Class Notes

Special Topics: Particle Theory

Aaron Hillman
November 9, 2016

Contents
1 Symmetries in Quantum Field Theory
1.1 What are the possible symmetries of a QFT? . . . . . . . . . . . . . . . .

2
2

2 Why Study CFT?

2.1 3D CFTs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2 4D CFTs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

4
4
5

3 Back to Symmetry
3.1 QFT . . . . . . . . . . . . . . . . . . .
3.1.1 Last Time . . . . . . . . . . . .
3.2 Quantization . . . . . . . . . . . . . .
3.3 The Emperors New Symmetries . . .
3.3.1 Dilatations . . . . . . . . . . .
3.3.2 Special Conformal . . . . . . .
3.3.3 The Conformal Algebra . . . .
3.3.4 Conformal Representations . .
3.3.5 Correlations Functions in CFTs

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

5
6
7
9
9
10
10
11
12
13

4 Radial Quantization

16

5 Unitarity and H.C.

19

6 Overview of AdS/CFT
6.1 What is AdS? . . . . . . . . . .
6.2 Boundary . . . . . . . . . . . .
6.3 Outline of the Correspondence
6.3.1 Path-Integral . . . . . .
6.3.2 Hamiltonian . . . . . . .

20
21
22
22
22
22

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

7 Scalar Fields in AdSd+1

7.0.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7.1 Boundary Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7.2 Interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

23
25
25
28

1 Symmetries in Quantum Field Theory

1.1 What are the possible symmetries of a QFT?
1. Spacetime symmetry
a) Translations: x x + a
b) Rotation/Boost: x x + x
These transformations comprise the Poincar Group and we can see how they
are realized on functions by exponentiating the generators of these transformations.
This is manifest as
p

F (x) ea

F (x) = F (x + a)

and this tells as that the generators of translations are the differential operators
(i.e. p = ). We can do the same for rotations
1

F (x) e 2 M F (x) = F (x + x)
and as such
= X X
M

We can realize these on a QFT Hilbert space via operators:

p |0i = M |0i = 0
where these are annihilating the vacuum. We can also look at the commutator of
our operators with other operators. For example
[p , O(x)] = O(x)
this is like the four-vector version of the Heisenberg picture of the Schrodinger
equation. Viewing this is as a differential equation, we can also write
O(x) = ex p O(0)ex p
Now, looking at rotations, we note that the action of M on an operator at the
origin will be different from its action somewhere else, because its a rotation. WE
have
[M , Oa (0)] = (S )ab Ob (0)

For a vector we could write

(S )ab = a b a b
We can get general transformations now if we use the rotations and translations.
We can rotate at the origin, and translate and just need the commutator of the
rotations with translations, which is captured in the symmetry algebra. This is
[M , p ] = p p
[p , p ] = 0
[M , M ] = M M + M M
employing Baker-Campbell-Hausdorff and grinding it out we have
[M , Oa (x)] = (x x )Oa (x) + (S )ab Ob (x)
This is the precise method of realizing the action of symmetries on a QFT Hilbert
space. We realize operators, and their commutation with operators generates transformation.
2. Internal Global Symmetries. These are global because they have nothing to do
with spacetime. Their symmetry is captured by some Lie Algebra (in particular,
the structure constants). Both these statements are captured by
[Ri , Rj ] = Fijk Rk
[Ri , p ] = [Ri , M ] = 0
then the commutations with operators is
[Ri , OIa (x)] = (Ti )JI OJa (x)
where I is a global index and a is a spacetime index.
Now we can ask, can there be any any symmetries which dont commute with
p and M ? In the 60s physicists were very curious about this question because
they wanted to understand the constraints of symmetries in QFTs so they could
understand things in their colliders. The Coleman-Mandula theorem (67) says
its very hard to find symmetries that dont commute. To understand the theorem,
lets create a QFT with some constraints:
a) Nontrivial scattering (particles cant just pass through each other).
b) Mass gap (Fock Space with a finite number of a particle types below any mass
m)
c) We are in > 1 + 1 dimensions.
d) The S-Matrix is analytic
if we have these constraints on a QFT then the theorem tells us: the only symmetries transforming as tensors are: p , M , Ri . The logic behind the proof of
this theorem is the following: We can look at some scattering process with some
interaction in the middle (two-to-two) scattering. We can then ask how Poincar

symmetries constrain this process. What we find is that the process is determined
up to the scattering angle. Suppose we have some extra conserved charge Q1 ,...,l ;
this charge would determine the angle. This is because there are 3 Mandelstam
invariants, and the Poincar symmetries leave one free parameter. This charge
would then leave only a discrete set of possibilities. The conclusion is that either
the theory is trivial or the S-matrix is non-analytic (because only a discrete set of
values are allowed).
There are loopholes around this
a) Conformal Field Theories (no mass gap no S-matrix)
b) Supersymmetry, where the conserved Q is not a tensor representation.
In particular, this is one of the motivations for studying conformal field theories.

2 Why Study CFT?

For one, CFTs are the building blocks of more general QFTs. We have also mentioned
that they are a loop hole in the Coleman-Mandula theorem. You may have heard of the
AdS/CFT correspondence.

2.1 3D CFTs
1. PHASE DIAGRAM OF WATER
We can define a correlation length of our fluid, which is a scale at which the
correlations die off exponentially. The correlation length will diverge at the critical
point since

1
|T Tc |

This behavior is universal (critical universality) and it turns that it is described by

a 3D conformal field theory.
2. We can also consider 3D Ferromagnets. In this case we have
X
H J
i j
hiji

depending on whether there is a preferred axis or the spins are free to rotate in
space we have 3D Ising or the O(3) model. In either case there is a critical value
of the coupling at which the correlation length diverges and we get scale-invariant
behavior. Whats so interesting is that the same 3D CFT describing the 3D Ising
model describes the critical behavior of water as well.
3. We could also just think of a scalar field in

Z
1
3
S= d x
()2 +
2

3D
1 2 2
1
m + 4
2
4!

at short distances with small, then the theory appears free as the energies are
much higher than the interaction energies (free in the U V limit). In the IR limit
2
this would not be true. Now if we tune m
2 to the right value then we get the
3D Ising CF T at long distances in IR (we get emergent scale invariance). We
can think of this point as being a transition between a massive phase and a higgs
phase.
4. The Landau-Ginsberg model of superconductivity is essentially the same thing
Z


S = d3 x ||2 + m2 ||2 + ||4
and we get the scale-invariance at the transition. This is called the O(2) model or
3D XY model.

