis, Eqs. (3.29)
(3.32). In Appendix C we find the classical solution associated to the fractional D0brane following the latter procedure: we solve the homogeneous field equations under
the assumption that all the fields are functions only of the radial transverse coordinate
of the six-dimensional space and that the metric obeys the standard extremal black 0-brane
ansatz.
The first corrections to the asymptotic behavior of the fields can be easily obtained
by solving iteratively the equations of motion. From this analysis one can learn that, up
to second order, the twisted fields do not get corrections with respect to their harmonic
asymptotic behavior. This appears to be an important physical feature of the twisted
fields. Therefore we take it as an ansatz for the full solution. Once made this assumption
the equations of motion are easily solved in terms of a single function H , which is a
very simple generalization of the harmonic function appearing in the classical solution of
ordinary bulk branes
H =1+

Q20
V
1 Q0 1

,
2 r3
2 (2)4 2 r 6

(4.1)

where

2 2 T0 orb
Q0 =
.
34V 1/2

(4.2)

More explicitly we have



3
1
ds 2 = H 4 dt 2 + H 4 dr 2 + r 2 d 2

(4.3)

for the metric,


1

e = H 4 ,

e a = H 4

(4.4)

for the dilaton and for the scalar fields,




C0 = H 1 1

(4.5)

for the untwisted RR vector, and finally


1 V 1/2 Q0
D =
,
2 4 2  r 3

A0 =

V 1/2 Q0
4 2  r 3

(4.6)

for the twisted NSNS scalar and the twisted RR vector, respectively.
The structure of the D0-brane solution is very simple: for the twisted fields the first
order correction to the background value is exact, while the untwisted fields have the same
expression of the fields associated to a bulk brane in terms of the function H . This fact

50

M. Frau et al. / Nuclear Physics B 602 (2001) 3960

suggests a natural generalization of our solution to the case of the fractional D2-brane,
namely


1
3
ds 2 = H 4 dt 2 + dx12 + dx22 + H 4 dr 2 + r 2 d 2
(4.7)
for the metric,
e = H 4 ,
1

e a = H 4

(4.8)

for the dilaton and for the scalar fields,




C012 = H 1 1

(4.9)

for the untwisted RR vector and


1 V 1/2 Q2
,
D =
2 4 2  r

A012 =

V 1/2 Q2
4 2  r

(4.10)

for the twisted NSNS scalar and for the twisted RR vector, respectively, where
H =1+
and

Q22
1 Q2 1
V

4
2
2 r
2 (2) r 2

2 2 T2 orb
Q2 =
.
2 V 1/2

(4.11)

(4.12)

Let us make some comments about the solution both for p = 0 and p = 2. First of all
one can notice that this solution is consistent with the BPS nature of fractional Dp-branes,
as seen from the mass-charge relation (3.28) and from the one loop no-force condition,
verified by the boundary state technique. In fact, by computing the worldvolume action
(3.34) for a probe fractional brane in the background of Eqs. (4.3)(4.6) or Eqs. (4.7)
(4.10), one can check that the distance dependent part identically vanishes, and therefore
there is no static force acting on the probe.
Another important observation is that the untwisted fields, and the metric in particular,
have a naked singularity at r = r+ where



Qp
V
3p
1 + 1 +
(r+ )
(4.13)
, p = 0, 2.
=
4
2 4 2
However the metric we have found is strictly analogous to the one studied in Ref. [20],
so we expect that this singularity is of repulson type and one may cure it by means of an
enhanon mechanism first proposed therein. To see this, one notices that at a distance re ,
where
(re )3p = 2Qp

V
> (r+ )3p
(2)4 2

(4.14)

the derivative of H vanishes and both gravitational and gauge forces change sign.
Therefore, even if a fractional Dp-brane probe feels no net force at any distance, it becomes
tensionless at r = re and acquires a negative tension at shorter distances. Indeed expanding

M. Frau et al. / Nuclear Physics B 602 (2001) 3960

51

the DBI part of the worldvolume action in the velocities and keeping only the lowest order
terms one gets




Tp
x i x j
2V
p+1

ij
D
d
1+

2 2 
2Vorb


3p 
Tp
re
x i x j
p+1
=
(4.15)
1 3p .
ij
d

r
2V orb
We see then that if D has the form of Eq. (4.6) or (4.10) the Dp-brane becomes tensionless
exactly at r = re . It is then clear that the classical solution cannot be trusted at r < re .
This may correspond to the fact that, as shown in Ref. [20], the Dp-branes building up
the classical background, rather than piling up at r = 0, are forced to cover uniformly
the hypersphere at r = re . If one identifies the (properly rescaled) transverse coordinates
with the Higgs fields, one may use the procedure just outlined to derive their kinetic term.
Then, for the (2 + 1) gauge theory dual to the D2 fractional brane, the factor ij (1 re /r)
can be interpreted as the metric in moduli space, which turns out to be rather independent
from the detailed geometry of D2-brane. Actually from this point of view the fact that
NSNS twisted field keeps its harmonic asymptotic form may be seen as expressing
in geometrical classical terms the quantum property of N = 2 SUSY theory allowing
only one-loop perturbative corrections. It is then worth investigating more closely the
correspondence between the detailed structure of fractional Dp-branes geometry and the
associated worldvolume gauge theory.

Acknowledgements
We would like to thank M. Bill, M. Bertolini, P. Di Vecchia, L. Gallot, A. Lerda,
R. Marotta, I. Pesando, F. Pezzella and R. Russo for very useful discussions. A.L.
acknowledges support from Fondazione Angelo Della Riccia, and thanks NORDITA and
Universit degli Studi di Napoli for kind hospitality.

Appendix A. Alternative derivation of the bulk action


In this appendix we give an alternative derivation of the truncated bulk action necessary
to reconstruct the geometry of a fractional brane. This alternative strategy consists in
exploiting the S-duality relating type IIA on T4 /I4 to heterotic theory compactified on T4 .
Furthermore our task is made easier by turning on only the fields which couple to fractional
Dp-branes at tree level. For the sake of definiteness we refer here to the case of D0-brane
where the fields are the graviton g , 4 scalar fields a , the dilaton , 1 untwisted RR
vector C0 , 16 twisted RR vectors AI0 (I = 1, . . . , 16) and 16 twisted NSNS scalars D I .
One may easily check that under the chain of dualities (implementing the 6-dimensional
S-duality from IIA on T4 /I4 to the heterotic string on T4 )
IIA

T4 T9
T4
T4 4T
S
S
IIB
IIB
Type I T4 heterotic T4 ,
F
I4
(1) L I4
I4

(A.1)

52

M. Frau et al. / Nuclear Physics B 602 (2001) 3960

the previous fields transform as follows


he
= he ,
g = e2he g
,
1/2

 he 1
Gaa =
Ghe
, for a = 9,
Gaa
bb

and G99 =

1/2
Ghe
bb

(A.2)
Ghe
99 ,

99
9 he
C = Ghe
9 Ghe A ,


I

AIhe + AI+1 he
=
,

(A.3)
(A.4)

AI9 he + AI9+1 he
,
(A.5)

2
where the label he refers to heterotic fields.
The effective action of the heterotic string compactified on a 4-torus of volume 4 is
given by


  

he
4
he
d 6 x g he e2 R g he + 4 he he + 14 Gab Gab
S he = 2
210

he

he
a
1
h h
14 Gab F
F b 12

he

 
bd he
14 Gab ha hb 14 Gab Gcd hac
h
DI =

4
2
4g10


d 6x

16


he  I
 F
I + 2Gab F
aI F
bI he ,
F
g he e2

(A.6)

I =1

where we have assumed that the heterotic gauge group is broken to U (1)16 and have
I he follows

defined the field F
 I
he
I he

F
(A.7)
= F
+ Aa[ ] AIa .
To make contact with the case considered in the previous sections, we first neglect in the
action (A.6) all terms that contain fields not dual to those of the truncated type IIA theory
we are interested in, which contains only one pair of twisted fields (say (AJ , D J ) for a
given J ). Doing this, we get


  
4
he
d 6 x g he e2 R g he + 4 he he
S he = 2
210

he

he 
9
+ 14 Gaa Gaa 14 G99 F
F 9


 

4
he
I I

I 9 F
9I he . (A.8)
2
F
+ 2G99 F
d 6 x g he e2
F
4g10
I =J,J +1
By performing the S-duality transformations (Eqs. (A.2)(A.5)) on the previous action, we
get the following truncated low energy action for IIA on T4 /I4 :



4
1
a
6
2

R(g) + 4 a
S= 2
e F F
d x g e
4 a
210

M. Frau et al. / Nuclear Physics B 602 (2001) 3960

4
2
4g10

53

J
J
F J + 2F
C [ ] D J + C[ ] D J C [ ] D J
d x g F
6

+ 2e2 D J D J


e

(A.9)

a
J = AJ . After extracting the dilaton vacuum expectation
where F = [ C] and F
[ ]
value, one may rewrite the previous action in the Einstein frame as follows:



1
a  
1
e F F
S= 2
d 6 x g R(g) a a e
4
2orb
a

1 J J
1
1 
J [ ] J

J C
[ ] D
J
e F
e F
C D e C
[ ] D
F
4
2
2


a
 D

D
e
,

(A.10)

a
2 = g 2 /4 . Moreover the following
where orb has been defined in Section 2 and gorb
10
rescalings on the fields in Eq. (A.9) have been made:

2 orb J
gs 2
J


C ,
A =
A ,
C = 
1/2
gorb
a (2Ra )


(2Ra )1/2 J
J = orb a
D
(A.11)
D .
gorb gs 2

Under the identifications


 = D
J ,
D

 = A
J
A

the action (A.10) coincides with the one obtained in Section 2 with a different procedure.

Appendix B. Untwisted and twisted states


In this appendix we briefly review the description of the orbifold theory we are interested
in, namely type IIA compactified on T4 /Z2 , and its spectrum.
In a generic orbifold theory the periodicity conditions of the closed string are satisfied
up to the action of a generic element of the orbifold group. Thus in the case of the discrete
group ZN which consists of N elements {1, g, . . . , g N1 } we may have
X(, + ) = g m X(, )g m ,

with m {0, . . . , N 1}.

(B.1)

For m = 0 one recovers the boundary conditions of closed string without orbifold
projection, called untwisted boundary conditions, while for m = 0 one has N 1 different
twisted boundary conditions. In the case of a Z2 projection, acting as a reflection over four
space coordinates, the ten-dimensional Lorentz group SO(1, 9) decomposes in SO(1, 5)
SO(4). We fix our convention as follows: M, P , Q are the ten-dimensional indices; a, b
{6, 7, 8, 9} are the indices corresponding to the coordinates which are reflected by I4 and

54

M. Frau et al. / Nuclear Physics B 602 (2001) 3960

, {0, . . . , 5} are the directions transverse to the orbifold projection. The theory has a
unique twisted sector for each one of the sixteen orbifold fixed planes, characterized by the
following boundary conditions
X (, + ) = X (, ),

and Xa (, + ) = Xa (, )

(B.2)

for the bosonic coordinates and

(0, ) = (, ),

a
a

(0, ) =
(, )

(B.3)

a
a

(0, ) =
(, )

(B.4)

for the RR twisted sector and

(0, ) = (, ),

for the NSNS twisted sector. Obviously different boundary conditions correspond to
different mode expansions of the string coordinates. The bosonic string mode expansion of
the twisted sector is given by




  n 2in( ) n 2in( + )


+
e
e
X (, ) = q + 2 p + i
(B.5)
2
n
n
n=0

for each {0, . . . , 5} and





  ra 2ir( ) ra 2ir( + )
a
X (, ) = i
e
e
,
+
2
r
r

(B.6)

r=1/2

for each a {6, . . . , 9}. For the fermionic coordinates one finds
 M 2it ( )
 M 2it ( + )
M
M
= 
t e
,
+
= 
,

t e
t

where


and

t
ta

t
ta

and t , t Z
and ta , t Z +

and t , t Z +
a
and t , t Z

(B.7)

1
2

1
2

RR twisted sector

(B.8)

NSNS twisted sector.

(B.9)

To construct the massless spectrum of the theory one has to impose the mass-shell
condition separately in each sector. For the sake of simplicity we consider a 4-torus of the
type T4 = T1 T1 T1 T1 ; in this case the mass-shell condition in the twisted NSNS
sector reads as follows


 
2 
  na 2
2 
R
w
a
a
 + N + N 1 +
|state = 0
+
N + N

Ra

a
(B.10)
and must be imposed together with the level matching condition



 N N |state =
 + N
(B.11)
na wa |state .
N
a

M. Frau et al. / Nuclear Physics B 602 (2001) 3960

55

with N and N being the bosonic and fermionic occupation numbers, Ra the compactification radius of the coordinate Xa and finally na and wa
, respectively, KaluzaKlein and
winding modes along the compact direction a. For R =  it turns out that the massless
states in the NSNS untwisted sector are the graviton which transforms as (3, 3) under
the action of the little group SO(4), the KalbRamond field ((1, 3) + (3, 1)), the dilaton
((1, 1)), 10 KaluzaKlein scalars coming from the compactification of the ten-dimensional
graviton (10(1, 1)), and 6 scalars coming from the compactification of the KalbRamond
field (6(1, 1)). The explicit expressions of these states are
|h = h

12 12

|0 1 |0 1 |k

1
|na = wa = 0 ,
a

1
B
1 |k
|B = 1
|na = wa = 0 ,
1 |0 1 |0
2
a
2 2
a

1 |k
| = ( k L k L ) 1
1 |0 1 |0
2
2
8

1
|na = wa = 0 ,

a
a

1
a
a
1 |k
|na = wa = 0 ,
|a = a
1 1 |0 1 |0
a
2 2
a

1
Bab a
b
1 |k
|Bab =
|na = wa = 0 ,
1 1 |0 1 |0
2 2
a
2
a

(B.12)
(B.13)

(B.14)
(B.15)
(B.16)

where a is the self-dual volume of the compact direction xa [29,32]. Moreover h ,


B and are graviton, KalbRamond and dilaton field, respectively, while Bab and a
are the scalars corresponding to the compact components of KalbRamond field and to
the compact diagonal components of the metric. In the RR untwisted sector instead the
mass-shell condition reads as follows


 
2 
  na 2
w
R
2 
a
a
 + N + N +
N + N
+
|state = 0
(B.17)

Ra

a
and it should be imposed together with the level-matching condition given in Eq. (B.11).
In type IIA theory compactified on T4 /Z2 the massless states turn out to be 8 vectors
transforming as (2, 2) under the action of the little group SO(4), whereas in type IIB theory
on T4 /Z2 one find 8 scalars, and 3 2-forms potential, transforming, respectively, as (1, 1),
and ((3, 1) + (1, 3)) under the little group. Their explicit expression is given by


1
C1 ...n C 1 ...n + AB cos(0 0 )
|C(n) =
2 2 n!



 3/2 |k ,
+ C 1 ...n AB sin(0 0 ) |A 1/2 |B

(B.18)

where = (1 11 )/2, and 0 and 0 are the superghost zero-modes.


Let us examine now explicitly the twisted mass spectrum. In this case the mass-shell
condition is independent of the KaluzaKlein and winding modes and it reduces to

56

M. Frau et al. / Nuclear Physics B 602 (2001) 3960


2 
 + N + N |state = 0
N + N
(B.19)


both in the NSNS and RR twisted sectors. The NSNS twisted states are a product of two
spinors of the internal SO(4) having the same chirality (this is true both in type IIA and
type IIB theory on T4 /Z2 ). Therefore, from the point of view of the internal space, these
states transform as (2, 1) (2, 1) = (3, 1) + (1, 1) which corresponds to the transformation
I and a scalar that we denote by D I ,
properties of a self dual 2-form that we denote by Dab
where the index I runs from 1 to 16. All these states transform as scalars under the little
group, thus there are 4 16 scalars more in the theory.
Following the same notations of Section 3 the explicit expression of these states is
 I 
 ...n

1
D
 1
+ |l 1 |
DaI 1 ...n C
m 1 |k .

(B.20)
(n) =
lm
2 n!
Finally, in the RR twisted sectors, one finds only one massless state for each fixed point
which is a vector in type IIA and an antiself-dual 2 form plus a scalar in type IIB theory.
These states have the following expression
 I
A

0 ,...,m

1 AI0 ,...,m   0 ,...,m  


C
=
+ ab cos(0 0 )
m!
2
 ,...,


m 
 0
3/2 |k .
+ C
ab sin(0 0 ) |a 1/2 |b

(B.21)

Appendix C. Field equations


In this appendix we write down and solve the homogeneous equations of motion and fix
all the integration constants in order to obtain a solution consistent with the large distance
behavior of all the fields. For the sake of simplicity let us start by rewriting in components
the low-energy action obtained in Section 2


9


1
a a
d 6 x g R(g)
S= 2
2orb
a=6


ea F1 2 F 1 2 + FA1 2 FA 1 2 2 2 FA1 2 C [1 2 ] D
4 a


[1 2 ]

a
e
,
+ 2C[1 2 ] DC D D D
(C.1)
a

where we have dropped all tildes from the fields, to simplify the notation. The
corresponding equations of motion read as follows:


 1

2

e a F1 2 F 1 2 + FA1 2 FA 1 2
g = e
g
4
a

A
[1 2 ]
[1 2 ]
2 2 F1 2 C D + 2C[1 2 ] D C D (C.2)

M. Frau et al. / Nuclear Physics B 602 (2001) 3960

57

for the dilaton


 e

2
e a F1 2 F 1 2
ea D D
g a =
g
4 a
a
for the scalar fields




a
g e
e F
= e D 2C [ ] D 2 FA

g
a
for RR untwisted field




g e FA 2 e C [ ] D = 0

(C.3)

(C.4)

(C.5)

for RR twisted field







1
a

[ ]

g
e D + e C C D FA
=0
2
a

(C.6)

for the NSNS twisted field. Finally the Einstein equations are:
 

 1  
a a 12 g a a
R 2 g R 12 g
a



 

e
a
g
g A
F1 2 F 1 2 + 2FA FA
F1 2 FA1 2

e 2F F
4 a
2
2



g
F A C [1 2 ] D
2 2 2FA C [ ] D
2 1 2



g
+ 2 2C[ ] DC [ ] D
C [1 2 ] DC[1 2 ] D
2



1
ea D D g D D .

(C.7)
2
a
To obtain the fractional D0-brane solution, we make the following ansatz for the metric


ds 2 = B 2 (r) dt 2 + F 2 (r) dr 2 + r 2 d 2
(C.8)
and take all the fields to be function only of r. Under this assumption, introducing the
quantity ln B + 3 ln F the previous equations become

e  4
r e r r
r4



1
a
2 1 1
2
2
2
(r A0 ) + 2 (r A0 )C0 r D + (C0 r D) +
= B e
e (r C0 )
2 2
2 a
(C.9)
for the dilaton fields.


1
1
a
e  4
r e r r a = B 2 e
ea (r C0 )2
e (r D)2
4
4
2
r
a
a

(C.10)

58

M. Frau et al. / Nuclear Physics B 602 (2001) 3960

for the scalar fields





e
2 4
a
r e B r e
e r C0 = e B 2 r D 2C0 r D + 2 r A0
4
r
a

(C.11)

for the RR untwisted gauge field,




r e r 4 B 2 e r A0 + 2 C0 r D = 0

(C.12)

for the RR twisted vector field








4
a
2
2
=0
r e r 2
e r D B e C0 2 r A0 + 2C0 r D

(C.13)

for the NSNS twisted scalar field. And finally






Rrr = F 2 (r )2 +
(r a )2 +
ea (r D)2
a


2
3
2
ea r C0
(BF ) e
8
a



I 2
I
2
+ r A0 + 2 2 (r D)C0 r A0 + 2(C0 r D) ,

R00

(C.14)

3
2
= (BF ) e
ea (r C0 )2 + (r A0 )2
8
a

2
+ 2 2 (r D)C0 r A0 + 2(C0 r D) ,

(C.15)

1
2
= (BF ) e
ea (r C0 )2 + (r A0 )2
8
a

2
+ 2 2 (r D)C0 r A0 + 2(C0 r D)

(C.16)

for the Einstein equations. To simplify the set of Eqs. (C.9)(C.16) it is convenient to obtain
an equation for the quantity . This is achieved by combining the Einstein equations for
the R00 and R components, obtaining:
7r
+ (r )2 = 0
(C.17)
r
As asymptotically = 0 we take this to be the solution everywhere. Then by using the
dilaton equation and the one for the Rrr component, taking into accounts the asymptotic
behavior, one gets:
r2 +

B = F 3 = e 2 .
3

(C.18)

Expanding the fields up to the second order around their asymptotic values, one sees that
the twisted fields do not get corrections with respect to their harmonic asymptotic behavior

M. Frau et al. / Nuclear Physics B 602 (2001) 3960

59

Taking this as an ansatz for the full solution, one can easily solve the field equations. The
solution is written in Section 4.

References
[1] J. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231, hep-th/9711200.
[2] L. Girardello, M. Petrini, M. Porrati, A. Zaffaroni, JHEP 9812 (1998) 022, hep-th/9810126;
L. Girardello, M. Petrini, M. Porrati, A. Zaffaroni, JHEP 9905 (1999) 026, hep-th/9903026;
L. Girardello, M. Petrini, M. Porrati, A. Zaffaroni, Nucl. Phys. B 569 (2000) 451, hepth/9909047.
[3] J. Distler, F. Zamora, Adv. Theor. Math. Phys. 2 (1998) 1405, hep-th/9810206.
[4] D.Z. Freedman, S.S. Gubser, K. Pilch, N.P. Warner, Renormalization group flows from
holography supersymmetry and a c-theorem, hep-th/9904017;
D.Z. Freedman, S.S. Gubser, K. Pilch, N.P. Warner, JHEP 0007 (2000) 038, hep-th/9906194.
[5] I.R. Klebanov, N. Nekrasov, Nucl. Phys. B 574 (2000) 263, hep-th/9911096.
[6] I.R. Klebanov, A.A. Tseytlin, Nucl. Phys. B 578 (2000) 123, hep-th/0002159.
[7] K. Oh, R. Tatar, JHEP 0005 (2000) 030, hep-th/0003183.
[8] I.R. Klebanov, M.J. Strassler, JHEP 0008 (2000) 052, hep-th/0007191.
[9] J. Polchinski, M.J. Strassler, The string dual of a confining four-dimensional gauge theory,
hep-th/0003136.
[10] M. Bertolini, P. Di Vecchia, M. Frau, A. Lerda, R. Marotta, R. Russo, Nucl. Phys. B 590 (2000)
471, hep-th/0007097;
M. Bertolini, A. Lerda, Stable non-BPS D-branes and their classical description, hepth/0012169.
[11] M. Grana, J. Polchinski, Supersymmetric three-form flux perturbations on AdS5 , hepth/0009211.
[12] L.A. Pando Zayas, A.A. Tseytlin, 3-branes on resolved conifold, hep-th/0010088.
[13] M. Bertolini, P. Di Vecchia, M. Frau, A. Lerda, R. Marotta, I. Pesando, Fractional D-branes and
their gauge duals, hep-th/0011077.
[14] C.V. Johnson, K.J. Lovis, D.C. Page, Probing some N = 1 AdS/CFT RG flows, hep-th/0011166.
[15] J. Polchinski, N = 2 gauge/gravity duals, hep-th/0011193.
[16] A. Rajaraman, Supergravity duals for N = 2 gauge theories, hep-th/0011279.
[17] M.R. Douglas, JHEP 9707 (1997) 004;
D. Diaconescu, M.R. Douglas, J. Gomis, JHEP 9802 (1998) 013, hep-th/9712230.
[18] P. Di Vecchia, M. Frau, I. Pesando, S. Sciuto, A. Lerda, R. Russo, Nucl. Phys. B 507 (1997)
259, hep-th/9707068;
P. Di Vecchia, M. Frau, A. Lerda, A. Liccardo, Nucl. Phys. B 565 (2000) 397, hep-th/9906214.
[19] P. Di Vecchia, A. Liccardo, D-branes in string theories I, hep-th/9912161;
P. Di Vecchia, A. Liccardo, D-branes in string theories II, hep-th/9912275.
[20] C.V. Johnson, A.W. Peet, J. Polchinski, Phys. Rev D 61 (2000) 086001, hep-th/9911161.
[21] A. Buchel, A.W. Peet, J. Polchinski, Gauge dual and noncommutative extension of an N = 2
supergravity solution, hep-th/0008076.
[22] N. Evans, C.V. Johnson, M. Petrini, The enhanon and N = 2 gauge theory/gravity RG flows,
hep-th/0008081.
[23] C.V. Johnson, D-brane primer, hep-th/0007170.
[24] P. Aspinwall, K3 surfaces and string duality, hep-th/9611137.
[25] D. Diaconescu, J. Gomis, JHEP 0010 (2000) 001, hep-th/9906242.
[26] M.R. Gaberdiel, J.B. Stefanski, Nucl. Phys. B 578 (2000) 58, hep-th/9910109.
[27] T. Takayanagi, JHEP 0002 (2000) 040, hep-th/9912157.

60

M. Frau et al. / Nuclear Physics B 602 (2001) 3960

[28] M. Bill, B. Craps, F. Roose, Orbifold boundary states from Cardys condition, hep-th/0011060.
[29] M. Frau, A. Lerda, I. Pesando, R. Russo, S. Sciuto, Phys. Lett. B 400 (1997) 52, hepth/9702037;
M. Bill, P. Di Vecchia, M. Frau, A. Lerda, I. Pesando, R. Russo, S. Sciuto, Nucl. Phys. B 526
(1998) 199, hep-th/9802088.
[30] E. Eyras, S. Panda, Nucl. Phys. B 584 (2000) 251, hep-th/0003033.
[31] P. Merlatti, G. Sabella, World-volume action for fractional branes, hep-th/0012193.
[32] A. Giveon, E. Rabinovici, G. Veneziano, Nucl. Phys. B 322 (1988) 167.

Nuclear Physics B 602 (2001) 6186


www.elsevier.nl/locate/npe

Solution of the complex action problem in


the Potts model for dense QCD
M. Alford a , S. Chandrasekharan b , J. Cox c , U.-J. Wiese c
a Department of Physics and Astronomy, Glasgow University, Glasgow G12 8QQ, UK
b Department of Physics, Duke University, Durham, NC 27708, USA
c Center for Theoretical Physics, Massachusetts Institute of Technology (MIT), Cambridge, MA 02139, USA

Received 23 January 2001; accepted 20 February 2001

Abstract
Monte Carlo simulations of lattice QCD at non-zero baryon chemical potential suffer from the
notorious complex action problem. We consider QCD with static quarks coupled to a large chemical
potential. This leaves us with an SU(3) YangMills theory with a complex action containing the
Polyakov loop. Close to the deconfinement phase transition the qualitative features of this theory, in
particular its Z(3) symmetry properties, are captured by the 3-d 3-state Potts model. We solve the
complex action problem in the Potts model by using a cluster algorithm. The improved estimator
for the -dependent part of the Boltzmann factor is real and positive and is used for importance
sampling. We localize the critical endpoint of the first order deconfinement phase transition line and
find consistency with universal 3-d Ising behavior. We also calculate the static quarkquark, quark
antiquark, and antiquarkantiquark potentials which show screening as expected for a system with
non-zero baryon density. 2001 Published by Elsevier Science B.V.
PACS: 12.38.Aw; 05.50.+q; 64.60.-i

1. Introduction and summary


Non-perturbative dense QCD can presently not be studied from first principles because
Monte Carlo simulations of lattice QCD with non-zero baryon chemical potential suffer from a severe complex action problem. The Boltzmann factor in the path integral can
then not be interpreted as a probability and standard importance sampling methods fail. In
particular, when the -dependent part of the Boltzmann factor is included in the measured
observables, due to severe cancellations the required statistics is exponentially large in the
spacetime volume [1,2].

This work is supported in part by funds provided by the U.S. Department of Energy (D.O.E.) under
cooperative research agreements DE-FC02-94ER40818 and DE-FG02-96ER40945.
E-mail address: cox@mit.edu (J. Cox).

0550-3213/01/$ see front matter 2001 Published by Elsevier Science B.V.


PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 0 6 8 - 2

62

M. Alford et al. / Nuclear Physics B 602 (2001) 6186

The complex action problem prevents the numerical simulation of a large class of
interesting physical systems including other field theories at non-zero chemical potential or
non-zero -vacuum angle as well as some fermionic field theories with an odd number of
flavors. A special case of the complex action problem is the so-called fermion sign problem
which arises for fermionic path integrals formulated in a Fock state basis. The problem is
due to paths that correspond to an odd permutation of fermion positions which contribute
negatively to the path integral. There are numerous condensed matter systems ranging from
the repulsive Hubbard model away from half-filling to antiferromagnetic quantum spin
systems in an external magnetic field that cannot be simulated with standard Monte Carlo
algorithms. Meron-cluster algorithms have been used to solve the sign or complex action
problems in several of these cases. For example, the first meron-cluster algorithm has led
to a solution of the complex action problem in the 2-d O(3) symmetric field theory at nonzero -vacuum angle [3]. In this model, some of the clusters are half-instantons, so they
are called meron-clusters. The complex action problem also arises in the 2-d O(3) model
at non-zero chemical potential. When formulated as a D-theory [47] i.e., in terms
of discrete variables that undergo dimensional reduction the complex action problem
has also been solved with a meron-cluster algorithm [8]. Recently, the meron concept has
been generalized to fermions [9]. Meron-cluster algorithms have led to a complete solution
of the fermion sign problem in a variety of models including non-relativistic spinless
fermions [9,10], relativistic staggered fermions [1113] and some models in the Hubbard
model family [8,14]. 1 Recently, a meron-cluster algorithm has been used to solve the sign
problem that arises for quantum antiferromagnets in an external magnetic field [15]. For
a review of these recent developments and a preliminary version of the present results
see [16,17].
In the conventional formulation of lattice QCD the quarks are represented by Grassmann
fields. When the quarks are integrated out, they leave behind a fermion determinant that
acts as a non-local effective action for the gluons. At zero chemical potential and for
an even number of flavors, the fermion determinant is real and positive and can thus be
interpreted as a probability for generating gluon field configurations. Despite the fact that
standard importance sampling techniques apply, the non-local nature of the effective gluon
action makes lattice QCD simulations with dynamical fermions very time consuming. With
a non-zero chemical potential for the baryon number, the fermion determinant becomes
complex and standard importance sampling techniques fail completely [1,2]. This is the
reason why non-perturbative QCD at non-zero baryon density can presently not be studied
from first principles.
It is natural to ask if a meron-cluster algorithm could be used to solve the complex action
problem in QCD. When one integrates out light quarks, one obtains a non-local effective
action for the gluons and it appears unlikely that the meron concept will apply. On the other
hand, when one describes the quarks in a Fock state basis, the complex action problem is
still present, in the form of a fermion sign problem. Our hope is that this problem will
eventually be solved by a meron-cluster algorithm applied to the D-theory formulation of
1 The models investigated so far only show s-wave superconductivity.

M. Alford et al. / Nuclear Physics B 602 (2001) 6186

63

QCD [47], since the quark and gluon degrees of freedom are then discrete and are much
easier to handle. In this paper we address a simpler situation first. We consider QCD in the
limit of very heavy quarks with a large chemical potential. These can be integrated out,
introducing Polyakov loops into the effective gluon action. When quarks are integrated out
at non-zero chemical potential we expect a complex action, and in this case it arises
because a Polyakov loop and its charge conjugate get different weights when = 0.
Polyakov loops are only non-local in the Euclidean time direction, so this effective gluon
action is more manageable than the one that arises for a general fermion determinant.
Indeed, Blum, Hetrick and Toussaint have simulated the theory in this form on lattices
of moderate size where the complex action problem is less severe [18]. Recently, Engels,
Kaczmarek, Karsch and Laermann have studied QCD with heavy quarks at fixed baryon
number. Again, for moderate baryon density and moderate volumes the complex action
problem is not too severe and simulations are possible [19]. Ultimately one would like
to be able to solve the complex action problem for this gluon action completely. At the
moment, we still cannot apply a meron-cluster algorithm to solve the problem, because
the construction of efficient cluster algorithms for non-Abelian gauge theories seems to
be impossible for Wilsons formulation of lattice field theory. Here we will simplify
the problem further by replacing the gauge dynamics by that of the Z(3) Potts model
representing the Polyakov loops [20,21]. We have found a cluster algorithm that solves
this complex action problem in the Potts model approximation to QCD.
The 3-d Z(3)-symmetric Potts model has often been used as an approximation to QCD
with static quarks. In particular, the phase transition to a broken Z(3) symmetry phase at
high temperature corresponds to the first order deconfining phase transition in QCD. As
has been noted by Condella and DeTar, a term that corresponds to a chemical potential can
also be included in the Potts model, explicitly breaking the Z(3) symmetry [22]. As the
coefficient of this term grows, the first order deconfinement phase transition persists but
it becomes weaker and ultimately disappears in a critical endpoint. This point is expected
to be in the universality class of the 3-d Ising model. In this paper we will confirm this
expectation with numerical simulations.
In principle, one can imagine deriving an effective 3-d 3-state Potts model directly from
QCD by integrating out all degrees of freedom except for the Z(3) phase of the Polyakov
loop. However, the resulting Potts model action would be very complicated and cannot
be derived in practice, except in the strong coupling limit. Here we approximate QCD
with heavy quarks by a 3-d Z(3)-symmetric Potts model with a standard nearest-neighbor
interaction. Universal features like the nature of the critical endpoint of the deconfinement
phase transition are correctly reproduced in this approximation. Fig. 1 contains the phase
diagram of the 3-d 3-state Potts model in the (h, )-plane. The parameter h represents
exp(( M)) in QCD with quarks of mass M at chemical potential . We study the
limit M, for any given M. Large h corresponds to > M and small h to
< M. Because M  M, we are always, for any h, in the immediate neighborhood
of the onset of non-zero density for the heavy quarks. This means that it does not matter
whether they are fermions or bosons, since they never move. The difference would only
become apparent above the onset, where either a Fermi surface or a degenerate Bose gas

64

M. Alford et al. / Nuclear Physics B 602 (2001) 6186

Fig. 1. The phase diagram of the Z(3) Potts model in the (h, )-plane. The ordinary deconfinement
phase transition at T = (0, 0.550565(10)) is a triple point from which a line of first order phase
transitions emerges. This line terminates in the critical endpoint E = (0.000470(2), 0.549463(13))
and continues only as a crossover.

would occur, and our order of limits is such that we never get that far from the onset.
The parameter is the standard Potts model coupling, which corresponds roughly to the
temperature T = 1/. The ordinary first-order deconfinement phase transition at h = 0
(point T in Fig. 1) extends into a line of first order transitions that terminates in the critical
endpoint E. This endpoint occurs at such a low value of h that the complex action problem
is not very severe there, and we found that the most efficient way to locate and study it
was to employ a reweighted Metropolis algorithm, which can in this case be applied at
volumes large enough to show the critical behavior. Similar methods were used recently
by Karsch and Stickan [23] in a version of the 3-d 3-state Potts model where the action
is real, and the endpoint was found to have the critical exponents of the 3-d Ising model.
We find that in the Potts model with complex action the endpoint has the same critical
properties. Furthermore its position is barely shifted in comparison to the model with real
action. In this paper we do not limit our attention to the endpoint, but develop a method
that solves the complex action problem everywhere in the phase diagram.
We also calculate the potentials between static quarks and antiquarks in the Potts model
approximation to QCD. In the confined phase at = 0 the static quarkantiquark potential
is linearly rising with the distance as a manifestation of confinement. For the same reason

M. Alford et al. / Nuclear Physics B 602 (2001) 6186

65

the quarkquark and antiquarkantiquark potentials are infinite at all distances. In the
deconfined phase the quarkantiquark potential reaches a plateau at twice the (now finite)
free energy of a quark. Similarly, the quarkquark and antiquarkantiquark potentials are
no longer infinite. It should be noted that quarkquark and antiquarkantiquark potentials
are usually not calculated in lattice simulations. This is because as a consequence of
the Z(3) Gauss law quark or antiquark pairs cannot exist in a finite spatial volume
with periodic boundary conditions [24]. Interestingly, this changes for = 0 because then
there are compensating background charges in the medium that can absorb the Z(3) flux
of an external quark. Since the chemical potential explicitly breaks the Z(3) symmetry,
there is no longer a clear distinction between confinement and deconfinement for = 0.
This manifests itself in the phase diagram by the fact that confined and deconfined phases
are analytically connected. Fig. 2 shows the quarkantiquark, quarkquark and antiquark
antiquark potentials on the confined side (a) and on the deconfined side (b) of the crossover.
Note that at = 0 even in the confined phase the quarkantiquark potential now reaches a
plateau. The plateau height corresponds to the sum of the free energies FQ of an external
quark and FQ
 of an external antiquark. For > 0 quarks are favored in the medium while
antiquarks are suppressed. As a consequence, the free energy of an external static quark is
larger than that of an external static antiquark. While an external static antiquark can bind
with a single background quark from the medium and form a meson, an external static
quark needs two quarks from the medium to form a baryon. Indeed, on the confined side
of the transition FQ is clearly larger than FQ
 , while on the deconfined side FQ and FQ

are more or less the same. We have normalized the potentials such that at zero distance
a static quarkantiquark pair has zero energy. In the Potts model, two quarks at zero
distance are indistinguishable from a single antiquark, and similarly, two antiquarks on top
of each other behave like a single quark. Hence, at zero distance the quarkquark potential
VQQ (0) agrees with the free energy of a single antiquark FQ
 and the antiquarkantiquark
potential obeys VQ
(0)
=
F
.
At
asymptotic
distances
the
potentials VQQ
Q

Q
 (), VQQ ()
and VQ
Q
 () take the values FQ + FQ
 , 2FQ and 2FQ
 , respectively. This behavior is
consistent with our numerical data shown in Fig. 2.
In the absence of a chemical potential, the Potts model can be simulated with the original
SwendsenWang cluster algorithm [25]. When a chemical potential is introduced, the
Potts model suffers from the complex action problem and standard importance sampling
methods including the cluster algorithm fail. In this paper, we construct an improved
estimator for the -dependent part of the Boltzmann factor by averaging analytically over
all configurations related to each other by cluster flips. In contrast to the original Boltzmann
factor, the improved estimator is real and positive and can be used for importance sampling.
This solves the complex action problem completely.
Although the Potts model inherits the complex action problem from QCD, it can be
transformed into a flux model that has no complex action problem [22]. The flux model
has been simulated in [22] and the disappearance of the first order deconfining phase
transition at large chemical potential has been observed numerically. These results may
give encouragement to the hope that QCD itself could be transformed into a model without
a complex action problem. In this paper we show that, at least for the Potts model, it is

66

M. Alford et al. / Nuclear Physics B 602 (2001) 6186

Fig. 2. The static quarkantiquark, quarkquark and antiquarkantiquark potentials (a) on the
confined side (at h = 0.01, = 0.50) and (b) on the deconfined side (at h = 0.01 and = 0.56)
of the crossover.

M. Alford et al. / Nuclear Physics B 602 (2001) 6186

67

more efficient to leave it in its usual form and solve the complex action problem with our
cluster algorithm than to transform it into a flux model and use conventional Metropolis
methods.
The paper is organized as follows. Section 2 contains a derivation of the effective gluon
action resulting from static quarks with large chemical potential as well as its Potts model
approximation. In Section 3 we describe the cluster algorithm that solves the complex
action problem. Section 4 contains the derivation of the flux representation of the Potts
model and a description of a Metropolis algorithm to simulate it. A comparison of the
Metropolis algorithm for the flux model and the cluster algorithm for the original Potts
model shows that the latter is more efficient. In Section 5 we present the physical results
concerning the critical endpoint E. Using finite-size scaling techniques, we are able to
determine the position of the critical endpoint of the deconfinement phase transition to
high accuracy. Our results are consistent with the expected universal 3-d Ising behavior.
Finally, Section 6 contains our conclusions.

2. QCD with heavy quarks and the 3-d 3-state Potts model
The partition function for a pure SU (3) YangMills theory is given by

Z = DA exp(S[A]),

(2.1)

where



S[A] =

dt

d 3x

1
Tr[F F ],
2g 2

(2.2)

is the Euclidean action for the gluons and is the inverse temperature. The action is
invariant under gauge transformations
g(
x , 0) = g(
x , )z,

(2.3)

that are periodic in Euclidean time up to an element z of the center Z(3) = {exp(2in/3),
n = 1, 2, 3} of the non-Abelian gauge group. In the presence of a single external heavy
quark of bare mass M at an undetermined position x the partition function turns into

ZQ = DA [A] exp(S[A]) exp(M),
(2.4)
where

[A] =

 

d x Tr P exp dt A4 (
x , t) ,
3

(2.5)

is the spatial integral of the Polyakov loop. Ultimately, the mass M will be sent to infinity.
Note that while the center transformation of Eq. (2.3) leaves the pure gluon action S[g A] =
S[A] invariant, the Polyakov loop transforms into
[g A] = z[A].

(2.6)

68

M. Alford et al. / Nuclear Physics B 602 (2001) 6186

This shows that in the presence of the external quark, the Z(3) symmetry is explicitly
broken. The partition function for a system of gluons in the presence of a single heavy
antiquark is given by

ZQ
(2.7)
DA [A] exp(S[A]) exp(M),
=
where denotes complex conjugation. Let us now consider a system of gluons in a
background of n static quarks and n static antiquarks. The partition function then takes
the form



1
1
.
Zn,n = DA [A]n ([A] )n exp(S[A]) exp M(n + n)
(2.8)
n!
n!

The factors 1/n! and 1/n!


appear because quarks are indistinguishable, as are antiquarks.
Introducing the quark chemical potential that couples to (n n),
i.e., three times the
baryon number, we obtain the grand canonical partition function



Zn,n exp (n n)

Z() =
n,n


n,n


=

DA



1
1
[A]n ([A] )n exp S[A] n(M ) n(M

+ )
n!
n!



DA exp S[A] + exp((M ))[A] + exp((M + ))[A] .
(2.9)

As expected, the presence of quarks and antiquarks leads to an explicit breaking of the
Z(3) center symmetry. Furthermore, in the presence of a non-zero chemical potential the
effective action for the gluons is complex. Note that in the SU(2) case the action remains
real because then the Polyakov loop itself is real, i.e., [A] = [A]. The action becomes
real even in the SU (3) case if is purely imaginary. Furthermore, one can see that the
chemical potential explicitly breaks the charge conjugation symmetry that replaces [A]
by [A] . In fact, under charge conjugation the action turns into its complex conjugate. We
have assumed that the quarks are static. Hence, to be consistent we must consider the limit
M . In order to obtain a non-trivial result, we simultaneously take the limit
such that M remains finite. The partition function then simplifies to





Z() = DA exp S[A] + exp (M ) [A] .
(2.10)
As discussed in [18] and [19], a similar result can be obtained by simplifying the full
QCD quark determinant in the static quark limit. In general the determinant would contain
all Wilson loops, but because M is large most of them are suppressed. The only ones that
survive are those for which the enhancement due to the chemical potential compensates for
the suppression due to the mass. These are the Polyakov loops that progress in a straight
line from Euclidean time t = 0 to t = at some position x . In the loop expansion of the
quark determinant, each of these has a weight exp((M )).

M. Alford et al. / Nuclear Physics B 602 (2001) 6186

69

Up to this point we have treated QCD consistently in the static quark limit. The resulting
effective action for the gluons is complex and we presently do not know how to simulate it
efficiently. For that reason we now replace the gluon system by a simple 3-d lattice 3-state
Potts model. The Potts spins x Z(3) replace the original Polyakov loop variables and
the partition function turns into




x ,
Z(h) = D exp S[] + h
(2.11)
x

where h replaces exp((M )). Note that the Potts model action is still complex. In
principle, one can imagine integrating out all QCD degrees of freedom except for the Z(3)
phase of the Polyakov loop and thus derive an effective Potts model action directly from
QCD. In practice this is impossible, except in the strong coupling limit. For simplicity, we
therefore replace the pure gluon action S[A] by a standard nearest-neighbor Potts model
interaction

x ,x+i .
S[] =
(2.12)
x,i

The coupling constant is not related in a simple way to the parameters of QCD. Still,
a large value of corresponds qualitatively to the high-temperature deconfined phase,
while small values correspond to the confined phase. As mentioned in the introduction,
the Potts model also retains the general features of the QCD phase diagram. At h = 0
(M infinite, finite) there is a first-order phase transition as a function of , between the
disordered (confined) phase that respects the Z(3) symmetry and the ordered (deconfined)
phase that spontaneously breaks it. An order parameter for this transition is . As h rises
from zero, the chemical potential term explicitly breaks the Z(3) symmetry, the phase
transition weakens, and then ends at a critical point. Correspondingly, in heavy-quark
QCD the quarks begin to contribute to the partition function when gets close to M, and
there is no longer an order parameter for deconfinement. The deconfining phase transition
terminates at a critical endpoint.

3. Cluster algorithm solution of the complex action problem


In this section we first discuss the general nature of the complex action problem and
then discuss the cluster algorithm that solves this problem for the Potts model. We also
construct improved estimators for various physical quantities.
3.1. The general nature of the complex action problem
When the action is complex the resulting Boltzmann factor cannot be interpreted as
a probability and hence standard importance sampling techniques fail. When one uses
just the absolute value of the Boltzmann factor for importance sampling and includes its
complex phase in measured observables O, expectation values take the form

70

M. Alford et al. / Nuclear Physics B 602 (2001) 6186





1
O =
x
D O[] exp S[] + h
Z
x




O exp ih x Im x R
 .
= 

exp ih x Im x R

(3.1)

The subscript R refers to a modified ensemble with a real action described by the partition
function




ZR = D exp S[] + h
(3.2)
Re .
x

By definition we have



exp ih
Im x
1
=
ZR

R
D exp ih


x

Im x exp S[] + h


x


Re



Z
exp V (f fR ) ,
(3.3)
ZR
where f and fR are the free energy densities of the original complex and the modified
real action systems, respectively, and V is the spatial volume. Hence, the denominator in
Eq. (3.1) becomes exponentially small as one increases the volume. The same is true for
the numerator, because O itself is not exponentially large in V .
Although, in principle, simulating the modified ensemble is correct, in practice this
method fails for large volumes. The reason is that observables are obtained as ratios
of exponentially small numerators and denominators which are themselves averages
of quantities of order one. This leads to very severe cancellations and requires an
exponentially large statistics in order to obtain accurate results. To see this, we estimate the
relative statistical error in the determination of the average phase of the Boltzmann factor

exp(ih x Im x ). Since exp(ih x Im x )R = Z/ZR the average itself is real. When
one generates N statistically independent field configurations in a Monte Carlo simulation,
the resulting error to signal ratio is given by



& exp ih x Im x



exp ih x Im x R
  

 

 2
exp ih

x Im x exp ih
x Im x R R
=



N exp ih x Im x R

 2


1 exp ih x Im x R
exp(V (f fR ))
= 

(3.4)
.

N exp ih x Im x R
N

For large V we have used exp(ih x Im x )R  1 as implied by Eq. (3.3). Consequently, in order to obtain an acceptable error to signal ratio one must generate at least
N exp(2V (f fR )) configurations. For large volumes this is impossible in practice.
=

M. Alford et al. / Nuclear Physics B 602 (2001) 6186

71

3.2. The cluster algorithm for the Potts model


Let us now outline the ideas that underlie the cluster algorithm that we use to solve
the complex action problem. It is based on the original SwendsenWang cluster algorithm
[25] for the Potts model without chemical potential. In fact, in the limit h = 0 our algorithm
reduces to that algorithm. The SwendsenWang cluster algorithm decomposes the lattice
into independent clusters of connected sites. Each spin belongs to exactly one cluster and
all spins within a cluster are assigned the same random Z(3) element. In this paper, we

construct an improved estimator for the h-dependent part exp(h x x ) of the Boltzmann
factor by analytically averaging it over all configurations related to each other by cluster

flips. Although, for an individual configuration exp(h x x ) is in general complex, its


improved estimator is always real and positive and can thus be used for importance
sampling. This completely solves the complex action problem.
Let us first describe the original SwendsenWang algorithm for h = 0. In this method
one introduces variables b = 0, 1 for each bond connecting neighboring lattice sites x and
y = x + i and one writes the nearest neighbor Boltzmann factor as




exp(x ,y ) =
(3.5)
b,1 x ,y e 1 + b,0 .
b=0,1

In the enlarged configuration space of spin and bond variables, the bond variables impose
constraints between the spin variables. When a bond is put (i.e., when b = 1), the spin
Boltzmann factor is x ,y (e 1) and hence the spin variables x and y at the two
ends of the bond must be identical. On the other hand, when the bond is not put (b = 0), the
spin Boltzmann factor is 1 and thus the variables x and y fluctuate independently. The
spin variables, in turn, determine the probability to put a bond. When the spins x and y
are different, the bond Boltzmann factor is b,0 and thus the bond is not put. On the other
hand, when x and y are the same, the bond Boltzmann factor is [b,1 (e 1) + b,0 ].
Consequently, a bond between parallel spins is put with probability p = 1 e . Note that
for = 0 no bonds are put, while for = parallel spins are always connected by a bond.
The SwendsenWang cluster algorithm updates bond and spin variables in alternating
order. First, for a given spin configuration, bonds are put with probability p between
parallel neighboring spins. No bonds are put between non-parallel spins. Then the spins are
updated according to the constraints represented by the resulting bond configuration. Spins
connected by bonds must remain parallel, while spins not connected by bonds fluctuate
independently. Hence, to update the spins, one must identify clusters, i.e., sets of spins that
are connected by bonds. All spins in a cluster are parallel and are assigned the same random
Z(3) element in the spin update. All spins belong to exactly one cluster. It should be
noted that a cluster may consist of a single spin. A configuration consisting of NC clusters
can be viewed as a member of a sub-ensemble of 3NC equally probable configurations
which result by assigning Z(3) elements to the various clusters in all possible ways. As
was already pointed out by Swendsen and Wang, one can construct improved estimators
for various physical quantities by averaging analytically over all 3NC configurations in a
sub-ensemble. Since the number of clusters is proportional to the volume, this effectively
increases the statistics by a factor that is exponentially large in V .

72

M. Alford et al. / Nuclear Physics B 602 (2001) 6186

Let us construct an improved estimator for the h-dependent part exp(h x x ) of the
Boltzmann factor. Although for an individual configuration this term is in general complex,
its average over a sub-ensemble of 3NC configurations is always real and positive. This
results from the following observations. The h-dependent part of the Boltzmann factor is a
product of cluster contributions







x =
exp h
x .
exp h
(3.6)
x

xC

Since the clusters are independent, the sub-ensemble average is a product









exp h
x
=
x
,
exp h
3NC

xC

(3.7)

of 3-state averages for the individual clusters





1 
exp h
x
=
exp(h|C|)
3
xC

Z(3)



1
= exp(h|C|) + 2 exp(h|C|/2) cos 3h|C|/2
3
= W (C),

(3.8)

which defines a weight W (C) for each cluster. We have used the fact that all spins x in

a given cluster C take the same value Z(3) so that xC x = |C| where |C| =

xC 1 is the cluster size. It is easy to show that the expression in Eq. (3.8) is always
positive and can hence be used for importance sampling. This is crucial for a complete
solution of the complex action problem.
For a given bond configuration one can integrate out the spin variables and one obtains


N

W (C)
Z = Db e 1 b 3NC

=

Db e 1

Nb 



exp(h|C|) + 2 exp(h|C|/2) cos 3h|C|/2 .

(3.9)

Here Nb is the number of bonds that are put (i.e., have b = 1). The factor 3NC represents the
number of allowed spin configurations for a given bond configuration and the factors W (C)
come from the improved estimator. The effective action for the bond variables depends only
on the sizes |C| of the clusters corresponding to a given bond configuration. Note that the
factor 1/3 per cluster in Eq. (3.8) cancels against the factor 3NC .
Our algorithm directly updates the partition function of Eq. (3.9), i.e., it only operates
on the bond variables while the spins are already integrated out analytically. 2 The bond
variables that define the clusters are updated with a local algorithm. A bond whose value
does not affect the cluster sizes is put with probability p = 1 e . This happens when
the two sites at its ends belong to the same cluster because they are connected indirectly
2 B. Scarlet was first to realize that the spin variables need not even be simulated.

M. Alford et al. / Nuclear Physics B 602 (2001) 6186

73

through other bonds. A bond whose value affects the cluster sizes is put with a probability
that depends on the sizes of the corresponding clusters. When the bond is not put (b = 0),
its endpoints x and y belong to two different clusters C1 and C2 of sizes |C1 | and |C2 | and
the corresponding Boltzmann weight is 32 W (C1 )W (C2 ). On the other hand, when the bond
is put (b = 1), its endpoints belong to the combined cluster C1 C2 of size |C1 | + |C2 |.
In that case, the Boltzmann weight is 3W (C1 C2 )(e 1). Hence, the bond is put with
probability
q=

W (C1 C2 )(e 1)
.
3W (C1 )W (C2 ) + W (C1 C2 )(e 1)

(3.10)

3.3. Improved estimators for physical quantities


In order to measure physical observables, it is crucial to construct improved estimators
for them as well. Here we construct improved estimators for the Polyakov loop x ,
its charge conjugate x , as well as for the correlators x y , x y and x y . The
expectation values

x  = exp(FQ ),
(3.11)
x = exp(FQ
 ),
determine the free energies FQ of a quark and FQ
 of an antiquark. The Polyakov loop
correlators determine the quarkantiquark potential VQQ
 (x y), the quarkquark potential
VQQ (x y) and the antiquarkantiquark potential VQ
Q
 (x y) via




exp VQQ
 (x y) = x y ,



exp VQQ (x y) = x y ,


exp VQ
(3.12)
Q
 (x y) = x y .
The improved estimator for the Polyakov loop is given by the sub-ensemble average





1 
z
=
exp(h|Cx |)
W (C),
x exp h
(3.13)
3
NC
z
3

Z(3)

C=Cx

where Cx is the cluster that contains the point x. Hence, we obtain






N
1
1
x  =
exp(h|Cx |) e 1 b 3NC
W (C),
Db
Z(h)
3W (Cx )
Z(3)

(3.14)
i.e., after integrating out the spin variables, the Polyakov loop is represented by
x =


1
exp(h|Cx |).
3W (Cx )

(3.15)

Z(3)

Similarly, the operator representing the charge conjugate Polyakov loop is


x =


1
exp(h|Cx |).
3W (Cx )
Z(3)

(3.16)

74

M. Alford et al. / Nuclear Physics B 602 (2001) 6186

The improved estimator for the Polyakov loop correlator x y is given by the subensemble average




1 
1 

x y exp h
z
=
exp(h|Cx |)
exp(h|Cy |)
3
3
z
Z(3)
Z(3)
3NC


(3.17)
W (C),
C=Cx ,Cy

if the points x and y belong to two different clusters Cx and Cy . If the points x and y
belong to the same cluster (i.e., if Cx = Cy ), the improved estimator is simply given by





x y exp h
(3.18)
z
=
W (C),
3NC

because then

x y

x y =

= 1. Hence, the operator representing the correlator is



1
exp(h|Cx |)
exp(h|Cy |),
9W (Cx )W (Cy )
Z(3)

Z(3)

if Cx = Cy ,
x y = 1,

if Cx = Cy .

(3.19)

Similarly, we have
x y =



1
exp(h|Cx |)
exp(h|Cy |),
9W (Cx )W (Cy )
Z(3)

if Cx = Cy ,

1
2 exp(h|Cx |),
x y =
3W (Cx )

Z(3)

if Cx = Cy ,

Z(3)

x y =



1
exp(h|Cx |)
exp(h|Cy |),
9W (Cx )W (Cy )
Z(3)

if Cx = Cy ,

1
( )2 exp(h|Cx |),
x y =
3W (Cx )

Z(3)

if Cx = Cy .

(3.20)

Z(3)

3.4. Severity of the complex action problem


In order to estimate the severity of the complex action problem, we like to determine the
expectation value of the complex phase of the Boltzmann factor in the modified real action
ensemble



Z
Im x
=
.
exp ih
(3.21)
Z
R
x
R

Rather than implementing this directly in a simulation that uses the absolute value of
the Boltzmann factor for importance sampling, one can measure ZR /Z with the cluster

M. Alford et al. / Nuclear Physics B 602 (2001) 6186

75

Fig. 3. The expectation value of the phase factor as a function of the volume at the critical endpoint E.


algorithm. In fact, an improved estimator for this quantity is given by C WR (C)/W (C),
where




1
WR (C) = exp h
(3.22)
Re x
= exp(h|C|) + 2 exp(h|C|/2) ,
3
xC

is the weight that replaces W (C) in the real action ensemble. Alternatively, one can
construct a cluster algorithm that simulates the real action ensemble. In that case, one

needs to measure C W (C)/WR (C) in order to obtain Z/ZR .

Fig. 3 shows exp(ih x Im x )R as a function of the volume V = L3 at the critical


endpoint of the transition line (point E in Fig. 1). Indeed, one finds an exponentially small
signal, as expected from Eq. (3.3). Defining a scale parameter L0 that measures the severity
of the complex action problem by





L3
Im x
exp 3
exp ih
(3.23)
L0
x
R
we get L0 80 at the endpoint E. This means that the complex action problem at the
endpoint is extremely mild. In fact, in practice it is not a problem at all up to volumes as
big as 1003 . Since the current computer hardware restricts the system size to a couple of
million degrees of freedom, one can also study the point E with an algorithm that does
not solve the complex action problem. On very large lattices the meron-cluster algorithm

76

M. Alford et al. / Nuclear Physics B 602 (2001) 6186

Fig. 4. The scale parameter L0 related to the severity of the complex action problem as a function of
h at = 0.5495. The solid line is a spline to guide the eye. The phase transition takes place at the
dotted vertical line.

will become superior also at the point E since its computational effort is polynomial in the
system size as opposed to exponential for the reweighted Metropolis algorithm.
It should be noted that the complex action problem is most severe for intermediate values
of h. While it is obvious that there is no complex action problem at h = 0, it is perhaps less
obvious that there is also no problem for large h. This is because
WR (C)
exp(h|C|) + 2 exp(h|C|/2)
=

W (C)
exp(h|C|) + 2 exp(h|C|/2) cos( 3h|C|/2)

(3.24)

approaches 1 in the limit h so that exp(ih x Im x )R R = ZR /Z 1. Fig. 4


shows the complex action problem scale parameter L0 as a function of h for fixed =
0.5495. It has a minimum at h 1 meaning that the complex action problem is the most
severe in that region. Defining the practical complex action problem as being present

when exp(ih x Im x )R < 0.01 on a 1003 lattice (i.e., L0 < 60) we see that at =
0.5495 there is no practical complex action problem for h < 0.003 as well as for h > 6.
A more physical definition of a practical complex action problem would compare L0 with
the correlation length but we have not measured the correlation length.

M. Alford et al. / Nuclear Physics B 602 (2001) 6186

77

4. Flux representation of the Potts model


In this section we map the Potts model to an equivalent flux model that does not suffer
from the complex action problem. Then we describe a Metropolis algorithm to update the
flux model and compare its efficiency with the cluster algorithm.
4.1. Mapping the Potts model to a flux model
As pointed out in [22] the Potts model can be rewritten as a flux model that does not
suffer from the complex action problem. As we have seen, the complex action problem can
also be solved in the cluster formulation. Hence, the question arises if the flux or the cluster
formulation leads to more efficient numerical simulations. Let us first match the flux model
to the original Potts model. The flux model is formulated in terms of electric charges
Qx {0, 1} defined on the lattice sites x and electric flux variables Ex,i {0, 1} living
on the links. The charge and flux variables are related by the Z(3) Gauss law constraint

Qx =
(4.1)
(Ex,i Exi,i
) mod 3.
i

The action of the flux model takes the form


S[E, Q] =



g2  2
MQ2x Qx .
Ex,i +
2
x

(4.2)

x,i

The mass M and the chemical potential of the Z(3) charges are not directly related to
the mass and chemical potential of quarks in QCD (which we also denoted by M and
in Section 2) but qualitatively they play the same role. The partition function of the flux
model takes the form
    
x exp(S[E, Q]).
Z=
(4.3)
x Qx {0,1} x,i Ex,i {0,1} x

The -function x imposes the Gauss law of Eq. (4.1) at the point x and can be written as
x =

Qx i (Ex,i Exi,i
1 
)
x
.
3

(4.4)

x Z(3)

Inserting this as well as Eq. (4.2) for the action in Eq. (4.3) one can integrate out the Ex,i
and Qx variables. The result of the Ex,i integration is

 2
 2
 

Ex,i

g
g 2

Re x x+i . (4.5)
x x+i
exp Ex,i = 1 + 2 exp
2
2
Ex,i {0,1}

In the Potts model (up to an overall factor) the corresponding term is exp(x ,x+i ). Thus,
the flux model matches the Potts model if
exp() =

1 + 2 exp(g 2 /2)
.
1 exp(g 2 /2)

(4.6)

78

M. Alford et al. / Nuclear Physics B 602 (2001) 6186

Note that the g 0 limit of the flux model corresponds to the limit of the Potts
model. When one integrates out the charges Qx one obtains



xQx exp MQ2x Qx
Qx {0,1}





= 1 + exp (M ) x + exp (M + ) x .

(4.7)

In the original Potts model (up to an overall factor A) the corresponding term is exp(hx ).
Hence, the flux model matches the Potts model if
1 
exp(h) = A,
3
Z(3)



1 
exp(h) = A exp (M + ) ,
3
Z(3)



1 
exp(h) = A exp (M ) .
3

(4.8)

Z(3)

These relations can be used to determine the parameters exp((M )) and exp((M
+ )) of the flux model in terms of the Potts model parameter h.
4.2. Metropolis algorithm for the flux model and comparison with the cluster algorithm
As first described in [22], the flux model can be updated with a simple Metropolis
algorithm. One basic move in the algorithm creates or annihilates a nearest-neighbor
charge-anti-charge pair across a given link. The other basic move creates or annihilates
an electric flux loop around an elementary plaquette. These basic moves are proposed
on every link and plaquette and are accepted or rejected in a Metropolis step. We have
implemented both the Metropolis algorithm for the flux model and the cluster algorithm
for the Potts model and we have verified that physical observables obtained with the two
algorithms agree with each other.
The question arises which of the two algorithms is more efficient. The Metropolis
algorithm is expected to suffer from critical slowing down at the endpoint of the first order
phase transition with a dynamical critical exponent z 2. Cluster algorithms are known
to drastically reduce critical slowing down, in some cases even to z 0. However, our
cluster algorithm cannot eliminate critical slowing down completely because the decision
to put a bond is more time-consuming than the one in the SwendsenWang algorithm. For
example, when a bond is removed, one must check if an old cluster decomposes into two
new clusters. To minimize the computational effort, we simultaneously grow two clusters
from the two ends of the bond. Once they touch each other, we know that the old cluster did
not decay. Still, in the less likely event that the old cluster does decay, one must completely
grow the smaller of the two clusters in order to decide if the bond can be deleted. Here we
do not attempt to determine the dynamical critical exponent z of our cluster algorithm at
the critical endpoint.

M. Alford et al. / Nuclear Physics B 602 (2001) 6186

79

An even more severe super-critical slowing down is expected close to the first order
phase transition line. Along that line, the deconfined phase coexists with the confined
phase, and the Monte Carlo simulation must tunnel between the two phases. In order to
tunnel between the confined and the deconfined phase, a local algorithm must go through
configurations containing both phases simultaneously. Since the interface that separates
the two phases has non-zero interface tension, such configurations are exponentially
suppressed. Hence, close to a first order phase transition, a local algorithm like the
Metropolis algorithm necessarily suffers from exponential slowing down. This supercritical slowing down is even more severe than the power-law critical slowing down at
a second order phase transition. Hence, we expect that the Metropolis algorithm for the
flux model is not very well suited to study the first order phase transition line. In some
cases, cluster algorithms can even eliminate super-critical slowing down. For example, in
the broken phase of the Potts model three distinct deconfined phases coexist with each
other. The SwendsenWang algorithm can efficiently tunnel from one deconfined phase
to another because it assigns the same random Z(3) element to all spins in a cluster in
a non-local spin update. Still, even the SwendsenWang algorithm suffers from supercritical slowing down at the first order phase transition that separates the confined from the
deconfined phase. Although cluster flips can naturally lead to tunneling between distinct
deconfined phases, they do not lead directly from a deconfined to the confined phase. To
cure this problem, Rummukainen [26] has combined the cluster algorithm with the multicanonical methods of Berg and Neuhaus [27] which can reduce the exponential supercritical slowing down to a power-law behavior. Although this may well be possible, we
have not yet attempted to combine our algorithm with multi-canonical methods. Hence,
we expect that our cluster algorithm still suffers from super-critical slowing down close to
the deconfinement phase transition.
We have compared the efficiency of the Metropolis algorithm for the flux model, the
cluster algorithm for the complex action Potts model, and the reweighted Metropolis
algorithm for the complex action Potts model at several points in the phase diagram.
In Fig. 5 we compare the computer time histories of the Polyakov loop for the three
algorithms mentioned above at (h, ) = (0.01, 0.5) which is in the confined region of
the phase diagram. Obviously, the flux model Metropolis algorithm decorrelates a lot
worse than the other two algorithms. The flux algorithm performs even worse when one
approaches the first order transition line. The reweighted Metropolis algorithm for the Potts
model suffers from the complex action problem. What is plotted is the time evolution

of Re[ x x exp(ih x Im x )]/V exp(ih x Im x )R . Its statistical fluctuations are


much larger than the expectation value which is a manifestation of the complex action
problem.

5. Universality class of the critical endpoint


In this section we present the results of our numerical simulations at the critical endpoint.
The complex action problem turned out to be very weak in the vicinity of the critical

80

M. Alford et al. / Nuclear Physics B 602 (2001) 6186

Fig. 5. The computer time history of the Polyakov loop (a) for the cluster algorithm for the
Potts model using the improved estimator for , (b) for the Metropolis algorithm for the Potts
model using directly, and (c) for the Metropolis algorithm for the flux model on a 203 lattice
at (h, ) = (0.01, 0.5). The horizontal straight line denotes the expectation value. The Metropolis
algorithm for the flux model has a much longer autocorrelation time than the other two algorithms.
In the case of the Metropolis algorithm for the Potts model the complex action problem manifests
itself by large fluctuations around the expectation value.

M. Alford et al. / Nuclear Physics B 602 (2001) 6186

81

endpoint (see Section 3.4). Therefore it is possible to use a simple reweighted Metropolis
algorithm even though that does not solve the complex action problem. It is usable up to
sufficiently large lattices so that critical exponents can be extracted from a finite size scaling
analysis. Of course, on even larger lattices the meron-cluster algorithm will eventually be
superior to the reweighted Metropolis algorithm. But since at the endpoint E the complex
action problem sets in only at volumes  1003 , simulations at E are not limited by the
complex action problem but by the ability to simulate large lattices on todays computers.
Fig. 1 shows the phase diagram of the model defined by Eq. (2.11). For h = 0 our
model reduces to the standard 3-d 3-state Potts model which has been studied extensively
in Monte Carlo simulations [28]. The model is known to have a weak first order phase
transition. The value of the coupling where the phase transition occurs (point T in Fig. 1)
has been determined with high precision. In [28], the phase transition was found to occur at
T = 0.550565(10). Above this value the Z(3) symmetry is spontaneously broken, i.e., for
> T three distinct deconfined phases coexist. When we switch on the parameter h, the
Z(3) symmetry gets explicitly broken. Positive values of h favor the deconfined phase with
a real value of . Hence, the line > T at h = 0 is a line of first order phase transitions
which cannot terminate in the deconfinement transition at the point T . In fact, T is a triple
point because two other first order transition lines emerge from it. For h > 0 a line of
first order transitions extends into the (h, )-plane and terminates in a critical endpoint
(E in Fig. 1). Negative values of h favor the two deconfined phases with complex values
of . Negative h are unphysical in the QCD interpretation of the Potts model because h
represents exp((M )) in QCD. Still, the Potts model at h < 0 makes perfect sense as
a statistical mechanics system (unrelated to QCD) and it has another first order transition
line emerging from the point T . Interestingly, with our method the complex action problem
can only be solved for h  0 since otherwise the improved estimator of Eq. (3.8) is not
necessarily positive. It should be noted that for h < 0 also the flux model suffers from the
complex action problem.
The line of first order phase transitions t (h) is determined by the condition that the
free energy densities of the confined and deconfined phases are equal, i.e., fc (h, t (h)) =
fd (h, t (h)). Close to the point T = (0, T ) the free energy density of the confined phase
is given by
fc (h, ) = fc,T + ec,T ( T ),

(5.1)

where fc,T = fc (0, T ) and ec,T = dfc /d(0, T ) is the energy density of the confined
phase at the point T . Note that to leading order fc (h, ) is independent of h because
 = 0 in the confined phase at h = 0. On the other hand, for the deconfined phase one
obtains
fd (h, ) = fd,T + ed,T ( T ) hT ,

(5.2)

where T is the value of the Polyakov loop at the point T in the deconfined phase that is
favored at h > 0. Using the condition fc,T = fd,T for the deconfinement phase transition
at h = 0, one finds

82

M. Alford et al. / Nuclear Physics B 602 (2001) 6186

t (h) = T

T
h
h = T .
ec,T ed,T
r

(5.3)

The Monte Carlo data of [28] and [29] imply r = 0.41(1). Our data are consistent with
the first order transition line being a straight line. Fitting the values of t (h) obtained from
the infinite volume extrapolation described below yields r = 0.430(6) in reasonable agreement with the number from above. Similarly, one can determine the angle at which the
third transition line leaves the point T in the direction of negative h. A similar argument
can be applied to the Potts model with real action that was studied in [23]. Also in that case
the first order phase transition line is consistent with a straight line and again the predicted
position for the transition line agrees with the numerical data. Interestingly, for the Potts
model with both real and complex action, information at h = 0 is sufficient to predict the
position of the transition line for h > 0. This is because the transition at h = 0 is rather
weak and the line ends already at small values of h. If the transition would extend deep
into the (h, )-plane one would expect deviations from a straight line that would be hard
to predict based on data at h = 0.
To determine the location of the transition line numerically we perform for given values
of h and the volume V simulations at 3 to 5 different values of . These simulations are
then combined with Ferrenberg Swendsen multi-histogram reweighting [30]. To estimate
the position of the transition line we use the specific heat
2 
 2 
 


1
S[] h
S[] h
x
x
C=
(5.4)
V
x
x
and determine the position of its maximum t (h, V ) for a given h and V . The transition
point t (h) is determined in the infinite volume limit using
A(h)
(5.5)
V
where t (h) = t (h, V = ). This ansatz is used successfully in the whole h-range, i.e.,
for the first order region as well as in the crossover region. The values for t (h) are plotted
with errorbars into the phase diagram (Fig. 1). On the first order transition line they are
consistent with a straight line which intersects the axis exactly at the point T . The
crossover line has a slight curvature.
After we have determined the tangent to the first order transition line close to the
endpoint we still have to find the exact location of the endpoint on that line. Also we want
to extract critical exponents. For that purpose we consider the Polyakov loop susceptibility
2
2 
 


1
x
x

=
(5.6)
V
x
x
t (h, V ) = t (h) +

along the transition line, i.e., at the points (t (h), h). One could consider a rotated
susceptibility instead, where an admixture of the kinetic energy term is added to the
Polyakov loop to diagonalize the fluctuation matrix. Nevertheless, this is not necessary,
since the magnetic field direction is the dominant one. We explicitly checked that the
results do not depend on the admixture we chose, unless one comes close to the linear

M. Alford et al. / Nuclear Physics B 602 (2001) 6186

83

combination where the discontinuities of the Polyakov loop and the kinetic energy cancel.
This linear combination would correspond to the kinetic energy term in the Ising model.
Of course, if one would be interested in observables related to the kinetic energy of the
Ising model, one would have to exactly choose the linear combination that corresponds to
the kinetic energy direction.
Close to the critical point the following scaling ansatz describes the susceptibility (see,
e.g., [31])
= L / f (x),

x = (h hc )L1/ .

(5.7)

For the fit, the function f (x) is expanded in a Taylor series around x = 0 up to second
order, f (x) = f0 + f1 x + f2 x 2 /2. We perform two fits:
Six parameter fit: It results in 1/ = 1.532(57), / = 2.064(70), hc = 0.000445(18),
f0 = 0.70(16), f1 = 3.99(95), and f2 = 16.4(60) with 2 /d.o.f. = 1.33. The results for the exponents agree with the estimates for the 3d-Ising universality class
1/ = 1.587(2) and / = 1.963(3) [31] almost within errorbars.
Four parameter fit: The critical exponents are fixed to the Ising model values. The
result is hc = 0.000470(2), f0 = 0.9775(46), f1 = 4.567(55), and f2 = 16.5(17)
with 2 /d.o.f. = 1.35. The good value for 2 supports again the universality class of
the 3-d Ising model.
The susceptibility and the four parameter fit are shown in Fig. 6. Fig. 7 shows the
function f (x).

Fig. 6. The susceptibility along the transition line plotted as a function of h for five different
volumes.

84

M. Alford et al. / Nuclear Physics B 602 (2001) 6186

Fig. 7. The scaling function f (x).

6. Conclusions
We have used a cluster algorithm to solve the notorious complex action problem in the
Potts model approximation to QCD with heavy quarks at large chemical potential. We use
a simple analytically constructed improved estimator that gives an exponential reduction
in the required statistics. Since the improved estimator is real and positive, importance
sampling techniques that fail for complex actions then become applicable. This makes it
possible to study the whole h > 0 parameter range of the Potts model, not just the h = 0
axis. (Recall that h corresponds to exp(( M)) in QCD in the limit M, at any
given M).
We compared our cluster algorithm with a flux model reformulation, and with the
reweighted Monte Carlo algorithm. We found that the cluster algorithm was more efficient
than using the flux model reformulation. In the large volume limit the cluster algorithm will
always be superior to Monte Carlo reweighting. However, at very small h it is sufficient to
use reweighting techniques to obtain physically relevant results. This turned out to be the
case for the endpoint of the first order line, which occurs at a very small h because the 3-d
3-state Potts model phase transition is rather weak at h = 0. We therefore used reweighted
Monte Carlo to locate the first-order line and its endpoint. However, we emphasize that as
computer power rises, and the maximum attainable volume becomes bigger, the meroncluster algorithm will eventually become superior at any h > 0.
We also calculated quarkquark, quarkantiquark, and antiquarkantiquark potentials,

M. Alford et al. / Nuclear Physics B 602 (2001) 6186

85

in the confined and deconfined regions of the phase diagram. We found the expected
behavior: the background density of heavy quarks screens color fields, so that all potentials
reach plateaux at long distances, whose values are simply related to the free energies of
external static quarks and antiquarks.
The algorithm that we have developed for the Potts model belongs to the class of meroncluster algorithms that has recently been used to solve a large variety of sign and complex
action problems. Of course, the ultimate goal is to construct a similar algorithm for QCD
at non-zero chemical potential and investigate the phase structure of QCD at = 0 from
first principles. The complex action problem in full QCD is more complicated than the one
in the Potts model. So far, meron-cluster algorithms have led to solutions of fermion sign
problems as well as complex action problems in bosonic theories, but have not yet solved
complex action problems in theories with fermions. We believe that this may ultimately
become possible when one uses the D-theory formulation of QCD.

Acknowledgements
We like to thank the INT in Seattle, where this work was initiated, for its hospitality.
U.-J.W. thanks F. Karsch and S. Stickan for helpful discussions and acknowledges the
support of the A.P. Sloan foundation.

References
[1] I.M. Barbour, S.E. Morrison, E.G. Klepfish, J.B. Kogut, M. Lombardo, Nucl. Phys. Proc. Suppl.
A 60 (1998) 220.
[2] M. Alford, A. Kapustin, F. Wilczek, Phys. Rev. D 59 (1999) 054502.
[3] W. Bietenholz, A. Pochinsky, U.-J. Wiese, Phys. Rev. Lett. 75 (1995) 4524.
[4] S. Chandrasekharan, U.-J. Wiese, Nucl. Phys. B 492 (1997) 455.
[5] B.B. Beard, R.C. Brower, S. Chandrasekharan, D. Chen, A. Tsapalis, U.-J. Wiese, Nucl. Phys.
Proc. Suppl. 63 (1998) 775.
[6] U.-J. Wiese, Nucl. Phys. Proc. Suppl. 73 (1999) 146.
[7] R. Brower, S. Chandrasekharan, U.-J. Wiese, Phys. Rev. D 60 (1999) 094502.
[8] J. Cox, C. Gattringer, K. Holland, B. Scarlet, U.-J. Wiese, Nucl. Phys. Proc. Suppl. 83 (2000)
777.
[9] S. Chandrasekharan, U.-J. Wiese, Phys. Rev. Lett. 83 (1999) 3116.
[10] S. Chandrasekharan, Nucl. Phys. Proc. Suppl. 83 (2000) 774.
[11] S. Chandrasekharan, J. Cox, K. Holland, U.-J. Wiese, Nucl. Phys. B 576 (2000) 481.
[12] J. Cox, K. Holland, Nucl. Phys. B 583 (2000) 331.
[13] S. Chandrasekharan, J.C. Osborn, Phys. Lett. B 496 (2000) 122.
[14] J.C. Osborn, hep-lat/0010097.
[15] S. Chandrasekharan, B. Scarlet, U.-J. Wiese, cond-mat/9909451.
[16] S. Chandrasekharan, J.C. Osborn, Springer Proc. Phys. 86 (2000) 28.
[17] S. Chandrasekharan, hep-lat/0011022.
[18] T.C. Blum, J.E. Hetrick, D. Toussaint, Phys. Rev. Lett. 76 (1996) 1019.
[19] J. Engels, O. Kaczmarek, F. Karsch, E. Laermann, Nucl. Phys. B 558 (1999) 307;
J. Engels, O. Kaczmarek, F. Karsch, E. Laermann, Nucl. Phys. Proc. Suppl. 83 (2000) 369.

86

[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]

M. Alford et al. / Nuclear Physics B 602 (2001) 6186

P. Hasenfratz, F. Karsch, I.O. Stamatescu, Phys. Lett. B 133 (1983) 221.


T.A. DeGrand, C.E. DeTar, Nucl. Phys. B 225 (1983) 590.
J. Condella, C.E. DeTar, Phys. Rev. D 61 (2000) 074023.
F. Karsch, S. Stickan, Phys. Lett. B 488 (2000) 319.
E. Hilf, L. Polley, Phys. Lett. B 131 (1983) 412.
R.H. Swendsen, J. Wang, Phys. Rev. Lett. 58 (1987) 86.
K. Rummukainen, Nucl. Phys. B 390 (1993) 621.
B.A. Berg, T. Neuhaus, Phys. Lett. B 267 (1991) 249;
B.A. Berg, T. Neuhaus, Phys. Rev. Lett. 68 (1992) 9.
W. Janke, R. Villanova, Nucl. Phys. B 489 (1997) 679.
R.V. Gavai, F. Karsch, B. Petersson, Nucl. Phys. B 322 (1989) 738.
A.M. Ferrenberg, R.H. Swendsen, Phys. Rev. Lett. 61 (1988) 2635;
A.M. Ferrenberg, R.H. Swendsen, Phys. Rev. Lett. 63 (1989) 1195.
H.W.J. Blte, E. Luijten, J.R. Heringa, J. Phys. A: Math. Gen. 28 (1995) 6289.

