IAS–TH–00/18, NSF-ITP-00-16

hep-th/0003136
March 2000
The String Dual of a Confining Four-Dimensional Gauge Theory

∗ †
Joseph Polchinski and Matthew J. Strassler
arXiv:hep-th/0003136v2 16 Mar 2000

Abstract

We study N = 1 gauge theories obtained by adding finite mass terms

to N = 4 Yang-Mills theory. The Maldacena dual is nonsingular: in each
of the many vacua, there is an extended brane source, arising from Myers’
dielectric effect. The source consists of one or more (p, q) 5-branes. In par-
ticular, the confining vacuum contains an NS5-brane; the confining flux tube
is a fundamental string bound to the 5-brane. The system admits a simple
quantitative description as a perturbation of a state on the N = 4 Coulomb
branch. Various nonperturbative phenomena, including flux tubes, baryon
vertices, domain walls, condensates and instantons, have new, quantitatively
precise, dual descriptions. We also briefly consider two QCD-like theories.
Our method extends to the nonsupersymmetric case. As expected, the N = 4
matter cannot be decoupled within the supergravity regime.

∗ Institute for Theoretical Physics, University of California, Santa Barbara CA 93106-4030

† School of Natural Sciences, Institute for Advanced Study, Olden Lane, Princeton NJ 08540

1
 I. INTRODUCTION

The proposal of ’t Hooft [1], that large-N non-abelian gauge theory can be recast as a
string theory, has taken an interesting turn with the work of Maldacena [2]. The principal
Maldacena duality applies not to confining theories but to conformal N = 4 gauge theories,
which are dual to IIB string theory on AdS5 ×S 5 . Starting with this duality one can perturb
by the addition of mass terms preserving a smaller supersymmetry, or none at all, and in
this way obtain a confining gauge theory. The problem is that the perturbation of the dual
string theory appears to produce a spacetime with a naked singularity [3]. As a consequence,
even basic quantities such a condensates are incalculable.
In this paper we show that the situation is actually much better. There is no naked sin-
gularity, but rather an expanded brane source, and all physical quantities are calculable. We
believe that this is the first example of a dual supergravity description of a four-dimensional
confining gauge theory. It is also gives new insight into the resolution of naked singularities
in string theory.
We focus on perturbations that preserve N = 1 supersymmetry, though in fact our
solutions are stable under the addition of small additional masses that break the supersym-
metry completely. The N = 4 vector multiplet contains an N = 1 vector multiplet and
three N = 1 chiral multiplets. We will add finite N = 1 supersymmetry-preserving masses
to the three chiral multiplets. For brevity we will refer to this theory as ‘N = 1∗ ’. This
theory has been studied by many authors [4–8]. It is known to have a rich phase structure
[4,5,9], which includes confining phases that are in the same universality class as those of
pure N = 1 Yang-Mills theory. We will show that the rich structure of this theory is reflected
in supergravity in remarkable ways.
To study pure N = 0 or N = 1 Yang-Mills theories would require working at small ’t
Hooft coupling and taking the masses of the extra multiplets to infinity. This is not tractable
without an understanding of classical string theory in Ramond-Ramond backgrounds at large
curvature. At large ’t Hooft coupling, where supergravity is valid, the masses of the extra
multiplets must be kept finite. However, we emphasize these multiplets are four-dimensional
and the ultraviolet theory is conformal. An alternative approach to obtaining a string dual
of confining theories is via high-temperature five-dimensional supersymmetric field theories
[10–13], whose low-energy limit is four-dimensional strongly-coupled non-supersymmetric
Yang-Mills theory. The dual spacetime is non-singular, and the infrared cutoff provided by
the temperature does indeed lead to confinement of electric flux tubes. In this case, however,
there is a full set of massive five-dimensional states that do not decouple.
Our work was motivated by the observation of Myers [14], that D-branes in a transverse
Ramond-Ramond (RR) potential can develop a multipole moment under fields that normally

2
couple to a higher-dimensional brane. This ‘dielectric’ property is analogous to the induced
dipole moment of a neutral atom in an electric field. For example, a collection of N D0-branes
in an electric RR 4-form flux develops a dipole moment under the corresponding 3-form
potential. One can think of them as blowing up into a spherical D2-brane, and in a strong
field the latter is the effective description. This happens because the D0-brane coordinates
become noncommutative. The original D0-brane charge N, which of course is conserved in
this process, shows up as a nonzero world-volume field strength on the D2-branes. Even
earlier, Kabat and Taylor [15] had observed that N D0-branes with noncommuting position
matrices could be used to build a spherical D2-brane in matrix theory, generalizing the flat
membranes of matrix theory [16]. For finite N the sphere is ‘fuzzy’; or better, perhaps, it is
somewhat granular. The equations describing this sphere bear a marked similarity to those
which appear in the N = 1∗ theory, which were first analyzed in [4].
It is then natural to guess that Myers’ mechanism is at work in this theory. The mass
perturbation corresponds to a magnetic RR 3-form flux, which is dual to an electric RR
7-form flux. The latter couples to the D3-brane in the same fashion as the electric 4-form
flux does to the D0-brane, and so the D3-branes polarize into D5-branes with world-volume
R4 × S 2 . One difference is that Myers considers D-branes in flat spacetime (gN small),
whereas for the gauge/gravity duality the background is AdS5 × S 5 . In Myers’ case a small
field produced a small D2-sphere, but in the conformal field theory there is no invariant
notion of a small mass perturbation, and on the supergravity side there is no such thing as
a small transverse two-sphere. Rather, the D5-spheres, which are dynamically (though not
topologically) stable, wrap an equator of the S 5 . We will show that there exist supergravity
solutions in which the only ‘singularity’ is that due to the D5-brane source on the S 5 .
However, this is far from the whole story. First, the classical N = 1∗ theory has many
isolated vacua [4]. For each partition of N into integers ni , there must be a separate solution
involving multiple D5-branes with D3-branes charges ni , each wrapped on an equator of S 5
but at different AdS radii ri proportional to ni . We will study these vacua, and their
properties, in our discussion below. Second, the quantum theory has even more vacua,
which are permuted under the SL(2, Z) duality the field theory inherits from N = 4 [5].
In particular, the transformation τ → − τ1 , which takes the maximally Higgsed vacuum into
the confining vacuum, will replace the D5-brane sphere with an NS5-brane sphere: this
is the effective string description of the confining vacuum. The confining flux tubes are
bound states of a fundamental string to the NS5-brane, or equivalently, instantons of the 5-
brane world-volume noncommutative gauge theory. Meanwhile, the leading nonperturbative
condensate corresponds to the three-form field generated by the NS5-brane’s magnetic dipole
moment.

3
 Our removal of the singularity resembles phenomena that occur on the Coulomb
branch [17–20] and with the repulson singularity that arises in N = 2 supergravity du-
als [21]. There are certainly connections which need to be developed further, but the de-
tailed mechanism is different. In particular, the appearance of NS-branes is new. Our result
also gives insight into perturbations of the Randall-Sundrum compactification [23], and into
recent proposals for the solution to the cosmological constant problem [24].
We begin in section II with a review of the classical and quantum field theory vacua,
and a discussion of the corresponding brane configurations. In fact, there are more brane
configurations than vacua, but later we will argue that only one configuration is applicable for
any given value of the parameters. In section III we review perturbations of the AdS/CFT
duality, with attention to the issue of the naked singularity. We show that there is a small
parameter: the system can be regarded as a perturbation of one that has only D3-brane
charges. This enables us to obtain a quantitative description even for the rather asymmetric
and nonlinear supergravity configuration that results from the expansion of the branes. In
section IV we study a simplified calculation, in which n ≪ N probe D3-branes are introduced
into a fixed background. We find that their potential has minima where they form a D5-
brane or NS5-brane, or more generally one or more (c, d) 5-branes, wrapped on an equator
of the S 5 . In section V we consider the case that all N D3-branes expand into 5-branes.
Although this substantially deforms the geometry, serendipitous cancellations allow us to
find the effective potential in a simple form: it is the same as in the probe case. We
discuss the stability of the solution, arguing that it survives even when supersymmetry is
broken completely. In section VI we use the dual description to discuss the physics of the
gauge theory, including flux tubes and confinement, baryons, domain walls, condensates,
instantons, and glueballs. In section VII we briefly discuss extensions, including the N = 0
case and orbifolds, and in section VIII we discuss implications and future directions.

II. N = 1∗ GROUND STATES

A. Field Theory Background

In the language of four-dimensional N = 1 supersymmetry, the N = 4 theory consists of a

vector multiplet V and three chiral multiplets Φi , i = 1, 2, 3, all in the adjoint representation
of the gauge group. In addition to the usual gauge-invariant kinetic terms for these fields,
the theory has additional interactions summarized in the superpotential1

1 The 2 )tr Φ Φ .
Kähler potential is normalized (2/gYM i i

4
 √
2 2
W = 2 tr([Φ1 , Φ2 ]Φ3 ) . (1)
gYM

The theory has an SO(6) R-symmetry which is partially hidden by the N = 1 notation;
only the U(1) R-symmetry of the N = 1 supersymmetry and the SU(3) that rotates the Φi
are visible. However, if we write the lowest component of Φi as
Ai+3 + iAi+6
φi = √ (2)
2
(the reason for this notation will become evident later), then the potential energy for the
scalar fields Am , m = 4, . . . , 9, is explicitly SO(6) invariant:
9
X
V (Am ) ∝ tr ([Am , An ][Am , An ]) . (3)
m,n=4

The theory is conformally invariant, and consists of a continuous set of theories indexed by
θ
a marginal coupling τ = 2π + i g4π
2 , where θ and gYM are the theta angle and gauge coupling
YM
of the theory.
We can partially break the supersymmetry by adding arbitrary terms to the superpo-
tential. Consider the addition of mass terms
1
∆W = 2
(m1 tr Φ21 + m2 tr Φ22 + m3 tr Φ23 ) . (4)
gYM

If m1 = m2 and m3 = 0 the theory has N = 2 supersymmetry; otherwise it has N = 1. If

m1 = m2 = 0 and m3 6= 0 then the theory flows to a conformal fixed point with a smooth
moduli space and SL(2, Z) duality [25,26,19,27]. With two nonzero masses, the theory has
a moduli space containing special subspaces where charged particles are massless and the
Kähler metric is singular. However, in N = 1∗ , where all three masses are non-zero, there
is no moduli space; the theory has a number of isolated vacua. In the limit

τ → i∞ , mi → ∞ , Λ3 = m1 m2 m3 e2πiτ /N fixed (5)

the theory becomes pure N = 1 Yang-Mills theory. For gauge group SU(N) the pure N = 1
theory has N vacua related by a spontaneously broken discrete R-symmetry. Note that this
R-symmetry is not present in N = 1∗ ; it is an accidental symmetry present only in the limit
Eq. (5).
The classical vacua were described by Vafa and Witten [4]. Assuming all masses are
nonzero, we may rescale the fields Φi so as to make all the masses equal; having computed
the vacua in this case one may undo this rescaling. In this case the F -term equations for a
supersymmetric vacuum read

5
 m
[Φi , Φj ] = − √ ǫijk Φk . (6)
2
Consider the case of SU(N). Recalling that the Φi are N × N traceless matrices, it is
evident that the solutions to these equations are given by N-dimensional, generally reducible,
representations of the Lie algebra SU(2). The irreducible spin (N − 1)/2 representation is
one solution; N copies of the trivial representation give another (Φi = 0). Since for every
positive integer d there is one irreducible SU(2) representation of dimension d, each vacuum
corresponds to a partition of N into positive integers:
N
X
{kd ∈ Z ≥ 0} such that dkd = N , (7)
d=1

where kd is the number of times the dimension d representation appears. The number of
classical vacua of the theory is given by the number of such partitions.
Generally, for a given partition, the unbroken gauge group is [⊗d U(kd )] /U(1). For
example, if kd = 1 and kN −d = 1, then the Φi are block diagonal with blocks of dimension
d and N − d; the diagonal traceless matrix which is 1 in each block generates an unbroken
U(1) gauge symmetry. Clearly we obtain U(1)k−1 if there are k such blocks. However, if
kd = 2, then the two blocks of size d can be rotated into each other by additional generators,
giving altogether an SU(2) instead of a U(1). More generally we obtain SU(kd ). Among
these vacua there is a unique one which we will call the ‘Higgs’ vacuum, in which the SU(N)
gauge group is completely broken. This is the only ‘massive vacuum’ (meaning that it has
a mass gap) at the classical level. For each divisor d < N of N we may take kd = N/d with
all others zero, giving a vacuum with a simple unbroken gauge group SU(N/d). All other
vacua have one or more U(1) factors; these are ‘Coulomb vacua.’
Quantum mechanically, the story is even richer. Donagi and Witten [5] found an inte-
grable system which permitted them to write the holomorphic curve and Seiberg-Witten
form describing the quantum mechanical moduli space of the N = 2 theory with m1 = m2
and m3 = 0.2 They considered the effect of breaking the supersymmetry to N = 1 through
nonzero m3 ≪ m1 , m2 , and showed that the theory has a number of remarkable properties.
Each classical vacuum which has unbroken gauge symmetry SU(k) splits into k vacua, all
of which have a mass gap. (Coulomb vacua with non-abelian group factors split as well,
although a complete accounting of these vacua was not given in [5]; since the photons remain
massless, such vacua do not have mass gaps.) The vacuum with SU(N) unbroken (k1 = N,
Φi = 0) splits into N massive vacua, exactly the number which would be needed in the

2 Itwould be very interesting to find this integrable system in the supergravity dual description
of this theory.

6
N = 1 Yang-Mills theory obtained in the limit Eq. (5). The massive quantum vacua are
those without U(1) factors, and as noted above are associated with the divisors of N. Their
total number is obviously given by the sum of the divisors of N; it therefore depends in an
interesting way, one which does not have a large-N limit, on the prime factors of N. The
√
number of Coulomb vacua is exponential in N .
Donagi and Witten showed the massive vacua were in a beautiful one-to-one correspon-
dence with the phases of gauge theories classified by ’t Hooft. Let us review this classification
[28]. SU(N) gauge theories with only adjoint matter can be probed by sources which carry
electric charges in the ZN center of SU(N) and magnetic charges in the ZN = π1 [SU(N)/ZN ]
which characterizes possible Dirac strings. We may think of these charges as lying in an
N × N lattice, a ZN × ZN group L. ’t Hooft showed that the possible massive phases
of SU(N) gauge theories are associated to the dimension-N subgroups P of L. In each
phase, the charges corresponding to the N elements of P are screened, and all others are
confined; the flux tubes which do the confining are represented by the elements of L/P . For
example, if the ordinary Higgs mechanism creates a mass gap, all sources with magnetic
charge are confined; the only unconfined elements of L are the (m, 0), m = 0, . . . , N − 1.
Thus P is generated by the single element (1, 0). Every magnetic flux tube carries a ZN
charge n = 0, . . . , N − 1 and confines the sources with charge (m, n) for any m. In an
ordinary confining vacuum, the roles of m and n are reversed, but otherwise the story is the
same. Vacua with oblique confinement are given by groups P generated by (m, 1), where
m = 0, . . . , N − 1.
More generally, however, the vacua are more complex. As mentioned earlier, each classi-
cal vacuum with unbroken SU(k) symmetry splits into k vacua. These vacua correspond to
subgroups P generated by (k, 0) and (s, d), where dk = N and s = 0, 1, . . . , k − 1. This map
of vacua to subgroups is one-to-one and onto. Note the Higgs vacuum is the case d = N,
while the N vacua which survive in the pure N = 1 Yang-Mills theory are the cases d = 1
for s = 0, 1, . . . , N − 1, with s = 0 being the confining vacuum.
The action of SL(2, Z) on the massive vacua is then straightforward [5]. The T trans-
formation τ → τ + 1 shifts each element (m, n) of the group L to (m + n mod N, n); all
electric charges shift by their magnetic charge, through the Witten effect [29]. The S trans-
formation τ → − τ1 reverses electric and magnetic charges [30]: (m, n) → (−n mod N, m).
Thus S and T map L to itself, but act nontrivially on its subgroups P . This action then
corresponds to a permutation of the massive vacua. In particular, note that the Higgs and
confining vacua are exchanged by S, while T rotates the confining and oblique confining
vacua into each other while leaving the Higgs vacuum unchanged. S and T then generate
the entire SL(2, Z) group and its action on the vacua. The Coulomb vacua have not been

7
fully classified, and the action of SL(2, Z) on them has not yet been understood.3
We close the discussion of field theory by noting that this theory is very different from
N = 1 Yang-Mills theory in certain respects. (Recently, many of these qualitative points
were emphasized in [8].) Although it is a four-dimensional theory, it still has massive degrees
of freedom (three Weyl fermions and six real scalars in the adjoint representation) with
masses of order m. These massive states ensure that far above the scale m (actually, as we
2
will see, above mgYM N in the confining phase) the theory becomes conformal, with gauge
coupling τ . The important Z2N non-anomalous R-symmetry of the pure N = 1 Yang-Mills
theory, a Z2 of which is unbroken and a ZN of which permutes the N vacua of the theory,
is broken explicitly by the presence of the massive fields. Consequently the confining and
oblique confining vacua, although still permuted by τ → τ + n with n an integer, are not
related by a discrete R-symmetry and are not isomorphic. In particular their superpotentials
have different magnitudes and the domain walls between them have a variety of tensions
[8]. In the limit of Eq. (5), for fixed N, the strong coupling scale and the corresponding
gluino condensate, domain wall tension, and string tension are all much below the scale m
of the masses, and so the strong dynamics is not affected by the massive fields. However,
2
we want to study the gravity dual of this theory, which requires large gYM N. In this limit
2 2
Λ = m exp(−8π /gYM N) is of order m, and so all of the physics of the theory takes place near
the scale m. We will not find the exponentially large hierarchy expected from dimensional
2
transmutation; this can only be seen at small gYM N, outside the supergravity regime.

B. Brane Representations

Consider the Higgs phase, in which

A7 = −mL1 , A8 = −mL2 , A9 = −mL3 , (8)

where Li is the N-dimensional irreducible representation of SU(2). The scalars Am are the
collective coordinates of the D3-branes, normalized xm = 2πα′ Am [31]. These are therefore
noncommutative, but lie on a sphere of radius r = πα′ mN

xm xm = (2πα′ m)2 Li Li ≈ π 2 α′2 m2 N 2 . (9)

The nonzero commutator of the collective coordinates corresponds to higher-dimensional

brane charge, a fact familiar from matrix theory. Specifically [15] the D3-branes can be

3 Insection VI.C we will show that some of the Coulomb vacua are transformed in a simple way
by certain elements of SL(2, Z). However, we will not obtain the full story.

