Quantum Field Theory in Curved Space-Time

Pieter Bomans
Mathematical Institute
University of Oxford

These notes are meant to complement the lectures and may be updated over the course of the term.
Feedback is very welcome, especially if there are typos or places where the text lacks clarity. Please
send any comments, corrections or questions to
[email protected].
 About this course

Welcome to the 2023-2024 edition of Quantum Field Theory in Curved Space-Time of the Oxford
Mathematical and Theoretical Physics master course. This course builds on the courses Quantum
Field Theory, Advanced Quantum Field Theory and General Relativity I and II. In these courses, the
fundamental concepts of quantum field theory and general relativity were introduced. Here we will
further develop these subjects more broadly by applying the tools of quantum field theory in some
of the curved space-times encountered in general relativity. Along the way several concepts from
differential geometry, representation theory as well as special functions will arise. Familiarity with
these topics is not required but will be helpful.

Synopsis

The course will consist of the following topics:

• Quantum field theory in flat space

• Lorentzian geometry and causality

• Quantum field theory in curved space-time

• The Unruh effect

• Hawking Radiation

• Quantum fields in de Sitter space

• Quantum cosmology

• Quantum fields in Anti-de Sitter space

• Holography

i
References

There are a variety of excellent textbooks and lecture notes available in the literature. This course
borrows from a number of them, the most relevant ones are listed below.
Standard textbooks on quantum field theory in curved spaces are:

• N.D. Birrell and P. Davies, Quantum Fields in Curved Space, (Cambridge University Press 1982)

• R. Wald, QFT in Curved Space-time and Black Hole Thermodynamics, (Chicago University Press
1994)

• V. Mukhanov and S. Winitzki, Introduction to Quantum Effects in Gravity, (Cambridge University

Press 2007)

• L. Parker and D. Toms, Quantum Field Theory in Curved Spacetime, (Cambridge University Press
2009)

The following lecture notes are also recommended for reference:

• L. Mason, Quantum Field Theory in Curved Space-Time, Lecture notes.

• M. Mariño, QFT in curved space, Lecture notes.

• L.H. Ford, Quantum Field theory in curved spacetime, arXiv:gr-gc/9707062.

• P.K. Townsend, Black Holes, arXiv:gr-gc/9707012.

• T. Jacobson, Introduction to Quantum Fields in Curved Spacetime and the Hawking Effect, arXiv:gr-
qc/0308048.

• E. Pajer, Field Theory in Cosmology, Lecture notes .

• N. Arkani-Hamed and Y. Kats, Lecture Notes on Quantum Mechanics and Spacetime. (PDF version
available upon request)

• C. Bär and K. Fredenhagen, Quantum Field Theory on Curved Spacetimes, (Springer-Verlag

Heidelberg 2009), DOI.

• D. Anninos, De Sitter Musings, Int. J. Mod. Phys. A 27 (2012), arXiv:hep-th/1205.3855.

• J. Penedones, TASI lectures on AdS/CFT, TASI 2015, 75-136, arXiv:hep-th/1608.04948.

ii
 Contents

About this course i

Synopsis i

References ii

I FUNDAMENTAL ASPECTS 1

1 Why you should take this course 2

1.1 What is in this course? 4

1.2 What is not? 5

2 Quantum fields in flat space 7

2.1 Canonical quantisation in Minkowski space 7

2.2 Particle interpretation 10

2.3 Vacuum energy 12

2.4 Two-point functions 13

2.5 Charged scalars, gauge fields and spinors 17

3 Lorentzian geometry 20

3.1 Lorentzian manifolds 20

3.2 Causal structure 21

3.3 Conformal infinity and Penrose diagrams 26

3.4 Asymptotics and peeling 30

4 Quantum fields in curved space 33

4.1 Classical fields in curved space-time 33

4.2 Canonical quantisation in curved spacetime 36

 4.3 Quantisation in generic curved spacetimes 40

4.4 Cosmological particle creation 41

5 Observables quadratic in fields 46

5.1 Two-point functions 46

5.2 Adiabatic expansion of the Green’s function 47

5.3 Stress tensor renormalisation 50

II APPLICATIONS 53

6 The Unruh effect 54

6.1 Thermal states 54

6.2 Particle detectors 56

6.3 The Unruh effect 58

7 Hawking radiation 66

7.1 Quantum fields in a black hole background 66

7.2 The Hawking thermal state and friends 73

7.3 Black hole thermodynamics 75

7.4 The information paradox 77

8 Quantum fields in de Sitter space 80

8.1 de Sitter basics 80

8.2 Quantum fields in de Sitter space 82

8.3 De Sitter temperature 88

8.4 De Sitter entropy 91

9 Cosmological perturbations 94

9.1 Cosmological models 94

9.2 QFT on a cosmological background 97

9.3 Cosmological correlators 99

9.4 Fluctuating gravitons 100

iv
10 Quantum fields in AdS 102

10.1 Geometry of AdS spaces 102

10.2 Quantum fields in AdS 105

10.3 Towards a conformal theory on the boundary 108

10.4 The AdS/CFT correspondence 113

10.5 Hawking radiation in AdS 117

III APPENDICES 119

A Conventions 120

A.1 Signs, signatures and curvature 120

B Differential forms 123

B.1 Connections and curvature 125

C Hypersurfaces 127

D Variational calculus 129

E Ingredients from general relativity 132

E.1 Maximally symmetric spaces 132

E.2 Warped product manifolds and FLRW spaces 140

E.3 Black holes 142

F Hypergeometric functions 151

G A mini introduction to CFT 152

G.1 Conformal Transformations 152

G.2 Local Operators 153

G.3 Ward identities 154

G.4 State-Operator Map 156

G.5 Operator Product Expansion 157

G.6 Conformal Bootstrap 157

v
G.7 Embedding Space Formalism 158

G.8 Large N Factorization 160

Bibliography 162

vi
 Part I

FUNDAMENTAL ASPECTS

1
 Chapter 1

Why you should take this course

The incorporation of gravity into quantum physics remains one of the most important outstanding
problems in theoretical physics. With the ever growing experimental support for both Einstein’s
theory of general relativity, which relies on the differential geometry of Lorentzian manifolds, as well
as the standard model of elementary particle physics, based on quantum field theory, the issue of their
mutual compatibility becomes increasingly pressing. In particular in situations where both theories
must be simultaneously applied, such as in early universe cosmology or in the vicinity of a black hole.
Early attempts to incorporate gravity into quantum field theory by treating gravity as one of
the quantum fields encountered conceptual and practical difficulties due to the intrinsic non-
renormalisability.1 As a result, new approaches such as string theory (and more controversial
approaches such as loop quantum gravity or causal dynamical triangulation theory) arose. Despite
their mathematical elegance, explicit computations in these frameworks are often prohibitively hard
and a fully satisfactory understanding of quantum gravity remains elusive. Therefore, developing
tools to access this regime where gravity and quantum effects coexist is extremely valuable.
To illustrate in what sort of regime we are interested, consider a body of mass M and size L. The
"quantumness" of the interactions is governed by the dimensionless constant MħhLc , where c is the
h is Planck’s constant.2 Quantum effects are important whenever
speed of light and ħ

h
ħ
≳ 1, (1.1)
M Lc
G M
and are suppressed when MħhLc ≪ 1. On the other hand, the curve cN2 L , where GN is the effective
gravitational Newton’s constant, divides the regions in parameter space where the dynamical effects
of gravity become important or can be ignored. When we have

GN M
≳ 1, (1.2)
c2 L

the dynamical effects of gravity are important. In this situation we need the laws of general relativity
G M
to describe gravity. When cN2 L ≪ 1 gravitational effects are suppressed and we can use classical
(Newtonian) gravity to describe gravitational interactions. In Figure 1.1 we sketch the various domains
in this parameter space and indicate the validity of the relevant approximation in each region.
In this course we are probing the bottom right region of this diagram where both (1.1) and (1.2)
1
A somewhat oversimplified but instructive way to see this is through counting the dimension of the coupling constant.
Recall that in d +1 dimensions Newton’s gravitational constant has mass dimension [GN ] = 2−d. Hence, the Einstein-Hilbert
action has dimension d − 2 which suggests that gravity is non-renormalisable in d > 2. There are various subtleties that
need to be taken into account to make this naive argument rigorous but in the context of gravity it turns out to be correct.
2
h = c = GN = kB = 1.
In the remainder of this course we almost exclusively use natural units where ħ

2
 h
L
ħ
M Lc =1
GM
c2 L
=1
Classical gravity

QFT

General Relativity
(0,1)

h
ħ
M Lc =1
Quantum gravity
M
(1,0)

Figure 1.1: This diagram describes the range of validity of various approximations. Right of
the blue hyperbola quantum effects are suppressed and we can approximate our system using
classical physics, resp. Newtonian mechanics or general relativity depending on the position
with respect to the red line. On the other hand, on the left of the hyperbola quantum effects
are important and have to be included in an accurate description. In this case we either need
QFT or a theory of quantum gravity.

are satisfied. To properly describe this situation we need an honest theory of quantum gravity.
However, due to the relative weakness of the gravitational force it is often possible to neglect the
backreaction of the quantum fields on the space-time as a first approximation. Consequently the
problem then boils down to understanding quantum fields on generic Lorentzian space-time manifolds.
In other words, when the space-time curvature is very small on the Planck scale3 we can describe
gravity classically, through general relativity, and consider fluctuations in the matter fields in the
gravitational background described by a solution to the equations of motion of general relativity.
In this approximation we therefore do not capture the dynamical nature of space-time itself but
nonetheless this modest extension of quantum field theory is surprisingly rich in consequences. Among
other things, it gives rise to the process of particle creation in cosmological and black hole space-times.
Moreover, we can go beyond this first application and perturbatively4 include the effects of backreac-
tion of the quantum fields on the gravitational background. We still treat the background as classical
but now take the backreaction into account. To do so we have to solve the Einstein equation,

1
Rµν − Rgµν = 8πGN Tµν ψ
(1.3)
2

where the right hand side is given by the vacuum expectation value of the stress-tensor in the state
|ψ〉. This approach takes into account the one-loop gravitational interactions. In principle one should
be able to step-by-step add higher loop contributions but this procedure is not well understood at
3
See Appendix A for a definition of the relevant length scales.
G l p2
4
The perturbative parameter here is LN2 ≃ L2
≪ α where α parameterises the interaction strength of the matter, such as
1
e.g. the fine structure constant α ∼ 137 .

3
present and would take us beyond the scope of these lectures.
From another point of view, studying quantum field theory in curved space-times can help to clarify
various conceptual issues and teach us general lessons about the structure of quantum field theories.
Traditionally, quantum field theory in Minkowski space relies on concepts such as a unique vacuum,
particles, Fock spaces and S-matrices. In curved space-time many of these concepts break down and
need to be reconsidered. For example, we will learn that particles are an observer-dependent concept.
Rethinking these concepts leads us to a deeper understanding of quantum field theory and eventually
quantum gravity. Even more, in some cases computations can be more tractable in curved space then
in flat space. Many quantum field theories, such as Yang-Mills theory, suffer from various IR problems
on flat space, making semi-classical methods such as instantons useless. When working on a compact
space, such as the sphere, the ’radius’ of the space provides a natural IR regulator and one can use
semi-classical intuition once again. A modern application of this phenomenon is supersymmetric
localisation, which allows one to exactly compute various observables non-perturbatively on a variety
of manifolds.
For a physicist, this should be more than enough motivation to jump into the study of quantum field
theory in curved spaces. It is an active area of research with many questions left unexplored. This
course is but the starting point and aims to optimally equip you to attack such problems.
If you are more mathematically inclined, and do not care so much about the rich physical applications
of quantum field theory in curved space, do not leave just yet. On top of all these physical applications,
quantum field theory in curved spaces is mathematically very interesting in itself. Through its
study many new directions in mathematics have opened up such as Donaldson-Witten theory and
Chern-Simons theory, providing various new invariants characterizing manifolds. A hallmark of
mathematical physics is the connection between geometry and gauge theories. Donaldson-Witten
theory aims to characterize four-manifolds through geometric observables. This study was initiated
by Donaldson but a breakthrough came when Witten showed that such invariants can be computed
efficiently using (supersymmetric) gauge theories on said manifolds. Similarly, Chern-Simons theory
in three-dimensions can be used to compute a variety of knot invariants characterising knots in
three-dimensional manifolds. In this course we will not have time to dive into these topics but this
course, together with the course on supersymmetry and supergravity, provides you with all the tools
you need to start exploring the literature.

1.1 What is in this course?

Before we start let us take a moment to outline the scope of this course in some more detail. As
this course is about quantum fields, we start in the first chapter with a brief recapitulation of the
theory of quantum fields in flat space, with emphasis on those aspects which are important for our
exploration in curved space. The wave equations we encounter, such as the Klein-Gordon equation
or Dirac equation are linear differential equations which in flat space are straightforward to solve.
Upon placing quantum field theories in curved space, a variety of subtleties arise concerning the
well-definedness of said linear equations as well as issues regarding causality which now have to be
addressed properly.

4
For this reason we continue with a careful introduction of Lorentzian geometries and their local and
global causal structure in Chapter 3. Following this, we come back to the main goal of this course and
proceed in Chapter 4 with the general procedure of canonical quantisation of free fields in curved
space-time. This chapter will introduce all the important concepts such as the non-uniqueness of the
vacuum and the concept of particle creation in a gravitational background. Following this Chapter 5
introduces how to compute Green’s functions and other observables quadratic in fields, such as the
stress tensor expectation value.
Having introduced the general framework, the remainder of the course will focus on applying it to a
plethora of physically interesting backgrounds. The first example is the Unruh effect, discussed in
Chapter 6, which describes the physics of an accelerated observer in flat space. This example perfectly
sets us up for Chapter 7, where we discuss maybe the most well-known consequence of quantum
field theory in curved space-time, the Hawking radiation of black holes.
Having deepened our understanding of quantum mechanical black holes we move on to cosmological
consequences of quantum field theory in curved space. In the distant future, as well as during the
inflationary era, our universe can be described as de Sitter universe. In Chapter 8 we start with a
discussion of quantum fields in de Sitter space and explain why the second law of thermodynamics
forces the universe towards a cold empty space. In Chapter 9 we take a look at more generic
cosmological space-times and explain how to deal with cosmological perturbations.
Finally, the last topic of this course is quantum field theory in Anti-de Sitter space. Up to a sign,
Anti-de Sitter space is identical to de Sitter space, but this sign turns out to have deep implications.
In particular Anti-de Sitter space is the space in which holography is best understood. In Chapter 10
we discuss various subtleties in describing quantum fields in Anti-de Sitter space with an eye towards
defining the holographic principle, which teaches us that the theory in the bulk of Anti-de Sitter can
equivalently be described by a conformal theory on the conformal boundary!
In order not to obscure the message in the main text some of the background information as well
as some technical details are relegated to the appendices. In Appendix A we introduce the main
conventions used in these notes as well as a comparison to other conventions used in the literature.
In Appendix B we review the theory of differential forms on a generic manifold M. Appendix C
contains various useful ingredients from general relativity which will be used in this course. The
background on special functions and various useful identities for hypergeometric functions can be
found in F. Finally, the last two appendices ?? and G contain a short review on how to deal with
spinors in curved space as well as a lightning introduction to conformal field theory.

1.2 What is not?

Unfortunately, time is limited so we necessarily have to skip a variety of interesting topics. As such
we will not have time to discuss the rich subject of one-loop effective actions, which describe the
one-loop quantum corrections to the partition function or on-shell action of a variety of gravitational
backgrounds. Similarly, we are forced to skip the rich subject of supersymmetric theories on curved
backgrounds. In general putting a supersymmetric theory on a curved background breaks all the
supersymmetry. However, some supersymmetry can be preserved by placing the theory in backgrounds

5
which are solutions not to Einsteins equations of motion but those of a supergravity theory [FS11].
In particular, these two topics together provide one of the main testing grounds for the holographic
correspondence and provide a active research direction furthering our understanding of the holography.
Furthermore, the theory of supersymmetric theories in curved spaces gives rise to rich mathematics
such as the theory of Donaldson-Witten or Chern-Simons invariants. We will not be able to cover this
topic in this course but the interested reader can have a look at the lecture notes [Mar01, Moo12,
Lab00, Mar05, LM05].
Another glaring omission from this course is the theory of interacting quantum fields in curved space-
times. For weakly coupled theories one can use similar tools as in flat space, such as perturbation
theory as will be hinted at in various instances in the main text. However, the presence of curvature
and/or non-trivial topology add additional subtleties which need to be treated carefully. When the
theory is strongly coupled many familiar tools break down, and even in flat space the problem of
solving these theories is extremely hard and in general not well-understood. However, by putting
the theories in curved space-times we can make significant progress, especially for supersymmetric
theories. In particular, supersymmetric localisation provides us with a rare tool allowing us to compute
a set of protected supersymmetric observables exactly.
Finally, in these lecture notes we took the approach of canonical quantisation. There are various
other approaches such as path integral quantisation or algebraic quantum field theory. Path integral
quantisation yields a simple, manifestly covariant framework which generalises the stationary action
principle of classical mechanics by integrating over all classical paths, weighted by their action.
Algebraic quantum field theory on the other hand starts from the algebra of observables acting on
some Hilbert space and formalises the quantum field theory as a C ∗ -algebra with a set of axioms
known as the Haag-Kastler axioms. Each of these approaches can be generalised to curved space-times
and highlights different aspects. Each of them has their advantages and disadvantages such as the
manifest (or not so manifest) presence of unitarity, covariance and locality.

6
 Chapter 2

Quantum fields in flat space

Quantum field theory in curved space-time is a generalisation of quantum field theory in flat space.
It is not surprising that in many respects the behaviour of quantum fields in curved space-time can
be directly inferred from the flat space theory. Local entities, such as commutation relations or field
equations are determined by the principle of general covariance and the equivalence principle and will
therefore remain unchanged. On the other hand, there are various global entities which will behave
radically different in curved space. For example, in Minkowski space the vacuum is unambiguously
determined by Poincaré invariance. However, as we will see in Chapter 4, the concept of a vacuum
becomes ambiguous in curved space.
For this reason we will take a moment to review certain aspects of quantum field theory in flat space,
fix our conventions and notation and highlight certain aspects which carry over to curved space as
well as those which lose their meaning. To avoid having to deal carefully with gauge invariance etc.
we mostly focus on scalar fields as they suffice to illustrate the properties we are interested in. For
more details on spinors, vector or higher spin fields in Minkowski space we refer the reader to their
favourite textbook on quantum field theory, for example [Wei95, PS95, Sre07].

2.1 Canonical quantisation in Minkowski space

In this chapter we exclusively deal with (d + 1)-dimensional Minkowski space R1,d . The signature is
(1, d) and we use mostly minus conventions such that the metric on Minkowski space is given by

ηµν = diag (1, −1, · · · , −1) . (2.1)

We denote the coordinates collectively by x = (x 0 , x 1 , · · · , x d ) and often we use x 0 = t for the time
direction. Sometimes it is useful to separate the time direction and denote it by x 0 = t while we
collect the space coordinates in the vector x = (x 1 , · · · , x d ).
Consider a classical scalar field φ(x) in (d + 1)-dimensional Minkowski space R1,d , satisfying the
Klein-Gordon field equation,
□ + m2 φ = 0 ,


(2.2)

where □ = ηµν ∂µ ∂ν is the d’Alembertian, and m is the mass of the scalar field. This field equation
can be derived from the action,
Z

1 µν
η ∂µ φ∂ν φ − m2 φ 2 ,



S= L φ, ∂µ φ dd+1 x , with L φ, ∂µ φ = (2.3)
R1,d 2

7
by demanding that the variation δS with respect to φ vanishes. The conjugate momentum π(x) is
defined through the following definition,

δL (φ, ∂ φ)
π(x) = , (2.4)
δ (∂ t φ)

such that for the free scalar we find π(x) = ∂ t φ.

We proceed by defining the (classical) phase space Vφ as the vector space of fields satisfying the linear
field equations, i.e. the Klein-Gordon equation in the case of the free scalar. There are some caveats
regarding the asymptotic fall-off of these fields at infinity for the various formulae that we use to
make sense. Without going into too much detail we assume that our fields have been appropriately
restricted.1 For the scalar field we denote the phase space as

Vφ = φ ∈ C ∞ (R1,d ) (□ + m2 )φ = 0 = (φ, ∂ t φ) ∈ C ∞ (Rd ) × C ∞ (Rd ) ,

 
(2.5)

where in the second equality we identified the phase space with the initial data along a time slice
t = t 0 . A similar phase space can be defined for fermions and gauge fields but we leave the precise
definition as an exercise to the reader.2 The phase space comes with a natural symplectic structure,
or skew symmetric form,3
Z Z
Ω(φ1 , φ2 ) = φ1 ⋆ dφ2 − φ2 ⋆ dφ1 = (φ1 ∂ t φ2 − φ2 ∂ t φ1 ) dd x (2.6)
Rd Rd

More concretely, consider a basis {uk (x)} of the phase space, and denote their conjugate momenta by
{πk (x)}. In this basis the symplectic form takes the canonical form Ω = k duk (x) ∧ dπk and the
P

expression above can be obtained straightforwardly by expressing the fields φi in this basis. This
skew symmetric form is dual to the Poisson bracket on Vφ which can be expressed as,

φ(x), π(x ′ ) = δ(d) (x − x ′ ) , φ(x), φ(x ′ ) = π(x), π(x ′ ) = 0 .

  
(2.7)

Canonical quantisation then proceeds by promoting the fields and canonical momenta to operators
on an appropriate Hilbert space and imposing the canonical (equal time) commutation relations, 4

φ(t, x), φ(t, x′ ) = 0 ,

 

π(t, x), π(t, x′ ) = 0 ,

 
(2.8)
φ(t, x), π(t, x′ ) = i δ(x − x′ ) .
 

To develop the quantum theory we relate the phase space to a one-particle Hilbert space H and define
1
For the interested reader: The relevant space is often taken to be the Schwartz space, which consists of infinitely
differentiable functions that at infinity fall off faster than any reciprocal power of x. Crucially, this space has the property
that it allows for the Fourier transform to be applied.
2
Note that the Dirac equation is first order so that the initial value problem only needs the value of the field as boundary
condition. For vector fields one has to deal with gauge invariance and quotient out gauge equivalent field configurations.
3
For fermionic fields the phase space comes with a symmetric form.
4
Note that here and throughout the text we put ħ h = 1 explaining the absence of the characteristic factor of ħ h on the
right hand side of the equations in (2.8).

8
the Fock space to be
∞
M
F = C ⊕ H ⊕ (H ⊗ H) ⊕ · · · = ⊗n H . (2.9)
n=0

We can make this more precise using the Fourier transform, by decomposing complex fields into
plane waves. An appropriately normalised set of solutions to the wave equation (2.2) is given by the
following plane waves,
1
uk (t, x) = p ei(k·x−ωt) , (2.10)
(2π) 2ω
d
p
with ω = k2 + m2 . The plane waves uk are called positive frequency or positive energy solutions
with respect to t, while the the complex conjugate solutions, u∗k , are the negative frequency/energy
solutions. They are eigenfunctions of the time translation operators with eigenvalues ∓iω,

∂ t uk = −i ω uk , ∂ t u∗k = i ω u∗k . (2.11)

Such plane waves, together with their complex conjugates, form a complete set of solutions and
therefore any solution to the Klein-Gordon equation can be Fourier expanded as
Z
dd k ak uk (x) + ak† u∗k (x) .
 
φ(x) = (2.12)

Such that the phase space for a complex scalar decomposes as C ⊗ Vφ = Vφ+ ⊕ Vφ− and the choice of ω
above is precisely made so that the fields satisfy the on-shell condition, k2 = m2 .
The skew symmetric form on the phase space translates to a (positive definite) inner product on the
Hilbert space defined as,
Z
〈φ1 , φ2 〉 = i dd x (φ1 (x)∂ t φ2 (x) − ∂ t φ1 (x)φ2 (x)) . (2.13)

The main property of this inner product is that for two solutions to the Klein-Gordon equation the
product is conserved under time translation. The plane waves defined in (2.10) are orthonormal with
respect to this product,
〈uk , uk′ 〉 = δ(k − k′ ) , 〈uk , u∗k′ 〉 = 0 . (2.14)

Exercise 2.1. Prove that the plane waves uk together with their complex conjugates form an orthonormal
basis of L 2 (R1,d ), i.e. square integrable functions on R1,d .5

We can therefore express the Fourier coefficients as follows,

ak = 〈uk , φ〉 , ak† = 〈u∗k , φ〉 . (2.15)

From the equal time commutation relations (2.8), it follows that

ak , ak†′ = δ(k − k′ ) , [ak , ak′ ] = ak† , ak†′ = 0 ,

   
(2.16)
5
Note that strictly speaking plane waves do not belong to L 2 (R1,d ) but nonetheless they form a complete basis for it.
Similar comments apply to the dual basis given by the delta functions, δ(x − x ∗ ).

9
Exercise 2.2. Prove the commutation relations, (2.16) , for the creation and annihilation operators
starting from (2.8).

Remark. In the above, k is a continuous parameter. Sometimes it is useful to introduce periodic

boundary conditions on a torus T d+1 of volume L d+1 , a.k.a. putting the theory in a box. In particular,
this avoids all kinds of volume divergences. Doing so, the wave-vectors become quantised k = 2π L n
and all integrals in the above get replaced by infinite sums.
It is often easier to compute physical quantities using such a (IR) regulator and taking the limit
L → ∞ at the end of the computation to obtain the infinite volume result.

When the quantum theory enjoys some global symmetries, there will be associated conserved currents.
The theories we consider in this course are all covariant under space-time diffeomorphisms. In
particular, they are invariant under space and time translations. The associated conserved current is
given by the stress tensor,

δL
∂ µ Tµν , where Tµν = ∂ν φ − δµν L . (2.17)
δ∂ µ φ

Note that in order for this to serve as a source for the Einstein equations we need to improve it to
be symmetric under interchanging the indices. We will present a manifestly symmetric stress tensor
formulation of the stress tensor later when dealing with curved space-times. Similarly, if the theory
has additional global symmetries we can write the conserved current as

δL
∂ µ Jµ = 0 , where Jµ = δφ − Tµν δx ν , (2.18)
δ∂ µ φ

where δφ = φ ′ − φ denotes the change of the field under an infinitesimal symmetry transformation
and similarly δx ν = x ν′ − x ν . For an example, see the complex scalar field below.

2.2 Particle interpretation

We can interpret the operators ak and ak′ as annihilation and creation operators for an infinite amount
of harmonic oscillators labelled by their momentum. In the Heisenberg picture, the states span a
Hilbert space. A convenient basis for this Hilbert space is given by the Fock representation introduced
above.
The state space consists of the vacuum |0〉, which is annihilated by all the annihilation operators,

ak |0〉 = 0 . (2.19)

All the excited states in the Hilbert space can be constructed by acting on the vacuum with the creation
operators ak† ,
ak† |0〉 = |1k 〉 . (2.20)

10
Successively acting with the creation operators we can then construct the most general states,
p
Y 1 € † Š ni (k ) (k ) (k ) (k p )
E
p ak |0〉 = n1 1 n2 2 n3 3 · · · n p . (2.21)
i=1 ni ! i

These basis vectors are normalised such that

(k p ) (kq )
D E
(k ) (k ) (k ) (k ) (k ) (k )
n1 1 n2 2 n3 3 · · · n p m1 1 m2 2 m3 3 · · · mq =
X
δ pq δn1 mσ(1) δn2 mσ(2) · · · δnp mσ(p) , (2.22)
σ

where the sum runs over all permutations σ of the integers 1, ..., p.
In analogy with the harmonic oscillator we can then introduce the number operator,
X
N= Nk , Nk = ak† ak , (2.23)

whose expectation value in a generic state is given by

X
〈ψ| N |ψ〉 = ni , (2.24)
i

(k p )
E
(k ) (k ) (k )
where |ψ〉 = n1 1 n2 2 n3 3 · · · n p . Hence, we can interpret Nk and N as counting the number of
quanta with momentum k and the total number of quanta respectively.
Note that the vacuum, as defined in (2.19) is unique. Naively it may seem to depend on a choice
of inertial frame but an easy argument shows otherwise. Indeed, consider a second inertial frame
x̃ µ = Λµ ν x ν with Λ a Lorentz transformation. Analogous to the above we can define the positive
frequency functions
1
ũk̃ ( t̃, x̃) = p ei(k̃·x̃−ω̃ t̃) . (2.25)
(2π) 2ω̃
d

and expand the field φ as h i

φ(x̃) = d k̃ d
ãk̃ ũk̃ (x̃) + ㆠũ∗ . (2.26)
k̃ k̃(x)

The "new" vacuum is then defined by the condition that

ãk 0̃ = 0 . (2.27)

To show that this is nothing but the old vacuum consider the mode function and notice that
‹ 12 ‹ 12
1 ω̃ 1 ω̃
 
i(k·x̃−ω t̃) i(k̃·x−ω̃t)
ũk = p e = e = uk̃ . (2.28)
ω ω
p
(2π)d 2ω (2π)d 2ω̃

Since we restrict to the orthochronous subgroup of the Lorentz group, to preserve time orientation,
we have that ω̃ > 0 implies ω > 0 and thus we find that

ak̃ 0̃ = 0 ∀k̃ ⇒ ak |0〉 = 0 ∀k , (2.29)

11
and the converse follows by symmetry. Hence the vacuum is indeed unique and independent of the
choice of frame.

2.3 Vacuum energy

To further explore the meaning of the Fock states we can compute their energy and momentum,
which can be obtained from the expectation value of the energy-momentum tensor which for a scalar
field is given by
1
Tµν = ∂µ φ∂ν φ − ηµν ∂ ρ φ∂ρ φ − m2 φ 2 .


(2.30)
2
We can define the conserved momentum operator as
Z
Pµ = Tµ0 dd x , (2.31)
t=t 0

with the time-like component P0 = H, the Hamiltonian. The expection value in a state |ψ〉 can then
be computed as 〈ψ| H |ψ〉.

Exercise 2.3. Give an expression for the Hamiltonian and conserved momentum in terms of the creation
and annihilation operators and show that they commute with the number operator N .
Naively compute the expectation value of the Hamiltonian density H and momentum density Pµ and
show that they are divergent.
Similarly, consider the expectation value of the energy-momentum density Tµν in the vacuum as well as
in a generic state and find an expression in terms of the mode functions.

Computing the energy we encounter our first infinite result. Such troubling results are well-known to
plague the subject of quantum field theory but can be cured through renormalisation. In this setup it
will suffice to define the normal ordering operation, : • :, which is understood to act on products of
annihilation and creation operators such that it puts all the annihilation operators on the right of the
creation operators.

Exercise 2.4 (Vacuum energy divergence). Compute the momentum and energy of the vacuum and
show that it naively diverges. Use the normal ordering prescription to regularise this result and find the
resulting vacuum energy.

In the above we regularised the vacuum energy by passing through a normal ordered prescription
which simply throws away the infinite contribution. However, as already mentioned before, in curved
space, especially when gravity is included, the energy of the vacuum is physical since it gravitates.
For this reason it is instructive to take a closer look at the vacuum expectation value of the energy in
a situation where it becomes important. Consider a massless neutral scalar in (3 + 1)-dimensional
Minkowski space in the presence of two parallel plates at x 3 = 0 and x 3 = a with a ≪ 1 and
impose boundary conditions φ(0) = φ(a) on the plates. In addition we impose periodic boundary
conditions in the two other spatial directions with x 1,2 ∼ x 1,2 + L, with L ≫ a. This setup represents
a modified version of the usual Casimir effect for two neutral conducting plates in a vacuum electric
field vanishing on the plates.

12
Exercise 2.5. Quantise the scalar field in the presence of the two plates and show that the average energy
density is given by
1 X
ρ(a) = a−1 L −2 〈0| T00 |0〉a = ωk , (2.32)
2a L 2 k

and give an expression for ωk .

We can regularise this sum by writing

1 d X −εωk
ρ(a) = − lim e (2.33)
2a L 2 ε→0 dε k

Remark. The parameter ε here is dimensionful so it is good practice to introduce an explicit length
scale ε → ε/Λ so that ε becomes dimensionless. In the final result all the Λ dependence should drop
out as can be easily verified.

Exercise 2.6. Compute the regularised vacuum energy and show that the sum S(ε, a) = L −2 k exp(−εωk ),
P

takes the form

π3
πS(ε, a) = aG(α) + ε + O(ε2 ) . (2.34)
45a3
Before taking the limit in (2.33), subtract the infinite part to arrive at the renormalised vacuum energy.
Doing so compute ρ(a).
What would have changed if we instead imposed vanishing boundary conditions at the plates?

Remark. In the last exercise we proceeded in a rather cavalier way. In order to obtain the same result
in a mathematically more satisfying manner one can employ ζ-function regularisation. We invite the
interested reader to explore this method and repeat this exercise in a more rigorous manner.

2.4 Two-point functions

On top of the vacuum energy and Hilbert space, much information about a quantum field theory is
encoded in its n-point functions. Since we are so far only working with free scalars, all the non-trivial
information is encoded in the two-point functions, i.e. the Green functions of the wave equations.
There are various types of Green functions depending on the choice of integration contour in the
complex plane.
A useful set of Green functions are given by the following expectation values,

iG(x 1 , x 2 ) = 〈0| [φ(x 1 ), φ(x 2 )] |0〉 ,

G (1) (x 1 , x 2 ) = 〈0| {φ(x 1 ), φ(x 2 )} |0〉 ,
(2.35)
G + (x 1 , x 2 ) = 〈0| φ(x 1 )φ(x 2 ) |0〉 ,
G − (x 1 , x 2 ) = 〈0| φ(x 2 )φ(x 1 ) |0〉 ,

13
G is known as the Pauli-Jordan, or Schwinger function, while G (1) is called the Hadamard elementary
function. These Green functions can be split into their positive and negative part,

iG = G + − G − , and G (1) = G + + G − , (2.36)

where the positive and negative part G ± are known as the Wightman functions. Finally, the Feynman
propagator G F and retarded/advanced Green functions GR/A are defined as

G F (x 1 , x 2 ) = 〈0| T [φ(x 2 )φ(x 1 )] |0〉 = θ (t 1 − t 2 )G + (x 1 , x 2 ) + θ (t 2 − t 1 )G − (x 1 , x 2 ) , (2.37)

where T denotes time ordering, and

GR (x 1 , x 2 ) = θ (t 1 − t 2 )G(x 1 , x 2 ) , GA(x 1 , x 2 ) = −θ (t 2 − t 1 )G(x 1 , x 2 ) . (2.38)

All these two-point functions have the same form in momentum space

−i
G(k)
e = , (2.39)
k2 + m2

but they have different iε prescriptions and are used in different contexts. Time ordered products,
such as the Feynman or retarded/advanced Green’s functions are relevant for S-matrix calculations.
The Green’s functions involving (anti-)commutators on the other hand are useful to describe how the
field responds to a source. Finally, the Wightman functions are useful since they describe the effect of
the field on a moving detector, as will be described in Chapter 6.
Exercise 2.7. Show that the average of the retarded and advanced Green functions G = 12 (GR + GA) is
given by
1
G F (x 1 , x 2 ) = iG(x 1 , x 2 ) + G (1) (x 1 , x 2 ) . (2.40)
2
Using the field equations (2.2) one can show that the Green functions G ∈ {G , G (1) , G ± } all satisfy
the homogeneous equation
(□ x 1 + m2 )G (x 1 , x 2 ) = 0 , (2.41)

while the Feynman and retarded/advanced Green functions satisfy

(□ x 1 + m2 )G F (x 1 , x 2 ) = − δ(d+1) (x 1 − x 2 ) ,
(2.42)
(□ x 1 + m2 )GA/R (x 1 , x 2 ) = δ(d+1) (x 1 − x 2 ) .

We can use translation invariance to translate one of the points, say x 2 to the origin in which case we
can denote the Green functions by G (x) = G (x, 0). Having done so, an integral representations for
the Green functions can be obtained by substituting the mode decomposition (2.12) in the definitions
above. All the Green functions can then be represented as

e−ik·x
Z
1
G (x) = dd+1 k . (2.43)
(2π)d+1 k2 − m2
p
This integral has poles at ω = ± |k2 + m2 | and the various Green functions correspond to various
contour prescriptions for the integration. In Figure 2.1, the contours are shown for the various Green

14
functions.

Figure 2.1: THe various Green functions are associated with the above contours for the integral
(2.43). The open contours should be interpreted as closed by an infinitely large semicirle in
the upper/lower plane.

From (2.36)–(2.38) we can see that we can obtain all the Green functions from the Wightman
functions G ± so we mostly focus on those.

Exercise 2.8. Use the mode expansion of the scalar field to show that the Green functions take the
manifestly Lorentz invariant form (2.43).

Example 2.1 (Massless scalar). To illustrate the formalism let us consider a massless scalar in four-
dimensional Minkowski space. To compute the Wightman functions perform the relevant contour integra-
tion as indicated in Figure 2.1. We focus on G + but the computation for G − is entirely analogous. Near
the pole the integrand reduces to
1 −ik·x
I+ = e , (2.44)
2ωk
where ωk = |k|. Using Cauchy’s residue theorem we obtain

d3 k
Z
+
G (x) = e−i(|k|t−k·x) . (2.45)
(2π) 2|k|
3

We can use spherical coordinates to rewrite this as

∞
|k|d|k| sin θ dθ dϕ i|k|(r cos θ −t)
Z Z
+ i
d|k| e−i|k|(t−r) − e−i|k|(t+r) .


