Introduction to Gauge/Gravity Duality

Lectures at TASI, June 1-7, 2010

Joseph Polchinski1
Kavli Institute for Theoretical Physics
arXiv:1010.6134v1 [hep-th] 29 Oct 2010

University of California
Santa Barbara, CA 93106-4030

Abstract
These lectures are an introduction to gauge/gravity duality, presented at TASI
2010. The first three sections present the basics, focusing on AdS5 × S 5 . The last
section surveys a variety of ways to generate duals of reduced symmetry.

1
[email protected]
Contents
1 Generalities 2
1.1 The greatest equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 A hand-waving derivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 A braney derivation, for AdS5 × S 5 . . . . . . . . . . . . . . . . . . . . . . . 6
1.4 Statement of the duality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2 The two sides 11

2.1 The gravity side . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 More about anti-de Sitter space . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3 Scalar fields in AdS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.4 Conformal field theories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.5 The D = 4, N = 4 theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

3 The dictionary 18
3.1 Symmetries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.2 Matching of states . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.3 Correlators I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.4 Correlators II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

4 Breaking symmetries 27
4.1 The Coulomb branch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.2 Renormalization group flows . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.3 Multitrace RG flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
4.4 Orbifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.5 Non-spherical horizons. II . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.6 Nonconformal branes [57] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.7 D2-D6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.8 D2-D8 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.9 Fractional D3-D7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

1
1 Generalities
1.1 The greatest equation
A few years back, Physics World magazine had a reader poll to determine the Greatest
Equation Ever, and came up with a two-way tie between Maxwell’s equations
d∗F =j, dF = 0 , (1.1)
and Euler’s equation
eiπ + 1 = 0 . (1.2)
The remarkable appeal of Euler’s equation is that it contains in such a compact form the
five most important numbers, 0, 1, i, π, e, and the three basic operations, +, ×, ˆ. But my
own choice would have been Maldacena’s equation
AdS = CFT , (1.3)
because this contains all the central concepts of fundamental physics: Maxwell’s equations,
to start with, and their non-Abelian extension, plus the Dirac and Klein-Gordon equations,
quantum mechanics, quantum field theory and general relativity. Moreover, in addition to
these known principles of nature, it contains several more that theorists have found appealing:
supersymmetry, string theory, and extra dimensions, and it ties these all together in an
irreducible way. Also, while Euler’s equation is a bit of an oddity, the relation AdS = CFT
is just the tip of a large iceberg, it can be deformed into a much large set of gauge/gravity
dualities.
So I get to teach you about this, trying to focus on things that you will need in the
many upcoming lectures which will use the left-hand side of the equation, string theory and
gravity, to learn more about the the right-had side, quantum field theory. Of course it’s
also interesting to use the equation in the other direction, and maybe some of that will
sneak in. Roughly speaking, today’s lecture will be a conceptual overview, lecture 2 will give
some essential details about the two sides of the duality, lectures 3 and 4 will work out the
dictionary between the two sides focussing on the familiar AdS5 × S 5 example, and lecture
5 will discuss generalizations in many directions.
Let me start by noting a few other reviews. The early MAGOO review [1] contains a
detailed summary of the early literature, in which many of the basic ideas are worked out.
The 2001 TASI lectures by d’Hoker and Freedman [2] are thorough and detailed, particularly
with regard to the constraints from supersymmetry and the conformal algebra, and the
calculation of correlation functions. McGreevy’s course notes [3] are similar in approach to
my lectures.

2
1.2 A hand-waving derivation
I am going to first motivate the duality in a somewhat unconventional way, but I like it
because it connects the two sides, gauge theory and gravity, without going directly through
string theory (as do many of the applications), and it allows us to introduce many ideas
that will be important later on. So let me start with the question, is it possible to make the
spin-2 graviton as a bound state of two spin-1 gauge bosons? With the benefit of generous
hindsight, we are going to make this idea work. To start off, there is a powerful no-go
theorem that actually forbids it [4]. Theories without gravity have more observables than
theories with gravity (local operators, in particular, since there is no invariant local way
to specify the position in general relativity), and this leads to a contradiction. Specifically,
Weinberg and Witten show that if there is a massless spin-2 particle in the spectrum, then
the matrix element
hmassless spin 2, k|Tµν |massless spin 2, k 0 i (1.4)
of the energy momentum tensor (which exists as a local observable in the gauge theory) has
impossible properties.
Of course, to prove a no-go theorem one must make assumptions about the framework,
and it often proves to be the case that there is some assumption that is so natural that
one doesn’t even think about it, but which turns out to be the weak link. The Coleman-
Mandula theorem, classifying all possible symmetries of the S-matrix [5], is a classic example.
This paper played an important role in its time, ruling out a class of ideas in which spin
and flavor were unified in SU (6). However, it made the unstated assumption that the
symmetry generators had to be bosonic,1 which was sufficient for the immediate purposes
but missed the possibility of supersymmetry. The more powerful a no-go theorem, the deeper
its counterexamples.2
The reason for going through this is that the no-go theorem is indeed wrong, but to violate
it we have to recognize a deep property of quantum gravity, the holographic principle [6, 7].
The entropy of a black hole is proportional to its area in Planck units, and this is the largest
possible entropy for a system with given surface area. This suggests that quantum gravity
in any volume is naturally formulated in terms of degrees of freedom on its surface, one per
Planck area. Thus we see the hidden assumption, that the graviton bound state moves in
the same spacetime as its gauge boson constituents; rather, it should move in one additional
1
Rereading Ref. [5], this assumption seems to have made its entrance at the point where the symmetry
generators are diagonalized, which can’t be done for a nilpotent operator.
2
Perhaps this is what Bohr meant when he said “It is the hallmark of any deep truth that its negation is
also a deep truth.”

3
dimension. Of course, there might be other ways to violate the theorem that will turn up in
the future.
So how is the two-gauge boson state in four dimensions supposed to correspond to a
graviton in five dimensions? With the benefit of hindsight, there are several places in QCD
phenomenology where the size of a gluon dipole, the magnitude z of the separation, behaves
like a spacetime coordinate. In color transparency [8], and in the BFKL analysis of Regge
scattering [9], interactions are approximately local in z and the pair wavefunction satisfies
a five-dimensional wave equation. So when you look at the gluon pair you picture it as a
graviton four of whose coordinates are the center of mass of the pair, and the fifth is the
separation.
We need just two more ingredients to make this idea work, but first we will make an
excursion and discuss the shape of the five dimensional spacetime. Quantum field theory is
nicest when it applies over a wide range of scales, so it is natural to consider scale-invariant
theories first. If we rescale the system, the c.m. coordinates x and the separations z scale
together. The most general metric respecting this symmetry and the symmetries of the
four-dimensional spacetime is
L2 dz 2 + L02 ηµν dxµ dxν 2 µ
2 dz + ηµν dx dx
ν
ds2 = → L , (1.5)
z2 z2
where I have rescaled z in the last form so as to emphasize that there is only one scale R.
This is the metric of anti-de Sitter space, in Poincaré coordinates. If we replace the
z with t in the denominator we get de Sitter space, the approximate geometry of our own
accelerating spacetime. Certainly one of the major frontiers in gauge/gravity duality, though
not one that I will focus on, is to figure out how to interchange z and t in the dual field
theory: it appears to require great new concepts. Anti-de Sitter spacetime is not expanding
but warped: clocks that run at the same rate in inertial coordinates run at different rates in
terms of x0 , depending on where they are in z.
To finish off our ‘derivation’ of the duality, we need two more ingredients. The first is a
large number of fields. We want the AdS scale L to be large compared to the Planck length
LP , so that we can use Einstein gravity. This means that we can fit a large black hole into
the space, one with many Planckian pixels and so a large entropy, and so the field theory
had better have a correspondingly large number of degreees of freedom. As we will discuss
in more detail later, the number of fields is a power of L/LP , depending on the example.
The other ingredient we need is strong coupling, so that the gauge boson pair behaves like
a graviton and not like a pair of gauge bosons.3 I should say VERY strong coupling, much
3
Of course, it will then mix with states having more constituents, but one can still retain a bit of the
basic idea that the graviton spin comes from two ‘valence’ gauge bosons.

4
larger than one, to get a limit in which the gravitational description is quantitative.
Thus we have two necessary conditions for the duality, many fields and very strong
coupling. The second condition can actually be made a bit stricter, as we will see: we need
that most operators get parametrically large anomalous dimensions. In this form, these
necessary conditions are actually likely to be sufficient, as shown in part in Ref. [10]. Large
anomalous dimensions clearly require very strong coupling, but do not necessarily follow
from it as we will see in an example later.
These conditions of many fields and strong coupling will reappear at various points in
these lectures. Clearly they play a controlling role in the applications, as to whether we
have a quantitive description or merely a ‘spherical cuprate,’ a solvable model that captures
the qualitative physics that we wish to understand. For example, in the application to
heavy ion physics, the coupling seems to be of order one, midway between weak and strong.
The strong coupling picture is not quantitative in a precise way, but seems to do better
than perturbation theory on many qualitative properties. Naively the number of fields is
N 2 − 1 = 8 for the gluon states, which is a modestly large number, though it has been noted
that the parameter N/Nflavor is only unity, and there should be corrections of this order.
Fortunately the large-N approximation seems to be fairly robust even at small values.
In the condensed matter applications, again the relevant couplings are of order one at a
nontrivial fixed point, where a strong coupling expansion has a chance to capture things that
a weak coupling expansion cannot. There is no large N , but condensed matter theorists in
the past have not been above introducing a large-N vector index in order to get a tractable
system. The large-N vector is a mean-field approximation; this is true for the large-N matrix
limit as well, though in a more subtle and perhaps more flexible way [11]: expectations of
products of color singlets factorize, but there is a very large number of color singlet fields.
Notice that I have not yet mentioned supersymmetry. However, it tends to enter nec-
essarily, through the requirement of very strong coupling. Quantum field theories tend to
become unstable at strong coupling, through the production of pairs whose negative po-
tential energy exceeds their kinetic energy. In continuum theories this can happen at all
scales, and the theory ceases to exist. Supersymmetry protects against this: schematically
the Hamiltonian is the sum of the squares of Hermitian supercharges, H = i Q2i , so the
P

energy is bounded below. Thus, we can break the supersymmetry softly and still have a
duality, but if we break it at high energy we lose the theory. We may have to be satisfied
with metastability, but that is OK, we probably live with that in our own vacuum.
We have also not mentioned strings, we seem to have found a theory of quantum gravity
that uses only gauge theory as a starting point, but in a range of parameters for N and

5
coupling that is not so familiar. But of course what happens is that when we get gravity
in this way we get everything else as well, the strings, branes, extra dimensions and so on.
Again this was anticipated by ’t Hooft [12], who argued that the planar structure of large-N
gauge theory made it equivalent to a theory of strings.

