AdSCFT 2010 01 22

Introduction to gauge gravity duality
Johanna Erdmenger
typed by Oliver Schlotterer
January 22, 2010
2
Contents
1 Introduction 7
2 Preparations 9
2.1 Conformal eld theory in d dimensions . . . . . . . . . . . . . . . . . . . . . . . 9
2.1.1 Conformal coordinate transformations . . . . . . . . . . . . . . . . . . . 9
2.1.2 Conformal elds and correlation functions . . . . . . . . . . . . . . . . . 11
2.1.3 The energy momentum tensor in a CFT . . . . . . . . . . . . . . . . . . 12
2.2 ^ = 4 super Yang Mills theory . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2.1 Non-abelian gauge theories . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2.2 The 1/N expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2.3 Supersymmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2.4 Field content of ^ = 4 supersymmetric eld theory . . . . . . . . . . . . 18
2.2.5 The superconformal algebra and its representations . . . . . . . . . . . . 19
2.3 Anti-de Sitter space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3 Introduction to superstring theory 25
3.1 Bosonic strings in Minkowski spacetime . . . . . . . . . . . . . . . . . . . . . . . 25
3.1.1 Closed strings in Minkowski spacetime . . . . . . . . . . . . . . . . . . . 26
3.1.2 Open strings in Minkowski spacetime . . . . . . . . . . . . . . . . . . . . 27
3.2 Bosonic string theory in background elds . . . . . . . . . . . . . . . . . . . . . 27
3.2.1 Background elds of the closed string sector . . . . . . . . . . . . . . . . 28
3.2.2 Background elds of the open string sector . . . . . . . . . . . . . . . . . 29
3.3 Superstring theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.3.1 Open superstrings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.3.2 Closed superstrings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.4 D branes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.4.1 Eective actions for D branes . . . . . . . . . . . . . . . . . . . . . . . . 32
3
4 CONTENTS
3.4.2 D branes in supergravity . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4 The AdS/CFT correspondence 35
4.1 Maldacenas original argument . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.1.1 D3 branes from the open string point of view . . . . . . . . . . . . . . . 35
4.1.2 D3 branes from the closed string point of view . . . . . . . . . . . . . . . 36
4.1.3 Dierent forms of the AdS/CFT correspondence . . . . . . . . . . . . . . 37
4.2 Field operator map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
4.2.1 CFT correlation functions . . . . . . . . . . . . . . . . . . . . . . . . . . 38
4.2.2 The dual elds of supergravity . . . . . . . . . . . . . . . . . . . . . . . . 39
4.2.3 AdS propagators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
5 Tests of the correspondence 45
5.1 Three point function of 1/2 BPS operators . . . . . . . . . . . . . . . . . . . . . 45
5.1.1 Correlation functions of 1/2 BPS operators . . . . . . . . . . . . . . . . . 45
5.1.2 The non-renormalization theorem . . . . . . . . . . . . . . . . . . . . . . 47
5.1.3 The three point function on the gravity side . . . . . . . . . . . . . . . . 48
5.2 The conformal anomaly . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
5.2.1 The conformal anomaly on the eld theory side . . . . . . . . . . . . . . 53
5.2.2 The conformal anomaly on the gravity side . . . . . . . . . . . . . . . . . 54
6 Generalizations of AdS/CFT 57
6.1 Holographic renormalization group ows . . . . . . . . . . . . . . . . . . . . . . 57
6.1.1 Renormalization group ow in supergravity . . . . . . . . . . . . . . . . 58
6.1.2 Leigh Strassler ow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
6.1.3 Holographic ows in supergravity . . . . . . . . . . . . . . . . . . . . . . 62
6.2 The holographic c theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
6.3 Last lecture of 2009 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
6.4 Applications 1: Field Theories at nite temperature . . . . . . . . . . . . . . . . 64
6.5 Gauge gravity duality at nite temperature and density . . . . . . . . . . . . . . 66
6.5.1 Finite density . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
6.5.2 Chemical potential in quantum eld theory . . . . . . . . . . . . . . . . . 67
6.6 Dissipative dynamics close to equilibrium . . . . . . . . . . . . . . . . . . . . . . 69
6.6.1 Retarded Green functions in QFT . . . . . . . . . . . . . . . . . . . . . . 69
6.6.2 The gravity side of Green functions . . . . . . . . . . . . . . . . . . . . . 70
6.6.3 Example: Holographic computation of Ohms law . . . . . . . . . . . . . 72
CONTENTS 5
6.7 Hydrodynamics and shear viscosity in AdS/CFT . . . . . . . . . . . . . . . . . . 73
6.7.1 Relativistic hydrodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . 73
6.7.2 Kubo formula from linear response theory . . . . . . . . . . . . . . . . . 74
6.7.3 Shear modes and sound modes . . . . . . . . . . . . . . . . . . . . . . . . 75
6.7.4 AdS/CFT calculation of the shear viscosity . . . . . . . . . . . . . . . . 76
6 CONTENTS
Chapter 1
Introduction
Let us start by giving the (rough) statement of gauge gravity duality:
Some quantum eld theories are equivalent to (quantum) gravity theories.
In particular limits, the gravity theory becomes classical and the corresponding quantum
eld theory (QFT) strongly coupled.
The second point makes the duality particularly useful since by other methods, dynamical
processes are inaccessible in the strongly coupled regime of QFTs: Normally, QFT calculations
are done by means of perturbation theory, but this only works at weak coupling. Lattice
gauge theory might be a powerful way out of this dilemma, but it is hard to use for capturing
dynamics. Also, for technical reasons, it is problematic at high temperature or large density
and chemical potential.
The purpose of this lecture is the following:
explain how and for which QFTs the gauge gravity duality works
work out the details of the subtle limit
give nice examples and applications
Gauge gravity duality originates from string theory. (However, there is a limit of the duality
in which string theory reduces to classical gravity, i.e. general relativity (GR).) The duality
generalizes the so-called AdS/CFT correspondence,
AdS anti deSitter spacetime, a solution of Einsteins equations
CFT conformal eld theory
a conjecture for equivalence between
7
8 CHAPTER 1. INTRODUCTION
string theory on certain ten dimensional backgrounds involving AdS spacetime
four dimensional QFT with conformal symmetry (supersymmetric SU(N) Yang Mills)
There is no mathematical proof for the AdS/CFT correspondence but overwhelming evidence
of its correctness. The conjecture states that these two theories are equivalent including ob-
servables, states, correlation functions and dynamics. It is interesting to ask in which way the
conformal symmetry could be dropped in order to cover non-conformal theories such as QCD.
The ten dimensional spacetime of the string theory side contains a ve dimensional anti deSitter
spacetime with a four dimensional boundary. The four dimensional QFT can be regarded as
living on this four dimensional boundary. In analogy to conventional holograms (which encode
three dimensional information on a lower dimensional surface), the AdS/CFT correspondence
is said to realize the holographic principle.
As in any eld theory, symmetrie are of central importance for gauge gravity duality. The
two equivalent theories have the same symmetries. Moreover, the correspondence provides a
one-to-one map between classical gravity elds and quantum operators of the eld theory, i.e.
some sort of holographic dictionary. This map then identies representations of the common
symmetry group.
As to the literature to this subject, there are (at present) no textbooks available. Let us
instead refer to the original papers [1], [2], [3], [4] which marked the birth of the AdS/CFT
correspondence. Several review articles followed [5], [6], [7] which assume lots of background
knowledge and usually emphasize particular aspects of the duality. Finally, at later stages of
this course, [10], [11] are helpful references for applications.
Chapter 2
Preparations
sec:preparations
In this section we elucidate the several subject areas which will be connected by the correspon-
dence.
2.1 Conformal eld theory in d dimensions
2.1.1 Conformal coordinate transformations
sec:CFT
Conformal coordinate transformations are dened as those local transformations x
(x)
that leave angles invariant. In a Euclidean d-dimensional space R
d
we therefore can write
dx
dx
=
2
(x) dx
dx
. (2.1.1) eq:metricTransf
The corresponding innitesimal coordinate transformation from old coordinates x to new ones
x
looks like
x
= x
+ v
(x) (2.1.2)
and we have
(x) = 1 (x) , (x) =
1
d
v(x) . (2.1.3) 2,3
Equivalently to (2.1.1) we can formulate an equation for the vector v, the conformal Killing
equation,
= 2 (x)
, (2.1.4) eq:confKilling
taking its trace yields the expression (2.1.3) for (x). We will work in d dimensional Euclidean
space where
. Solutions v to (2.1.4) are referred to as conformal Killing vectors, the

most general one reads
v
= a
+ x
+ b
x
2
2 (b x) x
. (2.1.5) eq:solutionsForv
9
10 CHAPTER 2. PREPARATIONS
This Killing vector leads to the scale factor (x) = 2(b x). Equation (2.1.5) is valid for
any d. Note that in the special case of d = 2 the conformal Killing equation (2.1.4) is nothing
but the Cauchy-Riemann equations
1
v
1
=
2
v
2
,
1
v
2
=
2
v
1
. (2.1.6)
Thus, in d = 2 all holomorphic functions v(x) are solutions and generate conformal coordi-
nate transformations. In this case we have an innite number of functions solving (2.1.5),
accompanied by an innite number of associated conserved quantities.
However, we will mostly consider theories in d = 4 dimensions, for example in Minkowski space
or on the boundary of Ads
5
. Here we have a nite amount of conserved quantities. Counting
the independent components of the factors in the solutions (2.1.5) amounts to a total number
of 15:
a
+ 6
+ 1
b
+ 4
total 15
The general conformal Killing vector (2.1.5) may be viewed as the combination of elementary
transformations. The group of large conformal transformation is generated by innitesimal
elements of the conformal algebra. We dene locally orthogonal tranformations 1 correspond-
ing to a group element g of the conformal group as
1
g
(x) :=
g
(x)
x
. (2.1.7)
One can easily show that 1 O(d), i.e. that 1
g
(x)1
g
(x) =
. The group multiplication

and the inverse are given as follows:
1
g
(gx) 1
g
(x) = 1
g
g
(x) ,
_
1
g
(x)
_
1
= 1
g
1
(gx) (2.1.8)
With these we can construct translations and rotations as
x
= 1
+ a
, (x) = 1 . (2.1.9)
Scale transformations () and special conformal transformations (b
) involve a non-trivial
factor:
x
= x
, (x) = (2.1.10)
2.1. CONFORMAL FIELD THEORY IN D DIMENSIONS 11
x
=
x
+b
x
2
g
(x)
,
g
(x) = 1 + 2b x +bx
2
. (2.1.11)
Together, these transformations form a group isomorphic to SO(d + 1, 1) (or SO(d, 2) in
Minkowski spacetime). All transformations belonging to this group can be constructed by
performing translations, rotations, and inversions; the latter are given by
x
=: (ix)
=
x
x
2
,
i
(x) = x
2
(2.1.12)
1
i
(x) =: I
(x) =
2
x
x
2
. (2.1.13)
Special conformal transformations can be composed by concatenating inversion+translation+
inversion.
2.1.2 Conformal elds and correlation functions
So far we examined coordinate transformations. Now we will investigate the behaviour of elds.
For instance, the ^ = 4 super Yang Mills theory (SYM) mentioned in the introduction only
contains elds transforming covariantly under the conformal group. In general QFTs (such as
QED or QCD), conformal symmetry is generically broken by quantum eects (anomalies).
Necessary condition for a eld theory to be conformally symmetric is a vanishing -function.
The latter describes the change of a coupling g with energy scales , i.e.
(g) =
g
, (2.1.14)
so (g) = 0 rephrases scale invariance.
A conformally covariant operator O of a conformal eld theory (CFT) transforms as follows
under innitesimal conformal transformations (with Killing vector v and = v/d):
v
O = (L
v
O) , L
v
= v(x) + (x)
1
2

[
v
]
(x) S
(2.1.15)
Here, denotes the scaling dimension of the operator O and S
a generator of O(d) in an
appropriate representation. It only aects spinor-, vector- and tensor elds but no scalars :
v
=
_
v(x) + (x)
_
(2.1.16)
In general QFTs, correlation functions are dened as time ordered vacuum expectation values,
e.g. a two point function of some eld is given by
(x) (y)) := 0[ T (x) (y) [0) , (2.1.17) 2,5
three-, four- and higher point functions by analogous expressions. Generically, their computa-
tion is quite involved and possible only in the framework of perturbation theory.
Let us also give the path integral analogue of the denition (2.1.17) in the operator approach. In
a scalar eld theory governed by action o[], the partition function Z and a general correlation
function O) is dened by the path integrals
Z :=
_
T e
S[]
, O) :=
1
Z
_
T Oe
S[]
. (2.1.18)
In CFTs, conformal symmetry is so strong that it determines the form of the two- and three
point correlation functions up to a managable number of parameters. In the notation (xy)
2
=
(x y)
(x y)
, the two- and three point functions of scalars

i
with scale dimensions
i
are
given by
1
(x)
2
(y)) :=
c
1
,
2
(x y)
2
1
(2.1.19)
1
(x)
2
(y)
3
(z)) :=
k
(x y)
1
+
2
3
(y z)
1
+
2
+
3
(x z)
2
+
3
(2.1.20)
with constants c, k determined by the eld content.
Four point correlators
1
(x)
2
(y)
3
(z)
4
(w)) are less constraint by the symmetry since they
involve dimensionless cross ratios
(xy)
2
(zw)
2
and
(xz)
2
(yw)
2
.
2.1.3 The energy momentum tensor in a CFT
The symmetric energy momentum tensor T
subject to the conservation law
= 0 (or
rather
= 0 in curved spacetime) generates the Noether currents associated with confor-

mal symmetry. The innitesimal transformations with conformal Killing vector v
gives rise to
the conserved current
j
= T
. (2.1.21)
In this subsection, we will now show an important property of the energy momentum tensor in
a conformal eld theory, namely its tracelessness T
= 0.
It is a common method in QFT to introduce sources for operators in a QFTs action, and then
express the operator (in correlation functions) as the functional derivative of the generating
functional. To do so, the action o
0
of our theory is modied by an additive term which couples
the operator to its source. For instance consider some scalar operator and its source J,
o[, J] = o
0
[] +
_
d
d
x (x) J(x) . (2.1.22)
Correlation function of that operator may now be calculated as the functional derivative of
the generating functional W[J] := ln Z[J] of the theory with respect to the source J, e.g.
(x))
W[J]
J(x)
. (2.1.23)
2.2. ^ = 4 SUPER YANG MILLS THEORY 13
One can also apply this procedure to vector- and tensor operators,
o = o
0
+
_
d
d
x
_
J + V
+ T
_
. (2.1.24)
It can be shown that the source of the energy momentum tensor is exactly the quantity that
has the properties of the metric. So the energy momentum tensor is obtained by calculating
T
(x) =
2
_
[ det g[
W[g]
g
(x)
. (2.1.25) eq:defEMTensor
The metric transforms under conformal coordinate transformations induced by a vector eld v
as
v
g
= 2g
, so requiring invariance of W implies

0 =
v
W[g] =
_
d
d
x
W[g]
g
(x)

v
g
(x) =
_
d
d
x
_
_
[ det g[ T
2
_
_
2 g
_
=
_
d
d
x
_
[ det g[ T

. (2.1.26)
Since T

vanishes upon integration against an arbitrary function , one can conclude the
announced tracelessness of the energy momentum tensor
T

= 0 . (2.1.27)
2.2 ^ = 4 super Yang Mills theory
sec:maxiSusy
In this section we want to develop the eld theory side of the AdS/CFT correspondence the
maximally supersymmetric SU(N) gauge theory. This ^ = 4 super Yang Mills theory is an
example for a d = 4 dimensional CFT. In the following, the ingredients will be introduced step
by step.
2.2.1 Non-abelian gauge theories
Super Yang Mills theory is a non-abelian gauge theory, i.e. its elds take values in the algebra
of a non-abelian gauge group. QED, on the other hand, is associated with the abelian gauge
group U(1). Let us take it as an introductory example for the necessity of a gauge eld:
Consider a complex scalar eld (x) transforming under local U(1) transformations as
(x) e
i(x)
(x) ,
(x)
_
e
i(x)
(x)
_
,= e
i(x)

(x) . (2.2.1)
The derivative
obviously does not transform like the eld itself, so a connection A
is
required in order to dene a gauge covariant derivative:
D
(x) :=
_
+ iA
_
(x) e
i(x)
D
(x) A
(2.2.2)
With A
transforming like that, we can use the covariant derivative D
to construct gauge
invariant objects (e.g. kinetic terms in the action). Furthermore, the eld strength tensor
F
:=
(2.2.3)
is unaected by gauge transformations of A
since
[
]
= 0.
The most important examples of non-abelian gauge groups in these lectures are SU(N) with
N 2. One has to distinguish two transformation properties of elds under the non-abelian
SU(N):
Fields transforming in the fundamental representation of the gauge group are elements of
an N dimensional vector space:
q
i
(x)
_
e
i
a
(x)T
a
_
i
j
q
j
(x) , i, j = 1, 2, ..., N (2.2.4)
The SU(N) generators T
a
are traceless hermitian N N matrices and ensure that e
i
a
T
a
is unitary. If the parameters
a
(x) are innitesimal, the eld q
i
is shifted by an algebra
element
q
i
(x) q
i
(x) + i
a
(x) (T
a
)
i
j
q
j
(x) (2.2.5)
Fields transforming in the adjoint representation of the gauge group are aligned into the
N
2
1 dimensional algebra su(N),
i
j

a
(T
a
)
i
j
_
e
i
b
T
b
_
i
k
a
(T
a
)
k
l
_
e
i
c
T
c
_
l
j
. (2.2.6)
Innitesimally, conjugation by a group element e
i
a
T
a
involves the commutator [T
a
, T
b
] =
if
abc
T
c
of the su(N) generators:
a
T
a

a
T
a
+ i
_
b
T
b
a
T
a

a
T
a
b
T
b
_
=
a
T
a
i
b
a
_
T
a
, T
b
=
a
T
a
+ f
abc
b
T
c
(2.2.7)
Non-abelian gauge elds A
= A
a
T
a
give rise to a non-abelian eld strength tensor in the
adjoint representation
F
:=
+ ig
_
A
, A
=
_
A
a
A
a
g f
abc
A
b
A
c
_
T
a
. (2.2.8)
The transformation properties of F
can be deduced from its alternative denition as a com-

mutator of (non-abelian) gauge covariant derivatives (with g denoting the gauge coupling)
(D
)
i
j
:=
i
j
+ ig A
a
(T
a
)
i
j
, F
=
i
g
_
D
, D
. (2.2.9)
One can thus form a gauge invariant action for the eld strength by taking a trace over the i, j
indices of the generators:
o[A]
_
d
4
x Tr
_
F
_
(2.2.10)
The non-linear contribution to F
gives rise to interactions with the vertices

