Introduction to the AdS/CFT correspondence

Alfonso V. Ramallo

arXiv:1310.4319v3 [hep-th] 1 Nov 2013

Departamento de Fsica de Partculas

Universidade de Santiago de Compostela
and
Instituto Galego de Fsica de Altas Enerxas (IGFAE)
E-15782 Santiago de Compostela, Spain

Abstract
This is a pedagogical introduction to the AdS/CFT correspondence, based on lectures
delivered by the author at the third IDPASC school. Starting with the conceptual basis
of the holographic dualities, the subject is developed emphasizing some concrete topics,
which are discussed in detail. A very brief introduction to string theory is provided,
containing the minimal ingredients to understand the origin of the AdS/CFT duality.
Other topics covered are the holographic calculation of correlation functions, quarkantiquark potentials and transport coefficients.

[email protected]

Contents
1 Introduction and motivation

2 The anti-de Sitter space

2.1 Counting the degrees of freedom in AdS . . . . . . . . . . . . . . . . . . . .

6
7

3 String theory basics

4 Large N expansion in gauge theories

14

5 D-branes

16

6 D-branes and gravity

18

7 The
7.1
7.2
7.3

AdS/CFT for N = 4
Conformal symmetry .
Supersymmetry . . . .
Reduction on the S5 .

SU (N ) SYM
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

21
22
23
23

8 A scalar field in AdS

8.1 Application to the KK spectrum in AdS5 S5 . . . . . . . . . . . . . . . . .
8.2 Normalizable and non-normalizable modes . . . . . . . . . . . . . . . . . . .
8.3 Higher spin fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

25
27
28
29

9 Correlation functions
9.1 One-point function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.2 Linear response theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.3 Two-point function for a scalar field . . . . . . . . . . . . . . . . . . . . . . .

29
30
32
33

10 Quark-antiquark potential
10.1 Quark-antiquark potential at finite temperature . . . . . . . . . . . . . . . .
10.2 Quark-antiquark potential in a confining background . . . . . . . . . . . . .

37
42
43

11 Black hole thermodynamics

11.1 Application to the Schwarzschild black hole . . . . . . . . . . . . . . . . . .
11.2 AdS black hole . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11.3 Comparison with field theory . . . . . . . . . . . . . . . . . . . . . . . . . .

45
47
48
50

12 Transport coefficients

52

13 Holographic viscosities

56

Introduction and motivation

The AdS/CFT correspondence is a duality relating quantum field theory (QFT) and gravity.
More precisely, the correspondence relates the quantum physics of strongly correlated manybody systems to the classical dynamics of gravity in one higher dimension. This duality
is also referred to as the holographic duality or the gauge/gravity correspondence. In its
original formulation [13], the correspondence related a four-dimensional Conformal Field
Theory (CFT) to the geometry of an anti-de Sitter (AdS) space in five dimensions.
In the study of collective phenomena in condensed matter physics it is quite common that
when a system is strongly coupled it reorganizes itself in such a way that new weakly coupled
degrees of freedom emerge dynamically and the system can be better described in terms of
fields representing the emergent excitations. The holographic duality is a new example of this
paradigm. The new (and surprising!) feature is that the emergent fields live in a space with
one extra dimension and that the dual theory is a gravity theory. As we will argue below,
the extra dimension is related to the energy scale of the QFT. The holographic description is
a geometrization of the quantum dynamics of the systems with a large number of degrees of
freedom, which makes manifest that there are deep connections between quantum mechanics
and gravity.
The gauge/gravity duality was discovered in the context of string theory, where it is quite
natural to realize (gauge) field theories on hypersurfaces embedded in a higher dimensional
space, in a theory containing gravity. However, the study of the correspondence has been
extended to include very different domains, such as the analysis of the strong coupling
dynamics of QCD and the electroweak theories, the physics of black holes and quantum
gravity, relativistic hydrodynamics or different applications in condensed matter physics
(holographic superconductors, quantum phase transitions, cold atoms, ...). In these lectures
we will concentrate on some particular topics, trying to present in clear terms the basic
conceptual ideas, as well as the more practical calculational aspects of the subject. For
reviews on the different aspects of the duality see [410].
We will start by motivating the duality from the Kadanoff-Wilson renormalization group
approach to the analysis of lattice systems. Let us consider a non-gravitational system in a
lattice with lattice spacing a and hamiltonian given by:
X
H =
Ji (x, a) Oi (x) ,
(1)
x,i

where x denotes the different lattice sites and i labels the different operators Oi . The Ji (x, a)
3

are the coupling constants (or sources) of the operators at the point x of the lattice. Notice
that we have included a second argument in J i , to make clear they correspond to a lattice
spacing a. In the renormalization group approach we coarse grain the lattice by increasing
the lattice spacing and by replacing multiple sites by a single site with the average value
of the lattice variables. In this process the hamiltonian retains its form (1) but different
operators are weighed differently. Accordingly, the couplings Ji (x, a) change in each step.
Suppose that we double the lattice spacing in each step. Then, we would have a succession
of couplings of the type:
Ji (x, a) Ji (x, 2a) Ji (x, 4a) .

(2)

Therefore, the couplings acquire in this process a dependence on the scale (the lattice spacing)
and we can write them as Ji (x, u), where u = (a, 2a, 4a, ) is the length scale at which
we probe the system. The evolution of the couplings with the scale is determined by flow
equations of the form:



Ji (x, u) = i Jj (x, u), u ,

(3)
u
u
where i is the so-called -function of the i th coupling constant. At weak coupling the i s
can be determined in perturbation theory. At strong coupling the AdS/CFT proposal is
to consider u as an extra dimension. In this picture the succession of lattices at different
values of u are considered as layers of a new higher-dimensional space. Moreover, the sources
Ji (x, u) are regarded as fields in a space with one extra dimension and, accordingly we will
simply write:
Ji (x, u) = i (x, u) .
(4)
The dynamics of the sources i s will be governed by some action. Actually, in the AdS/CFT
duality the dynamics of the i s is determined by some gravity theory ( i. e. by some
metric). Therefore, one can consider the holographic duality as a geometrization of the
quantum dynamics encoded by the renormalization group. The microscopic couplings of the
field theory in the UV can be identified with the values of the bulk fields at the boundary of
the extra-dimensional space. Thus, one can say that the field theory lives on the boundary
of the higher-dimensional space (see figure 1).
The sources i of the dual gravity theory must have the same tensor structure of the
corresponding dual operator Oi of field theory, in such a way that the product i Oi is a
scalar. Therefore, a scalar field will be dual to a scalar operator, a vector field A will be
dual to a current J , whereas a spin-two field g will be dual to a symmetric second-order
tensor T which can be naturally identified with the energy-momentum tensor T of the
field theory.
The holographic duality raises several conceptual issues which should be addressed in
order to fully understand it. The simplest of these issues is the matching of the degrees of
freedom on both sides of the correspondence. Let us consider a QFT in a d-dimensional
4

J|UV

UV

IR

Figure 1: On the left we illustrate the Kadanoff-Wilson renormalization of a lattice system.

In the AdS/CFT correspondence the lattices at different scale are considered as the layers
of the higher dimensional space represented on the right of the figure.
spacetime (i.e. in d 1 spatial dimensions plus time). The number of degrees of freedom of
a system is measured by the entropy. On the QFT side the entropy is an extensive quantity.
Therefore, if Rd1 is (d 1)-dimensional spatial region, at constant time, its entropy should
be proportional to its volume in d 1 dimensions:
SQF T Vol(Rd1 ) .

(5)

On the gravity side the theory lives in a (d + 1)-dimensional spacetime. How such a higher
dimensional theory can contain the same information as its lower dimensional dual?. The
crucial point to answer this question is the fact that the entropy in quantum gravity is
subextensive. Indeed, in a gravitational theory the entropy in a volume is bounded by the
entropy of a black hole that fits inside the volume and, according to the so-called holographic
principle, the entropy is proportional to the surface of the black hole horizon (and not to
the volume enclosed by the horizon). More concretely, the black hole entropy is given by the
Bekenstein-Hawking formula:
1
AH ,
(6)
SBH =
4GN
where AH is the area of the event horizon and GN is the Newton constant. In order to apply
(6) for our purposes, let Rd be a spatial region in the (d + 1)-dimension spacetime where the
gravity theory lives and let us assume that Rd is bounded by a (d 1)-dimensional manifold
Rd1 ( Rd1 = Rd ). Then, according to (6), the gravitational entropy associated to Rd
scales as:
SGR (Rd ) Area(Rd ) Vol(Rd1 ) ,
(7)
which agrees with the QFT behavior (5). In this lectures we will present several refinements
5

of this argument and we will establish a more precise matching of the degrees of freedom of
the two dual theories.

The anti-de Sitter space

In general, finding the geometry associated to a given QFT is a very difficult problem.
However, if the theory is at a fixed point of the renormalization group flow (with a vanishing
-function) it has conformal invariance (is a CFT) and one can easily find such a metric.
Indeed, let us consider a QFT in d spacetime dimensions. The most general metric in
(d + 1)-dimensions with Poincare invariance in d-dimensions is:
ds2 = 2 (z) (dt2 + d~x2 + dz 2 ) ,

(8)

where z is the coordinate of the extra dimension, ~x = (x1 , , xd1 ) and (z) is a function
to be determined. If z represents a length scale and the theory is conformal invariant, then
ds2 must be invariant under the transformation:
(t, ~x) (t, ~x) ,

z z .

(9)

Then, by imposing the invariance of the metric (8) under the transformation (9), we obtain
that the function (z) must transform as:
(z) 1 (z) ,

(10)

which fixes (z) to be:

L
,
(11)
z
where L is a constant. Thus, by plugging the function (11) into the metric (8) we arrive at
the following form of ds2 :
L2
ds2 = 2 (dt2 + d~x2 + dz 2 ) ,
(12)
z
which is the line element of the AdS space in (d + 1)-dimensions, that we will denote by
AdSd+1 . The constant L is a global factor in (12), which we will refer to as the anti-de Sitter
radius. The (conformal) boundary of the AdS space is located at z = 0. Notice that the
metric (12) is singular at z = 0. This means that we will have to introduce a regularization
procedure in order to define quantities in the AdS boundary.
The AdS metric (12) is a solution of the equation of motion of a gravity action of the
type:
Z
h
1
(13)
I =
dd+1 x g 2 + R + c2 R2 + c3 R3 + ] ,
16GN
with GN is the Newton constant, the ci are constants, g = det(g ), R = g R is the scalar
curvature and is a cosmological constant. In particular, if c2 = c3 = = 0 the action (13)
(z) =

becomes the Einstein-Hilbert (EH) action of general relativity with a cosmological constant.
In this case the equations of motion are just the Einstein equations:
R

1
g R = g .
2

(14)

Taking the trace on both sides of (14), we get that the scalar curvature is given by:
R = g R = 2

d+1
.
d1

(15)

Inserting back this result in the Einstein equation (14) we get that the Ricci tensor and the
metric are proportional:
2
g .
(16)
R =
d1
Therefore the solution of (14) defines an Einstein space. Moreover, the Ricci tensor for the
metric (12) can be computed directly from its definition. We get:
R =

d
g .
L2

(17)

By comparing (16) and (17) we get that the AdSd+1 space solves the equations of motion
(14) of EH gravity with a cosmological constant equal to:
=

d(d 1)
,
2L2

(18)

which is negative. It follows from (15) and (18) that the scalar curvature for the AdSd+1
space with radius L is given by:
d(d + 1)
R =
.
(19)
L2
In these lectures we will be mostly interested in gauge theories in 3 + 1 dimensions, which
corresponds to taking d = 4 in our formulas. The dual geometry found above for this case
is AdS5 . This was precisely the system studied in [1] by Maldacena, who conjectured that
the dual QFT is super Yang-Mills theory with four supersymmetries (N = 4 SYM).

2.1

Counting the degrees of freedom in AdS

After having identified the AdS space as the gravity dual of a field theory with conformal
invariance, we can refine the argument of section 1 to match the number of degrees of freedom
of both sides of the duality.
Let us consider first the QFT side. To regulate the theory we put both a UV and IR
regulator. We place the system in a spatial box of size R (which serves as an IR cutoff) and
we introduce a lattice spacing  that acts as a UV regulator. In d spacetime dimensions the
7

system has Rd1 /d1 cells. Let cQF T be the number of degrees of freedom per lattice site,
which we will refer to as the central charge. Then, the total number of degrees of freedom
of the QFT is:
 R d1
QF T
Ndof
=
cQF T .
(20)

The central charge is one of the main quantities that characterize a CFT. If the CFT is
a SU (N ) gauge field theory, such as the theory with four supersymmetries which will be
described below, the fields are N N matrices in the adjoint representation which, for large
N , contain N 2 independent components. Thus, in these SU (N ) CFTs the central charge
scales as cSU (N ) N 2 .
Let us now compute the number of degrees of freedom of the AdSd+1 solution. According
to the holographic principle and to the Bekenstein-Hawking formula (6), the number of
degrees of freedom contained in a certain region is equal to the maximum entropy, given by
AdS
Ndof
=

A
,
4GN

(21)

with A being the area of the region at boundary z 0 of AdSd+1 . Let us evaluate A by
integrating the volume element corresponding to the metric (12) at a slice z =  0:
Z
 L d1 Z
d1
d x g =
A =
dd1 x .
(22)

d1
d1
R
, z=
R
The integral on the right-hand-side of (22) is the the volume of Rd1 , which is infinite. As
we did on the QFT side, we regulate it by putting the system in a box of size R:
Z
dd1 x = Rd1 .
(23)
Rd1

Thus, the area of the A is given by:

A =

 RL d1

.
(24)

Let us next introduce the Planck length lP and the Planck mass MP for a gravity theory in
d + 1 dimensions as:
1
GN = (lP )d1 =
.
(25)
(MP )d1
Then, the number of degrees of freedom of the AdSd+1 space is:
1  R d1  L d1
AdS
Ndof
=
.
(26)
4 
lP
QF T
AdS
By comparing Ndof
and Ndof
we conclude that they scale in the same way with the IR
and UV cutoffs R and  and we can identify:
1  L d1
= cQF T .
(27)
4 lP

QFT side

Put the system in a box of size R with lattice spacing !