2.2 4D CFTs
1. High energy physicists think about 4D CFTs. There are many proposals for physics
beyond the Standard Model which involve scale invariance. One puzzle in the
Standard Model: why are there flavor heirarchies (mass difference in electron and
top quark)?
2. QCD: An SU (NC ) gauge theory with NF quarks (QCD of the real world as NC =
3). We get quark confinement, bound states corresponding to protons etc. In the
infrared limit we think the behavior is conformal. Open question: what is the
conformal window of QCD? In particular
< NF <

11
NC
2

once we have this, the coupling is strong in the IR. We can show that when these
two values are close (say differing by ) then there is conformal behavior in QCD.
There appears to be a transition between conformal behavior in QCD and confined
(real world) behavior.
3. There are also SUSY Theories.
a) In 4D N=1 SQCD then we know the conformal window because of Seiberg
duality (electric-magnetic duality). When we close to the bottom of the range,
there is a dual system which is weakly coupled. ()
b) In 4D N=2 SQCD there is one unique point with conformal behavior NF =
2NC . This is captured by Seiberg-Witten theory.
c) We can also consider 4D N=4 SYM. This actually turns out to be a great
example of AdS/CFT correspondence.

3 Back to Symmetry
If we have a classical field theory with a symmetry (meaning a symmetry of the action S)
then this implies that there is a corresponding conserved current (Noethers theorem).
We can show this by  S = 0 for a symmetry transformation . We can define the

variation on space time as (x) and integrate by parts and fairly assume we can throw
away the boundary term
Z
Z
d

(x) S = d xJ (x) = dd (x) ( J ) = 0 up to classical E.O.M

J = 0
Note: since the J can be written in terms of the fields of our theory and we impose
that the fields die off, we can ignore the boundary term.
Once we have our conserved current for a choice of action and symmetry transformation, we can get a conserved charge by integrating over a d 1 dimensional surface
Z
Q=
dS J
d1

integrating the flux through a vector normal to that surface. Often we will integrate the
current contracted with normal vectors to surfaces of constant time. In fact, it does not
matter which time slice you use (we didnt know this a priori. In fact, we could deform
the surface and it would not matter, the charge is independent of up to topological
equivalence. Because of this, the operator is a topological surface operator.
Lets concretely consider the translational symmetry xr xr + ar . For this symmetry we get the local conserved current T (x) and corresponding charge
Z

P () =
dS T
d1

so this is the charge we associate with the generator of translations. This current is the
stress tensor and we can always take it to be symmetric. Its ambiguous how we defined
this tensor, and we note that we can also always change
T T + k F [k ]
this is called adding improvement terms, which will make it symmetric if its not. This
stress tensor is also the object which couples to the metric. As such, a trick for finding
it is
p
2 ( |g|L)

T =p
|g| g (x)
The thinking is that doing a translation with a spatially dependent ar (x) term is equivalent to perturbing the metric. This makes sense because its changing distances nonuniformaly. Now lets make this field theory a quantum field theory.

3.1 QFT
In QFT,
T

which is true as an operator statement up to "contact terms". We can think of this in

terms of Ward Identities
Z
DO1 . . . On eS[]
and lets consider a transformation + (x)f (, , . . .)
Z
h
i
(x) DO1 . . . On eS[]
but since we can actually view this as a change of variables, we would expect to get 0.
We get the following terms expanding
Z
(x) hO1 . . . On i + d4 x(x) h J O1 . . . On i = 0
this shift is now taking in potential shifts from the measure and the action. Without this
first term coming from contact terms we would get the conserved current immediately.
In the case of the stress tensor, we get
h T

O1 (x1 ) . . . On (xn )i =

n
X

(x xi )i hO1 (x1 ) . . . On (xn )i

i=1

so they are contact terms because they only matter if two terms come into contact, this
is manifest by the delta functions. In particular if we have a ball B with boundary B
then we can integrate over the ball.
Z
dyy hT (y)O(x) . . .i
B

and we can use Stokes theorem to relate to a surface integral so it is

= hP ()O(x) . . .i
we can also integrate the right side from the Ward Identity
hP ()O(x) . . .i = x hO(x) . . .i
because we pick up a delta function in the integration. So inserting the momentum
operator is like taking a derivative!
3.1.1 Last Time
1. The stress tensor T can be identified as the Noether current for translations. We
then define the translation generator
Z
P () =
dS T
d1

and we can deform the surface and get the same operator.

Now, we would like to know whether we can construct any other conserved charges from
T . If we were to build another charge, it should have the form
Z
Q () =
dS  (x)T (x)
d1

for a vector field  =  . We say the charge is conserved if we have

( (x)T (x)) = 0
and by the symmetry of the stress tensor and the fact that it is divergenceless, the above
is
1
(  +  )T = 0
2
from this we get Killings equation for Killing vector fields
 +  = 0
We can go through solutions we have. We have
( (x)) = p
we can find another solution with two indices
( (x)) = x x
= X X
M
there arent any other solutions (Prove this??). Now, what if we have some different
conditions on T ? For example what if we have traceless stress tensor
T = 0
In this case our Killing condition becomes
(  +  ) = c(x)
this is called the Conformal Killing Equation. We have new solutions satisfying this
equation, as it is a weaker condition on the vector field. We have
 (x) = X
= X
D
which generates dilatations. There is one more solution
( (x)) = 2X X X 2
2X (x ) X 2
these are the special conformal transformations.
Summary: If T = 0, then we have two new conformal symmetries, giving us all the

conserved charges below

Z

P =

dS T (x)

d1

Z
=

dS (X T X T )

d1

Z
D=

dS X T

d1

K =

dS 2X X T X 2 T

d1

3.2 Quantization
In QFT, we can interpret the same path integral as evolution of different Hilbert spaces.
We can think of the path integral as evolution of slices of constant time, each corresponding to Hilbert spaces. The path integral can describe different Hilbert spaces in the sense
that, since operators are topological, it wont care if we make a Lorentz transformation or
some other topological deformation. We could also consider surfaces of constant radius
and define our topological surface operators on these surfaces of constant radius. This is
radial quantization. We are interested in computing correlation functions, and these
different pictures give us different ways of viewing the correlation functions. WE have
Z
h1 . . . n i = D 1 . . . n eS = h0|T {1 . . . n }|0i
the meaning of "Time Ordering" and "vacuum state" will vary depending on our scheme.
In particular, time ordering is radial ordering in the radial quantization scheme. If we
have an operator insertion inside a surface we can stretch out the surface into a top
and a bottom flat surface and this give us two translation terms (one for each new surface
1 and 2 ). We then get
h(P (1 ) P (2 ))O(x)i = h0|T {[P , O(x)] . . .}|0i = hP ()O(x)i
= hO(x) . . .i
this gives us
h0|T {O(x) . . .}|0i = h0|T {[P , O(x)] . . .}|0i
and as a consequence
[P , O(x)] = O(x)

3.3 The Emperors New Symmetries

What do our dilation and special conformal transformations do? First we can consider
dilatations.