8
equivalently represented as a single D5-brane of topology R4 × S 2 , the two-sphere having
radius r, with N units of world-volume magnetic field on the two-sphere. The Higgs vacuum
of the four-dimensional theory is represented by this D5-brane.
Similarly, a vacuum corresponding to the reducible representation {kd }, defined as in
Eq. (7), corresponds to concentric D5-branes, where kd have radius πα′ md for each d. Con-
sider the case of two spheres, with kd = kN −d = 1. If d 6= N − d then the spheres have
different radii; the gauge group of the field theory is U(1). However, if d = N − d, the
two spheres coincide and the field theory has gauge group SU(2). For N − 2d small, the
SU(2) is broken at a low scale and its W-bosons have mass proportional to N − 2d. More
generally, k coincident D5-branes correspond to a classical vacuum with SU(k) symmetry.
Just as in the case of flat branes with sixteen supercharges, the curved D5-branes with four
supercharges and only four-dimensional Lorentz invariance show enhanced gauge symmetry
when they coincide, and when separated have W-bosons with masses of order the separation
distance.4 Each classical vacuum of the theory is given by a set of D5 branes of radius ni ,
with i ni = N.
P

Quantum mechanically the situation is much more complicated. The S transformation

τ → − τ1 should exchange the Higgs and confining vacua; therefore by Type IIB duality the
confining vacuum is a single NS5-brane. A T n transformation (τ → τ + n) leaves D5-branes
unchanged and shifts an NS5-brane to a (1, n) 5-brane. It follows that the nth oblique
confining vacuum is given by a (1, n) 5-brane. The S-duality implies also that there should
be vacua with multiple NS5-branes, or generally (1, n) 5-branes, possibly coincident. In fact
we may expect there to be vacua in which different types of 5-brane coexist. For example,
suppose we partition N using ks = 1 and k1 = N − s, so that the lower (N − s) × (N − s)
block of the fields Φi is zero, leaving SU(N − s) unbroken. In this case we would expect
a D5-brane of radius s representing the broken part of the gauge group, and an NS5-brane
(or a (1, q) 5-brane) representing an (oblique) confining phase of the unbroken SU(N − s)
subgroup.
We will show that all of these brane configurations do indeed appear in the dual of the
N = 1∗ theory. This is a puzzle, however: the number of brane configurations is much larger
than the number of phases. For N = pq, for example, the vacuum with kp = q is described
by q D5-branes of radii πα′ mp. In supergravity, this is clearly S-dual to the vacuum with
kq = p, which is therefore described by q NS5-branes. However, from investigation of the

4 The absence of an overall center-of-mass U (1) in the brane configuration, in parallel with the
absence of a U (1) in the gauge theory, is not completely understood, although we will comment
on it in section VI.H.

9
field theory [5], this vacuum also has a description in terms of p D5-branes. We will see, in
this and other examples, that our solutions exist only in limited ranges of parameter space,
such that only one of the descriptions is valid at a time. Ideally, however, a more complete
understanding of how the theory resolves this puzzle would be desirable.

III. PERTURBATIONS ON AdS5 × S 5

In this section we first review deformations of the AdS/CFT duality with attention to the
issue of singularities, introduce the small parameter that makes the problem tractable, and
discuss the field theory perturbation and its supergravity dual. We then give the IIB field
equations, develop the necessary tensor spherical harmonics, and solve the field equations
to first order in an expansion around AdS5 × S 5 .

A. AdS/CFT and its Deformations

The d = 4, N = 4 Yang-Mills theory is dual to IIB string theory on AdS5 × S 5 [2].

2
The Yang-Mills coupling is related to the string coupling by gYM = 4πg, and the common
radius of the two factors of spacetime is R = (4πgNα′2 )1/4 . To each local operator Oi of
dimension ∆i in the CFT corresponds two solutions of the linearized field equations [32,33],
a nonnormalizable solution which scales as r ∆i −4 with r the AdS radius, and a normalizable
solution which scales as r −∆i . A supergravity solution which behaves at large r as

ai r ∆−4 + bi r −∆ (10)

is dual to a field theory with Hamiltonian

H = HCFT + ai Oi , (11)

and where the vacuum expectation value (vev) is [34]

h0|Oi |0i = bi . (12)

We will be interested in relevant perturbations, those with ∆ < 4. In the field theory
these are unimportant in the UV, while in the IR they become large and take the theory to
a new fixed point or produce a mass gap. Correspondingly the perturbation (10) is small at
large r, but at small r it becomes large and nonlinear effects become important.
For a theory with a unique (or at least isolated) vacuum, the dynamics should determine
the vev once the Hamiltonian is specified. This is in accord with the general experience with
second order differential equations, where some condition of nonsingularity at small r would
give one relation for each pair ai and bi .

10
 Now let us summarize what is known, with attention first to two special cases that make
sense:
1. In the N = 4 theory, ai = 0, it is actually possible to vary the particular b that
corresponds to O being a scalar bilinear. The point is that the N = 4 theory does not have
an isolated vacuum, and varying b gives a state on the Coulomb branch. It is important
to note that the supergravity solution is still singular, but that the singularity is physically
acceptable, corresponding to an extended D3-brane source [17–20].
2. Certain perturbations give a nontrivial fixed point in the IR. These correspond to
supergravity solutions with AdS behavior at large and small r, with a domain wall interpo-
lating [35,36,25,26,37]. The vacua do have moduli, but most or all analyses have imposed
symmetries which determine a unique vacuum and restrict to a single pair (a, b). In these
cases the differential equation does indeed determine b. The condition of AdS behavior in
the IR gives a boundary condition, which takes the form of an initial condition for damped
potential motion.
3. More generally, for perturbations that produce a mass gap and destroy the moduli
space, the known solutions are singular for all values of bi [38,3]; for a recent discussion
see [39]. It does not make sense, however, that such singularities can all be understood as
physically acceptable brane or other sources, because that would mean that the vevs are
undetermined even though the vacua are isolated. This is another example of the important
observation made by Horowitz and Myers in the context of negative mass Schwarzschild [40]:
string theory does not repair all singularities; many singular spacetimes do not correspond
to any state in string theory.
We will show that the perturbations corresponding to the masses (4) actually produce
spacetimes with extended brane sources. The spacetime geometry is singular, but in a way
that is fixed by the source, and so in particular the values of bi are determined.
This resembles the case 1 in that there are extended branes, and could in principle be
analyzed by supergravity means as in that case: for some subset of the supergravity solutions
the singularity will have an acceptable physical interpretation as a brane source. There has
in fact been a search for just such solutions [41,39]; it has thus far been unsuccessful, but
some features of our solution have been anticipated. This approach is extremely difficult,
and has generally been restricted to special solutions with constant dilaton. In fact, the
branes in our solution couple to the dilaton, which is therefore position-dependent.
We are able to treat these rather asymmetric geometries without facing the full nonlin-
earity of supergravity because of the existence of a small parameter. Consider the case of
a single D5-brane with D3-brane charge N, wrapped on an equator of the S 5 . The area of
the two-sphere is of order R2 , so the density of D3-branes is

11
 N N 1/2
∼ . (13)
R2 g 1/2 α′
Under the rather weak condition N/g ≫ 1, this is large in string units and the effect of
the D3-brane charge dominates that of the D5-charge charge.5 The system is therefore well
approximated by a Coulomb branch configuration of the parent N = 4 theory, where the
general solution is given by linear superposition in the harmonic function. Thus we can work
by treating the D5-brane charge, and the 3-form field strengths that are generated by it, as
perturbations. It is less obvious, but will be seen in section IV.A, that the full 3-form field
strength is effectively proportional to the same small parameter.
For the NS5 solution the corresponding condition is given by g → 1/g and so Ng ≫ 1.
This is precisely the condition for the gauge theory to be strongly coupled. We then recognize
the earlier condition N/g ≫ 1 as the condition for the dual gauge theory to be strongly
coupled. When both of these conditions are satisfied the supergravity description is valid,
so the D5 and NS5 solutions are both valid in the entire supergravity regime.
We will begin with a simpler problem, where we place a probe D5-brane of D3-brane
charge n into the linearized perturbation of the AdS5 × S 5 background. In this case the
condition for the D5-brane solution to be valid is similarly
n2
≫1. (14)
gN
We will use this condition at several points. In section V.B we will infer that this condition
is not just a convenience but in fact a necessity in order for the solution to exist.

B. Field Equations and Background

The IIB field equations can be derived from the Einstein frame action [42]
1 1
Z Z 
10 1/2
d x(−G) R− 2 dΦ ∧ ∗dΦ + e2Φ dC ∧ ∗dC +
2κ2 4κ
g2

ge−Φ H3 ∧ ∗H3 + geΦ F̃3 ∧ ∗F̃3 + F̃5 ∧ ∗F̃5 + g 2 C4 ∧ H3 ∧ F3 , (15)
2
supplemented by the self-duality condition

∗ F̃5 = F̃5 . (16)

5 This estimate (13) ignores the warping of the geometry by the expanded brane, but should be
correct in order of magnitude almost everywhere. In fact, very close to the surface of the two-
sphere the effect of the D5-brane dominates. However, in this regime we can match onto the exact
solution for a flat D5-brane with D3-brane charge, as we develop further in section V.D.

12
Here

F̃3 = F3 − CH3 , F3 = dC2 ,

F̃5 = F5 − C2 ∧ H3 , F5 = dC4 . (17)

We define the Einstein metric by (Gµν )Einstein = g 1/2 e−Φ/2 (Gµν )string , so that it is equal to
the string metric in this constant background. As a result g appears in the action, explicitly
and also through 2κ2 = (2π)7 α′4 g 2.
The field equations are [43]

ge−Φ geΦ
∇2 Φ = e2Φ ∂M C∂ M C − HM N P H M N P + F̃M N P F̃ M N P ,
12 12
geΦ
∇M (e2Φ ∂M C) = − HM N P F̃ M N P ,
6
d∗(eΦ F̃3 ) = gF5 ∧ H3 ,
d∗(e−Φ H3 − CeΦ F̃3 ) = −gF5 ∧ F3 ,
d∗F̃5 = −F3 ∧ H3 ,
1 e2Φ g2
RM N = ∂M Φ∂N Φ + ∂M C∂N C + F̃M P QRS F̃N P QRS
2 2 96
g −Φ
+ (e HM P QHN + e F̃M P Q F̃N P Q )
PQ Φ
4
g
− GM N (e−Φ HP QR H P QR + eΦ F̃P QR F̃ P QR) . (18)
48
We use indices M, N, . . . in ten dimensions. The Bianchi identities are

dF̃3 = −dC ∧ H3
dF̃5 = −F3 ∧ H3 . (19)

One class of solutions is

ds2 = ds2string = Z −1/2 ηµν dxµ dxν + Z 1/2 dxm dxm ,

1
F̃5 = dχ4 + ∗dχ4 , χ4 = dx0 ∧ dx1 ∧ dx2 ∧ dx3 ,
gZ
θ
eΦ = g , C = , (20)
2π
with g and θ constant and other fields vanishing. Here µ, ν = 0, 1, 2, 3, and m, n = 4, . . . , 9.
Also, Z is any harmonic function of the xm , ∂m ∂m Z = 0. For AdS5 × S 5 ,
R4
Z= , R4 = 4πgNα′2 . (21)
r4
This fails to be harmonic at the origin, but this is a horizon, dual to a D3-brane source at
the origin. More generally a nonharmonic Z corresponds to a distributed D3-brane source.

13
 We will need to expand the field equations around this solution. The equations for
linearized F3 and H3 perturbations are conveniently written in terms of

G3 = F3 − τ̂ H3 . (22)

Here

τ = C + ie−Φ , (23)

and aˆdenotes unperturbed fields, so that

θ i
τ̂ = + . (24)
2π g
The linearized field and Bianchi equations in a general background are

∗G3 + iG3 ∧ F̃ˆ 5 = 0 ,

dˆ
dG3 = 0 . (25)

We will only be interested in the transverse (mnp) components of G3 . For the back-
ground (20) and a transverse 3-form field,

∗ˆG3 = Z −1 ∗6 G3 ∧ dx0 ∧ dx1 ∧ dx2 ∧ dx3 . (26)

where the dual ∗6 acts in the six-dimensional transverse space with respect to the flat metric
δmn . Then, in the solution (20) with general Z, the field equation for a transverse 3-form
field can be written simply as

d[Z −1 (∗6 G3 − iG3 )] = 0 . (27)

The duality of the field strengths implies that the 7-form field strength is

− ∗F̃3 = dC6 − H3 ∧ C4 . (28)

This is parallel in form to the other field strengths (17). The relative sign of the two terms
on the right can be deduced by noting that the D5-brane action, which we will write in
section IV.A, and the field strength are both invariant under δC4 = dχ3 provided that
δC6 = −H3 ∧ χ3 . The relative sign of the two sides is obtained by acting with d and
comparing with the field equation (18).
For the 6-form we write
i
d(B6 − τ̂ C6 ) = ∗G3 + C4 ∧ G3 . (29)
g
The imaginary part of this equation is just Eq. (28), while the real part defines B6 ; the
meaning of B6 will become clear in the section IV.B. For the background (20) this becomes
i
d(B6 − τ̂ C6 ) = (∗6 G3 − iG3 ) ∧ dx0 ∧ dx1 ∧ dx2 ∧ dx3 . (30)
gZ

14
 C. Fermion Masses and Tensor Spherical Harmonics

The N = 4 theory has Weyl fermions λα transforming as a 4 of the SO(6) R-symmetry.

We will add a mass term

mαβ λα λβ + h.c. (31)

(spinor indices suppressed), which we can assume to be diagonal, mαβ = mα δ αβ . When one
of the masses, say m4 , vanishes, the Hamiltonian has an N = 1 supersymmetric completion,
as given by the superpotential (3). The fermion λ4 is then the gluino.6
The fermion bilinear transforms as the (4 × 4)sym = 10 of SO(6), and the mass matrix
as the 10. The 10 and 10 are imaginary-self-dual antisymmetric 3-tensors,
1
∗6 Tmnp ≡ ǫmnp qrs Tqrs = ±iTmnp , (32)
3!
with + for the 10 and − for the 10. The indices again run from 4 to 9.
To relate the fermion mass to a tensor, it is convenient to adopt complex coordinates z i :

x4 + ix7 x5 + ix8 x6 + ix9

z1 = √ , z2 = √ , z3 = √ . (33)
2 2 2
Under a rotation z i → eiφi z i the spinors in the 4 transform

λ1 → ei(φ1 −φ2 −φ3 )/2 λ1 ,

λ2 → ei(−φ1 +φ2 −φ3 )/2 λ2 ,
λ3 → ei(−φ1 −φ2 +φ3 )/2 λ3 ,
λ4 → ei(φ1 +φ2 +φ3 )/2 λ4 . (34)

From this it follows that a diagonal mass term transforms in the same way as the form

T3 = m1 dz 1 ∧ dz̄ 2 ∧ dz̄ 3 + m2 dz̄ 1 ∧ dz 2 ∧ dz̄ 3 + m3 dz̄ 1 ∧ dz̄ 2 ∧ dz 3 + m4 dz 1 ∧ dz 2 ∧ dz 3 .

(35)

In N = 1 language, m4 is a gluino mass and the other mα are chiral superfield masses. In
the supersymmetric case the nonzero components are

T12̄3̄ = m1 , T1̄23̄ = m2 , T1̄2̄3 = m3 (36)

6 Evenwhen all four masses are nonvanishing, this operator is still chiral and has a supersymmetric
completion to linear order in m. However, the Hamiltonian at order m2 is nonsupersymmetric.
This case will be discussed in section VII.B.

15
and permutations, and in the equal-mass case

Tı̄̄k = Ti̄k̄ = Tı̄j k̄ = mǫijk . (37)

These satisfy ∗6 T = −iT .

One might guess, correctly, that the fermion mass is associated with the lowest spherical
harmonic of the field G3 [32,44,45]. To make a 3-tensor field transforming in the same way
as any given tensor T , we can use the constant T itself, or combine it with the radius vector
to form
xq m
Vmnp = (x Tqnp + xn Tmqp + xp Tmnq ) (38)
r2
where r 2 = xm xm . Define the forms
1 1
T3 = Tmnp dxm ∧ dxn ∧ dxp , V3 = Vmnp dxm ∧ dxn ∧ dxp ,
3! 3!
1
S2 = Tmnp xm dxn ∧ dxp . (39)
2
One then finds

dS2 = 3T3 , d(ln r) ∧ S2 = V3 , d(r p S2 ) = r p (3T3 + pV3 ) ,

dT3 = 0 , dV3 = −3d(ln r) ∧ T3 , (40)

and

∗6 T3 = ±iT3 , ∗6 V3 = ±i(T3 − V3 ) . (41)

D. Linearized Solutions

We specialize to perturbations on the AdS5 × S 5 case Z = R4 /r 4 , which is invariant

under the transverse SO(6). This will be applicable to the probe calculation of the next
section. A general form for the perturbation is

G3 = r p (αT3 + βV3 ) , (42)

where for now we take T to be an arbitrary constant tensor in the 10 or 10. The Bianchi
identity gives

0 = dG3 = (pα − 3β)d(ln r) ∧ S2 ⇒ β = pα/3 , (43)

corresponding to

16
 G3 = (α/3)d(r pS2 ) . (44)

Using the duality properties (41) we then have

∗6 G3 − iG3 = −ir p (α/3)[(3 ∓ p ∓ 3)T3 + (p ± p)V3 ] , (45)

and so the equation of motion (27) gives

p2 − 10p + (12 ∓ 12) = 0 . (46)

For the lower sign, the 10, there are two solutions:

p = −4 , G3 = αr −4(T3 − 4V3 /3) ,

p = −6 , G3 = αr −6(T3 − 2V3 ) . (47)

In interpreting these, note that a factor Z −3/4 = (r/R)3 must be included to translate
the tensors to an inertial frame. These solutions then have the falloffs appropriate to the
nonnormalizable and normalizable solutions for a operator of ∆ = 3. The former thus
corresponds to the perturbation of m, and the latter to the vev of λ̄λ̄. The mass perturbation
therefore corresponds at first order to
4 " 4 #
ζ R ζ R
 
G3 = (T3 − 4V3 /3) = d S2 (48)
g r 3g r

with T3 given in Eq. (35). The factors of R are necessary for the dimensions, and the factor of
−2
g −1 arises from the overall gYM in the superpotential. The numerical coefficient ζ appearing
in the relation between the fermion bilinear and the supergravity field will eventually be
√
determined to take the value ζ = −3 2. Note also that as a consequence of the equation of
motion (27),

2iζ 2iζ
Z −1 (∗6 G3 − iG3 ) = T3 = dS2 (49)
3g 9g
is exact.
For fields in the 10, the upper sign, there are again two solutions:

p=0, G3 = αT3 ,
p = −10 , G3 = αr −10 (T3 − 10V3 /3) . (50)

The first of these corresponds to the coefficient of λ̄λ̄F 2 , and the second to the vev of λλF 2 .