G (x) = 3
e =− (2.46)
2(2π) 2(2π)2 r 0

This integral is divergent and needs to be regularised. We do so by shifting t → t − iε, with ε > 0 so that
Z ∞
i
d|k|e−i|k|(t−iε±r) = − . (2.47)
0 t ± r − iε

15
This has to be understood as a distribution, since we have (Sokhotski-Plemelj theorem)

1 1
lim = P ∓ iπδ(x) , (2.48)
ε→0 x ± iε x

where P denotes the Cauchy principal value. Using this we find the Wightman function

1 1 i
G + (x) = − 2
P 2 2
+ (δ(t + r) − δ(t − r)) . (2.49)
4π t −r 8πr

Exercise 2.9. Use this result to compute the Feynman propagator and Hadamard’s elementary function,

i 1 1 1 1
G F (x) = 2
P 2− δ(x 2 ) , G (1) = − 2
P 2. (2.50)
4π x 4π 2π x

Exercise 2.10 (Massive scalar). Use spherical coordinates in the spatial directions to simplify the integral
and find an explicit expression for the Wightman functions for a massive scalar.
Show that for space-like separated points, the Wightman function takes the form
m
G + (x 1 , x 2 ) =
Æ
K1 (ms) , s= −(x 1 − x 2 )2 , (2.51)
4π2 s

where s is the proper distance and K1 is a Bessel function of the second kind. Show that for timelike
separated points it becomes

im (2)
G + (x 1 , x 2 ) =
Æ
H (mτ) , τ= (x − y)2 , (2.52)
8πτ 1
(2)
where τ is the proper time and H1 is a Hankel function of the second kind.
What is the behaviour for large space-like separation r ≫ 1? Give a physical explanation of this behaviour.
What happens for light-like separated points? You can study this by taking the limit s → 0 in (2.51)
or τ → 0 in (2.52). Show that there is a branch point singularity in the Wightman distribution. This
essential singularity is known as the lightcone singularity of Wightman functions at null separation.

In order to perform the calculations of the (Feynman) Green’s function it is often useful to rotate the
contour by 90 degrees to obtain the Euclidean Green’s function G E . The integration variables are
changed to κ = −iω and similarly we replace τ = −it, such that we have the relation

G F (t, x) = −iG E (iτ, x) . (2.53)

where

ei(k·x+ωτ)
Z
1
G E (τ, x) = dκdd k , (□ − m2 )G E = −δ(d+1) (x) , (2.54)
(2π)d+1 ω2 + |k|2 + m2

where □ now denotes the d’Alembertian on (d + 1)-dimensional Euclidean space. The advantage of
the Euclidean theory is that the operator □ − m2 has a unique well-defined inverse because the poles
now lie on the imaginary rather than the real axis. Hence it is often much easier to work in Euclidean
space and Wick rotate the result to obtain the Lorentzian Feynman Green’s function. Note that this

16
only works for the Feynman propagator, as all of the other contours in Figure 2.1 can not be rotated
without crossing poles. A useful way to compute Euclidean Green’s functions is using heat kernel
regularisation, as demonstrated in the following exercise.

Exercise 2.11. An alternative and often useful way to compute Green’s functions is by analytically
continuing the expressions to Euclidean signature and using heat kernel regularisation.
Let us consider the Laplacian operator A = −□ in Euclidean space Rd . The heat kernel of an operator A
is defined as
K(x, x ′ ; τ) = 〈x| e−τA x ′ . (2.55)

Show that, in the case of the Laplacian, the heat kernel solves the heat equation

∂ K(x, x ′ ; τ)
□ x K(x, x ′ ; τ) = , (2.56)
∂τ
with boundary condition
K(x, x ′ ; 0) = δ(x − x ′ ) . (2.57)

Show that the Euclidean Green’s function of A is then given by

Z ∞
′
G(x, x ) = K(x, x ′ ; τ)dτ , (2.58)
0

and use the explicit expression for the heat kernel to derive the Euclidean Green’s function for a massless
scalar in Rd . Compare the result with a direct calculation of G + (x, x ′ ) in four-dimensional Minkowski
space-time.

The last Green’s function we want to introduce is the thermal Green’s function about which we will
have much more to say in Chapter 6. Instead of looking at the Green’s functions in the vacuum state,
the thermal Green’s functions Gβ (x) are obtained by considering a thermal state at temperature
T = β1 . These Green’s functions have the important property that they are periodic in imaginary time,

Gβ (t, x) = Gβ (t + iβ, x) . (2.59)

2.5 Charged scalars, gauge fields and spinors

To finish this chapter, we briefly consider charged scalars, gauge fields and Dirac spinors. Many details
are omitted as this section mainly serves to set our notation and conventions.
So far we considered real, neutral scalar fields. A charged scalar on the other hand can be described
by a pair of Hermitian scalar fields φ1 and φ2 which we can collect in a single complex field

φ(x) = φ1 (x) + iφ2 (x) . (2.60)

The Lagrangian density is given by the slightest generalisation of (2.3) as

L = ηµν ∂µ φ † ∂ν φ − m2 φ † φ , (2.61)

17
which is invariant under the global symmetry transformation

φ(x) → φ ′ (x) = eiα φ(x) , α ∈ R. (2.62)

Exercise 2.12. What is the conserved current for this symmetry Jµ ? And what is the symmetry generator
R
G(•) = dd J0 •?
Show that the generator can be written as
X
N− − N+ ,


G=α (2.63)
k

where N ± are respectively the number operators for positively and negatively charged particles.

We can couple the charged scalar field to an external electromagnetic field with field strength
F = dA.6 The minimal coupling prescription is defined by replacing the derivatives ∂ in L with
covariant derivatives D = ∂ + i gA, where g is the coupling constant. The Lagrangian is then invariant
under the local symmetry transformation

1
Aµ (x) →A′µ (x) = Aµ (x) − ∂µ θ (x) ,
g (2.64)
′ iθ (x)
φ(x) →φ (x) = e φ(x) .

If the gauge field Aµ is dynamic we must add to the Lagrangian a Maxwell term, gauging the original
global symmetry,
1
L = − Fµν F µν + Dµ φ † Dµ φ − m2 φ † φ . (2.65)
4

Remark. In the above we considered a U(1) gauge field. If φ is instead charged under a G symmetry
we can proceed analogously. The Lagrangian remains unchanged but the gauge field A now transforms
in the adjoint representation of the gauge group G,

Aµ (x) = Aaµ (x)T a , (2.66)

where T a are the generators of the gauge group which satisfy

 
T a , T b = i f ab c T c , (2.67)

with the structure constants f a b c . The (covariant) field strength in turn is given by

F = −ig −1 [D, D] = dA + ig [A, A] . (2.68)

In this case the kinetic term for the gauge fields, also known as the Yang-Mills action contains a
quartic interaction term, making this case significantly harder to analyse.

6
If this notation is unfamiliar, see Appendix B for a review on differential forms.

18
Finally, the Dirac spinor ψ is governed by the action

S = ψ (i∂/ − m) ψ , ∂/ = γµ ∂µ . (2.69)

where the gamma matrices γµ satisfy the following commutation relations

{γµ , γν } = 2ηµν . (2.70)

The Dirac adjoint is defined as ψ = ψ† γ0 and transforms such that the Lagrangian density is a scalar.
For more information on the quantisation of Maxwell theory, Yang-Mills theory or the (charged)
Dirac spinor we refer the reader to any textbook on quantum field theory cited above. For additional
information on representations of gamma matrices and minimal spinors in various dimensions
see [VP99, FVP12].

19
 Chapter 3

Lorentzian geometry

In the previous chapter, we reviewed some salient features of quantum field theory in flat space. In
particular we focused on the canonical quantisation of free theories and the concepts of particles
and vacuum energy. In a classical field theory, one obtains the physical field configurations through
variational principles, i.e. Euler-Lagrange equations etc. Once a solution has been found its stability
can be studied through a local analysis in field space. Quantisation on the other hand is a global
procedure where we need to take into account the full phase space. Indeed, already in quantum
mechanics there is the possibility for a particle to jump over any potential barrier allowing it to probe
the full phase space.
As we saw in the previous chapter, in order to quantise the theory (using canonical quantisation) we
need a complete set of solutions to certain (linear) wave equations. Before moving on to quantum
fields in curved space, we will therefore need some basic global notions from Lorentzian geometry.
In particular, we focus on carefully defining causality and various related concepts, such as Cauchy
hypersurfaces and global hyperbolic manifolds. On such manifolds, the above mentioned wave
functions behave particularly nicely. More details can be found in for example [Wal84, O’N83].

3.1 Lorentzian manifolds

Before stating the definition of a Lorentzian manifold let us start to define what we mean by a
Lorentzian scalar product on a vector space V .

Definition 3.1. Let V be a (d + 1)-dimensional real vector space. A Lorentzian scalar product on V is
a non-degenerate symmetric bilinear form, 〈·, ·〉, of signature (1, d).

This means that we can find a basis eµ , µ = 1, . . . , d + 1, of V such that

eµ , eν = ηµν . (3.1)

where ηµν is the usual Minkowski metric, ηµν = diag(1, −1, . . . , −1). With this definition at hand we
can now give a precise definition of a Lorentzian manifold.

Definition 3.2. A Lorentzian manifold is a pair (M, g), where M is a smooth (d + 1)-dimensional
manifold, and g is a Lorentzian metric, i.e. g associates with each point p ∈ M a Lorentzian scalar
product g p on the tangent space Tp M.

As usual in differential geometry we require that g p depends smoothly on p. For a choice of local
coordinates, (x 1 , . . . , x d+1 ) : U → V , where U ⊂ M and V ⊂ R1,d are open subsets, and for any

20
µ, ν = 1, . . . , d + 1, the functions gµν : V → R, defined by g(∂µ , ∂ν ), are smooth. Here ∂µ = ∂ ∂x
µ
denote the usual coordinate vector fields. With respect to these coordinates we write the line element
ds2 = gµν dx µ ⊗ dx ν .
Before we proceed let us give a few examples.

Example 3.1 (Minkowski space). Minkowski space (R1,d , η) is obviously a Lorentzian manifold.

Example 3.2 (Warped product spaces). Let (N , h) be a connected Riemannian manifold, and I ⊂ R
an open interval. For any smooth positive function f : I → (0, ∞), we can define a metric gµν with
line element ds2 = dt 2 − f (t)2 h on M = I × N . For any two vectors X i = (ai ∂ t ⊕ Yi ) ∈ T(t,p) (M), with
Yi ∈ Tp N we have g(X 1 , X 2 ) = a1 a2 − f (t)2 h(Y1 , Y2 ).1 This type of Lorentzian metrics is called a warped
product metric.

Many familiar Lorentzian manifolds are of the form of this second example. Friedman-Lemaître-
Robertson-Walker space-times [Fri22, Fri24, Lem31, Lem33, Rob35, Rob36a, Rob36b, Wal37] are
obtained by requiring (N , h) to be a maximally symmetric Riemannian manifold with a constant
curvature metric. This type of metric is of particular relevance when studying cosmological models
describing the big bang or the expansion of the universe. A special case is the de Sitter (dS) space-time
in global coordinates, where I = R, N = S n−1 , with h the canonical metric on the (n − 1)-sphere with
unit radius, and f (t) = cosh(t).
As a final example, consider the Schwarzschild black hole.

Example 3.3 (Schwarzschild black hole). For a fixed mass M > 0, consider the function

2M
h : R+ → R : r 7→ 1 − . (3.2)
r

This function has a pole at r = 0 and a root at r = 2M . On both patches PI = (r, t) ∈ R2 |r > 2M

and PI I = (r, t) ∈ R2 |0 < r < 2M we define the Lorentzian metric as

1
g = h(r)dt 2 − dr 2 . (3.3)
h(r)

The singularity of the metric g at r = 2M might seem problematic, but one can easily show (by going
to Kruskal coordinates for example) that this is simply a coordinate singularity. For more details on
the Schwarzschild black hole and its rotating and electromagnetically charged cousins we refer the
reader to the course GR II.

3.2 Causal structure

Given a Lorentzian manifold with associated metric g, we can associate to each point p ∈ M the
quadratic form
γ p : Tp M → R : γ p (X ) = g p (X , X ) . (3.4)
1
For any t ∈ I and p ∈ N we identify the tangent space at (t, p) as follows, T(t,p) M = Tt I ⊕ Tp N .

21
 Figure 3.1: The lightcone associated to the point p.

A vector X ∈ Tp M is called time-like, light-like or space-like respectively if

 γ p (X ) > 0 ,

time-like ,
γ p (X ) = 0 , light-like , (3.5)
γ p (X ) < 0 ,

space-like .

A vector is called causal if it is time-like or light-like. For d ≥ 1, the set of time-like vectors consists of
two connected components, assigning a time-orientation consists of choosing one of these components
which we denote by I+ (p) and call future-directed. The closure J+ (p) = I+ (p) consists of the set of
future-directed causal vectors. Analogously we call I− (p) resp. J− (p) the past directed light-like and
causal vectors. The double cone formed as such is called the lightcone at the point p, see Figure 3.1.
Similarly, we call a (piecewise) differentiable curve s ∈ C 1 (M) in M time-like, light-like or space-like,
if all of its tangent vectors are respectively time-like, light-like or space-like.
Having defined these concepts locally at each point p we now want to extend them to global properties
of the manifold. A first step in this direction is the following definition.

Definition 3.3 (time-oriented manifold). A Lorentzian manifold (M, g) is time-orientable if there

exists a continuous time-like vector field τ on M. A Lorentzian manifold together with such a vector
is called time-oriented.

We will call time-oriented Lorentzian manifolds space-times. It should be noted that the concept
of orientability depends only on the topology of M, while the notion of time-orientability depends
on the choice of Lorentzian metric. The question of whether a manifold can be equipped with
some time-orientable metric is ’topological’ indeed, we have the following equivalent statements
(see [O’N83]):

• M admits a smooth non-vanishing vector field.

22
 • M can be equipped with a smooth Lorentzian metric.

• M can be equipped with a time-orientable Lorentzian metric.

Example 3.4. Some (non-)time-orientable manifolds:

• The standard Minkowski space is orientable and time-orientable.

• The two-sphere is orientable but cannot be equipped with a time-orientable metric. Indeed, famously
S 2 does not admit a smooth non-vanishing vector field.

• The Mobius strip is not orientable but it can be equipped with Lorentzian metrics that are either
time-orientable or not.

• The cylinder R × S 1 is orientable and can be equipped with a metric that is time orientable,

ds2 = dt 2 − dx 2 . (3.6)

It can also be equipped with a metric that is not time orientable,

ds2 = − cos θ dθ 2 + 2 sin θ dxdθ + cos θ dx 2 . (3.7)

For future reference, we define the following causality relations. Let p, q ∈ M, we have

 p ≪ q ↔ ∃ a future directed timelike curve in M connecting p and q ,
p < q ↔ ∃ a future directed causal curve in M connecting p and q , (3.8)
p≤q ↔ p < q or p = q .


Unless otherwise mentioned, we always consider both space and time orientable manifolds. However,
time-orientability is not quite strong enough to rule out all problematic cases. For linear operators to
have well-defined unique, causal solutions we need something more.
In general relativity, worldlines of particles are modelled by causal curves. If the space-time is compact,
something strange happens. Namely, in every compact space-time M there exists a closed time-like
curve (CTC). When such curves exist, one easily runs into paradoxes as travel into the past is now
a clear possibility. Therefore, when dealing with Lorentzian space-times we want to exclude such
examples.

Definition 3.4 (Causal manifold). A space-time (M, g) is causal if it does not contain any closed
causal curve.

Definition 3.5 (Strongly causal manifold). A space-time (M, g) is strongly causal if for any point
p ∈ M and any neighbourhood U of p, there exists a causally convex neighbourhood V of p, contained
in U. A neighbourhood is said to be causally convex if any causal curve with endpoints in V is entirely
contained in V .

This property implies that there cannot be time-like curves that pass through V more then once. In
other words, it is not possible to return to the same point in space-time by following a time-like

23
curve, i.e. particles travelling slower then light cannot return to the same point in space-time. Strong
causality obviously implies causality. For technical reasons we will always assume strong causality.
The wave equations that we consider, such as the wave equation for a scalar field (□ + m2 )φ = 0,
are all hyperbolic (linear) partial differential equations. This means that an equation of order n
has a well-posed initial value problem for the first n − 1 derivatives.2 More precisely, the Cauchy
problem can be locally solved for arbitrary initial data along a non-characteristic hypersurface Σ. In
order to have well-defined global solutions on Lorentzian space-times we need some further technical
definitions.

Definition 3.6 (Achronal hypersurface). A hypersurface Σ is achronal if no pair of points p, q ∈ Σ

can be connected by a time-like curve.

Since solutions propagate along causal curves the data on the hypersurface Σ can only influence a
restricted region. We can define the regions

J+ (Σ) = p ∈ M ∃ a future directed causal curve starting on Σ to p ,
 (3.9)
J− (Σ) = p ∈ M ∃ a future directed causal curve from p ending on Σ .

These sets are sometimes denoted as the future/past domain of influence. Analogously we can define
I± (Σ) by restricting to the interior of J± (Σ).

Definition 3.7 (Domain of dependence). The domain of dependence of a subset Σ is defined as the
set of points, D(Σ), in M through which every inextendable causal curve in M meets Σ, i.e.

D(Σ) = {p ∈ M| every inextendable causal curve passing through p intersects Σ} . (3.10)

Analogously we define the future/past domain of dependence D± (Σ) as the intersection D(Σ) ∩ J± (Σ).

The domain of dependence is the region on which the initial value problem for wave equations can
be proved to be well-posed by various PDE techniques. If p is a point lying on a causal curve that
cannot be extended through Σ, then one can imagine waves coming in along that curve that are
not determined by the data on Σ and therefore they would violate the uniqueness assumption. The
interior of the domain of dependence is sometimes also called the Cauchy development of a set.

Definition 3.8 (Cauchy hypersurface). A subset Σ of a connected space-time M is a Cauchy hyper-

surface if each inextendable causal curve in M meets Σ at exactly one point.

In other words a Cauchy hypersurface is a hypersurface for which D(Σ) = M. Any two Cauchy
hypersurfaces in M are homeomorphic. They are topological.
In analogy with the nomenclature for PDEs, we call a space-time globally hyperbolic when the future
state of the system is entirely specified by initial conditions. There are several equivalent definitions
of global hyperbolicity of which we note two.

Definition 3.9 (globally hyperbolic manifold I). A space-time M is a globally hyperbolic manifold if
it is strongly causal and if for all p, q ∈ M the intersection J+ (p) ∩ J− (q) is compact.
2
We say that the initial value problem on some region U with given data on Σ is well-posed if there exists a unique
solution on U with given initial data on Σ.

24
 D(Σ)

J+ (Σ)
Σ J− (Σ)

Figure 3.2: The domains of dependence and influence of a set Σ. The green area is the future
domain of influence, the blue area the past domain of influence and the red area denotes the
full future + past domain of dependence.

Definition 3.10 (globally hyperbolic manifold II). A space-time M is a globally hyperbolic manifold
if it admits a Cauchy surface.

A very useful theorem about globally hyperbolic manifolds goes as follows.

Theorem 3.1. Let M be a connected time-oriented Lorentzian manifold. Then the following are
equivalent:

1. M is globally hyperbolic.

2. M has a Cauchy hypersurface.

3. M is isometric to R × Σ with metric ds2 = −βdt 2 + dsΣ

2
(t), where dsΣ
2
(t) depends smoothly on t
and each {t} × Σ is a Cauchy hypersurface in M.

The proof is rather technical so we do not state it here but the crucial step is proving that 1. follows
from 3.. The proof can be found in [BS05] using a theorem by Geroch [Ger70]. Once this step has
been proven the other implications follow straightforwardly.

Exercise 3.1. Show that on a globally hyperbolic manifold M there always exists a smooth function
h : M → R whose gradient is past-directed time-like at every point and all of whose level sets are space-like
Cauchy hypersurfaces. Such a function is called a Cauchy time function.

Globally hyperbolic manifolds are very useful since this property ensures that for a large class
of operators the Cauchy problems are well-posed, with initial data on a Cauchy hypersurface in
appropriate function spaces. In particular, most of the operators we will encounter are generalized
d’Alembertians P, of the form

d
X d
X
P= ij
g (x)∂ x i ∂ x j + a i (x)∂ x i + b(x) , (3.11)
i, j=0 i=0

where the inverse metric g i j , and the functions a i and b are smooth functions of x. On globally
hyperbolic manifolds, one can prove a whole range of global existence and uniqueness theorems
for the solutions of the homogeneous equation Pφ = 0 as well as for the Green’s functions. For a

25
more detailed treatment of linear wave equations on Lorentzian manifolds we refer the reader to the
lecture notes [BBB+ 09].
For these reasons, we will mostly consider globally hyperbolic manifolds in the subsequent chapters.
Minkowski space, de Sitter space, the Schwarzschild black hole as well as the FLRW solutions are
all globally hyperbolic space-times. However, there are various notable exceptions which will be
very interesting to study such as for example Anti-de Sitter (AdS) space. We can overcome the issues
accompanying the absence of global hyperbolicity by providing appropriate boundary conditions at
conformal infinity. In Chapter 10 we will come back to this example in more detail.

3.3 Conformal infinity and Penrose diagrams

To get a good grip on the global structure, a very useful tool is to study the asymptotics. A neat way
to do so is via conformal compactifications. This consist of introducing a conformal boundary which
corresponds to infinity in the physical space-time. Various aspects of this procedure were discussed in
GRII and we refer the reader to that course for more details and references. Here we restrict ourselves
to those aspects relevant in the remainder of this course. In particular we will not go into any detail
on the conformal compactifications of black hole space-times.

Definition 3.11 (Conformal transformation). A conformal transformation is a map from a space-time

(M, g) to another space-time (M
f, g̃), such that

g̃µν (x) = Ω(x)2 gµν (x) , (3.12)

where Ω is a non-vanishing smooth function of the coordinates, Ω(x) ̸= 0 for all x ∈ M.

One reason why conformal transformations are useful is because they preserve the causal structure of
space-time. Indeed, two space-times whose metrics are related by a conformal transformation have the
same null geodesics.3 We can use this fact to our advantage to study the causal structure of space-time
by using suitably chosen conformal transformations to bring infinity to a finite coordinate distance
allowing us to represent the causal structure on a finite sized diagram called a Penrose diagram.4
The process of mapping space-time to a compact domain is called conformal compactification.

Definition 3.12. A conformal compactification of a space-time (M, g) is a manifold M

f with boundary
I =M f and metric g̃ such that

1. g̃ is smooth on M
f.

2. M is diffeomorphic to the interior of M

f.

3. On M we have that g̃ = Ω2 g with Ω a smooth function on M

f.

f ≃ M we have Ω ̸= 0.
f\∂ M
4. In the interior, M

5. On the boundary ∂ M
f = I , we have Ω = 0, and dΩ ̸= 0.
3
Space-like and time-like geodesics on the other hand are not necessarily preserved under conformal transformations.
4
Or Carter-Penrose diagram if you ask someone from Cambridge.

26
The boundary I 5 is called conformal infinity. In addition to the hypersurface I the conformally
extended space-time might contain loci of higher co-dimension, these have to be considered separately
and are denoted by i0 or i± depending on their (time-like/space-like/null) causal properties.
To illustrate these concepts, let us consider in detail the example of Minkowski space.
Example 3.5 (conformal compactification of Minkowski space-time). Consider Minkowski space-time
in d + 1 dimensions with line element
d
X
ds2 = dt 2 − dx i2 , t, x i ∈ (−∞, ∞) . (3.13)
i=1

In spherical coordinates the metric becomes

ds2 = dt 2 − dr 2 − r 2 dsS2d−1 . (3.14)

Defining the light-cone coordinates u = t − r and v = t + r and performing a further coordinate

transformation6  π π
u = tan ũ , v = tan ṽ , ũ, ṽ ∈ − , , (3.15)
2 2
with ũ ≤ ṽ, the metric becomes

1
ds2 = − 4dũdṽ − sin2 (ṽ − ũ)dsS d−1 .
 
(3.16)
4 cos2 ũ cos2 ṽ

We can now use a conformal transformation to remove the prefactor. Since the metric is now regular at
the points at infinity we can now compactify the space by including the points ũ, ṽ = ± π2 . The Penrose
diagram for Minkowski space is illustrated in Figure 3.3.

Alternatively, changing coordinates ũ = 12 (T − ρ) and ṽ = 12 (T + ρ) we obtain the following metric

for the conformal compactification
e 2 = dT 2 − ds23 ,
ds (3.17)
S

known as the Einstein cylinder metric. Consequently, the rescaling procedure described above maps
Minkowski space into a (compact) region of the Einstein cylinder, see Figure 3.3.
Having constructed the conformal compactification of Minkowski space let us discuss the structure of
conformal infinity.

• Future and past null infinity are defined as the hypersurfaces,

I± = p ∈ M

f 0 < ρ(p) < π, T (p) = ±(π − ρ(p)) . (3.18)

By definition, on this hypersurface we have Ω = 0 and dΩ ̸= 0. Moreover, at this locus we have

g̃(dΩ, dΩ) = 0, (3.19)

so that I are null hypersurfaces.

5
Pronounced as scri as a shorthand for script I.
6
Note that in order for r to be positive we need to restrict both u, v and ũ, ṽ to the domain u < v and ũ < ṽ.

27
 i+
R × Sd
I+ I+

M
f i0 i0

I− I−

i−

Figure 3.3: The Penrose diagram of Minkowski space. On the left we show the conformal
compactification wrapped on the Einstein cylinder. In the Penrose diagram on the right, the
time-like geodesics, i.e. lines with constant r, are illustrated as blue lines, while the space-like
geodesics, lines with constant t, are illustrated as the dashed red lines.

• Spatial infinity is defined as

i0 = p ∈ M

f ρ(p) = π , T (p) = 0 . (3.20)

At these points, the radius of the (d − 1)-sphere vanishes in the usual (d)-sphere degeneration.
At this point, both Ω = dΩ = 0.

• Future and past time-like infinity are defined as

i± = p ∈ M

f ρ(p) = 0 , T (p) = ±π . (3.21)

Again, the (d − 1)-spheres at these points have vanishing radius and Ω = dΩ = 0.

The motivations for this nomenclature follows from the analysis of (inextensible) geodesics in the
conformally compactified space. Indeed, one can show that space-like geodesics all end and start at
i 0 , while time-like geodesics start at i − and end at i+ . Finally, null geodesics start at I − and end at
I +.
Next, let us consider the maximally symmetric space-times: Minkowski space, de Sitter space and
anti-de Sitter space. These are solutions to the vacuum Einstein equations with resp. vanishing,
negative or positive cosmological constant. See Appendix E for more details and a variety of metrics
on (A)dS space-times. We leave the explicit construction of the conformal compactification in these
cases as an exercise, and restrict ourselves here to a qualitative discussion.
The hypersurface at infinity, I has a rather different flavour depending on the value of the cosmological
constant. In particular it is null for flat space, space-like for de Sitter, and time-like for anti de Sitter.

28
Moreover, as all of them are conformally flat, they can be mapped to a portion of the Einstein cylinder.
de Sitter space occupies a horizontal strip, while Anti de Sitter maps to a vertical strip, in line with
the nature of conformal infinity. In Figure 3.4 we show the Penrose diagram for (A)dS space-time.
As a final comment, note that Minkowski space and de Sitter are clearly globally hyperbolic, but
that AdS is not. For AdS, we need to present, not just data on an initial t = const. hypersurface, but
also data, or at least boundary conditions at time-like infinity. Otherwise, one can imagine incoming
radiation from infinity which in turn might cause very problematic instabilities.

+
IdS

IAdS IAdS

−
IdS

Figure 3.4: The Penrose diagrams for de Sitter (left) and Anti de Sitter (right) space-times.
±
In the de Sitter case, the left and right vertical line can be identified. The topology of IdS is
2
R × S , while for AdS conformal infinity is conformal to Mink1,d−1 .

For the maximally symmetric spaces there is an elegant alternative construction for their conformal
compactification though their embedding in R2,d+1 . To see this let us consider the Lorentzian
conformal group in (d + 1) dimensions, SO(2, d + 1). This group only acts on the compactification, as
it interchanges points at finite distance with points at infinity. The conformal group acts on Rd+3 by
orthogonal transformations preserving the quadratic form

X 2 = η I J X I X J = s2 − w2 + ηµν x µ x ν , (3.22)

where the coordinates on Rd+3 are given by X I = (s, w, x µ ) where I = 0, . . . , d + 2 and µ = 0, . . . , d.

To obtain the conformal compactification of the maximally symmetric space-times, we choose a
non-zero constant vector K I ∈ R(d+3) and define the metric,

η I J dX I dX J
dsK2 2 = (3.23)
(K · X )2 X 2 =0

where we defined the product between to vectors X · Y = X I Y I . Up to SO(2, d + 1) transformations,

K I is distinguished only by its norm K 2 so there are only 3 cases K 2 = −1, 0, 1. Since we divided by a
quadratic function, the metric is invariant under constant rescalings of the embedding coordinates X I .
Moreover, on X 2 = 0 the form X I dX I = 12 dX 2 vanishes so it is easy to see that under X I → f (X )X I , the
metric ds2I is invariant for any (non-vanishing) function f (X ). Thus we can scale X so that K · X = 1
on the interior of the conformal compactification. At conformal infinity we find that K · X = 0. The

29
isometry group of the metric dsK2 2 can then be found as the subgroup of SO(2, d + 1) that preserves
the vector K I .
Exercise 3.2. Show that by taking the vectors:

K = (1, −1, 0, 0, 0, 0) , (3.24)

K = (1, 0, 0, 0, 0, 0) , (3.25)
K = (0, 1, 0, 0, 0, 0) , (3.26)

the metric (3.23) reduces resp. to the metric on Minkowksi space, de Sitter space or anti-de Sitter space.
Use this to construct the conformal compactifications. Show that the resp. preserved isometry groups are
SO(1, d), SO(1, d + 1) and SO(2, d).
Exercise 3.3. As a final example consider the FLRW backgrounds with spatial sections of constant
curvature. The metric is given by
ds2 = dt 2 − a(t)2 dsd,k
2
, (3.27)

where for k = 0, 1, −1 the spatial manifold is respectively flat Euclidean space, the sphere or hyperbolic
space.
R t dt
Show that these metrics are all conformally flat. Hint: consider the conformal time τ = 0 a(t) .

So far all our examples were conformally flat, allowing us to use a variety of tricks to easily construct
their conformal compactification. For non-conformally flat space-times many of our tricks fail making
the task of finding the conformal compactification more involved. Conceptually the procedure remains
identical, as described in Definition 3.12 but has to be discussed on a case by case basis. However, if
M is globally hyperbolic we can see that I = I + ∪ I − where future infinity I + is to the future of a
Cauchy hypersurface and past infinity I − to the past.

3.4 Asymptotics and peeling

All the examples discussed above had an important feature in common, namely that they all admit
a smooth conformal extension which attaches a conformal boundary to the space-time. A natural
question is to what extend this property is shared by more generic manifolds.
Definition 3.13 (Asymptotically simple space-times). A space-time (M, g) is asymptotically simple if
there exists a smooth, oriented, time-oriented causal conformal compactification (Mf, g̃) such that
each null geodesic of (M
f, g̃) acquires two distinct endpoints on I .

Note that the completeness requirement in this definition excludes singular space-times such as the
Schwarzschild black hole in which there exist null geodesics which do not reach I . Not only those
falling into the black hole but also those lying in the photon sphere are incomplete in this sense.
Moreover, even without singularities, the fact that a space-time is smooth and geodesically complete
does not guarantee that it admits a smooth conformal compactification.
Exercise 3.4. Consider the Nariai space-time M = R × S 1 × S 2 with metric

ds2 = dt 2 − cosh2 tdψ − dsS22 . (3.28)

30
The Nariai space-time is both geodesically complete and globally hyperbolic. Show that it does not allow
for a smooth conformal extension.
To do so note that under a conformal transformation the squared Weyl tensor transforms as

Cµνρσ C µνρσ = Ω4 Ceµνρσ Ceµνρσ . (3.29)

Use this fact to show that a smooth conformal extension cannot exist for this space-time.

The requirement of asymptotic simplicity is very natural from a physical point of view and can be
thought of as stating that the matter density has to die off quickly enough at infinity such that the
asymptotic geometry is purely described by the cosmological constant. More precisely, we only allow
conformally invariant matter in the neighbourhood of I .7 Furthermore, we can prove the following
theorem

Theorem 3.2. Let (M, g) have conformal compactification (M f, g̃), and suppose that the space-time
asymptotically satisfies the Einstein equations with conformally invariant matter (so that the trace of the
stress-energy tensor vanishes) with cosmological constant λ. Then I is space-like when λ > 0, time-like
for λ < 0 and null when λ = 0.

Proof: From Einstein’s equations in d + 1 dimensions we immediately obtain

2(d + 1)
R= λ, (3.30)
d −1

in a neighbourhood of I . Define the Schouten tensor as

1 1
 ‹
Pµν = − Rµν − Rgµν . (3.31)
d −1 2d
µ
d+1
Near I we have that Pµ = − d(d−1) λ and note that near I it transforms as

1
Pµν = P̃µν + Ω−1 ∇ ˜ ν Ω − Ω−2 g̃µν ∇
˜ µ∇ ˜ ρΩ ,
˜ ρ Ω∇ (3.32)
2

under a conformal rescaling of the metric. Remember that on I we have Ω = 0 so that,

λ
g µν Nµ N ν =
e , (3.33)
24

on I , where we defined the normal vector Nµ = −∇

˜ µ Ω. From this the proposition immediately
follows.
Asymptotically simple space-times satisfying the conditions in the theorem above are called asymp-
totically flat, asymptotically AdS or asymptotically dS, depending on the value of the cosmological
constant. Moreover, using the results above one can analyse the topology of conformal infinity in
each case, the de Sitter and anti de Sitter cases are rather straightforward. For de Sitter one can show
that topologically I ≈ S d−1 , while for anti de Sitter it is a compact time-like hypersurface. The case
7
Conformally invariant matter comes with a traceless stress tensor and includes massless matter fields such as for
example a massless scalar field or massless gauge fields. Therefore this criterion still allows for radiation to reach infinity.

31
of vanishing cosmological constant on the other hand is harder to analyse. However, one can prove
that in any asymptotically simple space-time for which I is everywhere null, the topology of each
component I ± is given topologically by I ± ≈ R × S 2 , where the R factor can be thought of as the
rays generating null infinity.
To finish this chapter we briefly discuss one of the most important results of the theory of asymptotics of
the gravitational field, the so-called peeling theorem. The peeling theorem is based on the observation
that in asymptotically simple space-times the Weyl tensor must vanish on I and quantises the allowed
decay. A precise statement of the peeling theorem requires the introduction of a variety of new
concepts and goes beyond the scope of these lectures. For this reason we only present a simplified
sketch and refer the interested reader to the textbooks [PR84, Kro23] for more information.
As already mentioned before, it is reasonable to expect conformally invariant and massless fields to
continue smoothly to I in the conformal compactification. Consider a (conformally coupled) scalar
d−1
field φ solving the (conformally invariant) wave equation. Then φ̃ = φ/Ω 2 should be smooth on
I (at least if it is in the domain of dependence of M). In an asymtotically de Sitter space, where
λ > 0, we can deduce that the massless scalar will evolve past I as if it wasn’t there and so φ̃ will be
smooth and generically non-vanishing near I in the conformally extended space-time. Translating
this back to the physical space-time this gives a sharp asymptotic fall-off of the physical scalar φ. It is
instructive to compare this fall-off in terms of the affine parameter r along an outward going geodesic
terminating on I . When I is null we find

1 1
Ω∼ → φ∼ d−1
. (3.34)
r r 2

On the other hand, in the de Sitter case, it is easily seen that Ω ∼ exp(−t), where t is the proper time
along a time-like geodesic ending on I + .8 Hence, for massive fields we find an exponential fall-off
for the physical fields. In the asymptotically de Sitter case these conclusions can straightforwardly be
extended to higher spin/helicity fields.
These considerations can straightforwardly be generalised to study the fall-off of scalars in asymptot-
ically AdS or flat spaces. When λ = 0 however, the analysis is more involved for higher spin fields as
they can scale differently according to whether they are aligned with I or transverse to it. Carefully
doing so for the Weyl tensor results in a proper statement of the peeling theorem in asymptotically
flat spaces, see for example [PR84, Kro23] for a careful statement.

Remark. As a final remark, note that in the peeling theorem one is usually focused on asymptotically
flat space-times and the decay of gravitational waves at null infinity. In asymptotically AdS space-times,
in particular in the context of holography, a more common way of analysing asymptotic expansions
proceed through the use of the Fefferman-Graham expansion [FG85]. See Chapter 10 for more details
in the context of holography.

dT
8
This can be easily see by noting that dt = T , where T is the time coordinate on the Einstein cylinder.

32
 Chapter 4

Quantum fields in curved space

Having introduced all the necessary global properties of Lorentzian spacetimes we are well-equipped to
move on to quantum fields in curved spacetime. This chapter introduces all the necessary ingredients
and discusses general features of quantisation in curved spacetimes. In the next chapter we continue
the general discussion and focus on observables quadratic in the fields, such as the stress tensor
vacuum expectation value. The following chapters will then be devoted to applying this general
framework to particularly interesting examples. The main new feature in curved space will be that
we will no longer have a clear grasp of the concept of particle. Associated is the lack of uniqueness in
defining the vacuum state.

4.1 Classical fields in curved space-time

To formulate a classical field theory in curved spacetime we need to know how the various fields
couple to the background metric. Let us once more emphasize that the metric will not be a dynamic
field and we only wish to consider fixed background metrics. This situation is very similar to the
charged scalar we considered in Chapter 2.
Analogous to that case, we can couple the theory to a non-trivial background metric simply by
changing all partial derivatives into covariant derivatives, ∂µ → ∇µ . However, this is not quite enough.
In order to guarantee that the action transforms as a scalar under Lorentz transformations we need
p
to simultaneously change the measure dd+1 x to the Lorentz invariant measure |g|dd+1 x. This
procedure is in a sense the minimal consistent way to couple the system to gravity and for that reason
goes under the name minimal coupling.1
Taking for example the free massive scalar field we find the action
Z
1
|g| g µν ∇µ φ∇ν φ − m2 φ 2 .
Æ

S= dd+1 x (4.1)
2 M

This leads to the equation of motion,

□ + m2 φ = 0 ,


(4.2)
1
Moreover, this coupling is consistent with Einstein’s principle of equivalence, according to which local gravitational
effects are not present in a neighbourhood of the spacetime origin of a locally inertial frame of reference. A similar comment
applies to the theory with conformal coupling.