PICTURE: 3 vertex g, 4 vertex g
2
Later, we will discuss QCD, an SU(3) Yang Mills theory of gluons together with fundamental
quarks. It has two essential features following from the negative sign of the function
asymptotic freedom, attenuation of the coupling in the UV region, i.e. lim
g() = 0
connement, the coupling g grows rapidly in the IR regime 0
2.2.2 The 1/N expansion
It was suggested by Gerald tHooft that non-abelian gauge theories may simplify when SU(N)
is studied in the limit N . The diagrammatic expansion of SU(N) eld theory suggests
that it is a free string theory in the N limit with string coupling 1/N.
To understand this, let us consider a toy model: let
a
denote a set of elds with an adjoint

index
a
and a label

for spin- or avour degrees of freedom. We assume that the interaction
vertices mimic Yang Mills theory a three point vertex g and a four point vertex g
2
. The
toy models Lagrangian then reads
/ Tr
_
d
_
+ g c
Tr
_
_
+ g
2
d
Tr
_
_
. (2.2.11)
A rescaling g
turns it into
/
1
g
2
_
Tr
_
d
_
+ c
Tr
_
_
+ d
Tr
_
_
_
. (2.2.12) 2,20
To have a well-dened N limit, it is convenient to introduce the tHooft coupling
:= g
2
N . (2.2.13)
If we send N at constant , the coecient of (2.2.12) diverges but the number N
2
1
of components in the elds diverges as well. This point becomes clearer after an analysis of
Feynman graphs in the tHooft limit. The propagator will have the following SU(N) index
structure to ensure tracelessness,
i
j
(x)
k
l
(y)
_

_
i
l
k
j
1
N

i
j
k
l
_
, (2.2.14)
regardless of the spacetime dependence. In the N limit, the second term can be safely
ignored, this suggests double line notation
PICTURE of a i-k over j-l double line
Feynman diagrams then become networks of double lines. Vertices scale as
N
, propagators
as

N
, and the sum over indices in a trace contributes a factor of N for each closed loop. If
we introduce shorthands (V, E, F) for the numbers of vertices, propagators (edges) and loops
(faces) respectively, diagrams are proportional to
diagram(V, E, F) N
V E+F
EV
= N
EV
. (2.2.15)
The power of the expansion parameter N is precisely the Euler characteristic
:= V E + F = 2 2 g , (2.2.16)
related to the surfaces number of handels (the genus) g.
Any physical quantity in this theory is given by a perturbative expansion of type
g=0
N
22g
i=0
c
g,i
i
=
g=0
N
22g
f
g
() (2.2.17)
with f
g
() a polynomial in the tHooft coupling. For large N, the series is clearly dominated by
surfaces of minimal genus, the so-called planar diagrams. As an example, compare the following
vacuum amplitudes
() with N
2
and non-planar (X) with N
0
The form of this expansion is the same as in a perturbative theory of closed oriented strings
with string coupling
1
N
. The propagator and the interaction vertex of a closed string is depicted
below.
PIC: Zylinder und Hose
In this simple toy model, one cannot say which string theory ts to the perturbative series.
For ^ = 4 SYM, however, the AdS/CFT correspondence tells us which string theory leads to
the correct expansion: ten dimensional type IIB superstring theory on AdS
5
S
5
.
2.2.3 Supersymmetry
We know to have Poincare symmetry in the at Minkowski spacetime, which is equipped with
a mostly positive metric of signature = diag(, +, +, +). Generators of translations and
Lorentz transformations will be denoted as P
and L
respectively. Supersymmetry now

enlarges the Poincare algebra
_
L
, P
= i (
) (2.2.18) L,P
_
L
, L
= i (
) (2.2.19) L,L
by including spinor supercharges Q. In so-called Weyl notation we have aleft-handed spinor Q
a
and its right-handed counterpart

Q
a
= (Q
a
where the SL(2, C) indices , take values 1, 2

and a counts the number of independent supersymmetries a = 1, . . . , ^. The Qs transform as
Weyl spinors of SO(1, 3)
= SL(2, C)/Z
2
.
The two-component Weyl spinor notation is related to the Dirac four-spinor notation by
Q
a
D
=
_
_
Q
a
Q
a
_
_
,
=
_
_
0

0
_
_
, (2.2.20)
where
= (1,
i
) and
= (1,
i
) are four vectors of 2 2 matrices with the standard
Pauli matrices
i
as their spatial entries.
The supercharges commute with the generators of translations but otherwise obey the algebra
_
Q
a
,

Q
b

_
= 2
a
b
,
_
Q
a
, Q
b
_
= 2
Z
ab
. (2.2.21) eq:susyAlgebra
Here the operators Z
ab
are referred to as central charges. They commute with all the Poincare-
and supersymmetry generators Q
a
and need to by antisymmetric Z
ab
= Z
ba
in order to respect
the anticommutators symmetry. Therefore, for ^ = 1 supersymmetry, we have Z = 0.
The supersymmetry algebra (2.2.21) is invariant under global phase rotations of the super-
charges Q
a
1,2
into each other. This forms an R symmetry group denoted as U(1)
R
. In addition,
when ^ > 1, the dierent supercharges may be rotated into one another under the unitary
group SU(N)
R
which extends the R symmetry.
The eld theory in the AdS/CFT dictionary has ^ = 4 supersymmetries. Let us briey
explain why this is the maximal supersymmetry for a pure gauge theory without gravity: Each
supercharge Q
a
,

Q
a
changes the spin of the state it acts on by 1/2. In absence of gravity,
helicities between -1 and +1 occur, hence no spin modication greater than 2 = ^
max
1/2 is
allowed.
In the ^ = 4 theory we have R symmetry SU(4)
= SO(6). Exactly this is the isometry group

of the sphere in the AdS
5
S
5
background of the string theory side of the correspondence.
The AdS
5
factor has the symmetries encoded by SO(4, 2) in Minkowski space or SO(5, 1)
in a Euclidean formulation. These groups are isomorphic to the conformal group in d = 4
dimensions according to our analysis in subsection 2.1.1.
eld range representation of SU(4)
R
vector A
(1) singlet
Weyl fermions
a
, a = 1, 2, 3, 4 (4) fundamental
real scalars X
i
, i = 1, 2, . . . , 6 (6) adjoint
Table 2.1: The eld content of the ^ = 4 supersymmetry multiplet and the representation in
which these elds transform with respect to the R symmetry group SU(4)
R

= SO(6)
R
tab:n4content
2.2.4 Field content of ^ = 4 supersymmetric eld theory
Representations of the supersymmetry algebra make up the SUSY multiplets. Their com-
ponents are spin 1 vector elds, spin
1
2
fermion elds and spin 0 scalar elds. In ^ = 4
supersymmetry we encounter maximal supersymmetry if s = 1 is the highest spin in a SUSY-
multiplet. This implies that we cannot describe gravity with this theory, because the graviton
is supposed to have spin 2.
For any ^ with 1 ^ 4 we encounter one gauge multiplet, which is a multiplet transforming
in the adjoint representation of the gauge group (while we are used to have matter elds in
the fundamental representation in non-supersymmetric theories). For ^ = 4 this is the only
possible multiplet.
Lower symmetry ^ = 1 and ^ = 2 also admits matter multiplets which we will not discuss
here, though. (But to make you familiar with the names, the multiplet in the fundamental
representation in ^ = 1 SUSY is called chiral multiplet, and the multiplet in the fundamental
representation in ^ = 2 SUSY is called the hypermultiplet). The content of the ^ = 4 multiplet
is given in table 2.1. Note that this theory is non-chiral. The Lagrangian may be written as
/ = Tr
_
1
2 g
2
F
+

I
8
2
F
i
D
X
i
D
X
i
+ g
a,b,i
C
ab
i

a
_
X
i
,
b
+ g
a,b,i
C
iab

a
_
X
i
,

+
g
2
2
i,j
_
X
i
, X
j
2
_
.
(2.2.22) eq:N4Lagrangian
Here the trace is summing over gauge indices ,

which are suppressed in the expression above.
They appear if we rewrite the adjoint elds correctly as linear combinations of the generators
T
A
of the gauge group, e.g. X
i
= X
iA
T
A

. The symbol
I
denotes the instanton number and
=
1
2
.
The C
ab
i
are the structure constants of SU(4)
R
. Note that there is only one coupling constant
g. On the classical level this theory is conformal with engeneering dimensions of the elds as
[A
] = 1, [] = 3/2, [X] = 1 and therefore [g] = 0. The dimensionless coupling and absence of
any mass term are necessary for conformal invariance.
The Lagrangian (2.2.22) is invariant under SUSY-transformations given by
(X
i
)
a
=
_
Q
a
, X
i
= C
iab
b
,
(
b
)
a
=
_
Q
a
,
b
_
= F
+

a
b
+
_
X
i
, X
j
(C
ij
)
a
b
(
)
a
=
_
Q
a
_
= C
ab
i

X
i
(A
)
a
=
_
Q
a
, A
a
.
(2.2.23)
Note that F
+
is the self-dual part

1
2
(F
+

F
) of the eld strength, and the constants (C

ij
)
a
b
are related to bilinears in Cliord Dirac matrices of SO(6)
R
.
Upon quantization of this theory, one nds that the -function vanishes to all orders of pertu-
bation theory (and even non-perturbatively), therefore we are left with a CFT even at quantum
level.
2.2.5 The superconformal algebra and its representations
The concept of supersymmetry together with the conformal group form the superconformal
group SU(2, 2[4). The SU(2, 2) part represents the symmetry of the Weyl spinors while the
SU(4) refers to the R symmetry group SU(4)
R
of the ^ = 4 supersymmetry.
The AdS/CFT map will provide a direct one to one mapping between operators on both sides
of the correspondence. This relies heavily on the fact that on both sides the operators fall into
representations of the same symmetry groups.
The generators of the superconformal group are given by
Conformal symmetry with generators P
, L
, D, K
: In addition to the Poincare algebra

(2.2.18) and (2.2.19), the conformal algebra involves commutators
_
D, P
= i P
_
D, K
= i K
_
L
, D
= 0
_
L
, K
= i (
)
_
P
, K
= 2i L
2i
D
(2.2.24)
R symmetry SO(6)
R

= SU(4)
R
with generators T
A
, A = 1, 2, . . . , 15. The SO(4, 2)- and
SU(4)
R
subgroups commute.
Poincare supersymmetry with generators Q
a
,

Q
a

, a = 1, 2, 3, 4 subject to (2.2.21).
Conformal supersymmetry generators S
a
and

S
a
which introduce the following anti-
commutation relations:
_
Q
a
, Q
b
_
=
_
S
a
, S
b
_
=
_
Q
a
,

S
b
_
= 0
_
Q
a
,

Q
b
_
= 2 (
a
b
_
S
a
,

S
b
_
= 2 (
a
b
_
Q
a
, S
b
_
=

a
b
D +
1
2

a
b
L
(2.2.25)
Central charges are assumed to vanish throughout the rest of these lectures.
The elds A
(x),
a
(x),

a

(x) and X
i
(x) of the SUSY multiplet (a = 1, 2, 3, 4 and i = 1, 2, ..., 6)
can be used to construct composite operators of ^ = 4 SYM. Some regularization prescription
is needed when multiplying elds at the same spacetime point.
We dene a superconformal primary operator O by
_
S , O
= 0 , (2.2.26) 2,40
i.e. the Os are the lowest dimensional operators in a representation of SU(2, 2[4). This is the
generalization of the primary operator condition [K
, O] = 0 in bosonic conformal eld theory

(which is in fact implied by (2.2.26) since two S generators anticommute to Ks).
An operator O
is a superconformal descendant of O if
O
=
_
Q, O
, (2.2.27)
O and O
then belong to the same superconformal multiplet, i.e. the same representation of
SU(2, 2[4). The scale dimension is shifted as
O
=
O
+
1
2
.
Of central importance are single trace operators (taking a trace is necessary to ensure gauge
invariance)
O = Tr
_
X
(i
1
X
i
2
... X
i
n
)
_
= sTr
_
X
i
1
X
i
2
... X
i
n
_
(2.2.28)
They are also referred to as half BPS states since they are annihilated by half the spinorial
generators S (but not by the other half Q).
2.3 Anti-de Sitter space
In this section we will examine the Anti-de Sitter spacetime and compare it to at Minkowski
spacetime. As mentioned earlier, one side of the AdS/CFT correspondence is so-called type
IIB string theory formulated on the spacetime AdS
5
S
5
. We will not discuss string theory
2.3. ANTI-DE SITTER SPACE 21
now. Instead we want to get familiar with the spacetime and see how it may be connected to
the more familiar Minkowski spacetime R
1,3
.
The most important facts about AdS
5
S
5
spacetime for us are of geometrical nature. We
already stated that the isometry group of this spacetime is the same as the symmetry group of
the quantum eld theory on the other side of the correspondence.
The key result of this section will be that the boundary of the Euclidian compactication
of AdS
5
spacetime is equal to compactied R
4
, which is the Euclidean compactication of
the Minkowski spacetime we live in. To see this equivalence we will make use of so called
conformal diagrams which enable us to draw an image of the entire spacetime on a single sheet
of paper making the causal structure of the spacetime visible. A short introduction to conformal
diagrams is for example given in appendix H of [8].
The (p + 2)-dimensional version AdS
p+2
of this spacetime can be dened as the embedding of
a hyperboloid (with AdS radius L)
X
2
0
+ X
2
p+2

p+1
i=1
X
2
i
= L
2
(2.3.1)
into a at (p + 3)-dimensional space R
p+3
with metric
ds
2
= dX
2
0
dX
2
p+2
+
p+1
i=1
dX
2
i
. (2.3.2) eq:adsEmbedMetric
The AdS radius is a measure for the constant curvature: Riemann tensor and cosmological
constant are given by
R
=
1
L
2
_
g
_
, =
d (d 1)
L
2
< 0 (2.3.3)
where d is the dimension of the boundary.
One possible parametrization of this spacetime is given by
X
0
= L cosh cos
X
p+2
= L cosh sin
X
i
= L
i
sinh
(2.3.4) eq:adsParamGlobal
with angular coordinates
i
, i = 1, . . . , p+1 such that
i

2
i
= 1 and ranges 0 , 0 < 2
for the remaining coordinates.
Inserted into (2.3.2), this yields the metric
ds
2
= L
2
_
cosh
2
d
2
+ d
2
+ sinh
2
d
2
p
_
. (2.3.5) eq:adsMetricGlobal
It features a timelike killing vector
on the whole manifold, so may be called the global

time coordinate. The isometry group SO(2, p +1) of AdS
p+2
has a maximal compact subgroup
SO(2) SO(p + 1), the former generating translations in , the latter rotating the X
i
s.
picture of AdS as hyperboloid?
Near = 0 we have cosh 1 and sinh , so in this environment the metric of AdS
5
looks like
ds
2
L
2
_
d
2
+ d
2
+
2
d
2
3
_
(2.3.6)
and thus is seen to be topologically S
1
R
4
. The S
1
parametrized by the time coordinate
represents closed timelike curves. To prevent inconsistencies concerning causality, AdS
5
is therefore regarded as the causal spacetime obtained by unwrapping these circles, taking
< < without any identication.
Introducing a new coordinate , the metric (2.3.5) becomes that of the Einstein static universe
R S
p
:
tan = sinh ds
2
=
L
2
cos
2
_
d
2
+ d
2
+ sin
2
d
2
3
_
. (2.3.7)
However, since 0 <

2
, only half of R S
p
. The causal structure remains unchanged
when scaling this metric to get rid of the overall factor. Further, adding the point =

2
corresponding to spatial innity results in the compactied spacetime
ds
2
= d
2
+ d
2
+ sin
2
d
2
3
, 0

2
, < < . (2.3.8) eq:adsCompact
If we specify boundary conditions on RS
p
at =

2
, then the Cauchy problem is well-posed.
As one can easily read o from (2.3.8), the =

2
boundary of conformally compactied AdS
p+2
is identical to the conformal compactication of (p + 1) dimensional Minkowski spacetime.
Let us take a quick look at the special case of conformally compactied (1 + 1) dimensional
Minkowski spacetime. It is convenient to introduce light cone coordinates,
u
:= t x ds
2
= dt
2
+ dx
2
= du
+
du
. (2.3.9)
If we furthermore restrict the coordinates to a nite range, a useful choice is
u
=: tan u
, u
=:

2
ds
2
=
d
2
+ d
2
4 cos
2
u
+
cos
2
u
. (2.3.10)
Another neat parametrization of AdS
p+2
are the Poincare coordinates which cover half of the
2.3. ANTI-DE SITTER SPACE 23
hyperboloid. Introduce (y, t, x) such that y > 0 and x R
p
, then:
X
0
=
1
2y
_
1 + y
2
(L
2
+ x
2
t
2
)
_
X
p+1
=
1
2y
_
1 y
2
(L
2
x
2
+ t
2
)
_
X
p+2
= Ly t
X
i
= Ly x
i
(2.3.11)
The boundary at y can be better analyzed in terms of a new variable u
u :=
1
y
ds
2
= L
2
_
du
2
u
2
+
1
u
2

ij
dx
i
dx
j
_
. (2.3.12) 2,last
After a conformal rescaling by u
2
, we obtain the Minkowski metric by freezing u = 0.
Chapter 3
Introduction to superstring theory
This chapter aims to give a brief introduction to selected aspects of superstring theory. Of
course, we cannot provide a self-contained course about this topic, the following sections will
shed light only on those aspects which are relevant for the AdS/CFT correspondence.
As we have emphasized before, the AdS/CFT map relates the ^ = 4 SYM eld theory to string
theory. Relations of that type have been known for some time, in fact the original motivation to
study string theory in the 1960s was to describe mesons and hadrons (bound states of quarks).
This picture gives a relation between mass m and spin J of hadrons, m
2
= J/
+ const. The
m
2
(J) plot is known as Regge trajectory and the parameter
as Regge slope. Mass and angular

momentum are assumed to come from a rotating relativistic string.
3.1 Bosonic strings in Minkowski spacetime
The basic idea behind the bosonic string is to take one-dimensional strings as the fundamental
objects rather than point particles. Such a string sweeps out a 1+1 dimensional worldsheet (cf.
worldline of point particles). Strings can be closed or open closed string will represent the
gravity side of the correspondence whereas open strings will cover the gauge sector.
The worldsheet is parametrized by two coordinates, proper time and the spatial extent of
the string. The embedding of the worldsheet of the fundamental string into the target spacetime
is dened by functions X
(, ).
The string action is simply given by the worldsheet area (similar to the length of a point
particles worldline),
o =
1
2
_
d
2
_
det
, (3.1.1) 3,1
where (
0
,
1
) (, ) and
is the inverse string tension.

