hep-th/0008142

HUTP-00/A035

Superstrings and Topological Strings at Large N

arXiv:hep-th/0008142 v2 28 Nov 2000

Cumrun Vafa

Jefferson Physical Laboratory

Harvard University
Cambridge, MA 02138, USA

Abstract
We embed the large N Chern-Simons/topological string duality in ordinary super-
strings. This corresponds to a large N duality between generalized gauge systems with
N = 1 supersymmetry in 4 dimensions and superstrings propagating on non-compact
Calabi-Yau manifolds with certain fluxes turned on. We also show that in a particular
limit of the N = 1 gauge theory system, certain superpotential terms in the N = 1 sys-
tem (including deformations if spacetime is non-commutative) are captured to all orders in
1/N by the amplitudes of non-critical bosonic strings propagating on a circle with self-dual
radius. We also consider D-brane/anti-D-brane system wrapped over vanishing cycles of
compact Calabi-Yau manifolds and argue that at large N they induce a shift in the back-
ground to a topologically distinct Calabi-Yau, which we identify as the ground state system
of the Brane/anti-Brane system.

August 2000
1. Introduction

The idea that large N gauge theories should have a phase described by perturbative
strings, set forth by ‘t Hooft [1], has been beautifully realized by various examples. The first
example of this kind was found by Kontsevich [2], which relates the bosonic string theory
coupled to certain matter ((1, 2) minimal model, which is equivalent to pure topological
gravity formulated by Witten [3]), to a matrix integral with cubic interaction (which can
be viewed as a particular gauge theory in zero dimensions)1 . Many more examples were
also found in the context of non-critical bosonic strings. For example, it was found [4] that
bosonic strings propagating on a circle with self-dual radius is equivalent to Penner matrix
model [5].
More recently it was recognized that ‘t Hooft’s conjecture is also realized even for
much more complicated and physically more interesting gauge theories [6][7][8]. In partic-
ular certain gauge theories at large N are equivalent to superstrings propagating on AdS
backgrounds. Another example of a string/large N duality was discovered in [9], where it
was shown that large N limit of Chern-Simons gauge theory on S 3 is equivalent to topo-
logical strings on a non-compact Calabi-Yau threefold which is a blow up of the conifold
(given by O(−1)+O(−1) bundle over P1 ). This duality was tested to all orders in the 1/N
expansion including checks at the level of Wilson Loop observables of the Chern-Simons
theory [10][11]. It is also known [12] that in some limit (large N , fixed Chern-Simons
coupling k) this theory has the same partition function as bosonic strings at the self-dual
radius.
This paper was motivated by trying to connect the duality discovered in [9] with the
dualities discovered in the context of AdS/CFT correspondences. The basic idea is to con-
sider type IIA superstring propagating in the conifold background (which is symplectically
the same as T ∗ S 3 ) in the presence of N D6 branes wrapped around S 3 and filling the
spacetime. It has been known [13] that the topological string amplitudes for the internal
theory on the non-compact Calabi-Yau compute superpotential terms on the left-over R4
worldvolume of the D6 brane. On the other hand it is also known that the internal topo-
logical string theory with N D-branes wrapped on S 3 is equivalent to Chern-Simons gauge
theory on S 3 [14]. Thus the duality found in [9] suggests that type IIA string on the coni-
fold with N D6 branes is equivalent to the blown up version of the conifold with no branes
1
This is not the same as the old matrix model which discretizes the worldsheet–rather it is
the target space description exactly in line with ‘t Hooft’s conjecture.

1
left over. At first sight this sounds strange, because having no D-branes left would naively
suggest a theory with N = 2 supersymmetry rather than N = 1. Moreover Ramond fluxes
should also be turned on in the blown up geometry corresponding to the flux generated by
the D6 brane. The main puzzle was why in the dual topological string theory discovered
in [9] there is no mention of RR fluxes? Indeed it is an ordinary topological string (the
A-model) on the blown up conifold.
The resolution turns out to be that turning on the RR flux does not affect the topo-
logical string amplitudes, and the dual string theory does involve RR fluxes. Turning on
RR flux, however, does generate an N = 1 superpotential term [15][16], which can be
computed in terms of the topological string amplitudes. Thus the duality found in [9]
can be viewed as an all order check in the 1/N expansion for the N = 1 superpotential
computations in the context of this type IIA superstring/gauge theory duality. One can
also consider the mirror symmetry acting on all these statements, which as noted in [9]
give rise to similar dualities. In the superstring realization, the mirror case (in a certain
limit) would correspond to considering type IIB string on the blow up of the conifold with
N D5 branes wrapped on P1 and we end up with type IIB on deformed conifold geometry
T ∗ S 3 but with RR flux turned on.
One can also consider wrapped D-brane in the context of compact Calabi-Yau man-
ifolds. However in this case we also need to put anti-D-branes, in order to have no net
D-branes. In this case we conjecture that the large N limit will correspond to having
a new Calabi-Yau with fluxes, which can decay as discussed in [17] to a theory with no
fluxes left-over and with supersymmetry increased to N = 2. The effect of the non-BPS
states has been to shift the background to a new background. This is a novel way of de-
forming backgrounds, and as we will suggest later in the paper may have many interesting
extensions.
The organization of this paper is as follows: In section 2 we review aspects of topolog-
ical string amplitudes and what they compute in the corresponding superstring theory. In
section 3 we revisit the duality of [9] and embed it in the context of Type IIA superstrings.
In section 4 we apply mirror symmetry to the statements in section 3 and discuss the
equivalent Type IIB superstring theory. In section 5 we discuss possible applications of
c = 1 non-critical bosonic strings to the question of generation of superpotential in the
large N limit of N = 1 supersymmetric gauge theory. In section 6 we discuss wrapped
brane/anti-brane systems in the context of compact Calabi-Yau manifolds and use the

2
above duality to make new predictions about the shift in the background. In section 7 we
discuss some generalizations of this work.
While preparing this paper, three papers appeared which have overlaps with different
aspects of our work. In particular [18][19] have some overlap with our work in the context
of large N duals of N = 1 gauge theories in the context of type IIB strings, which we will
briefly comment on in section 4. Also the same configuration of wrapped D-branes/anti-
D-branes considered in section 6 was also studied in [20] in a different context.

2. Topological Strings and Superstrings

In this section we discuss aspects of topological strings and their relevance for super-
potential computations in the corresponding superstring compactifications. We will divide
our discussion to two parts: Closed string case (i.e. without D-branes) and open string case
(i.e. including D-branes). We also point out the relevance of topological string amplitudes
for N = 1 superpotential computations when RR-fluxes are turned on.

2.1. Closed topological string and superstring amplitudes in 4d

Consider A-model topological strings on a Calabi-Yau manifold K (similar remarks

apply to the mirror B-model). For simplicity of notation let us assume that the CY
manifold has only one Kahler class, parameterized by the complexified Kahler parameter
t. Then closed topological string amplitude on K is given by
X
F (t, λs ) = λs2g−2 Fg (2.1)
g

X
Fg = Fd,g e−dt
d

where, roughly speaking Fd,g denote the “numbers” (Gromov-Witten invariants) of genus g
curves in class d. The topological strings compute certain amplitudes in the corresponding
type IIA superstring compactifications on the Calabi-Yau [13][21][22]. In particular they
compute terms in the action of the form
Z
d4 θW 2g Fg (t) = gR2 F 2g−2 Fg (t) + ...

where Wαβ denotes the graviphoton field strength multiplet, R2 and F 2g−2 denote certain
contractions of the self-dual part of the Riemann tensor and of the gravi-photon field

3
strength, and t denotes the vector superfield with the vev of the lowest component being
the Kahler parameter t. One way to derive this formula is to note that with 2g−2 insertions
of the spin operator, needed to compute the amplitude involving the F 2g−2 , the ordinary
sigma model is topologically twisted. At genus 0 what one gets is
Z
d4 θF0 (t) = ∂ 2 F0 (t)F t ∧ F t + ...

where F t denotes the (self-dual part of the) U (1) field strength in the same multiplet as t.
In the type IIA this arises from the 4-form field strength G by setting it to

G = F t ∧ ωt

where ωt denotes the Kahler form associated to t.