33
where now □ represents the d’Alembertian on the curved space,2

□φ = g µν ∇µ ∇ν φ = |g|−1/2 ∂µ |g|1/2 g µν ∂ν φ .
 
(4.3)

This is not the most general way of coupling the scalar field to a background metric. A slightly more
general action one can consider is,
Z
1
|g| g µν ∇µ φ∇ν φ − m2 φ 2 − ξRφ 2 ,
Æ

S= dd+1 x (4.4)
2 M

where R is the Ricci scalar of the curved manifold M. The term proportional to R disappears in flat
space so in this sense it is a generalization of the usual flat space action. The equation of motion after
including the curvature coupling is given by

□ + m2 + ξR φ = 0 ,


(4.5)

Remark. In principle one can consider more general higher derivative interactions in the Lagrangian.
Such terms could be of the form Rn φ 2 but equally well terms like (∂ µ φ∂µ φ) could be included. In
many context such terms are indeed present, but in this course we will mostly ignore them. However,
there are various arguments why to a first approximation this is a reasonable thing to do. First of
all, under the renormalisation group flow such terms will always be suppressed with respect to the
two-derivative couplings. In many cases these terms will be "irrelevant" and can therefore be ignored
at low energies. See the course on renormalisation for more details on this first point. Secondly, when
including higher degree terms in the Lagrangian a multitude of issues with causality arise which need
a careful treatment which goes beyond the scope of this course.

One reason why the curvature coupling Rφ is useful is that for a specific values of ξ its addition to
the Lagrangian of a massless scalar makes the action conformally invariant. For this value ξ = ξconf.
this coupling to the background metric is called conformal coupling.

Exercise 4.1. Consider a free massless scalar field in d + 1 dimensions. Show how the action transforms
under a conformal transformation of the background metric,

gµν → Ω(x)2 gµν , φ → Ω(x)∆ φ , (4.6)

for any function Ω(x) ̸= 0.

For which values of ξ and ∆ is the action invariant under conformal transformations? In conformally
invariant theories ∆ is called the scaling dimension of the field φ.

Next, let us briefly consider Maxwell theory, i.e. the Abelian gauge theory with gauge field Aµ (x) and
field strength Fµν = ∇[µ Aν] . For gauge fields, the differential form language really starts to show it’s
R
2
Written as a differential form, the kinetic term in the action takes the form dφ ∧ ⋆dφ, where ⋆ is the Hodge star
operator. In this language, the d’Alembertian takes the form □φ = ⋆ (d ⋆ dφ).

34
elegance. In this language, we write the gauge field as a one form

A = Aµ dx µ ∈ Ω1 (M) . (4.7)

In this language the field strength can be written as the two-form F = dA ∈ Ω2 (M). The action for
Maxwell theory coupled to a background metric is given by
Z Z
1 1
|g|dd+1 x Fµν F µν ,
Æ
S= F ∧ ⋆F = (4.8)
4 M 2 M

which is invariant under the gauge transformations, Aµ → Aµ + ∂µ f , or equivalently A → A + d f . The

Bianchi identity for F can be written as dF = ∇[µ Fνρ] = 0 while the equations of motion are given by

d ⋆ F = 0, or ∇µ Fµν = 0 . (4.9)

The gauge invariance noted above adds a redundancy to Maxwell theory. In order to obtain a
deterministic equation, one should first fix a gauge. A common choice of gauge in this situation is
provided by the Lorenz gauge defined by ∇a Aa = 0. Upon this gauge fixing, the equations of motion
reduce to the wave equation □Aa = 0. Note however that there is nevertheless still a residual gauge
freedom parameterised by A → A + d f provided that □ f = 0. Finally, one can also consider spinors in
curved spacetime. Since scalars and occasionally vector fields suffice for most of our purposes, we
defer a discussion of spinors in curved spacetimes to Appendix ??.

Exercise 4.2. Show how the Maxwell action transforms under conformal transformations. Demonstrate
that in d + 1 = 4 dimensions the action and equations of motion (and Bianchi identities) are conformally
invariant.

Just as in general relativity (and its generalisations such as e.g. Einstein-Maxwell theory) we can
define the energy momentum tensor as

2 δS
T µν = , (4.10)
|g|1/2 δgµν

which in general relativity this provides a source to the Einstein equation. In this course we consider
quantum fields in fixed background metric and do not consider fluctuations of the metric. One can
think of this as solving Einsteins equations with a source given by the vacuum expectation value of the
stress tensor 〈Tµν 〉. In a second step of our semi-classical treatment we can then consider fluctuations
of the dynamical fields around this background metric. Note that up to possible improvement terms
this reduces to the definitions in flat space presented in Chapter 2.
µ
Conformal invariance of a theory is manifested in the stress tensor as the property Tµ = 0. This is a
general consequence of the invariance of the action under variations of the form δgµν = σgµν .

Exercise 4.3. Show that the stress tensor of a conformally invariant theory is necessarily traceless.

Exercise 4.4. Compute the stress tensor for a minimal and conformally coupled scalar as well as Maxwell
theory in curved space.
Check that for the latter two the stress tensor is indeed traceless.

35
4.2 Canonical quantisation in curved spacetime

To avoid subtleties, in this chapter we always assume the spacetime to be globally hyperbolic with a
foliation by Cauchy hypersurfaces Σ t where t ∈ R is a time-like coordinate. Having discussed how
various fields couple to a background metric we repeat and modify where necessary the same steps
as in Chapter 2 to quantise the theory.
Concepts such as the phase space, the associated symplectic form and the commutation relations
generalise straightforwardly to curved space. The phase space for a scalar field on a spacetime M
can be defined as

Vφ (M) = φ ∈ C ∞ (M) (□ + m2 + ξR)φ = 0 = (φ, φ̇) ∈ C ∞ (Σ t ) × C ∞ (Σ t )

 
(4.11)

where we allow for an arbitrary coupling to the Ricci scalar. Note that here we defined the initial
data on the Cauchy hypersurface at time t and defined φ̇ = ∇n φ with nµ a unit normal vector to the
hypersurface at time t. Similarly, one can define the phase space for fermions or gauge fields entirely
analogous to the flat space case. The symplectic form is given by
Z Z
Ω(φ1 , φ2 ) = φ1 ⋆ dφ2 − φ2 ⋆ dφ1 = (φ1 φ̇2 − φ2 φ̇1 )dΣ (4.12)
Σ Σ

where dΣ is the volume form on the hypersurface Σ. For more details on hypersurfaces see Appendix
C.

Remark. Note that for the complex scalar field we can write the inner product more invariantly as
Z
〈φ1 , φ2 〉 = dΣµ Jµ , (4.13)
Σ

where dΣµ = dΣnµ and Jµ is the conserved current for the U(1) global symmetry,

Jµ = i φ1∗ ∇µ φ2 − φ2∗ ∇µ φ1 .


(4.14)

In order to proceed with the quantisation we need a Hamiltonian description of the theory, and
therefore a preferred time coordinate. Assuming there is a splitting (t, x m ) we will quantise the theory
using the hypersurfaces Σ t defined by t = constant. Let us be more explicit and write the metric as

ds2 = g00 dt 2 + 2g0m dtdx m − hmn dx m dx n , (4.15)

where hmn is the induced metric on the Cauchy hypersurface Σ. In terms of these coordinates, we
can write the momentum conjugate to φ as

δL
π= = |g|1/2 g 0µ ∇µ φ = |h|1/2 nµ ∇µ φ . (4.16)
δ(∂ t φ)

36
Exercise 4.5. Prove the second equality in Equation (4.16). The easiest way to do so is by explicit
computation.

Analogous to flat space we can now quantise the theory by promoting all fields to operators on a
Hilbert space and imposing the canonical commutation relations

φ(t, x), φ(t, x′ ) = 0 ,

 

π(t, x), π(t, x′ ) = 0 ,

 
(4.17)
φ(t, x), π(t, x′ ) = i δ(x − x′ ) .
 

where now the δ-function is normalised against the volume form on the hypersurface,
Z
′
f (x ) = δ(x − x′ ) f (x′ )dΣ . (4.18)
Σ

Similarly, this Hilbert space now comes with a generalised inner product,

〈φ1 , φ2 〉 = i Ω(φ1 , φ2 ) . (4.19)

Similar to flat space it follows immediately that this inner product is independent of the choice of
Cauchy hypersurface in the given foliation. For this reason it was justified not to specify the precise
time t in the above.
To further develop the quantum field theory we again want to relate the above phase space to a
one-particle Hilbert space H. However, in curved space there is no canonical analogue of the Fourier
transform so this step will need some more care.
Let us for the moment consider a complex scalar field φ. Given any solution f ∈ Vφ (M) to the Klein
Gordon equation we can associate to it an annihilation operator,

a : Vφ (M) → H : f 7→ a( f ) = 〈 f , φ〉 . (4.20)

Since φ is Hermitian, we have that the associated creation operator can be defined as

a† ( f ) = −a( f ∗ ) . (4.21)

More explicitly, we can write the annihilation operator (and similarly for the creation operator) in
terms of φ and its canonical momentum π as
Z Z
µ
1/2 ∗ ∗
dd x f ∗ π − |h|1/2 nµ ∂µ f ∗ φ .


a( f ) = i d
d x |h| n ( f ∂µ φ − ∂µ f φ) = i (4.22)
Σ Σ

37
From the canonical commutation relations introduced in Equation (4.17), we immediately find
Z Z
†
dd y ( f ∗ π − |h|1/2 nµ ∂µ f ∗ φ)(t, x)(gπ − |h|1/2 nµ ∂µ φ)(t, y)
   
a( f ), a (g) = d
d x
Z
(4.23)
dd x|h|1/2 nµ f ∗ (t, x)∂µ g(t, x) − ∂µ f ∗ (t, x)g(t, x)


=i

= 〈 f , g〉 .

Similarly, one can easily show that

[a( f ), a(g)] = − 〈 f , g ∗ 〉 , a† ( f ), a† (g) = − 〈 f ∗ , g〉 .

 
(4.24)

With this map a at hand we have provided an appropriate procedure to quantise the scalar field
and construct the one-particle Hilbert space. Similar to the flat space case, one can then proceed to
find the full Hilbert space in the Fock representation by consecutively applying creation operators
to the vacuum. However, this is not quite enough. In order to have a complete understanding of
this Hilbert space we need to find a complete orthonormal basis of the phase space, analogous to
the plane waves in flat space. I.e. we have to find a complete set of functions ui , solving the wave
equation and satisfying,
¬ ¶
ui , u j = δi j , u∗i , u j = 0 , u∗i , u∗j = −δi j . (4.25)

Having found such a basis, we can expand the field φ as follows,

X
ai ui + ai† u∗i ,


φ(x) = (4.26)
i

where the quantum annihilation and creation operators, ai and ai† , satisfy the standard commutation
relations, ” —
ai , a†j = δi j . (4.27)

A very important point is that on curved space, there is no natural way to perform this mode
decomposition. This property is key to many odd strange but interesting phenomena in the theory of
quantum field theory on curved space and will lie at the origin of many of the observations that will
follow.
A special scenario where there exists a "natural" choice of decomposition is when the spacetime
admits a globally well-defined, non-vanishing Killing vector K = ∂ t , which is irrotational, i.e. when
the spacetime is globally static. In this case one can write the metric as

ds2 (t, x) = g00 (x)dt 2 − hmn (x)dx m dx n , (4.28)

with h and g00 are respectively the metric and a positive function on the transverse (space-like)
manifold Σ, both independent of t. In this case the quantisation follows very closely in the steps of
quantisation in flat space. We can separate variables and analogous to flat space define the positive

38
modes Vφ+ as the set of functions Pn of the form

χn (x) −iωn t
Pn (x) = p e . (4.29)
2ωn

where the functions χn (x) collectively give a basis of functions on the manifold Σ and satisfy the
reduced wave equation,
∆ + ω2n χn = 0 ,


(4.30)

for some second order operator ∆ on Σ. The functions χn should satisfy the completeness relation,
X
χ̄n (x)χn (x ′ ) = δ(d) (x, x ′ ) , (4.31)
n

where the δ-function is understood to be normalised against the measure dd x

p
g00 h. Hence, it
follows that the inner product reduces to
Z
Æ
〈Pm , Pn 〉 = dd x g00 hχ̄m χn = δmn . (4.32)

Similarly, one can introduce negative frequency modes Nn ∈ Vφ− and complete the quantisation exactly
like in flat space. However, this limited approach does not generalise to generic curved spacetimes
and moreover, even though the Schwarzschild black hole is static we will see that this approach is
insufficient to fully understand quantum field theory on a black hole background.

Remark. In the above we silently assumed that Σ is compact such that the labels n take value in
a countably infinite set. More generally, on a non-compact spatial slicing such as R3 , n, m must be
replaced by continuous variables such as k, k′ for flat spatial slices. Similarly, the Kronecker delta,
δmn , should then be replaced by δ3 (k − k′ ) and n by R3 d 3 k/ωk . For the more formal development
P R

we will mostly stick to the discrete modes and introduce continuous alternatives when needed.

Another example of a natural choice of vacuum arises for conformally invariant theories in conformally
flat spacetimes. Consider a conformally flat metric, i.e. gµν (x) = Ω2 (x)ηµν . A solution to Klein-
Gordon equation for a conformally coupled massless scalar can then be obtained from the Minkowski
solution ukM (x) simply by using the conformal transformation of the scalar field. Therefore, we have
a natural choice of solutions of the wave equation given by
1−d
uk (x) = Ω(x) 2 ukM (x) . (4.33)

The vacuum associated to this choice of modes is called the conformal vacuum. In this case the analysis
of the two-point functions becomes particularly easy, as one can obtain the Wightman functions
simply as,
1−d 1−d
G± (x, x ′ ) = Ω(x) 2 G±M (x, x ′ )Ω(x ′ ) 2 , (4.34)

where G±M (x, x ′ ) is the Wightman function in Minkowski space introduced in Chapter 2. The Wightman
functions transform as bi-scalars under conformal transformations.

39
4.3 Quantisation in generic curved spacetimes

In general there does not exist such a natural choice of orthonormal basis. For example, consider a
spacetime that is asymptotically static both in the future and past region, respectively M+ and M− .
In both asymptotic regions one can write down a metric in the form (4.28) and define the natural
basis Pn± and Nn± resp. in the past and future region. This will lead to two notions of positive and
negative modes and in general we do not expect these to agree.
We can quantise the theory in the two regions as defined above. In order to construct the transition
between them we can impose that the quantum field φ is the same in both regions, i.e.
X X
an+ Pn+ + an+ Nn+ , .
† †
φ(x) = an− Pn− + an− Nn− = (4.35)
n n

The main question we’ll try to answer in this section is how to identify the Fock spaces F + and F − of
the different asymptotic regions. Similarly, in spacetimes with no asymptotically static regions, we
have to learn how to identify the Fock spaces at different times.
To do so, consider two different orthonormal bases of the phase space {un }n∈N and {vn }n∈N such that
X X
φ(x) = an un + an† u∗n = bn vn + bn† vn∗ , , (4.36)
n n

where an , an† and bn , bn† are the creation and annihilation operators defined with respect to either the
u basis or v basis. Since both sets of functions are independent bases of the C ⊗ Vφ we can relate
them through linear maps

αnm βnm
  X   X  
vn um um
∗ = ∗ ∗ = Snm † (4.37)
vn m
βnm αnm †
um m
um

where S is the classical S-matrix whose entries are called the Bogoliubov coefficients,

αmn = 〈un , vm 〉 , βmn = − u∗n , vm , (4.38)

and analogous for their complex conjugates. Since the u and v basis were both assumed to be
orthonormal, the Bogoliubov coefficients have to satisfy the normalisation relation,

αα† − ββ † = 1 , αβ † − βα† = 0 . (4.39)

Going back to (4.36) we find that the (transposed) Bogoliubov coefficients similarly encode the
relation between the a and b creation and annihilation operators,
 X ∗ ∗
αnm −βnm
  
bn am
† = (4.40)
bn m
−βnm αnm †
am

Exercise 4.6. Starting from (4.36) and (4.37), show that the Bogoliubov coefficients encode the relation
(4.40) between the a and b operators.
Exercise 4.7. Show that in order for the commutation relations of the b operators to be properly

40
normalised, we have to impose the condition (4.39).

Define the vacua with respect to the u or v expansion, |0〉a and |0〉 b in the usual way as the state
annihilated by all annihilation operators,

an |0〉a = 0 , bn |0〉 b = 0 , ∀n ≥ 0 . (4.41)

When β = 0 the two vacua are equivalent and can be identified up to possibly a phase. However, if
β ̸= 0, the b vacuum will contain a particles and vice versa! To see this more explicitly, define the
number operators, X
Nn(a) = an† an , N (a) = Nn(a) , (4.42)
n

and compute their expectation values in the b vacuum,

X
Nn(a) b
= |βmn |2 . (4.43)
m

Hence, we see that the changing gravitational field creates particles (in pairs). In a sensible physical
systems the number of quanta in any vacuum should be finite so we require that the Bogoliubov
transformation is such that Nn(a) b < ∞ for all n. Indeed, without this condition various problems
regarding the convergence of all the expressions in this section arise.
To find an explicit relation between the a and b vacuum we would like to extend the map S from the
classical phase space to the Fock space, S : Fa → F b . As a first step, we can construct the b vacuum
using the Bogoliubov coefficients. After some algebra, one can show that (up to phase) we have
¨ «
1 X
|0〉 b = exp † †
Mmn am an |0〉a , Mmn = (α∗−1 β ∗ )mn , (4.44)
| det α|2 m,n

Such states are sometimes called squeezed states, in analogy with the nomenclature for the ground
state for an oscillator with a different frequency. The full quantum evolution operator translating
between the two Fock spaces then takes the form
¨ «
1 X
† † ∗
S= exp Mmn am an − Mmn am an . (4.45)
| det α|2 m,n

This operator is now unitary as the exponent is skew Hermitian so it preserves norms.

Exercise 4.8. Show that the a and b vacua are related as in Equation (4.44). Hint: use the Baker-
Campbell-Hausdorff formula to expand expressions of the form as e−F an e F .

4.4 Cosmological particle creation

To illustrate the above let us consider the following cosmological FLRW model with a spatially flat
isotropically changing metric,
d
X
2 2 2 2
ds = dt − a(t) dx m , (4.46)
m=1

41
where the cosmological scale factor a(t) is an arbitrary non-vanishing function of time. We take the
cosmological scale factor to asymptotically approach the values3

a1 as t → −∞
a(t) = . (4.47)
a2 as t →∞

To analyse this setup it is convenient to change coordinates to conformal time

t
dt ′
Z
η(t) = , (4.48)
t0 a(t ′ )

such that the metric becomes

d
‚ Œ
X
ds2 = C(η) dη2 − 2
dx m , (4.49)
m=1

where C(η) = a2 (η). The d’Alembertian for this metric can be written as
d+1 d−1
€ Š
□φ = C(η)− 2 ∂η C(η) 2 ∂η φ − C(η)−1 ∇2 φ . (4.50)

The spatial translation symmetry allows the spatial dependence to be separated from the time
dependence so that we can consider solutions to the wave equation of the form,

eik·x 1−d
uk (η, x) = d
C(η) 4 χk (η) . (4.51)
(2π) 2

Direct computation then results in

3+d ‹2 
C(η)− C ′′ (η) d − 5 C ′ (η)
  
1−d

4
□uk = d
e ik·x
χk′′ (η) + k2 χk (η) + + χk (η)
(2π) 2 2 C(η) 4 C(η)
3+d
(4.52)
C(η)− 4
ik·x
χk′′ (η) + k2 χk (η) − ξ(d)C(η)Rχk (η)
 
= d
e .
(2π) 2

where R is the Ricci scalar of the FLRW metric and ξ(d) = 14 1−d
d . The above can also be obtained by
noting that the FLRW metric is conformally flat with conformal factor Ω2 = C(η) and performing
a conformal transformation. Putting the above together, we find that uk solves the Klein-Gordon
equation if χk solves
χk′′ + ω2 (η)χk = 0 , (4.53)

where
ω2 (η) = k2 + m2 C(η) + C(η)R(η) (ξ − ξ(d)) . (4.54)

The induced metric on the hypersurface at constant η is given by

hmn dx m dx n = C(η)δmn dx m dx n , (4.55)

3
As usual we are not very careful with the appropriate fall-off and smoothness conditions but assume the function a(t)
is sufficiently well-behaved.

42
so that the inner product between two solutions to functions is given by
Z
d−1
u∗k ∂η uk′ − ∂η u∗k uk′


〈uk , uk′ 〉 =i dd xC(η) 2

(4.56)
(d) ′
χk∗ ∂η χk′ − ∂η χk∗ χk′


=iδ (k − k ) .

Hence we find that the modes are properly normalised if

i χk∗ ∂η χk′ − ∂η χk∗ χk′ = 1 .

(4.57)

This is a normalisation condition on the Wronskian of the solutions. The fact that the Wronskian
is independent of η is a manifestation of the fact that the inner product does not depend on the
particular hypersurface we use.
Going back to the problem at hand, let us consider the special case when the function C(η) approaches
a constant value in the past and future. In this case the asymptotic past and future are Minkowski
spacetimes and the η-dependent frequencies have limiting values

ωin as η → −∞

ω(η) = . (4.58)
ωout as η → ∞

As discussed above, we have two natural vacua, one in the asymptotic past and one in the future.
in/out
These vacua are defined by considering bases of solution uk satisfying the following asymptotic
conditions,
1
χkin (η) → p exp (−iωin η) , as η → −∞ ,
2ωin
(4.59)
out 1
χk (η) → p exp (−iωout η) , as η → ∞ .
2ωout
in/out †
€ Š
in/out
These modes then define annihilation and creation operators ak and ak and the corres-
ponding vacua are annihilated by
in/out
ak 0in/out = 0 . (4.60)

To relate these vacua, we have to compute the Bogoliubov coefficients (4.38). Due to the spatial
homogeneity, we immediately see that αk,k′ ∝ δ(k − k′ ) and similarly, βk,k′ ∝ δ(k + k′ ). Hence we
have
αk,k′ = αk δ(k − k′ ) , βk,k′ = βk δ(k + k′ ) . (4.61)

Note that this means that the Bogoliubov coefficients only mix modes with wave vectors k and −k
and in addition they only depend on the magnitude of the wave vector due to rotational symmetry.
The Bogoliubov transformation then takes the form

∗out
uin out
k = αk uk + βk u−k (4.62)

The first condition in (4.39) is automatically satisfied, while the second gives

|αk |2 − |βk |2 = 1 (4.63)

43
The condition that the vacuum in the in region contains only a finite number of modes in the out
Hilbert space becomes, Z
dd k|βk |2 < ∞ . (4.64)

Remark. Note that since FLRW metric is conformally flat, in the case of a massless, conformally
coupled scalar we can always construct a conformal vacuum. Indeed, in this case we have

ω2k (η) = k2 (4.65)

and the solution simplifies drastically to

1
χk (η) = p e−ikη . (4.66)
2ωk

In this case the function uk can simply be obtained as a conformal transformation of the plane waves
in standard Minkowski space.

Exercise 4.9 (An exactly solvable model). To get a better feeling for these types of properties, it is
interesting to work out some non-trivial models where the equations for χk can be solved explicitly. One
such model is given by the two dimensional spacetime with ξ = 0 and C(η) = A + B tanh(ρη) studied
in [BD77].
Verify that the equations of motion for χk can be solved by the following two sets of functions,

1 iω−
• ˜
χkin (η) = p exp −iω+ η − log(2 cosh ρη) (4.67)
2ωin ρ
ρ + iω− iω− ρ − iωin 1
 ‹
× 2 F1 , , , (1 + tanh ρη) , (4.68)
ρ ρ ρ 2
1 iω
• ˜
−
χkout (η) = p exp −iω+ η − log(2 cosh ρη) (4.69)
2ωout ρ
iω− iω− iωout 1
 ‹
× 2 F1 1 + , ,1 − , (1 + tanh ρη) , (4.70)
ρ ρ ρ 2

where ω± are defined as ω± = 12 (ωout ± ωin ).

Show that these functions have the appropriate plane wave limit as η → ±∞ and show that the squared
Bogoliubov coefficients are given by

πω+ πω−
€ Š € Š
sinh2 ρ sinh2 ρ
|αk |2 = Š, |βk |2 = Š. (4.71)
πωin πωout πωin πωout
€ Š € € Š €
sinh ρ sinh ρ sinh ρ sinh ρ

Hint: Use the identities for hypergeometric functions introduced in Appendix F.

Remark. In curved space an interesting connection between statistics and dynamics appears that

44
is not present in Minkowski space. The statistics is determined by the algebra of creation and
annihilation operators. Commuting operators give rise to bosons, while anti-commuting operators
give rise to fermions. This is nothing new, but as we will argue, in curved spacetime this is the only
consistent option!
To see this, consider a spin-0 field, i.e. a scalar field in curved space and two possible vacua, the
a and b vacuum among which we can interpolate using the Bogoliubov coefficients defined above,
satisfying |α|2 − |β|2 = 1. Note that this relation follows purely from the field equations. Assume that
the a creation and annihilation operators satisfy the (anti-)commutation relations
 † †
am , an†
 
[an , am ]± = am , an ± = 0 , ±
= δmn , (4.72)

where the plus sign denotes the anti-commutator and the minus the commutator. Following the
above, one can show that in the b vacuum, the (anti-)commutation relations become

∗ ∗
 † †
[bm , bn ]± = bm , bn ± = (αmn βmn ± αmn βmn )δm,−n , (4.73)

and
bm , bn† = (|α|2mn ± |β|2mn )δm,n ,
 
±
(4.74)

Hence, we see that in order for the particles in the b vacuum to satisfy the same statistics as the
particles in the a vacuum it is necessary to pick the minus sign such that (4.73) vanishes identically
and (4.74) reduces to the standard commutation relations for bosons.
This is a purely curved space derivation, since in flat space β = 0 and the connection derived above is
absent and both statistics seem allowed. Similar conclusions can be reached for fermionic fields as
well as higher spin fields.

45
 Chapter 5

Observables quadratic in fields

In the previous chapter we discussed the Hilbert space and non-uniqueness of the vacuum. Another
important set of observables is given by the two-point functions. Indeed, a number of physically
interesting quantities, such as the action and the energy-momentum tensor are quadratic in the fields
and their derivatives, evaluated at a single point. As discussed in Chapter 2, the expectation values of
such quantities usually diverges and has to be regularized. In flat space this could be done through
normal ordering, but in curved space the implicit gravitational interactions introduce additional
divergences. Furthermore, the vacuum energy needs to be treated very carefully since it can give
rise to gravitational effects through the gravitational field equations. For these reasons more care is
needed in the process of renormalisation.

5.1 Two-point functions

Similar to the situation in flat space, the most elementary Green’s functions are the Wightman
functions,
G + (x, x ′ ) = 〈0| φ(x)φ(x ′ ) |0〉 , G − (x, x ′ ) = 〈0| φ(x ′ )φ(x) |0〉 , . (5.1)

Using these expression, we can obtain the various other Green’s functions, following the same
identities as in flat space, see (2.36), (2.37) and (2.38).
Given the mode expansion X
an un (x) + an† u∗n (x) ,


φ(x) = (5.2)
n

appropriate for a given choice of vacuum one can compute the Wightman function G + (and similarly
G − ) as X
G + (x, x ′ ) = un (x)u∗n (x ′ ) . (5.3)
n

Furthermore, since the field φ satisfies the wave equation we necessarily have

(□ x + m2 + ξR)G ± (x, x ′ ) = 0 . (5.4)

Although conceptually clear, finding the appropriate basis of functions and computing the Wightman
functions is often prohibitively hard and only in very special cases will we be able to obtain analytic
expressions.
Such cases are usually characterised by some sort of additional symmetry. As in flat space, each

46
symmetry gives rise to some sort of conserved quantity attached to a Cauchy surface Σ,
Z
Q= Jµ dΣµ , ∇µ Jµ = 0 . (5.5)
Σ

Due to Gauss law this charge is conserved, as long as the Cauchy surface does not cross any charged
objects. This formula is perhaps most familiar for internal global symmetries but is easily extended
to include space-time symmetries. Namely, when the space-time has a Killing vector K µ we imme-
diately find the conserved current Jµ = K µ Tµν with the associated charge being the momenta for
(space) translation invariant theories, angular momenta for rotationally invariant theories and the
Hamiltonian, or energy, for time translations.
The more symmetries are present in our setup the more constrained are the Green’s functions. This
will become most clear when working in maximally symmetric spaces, such as (A)dS, where the form
of the Green’s functions can be explicitly found after imposing all the symmetry constraints. See
Chapter 8 and 10 for more details in these cases.
Another case where computations simplify dramatically is when we have a conformal vacuum, i.e. a
conformal theory on a conformally flat manifold. In this case the Green’s functions can simply be
obtained through a conformal transformation from Minkowski space.

5.2 Adiabatic expansion of the Green’s function

In the absence of symmetries, we have little hope of explicitly finding the mode functions and
constructing the Green’s functions. However, not all is lost as in many cases we can make progress by
working perturbatively. We will be schematic and highlight the results while skipping many technical
steps in the derivation. More details can be found in [BD84, PT09].
When the space-time curvature is small and slowly changing, we do not expect particles with arbitrary
large energies to be created. In other words, if the metric changes sufficiently slow, or adiabatic, the
particle number in a given mode should not change. We can make this more precise by introducing a
formal parameter T in the metric,
 t x
T
gµν (t, x) → gµν (t, x) = gµν , . (5.6)
T T

We call the expansion in T −1 the adiabatic expansion, which roughly counts the number of derivatives
of the metric. Indeed, we find that the scalar curvature R is of ’second adiabatic order’. We will not
explicitly write this parameter but it can easily be introduced by counting derivatives. However, in
the expansions below we will sometimes write O(T −3 ) for example which can be understood as up to
terms of third adiabatic order.
To illustrate how this expansion can be used, let us consider again a free neutral scalar in d = 3 + 1
space-time dimensions, with equation of motion

(□ + m2 + ξR)φ = 0 . (5.7)

An all-powerful being would now proceed to find the mode functions and subsequently use them

47
to compute the Green’s function. For us, mere humans, this is usually not possible. However, one
can show that Feynman’s Green’s function can be obtained directly as an adiabatic expansion. As
discussed in flat space, the Feynman Green’s function satisfies

(□ + m2 + ξR)G F (x, x ′ ) = −δ(x, x ′ ) , δ(x, x ′ ) = |g(x)|−1/2 δ(x − x ′ ) . (5.8)

The minus sign is convention, and both δ(x, x ′ ) and G F (x, x ′ ) transform as biscalars. To obtain the
adiabatic expansion, rewrite the propagator as
Z ∞
2
′
G F (x, x ) = −i dse−im s K(x, x ′ ; s) , (5.9)
0

where m2 is understood to have a small imaginary part m2 − iε so that there is no divergence as

s → ∞. The kernel K satisfies the Schrödinger type equation

i∂s K(x, x ′ ; s) = (□ x + ξR)K(x, x ′ ; s) , (5.10)

with the boundary condition that K(x, x ′ ; s) ∼ |g(x)|−1/2 δ(x − x ′ ) as s → 0.

Exercise 5.1. Show that the kernel K(x, x ′ ; s) satisfies the equation (5.10) with the boundary conditions
mentioned above.

Equation (5.10) implies that we can expand the kernel K in powers of the parameter s. This can be
made explicit by writing the kernel K as

∆1/2 (x, x ′ ) σ(x,x ′ )

K(x, x ′ ; s) = i e 2is F (x, x ′ ; is) , (5.11)
(4π2 )(is)2

where ∆(x, x ′ ) is the Van Vleck-Morette determinant and σ(x, x ′ ) is related to the proper distance
along the geodesic from x to x ′ ,1

1
∆(x, x ′ ) = −|g(x)|−1/2 det −∂ x µ ∂ x ′ν σ(x, x ′ ) |g(x ′ )|−1/2 , σ(x, x ′ ) = τ(x, x ′ )2 ,


(5.12)
2

where τ is the proper distance along the geodesic. The adiabatic expansion of G F can now be
rephrased as the following expansion of the function F ,

F (x, x ′ ; s) ∼ a0 (x, x ′ ) + (is)a1 (x, x ′ ) + (is)2 a2 (x, x ′ ) + · · · . (5.13)

where the first coefficients are given in the coincidence limit x → x ′ as

1
‹
a0 (x) = 1 , a1 (x) = − ξ R, (5.14)
6
‹2
1 1 1 1 1 1
 ‹ 
µνρσ µν
a2 (x) = Rµνρσ R − Rµν R − − ξ □R + − ξ R2 . (5.15)
180 180 6 5 2 6

In this expansion we did not keep the parameter T explicit but by counting the derivatives on the
metric, one can easily see that the term an is of adiabatic order 2n. If the metric is smooth, one can
1
Here we assume x ′ is in a normal neighbourhood of x such that only one geodesic goes from x to x ′ .

48
continue this expression indefinitely and find a unique expansion. However, typically this expansion
is asymptotic so the solutions will in general not be uniquely determined.

Exercise 5.2. To make this more explicit, consider the FLRW metric with flat spatial slices,

ds2 = dt 2 − a(t)2 dx2 . (5.16)

The wave equation for the scalar field it

□ + m2 + ξR φ = 0 .


(5.17)

(Hint: Note that the Ricci scalar for this metric is R = 6(ȧ2 /a2 + ä/a))
Show that the modes for the scalar field can be written as

fk (x) = eik·x hk (t) , (5.18)

where hk (t) satisfies

ḧk + ω2 + σ hk = 0 ,


(5.19)
Ç
k2
where ω = a2
+ m2 . Find an expression for σ in terms of a and its derivatives.
The adiabatic expansion for the field modes is based on the usual WKB ansatz,
 Z t 
1 ′ ′
hk (t) = p exp −i Wk (t )dt , (5.20)
Wk (t)

where Wk (t) can be expanded in an adiabatic expansion,

Wk (t) = ω(0) + ω(0) + ω(0) + O(T −2 ) , (5.21)

where each ω(n) is of nth adiabatic order, i.e. contains n derivatives with respect to t.
Show that Wk (t) satisfies the following equation,

2
3 Ẇk 1 Ẅk
Wk2 =ω +σ+2
2
− , (5.22)
4 Wk 2 Wk

and solve this equation perturbatively in a large T expansion. (Hint: explicitly reintroduce T and
expand.)
Show that all odd ω(odd) vanish and that

ω(0) =ω , (5.23)
1 3 2 1
 ‹
ω(2) = σω 2
+ ω̇ − ω ω̈ , (5.24)
2ω3 4 2
1 3 1
 ‹
(2) 2
ω(4) = (2) 2 (2) (2) (2)


2σωω − 5ω ω + ω̇ ω̇ − (ω ω̈ + ω̈ω . (5.25)
2ω3 2 2

Exercise 5.3. Using the results from the previous exercise we can compute the Green’s function in the

49
adiabatic expansion. In the coincidence limit the Green’s function takes the form
Z ∞
G(x, x) ≃ dkk2 Wk−1 (5.26)
0

Expand this integral adiabatically and show that only the first to adiabatic orders contain divergences.
More precisely, show that the Green’s function can be written as,
Z ∞
1 1 1
•  ‹ ˜
R 2 R
G(x, x) = + dkk − ξ− . (5.27)
288π2 4π2 a3 0 ω 6 2ω3

After removing the divergent terms, we find a finite result for the Green’s function (at coincident points).

Note that the adiabatic expansion described above can be continued to arbitrary order and it is even
possible to find analytic expressions for all of the higher order terms in terms of first two [dRNS15]. The
perturbative expansion is therefore uniquely determined. However, this expansion is an asymptotic
expansion and is only uniquely defined up to non-perturbative terms.

5.3 Stress tensor renormalisation

In free theories, the Green’s function suffice to solve the full theory as all higher point functions can
simply be obtained by Wick contraction. As such, many interesting observables, like the stress tensor
vacuum expectation value, are quadratic in the fields and can be obtained through coincidence limits
of the Green’s functions.
However, in four dimensions, for arbitrary ξ one can easily see that the first two terms in the expansion
of the Green’s function (5.13) are divergent in the UV limit, i.e. when s → 0 and σ → 0. This is easy
to see after performing the ds integration for the first two terms,

|g(x)|1/4 p a1 (x, x ′ ) p
• ˜
(2) m
G F (x, x ′ ) = −i p K1 (m −2σ) + K0 (m −2σ) , (5.28)
4π2 −2σ 2

where Ki are Bessel functions of the second kind. Higher order terms do not involve UV divergences
for the two-point function but as we will see the fourth adiabatic order is necessary to tame the
logarithmic divergence of the stress tensor vacuum expectation value.
In order to obtain the physical expectation value for the two-point function evaluated at coincident
points, we need to subtract the divergent pieces. The adiabatic regularisation or subtraction method
tells us that we can do so in n = 4 dimension as

〈0| φ(x)φ(x ′ ) |0〉phys. = lim 〈0| φ(x)φ(x ′ ) |0〉 − i∆1/2 (x, x ′ )(4π)−n/2 ×
n→4
Z ∞  (5.29)
σ(x,x ′ )
−n/2 −im2 s+ ′


× ds(is) e 2is 1 + a1 (x, x )(is) .
0

This quantity has a smooth, well-defined limit as x → x ′ in any space-time. Note that we took a limit
n → 4, to regularise the infinite pieces. This is an example of dimensional regularisation.