1.3 A braney derivation, for AdS5 × S 5

Let me now outline, much more briefly, the original argument for the duality [13].4 This will
have the advantage that it gives specific constructions of gauge/gravity duals. Consider a
stack of N coincident D3-branes, in the IIB superstring. In string perturbation theory, each
additional world-sheet boundary on the branes brings in a factor of the string coupling g = eΦ
from the genus and a factor of N from the Chan-Paton trace. Thus perturbation theory is
good when gN  1 and breaks down when gN  1.5 Now, there are black branes [15] that
source the same Ramond-Ramond fluxes as D-branes [16]. The black 3-brane metric for N
units of flux (in string frame where the physical string tension is α0 ) is

ds2 = H −1/2 (r)ηµν dxµ dxν + H 1/2 (r)dxm dxm , µ, ν ∈ 0, . . . , 3 , m, n ∈ 4, . . . , 9 ,

L4
H = 1 + 4 , L4 = 4πgN α02 , r2 = xm xm . (1.6)
r
You will have to take my word for this solution here, but we will make a useful check later.
The harmonic function H is singular as r → 0, but the metric has a nice behavior in the
‘near-horizon’ small r limit,
r2 L2 2
ds2 → ηµν dxµ
dx ν
+ dr + L2 dΩ2S 5 . (1.7)
L2 r2
The first two terms are again the AdS5 metric, with the coordinate r = R2 /z, while the
last term is the metric of S 5 . The curvature radius is L for both factors. When gN  1
this length is large in string units, so the low energy supergravity theory is a good effec-
tive description, but it breaks down when gN  1. This is just complementary to the
perturbative description: the system looks like D-branes in one regime and black branes in
another. By varying g one can move adiabatically between these. This was the strategy
used to count the states of a black hole [17], though in a different brane system that has a
finite zero-temperature entropy, as we will discuss later.
This is not yet a duality; we have a single theory, the IIB string theory, describing
both limits, but in different approximations: perturbation theory around flat spacetime with
4
Note also Ref. [14], which reaches a similar picture by very different reasoning.
5
Of course the same is true for other Dp-branes, but the rest of the story is simplest for p = 3. We will
consider other cases later.

6
branes at small gN , and perturbation theory around the curved black brane spacetime (with
no branes and no open strings) at large gN . To find a duality let us look at the low energy
limit of both descriptions. In the small-gN description, this consists of the massless open
and closed strings. The massless open strings, ending on the D3-branes, are the usual U (N )
adjoint gauge field and collective coordinates, as well as their fermionic partners, while the
massless closed strings are the supergravity multiplet. The open strings remain interacting
at low energy, because the gauge coupling is dimensionless in 3+1 dimensions, but the closed
strings have irrelevant interactions and decouple. In the large-gN description, we again have
the massless closed strings away from the brane, but there are also states whose energy is
small because they are in the AdS5 × S 5 region at small r, where the warp factor g00 is going
to zero; these include not just the massless states, but any massive string state will have an
arbitrarily small energy as r → 0.
Let us ignore the massless closed strings away from the brane, which are decoupling in
both pictures. Concretely, consider the scaling r → r/ζ, xµ → ζxµ . The only effect on the
black brane metric (1.6) is that the 1 in the harmonic function becomes ζ −4 , and scales away
as ζ → ∞ leaving AdS5 × S 5 . In the D3-brane picture this is a symmetry of the low energy
gauge theory, while the closed string phase space volume scales to zero, and massive open
string effects are suppressed by powers of α0 /x2 ∼ ζ −2 . In Ref. [13] this limit is described
in terms of scaling α0 to zero while holding xµ fixed, with appropriate scalings elsewhere; I
have always found it harder to think this way, but it is the same in terms of dimensionless
ratios.
Now, if we make the innocuous-sounding assumption that taking the low energy limit
commutes with the adiabatic continuation in g we get a remarkable result. At weak coupling
we have the gauge theory, and at strong coupling we have all the string states in the AdS5 ×S 5
region. The assumption that the limits commute means that the strongly coupled gauge
theory is identical to the full string theory in AdS5 ×S 5 . This is now a duality, the statement
that two seemingly different theories describe the same system, but in different limits.
This ‘derivation’ of the duality seems rather slick, but it explained why very different
calculations in the two pictures were giving identical answers, and has been supported by
many further checks over time. How could the argument fail? The most obvious issue is that
the gauge theory could have a phase transition as we vary from weak to strong coupling, but
the supersymmetry strongly restricts this possibility. In particular, the property H = i Q2i
P

implies that any supersymmetric vacuum is a minimum of the energy, zero, so we can’t have
some other vacuum cross to lower energy at some intermediate coupling.6 We can imagine
6
This would become an issue however if we try to extend the argument to nonsupersymmetric configura-
tions, as we might want to in condensed matter systems. I will discuss an examples in Sec. 4.9.

7
more intricate ways that the duality might fail, and I will say a bit more about this later,
but it is very hard to think of an alternative that is consistent with all the evidence, and the
simplest conclusion is that the duality, remarkable though it be, is true.
There are many other weak-strong duality conjectures. Some relate one field theory to
another, and some relate one string theory to another or to M theory, but here we have a
duality between a quantum field theory and a string theory. From the point of view of trying
to construct a theory of quantum gravity this duality is particularly striking, because it allows
us to reduce it to the problem of constructing a QFT, solved by Wilson; of course there is
much more to say on this point, but it is not the focus of these lectures. What most of these
dualities have in common is that there is no explicit derivation, but rather there is the kind
of plausibility argument we have just made, combined with many tests giving circumstantial
evidence. Some QFT-QFT dualities can be derived explicitly, like bosonization and Ising
self-duality in 1+1 dimensions, and Abelian dualities in higher dimensions. These seem as
though they should be prototypes for all the other dualities, but somehow the direct steps
used to derive them are not enough, and some big new idea is needed.

1.4 Statement of the duality

The precise statement is that

D = 4, N = 4, SU (N ) Yang-Mills = IIB string theory on AdS5 × S 5 . (1.8)

We will give more detail about each side in the next lecture. Again, this is just one example
of a very large number of such dualities. We focus on it, as do most other introductions,
because it is both the simplest example, and the one whose field theory side is the most
familiar and relevant to many applications. The equality means a one-to-one mapping of
the spectra, at any given value of the energy and other quantum numbers. It also includes
equality of observables, namely the correlation functions of operators with an appropriate
dictionary between the two sides.
A minor aside that may occur to the reader: we have replaced the U (N ) of the brane
stack with SU (N ). The missing U (1) represents the supermultiplet containing the overall
collective motion of the stack. In the black brane picture, before we take the near-horizon
limit, this is a fluctuation of the spacetime geometry. The mode is peaked in the transition
region between flat spacetime, where the 1 in H(r) dominates, and the near-horizon region,
where the L4 /r4 dominates. In the AdS5 × S 5 limit, this moves away, and the collective
mode is trivial there.
Another aside: the holographic argument led to one extra dimension, but now we have

8
five more from the S 5 . Still, the Poincaré and radial dimensions will play a central role, and
for much of the discussion we can reduce on the S 5 . Note that the warping acts only on the
Poincaré dimensions, the S 5 having a constant radius.
2
On the gauge side we have the two parameters gYM and N . On the string side these map to
the string coupling g and again N , where the latter is now interpreted as the number of units
of five-form flux on the S 5 , which counts the D3-branes in the stack. The map between the
couplings is standard for D3-branes, g = gYM 2
/4π, with action normalized − 2g12 Tr Fµν F µν .
YM

The parameters on the gravity side are also usefully expressed in terms of ratios of length
scales. We have already noted that

L/Lstring ≡ L/α01/2 = (4πgN )1/4 = (gYM

2
N )1/4 = λ1/4 , (1.9)
2
where in the last line we have introduced the ’t Hooft parameter λ = gYM N . The other
relevant length is Planck’s. In general spacetime dimension D it is natural to work with
the reduced Planck length L̂P,D , such that the coefficient of the Ricci scalar in the action is
1/2L̂D−2
P,D ; in four dimensions L̂P,4 = (8π)
1/2
LP in terms of the usual Planck length. Then in
8 1 7 2 04
string theory L̂P,10 = 2 (2π) g α , and

L/L̂P,10 = 2−1/4 π −5/8 N 1/4 . (1.10)

A classical spacetime description requires the ratios (1.9) and (1.10) to be large, and so as
anticipated in the handwaving argument we need the coupling λ and the number of fields N
to be large.
The evidence for the duality takes many varied forms, some of which will be discussed as
we go along:

1. The symmetries on the two sides match.

2. The spectra of supersymmetric states match. This includes, for example, all modes of
the graviton in AdS5 × S 5 .

3. Amplitudes which are protected by supersymmetry and so can be compared between

the two sides are equal.

4. When we perturb the duality in ways that break some of the supersymmetry and/or
conformal symmetry, the geometry realizes the behaviors expected in the field theory,
such as confinement.

5. There are higher symmetries on both sides, which allow some quantities to be calculated
for all g, with apparent consistency.

9
 6. Matching long string states can be identified on both sides.

7. The predictions of the duality for strongly coupled gauge theories can be compared
with numerical calculations in those theories, using both light-cone diagonalization
and lattice simulation (these are computationally challenging, but progressing).

8. The predictions agree with experiment!

The next-to-last item, and especially the last, are currently at a crude level of accuracy.
Probably I have missed some good tests. Essentially, every time one applies the duality one
is testing it, because there is always the possibility of some absurd consequence.
It is sometimes asserted that the evidence supports only a weak form of the duality, but it
is not clear what a sensible weak form would be. Suggestions include only the supergravity
states, but the number of states in the gauge theory is much larger than this, and some
can be identified clearly with string states. Another weak form would be to hold only in the
extreme large-N limit. However, this is not consistent (unitary) by itself on either side of the
duality, and unitarity largely determines the 1/N expansion on the QFT side and the gravi-
tational loop expansion on the string side; it follows from (1.10) that these are expansions in
equivalent parameters. Yet another weak form would be to hold only perturbatively in 1/N
and not exactly. However, the most important nonperturbative phenomena is present on
both sides: the integer property of N . In the 1/N expansion 1/N is a continuous parameter,
but in reality it is discrete with an accumulation point at zero. The string side knows about
this because the 5-form flux satisfies a Dirac quantization condition.
Thus, by far the simplest way to account for all the facts is that the duality is an exact
statement. Of course, we only have an explicit construction of the theory on the QFT side,7
so I mean that the QFT must agree with all of the approximations we have to the string
theory, and with any future constructions of the theory. Anyway, the QFT is fully quantum
mechanical and consistent, and as we have noted it includes all the graviton states (with the
right trilinear interactions), so at the very least it is some theory of quantum gravity.
One should note that there are local symmetries on both sides of the duality, the SU (N )
gauge symmetry in the QFT and coordinate invariance and local supersymmetry on the AdS
side. These are different and neither contains the other. It is an important general principle
that dualities acts only on the physical quantities and not on the redundant variables that
we use to construct them.
7
When I say this, I am thinking of the lattice regulator, with supersymmetry broken but restored in the
continuum limit. For the 3+1 example that we are considering, a skeptic can still doubt whether this limit
is controlled, because the coupling is strong, but we will see other examples where the theory flows from a
superrenormalizable gauge theory and so the continuum limit is much simpler.

10
 We conclude with some homework problems to think about before the next lecture:

Ex. 1. It is obviously absurd to claim that a four-dimensional quantum field theory is the
same as a ten-dimensional string theory. Give one or more reasons why it can’t be
true.

Ex. 2. Figure out why your answer to the previous problem is wrong.

2 The two sides

2.1 The gravity side
The massless sector of the IIB string theory, IIB supergravity, controls the physics on long
distance scales on the gravity side. The fields are the metric GM N , the dilaton, and a 3-form
Neveu-Schwarz (NS) field strength HM N P , and 1-, 3- and self-dual 5-form Ramond-Ramond
(RR) field strengths FM , FM N P , and FM N P QR (I’ll use capital Roman indices for spacetime,
saving Greek for the Poincaré directions). You can find the action in your favorite textbook.
Of course the theory also includes excited string states, and D(−1), D1, D3, D5, and D7-
branes, as well as NS5-branes and the various (p, q) bound states. All these branes are much
heavier than the massless fields, but they exist in the AdS5 × S 5 region and represent various
excitations of the gauge theory.
We could write out the field equations here and verify that AdS5 × S 5 is a solution,
but instead I will do something a bit cruder, which gives some useful insight. Leaving out
numerical constants, the relevant terms in the action are
√
Z
0−4
S∼α d10 x −G(e−2Φ R − FM N P QR F M N P QR ) . (2.1)

The 5-form is self-dual, which complicates the action principle. One way to deal with this,
which is most efficient for the energetic argument that we are going to make, is to treat the
fully spatial components as the independent fields.8 Now, we are going to make a Kaluza-
Klein reduction on the S 5 , which is taken to have radius r. The Dirac quantization condition
gives FM N P QR ∼ N α02 , and so
Z p
0−4
S5 ∼ α d5 x −G5 r5 (e−2Φ R5 + e−2Φ r−2 − α04 N 2 r−10 ) . (2.2)

8
In fact, even for forms other rank, we have the option of using either a q-form or a Hodge-dual (D − q)
form, and for energetics it is always useful to work with the one that is spatial, else there are surface terms
that enter. Note that any dilaton or moduli dependence in the field strength action gets inverted for the
Hodge dual.