It is natural to ask what changes in the closed topological string computations when
we turn on some RR flux in the target space. The choices are2 the 2-form field strength
in the internal space F , 4-form field strength Gint along the internal CY directions and
the G along the spacetime directions G4 , which we equivalently study in terms of the dual
6-form field strength G6 = ∗G4 . It turns out that the topological string amplitudes in
the presence of RR fields is not modified at all! This is particularly simple to show in
the Berkovits formalism [23][22][24]. Instead of demonstrating it in this way we follow a
related idea, which we will need later in this paper, by studying the generation of N = 1
superpotential terms in the presence of RR fluxes, which we will discuss next.

2.2. Generation of superpotential due to internal field strength

RR fluxes have been studied in the context of CY compactifications [25][26][15][16].

In particular it has been shown in [15][16] that turning on internal field strength in the CY
leads to generation of superpotential terms in 4d N = 1 theory (see also similar situations
considered in [27][28]). In the context of type IIA theory with RR fluxes corresponding to
F and Gint and G6 discussed above, the superpotential is given by
Z Z Z
λs W = F ∧ k ∧ k + i G ∧ k + G6 (2.2)

2
We can also include the 0-form field strength dual to 10 form field strength in type IIA, but
since we will not deal with it in this paper we will not discuss it. It will give rise to an N = 1
R
superpotential of the form G0 ∧ k 3 .

4
where k is the complexified Kahler class. To see this one considers the BPS charge in the
presence of BPS domain walls which may be partially wrapped over the CY. For example,
considering a D6 brane wrapped over 4-cycles of CY gives a domain wall with BPS tension
1
R
λs
k ∧k integrated over the internal part of the 6-brane. This in turn shifts the dual F by
one unit. This BPS formula should be captured by a ∆W and we can see from the above
form of (2.2) that the first term above precisely captures this term. More precisely what
we mean by the formula (2.2) is the worldsheet quantum corrected formula for the kahler
forms (as is well known in the context of mirror symmetry the mass of the D-branes receives
corrections by the worldsheet instantons). In particular if t denotes the complexified area
of the basic 2 cycle, then the volumes of the 0, 2, 4 and 6 cycles are given by

∂F0 ∂F0
1, t, , 2F0 − t
∂t ∂t

where F0 is the genus zero topological string amplitude. So in particular suppose we have
N units of the F flux through the basic 2-cycle, where t denotes the complexified area of
this 2-cycle. Then the first term in (2.2) is equivalent to
Z
∂F0
F ∧k∧k =N
∂t

Similarly if we considered D4 branes wrapped over 2 cycles and D2 branes with no wrap-
pings, we deduce the existence of the second and third term in (2.2). In particular if we
denote the fluxes of F , Gint , G6 by integers N, L, P relative to integral 2, 4 and 6 cycles,
we have
∂F0
λs W = N + itL + P (2.3)
∂t
Note that equation (2.2) can also be written in the form
Z Z
λs W = (F + i ∗ G) ∧ k + G6 (2.4)

where again here by ∗ we mean the worldsheet quantum corrected ∗ operation.

Now we come to the discussion of why turning on RR fluxes should not modify the
topological amplitudes. We will concentrate on genus 0 amplitudes (similar arguments can
be advanced for the higher genus amplitudes as well). The vector superfield with bottom
component t has an auxiliary field in the superspace of the form

t + θ 2 (F + ∗iG) + ...

5
where F and G are the usual RR fluxes of the internal Calabi-Yau3 . In the usual supersym-
metric background they are set to zero. Now suppose we wish to turn them on. Suppose
for example we wish to turn on N units of F . Consider the topological string amplitude
F0 (t). We claim that this already yields the correct structure for the generation of N = 1
superpotential precisely if F0 is unmodified in the presence of RR flux. To see this note
that using the expansion of t in terms of the RR field strength auxiliary fields we have
Z Z
∂F0
d θF0 (t) = d2 θN
4
∂t

which is exactly the expected answer if F0 is unmodified. Similarly turning on the Gint
flux and using (2.4) we see that the term in (2.2) involving Gint will also have the correct
structure if F0 is unmodified.
There is another auxiliary field in the vector multiplet which come from the NS-NS
sector which is relevant for us. This corresponds to the field strength associated with the
lack of integrability of complex structure. In particular if we write D = ∂ + A∂, where A
is an anti-holomorphic one form taking values in the tangent bundle, then

2
D = (∂A + [A, A])∂ = F ∂

where F is an anti-holomorphic 2-form with values in the tangent bundle which is equiv-
alent, by lowering the vector index by the three form, to a (2, 2) form. If this is non-
vanishing it also corresponds to making the (3, 0) form in the CY not to be annihilated by
∂. These turn out effectively to add to F and Gint complex pieces of the form iF NS /λs
and iGNS
int /λs . A similar N S auxiliary field gives rise effectively to the complex part of
∗G6 .4 In other words, even with these fields turned on, the formula (2.3) remains correct
but now N, L, P also include imaginary pieces of the form iNI /λs , iL/λs , iP/λs . We will
continue to denote the superpotential as (2.3) and keep in mind that N, L and P can have
complex pieces given by an integer over λs .
Turning these vevs on breaks the N = 2 supersymmetry to N = 1. The field t is
now the bottom component of an N = 1 chiral multiplet whose auxiliary field descends
from another auxiliary field (which also comes from the NS sector) in the original N = 2
multiplet which is not turned on.
3
I have greatly benefited from discussions with Nathan Berkovits in connection with the
auxiliary field structure of the superfields.
4
Turning these fields on is mirror to turning on HN S on the mirror CY.

6
 Note also that the higher genus topological amplitudes also give rise to certain N = 1
superpotential terms when the auxiliary field of the N = 2 multiplet t takes a vev. In
particular with N units of RR flux for F we get
Z Z
4 2g ∂Fg
d θ W Fg (t) → N d2 θ W 2g (2.5)
∂t
where we continue to denote by Wαβ the reduction of the N = 2 multiplet to an N = 1
multiplet with the self-dual part of the graviphoton field strength as its bottom component.
So in conclusion we have learned that the topological string amplitudes are not sensi-
tive to turning on RR field strengths, but they are useful in determining the superpotential
terms that will be generated once certain RR and NS fields take a vev. This is captured
by equation (2.3).

2.3. Open Topological Strings and N=1 amplitudes in 4d

In the A-model, the open topological string corresponds to studying holomorphic

maps from worldsheet with boundaries to the target space where the boundary lies on a
3-dimensional Lagrangian subspace of the CY, i.e. the 3-dimensional topological version of
D-brane [14]. Moreover, it was shown that topological string field theory in this case is just
the Chern-Simons gauge theory on the corresponding Lagrangian submanifold (possibly
corrected with non-trivial worldsheet instantons). The implications of these theories for
superstring amplitudes has been studied as well. In particular if we consider type IIA
superstring in the presence of a CY with N D6 branes wrapping a lagrangian 3-cycle of
CY and filling the rest of the spacetime we get an N = 1 gauge theory with SU (N ) gauge
group in 4d. Then it was shown in [13] that for example the genus 0 open topological
string amplitudes compute corrections of the form
XZ
λs W = d2 θ F0,h [N h S h−1 ] + αS + β (2.6)
h

where F0,h is the partition function of the topological string at genus 0 with h holes, and
S = λs T rW 2 where Wα is the chiral superfield with gaugino as its bottom component5 .
5
The coefficient of N h in front arises because, as discussed in [13], we have to choose h − 1
holes to put the trW 2 fields and this can be done in h ways and also the trace over the hole without
a field gives a factor of N . Note also that T rW 2 which is a fermion bilinear is nilpotent, in the
sense that (T rW 2 )k = 0 for k > N 2 . It is relevant for us precisely because we are considering a
large N limit. Thus in the large N limit the gaugino bilinear can even have a classical vev.