25
26 CHAPTER 3. INTRODUCTION TO SUPERSTRING THEORY
In order to get rid of the square root in view of quatization, a worldsheet metric h
() is
introduced as an auxiliary eld subject to certain constraints. This gives rise to the Polyakov
action
o =
1
4
_
d
2
hh
. (3.1.2) 3,2
The equation of motion for X
is a relativistic wave equation. In the gauge with h
=
diag(1, 1), it takes the particularly simple form
(
2
) X
(, ) = 0 . (3.1.3) 3,3
This is supplemented by the Virasoro constraints
= 0 . (3.1.4) 3,4
3.1.1 Closed strings in Minkowski spacetime
For AdS/CFT, it is essential to have two dierent types of strings, closed and open ones,
depending on the boundary conditions they satisfy. Closed strings are equivalent to a circle,
they do not have endpoints and satisfy periodic boundary conditions. With [0, 2[ we have
X
(, 0) = X
(, 2) ,
(, 0) =
(, 2) (3.1.5) 3,5
and also h
(, 0) = h
(, 2). The mode expansion for the closed string is governed by the
solutions to the wave equation (3.1.3) which split into left- and right movers
X
(, ) = X
L
( +) + X
R
( ) . (3.1.6) 3,6
The periodic boundary conditions give rise to a discrete Fourier expansion
X
L
( +) =
x
0
2
+
L
( +) + i
_
n=0

n
n
e
in(+)
X
R
( ) =
x
0
2
+
R
( ) + i
_
n=0
n
n
e
in()
(3.1.7) 3,7
The x
0
and p
= p
L
= p
R
are center of mass positions and -momenta, the latter can be viewed
as the zero modes of the expansion via
0
=
0
=
_
2
p
. The constraint p
= p
L
= p
R
is
enforced by periodicity. Reality of X
requires
n
= (
n
)
and
n
= (
n
)
.
In the quantization procedure, the
n
modes become creation- and annihilation operators, e.g.
the graviton as the lowest closed string excitation corresponds to the massless spin 2 state
1
[0, 0, k).
3.2. BOSONIC STRING THEORY IN BACKGROUND FIELDS 27
3.1.2 Open strings in Minkowski spacetime
Open strings have two endpoints. The usual convention is to delimit [0, [ in this sector. In
each direction of spacetime, either Neumann- or Dirichlet boundary conditions are possible:
Neumann boundary conditions
(, 0) =
(, ) = 0 (3.1.8) 3,8
Momentum ow through the endpoints of the string is forbidden by Neumann boundary
conditions. This is reected in the Neumann mode expansion
X
(, ) = x
0
+ 2
+ i
n=0
n
n
e
in
cos(n) . (3.1.9) 3,9
Left- and right movers are reected into each other. Again,
0
=
.
Dirichlet boundary conditions
Now the endpoints of the string are xed
X
(, 0) = X
(, ) = x
0
, (3.1.10) 3,10
which gives rise to the mode expansion
X
(, ) = x
0
+
n=0
n
n
e
in
sin(n) . (3.1.11) 3,11
Open string boundary conditions (b.c.) can be interpreted as follows: Dirichlet b.c. dene a
hyperplane in target space, so-called Dp branes, on which open strings can end. In p spatial
dimensions and in the time direction, Dirichlet b.c. are used whereas in the other directions
Neumann b.c. are imposed.
The quantization procedure naturally leads to a massless spin 1 state
1
[0, k), the photon.
3.2 Bosonic string theory in background elds
Up to now, we have considered the propagation of open and closed strings in Minkowski space-
time. By coupling the fundamental string to the massless closed string excitations (which in-
volve the graviton), strings propagating through curved background spacetime (such as AdS
5
S
5
) can be described. In particular, the symmetric traceless part of the state
1
[0, 0, k)
g
can be identied with the metric of the target spacetime.

3.2.1 Background elds of the closed string sector
sec:3,1
Weyl invariance of string theory implies that spacetime has to satisfy the vacuum Einstein
equations. The Polyakov action becomes
o =
1
4
_
d
2
hh
(X) . (3.2.1) 3,12

In addition, we have a Kalb Ramond eld B
[]
and a dilaton associated with the remaining
irreducibles
[
1

]
1
[0, 0, k) and
1

1
[0, 0, k). Their action reads
o
B,
=
1
4
_
d
2
h
_
i
(X) +
R
h
(X)
_
. (3.2.2) 3,13
By comparison with the string theory perturbative expansion, we nd a string coupling g
s
= e
.
To ensure Weyl invariance of the quantized theory, we have to impose tracelessness of the
worldsheet energy momentum tensor. In bosonic string theory, this is possible in D = 26
spacetime dimensions only. The critical dimension of superstring theory is D = 10. The
worldsheet energy momentum trace reads
T
=
1
2
i
2
1
2

R
h
(3.2.3)
where R
h
denotes the Ricci scalar on the worldsheet (with respect to the metric h
) and the
functions are given as follows (to order
):
_
(1
g
)

1
4
H
1
2

1
2

2
+

1
2
H
_
.
(3.2.4) 3,16
(By the usual method of dierential forms, one denes a eld strength H = dB for the Kalb
Ramond eld,
H
:=
. ) (3.2.5) 3,14
The theory is Weyl invariant if
g
=
B
= 0. Remarkably, the vanishing of the

functions (3.2.4) may be derived as equations of motion from the target spacetime action
o =
1
2
0
_
d
26
X
_
[ det g[ e
2
_
1
g
+ 4

1
12
H
_
. (3.2.6) 3,15
with spacetime Ricci scalar 1
g
built from the g
metric. This is the eective action for the

massless string states
1
[0, 0, k) g
, B
, of the closed string sector. From the form

of o, the g
eld may be identied with the target spacetime metric.

3.3. SUPERSTRING THEORY 29
3.2.2 Background elds of the open string sector
Similarly, we may couple the open string to an abelian gauge eld living on a D brane. This is
achieved via worldsheet action
o
A
=
_
d A
(X)
(3.2.7) 3,17
where is the boundary of the worldsheet. The eective spacetime action for the open string
sector (to leading order in
) is given by
o = C
_
d
26
X e
. (3.2.8) 3,18
Therefore, the tree level open string physics is described by Yang Mills theory. Recall that
can be interpreted as the squared string length, so the
0 limit extracts the point

particle-like behaviour.
A single D brane gives rise to the gauge group U(1), but this can be generalized to non-abelian
symmetry by taking a stack of coinciding D branes. Superposition of branes introduces non-
dynamical degrees of freedom (from the worldsheet point of view) called Chan Paton factors.
They arise on a stack of N Dp branes and are therefore assigned to the endpoints of the string.
The Chan Paton factor
ij
labels strings stretching from brane i to j where i, j = 1, 2, ..., N. The
matrix is an element of some Lie algebra. It turns out that the only Lie algebras consistent
with open string scattering amplitudes is U(N). Note that the Chan Paton degrees of freedom
parametrize a global symmetry on the worldsheet but a local symmetry in target spacetime.
The theory of open strings ending on coincident Dp branes can eectively by described by a
non-abelian gauge theory.
3.3 Superstring theory
Bosonic string theory which we have described so far has two major shortcomings. Firstly,
it contains tachyons in both the open string- and the closed string sector which are states of
negative mass square. Secondly, the bosonic string lacks fermionic degrees of freedom necessary
to model particles observed in nature.
Let us give the supersymmetrized Polyakov action for the string position X
and its worldsheet

superpartner
in conformal gauge h
(, ) = e
(,)
:
o =
1
4
_
d
2
+ i
_
(3.3.1) 3,19
The
are spacetime vectors of two components spinors on the worldsheet,
=
_
+
_
T
with real entries
. The
denote worldsheet matrices for which one possible representation

is
0
=
_
_
0 1
1 0
_
_
,
1
=
_
_
0 1
1 0
_
_
. (3.3.2) 3,20
In terms of lightcone derivatives
:=

with
= , the fermionic part of the action

(3.3.1) may be rewritten as
o
f
=
i
2
_
d
2
+
_
, (3.3.3) 3,21
the equations of motion describe left- and right moving waves just like in the bosonic sector,
+
= 0 . (3.3.4) 3,22
The total action is invariant under the worldsheet supersymmetry transformations
and
where the parameter is an innitesimal constant Majorana spinor.

3.3.1 Open superstrings
Upon integrating the action (3.3.3) by parts, one encounters the boundary term
o
f
=
i
4
_
d
_
+
_
=
=0
. (3.3.5) 3,23
In the open string sector, we have to impose that the contributions from = 0 and =
vanish separately. This is equivalent to
=0,
= 0
_
+
_
2
=0,
=
_
+
_
2
=0,
= 0 . (3.3.6) 3,24
Since the overall sign of the spinor components can be chosen arbitrarily, we impose
+
(, 0) =
(, 0), then the boundary condition at = leaves two options corresponding to the Neveu
Schwarz- and the Ramond sector of the theory:
R :
+
(, ) = +
(, )
NS :
+
(, ) =
(, )
(3.3.7) 3,25
These boundary conditions give rise to the Fourier expansions
R :
(, ) =
1
nZ
d
n
e
in
NS :
(, ) =
1
rZ
1
2
b
r
e
ir
(3.3.8) 3,26
with Grassmann valued modes d
n
, b
r
. The string states are created by acting on the ground
state of the NS- and R sectors with creation operators. The NS ground state is tachyonic and
3.4. D BRANES 31
will be removed from the spectrum. The spectrum of both NS- and R sector can be truncated
in a specic way which eliminates the tachyons. This truncation prescription is called GSO
projection due to Gliozzi, Scherk and Olive. This projection leaves an equal number of fermions
and bosons at each mass level and therefore paves the way for spacetime supersymmetry.
3.3.2 Closed superstrings
The closed sector of superstring theory can be constructed in four dierent ways. Each of left-
and right movers may be taken from open string NS- or R sectors. From spacetime point of
view, we nd the following statistics for the states:
NS-NS, R-R sectors spacetime bosons
NS-R, R-NS sectors spacetime fermions
The NS-NS sector contains the elds g
, B
, which we had already discussed in bosonic

string theory whereas the mixed NS-R, R-NS sectors contain SUSY superpartners such as
gravitino and dilatino.
The R-R sector is more complicated due to the degenerate ground state. There are two possible
inequivalent R-R ground states (which dier by chirality), corresponding to type IIA- and type
II B superstring theory. In type IIB, left- and right moving sectors have the same chirality, this
leads to a scalar C
0
and antisymmetric tensor elds C
2
and C
4
of rank 2 and 4 at the massless
level. Type IIA (with R-R ground states of opposite chiralities) gives rise to C
1
, C
3
tensor
elds.
The Dp branes to which the C
p
forms couple of course also dier between the two theories
type II B D1, D3, D5, D7, D9 branes
type II A D0, D2, D4, D6, D8 branes
where the D3 branes play a major role in the AdS/CFT correspondence. Although types IIA/B
are inequivalent theories, they are related by dualities.
As it was shown in the 90s by Witten, there in fact three further consistent superstring theories
known as type I and heterotic string theories (with gauge groups SO(32) and E
8
E
8
). They
are connected with each other and the type II models by a web of dualities. For AdS/CFT
purposes, however, it is sucient to focus on type II.
3.4 D branes
D branes have a dual interpretation which is crucial for the AdS/CFT correspondence:
hyperplanes where open strings can end
To lowest order in
, massless excitations of D branes are described by supersymmetric

Yang Mills theory (with gauge group SU(N) in presence of N branes).
solitonic solutions of type IIB supergravity in D = 10 dimensions
D branes are very massive and curve spacetime around them. The lowest energy closed
string excitations are gravitons.
The AdS/CFT is based on the identication of these two pictures in a particular limit!
3.4.1 Eective actions for D branes
Just as fundamental strings, D branes can couple to background elds, in particular to gravity.
We aim to nd a world volume action describing their dynamics as a generalization of the
worldsheet action for strings. The background elds act as generalized couplings.
Let
a
denote the coordinates for the world volume of a Dp brane (which reduces to
0
=
and
1
= in case of the fundamental string). In direct analogy to the string worldsheet area
action, the bosonic part of the D brane action is given by
o
(p)
DBI
=
p
_
d
p+1
e
_
det
_
g
ab
+ B
ab
+ 2
F
ab
_
. (3.4.1) 3,27
The action (3.4.1) is known as Dirac Born Infeld action, or in short, DBI action. Its prefactor
p
= (2)
p
(p+1)/2
relates to the (genuinely non-perturbative) brane tension T
p
=
p
/g
s
,
and g
is the induced metric on the brane obtained via pullback of the spacetime metric to the
brane worldvolume,
g
ab
=
X
a
X
b
g
. (3.4.2) 3,28
The same applies to the B eld.
Expanding the DBI action in at spacetime (with g
ab
=
ab
) by means of det(1 + M) =
1
1
4
TrM
2
for antisymmetric matrices M, we see that the DBI action for D3 branes is a
generalization of Yang Mills theory
o
(p=3)
DBI

2
_
d
4
Tr
_
T
ab
T
ab
_
, T
ab
= B
ab
+ 2
F
ab
. (3.4.3) 3,29
D branes also carry some charge under the R-R p form elds C
p
. The full action describing
a charged BPS brane (named after Bogomolnyi, Prasad and Sommerfeld) involves a Chern
Simons term, o = o
DBI
o
CS
,
o
CS
=
p
_
d
p+1
q
C
q+1
Tr
_
e
F
_
, (3.4.4) 3,30
3.4. D BRANES 33
it describes the interaction of the R-R elds C
q+1
with the NS-NS eld B. The exponential of
the two form T has to be understood in terms of the wedge product.
BPS branes are stable due to charge conservation. In type IIA/B superstring theory, Dp branes
with p even/odd are BPS stable since R-R gauge potentials C
p+1
are present to which Dp branes
can couple. Unlike fundamental strings, D branes are non-perturbative objects since the tension
and therefore their energy scales as 1/g
s
, i.e. with the inverse string coupling.
3.4.2 D branes in supergravity
sec:Dsugra
We have discussed in subsection 3.2.1 that to leading order in
(i.e. at low energies when

only massless excitations contribute), Weyl invariance of the string worldsheet action in curved
background is equivalent to certain eld equations which can be derived from a gravity action.
In superstring theory, this eective target space action is precisely that of supergravity. For this
reason, the supergravity theories are referred to as type IIA/B although they can be motivated
independent of string theory.
In type IIB supergravity, the bosonic eld consists of the massless closed string states, g
, B
and from the NS-NS sector and the form the R-R form elds C
0
, C
2
and C
4
. In addition,
there are fermions with an equal number of degrees of freedom as in the bosonic part.
Moreover, we dene the axio-dilaton and a complex 3 form G
3
by
:= C
0
+ i e
, G
3
:= F
3
H
3
(3.4.5) 3,31
where F
3
, H
3
are the eld strengths of C
2
and B
2
(in dierential form notation F
3
= dC
2
and
H
3
= dB
2
). The C
4
potential is more conveniently represented by the eld strength
F
5
= dC
4
+
1
2
B
2
F
3

1
2
C
2
H
3
. (3.4.6) 3,32
Let us nally introduce the rescalings g
= e
(
0
)/6
and =
0
e
0
=
8G
N
into the Einstein
frame, then the type IIB supergravity action is given by
o
IIB
=
1
2
2
_
d
10
x
_
g
_
1
g

[
[
2
2 (Im)
2

[G
3
[
2
12 Im

[
F
5
[
2
4 5!
_
+
1
8i
2
_
C
4
G
3

G
3
Im
. (3.4.7) 3,33
The eld strength

F
5
has to be self-dual in the sense that
(F)
1
...
5
= F
1
...
5
(3.4.8) 3,34
where the Hodge dual ()
k
of a k form in D dimensions is dened by
()
1
...
Dk
=
[ det g[
k!