11
 The second term is from the curvature of the sphere, and in the third we include r-dependence
from the inverse metric. To interpret this in terms of an effective potential, rescale G5 →
r−10/3 e−4Φ/3 G05 to get
Z p
0−4
S5 ∼ α d5 x −G05 (R50 + e4Φ/3 r−16/3 − α04 N 2 e10Φ/3 r−40/3 ) . (2.3)

Then
1 α04 N 2
V (r, Φ) ∼ − + (2.4)
x4/3 x10/3
where x = r4 e−Φ . The negative term from the curvature of S 5 dominates at large x, while
the positive flux term dominates at small x, leaving a minimum at x ∼ α02 N , or

r4 ∼ α02 N eΦ . (2.5)

This is in agreement with the exact solution asserted before, both the scaling and the presence
of a flat direction in the potential so that the dilaton is undetermined. The minimum is at
negative potential, giving rise to an AdS5 solution.
This energetics argument is similar to the way one studies stabilization of moduli in
string compactification [18]. It is crude by comparison to the sophisticated methods that
are employed to find anti-de Sitter solutions in supergravity, but it is a useful complement
to these. Some examples:

Ex. 3. M theory has just a metric and a four-form flux, and the only length scale is the
Planck scale LM . Use the potential method to find AdS7 × S 4 and AdS4 × S 7 solutions,
and determine how the radii of the spheres scale with the number of flux units (note
footnote 8).

Ex. 4. In the IIB theory find an AdS3 × S 3 × T 4 solution with Q5 units of RR 3-form flux on
the S 3 and (again using footnote 8) Q1 units of RR 7-form flux on S 3 × T 4 . Identify
flat directions, and the scalings of the radii with the number of units of each of the
fluxes.

All these solutions have another important property that is more easily seen in 10 dimen-
sions than in the 5-dimensional reduction: the curvature radii of the spherical factors and
the AdS factor are of the same order. If one considers the Einstein equations along the AdS
space and the sphere, one can conclude that these curvature terms are of the same order
unless there are cancellations. In fact, in all explicitly known examples the radii are similar.
A framework for constructing AdS/CFT duals with a large hierarchy is given in Ref. [19],

12
but many details remain to be filled in. Again, from study of moduli stabilization [?] we
know that there are many string theory solutions where all the compact directions are much
smaller than the AdS radius, but the CFT duals are not known. Making progress in this
direction is important for the nonperturbative construction of the landscape, and also for
the top-down construction of applied dualities.

2.2 More about anti-de Sitter space

Anti-de Sitter space has unusual properties that play a key role in the duality. Referring
to the metric (1.5), the radial coordinate range is 0 < z < ∞, and the null geodesics
are z = ±(t − t0 ). These reach the boundary z = 0 in finite time, and so boundary
conditions must be imposed there in order to have a well-defined system. These boundary
conditions also provide a natural set of observables, which generally are complicated to define
in quantum gravity. On the other hand, the geodesics take infinite time to reach or emerge
from z = ∞.
The gravitational reshift means that invariant time intervals and coordinate time intervals
are related as dτ = Ldx0 /z. Correspondingly, the energy E of an excitation as seen by an
invariant observer, and its Killing (conserved) energy P0 conjugate to x0 are related
P0 = LE/z . (2.6)
A given bulk excitation has an energy that increases when it moves nearer the boundary,
which is another way of thinking about the origin of z in the CFT. For Kaluza-Klein exci-
tations near rest E ∼ 1/L and so P0 ∼ 1/z. It is closely related to the earlier interpretation
in terms of gluon separation: if we work in terms of the internal momenta for the gluon we
have pz ∼ 1/z ∼ P0 .9
The Poincaré coordinates (1.5, 1.7) are the natural ones for applications, but this space
is not geodesically complete: the z = ∞ limit is a horizon, and a bulk excitation can actually
fall through it. A nice way to describe the global space AdSD+1 is by embedding it in D + 2
dimensions with signature (D, 2),10
ds2 = L2 γab dX a dX b , γii = +1 , i = 1, . . . , D , γ00 = γD+1,D+1 = −1 . (2.7)
The embedding is
γab X a X b = −1 . (2.8)
9
Of course there are also bulk excitations associated with other energy scales, like the string scale, which
would introduce a possibly large dimensionless constant into the relation P0 ∼ 1/z. In this case it seems
more general to associate z with the size of the state [21].
10
I will always use D for the spacetime dimension of the field theory and D + 1 for the dimension of the
AdS space.

13
Defining U = X D − X D+1 and V = X D + X D+1 , this is

ds2 = L2 (ηµν dX µ dX ν + dU dV ) , ηµν X µ X ν + U V = −1 , (2.9)

where µ runs over the Poincaré directions. Solving for V and defining xµ = X µ /U , z = 1/U ,
the metric in the region U > 0 takes the Poincaré form (1.5). The extension to negative U
doubles this space, and we can go further: the sum (X 0 )2 + (X D+1 )2 is positive definite, so
there is a noncontractible circle and a larger covering space. In terms of
D
X
0 2 1/2 D+1 2 1/2 2
X = (1 + ρ ) cos τ , X = (1 + ρ ) sin τ , ρ = (X i )2 , (2.10)
i=1

the metric is
dρ2
ds2 /L2 = (1 + ρ2 )dτ 2 + + ρ2 dΩ2S D−1 . (2.11)
1 + ρ2
The range 0 < τ < 2π gives the periodic space, and −∞ < τ < ∞ gives global AdS space.
Falling through the horizon would not seem to be relevant to the applications, since these
always involve systems that are finite in size and duration: we turn the experiment off before
anything reaches the horizon! However, the global picture has several uses, some of which
will be developed later: 1) it makes the SO(D, 2) conformal symmetry manifest; 2) it is dual
to the CFT quantized on the space S D−1 rather than RD−1 , if this is what interests us; 3)
for this reason, it gives a mapping between the local operators of the CFT and the spectrum
of states; 4) if we are interested in studying quantum gravity, the global spacetime is the
natural place to formulate it. Curiously, although the Poincaré patch is a small subset of
the global space, the respective Wick rotations of the Poincaré x0 and the global τ yield the
same Euclidean version of anti-de Sitter spacetime.
Writing the SO(D, 2) as δX a = ab X b , we can identify the action of the symmetries on the
Poincaré coordinates: the µν are the Lorentz transformations, the µU are the translations,
the UU are the scale transformations, and the µV are the special conformal transformations.

2.3 Scalar fields in AdS

Consider a Klein-Gordon field, with Lorentzian action:
Z
η
S0 = − D−1 dz dd x (GM N ∂M φ∂N φ + m2 φ2 )
2L
dz dd x 2
Z
η 2 µν 2 2 2


=− z ∂z φ∂ z φ + z η ∂ µ φ∂ ν φ + m L φ . (2.12)
2 z D+1

14
I’ve left in an arbitrary normalization constant η for later use, but you can ignore it for now.
To discuss stability it is useful to define φ = z D/2 ψ, z = − ln y, so that
Z Z
η d −2y µν 2 2 1 2 2

ηD d
y=∞
2
S0 = − dy d x ∂y ψ∂y ψ + e η ∂µ ψ∂ν ψ + [m L + 4 D ]ψ + d xψ  .
2 4 y=−∞
(2.13)
If we add a boundary term
Z
ηζ
S = S0 + Sb = S0 − dD x ψ 2 |y=∞ (2.14)
2
D
with ζ ≥ 2
, and if m2 satisfies the Breitenlohner-Freedman (BF) bound

D2
m2 ≥ − , (2.15)
4
then the Hamiltonian is bounded below term-by-term and the system is stable [22]. In fact,
the Hamiltonian can be organized into a sum of squares under the weaker condition ζ ≥ ∆− ,
defined in Eq. (2.18) below.
We see that a range of tachyonic masses is allowed. If the mass-squared lies below the
BF bound the free-field energy is unbounded below regardless of boundary terms, though
it can be stabilized by higher order bulk terms. As you will hear from some of the other
speakers, this leads to interesting phase transitions.
The momentum space field equation is

z 1+D ∂z (z 1−D ∂z φk ) − (m2 + k 2 z 2 )φk = 0 . (2.16)

Near the z = 0 boundary the k 2 term can be neglected and the solutions behave as

φk ∼ z ∆ , ∆(∆ − D) = m2 L2 . (2.17)

There are two roots,

D2
 
D D
∆± = ± + m2 L2 ≡ ±ν, (2.18)
2 4 2
so the asymptotic behavior is

φ(z, x) ∼ α(x)z ∆− + β(x)z ∆+ , z → 0 . (2.19)

We will consider the ‘standard’ boundary condition α = 0 and the ‘alternate’ boundary
condition [23] β = 0. For the first of these the surface terms in the action and the equations
of motion vanish due to the falloff of the fields. For the second, the vanishing of the surface
term in the equation of motion requires ζ = ∆− ; as we have noted, this gives a stable system

15
 in the tachyonic case.11 With the alternate boundary condition both the boundary term and
the bulk action are divergent, but the divergences cancel (one should put the boundary at
z =  and then take  → 0) provided that ∆− > D2 − 1; beyond this point the k 2 terms in
the action diverge and only the standard boundary condition can be used. We refer to the
z ∆+ solution as normalizable and the z ∆− solution as non-normalizable, though the later is
normalizable in the more restrictive sense just discussed.

Ex. 5. Extend the discussion to the degenerate case ∆+ = ∆− . Show that the alternate
quantization is consistent with conformal invariance in the generic case but not in the
degenerate one.

The general solution to the field equation is

φ(z, x) = eik·x z D/2 J∆−D/2 (qz) , (2.20)

where q 2 = −k 2 < 0 and ∆ is whichever of ∆± we are using. These are plane-wave normal-
izable at the horizon z → ∞, and correspond to a particle that comes out of the horizon,
reflects off the boundary, and returns to the horizon. If conformal symmetry is broken in
such a way that the space is cut off near the horizon at some large value of z, the effective
four-dimensional mass spectrum becomes discrete.

2.4 Conformal field theories

Now we discuss the field theory side of the duality. Again, gauge/gravity duality is much
more general than AdS/CFT, but the conformal case gives the cleanest examples to start
from, so I will start with few general comments about CFT. The basic observables are the
correlation functions of local operators O(x). Under scale transformations, an operator of
dimension ∆ transforms as
O(x) → ζ ∆ O(ζx) . (2.21)
Thus scale invariance determines the form of the two-point function
1
h0|O(x)O(0)|0i = ζ 2∆ h0|O(ζx)O(0)|0i =⇒ h0|O(x)O(0)|0i ∝ . (2.22)
x2∆
Conformal invariance also determines the position-dependence of the three-point function,
so that the basic data is a single coefficient cijk for h0|Oi Oj Ok |0i. This is essentially the
11
One can think of the boundary condition in either of two ways: one can impose it on the fields from
the start, or one can take free boundary conditions and let the equation of motion determine the boundary
behavior. These differ only by contact terms in the operator correlations.