50
Next, let us apply this procedure to the stress tensor vacuum expectation value 〈T 〉 and let us focus
on a conformally coupled scalar, with ξ = 16 in the massless limit, m → 0. The trace of the stress
tensor is given by
1 1
T µ µ = −∇µ φ∇µ φ + 2m2 φ 2 + Rφ 2 + □φ 2 , (5.30)
6 2
which after applying the equations of motion simply becomes

T µ µ = m2 φ 2 . (5.31)

Formally one might therefore be tempted to identify the vacuum expectation values

T µ µ = m2 φ 2 , (5.32)

and conclude that in the massless case the expectation value of the trace vanishes. However, note that
(5.32) does not imply that T µ µ phys. = m2 φ 2 phys. ! The reason is that the separate components of
the stress tensor have residual logarithmic divergences after subtracting the counterterms (5.29).
Indeed, in general one has to be very cautious in using equations of motion in the physical expectation
values.
In order to cancel the additional logarithmic divergence we need to furthermore subtract the coun-
terterm proportional to a2 , in order to find,

µ
〈0| T µ (x) |0〉phys. = lim m 2
〈0| φ(x)2 |0〉 − i∆1/2 (x, x ′ )(4π)−n/2 ×
n→4
Z ∞  (5.33)
2 σ(x,x ′ )
ds(is)−n/2 e−im ′ ′ 2
s+ 2is


× 1 + a1 (x, x )(is) + a2 (x, x )(is) .
0

Note that in the massless limit, this expression reduces to

1
lim 〈0| T µ µ (x) |0〉phys. = − a2 (x) ξ= 16
. (5.34)
m→0 (4π)2
1
where for ξ = 6 the second adiabatic coefficient simplifies to

1 1 1
a2 (x) ξ= 16
= Rµνρσ Rµνρσ − Rµν Rµν − □R
180 180 180 (5.35)
1 1 1
= W− E− □R .
120 180 360

where we introduced the square of the Weyl tensor and Euler density (also known as the Gauss-Bonnet
term in four dimensions),

1
W =Cµνρσ C µνρσ = Rµνρσ Rµνρσ − 2Rµν Rµν + R2 ,
3 (5.36)
µνρσ µν 2
E =Rµνρσ R − 4Rµν R + R .

This quantum violation of the tracelessness of the stress tensor in a conformal theory is known as the

51
trace anomaly and is characterised in four dimensions by a and c which appear as

T µ µ = (a E − c W + b □R) . (5.37)

Hence with our chosen normalisation we see that the conformally coupled free scalar contribute as
follows to the anomalies,
1
(a, c)conformal free scalar = (1, 3) . (5.38)
90(8π)2
The coefficient b on the other hand is scheme dependent and can be removed by adding appropriate
UV counterterms. The constants a and c are scheme independent and therefore characterise the
theory. Similar to the c trace anomaly in two dimension, the a anomaly is monotonous along the RG
flow. Indeed, if we have a flow from a UV CFT to another IR CFT we have aUV > aIR [Zam86, KS11].
In four dimensions, the constant c does not universally satisfy such monotonicity properties. See your
course on CFT for more details on a-theorems and trace anomalies.
In this chapter we discussed how to renormalise the scalar two-point function and stress tensor
vacuum expectation value using the adiabatic subtraction scheme. In the literature a variety of
other renormalisation schemes has been used, such as Hadamard renormalisation, proper time
renormalisation or dimensional regularisation to name a few. All these schemes give a different
prescription to remove the infinities, however, on scheme independent quantities they should and are
in many case proven to produce identical results.

52
 Part II

APPLICATIONS

53
 Chapter 6

The Unruh effect

A fundamental application of the formalism introduced above can already be seen in flat space.
Indeed, the general principles of relativity state that it should be possible to express the laws of
physics as being the same for all observers, even those undergoing acceleration. Indeed, the effects
of constant acceleration are equivalent to those caused by the presence of a uniform gravitational
field. Thus one can ask the question: in flat space, how does the Minkowski vacuum appear to an
accelerating observer? The perhaps surprising answer to this question lies in the Unruh effect which
states that this observer will perceive a thermal state!
Studying the physics of an accelerated observer will highlight many of the properties we discussed in
the previous chapters. Due to its simplicity we will be able to exactly solve this problem and explicitly
see the theory at work. Moreover, as we will see in the next chapter, many of the curious properties
observed in this case immediately carry over to the study of an evaporating black hole.

6.1 Thermal states

Before deriving the Unruh effect, let us briefly review some aspects of thermal states in quantum field
theory. Thermal states are a feature of statistical physics at a temperature T , the equilibrium state is
a probability distribution of physical states. In quantum mechanics such a distribution is given in the
form of a density matrix,

Definition 6.1. A density matrix is an element ρ ∈ H ⊗ H∗ that is Hermitian, positive definite and
has unit trace tr ρ = 1.

Such matrices are always diagonalizable and an orthonormal basis of states |n〉 can be found such
that they can be expressed in the form
X
ρ= pn |n〉 〈n| (6.1)
n

where the coefficients are positive pn ≥ 0 and n pn = 1. In this context, the coefficients pn can be
P

thought of as the probability of the ensemble to be in the state |n〉 such that the expectation value of
an observable A can be computed as
X
〈A〉ρ = tr ρA = ρn 〈n|A|n〉 . (6.2)
n

A density matrix represents a pure state when ρ = |ψ〉 〈ψ| for some normalized state |ψ〉, i.e., ρ has
rank 1, otherwise states are said to be mixed. Mixed states arise naturally when part of a quantum

54
system is hidden. Consider for example a pure state |ψ〉 ∈ H = H L ⊗ HR where the Hilbert space H L
is hidden for an observer. The state observed by the observer is then obtained by the partial trace
over H L . More precisely, consider the state |ψ〉
X
|ψ〉 = ψ r,s |r〉 L ⊗ |s〉R . (6.3)
r,s

where {|r〉} represents a basis of H L and {|s〉} of HR . The partially traced density matrix perceived by
the observer is then given by
X
ρR = Tr L |ψ〉 〈ψ| = ψ̄ r,s1 ψ r,s2 |s2 〉 〈s1 | . (6.4)
r

The observed state is mixed if the density matrix ρR has rank greater than one. In this case the systems
R and L are said to be entangled. Given a Hamiltonian H, we define a thermal state as follows,1

Definition 6.2. A thermal state of temperature T is a state of the form

1 1 X −β En 1
ρβ = exp(−β H) = e |n〉 〈n| , β= , (6.5)
Zβ Zβ n T

where 〈n| is a basis of energy eigenstates of energy En and

X
Zβ = Tr exp(−β H) = e−β En , (6.6)
n

is the partition function.

Note that we use units where the Boltzmann’s constant kB = 1. This is in exact analogy with the
canonical ensemble in statistical mechanics.

Example 6.1 (Bose-Einstein distribution). In the context of a harmonic oscillator (i.e., a single mode of
a quantum field), we can compute the thermal expectation of the number operator N using En = (n+ 12 )ω
to obtain
−β nω
P
n ne 1 d 1
Tr (ρβ N ) = = log(1 − e−βω ) = βω . (6.7)
ω dβ
P
−β nω e −1
ne

A useful way to characterize thermal states in quantum mechanics and quantum field theory was
described by Kubo, Martin and Schwinger [Kub57, MS59] and is called the KMS condition. Similarly
a state satisfying the KMS condition is called a KMS state.

Definition 6.3 (KMS state). A KMS state is a state for which the time evolution of operators A → A t
can be continued to complex time in such a way that for a time-independent operator, B, we have

〈A t B〉K M S = BA t+iβ K MS
, (6.8)

where 〈Az B〉K M S and 〈BAz 〉K M S are analytic functions of z in the strip 0 < im z < β.
1
One can decorate this definition with chemical potentials when the particle number is allowed to change, i.e. in a
grand canonical ensemble.

55
For finite systems the definition of the KMS condition is equivalent to the the definition of thermal states
above. To see this recall that in the Heisenberg representation, an operator A has time dependence
A t = eiH t A0 e−iH t . For our thermal density matrix above we can compute

1 1
〈A t B〉β = tr(e−β H A t B) = tr(e−β H+iH t A0 e−iH t B)
Zβ Zβ
1 1
= tr(A t+iβ e−β H B) = tr(e−β H BA t+iβ )
Zβ Zβ
= 〈BA t+iβ 〉β ,

where we have used the cyclic property of the trace. This is often interpreted as the property that our
system can be analytically continued to Euclidean signature with periodicity in imaginary time. In
infinite dimensions these manipulations are a lot more subtle as we might encounter phase transitions,
spontaneous symmetry breaking, operators that are not trace class and so on. However, the content of
the KMS condition is precisely that a similar relation continues to hold in the thermodynamic limit.2
We can apply the same analytic continuation to Green’s functions. The thermal Green’s function can
be obtained as the analytic continuation of the Wightman function or alternatively as the Green’s
function for the relevant operator, i.e., the Laplacian on R3 × Sβ1 where now S 1 is a circle of length β.

Definition 6.4. The thermal Green’s function is defined as

Tr(e−β H φ(x)φ( y))

Gβ (x, y) = . (6.9)
Tr(e−β H )

Furthermore, when our system has a time-like symmetry, the thermal Green’s function Gβ = Gβ (t −
t ′ , x, x′ ) and satisfies the KMS condition, i.e. it is periodic in imaginary time with period iβ. This
property follows directly as above from the KMS condition.

Example 6.2. In flat space, the thermal propagator can be constructed by images in imaginary time of
period β. Thus we can identify the thermal greens function on Minkowski space for the massless wave
equation as
X 1
Gβ (x, x ′ ) = 2 ((t − t ′ + inβ + iε)2 − x · x)
. (6.10)
n∈Z
4π

6.2 Particle detectors

Since in general curved space-times the notion of a particle is observer-dependent it will prove useful
to give a coordinate independent characterisation of the temperature. A useful way to achieve this is
to consider an observer equipped with a so-called Unruh detector [Unr76, GH77b].
The detector will have some internal energy states and can interact with the scalar field by exchanging
energy, i.e. by emitting or absorbing scalar particles. The detector could for example be constructed so
that it emits a ‘ping’ whenever its internal energy state changes. All observers will agree on whether or
not the detector has pinged, although they may disagree on whether the ping was caused by particle
2
If a phase transition takes place or if some symmetry is spontaneously broken, the KMS state might not be uniquely
defined.

56
emission or absorption. Such a detector can be modelled by a coupling of the scalar field φ(x(τ))
along the world-line x(τ) of the observer to some operator m(τ) acting on the internal detector states
Z ∞
g dτm(τ)φ(x(τ)) , (6.11)
−∞

where g is the strength of the coupling and τ is the proper time along the observer’s world-line. Let
H0 denote the detector Hamiltonian, with energy eigenstates E j ,

H0 E j = E j E j , (6.12)

and let mi j be the matrix element of the operator m(τ) at τ = 0,

mi j = 〈Ei | m(0) E j . (6.13)

We will calculate the transition amplitude from a state |0〉 ⊗ |Ei 〉 ∈ Hφ ⊗ Hdet. in the tensor product of
the scalar field and detector Hilbert spaces to the state E j ⊗ 〈ψ|, where |ψ〉 is any state of the scalar
field. To first order in perturbation theory (for small g) the desired amplitude can be computed as
Z ∞
A= g dτ E j ⊗ 〈ψ| m(τ)φ(x(τ)) |0〉 ⊗ |Ei 〉 . (6.14)
−∞

Using (in the Heisenberg picture) that m(τ) = eiHτ m(0)e−iHτ , this can be written as
Z ∞
A = g m ji dτ ei(E j −Ei )τ 〈ψ| φ(x(τ)) |0〉 . (6.15)
−∞

Since we are only interested in the probability for the detector to make the transition from Ei to E j ,
we should square this amplitude and sum over the final state |ψ〉 of the scalar field, which will not be
measured. Using the resolution of identity ψ |ψ〉 〈ψ| = 1 we find the probability
P

Z ∞
′
2 2
dτdτ′ e−i(E j −Ei )(τ −τ) G+ x(τ′ ), x(τ) ,


P(Ei → E j ) = g |mi j | (6.16)
−∞

where G+ is the Wightman function. Notice that the prefactor in (6.16) depends on the details of the
detector, so it is useful to extract the piece which depends only on the scalar field and the world-line
trajectory. For this reason we define the detector response function
Z ∞
′
dτdτ′ e−iE(τ −τ) G+ x(τ′ ), x(τ) ,


F (E) = (6.17)
−∞

When the Wightman function only depends on ∆τ = τ′ − τ we can change variables to ∆τ and
′
τ̄ = τ 2+τ . The detector response function is then defined by removing the diverging volume factor
coming from the integration over τ̄,
Z ∞
f (E) = d∆τ e−iE∆τ G+ (∆τ) . (6.18)
−∞

57
Example 6.3. First consider Minkowski space in the vacuum state. The Wightman function is then given
by
1 1
G+ (∆τ) = . (6.19)
4π (∆τ − iε)2
2

If we plug this in the formula (6.18), we can calculate the integral by residues. Since E > 0 we should
close the contour for ∆τ in the lower half-plane, since in this case the integral at the half-circle at infinity
goes to zero due to the damping factor e−i E ∆τ (Jordan’s lemma) and we conclude that

f (E) = 0 . (6.20)

So, unsurprisingly, we find that there is no particle detection in the Minkowski vacuum.

Example 6.4. We can also arrive at the Bose-Einstein distribution from the Green’s function by computing
the detector responds function for a detector at x = 0. Inserting the thermal Green’s function in (6.18)
we find Z ∞
f (E) = d∆τ e−i E ∆τ Gβ (∆τ)
−∞
∞ (6.21)
−e−i E ∆τ
Z X
= d∆τ
−∞ n∈Z
4π2 (∆τ + inβ + iε)2

The integral can be analytically continued so as to be evaluated by residues along the negative imaginary
axis where it has double poles at t = −nβ i. Thus we obtain the sum of residues
X E
f (E) = E e−Enβ = , (6.22)
n∈Z
eβ E −1

in line with our expectation for a detector immersed in a thermal bath of temperature T = 1/β.

6.3 The Unruh effect

Having introduced these tools we are ready and well-equipped to study the Unruh effect. Here we
will perform the calculation for massless modes in 1 + 1 dimensions where we can use the conformal
invariance of the wave equation allowing us to explicit perform calculations. All the essential features
will already be present in this setup. The analysis can be performed explicitly also for massive modes
and be extended to higher dimensions, but only at the expense of having to deal with Bessel function
identities.3
Consider an observer Oa with constant acceleration a along the x-axis. The world-line for this
observer can be parametrised as

1
X (τ) = (t(τ), x(τ)) = (sinh aτ, cosh aτ) , (6.23)
a

where τ is the proper time of the observer. Note that these trajectories parametrise the hyperbolae,
x 2 − τ2 = a2 . We will be asking the question as to how the accelerating observer sees the Minkowski
3
Going to higher dimensions is no worse than introducing a mass, but we lose the conformal invariance that we exploited
in this section.

58
vacuum. To do so we want to find the natural coordinates in Minkowski space-time adapted to this
observer. I.e. we want the time coordinate to be its proper time, and the spatial coordinate to be
characterized by the fact that the observed is at rest in it. Using the clock and radar method, the
observer will set up coordinates (τ, ξ) related to inertial coordinates (t, x) by

eaξ
(t, x) = (sinh aτ, cosh aτ) , ds2 = dt 2 − dx 2 = e2aξ (dτ2 − dξ2 ) . (6.24)
a

Indeed, in these coordinates, the path followed by the observer is given by X (τ) = (τ, 0). The new
coordinates have range {τ, ξ} ∈ (−∞, ∞) but they only cover the part of Minkowski space with
R : {x > |t|}, called the Rindler wedge. This is the portion of space-time that the accelerating observer
can measure and see. Minkowski space equipped with this metric will be denoted Rindler space.
Note that Rindler space corresponds to the right wedge foliated by the world-lines of the accelerated
observers, labelled by R in Figure 6.1.

t
1,1
R
H+
R

H−

Figure 6.1: The Rindle wedge R in two-dimensional Minkowski space is denote by the orange
region. The uniformly accelerated observer Oa follows the blue hyperbolic trajectories in R.
H± denote the Killing horizons which are the boundaries of R1,1 perceivable for this observer.

The inertial light-cone coordinates (u, v) := (t − x, t + x) can be rewritten in terms of the light-cone
coordinates adapted to the accelerating observer (UR , VR ) := (τ − ξ, τ + ξ),

1
−e−aU , eaV , ds2 = dudv = ea(VR −UR ) dUR dVR .


(u, v) = (6.25)
a

The Rindler wedge is therefore the whole plane in the (UR , VR ) coordinates but only the quadrant
with −u, v > 0 in the (u, v) coordinates.
More generally, we see that the lines of constant ξ in the Rindler metric describe uniformly accelerated
observers with acceleration
α = ae−aξ , (6.26)

59
and proper time τ. Therefore, Rindler space can be regarded as a foliation of Minkowski space by the
trajectories of uniformly accelerated observers. Near the horizon where ξ → −∞, we have α → ∞
such that the observers feel an infinite proper acceleration.
Similarly, we can also cover the left wedge of Minkowski space, x < |t|, by defining the coordinates

eaξ eaξ
t =− sinh aξ , x =− cosh aξ . (6.27)
a a

Notice that the events happening in the left Rindler wedge are causally disconnected from the world-
lines of a Rindler observer in the right Rindler wedge, and the line u = 0 effectively behaves as an
event horizon. This observation will be relevant in the context of our discussion of black holes in the
next chapter.
Let us now consider the quantisation of a massless scalar field in Rindler space. Since the wave
equation in this case is conformal, we can trivially solve it by applying a conformal transformation to
the standard Minkowski space modes. In standard Minkowski space, we have respectively the left
and right moving modes4

1 1
φω (u) = p e−iωu , φ̃ω (v) = p e−iωv , (6.28)
4πω 4πω

which constitute the positive frequency modes for ω > 0. In the adapted coordinates introduced
above the metric is conformal to the standard Minkowski metric so the modes seen/measured by Oa
of frequency ω will be respectively

θ (−u) −iωUR θ (v) −iωVR

ΦRω (UR ) = p e , Φ̃Rω (VR ) = p e . (6.29)
4πω 4πω

where we added the Heaviside functions to denote that these modes are only non-zero in the right
Rindler wedge.
Similarly, considering the left Rindler wedge L with lightcone coordinates U L and VL we find the
modes for the quantum field there as,

θ (u) iωU L θ (−v) iωVL

L
Φω (U L ) = p e , L
Φ̃ω (VL ) = p e . (6.30)
4πω 4πω

Given that the left- and right-moving modes decouple, we can focus on the right-moving modes while
the results for left-moving modes will follow identically.5 In the standard Minkowski picture, we can
define a general right-moving field operator as
Z ∞
dω
φ̂ = p [e−iωu aω + eiωu aω
†
], (6.31)
0 4πω

while for an accelerating observer in the right Rindler wedge the right-moving field operators are
4
In two dimensions we have |k| = ω where k only has one component. The right-moving waves have k > 0 while the
left-moving ones have k < 0.
5
The left and right movers can never mix under Bogoliubov transformations, since ΦR/L
ω
(UR/L ) only depends on UR/L
while Φ e R/L (VR/L ) only depends on VR/L .
ω

60
defined as Z ∞
dλ
Φ̂R = p [e−iλUR ARλ + eiλUR AR†
λ
], (6.32)
0 4πλ
where aω and ARλ are the standard raising and lowering operators satisfying the commutation relations

†
[aω , aω ′
′ ] = δ(ω − ω ) , [Aλ , A†λ′ ] = δ(λ − λ′ ) , (6.33)

and analogous for the left Rindler wedge. Implicit in these definitions are the Minkowski vacuum
|0 M 〉, satisfying aω |0 M 〉 = 0, and the Rindler vacuum satisfying Aλ |0R 〉 = 0 for respectively all ω > 0
and λ > 0.
As by now standard, we can build a Fock space FR/L on the Rindler vacua 0R/L using the respective
R/L† R/L
creation and annihilation operatorsAλ and Aλ . These Fock spaces are based on the Rindler modes
(6.29) or (6.30) measured by an observer OaR/L respectively in the right or left Rindler wedge. In
particular the vacua |0〉R/L is the state in which OaR/L sees no particles. However, As it stands, there
can be no identification between the Minkowski Fock space F M and either of the Rindler Fock space
FR/L separately as they only determine the Minkowski fields respectively for u < 0 or u > 0 and
are not defined in the opposite patch. To determine φ̂ and F M from Rindler type data, we have to
consider both the left and right Fock spaces FR and F L such that the Minkowski Fock space is now a
tensor product
F M = F L ⊗ FR . (6.34)

Note however, that the Minkowski vacuum |0〉 M might be an entangled state in this product, i.e.
|0〉 M ̸= |0〉 L ⊗ |0〉R .
Having discussed in detail the various vacua in the picture we now wish to compute the distribution
of the number of particles of frequency λ detected by the observer Oa in the Minkowski vacuum. To
do so we need to compute the Bogoliubov coefficients relating the Minkowski and Rindler modes in
the right Rindler wedge,
Z
 
θ )(−u)φω (u) = dλ αRωλ ΦRλ (UR ) + βωλ
R
ΦR∗
λ (U R ) . (6.35)

Inserting the modes in the definition for the Bogoliubov coefficients, (4.38), 6 we find,
Z ∞
↔
αRωλ =i dUR ΦR∗
λ ∂u ξω
−∞
Z0
1 iλ
= p duλe−iωu (−au)− a −1 (6.36)
2π ωλ −∞
v iλ
1 t λ  a − a iλ πλ
 ‹
= Γ − e 2a .
2πa ω ω a
6
where the inner product is defined on the hypersurface ξ = constant with normal vector nµ = e−aξ (1, 0)

61
Similarly, for βωλ
R
we find
Z ∞
↔
R
βωλ =i dUR ΦRλ ∂u ξω
−∞
v (6.37)
iλ
1 tλ  a a iλ − πλ
 ‹
= Γ e 2a .
2πa ω ω a
The verification of the intermediate steps are left as an exercise for the reader but mainly consist of
rewriting the integral in terms of an integral representation of the Gamma function and using some
Gamma function identities.
The main takeaway from this calculation is the relation

|αωλ |2 = e2πλ/a |βωλ |2 , (6.38)

as this allows us to compute the expectation value of the number operator Nλ of the modes with
frequency λ detected by Oa in the Minkowski vacuum,

〈Nλ 〉 M := 〈0| M AR† AR |0 〉

λ λ M
Z∞
= dω|βωλ |2 (6.39)
0

This can be simplified by using the normalization condition for Bogoliubov coefficients which in our
context reads Z∞
dω(αωλ ᾱωλ′ − βωλ β̄ωλ′ ) = δ(λ − λ′ ) . (6.40)
0

Evaluating at λ = λ′ , we reinterpret the right hand side δ(0) = V as the volume of space, regularized
as usual by putting the system in a finite box. This allows us to deduce the particle number density as

〈Nλ 〉 M 1
= . (6.41)
V e2πλ/a −1

This is the main result of this section as we now recognize this as the Bose Einstein distribution with
Unruh temperature
a
TUnruh = . (6.42)
2π
We conclude that an observer moving with uniform acceleration through the Minkowski vacuum
a
observes a thermal spectrum of particles. The Unruh temperature T = 2π is the temperature that
would be measured by an observer moving along the path ξ = 0, which feels the acceleration α = a.
Any other path with ξ = constant feels an acceleration

α = ae−aξ , (6.43)

α
and will thus measure thermal radiation at temperature T = 2π . As ξ → ∞, the temperature
approaches 0, in line with the fact that near ∞ the Rindler observer is nearly inertial. We conclude
that not only does the choice of vacuum, and hence concept of particle become time-dependent, it is
also observer-dependent, even in flat space-time.

62
Coming back to the relation between the Minkowski and Rindler vacua we have that the Unruh state
ρU as measured by Oa is given by the density matrix

ρU = Tr F L |0〉 M 〈0| M . (6.44)

A key point to analyse this state is that time translation ∂τ for Oa in R is given by the boost Killing
vector on R but in L it is given by minus the boost Killing vector

∂ ∂
B := x∂ t + t∂ x = v∂ v − u∂u = =− (6.45)
∂ τR ∂ τL

However, since the Minkowski vacuum is Lorentz and hence boost invariant we must have an entangled
product of the form X
|0〉 M = f n |n〉 L ⊗ |n〉R , (6.46)
n

for some f n . Indeed, our calculations show that f n = e−β En , where β = 1/TUnruh .
Alternatively we can recognize the thermal nature of the Rindler vacuum by looking at the KMS
condition on the Wightman function in the Rindler vacuum. Since we are working in a conformal setup
we can straightforwardly obtain the Rindler space Wightman function from the Minkowski Wightman
function. Indeed, in two dimensions it is unchanged so we simply have to change coordinates to
obtain the Wightman function in Rindler space

1
G+M (x, x ′ ) = 〈0| M φ̂(x)φ̂(x ′ ) |0〉 M =
− (x 0
− iε)2 − (x − x′ )2
x ′0
−a2
= 2aξ
e + e2aξ + a2 ε2 − 2ea(ξ+ξ ) cosh a(τ − τ′ ) + 2aiε (eaξ sinh aτ + eaξ′ sinh aτ′ ) (6.47)
′ ′

= 〈0|R φ̂(x)φ̂(x ′ ) |0〉R

=G+R (x, x ′ ) .

From this expression is is clear that in Rindler coordinates, G+R is periodic in complex τ with period
β = 2π
a . Thus, for Oa , the Minkowski propagator is a thermal Green’s function of temperature TUnruh .

As before, we can make the previous remark more explicit by restricting to the accelerating world-line
x = x(τ), x ′ = x(τ = 0) and introducing a particle detector. Restricting to the world-line we obtain
the Green’s function

a2
G+R (x(τ), x(0)) =
(sinh aτ − iaε)2 − (cosh aτ − 1)2
(6.48)
−a2
= .
2(1 − cosh aτ + 2iaε sinh aτ + a2 ε2 /2)

With this expression at hand we can compute the detector response function f (E) characterising the

63
detection of field transitions at energy E as
Z ∞
f (E) = d∆τ e−iE∆τ G+R (∆τ)
−∞
∞ (6.49)
−a2 e−iE∆τ
Z
= d∆τ ,
−∞ 2(1 − cosh aτ + iaε sinh aτ)

The integral can be analytically continued so as to be evaluated by residues along the negative
imaginary axis where it has double poles at aτ = 2nπi. We obtain the sum of residues
∞
X 2πEn E
f (E) = E e− a = 2πE
, (6.50)
n=0 e a −1

as expected for a detector immersed in a thermal bath at the Unruh temperature TUnruh .
Before moving on let us introduce an alternative way to detect the thermal nature of space-times.
In terms of the Euclidean continuation of the space-time, it turns out thermal effects can be seen
as the need to periodically identify the imaginary time coordinate [GH77a, GH94]. It is easy to see
that the inverse is true. In a space-time which is periodic in imaginary time on can compute the
Euclidean Green’s function which consequentially will also be periodic. After analytically continuing
to Lorentzian signature, the Euclidean Green’s function becomes the Feynman Green’s function which
naturally inherits the complex periodicity of its Euclidean counterpart.
As an example, consider the the analytic continuation of the Rindler wedge metric,

ds2 = −a2 (dρ 2 + ρ 2 dθ 2 ) , (6.51)

where we defined the coordinate ρ = eaξ and θ = iτ a . This is usual metric on flat Euclidean space
but generically, it has a conical singularity at ρ = 0. In order to avoid this singularity we need to
periodically identify θ with period 2π. This is essential for regularity at the horizon and gives rise
to the imaginary periodicity τ ∼ τ + iβ for β = 1/TUnruh . This is a theme that can be taken much
further in curved space-times where similar consideration prove very useful in studying black hole
backgrounds.

Remark. Let us finish this chapter with some apparent paradoxes, and their resolutions, in Rindler
space. First, note that a Rindler observer with smaller constant ξ coordinate are accelerating faster to
keep up. This may seem surprising because in Newtonian physics, observers who maintain constant
relative distance must share the same acceleration. In relativistic physics, this is no longer true and
we see that the trailing endpoint of a rod which is accelerated by some external force (parallel to
its symmetry axis) must accelerate a bit faster than the leading endpoint, or else it must ultimately
break. This is a manifestation of Lorentz contraction. As the rod accelerates, its velocity increases
and its length decreases. Since it is getting shorter, the back end must accelerate harder than the
front. Another way to look at it is: the back end must achieve the same change in velocity in a shorter
period of time. This leads to a differential equation showing that, at some distance, the acceleration
of the trailing end diverges, resulting in the Rindler horizon. This phenomenon is the basis of a

64
well known "paradox", Bell’s spaceship paradox. However, it is a simple consequence of relativistic
kinematics. One way to see this is to observe that the magnitude of the acceleration vector is just the
path curvature of the corresponding world line. But the world lines of our Rindler observers are the
analogues of a family of concentric circles in the Euclidean plane, so we are simply dealing with the
Lorentzian analogue of a fact familiar to speed skaters: in a family of concentric circles, inner circles
must bend faster (per unit arc length) than the outer ones.
The main observation of this chapter was that an accelerated observer detects particles in the
Minkowski vacuum state. An inertial observer would say that the same state is completely empty, the
expectation value of the energy momentum tensor Tµν M = 0. If there is no energy momentum how
can the Rindler observer detect particles? If the Rindler observer is to detect background particles,
they must carry a detector. This must be coupled to the particle being detected. However, if a detector
is being maintained at constant acceleration, energy is not conserved. From the point of view of
the Minkowski observer the Rindler detector emits as well as absorbs particles, once the coupling
is introduced the possibility of emission is unavoidable. When the detector registers a particle the
inertial observer would say that it had emitted a particle and felt a radiation-reaction force in response.
Ultimately the energy needed to excite the Rindler detector does not come from the background
energy momentum tensor but from the energy we put into the detector to keep it accelerating.

65
 Chapter 7

Hawking radiation

The creation of particles by black holes is necessary for maintaining the second law of thermodynamics
in their presence. This process of radiation and evaporation of black holes is an important facet in
the fundamental search for a microscopic explanation of the entropy of black holes; a search which
appears to be leading to new and exciting physics connecting gravitation and quantum theory. In this
chapter we will explore the effect of Hawking radiation and introduce some of the challenges this
phenomenon generates.

7.1 Quantum fields in a black hole background

The goal of this section is to explore quantum fields in a black hole background. The simplest, and
prototypical example of such a background is the Schwarzschild background, with metric,

2M 2M −1 2
 ‹  ‹
ds2 = 1 − dt 2 − 1 − dr − r 2 dΩ2 . (7.1)
r r

Black holes, such as the Schwarzschild black hole and its rotation or charged cousins were discussed
at length in the course general relativity 2. In Appendix C we collect all the necessary background
information for these notes to be self-contained but we refer the reader to the lecture notes of general
relativity 2 for more information and references.
One might be surprised that anything interesting can happen since the Schwarzschild black hole is
a static space-time. Surely one can use the Schwarzschild time-like Killing vector (at least at large
distances) to define positive and negative frequency and proceed with the quantisation just like in
Minkowski space. The point of this chapter is to show that interesting things do happen! We will do
so in steps which will become increasingly realistic at the expense of a loss of explicit solvability.

(1 + 1)-dimensional toy model

We start our exploration with a massless scalar field in a two-dimensional "black hole" background
with the same time-radial part of the metric as the Schwarzschild black hole

2M 2M −1 2
 ‹  ‹
2 2
ds = 1 − dt − 1 − dr
r r
2M
 ‹
= 1− (dt 2 − dr∗2 ) (7.2)
r
2M − r
= e 2M dUdV .
r

66
 r


where we introduced the tortoise coordinate r∗ = r + 2M log 2M − 1 and the Kruskal-Szekeres and
Eddington-Finkelstein null coordinates U, V and u, v are defined as,
u v
U = −4M e− 4M , V = 4M e− 4M , u = t + r∗ , v = t − r ∗, . (7.3)

See Appendix C for more details on the relevant coordinates and their properties. Note that this
(1 + 1)-dimensional model merely serves to illustrate some properties of the (3 + 1)-dimensional black
hole and should not be taken seriously on its own. In itself it is not even a solution to the (vacuum)
Einstein equation.
A first hint that this picture has something to do with the Unruh effect can already be seen from the
coordinate change (7.3), which is identical to the transformation between null coordinates in Rindler
and Minkowski space upon substituting a = (4M )−1 . Indeed, the problem is very similar to the one in
Rindler space. The two coordinate systems we consider, Eddington-Finkelstein and Kruskal-Szekeres
are respectively very similar to the Rindler and Minkowski coordinates.
Consider for simplicity the massless minimally coupled scalar. We could introduce mass and a coupling
to the Ricci scalar but note that in the four-dimensional the Ricci scalar vanishes so we want consider it.
The addition of mass breaks conformality so for the sake of keeping things simple we will not include
it here but comment on it later. The Eddington-Finkelstein coordinates are adapted to an observer
sitting very far from the black hole, where the metric approaches Minkowski space ds2 → dudv. In
these coordinates it’s straightforward to solve the wave equation and find a complete set of incoming
and outgoing modes
1 1
ψω − p e−iωu , φ̃ω = p e−iωv . (7.4)
4πω 4πω
To these modes we can associate a vacuum called the Boulware vacuum which is defined by

bω |0〉B = 0 . (7.5)

The Boulware vacuum contains no particles from the point of view of a distant observer. However,
since the Eddington-Finkelstein coordinates do not cover the whole of space-time, similar to the
Rindler coordinates, this vacuum is similar to the Rindler vacuum of an accelerated observer.
Similarly, in Kruskal-Szekeres coordinates we can solve the wave equation finding the following set
of positive frequency, incoming and outgoing, modes,

1 1
ξω (U) = p e−iωU , ξ̃ω (V ) = p e−iωV . (7.6)
4πω 4πω

The corresponding Kruskal vacuum is defined as,

aω |0〉K = 0 . (7.7)

In Kruskal-Szekeres coordinates, the metric near the black hole horizon approaches ds2 → dUdV , so
the Kruskal vacuum is the appropriate one for an observer sitting next to the black hole horizon. Since
Kruskal–Szekeres coordinates cover the whole of space-time, they are the analogue of the Minkowski
vacuum that we studied in the quantization of a scalar field in Rindler space.

67
We can now ask the following question: if a Kruskal observer is in the vacuum state, what does the
Boulware observer see? Since the relation between both systems is the same as we found before in
the case of the Unruh effect, the calculation of the Bogoliubov coefficients will be the identical as the
one for the Unruh effect. The only difference lies in replacing the acceleration a by the surface gravity
1
a → κ = 4M . We conclude that the Boulware observer sees a thermal spectrum with temperature

κ 1
TH = = . (7.8)
2π 8πM

Far from the black hole, the factor g00 in Eddington–Filkenstein coordinates goes to 1, and T0 = T ,
so (7.8) is the physical temperature observed by an observer at infinity.

(3 + 1)-dimensional Schwarzschild

The two-dimensional toy model discussed above was extremely simple but included all the necessary
ingredients to observe the Hawking temperature. However, we are really interested in the four-
dimensional Schwarzschild black hole. In this case we lose conformality and we will not be able to
exactly solve the problem. However, using some approximations we will still be able to come to a
similar conclusion as in the toy model above.
Let us consider again a massless scalar field but now in the full Schwarzschild background (7.1).
As in Schwarzschild, and similarly in the Kerr black hole we have R = 0 we can safely ignore the
coupling to the Ricci scalar.1 As the problem is entirely symmetric on the two-sphere it will prove
useful to decompose our field in spherical harmonics,

φ = f lm (t, r)Ylm (θ , φ) , (7.9)

where Yl m (θ , φ) are the spherical harmonics. Substituting this expansion in the wave equation results
in
□(4) φ = 0 ⇒ □(2) + Vl (r) f lm (r, t) ,


(7.10)

where □(d+1) denotes the (d + 1)-dimensional Laplacian on respectively the full Schwarzschild
space-time or the 2d time-radial slice considered in the above. The potential Vl (r) is given by

2M 2M l(l + 1)
 ‹ ‹
Vl (r) = 1 − + . (7.11)
r r3 r2

So we see that the massless scalar in (3 + 1) dimensions decomposes in infinitely many massless
scalars in (1 + 1) dimensions in the presence of a potential. The only change from the story above
is therefore that a wave escaping the black hole needs to propagate through the potential barrier
caused by Vl (r). Even though we cannot solve this problem analytically, note that the potential falls
off exponentially in r ∗ as r ∗ → −∞, i.e. when one approaches the horizon, and falls off polynomially
as r → ∞. For this reason we can use the same asymptotic states as above. Hence, the only effect
of the potential is that it decreases the intensity of the wave and changes the resulting spectrum of
1
In the collapse picture below, in the collapsing phase the Ricci scalar might be non-vanishing. However, this will not
change the late time spectrum and so we will keep ignoring this coupling.

68
emitted particles by a greybody factor 0 < Γl (ω) < 1,

Γl (ω)
〈nω 〉 = ω . (7.12)
e TH − 1

The greybody factor is entirely due to the potential outside the black hole horizon. It is clear that this
factor is not directly related to the quantum origin of the Hawking radiation and therefore the basic
features of the derivation above survive without significant alterations. This result can be generalized
for the case of a massive scalar field, and also for vector and spinor fields. The conclusion is that the
black hole must emit all possible species of particles, each having the Hawking thermal spectrum
corrected by the corresponding greybody factor.

Vl (r)
l =0

l =1

l =2

r
2M

Figure 7.1: The effective potential Vl (r) experienced by the spherically symmetric modes f lm
for the values of l = 0, 1, 2.