17
 IV. FIVE-BRANE PROBES

In this section we consider probes in the background given by AdS5 × S 5 plus the linear
G3 perturbation. The probes are 5-branes with world-volume R4 × S 2 and D3-brane charge
√
n ≪ N, with n ≫ gN. We consider first D5-brane probes, and then use SL(2, Z) duality to
extend to a general (c, d) 5-brane. For all such probes we find that there is a supersymmetric
minimum at nonzero AdS radius r.

A. The D5 Probe Action

The relevant terms in the action for a D5-brane are [46,31,47]

µ5 Z 6 h −1/2 Φ/2 ′
i1/2 Z
S=− d ξ − det(Gk ) det(g e G⊥ + 2πα F ) + µ5 (C6 + 2πα′ F2 ∧ C4 ) ,
g
(51)

where

2πα′ F2 = 2πα′ F2 − B2 . (52)

Here Gk is the metric in the R4 directions of the world-volume and G⊥ is the metric in the
S 2 directions, pulled back from spacetime. It is convenient to note that det Gk = Z −2 and
that det F = 12 Fab F ab det G⊥ .
The D3-brane charge of the probe is n, so that
Z
F2 = 2πn . (53)
S2

This is assumed in this section to be small compared to N so that the effect of the probe on
the background can be ignored. If the internal directions are a sphere, rotational symmetry
and the quantization (53) give Fθφ = 21 n sin θ, or Fab F ab = n2 /2Zr 4.
Let us first consider the action in the absence of the G3 background so in particular
Fab = Fab . The first term in the Born-Infeld action is dominated by the second, since for
Z = R4 /r 4

4π 2 α′2 Fab F ab = 2π 2 α′2 n2 /2R4 ∼ n2 /gN ≫ 1 (54)

That the field strength dominates reflects the physical input that the D3-brane charge dom-
inates. It is then useful to write
√
" #
q
′ 1
det(G⊥ + 2πα′ F ) = 2πα det F 1 +
(2πα ) Fab F ab
′ 2

√ det G⊥
= 2πα′ det F + √ . (55)
4πα′ det F

18
If the D5-brane is a sphere in the x⊥ directions, then in spherical coordinates det G⊥ =
Zr 4 sin2 θ. Since a D3-brane probe feels no force from D3-branes, there is a large cancellation
between the Born-Infeld and Chern-Simons terms. The leading nonvanishing term in the
D5-action gives a potential density of the form
q
µ5
Z
2
det Gk det G⊥ µ5
Z
r4 µ5 2r 4
d ξ √ = d cos θ dφ = , (56)
g S2 4πα′ det F g S2 2πα′ n g nα′
where in the last two equations we have assumed the 5-brane is a two-sphere in the x⊥
directions. Notice the Z factors cancel explicitly; if the metric takes the form in Eq. (20),
the energy density of the 5-brane goes as r 4 . This is consistent with the fact that the
D3-branes see this energy as coming from the square of a commutator term, ([Φ, Φ† ])2 .
Now let us add the perturbation back in. For the linear perturbation, Eq. (48) immedi-
ately gives the potentials (up to an irrelevant gauge choice) as
4
ζ R

C2 − τ̂ B2 = S2 . (57)
3g r
For the 6-form, Eqs. (49) and (30) then give
2ζ 0
C6 = dx ∧ dx1 ∧ dx2 ∧ dx3 ∧ Im(S2 ) , (58)
9g
up to gauge choice.
The effect of B2 in the D5-brane action is subleading and can be ignored. Using the
flux (53) and the potential (57), one finds the ratio of the two terms in Fab is

mR4 α′ n mgNα′

′
Bab /2πα Fab ∼ 3 ∼ . (59)
r r2 nr
Looking ahead, the minimum of interest is located at

r ∼ mnα′ , (60)

and so the ratio (59) becomes gN/n2 which is just the small parameter. Thus, at the AdS
radii (60) or greater, the field strength term in Fab dominates: Fab ≈ Fab . The cancellation
between the Born-Infeld and Chern-Simons terms is unaffected; B need merely be inserted
in Eq. (56), where it is negligible.
Inserting the perturbed C6 from Eq. (58) into the D5-brane action gives an additional
potential density
∆S µ5 2ζ
Z
− =− Im(S2 ) . (61)
V g S2 9
which is cubic in r, linear in m, and independent of Z.

19
 The two terms in Eqs. (56) and (61) can be identified with the quartic φ4 and cubic mφ3
terms in the N = 1 supersymmetric potential, as we will see in more detail in section IV.C.
For consistency we must also keep the term of order m2 φ2 . This arises from the second-order
perturbations of the dilaton, metric, and four-form potential. In fact, supersymmetry makes
it possible to write the second-order term in the potential directly:
q
det Gk det G⊥ ζ Z
(Z
S µ5
− = d2 ξ √ − Im(Tmnp xm dxn ∧ dxp )
V g S2 ′
4πα det F 9 S 2

′ 2
)
πα ζ
Z
i ¯l
+ T T F2 z z̄ . (62)
18 i̄k̄ l̄jk S 2

The form of this term is readily understood. The integral F2 essentially sums over D3-
R
S2
branes, while the tensor structure gives the N = 1 scalar mass |mi |2 |φi |2 . The coefficient
P
i
will be deduced in section IV.C.
Before we go on, let us address two puzzles. The first is the expansion around AdS5 × S 5 ,
and why we need to keep terms precisely through second order. A measure of the square of
the size of the perturbation is the ratio of the energy density in the perturbation |F3 |2 with
that in the unperturbed |F5 |2 :

m2 R2 1 m2 gNα′2 gN

2 2
|F3 | /|F5 | ∼ 2 2 2 1/2 2
∼ 2
∼ 2 , (63)
g r g Z r r n
which is the controlling small parameter, basically the effective ratio of brane charge densities
σ52 /σ32 . The three terms in the potential (62) are respectively of zeroth, first, and second order
in the perturbation. The zeroth order term is the remainder after cancellation between the
Born-Infeld and Chern-Simons terms, and, since the D5 and D3 tensions add in quadratures,
is of order
q σ52
σ52 + σ32 − σ3 ∼ . (64)
σ3
The linear perturbation is of order σ5 /σ3 and couples to σ5 , so the first order term is again
of magnitude (64). The second order perturbation is felt by the D3-branes and so this term
is of order σ3 (σ5 /σ3 )2 , again the same. Note that this analysis does not use supersymmetry,
and so will apply to the N = 0 case as well.
The second puzzle is that the second order term in the potential (62) makes reference to
complex coordinates in spacetime, and these are not intrinsic. In particular, when all four
fermion masses are nonvanishing (N = 0) there is no special complex structure. The point7

7 See also section 5 of ref. [37].

20
is that the supergravity equations have homogeneous second order solutions, corresponding
to the traceless scalar bilinear Am An − 61 δmn Ap Ap . The coefficients of these solutions are
determined by boundary conditions, so the inhomogeneous solution with (G3 )2 as source
determines only the trace part Am Am . Thus, the general form for the second order term,
not imposing N = 1 supersymmetry, is given by replacing
¯ r2
+ µmn xm xn
Ti̄k̄ T l̄jk z i z l → Tmnp T mnp (65)
18
with arbitrary traceless µmn . Note that both Tmnp and µmn are intrinsic (determined by the
boundary conditions).

B. The (c, d) Probe Action

A given background can also be given in an S-dual description,

aτ + b
τ′ = . (66)
cτ + d
Specifically,
g ′ = g|M|2 , G′M N = GM N |M| , C4′ = C4 ,
G′3 = G3 M −1 , B6′ − τ̂ ′ C6′ = (B6 − τ̂ C6 )M −1 , (67)
where M = cτ + d. A D5-brane in the primed description has the action
q
det G′k det G′⊥ Z
( )
1
Z Z
− S = µ5 d4 x ′ dξ √2
− C6′ + O(T 2) . (68)
g
S 2 ′
4πα det F S 2

Under the duality (67), this translates into

q
det Gk det G⊥ ζ Z
(
S µ5
Z
− = |M|2 d2 ξ √ − Im(MTmnp xm dxn ∧ dxp )
V g S2 ′
4πα det F 9 S 2

πα′ ζ 2
Z )
i ¯
l
+ T T F2 z z̄ . (69)
18 i̄k̄ l̄jk
The probe couples to
C6′ = −g ′ Im(B6′ − τ̂ C6′ ) = −gIm(M [B6 − τ̂ C6 ]) = B6 c + C6 d . (70)
This is the coupling of a (c, d) 5-brane, a bound state of c NS5-branes and d D5-branes.
In the first term of the potential, the factor |cτ + d|2 is the tension-squared of the (c, d)
5-brane, squared from the addition in quadratures in the Born-Infeld term. The second is
the coupling to the background (70). The final term has again been added by hand in the
form required by supersymmetry, which is in fact independent of (c, d). This is because it is
the interaction of the D3-brane charge with the second-order background, and so does not
depend on the 5-brane quantum numbers. The duality transformation only gives relatively
prime (c, d), but the result holds generally, by superposition.

21
 C. The Probe Potential and Minima

We now focus on the SO(3)-invariant N = 1 equal-mass case. The general SO(3)-

invariant brane configuration is

z i = zei , ei = ei , ei ei = 1 . (71)
√
This is a sphere of coordinate radius |z|/ 2, obtained from the sphere (x4 )2 + (x5 )2 +
(x6 )2 = 12 |z|2 by a simultaneous phase rotation of the z i . Rotational symmetry and the
quantization (53) give Fθφ = 21 n sin θ, or Fab F ab = n2 /2Zr 4. Inserting this configuration
into the action (69) gives

2π 2 nα′ ζ 2
" #
S µ5 8 2 4 8πζ 2
− = ′
|M| |z| + Im(M z̄ mz) + |m|2 |z|2
V g αn 3 9
4
= |Mφ2 + iζmnφ/12|2 . (72)
πgn
Here φ = z/2πα′ is the normalization of the gauge theory scalar relative to the D3-brane
collective coordinate. This is of the form required by N = 1 supersymmetry; the second
order term was normalized to give this result.
For M = 1, the D5-brane, we can compare to the classical N = 1 potential. We can use
the Ansatz
2
Φi = ΦLi , (73)
n
where Φ is a scalar (not a matrix) complex superfield, so that Φi Φi = Φ2 1. The Kähler
P
i
potential and superpotential are then
√
n mn 2 i 2 3
K= ΦΦ , W = Φ + Φ . (74)
2πg 4πg 3πg
√
The potential then agrees with that found in the brane calculation provided ζ = −3 2.
This could be checked by various independent means, such as the fermionic terms in the
D3-brane action in a G3 background.
Returning to general M, there is supersymmetric minimum at
πα′ imn
z= √ . (75)
2M
√
For a D5-brane, (c, d) = (0, 1) and z = iπα′ mn/ 2. For illustration let m be real. The
i reflects the fact that the two-sphere lies in the 789-directions, where F̃3 is maximized.
√
For an NS5-brane, taking C = 0 for convenience, z = πα′mgn/ 2. This is smaller by g,
and lies in the 456-directions where H3 is maximized. Note that the potential in each case

22
has another minimum at z = 0, where the probe has dissolved into the source branes; our
approximation is not valid at z = 0, but it is valid far enough to show that the potential
becomes attractive at small z.
We can also introduce several probes of arbitrary types, and each will independently sit
at the minimum of its own potential. Note that in the AdS geometry we should not think
of these as concentric, but rather arranged along the AdS coordinate r while wrapped at
various angles on equators of the S 5 .
An S 2 on S 5 can be contracted to a point, but it is energetically unfavorable to do so.
The first term in the potential vanishes in this limit (since det G⊥ goes to zero), and the
second does as well, leaving only the positive third term. This is because the pointlike
D5-brane retains only its D3 charge, which feels a positive potential.

V. THE FULL PROBLEM

We now consider the fields and self-energy of the full set of N D3-branes, when these are
in the configuration R4 × S 2 (or a sum of several two-spheres) with 5-brane charges. As an
intermediate step we consider a probe moving in such a background. One might expect these
calculations to be much harder that the previous probe problem, as the symmetry is greatly
reduced. Remarkably, however, all of the work has already been done. The expanded brane
configuration is reflected in a less symmetric warp factor Z, but we will see that this drops
out of all terms in the potential.
In this section we also work out the first-order correction to the background. In addi-
tion we show that our approximation breaks down close to the 5-brane shell, and give the
corrected form.

A. The Warped Geometry

Consider N D3-branes spread on a two-sphere of AdS radius r0 in some 3-plane in the

six transverse dimensions. This Coulomb branch background is again of the form (20), with
the Z-factor given by harmonic superposition. The Z-factor at any point can depend only
on its radii w in the 3-plane and y in the orthogonal 3-plane:
1Z 1 R4
Z= d cos θ 2
2 −1 (w + y 2 + r02 − 2r0 w cos θ)2
R4
= 2 . (76)
(y + [w + r0 ]2 )(y 2 + [w − r0 ]2 )
This is normalized to agree with the AdS Z-factor at large w, y. When the D3-brane charge
is divided among several two-spheres, then Z is a sum of such terms, with total coefficient

23
R4 . At r = 0 this Z goes to a constant, so for w, y ≪ r0 we find flat ten-dimensional
spacetime, with no nontrivial topology.
To next order we consider linearized G3 fields in this background. The field equation is
again

d[Z −1 (∗6 G3 − iG3 )] = 0 , (77)

and the Bianchi identity is dG3 = 0. The origin is now a smooth point and the pertur-
bation will be nonsingular there. It has a specified nonnormalizable behavior at infinity,
corresponding to the perturbation of the gauge theory Hamiltonian, and a specified source
at the 5-branes. Note that this is a magnetic source, appearing in the Bianchi identity but
not the field equation. Note also that

d∗6 [Z −1 (∗6 G3 − iG3 )] = d[−iZ −1 (∗6 G3 − iG3 )] = 0 . (78)

Thus, the combination Z −1 (∗6 G3 −iG3 ) is annihilated by both d and d∗6 . Further, at infinity
it approaches the constant value (49) which is just governed by the boundary condition on
the nonnormalizable solution:
√
−1 2 2
Z (∗6 G3 − iG3 ) → −i T3 . (79)
g
It follows that it takes this constant value everywhere, independent of the warp factor Z
and of the configuration of the brane.
The field G3 itself does depend on the brane configuration, and we will determine it in
section V.C, but it is not relevant here. The brane dominantly couples only to the integral
of the potential B6 − τ̂ C6 which is already determined by Eq. (30) to be independent of Z.
Thus it too is independent of the brane configuration.

B. The Potential and Solutions

Let us consider again a probe, but now moving in the warped geometry just described.
The potential felt by the D3-brane charge of the probe is again zero, for the usual super-
symmetric reasons, so the Born-Infeld and Chern-Simons terms again nearly cancel, leaving
behind the first term in the potential (62). As we noted, this term is independent of Z.
The second term in the potential comes from the coupling to B6 − τ̂ C6 , and we have found
that this too is independent of Z. The third term, given by supersymmetry, must then also
be Z-independent. Thus, a probe feels exactly the same potential in the warped geometry
formed by R4 × S 2 sources, as when all the sources are at the origin.
Now consider the potential felt by the full set of N D3-branes with 5-brane charges.
As is familiar from electrostatics, we cannot simply take the coupling of the branes to their

24
self-field. Rather, we must think of dividing them into infinitesimal fractions and assembling
the configuration by bringing these together one at a time; in electrostatics this produces
the familiar factor of 12 . In the present case, however, there is no ‘charging up’ effect because
as just shown the potential felt by each fractional ‘probe’ is unaffected by the distribution
of the earlier fractions. Thus the potential is the same as in the probe case. If the brane
configuration consists of two-spheres of respective D3-charges nI (with I nI = N), 5-brane
P

charges (cI , dI ), and radii and orientations zI = 2πα′ φI , the potential is

S X 4 √
− = |MI φ2I − imnI φI /2 2|2 . (80)
V I πgnI

Thus, for every collection of 5-branes of total D3-brane charge N there is a solution with
nonzero radii,
πα′imnI
zI = √ . (81)
MI 2
For a D5 sphere this is AdS coordinate radius r = πα′ mnI . For an NS5-sphere it is r =
πα′ mgnI , smaller by a factor g (when C = 0).
It is important to check the validity of these solutions. We have already argued that
for all N D3-branes in a single D5 or NS5 two-sphere the solution is valid in the entire
supergravity regime. Now let us consider the problematic case discussed in section II.B,
namely p D5-branes each of charge q, which is supposed to represent the same state as q
NS5-branes each of charge p. For the former solution, each D5-brane has charge N/p = q
and so the central condition (14) becomes

q2 q
= ≫1. (82)
gN gp
For the NS5-brane solution we can simply interchange g ↔ 1/g and p ↔ q via S-duality to
obtain
gp
≫1. (83)
q
The conditions (82) and (83) are beautifully complementary, so that only one solution is
valid at a time. At weak coupling the state is described by a D5-brane and at strong coupling
by an NS5-brane.
This example also provides the evidence that the condition (14) is a necessity, not a con-
venience: if the solutions persisted beyond this range there would be too many, as compared
to the known vacua of the gauge theory. Thus, we require that for each sphere
nI
≫1. (84)
g|MI |2

25
 It would be extremely interesting to understand the crossover between the D5 and NS5
representations of the above phase. At a minimum this will require the full nonlinear su-
pergravity solutions, but it may involve nonperturbative brane dynamics beyond this. Note
that at the crossover coupling the D5 and NS5 two-spheres have the same AdS radii but
different and nonoverlapping orientations.
There should be a similar story for the minima of the potential at φ = 0. These are out-
side the range of validity of the approximation, and should not correspond to true solutions
because these would again have no duals in the gauge theory. Rather, a 5-brane at small φ
should transmute into a different kind of 5-brane.
As another example consider the oblique solutions (c, d) = (1, s). The condition that
the 5-brane energy density, added in quadratures, be much less than the D3-brane energy
density, is [see Eq. (63)]

s2 N 1/2
1/2
1 1

+ ≪ ⇒ 1 + g 2s2 ≪ gN . (85)
α′ g 4 g 2 g 3/2 α′
For small s this is valid in most of the supergravity regime, but for s ∼ N it is valid nowhere.
This resolves the overcounting, that (1, s) and (1, s + N) represent the same state. Note that
for s ≫ 1 there is a range of g where supergravity is valid but the (1, s) brane solution is not;
the SL(2, Z) duality (which acts on these vacua in an intricate way) gives other candidate
brane configurations.
There is one final issue connected with the stability of the brane solutions. Let us
focus on the D5-brane. At opposite points on the two-sphere, the D5 world-volumes are
antiparallel. Intuition from flat space D5-branes [31] would suggest that this configuration
is not supersymmetric, but this must be wrong. The supersymmetry transformation related
to the D5 charge must be offset by the effect of the background on the much larger D3
charge.
We leave the analysis of supersymmetry for the future, but do address a related point:
the self-force of the D5-brane. Again, intuition suggests that there should be an attractive
force between opposite sides of the two-sphere, rendering the state unstable, but if the
configuration is supersymmetric then this must vanish. Let us see how this works. In the
D5-brane action (51), the strongest couplings to bulk fields are those of the D3-brane charge
to Gk and to C4 . The self-force from these cancels as usual due to the supersymmetry of
D3-branes. The next strongest coupling is of the D5-brane charge to G3 . It is this that
might give an attractive force, but in fact it does not: Eq. (30) shows that the field sourced
by the D5-brane does not act back on the D5-brane. The C4 background induces mixing

26
between F3 and H3 in such a way that the self-force cancels for any orientation!8 Finally,
the dilaton and metric couple to the quadrature term; this is second order in σ5 /σ3 , and so
the exchange force would be fourth order. In the supersymmetric case this should actually
vanish, but because it is in any event small we will not show this. Moreover, even for a
nonsupersymmetric perturbation the arguments for the vanishing of the forces from Gk , C4 ,
and G3 continue to hold, so only the small residue from the dilaton and G⊥ remains. This
is too small to destabilize the solution, as the potential (80) is a second order effect.