QF T
Ndof

! R "d1
!

cQF T central charge

cQF T

# dof per cell

In SU (N ) YM cSU (N ) N 2

Gravity side
This gives the matching
condition between
and QFT that we were looking
!
!
" # gravity
! for."Notice

d1
L d1

RL In
that a theory
the coefficient
multiplying
its action
is large.
this
=
dd1
x
A = is (semi)classical
dd1 x gwhen
A
=

!
Rd1
Rd1 , z="
!
integral is dominated by a saddle point. The action of our gravity theory
Area of thecase
AdSthe path
!
of radius L contains a factor Ld1 /GN . Thus, taking into account the
boundary in the AdSd+1 space
dd1 x = Rd1
definition ofRthe
d1 Planck length in (25), we conclude that the classical gravity theory is reliable
if:
 
1 L d1  1 ,
d1
(28)
classical
gravity
in
AdS
Planck length and mass
GN = (lP )
=
(MP )d1 lP
which happens when the AdS radius is large in Planck units. Since the scalar curvature
! L "d1
goes like 1/L2 , the curvature is small in1 !
Planck
Thus, a QFT has a classical gravity
R "d1units.
AdS
=
N
From the entropy
formula
dof
dual when cQF T is large, or equivalently
is al large number of degrees of freedom per
4 if there
!
P
unit volume or a large number of species (which corresponds to large N for SU (N ) gauge
! L "d1
theories).
! L "d1
>> 1
classical gravity in AdS
lP
cQF T

3 lP String theory basics

large # dof per unit volume

As mentioned in the introduction, the AdS/CFT correspondence was originally discovered

in the context of string theory. Although it can be formulated without making reference
to its stringy origin, it is however very convenient to know its connection to the theory
of relativistic strings in order to acquire a deep understanding of the duality. We will
start reviewing qualitatively this connection in this section (refs [1116] contain systematic
expositions
with different
of technical
details). of high spin (60s)
Historical
originof string theory
Description
of levels
hadronic
resonances

String theory basics

M2

Regge trajectories

M2 J

CHAPTER 1 Introduction

13

Figure extended
2: In a Regge
trajectory
mass square M distance
of the particles
Basic objects
along
somethe
characteristic
ls grows linearly with their
Figure 1.5

spin J.
CHAPTER 1 Introduction

Fundamental particles are extended one-dimensional objects

called strings.

13

Historically, string theory was introduced in the sixties as an attempt to describe the
non-local theory
hadronic resonances of high spin observed in the experiments. Experimentally, the mass

Figure 1.5 Fundamental particles are extended one-dimensional objects

called strings.

open

Figure 1.6 A closed string has no loose ends.

closed

Strings can be open (Fig. 1.5) or closed (Fig. 1.6), the latter meaning that the
ends are connected.
Excitations of the string give different fundamental particles. As a particle moves
through space-time, it traces out a world line. As a string moves through spacetime, it traces out a worldsheet (see Fig. 1.7), which is a surface in space-time
parameterized by ( , ). A mapping x ( , ) maps a worldsheet coordinate
Figure 1.6 A closed string has no loose ends.
( , ) to the space-time coordinate x.
So, in the world according to string theory, the fundamental objects are tiny
strings with a length on the order of the Planck scale (10 33 cm). Like any string,
Strings can be open (Fig. 1.5) or closed (Fig. 1.6), the latter meaning that the
ends are connected.
t
t
Excitations of the string give different fundamental particles. As a particle moves
through space-time, it traces out a world line. As a string moves through spacetime, it traces out a worldsheet (see Fig. 1.7), which is a surface in space-time
parameterized by ( , ). A mapping x ( , ) maps a worldsheet coordinate
x
x
( , ) to the space-time coordinate x.
So, in the world according to string theory, the fundamental objects are tiny
Schematic representation of a
A particle moving through
33
strings with a length on the order of the Planck scale (10 cm). Like any string,
string moving through spacespace-time has a world

The rotational degree of freedom gives rise to high spins and Regge trajectories

In modern language a meson is a quark-antiquark pair joined by a string

HAPTER 1 Introduction
t

time, it is represented by a

13worldsheet (a tube for a closed

string)

line

square of these particles is linearly related to its spin J:

M2 J .

(29)

It is then said that the hadrons are distributed along Regge trajectories (see figure 2).
String theory was introduced to reproduce this behavior. Actually, it is not difficult to verify
qualitatively that the rotational degree of freedom of the relativistic string gives rise to Regge
trajectories as in (29). Indeed, let us suppose that we have an open string with length L
and tension T which is rotating around its center of mass. The mass of this object would
be M T L, whereas its angular momentum J would be J P L, with P being its linear
momentum. In a relativistic theory P M , which implies that J P L M L T 1 M 2
or, equivalently, M 2 T J. Thus, we reproduce the Regge behavior (29) with the slope
being proportional to the string tension T .
The basic object of string theory is an object extended along some characteristic distance
ls . Therefore, the theory is non-local. It becomes local in the point-like limit in which the
size ls 0. The rotation degree of freedom of the string gives rise to Regge trajectories
similar to those observed experimentally. In modern language one can regard a meson as a
quark-antiquark pair joined by a string. The energy of such a configuration grows linearly
with the length, and this constitutes a model of confinement.
The classical description of a relativistic string is directly inspired from that of a point
particle in the special theory of relativity. Indeed, let us consider a relativistic point particle
of mass m moving in a flat spacetime with Minkowski metric . As it moves the particle
describes a curve in spacetime (the so-called worldline), which can be represented by a
function of the type:
x = x ( ) ,
(30)
where x is the coordinate in the space in which the point particle is moving (the target
space) and parameterizes the path of the particle (the worldline coordinate). The action
of the particle is proportional to the integral of the line element along the trajectory in
spacetime, with the coefficient being given by the mass m of the particle:
Z
Z 1
p
S = m ds = m
d x x .
(31)
0

Let us now consider a relativistic string. A one-dimensional object moving in spacetime

describes a surface (the so-called worldsheet). Let dA be the area element of the worldsheet
(see figure 3). Then, the analogue of the action (31) for a string is the so-called Nambu-Goto
action:
Z
SN G = T
dA ,
(32)
where T is the tension of the string, given by:
T =

1
,
20
10

(33)

and 0 is the Regge slope (0 has units of (length)2 or (mass)2 ). The string length and
mass are defined as:

1
.
(34)
ls = 0 =
Ms
x0

dA

x2

x1

Figure 3: The string sweeps in spacetime a two-dimensional worldsheet.

Then, the string tension is related to ls and Ms as:
T =

1
Ms2
=
.
2ls2
2

(35)

Let us write more explicitly the Nambu-Goto action (32). With this purpose we will take
two coordinates ( = 0, 1) to parameterize the worldsheet ( ( 0 , 1 ) = (, )). Let us
assume that the string moves in a target space M with metric G . The embedding of
in the spacetime M is characterized by a map M with X ( ). The induced
metric on is:
G X X .
G
(36)
Thus, the Nambu-Goto action of the string is:
Z q
d2 .
det G
SN G = T

(37)

The action (37) depends non-linearly on the embedding functions X (, ). The classical
equations of motion derived from (37) are partial differential equations which, remarkably,
can be solved for a flat target spacetime with G = , both for Neumann and Dirichlet boundary conditions. Actually, the functions X (, ) can be represented in a Fourier
expansion as an infinite superposition of oscillation modes, much as in the string of a violin.
The quantization of the string can be carried out by using the standard methods in quantum physics. The simplest way is just by canonical quantization, i.e. by considering that
the X are operators and by imposing canonical commutation relations between coordinates
11

and momenta. As a result one finds that the different oscillation modes can be interpreted
as particles and that the spectrum of the string contains an infinite tower of particles with
growing masses and spins that are organized in Regge trajectories, with 1/ls being the mass
gap.
The close scrutiny of the consistency conditions of the quantum string reveals many surprises (quantizing the string is like opening Pandoras box!). First of all, the mass spectrum
contains tachyons (particles with m2 < 0), which is a signal of instability. In order to avoid
this problem one must consider a string which has also fermionic coordinates and require that
the system is supersymmetric (i.e. that there is a symmetry between bosonic and fermionic
degrees of freedom). This generalization of the bosonic string (37) is the so-called superstring. Another consistency requirement imposed by the quantization is that the number D
of dimensions of the space in which the string is moving is fixed. For a superstring D = 10.
This does not mean that the extra dimensions have the same meaning as the ordinary ones
of the four-dimensional Minkowski spacetime. Actually, the extra dimensions should be regarded as defining a configuration space (as the phase space in classical mechanics does).
We will see below that this is precisely the interpretation that they have in the context of
the AdS/CFT correspondence.
Another important piece of information about the nature of string theory is obtained
from the analysis of the spectrum of massless particles. Massive particles have a mass
which a multiple of 1/ls and, therefore, they become unobservable in the low-energy limit
ls 0. Then, after eliminating the tachyons by using supersymmetry, the massless particles
are the low-lying excitations of the spectrum. We must distinguish the case of open and
closed strings. The spectrum of open strings contains massless particles of spin one with the
couplings needed to have gauge symmetry. These particles can be naturally identified with
gauge bosons (photons, gluons,...). The big surprise comes when looking at the spectrum
of closed strings, since it contains a particle of spin two and zero mass which can only be
interpreted as the graviton (the quantum of gravity). Besides, quantum consistency of the
propagation of the string in a curved space implies Einstein equations in ten dimension plus
corrections:
R + = 0 .
(38)
Then, one is led to conclude that string theory is not a theory of hadrons but a theory of
quantum gravity!!. Thus, the string length ls should be of the order of the Planck length
lP and not of the order of the hadronic scale 1 fm). Elementary strings without internal
structure and zero thickness were born for the wrong purpose. Moreover, in addition to the
graviton, the massless spectrum contains antisymmetric tensor fields of the form A1 p+1 ,
which will become very relevant in formulating the AdS/CFT correspondence.
As strings propagate through the target space M, they can suffer interactions by splitting
in two or more parts or by joining with other strings. The worldsheet for one of these
interactions is just a two-dimensional surface with holes and boundaries which can be thought
12

~g
s

Figure 4: A vertex in field theory corresponds to a two-dimensional surface with boundaries

(left). The triple vertex for three closed strings is represented on the right as a pants surface.
as a vertex in QFT perturbation theory in which the lines has been thickened, as shown in
figure 4. In closed string theory the basic interaction is a string splitting into two (or the
inverse process of two closed strings merging into one). The corresponding worldsheet has
the form of pants, which can be obtained by thickening a triple vertex in QFT. A loop
in string perturbation theory can be obtained with two triple vertices, which produces a
worldsheet which is a Riemann surface with a hole. Higher order terms in perturbation
theory correspond to having more holes on the worldsheet (see figure 5). Therefore, the
perturbative series in string theory is a topological expansion!. In the topology of twodimensional surfaces, the number h of holes (or handles) of the surface is called the genus of
the surface. In string perturbation theory, h is just the number of string loops.
Let us suppose that we weigh every triple vertex with a string coupling constant coupling
gs . It is clear from the discussion above that string perturbation theory is an expansion of
the type:

X
gs2h2 Fh (0 ) ,
(39)
A =
h=0

where A is some amplitude. In the next section we will identify a topological expansion in
gauge theories which has the same structure as the series (39).

+ ...

Figure 5: Perturbative expansion of the amplitude corresponding to four closed strings.

13

Large N expansion in gauge theories

Let us consider U (N ) Yang-Mills theory with lagrangian

L =


1 
Tr F F ,
2
g

(40)

where the non-abelian gauge field strength is given by:

F = A A + [A , A ] .

(41)

In the equations above A is an N N matrix whose elements can be written as Aa , b , where

a, b are indices that run from 1 to N . Let us rewrite L as:

Topological expansion of gauge theories

Topological expansion of gauge

N theories

L =

Tr F F

(42)

!
"
1

!
"
U (N ) Yang-Mills theory
F
L=
1 2 Tr F
U (N ) Yang-Mills
theory
F F
L = 2 gTr
g
where = g 2 N is the so-called t Hooft
coupling. The t Hooft expansion [17] corresponds
Fkeeping
= Afixed
A and
+ [Ato
, Aperform
]
Aan
N N matrix
to
expansion
of the amplitudes
in powers of N . It turns
Aa

F
= A A + [A , A ]
A
N N matrix
Aa , b , b

out that the different powers of N correspond to the different topologies. One can prove this
!
"
N !notation
by adopting
a double
line
Then,
one can verify that the
" for the gauge
N
=
g 2propagator.
N
t Hooft
coupling
Rewrite
Tr F F
F
L=
=
g2 N

t Hooft
coupling
Rewrite
LL
L = Tr F

color lines form the perimeter of an oriented polygon (a face). Polygons join at a common
edges
such a way expansion
that every vacuum
graph of
is associated
to a triangulated two-dimensional
Hooftin
topological
powers
fixed
ttHooft
topological
expansion ininpowers
of N N
at at
fixed

surface formed by sewing all polygons along the edges.

doubleline
linenotation
notation
double
gaugepropagator
propagator
gauge

a a
b b

N
N

c c
d d

vertices
vertices

N N

index
indexloop
loop
color
colorsum
sumNN
Figure 6: Gauge propagator and vertices in the double line notation of Yang-Mills U (N )
theories.
The
thethe
perimeter
of of
an an
oriented
polygon
(a face)
Thecolor
colorlines
linesform
form
perimeter
oriented
polygon
(a face)
It is not difficult
to find
the
powers
of N
and appearing in a given diagram D with no
Polygons
join
atat
a common
edge
Polygons
join
a common
edge
external lines. The contributions of the gauge propagator and the vertices are displayed in
vacuum
graph
is associated
to to
a triangulated
2d surface
Every
vacuum
graph
iscontributes
associated
a triangulated
2dofsurface
figure 6. Moreover,Every
every
index
loop
with
a power
N . Suppose that E is the
number of propagators (edges) of D connecting two vertices, V is the number of vertices and
Diagram
DDnumber
with
F is the
of index loops (faces). Then, one can show that:
Diagram
with
EE
connecting
vertices
Vvertices
 E two
number
numberofofpropagators
propagators(edges)
(edges)
connecting
two
N
N F = N F E+V EV .
(43)
D
V number of vertices
V number of vertices
N
! !"E"! N!"V "
F number of index loops (faces)
F number of index loops (faces)

Power of N
Power of N
= 2 2h
= 2 2h

F E+V =
F E+V =

D
D N
14N

N NVF =F N F E+V
EV
N = N F E+V EV

Euler characteristic (topological invariant)

Euler characteristic (topological invariant)

h genus number of handles

h genus number of handles

Let us denote by the combination of F , E and V that appears in the exponent of N in

(43):
F E+V = ,
(44)
which is nothing but the Euler characteristic of the triangulation, which only depends on the
topology of the surface. If the diagram triangulates a surface with h handles (an h-torus),
is given by:
= 2 2h .
(45)
Thus the diagrams are weighed with a power of N determined by the number of handles
of the surface that they triangulate (see figure 7). The planar diagrams are those that can
be drawn on a piece of paper without self-crossing. They correspond to h = 0 and their
contribution is of the type N 2 n , for some power n which depends on the diagram considered.
Clearly, as the dependence on N goes like N 22h the diagrams with h = 0 are the dominant
ones in the large N expansion. For this reason the large N limit is also called the planar
limit of the gauge theory (see [18] for more details and examples).