Black hole formed through collapse

The eternal black hole described above is rather unphysical and we don’t expect to see the full Kruskal
extension. A more physical picture would be to consider a ball of spherically symmetric dust collapsing
to form a black hole. The Penrose diagram for this space-time is given in Figure 7.2.
We can now proceed as before and quantise the massless scalar field in this background. The past
null hypersurface I− is a Cauchy hypersurface, hence we can quantise the scalar field using this
hypersurface and write Z
† ∗


φ = dω aω fω + aω fω , (7.13)

where the fω are a complete set of orthonormal solutions to the wave equation with associated
†
annihilation and creation operators aω and aω . Far outside the collapsing body at early times, the
definition of physical particles that would be detected by inertial observers, or equivalently of positive
frequency solutions of the wave equation, is unambiguous. We choose the fω such that they form a
complete set of incoming positive frequency solutions of energy ω. Their asymptotic form on past
null infinity is
1
fω ∼ p e−iωv Ylm (θ , φ) , 〈 fω , fω′ 〉 = δ(ω − ω′ ) , (7.14)
16π ω
3

where we suppress the discrete quantum numbers l and m in labelling the functions fω .

69
 Singularity

u(v)
H

I+

I−
v0
v

Figure 7.2: The Penrose diagram for collapse to the Schwarzschild black hole. The singularity
is located on top and shielded by the horizon H in orange. The collapsing cloud of dust is
pictured in blue. Once the cloud enters the horizon, a black hole is formed. The incoming ray
with v = v0 is the last one that reaches the centre of the collapsing body and makes it to I + .
Rays with v < v0 fall into the black hole.

At late times on the other hand, we know that I + is not a Cauchy hypersurface. Instead we have to
consider boundary data both at future null infinity and the event horizon H. On I + , just like on I − ,
∗
the definition of positive frequency modes is unambiguous and we can find a complete set {pω , pω }
+ +
of orthonormal solutions on I . The asymptotic form of these functions on I is

1
pω ∼ p e−iωu Ylm (θ , φ) , (7.15)
16π ω
3

where u is the outgoing null coordinate at I + . A general solution, incoming from the past, will also
have a part that is incoming at the event horizon. Therefore we must introduce a second complete
basis of orthonormal functions qω on the horizon which have zero Cauchy data on I + . Since the
functions pω and qω are supported in disjoint regions at late times, their (conserved) scalar product
must vanish 〈qω , pω′ 〉 = 0 and similarly for their complex conjugates. For this reason the precise form
of the functions qω will not affect observations on I + . The details are therefore not important since
we will trace over the modes at the horizon. We can thus expand the field φ in the entire space-time

70
as
† ∗ † ∗


φ = dω bω pω + cω qω + bω pω + cω qω , (7.16)

with bω and cω the annihilation operators for outgoing particles at late times. The vacuum at I+
defined by bω |0〉B is the Boulware vacua as defined before, while the vacuum at past null infinity
|0〉− , defined by aω |0〉− will take up the role of the Kruskal vacuum. The task we have to do is then
clear, we want to compute the number density of particles observed by a Boulware observer in the
"Kruskal" vacuum. Although conceptually clear the computation is rather involved and some details
will be left to fill in by the reader.
To compute the density of emitted particles we have to compute the Bogoliubov coefficients

αωω′ = 〈 fω′ , pω 〉 , βωω′ = − fω∗ ′ , pω . (7.17)

To determine these coefficients we need to trace back in time the function pω along an outgoing
geodesic have a large value of u, close to the horizon. Such a geodesic is illustrated in Figure 7.2 as
the red line and passes through the center of the collapsing cloud just before the event horizon is
formed and emerges as an incoming geodesic characterised by a value of v close to v0 . The value of u
depending at I + depending on v can be computed by analysing the null geodesics in this space-time,
see Appendix C, and is given by v − v
0
u(v) = −4M log , (7.18)
K
where K is some positive constant. Inserting this expression into expressions (7.15) for the functions
pω we can compute the Bogoliubov coefficients as

v0 ‹1
ω′ 2 iω′ v−iωu(v)
Z 
αωω′ =C dv e ,
−∞ ω
Z v0  ′ ‹ 12 (7.19)
ω ′
αωω′ =C dv e−iω v−iωu(v) ,
−∞ ω

where C is a constant. Substituting s = v0 − v = iz we can compute

0 ‹ 12
ω′
Z 
′
 s
αωω′ = −C ds e−iω (s−v0 ) exp 4iωM log
∞ ω K
Z0  ′ ‹ 12
ω iz
 ‹
iω′ v0 ω′ z
= − iCe dz e exp 4iωM log (7.20)
−∞ ω K
 ′ ‹ 12 Z 0
iω′ v0 2πωM ω |z|
 ‹
ω′ z
= − iCe e dz e exp 4iωM log ,
ω −∞ K

and similarly,

‹ 12 Z 0
ω′ |z|
  ‹
−iω′ v0 −2πωM ′
βωω′ = iCe e dz eω z exp 4iωM log . (7.21)
ω −∞ K

We immediately find that

|αωω′ |2 = e8πM ω |βωω′ |2 , (7.22)

71
for the part of the wave packet that was propagated back in time through the collapsing body just
before it formed a black hole.
For the components pω of this part of the wave packet, we have the scalar product,

〈pω , pω′ 〉 = Γ (ω)δ(ω − ω′ ) , (7.23)

where Γ (ω) is the fraction of the wave packet that would propagate back in time through the collapsing
body. Indeed, we can divide the functions pω in two parts,

(1) (2)
pω = pω + pω . (7.24)

(1)
The part pω propagates backwards in time outside of the collapsing body and reaches I− at some value
v > v0 . This part of the wave will interact minimally with the collapsing matter and consequentially
the frequency will not change significantly from I− to I+ . For this reason we can ignore this part of
(1) (2)
the wave when asking questions about particle production. Indeed, since pω and pω have disjoint
(1)
support on I− (resp. v > v0 and v < v0 ), they do not interact and we can safely ignore the parts pω .
Their only effect is the introduction of the function Γ . From the normalisation condition (7.23) we
therefore find Z
Γ (ω)δ(ω − ω′ ) = dλ α∗ωλ αω′ λ − βωλ ∗


βω′ λ (7.25)

(2)
where now the Bogoliubov coefficients refer to the coefficients in the expansion of pω only.
As before, this allows us to compute the density of emitted particles as
Z
(2)† (2)
〈Nω 〉− = †
〈0|K bω bω |0〉− ≃ 〈0|− bω bω |0〉− = dω′ |βωω′ |2 . (7.26)

The resulting integral is again divergent but can be regularised by putting the system in a box and
computing the density of emitted particles instead,

1 Γ (ω)
nω = 〈Nω 〉− = . (7.27)
V 2π (e8πM ω − 1)

Hence after this long computation we come to exactly the same conclusion as before and find that
the collapsing black hole emits and absorbs radiation exactly like a gray body of absorptivity Γlm (ω)
and Hawking temperature TH = (8πM )−1 !
For large black holes this temperature TH ∼ 6 × 10−8 MM⊙ K is extremely small for large black holes with
M ≫ M⊙ , where M⊙ is the mass of the sun. For this reason our assumption that the background does
not back-react against this radiation seems to be justified. For small black holes on the back-reaction
cannot be ignored and a more sophisticated treatment is needed.
Indeed, from energy conservation one can estimate the rate of loss of mass. Stefan’s law for the
evaporation of a black body states that

dE
∝ −AT 4 , (7.28)
dt

72
where A is the area. With E ∝ M 2 and T ∝ M −1 this leads to the rate of mass loss to be proportional
to
dM c
≃− 2, (7.29)
dt M
where c is a positive constant that depends on the number and type of quantised matter fields
that couple to gravity. From this expression we it becomes indeed apparent that for large black
holes M ≫ dMdt justifying our assumption of ignoring back-reaction. For reference, this leads to the
€ Š3
black-hole evaporating in a finite time of the order of 1071 MM⊙ seconds.
Finally, before moving on, note that a static observer O at finite radius r measures a blue-shifted
T
temperature TO = |g H | . As r → ∞ this approaches the Hawking temperature, but it diverges at the
00
horizon where |g00 | → 0 due to the infinite acceleration of the static observer at the horizon. This is
precisely the Unruh effect we observed in the previous section. A freely falling observer however sees
no divergence as they cross the horizon.

7.2 The Hawking thermal state and friends

In analogy with the Rindler case, we can easily observe the thermal nature of the |0〉− vacuum for a
Boulware observer. The observer in the Boulware vacuum |0〉B has access to a Fock space FB built
†
by acting on |0〉B with the creation operators bω . As in the Rindler case however, and from the fact
+
that I is not a Cauchy surface, we know that this is not enough to construct the full Fock space as
seen by an observer at past null infinity. Indeed, the full Fock space is obtained as the tensor product
F− = F B ⊗ F H .

As before for Rindler, the late time vacuum is a complicated state of the form
X
|0〉− ∝ f n |n〉H |n〉B , (7.30)
n

Considering the associated density matrices, we construct the Boulware vacuum by tracing over the
horizon modes X
ρB = Tr FH ρ− = Tr FH |0〉− 〈0|− ∝ e−nπΩ/κ |n〉B B 〈n| (7.31)
n

which gives the desired thermal state.

The Hawking state was what arose from an essentially Minkowskian vacuum at I − in the collapsing
scenario, but other states are natural for the eternal Schwarzschild black hole where we start from
the Kruskal vacuum defined near the horizon. In this case the Cauchy surface in the far past consists
of two components, I− ∪ H− , as can be seen from the Penrose diagram in Figure E.1. Having said so,
it becomes clear that there are various ’natural’ choices for the vacuum in the past depending on what
we define as positive frequency states. The options are summarised in Table 7.1 below for positive
frequencies ω > 0.
The three options defined in this table each have a distinct physical interpretation.

• The Boulware vacuum corresponds to our familiar concept of an empty state defined far away
from the black hole and is defined with respect to a static observer. It is pathologic in the sense

73
 Vacuum Positive modes on H− Positive modes on I−
Boulware vacuum |B〉 e−iωu e−iωv
Unruh vacuum |U〉 e−iωU e−iωv
Hartle-Hawking vacuum H 2 e−iωU e−iωV

Table 7.1: The three natural vacua in the eternal Schwarzschild black hole. The options differ
by the choice of positive frequency modes in the two component of the far past Cauchy surface
H ∪ I− .

that the expectation value of the stress tensor diverges at the horizon. This is similar to the
Rindler vacua becoming singular at the Killing horizon seen in the previous chapter.

• The Unruh vacuum is regular on the future horizon, but not on the past horizon. At infinity, this
vacuum corresponds to an outgoing flux of blackbody radiation at the black hole temperature.
The black hole collapse studied in the previous section brings about the Unruh state.

• The Hartle-Hawking vacuum does not correspond to our usual notion of a vacuum. It is well-
behaved both on the future and past horizon but the price we have to pay for this is that the
state is not empty at infinity, but instead corresponds to a thermal distribution of quanta at
the Hawking temperature. That is, the Hartle-Hawking vacuum corresponds to a black hole in
(unstable) equilibrium with an infinite bath of blackbody radiation.

All these vacua are interesting in their own right and have been studied for a variety of reasons.
However, from the point of view of the ’physical’ collapse picture described above, it seems that the
Unruh vacuum best approximates the state obtained following the gravitational collapse of a massive
cloud of dust.
There are various ways to investigate the thermal nature of the various vacua above. As mentioned
above, the Hartle-Hawking state is a thermal state both at I± . Moreover, it is an example of a Thermal
Green’s function
2
G H (x, x ′ ) := H 2 φ̂(x)φ̂(x ′ ) H 2 = Gβ (x, x ′ ) . (7.32)

We will not attempt to explicitly compute the Green’s function (see for example [CJ86] for the result)
but a key statement is that it analytically extends to complex time and is periodic in imaginary time
with period,
β = 1/TH = 8πM , (7.33)

An alternative, and easier way to recognize the thermal nature of black holes is to study the Euc-
lideanised background. Wick rotating t → iτ, we find the positive definite background with metric2

2M 2M −1 2
 ‹  ‹
2 2
ds = 1 − dτ + 1 − dr + r 2 dΩ2 . (7.34)
r r
2M r
2
It’s even easier to see this in Kruskal coordinates, where the metric is given by ds2 = r e− 2M dUdV . Remembering that
r∗ iτ
U = −4M e 4M e− 4M it is equally clear that unless τ has period 8πM this metric has a conical singularity at the origin.

74
Substituting r = 2M + ε and expanding in small ε we find

2M 2 ε
ds2 ≈ dε + dτ2 + 4M 2 dΩ2 . (7.35)
ε 2M
p
At ε = 0 we see that the angular part nicely factorises out. Changing coordinates to ρ = 8M ε the
metric becomes
ρ2
ds2 ≈ dρ 2 + dτ2 + 4M 2 dΩ2 . (7.36)
16M 2
Hence we clearly see that in order to avoid a conical singularity at the origin we need to impose the
periodicity τ ∼ τ + 8πM . This gives alternative evidence for the Hawking temperature, analogous to
the Euclideanisation argument for the Unruh temperature in the previous section. In Figure 7.3 we
sketch the topology of the various Euclideanisations discussed so far. This property is an important red
herring for the presence of thermal states and remains valid much more general in various dimensions
with various types of matter content. The Euclideanisation of a black hole has the characteristic
topology of a cigar.

Figure 7.3: From left to right, the Euclideanised Minkowski space R×S 1 , Rindler space R2 and
(r, t) plane of the Schwarzschild black hole. The cigar topology background is characteristic
for black hole backgrounds.

7.3 Black hole thermodynamics

Prior to the discovery of Hawking radiation of black holes Bekenstein already conjectured that black
holes must have a non-vanishing intrinsic entropy [Bek73]. He came to this conclusion through the
following thought experiment. Consider a black hole that absorbs matter with non-zero entropy. If
the black hole entropy were vanishing then the total entropy in the system would decrease, violating
the second law of thermodynamics. Based on this reasoning Bekenstein concluded that the second
law can only be preserved if a black hole has an intrinsic entropy SBH proportional to its surface area.
However, the proportionality constant could not be fixed until the discovery of Hawking radiation.
Differentiating the expression for the surface area A = 16πM 2 , we find

1 A
dM = d . (7.37)
8πM 4

75
Recognising the coefficient on the left hand side as the Hawking temperature this looks precisely like
the first law of thermodynamics
dE = T dS , (7.38)

Following this analogy we conclude that the black hole (or Bekenstein-Hawking) entropy must be
equal to
A
SBH = = 4πM 2 . (7.39)
4
In line with its thermodynamic counterpart, the first law of black hole thermodynamics can be
generalised to closed systems with rotation and charge as follows,

dE = T dS + ΩdJ + ΦdQ , (7.40)

where we interpret Φ as the electric potential at the horizon and Q the total charge. Similarly, J is the
angular momentum and Ω the angular velocity.

Exercise 7.1. Consider the Reissner-Nordstrom solution

∆(r) 2 r2
ds2 = dt − dr 2 − r 2 dΩ2 , ∆(r) = r 2 − 2M r + Q2 , (7.41)
r2 ∆(r)

This is a solution to the Einstein-Maxwell equations with electromagnetic potential

Q
A= dt . (7.42)
r
Assuming Q < M , state the second law of thermodynamics by differentiating the area as a function of
mass and show that the coefficient is indeed equal to the Hawking temperature. (Hint: the Hawking
temperature has the same expression as for Schwarzschild when expressed in terms of the surface gravity.)

Exercise 7.2. If you are feeling courageous, repeat the previous exercise for the Kerr black hole.

The entropy of astrophysical black holes is extremely large, for a solar mass black hole for example one
⊙
finds SBH ∼ 1076 . Interpreting this as a statistical entropy implies that a quantum mechanical black
hole has an enormous number of microstates corresponding to the unique classical black hole. Finding
a microscopic derivation of this entropy is an active area of modern research. In asymptotically flat
space-time such a derivation has been given through string theory [SV96] but in asymptotically AdS
or dS space-times this remains an open question.
Taking into account the entropy of a black hole, we can state the generalised second law of thermody-
namics as follows.
δStotal = δSmatter + δSBH ≥ 0 . (7.43)

I.e. the total entropy of all black holes and matter combined can never decrease. In classical general
relativity, one can prove that the combined area of all black hole horizons cannot decrease. This
applies not only to adiabatic processes but also to strongly out of equilibrium processes such as
collisions and mergers of black holes.
Ordinary thermodynamic systems can be in a stable equilibrium with an infinity heat reservoir.
However, this is not true for black holes because they have a negative heat capacity! In other words,

76
black holes get colder when they absorb energy. Indeed, with E(T ) = M = (8πT )−1 , we find

∂E 1
CBH = =− < 0. (7.44)
∂T 8πT 2

This means that a black hole surrounded by an infinite thermal bath at temperature T < TH will emit
radiation and become even hotter. The process of evaporation is not halted in an infinite thermal
reservoir with constant temperature. Similarly, putting a black hole in a bath with T > TH will make
the black hole colder! In either case no stable equilibrium is possible. Stable equilibrium is only
possible in a finite reservoir. In this case the radiation of the black hole changes the temperature of
the bath until both reach the same temperature.

7.4 The information paradox

In this chapter we have discussed quantum fields in a black hole background and discovered that black
holes have a temperature. But where precisely does this radiation come from? The answer, discovered
by Hawking, is that we must consider quantum processes, more precisely quantum fluctuations of
the vacuum. In the vacuum pairs of particles and antiparticles are continuously being created and
annihilated. Consider such fluctuations for electron-positron pairs. Suppose we apply a strong electric
field in a region which is pure vacuum. When an electron–positron pair is created, the electron gets
pulled one way by the field and the positron gets pulled the other way. Thus instead of annihilation of
the pair, we can get creation of real (instead of virtual) electrons and positrons which can be collected
on opposite ends of the vacuum region. Thus we get a current flowing through the space even though
there is no material medium filling the region where the electric field is applied. This is called the
‘Schwinger effect’.
A similar effect happens with the black hole, with the effect of the electric field now replaced by the
gravitational field. We do not have particles that are charged in opposite ways under gravity. But the
attraction of the black hole falls off with radius, so if one member of a particle–antiparticle pair is
just outside the horizon it can flow off to infinity, while if the other member of the pair is just inside
the horizon then it can get sucked into the hole. The particles flowing off to infinity represent the
‘Hawking radiation’ coming out of the black hole. Doing a detailed computation, one finds that the
rate of this radiation is given by (7.28). Thus we seem to have a very nice thermodynamic physics of
the black hole. The hole has entropy, energy, and temperature and radiates as a thermal body should.
So far, so good, but there is a deep problem arising out of the way in which this radiation is created by
the black hole [Haw76]. The radiation which emerges from the hole is not in a ‘pure quantum state’.
Instead, the emitted quanta are in a ‘mixed state’ with excitations which stay inside the hole. There is
nothing wrong with this in this by itself, but the problem comes at the next step. The black hole loses
mass because of the radiation and eventually disappears. Then the quanta in the radiation outside the
hole are left in a state that is ‘mixed’, but we cannot see anything that they are mixed with! Thus the
state of the system has become a ‘mixed’ state in a fundamental way. This does not happen in quantum
mechanics. If we start with a pure state |ψ〉 and evolve it by some Hamiltonian H to ψ′ = e−iH t |ψ〉
we obtain another pure state at the end. Mixed states arise in usual physics when we coarse-grain
over some variables and thereby discard some information about a system. This coarse-graining is

77
done for convenience, so that we can extract the gross behaviour of a system without keeping all its
fine details, and is a standard procedure in statistical mechanics. But there is always a ‘fine-grained’
description available with all information about the state, so that underlying the full system there is
always a pure state. With black holes we seem to be getting a loss of information in a fundamental
way. We are not throwing away information for convenience; rather we cannot get a pure state even
if we wanted.
To make this discussion a bit more quantitative, let us introduce the von Neumann entropy, which is
an extension of the Gibbs entropy from statistical mechanics to quantum statistical mechanics. For a
quantum system described by a density matrix ρ, the von Neumann entropy is defined as,

S = −Tr ρ log ρ , (7.45)

In a finite dimensional system we can always write the density matrix in a basis of eigenvectors |n〉 as
X
ρ= pn |n〉 〈n| , (7.46)
n

which makes it clear that for pure state the von Neumann entropy vanishes, while its maximal value
S = log dim H is reached for the maximally mixed state ρ = dim1 H |n〉 〈n|. Now, let us consider the
P

evaporation of a black hole à la Hawking. The black hole starts in a pure state, hence initially we
have S = 0. After some time part of the black hole has evaporated where the radiation is in a mixed
state. Hence, during the evaporation, the von Neumann entropy gradually increases until it reaches
its maximum when the black hole is fully evaporated into thermal radiation. See Figure 7.4 for a
graphical representation of the entropy as a function of time.
This paradox has been a guiding post for progress on quantum gravity since its discovery by Hawking
in 1975. Hawking initially advocated that in the presence of gravity we should change our ideas
about quantum mechanics and loosen the our demand of having purely unitary evolution. However,
this is a very unsettling proposal which opens a Pandora’s box of unwanted consequences and most
physicists are not willing to abandon ordinary quantum mechanics when it works so well in all other
contexts. Luckily, in the 90s and 2000s string theory provided various hints that information is not
lost! But how can it be that we need string theory for this? The gravitational interactions at the event
horizon for a large black hole are so incredibly small that we would expect that our semiclassical
intuition should be valid here.
Around 2020 a new perspective emerged in the papers [Pen20, AEMM19] and many papers after
that. In this paper an alternative semiclassical computation was performed that instead of Hawkings
entropy curve produces the Page curve. As can be seen from Figure 7.4 this curve descends back
to zero entropy at the end of the evaporation, therefore restoring unitarity! The fundamental idea
behind these computations is to introduce a new tool called the ’quantum extremal surface’ which
takes into account the microscopic structure of the black hole as well as the coupling with the external
fields. Performing the semiclassical computation using this surface, instead of the event horizon as
in Hawkings computation results in a different prediction for the entropy where at the Page time a
transition takes place after which the entropy starts to shrink, reproducing the Page curve. A full
discussion of their formalism would lead us beyond the scope of this course so we refer the reader to

78
 S

t Page t
BH R BH R BH R BH R

Figure 7.4: The red line represents the entropy of the radiation following Hawking’s calculation,
while the blue line is the page curve. The turning point at the page time occurs at the point in
time where the entropy of Hawking radiation is equal to the Bekenstein-Hawking entropy of
the black hole. The dots give a cartoon picture of the qubits of information transferred from
the black hole to the radiation.

the original literature.

This approach immediately brings us a whole range of new questions. Why are the equations for
various quantities modified by quantum gravity when a black hole is involved? And should they
then also be modified when studying the sun or Mercury? The key in answering this question turns
out to be complexity. Black holes are incredibly complex objects, they are maximally chaotic and
pack information in the densest possible way. This characteristic sets them apart from the other
astrophysical object where we find a similar curvature as at the event horizon of a black hole. The
computation of the quantum extremal surface turn out to crucially depend on complexity. It turns
out that usual semi-classical gravity is valid at low curvature and low complexity. However, at large
complexity our semi-classical intuition has to be modified in order to predict the correct physical
behaviour. Research in this direction continues until today and is an exciting area of new developments
in quantum gravity.

79
 Chapter 8

Quantum fields in de Sitter space

Our universe is expanding. Moreover, it is expanding at an accelerating pace. If this persists, we will
eventually head towards a cold and lonely world. All matter will drift apart, until we are left only
with our Milky Way, which by then will most likely have merged with Andromeda. After even longer
time scales, all cosmic radiation will have stretched to sizes beyond the cosmic horizon, leaving us
with a universe dominated by the non-diluting cosmological constant. We will end up in a de Sitter
universe.
We have observational evidence for two periods of exponential expansion. The first is the early
inflationary era, during which the universe dramatically expanded in the first moments after the big
bang while the second is the late time era dominated by the cosmological constant. This chapter will
deal with quantum fields in pure de Sitter space, while the next chapter will look at more general
cosmological backgrounds. For more details and references, see [Ann12, SSV01, Har17].

8.1 de Sitter basics

The de Sitter space is the maximally symmetric spacetime of positive curvature

d(d + 1)
R= , (8.1)
L2
where L is the characteristic de Sitter length scale. This space is a solution to the vacuum Einstein
d(d+1)
equations with positive cosmological constant Λ = 2L 2 . To get a better grip on the structure of the
de Sitter space, a variety of coordinates systems can be used. The most common ones are summarised
in Appendix E to which we refer for more details.
As we discussed in detail in Chapter 3, a lot of information about the causal structure of a spacetime
is encoded in its Penrose diagram. The Penrose diagram for de Sitter space is given in Figure 8.1.
Note that for dSd+1 , each point in this diagram is actually an S d−1 , except for the points on the left
and right boundary, where the (d − 1)-spheres shrink. Future and past null infinity, where all null
geodesics end and start, are located at the top and bottom of the diagram and are both spacelike
hypersurfaces. Moreover they, or any other horizontal slice of this diagram, are Cauchy hypersurfaces
and hence, de Sitter space is clearly globally hyperbolic.
One peculiar feature of de Sitter space is that no single observer can access the entire space-time. De
Sitter space describes a exponentially expanding universe and for this reason, there is an observer-
dependent horizon, called the cosmological horizon beyond which space-time is expanding faster
than the speed of light. This is a null hypersurface beyond which the observer can never receive or
send a signal. For example, an observer sitting at the right edge of the Penrose diagram will never

80
 I+

O+

O−

I−
Figure 8.1: The Penrose diagram for dSd+1 . The left and right boundary are time-like lines
which can be identified, while every point in the interior represents an S d−1 . The dashed lines
are the (cosmological) past and future horizons for an observer at the left or right edge of the
diagram. Conformal infinity I ± is space-like.

be able to see anything past the dashed line stretching from the bottom-left to top-right. This is
qualitatively very different from Minkowski space, where a time-like observer will eventually have
access to the entire history of the universe in their past lightcone.
The blue region O+ in the Penrose diagram is the part of spacetime that this observer can send
signals to, while the red region is the part from which they can receive signals. Their intersection
D = O+ ∩ O− is called the causal diamond.

Black holes in de Sitter space

Imposing that the space-time is only asymptotically de Sitter allows us to discuss a variety of more
general solutions. The simplest one of which is the Schwarzschild-de Sitter solutions. This solution
describes an electrically neutral non-ratating black hole in de Sitter space and can be described by
the metric,
−1
r2 r2
  
2 2M 2 2M
ds = 1 − 2 − d−2 dt − 1 − 2 − d−2 dr 2 − r 2 dΩ22 . (8.2)
L r L r
The g t t component of the metric has two positive real zeros, rc (M ) and r+ (M ) with rc > r+ which
are the cosmological and black hole horizons. As the parameter M > 0 increases, rc and r+ tend
to each other and eventually meet at a critical mass Mc = 3pL 3 . This is known as the Nariai limit.
For M > Mc one finds a cosmological solution with no horizons reaching all the way to I + with a
space-like singularity at r = 0, where r is now a time coordinate. Indeed, for M > Mc one can identify
t ∈ R without introducing closed timelike curves. For M < 0, one finds only one horizon and a naked
timelike singularity at r = 0. Note that as we take L → ∞ and d → 3 we recover the asymptotically
flat Schwarzschild black hole discussed in the previous chapter. As in asymptotically flat space, one can
further generalise these solutions to include rotation and/or charge by appropriately modifying the

81
Kerr(-Newman) and Reissner-Nordstrom black holes. One reason to introduce the Schwarzschild-de
Sitter solution is that it plays an important role in determining the entropy and temperature of pure
de Sitter, as will be reviewed below.
One reason why we introduced the Schwarzschild-de Sitter black hole is that it plays an important
role in the work of Gibbons and Hawking [GH77b] determining the entropy of pure de Sitter (see
below for more). To make things more concrete it will be useful to focus on the (2 + 1)-dimensional
case,
dr 2
ds2 = (1 − 8G E − r 2 )dt 2 − − r 2 dφ 2 , (8.3)
(1 − 8G E − r 2 )
where for convenience we set L = 1 and normalised the energy E appropriate for a three-dimensional
black hole. In three dimensions many things simplify, for example, in this case we only have one
p
horizon at rH = 1 − 8G E, and as E goes to zero this reduces to the usual horizon in empty de Sitter
space. The fact that there is no black hole horizon is not so surprising given that in three-dimensions
gravity is not really dynamical and there are no black holes. We can learn a bit more about this
solution by looking at it near r = 0 where we see that the metric behaves as

dr 2
ds2 ≃ rH2 dt 2 − + r 2 dφ 2 . (8.4)
rH2

Defining the rescaled coordinates,

t ′ = rH t , r ′ = r/rH , φ ′ = rH φ , (8.5)

this metric reduces to

ds2 = dt ′2 − dr ′2 − r ′2 dφ ′2 . (8.6)

This looks like flat space but it is not quite. The periodicity of φ was 2π so that the periodicity of φ ′
becomes 2πrH . Therefore we find that the space has a conical singularity with a positive deficit angle
at the origin. This might look familiar, if you remember that in three-dimensions if you put a point-like
mass you get a conical deficit angle at the position of the particle. Hence, the three-dimensional
Schwarzschild-de Sitter background behaves more like a particle then a black hole.

Exercise 8.1. Show that the Schwarzschild-de Sitter black hole is locally isometric to de Sitter space.
More precisely, show that it corresponds to an orbifold of the universal covering space / maximal extension
of de Sitter space.

8.2 Quantum fields in de Sitter space

In this section we discuss the quantization of a non-interacting massive scalar field in a fixed de Sitter
background. As discussed in Chapter 4, when working in curved space, it is much harder to define
what one means by energy. Similarly, the vacuum is no longer a non-ambiguously defined state. In
de Sitter space, thanks to it being maximally symmetric, we can work things out very explicitly.
Let us for simplicity continue with the free scalar field φ(x) with mass m in a fixed de Sitter background

82
in planar coordinates. The action of such a scalar is given by:
Z
1
|g| g µν ∂µ φ∂ν φ + m2 φ 2 + ξRφ 2 .
Æ

S= dd x (8.7)
2

Using the conformal time coordinate η, the de Sitter metric becomes a FLRW metric where the
function C(η) is given by
L2
C(η) = 2 . (8.8)
η
In Chapter 4 we solved the problem of quantising the scalar field in such a background. In de Sitter
space, we have
d(d + 1)
C(η)R(η) = , (8.9)
η2
so that the Klein-Gordon equations reduces to

m2 L 2 d(d + 1)(ξ − ξ(d))

 
χk′′ + k + 2
+ χk = 0 . (8.10)
η2 η2

Performing the change of coordinates s = −kη and defining

p
χk = s f (s) , (8.11)

it is easy to see that f (s) satisfies the Bessel equation,

s2 f ′′ (s) + s f ′ (s) + (s2 − ν2 ) f (s) = 0 , (8.12)

where
1 d2
ν2 = + (ξ(d) − ξ) d(d + 1) − m2 L 2 = − d(d + 1)ξ − m2 L 2 . (8.13)
4 4
We can then write χk in terms of Bessel functions, for η < 0, as
Æ
χk (η) = k|η| [Ak Jν (k|η|) + Bk Yν (k|η|)] . (8.14)

and the normalisation condition leads to

−ik2 η Ak Bk∗ − A∗k Bk W [Jν (k|η|), Yν (k|η|)] = 1 .

(8.15)

Here W [ f , g] = f g ′ − f ′ g is the Wronskian, where the derivative is with respect to the full argument
k|η|. The Wronskian can be computed to be

2
W [Jν (k|η|), Yν (k|η|)] = , (8.16)
πx

so that we find
iπ
Ak Bk∗ − A∗k Bk = − . (8.17)
2k
To get a better feeling for the behaviour of these functions consider the asymptotic region near I+ ,

83
i.e. η → 0. In this region, the modes asymptotically behave as
1
χk ∝ (−kη) 2 ±ν . (8.18)

Notice that for small mass m and ξ, ν is real and the modes either blow up or vanish asymptotically
while for large masses and/or ξ it becomes an oscillatory mode with positive frequency |ν|.

de Sitter vacua

De Sitter space is a static space-time, and moreover maximally symmetric. Hence, guided by the
discussion in the previous sections we look for the ’preferred’ vacuum which preserves all said
symmetries. For this reason, let’s consider the behaviour in the asymptotic past I− , i.e. k|η| ≫ 1,
where we have ωk ≃ k. In this regime we want to consider modes which behave like Minkowski
modes in conformal time η,
1
χk ∝ p e−ikη . (8.19)
2k
In order to find solutions to the Klein-Gordon with this behaviour, consider the asymptotic behaviour
of the Bessel functions. We have
v 
t 2 A + iB A − iB iλ
‹
−iλ
χk ∼ e + e , η → −∞ , (8.20)
π 2 2

where
νπ π
λ = k|η| − − . (8.21)
2 4
Therefore, we require that A + iB = 0, which together with the condition (8.17) results in
π
|A|2 = . (8.22)
4k

Inserting this in the expression for the modes we find,

1 1 1 1
χk (η) = (π|η|) 2 (Jν (k|η|) + iYν (k|η|)) = (π|η|) 2 Hν(1) (k|η|) , (8.23)
2 2

where Hν(1) is the Hankel function of the first kind. This vacuum is called the Euclidean vacuum or
Bunch-Davies vacuum |0〉BD , after [BD78]1
Positive frequency modes in the Bunch-Davies vacuum state are those which become the positive
frequency modes in Minkowski space upon taking the limit k|η| → ∞ [BD78] (see also [STY95]).
We can thus expand our quantum field in terms of the creation and annihilation operators associated
to |0〉, X” —
φ(η, x⃗ ) = ak u E,k (η, k) + a⃗† u∗E,k (η, k) , (8.24)
k
k
1
History has its ways so that Bunch and Davies got the honour of naming this vacuum. However, it was already described
earlier in various papers such as [CT68, SS76].

84
where the creation and annihilation operators satisfy the usual properties:

ak , ak†′ = δkk′ .
 
ak |0〉 = 0 , (8.25)

This vacuum has various interesting properties. Namely, it is invariant under the de Sitter isometry
group SO(1, 4). Clearly, it is invariant under rotations of the spatial coordinates x, since χk (η) only
depends on the modulus of k. It is also invariant under the dilatation,

η → λη , x → λx , λ ∈ R\{0} (8.26)

Indeed, under this transformation the wave-vector transforms as k → λ1 k such that the argument of
1
the Hankel function remains invariant. Collecting the overall factor |η| 2 in χk (η), together with the
d−1 d
factor C 4 /4, we get a total factor of |η| 2 . However, this factor gets cancelled against the factor of
1
1/V 2 in the wave-function uk , where V = L d is the spatial volume. This factor combines with the η
factor to produce (|η|/L)d/2 , showing that the modes are invariant under dilatations. Below we will
see that the invariance of this vacuum manifests itself in the O(1, 4) invariance of the corresponding
Wightman functions.
Only demanding that the vacuum state is invariant under the de Sitter isometries does not uniquely pick
out the Bunch-Davies vacuum. Indeed, the vacuum state of the quantum field could be rather different.
A more general family of vacua is given by the so-called the α-vacua, |α〉 [Mot85,All85], parameterised
by a complex number α. These vacua and their properties are reviewed in [BMS02, SSV01] but as
we’ll see in a bit, these vacua have some funny properties for which reason we usually discard them.
In this course we will mostly focus on the Bunch-Davies vacuum as defined above but surely at some
point these funny extra vacua will turn out to have some purpose in life.

Remark. The Bunch-Davies vacuum is picked out for another reason. We can prepare a de Sitter
vacuum by starting with a Euclidean sphere, cutting it in half and gluing it to ’half’ of the Sitter space.
The state prepared by this procedure is precisely the Bunch-Davies vacuum. This origin furthermore
elucidates where the name ’Euclidean vacuum’ comes from. This construction is motivated by the ’no
boundary wave function’ proposal of Hartle and Hawking [HH83], where a conjectural description
for the wavefunction of the universe was given. According to their prescription, the universe has no
origin. If we would travel back towards the beginning of the universe we would note that, similar
to in the interior of the black hole, our notion of space and time changes. In this case they propose
that close enough to the singularity time stops to exist and we are left with a Euclidean space which
smoothly caps off.

Finally, note that when m = 0 and ξ = ξ(d), the Bunch-Davies vacuum reduces to the conformal
vacuum. Indeed, in this case the index of the Hankel function is ν2 = 14 and we have

‹ 12
2

Hν(1) (z) = −i eiz . (8.27)
πz

so that the modes reduce up to a phase to the Minkowski modes, χk − pi e−ikη , appropriate for a
2k

85
 dSd+1

|0〉BD

S d+1

Figure 8.2: The Bunch-Davies state is prepared by gluing the de Sitter space to a sphere, i.e.
Euclidean de Sitter space. This construction can be though of as a toy model for the universe.

conformal vacuum.

Green’s functions

Having defined the vacuum we would like to compute the Wightman functions. This can be done in
two ways. One way is the brute-force approach, by explicitly plugging the equations for the modes
in (5.3) and performing the integration over momenta. This has been done in [BD78, SS76, CT68],
but it’s rather tedious. Another, more elegant approach is to solve the Klein-Gordon equation for
G+ (x, x ′ ) while assuming O(1, d + 1) invariance along the way. This in fact gives the answer not only
for the Bunch-Davies vacuum but also for the more general family of α-vacua |α〉.
If the Wightman function is computed in a state invariant under the de Sitter isometry group it
should only depend on the O(1, d + 1) invariant distance between the two points. This is nothing
but the geodesic distance P(x, x ′ ) defined in more detail in Appendix E. we can then try to solve the
homogeneous Klein-Gordon equation

□ x + m2 + ξR G + P(x, x ′ ) = 0 .