C. First Order G3 Background

Here we work out the first order correction to the background, which appears only in the
field G3 . In addition to the earlier result (30),
√
2 2
∗6 G3 − iG3 = −i ZT3 , (86)
g
we have the Bianchi identity with magnetic source,

dG3 = J4 . (87)

Let us adopt a coordinate system in which the brane is a sphere of radius r0 in the w 1,2,3
directions and at the origin in the y 1,2,3 directions. Then

J4 = 4π 2 α′ Mδ 3 (y)δ(w − r0 )dw ∧ d3 y , (88)

where w is the radius in the w-plane, d3 y = dy 1 ∧ dy 2 ∧ dy 3, and the factor 4π 2 α′ arises as

2κ2 µ5 /g 2 . Note that the quantum numbers M appear in a simple way. In place of Eq. (86),
we can use its exterior derivative,
√
2 2
d∗6 G3 = iJ4 − i dZ ∧ T3 . (89)
g
This and the Bianchi identity determine G3 ; they can be solved in terms of potentials.
Write

G3 = ∗6 dω2 + idω2 + dη2 (90)

8 This might seem to contradict claims that there is a large-N limit of AdS space which gives flat-
spacetime physics [48], since nonparallel D5-branes do attract in flat spacetime. The point is that
this large-N limit includes going to small AdS distances. This would bring us into the ‘near-shell’
region of the D5-brane (to be discussed in section V.D), where the above no-force analysis does
not apply.

27
with the gauge choice

d∗6 ω2 = d∗6 η2 = 0 . (91)

Then
2π 2 α′M 3
∂m ∂m ω2 = ∗6 J4 = δ (y)δ(w − r0 )ǫijk w i dw j ∧ dw k ,
r0
√ √
2i 2 2
∂m ∂m η2 = − ∗6 (dZ ∧ T3 ) = − Tmnp ∂m Zdxn ∧ dxp . (92)
g g
The solutions are
α′ M 1 y 2 + w 2 + r02 + 2r0 wt 
 
i j k
ω2 = − 3 ǫijk w dw ∧ dw ∂t ln ,
4w t y 2 + w 2 + r02 − 2r0 wt t=1
R4 1 y 2 + [w + r0 ]2
 
n p
η2 = √ Tmnp dx ∧ dx ∂m ln . (93)
8gr0 2 w y 2 + [w − r0 ]2

These do not seem very enlightening, but we can obtain their forms at large r 2 :

8α′ Mr03
ω2 ≈ − ǫijk w idw j ∧ dw k ,
3r 6
R4
η2 ≈ − √ 4 Tmnp xm dxn ∧ dxp . (94)
g 2r
These scale as the normalizable and nonnormalizable solutions respectively. The latter, η2 ,
matches the boundary condition (48).

D. The Near-Shell Solution

Our small parameter guarantees that our solution is good over most of spacetime, but it
must break down as we approach the 5-brane shell. The metric in the directions parallel to
the 5-brane and orthogonal to the D3-branes expands, diluting the D3-brane charge so that
close to the 5-brane it no longer dominates. One also sees this in the ratio of energy densities,
where the metric has the same effect. Since this occurs only close to the 5-brane, we can
approximate the solution in this region by a flat 5-brane+D3-brane solution. Specializing
to p D5-branes,9

9 See for example Eq. (6), and for the NS5 brane Eq. (35), of Ref. [49]. Note that these equations
arise after taking the limit where a noncommutative gauge theory describes the 5-brane dynamics.
We will return to this issue briefly in our conclusions.

28
 α′ u h i α′ agp
ds2string = ηµν dxµ dxν + h(dx̃4 dx̃4 + dx̃5 dx̃5 ) + (du2 + u2 dΩ23 ) ,
agp u
2 2
au
e2Φ = g2 , ds2 = g 1/2 e−Φ/2 ds2string , (95)
1 + a2 u2
where

h = (1 + a2 u2 )−1 . (96)

Let us compare with the near-shell metric based on the harmonic function (76), near the
point (w1 , w2 , w3 ) = (0, 0, r0):

2r0 ρ R′2
ds2string = ηµν dxµ
dxν
+ (dw · dw + dy · dy) , (97)
R′2 2r0 ρ
where

ρ2 = (w3 − r0 )2 + y 2 . (98)

We have also defined R′4 = 4πgnα′2 to include the case that the shell does not carry the
full D3 charge N; we do not assume that n is small. The metrics agree away from the shell,
au ≫ 1, provided that

ρ R′2 R′8
u= , a= , x̃4,5 = w 1,2 . (99)
α′ 2gpr0 16g 3p2 r04 α′2

With these identifications, the solution (95) gives the continuation to au < 1. As a check,
the crossover distance au = 1 is

′ −1 2gpr0α′ pg 1/2
ρc = α a = ′2
∼ 1/2 r0 ∼ pm(gnα′ )1/2 . (100)
R n
Thus the shell is indeed thin: ρc is smaller than the radius r0 by p(g/n)1/2 , which is precisely
our controlling parameter (82) for the D5 solution. As a reminder, r0 = mπα′ n/p for this
shell. In summary, the components of the metric tangent to the two-sphere, and the dilaton,
are multiplied by a factor ρ2 /(ρ2 + ρ2c ),

2r0 ρ R′2 ρ R′2

ds2string = ηµν dxµ
dxν
+ (dw 1
dw 1
+ dw 2
dw 2
) + (dw 3dw 3 + dy · dy) ,
R′2 2r0 (ρ2 + ρ2c ) 2r0 ρ
ρ2
e2Φ = g2 2 , ds2 = g 1/2 e−Φ/2 ds2string . (101)
ρ + ρ2c

This interpolates between the D3- and D5-brane metrics.

Similarly for q NS5-branes, the solution interpolates between the D3- and NS5- solutions.
The crossover radius is now

29
 2r0 qα′ q
ρ′c = ′2
∼ 1/2
r0 ∼ qm(gnα′ )1/2 , (102)
R (gn)
the AdS radius is r0 = mπα′ gn/q for this shell, and the solution is
2r0 (ρ2 + ρ′2
c )
1/2
R′2
ds2string = ηµν dx µ
dx ν
+ (dw 1 dw 1 + dw 2 dw 2 )
R′2 2r0 (ρ2 + ρ′2
c ) 1/2

′2 2 ′2 1/2
R (ρ + ρc )
+ 2
(dw 3 dw 3 + dy · dy) ,
2r0 ρ
2 ′2
ρ + ρc
e2Φ = g2 2
, ds2 = g 1/2 e−Φ/2 ds2string . (103)
ρ
For ρ < ρ′c the metric develops the usual throat for q NS5-branes [50]. The string coupling
becomes strong at ρ/ρ′c ∼ g, a proper distance ln 1/g from the crossover region.
It is important to see where the supergravity solution is valid. A crude but simple
measure is that the radius of a transverse sphere (fixed ρ) must be large in string units. (We
assume g ≤ 1 so that the F-string scale is the relevant one.) At the crossover point, the D5
and NS5 radii-squared are respectively

gpα′ , qα′ . (104)

The NS5 solution is valid for q ≫ 1 and marginal for q = 1 (these properties continue to
hold down the throat, until the dilaton diverges). The D5 solution has a limited range of
validity for p ≫ 1 but none for p = 1 (not even g ≫ 1, because the dual string theory is
strongly curved). Thus the low energy physics of the Higgs phase is given by the dual field
theory description.

E. The Complete Metric and Dilaton

The pieces of our solution are scattered through this paper. The zeroth order solution
is the D3-brane background (20) with harmonic function (76), with the brane locations and
orientations (81). The first order correction is given by Eqs. (90) and (93). The correction
near the brane is given in Eqs. (101) and (103). For convenience we give here the full
solution for the metric and dilaton in a form that interpolates between the zeroth order
solution and the near-shell solution. We emphasize that these have overlapping ranges of
validity, ρ > ρc , ρ′c versus ρ < r0 .
We focus on a single shell of D5 or NS5 type, but the generalization is straightforward.
The solution is be conveniently written using coordinates xµ for spacetime, w i for the three
coordinates in which the brane is embedded, and y i for the other three. Write w, Ωw as
spherical coordinates for the w i , and similarly for the y i. Both the Higgs and confining
metrics, in string frame, can be conveniently written

30
 1/2
Zx−1/2 ηµν dxµ dxν + Zy1/2 (dy 2 + y 2 dΩ2y + dw 2 ) + ZΩ w 2 dΩ2w . (105)

For the Higgs (D5) vacuum, the w i are x7,8,9 and the y i are x4,5,6 ; for the confining (NS5)
vacuum at θ = 0 this is reversed.
For the D5 brane we have
#2
R4 ρ2−
"
Zx = Zy = Z0 ≡ 2 2 , ZΩ = Z0 , (106)
ρ+ ρ− ρ2− + ρ2c

where
2gr0 α′
R4 = 4πgN , ρ± = (y 2 + [w ± r0 ]2 ) , ρc = , r0 = πα′ mN . (107)
R2
The dilaton is
ρ2−
e2Φ = g 2 . (108)
ρ2− + ρ2c

For the NS5-brane, we have

ρ2− ρ2− + ρ2c

Zx = ZΩ = Z0 , Zy = Z0 , (109)
ρ2− + ρ2c ρ2−

where
2r0 α′
ρc = , r0 = πα′ mgN . (110)
R2
Meanwhile the dilaton is
ρ2− + ρ2c
e2Φ = g 2 . (111)
ρ2−
√
Note ρc = mR2 /2 ∼ m gNα′ for both branes.

VI. GAUGE THEORY PHYSICS

In this section, we consider some of the non-perturbative objects in the field theory —
strings, baryon vertices, domain walls, condensates, instantons and glueballs, — and discuss
their appearance in the supergravity representation. Although objects of this type have
appeared in a number of previous incarnations [10–12,51,13,52,53], they arise here in novel
forms. We will also consider a vacuum with massive fundamental matter and mention some
of its amusing properties.

31
 A. Flux Tubes: A First Pass

Many of the vacua of the N = 1∗ field theory have stable flux tubes. At weak coupling,
the Higgs vacuum, where the SU(N)/ZN gauge group is completely broken, has semiclassical
vortex solitons in which certain components of the adjoint scalars wind at infinity. The
topological charge associated with this winding takes values in π1 (SU(N)/ZN ) = ZN ; it
measures the magnetic flux carried by the vortex. The confining vacuum has electric flux
tubes carrying flux in the ZN center of SU(N). These become semiclassical solitons in the
S-dual description of the theory as τ → 0. Similar statements apply for the oblique confining
vacua. In the other massive vacua [5] there are both electric and magnetic flux tubes, and
in the Coulomb vacua there may or may not be any stable flux tubes. We will return to
these cases in a later section. For the moment we focus our attention on the strings of the
Higgs and confining vacua.
One of the surprising features of Maldacena’s duality is that it relates string theory to
a conformal rather than a confining gauge theory. Unconfined electric flux lines between
two charged sources in the conformal N = 4 field theory are represented by a string in the
gravity dual [54,55]. The string in question droops into the AdS5 space, rather than lying
at a fixed AdS radius r. Since small r corresponds to large distances in the field theory, the
drooping string represents flux lines which spread out in the region between the sources, as
expected in a nonconfining theory. The symmetries of AdS space suffice to show that the
energy of the string scales as a constant plus a term inversely proportional to the separation
of the sources.
In the realization of confining gauge theories via high-temperature five-dimensional field
theories [10–12], the temperature provides an IR cutoff on r. The flux between two charged
sources in the field theory now is represented by a string which droops only part way into
the AdS5 space, becoming stuck at a radius of order the temperature R2 T ; consequently the
string represents flux lines trapped in a physical string-like object, of definite tension and
width. In this way the confinement of this theory, which is hoped to be in the same univer-
sality class as asymptotically free Yang-Mills theory, was established. The same happens in
our dual description of N = 1∗ gauge theory.
Before treating the supergravity picture carefully, we begin with an intuitive argument.
Let us assume, as we will shortly show, that a (p, q) string, with its world-sheet oriented
in the xµ directions, can bind to a (p, q) 5-brane with D3-brane charge, in a state of finite
width and nonzero tension. We claim that this object is a confining flux tube of the gauge
theory; since its AdS radius is by construction constant, it certainly has a definite tension.
Let us consider p = 0, q = 1, the Higgs vacuum. The potential between charged electric
sources, given by suspending a fundamental string from two points on the AdS boundary, is

32
highly suppressed: the string can split into two strings joining the D5-brane to the boundary,
meaning there is little energy cost to moving the endpoints of the string apart. By contrast,
a D1-brane cannot end on the D5-brane. However, it can link up with our putative D1-
D5/D3 bound state. This makes the potential between two magnetic sources linear in the
distance between them, with a coefficient set by the tension of the bound state. Note also
that any (p, q) string with q 6= 0 is similarly confined — its p F1 charges ending on the D5,
its q charges connected to q flux tubes (or a bound state of such tubes) on the D5 brane. It
follows that monopoles and dyons, represented by strings with D-charge, are confined in the
Higgs vacuum, while electric charges are screened. This is as expected on general grounds
from the field theory.
By S-duality, the confining vacuum sports F1-NS5 bound states. All strings except
those having only D1-charge will bind to the D3-NS5-brane. These bound states are the
electric flux tubes of the gauge theory. In this vacuum it is fundamental string charge
which is confined and D-charge which is screened, in agreement with expectations. Similar
conclusions hold in the oblique confining vacua.
We now turn to the supergravity description of this physics, and demonstrate that these
bound states truly exist. In our solutions the function Z, given in Eq. (76), diverges at the
branes, so all strings can lower their tensions by drooping inward toward one of the branes.
However, we have seen that there is a crossover point near each brane, where the universal
D3-brane behavior ceases to hold and 5-brane behavior takes over. An F-string, representing
electric flux, couples to the string metric. The string stretches in a noncompact direction,
so the relevant metric component is Gµν . In the D5-solution (101) this still goes to zero at
ρ = 0, so electric flux is unconfined. In the NS5-solution (103) it takes the minimum value
r02 q/πgnα′ = πα′ m2 gnq, so for the confining phase, where n = N and q = 1, the tension is
1 m2 gN
τe = πα′ m2 gN = . (112)
2πα′ 2
This satisfies ’t Hooft scaling, as expected in a confining vacuum. The F-string lowers its
tension, but only by a finite amount, by binding to the NS5-brane.
A D-string, representing magnetic flux, couples to e−Φ times the string metric. For the
NS5-brane this now vanishes at the brane,10 but for the D5-brane there is a minimum value

10 This ‘magnetic screening’ is required both by physical intuition and by S-duality, but notice that
it requires that the string coupling diverge in the NS5-brane throat. For multiple coincident NS5-
branes, this can be seen in supergravity alone. For one NS brane, however, the very nature of the
throat is in dispute [56] and it is not clear whether supergravity, worldsheet CFT, or semiclassical
brane physics gives a good description. In any case, the D-string must dissolve in the NS5-brane,
one way or another.

33
r02 p/πnα′ = πα′ m2 nq. The Higgs phase magnetic tension is then

1 m2 N
τm = πα′ m2 N = . (113)
2πα′ g 2g
The g and N scaling appears to be the same as for a classical Nielsen-Oleson vortex. The
action scales as N 3 /g, and the change in the field, which appears squared, is presumably of
order 1/N for a ZN vortex. The D-string is bound to the D5-brane.
Though satisfying, these results are partly outside the range of validity of the supergravity
description. We have seen in section V.D that for D5-branes this description breaks down
before the crossover point, while for NS5-branes it is marginal (we assume g < 1; for g > 1
the S-dual is true.)
Let us discuss the bound state more carefully in the D5 case, by considering the limit in
which the D5-brane is flat. Note first that D1-branes outside a D5-brane are BPS saturated
and not attracted to the D5-brane. By contrast, D1-branes outside and parallel to a set
of D3-branes are attracted to the D3-branes; upon reaching the D3-branes they appear
as tubes of magnetic flux inside an N = 4 gauge theory, which, since flux is unconfined,
expand to infinite radius. Combining the D5 and D3 branes, the D1-brane is attracted to
the D5/D3 object but upon reaching it cannot expand to arbitrary size. Its behavior within
the D3-brane field theory is determined by the semiclassical calculation of the vortex soliton
which confines magnetic flux. Alternatively, it should be a semiclassical instanton of the
noncommutative field theory on the 5-brane [57].
Another intuitive way to see the bound state is to use T -duality. Begin with a D1-brane
extended in the 01-directions, a 012345 D5-brane, and 0123 D3-branes which are distributed
in the 45-directions with density σ. A T -duality in the 5-direction converts the D1-brane
into a 015 D2-brane and the D5/D3 system into a D4 brane, which fills 0123 plus a line in
the 45-plane. The line makes an angle θ with the 5-axis, where

tan θ = (4π 2 α′ σ)−1 . (114)

If θ = π/2 (σ = 0), the D2 and D4 are perpendicular and BPS, so there is no force on
the D2. If θ = 0 (σ → ∞), then the D2 can be absorbed by the D4. But if 0 < θ < π/2
the D2 brane is attracted to the D4 but is misaligned with it, and so cannot be completely
absorbed. Instead, only a part of it is absorbed, leading to a D2-D4 bound state of finite
energy and size. As shown in figure 1, the tension is reduced by a factor sin θ.

34
 x5
x4

FIG. 1. D4/D2 system projected on the 45-plane. The D4-brane, which is also extended in
0123, wraps multiple times (tan θ). The D2-brane, which is oriented in the 05 directions, can partly
dissolve, leaving a piece connecting two leaves of the D4-brane.