3.3.1 Dilatations
We can exponentiate this generator to get a dilatation and observe how to acts on a
function

eD f (x) = eX

f (x)

We can consider a few different functions to gain some intuition. First we consider
= x
f (x) = x Dx

eD x = e x
so we get a rescaled coordinate. How about something more complicated, like
x = 2x x
f (x) = x x Dx

eD x x = e2 x x
If we have

f (x) = (x x ) 2 eD f (x) = e f (x)

we call the scaling dimension of f (x). The first function have scaling dimension 1,
the second had scaling dimension 2.
In the context of field theory, we can consider a free scalar field (x)
Z
S = dd x()2
because we have our dilatation symmetry, this is invariant under
x ex
e
and so we get scaling dimension =
transformations.

d2

d2
2 .

Now we examine the special conformal

3.3.2 Special Conformal

= 2X (X ) X 2 = I P I
K
where P = and I performs inversions as
I : X

X
X2

we can show by comparing their actions on arbitrary functions of X that these yield the
same answer. This second expression provides a more geometrically intuitive picture of
what the special conformal transformation does.

10

We can deduce that a special conformal transformation has the form

K

ea

X =

X a X 2
1 2(a X ) + a2 X 2

all of this is captured by the conformal algebra.

3.3.3 The Conformal Algebra
There are two ways to view the conformal algebra, focusing on the killing vectors or
????.
For any killing vectors i , we have the operator algebra
[Q1 , Q2 ] = Q[1 ,2 ]
where the minus sign is there because differential operators act in the opposite order
from charges, in the sense
[Q1 , [Q2 , O]] = D2 D1 O
what about
[D, P ] = +P
taking the derivative raises something scaling dimension by 1. That is, P raises the
scaling dimension.
[D, K ] = K
Here we deduce K lowers the scaling dimension by 1. We also have
[K , P ] = 2 D 2M
These new commutations, on top of the ones we already new, are needed to understand
the conformal algebra (along with a couple not listed). This algebra is isomorphic to
SO(d + 1, 1) if we are dealing with a Euclidean conformal field theory. If we wick rotate
to a Lorentzian field theory, then we would get an isomorphism to SO(d, 2).
We can generate this isomorphism with a d-dimension rotation M matrix as a sub
matrix and L01 = L10 = D and Li0 = L0i = 12 (Pi + Ki ) and Li1 = L1i = 12 (Pi Ki ).
This hints at the notion of an embedding space which may be considered later. We can
now consider how our algebra acts on operators at the origin. WE recall that
[P , O(0)] = O(0)
[M , Oa (0)] = (S )ab Ob (0)
[K , O(0)] = 0 if O is a primary operator
[D, O(0)] = O(0) if O has scaling dimension
we can ask how the operator translates, what is scaling dimension is, whether is primary,
and how it rotates. To find an operator away from the origin we only need consider
O(x) = eX P O(0)eX P

11

 [D, O(x)] = ( + X )O(x)

Knowing these commutation relations in terms of differential operators will let us understand correlation functions, the observables of conformal field theory, in terms of
differential equations. Lets further examine them


[K , O(x)] = 2X + 2X S + 2X (X ) X 2 O(x)
and we note that the differential operator is the killing vector associated with K.
We can consider some finite transformation corresponding to a conformal killing vector
tilde
e : x x0
and we will find that the transformation is of the form

x0
= (x0 )R (x0 )
x
and so it is a combination of a rotation and a rescaling. We can also examine how the
matrix transforms. Well we know

x0 x0
=
x x

which is a local rescaling of the metric. So conformal transformations map right angles
to right angles.
We can now consider exponentiating a charge
U = eQ()
U ()Oa (x) U Oa (x)U 1 = (x0 ) D(R(x0 ))ab Ob (x0 )
where the the arrow is indicating a correspondence between left object in the path integral
formalism and the right object on a Hilbert spaces of constant time or radius (depending
on the picture.) So exponentiating the charge and acting on the operator in the path
integral, is a conjugation of the operator on the Hilbert space. We can now build up
conformal representations starting with a primary operator.
3.3.4 Conformal Representations
Operator
O
O
{ } O
..
.

Dimension

+1

..
.

We can consider the example of a free scalar theory and we can try to construct primary
operators. The first operator is the parent and the ones listed after are its descenendents
{, , , . . .}
{2 , 2 , 2 , . . .}

12

..
.
{n , n , . . .}
and we can get the stress tensor as well
{T , [ T ] , T = 0, T , . . .}
3.3.5 Correlations Functions in CFTs
Our questions is
hO(x)i = h0|O(x)|0i = ?
We know
hP O(x)i = h0|P O(x)|0i = hO(x)i = 0
because the vacuum should be translation invariant we would just apply our translation
to the vacuum. We can see how were starting to build a nice differential equation.
hDO(x)i = ( + X ) hO(x)i = 0
hO(x)i = 0

6= 0

intuitively we could think that if this operator has some scaling dimension, then having a
vacuum expectation value would set a scale and break conformal symmetry. So one-point
functions need to vanish. What about two-point functions? We consider
hO1 (x1 )O2 (x2 )i
which we take to scalar operators (spinless) for simplicity at the moment. Now, what if
we insert P and right down the Ward identity associated with P . By the ward identity
we get that
hO1 (x1 )O2 (x2 )i = f (x1 x2 )
meaning that the correlation must be a function of the distance only. Symmetry under
M tells us that it can only depend on the magnitude of the distance so we now say
hO1 (x1 )O2 (x2 )i = f ((x1 x2 )2 )
The action of D tells us
hDO1 (x1 )O2 (x2 )i = (1 + 2 + X1 1 + x2 2 ) f (x212 )
The solution is
f (x212 ) =

c12
(x212 )