(8.28)

To do so let us note two useful properties of the geodesic distance,

P2 − 1 P
∇µ P∇µ P = , ∇µ ∇ν P = gµν . (8.29)
L2 L2
Exercise 8.2. Prove the two properties above. You can do so abstractly or by explicit computation in
your favourite coordinate system. Hint: in conformal coordinates (η, x) these take a particularly simple
form. To do so you will need to compute the Christoffel symbols.

Consider now a function F (P) depending on x only through P(x, x ′ ). We then have,

P 2 − 1 ′′ d +1 ′
□ x F (P) = 2
F (P) + P F (P) . (8.30)
L L2

86
Using these properties, it follows that the Wightman function for a scalar field in de Sitter space
satisfies,
(P 2 − 1)∂ P2 G + + (d + 1)P∂ P G + + m2 L 2 + ξd(d + 1) G + = 0 .


(8.31)
1+P
After a change of variables, z = 2 this becomes a hypergeometric equation,

d +1
 ‹  2 
m
z(1 − z)∂z2 G + + +
− (d + 1)z ∂z G − 2
+ ξd(d + 1) G + = 0 . (8.32)
2 H

Comparing this to the standard form (see Appendix F), we find that

α+β = d, αβ = m2 L 2 + ξd(d + 1) , (8.33)

so that we find the general solution

d +1
 ‹
+
G = cm,d 2 F1 h+ , h− , ,z , (8.34)
2

where cm,d is a normalisation constant to be determined shortly and we defined

1” Æ —
h± = d ± d 2 − 4 (m2 L 2 + ξd(d + 1)) . (8.35)
2

The hypergeometric function in (8.34) has a pole at z = 1, or P = 1 and a branch cut for 1 < P < ∞.
The pole occurs when the points x and x ′ are separated by a null geodesic. At short distances, the
scalar field is insensitive to the fact that it lives in a de Sitter space and the form of the singularity
should be the same as that of the propagator in flat Minkowski space. We can use this fact to fix the
normalisation constant. Near z = 1, the hypergeometric function behaves as

d+1

d−1

d +1 ζ(P(x, x ′ ))1−d Γ 2 Γ 2
 ‹
2 F1 h+ , h− , ,z ∼ , (8.36)
2 21−d Γ (h+ ) Γ (h− )

where ζ(P) = cos−1 P is the geodesic separation between the two points. Comparing this expression
with the usual short distance singularity,

d−1


d−1 Γ 1
G +,flat (x, x ′ ) ∼ (−1)
2
2
d+1 d−1
, l 2 = (x − x ′ )2 . (8.37)
4π 2 (l 2 ) 2

Thus we find the normalisation constant

Γ (h+ )Γ (h− )
cm,d = L 1−d d
. (8.38)
(4π) 2 Γ d2

Exercise 8.3. Use the properties of hypergeometric functions to show that near the pole the hypergeometric
(∆η)2 −(∆x)2
function behaves as in (8.36). (Hint: to do so write P = 1 + δ, where δ = 2ηη′ and expand for
small δ.)

Exercise 8.4. Show that for m = 0 and ξ = ξ(d) this expression reduces to the expected expression for a
two-point function in a conformal vacuum.

87
As noted above, the hypergeometic function (8.34) has a branch cut along the semi-infinite axis
running from 1 to ∞. This corresponds to points where P(X , X ′ ) ≥ 1, i.e. points inside the lightcone.
The prescription for avoiding the singularity at the lightcone is the same as in Minkowski space and
simply consists of changing
(η − η′ )2 → (η − η′ − iε)2 . (8.39)

Finally, note that the equation (8.31) is symmetric under interchanging P ↔ −P. So if G(P) is a
solution, then so is G(−P). We therefore find a second linearly independent solution,

d +1 1−P
 ‹
(2)
G+ = cm,d 2 F1 h+ , h− , , . (8.40)
2 2

The singularity now lies at P = −1, which corresponds to X being null separated from the antipodal
point to X ′ . This singularity sounds rather nonphysical at first, but we should recall that antipodal
points in de Sitter space are always separated by a cosmological horizon. The Green’s function (8.40)
can thus be thought of as arising from an image source behind the horizon, and is non-singular
everywhere within an observer’s horizon. Hence the nonphysical singularity can never be detected
by any experiment. The de Sitter space therefore has a one parameter family of de Sitter invariant
Green’s functions
(α)
G+ = 〈α| φ(x)φ(x ′ ) |α〉 = sin α G + + cos α G +(2) , (8.41)

corresponding to the α vacua |α〉 discussed above. Putting α = 0 the antipodal singularity disappears
and the vacuum reduces to the Bunch-Davies vacuum.

8.3 De Sitter temperature

Having discussed some properties of quantum fields in a fixed de Sitter background, let us proceed
to give a brief account of some semi-classical aspects. In particular, we will show that an observer
moving along a time-like geodesic observes a thermal bath of particles when the scalar field is in the
(Bunch-Davies) vacuum state |0〉. We conclude that the de Sitter space is naturally associated with a
temperature TdS.
In Chapter 6 we introduced a variety of tools to diagnose the thermal nature of spacetimes. Now the
time has come to put them to good use. Let us consider an observer, equipped with an Unruh detector,
moving along a time-like geodesic. Along a time-like geodesic, the geodesic distance P is given by

P = cosh ∆τ/L , (8.42)

where ∆τ is the difference in proper time between the two points under consideration. Let us now
consider the detector response function in de Sitter space,
Z
f (E) = d∆τ e−iE∆τ G + (∆τ/L)
(8.43)
d +1 1
 ‹
=cm,d 2 F1 h+ , h− , , (1 + cosh ∆τ/L) .
2 2

88
As discussed above, the hypergeometric function has poles at P = 1 or equivalently at ∆τ = 2πiLn,
for any integer n ∈ Z. To compute the integral in (8.43), consider the contour depicted in Figure 8.3.
Since the contour does not contain any poles the total integrand vanishes
Z ∞ Z −∞−iβ
+
d∆τ e−iE∆τ
G (cosh ∆τ) + d∆τ e−iEτ G + (cosh ∆τ) = 0 , (8.44)
−∞ +∞−iβ

where β = 2πL. After changing variables in the second integral ∆τ → −∆τ − iβ this implies that

f (E) = e−β E f (−E) , (8.45)

which is nothing but the principle of detailed balance in a thermal ensemble!

∆τ

Figure 8.3: The integrand in (8.43) has singularities at ∆τ = 2πin for any integer n. The
contour C will be used to compute the integral.

Remark. Let us suppose that we are in a state where we have the following relation between
probabilities,
P(Ei → E j ) = P(E j → Ei )e−β(E j −Ei ) , (8.46)

for all i and j. Furthermore suppose that the energy levels of the detector are thermally populated,
i.e.
ni = N e−β Ei , (8.47)

for some N . Then it is clear that the total transition rate from Ei to E j is the same as that from E j to
Ei ,
R(Ei → E j ) = ni P(Ei → E j ) = N e−β Ei P(E j → Ei ) = R(E j → Ei ) . (8.48)
1
Hence the state we describe is a thermal state at equilibrium at a temperature T = β. This is the
principle of detailed balance applied to a thermal state.

Since the Green function used in this calculation is invariant under all de Sitter isometries, the result
for the temperature is the same for any observer moving along a time-like geodesic. We conclude

89
that any geodesic observer in de Sitter space will feel a thermal bath of particles at temperature,

1
TdS = . (8.49)
2πL

As discussed in the previous chapter, the laws of black hole thermodynamics [BCH73, Bek73] were
a crucial stepping stone paving the way toward a better understanding of the microscopic nature
of a black hole. The basic observation was that classically any physical process can only lead to
an increase in the area ABH of the black hole horizon (area law) and the way the area responds to
classical processes follows an equation equivalent in form to the first law of thermodynamics,

δM = TBH δSBH + ΩH δJ , (8.50)

where δM is the change in the ADM mass, TBH is the surface gravity divided by 2π, ΩH is the angular
velocity of the horizon, δJ is the change in ADM angular momentum and SBH = ABH /4G. Hawking
famously discovered that TBH is in fact the temperature of the black hole as measured by a far away
observer. The appearance of the first law was no coincidence and a statistical mechanics interpretation
was eventually provided by string theory [SV96]. The black hole entropy is interpreted as a count
of the number of microstates with macroscopic quantities equivalent to those of the black hole they
constitute.
In analogy to the case of black holes, it was proposed by Gibbons and Hawking [GH77b] that there
exists an entropy associated to the pure de Sitter horizon,

πL 2
SdS = . (8.51)
GN

This de Sitter entropy is proportional to the area of the horizon, as in the case of black hole entropy.
Furthermore, as with all horizons, Hawking’s famous result tells us that the de Sitter horizon has a
temperature associated to it, given by TdS = 1/2πL. As before this thermal behaviour is manifested
in the Euclideanised version of de Sitter. In particular, considering the static patch metric and
transforming t → iτ we find that in order to avoid conical singularities in the Euclidean geometry we
have to impose the following periodicity on the Euclidean time coordinate,

τ ∼ τ + 2πL . (8.52)

Once again we see that the temperature is indeed matched by the period of the imaginary time. For
the cosmological constant measured in our own universe one finds SdS ∼ 10120 , which far exceeds
the present entropy2 of the remaining matter content.
When studying black hole thermodynamics, one typically asks how the black hole responds to some
classical process such as absorbing some mass. What is the analogous question in the case of a
cosmological horizon surrounding the static patch observer? We begin with some mass M localized
at the center of the static patch. From analysing the metric (8.2), we see that this has the effect of
reducing the size of the cosmological horizon. One can indeed check that if the mass falls outside the
2
The entropy of the present day universe, including the Bekenstein-Hawking entropy of supermassive black holes, is
estimated around 10104 (see [EL10] and references therein).

90
cosmological horizon
−δM = TdSδScos . (8.53)

This is the analogue law of thermodynamics for a de Sitter horizon. It can be naturally generalized to
include angular momentum and other conserved quantities. The point is that the de Sitter horizon
responds to physical processes very much like any other horizon. Notice however the minus sign in
front of δM . The entropy increases when we throw mass outside the cosmological horizon surrounding
us. Indeed, the less information we have about the interior of the cosmological horizon, the higher
its entropy will be.
We end with an important remark. Though it seems that much of the black hole picture carries
forward to the case of the de Sitter horizon, there are some crucial differences. For instance, an
observer can never approach her own cosmological horizon and probe it. In other words, there is
no sense in which the cosmological horizon is an object localized in space as would be the case for
a black hole. Also, black holes in flat space decay when emitting Hawking radiation. On the other
hand, at least naively, the Hawking radiation of a de Sitter horizon is reabsorbed by the de Sitter
horizon itself, leading to no overall evaporation of the horizon.

8.4 De Sitter entropy

To finish this chapter, let us discuss the entropy associated to de Sitter space in some more detail. For
simplicity, let us again restrict our attention to dS3 , where the analysis simplifies considerably.
For the case of black holes one can use similar methods as those in the previous chapter to calculate
the temperature TBH of the black hole. The black hole entropy SBH can then be found by integrating
the thermodynamic relation
dSBH 1
= , (8.54)
dEBH TBH
where EBH is the energy or mass of the black hole. So if you know the value of the temperature just
for one value of EBH you will not be able to get the entropy, but if you know it as a function of the
black hole mass then you can simply integrate (8.54) to find the entropy. The constant of integration
is determined by requiring that a black hole of zero mass has zero entropy.
So for de Sitter space one would expect to use the relation

dSdS 1
= (8.55)
dEdS TdS

to find the entropy SdS. The problem in de Sitter space is that once the coupling constant of the
theory is chosen there is just one de Sitter solution, whereas in the black hole case there is a whole
one parameter family of solutions labeled by the mass of the black hole, for fixed coupling constant.
In other words, what is EdS in (8.55)? One might try to vary the cosmological constant, but that is
rather unphysical as it is the coupling constant. One would be going from one theory to another
instead of from one configuration in the theory to another configuration in the same theory.
Let us instead follow Gibbons and Hawking [GH77b] and use the one parameter family of
Schwarzschild-de Sitter solutions to see how the temperature varies as a function of the parameter E

91
labeling the mass of the black hole.

Exercise 8.5. The three-dimensional Schwarzschild-de Sitter in static coordinates is

dr 2
ds2 = (1 − 8G E − r 2 )dt 2 − − r 2 dφ 2 . (8.56)
(1 − 8G E − r 2 )

Find a Green function by analytic continuation from the smooth Euclidean solution. Show that this Green
function is periodic in imaginary time with periodicity

2πi
t∼t+p . (8.57)
1 − 8G E

From this exercise and the discussion above we conclude that the temperature associated with the
Schwarzschild-de Sitter solution is p
1 − 8G E
TSdS = . (8.58)
2π
Using the formula
dSSdS 1
= , (8.59)
dE TSdS
p
and writing the result in terms of the area AH of the de Sitter horizon at rH = 1 − 8G E which is
given by
p A
1 − 8G E = H , (8.60)
2π
one finds that the entropy is equal to
AH
SSdS = − . (8.61)
4G
This differs by a minus sign from the famous Bekenstein-Hawking formula! What did we do wrong?
Gibbons and Hawking suggested that to get the de Sitter entropy we should use not (8.59) but instead

dSSdS 1
= . (8.62)
d(−EdS) TSdS

This looks funny but in fact there is a very good reason for using this new formula.
The de Sitter entropy, although we don’t know exactly how to think about it, is supposed to correspond
to the entropy of the stuff behind the horizon which we can’t observe. Now in general relativity the
expression for the energy on a surface is the integral of a total derivative, which reduces to a surface
integral on the boundary of the surface, and hence vanishes on any closed surface. Consider a closed
surface in de Sitter space such as the one shown in figure 8.4. If we put something with positive
energy on the south pole, then necessarily there will be some negative energy on the north pole.
This can be seen quite explicitly in the Schwarzschild-de Sitter solution. With no black hole, the
space-like slice in figure 8.4 is an S 2 , but we saw that in the Schwarzschild-de Sitter solution there is
a positive deficit angle at both the north and south poles. If we ascribe positive energy to the positive
deficit angle at the south pole, then because the Killing vector ∂ /∂ t used to define the energy changes
direction across the horizon, we are forced to ascribe negative energy to the positive deficit angle at
the north pole.

92
 I

North Pole

South Pole
t= 0

I
Figure 8.4: The energy associated to the Killing vector ∂ /∂ t (indicated by the arrows) along
the spacelike slice t = 0 (solid line) must vanish. If we ascribe positive energy to a positive
deficit angle at the south pole, then we must ascribe negative energy to a positive deficit
angle at the south pole since the Killing vector ∂ /∂ t runs in the opposite direction behind the
horizon.

Therefore the northern singularity of Schwarzschild-de Sitter behind the horizon actually carries
negative energy. In (8.59) we varied with respect to the energy at the south pole, and ended up with
the wrong sign in (8.61), but if we more sensibly vary with respect to the energy at the north pole,
then we should use the formula (8.62). Then we arrive at the entropy for Schwarzschild-de Sitter

AH π p
SSdS = = 1 − 8G E . (8.63)
4G 2G
1
The integration constant has been chosen so that the entropy vanishes for the maximal energy E = 8G
at which value the deficit angle is 2π and the space has closed up. In conclusion, we see that the
Bekenstein-Hawking area-entropy law indeed continues to apply in three dimensional Schwarzschild-
de Sitter. Moreover, we see that the de Sitter entropy is maximal for an empty, pure de Sitter space.
After a moment of thought this is very logical though. Our universe is expanding, so eventually all
matter will dilute and end up behind the cosmological horizon. In order to preserve the second law
of thermodynamics we are automatically lead to the conclusion that empty de Sitter has to have
maximal entropy.
This is in stark contrast to flat space or anti-de Sitter where if we throw more matter into the space,
or build a bigger black hole the entropy continues to increase. Indeed, in these spaces the universe is
not expanding and eventually the gravitational force will win and pull all the matter together into
one huge black hole, which according to the second law of thermodynamics is then the most entropic
state in (asymptotically) flat or anti-de Sitter spaces.

93
 Chapter 9

Cosmological perturbations

In a maximally symmetric universe such as Minkowski or (A)dS, there cannot be a beginning or

end of time. There cannot be any ’history’ because each time is equivalent to any other time. While
such eternal universes might be appealing from a philosophical or aesthetic point of view, it is in
contradiction with the last century of cosmological observations. This simplest possible non-trivial
cosmological toy models are FLRW universes.
The same process that creates particles in an expanding universe is responsible for creating, in the
context of an early inflationary expansion, the primordial fluctuations that are now observed with
astonishing accuracy in the cosmic microwave background (CMB) radiation. These same primordial
fluctuations also appear responsible for the large-scale structure of the universe.

9.1 Cosmological models

The FLRW models are models in which we assume a collection of co-moving observers for whom the
universe is homogeneous (the same for each observer) and isotropic (the same in every direction).
In Appendix E we introduce some basic facts about FLRW spaces as well as discuss the Friedmann
equations in general dimensions. In this chapter we will mostly restrict ourselves to d = 3 spatial
dimensions. For more details we refer the reader to one of the standard textbooks on modern
cosmology, such as [Dod03].
The FLRW metric has been introduced before and can be written as

ds2 = dt 2 − a(t)2 dsM

2
, (9.1)
k,3

where Mk,3 is the maximally symmetric spatial manifold with curvature k = 0, ±1. For any such spatial
manifold, the FLRW metric is conformally flat, as can immediately be seen by going to conformal
coordinates Z t
2 2
€
2 2
Š dt ′
ds = C(η) dη − dsM , η(t) = , (9.2)
k,3 a(t ′ )
where C(η) = a(t(η)). In the context of these cosmological models, the conformal coordinate is
particularly useful to understand the causal structure of the model. In our own universe, the spatial
curvature is tantalisingly close to k = 0, but so far experiments are not yet able to determine whether
the spatial curvature is exactly vanishing or not. A particularly important quantity in such models is
the Hubble parameter,
ȧ(t) Ċ(η)
H(t) = = , (9.3)
a(t) C(η)2

94
which captures the time dependence of the scale factor. All the cosmological phenomena we will
encounter will have an crucial dependence on H. As long as the so-called null energy condition (NEC)
is met, H will be decreasing during the expansion of the universe. A purely (positive) cosmological
constant saturates the NEC and consequentially, de Sitter space has a constant hubble parameter,

1
HdS = . (9.4)
L

A natural assumption for the stress tensor, respecting the symmetries of the problem, is that of a
perfect fluid,
Tµν = diag(ρ, −p, −p, −p) , (9.5)

The Einstein equations for an FLRW metric are usually referred to as the Friedman equations and in
terms of the conformal coordinate η take the form,
 2
Ċ 8π k C̈ + C k
=− ρ− 2 , 6 3
= 8πGN (ρ − 3p)C + . (9.6)
C 3 C C C

Together with the continuity equation,

Ċ
ρ̇ + 3 (ρ + p) = 0 , (9.7)
C

these determine the full evolution of the FLRW universe. However, the form of the equation of state
p(ρ) is not determined by these equations and has to be put in by hand. For non-relativistic matter
(dust), relativistic matter (radiation) or a cosmological constant the equation of state takes the form
p = wρ , where w ∼ 0 for non-relativistic matter, w = 13 for relativistic matter and w = −1 for the
cosmological constant.

Example 9.1. As an example, let us consider dust in an FLRW universe with k = 1. After solving the
Friedman equations we find,

C(η) = c(1 − cos η) , t = c(η − sin η) . (9.8)

Hence, we find a big bang (C = 0) at η = 0, followed by a big crunch at η = π.

Why inflation?

To finish the discussion of cosmological models, let us recall a particular problem for universes
undergoing a decelerated expansion. In the current standard cosmological model (known as ΛCDM)
an accelerated expansion is induced at late times by the cosmological constant. At earlier times, when
the universe is radiation or matter dominated, we expect a period of decelerated expansion. We refer
to such models as hot big bang models, where the adjective hot refers to the temperature of radiation.
In general, in order to justify the homogeneity of the spatial slices, we would like to have that the
distance between regions of space that look the same is much smaller then the maximal distance
travelled by light since the beginning of time. Otherwise it is hard to explain why the two region can
look similar, since their causal past is disconnected. However, in hot big bang models, this desired

95
inequality is dramatically violated. More precisely, cosmological observations of far away objects
allow us to see regions in the past that are separated by much more that than the particle horizon at
the time, which is the furthest a signal can travel. Any mechanism attempting to explain homogeneity
across these regions then necessarily violates causality and/or locality, leading to the horizon problem.
This apparent tension is resolved by inflation. This scenario posits that there is a surface of last
scattering at some time t s soon after the big bang, before which we cannot clearly see what was
going on. This surface is where radiation decouples from matter and so after this time, we can see
what is going on, whereas before, we just have what we see from the cosmic microwave radiation.
Inflationary models are then obtained by gluing in an exponentially expanding region of de Sitter
space, before the surface of last scattering. This inflationary phase gives the past of distant regions
time to mix and homogenise so as to explain the homogeneity and isotropy of the universe.
Guided by experimental evidence, it is by now firmly believed that the cosmological constant of our
universe is positive. This implies that at large times the scale factor will diverge. At such time the
contributions from the cosmological constant will dominate the Friedmann equations which in this
regime (for k = 1) we can approximate by,
v
tΛ 1
Ċ = C2 , C∼ . (9.9)
3 ηI − η

We find that C ∼ − Λ3 η−η

q
1
I
+ · · · such that the scale factor has a pole at conformal infinity η = ηI .
This implies that near I + , conformal infinity looks like that of de Sitter, i.e. we have an asymptotically
de Sitter universe. For k = 0, −1 the result is similar. This exponential expansion arises because a
positive cosmological constant has a stress-energy tensor Tµν = diag(Λ, −Λ, −Λ, −Λ), so that although
the effective energy density is positive, the pressure is negative. At the current age of the universe,
the contribution of Λ to the energy density is thought to be of the same order as that of the matter
including dark matter (visible matter being thought to be 3%, dark matter 30% and cosmological
constant about 67% of the critical mass of the universe). Such a ratio of matter to cosmological
constant is extremely high at early times, and extremely low at late times, and this sometimes leads
to the ‘why are we alive now?’ question. The later periods are however, very cold and boring, and
the early periods rather hot, and too early for structure to form, leading to the answer in the form of
an anthropic principle.
In this section we introduced just one motivation for considering inflation, i.e. the horizon problem.
However, there are various other problems arising in hot big bang models, such as

• The curvature related to the approximate spatial flatness of our current universe,

• The particle horizon problem related to the statistical isotropy of our universe,

• The phase coherence problem related to the homogeneity of the CMB,

• The scale invariance problem related to the scale invariance of the CMB.

Discussing all these problems in more depth would take us too far, but importantly inflation offers a
way out for each of them and provides us with a plausible explanation for the current state of the
universe. For more details we refer to [Dod03, Paj20].

96
9.2 QFT on a cosmological background

Having briefly introduced the cosmological FLRW models, let us apply the methods introduced in
this course to such space-times. In chapter 4 we already consider the quantum scalar field in such
backgrounds and we refer the reader to that chapter for details on the quantisation. Here we only
repeat the results we need for the analysis at hand.
Let us for simplicity consider the FLRW metric with flat spatial slices

ds2 = dt 2 − a(t)2 dx · dx = C 2 (dη2 − dx · dx) , (9.10)

where t is the proper time of the co-moving observers, a the scale factor, and η the conformal time.
Writing the modes as
eik·x
uk (η, x) = χk (η) , (9.11)
(2π)3/2 C(η)1/2
The Klein-Gordon reduces to the following equation for

χk′′ (η) + ω2 (η)χk (η) , (9.12)

where
C̈
ω(η)2 = k2 + m2 C(η)2 − (1 − 6ξ) . (9.13)
C
The inner produce, or equivalently the symplectic form on the phase space, is conserved in time.
However, now that the space-time is explicitly time-dependent, energy is not conserved and in the
absence of asymptotically static patches, particles will be created at any time and there is no fixed or
preferred vacuum.
However, at each time η there is some notion of vacuum, i.e. the state without particles, or equivalently
the instantaneous low energy state. Expanding the quantum scalar field as

d3 k
Z
ak χk eik·x + ak† χk∗ e−ik·x ,


φ= (9.14)
(2π)3 C(η)
p

we find the Hamiltonian

Z
1
d3 k Fk ak a−k + Fk∗ ak† a−k
†
+ (ak† ak + ak ak† )Ek ,


H= (9.15)
4

where
Ek = |χ̇|2 + ω2 |χ|2 , Fk = χ̇k2 + ω2 χ 2 . (9.16)

With this implicit choice of vacuum we find

Z
E 1
= 〈0| H |0〉 = d3 kEk , (9.17)
V 4

where as usual we took out a diverging volume factor V ∼ δ(3) (0) to get the local energy density.
Minimising Ek for each mode subject to the appropriate normalisation condition a brief computation

97
shows that we obtain the following initial data to impose at the hypersurface η = η0

1
(χk , χ̇k ) η=η0
=p (1, −iω(η0 )) . (9.18)
2ω(η0 )

The particle content measured in this vacuum is that of the instantaneous ‘static’ observer.

Exercise 9.1. Derive the initial conditions (9.18) starting from the reduced Klein-Gordon equation
(9.12) by demanding that the energy is minimized for each mode.

When the frequencies ω(η) are varying slowly enough, we can use an adiabatic approximation to
define also an evolution of the vacuum. This arises as the WKB approximation at leading order; more
generally, the adiabatic approximation takes the WKB approximation beyond leading order and then
there will be particle creation. 1
Having defined the instantaneous vacuum at each time η and ideally also have an evolution equation,
we can define the Bogoliubov transformations between any two times η1 and η2 and compute the
particle creation just like we did in the previous chapters. As we discussed above, the Ek determines
the energy at some time η. The Fk on the other hand determines the instantaneous particle creation
at each time. If we are in the lowest energy vacuum at some time η0 the Fk at that time vanishes.
A novel feature of having the time dependent frequencies is that for large length scales and small
enough k and m, the frequency ω2k can become negative. This, as we will see now, leads to the
corresponding modes ceasing to oscillate and essentially freezing out.
To see this in more detail, let us consider the simplest possible cosmological model, namely de Sitter.
Looking at the planar coordinates (see Appendix E) we see that de Sitter is an FLRW space with scale
factor a = Le t/L and constant Hubble parameter H = 1/L. In particular, the Hubble parameter is
proportional to the radius of the cosmological horizon.
The de Sitter mode functions, discussed in the previous chapter, behave very different from their
Minkowski counterparts when the wave-number k becomes smaller that the co-moving Hubble
parameter
1
k < kHC = aH = , (9.19)
|η|
where HC stands for Hubble crossing, sometimes also called horizon crossing. In physical length
scales this means that the physical wavelength λ = a/k is stretched by the expansion to become
larger than the Hubble radius 1/H. Since k and H are constant, while a = e t/L grows with time, all
modes cross the Hubble radius as time proceeds and eventually become "super-Hubble" modes. Unlike
"sub-Hubble" modes with k ≫ aH, which oscillate, super-Hubble modes freeze out and asymptote to
a constant.
1
This requires a mechanism for the dissipation of the particles created and otherwise we are working in the full quantum
theory in the Heisenberg representation where the state is fixed and the evolution is carried by the operators.

98
9.3 Cosmological correlators

Having discussed some general properties of quantum fields in cosmological backgrounds, a natural
question is: What are the observables for these theories. As familiar from the above, these are given
by the expectation values of operators. However, in cosmology, we have observational access only
to expectation values in the infinite future η → 0. In this limit, observables become approximately
constant and so we will only be interested in the expectation values of products of local operators
inserted at equal times. Such operators are often called cosmological correlators

lim 〈O(x1 , η)O(x2 , η) · · · O(xn , η)〉 . (9.20)

η→0

Since we are studying free theories, all information is contained in the two-point correlators of φ
and its conjugate momentum π. All odd point correlators vanish by the symmetry φ ↔= φ and all
higher even-point correlators can be reduced to two-point correlation functions using Wick’s theorem.
Let us therefore consider the two-point correlator in a generic cosmological background,

†
lim φ(k)φ(k′ ) =|uk |2 ak a−k ′
η→0
(9.21)
=(2π)3 δ(k − k′ )P(k) ,

where we Fourier transformed to momentum space and introduced the power spectrum

H2
P(k) = , (9.22)
2k3

for a massless scalar. Note that the Dirac delta simply encodes the momentum conservation and that
the power spectrum does not depend on the direction of k, due to the isotropy of the background.
Furthermore, the fact that the power spectrum asymptotes to some non-vanishing constant at η → 0
is related to the absence of mass. Similarly, the k-dependence P ∝ k−3 is corresponding to the fact
that the massless scalar is scale invariant. Indeed this can be seen by Fourier transforming back to
real space where the correlation function is constant and independent of the separation between the
two points.
Introducing mass, we find the power spectrum to be

H2 (−kη)3−2ν
P(k) = |uk |2 = , (9.23)
π22(ν−1) Γ (ν)2 k3
Ç
2
for m2 < 94 H 2 , where ν = 94 − m H2
, for a minimally coupled scalar (see previous chapter). Because
of the mass, the power spectrum is not scale invariant anymore. Indeed, P ∝ k−2ν . Also, P has
acquired a time dependence. For positive m2 > 0, one finds 3 − 2ν > 0 and the power spectrum
decays with time and vanish at future infinity. This is to be expected because the quadratic potential
pushes the field towards φ = 0. For negative m2 we would expect an instability and indeed the power
spectrum grows with time and diverges at future infinity. The second case is when the mass square is
large and positive, m2 > 43 H 2 , then ν becomes complex and the the power spectrum oscillates while
decaying as η3 . In cosmology, we are mostly interested in massless or almost massless fields, which

99
do not create large instability and whose perturbations survive long enough to be observable at late
times.

9.4 Fluctuating gravitons

By following a very similar procedure for the scalar field, it is straightforward to quantize metric
fluctuations as well. To do so, we divide the metric into a classical background and small quantum
fluctuations,

gµν (t, x) = ḡµν (t, x) + hµν (t, x) . (9.24)

A priori, there are ten independent components of the fluctuations. Four of them however obey
constraint equations which are at most first order in time derivatives and therefore not dynamical. To
see this recall the Bianchi identity
∇µ Gµν = 0 . (9.25)

Writing this out, we obtain

α νγ ν
∂ t G tν = −∂m G mν − Γαγ G + Γαγ G αγ . (9.26)

Since the right hand side has at most second order derivatives of the metric, we see that G tν has at
most one time derivative. Hence the metric must appear with just one time derivative in four of the
Einstein equations and we must have some constraint equations that limit the set of consistent initial
data for the fluctuations. It takes a bit more work to which has been done in GRII to fully linearise the
fluctuations and Einstein equations but let us here just summarise the result for spatially flat FLRW
metrics.
A convenient gauge choice is given by

ds2 = dt 2 − a2 (δmn + γmn )dx m dx n , (9.27)

where γ is transverse, i.e. ∂ m γmn = 0 and traceless γm

m = 0. So we are left with two independent
components. Expanding the Einstein Hilbert action to quadratic order in the fluctuations we find

2 Z
MPl
d3 x dt a2 γ′mn γ′mn − ∂m γnk ∂ m γnk .


S2 = (9.28)
8

This action can be derived by rigorously linearising Einsteins equations, but alternatively it could
easily have been guessed by writing the simplest action consistent with the symmetries of the problem.
As we did for the scalar we can expand the graviton in plane waves by writing
Z X
γmn (x) = d3 k εsmn (k)γs (k)eik·x , (9.29)
s=±

100
where εsmn (k) are the polarisation tensors, which are generally complex and satisfy

εsmm (k) =km εsmn (k) = 0 , Transverse and traceless ,

εsmn (k) =εsnm (k) , Symmetric ,
εsmn (k)εsnk (k) =0 , Light-like , (9.30)
′ ′
εsmn (k)εs mn (k) =δss , Unit normalised ,
εsmn (k)∗ =εsmn (−k) , γi j is real .

From these properties we can derive explicit expressions for the polarisation vector and rewrite the
action as
2
k2
Z X 
MPl 3 2′ ′
S2 = d k dt a γs (k)γs (−k) − 2 γs (k)γs (−k) . (9.31)
4 s=±
a

Now this action consists of two independent copies of the action for a (canonically normalised)
M2
massless scalar field, up to a normalisation factor 2Pl . To quantise the gravitons we therefore can
proceed exactly as above. We can promote γs (k) to an operator and expand it in creation and
annihilation operators, p
2
fk aks + fk∗ aks† ,


γs (k) = (9.32)
MPl
where for both signs s = ± the creation and annihilation operators satisfy the canonical commutation
relations. If we now assume a de Sitter background we can explicitly compute the graviton power
spectrum in the same way as we did for the massless scalar field. We find
X ′
γmn (k)γmn (k′ ) = εsmn (k)εsmn (k′ ) γs (k)γs′ (k′ )
s,s′
2 X ′
= 2
εsmn (k)εsmn (k′ )(2π)3 δ(k + k′ )| fk |2
MPl s,s′ (9.33)
2
2 H X
= 2 δss′ (2π)3 δ(k + k′ )| fk |2
MPl 2k3 s,s′

=(2π)3 δ(k + k′ )PT ,

with
4 H2
PT = 2
. (9.34)
k3 MPl

This power spectrum provides us with a clear prediction for the CMB spectrum after a period of
inflation starting from the Bunch-Davies vacuum. This spectrum can then be compared to the
observational data from cosmological experiments and gives an excellent match. However, the match
is not exact. This is to be expected, since in this course we take the coarse approximation that all
fields are free. It turns out that this is an excellent approximation to predict the CMB radiation but
recent experiment have nonetheless found tiny non-Gaussianities hidden in the observed radiation.
To properly account for such effects one has to introduce interactions. This goes beyond the scope of
these lectures, but we refer the interested reader to [Paj20] for more details.

101
 Chapter 10

Quantum fields in AdS

de Sitter space is important because it describes the early and late time behaviour of the real universe.
Anti de Sitter space is important for an entirely different reason. It is the background in which the
holographic principle is best understood. Thanks to the isometries of AdS, the observables in these
theories are constrained by the SO(2, d) conformal group, even in the presence of mass deformations,
allowing us to make progress in explicitly solving them. In addition, the AdS length scale provides a
convenient IR regulator for interacting quantum field theories.
In many ways putting a quantum theory in anti de Sitter space makes things behave better. We can
think of AdS as putting the system in a box. However, this will come at the cost of introducing some
subtleties mostly resulting from the fact that AdS is not a globally hyperbolic space and hence we
need to impose appropriate boundary conditions at the conformal boundary.

10.1 Geometry of AdS spaces

Anti de Sitter space is the maximally symmetric, negatively curved space. An easy way to construct it
is as the isometric immersion of an (d + 1)-dimensional hyperboloid in a (d + 2)-dimensional ambient
space. More concretely it is defined as the universal cover of the manifold

2
2
2
2
− X0 + X1 + ··· + Xd − X d+1 = −L 2 , (10.1)

embedded in Rd,2 . The various coordinate systems are summarised in Appendix E, to which we refer
the reader for more details. For concreteness let us here consider AdS space in global coordinates. In
this case we choose the embedding coordinates as

X0 = L cos t cosh ρ
µ
X = L ωµ sinh ρ (10.2)
X d+1 = L sin t cosh ρ

where ωµ (µ = 1, . . . , d) parameterises a unit (d − 1)−dimensional sphere. Doing so we obtain the

metric with metric,
ds2 = LAdS
2
cosh2 ρ dt 2 − dρ 2 − sinh2 ρ dΩ2d−1 .


(10.3)

By taking ρ ≥ 0 and 0 ≤ t < 2π, we cover the entire hyperboloid once. However, since the t direction
has the topology of an S 1 , this space however contains closed time-like curves. To restore the causality
we can simply unwrap the S 1 , i.e. take −∞ < t < ∞ in (10.3) to obtain the universal covering
of the hyperboloid without closed time-like curves. This coordinate system can teach us several

102
interesting facts about the geometry of anti de-Sitter space. For example, we can see that any light
ray will reach spatial infinity in finite time t. Indeed, if we consider a light ray at constant position on
the (d − 1)-sphere we find
∞
πLAdS
Z Z
dρ
∆t = LAdS dt = LAdS = . (10.4)
0 cosh ρ 2

This has the important implication that all observers in AdS space can communicate with each other
from any point in space within finite (proper) time.

Causality and the conformal boundary

Notice that as we approach ρ → ∞, the metric blows up. The locus ρ = ∞ is strictly not a part
of AdS. However, as discussed in Chapter 3 we can consider the conformal compactification of AdS
including the hypersurface IAdS at infinity. IAdS is also called the conformal boundary. More precisely,
the conformal boundary is defined as the conformal equivalence class of metrics ds̃2 = e−2ρ ds2 with
boundary R1,d−1 at ρ = ∞.
Particularly interesting for our purposes is the relation between the conformal compactification of
AdS and flat space. It is well-known that Euclidean flat space can be compactified to the d-sphere, S d
by adding a point at infinity. On the other hand, Euclidean AdSd+1 , which is simply the hyperbolic
space, can be conformally mapped to the (d + 1)-dimensional disk. Therefore the boundary of the
compactified Euclidean AdS space is the compactified Euclidean plane.
Similarly, in Lorentzian signature, by changing to hyperspherical coordinates, i.e. performing the
coordinate change sinh ρ = tan θ from global coordinates, we obtain

2
LAdS
ds2 = −dt 2 + dθ 2 + sin2 θ dΩ2d−1 ,


(10.5)
cos2 θ

which after a conformal rescaling becomes the metric of the Einstein static universe. However, it is
only half of the Einstein static universe since θ is restricted to the range [0, π/2) rather then [0, π).
The boundary of this space is at θ = π/2 and is given by R × S d−1 . This is identical to the conformal
compactification of d-dimensional Minkowski space. This identification will play an essential role in
the AdS/CFT correspondence.
Before moving on to considering quantum fields in AdS there are a few causal issues that need our
attention. A peculiarity of AdS spaces is that an initially radially outward trajectory from ρ = 0 will
begin to re-converge to its starting point following a period of π2 . Remembering the discussion in
Chapter 3 a Lorentzian spacetime is globally hyperbolic if and only if it contains a Cauchy hypersurface,
i.e. a hypersurface whose domain of dependence covers all of the spacetime. Famously, AdS is not
globally hyperbolic. This becomes clear by looking at Figure 10.2, where the domain of dependence
for the spatial hypersurface Σ is indicated by the blue diamonds.
Therefore, there are regions for which a knowledge of events on Σ does not allow any prediction. The
underlying cause of these issues is the fact that conformal infinity in AdS is a time-like hypersurface.
Indeed, as a massless particle can reach spatial infinity in finite time, it can then propagate along I

103
 Figure 10.1: AdSd+1 can be conformally mapped into one half the Einstein static universe.
This space has boundary R×S d−1 which is exactly the conformal compactification of Minkowski
space.