Figure 7: The planar diagram on the left can be drawn on a sphere, whereas the non-planar
diagram on the right must be drawn on a torus.
The effective action Z can be obtained by computing the sum over the connected vacuumto-vacuum diagrams. It has the structure:
log Z =

X
h=0

22h

cl,h =

l=0

N 22h fh () ,

(46)

h=0

where fh () is the sum of Feynman diagrams that can be drawn in a surface of genus h.
This clearly suggest a connection with the perturbative expansion of string theory written
in (39) if one identifies the string coupling constant gs as:
gs

1
.
N

15

(47)

Heuristically one can say that gauge theory diagrams triangulate the worldsheet of an effective string. The AdS/CFT correspondence is a concrete realization of this connection in the
limit (N, ) (i.e. for planar theories in the strongly coupled regime). This dual stringy
description of gauge theory is nothing but a quantum version of the familiar description of
electromagnetism in terms of the string-like lines of force.

D-branes

Besides the perturbative structure reviewed above, string theories have a non-perturbative
sector, which plays a crucial role in connecting them with gauge theories. The relevant objects in this non-perturbative sector are the solitons, which are extended objects. The most
important solitons for the AdS/CFT correspondence are the Dp-branes, that are objects
extended in p + 1 directions (p spatial + time). The Dp-branes can be defined as hypersurfaces where strings end (see figure 8). They can be obtained by quantizing the string with
fixed ends along hyperplanes (Dirichlet boundary conditions). In addition, they can also
be understood as objects charged under the antisymmetric tensor fields A1 p+1 of string
theory, which naturally couple to the Dp-brane worldvolume as:
Z
A1 p+1
A1 p+1 dx1 dxp+1 ,
(48)
Mp+1

with Mp+1 being the worldvolume of the Dp-brane. The Dp-branes are dynamical objects
that can move and get excited. Schematically, their action takes the form:
Z
SDp = TDp
dp+1 x [ ] ,
(49)
with TDp being the tension of the Dp-brane, which in terms of the string coupling constant
gs and the string length ls is given by:
TDp =

(2)p

1
.
gs lsp+1

(50)

Notice from (50) that TDp gs1 , which confirms that the Dp-branes are non-perturbative
objects in string theory. The dependence of TDp on ls in (50) is fixed by dimensional analysis.
The Dp-branes can have two classes of excitations. The first one correspond to rigid
motions and deformations of their shape. These degrees of freedom can be parameterized by
the 9p coordinates i (i = 1, , 9p) transverse to the (p+1)-dimensional worldvolume in
the ten-dimensional target space. The i s are just scalar fields on the Dp-brane worldvolume.
Besides, the Dp-branes can have internal excitations. To get a clue on how to represent these
excitations, let us recall that the endpoint of the string is a charge. When there is a charge,
16

re scalar fields on the worldvolume

ations
A

the string is a charge that

field on the worldvolume

d action

ers of

!
" open strings can end. The endpoints of the
Figure 8: The D-branes are hyperplanes
where
p+1
strings source
S gauge
= fields
T A on the
det(g + 2l2 F )
d D-brane
x worldvolume.
DBI

Dp

a gauge field is sourced, as illustrated in figure 8. Then, it follows that a Dp-brane has an
abelian gauge field A ( 2= 20, , p)
i living
i in its worldvolume. The action of the Dp-brane
)

g
=

+
(2l

takes
which
into accountsthese two types
of excitations is the so-called Dirac-Born-Infeld
action, which can be written as:
Z
q
p+1
SDBI = TDp
d x det(g + 2 ls2 F ) ,
(51)
F and

where g is the induced metric on the worldvolume and F = A A is the field

"
1
strength

i of the
i gauge field A . When the worldvolume gauge field is not excited (F = 0)
F F
+ +
the
2 DBI action (51) becomes the natural generalization of the Nambu-Goto action (37) for
an object extended along p spatial directions.
Let us consider a Dp-brane in flat space. The induced metric in this case takes the form:

gauge+scalars+(fermions)

gY2 M = 2(2)p2 lsp3 ggs

= + (2ls2 )2 i i ,

(52)

where the i are the coordinates that parameterize the embedding of the brane. Let us
now expand the square root of (51) in powers of F and . The quadratic terms in this
expansion can be written as:

1 1
1
(2)
SDBI = 2
F F + i i + ,
(53)
gY M 4
2
which is just the ordinary action of a gauge field and 9 p scalar fields. In (53) gY M is the
Yang-Mills coupling, which, in terms of ls and gs , is given by:
gY2 M = 2(2)p2 lsp3 gs .
17

(54)

In addition to A and , in superstring theory the branes have fermionic excitations, which
can be represented in terms of fermionic fields.
We have just discovered one of the most important features of the Dp-branes: they contain
a gauge theory living in their wordlvolume!. In the case of a single Dp-brane the gauge group
is U (1), as in (53). However, if we consider a stack of coincident parallel Dp-branes the gauge
group gets promoted to U (N ). Indeed, one can verify in this case that the fields A and i
are matrices transforming in the adjoint representation of U (N ). The non-diagonal elements
of these fields correspond to excitations connecting different branes, whereas the diagonal
ones are excitations on a single brane. Actually, the U (1) component of the U (N ) gauge
theory can be decoupled from the SU (N ) fields. Therefore, we conclude that a stack of N
Dp-branes realizes a SU (N ) gauge theory in p + 1 dimensions.
The D-branes have brought a completely new perspective on gauge theories which gives
rise to what is called brane engineering. The geometric realization of the gauge symmetry
that they provide allows to perform a series of transformations which lead to discover new
unexpected properties of gauge theories. For example, one can move the branes, place
them in different spaces, etc. In the gauge theory, these transformations lead to dualities,
reductions of the amount of supersymmetry, changes of the field content, etc. In general
we have a novel geometric insight on gauge dynamics. The AdS/CFT correspondence is an
important outcome of this new realization of the gauge symmetry.
The particular case of the D3-branes is specially relevant in what follows. In this case we
have a 3 + 1 worldvolume and 10 4 = 6 scalar fields. The corresponding four-dimensional
SU (N ) gauge theory can be identified with super Yang-Mills theory with four supersymmetries (N = 4 SYM). This theory is an exact CFT at the quantum level and will be the basic
example of the AdS/CFT correspondence. Notice from (54) that the Yang-Mills coupling is
dimensionless in this case and is related to the string coupling constant gs as:
gY2 M = 4gs ,

(D3 branes) .

(55)

D-branes and gravity

String theory is a gravity theory in which all types of matter distort the spacetime. This
distortion is determined by solving Einstein equations which follow from the action:
Z

1
d10 x g R + .
(56)
S =
16 G
The ten-dimensional Newton constant in (56) is related to string parameters as:
16 G = (2)7 gs2 ls8 .

18

(57)

The dependence of G on ls in (57) follows from dimensional analysis. The dependence on gs

follows by comparing two-to-two scattering amplitudes in string theory and in supergravity
(see figure 9).

string

graviton

~ gs2

~G

Figure 9: In string theory the exchange of a graviton is obtained as the low-energy limit of
the exchange of a closed string.
The Dp-branes (and other extended objects) are solutions of the Einstein equations. Let
us consider these solutions at the linearized level of weak gravity (i.e. when we are far from
the object). The linearized metric for a point-like object in a D-dimensional spacetime is:



2
ds2 (1 + 2) dt2 + 1

dx21 + + dx2D1 ,
(58)
D3
where the function parameterizes the deviation of the metric from the flat Minkowski metric
in D dimensions. The specific form (58) can be obtained by solving the linearized Einstein
equations and generalizes the standard result in general relativity in four dimensions. As in
this latter particular case, one can identify the function with the Newtonian gravitational
potential, as seen by comparing the motion along a geodesic and Newtons law. Thus:
GM
,
(59)
rD3
q
where M is the mass of the particle and r = x21 + + x2D1 is the radial coordinate of
the space. The function is a solution of the Poisson equation in D 1 dimensions. Notice
that the power D 3 is equal to the number dT of dimensions transverse to the object (i. e.
dT = D 1) minus 2:
D 3 = dT 2 .
(60)

For an extended object (along p spatial directions) the corresponding linearized metric takes
the form:




2(p + 1)

ds2 (1 + 2) dt2 + dx21 + + dx2p + 1
dx2p+1 + + dx2D1 . (61)
Dp3
Since now the number of transverse directions is dT = D 1 p, the function must be:

GM
rDp3
19

(62)

The D3-brane solution is a solution of 10d supergravity corresponding to a stack of N

coincident D3-branes. The (exact) metric takes the form:


1
1
(63)
ds2 = H 2 dt2 + dx21 + dx22 + dx23 + H 2 (dr2 + r2 d25 ) ,
with d25 is the line element of a unit S5 . Notice that dr2 + r2 d25 is the flat metric of R6 .
The function H(r) is the so-called warp factor:
H =1+

L4
,
r4

(64)

with the constant L being given by:

L4 = 4gs N ls4 .

(65)

At linearized level (when r ) the solution (63) corresponds to the general solution in
(61) with D = 10, p = 3 and the function given by:
1 L4
=
.
4 r4

(66)

Comparing this expression of with the one written in (62) for D = 10 and p = 3 we
conclude that:
GM L4 .
(67)
Let us find this last relation from a different reasoning. As G gs2 ls8 (see (57)) and since M
should be the mass of a stack of N D3-branes with tension TD3 1/(gs ls4 ), we should have:
GM gs2 ls8

N
N gs ls4 ,
gs ls4

(68)

which, apart from a numerical factor is precisely the value of L4 written in (65).
The geometry of this solution is asymptotically the Minkowski spacetime in 10d with a
throat of infinite size (see figure 1). In the throat we can take r << 1 and neglect the 1 in
the function H. This defines the so-called near-horizon limit. Actually, taking
L4
,
r4

(69)

L2 2
r2
2
2
2
2
+
dx
)
+
(dt
+
dx
+
dx
dr + L2 d25 .
2
3
1
2
2
L
r

(70)

H
the metric (63) becomes:
ds2 =

To identify this near-horizon metric, let us change variable as:

r=

L
.
z

20

(71)

Then, the line element (70) can be rewritten as:

ds2 =

L2
(dt2 + dx21 + dx22 + dx23 + dz 2 ) + L2 d25 ,
z2

(72)

which is precisely the metric of the product space AdS5 S5 , with L being the radius of
both factors.

The AdS/CFT for N = 4 SU (N ) SYM

We are now in a position to formulate the AdS/CFT correspondence, as originally proposed

by Maldacena in [1]. We have seen two different descriptions of the D3-brane: as a gauge
theory and as a gravity solution. Maldacena proposed that the two descriptions are dual to
each other. More generally, it was conjectured in [1] that N = 4 SYM theory with gauge
group SU (N ) is equivalent to string theory in AdS5 S 5 .
In order to make the duality more precise, it is quite convenient to find the relation of the
two parameters on the two sides of the correspondence. First of all, by combining (65) and
(55) we can write the ratio between the AdS radius L and the string length ls in terms of
gauge theory quantities:
!
L
ls

= N gY2 M .

(73)

Notice that the right-hand side of (73) is just the t Hooft coupling = N gY2 M . Therefore,
we can rewrite (73) as:
ls2
1
(74)
= .
2
L

Moreover, by combining (57) and (55) we can write the ten-dimensional Newton constant
G, and the corresponding Planck length lP , in terms of gY M , namely:
G = lP8 =

4 4 8
g
l .
2 YM s

(75)

Then, the relation between the Planck length and the AdS radius is given by
 l 8
P

4
.
2N 2

(76)

The relations (74) and (76) are essential to delimitate the domain of validity of the dual
description in terms of classical gravity. As discussed above, we must require that lP /L  1
if we want to avoid having quantum gravity corrections (this is equivalent to the condition
of having small curvature). From (76) is clear that one should require that N  1. Notice
this conclusion agrees with our general discussion at the end of section 2. Moreover, in order
21

to avoid stringy corrections due to the full tower of massive states of the string, we must
require that ls /L  1 which, in view of (74), means that the t Hooft coupling must be large.
Therefore, we conclude that the gravitational description of N = 4 SYM is reliable if:
N 1,

1.

(77)

Then, the planar (large N ), strongly coupled (large ) SYM theory can be described as
classical gravity. As a first check of the correspondence, let us look at the symmetries on
both sides.

7.1

Conformal symmetry

The N = 4 SYM has an exact vanishing -function and, therefore is a CFT (invariant under
the conformal group, which includes dilatations and special conformal transformations).
On the gravity side, the space AdS5 has the four-dimensional conformal group SO(2, 4) as
isometry group. The SO(2, 4) group acts on the boundary of AdS5 as the conformal group
in four dimensions. In particular, the dilatations act as:
(t, ~x) (t, ~x) ,

z z .

(78)

Therefore, the conformal symmetry is also realized on the gravity side if one identifies the
field theory Minkowski spacetime as the boundary of AdS5 .
It is also interesting to relate the different scales on both theories. From the AdS metric
we see (due to the L2 /z 2 factor in front of the metric) that the proper distance d on the bulk
and the distance dY M on the Minkowski coordinates are related as:
L
dY M .
(79)
z
For the energies this relation is inverted, since the energy is conjugated to the time. Then:
z
E = EY M .
(80)
L
Thus, the high-energy limit (UV) in the field theory (EY M ) corresponds to the nearboundary region z 0. Conversely, as EY M 0 is equivalent to z , the low energy
limit (IR) corresponds to the near-horizon region z .
In a conformal theory, there exist excitations at arbitrary low energies. This corresponds
to the fact that in AdS the geometry extends all the way to z . On the contrary, in a
non-conformal theory there should be a minimal scale and the geometry should end smoothly
at some z0 . This is important when one generalizes the correspondence to non-conformal
theories. There are two prominent examples of this generalized duality . The first case
corresponds to the confining theories with a mass gap m. In this case z0 1/m. The second
case are the theories at finite temperature. Let T denote the temperature. In this case the
geometry is a black hole with a horizon and z0 1/T .
d=

22

7.2

Supersymmetry

Let us now see how one can match the supersymmetry on both sides of the correspondence.
On the field theory side, the N = 4 SYM theory is maximally supersymmetric. It has 32
A
fermionic supercharges generated by 4 sets of complex Majoranas QA
, Q ( A = 1 , 4).
This corresponds to having N = 4 supersymmetry in four dimensions. The QA can be
rotated with the group SU (4), which is the so-called R-symmetry group. Under this group
the supercharges transform as:
A

Q
4 .