7
Here we have shown explicitly the contribution coming from h = 2 in the form of αS. This
term, coming from annulus, is typically divergent, signifying the RG flow of the coupling
constant of the gauge theory and needs regularization. Also we have added a constant β
to remind us that we cannot fix that from open topological string considerations. Here h
is the number of holes on the sphere and F0,h denotes the topological amplitude on the
sphere with h holes. If the target space has some Kahler moduli t they will correspond
to chiral fields in the N = 1 theory in 4 dimensions and F0,h will depend on t. The case
of h = 1 in the above formula was recently discussed in [29][30][10]. Some superstring
implications of higher genus open topological strings, i.e. Fg,h with arbitrary g, has also
R
been noted in [10]. In particular they compute terms of the form d2 θ Fg,h W 2g [N hS h−1 ].
Let us define the open topological string amplitude summed over all holes at a fixed
genus by
X
Fgopen (r) = Fg,h r h .
h

Then, for example the genus 0 open topological string amplitude computes the following
correction to the superpotential
Z
2 ∂F0open (S)
λs W = N d θ + αS + β
∂S

This is strikingly similar to the form obtained in (2.3) in the context of closed topological
string amplitudes. Similarly the higher genus correction computes terms of the form
Z
∂Fgopen (S)
N d2 θ W 2g
∂S
which is also similar to the higher genus correction obtained in the closed string context
in the presence of flux (2.5). The main difference being that S is an operator for the open
string amplitudes but t is a parameter in the closed string setup. Nevertheless we will see
in the next section why this is not an accidental similarity and provides the superstring
interpretation of the duality found in [9], when S takes an expectation value equal to t.

3. Embedding Large N Topological String duality in Superstrings

Consider Type IIA strings in a non-compact CY 3-fold geometry of the form of the
conifold times the Minkowski space M 4 : The internal geometry is given by

f = x21 + ... + x24 = ρ

8
where each xi parameterizes C. The real subspace of the above geometry is S 3 (for real
ρ) and the imaginary directions sweep the cotangent direction of S 3 . The volume of S 3 in
Q
string units is given by ρ (here we are taking the canonical 3-form Ω = dxi /df , which
scales as ρ to give the volume). Thus symplectically the conifold is T ∗ S 3 . Consider N D6
branes wrapped over the S 3 of the conifold and filling the rest of the spacetime. On the
uncompactified worldvolume of the D-brane we have an N = 1 supersymmetric SU (N )
gauge theory. Note that to leading order the action on the uncompactified worldvolume of
the D-branes is given by the superpotential
Z
1
d2 θ SY (3.1)
λs

where S = λs T rW 2 and where Y denotes the N = 1 chiral superfield with its bottom
component given by iC+ λρs , where C is the vev of the 3-form gauge field on IIA (normalized
with periodicity 2π) and plays the role of the theta angle for the gauge theory and ρ denotes
the volume of the S 3 .
The choice of this type IIA geometry is based on the desire to utilize the topological
open/closed string duality. In particular as discussed in the previous section the open
topological string in this case computes corrections to the superpotential of the form
Z
N
d2 θ ∂S F0open (S)
λs

The topological A-model is insensitive to complex structure. In particular F open is inde-

pendent of ρ except for a linear terms in S (coming from the annulus) written in (3.1),
which is related to the ambiguity of open topological string at the level of annulus. There
is also a divergence of the annulus amplitude corresponding to the running of the gauge
coupling constant, which, in the regularized form, can be viewed as addition of a linear
term in S. The corrections above to the simple Y S = Y T rW 2 involve higher dimension
operators (more powers of S) and are captured by the open string amplitude which coin-
cide with the large N expansion of the Chern-Simon amplitudes on S 3 . Note also that
the fact that they are independent of Y implies that they survive no matter what the size
of the S 3 is.
Now we wish to consider the limit where we consider the N → ∞ limit keeping N λs
fixed. In this limit, the analog of ‘t Hooft coupling for the gauge system is given by

1 Y
2 → .
N gY M N λs

9
which remains fixed in this limit. We would like to consider the gravity dual of this gauge
system. In the spirit of AdS/CFT correspondence we will have to consider the near horizon
geometry. What the precise notion of “near” horizon geometry in this case should be is
more subtle because the expectation value of Y undergoes an RG flow, as noted above,
and it will depend at which scale we are probing it. In other words we have to readjust
the size of Y depending on how close we are approaching the branes. The limit should be
such that the S 3 has zero size when we probe it in the UV of the gravitational side but
finite size in the IR. To avoid such subtleties we try to look for a consistent gravitational
background which the branes create. In particular we should find an S 2 of finite size
emerging, surrounding the S 3 , with the D-branes completely disappeared and replaced by
the corresponding fluxes. In the case at hand, since we have N D6 branes wrapping the S 3
in the geometry after transition we should get N units of the 2-form RR flux F through
the dual S 2 . We will now turn to studying this geometry.

3.1. Type IIA Superstring on the blown up geometry

We thus seek the dual large N stringy description of the above gauge system, in the
form of the Type IIA background with the blown up conifold geometry, i.e. the geometry
corresponding to O(−1) + O(−1) bundle over P1 , with N units of 2-form F flux through
P1 . However we must also have internal 4-form and 6-form fluxes (in the form of NS and
RR fields discussed before).
That there should be an NS 4-form flux corresponds to the fact that the size of the
R
S 3 is changing (i.e. that Ω is no longer closed and ρ = S 3 Ω changes), inducing a running
of the gauge coupling constant. Moreover to preserve N = 1 supersymmetry for a finite
value t of the complexified area of the blown up P1 we need both 4-form as well as 6-form
fluxes. In fact, as discussed in the previous section and summarized in equation (2.3) we
have a superpotential of the form

W = N ∂t F0 (t) + itL + P (3.2)

where for the geometry at hand F0 (t) is, up to a cubic polynomial, the tri-logarithm
function, given by
1 3 X e−nt
F0 (t) = t − + P2 (t). (3.3)
6 n>0
n3

10
(where P2 (t) is a polynomial of order 2 in t and is somewhat ambiguous). Similarly for
higher genus Fg we have

B2g X B2g B2g−2

Fg (t) = n2g−3 e−nt + g>1
2g(2g − 2)! n>0 2g(2g − 2)(2g − 2)!

t 1
F1 (t) = + log(1 − e−t )
24 12
where B2g denotes the Bernoulli numbers. The terms involving e−nt in the above formula
reflects the corrections due to worldsheet instantons wrapping n times over the P1 .
The content of the duality obtained in [9] is that

FgOpen (S) = Fg (t)

for all g if we set S = t. We now try to interpret this statement in the superstring context.
For this we need to study solutions to the gravitational equations.
Typically in physics and mathematics when we try to solve some system of equations,
there are topological obstructions that have to be shown to be absent. Once they are
shown to be absent then a solution exists. For example, when we are trying to find Ricci-
flat metrics on Kahler manifolds we need the first chern class of the manifold to be zero.
In fact this is also sufficient for being able to find a Calabi-Yau metric. Of course explicit
solution for the metric has not been possible in almost all cases and in fact the Ricci-flat
metric is only an approximate metric which gives rise to a conformal worldsheet theory.
In a sense the topological condition, guaranteeing the existence of the solution is more
fundamental than the solution itself.
Now we come to the case at hand. To preserve N = 1 supersymmetry we need
W = dW = 0. Once these are satisfied, we expect physically that there must be a
solution to the gravitational system. In fact a very similar example with the same number
of supercharges (namely 4) was already studied from this point of view. Namely if we
consider M-theory on Calabi-Yau 4-fold with G-flux turned on, the gravitational equations
have been studied in [31]. The topological conditions they find for the existence of the
gravitational solution has been shown to be identical to the condition that W = dW = 0
[27].
Of course the low energy gravitational equation in the present case can also be studied
similar to what was done in [31] and will involve warped geometries mixing the spacetime
with the Calabi-Yau. Even though solving the gravity equations would be interesting, we

11
have to remember that due to worldsheet instantons wrapping the P1 there are important
corrections to the gravity equations, and so at best we can trust the low energy gravity
description in the limit of large t. Nevertheless, as already noted above, the superpotential
terms including the corrected string geometry, can be incorporated to all orders in the W
which is computable by topological string amplitudes.
Before even solving the conditions W = dW = 0 we can already interpret the duality
of [9] in the superstring context. If we compare the equation (3.2) with the superpotential
given in the gauge theory side namely W = N ∂S F0open (S) + αS + β we see that they are
identical in form with an appropriate identification of α and β with L and P . Therefore,
since the vacuum in the gauge theory side, as well as the moduli in the gravity side
correspond to W = dW = 0, and W has the same form for the gauge as well as the
gravitational system, this will identify

hSi = hλs T rW 2 i = t Fgopen (S) = Fg (t) (3.4)