16
OPE coefficient X
Oi (x)Oj (0) = cijk O(x∆k −∆i −∆j )Ok (0) , (2.23)
k

the two-point function being used to raise and lower indices (also, the OPE of descendant
fields is determined in term of their primaries). The OPE has a nonzero radius of convergence,
as it can be regarded under the state-operator mapping as simply the insertion of a complete
set of states. Then higher n-point functions can then be reduced to the two-point function
using the OPE, and so are determined by the cijk . The products in different channels have
overlapping regions of convergence, so there is an associativity relation
X X
cijk cklm ∼ ciln cnjm , (2.24)
k k

with appropriate functions of x inserted. There is always an energy-momentum tensor Tµν

which is traceless and has ∆ = D. Its OPE coefficients are determined by the Ward identities.
This gives a purely algebraic description of a CFT, not referring directly to a Lagrangian.

2.5 The D = 4, N = 4 theory

The particular CFT of interest for the AdS5 × S 5 duality can be compactly defined as the
dimensional reduction of the D = 10 super-Yang-Mills theory, whose action is simply
Z
1
d10 x Tr − 12 FM N F M N + iχ̄γ M DM χ .


S= 2 (2.25)
gYM

To reduce to four dimensions, ignore the coordinates xM for M > 3 and set the corresponding
derivatives to zero. The ten-dimensional gauge field separates into a four-dimensional gauge
field and six scalars,
AM → Aµ , µ ≤ 3 , A m , m ≥ 4 . (2.26)
The Majorana-Weyl spinor χ, with 16 real components, separates into four D = 4 Weyl
spinors. Further, the supersymmetry generators, also a Majorana-Weyl spinor in D = 10,
separates into four sets of D = 4 generators. The scalars and spinors, like the gauge fields,
are SU (N ) adjoints.
The one-loop β-function vanishes by cancellation between the gauge field and matter
contributions; we will give a nonperturbative agrument for this in Sec. 4.1.
It is sometimes useful to single out one supersymmetry and use N = 1 superfields. There
is one gauge multiplet and three chiral multiplets Φ1,2,3 , and the superpotential is

W ∝ Tr(Φ1 [Φ2 , Φ3 ]) . (2.27)

17
The scalar potential X
V ∝ Tr|[Am , An ]|2 (2.28)
m,n

arises from the non-Abelian terms in the field strength in the dimensional ,reduction and as
the sum of F-terms and D-terms in the N = 1 form.
Writing the prefactor of the action (2.25) as N/λ, the action is of order N in the ’t Hooft
limit of large-N fixed λ. Each propagator then contributes a factor N −1 and each interaction
a factor N . We are interested in expectation values of gauge-invariant operators,

O = N Tr(. . .) . (2.29)

With this, an operator behaves like N −1 times an interaction. There is also a factor of N
for each index loop. Filling the index loops in to make the faces of a simplex the total
N -dependence is
N V −O+F −P = N χ−O (2.30)
where V, F, P, O count the vertices, faces, propagators, and operators, and χ is the Euler
number of the simplex. Thus for the O4 expectation value, the leading part of the fully
disconnected term (the product of four one-point functions) has the topology of four two-
spheres for N 4 , the terms with two connected two-point functions has leading behavior
N 0 , and the fully connected amplitude has leading behavior N −2 For the fully connected
contribution for O operators the leading connected amplitude is N 2−O .

3 The dictionary
In addition to classic references [24, 25], the review [2] is good for further details.

3.1 Symmetries
The first check is a comparison of the symmetries of the two systems. On the AdS side there
are the SO(4, 2) × SO(6) symmetries of the AdS5 and S 5 spaces. In the CFT, SO(4, 2)
is the conformal group: we have already discussed at the end of Sec. 2.2 how the various
symmetry transformations are realized. The SO(6) is the symmetry of the scalar field space
Am , with the four fermions transforming as the spinor representation. Equivalently, the
fermions are in the fundamental representation of SU (4) = SO(6), and the six scalars in
the antisymmetric product of two fundamentals. On both sides of the duality this bosonic
symmetry group is extended by supersymmetry to the superconformal P SU (2, 2|4). Again,

18
there are gauge/gravity duals with much less symmetry, but it is useful to take account of
it in this prototype example.
There is also an SL(2, R) weak-strong duality on both sides, in the D = 4, N = 4 gauge
theory and in the IIB string theory.

3.2 Matching of states

The duality implies a 1-1 mapping of the states on the two sides. On the AdS side, since we
are expanding in the loop parameter LP /L, we can count particle states at weak coupling.
On the CFT side, there is the usual isomorphism between states and operators: drawing
a (D − 1)-sphere around an operator, we can go to radial quantization and so we have an
isomorphism between local operators and states on S D−1 . Combining these, we can express
the duality as a 1-1 mapping between the particle species in AdSD+1 and the single-trace
chiral primary operators in the CFT.
The statement is that a field whose boundary behavior is z ∆ maps to an operator of
dimension ∆. The mapping is simply that the scaled boundary limit of the bulk operator is
the CFT operator,
O(x) = CO lim z −∆ φ(x, z) . (3.1)
z→0

This fits with our original picture of the z coordinate emerging from the size of the composite
state: a local operator creates a state with zero size. The constant CO depends on convention,
which we will set later. The scale transformation in the bulk takes φ(z, x) → φ(ζz, ζx) and
so
O(x) → CO lim z −∆ φ(ζx, ζz) = CO lim (z/ζ)−∆ φ(ζx, z) = ζ ∆ O(ζx) , (3.2)
z→0 z→0

which is the scale transformation of an operator of dimension ∆. Recall that the boundary
behavior is related to the mass of the field by m2 L2 = ∆(∆ − D), so that a given mass
corresponds to two possible dimensions depending on the boundary condition taken. The
condition ∆ > D2 − 1 is precisely the lower limit on the dimension of an interacting scalar
field in a unitary CFT.
We have focused on scalar fields, but the scalings work the same way for tensor fields if
we use tangent indices rather than coordinate indices. For example, every CFT contains the
energy-momentum tensor, of dimension D, which maps to the graviton on the AdS side. In
terms of tangent space indices we then have for the metric perturbation hµ̂ν̂ ∼ az D + b, or
hµν = (z/L)−2 hµ̂ν̂ ∼ az D−2 + bz −2 . Note that the larger behavior is the same as that of the
AdS metric itself.
When the CFT has conserved currents, there is a corresponding gauge field in the bulk.

19
The dimension of a conserved current is D − 1, so Aµ̂ ∼ az D−1 + bz and Aµ ∼ az D−2 + b.
The constant term can be identified directly with a background gauge field in the CFT. The
AdS5 × S 5 theory has the SO(6) global symmetry noted earlier, and correspondingly there
are SO(6) Kaluza-Klein gauge fields from the S 5 .
There is an important distinction between states that would be massless in D = 10 and
get effective D = 5 masses only due to the spacetime curvature, and those that are already
massive in D = 10. The former have masses of order 1/L and so dimensions of order one.
In the D = 4, N = 4 case, there is enough supersymmetry in fact to guarantee that these
dimensions are actually independent of the coupling. This is because the maximum spin of
such an operator is 2, which requires it to live in a BPS multiplet of the supersymmetry. For
duals with less supersymmetry the dimensions of such operators will be O(1) at weak and
strong coupling, but can depend on the coupling.
Excited string states, on the other hand, have masses of order α01/2 and so dimensions
at strong coupling of order λ1/4 . This is a striking prediction of the duality. It is plausible
that anomalous dimensions would become large at strong coupling, but there is no simple
analytic argument for this particular behavior. Results based on integrability, which are
supposed to interpolate over all couplings, are consistent with it [26], but there still seems
to be some guesswork involved. Eventually we might hope that the numerical approaches
will reproduce it.
These large dimensions are essential to have a supergravity limit: in supergravity the
maximum spin is two, so any operator of higher dimension must get a large anomalous
dimension.12 This requires very strong coupling, but it is really a stronger statement, as there
are theories with couplings that are much larger than one but still without parametrically
large anomalous dimensions for most operators. Thus we can now elaborate on how we
define strong coupling as a necessary condition for the duality: we need all stringy states, in
particular those with spins greater than two, to get large anomalous dimensions.
Focusing now on the specifics of the D = 4, N = 4, theory, the gauge kinetic term Fµν F µν
maps to the dilaton; it has dimension 4 at weak coupling, and (with appropriate additional
pieces) it is BPS and so its dimension is exactly four. The topological term Fµν F̃ µν maps to
the RR scalar and has dimension 4.
The traces of products of scalars, Tr(Am An . . . Ap ), give a large set of operators. The
linear combinations that are traceless on the SO(6) indices map to shape fluctuations of the
S 5 , mixed with other modes. The operator with l scalars has exact dimension l. The modes
with traces, such as the Konishi operator Tr(Am Am ) must map to excited string states.
12
The higher spin theories constructed by Vasiliev may allow one to go beyond this in some cases.

20
 The superpotential perturbations δW (Φ1,2,3 ) map to the perturbations of the harmonics
of the NS and RR 3-form fluxes FM N P , HM N P (which mix), and an operator with m fields
has dimension m + 1.
There is a one-to-one correspondence between BPS operators of the CFT and supergrav-
ity modes in AdS5 , as required by the duality. It might seem impossible for a four-dimensional
field theory to have as many states as a ten-dimension supergravity theory, much less a string
theory, and this would be many peoples’ answer to Ex. 1, e.g. [27]. The key is that there
are a lot of states due to the large matrices. We have seen that the Kaluza-Klein excitations
come from traces of many operators, and these traces are all independent as N → ∞. At one
level it seems trivial that one can encode anything in a large-N matrix; what is remarkable
is that the codebook is just the N = 4 path integral.
This also provides another nice illustration of the duality working nonperturbatively in
N . At finite N the the traces with more than N operators are not independent, and so there
must be a cutoff on the momentum parallel to the S 5 . One can make harmonics up to l = N ,
corresponding to momenta of order N/L ∼ N 3/4 /LP . It is might seem wrong that the cutoff
is much larger than the Planck scale, but there is no bar to boosting a particle up to highly
super-Planckian momenta. The cutoff arises rather from a curious phenomenon [28]: the
interaction of the particle with the flux on the S 5 causes it to blow up to the point that it
can no longer fit in the space.
Rather than counting supergravity states, we could consider the total partition function
of the theory. In the gravity picture this is dominated by the black 3-brane, whose entropy
per unit volume is
π2
sblack D3 = N 2 T 3 , (3.3)
2
to be compared with the free-field entropy of the CFT,

2π 2 2 3
sfree CFT = N T . (3.4)
3
These agree up to the famous factor of 3/4 [29]: the duality is saying that as the coupling is
increased from zero to infinity in the gauge theory, there is this small shift in the density of
states. The agreement of the temperature dependences is confirming that the AdS theory
is behaving like a four-dimensional CFT, and the agreement of the N -scaling is another
satisfying check. However, there could have been a nontrivial dependence on the ’t Hooft
coupling λ, so that the weak and strong coupling entropies could have differed for example
by a power of λ. There is no independent check of this result, so for now it is a prediction
of the duality rather than a check.