1 +2
2

we can then examine how the special conformal transformation acts on this two point
correlation function. We will set one coordinate equal to the origin to make the algebra

13

simpler. We have
hK O1 (x)O2 (0)i = 0 = (2X (X ) + 1 ) X 2 )f (x212 )
there is only the asymmetry in 1 and 2 dependence because of our choice of origin,
which doesnt matter. The result is that
hK O1 (x)O2 (0)i = 2X (1 2 )f (x212 ) 1 = 2
because we want the above to vanish by the Ward identity. Lastly we can observe the
following,
hO1 (x1 )O2 (x2 )i = 2

I (x12 )
x2
12

where
I (x) = 2

x x
x2

and this has a nice relation to inversions, where

I:

x0
= (x0 )I
x

Lets provide a summary of what weve found about constraints on correlation functions
due to conformal symmetry
1. 1-point functions vanish
2. 2-points are very constrained. 2-point function sof two primary operators satisfy
hO1 (x1 )O2 (x2 )i = c

1 ,2
|x12 |2

where is the scaling dimension.

3. For spinning operators we get
hO11 ...l (x1 )O21 ...l (x2 )i

I11 (x12 ) . . . Ill (x12 )

=
1 ,2
|x12 |2

now we can ask about three-point functions. WE consider scalar primary operators for
simplicity. We consider
hO1 (x1 )O2 (x2 )O3 (x3 )i = f (x212 , x223 , x213 )
where we are saying that our three point function depends on a function of the invariant
distance because of translation and rotational symmetry in our theory. Then, dilation
symmetry can give us a differential equation for this function. This tells us we have a
function of the form
f=

123
a
|x12 | |x23 |b |x13 |c

14

with the constraint (a + b + c = 1 + 2 + 3 ). Then, special conformal symmetry tells

us that we cannot just have arbitrary exponents satisfying this equation, we must have
f=

123
|x12 |1 +2 3 |x23 |2 +3 1 |x13 |1 +3 2

and so we have fixed the three-point functions up to an overall constant 123 . We should
also note that the 2 and 3-point function constants are not independent of each other.
The constants for the 2-point functions could fix the 3-point function constants.
How about four-point functions? We can examine
h(x1 )(x2 )(x3 )(x4 )i
and inserting a dilation operator in front of the first and commuting it through the
different functions, we have
0=

4
X

( + Xi i ) hi

i=1

We know that h(x1 )(x2 )i solves the equation

2
X

( + Xi i ) hi = 0

i=1

This is useful, as
h1 2 3 4 i = h1 2 i h3 4 i
h1 3 i h2 4 i
h1 4 i h2 3 i
This tells us that we can actually multiply by any function of the cross ratios
u=

x212 x234
x213 x224

and

v=

x214 x223
x213 x224

we can workout that these are invariant under the conformal group, and its in fact quite
easy to see its true other than the special conformal transformation. And so we have
h(x1 )(x2 )(x3 )(x4 )i =

g(u, v)
2 2
x12 x34

This function is not independent from the data on scaling dimensions and spins in our
field theory, its related. We have {i , Si , ijk }, our scaling dimensions, spins, and scalars
for our three point functions. These are related via the operator product expansion
(OPE). This allows us to write our four-point function as a sum over 3-point functions.
In turn we get a conformal block decomposition. Before we can utilize this, we
should further develop radial quantization.

15

4 Radial Quantization
In our d-dimensional space, we can consider Hilbert spaces living on hyper-spheres of
dimension d 1. We have
1 (x1 ) . . . O
2 (x2 )}|0i
hO1 (x1 ) . . . On (xn )i = h0|R{O
If we do the path integral through a sphere with no operator insertions, we get the
vacuum. Lets make a connection between a state in the Hilbert space and the variables
we like to use in the path integral. We consider a basis of field eigenstates {b } on S d1 .
Z
|i =
Db |n i hb |i
B

What are these expansion coefficients?

Z
hb |0i =

D(r, ~n)eS[]

rr0

where r0 is the radius of the sphere we are integrating throughout and we have the
boundary condition (r0 , ~n) = b (~n). We can also examine overlaps in the case where
we have some operator insertion,
Z
hb |O(x)|0i =
D(r, ~n)O(x)eS[]
rr0

again with boundary condition (r0 , ~n) = b (~n). For ever local operator O we can define
a state labeled by the operator
|Oi = O(0) |0i
which is the state you get by inserting the operator at the origin and having it act on
the vacuum of radial quantization. This is analagous the Heisenberg picture of quantum
mechanics where my operators are radially dependent and my states are radius independent. We also want to say that for every state we can define a local operator and this
way we get the state operator correspondence.
We can consider the dilatation operator. For every state
D |Oi i i |Oi i
we can define a local operator via cutting out a ball around the state and gluing in
an operator like a local operator insertion and shrink the new operator ball down with
conformal transformations (and imposing the right boundary conditions on these balls).
(CLARIFY THIS). In an equation, we might say
Z Y
hO1 (x1 ) . . . On (xn )i =
Dbi hbi |Oi i
Zi
=

D(x)eS[]

i =bi

16

where x
/ Bi . The part of the path integral which goes inside the operator balls is
captured in the state |Oi i. The point of having eigenstates of the dilatation operator is
that we can do a conformal transformation and get
P

hO1 (x1 ) . . . On (xn )i

so that the Bi are not overlapping and our object is always well defined. So at the end
of the day we can map operators to states and states to operators. This is the state
operator correspondence.
To reiterate, given an operator at the origin we can define a state by acting on the
vacuum
O(0) O(0) |0i |Oi
We can consider a primary state, where it is an eigenstate of the dilatation operator:
K |Oi = 0
D |Oi = |Oi
M |Oi = S |Oi
and we can make a multiplet of states and their descendants
{|Oi , P |Oi , P P |Oi , . . .}
and we have
O(x) |0i =

X 1
(x P )N |0i
N!
N

Remark Radial quantization on Rd is equivalent to ordinary Equal "time" (Euclidean

time) quantization on R S d1 (a cylinder). This picture can give us a lot of intuition
for what happens when we do radial quantization.
Lets examine metric in this cylinder case
ds2Rd = dr2 + r2 ds2Sd1
 2