τ=π

I I

γ γ′

τ=0

τ = −π

Figure 10.2: The Penrose diagram of AdS. The blue domain blue diamonds denote the domain
of dependence of the spatial hypersurface Σ. Two geodesics, γ and γ′ are denoted in yellow.
In the universal covering the diagram continues indefinitely in the vertical direction.

and move outside of D(Σ). In this way information is ’lost’ from Σ to spatial infinity. Similarly AdS
space allows for information to be introduced from spatial infinity.
To resolve these issues, and restore the predictability in the entire space, we have to provide boundary
conditions at spatial infinity. In these lectures we want to obtain a closed system hence reflective
boundary conditions are the natural choice [AIS78]. In the next section, when discussing the solutions

104
to the wave equation we will discuss these boundary conditions at length. Note that such boundary
conditions amount to requiring that there is no net flux across spatial infinity.

Asymptotically locally AdS spaces

When we consider matter fields coupled to gravity they will back-react on the metric and we will no
longer have an exact AdS space. However, a lot of the machinery developed for AdS spaces will still
be valid. By adding matter to the theory the bulk of AdS will change but the conformal boundary is
a rather robust characteristic of spaces with a negative cosmological constant. It takes an infinite
energy to change the asymptotics of such spaces. Therefore we will be able to extend the analysis of
pure AdS to the class of asymptotically locally AdS (AlAdS) spaces. In particular they all have the
same conformal boundary.
In Poincaré coordinates this statement implies that the metric of any AlAdS space near the conformal
boundary is of the form
2
LAdS
ds = 2 dz 2 + g mn (z, x)dx m dx n ,
2


(10.6)
z
where g mn (z, x) is smooth and finite as z → 0. This function can be expanded in powers of the radial
coordinate near the conformal boundary as
∞
(n)
X
g a b (z, x) = z n g ab (x) . (10.7)
n=0

Similarly, the matter fields coupled to gravity can be expanded near the conformal boundary. This
expansion is called the Fefferman-Graham expansion.

10.2 Quantum fields in AdS

We have seen that AdS acts like a box for classical massive particles. Quantum mechanically, this
confining potential gives rise to a discrete energy spectrum. Consider the Klein-Gordon equation
€ Š
□ − m2ξ φ = 0 , (10.8)

in global coordinates. Since the Ricci scalar is a constant in AdS we define the effective mass mξ as

m2ξ = m2 + ξL . (10.9)

To avoid excessive notation we will mostly suppress the subscript ξ. To solve this problem we will use
an indirect route which has the advantage that it makes the correspondence with holography more
explicit. Consider the action of the quadratic Casimir of the AdS isometry group on a scalar field1

1
JAB J BAφ = −X 2 ∂X2 + X · ∂X (d + X · ∂X ) φ .
 
(10.10)
2
Formally, we are extending the function φ from AdS, defined by the hypersurface X 2 = −L 2 , to the embedding space.
1

However,the action of the quadratic Casimir is independent of this extension because the generators JAB are interior to
AdS, i.e. JAB , X 2 + L 2 = 0.

105
By foliating the embedding space R2,d with AdS surfaces of different AdS radii L, we can obtain the
Laplacian in the embedding space as

1 ∂ d+1 ∂
∂X2 = − L + □AdS . (10.11)
L d+1 ∂L ∂L
Substituting this in (10.10) and noticing that X · ∂X = L∂ L we conclude that

1
JAB J BAφ = L 2 □AdSφ . (10.12)
2

Therefore, we should identify m2ξ L 2 with the quadratic Casimir of the conformal group. The Lorentzian
version of the conformal generators is

D = −J0,d+1 , Pµ = Jµ0 + i Jµ,d+1 , (10.13)

Mµν = Jµν , Kµ = Jµ0 − i Jµ,d+1 . (10.14)

Exercise 10.1. Show that, in global coordinates, the conformal generators take the form

∂
D=i ,
∂t
∂ ∂
‹
Mµν = −i ωµ − ων ,
∂ ων ∂ ωµ
1
• ˜
−i t


Pµ = −ie ωµ ∂ρ − i tanh ρ ∂ t + ∇µ ,
tanh ρ
1
•
˜
it
Kµ = ie ωµ −∂ρ − i tanh ρ ∂ t − ∇µ ,
tanh ρ

∂
where ∇µ = ∂ ωµ − ωµ ων ∂ ∂ων is the covariant derivative on the unit sphere S d−1 .

In analogy with CFT construction we can look for primary states, which are annihilated by Kµ and are
eigenstates of the Hamiltonian, Dφ = ∆φ. The condition Kµ φ = 0 splits in one term proportional to
ωµ and one term orthogonal to ωµ . The second term implies that φ is independent of the angular
variables ωµ while the first term reduces to ∂ρ + ∆ tanh ρ φ = 0. This implies that

∆ ‹∆
e−i t
 
R
φ∝ = . (10.15)
cosh ρ 0
X − X d+1

This is the lowest energy state with eigenvalue ∆. Starting from this state one can construct excited
states by acting on it with the generator Pµ . Notice that all such states have the same value for the
quadratic Casimir
1
JAB J BAφ = ∆(∆ − d)φ . (10.16)
2
Hence in this way we can generate all normalisable solutions of the Klein-Gordon equation with
m2 L 2 = ∆(∆ − d). This shows that the one-particle energy spectrum is given by ω = ∆ + l + 2n
where l = 0, 1, 2, . . . is the spin, generated by acting with the traceless generators, Pµ1 . . . Pµl − traces,

n
and similarly, the quantum number n = 0, 1, 2, . . . is generated by acting with P 2 .

106
Note that, for a given (effective) mass we can solve the equation m2 L 2 = ∆(∆ − d) as
v
d t d2
∆ = ∆± = ± + m2 L 2 . (10.17)
2 4
d
Demanding that ∆± ∈ R requires having m ≥ − 2L . Hence, we see that a range of tachyonic masses is
allowed. In Minkowski space this would lead to an instability of the perturbative vacuum. In AdS
space, whenever the mass-squared lies above this bound the free energy of the field is bounded from
below and no instabilities arise. This bound is called the Breitenlohner-Freedman bound after [BF82].

Exercise 10.2. Given the symmetry of the metric (E.46) we can look for solutions of the form

φ = e iωt Yl (Ω)F (r) , (10.18)

where Yl (Ω) is a spherical harmonic with eigenvalue −l(l + d − 2) of the Laplacian on the unit sphere
S d−1 . Derive a differential equation for F (r) and show that it is solved by

l +∆−ω l +∆+ω
 ‹
d
F (r) = (cos r)∆ (sin r)l 2 F1 , , l + , sin r , (10.19)
2 2 2
p
with 2∆ = d + d 2 + 4(mL)2 . We chose this solution because it is smooth at r = 0. Now we also need to
impose another boundary condition at the boundary of AdS, i.e. r = π2 . Imposing that there is no energy
flux through the boundary leads to the quantization of the energies |ω| = ∆ + l + 2n with n = 0, 1, 2, . . .
(see for example [AGM+ 00]).

If there are no interactions between the particles in AdS, then the Hilbert space has a Fock represent-
ation and the energy of a multi-particle state is just the sum of the energies of particles. Turning on
small interactions leads to small energy shifts of the multi-particle states.

Green’s functions

For the computation of the Green’s function, let us now return to Euclidean signature and afterwards
analytically continue the result to obtain the Lorentzian Feynman Green’s function. For simplicity, we
consider a free scalar field with action
Z
1
dd+1 x (dφ)2 + m2 φ 2 .
 
S= (10.20)
2 AdS

Similar as for the de Sitter space, we can obtain the two-point function 〈φ(X )φ(Y )〉 by exploiting
the conformal symmetry of AdS. The Euclidean Green’s function denoted by the propagator Π(X , Y ),
has to obey,
□X − m2 Π(X , Y ) = −δ(X , Y ) .
 
(10.21)

From the symmetry of the problem it is clear that the propagator can only depend on the invariant
X · Y or equivalently on the chordal distance ζ = (X − Y )2 /L 2 . From now on we will set L = 1 and
all lengths will be expressed in units of the AdS radius.

107
Exercise 10.3. Use (10.10) and (10.12) to show that

C∆ d 1 4
 ‹
Π(X , Y ) = ∆ 2 F1 ∆, ∆ − + , 2∆ − d + 1, − , (10.22)
ζ 2 2 ζ
p
where 2∆ = d + d 2 + (2m)2 and

Γ (∆)
C∆ = d
. (10.23)
d
2π Γ ∆ −
2
2 +1

For a free field, higher point functions are simply given by Wick contractions. For example,

φ(X 1 )φ(X 2 )φ(X 3 )φ(X 4 ) = Π(X 1 , X 2 )Π(X 3 , X 4 ) + Π(X 1 , X 3 )Π(X 2 , X 4 )

+ Π(X 1 , X 4 )Π(X 2 , X 3 ) . (10.24)

Weak interactions of φ can be treated perturbatively. Suppose the action includes a cubic term,
Z
1 1 2 2 1
• ˜
2 3
S= dX (∇φ) + m φ + gφ . (10.25)
AdS 2 2 3!

Then, there is a non-vanishing three-point function

Z
〈φ(X 1 )φ(X 2 )φ(X 3 )〉 = −g d Y Π(X 1 , Y )Π(X 2 , Y )Π(X 3 , Y ) + O(g 3 ) ,
AdS

and a connected part of the four-point function of order g 2 . This is very similar to perturbative QFT
in flat space.

10.3 Towards a conformal theory on the boundary

Given a correlation function in AdS we can consider the limit where we send all points to infinity.
More precisely, we introduce

1
O(P) = p lim λ∆ φ (X = λP + . . . ) , (10.26)
C∆ λ→∞

where P is a future directed null vector in Rd+1,1 and the . . . denote terms that do not grow with λ
whose only purpose is to enforce the AdS condition X 2 = −1. In other words, the operator O(P) is
the limit of the field φ(X ) when X approaches the boundary point P of AdS. Correlation functions of
O are naturally defined by the limit of correlation functions of φ in AdS. For example, the two-point
function is given by
1
〈O(P1 )O(P2 )〉 = ∆
+ O(g 2 ) , (10.27)
(−2P1 · P2 )

108
which is exactly the CFT two-point function of a primary operator of dimension ∆. The three-point
function is given by
Z
−3
〈O(P1 )O(P2 )O(P3 )〉 = − g C∆ 2 d X Π(X , P1 )Π(X , P2 )Π(X , P3 ) + O(g 3 ) , (10.28)
AdS

where
C∆
Π(X , P) = lim λ∆ Π (X , Y = λP + . . . ) = (10.29)
λ→∞ (−2P · X )∆
is the bulk to boundary propagator.

Exercise 10.4. Write the bulk to boundary propagator in Poincaré coordinates.

Exercise 10.5. Compute the following generalization of the integral in (10.28),

Z 3
Y 1
dX , (10.30)
AdS i=1 (−2Pi · X )∆i

and show that it reproduces the expected formula for the CFT three-point function 〈O1 (P1 )O2 (P2 )O3 (P3 )〉.
It is helpful to use the integral representation
Z ∞
1 1 ds ∆ 2sP·X
∆
= s e (10.31)
(−2P · X ) Γ (∆) 0 s

to bring the AdS integral to the form Z

d X e2X ·Q (10.32)
AdS

with Q a future directed timelike vector. Choosing the X 0 direction along Q and using the Poincaré
coordinates (E.42) it is easy to show that
Z Z ∞
2X ·Q d dz − d −z+Q2 /z
dX e =π 2 z 2e . (10.33)
AdS 0 z

To factorize the remaining integrals over s1 , s2 , s3 and z it is helpful to change to the variables t 1 , t 2 , t 3
and z using p
z t1 t2 t3
si = . (10.34)
ti

State-Operator Map

We have seen that the correlation functions of the boundary operator (10.26) have the correct
homogeneity property and SO(d + 1, 1) invariance expected of CFT correlators of a primary scalar
operator with scaling dimension ∆. We will now argue that they also obey an associative OPE. The
argument is very similar to the one used in CFT. We think of the correlation functions as vacuum
expectation values of time ordered products

〈φ(X 1 )φ(X 2 )φ(X 3 ) . . . 〉 = 〈0| . . . φ̂(τ3 , ρ3 , Ω3 )φ̂(τ2 , ρ2 , Ω2 )φ̂(τ1 , ρ1 , Ω1 ) |0〉 ,

109
where we assumed τ1 < τ2 < 0 < τ3 < . . . . We then insert a complete basis of states at τ = 0,

〈φ(X 1 )φ(X 2 )φ(X 3 ) . . . 〉 (10.35)

X
= 〈0| . . . φ̂(τ3 , ρ3 , Ω3 ) |ψ〉 〈ψ| φ̂(τ2 , ρ2 , Ω2 )φ̂(τ1 , ρ1 , Ω1 ) |0〉 .
ψ

Using φ̂(τ, ρ, Ω) = eτD φ̂(0, ρ, Ω)e−τD and choosing an eigenbasis of the Hamiltonian D = − ∂∂τ it is
clear that the sum converges for the assumed time ordering. The next step, is to establish a one-to-one
map between the states |ψ〉 and boundary operators. It is clear that every boundary operator (10.26)
defines a state. Inserting the boundary operator at P A = P 0 , P µ , P d+1 = 12 , 0, 12 , which is the

boundary point defined by τ → −∞ in global coordinates, we can write

〈. . . φ(X 3 )O(P)〉 = 〈0| . . . φ̂(τ3 , ρ3 , Ω3 ) |O〉 , (10.36)

where

∆
|O〉 = lim e−τ cosh ρ φ̂(τ, ρ, Ω) |0〉 (10.37)
τ→−∞
X
= |ψ〉 (cosh ρ)∆ lim 〈ψ| eτ(D−∆) φ̂(0, ρ, Ω) |0〉 .
τ→−∞
ψ

The limit τ → −∞ projects onto the primary state with wave function (10.15).
The map from states to boundary operators can be established using global time translation invariance,

〈0| . . . φ̂(τ3 , ρ3 , Ω3 ) |ψ(0)〉 (10.38)

= lim 〈0| . . . φ̂(τ3 , ρ3 , Ω3 )eτD |ψ(τ)〉 ≡ . . . φ(X 3 )Oψ (P)
τ→−∞

where |ψ(τ)〉 = e−τD |ψ〉 and P A = 21 , 0, 12 is again the boundary point defined by τ → −∞ in global

coordinates. The idea is that |ψ(τ)〉 prepares a boundary condition for the path integral on a surface
of constant τ and this surface converges to a small cap around the boundary point P A = 12 , 0, 12

when τ → −∞. This is depicted in figure 10.3.

The Hilbert space of the bulk theory can be decomposed in irreducible representations of the isometry
group SO(d + 1, 1). These are the highest weight representations of the conformal group, labelled
by the scaling dimension and SO(d) irrep of the the primary state. Therefore, the CFT conformal
block decomposition of correlators follows from the partial wave decomposition in AdS, i.e. the
decomposition in intermediate eigenstates of the Hamiltonian organized in irreps of the isometry
group SO(d + 1, 1). For example, the conformal block decomposition of the disconnected part of the
four-point function,

1 1 1
O(P1 ) . . . O(P4 ) =
∆ +
∆ +
∆ , (10.39)
P12 P34 P13 P24 P14 P23

where Pi j = −2Pi · P j , is given by a sum of conformal blocks associated with the vacuum and

110
 τ = −∞ τ =∞

Figure 10.3: Curves of constant τ (in blue) and constant ρ (in red) for AdS2 stereographically
projected to the unit disk (Poincaré disk). This shows how surfaces of constant τ converge to
a boundary bound when τ → −∞. The cartesian coordinates in the plane of the figure are
(cosh ρ sinh τ,sinh ρ) ⃗2
4d w
⃗ = 1+cosh ρ cosh τ which puts the AdS2 metric in the form ds2 = 1−
given by w ⃗2
w
.

two-particle intermediate states

∞ X
X ∞
O(P1 ) . . . O(P4 ) = G0,0 (P1 , . . . , P4 ) + cn,l G2∆+2n+l,l (P1 , . . . , P4 ) .
l=0 n=0
even

Exercise 10.6. Check this statement in d = 2 using the formula [DO01]

k(E + l, z)k(E − l, z̄) + k(E − l, z)k(E + l, z̄)

G E,l (P1 , P2 , P3 , P4 ) =
∆ (10.40)
(−2P1 · P2 )∆ −2P3 · P4


1 + δl,0

where
k(2β, z) = (−z)β 2 F1 (β, β, 2β, z) . (10.41)

Determine the coefficients cn,l for n ≤ 1 by matching the Taylor series expansion around z = z̄ = 0.
Extra: using a computer you can compute many coefficients and guess the general formula.

Generating function

There is an equivalent way of defining CFT correlation functions from QFT in AdS. We introduce the
generating function ¬ R ¶
W [φ b ] = e ∂ AdS d Pφ b (P)O(P) , (10.42)

where the integral over ∂ AdS denotes an integral over a chosen section of the null cone in Rd+1,1
with its induced measure. We impose that the source obeys φ b (λP) = λ∆−d φ b (P) so that the integral
is invariant under a change of section, i.e. conformal invariant. For example, in the Poincaré section
R
the integral reduces to d d xφ b (x)O(x). Correlation functions are easily obtained with functional

111
derivatives
δ δ
〈O(P1 ) . . . O(Pn )〉 = ... W [φ b ] . (10.43)
δφ b (P1 ) δφ b (Pn ) φ b =0

If we set the generating function to be equal to the path integral over the field φ in AdS

[dφ] e−S[φ]
R
φ→φ b
W [φ b ] = R , (10.44)
φ→0
[dφ] e−S[φ]

with the boundary condition that it approaches the source φ b at the boundary,

1 1
lim λd−∆ φ(X = λP + . . . ) = p φ b (P) , (10.45)
λ→∞ 2∆ − d C∆

then we recover the correlation functions of O defined above as limits of the correlation functions of
φ.
For a quadratic bulk action, the ratio of path integrals in (10.44) is given e−S computed on the classical
solution obeying the required boundary conditions. A natural guess for this solution is

φ b (P)
Z
φ(X ) =
p
C∆ dP . (10.46)
∂ AdS (−2P · X )∆

This automatically solves the AdS equation of motion ∇2 φ = m2 φ, because it is an homogeneous

function of weight −∆ and it obeys ∂A∂ Aφ = 0 in the embedding space (see equations (10.10) and
(10.12)). To see that it also obeys the boundary condition (10.45) it is convenient to use the Poincaré
section.

Exercise 10.7. In the Poincaré section (G.31) and using Poincaré coordinates (E.42), formula (10.46)
reads
z ∆ φ b ( y)
Z
φ (z, x) = C∆ d d y
p
(10.47)
(z 2 + (x − y)2 )∆
and (10.45) reads
1 1
lim z ∆−d φ(z, x) = p φ b (x) . (10.48)
z→0 2∆ − d C∆
Show that (10.48) follows from (10.47). You can assume 2∆ > d.

1
The cubic term 3! gφ 3 in the action will lead to (calculable) corrections of order g in the classical
solution (10.46). To determine the generating function W [φ b ] in the classical limit we just have to
compute the value of the bulk action (10.25) on the classical solution. However, before doing that,
we have to address a small subtlety. We need to add a boundary term to the action (10.25) in order
to have a well posed variational problem.

Exercise 10.8. The coefficient β should be chosen such that the quadratic action 2

p p
Z Z
1 1 2 2
• ˜
S2 = dw G 2
(∇φ) + m φ + β d w G ∇α (φ∇α φ) (10.49)
AdS 2 2 AdS
2
Here w stands for a generic coordinate in AdS and the index α runs over the d + 1 dimensions of AdS.

112
is stationary around a classical solution obeying (10.48) for any variation δφ that preserves the boundary
condition, i.e.
δφ(z, x) = z ∆ [ f (x) + O(z)] . (10.50)
∆−d
Show that β = d and that the on-shell action is given by a boundary term
Z
2∆ − d p
S2 = d w g ∇α (φ∇α φ) . (10.51)
2d AdS

Finally, show that for the classical solution (10.47) this action is given by 3
Z
1
S2 = − d d y1 d d y2 φ b ( y1 )φ b ( y2 )K( y1 , y2 ) , (10.52)
2

where

z∆ z∆
Z
2∆ − d dd x
K( y1 , y2 ) = C∆ lim ∆
∂z
d z→0 z d−1 (z 2 + (x − y1 )2 ) (z 2 + (x − y2 )2 )∆
1
= (10.53)
( y1 − y2 )2∆

is the CFT two point function (10.27).

Exercise 10.9. Using φ = φ0 + O(g) with φ0 given by (10.46), show that the complete on-shell action
is given by
Z Z
1 1
S=− d d
d y1 d y2 φ b ( y1 )φ b ( y2 )K( y1 , y2 ) + g d X [φ0 (X )]3 + O(g 2 ) ,
2 3! AdS

and that this leads to the three-point function (10.28).

Extra: Compute the terms of O(g 2 ) in the on-shell action.

10.4 The AdS/CFT correspondence

We have seen that QFT on an AdS background naturally defines conformal correlation functions
living on the boundary of AdS. Moreover, we saw that a weakly coupled theory in AdS gives rise to
factorization of CFT correlators like in a large N expansion. However, there is one missing ingredient
to obtain a full-fledged CFT: a stress-energy tensor.
To add this will radically change the theory though, as it turns out this requires us to make gravity
dynamical in the bulk. Indeed, if we have some conserved current Jµ for a global current in the
boundary theory this couples to bulk fields as
Z
S=S+ J µ Aµ , (10.54)
∂ AdS
3
This integral is divergent if the source φ b is a smooth function and ∆ > d2 . The divergence comes from the short
distance limit y1 → y2 and does not affect the value of correlation functions at separate points. Notice that a small value of
z > 0 provides a UV regulator.

113
where Aµ becomes a dynamic gauge field in the bulk. Repeating this for the stress tensor we find that
it couples as
Z
S=S+ T µν gµν , (10.55)
∂ AdS

where the symmetric tensor gµν is the dynamic bulk metric! Having a stress tensor on the boundary
therefore automatically induces dynamical gravity in the bulk!
The next exercise also shows that a free QFT in AdSd+1 can not be dual to a local CFTd .

Exercise 10.10. Compute the free-energy of a gas of free scalar particles in AdS. Since particles are free
and bosonic one can create multi-particle states by populating each single particle state an arbitrary
number of times. That means that the total partition function is a product over all single particle states
and it is entirely determined by the single particle partition function. More precisely, show that
‚∞ Œ ∞
Y X
kEψsp
X 1
F = −T log Z = −T log q = −T Z1 (q n ) , (10.56)
ψsp k=0 n=1
n
X
Eψsp q∆
Z1 (q) = q = , (10.57)
ψsp
(1 − q)d

where q = e− RT and we have used the single-particle spectrum of the hamiltonian D = − ∂∂τ of AdS in
1

global coordinates. Show that

F ≈ −ζ(d + 1)Rd T d+1 (10.58)

in the high temperature regime and compute the entropy using the thermodynamic relation S = − ∂∂ TF .
Compare this result with the expectation

S ∼ (RT )d−1 , (10.59)

for the high temperature behaviour of the entropy of a CFT on a sphere S d−1 of radius R. See section 4.3
of reference [ESP12] for more details on this point.

Gravity with AdS boundary conditions

Consider general relativity in the presence of a negative cosmological constant

p
Z
1
I[G] = d d+1 w G [R − 2Λ] . (10.60)
ℓd−1
P

The AdS geometry

dz 2 + d x µ d x µ
ds2 = R2 , (10.61)
z2
d(d−1)
is a maximally symmetric classical solution with Λ = − 2R2 . When the AdS radius R is much larger
than the Planck length ℓ P the metric fluctuations are weakly coupled and form an approximate Fock
space of graviton states. One can compute the single graviton states and verify that they are in one-
to-one correspondence with the CFT stress-tensor operator and its descendants (with AdS energies

114
matching scaling dimensions). One can also obtain CFT correlation functions of the stress-energy
tensor using Witten diagrams in AdS. The new ingredients are the bulk to boundary and bulk to bulk
graviton propagators [LT98, LT99, DFM+ 99a, DFM+ 99b, CGP14].
In the gravitational context, it is nicer to use the partition function formulation
Z
Z[gµν , φ b ] = [d G] [dφ] e−I[G,φ] (10.62)
G→g
φ→φ b

where
p
Z
1 1 1 2 2
• ˜
2
I[G, φ] = d d+1
w G R − 2Λ + (∇φ) + m φ (10.63)
ℓd−1
P
2 2
and the boundary condition are

µ ν
 
dz 2
+ d x d x g µν (x) + O(z)
ds2 = Gαβ d wα d wβ = R 2
, (10.64)
z2
z d−∆
φ = [φ b (x) + O(z)] .
2∆ − d

By construction the partition function is invariant under diffeomorphisms of the boundary metric gµν .
Therefore, this definition implies the Ward identity (G.19). The generating function is also invariant
under Weyl transformations

Z Ω2 gµν , Ω∆−d φ b = Z gµν , φ b

   
(nai ve) (10.65)

This follows from the fact that the boundary condition

µ ν
 2 
dz 2
+ d x d x Ω (x)g µν (x) + O(z)
ds2 = R 2
(10.66)
z2
z d−∆  ∆−d 
φ = Ω (x)φ b (x) + O(z)
2∆ − d

can be mapped to (10.64) by the following coordinate transformation

1
2
z → z Ω − z 3 Ω ∂µ log Ω + O(z 5 ) (10.67)
4
1
xµ → x µ − z 2 ∂ µ log Ω + O(z 4 )
2
where indices are raised and contracted using the metric gµν and its inverse. In other words, a bulk
geometry that satisfies (10.64) also satisfies (10.66) with an appropriate choice of coordinates. If the
partition function (10.62) was a finite quantity this would be the end of the story. However, even
in the classical limit, where Z ≈ e−I , the partition function needs to be regulated. The divergences
originate from the z → 0 region and can be regulated by cutting off the bulk integrals at z = ε (as it
happened for the scalar case discussed above). Since the coordinate transformation (10.67) does
not preserve the cutoff, the regulated partition function is not obviously Weyl invariant. This has
been studied in great detail in the context of holographic renormalisation [?, ?]. In particular, it leads
to the Weyl anomaly g µν Tµν ̸= 0 when d is even. The crucial point is that this is a UV effect that

115
does not affect the connected correlation functions of operators at separate points. In particular, the
integrated form (G.22)=(G.24) of the conformal Ward identity is valid.
We do not now how to define the quantum gravity path integral in (10.62). The best we can do is a
semiclassical expansion when ℓ P ≪ R. This semiclassical expansion gives rise to connected correlators
of the stress tensor Tµν that scale as
 ‹d−1
R
Tµ1 ν1 (x 1 ) . . . Tµn νn (x n ) ∼ . (10.68)
c ℓP
€ Šd−1
This is exactly the scaling (G.47) we found from large N factorization if we identify N 2 ∼ ℓR .
P
This suggests that CFTs related to semiclassical Einstein gravity in AdS, should have a large number
of local degrees of freedom. This can be made more precise. The two-point function of the stress
tensor in a CFT is given by

CT 1 1 1 1
• ˜
Tµν (x)Tσρ (0) = I I
µσ νρ + I I
µρ νσ − δ δ
µν σρ , (10.69)
Sd2 x 2d 2 2 d

2πd/2
where Sd = Γ (d/2) is the volume of a (d − 1)-dimensional unit sphere and

xµ xν
Iµν = δµν − 2 . (10.70)
x2

The constant C T provides an (approximate) measure of the number of degrees of freedom.4 For
instance, for nϕ free scalar fields and nψ free Dirac fields we find [?]

d
+ nψ 2[ 2 ]−1 d ,
d
C T = nϕ (10.71)
d −1

where [x] is the integer part of x. If the CFT is described by Einstein gravity in AdS, we find [?]
d
d + 1 π 2 Γ (d + 1) Rd−1
CT = 8 , (10.72)
d − 1 Γ3 d ℓd−1


P 2

which shows that the CFT dual of a semiclassical gravitational theory with R ≫ ℓ P , must have a very
large number of degrees of freedom.
In summary, semiclassical gravity with AdS boundary conditions gives rise to a set of correlation
functions that have all the properties (conformal invariance, Ward identities, large N factorization)
expected for the correlation functions of the stress tensor of a large N CFT. Therefore, it is natural to
ask if a CFT with finite N is a quantum theory of gravity.

The AdS/CFT dictionary

Work in progress...
4
However, for d > 2, C T is not a c-function that always decreases under Renormalization Group flow.

116
10.5 Hawking radiation in AdS

In this last section, which is left entirely as an exercise, we derive the Hawking effect for black holes
in AdS space.
Exercise 10.11. Consider the following metric for a black hole in AdSd+1 space,

dr
ds2 = f (r)dt 2 − + r 2 dΩ2d−1 , (10.73)
f (r)
µ
where f (r) = r 2 + 1 − r d−2
. The mass is related to the parameter µ as
 ‹
2−d d
(d − 1)µ = 8π 2 Γ M. (10.74)
2

Argue why this metric represents a black hole and show that asymptotically it reduces to a metric on
standard AdSd+1 .
To analyse the Hawking effect in this background, define the tortoise coordinate satisfying dr ∗ = fdr
(r)
which tends to −∞ near the horizon and approaches a constant value near the asymptotic boundary.
In AdS, it is natural to impose normalizable boundary conditions on the field. Given a scalar field, φ,
p
of mass m, we demand that, as r → ∞, the field dies off as r ∆+ , where 2∆+ = d + d 2 + 4m2 . The
relevance of this boundary condition is that it relates the left- and right-moving modes near the boundary.
Show that this implies that at late times we can expand the field as

XZ
φ= dωaωl f (ω, l, r ∗ )e−iωt Yl (Ω) + h.c. . (10.75)
l

Notice that the boundary condition at conformal infinity reduces the number independent waves to one
’standing wave’.
How does the function f behave near the horizon and near the asymptotic boundary?
The analysis near the horizon is identically as in flat space. Define Kruskal coordinates and expand the
scalar field near the horizon.
Use all this to compute the Bogoliubov coefficients and argue that, just like in the asymptotically flat
Schwarzschild black hole, the modes outside of an AdS black hole must be populated thermally. Note
however that, due to the reflective boundary conditions in AdS, this time there is no flux near the boundary
since the global modes multiply standing waves. For this reason black holes in AdS do not evaporate and
can be in thermal equilibrium with a thermal bath.
Exercise 10.12. An alternative way to detect the thermal nature of black hole backgrounds is to study
the periodicity in Euclidean time. Show that for a generic black hole background with metric (10.73),
the absence of conical singularities predicts a Hawking temperature T = β1 with

4π
β= . (10.76)
V ′ (rh )

Hint: expand the metric around the horizon and rewrite in the standard metric on R1,1 × S d−1 .

117
Use this to verify your computation of the Hawking temperature of the AdS Schwarzschild black hole in
the previous exercise.

118
 Part III

APPENDICES

119
 Appendix A

Conventions

h = c = GN = kB = 1. In these units the

In these notes we use natural (or Planck) units in which ħ
natural scales are given by

Quantity Expression Metric value

Ç
h GN
ℓP = 1.616 · 10−35 m
ħ
Length
Ç c3
Mass m P = GħhNc 2.176 · 10−8 kg
Ç
h GN
tP = 5.391 · 10−44 s
ħ
Time
r c 55
Temperature TP = Għhck2 1.417 · 1032 K
N B

Using these normalised units, the cosmological constant of our observable universe is Λ ∼ 2.888 ·
10−122 ℓ−2
P .

These notes we employ a plethora of indices, each with its own meaning. The various uses of indices
are summarised in Table A.1 below.

Index Range Meaning

µ, ν, . . . 0, . . . , d Curved spacetime indices
m, n, . . . 1, . . . , d Space-like curved spacetime indices
M, N, . . . 0, . . . , d + 1 Embedding space curved indices
a, b, . . . 1, . . . , d Tangent space indices
α, β, . . . 1, 2 SU(2) L indices
α̇, β̇, . . . 1, 2 SU(2)R indices

Table A.1: The various indices used in these lecture notes. d is the dimension of spacetime.
When considering four-dimensional spacetime we sometimes employ the exceptional isomorph-
ism so(4) = su(2) L × su(2)R .

A.1 Signs, signatures and curvature

Let (M, gµν ) be our space-time. For the most part, we will take to be a four-dimensional manifold
with metric gµν and here we restrict to this situation only. As usual curved indices are raised and
lowered with respectively the metric and its inverse g µν while flat tangent space indices are raised
and lowered with the Minkowski metric ηab and its inverse.
A second set of conventions and possible source of confusion is related to the signature of spacetime
and the curvature tensor. In order to easily compare with the literature we keep all the signs explicit

120
in this appendix while in the main text we fix all signs to be one,

s1 = s2 = s3 = s4 = s5 = 1 . (A.1)

The first choice of sign comes from the signature of the metric, which can be either mostly plus or
mostly minus,
ηa b = s1 diag (+, −, −, −) . (A.2)

A second sign appears in the definition of the Riemann tensor,

€ Š
Rµν ρ σ = −s2 ∂µ Γνσ
ρ ρ
− ∂ν Γµσ ρ τ
+ Γµτ ρ τ
Γνσ − Γντ Γµσ . (A.3)

A third sign appears in the definition of the Ricci tensor

s2 s3 Rµν = Rρ νρµ , R = g µν Rµν . (A.4)

This sign in turn gives rise to a sign in the Einstein equation

1
 ‹
s3 Rµν − gµν R = −κ2 Tµν , (A.5)
2

where by definition, T00 is always positive and κ2 = 8πGN . The signs s1 and s3 determine the signs
of the kinetic terms of scalars and gravitons

1 1 1
L = s1 ∂µ φ∂ µ φ − Fµν F µν + s1 s3 2 R . (A.6)
2 4 2κ

Hence, looking at the Lagrangian one can easily recognize the values of these signs.
When working with frame fields the curvature can also be obtained from the spin connection ωµ ab .
The usual convention is that it is related to the curvature defined above in terms of the Christoffel
symbols as
Rρσ µ ν (Γ ) = Rρσ ab (ω)eaµ eνb . (A.7)

However, there is an independent sign in

€ Š
Rµν a b = −s4 ∂µ ωνab − ∂ν ωµab + ωµac ωνc
b
− ωνac ωµc
b
. (A.8)

This sign is relevant when considering the covariant derivatives of vectors and spinors, which are
given by
1
 ‹
∇µ ψ = ∂µ + s2 s4 ωµ a b γab ψ , ∇µ V a = ∂µ V a + s2 s4 ωµ ab Vb . (A.9)
4
Furthermore we always (anti-)symmetrize with weight one, i.e.

1 1
A[a b] = (Aa b − A ba ) , A(ab) = (Aab + A ba ) . (A.10)
2 2

In some references the factor of 2 is omitted. Finally, the Levi-Civita tensor is defined as ε0123 = s5
and ε0123 = −s5 .

121
To illustrate all these conventions above here are some useful formulae which depend on the choices
of sign

ρ a
∇µ eνa = ∂µ eνa + s2 s4 ωµ ab eνb − Γµν eρ = 0 , (A.11)
€ Š
ωµ a b = s2 s4 2eν[a ∂[µ eν] − eν[a e b]σ eµc ∂ν eσ
b] c
, (A.12)
∇µ , ∇ν Vρ = s2 Rσ ρµν Vσ
 
(A.13)

122
 Appendix B

Differential forms

Differential forms often simplify formulae both computationally and conceptually. In this appendix
we briefly review the essentials of this framework, for a more comprehensive treatment, see for
example [Nak90, BT13, Nab10].
On any manifold we can define the formal objects, dx µ , called differentials. The composition of such
differential forms is done through the exterior product and denoted by a wedge, ∧. This product is
associative and anti-symmetric,

1 X
dx µ1 ∧ . . . ∧ dx µp = dx [µ1 ∧ . . . ∧ dx µp ] ≡ (−1)|σ| dx µσ(1) ∧ . . . ∧ dx µσ(p) (B.1)
p! σ∈S
p

where |σ| denotes the signature of the permutation σ. We define a p-form as an element of the linear
vector space ∧ p (M) spanned by the the external composition of p differentials. Any p-form can thus
be represented as a homogeneous polynomial of degree p in the exterior product of differentials,

α = αµ1 µ2 ...µp dx µ1 ∧ . . . dx µp = α[µ1 µ2 ...µp ] dx µ1 ∧ . . . dx µp ∈ ∧ p (M) . (B.2)

In a d-dimensional manifold, the direct sum of vector spaces ∧(M) = dp=0 ∧ p (M) is called the
L

exterior algebra. In the exterior algebra, the exterior product is a map ∧(M) × ∧(M) → ∧(M)
defined as
α ∧ β ≡ α[µ1 ...µp βµp+1 ...µp+q ] dx µ1 ∧ . . . ∧ dx µp+q ∈ ∧ p+q (M) , (B.3)

where α is a p-form and β a q-form. This product is graded commutative

α ∧ β = (−1) pq β ∧ α . (B.4)

We also have the exterior derivative defined by

dα ≡ ∂[µ1 αµ2 ...µp+1 ] dx µ1 ∧ . . . ∧ dx µp+1 ∈ ∧ p+1 (M) . (B.5)

Key features of the exterior derivative are:

Lemma B.1. The exterior derivative does not depend on the choice of torsion-free covariant derivative.
We have d 2 α = 0 for all α as a consequence of the commutation of partial derivatives (or symmetry of a
torsion-free connection).