21
3.3 Correlators I
I am going to take the operator relation (3.1) as the starting point for the dictionary. First, I
would like to dispel a myth, that the AdS/CFT dictionary is somehow naturally Euclidean,
and that there are difficulties with extending it to Lorentzian spaces.13 In the first place,
when we compare the spectra of the Hamiltonian on the two sides, this makes no reference
to any signature, since the difference is just whether we evolve forward with e−iHt or e−Hτ .
Secondly, the basic relation (3.1) holds equally in any signature. It is true that in Lorentzian
signature there are several correlators of interest (time-ordered, advanced, retarded), and the
prescriptions for their calculation, at finite temperature, are intricate. This will be discussed
in the lectures by Son; for a recent overview see Ref. [30].
But these questions are not issues for the duality, which via (3.1) just relates any given
correlator in the CFT with the same correlator in the AdS space.
I want to calculate the time-ordered 2-point correlator in several ways. First, the propa-
gator of φ in the bulk of AdSD+1 is
1
h0|T φ(z, x)φ(z 0 , x0 )|0i = G∆ (ξ) , (3.5)
η
where ∆ = ∆+ or ∆− according to the choice of boundary condition. Here
2zz 0
ξ= (3.6)
z 2 + z 02 + (x − x0 )2
is the unique conformal invariant that can be constructed from the two positions. It is a
measure of the distance between them, approaching 1 when they become coincident and 0
when they are far apart. Emphasizing again the signature of spacetime, this is equally valid
in the Euclidean and Lorentzian theories, with
D
X D−1
X
0 2 0i 2 00 2
(x − x ) ≡E i 2 0
(x − x ) ≡M (i + ) (x − x ) + (xi − x0i )2 , (3.7)
i=1 i=1

and we are working in the Poincaré patch.

The propagator G∆ (ξ) is hypergeometric, and Ref. [2] is a good source for more details.
When the points are far apart, and this includes the case that z and/or z 0 approach the
boundary with x − x0 fixed and nonzero, then
C∆ Γ(∆)
G∆ (ξ) → (ξ/2)∆ , C∆ = . (3.8)
2∆ − D π D/2 Γ(∆ − D/2)
13
If anything the reverse is true, since there is still no comprehensive treatment of the contour for the
conformal factor in the path integral for Euclidean gravity.

22
In particular, it behaves as z ∆ z 0∆ as required by the boundary conditions. Thus we can form
the limit (3.1) and conclude that
C∆ CO2
h0|T O(x)O(x0 )|0i = (x − x0 )−2∆ . (3.9)
η(2∆ − D)
Of course the 2-point correlator is determined up to normalization by scale invariance, but
we are just warming up for the later lecturers, who will look at the 2-point function after
turning on temperature, densities, or background fields, where it captures much interesting
physics.
In some cases the normalization of the 2-point function is interesting, but here it is just
the product of a bunch of conventional factors. We now introduce a standard convention. If
we take just the primed field to the boundary we get
 ∆
0 C∆ CO z
h0|T φ(z, x)O(x )|0i = . (3.10)
η(2∆ − D) z 2 + (x − x0 )2
If we now take z → 0, the correlator goes to zero pointwise except at x = x0 , and its
value integrated dD x is z D−∆ CO /η(2∆ − D). It is therefore conventional to set CO =
η(2∆ − D)[N −1 ] (ignore for now factors in square braces!), so that
z→0
h0|T φ(z, x)O(x0 )|0i → z D−∆ δ D (x − x0 )[N −1 ] . (3.11)

At this point I am going to switch to a Euclidean metric, not because I have to but because
it is conventional in this subject (also, the Lorentzian form would have a factor of −i in
Eq. (3.11) and compensating factors elsewhere); anyway, I hope I have already made my
point about the duality being just fine in Lorentizian signature. Note that this also requires
that I flip the sign of the action due to standard conventions. Noting that D − ∆+ = ∆−
(and D − ∆− = ∆+ ), the operator O can be regarded as a delta-function source for the mode
that was previously set to zero.
With this normalization we also have
η(2∆ − D)C∆ −2
h0|T O(x)O(x0 )|0i = [N ] ,
(x − x0 )2∆
 ∆
0 z
h0|T φ(z, x)O(x )|0i = C∆ 2 [N −1 ] . (3.12)
z + (x − x0 )2
The constant η arises from the normalization of the Klein-Gordon action.14
14
Noting the discussion of operator normalization in Sec. 2.5, we see that η must be of order N 0 . On the
other hand, it is often convenient to normalize all closed string fields like the graviton, with a 1/G in the
action, so η ∼ LD−1 /GD+1 ∼ L8 /G10 ∼ N 2 . In this case one must include the factors in braces. In the
remainder of these notes we use only the η ∼ N 0 convention.

23
 The relation (3.11) allows us to connect with the more standard way to write the dic-
tionary. Noting that D − ∆+ = ∆− (and D − ∆− = ∆+ ), the operator O can be regarded
as a delta-function source for the mode that was previously set to zero. In particular, if we
introduce into the CFT path integral a factor
dd x j(x)O(x)
R
e , (3.13)

then we now have

α(x) → j(x) (standard quantization) ; β(x) → j(x) (alternate quantization) . (3.14)

In other words, the coefficient of the fixed (z D−∆ ) mode is that with which the operator O
is added to the CFT, while the coefficient (3.1) of the z ∆ quantized mode is the expectation
value of CO−1 O.
Evaluating the bulk path integral then gives a generating functional
dD x j(x)O(x)
R
h0|T e |0i = Zj → e−Scl . (3.15)

Here Zj is the bulk path integral with boundary condition (3.14), and in the last form we
have evaluated it in the semiclassical approximation, extremizing with respect to the field
with given boundary conditions. Using Eqs. (3.12, 3.14), the extremum is
Z  ∆
D 0 0 z
φcl (z, x) = C∆ d x j(x ) (3.16)
z 2 + (x − x0 )2
This corresponds to the planar approximation in this CFT, but is also exact for the quadratic
theory being studied. This is the form in which the dictionary is often given.
Integrating by parts, the classical bulk action can be written as a surface term (again,
all signs flipped from the previous discussion)
Z Z
η D 2 η 1−D
S0,cl = − D−1 dz d x φcl ( − m )φcl +  dd x φcl (, x)∂ φcl (, x) , (3.17)
2L 2
the first term vanishing by the equation of motion. This is not the whole story because of the
need for a boundary term in the action. To regulate potential divergences, we temporarily
move the boundary in to a small value z = . The boundary term must be such that
the boundary terms in the variation of the action vanish for variations that respect the
boundary conditions; this is in order to have a good variational principle.15 One good choice
15
This may still leave some freedom in the choice of the boundary action, which would just correspond to
changes of convention such as operator redefinition. Also, finiteness of the total bulk plus boundary action
as  → 0 might be desirable, but the divergences are local contact terms and so easily subtracted by hand.

24
of boundary action for the standard quantization is
Z
η∆− −D
Sb =  dD x φ2 (, x) , (3.18)
2
which also guarantees stability as discussed earlier. Then
Z
−D
δ(S0 + Sb ) = −η dD x δφ(, x)(∂ − ∆− )φ(, x) . (3.19)

The boundary condition fixes α(x), so δφ(, x) has only a δβ term, and then only the cross
term with α in the second field survives at the boundary, and this is annihilated by ∂z −∆− .
One can also check that the potentially divergent α2 terms cancel for the same choice of
boundary term. In all
Z
η −D
Scl = −  dD x φcl (, x)(∂ − ∆− )φcl (, x) . (3.20)
2
Inserting φcl leaves a convolution to do [25, 3], but it is not hard to deduce the answer. The
∂ − ∆− kills the α part of the second φcl , so we must get precisely the α part of the first
φcl in order that the term survive at the boundary, and this is just j(x). So
Z
η −∆
Scl = −  dD x j(x)(∂ − ∆− )φcl (, x)
2
Z Z
η 1
→ − (2∆ − D)C∆ d x dD x0 j(x)j(x0 )
D
, (3.21)
2 (x − x0 )2∆
correctly generating the earlier result. The extensions to the alternate quantization, and to
the more general nonconformal boundary condition α = f β + j, are left as exercises. The
solution for the latter is given below Eq. 3.22 of Ref. [31].

3.4 Correlators II
For gauge fields and metric perturbations it is much the same, except that the symmetries
give a natural normalization. We have noted in Sec. 3.2 that the non-normalizable mode of
a gauge field scales as z 0 , so we can interpret its limit directly as a gauge field in the CFT,

lim Aµ (z, x) = Abµ (x) . (3.22)

z→0

This identification is necessary in order that gauge transformations act consistently on

charged operators. This mode couples to the corresponding conserved current in the CFT as
R D
d x Abµ (x)j µ (x). Similarly the nonnormalizable mode of the metric has the same scaling
as the AdS metric, and so we interpret

lim z 2 hµν (z, x) = hbµν (x) , (3.23)

z→0

25
 1
R
which couples to the energy-momentum tensor in the CFT as 2
dD x hbµν (x)T µν (x).
Since later lecturers will be looking the the T T correlator, let us look at one component
of this, at zero temperature. We excite h ≡ h12 with momentum k in the 3-direction. So
far we have worked in position space, but with less symmetry there will not be a simple
coordinate form and so we go to momentum space. The relevant terms in the Euclidean
action in AdS5 are
L3
Z 4
d x dz
d−1
(∂z h∂z h + k 2 h2 ) . (3.24)
3
4L̂P,5 z
This is exactly like the massless Klein-Gordon field. The usual gravitational extrinsic curva-
ture term has been included implicitly to make the action first order [32], but no additional
surface term is needed because ∆− = 0.
The extremum can again be written as a surface term,
L3 −3
Z
Scl = − 3  dd x ∂ (h2 ()) . (3.25)
8L̂P
The classical solution that approaches hb at the boundary and remains bounded at z = ∞
is
k2z 2
hcl (z) = hb K2 (kz) . (3.26)
2
Expanding the Bessel function at small z,
k4z 4
 
2 2 4 4
hcl (z) = hb 1 + O(k z ) − ln k + O(k z ln z) + . . . (3.27)
16
There is a quadratically divergent −2 k 2 piece; this is analytic in k 2 and so is a contact term.
There are also ln  and finite k 4 contact terms. The term of interest ir k 4 ln k, leading to
h2b L3 k 4 ln k
Scl = , (3.28)
16L3P
and
L3
h0|T12 (k)T12 (0)|0i = k 4 ln k . (3.29)
8L̂3P
Using L̂3P,5 = L̂8P,10 /VS 5 where VS 5 = π 3 L5 , this becomes [24]
π 3 L8 N2 4
h0|T12 (k)T12 (0)|0i = − k 4 ln k = − k ln k . (3.30)
8L̂8P,10 32π 2

In position space this goes as 1/x8 as it must for an operator of dimension 4. The central
charge c is defined as −8π 2 times the coefficient of k 4 ln k, i.e. N 2 /4. It must be independent
of the coupling because it is related by supersymmetry to the SO(6) R-anomaly, and indeed
agrees with the weak coupling result [33].

26
4 Breaking symmetries
In this final section I want to generalize the example we have been studying in various ways.
Two challenging goals are to construct duals to landscape AdS states, and to construct non-
Fermi liquids. We will not reach these, but I will discuss some issues. The landscape states,
as mentioned before, have the property that the compact dimensions can be much smaller
than the AdS space, and also that one can uplift them to higher AdS states, and also dS, by
breaking supersymmetry (though as Tom Banks emphasizes, the uplifted states may not be
visible in the AdS construction). The non-Fermi liquids represent a set of observed strongly
coupled fixed points that have defied a clear description.
In the applications to condensed matter physics there are two approaches being taken.
One is to postulate a gravitational theory with some set of fields, implying that the CFT
has the corresponding operators. There will always be the gravitational field, corresponding
to Tµν , and there may be some gauge fields in the bulk, mapping to symmetry currents
in the CFT, and also some charged matter fields, mapping to charged operators of various
dimensions. In this case there is no specific Lagrangian in the CFT, the results apply to any
theory with the given set of operators: once one assumes some specific operators (and their
dimensions and maybe some couplings) the dual allows one to calculate how the system
responds to temperature, electric and magnetic fields, and other perturbations. This has
been instructive, and led to interesting work on non-Fermi liquids which Hong Liu will
discuss, but it would seem more satisfying to start from some brane system and deduce the
Lagrangian and its gravity dual. This top-down approach has been challenging, as I will
describe.