2 dr
2
=r
+ dsSd1
r2
the idea is to show that this metric is conformally related to the metric on the cylinder.
We use r = e and we get


ds2Rd = e2 d 2 + dSS2d1
so these metrics are equivalent up to a local rescaling. We can picture our cylinder, and
we see relations

r
0
1

17

we can also consider how the conformal group acts on each of these. We see
r
r r
r 1/r

D
I

+ log

We can also examine how our correlation functions change going into this new conformally equivalent geometry. We have, including normalization in the denominator
!
Y
hO1 (x1 ) . . . On (xn )i
hO1 (x1 ) . . . On (xn )i
i
=
(xi )
h1ig
h1i2 g
i

This motivates us to define a cylinder operator

Ocyl (, ~n) = e Oflat (~x = e ~n)
and now we can start asking about correlation functions of these cylinder operators. For
example, a 2-point funciton
h0|Ocyl (2 , ~n2 )Ocyl (1 , ~n1 )|0i
These should be some linear combination of states with energies EN = O + N and we
get for the correlation function
X
CN (~n1 , ~n2 )eEN (2 1 )
N

this is computing the overlap between the two states, and in order to do that we need
to evolve by exponentiating the Hamiltonian. The farther apart these are in Euclidean
time, the more damped they will be. Moreover, these energies are the way they are
because of the relation between dilatation and time. Taking the power law for the CFT
2-point function, we combine this and get that the correlation function here is
(r1 /r2 )
|1 2 rr12 (~n1~n2 ) + (r1 /r2 )2 |
where


r1
r2

= e(1 2 )

proving this is a good exercise. We can produce the previous sum by taylor expanding
the denominate in power of r1 and r2 . Moreover, from this expansion we could begin
computing the coefficient CN . We can think of this correlator as capturing the overlaps
between the desendents of each insertion. We may write
CN (~n1 , ~n2 ) = hN, ~n2 |N, ~n1 i
1
where |N, ~ni =
(P n )N |Oi
N!
To be able to do this computation, we need to be clear on how P transforms when we
take its conjugate. We have not clarified this point, but it is now in order. This brings
us to discuss unitarity.

18

5 Unitarity and H.C.

These properties will depend both on the metric signature and the quantization scheme.
Starting with something we are familiar with, we may consider a Lorentzian QFT. In this
case we want to demand that the charges are Hermitian operators so that they generate
unitary transformations We can consider

OL (t, ~x) = eiHti~x PL OL (0, 0)eiHt+i~x PL = OL

(t, ~x)
~

as long as we have a basis of Hermitian operators. The statement is that H = H ,

P~L , = P~L . We can do a Wick rotation and go to Euclidean theory (t = itE ). We can
start simple and consider a scalar operator
OE (tE , ~x) = OL (itE , ~x)
expanding this out we get

eHtE i~x PL OL (0, 0)eHtE +i~x PL = OL

(t, ~x)
~

now if we take the dagger we get

(tE , ~x) = OE (tE , ~x)

OE

this would be more complicated with spindices.

19

6 Overview of AdS/CFT
The idea is that we have a correspondence
CF Td Quantum Gravity in AdSd+1
There are multiple things we can match up between these two sides.
CF Td
Conformal Group SO(d, 2)
Dilatations D
Operator s
"single-trace operators"
"multi-trace operators"
Scalar operator
Current J
Stress-tensor

AdSd+1
Isometries of AdSd+1
Hamiltonian H (time translations)
Energies of states in AdSd+1
1-particle states (fields in AdS)
multi-particle states
Bulk scalar
Bulk gauge field A
graviton g

Remark If we have a true CF T on the boundary we do not necessarily need to have

Einstein gravity in the bulk, but we will not have dilatonic gravity or antisymmetry. The
difference would be that we can have higher order curvature terms in our action. This
can be seen as coming from tracelessness of the T on the boundary.
One of the most studied examples is N = 4 Supersymmetric Yang-Mills (SYM). In this
case we have an SU (Nc ) gauge group in d = 4. We have
Nc (RMl )3/2
where R is some length scale and not the Ricci curvature and M is the AdS planck
scale.. This tells us about measuring our interactions in units of this R. Making the
Planck scale big corresponds to making gravity very weakly coupled (like the real world)
and this corresponds to making Nc very big, which regulates the strength of interactions.
What is meaningful is the relationship between R and M since an individual length scale
is not meaningful on its own.
Remark Supersymmetric Yang-Mills is like pure Yang-Mills but we had fermions and
massless particles to make it supersymmetric.
Another important parameter, the Thooft coupling, has the relationship
g2N
=

16 2

 4
R
ls

where ls is the string scale regulating the excitations in the bulk, and in particular the
gaps in the spectrum. So there are two things on the gauge-theory side, Nc and g,
controlling two things on the AdS side, R and ls .
If we look at the limit Nc >> 1 and >> 1 we have a weakly coupled effective field theory
in the bulk. If we have a big gap in the bulk then we have a good effective field theory
description, so we have a good description for weak coupling. Where supersymmetry is
helpful is that it gives us a vanishing beta function no matter what we tune the coupling
to and so we still sit at a conformal fixed point, with no RG flow. Now that we have a
general notion of AdS/CF T , we can explore the correspondence in more detail.

20

6.1 What is AdS?

A useful way to describe AdS space is the embed this space in higher dimensions, much
like embedding S 2 in R3 . We will start in a flat space with a weird metric signature
( + + + +) with d +s and 2 minuses. The line element is
2

ds =

dX02

d
X

2
dX2 dXd+1

=1

we may consider the (d + 1)-dimensional slice:

X02 +

d
X

2
X2 Xd+1
= R2

=1

and we call R the curvature scale. The left-hand side is just the invariant magnitude
for this space, and so this subspace is invariant under SO(d, 2) transformations. This R2
will set the scale of the cosmological constant in our AdS gravity theory. For d = 0 we
just have a circle, but if we start including the other direction we get hyperbolas. You
can picture moving a circle and shrinking and then again growing its radius to make a
solid hyperbola in 3D. We want to look at theories on this space.
It is prudent to introduce global coordinates




cos t
sin t
X0 =
R, Xd+1 =
R
cos
cos



X = tan() R,
= 1
These s parameterize a d-sphere. If we take this parameterization and calculate the
metric on space we have
ds2AdS =