Nuclear Physics B 602 (2001) 87108


www.elsevier.nl/locate/npe

QCD isospin breaking in meson masses, decay


constants and quark mass ratios
G. Amoros a,b , J. Bijnens a , P. Talavera c
a Department of Theoretical Physics 2, Lund University, Slvegatan 14A,

S22362 Lund, Sweden


b Departament de Fsica Terica, IFIC, Universitat de Valncia-CSIC, Apt. Correus 22085,

E-46071 Valncia, Spain


c Groupe de Physique Thorique, Institut de Physique Nuclaire, Universit Paris Sud,

F-91406 Orsay Cedex, France


Received 15 January 2001; accepted 9 March 2001

Abstract
The procedure to calculate masses and matrix-elements in the presence of mixing of the basis states
is explained in detail. We then apply this procedure to the two-loop calculation in Chiral Perturbation
Theory of pseudoscalar masses and decay constants including quark mass isospin breaking. These
results are used to update our analysis of K4 done previously and obtain a value of mu /md in
addition to values for the low-energy-constants Lri . 2001 Published by Elsevier Science B.V.
PACS: 12.15.Ff; 13.20.Eb; 11.30.Rd; 12.39.Fe

1. Introduction
Experiments in low-energy QCD are becoming more and more precise. It is therefore
needed that theoretical calculations are performed to similar accuracy. In the purely
mesonic sector the low-energy effective theory of QCD is known as an effective theory
in terms of pseudoscalar mesons only. It is known as Chiral Perturbation Theory (CHPT)
and was put on a solid theoretical basis in [1]. In the two-flavour sector calculations to
two-loop order are now customary. In our earlier work [2,3], where also references to other
calculations can be found, we have performed the main calculations in the three flavour
sector also to two loops in the isospin symmetric approximation.
One of the remaining uncertainties was the importance of isospin breaking effects.
Basically in our previous work and many other theoretical calculations isospin breaking
was only included by guessing its effect and putting it in the uncertainty of the final result.
E-mail address: bijnens@thep.lu.se (J. Bijnens).
0550-3213/01/$ see front matter 2001 Published by Elsevier Science B.V.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 2 1 - 3

88

G. Amoros et al. / Nuclear Physics B 602 (2001) 87108

An example where we definitely need to go beyond this is in the use of K3 form-factors [4]
to extract an accurate value of the |Vus | CKM matrix-element. Here we take a first step in
that we evaluate within CHPT the strong isospin breaking in masses and decay constants
to next-to-next-to-leading order.
This work serves two purposes: it checks the dependence on isospin breaking of the
determination of the CHPT low-energy constants (LECs) at O(p4 ) performed in our earlier
work [2,3] and allows us to extract information on the quark mass ratio mu /md . The latter
is one of the fundamental input parameters in QCD and thus needs to be determined as
accurately as possible.
Another motivation is to see how our previous results [2,3] can change with the
new experimental data on K4 [5]. In particular, they contain one more independent
measurement, the slope of the G form-factor, allowing for a cross-check on the CHPT
description and we will use it to have a first check on the large Nc assumptions made in
the theoretical analysis.
The plan of the paper is the following: We describe in Section 2 the changes needed in
the presence of mixing of the external states. We first discuss it for the masses and second
for matrix-elements. We then sketch all the computed quantities, Section 3, and follow
it with a discussion of the electromagnetic corrections in Section 4. The latter remain
a sizable source of uncertainty in our results. In Section 5 we list all the assumptions
on which the calculation relies. Section 6 introduces one of the main inputs, the K4 formfactors, discussing the main differences in the data and parametrization used in this analysis
w.r.t. the previous one. Section 7 presents a brief summary of our fitting procedure with all
the inputs and output variables. Section 8 is devoted to the update of the results presented
already in [2,3]. In Section 9 we discuss the impact of the outputs on the quark mass ratios.
And finally we briefly summarize our findings in Section 10.
The formulas are of such a length that they cannot be presented in a manuscript of
reasonable length and we have therefore not included them. For an introduction to CHPT
we refer to [6].

2. Formalism
In the presence of mixing, where each external field can couple to more than one-particle
state, we have to generalize the discussion of masses and amplitudes given in [2] somewhat.
We first discuss how the masses can be obtained from a general two-point function
and later the generalization of the wave-function-renormalization part for matrix-element
calculations. The notation will be chosen appropriate for the 0 mixing case, where the
inverse of a two-by-two matrix can be written explicitly in a simple manner.
2.1. Masses
In terms of the lowest order fields i we define the two-point function





Gij p2 , m2i0 , F0 = d d x eipx 0|T i (x)j (0) |0.

(1)

G. Amoros et al. / Nuclear Physics B 602 (2001) 87108

89

Fig. 1. The diagrams contributing to the two-point function Gij . The filled circles correspond to
one-particle-irreducible diagrams and the lines to lowest order propagators.

We have suppressed the dependence on everything except p2 , F0 and the lowest order
masses mi0 . The diagrams contributing to this are depicted in Fig. 1. Writing the lowest
order propagator as iPij and the one-particle-irreducible contributions as ikl , pictorially
ikl = k  l , we can obtain the full two-point function in matrix notation as
G = iP + iP (iiP ) + iP (iiP )2 + = iP (1 + P )1 .

(2)

In the case of no mixing, , P and G are all diagonal matrices and the solution is simple.
The masses of the particles correspond to the poles which in the diagonal case are given
by the zeros of p2 m20 as described in [2]. In the general case they are again given
by the poles in G as a matrix. These can be most easily found as places where the inverse
of G has a zero eigenvalue or the determinant vanishes






det G1 = det (1 + P )P 1 = det P 1 + = 0.
(3)
It is this equation we now like to solve perturbatively in our case of 0 mixing. The
isospin eigenstates used in the Lagrangian are denoted by 30 and 8 . To lowest order, we
can diagonalize G and thus by definition P by a simple rotation
0 cos( )30 + sin( )8 ,
sin( )30 + cos( )8 ,

(4)

where the lowest order mixing angle satisfies

3 md mu
tan(2 ) =
(5)
,
2 ms m

with m
 = (mu + md )/2, the average of the lightest quark masses. In the basis defined by
0
, , the Kaon and charged pion fields, we have
Pij =

ij
p2 m2i0

(6)

and starts by definition only at next-to-leading order, O(p4 ), but does not need to be
diagonal. The lowest order mixing angle does not need to be small, we will keep it to all
orders throughout our calculations.
With the previous prescriptions P is diagonal and is blockdiagonal enforced by the
various other conserved quantities like, for instance, parity. Then each block can be treated
separately. The charged pion and Kaon do obey a simple diagonal equation without mixing,
therefore only the 0 block needs special consideration.

90

G. Amoros et al. / Nuclear Physics B 602 (2001) 87108

In order to obtain the chiral expansion of the masses, we need to find the zeros of the
determinant in Eq. (3) as a function of
 (4)  2 (6)
p2 = m2phys = m20 + m2
(7)
+ m
+ ,
where the superindices refer to the chiral order. In the same way we expand the oneparticle-irreducible self interaction, (m2phys, m2i0 , F0 ), in chiral orders






m2phys, m2i0 , F0 = (4) m2phys, m2i0 , F0 + (6) m2phys, m2i0 , F0 + .
(8)
Using (3) and (8) as labels for the 0 and fields defined in Eq. (4) and temporarily
suppressing the functional dependence in the equation becomes to O(p6 )
 2
(4)
(6) 
(4)
(6) 
mphys m230 + 33 + 33 m2phys m280 + 88 + 88
 (4)
(6) 2
38 + 38 = 0,
(9)
where the matrix is symmetric. We insert at this point the expansion of m2phys and then
either the lowest order mass is m230 or m280 . For definiteness hereafter only the pion mass
case will be considered. Thus we obtain
m20 = m230 ,
 2 (4)

(4) 
m
= 33 m230 , m2i0 , F0 ,

 2 (6)
  2 (4)

(6)  2
(4)  2
2
2
m
= 33 m30 , mi0 , F0 m

p , mi0 , F0 
p2 33
p 2 =m2
 (4)  2
2
1
2

m
+ 2
,
m
,
F
.
0
30
i0
38
m30 m280

30

(10)

Notice that we have chosen here to use the strict expansion where we always express the
arguments of the selfenergy in lowest order quantities. For numerical results later we will
express all masses in terms of their physical masses and F0 in terms of F and reorder
the series accordingly.
We have only used the fact here that a lowest order mixing angle is well-defined. There
is of course no reason for this to be true at higher orders and indeed it is not the case already
at order p4 [1]. When photonic corrections are included, the possibility of mixing of states
with additional soft photons of course introduces the usual infrared problems and provides
another reason for not having a simple mixing angle for the physical states [7].
Another alternative is to make a second, p2 -dependent, field redefinition such that Pij1 +
(4)

ij (p2 , m2i0 , F0 ) (i = j ) vanishes. In principle in terms of those fields the expressions for
the matrix-elements derived below are somewhat simpler. The additional complications in
calculating the diagrams is the reason we work with fields that are only diagonalized to
lowest order.
2.2. Matrix-elements
To calculate any observable we follow the usual LSZ reduction formalism. The matrixelement in momentum space for any n incoming or outgoing states is

G. Amoros et al. / Nuclear Physics B 602 (2001) 87108


Ai1 ...in =

(i)n

Zi1 . . . Zin



lim ki2 m2i Gi1 ...in (k1 , . . . , kn ).

2
2
i=1 ki mi

91

(11)

The function Gi1 ...in (k1 , . . . , kn ) is the full n-point Green function generated by the n fields
i1 (k1 ), . . . , in (kn ). The coefficients Zi are defined via


Gii p2 m2iphys =

p2

Zi
+ O(1)
m2iphys

(12)

and are often referred to as wave-function-renormalization. In the diagonal case this is


the procedure described in detail in [8]. In the case where mixing is present, the formulas
become more cumbersome. We will for simplicity discuss only the case of one external
leg but the generalization should be obvious. In terms of the Green-function Gj which is
one-particle-irreducible and amputated on the j -leg which can connect to the i external
state we then get

  
i
lim ki2 m2i Gij ki2 Gj
Ai =
Zi ki2 m2i

(13)

with an implicit summation over j .


The quantity, simply calculable in terms of , was G1 = iP, P = P 1 + . Now
for a two-by-two matrix the inverse is fairly easy


 2
i
P88 (p2 ) P38 (p2 )
G=G p =
(14)
,
2 (p2 ) P (p2 )
P33 (p2 )
P33 (p2 )P88 (p2 ) P38
38
where we have indicated the p2 dependence explicitly. At the physical mass det P(p2 )
vanishes and thus the pole in Eq. (12) appears naturally. For the and case we obtain
Z3 =
Z8 =

 
P88 m2 ,

 
P33 m2 .

(det P(p2 ))p2 =m2


p 2

(det P(p2 ))p2 =m2


p 2

(15)

The final normalized amplitude for the pion case is then


A3 =

1
P88 p 2 (P33 P88

2 )
P38

{P88 G3 P38 G8 },

(16)

all functions are taken at the pion mass. Gj refers to the l.h.s. of Fig. 2, the dark (red)
bubble, which contains all the one-particle-irreducible in leg j diagrams.
The amplitude in Eq. (16) contains formally all the chiral orders. Obviously for practical
purpose one has to restrict to a given order in the chiral series. Therefore we expand P and
Gi inside Eq. (16) up to O(p6 ), but we do not expand these functions in .
In our basis we have that (P 1 )ij = ij (p2 m2i0 ) and only starts at O(p4 ) which
allows to evaluate the expansions explicitly in terms of
Gj = Gj(2) + Gj(4) + Gj(6) ,

92

G. Amoros et al. / Nuclear Physics B 602 (2001) 87108

Fig. 2. A general contribution to Gi1 i2 ...in . We choose one of the legs, il = i, and show how the full
two-point function Gij contributes on that leg, denoted by the filled circle on the right-hand side.
The remainder of the diagrams, the Green-function, reduced on the ih leg is depicted by the oval on
the left-hand side and denoted in the text by Gi1 i2 ...in with ih = j .

ij = ij(4) + ij(6) ,
Zij

(2)
(4)
(6)
ij = Zij + Zij + Zij .
p2

(17)

Using the fact that det P = 0 to simplify further the expression, one explicitly obtains for
the pion case


(4)
1 (4) (2) 38 (2)
(2)
(4)
A3 = G3 + G3 Z33 G3
G
2
&m2 8

(4) (4)
1 (6) (2) 1 (4) (4) 3  (4) 2 (2) Z38 38 (2)
(6)
+ G3 Z33 G3 Z33 G3 + Z33 G3 +
G3
2
2
8
&m2
 (4)

(4)
(6)
1 38 2 (2) 38 (4) 38 (2)

G
3
2 &m2
&m2 8
&m2 8

(4) (4)
(4)

1 (4) 38 (2)
,
+ 38 288 G8(2) + Z33
G
(18)
&m
2
&m2 8
where the terms displayed between curly brackets are of O(p4 ) and the ones between
squared brackets are O(p6 ). The formulas are at ki2 = m2 , the physical pion mass and
&m2 (P 1 )88 (m2 phys) = m2 phys m20 . We can then rewrite the dependence on the
lowest order masses and decay constants as before as well in the physical masses and
decay constants.
We checked our results in an analytical and numerical way: in terms of the general
functions sin( ) and cos( ) we find a scale-independent result, while setting = 0 we
recover the results of [2].

3. Analytical results
In [2] we presented the masses and decay constants and in [3] in addition the vacuum
expectation values and the K4 form-factors at next-to-next-to-leading order. Even in the
latter case we had to resort to numerical approximations for a large part of the expressions
involving integrals in order to keep formulas to a reasonable size. In the present case, due

G. Amoros et al. / Nuclear Physics B 602 (2001) 87108

93

to the presence of isospin breaking effects even the expressions for the masses become very
cumbersome. In this section we therefore only define the quantities we have calculated and
how the various chiral orders have been split.
We have followed as much as possible the philosophy of rewriting the quark masses in
terms of physical meson masses at O(p4 ), with the exception of mu md and the lowest
order mixing angle . We have however used Eq. (5) as much as possible to reduce the
number of terms. In addition the lowest order mass relations are extensively used at O(p6 ).
The quantities calculated up to O(p6 ) including isospin breaking due to the quark
masses to all orders are
The 5 meson masses, m2 , m2 0 , m2 , m2K and m2K 0 .
The decay constants F , FK and FK 0 .
The 4-decay constants in the neutral sector. These are the couplings of the physical 0
and states to the triplet and octet axial current, labeled as F 0 3 , F 0 8 , F3 and F8 .
In addition we define the quantities = F 0 8 /F 0 3 and = F3 /F8 .
and s s.
The three vacuum expectation values uu,

dd

4. Treatment of electromagnetic corrections


Isospin breaking has two different sources, the quark mass difference mu md and the
electromagnetic interaction. The former breaking introduces only one new parameter as
compared to the isospin symmetric analysis of [2], mu /md . The latter can in principle
also be treated within the framework of CHPT. The corrections to the masses were treated
in [9] and the framework needed for the decay constants and K4 was worked out in [10].
Unfortunately, the general treatment brings in a large number of new unknown constants.
Estimating these brings into the game the evaluation of nonleptonic matrix-elements,
a problem that is beyond our present scope.
The lowest order is O(e2 p0 ) and brings in only one more parameter which can be fixed
by the difference m2 m2 0 . The quark mass contribution to this mass difference is very
small and can be neglected as a good first approximation. This leads to the prediction
 2

mK m2K 0 em = m2 m2 0 ,
(19)
a result often referred to as Dashens theorem [11]. At this order there are no corrections
to the decay constants.
At the next chiral order, O(e2 p2 ), there are corrections to Eq. (19). The chiral logarithms
are of the expected size [9,12]. The earlier indications of a large contribution also from the
O(e2 p2 ) LECs can be found in [13,14]. These results, but in a less pronounced manner,
were confirmed by the lattice calculations of [15]. In what follows we will use the results
of [16] which are compatible with the averaged value of [13,14] since it incorporates both a
correct long-short-distance-matching 1 and the chiral logarithms at leading and nonleading
order in 1/Nc . It also agrees with the results of [15]. The final result of [16] was
1 It is precisely this matching that invalidates the results of [17].

94

G. Amoros et al. / Nuclear Physics B 602 (2001) 87108

m2K m2K 0


em



= (1.84 0.25) m2 m2 0 .

(20)

We will use the central value of Eq. (20) in our numerical results and Eq. (19) to show the
dependence of the result on this quantity. Numerically the above two correspond to


mK mK 0 em 2.3 MeV [Eq. (20)] and 1.3 MeV [Eq. (19)].
(21)
We calculate the masses and decay constants with the strong meson masses, i.e., with the
electromagnetic effects above removed.
The electromagnetic corrections to the decay constants are only partially treated within
CHPT. We will use the values as calculated by [18,19] as F = 92.4 MeV and FK /F
where the electromagnetic radiative corrections have been removed.
The experimental results of [20] were already corrected for radiative corrections. In the
absence of better calculations and measurements we therefore use our earlier procedure [3]
again. We calculate this process with the physical mass of the charged Kaon and pion.

5. Other assumptions
The calculations in CHPT we have performed are all to O(p6 ). We have not used any
improvement procedure to guess the size of the higher order terms nor can we determine
experimentally the O(p6 ) tree terms. At present the data are simply not constraining
enough to allow us to do so. The total number of parameters is very large. Thus these
have to be estimated using a theoretical procedure as discussed in Section 5.1. In addition,
the available data and O(p6 ) calculations are not sufficient to determine all other inputs,
the assumptions we use here are given in Section 5.2.
5.1. At next-to-next-to-leading order, O(p6 )
When the full fits of the O(p4 ) LECs became available [1] an understanding of
their general pattern and sizes within about 30% could be obtained using resonance
saturation [21]. These results were later refined in the framework of quark-models and
large Nc [22].
We will use the former procedure, a few tests at O(p6 ) exist, in two-point-functions
some parameters could be determined [23] and in the pion form-factors two additional tests
were possible [24]. All of these were successful within about 30%. The precise estimates
we use are described in [2,3]. We test this assumption by varying the input in several ways:
1. Keeping only the two main contributions: vectors and  .
2. Varying the input parameters of the resonances by a factor of two.
3. Choosing different renormalization points where we do the estimates.
The combination of the above 3 variations results in a rather large variation of the O(p6 )
constants used. We thus expect that this part does not affect our result within the quoted
errors.

G. Amoros et al. / Nuclear Physics B 602 (2001) 87108

95

5.2. At next-to-leading order, O(p4 )


Furthermore, using the data presented in [20] we need to fix four more inputs, we choose
the quark mass ratio ms /
m, and three of the LECs: Lr4 , Lr6 and Lr9 . The new free parameter
compared to [3] is mu /md but now there are two Kaon masses instead of one as fitting
parameters so mu /md can be determined as a function of the inputs.
For the quark mass ratio ms /
m we use as standard value 24. This is a reasonable average
of the ratio of the sum-rule calculations of m
 [25] and ms [26]. We will also check the
changes of the fits with a rather broad range of inputs for ms /
m [20, 30].
The constant Lr9 driving the pion charge radius is saturated up to some good extent by
a Vector Dominance expression. The pole due to the exchange represents the dominant
low-energy singularity while the and K K cuts generate a tiny correction. This LEC is
fixed at its standard value of 6.9 103 . Even putting it to zero hardly changes the results
so this assumption does not bring extra uncertainty.
The largest uncertainty w.r.t. the masses and decay constants is in the values of Lr4
and Lr6 . In the large Nc limit they vanish and varying the scale from 500 MeV to 1 GeV
changes Lr4 by about 0.5 103 and Lr6 by about 0.3 103 . A range within zero
of that size is thus an acceptable range. There are some indications that the values are
at the end of this region and even somewhat outside for Lr6 [27,28]. We will make use
of two experimental data analysis: in the first one we will use [20] (otherwise using [29]
barely will change the conclusions) and will keep both of them, Lr4 and Lr6 at zero for
our main fit but discussing the variation of the results below. In second place we will
use the analysis of [5], this will allow to remove one of the assumptions for the large
Nc behaviour of Lr4 or Lr6 and compare the result with the previous one. Let us note
here that there are possibly large corrections to the large Nc limit in the scalar sector
but they are to a large extent due to the chiral loop corrections and are thus included in
the present results. Examples are scattering [8] and the pion scalar form-factor [24].
A large Nc saturation of the constants in the latter case provided excellent agreement with
experiment. This question does however deserve further study. But if we assume that the
SU(2) quantities should not be too sensitive to ms then we need to assume that Lr4 and Lr6
are small. We also obtained [3] that the 1/Nc suppressed combination (2Lr1 Lr2 )/Lr3 was
small providing one more test of the large Nc assumption. Let us comment briefly about
the resonance saturation pattern of the LEC. The symmetry breaking part of the effective
Lagrangian at O(p4 ) is determined by Lr4 , . . . , Lr8 . Something in common to all of these
constants is the fact that in the low-energy regime none is affected by the exchange of spin
1 resonances but of scalars and pseudoscalar particles, sector in which the experimental
data are still fuzzy. Further more, for Lr4 , . . . , Lr8 the natural scale (MS  M  1 GeV)
differs from the one of Lr1 , Lr2 and Lr3 , set in the last ones by Vector exchange (MV  M 
0.8 GeV). This obviously is reflected in the expansion in powers of ms , some quantities
are controlled by the parameter (MK /MS )2 1/9 while others present a worse behaviour
(MK /M )2 1/4.

96

G. Amoros et al. / Nuclear Physics B 602 (2001) 87108

6. K4 decay parametrization and data


The experiment E865 was not primarily designed to collect K4 data but it recorded it
as a background anyway [5]. These data were previously analyzed in a way similar to the
older experiment [20] and presented in [29]. The second and latest analysis [5] uses a more
suitable parametrization of the form-factors to fit the data at low-energy [30]
0

F = {fs (s ) + f s }ei0 (s ) + p-wave,


1

G = {gp (s ) + g s }ei1 (s ) + d-wave,


where K

+ (p+ ) (p )(p )(p )

s = (p + p )2 ,

(22)

decay variables are defined as

s = (p+ + p )2 .

(23)

The analysis of [5] assumes only dependence in the lowest partial wave decomposition,
i.e., s-wave in F and p-wave in G. Furthermore, any sensitivity on s is discarded. As
previous work showed, [3], those assumptions where valid in the analysis of [20]. But in
view of the errors quoted in [5] for the form-factor F at threshold, fs (0), this assumption
is borderline for the new data.
The form-factors in [5] where fitted to the following expressions
 0

F = fs + fs s + fs s2 ei0 (s ) ,

 1
G = gp + gp s ei1 (s ) ,
(24)
where fs , fs , fs , gp and gp do not depend on any kinematical variable leading to the result
fs (0) = 5.75 0.02 0.08,

fs (0) = 1.06 0.10 0.40,

gp (0) = 4.66 0.47 0.07,

gp (0) = 0.67 0.10 0.04.

(25)

The quadratic slope is rather larger than we expect from our CHPT results [3]. We therefore
use the results of [5] with a linear fit only
fs (0) = 5.770 0.097,

fs (0) = 0.47 0.15,

gp (0) = 4.684 0.092,

gp (0) = 0.54 0.20,

(26)

and regards these fits as preliminary till the data are finalized.

7. The fitting procedure and fit results


The fitting procedure is the same as the one used in [3] but slightly adapted to the present
case. We refer to [3] and the previous sections for a description of the details about extra
assumptions, besides CHPT, and how electromagnetic corrections were taken into account
in all the computed physical quantities. Here we only sketch briefly the inputs and outputs
of the calculations.
In addition to the lowest order Lagrangian that contains: B0 mu , B0 md , B0 ms and F0 as
free constants (see below) the O(p4 ) Lagrangian is parametrized in terms of 10 unknown

G. Amoros et al. / Nuclear Physics B 602 (2001) 87108

97

LECs not restricted by symmetry: Lr1 , . . . , Lr10 . All in all this amounts to 14 free parameters
in total. 2 The constant Lr10 does not appear to O(p6 ) in any of our quantities, thus remain
13 free parameters to be determined at O(p6 ). Taking into account the restrictions of
Section 5 the number of parameters is reduced to 10 or 9 depending if we use the results
of [20] or [5], respectively. Hereafter we will quote the quantities referring [5] between
parenthesis. The experimental inputs used are four of the five pseudoscalar masses, i.e., we
discard the use of m2 because its difference with m2 0 is purely of electromagnetic origin.
Two decay constants F , FK . And three (four) inputs from K4 decays, Section 6,
two form-factors at threshold, fs (0), gp (0), and one (two) slope fs (0), (gp (0)). In all
there are nine (ten) inputs and therefore this allows to determine nine (ten) parameters.
Schematically

F , FK ,
2 , M 2 , M 2,
Inputs M2 0 , MK
(27)

K0

fs (0), gp (0), fs (0), (gp (0)),

F0 ,

,
B0 m
Outputs
(28)

m
/m ,

u d
L1 , L2 , L3 , (L4 ), L5 , L7 , L8 .
Notice that using [5] we obtain between our outputs the order parameter Lr4 .
Unfortunately the K4 quantities are not very sensitive to Lr4 [31].
The errors used in the fit were the experimental errors on fs , fs and gp , (gp ), 0.01 for
FK /F and we required the quark mass ratios determined from the lowest order meson
masses to be satisfied within 10% accuracy.
The procedure for the latter was to take the physical meson mass, calculate the O(p4 )
and O(p6 ) corrections to obtain the lowest order masses of , 0 , K , K 0 and .
Since the lowest order masses actually were needed in the calculations, they surface in
the formulas via and &m2 , we iterated the procedure till it converged.
The change in the Lri using [20] are compared to those reported in [3] for all the fits was
significantly smaller than the errors quoted there, indicating that the neglect of isospin
violation was a good approximation. 3 The changes are always below 0.02 103 in
magnitude. We therefore still consider the main fit of [3] as the standard values for the Lri .
The remaining electromagnetic effects are estimated by also performing a fit, this time
with the masses of all the propagating particles inside the loop set to their physical mass
rather than to the strong mass only. Again, the changes in the results were rather small
compared to the quoted errors.
The fit results using [20] are presented in Table 1. This can be compared directly with
Table 2 in [3]. 4 With respect to this we have discarded the comparison with previous
2 Notice that we have one more parameter than in [3] due to the isospin breaking.
3 Note that the change in value is due to an error in the numerical program used in [3] as described in the

erratum.
4 See footnote 3.

98

103 Lr1
103 Lr2
103 Lr3
103 Lr5
103 Lr7
103 Lr8
F0 [MeV]
 [GeV2 ]
2B0 m
m [MeV]
mu /md (full)
mu /md (simple)
Changed
quantity
Unit

Main fit

O(p4 )

fit 2

fit 3

fit 4

fit 5

fit 7

fit 8

fit 9

0.53 0.25
0.71 0.27
2.72 1.12
0.91 0.15
0.32 0.15
0.62 0.20
87.1
0.0136
0.32
0.46 0.05
0.52 0.05

0.46
1.49
3.18
1.46
0.49
1.00
81.1
0.0181
0.27
0.52
0.58

0.53
0.72
2.73
0.87
0.25
0.48
87.2
0.0141
0.16
0.43
0.50

0.43
1.01
2.90
1.48
0.30
0.75
82.9
0.0158
0.07
0.47
0.54

0.63
0.72
2.72
0.72
0.30
0.56
90.9
0.0110
0.63
0.33
0.43

0.66
0.85
3.32
0.86
0.32
0.62
86.7
0.0136
0.36
0.46
0.53

ms /
m
26

0.50
0.73
2.76
0.90
0.32
0.62
86.6
0.0136
0.32
0.46
0.53

s
0.293
GeV

0.53
0.80
2.76
0.91
0.32
0.63
98.9
0.0138
0.23
0.46
0.53

O(p4 )

0.50
0.66
2.60
0.91
0.32
0.62
87.1
0.0136
0.32
0.45
0.53

s
0.1
GeV

Lr4 ; Lr6
0.3; 0.2
103

0.5
GeV

1.0
GeV

g(0)
4.93

G. Amoros et al. / Nuclear Physics B 602 (2001) 87108

Table 1
 and mu /md for the various fits described in the main text using [20]. Errors are fitting errors as quoted by MINUIT. All Lri ()
Results for Lri (), F0 , B0 m
values quoted have been brought to the scale = 0.77 GeV. The first
line with mu /md (full) is with Eq. (20) while the second line, mu /md (simple) is
m = 24, s = 0.336 GeV, s = 0, Lr4 = Lr6 = 0 and Lr9 = 6.9 103
with Dashens theorem, Eq. (19). The standard values are ms /

G. Amoros et al. / Nuclear Physics B 602 (2001) 87108

99

works [1,32] (second column in Table 2 in [3]) and fit 6, performed only with vectors and
 in the Cir saturation, otherwise we keep the same notation and only include mu /md , F0
and B0 m
 as output parameters. We also performed the fit with varying the resonance input
parameters by a factor of two with similar results.
Let us summarize our findings in Table 1: In the main fit we use a strict standard
approach, considering the central value of the Zweig Rule violating term, Lr4  Lr6 = 0,
r = 24 and the scale is set to M . The second column compares each quantity with the
corresponding O(p4 ) results using the same input assumptions. Fits 3 and 4 correspond
to different choices in the kinematical points for the K4 variables. In fit 5 we allow for
a small violation of the Zweig Rule. Fits 7 and 8 show different choices of the scale
dependence . Our results are fully -independent, but numerically the change reflects
our lack of knowledge about the scale where the resonance saturation for the LECs, Cir , is
valid. Finally fit 9 presents a weighted average of the p-wave form-factor gp (0) [20,33].
There are no changes w.r.t. the numbers in [3]. 5 Our results are therefore stable against
the isospin breaking effects. Furthermore, the values of Lr5 and Lr8 are compatible with the
recent lattice simulations [34].
In order to show the changes introduced by the analysis of [5] we also include Table 2.
Notice that the values quoted there are only indicative and should be taken as preliminary
till the final E865 analysis is available. As explained in Section 6 we use the linear fit of [5]
to perform the fit.
In Table 2 we have displayed the following results: fit 10 is the direct equivalent of the
main fit, and the next column are the corresponding O(p4 ) values. Fit 11 is the equivalent
Table 2
 and mu /md for the various fits described in the main text using the
Results for Lri (), F0 , B0 m
inputs of [5]. Lr4 and Lr6 are given, means it is an input value. All other inputs are as in Table 1

103 Lr1
103 Lr2
103 Lr3
103 Lr4
103 Lr5
103 Lr6
103 Lr7
103 Lr8
F0 [MeV]
 [GeV2 ]
2B0 m
m [MeV]
mu /md (full)
mu /md (simple)

5 See footnote 3.

fit 10

O(p4 )

fit 11

0.43 0.12
0.73 0.12
2.35 0.37
0
0.97 0.11
0
0.31 0.14
0.60 0.18
87.7
0.0135
0.28
0.45 0.05
0.52 0.05

0.38
1.59
2.91
0
1.46
0
0.49
1.00
81.1
0.0181
0.27
0.52
0.58

0.43 0.12
0.77 0.20
2.36 0.40
0.18 0.85
1.08 0.61
0
0.35 0.30
0.70 0.59
93.9
0.0104
0.83
0.37 0.12
0.45 0.27

fit 12
0.43
0.85
2.40
0.46
1.07
0.2
0.35
0.70
105.7
0.0106
0.53
0.38
0.45

fit 13
0.43
0.70
2.32
0.10
1.11
0.2
0.36
0.70
84.1
0.0104
1.12
0.38
0.47

100

G. Amoros et al. / Nuclear Physics B 602 (2001) 87108

of the main fit but leaving Lr4 as a free parameter. As one can see there are no significant
changes w.r.t. assuming Lr4 0 and we only obtain a very weak limit on Lr4 .
This is not conclusive, i.e., one can not claim that Lr6 = 0 Lr4 0 is the only possible
solution because both constants can be highly correlated. For this reason we show in fits
12 and 13 the shift due to different choice of Lr6 . As one can see from the last three
columns of Table 2 the values of Lr6 and Lr4 follow almost a linear relation. Therefore
a big deviation from the Zweig Rule value of Lr6 will probably signal also a deviation
for Lr4 . The possibility of a small value of Lr6 and a large value of Lr4 fixed via the scalar
pion radius is thus unlikely.

8. Survey of applications
In this section we collect various results in the light of the new values of the main fit in
Table 1. We want to stress that we do not study the convergence of CHPT here, we only
want to check that the total size of corrections is not very large as compared to the lowest
order. The masses typically have small O(p4 ) corrections but sizable O(p6 ) ones. As we
mentioned in [2,3] there can be several reasons for this kind of behaviour, we rephrase
them here for sake of completeness: there is a strong suppression between the LECs and
chiral logarithms at O(p4 ) and the assumption of resonance saturation using our naive
scalar picture at O(p6 ) might also be incomplete.
8.1. Meson masses
The numerical results for the masses can be presented in various ways. Using the
definition equation (7) we obtain


m2 / m2 QCD = 0.742 + 0.007 + 0.251,


m2K / m2K QCD = 0.693 + 0.027 + 0.280,
 
m2 / m2 phys = 0.741 0.029 + 0.289.
(29)
These results are the same within errors as those given in [3]. Hereafter the first quoted
number refers to the Born approximation, the second to the next-to-leading correction
and the third to the next-to-next-to-leading contribution. Notice, that as previous results
concerning the masses, even if the O(p6 ) contributions are bigger than the O(p4 ) all
three appear to be of the same size, (m2 /(m2 )QCD )(6) (m2K /(m2K )QCD )(6)
(m2 /(m2 )phys)(6) . Furthermore, almost all the quantities (besides the O(p6 ) LECs)
contributing to the masses have the same sign. Thus in order to make O(p4 )  O(p6 ) there
must be some fine tuning of the O(p6 ) LECs. Our simple not well controlled estimate of
the LECs at O(p6 ) can thus be the reason for large O(p6 ) corrections.
The isospin breaking quantities have as expansion

 2
mK 0 m2K QCD = (5.09 0.34 + 1.53) 103 GeV2 ,
 2

m m2 0 QCD = (3.96 + 1.54 + 3.17) 105 GeV2 .
(30)

G. Amoros et al. / Nuclear Physics B 602 (2001) 87108

101

The Gell-MannOkubo relation in the presence of isospin breaking is


 2
2 2
(31)
mK + m2K 0 m2 = 0.
3
3
The lowest order masses of course obey this identically and we use it extensively
throughout the calculation. The deviation is given by
GMO m2 + m2 0

GMO = (0 0.0172 0.0033) GeV2 = 0.0205 GeV2 .

(32)

8.2. Decay constants


The decay constant have as expansion
F /F0 = 1 + 0.135 0.075,
FK /F = 1 + 0.162 + 0.058,
F8 /F = 1.000 + 0.242 + 0.066 = 1.308,

(33)

and for the isospin violating quantities


F 0 3 F
= (1.0 + 0.3 + 0.8) 104 ,
F
F 0 8
= 0.0141 + 0.0039 0.0021 = 0.0159,
=
F 0 3
F3
= 0.0141 0.0010 0.0026 = 0.0105.
=
F8

(34)

8.3. Vacuum expectation values


The effective Lagrangian at lowest order contains two order parameters F0 and B0
both of them sensitive to the infrared end of the Dirac spectrum [35] but their role in
the spontaneous chiral symmetry breaking pattern is quite different. The only necessary
and sufficient condition of SB S is a nonzero value of the correlator function

1
lim i d 4 x | : Li (x)Rj (0) : | = ij F02 + O(mq ),
(35)
m0
4
where L = 12 (V A ) and R = 12 (V + A ) are Noether currents. There is no such
constraint on B0 whose size is determined by the vacuum expectation value
0|q i q j |0 = F02 B0 ij O(mq ),

(36)

and in principle has no lower bound other than vacuum stability, B0  0. Even though there
is no proof available is rather likely that QCD is not realized in the phase B0 = 0 [36].
The chiral expansion of the vacuum expectation value (36) can be ordered as


0|q i q j |0 = F02 B0 1 + 0|q i q j |0(4) + 0|q i q j |0(6) + ,
(37)
where the superscripts refer to the chiral order. Although the result is scale independent
order by order in the CHPT expansion it depends on the QCD renormalization scale QCD .

102

G. Amoros et al. / Nuclear Physics B 602 (2001) 87108

It follows from the fact that the quark scalar current needs to be defined precisely in QCD.
Already at O(p2 ) B0 depends on QCD . At O(p4 ) there appears an additional ambiguity
via the high energy constant H2 which is forced by the ambiguity in the subtraction of the
scalar two-point function
0|qq|0

= 0|qq|0|

chiral limit i



d 4 x 0| : (qMq)(x)(

qq)(0)

: |0 + O M2 ,

(38)

at zero distance. At O(p4 ) one can eliminate the constant H2 and thus obtain a sum rule

relating the ratios of the isospin asymmetry 0|dd|0/0|


uu|0

to the SU (3) asymmetry


0|s s|0/0|uu|0.

In order to quote numbers we fix H2 assuming scalar dominance


H2r = 2Lr8 .

(39)

With this approach we obtain the following expansion


0|uu|0

= B0 F02 [1 + 0.271 + 0.106],

0|dd|0
= B0 F02 [1 + 0.284 + 0.110],
0|s s|0 = B0 F02 [1 + 0.964 + 0.289],
where the quoted numbers correspond to the lowest
The dependence on H2r can be judged from 6

(40)
order, O(p4 )

and

O(p6 ),

respectively.

H2r = 0,
0|uu|0

= B0 F02 [1 + 0.264 + 0.104],

0|dd|0
= B0 F 2 [1 + 0.271 + 0.107],
0

0|s s|0 = B0 F02 [1 + 0.693 + 0.222].

(41)

The vacuum asymmetries are defined as

0|dd|0
0|s s|0
(42)
1 d,
1 s.
0|uu|0

0|uu|0

Using Partially Conserved Axial Currents (PCAC) these ratios are fixed to be 1, then d and
s measure the SU(2) and SU(3) breaking of the non-perturbative vacuum, respectively.
We get for them
d

= 0.0129 0.0011,

= 0.693 + 0.004,

(43)
O(p4 )

O(p6 ).

where the first number corresponds to


and the second to
As becomes clear
the breaking inside the SU(2) doublet driven by d is very small while the breaking due to
the presence of ms , s , is rather huge. There is a plethora of phenomenological calculations
determining Eq. (43) from sum-rules. Their results are very spread and no clear conclusion
can be reached. Otherwise the results in Eq. (43) are in the ball park of those quoted in [37]
for d and [38] for s while the remaining results [39] are quite off of Eq. (43).
6 The O(p 6 ) dependence on H r is from rewriting F 2 in F 2 in 0|qq|0.

0
2

G. Amoros et al. / Nuclear Physics B 602 (2001) 87108

103

9. Quark mass ratios


A review of the early quark mass determinations can be found in [40]. A more recent
list of references is the quark mass section in the Review of Particle Properties [41].
While sum-rule or lattice techniques allow to extract the value of quark masses via some
hadronic observable, the use of an effective Lagrangian, as CHPT, only allows to determine
the relative size of the quark masses, mu , md and ms . This approach assumes that the quark
masses can be treated as a perturbation on the massless QCD Hamiltonian, H0

+ ms s s].
HQCD = H0 d 3 x[mu uu
+ md dd
(44)
Quark masses depend on the renormalization scale, but if we use a mass-independent
renormalization scheme like MS they are renormalized multiplicatively and their ratios
are renormalization-scale-independent. Due to the locality of the symmetries in the chiral
Lagrangian the quark masses in CHPT can be identified with those in QCD via the Ward
identities, thus by construction they are the same as those in QCD.
The values of the relevant LECs can depend on the precise definition of the scalar
currents in QCD. In practice, since we have to determine the LECs from the same data as
we use to obtain the quark masses some ambiguities might occur [42]. These ambiguities
are discussed in [43] where we refer for a more detailed discussion. In the present context
they are fixed by our large Nc assumptions on the values of Lr4 and Lr6 .
To lowest order in the chiral expansion and neglecting electromagnetic corrections we
have
m2 0 = m2 = B0 (mu + md ),
m2K = B0 (mu + ms ),
m2K 0 = B0 (md + ms ),
1
m2 = B0 (mu + md + 4ms ) + ,
3

(45)

where is the effect of 0 mixing. The constant of proportionality is determined by the


2

quark condensate B0 = |0|uu|0|/F


+ O(mq ). We can use Eq. (45) to get
m2 + m2K 0 m2
ms
= K
24.2 (real); 25.8 [Eq. (19)]; 25.7 [Eq. (20)],
m

m2
2
2
2
mu mK mK 0 + m
= 2
0.66 (real); 0.56 [Eq. (19)]; 0.49 [Eq. (20)].
md
mK 0 m2K + m2

(46)

The three quoted numbers correspond to using the physical masses, electromagnetic
corrections using Dashens theorem, Eq. (19), and e2 ms corrections included as well,
Eq. (20).
To O(p4 ) we do not have quite enough information anymore to determine both quark
mass ratios and all the relevant LECs. To first order in isospin breaking it was noted
in [1,43] that the particular combination

104

G. Amoros et al. / Nuclear Physics B 602 (2001) 87108


2
m2s m
1
m2s 1 + (mu /md ) m2K m2K m2
=
Q 2

1+ 2
m
 1 (mu /md ) m2 m2 0 m2K
md m2u 4
K
2

(47)

could be related to the meson masses without corrections at this chiral order, thus providing
a tight constraint on the quark mass ratio. As was pointed out in [42] if the terms m
2 /m2s
are discarded Eq. (47) leads to an ellipsis where the variables are the ratios mu /md and
ms /md . The major semi-axis is given by Q while the minor one is equal to 1. Plugging the
masses in Eq. (47) gives
Q = 22.0(21.3) [Eq. (20)];

24.1(23.4) [Eq. (19)].

(48)

The numbers are by using the strong meson masses in the last part of Eq. (47). The presence
of higher orders corrections can be seen from the number in brackets, these are calculated
m = 24 at O(p4 ) of Table 1 and the
using the quark mass ratios from the fit with ms /
definition of Q in terms of the quark masses. Again the two sets of numbers correspond to
using the electromagnetic corrections estimated with e2 ms corrections included, Eq. (20),
or O(e2 ) only using Dashens theorem, Eq. (19).
The quantity Q can be determined to the same order from 3 decays giving
Q = 22.7 0.8.

(49)

The decay 3 is known in CHPT to O(p4 ) [44] and higher order corrections were
estimated using dispersion relations [45]. In view of the discrepancy between dispersive
estimates [32] and the full calculation to O(p6 ) [3] in the K4 process the error in Eq. (49)
is probably somewhat underestimated. The value is also obtained using the average value
of ( ) of [41].
m = 24 gives
Let us now turn to the O(p6 ) results. Our main fit with ms /
Q = 19.6 [Eq. (20)];

21.5 [Eq. (19)]

(50)

substantially lower than the values quoted in Eq. (48). On the other hand, Q is rather
m. In Fig. 3(a) we have shown how Q depends on the
sensitive to the input value for ms /
m. Otherwise the inputs are as in our main fit. Fig. 3(b) shows
input values chosen for ms /
similarly the dependence of mu /md on ms /
m.
The position of the previous mentioned ellipsis can be described as the ratio of SU(3)
breaking effects versus SU(2) isospin breaking using

ms m
.
(51)
md mu
The standard value [40] used mixing and combinations of the baryon masses to obtain
R = 43.5 2.2. Both of these inputs are now known to have rather large corrections
of O(mq ). mixing is treated in [46] and results for the baryon masses with earlier
references can be found in [47]. We obtain for this quantity
R

R 31 [Eq. (20)];

37 [Eq. (19)].

One of the main results from Table 1 is


mu
= 0.46 0.09.
md

(52)

(53)

G. Amoros et al. / Nuclear Physics B 602 (2001) 87108

105

Fig. 3. (a) Q, (b) mu /md as a function of ms /


m but other input fixed as in the main fit.

In the previous estimate the error is increased w.r.t. the one of the fit in order to include two
other sources of errors, about 0.02 from the error on the electromagnetic correction due to
m2K m2K 0 and another 0.02 from the rather large change that happened with the scale
variation at = 1 GeV. The other new result we obtain is the strong mass difference
m = (m m 0 )QCD = 0.32 0.20 MeV.

(54)

The change with the result quoted in [1] has two sources, about 0.04 MeV from the
inclusion of the mass-corrections in the Kaon electromagnetic mass difference and the
remainder comes from the O(p6 ) effects.
A possibly more serious variation with input is due to the assumptions on Lr4 and Lr6 .
Our tests of large Nc as described in Section 5.2 are mainly in quantities dominated by
vector exchange. We have therefore checked what happens when we vary the input value
of Lr4 and Lr6 over a fairly wide range. The resulting values for mu /md are plotted in Fig. 4.
It should be kept in mind that not all of the points shown have good fits to all inputs and
most of the points actually fall in the range of Eq. (53). It is a rather clear conclusion that
the value mu /md = 0 is far away from the numbers we have obtained. Unfortunately this
means that the solution to the strong CP problem needs to be found elsewhere.

10. Summary
In this paper we have explained how isospin breaking effects can be incorporated at
next-to-next-to-leading order in CHPT. The expectation that isospin breaking would not
significantly alter our previous results has been confirmed and the results presented here
are a first step in understanding how isospin breaking effects can be added consistently.
We have reanalyzed the main fit at O(p6 ) of the CHPT LECs in the presence of isospin
breaking. The determination of the Lri is only marginally changed compared to the isospin

106

G. Amoros et al. / Nuclear Physics B 602 (2001) 87108

Fig. 4. The ratio mu /md versus the inputs for Lr4 , Lr6 the rest of parameters correspond to those of
our main fit.

symmetric analysis. As by product we obtain control on the quantity mu /md at next-tonext-to-leading order in terms of meson masses. This constitutes one of our main new
results.
Furthermore, using the preliminary new experimental results on K4 we can obtain one
of the Zweig Rule suppressed LEC, namely Lr4 . It follows approximately a linear relation
together with Lr6 .
We also revised the value of the isospin breaking quantities Q and R discussing the
electromagnetic effects. In view of the low value of Eq. (50) a reanalysis of the 3
decay at the same order in the quark mass expansion as was done here seems necessary.

Acknowledgements
G.A. and P.T. were supported by TMR, ECContract No. ERBFMRX-CT980169
(EURODAPHNE).

References
[1] J. Gasser, H. Leutwyler, Nucl. Phys. B 250 (1985) 465.
[2] G. Amoros, J. Bijnens, P. Talavera, Nucl. Phys. B 568 (2000) 319, hep-ph/9907264.
[3] G. Amoros, J. Bijnens, P. Talavera, Phys. Lett. B 480 (2000) 71, hep-ph/9912398;
G. Amoros, J. Bijnens, P. Talavera, Nucl. Phys. B 585 (2000) 293, hep-ph/0003258.
[4] H. Leutwyler, M. Roos, Z. Phys. C 25 (1984) 91.
[5] P. Truol et al., E865 Collaboration, hep-ex/0012012.
[6] G. Ecker, Strong interactions of light flavours, hep-ph/0011026;
A. Pich, Effective field theory, hep-ph/9806303.
[7] G. Ecker, G. Muller, H. Neufeld, A. Pich, Phys. Lett. B 477 (2000) 88, hep-ph/9912264.

G. Amoros et al. / Nuclear Physics B 602 (2001) 87108

107

[8] J. Bijnens, G. Colangelo, G. Ecker, J. Gasser, M.E. Sainio, Nucl. Phys. B 508 (1997) 263,
hep-ph/9707291.
[9] R. Urech, Nucl. Phys. B 433 (1995) 234, hep-ph/9405341.
[10] M. Knecht, H. Neufeld, H. Rupertsberger, P. Talavera, Eur. Phys. J. C 12 (2000) 469, hepph/9909284, and work in progress.
[11] R. Dashen, Phys. Rev. 183 (1969) 1245.
[12] H. Neufeld, H. Rupertsberger, Z. Phys. C 68 (1995) 91;
H. Neufeld, H. Rupertsberger, Z. Phys. C 71 (1996) 131, hep-ph/9506448.
[13] J. Bijnens, Phys. Lett. B 306 (1993) 343, hep-ph/9302217.
[14] J.F. Donoghue, B.R. Holstein, D. Wyler, Phys. Rev. D 47 (1993) 2089.
[15] A. Duncan, E. Eichten, H. Thacker, Phys. Rev. Lett. 76 (1996) 3894[hep-lat/9602005];
A. Duncan, E. Eichten, H. Thacker, Nucl. Phys. Proc. Suppl. 53 (1997) 295, hep-lat/9609015.
[16] J. Bijnens, J. Prades, Nucl. Phys. B 490 (1997) 239, hep-ph/9610360.
[17] R. Baur, R. Urech, Phys. Rev. D 53 (1996) 6552, hep-ph/9508393.
[18] B.R. Holstein, Phys. Lett. B 244 (1990) 83.
[19] M. Finkemeier, Phys. Lett. B 387 (1996) 391, hep-ph/9505434.
[20] L. Rosselet et al., Phys. Rev. D 15 (1977) 574.
[21] G. Ecker, J. Gasser, A. Pich, E. de Rafael, Nucl. Phys. B 321 (1989) 311;
G. Ecker et al., Phys. Lett. B 223 (1989) 425.
[22] D. Espriu, E. de Rafael, J. Taron, Nucl. Phys. B 345 (1990) 22;
J. Bijnens, C. Bruno, E. de Rafael, Nucl. Phys. B 390 (1993) 501, hep-ph/9206236;
J. Bijnens, Phys. Rept. 265 (1996) 369, hep-ph/9502335;
S. Peris, M. Perrottet, E. de Rafael, JHEP 9805 (1998) 011, hep-ph/9805442.
[23] E. Golowich, J. Kambor, Nucl. Phys. B 447 (1995) 373, hep-ph/9501318;
E. Golowich, J. Kambor, Phys. Rev. D 58 (1998) 036004, hep-ph/9710214;
S. Durr, J. Kambor, Phys. Rev. D 61 (2000) 114025, hep-ph/9907539.
[24] J. Bijnens, G. Colangelo, P. Talavera, JHEP 9805 (1998) 014, hep-ph/9805389.
[25] C.A. Dominguez, E. de Rafael, Ann. Phys. 174 (1987) 372;
J. Bijnens, J. Prades, E. de Rafael, Phys. Lett. B 348 (1995) 226, hep-ph/9411285;
S. Narison, Nucl. Phys. Proc. Suppl. 86 (2000) 242, hep-ph/9911454.
[26] K.G. Chetyrkin, C.A. Dominguez, D. Pirjol, K. Schilcher, Phys. Rev. D 51 (1995) 5090, hepph/9409371;
M. Jamin, M. Munz, Z. Phys. C 66 (1995) 633, hep-ph/9409335;
K.G. Chetyrkin, D. Pirjol, K. Schilcher, Phys. Lett. B 404 (1997) 337, hep-ph/9612394;
P. Colangelo, F. De Fazio, G. Nardulli, N. Paver, Phys. Lett. B 408 (1997) 340, hep-ph/9704249;
K. Maltman, Phys. Lett. B 462 (1999) 195, hep-ph/9904370;
A. Pich, J. Prades, JHEP 9910 (1999) 004, hep-ph/9909244;
J. Kambor, K. Maltman, Phys. Rev. D 62 (2000) 093023, hep-ph/0005156.
[27] B. Moussallam, Eur. Phys. J. C 14 (2000) 111, hep-ph/9909292.
[28] S. Descotes, J. Stern, Phys. Lett. B 488 (2000) 274, hep-ph/0007082;
S. Descotes, hep-ph/0012221.
[29] S. Pislak, in: Talk given at Laboratori Nazionali di Frascati, Rome, Italy, June 22, 2000,
unpublished.
[30] G. Amoros, J. Bijnens, J. Phys. G 25 (1999) 1607, hep-ph/9902463.
[31] J. Bijnens, Nucl. Phys. B 337 (1990) 635;
C. Riggenbach et al., Phys. Rev. D 43 (1991) 127.
[32] J. Bijnens, G. Colangelo, J. Gasser, Nucl. Phys. B 427 (1994) 427, hep-ph/9403390.
[33] G. Makoff et al., Phys. Rev. Lett. 70 (1993) 1591;
G. Makoff et al., Phys. Rev. Lett. 75 (1995) 2069, Erratum.
[34] J. Heitger, R. Sommer, H. Wittig, ALPHA Collaboration, Nucl. Phys. B 588 (2000) 377, heplat/0006026.

108

G. Amoros et al. / Nuclear Physics B 602 (2001) 87108

[35] H. Leutwyler, A. Smilga, Phys. Rev. D 46 (1992) 5607.


[36] J. Stern, hep-ph/9801282;
J. Stern, Talk given at Workshop on Chiral Dynamics: Theory and Experiment (ChPT 97),
Mainz, Germany, 15 September, 1997, hep-ph/9712438.
[37] M.A. Shifman, A.I. Vainshtein, V.I. Zakharov, Nucl. Phys. B 147 (1979) 519.
[38] S. Narison, Phys. Lett. B 104 (1981) 485;
E. Bagan, A. Bramon, S. Narison, N. Paver, Phys. Lett. B 135 (1984) 463.
[39] P. Pascual, R. Tarrach, Phys. Lett. B 116 (1982) 443;
L.J. Reinders, H.R. Rubinstein, S. Yazaki, Phys. Lett. B 120 (1983) 209;
L.J. Reinders, H.R. Rubinstein, S. Yazaki, Phys. Lett. B 122 (1983) 487, Erratum.
[40] J. Gasser, H. Leutwyler, Phys. Rept. 87 (1982) 77.
[41] D.E. Groom et al., Eur. Phys. J. C 15 (2000) 1.
[42] D.B. Kaplan, A.V. Manohar, Phys. Rev. Lett. 56 (1986) 2004.
[43] H. Leutwyler, Non-lattice determinations of the light quark masses, hep-ph/0011049;
H. Leutwyler, Phys. Lett. B 378 (1996) 313, hep-ph/9602366.
[44] J. Gasser, H. Leutwyler, Nucl. Phys. B 250 (1985) 539.
[45] J. Kambor, C. Wiesendanger, D. Wyler, Nucl. Phys. B 465 (1996) 215, hep-ph/9509374;
A.V. Anisovich, H. Leutwyler, Phys. Lett. B 375 (1996) 335, hep-ph/9601237.
[46] J. Bijnens, P. Gosdzinsky, P. Talavera, Nucl. Phys. B 501 (1997) 495, hep-ph/9704212.
[47] B. Borasoy, U. Meissner, Ann. Phys. 254 (1997) 192, hep-ph/9607432.

Nuclear Physics B 602 (2001) 109131


www.elsevier.nl/locate/npe

Free energy and critical temperature in


eleven dimensions
Jorge G. Russo
Departamento de Fsica, Universidad de Buenos Aires, Ciudad Universitaria, Pab. I,
1428 Buenos Aires, Argentina
Received 13 February 2001; accepted 15 March 2001

Abstract
We compute the one-loop contribution to the free energy in eleven-dimensional supergravity, with
the eleventh dimension compactified on a circle of radius R11 . We find a finite result, which, in a
small radius expansion, has the form of the type IIA supergravity free energy plus non-perturbative
corrections in the string coupling gA , whose coefficients we determine. We then study type IIA
superstring theory at finite temperature in the strong coupling regime by considering M-theory on
R 9 T 2 , one of the sides of the torus being the euclidean time direction, where fermions obey
antiperiodic boundary conditions. We find that a certain winding membrane state becomes tachyonic
above some critical temperature, which depends on gA . At weak coupling, it coincides with the
Hagedorn temperature, at large coupling it becomes Tcr
= 0.31 lP1 (so it is very small in string
units). 2001 Published by Elsevier Science B.V.
PACS: 11.25; 04.65; 04.50.+h

1. Introduction
A problem of interest in type IIA superstring theory is to understand the evolution of the
degrees of freedom of the system as the coupling is increased from weak to strong values.
For weak couplings, the theory can be described in terms of a supersymmetric relativistic
string, but for strong coupling the relevant degrees of freedom are not well understood.
The study of a system at finite temperature can give some non-trivial information about
its microscopic degrees of freedom, and about their behavior as the system is heated up
to high temperatures. In this paper we will discuss some features of M-theory at finite
temperature, with the eleventh dimension X11 compactified on a circle of radius R11 .
Given our limited knowledge of M-theory, a complete treatment is of course presently
E-mail address: russo@df.uba.ar (J.G. Russo).
0550-3213/01/$ see front matter 2001 Published by Elsevier Science B.V.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 2 8 - 6

110

J.G. Russo / Nuclear Physics B 602 (2001) 109131

impossible. Nevertheless, we will find, somewhat surprisingly, that very interesting aspects
can be revealed by simple calculations.
For small radius R11 , one must recover the thermodynamics of string theory, which is
well understood qualitatively [1]. There is a first order phase transition at some T0 < TH ,
where TH is the Hagedorn temperature [2], with a large latent heat leading to a gravitational
instability of the thermal ensemble. The way the Hagedorn temperature is calculated in
string theory basically involves free string theory methods, so an important question is how
interactions modify this picture. In particular, one would like to understand what happens
to the Hagedorn temperature in the type IIA superstring theory at strong coupling, in other
words, what is the fate of the Hagedorn transition in D = 11.
The presence of phase transitions is usually reflected as infrared divergences in the oneloop free energy. This calculation is difficult to generalize to eleven dimensions, for a
number of reasons. In string theory, the one-loop free energy is essentially the sum over
free energy contributions of each individual physical string mode. In order to compute a
one-loop free energy in M-theory in this way, a more detailed knowledge of the relevant
physical degrees of freedom would be required. In addition, in the eleven-dimensional
theory there is no coupling constant parameter, and higher loops will give contributions
to the free energy of order one. Nevertheless, as a first step, one could try to determine
the free energy in eleven-dimensional supergravity. As a physical application, the elevendimensional supergravity result may be then used to incorporate non-perturbative D0 brane
contributions to the one-loop free energy of string theory. D = 11 supergravity is not
renormalizable as a quantum field theory. However, as we show in Section 2, the one-loop
free energy is finite, thanks to a cancellation between boson and fermion contributions
(a general discussion of one-loop divergences in supergravity can be found in [3]). This is
important, since the presence of an UV divergence (as happens in a purely bosonic theory),
would prevent us from recovering the ten-dimensional physics in the small radius limit.
One of the difficulties in understanding details of the Hagedorn transition in string theory
is that gravitational effects cannot be neglected shortly above the Hagedorn temperature,
due to a large genus zero contribution to the free energy [1]. The description in terms of
a free string gas in a flat background is not applicable; rather, one expects gravitational
collapse near the Hagedorn temperature. One can give a qualitative description of what
physical picture should be expected (e.g., by using the microcanonical ensemble one
can argue which string configurations are favored and dominate the density of states),
but the arguments are mostly based on string propagation in flat space. A more detailed
understanding was recently achieved in non-commutative open string theory, which does
not contain gravitation [4].
In the case of M-theory at large radius R11 , a flat-theory description of the thermal
ensemble can be justified only for temperatures T  O(lP1 ), where lP is the elevendimensional Planck length. The reason is the following one. Statistical mechanics is valid
provided the system has a large volume, so that it contains many degrees of freedom.
In eleven dimensions, large volume means that the size R of the system is R  lP . On
the other hand, a flat theory description requires that corrections to the flat Minkowski
metric are small, i.e., GN E/R 8  1, with GN lP9 . Now consider the thermal ensemble at

J.G. Russo / Nuclear Physics B 602 (2001) 109131

111

temperature T = O(lP1 ). Then the energy density is E/R 10 = const lP11 , since there is no
other parameter in the theory. Hence gravitational effects are of order GN E/R 8 R 2 / lP2 ,
so they are important for R > lP . Thus it is not possible to have a statistical description of
the thermal ensemble in flat space near the Planck temperature. A temperature T = O(lP1 )
is in fact the maximum temperature that a statistical system can reach in eleven dimensions
without gravitational collapse. For any T > O(lP1 ), a thermal ensemble with size R > lP
will be inside its Schwarzschild radius, so it will collapse into a black hole. This can be
compared with the situation in string theory, where one has the string coupling gs as a free
parameter, and for sufficiently small gs gravity can be ignored at any T < TH (but not at
T > TH due to the appearance of a genus zero contribution F0 1/gs2 [1]).
Despite these complications, using flat-space methods to study the Hagedorn transition
has led to important insights on the nature of string theory and its physical degrees of freedom. One may then expect that a similar simplified study in eleven dimensions can teach us
important lessons about M-theory. Here we will find that the Hagedorn temperature admits
a straightforward generalization to eleven dimensions. In string theory, the Hagedorn temperature can be found as the temperature at which a certain winding state becomes tachyonic [5] (see Section 3). In eleven dimensions, this winding state is a winding membrane.
We find that it becomes tachyonic at some critical temperature Tcr = Tcr (gA ), where gA is
the type IIA string coupling, which smoothly interpolates between the Hagedorn temperature (gA  1) and a critical temperature T = O(lP1 ) (gA  1). This is done in Section 4,
where we also include some remarks about a duality to type 0A string theory [6].
Other discussions about the Hagedorn temperature in string theory can be found, e.g., in
[79]. There have also been some discussions on membrane theory at finite temperature in
[10], where there is an attempt of computing the free energy. The matrix theory approach
[11] has also been considered at finite temperature in Refs. [1215], which discuss other
aspects of the Hagedorn transition. There is no overlap with the present work.
In Appendix A we review standard properties of the non-holomorphic Eisenstein
series [16]. Appendix B contains new Eisenstein-type series with alternating signs in the
sum. We derive formulas for the expansions at large and small values of the modular
parameter. In the main text, these series arise as the contribution of antiperiodic fermions.

2. Free energy in eleven dimensions


The thermal ensemble at temperature T can be studied as usual by considering the theory
in euclidean space where the time coordinate is compactified on a circle of circumference
1/T , i.e.,
X0 = X0 + 2R0 ,

T = (2R0 )1 ,

(2.1)

where fermions obey antiperiodic boundary conditions. In order to have a description of


type IIA superstring theory which can be extended beyond perturbation theory, one should
thus consider euclidean M-theory on a 2-torus X0 , X11 ,
X0 = X0 + 2R0 ,

X11 = X11 + 2R11 ,

112

J.G. Russo / Nuclear Physics B 602 (2001) 109131

where fermions are antiperiodic around X0 , and periodic around X11 . Here we will
consider the case of a rectangular torus.
The supersymmetric compactification of M-theory on a 2-torus gives rise to a theory
that inherits the SL(2, Z) isometry of the 2-torus, because this symmetry is not broken by
boundary conditions of the fields. In the case of finite temperature M-theory, the different
boundary conditions for fermions in the directions X0 , X11 break the SL(2, Z) symmetry.
This will be reflected in the present calculation of the free energy. Instead, in a purely
bosonic theory, the exact partition function must have the symmetry
Zbos (geff ) = Zbos(1/geff ),

(2.2)

with

R11
= 2
T gA .
(2.3)
R0

Here gA = R11 /
is the type IIA string coupling. This implies for the free energy the
relation:


1
Fbos (geff , A, lP ) = geff Fbos
(2.4)
, A, lP ,
geff
geff

where A = R0 R11 . The symmetry (2.4) can also be expressed in terms of string theory
parameters {gA , T ,
} and relates low and high temperature regimes, as well as weak and
strong coupling regimes.
Here we will obtain the one-loop contribution to the free energy in eleven-dimensional
supergravity compactified on a circle by adding to the ten-dimensional supergravity

2
2
expression an extra factor containing a sum over KaluzaKlein modes m e m /R .
The free energy in ten-dimensional supergravity can be obtained from superstring theory
as a limit
0 (deriving the one-loop contribution to the free energy directly from the
component formulation of D = 11 supergravity is more complicated). This is similar to
[17,18], where the one-loop 4-graviton amplitude in D = 11 supergravity was computed
by adding the KaluzaKlein modes to the D = 10 supergravity amplitude.
2.1. Free energy in a simplified model
Before considering the free energy in string theory, it is instructive to study a simplified
model in which we sum up the individual free energies of KaluzaKlein scalar fields. That
is
  d D2 p


m2
p /T
2
2
,

log
1

e
=
p

+
.
F (T ) = V T
(2.5)
p
2
(2)D2
R11
m
Expanding the logarithm and using
e

2 ab


=


0

dt at b/t
e
,
t 3/2

J.G. Russo / Nuclear Physics B 602 (2001) 109131

113

one can write




F (T ) = V

D+1
2

w2 R02
m2 t
exp 2
,
t
R11
w=1

dt 

R0 = (2T )1 ,

(2.6)

where we have ignored a multiplicative numerical constant. Let us now set D = 11 (so that
[V ] = cm9 ). Including the vacuum part w = 0, and making a Poisson resummation in w,
i.e.,


2
t  t k22
Rt w2
R ,
(2.7)
e
=
e
R
w=
k=

we get

F (T ) = T V


 2
k
m2
.
exp t
+ 2
R02 R11
k,m

dt 
t 11/2

(2.8)

This can be recognized as the one-loop contribution of KaluzaKlein scalars associated


with the 2-torus X0 , X11 . The integral diverges in the ultraviolet region t 0. We can
isolate the divergent piece by performing Poisson resummation in k, m and introducing the
new integration variable s = 1/t, so that

F (T ) = 12 V R11
0




ds 11/2 
2
s
exp s w2 R02 + w
2 R11
.
s

(2.9)

w,w

Now the UV divergence is in the term (w, w


) = 0. In string theory the analog term will
cancel against a fermion contribution. Thus we get


V R11 (11/2)
2 11/2
F (T ) =
w2 R02 + w
2 R11
2 11/2

(w,w )=(0,0)

+ divergent term.

(2.10)

This can be expressed in terms of an Eisenstein series (see Appendix A)


F (T ) =

V R11 (11/2)
(11)E11/2(geff ) + divergent term.
(R11 R0 )11/2

(2.11)

It satisfies the symmetry relation mentioned above, F (geff , A) = geff F (1/geff , A) (this
symmetry still holds for the regularized divergent part for a cutoff proportional to lP1 , i.e.,
independent of the radii R0 , R11 ). To study the behavior at geff  1 and geff  1 we use
the expansions (A.7), (A.8). We obtain



945 (11) 24 (10)
R0
F (T )
=

+
O
exp
2
, R11  R0 ,
(2.12)
10
V
R11
32 5 R11
5 R010



945 (11)R11
24(10)
R11
F (T )
=
, R11  R0 . (2.13)

+ O exp 2
9
V
R0
32 5R011
5 R0 R11

114

J.G. Russo / Nuclear Physics B 602 (2001) 109131

The leading term in Eq. (2.13) has the correct form for the free energy of a massless field
theory in D = 11. The expression (2.12) contains two terms with power-like dependence on
R11 /R0 . The subleading term proportional to 1/R010 gives the expected expression for the
free energy of a D = 10 massless field theory. However, there is a leading term proportional
10 . The presence of a term of the form 1/R 10 in the supergravity calculation at
to 1/R11
11
R11  R0 would be problematic because there is no such contribution in superstring theory
at weak coupling. As we shall see below, in the supergravity calculation the analog term
cancels out.
2.2. One-loop free energy in D = 11 supergravity
Let us now consider the supergravity computation. The calculation of the free energy in
type II superstring theory was carried out in [1] in the genus one approximation (valid for
gA  1), with the result

 r02 |w
+w |2

5 d 2
1
( ) 24
Fstring = V 4 2

e 2
4
26
w
,w
F



|2 |8 + |3 |8 + |4 |8 (0, ) + ei(w+w ) 24 44 + 44 24 (0, )




eiw 24 34 + 34 24 (0, ) eiw 34 44 + 44 34 (0, ) ,
R02
(2.14)
.

In order to obtain the ten-dimensional supergravity result, we first separate the term with
vanishing winding w = 0, writing
r02

Fstring = Fstring
+ F0 ,

where Fstring
is as in (2.14) with the omission of the w = 0 term in the sum, which we
call F0 . The free energy in ten-dimensional supergravity is obtained by taking the limit

0 in F0 . Taking
0 implies keeping the leading terms of the theta functions and
Dedekind function at large 2 , that is




( )
2 (0, )
= q 1/12 1 q 2 ,
= 2q 1/4 1 + q 2 , q ei ,


4 (0, )
3 (0, )
= 1 + 2q + O q 3 ,
= 1 2q + O q 3 .

We obtain
(10)
FSG


5
= lim F0 = 256V 4 2

2
 r0
2
d2 d1 
w
2 w
,
1 (1) e
26
w

(2.15)

where the integration region is now the whole strip 2 > 0, |1 | < 1/2.
Integrating over 1 and making a Poisson resummation in w
, we get
(10)
FSG

5
V
= 256 4 2

r0


0

d2 
11/2
2

2 2
k
r02

2
(k+ 21 )2
r02


.

(2.16)

J.G. Russo / Nuclear Physics B 602 (2001) 109131

115

The appearance of half-integer momentum modes is a well-known distinctive feature


of having fermions with antiperiodic boundary conditions. Similar expressions for the
partition function or for the free energy appear in string compactifications where fermions
obey antiperiodic boundary conditions around some spatial dimension [19] (for recent
discussions on string compactifications with antiperiodic fermions, see, e.g., [20,21]).

Taking a similar limit on Fstring


, and making a Poisson resummation in w
, gives

Fstring

VT


9/2
4 2

d2 d1
11/2


 2
1 (1)w e

e22

w2 r02 + k2


e2i1 kw .

r
0

(2.17)

k,w

One can see the presence of the thermal tachyon corresponding to the term k = 0, w = 1,
which reflects as an infrared divergence of Fstring for T > TH : the integral is divergent at
2 for r02 < 2 (this is precisely the critical radius that one obtains by examining the
spectrum, see Eq. (3.4) with aL = aR = 1/2).
The KaluzaKlein modes associated with the eleventh dimension have masses |m|/gA in
(10)
string units, m = integer. By adding their contribution to FSG , given in (2.16), we get the
one-loop contribution to the free energy in eleven-dimensional supergravity compactified
on a circle X11 . This is
(11)
FSG

5
V
= 256 4 2

r0

d2 
11/2

2 2
2 m
gA


e

2 2
k
r2
0

2
(k+ 21 )2
r2
0


.

(2.18)

k,m

The first term (containing exp[(2 /r02 )k 2 ]) is essentially the same as the expression
obtained in the previous bosonic example (2.8). The second term (with exp[(2 /r02 )(k +
1 2
2 ) ]) represents the fermion contribution. Making Poisson resummation in both k, m we
get
(11)
FSG


5
= 256V gA 4 2


0

ds 11/2 

2 2
2 2
s
1 (1)w es(w r0 +n gA ) , (2.19)
s

w ,n

i.e.,

5 (11/2)
(11)
FSG = 256V gA 4 2

(r0 gA )11/2
11/2

geff


1 (1)w

,
2 )11/2
(w
2 + n2 geff
w
,n

(2.20)

with geff = R11 /R0 = gA /r0 . This can be written in terms of the Eisenstein-type series
defined in appendices A, B as follows:
(11)

FSG
29 (11/2) (11) 
= T 10
E11/2(geff ) F11/2 (geff ) .
9/2
V
11/2 g
eff

(2.21)

116

J.G. Russo / Nuclear Physics B 602 (2001) 109131

Using the formulas (A.7), (A.8), (B.11), (B.12) for the weak and strong coupling
expansions, we obtain
(11)



FSG
24 (10) 10
=
2 1 T 9 + O e2/geff ,
VT
5

geff  1,

(2.22)

and
(11)



FSG
945 (11) 11
3 212(10) T 10
=
2 1 T 10
+ O e2geff ,
5
5
10
2R11 V T
64

geff

geff  1.

(2.23)

The weak coupling expression (2.22) has the expected field theory behavior F /(V T )
T D1 for a free energy of a D = 10 dimensional massless field theory. The leading
(10)
given in (2.16). The strong geff  1 coupling expression (2.23) has
term is in fact FSG
the expected field theory behavior F /(R11 V T ) T D1 for a free energy of a D = 11
dimensional massless field theory. This agrees with the expectation that varying gA from
small to large values should lead to an interpolation of a ten-dimensional and an elevendimensional theory. This does not happen in the bosonic theory, which has an extra term
at small coupling (the underlying reason for which in the bosonic theory the small radius
limit does not give the ten-dimensional result is the UV divergence, which is different in
ten and eleven dimensions; consequently, some memory of the KK modes survives even
at small radius R11 ). Note also that the only power-like correction in Eq. (2.23) is always
subleading, since geff > 1. It is independent of the temperature. Indeed, as a function of
T , gA , the free energy in (2.23) has the form
(11)



FSG
945 (11) 11
12(10)
1

=
2 1 T 10
+ O e2geff ,
5
15

10
2R11 V T
64

( gA )
geff  1.
(2.24)

Finally, one can use the above results to define an improved expression for the free
energy of type IIA superstring theory by adding to the one-loop expression (2.14)
(representing the contribution of perturbative string modes) the contribution of D0 branes
represented by the exponentially small terms in (2.22). Their explicit form is obtained using
Eqs. (A.7), (B.11). We find


 

 m 5
V T 10 211  
wm
w
1 (1)
K5 2
.
Fstr+D0 = Fstring
(2.25)
5
w
geff
geff
w,m=1

3. Hagedorn temperature in string theory


It is useful to recall the way the Hagedorn temperature in superstring theory is
manifested in the spectrum, as the temperature at which a certain winding mode becomes
massless [5]. We consider the theory in Euclidean space where the time coordinate X0 is
compactified on a circle of circumference 1/T . In string theory, the presence of coordinates

J.G. Russo / Nuclear Physics B 602 (2001) 109131

117

compactified on circles gives rise to winding string states. The string coordinate X0 (, )
can be expanded as follows:
 ),
X0 (, ) = x 0 + 2
p0 + 2R0 w0 + X(,
m
0
, m0 , w0 = 0, 1, 2, . . .,
p0 =
(3.1)
R0

 ) is a single-valued function of and d X
0 = 0. The Hamiltonian and
where X(,
0
level matching constraints are
w02 R02
m2
+
20 + 2(NL + NR aL aR ) = 0,

R0
NL NR = m0 w0 .
H =
pi2 +

(3.2)

Here aL , aR are the normal ordering constants, which represent the vacuum energy of the
(1 + 1)-dimensional field theory (e.g., for the bosonic string, aL = aR = 1). The Hagedorn
temperature can be obtained as usual by determining the radius R0 at which infrared
instabilities first appear. We have seen this effect in Section 2 in the one-loop contribution to
the free energy; in the presence of infrared instabilities, the integral over the torus modular
parameter 2 diverges at large 2 . This happens when some state has negative H , i.e.,
when a tachyon first appears in the spectrum (apart from the usual bosonic string tachyon).
By examining the form of the Hamiltonian, one immediately sees that the first tachyon
that appears as the temperature T = (2R0 )1 is increased from zero has NL = NR = 0,
m0 = 0 and w0 = 1. For such states, the critical R0 is determined by
H =0=

R02
2(aL + aR ),

(3.3)

whereby
TH =

1
1
.
=

2R0 2 2
(aL + aR )

(3.4)

In the NSR formulation of type II superstring theory the calculation is similar. The
tachyon appears in the NSNS sector, where the normal-ordering constants are aL = aR =
1/2. GSO projection does not remove this tachyon state, because for odd winding number
the GSO condition is reversed [1] (this is explicit in the one-loop expression for the free
energy in Section 2; the tachyon state with w0 = 0 is projected out by GSO, but not this
thermal tachyon with w0 = 1).
In order to reproduce this calculation in the GreenSchwarz formulation of the
superstring (which is more suitable for the generalization to membrane theory), we
need to calculate the normal ordering constant for the Euclidean theory on R 9 S 1 .
At zero temperature, the normal ordering constant vanishes because of a cancellation
between bosons and fermions. In the thermal ensemble at finite temperature, fermions obey
antiperiodic boundary conditions under X0 X0 + 2R0 . As a result, supersymmetry
is broken and the vacuum energy will not vanish. In type II superstring theory with
antiperiodic fermions, the number operators in the sector w0 = 1 are given by

118

J.G. Russo / Nuclear Physics B 602 (2001) 109131

NL =
NR =



n=1

a a

i
n
ni + n 12 Sn
Sn ,

a a

i
n
Sn ,
ni + n 12 Sn

n=1

i = 1, . . . , 8,

a = 1, . . . , 8.