4.1 The Coulomb branch

Starting from the D = 4, N = 4 example, there is a way to break the conformal symmetry
without changing the theory, but just going to a different vacuum. The scalar potential
2
P
m,n Tr|[Am , An ]| vanishes whenever the six matrices Am commute, but they can otherwise
take any values. This corresponds to separating some or all of the D3-branes in the six
transverse directions of the Poincaré picture. There is no force between them due to su-
persymmetry; the potential gets no quantum corrections. Separating the D3-branes breaks
the gauge invariance to a block for each set of coincident D3-branes, a product of U (1) and
SU factors, and it breaks the scale and conformal invariance because some of the branes are
sitting at nonzero r. A fairly recent discussion of this subject is Ref. [34].
From the point of view of the applications, this is more of a nuisance perhaps: most of

27
them don’t have fundamental scalars but in the duality these come along due to supersym-
metry. And, when we break the supersymmetry, it may be that the vacuum that we want
(where the scalar vacuum expectation values are zero) is no longer stable. The top-down
construction of a Fermi sea has been a goal of mine for a long time because of the possibility
of constructing a non-Fermi liquid, but these constructions often seem to have this problem.
We can use these vacua to give a modern nonperturbative proof of the conformal invari-
ance of the N = 4 theory. If we make all of the vevs distinct, the gauge group is U (1)N −1
and the low energy effective fields are just the gauge fields and their scalar Am and fermionic
partners. The low energy effective action for the gauge fields is
X 1
− F F µν .
2 iµν i
(4.1)
i
4gi

If the couplings run, then the gi should be evaluated at a scale of order the masses of the
higged gauge bosons, which are proportional to the Am . But the action will then depend on
the scalars, and this is forbidden by the N = 4 supersymmetry.

4.2 Renormalization group flows

If we start with the D = 4, N = 4 theory and perturb the Hamiltonian by a relevant
operator, ∆ < D, the effect of the perturbation becomes large at low energies. This is
R
readily studied in the dual theory: adding g dD x O(x) to the action for some operator O
shifts the corresponding fixed mode. For an operator of dimension ∆ this mode behaves as
z D−∆ which is just right: for relevant operators, ∆ < D, their effect grows as we move away
from the boundary.16
The interesting question is, what happens at low energies where the perturbation gets
large? The two broad classes of possibility are that we might flow to a new conformal theory,
or that everything just gets massive. We can find examples of each within the N = 4 theory.
A perturbation that preserves N = 1 supersymmetry is to add a mass term for one or more
of the chiral superfields,
X 3
δW = mα Φ2α . (4.2)
α=1

This is a nice example because the low energy physics depends in an interesting way on how
many of the fermions are massive.
16
If we try to add an irrelevant operators ∆ > D, the effect grows as we approach the boundary, the
nonlinearities become large, and generically we will lose the AdS geometry and the duality.

28
 First on the field theory side, for one nonzero mass there are field theory arguments that
the system flows to a new conformal fixed point [35]. For two equal nonzero masses, there is
a N = 2 supersymmetry and the massless sector is just the pure N = 2 super- Yang Mills,
whose low energy physics is the SU (N ) version of Seiberg-Witten [36], where the massless
fields are described by a U (1)N −1 effective theory. (For two unequal masses I am not sure
what happens, but I’m guessing that it flows to the same fixed point). For three nonzero
masses [37], there are actually many supersymmetric vacua: the superpotential equations

mΦα = αβγ [Φβ , Φγ ] (4.3)

give a vacuum for each N -dimensional representation of SU (2). The vacuum corresponding
to the trivial solution Φα = 0 is expected to be confining, while all the others have the gauge
group broken to some subgroup containing non-Abelian factors that confine and always some
unbroken U (1)’s.
On the gravity side, the mass perturbation corresponds to a perturbation of the 3-form
flux on the S 5 . One can analyze this in terms of an effective Lagrangian for just this
mode coupled to gravity. This seems like a cheat, because this is not a real effective field
theory, there is no symmetry that allows us to truncate to just this mode, but there is some
supergravity magic that allows one to extend any solution of this truncated system to the
full theory. The effective potential looks like the curve x3 − x, with one maximum and one
minimum. The AdS5 × S 5 theory actually sits at the maximum, which looks unnerving but
is fine because of the BF bound: the field can’t roll down in time, but it can roll in the radial
direction, ending up at large z at the minimum, which is a new CFT. The overall geometry
in the infrared is AdS5 × X where X is some deformed S 5 with three-form flux [38].
One can repeat this with two masses, but now the field flows off in a different direction,
to infinity. The supergravity magic allows this to be lifted to a ten-dimensional solution
which is now singular, but the singularities have a nice interpretation: they are just N D3-
branes [39, 40]. The z-coordinate (defined so that the metric is L2 ηµν /z 2 + . . .) is cut off at a
finite upper limit, meaning that there is a lower bound on the energies P0 = LE/z. However,
there are massless states on the D3-branes, giving just the expected U (1)N −1 theory.
When masses are turned on for all three superfields, one can try to repeat the above
effective Lagrangian strategy. It again shows the geometry ending before z = ∞, but now
the five-dimensional solution [41] lifts to a ten-dimensional solution with an unphysical sin-
gularity, it is not quite right. One must work in fully ten-dimensional terms. With some
guesswork [37], guided by the field theory, one can find a sensible solutions where the geom-
etry ends at some finite z, and there are explicit NS5- and/or D5-branes, with D3-branes
dissolved in them. The D5’s correspond to non-commuting coordinate matrices Am for the

29
original D3’s, as suggested by (4.3): this is the Myers effect [42], where in a background field
it can become energetically favorable for the D3 coordinates not to commute. For the NS5’s,
there is some strong-coupling dual of this effect, which has no such classical description. The
confining vacuum is the one with just a single NS5-brane.
To talk about confinement we need another observable, which measures the force between
quark-like sources [45]. This is simply the energy of a string that starts and ends on the the
sources (one can give a longer justification, but this is just part of ’t Hooft’s large-N story,
where the string corresponds to a tube of color flux). Now, the effective tension of such a
string, as seen in the gauge theory, is µ(z) = 2πL2 /α0 z 2 , since the tension seen by a local
inertial observer is 2π/α0 . In the conformal theory, as one move the sources apart, the string
can drop down closer and closer to the horizon and reduce its tension indefinitely. One then
gets V (r) ∝ 1/r (as one must in a conformal theory, though it is interesting that at strong
coupling the potential goes as λ1/2 rather than as λ).
In the confining vacua, there is a maximum z and so a minimum tension 2πL2 /α0 zmax 2
.
There are many other realizations of confinement in AdS/CFT [25, 43, 44]; others are simpler
on the gravity side, but generally more complicated on the CFT side. In the Coulombic
vacua, as with two masses, strings hang down from each source and end on a D3-brane,
where they source an explicit flux that can spread out coulombically.

4.3 Multitrace RG flows

If the CFT has a scalar operator of dimension D2 − 1 < ∆ < D2 , this is described in the bulk
by the alternate quantization in which the larger mode α fluctuates and the smaller mode
β is fixed. There are actually no such operators in the N = 4 theory, but there are in many
other examples.
In this case the double-trace operator O2 is relevant [46, 47]:

∆O2 = 2∆ + O(1/N 2 ) , (4.4)

because the large-N factorization means that any graph that connects the two single-trace
operators is down by 1/N 2 . Thus if we add this to the action there will be some RG flow.
To see where it goes, write [47]
D R D 2
E
Zj (g) = e d x {−gO (x)+j(x)O(x)}
β=0
Z D R D E
2
= Dσ e d x {−σ (x)/4g+(j(x)+iσ(x))O(x)}
β=0
Z D R D 2 E
= Dσ e− d x σ (x)/4g
β=j+iσ

30
 Z
dD x σ 2 (x)/4g
R
= Dσ e− h1iβ=j+iσ (4.5)

In the first line we have written the generating functional for expectation values of O with
the perturbation gO2 (x) in the Hamiltonian,17 and in the second line we have introduced
an auxiliary field to relate this to the theory with just the single-trace perturbation. Now
let use integrate out the auxiliary field. In the planar approximation we can just find the
saddle point,18

σ(x) = −2g∂σ Zβ=j+iσ

= 2gi hOiβ=j+iσ
= 2giCO αβ=j+iσ . (4.6)

In the second line we have used the fact that differentiating with respect to j is the same as
with respect to σ, and pulls down a O. In the last line we have used the basic dictionary (3.1).
Eliminating σ
β(x) = j(x) − 2gCO α(x) . (4.7)
In other words, the earlier pure boundary conditions are now replaced by a mixed condition
with a linear relation between α and β [46, 47]. In terms of the field we have

φ(z, x) = z D−∆ {j(x) − 2gCO α(x)} + z ∆ α(x) . (4.8)

At high energy z → 0 the z D−∆ term vanishes (we are in the regime ∆ < D/2) and we have
the alternate quantization. At low energy the z D−∆ term dominates and we have the usual
quantization. Thus we flow from one to the other.
By the way, in the middle two lines of Eq. (4.5) we see an interesting construction: a
field σ coupled to the boundary of the supergravity. In this case the field is auxiliary, but
we could consider more general actions for σ. This corresponds to coupling some strongly
coupled field theory, with a gravity dual, to another field theory (generally weakly coupled)
with an explicit Lagrangian description.

4.4 Orbifolds
Tom Banks discussed some features of these, let me add a little. If we identify the R6 space
transverse to the D3-branes under some discrete group Γ ⊂ SO(6) then the origin becomes
17
It was asked whether g must be positive. Naively, negative g leads to instability, but recently it has been
shown [48] that nonlinear effects of backreaction can stabilize the system.
18
Loop corrections for auxiliary fields are generally uninteresting, being dominated by the cutoff scale for
dimensional reasons, and so absorbable into redefinitions.

31
a singularity. If we then take the near horizon limit, then we get a new duality [49]. On the
supergravity side S 5 is replaced by S 5 /Γ. On the gauge side, the orbifolding retains only the
Γ-invariant fields from the original gauge theory, leaving a quiver gauge theory [50], e.g. for
Γ = Zk we get SU (N )k , with bifundamental matter that depends on how Γ acts on the
different planes in R6 . The surviving supersymmetry can be N = 2, 1, or 0.
In the nonsupersymmetric cases, Tom mentioned that at weak coupling conformal invari-
ance is destroyed by the flow of double-trace interactions. The pathology takes a different
form at strong coupling. If a nonsupersymmetric element of Γ has a fixed point there will
be a localized tachyon, with the result that a hole develops and consumes the space [51]. If
there is no fixed point the same process occurs via tunneling [52]. These are examples of
the general phenomenon that nonsupersymmetric states in string theory are almost always
unstable on long enough time scales.
There is an interesting generalization where SU (N )k is replaced by SU (N1 ) × . . . ×
SU (Nk ). On the supergravity side this corresponds to wrapping higher dimensional branes
(D5 in particular) on cycles hidden in the singularity. The different gauge factors now have
different nonzero β-functions and the couplings flow. Figuring out where the flow ends up
has been a rich subject [43].

4.5 Non-spherical horizons. II

If we combine orbifolding and RG flow we get something new. Some supersymmetric orbifolds
have relevant operators in the twisted sector, corresponding to smoothing the fixed point.
By turning these on one gets to new manifolds, such at the famous conifold [53]. Here the S 5
is replaced by a manifold T 1,1 . In fact one gets a solution to the supergravity field equations
with just metric and five-form if S 5 is replaced by any Einstein space (Rmn = cgmn with c a
positive constant). Supersymmetry requires a more restrictive Einstein-Sasaki condition [54].
The dual gauge theory can be determined in many cases by a combination of orbifolding,
RG flow, moving to the Coulomb branch, and so on. The toric case, where X has a U (1)3
symmetry, is understood best [55]. There has been recent progress on the more general
case [56].