QA
4 ,

(81)

In addition, the theory has six scalars 1 , 6 , which transform as the fundamental representation 6 of SO(6). Notice the Lie algebra isomorphism SU (4) SO(6).
The AdS5 S5 space is also a maximally supersymmetric solution of ten-dimensional
supergravity. It has 32 Killing spinors, which correspond to the supercharges of N = 4
SYM. The rotational symmetry of the five-sphere is SO(6), which can be identified with the
R-symmetry of N = 4 SYM. The directions along S5 correspond to the scalar fields on SYM.
Thus, we have a perfect matching with the fields and symmetries of N = 4 SYM. Notice
that the scalar fields of the field theory are extra coordinates on the gravity side and the
isometries of the compact space are interpreted as internal rotations of the scalar fields and
supercharges.

7.3

Reduction on the S5

Let us now discuss further the role of the five-sphere in the AdS dual of N = 4 SYM. We
begin by noticing that any field on AdS5 S5 can be reduced to a tower of fields on AdS5
by expanding it in terms of the harmonics on S5 :
X
(x, ) =
l (x) Yl () ,
(82)
l

with x being coordinates of AdS5 , are coordinates of S5 and Yl () are spherical harmonics
on S5 . The gravity action, after reducing on S5 , becomes:
Z
h
i
1
S =
d5 x Lgrav + Lmatter ,
(83)
16G5
where G5 is the five-dimensional Newton constant, and the gravitational part of the L is:
h
12 i
(84)
Lgrav = g R + 2 ,
L
which corresponds to a negative cosmological constant = 6/L2 . The five-dimensional
Newton constant can be related to the 10d constant by considering the reduction of the
23

Einstein-Hilbert term:
1
16G

L5 5
d x d g10 R10
16G
5

d5 x g5 R5 ,

(85)

where 5 = 3 is the volume of a unit S5 . Then it follows that G5 and G can be related as:
G5 =

G
L5 5

G
3 L5

(86)

Using the value of the ten-dimensional Newton constant G written in (57), we get that G5
is given by:

G5 =
L3 .
(87)
2N 2
From this formula and (27) we can compute the central charge of the SYM theory:
cSY M =

N2
1 L3
=
.
4 G5
2

(88)

Then, the large N limit corresponds to having a large central charge, in agreement with our
previous discussions on the validity domain of the gravitational description.
Let us now consider the reduction (82) of a massless scalar field in AdS5 S5 . The
Klein-Gordon equation in ten dimensions is:
2 = 0 .

(89)

Since the metric factorizes into a AdS5 and S5 part, the DAlembertian is additive
2 = 2AdS5 + 2S5 ,

(90)

where 2S5 is nothing but the quadratic Casimir operator in SO(6). The eigenvalues of 2S5
acting on the spherical harmonics are:
(5)

2S5
(5)

where the Cl

C
Yl () = l 2 Yl () ,
L

(91)

are given by:

(5)

Cl

l = 0, 1, 2, .

= l(l + 4) ,

(92)

Thus, the reduced AdS5 fields l satisfy the massive Klein-Gordon equation
l(l + 4)
.
(93)
L2
Therefore, we have a tower of massive fields l , with a particular set of masses, which
originate, after dimensional Kaluza-Klein (KK) reduction on the five-sphere, from a single
massless scalar field in ten-dimensions. These fields should have a field theory dual in the
N = 4 theory and the mass spectrum (93) should have a counterpart on the field theory
side. We will see that this is indeed the case in the next section.
2AdS5 l = m2l l ,

m2l =

24

A scalar field in AdS

We argued in section 1 that fields in AdS correspond to sources of operators on the field
theory side and that we can learn about these dual operators by analyzing the dynamics of
the sources in the curved space. In this section we study the simplest case of a scalar field
in AdS. Accordingly, let us consider the AdSd+1 space in euclidean signature with metric:
ds2 =

L2
[dz 2 + dx dx ] .
2
z

The action of a scalar field in the AdSd+1 space is:

Z
i
1
h
S =
dd+1 x g g M N M N + m2 2 .
2

(94)

(95)

The equation of motion derived from the action (95) is:



1
MN
g g N m2 = 0 .
M
g
More explicitly, after using the metric (94), this equation becomes :


z d+1 z z 1d z + z 2 m2 L2 = 0 .
Let us perform the Fourier transform of in the x coordinates:
Z
dd k ikx

e fk (z) .
(z, x ) =
(2)d

(96)

(97)

(98)

Then, the equation of motion becomes:

z d+1 z z 1d z fk

k 2 z 2 fk m2 L2 fk = 0 .

(99)

Let us solve (99) near the boundary z = 0. We put fk z for some exponent and keep
the leading terms near z = 0. Then, it is straightforward to find that must satisfy the
following quadratic expression:
( d) m2 L2 = 0 ,

(100)

whose solutions are:

r
d2
d
=
+ m2 L2 .
2
4
Therefore, near z 0 the function fk (z) behaves as :
fk (z) A(k) z d + B(k) z ,
25

(101)

(102)

where is given by:

r

d
= + ,
2

d2
+ m2 L2 .
4

(103)

By performing the inverse Fourier transform we can write the expansion near the boundary
in position space:
(z, x) A(x) z d + B(x) z ,
z0.
(104)
Notice that is real if R, which happens if the mass m satisfies the so-called BreitenlohnerFreedman (BF) bound:
 d 2
m2
.
(105)
2L
This means that m2 can be negative (and the field can be tachyonic) but it must satisfy
the BF bound. In what follows we will suppose that the BF bound is satisfied. Moreover,
notice that:
d

= 2 d 0 ,
(106)
which is obviously satisfied in the mass is above the BF bound. Then, the term behaving as
z d in (104) is the dominant one as z 0. Let us take the boundary as z =  and neglect
the subdominant term. We have:
(z = , x) d A(x) .

(107)

As d is negative if m2 > 0, the leading term is typically divergent as we approach the

boundary at z =  0. In order to identify the QFT source (x) from the boundary value
of the field (z, x) we have to remove the divergences of the latter. We will simply do it
by extracting the divergent multiplicative factor from (107), i.e. the QFT source (x) is
identified with A(x). Equivalently, we define:
(x) = lim z d (z, x) .
z0

(108)

Clearly, with this definition (x) is always finite. Conversely we can write (z, x) =
z d (x) at leading order. In order to interpret the meaning of , let us look at the
boundary action. If O is the operator dual to , this action is given by:
Z

(109)
Sbdy dd x  (, x) O(, x) ,
where  =

 2d
L


is the determinant of the induced metric at the z =  boundary. Then,

by plugging (, x) = d (x) inside Sbdy , we get:

Z
d
dd x (x)  O(, x) .
Sbdy L
26

(110)

In order to make Sbdy finite and independent of  as  0 we should require:

O(, x) =  O(x) .

(111)

But, passing from z = 0 to z =  is a scale transformation in the QFT. Thus, must

be interpreted as the mass scaling dimension of the dual operator O. In other words, the
behavior O(, x) =  O(x) is the wave function renormalization of O as we move into the
bulk. Similarly, from the relation (, x) = d (x) it follows that d is the mass scaling
dimension of the source . Remember from (103) that:
r 
d 2
d
= +
+ m2 L2 .
(112)
2
2
 2
d
, the corresponding scaling dimension is real. We consider three
Then, if m2 2L
different situations, depending on the value of the mass m. The first case we will study is
the one in which m2 > 0. In this case > d and the corresponding operator O is called
an irrelevant operator. By deforming the CFT with O, we have the following term in the
action:
Z
Z
d
d
S = d x (mass)
O = dd x (mass)negative number O .
(113)
When computing
amplitudes
with this action one gets that the effect of the interaction


Energy
depends on mass
, for some > 0. Then, the effects of this interaction can be neglected
for low energies since they are suppressed by powers. Therefore, the interaction goes away
in the IR but changes completely the UV of the theory.
When m2 = 0 we have from (112) that = d and the corresponding operator is called
marginal. Finally, if m2 is negative and takes value if the range (d/2L)2 < m2 < 0, the
dual operator it is called relevant, since it has < d and changes the IR of the theory.

8.1

Application to the KK spectrum in AdS5 S5

We have seen in section 7.3 that a massless scalar in ten-dimensions gives rise to the tower
(93) of massive KK scalars in AdS5 . We will now apply the formula (112) to these fivedimensional scalar fields. First of all, the dimension/mass relation for the case d = 4 is:
p
(114)
= 2 + 4 + (m L)2 .
Let us first consider a massless scalar, which corresponds to the s-wave (l = 0) in (7.3). In
this case (114) with m = 0 gives = 4. Then, the QFT dual operator should be a scalar
operator of dimension 4. Since this s-wave operator is singlet under the SO(6) symmetry of
the S5 , it must not contain the i scalars of the dual N = 4 QFT. The only candidate with
these characteristics is the glueball operator:


O = Tr F F .
(115)
27

Notice that dim[]=dim[A] = 1, so, indeed, dim( O)=4. For higher order KK modes, the
masses are m2 L2 = l(l + 4) (see (93)). Then, the dimensions are:
p
l = 2 + 4 + l(l + 4) = 4 + l .
(116)
In this case, the dual operator should transform under the corresponding representation of
SO(6) (a symmetric tensor with l indices). One can construct such a tensor by multiplying
N = 4 SYM scalar fields i (they transform as vectors of SO(6)). Then, the natural operator
dual to the lth KK mode is:


(117)
Oi1 , ,il = Tr (i1 , ,il ) F F ,
with (i1 , ,il ) being the traceless symmetric product of l scalar fields i of N = 4 SYM.
As dim[] = 1, one can check immediately that the dimension of the operator Oi1 , ,il in
(117) is indeed 4 + l, in agreement with the AdS/CFT result. It has been checked that this
agreement can be extended to all the KK modes of 10d supergravity on AdS5 S5 (including
fermions, forms,...).

8.2

Normalizable and non-normalizable modes

The natural inner product for solutions of the Klein-Gordon equation in a curved space is:
Z

(1 , 2 ) = i
dz dd x g g tt (1 t 2 2 t 1 ) ,
(118)
t

with t being a constant-t slice. Let us consider in particular a field that behaves as z

near z 0. In the AdSd+1 metric g g tt z d+1 . Then, t z 2 as z 0 and the

integrand in the norm of behaves near z = 0 as:

g g tt t z 2d+1 ,

(119)

which leads to a convergent integral if 2 d + 2 > 0.

Let us now consider the two types of modes in (104). We begin with the subleading modes
B(x)z . In this case the exponent is = = d2 + and, therefore, 2 d + 2 = 2 + 2,
which is always positive. This mode is normalizable and can be considered as an element of
the bulk Hilbert space and, given the equivalence with the dual theory, it should be identified
with some state in the boundary theory. Next, we analyze the normalizability of the leading
modes A(x)z d . For these modes = d and, thus, 2 d + 2 = 2(1 ), which is
2
negative if 1. Notice that this condition is equivalent to m2 L2 d4 + 1. Then, the
non-normalizable modes correspond to sources in the boundary theory. Interestingly, for
2
2
masses in the range d4 m2 d4 + 1 both types of modes are normalizable and they
give rise to different Fock spaces of physical states.
28

8.3

Higher spin fields

The results for the scalar field can be generalized to any p-form field, i.e. to any antisymmetric
tensor A1 p with p indices. For a p-form field of mass m, the dimension of the dual
operator is the largest root of the quadratic equation:
( p)( + p d) = m2 L2 .

(120)

This equation can be solved to give the following dimension/mass relation:

r
d 2p 2
d
= +
+ m2 L2 .
2
2

(121)

In particular, for a massive vector field A , the previous formula (121) with p = 1 gives:
r
d
d 2 2
(122)
= +
+ m2 L2 .
2
2
Taking m = 0 in (122), we get = d 1, which is the dimension of a conserved current j
in d dimensions. Finally, one can prove that the dimension/mass relation for a spin 1/2 field
is:
d
+ |m L| .
(123)
=
2

Correlation functions

Let us now see how one can compute correlation functions in Euclidean space from gravity.
The objective is to obtain Euclidean correlation functions of the type:
hO(x1 ) O(xn )i .

(124)

In field theory these correlators can be calculated from a generating function, which is
obtained by perturbing the lagrangian by a source term:
L L + J(x) O(x) L + LJ .

(125)

The generating functional is just:

ZQF T [J] =

Z
E
exp[ LJ ]

The connected correlators are obtained from the functional derivatives of Z:

DY
E Y
log ZQF T [J]
O(xi ) =
J=0
J(x
)
i
i
i
29

(126)

QF T

(127)

Let us consider now any bulk field (z, x) fluctuating in AdS. Let 0 (x) be the boundary
value of :
0 (x) = (z = 0, x) = |AdS (x) .
(128)
The field 0 is related to a source for some dual operator O in the QFT. As we know the
actual source is not the value of at z = 0, which is typically divergent, but the limit:
lim z d (z, x) = (x) .

(129)

z0

Then, the AdS/CFT prescription for the generating functional is [2, 3]:
Z
D

E
ZQF T [0 ] = exp
0 O
= Zgravity [ 0 ] ,

(130)

QF T

where Zgravity [ 0 ] is the partition function (i.e. the path integral) in the gravity theory
evaluated over all functions which have the value 0 at the boundary of AdS:
X
Zgravity [ 0 ] =
eSgravity .
(131)
{0 }

In the limit in which classical gravity dominates, one can substitute the sum by the term
corresponding to the classical solution. In this case the generating function becomes:
onshell

ZQF T [0 ] eSgravity

[0 ]

(132)

One should be careful when evaluating the on-shell gravity action because it typically diverges
and has to be renormalized following the procedure of holographic renormalization [19, 20]
(see below and the review [21]). Thus, the classical action must be substituted by a renorren
and the generating functional becomes:
malized version, which will be denoted by Sgrav
ren
log ZQF T = Sgrav
[ 0 ] .