(3.5)

Thus the normal ordering constant is as in the NS sector of the NSR formulation, i.e.,
aL = aR = 1/2. In this way we reproduce the result for the Hagedorn temperature in the
GreenSchwarz formulation.
The calculation of the normal ordering constants can be done by -function regularization. For the operators in (3.5), one has


1
aL = aR = (D 2) EB + EF ,
2
with
EB =

n,




n + 12 .
EF =

n=1

n=0

Using the formulas


1
= (),
n
n=1

n=0

(n +

1
2)



= 2 1 (),

and (1) = 1/12, we find


1
1
,
EF = ,
12
24
so that for D = 10 one has aL = aR = 1/2.
EB =

4. M-theory at finite temperature


We will describe M-theory at finite temperature as in Section 2 by considering the
eleven-dimensional theory in euclidean target space with periodic time X0 , and periodic
coordinate X11 . Fermions are antiperiodic around X0 and periodic around X11 . Thus
we are to consider a toroidal compactification of Euclidean M-theory with (, +) spin
structure. Having the topology R 9 T 2 , membranes can wrap on a 2-torus X0 , X11 .
When viewed in eleven dimensions, the winding string that in Section 3 led to a
tachyon instability is a membrane wrapped around X0 , X11 with winding number equal
to 1. Small oscillations of this membrane are effectively described by the D = 11
supermembrane theory [22]. The observation that there is a tachyon instability at T > Tcr
will be independent of many details of the membrane Hamiltonian, depending only on the
net vacuum energy. Although we will not determine the exact spectrum, being outside of
the scope of this paper, it is interesting to note that, because of the supersymmetry breaking

J.G. Russo / Nuclear Physics B 602 (2001) 109131

119

boundary conditions, no flat direction remains in the membrane Hamiltonian, so the exact
supermembrane spectrum must be discrete (see discussion in Section 4.2).
In the sector with zero winding, there can be other types of configurations which give rise
to low energy excitations, related to D0 brane configurations. We emphasize, however, that
the aim here is not to provide a complete account of all relevant excitations of M-theory at
a given temperature and radius R11 , but rather to point out the existence of a winding mode
that becomes tachyonic at some critical temperature Tcr (which will depend on the radius
of the eleventh dimension). This does not exclude that there could be other instabilities. In
particular, in matrix model calculations it has been shown [15] that at some sufficiently high
temperature there are D0 branes which cluster at one point. This configuration might lead
to a gravitational instability, but estimating the temperature at which such configuration
occurs does not appear to be simple [15].
4.1. Toroidal membranes
Before considering the finite temperature case, it is convenient to briefly review the
light-cone Hamiltonian formalism for membranes wrapped on a torus in Minkowski space,
where X10 and X11 are compact. Let , [0, 2) be the membrane worldvolume
coordinates. We can write
10 (, ),
11 (, ),
X11 (, ) = R11 + X
X10 (, ) = w0 R10 + X
(4.1)
10 , X
11 are single-valued functions of and . The transverse coordinates
where X
i
X (, ), i = 1, 2, . . . , 8 are all single-valued (we use the notation where the eleven
bosonic coordinates are {X0 , Xi , X10 , X11 }). They can be expanded in a complete basis
of functions on the torus,
 i
X(k,m) eik +im ,
Xi (, ) =

k,m

P i (, ) =

(2)2

1


= 4 2 R11 T2 ,

i
P(k,m)
eik +im ,

k,m

(4.2)
] = cm3 ).

where T2 is the membrane tension ([T2


The membrane light-cone Hamiltonian
[23,24] takes the form H = H0 + Hint , with [25,26]

1  i i
2 2
2
i

H0 = 8 4
T22 R10
Pn Pn + km
,
R11 w02 +
Xni Xn
2 n
1 
j
j

Hint = 2
(n1 n2 )(n3 n4 )Xni 1 Xn2 Xni 3 Xn4 ,
4gA
11
X0 + X
X+ =

= x + +
p+ ,
2
n (k, m),
n n
= km
mk
,

2
R
R2
2
2 3
2 + w 2 m2 10 .
11
=
4
R
T
,

=
k
gA
(4.3)
2
km
11
0
2

R11

120

J.G. Russo / Nuclear Physics B 602 (2001) 109131

Here only the bosonic modes have been written explicitly (fermion modes will be included
later). The constant gA represents the type IIA string coupling. One can introduce mode
operators as follows:
 i

i
i
i
(k,m) + (k,m)
,
=
X(k,m)
2 w(k,m)

1  i
i
i
P(k,m)
,
= (k,m)
(k,m)
2

i
i
i
i
X(k,m) = X(k,m)
,
P(k,m) = P(k,m)
,

w(k,m) 9(k)km ,

(4.4)

where 9(k) is the sign function. The canonical commutation relations imply
 i

j
X(k,m) , P(k
,m
) = ik+k
m+m
ij ,

 i
j
(k,m) , (k
,m
) = w(k,m) k+k
m+m
ij ,

(4.5)

i
and similar relations for the (k,m)
.
The mass operator is given by
2
2
2
2
= 2H0 + 2Hint pi p10
.
M 2 = 2p+ p pi p10

(4.6)

The Hamiltonian is non-linear. There are two situations where one can extract useful
information from this Hamiltonian. One is the limit of large gA , with R10 /R11 fixed, in
2 and can be considered
which the non-linear terms are multiplied by the small number 1/gA
in perturbation theory. The other limit is gA 0 at fixed R10 /R11 . This is related to the
zero torus area limit of M-theory on T 2 , which leads to ten-dimensional type IIB string
theory. Hint is positive definite, and any state |  with  |Hint|  = 0 will have infinite
mass in the zero area limit, where gA 0 (with T2 , so that
= (4 2 R11 T2 )1
and R10 /R11 remain fixed). The only states that survive are those states made of operators
i
i
, n(p,q)
with the same value of (p, q) [26]. They satisfy  |Hint|  = 0, so that
n(p,q)
Hint drops out from  |M 2 | . They describe the (p, q) strings of type IIB superstring
theory (the proposal that the (p, q) string bound states of type IIB string theory originate
from membranes was first made by Schwarz [27]).
Let us now focus on the situation of large gA . To leading order in perturbation theory
2 , the interaction term can be dropped. The solution to the membrane equations of
in 1/gA
motion is given by

Xi (, , ) = x i +
pi + i
/2




(4.7)
wn1 ni eik +im + ni eik im eiwn .
n=(0,0)

Let the momentum components in the directions X10 and X11 be given by
p10 =

l10
,
R10

p11 =

l11
,
R11

where l10 , l11 are integers. The nine-dimensional mass operator takes the form M 2 = H,
with

J.G. Russo / Nuclear Physics B 602 (2001) 109131

H=

2
l10

2
l11

2
w02 R10
1
+
H,

2
2
R10
R11


i
i
i
i
(k,m)
.
(k,m)
+ (k,m)
(k,m)
H=

121

(4.8)
(4.9)

k,m

The level-matching conditions are given by [25,28]


N+ N = w0 l10 ,

N+ N = l11 ,

(4.10)

where
N+ =


k i
i
(k,m) (k,m)
,

km
m=
k=1


k i

i
N =
(k,m) (k,m)
,

km
m=
k=1



m  i
i
i
i
N+ =
(k,m) (k,m)
,
+ (k,m)
(k,m)
km

N =

m=1 k=0



m=1 k=0


m  i
i
i
i
(k,m) (k,m)
.
+ (k,m)
(k,m)
km

To define the operator H in the quantum theory, we have to specify the normal ordering
prescription. This will give rise to a vacuum energy. The annihilation operators are
i
i
, (k,m)
with k > 0 for all m, and k = 0, m > 0. Defining
(k,m)


i
i
i
i
=
:(k,m)
(k,m)
: + : (k,m)
(k,m)
:,
H
(4.11)
n

where the normal ordering symbol : : means as usual taking the annihilation operators to
the right, one finds the relation
 + 2(D 3)E,
H=H
1
E=
km .
2

(4.12)

k,m

This constant shift represents the purely bosonic contribution to the vacuum energy of
the (2 + 1)-dimensional field theory (discussed in [29]). If one chooses supersymmetry
preserving boundary conditions for fermions, then the fermion and boson contributions
to the vacuum energy cancel out [22,28]. Being a consequence of the underlying
supersymmetry, this result also holds when non-linear terms are included.
4.2. Vacuum energy for the finite temperature theory
Let us now extend this to the supermembrane theory at finite temperature. The euclidean
time coordinate X0 plays role of X10 . Fermions will obey antiperiodic boundary conditions
around X0 , and periodic boundary conditions around X11 . We are interested in the sector

122

J.G. Russo / Nuclear Physics B 602 (2001) 109131

w0 = 1, where fermions are antiperiodic under + 2 . This implies that the


frequencies k in the Fourier expansions will be half integers, and the frequencies m will
be integers (since fermions are periodic under + 2 ). The Hamiltonian operator is
(cf. (3.5))
H=
where
=
H

l02
R02

2
l11

2
R11

R02
1 
+ (H
+ 2(D 3)E),

2

(4.13)


a a

i
i
a a
:n
Sn : ,
ni : + : n
ni : + k+1/2,m :Sn
Sn : + :Sn
n

and
1
(km k+1/2,m ),
2
k,m


m2 1/2
.
km = k 2 + 2
geff
E = EB + EF =

(4.14)

(4.15)

The sums in (4.14) are divergent, but they can be defined by analytic continuation.
The procedure generalizes the usual zeta-function regularization used in Section 3 for
the superstring case, and it is equivalent to the functional relation E () = cE1 ()
allowing the definition of Eisenstein series with < 1/2 [16]. We write


1
1
1

E = lim
1/2 2
(km )2 (k+1/2,m )2
k,m



2 ()

k,m 0


d  2
2
km e (k+1/2,m )
e
.

(4.16)

Then, using the Poisson formula (2.7) one obtains







geff  ds 1
2
E=

1 (1)w exp s w2 + w
2 geff
,
2 () w,w
s

(4.17)

where we have made the change of integration variable, s = 1/ . Hence


E=


geff 21 (1 ) 
1
1 (1)w
.
2 )1
2
2 ()
(w + w
2 geff

(4.18)

w,w

Setting now = 1/2 we get





geff
2 3/2
.
1 (1)w w2 + w
2 geff
E = 2
8

(4.19)

(w,w )=(0,0)

Thus we have (see Eqs. (A.1), (B.1))


E =

4 2 geff



(3) E3/2 (geff ) F3/2 (geff ) .

(4.20)

J.G. Russo / Nuclear Physics B 602 (2001) 109131

123

Using the expansions (A.5), (A.6), (B.9), (B.10) the vacuum energy takes the form


1
E = + O e2/geff , geff  1,
(4.21)
8
and


7
1
E =
(4.22)
(3)geff
+ O e2geff , geff  1.
2
12geff
16
The explicit analytic form for the exponentially small terms can be read from the formulas
in the appendices.
Notably, Eq. (4.21) implies that at geff  1 (i.e., small type IIA coupling gA or
sufficiently low temperatures), the vacuum energy is identical to that of type II superstring
1
1
+ 24
), due to a cancellation of the term
theory, i.e., 2(D 2)E = 2(D 2)( 12
3/2
proportional to (3)/geff in the expansions (A.5), (B.9).
A question is why the vacuum energy gives the correct result in the weak coupling limit
gA  1. In the derivation of the vacuum energy, we have used the assumption that gA  1
to neglect the contribution of the non-linear terms. The fact that the correct result emerges
at weak coupling indicates that a possible extra contribution coming from the non-linear
terms in the Hamiltonian may cancel out between fermion and boson contributions.
Another interesting point is the issue of flat directions in the membrane Hamiltonian for
the wrapped membrane. Consider first the case of supersymmetric boundary conditions. In
the strict limit gA , one has a Hamiltonian which is a sum of harmonic oscillators,
so there is no flat direction and the spectrum is discrete. For any finite gA  1, states
representing small oscillations should be almost stable, since they only see the harmonic
potential. However, if flat directions are present, they may decay by tunnel effect (see also
discussion in [30]). This effect should be exponentially small for large gA . Now, in the
present case of non-supersymmetric boundary conditions, possible flat directions will be
removed by the same effect flat directions are removed in the bosonic theory (described
in [31]). The motion is confined to some finite region, and the exact spectrum of the
supermembrane must be discrete. For large values of gA , most of the states are confined to
the harmonic region of the potential, so this effect should not play a significant role.
4.3. Critical temperature in M-theory
As in the string theory case, there will be a tachyonic instability when H < 0 for some
state (see (4.13)). The first state that solves H = 0 is a state with l0 = l11 = 0, w0 = 1,
which is annihilated by all annihilation operators ni , ni (this is nothing but the uplift of
the winding tachyon of string theory to eleven dimensions). From Eq. (4.13), we thus find
that the critical temperature is determined by the solution of the equation
R02
+ 2(D 3)E.

Using Eq. (4.20) and setting D = 11, this becomes




16
1
=

(3) E3/2 (geff ) F3/2 (geff ) .
2
geff
Tcr
H=0=

(4.23)

(4.24)

124

J.G. Russo / Nuclear Physics B 602 (2001) 109131

Here geff = 2
Tcr gA , so this is a transcendental equation for Tcr . It can be solved
analytically in two regimes, geff  1 and geff  1, using the expansions (4.21) and (4.22).
At weak coupling geff  1, Eq. (4.23) becomes


2/g
R02
1
eff
+
O
e
,
=
16

(4.25)

8
i.e.,
1
(4.26)
.
2 2

This coincides with the Hagedorn temperature. This is a consequence of the observation of
the previous subsection that the vacuum energy reduces to the weak coupling superstring

value at geff  1. Since the critical temperature in this regime is of order 1/


, the
condition geff  1 implies gA  1.
It is easy to see that in a regime gA  1, the coupling geff will be large at the critical
temperature T = Tcr . This means that, in order to determine the critical temperature, we
have to use (4.22). Using (4.22) and keeping only the leading term, the condition (4.24)
determining the critical temperature becomes
Tcr =

3/2
= 28 (3)geff = 56
(3)gA Tcr .
Tcr2

(4.27)

Thus the critical temperature at strong coupling is


1
1
Tcr
= (2T2 )1/3 ,
=

1/3
a
a (2gA )
1/3
a = [28 (3)] = 3.23,

gA  1,
(4.28)

where T2 is the membrane tension (we have used Eqs. (4.2), (4.3)). In terms of the eleven2 = 2R 3 / l 3 , so
dimensional Planck length lP , gA
11 P
Tcr =

1
= 0.31lP1.
alP

(4.29)

2 = 16 5 l 9 . As a check,
Here lP is normalized so that the gravitational coupling is 11
P

2/3
note that geff (Tcr ) = 2
Tcr gA gA  1, which is consistent with approximating the
Eisenstein functions by Eq. (4.22). The same result (4.28) is obtained by solving (4.24)
numerically at gA  1. Thus the critical temperature for the type IIA superstring decreases
at strong coupling. In terms of the eleven-dimensional Planck length, at large radius, it
approaches a constant value, Tcr
= 0.31lP1 .
By studying an expression for the free energy computed in a semiclassical approximation, in Ref. [10] a regularized Hagedorn temperature was proposed for the D = 11
theory, which becomes infinity as the cutoff is sent to zero. The numerical coefficient contains a similar factor 7 (3) as in (4.27). The appearance of this factor in [10] is also related
to the vacuum energy of the worldvolume theory in D = 11.
The critical temperature can be obtained for all values of the coupling by solving
Eq. (4.24) numerically. Fig. 1 is a plot of the critical temperature as a function of geff ,

J.G. Russo / Nuclear Physics B 602 (2001) 109131

125

Fig. 1. Critical
temperature as a function of geff (with = 1). The value at geff = 0 is
TH = 1/(2 2
).

Fig. 2. Critical temperature as a function of gA .

and Fig. 2 is a plot of Tcr as a function of gA . At small couplings, the plots have the
same behavior, since Tcr is approximately constant. At strong coupling, Tcr goes to zero as
1/2
1/3
1/geff in Fig. 1, and as 1/gA in Fig. 2.
of
The natural mass scale in eleven dimensions is lP1 . This means that the energies

elementary excitations of the system at large gA must be proportional to lP , not 1/ . So


temperature at gA  1 is more properly measured in units of lP1 . An analogous situation
happens in type IIB superstring theory. The Hagedorn temperature at gB  1 can be found
by S-duality. The strong coupling limit of type IIB theory is known to be the same theory,

= g
,
where fundamental strings are replaced by D-strings, gB 1/gB and
by D
B
so that the Hagedorn termperature is
TH =

1
1


=
.

2 2
gB
2 2D

The Hagedorn temperature goes to zero for large gB at fixed


, but it should be measured
with
 respect to the D-string tension (since elementary excitations have energies of order

), in which case it is a constant independent of the coupling. In principle, it seems


1/ D
possible to generalize the present method to determine a critical temperature in type IIB
superstring theory at some intermediate coupling gB . In order to connect M-theory at finite
temperature with type IIB theory one needs to study M-theory on an euclidean 3-torus.

126

J.G. Russo / Nuclear Physics B 602 (2001) 109131

Now membranes can wrap in different ways.


4.4. Duality connections
The Hagedorn temperature in string theory was first understood as a consequence of the
exponential growth of the asymptotic level density with the mass, (m) econst m . The
existence of a finite critical temperature at gA  1 can be explained if there is a stringtheoretic description of M-theory in this limit. The tension 1/
of such string can be read
(up to a numerical constant) from the critical temperature. We have the formula:
Tcr2 =

1
const
=
.
28 (3)
geff

(4.30)

How can a string-theory description arise at large R11 ? In this limit, near the critical
temperature we have R0  R11 . Therefore, the relevant low energy degrees of freedom
of the system are more appropriately described by making dimensional reduction along
the Euclidean time direction X0 , with X11 now playing the role of a compact spatial
dimension of the resulting ten-dimensional theory. Because of the antiperiodic boundary
conditions around X0 , the resulting ten-dimensional string theory is a non-supersymmetric
string theory. According to [20], a compactification of M-theory on a circle where fermions
obey antiperiodic boundary conditions gives type 0A string theory (the relation between
finite temperature type IIA theory and type 0A theory was noted already in [1]). Therefore
the strong coupling limit of type IIA superstring theory at finite temperature T would
2 =
be described by euclidean type 0A string theory, where the string coupling is gA
3 3
3 1
2
3
2R0 / lP = (4 T lP ) , and the string tension is obtained from the usual formula
=
lP3 /(2R0 ) = lP3 T =
geff . This agrees with the identification in (4.30). In the strict limit
R11 , this duality implies that uncompactified M-theory at temperature T is described
by a ten-dimensional euclidean string theory.
Reproducing the numerical coefficient in (4.30) using string theory techniques may not
be simple, because the type 0A theory is strongly coupled below the critical temperature,
i.e., gA > O(1) for T < Tcr . However, Eq. (4.30) predicts that the type 0A tachyon must
disappear at a coupling
2
2
gA
> gA
cr =

7 (3)
= 0.85.
2

The precise numerical value may be subject to corrections, for reasons explained in
Section 1. This agrees with the suggestion of [20], that the type 0A tachyon should
become massive at strong coupling. Conversely, the existence of a critical coupling g A cr
in type 0A string theory at which the type 0A tachyon becomes massless implies the
existence of the critical temperature T = O(lP1 ) found in this paper in uncompactified
M-theory at finite temperature. In terms of the critical coupling, the critical temperature is
Tcr = (2gA cr )2/3 lP1 . It would be interesting to investigate further consequences of this
connection in more detail.

J.G. Russo / Nuclear Physics B 602 (2001) 109131

127

Acknowledgement
This work is supported by Universidad de Buenos Aires and Conicet.

Appendix A. Non-holomorphic Eisenstein series


The non-holomorphic Eisenstein series is defined by [16]

2r
1
, r> ,
2 (2r)Er () =
|k + m|2r
2

(A.1)

(k,m)=(0,0)

where = 1 + i2 is a complex parameter describing the upper half complex plane.


The Eisenstein series Er () is invariant under SL(2, Z) transformations of ,
a + b
, ad bc = 1, a, b, c, d Z.
c + d
At large 2 , one has the expansion

(A.2)

Er () = 2r + r 21r
+

1/2
 
42 r  w r1/2
cos(2wn1 )Kr1/2 (2wn2 ),
(2r) (r)
n
n,w=1

(r 1/2) (2r 1)
.
r =
(A.3)
(r) (2r)
The derivation is as in the analogous case given in Appendix B. Using the asymptotic
expansion for the Bessel function Kr1/2 ,


1
(r + m)
1
,
Kr1/2 (2wn2 ) =
e2wn2
m
(4wn2 ) (r m)m!
4wn2
m=0

(A.4)

we see that the terms in (A.3) involving the Bessel function will be exponentially
suppressed. In the present case with 2 = 1/geff , such exponentially suppressed terms
represent non-perturbative contributions originating from D0 branes, whose coefficient
is therefore exactly determined by the above expansion of the Bessel function. In a
strong coupling expansion obtained by the modular transformation 1
the exponentially suppressed terms are instead of the form e2geff .
From Eq. (A.3), we obtain the following expressions for the expansions of Eisenstein
series appearing in Sections 2 and 4:



wn
(3) 2 1/2
8  w
(3)E3/2(geff ) = 3/2 +
g +
K1 2
,
3 eff
geff
n
geff
g
n,w=1

eff

for geff  1,

(A.5)
3/2

(3)E3/2(geff ) = (3)geff +
for geff  1,

1/2
geff

+ 8 geff


n,w=1

w
K1 (2wngeff ),
n
(A.6)

128

J.G. Russo / Nuclear Physics B 602 (2001) 109131

and
256(10) 9/2
geff
315






4 11/2
w 5
wn
+
K5 2
,

n
geff
( 11
2 ) geff n,w=1
11/2

(11)E11/2(geff ) = (11)geff

256(10) 9/2
11/2
(11)E11/2(geff ) = (11)geff +
geff
315



4 11/2 geff  w 5
+
K5 (2wngeff ),
n
( 11
2 )
n,w=1

geff  1,

geff  1.

(A.7)

(A.8)

Appendix B. Generalized Eisenstein series for fermion contributions


In the calculations performed in the main text, fermion contributions (either to the free
energy or to the vacuum energy) led to Eisenstein-type series of the form

2r
1
(1)m
, r> .
2 (2r)Fr ()
(B.1)
2
|k + m|2r
(k,m)=(0,0)

Here we will derive some basic properties that we need, such as the weak and strong
coupling expansions.
Note that Fr () is not SL(2, Z) invariant, so the weak and strong coupling expansions
will be different. In particular, the modular transformation 1/ gives

2r
2 (2r)Fr (1/)
(B.2)
(1)k
.
|k + m|2r
(k,m)=(0,0)

Let us first derive an expansion of (B.1) for 2  1. We will make use of the formulas:


1
= (s),
ns
n=1


(1)n
n=1

ns



= (s) 1 21s .

Separating the m = 0 term in Eq. (B.1), we get


(2r)Fr () = (2r)2r

+ 2r


k m=1

r
(1)
(r)

dx r x|k+m|2
x e
.
x

(B.3)

We now use the Poisson resummation formula,



k

f (k) =



d f ()e2ik .

(B.4)

We get
(2r)Fr () = (2r)2r +

2r r  

(1)m e2ik m1
(r)

m=1

J.G. Russo / Nuclear Physics B 602 (2001) 109131

dx r 1 xm2 2 k
2
2
x .
x 2e
x

129

(B.5)

Separating the term

= 0, and performing the integrations, we finally obtain

(2r)Fr () = (2r)2r + r 21r

 r1/2
1/2

42 r  
m k
+
(1)
(r)
m
k=1 m=1

cos(2km1)Kr1/2 (2km2 ),



(r 1/2)
(2r 1) 1 222r .
r
(r)

(B.6)

For = i/geff , Eq. (B.6) is an expansion which is applicable in the regime geff  1.
An expansion for the opposite regime, geff  1, can be obtained by proceeding in a
2 =
similar way, but separating the k = 0 term in Eq. (B.1). Define = 1/, so that
2
2

2 /|| , 1 = 1 /|| . We get
r

2
(2r)Fr () = (2r) 1 212r



r
dx r x|m+k |
2
x e
(1)m
.

(r)
x
m

2r
+

k=1

(B.7)

Now we make Poisson resummation in m, and then perform the integration over x. We
obtain
2r
(2r)Fr () = (2r)(1 212r )


1/2 r  
|m + 12 | r1/2 2ik(m+ 1 )1
2
2
2
+
e
(r) m
k
k=1



2 .
Kr1/2 2k m + 12

(B.8)

2 . In our case, we have = i/geff and


 = igeff ,
This expansion is applicable for large
so Eq. (B.8) gives an expansion for geff  1. Note that there is only one power-like term,
2 .
and the remaining terms are exponentially suppressed at large
Summarizing, we obtain for r = 3/2 and r = 11/2 the following expansions (cf. (A.5)
(A.8)



km
(3) 2 1/2
8 
mk
g +
K1 2
,
(1)
(3)F3/2 (geff ) = 3/2
6 eff
geff
m
geff
g
eff

k,m=1

geff  1,
3

3/2
(3)F3/2 (geff ) = (3)geff + 4 geff
4
geff  1,

(B.9)


m k=1

|m +
k

1
2|



K1 2geff k|m + 12 | ,
(B.10)

130

J.G. Russo / Nuclear Physics B 602 (2001) 109131

and

256
9/2
1 29 (10)geff
315
 5


4 11/2
m k
+
(1)

m
( 11
2 ) geff k,m=1


km
, geff  1,
K5 2
geff


11/2
(11)F11/2(geff ) = 1 210 (11)geff

2 11/2 geff   |m + 12 |5
+
k5
( 11
m k=1
2 )




K5 2geff k m + 12 , geff  1.

(11)F11/2(geff ) =

(11)
11/2
geff

(B.11)

(B.12)

References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]

J.J. Atick, E. Witten, Nucl. Phys. B 310 (1988) 291.


R. Hagedorn, Nuovo Cimento Suppl. 3 (1965) 147.
E.S. Fradkin, A.A. Tseytlin, Nucl. Phys. B 227 (1983) 252.
S.S. Gubser, S. Gukov, I.R. Klebanov, M. Rangamani, E. Witten, The Hagedorn transition in
non-commutative open string theory, hep-th/0009140.
I. Kogan, JETP Lett. 45 (1987) 709;
B. Sathiapalan, Phys. Rev. D 35 (1987) 3277.
L.J. Dixon, J.A. Harvey, Nucl. Phys. B 264 (1986) 93;
N. Seiberg, E. Witten, Nucl. Phys. B 276 (1986) 272.
M. Bowick, S. Giddings, Nucl. Phys. B 325 (1989) 631.
D. Amati, J.G. Russo, Phys. Lett. B 454 (1999) 207, hep-th/9901092.
S. Chaudhuri, Deconfinement and the Hagedorn transition in string theory, hep-th/0008131.
A.A. Bytsenko, S.A. Ktitorov, Phys. Lett. B 225 (1989) 325;
A.A. Bytsenko, S.D. Odintsov, Phys. Lett. B 243 (1990) 63.
T. Banks, W. Fischler, S.H. Shenker, L. Susskind, Phys. Rev. D 55 (1997) 5112, hepth/9610043.
H. Liu, A.A. Tseytlin, JHEP 9801 (1998) 010, hep-th/9712063.
M.L. Meana, M.A. Osorio, J.P. Pealba, Nucl. Phys. B 531 (1998) 613, hep-th/9803058.
S. Chaudhuri, D. Minic, Phys. Lett. B 433 (1998) 301, hep-th/9803120.
B. Sathiapalan, Mod. Phys. Lett. A 13 (1998) 2085, hep-th/9805126.
A. Terras, Harmonic Analysis on Symmetric Spaces and Applications, Vol. I, Springer-Verlag,
1985.
M.B. Green, M. Gutperle, P. Vanhove, Phys. Lett. B 409 (1997) 177, hep-th/9706175.
J.G. Russo, A.A. Tseytlin, B 508 (1997) 245, hep-th/9707134.
R. Rohm, Nucl. Phys. B 237 (1984) 553.
O. Bergman, M.R. Gaberdiel, JHEP 9907 (1999) 022, hep-th/9906055.
M. Fabinger, P. Horava, Nucl. Phys. B 580 (2000) 243, hep-th/0002073.
E. Bergshoeff, E. Sezgin, P.K. Townsend, Ann. Phys. 185 (1988) 330.
E. Bergshoeff, E. Sezgin, Y. Tanii, Nucl. Phys. B 298 (1988) 187.

J.G. Russo / Nuclear Physics B 602 (2001) 109131

131

[24] B. de Wit, J. Hoppe, H. Nicolai, Nucl. Phys. B 305 [FS23] (1988) 545.
[25] J.G. Russo, in: Y.M. Cho, S. Nam (Eds.), Proc. of the APCTP Winter School, Dualities in Gauge
and String Theories, World Scientific, 1997, hep-th/9703118.
[26] J.G. Russo, Nucl. Phys. B 535 (1998) 116, hep-th/9802090.
[27] J.H. Schwarz, Phys. Lett. B 360 (1995) 13, hep-th/9508143.
[28] M.J. Duff, T. Inami, C.N. Pope, E. Sezgin, K.S. Stelle, Nucl. Phys. B 227 (1988) 515.
[29] K. Fujikawa, J. Kubo, Phys. Lett. B 199 (1987) 75.
[30] B. de Wit, K. Peeters, J. Plefka, Phys. Lett. B 409 (1997) 117.
[31] B. de Wit, M. Lscher, H. Nicolai, Nucl. Phys. B 320 (1989) 135.

Nuclear Physics B 602 (2001) 132146


www.elsevier.nl/locate/npe

The topology of the AdS/CFT/RandallSundrum


complementarity
Brett McInnes
Department of Mathematics, National University of Singapore, 10 Kent Ridge Crescent,
Singapore 119260, Singapore
Received 17 October 2000; accepted 15 March 2001

Abstract
The background geometries of the AdS/CFT and the RandallSundrum theories are locally similar,
and there is strong evidence for some kind of complementarity between them; yet the global
structures of the respective manifolds are very different. We show that this apparent problem can
be understood in the context of a new and more complete global formulation of AdS/CFT. In this
picture, the brane-world arises within the AdS/CFT geometry as the inevitable consequence of recent
results on the global structure of manifolds with infinities. We argue that the usual coordinates give
a misleading picture of this global structure, much as Schwarzschild coordinates conceal the global
form of KruskalSzekeres space. 2001 Elsevier Science B.V. All rights reserved.
PACS: 11.25.-w; 11.25.Hf
Keywords: RandallSundrum; AdS/CFT; Euclidean quantum gravity

1. Introduction
In the best-known special case, the AdS/CFT correspondence [1] is a duality between
quantum gravity on an (n + 1)-dimensional AdS background (for certain n) and a
YangMills superconformal field theory on the n-dimensional boundary of the Penrose
compactification. That this is indeed a special case is emphasised in Ref. [2], where it is
also argued that more general cases are best treated by Euclidean methods. In the Euclidean
picture, the bulk is an (n + 1)-dimensional hyperbolic space H n+1 (topology Rn+1 ); it may
be regarded as the warped product of R with Rn , with metric
g H = dy dy + e+2y/LgijF dx i dx j ,
E-mail address: matmcinn@nus.edu.sg (B. McInnes).
0550-3213/01/$ see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 2 3 - 7

(1.1)

B. McInnes / Nuclear Physics B 602 (2001) 132146

133

where y (, ), L is a constant, gijF dx i dx j is a flat Euclidean metric, and the


Ricci curvature of g H satisfies
 
n
Ric g H = 2 g H .
(1.2)
L
The conformal boundary consists of a copy of Rn at y = + together with a point at
y = , and so it has the topology of S n , the n-sphere, the one-point compactification
of Rn . (Notice that the sections y = constant have a different topology: each is a copy
of Rn .) The metric g H induces a conformal structure rather than a metric on the boundary.
This is the purely geometric aspect of the duality.
The AdS/CFT correspondence has proved to be an extremely useful tool for exploring
otherwise inaccessible aspects of quantum gravity. It is natural to ask whether it is more
than this: can something resembling (an Euclidean version of) the real world arise within
the AdS/CFT geometry? One immediately thinks, of course, of the RandallSundrum
brane-world [3], which sits at the origin of a 5-dimensional space with metric
g RS = dy dy + e2|y|/LgijF dx i dx j .

(1.3)

Indeed, evidence for some kind of AdS/CFT/RandallSundrum complementarity has


been adduced [416]. For example, in Ref. [10] the corrections to the Newtonian
gravitational potential are computed from both points of view, and are found to have
the same form. The agreement becomes precise provided that the curvature parameter L
in the RandallSundrum metric is equated to its counterpart in g H again strongly
suggesting that we are dealing here with two aspects of one geometry. A clearer and more
concrete interpretation would be that we are dealing with two different regions of a single
Riemannian manifold with curvature parameter L.
While g H and g RS are obviously closely related, the global structures of the respective
manifolds are of course very different; and these differences are crucial. First, the Randall
Sundrum space has no access to any region analogous to the region of H 5 where y
takes large positive values, and this is essential for trapping gravity on the braneworld. Second, this excision of the large end of H 5 results in a manifold which is not
geodesically complete. Again, this incompleteness (due to the jump at y = 0) may well
play an essential role in gravity trapping (see [14,16], and the references therein). Thirdly,
as was pointed out above, the topology of the conformal boundary of H 5 is that of S 4 ,
and thus is very different to that of the brane-world; this is no mere technical bagatelle, for
the compactness of the boundary ensures the uniqueness of the correlation functions of the
conformal field theory defined there.
Despite these global differences, there is no denying the appeal of the notion that
the brane world results from a UV cutoff of AdS/CFT, with a corresponding non-trivial
coupling of the CFT to gravity [4]. From this point of view, AdS/CFT should be regained
from the RandallSundrum space by some kind of limiting process. But what is this
process? How can it resolve the global differences discussed above? The metrics g H
and g RS appear to have no adjustable parameter other than L; how can we introduce a
new parameter into g RS , such that g H is obtained in the limit? What is the geometric
significance of this new parameter?

134

B. McInnes / Nuclear Physics B 602 (2001) 132146

We wish to argue that these questions are best answered in the context of a radical
reformulation of the geometric background of AdS/CFT. The fact that the familiar
bulk/boundary formulation of AdS/CFT must be generalised in some way is already
recognised in [2]: ultimately, we expect a fully general AdS/CFT correspondence to relate
gauge theory on an arbitrary compact manifold to string theory on a related higherdimensional Einstein manifold of negative scalar curvature. However, as we move beyond
simple manifolds like S n and S 1 S n1 , we soon encounter manifolds of non-zero
Hirzebruch signature, and these simply cannot be represented as the boundary of any
manifold-with-boundary [17]. A familiar example of this is the K3 manifold. Even if we
leave this problem to one side, there are serious practical difficulties in carrying out this
programme. For example, every four-dimensional compact flat manifold is the quotient of
the torus T 4 by some finite but possibly very complicated group. All such manifolds can
be represented as the boundary of some five-dimensional manifold-with-boundary, but this
fact is established in an abstract way using index theory, not by a concrete construction;
to exhibit an Einstein metric on the interior of this space (inducing on the boundary the
conformal structure corresponding to the given flat metric) is, in general, a formidable
task indeed. We have shown elsewhere [18] that the first of these problems is solved
by interpreting infinity as a compact submanifold of a compact manifold rather than
as the boundary of a manifold-with-boundary. (As we shall explain below, cobordism
theory indicates that this is really the only natural way of dealing with this problem.) The
bulk is then the (non-compact) complement of the infinity hypersurface. This is indeed
a generalisation of the usual approach; for while it is essential if we are to deal with nonboundary manifolds like K3, it also works perfectly for manifolds like spheres and tori
which happen to be boundaries. The second problem is then automatically solved, since,
as we shall show here, it is quite easy to exhibit Einstein metrics defined on the bulk if the
latter is interpreted in this way.
If we accept that this is the correct geometric formulation of AdS/CFT, then a pair of
beautiful results due to Witten and Yau [19] and Cai and Galloway [20] essentially imply
the following remarkable result: if we require the bulk to be Einstein with negative scalar
curvature, then the bulk cannot be geodesically complete. Something like the brane-world
arises in the bulk not because it has been inserted by hand, but because the geometry
requires it to do so. The Einstein condition on the bulk strongly constrains the local
geometry, and yet, by forcing us (via incompleteness) to cut and paste, it opens up a
host of interesting new global structures for the higher-dimensional space. A very simple
and natural example of this construction will allow us to represent the brane-world as
residing in a space which is essentially connected to the AdS/CFT region by a wormhole.
The solution contains, as so often, negative-tension branes, but these are confined to the
throats of the wormhole where, one might say [21], they belong. The parameter which
controls the extent to which the brane-world resembles AdS/CFT infinity is the width of
the wormhole, or, equivalently, the tension of the wormhole branes; the topology of infinity
is always identical to that of the brane world; and there is a clear and precise sense in which
the AdS/CFT picture is obtained from the brane world as the parameter tends to its limiting
value. Finally, the wormhole approach can be reconciled with the usual bulk/boundary

B. McInnes / Nuclear Physics B 602 (2001) 132146

135

formulation of AdS/CFT by taking another limit of this parameter, in which the wormhole
effectively pinches off and converts the infinity hypersurface to an effective boundary,
destroying all the branes and restoring geodesic completeness. Thus, the new approach
does not completely replace the standard one: instead it reveals the latter as a special,
limiting case of a more general geometry in which the bulk is generically non-smooth. In
short, the reformulation advocated here allows us to understand the differences between
the AdS/CFT picture and the RandallSundrum cosmology, and provides a suitable arena
for discussing their relationship.
We begin with a rapid summary of the relevant geometric ideas.

2. Spin cobordism and Einstein metrics


Let N n be a smooth, compact, n-dimensional submanifold of a compact (n + 1) n+1 . Let f be a smooth function on M
n+1 which is positive on
dimensional manifold M
n+1 N n but which vanishes to first order on N n . Suppose
the complement M n+1 = M
M
that we can find a metric g on M n+1 and a function f as above, such that f 2 g M extends
continuously to N n . Then N n is at infinity for points in the bulk, M n+1 . Any instance
of the usual bulk/boundary formulation of AdS/CFT can be interpreted in this way by
simply taking two copies and identifying them along the boundary. A more interesting
construction results, however, when one considers how to deal with a CFT defined on a
manifold which is not a boundary.
Let P n and Qn be compact n-dimensional spin manifolds. Let Qn be obtained from
Qn by reversing oriental ion, and let P n + (Qn ) be the disjoint union. One says that P n
and Qn are spin cobordant [22] if there exists at least one compact manifold-with-boundary
n+1 with P n + (Qn ) as boundary and with an interior R n+1 which is spin and which
R
induces the given spin structures on P n and Qn . The spin cobordism equivalence classes
spin
spin
in dimension n form an Abelian group, the spin cobordism group n . In particular, 4
is not trivial: there are (infinitely) many distinct 4-manifolds which are not spin boundaries.
Let N n be a compact manifold on which we wish to investigate the generalised AdS/CFT
correspondence [2], but suppose that N n is not spin cobordant to zero, so it is not the spin
spin
boundary of any manifold-with-boundary. The fact that n is a group points the way to
a solution of the problem, for clearly the class represented by N n has an inverse: that is,
the non-boundary N n can always be cancelled by another non-boundary, say P n . That is,
n+1 with N n + (P n ) as boundary. Let g M be
there exists a manifold-with-boundary M
a metric on the interior, M n+1 , such that N n + (P n ) is the conformal boundary relative
to g M ; thus we have represented N n as one connected component of the boundary, which
is the best we can do. However, there are strong objections to disconnected boundaries
in the generalised AdS/CFT correspondence [19], so this procedure only exchanges one
problem for another. The only way to avoid this conclusion is to pick P n to be N n itself,
 n+1 (without
and then perform a topological identification, obtaining a compact manifold M
n
a boundary) which contains N as a hypersurface at infinity. The fact that N n is not a
boundary is no longer relevant. Clearly this is a simple and compelling way of dealing

136

B. McInnes / Nuclear Physics B 602 (2001) 132146

with non-boundary manifolds; it is hard to see how else they can be handled. Furthermore,
we shall see below that the more familiar bulk/boundary formulation of AdS/CFT can be
regarded as a limiting case of this geometry. In view of this, we advocate the hypersurface
interpretation of infinity in all cases, even when N n is a boundary.
Now let us consider the geometry of the compact manifold N n . It is clear that, because
the function f introduced above is not unique, the metric g M on the bulk, M n+1 , does not
induce a Riemannian structure on N n , but rather a conformal structure. We may therefore
take it, without loss of generality [23], that the relevant metric on N n has constant scalar
curvature. It was pointed out in [24] that this constant must not be negative, since otherwise
the conformal field theory at infinity will become unstable. Now this constraint on N n
has extraordinary consequences for the bulk, M n+1 . It turns out [18] that if [a] the compact
 n+1 is constructed via the spin cobordism argument given above, [b] the bulk
manifold M
M n+1 is orientable, [c] N n has positive or zero scalar curvature, and [d] the metric g M on
the bulk has Ricci tensor Ric(g M ) = Ln2 g M , then M n+1 cannot be geodesically complete.
This follows from the very remarkable work of WittenYau [19] and CaiGalloway [20].
In simple examples, the incompleteness manifests itself as one or more regions where g M
fails to be smooth. In other words, in this formulation of AdS/CFT, RandallSundrum
style jumps are perfectly natural indeed, they are inevitable. Physically, this means that
a fully general formulation of AdS/CFT necessarily entails the presence of branes in the
bulk as foreseen in [2]. Mathematically, it means that, when constructing bulk solutions
of Ric(g M ) = Ln2 g M , we are free of the (very restrictive) condition of smoothness, and
this allows new and interesting geometries for the bulk.
The simplest possible cobordism from N n to itself is given by a cylinder. Identifying the
 n+1 = S 1 N n , where S 1 is a circle. Even in this simplest possible
two copies, we obtain M
case, there are many interesting geometric structures, so we shall confine our investigation
to this case. Again, we stress that even if N n is a boundary, there are definite advantages
in constructing the bulk in this way, since, as we shall see, it is easy to construct Einstein
metrics on the bulk even if the metric of N n is very complex or not known explicitly. By
contrast, this could be very difficult in the usual bulk/boundary approach.
Parametrise the circle by running from to + , and think of infinity as residing
at = 0. We shall look for metrics of the general form


g M = A2 ( ) L2 d d + B 2 ( )g N ,

(2.1)

where L is a constant, g N is a Riemannian metric on N n , B( ) is a positive function which


does not tend to zero as tends to zero, and A( ) is a positive function such that 1/A( )
has a first-order zero at = 0. Then g M will be a metric on the non-compact bulk,
M n+1 = [S 1 {0}] N n , such that points in the hypersurface = 0 are infinitely far
from any point in the bulk. As we have stressed, the compact manifold at infinity can have
an otherwise arbitrary topology it could be the K3 manifold, for example.
Our objective is to solve the bulk Einstein equation, Ric(g M ) = Ln2 g M . In this we
will be guided by the following lemma, the proof of which is straightforward and will be
omitted.

B. McInnes / Nuclear Physics B 602 (2001) 132146

137

Lemma. With the above notation, g M is an Einstein metric only if g N is an Einstein


metric.
In view of the requirement that the scalar curvature of N n should not be negative if the
CFT at infinity is to be stable, this means that we have two cases to consider. The more
interesting of the two is the Ricci-flat case, so we begin with it.

3. Ricci-flat infinity
We now set n = 4 and take N 4 to be Ricci-flat: thus N 4 could be a four-torus T 4 (in
 5 is T 5 ), or a quotient of T 4 by a
which case the compact five-dimensional manifold M
finite group, or K3, and so on. As we know, there are no smooth solutions of Ric(g M ) =
Ln2 g M of the kind we require, and we must bear this in mind when solving the equation;
we must do it piecewise.
When N 4 is Ricci-flat, the solutions have the following local form:
A( ) = 1/(K ),

B( ) = J,

(3.1)

where K and J are constants. The global solution has therefore to be patched together from
pieces of this form, so that A( ) and B( ) are continuous functions on S 1 {0}, that is,
on the set (, 0) (0, ]. An interesting global solution of this kind is the following:


1 2

L d d + (2 )2 g RF ,
(0, ] [, 0),

 2

1
L d d + (2 )2 g RF , [, ],
gM =
(3.2)
2
(2 )

 2

1

L d d + (2 )2 g RF , (, ].
(2 + )2
Here is a constant angle in the range (/2, ), and g RF is Ricci-flat. Note that the
introduction of the parameter is the inevitable consequence of the singularity theorem
discussed earlier: it describes the relative positions of the points on the circle where the
metric fails to be smooth.
5 = S 1 N 4 , except at = 0, which is
Clearly g M is well-defined everywhere on M
M
infinity. However, g does induce a conformal structure on infinity, represented by the
Ricci-flat metric g RF . Elsewhere, g M is smooth except at = , = , and, perhaps a
little less obviously, at = . At = there is a positive-tension brane which we interpret
as the brane-world; the topology is that of N 4 , and the metric induced by g M is precisely
g RF . As we move away from the brane-world (in either direction), the sections = constant
rapidly shrink, the induced metric being g RF multiplied by the factor (2 )2 /(2 )2 ,
which is less than unity. This continues until = is reached, where the scaling factor
reaches the value (2 )2 / 2 ; beyond this, the sections increase in size without limit as
= 0 is approached. If is assumed to be only slightly larger than /2, then (2 )2 / 2
is positive but very small. The physical picture is then as follows. We have a Randall
Sundrum brane-world at = , with the manifold rapidly shrinking away to either side.

138

B. McInnes / Nuclear Physics B 602 (2001) 132146

The narrow region, however, proves to be the entrances of a wormhole with negativetension branes at the throats; the other side of the wormhole is an infinite region of the kind
familiar in AdS/CFT theory. Thus, the RandallSundrum world can indeed arise naturally
in an AdS/CFT geometry: it lives deep inside a wormhole. The extent to which each
world can influence the other is controlled by the parameter , which determines the
width of the wormhole throats.
Let us examine the ways in which this global structure addresses the questions raised
in the Introduction. First, the geometry does clearly reconcile the differing natures of the
RandallSundrum and AdS/CFT geometries; the former arises within a wormhole in the
latter; this explains the equality of the L parameters on the two sides of the calculation
in [10]. Second, the lack of smoothness in the RandallSundrum geometry is now
explained by AdS/CFT it is an inevitable consequence of the Witten/Yau/Cai/Galloway
theorems. Third, the topology of the brane world ( = ) is the same as that of AdS/CFT
infinity ( = 0): both are diffeomorphic to N 4 , and both have essentially the same
geometry (given by the Ricci-flat metric g RF on the brane world, and by the conformal
structure represented by g RF at = 0). This is in contrast to the usual picture, in which
infinity is S 4 and thus has a very different topology and geometry to the brane-world.
Thus, in the new picture, it is at least topologically possible to imagine a limiting process
whereby the brane-world comes to resemble the AdS/CFT picture. Before discussing that,
however, we wish to point out the following simple yet very appealing property of the
above geometry. Notice that there is an obvious asymmetry between the AdS/CFT and the
RandallSundrum sides of the wormhole: on one side the manifold flares out to infinity,
while on the other it begins to do so but is abruptly halted at the brane-world. This happens
simply because is bounded values beyond are meaningless. Thus we can say that
the UV cutoff implicit in the brane-world is a consequence of the topology of the fivedimensional world.
Now let us ask what happens if (which always lies in the range /2 < < ) is taken
to be arbitrarily close to /2. Note that the brane-world, at = , has a metric g RF which
2
does not depend on , while the metric at = is (2)
g RF . For an observer at the
2
throat, therefore, the brane-world tends to become infinitely large as tends down to /2.
The distance d(, ) from any point at a fixed value of [, ] to the brane world is

2
,
d(, ) = L ln
(3.3)
2
and this becomes arbitrarily large; the brane-world recedes to infinity. In short, by taking
sufficiently close to /2, we can effectively (though not truly) isolate the brane-world
from AdS/CFT infinity; by taking it still closer to /2, we can cause the interior of the
wormhole to resemble its exterior. Thus we have a precise formulation of the notion that
AdS/CFT can be recovered from the RandallSundrum cosmology by some continuous
process. The relevant parameter is essentially the width of the wormhole throats, or, if one
prefers, the magnitude of the tension of the branes there (which is also related to 2 ).
The reader is entitled to object that the metric (3.2) bears no obvious relation to the
RandallSundrum metric (1.3) or to the Euclidean AdS metric (1.1). Before we explain

B. McInnes / Nuclear Physics B 602 (2001) 132146

139

this, the reader may wish to reflect on the respective roles played by Schwarzschild
coordinates and KruskalSzekeres coordinates in the study of the Schwarzschild solution
in general relativity. The former have their uses, but they are profoundly misleading
regarding the true global structure of the underlying spacetime manifold [25]. In particular,
they conceal the wormhole structure of the full spacetime. We claim that the familiar
coordinate y does a similar disservice to the study of (1.1) and particularly (1.3). To see
the point, take (3.2) and change the coordinate as follows. If [, ], define y by

y/L
e
(3.4)
, y [0, y ], y = L ln
.
=
2
2
A straightforward calculation reveals that, for these values of , the metric (3.2) becomes
g M = dy dy + e2y/Lg RF .

(3.5)

If (, ], define y by
2 +
, y [y , 0).
2
In this range, (3.2) becomes
ey/L =

(3.6)

g M = dy dy + e2y/Lg RF ,

(3.7)

and so, for y [y , +y ] we have


g M = dy dy + e2|y|/Lg RF ,

(3.8)

and so we have recovered the RandallSundrum metric, except that the flat metric g F
in (1.3) has been replaced by a Ricci-flat metric g RF . (The fact that this replacement does
not affect the Einstein equation Ric(g M ) = L42 g M is explained in [26, p. 268].)
For (0, ], define y by

2
ey/L, y [y , ),
=
(3.9)
2
and for [, 0) define it by

2
e+y/L, y (, y ].
=
2

(3.10)

Then (3.2) becomes




g M = dy dy + e+2|y|/L e4y /L g RF ,

y (, y ] [y , ).

(3.11)

Bearing in mind the fact that e4y /L g RF is Ricci-flat since g RF is Ricci-flat, we essentially
have here two copies of the (Ricci-flat generalisation of) the Euclidean AdS metric (1.1).
If we take the manifold N 4 to be the torus T 4 , so that the five-dimensional world is T 5 ,
and if we take g RF to be a flat metric on T 4 , then (3.8) is locally identical to (1.3), and
(apart from a constant scaling factor) (3.11) is locally identical to (1.1). The true global
structure of this space is, however, utterly unlike the picture suggested by (3.8) and (3.11).
In particular, (3.11) very misleadingly suggests that infinity (at y = and y = +) is

140

B. McInnes / Nuclear Physics B 602 (2001) 132146

disconnected, which would lead directly to the paradox discussed by Witten and Yau [19].
In fact, of course, infinity is connected. (Let y in (3.9) and let y in (3.10).)
Our claim, then, is that the familiar y coordinate in the Euclidean AdS metric (1.1) and
in the RandallSundrum metric (1.3) is akin to Schwarzschild coordinates: very useful
for many applications, but profoundly misleading as an indicator of global structure. The
analogue of KruskalSzekeres coordinates here is the circular coordinate , and the global
structure is to be read off from (3.2). We have found that this global structure accounts for
AdS/CFT/RandallSundrum complementarity in a way that answers all of the questions
raised in the Introduction.

4. Ricci-positive infinity
The other potentially interesting solutions of Ric(g M ) = L42 g M of the form (2.1) are
obtained when N 4 is a compact Einstein manifold of positive scalar curvature. In fact, the
metric g P on N 4 must be normalised to satisfy
 
3
Ric g P = 2 g P .
(4.1)
L
As in the Ricci-flat case, we know that there are no geodesically complete solutions, so we
must be prepared to patch a global solution together from local pieces. A local solution of
the form (2.1) is given by




1
cosec K 2 ,
B( ) = 2 cos K 2 ,
(4.2)
2
where K is a constant, and so we can obtain a global solution on (, 0) (0, ] of the
following kind:

 
  
cosec2 2 14 L2 d d + cos2 2 g P ,
(0, ] [, 0),






g M = cosec2 2 14 L2 d d + cos2 2 g P , [, ],
(4.3)







cosec2 + 2 14 L2 d d + cos2 + 2 g P , (, ],
A( ) =

where is a constant angle in (/2, ). This is a solution of Ric(g M ) = L42 g M provided


g P satisfies (4.1). Clearly g M is well-defined and continuous everywhere in the bulk,
and it induces the conformal structure represented by g P on infinity, which is at = 0.
This solution shares many of the virtues of (3.2): there is a brane-world at = , deep
inside a wormhole with throats at = ; the brane-world has the same topology as
infinity; as before, the characteristic RandallSundrum lack of smoothness is explained
by AdS/CFT, via the Witten/Yau/Cai/Galloway theorems; and so on. In (4.3), the metric
at = is cot2 ( 2 )g P , which tends to g P as is taken closer to /2. On the other hand,
the metric induced on the brane-world is tan2 ()g P , which increases without bound as
decreases towards /2. Relative to the length scales used by an observer on the braneworld, therefore, the throats of the wormhole shrink to zero size in this limit; the distance
from them to the brane-world tends to infinity; as before, the interior of the wormhole

B. McInnes / Nuclear Physics B 602 (2001) 132146

141

comes to resemble its exterior, in a clear-cut way; there is a clear sense in which AdS/CFT
is a limit of the brane-world.
If we choose N n to be the four-sphere S 4 with its usual metric g S of constant curvature
1/L2 , then (4.3) has another very interesting limit, namely that in which tends to . In
this limit, the brane-world shrinks to zero size, the negative-tension branes coincide with
the brane-world, and the wormhole pinches off. Surprisingly, however, the metric becomes
more rather than less smooth as this occurs. To see this, for [, ], define a coordinate x
by


2 tan 2 =

L
2x x

.
2x x
L

(4.4)

Here x [x , x ], where x and x are the values of x at = , respectively. Then (4.3)


becomes
g


2x x 2 S
=
g .

2 2 dx dx +
L
1 2x x
4

(4.5)

This represents a piece of hyperbolic space, H 5 , corresponding to one side of the bulk
adjacent to the brane-world. For [0, ], define x by
  L x
2 tan 2 = ,
x
L

x [x , L],

(4.6)

and now (4.3) is


g



x2 S
=
dx dx + 2 g ,
2 2
L
1 Lx 2
4

(4.7)

representing another piece of H 5 , in this case extending to infinity at x = L. As before,


these coordinates put (4.3) into a more familiar form, at the cost of completely concealing
the topology of the underlying five-dimensional space which, in this case, is that of
S 1 S 4 . Notice however that if tends to , then x tends to zero, meaning that (4.7) is
valid over the whole range of values of x from zero to L, and so, as claimed, we obtain all
of H 5 , with its usual complete metric. Since the wormhole has pinched off, the sphere at
infinity may now be regarded as a boundary rather than merely as a hypersurface. [To be
precise, we now have two copies of H 5 identified along their conformal boundaries.] The
usual bulk/boundary geometry is now seen as a limiting case of the wormhole geometry.
[Strictly speaking, we cannot let the wormhole actually pinch off, since that would change
the topology; however, we can take so close to that, from a physical point of view, the
wormhole has pinched off for all practical purposes.]
We conclude this section with the observation that there are, of course, many more
solutions of Ric(g M ) = L42 g M compatible with the topology S 1 N 4 . For example, (4.3)
may be replaced by a metric having the form

142

gM

B. McInnes / Nuclear Physics B 602 (2001) 132146



cosec2 ( ) L2 d d + cos2 ( )g P , (0, ] [, 0),

 2


d + cos2 (2 )g P , [, ],

cosec (2 ) L d


= cosec2 (2 2 + ) L2 d d + cos2 (2 2 + )g P , [, ], (4.8)




cosec2 (2 + ) L2 d d + cos2 (2 + )g P , [, ],




cosec2 (2 2 ) L2 d d + cos2 (2 2 )g P , (, ],

where and are constant angles which must satisfy

(4.9)
<< ,
< < 2.
4
2
2
This metric corresponds to a wormhole with throats at = . The throats can be made
arbitrarily narrow by choosing sufficiently close to /2. The wormhole contains two
brane-worlds, at = ; these can be made arbitrarily large by taking sufficiently close
to 2. The two brane-worlds are separated by a further throat at = , where there is
another negative-tension brane. It is clear that one can construct arbitrarily complicated
solutions in this way: the point to bear in mind is that, since it is impossible to construct a
smooth solution, there is nothing unnatural about metrics like (4.8).

5. Conclusion
The familiar bulk/boundary formulation of AdS/CFT must obviously be generalised to
accommodate manifolds (such as K3 ) which are not boundaries. It was predicted in [2]
that doing this will require the introduction of branes or stringy impurities of some kind
into the bulk. This indeed proves to be the case [18].
The objective of this work has been to argue that Wittens stringy impurities include
our Universe. We saw that, within a framework sufficiently general to allow for nonboundary manifolds, an Einstein metric on the bulk cannot be geodesically complete. We
are led to consider 5-dimensional compact manifolds with metrics built up from pieces
of the form (2.1). An example of an Einstein metric constructed so that AdS/CFT can be
explored for K3 is given by (3.2). We find that the generalised AdS space contains a
wormhole, inside which there naturally arises a RandallSundrum type brane surrounded
by a region with the familiar metric (3.8). Finally, we have seen that the wormhole approach
does not require us to jettison the standard bulk/boundary formulation of AdS/CFT.
Instead, the latter is now seen to be a special limiting case, the case in which the wormhole
effectively pinches off, destroying the branes, thereby restoring geodesic completeness
and effectively converting the infinity submanifold to a boundary. [This pinching off is
reminiscent of the constructions discussed in [27].]
We do not, of course, claim to have a realistic cosmological model here. In order
to construct such a model, one would have to begin by understanding the Lorentzian
analogues of the WittenYau and CaiGalloway results. [For ideas on how this might be
done, using techniques given in [28], see [29].] It is not unreasonable to hope that, in
the spirit of [2], our investigations may serve as a useful guide. In particular, our results
suggest:

B. McInnes / Nuclear Physics B 602 (2001) 132146

143

[a] That the RandallSundrum coordinate y may conceal considerable topological


complexity in the underlying 5-dimensional manifold. (See also [30].)
[b] That wormholes in the AdS/CFT bulk may be important. (See also [31].)
[c] That the AdS/CFT/RandallSundrum complementarity should be interpreted as
evidence that AdS/CFT explains the existence of the brane-world that is, of our
Universe.
The RandallSundrum geometry strikes many theorists as somewhat contrived: how
could such a peculiar structure arise? The answer advanced here is that, if wormholes
develop in the AdS/CFT bulk, then a structure like the brane-world must develop inside the
wormhole. On the other hand, the picture we have developed here does have two principal
drawbacks. (1): One of the advantages of the second RS scenario is that, unlike the
first [32], it does not involve negative-tension branes. Here we have reintroduced them, at
the wormhole throats. (2): We have assumed that the angle is slightly larger than /2.
Why should that be so?
Regarding point (1): any in-depth discussion of brane tensions must take into account
the remarkable results of [33], which shows that correctly identifying brane tensions in
the string context can be a very subtle matter. Leaving those issues to one side, however,
we feel that Visser and Barcelo [34] have argued very persuasively that violations of the
familiar positivity conditions on energy densities and tensions are virtually inevitable when
scalar fields are present as they are when one attempts to make brane-world models
more realistic [35]. Indeed, it is argued in [35] that RandallSundrum style kinks can
be smoothed if all tensions are positive [though it is questionable whether even this can
be achieved in a fully supersymmetric treatment [14]]. Since the kinks in our model
certainly cannot be smoothed, we take this as further evidence that negative-tension branes
are unavoidable in our approach as indeed one would expect in any theory involving
wormholes [21].
In fact, it is very questionable whether the traditional aversion to negative tensions is
really justified here. It is argued in [35] that negative tension branes are acceptable if
they lie on spacetime boundaries generated by taking a quotient of some manifold by
a symmetry group which has fixed points. In our picture, the negative tension branes
lie on the boundaries of two such spaces [one containing the brane-world, the other
containing AdS/CFT infinity] which have been sewn together along their boundaries, so
that the negative branes correspond to internal boundaries [27]. Therefore, our negative
branes are like those in [35]: . . . just part of a background, not something which can be
dynamically created anywhere in space. There is however one difference: the negative
tension branes of [35] are at the end of the Universe, so we cannot ask what happens
if some object passes through them in either direction. Our negative branes lie between
two regions of the five-dimensional world, so for us this is potentially very important. We
return to this below.
Whether or not our negative branes are as innocuous as we claim, they are in any case
hidden in the throats of the wormhole; if the latter are sufficiently narrow and far away,
then the negative branes are effectively inaccessible. This brings us to the second point
raised above, the fact that is so close to /2. We can show that this is indeed the case

144

B. McInnes / Nuclear Physics B 602 (2001) 132146

as follows. Eq. (4.1), together with the fact that the Ricci tensor is invariant under constant
re-scalings of the metric, implies that




3
Ric tan2 ()g P = 2 cot2 () tan2 ()g P .
L
Here tan2 ()g P is the metric induced on the brane-world by the metric (4.3). Thus the
cosmological constant of the brane-world is L32 cot2 () [unless it is exactly zero, as in
Section 3]. This is positive and very small, as observations suggest, only if is slightly
larger than /2.
Since is indeed so close to /2, the throats of the wormhole are very narrow and very
far away from the brane-world. As the RandallSundrum scenario is, despite appearances
to the contrary, entirely local [36], the existence of the wormhole cannot be detected.
For example, the presence of the wormhole will not affect the detailed comparison made
in [10], with which we began. [Note incidentally that [10] uses the exact AdS/CFT relation
N 2 = L3 /2G5 , which is only consistent if the cutoff is taken at a large value of y in (1.1);
as we have seen, this means, once again, that is slightly larger than /2.] In other words,
despite the fact that the real structure of the space used in this work is very different to the
one used in [5,6,9], and [10], it leads to essentially the same results.
Can our approach ever lead to something quantitatively new? We believe that the answer
is yes, because we hope that it will eventually be possible to understand the value of in a
dynamical way. That is, possibly was not close to /2 in the early Universe indeed,
the formula for the cosmological constant, L32 cot2 (), suggests that was originally quite
small, if the Universe had an inflationary period [37]. In that case, in the early Universe the
throats of the wormhole were nearby and relatively large, so that access to the AdS/CFT
region was not impeded. The consequences of this might well have been profound. To
analyse them, we will need a good understanding of the dynamics of moving, asymmetrical
branes [38], as well as of the precise ways in which signals can be transmitted from one
side of a negative brane to the other. This latter problem will be particularly challenging,
but very recent advances in the relevant mathematical techniques [39] offer hope that
the wormhole picture will eventually make some specific predictions related to the early
Universe. Even at this point one can see the kind of influence the AdS/CFT parent universe
can exercise on the brane-world, as follows. The topology of the brane-world must be
related [via cobordism] to that of the AdS/CFT world. The AdS manifold itself can be
generalised to other negatively curved non-compact spaces with other topologies, but many
possible topologies are forbidden by stability considerations [40]. Thus, the existence of
the wormhole will directly constrain the topology of the brane-world.
The overall picture presented here may therefore be described as follows. One can
imagine that spacetime foam in the AdS/CFT background could lead to the formation
of a wormhole. Inside this wormhole, the Universe forms as a positive-tension kink. At
this early stage it is strongly influenced by the parent universe, but the dynamics causes
to approach /2, effectively cutting off the RandallSundrum space from the parent. The
smallness of the observed cosmological constant is now simply a reflection of the fact that
the wormhole throats are far away and very narrow. The only current traces of the AdS/CFT
parent are to be found in relics including cosmic topology of the very early Universe.

B. McInnes / Nuclear Physics B 602 (2001) 132146

145

References
[1] O. Aharony, S.S. Gubser, J. Maldacena, H. Ooguri, Y. Oz, Large N field theories, string theory
and gravity, Phys. Rep. 323 (2000) 183, hep-th/9905111.
[2] E. Witten, Anti-de Sitter space and holography, Adv. Theor. Math. Phys. 2 (1998) 253, hepth/9802150.
[3] L. Randall, R. Sundrum, An alternative to compactification, Phys. Rev. Lett. 83 (1999) 4690,
hep-th/9906064.
[4] E. Witten, http://online.itp.ucsb.edu/online/susy_c99/discussion/.
[5] H. Verlinde, Holography and compactification, Nucl. Phys. B 580 (2000) 264, hep-th/9906182.
[6] S.S. Gubser, AdS/CFT and gravity, hep-th/9912001.
[7] C.S. Chan, P.L. Paul, H. Verlinde, A note on warped string compactification, Nucl. Phys. B 581
(2000) 156, hep-th/0003236.
[8] S. Nojiri, S.D. Odintsov, S. Zerbini, Quantum (in)stability of dilatonic AdS backgrounds
and holographic renormalization group with gravity, Phys. Rev. D 62 (2000) 064006, hepth/0001192.
[9] S.W. Hawking, T. Hertog, H.S. Reall, Brane new world, Phys. Rev. D 62 (2000) 043501, hepth/0003052.
[10] M.J. Duff, J.T. Liu, Complementarity of the Maldacena and RandallSundrum pictures, hepth/0003237.
[11] S. Nojiri, S.D. Odintsov, Brane world inflation induced by quantum effects, Phys. Lett. B 484
(2000) 119, hep-th/0004097.
[12] S. Nojiri, O. Obregon, S.D. Odintsov, (Non-)singular brane-world cosmology induced by
quantum effects in d5 dilatonic gravity, hep-th/0005127.
[13] L. Anchordoqui, C. Nuez, K. Olsen, Quantum cosmology and AdS/CFT, hep-th/0007064.
[14] M.J. Duff, J.T. Liu, K.S. Stelle, A supersymmetric type IIB RandallSundrum realization, hepth/0007120.
[15] R. Akhoury, A variational principle for radial flows in holographic theories, hep-th/0007041.
[16] J. Maldacena, C. Nuez, Supergravity description of field theories on curved manifolds and a
no go theorem, hep-th/0007018.
[17] J.W. Milnor, J.D. Stasheff, Characteristic Classes, Annals of Mathematics Studies, Vol. 76,
Princeton Univ. Press, 1974.
[18] B. McInnes, AdS/CFT for non-boundary manifolds, JHEP 0005 (2000) 025, hep-th/0003291.
[19] E. Witten, S.T. Yau, Connectedness of the boundary in the AdS/CFT correspondence, hepth/9910245.
[20] M. Cai, G.J. Galloway, Boundaries of zero scalar curvature in the AdS/CFT correspondence,
hep-th/0003046.
[21] C. Barcelo, M. Visser, Brane surgery: energy conditions, traversable wormholes, and voids,
Nucl. Phys. B 584 (2000) 415, hep-th/0004022.
[22] H.B. Lawson, M.L. Michelsohn, Spin Geometry, Princeton Mathematical Series, Vol. 38,
Princeton Univ. Press, 1989.
[23] R. Schoen, Conformal deformation of a Riemannian metric to constant scalar curvature, J. Diff.
Geom. 20 (1984) 479.
[24] N. Seiberg, E. Witten, The D1/D5 system and singular CFT, JHEP 9904 (1999) 017, hepth/9903224.
[25] R.M. Wald, General relativity, Chicago Univ. Press, 1984.
[26] A.L. Besse, Einstein Manifolds, Springer-Verlag, 1987.
[27] K. Kirklin, N. Turok, T. Wiseman, Singular instantons made regular, hep-th/0005062.
[28] G.J. Galloway, K. Schleich, D.M. Witt, E. Woolgar, Topological censorship and higher genus
black holes, Phys. Rev. D 60 (1999) 104039, gr-qc/9902061;
See also hep-th/9912119.

146

B. McInnes / Nuclear Physics B 602 (2001) 132146

[29] E. Witten, http://online.itp.ucsb.edu/online/susy_c99/witten/.


[30] H. Boschi-Filho, N.R.F. Braga, Quantum fields in anti-de Sitter spacetime and degrees of
freedom in the bulk/boundary correspondence, hep-th/0009039.
[31] S. Nojiri, S.D. Odintsov, K.E. Osetrin, Dilatonic quantum multi-brane-worlds, hep-th/0009059.
[32] L. Randall, R. Sundrum, A large mass hierarchy from a small extra dimension, Phys. Rev.
Lett. 83 (1999) 3370, hep-ph/9905221.
[33] M. Cvetic, M.J. Duff, J.T. Liu, H. Lu, C.N. Pope, K.S. Stelle, RandallSundrum brane tensions,
hep-th/0011167.
[34] M. Visser, C. Barcelo, Energy conditions and their cosmological implications, gr-qc/0001099.
[35] O. DeWolfe, D.Z. Freedman, S.S. Gubser, A. Karch, Modeling the fifth dimension with scalars
and gravity, Phys. Rev. D 62 (2000) 046008, hep-th/9909134.
[36] A. Karch, L. Randall, Locally localized gravity, hep-th/0011156.
[37] S.M. Carroll, TASI lectures: cosmology for string theorists, hep-th/0011110.
[38] R.A. Battye, B. Carter, Generic junction conditions in brane-world scenarios, hep-th/0101061.
[39] P.B. Gilkey, K. Kirsten, D.V. Vassilevich, Heat trace asymptotics with transmittal boundary
conditions and quantum brane-world scenario, hep-th/0101105.
[40] B. McInnes, Topologically induced instability in string theory, hep-th/0101136.

Nuclear Physics B 602 (2001) 147171


www.elsevier.nl/locate/npe

Supercurrents in Matrix theory and the generalized


AdS/CFT correspondence
Yasuhiro Sekino
Institute of Physics, University of Tokyo, Komaba, Tokyo 153, Japan
Received 22 December 2000; accepted 15 March 2001

Abstract
We investigate Matrix theory in the large-N limit following the conjectured correspondence
between Matrix theory and supergravity on the near-horizon limit of the D0-brane background.
We analyze the complete fermionic spectrum of supergravity and obtain two-point functions of
the supercurrents in Matrix theory. By examining the large-N scaling properties of the correlators,
we analyze the behavior of the supercurrents under the boost in the 11th direction and discuss the
consistency of the 11-dimensional interpretation of the supersymmetry of Matrix theory. 2001
Published by Elsevier Science B.V.
PACS: 12.10.-g

1. Introduction
According to the Matrix-theory conjecture of Banks, Fischler, Shenker and Susskind
[1], M-theory in the infinite momentum frame is defined by the large-N limit of the
one-dimensional U(N ) super YangMills theory which describe the collection of N D0branes. Since the D0-brane is believed to be the KaluzaKlein particle corresponding
to the compactification of 11th direction and has positive longitudinal momentum, it is
considered as a natural candidate for the elementary degrees of freedom in the infinite
momentum frame. Also, extended objects of M-theory such as membranes and 5-branes
can be described as classical configurations of matrices in SYM theory. However, what
is highly non-trivial is that the SYM theory which is the description of D-branes in
the low-energy, short-distance and the weak-coupling limit is proposed as the exact
non-perturbative definition of M-theory. Taking the large-N limit must be an important
ingredient. Subsequently, it was proposed that the finite-N version of Matrix theory
can be interpreted as the compactification of M-theory on a lightlike circle [2]. The
E-mail address: sekino@hep1.c.u-tokyo.ac.jp (Y. Sekino).
0550-3213/01/$ see front matter 2001 Published by Elsevier Science B.V.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 2 6 - 2

148

Y. Sekino / Nuclear Physics B 602 (2001) 147171

compactification in the lightlike direction is equivalent to the compactification in the


spacelike direction with small compactification radius and small transverse length scale
[3,4]. Since M-theory compactified on the small spatial circle is the weakly-coupled type
IIA string theory, it provides the explanation for the appearance of SYM theory for the
D0-branes. Various quantitative checks for the finite-N interpretation has been performed,
including the consistency with the supergravity interactions up to two-loop perturbations
in Matrix theory [57].
As Matrix theory is intrinsically formulated in 10-dimensional language, the crucial
issue is how it realizes the 11-dimensional Lorentz invariance. In Matrix theory, the
longitudinal momentum is given by P = N/R where R = gs s and the boost in the 11th
direction rescales N with fixed gs , according to the proposal of Banks et al. Thus, for the
boost invariance, the theory must have a kind of scale invariance with respect to N in the
large N limit. However, analysis of gauge theory in the large N limit (with fixed gs ) is
difficult and there was no clue for this scale invariance.
In the previous paper [11], with the motivation for understanding the boost invariance
of Matrix theory, we have investigated Matrix theory in the large-N limit using the
conjectured correspondence with supergravity. For non-dilatonic branes such as D3branes, exact equivalence between super YangMills theory describing the branes and the
superstring theory on the near-horizon limit of the classical solution is proposed [8]. The
same kind of correspondence is conjectured for dilatonic branes such as D0-branes [9].
The conjecture was made more precise by Jevicki and Yoneya [10] who pointed out that
Matrix theory and the near-horizon limit of the D0-brane solution have the same symmetry
called the generalized conformal symmetry. In [11], following this generalized AdS/CFT
correspondence and assuming the relation between the supergravity action evaluated on
the near-horizon limit of D0-brane solution and the generating functional for Matrix-theory
correlators, following the general method proposed by Gubser, Klebanov and Polyakov
[12] and by Witten [13] for non-dilatonic branes, we have predicted the 2-point functions
of bosonic operators of Matrix theory. We have examined the N dependence of the
correlators and analyzed the behavior of Matrix-theory operators under the boost in the
11th direction. We have found that the scaling of N can be interpreted as a spacetime
symmetry, but there was an unexpected anomalous behavior for interpreting it as the
Lorentz boost. However, this anomalous behavior does not necessarily mean the violation
of 11-dimensional Lorentz invariance in Matrix theory. We have to point out the possibility
that the generalized AdS/CFT correspondence does not give the exact information for
Matrix theory, which may be due to the fact that we have to restrict ourselves to the nearhorizon region of D0-branes in the supergravity calculation in this formulation. We shall
review these discussions in Section 4 of this paper for completeness.
In this paper, we study the Matrix-theory supercurrents. The boost properties of the
supercurrents are particularly important, for they form the basis for the following 11dimensional interpretation of the supersymmetry of Matrix theory. The SUSY in 11
dimensions has 32 real supercharges. In the lightcone frame where SO(9) of SO(10, 1)
SO(1, 1) SO(9) is manifest, supercharges are divided into two classes which are
called the kinematical SUSY and the dynamical SUSY, according to the SO(1, 1) weight

Y. Sekino / Nuclear Physics B 602 (2001) 147171

149

(longitudinal boost weight). The explanation for the 11-dimensional SUSY in Matrix
theory is as follows. Matrix theory has two kinds of fermionic symmetries: one is a constant
shift of fermionic field only, and the other is the usual SUSY transformation of SYM
theory. The former is interpreted as the kinematical SUSY and the latter is interpreted as the
dynamical SUSY in 11 dimensions. Indeed, the Dirac-bracket algebra of the supercharges
calculated in Matrix theory reproduces the 11-dimensional lightcone SUSY algebra under
this identification [14]. In this paper, we try to give further evidence for the 11-dimensional
SUSY of Matrix theory by checking whether the supercharges (supercurrents) have the
correct SO(1, 1) weight in the sense of the large-N scaling behavior. We continue the
analysis of [11] and compute the two-point functions of the supercurrents of Matrix
theory following the generalized AdS/CFT correspondence. For that purpose, we work out
the complete spectrum of the fermionic degrees of freedom of 10-dimensional type IIA
supergravity on the near-horizon limit of D0-brane background. Analysis of the spectrum
is performed on the equivalent 11-dimensional background. We analyze the N -dependence
of the correlators which are obtained form the generalized AdS/CFT correspondence and
discuss the behavior of the supercharges under the boost in the 11th direction.
This paper is organized as follows. In Section 2, we review the 10- and 11-dimensional
descriptions of D0-brane solution. In Section 3, the fermionic spectrum of supergravity
is analyzed. In Section 4, we calculate Matrix-theory two-point functions following
the generalized AdS/CFT correspondence. In Section 5, we discuss the 11-dimensional
Lorentz invariance and supersymmetry of Matrix theory by analyzing the scaling behavior
of the correlators with respect to N . We conclude in Section 6.

2. Ten- and eleven-dimensional descriptions of the solution


2.1. The D0-brane solution
D0-brane solution in the string frame in 10 dimensions is given by

2
2
= e2/3 dt 2 + e2/3 dxm
,
ds10


1
1

1 ,
e = gs e ,
A0 =
gs 1 + h

e = (1 + h)3/4 ,

(2.1)

where
q
,
q = 60 3 (  )7/2 gs N
r7
and m(= 1, . . . , 9) denotes the indices of the Cartesian coordinates in the transverse
directions throughout this paper. String coupling gs is defined by e at infinity and the

part of the dilaton which vanish at infinity is denoted by .


The D0-brane solution can be considered as the dimensional reduction of the solution in
11 dimensions. 10-dimensional description and 11-dimensional description are equivalent
if we note a few points remarked in the next subsection. We use the 11-dimensional
h=

150

Y. Sekino / Nuclear Physics B 602 (2001) 147171

description for the analysis of the spectrum in the next section, and we convert to the
10-dimensional language when studying the generalized AdS/CFT correspondence which
is formulated in 10 dimensions. The metric in 11 dimensions and the metric, dilaton and
one-form field in 10 dimensions are related by

2
2
2
ds11
(2.2)
= e2/3 ds10
+ e4/3 dx 11 A dx .
The metric which yields the D0-brane solution upon dimensional reduction on x 11 is given
by 1
2
ds 2 = dx + dx + h dx dx + dxm
,

(2.3)

where x = t + x 11 . We choose the following form of vielbein:


1

e0 + = ,
2 h

1

e11 + = ,
2 h

e0 = 0,

e11 =

h.

(2.4)

Indices with hat refers to those for the local Lorentz frame throughout this paper. We
summarize the connections and the curvatures of this background in Appendix A.
2.2. Near-horizon limit
The 11-dimensional geometry given by (2.3) corresponds to the full D0-brane solution
(without taking near-horizon limit). The near-horizon limit (r
q 1/7 ) of the D0-brane
solution in 10 dimensions is given by substituting h for 1 + h in (2.1). That metric is
formally obtained from the metric
2
ds 2 = 2 dt dx + h dx dx + dxm

(2.5)

by dimensional reduction on the x 11 direction. (2.5) means that we can analyze the
near-horizon limit of the 10-dimensional solution in the 11-dimensional language, by
reinterpreting the coordinate as follows:
2t (n.h.) = x (full)+ ,

x (n.h.) = x (full) ,

(2.6)

where x (full) is the coordinate which was used in the last subsection. We use the coordinate
x (full) in the calculation and make this reinterpretation at the end.
There is one thing which needs care. We want to analyze the spectrum of the fluctuations
which does not depend on the 11th coordinate, for it is directly related to the 10dimensional interpretation. Here, 11th coordinate mean x (n.h.)11 when we want to study the
near-horizon limit. As the fluctuations are independent of x (n.h.)11 = 12 x (full)+ + x (full)
(but depends on t (n.h.) = 12 x (full)+ and other transverse coordinates), in the intermediate
1 To derive (2.3) from (2.1) using (2.2), we have rescaled the value of the coordinates
1/3 (10)
x
,

x (11) = gs

( = 0, . . . , 9),

2/3 (10)11
x
,

x (11)11 = gs

where x (11) , x (11)11 are the coordinates used in (2.3). With the rescaling of x , asymptotic flat 11D metric
corresponds to asymptotic flat 10D metric. The rescaling of the 11th coordinate is such that the period for the
3/2
compactification is R = gs P = gs s (not = P ) for x (11) .

Y. Sekino / Nuclear Physics B 602 (2001) 147171

151

stage where we are calculating using the (full)coordinate, we must take the fluctuations
to be independent of x (full) . 2

3. Fermionic spectrum on the near-horizon D0-brane background


3.1. Equations of motion and SUSY invariance
We shall analyze the spectrum of the fermionic fluctuations on the background (2.3).
Fermionic fields of the 11-dimensional supergravity are the RaritaSchwinger fields M
which are Majorana spinors having vector indices. (We use M = 0, 1, . . . , 9, 11 for the
indices of the 11-dimensional coordinate throughout this paper.) Linearized equation of
motion for the RS field is
MNP DN P = 0,

(3.1)

where DM is the covariant derivative containing the torsionless spin connection at the
linearized level. Equations of motion for each component is given as follows. We use the
spherical coordinates on the transverse S 8 , which is denoted by x i (i = 1, . . . , 8). In the
i which is made
following equations, we use the covariant derivative on the unit sphere D
8
from the connections on S . Covariant derivative Di on the background (2.3) is not equal
i , for Di has extra contributions from r ij , i and i r . Note that the background
to D
rj
(2.3) is not a direct product of a sphere and something. (See Appendix A.) The (M = +)
component of (3.1) is
i 4 r + + r i D
i r
+ r (D r Dr ) + i D
r
4
i i + + D i i + + r Dr i i + + j D
j i i
+ r + D
r
9
+ + r i i = 0.
(3.2)
2r
The (M = ) component is
i + 4 r + + r i D
i r
+ r (D+ r Dr + ) + i D
r
4
i i + + D+ i i + r Dr i i + j D
j i i
r D
r
9
+ r i i = 0.
(3.3)
2r
The (M = r) component is
i + + 4 r + + i D
i + 4 r
+ (D+ D + ) + i D
r
r
7
i i + + D+ i i + D i i + j D
j i i + r i i = 0. (3.4)
D
2r
2 Compactification of the supergravity solution (2.3) on the lightlike circle is also discussed in [1517].

152

Y. Sekino / Nuclear Physics B 602 (2001) 147171

The (M = i) component contracted with i (after subtraction of a certain combination of


(3.2), (3.3) and (3.4)) is
i + + 4 r + + i D
i + 4 r r i D
i r
+ i D
r
r
4
i i + + D+ i i + D i i + r Dr i i + 2 j D
j i i
r 2 D
r
8
+ r i i = 0.
(3.5)
r
The i -transverse part of (M = i) component (which vanishes when contracted with i )
is
(i) + D (i) r D
(i) r + r Dr (i) + j D
j (i) + 4 r (i) = 0,
D
r

(3.6)

where the subscript (i) means the appropriate term proportional to i is subtracted to make
the expression i transverse.
The equations of motion are invariant under the local supersymmetry under which the
linearized fluctuations transform as M = DM #. For general modes, we fix this gauge
invariance by imposing the condition
i i = 0.

(3.7)

For the modes with lowest angular momentum, the above condition cannot be imposed
since some of the components of the fluctuations are invariant under the SUSY
transformation, as we shall see in Section 3.4. That case will be treated separately.
We take the following explicit representation for the matrices in 11 dimensions which
have 32 32 components:


0 = i 1 9 ,

r = 2 9 ,

11 = 3 9 ,

= 1 , (3.8)

where are Gamma matrices in 8 dimensions (16 16 matrices) which satisfy { , } =


2 and 9 = 1 , . . . , 8 .
3.2. Expansions into spinor and vector-spinor spherical harmonics
Since the background has SO(9) symmetry in the transverse space, it is appropriate to
expand each field into spherical harmonics which transform irreducibly under SO(9).
We decompose the spinor fields on S 8 (+ , and r ) into spinor spherical harmonics
( k,+ , k, ) on S 8 which are 16-component Majorana spinors. The coefficients of the
 and
r ) are 2-component spinors:
+ ,
expansion (
+ =
r =



+, k=0



+k, ( 9 )1/2 k, ,

rk, ( 9 )1/2 k, .