1
...
k
1
...
Dk

1
...
k
, (3.4.9) 3,35
e.g. F
=
| det g|
2

in D = 4 dimensions.
Now let us look for solitonic solutions of the equations of motion due to (3.4.7). A Dp brane
is a BPS solution of 10 dimensional supergravity, i.e. it is annihilated by half the Poincare
supercharges Q
. It has a p + 1 dimensional at hypersurface with Poincare invariance group

R
p+1
SO(1, p). The transverse space is then of dimension D p 1.
Here, it would be nice to have a few more words about the BPS term and about the
relation between SUSY conditions and eq. of motion...
A p brane in 10 dimensions has symmetries R
p+1
SO(1, p) SO(9 p). An ansatz which
solves the equations of motion of type IIB supergravity is
ds
2
=
1
_
H(y)
dx
dx
+
_
H(y) dy dy (3.4.10) 3,36
where x
are the coordinates on the brane world volume and y denote the coordinates perpen-
dicular to the brane. It turns out by means of the supergravity equations of motion that
e
(y)
=
_
H(y)
3p
4
, H harmonic function of y =
_
y y . (3.4.11) 3,37
Far away from the brane, i.e. at y , at space has to be recovered, this boundary condition
uniquely xes H to be
H(y) = 1 +
_
L
y
_
Dp3
. (3.4.12) 3,38
L is a length scale related to the only dimensionful parameter
. For a stack of N coincident

Dp branes, one nds
L
Dp3
= N g
s
(4)
(5p)/2
_
7p
2
_

(Dp3)/2
. (3.4.13) 3,39
Let us nally summarize the special features of D3 branes:
its worldvolume has 1+3 dimensional Poincare invariance
axion- and dilaton elds (C
0
, ) are constant with relation to the coupling g
2
YM
= g
s
= e
it is a regular supergravity solution for y 0

it couples to a self-dual ve form dC
4
= F
5
= F
5
string theory implies (since g
2
YM
= g
s
) that
L
4
= 4 g
s
N
2
= 4
2
, = g
s
N tHooft coupling (3.4.14) 3,42
Chapter 4
The AdS/CFT correspondence
4.1 Maldacenas original argument
Following the arguments of [5], let us consider type IIB string theory in 9+1 dimensional
spacetime with a stack of N D3 branes. There are two kinds of excitations:
closed strings: excitations of empty space with the graviton as the massless mode
open strings ending on the D3 branes: excitations of D branes
At energies below the string mass scale (
)
1/2
, only massless string states are excited:
massless closed string states gravity multiplet of type IIB supergravity
massless open strings states ^ = 4 vector multiplet in 3+1 dimensions, SU(N) SYM
4.1.1 D3 branes from the open string point of view
The low energy eective action for the massless excitations of N D3 branes in at ten dimen-
sional space has the schematic form
o = o
bulk
+ o
brane
+ o
int
o
bulk
D = 10 supergravity including higher derivative terms, i.e.
corrections
o
brane
DBI- and CS action dened on 3+1 dimensional brane world volume:
for small
, we get SYM TrF
plus interactions
TrF
4
+...
o
int
bulk-brane interaction: leading term is appearance of the background
metric g in the brane action
35
36 CHAPTER 4. THE ADS/CFT CORRESPONDENCE
For
0, the bulk action becomes the Einstein Hilbert action with coupling g
s
2
. In
the expansion g
+h
about at space (with Minkowski metric ), the leading terms

are
o
bulk
=
1
2
2
_
d
10
x
_
[g[ 1
g

_
d
10
x
_
(h)
2
+ (h)
2
h + ...
_
. (4.1.1) 3,43
In the low energy limit g
s
2
0, the interaction terms O() drop out, so gravity becomes
free at long distances. Similar behaviour can be observed in the o
int
sector. The term low
energy limit should not be taken too literally: the relevant energies E are certainly kept xed
but we send the dimensionful parameter
0, therefore various dimensionless quantities such

as
E
2
are suppressed.
4.1.2 D3 branes from the closed string point of view
In their solitonic interpretation, D branes are viewed as massive charged objects which act as
sources for the various supergravity elds. Specializing (3.4.12) to D3 branes in ten dimensions
(D = 10 and p = 3) yields the metric
ds
2
=
1
_
H(y)
dx
dx
+
_
H(y)
_
dy
2
+ y
2
d
2
5
_
H(y) = 1 +
_
L
y
_
4
. (4.1.2) 3,44
Let us discuss the limits of this metric: When y
4
L
4
= 4g
s
N
2
, one recovers at 10 D
space. When y < L, on the other hand, the metric appears to be singular as y 0. To
examine this limit more carefully, let us dene a new coordinate u := L
2
/y. In the limit of
large u (where H = 1 +u
4
/L
4
u
4
/L
4
), the metric takes the asymptotic form
ds
2
u
= L
2
_
1
u
2

dx
dx
+
du
2
u
2
+ d
2
5
_
. (4.1.3) 3,41
In this near horizon limit y 0 u L, the geometry close to the brane is regular and highly
symmetrical (with isometry group SO(4, 2) SO(6)). Apart from the S
5
sphere represented
by d
2
5
, we rediscover the AdS
5
metric (2.3.12).
An important property of the metric (4.1.2) is its non-constant redshift factor
_
H(y)
_
1/4
= g
tt
with an interesting near horizon limit:
_
H(y)
_
1/4
=
_
1 + L
4
/y
4
_
1/4
=
_
_
_
1 : large y
y/L : small y
(4.1.4) 3,44a
The energy E
p
of an object measured by an observer at constant position y diers from the
energy E
i
of the same object, this time measured by an observer at innity,
_
H(y)
_
1/4
E
p
= E
i
. (4.1.5) 3,45
4.1. MALDACENAS ORIGINAL ARGUMENT 37
When the object approaches y 0, it appears to have lower and lower energy to the observer
at innity. This gives another, geometric notion of low energy regime. We have to distinguish
two kinds of low energy excitations:
particles approaching y 0
massless particles propagating in the bulk (away from y = 0)
Their excitations decouple from each other in the low energy limit: Bulk massless particles
decouple from the near horizon region around y 0. Excitations close to y = 0 are trapped
by the gravitational potential to the AdS
5
S
5
region.
4.1.3 Dierent forms of the AdS/CFT correspondence
Both from the point of view of open strings eld theory limit and from the supergravity
point of view, there are two decoupled theories in the low energy regime. One of them is free
supergravity in at space, and we are led to identify it with the supersymmetric gauge theory
which appears in both descriptions:
^ = 4 SYM with gauge group SU(N)
()
type IIB supergravity
The () above the arrow indicates that the correspondence claimed in this AdS/CFT conjecture
holds in the N limit at large and xed tHooft coupling = g
s
N. Maldacena generalized
this idea to conjecture that the duality goes beyond the supergravity approximation.
The strongest form of the AdS/CFT correspondence conjectures that the duality between the
supersymmetric SU(N) gauge theory and type IIB supergravity holds for any value of N and
g
s
. This implies that ^ = 4 SYM is exactly equivalent to the full type IIB superstring theory
on AdS
5
S
5
. However, it is at present not possible to test the strongest form since there is
no consistent non-perturbative quantization of string theory yet, in particular not in curved
spacetime.
In the (modestly) strong form of the AdS/CFT conjecture, one keeps = g
s
N xed while
sending N . In this case the ground state is classical type IIB string theory on AdS
5
S
5
. The perturbative expansion parameter is g
s
= /N 1 on the string theory side, this
corresponds to a perturbative 1/N expansion on the eld theory side.
Finally, there is the weak form of the AdS/CFT conjecture described above. It states that the
correspondence is only valid in the Maldacena limit N and very large. It relates ^ = 4
SYM at strong coupling and N with classical supergravity. In contrast to previous forms,
is assumed to be small now, and the
expansion of supergravity is dual to a eld theory

expansion in
1/2
powers around the strong coupling limit.
weak form of AdS/CFT correspondence :
_
_

N
_
_

_
_
g
s
0
0
_
_
The AdS/CFT map provides a weak/strong coupling duality:
more complicated to test: only direct tests based on objects which are independent of the
coupling
interesting predictive power: non-trivial prediction for strongly coupled gauge theories
4.2 Field operator map
The aim of this section is to work out the precise dictionary between objects of the two equiv-
alent theories,
_
_
^ = 4 SYM
N,
_
_

_
_
type IIB supergravity
on AdS
5
S
5
_
_
in particular between representations of the common symmetry groups. We will relate eld
theory operators to supergravity elds which transform in the same representation of the su-
perconformal group SU(2, 2[4) or its bosonic subgroup SO(6) SO(4, 2). This provides a
one-to-one map between gauge invariant operators in ^ = 4 SYM and classical elds in IIB
supergravity on AdS
5
S
5
.
4.2.1 CFT correlation functions
A crucial role in testing the AdS/CFT correspondence is played by the computation and com-
parison of correlation functions. Correlators which obey non-renormalization theorems (i.e.
which are independent) will be of particular interest. Let us give a brief review of correlation
functions in QFT.
Composite operators with coinciding arguments such as
_
(x)
_
2
require regularization, the
regularized version will be denoted by
_
2
(x)
. Consider an n point function of composite

regularized gauge invariant operators O
k
(x),
O
1
(x
1
) O
2
(x
2
) ... O
n
(x
n
)) .
4.2. FIELD OPERATOR MAP 39
An important tool to compute this correlator is the generating functional Z[J] (and its analogue
W[J] for connected diagrams) dened by
Z[J] :=
_
exp
_
_
d
D
x /
J
__
= e
W[J]
(4.2.1) 4,1
where /
J
is the Lagrangian of a given QFT with added source term coupled to a basis O
i
of
gauge invariant local operators:
/
J
= / +
i
J
i
O
i
(4.2.2) 4,2
The n point function is then given by
O
1
(x
1
) O
2
(x
2
) ... O
n
(x
n
)) =

n
ln Z[J]
J
1
(x
1
) J
2
(x
2
) ... J
n
(x
n
)
J
i
=0
. (4.2.3) 4,3
To calculate correlation functions in AdS
5
S
5
, it is convenient to work in Euclidean AdS
5
with Poincare coordinates
H :=
_
(z
0
, z), z
0
> 0, z R
4
_
, H = R
4
. (4.2.4) 4,4
The metric
ds
2
=
1
z
2
0
_
dz
2
0
+ dz
2
_
(4.2.5) 4,5
diverges at the boundary z
0
0, but it is merely a coordinate singularity, not a curvature
singularity. The divergence may be removed by a Weyl rescaling. As we will see later, however,
sometimes it is necessary (and useful) to consider a cuto at xed z
0
= . The UV cuto =
1
is mapped to an IR cuto in AdS. It is natural to assume that ^ = 4 SYM lives on the

boundary of AdS
5
.
Typical gauge invariant operators in SU(N) SYM with ^ = 4 in D = 4 are
O
(x) := sTr
_
X
i
1
X
i
2
... X
i
_
= N
(1)/2
C
i
1
...i
Tr
_
X
i
1
X
i
2
... X
i
_
. (4.2.6) 4,6
Here, denotes the conformal dimension of the operators, X
i
are the elementary scalar elds of
^ = 4 SYM transforming in the representation (6) of SO(6)
= SU(4) and C
i
1
...i
fall into the

totally symmetric rank tensor representation of SO(6). The trace is taken over color indices
(recall that all the elds transform in the adjoint representation of SU(N)). The normalization
is chosen such that all planar graphs scale with N
2
.
4.2.2 The dual elds of supergravity
On the AdS side, we decompose all elds into Kaluza Klein towers on S
5
, i.e. we expand the
elds in spherical harmonics Y
(y) of S
5
:
(z, y) =
=0
(z) Y
(y) (4.2.7) 4,7

The ten dimensional Klein Gordon equation implies a massive wave equation in the ve dimen-
sional AdS sector,
_
5
+ m
2
(z) = 0 , m
2
= ( 4) . (4.2.8) 4,8
It has two independent solutions which can be characterized by their asymptotics as z
0
0:
(z
0
, z)
_
_
_
z
0
: normalizable
z
4
0
: non-normalizable
(4.2.9) 4,9
The non-normalizable elds dene associated boundary elds [3] by virtue of

(z) := lim
z
0
0
(z
0
, z) z
4
0
. (4.2.10) 4,10
We may identify the normalizable AdS modes
as vacuum expectation values of the eld

theory operators O
and the non-normalizable modes
as sources for these operators:
(z
0
, z) O
) z
0
+
z
4
0
(4.2.11) 4,11
The mapping between correlation functions in SYM theory and the supergravity dynamics is
given as follows: The generating functional W[
] for all correlators of single trace operators

O
in SYM is given in terms of the source elds
. The boundary values of these supergravity

elds become the sources for the QFT. In other words, on the eld theory side we have
e
W[
]
=
_
exp
_
_
H
d
4
z
__
. (4.2.12) 4,12
The AdS side is governed by an action in terms of the bulk elds o[
] in the framework of type

IIB supergravity on AdS
5
S
5
. The AdS/CFT conjecture for correlation functions says that
precisely this classical gravity action enters the generating functional for the subclass O
of
operators in the ^ = 4 QFT. The AdS boundary conditions have to be adjusted to meet the
eld theory values of the source elds:
W[
] = o[
lim
z
0
0

(z
0
,z) z
4
0
=
(z)
(4.2.13) 4,13
The action o is the generating functional for tree diagrams on AdS space, i.e. for the classical
expansion of correlators. These tree level graphs in AdS are referred to as Witten diagrams [2],
let us give the corresponding Feynman rules:
Each external source
(z) is located at the boundary.

Propagators depart from the external sources either to another boundary point or to an
interior interaction point (in which case they are called bulk-to-boundary propagators)
The structure of the interior interaction points is governed by the interaction vertices of
the supergravity action. These are obtained from the Kaluza Klein reduction on S
5
.
Two interior interaction points may be connected by bulk-to-bulk propagators.
picture of 2pt, 3pt and 2 times 4pt (1 or 2 vertices)
4.2.3 AdS propagators
sec:adsprop
In this section we will derive the scalar propagator in Euclidean AdS spacetime H as dened
in (4.2.4). For simplicity, the AdS radius is set to L = 1. The four vector z in the metric
ds
2
=
1
z
2
0
(dz
2
0
+ dz
2
) parametrizes the boundary H. The geodesic distance is obtained by
solving the geodesic equation (where the parameter is called chordal distance):
d(z, w) =
w
_
z
ds = ln
_
1 +
_
1
2
_
, =
2 z
0
w
0
z
2
0
+ w
2
0
+ (z w)
2
(4.2.14) 4,15
Let us start from the scalar part of the action which we obtain by Kaluza Klein reduction of
the ten dimensional IIB supergravity on S
5
. Schematically we get
o[
] =
_
d
5
z
_
[g[
_
1
2
g
+
m
2
+ /
int
_
(4.2.15) 4,16
where /
int
denotes higher order interaction terms from KK reduction. Now the propagators
are represented by integral kernels K
, G
subject to
(z) =
_
H
d
4
x K
(z, x)
(x) bulk-to-boundary propagator (4.2.16) 4,17
(z) =
_
H
d
5
x G
(z, x) J(x) bulk-to-bulk propagator (4.2.17) 4,18

The scalar Green function satises
_
g
+ m
2
_
G
(z, x) =
5
(z, x)
5
i=1
(z
i
x
i
)
_
[g[
, m
2
= ( 4) (4.2.18) 4,19
where the action of the Laplacian
g
on scalar elds is in general given by
g
=
1
_
[g[
_
[g[ g
(4.2.19)
and reduces to the following expression in the AdS metric (??):
AdS
= z
2
0

2
0
+ (d 1) z
0
0
z
2
0
d
i=1
2
i
(4.2.20) 4,20
This turns (4.2.18) into a hypergeometric equation. The Green function which solves it is thus
given by a hypergeometric function in the argument from (4.2.14):
G
(z, w) = G
() =
C
(2d)

F
2,1
_
2
,
+1
2
, 1;
2
_
(4.2.21) 4,21
C
=
()
2
(2)
When x is located at the boundary, G
reduces to the bulk-to-boundary propagator

K
(z, x) = C
_
z
0
z
2
0
+ (z x)
2
_
. (4.2.22) 4,22
Calculation of the two point function requires careful treatment of potential divergences at the
boundary. We Fourier transform the boundary coordinates to momentum space. The d + 1
dimensional bulk action
o[] =
_
d
d+1
z
_
[g[
_
1
2

+
m
2
2

2
_
(4.2.23) 4,23
gives rise to a boundary term after integration by parts,
o[ ] =
1
2
d1
_
d
d
z (z)
0
(, z) , (4.2.24) 4,24
it is regularized by cutting o the z
0
integral at z
0
= . In the notation (, p) ( p)
d
, we
Fourier transform
(z
0
, z) =
_
d
d
p e
i pz
(z
0
, p) (4.2.25) 4,25
in order to simplify the equations of motion:
_
z
2
0

2
0
(d 1) z
0
0
( p
2
z
2
0
+ m
2
)
_
(z
0
, p) = 0 (4.2.26) 4,26
This is a Bessel equation with solutions z
d/2
0
K
(z
0
p) (where = d/2 and p =
p p). The
asymptotics is governed by lim
z
0
z
d/2
0
K
(z
0
p) = 0 and K
(z
0
0) z
d
0
.
The normalized solutions of the boundary problem read
(z
0
, p) =
z
d/2
0
K
(z
0
p)
d/2
K
(p)
( p)
d
. (4.2.27) 4,27
The rst term of the expansion of the supergravity action in correlation functions is
o
p
[ ] =
_
d
p
p d
d
q (2)
d
d
( p +q) (, p)
0
(, q) , (4.2.28)
This yields the following two point functions for the dual CFT operators:
O
( p) O
(q))
=

2
o
p
[ ]
( p) (q)
=
(2)
d
d
( p +q)
2d1
d
d
ln
_
d/2
K
( p)
_
(4.2.29)
The Bessel index is a positive integer whenever the associated CFT operator O
with =
+ d/2 is a chiral primary. Bessel functions have an asymptotic u 0 expansion of the
schematic form
K
(u) u
(a
0
+ a
1
u
2
+ a
2
u
4
+ ...) + u
ln u(b
0
+ b
1
u
2
+ b
2
u
4
+ ...) , (4.2.30) 4,30
this translates as follows to the level of two point functions:
O
( p) O
(q))
=
(2)
d
d
( p +q)
2d
_
d
2
+ (1 + c
2
+
2
p
2
+ c
4
4
p
4
+ ...)
2 b
0
a
0
2
p
2
ln(p) (1 + a
2
2
p
2
+ ...)
_
(4.2.31) 4,31
Explicitly, we have
2b
0
a
0
=
(1)
+1
2
2(1)
()
2
and
2
=
2d
such that
O
( p) O
( p))
=

0
+
1
2
p
2
+ ... +
(p)
2(1)
2d

2 b
0
a
0
p
2
ln(p) + O(
2
) .
(4.2.32) 4,32
The eld theory of the rst terms is governed by scheme dependent contact terms
m
d
(xy)
and the second term gives the correct non-local result
O
( p) O
( p)) =
2 b
0
a
0
p
2
ln(p) (4.2.33) 4,33
independent on . Transforming back to position space yields
O
( p) O
( p)) =
()
(d/2)
2 d
d/2
[x y[
2
. (4.2.34) 4,34
Chapter 5
Tests of the correspondence
5.1 Three point function of 1/2 BPS operators
An impressive test of the AdS/CFT correspondence is the agreement of the three point functions
of 1/2 BPS operators in ^ = 4 SYM at large N with the corresponding elds in supergravity.
To demonstrate this result, we will proceed as follows:
look at two point functions to x the normalization
calculate three point function in SYM to zeroth order in the coupling
check that this is not normalized at higher orders, i.e. prove a non-renormalization
theorem to show independence of the correlator on the coupling
calculate the correlation function on the gravity side (spacetime dependence from the
Green function and couplings from KK reduction)
5.1.1 Correlation functions of 1/2 BPS operators
For the purpose of this section, it is convenient to modify the notation: An 1/2 BPS operator
of ^ = 4 SYM will be denoted by
O
I
k
= C
I
i
1
...i
k
Tr
_
X
i
1
... X
i
k
_
(5.1.1) 5,1
where k and the C
I
are totally symmetric traceless rank k tensors of SO(6).
The SYM action is normalized such that g
2
YM
= 4g
s
, and the normalization TrT
a
T
b
=
ab
/2 of the SU(N) generators T
a
allows to recast it into the form
o =
1
2 g
2
YM
_
d
4
x Tr
_
F
_
+ SUSY completion
45
46 CHAPTER 5. TESTS OF THE CORRESPONDENCE
=
1
4 g
2
YM
_
d
4
x F
a
F
a
+ SUSY completion (5.1.2) 5,2
this gives rise to the following scalar propagators:
X
ia
(x) X
jb
(y)) =
g
2
YM
ij
ab
(2)
2
[x y[
2
(5.1.3) 5,3
The two point function on the eld theory side to lowest order in perturbation theory is therefore
given by
O
I
k
(x) O
J
k
(y)) = C
I
i
1
...i
k
C
J
j
1
...j
k
Tr
_
X
i
1
(x) ... X
i
k
(x)
_
Tr
_
X
j
1
(y) ... X
j
k
(y)
_
)
= C
I
i
1
...i
k
C
J
j
1
...j
k
N
k
g
2k
YM
_
i
1
j
1
i
2
j
2
...
i
k
j
k
+ cyclic permutations
_
(2)
2k
[x y[
2k
=
k
k
IJ
(2)
2k
[x y[
2k
(5.1.4) 5,4
line notation diagram
The last equality only holds to leading order in N.
By an appropriate generalization, one can obtain a nice result for the three point function to
lowest order in perturbation theory and in the limit of large N:
O
I
k
1
(x) O
J
k
2
(y) O
K
k
3
(z)) =