(133)

Moreover, the n-point function can be obtained by computing the derivatives with respect
to = z d :

ren
(n) Sgrav
[] 
hO(x1 ) O(xn )i =
.
(134)

(x1 ) (xn ) 
=0

9.1

One-point function

It is also interesting to compute the one-point function of an operator O in the presence of

the source :
ren
Sgrav
[]
hO(x)i =
.
(135)
(x)
30

Taking into account the relation between and (eq. (129)), we get:
hO(x)i = lim z

z0

ren
Sgrav
[]
.
(z, x)

(136)

The functional derivative of the classical on-shell action can be computed in closed form.
Indeed, let Sgrav be represented as:
Z Z
Sgrav =
dz dd xL[, ] ,
(137)
M

with M being a (d + 1)-dimensional manifold whose boundary is located at z = 0. Under a

general change + , the classical action Sgrav varies as:
Z Z
h L
i
L
dz dd x
Sgrav =
+
( ) .
(138)

( )
M
Let us now use in (138) that [ ] = () and let us integrate by parts. We get:
Z Z
 L 
 L
i
h  L

+

.
Sgrav =
dz dd x

( )
( )
M

(139)

The first term in the previous equation vanishes on-shell due to the Euler-Lagrange equations.
As the boundary is at z =  0, we can write:
Z Z
Z

 L

L

onshell
d
=
d x z
=
dd x

.
(140)
Sgrav
(z )
(z ) z=

M
Let us next define as:
=

L
,
(z )

(141)

which is the canonical momentum if z is taken as time. Then (140) can be rewritten as:
Z
onshell
Sgrav
=
dd x (, x) (, x) .
(142)
M

Thus, it follows that:

onshell
Sgrav
L
= (, x) =
.
(, x)
(z )

(143)

In general the renormalized action can be written as:

onshell
+ Sct ,
S ren = Sgrav

(144)

where Sct is the action of couterterms, defined at the boundary z = . Let us define the
renormalized momentum as:
S ren
ren (z, x) =
.
(145)
(z, x)
31

Clearly, by taking z =  in (145) and (144) we have:

ren (, x) =

L
Sct
+
.
(z (, x)) (, x)

(146)

Therefore, from the definition of ren in (145) and eq. (136), we obtain the one-point function
of O in presence of the source as the following limit in the AdS boundary:
hO(x)i = lim z d ren (z, x) .
z0

9.2

(147)

Linear response theory

The field theory path integral representation of the one-point function with a source is:
Z
R d
hO(x)i = [D] O(x) eSE []+ d y (y)O(y) ,
(148)
where denotes the fields of the QFT. Let us expand the exponent of this expression in a
power series of the source and let us keep the terms up to linear order:
Z
hO(x)i = hO(x)i=0 +
dd y hO(x) O(y)i (y) + .
(149)
Next, we define the euclidean two-point function GE (x y) as:
GE (x y) = hO(x) O(y)i .

(150)

Then, equation (149) can be rewritten as:

Z
hO(x)i = hO(x)i=0 +

dd y GE (x y) (y) .

(151)

We will consider normal-ordered observables such that hO(x)i=0 vanishes. Notice that this
always can be achieved by subtracting to O its vacuum expectation value (VEV) without
source. Then, hO(x)i measures the fluctuations of the observable away from the expectation
value, i.e. the linear response of the system to the external perturbation and we can write:
Z
dd y GE (x y) (y) .
(152)
hO(x)i =
In momentum space this expression can be written as:
hO(k)i = GE (k) (k) ,

(153)

and, thus, we can obtain the two-point function in momentum space by dividing the onepoint function by the source:
hO(k)i
.
(154)
GE (k) =
(k)
32

In the framework of the AdS/CFT correspondence, since = z d (z, x) with z 0, we

have the following formula for the two-point function in momentum space:
GE (k) = lim z 2(d)
z0

9.3

ren (z, k)
.
(z, k)

(155)

Two-point function for a scalar field

Let us apply the developments of the previous sections to the case in which the source is
a scalar field in Euclidean AdSd+1 . We will assume that the action of is:
Z
h
i

d
MN
2 2
S =
dz d x g g M N + m ,
(156)
2
where is a normalization constant. In order to evaluate the on-shell action of we rewrite
S as follows:
Z
Z
h
i


i

h 1
d
MN
d
MN
2
dz d xM g g N +
dz d x g M
gg N m .
S =
2
2
g
(157)
The second term in (157) vanishes when the equation of motion of is used. Then, the
on-shell action is:
Z
h
i

onshell
dz dd x M g g M N N .
(158)
S
=
2
Taking into account that the boundary is at z = , which is the lower limit of the integration,
the on-shell action (158) can be written as:
Z



onshell
dd x g g zz z
S
=
.
(159)
2
z=
Let us define the canonical momentum as in (141):

= g g zz z .
(z )

(160)

Then the on-shell action of becomes:

S

onshell

1
=
2

dd x (z, x) (z, x) .

(161)

z=

Let us work in momentum space and Fourier transform the field and the canonical momentum :
Z
Z
dd k ikx
dd k ikx
(z, x) =
e fk (z) ,
(z, x) =
e k (z)
(162)
(2)d
(2)d
33

Then, Parseval identity allows to rewrite S onshell as:

Z
dd k
1
onshell
S
=
k (z = ) fk (z = ) .
2
(2)d

(163)

Recalling that the function fk (z) behaves near z 0 as in (102) and the definition of , we
obtain:
h
i
(z, x) Ld1 (d ) A(x) z + B(x) z d ,
(z 0) ,
(164)
which, in momentum space becomes:
h
i
k (z) Ld1 (d ) A(k) z + B(k) z d ,

(z 0) ,

(165)

Let us use these results to compute the on-shell action, keeping the terms that do not vanish
when  0. We get:
Z
i
d1
dd k h 2
onshell

(d ) A(k) A(k) + d A(k) B(k) .
(166)
S
= L
2
(2)d
Notice that the first term in (166) is divergent. We will now find a counterterm to renormalize
this divergence of the on-shell action. It must be a local quadratic functional defined at the
boundary of AdS. The natural candidate would be a term proportional to:
Z

dd x 2 (, x) ,
(167)
AdS

where is the determinant of the induced metric on the boundary:

ds2z= = dx dx =
It is easy to prove that:
Z
Z
d
2
d
d x (, x) = L
AdS

L2
dx dx .
2

i
dd k h 2

A(k)
A(k)
+
2
A(k)
B(k)
,
(2)d

(168)

(169)

Let us adjust the coefficient of the counterterm action in such a way that the leading divergence is cancelled. It is immediate to check that this counterterm action must be:
Z
d

Sct =
dd x 2 ,
(170)
2 L
AdS
or, equivalently, in momentum space:
Z
i

dd k h 2
d1
Sct = (d ) L

A(k)
A(k)
+
2
A(k)
B(k)
.
2
(2)d
34

(171)

The renormalized action S ren = S onshell + Sct can be obtained by adding (166) and (171),
with the result:
Z
dd k
d1
ren
S
= L (2 d)
A(k) B(k) .
(172)
2
(2)d
In order to extract the one-point function from S ren we have to compute the functional
derivative with respect to (x)(recall that (x) = A(x)). However, the coefficient B(x) also
depends functionally on A(x). To illustrate this point, let us represent fk (z) for arbitrary z
(not necessarily small) as:
fk (z) = A(k) 1 (z, k) + B(k) 2 (z, k) ,

(173)

where 1 (z, k) and 2 (z, k) are independent solutions of the equation satisfied by fk (z),
normalized in such a way that for z 0 they behave as:
1 (z, k) z d ,

2 (z, k) z .

(174)

(the explicit expressions of 1 (z, k) and 2 (z, k) are given below). To determine completely
we have to impose regularity conditions in the deep IR z . As we will see soon, this
fixes uniquely the ratio B/A to a value which is independent of the value of the field at the
boundary z = 0. Let us denote this ratio by :
=

B
.
A

Clearly, we can write the renormalized action (172) as:

Z
d1
dd k
ren
S
= L (2 d)
(k)(k) (k) ,
2
(2)d
where we have used the fact that (k) = A(k). Then, as the functional derivative

equivalent to (2)d (k)

in momentum space, we have:
hO(k)i = (2)d

S ren
= Ld1 (2 d) (k)(k) .
(k)

(175)

(176)

(x)

is

(177)

Taking into account the definition of in (175) and that 2 d = 2, we can write (177)
as:
hO(k)i = 2 Ld1 B(k) .
(178)
Thus, the subleading contribution near the boundary (B(k)) determines the VEV of the
operator. The two-point function GE (k) can be obtained by dividing by the source (see
(154)):
B(k)
GE (k) = 2 Ld1
.
(179)
A(k)
35

Let us now calculate explicitly A(k) and B(k). We first define the function gk (z) as:
d

fk (z) = z 2 gk (z) .

(180)

Then, one can easily check from (97) that gk (z) satisfies the equation:
z 2 z2 gk + z z gk ( 2 + k 2 z 2 )gk = 0 .

(181)

Eq. (181) is just the modified Bessel equation, whose two independent solutions can be taken
to be gk = I (kz), where I are modified Bessel functions. Thus, the two independent
solutions for fk (z) are:
d
(182)
z 2 I (kz) .
Notice that for z 0 the modified Bessel functions behave as:
 z 
1
,
(z 0) .
I (z)
(1 ) 2

(183)

Then, we can take 1 (z, k) and 2 (z, k) in (173) as:

1 (z, k) = (1)

 k 
2

z 2 I (kz) ,

2 (z, k) = (1+)

 k 
2

z 2 I (kz) . (184)

One can check that these functions have, indeed, the correct behavior when z 0. Then,
by using (173) we get:
h
 k 
 k 
i
fk (z) = z (1 )
A(k) I (kz) + (1 + )
B(k) I (kz) .
2
2
d
2

(185)

Let us now impose that fk (z) is finite when z . This condition determines a precise
relation between the coefficients A(k) and B(k), as already mentioned above. When z ,
the functions I (z) behave as:
ez
,
I (z)
2z

(z ) .

(186)

Then, after changing z kz, we get for large z:

d
 k 
 k 
i
z 2 ekz h
fk (z)
(1 )
A(k) + (1 + )
B(k) ,
2
2
2kz

(187)

which diverges when z unless the coefficient in brackets vanishes. Then, we must
require:
B(k)
(1 )  k 2
()  k 2
=
=
.
(188)
A(k)
(1 + ) 2
() 2

36

Using this result we can compute the euclidean two-point function. We get:
GE (k) = 2 L

d1

()  k 2
.
() 2

(189)

Let us write this result in position space. The relation between GE (x) and GE (k) is:
Z
dd k ikx
GE (x) =
e GE (k) .
(190)
(2)d
We now use the formula:
Z

1
dd k ikx n
2n d+n
2

e
k
=
,
d
n
(2)d
2 2 |x|d+n

(191)

to obtain the correlator in momentum space:

1
2Ld1 d2 +


.
hO(x)O(0)i =
d
|x|2
2

(192)

The behavior hO(x)O(0)i |x|2 in (192) confirms that is, indeed, the scaling dimension
of the operator O(x).

10

Quark-antiquark potential

Let us consider an external charge moving along a closed curve C in spacetime in QED. The
action for such a charge is:
I
SC =
A dx .
(193)
C

Adding this term to the action is equivalent to insert in the path integral the quantity:
eiSC = ei

H
C

A dx

W (C) ,

(194)

where W (C) is the so-called Wilson loop, which is just the holonomy of the gauge field A
along the closed curve C. Notice that W (C) is the Aharonov-Bohm phase factor due to the
propagation of a quark along the closed curve C. The non-abelian analogue of the previous
formula is:
h I
i
W (C) = Tr P exp i
A dx ,
(195)
C

Aa

where A =
T , Tr is the trace over the group indices and P is the path ordering
operator. Notice that hW (C)i can be regarded as the amplitude for a creation of a q q pair
which propagates and is annihilated afterwards.
37

In particular, we can take a rectangular Wilson loop in Euclidean space. Let T and d be
the sizes of the rectangle. In this case the Wilson loop is the amplitude for the propagation
of a q q pair separated a distance d, which can be calculated by means of the hamiltonian
formalism. Actually, when T one has:
lim hW (C)i eT E(d) ,

(196)

where E(d) is the energy of a q q pair separated a distance d. In a confining theory the energy
grows linearly with the q q distance d:
E(d) d ,

constant .

(197)

Then, the Wilson loop in a confining theory satisfies the so-called area law:
lim hW (C)i e T d e (area enclosed by the loop) .

(198)

Let us see how we can compute VEVs of Wilson loops in the AdS/CFT correspondence
[22, 23]. Recall that the endpoint of an open string ending on a D-brane is dual to a quark.
Thus, in string theory, the Wilson loop is represented by an open string whose worldsheet
has a boundary which is in a D-brane and describes the curve C, as shown in figure 10.
Then:


hW (C)i = Zstring = C .
(199)
To have infinitely heavy (non-dynamical) quarks we push the D-brane to the AdS boundary
and, therefore, = C lies within the AdS boundary z = 0. Then, in the t Hooft limit, we
have:


Zstring = C = eS(C) ,
(200)
where S(C) is the on-shell extremal Nambu-Goto action for the string worldsheet satisfying
the boundary condition that ends on the curve C. The Wilson loop becomes simply:
hW (C)i = eS(C) .