+, k=0


+, k=0

k, ( 9 )1/2 k, ,

(3.9)

Y. Sekino / Nuclear Physics B 602 (2001) 147171

153

Spinor spherical harmonics are the eigenfunctions of the Dirac operator on the sphere
i k, = i(k + 4) k,
iD

(k = 0, 1, . . .).

(3.10)

As we can see from (3.10), k,+ and k, transform to each other when multiplied
by 9 . We have to take a linear combination of k,+ and k, in the expansion (3.9)
for diagonalizing the equations of motion. (We have denoted the combination by ( 9 )1/2
following [18]. ( 9 )1/2 has the property i ( 9 )1/2 = i( 9 )1/2 i .) The explicit forms
of the harmonics k, are given by


i )Y k ,
k, = (k + 7 i i D
(3.11)
2 Y k = k(k + 7)Y k .
where Y k is the scalar spherical harmonics of order k which satisfy D
k

8

Di acts only on Y in (3.11). are the Killing spinors on S which satisfy
i = i i .
D
2

(3.12)

The k = 0 modes of the spinor spherical harmonics are the Killing spinors themselves.
We expand the vector-spinor on S 8 , i which satisfy i i = 0 as
i =


+, k=1


 
 k,  9 1/2 D
k, 9 1/2 k, +
(i) k, .

(3.13)

+, k=1

The vector spinor harmonics ik,+ , ik, are 16-component Majorana spinors and have
the properties
i ik, = 0,

i k, = 0
D
i

(3.14)

and are the eigenfunctions of the Dirac operator on the sphere


j k, = i(k + 4) k,
jD
i
i

(k = 1, 2, . . .).

(3.15)

(i) k, is defined by
D
i k, i (k + 4)i k,
(i) k, = D
D
8

(k = 1, 2, . . .)

(3.16)

i (satisfies i D
(i) = 0). The k = 0 mode is an
and is the i -transverse part of D
(i) k, vanishes
exception. Since the vector-spinor harmonics are defined for k  1 and D
i
for k = 0 (as we can see from (3.12) and (3.16)), i which satisfy i = 0 does not have
the k = 0 mode. However, we cannot impose the condition i i = 0 in this case (as we
will see in Section 3.4), and we expand i for the k = 0 mode without that condition as

0, i ( 9 )1/2 .

i =
(3.17)
Since the spherical harmonics of different kind, or with different k, +, are orthogonal
to each other, we can separately analyze each mode. The superscript (k, +), (k, ) on the
spherical harmonics will be often omitted when there is no source of confusion.

154

Y. Sekino / Nuclear Physics B 602 (2001) 147171

3.3. Transverse vector-spinor modes


(k,)

First, we analyze the mode corresponding to the vector-spinor harmonics i


. We
impose the gauge condition i i = 0 and substitute the harmonic expansions (3.9) and
(3.13) into the equations of motion. The vector-spinor modes only appear in (3.6) where
the coefficient of ( 9 )1/2 ik, is


4
1 
h
1
 = 0, (3.18)
+ 3 (k + 4)
2 hi 1 + + i 1 + 2 + 3 r +
4h r
r
h
where h = r h = 7h/r.
To analyze the near-horizon limit of the D0-brane solution, we reinterpret x + 2t and
assume the fields to be independent of x , as explained in Section 2.2. Then, we change
the radial variable to
2
z = q 1/2r 5/2
(3.19)
5
and Euclidize the time coordinate (t = i , M = iE )



7 1  8/5 
2
1
3
z
z + (k + 4) +
(3.20)
= 0.
5
10 z
The solution which is regular at r 0 (z ) is given by

21
 = d ei z 10 C (1) ()K 2 1
(||z)

5 | 2 (k+4)|
2

+ C (2) ()K 2 |3(k+4)| (||z) ,
5

(3.21)

where the superscript (1) denotes the upper component of the two component spinors and
(2) denotes the lower component ( 3 C (1) = +C (1) and 3 C (2) = C (2) ). To be a solution
of (3.20), C (1) and C (2) must satisfy
i 1 C (1) = ||C (2) .

(3.22)

Note that as is the case for the 1st-order action, half of the components are physical and we
(1) or
(2) by a boundary condition.
can only fix either
3.4. Spinor modes at the lowest level (k = 0)
Next, we analyze the spinor modes on the sphere with lowest angular momentum,
k=0, . First of all, we shall discuss the supersymmetry on the D0-brane background.
The transformation M = DM # is written in components as


h 
= # + i 1 2 1 #,
4 h





h
i # 1 3 9 i #.
r = r # +
i = D
3 1 #,
4h
2
When # is in the k = 0 mode (proportional to the Killing spinor)
+ = + #,

# = # ( 9 )1/2 ,

(3.23)

(3.24)

Y. Sekino / Nuclear Physics B 602 (2001) 147171

155

we can see from the property of the Killing spinor (3.12) that i transforms as

 1/2
1
.
(3.25)
1 3 # i 9
2
That is, half of the components of i are invariant under the SUSY and we cannot achieve
i i = 0 for the k = 0 modes. We shall first analyze the equations for the k = 0 modes
without gauge fixing and later specify a gauge conditions which are convenient.
We have to analyze four equations: (3.2)(3.5). ( i -transverse part (3.6) is identically
zero for k = 0 modes.) Substituting the harmonic expansions (3.9) and (3.17) for + , ,
r and i , we obtain the following equations for k = 0 modes without gauge fixing. The
upper signs are for k=0,+ mode and the lower signs are for k=0, . From (3.2),
i =






i
 + 4 1 3
r
r r
 ) h 2
 + 4 i 1 2
1 (
3/2
4h
r
h
hr

h 
8
8
2h 3
r 2
 3 r


1 + 3

+
4h
r
hr
hr

28 
 = 0.
2 1 + 3
(3.26)
r
From (3.3),



h 3
8
4  1
r
+ + (1 1)
+ + 3 1

i 2
2h
r
hr


8  1
2h  1
16

+

3 +
i 2
i 2 r
3/2
r
h r
hr

28  1
 = 0.
i 2
+ (1 1)
(3.27)
2
hr
From (3.4),



h  1
8 h
3
2 
+


2 ( + + ) +
i + +
i 1 2
2h
r



8 
4 
16 h 1
1
2 


i +
+ (1 1)
i +
i 1 + 2
r
hr
hr

28 
 = 0.
2 1 + 3
(3.28)
r
From (3.5),




1 
1
h
+ + (1 1)

r
i 1 + 2
1 3

i 1 + 2
r
2r
2 hr


2 h 1
1 
1
h


 3 r


i 1 + 2
i +

r
r
4h r
hr

7
 = 0.
2 1 + 3
(3.29)
r
 ,
+ ,
We have 8 first-order equations (four 2-component equations) for 2 4 variables (
r ,
 ). As is the case for RaritaSchwinger field, 3 2 degrees of freedom are reduced by

r r
+ )
2(+

156

Y. Sekino / Nuclear Physics B 602 (2001) 147171

gauge fixing and constraints and remaining degrees of freedom are halved due to the Dirac
constraint. Thus, there is one physical degree of freedom. We have obtained the Klein
Gordon-type 2nd-order equation for physical mode which should correspond to the result
of elimination of unphysical variable from Dirac-type 1st-order equations.
The physical mode corresponding to k=0,+ (the upper signs in the above equations)
satisfies
 2

1
21
1  7/2 (1) 
 = 0,
t2 + z2 + z
(3.30)
z

z
5
z2
(1) is the upper component of
(1) = +
 ( 3
(1) ). The derivation of (3.30) is
where
presented in Appendix B. We remark here that we have done the reinterpretation x + 2t
and set = 0 which corresponds to considering the near-horizon limit in 10 dimensions.
(1) ) is not unique for it can
Of course, our choice of the physical degree of freedom (
be expressed by (the combination of) other variables using the constraint equations. In
addition, we have to note that the order of the Bessel function (= 21/5 in this case) is
not unique. That is, when there is a field which satisfy Bessel equation with order


1
2
t2 + z2 + z 2 = 0.
(3.31)
z
z
The following combination satisfies the Bessel equation with the order shifted by 1: 3



1
( 1)2

= 0.

t2 + z2 + z

(3.32)
z
z
z2
z
This fact corresponds to the identity of the Bessel function
zK (z) K (z) = zK1 (z).

(3.33)

In (3.30), we have chosen the representative for physical mode whose order allows
interpretation in terms of the generalized AdS/CFT correspondence, as we shall see in
Section 4.
As a result of a similar analysis for k=0, mode (the lower signs in the field equations),
we obtain the physical mode for this case
 2

1
14
1  7/2 (2) 
= 0,
z
t2 + z2 + z
(3.34)
z
5
z2
(2) =
(2) . The same remark as in the k=0,+ case for the order of the Bessel
where 3
function applies.
3.5. Spinor modes at general level (k = 1, 2, . . .)
In this case, we analyze five equations: (3.2)(3.6). Under the gauge condition i i = 0,
the equations for k, modes (k = 1, 2, . . .) are the following. The upper signs are for the
k,+ mode and the lower signs are for the k, mode. From (3.2),
3 The difference of the orders for the upper (C (1) ) and the lower (C (2) ) component of the vector-spinor mode
 in (3.21) is an example of this ambiguity of the order of the physical degree of freedom.

Y. Sekino / Nuclear Physics B 602 (2001) 147171



 h
2

r r
 + 3
 + (k + 4) 3 + 4
2
2h
r




h 
r 2 h (k + 4) 2 + 4i 1
r
+ i 1 + 2
r
2 h

h
7
 = 0.
+ k(k + 8) 2 i 1
4
r
From (3.3),

157

(3.35)



 h
2
+
r r
+ 3
+ + (k + 4) 3 4
2 +
2h
r


1 
r
+ (k + 4) 4 i 1 + 2
r h

7
1 
= 0.
+ k(k + 8) i 1 + 2
(3.36)
2
8
r h
From (3.4),





h 
2 h
3
1
2
+ +
+

+ i +
2 +
(k + 4)i 1 4 2
r
2 h


1 
 = 0.
 + 7 k(k + 8) 1
+ (k + 4) + 4 i 1 + 2
(3.37)
8
r2
r h
From (3.5),




1 
+ + (k + 4) 4 i 1 + 2

2 h (k + 4)i 1 + 4 2
h


1
7
r k(k + 8)
 = 0.
+ (k + 4) 3 + 4
(3.38)
4
r
From (3.6),


1 

+ i 1 + 2
r + 2 h i 1 +
 3

2 h i 1
h

 
1 
+ h

+ 3 r
+ i 1 + 2
4h
h
1 
3
 = 0.
+ (k + 4) + 4 3
(3.39)
4
r
There are two independent physical degrees of freedom. This is as expected, for the total
number of degrees of freedom for k  1 mode equals that of the RS field. That is, 2 16
for spinor modes ( ) plus (8 2) 16 for vector-spinor modes (i ) (2 for the conditions
D i i = i i = 0) equals the total degrees of freedom of RS fields (11 3) 16. We shall
explain the outline of the calculation of the spectrum and present the results. First, we note
that there are 5 sets of equations for 4 sets of variables. We can see that the system of the
equations is consistent by noticing that (3.39) is obtained by taking a linear combination of
derivatives of other four equations. Thus, we consider only the first four of the equations.
i ,
r and
+ algebraically
Setting = 0 (and reinterpreting x + 2t), we can eliminate
 :
using (3.38), (3.37) and (3.35) and obtain the following equation for

158

Y. Sekino / Nuclear Physics B 602 (2001) 147171


 

4
2
2
32
1
 + r2
 +

+ (k + 4) 1 3 r
ht2
7
7
7
7
r

 

30
15 3 1
2
753

(k + 4)2 (k + 4) +
+ (k + 4)

+
7
56
7
2
r2





8
((k + 4) + 4 3 )
58
2
1
(k + 4) 2(k + 4)
i + 2(k + 4) 10
+
k(k + 8)
7
7

h
 = 0.

(3.40)
+
r
(1) and
(2)
After diagonalizing this second-order coupled equations for two variables (
(1)
(1)
(2)
(2)
 = +
 =
 and 3
 ), we found two physical modes which are
where 3
solved by the modified Bessel functions of the following order . For k,+ modes,
21
2
(k = 0, 1, . . .),
1 = k +
5
5
7
2
(k = 1, 2, . . .),
2 = k +
(3.41)
5
5
where we have included the case k = 0 in the formula. Explicit forms of the diagonalized
fields 1k,+ and 2k,+ corresponding to 1 and 2 respectively are given (for k  1) by




21 1 (1)
(k + 8) 5/2
2
 + z5/2 t
(1) +
(2)
z
k

z
1k,+ = 2i
(3.42)
(2k + 9)
5
10 z
and
2k,+ = 2i



( 31 k + 119
(k + 8) 5/2
5 )  (1)
(1) 10
(2) .
z
+ z5/2 t
z
(2k + 9)
(k + 8)z

(3.43)

For k, modes, the orders of the Bessel functions are


2
1 = k (k = 1, 2, . . .),
5
14
2
(k = 0, 1, . . .).
2 = k +
5
5

(3.44)

Explicit forms of the diagonalized fields (1k, and 2k, ) for k  1 are obtained from the
following expressions:




53 1 (1)
2ik
2
(1)
(2) ,
 + z5/2 t
z5/2 z
k+
1k, =

(2k + 7)
5
10 z


( 31
2ik
(1)
k,
10 k + 1) (1)
5/2

(2) .
2 =
(3.45)
z
+ z5/2t
z
(2k + 7)
kz
1k, and 2k, in (3.45) are the solutions with orders 1 = 2k/5 1 and 2 = 2k/5 + 9/5
respectively which are related to the orders given in (3.44) by = + 1. Thus, we can
construct the solutions with order using the method described in the last subsection. 4 The
4 Different expressions for k, , k, having the orders (3.44) may be possible by using (3.40) again.
1
2

Y. Sekino / Nuclear Physics B 602 (2001) 147171

159

orders of (3.44) are the ones which allow the interpretations in terms of the generalized
AdS/CFT correspondence.

4. Generalized AdS/CFT correspondence


4.1. Matrix-theory correlators form supergravity
As we did in the case of bosons, we shall calculate the correlators in Matrix theory by
evaluating the supergravity action. First, we recall in what situations the calculation of
classical supergravity is valid [11]. The curvature of the D0-brane background must be
small in string unit which gives
r
(gs N)1/3 s

(4.1)

and the effective string coupling (dilaton background) must be small


1/3

gs N 1/7 s
r.

(4.2)

Those two conditions are simultaneously satisfied in the whole near-horizon region
r
(gs N)1/7 s ,

(4.3)

N (gs N) > 1.

(4.4)

if

We assume the following relation between the supergravity action and the generating
functional of the Matrix-theory correlators:
 


dt
I (t)OI (t) ,
eSSG [ ] = exp
(4.5)
I

where we impose the boundary conditions on the physical modes of supergravity at the end
of the near-horizon region (z = q 1/7 , i.e., r (gs N)1/7 s ) and evaluate the action on the
near-horizon limit of the D0-brane background as a functional of the boundary conditions.
In this paper, we obtain the two-point functions of the fermionic operators of Matrix theory.
We first analyze the transverse vector-spinor mode with the + choice of the eigenvalue
of the Dirac operator on the sphere (i (k,+) mode) and then present the results for general
modes. The relevant part of the 10-dimensional action becomes (up to a numerical factor)



 I
1
I (i 1 ) i 1 h t + 3 r +
 .

S = i 2 dt dr r 8
(4.6)

We have used the orthonormality of the spherical harmonics


d8 g ij iI jJ = C I J ,

(4.7)

where C is a numerical constant and I, J are the labels for the harmonics. Also note that
we have chosen the normalization (gs dependence) of the fields such that the action in 10
dimensions has 2 1/gs2 8s as a prefactor.

160

Y. Sekino / Nuclear Physics B 602 (2001) 147171

Euclidizing the time coordinate (t = i ) and using the radial coordinate z =


2q 1/2r 5/2 /5,



 I
1
 .
I ( 1 ) 1 + 3 z +
S = 2 d dz q 8/5z16/5
(4.8)

As noted in Section 3.3, we can impose the boundary condition to one of the two
 . Following the choice which is assumed in the case of ordinary AdS/CFT
components of
correspondence [19], we fix the component which is more divergent as we take the
(1) for the k,+ mode (by noting K (z)
boundary to infinity (z 0). That is, we fix
i
(2) in the case of the k, mode. The action is of
z ). For the same reason, we fix
i
first order and the bulk contribution vanishes on shell, but we need to add the following
boundary term (at z = q 1/7 ) to the action

1 8/516/35

 1
S = Sboundary = 2 q
(4.9)
d
2
(1)
to ensure that the classical solution is really an extremum of the action when we fix
at the boundary [20].
Substituting the solution for the i(k,+) mode


21/10 
i 1
d i 
=
z/q 1/7
e
K 2 (k+1) (||z)

K 2 (k+ 7 ) (||z) +
5
2
5
2
||
(1)
 ()

(4.10)
K 2 (k+ 7 ) (||q 1/7)
5


(1) (, z = q 1/7 ) = d ei
(1)()/2 , the action
which satisfy boundary condition
b
reads

1
i K1 (||q 1/7)
d (1)
 ()
(1) ()

S = 2 q 8/7
b
b

2
|| K (||q 1/7)



(1 ) 1 1+2/7 d (1)
(1) ()i ||22 + , (4.11)
 ()
q
= 22+1

b
b
2
()
2
where
2
7
= k+ ,
5
5
and we have retained only the leading part which is non-analytic in . We have used the
T 3 ).
 =
Majorana condition (for our representation,
Following the ansatz of the generalized AdS/CFT correspondence (4.5), we obtain the
two-point function of the operator O which couple to the mode (1)
O()O() =

SSG
(1) ()
(1) ()

b
b

22+1 (1 ) 1 1+2/7 22
=
q
i ||
+ .
2

()

(4.12)

Y. Sekino / Nuclear Physics B 602 (2001) 147171

161

Fourier transforming to the position space,




d i
d i
e
e
O(1 )O(2 ) =
O()O( )
2
2
1 ( + 1/2) 1 1+2/7 (1 2 )
=
(4.13)
q
.
3/2
2
()
2
|1 2 |2+1
The 2-point functions corresponding to the other physical modes are given by substituting
into (4.13) the order obtained in Sections 3.3, 3.4 and 3.5.
4.2. Identification of the Matrix-theory operators
Following the generalized AdS/CFT correspondence (4.5), we have calculated the 2point functions of Matrix-theory operators which couple to RaritaSchwinger fields of
the 11-dimensional supergravity. Now we would like to identify the operators which
couple to each physical mode of supergravity. RS fields are the gauge fields of the
supersymmetry and should couple to the supercurrents of Matrix theory, but we do not have
principles which determine the coupling. We shall determine the Matrix-theory operator by
requiring (1) the scaling dimension (with respect to the generalized conformal symmetry
which is explained below) agrees with the one read off from the correlator which was
obtained form the supergravity calculation, and (2) the representation with respect to the
transverse rotation group SO(9) agrees with the supergravity mode. Note that this situation
is essentially the same as in the ordinary AdS/CFT correspondence such as the AdS5 /CFT4
case, where the representation with respect to the N = 4 superconformal symmetry is the
guideline for determining the coupling.
Matrix theory is defined by the following action:


1
i
1
1
S=
dt Tr Dt Xm Dt Xm + Dt + 4 [Xm , Xn ]2
gs s
2
2
4s

1
(4.14)
2 m [, Xm ] ,
2s
where Dt = t + i[A, ] and m are the 9-dimensional Gamma matrices which are real and
symmetric. It has the generalized conformal symmetry which allows the coupling constant
gs to be transformed. The scale transformation is given by
t 1 t,

Xm Xm ,

3/2 ,

gs 3 gs .

(4.15)

Together with the time translation and the special conformal transformation (treating as gs
as a field with dimension 3), they constitute the conformal algebra. The near-horizon limit
of the D0-brane solution has the same kind of symmetry [10].
Matrix theory has two kinds of fermionic symmetries. One is a constant shift of
fermionic field only
= #.

(4.16)

The other is the usual supersymmetry of the SYM theory


Xm = i# T m ,

A = i# T ,

162

Y. Sekino / Nuclear Physics B 602 (2001) 147171



1
i
2Dt Xm m + 2 [Xm , Xn ] mn #.
2
s

(4.17)

The explicit forms of the supercurrents of Matrix theory are obtained by Taylor and
van Raamsdonk [21]. They are identified by reinterpreting the one-loop effective potential
between a pair of diagonal blocks in Matrix theory as the potential between the currents in
the linearized supergravity compactified on x . The (M = +) and (M = m) components
of the supercurrents are given as follows up to numerical constant factors and possible
three-fermion terms. The currents are assumed to be integrated in the x -direction:
1
Tr( ),
R
 
1 
q m = Tr X m Fmn n ,
R
 
1 
+
q = Tr X m m + 12 Fmn mn ,
R

1
m
q = STr 2X m X n n + X m Fnp n p
R
 
+ 2X n Fpm n p + Fnp Fqm n p q ,
q + =

(4.18)

where Fmn = i[Xm , Xn ] and STr is the symmetrized trace in which the average is taken
over all possible orderings of matrices (treating Fmn as a single unit). The above expression
is for the A = 0 gauge and we have set s = 1. Note that q + and q + agree with
the supercharges in the SYM theory corresponding to (4.16) and (4.17). The (M = )
components of the supercurrents are of the form (up to 3-fermion terms)

 
1
q = STr terms of 2nd order in Fmn , X m ,
R

 
1
q = STr terms of 3rd order in Fmn , X m .
(4.19)
R
To obtain the explicit forms of q and q , analysis of the 1/r 8 terms of Matrixtheory effective potential is needed. (Terms in (4.18) are obtained from the 1/r 7 terms.)
In addition to the operators (4.18) and (4.19), there are the moments of the currents
which couple to the transverse derivatives of the supergravity fields. The kth moment
M{m ,...,mk }
M{m ,...,mk }
, q(k) 1
contain a term which is obtained by inserting k times the
q(k) 1
m
m

transverse field X = X /q 1/7 into the traces in (4.18) and (4.19).


The consistency of these couplings between the branes and the supergravity fields
are confirmed in various contexts: they are consistent with the couplings obtained by
expanding the Matrix-regularized membrane action in the lightcone gauge around the flat
background [22]; the couplings for D0-branes which are obtained through the Seiberg
Sen argument are consistent with the BornInfeld action [23]; the couplings for Dpbranes obtained from D0 couplings by T-duality indeed have (p + 1)-dimensional Lorentz
covariance [24]; and notably, the leading part of the absorption cross sections by D3-branes
in the small (gs N), large N limit which is calculated in the D = 4, N = 4 SYM theory
using the T-dualized couplings exactly reproduce the ones calculated in the semiclassical
supergravity including the cases of higher partial waves [25].

Y. Sekino / Nuclear Physics B 602 (2001) 147171

163

Table 1
Physical modes of supergravity and the corresponding Matrix-theory operators
i

SUGRA fields

+ , , r , i

ik,+

ik,

k,+

k,+

k,

k,

(1)
z7/2

 (2)
z7/2

1k,+

2k,+

2k,

1k,

2k + 7
5
5

2 k + 14
5
5

2
21
5k + 5

2
7
5k + 5

2
14
5k + 5

2k
5

Regions of k

k1

k1

k 0

k1

k0

k1

Operator O

m
q(k)
1 + 4k
7
2

m
q(k)
5 + 4k
7
2

q(k)
9 + 4k
7
2

+
q(k)
1 + 4k
7
2

q(k)
5 + 4k
7
2

+
q(k)

Harmonics
Physical modes
Order

Dimensions of O

32 + 47 k

We assume that these operators are the candidates for the operators which couple to the
supergravity modes in our case. We shall compare the scaling dimensions (with respect to
the generalized conformal symmetry (4.15)) of the supercurrents and that of the correlators
obtained in the last subsection. From the right hand side of (4.13), we see that the scaling
dimension of the operator is




1
2
=
3 (2) + 3 1 +
+ 2
2
7
3 10
= + .
(4.20)
2
7
We can consistently identify the Matrix-theory operator which have the dimension (4.20)
m
for every supergravity modes. The results are summarized in Table 1. Operators q(k)
and
m
q(k) in Table 1 are the ones which have the same SO(9) quantum numbers as the harmonics,
m{m1 }
e.g., q
1 STr{(X m Fmn n )Xm1 } (m m1 ).
(1)

5. 11-dimensional Lorentz invariance and supersymmetry of Matrix theory


In this section we consider the implications of our results for the Matrix-theory
conjecture, following the line of arguments of the previous paper [11]. Some of the results
of [11] are reviewed for completeness. According to the conjecture of Banks, Fischler,
Shenker and Susskind, infinite momentum frame (P ) is achieved by taking the
limit
N with

gs : fixed.

(5.1)

In this region, Matrix theory is expected to have the boost invariance in the x 11 direction,
which rescales N with fixed gs . As argued in Section 4.1, our results obtained from
supergravity is valid when N and (gs N) > 1, which will allows us to investigate

164

Y. Sekino / Nuclear Physics B 602 (2001) 147171

the infinite momentum limit. 5 We shall check the boost invariance in the sense of the
large-N scaling behavior of the operators.
Under the boost transformation
x + e x + ,

x e x ,

(5.2)

where P scales as N , the time coordinate of Matrix theory which is the lightcone time
scales as N , Matrix-theory Hamiltonian (P+ ) must scale as 1/N while the transverse
coordinates and the compactification radius are kept fixed. We shall analyze the behavior
of the operators under the boost from the scaling behavior of the correlators. We substitute
N in the correlators and examine the N dependence
O(1 )O(2 ) N 2dIMF G(1 2 ),

(5.3)

where G( ) is a function independent of N and we have defined the scaling dimension


with respect to the boost transformation dIMF of the operator O.
First, we review the behavior of the correlators of the bosonic operators. From the
generalized AdS/CFT correspondence we have obtained [11]
O(1 )O(2 ) =

1
gs2 8s

(gs N7s )1+2/7

1
,
|1 2 |2+1

(5.4)

where is the order of the modified Bessel function of the corresponding supergravity
mode. By examining the values of , we have found that the dimension dIMF is given by




1
6
1 1
+
dIMF = = 1 +
(5.5)
(n+ n 1)
k,
7
5
7 5
where n+ (n ) is the number of the upper +() index on the operator and k means the kth
moment. The fact that dIMF is determined solely from the spacetime index structure of the
operators suggests that the scaling of N is indeed related to the spacetime symmetry. The
tensor with the + index indeed scales inversely as the tensor with the index. However, the
weights 6/5 are different from the natural expectation 1 by 1/5. The interpretation
of the remaining terms are as follows. The kth moment has a contribution (1/7 + 1/5)k.
The factor 1/7 is explained by the fact that the moments which we are considering are
m Xm /((gs N)1/7 s ) (but not Xm ) k times, but there is also an
made by multiplying X
anomalous behavior 1/5 for each transverse field. The constant part ((1 + 1/5)) is
interpreted as coming from the x integration when we define the currents, also with the
anomalous behavior for x (1/5).
Now we shall analyze the scaling behavior of the supercurrents. From the correlator
obtained in the last section
O(1 )O(2 ) =

1
(1 2 )
(gs N7s )1+2/7
gs2 8s
|1 2 |2+1

(5.6)

5 Strictly speaking, the condition for the validity of supergravity (4.4) is different from the infinite momentum

limit (5.1). To satisfy (4.4), we need to take gs 0 as N rather than fixing gs . We are assuming that this
difference does not affect the results.

Y. Sekino / Nuclear Physics B 602 (2001) 147171

165

we find the following scaling properties of the fermionic operators of Matrix theory:
1 6

2
 7 




1
1 1
1
1
1
= 1+
(n+ n 1)
+
k
1+
,
5
7 5
2
5
10

dIMF =

(5.7)

where the () sign on 12 (1 + 15 ) is for the current q M and the (+) sign is for the current q M .
The interpretations of the scaling for the lightcone indices, for x -integration and for
the moments are the same as in the bosonic case. In addition, we see that q M and q M
transform inversely. This is indeed needed for spacetime interpretation. The current q M
is the supercurrent corresponding to the constant shift of fermionic field (4.16) in Matrix
theory, which is interpreted as the kinematical SUSY in 11 dimensions. The current q M
is the supercurrent corresponding to the supersymmetry of the SYM theory (4.16), which
is interpreted as the dynamical SUSY in 11 dimensions. The kinematical supercharge (q)

satisfy q = 0 and the dynamical supercharge (q) satisfy + q = 0 to form the lightcone
q} Pm m , {q, q} P+ + ). Under the boost (5.2),
SUSY algebra ({q,
q}
P , {q,
the kinematical and dynamical SUSY charges transform as
q e

0,11 /2

q = e/2 q,

q e

0,11 /2

q = e/2 q,

(5.8)

which can be shown from the properties of q, q mentioned in the above sentences. Thus we
expect that the current q M has a contribution 1/2 for the boost weight, and the current
q M has +1/2. Our results shows anomalous behavior 1/10 also in this case. We do not
have an interpretation for the last term 1/10 of (5.7).
The interpretation of the anomalous behavior is not clear. One possibility is that
Matrix theory does not have 11-dimensional Lorentz invariance or supersymmetry, but
the following interpretation seems more appropriate [11]. We have studied Matrix theory
by analyzing the supergravity inside the near-horizon region. However, it is possible that
we could not take into account the necessary degrees of freedom of Matrix theory by
restricting to the near-horizon region. Indeed, average transverse size of Matrix theory
L2 Tr(Xm Xm )/N is estimated to be L (gs N)1/3 s [26], which is larger than the
boundary of the near-horizon region rb (gs N)1/7 s in the case of interest (gs N) > 1.
Further argument supporting this interpretation was given in [27]. Suppose we shift
the boundary rb N 1/7 to some larger value rb N . Assuming that the system is
homogeneous (though it will not really be the case), the field Xm will be rescaled to rrbb Xm .
If we require the factor just cancels the anomalous behavior for transverse fields N 1/5 ,
the shifted boundary must be
rb N 1/7+1/5 = N 1/3+1/105,

(5.9)

which is very close to L N 1/3 . It seems to suggest that we need to have radial size at
least L (gs N)1/3 s to analyze Matrix theory.

166

Y. Sekino / Nuclear Physics B 602 (2001) 147171

6. Discussions
In this paper, we have predicted the 2-point functions of the supercurrents of Matrix
theory following the conjecture of generalized AdS/CFT correspondence. We have
analyzed the scaling behavior with respect to N and considered whether the Matrixtheory supercurrents have correct weights when we interpret the scaling of N (with fixed
gs ) as the boost in the 11th direction following the original proposal of Banks, Fischler,
Shenker and Susskind [1]. We have reached to a result which is not very definitive: The
supercurrents corresponding to the two kinds of fermionic symmetries of Matrix theory can
be indeed interpreted as having inverse weights under the Lorentz boost, which suggests the
correctness of the interpretation of them as kinematical and dynamical SUSY currents of
the 11-dimensional SUSY. However, the weights are slightly different from the canonical
behavior, which seem to be of the same nature as the anomalous behavior for the spacetime
tensors which were found in the previous paper. The origin of these behavior are not clear
but it is likely that the generalized AdS/CFT correspondence is not able to give the exact
information of Matrix theory, possibly due to the IR cutoff of the supergravity calculation
which is the boundary of the near-horizon region.
To understand the effect of this cutoff more precisely, we will have to study the cutoff
dependence from both supergravity and gauge theory. On the supergravity side, one
possibility is to try to extend further the generalized AdS/CFT correspondence assuming
that the bulk/boundary correspondence continues to hold for outside the near-horizon
region. The holographic renormalization group [28] is a suitable framework for analyzing
the effect of shifting the position of the boundary. On the gauge-theory side, on the other
hand, analyzing the dependence on the cutoff (for transverse fields Xm ) may be as difficult
as solving the theory directly. Up to now, quantitative study of the large-N Matrix theory
in the strong coupling is only attempted through the Gaussian approximation [29,30]. We
hope that the cutoff dependence can be incorporated in some approximation scheme.
Though we have mentioned the possibility for the limitation of the generalized AdS/CFT
correspondence, our results which exhibit the large-N scaling behavior should be the
consequence of the collective dynamics of Matrix theory. Though the near-horizon region
r < (gs N)1/7 s is smaller than the average transverse size L (gs N)1/3 s , it should be
noted that it is larger than the characteristic size of N -body bound state obtained by a
mean field analysis [1] Lm.f. (gs N)1/9 s . We would like to note a fact which may be an
indication [11] that the generalized AdS/CFT correspondence takes into account certain
amounts of degrees of freedom of Matrix theory. Assume that the coefficient of the twopoint function of the stress-energy tensor gives the entropy of the theory. The coefficient
(c) of the two-point function of Tij in the lowest angular momentum (k = 0) mode which
is obtained from the generalized AdS/CFT correspondence c N 2 (gs N)3/5 agrees with
the gs and N dependence of the entropy which is calculated from the BekensteinHawking
formula S N 2 (gs N3s /TH3 )3/5 where TH is the Hawking temperature. We hope to
clarify the meaning of this agreement in future works.
Once the nature and the validity of the generalized AdS/CFT correspondence for Matrix
theory are understood, we can use this formalism to investigate other dilatonic D-branes,

Y. Sekino / Nuclear Physics B 602 (2001) 147171

167

for the general Dp-branes have the generalized conformal symmetry. Particularly interesting case is that of D1-branes. The U(N ) super YangMills theory in 2 dimensions describing D1-branes is called Matrix string theory and is conjectured to be another nonperturbative definition of the type IIA string theory [31]. In the large-N limit with small
gs , this theory is believed to be described effectively by a conformal field theory, and an
effective interaction vertex operator is proposed which is claimed to have leading scaling dimension. The generalized AdS/CFT correspondence for D1-branes will be useful
in checking these conjectures and in investigating the non-perturbative behavior of string
theory.

Acknowledgement
I would like to thank T. Yoneya for valuable discussions at every stage of this work and
for careful reading of the manuscript. I also thank W. Taylor for reading the manuscript
and for giving important comments.

Appendix A. The connections and the curvatures of the background


We summarize here the connections and the curvatures of the background in 11
dimensions which is given by the metric
2
,
ds 2 = dx + dx + h dx dx + dxm

where h = q/r 7 . Using the spherical coordinate on S 8 , the metric reads


g++ = 0,

1
g+ = ,
2

g = h,

grr = 1,

gij = r 2 kij ,

(A.1)

where kij is the metric of the unit sphere S 8 and x i (i, j = 1, . . . , 8) are the (angular)
coordinate along S 8 .
Non-vanishing components of the Christoffel symbols are
1
r = h ,
2
1
r
ij = gij ,
r

1
i j r = ji ,
r

+ r = h ,
8

i j k = i j k (S ) ,

(A.2)

where i j k (S ) means the Christoffel symbol on the unit sphere S 8 and h = r h. Nonvanishing components of the curvature tensors are
8

h i 7 q i
=
,
2r k 2 r 9 k
With the choice of the vielbein
R i j =

1

e0 + = ,
2 h

1

e11 + = ,
2 h

1
q
R r r = h = 28 9 .
2
r


e0 = 0,

e11 =

(A.3)

(A.4)

168

Y. Sekino / Nuclear Physics B 602 (2001) 147171

non-vanishing components of the spin connections are


h
0r = ,
2 h
1
i r = ei er r ,
r

h
11
r = ,
2 h

r011
 =

h
,
2h

i k = i k (S ) ,

(A.5)

where i k (S ) means the spin connection on the unit sphere S 8 .


i . The
We define the covariant derivative which is made from the connections on S 8 , D
i
r
covariant derivative Di has extra contributions from ij , rj , i r . For example, the
covariant derivative of a spinor # is
1

i # + 1 i r r #
Di # = i # + i M N M N # = D
4
2

 3
1
i # 9 i #
=D
2
using the representation of the matrix (3.8).

(A.6)

Appendix B. Spectrum of the k=0,+ mode


In this appendix, we describe the details of the analysis of the spectrum of the k=0,+
mode and show that there is one physical degree of freedom which is solved by the Bessel
equation with order = 21/5. We have 2 4 components of Eqs. (3.2)(3.5) for 8 variables
+(2) ,
(1) ,
(2) ,
r(1) ,
r(2) ,
(1) and
(2) , where the superscripts (1) and (2)
+(1) ,
(
denote upper and lower component of the two-component spinors, respectively).
M = DM # under which the fields transform as
There is a local supersymmetry

h 
 = # + 2 + i 1 # ,

4 h



h
r = r # + 3 #,
 = 1 1 3 #.

(B.1)

4h
2
The parameter # is defined by # = #(
9 )1/2 where is the Killing spinor on S 8 . We
(2) = 0. Other half will be fixed in the following.
use half of the gauge freedom and set
(2) = 0 and = 0 are the
The equations of motion in components with the condition
following. From (3.2),
 



i
h
8
8
56 2h
(2)
(2)
(1)



(1) = 0, (B.2)
r + i

+
r

4h3/2
r
r2
rh
h
hr


h
i
ih
8
(1)
(1)
(2) = 0.


+

r 3/2
(B.3)
r
2h r
4h
h
From (3.3),
+ = + # ,

h (1)
16i (2) 16
r +
+ +
(1) = 0,

2h
r
r h


r(2) 2r
+(2) + 16 + h
(2) = 0.
2+
r
2h +

r(1) 2r
+(1)
2+

(B.4)
(B.5)

Y. Sekino / Nuclear Physics B 602 (2001) 147171

169

From (3.5),



16 h
16i (2) 56 (1)

+(2)
(1) + i h
 = 0,
2+
+

r
h
r2
hr

16i h
(2)

(1) = 0.
+
2+
r
From (3.6),



2i
2i h (2) 1
14
h
(2) +
(1)
+ r
(1) = 0,

+

r
r
r2
4rh
r h

1 (2) 2i h

(1) = 0.
+
r
r
r
To solve this set of equations, first note that (B.7) can be written as



8i h (1)
(2)



=0
+
r

(B.6)
(B.7)

(B.8)
(B.9)

and (B.2) can be written as




8i h (1)
8 1/4  (2)

r r h

= 0,

r
thus, we can set

8i h (1)
(2)

=
.
r
From (B.9), we find

r(2) = 2i h+
(1) .

From (B.8) and (B.10),






h (2)
r
1
(2)
(2)


+
+ =
.
r

16h
r
4h

(B.10)

(B.11)

(B.12)

(1) ,
r(2) and
+(2) (and using h = 7h/r), (B.3)
Using (B.10)(B.12) to eliminate
can be written as

 9
(1) = h3/4 +
(2)
ir h1/4
(B.13)
8
and (B.6) can be written as
(1) =
+

9i
603i
(2).
(2)
r

32 h
128r h

(1) which satisfy


This set of Dirac equation is solved by

 2
1 7/2 (1)
1
21
2
t + z + z
z
= 0,
z
5
z2

(B.14)

(B.15)

where we have reinterpreted x + 2t and changed the radial variables z = 2q 1/2r 5/2 /5.

170

Y. Sekino / Nuclear Physics B 602 (2001) 147171

(1) and we have seen that the variables


We have found one physical degree of freedom
(1) ,
r(2) ,
+(2) and
(2) are determined using Eqs. (B.10)(B.13). Remaining 2 variables,

(1)
(1)
+ and
r are determined by (B.4) and using the remaining gauge freedom (to set, e.g.,

(1)
+ = 0). We can check that the set of solutions satisfy Eq. (B.5) which we have not used

for the derivation, which shows that the system of the equations is consistent.

References
[1] T. Banks, W. Fischler, S.H. Shenker, L. Susskind, M-theory as a Matrix model: a conjecture,
Phys. Rev. D 55 (1997) 5112, hep-th/9610043.
[2] L. Susskind, Another conjecture about M(atrix) theory, hep-th/9704080.
[3] N. Seiberg, Why is the Matrix model correct?, Phys. Rev. Lett. 79 (1997) 3577, hep-th/9710009.
[4] A. Sen, D0 branes on Tn and Matrix theory, Adv. Theor. Math. Phys. 2 (1998) 51, hepth/9709220.
[5] K. Becker, M. Becker, A two-loop test of M(atrix) theory, Nucl. Phys. B 506 (1997) 48, hepth/9705091.
[6] K. Becker, M. Becker, J. Polchinski, A. Tseytlin, Higher order graviton scattering in M(atrix)
theory, Phys. Rev. D 56 (1997) 3174, hep-th/9706072.
[7] Y. Okawa, T. Yoneya, Multi-body interactions of D-particles in supergravity and Matrix theory,
Nucl. Phys. B 538 (1998) 67, hep-th/9806108;
Y. Okawa, T. Yoneya, Equations of motion and Galilei invariance in D-particle dynamics, Nucl.
Phys. B 541 (1999) 163, hep-th/9808188.
[8] J. Maldacena, The large N limit of superconformal field theories and supergravity, Adv. Theor.
Math. Phys. 2 (1998) 231, hep-th/9711200.
[9] N. Itzhaki, J. Maldacena, J. Sonnenschein, S. Yankielowicz, Supergravity and the large N limit
of theories with sixteen supercharges, Phys. Rev. D 58 (1998) 046004, hep-th/9802042.
[10] A. Jevicki, T. Yoneya, Spacetime uncertainty principle and conformal symmetry in D-particle
dynamics, Nucl. Phys. B 535 (1998) 335, hep-th/9805069.
[11] Y. Sekino, T. Yoneya, Generalized AdS-CFT correspondence for Matrix theory in the large N
limit, Nucl. Phys. B 570 (2000) 174, hep-th/9907029.
[12] S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Gauge theory correlators from noncritical string
theory, Phys. Lett. B 428 (1998) 105, hep-th/9802109.
[13] E. Witten, Anti de Sitter space and holography, Adv. Theor. Math. Phys. 2 (1998) 505, hep-th/
9802150.
[14] T. Banks, N. Seiberg, S. Shenker, Branes from matrices, Nucl. Phys. B 490 (1997) 91, hepth/9612157.
[15] S. Hyun, Y. Kiem, H. Shin, Infinite Lorentz boost along the M-theory circle and nonasymptotically flat solutions in supergravities, Phys. Rev. D 57 (1998) 4856, hep-th/9712021.
[16] S. Hyun, The background geometry of DLCQ supergravity, Phys. Lett. B 441 (1998) 116, hepth/9802026.
[17] S. Hyun, Y. Kiem, Background geometry of DLCQ M theory on a p-torus and holography,
Phys. Rev. D 59 (1999) 026003, hep-th/9805136.
[18] P. van Niewenhuizen, The complete mass spectrum of d = 11 supergravity compactified on S4
and a general mass formula for arbitrary cosets M4 , Class. Quantum Grav. 2 (1985) 1.
[19] M. Henningson, K. Sfetsos, Spinors and the AdS/CFT correspondence, Phys. Lett. B 431 (1998)
63, hep-th/9803251.
[20] M. Henneaux, Boundary terms in the AdS/CFT correspondence for spinor fields, hepth/9902137.

Y. Sekino / Nuclear Physics B 602 (2001) 147171

171

[21] W. Taylor, M. Van Raamsdonk, Supergravity currents and linearized interactions for Matrix
Theory configurations with fermionic backgrounds, JHEP 9904 (1999) 013, hep-th/9812239.
[22] A. Dasgupta, H. Nicolai, J. Plefka, Vertex operators for the supermembrane, JHEP 0005 (2000)
007, hep-th/0003280.
[23] W. Taylor, M. Van Raamsdonk, Multiple D0-branes in weakly curved backgrounds, Nucl. Phys.
B 558 (1999) 6395, hep-th/9904095.
[24] W. Taylor, M. Van Raamsdonk, Multiple Dp-branes in weak background fields, Nucl. Phys.
B 573 (2000) 703734, hep-th/9910052.
[25] I. Klebanov, W. Taylor, M. Van Raamsdonk, Absorption of dilaton partial waves by D3-branes,
Nucl. Phys. B 560 (1999) 207229, hep-th/9905174.
[26] J. Polchinski, M-theory and the light cone, Prog. Theor. Phys. Suppl. 134 (1999) 158, hep-th/
9903165.
[27] T. Yoneya, Generalized conformal symmetry and oblique AdS/CFT correspondence for Matrix
theory, hep-th/9908153.
[28] J. de Boer, E. Verlinde, H. Verlinde, On the holographic renormalization group, JHEP 0008
(2000) 003, hep-th/9912012.
[29] D. Kabat, G. Lifschytz, Approximations for strongly-coupled supersymmetric quantum
mechanics, Nucl. Phys. B 571 (2000) 419, hep-th/9910001.
[30] D. Kabat, G. Lifschytz, D.A. Lowe, Black hole thermodynamics from calculations in stronglycoupled gauge theory, hep-th/0007051.
[31] R. Dijkgraaf, E. Verlinde, H. Verlinde, Matrix string theory, Nucl. Phys. B 500 (1997) 43, hepth/9703030.

Nuclear Physics B 602 (2001) 172200


www.elsevier.nl/locate/npe

Four-flux and warped heterotic M-theory


compactifications
Gottfried Curio a , Axel Krause b
a Humboldt-Universitt zu Berlin, Institut fr Physik, D-10115 Berlin, Germany
b Physics Department, University of Crete, 71003 Heraklion, Greece

Received 30 January 2001; accepted 9 March 2001

Abstract
In the framework of heterotic M-theory compactified on a CalabiYau threefold times an interval,
the relation between geometry and four-flux is derived beyond first order. Besides the case with
general flux which cannot be described by a warped geometry one is naturally led to consider two
special types of four-flux in detail. One choice shows how the M-theory relation between warped
geometry and flux reproduces the analogous one of the weakly coupled heterotic string with torsion.
The other one leads to a quadratic dependence of the CalabiYau volume with respect to the
orbifold direction which avoids the problem with negative volume of the first order approximation.
As in the first order analysis we still find that Newtons constant is bounded from below at just
the phenomenologically relevant value. However, the bound does not require an ad hoc truncation
of the orbifold-size any longer. Finally we demonstrate explicitly that to leading order in 2/3 no
Cosmological constant is induced in the four-dimensional low-energy action. This is in accord with
what one can expect from supersymmetry. 2001 Elsevier Science B.V. All rights reserved.
PACS: 04.65.+e; 11.25.Mj; 12.60.Jv

1. Introduction and summary


For heterotic M-theory compactified on a CY 3 the 4-form field-strength G does
not vanish if higher order corrections in 2/3 are taken into account. The reason is
that the boundary super YangMills (SYM) theories represent magnetic sources which
show up in the Bianchi-identity for G and require a G of order 2/3 . Hence, by
requiring SUSY which connects via the Killing-spinor equation metric quantities with
the G-flux, we expect warped geometries to arise at this order, which is indeed the
case [8]. An interesting interplay arises between the physics of G-fluxes and warped
geometry (similar considerations for M- and F-theory on CY fourfolds can be found
E-mail addresses: curio@physik.hu-berlin.de (G. Curio), krause@physics.uoc.gr (A. Krause).
0550-3213/01/$ see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 1 6 - X

G. Curio, A. Krause / Nuclear Physics B 602 (2001) 172200

173

in [1520]; see also [21] for dynamical topology changes within heterotic M-theory).
An important phenomenological issue is related to the value of Newtons Constant GN .
From dimensional reduction of heterotic M-theory on a CY 3 with volume V (x 11 ) (in the
11-dimensional metric), one can infer [8] (note that these expressions will become more
refined if a non-trivial dependence on internal coordinates is kept, as will be explored in
Section 5 below)
GN =

2
,
16V d

i =

(4 2 )2/3
,
2Vi

(1)

where i is the gauge-coupling of the two (i = 1, 2) boundary E8 SYM theories separated


by a distance d and V1 = V (0), V2 = V (d). Because GN is related to gravity in the
d
bulk, we use for its determination a bulk-averaged volume V  = d1 0 dx 11 V (x 11 ).
A determination of the warped geometry allows to calculate V (x 11 ) and thereby i and
GN . This had been undertaken in [8] to linearized order (first order in 2/3 ) with the


1
result that V (x 11) = ax 11 + V0 , where the slope a =
d 6 x g lm np Glmnp > 0
4 2 CY 3
is controlled by the G-flux. Here lm denotes the Khler-form on CY 3 . The surprising
observation [8] has been that when the linear function V (x 11 ) becomes zero, the
corresponding distance for d just gives rise to the correct value for GN , whereas generically
in heterotic string compactifications GN is predicted too large by a factor of 400. Placing
the second boundary at that distance means 2 . Hence, the SYM there becomes
strongly coupled and instanton contributions become relevant, which is the reason why this
second boundary corresponds to a hidden world rather than our observable world. The
first boundary at x 11 = 0 instead allows for a perturbative SYM on it if V1 is chosen huge
enough such that 1 1 and consequently can be regarded as our observable world.
In this context some questions arise:
How is the linear behaviour of V (x 11), which leads to an unphysical negative volume
beyond a certain distance, changed in the full theory, i.e., beyond the leading 2/3
order?
Does V (x 11) still keeps its attractive feature of becoming zero just at the phenomenological relevant distance? One has to recall that actually the linear approximation
breaks down at the position of the zero. Is Newtons constant still bounded from below
beyond leading order?
Does the phenomenologically relevant distance d get stabilized by an effective
potential?
What is the relation of the compactification of heterotic M-theory with G-flux to the
weakly coupled heterotic string with torsion [3]?
The trouble with the linear behaviour is the following. Though one expects that quantum
corrections will shift the actual value of V (x 11 ) slightly, small distortions of a linear
function can never lift its zero they can only shift its x 11 position slightly, but the
zero remains, as does the problem with the unphysical negative volume. Therefore, it is
important to determine the warp-factors and thereby V (x 11 ) beyond the leading order in
2/3 , which we will undertake in this paper.

174

G. Curio, A. Krause / Nuclear Physics B 602 (2001) 172200

It may sound surprising how a result beyond order 2/3 can be achieved within the
framework of heterotic M-theory whose effective action is only known to order 2/3 .
Let us, therefore, briefly indicate where and in which way features of heterotic M-theory
will enter our analysis. By imposing supersymmetry, we are going to solve the gravitino
Killing-spinor equation of M-theory. The heterotic M-theory characteristics enter on the
one hand through specific G-fluxes originating from boundary or M5-brane sources and on
the other hand through the chirality condition 11 = on the susy-variation Majoranaparameter . The important point is that the information which is restricted to order 2/3
becomes only relevant if knowledge about the actual source strengths is required. However,
to obtain the functional behaviour of V (x 11) this knowledge is not needed. It suffices to
assume that in the full heterotic M-theory the relevant sources can be localized in the x 11
direction (as suggested by the anomaly considerations of [12] leading to the two E8 gauge
groups), i.e., they appear as dG = (x 11 z)S(x m ) dx 11 in the Bianchi-identity. Thus,
we will be able to answer the first question posed above. The actual value (and thereby the
complete knowledge about heterotic M-theory beyond 2/3 order) for the 4-form source
strength S only becomes indispensable if, e.g., questions about the precise value of a zero
or a minimum of V (x 11) should be answered. This is necessary if one wants to quantify
the resulting value for Newtons constant.
The main results of this paper concerning the questions posed above are:
The full dependence of the CalabiYau volume on the orbifold coordinate is quadratic
with a manifest non-negative volume throughout the whole interval. The minimum of
the parabola is a zero which can now be lifted by quantum effects.
This zero of the nonlinear analysis corrects the zero of the first order analysis
by a factor two. The critical lower-bound on Newtons constant in the first order
approximation is thereby further decreased by a factor 2/3. This ameliorates the slight
discrepancy between the lower bound on Newtons constant and its observed value.
The distance d between the walls cannot be stabilized by means of an effective
potential, which is obtained by integrating out the internal dimensions of the elevendimensional heterotic M-theory action on the warped gravitational background. We
demonstrate this explicitly at the 2/3 level by showing the vanishing of the fourdimensional cosmological constant.
In the case where the volume does not depend on the orbifold coordinate one is
able to reproduce the relation between warp-factor and torsion of the weakly coupled
heterotic string.

2. Deriving the full relation between warped geometry and G-flux in heterotic
M-theory
Let us consider heterotic M-theory compactified on CY 3 S 1 /Z2 with four external coordinates x ; , , , . . . = 1, 2, 3, 4 and seven internal coordinates x u ; u, v, w, x, y, z =
5, . . . , 11. In the absence of any G-flux (for heterotic M-theory this amounts to considering
only the leading order which is M-theory itself without boundary or M5-brane sources)

G. Curio, A. Krause / Nuclear Physics B 602 (2001) 172200

175

the metric solution to the Killing-spinor equation, which describes a supersymmetrypreserving vacuum, is given by
ds 2 = dx dx + guv (x w ) dx u dx v ,

(2)

where guv decomposes into a direct product of the CalabiYau metric ga b and the metric
g11,11 of the eleventh dimension. Without loss of generality one can set g11,11 = 1. We
m,
will denote the six real CalabiYau indices l, m, n, p, q, . . . , while l,
n,
p,
q,
. . . , are
the, respective, flat tangent space indices. The alternative choice of holomorphic and
c . . . . Genuinely we have
anti-holomorphic indices will be denoted a, b, c . . . and a,
b,
to take the boundary sources into account which require turning on a G-flux in the internal
directions. This necessitates a more general metric, for which we choose the warp-factor
ansatz
ds 2 = gMN dx M dx N ,
ds = e
2

b(x n,x 11 )

+ ek(x

M, N = 1, . . . , 11,

dx dx + ef (x

n ,x 11 )

dx 11 dx 11.

n ,x 11 )

glm (x n ) dx l dx m
(3)

It will turn out that the appropriate G-flux of the relevant sources can be accommodated
with this ansatz. The most general ansatz which allows for arbitrary G-flux compatible
with supersymmetry will be considered in the last section.
The initial CalabiYau manifold possesses a closed Khler-form a b . However, a nonzero G-flux entails a non-trivial internal warp-factor ef , thereby rendering the deformed
Khler-form a b = ef a b non-closed. In this respect, the warp-factor ef serves as a
measure for the deviation from Khlerness of the internal complex threefold.
To preserve 4-dimensional Poincar-invariance, we set all components of G with at least
one external index to zero. 1 An important point is that in order to preserve supersymmetry
the magnetic sources on the right-hand-side of the Bianchi-identity must be (2, 2, 1) forms
[8]. I.e., they are forms with two holomorphic, two anti-holomorphic indices and one x 11 index. This is clear for the boundary sources and amounts for the M5-brane sources to an
orientation parallel to the boundaries. Solving the Bianchi-identity, we see from the fact
can
that the sources are (2, 2, 1) forms, that only the components Ga bc
, Ga b c11
d , Gabc11

become non-zero.
2.1. The Killing-spinor equation
The supersymmetry-variation of the gravitino in low-energy M-theory is given in the
full metric (3) by



I + 2 I J KLM 8gI J KLM GJ KLM ,
I = D
(4)

288
1 An external G
m 11
1234 = 21234 (x , x ) would be compatible with Poincar-symmetry but the known sources
(boundaries, M5-branes) do not give rise to such a flux. Moreover, compatibility with the Bianchi-identity allows
only a constant G1234 . Though such a sourceless constant field-strength is allowed by the non-compact external
spacetime, we will set it to zero subsequently.

176

G. Curio, A. Krause / Nuclear Physics B 602 (2001) 172200

where = e(x ,x ) . Here, is the original covariantly constant spinor and the
exponential-factor accounts for the correction if G-flux is turned on. We will assume
to be real and see later on that this is indeed compatible with supersymmetry in the warped
background. Subsequently, indices are raised and lowered with the full metric gMN , which
is also how contractions are performed in (4). Setting the variation to zero in order to obtain
a supersymmetry preserving solution, we obtain the Killing-spinor equation.
m

11

2.1.1. Covariant derivative contribution


Let us first deal with the part containing the covariant derivative of the Majorana-spinor
.
Using the definition of the spin-connection for the warped-metric 2

1 J

I J K (e)
I K J (e)
eJ
eK K
eJ J eK K eI I
,
J K I (e)
2
I J K (e)
= I eKJ

 J eKI
 ,

I JK (e)
=

(5)

allows to express the warped spin-connection through the initial one


1
l(e)
= el m e m b,
2

1
11 (e)
= e e11 11 11 b,
2
1
11
p
= e[m en ]l p f + l m n (e),
l m11
= eml
l m n (e)
e11 11 f,
(e)
2
1
11l11
(6)
= el m e11,11 m k,
(e)
2
and all other terms are zero. This is done to employ the covariant constancy DI =

(I + 14 I JK (e) J K ) = 0 of the initial spinor-parameter, which brings us to



I = dx u u + 1 dx  l l b +  11 11 b
dx I D
4

 1
1 l m

+ dx l m f + l 11 11 f + dx 11 11 l l k .
(7)
4
4
Let us now specify, that our internal space actually consists of a CalabiYau and a separate
eleventh dimension. The positive chirality condition 11 = on the original space
translates into

11 = ek/2 ,

(8)

on the warped space. The condition that we have a covariantly constant spinor (and its
complex conjugate) on the CalabiYau gives a = 0, a = 0 and translates into a =

0, a = 0. Using these relations plus the Dirac-algebra {a , b } = 2g a b , we end up


with 3
2 Here J, K,
 . . . denote flat 11-dimensional indices.
3 It is interesting to look at the Killing-spinor equation with general internal G-flux but without applying the
condition of Z2 invariance, i.e., 11 = , or using the CalabiYau condition a = 0 (which means that we

are considering an M-theory compactification on a smooth 7-manifold). In this case, which will be briefly treated
in Appendix A.1, the Killing-spinor equation can only be solved by trivial warp-factors and a vanishing G-flux.
Hence, M-theory compactifications on smooth 7-manifolds neither allow for nontrivial warped metrics nor for
nonzero internal G-flux. This agrees with the result of [2].

G. Curio, A. Krause / Nuclear Physics B 602 (2001) 172200

I =
dx I D

177






f
f
a +
dx a + a +
dx a 11 dx 11
4
4




1
1
1
+ ek/2 11 b dx  + ek/2 11 f dxa ek/2 a k dx11  a
4
4
4





1
1

+ a b dx a + b f dxa  a b .
(9)

4
4

2.1.2. G-flux contributions


Next, let us deal with the second term in the Killing equation, containing the G-flux.
To obtain condense expressions, it proves convenient to parameterize the three sorts of
allowed fluxes by defining
= lm np Glmnp ,

(10)

l =

(11)

mn

Glmn11 ,

lm = Glmnp ,
np

(12)

where a b = iga b , a b = ig a b denotes the Khler-form of the initial CalabiYau

manifold. The warped metric is related to the Khler-form by g a b = ief a b .


Subsequently, we will make use of
1 2f
i f

g a b g cd Ga bc
,
g bc Glbc11
l ,
= e
d = e
4
2
i

g cd Glmcd = ef lm ,
(13)
2
to express the occurring contractions through the above defined parameters. In order to
handle the various contractions of -matrices with the G-flux, it is convenient to evaluate

the expressions by first letting the matrices act on and employ g a11 = 0, 11 = ek/2 ,
a = 0. Taking as the ground state, a and  a can be regarded as annihilation and
creation operators, respectively. This leads to some useful identities (A.3) collected in
Appendix A. With their help, we establish the various contractions (A.4) of the five-index
-matrices with the G-flux and also the contractions (A.5) of the three-index -matrices
with G. These can be found in Appendix A, as well. Putting all this, together, we arrive at
the following expression for the second part of the Killing-spinor equation


dxI I J KLM 8g I J KLM GJ KLM



= 3ek/2f 4ia dx a + 12ia dx a ef dx11 3e2f dx 



+ 3ef ef dxa + 12i b a dxb 8ia dx 11  a



 a b 
 .
12iek/2f a dx a + 3ek/2 4ief a dxb 12Gc a b11

dxc

(14)
2.1.3. Complete Killing-spinor equation
Now, the complete Killing-spinor equation can be composed out of the two pieces (9)
and (14) and is given by

178

G. Curio, A. Krause / Nuclear Physics B 602 (2001) 172200


2 I

dx DI +
dx I J KLM 8gI J KLM GJ KLM
288



2 k/2f
1
=
a a f + i
e
a dx a
4
8



2 k/2f
1
e
+ a + a f + i
a dx a
4
24



2 k/22f
11
e

+ 11 dx
96



2 2f
1 k/2
e
+ e
11 b
dx 
4
24


2 2f
1
1

e
+ ek/2 11 f dxa
dxa + i ef b a dxb
4
24
2




2
ef k a dx11  a
ek/2 a k + i
3



2 k/2f
1
e
+ a b i
a dx a
4
6




1
1 k/2 c
2 k/2f
a b

e
+ b f dxa + i
a dxb e
G a b11

dxc
4
6
2
= 0.
I

2.2. The warp-factor flux relations


Setting the coefficients of the various -matrices to zero, we have to distinguish
carefully between the imaginary and the real part of the equations. For arbitrary vectors
AU , BU , the sum Aa Ba + Aa Ba is real, whereas the difference Aa Ba Aa Ba is purely
imaginary. Furthermore, is a real parameter.
2.2.1. The I,  and a -terms
From the terms proportional to the unit-matrix,  and a we receive the relations

2 k/2f
e
8a = a f = 2 a b = i
(15)
a ,
3

2 k/22f
e
411 = 11 b =
(16)
.
24
2.2.2. The  a -terms
The terms proportional to  a lead to

2 k/2f
e
a k = i
(17)
a ,
3
which shows that the warp-factors f and k are equal up to an additive function F depending
merely on x 11

G. Curio, A. Krause / Nuclear Physics B 602 (2001) 172200

179





 
f x W , x 11 = k x W , x 11 + F x 11 .

(18)

In the following we will set F (x 11 ) to zero since it can be eliminated by a simple


reparameterization of x 11 . Furthermore, the  a terms yield the relation

2 k/22f
1
e
11 f =
(19)
i ek/2f a a , no sum over a
24
2
together with

b a = 0,

b = a.

(20)

Note that in (19) there is no summation over the antiholomorphic indices a.


Hence (19)
implies the following isotropy-condition

1 1 = = n n ,
with n the complex dimension of the CalabiYau manifold. Using the identity
= 2i ef , it then follows that (19) simplifies to


2 1 1 k/22f
e

11 f =
.
4 6 n

n

a=
1

(21)
a a

(22)

2.2.3. The  a b -terms

Finally the  a b terms lead to an equation, which can be simplified, using the relation
for a f from (15), to

ief [a dxb ] = Gc a b11


dxc .

(23)

The component of this equation with c = a,


b leads to the following G-flux constraint
= 0,
Gcab11

c = a,
b,

(24)

whereas the c = a and c = b components simply reproduce the defining relation (11)
for a .
To summarize, the Killing-spinor equation leads to the set of (15), (16), (18), (22)
together with the G-flux constraints (20), (21), (24).
We are now in a position to briefly check that our assumption of choosing real does
not lead to inconsistencies. For this purpose it is enough to show that Im is constant,
which in particular means that a zero value can be maintained. Following [8], we use the
above equation for a and obtain





2 a b   3f/2 
a b
a b
g b e
a + a e3f/2b
g a b Im = g a b
=
2i
48

2 3f/2 m
e
D m .
=
48
Employing D m m = 0, which can be obtained from the field equation for G, one
establishes that Im is a harmonic function on a compact space and therefore has to be
constant.

180

G. Curio, A. Krause / Nuclear Physics B 602 (2001) 172200

3. Implications of the warped geometry


Let us now analyze the above equations in more detail. Notice, that up to now we
were not forced to specify whether we compactify on a CY 2 or a CY 3 the complex
dimension n of the CY n entered as a free parameter.
Arbitrary G-flux parameters , are only compatible with a pure warp-factor description of the internal deformed CalabiYau in the 6-dimensional case with n = 2, as we will
now see. For n = 2 we obtain

2 k/22f
,
e
11 f =
(25)
12
which says, together with (15), (16) that
8a = a f = 2a b,

(26)

811 = 11 f = 211 b,

(27)

and implies 8 = f = k = 2b. Here the warp-factors depend on both x m and x 11 . For
n = 2 the 11 f part receives a different prefactor and does not allow for this conclusion.
Instead as we will see explicitly for the case of n = 3 below one has to set either (26)
or (27) to zero to obtain a consistent solution.
The difference between the 4- and 6-dimensional cases was also pointed out in [8].
In contrast to CY 3 compactifications, a compactification on K = K3 S 1 /Z2 to six
dimensions allowed the inclusion of a general G-flux without the need to treat the Calabi
Yau coordinates and the orbifold coordinate differently. This led to a full derivation of
the relation between warp-factors and G-flux without the need for a first order truncation.
Starting on M 11 = R6 K with
ds 2 = eb dx dx + ef guv dx u dx v ,

(28)

the equation of motion


D u Guvwx = 0,

(29)

is solved by the ansatz (with the 2 (0) tensor in the original metric)
(0)
y w,
Guvwx = 2uvwxy

(30)

together, with the condition


(0) w = sources,

(31)

with the sources derived from the sources in the Bianchi equation dG = sources. The
searched for connection between the warp factor and the G-flux then takes the form
1/3
2/3


eb = c + 2 2 w
(32)
,
ef = c + 2 2 w
.
The fact that ef = (eb )2 which we also obtain from (26), (27) will again show up in
the 4-dimensional case without dependence on x 11 . The 6-dimensional metric is related
via the decompactification limit to the 11-dimensional extreme M5-brane metric ds 2 =

G. Curio, A. Krause / Nuclear Physics B 602 (2001) 172200

181

1/3 dx dx + 2/3uv dx u dx v where 1 = c + 2 2 w with w = q/R 3 and R =

xu x u for G taken to be the magnetic field of a point charge at x A = 0 of charge q.


But let us now turn to the 4-dimensional case. If we choose n = 3, we have

2 k/22f
e
.
11 f =
(33)
24
Taking mixed derivatives of f and b this implies that a 11 f = 0. A non-trivial solution
is either obtained from 11 f = 0 or a f = 0. The implications of these two cases will be
analyzed in more detail in the following two sections.

4. Connection between strong and weak coupling


In this section, we will present the connection to the heterotic string with torsion [3].
The choice, 11 f = 0, requires = 0, a = 0 and leads to
8(x m ) = f (x m ) = k(x m ) = 2b(x m)

(34)

without any dependence on x 11 . The required sort of fluxes is obtained by solving



11 z )S (x m ) dx 11 with m sources by G =
the Bianchi-identity dG = m
i i
i=1 (x
m
11
m
11
(x

z
)P
(x
)

dx
with
dP
=
Si . This type of geometry seems tailor-made
i i
i
i=1
for a smooth transition to the weakly coupled heterotic string, since any x 11 dependence
is lost. Indeed, we will now show that the heterotic M-theory relation between warpfactor and G-flux reproduces the corresponding relation (A.17) for the heterotic string
with torsion.
The warp-factor belonging to the 4-dimensional external part multiplies the Minkowskimetric both in the string and the M-theory case and is, therefore, fixed in the sense
that one does not have to take into account further coordinate-reparameterizations for a
comparison. Let us, therefore, start with the relation between external warp-factor and Gflux by using (15) for a b plus f = k = 2b and employing the definition of a to obtain

 b 
2
Gab b 11 .
a e
(35)
=
3
The contraction on the right-hand-side is with respect to g bc . To compare M-theory with
string-theory [11] one has to perform an overall Weyl-rescaling involving the dilaton,

= e2/3 gMN , which brings us to the string-frame. Here we have


gMN

 b 
2
2 2
e 3 Gab b 11 ,
a e
=
ds 2 = eb+ 3 dx dx + .
3
E = e /2 g in which
Finally, let us go over to the 10-dimensional Einstein-frame via gAB
AB
we obtained the heterotic string relation between warp-factor and torsion. We thus arrive
at



2
e 6 Gab b 11 ,
a eb =
(36)
3

ds 2 = eb+ 6 dx dx + ,

(37)

182

G. Curio, A. Krause / Nuclear Physics B 602 (2001) 172200

where again the contraction is performed with the metric of the actual frame, (g E )bc .
A comparison of the above metric with the heterotic string metric (A.8) shows that we
have to identify 2 with b + /6, which gives
11
(38)
.
6
If we use this in (36), we receive the heterotic M-theory warp-factor flux relation

 2 
4 2
a e
(39)
Gab b11 .
=
11
b=

Setting Gab b 11 equal to Hab b up to some constant normalization factor, we see that indeed
the relation between warp-factor and torsion of the heterotic string (see Appendix A.3
for relevant facts about the heterotic string with torsion and a derivation of the following
formula in that context)


1
a e2 = Hab b ,
(40)
2
can be reproduced from heterotic M-theory including G-flux. This represents a non-trivial
check on the duality between the strongly and the weakly coupled heterotic regions in the
presence of torsion.
The choice of fluxes treated in this subsection leads to a CalabiYau volume which does
not depend on x 11 . Moreover, due to the deformation with the warp-factor ef the Khlerform is no longer closed (cf., Eq. (A.13)). In addition the Ricci-tensor for the internal
six-manifold becomes




Ra b ef gmn = Ra b (gmn ) + ga b g cd 2c f d f + c d f a f b f + 2a b f, (41)
where the derivatives of f are determined through the G-flux by (15). Though Ra b (gmn )
= 0 due to the Ricci-flatness of the initial CalabiYau space, we recognize that in the
presence of G-flux the internal six-manifold also looses its property of being Ricci-flat.

5. The analysis of the warped geometry in the strong coupling case and Newtons
constant
The second choice, a f = 0, requires = 0, a = 0 and implies
 
 
 
 
4 x 11 = f x 11 = k x 11 = b x 11 ,

(42)

without any x m dependence. The necessary non-vanishing Ga bc


d and vanishing Gabc11
are

11 z )S (x m ) dx 11 through 4
obtained by solving the Bianchi-identity dG = m
(x
i
i
i=1

11
m
m
G= m
i=1 (x zi )Si (x ). Again Si (x ) is a closed 4-form representing the strength
of the ith magnetic source.
Note that for = 0, a = 0 the internal six-manifold remains Khler. This is due to the
fact that the warp-factor f does not depend on x m . Furthermore, we see from the general
4 The Heaviside step-function (x) is defined by (x < 0) = 0 and (x > 0) = 1.

G. Curio, A. Krause / Nuclear Physics B 602 (2001) 172200

183

formula (41) for the Ricci-tensor that in this case the six-manifold also keeps its property
of being Ricci-flat. In other words the six-manifold is still a CalabiYau space with volume
depending on the parameter x 11 .
5.1. The volume dependence on the orbifold direction
volume of theCalabiYau, as measured by the warped metric, is given by V (x 11 ) =

 The

6
d x gCY 3 = e3f d 6 x gCY 3 . The decisive part, which is responsible for the variation
11
of the volume with x , is the factor e3f . For its determination, we use f = k and the
equation for 11 f


1
11 e3f/2 = ,
8 2
which is solved by
e

3f (x 11 )/2

1
= e3f (0)/2
8 2

(43)

x 11
dz (z).

(44)

Notice that does not depend on the CalabiYau coordinates, which can be easily seen
by acting with a on (43) and taking into account that a f = 0. Hence the variation of the
CalabiYau volume with x 11 is given by

2
x 11
 11 
 m 
1 a b cd
V x = 1
dz Ga bc
V1 ,
d x , z
2 2

(45)

where V1 = d 6 x gCY 3 is the CalabiYau volume in the initial metric. The integration
constant e3f (0)/2 has been set to 1 to obtain a smooth transition from V (x 11) to V1 in case
that we turn off any G-flux. The only assumption about the full heterotic M-theory that we
will have to make is that the sources can still be localized at x 11 = zi in the eleventh

11 z )
direction, i.e., that the Bianchi-identity possesses the form dG = m
i
i=1 (x

m
Si (x m ) dx 11 . Its solution G = i=1 (x 11 zi )Si (x m ) then leads to the following
behaviour of the CalabiYau volume
2

m

 11 
 11
  11

V x = 1
(46)
x zi x zi Si V1 ,
i=1

1
m

a b cd (Si )a bc
d (x ).
2 2

where Si =
Thus we get the remarkably simple result that in
the full treatment the linear behaviour of the first order approximation gets replaced by
a quadratic behaviour.
For
the simplest case with only the two boundary sources at z1 = 0, z2 = d, we obtain
= 8 2 (x 11)S1 with S1 representing the magnetic source of the visible boundary.
This gives the warp-factor (beyond x 11 = x011 = 1/S1 , where the right-hand-side (rhs)
becomes negative, we assume an analytic continuation of the left-hand-side through
the rhs)

184

G. Curio, A. Krause / Nuclear Physics B 602 (2001) 172200

Fig. 1. The quadratic dependence of the CalabiYau volume on the orbifold direction in the full
geometry and its linear approximation to order 2/3 . If higher order contributions are negligible then
the linear approximation is valid for small x 11 . The left boundary corresponds to the visible world.

e3f (x

11 )/2

= 1 S1 x 11 ,

which leads to the following volume dependence (see Fig. 1)


2
  
V x 11 = 1 S1 x 11 V1 .

(47)

(48)

Moreover as becomes clear from the figure with the quadratic volume behaviour
tiny quantum effects are now able to resolve the zero volume as opposed to the linearized
case (cf. in this respect also [9,10]). We believe that the perfect square structure of the
volume and thereby its zero at the minimum is related to supersymmetry. The reason why
e3f/2 is linear in x 11 can be traced back through the derivation of the warp-factor-flux
relations to the fact, that we had imposed the condition (8). But this condition is nothing
but the chirality condition 11 = for the initial fermionic parameter , which had to
be imposed to preserve N = 1 supersymmetry in ten dimensions. On a more qualitative
level one may argue that in general a solution of the Einstein equations is a solution to a
second order differential equation with two integration constants. One gets a subclass of
supersymmetry-preserving solutions by solving the Killing-spinor equation instead. This
special subclass exhibits in contrast only one integration constant. We suppose that the
additional integration constant in the non-susy case parameterizes the ordinate position of
the locus of minimal volume and becomes zero in the supersymmetric limit. This indicates
that the zero could be lifted, not only by quantum corrections, but also classically by an
appropriate Susy-breaking.
Here and in the following the right boundary will not be depicted it would cut off
the solution at some finite distance d. To determine the actual value of S1 (and thereby
Newtons constant, see below) would require the actual knowledge of heterotic M-theory
to all orders in 2/3 . As we see, it is possible to parameterize our ignorance about these
higher-order terms by S1 . In the phenomenological relevant case where S1 > 0 (a positive
S1 is in accordance with the leading order approximation), a zero volume develops at
x011 = 1/S1 .
The x 11 dependence of the full metric reads 5
5 It is interesting to compare this solution to the domain-wall solution which arises in the effective

G. Curio, A. Krause / Nuclear Physics B 602 (2001) 172200


2/3
ds 2 = 1 S1 x 11
dx dx
2/3 


glm (x n ) dx l dx m + dx 11 dx 11 ,
+ 1 S1 x 11

185

(50)

or in terms of the CalabiYau volume




V (x 11) 1/3
ds 2 =
dx dx
V1



V (x 11) 1/3 
glm (x n ) dx l dx m + dx 11 dx 11 .
+
V1
5.2. Extracting the first-order results
Finally, one would like to see the transition of the full expression for V (x 11 ) to the
linearized expression which was derived in [8]. For this purpose one has to expand the
sources into a power-expansion in 2/3 . If we are only interested in sources coming from
the boundary, then we know that they start at the order 2/3
(1)

(2)

S1 = S1 2/3 + S1 4/3 + .

(51)

For the first order approximation, we have to truncate this series after the first term, which
indeed gives rise to a linear volume dependence
  



(1)
V x 11 = 1 2S1 2/3x 11 V1 + O 4/3 ,
(52)
11 = 1/(2S (1) 2/3 ) in the
as found in [8]. If we now read off the zero of V (x 11), we get xlin
1
linearized case, while the full solution gives a different first order zero



 (1)
x011 = 1/ S1 2/3 + O 4/3 .
(53)

This little puzzle is resolved by noticing that the linear approximation (52) holds true
only as long as S1(1) 2/3x 11 1 (plus similar conditions for the higher S1(i) , i  2
(1)
11 S (1) 2/3 x 11 = 1,
contributions). Because at the position of the zero, we face S1 2/3 xlin
0
1
the linear approximation (52) breaks down and cannot be used to determine reliably the
zero of V (x 11). Therefore, in contrast to the first order analysis, the actual zero at the first
order level becomes larger by a factor 2
11
.
x011 = 2xlin

(54)

5-dimensional M-theory [9]. In the 11-dimensional decompactification limit it is given by (use Eq. (4.19) of [9]
and set H1 = H2 = H3 = H (y))
ds 2 =

1
dx dx + H (d1 + d2 + d3 + H dy 2 ).
H

(49)

Here di are 2-dimensional line-elements spanning the internal six dimensions. Comparing both external
4-dimensional partsgives H V 1/3 , while from a comparison of the eleventh coordinate parts we gain the
reparameterization H dy = dx 11 . Noting that H y [9], we obtain x 11 y 3/2 and thus H (x 11 )2/3 .
Therefore, (50) and (49) show the same x 11 dependence for both the external and the internal parts. The
coordinate x 11 which we are using instead of say y is distinguished by simple Bianchi-identities of the form
dG = (x 11 zi )S(x m ) dx 11 .

186

G. Curio, A. Krause / Nuclear Physics B 602 (2001) 172200

(a)

(b)

Fig. 2. The figure shows the volume dependence in the presence of an additional M5-brane at zM5 ,
where we assume a positive M5-brane source SM5 . In (a) the situation for x011 < x011 < zM5 is
depicted while (b) shows the behaviour for x011 > x011 > zM5 . The second boundary at x 11 = d is
not depicted it would truncate the solution at the finite distance d.

This shows that if we place the hidden boundary not very close (such that S1(1) 2/3d
is not much smaller than 1) to the visible boundary, we are forced to take into account
higher order terms in 2/3 . In particular this applies for the phenomenologically interesting
region around the volume zero.
5.3. Inclusion of M5-branes
Let us briefly consider the case with three sources two from the boundaries
S
1 , S2 plus a further one SM5 from an M5-brane placed in between at zM5 . With =
8 2 [(x 11)S1 + (x 11 zM5 )SM5 ] we get a warp-factor
 


11
e3f (x )/2 = 1 x 11 S1 x 11 zM5 x 11 zM5 SM5 ,
(55)
and the following volume dependence (see Fig. 2)

2
1 S1 x 11 V0 ,
11
V (x ) = 
2
1 (S1 + SM5 )x 11 + SM5 zM5 V0 ,

x 11 < zM5 ,
x 11  zM5 .