In fact, this bound state is supersymmetric: as one sees from figure 1, after the flux
dissolves to the maximal extent, the remaining state is a D2-brane ending on a D4-brane, a
familiar supersymmetric configuration. The BPS bound [31] is
h i1/2 h i1/2 τD2
τ ≥ (V τD4 + τD2 )2 + V τD4
2
cot2 θ 2
− V τD4 2
+ V τD4 cot2 θ = , (115)
sin θ
where V is a large regulator volume in the 123 directions.
Having established that a BPS state exists in this limit, we should now verify that a
nearly-BPS state of essentially the same mass is still present when the D5-brane has the
shape of a two-sphere. The D1-D5/D3 bound state is in a rather difficult region of parameter
space because the effect of gravity is large (because the D3 charge is large) but the gravity
description is not valid everywhere (because the D1 and D5 charges are small). At large r
the supergravity description is valid, while at small r the effective description is the field
theory on the brane, as in examples in ref. [58]. It appears that a correct treatment requires
that we match these two descriptions, in the spirit of the correspondence principle [59]. By
the logic of section V.D, the crossover between the two descriptions occurs at a radius
α ′ r0
ρ̂ = η . (116)
R2
We will see that it is interesting to retain an undetermined constant η in the crossover point.
At the crossover, the metrics in the 0123 and the 45 (w 1,2) directions are
2ηα′ r02 µ ν R4
ηµν dx dx + (dw 1 dw 1 + dw 2 dw 2) . (117)
R4 2ηα′r02
The area of the two-sphere is then
R4 8π 2 gNα′
4πr02 = , (118)
2ηα′r02 η

35
giving
η
4π 2 α′ σ = . (119)
2g
Combining the D1 tension, the rescaling of Gµν , and the effect shown in figure 1 gives the
tension
1 2ηα′r02 2g m2 N
τm = = . (120)
2πα′ g R4 η 2g

This is the same as the estimate (113) which came from the purely gravitational picture;
it is independent of the precise crossover η (a necessary, though not sufficient, condition for
correspondence arguments to give a correct numerical value); and, one gets the same result
if one ignores the gravitational effect entirely and takes unit coefficients in the metric (117).
It has been suggested that the ZN strings of supersymmetric QCD might be nearly BPS
saturated in the limit of infinite N. In N = 1∗ this hope is realized, although we see that
large gN is necessary for this to be the case. But we have not yet explained why the strings
carry charges which are conserved only mod N. To do so, we turn to the construction of
the baryon vertex.

B. Baryon Vertex

To put N sources in the fundamental representation into a gauge invariant configuration

requires a baryon vertex. In the AdS5 ×S 5 supergravity dual of N = 4 Yang-Mills this vertex
is given by a D5-brane which wraps the entire S 5 [51,13]. We will see that in N = 1∗ , by
contrast, the baryon vertex is a D3-brane with the topology of a ball B 3 , whose boundary is
the two-sphere of the 5-branes which form the vacuum. The link between these two pictures
is the Hanany-Witten brane-creation mechanism [60].
One way to derive the nature of the baryon vertex in the N = 1∗ theory is to begin
in the ultraviolet. The ultraviolet theory is N = 4 supersymmetric and the spacetime
is approximately AdS5 × S 5 . Let us consider the confining vacuum, represented by an
NS5-brane two-sphere. A baryon vertex joining charged sources in a small spatial region
corresponds to a D5-brane wrapping the S 5 near the AdS5 boundary, with N fundamental
strings joining the boundary and the D5-brane. Now let the region containing the sources
grow comparable to the IR cutoff distance m−1 (we will give a more precise estimate in
section VI.H). The AdS5 radius of the D5-brane decreases until it eventually crosses the
NS5-brane. Since drawing the sources further apart than m−1 should lead to a large energy
cost, something dramatic must happen at this scale. And indeed, it does: the crossing of
the D5- and NS5-brane produces a D3-brane which connects the two.

36
 To see this, consider the configuration more carefully. The brane-creation process is
local, so let us consider a nearly-flat portion of the NS5-brane, which extends in the 12345-
directions. The D5-brane locally extends in the 45678-directions. The distance vector be-
tween the two branes lies in the 6-direction. In this arrangement, the crossing of the branes
leads to the creation of a D3-brane which fills the dimensions 456. The transition is shown
in figure 2.

S5
S5
S2 S2

a) b)

S2

c)
FIG. 2. a) A small baryon vertex, at r > r0 : the S 5 baryon vertex is outside the S 2 of the
vacuum brane (the vacuum brane is also extended in 123, while the baryon is not). b) The baryon
S 5 contracted to r < r0 : a 3-brane (shaded) has been created. c) The S 5 has contracted to nothing,
leaving a 3-brane baryon filling the S 2 .

Looking globally at the two-sphere, we see that the D3-brane fills the part of the NS5-
brane two-sphere which lies outside the D5-brane. But the space inside the NS5-brane is
topologically flat; the radius of the five-sphere shrinks to zero inside. The D5-brane therefore
is topologically unstable and can be shrunk to zero radius, leaving a D3-brane which fills the
entire two-sphere inside the NS5-brane. Like the D5-brane which created it, this D3-brane
is a particle in the four-dimensional spacetime; more precisely, it is a localized object whose
size is of order m−1 . If the charged sources are taken to lie further apart than this, then they

37
will connect to the D3-brane not directly but through F1-NS5 flux tubes; thus the baryon
vertex behaves dynamically as we would expect in a confining theory.
The D5-brane baryon vertex in the N = 4 theory has no preferred size or energy. Here,
the D3-brane actually represents a physical excitation of definite size and mass. The mass
is
µ3 Z 1/2 µ 3 R 2 Z r0 4πw 2
d3 x e−Φ Gstring = dw 2
g B3 g 0 (r0 + ρ′2 2
c −w )
√
m gN
≈N ln(gN) . (121)
2π 3/2
Note that this diverges, due to a net factor Z 1/2 in the integrand, until the near-shell form is
taken into account. The result is a factor of N times ’t Hooft scaling, as would be expected.
To see directly that the D3-ball is a baryon vertex, note that the NS5-brane world-volume
action includes a Chern-Simons term
Z
F2 ∧ F2 ∧ B2 . (122)

This is the S-dual of the term

Z
F2 ∧ F2 ∧ C2 (123)

in the D5-brane action, which is familiar as it implies that a world-volume gauge instanton
is a dissolved D1-brane. The D3-brane ending on the NS5-brane is a magnetic monopole
source for F2 (the S-dual of a familiar fact for D3- and D5-branes), while the dissolved
D3-branes become N units of F2 . Under δB2 = dχ1 ,
Z Z
δ F2 (monopole) ∧ F2 (dissolved) ∧ B2 = − dF2 (monopole) ∧ F2 (dissolved) ∧ χ1 . (124)

This violation, proportional to the number of dissolved D3-branes, must be offset by N

fundamental strings ending at the D3-NS5 junction.
Note that this now also explains the ZN quantum numbers of the flux tubes. If we place
N flux tubes close and parallel to each other, pair-creation of these D3-branes can occur.
This allows the flux tubes to annihilate in groups of N.
Application of S-duality allows us to form the same construction for other similar vacua.
Notice that the baryon vertex is always a D3-brane. However, the 5-brane on which it ends
determines its properties. For example, if we are in the Higgs vacuum, the magnetic flux and
its magnetic baryon form the S-dual of what we just considered. By contrast, the electric
baryon is completely shielded: the N fundamental string sources, which in the confining
vacuum were forced to end on the D3-brane, are no longer forced to do so, since they may
end anywhere on the D5-brane. This is of course consistent with field theory expectations.

38
 Now let us consider some other vacua. Suppose that we take a vacuum where the classical
√
unbroken gauge group was SU(N/k), with k ≪ N a small divisor of N. Since only the
SU(N/k) confines, and since a fundamental representation of the SU(N) parent breaks
up into k copies of the fundamental representation of SU(N/k), we should expect that N
sources would now be joined by not one but k different baryon vertices. To see this in the
supergravity is straightforward. The relevant vacuum is given by k coincident NS5-branes,
so when the D5-brane baryon vertex of N = 4 crosses the NS5-branes, k D3-branes are
created. Each of these carries N/k units of string charge (since each NS5-brane has N/k
units of D3-brane charge) and so N/k strings must end on each of them. On the other hand,
the k D3-branes are not bound together and may be separated spatially from one another.
Each one represents a separate, dynamical, massive baryon vertex of SU(N/k). Note also
that pair creation of these objects ensures that the electric flux tubes in this vacuum carry
only ZN/k quantum numbers.

C. Flux Tubes: A Second Pass

Here we will look at Coulomb vacua to understand how the baryons and strings behave,
and obtain the correct flux tube quantum numbers.
We have already noted the flux tubes present when the vacuum is massive — that is,
when the classical vacuum is given by k copies of the N/k-dimensional representation of
SU(2). The baryons ensured that the electric flux tubes carry flux in Zk and the magnetic
or dyonic flux tubes have charge in ZN/k .
Let us consider instead a general Coulomb vacuum, given by choosing pi copies of the qi -
dimensional representation, with pi qi = N. The unbroken gauge group is [U(p1 )×U(p2 )×
P

... × U(pk )]/U(1), corresponding to pi D5-branes of AdS radius proportional to qi . Let r to

be the greatest common divisor of the pi , and s = gcd(qi ). Simple field theoretic arguments
then determine the properties of possible flux tubes. The topology of the breaking pattern
of the gauge group permits magnetic flux tubes to carry a Zs charge, while the massive
W -bosons of the theory will break all electric flux tubes down to those carrying a Zr charge.
How do we see these flux-charges in the supergravity? For the magnetic flux, it is
straightforward. Consider a collection of k magnetic flux tubes. A magnetic flux tube can
be moved with impunity from one D5-brane to another, since two flux tubes on different D5-
branes can be connected by a D1-string in the radial direction, corresponding to a magnetic
gauge boson. A magnetic baryon vertex connecting to a D5-brane of radius q1 can remove
or add q1 flux tubes from the k that we started with. Since we may move all the flux tubes
from the first group of D5-branes to the second group, we may also remove any multiple of
q2 flux tubes from our collection. Removing qi flux tubes in any combination, we are left

39
with a number k̂ with 0 ≤ k̂ < gcd({qi }). This confirms what we set out to prove.
To see the charges of the electric flux tubes requires S-duality, which in not understood
for the general Coulomb vacuum. However, we conjecture that the τ → −1/τ transformation
acts in a simple way in an important subclass of the vacua. In particular, consider those
classes of vacua where all {pi } are distinct integers and all {qi } are distinct integers. In
this case we claim that the S-dual of this vacuum is that with qi NS5-branes of radius pi .
This is of course consistent with the known transformation of the massive vacua [5], for
which p1 q1 = N. The S-dual of the argument in the previous paragraph then shows that
the electric flux tubes for these vacua is indeed Zr . Indeed, this is our main evidence for the
conjecture.
If the pi or the qi are not distinct integers, then the S-duality transformation we have
suggested is ambiguous. We do not know what happens in this case, either in field theory
or in supergravity.

D. Domain Walls

Since the theory has many isolated vacua, it also has a large number of domain walls
which can separate two spatial regions in different vacua. If the walls are spatially uniform
then they may be BPS saturated [61,62].
Between the oblique confining vacuum represented by a (1, 1) 5-brane sphere and the
confining vacuum represented by an NS5-brane, there must be a BPS domain wall which
carries off one unit of D5-brane charge. We may therefore conjecture that a BPS junction
of three 5-branes — the NS5-brane sphere for x1 > 0, the (1, 1) 5-brane sphere for x1 < 0,
and a D5-brane at x1 = 0 which fills the two-sphere — describes this domain wall. That
is, the world-volume of the D5-brane is the 023-plane of the domain wall times the three-
ball spanning the two-sphere. At small g the NS5-brane and the (1, 1) brane are nearly
coincident (their AdS radius and orientation on the S 5 differ only at order g) and the effect
of the D5-brane on the much denser NS5-brane is very small.
We can see that this reproduces some known properties of the domain wall. First [61,62],
the flux tubes of the theory (F1 strings) obviously can end on the domain wall (a D5-brane).
Furthermore, consider dragging a (1, 1) dyonic string, representing a dyonic source in the
gauge theory, across the wall. For x1 < 0, the dyon is screened; it can end happily on the
(1, 1) 5-brane. For x1 > 0, the dyon is confined; its monopole charge ends on the NS5-brane,
but its electric charge must join onto a flux tube — an F1-NS5 bound state — which in turn
ends on the D5-brane domain wall. Finally, note that we may dissolve an N-string vertex
(a D3-brane) into this domain wall (a D5-brane), leaving N strings which end on the wall
and are free to move around on it. If we then permit this domain wall to annihilate with

40
an antidomain wall with no strings attached, then the annihilation will leave a D3-brane
behind on which the strings may end, as in the well-known process described in [63].
More generally, if for x1 < 0 the system is in the phase corresponding to a (c, d) 5-brane,
and for x1 > 0 it is in the phase corresponding to a (c′ , d′) 5-brane, then a (c − c′ , d − d′)
5-brane must fill the 2-sphere where they meet. In general the branes on the right and left
have different orientations and radii, and so must bend as they meet as depicted in figure 3.

z z z
C x1
B

FIG. 3. Triple 5-brane junction, corresponding to a domain wall. Depicted is z(x1 ); the full
geometry is obtained by translating in x2,3 and rotating in the transverse SO(3) symmetry. At
x1 < 0 the system is in the vacuum corresponding to 5-brane A; at x1 > 0 it is in the vacuum
corresponding to 5-brane B, with different radius and orientation. The domain wall lies in the
shaded plane x1 = 0 and is a 5-brane of type C. The bending of branes A and B is described by
the BPS equation (135).

When the left and right phases involve multiple spheres, there will be a more complicated
domain wall, constructed from multiple triple 5-brane junctions.
To discuss the domain wall tensions quantitatively we need the kinetic term for the
collective coordinate z = 2πα′ φ. This arises in the Born-Infeld action, from

Gµν (induced) = Gµν + Gmn ∂µ xm ∂ν xn . (125)

Then
S µ5 n µν
Z
1/2
= − 2πα′ d2 ξ G⊥ (Fab F ab )1/2 η µν ∂µ xm ∂ν xn = − η ∂µ φ̄∂ν φ . (126)
V 2g 2πg
This gives the Kähler potential K = nΦ̄Φ/2πg. This is the same as Eq. (74) for the classical
gauge theory, but by an SL(2, Z) transformation one can show that it holds for all (c, d).
This makes sense, as the main kinetic effect comes from the D3-branes, which are self-dual.
With this normalization the potential (80) implies the superpotential
√
1 4 2 3
W = (i Φ + mnΦ2 ) . (127)
4πg 3M

41
as in Eq. (74). At the nonzero stationary point this takes the value

m3 n3
W →− . (128)
96πgM 2
For a multi-brane configuration it is summed over I.
We cannot rule out an additional additive contribution to this superpotential, although
from our semiclassical reasoning we know that any such contribution must be subleading in
the Higgs vacuum and others containing only large D5-branes.11 Field theory also suffers
from the same ambiguity [8]. Up to these additive contributions, the field theory and
supergravity agree. Consider the massive vacuum corresponding to p D5 branes of radius q;
for this vacuum M = p. Using [7], it is easy to obtain a slight generalization of [8] (adjusted
to match our conventions, and with the function A(τ, N) in Eq. (5) of [8] set to zero)

m3 N 2 m3 N 2 E2 (τ ) m3 N 3
" !#
q q
W = 2
E2 (τ ) − E2 τ → − . (129)
24gYM p p 96πg 96πgp2

Here E2 is the second Eisenstein series, and we have used Eq. (82) and E2 (i∞) = 1. Note
√
the first term is p independent and is subleading for p ≪ N. For the massive vacuum
given by q (1, k) 5-branes of radius p, k ≪ p, in which M = q(τ + k), the formula is

m3 N 2
" !#
q q
W = 2
E2 (τ ) − E2 (τ + k) , (130)
24gYM p p

but since Eqs. (83) applies and xE2 (x) = x−1 E2 (−x−1 ) + (6i/π),

m3 N 2 E2 (τ ) m3 N 3
W → − . (131)
96πg 96πgq 2(τ + k)2

The first term in this expression is the same as in the D5-brane vacua, so field theory and
supergravity agree up to a classically-subleading M-independent function.
For a supersymmetric domain wall between two phases, the tension is
 
|m3 | n3I X n′3
X 
 J 
τDW = 2|∆W | = 
2
− ′2 
. (132)
48πg 
I left MI J right MJ 

Let us consider two examples, to see that the 5-brane junction construction reproduces this
tension. For both examples we take g ≪ 1.
The first is the domain wall described above, between the confining and first oblique
confining phase. The general result (132) becomes

11 We thank O. Aharony for pointing out an error in our orginal approach.

42
 |m3 |N 3  −1 −2  |m3 |g 2 N 3
τDW = (ig ) − (ig −1 + 1)−2  ≈ . (133)

48πg 24π
In the brane picture, the tension comes from the spanning D5-brane. This has three trans-
verse and three longitudinal dimensions and so feels no warp factor, giving simply
4πr03 µ5 |m3 |g 2 N 3
τDW = · = . (134)
3 g 24π
The agreement is quite beautiful, given the very different physics that has gone into the two
calculations.
The second example is the domain wall between the confining and Higgs phases: a
junction between an NS5-brane and a D5-brane, spanned by a (1,1) 5-brane. The NS5-
brane is at much smaller radius than the D5-brane (by a factor g) and has much greater
tension, so the predominant effect is that the D5-brane bends down to join the NS5-brane.
The bending is described by the BPS equation

∂W
∂1 φ = Ω (135)
∂φ
with Ω any phase. For m real, Ω = −1 gives a solution that passes through the origin and
the nonzero stationary point, approximating to order g the solution needed. The tension
comes primarily from this bending,
∞ N 2πg
Z  
1
τDW = dx |∂1 φ|2 + |Wφ |2 . (136)
0 2πg N
In this case the general result (132) follows by construction

|m3 |N 3
τDW = , (137)
48πg
where the superpotential in the confining phase is smaller by O(g 2). It will be an interest-
ing exercise to show that the brane construction reproduces Eq. (132) in the general case,
without assuming small g.
Note that a domain wall is the same as a baryon vertex extended in two additional
directions. By analogy, we might expect the D5-brane three-ball which acts as a domain
wall to be associated with passing a D7-brane through the NS5-brane. This suggests that
D7-branes should be reexamined in the original AdS5 × S 5 context.