4.6 Nonconformal branes [57]

Suppose we try to repeat the near-horizon limit for D2-branes. Here the gauge theory is
2
superrenormalizable: at high energy it is weak, and the dimensionless coupling gYM N/E
becomes small. In the IIA supergravity theory, the curvature is small in some range, but

32
becomes large in string units near the boundary. Thus we have the same complementarity
as for D3-branes, but at different scales within a given theory rather than by varying the
coupling. The superrenormalizable gauge theory is the Lagrangian description, and the
gravitational theory describes the low energy physics.
There is more to the story, though. The growth of the dimensionless gauge coupling is
also reflected in the blowup of the dilaton at large z. In this regime one must lift to M
theory, and the extreme low energy limit is the M2-brane geometry. Thus the extreme IR of
D = 3, N = 8 gauge theory provides one Lagrangian description of this geometry. I like this
D = 3 theory as an ‘existence proof’ of AdS/CFT as a theory of quantum gravity, because we
can regulate it on the lattice without exact supersymmetry, and the superrenormalizability
makes it easy to see the restoration of the symmetry in the continuum limit, since only a
finite number of perturbative graphs need be canceled.
If we periodically identify one of the Poincaré directions of the D = 4 theory, we would
expect its low energy physics to be given by the D = 3 theory. Correspondingly on the
gravity side, the periodic dimension pinches off near the horizon due to the L2 /z 2 in the
metric, and a T -duality brings us to the D2-brane geometry. One can repeat this to get
down to D0-branes, and I think that it is accurate to say that the BFSS Matrix Theory
conjecture [58] is implied by AdS/CFT duality. It is less clear that the reverse is true,
because the BFSS duality focuses on certain extreme low energy observables.
If we move up in dimension to D4-branes, the order of things is reversed. The gauge
theory is classical at low energy, while at higher energies the effective description is the D4-
brane metric, and at higher energies still the dilaton becomes large and we go over to the
M5-brane geometry. We do not need the duality to solve the low energy physics because it
is free, but it provides a UV completion. Higher dimensional branes become more intricate.
Question: How can the N 2 D2-brane entropy go over to the N 3/2 M2-brane entropy?
Ref. [57] finds that the black 2-brane entropy is

sblack D2 ∼ N 2 T 2 (λ/T )−1/3 . (4.9)

Unlike the D3-brane, this depends on the ’t Hooft coupling, and as λ increases more degrees
of freedom freeze out. Why the D2 should be different from the D3 in this regard is a puzzle!
But because λ is dimensional, this implies an extra T 1/3 dependence. Exercise: by following
through the calculations in Ref. [57], show that this suppression at low temperatures is just
enough to make the D2 result match on to N 3/2 at the crossover temperature between the
D2 and M2 pictures.

33
4.7 D2-D6
(The discussion in this and the following section is based in part on unpublished work with
E. Silverstein.)
For a D = 3 gauge theory with matter, the one-loop running is
µ∂µ γ 2 = −γ 2 + b0 γ 4 , (4.10)
where γ 2 = gYM
2
/µ is the dimensionless gauge coupling. The constant b0 is proportional to
the one-loop β-function from D = 4. If b0 > 0, meaning the that D = 4 theory is free in
the IR, then the one-loop equation (4.10) exhibits an IR fixed point at nonzero coupling
γ 2 = 1/b0 . This is not reliable in general, but if we take an N -vector of matter fields the
one-loop term dominates higher orders and we can believe the result. Thus we generate a
large class of conformal theories in D = 3 [59], gauge theory analogs of the Wilson-Fisher
fixed point.
We can build such theories with branes by taking D2-branes lying within D6-branes, i.e.
they are extended in the directions
0 1 2 3 4 5 6 789
D2: × × × (4.11)
D6: × × × × × × ×
This preserves half the supersymmetry of the D2’s, namely D = 3, N = 4. The 2-2 strings
are the SU (N2 ) reduction of the D = 4 theory and give b0 = 0 for this group. The 2-6
strings provide N6 fundamentals and give a positive contribution to b0 . The 6-6 strings are
free at low energy, because of the high dimension of the D6-brane, and do not of course
contribute to the SU (N2 ) running. Thus, for N6  N2 there is a weakly coupled fixed point.
In general one cannot guarantee that this fixed point survives to strong coupling, but here
there is enough supersymmetry to assure that it does. Moreover there is a gravity dual [60].
Normally it is difficult to describe such localized branes explicitly, but here the M-theory lift
of the D6-brane allows to construct the solution as the ZN6 orbifold of the M2-geometry of
charge N2 N6 [61]. For N6  N2  N65 there is a weakly coupled IIA description, and for
N65  N2 the M theory picture is the valid one. These same scalings are found in a different
orbifold of this geometry studied in Ref. [62].

4.8 D2-D8
Now consider D2-D8,
0 1 2 3 4 5 6 7 8 9
D2: × × × . (4.12)
D8: × × × × × × × × ×

34
This case is similar to the preceding, except that there is no supersymmetry, and the 2-8
strings are all fermionic. This sounds a good starting point for a Fermi liquid: turn on a
chemical potential for these strings and the geometry will tell us what happens when they
are in interaction with a strong gauge coupling (in the D2-D6 case, there are also 2-6 bosons,
and energetically it is likely that the charge will be carried predominantly be these).
An effective potential analysis along the lines of that in Sec. 2.1 indicates that there is
the possibility of an AdS4 solution before we turn on the chemical potential. The analysis
is crude, though, because there is less symmetry (the D8’s lie on a particular S 5 within the
S 6 surrounding the D2’s).
Further, there are several instabilities that may keep us from getting to the desired strong
coupling phase. The system may be unstable along the Coulomb branch direction, the D2-
branes separating (E. Silverstein has christened this seasickness because the Fermi sea make
the AdS throat eject the branes). The D8’s may be repelled from the D2’s, which is chiral
symmetry breaking, a long-standing issue in carrying these D = 3 gauge fixed points to strong
coupling. Shape modes of the S 6 surrounding the D2’s may become BF-forbidden tachyons;
this is a common occurence in nonsupersymmetric AdS spaces [63]. These instabilities may
be present even before the chemical potential is turned on, and the fermion density seems
to make some worse (e.g. the seasickness along the Coulomb branch).
It seems likely that one can eventually, by adding additional ingredients, stabilize any of
these and get to a Fermi liquid dual, at least one that is metastable (one additional knob
is to add D6-branes in various orientations). Therefore let us ignore them as far as we can
and turn on a chemical potential. Something nice then happens [64], in that the system
does exhibit at least one feature of a Fermi surface: there are particle-hole states of finite
momentum and arbitrarily small energy, as one gets by taking a fermion from just below
the surface and moving it to a point just above the surface somewhere else. The calculation
is this: turning on a chemical potential introduces a radial electric field on the D6-brane.
One measures the correlator of the fermion current (dual to a 6-6 string) by expanding the
D-brane Dirac-Born-Infeld action, and because of the special form of this action (which in
particular has a maximum electric field) the correlator has the distinctive behavior of an
imaginary part that goes down to zero frequency at finite momentum.
There are still several puzzles, however: 1) this behavior is also seen in the D2-D6 system,
where a Fermi liquid might not have been expected; 2) in Fermi liquids there is a maximum
momentum of 2kF , the maximum distance an electron can be moved, while here there is
no maximum — perhaps this is a strong-coupling effect; 3) the large field near the origin
backreacts on the metric in a divergent way, and the resolution of this singularity is not

35
understood in general, it could gap the surface; 4) even when the backreaction is paramet-
rically suppressed by taking N2 large, the DBI action breaks down near the horizon due to
the field gradients.
Question: What about T -dual configurations, like D3-D7 in the nonsupersymmetric ori-
entation
0 1 2 3 4 5 6 7 8 9
D3: × × × × ? (4.13)
D7: × × × × × × × ×
Good question: the intersection is still D = 3, and the fermions live there, but the gauge
fields now move in D = 4. Things are less IR singular but still interesting: the perturbative
analysis seems more tractable. So maybe this is another interesting case. For D4-D6,

0 1 2 3 4 5 6 7 8 9
D4: × × × × × , (4.14)
D8: × × × ××××

the gauge fields move in D = 5 and are classical at low energy, they do not give interesting
dynamics.

4.9 Fractional D3-D7

Finally I mention a case that addresses the generality of AdS/CFT. If we replace the N = 4
gauge theory with N = 2, and add Nf = 2Nc fundamentals, we get another theory which
is conformal for all values of the coupling. We can get this from Nc fractional D3-branes at
an orbifold point plus Nf D7-branes. However, the conformal limit is singular and suggests
a T -duality to a IIA configuration [65]. Even then, an effective potential analysis suggests
that the scales of both the compact space and the AdS space are stringy, meaning that
the anomalous dimensions remain of order one even at large coupling, they do not grow as
a power of the coupling. Thus, it appears that the necessary condition of large operator
dimensions is rather special. Further this seems to go against the principle that whenever a
coupling becomes very large we can find a small expansion parameter in a different picture.
Ref. [66] discusses this system in much more detail, and proposes a more intricate dual
picture.

Acknowledgments
I would like to thank Allan Adams, Alex Buchel, Tom Faulkner, Sean Hartnoll, Idse Heemskerk,
Gary Horowitz, Jacopo Orgera, Amanda Peet, Joao Penedones, Eva Silverstein, Matt Strassler

36
and James Sully for collaborations on various aspects of this subject, and Tom Banks, Gary
Horowitz, Hong Liu, Don Marolf, Dam Son, and Sho Yaida for useful discussions. I would
also like to thank the students at TASI for their incessant questions and all-around enthu-
siasm, which made this lecture series a rewarding experience. This work was supported in
part by NSF grants PHY05-51164 and PHY07-57035.

References
[1] O. Aharony, S. S. Gubser, J. M. Maldacena, H. Ooguri and Y. Oz, “Large N field the-
ories, string theory and gravity,” Phys. Rept. 323, 183 (2000) [arXiv:hep-th/9905111].

[2] E. D’Hoker and D. Z. Freedman, “Supersymmetric gauge theories and the AdS/CFT
correspondence,” arXiv:hep-th/0201253.

[3] John McGreevy, 8.821, String Theory, Fall 2008,

http://ocw.mit.edu/courses/physics/8-821-string-theory-fall-2008/.

[4] S. Weinberg and E. Witten, “Limits On Massless Particles,” Phys. Lett. B 96, 59 (1980).

[5] S. R. Coleman and J. Mandula, “All Possible Symmetries of the S Matrix,” Phys. Rev.
159, 1251 (1967).

[6] G. ’t Hooft, “Dimensional reduction in quantum gravity,” arXiv:gr-qc/9310026.

[7] L. Susskind, “The World As A Hologram,” J. Math. Phys. 36, 6377 (1995) [arXiv:hep-
th/9409089].

[8] G. Bertsch, S. J. Brodsky, A. S. Goldhaber and J. F. Gunion, “Diffractive Excitation

In QCD,” Phys. Rev. Lett. 47, 297 (1981);
A. H. Mueller, “Color Transparency and Nuclear Shadowing,” CU-TP-412.

[9] L. N. Lipatov, “The Bare Pomeron In Quantum Chromodynamics,” Sov. Phys. JETP
63, 904 (1986) [Zh. Eksp. Teor. Fiz. 90, 1536 (1986)];
R. C. Brower, J. Polchinski, M. J. Strassler and C. I. Tan, “The Pomeron and
Gauge/String Duality,” JHEP 0712, 005 (2007) [arXiv:hep-th/0603115].

[10] I. Heemskerk, J. Penedones, J. Polchinski and J. Sully, “Holography from Conformal

Field Theory,” JHEP 0910, 079 (2009) [arXiv:0907.0151 [hep-th]].

[11] E. Witten, “The 1 / N Expansion In Atomic And Particle Physics,” S. R. Coleman,

“1/N,”

37
[12] G. ’t Hooft, “A planar diagram theory for strong interactions,” Nucl. Phys. B 72, 461
(1974).