(56)

The zero of the parabola for x 11  zM5 lies at x011 = (1 + SM5 zM5 )/(S1 + SM5 ). Thus we
see that an additional M5-brane will increase or decrease the slope of the volume parabola
depending on whether it contributes a positive or negative SM5 .
5.4. Newtons constant
Let us briefly recall the evaluation of Newtons constant in the first order analysis [8].
It is obtained by a dimensional reduction procedure together with an average CalabiYau
d
volume V  = d1 0 dx 11 V (x 11 ) through (1). It had been found [8] that



tr F 2 12 tr R 2 
12 
2
crit,lin

GN  GN
(57)

.
11

16 2 
8 2
8V1 xlin
CY 3

G. Curio, A. Krause / Nuclear Physics B 602 (2001) 172200

187

2
An approximation of the integral by 1/MGUT
delivers (with MGUT = 2 1016 GeV and
1 = 1/25) the lower bound


Gcrit,lin
N

1
,
(6.3 1018 GeV)2

(58)
phen

which is only slightly bigger than the actual phenomenological value GN = 1/(1.3
1019 GeV)2 .
In order to obtain effective 4-dimensional entities like GN or the hidden gaugecoupling 2 beyond the linear approximation, we have to keep the full x 11 dependence
of the fields and integrate out the seven compact dimensions (in contrast to performing
a simple dimensional reduction). Starting with the 11-dimensional EinsteinHilbert term






 
1
4
6
S = 2 d x d x dx 11 g (4) gCY 3 g11 R x 11 ,
(59)
2
we can use the full metric information (50) to explicitly integrate over x 11 and the Calabi
Yau coordinates (the integration over the CalabiYau coordinates will be contained in the
volume V (x 11 )). To gain a non-vanishing 4-dimensional curvature scalar, we consider
(4)
slight perturbations around the flat 4-dimensional spacetime, g . Keeping only
the 4-dimensional curvature-scalar part from the reduction of the 11-dimensional curvature
(4)
scalar, R(gMN ) = eb R(g ) + , we match the effective 4-dimensional Einstein
Hilbert action


 (4) 
1
(4)
4
,
d x g (4) R g
S =
(60)
16GN
with the following effective Newtons constant (in the downstairs picture [13], which we
are employing, an additional factor of two multiplying the integral over x 11 has to be taken
into account)
GN =

16

d
0

2
dx 11eb(x

11 )/2

V (x 11)

(61)

Note that it is not only the volume of the internal 7-fold but also an additional warpfactor stemming from the external 4-dimensional metric which enters the expression for
the effective GN . Though the CY-volume is manifestly positive, the square-root of the
warp-factor eb becomes negative beyond x011 if we use the analytic continuation mentioned
before (47). Notice, however, that it is only the square root of the warp-factors which
is continued into the negative region. The warp-factors themselves stay positive under
this continuation, as can be seen explicitly by comparing (3) with (50). Also physical
entities like the Ricci-tensor or the Riemann curvature scalar are well-behaved under this
continuation.
If we now insert the known warp-factor and volume dependences, we arrive at
2
1
2
  Gcrit
GN = Gcrit
,
N 
11
8/3
3
3 N
1 (1 d/x0 )
where Gcrit
N =

2
4V1 x011

(62)

approaches the lower bound of the first order analysis Gcrit,lin


if
N

11 . Notice that 2 Gcrit places a lower bound on Newtons constant which depends
x011 2xlin
3 N

188

G. Curio, A. Krause / Nuclear Physics B 602 (2001) 172200

(a)

(b)

Fig. 3. The dependence of GN on d is shown in (a). Qualitatively (modulo an external warp-factor)


this can be understood from the corresponding variation of the seven-fold CY 3 S 1 /Z2 volume
V (7-fold) with d, figure (b). Note that the decrease of V (7-fold) beyond x011 results from the analytic
continuation of the warp-factors into the negative region.

via x011 = 1/S1 on the source strength. In the first order approximation this lowers the
previously [8] obtained lower bound by a further factor 2/3, which is welcomed for
phenomenological purposes.
The dependence of GN on d is symmetric around the zero-position x011 (see Fig. 3(a)).
This implies that the effective Newtons constant does not only diverge at d = 0 (as it
already does at the zeroth-order approximation (1)) but also at d = 2x011 . Hence, the length
of the orbifold-interval has to be upper-bounded by d  2x011 . Notice, furthermore, that
qualitatively the inverse of the seven-fold CY 3 S 1 /Z2 volume

V (7-fold) =

d
6

d x2

dx
0

3
= V1 x011
5

11

gCY 3

d
g11,11 = 2




d 10/3
1 1 11
,
x0

dx 11ek(x

11 )/2

V (x 11 )

(63)

reflects the main features of GN (see Fig. 3(b)), as one would expect from the zerothorder formula (1). However, quantitatively they differ by the contribution of an additional
external warp-factor eb which appears under the x 11 integral of GN .
Both the CalabiYau volume and the warp-factors are symmetric with respect to x011 .
Therefore, it is clear that this property also holds for the gauge-coupling (as a function
of d) of the hidden boundary. Let us briefly derive the corresponding expression for the
hidden gauge-coupling. Starting with the 11-dimensional gauge-kinetic term



1
10
AC BD a
a
(4) g
S =
(64)
x
g

g FAB , FCD
,
d
CY 3 g
8(4 2)2/3
a components, which are supposed to depend solely on
we focus on the 4-dimensional F
x . Using (50), one obtains

G. Curio, A. Krause / Nuclear Physics B 602 (2001) 172200

189



V2
4
a
a
x
g (4) g (4) g (4) F
F
,
(65)
d
8(4 2)2/3
where again we denote V2 = V (d). The fact that the warp-factor contributions arising from
the 4-dimensional part of the full metric cancel each other is special to four dimensions and
leads to the simple 4-dimensional hidden gauge-coupling
S =

(4 2)2/3
(66)
.
2V2
Notice that the formula is the same as for the dimensional reduction (1). But here, we have
to take the full quadratic CalabiYau volume at hidden boundary position x 11 = d instead
of the first order linear volume. Again, we see that the hidden gauge-theory becomes
strongly coupled if the second boundary is placed in the vicinity of the zero of the Calabi
Yau volume.
2 =

6. The cosmological constant vanishes at 2/3 order


We saw previously that for the case with fluxes = 0, a = 0 we gain a quadratic
behaviour in x 11 for the CalabiYau threefold volume. The value of its zero is closely
linked to the value of the 4-dimensional Newtons constant. Hence, it is important to ask
whether such a phenomenologically favoured value for the modulus d could be stabilized
by means of an effective 4-dimensional potential. One could attack this problem by means
of the effective 4- or 5-dimensional action of heterotic M-theory derived in [14]. Here we
are going to examine the resulting potential for d for the derived warped gravitational
background by integrating out the internal dimensions of the original 11-dimensional
heterotic M-theory action. We will find, as one expects from supersymmetry arguments,
that this potential vanishes to first order in 2/3 . Thus, in agreement with the 4-dimensional
Minkowski-solution in (50) the effective 4-dimensional cosmological constant is zero at
this order. Unfortunately, the analysis at higher orders cannot be performed, yet, since
it would presuppose the complete knowledge of the heterotic M-theory action at the
appropriate orders, which is still lacking.
Let us start with the bosonic action of 11-dimensional heterotic M-theory [13], which
under the condition that only the G-flux component Ga bc
d contributes reads (i = 1, 2)
(1)

(2)

S = S11 + S10 + S10 ,



 
1
d 
 G
a bc
 gMN ) 1 G
S11 = 2 d 11 x g R(
,
a bcd
4
2




1
1
(1)
(2)
S10 = 2 d 10 x g (2) L(2) ,
S10 = 2 d 10 x g (1) L(1) ,
(67)
2
2
where the hats are meant to indicate that the full warped metric is used for determination
or contractions. To leading order in 2/3 the bosonic part of the boundary actions is given
by
 2/3


1

(i)
L(i) =
(68)
tr Fa b F (i)a b + O 4/3 ,
2 4

190

G. Curio, A. Krause / Nuclear Physics B 602 (2001) 172200

where to leading order in 2/3 the hats over the field-strengths have been omitted since
contractions must now be performed with the original zero-order metric.
6.1. The measure-factors
For the case with varying volume we found in the last
warp-factors relation
 section the

f (x 11) = k(x 11) = b(x 11). This allows to express g = e3b/2 gCY 3 in terms of the
measure on the original CalabiYau threefold without warp-factors. The condition to
preserve supersymmetry gave

 3b/2
2
,
11 e
=
16

which together with = 8 2 (x 11)S1 (note that for the case under consideration, we
have a = 0 which means S1 is a constant) leads to
e3b(x

11 )/2

= 1 S1 x 11 ,

thus determining the measure-factor as





g = 1 S1 x 11 gCY 3 .

(69)

11

Analogously the boundary measures are given by g (1) = eb(x =0,d) gCY 3 , which
leads to



(1)
(70)
g = gCY 3 ,
g (2) = (1 S1 d)2/3 gCY 3 .
6.2. The curvature-scalar
Next let us express the 11-dimensional curvature-scalar for the warp-metric


P Q
 gMN ) = g KL M M gKL K L gKL + KL
MN gP Q g KL g MN g KM g LN ,
R(
through the warp-factors b, f and the original CalabiYau curvature scalar. This gives

D(D + 1)
N (N + 1)
k
2
2

(11 b)2 +
(11 f )2
b + N 11
f+
R=e
D 11
4
4
DN
D
N
11 b 11 f 11 b 11 k 11 f 11 k + ef R(gmn ),
+
2
2
2
where D represents the real dimension of the non-compact external spacetime, while
N = 2n denotes the real dimension of the complex n-dimensional internal CalabiYau
manifold. For our concrete case with D = 4, n = 3 and f = k = b plus a Ricci-flat
CalabiYau manifold, we arrive at


2
 = eb 211
R
b + 52 (11 b)2 .
Using e3b/2 = 1 S1 x 11 plus the Z2 symmetry of the orbifold (which introduces a jump
in the first derivative of the metric at the orbifold fixed-points), we finally obtain


S12
2
8
(x 11 d)
11

R=
(71)
+ S1 (x ) +
9 (1 S1 x 11)8/3 3
(1 S1 d)5/3
for the curvature.

G. Curio, A. Krause / Nuclear Physics B 602 (2001) 172200

191

6.3. The field-strengths


Let us deal next with the field-strengths appearing in the bulk and boundary action.
Using Einsteins equation


1
N KLP
MKLP G
N
KLP Q G
KLP Q
MN 1 R
gMN = 1 4G
gMN G
R
2
24
2
 11 (1)
(2)  A B
11
M N ,
(x )TAB + (x d)TAB
(72)
with the boundary energymomentum tensor


1
1
L(i)
(i)
g AB L(i) AB
TAB
=
g11,11 2
g
 2/3 

1
1
(i) (i)C
(i) (i)CD
=
tr FAC
FB gAB tr FCD
F
g11,11 2 4
4
 4/3 
+O
,

(73)

we may substitute the G-flux kinetic term in the bulk action by the expression


1
d
 2 (x 11 )T (1)A + (x 11 d)T (2)A .
a bc
= 3R
G
(74)
d G
a bc
A
A
4
3
Thus the tree-level action becomes





(i)

1 
1
11
10
(i)AB L
(i)
(i)

7L
d x g 2R +
d x g
S= 2
2g
.
2
3
g (i)AB
i=1,2
(75)
For the boundary contributions let us concentrate on terms with two metric contractions
which take expectation values on the CalabiYau manifold. It is this class of terms which
appear at 2/3 order to which we ultimately have to truncate our results. This amounts
11
to an additional warp-factor e2f (x =0,d) = 1/(1 S1 x 11 )4/3 |x 11 =0,d from the metric
contractions.
6.4. The effective potential
Putting everything together, we can now integrate over the internal dimensions, which
can be done explicitly for the eleventh dimension. 6 We thus obtain for the effective treelevel action




(i)

1
1
10
lm L
(2)
S = 2 d x gCY 3

7L
2g
6
g lm
(1 S1 d (i) )2/3
i=1,2


1
(76)
+ 28S1 1 +
,
(1 S1 d)2/3
where we have defined d (1) = 0, d (2) = d. At this point, since the complete higher order in
2/3 terms of the boundary actions L(i) are still unknown, we have to truncate the action to


6 We work in the downstairs picture and employ d dx 11 = 2 d dx 11 .
d
0

192

G. Curio, A. Krause / Nuclear Physics B 602 (2001) 172200

the first non-trivial 2/3 order to proceed further. First for the boundary contributions, this
truncation gives
 2/3
3
L(i)

(i)
2g lm lm 7L(i) =
(77)
tr Fa b F (i)a b .
g
2 4
To proceed further we are going to derive a relationship between both tr F (i)2 terms
under the integral over the CalabiYau manifold. Let us start with the CY-integral




(i) (i)lm
6
lm
tr Flm F
tr Rlm R
,
d x gCY 3
(78)
i=1,2

and promote it to an integral over the CY 3 S 1 /Z2 manifold (we are working now at 2/3
order, which means that the integral measure as well as the contractions and the Riemann
curvature tensor have to be taken in the original unwarped geometry)







1

(i)
d 7 x gCY 3
(79)
tr Flm F (i)lm tr Rlm R lm x 11 d (i) .
2
i=1,2

Exploiting the supersymmetry-condition resulting from the vanishing of the gauginovariation, lm Flm = 0, plus the CalabiYau SU(3) holonomy condition, lm Rlm = 0, we
see that the integral is proportional to







1

(i)
(i)
d 7 x gCY 3 lm np
tr F[lm Fnp] tr R[lm Rnp] x 11 d (i)
2
i=1,2







1

tr F (i) F (i) tr R R x 11 d (i) dx 11 .

2
i=1,2

CY 3 Z2

The Bianchi-identity at this order reads [13]



 2/3  


1
1

(i)
(i)
dG =
tr F F tr R R x 11 d (i) dx 11,
2
2 2 4
i=1,2

which says that (78) is simply proportional to






dG =
d( G) =
[ G]x 11=0
[ G]x 11=d .
CY 3 Z2

CY 3 Z2

CY 3

CY 3

By the inverse operation as before we convert the last integral to






6
lm np
G
d x gCY 3 Glmnp
d 6 x gCY 3 .
CY 3

CY 3

(80)

CY 3

However, since this integral in general (hence in particular for our choice of fluxes) does
not depend on x 11 at 2/3 order [8], the integral is the same for both boundaries thus
rendering (78) vanishing.

G. Curio, A. Krause / Nuclear Physics B 602 (2001) 172200

193

If we neglect the tr R 2 contributions in (78) for the low-energy boundary action because
they are of higher order in derivatives, we then obtain the desired relation



(81)
d 6 x gCY 3 tr F (1)2 =
d 6 x gCY 3 tr F (2)2 .
CY 3

CY 3

2/3

Together with the


order value for the source (again we suppress the higher order in
derivative Riemann-curvature term for the low-energy action)
 2/3
1

(1)
S1 =
(82)
tr Fa b F (1)a b ,
4 4
this gives the following tree-level effective action


 2/3 
1
11

(1)
S = 1 +
(83)
d 10 x gCY 3 tr Fa b F (1)a b .
2/3
2
(1 S1 d)
12 4
Recognizing that the whole expression has to be truncated to 2/3 order, the expansion of
the prefactor for S1 d 1 simply results in 1 + 1 = 0, which shows the desired result,
namely that at this order the (tree level) cosmological constant vanishes.

7. The case with general flux: beyond warped geometries


If one wants to relax the constraint that either or a is zero (which we adopted up to
now), then one has to generalize the previous pure warped geometry to a geometry which
exhibits a deviation from the CalabiYau metric and is not describable by a warp-factor
alone. With such a generalized ansatz
ds 2 = gMN dx M dx N
= eb(x

W ,x 11 )

+ ek(x




dx dx + glm (x n ) + hlm x n , x 11 dx l dx m

W ,x 11 )

dx 11 dx 11,

(84)

we will see that the inconsistency which arose for the warp-factor f if both and a were
present, disappears and instead leads to constraints on the internal spin-connection.

The CY 3 metric split entails a corresponding split for the internal Vielbein ellm =

ellm (x n ) + fllm (x n , x 11 ). Again, we will express the spin-connection through the one
belonging to the initial metric
1
1
l(e)
= el m e m b,
11 (e)
= e , e 11 11 11 b,
2
2
(d)
= l m n (e) + l m n (e, f ),
l m n (e)


1
m m l
l m11
= e 11 11 11 fml
+ em
,
el l 11 flm
(e)
2
1
11lm (e)
(85)
= e[l l |11| fm ]l ,
11l 11 (e)
= ell e 11,11 l k,
2
with all remaining terms vanishing. Now the deviation from the initial CY-geometry

is characterised by fllm and l(d)


m
n (e, f ). Both go to zero if we turn off the G-fluxes.

194

G. Curio, A. Krause / Nuclear Physics B 602 (2001) 172200




Employing again the covariant constancy DI = (I + 14 I JK (e) J K ) = 0 of the


original spinor-parameter, leads to



I = dx u u + 1 dx l l b + 11 11 b
dx I D
4

1 l  (d) mn
+ dx lmn + 2lm11(e)
m 11
4


1 11 l
lm



.

+ dx 11 l k + 11lm (e)
(86)
4
Again, specifying that our internal space consists of a CalabiYau times an interval, we

employ 11 = ek/2 and a = 0, a = 0 plus the Dirac-algebra {a , b } = 2g a b to


obtain



1 (d)a 1 (d)a
I 

dx DI =
l a l dx l
4 la
4



1
1
11
a
a
11a (e)
+
11a (e)
11 dx
4
4


1
+ ek/2 11 b dx 
4


1 k/2
1 k/2
l
+ e
l a11
dx e
a k dx11  a
(e)
2
4





1
1 (d) l 1
a
11  a b

+ a b dx + l a b dx + 11a b (e)
dx
.
(87)
4
4
4
For the second part of the Killing-equation which consists of the contractions of matrices with the G-flux, it will be convenient to use the following abbreviations

G = g a b g cd Ga bc
d ,

(88)

bc

Gm = g Gmbc11
,

(89)

Gmn = g Gmncd .

(90)

cd

In order to eventually extract the real and imaginary parts of the Killing-spinor equation,
we have to know their behaviour under complex conjugation, which is given by
Ga b = Ga b .

(91)

Analogously, to the treatment in the previous section, we then arrive at




dxI I J KLM 8g I J KLM GJ KLM



= 3ek/2 24Ga dx a 8Ga dx a + 4G dx11 + 12G dx 



+ 12 G dxa 6Gb a dxb + 4Ga dx 11  a + 24ek/2Ga dx a



 a b 
 .
+ 12ek/2 2Gb dxa 3Gc a b11
dxc

(92)

G = G,

Ga = Ga ,

With (87) and (92) we are then able to decompose the complete Killing-equation (4) into
its external, CY- and 11-components. Thus unbroken supersymmetry finally translates into
the following constraints on the spin-connection

G. Curio, A. Krause / Nuclear Physics B 602 (2001) 172200


(d) b
ab

b
a(d)
b

2 2 k/2
=
Ga ,
e
3

(d) a b c = 0,


2 k/2 
(d) c
cd
e
+ 3Gc ab11 ,

Gd ab
ab =
6
11a a (e)
= 11a a (e),

= 11ab (e)
= 0,
a11
(e)


2 k/2 b
e
Ga 6Gb a ,
=
b a11 (e)
12

195

(93)
(94)
(95)
(96)
(97)
(98)

cd = c d c d . Additionally, the solution to the Killing-spinor equation provides


where ab
a b
b a
us with further equations, which determine the warp-factors and covariant-spinor deviation
in terms of the G-flux parameters

2 k/2
e
Ga ,
a b =
(99)
3

2 k/2
e G,
11 b =
(100)
6

2 2 k/2
e
a k =
(101)
Ga ,
3

2 k/2
e
Ga ,
a =
(102)
12
2 k/2
e G.
11 =
(103)
24
Similarly to the last section we obtain






8 x m , x 11 = k x m , x 11 = 2b x m , x 11 ,
(104)

but this time a dependence on both x m and x 11 is allowed. Note that this relation is in
accordance with the result of the first
 order approximation derived in [8].
= 1 v g between the Christoffel-symbols and the metric
The relation u uv (g)
g

determinant enables us to to find







 1
u
11 gCY 3 = gCY 3 u11 g CY 3 I 11 k .
2

(105)

Via the relation between the Christoffel-symbols and the spin-connection, eu x x vw =


v eu w + v u x ex w , we get u u11 (gCY 3 I ) 12 11 k = u u11 (eCY 3 I ) and thereby


11 gCY 3 = gCY 3 l l11 (e),
(106)

(e)
= 0. Together, with the constraint on b a11 (e),
which
where we have used that 11 1111

6+n k/2
a
a

= a11
= 2 12 e G (n = dimC CY n ), we obtain ultimately
gives a11 (e)
(e)



6 + n k/2
11 ln gCY 3 = 2
(107)
e G.
6

196

G. Curio, A. Krause / Nuclear Physics B 602 (2001) 172200

Employing the equation for 11 b, we can integrate this equation to obtain the following
expression for the CalabiYau dependence on x 11 (with n = 3)



 11 
11 m
6
V x =
(108)
d x gCY 3 =
d 6 x e9b(x ,x ) C(x m ),
CY 3

CY 3

where C(x m ) arose as an integration constant by integrating (107) over x 11 . We see that
now the specification of the sources simply by means of their location in the eleventh
direction is not enough to determine V (x 11 ). This stems from the fact that G, which

determines the warp-factor b contains contractions with g a b which itself is x 11 dependent.


11 z ) does not fully determine the x 11
Therefore, the specification Ga bc
i
d (x
behaviour of G. However, it is definitely true that a non-trivial V (x 11) requires G = 0
and, therefore, in view of (100) Ga bc
d = 0.
Another interesting aspect of turning on both Ga bc
derives from the fact,
d and Gabc11
that we saw above how the fluxes determine the internal spin-connection (e).
Now it
is well-known that the spin-connection determines the holonomy-group H of a manifold
through the path-ordered exponential of (e)
around a closed curve
Pe

m
m (e)dx

H.

(109)

This is an interesting further link between the physics of G-fluxes and the geometry
of the compactification space. A complex 3-dimensional Khler-manifold exhibits U(3)
holonomy. But we already saw above that turning on a G-flux in general ruins the closeness
property of the Khler-form therefore, the new deformed manifolds are no longer
Khler. This means that we do expect a more general holonomy than U(3).

Acknowledgements
We would like to thank K. Behrndt, S. Gukov, D. Lst, A. Lukas, B.A. Ovrut and in
particular E. Kiritsis for discussion. A.K. is supported by RTN project HPRN-CT-200000122.

Appendix A
A.1. M-theory on a 7-manifold
In this appendix let us consider as an aside the compactification of M-theory on a smooth
7-manifold. Our aim is to investigate which kind of internal G-flux can be turned on if at the
same time supersymmetry shall be preserved and which sort of warp-factors will appear.
Let us, therefore, start with the warp-ansatz
w

ds 2 = gMN dx M dx N = eb(x ) dx dx + ef (x ) guv dx u dx v ,

(A.1)

G. Curio, A. Krause / Nuclear Physics B 602 (2001) 172200

197

for an initial 7-manifold with metric guv . We will assume only an internal nonzero G-flux.
To solve the equation of motion for G, we write G =J dU , with U a 2-form and the Hodgeoperation is taken with respect to the 7-manifold. In components this reads
1 f/2
2uvwxz1 z2 z3 z1 U z2 z3 ,
e
(A.2)
4!
where 2uvwxz1 z2 z3 is defined as a tensor (rather than a tensor-density) containing a factor

det g uv . The effect of a nonzero G on the fermionic supersymmetry-variation parameter
u
will be considered bu using = e(x ) instead of the initial . Plugging this into the
Killing-spinor equation for the gravitino eventually leads to

1
1
dx u u + dx u u b + dx u uv v f
4
4

2 f/2
2 f/2 u vw
 uvw u Uvw

e
dx
e
dx  [u Uvw]

48 73
48 72


2 f/2 u  vwx
+
e
dx

u
v wx = 0.
6 74
Guvwx =

This equation is solved by setting the various coefficients in front of the independent
Gamma-matrices to zero, which shows that the warp-factors b, f and the spinor-correction
factor all have to be trivially constant. In addition [u Uvw] has to be zero, which amounts
to a zero internal G-flux. Thus M-theory compactifications on smooth 7-manifolds which
are supposed to preserve some supersymmetry do not allow for non-zero internal G-flux
and require at the same time trivial warp-factors.

A.2. G-flux contractions


First, we present those identities, which are used in the evaluation of the Killing-spinor
equation

ea bcd = 0,
 

d



bc
= e g a b g cd g a d g cb + b g a d g ce g a e g cd
ea



+ d g a e g cb g a b g ce ,


abcd = 0,
a bcd = g a b g cd g a d g cb ,




a bcd =  a b g cd  a d g cb + b d g ca ,

abc11
= 0,


ab
c11

h/2  a bc
c ba
abc




= 0,

= e
g g ,
 a b c


b
a
b
c

a
c

a
b
 = g  g  ,

 = g a b .

With their help and the definitions (10), (11), (12), we arrive at the contractions



uvwx Guvwx = 3 e2f  + 4iek/2f a   a ,


euvwx
k/2f
e

Guvwx = 12ie
,

(A.3)

198

G. Curio, A. Krause / Nuclear Physics B 602 (2001) 172200

euvwx
Guvwx = 3 e2f e 4ief e a  a 4iek/2f e

+ 4iek/2f a e  a + 4ek/2 a b Ge a b11

11uvwx Guvwx = 3ek/22f ,

(A.4)

and
g au vwx Guvwx = 3iek/2f a ,

g u vwx Guvwx = 0,


au
vwx
k/2f a
f a e
k/2 b c a
G bbc11
Guvwx = 3 ie
+ ie e e
,

g 11u vwx Guvwx = 3ief c c g 11,11 ,

(A.5)

which are used in the main text.

A.3. The heterotic string with torsion


The ansatz traditionally used [1] in compactifications of the 10-dimensional heterotic
E8 E8 string on CY 3 to four dimensions with N = 1 supersymmetry is to make the
susy-variations of the gravitino M , the dilatino and the gluino a (i, j = 5, . . . , 10)

1
 j kl
j
i = Di +
i 9i kl Hj kl ,

32g 2

1
1
ij k Hij k , = ij Fij
= ( ) +
(A.6)
4g
2
8 2 g2
vanish by assuming that H = d = 0. Here is the dilaton and H the gauge-invariant field
strength of the NSNS 2-form B, which in addition has to fulfill the Bianchi identity
1
tr F F.
(A.7)
30
This leads to the consequence that CY 3 is a Khler manifold with c1 (CY 3 ) = 0 and SU(3)
holonomy (and the gauge field A being a holomorphic connection on a holomorphic vector
bundle V over the CalabiYau threefold CY 3 obeying the DonaldsonUhlenbeckYau
equation).
This ansatz was generalized in [3] to include a non-vanishing torsion H = 0, where
solutions leading again to N = 1 supersymmetry in D = 4 were obtained by allowing for
a warp-factor e2D(y) in the metric (in Einstein-frame)


0
E
(x, y) = e2D(y)gAB (x, y) = e2D(y)
gAB
(A.8)
,
0 gmn (y)
dH = tr R R

where we denote 10-dimensional indices by A, B, C, . . . . It turns out that D has to be the


dilaton . The torsion and the dilaton are determined by
i
H = ( )J,
2
e8 = e80 ||||,

(A.9)
(A.10)

where the fundamental (1,1) form J is built out of the complex structure Jm as J =
1
b
m
p
a
n
2 Jm gnp dy dy = iga b dz d z (in our conventions J equals up to a minus-sign the
n

G. Curio, A. Krause / Nuclear Physics B 602 (2001) 172200

199

Khler-form ) and is the (determined up to an overall constant) holomorphic 3-form


 g a1 b1 g a2 b2 g a3 b3 )1/2 . To recognize the relation between
with norm |||| = (a1 a2 a3
b1 b2 b3
H (commonly called torsion) and the original torsion, we note that the metric torsion of a
complex manifold is specified by the expression

a
Tbc
= 2g da gd[b,c]

(A.11)

and its complex conjugate. Hence, the above expression for H can be explicitly expressed
through the metric torsion via

1

a
b
c
H = Ta b c dza dzb dzc + Tcab
(A.12)
dz dz dz .
4
Finally, the link between H and the warp-factor is given implicitly by the dilatino equation
= 0, which manifests itself in the following relationship
ln |||| = 8i( )(

d J = i( )
0 ).

(A.13)

From the left-hand side of this equation it can be easily discerned, that the right-hand side
serves as a measure for the non-Khlerness of the compactification manifold. Therefore,
by turning on H -torsion, the compactification manifold becomes deformed to a manifold
which is no longer Khler.
To gain a more explicit relation between the H -torsion and the resulting warp-factor, we
note that the dilatino equation = 0 can be alternatively written as [3]
8m = Jm n p Jn p .

(A.14)

Here, the covariant derivative is constructed out of the initial metric gMN without warpfactor. The H -covariant constancy of the complex structure [3]
m Jm n Hqm p Jn q H q mn Jq p = 0,
plus its property to
iga b to derive

square to minus the identity, Jm n Jn p

8a = Hab b Ha b b .

(A.15)
p
= m ,

serve together with Ja b =


(A.16)

The contraction is with respect to the initial metric in whose frame the relation holds.
Eq. (A.9), which relates H -torsion with metric torsion, reads in components Habc =

and leads to Ha b b = Hab b . Finally, to obtain the relation between the warpgc[a,b]

factor and H -torsion in the Einstein-frame, we have to transform the contractions


E from the initial frame to the Einstein-frame
according to the rescaling gAB = e2 gAB
and gain
1
a (e2 ) = Hab b .
(A.17)
2
E .
The contraction on the right-hand-side is now understood to be carried out with gAB
Furthermore, warp-geometries appear in heterotic five-brane solutions preserving
supersymmetry. They were obtained ([46]; cf., also the axionic instantons in [7]) with
the Ansatz (k, l, m, n = 7, . . . , 10)
gmn = e2 mn ,

Hmnl = 2mnl k k ,

200

G. Curio, A. Krause / Nuclear Physics B 602 (2001) 172200

showing again that turning on torsion leads to a warp factor.

References
[1] P. Candelas, G. Horowitz, A. Strominger, E. Witten, Vacuum configurations for superstrings,
Nucl. Phys. B 258 (1985) 46.
[2] B.S. Acharya, B. Spence, Flux, supersymmetry and M-theory on seven-manifolds, hepth/0007213.
[3] A. Strominger, Superstrings with torsion, Nucl. Phys. B 274 (1986) 253.
[4] A. Strominger, Heterotic solitons, Nucl. Phys. B 343 (1990) 167.
[5] C. Callan, J. Harvey, A. Strominger, Worldsheet approach to heterotic instantons and solitons,
Nucl. Phys. B 359 (1991) 611.
[6] C. Callan, J. Harvey, A. Strominger, Worldbrane actions for string solitons, Nucl. Phys. B 367
(1991) 60.
[7] C. Kounnas, E. Kiritsis, D. Lst, A large class of new gravitational and axionic backgrounds for
four-dimensional superstrings, Int. J. Mod. Phys. A 9 (1994) 1361, hep-th/9308124;
C. Kounnas, E. Kiritsis, D. Lst, Non-compact CalabiYau spaces and other non-trivial
backgrounds for 4D-superstrings, essays on mirror manifolds II, hep-th/9312143.
[8] E. Witten, Strong coupling expansion of CalabiYau compactification, Nucl. Phys. B 471
(1996) 135, hep-th/9602070.
[9] K. Behrndt, S. Gukov, Domain walls and superpotentials from M-theory on CalabiYau
threefolds, Nucl. Phys. B 580 (2000) 225, hep-th/0001082.
[10] A. Strominger, Loop corrections to the universal hypermultiplet, Phys. Lett. B 421 (1998) 139,
hep-th/9706195.
[11] E. Witten, String theory dynamics in various dimensions, Nucl. Phys. B 443 (1995) 85, hepth/9602070.
[12] P. Horava, E. Witten, Heterotic and type I string dynamics from eleven dimensions, Nucl. Phys.
B 460 (1996) 506, hep-th/9510209.
[13] P. Horava, E. Witten, Eleven-dimensional supergravity on a manifold with boundary, Nucl.
Phys. B 475 (1996) 94, hep-th/9603142.
[14] A. Lukas, B.A. Ovrut, K.S. Stelle, D. Waldram, Heterotic M-theory in five dimensions, Nucl.
Phys. B 552 (1999) 246, hep-th/9806051;
A. Lukas, B.A. Ovrut, D. Waldram, On the four-dimensional effective action of strongly
coupled heterotic string theory, Nucl. Phys. B 532 (1998) 43, hep-th/9710208.
[15] K. Becker, M. Becker, M-theory on eight-manifolds, Nucl. Phys. B 477 (1996) 155, hepth/9605053.
[16] K. Dasgupta, G. Rajesh, S. Sethi, M-theory, orientifolds and G-flux, JHEP 9908 (1999) 023,
hep-th/9908088.
[17] C.S. Chan, P.L. Paul, H. Verlinde, A note on warped string compactification, hep-th/0003236.
[18] B.R. Greene, K. Schalm, G. Shiu, Warped compactifications in M- and F-theory, hepth/0004103.
[19] K. Becker, M. Becker, Compactifying M-theory to four dimensions, hep-th/0010282.
[20] K. Becker, A note on compactifications on spin(7)-manifolds, hep-th/0011114.
[21] B.R. Greene, K. Schalm, G. Shiu, Dynamical topology change in M-theory, hep-th/0010207.

Nuclear Physics B 602 (2001) 201212


www.elsevier.nl/locate/npe

Note on noncommutative tachyon in matrix models


Miao Li a,b
a Institute for Theoretical Physics, Academia Sinica, Beijing, PR China
b Department of Physics, National Taiwan University, Taipei 106, Taiwan

Received 7 November 2000; accepted 15 March 2001

Abstract
The solution representing a braneanti-brane system in matrix models breaks the usual matrix
spacetime symmetry. We show that the spacetime symmetry on the branes is not breaking, rather
appears as a combination of the matrix spacetime transformation and a gauge transformation. As a
result, the tachyon field, itself an off-diagonal entry in longitudinal matrices, transforms nontrivially
under rotations, decomposing into tensors of different ranks. We also show that the tachyon field can
never be gauged away, and conjecture that this field is related to the usual complex scalar tachyon
by a field redefinition. We also briefly discuss tachyon condensation. 2001 Published by Elsevier
Science B.V.

1. Introduction
Tachyon condensation has been one of the focus points of recent research in string
theory [1,2]. One apparent motivation in studying this issue is to understand the dynamic
process of the braneanti-brane annihilation. More profoundly, revelation of elements in
this problem will shed new lights on the deep connection between the open string sector
and the closed string sector, and on the old problem of background independence in string
theory. The tachyon condensation problem has been proven considerably simplified when
a large B field is present [36], thanks to some new features of noncommutative field
theory [8].
As conjectured recently by Witten [7], it may be possible to realize any configuration
in, for instance the IIB theory, by starting with infinitely many D9 and anti-D9 branes.
It is also very interesting to ask whether it is possible to derive the IIB matrix model
or some revision of it [9,10], by starting with D9anti-D9 system and turning on tachyon
condensation. We believe that the answer to this question is likely yes. The reason for
this belief is the following. Consider for instance a single D9-brane with a large B field
E-mail address: mli@phys.ntu.edu.tw (M. Li).
0550-3213/01/$ see front matter 2001 Published by Elsevier Science B.V.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 2 5 - 0

202

M. Li / Nuclear Physics B 602 (2001) 201212

switched on. The open string sector decouples from the closed string sector in the Seiberg
Witten scaling limit [8]. If the rank of B is maximal (in the Euclidean spacetime), then
one can rewrite the noncommutative YangMills theory as a 0-dimensional matrix model
in an operator formulation. Since there is no tachyon in this system, it is not possible to
realize other D-branes and closed strings yet. By adding anti-D9-branes, one would get a
matrix model different from the IIB matrix model: if both D9-branes and anti-D9-branes
are immersed in the same B field background, again one has a noncommutative field theory,
one that is different from the IIB matrix model in an operator formulation.
One can study this problem by reversing the above procedure. One starts with the IIB
matrix model and constructs D9anti-D9 solution. This solution is a system quite different
from D9anti-D9 embedded in the same constant B field background. The reason for this
is quite simple. If the D9-brane is represented by a solution [Xi , Xj ] = i ij , then an antiD9-brane is represented by a solution [Xi , Xj ] = i ij , corresponding to reversing the
orientation of the D9-brane solution by making reflection of 5 dimensions. However, a
D-brane solution in matrix models does not come for free, it always carries a background
gauge field Fij which is just the inverse of ij . Indeed for this solution, as first observed
in [11], a noncommutative super YangMills theory directly results from the matrix model.
For the braneanti-brane solution, there are different background gauge fields on D9branes and anti-D9-branes. It is not known how to write down a simple field theory with
an associative algebra generalizing the star product. On the D9-branes, the star product
is defined with the noncommutative parameters ij , while on the anti-D9-branes, the star
product is defined with ij . If one naively generalizes the product to include both, one
will not get an associative algebra.
As we mentioned before, if one starts with a background in which the gauge field
strengths on all branes are the same, then one has the usual noncommutative field theory.
Other different field strengths can be achieved by adding perturbations. To simplify the
situation, we shall focus on two coordinates, and the resulting theory can either be
regarded as D-string anti-D-string in the IIB matrix model, or D2-brane anti-D2-brane
in matrix theory. The matrix model solution of this configuration explicitly realizes the
above idea. As a result, the solution breaks the usual matrix model rotational symmetry.
This can be seen in two different ways. The simpler one is to simply take a look at
the solution X1 = diag(x 1 , x 1 ), X2 = diag(x 2 , x 2 ) with [x 1, x 2 ] = i . The first entries
represent the D-string, and the second entries represent the anti-D-string. This solution
is not invariant under the rotation Xi = ij Xj . The second way to see this is to write
X2 = diag(x 2 , x 2 ) + diag(0, 2x 2). The second term can be regarded as a perturbation in
the gauge field on the second string and this field depends explicitly on coordinates. This
is a puzzle, since we believe that even though the field strengths on different strings are
different, the whole system is invariant under a rotation. We will see in the next section
that the rotational symmetry is still present in the solution. This symmetry is a mixture of
the matrix rotation and a gauge transformation. In Section 3, we shall see that with the
presence of gauge fields, the mixture is more complicated. This is not surprising, since
gauge fields can be understood as longitudinal oscillations of branes, thus it is intimately
related to spacetime symmetry.

M. Li / Nuclear Physics B 602 (2001) 201212

203

The scalar nature of the tachyon presents another puzzle. We expect this mode to arise
from the off-diagonal entries of Xi , and these entries carry longitudinal spacetime indices.
This puzzle again is related to the lacking of matrix rotational symmetry. We will in
Section 4 see that one of the complex off-diagonal entry can be gauged away, the remaining
off-diagonal entry is tachyonic. Under the rotation about a point, the tachyon can be divided
into modes with different tensor structures. We attribute this unusual feature to the bad
representation of braneanti-brane system in matrix models. We believe that if one can
write down an associative algebra without breaking the matrix rotational symmetry, the
scalar nature of the tachyon will become apparent. We briefly discuss tachyon condensation
in the final section.
A paper discussing related issues appeared very recently [14].

2. Rotational invariance
In the Euclidean signature, the braneanti-brane solution can be represented as




X2 = diag x 2 , x 2 .
X1 = diag x 1 , x 1 ,
Here x 1 , x 2 are noncommutative coordinates satisfying
 1 2
x , x = i.

(2.1)

(2.2)

Apparently, the braneanti-brane system still respects the rotational invariance on the
(x 1 , x 2 ) plane, this is a consequence of the first quantized strings for them the world-sheet
action is invariant under rotations. For an infinitesimal rotation, x 1 = x 2 , x 2 = x 1 ,
we expect the matrices X1 , X2 transform in the same way. This is not the case for the
ansatz (2.1), since X1 is proportional to the identity matrix while X2 is proportional to
3 . Thus we can hardly obtain a manifestly rotationally invariant action or Hamiltonian
starting with ansatz (2.1).
In matrix models, in addition to translational and rotational symmetry, the only other
symmetry readily available is gauge symmetry. Applying a rotational transformation
directly to X1 , X2 would bring us away from the ansatz (2.1). To come back to (2.1), one
may try to apply a gauge transformation. It is easy to guess what kind of gauge parameter
is needed. Since the first diagonal entries transform correctly under the matrix spacetime
transformation, thus the gauge parameter has a vanishing first diagonal entry. Assume g =
diag(0, f ), where f is a function of x 1,2 , the gauge transformation takes the form g Xi =
i[g, Xi ]. Now apply the rotational transformation




X1 = diag x 2 , x 2 ,
X2 = diag x 1 , x 1 .
Applying further a gauge transformation, in order to keep the ansatz (2.1) invariant, we
need


x 2 + i f, x 1 = x 2 ,


x 1 i f, x 2 = x 1 .
(2.3)

204

M. Li / Nuclear Physics B 602 (2001) 201212

The apparent solution to the above conditions is




f = (x 1 )2 + (x 2 )2 .

The corrected rotational transformation therefore is




X1 = ( + g )X1 = diag x 2 , x 2 ,


X2 = ( + g )X2 = diag x 1 , x 1 .

(2.4)

(2.5)

This is of course a symmetry since both terms on the R.H.S. are symmetries in the matrix
model.
Note that the gauge transformation with parameter (2.4) is rather singular in the usual
sense, for f diverges for large r 2 . This is not surprising however, since the rotational
transformation itself diverges too.
Similarly, ansatz (2.1) does not conform with the conventional translational transformation Xi = a i 12 , where 12 is the identity matrix. To preserve the form (2.1), the
conventional transform is to be accompanied by a gauge transformation with a gauge parameter


g = diag 0, 2(a 2/ )x 1 .
(2.6)

3. The gauge fields


The same issue of maintaining rotational transformation remains when we perturb from
the constant background (2.1). For instance, consider switching on the U (1) gauge field
corresponding to the center of mass degree of freedom. In case of two coincident D-branes,
the ansatz in matrix models is [11]


Xi = x i + ij Aj 12 .
(3.1)
The commutator [X1 , X2 ] is given by
 1 2
X , X = i (1 + F12 )12 ,

(3.2)

where F12 is the noncommutative field strength


F12 = 1 A2 2 A1 i[A1 , A2 ],

(3.3)

and we implicitly used the star product defined by the noncommutative parameter . Thus,
the action is proportional to
 2
d x 2
(1 + F12 )2 .
(3.4)
2
As noticed in [12], 2 is to be interpreted as the open string metric factor Gij . The integral
factor is simply the phase space factor corresponding to the matrix trace.
The correct ansatz generalizing both (2.1) and the usual brane-brane system is




X2 = x 2 A1 3 .
X 1 = x 1 + A2 12 ,
(3.5)

M. Li / Nuclear Physics B 602 (2001) 201212

205

The commutator [X1 , X2 ] is the same as in (3.2) with an additional factor 3 . This does not
change the action. Again, under the usual rotational transformation Xi = ij Xj , the
above ansatz is not preserved. We expect that with the coordinates rotation x i = ij x j ,
the gauge field Ai transform in the same fashion, namely, Ai = ij Aj . This together
with the ansatz (3.5) implies that an additional gauge transformation is again needed:


Xi = Xi + i g, Xi .
(3.6)
Again, g is diagonal and assumes the form g = diag(0, 2 f ). The conditions that f must
meet are




i f, x 1 + A2 = x 1 A1 ,


i f, x 2 A1 = x 1 + A2 .
(3.7)
The commutators in the above can be written as covariant derivatives, so formally these
conditions can be written as
Di f = i ij Dj .

(3.8)

With the presence of the other U (1) gauge field Bi corresponding to the difference of gauge
fields on the two branes, the correct ansatz is


X1 = x 1 + A2 12 + B2 3 ,


X2 = x 2 A1 3 B1 12 .
(3.9)
To see that this is indeed the correct ansatz, we write the above formulas in the component
form


X1 = diag x 1 + (A2 + B2 ), x 1 + (A1 B1 ) ,



X2 = diag x 2 (A1 + B1 ), x 2 (A1 B1 ) .
(3.10)
Indeed we see that Ai + Bi is the gauge field living on the brane, and Ai Bi is the gauge
field living on the anti-brane. Again it is the second entries on the anti-brane that break the
usual matrix rotational symmetry. To restore the coordinates rotational symmetry, we need
to form a gauge transformation with a gauge parameter of the form g = diag(0, 2 f ) and
f satisfies




i f, x 1 + (A2 B2 ) = x 2 (A1 B1 ) ,


i f, x 2 (A1 B1 ) = x 1 + (A2 B2 ).
(3.11)
Or more compactly,
Di (A B)f = i ij Dj (A B),

(3.12)

where Di (A B) is the covariant derivative defined with respect to the gauge field A B.
The commutator [X1 , X2 ] assumes a little more complicated form than for the brane
brane system



 1 2
X , X = i F12 + 1 3 + F12 12 ,
(3.13)

206

M. Li / Nuclear Physics B 602 (2001) 201212

where it is not surprising to see the term 1 since the B field is 1 . The two U (1) field
strengths are defined by
F12 = 1 A2 2 A1 i[A1 , A2 ] i[B1 , B2 ],
12 = D1 (A)B2 D2 (A)B1 ,
F

(3.14)

where the covariant derivatives Di (A) are defined against the gauge field Ai . It is easy
12 , since everything is charged
to understand why these covariant derivatives appear in F
adjointly with respect to the center of mass U (1) gauge field. However, it is rather unusual
that the gauge field Bi contributes to F12 , the field strength of the center of mass degree
of freedom. The cause of this is the asymmetric fashion in which we are dealing with the
brane and the anti-brane.
In a noncommutative field theory it is often convenient to work with the creation and
annihilation coordinates


1  1
1  1
a=
(3.15)
a+ =
x + ix 2 ,
x ix 2
2
2
satisfying [a, a +] = 1. These are the analogue of complex coordinates z = x 1 + ix 2 , z =
x 1 ix 2 . The gauge fields we introduced can also be written in the complex form
Az = A1 iA2 ,

Az = A1 + iA2 .

It is straightforward to see that Az always appears together with


as z , and Az always appears with a which can be written as z .

(3.16)
a+

which can be regarded

4. The tachyon
We expect the tachyon to arise as the off-diagonal entries of matrices X1 , X2 . Modes
from other matrices remain massless. In the braneanti-brane system, the tachyon field is
a complex scalar. However, one naively expects that any field arising from Xi carries a
spacetime index tangent to the branes. The resolution of this puzzle is connected to the
fact that the background (2.1) breaks the usual spacetime symmetry in the matrix model,
and the conventional world-volume spacetime symmetry, as explained in the previous two
sections, is a combination of the matrix model spacetime symmetry and gauge symmetry.
We work again with the harmonic representation. The gauge parameter associated with
rotation generator can be written as


0
0
g=
,
(4.1)
0 (2N + 1)
where N = a + a is the number operator, is the infinitesimal rotation angle. As explained
before, the rotation around point z = x 1 + ix 2 = 0 is generated by


Xi = ij Xj + i g, Xi .
(4.2)
It is convenient to work with the complex matrices



1 

= 1 X1 iX2 .
Z = X1 + iX2 ,
Z
2
2

(4.3)

M. Li / Nuclear Physics B 602 (2001) 201212

207

is the Hermitian conjugate of Z, it is enough to work with Z. In the absence of the


Since Z
gauge fields,



a T
Z=
(4.4)
.
T a+
Under the rotational transformation (4.2), Z transforms as



0
T(2N + 1)
.
Z = i Z i
(2N + 1)T
2a +

(4.5)

We read from above that a = i a and a + = i a + , meaning that a transforms as z and


a + as z . The off-diagonal fields transform according to
T = 2i NT ,

(N + 1).
T = 2i T

(4.6)

These formulas say that under the rotation, both T and T transform quite nontrivially, and
that one can not simply classify them by a tensor of a fixed rank.
 are both operator-valued, so it is convenient to expand them in terms of the
T and T
basis |m n|, where N|m = m|m


T=
Tmn |m n|,
(4.7)
T =
Tmn |m n|.
Thus under a rotation, Tmn transforms as z 2m and Tmn as z2(n+1) . Since m, n  0, one can
never get a scalar from T, while T0n are invariant under the rotation.
For a fixed m, one can write

Tmn n| = 0|Tm
(4.8)
n

with an operator Tm , so that



T=
|m 0|Tm .

(4.9)

Now Tm transforms as z 2m . In particular, T0 is a scalar under the rotation. Similarly,



T =
Tm |0 m|,
(4.10)
and Tm transforms as z2(m+1) .
Although T0 is invariant under rotation about z = 0, it is not translationally invariant.
This is not a surprise, as T0 itself is not a fully-fledged two dimensional field. The state |0
can be regarded as localized at z = 0. One can construct a similar state localized at another
point z0 satisfying
1
a|z0 = z0 |z0 .
2
This state can be expressed as a coherent state
|z0 = e

z
0
2

a+

|0 .

(4.11)

(4.12)

In a sense, a + can be regarded as the generator of a translation in the z direction in the


noncommutative plane. This observation agrees well with the gauge parameter (2.6), since

208

M. Li / Nuclear Physics B 602 (2001) 201212

x 1 a + a + . Just like |0 , |z0 is not well-localized in the z direction, as the plane is


noncommutative. Now a tachyon mode
T = |z0 0|T

(4.13)

is invariant under the rotation about the point z0 . The corresponding gauge parameter
similar
to (4.1) can be expressed
in terms new creation and annihilation operators a
a z0 / 2 , a + a + z 0 / 2 .

2:
The action (of strings) or the Hamiltonian (of D2-branes) is proportional to [Z, Z]



1 
2 
T+ 2 + 1 T T + + T+ T 2
Z, Z = 1 T + T + T
2

2
+ 2
T a + aT + a + T+ T+ a
,

(4.14)

where the first two terms on the R.H.S. tell us that T is the tachyon while T is massive. That
T is tachyonic was first noticed in [13]. The massive mode T should not be a propagating
mode in the braneanti-brane system. We now show that indeed this mode can be gauged
away.
Although in the braneanti-brane system, the gauge symmetry is U (1) U (1), the
 can be regarded as
full gauge symmetry in the matrix model is still U (2), and T and T
the corresponding off-diagonal gauge fields. Since the off-diagonal gauge parameter is a
complex field, one can gauge away one of them. One would guess that one can gauge away
any of them, or any linear combination of them, as in a branebrane system. This is not
the case here. We can gauge away only T, the massive mode. To see this, let us apply the
gauge transformation with the gauge parameter


0 f+
g=
(4.15)
f
0
to Z, and we find


T = i f a a + f ,


T = i f + a + af + .

(4.16)

Similar to (4.9) and (4.10), we expand



f=
|m 0|fm ,

fm+ |0 m|,
f+ =

(4.17)

where fm , fm+ are operators. Using

a|m = m |m 1 ,
a + |m = m + 1 |m + 1
we find, in terms of the components


+
Tm |0 = i m + 1fm+1
afm+ |0 ,



0|Tm = i 0| fm a mfm1 .

(4.18)

M. Li / Nuclear Physics B 602 (2001) 201212

209

For an arbitrary Tm , it is possible to satisfy the first equation of (4.18), since one simply
needs the following recursion relation

+
(4.19)
m + 1fm+1
= i Tm + afm+ .
Thus it is possible to gauge away any infinitesimal Tm by letting Tm = Tm . On the other
hand, for an arbitrary Tm , it is generally impossible to solve the second equation in (4.18).
The reason is that in the recursion relation



0|fm a = 0| iTm + mfm1


(4.20)
a is not invertible. For instance, for a nonvanishing c-number valued T0 , there is no
solution to the equation
0|f0 a = iT0 0|.

(4.21)

 = 0. The residual gauge symmetry is U (1) U (1).


Now we choose a gauge in which T
This is the desired gauge symmetry in the braneanti-brane system. The action for T
becomes quite simple in this gauge

2
2
2 
1 
2 
Z, Z = 1 T + T + 1 T T + + 2
T a + aT
.
(4.22)
2

Again, this action is not invariant under the world-volume rotation if we simply regard T
as a scalar. As we showed before, T actually transforms nontrivially under a rotation. The
breaking-down of an explicit rotational symmetry when T is treated as a scalar is due to
the fact that in the matrix model the background forces us to use the star product which is
natural only for the D-brane, not for the anti-D-brane. If an algebraic structure symmetric
for the pair of brane and anti-brane exists, the tachyon should appear as a scalar, and should
be related to the tachyon discussed here by a field redefinition, quite similar to the Seiberg
Witten field redefinition.

5. Tachyon condensation
As discussed in the previous section, the complex tachyon field does not appear as
a scalar in the matrix model solution. Naturally therefore, we do not expect a uniform
tachyon condensation in this context, as also pointed out in [14]. First of all, we want to

= 0, the absolute
show that there does not exist a tachyon configuration making [Z, Z]
minimum, when gauge fields are set to zero.

= 0 are
From (4.22) the conditions for [Z, Z]
T + T = 1,

a + = T 1 aT .

(5.1)

The first equation says that T is a unitary operator, and the second equations says that
a + is similar to a. Taking Hermitian conjugate of this condition we have a = T 1 a + T or
a + = T aT 1 . This together with the second equation in (5.1) implies
 2 
T , a = 0.
(5.2)

210

M. Li / Nuclear Physics B 602 (2001) 201212

Similarly, one can derive [T 2 , a + ] = 0. Thus T 2 commutes with all operators constructed
of a and a + and must be a c-number. Unitarity of T further implies that T 2 must be a
pure phase: T 2 = exp(2i). In general, T = diag( exp(i)). But this form cannot satisfy
a + = T 1 aT . We conclude that there is no absolutely minimal tachyon condensation.
The cause of the absence of the minimal tachyon condensation is rather obvious, that
is in the braneanti-brane solution the gauge field strengths on the brane and the antibrane have opposite sign. The lacking of a uniform minimal tachyon condensation is also
noticed in [14]. Due to the nonscalar nature of T , we thus expect localized minimum at
best, coming from solving the equation of motion derived from the variational principle

2 = 0. The equation of motion is just


tr[Z, Z]


T 1 T + T = NT + T N aT a a + T a +
(5.3)
and its Hermitian conjugate. It is not easy to find a general solution to the above equation.
However, a simple solution making both sides of (5.3) vanish exists, it is simply
T = c|0 0|

(5.4)

with |c|2 = 1. This is a vortex localized at the origin. Unlike in the uniform situation studied
in [3], T |m m| can not be an exact solution, since both aT a and a + T a + bring T out
of the projection operator. According to the discussion in the previous section, this is not
surprising, since these modes are not rotationally invariant at all. Nevertheless, T |m m|
can be regarded as an approximate solution, since both aT a and a + T a + are slightly offdiagonal and quite close to NT and T N , and the R.H.S. of (5.3) almost cancel.
The energy difference between the braneanti-brane configuration and a vortex on this
system is given by



2 = 2 2 tr |0 0| = 2 2 ,
$ tr Z, Z
(5.5)
thus indeed the vortex solution is a lower energy configuration. On the other hand, the
approximate solution T = |m m| carries more energy



2 = 2(m 1) 2 .
$ tr Z, Z
(5.6)
So the system will not tend to create these anisotropic vortices.
More vortex solutions can be simply generated by translating |0 0| to other points,
using the matrix translation together with a gauge transformation discussed in Section 2.
It would be interesting to consider the situation when both gauge fields and the tachyon

= 0. This will help to


are turned on, and to look for nontrivial solutions to [Z, Z]
understand the enigmatic nature of the nothing state [15,16]. Since the tachyon can never
be gauged away, as shown in the previous section, a nothing state involving nonvanishing
tachyon can not be equivalent to a pure gauge field configuration. It is also interesting to
see how the U (1) confinement problem is resolved in the present context [17,18].
In conclusion, we have found that the braneanti-brane system in matrix models has a
rather strange representation. This strangeness is due to the asymmetric treatment of of
the brane and the anti-brane, and the noncommutative field theory is defined in favor of
the brane. It is highly desirable to find a new algebraic structure incorporating different

M. Li / Nuclear Physics B 602 (2001) 201212

211

noncommutative structures on both branes. It may help to follow the line of [19] to use a
similarity transformation to figure out this structure.

Acknowledgements
This work was supported by a grant of NSC and by a Hundred People Project grant of
Academia Sinica.

References
[1] A. Sen, SO(32) spinors of type I other solitons on braneantibrane pair, JHEP 9809 (1998) 023,
hep-th/9808141;
A. Sen, Stable non-BPS bound states of BPS D-branes, JHEP 9808 (1998) 010, hep-th/9805019;
A. Sen, Tachyon condensation on the braneantibrane system, JHEP 9808 (1998) 012, hepth/9805170;
A. Sen, BPS D-branes on non-supersymmetric cycles, JHEP 9812 (1998) 021, hep-th/9812031;
A. Sen, Descent relations among bosonic D-branes, Int. J. Mod. Phys. A 14 (1999) 4061, hepth/9902105.
[2] E. Witten, D-branes and K-theory, JHEP 9812 (1998) 019, hep-th/9810188;
P. Horava, Type IIA D-branes, K-theory, and matrix theory, Adv. Theor. Math. Phys. 2 (1999)
1373, hep-th/9812135.
[3] R. Gopakumar, S. Minwalla, A. Strominger, Noncommutative solitons, JHEP 0005 (2000) 020,
hep-th/0003160.
[4] K. Dasgupta, S. Mukhi, G. Rajesh, Noncommutative tachyons, JHEP 0006 (2000) 022, hepth/0005006.
[5] J.A. Harvey, P. Kraus, F. Larsen, E.J. Martinec, D-branes and strings as non-commutative
solitons, JHEP 0007 (2000) 042, hep-th/0005031.
[6] E. Witten, Noncommutative tachyons and string field theory, hep-th/0006071.
[7] E. Witten, Overview of K-theory applied to strings, hep-th/0007175.
[8] N. Seiberg, E. Witten, String theory and noncommutative geometry, JHEP 9909 (1999) 032,
hep-th/9908142.
[9] T. Banks, W. Fischler, S. Shenker, L. Susskind, M theory as a matrix model: a conjecture, Phys.
Rev. D 55 (1997) 5112, hep-th/9610043.
[10] N. Ishibashi, H. Kawai, Y. Kitazawa, A. Tsuchiya, A large-N reduced model as superstring,
Nucl. Phys. B 498 (1997) 467, hep-th/9612115.
[11] M. Li, Strings from IIB matrices, Nucl. Phys. B 499 (1997) 149, hep-th/9612222.
[12] N. Seiberg, A note on background independence in noncommutative gauge theories, matrix
model and tachyon condensation, hep-th/0008013.
[13] H. Awata, S. Hirano, Y. Hyakutake, Tachyon condensation and graviton production in matrix
theory, hep-th/990215.
[14] P. Kraus, A. Rajaraman, S. Shenker, Tachyon condensation in noncommutative gauge theory,
hep-th/0010016.
[15] R. Gopakumar, S. Minwalla, A. Strominger, Symmetry restoration and tachyon condensation
in open string theory, hep-th/0007226.
[16] A. Sen, Some issues in non-commutative tachyon condensation, hep-th/0009038;
A. Sen, Uniqueness of tachyonic solitons, hep-th/0009090.

212

M. Li / Nuclear Physics B 602 (2001) 201212

[17] P. Yi, Membranes from five-branes and fundamental strings from Dp branes, Nucl. Phys. B 550
(1999) 214, hep-th/9901159;
O. Bergman, K. Hori, P. Yi, Confinement on the brane, Nucl. Phys. B 580 (2000) 289, hepth/0002223.
[18] A. Sen, Supersymmetric world-volume action for non-BPS D-branes, JHEP 9910 (1999) 008,
hep-th/9909062;
A. Sen, Universality of the tachyon potential, JHEP 9912 (1999) 027, hep-th/9911116.
[19] C.S. Chu, P.M. Ho, M. Li, Matrix theory in a constant C field background, Nucl. Phys. B 574
(2000) 275, hep-th/9911153.

Nuclear Physics B 602 (2001) 213237


www.elsevier.nl/locate/npe

Light sgoldstino: precision measurements versus


collider searches
D.S. Gorbunov
Institute for Nuclear Research of the Russian Academy of Sciences, 60th October Anniversary prospect 7a,
Moscow 117312, Russia
Received 25 August 2000; accepted 15 March 2001

Abstract

We study sensitivity of low-energy experiments to the scale of supersymmetry


breaking F in

models with light sgoldstinos superpartners of goldstino. The limits on F may be obtained from
direct and indirect measurements of sgoldstino coupling to photons, leptons, mesons and nucleons.
There are three sources of constraints: (i) astrophysics and cosmology; (ii) precision laboratory
experiments at low energies; (iii) rare decays. We discuss only processes with real sgoldstinos.
For sgoldstino lighter than a few MeV
and superpartner masses of the order of electroweak scale,
astrophysical and reactor bounds on F are significantly stronger than limits which may be
reached at future colliders. In models with heavier sgoldstino (up to 5 GeV), constraints from flavor
conserving decays of mesons are complementary to ones coming from collider experiments. The
most sensitive probes of sgoldstinos are flavor violating processes, provided that flavor is violated in
squark and/or slepton sector. It is shown that sgoldstino contributions into FCNC
and lepton flavor
violation are strong enough to probe the supersymmetry breaking scale up to F 107 GeV, if offdiagonal entries in squark (slepton) mass matrices are close to the current limits in MSSM. 2001
Published by Elsevier Science B.V.
PACS: 12.60.Jv; 13.90.+i; 14.80.-j

1. Introduction
Superpartners of goldstino longitudinal component of gravitino may be fairly
light. In a variety of models with low energy supersymmetry they are lighter than a few
GeV. Such pattern emerges in a number of non-minimal supergravity models [1,2] and
also in gauge mediation models if supersymmetry is broken via non-trivial superpotential
(see, e.g., Ref. [3] and references therein). To understand that superpartners of goldstino
may be light, it suffices to recall that in globally supersymmetric theories with canonical
E-mail address: gorby@ms2.inr.ac.ru (D.S. Gorbunov).
0550-3213/01/$ see front matter 2001 Published by Elsevier Science B.V.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 2 2 - 5

214

D.S. Gorbunov / Nuclear Physics B 602 (2001) 213237

Khler potential and in the absence of anomalous abelian gauge factors, the sum of scalar
squared masses is equal to the sum of fermion squared masses in each separate sector of the
spectrum. Since goldstino is massless, its spinless superpartners (scalar and pseudoscalar
particles, S and P , hereafter, sgoldstinos) are massless too; they are associated with a
non-compact flat direction of the scalar potential. Higher order terms from the Khler
potential contribute to sgoldstino masses. Provided these terms are sufficiently suppressed,
sgoldstinos remain light. Of course, these arguments in no way guarantee that sgoldstinos
are always light, but they do indicate that small sgoldstino masses are rather generic. The
theoretical discussion of sgoldstino masses is contained, e.g., in Ref. [4]; here we merely
assume that sgoldstinos are light and consider their phenomenology.
Sgoldstinos couple to MSSM fields in the same way as goldstino [5]; constraints on their
couplings may be translated into the limits on the supersymmetry breaking parameter F .
There are several papers devoted to astrophysical [6], cosmological [7] and collider [8
10] constraints on models with light sgoldstinos. However, the role of light sgoldstinos in
low-energy laboratory measurements has not been studied in detail. To the best of authors
knowledge, the only paper discussing this issue, Ref. [11], concentrated on sgoldstino
contribution (as well as the contribution from light gravitino) into anomalous magnetic
moment of muon. Here we consider a variety of low energy experiments sensitive to light
sgoldstinos.
In this paper we identify those experiments which are most sensitive to different
sgoldstino vertices for various sgoldstino masses. These experiments provide constraints
on the corresponding coupling constants. These constants are in fact proportional to the
ratios of soft terms (squark and gaugino masses, trilinear coupling constants) and F .
The latter parameter is related to the gravitino mass m3/2 in a simple way, F =

3/(8) m3/2 Mpl ; small F corresponds to light gravitino (m3/2 < MSUSY ). Hence, the
constraints derived in this paper are of importance for models with light gravitino, whereas
sgoldstino effectively decouple from the visible sector in models with heavy gravitino.
In principle, there are both flavor-conserving and flavor-violating
sgoldstino couplings
to fermions. We present our results in the form of bounds on F setting soft flavorconserving terms to be of the order of electroweak scale, as motivated by the supersymmetric solution to the gauge hierarchy problem. Flavor-violating couplings are governed
by soft off-diagonal
entries in squark (slepton) squared mass matrices. When evaluating

bounds on F we set these off-diagonal entries equal to their current limits derived from
the absence of FCNC and lepton flavor violation in MSSM [12]. In this way we estimate
the sensitivity of various experiments to the supersymmetry breaking scale.
We consider only low-energy processes with sgoldstinos on mass-shell. Processes
with sgoldstino exchange deserve separate discussion, though we do not expect that the
results obtained in this paper will be altered significantly. Also, behind the scope of this
 0paper are loop processes with virtual sgoldstinos running in loops (for instance, K 0 K
mixing, e , etc.). These processes were analyzed in Ref. [13] in models with heavy
sgoldstinos. Constraints on F obtained in Ref. [13] are significantly weaker than ones
presented in our paper, so the loop processes are less sensitive to F in models with heavy
sgoldstinos. However, models with light sgoldstinos have not been analyzed in detail yet,

D.S. Gorbunov / Nuclear Physics B 602 (2001) 213237

215

though it was pointed out in Ref. [13] that enhancement effects may appear if sgoldstinos
are light. In view of the results obtained in this paper we also find it conceivable that
light virtual sgoldstinos may give significant contributions into rare processes considered
in Ref. [13].
Let us briefly review the current status of experimental limits on F . If one ignores
sgoldstino, then in modelswith light gravitino the strongest direct current bound on F
is obtained from Tevatron, F > 217 GeV [14]. In models with light sgoldstinos, collider
experiments become more sensitive to the scale of supersymmetry breaking. Namely, LEP
and Tevatron provide constraints at the level of 1 TeV on the supersymmetry breaking
scale in models with mS(P ) of order of 20 GeV [9,10]. The most stringent cosmological
constraint comes from Big
Bang Nucleosynthesis [7]: models with light gravitino, m3/2 <
1 eV, that corresponds to F < 7 104 GeV, are disfavored if sgoldstinos decouple at
temperature not less than O(100) MeV (mS(P )  1 MeV). It has been argued in Ref. [6]
that among the astrophysical constraints, the strongest one comes from SN1987A: the
gravitino mass is excluded in the range 0.03 eV < m3/2 < 30 eV for 1 keV< mS(P ) <
These
10 MeV and in a wider range 3 106 eV <
m3/2 < 50 eV for mS(P ) < 1 keV.
excluded intervals correspond to 104 GeV < F < 4 105 GeV and 120 GeV < F <
5 105 GeV, respectively.
In this paper we consider various constraints on couplings of light (mS(P )  5 GeV)
(pseudo)scalars to SM fields coming mostly from astrophysics and direct precision
measurements. So, we partially fill the gap between constraints coming from collider
experiments and cosmology.
As there are flavor-conserving and flavor-violating interactions of sgoldstino fields, we
have to consider both flavor-symmetric and flavor asymmetric processes. Let us outline our
results referring to these two cases in turn.
We begin with constraints independent of assumptions concerning breaking of flavor
symmetry.
As expected, strongest bounds arise from astrophysics and cosmology, that is

6
F  10 GeV, or m3/2 > 600 eV, for models with mS(P ) < 10 keV and MSSM soft
flavor-conserving terms being of the order of electroweak scale. For the intermediate
sgoldstino masses (up to a
few MeV) constraints from the study of SN explosion and
reactor experiments lead to F  300 TeV. We will find that for heavier sgoldstinos, low
energy processes (such as rare decays of mesons) provide limits comparable to ones from
colliders but valid for different sgoldstino masses.
As concerns flavor-asymmetric processes, we find that these are generally very sensitive
to light sgoldstino. Namely, with flavor-changing off-diagonal entries in squark (slepton)
squared mass matrix close to the current bounds,
direct measurements of decays of mesons
(leptons) provide very strong bounds, up to F  900 (15 000) TeV (valid at mS  5
(0.34) GeV), which is much higher than
bounds expected from future colliders. If offdiagonal entries are small, the limits on F become weaker: they scale as square root of
the corresponding off-diagonal elements.
We will see that the rates of processes with one sgoldstino in final state are proportional
to F 2 , whereas the rates of processes with two sgoldstinos in final state are proportional to
F 4 . Hence, under similar assumptions about soft terms governing sgoldstino couplings,

216

D.S. Gorbunov / Nuclear Physics B 602 (2001) 213237

processes with one sgoldstino are more sensitive to the supersymmetry breaking scale.
Nevertheless, the coupling constants entering one-sgoldstino and two-sgoldstino processes
are generally determined by different parameters, so the study of two-sgoldstino processes
is also important.
Further progress in the search for sgoldstino is expected in several directions. Among
the laboratory experiments, the most sensitive to flavor-conserving sgoldstino coupling
for sgoldstino lighter than a few MeV are experiments with laser photons propagating
in magnetic fields and reactor experiments. For heavier sgoldstinos, measurements of
partial widths exhibit the best discovery potential. If flavor violation in MSSM is
sufficiently strong (say, at the level of current limits), the most promising is the study
of kaon decays.
This paper is organized as follows. In Section 2 the effective Lagrangian for sgoldstinos
is presented and sgoldstino decay modes are described. In
Section 3 we derive various
constraints on the parameter of supersymmetry breaking F by considering low energy
processes. There we study separately processes with one and two sgoldstinos in final states
(Subsections 3.1 and 3.2, respectively). First, we discuss astrophysical and cosmological
limits on sgoldstino interactions (Subsection 3.1.1). Then we present laboratory bounds
coming from search for light (pseudo)scalars in electromagnetic and strong processes
(Subsection 3.1.2). In Sections 3.1.3 and 3.1.4. we discuss rare decays with one sgoldstino
in final state due to flavor-conserving and flavor-violating sgoldstino couplings to SM
fermions, respectively. Subsections 3.2.1 and 3.2.2 are devoted to rare meson decays with
two sgoldstinos in final state. Our conclusions and comparison of the results with ones
coming from collider experiments are presented in Section 4.

2. Effective Lagrangian
Let us introduce the effective Lagrangian for light goldstino supermultiplet: scalar S,
 The free part reads
pseudoscalar P and goldstino G.
L=

 1
 i 
1

 G.
S S m2S S 2 + P P m2P P 2 + G
2
2
2

There exist two types of interactions in the low-energy effective theory involving sgoldstino
fields: these are terms that couple one sgoldstino [5,911] and two sgoldstinos [9],
respectively, to SM gauge fields (photons, gluons) and matter fields (leptons fL , up- and
down-quarks fU and fD ). Terms involving one sgoldstino are

1  2 


G
 + im2P P G
 5G
mS S G
Leff =
2 2F
1 M
1
P  F F
+
4 2 F
2 2

1 M
SF F

2 2 F
M3
SG G
F

2
2
m
LR
m
LR
1 M3
Dij
Dij
P  G G
S fDi fDj i
P fDi 5 fDj
+
4 2 F
2F
2F

D.S. Gorbunov / Nuclear Physics B 602 (2001) 213237


2
2
2
m
LR
m
LR
m
LR
Uij
Uij
Lij
S fUi fUj i
P fUi 5 fUj
S fLi fLj

2F
2F
2F
2
m
LR
Lij
i
P fLi 5 fLj .
2F
The direct coupling of two sgoldstinos is described by

Leff =

1
(S P P S)
4F2
 LL 2


 LL 2
2

m
Lij + m
RR
Lij m
RR
Lij fLi 5 fLj + m
Lij fLi fLj


 LL 2

2

2
2

+ m
Dij + m
RR
LL
RR
Dij fDi 5 fDj + m
Dij m
Dij fDi fDj



 LL 2
 LL 2
2

+ m
Uij + m
RR
Uij m
RR
Uij fUi 5 fUj + m
Uij fUi fUj .

217

(1)

(2)

Here M = M1 cos2 W +M2 sin2 W and Mi are gaugino masses; for down-quarks i = d,
2 LL 2 and m
2 are LR-, LL-, and RR-soft
RR
s, b, whereas for up-quarks i = u, c, t; m
LR
ij , m
ij
ij
mass terms in squark squared mass matrix and for convenience we take them real. In what
follows we do not discuss neutrino, so the corresponding couplings are omitted. Note that
2=m A ,
in MSSM the flavor-conserving one-sgoldstino coupling constants satisfy m
LR
fi fi
ii
where mfi are fermion masses and Afi are corresponding soft trilinear coupling constants.
2 LL 2 and m
2 are subject to constraints from the absence
RR
Off-diagonal soft terms m
LR
ij , m
ij
ij
of FCNC and lepton flavor violation (see, e.g., Ref. [12]).
The first part of the effective Lagrangian, Eq. (1), is suppressed by F 1 , whereas
the second one, Eq. (2), is proportional to F 2 , so processes with two sgoldstinos are
very rare. The most stringent bounds on F come from processes with one sgoldstino in
final state. Nevertheless, as we will see, the absence of processes with two sgoldstinos
gives rise to constraints on supersymmetry breaking parameter F comparable to bounds
from high-energy experiments. The latter constraints are, strictly speaking, independent
of the constraints coming from one-sgoldstino processes: one-sgoldstino and direct two LL 2 , m
RR 2 , respectively.
sgoldstino processes are governed by m
LR 2 and m
Let us discuss decay modes of light sgoldstino. First, sgoldstino decay into two photons
is always open [9],

 m3S(P ) M2
.
S(P ) =
(3)
32F 2
Second, in models where 2m3/2 < mS(P ) sgoldstinos may decay into gravitino pairs;
however, the corresponding rates are suppressed by squared ratio of sgoldstino mass
mS(P ) and M in comparison with the decay into two photons, hence this mode may be
disregarded. Third, relatively heavy sgoldstinos (mS(P )  QCD ) decay into gluons (light
mesons) with larger width than into photons because of color enhancement and because
the corresponding coupling is proportional to gluino mass which is usually the largest
among the gaugino masses, i.e., M3 > M . When analyzing hadronic decay modes of
light sgoldstinos (mS(P ) < a few GeV), corresponding rates into quarks and gluons should
be rewritten in terms of light mesons. This step will be presented below. Fourth, sgoldstino

218

D.S. Gorbunov / Nuclear Physics B 602 (2001) 213237

can decay also into light leptons if this process is allowed kinematically (mS(P ) > 2ml ).
Since the corresponding coupling constants are proportional to fermion masses these rates
are suppressed by a factor m2l /m2P (S) apart from the phase space volume [10],


4m2 3/2
1 2l
,
mS


m3P A2l m2l
4m2l 1/2

.
(P l l ) =
1 2
16F 2 m2P
mP
(S l l ) =

m3S A2l m2l


16F 2 m2S

(4)

Consequently, depending on MSSM mass spectrum, sgoldstino masses and the value of
the supersymmetry breaking parameter F , there are three possible situations in experiments
where light (pseudo)scalar particle appears. This particle may live long enough to escape
froma detector. For instance, in the theory with the superpartner scale of order 100 GeV
and F = 1 TeV this behavior would be exhibited by (pseudo)scalar particle with mass
less than 10 MeV, at which sgoldstino width is saturated by two-photon mode. Another
case is when (pseudo)scalar particle decays within detector into two photons or leptons.
Apart from these cases, there is also a possibility of the decay into two gluons (quarks).
For relatively light sgoldstinos (but with masses exceeding 270 MeV), the dominant
hadronic decay is into two pions, while for heavier sgoldstinos KK and channels become
available. Furthermore, there would be effects emerging due to P 0 (, K 0 ) mixing.
Let us estimate branching ratios of hadronic and photonic decay channels neglecting
threshold factor. In order to estimate sgoldstino coupling to hadrons we make use of chiral
theory of light hadrons. There are two different sources of sgoldstinomeson couplings
in the effective Lagrangian (1): interaction terms with gluons and coupling to quarks.
We evaluate contributions from these two sources into mesonsgoldstino interactions
separately.
First, we have to relate gluonic operators entering Eq. (1) to meson fields. We make use
of the correspondence
(s ) a a

1
G G
|0 = q 2 ,
()J =0
8s
2

(5)

derived in Ref. [15]. Here q 2 is momentum of pion pair created with zero total angular
momentum, J = 0; (s ) is the -function of QCD, is the pion isotopic amplitude,
= 2 + + 0 0 ,
and quarks and mesons are considered massless. At higher energies also KK and pairs
may be created by gluonic operator.
There is one more relation [16],
A|

Nf s a
 a |0 = const  fA m2A A ,
G G
4

(6)

 a is a tensor dual to gluonic one, A is a neutral pseudoscalar meson ( 0 , ) and


where G
const is a normalization factor; fA = f = 130 MeV and  is a parameter responsible for
SU(Nf ) flavor symmetry breaking ( = (mu md )/(mu + md ) for 0 ,   1 for ).

D.S. Gorbunov / Nuclear Physics B 602 (2001) 213237

219

In fact, the Lagrangian (1) describes sgoldstino interactions at the superpartner scale.
Sgoldstino coupling constants at low energies may be obtained by making use of
renormalization group evolution. Thus for the gluonic operator one has
G2 (M3 ) = G2 ()

(s ()) s (M3 )
.
s () (s (M3 ))

Hence, we estimate the matrix element of the gluonic operator between the scalar and
meson pair as


M3
s (M3 )
M3
Ga Ga S|S =
(AA)J =0
q 2 2 A A
S ,
(s (M3 ))
F
2 2F

(7)

and in a similar way we estimate the matrix element of another gluonic operator between
the pseudoscalar and meson

M3
 const 2
M3
a  a
P .
G G
P |P =
fA m2A A
A|
(8)

(M
)
N
F
2 2F
s
3
f
Note, that these matrix elements are highly suppressed by squared sgoldstino or meson
masses.
Since direct sgoldstino coupling to quarks contributes also to meson production, we
remind basic relations of chiral theory

i
0
q ) = f q ,
0|J (0) 0 (
2

+
0|J (0) + (
q ) = if q .

(9)

where

1
+
5 d ,
5 u.
u
5 u d
J = d
2
If we parameterize sgoldstino couplings to the triplet of light quarks q as


S i5 P
P q,
L = q S
0

J =

(10)

S and
P being 3 3 matrices of the corresponding coupling constants (which are
with
read off from Eq. (1)), then the standard procedure (see, e.g., Ref. [17]) gives the following
low-energy effective Lagrangian



P P S
2
S
Lmeson = B0 Tr f
(11)
The constant B0 is related to
to the leading order in mesonic fields included in matrix .
1
2
quark condensate as 0|qq|0

= 2 B0 f and may be evaluated from the masses of kaon


2 /(m + m ). We account only for one-sgoldstino terms since others
and quarks, B0 = MK
d
s
0
are suppressed by sgoldstino masses and additional inverse power of F .
The Lagrangian (11) consists of two parts. The first one,



 LR 2
B0 f 0  LR 2
2
2
2
U11 m
U11 + m
m
LR
LR
LR
Lmeson-1 =
D11 + m
D11 2m
D22
2F
2
6

2
2
 0 LR
+ K 0m
(12)
LR
D21 + K m
D12 P ,

220

D.S. Gorbunov / Nuclear Physics B 602 (2001) 213237

 0 mesons, while the second one,


is pseudoscalar sgoldstino mixing with 0 , , K 0 and K



 +
B0  LR 2
1 0 0
2
+
0 0
Lmeson-2 =

K
LR

+
K
K
+
K
+

m
U11 + m
D11
2
2F



1
1 2  LR 2
2
2
0
2
2
U11 + m
LR
LR
LR
LR
+ m
D11 + 4m
D22 + m
U11 m
D11
6
3
1 0  0 LR 2
1 0 0 LR 2
1  0 LR 2
K m
D12 K m
D21 K
m
D12
6
2
2

1
2
+ LR 2
2
K 0 m
LR
D12 + K + m
LR
(13)
D21 + K m
D21 S,
6
describes scalar sgoldstino decays into mesons. Note that sgoldstino couplings with two
different mesons is suppressed by off-diagonal term in squark mass matrix. In what follows
we will not consider processes where real sgoldstino decays into such modes.
Now let us estimate matrix elements between sgoldstino and meson (i.e., sgoldstino
meson mixing) as a sum of two quantities, Eqs. (8) and (12), while the amplitude of the
scalar sgoldstino decay into pairs of light mesons is evaluated as a sum of Eqs. (7) and (13).
Let us compare contributions of gluon and quark operators into sgoldstino couplings to
mesons. As an example, for the ratio of the corresponding contributions into coupling of
the scalar to neutral pions and into pionpseudoscalar mixing we obtain
0 0 |S gluon
mS M3
mS
s (M3 )
4
=
,
0
0
|S quark (s (M3 ))
B0 AQ mu + md

0 |P gluon
2 2 M3
=
.
0 |P quark s (M3 ) 3AQ
These ratios are larger than 10 for M3 = AQ . Hence, gluonic operators give rise to stronger
coupling of light sgoldstinos to light mesons, as compared to sgoldstinoquark interactions.
Let us evaluate the rate of the scalar sgoldstino decay into light mesons, assuming that
this decay is allowed kinematically. As an example, for the neutral pion mode we obtain


S 0 0 =

s2 (M3 ) mS m2S M32


2 (s (M3 )) 324 F 2


2
2
(s (M3 )) 9 B0 mu + md AQ
1 4m 0 .
1
s (M3 ) 4 mS
mS
M3
m2S

Taking into account only the largest contribution from the gluon operator and neglecting
the threshold factor we estimate the ratio of rates of sgoldstino decays into photons and
mesons,
(S )
81 2 (s (M3 )) M2
.
=
(S 0 0 ) 8 2 s2 (M3 )
M32
We see that this ratio is smaller than 1 at M = M3 . Since in most models gluino is several
times heavier than photino, for sufficiently heavy sgoldstinos hadronic modes usually
dominate over photonic one.