/2
k
1
k
2
k
3
C
I
C
J
C
K
)
N (2)
[x y[
2
3
[y z[
2
1
[x z[
2
2
(5.1.5) 5,5
Note that the spacetime dependence is completely determined by conformal invariance. We
have used shorthands
= k
1
+ k
2
+ k
3
,
i
=

2
k
i
(5.1.6) 5,6
(such that e.g.
1
=
k
2
+k
3
k
1
2
) and C
I
C
J
C
K
) denotes a uniquely dened SO(6) tensor con-
traction of indices determined by the Feynman graph.
anotherlinepicture
The notation can be streamlined by dening normalized operators

O
I
:=
(2)
k
k/2
k
O
I
. Their two
point function is normalized to one,
O
I
k
(x)

O
J
k
(y)) =

IJ
[x y[
2k
, (5.1.7) 5,7
and the three point function reads
O
I
k
(x)

O
J
k
(y)

O
K
k
(z)) =
k
1
k
2
k
3
C
I
C
J
C
K
)
N [x y[
2
3
[y z[
2
1
[x z[
2
2
. (5.1.8) 5,8
This holds for large values of N, otherwise non-planar corrections of order
1
N
2
arise.
5.1. THREE POINT FUNCTION OF 1/2 BPS OPERATORS 47
5.1.2 The non-renormalization theorem
Next we demonstrate the absence of O() terms both in OO) and in OOO). The argument
will hold for any N [9].
Dene complex scalar elds Z
i
:= X
i
+ iX
i+3
making use of the embedding SU(3) SU(4).
The Euclidean version of the ^ = 4, SU(N) SYM Lagrangian then reads
/ = Tr
_
1
4
F
+
1
2
/ D + D
Z
i
D

Z
i
+
1
2
i
/ D
i
+ i
2 g f
abc
_
a

Z
i
b
L
i
c

i
a
RZ
i
b
c
_

g
2
f
abc
ijk
_
i
a
LZ
j
a

k
c
+

i
a
R

Z
i
b
k
c
_
g
2
2
f
abc

Z
i
b
Z
i
c
f
ade

Z
jd
Z
je
+
g
2
2
f
abc
f
ade
ijk

ilm
Z
i
b
Z
k
c
Z
l
d
Z
m
e
_
(5.1.9) 5,9
where L, R denote the left- and right handed chirality projectors.
Due to supersymmetry, it is sucient to consider
Tr
_
(Z
1
)
k
(x)
_
Tr
_
(

Z
1
)
k
(y)
_
) =
P
k,k,0
(N)
_
4
2
[x y[
_
k
(5.1.10) 5,10
with the following polynomial in N
P
k,k,0
(N) =
S
k
Tr
_
T
a
1
T
a
2
... T
a
k
_
Tr
_
T
a
(1)
T
a
(2)
... T
a
(k)
_
= k
_
N
2
_
k
+ lower order in N . (5.1.11) 5,11
There are various eects to consider at leading order in the coupling:
self energy corrections
a blob a
=
aa
N A(x, y) G(x, y) (5.1.12) 5,12

with A(x, y) = a
0
+a
1
ln
_
2
(x y)
2
_
and the scalar propagator G(x, y) =
1
4
2
|xy|
2
two particle exchange interactions
(a to b) and (a
to b
) + quartic vertex =
_
f
pab
f
pa
+ f
pab
f
pa
b
_
B(x, y) G(x, y)
2
(5.1.13) 5,13
B(x, y) = b
0
+b
1
ln
_
2
(x y)
2
_
The possible corrections to the rainbow graph at order g
2
YM
schematically look like
sum up graphs
and it turns out that these three graphs cancel each other for all N and for all k. The proof
goes as follows:
use a trace identity valid for any matrices N and M
i
:
n
i=1
Tr
_
M
1
... M
i1
_
M
i
, N
M
i+1
... M
n
_
(5.1.14) 5,14
combinatorics for color indices
insert (5.1.12) between all pairs of adjacent lines using [T
a
, T
b
] = if
abc
T
c
result for all exchange graphs (with S
k
permutation ):
1
4
(2 B) Tr
_
T
a
1
... T
a
k
_
k
i=j=1
Tr
_
T
a
(1)
...
_
T
a
(i)
, T
p
...
_
T
a
(j)
, T
p
... T
a
(k)
_
apply (5.1.14) to one of the two commutators to nd
B
2
Tr
_
T
a
1
... T
a
k
_
k
i=1
Tr
_
T
a
(1)
...
_
_
T
a
(i)
, T
p
, T
p
_
... T
a
(k)
_
=
N B
2
Tr
_
T
a
1
... T
a
k
_
k
i=1
Tr
_
T
a
(1)
... T
a
(i)
... T
a
(k)
_
(5.1.15) 5,16
The last step follows from the fact that
_
[, T
p
], T
p
is the Casimir operator of the adjoint

representation of SU(N) such that
_
[T
a
, T
p
]T
p
= NT
a
and the sum over i yields k
identical terms. In the self energy corrections, one also gets a factor of k by similar
argument such that the overall contribution is
k N (B + 2A)
2
S
k
Tr
_
T
a
1
... T
a
k
_
Tr
_
T
a
(1)
... T
a
(k)
_
=
k N (B + 2A) P
k,k,0
(N)
2
.
(5.1.16) 5,17
It follows from the non-renormalization theorem that B + 2A = 0. The reason for that
is the following: The two point function TrX
2
falls into the same SUSY multiplet as
the energy momentum tensor T
. It can be shown that the latter is not renormalized (in

agreement with momentum conservation), so by supersymmetry, TrX
2
is protected as
well.
On the other hand, it suces to consider k = 2, i.e. to check explicitly that there are no
quantum corrections to O
k
(x)O
k
(y)) at order O(g
2
YM
) = O().
5.1.3 The three point function on the gravity side
Having obtained an exact result for the three point function of 1/2 BPS operators on the
eld theory side, we are ready to compare with a gravity counterpart. Let us consider three
point functions of scalar elds in AdS spacetimes. Their Feynman diagram has the structure
of a Mercedes star. It is specied by three edge points x, y, z, by three bulk-to-boundary
propagators and a coupling in the center determined by Kaluza-Klein reduction of S
5
.
Recall from section 4.2.3 that the bulk-to-boundary Green functions in AdS
d+1
is given by
K
(z
0
, z, x) =
()
d/2
(2)
_
z
0
z
2
0
+ (z x)
2
_
(5.1.17)
Because of its dening property lim
z
0
0
_
z
d
0
K
(z
0
, z, x)
=
d
(x z) we can express a bulk
eld in terms of its values at the boundary
(z
0
, z) =
()
d/2
(2)
_
d
d
x
_
z
0
z
2
0
+ (z x)
2
_
0
(x) . (5.1.18)
Now the Mercedes diagram of the gravity three point functions is evaluated as
A(x, y, z) :=
_
dw
0
d
d
w
1
w
d+1
0
_
w
0
(w x)
2
_
1
_
w
0
(w y)
2
_
2
_
w
0
(w z)
2
_
3
(5.1.19) 5,20
Here, we use the notation (w x)
2
:= w
2
0
+ ( w x)
2
.
The number of functions in denominator can be reduced using the trick of inversion: Reex-
press integration variable as w
=
w
(w
)
2
and similarly set x =
x
|x
|
2
, y =
y
|y
|
2
and z =
z
|z
|
2
.
Consequently, the propagators are aected as
K
(w, x) = [x
[
2
K
(w
, x
) . (5.1.20)
The factor [x
[
2
is a rst parallel to eld theory since [x
[
2
=
1
|x|
2
. Note that inversion is an
isometry of AdS, so its volume element is invariant
d
d+1
w
w
d+1
0
=
d
d+1
w
(w
0
)
d+1
. This causes the Mercedes
integral to transform as
A(x, y, z) = [x
[
2
1
[y
[
2
2
[z
[
2
3
A(x
, y
, z
) . (5.1.21)
To reduce the number of functions in the denominator of (5.1.19) from three to two, proceed
as follows:
set one argument to zero z 0 using translation invariance,
A(x, y, z) = A(x z, y z, 0) =: A(u, v, 0) (5.1.22)
This brings the third terms into the nice form
_
w
0
(wz)
2
_
3
=
_
w
0
w
2
_
3
= (w
0
)
3
.
apply an inversion to nd
A(u, v, 0) =
1
[u[
2
1
[v[
2
2
_
d
d+1
w
(w
0
)
d+1
_
w
0
(w
)
2
_
1
_
w
0
(w
)
2
_
2
(w
0
)
3
(5.1.23) 5,21
By translation invariance of the w integration variable, the integral can only depend on the
dierence u
, and dimensional analysis xes the power to be [u
2
. Hence, we
have already found the spacetime dependence:
A(u, v, 0)
[u
2
[u[
2
1
[v[
2
2
=
1
[x y[
1
+
2
3
[y z[
2
+
3
1
[z x[
3
+
1
2
=: f(x, y, z) (5.1.24) Af1
(Note that good care has to be taken to restore the old variables before the inversion transfor-
mation. A useful formula is: (u
)
2
=
(xy)
2
(xz)
2
(yz)
2
.)
An exact calculation of A(u, v, 0) can be done using Feynman parameter methods [4], the
prefactor in A(x, y, z) = a f(x, y, z) is found to be
a =
[
1
2
(
1
+
2
3
)] [
1
2
(
2
+
3
1
)] [
1
2
(
3
+
1
2
)] [
1
2
(
i

i
d)]
2
d
[
1
d
2
] [
2
d
2
] [
3
d
2
]
.
(5.1.25) Af2
The Gamma functions due to the Feynman parameter method have a number of poles.
Now need to consider coupling with which the Mercedes integral (5.1.19) enters the three point
function:
O
I
(x) O
J
(y) O
K
(z)
_
=
IJK
A(x, y, z) (5.1.26)
The A part was just calculated, we will next treat the cubic coupling
IJK
coming from KK
reduction in supergravity [12].
Recall from section 3.4.2 that type IIB supergravity contains a self dual ve form eld F. It
enters the equations of motion for the graviton via
R
mn
=
1
3!
F
mijkl
F
n
ijkl
. (5.1.27)
In the at AdS
5
S
5
background solution, the ve form takes particularly simple values. Denote
the AdS
5
indices by
i
, i = 1, 2, ..., 5 and the S
5
indices by
i
, i = 1, 2, ..., 5, then the solution
reads
ds
2
=
1
z
2
0
_
dz
2
+ dz
2
0
+ d
2
5
_
=: g
mn
dx
m
dx
n
1
...
5
=
1
...
5
,

F
1
...
5
=
1
...
5
(5.1.28)
Note that the curvatures of the AdS
5
and S
5
factors cancel:
AdS
5
: R
= (g
) , R
= 4 g
, 1
AdS
5
= 20
S
5
: R
= +(g
) , R
= +4 g
, 1
S
5 = +20
(5.1.29)
Observe that 1 = 1
AdS
5
+1
S
5 = 0.
Next we need to look at uctuations
0
about this background which couple to operators O in
the dual eld theory via interaction terms o
int
=
_
d
d
x
0
(x)O(x). It was investigated in [13]
how to decompose the supergravity equations of motion and how to decouple them from the
uctuations.
Starting point is the ansatz
G
mn
= g
mn
+ h
mn
, F =

F + F (5.1.30)
where the uctiations h, F are organized as
h
= h
()
+
h
2
5
g
, g
h
()
= 0
h
= h
()
+
h
5
g

h
2
3
g
, g
()
= 0 (5.1.31)
F
ijklm
= 5
[i
a
jklm]
(5.1.32)
It is convenient to work in de-Donder gauge (with respect to S
5
) where
3
= 0 . (5.1.33)
The KK programme requires to expand this ansatz in spherical harmonics Y
I
on S
5
:
h
I
Y
I
(h
)
I
, h
2
=
I
Y
I
h
I
2
a
1
...
4
=
Y
I
4
b
I
(5.1.34)
a
1
...
4
=
I
Y
I
a
I
1
...
4
Inserting this ansatz into the ten dimensional equations of motion leads to diagonalization and
decoupling. The modes which couple to the eld theory 1/2 BPS operators O
I
are given by
S
I
=
1
20 (k + 2)
_
h
I
2
10 (k + 4) b
I
_
. (5.1.35)
Note that k = in the dierent notations of the original papers. These S
5
modes satisfy a
ve dimensional equation of motion in AdS space
_
k (k 4)
_
S
I
=
IJK
S
J
S
K
(5.1.36)
where
IJK
is given by
IJK
= a(k
1
, k
2
, k
3
)
128
_
(/2)
2
1
_ _
(/2)
2
4
_
C
I
C
J
C
K
_
(k
1
+ 1)(k
2
+ 1)(k
3
+ 1)
. (5.1.37)
We are using the usual shorthands = k
1
+ k
2
+ k
3
and
1
=
k
2
+k
3
k
1
2
(as well as cyclic
variations thereof) and the numbers a(k
1
, k
2
, k
3
) relate S
5
integrals of spherical harmonics with
the SO(6) tensors
C
I
C
J
C
K
_
of (5.1.5),
_
S
5
d Y
I
() Y
J
() Y
K
() = a(k
1
, k
2
, k
3
)
C
I
C
J
C
K
_
a(k
1
, k
2
, k
3
) =

3
(/2)! 2
d/2(2)
k
1
! k
2
! k
3
!
1
!
2
!
3
!
(5.1.38)
This gives rise to the following dimensionally reduced supergravity action for the S
I
modes:
o =
4 N
2
(2)
5
_
d
5
x

g
_
A
I
2
(S
I
)
2
k (k 4) (S
I
)
2
+
1
3

IJK
S
I
S
J
S
K
_
(5.1.39)
We can identify the lower dimensional gravitation coupling and the AdS radius as
1
2
2
=
4 N
2
(2)
5
, L
AdS
= 1 , (5.1.40)
and the constant A
I
is determined from IIB 10d SUGRA action to be
A
I
= 32
k (k 1) (k 2)
k + 1
Z(k) ,
_
S
5
d Y
I
() Y
J
() =: Z(k)
IJ
. (5.1.41)
Let us now use the action o as given above to calculate the the two point function,
S
I
(x) S
J
(y)
_
=
4 N
2
(2)
5
2
k7
k (k 1)
2
(k 2)
2
(k + 1)
2
IJ
(x y)
2k
(5.1.42)
then dene normalized operators

O
I
(x) S
I
(x) such that
O
I
(x)

O
J
(y)
_
=

IJ
(xy)
2k
. The three
point function is computed on the basis
IJK
, the operators normalization as given above and
the result (5.1.24), (5.1.25) for A(x, y, z):
_
O
I
(x)

O
J
(y)

O
K
(z)
_
=
1
N
k
1
k
2
k
3
C
I
C
J
C
K
_
[x y[
2
3
[y z[
2
1
[z x[
2
2
(5.1.43)
Remarkably, this gravitational correlator coincides with the eld theory result!
5.2 The conformal anomaly
As a second example of astonishing agreement between computations in AdS gravity and ^ = 4
SYM, we will now compute the conformal anomaly using both approaches. The conformal
anomaly parametrizes the failure of the energy momentum tensor to remain traceless under
5.2. THE CONFORMAL ANOMALY 53
quantum corrections in a clasically conformal eld theory. Recall that the energy momentum
tensors expectation value can be derived from the eective action via
T
(x)) =
2
g
W
g
(x)
. (5.2.1) eq:
In this denition, we can regard g
as a classical background eld, it does not propagate, but

is a source for T
.
5.2.1 The conformal anomaly on the eld theory side
Let us consider a classical eld theory with conformal symmetry with action functional for the
matter elds
S
mat
=
_
d
4
x

g /
mat
(5.2.2) eq:1
Under variation of the metric g
+g
the chain rule implies

o
mat
=
1
2
_
d
4
x

g T
(5.2.3) eq:2
by denition of the energy momentum tensor. In Poincare invariant theories, it is symmetric
T
= T
and satises the conservation law
= 0. In conformal theories, it is also

traceless T
= 0. This can be seen from a Weyl transformation g
= 2g
in (5.2.3).
Next we proceed to the quantized theory. Matter elds will be promoted to quantum elds
whereas the metric is still regarded as an external, classical eld. The generating functional is
given by
Z[g] = e
W[g]
=
_
T exp
_
_
d
4
x

g /
mat
_
. (5.2.4) eq:
Then the Weyl transformation gives rise to the following variation in the eective action:
W[g] =
1
2
_
d
4
x

g T
=
_
d
4
x

g T
) (5.2.5)
Generically, one has T
) , = 0 at the quantum level since counterterms needed for regularization

give nite contribution to T
). This phenomenon is referred to as conformal anomaly.