(201)

D-brane

Figure 10: A Wilson loop along a closed curve C is holographically realized as the boundary
of a surface .
38

We will now apply these ideas to the holographic calculation of the rectangular Wilson
loop and the corresponding q q potential energy in N = 4 SYM. With this purpose, let us
consider a string ending on the boundary of AdS5 , where it extends along one of the gauge
theory coordinates (say x1 x). We shall parameterize the embedding of the string in AdS5
by a function z = z(x) and we will denote by z 0 the derivative of z with respect to x. The
induced metric in Euclidean signature on the string worldsheet is:

L2  2
(202)
dt + (1 + z 0 2 ) dx2 .
2
z
Therefore, the Nambu-Goto action for the string is:

Z
Z
Z
1
1 + z0 2
T L2

S =
g
=
dt
dx
dx
,
(203)
20
20
z2
R
where T = dt. The action (203) corresponds to a problem in classical mechanics where x
is the time and the lagrangian is given by:

T L2 1 + z 0 2
.
(204)
L =
20
z2
Since this lagrangian does not depend on the time x , the energy is conserved, which
means that:
L
(205)
z 0 0 L = constant .
z
By computing the derivative of L appearing on the left-hand side of (205), the conservation
law written above can be proved to be equivalent to the following first integral of the equation
of motion:

z 2 1 + z 0 2 = constant .
(206)
ds2 =

For the hanging string configuration we are interested in, the boundary conditions that the
function z(x) must satisfy are:




z x = d/2 = z x = d/2 = 0 ,
(207)
where d is the quark-antiquark separation. Clearly, there is a value of x for which z is
maximal. By symmetry this value is just x = 0 and, since z(x) has a maximum implies that
z 0 (x = 0) = 0. Let z be the maximal value of z, i.e. z = z(x = 0). Then, the constant
appearing in the conservation law is just z2 and we have:
p
z4 z 4
0
,
(208)
z =
z2
where the two signs correspond to the two sides of the hanging string. The equation written
above can be integrated immediately to give x as a function of z:
Z z
2
p
x =
d .
(209)
z4 4
z
39

By introducing a new rescaled variable y by means of the relation = z y, one can write:
Z z
z
y2
p
x = z
dy .
(210)
1 y4
1
By imposing the boundary conditions (207), we can obtain the quark-antiquark separation
as a function of z :
Z 1
d
y2
p
= z
dy .
(211)
2
1 y4
0
The integral over y in (211) can be computed exactly and one gets:
  1 2
d
z = 3
.
4
2 22

(212)

Then, it follows that d and z are proportional. Let us now compute the on-shell action
for the string, from which we will obtain the quark-antiquark potential. By plugging the
conservation law (206) inside the expression of the action in (203), we get:
Z
dx
T L2 z2
.
(213)
S =
0
2
z4
Let us change variables in the integral (213) from x to z. The jacobian for this change of
variables is:
dx
1
z2
= 0 = p
,
(214)
dz
z
z4 z 4
where we have taken into account the value of z 0 written in (208). Taking into account that
the variable z is double-valued, the total action is:
Z
T L2 z2 z
dz
p
S = 2
.
(215)
0
2  z 2 z4 z 4
Let us next work on a rescaled variable y, related to z as z = z y. Then, the on-shell action
becomes:
T L2
I ,
(216)
S =
0 z
where I is the following integral:
Z 1
dy
p
I =
.
(217)
2
1 y4
/z y
The integral I in (217) diverges when  0. Indeed, one can prove that, for small :
3
2 2
z
.
(218)
I =   2 +

1
4
40

The on-shell action S is just T E, where E is the energy of the quark-antiquark. It follows
that E is given by:
L2 1
4 2 L2
1
+
.
(219)
E =   4
d
0 
14
0
The divergent term in (219) corresponds to the quark and antiquark masses (which are
considered to be infinitely large in the static limit). To check this, let us compute the
euclidean action of a single string which goes straight from the boundary z =  to z = at
fixed x. This configuration can be described by using t and z as worldvolume coordinates.
The induced metric is obtained by keeping constant the x coordinate and is given by:
ds2 =

L2
(dt2 + dz 2 ) .
z2

(220)

The on-shell Nambu-Goto action for this configuration is:

Z 2
T L2
L
T
2
dz
=
.
S| =
20  z 2
20 

(221)

Therefore, the energy for a configuration of two straight parallel strings in AdS5 is:
E| | = 2

L2
,
20 

(222)

which is equal to the divergent term in (219), as claimed. The quark-antiquark potential Vqq
is obtained by subtracting the divergent contribution due to the quark masses:
Vqq = E E| | .

(223)

By subtracting (219) and (222), one gets:

4 2 L2
1
Vqq =   4
.
d
41
0
Let us write this result
in terms of gauge theory quantities. Recall that L2 =

or, equivalently L2 = 0 . Then, (224) can be rewritten as:

4 2 1
Vqq =   4
.
d
14

(224)

N 2 gY M 0

(225)

The Coulombic 1/d dependence in (225) is consequence of the conformal invariance of the
theory. The non-analytic dependence on the coupling is a non-perturbative effect. In field
theory this non-analyticity results from the summation of an infinite number of Feynman
41

diagrams. It is interesting to compare (225) with the perturbative q q potential, valid for
small , which is given by:

.
(226)
Vqq =
d
Remarkably we have been able to find a non-perturbative result in a interacting quantum
field theory by studying the catenary curve for a hanging string in classical gravity.
In the next two sections we will generalize the result (225) to the case of N = 4 SYM at
finite temperature and to the gravity dual of a theory in which the quarks are confined. In
the former case we have to study a string hanging from the boundary of an anti-de-Sitter
black hole, whereas in the latter we must deal with a geometry obtained from the AdS black
hole by a double analytic continuation.

10.1

Quark-antiquark potential at finite temperature

Let us suppose that we put our gauge theory in a thermal bath at a temperature T . In the
holographic correspondence the heat bath is dual to a bulk geometry with an event horizon
(a black hole) and the Hawking temperature of the black hole is identified as the temperature
of heat bath. In particular, the dual of N = 4 SYM at finite temperature is a black hole in
AdS5 , whose metric in euclidean space is (see below):
ds2 =

L2 h
dz 2 i
2
2
f
(z)
dt
+
d~
x
+
,
z2
f (z)

(227)

with f (z) being the following function:

f (z) = 1

z4
.
z04

(228)

The parameter z0 corresponds to the position of the horizon and is related to the black hole
temperature T as T = (z0 )1 (see section 11.2 below). In the context of the holographic
duality one naturally identifies the black hole temperature T with the temperature of the
gauge theory.
As in the zero temperature case, we will parametrize the embedding of the string by a
function z = z(x). The induced metric is:
ds2 =


L2 h
z0 2  2 i
2
f
dt
+
1
+
dx .
z2
f

The Nambu-Goto action for this embedding becomes:

p
Z
f (z) + z 0 2
L2
dx
,
S =
20
z2
42

(229)

(230)

R
where = dt. Following the same steps as in the zero temperature case, it is straightforward to derive the following first integral of the equation of motion:
p
z 2 f (z) + z 0 2
z2
= constant = p
.
(231)
f (z)
f (z )
From (231) we obtain readily the value of z 0 :
s
p
z4 z 4
f (z)
0
z =
.
f (z )
z2

(232)

Let us now define the constant as follows:

Then, we have:
Z
p
x = z 1

 z 4
0

z
z
z

y 2 dy
p
,
(1 y 4 )( y 4 )

and the quark-antiquark distance d is equal to:

Z 1
p
y 2 dy
p
d = 2z 1
.
(1 y 4 )( y 4 )
0

(233)

(234)

(235)

Notice that these expressions become the ones at zero temperature as , as it should.
Moreover, as z z0 , 1 and the distance d 0 (see figure 11). Actually, by varying
one can see that there is a maximal value of d (dmax z0 ) and the q q bound state becomes
unbound due to thermal screening, in agreement with the behavior expected from the point
of view of the gauge theory at non-zero temperature.

10.2

Quark-antiquark potential in a confining background

Let us consider the AdS black hole in Euclidean signature and let us go back to Minkowski
signature by analytic continuation along a different direction, namely by making x3 it. If
we call u to the original euclidean time, we arrive at the following metric:
L2 h
dz 2 i
2
2
2
2
ds = 2 dt + dx1 + dx2 + f (z)du +
.
z
f (z)
2

(236)

In this case the space ends smoothly at z = z0 , which should be though as a IR mass scale. We
will now compute the q q potential for the metric (236) and we will verify that it corresponds
to a confining background. With this purpose, let us consider a fundamental string in the
43

First integral

f (z) + z ! 2
z
= constant = !
f (z)
f (z )

"
!
d = 2z 1

q q distance

d
z

y 2 dy
!
(1 y 4 )( y 4 )

!z "
0

0.8

0.6

0.4

0.2

10

Horizon

z z0
d0
1
Figure 11: In a black hole metric, the string ends on the horizon when the q q separation is
Thered isreaches
a maximal
value of value,
d (dmax
z0 )plot on the
large enough. Correspondingly, the distance
a maximum
as
the
right shows.
At high T disconnected configuration energetically favor

Models
screeningcoordinates.
in a plasma!The
euclidean version of the above metric and let
us use thermal
(t, x) as worldsheet
induced metric is:

z0 2  2 i
L2 h
dx ,
(237)
ds2 = 2 dt2 + 1 +
z
f
and the Nambu-Goto action becomes:
2

S =
where =

L
20

dx
z2

s
1+

z0 2
,
f (z)

(238)

dt. The first integral corresponding to the action (238) is:

z2 p
p
f (z) + z 0 2 = z2 ,
f (z)

(239)

p
z4 z 4
.
z2

(240)

from which we get:

p
z 0 = f (z)
This equation can be integrated as:

x = z

Z
1

z
z

y 2 dy
p
,
(1 y 4 )( y 4 )

where the constant is the same as in (233). It follows that the q q distance d is:
Z 1
y 2 dy

p
d = 2z
.
(1 y 4 )( y 4 )
0
44

(241)

(242)

In this case, d grows without limit as 1, or equivalently as the turning point approaches
the end of the space (i.e. when z z0 ) (see figure 12). In the large d limit the profile of
the hanging string is approximately rectangular. The energy due to the vertical parts of the
profile can be identified with the masses of the static quarks, which have to be subtracted
to get the potential energy. The q q potential is just due to the horizontal part of the profile.
Since in this part z is approximately constant and equal to z0 , we get that the contribution
to the euclidean action is:
L2 d
,
(243)
Shorizontal =
20 z02
which corresponds to an area law (it is proportional to d) and gives rise to a confining
potential of the type:
V = s d ,
(244)
with s being the effective string tension, given by:
s =

L2 1
.
20 z02

In terms of gauge theory quantities, s can be written as:

.
s =
2z02

(245)

(246)

Notice that M 1/z0 is an IR scale that can be identified with the mass gap of the theory
(and z0 with the glueball size). The previous formula for the tension is simply:

s M 2 .
(247)

11

Black hole thermodynamics

The partition function in statistical mechanics in the canonical ensemble is given by:
H

Z = Tr e T ,

(248)

where, H is the hamiltonian operator, T is the temperature and we are taking the Boltzmann
constant kB = 1. The thermal average of an operator O at the temperature T is:

H
Tr O e T
hOiT =
.
(249)
Z
In the path integral approach, the average hOiT can be written as:
Z
H
hOiT [D] h(x), t |O e T |(x), ti ,
45

(250)

d = 2z

d
z

y 2 dy
"
(1 y 4 )( y 4 )

! z "4
0

3.0

2.5

2.0

1.5
Z=0

1.1

1.2

1.3

1.4

1.5

Z=Z

Figure 12: In a confining metricNo

the maximal
space endsdistance
smoothly and the profile of the hanging string
becomes almost rectangular for large q q separation (left). The distance d is not bounded, as
shown on the plot on the right.The distance diverges as we approach the end of the

space

where the expectation value is taken between the same initial

and final state |(x), ti (as
z = 0.
it corresponds to a trace in Hilbert space). Equivalently, since the hamiltonian operator
Qualitative
picture(250) as:
implements time evolution,
we can rewrite
Z
z = z0
i
(251)
hOiT [D] h(x), t |O|(x), t + i .
T

When
d wethe
profile
is almost
rectangular
Then, to perform thermal
averages
have
to consider
imaginary
time evolution and we
have to impose periodic boundary
in the Hilbert
for fermions).
Verticalconditions
parts masses
of the space
static (antiperiodic
quarks
Then, the euclidean time tE must be periodically identified
Horizontal part q q potential
1
tE tE + .
(252)
T
2
L d
horizontal =time ! is 2equivalent
Area
law T 6=
Confinement
Thus, the compactification ofSEuclidean
to having
0.
2 z0
Let us apply these ideas to obtain the Hawking temperature of a black hole. We will

assume that we have an euclidean metric of the type:

L2 1

V = s d
s =
s =
! z2
2


1
0
2z 2
ds2 = g(r) f (r) dt2E + d~x 2 +
dr2 ,
(253) 0
h(r)

effective string tension

1
where the functions f (r) and h(r) have a M
first-order
zero at r = r0 , which is the location of

mass gap

z
0
the horizon, and g(r0 ) 6= 0. Then, for r r0
s M 2

glueball size
f (r) fz00(r
0 ) (r r0 ) ,

h(r) h0 (r0 ) (r r0 ) ,

(254)

while we can take g(r) = g(r0 ). Then, the near-horizon euclidean metric can be written as:


ds2 g(r0 ) f 0 (r0 ) (r r0 )dt2E + d~x 2 +
46

1
dr2
.
h0 (r0 ) r r0

(255)

Let us define a new radial variable such that:

1
dr2
= d2 ,
h0 (r0 ) r r0

(256)

which can be integrated to give the following relation between and r:

r
r r0
.
= 2
h0 (r0 )

(257)

Next, we define an angular coordinate such that

g(r0 ) f 0 (r0 ) (r r0 )dt2E = 2 d2 ,

(258)

which is equivalent to defining as:

=

1p
g(r0 ) f 0 (r0 ) h0 (r0 ) tE .
2

(259)

In the new variables, the (tE , r) part of the metric takes the form d2 +2 d2 , which is locally
like the flat metric of a plane. In order to have = 0 (i.e. the horizon) as a regular point
without any curvature singularity, the variable must be a periodic variable with period
2. Otherwise we would have a conical singularity at the origin, due to the defect angle.
We have argued above for a general system that the compactification of the euclidean time
is equivalent to having a non-zero temperature T . Thus, it follows that we can assign a
temperature T to a black hole (the Hawking temperature). In order to find the value of
T , let us notice that the periodicity under + 2 is equivalent to periodicity under
tE tE + T1 , where T is the Hawking temperature given by:
1
4
= p
.
T
g(r0 ) f 0 (r0 ) h0 (r0 )

(260)

In the next subsections we will apply this formula to determine the Hawking temperature of
a couple of black holes.

11.1

Application to the Schwarzschild black hole

Let us consider the ordinary Schwarzschild metric in four dimensions:


2GM  2
dr2
2
dt +
+ r2 d22 ,
ds = 1
2GM
r
1 r

(261)

which is a particular case of the general expression written in (253), with the functions f , h
and g being given by:
g(r) = 1 ,

f (r) = h(r) = 1
47

2GM
.
r

(262)

In (261) G is the four-dimensional Newton constant and M is the mass of the black hole.
The horizon in the geometry (261) is located at r = r0 = 2GM . Since:
f 0 (r0 ) = h0 (r0 ) =

2GM
r02

(263)

it follows that the Hawking temperature for the Schwarzschild black hole is given by:
T =

1
.
8GM

(264)

Let us use the expression of T in (264) to obtain the black hole entropy S of the Schwarzschild
black hole. We will identify the mass M with the internal energy and we will make use of
the first law of thermodynamics:
dM = T dS =

1
dS ,
8GM

(265)

which can be integrated to give the entropy S:

S = 4 G M 2 ,

(266)

The horizon area AH for the metric (261) is the area of the surface t = constant, r = r0 . It
is straightforward to demonstrate that AH is:
AH = 4r02 = 16 G2 M 2 .

(267)

It is now immediate to check that the relation between S and AH is:

S =

AH
,
4G

(268)

which is noting but the celebrated Bekenstein-Hawking entropy formula which relates the
entropy of a black hole with the area of its horizon (see (6)).