D.S. Gorbunov / Nuclear Physics B 602 (2001) 213237

221

Let us estimate now the contribution of pionsgoldstino mixing into pseudoscalar


sgoldstino width. Recall that the pion width is almost saturated by the two-photon decay
mode. Then


M32 m4 0 mu md 2


2 f2
1
0
P = 2
s (M3 ) 4(m2P m2 0 )2 F 2
mu + md

 0


,
(14)
where the two-photon width of virtual pion is taken at p2 0 = m2P and may be approximated
as


  m3
0 tot 0 3P
m 0
With account of only leading contributions from gluonic operator we obtain
2

M2 m2 0 m2P
direct (P )
s2 (M3 )
=
0m 0
1 .
(P 0 )
8 3 M32 f2 m2 0

(15)

2 = m A , so at M
As discussed above, m
LR
i Qi
= M3 and light P (mP  m 0 ) we obtain
ii
2
that the ratio (15) is numerically 8 10 . In the opposite case of heavy P (mP  m 0 )
the ratio becomes even larger. Hence mixing with pions gives negligible contribution to
sgoldstino decay into photons (unless M  M3 /30; we do not consider this case). The
only exception is the degenerate case, when sgoldstino and pion masses are close and this
branching becomes of order 1. (In the case of strong degeneracy there is also a correction to
pion life-time which may give rise to a constraint on F ). We do not consider this unrealistic
situation. The interference with -meson gives nothing new. Indeed, Eq. (15) scales as
meson m3meson which is invariant under the variation of meson mass, if the meson width is
(almost) saturated by anomalous decay into two photons. Decay via neutral kaon is also
negligible because of large kaon life-time.
To conclude this section we summarize the situation with sgoldstino branching ratios.
Let us begin with scalar sgoldstino. In Fig. 1 we present scalar sgoldstino branching ratios
into photons, leptons and neutral pions evaluated for four different sets of supersymmetry
breaking soft terms, A, M , M3 . To determine photonic and leptonic widths we make
use of Eqs. (3) and (4), while the width into two neutral pions is calculated according
to Eq. (5) generalized to non-zero pion masses. Estimating hadronic sgoldstino partial
width we account only for + and 0 0 decay modes. Other hadronic modes may be
considered in the same way. Ratios between different hadronic channels are determined by
chiral theory.
Scalar sgoldstino lighter than 270 MeV almost always predominantly decays into two
photons. At sgoldstino mass close to 2me or 2m , rates of the decays into pairs of
corresponding leptons become comparable to the two-photon rate and even exceed the
latter in models with large trilinear soft terms. Far from the lepton mass, the corresponding
lepton branching ratio decreases as m2l /m2S(P ) . At sgoldstino masses exceeding 270 MeV
hadronic modes emerge. Their rates are somewhat higher than the rate of the two-photon
decay except for models with large M , in which the photonic mode always dominates.

222

D.S. Gorbunov / Nuclear Physics B 602 (2001) 213237

Fig. 1. Branching ratios of scalar sgoldstino decays into photons (thick line), e+ e (long dashed
line), + (thin line) and 0 0 (short dashed line) in models with: (a) |A| = M = M3 =
100 GeV; (b) M = M3 = 100 GeV, |A| = 1 TeV; (c) |A| = M3 = 100 GeV, M = 1 TeV;
(d) |A| = M = 100 GeV, M3 = 1 TeV.

As regards pseudoscalar sgoldstino, it does not have the decay mode into two
pseudoscalar mesons to the zero order in GF . Hence at M M3 its hadronic decay
modes are suppressed unless mP is quite large (well above 1 GeV). In what follows we
consider photonic and leptonic decay channels of the pseudoscalar sgoldstino only.

3. Searches for light sgoldstino


In accordance with the discussion of sgoldstino effective Lagrangian presented in the
previous section, there are two types of processes we are interested in. In the processes of
the first type only one sgoldstino emerges while in the processes of the second type a pair
of sgoldstinos appears in the final state. These processes are governed by different coupling
constants and will be considered in turn.
3.1. Processes with one sgoldstino in the final state
3.1.1. Bounds from astrophysics and cosmology
In Subsections 3.1.1 and 3.1.2 we consider mainly pseudoscalar sgoldstino, though
almost all constraints are valid for the scalar sgoldstino as well; a few exceptions will
be pointed out.

D.S. Gorbunov / Nuclear Physics B 602 (2001) 213237

223

Light pseudoscalar particles appear in particle physics models in various contexts;


a well known example is an axion. There are numerous cosmological, astrophysical and
laboratory bounds on interactions of light pseudoscalars which apply to sgoldstino. For
completeness we collect in Subsections 3.1.1 and 3.1.2 the most stringent of these bounds
and translate the bounds
on sgoldstino coupling constants into bounds on supersymmetry
breaking parameters F and m3/2 . Let us write the interactions of sgoldstino with photons
and fermions as follows
M

L P = 2
P E B g P E B,
F
mf A f
P f5 f gf P P f5 f.
Lf P =
(16)
2F

Then the limits on g and gf P imply limits on F .


In what follows we set M = Af = 100 GeV in our quantitative estimates, since
superpartner scale MSUSY is expected to be close to the electroweak scale as motivated
by supersymmetric solution to the hierarchy problem in SM.
One of the sources of pseudoscalars are stars: light pseudoscalars are produced there by
Primakoff process, that is P conversion in external electromagnetic field. Another
place of sgoldstino creation is galactic space where magnetic fields produce pseudoscalars
from propagating photons.
In helioscope method, a dipole magnet directed towards the Sun is used. Inside
the volume with strong magnetic field, solar pseudoscalars can transform into X-rays
by inverse Primakoff process. An alternative method, Bragg diffraction, was applied
in SOLAX experiment to detect solar pseudoscalars. The absence of anomalous X-ray
fluxes from SN1987A related to possible pseudoscalar conversion into photons in galactic
magnetic field gives the strongest constraint on g . Since this limit is valid only for
unrealistically light pseudoscalar (mP < 109 eV), we consider the helium-burning lifetime of Horizontal Branch Stars (HBS) in globular clusters as the most sensitive probe of F
at very small mP .
There are two more constraints on g coming from cosmology and astrophysics. Light
sgoldstinos are thermally produced in the early Universe via Compton process e
eP (S). Photons from sgoldstino decays contribute to the photon extragalactic background,
if sgoldstinos outlive matterradiation decoupling. If, on the other hand, sgoldstinos
decay before matterradiation decoupling, produced photons may heat electrons leading
to distortion of CMBR spectrum, which is experimentally studied well enough to
exclude wide range of F at corresponding sgoldstino masses. The experiments on photon
background and cosmic microwave background radiation, being combined, exclude a strip
in (mP , g ) plane (see Ref. [22]). In Table 1 we present the corresponding limits for two
typical values of mP .

All these constraints 1 on g are collected in Table 1. The limits on F are obtained
at M = 100 GeV and scale as square root of M . For completeness, we included in
Table 1 also the limits obtained in Ref. [6] by considering SN1987A.
1 See also Ref. [22] for constraints on g coming from deuterium fission by scalars decaying into two photons.

224

D.S. Gorbunov / Nuclear Physics B 602 (2001) 213237

Table 1
Constraints from astrophysics and cosmology on SUSY models with light sgoldstinos coupled to
photons
g , GeV1

F , GeV

 0.03 eV

< 6 1010

> 0.5 106

> 60 eV

SOLAX [19]

< 1 keV

< 2.7 109

> 2.3 105

> 12 eV

SN [20]

< 109 eV

< 1011

> 4 106

> 3.5 keV

HBS [21]

< 10 keV

< 6 1011

> 1.5 106

> 550 eV

Photon
background [22]

1 keV

< 1014
or

> 1.2 108


or

> 3.5 MeV


or

> 105

< 4 103

< 3.5 meV

< 1020
or

> 1.2 1011


or

> 3.5 TeV


or

> 109

< 4 105

< 35 eV

< 5.6 1010

> 5 105

> 50 eV
or

mP

Experiment
Helioscope [18]

Distortion of
CMBR

1 MeV

spectrum [22]
SN1987A [6]

< 1 keV

or
SN1987A [6]

1 keV < mP < 10 MeV

or

m3/2

> 102

< 120

< 3 106 eV

< 9 1010
or

> 4 105
or

> 30 eV
or

> 8 107

< 1.3 104

< 30 meV

Let us proceed with sgoldstino coupling to electrons. Restrictive limits come from delay
of helium ignition in low-mass red giants. There are also two limits on coupling to electrons
from bremsstrahlung process e + (A, Z) (A, Z) + e + P and Compton process
+ e e + P in stars: these processes lead to energy loss of stars and are constrained
by helium-burning life-time of Horizontal Branch Stars. Note that the life-time of HBS
gives stronger constraints on electron coupling to scalar than to pseudoscalar.
Let us turn to (pseudo)scalar coupling to nucleons. In order to relate the corresponding
constant gN to F we make use of the analogy to axion. Then effective Lagrangian reads


5 g (0) + g (3) 3 P ,
Leff = i
(17)
N
N
where denotes the nucleon dublet and
A Q mN
(0)
gN
,

2F

(3)
gN

mu md (0)
g .
mu + md N

(18)

The energy loss of Horizontal Branch Stars via Compton process +4 He 4 He + S gives
rise to a bound on F . Also, nucleonsgoldstino coupling leads to shortening of SN1987A
neutrino burst.
Astrophysical
constraints on sgoldstinofermion interactions arepresented in Table 2.
Bounds on F are obtained at Ae = AQ = 100 GeV and scale as Af , f = Q, e. Note

D.S. Gorbunov / Nuclear Physics B 602 (2001) 213237

225

Table 2
Constraints from Astrophysics on SUSY models with light sgoldstinos coupled to fermions
Experiment

mP

gf

F , GeV

m3/2

< 10 keV

ge < 2.5 1013

> 3.8 105

> 35 eV

HBS [23,24]

 10 keV

geP

< 0.5 1012

> 2.7 105

> 17 eV

HBS [25]

 10 keV

geS < 1.3 1014

> 1.6 106

> 650 eV

 1 keV

(0)
gN < 4.3 1011
(0)
gN < 3 1010

 1.2 106

> 370 eV

 5 105

> 50 eV

or

or

Red Giants [23]

HBS [25]
SN1987A [21]

 10 MeV

(0)
gN < 3 107

 1.5 104

or
< 0.05 eV

Table 3
Constraints from direct measurements on SUSY models with light sgoldstinos
Experiment

mP

g , GeV1

F , GeV

Laser [26]

< 103 eV

< 3.6 107

> 2.0 104

> 93

-regeneration [26]

< 103 eV

< 6.7 107

> 1.5 104

> 50

NOMAD [27]

 40 eV

< 1.5 104

> 970

> 0.22

m3/2 , MeV

that the region F  1.5 104 GeV allowed by SN explosion [21] is not ruled out by
astrophysical arguments or direct measurements if sgoldstino is relatively heavy (10 keV 
mS(P )  10 MeV) and its interactions conserve flavor (see below).
For constraints coming from Big Bang Nucleosynthesis see Ref. [7].
3.1.2. Laboratory bounds on very light sgoldstinos
Let us now consider direct laboratory limits on couplings of very light sgoldstinos.
The first set of bounds on F comes from the study of laser beam propagation through
transverse magnetic field. The production of real sgoldstinos would induce the rotation
of the beam polarization, while the emission and absorption of virtual sgoldstinos would
contribute to the ellipticity of the laser beam. Such effects have not been observed and their
absence implies a constraint on pseudoscalarphoton coupling. There is also a constraint
on the interaction of a pseudoscalar particle with photons coming from experiments on
photon regeneration. In these experiments, light pseudoscalars produced via Primakoff
effect penetrate through optic shield and then transform back into photons (invisible light
shining through walls). Similar scheme is applied
in NOMAD experiment. The results are

presented in Table 3 at M = 100 GeV; limits on F scale as M whereas bounds on
m3/2 scale as M .
Another set of constraints is obtained from reactor experiments, where nuclear deexcitation is studied. Let us again make use of Eqs. (17) and (18). Then we obtain for the

226

D.S. Gorbunov / Nuclear Physics B 602 (2001) 213237

Table 4
Constraints from reactor experiments on SUSY models with light sgoldstinos
(0)

1/2

1/4
F Br(P X) , GeV

1/2

mP

Final state, X

gN Br(P X)

< 1.5 MeV

e+ e

 7 1010 [28]

 3 105

 25 eV

 8 107 [29]

 9 103

 20 MeV

< 1 MeV

m3/2 Br(P X)

isoscalar transition from excited nucleon state with the change of spin by J and isospin by
T and with emission of photons and pseudoscalars with momenta k and kP , respectively,
the following ratio of rates [30]
PJ =1,T =0
M1,J =0


6

kP
k

3

2
g (0)2 gNN
f2

4MN2

where M1 refers to the type of electromagnetic transition and effective pionnucleon


2
coupling constant is gNN
/4 = 14.6. Products of pseudoscalar decay (two photons
+

or e e ) are observed in detectors. In this way two constraints on the coupling of


a pseudoscalar to nucleon have been obtained: P / Br(P e+ e ) < 1016 [28]
and P / Br(P ) < 1.5 1010 [29] (we set the
pseudoscalar momentum equal
to photon frequency, kP = k ). Corresponding bounds on F are presented in Table 4 at

AQ = 100 GeV and scale as AQ . The first constraint is valid for mP < 1.5 MeV and
becomes weaker for heavier sgoldstinos, while the second limit is relevant only for light
sgoldstino,mP < 1 MeV. The larger the branching ratio the stronger the corresponding
bounds on F : these bounds scale as quartic root of branching ratios. Although sgoldstino
branching into e+ e is usually very small (see Fig. 1), current experimental bounds on F
from sgoldstino decaying into e+ e are stronger than limits from decay into two photons.
Note that reactor experiments give fairly strong bounds on F but they should be considered
as order-of magnitude estimates, as obtaining exact numbers requires accurate calculations
involving nuclear matrix elements.
3.1.3. Flavor conserving rare decays
Numerous bounds arise from precise measurements of partial widths of mesons and
leptons (see Tables 58), if corresponding processes are allowed kinematically.
We begin with constraints independent of flavor violating terms in squark (slepton)
mass matrix. One obtains a set of limits on supersymmetry breaking scale by considering

Wilczek mechanism [31] decay of neutral vector meson VQQ


 (1 state) into photon and
(pseudo)scalar S(P ). There are two types of contributions into this process (see Fig. 2).
The first one is emission of real photons and (pseudo)scalars by quarks, while the second
is decay of virtual photons, emitted by quarks, into photons and (pseudo)scalars. The first
process is governed by fermionsgoldstino coupling, while the second one emerges due to
interaction with a pair of photons. The relevant candidates on the role of VQQ
 are J /,
and -, -, -mesons.

D.S. Gorbunov / Nuclear Physics B 602 (2001) 213237

227

Fig. 2. Diagrams contributing to vector meson decay into sgoldstino and photon.
Table 5
Constraints from decays of vector mesons on SUSY models with light sgoldstinos decaying inside
detector
Experimental limit

Br(J / S (S X)) < 5.5 105 [32]

mS

1/4
F Br(SX)

< MJ /

Br( (1S) S (S X)) < 3.1 104 [33]

< 0.1 GeV > 170 GeV

Br( (1S) S (S X)) < 3.1 104 [33]

e+ e

< 1.5 GeV > 170 GeV

Br( (1S) S (S X)) < 4 104 [33]

> 180 GeV

+ , K + K < 1.5 GeV > 160 GeV

Br( (1S) S (S X)) < 4 104 [33]

< 1.5 GeV > 160 GeV

Br( (1S) S (S X)) = (6.3 1.8) 105 [34]

0 0

> 1.0 GeV > 440 GeV

Br( (1S) S (S X)) = (2 2) 105 [35]

2K + 2K

< MJ /

> 330 GeV

Let us first consider heavy mesons, which may be described as quasistationary systems.
With account of effective Lagrangian (1) we obtain
(VQQ
 S(P ) )
+
(VQQ
 e e )

MV2 (AQ M )2
,
16F 2

(19)

where (+) refers to decay into S(P ). We should compare the rate (VQQ
 S(P ) )
with current data on the rates (VQQ
 + missing energy), (VQQ
 3 ) or
(VQQ
 + pair(s) of leptons (light mesons)) depending on mS(P ) and superpartner
mass spectrum (see discussion of sgoldstino decay modes in Section 2). For illustration
we set M = AQ = 100 GeV in our quantitative estimates, so vector mesons would
decay
only into scalar sgoldstino. Eq. (19) shows that the corresponding constraints on

F scale as a square root of the absolute value of the difference (sum) of AQ and M , if
one considers decay into scalar (pseudoscalar).
It turns out that constraints on F from decay into photons (leptons or light mesons),
summarized in Table 5, are of the same order as limits from processes with single photon
and missing energy (see Table 6) if corresponding branching ratios for sgoldstino decay
are roughly of order one. The first type of constraints (Table 5) is relevant for sgoldstino
decaying within detector (which is the case for mS(P )  10 MeV if M = A = 100 GeV);
these constraints scale as quartic root of the corresponding sgoldstino branching ratios. The
second type of bounds (Table 6) applies to lighter
sgoldstino flying away from detector. We
present in Table 5 only strongest constraints on F . Besides these, there is a number of

228

D.S. Gorbunov / Nuclear Physics B 602 (2001) 213237

Table 6
Constraints from decays of vector mesons on SUSY models with light sgoldstinos flying away from
detector
mS

F , GeV

Br(J / S ) < 1.4 105 [36]

 MJ /

> 260

Br( (1S) S ) < 1.3 105 [37]

< 5 GeV

> 370

Experimental limit

other decay modes providing somewhat weaker constraints on F : + K + K ,


2 K + K .
2 + 2 , 3 + 3 , 2 +
One can show that limits on F from decays of light vector mesons (, , ) are weaker
at least by an order of magnitude.
Decays of seem to have the best sensitivity to flavor-conserving sgoldstino couplings
if M  mS(P )  a few MeV. Since in the most part of the parameter space sgoldstino
decays predominantly into two photons or two mesons, the most promising decays are
into three photons and into a photon and a pair of mesons. In models with large trilinear
soft terms, leptonic widths of sgoldstinos become larger, and these modes become also
interesting.
3.1.4. Flavor violating rare decays
There is another type of processes to be considered. These are decays of charged
particles: leptons or pseudoscalar mesons. The rates of these processes are more model
dependent because they
are governed by flavor violating soft terms.
While the bounds on F coming from decays of leptons are the same irrespectively of
whether scalar or pseudoscalar sgoldstino is created in the final state, in the flavor-violating
hadronic processes the creation of scalar sgoldstino is more important than the creation of
pseudoscalar sgoldstino (if they have similar masses). When we discuss hadronic processes
in what follows, we consider the emission of scalar sgoldstino only. The simplest example
is kaon decay K + + S. In chiral theory kaon conversion into pion is described by
matrix element



 +
s d K + = f+ (kK + k ) + f (kK k ) ,
where f+ = 1 and f = 0 in the case of exact SU(3) flavor invariance. Then


m2S
m2 m2
+ sd K + = f+ K
+ f
md + ms
md + ms
and in what follows we neglect f contribution.
In principle, there are two mechanisms of the decay of charged particles with sgoldstinos
in the final state. The first one is due to flavor-conserving sgoldstino interactions (with
fermions and intermediate W-boson). The second one is due to flavor-changing terms in the
low-energy effective interactions of light sgoldstinos (see Eqs. (1), (2); for instance, decay
m
LR
D

K + + S is due to Leff =

2
12

2F

S s d). Hence, the second contribution emerges because

D.S. Gorbunov / Nuclear Physics B 602 (2001) 213237

229

Table 7
Constraints on SUSY models with light sgoldstinos flying away from detector; bounds come
from flavor changing decays of charged particles, if they are allowed kinematically; we set flavor
d =m
l =m
LR2
2Q , 12
LR2
2L to be equal to their current bounds [12] at equal
violating terms 12
D12 /m
L12 /m
Q =500 GeV and equal masses of slepton and photino, m
l =
masses of squark and gluino, M3 = m
M =100 GeV
mS(P )

Experimental limit
Br( eS(P )) < 2.6 106 [38]

< m

Br(K + + S) < 3 1010 [39]

0

Br(K + + S) < 5.2 1010 [40]

< 80 MeV

Br(K + + S) < 108 [41]

 180240 MeV

(ij )LR

F , GeV

l = 1.7 106
12

> 3 104

d = 2.7 103
12
d = 2.7 103
12
d = 2.7 103
12

> 3.7 107


> 3.3 107
> 1.6 107

of flavor violating interactions with fermions originating from off-diagonal insertions in


squark (slepton) mass matrix.

As regards the first mechanism, it gives rise to constraints on F at the level of 100
250 GeV. We do not present these constraints explicitly, as they are at the same level or
weaker than those summarized in Tables 5, 6.
The second mechanism is more interesting. The corresponding constraints are presented
in Tables 7, 8, where for definiteness we take flavor violating off-diagonal insertions
in squark (slepton) mass matrix to be equal to their current experimental limits [12] at

m
squark = M3 = 500 GeV, m
slepton = 100 GeV. The limits on F scale as inverted quartic
2
root of bounds on meson branchings and as square root of the off-diagonal elements mLR
ij
in squark squared mass matrix; they depend crucially on the strength of flavor violation in

MSSM. Since hadronic and photonic modes usually dominate, limits on F coming from
meson decays with a pair of leptons in the final state (say, K + + S(S e+ e )) are
weaker, but not more than by one or two orders of magnitude, as compared to photonic and
mesonic modes. Note, that similar constraints from three-body decays of neutral mesons
(like B 0 K 0 S(S + )) depend on the same coupling constants and are generally
weaker than limits from rare decays of charged mesons.
From bounds presented in this section we conclude that sgoldstino interactions may give
large contributions into flavor changing rare decays, including those forbidden in SM. In
particular, in the case F = 1 TeV2 , the constraints from processes with final light sgoldstino
significantly strengthen the bounds on off-diagonal elements in squark and slepton mass
matrices in comparison with models where sgoldstinos decouple at low energies.
Our analysis suggests that contributions of intermediate (virtual) sgoldstinos into FCNC
and lepton flavor violating processes may be also significant. For instance, pseudoscalar
mesons may decay through light sgoldstino exchange. Also, there are potentially important
 0 , B 0 B
 0 mixings, etc. These issues will be
contributions to loop processes like K 0 K
considered elsewhere.

230

Table 8
Constraints on SUSY models with light sgoldstinos decaying within detector, from search for flavor changing decays of charged
particles; flavor violating terms (ij )LR are the same as in Table 7

1/4
F BrSX >

l = 1.7 106
12

3.4 105 GeV

> 2me

l = 1.7 106
12

1.2 105 GeV

< 100 MeV

d = 2.7 103
12

1.0 107 GeV

Br(K + + S(S X)) < 1.1 108 [45]

e+ e

 150340 MeV

d = 2.7 103
12

1.5 107 GeV

Br(K + + S(S X)) = (7.6 2.1) 108 [46]

> 2m

d = 2.7 103
12

1.3 107 GeV

Br(D + + S(S X)) < 5.2 105 [47]

e+ e

> 2me

u = 3.1 102
12

5.2 105 GeV

Br(D + + S(S X)) < 1.5 105 [47]

> 2m

u = 3.1 102
12

7.0 105 GeV

Br(D + + S(S X)) = (2.2 0.4) 103 [47]

> 2m

u = 3.1 102
12

3.1 105 GeV

Br(Ds+ K + S(S X)) < 1.4 104 [47]

> 2m

u = 3.1 102
12

3.4 105 GeV

Br(Ds+ K + S(S X)) < 6 104 [48]

K +K

> 2mK

u = 3.1 102
12

2.4 105 GeV

Br(B + K + S(S X)) < 6 105 [49]

e+ e

> 2me

d = 1.6 102
23

4.8 105 GeV

Br(B + K + S(S X)) < 5.2 106 [50]

> 2m

d = 1.6 102
23

9.0 105 GeV

Br(B + K + S(S X)) < 2.8 105 [51]

> 2m

d = 1.6 102
23

5.9 105 GeV

Br(B + + S(S X)) < 7.5 105 [51]

K +K

> 2mK

d = 3.3 102
13

6.6 105 GeV

Br(B + + S(S X)) < 4.1 105 [51]

> 2m

d = 3.3 102
13

7.7 105 GeV

Br( eS(P )(S(P ) X) < 7.211 [42]

Br( eS(P )(S(P ) X) < 1012 [43]

e+ e

Br(K + + S(S X)) < 5 108 [44]

mS(P )

D.S. Gorbunov / Nuclear Physics B 602 (2001) 213237

(ij )LR

Experimental limit

D.S. Gorbunov / Nuclear Physics B 602 (2001) 213237

(a)

231

(b)

Fig. 3. (a) The diagram illustrating 0 decay into two sgoldstinos due to two-sgoldstino interaction;
(b) diagram of 0 decay into two photons and sgoldstino due to one-sgoldstino interaction.

3.2. Processes with two sgoldstinos


Processes with two final sgoldstinos appear due to the presence of two-sgoldstino
interactions in low-energy effective Lagrangian, Eq. (2), and due to the double contribution
of one-sgoldstino interaction, Eq. (1). Of course, the corresponding amplitudes are highly
suppressed (by additional F 1
). Nevertheless, some of these processes are sensitive
enough to place constraints on F at the level of 1 TeV.
Recall that two-sgoldstino coupling constants (2) differ from one-sgoldstino constants (1). Indeed, they are proportional to LL and RR insertions in scalar squared mass
matrix, while one-sgoldstino coupling constants are proportional to LR insertions. Note
2 and mRR 2
in this regard, that the current limits on flavor changing squark masses mLL
ij
ij
2 . Hence, it makes sense to consider processes where twoare weaker than limits on mLR
ij
sgoldstino couplings could be observed.
Complete analysis of low-energy processes with two sgoldstinos may be carried out
along the same lines as for processes with one sgoldstino. Instead of going through the
limits systematically,
we discuss here only some examples in order to get the feeling of

sensitivity to F .
3.2.1. Light neutral mesons
We begin with pion decay into two light sgoldstinos (see Fig. 3(a)). The relevant part of
the effective Lagrangian reads


 5

 LL 2
1
LL 2
RR 2
RR 2 5
u

d
(S
P

S)

+
m

u
+
m

+
m

d
.
L=

U11
U11
D11
D11
4F 2
Then by making use of Eqs. (9), (10) we obtain
 LL 2

2
2
2
2 2  2
U11 + m
RR
LL
RR
 f2 m
 0
mS m2P
U11 m
D11 m
D11
SP =
m
128F 2
F2


m2 m2 2
m2
1+ P 2 S

4 P2 .
(20)
m
m
This rate is proportional to (m2S m2P ), so, as expected, it vanishes in the massless limit,
mS , mP 0. In order to examine the sensitivity of this process, let us neglect the phase
volume dependence and take |m2S m2P | m2 /4. If we set the value in the square bracket

232

D.S. Gorbunov / Nuclear Physics B 602 (2001) 213237

Table 9
Constraints on SUSY models with light sgoldstinos from neutral pion decay due to two-sgoldstino
Q = 500 GeV
coupling to matter fields; these constraints are evaluated at |m2S m2P | = m2 /4 and m
(see text)
Experimental limit
Br( 0 SP ) < 8.3 107 [52]
Br( 0 SP (S 2 , P 2 )) < 2 108 [53]

F , GeV
> 150
1/8
1/8
> 240 BrS BrP

equal to 2m
2Q and choose m
Q = 500 GeV, we obtain the limits presented in Table 9. A few

remarks are in order. First, these bounds may be irrelevant in some theories because F
should not be significantly smaller than any of the soft terms. Second, the constraint from
pion disappearance (i.e., from Br( 0 SP )) is valid only if sgoldstinos fly away from
detector. For mS(P )  m /2 this is the case if M , Ae < 10 GeV, which is not forbidden
by current experiments. Third, these limits are obtained at tuned sgoldstino masses and, in
general, they are weaker (see Eq. (20)).
These results do not depend on flavor-violating couplings and are of the same order
of magnitude as the limits presented in Tables 5, 6. However, the limits presented in
Table 9 scale as inverted octopic root of the corresponding pion partial width (see
Eq. (20)).
To illustrate that two-sgoldstino processes may impose more stringent constraints than
one-sgoldstino processes with the same content of final SM particles, let us estimate
the one-sgoldstino contribution to pion decay into four photons. Namely let us consider
emission of sgoldstinos from the photon legs of pion (see Fig. 3(b)). If sgoldstino decays
within detector into photons, this would correspond to four-photon decay of pion. Pion
photon anomalous amplitude reads

 1 2 q3 q2 ,
A( ) =
f
where q1 , q2 are the photon momenta. Then the corresponding squared matrix element of
0 S(P ) is
2
2

2 M f 
(q1 p)2 + (q1 q3 )2 ,
2
2

F
where p and q1 , q3 are momenta of sgoldstino and outgoing photons, respectively.
Neglecting sgoldstino mass we estimate the decay width as

|M|2 = 4

2
2


1 2 m3 M m
0 S(P ) =
.
32 f2 4
F2

(21)

One can check that Eq. (21) gives weaker bound on F than the limit presented in the
Q = 500 GeV.
second row of Table 9 if M = 100 GeV and m
Let us now evaluate the bounds from decays of neutral kaons due to two-sgoldstino
flavor violating couplings. The effective Lagrangian reads

D.S. Gorbunov / Nuclear Physics B 602 (2001) 213237

L=

233


 5

 LL 2
1
2
2
2
(S
P

S)

m
LL
RR
D21 m
RR

D21 + m
D21 s d + m
D21 s d
2
4F



 LL 2
 LL 2
2 5
2
RR
D12 m
RR
+ m
D12 + m
D12 d s + m
D12 d s .

One can show, that only the measurements of branching ratios of KL0 impose interesting
constraints
on F , whereas current limits on rare KS0 decays provide weak constraints

on F . We obtain by making use of chiral theory


 LL 2

2
2
2
2 2  2
D21 + m
LL
RR
RR
 0
 fK2 m
mS m2P
D12 + m
D21 + m
D12
KL SP =
mK
512F 2
F2


m2 m2 2
m2

4 2P .
1+ P 2 S
mK
mK
Note that in the limit of CP conservation, LL and RR squark mass matrices are real and
2
2
symmetric, and the sum in the bracket is equal to 2(m
LL
RR
D21 + m
D21 ).
In analogy to the discussion of pion decays, let us neglect the phase volume dependence
and set |m2S m2P |  m2K /4, fK = 160 MeV. If we set the sum in the bracket equal to
2
2
4Re m
LL
LL
D21 and impose on Re m
D21 current constraints from the absence of FCNC [12]
at squark mass m
Q = 500 GeV, we obtain the limits presented in Table 10. Note that
limits from kaon decays into a leptonic pair and a pair of mesons (photons) are usually
more significant than limits from decays into four leptons, because
of small sgoldstino
decay branching ratio into leptons (see Fig. 1). These bounds on F are obtained at tuned
sgoldstino masses, |m2S m2P |  m2K /4, and generally the bounds are somewhat weaker.
We are not aware of limits on decays KL0 4 and KL0 + . If it would be
possible to measure (or limit) their branching ratios at the level of 107 , the sensitivity
Table 10
0 due to two-sgoldstino
Constraints on SUSY models with light sgoldstinos coming from decays of KL
flavor-violating coupling to matter fields; we set real parts of the flavor violating term, (12 )LL =
2 2 , equal to its current bound, Re( )
2 [12] at equal masses of squarks and
m
LL
12 LL = 4.6 10
D /m
Q
12

gluino, M3 = m
Q = 500 GeV; these constraints are evaluated at |m2S m2P | = m2 0 /4
K

Experimental limit
F , GeV
1/8

1/8

1/8

1/8

0 SP e+ e )
Br(KL
= (6.9 1.0) 107 [46]

> 1.9 103 BrS(P )e+ e BrP (S)

0 SP e+ e e+ e )
Br(KL

> 2.7 103 BrS(P )e+ e BrP (S)e+ e

0 SP + e+ e )
Br(KL

> 3 103 BrS(P )+ BrP (S)e+ e

= (4.1 0.8) 108 [46]

9 [54]
= (2.9+6.7
2.4 ) 10
0 SP e+ e + )
Br(KL
= (3.5 0.6) 107 [46]

1/8

1/8

1/8

1/8

> 2.1 103 BrS(P ) + BrP (S)e+ e

234

D.S. Gorbunov / Nuclear Physics B 602 (2001) 213237

Table 11
Constraints on SUSY models with light sgoldstinos from decays of heavy neutral mesons due to
two-sgoldstino coupling to matter fields
|m2S m2P |

(ij )LL

F >, GeV

Br(D 0 SP 2 + 2 )
= (7.3 0.5) 103 [46]

 m2 0 /4
D

u = 1.0 101
12

360 BrS + BrP +

Br(D 0 SP K + K + )
< 8 104 [55]

 m2 0 /4
D

u = 1.0 101
12

340 BrS + BrP +

Br(D 0 SP 3 + 3 )

 m2 0 /4

u = 1.0 101
12

390 BrS(P )2 + 2

Experimental limit

1/8

1/8

1/8

1/8

1/8

= (4 3) 104 [56]
Br(B 0 SP 2 + 2 )
< 2.3 104 [57]

1/8

BrP (S) +
 m2 0 /4
B

d = 9.8 102
13

1/8

1/8

710 BrS + BrP +

of experiments to two-sgoldstino couplings would increase, because the photonic decay


usually dominates over leptonic decay of sgoldstinos.
3.2.2. Decays of heavy mesons
In analogy with light mesons we consider now heavy neutral mesons living sufficiently
long, D 0 , B 0 and Bs0 . We make use of the approach similar to the chiral theory in order
to describe their interaction with sgoldstinos; in the following we set fBs0 = fB 0 = fD 0 =
200 MeV. The limits obtained with the same assumptions as above about sgoldstino masses
and values of flavor violating couplings are listed in Table 11. All remarks concerning these
assumptions given in Subsection 3.2.1 may be repeated here.
It turns out that constraints on two-sgoldstino coupling from current bounds on Bs decay
modes are rather weak.
To conclude this subsection, we note that two-sgoldstino bounds scale as inverted
octopic root of meson branching ratios, so eight order improvements (!) in
measurements
of corresponding partial widths are required to strengthen the bounds on F by an order
of magnitude.

4. Discussions and conclusions


We have considered models with light (mS(P ) < a few GeV) superpartners of goldstino.
The constraints on their effective couplings to SM particles have been presented.
By making use of these constraints we have estimated the sensitivity of low energy
experiments to the scale of supersymmetry breaking and gravitino mass.
Let us compare our results with constraints coming from high energy processes [810].
If we ignore sgoldstino, then in models with light gravitino
current direct bound on
supersymmetry breaking scale is obtained from Tevatron, F > 217 GeV [14]. The

D.S. Gorbunov / Nuclear Physics B 602 (2001) 213237

235

upgraded Tevatron may be able to


cover the range of F up to F  290 GeV [58],
while LHC will be capable to reach F  1.6 TeV [59]. In models with light sgoldstinos,
collider experiments become more sensitive to the scale of supersymmetry breaking. Most
powerful among the operating machines, LEP and Tevatron, give a constraint of order
1 TeV on supersymmetry breaking scale in models with light sgoldstinos.
Indeed, it was

found in Ref. [9] that with the LEP luminosity of 100 pb1 , at F = 11.5 TeV
one
+

e e SZ( ) event would occur, and ten e e e e S events would appear at F =


1.5 TeV. At Tevatron, about 10
events in pp S (Z) channel, and 105 events in pp S
channel would be produced at F = 1 TeV [10]. This gives
rise to a possibility to detect
sgoldstino, if it decays inside detector into photons and F is not larger than 1.52 TeV.
However, these numbers have been obtained in a model with heavier superpartner scale,
and, hence, with larger
sgoldstino couplings, than we assumed in our paper. For that set of
parameters bounds on F derived in our paper from processes originating due to flavorconserved sgoldstino couplings should be stronger at least by a factor of 1.5.
One important remark concerns the sensitivity of collider experiments to light particles.
The currently available analyses carried out by experimental collaborations are relevant
only for fairly heavy sgoldstino (M  20 GeV). In this paper we have discussed lighter
particles; in this sense our results may be considered as complementary to those derived
up to now from LEP and Tevatron experiments.
From constraints presented in this paper we conclude that the sgoldstino signal is not
likely to be observed at LEP and Tevatron if sgoldstinos are lighter than a few GeV
and flavor-violating processes in MSSM are not extremely suppressed. One observes
that precision measurements at low energies are promising for confirming directly such
a model. The astrophysical bounds are usually stronger than laboratory ones, though they
become invalid for mS and mP larger than a few MeV.
Among the laboratory processes, the most sensitive to very light sgoldstinos are
propagation of laser beam in magnetic field and reactor experiments. For heavier
sgoldstinos the most sensitive processes are decays for flavor-conserving sgoldstino
couplings and kaon decays for flavor-violating sgoldstino couplings.

Acknowledgements
The author is indebted to F. Bezrukov, P. Onyisi, A. Ovchinnikov, A. Penin, A. Pivovarov
and V. Rubakov for useful discussions. This work was supported in part by RFBR grant
99-01-18410, CRDF grant 6603, Swiss Science Foundation, grant 7SUPJ062239, Russian
Academy of Science, JRP grant # 37 and by ISSEP fellowship.

References
[1] J. Ellis, K. Enqvist, D. Nanopoulos, Phys. Lett. B 147 (1984) 99;
J. Ellis, K. Enqvist, D. Nanopoulos, Phys. Lett. B 151 (1985) 357.
[2] T. Bhattacharya, P. Roy, Phys. Rev. D 38 (1988) 2284.

236

D.S. Gorbunov / Nuclear Physics B 602 (2001) 213237

[3] G. Giudice, R. Rattazzi, Phys. Rep. 322 (1999) 419;


S. Dubovsky, D. Gorbunov, S. Troitsky, Phys. Usp. 42 (1999) 705.
[4] A. Brignole, F. Feruglio, F. Zwirner, Phys. Lett. B 438 (1998) 89.
[5] T. Bhattacharya, P. Roy, Phys. Lett. B 206 (1988) 655.
[6] J. Grifols, R. Mohapatra, A. Riotto, Phys. Lett. B 400 (1997) 124.
[7] T. Gherghetta, Phys. Lett. B 423 (1998) 311.
[8] D. Dicus, S. Nandi, J. Woodside, Phys. Rev. D 41 (1990) 2347;
D. Dicus, S. Nandi, Phys. Rev. D 56 (1997) 4166.
[9] E. Perazzi, G. Ridolfi, F. Zwirner, Nucl. Phys. B 574 (2000) 3.
[10] E. Perazzi, G. Ridolfi, F. Zwirner, Signatures of massive sgoldstinos at hadron colliders, hepph/0005076.
[11] A. Brignole, E. Perazzi, F. Zwirner, JHEP 9909 (1999) 002.
[12] F. Gabbiani et al., Nucl. Phys. B 477 (1996) 321;
M. Ciuchini et al., JHEP 9810 (1998) 008.
[13] A. Brignole, A. Rossi, Flavour non-conservation in goldstino interactions, hep-ph/0006036.
[14] CDF Collaboration, T. Affolder et al., Phys. Rev. Lett. 85 (2000) 1378.
[15] M. Voloshin, V. Zakharov, Phys. Rev. Lett. 45 (1980) 688.
[16] D. Gross, S. Treiman, F. Wilczek, Phys. Rev. D 19 (1979) 2188;
V. Novikov et al., Nucl. Phys. B 165 (1980) 55.
[17] A. Pich, Effective field theory, Lectures at the 1997 Les Houches Summer School, Probing the
Standard Model of Particle Interactions, hep-ph/9806303.
[18] S. Moriyama et al., Phys. Lett. B 434 (1998) 147.
[19] SOLAX Collaboration, F. Avignone et al., Phys. Rev. Lett. 81 (1998) 5068;
R.J. Creswick et al., Phys. Lett. B 427 (1998) 235.
[20] J. Brockway, E. Carlson, G. Raffelt, Phys. Lett. B 383 (1996) 439.
[21] G. Raffelt, Stars as Laboratories for Fundamental Physics, University of Chicago Press,
Chicago, 1996.
[22] E. Masso, R. Toldra, Phys. Rev. D 52 (1997) 7967.
[23] G. Raffelt, A. Weiss, Phys. Rev. D 51 (1995) 1495.
[24] D. Dicus et al., Phys. Rev. D 18 (1978) 1829.
[25] J. Grifols, E. Masso, Phys. Lett. B 173 (1986) 237;
J. Grifols, E. Masso, S. Peris, Mod. Phys. Lett. A 4 (1989) 311.
[26] R. Cameron et al., Phys. Rev. D 47 (1993) 3707.
[27] NOMAD Collaboration, P. Astier et al., Phys. Lett. B 479 (2000) 371.
[28] M. Altmann et al., Z. Phys. C 68 (1995) 221.
[29] H. Koch, O. Schult, Nuovo Cimento A 96 (1986) 182.
[30] T. Donnelly et al., Phys. Rev. D 18 (1978) 1607.
[31] F. Wilczek, Phys. Rev. Lett. 39 (1977) 1304.
[32] R. Partridge et al., Phys. Rev. Lett. 44 (1980) 712.
[33] ARGUS Collaboration, H. Albrecht et al., Phys. Lett. B 179 (1986) 403.
[34] CLEO Collaboration, A. Anastassov et al., Phys. Rev. Lett. 82 (1999) 286.
[35] R. Fulton et al., Phys. Rev. D 41 (1990) 1401.
[36] C. Edwards et al., Phys. Rev. Lett. 48 (1982) 19903.
[37] CLEO Collaboration, R. Balest et al., Phys. Rev. D 51 (1995) 2053.
[38] A. Jodidio et al., Phys. Rev. D 34 (1986) 1967;
A. Jodidio et al., Phys. Rev. D 37 (1988) 237, Erratum.
[39] E787 Collaboration, S. Adler et al., Phys. Rev. Lett. 79 (1997) 2204.
[40] S. Adler et al., Phys. Rev. Lett. 76 (1996) 1421.
[41] M. Atiya et al., Phys. Rev. Lett. 64 (1990) 21.
[42] R. Bolton et al., Phys. Rev. D 38 (1988) 2077.
[43] SINDRUM Collaboration, U. Bellgardt et al., Nucl. Phys. B 299 (1988) 1.

D.S. Gorbunov / Nuclear Physics B 602 (2001) 213237

[44]
[45]
[46]
[47]
[48]
[49]
[50]
[51]
[52]
[53]
[54]
[55]
[56]
[57]
[58]

237

E787 Collaboration, P. Kitching et al., Phys. Rev. Lett. 79 (1997) 4079.


C. Alliegro et al., Phys. Rev. Lett. 68 (1992) 278.
Particle Data Group, Eur. Phys. J. C 15 (2000) 1.
E791 Collaboration, E.M. Aitala et al., Phys. Lett. B 462 (1999) 401.
E687 Collaboration, P.L. Frabetti et al., Phys. Lett. B 363 (1995) 259.
P. Avery et al., Phys. Lett. B 223 (1989) 470.
CDF Collaboration, T. Affolder et al., Phys. Rev. Lett. 83 (1999) 3378.
CLEO Collaboration, T. Bergfeld et al., Phys. Rev. Lett. 77 (1996) 4503.
M. Atiya et al., Phys. Rev. Lett. 66 (1991) 2189.
J. McDonough et al., Phys. Rev. D 38 (1988) 2121.
P. Gu et al., Phys. Rev. Lett. 76 (1996) 4312.
E687 Collaboration, P.L. Frabetti et al., Phys. Lett. B 354 (1995) 486.
ACCMOR Collaboration, S. Barlag et al., Z. Phys. C 55 (1992) 383.
DELPHI Collaboration, W. Adam et al., Z. Phys. C 72 (1996) 207.
S. Ambrosanio et al., Report of the beyond the MSSM subgroup for the tevatron run II
SUSY/higgs workshop, hep-ph/0006162.
[59] A. Brignole et al., Nucl. Phys. B 526 (1998) 136.

Nuclear Physics B 602 (2001) 238260


www.elsevier.nl/locate/npe

Perturbative and instanton corrections to the OPE of


CPOs in N = 4 SYM4
Gleb Arutyunov a,1 , Sergey Frolov b,1 , Anastasios C. Petkou c
a Max-Planck-Institut fr Gravitationsphysik, Albert-Einstein-Institut,

Am Mhlenberg 1, D-14476 Golm, Germany


b Department of Physics, The Ohio State University, Columbus, OH 43210-1106, USA
c Dipartimento di Fisica dellUniversit di Milano, Via Celoria 16, 20133 Milano, Italy

Received 6 November 2000; accepted 9 March 2001

Abstract
We study perturbative and instanton corrections to the Operator Product Expansion of the
lowest weight Chiral Primary Operators of N = 4 SYM4 . We confirm the recently observed nonrenormalization of various operators (notably of the double-trace operator with dimension 4 in
the 20 irrep of SU(4)), that appear to be unprotected by unitarity restrictions. We demonstrate the
splitting of the free-field theory stress tensor and R-symmetry current in supermultiplets acquiring
different anomalous dimensions in perturbation theory and argue that certain double-trace operators
also undergo a perturbative splitting into operators dual to string and two-particle gravity states,
respectively. The instanton contributions affect only those double-trace operators that acquire finite
anomalous dimensions at strong coupling. For the leading operators of this kind, we show that the
ratio of their anomalous dimensions at strong coupling to the anomalous dimensions due to instantons
is the same number. 2001 Published by Elsevier Science B.V.
PACS: 11.25.Hf; 11.30.Pb

1. Introduction
N = 4 supersymmetric YangMills theory (SYM4 ) provides a concrete example of
a supersymmetric quantum field theory where the idea of the AdS/CFT duality [13]
can be successfully explored. According to the duality conjecture, SYM4 with a gauge
2 N is dual to type IIB
group SU(N) at large N and at strong t Hooft coupling = gYM
5
supergravity on the AdS5 S background. Unifying the results obtained in the context of
E-mail addresses: agleb@aei-potsdam.mpg.de (G. Arutyunov), frolov@pacific.mps.ohio-state.edu
(S. Frolov), anastasios.petkou@mi.infn.it (A.C. Petkou).
1 On leave of absence from Steklov Mathematical Institute, Gubkin str. 8, 117966, Moscow, Russia.
0550-3213/01/$ see front matter 2001 Published by Elsevier Science B.V.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 1 8 - 3

G. Arutyunov et al. / Nuclear Physics B 602 (2001) 238260

239

the usual weak coupling expansion with the predictions of AdS/CFT duality allows us to
conceive basic dynamical features of the theory.
In the superconformal phase the non-trivial dynamics of the SYM4 is encoded
into correlation functions of gauge-invariant composite operators, which may acquire
perturbative as well as non-perturbative (instanton) corrections. An important class of
local operators in SYM4 is given by the Chiral Primary Operators (CPOs) of the form
OkI = tr( (i1 ik ) ), where i are the YangMills scalars. Under supersymmetry these
operators generate short multiplets of the superconformal algebra SU(2, 2|4) that are dual
to multiplets of type IIB supergravity compactified on AdS5 S 5 . Unlike 2- and 3-point
correlation functions of CPOs that are subject to the known non-renormalization theorems
[412], 4-point functions receive in general perturbative and instanton corrections. As such,
they contain important dynamical information for the supersymmetry multiplets which
appear in the Operator Product Expansion (OPE) of two CPOs.
Recently, the 4-point function of the lowest weight CPOs O2I has been computed
in the supergravity approximation [13,14], and has been used in [15] to analyze their
OPE at strong coupling. 2 The structure of the OPE obtained in [15] for the first few
low-dimensional operators was found to be in complete agreement with the predictions
of AdS/CFT correspondence. Recall that the transformation properties of local gaugeinvariant operators of SYM4 with respect to the superconformal algebra allow one to
classify them into three categories:
(i) Single-trace chiral operators which belong to short representations and have
conformal dimensions protected from quantum corrections.
(ii) Operators which are obtained as normal-ordered products of the chiral operators.
They may belong either to short or long representations, the former have protected
conformal dimensions, while the dimensions of the latter are restricted from above.
(iii) Operators which belong to long representations and whose conformal dimensions
grow without bound in the strong coupling limit.
According to the AdS/CFT duality, the operators in (i) are dual to the type IIB
supergravity fields while operators in (ii) are dual to multi-particle supergravity states. For
the operators in (iii) the duality predicts the growth of their conformal dimensions as 1/4
when . The latter operators are interpreted as being dual to string states (single- or
multi-particle), which decouple in the strong coupling limit.
Comparison of the OPE of the two lowest weight CPOs in free-field theory and at strong
coupling [15] has enabled us to make the following predictions for the structure of the OPE
at finite and N :
(i) The R-symmetry current and the stress tensor of the free-field theory, which involve
only the six SYM scalars i , undergo splitting into 2 and 3 operators, respectively,
belonging to different supermultiplets. Only one operator in each splitting is dual to
a supergravity field and has protected conformal dimension, while all others decouple at
strong coupling as their anomalous dimensions grow without bound.
(ii) The only double-trace operator with free-field conformal dimension 4 that acquires an
2 Various aspects of 4-point functions involving operators descendent to O I were discussed in [1630].
2

240

G. Arutyunov et al. / Nuclear Physics B 602 (2001) 238260

anomalous dimension at strong coupling is O1 = :O I O I : + transforming in the trivial


representation of the R-symmetry group. We argue that the free-field theory operator O1fr
also undergoes splitting into a sum of an operator dual to a gravity state and operators
dual to string modes. The same kind of splitting also occurs for the scalar operator in
the 84 irrep. The double-trace scalar operators in the 20 and 105 irreps do not split. The
operator in the 20 irrep saturates the unitarity bound A) in the classification of [31] and is
not protected from acquiring anomalous dimension. However, our analysis shows that this
operator retains its canonical dimension and hence its non-renormalization is a genuine
dynamical effect.
(iii) The double-trace operator with free-field dimension 5 in the 15 irrep acquires
anomalous dimension, while the one in the 175 has protected dimension. They both split
at finite and N .
(iv) There are several towers of traceless symmetric tensor operators in the 105, 84 and
175 irreps, whose anomalous dimensions vanish.
Here we confirm the above predictions by analyzing the 4-point function of the CPOs
O2I computed at 2-loops in perturbation theory [3235] (three-loop results were obtained
in [36,37]). We also study the instanton contribution to the 4-point function. In [38] the
correlation functions of the four N = 2 singlet scalar fields and of the sixteen dilatinos
were computed in SYM4 with gauge group SU(2) in the sector with instanton number
k = 1. In [39] these results were further generalized to the group SU(N) and in [40,
41] to arbitrary k in the large N limit. With the above results at hand, we then use the
recently obtained non-renormalization theorem of [42] to restore the complete 4-point
function of the CPOs O2I and perform its OPE analysis. Our results are in agreement
with the earlier considerations of [43] and show the absence of instanton contributions
to the anomalous dimensions of single-trace operators in the Konishi multiplet [35].
Pictorially, we observe that the instanton contribution is seen only by those operators
whose anomalous dimensions are non-zero and finite at strong coupling. In particular, the
double-trace operator in 20 does not receive instanton corrections. Such a picture points to
an interesting relation between the mysterious multi-particle supergravity states and the
D-particle modes.
The plan of the paper is as follows. In Section 2 we recall the OPE structure of
two CPOs at weak and strong coupling. In Section 3 we analyze the two-loop 4-point
function of the lowest weight CPOs. We compute the anomalous dimensions of singleand double-trace operators and demonstrate the splitting of the free-field operators into
distinct supermultiplets acquiring different anomalous dimensions. In Section 4 we study
the instanton contribution to the 4-point function of the CPOs and show that instantons do
not contribute to the anomalous dimensions of neither operators dual to string-modes nor
operators with protected dimension. The dynamically protected operator in the 20 does
not receive instanton contributions, which indicates that only operators receiving finiteanomalous dimensions at strong coupling see instantons. In the conclusion we discuss
the results obtained.

G. Arutyunov et al. / Nuclear Physics B 602 (2001) 238260

241

2. OPE algebra of CPOs at weak and strong coupling


In this section we review the structure of the OPE algebra of the lowest weight CPOs at
both the weak and the strong coupling regimes. We follow the notation of [15] and also use
2 N
gYM

=
.
(2)2
(2)2
The normalized lowest weight CPOs in SYM4 are operators of the form

O I (x) =

1
21/2

(1)

CijI tr(: i j :),

where the symmetric traceless tensors CijI , i, j = 1, 2, . . . , 6 form an orthonormal basis of


the 20 of SO(6). As was shown in [15] the leading terms in the OPE of two O I s in free
field theory take the form
O I1 (x1 )O I2 (x2 ) =

2 I1 I2
I1 I2
23/2 C I1 I2 I  I 
O
+
+
[K]
4
2
2
N x12
31/2N x12
x12



23/2 x12 I1 I2  J15  I1 I2 x12 x12
free
T
CJ15 J
2
2

x12
x12
6N
N

1 x12 x12 I1 I2 I  I 
+
T + I1 I2 [ O1 ]
C
2
N x12
I1 I2  J20 
I1 I2  J105 
+ CJ
O
+
C
O
J
20
105


I1 I2
I1 I2  J15 
I1 I2  J175 
+ CJ
O J84 + CJ
+ . (2)
x O + CJ
x O
84
15 12
175 12

free and J 15 are, respectively, the stress tensor and the normalized R-symmetry
Here T

current of the free field theory (including only six scalar fields), K is the normalized
I is a traceless second rank tensor in 20 and O J denote generically
Konishi scalar, T
double-trace operators in the corresponding representation J of the R-symmetry. In
addition to the above fields the OPE contains infinite towers of both single-trace as well as
double-trace operators.
The strong coupling OPE compatible with the 4-point function of [13] is different from
(2) and reads

O I1 (x1 )O I2 (x2 ) =

I1 I2
23/2 C I1 I2 I  I  23/2 x12 I1 I2  J15 
O
+
+
CJ15 R
4
2
2
N x12
x12
3 N x12

(s)


x x 
1

I1 I2 122 12 T + I1 I2 x121 O1

x12
30N

(s)
(s)
(s)



I I 
I I 
I I 
+ CJ1202 x1220 O J20 + CJ11052 x12105 O J105 + CJ1842 x1284 O J84
(s)
(s)
I1 I2 15  J15 
I1 I2 175  J175 
+ CJ
O
+
C
+ .
x
x
x
x12 O

12
12
12
J
15
175

(3)

Here R 15 is the R-symmetry current and T is the stress tensor of the full N = 4 SYM4
and (s)
J is the anomalous dimension of the corresponding double-trace operator at strong

242

G. Arutyunov et al. / Nuclear Physics B 602 (2001) 238260

coupling. The conformal blocks appearing in (3) encode all the strong coupling information
for the anomalous dimensions and the couplings of the corresponding operators. In the
place of an infinite number of single-trace operators in (2), (3) contains instead only three
single-trace operators giving rise to the most singular terms. Note that the coefficients in
front of the R-symmetry current and the stress tensor in (3) are different from the ones in
J
free constructed only from scalars
(2). The reason is that the free-field operators J 15 and T
are split into operators belonging to different supersymmetry multiplets. Multiplets that are
dual to string modes decouple in the strong coupling limit, while operators from the stress
tensor multiplet are non-renormalized and show up at strong coupling.
The leading double-trace operators receive anomalous dimensions whose value at strong
coupling was found to be
16
16
(s)
,
O15 = 2 ,
(4)
N2
5N
while all the other operators shown in (3) have vanishing anomalous dimensions. The
double trace operators in 84, 105 are in short multiplets and they are protected. The doubletrace operator in 20 is not protected by unitarity and is allowed to acquire an anomalous
dimension. Nevertheless, it was found to have vanishing anomalous dimension at strong
coupling.
Comparison of the free-field and strong coupling OPEs (2) and (3) enabled us to make
the predictions for the OPE structure at finite N and discussed in the Introduction.
In the next section we verify that these predictions are in agreement with the 2-loop 4point function of CPOs. To this end we study the asymptotic behavior of the 4-point
2 , x 2 0, which in terms of the biharmonic ratios u =
function in the direct channel x12
34
2 x 2 /(x 2 x 2 ), v = x 2 x 2 /(x 2 x 2 ) and the variable Y = 1 v/u, amounts to taking
x12
34
13 24
12 34
14 23
the short-distance limit u, v, Y 0. Our analysis closely follows [15] and is based on
the knowledge of the conformal partial wave amplitudes of quasi-primary operators. In
particular, consider the contributions to the OPE of two CPOs coming from a scalar, vector
and second rank symmetric traceless tensor. Schematically this is given by


 J
1  J  COOT x12 x12
I1 I2 COOS
S

T
O I1 (x1 )O I2 (x2 ) = CJ
4
6
S
T
CS x
CT x
12
12


COOV x12  J 
V
+

.
+
(5)
CV x 5V
12
(s)

O1 =

I1 I2
Here J denotes an index of an irrep of the R-symmetry group SO(6), CJ
are the
ClebschGordan coefficients and S , T , V are the conformal dimensions of the scalar,
tensor and vector operators, respectively. For any operator in the OPE, CO and COO O
denote the normalization constant in the 2-point function O(x1 )O(x2 ) and the coupling
constant in the three-point function O I (x1 )O J (x2 )O(x3 ), respectively. Then, the shortdistance expansion of the conformal partial amplitudes (CPWA) of the scalar S, tensor T
and vector V operators can be written as (cf. [15])

 I
O 1 (x1 )O I2 (x2 )O I3 (x3 )O I4 (x4 )

G. Arutyunov et al. / Nuclear Physics B 602 (2001) 238260

243


I1 I2 I3 I4  2
CJ
CJ
COOS S
S
2
Y
v
1+
4 x4
C
4
x12
S
34



3S
S + 2
v 1+
Y +
+
16(S 1)(S + 1)
4


2
COOT T 1 1 2 1
T
2
+
Y v
vY
v
CT
4
4
16


V 1
C2
1
Y + ,
+ OOV v 2
CV
2
(1)

(6)
(1)

where we assumed that T = 4 + T and kept only terms linear in T . The formulas
for the leading contributions of a rank-2 traceless symmetric tensor and a vector can be
generalized to the case of a rank-l traceless symmetric tensor of dimension l and one
gets a leading term of the form v (l l)/2 Y l .
If we decompose the conformal dimension of an operator into a canonical part (equal
to its free-field conformal dimension) and an anomalous part, taken to be a small
parameter (see [15]), then (6) shows that the anomalous dimensions are related to terms
(0)
(0)
of the form v S /2 log v for scalar operators, v (V 1)/2 Y log v for vector operators and
(0)
v (T 2)/2 Y 2 log v for rank-2 tensor operators. Formula (6) is the basic tool in our analysis
of the 2-loop 4-point function in Section 3 and the instanton contribution in Section 4.

3. OPE analysis of the 2-loop 4-point function


The 2-loop 4-point functions of the CPOs O2 were computed in [3234] and the results
obtained there can be represented in terms of a basic function (1) (v, u) that can be written
in the form of a MellinBarnes integral as

1
(1) (x, y) =
(7)
ds dt 2 (s) 2 (t) 2 (1 + s + t)x s y t ,
(2i)2
C

where the contour(s) C run parallel to the imaginary axis. Performing the integrations, we
may cast it in a form suitable for studying the OPE as
(1) (v, Y ) =

n,m=0

v n Y m 2 (1 + n) 2 (1 + n + m)
(2 + 2n + m)
(n!)2 m!
[ log v + 2(2 + 2n + m) 2(1 + n + m)].
(8)

Then, the 4-point function of the CPOs O I reads


 I

O 1 (x1 )O I2 (x2 )O I3 (x3 )O I4 (x4 ) = I1 I2 I3 I4 a1 + I1 I3 I2 I4 a2 + I1 I4 I2 I3 a3
+ C I1 I2 I3 I4 b2 + C I1 I3 I2 I4 b1 + C I1 I3 I4 I2 b3 , (9)
where up to 2-loops the various coefficients are given by

244

G. Arutyunov et al. / Nuclear Physics B 602 (2001) 238260



2
1
(1)
a1 = 4 4 1 2 v (v, Y ) ,
N
x12x34



1
4
(1)
b1 = 2 4 4 vu v(vu v u) (v, Y ) ,
2
N x12 x34


1
2
a2 = 4 4 u2 2 vu (1) (v, Y ) ,
N
x12x34



1
4
b2 = 2 4 4 v + v(v + Y ) (1) (v, Y ) ,
2
N x12 x34


1
2
a3 = 4 4 v 2 2 v 2 (1) (v, Y ) ,
N
x12x34


1
4
v Y
v (1) (v, Y ) .
b3 = 2 4 4 u +
N x12 x34
2 1Y
Using the above result we may now study the OPE at 2-loops. We start with the projection
into the singlet which includes important fields such as the stress tensor, the Konishi scalar
and the double-trace operator O1 with canonical dimension 4.
3.1. Projection in the singlet
Using the properly normalized projector in the singlet [15], we obtain for the first few
terms in the short-distance expansion

 I
O 1 (x1 )O I2 (x2 )O I3 (x3 )O I4 (x4 ) 1



I1 I2 I3 I4
3
4

log v 3
= 4 4
v 1+
1+
3N 2
2
x12 x34




2
3
5
3
3
2
2
+
log v
log v
vY 1 +
vY 1 +
+
3N 2
2
2
3N 2
2
3


2
230
2
1 2

log v +
+ v 1+
10
3N 2 N 2
45N 2


2
2
37
1 2

log v +
+ v Y 1+
(10)
.
10
3N 2 N 2
9N 2
The expansion (10) should be matched with the contributions coming from the first few
low-dimensional operators in the singlet projection of the OPE (2). In the free-field theory
limit the first fields which appear in the above OPE are the Konishi scalar K with freefree and a double-trace operator
field dimension 2, the stress tensor of 6 free scalar fields T
free
O1 with free-field dimension 4. It is natural to assume that these are exactly the first
few operators which appear also in the 2-loop OPE, allowing only for possible small
corrections in their free-field dimensions and normalization constants in order to account
for the logarithmic terms in (10). Although such an assumption seems to work for the
Konishi scalar, it does not quite fit the 2-loop result (10) as there is a discrepancy in the
coefficients in front of the stress tensor in free field theory and at 2-loops.

G. Arutyunov et al. / Nuclear Physics B 602 (2001) 238260

245

In order to properly interpret (10) one should realize that the stress tensor T expected
free , since it receives contributions not only from the six
to appear in it is different from T
scalars but also from the four Weyl fermions and the vector field of N = 4 SYM. It has
been argued in [44], following [45], that the general stress tensor of an interacting CFT
involving scalars, fermions and vectors involves three, linearly independent and mutually
orthogonal structures. 3 For the specific case of N = 4 SYM, simple manipulations allow
one to write the free-field stress tensor (i.e., the stress tensors of six massless free scalars)
free (x) as follows
T
1
10
18
free
T
(11)
(x) = T (x) + K (x) + (x),
5
35
35
where the three structures depicted in (11) are mutually orthogonal and linearly independent. The idea of [44] is that the orthogonality and linear independence property is preserved by perturbation theory, i.e., the structures in (11) do not mix under renormalization.
The (symmetric and traceless) tensor K belongs to the Konishi supermultiplet while the
(symmetric and traceless) is the lowest component of a new supermultiplet. The full
stress tensor T is expected to remain conserved, therefore, it retains its canonical dimension at any order in perturbation theory. However, K and can, and do, acquire
anomalous dimensions.
We preface the more detailed analysis of the OPE with some necessary comments.
The fact that T is canonically normalized allows one to find the free-field value of the
normalization constants of the 2-point functions of the three operators in (11) as
CT = 5Ct ,

7
CK1 = Ct ,
2

C =

35
Ct ,
18

(12)

free . The value of the coupling C


where Ct = 32 2 is a normalization constant for T
OOT =

16/(3N) is fixed by the conformal Ward identity. The free-field theory OPE (2) together
with (11) and (12) gives the free-field value of the normalization constants

16
.
3N
Recall that the Konishi field is canonically normalized, i.e., CK = 1 and the free-field result
for COO K is COO K = 2/(31/2N 2 ).
In the sequel we assume that for any operator O in the OPE the ratio COO O /CO is kept
equal to its free-field value. The correction to a coupling dependent normalization constant
is introduced in the following way
COO O ()


= COO O 1 + C (1) ,
COO O ()
(13)
OO O
COO K1 = COO =

where COO O stands for the free-field value.


Now taking into account the splitting (11), using (6) for the contributions of scalars and
symmetric traceless tensors to the OPE and expanding the anomalous dimensions and the
3 This is easily seen in free-field theory, where the stress tensors for free scalars, fermions and vectors provide
three linearly independent and orthogonal to each other structures [45].

246

G. Arutyunov et al. / Nuclear Physics B 602 (2001) 238260

normalization constants, we find for the leading terms of the short-distance expansion of
the singlet projection
OPE
Psinglet
=


I1 I2 I3 I4 
1 + A10 v + A11 vY + A12 vY 2 ,
4
4
x12 x34

where the coefficients A are given by




2
COO
K
(1)
K
1+
A10 =
log v + COO K ,
CK
2


2
K
1 COO
K
(1)
K
log v +
+ COO K ,
A11 =
1+
2 CK
2
2


2
2
K
1 COOT
1 COO K
2
(1)
log v + COO K + K +
A12 =
1+
3 CK
2
3
4 CT


2
1 COO K1
1
(1)
+
1 + K1 log v + COO K1
4 CK1
2


2
1 COO
1
(1)
+
1 + log v + COO .
4 C
2

(14)

(15)
(16)

(17)

Here the parameters K , K1 and correspond to the small corrections to the canonical
(1)
(1)
dimensions of the operators K, K and , respectively, while COO
K , COO K1

(1)
and COO
denote the small corrections to the corresponding free-field normalization
constants. According to the discussion above the free-field values of the ratios of the 3and 2-point normalization constants occurring in (15)(17) are given by
2
COO
4
K
=
,
CK
3N 2

2
COO
K1

CK1

16
,
63N 2

2
COO
16
=
.
C
35N 2

(18)

Requiring consistency of the terms carrying equal powers of v and Y in (10) and (14)
we then obtain the anomalous dimensions and corrections to the coupling constants of the
operators discussed above.
Consistency of the terms proportional to v in (10) and (14) gives

K = 3,

(1)

COO K = 3.

(19)

The value of K coincides with the well-known value for the 2-loop anomalous dimension
for the Konishi scalar [44].
Using the result (19), we immediately see that the terms proportional to vY in (10) and
(14) are consistent.
Consistency of the terms proportional to vY 2 in (10) and (14) gives
1
3
9
= K1 + ,
(20)
2
7
35
1 (1)
9 (1)
3
+ COO
.
= COO
(21)
K
1
2
7
35
This shows that the consistency of the short-distance expansion with the OPE is
not sufficient to determine the individual anomalous dimensions and corrections to

G. Arutyunov et al. / Nuclear Physics B 602 (2001) 238260

247

the couplings of the split fields K and . However, here comes the input of
supersymmetry which rectifies the situation. Namely, requiring that K is in the same
supermultiplet as the Konishi scalar K we fix its anomalous dimension to be the same as K
K1 = 3 .

(22)

Then, we easily find from (20) that


=

25

(23)

(1)
in complete agreement with [44]. In the same way we also have COO K1 = 3 and from
(1)

(21) we obtain C
= 25 .
OO

The terms in (10) proportional to v 2 and v 2 Y encode the information about scalar
operators of free-field dimension 4. Recall that in free-field theory the corresponding terms
match with the contribution of a unique operator
O1free =



1 i j i j
1
: tr tr : : tr i i tr j j :
2
6
40

with the 2-point function







1
1
2
O1free (x1 )O1free (x2 ) =
,
1+
8
10
3N 2 x12

(24)

while at strong coupling the singlet channel was shown [15] to contain a scalar operator
O1 of approximate dimension 4 with the following 2-point function:


38
1
1
1+
.
O1 (x1 )O1 (x2 ) =
(25)
(s)
2
10
15N
8+2
x12 1
Although the difference of 2-point functions of O1free and O1 might be explained by the
fact that they are computed in different regimes and the operator is not protected, it is
more natural to assume that O1free splits in perturbation theory into a sum of operators
such that only one of them is dual to a gravity state. Indeed, in free-field theory one
finds a number of linearly independent operators of dimension 4, e.g., : tr( i j ) tr( i j ):,
: tr( i i ) tr( j j ): and : tr( i j i j ): that may mix under renormalization. To find the
individual anomalous dimensions at two loops one should diagonalize their mixing matrix.
We then expect to find a unique operator O1 (dual to a two-particle gravity state), whose
2 , while the other operators (dual to string modes)
anomalous dimension behaves as /N
should have the anomalous dimensions of the Konishi type . Such a splitting, similar in
spirit with the above discussed splitting of the stress tensor, seems to be necessary in order
to account for the fact that at strong coupling we find only one operator with approximate
dimension 4 while at any order in perturbation theory we expect an operator mixing.
However, the knowledge of the correlation functions of CPOs alone is not sufficient
in order to establish the mixing matrix and additional information is needed, e.g., the
knowledge of correlation functions of four Konishi scalars or other operators.

248

G. Arutyunov et al. / Nuclear Physics B 602 (2001) 238260

3.2. Projection in 20
Projecting the 4-point function in the 20 irrep we obtain for the leading in v, Y terms the
following answer

 I
O 1 (x1 )O I2 (x2 )O I3 (x3 )O I4 (x4 )
=

I3 I4 

20


1 2
1
40
Y
+
Y
v
1
+
4 x4
3N 2
2
2
x12
34




2
20
3
2
v
Y
vY
(1
+
Y
)

+ v2 2 +
4
3N 2
N2 3

10 2

+ 2 v Y v vY log v .
N 3
I1 I2

CJ CJ
20

20

(26)

According to our discussion of the free-field theory OPE in Section 2, the first three lowdimension operators contributing to 20 are the CPOs themselves, the double-trace operator
I K of approximate dimension 4. On the
O20 , and a symmetric second rank tensor K
20
other hand at strong coupling and in the large N limit we found that only the CPOs and the
O20 survive and keep their free-field dimension. While the non-renormalization property
of CPO is well-known, the non-renormalization of O20 is a new phenomenon that cannot
be explained on the basis of unitarity. A natural suggestion made in [15] is that O20 is
non-renormalized in perturbation theory at finite N . As far as K20 is concerned, being dual
to a string mode it receives perturbatively large anomalous dimension and decouples from
the spectrum at strong coupling. 4 Let us see how this picture is compatible with two-loop
result (26).
The last two terms in (26) are proportional to and we interpret them as loop
contribution to the coupling COO K20 and to the anomalous dimension of K20 , respectively.
(1)
Indeed, if we denote the anomalous dimension of K20 as K20 then the log v-dependent
term in (26) occurs due to the contribution of the conformal block of the second rank tensor
with free-field dimension 4 (cf. (6)). Therefore, the other two operators, CPO and O20 ,
do not receive anomalous dimensions. To compute (1)
K20 one needs to know the free-field
2

value of the ratio C


/CK . This can be found by considering, e.g., the -independent
OO K20

20

vY 2 terms in (26). By using the CPWA of the scalar with dimension 2 and comparing vY 2
terms in (6) with the ones in (26) one gets
2
2
COO
COOO
20
K20
+
=
.
3CO
4CK20
3N 2
2
/CO = 40/(3N 2 ) [15] one finds the following free-field value
Since COOO
2
COO
K20

CK20

80
.
9N 2

(27)

4 In principle one could expect a splitting of K into a sum of operators dual to string modes. However, our
20
analysis will show that this does not happen.

G. Arutyunov et al. / Nuclear Physics B 602 (2001) 238260

249

Analogously, analysis of the v 2 terms in (6) and in (26) produces the free-field relation
2
2
2
COO
COOO
COOO
2
K20
20
+

=2+
6CO
CO20
4CK20
3N 2

(28)

that further gives


2
COOO
20

CO20

=2+

2
.
3N 2

Note that the same answer was found by studying the 4-point function at strong coupling
[15], that agrees with the conjectured non-renormalization of the operator O20 .
(1)

Now the log v-dependent term allows to find K20 :

2
COOK

20

8CK20

(1)

K20 =

10
,
3N 2

i.e.,

(1)
K20 = 3.

(29)

Thus, the anomalous dimension of K20 is the same as the dimension of the Konishi field,
hence they are from the same multiplet.

Finally the -dependent


terms without log v are due to the loop correction to the free(1)

field value of COO K20 . Indeed, the vY 2 term in (26) allows one to find COO
K20 = 3.
We can check the consistency of the assumption that there is only one tensor operator in
the 20 which receives corrections to its anomalous dimension and structure constant. To
this end we compute the term of order v 2 Y by using the found anomalous dimension and
correction to the structure constant, and see that it coincides with the corresponding term
in (26).
3.3. Projection in 84
Projecting in 84 we get for the leading terms in the short-distance expansion

 I
O 1 (x1 )O I2 (x2 )O I3 (x3 )O I4 (x4 )
I1 I2

I3 I4 

CJ84 CJ84
4 x4
x12
34

84

6 2
v (1 + Y ) log v + v 2
N2




9
2
12
2 2 (1 + Y ) 2 2 Y .
N
N
N
(30)

A strong coupling result suggests that at finite the OPE of CPOs contains two operators
O84 and K84 transforming in the irrep 84. The operator O84 has protected both the
dimension and the normalization constants of the 2- and 3-point functions, while K84 is
from the Konishi multiplet and receives anomalous dimension.
the non-logarithmic term in (30) gives for the free-field values of
At zeroth order in ,
the normalization constants 5
1
2
CO84 + 2 CK84 = 2 2 .
N
N
5 We exhibit a coefficient 1/N 2 in front of C
fr
K84 to emphasize the fact that a free-field operator O undergoes
1
fr
gr
str
gr
a splitting into a sum of operators as O = O + N O , where O is dual to a supergravity two-particle state
and O str dual to a string state [15]. The same splitting applies to operators in irrep 175.

250

G. Arutyunov et al. / Nuclear Physics B 602 (2001) 238260

The constant CO84 is non-renormalized and is found [15] to be CO84 = 2 6/N 2 .


Therefore, CK84 = 4. The log v term in (30) allows one to read off the anomalous
(1)

dimension (1)
K84 of K84 : K84 = 3, as it should be for the member of the Konishi
2

term in (30) we can find a correction C (1)


= 3.
multiplet. Finally, from the v
OO K84

3.4. Projection in 105


For the leading terms of the projection of the 4-point function in 105 we find
 I

O 1 (x1 )O I2 (x2 )O I3 (x3 )O I4 (x4 ) 105

I1 I2 I3 I4 




CJ
C
2 3
3
4
105 J105
2
=
v 1 + Y log v + v 2 + 2 (1 + Y ) .
4 x4
2
N2
N
x12
34

(31)

The last formula shows that the first log v-term appears at order v 3 . Therefore, all
symmetric traceless rank-2k tensor operators of dimension 4 + 2k transforming in the
105 have protected conformal dimensions, the lowest operator among them is the doubletrace operator O105 . The log v-term in (31) indicates the appearance of the anomalous
dimensions for the symmetric traceless rank-2k tensors of the canonical dimensions 6 + 2k.
However, at strong coupling the first log v-term appears only at order v 4 (see (4.16) of
[15]). Thus, a free-field tensor operator of dimension 6 + 2k undergoes a splitting into two
operators, one has a protected dimension and normalization constants, another one receives
perturbatively an anomalous dimension and disappears at strong coupling.
3.5. Projection in 15
Here we comment briefly on the projection in the irrep 15 whose leading terms have the
form
 I

O 1 (x1 )O I2 (x2 )O I3 (x3 )O I4 (x4 )
I1 I2

I3 I4 

CJ15 CJ15
4 x4
x12
34

15


16
8
8
vY 2 vY + 2 vY log v .
N2
N
N

(32)

The presence on the term vY log v shows the appearance of the anomalous dimension for
J
the vector operator K 15 of dimension 3. At strong coupling, however, the term vY log v is
J
absent and the dimension 3 operator which is the R-symmetry current R 15 has protected
conformal dimension. Thus, at finite the contribution to 15 comes from two operators,
J
J

K 15 and R 15 . From the -independent


term vY in (32) we read off the relation for the
free-field values of ratios of the normalization constants of these operators:
2
COO
K

2CK

2
COOR

2CR

8
.
N2

(33)

2
/(2CR ) = 8/(3N 2 ) we find
Taking into account that COOR

2
COO
K

2CK

16
.
3N 2

(34)

G. Arutyunov et al. / Nuclear Physics B 602 (2001) 238260

251

J15
Finally, from the vY log v term one obtains the anomalous dimension for K
:
(1)

K = 3,

(35)

J15
i.e., the vector operator K
is in the Konishi multiplet.