R2
dt2 + d2 + sin2 d2d1
2
cos

and we restrict 0 < 2 since we see = 2 creates a divergence. We will see that
= 2 will define the boundary. We note that t was parameterizing a circle as a cyclic
coordinate. We willy actually allow < t < . This picture will give us a cylinder
moving up with t which we will identify with time and will allow us to make nice fieldtheory connections. Then is the radius of this cylinder and on the boundary of the
cylinder at constant t we have angular directions. We note that d2d1 is d 1 because
we have the condition = 1.
We can now discuss the symmetry in our space, which is SO(d, 2) invariant. We have
the special conformal generators
J = i(X X )
We can identify with conformal group elements and then use the parameterization introduce
D = J0,d+1 = i(X0 d+1 Xd+1 0 )

D=i
t

21

and this way we recover a correspondence between dilatations and time translations in
AdS global coordinates.

6.2 Boundary
We can discuss the boundary of this space as we take

and we get

ds2bdy (dt2 + d2d1 )

where we have coordinates on a cylinder R S d1 , which is analogous to the cylindrical
quantization we developed earlier. And so we already see based on symmetry a connection between an AdS bulk theory and a CF T in radial quantization on the boundary.
Lets ask, if we send a beam of light out toward the boundary, what happens? Well, we
are looking at a massless geodesic in AdS meaning ds2 = 0. Well, we are free to fix our
angle in doing so, and then null geodesic implies
dt2 = d2
so we can send single out to the boundary and back again in a finite amount of time on
45 angles in , t coordinates. In flat space these light rays would just go off to infinity,
but AdS in these coordinates is in some sense like a finite dimensional box where light
can bounce back in a finite amount of time, and this is important for its usefulness.

6.3 Outline of the Correspondence

We can consider putting a scalar field (r, , ) in AdS. There are two formulations of
the correspondence, the Hamiltonian and the path-integral formulation.
6.3.1 Path-Integral
This formulation tells us
Z

AdS


(= ) =
2

DeiS[] = hei

/2 O

/2

where the right side is an expectation value in the CF T for an operator, which is sourced
by the field on the boundary.
6.3.2 Hamiltonian
We can consider quantizing on slices of constant t. We can define some operators

X
, ) =
(r,
i (r, , )ai + i (t, , )ai
i

The states correspond to some Fock space

|0i a |0i a a |0i . . .
and the correspondence says that this Fock Space is the CF T Hilbert space in radial
quantization.

22

7 Scalar Fields in AdSd+1

We can write the action
Z
S=

dd+1 x g g + m2 2

we can derive the equation of motion as usual and we get the wave equation
 = m2
where

1
 =
gg
g
and so our objective is now to solve this wave equation in AdSd+1 . In global coordinates
R2  = cos2 tt + cos2 2 + (d 1) cot + cot2 2S d1
Where the nabla defines the laplacian on a a d 1 sphere. It obeys
=

a very detailed solution can be found in arxiv:hep-th/9805171. The idea is that we find
some nice factorized ansatz and break it up into little pieces. We can try a solution of
the form
= eit Yl{m}
~ ()()
where would correspond to a definite energy of the field and we intuit a spherical
harmonic component from the spherical laplacian and tack on a dependence as well.
These spherical harmonics are eigenfunctions of the spherical laplacian obeying
2S d1 Yl{m}
~ = l(l + d 2)Yl{m}
~
and thanks to this, we can take a three-variable differential equation into a one-variable
differential equation which looks like




1
l(l + d 2)
m
2
d1
2
(tan ) + w =

=0
cos2
(tan )d1
sin2
and m
is such going to the mass in units of the AdS curvature scale. That is m
mR.
This messy, second-order differential equation is solved by a hypergeometric function.
We get


+ + l w + + l + w
d
() = (cos )+ (sin )l F1
,
, l + , sin2
2
2
2
and we require
d
m
= ( d) =
2
2

r
m
2 +

d2
4

23

in order to be able to find a solution to this equation. a priori it seems like would
also provide a solution, but this leads to a negative power for sin which makes the
solution blow up at the origin. We can consider the behavior at the boundary ( 2 )
and we get
() C+ (cos )+ + C (cos )
with
1

C+

+lw
2

+ +lw
2

We would think that we can have general values of , but not all will be well-behaved.
Another way to think about the energy is
Z

E = dd x gT 00
and
q eq require that it be finite; not all values of produce this. We introduce =
2
m
2 + d4 and we say that if > 1 then we require that C vanish or else will not die
off sufficiently fast. To make C vanish we need the Gamma function in the denominator
to approach infinity and so we choose so that we are at a pole of the gamma function.
This gives us a quantization condition
= + + l + 2n
It is worth noting that the same quantization condition would come from making the
energy flux into the boundary vanish, i.e.
Z

g
n T 0 = 0

If 0 < < 1 then both the + and behaviors are allowed, each corresponding to
different boundary conditions that you might impose and your scalar field. What we
mean by different boundary conditions, we mean that, when we wrote the actions S in
the beginning, we could change the action by some boundary term. E.g. writing versus
would lead to different conditions on C+ and C . We get
C = 0 = + + l + 2n
C+ = 0 = + l + 2n
We can normalize using an inner product
Z

0
0
0
(nlm
,

)
=
dd x gg 00 nlm
~ 0 n0 l0 m
~ 0 = nn0 ll0 m
~m
~0
~
nlm
~
where this integration is happening over a single time-slice, hence the d-dimensional
differential. Where the field solutions have energy label nl = + 2n + l.
1) Solutions make sense for all
m
2 >

d2
4

negative masses

24

this is called the Breitenlohner-Freedman Bound. This comes from asking under
what conditions the vacuum is stable. This is analagos to notions of centrifugal or
effective potentials experienced by particles. For a particle sitting in AdS space,
an effective potential is felt due to the negative curvature of the space.