Generically in a d = 4 QFT, the conformal anomaly is of the form:
T
(x)) =
c
16
2
C

a
16
2

(5.2.6) anomaly
Here, C is the Weyl tensor, obtained from the Riemann tensor by subtracting the traces such
that C
= 0. The second contribution
is called the Euler density and

gives a topological term (Gauss-Bonnet-term)
_
d
4
x
g RR = 4 proportional to the Euler

characteristic of the manifold. The coecients c and a are model-dependent.
Note that in two dimensions, the conformal anomaly is of purely topological nature, T
(x)) =
c
24
R. The prefactor c is the Virasoro central charge. This aects the energy momentum tensors
two point functions with a trace involved: Taking

g
(y)
gives T
(x)T
(y)) =
c
24
(2)
(x
y) in d = 2 two dimensions. In d = 4, however, taking rst g
derivatives of (5.2.6) we get

zero in at space (one curvature term remains in the derivative of the quadratic expression).
Nonzero results arise from second derivatives, i.e. the d = 4 conformal anomaly manifests itself
in the three point function T
(x)T
(y)T
(z)).
We are particularly interested in the case of ^ = 4 SYM. Many explicit calculations methods
(for instance heat-kernel) have been developed prior to AdS/CFT. To lowest order in one
nds that c and a depend on the number of degrees of freedom (vectors, fermions, scalars) but
not on :
c = a =
1
4
(N
2
1)
N
1
4
N
2
(5.2.7)
In total, the agreement of c and a yields the following conformal anomaly in ^ = 4:
T
(x)) =
c
8
2
_
R

1
3
R
2
_
N
N
2
32
2
_
R

1
3
R
2
_
(5.2.8) YManomaly
5.2.2 The conformal anomaly on the gravity side
The gravity counterpart, i.e. the conformal anomaly from AdS space, was computed in [14] for
the rst time. Starting point is the action of d = 5 AdS gravity,
o =
1
16 G
_
d
5
z

g
_
R +
12
L
2
_
. (5.2.9) AdSaction
Recall that the metric for AdS
5
is given by ds
2
= L
2
_
d
2
4
2
+
1
dx
dx
_
as long as the = 0
boundary remains at. If we allow for curvature terms at the boundary, this generalizes to
ds
2
= L
2
_
d
2
4
2
+
1
(x, ) dx
dx
_
, lim
0
g
(x, ) = g
(x) . (5.2.10) eq:EquationStar

Fluctuations about the at case are parametrized as g
(x) =
+h
(x), the energy momen-

tum is then obtained by T
(x)) =
2
g
W
h
.
The metrics coordinate singularty at 0 can be avoided by means of a cuto at = . The
integration region in the action is then restricted to . The induced metric at = will be
denoted by
(x) :=
g
(x, = )
. (5.2.11)
As usual, Weyl transformation of the metric gives the trace of the energy-momentum tensor.
Therefore, we need to translate a Weyl transformation in the boundary theory into a trans-
formation in the bulk, i.e. in d = 5 AdS. The task is to nd a d = 5 dieomorphism which
5.2. THE CONFORMAL ANOMALY 55
reduces to a Weyl transformation on the boundary. The desired dieomorphism is known as
the Penrose-Brown-Henneaux transformation:
=
_
1 2 (x
)
_
, x
= (x
+ a
(x
) (5.2.12)
One has to make sure that the form of the d = 5 metric (5.2.10) is covariant under this
transformation, i.e. that g
55
= g
55
and g
5
= g
5
(where 5 denotes the index). This imposes
the constraints
5
a
=
L
2
2
g
(5.2.13) diff
on the functions a
and of (5.2.13). It follows that

a
(x, ) =
L
2
2
_
0
d g
(x, )
(x) . (5.2.14) diff1

Under this dieomorphism, the d = 4 part g
(x, ) of the metric transforms as

g
+ 2
_
1

_
g
(5.2.15)
such that at the boundary (where 0) we have a
0 and

0 and therefore
g
(x) = 2(x)g
(x)
Inserting all that into action (5.2.9) yields
W =
_
d
4
x T
) g
, g
= 2 g
. (5.2.16)
The divergence of o at the boundary is regularized by introducing the cuto , subtracting
counterterms make o
ren
= o o
ct
nite. To get the explicit form of o
ct
, we need some more
information about the form of g
(x, ). This is provided by the Feerman-Graham theorem:

If a metric of the form (5.2.10) satises the Einstein equations, then g
(x, ) may be expanded

as:
g
(x, ) = g
(x) + g
(2)
(x) +
2
g
(4)
(x) +
2
ln() h
(4)
+ ... (5.2.17)
The coecients g
(n)
are built out of the curvature for the boundary metric g
(x). They are

calculated by inserting the expansion into the vacuum Einstein equation
_
R
1
2
g
R
_
= g
.
For example, the linear coecient is found to be g
(2)
(x) =
1
2
_
R
1
6
Rg
_
. The lowest order
divergent terms in o then lead to
S
ct
=
1
4 G
_
d
4
x

_
3
2L

LR
8

L
3
ln
32
_
R

R
2
3
__
(5.2.18)
in the minimal subtraction renormalization scheme. The action o we started with is dieomor-
phism invariant, but the introduction of the counterterms spoils this symmetry. More precisely,
the Penrose-Brown-Henneaux transformation of o o
ct
is given by
(o o
ct
) = 2

o
ct
=
L
3
64G
_
R

R
2
3
_
. (5.2.19)
The Weyl variation at the boundary gives rise to the energy momentum trace
g
) =
L
3
8G
_
R
8

R
2
24
_
=
N
2
32
2
_
R

R
2
3
_
(5.2.20)
with d = 5 Newton constant G =
G
10
vol(S
5
)
=
L
3
2N
2
. Again, the gravity result conincides with the
N limit of the eld theory pendant (5.2.8).
Chapter 6
Generalizations of AdS/CFT
6.1 Holographic renormalization group ows
In the context of the AdS/CFT correspondence, the term holography represents the fact that the
number of bulk degrees of freedom equals the number of boundary degrees of freedom. So far, we
have checked this phenomenon for ^ = 4 SYM theory with gauge group SU(N) and symmetry
group SO(4, 2) SO(6) SU(2, 2[4) and its gravity dual. However, this correspondence is far
away from reality because of the following points:
correspondence valid in the N limit
conformal symmetry and supersymmetry
elds in the adjoint representation of the gauge group
On the other hand, with QCD, we have the well-established theory of strong interactions which
diers from ^ = 4 SYM in several points:
gauge group SU(3), i.e. small nite N
no supersymmetry
connement incompatible with conformal symmetry
quarks in the fundamental representation of the gauge group
Adapting the gauge group SU(N) SU(3) in the AdS/CFT correspondence will probably not
be achieved in the near future. But looking at large N QCD and performing a 1/N expansion
is still a useful approximation.
57
58 CHAPTER 6. GENERALIZATIONS OF ADS/CFT
As to the other features in the list: We may break conformal symmetry and supersymmetry in
a controlled way, and we may add quark degrees of freedom. To nd a modest beginning, let
us rst of all break some of supersymmetry (^ = 4 to ^ = 1) as well as conformal invariance.
On the eld theory side, this is done by adding relevant operators to the Lagrangian:
/ = /
N=4
+
m
ij
2
Tr
_
X
i
X
j
_
+
M
ab
2
Tr
_
b
_
+ b
ijk
Tr
_
X
i
X
j
X
k
_
(6.1.1) 6,1
This modication triggers renormalization group ows. In general QFTs, renormalization group
equations express the invariance of physics under dierence choices of the renormalization scale
. Let denote some vertex function (depending on couplings g
i
), then the RG equation
assumes the form
0 =
d[, g
i
]
d
=
[, g
i
]
+
i
(g)
i
[, g
i
] ,
i
(g) =
dg
i
()
d
. (6.1.2) 6,2
6.1.1 Renormalization group ow in supergravity
To nd an AdS analogue of eld theory RG equations we now look for a toy model of a
supergravity RG ow. The idea is to obtain equations like (6.1.2) as equations of motion in
extra dimensions. For that purpose, consider ve dimensional gravity with a single scalar eld:
o =
1
4 G
_
d
5
x
_
[g[
_
R
4
+
1
2

+ V ()
_
(6.1.3) 6,3
For simplicity set 4G = 1 and split spacetime as d
5
x = d
4
x dr, then is dimensionless.
Moreover, choose the potential V () such that is has one or more critical points with V
() = 0.
The equations of motion for and g
read
1
_
[g[
_
_
[g[ g
_
V
() = 0 (6.1.4) 6,4
as well as
R

R
2
g
= 2
+ 2 V () =: 2 T
(6.1.5) 6,5
At the critical points
i
, there is a trivial solution of the scalar equation of motion (r) =
i
.
Here, the Einstein equation reduces to R

R
2
g
= 2g
V (
i
). This is identical to the
Einstein equation of AdS space G
= 0 if we identify
i
= 2V (
i
) =
d(d1)
L
2
i
. In other
words, constant scalar elds with AdS
d+1
geometry of scale L
i
are critical solutions.
A more general ansatz for solving the equations of motion to (6.1.3) involves a metric with
so-called warp factor A(r):
ds
2
= e
2A(r)
dx
dx
+ dr
2
(6.1.6) 6,6
6.1. HOLOGRAPHIC RENORMALIZATION GROUP FLOWS 59
This is known as the domain wall ansatz for the metric, it reduces to another form of the AdS
metric if we make the linear choice A(r) = r/L. The idea is to identify the radial coordinate
with the RG scale r =
1
.
The components of the Riemann tensor due to (6.1.6) read
R
ij
kl
= A
(r)
2
_
i
k

j
l

i
l

j
k
_
R
i5
j5
=
_
A
(r) + A
(r)
2
_
i
j
(6.1.7) 6,7
R
ij
k5
= 0
(where i, j, k, l 0, 1, 2, 3 and r 5) and the resulting Ricci tensor is given by
R
ij
= e
2A(r)
_
A
(r) + d A
(r)
2
_
ij
R
55
= d
_
A
(r) + A
(r)
_
2
(6.1.8) 6,8
R
i5
= 0 .
This gives rise to Einstein equations
G
i
j
= (d 1)
i
j
_
A
+
d
2
(A
)
2
_
= 2 T
i
j
G
5
5
=
d (d 1)
2
(A
)
2
= 2 T
5
5
. (6.1.9) 6,8a
By carefully considering the dierence G
i
i
G
5
5
(without sum over i), one can extract a bound
on the second derivative of the warp factor from (6.1.9):
A
=
2
d 1
_
T
i
i
T
5
5
_
=
2
d 1
_
_
2
A
< 0 (6.1.10) 6,10

This is consistent with the weak energy condition saying that every Poincare invariant matter
distribution satises T
i
i
T
5
5
< 0.
The d = 4 equations of motion in terms of A read
+ 4 A
=
dV ()
d
(
)
2
2 V () = 6 (A
)
2
. (6.1.11) 6,11
Note that A(r) =
r
L
i
at critical points.
Our goal is to nd a general solution of (6.1.11) which interpolates between two critical points.
In AdS
5
S
5
language, we are looking for a domain wall solution interpolating between AdS of
radius L
1
for r + and AdS of radius L
2
for r . This is dual to an RG ow between
two xed points.
6.1.2 Leigh Strassler ow
A very nice example is the Leigh Strassler ow which can be best described in supereld
language. It establishes a eld operator map in particular for the relevant operators added
to /
N=4
. We can describe the ^ = 4 theory in ^ = 1 notation by reorganizing the ^ = 4
supermultiplet (one vector A
, four fermions
a
and six real scalars X
i
) as follows
one ^ = 1 vector multiplet (A
,
a=4
)
three ^ = 1 chiral multiplets (
a
,
a
= X
2a1
+iX
2a
) with a = 1, 2, 3
Superspace is an 8 dimensional space spanned by four standard spacetime coordinates x
, =
0, 1, 2, 3 and two Weyl spinorial Grassmann variables
(where ,

= 1, 2). Scalars and
three fermions of ^ = 4 SYM are aligned into ^ = 1 chiral superelds with theta expansion
i
=
i
+
+ higher order in . (6.1.12) 6,12

Chiral superelds are annihilated by half the supercharges

Q

= 0. Clearly, the expansion
terminates after a

term.
In addition, a vector supereld V captures the ^ = 4 vector A
and the fourth fermion. One

can bring the theta expansion into the form
V =
+ ( )

4
+ (
,4
(6.1.13) 6,12a
Let us write down the Lagrangian of the ^ = 4 theory in ^ = 1 superspace language:
/
N=4
=
3
i=1
_
d
2
d
2
Tr
_
e
gV

i
e
gV
i
_
+
_
g
_
d
2
Tr
_
3
_
1
,
2
_
+
1
g
2
_
d
2
Tr
_
W
_
+ h.c.
_
(6.1.14) 6,13
In this formulation, /
N=4
exhibits manifest ^ = 1 supersymmetry with R symmetry group
SU(3) U(1). The rst d
2
integrand g
3
_
1
,
2
is referred to as a superpotential. The

gauge kinetic term involves superelds W
which contain the non-abelian eld strength F
derived from A
.
The general relevant deformation which preserves ^ = 1 supersymmetry is obtained by adding
an additional superpotential contribution
U = hTr
_
3
_
1
,
2
_
+
M
AB
2
Tr
_
B
_
. (6.1.15) 6,14
What is left to show is the fact that this theory ows to a non-trivial xed point in the IR.
The Leigh Strassler ow is triggered by the deformation
U
LS
:= hTr
_
3
_
1
,
2
_
+
m
2
Tr
_
(
3
)
2
_
. (6.1.16) 6,15
Because of the scaling dimensions [h] = 0 and [m] = 1, the former term is referred to as
marginal, the mass term as relevant.
The deformation (6.1.16) leads to a reduced R symmetry SU(2) U(1), the former acting on
the
1,2
elds. The U(1) charges of the chiral superelds
1,2,3
are (1/2, 1/2, 1).
A necessary condition for an IR xed point is
i
= 0, and luckily the beta function for the
gauge coupling (g) is well-known for ^ = 1 theories to all orders in perturbation theory. It is
given by the NSVZ beta function (named after Novikov, Shifman, Vainstein and Zakharov)
(g) =
g
3
T(R
A
)
8
2
3 T(G)

A
T(R
A
) (1 2
A
)
1 g
2
T(G)/(8
2
)
(6.1.17) 6,16
Here,
A
denotes the anomalous dimension of the supereld
A
and the Dynkin index T(R
A
)
of the representation R
A
is dened by the normalizaton of the two-trace,
Tr
A
_
T
a
T
b
_
= T(R
A
)
ab
. (6.1.18) 6,17
Here we are dealing with G = SU(N) and all the elds transform in the adjoint representation.
Therefore, T(R
A
) = T(G) = N and
(g) 2N (
1
+
2
+
3
) . (6.1.19) 6,18
The functions for matter elds are simple due to non-renormalization theorems in SUSY
theories. Hence, the running of the parameters h, m in (6.1.16) is governed by
h
=
1
+
2
+
3
,
m
= 1 2
3
. (6.1.20) 6,19
The condition (g) =
h
=
m
has a unique SU(2) invariant solution
1
=
2
=
3
2
=
1
4
. (6.1.21) 6,20
The IR xed point theory with given values of the anomalous dimensions has ^ = 1 super-
conformal symmetry under SU(2, 2[1). The engineering dimensions of the superelds are given
by
A
= 1 +
A
. According to the eld operator map, conformal primary operators O are
constructed from gauge invariant combinations of
1
,
2
and the eld strength supereld W
(which contains F
).
O Tr
_
j
_
Tr
_
W

i
_
Tr
_
W
_
Tr
_
i
(T
a
)
i
j

j
_
Tr
_
W
_
3/2 9/4 3 2 3
The
i=1,2
form an SU(2) doublet, and T
a
denote the associated SU(2) generators.
6.1.3 Holographic ows in supergravity
The next goal is to construct a gravity dual to the Leigh Strassler ow. For that purpose, let
us start with some general remarks about holographic ows in supergravity. Consider a metric
of domain wall type
ds
2
= e
2A(r)
dx
dx
+ dr
2
(6.1.22) 6,21
with boundary at r . Recall that it reduces to an AdS spacetime if A(r) =
r
L
, i.e. these
cases are dual to conformal eld theories. An RG ow between two CFTs can be mapped to
an A(r) solution which interpolates between two linear regimes of dierent slope.
As a simplest case, we put a single scalar eld into the spacetime (6.1.22), the equations of
motion for this system have been given in (6.1.11). They can be simplied by introducing an
auxiliary function W(),
V () =
1
2
_
dW
dr
_
2
4
3
W
2
, (6.1.23) 6,22
namely they become a rst order gradient ow (after using A
=
2
3
W)
d
dr
=
dW
d
. (6.1.24) 6,23
We look for solutions of the equations of motions which interpolate between two conformal
xed points on the eld theory side:
lim
r
A(r) =
r
L
1
, lim
r
A(r) =
r
L
2
(6.1.25) 6,24
The two AdS radii L
i
lead to dierent central charges for the CFTs, say L
1
c
UV
and L
2
c
IR
.
Recall that in d = 4 SYM, the central charge is the prefactor relating T
with R
R
2
3
.
In quadratic approximation, the gravity potential V () in the neighbourhood of the UV critical
point
i
is given by
V () = V (
i
) +
m
2
i
2 L
2
i
h
2
+ O(h
3
) . (6.1.26) 6,25
in terms of h :=
i
with masses m
2
i
= L
2
i
V
(
i
). The value at
i
itself is the cosmological
constant,
V (
i
) =
d (d 1)
4 L
2
i
. (6.1.27) 6,26
The uctuation h(r, x) can be viewed as the gravity dual to some operator O
(x) where m
2
i
=
(4).
The marginal operator
1
g
2
TrF
for instance has dimension = 4, so it must couple to

some massless eld. A careful analysis of the d = 10 DBI action identies the dilaton with
that scalar coupling to TrF
2
. The dilaton being constant then leads to the vanishing of the
function.
The asymptotic behaviour of h(r, x) at the boundary of (d + 1) dimensional AdS space is
governed by
h(r, x)
r
e
(d)r
_
i
+

h(x)
_
, (6.1.28) 6,27
cf. the earlier interpretation (r)
0
e
(d)r
+O) e
r
.
As a generalization of the AdS/CFT generating functional, consider
_
exp
_
o
CFT
+
_
d
d
x O
(x)
_
i
+

h(x)
_
__
= e
S
grav
[
h]
. (6.1.29) 6,28
The new piece o :=
_
d
d
x O
(
i
+
h) belongs to a non-conformal action o = o

CFT
+o, i.e.
o is an operator deformation of the CFT.
Correlation functions on the eld theory side can be obtained from the gravitational action
O
(x
1
) ... O
(x
n
)) =
(1)
n1
h(x
1
) ...
h(x
n
)
o
grav
[
i
+

h]
h=0
. (6.1.30) 6,29
Negative mass squares occur if the critical point is a local maximum. If 0 > m
2
>
d
2
4
, then
d > >
D
2
and O
is a relevant deformation driving the eld theory away from the xed
point.
If V has a local minimum at the xed point then m
2
> 0 and O
has conformal dimension