E. QCD-like vacua

It is amusing that N = 1∗ is rich enough to permit us to study a theory similar to QCD

with heavy quarks. Suppose we consider a vacuum with a D5-brane of radius n and one

43
NS5-brane of radius N − n. Here we assume the usual condition on n, n2 ≫ gN, but take
n ∼ gN, so the D5-branes have comparable AdS radius to the NS5-brane. In the field
theory this corresponds to a vacuum with a broken SU(n) sector, a U(1) vector multiplet,
and a confining SU(N − n) sector. Among the massive vector multiplets are spin-1 bosons,
along with fermions and scalars, charged as (n̄, N − n) under SU(n) × SU(N − n). These
are strings connecting the D5-brane to the NS5-brane. We will refer to these as ‘quarks’.
Clearly these theories have no free quarks: the D5-NS5 strings cannot exist in isolation, since
they cannot actually end on the NS5-brane, and instead must be connected to a flux tube.
Note that the SU(n) acts as a sort of flavor group for the quarks (analogous to broken weak
isospin), and we will refer to its massive adjoint representation as ‘flavor’ gauge multiplets.
In order that the supergravity solution be valid, we must have n ≫ (gN)1/2 . When
n = gN, so that the D5 and NS5 sit at equal AdS radii, the quark has mass of order
Z q
(α′ )−1 G00 Gyy r02 dζ = r0 /α′ = mn

which agrees with field theory. However, it is easy to see that if n is smaller than gN, the
quark retains mass ∼ mgN; indeed, as an extreme, note that if a D3-brane sits exactly at
r = 0, corresponding to (gN)1/2 ≪ n ≪ gN, the quark mass is just proportional to the
coordinate length of the string, mgN. We therefore see signs that the physical ‘constituent’
masses of the quarks can be much larger than their current values. Of course the quarks are
never light compared to m.
We can see easily that QCD-like theories have no stable flux tubes due to quark pair
production. Recall that we measure the potential V (L) between two electric sources by
hanging a probe string by its ends from the AdS boundary, with the ends a distance L
apart. Take L ≫ m−1 (gN)−1/2 ; then in the absence of the D5-brane, the probe would bind
to the NS5-brane forming a confining flux tube between the sources. In the presence of the
D5-brane, however, pair production of the D5-NS5 strings can occur. This breaks the flux
tube, which shrinks away allowing the quark to screen the source. The probe string ends up
as two strings a distance L apart, each attached to the D5-brane. Note however that if we
take n ≫ gN, the quarks become very heavy, and the time scale for their pair production
becomes very long. In this limit the confining flux tubes are metastable.
Low-lying quark-antiquark mesons are not stable in this theory. Highly excited mesons
are represented by two D5-NS5 strings joined by a long F1-NS5 flux tube. However, as the
mesons deexcite by emission of glueballs (either supergravity or string states), it eventually
becomes energetically preferable for them to decay to a D5-D5 string, bypassing the NS5-
brane altogether. In short, the lowest lying mesons between two quarks always mix with and
decay to a massive, but lighter, flavor particle, in a vector multiplet of the broken SU(n)
group. (Indeed this almost happens in nature; charged pions decay through isospin gauge

44
multiplets, although not because those gauge bosons are light but because they couple to
light leptonic states — which could also be represented here, if there were a need.)
Baryons, on the other hand, carry a conserved charge and are both stable and interesting.
N − n D5-NS5 strings can end on a D3-brane filling the NS5-brane two-sphere, forming an
object whose mass can be computed. If we arrange for a more complicated spectrum of
quarks by choosing to use multiple D5-branes of various radii, then there are processes by
which baryons can be built from quarks of different masses, and can decay by emission of
flavored mesons (or the corresponding gauge multiplets of the ‘flavor’ group.) Scattering of
baryons, or of baryons and antibaryons, could also be studied. In addition, it is possible that
these baryons have residual attractive short range interactions (different from the physical
case in that they are dominated by the ‘flavor’ gauge multiplets) which can cause them to
form nuclei. It would be amusing to look for such baryon-baryon bound states. Furthermore,
these baryons and nuclei carry U(1) gauge charges, and in some vacua there are lepton-like
objects which presumably can combine with them to form atoms.
We cannot resist mentioning one more possibility, although it admittedly may not be
realized. Namely, our baryons act as D0 branes in spacetime, and our domain walls as
D2-branes (their structure in the extra dimensions is identical, so we suppress it.) While
we are not used to thinking of baryons as places where strings can end, this is quite natural
if there are no light quarks; if all quarks are heavy then short flux tubes are stable, and
physical baryons can be linked by them. Turning on condensates of these flux tubes makes
the positions of the baryons noncommuting (note these branes have no massless world-
volume gauge fields, but still have massless scalars), and through the Kabat and Taylor
mechanism [15], an assembly of baryons can be arranged into a spherical domain wall! Thus
the properties of the gauge theory recapitulate the method we have used to solve it. In
practise, one should try to implement this process physically, through Myers’ mechanism
[14]. Here we have a difficulty, as the required three-form potential is a massive state, a
glueball which couples to domain walls [64], so we cannot create a long-range field to induce
a dipole charge. However, there may be ways to circumvent this problem, and create this
effect as a thought experiment or even in a lattice simulation, where hints of domain walls
have been observed [65].

F. Condensates

With the naked singularity banished, the coefficients of the normalizable terms in the
supergravity fields, and so the dual condensates, become calculable. We have already deter-
mined the superpotential in our discussion of domain walls, and in principle the condensates
can be determined directly from this function. The full field theoretic superpotential, and

45
corresponding condensates, are also known [8]. However there are subtleties [39,66], and our
understanding is only partial.
The condensates of the operators λλ, tr[Φ1 , Φ2 ]Φ3 and mi trΦi Φi are all related by the
chiral anomaly and operator mixing. A linear combination of these must couple, by the
AdS/CFT correspondence, to the mode of G3 which falls off as 1/r 3 in invariant units,
which we have identified in Eq. (47). More generally, higher modes of G3 should give the
expectation values of all of the chiral operators λλφk + · · ·. For the lowest mode of G3 ,
Eq. (94) for ω2 gives the m and phase dependent parts as
m3 N 3
Mr03 ∝ . (138)
M2
For multiple shells, superposition gives
n3I
m3
X
. (139)
I MI2

Note that there are two SO(3)-invariant fermion bilinears, namely 3i=1 λi λi and λ4 λ4 . These
P

correspond to the polarization tensors ǫi̄k̄ and ǫijk , which have equal overlap with the actual
field ǫw1 w2 w3 .
In the Higgs vacuum, and other vacua with only large D5-branes, all condensates can be
described semiclassically. The expectation values for tr[Φ1 , Φ2 ]Φ3 and mi trΦi Φi are known,
and their m, n and M scaling agrees with (139). In the confining cases, however, the
situation is more subtle, since these operators have condensates of different sizes and since
the gluino bilinear is also expected to play a role. We note the following facts. First,
careful examination of the field theory superpotential given in [7,8] reveals no obvious linear
combination of the operators whose expectation values would have this property — except
the second term in the superpotential itself, as discussed in section VII.D. Second, there is
every indication from our study of domain walls that the second term in the superpotential,
proportional to the two-sphere volume times the 5-brane’s charge under ∗G3 , measures the
dipole moment of the 5-branes. Naturally, the lowest normalizable mode of G3 couples to the
lowest allowable 5-brane multipole moment, which is indeed a magnetic dipole. We therefore
speculate that perhaps the ijk components of G3 couple to this part of the superpotential,
and to the worldsheet instantons which we will discuss in section VII.G below.
In any case, the supergravity clearly shows that there are expectation values for some
dimension-three operators which classically would have had vanishing vevs. It also shows
these vevs differ from one confining vacuum to the next. These qualitative features certainly
agree with field theory.
The chiral operators F 2 φk are determined by the dilaton background. The dilaton is
nontrivial and is obtained from

46
 2πgr04|M|2 3 g2
∇2 Φ = δ (y)δ(w − r 0 ) + Re(Gmnp Gmnp ) . (140)
NR2 12
The first term comes from the coupling of the dilaton to the 5-brane through the Born-Infeld
action. The second is directly from the coupling to the bulk fields. Taking the value of r0
appropriate to a single shell of quantum numbers M, and integrating over a volume of order
r06 , both terms are of order m6 R2 r04 and must be retained. One can argue that they must
cancel in the dilaton monopole, which gives the expectation value of the derivative of the
Lagrangian with respect to the coupling; this must vanish in a supersymmetric vacuum.
Many of the higher operators F 2 φk are also highest components of superfields, and for them
a similar argument should apply. We have not yet shown this cancellation directly for the
background (140), and it is possible that there are subtleties. It is a challenge to include
this varying dilaton in the full nonlinear treatment of supergravity.

G. Instantons

At weak coupling, many quantities in N = 4 Yang-Mills theory receive contributions

from instantons. Holomorphic objects can be written as an instanton expansion, given by
an infinite series in powers of the parameter q = e2πiτ . This expansion is not useful at strong
coupling, but in that case the same objects can typically be reexpressed, using a modular
transformation, in terms of q̃ = e−2πi/τ (or some other SL(2, Z) variant.) In string theory,
both in perturbation theory for flat D3 branes and in the AdS5 × S 5 language, D-instantons
[D-(−1) branes], whose action is 2π/g, play the role of these field theory instantons. For
large g one uses an expansion in magnetic D-instantons, with action 2πg, or more generally
in dyonic D-instantons.
In N = 1∗ , the situation is slightly different. Consider first weak ’t Hooft coupling, such
that the strong coupling scale Λ is much less than m. In this case the Higgs vacuum has a
superpotential which is an expansion in q = e2πiτ , but the superpotential of the confining
vacuum has an expansion in q 1/N . The 1/N in the exponent is responsible, in the m → ∞
limit, for SU(N) Yang-Mills having N vacua, related by θ → θ + 2πk. It has long been
suggested that this behavior implies the existence of fractional instantons, carrying 1/N
units of instanton charge, which, unlike instantons themselves, remain important in the
large N limit. Some evidence for these objects has been found in MQCD [67] (note that
these are distinct from similar objects which require compactification of a dimension of
spacetime for their existence) and there is even a claim that they have been seen in lattice
nonsupersymmetric Yang-Mills [68].
We might hope to find such objects here, but for the same reason that Λ ∼ m, we cannot
do so, for q 1/N is of order one, and the fractional-instanton expansion fails. We also cannot

47
use the magnetic instanton expansion, since g ≪ 1. But remarkably, as can be inferred
from the results of [8], there is yet another expansion, one which is dual, in the sense of
gN ↔ (gN)−1 , to the expansion in fractional instantons.
In particular, Dorey and Kumar show that the superpotential for the confining vacuum
is proportional to
1 τ
 
E2 (τ ) − E2 (141)
N N
where E2 is the second Eisenstein series. For large imaginary arguments, E2 (z) can be
written as an expansion in e2πiz . The first term in (141) can therefore be interpreted as a
sum over ordinary instantons. The second term, by contrast, cannot be expanded in this
way, since |τ /N| ≪ 1. However, since N/τ is large and imaginary, we may make progress as
in [8] by using the anomalous modular transformation
1 1 6i
 
E2 (z) = 2
E2 − + . (142)
z z πz
from which we learn that the second term in Eq. (141) dominates the first and that it can
be expanded in power of e−2πiN/τ = e−2πgN . This is an expansion in a small quantity, and
we need only provide an interpretation for it.
This is not difficult to obtain, for an NS5-brane of the form S 2 times Minkowski space
permits the string world-sheet to wrap the S 2 , producing instantons. From the metric (109)
the proper area of the NS5-brane sphere times the tension of the fundamental string is
minimized by
R2 1 R4
(4πr02 ) = = 2πgN (143)
2r0 ρc 2πα′ 2α′2
which is just as required to explain the expansion in the superpotential. As a check, note
that if we repeat the calculation for a vacuum with q coincident spheres, both the area of
the spheres and the exponent in the field theory are reduced by a factor q. For the Higgs
vacuum, the metric and dilaton in Eqs. (106) and (108) imply the area of the D5-sphere
times the tension of a D1 brane goes at small ρ to
R4
= 2πN/g . (144)
2g 2α′2
Since the Higgs vacuum superpotential is proportional to [8]

E2 (τ ) − NE2 (Nτ ) (145)

we may again interpret the second term (now much smaller than the first term, except
for its leading τ -independent contribution) as an expansion in D1-instantons wrapping the
D5-brane two-sphere.

48
 It would be interesting to find a string theory interpretation for the coefficients in the
expansion of of the superpotential, and especially for the anomalous term in the modular
transformation of E2 .

H. Glueballs and Other Particle States

The spectrum of states in this theory is complicated, and we do not yet have a physical
understanding of its features. We will outline its structure and point out some puzzles and
problems which must be solved in future.
First, there is already surprising structure in the N = 4 theory in the vacuum with Z
given by Eq. (76), where the D3-branes form a 2-sphere of radius r0 . The typical warp factor
in the region r ∼ r0 ∼ mN(g)α′ is Z ∼ R4 /r04 . A supergravity state then has typical kµ
given by [69]

Gµν kµ kν ∼ Gmn km kn (146)

so that
r0
kµ ∼ Z −1/2 km ∼ ∼ mg ±1/2 N 1/2 . (147)
R2
where the minus (plus) sign applies for the D3-sphere radius appropriate to the Higgs (con-
fining) vacuum. Immediately we have a puzzle. The semiclassical field theory of this N = 4
vacuum would have led us to expect physics from the W -bosons whose masses lie between
the scales m and Nm. There are also monopoles of masses m/g and Nm/g. These states
are present on the supergravity side as F- and D-strings stretched between the D3-branes.
But there is no sign of gauge theory states with masses given in the previous equation. The
situation is not improved by consideration of excited string states in the bulk, for which one
has similarly

Gµν kµ kν ∼ 1/α′ (148)

and
r0
kµ ∼ Z −1/4 α′−1/2 ∼ ∼ mg ±1/2 N 1/2 × (gN)1/4 , (149)
Rα′1/2
an odd-looking scale.
Now, what changes when the D5- or NS5-brane charge is added? For the D5-brane,
essentially nothing happens to these arguments. This is despite the fact that a magnetic
flux tube has formed, with a tension whose square root is given by Eq. (147) with the minus
sign. Presumably the details shifts around slightly, and of course the massless photons of

49
the D3-branes now develop mass of order m, but apparently the spectrum is otherwise little
changed.
For the NS5-brane, the situation is more subtle. The electric flux tube has a tension
whose square root is given, as in the D5 case by Eq. (147), now with the plus sign. But in
addition, unlike the D5-brane, the NS5-brane has a throat region. If we consider a vacuum
with several coincident NS5-branes, then the throat region is reasonably well understood,
since it can be described in conformal field theory in the region where the string coupling is
small [50]. From this it is known that supergravity states get string-scale masses from the
coupling to the throat geometry. Using the metric (103) in the calculation (148) gives
1/2 r0
kµ ∼ r0 ρc′1/2 R−1 α′−1/2 ∼ 2
∼ mg 1/2 N 1/2 . (150)
R
This applies to both supergravity and excited string states. Notice that this is the same
scale as for supergravity states in the bulk, and for the string tension.
Unfortunately, the confining vacuum has only one NS5-brane, and there is a long-standing
controversy over this object, reflecting the difficulty of doing any reliable calculations in its
presence [56]. There are disputes over whether the throat and its region of strong coupling
even exist. In any case, neither supergravity nor conformal field theory is reliable, and we
simply do not know what the spectrum will do in this regime. We note this may hint at a
profound obstacle to using string theory as a practical computational tool in QCD.
In both the D5 and NS5 cases, there is one more class of light states, arising from the
massless gauge fields on the 5-brane. Relative to the bulk states, the magnetic field on the
D5-brane reduces the velocity of open string states by a factor of the dimensionless field,
v ∼ (N/g)−1/2 : restoring Fµν F µν and Fµa F µa to the brane action, one finds that the latter
is multiplied by v 2 . The mass gap is reduced by the corresponding factor, and so is simply

kµ ∼ m . (151)

This agrees with the classical result, since these are the W bosons and this is the scale of
SU(N) breaking. Meanwhile, the NS5-brane has a normalizable zero mode [70], which gives
rise to the massless vector required by S-duality to the D5-brane. We assume that this mode
survives when D3-branes are dissolved in the NS5-brane. Then
1 w1
Gµν kµ kν ∼ Gw kw 1 kw 1 , (152)

and
r0 ρ′c ρ′c
kµ ∼ k w 1 ∼ ∼m. (153)
R2 R2
Again we find a lighter branch of states localized at the brane. By analogy to the Higgs
vacuum, one might interpret them as massive magnetic gluons, but in the NS system they

50
are not open strings but rather closed string states in localized wavefunctions on the throat.
Most of them carry SO(3) quantum numbers and are therefore not seen in N = 1 Yang-Mills
theory. Some or all of the remainder may mix with bulk states, as we discuss below.
Before doing so, we note that all of the masses we have obtained in the confining vacuum
are consistent with ’t Hooft scaling, as they are proportional to m times a power of gN.
This is pleasing, although it means of course that all of the states merge and mix as gN is
taken small, making any quantitative computations in this regime essentially irrelevant for
N = 1 Yang-Mills.
Let us note one important interplay between the open and closed string states. There
would appear to be one unbroken U(1) gauge group per (noncoincident) brane, from the
world-sheet gauge field on each brane. For an SU(2) representation given by the sum of
k distinct irreducible blocks, SU(N) is broken to U(1)k−1 . On the string side there are k
D5-spheres and so apparently a U(1)k . In fact, one U(1) should be lifted by the coupling to
the bulk states.12 We have not understood all the details, but will indicate the ingredients.
There is a massless tensor in 3 + 1 dimensions, with the field Bµν independent of xm . Its
kinetic term is
Z √ Z
′ ′ ′
Z
d10 x GHµνλ H µνλ = d4 xµ η µµ η νν η λλ Hµνλ Hµ′ ν ′ λ′ d6 xm Z 2 . (154)

The xm integral converges both at the branes and at infinity, so this is a discrete state. By
itself, the coupling of this field to Fµν in the Born-Infeld action would generate a mass for
one linear combination of U(1)’s via the Higgs mechanism. However, the field Cµν also has
a zero mode, seemingly lifting a second U(1). The actual story must be more complicated,
with the bulk Chern-Simons term playing a role, because from the point of view of a single
brane this would be simultaneous electric and magnetic Higgsing, an impossibility.