2
2) Smallest is d2 1, solution at m
2 = 1 d4 . Becoming smaller than this
would require > 1. We note that this is precisely the CF T unitarity bound.
3) We can check 0,0 is "primary" in the sense that this annihilated by the K generator. On the other hand, the other n,l s are not primary. The functional form
we would have


00 eit cos
So, to get operators from this fields, we can expand

X

, ) =
nlm
(t,
~ anlm
~
~ + nlm
~ anlm
nlm
~

where anlm creates a 1-particle state w/ energy nl = + 2n + l and angular momentum

{l, m}.These
~
states give us a nice correspondence with CF T states,
1
|n, l, mi
~ = anlm
. . . l 2n O(0) |0i
~ |0i CFT states

and so we can say that each wave-function in the bulk corresponds to some CF T state
on the boundary. We can extend this to multiparticle states and operators next time.
7.0.1 Summary
So we put a massive scalar field in the bulk of AdS and got a wave equation. We then
solved for a set of normalizable solutions of this wave equation and got an expression for
the field operator in the quantum theory as a function of AdS coordinates which we could
expand in basis of creation and annihilation operators times our basis of solutions to the
wave equation. Now how do we move forward? Well we could begin adding interactions
to our theory. But the way to really get to the meat of the correspondence is examine
correlation functions and try to look at these correlation functions in the context of CFT.

7.1 Boundary Operators

The way to move forward is to take a bulk field operator and then get a boundary
operator O(t, ). We want to define this boundary operator by
, )
O(t, ) = lim (t,
2

Now, because of the cos() behavior in the n,l,m

~ functions we found, this limit would
vanish at the boundary. So really we define the limit as
, )
O(t, ) = cO lim (cos ) (t,
2

25

We note that this limit may not be well-defined in the strongly couple regime, but we
wont need to worry about that here. We also note the presence of a cO constant which is
not too important, but its typically defined to remove dependence on the AdS curvature
scale, R. We will define boundary coordinates collectively as b and bulk coordinates as
x. Now, we have various propagators
1 )(x
2 )i = GBB (x1 , x2 )
hT (x
1 )O(b2 )i = GB (x1 , b2 ) = CO lim (cos 2 ) GBB (x1 , x2 )
hT (x

2 2

hT O(b1 )O(b2 )i = G (x1 , b2 ) = CO lim (cos 1 ) GB (x1 , b2 )

1 2

So we want to think about GBB (x1 , x2 ) which only depends on the geodesic distance
between x1 and x2
#
"
Z x2

1 cos(t1 t2 ) sin 1 sin 2 1 2
ds = R cosh
S(x1 , x2 ) =
cos 1 cos 2
x1
and so our expectation is dependence only on this invariant distance. Now, how can we
calculate this bulk-bulk propagator? Well one way is simply by brute force. We have the
expansion of the fields and we can use canonical commutation relations of the creation
and annihilation operators. We get
X

GBB (x1 , x2 ) =
nlm
~ (x1 )nlm
~ (x2 )
if we do some hard work, we can compute this sum directly. This is done in a paper
by Gary and Giddings 0904.3544. It is should be familiar from flat-space solutions of
Feynman propagators that this propagator is a Greens function obeying
( m
2 )GBB = (x1 , x2 )
In flat-space we have different pole prescriptions: retarded, advanced, and Feynman. In
this case we are discussing Feynman. This way we can get an expression for the bulk-bulk
propagator


cBB
+1
d
1
GBB (x1 , x2 ) =
,
,

+
1;

i
F
1
2
2
2
cosh2 (s/R)
cosh (s/R)
()
cBB = d1 +1 d/2
R 2
( d/2 + 1)
it is interesting to note that this formalism is quite general, and we could perform these
calculations to look at free field theory near a black hole or deSitter space or any other
geometry.
Now we focus on the bulk-boundary propagator. This is a useful object since it
obviously tells us how to travel from bulk to boundary. Its also useful since it tells how
to compute
Z

(x1 ) = GB (x, b)O(b)

26

We can think of these boundary operators as giving some boundary conditions for these
bulk fields which we convolute with this Greens function. The logical flow of this is a bit
murky, but its an important conceptual point that this correspondence between fields
exists. Now the thing we want to focus on is that given a bulk description, how do we
reconstruct all the boundary information.
So another direction we can take is using GB to evaluate CF T correlation functions in terms of bulk data. We might have the correlation function
hO(b1 ) . . . O(bn )i
Say in the case of 4-point function on the boundary, we can imagine propagating into
the bulk and computing all the bulk Feynman diagrams to get a result. We can also
write this as an integral
Z Y
n

dxi GB (bi , xi )GT (x1 , . . . , xn )

where GT is a truncated, Bulk n-point function where we have stripped off the external
legs. So we will need the GB explicitly, and we can get it from the limit relation before.
Explicitly we find that it is
"
0

GB , t, ; , t , = CO CBB

#/2

cos2

0
cos(t t0 ) sin

and from this we can then get G and we have

G (b, b0 ) = c2O cBB 

1

0
cos(t t0 )

 =

const
b2
12

to get in the standard coordinates we would have to use a conformal mapping from
R S d1 to Rd1,1 which will send


2
2 = r2 dr2 + r2 d
dt2 + d
r = eit
and so this conformal mapping, leaving the theory invariant, gives us the standard metric
and lets us write the correlation function the way we want. This is all to show that if
we compute the propagator in the bulk that we recover the standard form we are used
to in CFT for two-point functions of local primary operators.
Its also interesting to look at higher point functions from points on the bulk. But
in this free theory, there is no way to compute three-point functions since we can only
connect two of the legs. And so three-point functions for these identical operators must
vanish, so OOO = 0. So we move to 4-point functions. There are interesting bulk
diagrams to compute, three ways to connect the four legs in-fact. These bulk Feynman
diagrams are referred to as "Witten Diagrams". We see that these free theory four-point
functions are products of two-point functions. So
hO(b1 )O(b2 )O(b3 )O(b4 )i = hO(b1 )O(b2 )i hO(b3 )O(b4 )i + 2 more permutations

27

1
2
b2
12 b34

1
2
b2
13 b24

1
2
b2
14 b23

which looks like the four-point function for a "generalized free field". We can use our
usual techniques if we like to decompose this into conformal blocks and we will get
#
"


X
1
z z

+
(z
z

)
=
1
+
2n,l g2+2n+l,l (z, z)
1
+
2
(1

z)(1

b2
b
12 34
n,l

And we can think of

primary operators (n, l) = 2 + 2n + l 2-particle states in AdS
X
O 1 . . . l 2n O(0)
hOOOn,l i n,l 6= 0
O(x)O(0) 1 +
n,l

this general sturcture where correlation functions factorize is a general property of large
N theories. If we have some large N theory with many degrees of freedom, if you look at
their correlation functions they will generally always factorize. So an understanding of
this structure is quite useful. Another thing we can note is that there is no stress tensor,
which is another thing generic in large N CFTs (because there is no spin-2 operator of
dimension d). One way of appreciating this is that

OO T
CT
and generally the central charge CT grows with N and so large N decouples the stresstensor. To get a stress-tensor back we add gravity.