> d. This is an example for an irrelevant operator which drives the ow into an IR xed
point. We look at interpolating ows, i.e. at solutions of the combined equations of motion (of
Einstein- and scalar type) for which the scalar eld (r) corresponds to a maximum of V ()
in the UV (r ) and to a minimum in the IR (r ).
Let us revisit the expansion around the critical point, this time in the form
(r) =
i
+ h(r) , A
=
1
L
i
+ a
(r) (6.1.31) 6,30

where a
= O(h
2
). The linearized equation of motion for the scalar uctuation reads
h
+
d
L
i
h

m
2
i
L
2
i
h = 0 (6.1.32) 6,31
with general solution
h(r) = Be
(
i
d)r/L
i
+ C e
i
r/L
i
, m
2
i
=
i
(
i
4) . (6.1.33) 6,32
In the limit r we have
(r +) =
1
+ B
1
e
(
1
d)r/L
1
+ C
1
e
1
r/L
1
(r ) =
2
+ B
2
e
(
2
d)r/L
2
+ C
2
e
2
r/L
2
. (6.1.34) 6,33
One gets the impression that the domain wall ow sees the IR xed point only in the deep
interior r . To establish the eld operator map for CFT at the IR xed point, we have
to extend the IR geometry to a complete AdS space with radius L = L
IR
. Note that the bound
A
< 0 implies L
UV
> L
IR
.
6.2 The holographic c theorem
TO BE FILLED BY JOHANNA
6.3 Last lecture of 2009
Domain wall ow ds
2
= e
2A(r)
ij
dx
i
dx
j
+ dr
2
. Case A(r) =
r
L
corresponds to AdS.
Other possibility: Connement
Plots: - A(r) with pole at nite negative value of r, and linear for r - Coupling in
QCD
Characterise connement by calculating Wilson loop (area law)
What we also want to do: So far: N=4, all elds in adjoint rep of gauge group e
i
e
i
Need for QCD: Quarks are in fundamental rep. e
i
Also on gravity side, need more degrees of freedom. One possibility: Embedding of probe
D7 branes avour (global symmetry)
In the Maldacena limit, the gauge coupling on the D7 brane goes to zero global symmetry.
In the probe limit (N D3-branes, N , 1 or 2 D7-branes) we may ignore the backreaction of
the D7-branes on the D3-branes. Separation of D3, D7 in 8,9 directions corresponds to quark
mass.
New duality: 1) Standard AdS/CFT 2) N=2 SUSY hypermultiplet in fundamental rep of
gauge group uctuations of D7 in AdS5xS5 as described by the Dirac-Born-Infeld action for
the D7.
SUSY embeddings
D3 x x x x 0 0 0 0 0 0
D3 x x 0 0 x x 0 0 0 0 AdS
3
S
1
D5 x x x 0 x x x 0 0 0 AdS
4
S
2
D7 x x x x x x x x 0 0 AdS
5
S
3
Black Hole in AdS
5
- Field theory at nite temperature Hawking Temperature - temperature
in QFT
6.4 Applications 1: Field Theories at nite temperature
Switching on a temperature breaks all of the supersymmetry (still we keep the degrees of
freedom of ^ = 4, with possibly dierent masses, and the N planar limit!). Simplest
6.4. APPLICATIONS 1: FIELD THEORIES AT FINITE TEMPERATURE 65
case: t i, e
iHt
e
H
, =
1
k
B
T
, eld theory in thermal equilibrium.
2nd step: consider black hole in Minkowski signature AdS space dynamical processes
(transport, relaxation) near-equilibrium.
Quark-Gluon plasma: The to date most successful application of generalized AdS/CFT is to
descriptions of the QGP. This is a state of (QCD) matter at nite temperature and/or density.
Phase diagram of QCD: See plot
Perturbative QCD is not suited for describing the strongly coupled quark-gluon-plasma.
Lattice gauge theory is dicult at nite temperature and density, and not suited for describing
dynamical processes such as scattering.
AdS/CFT at nite temperature is well-suited for describing strongly coupled ^ = 4 theory
in the planar limit, in particular dynamical processes. In some important examples, comparison
to QCD is possible.
Most famous example: (Shear viscosity / entropy density) ratio.
=
1
_
d
4
p e
it
T
xy
( p) T
xy
( p)) (6.4.1)
(Kubo formula)
AdS/CFT:

s
=
1
4
very small value, almost ideal uid!
Perturbative QCD result: factor 10 larger
AdS/CFT result is in agreement with measurements at RHIC (Large experimental error,
however).
Review of Black Holes:
1) Flat space
Schwarzschild Metric
ds
2
=
_
1
2GM
r
_
dt
2
+ (1
2GM
r
)
1
dr
2
+r
2
d
2
(6.4.2)
r = 0 is a singularity (true curvature singularity), R
=
48G
2
M
2
r
6
.
r = 2GM Schwarzschild radius
At the Schwarzschild radius, the curvature is nite. It corresponds to the event horizon of
the black hole. For an external observer, the light cone closes up as r 2GM.
Other coordinates (Eddington-Finkelstein): Light cones tilt over
Hawking Temperature: Perform a Bogolubov transformation between states in coordinate
systems at and far away from the black hole.
2) Calculate number density thermal spectrum. n
=
1
exp(
E
T
H
)1
, T
H
=
kappa
2
=
1
8GM
.
surface gravity: acceleration needed to keep an object at the horizon (Schwarzschild radius).
6.5 Gauge gravity duality at nite temperature and den-
sity
At nite density and temperature: Let
be a saddle point of some Euclidean action o

E
[],
then we can approximate the generating functional semiclassically as
Z =
_
T e
S
E
[]
e
S
E
[
]
. (6.5.1) 8,1
According to the weak form of the AdS/CFT correspondence, the partition function of the
classical bulk theory with asymptotically AdS boundary conditions is equivalent to the partition
function of the large N QFT. The metric g then takes the role of the eld above:
Z
grav
= e
S
E
[g
]
(6.5.2) 8,2
The gravitational action contains a Gibbons Hawking boundary term required for niteness,
o
E
[g] =
1
2
2
_
d
d+1
x

g
_
1 +
d (d 1)
L
2
_
+
1
2
2
_
r0
d
d
x

g
_
2 / +
2 (d 1)
L
2
_
.
(6.5.3) 8,3
Here, / denotes the trace of the extrinsic curvature,
/ =
, (6.5.4) 8,4
where
is the induced metric on the boundary at r 0 and n
an outward pointing unit

normal vector on the boundary.
A saddle point, i.e. a solution to the equations of motion, is obtained by analytic continuation
of the AdS Schwarzschild metric
ds
2
=
L
2
r
2
_
f(r) d
2
+
dr
2
f(r)
+ dx
i
dx
i
_
, f(r) = 1
r
4
r
4
H
. (6.5.5) 8,5
The periodicity requirement for regularity xes the temperature to
T =
d
4 r
H
. (6.5.6) 8,6
We obtain further thermodynamic quantities by evaluating the partition function at the saddle
point e
S
E
[g
]
. The action as given in (6.5.3), evaluated at the Euclidean Schwarzschild metric,
is found to be
o
E
=
L
d1
2
2
r
d
H
V
d1
T
=
(4)
d
L
d1
V
d1
T
d1
2
2
d
d
(6.5.7) 8,7
where V
d1
is the spatial volume of the associated QFT.
In order to be in the classical gravity regime, we need that the spacetime is weakly curved in
Planck units, i.e. that
L
d1
2
1. The dual eld theory analogue of
L
d1
2
1 is N , recall
that L
4
= 4g
s
N
2
.
6.5. GAUGE GRAVITY DUALITY AT FINITE TEMPERATURE AND DENSITY 67
From the action given by (6.5.7) we obtain the free energy and entropy as
F = T ln Z = T o
E
[g
] =
(4)
d
L
d1
V
d1
T
d
2
2
d
d
(6.5.8)
S =
F
T
=
(4)
d
L
d1
V
d1
T
d1
2
2
d
d1
. (6.5.9)
The expression for the entropy is equal to the area of the event horizon divided by 4G
N
=

2
2
.
This area entropy relation is universally expected to be true for event horizons.
6.5.1 Finite density
Consider gravitational theories which are dual to a QFT with an additional global U(1) sym-
metry. What is the gravity dual of this symmetry? Generically, in gauge gravity duality, the
correspondence is
global symmetry of eld
theory in d dimensions
_
_
_

_
_
_
local symmetry of gravity
in d + 1 dimensions
(6.5.10)
The current J
of global U(1) symmetry in eld theory is dual to a gauge eld A

M
in the d +1
dimensional gravity theory, more precisely to its pull back A
to the boundary.
To give another example of the global-local dictionary: A global SO(d 1) rotation symmetry
in the spatial directions of a QFT becomes part of the dieomorphisms of general relativity.
Gauge symmetries include subgroups of large gauge transformations which act non-trivially
as global symmetries on the boundary of spacetime. In the AdS/CFT correspondence, this is
precisely the global symmetry at the boundary.
To describe the physics of the global U(1) symmetry, we therefore have to add a Maxwell eld to
the bulk spacetime. This leads to Einstein Maxwell theory (which for instance arises naturally
from the graviphoton in supersymmetric theories). In Minkowski signature, its action reads
o[g, A] =
1
2
2
_
d
d+1
x

g
_
1 +
d (d 1)
L
2
_

1
4 g
2
_
d
d+1
x

g F
. (6.5.11) 8,8
If the Einstein Maxwell action is derived from a supersymmetric theory, then the couplings
and g are related. Moreover, the supercurrent of the four dimensional supergravity theory
contains both the R symmetry current R
and the energy momentum tensor in its expansion:

J
= R
(6.5.12) 8,8a
6.5.2 Chemical potential in quantum eld theory
Consider a QFT containing a scalar, fermion and a gauge eld with Lagrangian
/ = (D
+ i

/ D +
1
g
2
F
. (6.5.13) 8,9
The U(1) gauge eld A
enters the covariant derivative via D
+iA
. Let us give its time

component a non-vanishing VEV of the form A
0
) = such that
A
0
= A
0
) + A
0
. (6.5.14) 8,10
then a potential is generated of the form
V =
2
. (6.5.15) 8,11
This is an upside down mass term for the scalar causing instability, and the extra term for
the fermion can be interpreted as a density operator ^
. The coecient of the overall

potential ^ is interpreted as the chemical potential.
The corresponding thermodynamical potential of the grand canonical ensemble is the Gibbs
free energy
= E T S ^ . (6.5.16) 8,12
A similar structure is present in the gravity dual. For this, we have to nd a solution to the
equation of motion of Einstein Maxwell theory with A = A
t
(r)dt. The background Maxwell
potential of the eld theory is read o from the boundary values of the bulk Maxwell eld
A
(r) = A
(0)
+ ... as r 0. The Einstein equations of motion involve the energy momentum

tensor of the eld strength F

1
2
g

d (d 1)
2 L
2
g
=

2
g
2
_
F
1
4
g
_
(6.5.17) 8,13
whereas the Maxwell equations remain in their standard form
= 0.
A particular solution of the Maxwell eld of the form A = A
t
(r)dt is the Reissner Nordstrom
AdS black hole
ds
2
=
L
2
r
2
_
f(r) dt
2
+
dr
2
f(r)
+ dx
i
dx
i
_
f(r) = 1
_
1 +
r
2
H
2
_ _
r
r
H
_
d
+
r
2
H
2
_
r
r
H
_
2(d1)
(6.5.18) 8,14
=
(d 1) L
2
g
2
(d 2)
2
This satises the boundary condition that A
t
(r) has to vanish at the horizon since
t
is not
well-dened as a Killing vector there. Moreover we have
A
t
(r) =
_
1
_
r
r
H
_
d2
_
. (6.5.19) 8,15
6.6. DISSIPATIVE DYNAMICS CLOSE TO EQUILIBRIUM 69
This identies the parameter in the solution (6.5.18) with the chemical potential. The
temperature is again xed by analytic continuation to the Euclidean regime and is given by
T =
1
4 r
H
_
d
(d 2) r
2
H
2
_
. (6.5.20) 8,16
In the grand canonical ensemble, by evaluating the Euclidean action on the solution, we nd
the following Gibbs free energy
= T ln Z =
L
d1
2
2
r
d
H
_
1 +
r
2
H
2
_
V
d1
= T
_
T
_
V
d1
T
d
. (6.5.21) 8,17
From this, we may obtain the charge density (wlog in d = 3 dimensions)
=
1
V
2
=
2 L
2
2
r
H
2
. (6.5.22) 8,18
6.6 Dissipative dynamics close to equilibrium
So far, we have considered time independent homogeneous backgrounds. A natural next step
is to include small space- and timedependent perturbations about equilibrium. The idea of
linear response theory will be implemented in the context of the AdS/CFT correspondence in
the following. This is particularly useful for describing experimentally relevant processes such
as transport coecients and spectroscopy.
The basic object in the linear response theory is the so-call retarded Green function relating
linear sources to corresponding expectation values. The theory allows to relate two point
correlation functions to transport coecients.
6.6.1 Retarded Green functions in QFT
Consider the response of a system to the presence of weak external elds
i
(with possible
Lorentz indices suppressed) coupled to a set of operators O
i
(x). The Hamiltonian is the mod-
ied by a term of the form
H =
_
d
d
x
i
(t, x) O
i
(t, x) . (6.6.1) 8,19
From time dependent perturbation theory, we know that these external elds will produce a
change in the expectation values of the operators,
O
i
) =
_
d
d+1
y G
ij
R
(x, y)
j
(y) + O(
2
) (6.6.2) 8,20a
G
ij
R
(x, y) = i (t
x
t
y
)
_
O
i
(x) , O
j
(y)
) , (6.6.3) 8,20
where G
ij
R
(x, y) is the retarded Green function. It is nonvanishing only in the forward light cone
and therefore provides a causality structure. In Fourier space we have
O
i
(k)) = G
ij
R
(k)
j
(k) + O(
2
) , G
ij
R
(k) =
_
d
d+1
x e
ikx
G
ij
R
(x, 0) . (6.6.4) 8,21
To explain the relation between Green functions and transport coecients, it is convenient
to start with the instructive example of Ohms law. It relates an electric source eld E
j
representing a linear perturbation to the (electric) response current J
i
via conductivity matrix
:
J
i
() =
ij
() E
j
() (6.6.5) 8,22
To make contact with the general notation, let us identify
i
with the external vector potential
A
and the operator O

i
with the conserved current J
. Choosing temporal gauge with A

t
= 0,
the electric eld becomes E
i
=
t
A
i
. In Fourier space, using A
i
e
it
, we have E
i
= iA
i
.
Comparing Ohms law with (6.6.4) we see that the conductivity and the current-current Greens
function as dened by (6.6.3) are proportional:
ij
() =
G
ij
R
()
i
(6.6.6) 8,23
6.6.2 The gravity side of Green functions
The AdS/CFT correspondence conjectures that the current-current correlation can be com-
puted from a higher dimensional gravity theory by varying its action with respect to boundary
values A
(0)
of a source eld, i.e.

J
(x)J
(y))

2
A
(0)
(x)A
(0)
(y)
e
iS
grav
. (6.6.7) 8,24
In general on the gravity side, taking operator mixing into account we have
O
A
(, k)) = G
O
A
O
B
R
(, k)
(0)
B
(, k) . (6.6.8) 8,25
The source is now the boundary value
(0)
of a eld in curved space. Consider uctuations of
the bulk elds of the form
A
(r)
A
(r) +
A
(r)e
it+ikx
where
A
solves the equations of
motion in gravity dual space.
The equation of motion for
A
is obtained by substituting the perturbed solution into the
equations of motion and by then linearizing. Take boundary conditions at the AdS boundary
r = 0
A
(r) = r
d
(0)
A
+ ... : r 0 . (6.6.9) 8,26
6.6. DISSIPATIVE DYNAMICS CLOSE TO EQUILIBRIUM 71
Since we are interested in the eld theorys behaviour at nite temperature, we assume the
gravity dual to possess an AdS black hole background. The second boundary condition sup-
plementing (6.6.9) has to be imposed at the black holes horizon. As one can read o from the
conformal diagram below, there are two possibilities leading to retarded and advanced Green
functions:
conf diagram
Fluctuations of the time slice ending on the future event horizon are associated with the retarded
Green function. The future horizon at r = r
H
(where g
tt
= 0) is a null surface beyond which
events cannot causally propagate to the asymptotically AdS region (the boundary region) any
more. On a future horizon, regularity requires that modes are ingoing (they can propagate into
the black hole but cannot escape from it). For the horizon at r = r
H
with nonzero temperature,
this implies
A
(r) = (r r
H
) e
4i/T
_
const + ...
_
: r r
H
. (6.6.10) 8,27
In all of the subsequent, we impose the given boundary conditions (6.6.9) and (6.6.10).
Given a mode
A
satisfying the required boundary conditions and linearized equations of
motion, we obtain from (6.6.8) that
G
O
A
O
B
R
(, k) =
O
A
)
(0)
B
=0
= lim
r0
(0)
B
=0
(6.6.11) 8,28
where
A
is obtained in the following way:
O
A
) = i
Z
bulk
[
(0)
]
(0)
A
N
= i
o
grav
[
(0)
]
(0)
A
(6.6.12) 8,29
Taking regularization at the boundary into account, we have
o
grav
[
(0)
]
(0)
A
= lim
r0
_
o
grav
[
(0)
]
(
r
(0)
A
)
+
o
bdy
[
(0)
]
(0)
A
_
= lim
r0
A
[
(0)
] . (6.6.13) 8,30
The boundary version of the action contains appropriate counterterms necessary to make o
grav
nite when evaluated at the boundary. The underlying procedure is known as holographic
renormalization. For a scalar eld, the boundary term is
o
bdy
=
_
r0
d
d
x

_
n
+

2 L

2
_
. (6.6.14) 8,31
Inserting the near boundary value of the scalar eld ...
HERE THE LAST LINES FROM LAST LECTURE ARE MISSING
6.6.3 Example: Holographic computation of Ohms law
Consider the gravity dual for a strongly interacting 2+1 dimensional eld theory, ^ = 8, d =
3, SU(N) SYM theory in 2+1 dimensions. The coupling g
2
YM
has dimensions of mass in d = 3,
so it has to appear in the dimensionless ration
g
2
YM
E
with E denoting the energy scale considered.
We expect the eld theory to be strongly coupled at low energies and to have an IR xed point.
This eld theory is conjectured to be dual to M theory on AdS
4
S
7
. A certain sector of the
dual gravity theory is described by the four dimensional eective action
o =
1
2
2
_
d
4
x
_
[g[
_
1 2
_