Thus the condition of vacuum configuration which sets hSi = t also translates the duality
found in [9] to the match between amplitudes in the gravity side and the gauge theory side
to all orders in 1/N at least as far as superpotential terms are concerned!
Note that the idea that hSi =
6 0, i.e. that we have gaugino condensation, is very
natural for the open string theory under discussion as it does have an N = 1 Yang-Mills
theory associated with it. Part of the above check involves, on the gauge theory side, the
statement that gaugino condensation generates superpotential terms captured through
topological open string amplitudes and in fact this was already pointed out in [13]. Of
course here we have a more refined gauge theory than just the N = 1 Yang-Mills theory
and in particular we have in the open string system, also the higher dimension operators
present, which are captured by the topological string amplitude. At any rate, the result
of [9] strongly suggests not only the existence of a large N duality involving this N = 1
brane system with this closed string background, but also that the gaugino condensation
takes place.
We now come to finding solutions to the equations W = dW = 0 on the gravity
side, which should guarantee the existence of a solution. There are four parameters under
control: the modulus t and the fluxes N, L, P . The two equations W = dW = 0

∂t W = 0 → N F0′′ + iL = 0 → L = iN F0′′

12
 W = 0 → P = −N F0′ + N tF0′′

imply that two of these four quantities are fixed in terms of the other two. The N is of
course fixed for us by the number of D6 branes. As is clear from our description of the dual
gauge system the choice of a shift in L is related to a shift in the bare coupling constant
of the gauge system. In particular in order to agree with the bare coupling constant of
the gauge theory iL = ρ/λs , where ρ is the volume of the S 3 where the D6 branes are
wrapped around. Thus the value of t (and also of P ) is fixed and from (3.3) and ∂t W = 0
the solution for t is given by

[c(et − 1)]N = exp(−ρ/λs ) (3.5)

The constant c depends on the ambiguities hidden in the P2 (t). As we will argue from the
dual gauge theory description, it should be fixed in our case (by a suitable regularization
of the one loop divergence of the gauge theory) to be c ∼ N λs .
Next we turn to the question of how the dynamics of the gauge system is reflected in
the W and the other superpotential terms. What do we expect for the dynamics of the
N = 1 supersymmetric theory living on the D6 brane? If we ignore the higher powers of S
in the superspace integral, i.e. if we ignore the higher order operators, as already discussed,
the leading term with the lowest number of derivatives, is given by the superpotential term
Z
1
d2 θ SY
λs

where S = λs T rW 2 and Y = iC + λρs where ρ denotes the size of the S 3 in the string frame.
In the usual geometric engineering of standard N = 1 gauge theories, and in particular
the ones discussed in [32] one considers the limit where ρ is large, in which case the field Y
gets demoted to a parameter in the lagrangian (the corresponding D-term involving Y Y
becomes very large). However here we are not necessarily interested only in a regime where
Y is very large. In other words we consider the field Y to be a dynamical field. Thus we
have a non-standard N = 1 supersymmetric gauge theory with its coupling constant as a
dynamical field. Even though it is somewhat unconventional, as we will now argue some of
the basic features of this theory are similar to that of N = 1 QCD, in the limit where we
R
ignore the higher derivative terms of the form d2 θ[trW 2 ]k . In other words, if we consider
the field space where S = trW 2 << 1 (in string units) we have a theory which is more
or less similar to N = 1 supersymmetric Yang-Mills theory. Even though we do not have
to restrict our attention only to this limit, and the duality with gravity holds regardless

13
of which field configuration in S we consider, it is first instructive to consider the small S
region to gain intuition for what this theory is.
In the dynamics of N = 1 supersymmetric gauge theory, a prominent role is played
by instantons. Here a similar effect exists: In particular if we consider Euclidean D2 brane
instantons wrapping the S 3 the superpotential gets corrected. Moreover this can also be
viewed as point-like instantons for the SU (N ) gauge theory. To have the right number
of fermionic zero modes to lead to a chiral superspace potential we need 1/N -th of this
instanton. Since the action for this instanton is e−Y , the term that can appear in the
action is e−Y /N . The coefficient in front of it should be of order N 2 (as argued in [33]).
So we must have the effective superpotential given by
Z
1
W = d2 θ( SY + iN 2 αe−Y /N ) (3.6)
λs
where the constant α, by a shift in Y → Y + Y0 , can be identified with a shift in the bare
coupling constant of S, i.e.,
α = e−Y0 /N (3.7)

(The choice of α is also related to how we regularize the one-loop divergence which corrects
R
the action with a term a d2 θ S).
This effective superpotential has the same structure as that encountered in the proof of
mirror symmetry in 2 dimensions [34]. This same superpotential structure was encountered
in [10] in the context of N = 1 domain walls in 4d, which we will also need in this paper,
and we will discuss further below. Notice that here since Y is a dynamical field, we can
integrate it out by setting
1
∂Y W = 0 → S = iN αe−Y /N
λs
which leads to
S
Y = log( )−N
iN αλs
plugging it back to the superpotential gives the effective superpotential for S:
1 S
Wef f (S) = [Slog( )−N + N S]
λs iN αλS
This is the familiar effective superpotential expected for the gaugino bilinear S in the
standard N = 1 supersymmetric gauge theory. Indeed setting ∂S W = 0 leads to
S
[ ]N = 1 → S = iN αλs e2πil/N = iN λs e(−Y0 +2πil)/N (3.8)
iN αλs

14
Note that we see the N vacua of SU (N ) Yang-Mills, in the standard way.
Let us compare the vev we found for S = λs T rW 2 with the gaugino condensate for
standard N = 1 Yang-Mills, which is of the form

[ −1
]+ 2πil
N g2
T rW 2 = iN Λ3 e
N
YM

In comparison with what we have above, note that this is in perfect agreement with (3.8),
where Λ corresponds to the string scale and 1/gY2 M → Y0 .
Note that the effective superpotential we have found for S, for small S, also follows
from either the open topological string amplitudes in the limit S → 0, or the dual closed
topological string amplitude in the limit S = t → 0 which is given by
1
F0 (t) → − t2 logt + at2 + bt + c
2
and so
1 1
W (S) = [N ∂S F0 (S) + αS + β] = (SlogS −N + N · const.S + N · const.)
λs λs
in perfect agreement with expectations based on the gauge theory analysis as well as
with the contribution of the Euclidean D2 brane instantons in the string context. This
comparison with gauge theory and recalling that t ↔ λs T rW 2 , also fixes the value c in
(3.5) to be c ∼ N λs . Note that the choice of α, the linear terms in S, on the gravity side
is controlled by the 4-form fluxes dual to the P1 , as discussed before.
Having discussed the geometry of the vacua of N = 1 theory, we now turn to another
important feature of N = 1 theories, namely the domain walls interpolating between
various vacua.

3.2. Domain Walls

N = 1 Yang-Mills theory admits BPS domain walls interpolating between various

vacua. As noted in [33] at large N they behave as D-branes for QCD string. In particular
their tension is of the order of N . Since in the present context the QCD string is realized by
the fundamental string, ordinary D-branes of string theory should play the role of domain
walls. This is indeed the case: On the gravity side we have a blown up P1 . If we consider
D4 branes wrapped over P1 they correspond to domain walls. Their tension goes as

1 |t|
T ∼ |t| = N
λs N λs

15
As discussed before |N λs | ∼ |t| so we obtain the expected behavior. For the QCD domain
wall the phase of the S field should change as we go from one vacuum to another. In
particular it should shift by exp(2πi/N ) for domain walls interpolating adjacent vacua.
Let us see how this is realized in the gravity setup. Since we have identified the domain
wall with D4 brane wrapped over S 2 we should note that the value of the G flux shifts
as we cross the domain wall. Consider in particular the imaginary part of the Y field
introduced earlier, which was identified with

ImY = CS 3

i.e. the vev of the C field along the S 3 . We now discuss how this changes from the left-side
of the domain wall to the right-side. Since the G flux should be equal to one for the D4
brane, it implies that ImY should shift by 2π, i.e.