11.2

AdS black hole

The metric of a black hole in AdSd+1 is:

ds2 =

L2 h
dz 2 i
2
2
f
(z)
dt
+
d~
x
+
,
E
z2
f (z)

(269)

where f (z) is the following function:

f (z) = 1
48

zd
,
z0d

(270)

with z0 being a constant (z = z0 is the location of the horizon). This metric is just of our
general form (253) with r z, f given as above and g and h being:
g=

L2
,
z2

h =

z2
f .
L2

(271)

In this case the derivatives of f and h at the horizon are:

f 0 (z0 ) =

d
,
z0

h0 (z0 ) =

dz0
,
L2

(272)

and, therefore, we have:

g(z0 ) f 0 (z0 ) h0 (z0 ) =

d2
.
z02

(273)

Thus, the Hawking temperature is related to z0 by means of the relation:

T =

d
.
4z0

(274)

The horizon is the hypersurface z = z0 and t constant, whose area in the metric (269) is:
AH =

 L d1
z0

Vd1 ,

(275)

where Vd1 is the volume along the Minkowski spacial directions. In terms of the temperature,
AH can be written as:
 4 d1
AH =
Ld1 T d1 Vd1 ,
(276)
d
and the entropy can be computed from the Bekenstein-Hawking formula, with the result:
AH
1  4 d1 d1 d1
=
L T
Vd1 .
(277)
S =
4Gd+1
4Gd+1 d
We now define the entropy density s as:
s =

S
Vd1

(278)

Let us write s in terms of the QFT central charge cQF T = 14 (L/lP )d1 . We get:
s =

 4 d1
d

cQF T T d1 .

(279)

In the case of N = 4 SYM, we take d = 4 and cSY M = N 2 /2 and the entropy density is
given by:
2 2 3
sSY M =
N T .
(280)
2
49

From the expression of the entropy density in (280) we can obtain the value of the pressure
by means of the thermodynamic relation:
s =

p
.
T

(281)

We get:
2 2 4
N T .
(282)
8
Moreover, the energy density  can be obtained from p and s by means of the standard
thermodynamic relation:
 = p + T s .
(283)
p =

Using the values of p and s computed from holography, we arrive at the following value of
the energy density:
3 2 2 4
 =
N T .
(284)
8

11.3

Comparison with field theory

Let us compare the strong coupling values of s, p and  found above with those corresponding
the N = 4 SYM at zero coupling. The partition function in the canonical ensemble for a
gas of non-interacting relativistic bosons and fermions is
Z


(p)
d3 p
T
log Z = V3
,
(285)
log
1

e
(2)3
where the
p upper minus (lower plus) signs corresponds to bosons (respectively, fermions) and
(p) = p~ 2 + m2 . In the massless case, we just take (p) = |~p| and we get:
Z


dp 2
log Z
Tp
=
p log 1 e
.
(286)
V3
2 2
0
Let us change variables in the integral (286) from p to x = p/T . After integrating by parts
we find:
Z
T3
x3
log Z
=
dx
.
(287)
V3
6 2 0
ex 1
To calculate these integrals we use the general results, valid for n Z:
Z
Z
x2n1
22n1 1 2n
x2n1
(2)2n Bn
dx x
=
Bn ,
dx x
=
,
(288)
e +1
2n
e 1
4n
0
0
where Bn denotes the Bernouilli numbers. In particular for n = 2, since B2 = 1/30, we
have:
Z
Z
x3
4
x3
7 4
dx x
=
,
dx x
=
.
(289)
e 1
15
e +1
120
0
0
50

Thus, for bosons:

log Z
2 3
=
T ,
V3
90

(bosons) ,

(290)

while for fermions:

7 2 3
log Z
=
T ,
(fermions) .
(291)
V3
720
The entropy density can be obtained from the partition function by means of the standard
statistical mechanics relation:
h log Z i
log Z
T
= 4
.
(292)
s =
T
V3
V3
Then, it follows that:
2 2 3
7 2 3
T ,
sf ermion =
T .
45
180
In N = 4 SYM the number of bosons we have is:


2(gauge field) + 6(scalar field) N 2 = 8N 2 ,
sboson =

while the number of fermions is:



2 4(Weyl spinors) N 2 = 8N 2 ,

(293)

(294)

(295)

which, due to the supersymmetry, is the same as the number of bosons. Therefore, the total
entropy density of N = 4 SYM when the coupling constant is zero is:
h 2
7 2 i 3
2 2 2 3
2 2
+
T =
N T .
(296)
sN =4 f ree gas = 8 N
45
180
3
By using the thermodynamic formulas (283) and (281), we obtain the values of the pressure
and energy density:
2 2 4
2 2 4
N T ,
N =4 f ree gas =
N T .
6
2
Then, by comparing with the result given by the black hole, we obtain:
pN =4 f ree gas =

3
sN =4 f ree gas ,
4
and similarly for the pressure and energy density:
sblack hole =

(297)

(298)

3
3
pN =4 f ree gas ,
black hole = N =4 f ree gas .
(299)
4
4
Therefore, the AdS/CFT correspondence predicts that the values of s, p and  at infinite
coupling differ from their values at zero coupling by a multiplicative factor 3/4. These
reductions of the entropy, pressure and energy density as the coupling is increased are in
very good agreement with the results obtained in lattice simulations for different gauge
theories.
pblack hole =

51

12

Transport coefficients

Let us begin by writing the linear response formulas of section 9.2 in real time. We consider
a system in QFT to which we couple a source (x) to a local operator O(x):
Z
S = S0 +
dd x O(x) (x) ,
(300)
where the d-dimensional spacetime has Minkowski signature. We will assume that the unperturbed VEV of the operator O vanishes. Then, the one-point function of the operator O
in the presence of the source is given by:
Z
hO(x)i = GR (x y) (y) dy ,
(301)
where GR (x y) is the retarded Greens function, defined as:


i GR (x y) (x0 y 0 ) h O(x), O(y) i .

(302)

The fact that the linear response is determined by the retarded correlator is a consequence
of causality since the source can influence the system only after it has been turned on. In
momentum space the relation (301) between the one-point function and the source becomes:
hO(, ~k)i = GR ( , ~k ) ( , ~k ) .

(303)

We are interested in analyzing the long wavelength hydrodynamic limit. In this case one
can take the zero spatial momentum and zero frequency limit of the retarded correlator. In
this limit one studies the response of the system to a time varying source (t), which can be
approximately represented as:
hOi t ,
(304)
where the real constant is the so-called transport coefficient. In frequency space, for 0,
the relation (304) is given by:
hOi i () .
(305)
The linear response equation for this quantity is obtained from the long wavelength limit
~k 0 of (303):
hOi = GR ( , ~k = 0 ) ( ) .
(306)
By comparing (305) and (306) we get:
GR (, ~k = 0) = i ,

( 0) ,

(307)

or, taking the imaginary part:

Im GR (, ~k = 0) = ,
52

( 0) .

(308)

From (308) we immediately get the so-called Kubo formula for :

= lim lim

0 ~k0

1
Im GR (, ~k) .

(309)

Let us now see how we can make use of (309) to compute the transport coefficient by
using holographic methods. Let us consider a (d + 1)-dimensional metric of the form:
ds2 = gtt dt2 + gzz dz 2 + gxx ij dxi dxj .

(310)

We will assume the metric (310) has an horizon at z = z0 and that gtt and that gzz behave
near z = z0 as:
gtt c0 (z0 z) ,

gzz

cz
,
z0 z

z z0 ,

(311)

with c0 and cz being constants. Let us consider a massless scalar field in this metric such
that its action is:
Z

M M
1
,
(312)
dd+1 x g
S =
2
q(z)
where the function q(z) is an effective coupling of the mode. The Euler-Lagrange equation
of motion derived from (312) is:
 g

M
g M N M = 0 .
(313)
q
In terms of the canonical momentum , defined as (see (141)):

L
g zz
=
g z ,

(z )
q

(314)

the equation of motion (313) becomes:

z =

g  t2 i2 
+
.
q
gtt
gxx

Let us now Fourier transform and in the variables t, ~x:

Z
d dd1 k i(~kxt)
(z, t, ~x) =
e
(z, , ~k) ,
(2)d
Z
d dd1 k i(~kxt)
(z, t, ~x) =
e
(z, , ~k) .
(2)d

(315)

(316)

Since in this massless case the scaling dimension is equal to d (see (112) for m = 0), the
one-point function is just the limit of the momentum at the boundary z = 0 and the
53

boundary field is just obtained by taking the limit z 0 of the bulk field , without any
multiplicative factor (see eqs. (147) and (129)). It follows from (305) that the transport
coefficient is obtained as:
"
#
(z, k )
(z, k )
= lim lim Im
= lim lim
,
(317)
k 0 z0
k 0 z0 i (z, k )
(z, k )
where, in the last step, we used the fact that Re/ 0 as 0 (see below). It turns
out that evaluating /( ) at the boundary is equivalent to evaluate it at the horizon. In
order to prove this, let us consider the general equation:


z A(z) z = B(z) (z) .
(318)
Let us write (318) in hamiltonian form. We first define:
P (z) A(z) z (z) .

(319)

z P (z) = B(z) (z) .

(320)

Then, (318) can be written as:

Then, one can readily prove that (318) and (320) can be combined in the following first-order
Riccati equation:
 P (z) 
1  P (z) 2
= B(z)
.
(321)
z
(z)
A(z) (z)
For the equation of motion of the scalar field in momentum space, the functions A(z) and
B(z) are given by:

A(z) =

~k 2 i
g h 2
B(z) =
+
,
q
gtt
gxx

g zz
g ,
q

(322)

and P (z) = (z) (see (314)). Then, we can write:

"
z

"

qgzz
=
g

!2

g 
gtt ~k 2 

1+
q gtt
gxx 2

#
.

(323)

The right-hand side of equation (323) vanishes when the ordered limit lim0 lim~k0 is
taken. It follows that /() is independent of z in this limit and, as claimed above, it can
be evaluated at the horizon. Thus, we can write the transport coefficient as:
= lim lim

k 0 zz0

54

(z, k )
.
i (z, k )

(324)

In order to evaluate the right-hand side of (324), let us study the equation of motion near
z = z0 , where becomes:
p
g(z0 )
1
(z0 z) z ,
(325)

cz q(z0 )
and the equation of motion takes the form:
h
i
h
z (z0 z) z (z, k ) + cz

2
k2 i

(z, k ) = 0 .
c0 (z0 z)
gxx (z0 )

(326)

To find an approximate solution of (326) near the horizon z = z0 , we neglect the last term
in the previous equation and try to find a solution of the type:
= (z0 z) .

(327)

Plugging the ansatz (327) in (326), we get that the exponent can take the following two
values:
r
cz
= i
,
(328)
c0
which correspond to the following two solutions:
i

(z0 z)

cz
c0

(329)

Only one of the two solutions in (329) is compatible with causality. Indeed, let us define
a new variable r as z0 z = er . In the r variable the horizon is located at r . By
inserting the t dependence, the two solutions are:
ei( t r) ,

(330)

p
where = cz /c0 . Clearly, is an incoming wave at the horizon since if we increase
t t+, we must decrease r as r r / to keep constant. Then, the wave moves
towards the horizon r and is an infalling wave at the horizon. Similarly, + is an
outgoing wave at the horizon. Causality on the gravity side, is implemented if we choose our
solution to be the infalling one . One can show that this corresponds to having retarded
Greens functions on the field theory side. This infalling solution near z = z0 satisfies:
s
gzz (z0 )
i ,
(331)
z =
gtt (z0 )
and, therefore, one has:



i 

z0

1
=
q(z0 )
55


g 

gzz gtt 

.
z0

(332)

It follows that the transport coefficient is given by:


r
g 
1
=

q(z0 ) gzz gtt 

(333)

z0

Notice that the square root in this last equation is just the area of the horizon AH divided
by the spatial volume V . Then, we can alternatively write the transport coefficient as:
=

1 AH
.
q(z0 ) V

In particular, for an AdS black hole, the previous expression becomes:

1  L d1
=
,
(AdS black hole) .
q(z0 ) z0

(334)

(335)

More interestingly, one can compare the general formula (334) with the entropy density, as
given by the Bekenstein-Hawking formula s = AH /(4GN V ). By computing the ratio /s,
we get the simple result:
4GN

=
.
(336)
s
q(z0 )

13

Holographic viscosities

Hydrodynamics can be thought as an effective theory describing the dynamics of a continuous

system at large distances and time scales. In order to study the dynamics of the system we
have to analyze the energy-momentum tensor T which is conserved ( T = 0). We will
assume that the system is at local thermal equilibrium and that the state of the system at
a given time is determined by the local temperature T (x) and the local fluid velocity u (x),
which satisfies the condition u u = 1.
The relation between T and u and the thermodynamic functions is expressed by means
of the so-called constitutive relations, which for isotropic fluids takes the form:
T = ( + p) u u + p g ,

(337)

where  is the energy density and p is the pressure, while is the so-called dissipative
part of T and depends on the derivatives of T (x) and u . In order to parametrize
at first-order in the derivatives, let us consider a local rest frame in which ui (x) = 0 (and
u = (1, 0, 0, 0)). One can choose this frame in such a way that the dissipative corrections to
the energy-momentum tensor components T 0 vanish. Then, 00 = 0i = 0 or, equivalently,
T 00 = , T 0i = 0. The only non-zero elements of the dissipative energy-momentum tensor
are ij . At first order in derivatives ij can be written as:
ij = i uj + j ui

2
ij k uk + ij k uk ,
3

56

(338)

where is the so-called shear viscosity and is the bulk viscosity.

Let us write in covariant form. We first define the projector onto the directions
perpendicular to u , as:
P = g + u u .
(339)
Then, in an arbitrary curved metric can be written as:
h

i


2 
= P P u + u + g u ,
3

(340)

where the covariant derivatives of the vectors u are defined as:

u = u u ,

(341)

with being the Christoffel symbols, defined as:

1 h g
g
g i
= g
.
+

2
x
x
x

(342)

We will find as (minus) the one-point function of T in the presence of a metric perturbation of the type g + h . By using linear response theory (eq. (301)), we
obtain:
Z

(x) =
G,
(x y) h (y)dy ,
(343)
R
where the retarded correlator is just:


i G,
(x y) = (x0 y 0 ) h T (x), T (y) i .
R

(344)

Let us now consider the following metric perturbation:

g00 (t, ~x) = 1 ,

g0i (t, ~x) = 0 ,

gij (t, ~x) = ij + hij (t) .

(345)

with hij << 1 and such that is traceless (hii = 0). We will assume that hij is a function of t
and that this variation with t is slow. The inverse metric is just:
g 00 (t, ~x) = 1 ,

g 0i (t, ~x) = 0 ,

g ij (t, ~x) = ij hij (t) ,

(346)

from which we get the different components of the projector P :

P 00 = 0 ,

P 0i = 0 ,

P ij = ij hij .