3.6. Projection in 175


Finally, projecting the 4-point function on irrep 175 one obtains for the leading terms
the following expression:
 I

O 1 (x1 )O I2 (x2 )O I3 (x3 )O I4 (x4 ) 175
I1 I2 I3 I4 

CJ
C
4 2
2 2
175 J175
2
2v
=
Y

v
Y
+
v
Y
log
v
.
(36)
4 x4
N2
N2
x12
34
At strong coupling the first log v term occurs at order v 3 Y (cf. Section 4.6 of [15]), while
here it appears at order v 2 Y . Thus, at finite the contribution of the lowest dimension
operators to the irrep. 175 comes from two operators K175 and O175 , both with approximate
dimension 5. The first operator receives infinite anomalous dimension at strong coupling,
while the second one is non-renormalized due to the shortening condition [31]. The v 2 Y
term in (36) produces for the free-field constants the following relation
1
COO K175 + COOO175 = 4.
N2

(37)

As was found in [15] COOO175 = 4 8/(3N 2 ) and, therefore,


8
COO K175 = .
3

(38)

Then the log v term allows one to find K175 = 3 justifying thereby that K175 belongs to
the Konishi multiplet.
(1)

4. Instanton contribution
To analyze the instanton contribution to the 4-point function of the lowest weight CPOs
we use the results of [3841]. Firstly, we follow [42] to write the 4-point function of the
CPOs (2) as

 ij
1 1 (x1 ) i2 j2 (x2 ) i3 j3 (x3 ) i4 j4 (x4 )
i j

= a1 (s, t)

i j

{i21 j21 } {i43 j43 }


4 x4
x12
34

i j

+ a2 (s, t)

4 x4
x13
24

{i {j i }j }

+ b1 (s, t)

{i13j14}{i42 j23}

2 x2 x2 x2
x13
14 23 24

i j

{i31 j31 } {i42 j42 }

i j

+ a3 (s, t)

4 x4
x14
23

{i {j i }j }

+ b2 (s, t)

{i12j14}{i43 j32}

2 x2 x2 x2
x12
14 23 34

i j

{i41 j41 } {i32 j32 }


{i {j i }j }

+ b3 (s, t)

{i12j13}{i34 j42}

2 x2 x2 x2
x12
13 24 34

, (39)

252

G. Arutyunov et al. / Nuclear Physics B 602 (2001) 238260

where ij =
s=

1
21/2

2 x2
x12
34
2 x2
x13
24

tr[ i j 16 ij 2 ], i, j = 1, 2, . . . , 6 and
t=

2 x2
x14
23
2 x2
x13
24

(40)

The traces are over SU(N) adjoint indices and the SO(6) group-theoretic -factors in (39)

are products of Kroenecker s (cf. [42]).
Superconformal invariance implies that (39) is actually determined in terms of only two
arbitrary functions (e.g., a1 and b2 ) of s and t. This fact allow us to restore the instanton
contribution to the full 4-point function (39) from the results of [3841] as follows. In the
N = 2 formulation of N = 4 the six scalars in the fundamental of SO(6) are decomposed
in one complex scalar and four scalars comprising the N = 2 matter hypermultiplet. The
complex scalar is = 5 + i 6 and one defines






tr 2 = tr 55 tr 66 + 2i tr 56 = Y ij tr i j ,
(41)
where Y ij = i5 j 5 i6 j 6 + i( i5 j 6 + i6 j 5 ). Then, from (39) using the nilpotency of
Y ij we obtain
 2




tr (x1 ) tr 2 (x2 ) tr 2 (x3 ) tr 2 (x4 )
16
16
16
= a2 (s, t) 4 4 + a3 (s, t) 4 4 + b1 (s, t) 2 2 2 2 .
(42)
x13 x24
x14 x23
x13 x14 x23x24
The last correlator is precisely the one computed in [3841] and the result reads (omitting
the anti-instanton contributions)




 2
4 4
x34 D4444(x1 , x2 , x3 , x4 ), (43)
tr (x1 ) tr 2 (x2 ) tr 2 (x3 ) tr 2 (x4 ) = 16Qx12
where the D-functions are defined in [15] and we have absorbed all the normalization
factors into Q defined as

1 230 34
N gYM
1
Q=
(44)
,
33 27/2
k 1/2 e2ik

16
d2
4 4 2
d|k

with the usual complex YangMills coupling. The first factor in (44) is due to the
normalization of the CPOs, the second factor comes from the k-instanton measure in the
large N limit and also takes into account the R-weight of the CPOs and the last one is the
result of the integration of the fermionic zero modes. Q is not a modular invariant function,
it is only the leading term of the modular invariant expression in the large gYM limit.
In order now to read off the functions a1 , a3 and b2 in (42) from (43) we can exploit the
result of [42] according to which the above three functions are in fact expressed in terms
of only one function F (s, t) as
b1 (s, t) = (s t 1)F (s, t),

a2 (s, t) = F (s, t),

a3 (s, t) = tF (s, t).

(45)

The function F (s, t) should satisfy the following crossing-symmetry properties


1
F (s, t) = F (t, s) = F (s/t, 1/t).
t

(46)

G. Arutyunov et al. / Nuclear Physics B 602 (2001) 238260

253

From (42) and (45) we then obtain


F (s, t) = Q

s
v3 
D(s, t) = Q 2 D
4444 (v, Y ),
3
t
u

(47)

where
4444 (s, t)
D(s, t) = D

= 2K dt1 dt4 (t1 t2 t3 t4 )3


1
s
exp t1 (t2 + t3 + t4 ) t2 t3 t2 t4 t3 t4 ,
t
t

(48)

and K was defined in [15]. Using the above integral representation it is easy to check that
the function F (s, t) does satisfy the relations (46).
Multiplying (39) with CijI we find the instanton contribution to the complete 4-point
function of the lowest weight CPOs as
 I

O 1 (x1 )O I2 (x2 )O I3 (x3 )O I4 (x4 ) inst
1 
= 4 4 I1 I2 I3 I4 A1 (v, Y ) + I1 I3 I2 I4 A2 (v, Y ) + I1 I4 I2 I3 A3 (v, Y )
x12 x34

+ C I1 I2 I3 I4 B2 (v, Y ) + C I1 I3 I2 I4 B1 (v, Y ) + C I1 I3 I4 I2 B3 (v, Y ) ,
(49)
where
v3 
D4444 (v, Y ),
u
4444 (v, Y ),
A2 (v, Y ) = u2 a2 (s, t) = Qv 3 D
4
v 
A3 (v, Y ) = v 2 a3 (s, t) = Q D
4444 (v, Y ),
u
 4

v 
u
D4444(v, Y ),
B1 (v, Y ) = uvb1 (s, t) = Q u 1
v
u


u v4 
B2 (v, Y ) = vb2 (s, t) = Q 1 u
D4444 (v, Y ),
v u2
 3

v 
u
B3 (v, Y ) = ub3 (s, t) = Q
u1
D4444 (v, Y ).
v
u
A1 (v, Y ) = a1 (s, t) = Q

(50)

4444 function has the following decomposition


The D

2
Y m v n 2 (n + 4) 2 (n + m + 4)
4444 = 5
D
108
m! (n!)2
(8 + 2n + m)
n,m=0

log v + 2(n + 1) 2(n + 4)

2(n + m + 4) + 2(8 + 2n + m) .

(51)

We are now ready to analyze the contribution of the instantons to OPE of the lowest
weight CPOs. Firstly we consider the short-distance expansion for the projection in the
singlet. We find that the leading terms are given by

254

G. Arutyunov et al. / Nuclear Physics B 602 (2001) 238260


O I1 (x1 )O I2 (x2 )O I3 (x3 )O I4 (x4 ) 1



139
2 Q I1 I2 I3 I4
1
451
2
2
(1
+
Y
)v
+
Y
.
=
log
v

4 x4
84
17640 4410
x12
34

(52)

This clearly shows that the leading contribution to the OPE comes from a scalar operator
of approximate dimension 4. Our experience at strong coupling teaches us that the only
such operator is the double-trace operator O1 discussed in Section 3. On the other hand,
due to the absence of v log v and vY 2 log v-terms, the contribution of the Konishi fields
K and K1 , and of the operator are absent. We conclude that the Konishi multiplet as
well as the multiplet built on receive only perturbative but not instanton corrections. 6
Furthermore, we observe in (52) the absence of the contribution of the stress tensor, in
agreement with the known non-renormalization theorem for this operator.
For the projection on 20 the leading contribution reads as

 I
O 1 (x1 )O I2 (x2 )O I3 (x3 )O I4 (x4 ) 20
I1 I2 I3 I4 

2 QCJ
CJ20

5 2 2
451 2 2
20
v Y log v
v Y v

=
(53)
4 x4
252
10584
x12
34
and comes from a second rank tensor of the canonical dimension 6. The contribution
from the Konishi multiplet is again absent. Recall that at strong coupling we have found
that the first operators receiving anomalous dimension are scalar and tensor operators of
approximate dimension 6 which we therefore identify with the operators appearing in (53).
These two examples at hand, i.e., projection in the singlet and in 20, allow us to make
a general observation: the (double-trace) operators receiving finite anomalous dimensions
at strong coupling also receive instanton contributions. The instanton contribution to the
(single-trace) operators with infinite anomalous dimensions at strong coupling is absent.
Let us examine the other irreps. One gets the following leading behavior
for 84:
I1 I2 I3 I4 

CJ84
2 QCJ

 I
451 3
1 3
I2
I3
I4
84
1

v ,
v log v
O (x1 )O (x2 )O (x3 )O (x4 ) 84 =
4 x4
28
5880
x12
34
for 105:
 I

O 1 (x1 )O I2 (x2 )O I3 (x3 )O I4 (x4 ) 105
I1 I2 I3 I4 

2 QCJ
C
1 4
451 4
105 J105
v ,
v log v
=
4 x4
84
17640
x12
34
for 15:

 I
O 1 (x1 )O I2 (x2 )O I3 (x3 )O I4 (x4 ) 15

I1 I2 I3 I4 
CJ
2 QCJ
451 2
1 2
15
15
v Y ,
=
v Y log v
4 x4
21
4410
x12
34
6 The absence of the instanton corrections to the Konishi multiplet was already noted in [35].

G. Arutyunov et al. / Nuclear Physics B 602 (2001) 238260

255

for 175:
 I

O 1 (x1 )O I2 (x2 )O I3 (x3 )O I4 (x4 ) 175

I1 I2 I3 I4 
C
2 QCJ
451 3
1 3
175 J175
=
v Y .
v Y log v
4 x4
84
17640
x12
34
Comparison of these formulae with the analogous strong coupling results confirms the
above observation. Moreover, we see that the instanton contributions do not spoil the nonrenormalization property of certain towers of double-trace operators found in [15]. Indeed,
the scalar operator O20 in 20, all the rank 2k tensors of dimension 4 + 2k in 84 and of
dimension 4 + 2k, 6 + 2k in 105, and all the rank 2k + 1 tensors of dimension 5 + 2k in
175 are non-renormalized in the instanton background.
The absence in the instanton OPE of corrections to the operators in the Konishi multiplet
can be easily understood at follows. Instanton corrections to the normalization constant
COO K and to the anomalous dimension of K are encoded into the 3- and 2-point functions
of K in the instanton background. According to the prescription of [3841], to calculate
the correlation functions of composite operators in the instanton background we replace
the latter by their instanton background expressions; the resulting correlation function is
then non-zero only if it contains all 16 fermionic zero modes. This is needed in order to
saturate the fermionic integration measure. It is then easy to see that the following 3-point
function




OOK tr (i1 i2 ) tr (i1 i2 ) tr k k ,
(54)
contains only 12 fermionic zero modes and therefore it should be zero in the instanton
background. Such arguments can be generalized to the whole Konishi multiplet. Indeed, as
was shown in [43], the number of the fermion zero modes for an operators O (q) from the
Konishi multiplet is 4 |q|, where q is a U (1)Y charge. The 3-point function of O (q) with
two lowest-weight CPOs provides only 12 |q| zero modes and, therefore, vanishes.
Application of the same arguments to the double-trace operator O20 shows that it can, in
principle, receive instanton corrections. However, our explicit OPE calculation shows that
it is not the case. This means that the particular dynamics that keeps O20 non-renormalized
is not affected by instantons.
Concerning instanton corrections to the other multiplets we found, their existence is
in agreement with the general considerations of [43]. An operator O (q) from a generic
long multiplet provides 8 |q| zero modes, so that the 2- and 3-point correlation
functions involving O (q) are non-zero only if q = 0. Thus, in our OPE the leading
operators that receive instanton corrections and survive at strong coupling with finite
anomalous dimensions, are either primary operators (or conformal descendants of the
k Op .
primary operators), Op generating long multiplets or have the form Qk Q
Coming back to the singlet projection we now compute the instanton contribution (i)
to the anomalous dimension of O1 . If we assume that the small parameter at hand is
N 1/2 /N 4 , then

256

G. Arutyunov et al. / Nuclear Physics B 602 (2001) 238260


2
COOO
1

2CO1

(i)
1 =

2Q
.
84

(55)

Substituting for COOO1 /CO1 its free-field value 1/10 we get the anomalous dimension
(i)

1 =

5 2 Q
.
21

(56)
(i)

Similarly, we can define the instanton contribution 15 to the anomalous dimension of


J
the vector current O 15 as
2
COOO
15

4CO15

(i)

15 =

2Q
,
21

(57)

i.e., one obtains


2Q
.
(58)
21
For the above two operators we observe that the ratio of the strong coupling anomalous
dimensions to the anomalous dimensions due to instantons is the same number
(i)
15 =

(i)

(s)

(i)
15
(s)

15

5 2
QN 2 .
336

(59)

By using the results of [15] we have also checked that the same ratio occurs for the leading
operators in 84, 105 and 175 indicating thereby a universal behavior
(i)
1
(s)

(i)
15
(s)

15

(i)
84
(s)

84

(i)
105
(s)

105

(i)
175
(s)

(60)

175

5. Concluding remarks
In this work we have extended the OPE analysis of the lowest weight CPOs initiated in
[15], to include 2-loop and instanton contributions. At the perturbative level, we found
that it is not sufficient to simply deform the free-field operator algebra by allowing
for anomalous dimensions and corrections to the coupling in order to account for the
2-loop corrections. What is required is a splitting of various free-field operators into
operators belonging to distinct supermultiplets which behave in a different way under the
RG-flow. We have explicitly demonstrated this splitting in the case of the stress tensor
and the R-current of the theory. Our results are consistent with earlier calculations by
Anselmi in [44,46]. Furthermore, we argued that a similar splitting occurs also for freefield theory double-trace operators, e.g., they split into supermultiplets which behave in
perturbation theory either like the Konishi multiplet or acquire anomalous dimensions
2 . The latter multiplets are the ones which survive at strong coupling and get
/N
non-zero anomalous dimensions. The above splitting seems to be necessary in order to
explain the fact that while at any order in perturbation theory we expect a mixing of many
operators with the same free-field dimension, at strong coupling only one of the operators

G. Arutyunov et al. / Nuclear Physics B 602 (2001) 238260

257

is present while all the others decouple. Nevertheless, an explicit calculation of the the 2loop anomalous dimension of the split operators would require the knowledge of 4-point
functions including operators other than the lowest weight CPOs, e.g., the 4-point function
of Konishi scalars. We believe that this is an interesting project.
We also found that the instantons give contributions only to operators which acquire nonzero and finite anomalous dimensions at strong coupling. In particular, instantons do not
contribute neither to protected nor to operators in the Konishi multiplet. This is consistent
with general arguments given in [43] concerning the vanishing of the three-point functions
of operators with non-zero U (1)Y -charge in the instanton background.
The fact that instantons contribute only to operators which correspond to two-particle
modes of classical supergravity, points to an interesting connection between the latter
modes and D-particles modes which is worth further study. Intuitively, the corrections to
the anomalous dimensions of the supergravity modes come from a corresponding change
of their energy in the presence of D-particles. On the other hand, it can be seen that the
universal behavior (60) for the leading operators in the OPE is a consequence of the fact
that the correlation function of CPOs is defined (up to the free-field contribution) by a
unique function F (v, Y ). If one projects in a given irrep J , one gets



O I1 (x1 )O I2 (x2 )O I3 (x3 )O I4 (x4 ) J
=

I1 I2 I3 I4
CJ
CJ
4 x4
x12
34

[hJ (v, Y ) + fJ (v, Y )F (v, Y )],

where hJ (v, Y ) is a free-field contribution, and fJ (v, Y ) is some function depending


on the irrep we consider. Now we see that if the strong coupling and instanton 4-point
functions are described by F (s)(v, Y ) and F (i) (v, Y ), respectively, then after subtracting
the free-field theory contribution, the ratio of the strong to the instanton contributions
will be one and the same for all irreps since the function fJ (v, Y ) does not depend
on the regime we consider and, therefore, cancels out. What is more important is
that the short-distance expansion of the both functions fJ (v, Y )F (i,s) (v, Y ) starts as
(Y ) log v + , where the function Gi,s
fJ (v, Y )F (i,s) (v, Y ) = v (l)/2 G(i,s)
l (Y ) may
l
(i,s)
(i,s)
l
be represented as Gl (Y ) = aJ (Y )f
(Y ) = aY + and it encodes the leading
contribution of the rank-l tensors of a canonical dimension . Therefore, the equalities
(60) stem from the fact that the ratio of the instantons to strong-coupling corrections
(i)
(s)
Gl (Y )/Gl (Y ) does not depend on an irrep we consider.
Finally, we wish to comment on the fact that we have confirmed both in perturbation
theory and in the instanton background the non-renormalization of various operators, most
importantly the scalar double-trace operator in the 20 with dimension 4. The dimension
of the latter operator is not protected by unitarity constraints and therefore it is subject to
a genuine dynamical protection. Being a double-trace operator, it is difficult to find the
corresponding supergravity mode. Furthermore, since it is a marginal operator we may
use it to deform the N = 4 SYM4 action preserving the conformal invariance to leading
order in the deformation. It would be of great interest to find out if this operator is exactly

258

G. Arutyunov et al. / Nuclear Physics B 602 (2001) 238260

marginal as in this case it defines a particular RG flow (fixed line), that might lead to a new
non-trivial N = 0 CFT.

Acknowledgements
We would like to thank A. Tseytlin for useful comments on the manuscript. G.A. and T.P.
are grateful to S. Ferrara and P. Fre for useful discussions. G.A. would like to acknowledge
helpful conversations with S. Theisen. The work of G.A. was supported by the DFG and
by the European Commission RTN programme HPRN-CT-2000-00131 in which G.A. is
associated to U. Bonn, and in part by RFBI grant N99-01-00166. The work of S.F. was
supported by the U.S. Department of Energy under grant No. DE-FG02-91ER-40690 and
in part by RFBI grant N99-01-00190. T.P. was supported by the E.U. under the program
RTN1-1999-00116.

References
[1] J. Maldacena, The large N limit of superconformal field theories and supergravity, Adv. Theor.
Math. Phys. 2 (1998) 231.
[2] G.G. Gubser, I.R. Klebanov, A.M. Polyakov, Gauge theory correlators from noncritical string
theory, Phys. Lett. B 428 (1998) 105, hep-th/9802109.
[3] E. Witten, Anti-de Sitter space and holography, Adv. Theor. Math. Phys. 2 (1998) 253, hepth/9802150.
[4] T. Banks, M.B. Green, Non-perturbative effects in AdS(5) S 5 string theory and d = 4 SUSY
YangMills, JHEP 9805 (1998) 002, hep-th/9804170.
[5] S. Lee, S. Minwalla, M. Rangamani, N. Seiberg, Three-point functions of chiral operators in
D = 4, N = 4 SYM at large N , Adv. Theor. Math. Phys. 2 (1998) 697, hep-th/9806074.
[6] E. DHoker, D.Z. Freedman, W. Skiba, Field theory tests for correlators in the AdS/CFT
correspondence, Phys. Rev. D 59 (1999) 045008, hep-th/9807098.
[7] W. Skiba, Correlators of short multi-trace operators in N = 4 supersymmetric YangMills,
Phys. Rev. D 60 (1999) 105038, hep-th/9907088.
[8] F. Gonzalez-Rey, B. Kulik, I.Y. Park, Non-renormalization of two point and three point
correlators of N = 4 SYM in N = 1 superspace, Phys. Lett. B 455 (1999) 164, hep-th/9903094.
[9] A.C. Petkou, K. Skenderis, A non-renormalization theorem for conformal anomalies, Nucl.
Phys. B 561 (1999) 100, hep-th/9906030.
[10] S. Penati, A. Santambrogio, D. Zanon, Two-point functions of chiral operators in N = 4 SYM
at order g 4 , JHEP 9912 (1999) 006, hep-th/9910197.
[11] S. Penati, A. Santambrogio, D. Zanon, More on correlators and contact terms in N = 4 SYM at
order g 4 , hep-th/0005223.
[12] P.S. Howe, E. Sokatchev, P.C. West, 3-point functions in N = 4 YangMills, Phys. Lett. B 444
(1998) 341, hep-th/9808162.
[13] G. Arutyunov, S. Frolov, Scalar quartic couplings in type IIB supergravity on AdS5 S 5 , Nucl.
Phys. B 579 (2000) 117, hep-th/9912210.
[14] G. Arutyunov, S. Frolov, Four-point functions of lowest weight CPOs in N = 4 SYM4 in
supergravity approximation, Phys. Rev. D 62 (2000) 064016, hep-th/0002170.
[15] G. Arutyunov, S. Frolov, A.C. Petkou, Operator product expansion of the lowest weight CPOs
in N = 4 SYM4 at strong coupling, Nucl. Phys. B 586 (2000) 547, hep-th/0005182.

G. Arutyunov et al. / Nuclear Physics B 602 (2001) 238260

259

[16] H. Liu, A.A. Tseytlin, On four-point functions in the CFT/AdS correspondence, Phys. Rev.
D 59 (1999) 086002, hep-th/9807097.
[17] D. Freedman, S.D. Mathur, A. Matusis, L. Rastelli, Comments on 4-point functions in the
CFT/AdS correspondence, Phys. Lett. B 452 (1999) 61, hep-th/9808006.
[18] G. Chalmers, K. Schalm, The large Nc limit of four-point functions in N = 4 super YangMills
theory from anti-de Sitter supergravity, Nucl. Phys. B 554 (1999) 215, hep-th/9810051.
[19] E. DHoker, D. Freedman, Gauge boson exchange in AdSd+1 , Nucl. Phys. B 544 (1999) 612,
hep-th/9809179.
[20] J.H. Brodie, M. Gutperle, String corrections to 4-point functions in the AdS/CFT correspondence, Phys. Lett. B 445 (1999) 296, hep-th/9809067.
[21] H. Liu, Scattering in anti-de Sitter space and operator product expansion, Phys. Rev. D 60
(1999) 106005, hep-th/9811152.
[22] E. DHoker, D. Freedman, General scalar exchange in AdSd+1 , Nucl. Phys. B 550 (1999) 612,
hep-th/9811257.
[23] E. DHoker, D. Freedman, S. Mathur, A. Matusis, L. Rastelli, Graviton exchange and complete
4-point functions in the AdS/CFT correspondence, Nucl. Phys. B 562 (1999) 353, hepth/9903196.
[24] E. DHoker, D. Freedman, L. Rastelli, AdS/CFT 4-point functions: how to succeed at
z-integrals without really trying, Nucl. Phys. B 562 (1999) 395, hep-th/9905049.
[25] Sanjay, On direct and crossed channel asymptotics of four-point functions in AdS/CFT
correspondence, Mod. Phys. Lett. A 14 (1999) 1413, hep-th/9906099.
[26] E. DHoker, S.D. Mathur, A. Matusis, L. Rastelli, The operator product expansion of N = 4
SYM and the 4-point functions of supergravity, hep-th/9911222.
[27] L. Hoffmann, A.C. Petkou, W. Rhl, A note on the analyticity of AdS scalar exchange graphs
in the crossed channel, Phys. Lett. B 478 (2000) 320, hep-th/0002025.
[28] C.P. Herzog, OPEs and 4-point functions in AdS/CFT correspondence, hep-th/0002039.
[29] L. Hoffmann, A.C. Petkou, W. Rhl, Aspects of the conformal operator product expansion in
AdS/CFT correspondence, hep-th/0002154.
[30] L. Hoffmann, L. Mesref, W. Ruhl, AdS box graphs, unitarity and operator product expansions,
hep-th/0006165.
[31] S. Ferrara, A. Zaffaroni, Superconformal field theories, multiplet shortening and the
AdS5 /SCFT4 correspondence, hep-th/9908163.
[32] F. Gonzalez-Rey, I. Park, K. Schalm, A note on four-point functions of conformal operators in
N = 4 super-YangMills, Phys. Lett. B 448 (1999) 37, hep-th/9811155.
[33] B. Eden, P.S. Howe, C. Schubert, E. Sokatchev, P.C. West, Four-point functions in N =
4 supersymmetric YangMills theory at two loops, Nucl. Phys. B 557 (1999) 355, hepth/9811172.
[34] B. Eden, P.S. Howe, C. Schubert, E. Sokatchev, P.C. West, Simplifications of four-point
functions in N = 4 supersymmetric YangMills theory at two loops, Phys. Lett. B 466 (1999)
20, hep-th/9906051.
[35] M. Bianchi, S. Kovacs, G. Rossi, Y.S. Stanev, On logarithmic behaviour in N = 4 SYM theory,
JHEP 9908 (1999) 020, hep-th/9906188.
[36] B. Eden, C. Schubert, E. Sokatchev, Three-loop four-point correlator in N = 4 SYM, Phys.
Lett. B 482 (2000) 309, hep-th/0003096.
[37] M. Bianchi, S. Kovacs, G. Rossi, Y.S. Stanev, Anomalous dimensions in N = 4 SYM at order
g 4 , Nucl. Phys. B 584 (2000) 216, hep-th/0003203.
[38] M. Bianchi, M. Green, S. Kovacs, G. Rossi, Instantons in supersymmetric YangMills and Dinstantons in IIB superstrings theory, JHEP 9808 (1998) 013, hep-th/9807033.
[39] N. Dorey, V. Khoze, M. Mattis, S. Vandoren, YangMills instantons in the large N limit and
the AdS/CFT correspondence, Phys. Lett. B 442 (1998) 145, hep-th/9808157.

260

G. Arutyunov et al. / Nuclear Physics B 602 (2001) 238260

[40] N. Dorey, T. Hollowood, V. Khoze, M. Mattis, S. Vandoren, Multi-instantons and Maldacenas


conjecture, JHEP 9906 (1999) 023, hep-th/9810243.
[41] N. Dorey, T. Hollowood, V. Khoze, M. Mattis, S. Vandoren, Multi-instanton calculus and the
AdS/CFT correspondence in N = 4 superconformal field theory, Nucl. Phys. B 552 (1999) 88,
hep-th/9901128.
[42] B. Eden, A. Petkou, C. Schubert, E. Sokatchev, Partial non-renormalization of the stress-tensor
four-point function in N = 4 SYM and AdS/CFT, hep-th/0009106.
[43] K. Intriligator, W. Skiba, Bonus symmetry and the operator product expansion of N = 4 superYangMills, hep-th/9905020.
[44] D. Anselmi, The N = 4 quantum conformal algebra, Nucl. Phys. B 541 (1999) 369, hepth/9809195.
[45] H. Osborn, A.C. Petkou, Implications of conformal invariance in field theories for general
dimensions, Ann. Phys. 231 (1994) 311, hep-th/9307010.
[46] D. Anselmi, Quantum conformal algebras and closed conformal field theory, Nucl. Phys. B 554
(1999) 415, hep-th/9811149.

Nuclear Physics B 602 (2001) 261288


www.elsevier.nl/locate/npe

Eikonal propagators and high-energy partonparton


scattering in gauge theories
Enrico Meggiolaro
Dipartimento di Fisica, Universit di Pisa, Via Buonarroti 2, I-56127 Pisa, Italy
Received 4 December 2000; accepted 7 March 2001

Abstract
In this paper we consider soft high-energy partonparton scattering processes in gauge theories,
i.e., elastic scattering processes involving partons at very high squared energies s in the center of
mass and small squared transferred momentum t (s , t  s, typically |t|  1 GeV2 ). By a
direct resummation of perturbation theory in the limit we are considering, we derive expressions for
the truncatedconnected quark (antiquark) propagator in an external gluon field, as well as for the
residue at the pole of the full unrenormalized propagator, both for scalar and fermion gauge theories.
These are the basic ingredients to derive high-energy partonparton scattering amplitudes, using the
LSZ reduction formulae and a functional integral approach. The above procedure is also extended to
include the case in which at least one of the partons is a gluon. The meaning and the validity of the
results are discussed. 2001 Published by Elsevier Science B.V.
PACS: 11.15.-q; 11.15.Bt; 11.80.Fv; 12.38.Cy

1. Introduction
Since the late 1950s a lot of models have been proposed to describe the physics of
hadronhadron elastic scattering at high energies. Some of these are pre-QCD models
(see, e.g., Refs. [1] and [2]), while others are QCD-inspired models (see, e.g., Ref. [3]).
With the advent of Quantum Chromo-Dynamics (QCD), which is now believed to be
the correct theory of hadrons and their interactions, many theoretical physicists started
to study the high-energy behaviour of gauge theories directly from their first principles. In
particular, a lot of work has been done within the framework of perturbation theory in order
to find systematic procedures for extracting the high-energy behaviour of each amplitude
and for summing these contributions using a leading-log or eikonal approximation scheme
[4,5]. Even if partially successful, the results obtained using these procedures are not
completely satisfactory and are not able to explain the most relevant phenomena.
E-mail address: enrico.meggiolaro@df.unipi.it (E. Meggiolaro).
0550-3213/01/$ see front matter 2001 Published by Elsevier Science B.V.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 1 1 5 - 8

262

E. Meggiolaro / Nuclear Physics B 602 (2001) 261288

In particular, for soft high-energy scattering processes, i.e., elastic scattering processes
at very high squared energies s in the center of mass and small squared transferred
momentum t (s , t  s, typically |t|  1 GeV2 ), QCD perturbation theory cannot
be safely applied, since t is too small, and one has therefore to appeal to nonperturbative
QCD. P.V. Landshoff and O. Nachtmann were maybe the first who argued [6] that the
theoretical description of measurable quantities of soft high-energy reactions (like the
total cross sections, for example) should involve in an essential way nonperturbative QCD.
Later on, Nachtmann developed a nonperturbative analysis, based on QCD, of these soft
high-energy scattering processes [7,8]: he derived formal expressions for the quarkquark
(and also quarkantiquark and antiquarkantiquark) scattering amplitudes in the abovementioned limit, by using a functional integral approach and an eikonal approximation to
the solution of the Dirac equation in the presence of an external non-Abelian gauge field.
In a previous paper [9] we proposed an alternative approach to high-energy quarkquark
scattering based on a first-quantized path-integral description of quantum-field theory
developed by Fradkin in the early 1960s [10]. In this approach one obtains convenient
expressions for the truncatedconnected scalar propagators in an external (gravitational,
electromagnetic, etc.) field, and the eikonal approximation can be easily recovered in
the relevant limit. Knowing the truncatedconnected propagators, one can then extract,
in the manner of Lehmann, Symanzik and Zimmermann (LSZ) [11], the scattering matrix
elements in the framework of a functional integral approach. (We remind the reader that this
method was originally adopted in Ref. [12] in order to study Planck-energy gravitational
scattering. In Ref. [9] we have translated the procedure and some results of Ref. [12] to
the case of Quantum Chromo-Dynamics, i.e., the case of quarks coupled to an external
non-Abelian gauge field in a flat spacetime.)
In this paper we shall address the same problem in an even more immediate way:
by a direct resummation of perturbation theory in the high-energy limit that we are
considering, we shall derive expressions for the truncatedconnected quark (antiquark)
propagator in an external gluon field, as well as for the residue at the pole of the full
unrenormalized propagator. (This last is also a basic ingredient to derive partonparton
scattering amplitudes, since it appears in the LSZ reduction formulae [11].) The idea of
using the background-field method in (perturbative) QCD calculations is certainly not new:
for example, the light-cone expansions of the quark and gluon propagators in external fields
are worked out in the Appendix A of Ref. [13] (see also [14]). Moreover, the backgroundfield expansion of the quark propagator in the high-energy limit has been proposed (and
derived in great detail) in a series of recent works by Balitsky, e.g., Ref. [15] (see also the
recent review [16] and references therein).
The paper is organized as follows. In Section 2 we begin, for simplicity, with the case
of scalar QCD, i.e., the case of a spin-0 quark coupled to a non-Abelian gauge field. In
Section 3 we extend the results to the case (more interesting from the physical point of
view) of real fermion QCD: that is a spin-1/2 quark coupled to a non-Abelian gauge
field. In Section 4 the above procedure is also extended to include the case in which at
least one of the partons is a gluon: the meaning and the validity of the results so obtained
are discussed. The truncatedconnected propagators in an external gluon field are the

E. Meggiolaro / Nuclear Physics B 602 (2001) 261288

263

basic ingredients to derive high-energy partonparton scattering amplitudes, using the LSZ
reduction formulae and a functional integral approach. This was done in Ref. [9] and it will
be quickly reviewed in Section 5 for the convenience of the reader. The results obtained in
Section 5 are in agreement with those of Refs. [79], where they were derived with different
methods. In Section 6 we give a summary of the main results, a detailed discussion of the
approximations involved and the conclusions.

2. The scattering of scalar quarks and antiquarks


We begin, for simplicity, with the case of scalar QCD, i.e., the case of a spin-0 quark

(described by the scalar field ) coupled to a non-Abelian gauge field A Aa Ta , Ta


(a = 1, . . . , Nc2 1) being the generators of the Lie algebra of the colour group SU(Nc ).
We take to stay in the vector space of the fundamental representation of SU(Nc ): i.e.,
stands for a vector i (i = 1, . . . , Nc ) in the colour space and Ta are the generators in the
fundamental representation [(Ta )ij , i, j {1, . . . , Nc }]. We limit ourselves to the case of
one single flavour. The unrenormalized Lagrangian is:


1 a a
F
,
L(, , A) = D D m20 F
4
where D = + igA is the covariant derivative.
The unrenormalized scalar quark propagator will be denoted as

 
T i (x)j (y) = Sij (x y).

(2.1)

(2.2)

Let us define the physical quark mass m, taken to be the pole mass, and the residue ZW
at the pole of the unrenormalized quark propagator by the following two equations:

1 
iZW ij

,
Sij (p)
S(p)
(2.3)
 2 2 = 0,
2
2
p =m
p 2 m2 p m + i
where Sij (p) is the unrenormalized propagator in the momentum space:

Sij (p) d 4 z eipz Sij (z).

(2.4)

[We remind the reader that ZW is not, in general, equal to the scalar-field renormalization
1/2
constant Z2 (defined as = Z2 R , where is the bare field and R is the renormalized
field); but it is equal to Z2 z , where z is the residue at the pole of the renormalized
propagator, defined as:
SRij (p)

p 2 m2

i z ij
.
p2 m2 + i

(2.5)

In fact one has that:


Sij (p) = Z2 SRij (p)

p 2 m2

that is, ZW = Z2 z , Q.E.D.]

p2

iZW ij
iZ2z ij
2
,
2
m + i p m2 + i

(2.6)

264

E. Meggiolaro / Nuclear Physics B 602 (2001) 261288

In order to derive the scattering matrix elements following the LSZ approach [11], we
need to know the on-shell truncatedconnected Green functions, which are obtained from
the connected Green functions by removing the external legs calculated on-shell. We first
consider the scattering of a quark in a given external gluon field A :
(p, j ) (p , i),

(2.7)

where i, j are colour indices (i, j = 1, . . . , Nc ). We define the truncatedconnected


propagator in the external gluon field A , in the momentum space as:
Sij(tc) (p, p |A)

lim

p 2 ,p 2 m2

p 2 m2
p  2 m2
,
Sij (p, p |A)
i
i

(2.8)

where m is the physical mass defined above and Sij (p, p |A) is the Fourier transform
of Sij (x, y|A), the scalar propagator in an external gluon field, in the coordinate
representation:




Sij (p, p |A) d 4 x d 4 y exp i(p x py) Sij (x, y|A).
(2.9)
[Let us observe that, for convenience and simplicity reasons, we have not included a factor
2
ZW
in the definition (2.8) of the truncatedconnected propagator. These omitted factors
must be properly included when deriving the scattering amplitudes (see Section 5).] In
the following we shall compute the truncatedconnected propagator Sij(tc) (p, p |A) in the
so-called eikonal approximation, which is valid in the case of scattering particles with
very high energy (E p0 |p |  m) and small transferred momentum q p p (i.e.,

|t|  E, where t = q 2 ). For example, if p p (E, E, 0, 0), one has that p+



 0, where the following general notation has been used for a
p+ 2E and p p
given four-vector V :
V+ V 0 + V 1 ,

V V 0 V 1 .

(2.10)
V ,

We shall call V+ and V the longitudinal components of the four-vector


while
2
3


V (V , V ) is the component of V in the transverse plane (y, z).
(tc)
Our strategy consists in evaluating the truncatedconnected propagator Sij (p, p |A) in
each order in perturbation theory considering L [D ] D m20 = L0 + Lint ,
where
L0 = m2

(2.11)

is the free (i.e., unperturbed) quark Lagrangian, which defines the free quark
propagator i/(p2 m2 + i), with the physical mass m, and


Lint = ig A + g 2 A A + m2
(2.12)
is the interaction Lagrangian, i.e., the perturbation. The squared-mass shift m2 is
defined as
m2 m2 m20 .

(2.13)

Let us start, therefore, by evaluating the nth order term (n  1) in the perturbative
expansion of the truncatedconnected scalar propagator in an external gluon field A , in

E. Meggiolaro / Nuclear Physics B 602 (2001) 261288

265

(a)

(b)
Fig. 1. (a) The Feynman diagram corresponding to the nth order term (n  1) in the perturbative

expansion of the truncatedconnected quark propagator in an external gluon field Aa , in the eikonal
approximation. [See Eq. (2.14) for the scalar case and Eqs. (3.12) and (3.16) for the fermion case.]
(b) The Feynman diagram which defines the nth order perturbative term (n  1) of the gluon

matrix element (4.1). Crosses represent insertions of the external gluon field Aa . The four-momenta
q1 , . . . , qn are taken to be flowing into the diagram.

the eikonal approximation. This contribution, that we shall indicate as [Sij (p, p |A)](n) ,
is schematically represented in Fig. 1(a): only the quarkquarkgluon vertex [the first term
appearing in Lint in Eq. (2.12)] contributes to the propagator in the eikonal limit that we are
considering. Let us discuss in detail how this approximation is justified. The key-point (see
also Ref. [4] and references therein) is that, in the high-energy limit we are considering,
p p and quarks retain their large longitudinal momenta during their scattering process.
In other words, we are assuming that the external gluon field has a frequency distribution,
i.e., Fourier transform, A (k) [defined by Eq. (2.15) below], such that the relevant phasespace region in d 4 k is the one where k is negligible when compared to p. In Refs. [7,8],
the eikonal approximation has been done under the hypothesis that the external gluon field
contains only a limited range of frequencies: in other words, A (x) is assumed to vary
slowly on the scale set by the wavelength of the incoming waves. We shall come back to a
detailed discussion on all these approximations and hypotheses in Section 5.
One must observe (see Table 1) that the scalar theory, in addition to the quarkquark
gluon vertex, has also two other types of vertices of the form g 2 A A (quarkquark
gluongluon) and m2 (squared-mass shift). Yet one can easily be convinced that the
contributions due to these additional scalar vertices to the nth order term in the perturbative
(tc)
expansion of the truncatedconnected scalar propagator Sij (p, p |A) in the high-energy
limit are suppressed with respect to the contribution coming from the quarkquarkgluon
(tc)

266

E. Meggiolaro / Nuclear Physics B 602 (2001) 261288

Table 1
The relevant vertices in scalar QCD
Vertex

Feynman rule

i (p2 )j (p1 )Aa

i j Aa Ab
m2 i j

ig(p1 + p2 ) (Ta )ij


ig 2 g {Ta , Tb }ij
im2 ij

couplings. This is essentially due to the fact that the quarkquarkgluongluon vertex and
the squared-mass shift vertex do not carry momentum (see Table 1).
(tc)
By virtue of the above-mentioned approximations, the expression for [Sij (p, p |A)](n)
reads as follows (see Fig. 1(a)):
 (tc)

Sij (p, p |A) (n)

 4
d 4 q1
d qn

(2)4 (4) (q q1 qn )
(2)4
(2)4

 


ig2pn A n (qn ) ig2p1 A 1 (q1 ) ij

i
i

,
(p + q1 + + qn1 )2 m2 + i
(p + q1 )2 m2 + i

(2.14)

where q = p p is the transferred momentum. We have indicated with A (k) the


Fourier transform at four-momentum k of the external gluon field A (z) in the coordinate
representation:

A (k) d 4 z eikz A (z).
(2.15)
In the eikonal limit, one also finds that:
(p + q1 + + qn1 )2 m2 2E(q1 + + q(n1) ),

(2.16)

and similarly for the other expressions. Therefore, Eq. (2.14) becomes, using also
Eq. (2.15) to express the external gluon field in the coordinate representation:

 (tc)
Sij (p, p |A) (n)



 


d 4 bn d 4 b1 eiqbn ig2pn An (bn ) ig2p1 A1 (b1 ) ij


 4

d qn1
ieiqn1 (bn1 bn )
d 4 q1 ieiq1 (b1 bn )

. (2.17)

4
4
(2) 2Eq1 + i
(2) 2E(q1 + + q(n1) ) + i
The last integration can be easily performed and one obtains:
 (tc)

Sij (p, p |A) (n)


 



d 4 bn d 4 b1 eiqbn ig2pn An (bn ) ig2p1 A1 (b1 ) ij

E. Meggiolaro / Nuclear Physics B 602 (2001) 261288

267


1 (2) 
b(n1) bn (b(n1) bn ) (bn+ b(n1)+ )
2E
 4

d qn2
d 4 q1 ieiq1 (b1 bn1 )
ieiqn2 (bn2 bn1 )

. (2.18)

(2)4 2Eq1 + i
(2)4 2E(q1 + + q(n2) ) + i

In the derivation of this result, we have used the following integral expression for the step
function [ () = 1, for > 0; () = 0, for < 0]:
() =

1
2i

+

ei
d.
i

(2.19)

Therefore, if we proceed recursively, we obtain the following result:


 (tc)

Sij (p, p |A) (n)





d 2 bn dbn dbn+ db(n1)+ db1+ eiqbn
1
(bn+ b(n1)+ ) (b2+ b1+ )
(2E)n1

 



n
1
igp An (bn ) igp A1 (b1 ) ij 

bi =bn ;bi =bn

(2.20)

Obviously, we can parametrize the coordinates bi (i = 1, . . . , n) in the integral as follows:


bi = b + pi

(i = 1, . . . , n),

(2.21)

where i (or, better, i mi ) are proper-time variables and b is a four-vector with b =


bn , b = bn and a fixed b+ . In fact, using the fact that p (E, E, 0, 0), i.e., p 0
and p 0 , we find that bi = b and bi = b , i = 1, . . . , n. Moreover, q = p p,
so that q+ , q 0 and qbn 
q bn = 
q b qb. Therefore, using also the fact
that dbi+ = p+ di = 2Edi , we can re-write Eq. (2.20) as follows:
 (tc)

Sij (p, p |A) (n)



 3  iqb
2E
d b e
d1 dn (n n1 ) (2 1 )
 



igpn An (b + pn ) igp1 A1 (b + p1 ) ij ,

(2.22)

where we have used the notation:


 3 
d b d 2 b db .

(2.23)

As a general rule, in [d 3b] one must not include the longitudinal component of b which
is parallel to p . In other words, if p p (E, E, 0, 0) (i.e., p+ 2E, p 0), one
has that [d 3 b] d 2 b db , while, if p p (E, E, 0, 0) (i.e., p+ 0, p 2E),
then [d 3 b] d 2 b db+ .
(tc)
Eq. (2.22) is the nth order term of Sij (p, p |A). Summing all orders (n  1), we finally
obtain (see also Ref. [9]):

268

E. Meggiolaro / Nuclear Physics B 602 (2001) 261288

Sij(tc) (p, p |A)



2E

= 2E





+

A (b + p )p d 1
d 3 b eiqb T exp ig




d 3 b eiqb Wp (b) 1 ij ,

ij

(2.24)

where Wp (b) = T exp( ) is the time-ordered exponential, defined as:




+
Wp (b) T exp ig




A (b + p )p d

d1

dn (n n1 ) (2 1 )

n=0

 


igpn An (b + pn ) igp1 A1 (b + p1 ) .

(2.25)

Eq. (2.24) gives the expression for the truncatedconnected scalar propagator in an external
gluon field, in the eikonal approximation.
Proceeding exactly in the same way, we can also derive the following expression for the
nth order term (n  1) in the perturbative expansion of the truncatedconnected propagator
of a scalar antiquark in an external gluon field A , (p, j ) (p , i), in the eikonal
approximation:
 
 (tc) 
p,j p  ,i A (n)
S



 3  iqb
2E
d b e
d1 dn (n n1 ) (2 1 )

 


igp 1 A1 (b + p1 ) igpn An (b + pn ) j i .
(2.26)
And therefore, summing all orders:
 

p  ,i A
S (tc) p,j



+

 3  iqb
2E
d b e
A (b + p )p d 1 .
T exp ig

(2.27)

ji

We have denoted with T exp( ) the antitime-ordered exponential, defined as:


+

T exp ig
A (b + p )p d

d1


dn (n n1 ) (2 1 )

n=0


 

igp1 A1 (b + p1 ) igpn An (b + pn ) .

(2.28)

Let us observe that in the antitime ordering the matrices A are ordered from left to right as
they appear along the path going from = to = + (i.e., when increasing the

E. Meggiolaro / Nuclear Physics B 602 (2001) 261288

269

proper time). On the contrary, in the time ordering they are ordered from left to right as
they appear along the path going from = + to = (i.e., when decreasing the
proper time).
Now, using the fact that A = Aa T a , with (T a ) = T a , i.e., (T a )j i = (T a )ij , we find:




A1 (b + p1 ) An (b + pn ) j i = An (b + pn ) A1 (b + p1 ) ij . (2.29)
Therefore, we can write Eq. (2.27) as follows:
 

S (tc) p,j
p  ,i A

+



 3  iqb

T exp ig
2E
d b e
A (b + p )p d 1

2E

ij




d 3 b eiqb Wp (b) 1 ij ,

(2.30)

where Wp (b) has been defined in Eq. (2.25).


When comparing with the result (2.24), we see that the scattering amplitude of an
antiquark in the external gluon field A is equal to the scattering amplitude of a quark
in the charge-conjugated (C-transformed) gluon field A = At = A , as expected.
3. The scattering of quarks and antiquarks
We want now to extend the results obtained in the previous section to the case (more
interesting from the physical point of view) of real fermion QCD: that is a spin-1/2
quark coupled to a non-Abelian gauge field.
As before, we limit ourselves to the case of one single flavour. The unrenormalized QCD
Lagrangian is:




1 a a
L , , A = i D m0 F
F
,
4

(3.1)

where D = + igA is the covariant derivative [A Aa Ta , Ta (a = 1, . . . , Nc2 1)


being the generators of the Lie algebra of the colour group SU(Nc )] and stands for a
vector i (i = 1, . . . , Nc ) in the colour vector space of the fundamental representation.
The unrenormalized quark propagator will be denoted as
 

T i (x) j (y) = Gij (x y).
(3.2)
Let us define the physical quark mass m, taken to be the pole mass, and the residue ZW
at the pole of the unrenormalized quark propagator by the following two equations:

1 
iZW ij

ij (p)
,
G(p)
(3.3)
G
 2 2 = 0,
2
2
p =m
p

m + i
p m
ij (p) is the unrenormalized propagator in the momentum space:
where G

ij (p) d 4 z eipz Gij (z).
G

(3.4)

270

E. Meggiolaro / Nuclear Physics B 602 (2001) 261288

In Eq. (3.3) we have used the notation: a a .


As in the previous section, we consider the scattering of a quark in a given external gluon
field A . The truncatedconnected fermion propagator in the momentum space is defined
as:
p  m
p m
(tc) (p, p |A)
lim
G
(3.5)
Gij (p, p |A)
,
ij
i
i
p 2 ,p  2 m2
ij (p, p |A) is the Fourier
where m is the physical quark mass defined above and G
transform [see Eq. (2.9)] of Gij (x, y|A), the truncatedconnected fermion propagator in
the coordinate representation. The matrix element for the scattering of a quark in a given
external gluon field A ,
(p, j, ) (p , i, ),

(3.6)

(tc) (p, p |A)


where i, j are colour indices and , are spin indices, is given by u (p )G
ij
u (p), where u (p) are the positive-energy spinors [(p m)u (p) = u (p)(p m) =
0] with the usual relativistic normalization:
u (p) u (p) = 2p ,

u (p)u (p) = 2m .

(3.7)

In the high-energy limit we are considering, we can make the following replacement:
(p, p |A)u (p) S (p, p |A),
u (p )G
ij
ij
(tc)

(tc)

(3.8)

where Sij(tc) (p, p |A) is the truncatedconnected propagator for a scalar (i.e., spin-0) quark
in the external gluon field A , which was discussed in the previous section. Let us see how
this approximation is justified. As in the scalar case, our strategy consists in evaluating the
(tc) (p, p |A), or better the quantity (3.8), in each order
truncatedconnected propagator G
ij
D m ) = L + L , where

in perturbation theory, considering L (i

0
0
int

m)
L0 = (i

(3.9)

is the free (i.e., unperturbed) quark Lagrangian, which defines the free quark
propagator i/(p m + i), with the physical mass m, and
A + m

Lint = g

(3.10)

is the interaction Lagrangian, i.e., the perturbation. The mass shift m is defined as:
m m m0 .

(3.11)

Using the fact that u(p


 )(p  m)u(p) = u(p
 )(p m)u(p) = 0 and the definition
(3.5) of the truncatedconnected fermion propagator, one can easily derive the following
expression for the nth order term in the perturbative expansion (n  1):


(tc) (p, p |A)u (p)
u (p )G
ij
(n)

 4
4
d q1
d qn
(ferm)
=

(2)4 (4)(q q1 qn )N, ij (q1 , . . . , qn )


(2)4
(2)4
i
i

(3.12)

,
2
2
2
(p + q1 + + qn1 ) m + i
(p + q1 ) m2 + i

E. Meggiolaro / Nuclear Physics B 602 (2001) 261288

271

(ferm)
where N,
ij (q1 , . . . , qn ) is given by:
(ferm)
N,
ij (q1 , . . . , qn )


u (p ) ig n A n (qn ) + im(2)4 (4) (qn ) 1 (p + q1 + + qn1 + m)



(p + q1 + m) ig 1 A 1 (q1 ) + im(2)4 (4)(q1 ) 1


u (p).
ij

(3.13)
By virtue of the eikonal approximation, the relevant phase-space region in the nth order
term of the perturbative expansion (3.12) has the property that qi , m and m are negligible
(ferm)
(q1 , . . . , qn ). Therefore, the numerator
when compared to p in the numerator N,ij
(ferm)
(q1 , . . . , qn ) in Eq. (3.12) can be approximated as:
N,ij

(ferm)
(ferm)
N,ij
(q1 , . . . , qn ) ig2pn A n (qn ) N
(q1 , . . . , qn1 ) ij .

(3.14)

Using the fact that:


(ferm)
N,ij
(q1 ) = u (p ) ig 1 A 1 (q1 ) + im(2)4 (4)(q1 ) 1 ij u (p)


ig2p1 A 1 (q1 ) ij ,
we find, proceeding recursively from Eq. (3.14):

 


(ferm)
N,ij (q1 , . . . , qn ) ig2pn A n (qn ) ig2p1 A 1 (q1 ) ij .

(3.15)

(3.16)

Apart from the delta function in front, which simply reflects the fact that fermions retain
their helicities during the scattering process in the high-energy limit, this is exactly the
term we would have expected at the numerator of the nth order term in the perturbative
expansion of the truncatedconnected scalar propagator Sij(tc) (p, p |A) in the high-energy
limit [the denominators in (3.12) are already equal to the scalar case!]. In fact, as reported
in Table 1, the factors (ig2p ) in (2.14) come from the quarkquarkgluon vertex of the
scalar theory in the high-energy limit (when p p ). From Eqs. (3.12), (3.16) and (2.14)
we derive

 (tc)


(tc) (p, p |A)u (p)
Sij (p, p |A) (n) ,
u (p )G
(3.17)
ij
(n)
for each order n in the perturbative expansion, so proving Eq. (3.8). Therefore, using
the result (2.24) derived in the previous section, we find the following expression for the
quantity (3.8) in the eikonal approximation (see also Refs. [9,15,16]):
(p, p |A)u (p) S (p, p |A)
u (p )G
ij
ij





+
 3  iqb

T exp ig
2E
d b e
A (b + p )p d 1
(tc)

(tc)


= 2E


d 3 b eiqb [Wp (b) 1]ij ,

ij

(3.18)

where Wp (b) is the time-ordered Wilson string along the path x( ) = b + p ,


[, +], defined in Eq. (2.25).

272

E. Meggiolaro / Nuclear Physics B 602 (2001) 261288

Proceeding exactly in the same way, we can also derive the following expression
for the truncatedconnected propagator of an antiquark in an external gluon field A ,

 , i, ), in the eikonal approximation:


(p,
j, ) (p
 

(tc) ( p,j p ,i |A)v (p ) S (tc)  A
v (p)G
p,j
p ,i

+



 3  iqb

T exp ig
2E
d b e
A (b + p )p d 1

= 2E



d 3 b eiqb Wp (b) 1 ij ;


ij

(3.19)

v (p) are the negative-energy spinors [(p + m)v (p) = v (p)(p + m) = 0] with the
usual relativistic normalization:
v (p) v (p) = 2p ,

v (p)v (p) = 2m .

(3.20)

In fact, one can easily derive the following expression for the nth order term in the
perturbative expansion (n  1):


(tc) ( p,j p ,i |A)v (p )
v (p)G
(n)

 4
d 4 q1
d qn
(ferm) (q1 , . . . , qn )
=

(2)4 (4)(q q1 qn )N
,ij
(2)4
(2)4
i
i

(3.21)

,
(p + q1 )2 m2 + i
(p + q1 + + qn1 )2 m2 + i
(ferm) (q1 , . . . , qn ) is given by:
where N
,ij
(ferm) (q1 , . . . , qn )
N
,ij


v (p) ig 1 A 1 (q1 ) im(2)4 (4)(q1 ) 1 (p + q1 m)



(p + q1 + + qn1 m) ig n A n (qn ) im(2)4 (4) (qn ) 1 j i v (p ).


(3.22)
Proceeding as for the derivation of Eq. (3.16) and making use of Eq. (2.29), we find:
(ferm) (q1 , . . . , qn )
N
,ij

 


ig2p1 A 1 (q1 ) ig2pn A n (qn ) j i



 


= ig2pn A n (qn ) ig2p1 A 1 (q1 ) ij .

(3.23)

Apart from the delta function in front, this is exactly the term we would have expected
at the numerator of the nth order term in the perturbative expansion of the truncated

connected scalar antiquark propagator S (tc) (p,j


p  ,i |A) in the high-energy limit.
From Eqs. (3.21) and (3.23) we derive





(tc) ( p,j p ,i |A)v (p )


v (p)G
S (tc) (p,j
p  ,i |A) (n) , (3.24)
(n)
for every order n in the perturbative expansion, so proving Eq. (3.19). As for the scalar
case, going from quarks to antiquarks corresponds just to the change from the fundamental
representation Ta of SU(Nc ) to the complex conjugate representation Ta = Ta .

E. Meggiolaro / Nuclear Physics B 602 (2001) 261288

273

4. The scattering of gluons


By using the same techniques developed in the previous sections, we shall evaluate the
truncatedconnected gluon propagator in a given external gluon field Ab , in the eikonal

approximation. We call this quantity D (tc)
 ,a  a (k, k |A), whose nth order perturbative
term is defined schematically in Fig. 1(b), where the external legs are supposed to be
truncated on-shell (k 2 , k 2 0). More precisely, we shall evaluate the following quantity:



( ) (k  )D (tc)
 ,a  a (k, k |A)()(k),

(4.1)

are the polarization four-vectors (,  {1, 2}):




k ()(k)k 2 =0 = 0.
()(k) (
 ) (k) =  ,

where

() (k)

(4.2)

This quantity should describe (under certain approximations that will be discussed below
and also in the next section) the scattering matrix element of a gluon in a given external
gluon field Ab :
g(k, a, ) g(k  , a  ,  );

(4.3)

a, a  {1, . . . , Nc2 1} are colour indices and ,  {1, 2} are spin indices. In the eikonal
approximation the dominant interaction between the incident gluon and the external gluon
field is represented by the three-gluon vertex, which is linear in the four-momentum of the
gluon (while the four-gluon vertex is not dependent on the momentum). Its Feynman rule
is given by:
Va11a22a33 (k1 , k2 , k3 )


= gf a1 a2 a3 g1 2 (k1 + k2 )3 + g2 3 (k2 k3 )1 + g3 1 (k3 + k1 )2 , (4.4)
where the four-momenta k1 and k2 are taken to be flowing into the vertex, while the fourmomentum k3 is taken to be flowing out from the vertex. The explicit expression of the nth
order perturbative term of the quantity (4.1), which is schematically defined in Fig. 1(b), is
given by:
   (tc)


( ) (k )D  ,a  a (k, k  |A)()(k) (n)



 4
d 4 q1
d qn
(gluon)

(2)4 (4)(q q1 qn )N ,a  a (q1 , . . . , qn )


(2)4
(2)4
i
i

(4.5)

,
2
(k + q1 + + qn1 ) + i
(k + q1 )2 + i
where q k  k is the transferred momentum and N ,a  a (q1 , . . . , qn ) is given by:
(gluon)

(gluon)

N ,a  a (q1 , . . . , qn )

ab1 c1
()(k)V
(k, q1 , k + q1 )A b11 (q1 )g 1 2 c1 c2
1 1

Vc22b22c33 (k + q1 , q2 , k + q1 + q2 )
b a

2n2 n
 n

V2n2
n  (k + q1 + + qn1 , qn , k )Abn (qn )( ) (k ).

(4.6)

274

E. Meggiolaro / Nuclear Physics B 602 (2001) 261288

It is not difficult to verify that, in the eikonal approximation, using the properties (4.2) for
the polarization four-vectors, the quantity (4.6) simplifies as follows:

 


(gluon)
N ,a  a (q1 , . . . , qn )  ig2k n A n (qn ) ig2k 1 A 1 (q1 ) a  a , (4.7)
where we have used the notation:
b
A Ab T(adj)
,

(4.8)

b
T(adj)
being the Nc2 1 matrices of the SU(Nc ) Lie algebra in the adjoint representation:
 a 
T(adj) bc = if abc .
(4.9)

The expression (4.5), with the result (4.7), is perfectly analogous to the corresponding
expression appearing in Eq. (2.14) for the case of the scalar quark. By proceeding exactly
as in Section 2, we can thus further approximate the above-written Eq. (4.5) as follows:
   (tc)


( ) (k )D  ,a  a (k, k  |A)()(k) (n)





 3  iqb
 2E
d b e
d1 dn (n n1 ) (2 1 )
 



igk n An (b + kn ) igk 1 A1 (b + k1 ) a  a .
(4.10)
Therefore, summing all orders (n  1), we finally obtain:


(tc)
( ) (k  )D  ,a  a (k, k  |A)()(k)

 3  iqb 

Vk (b) 1 a  a ,
d b e
 2E

(4.11)

where Vk (b) is the Wilson string along the path x( ) = b + k ( [, +]), in the
adjoint representation, defined as:


+
Vk (b) T exp ig




A (b + k )k d

d1

dn (n n1 ) (2 1 )

n=0


 

igk n An (b + pn ) igk 1 A1 (b + k1 ) .

(4.12)

Eq. (4.11) gives the expression for the scattering matrix element of a gluon in a given
external gluon field, in the eikonal approximation.

5. Partonparton scattering amplitudes


In the previous sections we have derived expressions for the truncatedconnected quark
(antiquark) propagator in a given external gluon field A , by a direct resummation of
perturbation theory in the limit of very high energy and small transferred momentum. The

E. Meggiolaro / Nuclear Physics B 602 (2001) 261288

275

procedure has been also extended to include the case in which the scattering parton is a
gluon: additional approximations are necessary and they will be discussed below.
The truncatedconnected propagators in an external gluon field are the basic ingredients
to derive high-energy partonparton scattering amplitudes, using the LSZ reduction
formulae and a functional integral approach. This was done in Ref. [9] and it will be
quickly reviewed here for the convenience of the reader. Let us consider, for example,
the elastic scattering process of two scalar quarks with initial four-momenta p1 and p2 and
final four-momenta p1 and p2 :
j (p1 ) + l (p2 ) i (p1 ) + k (p2 )

(5.1)

[i, j, k, l {1, . . . , Nc } are colour indices]. In the center-of-mass reference system (c.m.s.),
taking the initial trajectories of the two quarks along the x 1 -axis, the four-momenta p1 , p2 ,
p1 and p2 are given, in the limit of soft high-energy scattering, s = (p1 + p2 )2
and t = (p1 p1 )2  s, by:
p1 p1 (E, E, 0 t ),

p2 p2 (E, E, 0 t ).

(5.2)

Using the LSZ reduction formulae and a functional integral approach, one finds the
following expression for the scattering matrix element [with the plane wave functions
normalized as p (x) = exp(ipx)]:


i (p1 )k (p2 )|(S 1)|j (p1 )l (p2 )

 (tc)


1 
(tc)
2 Sij(tc) (p1 , p1 |A)Skl
(p2 , p2 |A) A + Skj
(p1 , p2 |A)Sil(tc) (p2 , p1 |A) A ,
ZW
(5.3)
2
2
where ZW is the residue at the pole (i.e., for p m ) of the unrenormalized quark
propagator [see Eq. (2.3)]. The expectation value O(A)A of an arbitrary functional O(A)
of the gluon field A is defined as:






1
1
i
O(A)A
,
[dA] O(A) exp
d 4 x Fa Fa det D D + m2
ZQCD
4
(5.4)
where ZQCD is the partition function for scalar QCD:
 


ZQCD [dA][d][d ] exp i d 4 x L(, , A)





1

i
.
= [dA] exp
(5.5)
d 4 x Fa Fa det D D + m2
4
The determinant in Eqs. (5.4) and (5.5) comes from the integration over the scalar degrees
of freedom. In fact, the Lagrangian L(, , A) is bilinear in the scalar fields and :






d 4 x (D ) D m2 = d 4 x D D + m2 .
(5.6)
Therefore, the functional integral over the scalar fields and is an ordinary Gaussian
integral and can be performed in the standard way, to give (apart from an irrelevant
constant) {det[D D + m2 ]}1 .

276

E. Meggiolaro / Nuclear Physics B 602 (2001) 261288

The first expectation value in Eq. (5.3) corresponds to the t-channel scattering of the two
particles, with squared transferred momentum equal to: (p1 p1 )2 t  s. The second
expectation value corresponds instead to the u-channel scattering of the two quarks. In
other words, the squared transferred momentum, flowing from one quark to the other, is
equal to:
(p1 p2 )2 u = 4m2 s t s,

(5.7)

in the limit s with t, m2  s. In this high-energy limit the contribution coming from
the second expectation value in Eq. (5.3) is smaller by at least a factor of s, when compared
with the first expectation value, and hence is negligible. [One can be easily convinced
of this by considering the Feynman diagrams of the process in the perturbation theory:
the diagrams which correspond to the second piece in Eq. (5.3) have intermediate gluons
carrying a big squared transferred momentum u s, so that their propagators suppress
the corresponding amplitude.] Therefore, in our limit:




1  (tc)
(tc)
i (p1 )k (p2 )|(S 1)|j (p1 )l (p2 ) 2 Sij (p1 , p1 |A)Skl (p2 , p2 |A) A ,
ZW

(5.8)
(tc)


where for the truncatedconnected propagator S (p, p |A) we can use the expression (2.24) derived in Section 2 in the eikonal limit. Also the quantity ZW can be evaluated
in the same approximation: the result and the details of the calculations are reported in the
Appendix A. If one defines the diffusion amplitude Tf i = f |T |i by


i (p1 )k (p2 )|(S 1)|j (p1 )l (p2 )
= i(2)4 (4) (Pfin Pin )i (p1 )k (p2 )|T |j (p1 )l (p2 ),

(5.9)

where Pin = p1 + p2 is the initial total four-momentum and Pfin = p1 + p2 is the final total
four-momentum, the following results is obtained at the end [9]:


i (p1 )k (p2 )|T |j (p1 )l (p2 )



i
2 2s d 2 z ei q z [Wp1 (zt ) 1]ij [Wp2 (0) 1]kl A ,
(5.10)
ZW
q2 , is the transferred four-momentum,
where q = p1 p1 (0, 0, q ), with t = q 2 = 
and zt = (0, 0, z ) is the distance between the two trajectories in the transverse plane.
One can proceed exactly in the same way when considering the elastic scattering process
of two real (i.e., spin-1/2) quarks with initial four-momenta p1 , p2 and final fourmomenta p1 , p2 , in the limit of soft high-energy scattering:
j (p1 ) + l (p2 ) i (p1 ) + k (p2 ).

(5.11)

The following expression for the scattering matrix element is derived using the LSZ
reduction formulae and a functional integral approach [with the plane wave functions
normalized as: p (x) = u(p) exp(ipx), with u (p)u (p) = 2m ]:


i (p1 )k (p2 )|(S 1)|j (p1 )l (p2 )

E. Meggiolaro / Nuclear Physics B 602 (2001) 261288


1 
(tc) (p1 , p |A)u (p1 ) u (p )G
(tc) (p2 , p |A)u (p2 ) ,
u (p1 )G
1
2
2
ij
kl
A
2
ZW

277

(5.12)

i, j, k, l {1, . . . , Nc } are colour indices, while , , , {1, 2} are spin indices.


The expectation value O(A)A of an arbitrary functional O(A) of the gluon field A is
now defined as:






1
i
O(A)A
[dA] O(A) exp
d 4 x Fa Fa det i D m ,
ZQCD
4
(5.13)
where ZQCD is the partition function for fermion QCD:

 

 


4

ZQCD [dA][d] d exp i d x L , , A








i
4

= [dA] exp
(5.14)
d xFa Fa det i D m .
4
The determinant in Eqs. (5.13) and (5.14) comes from the integration over the fermion
degrees of freedom.
(tc) (p, p |A)u (p), one finally finds the
If one uses the result (3.18) for u (p )G
ij
following expression for the high-energy quarkquark elastic scattering amplitude in
(fermion) QCD [79]:


i (p1 )k (p2 )|T |j (p1 )l (p2 )



i
2 2s d 2 z ei q z [Wp1 (zt ) 1]ij [Wp2 (0) 1]kl A .
(5.15)
ZW
The notation is the same as for Eq. (5.10). In a perfectly analogous way, one can also derive
the high-energy scattering amplitude for an elastic process involving two partons, which
can be quarks, antiquarks or gluons. One simply has to insert in the matrix element, for
each of the two partons involved, the corresponding quantity among (2.24), (2.30), (3.18),
(3.19) and (4.11), which, as we have said before, describe the scattering amplitude of the
parton in a given external gluon field. So, for example, the amplitude for the quarkgluon
scattering
j (p1 ) + ga (p2 ) i (p1 ) + ga   (p2 )

(5.16)

has the following expression in the high-energy limit [8]:




i (p1 )ga   (p2 )|T |j (p1 )ga (p2 )


1 


(tc) (p1 , p1 |A)u (p1 )  (p2 )D (tc)
u (p1 )G

 ,a  a (p2 , p2 |A)() (p2 )
ij

)
(
ZW ZV



i

 2s d 2 z ei q z [Wp1 (zt ) 1]ij [Vp2 (0) 1]a  a A .
ZW ZV
(5.17)
(The renormalization constant ZV is defined and evaluated in Appendix A.) Let us
observe at this point that, differently from their quark counterparts (which really come
from the integration over the quark degrees of freedom in the functional integral), the

278

E. Meggiolaro / Nuclear Physics B 602 (2001) 261288





gluon matrix element ( ) (k  )D (tc)
 ,a  a (k, k |A)()(k) can only be defined as the proper
functional of A which, when inserted in the functional average  A (together with
another parton matrix element) reproduces the corresponding gluonparton scattering
amplitude in the high-energy limit. We have derived the expression (4.11) for the gluon
matrix element by resumming all diagrams of the type reported in Fig. 1(b) in the eikonal
limit. Therefore, e.g., Compton-like diagrams (and their perturbative corrections) are not
included in Eq. (5.17): indeed, one can easily convince oneself (see, for example, Ref. [4])
that these diagrams (which are present also in the Abelian case) are of order O(s 0 ) in the
high-energy limit s ; while, for example, the O(g 2 ) diagram described by Eq. (5.17)
(i.e., the one obtained taking the first term in the perturbative expansion of the gluon matrix
element and of the quark matrix element) is of order O(s). However, the expression (5.17),
by its own construction, is not able to reproduce all those diagrams where the scattering
gluon converts in a quarkantiquark pair during the diffusion process. Without resorting
to large-Nc approximations, where these diagrams are of course sub-leading corrections, a
reason for not including these diagrams in the high-energy scattering amplitude can come
if we first study soft high-energy partonparton scattering in the femto-universe, in
the spirit of Refs. [7,8]. In other words, we first consider the scattering of the partons
over the finite time interval t0 /2  t  t0 /2 (t = 0 being the nominal collision time of
the hadrons h1 + h2 h1 + h2 in the the c.m.s.) of length t0 2 fm: t0 /2 is the time
when, in an inelastic collision, the first produced hadrons appear. (The estimate t0 2
fm is discussed in Ref. [7].) As discussed in Refs. [7,8], one can assume that over that
time interval: (a) parton annihilation and production processes can be neglected (i.e., the
parton state of the hadrons does not change qualitatively in this time); (b) partons travel
in essence on straight lightlike world lines and they undergo soft elastic scattering. This
was the strategy adopted in Ref. [7] in order to study partonparton soft high-energy
scattering in QCD. Of course, free quarks and gluons do not exist in (zero-temperature)
QCD, but, assuming that t0 2 fm is nearly infinitely long on the scale of the femtouniverse, one can use the standard LSZ reduction formulae to relate the partonic S-matrix
element to an integral over the four-point function of the quark/gluon fields. After having
solved the problem of partonparton scattering, one has to fold the partonic S-matrix with
the hadronic wave functions of the appropriate resolution to get the hadronic S-matrix
elements. We want to stress that the result (5.15) for quarkquark scattering at high energies
was derived in Ref. [7] under the crucial assumption that only gluon modes up to a fixed
frequency contribute in the functional integral  A . It was argued at length in Ref. [7]
that this should be a valid approximation for the scattering of partons over the time interval
(t0 /2, t0 /2), since, for these scattering processes, the relevant scale for the frequency
of the exchanged quanta is given by a 1 , independent of s, where a is the correlation
length of the (gauge-invariant) two-point function of the gluon field-strength tensor. In the
hypothetical scattering amplitude for real quarks, all the splitting processes with long
time scales will play an important role and therefore must be included.

E. Meggiolaro / Nuclear Physics B 602 (2001) 261288

279

6. Conclusions and outlook


In this paper we have derived the same results already found in Refs. [79] in a different
and even more immediate way, i.e., by a direct resummation of perturbation theory,
in a background gluon field, in the limit of very high energy and small transferred
momentum [1316]. The approximations used in the previous sections in order to derive
the eikonal propagators (and, as a consequence, the partonparton scattering amplitudes)
in the above-mentioned limit are exactly the same adopted in Ref. [4] in order to derive the
high-energy asymptotic expressions for the elastic partonparton scattering amplitudes at
a given order in perturbation theory, using standard Feynman-diagram techniques. (These
approximations essentially consist in neglecting the recoil of the scattering partons in the
diffusion process.) Therefore, in this sense, we can claim that we have given a proof that
the nonperturbative expressions (5.10), (5.15), etc., for the high-energy elastic scattering
amplitude are a resummation of the corresponding perturbative results found in Ref. [4].
(In a previous paper [17] we gave a direct proof that this is true up to the fourth order in
the expansion in the coupling constant.)
We remind the reader that the s dependence of the scattering amplitude is not all
contained in the kinematical factor 2s in front of the integral in Eqs. (5.10), (5.15), etc. In
fact, as was first pointed out by Verlinde and Verlinde in [23], it is a singular limit to take the
Wilson lines in (5.10), (5.15), etc., exactly lightlike. It turns out that a proper regularization
of these infrared singularities (so called because they essentially come from the limit
m 0, m being the quark mass) gives rise to a log s dependence of the amplitude, as
obtained by ordinary perturbation theory [4,5] and as confirmed by the experiments on
hadronhadron scattering processes. In practice, the regularization procedure consists in
letting each Wilson line have a small timelike component (so that they coincide with the
classical trajectories for quarks with a finite mass m) and letting them end after some finite

proper time T , i.e., after some time |t| T s/2m in the c.m.s. (so that, when one takes
the limit s , the trajectories of the Wilson lines will again become infinitely long and
lightlike). We refer the reader to Refs. [23] and [1719] for a detailed discussion about this
point.
We want also to stress that the expressions (5.10), (5.15), etc., for the scattering
amplitudes are not limited by our approximations to be quenched: indeed, in the
functional average  A also the determinant of the quark matrix is included and it gives
rise to Feynman diagrams with dynamical quark loops in the perturbative expansion. So,
for example, the so-called tower diagrams for quarkquark elastic scattering, which give
amplitudes as large as s(log s)n , where n is the number of quark loops joined together
vertically, with each loop having four vertices (see, e.g., Ref. [4] and references therein),
are expected to be reproduced in the perturbative expansion of Eqs. (5.10), (5.15), etc.,
provided you integrate from t = to t = +, i.e., you take infinitely long Wilson
lines, since the time associated with the hard fluctuations in these eikonal processes is
very large. [If we truncate the Wilson lines from t = t0 to t = +t0 (in the c.m.s. of the
reaction), with t0 fixed as s , we can loose in our description all those processes in
which there is production of hard partons, whose fluctuation time is larger than t0 (in the

280

E. Meggiolaro / Nuclear Physics B 602 (2001) 261288

asymptotic limit s ). These processes must then be included in the wave functions of
the fermions at t = t0 .]
As we have said before, free asymptotic states of quarks and gluons do not exist in
(zero-temperature) QCD. Therefore, in this case, the correct procedure should be the one
suggested in Refs. [7,8], where Eqs. (5.10), (5.15), etc., are considered to be the partonic
S-matrix elements in the femto-universe (with a fixed frequency cutoff for the gluon
modes in the functional average): they must be folded with the hadronic wave functions of
the appropriate resolution to get the hadronic S-matrix elements.
Vice versa, in QED one can have the free-electron states as IN and OUT asymptotic
states: there is no complication due to confinement and a free-electron state is a well
defined asymptotic state of the theory. Therefore, the eikonal formula (5.15) in QED,
with infinitely long Wilson lines, can be really considered as the asymptotic expression
for the fermionfermion elastic scattering amplitude in the high-energy limit and it should
reproduce all perturbative results evaluated by Cheng and Wu in Ref. [4]. We want to stress
that the eikonal formula (5.15) is not identical to what, in the literature, is usually called
the eikonal amplitude of the high-energy scattering in QED (see Refs. [2022]): this
last was obtained in the so-called quenched approximation, where vacuum polarization
effects, arising from the presence of dynamical fermion loops, are neglected. It was proved
in Ref. [9] (see also Refs. [25,26]) that, when evaluating Eq. (5.15) in the quenched
approximation, one correctly reproduces the eikonal result of Refs. [2022]. However, as
we have said before, the amplitude (5.15) is not limited, in general, to be quenched and so
it is expected to be more correct and to contain more information than the eikonal result of
Refs. [2022]).
Once we have found the nonperturbative expressions (5.10), (5.15), etc., for the highenergy scattering amplitudes, the natural question which arises is: How can we evaluate
them directly? The answer to this question is highly nontrivial and it is also strictly
connected with the renormalization properties of Wilson-line operators [24,25]. Some
nonperturbative approaches for the calculation of (5.15) were proposed in Refs. [26,27]. In
particular, in Refs. [27,28] a nonperturbative numerical estimate of high-energy hadron
hadron scattering amplitudes was obtained, using Eq. (5.15) as a basic ingredient, in the
framework of the so-called stochastic vacuum model. For an alternative approach to the
problem, we refer the reader to Refs. [1719], where interesting analytic properties of
the high-energy scattering amplitude were derived, going from Minkowskian to Euclidean
spacetime, so opening the possibility of studying the high-energy scattering amplitude
using the Euclidean formulation of the theory, e.g., on the lattice. (See also Ref. [29],
where a similar analytic continuation from Minkowskian to Euclidean theory was proposed
to study the small-xBj behaviour of the structure functions of deep inelastic leptonnucleon
scattering.) The analytic continuation proposed in Refs. [1719] has been recently adopted
in Ref. [30], in order to study the high-energy scattering in N = 4 supersymmetric SU(Nc )
gauge theories (in the strong coupling, large-Nc limit) using the AdS/CFT correspondence,
and also in Ref. [31], in order to investigate instanton-induced effects in QCD highenergy scattering. In our opinion, a considerable progress could be achieved by a direct
investigation of these problems on the lattice in the near future.

E. Meggiolaro / Nuclear Physics B 602 (2001) 261288

281

Acknowledgements
I would like to thank Prof. Otto Nachtmann for his useful suggestions and comments
and also for having encouraged me to write this paper.