Y → Y + 2πi

as we go across the domain wall. In fact we can find the geometry of the BPS domain
walls by the usual technique of the LG theory in 2d with N = 2 susy [35]. In fact for the
case at hand similar BPS domain walls were considered in [36]. These domain walls also
featured in the discussion of N = 1 generation of superpotential in [10]. Note that since
we have
S = λs N exp(−(Y + Y0 )/N )

this implies that the phase of S changes by exp(−2πi/N ) as expected. Of course this is
suppressed at large N , in agreement with the fact that classically the wrapped D-brane
does not change the value of t.
It is also easy to see from the form of the action (3.6) that the BPS tension, which is
given by ∆W is given by
1
∆W = S∆Y
λs
Since S is identified with t, this corresponds to

2πi
∆W = t
λs

as expected for the tension of the BPS wrapped D4 brane.

16
3.3. Subleading Corrections in the 1/N Expansion
So far we have concentrated on the interpretation of the leading corrections in large
N . In the context of topological strings also the subleading terms to all orders in 1/N were
found to agree between the Chern-Simons gauge theory and the closed topological string
expansion. What is the interpretation of these higher terms for the gauge theory system?
In the limit of small t the topological string amplitudes is given by
X
F (t) = Fg λs2g−2 t2−2g
g
B2g
where Fg = 2g(2g−2) and B2g are the Bernoulli numbers (Fg turns out to be equal to the
Euler characteristic of the moduli space of genus g Riemann surfaces). In this limit the
topological string partition function coincides with that of non-critical bosonic strings on
a circle with self-dual radius (this connection is well understood and will be reviewed in
section 5). The N = 2 amplitude that this computes is given by
Z
d4 θ W 2g Fg t2−2g = gR2 F 2g−2 Fg t2−2g + ... (3.9)

This correction has been physically understood by considering turning on constant

graviphoton field strength in the Minkowski space and computing the effect of wrapped
D2 branes on P1 to the R2 term [37]. In the present context the wrapped D2 branes cor-
respond to the baryon vertex, as in the usual AdS/CFT correspondence [38]. The Baryon
fields are charged under the graviphoton field with charge proportional to their BPS mass
t. Thus turning on graviphoton F effectively turns on a background field strength for Fv ,
i.e. the U (1) ⊂ U (N ) living on the D6 branes, which can be identified with a global U (1)
symmetry (the ‘Baryon number’ symmetry). Let us try to see how this can come about
from the gauge theory side.
On the worldvolume of the D6 branes we have terms of the form
Z Z
[G4 + F ∧ Fv ]( [CS(ω) − CS(A)])
R4 S3

where ω denotes the spin connection on S 3 and A is the internal gauge field on S 3 . This
term arises (by integrating by parts) from the usual inducement of brane charge by gravi-
tational and gauge curvature on the brane (see [39] and references therein). Thus shifting
F effectively shifts Fv 6 .
6
Note that if we change the G4 flux this is equivalent to turning on an internal Chern-Simons
action for the supersymmetric system on the brane. It should be possible to derive directly the
relation between generation of superpotentials on the brane and the Chern-Simons theory on S 3
from this fact.

17
 There is another term that is also generated from (3.9) when we recall that t has some
auxiliary field turned on. In particular this gives rise to the term
Z Z
N d θ W ∂t Fg (t) = N d2 θ F 2g ∂t Fg (t) + ...
2 2g

Recalling that in the gauge theory setup t is replaced by S, the gaugino bilinear superfield,
the above term corresponds to the superpotential term
Z Z
B2g 2g 1−2g
λs W = N d θ F ∂S Fg (S) = −N d2 θ
2 2g
F S
2g

So turning on the (self-dual) graviphoton field strength in 4 dimensions deforms the su-
perpotential. What is the gauge theory interpretation of this? As noted above turning
on F has the effect of turning on the field strength Fv in the U (1) ∈ U (N ), which is also
equivalent to turning on B field in spacetime. Thus this seems to be related to considering
the non-commutative version of the above gauge system [40]. In particular considering
a self-dual non-commutativity in spacetime presumably generates a superpotential, as is
predicted from the above formula. Note that this is consistent with the fact that in the
UV where S is smaller this modification of the superpotential is a more pronounced effect,
and it disappears in the IR where S is larger. It would be interesting to derive this result
directly in the context of the non-commutative N = 1 Yang-Mills theory. Moreover the
dependence of the genus g partition function on the non-commutativity parameter is iden-
tical7 to that obtained in [41]. Namely, in the large N expansion, there is no modification
at the level of planar diagrams, i.e. at g = 0. Moreover at genus g the amplitudes are
expected (when we have a self-dual non-commutativity) to scale (in the leading order) as
1
θ4g ∼ B 4g which in our case translates to an F 4g dependence. This is in agreement with
the fact that |∂S W |2 indeed scales as F 4g .

3.4. More General Values of S

So far, in the context of gauge theory discussion we mainly considered the limit where
< S > is small compared to the string scale. However the duality we are proposing holds
for arbitrary < S >. If < S > is not small, on the gauge theory side we get modification
to the form we have written above, which is computed by the Chern-Simons theory on
S 3 . What kind of gauge theory does this correspond to? The gravity side provides a
7
We thank R. Gopakumar for pointing this out to us.

18
hint: If we consider wrapped D4 brane domain walls, we have infinitely many species
of domain walls. The reason for this is that we can consider the bound state of n D2
branes with the D4 brane manifested through turning on n units of U (1) flux through
the S 2 part of the worldvolume of the D4 brane. This can also be viewed as the effect of
changing the B-field on the P1 by 2πin. The effect of such domain walls is thus shifting
t = S → Sexp(2πi/N )+2πin. In other words we have the vev’s of S, not only taking values
around a circle about the origin, but also circles about 2πin for any integer n. Moreover
the BPS tension for such domain walls is given by

1
∆W = (S + 2πin)
λs

The geometry of these domain walls can be recovered from an enlarged field content [10]:
We can introduce one variable Yn for each n, capturing the corresponding domain wall by
its shift in the argument, and consider the superpotential
Z X
W = d2 θ [(S + 2πin)Yn + iN 2 αe−Yn /N ] (3.10)
n

the domain wall with 1 D4 brane wrapped over P1 bound to n D2 branes will now corre-
spond to shifting Yn → Yn + 2πi. Integrating the Yn ’s out will give

1 X
W = (S + 2πin)log(S + 2πin)−N + a(S + 2πin) + b
λs n

which is indeed equal to Z

1 ∂F0 (S)
W = d2 θ N .
λs ∂S
The variables Yn were introduced to incorporate the kinks, but their appearance on the
original gauge theory side, except for Y0 seems mysterious. It would be interesting to see
if one can find a direct interpretation of all the Yn ’s. We expect that to be related to the
possibility of doing large SU (N ) gauge transformations on the S 3 part of the worldvolume
of D6 brane.
The higher genus corrections in the case of large S are also similar to the modification
at the genus 0 case. In particular we get an infinite sum with S replaced by S + 2πin. This
in particular is related to the fact that we can have a new baryon vertex for each wrapped
D2 brane with D0 brane turned on [42].

19
3.5. Adding Matter

In the context of geometric engineering of N = 1 supersymmetric gauge theories

realized as D6 branes wrapped around S 3 cycles of CY manifolds [32] matter can be
realized as extra D6 branes wrapped around other S 3 ’s intersecting the gauge theory S 3
along a circle (where the vev of the Wilson line around the circle on the probe brane plays
the role of mass for the matter). How does our duality extend to this case? In fact in the
topological string the duality does extend to this case [10]. In particular in computation
of Wilson loop observable for the Chern-Simons theory one adds extra topological branes
intersecting the original S 3 along a knot, and it was shown that the closed topological
string amplitudes agrees with the expected result for knot invariants for Chern-Simons
theory. More checks have been made in [11] for a large number of distinct knots. In the
context of embedding the topological string dualities in the superstring what this means
is that the dual gravitational system will not only have a blown up S 2 but will also have
additional D6 branes (which for algebraic knots will intersect the S 2 along a circle). The
fact that the topological computations agree on both sides translates to the statement
that the superpotential computations on both sides agree and is further evidence for this
duality in the superstring context. Note that for each knot we obtain a different “matter”
system for this generalized gauge theory, which in the limit of large Y give rise to the same
low energy physics, but are distinct theories in the context of generalized gauge theories
we have been considering. The gauge theoretic interpretation of these results is currently
under investigation [43].