VII. EXTENSIONS

A. Unequal Masses

Now we consider the general N = 1 case, three masses not necessarily equal. Examina-
tion of the classical F -term equations (6) suggest use of the coordinates
√
z1 = m2 m3 χ cos θ ,
√
z2 = m1 m3 χ sin θ cos φ ,
√
z3 = m1 m2 χ sin θ sin φ . (155)

12 This was suggested by E. Witten.

51
We will study the potential with the Ansatz that χ is constant and also that Fθφ = 21 n sin θ.
We insert this into the potential (69). Noting that

det G⊥ = 4m̃|m1 m2 m3 χ4 |Z sin2 θ ,

m̃ ≡ |m1 | cos2 θ + |m2 | sin2 θ cos2 φ + |m3 | sin2 θ cos2 φ ,
Tmnp xm dxn ∧ dxp = |m1 m2 m3 |(|m1 | + |m2 | + |m3 |)χχ̄2 ,
¯
Ti̄k̄ T l̄jk z i z̄ l = 2m̃|m1 m2 m3 χ2 | , (156)

the potential becomes

S 4|m1 m2 m3 | |m1 | + |m2 | + |m3 | 2 ′
√ 2
− = |Mχ − 2πα inχ/2 2| . (157)
V πgn(2πα′)4 3
√
This has a supersymmetric minimum at χ = 2πα′in/2 2M. The two-sphere is now an
ellipsoid with axes
πα′n q πα′ n q πα′ n q
|m2 m3 | , |m1 m3 | , |m1 m2 | . (158)
|M| |M| |M|
When m3 → 0 with m1 = m2 fixed, we obtain N = 2 Yang-Mills; here the ellipsoid
degenerates into a line of length m1 . This is very easy to understand, classically, for the
Higgs vacuum. N = 2 Yang-Mills with a massive hypermultiplet has a moduli space, which
classically is just given by the positions of D3-branes (suitably modified so that they can
only move in two dimensions — this remains to be understood in our present context.) On
this space is a single point where there are N − 1 massless electrically charged particles,
from the hypermultiplet. At this point, the N D3-branes are arranged in a line. From the
classical equations, it is easy to see that an N = 2 breaking mass parameter for the vector
multiplet causes this line to become a noncommutative ellipsoid analogous to the sphere of
[15]. In addition, the electromagnetic dual of this transition corresponds to a well-known
fact, both in field theory [71–73] and in MQCD [62,74–76], concerning the breaking of pure
N = 2 Yang-Mills to N = 1. The moduli space of N = 2 has an associated Seiberg-
Witten auxiliary Riemann surface of genus g. There are N special isolated points on the
moduli space where N − 1 mutually local dyons become massless, and the Riemann surface
completely degenerates. In this degeneration, the N handles of the surface join along a
singular line segment. The addition of a mass parameter breaking N = 2 supersymmetry to
N = 1 engenders a transition whereby the N handles join to form a single surface of genus
zero: the line segment opens up into a closed curve. In M theory this is represented by a
multigenus M5-brane making a transition to a genus zero M5-brane. Here we see signs of
a similar phenomenon; the 3-branes which represent the moduli space of the N = 2 theory
presumably align along a line segment, then join and expand to form an NS5-brane ellipsoid.

52
 By contrast, when m1 = m2 → 0 with m3 fixed, the ellipsoid becomes a disk while its
overall size shrinks to zero. In the field theory, one expects an infrared fixed point, and
indeed the supergravity should go over smoothly to the kink solution of [37] and at small
r to the ten-dimensional space of [77]. Presumably the G3 background in [77] is related to
the linearized one we have obtained in Eq. (36), although we have not checked this.
It would be interesting to understand these connections in more detail, but we should
note that our approximations will break down in these limits. We have seen that the 5-brane
shell has a finite thickness, and we need this to be less than the shortest axis of the ellipsoid
for our linearized approach to be consistent.
We note also that most of our results generalize easily to this case. The superpotential,
the domain wall tensions and the condensates, are all related, as is the dipole moment of
the 5-brane, to the volume of the ellipsoid ∝ m1 m2 m3 . The flux tubes will presumably show
signs of the extra metastable Regge trajectories seen in [73] by localizing on the ellipsoid,
along the lines observed in MQCD in [75].

B. N = 0

Let us make a few brief remarks about the nonsupersymmetric case, m4 = m′ with
m1 = m2 = m3 = m kept equal. The potential is now
2πα′n h (2πα′ n)2
( )
S 4 2 4 2 ′ 3
i
− = |M| |z| + √ Im (3mzz̄ + m z )M + O(z 2 ) . (159)
V πgn(2πα′)4 3 2 8
The quadratic term depends on the boundary conditions, as discussed at the end of sec-
tion IV.A. Its general form is given by
4
1
O(z 2 ) = |z|2 |mi |2 + (L = 2) .
X
(160)
3 i=1

In the absence of supersymmetry we must make a particular choice of the L = 2 harmonic

µmn , defined in Eq. (65), which represents a traceless combination of masses for the scalar
bilinears. This harmonic reduces the masses of some of the scalars, and if too large it can
cause the gauge symmetry to break. However, it represents an adjustable parameter in the
Lagrangian, so we are completely free to choose it in such a way that it preserves a stable
vacuum, assuming such a choice exists. Further, if we maintain SO(3) invariance, the choices
are greatly reduced.
Since µmn = 0 in the SO(3)-invariant supersymmetric case, and since the vacua in
the supersymmetric case are stable, it is evident that for small m′ /m the continued use
of µmn = 0 leads to stable vacua at nonzero z. Of course, the vacua need no longer be
degenerate. Whether the single NS5-brane is the preferred vacuum is less obvious, although

53
it seems likely, since it is an extreme case among the vacua. In such a vacuum the spectrum
would be altered and the stable domain walls would be lost, but most of the other features of
the confining vacuum — flux tubes, baryon vertices, condensates and instantons — would be
qualitatively unchanged. The new features would be the appearance of gluon condensates,
such as trF 2 , which must be zero in the supersymmetric case, and nontrivial dependence on
the phase of m′ /m.
While the situation is less clear when m′ ∼ m, there are reasons to expect, on purely
physical grounds, that a confining vacuum does in fact exist. It seems a worthwhile challenge
to seek it in supergravity.

C. Orbifolds and QCD

Many authors have studied supersymmetric and nonsupersymmetric orbifolds of the

N = 4 Yang-Mills theory, its D3-brane representation and its supergravity dual. (A list of
references may be found in [78].) Here we briefly consider two orbifolds of the results we
have obtained above.
In the N = 4 theory, a Z2 orbifold on four coordinates leaves an N = 2 SU(N) × SU(N)
theory with (N, N) + (N, N) hypermultiplets. We can combine this action with the mass
perturbation in two distinct ways. The first is a simultaneous rotation by π in the 45- and
78-planes. This is part of the SU(2) that acts on the chiral superfields, and so commutes
with the N = 1 supersymmetry and leaves it unbroken. The second is a rotation by π in
the 45-plane and −π in the 78-plane. Since it differs from the first rotation by a 2π rotation
in the 78-plane, we can think of it as the first rotation times (−1)F , with F being spacetime
fermion number. The first rotation commutes with the N = 1 supersymmetry generator,
so the second anticommutes with it and leaves a nonsupersymmetric theory. Note that the
two rotations are conjugate to one another, and so in the absence of the mass term would
give equivalent theories.
We are assuming that the Z2 acts on the Chan-Paton factors as
IN 0
 
. (161)
0 −IN
More generally the blocks could be different sizes, leading to an SU(N) × SU(M) gauge
theory; this gives rise to a twisted state tadpole and so is more complicated [79]. For the
supersymmetric orbifold, the massless fields that survive are
Aµ 0 Φ 0
   
Aµ = , Φ3 = ,
0 A′µ 0 Φ′
0 Q1 0 Q2
   
Φ1 = , Φ2 = . (162)
Q̃1 0 Q̃2 0

54
Thus there are SU(N)×SU(N) vector multiplets, chiral multiplets Φ and Φ′ in the respective
adjoints, and bifundamental chiral multiplets Q1 , Q2 in the (N, N) and Q̃1 , Q̃2 in the (N, N).
The N = 1 superpotential is W ∝ φ(Q1 Q̃2 −Q2 Q̃1 )+ φ̂(Q̃1 Q2 − Q̃2 Q1 ). The AdS description
of this theory is AdS 5 × S 5 /Z2 . There is a fixed plane at x4 = x5 = x7 = x8 = 0, which
in the supergravity is AdS5 × S 1 where the second factor is a fixed S 1 on the S 5 . We
may preserve supersymmetry and add the superpotential m(φ2 + φ̂2 + Q̃1 Q1 + Q̃2 Q2 ). In
the gravity description, this is simply the perturbation we studied earlier. The low energy
theory consists of two separate N = 1 Yang-Mills theories with no massless matter. There
are two gluino bilinears operators, one even and one odd under the Z2 .
The second rotation differs by (−1)F , so the action on the bosons is the same: they are
the same as for the theory (162). The action on the fermions is opposite, so the fermionic
partners are all absent, while fermions appear in all the blocks with 0’s. Thus the low-energy
field theory in this case is nonsupersymmetric SU(N) × SU(N) gauge theory with a Dirac
fermion Ψ in the (N, N). This is a Z2 orbifold of N = 1 Yang-Mills [80]. It has no fermion
bilinear odd under the Z2 . Note that for N = 3 it is QCD with three massless quarks and
with the vector SU(3) of the flavor group gauged. This gauging removes all but a Z3 axial
symmetry, or more generally a ZN .
In both cases, the renormalization group flow is from an SU(N) × SU(N) N = 2 su-
persymmetric theory in the ultraviolet to a gauge theory with massless fermions but no
massless scalars. In the limit m → ∞, with N and Λ3 ≡ m3 exp(−8π 2 /gYM2
N) fixed, the
supersymmetric orbifold has N 2 vacua; these confining vacua, in which the two gluino bi-
linears have separate condensates, should survive to the finite m case. The presence of the
massive matter in the bifundamental representation assures, however, that there is only one
type of ZN flux tube, and only one type of baryon vertex. Only the condensates can distin-
guish vacua separated by N consecutive domain walls. By contrast, the nonsupersymmetric
orbifold has only one fermion bilinear, with a consequently unique expectation value. As we
mentioned, there is an accidental ZN axial symmetry in the limit m → ∞, N, Λ fixed if Ψ
is held massless — more on this below. In this limit we expect the ZN to be broken by a
fermion bilinear (note this is consistent with the dynamics of physical QCD) so we expect N
confining vacua. Again there is only one ZN string and one baryon vertex. (In both theories
we expect physical baryon states; however they are rather similar in the two cases, differing
in mass only slightly.)
On the string side of the duality, all of our brane configurations are invariant under Z2
and so survive in the orbifold theories. We do not see a mechanism for new phases to arise
geometrically, so the extra vacua in the supersymmetric case are presumably associated
with expectation values of fields in the twisted sector. Certainly the difference of the gluino

55
condensates (and corresponding scalar operators) is in the twisted sector, while the sum is
discussed in section V.F. The two supergravity orbifolds differ in the behavior of fermionic
fields under the reflection, so that the spherical harmonics, and spectra, will be different in
the two cases. The physics involving strings and baryons is essentially the same as before,
but it would be interesting to understand how the domain walls are different.
The nonperturbative condensate in the unorbifolded theory survives, as a gluino bilinear
expectation value, to small gN, where it can break the ZN nonanomalous R-symmetry of
N = 1 Yang-Mills. Unfortunately we cannot make the same claim for the condensate in the
nonsupersymmetric orbifold. Only if Ψ is massless for m → ∞, N and Λ fixed, does the
theory have a discrete axial symmetry, which can be broken by a Ψ̄Ψ condensate. However,
while a gluino is always massless by supersymmetry, no such symmetry protects the fermion
Ψ. We must therefore assume that Ψ can obtain a mass in perturbation theory through
nonplanar graphs. Fine tuning is required to obtain the axial symmetry and the massless
fermion at small gN, and thus any connection between chiral symmetry breaking in QCD
and the large-gN condensate is tenuous.
As an aside, we emphasize the physical interest of these questions. Nonsupersymmetric
SU(N) × SU(N) with a single massless bifundamental fermion, treated as a function of N
and the ratio of gauge couplings, is a very interesting theory worthy of further attention.
First, if the gauge couplings are very different, the physics between the two strong coupling
scales approximates physical QCD. Second, the theory exhibits both confinement and chiral
symmetry breaking, with the breaking of a discrete axial symmetry and interesting domain
walls. As such, it closely resembles N = 1 supersymmetric Yang-Mills theory. Third, in
contrast to N = 1 Yang-Mills, the low-energy theory has only vector-like fermions and can
be investigated with relative ease on the lattice. To our knowledge it has not been previously
studied. Of course, lattice studies of weakly broken N = 1 Yang-Mills are not impossible
[65]; and one may hope, by comparing the SU(N) × SU(N) theory with a massive bilinear
fermion to broken N = 1 Yang-Mills, to study the extent to which nonperturbative physics
survives orbifolding in the large N limit. In short, this theory is physically interesting,
tractable on the lattice, qualitatively related to supersymmetric Yang-Mills theory even for
small N, and perhaps quantitatively related to it at large N.

VIII. DISCUSSION AND FUTURE DIRECTIONS

As with all dualities, our work has implications in both directions — for supergravity
and string theory, and for gauge theory.

56
 A. Strings, gravity, and singularities

We have found one more example of a recurrent pattern, the resolution of a naked
singularity by brane physics. Earlier examples are the nonconformal Dp geometries [58], the
Coulomb branch singularities [17,19,20], and the enhancon [21].
It is not clear how earlier work on the N = 1∗ theory is related to ours, because it was all
in the context of five-dimensional supergravity. The solutions of ref. [3] might lift to ours,
but this requires that a great deal of physics, the entire brane configuration, be hidden in
the ‘consistent oxidation’ of the five-dimensional solution. We note that the solutions [3]
all have constant dilaton while our 5-branes will produce a locally varying dilaton even if
not a dilaton monopole moment. As far as is known, the oxidation cannot produce such
an effect. In ref. [39] a general criterion was proposed for identifying physically acceptable
naked singularities. Again this was expressed in terms of the five-dimensional theory and so
cannot be applied to our solutions without substantial additional work.
There are a number of obvious loose ends in our work. We have not obtained the
full supergravity background. We had the good fortune that we could obtain the relevant
physics by working only to first order, and partly to second order, in the mass perturbation.
It remains at least to solve the supergravity equations to second order. We have observed
that the equations of the supergravity have many simplifications in our situation, but we
have not understood their origin. It seems extremely likely that these will extend to the full
second order calculation. It may even be possible, with the guidance from our approximate
solution, to find the exact supergravity solutions. A related question is the analysis of the
supersymmetry properties of our solutions, to establish why the supersymmetry is preserved.
This involves the 5-brane world-volume as well as the bulk supergravity.
We should remind the reader that we have not fully explained a key point, namely the
fact that there are far more brane configurations than there are field theory vacua: many
brane configurations represent the same vacuum. For example, we have not explained why
the massive vacuum with a (1, q) 5-brane is the same as that with a (1, q + N) 5-brane; we
have merely shown that the question never arises purely within the supergravity regime. To
understand the transition between these and other descriptions as τ is varied remains an
interesting challenge. A related issue is the minimum of the brane potential at z = 0. This is
outside the range of validity of our approximation, but would naively correspond to an unex-
panded brane and so a singular spacetime. Since there exist expanded brane representations
for all ground states, we presume that a correct interpretation involves transmutation into
a different brane configuration. A complete accounting of these transitions is greatly to be
desired.
Another related fact is that many brane configurations (such as p coincident D5-branes,

57
 √
p ≪ N) represent multiple vacua, which are only split from one another by strong coupling
SU(p) physics that we cannot see in supergravity because p is small. This physics will have
to be understood separately, and perhaps is only described by field theory.
Perturbed AdS spacetime is relevant to the generation of hierarchies in Randall-Sundrum
compactification [23,81]. In the latter paper it was argued that a large number of D3-branes
localized in a compact space will generate an effective AdS throat. That work was in the
context of an N = 4 compactification and so the throat was stable: no relevant perturbation
exists. The situation becomes more interesting in more realistic cases. If the D3-branes are
localized on a space that leaves N = 1 unbroken, and the compactification is generic, the G3
mode that we have considered will have a nonzero boundary value, proportional to m, which
is of order order one in four-dimensional Planck units. Since this is a relevant perturbation
it becomes nonlinear essentially immediately, and the throat disappears. If for some reason
the perturbation is anomalously small, then of course, as in any supersymmetric field theory,
the ratio m/mPl is quantum mechanically stable and the throat will be larger. Our work
shows that the incipient throat is capped by an expanded brane. It is also possible to avoid
the instability using, for example, a discrete symmetry G3 → −G3 .13
It would be most interesting to study marginally relevant interactions. These may not
exist in the N = 4 theory, but are present in various elaborations. On general grounds,
not requiring supersymmetry, such perturbations would naturally become nonlinear only
well down the AdS throat, where again an expanded brane would presumably form. This
behavior would bear some similarity to the hierarchy mechanism of ref. [82]. Examples of
such interactions were considered in [83,84], and there is strong reason to expect an expanded
brane there as well.
Our work bears directly on recent proposals regarding the vanishing of the cosmological
constant [24], which involve a naked singularity in the compact dimensions. There have been
many criticisms, published and unpublished, of the assumptions made in ref. [24] regarding
the properties of the singularity, but we can add to these the specific example of what string
theory does in our case. In terms of our notation from section III.B, the authors of refs. [24]
assume that the parameter b can take arbitrary values. That is, they integrate in the
coordinate r, and assume that any singularity encountered gives an acceptable spacetime.
Further, they must assume that b is a fixed parameter, rather than a dynamical quantity. As
we see from our discussion, this is inconsistent with the requirements of AdS/CFT duality,
and it is not what happens in our case: b takes discrete values, depending on the particular
vacuum, and can make dynamical transitions from one value to another.

13 This has also been noted by O. Aharony.

58
 Thus the singularity in the end is replaced by an ordinary physical object, like a hydrogen
atom. For hydrogen, too, one can integrate the wave equation inward in r for any energy,
obtaining a generally singular solution. The experimental discreteness of the spectrum
indicates that nature abhors a generic singularity.
√
It is notable that our system has a large number of ground states,14 of order e N . In
the supersymmetric case these are all degenerate, but once supersymmetry is weakly broken
they will form a closely spaced, near-continuum of discrete states. If a singularity is of
this type, the system may have vacua of exceedingly small cosmological constant, and the
mechanism of [85] may be realizable. In order for such a mechanism to solve the problem,
these states must be sufficiently metastable, and there must be a dynamical mechanism to
select a state with a small net cosmological constant (the same problems are in any event
present in the continuum case); for further discussion of the latter issue see [86].
Finally, we should note that the Myers dielectric effect will arise in many other situations,
such as perturbations of other conformal and nonconformal theories. As one example, con-
sider the perturbation of the BFSS matrix theory given by the dimensional reduction of the
N = 1∗ mass term. This has been used as a means of analyzing the structure of the matrix
theory bound state [87]. Now we see that this deformation has a physical interpretation in
its own right: it is the matrix theory for M theory with a nontrivial boundary condition on
the 4-form field strength. In this background the graviton will blow up into a finite-sized
M2-brane sphere. This same mechanism, the Myers expansion of a highly boosted graviton
in a background field strength, has recently led to a remarkable explanation of the stringy
exclusion principle [88].

B. Gauge theory

The brane solution gives a beautiful representation of a confining gauge theory and many
of the associated phenomena. There are many artifacts of the massive N = 4 matter, so
this is still far from QCD, but we emphasize that it is a confining four-dimensional gauge
theory in its own right. We should perhaps note that we use the Maldacena duality freely
in the entire range 1/N < g < N. Discussion often focuses on large N with fixed gN, but
this is only one part of the interesting range.
Without too much additional work, it should be possible to understand the breaking of
N = 4 to N = 2 Yang-Mills. Much of the moduli space, including the part corresponding

14 N.Arkani-Hamed, S. Dimopoulos, J. Feng, S. Gubser, N. Kaloper, E. Silverstein and F. Wilczek

have emphasized the importance of this point.