Appendix A. The residue at the pole of the unrenormalized eikonal propagators


A.1. The scalar propagator
Let us consider now the unrenormalized full scalar propagator (not truncated!) in the
vicinity of the pole (i.e., for p2 m2 ). We start from the scalar propagator Sij (x, y|A) in
an external gluon field A and then we take the functional average over the gluon field in
order to get the full propagator (2.2):




Sij (x, y|A) A = T [i (x)j (y)] = Sij (x y).
(A.1)
Therefore, using the definition (2.9), we find:


Sij (p, p |A) = (2)4 (4) (p p)Sij (p),

(A.2)

where Sij (p) is the unrenormalized propagator in the momentum space, defined by
Eq. (2.4). We shall evaluate the quantity (A.2) starting from the perturbative expansion
(tc)
of Sij (p, p |A) in the eikonal approximation, that we have computed in Section 2. At the
0th order we find the following expression for Sij (p, p |A)A :


 
Sij (p, p |A) A (0) = (2)4 (4) (p p)

i
.
p2 m2 + i

(A.3)

At the nth order (n  1) we find, using the corresponding expression for [Sij(tc) (p, p |A)](n)
that we have derived in Section 2 [see Eq. (2.20)]:

 
Sij (p, p |A) A (n)




i
2
2

db1+ eiqbn
db
db
b
d
n
n
n+
p m2 + i
1

(bn+ b(n1)+ ) (b2+ b1+ )


(2E)n1

 
 

 
n
1
igp An (bn ) igp A1 (b1 ) ij 
A


bi =bn ;bi =bn

i
,
2
p m2 + i

(A.4)

where q = p p. By virtue of the invariance under the Poincar group (in particular,
under spacetime translations):


An (bn )An1 (bn1 ) A1 (b1 ) A


= An (0)An1 (bn1 bn ) A1 (b1 bn ) A .
(A.5)

282

E. Meggiolaro / Nuclear Physics B 602 (2001) 261288

Therefore, making the change of variables


ci = bi bn , i = 1, . . . , n 1;

cn = b n ,

(A.6)

the integration over cn gives 2(2)4 (4) (q) and we can write Eq. (A.4) as follows:

 
Sij (p, p |A) A (n)


i

dc1+
dc
2(2)4 (4) (q)  2
(n1)+
p m2 + i
1

(c(n1)+ ) (c(n1)+ c(n2)+ ) (c2+ c1+ )


(2E)n1
 


igpn An (0) igpn1 An1 (cn1 )


 

i
1
.
igp A1 (c1 ) ij 
2
2
A
 p m + i
ci =0;
ci =0

(A.7)

In the vicinity of the pole, i.e., for p2 m2 , one has that: p 2 m2 2pq + q 2 . Moreover,
in the eikonal limit, p+ 2E, p 0 and p 0 , so that:
p 2

1
i
1

i
i
=
P
(q ),
+

2
m + i 2E q + i 2E q 2E

(A.8)

where P stands for principal-part value. While (4) (q)P(1/q) = 0, we can write,
formally:
2

(4)

1
(q ) = (4) (q)
(q)
2E
2E

+
+
1
icn+ q
(4)
dcn+ e
= (q)
dcn+ .
2E

(A.9)

(The origin of this infrared singularity is discussed in Section 6 and can be cured by
a proper regularization of the Wilson lines, as suggested in Refs. [23] and [1719].)
Substituting in Eq. (A.7) and changing again the integration variables from ci to bi =
ci + cn , i = 1, . . . , n 1 and bn = cn , we obtain:
 

Sij (p, p |A) A (n)


i
(2)4 (4) (q) 2

db1+
db
n+
p m2 + i
1

(bn+ b(n1)+ ) (b2+ b1+ )


(2E)n

 
 

 
igpn An (bn ) . . . igp1 A1 (b1 ) ij 
(A.10)
.
A


bi =bn ;bi =bn

(We have again made use of the invariance under spacetime translations in  A .) We can
parametrize the coordinates bi (i = 1, . . . , n) as in Section 2, i.e., in the form bi = b + pi
(i = 1, . . . , n), with a fixed b. In fact, using p+ 2E, p 0 and p 0 , one has that
bi = 0, bi = 0 and dbi+ = 2Edi . Eq. (A.10) becomes:

E. Meggiolaro / Nuclear Physics B 602 (2001) 261288



283

 
Sij (p, p |A) A (n)



i
(2)4 (4) (q) 2

dn (n n1 ) (2 1 )
d
1
p m2 + i
 
 


igpn An (b + pn ) . . . igp1 A1 (b + p1 ) ij .
(A.11)
A

In conclusion, summing all perturbative orders [Eqs. (A.3) and (A.11)], we find the
following expression for Sij (p, p |A)A :


Sij (p, p |A) A


i
[Wp (b)]ij A
(2)4 (4) (q) 2
2
p m + i

ij 
i
= (2)4 (4) (q) 2
(A.12)

Tr[Wp (b)] A ,
2
p m + i Nc
where Wp (b) has been defined in Eq. (2.25). (In the last passage we have used the fact that
the vacuum is gauge-invariant.) By virtue of Eq. (A.2), we derive the following expression
for Sij (p) in the vicinity of the pole:
Sij (p)

p 2 m2


ij 
i
Tr[W
(b)]
.

p
A
p2 m2 + i Nc

(A.13)

Moreover, from the definition of the physical mass m and of the residue ZW at the pole
for p2 m2 of the unrenormalized quark propagator, we must have [see Eq. (2.3)]:
Sij (p)

p 2 m2

p2

iZW ij
.
m2 + i

(A.14)

From the comparison of Eqs. (A.13) and (A.14), we derive the following expression for
the residue ZW at the pole:


1 
1 
ZW =
(A.15)
Tr[Wp (b)] A =
Tr[Wp (0)] A ,
Nc
Nc
where we have again made use of the invariance under spacetime translations.
A.2. The fermion propagator
Let us consider now the unrenormalized full fermion propagator (not truncated!) in the
vicinity of the pole (i.e., for p2 m2 ). We start from the fermion propagator Gij (x, y|A)
in an external gluon field A and then we take the functional average over the gluon field
in order to get the full propagator (3.2):




Gij (x, y|A) A = T [i (x) j (y)] = Gij (x y).
(A.16)
Going to the momentum representation [see Eq. (2.9)], we obtain:


ij (p, p |A) = (2)4 (4) (p p)G
ij (p),
G
A

(A.17)

ij (p) is the unrenormalized propagator in the momentum space, defined by


where G
Eq. (3.4). More precisely, we shall evaluate the following quantity:


ij (p, p |A) u (p) = (2)4 (4) (p p)u (p)G
ij (p)u (p),
u (p ) G
(A.18)
A

284

E. Meggiolaro / Nuclear Physics B 602 (2001) 261288

(tc) (p, p |A) in the eikonal approximation,


starting from the perturbative expansion of G
ij
that we have computed in Section 3. At the 0th order we find the following expression for
the quantity (A.18):




ij (p, p |A) u (p)
u (p ) G
A
(0)
i
u (p)
p m + i
i
= (2)4 (4) (p p) 4m2 2
.
p m2 + i

= (2)4 (4) (p p)u (p )

(A.19)

(p, p |A)](n)
At the nth order (n  1) we find, using the corresponding expression for [G
ij
that we have derived in Section 3 [see Eq. (3.12)]:




ij (p, p |A) u (p)
u (p ) G
A
(n)
 4

4
i
d qn
d q1
= 2

(2)4 (4) (q q1 qn )
p m2 + i
(2)4
(2)4
i
Q(ferm)

,ij (q1 , . . . , qn )
(p + q1 + + qn1 )2 m2 + i
i
i
2
,

(A.20)
2
2
(p + q1 ) m + i p m2 + i
(tc)

where q = p p and Q,ij (q1 , . . . , qn ) is given by:


(ferm)

(ferm)

Q,ij (q1 , . . . , qn )



u (p )(p  + m) ig n A n (qn ) + im(2)4 (4) (qn ) 1
(p + q1 + + qn1 + m) (p + q1 + m)




ig 1 A 1 (q1 ) + im(2)4 (4) (q1 ) 1 ij (p + m)u (p)


A
 (ferm)

2
= 4m N,ij (q1 , . . . , qn ) A ,

(A.21)

(ferm)

where N,ij (q1 , . . . , qn ) has been defined in Eq. (3.13). By using the result (3.16), we
find that the expression (A.20) simplifies as follows:





 
ij (p, p |A) u (p)
u (p ) G
4m2 Sij (p, p |A) A (n)
A
(n)

i
4m2 (2)4 (4) (q) 2
d1
p m2 + i

dn (n n1 ) (2 1 )
 
 


igpn An (b + pn ) igp1 A1 (b + p1 ) ij .
(A.22)
A

In conclusion, summing all perturbative orders [Eqs. (A.19) and (A.22)], we obtain the
following expression for the quantity (A.18), evaluated in the vicinity of the pole:

E. Meggiolaro / Nuclear Physics B 602 (2001) 261288



ij (p, p |A) u (p)
u (p ) G
A


4m2 Sij (p, p |A) A
4m2 (2)4 (4) (q)

p2


ij 
i

Tr[Wp (b)] A ,
2
m + i Nc

285

(A.23)

where Wp (b) has been defined in Eq. (2.25). (In the last passage we have used the fact that
the vacuum is gauge-invariant.) By virtue of Eq. (A.18), we derive the following expression
ij (p)u (p) in the vicinity of the pole:
for u (p)G
ij (p)u (p)
u (p)G

p 2 m2

4m2

p2


ij 
i

Tr[Wp (b)] A .
2
m + i Nc

(A.24)

Moreover, from the definition of the physical mass m and of the residue ZW at the pole
for p2 m2 of the unrenormalized quark propagator, we must have [see Eq. (3.3)]:
ij (p)
G

p 2 m2

iZW ij (p + m)
iZW ij
= 2
,
p m + i
p m2 + i

(A.25)

and, therefore:
ij (p)u (p)
u (p)G

p 2 m2

4m2

iZW ij
.
p2 m2 + i

(A.26)

From the comparison of Eqs. (A.24) and (A.26), we derive the following expression for
the residue ZW at the pole:


1 
1 
(A.27)
Tr[Wp (b)] A =
Tr[Wp (0)] A ,
Nc
Nc
where we have made use of the invariance under spacetime translations. This expression
is formally identical to the one we have obtained for the scalar case: however, we must
remember that now  A has to be intended as the functional average over the gluon field
in the theory with fermions. The same result (A.27) was also derived in Ref. [7] using
a different method.
ZW =

A.3. The gluon propagator


Let us consider, finally, the unrenormalized full gluon propagator (not truncated!) in the

vicinity of the pole (i.e., for k 2 0). We start from the gluon propagator Da a (x, y|A)
in an external gluon field Ab and then we take the functional average over the gluon field
Ab in order to get the full propagator:
 a a

  


D (x, y|A) A = T Aa (x)Aa (y) = Da a (x y).
(A.28)
Going to the momentum representation [see Eq. (2.9)], we obtain:

 a a

D  (k, k  |A) A = (2)4 (4) (k  k)D a a (k),

(A.29)

a a


where D

(k) is the full propagator in the momentum space:





D a a (k) d 4 z eikz Da a (z).

(A.30)

286

E. Meggiolaro / Nuclear Physics B 602 (2001) 261288

More precisely, we shall evaluate the following quantity:


 





( ) (k  ) D a a (k, k  |A) A ()(k) = (2)4 (4) (k  k)( ) (k)D a a (k)()(k), (A.31)


(tc)
starting from the perturbative expansion of D  ,a  a (k, k  |A) in the eikonal approximation,
that we have computed in Section 4. At the 0th order we find the following expression for
the quantity (A.31):

 



( ) (k  ) D a a (k, k  |A) A () (k) (0) = (2)4 (4) (k  k) a  a

i
.
(A.32)
+ i
At the nth order (n  1) we find, using the corresponding expression for the truncated

connected propagator [D a a (k, k  |A)](n) that we have derived in Section 4 [see Eqs. (4.5)
(4.7)]:
    a  a


( ) (k ) D  (k, k  |A) A () (k) (n)


i
d 4 qn
d 4 q1
 2

(2)4 (4)(q q1 qn )
4
k + i
(2)
(2)4
 
 


ig2k n An (qn ) ig2k 1 A1 (q1 ) a  a
k2

i
i
i

(k + q1 + + qn1 )2 + i
(k + q1 )2 + i k 2 + i

(A.33)

where q k  k is the transferred momentum. This expression is perfectly analogous to


the corresponding expression obtained in Section A.1 for the scalar case. We can proceed
as in that case to simplify Eq. (A.33) as follows:


    a  a
( ) (k ) D  (k, k  |A) A () (k) (n)


i
4 (4)
(2) (q) 2
d1 dn (n n1 ) (2 1 )
k + i
 
 


igk n An (b + kn ) igk 1 A1 (b + k1 ) a  a .
(A.34)
A

In conclusion, summing all perturbative orders [Eqs. (A.32) and (A.34)], we find the
following expression for the quantity (A.31), evaluated in the vicinity of the pole:
 


( ) (k  ) D a a (k, k  |A) A ()(k)

a  a 
i
2
Tr[Vk (b)] A ,
(2)4 (4) (q) 2
(A.35)
k + i Nc 1
where Vk (b) has been defined in Eq. (4.12) and we have made use of the gauge-invariance
of the vacuum (i.e., of the functional average  A ). By virtue of Eq. (A.31), we derive



the following expression for ( ) (k)D a a (k)()(k) in the vicinity of the pole:


( ) (k)D a a (k)()(k) 

k 2 0

k2


a  a 
i
2
Tr[Vk (b)] A .
+ i Nc 1

(A.36)

Moreover, from the definition of the residue ZV at the pole for k 2 0 of the
unrenormalized full gluon propagator, we must have that:

E. Meggiolaro / Nuclear Physics B 602 (2001) 261288

iZV g a  a
,
k 2 + i
k 2 0


D a a (k)

287

(A.37)

and, therefore:


( ) (k)D a a (k)()(k)  a  a

k 2 0

iZV
.
k 2 + i

(A.38)

From the comparison of Eqs. (A.36) and (A.38), we derive the following expression for
the residue ZV at the pole:
ZV =

Nc2



1 
1 
Tr[Vk (b)] A = 2
Tr[Vk (0)] A ,
1
Nc 1

where we have made use of the invariance under spacetime translations.

References
[1] T. Regge, Nuovo Cimento 14 (1959) 951;
T. Regge, Nuovo Cimento 18 (1960) 947.
[2] N. Byers, C.N. Yang, Phys. Rev. 142 (1966) 976.
[3] F.E. Low, Phys. Rev. D 12 (1975) 163;
S. Nussinov, Phys. Rev. Lett. 34 (1975) 1286.
[4] H. Cheng, T.T. Wu, Expanding Protons: Scattering at High Energies, MIT Press, Cambridge,
MA, 1987.
[5] L.N. Lipatov, in: A.H. Mueller (Ed.), Review in Perturbative QCD, World Scientific, Singapore,
1989, and references therein.
[6] P.V. Landshoff, O. Nachtmann, Z. Phys. C 35 (1987) 405.
[7] O. Nachtmann, Ann. Phys. 209 (1991) 436.
[8] O. Nachtmann, in: H. Latal, W. Schweiger (Eds.), Perturbative and Nonperturbative aspects of
Quantum Field Theory, Springer-Verlag, Berlin, 1997.
[9] E. Meggiolaro, Phys. Rev. D 53 (1996) 3835.
[10] E.S. Fradkin, in: Proceedings of the International Winter School on Theoretical Physics at JINR,
Dubna, 1964;
E.S. Fradkin, Acta Phys. Hung. XIX (1964) 175.
[11] H. Lehmann, K. Symanzik, W. Zimmermann, Nuovo Cimento 1 (1955) 205.
[12] M. Fabbrichesi, R. Pettorino, G. Veneziano, G.A. Vilkovisky, Nucl. Phys. B 419 (1994) 147.
[13] I.I. Balitsky, V.M. Braun, Nucl. Phys. B 311 (1989) 541.
[14] I.I. Balitsky, V.M. Braun, Nucl. Phys. B 361 (1991) 93.
[15] I. Balitsky, Nucl. Phys. B 463 (1996) 99.
[16] I. Balitsky, High-Energy QCD and Wilson Lines, hep-ph/0101042.
[17] E. Meggiolaro, Z. Phys. C 76 (1997) 523.
[18] E. Meggiolaro, Eur. Phys. J. C 4 (1998) 101.
[19] E. Meggiolaro, Nucl. Phys. B Proc. Suppl. 64 (1998) 191.
[20] H. Cheng, T.T. Wu, Phys. Rev. Lett. 22 (1969) 666.
[21] H. Abarbanel, C. Itzykson, Phys. Rev. Lett. 23 (1969) 53.
[22] R. Jackiw, D. Kabat, M. Ortiz, Phys. Lett. B 277 (1992) 148.
[23] H. Verlinde, E. Verlinde, Princeton University Report No. PUPT-1319 (revised 1993), hepth/9302104.
[24] I.Ya. Arefeva, Phys. Lett. B 93 (1980) 347.
[25] G.P. Korchemsky, Phys. Lett. B 325 (1994) 459;
I.A. Korchemskaya, G.P. Korchemsky, Nucl. Phys. B 437 (1995) 127.

288

E. Meggiolaro / Nuclear Physics B 602 (2001) 261288

[26] I.Ya. Arefeva, Phys. Lett. B 325 (1994) 171;


I.Ya. Arefeva, Phys. Lett. B 328 (1994) 411.
[27] H.G. Dosch, E. Ferreira, A. Krmer, Phys. Rev. D 50 (1994) 1992.
[28] E.R. Berger, O. Nachtmann, Eur. Phys. J. C 7 (1999) 459.
[29] A. Hebecker, E. Meggiolaro, O. Nachtmann, Nucl. Phys. B 571 (2000) 26.
[30] R.A. Janik, R. Peschanski, Nucl. Phys. B 565 (2000) 193;
R.A. Janik, R. Peschanski, Nucl. Phys. B 586 (2000) 163.
[31] E. Shuryak, I. Zahed, Phys. Rev. D 62 (2000) 085014.

Nuclear Physics B 602 (2001) 289306


www.elsevier.nl/locate/npe

Rare decays B Xs + and Bs + in


technicolor with scalars
Zhaohua Xiong a,b,c , Jin Min Yang b
a CCAST (World Laboratory), P.O.Box 8730, Beijing 100080, PR China
b Institute of Theoretical Physics, Academia Sinica, Beijing 100080, PR China
c Institute of High Energy Physics, Academia Sinica, Beijing 100039, PR China

Received 19 December 2000; accepted 12 March 2001

Abstract
We examine the rare decays B Xs + and Bs + in the framework of technicolor
with scalars. The contributions from both the neutral and charged scalars predicted in this model are
evaluated. We find that the branching ratios could be enhanced over the standard model predictions by
a couple of orders of magnitude in some part of parameter space. The forwardbackward asymmetry
and the distributions of differential branching ratios are also found to differ significantly from the
standard model results. Such large new physics effects might be observable in the new generation of
B experiments. 2001 Elsevier Science B.V. All rights reserved.
PACS: 12.60.NZ; 13.25.Hw
Keywords: Technicolor with scalars; Decay B Xs + ; Decay Bs + ; Forwardbackward
asymmetry

1. Introduction
One intriguing puzzle in particle physics is the regular pattern of three lepton and
quark families. The existence of families gives rise to many parameters of the standard
model (SM). Flavor-changing neutral-currents (FCNC) induced B-meson rare decays
provide an ideal opportunity for extracting information about the fundamental parameters
of the SM, such as the CabibboKobayashiMaskawa (CKM) matrix elements, and
for testing the SM predictions at loop level and probing possible new physics. The
experimental discovery of the inclusive and exclusive rare decays b Xs and B K
[1] stimulated the study of radiative rare B-meson decays with a new momentum.
The inclusive decays B Xs + ( = e, ) have been well studied in the frameworks
of minimal supersymmetric model [2], the two Higgs doublet model (2HDM) [35] and
E-mail address: xiongzh@itp.ac.cn (Z. Xiong).
0550-3213/01/$ see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 0 9 2 - X

290

Z. Xiong, J.M. Yang / Nuclear Physics B 602 (2001) 289306

the technicolor models [6,7]. It was shown that the matrix elements are strongly suppressed
by a factor m /mW and the contributions from exchanging neural scalars can be safely
neglected. However, the situation is different in the case of = . The branching ratio
Br(Bs + )  8 107 [8] in the SM is large enough to be observable in future Bfactories. The contributions from neutral scalars exchange to B Xs + may no longer
be negligible and thus also have to be examined.
On the other hand, among rare B-meson decays, Bs + is of special interest due
to its relative cleanliness and sensitivity to models beyond the SM [9,10]. We emphasize
that when photon is emitted in additional to the lepton pair, no helicity suppression exists,
and large branching ratio is expected. As in the decay of Bs + , we could expect
that for Bs + the contributions from exchanging scalars could be sizable.
In this work, we will study the inclusive and exclusive decays B Xs + and
Bs + in technicolor model with scalar. Since several scalars are predicted in this
model, they are expected to cause sizable effects in these decays. Taking into account the
contributions from both the neutral and charged scalars predicted in this model, we will
evaluate the branching ratios, the forwardbackward asymmetry as well as the distributions
of differential branching ratios. The paper is organized as follows. Section 2 is a brief
review of the model. The detailed calculations of the contributions from the scalars are
presented in Sections 3 and 4 for the decays B Xs + and Bs + , respectively.
Finally, in Section 5 we give some numerical results and conclusions.

2. Technicolor with scalars


In the technicolor model with scalars, the rare decay processes we will study receive
contributions not only from the SM particles but also from charged and neutral physical
scalars predicted in such a technicolor model. In this section we will briefly discuss the
model and give the relevant Lagrangian which are needed in our calculation. More details
of the model have been described in Refs. [11,12].
The gauge structure of the technicolor model with scalars is simply the direct product of
the technicolor and standard model gauge groups: SU(N)T C SU(3)C SU(2)W U (1)Y
[12]. The ordinary techni-singlet fermions are exactly as those in the SM. The technicolor
sector consists of two techniflavors p and m that also transform under SU(2)W . In addition
to the above particle spectrum, there exists a scalar doublet to which both the ordinary
fermions and technifermions are coupled. Unlike the SM Higgs doublet, does not cause
electroweak symmetry breaking but obtains a non-zero effective vacuum expectation value
(VEV) when technicolor breaks the symmetry.
Writing the matrix form of the scalar doublet as
 0

( + f )

+
=
(2.1)

,

2
we then have the kinetic terms for the scalar fields given by


 1

1
1
LK.E. = + f 2 Tr D D + ( + f )2 Tr D D . (2.2)
2
4
4

Z. Xiong, J.M. Yang / Nuclear Physics B 602 (2001) 289306

291

Here the non-linear representation = exp(2i/f ) and = exp(2i /f ) are adopted


for the technipion fields. is an isosinglet scalar field, f and f are the technipion decay
constant and the effective VEV, respectively. The covariant derivative is defined as D =
3
a
igWa 2 + ig B 2 with a /2 (a = 1, 2, 3) being the SU(2) generators, and
Wa (B ) denote the SU(2) (U (1)) vector fields with the gauge coupling constant g (g ).
The definition of D is analogous to that of D .
The mixing between and gives
f + f
a = 
,
f 2 + f 2
f + f
p = 
,
f 2 + f 2

(2.3)
(2.4)

with a becoming the longitudinal component of the W and Z, and p remaining in the
low-energy theory as an isotriplet of physical scalars. From Eq. (2.2) one can obtain the
correct gauge boson masses providing that f 2 + f 2 = V 2 with the electroweak scale V =
246 GeV.
Additionally, the contributions to scalar potential generated by the technicolor interactions should be included in this model. The simplest term one can construct is
 
 
h+ 0
3
+ h.c.,
LT = c1 4f Tr
(2.5)
0 h
where c1 is a coefficient of order unity, h+ and h are the Yukawa couplings of scalars to
p and m. From Eq. (2.5) the mass of the charged scalar at lowest order is obtained as

m2p = 2 2 (4f/f )v 2 h
(2.6)
with h = (h+ + h )/2.
In general, f and f depend on h+ , h , M and , where M is the mass of the scalar
doublet , and is 4 coupling. Two limits of the model have been studied previously
in the literatures: (i) the limit in which is small and can be neglected [11], and (ii)
the limit in which M is small and can be neglected [12]. When the largest Coleman
Weinberg corrections for the field are included in the effective chiral Lagrangian [12],
 and coupling .
In this case, one obtains
M , are replaced by the shifted scalar mass M
the constraint

2 f + f 3 = 8 2 c1 hf 3
M
2
and the isoscalar mass as
  4

4
2 + 2 6 mt
f 2
+
Nh
m2 = M
f
3 2

(2.7)

(2.8)

in limit (i), and


  4

3 2
1
mt
4

+
Nh
m2 = f
6
f 2
2
4 2
f

(2.9)

292

Z. Xiong, J.M. Yang / Nuclear Physics B 602 (2001) 289306

in limit (ii). The advantage of working in these two limits is that at the lowest order the
phenomenology depends on h, not on the difference of h+ and h , and can be described
h) in limit (ii). In this paper, we will work in the
 , h) in limit (i) and (,
in terms of (M
unitary gauge, where the particle spectrum consists of p , and the massive weak gauge
bosons. We choose two parameters (f/f , mp ) in both limits of the model, and assume
N = 4, c1 = 1 in numerical calculations.
The interactions relevant to our calculations can be extracted from Eqs. (2.2) and (2.5)
and are given by
 
 
f
f gmW
gmZ
+

W
W
+
Z Z
L=


f
2
V cos W
 

f g +
W p + W+ p

V 2
 
  gm2

ig
V
V
p
+
U + mD DD

p p +
mU U
+


f 2mW
f 2mW

ig cos 2W
ig + 0
W p p + W+ p p0 +
Z p+ p
+
2
2 cos W
 

f
ig 0
5D
eA p+ p +
p mU U 5 U mD D

f 2mW
 

ig
f
i [(mU mD ) (mU + mD )5 ]Vij Dj
+U

f 2 2 mW p

i Vij [(mU mD ) + (mU + mD )5 ]Uj ,
p D
(2.10)
where U, D and mU , mD represent the column vector and the diagonal mass matrix for
up- and down-quarks, respectively. p stands for the scalar field and Vij are the elements of
the CKM matrix. The physical scalar-lepton couplings can be read off from the expression
above by replacing (U, D) with the corresponding lepton fields, replacing quark mass
matrices with the corresponding diagonal lepton mass matrices, and setting Vij = 1.
3. B Xs + in technicolor model with scalars
It is well known that inclusive decay rates of heavy hadrons can be calculated in heavy
quark effective theory [13], and the leading terms in 1/mQ expansion turn out to be
the decay of a free heavy quark and corrections stem from the order 1/m2Q [14]. In the
technicolor model with scalars, the short distance contribution to b s + decay can
be computed in the framework of the QCD corrected effective weak Hamiltonian, obtained
by integrating out heavy particles, i.e., top quark, scalar , p and W , Z bosons

10


4GF

Heff = Vt b Vt s
(3.1)
Ci ()Oi () + CQi ()Qi () ,
2
i=1
where Oi are the same as these given in Ref. [3]. The additional operators Qi [4] are due
to the neutral scalars exchange diagrams, which give considerable contributions in the case

Z. Xiong, J.M. Yang / Nuclear Physics B 602 (2001) 289306

293

Fig. 1. Feynman diagrams for the charged scalar contributions in technicolor with scalars.

that the final lepton pair is + . Here we only present the explicit expressions of the
operators governing B Xs + . They read


e
mb s Rb F ,
O7 =
16 2

e 
O8 =
s Lb ( ),
2
16

e 
s Lb ( 5 ),
O9 =
2
16
e2
Q1 =
(s Rb )( ),
16 2
e2
Q2 =
(3.2)
(s Rb )( 5 ),
16 2
where L, R = (1 5 )/2, is the SU(3) color index and F the field strength tensor of
the electromagnetic interaction.
In general in theories beyond the SM there will be additional contributions, which are
characterized by the values of the coefficients Ci and CQj at the perturbative scale mW .
Using the Feynman rules presented in the preceding section, we can calculate the additional
contributions arising from the scalars , p0 and p . At the scale of mW , the Feynman
diagrams for the charged scalar contributions are depicted in Fig. 1, while Fig. 2 shows the
additional contributions from the neutral scalars.
The contributions of Fig. 1 to the Wilson coefficients at leading order read 1
 2
f
C7 (mW )T C =
H1 (xp ),
f
1 The contributions of take a similar form from those contributions of the color-singlet charged pseudop
Goldstone boson in the one-generation technicolor model [15]. The typical difference is the factor f/f in these

new contributions.

294

Z. Xiong, J.M. Yang / Nuclear Physics B 602 (2001) 289306

Fig. 2. Feynman diagrams for the additional contributions from the neutral scalars in technicolor with
scalars.

C8 (mW )T C =

C9 (mW )T C =

f
f
f
f

2
2

4 sin2 W 1
sin2 W
xW
sin2 W

[xW H2 (xp ) xp H3 (xp )],

H2 (xp ),

(3.3)

where xi = m2t /m2i , W is the Weinberg angle and the functions Hi can be expressed as


x
22x 2 53x + 25 3x 2 8x + 4

ln x ,
H1 (x) =
12(x 1)3
6
x1


1
x
H2 (x) =
1 +
ln x ,
8(x 1)
x1


47x 2 79x + 38 3x 3 6x + 4
1
H3 (x) =
(3.4)

ln
x
.
6
x1
18(x 1)3
Theoretical calculations show that the contributions of Fig. 2 are significant only for
large f/f . Keeping only the leading terms in large f/f limit, the CQj (mW ) induced
from these diagrams are given by
 4
mb m
xW
f
CQ1 (mW ) =
H4 (xp ),
f
m2 4 sin2 W
 4
mb m
xW
f
CQ2 (mW ) =
(3.5)
H5 (xp ),

2
f
mp 4 sin2 W
with


 2
4x 7x + 1 x 2 2x
1

ln
x
,
2
x 1
2(x 1)2


x
1
x +1

ln x .
H5 (x) =
(x 1)
2
x 1
H4 (x) =

(3.6)

Z. Xiong, J.M. Yang / Nuclear Physics B 602 (2001) 289306

295

It is noticeable that the contributions of Fig. 2 are proportional to (f/f )4 , while those of
Fig. 1 proportional to (f/f )2 . So for a sufficiently large f/f , the contributions of neutral
scalars in Fig. 2 are relatively enhanced and become comparable with those from charged
scalars in Fig. 1.
Neglecting the strange quark mass, the effective Hamiltonian (3.1) leads to the following
matrix element for the inclusive b s + decay,

em GF
mb
M=
Vt b Vts 2C7eff 2 s i p (1 + 5 )b
p
2 2
+ C8eff s (1 5 )b + C9 s (1 5 )b 5

+ CQ1 s (1 + 5 )b + CQ2 s(1 + 5 )b 5 .

(3.7)

The Wilson coefficients Ci , CQj are to be evaluated from mW down to the lower scale
of about mb by using the renormalization group equation. When evolving down to b
quark scale, the operators O1,2 and Q3 can mix with Oi , (i = 7, 8); however, they can
be included in an effective O7,8 because of their same structures contributing to the
b s + matrix element. At leading order, the Wilson coefficients are [35]





58  10/23
29  28/23
eff
16/23
1 +
1

C7 (mb ) =
C7 (mW )
135
189

C2 (mW ) 0.012CQ3 (mW ) ,
(3.8)




4  11/23
4
8
C8eff (mb ) = C8 (mW ) +
1 +

1 29/23
s (mb ) 33
87

 


mVi (Vi + )
m2c p2
3
,
+ g
2
em
m2b m2b
m2 p2 imVi Vi
Vi = , ,... Vi

3C1 (mb ) + C2 (mb ) ,


(3.9)
C9 (mb ) = C9 (mW ),
CQi (mb ) =

Q /0

CQi (mW ).

(3.10)
(3.11)

Here p is the momentum transfer, and

mb e 2
(3.12)
CQ1 (mW ) + CQ2 (mW ) ,
2
m g
where Q = 4 is the anomalous dimension of s Rb, 0 = 11 2nf /3, = s (mb )/
s (mW ), C2 (mW ) = 1 and C1,2 (mb ) = (6/23 12/23)/2. g(m2c /m2b , s) in Eq. (3.9)
arises from the one-loop matrix elements of the four-quark operators, and



8
16x 4
2x 
4x 1/2
4
+

1+
1

g(x, y) = ln x +
9
27
9y
9
y 
y 

ln Z(x, y) i, for 4x/y < 1,
2 arctan 1
(3.13)
, for 4x/y > 1,
4x/y1
CQ3 (mW ) =

where

296

Z. Xiong, J.M. Yang / Nuclear Physics B 602 (2001) 289306

Z(x, y) =

1+
1




4x
y

4x
y

(3.14)

The second term in brace in Eq. (3.9) estimates the long-distance contribution from the
intermediate , , . . . [3]. The phenomenological parameter is taken as 2.3 [16] in our
numeral calculations.
The formula of invariant dilepton mass distribution has been derived in [4], which is
given by
d (B Xs + )
ds
= Br(B Xc )
with



1/2

2  V V 2
em
 t b t s  f 1 (mc /mb )(1 s)2 1 4r
D(s)
s
4 2  Vcb 








2
2
2
2r
2r
1+
+ C8eff  1 +
(1 + 2s)
D(s) = 4C7eff  1 +
s
s
s






2r
2r
+ |C9 |2 1 8r + 2s +
+ 12 Re C7eff C8eff 1 +
s
s


3
3

+ |CQ1 |2 (s 4r) + |CQ2 |2 s + 6 Re C9 CQ


r 1/2.
2
2
2

(3.15)

(3.16)

Here s = p2 /m2b , r = m2 /m2b . Function f (x) = 1 8x 2 + 8x 6 x 8 24x 4 ln x is the


phase-space factor.
The angular information and the forwardbackward asymmetry are also sensitive to the
details of the new physics. Defining the forwardbackward asymmetry as
1
0
2
2
0 d cos (d /ds d cos ) 1 d cos (d /ds d cos )
,
AFB (s) =  1
(3.17)
0
2
2
0 d cos (d /ds d cos ) + 1 d cos (d /ds d cos )
where is the angle between the momentum of B-meson and + in the center of mass
frame of the dilepton, we obtain
AFB =

6(1 4r/s)1/2
D(s)


1/2
1/2
r + C8eff CQ
r
Re 2C7eff C9 + C8eff C9 s + 2C7eff CQ
.
2
1

(3.18)

4. Bs + in technicolor model with scalars


Now let us turn to rare radiative decay Bs + . The exclusive decay can be
obtained from the inclusive decay b s + , and further, from b s + . To achieve
this, it is necessary to attach photon to any charged internal and external lines in the
Feynman diagrams of b s + . As pointed out in Ref. [9], contributions coming from
the attachment of photon to any charged internal line are strongly suppressed and we can

Z. Xiong, J.M. Yang / Nuclear Physics B 602 (2001) 289306

297

neglect them safely. However, since the mass of -lepton is not much smaller than that of
Bs -meson, in Bs + decay, the contributions of the diagrams with photon radiating
from final leptons are comparable with those from initial quarks. When a photon is attached
to the initial quark lines, the corresponding matrix element for the Bs + decay can
be written as
3/2




em GF
M1 =
Vt b Vts A D p q + iB D (pq) (D p)q
2





+ C D p q + iD D (pq) (D p)q 5 ,
(4.1)
where



 2
mb  2 
eff
p

2C
p
g
g
C
,
7 2 2
8 1
p
m2Bs


 2
mb  2 
1
eff
B = 2 C8 f1 p 2C7 2 f2 p ,
p
m Bs


C9
C = 2 g1 p2 ,
m Bs
C9  
D = 2 f1 p2 .
m Bs
A=

(4.2)

In obtaining Eq. (4.1) we have used


 
e
D p q g1 p2
2
m Bs

 

i D (pq) (D p)q f1 p2 ,
 
e
 |s i p (1 5 )b|Bs  = 2 D p q g2 p2
m Bs

 

i D (pq) (D p)q f2 p2 ,
 |s (1 5 )|Bs  =

(4.3)

(4.4)

and
 |s (1 5 )|Bs  = 0.

(4.5)

Here D and q are the four vector polarization and momentum of photon, respectively;
gi , fi are form factors [8,18]. Eq. (4.5) can be obtained by multiplying p in both sides
of Eq. (4.4) and using the equations of motion. From Eq. (4.5) one can see that the neutral
scalars do not contribute to the matrix element M1 .
When a photon is radiated from the final -leptons, the situation is different. Using the
expressions
0|s b|Bs  = 0,
0|s (1 + 5 )b|Bs  = 0,
0|s 5 |Bs  = ifBs PBs

(4.6)

and the conservation of the vector current, one finds that only the operators Q1,2 and O9
give contribution to this Bremsstrahlung part. The corresponding matrix is given by [10]

298

Z. Xiong, J.M. Yang / Nuclear Physics B 602 (2001) 289306


 

3/2
m2Bs
em GF
D/P
/ Bs P/ Bs D/

M2 =
CQ2
C9 +
Vt b Vt s i2m fBs
5
2m mb
2p1 q 2p2 q
2



m2Bs
1
1
+
+
D/
CQ1 2m
2m mb
2p1 q 2p2 q


 
D/P
/ Bs P/ Bs D/
+

5 .
2p1 q 2p2 q

(4.7)

Here PBs , fBs are the momentum and the decay constant of the Bs meson.
Finally, the total matrix element for the Bs + decay is obtained as a sum of
the M1 and M2 . After summing over the spins of the -leptons and polarization of the
photon, we get the square of the matrix element as


|M|2 = |M1 |2 + |M2 |2 + 2 Re M1 M2
(4.8)
with 2
2 
 3/2

 em GF






Vt b Vt s  |A|2 + |B|2 p2 (p1 q)2 + (p2 q)2 + 2m2 (pq)2
|M1 | = 4
2





+ |C|2 + |D|2 p2 (p1 q)2 + (p2 q)2 2m2 (pq)2


 
+ 2 Re B C + A D p2 (p1 q)2 (p2 q)2 ,
(4.9)
2


2

 3/2
m2Bs CQ2 

 2
 em GF




2 


Vt b Vt s  m fBs (pq) C9 +
2 Re M1 M2 = 16
2m mb 
2


(p1 q p2 q)
(p1 q + p2 q)
Re(D)
Re(A)
(p1 q)(p2 q)
(p1 q)(p2 q)

 2

2
 mBs CQ1  3m2Bs + 2m2 5(pq)
2p

+ Re(B)

2m mb 
(p1 q)(p2 q)
(pq)2

 2

 mBs CQ1  (p1 q p2 q)
2p2


(1 +
+ Re(C)
) ,
2m mb  (p1 q)(p2 q)
(pq)2

(4.10)


2

 3/2
m2Bs CQ2 2
 2 2 
 em GF


|M2 | = 8
Vt b Vt s  m fBs C9 +
2m mb 
2

 2 2
m mBs (pq 2 )
m2Bs p2 + 2(pq)2

(p1 q)2 (p2 q)2


(pq )(p2 q)
2  2 2
 2
 mB CQ1  m (mBs 4m2 )(pq)2

 s
2m mb 
(p1 q)2 (p2 q)2
2

2 There are some errors in Eqs. (15) and (17) in Ref. [10]. We believe the expressions of C H (m ) and
W
Q1

H (m ) are not correct in Two-Higgs-Double model I.


CQ
W
2

Z. Xiong, J.M. Yang / Nuclear Physics B 602 (2001) 289306

(m2Bs 4m2 )p2 + 2(pq)2

(p1 q)(p2 q)


.

299

(4.11)

Here p1 , p2 are momenta of the final -leptons. It is obvious that the quantity |M|2
depends only on the scalar products of the momenta of the external particles.
In the rest frame of the Bs , the photon energy E and the lepton energy E1 are restricted
by
0  E 

m2Bs 4m2

,
2mBs

E
m Bs E
4m2

1 2
2
2
mBs 2mBs E

E
m Bs E
4m2
+
 E1 
1 2
.
2
2
mB 2mBs E

(4.12)

However, in |M2 |2 it appears an infrared divergence, which originates in the Bremsstrahlung processes when photon is soft and in this case, the Bs + can not be
distinguished from Bs + . Therefore, both processes must be considered together
in order to cancel the infrared divergence. Taking the fact that the infrared singular
terms in |M2 |2 exactly cancel the O(em ) virtual correction in Bs + amplitude in
account [9], we follow Ref. [9] and consider the photon in Bs + as a hard photon
and impose a cut on the photon energy E , which correspond to the radiated photon can
be detected in the experiments. This cut requires E  mBs /2 with = 0.02.
After integrating over the phase space and the lepton energy E1 , we express the decay
rate as

 3/2
2 5  2 1
 em GF
 m Bs m Bs

4r  2

3

=
(1 s ) d s 1
Vt b Vt s 
|A| + |B|2 (s + 2r )
3
12
s
(2)
2 2
4r


 1
m2Bs CQ2  


r
(1 s )2 d s Re(A) ln z
+ |C| + |D| (s 4r ) 2fBs C9 +
2m mb 


4r


 2
 1


 mBs CQ1 
4r


2fBs 
r
(1 s )d s Re(B) (1 + 4r + 5s ) ln z + s 1
2m mb 
s
4r

4fB2s
m2Bs



 1 


m2Bs CQ2 2 

4
r

2)
4r
2
s
C9 +
 r
d
s

1
+
s

+
1

ln
z

+

2m mb 
1 s
1 s
s



 CQ 2
+ fB2s  1 
mb

4r

1
d s
4r



2 10r + 8r 2
ln z
1 8r + s
1 s

300

Z. Xiong, J.M. Yang / Nuclear Physics B 602 (2001) 289306

Table 1
The value of the input parameters used in the numerical calculations (mass and decay constant in
unit GeV)
mt
176

mc
1.4

mb
4.8

m
1.78

mBs
5.26

mW
80.448

fBs [19]
0.14

|Vt b Vts |
0.045

|Vt b Vts /Vcb |2


0.95

1
em
137

(Bs )
1.64 1012 s

sin2 W
0.2325



2(1 4r )s
4r
+
1
,
1 s
s

(4.13)

where s = p2 /m2Bs , r = m2 /m2Bs . z Z(r , s ) takes the form given in Eq. (3.14).
5. Numerical results and conclusion
In this section we give some numerical results and discussions. For reference, we present
our SM predictions 3
Br(B Xs + ) = 2.59 106 ,
Br(Bs + ) = 5.18 108 .

(5.1)

These values are obtained for the fixed input parameters [17] listed in Table 1 and the QCD
coupling constant s (mb ) = 0.218 which is calculated via
s () =

s (mZ )
(mZ )
1 0 s 2
ln mZ

(5.2)

with s (mZ ) = 0.119 [17] and mZ = 91.19 GeV.


In addition, we use the masses, decay widths and branching rates of J / family in
Ref. [17], the normalized factor, branching rate Br(B Xc ) = 10.2%, and take the
dipole forms of the form-factors given by Ref. [18]
 
 
1 GeV
3.74 GeV
g1 p2 =
, g2 p2 =
,
2
2
2
(1 p /5.6 )
(1 p2 /40.5)2
 
 
0.8 GeV
0.68 GeV
, f2 p2 =
.
f1 p2 =
(5.3)
(1 p2 /6.52 )2
(1 p2 /30)2
As mentioned in Section 2, in limit (i) and (ii) there are only two independent parameters
f/f and mp in technicolor model with scalars. The limit on f/f can be obtained from
the studies of b Xc [20], b Xs [20,21], Z bb [21], B Xs + [22] and
mixing in technicolor model with scalars [11,12,21], which is given by [20]
BB


mp
f
(5.4)

0.03
(95% C.L.).
f
1 GeV
3 Switching off the scalars contributions, our formula for Br(B + ) is the same as given in Ref. [9], but
s
different from their result. In fact, the Bremsstrahlung part is 3.98 108 for = 0.02.

Z. Xiong, J.M. Yang / Nuclear Physics B 602 (2001) 289306

301

Fig. 3. Branching ratio of B Xs + versus mp for f/f = 10 (a), and versus f/f for
mp = 500 GeV (b). The dot-dashed line stands for the SM prediction. The dotted (dashed) lines
denote the new physics contributions from , Z exchange (neutral scalar exchange) diagrams shown
in Fig. 1 (Fig. 2). The solid one is the total values.

This indicates that f/f could be large for large mp . On the other hand, we should notice
the induced values of m in Eqs. (2.8), (2.9). Although the lower experimental bound of
107.7 GeV [17] on the SM Higgs boson may not be applied directly to m , we ensure that
the neutral scalar is not lighter than this value in our numerical calculations.
We found that the behaviors of the quantities we studied are similar in limit (i) and (ii).
So for illustrations, we only present some numerical results in limit (ii). Some numerical
examples are presented in Figs. 36.

302

Z. Xiong, J.M. Yang / Nuclear Physics B 602 (2001) 289306

Fig. 4. The same as Fig. 3, but for Bs + .

Figs. 3 and 4 show the branching ratios of B Xs + and Bs + ,


respectively. As expected, the branching ratios drop with mp and increase with f/f . For
the specified values of mp and f/f in the figures, the branching ratios can be enhanced
over the SM results by a couple of orders of magnitude.
Fig. 5 show the forwardbackward (FB) asymmetry as a function of the scaled invariant
dilepton mass squared s. We should point out that since some common contributions
appearing in both the numerator and the denominator cancel out to some extent, the FB
asymmetry is a sensitive, relatively model-independent probe of these models. 4 Fig. 5
4 Although the FB asymmetry is less model-dependent for this reason, it is not completely model-independent
since the model-dependent contributions cannot completely cancel out.

Z. Xiong, J.M. Yang / Nuclear Physics B 602 (2001) 289306

303

Fig. 5. The same as Fig. 3, but for the forwardbackward asymmetry of B Xs + versus the
scaled invariant dilepton mass squared s with f/f = 10 and mp = 500 GeV.

shows a significant difference between the SM and the technicolor predictions, especially
in the region of large invariant dilepton mass. One can see that, unlike the branching
ratio, the forwardbackward asymmetry is enhanced significantly due to the neutral scalars
contributions.
The differential branching ratios of B Xs + and Bs + versus the
scaled invariant dilepton mass squared are shown in Fig. 6, such distributions also differ
significantly from the SM predictions. We stress that all these distributions would be
useful for fitting the future experimental results in the framework of such a technicolor
model, especially when some deviations from the SM predictions are discovered in future
experiments. Different models, such as technicolor models and 2HDMs, may all predict
same enhancements in branching ratios, but they may give different behaviors for some
distributions. To claim a given model is experimentally favored or disfavored, all these
distributions would be useful.
We realized that since there are new model assumptions introduced here, such as the
form-factor (5.3), as well as a large number of experimental inputs, each of which comes

304

Z. Xiong, J.M. Yang / Nuclear Physics B 602 (2001) 289306

Fig. 6. The same as Fig. 3, but for the differential branching ratios of B Xs +
and Bs + versus the scaled invariant dilepton mass squared with f/f = 10 and
mp = 500 GeV.

with its own uncertainty, our conclusions may only be qualitatively reliable. These are:
1. The branching ratios predicted by the technicolor model with scalars can be enhanced over the SM results by a couple of orders of magnitude in parameter space we
studied.
2. The dominant contributions are from the exchange of charged scalars as shown in
Fig. 1. This could be well understood since the contributions from the charged scalars,
in contrast to the decays studied in 2HDM-II [10], are not suppressed but enhanced by a
factor (f/f )2 when f/f is large.
3. The neutral scalars play no important role except for sufficiently large f/f . For
the illustrative values in the numerical results, i.e., 10  f/f  15, the contributions of

Z. Xiong, J.M. Yang / Nuclear Physics B 602 (2001) 289306

305

neutral scalars are still much smaller than those of charged scalars. However, as showed
in the analytical expressions, the contributions of neutral scalars in Fig. 2 are two orders
higher in f/f than those of charged scalars in Fig. 1. Therefore, for sufficiently large f/f
the contributions of neutral scalars could also be important.
4. The possible large enhancements over those predicted by the SM might be detectable in future experiments since the sensitivity of the new generation of B experiment
to these processes should be quite high. If such large enhancements are observed in future
experiments, they could be interpreted in such a technicolor model although this model
might not be the unique one to explain them. If not observed, further stringent constraints
on the model parameter space could be obtained and thus this model would be severely
disfavored.

Acknowledgement
We thank C.-H. Chang and H. J. Yang for discussions and comments. This work is
supported in part by a grant of Chinese Academy of Science for Outstanding Young
Scholars.

References
[1] R. Ammar et al., CLEO Collaboration, Phys. Rev. Lett. 71 (1993) 674;
M.S. Alam et al., CLEO Collaboration, Phys. Rev. Lett. 74 (1995) 2885.
[2] Y. Grossman, Z. Ligeti, E. Nardi, Phys. Rev. D 55 (1997) 2768.
[3] B. Grinstein, R. Springer, M.B. Wise, Phys. Lett. B 202 (1988) 138;
B. Grinstein, R. Springer, M.B. Wise, Nucl. Phys. B 339 (1990) 269.
[4] Y.B. Dai, C.S. Huang, H.W. Huang, Phys. Lett. B 390 (1997) 257.
[5] J.L. Hewett, Phys. Rev. B 53 (1996) 4964.
[6] S. Weinberg, Phys. Rev. D 19 (1979) 1277;
L. Susskind, Phys. Rev. D 20 (1979) 2619.
[7] S. Dimopoulos, L. Susskind, Nucl. Phys. B 155 (1979) 237;
E. Eichten, K. Lane, Phys. Lett. B 90 (1980) 125.
[8] G. Buchalla, A.J. Buras, Nucl. Phys. B 400 (1993) 225;
D. Du, C. Liu, D. Zhang, Phys. Lett. B 317 (1993) 179.
[9] T.M. Aliev, A. Zpineci, M. Savci, Phys. Rev. D 55 (1997) 7059;
T.M. Aliev, N.K. Pak, M. Savci, Phys. Lett. B 424 (1998) 175.
[10] E.O. Iltan, G. Turan, Phys. Rev. D 61 (2000) 034010.
[11] E.H. Simmons, Nucl. Phys. B 312 (1989) 253.
[12] C.D. Carone, E.H. Simmons, Nucl. Phys. B 397 (1993) 591;
C.D. Carone, H. Georgi, Phys. Rev. D 49 (1994) 1427.
[13] M. Neubert, Phys. Rep. 245 (1994) 396.
[14] A. Falk, M. Misiak, M. Munz, Nucl. Phys. B 49 (1994) 3367;
A. Falk, M. Luke, M. Savage, Phys. Rev. D 53 (1996) 2491;
I. Bigi, M. Shifman, N.G. Uraltsev, A.I. Vainshtein, Phys. Rev. Lett. 71 (1993) 496;
A.V. Manohar, M.B. Wise, Phys. Rev. D 49 (1994) 1310.
[15] G. Lu, Z. Xiong, Y. Cao, Nucl. Phys. B 487 (1997) 43.

306

Z. Xiong, J.M. Yang / Nuclear Physics B 602 (2001) 289306

[16] A. Ali, C. Creub, Z. Phys. C 49 (1991) 431;


A. Ali, C. Creub, Phys. Lett. B 259 (1991) 182;
A. Ali, C. Creub, Phys. Lett. C 361 (1995) 146.
[17] Particle Physics Group, Eur. Phys. J. C 15 (2000) 274.
[18] G. Eilam, I. Halperin, R.R. Mendel, Phys. Lett. B 361 (1995) 137.
[19] V.M. Belyaev et al., Phys. Rev. D 51 (1995) 6177.
[20] Z. Xiong, H. Chen, L. Lu, Nucl. Phys. B 561 (1999) 3.
[21] C.D. Carone, Y. Su, Phys. Lett. B 344 (1995) 2015.
[22] Y. Su, Phys. Rev. D 56 (1997) 335.

Nuclear Physics B 602 (2001) 307328


www.elsevier.nl/locate/npe

Saltatory relaxation of the cosmological constant


Jonathan L. Feng a , John March-Russell b , Savdeep Sethi a,c ,
Frank Wilczek a
a School of Natural Sciences, Institute for Advanced Study, Princeton, NJ 08540, USA
b Theory Division, CERN, CH-1211, Geneva 23, Switzerland
c Enrico Fermi Institute, University of Chicago, Chicago, IL 60637, USA

Received 5 March 2001; accepted 12 March 2001

Abstract
We modify and extend an earlier proposal by Brown and Teitelboim to relax the effective
cosmological term by nucleation of branes coupled to a three-index gauge potential. Microscopic
considerations from string/M-theory suggest two major innovations in the framework. First, the
dependence of brane properties on the compactification of extra dimensions may generate a very
small quantized unit for jumps in the effective cosmological term. Second, internal degrees of
freedom for multiply coincident branes may enhance tunneling rates by exponentially large density
of states factors. These new features essentially alter the relaxation dynamics. By requiring stability
on the scale of the lifetime of the universe, rather than absolute stability, we derive a non-trivial
relation between the supersymmetry breaking scale and the value of the cosmological term. It is
plausibly, though not certainly, satisfied in Nature. 2001 Published by Elsevier Science B.V.

1. Introduction
Although both the Casimir effect of quantum theory and the existence of symmetrybreaking condensates in both the strong and the electroweak sectors of the Standard Model
indicate that empty space is a dynamical medium that ought to have a large mass density,
gravity, which couples universally to mass, does not reveal it. This is the problem of the
cosmological term [1]. Various mechanisms have been proposed to address this problem,
but so far none has won wide acceptance. This situation is especially challenging for string
theory, and its conjectured non-perturbative definition M-theory, in as much as string theory
is proposed as a fully specified dynamical theory of gravity.
An interesting approach to the solution of the cosmological term problem is the proposal
that it is relaxed by jumps (saltation) associated with some rather exotic dynamics. There
E-mail addresses: feng@ias.edu (J.L. Feng), jmr@nxth04.cern.ch (J. March-Russell),
sethi@theory.uchicago.edu (S. Sethi), wilczek@ias.edu (F. Wilczek).
0550-3213/01/$ see front matter 2001 Published by Elsevier Science B.V.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 0 9 7 - 9

308

J.L. Feng et al. / Nuclear Physics B 602 (2001) 307328

is an important conceptual advantage to having the relaxation connected to some


necessarily exotic dynamics that responds only to a source taking the form of an
effective cosmological term. For if the dynamics responds to several influences, it is
difficult to see how a particularly simple value for one partial determinant of its behavior
can become overwhelmingly preferred. A model for this logic is the advantage one gains
from PecceiQuinn symmetry in relieving the -problem of QCD. Indeed, that symmetry
produces an exotic dynamics (the axion field) which responds only to a source having
the form of an effective term. Note that this logic also applies, in connection with
continuous relaxation of scalar fields, to self-interactions (including kinetic terms), as has
been quantified by Weinberg [1].
Saltation in the effective cosmological term has been considered in the context of
stepwise false vacuum decay [2,3] in a quasi-periodic staircase potential by Abbott [4], or
alternatively by Brown and Teitelboim (BT) through nucleation of fundamental membrane
degrees of freedom [5]. Since the membrane formulation is close in spirit to the ones that
appear to arise naturally in string theory, we shall phrase our discussion in its framework.
The essential ingredients in the BT model are a fundamental membrane degree of
freedom (in modern language, a 2-brane) and a 4-form gauge field strength F4 (deriving
from a 3-form potential). The world-volumes of the membranes couple to the 3-form
potential. The field strength F4 has no other couplings and no local dynamics in the fourdimensional spacetime, but its expectation value contributes to the effective cosmological
term. The presence of an expectation value for F4 induces the nucleation of the 2-brane
to which it couples, in a manner analogous to the Schwinger mechanism for electric field
decay through nucleation of e+ e pairs. When a membrane is nucleated, say as a spherical
shell, the effective value of the cosmological term on the inside differs from its previous
value (the value on the outside) by an amount proportional to the coupling constant of the
membrane. If a membrane of the correct sign is nucleated, the contribution of the 4-form to
the effective value of the cosmological term will be reduced. For a suitable value of F4 ,
this contribution can cancel that arising from the combined effects of all other degrees of
freedom in the theory. If the steps between adjacent false vacua are sufficiently small, and
if there is a reason why vacuum decay stops or slows down dramatically as eff 0, this
mechanism could in principle relax a large microscopic cosmological term (arising from
all sources other than the membrane F field dynamics) to a value within observational
bounds.
In their pioneering work, Abbott, and Brown and Teitelboim, postulated dynamical
entities ad hoc, with no richer context to enhance their credibility or connect them to
other problems and facts of physics. On the other hand, developments in string theory
in the past few years have emphasized the importance of extended objects branes
of various types. It, therefore, seems appropriate to consider whether these branes might
allow improved mechanisms for saltatory relaxation of the cosmological term. We believe
that this is the case, for several reasons:
A wide variety of membranes play a fundamental role in string theory. There are
corresponding gauge fields, which naturally couple (only) to these extended objects.

J.L. Feng et al. / Nuclear Physics B 602 (2001) 307328

309

Since the theory is naturally formulated in higher dimensions, the couplings of these
membranes as seen in four dimensions are not fixed and quantized, but rather are
determined in terms of the fundamental (fixed, quantized) couplings together with
properties of the extra-dimensional compactification.
Similarly, the effective tension as seen in four dimensions depends on the properties
of the extra-dimensional compactification.
There are significant and, in suitable cases, exponentially large, density of states
factors associated with semi-classical brane processes.
Small tension, which may be favored for dynamical reasons, and large density of
states factors make possible rapid relaxation of the cosmological constant.
We should stress that regardless of whether membrane nucleation is the final solution,
or perhaps an ingredient, in solving the riddle of the cosmological constant, nucleation
processes involving extended objects generically occur in string theory. Studying these
processes is likely to give some insight into the question of vacuum selection and into the
question of a background independent formulation of string theory.
In the following section, we review the BT formalism for relaxation of the cosmological
constant through brane nucleation and the primary obstacles encountered. In the next
two sections, we discuss features of string theory that may alleviate these difficulties: in
Section 3 we describe the possibility of small charge densities and tensions arising from
compactification, and in Section 4 we note the relevance of exponentially large density
of states factors. Motivated by these features, we describe two possible scenarios for the
cosmological constant in Section 5. In Section 6, we discuss a number of outstanding issues
and summarize.

2. Relaxation dynamics
2.1. Basic mechanism
We now recall the basic dynamics of the BT mechanism [5], which we have been able
to express in a somewhat simplified fashion. For other recent discussions of membrane
nucleation, see [6].
Consider gravity in D = 4 spacetime dimensions with a 2-brane X coupled to a 3-form
gauge potential A3 . The Minkowski action is



X X X abc
2

SM = 2 d 3 det gab +
d 3 A
6
a b c





1
1

d 4 x g F F +
d 4 x g F A
48
6



1
+
(1)
d 4 x g M 2 (R 2) M 2 d 3 x h K,
2
where the a parameterize the membrane world-volume, and gab = a X b X is the
induced world-volume metric. The surface integral is over spacetime boundaries with
h and K the induced metric and extrinsic curvature, respectively. This term and the

310

J.L. Feng et al. / Nuclear Physics B 602 (2001) 307328

total derivative integral ensure that the action has well-defined functional derivatives with
respect to the metric and gauge field. An important point is that in four dimensions the
4-form field strength contains no independent propagating degrees of freedom, its value,
up to a constant, being fully determined by the background of sources charged with respect
to A .
The parameters entering this action, and their mass dimensions in D = 4, are
2-brane tension

2 ,

3,

2-brane charge density

2 ,

2,

(Reduced) Planck mass

M,

1,

Bare cosmological constant

2.

= (8G)1/2

(2)

= 2.4 1018

Numerically, M
GeV. Here, in agreement with BT, we use the
canonical positive energy sign for the F 2 term. In Section 3, we shall see that this is
appropriate for the branes relevant to us.
Rotating to Euclidean space, we find



X X X abc
2

SE = 2 d 3 det gab +
d 3 A
6
a b c




1
1

d 4 x g F F +
d 4 x g F A
48
6



1
+ d 4 x g M 2 (R + 2) + M 2 d 3 x h K.
(3)
2
The sign of the F 2 term in Eq. (3) depends on the Euclideanization procedure. Here,
following Ref. [5], we make the conventional rotations x 0 ix 0 , X0 iX0 for
time-like quantities, but take A02 D1 A02 D1 and A1 D1 iA1 D1
for the space-like components so that the field strength F is invariant. Alternatively,
one may adopt the prescription A02 D1 iA02 D1 and keep the space-like
components invariant. In this case, the sign of the F 2 term in Eq. (3) changes. However,
the field strength is not invariant under this prescription; taking F4  in Eq. (3) to be pure
imaginary [7] leaves the following analysis unchanged.
The instanton solution is a membrane that divides space into two regions, an outside O
and an inside I. In each region, the field strength is a constant
  cO,I
,
FO,I =
(4)
g
and the field strengths are matched across the membrane boundary via
cI = cO 2 .

(5)

The effective cosmological terms are


1 2
(6)
c ,
2M 2 O,I
where the field strength contribution follows from Einsteins equations. Alternatively, it
may be read off from the action if one is careful to include the on-shell contribution from
O,I = +

J.L. Feng et al. / Nuclear Physics B 602 (2001) 307328

311

the total derivative term, which is double in magnitude and opposite in sign relative to the
usual F 2 term. From Eq. (6), it is clear that if the bare cosmological term is to be canceled,
it must be negative, and we therefore assume < 0.
The tunneling probability is P eB , where the bounce action [2,3] for this false
vacuum decay is

,
if O , O < 0,
B = 12 2M 2  1 (1 b ) 1 (1 b ), otherwise.
(7)
O

Here the bubble radius, defined so that the area of the bubble slice when continued back to
Minkowski signature is 4b2 , is
b=
and

1
1
3 O

2
+ O

1
1
3 I

(8)

+ I2

1
1 3 2
x 2 ,
cO
I
3xM
2 4

1  2
2 22 cO ,
I = O +
2
2M
2
.
x=
2 M

O =

(9)
(10)
(11)

In the following sections, we will often work in Planck units with M 1.


2.2. Naturalness
The value of the cosmological term at present is
obs M 2  (2 103 eV)4 10120M 4 .

(12)

Present data prefer a non-zero, positive value; for recent reviews, see [8]. One virtue of
saltatory relaxation is that, in contrast to mechanisms involving symmetries or continuous
relaxation, a small non-zero value may emerge naturally as a consequence of a nonvanishing jump size.
For the observed value to be natural in the framework of brane nucleation, the spacing
between allowed values of the effective cosmological term, near the observed value, cannot
be much larger. This translates into the condition
2 

obs
.
||1/2

(13)

This is an extremely stringent condition on the microphysics, even for plausible ||  1,


since the observed cosmological constant is so small. For example, even if the bare cosmological constant is generated only at the TeV scale through low-energy supersymmetry
breaking so that || 1060 , one still requires 2  1090 , which translates into an associated mass scale of 1018 eV. The requirement of such a small coupling is, at best,
unsettling, and one might hope for an explanation in some more fundamental framework.

312

J.L. Feng et al. / Nuclear Physics B 602 (2001) 307328

This is especially true in string theory, where, in the absence of such an explanation, one
naively expects couplings to be of order the Planck scale.
As an aside, note that throughout this study, we consider evolution of the cosmological
term down the staircase of values allowed by brane nucleation with fixed charge density 2 ,
and thus with essentially fixed step size. However, very small fractional changes in 2 , of
order
2 obs
,

2
||

(14)

assuming 2  ||, are sufficient to bring the observed value of the effective cosmological term into range by distorting the size of the steps near = 0. Another possibility is
that the entire staircase moves up or down by some suitably small amount. Although we
will not attempt to exploit these features here, they may play an important role in future
work, since, as we shall emphasize below, 2 is in principle a dynamical variable. It depends, in particular, on the expectation values of the string theory compactification moduli,
including the dilaton.

2.3. Absolute stability

The cosmological term must not only relax to within its observational bounds, but it
must also stop evolving once it reaches this interval. For de Sitter space, additional bubble
nucleations are always possible. However, for O < 0, transitions can take place only when
2 > 0,
O > 0. It is not hard to show that this constraint, along with the condition 13 O + O
implies that further lowering of the effective cosmological term will not occur beyond the
first anti-de-Sitter step if the tension is sufficiently large. This remarkable result is closely
related to the Coleman de Luccia [3] gravitational suppression of false vacuum decay.
BT hypothesized that our universe is at the endpoint of such an evolution.
A sufficient condition to ensure absolute vacuum stability is 22 > 43 2 cO or, in terms of
the tension to charge density ratio,

2
x=

2

||
.
obs

(15)

Thus, even for the smallest plausible ||, a large hierarchy between tension and charge
density is required.
More problematic still, BT showed that, upon combining the stability and naturalness
conditions, the time required to reach the endpoint is excessively large, so large that even
the very slow inflation that occurs in the penultimate vacuum would leave the universe
entirely devoid of matter and energy.

J.L. Feng et al. / Nuclear Physics B 602 (2001) 307328

313

3. Saltation in string theory: tension and charge density


3.1. Framework
In its long wavelength approximation, M-theory supports BPS M2-branes and M5branes [911]. The M2-branes couple electrically to the 3-index gauge potential of
11-dimensional supergravity, which we denote by A3 . We define a dual gauge potential
A6 in the standard way:
F = dA3 = F7 = dA6 .

(16)

The M5-branes couple magnetically to A3 , or directly to A6 . In terms of the 11-dimensional


Planck scale lp , M2-branes have a tension T2 (lp )3 while M5-branes have a tension
T5 (lp )6 .
The simplest way to arrive at a 4-dimensional world is via compactification on a
7-dimensional internal space M. We obtain 2-branes from either the fundamental M2branes of M-theory, or by wrapping M5-branes on a 3-cycle a3 of the internal space M.
Our spectrum of 4-forms in spacetime comes about in the following way: let i be a basis
for H 3 (M, Z). We require integer forms so that the resulting 4-form fields satisfy Dirac
quantization. We can then expand F7 in this basis,
i i .
F7 = F

(17)

Note that we get many 4-forms from a generic compactification of this kind. There is a
natural geometric picture associated to this expansion. To each form i , we can associate
a dual 3-cycle on which we wrap an M5-brane to obtain a 2-brane.
After setting the fermions to zero, the 11-dimensional metric satisfies the equation of
motion


1
1
1
Facde Fbcde gab Fcdef F cdef = 0.
Rab gab R +
(18)
2
6
8
The sign of the four-form contribution is fixed by 11-dimensional supergravity. For both
) to the effective
kinds of effective 2-brane described above, the contribution of F (or F
cosmological constant is positive. We therefore require a negative bare cosmological
constant , as assumed in Section 2.
If M is circle-fibered then we can reduce M-theory on the fiber to obtain various string
compactifications. For example, if we take M to have the form S 1 K then for a small S 1
of size R11 , M-theory reduces to type IIA string theory on K [9]. The string scale  and
string coupling gs are given in terms of lp and R11 ,
 = lp3 /R11 ,

gs = (R11 / lp )3/2 .

(19)

Type IIA string theory contains an NSNS (NeveuSchwarz) 3-form field strength
H3 which measures fundamental string charge. The magnetic dual H7 = H3 measures
the charge of NS 5-branes. These 5-branes have a tension proportional 1/gs2 and so are
very heavy at weak string coupling. In addition, there are RamondRamond (RR) field
strengths F2 and F4 (and a non-dynamical F10 [17]) together, with their 8-form and 6-form

314

J.L. Feng et al. / Nuclear Physics B 602 (2001) 307328

(and 0-form) magnetic duals. These (p + 2)-form RR field strengths couple electrically to
dynamical Dirichlet p-branes [17,18] through the world-volume action

d p+1 Ap+1 .
(20)
In type IIA, we, therefore, have the additional possibility of wrapped D6-branes and D8branes giving rise to effective 2-branes. In 9 + 1 dimensions, the low-energy string and
Dp-brane effective action is


  1 2
1
10
2
2
F +
R + 4()
d x G e
S10 = 2
2n! n
210



Tp d p+1 e det Gab + p d p+1 Ap+1 + ,
(21)
where is the dilaton, Gab is the induced metric on the brane, and only the relevant bosonic
terms are displayed. The tension and charge density of type II Dp-branes is
3p

.
Tp2 = p2 = 2 4 2 
(22)
10
2 = 26 7 4 and defining
This may be written more conveniently by using the relation 10
the string length 6s in terms of the fundamental string tension through TF1 = 1/(2  ) =
2/62s . With these conventions,

Tp = p =

2
p+1

(23)

6s

(Note that there is no factor of 1/gs in these expressions because of the conventions
employed in Eq. (21).)
The 4-dimensional physical brane tensions and charge densities are proportional to Tp
and p . As we discuss in the next subsections, the exact relations depend on the
complicated dynamics of compactification which are not well understood. For this reason,
when we consider scenarios in Section 5, we will take a 4-dimensional effective action
approach, and simply assume suitable values for the 2-brane tension and charge density.
In the rest of this section, however, we explore the possible effects of compactification on
these quantities.
There are also other ways of arriving at a 4-dimensional world that do not fit into the
above framework. For example, compactifying F-theory [12] on a CalabiYau 4-fold [13]
naturally gives a large class of N = 1 compactifications. The 4-fold must be ellipticallyfibered with a section. If B denotes the 6-dimensional base of the fibration then over
each point of B, the structure of the elliptic fibration specifies a torus. By F-theory
compactified on the 4-fold, we mean type IIB string theory compactified on B. However,
the complexified string coupling constant is not constant but varies over B in a manner
determined by the shape of the torus fiber. Therefore, the string coupling constant is
typically not a tunable modulus for these compactifications.
To determine, the spectrum of 2-branes for an F-theory compactification, it is natural to
use the duality between M-theory on T 2 and the type IIB string. From this duality, we see

J.L. Feng et al. / Nuclear Physics B 602 (2001) 307328

315

that 2-branes arise only from M5-branes. An M5-brane wrapping the elliptic fiber and a
1-cycle in B would look like a D3-brane wrapping the 1-cycle in B. There are additional
possibilities. For example, an M5-brane with a leg on the elliptic fiber and 3 legs on B
can give rise to a 2-brane. From the IIB perspective, it would appear to be a combination
of (p, q) 5-branes wrapping a 3-cycle of B. In backgrounds such as these with N = 1
supersymmetry, we also need to worry about the spacetime superpotential. However, our
subsequent discussion will be largely classical. Our aim is to be as simple as possible
in order to isolate the aspects of compactifications that are most relevant for the brane
nucleation mechanism.
We also note that there is a much broader class of compactifications that involve background fluxes. These compactifications, which are actually generic string compactifications, typically warp spacetime [1416]. These backgrounds can potentially give rise to
the kind of novel infra-red physics that could make saltatory relaxation a viable mechanism for reducing the cosmological constant.
3.2. Direct descent
One might suppose that M2-branes are made to order for saltatory relaxation of the
cosmological term. However, as we shall see, they do not have the right properties, at least
when taken in their straightforward form. The wrapped M5-branes appear more promising.
Let us first consider the case of a D2-brane in 10 dimensions that descends directly to
a 2-brane in 4 dimensions. Our discussion will be in the context of 10-dimensional type
IIA supergravity compactified on a CalabiYau manifold K with string-frame volume V6 .
This compactification preserves N = 2 supersymmetry. The physical effective tensions
and charge densities are then determined in 4-dimensional Einstein frame, where the
gravitational action takes the conventional EinsteinHilbert form. This frame follows after
performing the Weyl rescaling g e2 g on the 4-dimensional metric g . The
4-dimensional action is then



 1
V6
4
2
2(n2) 2
S4 = 2
e
Fn +
d x g R 2()
2n!
210



Tp d p+1 ep det gab + p d p+1 Ap+1 + .
(24)
The 4-dimensional effective tension is p |4D,eff = Tp gs , where gs = e is the string
coupling. To determine the effective charge density, note that the Fn2 kinetic terms are
not canonically normalized. Upon restoring
thep normalization, we find a 4-dimensional
effective charge density of p |4D,eff = 2 p gs /M, where the reduced Planck mass is
2
2
fixed
by M = V6 /10 . The tension to charge density ratio, in Planck units, is then x =
1/ 2, as we expect for BPS branes.
For the 2-brane case of interest,


2 2 gs2

2 4D,eff =
(25)
.
M63s
p

316

J.L. Feng et al. / Nuclear Physics B 602 (2001) 307328

Clearly, to obtain a sufficiently small charge density, we must have extreme values for 6s
and/or gs . For example, for the canonical choice of string scale 6s  (1017 GeV)1 , we
find that a charge density of 2  1090 requires gs  1044 . Alternatively we could
take 6s to be a larger length scale. These non-canonical cases include the large extra
dimension scenario [25] with gs 1 and some number of sub-millimeter dimensions.
However, given the success of quantum field theory at colliders such as LEP and the
Tevatron, 6s  (TeV)1 at the very best. Although an improvement, this still requires a
tiny string coupling gs  1023 to generate a sufficiently small 2 . It is possible that this
string coupling is unrelated to the gauge couplings of the Standard Model; for example, the
Standard Model may arise from some non-perturbative sector of string theory. However,
if gs is related to the Standard Model gauge couplings in a straightforward way, it is difficult
to understand how to accommodate the Standard Model in such an extreme corner of string
theory moduli space, much less gauge coupling unification with unif  1/25 [26].
3.3. Degenerating cycles
A more promising alternative to direct descent branes are branes wrapped on homology
cycles. These branes become tensionless when the volume of these cycles approaches zero
classically, as for conifolds [27]. For compactifications with N  2 supersymmetry, these
branes give rise to BPS states and so quantum corrections do not change this conclusion.
Specifically, if a p-brane of tension Tp wraps a k-cycle ak of the compactification
manifold, where k  p and the volume of ak is Vol(ak ), then the result in the effective
4-dimensional theory is a (p k)-brane of tension (pk) Tp Vol(ak ). In particular, for
an effective 2-brane in 4 dimensions coming from the wrapping of a Dp-brane on a cycle
ap2 , the Einstein frame effective tension is

2 4D,eff = Tp Vol(ap2 )gs2 .
(26)
If Vol(ap2 ) approaches zero, i.e., ap2 is a degenerating cycle, a nearly tensionless object
exists in the 4-dimensional theory. The analogous formula for the tension of a wrapped
NS 5-brane just differs by a factor of gs . This is consistent with the higher-dimensional
quantization rules for brane properties.
What determines the charge density? As we saw in the case of the direct reduction of
an M2-brane to a D2-brane, the kinetic terms for the 3-form gauge-field become modulidependent. The term,

d p+1 Ap+1 ,
is reduced on integer classes and so contains no moduli dependence. For purposes of
determining the scaling of the charge density with the moduli, we only need the behavior
of the moduli space metric near the singularity. At the level of classical geometry, this is
determined by the behavior of the WeilPetersson metric. For N = 2 compactifications,
this metric behavior is not changed by quantum corrections, although it is certainly
renormalized in situations with less supersymmetry. Let us take the case of the conifold

J.L. Feng et al. / Nuclear Physics B 602 (2001) 307328

317

where an S 3 shrinks to zero size in K. We choose coordinates for the moduli space so that
the singularity is located at Z = 0. The tension of a wrapped brane goes like,

2 4D,eff |Z|.
However, the moduli space metric scales like ln(Z) and so the charge density behaves
like [28],


2 4D,eff 1/ | ln(Z)|.
Thus although the tensions can be vanishingly small, the charge densities generated are
only slightly smaller than in the direct descent case. We would therefore need to stabilize Z
at extremely small values in order to obtain an acceptably small charge density. 1
The log-singularity in the metric has a simple spacetime interpretation as arising from
integrating out a massless hypermultiplet [27]. We desire a singularity at finite distance
in the moduli space whose behavior is worse than that of a conifold. It is not hard to see
that in the context of N = 2 compactifications, such singularities cannot arise by simply
tuning vector multiplet moduli. For example, one can try generalizations of the conifold
with multiple simple nodes. These singularities are all at finite distance in the moduli
space. However, in each case and in general, it appears that gravity can be decoupled.
The resulting theory of vector multiplets has a moduli space metric with singularities than
metrically cannot be worse than logarithmic by standard non-renormalization theorems.
Too little is currently known about the hypermultiplet moduli space to decide whether a
singularity with the right metric behavior exists. Some results on classifying singularities
at finite distance have appeared in [29]. However, when we relax the condition of N = 2
supersymmetry and consider the far larger class of N = 1 compactifications, it seems far
more likely that a sufficiently bad singularity exists (classically). This is quite exciting
since it connects the possibility of novel infra-red physics with a mechanism for reducing
the cosmological constant. A recent F-theory example with unusual infra-red physics
arising from a bad singularity appeared in [30]. This kind of example certainly has the
right qualitative features, and it actually seems quite hard to rule out the existence of
a compactification with the features we desire. Clearly more work along these lines is
needed.
We have yet to discuss supersymmetry breaking. One natural mechanism worth
mentioning in this context breaks supersymmetry by turning on RR and NS fluxes on K.
For example, we can give (3 + 1)-Lorentz invariant expectation values to the RR fields Fn
by taking,
(2)

F2  = v2i i ,

(4)

F4  = v4i i

+ v4 (4) .

The i are harmonic forms on K, and (4) is the spacetime volume element. If we take the
i
10-dimensional Poincar duals of these expectation values v(4,2)
, we then find expectation
i
(4)
(2,4)
values of the general form F = v(4,2) 

, which couple magnetically to the Dpbranes. Let us denote the expectation values collectively by va .
1 We thank J. Polchinski for correcting an error in an earlier version of this paper, and J. Maldacena for
discussions.

318

J.L. Feng et al. / Nuclear Physics B 602 (2001) 307328

Generically, when va = 0, supersymmetry is broken [1923], and a scalar potential is


generated that depends on the CalabiYau moduli t and the expectation values va . What
is interesting to us, however, is that the potential naturally tends to drive us to singular
compactifications. In some examples involving just RR fluxes, two types of critical point
for the scalar potential have been found [20,21]: either the CalabiYau runs away to
infinite volume, or the theory is driven to conifold-like points where homology cycles
of K degenerate (classically approach zero volume). The minima tend to either break
supersymmetry completely or restore the full N = 2. Moreover, since the configurations
with va = 0 are not iso-potential, there are dynamical processes whereby the values of the
va s change. The va discharge by the nucleation and expansion of charged membranes,
the D-branes of string theory.
Classically, the vacuum expectation values va of the RR fields can vary continuously.
Quantum mechanically, they satisfy quantization rules that follow from the quantization of
brane properties, which are standard results in string theory [24]. These brane properties are
fixed, given information about the compactification and the parameters of our low-energy
effective Lagrangian.
There is an additional issue concerning this means of supersymmetry breaking that
deserves mention. In the context of M-theory on K S 1 , a class of 2-branes arise from M5branes wrapping 3-cycles. In type IIA, these 2-branes arise from either D4-branes wrapping
2-cycles of the compactification space K, or from NS 5-branes wrapping 3-cycles of K.
The two cases are quite different. In the case of the wrapped D4-branes, we must also
consider D2-branes which can also wrap the shrinking 2-cycle. These wrapped branes give
rise to particle states with mass 0 Vol(a2 )/63s 0. With multiple collapsed nodes, we
will find an interacting gauge theory in 3 + 1 dimensions. In this case, it seems possible that
the wrapped D4-branes will correspond to collective excitations of the gauge theory. This
is not an issue, however, for the wrapped NS 5-branes since there are no BPS D3-branes in
type IIA string theory. These wrapped NS-branes give rise to inherently non-perturbative
stringy excitations.
The phenomenon of classically vanishing tensions arising from branes wrapped on
degenerating cycles is intriguing. It is also worth noting that the classical phenomenon
of true degeneration and corresponding tensionless, or massless, states is typically not
realized in the full quantum theory. Instead the effective (3 + 1)-dimensional 2-brane
will have a dynamically generated non-perturbative tension, which may be exponentially
small. It is known that in some cases a tension is generated from the dynamics of the
would-be massless particle states arising from a D2-brane wrapping the 2-cycle. These
particle states realize a non-Abelian gauge theory, presumably in a sector hidden with
respect to the Standard Model, whose low-energy non-perturbative dynamics can break
supersymmetry. (This sector is conceptually distinct from hidden sectors postulated to
provide supersymmetry breaking for the supersymmetric Standard Model.) This leads
to a potential for the volume modulus of the cycle, which stabilizes it at a scale
2 )M [22,23,31,32]. Once the cycle is stabilized at this small scale,
exp(8 2 /b1 gYM
membranes wrapping this cycle have a tension that is also proportional to and thus can
be very small, even for gYM 1. The precise conditions under which very small tensions

J.L. Feng et al. / Nuclear Physics B 602 (2001) 307328

occur deserve much further study.


Some years ago, one of us made the numerical joke [33]

4
obs M 2 e/unif M ,

319

(27)

with unif  1/25. Our present considerations suggest an intimate connection between
non-perturbative effects and the value of the observed cosmological term, which could
conceivably lead to a relation of just this form.

4. Saltation in string theory: density of states


4.1. Multi-bounce properties
An essentially new feature is introduced by multi-bounce solutions arising from
coincident branes. Such coincident branes support low-energy internal degrees of freedom.
Thus there are potentially large density of states factors accompanying their nucleation.
Calculations [34,35] performed in the context of checking duality between type I and
heterotic SO(32) string theory demonstrate that D-branes do make contributions that can be
interpreted semi-classically as incorporating degeneracy factors reflecting the non-Abelian
structure of coincident D-branes. Another aspect of this is that many coincident branes with
large total charge can be described in appropriate limits as black objects, similar to black
holes, with event horizons, and with associated BekensteinHawking (BH) entropy [36].
Consider first the case of k coincident D3-branes. Such a configuration possesses a
U (k) super YangMills (SYM) gauge theory on its world-volume. In the limit where the
interactions with the bulk string theory are weak, and where the temperature (or excitation
energy) of the SYM is small, one can compute the entropy of this system.
2  kg 2 is small, the entropy of the gas of
When the effective SYM gauge coupling geff
YM
massless gauge bosons and their superpartners at temperature T is simply
2 2 2
k V T 3,
(28)
3
where V is the spatial volume of the 3-branes. What happens when the effective coupling
is large? In this case one can use the type IIB supergravity solution describing the k
coincident 3-branes. These classical solutions with the asymptotic geometry and quantum
numbers appropriate for k coincident 3-branes contain a non-extremality parameter upon
which their masses and horizon areas depend. If we associate the area of the horizon
with BH entropy, we can derive a temperature by taking an appropriate derivative. By
this procedure, the supergravity picture yields the strong coupling form of the entropy.
In this case, the entropy agrees with the preceding weak coupling formula up to a
numerical prefactor 3/4, which is then a prediction for the strong coupling behavior of
the theory.
For k coincident D2-branes the UV theory (in the decoupling limit) is again a weaklycoupled U (k) SYM theory, so the UV entropy again scales as k 2 . However, we are
interested in the IR entropy since, as we will argue in the next subsection, the physically
S3 =

320

J.L. Feng et al. / Nuclear Physics B 602 (2001) 307328

motivated temperatures are the ambient de Sitter temperatures which are small (vanishingly
small as O obs ). In the IR the SYM theory on the (2 + 1)-dimensional world-volume
becomes strongly coupled, so one must switch over to the supergravity description. As
shown in Ref. [37], the theory flows in the far IR to that of the M2-brane with BH entropy
inferred from the horizon area given by [36]
S2  k 3/2AT 2 ,

(29)

with A the 2-brane area. This strongly suggests that such strongly-coupled brane
configurations support O(k 3/2 ) light degrees of freedom, thoug