4. The Mirror Type IIB Description

As is well known, type IIA on a CY is equivalent to type IIB on a mirror CY. This
implies that everything we have said above in the context of type IIA has a type IIB
counterpart.
For example instead of D6 branes of type IIA wrapped around S 3 we consider D5
branes of type IIB wrapped around S 2 . Also turning on even-form fluxes in type IIA is
mirror to turning on 3-form HRR and HNS flux in the type IIB side and the superpotential
that gets generated in this context is given by
Z
1
W = Ω ∧ [HRR + τ HNS ]
λs

20
where τ is the complex coupling constant of type IIB, and Ω is the holomorphic 3-form of
the CY. The above integral can be done and yields the formula in terms of the prepotential
of the corresponding N = 2 theory, as discussed in the Type IIA case. Note however, that
the type IIB system is simpler in that by mirror symmetry the worldsheet instantons
that were relevant in the context of type IIA theory in computing the prepotential, are
absent for the type IIB case, and classical geometry already captures these corrections.
In particular the B-topological theory (known as the Kodaira-Spencer theory of gravity)
simply involves aspects of complex geometry of Calabi-Yau.
So as far as writing a classical gravitational background, the type IIB description
would be more useful because the worldsheet instanton effects are absent. However, as far
as the conformal theory on the string worldsheet, i.e. the large N expansion description
of the gauge system, the type IIA and type IIB theories are of course identical.
In the above context we would need to know the mirror of local CY: O(−1)+O(−1) →
P1 . The mirror of this is known and it is essentially the conifold with one subtlety [36]: The
conifold has only one compact 3-cycle, whereas O(−1) + O(−1) → P1 has two compact
even cycles, namely 0 and 2 cycle. As was noted in [36] in the limit where the Kahler
class of P1 approaches zero, i.e., t → 0, the mirror becomes effectively the conifold (the
actual mirror differs from the conifold by having some variables being C∗ variables rather
than C variables)8 . Similar observations were made in [44]. Even though in principle we
can consider the full mirror geometry, since the complex geometry of the conifold is more
familiar and better studied we restrict our attention to this case9 . This will correspond to
a particular limit of our Type IIA theory, where we consider only the small < S > region.
Recall that this was the regime where the theory retained only the leading dimension
operators in the action and led to a theory which was similar to the standard N = 1
supersymmetric gauge theory.
We will be brief for this case, as most of the discussion can be literally borrowed from
our discussion in the previous section. We start with N D5 branes wrapped over the P1
8
The actual mirror is given by x1 + x2 + x1 x2 e−t + 1 − uv = 0 where x1 , x2 are C∗ variables
and u, v are C variables.
9
In this limit the internal topological theory corresponds to a G/G model on S 2 (coming from
the holomorpohic Chern-Simons theory on S 2 [14][45]) which should also be equivalent, by mirror
symmetry, to the large N fixed k limit of the Chern-Simons theory. This topological theory should
also be equivalent to the Penner Matrix model. It would be interesting to verify these equivalences
among these topological gauge theories more directly.

21
in the O(−1) + O(−1) → P1 geometry. The large N limit of this, in the limit of shrinking
P1 corresponds to blowing up an S 3 with N units of HRR flux through the S 3 . Let us
write the conifold geometry as
z1 z2 − z3 z4 = µ

Then the genus 0 prepotential is given as

−1 2
F0 (µ) = µ logµ + P2 (µ)
2

where P2 (µ) is an undetermined polynomial of degree 2 in µ. Now we consider turning on

fluxes: The mirror of turning on N S 4-form flux corresponds here to turning on HNS and
in the cycle dual to S 3 . Thus as far as the superpotential is concerned we have

1 1
W = [N ∂µ F0 (µ) + M µ] = [−N µlogµ + aµ + b]
λs λs

where M = M1 + τ M2 , and the discussion reduces to the small S limit of the discussion
in the previous section.
While this paper was being prepared two papers [18][19] appeared which are related
to this type IIB construction. In particular (among other things) they consider the grav-
itational background corresponding to D5 branes wrapped on 2-cycles of CY and their
results are consistent with the superpotential analysis here.

5. c=1 Non-critical Bosonic String and N=1 Superpotentials at Large N

As discussed above the type IIA or type IIB near a conifold background with some
fluxes turned on can be interpreted as large N limit of certain N = 1 supersymmetric
gauge theories. In particular the string expansion is equivalent to the large N expansion
of a gauge theory. Moreover certain superpotential corrections of the gauge theory can be
viewed as computations of the corresponding topological strings in the CY background.
These are readily computed and thus carry a large amount of information to all orders in
1/N , for the gauge theories in question. In particular here we will explain how the Type IIA
superstring near the small blow up of conifold, or equivalently the Type IIB superstrings
in the conifold geometry relate the non-critical bosonic string amplitudes with that of
the superpotential computations at the large N limit of the corresponding N = 1 gauge
systems.

22
 It was shown in [46] that the conformal theory near the conifold is given by the
same system found in [47] in connection with non-critical bosonic strings on a circle of
self-dual radius. This conformal theory is that of a supersymmetric Kazama-Suzuki coset
construction
SL(2)/U (1)

at level k = 3, and the relation with non-critical bosonic strings is that the topological
twisting of this system is equivalent to considering bosonic string propagating on a circle
of self-dual radius with the fermions of the coset model playing the role of the ghosts
in the bosonic string. This relation between bosonic string on a self-dual circle and the
superconformal theory of a conifold is in agreement with the fact [48] that the ground ring
of the bosonic string for this background is isomorphic to the holomorphic function on the
conifold (which is generated by z1 , z2 , z3 , z4 subject to the relation z1 z2 − z3 z4 = µ) where
the cosmological constant of the bosonic string is mapped to the deformation parameter
of the conifold. Moreover the observables of the c = 1 theory are mapped to deformations
of the conifold geometry:
X
ǫn (z1 , z2 , z3 , z4 ) + z1 z2 − z3 z4 = µ
n

where ǫn is a polynomial of degree n is zi . These deformation parameters get mapped to

states of the bosonic string which are indexed by a representation of SU (2)L × SU (2)R of
this system, viewing zi as entries of a 2 × 2 matrix M with the conifold being defined as
detM = µ and where the SU (2)L and SU (2)R are realized by left and right multiplication
of M with SU (2). In particular the degree n polynomial ǫn decomposes into representation
of spin (jL , jR ) = (n/2, n/2) with |mL , mR | ≤ n/2. Let us denote the totality of these
parameters by µi (except for µ). The bosonic string amplitudes compute topological B-
twisting of the deformed conifold10 . For various aspects of c = 1 non-critical bosonic string
see [50]. As already discussed the genus g partition function will be a function Fg (µ, µi )
of these parameters deforming the conifold background. Recall that in the gauge theory
context µ is identified with S and we will thus denote Fg (S, µi ). The topological string
computes, at genus g, the term in the effective action given by
Z Z
N
d θ[W ] ∂S Fg (S, µi ) = d2 θF 2g ∂S Fg (S, µi ) + ...
2 2 g
λs
10
Aspects of this relation has recently been verified and certain results of bosonic strings have
been recovered directly using the Calabi-Yau picture and the Kodaira-Spencer theory [49].