(347)

At first order in the perturbation, the Christoffel symbols are given by:
000 = 00i = 0 ,

0ij =
57

1
0 hij .
2

(348)

Therefore the covariant derivatives of the velocity are:

0 u0 = 0 ui = 0 ,

i uj =

1
0 hij .
2

(349)

From these values and the hypothesis that the metric perturbation is traceless it follows that
the covariant divergence of the velocity vanishes:
u=

1
0 hii = 0 .
2

(350)

Let us assume that the only non-zero value of hij is h12 . Then, the linear response value of
12 in frequency space is:
12 () = G12,12
(, ~k = 0) h12 () .
R

(351)

Moreover, by using the values of the covariant derivatives of the velocity we get
12 (t) = 0 h12 (t) ,

(352)

12 () = i h12 () .

(353)

or, in frequency space (for low ):

By comparing these two expressions of 12 , we obtain again Kubo formula for the shear
viscosity, namely:
h1
i
~k = 0) .
= lim
Im G12,12
(,
(354)
R
0
In order to compute holographically the retarded correlator of T12 , let us consider a general
d + 1-dimensional diagonal metric and let us perturb it by adding a non-diagonal element
along x1 x2 :


ds2 = gtt dt2 + gzz dz 2 + gxx ij dxi dxj + 2 dx1 dx2 ,
(355)
where is small and independent of x1 and x2 . Notice that, in the perturbed metric at first
order in , one has:
g12 = gxx ,
g 12 = .
(356)
Let us write the x1 x2 part of the metric (at first order) as:
gxx (dx1 )2 + gxx (dx2 + dx1 )2 ,

(357)

which is clearly similar to the ansatz corresponding to a Kaluza-Klein reduction along x2

with KK gauge field connection:
A = dx1 .
(358)

58

Therefore, we can use the known results of the KK reduction to write the quadratic action
for . Indeed, the Einstein-Hilbert action for the metric leads to the following expression for
the action of the gauge connection A (or equivalently the metric perturbation ):
Z

F2
1
dd+1 x g gxx
,
(359)
S =
16GN
4
where g is the determinant of the unperturbed metric (with = 0), F = dA is the field
strength of A and in F 2 = F F the indices are raised with the unperturbed metric. As:
F = t dt dx1 + 3 dx3 dx1 + z dz dx1 ,

(360)

we have:

i
2 h tt
2
xx
2
zz
2
F =
g (t ) + g (3 ) + g (z ) ,
gxx
or, taking into account that 1 = 2 = 0:
2

F2 =

2 MN
g
M N .
gxx

Thus, the action for the perturbation is:

Z

1
1
S =
dd+1 x g g M N M N ,
16GN
2

(361)

(362)

(363)

which is the canonical form of the action for a scalar field, with normalization constant
q = 16 GN .

(364)

It follows from our general calculation of section 12 of the transport coefficients for a scalar
field that the shear viscosity is given by:
=

1
AH
,
16 GN V

(365)

and, therefore, the ratio /s is just [24, 25]:

1
=
.
s
4

(366)

In ordinary units the ratio /s is given by:

~
=
,
s
4kB

(367)

where kB is the Boltzmann constant. Notice that this result does not depend on the metric
chosen. It is valid for any theory with a gravity dual given by Einstein gravity coupled to
59

matter fields. In this sense it is a universal result valid at (infinite coupling limit).
The finite coupling corrections can also be calculated. For N = 4 SYM one gets:


1
15 (3)
+

,
=
1+
3
s
4
2

(368)

where (x) is the Riemann zeta function ((3) = 1.2020). In general, /s at is very
small:

= 0.07957 .
(369)
s
The finite coupling corrections make /s increase. It is interesting to compare with the weak
coupling calculation, valid when 0:

A
  ,
=
s
2 log B

(370)

where A and B are constant coefficients that depend on the theory. Notice that in (370)
/s as 0. This is because a weakly coupled gauge theory is a gas with strong
dissipative effects, in which momentum can be transported over long distances due to the
long free path. In contrast a strongly coupled plasma is an almost perfect fluid in which
momentum is rapidly transferred between layers of sheared fluid.
Kovtun, Son and Starinets (KSS) [24] conjectured that 1/(4) is the lower bound for /s.
The lowest values of the /s ratio found experimentally occur in two physical systems: the
quark-gluon plasma created in heavy ion collisions at RHIC and the ultracold atomic Fermi
gases at very low temperature. Both systems have /s which is slightly above 1/(4).
Changing the gravity theory the KSS bound is violated. For example, by adding higher
curvature terms as in the Gauss-Bonnet gravity, whose action is given by:
Z

i
h
3
1
d5 x g R 2 GB R2 4R R + R R . (371)
SGB =
16G5

For this gravity theory, the following /s ratio is obtained:

1 4GB
=
,
s
4

(372)

which violates the KSS bound if GB > 0.

Final remarks
In these lecture notes we have reviewed the basic features of the AdS/CFT duality, focusing
on its conceptual foundations and on some particular applications. We have only scratched
60

the surface of the subject, which in the last years has become a highly diversified field with
many ramifications and connections. For the reader interested in knowing more on some of
these applications of the holographic duality, let us quote some review articles, where the
reader can find detailed accounts and the original references.
In this review we have mostly dealt with the gravitational description of N = 4 SU (N )
gauge theory, which only contains fields transforming in the adjoint representation of the
gauge group. In this sense N = 4 SYM is a theory of pure glue. In order to extend the
duality to theories closer to particle physics phenomenology one should be able to include
flavor fields transforming in the fundamental representation of the gauge group (i.e. quarks).
This can be done by adding the so-called flavor branes, as reviewed in [8, 26]. Moreover, the
gauge/gravity duality can be extended to include less supersymmetric theories exhibiting
confinement (see [27]) and one can construct holographic duals of quantum cromodynamics
( [8, 9, 28, 29]). The holographic methods can also be used to study dynamical electroweak
symmetry breaking, in the framework of walking technicolor models [30]. On the other hand,
the AdS/CFT correspondence has unveiled the integrable character of planar N = 4 SYM
and has allowed to extend the duality beyond the supergravity regime [31].
One of the more interesting recent developments of the gauge/gravity duality is the application of the string theoretical ideas to the down-to-earth problems of condensed matter
physics. Indeed, condensed matter physics is full of strongly-coupled systems which display
quantum criticality with specific scaling laws. The gauge/gravity duality allows to map this
scaling behavior to the general covariance of a gravity theory, for which one can apply the
calculational tools and physical intuition of general relativity. In this way one can model
the behavior of unusual phases of matter, such us strange metals or unconventional superconductors [6, 3234]. In this context holography is emerging as a new tool to understand
the collective quantum behavior not explained by the conventional paradigms, such as the
Fermi-liquid theory.
It is also interesting to point out the connection between the AdS/CFT correspondence
and quantum information theory. In particular, there is a holographic proposal for the
entanglement entropy [35, 36], which allows a simple geometrical calculation of the latter as
the area of a minimal surface. Let us finally mention that holography has also been applied
to the study of strongly-coupled hydrodynamics. This particular version of the duality is
called the fluid-gravity correspondence [37, 38].
Hopefully these lectures will stimulate the reader to explore some of the topics listed above.

Acknowledgments
I am grateful to Yago Bea, Niko Jokela, Javier Mas and Ricardo Vazquez for their comments
and help in the preparation of these lecture notes. I also thank Carlos Merino for his invi-

61

tation to deliver the course on the AdS/CFT correspondence at the third IDPASC school.
This work is funded in part by the Spanish grant FPA2011-22594, by Xunta de Galicia (Consellera de Educacion, grant INCITE09 206 121 PR and grant PGIDIT10PXIB206075PR),
by the Consolider-Ingenio 2010 Programme CPAN (CSD2007-00042), and by FEDER.

References
[1] J. M. Maldacena, The Large N limit of superconformal field theories and supergravity,
Adv. Theor. Math. Phys. 2 (1998) 231 [hep-th/9711200].
[2] S. S. Gubser, I. R. Klebanov and A. M. Polyakov, Gauge theory correlators from
noncritical string theory, Phys. Lett. B 428, 105 (1998) [hep-th/9802109].
[3] E. Witten, Anti-de Sitter space and holography, Adv. Theor. Math. Phys. 2, 253
(1998) [hep-th/9802150].
[4] O. Aharony, S. S. Gubser, J. M. Maldacena, H. Ooguri and Y. Oz, Large N field
theories, string theory and gravity, Phys. Rept. 323, 183 (2000) [hep-th/9905111].
[5] E. DHoker and D. Z. Freedman, Supersymmetric gauge theories and the AdS / CFT
correspondence, hep-th/0201253.
[6] S. A. Hartnoll, Lectures on holographic methods for condensed matter physics, Class.
Quant. Grav. 26, 224002 (2009) [arXiv:0903.3246 [hep-th]].
[7] J. McGreevy, Holographic duality with a view toward many-body physics, Adv. High
Energy Phys. 2010, 723105 (2010) [arXiv:0909.0518 [hep-th]].
[8] J. Casalderrey-Solana, H. Liu, D. Mateos, K. Rajagopal and U. A. Wiedemann,
Gauge/String Duality, Hot QCD and Heavy Ion Collisions, arXiv:1101.0618 [hepth].
[9] Y. Kim, I. J. Shin and T. Tsukioka, Holographic QCD: Past, Present, and Future,
Prog. Part. Nucl. Phys. 68 (2013) 55 [arXiv:1205.4852 [hep-ph]].
[10] A. Adams, L. D. Carr, T. Schafer, P. Steinberg and J. E. Thomas, Strongly Correlated
Quantum Fluids: Ultracold Quantum Gases, Quantum Chromodynamic Plasmas, and
Holographic Duality, New J. Phys. 14 (2012) 115009 [arXiv:1205.5180 [hep-th]].
[11] M. B. Green, J. H. Schwarz and E. Witten, Superstring theory, 2 vols. (Cambridge
University Press, Cambridge, 1987).
[12] J. Polchinski, String theory, 2 vols. (Cambridge University Press, Cambridge, 1998).
62

[13] K. Becker, M. Becker and J. H. Schwarz, String theory and M-theory. (Cambridge
University Press, Cambridge, 2007).
[14] E. Kiritsis, String theory in a nutshell. (Princeton University Press, Princeton, 2007).
[15] B. Zwiebach, A first course in string theory. (Cambridge University Press, Cambridge,
2009).
[16] L. E. Iba
nez and A. Uranga, String theory and Particle Physics. (Cambridge University
Press, Cambridge, 2012).
[17] G. t Hooft, A planar diagram theory for strong interactions, Nucl. Phys. B 72, 461
(1974).
[18] A. V. Manohar, Large N QCD, hep-ph/9802419.
[19] M. Henningson and K. Skenderis, The Holographic Weyl anomaly, JHEP 9807 (1998)
023 [hep-th/9806087].
[20] S. de Haro, S. N. Solodukhin and K. Skenderis, Holographic reconstruction of spacetime and renormalization in the AdS / CFT correspondence, Commun. Math. Phys.
217 (2001) 595 [hep-th/0002230].
[21] K. Skenderis, Lecture notes on holographic renormalization, Class. Quant. Grav. 19
(2002) 5849 [hep-th/0209067].
[22] J. M. Maldacena, Wilson loops in large N field theories, Phys. Rev. Lett. 80, 4859
(1998) [arXiv:hep-th/9803002].
[23] 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 (2001) 379 [arXiv:hep-th/9803001].
[24] P. Kovtun, D. T. Son and A. O. Starinets, Viscosity in strongly interacting quantum field theories from black hole physics, Phys. Rev. Lett. 94, 111601 (2005) [hepth/0405231].
[25] D. T. Son and A. O. Starinets, Viscosity, Black Holes, and Quantum Field Theory,
Ann. Rev. Nucl. Part. Sci. 57 (2007) 95 [arXiv:0704.0240 [hep-th]].
[26] J. Erdmenger, N. Evans, I. Kirsch and E. Threlfall, Mesons in Gauge/Gravity Duals
- A Review, Eur. Phys. J. A 35 (2008) 81 [arXiv:0711.4467 [hep-th]].
[27] J. D. Edelstein and R. Portugues, Gauge/string duality in confining theories, Fortsch.
Phys. 54 (2006) 525 [hep-th/0602021].

63

[28] K. Peeters and M. Zamaklar, The String/gauge theory correspondence in QCD, Eur.
Phys. J. ST 152 (2007) 113 [arXiv:0708.1502 [hep-ph]].
[29] D. Mateos, String Theory and Quantum Chromodynamics, Class. Quant. Grav. 24
(2007) S713 [arXiv:0709.1523 [hep-th]].
[30] M. Piai, Lectures on walking technicolor, holography and gauge/gravity dualities,
Adv. High Energy Phys. 2010 (2010) 464302 [arXiv:1004.0176 [hep-ph]].
[31] N. Beisert, C. Ahn, L. F. Alday, Z. Bajnok, J. M. Drummond, L. Freyhult, N. Gromov
and R. A. Janik et al., Review of AdS/CFT Integrability: An Overview, Lett. Math.
Phys. 99 (2012) 3 [arXiv:1012.3982 [hep-th]].
[32] S. Sachdev, Condensed Matter and AdS/CFT, Lect. Notes Phys. 828 (2011) 273
[arXiv:1002.2947 [hep-th]].
[33] S. Sachdev, What can gauge-gravity duality teach us about condensed matter
physics?, Ann. Rev. Condensed Matter Phys. 3 (2012) 9 [arXiv:1108.1197 [condmat.str-el]].
[34] A. G. Green, An Introduction to Gauge Gravity Duality and Its Application in Condensed Matter, Contemporary Physics , Volume: 54, Issue: 01, pages 33 - 48 (2013)
[arXiv:1304.5908 [cond-mat.str-el]].
[35] T. Nishioka, S. Ryu and T. Takayanagi, Holographic Entanglement Entropy: An
Overview, J. Phys. A 42 (2009) 504008 [arXiv:0905.0932 [hep-th]]
[36] T. Takayanagi, Entanglement Entropy from a Holographic Viewpoint, Class. Quant.
Grav. 29 (2012) 153001 [arXiv:1204.2450 [gr-qc]].
[37] M. Rangamani, Gravity and Hydrodynamics: Lectures on the fluid-gravity correspondence, Class. Quant. Grav. 26 (2009) 224003 [arXiv:0905.4352 [hep-th]].
[38] V. E. Hubeny, S. Minwalla and M. Rangamani, The fluid/gravity correspondence,
arXiv:1107.5780 [hep-th].

64