[13] J. M. Maldacena, “The large N limit of superconformal field theories and supergravity,”
Adv. Theor. Math. Phys. 2, 231 (1998) [Int. J. Theor. Phys. 38, 1113 (1999)] [arXiv:hep-
th/9711200].

[14] A. M. Polyakov, “String theory and quark confinement,” Nucl. Phys. Proc. Suppl. 68,
1 (1998) [arXiv:hep-th/9711002].

[15] G. T. Horowitz and A. Strominger, “Black strings and P-branes,” Nucl. Phys. B 360,
197 (1991).

[16] J. Polchinski, “Dirichlet-Branes and Ramond-Ramond Charges,” Phys. Rev. Lett. 75,
4724 (1995) [arXiv:hep-th/9510017].

[17] A. Strominger and C. Vafa, “Microscopic Origin of the Bekenstein-Hawking Entropy,”

Phys. Lett. B 379, 99 (1996) [arXiv:hep-th/9601029].

[18] E. Silverstein, “TASI / PiTP / ISS lectures on moduli and microphysics,” arXiv:hep-
th/0405068.

[19] J. Polchinski and E. Silverstein, “Dual Purpose Landscaping Tools: Small Extra Di-
mensions in AdS/CFT,” arXiv:0908.0756 [hep-th].

[20] S. Kachru, R. Kallosh, A. D. Linde et al., “De Sitter vacua in string theory,” Phys. Rev.
D68, 046005 (2003). [hep-th/0301240].

[21] L. Susskind, E. Witten, “The Holographic bound in anti-de Sitter space,” [hep-
th/9805114];
A. W. Peet, J. Polchinski, “UV / IR relations in AdS dynamics,” Phys. Rev. D59,
065011 (1999). [hep-th/9809022].

[22] P. Breitenlohner and D. Z. Freedman, “Positive Energy In Anti-De Sitter Backgrounds

And Gauged Extended Supergravity,” Phys. Lett. B 115, 197 (1982);
P. Breitenlohner and D. Z. Freedman, “Stability In Gauged Extended Supergravity,”
Annals Phys. 144, 249 (1982).

[23] I. R. Klebanov and E. Witten, “AdS/CFT correspondence and symmetry breaking,”

Nucl. Phys. B 556, 89 (1999) [arXiv:hep-th/9905104].

[24] S. S. Gubser, I. R. Klebanov and A. M. Polyakov, “Gauge theory correlators from

non-critical string theory,” Phys. Lett. B 428, 105 (1998) [arXiv:hep-th/9802109].

38
[25] E. Witten, “Anti-de Sitter space and holography,” Adv. Theor. Math. Phys. 2, 253
(1998) [arXiv:hep-th/9802150].

[26] N. Gromov, V. Kazakov and P. Vieira, “Exact AdS/CFT spectrum: Konishi dimension
at any coupling,” arXiv:0906.4240 [hep-th].

[27] R. Penrose, The Road to Reality, Knopf, 2004.

[28] J. McGreevy, L. Susskind and N. Toumbas, “Invasion of the giant gravitons from anti-de
Sitter space,” JHEP 0006, 008 (2000) [arXiv:hep-th/0003075].

[29] S. S. Gubser, I. R. Klebanov and A. W. Peet, “Entropy and Temperature of Black

3-Branes,” Phys. Rev. D 54, 3915 (1996) [arXiv:hep-th/9602135].

[30] K. Skenderis and B. C. van Rees, “Real-time gauge/gravity duality: Prescription, Renor-
malization and Examples,” JHEP 0905, 085 (2009) [arXiv:0812.2909 [hep-th]].

[31] I. Heemskerk, J. Polchinski, “Holographic and Wilsonian Renormalization Groups,”

[arXiv:1010.1264 [hep-th]].

[32] G. W. Gibbons and S. W. Hawking, “Action Integrals And Partition Functions In

Quantum Gravity,” Phys. Rev. D 15, 2752 (1977).

[33] S. S. Gubser and I. R. Klebanov, “Absorption by branes and Schwinger terms in the
world volume theory,” Phys. Lett. B 413, 41 (1997) [arXiv:hep-th/9708005].

[34] K. Skenderis and M. Taylor, “Holographic Coulomb branch vevs,” JHEP 0608, 001
(2006) [arXiv:hep-th/0604169].

[35] R. G. Leigh and M. J. Strassler, “Exactly Marginal Operators And Duality In Four-
Dimensional N=1 Supersymmetric Gauge Theory,” Nucl. Phys. B 447, 95 (1995)
[arXiv:hep-th/9503121].

[36] M. R. Douglas and S. H. Shenker, “Dynamics of SU(N) supersymmetric gauge theory,”

Nucl. Phys. B 447, 271 (1995) [arXiv:hep-th/9503163].

[37] J. Polchinski and M. J. Strassler, “The string dual of a confining four-dimensional gauge
theory,” arXiv:hep-th/0003136.

[38] A. Khavaev, K. Pilch and N. P. Warner, “New vacua of gauged N = 8 supergravity in

five dimensions,” Phys. Lett. B 487, 14 (2000) [arXiv:hep-th/9812035];
A. Karch, D. Lust and A. Miemiec, “New N = 1 superconformal field theories and their
supergravity description,” Phys. Lett. B 454, 265 (1999) [arXiv:hep-th/9901041];

39
 D. Z. Freedman, S. S. Gubser, K. Pilch and N. P. Warner, “Renormalization group
flows from holography supersymmetry and a c-theorem,” Adv. Theor. Math. Phys. 3,
363 (1999) [arXiv:hep-th/9904017].

[39] K. Pilch and N. P. Warner, “N = 2 supersymmetric RG flows and the IIB dilaton,”
Nucl. Phys. B 594, 209 (2001) [arXiv:hep-th/0004063].

[40] A. Buchel, A. W. Peet and J. Polchinski, “Gauge dual and noncommutative exten-
sion of an N = 2 supergravity solution,” Phys. Rev. D 63, 044009 (2001) [arXiv:hep-
th/0008076].

[41] L. Girardello, M. Petrini, M. Porrati and A. Zaffaroni, “The supergravity dual of N =

1 super Yang-Mills theory,” Nucl. Phys. B 569, 451 (2000) [arXiv:hep-th/9909047].

[42] R. C. Myers, “Dielectric-branes,” JHEP 9912, 022 (1999) [arXiv:hep-th/9910053].

[43] I. R. Klebanov and M. J. Strassler, “Supergravity and a confining gauge theory: Du-
ality cascades and chiSB-resolution of naked singularities,” JHEP 0008, 052 (2000)
[arXiv:hep-th/0007191].

[44] J. M. Maldacena and C. Nunez, “Towards the large N limit of pure N = 1 super Yang
Mills,” Phys. Rev. Lett. 86, 588 (2001) [arXiv:hep-th/0008001].

[45] S. J. Rey and J. T. Yee, “Macroscopic strings as heavy quarks in large N gauge theory
and anti-de Sitter supergravity,” Eur. Phys. J. C 22, 379 (2001) [arXiv:hep-th/9803001];
“Wilson loops in large N field theories,” Phys. Rev. Lett. 80, 4859 (1998) [arXiv:hep-
th/9803002].

[46] E. Witten, “Multi-trace operators, boundary conditions, and AdS/CFT correspon-

dence,” arXiv:hep-th/0112258.

[47] M. Berkooz, A. Sever and A. Shomer, “Double-trace deformations, boundary conditions

and spacetime singularities,” JHEP 0205, 034 (2002) [arXiv:hep-th/0112264];
S. S. Gubser, I. R. Klebanov, “A Universal result on central charges in the presence of
double trace deformations,” Nucl. Phys. B656, 23-36 (2003). [hep-th/0212138].

[48] T. Faulkner, G. T. Horowitz and M. M. Roberts, “Holographic quantum criticality from

multi-trace deformations,” arXiv:1008.1581 [hep-th].

[49] S. Kachru and E. Silverstein, “4d conformal theories and strings on orbifolds,” Phys.
Rev. Lett. 80, 4855 (1998) [arXiv:hep-th/9802183].

40
[50] M. R. Douglas and G. W. Moore, “D-branes, Quivers, and ALE Instantons,” arXiv:hep-
th/9603167.

[51] A. Adams, J. Polchinski and E. Silverstein, “Don’t panic! Closed string tachyons in
ALE space-times,” JHEP 0110, 029 (2001) [arXiv:hep-th/0108075].

[52] G. T. Horowitz, J. Orgera and J. Polchinski, “Nonperturbative Instability of

AdS5 xS 5 /Zk ,” Phys. Rev. D 77, 024004 (2008) [arXiv:0709.4262 [hep-th]].

[53] I. R. Klebanov and E. Witten, “Superconformal field theory on threebranes at a Calabi-

Yau singularity,” Nucl. Phys. B 536, 199 (1998) [arXiv:hep-th/9807080].

[54] A. Kehagias, “New type IIB vacua and their F-theory interpretation,” Phys. Lett. B
435, 337 (1998) [arXiv:hep-th/9805131];
D. R. Morrison and M. R. Plesser, “Non-spherical horizons. I,” Adv. Theor. Math. Phys.
3, 1 (1999) [arXiv:hep-th/9810201].

[55] S. Franco, A. Hanany, D. Martelli, J. Sparks, D. Vegh and B. Wecht, “Gauge theories
from toric geometry and brane tilings,” JHEP 0601, 128 (2006) [arXiv:hep-th/0505211].

[56] P. S. Aspinwall and D. R. Morrison, “Quivers from Matrix Factorizations,”

arXiv:1005.1042 [hep-th].

[57] N. Itzhaki, J. M. Maldacena, J. Sonnenschein and S. Yankielowicz, “Supergravity and

the large N limit of theories with sixteen Phys. Rev. D 58, 046004 (1998) [arXiv:hep-
th/9802042].

[58] T. Banks, W. Fischler, S. H. Shenker and L. Susskind, “M theory as a matrix model:

A conjecture,” Phys. Rev. D 55, 5112 (1997) [arXiv:hep-th/9610043].

[59] T. Appelquist and D. Nash, “Critical Behavior in (2+1)-dimensional QCD,” Phys. Rev.
Lett. 64, 721 (1990).

[60] O. Pelc and R. Siebelink, “The D2-D6 system and a fibered AdS geometry,” Nucl. Phys.
B 558, 127 (1999) [arXiv:hep-th/9902045].

[61] N. Itzhaki, A. A. Tseytlin and S. Yankielowicz, “Supergravity solutions for branes lo-
calized within branes,” Phys. Lett. B 432, 298 (1998) [arXiv:hep-th/9803103].

[62] O. Aharony, O. Bergman, D. L. Jafferis and J. Maldacena, “N=6 superconformal Chern-

Simons-matter theories, M2-branes and their gravity duals,” JHEP 0810, 091 (2008)
[arXiv:0806.1218 [hep-th]].

41
[63] O. DeWolfe, D. Z. Freedman, S. S. Gubser, G. T. Horowitz and I. Mitra, “Stability of
AdS(p) x M(q) compactifications without supersymmetry,” Phys. Rev. D 65, 064033
(2002) [arXiv:hep-th/0105047].

[64] M. Kulaxizi and A. Parnachev, “Holographic Responses of Fermion Matter,” Nucl.

Phys. B 815, 125 (2009) [arXiv:0811.2262 [hep-th]].

[65] M. Grana and J. Polchinski, “Gauge / gravity duals with holomorphic dilaton,” Phys.
Rev. D 65, 126005 (2002) [arXiv:hep-th/0106014].

[66] A. Gadde, E. Pomoni and L. Rastelli, “The Veneziano Limit of N=2 Superconformal
QCD: Towards the String Dual of N=2 SU (Nc ) SYM with Nf = 2Nc ,” arXiv:0912.4918
[hep-th].

42