23
What is the interpretation of this for the gauge theory? As in the usual AdS/CFT corre-
spondences, we would expect that the µi will be related to operators on the gauge theory
side, deforming the gauge theory action by terms µi Oi . In fact, in the context of 3-brane
probes of the conifold [51] aspects of such deformations for the gravity side have been
studied in [52] and a similar analysis should be extendable to the case at hand. Thus
the topological strings compute the response of the system upon such deformations. In
particular at genus 0, turning on the µi modifies the superpotential for the gaugino super-
field. Also turning on F will give rise to 1/N correction to the superpotential. It would
be extremely interesting to understand the source of these corrections on the gauge theory
side.
It would also be interesting to find the conformal theory associated with the RR and
NS fluxes turned on in the conifold geometry. This is very interesting in view of the
fact that before turning fluxes on we have an exactly solvable conformal theory given by
SL(2)/U (1) KS model. It would be very interesting to find the deformation of this theory.
It is likely to involve ingredients similar to the ones encountered in [53][54].

6. Wrapped D-branes and Compact CY

Consider type IIA superstrings compactified on a Calabi-Yau threefold. In the above,

we considered a situation where we take a large number of D6 branes wrapped over an S 3
in the CY, and taking the analog of the near horizon geometry, to decouple the gravity,
and then proposing a dual gravity description for the gauge system.
If we wish to repeat what we did in the previous sections, by considering D6 branes
wrapped on some S 3 , which is part of a compact Calabi-Yau, and filling the rest of the
spacetime, we immediately run into a problem. We cannot wrap a D6 brane over a 3-cycle
as there would be nowhere for the flux to go for compact internal CY. However suppose
we consider a CY manifold with some number of S 3 ’s and we wrap D6-branes and anti-D6
branes over them, in such a way that the net D6 brane charge is zero. This is of course a
non-supersymmetric situation.11 We expect that the branes will eventually annihilate each
other leaving us with an N = 2 background. If the S 3 ’s are rigid then this annihilation
process takes some time, because there is a potential barrier for the wrapped D6 branes

11
We can also consider the type IIB mirror of this where we consider D5/anti-D5 branes
wrapped around vanishing S 2 ’s of the CY.

24
to move in the CY (i.e. there is a potential for the scalar corresponding to moving them
in the normal direction in the CY).
We now wish to apply the considerations of this paper and propose a large N dual
in this context. For the considerations of the gauge theory to be applicable we have to
consider the limit where S 3 has shrunk to zero size. This is the analog of Y0 = 0 in
the formula in section 3). What we will find is that taking the large N limit induces a
transition to a topologically distinct CY, with some fluxes turned on. Moreover the fluxes
can disappear as in [17] leaving us with an N = 2 supersymmetric vacuum. Thus the effect
of the brane annihilation at large N has been to shift the background.
Consider a Calabi-Yau with R vanishing S 3 cycles [Ci ] which span a K < R dimen-
sional subspace of H3 . In other words assume
X
Qji [Ci ] = 0 f or j = 1, ..., R − K

for some integral matrix Q. Let us consider Ni D6 branes partially wrapped around
Ci , where we allow some Ni to be negative, in which case we mean the number of the
corresponding anti-branes. The condition that the net D6 brane flux is zero implies that

Ni = lj Qji

for R − K integers lj . Now, let us consider the limit where the S 3 ’s are vanishingly small.
In this limit, applying the discussion of the near horizon geometry, we are naturally led
to consider the S 2 blown up geometry where the blow up parameter for the i-th sphere is
given by
ti = iλs Ni = iλs lj Qji

Notice that not all the ti are independent. In particular there are only R − K independent
parameters lj which determine them. This is exactly as it should be for the local geometry
to have a blowup12 . In other words the blow up geometry is a CY with K less dimension
of h2,1 but R − K more dimensions of h1,1 . Moreover the condition on the various Kahler
classes of the P1 ’s is exactly the same as that found above for the ti . This gives further
support for the conjecture that the large N limit of the wrapped brane-anti-brane geometry,
12
In fact this shift in the hodge numbers can be understood from the viewpoint of inverse
process of Higgsing of U (1)R−K by R charged fields [55] where the charged fields can be viewed
as wrapped D2 branes in the blown up geometry.

25
when we have no net branes is equivalent to a blown up CY with the Kahler parameters for
the blown up spheres given as above. However, here we will also have RR fluxes through
the S 2 ’s. In this case the supersymmetry is completely broken [25][26] by the RR fluxes.
The fluxes can disappear as recently studied in [17].
It would be interesting to check this generalized conjecture in the topological string
setup: Namely this suggests that the topological open string amplitudes in the context
of compact CY manifolds when there is no net topological D-brane, and when the D-
branes are wrapped over spheres is easily computable by a related closed string theory
computation on the blown up CY with different Hodge numbers.

7. Generalizations

There are many natural generalizations of this work. In particular it is natural to

consider transitions among topologically distinct manifolds, going through vanishing cycles,
and find a large N brane system/gravity duals, where the large N gauge system will lie on
one side of the transition and the dual gravitational system will lie on the other side (this
was in fact the philosophy advocated in [9])13 . In other words the large N brane systems
can be viewed as inducing transitions in the background geometry. For examples there
are transitions in the CY which involve the vanishing of certain 3-cycles and blowing up
4 manifolds, such as Del-Pezzo manifolds. In this context it is natural to conjecture the
existence of a duality involving a large N limit of wrapped D6 branes about the 3-cycles
in the context of type IIA with the 4 manifold resolution of the singularity on the gravity
side, with certain fluxes turned on14 . Or in the context of M-theory on 4-folds it is natural
13
If we consider a Morse function f on a manifold the critical points of it encode certain
topological aspects of the manifold. Near a critical point with p positive and q negative eigenvalues
for ∂i ∂j f , for f 6= fcritical the manifold near the critical point has the geometry of a filled
S p−1 × S q−1 . For f > fcritical the S q−1 is filled and for f < fcritical the S p−1 is filled. This is the
general kind of transition expected for large number of branes replaced by fluxes. If we consider
two manifolds in the same cobordism class, and consider a Morse function on the interpolating
manifold the above picture suggests that branes can induce the transition. So if the cobordism
classes are trivial and we have suitable branes we can interpolate between any two manifolds in
this way.
14
In fact there is already evidence for some such cases based on quotienting the Chern-Simons
duality on S 3 by finite groups on both sides [56]. For example Chern-Simons on S 3 /Z2 should be
equivalent to P1 × P1 blow up inside a Calabi-Yau. Some aspects of this predictions have already
been checked.

26
to look for transitions involving shrinking S 3 ’s and growing 4-cycles, where we consider a
large number of M 5 branes wrapped over the S 3 ’s and filling the 3-dimensional spacetime,
which should be dual to the geometry involving the blowup of the 4-cycle with some G
flux turned on (in fact in this context the gravity solutions are already worked out in [31]).
It is also natural to extend our results for the case of SU (N ) systems to SO and Sp
groups by including orientifolds. In fact it has been shown in [57] that the large N duality
of Chern-Simons theory for SU (N ) groups extend to the SO and Sp case as well.
Finally, the idea that studying BPS/anti-BPS systems are important for a more fun-
damental understanding of basic degrees of freedom for string theory as advocated by Sen
is in line with the example we have found: We can describe one string background in terms
of the ground state of a different one in the presence of D-brane/anti-D-brane systems. In
a sense, this idea, combined with the idea that various transitions among manifolds can be
induced by large N limit of brane systems, suggests that if we start with any background in
string theory, and consider complicated enough configurations of branes and anti-branes,
we can effectively be discussing arbitrary backgrounds of string theory.
Acknowledgements:
I have greatly benefited from discussions with N. Berkovits. I would also like to
thank M. Aganagic, M.F. Atiyah, R. Gopakumar, K. Intriligator, S. Katz, A. Klemm, J.
Maldacena, S. Minwalla, N. Seiberg, S. Sinha and A. Strominger for valuable discussions.
This research is partially supported by NSF grant PHY-98-02709.