59
to the repulson singularity studied in [21], should be visible in supergravity. We might
hope that, as in [22], the Seiberg-Witten Riemann surface will appear in a natural way.
However, where the surface degenerates (at points with light charged particles) we expect
there will be physics lying outside the supergravity regime. Since the transition from N = 2
to N = 1 is described simply in MQCD, we may hope that this breakdown will be described
simply using semiclassical branes, giving a picture of the physics studied in [73]. We also
expect a breakdown at Argyres-Douglas fixed points [89], where the physics, presumably a
supergravity kink connecting one AdS region to another, may not be described in a linear
approximation.
Once the supersymmetric gravity background is completely understood, it should also be
possible to fully explore the nonsupersymmetric case. Since even the confining vacuum is a
small perturbation of an N = 4 Coulomb branch configuration, we do not believe that small
breaking of supersymmetry is likely to have a large impact on the theory. New condensates
and new θ dependence, and a loss of the degeneracy between vacua, should certainly be
seen, but otherwise we see no reason why the nonsupersymmetric confining vacuum must
appear much different from the supersymmetric one. However, while this will be true if the
supersymmetry breaking scale m′ is small compared to m, it will surely be false if m′ ≫ m.
The quantitative question of where the transition lies, as a function of gN, remains to be
explored. Even assuming, however, that m ∼ m′ is within the supergravity regime, this does
not make the study of pure nonsupersymmetric Yang-Mills any easier, since it is a long way
from gN ≫ 1 to the required regime.
Assuming the nonsupersymmetric case can be studied to a degree, an obvious next step
in studying the gravity–gauge theory correspondence is to add massless charged fermions.
It would be most interesting to see chiral symmetry breaking and to determine if and when
the properties of pions lie within the supergravity regime. Adding supersymmetric charged
matter is unfortunately not of great use, since SU(N) with Nf ≪ N massless chiral multi-
plets in the fundamental plus antifundamental representation has no stable vacuum, while
if Nf ∼ N there is as yet no dual string description. The low flavor case might be stud-
ied by taking type IIB on an orientifold, which gives an N = 2 Sp(N) gauge theory with
a hypermultiplet A in the antisymmetric tensor representation and four Fi in the funda-
mental representation [90,91]. As in N = 1∗ , the ultraviolet theory is finite, and could be
perturbed by supersymmetric masses m for the adjoint chiral multiplet and A, and µi for
the Fi . When µi ∼ m, the physics probably resembles N = 1∗ , but the theory will exit the
supergravity regime as any one of the µi go to zero. Breaking the supersymmetry, leaving
only the fermions in Fi and the gauge bosons massless, would be a challenge, but might be
tractable.

60
 In this paper we have not pursued the obvious and important connections of our work
with noncommutative field theory in two additional dimensions. Similar relations are present
already in MQCD. We are especially intrigued that the confining flux tube of large N Yang-
Mills is a nearly-BPS instanton string of a sphere-compactified six-dimensional noncom-
mutative gauge theory (supersymmetric or not.) Note that the baryon vertex is also an
interesting object in this theory. The massive vacua with multiple coincident 5-branes, and
their solitons, may be of considerable interest for the study of nonabelian noncommutative
field theory. This may also be true for lower-dimensional cases, where the classical vacua
are still described by spherical D(p+2) branes.
It is important to note that many of our results bear some resemblance to those seen in
MQCD, which is an unusual compactification of the (2, 0) theory on an M5-brane. Both the
breaking of N = 4 to N = 2 Yang-Mills [22] and the breaking of N = 2 to N = 1 [62] have
been considered, although the full N = 1∗ model has not been constructed in MQCD. It is
interesting to consider some of the similarities and differences with the supergravity dual of
N = 1∗ . In our picture, the vacuum is represented by an NS5/D3 hybrid; in MQCD the
vacuum is given by a multisheeted M5-brane, which, when the radius R10 of the M-theory
circle is small, is an NS5/D4 hybrid. Second, the flux tubes in N = 1∗ are F1 strings
bound to the NS5/D3 hybrid; in MQCD they are M2-branes bound to the M5 brane. The
baryon vertex in MQCD is an M2 brane which extends off of but has a boundary on the M5
brane, analogous to our D3 brane whose boundary is the NS5-brane two-sphere. Domain
walls are in both approaches are 5-branes mediating transitions between different 5-brane
vacua. There are a number of important differences that this list understates; but the most
interesting difference, perhaps, is the instantons at large N. As we have seen, the fractional
instantons noted in [67] are resummed at large gN into strings wrapping the NS5-brane
two-sphere. All of these connections hint at the usual duality between M theory on a torus
and type IIB string theory, but the connection between the two pictures is not as simple
as this, given the absence of a torus in both constructions. It would be interesting to make
these connections precise.
One of the most interesting aspects of the N = 1∗ theory, and any similar gravity dual
of N = 4 broken to N = 2, is that it can be used to great effect to understand more deeply
the connection of field theory with gravity and string theory. Our work and that of [8,66]
points in this direction. The holomorphic properties of these field theories can be completely
understood using field theory and/or M theory, as in [5,22,7], for all values of g and all values
of gN. Consequently, one can distinguish clearly those regimes of parameter space in which
the theory is well-described by electric variables, by magnetic variables, or by IIB string
variables respectively. The various universal and quasiuniversal physical properties of the

61
theory are given different descriptions in each regime (see for example our discussion of the
instanton expansion of the superpotential, section VI.G,) and there are surely many things
to be learned by studying how one regime converts to another. In particular, the possibility
of quantitatively matching one regime to another deserves attention. It is probable that this
will be required if one is to study QCD, which lies mostly outside the supergravity regime. If
such matching is only possible for holomorphic and BPS quantities, as so far seems likely, it
poses yet another obstacle in the quest for a description of the strong interactions. However,
even if the goal of doing calculations in large-N QCD, using string variables, is shown to
be unrealizable, the fact that field theories exhibit behavior which differs greatly from the
physics we have observed so far in nature is of potentially great importance, in that it may
open new avenues for solving old problems.

ACKNOWLEDGMENTS

We would like to thank O. Aharony, N. Arkani-Hamed, I. Bena, M. Berkooz, K. Das-

gupta, S. Dimopoulos, E. Gimon, S. Gubser, A. Hashimoto, S. Hellerman, N. Itzhaki, D.
Kabat, N. Kaloper, R. Myers, N. Nekrasov, K. Pilch, L. Randall, S. Sethi, E. Silverstein, W.
Taylor, N. Warner, E. Witten, and many participants of the ITP Program on Supersymmet-
ric Gauge Theories for helpful discussion of various aspects of this work. M.J.S. is especially
grateful to D. Kabat for discussions and assistance at early stages of this work. We also
especially thank O. Aharony for pointing out important deficiencies in our discussion of the
superpotential and the condensates. The work of J.P. was supported by National Science
Foundation grants PHY94-07194 and PHY97-22022. The work of M.J.S. was supported by
National Science Foundation grant NSF PHY95-13835 and by the W.M. Keck Foundation.

62
 REFERENCES

[1] G. ’t Hooft, Nucl. Phys. B72, 461 (1974).

[2] J. Maldacena, Adv. Theor. Math. Phys. 2, 231 (1998) [hep-th/9711200].
[3] L. Girardello, M. Petrini, M. Porrati and A. Zaffaroni, hep-th/9909047.
[4] C. Vafa and E. Witten, Nucl. Phys. B431, 3 (1994) [hep-th/9408074].
[5] R. Donagi and E. Witten, Nucl. Phys. B460, 299 (1996) [hep-th/9510101].
[6] M. J. Strassler, Prog. Theor. Phys. Suppl. 131, 439 (1998) [hep-lat/9803009].
[7] N. Dorey, JHEP 9907, 021 (1999) [hep-th/9906011].
[8] N. Dorey and S. P. Kumar, JHEP 0002, 006 (2000) [hep-th/0001103].
[9] M. J. Strassler, Nucl. Phys. Proc. Suppl. 73, 120 (1999) [hep-lat/9810059].
[10] E. Witten, Adv. Theor. Math. Phys. 2, 505 (1998) [hep-th/9803131]
[11] S. Rey, S. Theisen and J. Yee, Nucl. Phys. B527, 171 (1998) [hep-th/9803135].
[12] A. Brandhuber, N. Itzhaki, J. Sonnenschein and S. Yankielowicz, JHEP 9806, 001
(1998) [hep-th/9803263].
[13] D. J. Gross and H. Ooguri, Phys. Rev. D58, 106002 (1998) [hep-th/9805129].
[14] R. C. Myers, JHEP 9912, 022 (1999) [hep-th/9910053].
[15] D. Kabat and W. I. Taylor, Phys. Lett. B426, 297 (1998) [hep-th/9712185].
[16] T. Banks, W. Fischler, S. H. Shenker and L. Susskind, Phys. Rev. D55, 5112 (1997)
[hep-th/9610043].
[17] P. Kraus, F. Larsen and S. P. Trivedi, JHEP 9903, 003 (1999) [hep-th/9811120].
[18] I. R. Klebanov and E. Witten, Nucl. Phys. B556, 89 (1999) [hep-th/9905104].
[19] D. Z. Freedman, S. S. Gubser, K. Pilch and N. P. Warner, hep-th/9906194.
[20] A. Brandhuber and K. Sfetsos, hep-th/9906201.
[21] C. V. Johnson, A. W. Peet and J. Polchinski, hep-th/9911161;
L. Jarv and C. V. Johnson, hep-th/0002244.
[22] E. Witten, Nucl. Phys. B500, 3 (1997) [hep-th/9703166].
[23] L. Randall and R. Sundrum, Phys. Rev. Lett. 83, 4690 (1999) [hep-th/9906064].
[24] N. Arkani-Hamed, S. Dimopoulos, N. Kaloper and R. Sundrum, hep-th/0001197;
S. Kachru, M. Schulz and E. Silverstein, hep-th/0001206.
[25] A. Khavaev, K. Pilch and N. P. Warner, hep-th/9812035.
[26] A. Karch, D. Lust and A. Miemiec, Phys. Lett. B454, 265 (1999) [hep-th/9901041].
[27] P. C. Argyres, K. Intriligator, R. G. Leigh and M. J. Strassler, hep-th/9910250.
[28] G. ’t Hooft, Nucl. Phys. B138, 1 (1978); Nucl. Phys. B153, 141 (1979); Nucl. Phys.
B190, 455 (1981).
[29] E. Witten, Phys. Lett. B86, 283 (1979).
[30] C. Montonen and D. Olive, Phys. Lett. B72, 117 (1977).

63
[31] J. Polchinski, Phys. Rev. Lett. 75, 4724 (1995) [hep-th/9510017]; hep-th/9611050.
[32] E. Witten, Adv. Theor. Math. Phys. 2, 253 (1998) [hep-th/9802150].
[33] S. S. Gubser, I. R. Klebanov and A. M. Polyakov, Phys. Lett. B428, 105 (1998) [hep-
th/9802109].
[34] V. Balasubramanian, P. Kraus and A. Lawrence, Phys. Rev. D59, 046003 (1999)
[hep-th/9805171]; T. Banks, M. R. Douglas, G. T. Horowitz and E. Martinec, hep-
th/9808016.
[35] L. Girardello, M. Petrini, M. Porrati and A. Zaffaroni, [hep-th/9810126].
[36] J. Distler and F. Zamora, [hep-th/9810206].
[37] D. Z. Freedman, S. S. Gubser, K. Pilch and N. P. Warner, hep-th/9904017.
[38] L. Girardello, M. Petrini, M. Porrati and A. Zaffaroni, [hep-th/9903026].
[39] S. S. Gubser, hep-th/0002160.
[40] G. T. Horowitz and R. Myers, Gen. Rel. Grav. 27, 915 (1995) [gr-qc/9503062].
[41] O. de Wolfe, D. Freedman, S. Gubser, K. Pilch, N. Warner, E. Witten, private commu-
nications.
[42] E. Bergshoeff, H. J. Boonstra and T. Ortin, Phys. Rev. D53, 7206 (1996) [hep-
th/9508091].
[43] J. H. Schwarz, Nucl. Phys. B226, 269 (1983).
[44] H. J. Kim, L. J. Romans and P. van Nieuwenhuizen, Phys. Rev. D32, 389 (1985).
[45] M. Gunaydin and N. Marcus, Class. Quant. Grav. 2, L11 (1985).
[46] R. G. Leigh, Mod. Phys. Lett. A4, 2767 (1989);
M. Li, Nucl. Phys. B460, 351 (1996) [hep-th/9510161];
M. R. Douglas, hep-th/9512077.
[47] M. Cederwall, A. von Gussich, B. E. Nilsson, P. Sundell and A. Westerberg, Nucl. Phys.
B490, 179 (1997) [hep-th/9611159];
E. Bergshoeff and P. K. Townsend, Nucl. Phys. B490, 145 (1997) [hep-th/9611173].
[48] J. Polchinski, hep-th/9901076;
L. Susskind, hep-th/9901079.
[49] M. Alishahiha, Y. Oz and M. M. Sheikh-Jabbari, JHEP 9911, 007 (1999) [hep-
th/9909215].
[50] C. G. Callan, J. A. Harvey and A. Strominger, Nucl. Phys. B359, 611 (1991).
[51] E. Witten, JHEP 9807, 006 (1998) [hep-th/9805112].
[52] S. S. Gubser and I. R. Klebanov, Phys. Rev. D58, 125025 (1998) [hep-th/9808075].
[53] S. Gukov, M. Rangamani and E. Witten, JHEP 9812, 025 (1998) [hep-th/9811048].
[54] S. Rey and J. Yee, hep-th/9803001.
[55] J. Maldacena, Phys. Rev. Lett. 80, 4859 (1998) [hep-th/9803002].

64
[56] We thank O. Aharony, M. Berkooz and S. Sethi for their insights.
[57] N. Nekrasov and A. Schwarz, Commun. Math. Phys. 198, 689 (1998) [hep-th/9802068];
B. Pioline and A. Schwarz, JHEP 9908, 021 (1999) [hep-th/9908019].
[58] N. Itzhaki, J. M. Maldacena, J. Sonnenschein and S. Yankielowicz, Phys. Rev. D58,
046004 (1998) [hep-th/9802042].
[59] G. T. Horowitz and J. Polchinski, Phys. Rev. D55, 6189 (1997) [hep-th/9612146].
[60] A. Hanany and E. Witten, Nucl. Phys. B492, 152 (1997) [hep-th/9611230].
[61] G. Dvali and M. Shifman, Nucl. Phys. B504, 127 (1997) [hep-th/9611213]; Phys. Lett.
B396, 64 (1997) [hep-th/9612128].
[62] E. Witten, Nucl. Phys. B507, 658 (1997) [hep-th/9706109].
[63] A. Sen, JHEP 9809, 023 (1998) [hep-th/9808141].
[64] G. Dvali, G. Gabadadze and Z. Kakushadze, Nucl. Phys. B562, 158 (1999) [hep-
th/9901032].
[65] R. Kirchner, I. Montvay, J. Westphalen, S. Luckmann and K. Spanderen [DESY- Mun-
ster Collaboration], Phys. Lett. B446, 209 (1999) [hep-lat/9810062];
A. Feo, R. Kirchner, I. Montvay and A. Vladikas [DESY-Munster Collaboration], hep-
lat/9909071.
[66] O. Aharony, N. Dorey and S. P. Kumar, to appear.
[67] J. H. Brodie, Nucl. Phys. B532, 137 (1998) [hep-th/9803140].
[68] R. G. Edwards, U. M. Heller and R. Narayanan, Phys. Lett. B438, 96 (1998) [hep-
lat/9806011].
[69] A. W. Peet and J. Polchinski, Phys. Rev. D59, 065011 (1999) [hep-th/9809022].
[70] C. G. Callan, J. A. Harvey and A. Strominger, Nucl. Phys. B367, 60 (1991).
[71] N. Seiberg and E. Witten, Nucl. Phys. B426, 19 (1994) [hep-th/9407087].
[72] A. Klemm, W. Lerche, S. Yankielowicz and S. Theisen, Phys. Lett. B344, 169 (1995)
[hep-th/9411048]; P. C. Argyres and A. E. Faraggi, Phys. Rev. Lett. 74, 3931 (1995)
[hep-th/9411057].
[73] M. R. Douglas and S. H. Shenker, Nucl. Phys. B447, 271 (1995) [hep-th/9503163].
[74] K. Hori, H. Ooguri and Y. Oz, Adv. Theor. Math. Phys. 1, 1 (1998) [hep-th/9706082].
[75] A. Hanany, M. J. Strassler and A. Zaffaroni, Nucl. Phys. B513, 87 (1998) [hep-
th/9707244].
[76] J. de Boer and Y. Oz, Nucl. Phys. B511, 155 (1998) [hep-th/9708044].
[77] K. Pilch and N. P. Warner, hep-th/0002192.
[78] O. Aharony, S. S. Gubser, J. Maldacena, H. Ooguri and Y. Oz, Phys. Rept. 323, 183
(2000) [hep-th/9905111].
[79] M. R. Douglas and G. Moore, hep-th/9603167.

65
[80] M. Schmaltz, Phys. Rev. D59, 105018 (1999) [hep-th/9805218].
[81] H. Verlinde, hep-th/9906182.
[82] W. D. Goldberger and M. B. Wise, Phys. Rev. Lett. 83, 4922 (1999) [hep-ph/9907447].
[83] I. R. Klebanov and N. A. Nekrasov, hep-th/9911096.
[84] I. R. Klebanov and A. A. Tseytlin, hep-th/0002159.
[85] J. D. Brown and C. Teitelboim, Phys. Lett. B195, 177 (1987).
[86] R. Bousso and J. Polchinski, hep-th/0003xyz.
[87] M. Porrati and A. Rozenberg, Nucl. Phys. B515, 184 (1998) [hep-th/9708119];
G. Moore, N. Nekrasov and S. Shatashvili, Commun. Math. Phys. 209, 77 (2000) [hep-
th/9803265];
N. A. Nekrasov, hep-th/9909213.
[88] J. McGreevy, L. Susskind and N. Toumbas, hep-th/0003075.
[89] P. C. Argyres and M. R. Douglas, Nucl. Phys. B448, 93 (1995) [hep-th/9505062].
[90] O. Aharony, J. Sonnenschein, S. Yankielowicz and S. Theisen, Nucl. Phys. B493, 177
(1997) [hep-th/9611222].
[91] O. Aharony, A. Fayyazuddin and J. Maldacena, JHEP 9807, 013 (1998) [hep-
th/9806159].

66