7.2 Interactions
We can consider interactions
S = S0 + Sint
H = H0 + Hint
A concrete example would be
Sint

=
3!

dd+1 x g3

and then we might ask for the three point function

Z
hO(b1 )O(b2 )O(b3 )i = hO(b1 )O(b2 )O(b3 )

dd+1 x g 3 i
3!
0

And in terms of propagators, we get a diagram looking like a peace sign and the contribution is
Z

dd+1 x gGB (b1 , x)GB (b2 , x)GB (b3 , x)

28

We can also ask about the correction to the four-point function

hO(b1 )O(b2 )O(b3 )O(b4 )i
and we have sums over the s, t and u channels. These diagrams have two vertices, and
for the s channel we get a contribution
Z
Z

2
d+1

d x g dd+1 y gGB (b1 , x)GB (b2 , x)GBB (x, y)GB (b3 , y)GB (b4 , y)
and we get similar contributions for the t and u channel diagrams
We can also examine 4 with interaction
Z

Sint =
dd+1 x g3
4!
We can see that the three point function correction must vanish by symmetry considerations. Then the tree-level four-point function gives us
Z

dd+1 x gGB (b1 , x)GB (b2 , x)GB (b3 , x)GB (b4 , x)

these show up often in holography computations and are called D functions. In general
they are of the form
Z

N
1
D1 ...N (b1 , . . . , bN ) = dd+1 x gG
B (b1 , x) . . . GB (bN , x)
where the i labels are the scaling dimensions of the boundary operators.
Remark How do you we know we still get a CFT on the boundary after we introduce
interactions? Well the interaction will transform nicely SO(d, 2) transformations and so
we dont lose the symmetry we had before. For example, we identify the Hamiltonian
with the Dilatation operator and we have no explicit time-dependence so this is preserved.
Now for these D-functions, we could consult a text to find tricks for evaluating them.
For N = 3 we would get
D1 2 3 (b1 , b2 , b3 ) =

const
1 +2 3 2 +3 1 1 +3 2
b
b23
b13
12

So if we have a specific CF T and we know three-point function coefficients for the CF T

we can read off information about the bulk theory. Going further, we can consider a
four-point function of the form
D (b1 , b2 , b3 , b4 ) =

const
D
(u, v)
2
b2
12 b34

are called reduced D-functions and are functions of the conformal crosswhere the D
ratio. You can take these reduced D-functions, expand them in conformal blocks, and
read off corrections. Thats quite messy, but theres also a much easier route to computing
the corrections to the CF T data. This is by using old-fashioned perturbation theory.
One way to think about the perturbative expansion were doing is to look at
H = H0 + Hint

29

|i = |i0 + |i1
E = E0 + h|0 Hint |i0
and in our theory we are identifying the Hamiltonian with the dilatation operator. We
can identify energy corrections with corrections to the scaling dimensions of boundary
operators. So we can perturb away getting
(n, l) = 2O + 2n + l(n, l)
where (n, l) is an anomalous dimension acquired and we compute it by
(n, l) = hn, l|2 Hint |n, li2
where we label with 2 because we are looking at 2-particle states. We can normalize
these states such that
hn, l|2 |n0 , l0 i2 = nn0 ll0
then working in 4 we get
(n, l) = hn, l|2

4!

dd x g4 (x)|n, li
2

we note that we have a d-dimensional integral because we are writing the Hamiltonian
and not integrating over the time direction. The Hamiltonian generates the time translation. We also recall the mode expansion of in the free theory in terms of creation
and annihilation operators and we get
Z

(n, l) =
dd x g hn, l|2 2 (x)|0i h0|2 (x)|n, li2
4
So we have reduced this computation to the computation of these 2 (x) inner products.
These are like two-particle wave-functions and we have means of computing these. We
use the following trick: use the primary property of states and annihilate these states
with K . In particular we say K annihilates this two-particle state. So what is K ?
We can write it as
K = J,0 + iJ,d+1

(J = i(X X ))

We can see that this kills the combination

1
R
X0 + iXd+1 = eit cos
where we use the definitions of the embedding space coordinates. We conclude from this
that
h0|2 (x)|n, li2 = F (eit cos )
Remark In general we are perturbing the dilatation and special conformal operators,
but in this old fashioned perturbation theory to first order we are allowed to just use the
free theory, so its not a worry.

30

We note that the expression has no angular dependance, so l = 0. To argue this explicitly
we can write down the rotation generator as an operator in AdS global coordinates, and
we would see that this differential operator would have to give something non-zero if l is
non-zero. So from that we deduce l = 0. Next, we can figure out this function F up to
a constant because we know the energy and the exponent of it must have a a power of
the energy. So we write
h0|2 (x)|n, 0i2 =

2 +2n
1
eit cos O
Nn

We can get the normalization in specific cases

(
2
q

Nn =
2O +2n3
2
(O +n1)

(n+1)(2o p+n3)

d=2
d=4

carrying out the computation for this anomalous dimension we have

(n, 0) =
4

2
Z/2  d1  
sin

1
2O +2n
d
d
(cos
)
Nn
cosd+1
0

this integral can actually be straightforwardly done and we would get

(n, 0) =

d/2 (2O + 2n d/2)

Nn2
(2O + 2n)

its interesting to ask how this behaves as we start varying n. Well, it depends on the
spacetime dimension.
(

n d = 2
(n, 0) = 8(2O +2n1)
n
d=4
and so in four dimensions perturbation theory can break down. Since n is identified
with the energy we can think of this meaning that perturbation theory breaking down
when these two particles have a lot of radial motion and momentum energy. What we
are seeing is that in two dimensions we have a renormalizable interaction and in four
dimensions this interaction is non-renormlizable. This is like effective field theory, where
we can get non-renormalizable interactions and our theory will break down.
Remark On the topic of dimensions
This basically tells us that if we have an effective field theory in AdS then we have a
description of the low dimension () sector but i tbreaks down for the large operators.

31