27
 References

[1] G. ‘t Hooft, “A Planar Diagram Theory for Strong Interactions,” Nucl. Phys. 72
(1974) 461.
[2] M. Kontsevich, “Intersection Theory on the Moduli Space of Curves and the Matrix
Airy Function,” Comm. Math. Phys. 147 (1992) 1.
[3] E. Witten, “On the Structure of the Topological Phase of Two-Dimensional Gravity,”
Nucl. Phys. B340 (1990) 281.
[4] J. Distler and C. Vafa, “A Critical Matrix Model at c = 1,” Mod. Phys. Lett. A6
(1991) 259.
[5] R.C. Penner, “Perturbative Series and the Moduli Space of Riemann Surfaces,” UCSD
preprint (1986).
[6] J. Maldacena, “The Large N limit of Superconformal Field Theories and Supergrav-
ity,” Adv. Theor. Math. Phys. 2 (1998) 231.
[7] S.S. Gubser, I.R. Klebanov and A.M. Polyakov, “Gauge Theory Correlators From
Non-critical String Theory,” Phys. Lett. B428 (1998) 105.
[8] E. Witten, “Anti-de Sitter Space and Holography,” Adv. Theor. Math. Phys. 2 (1998)
253.
[9] R. Gopakumar and C. Vafa, “On the Gauge Theory/Geometry Correspondence,” hep-
th/9811131.
[10] H. Ooguri and C. Vafa, “Kont Invariants and Topological Strings,” Nucl. Phys. B577
(2000) 419.
[11] J.M.F. Labastida and M. Marino,“Polynomial Invariants for Torus Knots and Topo-
logical Strings,” hep-th/0004196.
[12] V. Periwal, “Topological Closed-String Interpretation of Chern-Simons Theory,” Phys.
Rev. Lett. 71 (1993) 1295.
[13] M. Bershadsky, S. Cecotti, H. Ooguri and C. Vafa, “Kodaira-Spencer Theory of Grav-
ity and Exact Results for Quantum String Amplitudes,” Comm. Math. Phys. 165
(1994) 311.
[14] E. Witten, “Chern-Simons Gauge Theory as a String Theory,” hep-th/9207094.
[15] T.R. Taylor and C. Vafa,“RR Flux on Calabi-Yau and Partial Supersymmetry Break-
ing,” Phys. Lett. B474 (2000) 130.
[16] P. Mayr, “On Supersymmetry Breaking in String Theory and its Realization in Brane
World,” hep-th/0003198.
[17] R. Bousso and J. Polchinski, “Quantization of Four-Form Fluxes and Dynamical Neu-
tralization of the Cosmological Constant,” hep-th/0004134.
[18] I. Klebanov and M.J. Strassler,“Supergravity and a Confining Gauge Theory: Duality
Cascades and χSB-Resolution of Naked Singularities,” hep-th/0007191.

28
[19] J. Maldacena and C. Nunez, “Towards the Large N Limit of Pure N = 1 Super
Yang-Mills,” hep-th/0008001.
[20] T. Banks, M. Dine and L. Motl, “On Anthropic Solutions of the Cosmological Constant
Problem,” hep-th/0007206.
[21] I. Antoniadis, E. Gava, K.S. Narain and T.R. Taylor, “Topological Amplitudes in
String Theory,” Nucl. Phys. B413 (1994) 162.
[22] N. Berkovits and C. Vafa, “N=4 Topological Strings,” Nucl. Phys. B 433 (1995) 123.
[23] N. Berkovits, “Covariant Quantization of the Green-Schwarz Superstring in a Calabi-
Yau Background”, Nucl. Phys. B431 (1994) 258.
[24] N. Berkovits and W. Siegel, “Superspace Effective Actions for 4D Compactifications
of Heterotic and Type II Superstrings”, Nucl. Phys. B462 (1996) 213.
[25] J. Polchinski and A. Strominger, “New Vacua for Type II String Theory,” Phys. Lett.
B388 (1996) 736.
[26] J. Michelson, “Compactifications of Type IIB Strings to Four Dimensions with Non-
trivial Classical Potential,” Nucl. Phys. B495 (1997) 127.
[27] S. Gukov, C. Vafa and E. Witten, “CFT’s from Calabi-Yau Four-folds,” hep-
th/9906070.
[28] S. Gukov,“Solitons, Superpotentials and Calibrations,” hep-th/9911011.
[29] I. Brunner, M.R. Douglas, A. Lawrence and C. Romelsberger, “D-branes on the Quin-
tic,” hep-th/9906200.
[30] S. Kachru, S. Katz, A. Lawrence and J. McGreevy, “Open String Instantons and
Superpotentials,” Phys. Rev. D62 (2000) 0260001.
[31] K. Becker and M. Becker, “M-theory on Eight-Manifolds,” Nucl. Phys. B477 (1996)
155.
[32] H. Ooguri and C. Vafa, “Geometry of N=1 Dualities in Four Dimensions,” Nucl. Phys.
B500 (1997) 62.
[33] E. Witten, “Branes and the Dynamics of QCD,” Nucl. Phys. B507 (1997) 658.
[34] K. Hori and C. Vafa, “Mirror Symmetry,” hep-th/0002222.
[35] S. Cecotti and C. Vafa, “On Classification of N = 2 Supersymmetric Theories,”
Comm. Math. Phys. 158 (1993) 569.
[36] K. Hori, A. Iqbal and C. Vafa, “D-branes and Mirror Symmetry,” hep-th/0005247.
[37] I. Antoniadis, E. Gava, K.S. Narain and T.R. Taylor, “N=2 Type II-Heterotic Duality
and Higher Derivative F-terms,” Nucl. Phys. B455 (1995) 109.
[38] E. Witten, “Baryons and Branes in Anti de Sitter Space,” JHEP 9807 (1998) 006.
[39] D. Diaconescu, G. Moore and E. Witten, “A Derivation of K-Theory from M-theory,”
hep-th/0005091.
[40] N. Seiberg and E. Witten, “String Theory and Non-commutative Geometry,” hep-
th/9908142 .

29
[41] S. Minwalla, M. Van Raamsdonk and N. Seiberg, “Noncommutative Perturbative
Dynamics,” hep-th/0002186.
[42] R. Gopakumar and C. Vafa, “M-theory and Topological Strings I,II,” hep-th/9809187,
hep-th/9812127.
[43] C. Vafa, work in progress.
[44] M. Aganagic, A. Karch, D. Lust and A. Miemiec, “Mirror Symmetries for Brane
Configurations and Branes at Singularities,” Nucl. Phys. B569 (2000) 277-302.
[45] C. Vafa, “Extending Mirror Conjecture to Calabi-Yau with Bundles,” hep-th/9804131.
[46] D. Ghoshal and C. Vafa, “c=1 String as the Topological Theory of the Conifold,”
Nucl. Phys. B453 (1995) 121.
[47] S. Mukhi and C. Vafa, “Two Dimensional Black Hole as a Topological Coset Model
for Two-Dimensional String Theory,” Nucl. Phys. B407 (1993) 667.
[48] E. Witten, “Ground Ring of Two Dimensional String Theory,” Nucl. Phys. B373
(1992) 187 ;
E. Witten and B. Zwiebach, “Algebraic Structures and Differential Geometry in 2-d
String Theory,” Nucl. Phys. B377 (1992) 55.
[49] R. Dijkgraaf and C. Vafa, unpublished.
[50] P. Ginsparg and G. Moore, “Lectures on 2d Gravity and 2d String Theory,” TASI
summer school 1992, hep-th/9304011.
[51] I. Klebanov and E. Witten,“ Superconformal Field Theory on Threebranes at a Calabi-
Yau Singularity,” Nucl. Phys. B536 (1998) 199.
′
[52] D.P. Jatkar and S. Randjbar-Daemi, “Type IIB String Theory on AdS5 ×T nn ,” Phys.
Lett. B460 (1999) 281.
[53] N. Berkovits, C. Vafa and E. Witten, “Conformal Field Theory of AdS Background
with Ramond-Ramond Flux,” JHEP 9903 (1999) 018.
[54] N. Berkovits, M. Bershadsky, T. Hauer, S. Zhukov and B. Zwiebach, “Superstring
Theory on AdS2 × S 2 as a Coset Supermanifold,” Nucl. Phys. B567 (2000) 61.
[55] B.R. Greene, D.R. Morrison and A. Strominger, “Black Hole Condensation and the
Unification of String Vacua,” Nucl. Phys. B451 (1995) 109.
[56] R. Gopakumar, A. Klemm, S. Sinha and C. Vafa, unpublished.
[57] S. Sinha and C. Vafa, to appear.

30