1
4 g
2
_
d
4
x
_
[g[ F
AB
F
AB
(6.6.15) 8,32
with negative cosmological constant = 3/L
2
.
Take A
to be the dual eld to the current J
of a U(1) subgroup of the global SO(8) R

symmetry. The classical gravitational description is valid at large N where
1
2
N
3/2
. The
couplings and g in the Einstein Maxwell action (6.6.15) are related by supersymmetry
2
=
2g
2
L
2
.
An important solution to the equations of motion for the action o is the dyonic black hole
(which is only possible in d = 4 on the gravity side), i.e. a black hole with both electric and
magnetic charge (q, h):
ds
2
L
2
=
1
r
2
_
f(r) dt
2
+ dx
2
+ dy
2
_
+
1
r
2
dr
2
f(r)
A =
hx
r
H
dy q
_
1
r
r
H
_
dt (6.6.16) 8,33
f(r) = 1 + (h
2
+ q
2
)
r
4
r
4
H
_
1 + (h
2
+ q
2
)
_
r
3
r
3
H
We have dened :=

2
r
2
H
2g
2
L
2
. This solution is dual to the 2+1 dimensional eld theory at nite
T, B and electric charge density n:
T =
3 (h
2
+q
2
)
4 r
H
, B =
h
r
H
, n = J
t
) (6.6.17) 8,34
Now let us compute the current. Near the boundary, we may expand A
= a
+ rb
+ .... On
shell, the Maxwell part of the action reduces to a boundary term of the form
o
max
=
1
g
2
_
d
3
x
r
A
r=0
=
1
g
2
_
d
3
x
. (6.6.18) 8,35
We use the radial gauge A
r
= 0. Then, the current takes the expectation value
J
) =
o
max
a
=
b
g
2
. (6.6.19) 8,36
6.7. HYDRODYNAMICS AND SHEAR VISCOSITY IN ADS/CFT 73
For the dyonic black hole, b
t
=
q
r
H
and a
t
= q =: such that
J
t
) = n) =
q
g
2
r
H
=

g
2
r
H
. (6.6.20) 8,37
On the way towards Ohms law, we dene J
= J
x
iJ
y
and E
= E
x
iE
y
. The previous
calculation gave
J
) =
1
g
2
lim
r0
r
A
. (6.6.21) 8,38
We can think of i
r
A
as a bulk magnetic eld B
and of E
as the boundary limit of a bulk

electric eld c
. From the gravity point of view, Ohms law may be written as
=
i J
)
E
= lim
r0
B
(r)
g
2
c
(r)
. (6.6.22) 8,39
In the Maxwell part of the action (6.6.15), either F
AB
or its dual

F
AB
=
1
2
ABCD
F
CD
could
be the fundamental eld strength. The action is classically invariant under switching the
electric and magnetic eld. Note that this electric magnetic duality is a special feature of four
dimensional spacetime.
For the dyonic black hole, the duality transformation is
B
, c
, (q, h) (h, q) . (6.6.23) 8,40

For the numerical computation of
, see Hartnoll, Herzog. The result satises the constraints

from electric magnetic duality
(q, h) = lim
r0
B
(q, h)
g
2
c
(q, h)
= lim
r0
c
(h, q)
g
2
B
(h, q)
=
1
g
4
(h, q)
. (6.6.24) 8,41
6.7 Hydrodynamics and shear viscosity in AdS/CFT
6.7.1 Relativistic hydrodynamics
According to the work of Son and Starinets, we will work with a metric of signature
=
(, +, +, +). In hydrodynamics, one considers a system in equilibrium subject to small per-
turbations. This is a perfect framework for an eective theory describing dynamics at large
distances and time scales. It describes dissipation in thermal media.
In the simplest case, the hydrodynamic equations are just the laws of energy momentum conser-
vation
= 0. The number of independent components of T
is reduced by the assumption

of local thermal equilibrium: If perturbations have long wavelengths or small frequencies, then
the state of the system considered at a given time is determined by the temperature as a func-
tion of the coordinates, T(x), and the local uid four velocities u
(x). Because u
= 1,
only three components are actually independent.
The number of variables is thus four (e.g. u
1
, u
2
, u
3
, T), equal to the number of independent
equations
= 0. To express T
(x) in terms of T(x), u
(x), it is convenient to expand in

powers of spatial derivatives. To zeroth order, we have an ideal uid (without dissipation) for
which
T
= ( + P) u
+ P g
+ O() , (6.7.1) 8,42

and P denote the energy density and the pressure. From thermodynamic laws d = TdS,
dP = sdT and +P = Ts (with entropy density s), one can deduce conservation of the entropy
current
(s u
) = 0 . (6.7.2) 8,43
To describe dissipation or entropy production, we have to proceed to the next order in the
derivative expansion,
T
= ( + P) u
+ P g
+ O(
2
) . (6.7.3) 8,44
For simplication, we go to a local rest frame in which u
i
(x) = 0. In this frame,
00
=
0i
=
0 equivalent to T
00
= and T
0i
= 0. The only nonzero entries of the dissipative energy
momentum contribution are
ij
=
_
i
u
j
+
j
u
i

2
3

ij

k
u
k
_
+
ij

k
u
k
, (6.7.4) 8,45
parametrized by shear viscosity and bulk viscosity . In a general frame, this leads to
= P

2
3

_
+
_
(6.7.5) 8,46
where P
:= g
+u
is the projector onto directions perpendicular to u
.
Charged uids have an additional conserved U(1) current
= 0 given by
j
= u
DP
. WHAT IS ??? (6.7.6) 8,47

In the uids rest frame, we rediscover Ficks law of diusion

j = D
.
6.7.2 Kubo formula from linear response theory
Let us now concentrate on the particular case when metric perturbatiions are time dependent
but homogeneous in space, i.e.
g
ij
(t, x) =
ij
+ h
ij
(t) , h
ii
= 0 (6.7.7) 8,48
g
00
(t, x) = 1 , g
0i
(t, x) = 0 . (6.7.8) 8,49
The velocity vector hence depends on time only, u
i
= u
i
(t). Consider the case where the uid
remains at rest at all times, u
= (1, 0, 0, 0).
In curved spacetime, equation (6.7.5) for the O() contributions to T
generalizes to
= P
_
+
_

2
3
_
g
_
. (6.7.9) 8,50
In the situation considered above, this simplies to
xy
= 2
0
xy
=
0
h
xy
. (6.7.10) 8,51
By comparison with linear response theory, we nd the zero spatial momentum, low frequency
limit of the retarded Green function of T
xy
:
G
R
xy,xy
(,
0) =
_
dt d
3
x e
it
(t)
_
T
xy
(t, x) , T
xy
(0,
0)
) = i + O(
2
) (6.7.11) 8,52
The associated Kubo formula is
= lim
0
1
Im
_
G
R
xy,xy
(,
0)
_
. (6.7.12) 8,53
Let us now explain the notion of hydrodynamic modes, determined by the poles of the retarded
Green function. They also give for instance the poles in the spectral function. Poles of cor-
relators are obtained from solutions of the linearized hydrodynamic equations, i.e. from plane
wave solutions e
it+i
kx
. Dissipation is described by complex with negative imaginary part.
figure spectral function and poles
Charge diusion is governed by the following dispersion relation,
_
t
D

2
_
= 0 = iD
k
2
, (6.7.13) 8,54
which determines a pole in the current-current Green function J
(p)J
(p)).
6.7.3 Shear modes and sound modes
Shear modes correspond to uctuations of pairs of components T
0a
and T
3a
where a = 1, 2.
T
3a
=
3
u
a
=

+P

3
T
0a
0
T
0a
=

+P

2
3
T
aa
= 0 (6.7.14) 8,55
For plane waves h e
it+i
kx
, we nd =
i
k
2
+P
.
Sound waves, on the other hand, are longitudinal uctuations of T
00
, T
03
, T
33
with speed
c
S
=
_
dP
d
and frequency
= c
S
k
i
2
_
4
3
+
_

k
2
+P
. (6.7.15) 8,56
In weakly coupled theories, the viscosity is governed by the mean free path
mfp
(nv)
1
where n denotes the density, the cross section for interactions and v a typical velocity. In
4
theory at nite temperature, one can derive perturbatively that n T
3
and
_
T
_
2
.
The viscosity is then obtained by multiplying with the energy density (for which a Stefan
Boltzmann law T
4
is assumed):

mfp

T
3
2
(6.7.16) 8,57
The entropy density scales in the same way with temperature, s T
3
, so the quotient
s

1
2
(6.7.17) 8,58
depends on only and becomes large at weak coupling 1.
6.7.4 AdS/CFT calculation of the shear viscosity
In order to compute T
xy
T
xy
) in the eld theory, we have to examine the propagation of the dual
graviton h
xy
in AdS spacetime. For this purpose, let us start from the Einstein Hilbert action
in ve dimensions. Consider a scalar metric uctuation h
xy
, denote it by in the following.
The quadratic part of the Einstein Hilbert action in is given by
o
quad
[] =
N
2
8
2
L
3
_
d
4
x dr
g
_
1
2
g
_
(6.7.18) 8,59
and gives rise to the linearized equation of motion
gg
) = 0. Perform a Fourier
transformation of the boundary coordinates, then a boundary condition of type (p, r = 0) =
0
(p) can be imposed. It is convenient to factorize
(p, r) = f
p
(r)
0
(p) (6.7.19) 8,60
where f
p
is called zero mode function and satises
_
f
p
r
3
_
p
2
r
3
f
p
= 0 , f
p
(0) = 1. (6.7.20) 8,61
An exact solution in terms of a Bessel function K
2
exists,
f
p
(r) =
(pr)
2
2
K
2
(pr) = 1
(pr)
2
4

(pr)
4
16
ln(pr) + O
_
(pr)
4
_
. (6.7.21) 8,62
The other solution (pr)
2
I
2
(pr) is ruled out since it blows up for r . The on shell action
for f
p
reads
o
quad
[] =
N
2
16
2
_
d
4
x
1
r
3
(x, r)
(x, r)
r=0
=
_
d
4
p
(2)
4

0
(p) T(p, r)
0
(p)
r=0
(6.7.22) 8,59
T(p, r) =
N
2
16
2
r
3
f
p
(r)
r
f
p
(r) . (6.7.23) 8,60
It then follows that
T
xy
(p) T
xy
(p)) = 2 lim
r0
T(p, r) =
N
2
64
2
p
4
ln p
2
. (6.7.24) 8,61
At nonzero temperature, the metric
ds
2
=
r
2
L
2
_
f dt
2
+ dx
2
_
+
L
2
r
2
f
dr
2
+ L
2
d
2
5
(6.7.25) 8,62
gives rise to the Hawking temperature T
H
=
r
H
L
2
. The entropy is given by the Bekenstein
Hawking formula S =
A
4
with A the area of the black hole horizon. For the density, one nds
s =
S
V
=

2
2
N
2
T
3
. (6.7.26) 8,62a
This is 3/4 of the entropy density in ^ = 4 SYM theory at vanishing tHooft coupling. Dene
a new coordinate u =
r
2
H
r
2
, then the boundary is situated at u = 0, the horizon at u = 1. In
terms of u, the metric is
ds
2
=
(TL)
2
u
2
_
f(u) dt
2
+ dx
2
_
+
L
2
4 u
2
f(u)
du
2
+ L
2
d
2
5
. (6.7.27) 8,62b
In real time AdS/CFT, consider again the factorization (6.7.23), then the equation of motion
implies (HIER IST ETWAS UNGUENSTIG, DASS f DOPPELT VERGEBEN IST...)
f
p

1 +u
2
u f(u)
f
p
+
W
2
u f
2
(u)
f
p

q
2
u f(u)
f
p
= 0 (6.7.28) 8,63
with shorthands W =

2T
and q =
k
2T
. Near u = 0, the two solutions behave as f
1
1 and
f
2
u
2
. In Minkowski space, there are two nite solutions near the horizon f
p
(1 u)
iW/2
and f
p
(1 u)
iW/2
. Any linear combination is possible, the solution is not unique! This is a
problem in dening the Green function.
A thorough analysis of the real time formalism in AdS/CFT leads to the result that the retarded
Green function is related to T by the same formula that was found at zero temperature:
G
R
(p) = 2 lim
u0
T(p, u) (6.7.29) 8,64
This has no contribution from the horizon, but to obtain f
p
, we need infalling boundary con-
ditions of course. The correlator T
xy
(p)T
xy
(p)) comes from h
xy
= with
p

1 +u
2
u f

p
+
W
2
q
2
f
u f
2

p
= 0 (6.7.30) 8,65
giving rise to the incoming wave solution f
p
(r) (1 u
2
)
iW/2
+ O(W
2
, q
2
). Using (6.7.29),
the resulting Green function is
G
R
(,
k) =
2
N
2
T
4
4
i . (6.7.31) 8,66
This is the famous Kuba formula for the viscosity
=

8
N
2
T
3
,

s
=
1
4
. (6.7.32) 8,67
Bibliography
pap1 [1] J. M. Maldacena, The large N limit of superconformal eld theories and supergravity,
Adv. Theor. Math. Phys. 2 (1998) 231 [Int. J. Theor. Phys. 38 (1999) 1113] [arXiv:hep-
th/9711200].
pap2 [2] E. Witten, Anti-de Sitter space and holography, Adv. Theor. Math. Phys. 2 (1998) 253
[arXiv:hep-th/9802150].
pap3 [3] S. S. Gubser, I. R. Klebanov and A. M. Polyakov, Gauge theory correlators from non-
critical string theory, Phys. Lett. B 428 (1998) 105 [arXiv:hep-th/9802109].
pap4 [4] D. Z. Freedman, S. D. Mathur, A. Matusis and L. Rastelli, Correlation functions in
the CFT(d)/AdS(d + 1) correspondence, Nucl. Phys. B 546, 96 (1999) [arXiv:hep-
th/9804058].
AG [5] O. Aharony, S. S. Gubser, J. M. Maldacena, H. Ooguri and Y. Oz, Large N eld theories,
string theory and gravity, Phys. Rept. 323 (2000) 183 [arXiv:hep-th/9905111].
DF [6] E. DHoker and D. Z. Freedman, Supersymmetric gauge theories and the AdS/CFT
correspondence, arXiv:hep-th/0201253.
PE [7] J. L. Petersen, Introduction to the Maldacena conjecture on AdS/CFT, Int. J. Mod.
Phys. A 14 (1999) 3597 [arXiv:hep-th/9902131].
Carroll:2004st [8] S. M. Carroll, Spacetime and geometry: An introduction to general relativity, Addison-
Wesley.
new [9] E. DHoker, D. Z. Freedman and W. Skiba, Field Theory Tests for Correlators in the
AdS/CFT Correspondence, [arXix:hep-th/9807098].
SS [10] Son & Starinets: Viscosity, Black Holes, and Quantum Field Theory, arXiv:0704.0240
EE [11] Erdmenger, Evans, Kirsch & Threllfall: Mesons in Gauge/Gravity Duals - A Review,
arXiv:0711.4467
79
80 BIBLIOGRAPHY
Lee [12] Lee, Minwalla & Rangamani: Three Point Functions of Chiral Operators in D = 4, ^ = 4
SYM at Large N, Adv. Theor. Math. Phys. 2 (1998) 697, [arXix:hep-th/9807098]
KRN [13] Kim, Romans & Von Nieuwenhuizen: The Mass Spectrum of Chiral ^ = 2, D = 10
Supergravity on S
5
, Phys. Rev. D 32 (1985), 389
HS [14] M. Henningson and K. Skenderis, Fortsch. Phys. 48 (2000) 125 [arXiv:hep-th/9812032].
CFTJEHO [15] J. Erdmenger and H. Osborn, Conserved Currents and the Energy Momentum Tensor in
Conformally Invariant Theories for General Dimensions, Nucl.Phys. B483 (1997) 431-474
[arXiv:hep-th/9605009]
CFTHOAP [16] H. Osborn and A. Petkos, Implications of Conformal Invariance in Field Theories for
General Dimensions, [arXiv:hep-th/9307010]

AdSCFT 2010 01 22

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

AdSCFT 2010 01 22

Uploaded by

Copyright:

Available Formats

Introduction to gauge gravity duality

. Solutions v to (2.1.4) are referred to as conformal Killing vectors, the

. The group multiplication

, the two- and three point functions of scalars

subject to the conservation law

= 0 in curved spacetime) generates the Noether currents associated with confor-

, so requiring invariance of W implies

obviously does not transform like the eld itself, so a connection A

transforming like that, we can use the covariant derivative D

can be deduced from its alternative denition as a com-

gives rise to interactions with the vertices

denote a set of elds with an adjoint

respectively. Supersymmetry now

and its right-handed counterpart

where the SL(2, C) indices , take values 1, 2

= SO(6). Exactly this is the isometry group

is the self-dual part

) of the eld strength, and the constants (C

: In addition to the Poincare algebra

, O] = 0 in bosonic conformal eld theory

on the whole manifold, so may be called the global

as Regge slope. Mass and angular

is the inverse string tension.

is a relativistic wave equation. In the gauge with h

can be identied with the metric of the target spacetime.

(X) . (3.2.1) 3,12

= 0. Remarkably, the vanishing of the

metric. This is the eective action for the

, of the closed string sector. From the form

eld may be identied with the target spacetime metric.

can be interpreted as the squared string length, so the

0 limit extracts the point

and its worldsheet

are spacetime vectors of two components spinors on the worldsheet,

denote worldsheet matrices for which one possible representation

= , the fermionic part of the action

where the parameter is an innitesimal constant Majorana spinor.

, which we had already discussed in bosonic

, massless excitations of D branes are described by supersymmetric

(i.e. at low energies when

. It has a p + 1 dimensional at hypersurface with Poincare invariance group

. For a stack of N coincident

it is a regular supergravity solution for y 0

, we get SYM TrF

about at space (with Minkowski metric ), the leading terms

0, therefore various dimensionless quantities such

is assumed to be small now, and the

expansion of supergravity is dual to a eld theory

. Consider an n point function of composite

is mapped to an IR cuto in AdS. It is natural to assume that ^ = 4 SYM lives on the

fall into the

(y) (4.2.7) 4,7

as vacuum expectation values of the eld

and the non-normalizable modes

as sources for these operators:

] for all correlators of single trace operators

in SYM is given in terms of the source elds

. The boundary values of these supergravity

] in the framework of type

(z) is located at the boundary.

(x) bulk-to-boundary propagator (4.2.16) 4,17

(z, x) J(x) bulk-to-bulk propagator (4.2.17) 4,18

reduces to the bulk-to-boundary propagator