Thus it is metric independent and can be defined just using the coordinate derivative in any coordinate

123
system. The fact that d 2 = 0 allows us to define cohomology groups

H p (M ) = {α ∈ Ω p |dα = 0}/{α = dβ} , (B.6)

since the exact forms, i.e. those that can be expressed as dβ, are a subset of the closed forms, those
that satisfy dα = 0. Such cohomology groups encode important information about the topology of M
because dα = 0 implies that locally there exists a β with α = dβ (Poincaré lemma).

Example B.1. As an example, consider the circle S 1 . Since the circle is connected, every two points are
connected by a segment and are cohomologically equivalent. Indeed, this implies that H 0 (S 1 ) = R which
remains to be true for any connected manifold. Next, let us compute H 1 (S 1 ). Consider a generic one-form
ω = f (θ )dθ ∈ Ω1 (S 1 ). This form is clearly closed so we are left to investigate whether it is exact, i.e. if
we can find a globally well-defined function F such that ω = dF . Locally it is easy to see that we can find
such a function,
Zθ
F (θ ) = f (θ ′ )dθ ′ . (B.7)
0

In order for F to be globally well-defined we need to impose that F (2π) = 0. Defining the function
Z 2π
1 1
λ : Ω (S ) → R : ω = f (θ )dθ 7→ f (θ ′ )dθ ′ , (B.8)
0

it is easy to see that the first cohomology group is given by

H 1 (S 1 ) = Ω1 (S 1 )/ ker λ = im λ = R . (B.9)

The exterior derivative satisfies the graded Leibniz rule

d(α ∧ β) = (dα) ∧ β + (−1) p α ∧ dβ . (B.10)

Furthermore, we can also define the interior product with a vector V a that takes a p-form α to the
p − 1-form1
(V ⌟α)a2 a3 ...ap = pV a1 αa1 ...ap . (B.11)

This also satisfies a graded Leibniz property,

V ⌟(α ∧ β) = (V ⌟α) ∧ β + (−1) p α ∧ (V ⌟β) . (B.12)

It plays a role in the Cartan formula for the Lie derivative of a form

LV α = V ⌟dα + d(V ⌟α) . (B.13)

When we have a metric, we can define Hodge duality: in d dimensions a p-form α is dualized to a
d − p form ∗ α by
1
(∗ α)ap+1 ...ad := ϵa1 ...ad αa1 ...ap (B.14)
p!
1
Another common notation for the inner product is given by ιV α = V ⌟α.

124
 p
where ϵa1 ...ad = ϵ[a1 ...ad ] and ϵ0 1...d−1 = −g is the metric volume form.
A key application is to integration. Being a covariant tensor, a p-form naturally ‘pulls back’ under a
map, and restricts to provide a p-form on a submanifold. On a p-dimensional submanifold, it can
naturally be integrated subject to the choice of an orientation on the surface.

Definition B.1. A p-surface Σ p is said to be orientable if it is possible to choose a non-vanishing

p-form. Such a choice provides an orientation on Σ p .

The key point is that under a change of coordinates on the p-surface Σ p , a p-form transforms with
the determinant of the Jacobian of the coordinate transformation, whereas the change of variables
formula for integration requires the modulus of the determinant which can introduce additional signs,
and so we must restrict the coordinate transformations to those that preserve the sign of the chosen
form making sure that the sign in question is positive.2 The standard example of a non-orientable
manifold is RP2n = S 2n /Z2 where the Z2 acts by the antipodal map which reverses the sign of the
volume form.
The main theorem concerning integration on manifolds is Stoke’s theorem:

Theorem B.2 (Stokes). Let Σ be a p-surface with boundary S with compatible orientations (i.e., the
orientation on S is obtained from that on Σ by use of an outward pointing normal vector), and let α be a
p − 1-form on Σ, then Z Z
dα = α. (B.15)
Σ S

Another application is the Cartan formulation of connections and curvature.

B.1 Connections and curvature

Instead of working with the metric, it is often useful to define a orthonormal frame of one-forms, or
vielbeine, e a = eµa dx µ satisfying
gµν = ηab eµa eνb , (B.16)
µ
where ηa b = diag(1, −1, · · · , −1) is the flat Lorentz metric. The vielbeine eµa and their inverses ea can
be used to freely convert curved spacetime indices to flat tangent space indices. Note that the global
structure of spacetime manifolds does not always allow the vielbeine to be chosen globally. In other
words, generic spacetimes do not admit a global framing. In general this description is only valid
locally. However, for globally hyperbolic spacetimes with orientable spatial slices, it is valid globally.
The connection acting on this frame can be obtained from the Cartan structural equation

de a + ωa b ∧ e b = 0 , (B.17)
2
The issue is seen in one dimension: under the transformation y = −x,
Z b Z −b Z −a

f (x)d x = − f (− y)d y = f (− y)d y ,

a −a −b

so that there is no sign change if we are to integrate from the lower limit to the upper in each case.

125
where
ωa b = ω[ab] = dx µ ωµ ab , (B.18)

is the 1-form spin connection.3 In terms of the spin connection, we can define the curvature 2-form,

R a b = dx µ ∧ dx ν Rµνa b = dωa b + ωa c ∧ ωc b . (B.19)

Consistency then requires that this form satisfies the Bianchi identities

Ra b ∧ e b = 0 , dRa b + ωa c ∧ Rc b − Ra c ∧ ωc b = 0 . (B.20)

For a general 1-form we can then write the covariant derivative as

∇a A b = (∂a A b − ωa c b Ac ) , (B.21)

and similar for higher forms. In addition, this formulation allows us to consider spinors in general
spacetimes. For more on this see Appendix ??.

3
Note that here, as always in this course, we assume the connection to be torsionless. In the presence of torsion, (B.17)
has to be modified to de a + ωa b ∧ e b = Θ, where Θ is the torsion 2-form. Similarly, in the presence of torsion the Bianchi
identities have to be modified.

126
 Appendix C

Hypersurfaces

This appendix reviews some useful facts on hypersurfaces which will be useful in the computations
done in this course. More details and examples can be found in the book [Poi04].
Let M be a (d + 1)-dimensional Lorentzian manifold. A hypersurface Σ can be defined by parametric
equations of the form
x µ = x µ( y p ) , (C.1)

where x are coordinates on M and y p , p = 1, . . . d are intrinsic coordinates on Σ. Alternatively, the

hypersurface can be defined by implicit equations

Φ(x µ ) = 0 . (C.2)

Exercise C.1. As in standard Euclidean geometry, show that the vector ∂µ Φ is always normal to the
hypersurface.

A hypersurface is null if g µν ∂µ Φ∂ν Φ = 0 and space/time-like if the vectors in the tangent space at
each point are space/time-like. When the hypersurface is not null, we can introduce the unit normal
vector, defined by
+1 , if Σ is space-like,

µ
n nµ = ε = (C.3)
−1 if Σ is time-like.

When the hypersurface is defined implicitly, the normal is proportional to ∂µ Φ. By definition, the
normal is pointed in the direction of increasing Φ, i.e. nµ ∂µ Φ > 0.

Exercise C.2. Show that the normal can be written as

ε ∂µ Φ
nµ = . (C.4)
|g µν ∂µ Φ∂ν Φ|

Define now the tangent vectors

xµ
Eµ p = , nµ E µ p = 0 . (C.5)
∂ ya

The pull-back of the metric to Σ is then given by

ds2 Σ
= gµν E µ p E ν q d y p d y q = −εh pq d y p d y q . (C.6)

This defines the so-called induced metric, or first fundamental form, h pq . For non-null surfaces we

127
can then define the surface element as

dΣ = |h|1/2 dd y , dΣµ = εnµ dΣ . (C.7)

One then has the following Lorentzian (or Pseudo-Riemannian) version of Stokes’s theorem,
Z Z
µ
dΣµ Aµ .
Æ
d+1
d x |g|∇µ A = (C.8)
M ∂M

Exercise C.3. Show that the ambient metric g µν , when restricted to Σ, can be decomposed as

g µµ = ε nµ nν − h pq E µ p E ν q .


(C.9)

The second fundamental form, or extrinsic curvature, is defined as

K pq = ∇ν nµ E µ p E ν q . (C.10)

Exercise C.4. Show that K pq defined in (C.10) is a symmetric tensor and can furthermore be written as

1
K pq = Ln gµν E µ p E ν q , (C.11)
2
where Ln is the Lie derivative along the normal vector n.

The trace of the extrinsic curvature is given by

K = h pq K pq = (nµ nν − εg µν ) ∇ν nµ = h pq E µ p E ν q ∇ν nµ . (C.12)

Exercise C.5. As an example, and to get familiar with the concepts introduced above, consider the
spacetime M with metric,
ds2 = V (r)dt 2 − V (r)−1 dr 2 − r 2 dΩ2d−1 . (C.13)

and consider the hypersurface defined by r = constant.

1. Compute the tangent and normal vectors.

2. Compute the induced metric and extrinsic curvature.

128
 Appendix D

Variational calculus

To keep the discussion in these notes self-contained, this appendix includes a short discussion of
variational calculus, in particular as applied to the Einstein-Hilbert action. Even though we do not
discuss dynamical gravity in these notes this will be useful to revisit the general principles.
The Einstein–Hilbert action for the gravitational field in d + 1 dimensions is
Z
1 Æ
SEH = dd+1 x |g|R . (D.1)
16πGN

Its variation leads to the vacuum Einstein equations. After including additional matter fields these
give rise to a non-trivial stress-tensor. The derivation of Einstein’s equations is standard and can
be found in many textbooks. However, the standard derivations often do not carefully include the
contributions due to boundary terms. Such contributions are not so important if your main interest is
the study of the solutions to Einstein’s equations in classical GR. However, in a quantum theory, the
action becomes a crucial object since it gives the weight of a field configuration to the path integral.
For this reason let us revisit this derivation with particular care paid to the boundary terms.
Let us consider the gravitational action in a region M of space-time, with boundary ∂ M and analyze
the variation of the action as we vary the metric, with the condition that the metric variation vanishes
at the boundary:
δgµν = 0. (D.2)
∂M

To study the variation of the Einstein Hilbert action, note that

1Æ
|g|gµν δg µν ,
Æ
δ |g| = − (D.3)
2
such that €Æ Š Æ
|g|R = |g|Gµν δg µν + |g|g µν δRµν ,
Æ
δ (D.4)

where Gµν = Rµν − 21 gµν R is the Einstein tensor. The vanishing of the first term leads to Einstein’s
equations in empty space, while the last term is usually neglected by arguing that it vanishes on a
boundary at infinity. However, let us have a closer look at this term. Using a local frame where the
Christoffel symbols vanish, i.e. using Gauss’s normal coordinates, we have
€ Š
δRµν = δRρ µρν = δ ∂ν Γνρ
ρ ρ
− ∂ρ Γµν ρ
= ∇ν δΓµρ ρ
− ∇ρ δΓµν . (D.5)

This last expression is covariant and hence applies in any coordinate system. We can therefore write

g µν δRµν = ∇µ δv µ , δv µ = g µρ δΓρν
ν
− g ρν δΓρν
µ
. (D.6)

129
Using Stokes’ theorem we can then write
Z Z Z
µν µ
dΣµ δv µ .
Æ Æ
d+1
d x |g|g δRµν = d d+1
x |g|∇µ δv = (D.7)
M M ∂M

Let us now proceed to compute the variation of v. Since the variation of the metric, but not of its
derivatives, vanishes at the boundary, we find

µ 1 µν

δΓρσ = g δ∂σ gνρ + δ∂ρ gνσ − δ∂ν gρσ , (D.8)
2

and therefore we have

δvµ = g ρσ δ∂µ gρσ − δ∂ρ gµσ .


(D.9)

Finally, before we can plug this into (D.7), we need to contract this with the (ortho)normal vector n
to the boundary,

nµ δvµ = −nµ ε nρ nσ − h pq E ρ p E σ q



δ∂σ gµρ − δ∂µ gσρ . (D.10)

Now notice that the last bracket is anti-symmetric in µ and σ, while nµ nσ . Therefore, the part of
the metric which involves the normal vectors drops out. Next, we note that δ∂σ gνρ E σ p = 0, since
the variation of the metric vanishes everywhere on the boundary and therefore the variation on its
tangential derivatives has to vanish as well. Finally we have

nµ δvµ = −εhρσ δ∂µ gρσ nµ . (D.11)

This involves the derivative of the metric variation along the normal direction to the boundary so it is
in general non-vanishing.
Putting everything together, we find
Z Z
µν
|h|hµν δ∂ρ gµν nρ .
Æ Æ
16πGN δS EH = d d+1
x |g|Gµν δg − dd y (D.12)
M ∂M

Therefore, in the presence of a boundary, Einstein’s equations are not sufficient to guarantee the
vanishing of the variation of the action, due to the second term in (D.12). To remedy this, we must
add an explicit boundary term to the action,
Z
1 Æ
SGH = dd y |h|K , (D.13)
8πGN ∂M

called the Gibbons-Hawking counterterm. To see that this counterterm has the correct properties, let’s
have a quick look at its variation. Since the metric is fixed at the boundary the variation of h vanishes
and the only variation comes from the extrinsic curvature K. Computing its variation we find

1 µν
δK = hµν δ∂ν nµ = h δ∂ρ gµν nµ . (D.14)
2

130
Varying the Gibbons-Hawking counterterm we then find
Z
|h|hµν δ∂ρ gµν nρ ,
Æ
16πGN δSGH = dd y (D.15)
M

which exactly cancels the boundary term in the original variation and thus rendering the variational
principle well-defined.
In the context of black hole physics and AdS/CFT this is not always enough. In general, the action,
supplied with the Gibbons-Hawking counterterm, will lead to divergences when evaluated, even
in flat space. In order to obtain finite values for the on-shell action, it is usually necessary to add
additional counterterms removing the divergences. In the context of AdS/CFT the prescription to
remove said divergences goes under the name of holographic renormalisation.

131
 Appendix E

Ingredients from general relativity

In this appendix we review a variety of useful facts from general relativity that will come in handy in
this course.

E.1 Maximally symmetric spaces

Let us start with the maximally symmetric manifolds. In Euclidean signature these are given by the
sphere S d , flat Euclidean space Rd and hyperbolic space H d , which are respectively positively curved,
flat or negatively curved. We will sometimes collectively denote them by Mk,d , where k = 0, ±1 and
write their metrics as
2
k = 1,

 dΩ d
,
2 d
ds M = 2
k = 0,
P
i=1 dx i ,
(E.1)
k,d
2
k = −1 ,

dΞd ,

where dΩ2d and dΞ2d are the standard metrics on the d-sphere and d-dimensional hyperbolic space
respectively, !
d i−1 d
2
X Y
2 2 2 1 X 2
dΩd = sin φ j dφi , dΞd = 2 dx . (E.2)
i=1 j=1
x d i=1 i

There are many other coordinate choices but unless explicitly stated otherwise we will always use the
metrics above.
In most of this course we are interested in spaces with Minkowskian signature. The maximally
symmetric Lorentzian spacetimes are de Sitter space, Minkowski space and anti-de Sitter space, which
are respectively positively curved, flat or negatively curved.

de Sitter space

The de Sitter space is the maximally symmetric spacetime of positive curvature

d(d + 1)
R= , (E.3)
L2

where L is the characteristic length scale of the space. This space describes a exponentially expanding
universe. For this reason, there is an observer-dependent horizon, called the ‘cosmological horizon’
beyond which spacetime is expanding faster than the speed of light. This is a null surface beyond
which the observer can never receive a signal.
In d +1 dimensions de Sitter space can be described by a hypersurface in d +2 dimensional Minkowski

132
space. consider the embedding space R1,d+1 with metric,

d+1
X
ds2 = dX 02 − dX i2 . (E.4)
i=1

The de Sitter space of radius L is then defined as the hyperboloid

d+1
X
X µ X µ = X 02 − X i2 = −L 2 , (E.5)
i=1

In the same way that the two sphere embedded in R3 inherits the O(3) symmetry of its ambient space,
de Sitter space inherits an O(1, d) symmetry from the ambient Minkowski space. There are various
useful coordinate systems to describe the de Sitter space of which we list a few below:

• Global coordinates. These coordinates {τ, ωi } are defined by

τ τ
X 0 = L sinh , X i = Lωi cosh , (E.6)
L L

where i = 1, . . . d + 1 and the ωi are constrained coordinates on a round unit sphere S d such
that i ω2i = 1. In these coordinates, the metric on the de Sitter space reads,
P

τ
ds2 = dτ2 − L 2 cosh2 dΩ2d . (E.7)
L

Global coordinates are sometimes also called the closed slicing of de Sitter, especially in the
context of cosmology. This terminology comes from the FLRW metric (see below) since for
these coordinates the space-like slices are closed.

• Planar coordinates. These coordinates {t, x} are defined as

t x2 t
X 0 =L sinh + eL ,
L 2L
t x2 t (E.8)
X d+1 =L cosh − eL ,
L 2L
t
X i =x i e L ,

where i = 1, . . . , d. These coordinates do not cover the full de Sitter space but only the patch
t
X 0 + X d+1 = Le L > 0 . (E.9)

In these coordinates, the metric reads,

d
2t
X
ds2 = dt 2 − e L dx i2 . (E.10)
i=1

• Static coordinates. de Sitter space enjoys various time-like isometries inherited from the boosts
in embedding space. Yet, the metrics considered so far are time dependent. Since there is a
time-like Killing vector, there must exist coordinates such that time does not appear explicitly

133
 in the metric. These are static coordinates and are defined as
p t
X0 = L 2 − r 2 sinh
,
L
p t (E.11)
X d+1 = L 2 − r 2 cosh ,
L
X 0 =rωi ,

with i = 1, . . . , d and 0 ≤ r < L. They only cover the region

X
X d+1 > 0 , X i < L2 . (E.12)
i

The resulting metric reads

r2 dr 2
 
ds = 1 − 2 dt 2 −
2
2
− r 2 dΩ2d−1 . (E.13)
L 1 − Lr 2

This metric is manifestly static. In static coordinates the cosmological horizon is located at r = L
Therefore these coordinates cover precisely the patch that is accessible to a single observer, in
the sense that the observer can both send and receive signals to/from this entire region.

• Hyperbolic coordinates. Global coordinates foliate de Sitter with spheres, while planar co-
ordinates foliate with planes. To cover the third possibility we introduce hyperbolic coordinates
which foliate de Sitter by hyperbolic spaces. The embedding coordinates are defined by

X 0 = sinh τ cosh ψ ,
X d+1 = cosh τ , (E.14)
X i = sinh τ sinh ψωi ,

such that the metric takes the form

ds2 = dτ2 − sinh2 τdΞ2d . (E.15)

• Conformal coordinates. Finally, to obtain the conformal coordinates we start from planar
coordinates and perform a coordinate transformation to conformal time
t
dt ′
Z
t
η= = −Le− L , (E.16)
∞ a(t ′ )

so that −∞ < η < 0 and η → −∞ corresponds to the infinite past t → −∞. However, one
can extend this spacetime to almost all of the de Sitter by extending the range of η to the full

134
 real line. These coordinates can be parameterised as

1
η2 − x2 − L 2 ,


X0 =
2η
1
η2 − x2 + L 2 ,


X d+1 = − (E.17)
2η
L
X0 = − xi ,
η

so that the metric is given by

d
‚ Œ
2
L X
ds2 = 2 dη2 − dx i2 (E.18)
η i=1

These coordinates cover only the submanifold X 0 +X d+1 = 0. When written in these coordinates,
the metric of the de Sitter space, is manifestly conformally flat.

An important quantity in the geometry of the de Sitter space is the geodesic distance between two
points ζ(X , X ′ ). As in the case of the sphere or the hyperbolic plane in two dimensions, the distance
between two points of the de Sitter space is closely related to the distance defined in the embedding
space. Therefore, let us define
P(X , X ′ ) = −H 2 ηab X a X b . (E.19)

Notice that, if X = X ′ are identical, we have P = 1. However, if X and X ′ are antipodal, i.e. X ′ = −X ,
one has P = L12 ηa b X a X b = −1. Plugging in the explicit parameterisation, we find an expression of
the geodesic distance in conformal coordinates:

(η − η′ )2 − (x − x′ )2
P(X , X ′ ) = 1 + . (E.20)
2ηη′

One important property of P(X , X ′ ) is that it is a manifestly O(1, d) invariant function on de Sitter
space, since it is constructed out of the Lorentz invariant product in R1,d+1 . Depending on the causal
relationship between X and X ′ , we have the following behaviour for ζ(X , X ′ ):

• If X and X ′ are joined by a time-like geodesic P(X , X ′ ) > 1 and the geodesic distance is given by

1
ζ(X , X ′ ) = cosh−1 (P) . (E.21)
H

• If X and X ′ are space-like separated, |P(X , X ′ )| < 1 and

1
ζ(X , X ′ ) = cos−1 (P) . (E.22)
H

• If X and X ′ are light-like separated, P(X , X ′ ) = 1 and ζ(X , X ′ ) = 0.

Notice that there are points of de Sitter space which cannot be joined by geodesics to a given point X .
These are the points in the interior of the past and future light cones of −X , the antipodal point of X .
For these points, we have that P(X , X ′ ) < −1. The results listed above can be obtained by an explicit
analysis of geodesics in de Sitter space.

135
Exercise E.1. The geodesics in the de Sitter space can be obtained by minimizing the distance in the
embedding space subject to the constraint (E.5). Employ an appropriate Lagrange multiplier to solve this
minimisation problem and explicitly find the geodesics in the de Sitter space.

Anti-de Sitter space

The anti-de Sitter space is the maximally symmetric spacetime with negative curvature. Its scalar
curvature is given by
d(d + 1)
R=− . (E.23)
L2
Analogous to the de Sitter space we can describe Anti-de Sitter space in d + 1 dimensions by the
hyperboloid
d
X
X 02 + X 12 − X i2 = L 2 , (E.24)
i=1

embedded in a (d + 2) dimensional ambient space with metric

d+1
X
2
ds = dX 02 + dX 12 − dX i2 . (E.25)
i=2

The constant L is the AdS length parameterising the characteristic scale of the anti-de Sitter space.
The anti-de Sitter space inherits an O(2, d) symmetry from the ambient space. There are a variety of
useful coordinates on the Anti-de Sitter space, similar to the de Sitter space.

• Global coordinates. These coordinates {τ, ρ, ωi } are defined by

X 0 =L cosh ρ cos τ ,
X 1 =L cosh ρ sin τ , (E.26)
X i =L sinh ρωi ,

where i = 2, . . . d + 1 and the ωi are embedding coordinates on a round unit sphere S d−1 such
that i ω2i = 1. In these coordinates, the metric on the Anti-de Sitter space reads,
P

ds2 = L 2 cosh2 ρdτ2 − dρ 2 − sinh2 ρdΩ2d−1 .

(E.27)

In the limit ρ → ∞ one approaches the conformal boundary which in global coordinates is
given by R × S d−1 .

• Poincaré coordinates. These coordinates {z, x} are defined as

L 2 − t 2 + x2 + z 2
X0 = ,
2z
Lt
X1 = ,
z
(E.28)
L xi
Xi = ,
z
−L 2 − t 2 + x2 + z 2
x d+1 = ,
2z

136
 where i = 2, . . . , d. These coordinates do not cover the full anti-de Sitter space but only the
patch
L2
X 0 − X d+1 = > 0. (E.29)
z
In these coordinates, the metric reads,
‚ Œ
2
L X
ds2 = 2 dz 2 − dt 2 + dx i2 . (E.30)
z i

In these coordinates the conformal boundary is located at z → 0 and the geometry of the
boundary is that of d-dimensional Minkowski space.

• Static coordinates. These coordinates are given by

v
t r2 t
X0 = L 1+ sin ,
L2 L
v
t r2 t (E.31)
X1 = L 1+ 2
cos ,
L L
X i = rωi ,

with i = 2, . . . , d + 1. In these coordinates the metric reads

r2 dr 2
 
2
ds = 1 + 2 dt 2 − r2
− r 2 dΩ2d−1 . (E.32)
L 1 + L2

The conformal boundary is located at r → 0 and its geometry is given by R × S d−1 .

• de Sitter slicing. Finally, we can slice the anti-de Sitter space with de Sitter slices. The
embedding coordinates are given by

X 0 =L sinh ρ sinh t cosh ξ ,

X 1 =L cosh ρ ,
(E.33)
X 2 =L sinh ρ cosh t ,
X i = L sinh ρ sinh t sinh ξωi

so that the metric is given by

€ Š
ds2 = L 2 dρ 2 + sinh2 ρdsdS
2
. (E.34)
d−1

2
where dsdS is a metric on de Sitter space with Hubble scale H = 1.
d−1

• Hyperspherical coordinates. In these coordinates the embedding is parametrised as

X 0 =L sec ρ cos τ ,
X 1 =L sec ρ sin τ , (E.35)
X i =L tan ρωi ,

137
 with ωi again parametrising a unit d − 1 sphere. The metric in these coordinates is given by

ds2 = L 2 sec2 ρ dτ2 − dρ 2 − sin2 ρdΩ2d−1 .

(E.36)

In these coordinates the boundary of (the universal covering of) AdS is the Einstein static
universe.

An important quantity in the geometry of the anti-de Sitter space is the geodesic distance between
two points,
1
P(X , X ′ ) = − 2 ηab X a X ′b . (E.37)
L
where ηa b is the metric in the ambient space. If X = X ′ are identical, we have P = 1. However, if X
and X ′ are antipodal one has P = −1.
One important property of P(X , X ′ ) is that it is a manifestly O(2, d) invariant function on anti-de
Sitter space, since it is constructed out of the Lorentz invariant product in R2,d . Depending on the
causal relationship between X and X ′ , we have the following behaviour for P(X , X ′ ):

• If X and X ′ are joined by a time-like geodesic |P(X , X ′ )| < 1 and the geodesic distance is given
by
d(X , X ′ ) = L cos−1 (P) . (E.38)

• If X and X ′ are space-like separated, P(X , X ′ ) > 1 and

d(X , X ′ ) = L cosh−1 (P) . (E.39)

• If X and X ′ are light-like separated, P(X , X ′ ) = 1 and d(X , X ′ ) = 0.

Notice that this is opposite of the de Sitter case. Furthermore, notice that in the anti-de Sitter space
it is possible to reach the conformal boundary in a finite time. I.e. there exists a time-like geodesic
connecting any point X with the conformal boundary. This is why AdS is often thought of as a finite
’box’.

Exercise E.2. The geodesics in the anti-de Sitter space can be obtained by minimizing the distance in the
embedding space subject to the constraint (E.24). Employ an appropriate Lagrange multiplier to solve
this minimisation problem and explicitly find the geodesics in the de Sitter space.

Euclidean AdS

In some situations it is more convenient to perform computations in Euclidean signature and after
Euclidean AdS spacetime is the hyperboloid

d+1
X
−X 02 + X i2 = −R2 , X0 > 0 , (E.40)
i=2

138
embedded in Rd+1,1 . From this definition it is clear that Euclidean AdS is invariant under SO(d + 1, 1).
Let us be more explicit it this case and write out the symmetry generators as

∂ ∂
 ‹
JAB = −i X A − X B . (E.41)
∂ XB ∂ XA

Analogous to AdS , we define the Poincaré coordinates by

1 + x 2 + z2
X0 = R
2z
xµ
Xµ = R (E.42)
z
1 − x 2 − z2
X d+1 = R
2z

where x µ ∈ Rd and z > 0. In these coordinates the metric reads

2 2
dz 2 + δµν d x µ d x ν
ds = R . (E.43)
z2

This shows that EAdS is conformal to R+ × Rd whose boundary at z = 0 is just Rd . These coordinates
make explicit the subgroup SO(1, 1) × ISO(d) of the full isometry group of EAdS. These correspond
to dilatation and Poincaré symmetries inside the d−dimensional conformal group. In particular, the
dilatation generator is

∂ ∂ ∂ ∂
D = −i J0,d+1 = −X 0 + X d+1 = −z − xµ . (E.44)
∂X d+1 ∂X 0 ∂z ∂ xµ

Global coordinates in Euclidean AdS can simply be obtained from the global coordinates in AdS by
analytically continuing τ → iτ such that the metric is given by

ds2 = R2 cosh2 ρ dτ2 + dρ 2 + sinh2 ρ dΩ2d−1 .

 
(E.45)

To understand the global structure of this spacetime it is convenient to change the radial coordinate
via tanh ρ = sin r so that r ∈ [0, π2 [. Then, the metric becomes

R2  2
ds2 = 2 2 2

dτ + d r + sin r dΩ d−1 , (E.46)
cos2 r

which is conformal to a solid cylinder whose boundary at r = π2 is R × S d−1 . In these coordinates, the
dilatation generator D = −i J0,d+1 = − ∂∂τ is the Hamiltonian conjugate to global time.

Exercise E.3. Explicitly write out the symmetry generators for (Lorentzian) (A)dS spacetime, analogous
to the discussion in this last subsection.

139
E.2 Warped product manifolds and FLRW spaces

Apart from the maximally symmetric spacetimes, our second most loved example is given by warped
product manifolds of the form R × M with metric

ds2 = dt 2 − a(t)2 dsM

2
, (E.47)

When M is a maximally symmetric Euclidean manifold, i.e. M = Mk,d these are the FLRW manifolds
introduced in the main text. When a(t) is a constant function these represent static spacetimes. In
the case k = +1 this spacetime is often called the static Einstein universe.
On the other hand, when a(t) is a non-trivial function of time, these manifold provide an excellent
toy model for cosmology. The spatial section of the universe contracts or expands according to the
scale factor a(t). It is often useful to define the conformal time coordinate,
t
dt ′
Z
η= , (E.48)
−∞ a(t ′ )

in terms of which the FLRW metric becomes

€ Š
ds2 = a2 (η) dη2 − ds2M . (E.49)
k,d

Note that the de Sitter space can be thought of as a FLRW space.

In maximally symmetric spacetimes, such as Minkowski or (A)dS, there can be no beginning or end
of time. There cannot be any history because every time is equivalent. The simplest way to introduce
some time dependence is to consider FLRW spacetimes. For this reason they are natural toy models
in cosmology. The coordinate t introduced above, corresponds to the proper time of an observer at
rest with respect to the co-moving spatial coordinates. The spatial manifold in an FLRW space is
maximally symmetric hence such spacetimes describe a homogeneous and isotropic universe, i.e. a
universe that looks the same at every point in space. This metric is simple enough that it allows for a
very explicit, often exact, analysis. A most remarkable fact that should blow your mind is that this
most simple spacetime for k = 0 is in fact a very good description of our own universe on distances
much larger than average distance between galaxies, about a few Megaparsec (Mpc). Of course there
are small deviations from perfect homogeneity in our universe but on large enough scales this gives
an excellent description.
In contrast to the maximally symmetric spacetimes, in an FLRW universe time translation and Lorentz
boost fail to be isometries because of the time dependence of the scale factor a(t). A useful way to
capture this dependence is by defining the Hubble parameter

ȧ
H(t) = . (E.50)
a

The absence of time translations has profound implications for constructing QFTs in these backgrounds,
as energy is not preserved.
Next, let us quickly review the Einstein equations in such backgrounds. In order to have any hope to

140
solve them, we need to impose some particularly symmetric stress tensor. The most general stress
tensor consistent with the symmetries of FLRW spaces is given by

Tµν = diag (ρ, −p, · · · , −p) , (E.51)

where the energy density ρ and pressure p are functions of time only. We can interpret this as the
energy-momentum tensor of a homogeneous perfect fluid in its rest frame,

Tµν = (ρ + p)uµ uν − gµν p , (E.52)

where uµ is the normalised fluid velocity, |u|2 = 1, which in rest frame would be uµ = δµt . Einsteins
equation imply that the energy momentum tensor is covariantly conserved,

∇ν T µν = ∂ν T µν + Γαν
µ αν ν
T + Γαν T µν = 0 . (E.53)

Plugging in the FLRW solution this reduces to the so-called continuity equation,

ρ̇ + d H(ρ + p) = 0 . (E.54)

This equation tells us that the energy density changes only if the universe expands or contracts, i.e. if
H ̸= 0. The Einstein equations however will not tell us what kind of matter permeates the universe.
For that we need to specify an equation of state giving a relation between the pressure, density and
possibly other thermodynamic variables. Most systems of interest in cosmology can be described to a
good approximation by the very simple equation of state,

p = wρ , (E.55)

with a single parameter w. For this equation of state we can immediately solve the continuity equation
giving us
ρ(t) = ρ0 a(t)−d(1+w) . (E.56)

• Non-relativistic matter, a.k.a. dust, has a velocity much smaller than the speed of light. For this
type of matter, the pressure is negligible compared to the energy density, p ≪ ρ, or 0 < w ≪ 1.
Therefore in an expanding universe, dust dilutes as ρ ∝ a−d .

• Relativistic matter, a.k.a. radiation, on the other hand has pressure and energy density of the
same order. A statistical mechanics analysis furthermore predicts that p = ρ/d and so w = 1/d.
This is precisely the proportionality constant to make the matter conformal. Hence we find
ρ ∝ a−(d+1) .

• Finally, a cosmological constant, or vacuum energy has Tµν = −Λgµν and hence p = −ρ = −Λ,
or w = −1. Since in this case we have p + ρ = 0, the continuity equation teaches us that the
cosmological constant does not dilute, ρ ∝ a0 .

Solving the Einstein equations for an FLRW metric results in the Friedmann equations, which can be

141
written as
16πGN k 8πGN k
H2 = ρ− 2 , Ḣ = − (ρ + p) + 2 . (E.57)
d(d − 1) a d −1 a
Here p and ρ can be thought of as the effective energy density and pressure build from the combination
of all the matter present in the universe, together with the cosmological constant,
X Λ X Λ
p= pm − , ρ= ρm + . (E.58)
m
8πGN m
8πGN

The first Friedmann equation can be used to estimate the age of the universe, while the second
encodes the acceleration of the universe. Since most cosmological matter respects the null energy
condition, which in this case reads ρ + p > 0, we find that H typically decreases during the expansion
of the universe.

E.3 Black holes

Another important set of spacetimes that play a key role in this course are black holes. contrary to
the above examples, such spacetimes are singular and hence not complete.
In this course we restrict ourselves to the simplest of black holes, the Schwarzschild black hole, as it will
suffice to illustrate the relevant phenomena. This is the unique non-rotating neutral, asymptotically
flat black hole. More generally, black holes can have mass and/or charge. For such more general
solutions we refer the reader to the course General Relativity II.

The Schwarzschild black hole

The Schwarzschild black hole, with metric

2M 2M −1 2
 ‹  ‹
2 2
ds = 1 − dt − 1 − dr − dΩ22 , (E.59)
r r

is the unique non-rotating asymptotically flat black hole. This metric has a singularity at r = 2M , the
location of the event horizon, and therefore only describes the exterior of the black hole. To see that
the singularity at the event horizon is not a physical singularity it is useful to introduce the tortoise or
Regge-Wheeler coordinate r ∗ , which is defined such that massless free falling observers follow the
path t = r ∗ + constant. For such an observer one has,

1
ds2 = 0 → dt = 2M
dr ≡ dr ∗
1− r
 r  (E.60)
∗
→ r = r + 2M log −1 ,
2M

where we put the arbitrary additive constant to zero. In these coordinates the metric takes the form,

2M
 ‹
ds2 = 1 − dt 2 − dr ∗2 − r 2 dΩ22 .


(E.61)
r

142
From this line element we see that the two-dimensional metric is conformally equivalent to Minkowski
space. Next we introduce the retarded and advanced Eddington-Finkelstein coordinates

u = t + r∗ , v = t − r∗ , (E.62)

which places the horizon at (u, v) → (∞, −∞). In these coordinates the metric becomes

2M
 ‹
ds2 = 1 − dudv − r 2 dsS22 , (E.63)
r

where r can be expressed as a complicated function of u − v. In these coordinates the metric is still
singular at r = 2M but we can introduce one more coordinate transformation
 u   v 
U = −4M exp − , V = 4M exp , (E.64)
4M 4M

called Kruskal-Szekeres coordinates. The final metric is then given by

2M − r
ds2 = e 2M dUdV − r 2 dΩ22 . (E.65)
r
which makes it clear that there is no singularity at r = 2M . In these coordinates the event horizon
is at U = 0 or V = 0 and the original Schwarzschild metric only covers the patch U < 0 and V > 0.
However, there is no obstruction to extend U, V to the full real line. The fully extended metric covers
both the inside and outside of the Schwarzschild black hole and is the maximal extension of this
spacetime. Finally, an explicit map between U, V and r, t is given by

r
 r  U t
U V = e 2M 1 − , = e 2M . (E.66)
2M V

We see in fact that r = 2M is a null hypersurface ruled by outgoing null geodesics. The presence of
the event horizon means that not all light rays escape to infinity. For r > 2M , light rays with ṙ > 0
can and do escape. However, for r < 2M , all causal geodesics have future end point at the true
singularity at r = 0. More precisely, we can define the event horizon is as
Definition E.1. The event horizon is the boundary of the past of I + .

this can easily be seen from the metric using the retarded Eddington-Finkelstein coordinate

2M
 ‹
2
ds = 1 − du2 + 2dudr − r 2 dΩ22 , (E.67)
r

as the boundary is at u → ∞, which corresponds to r → 2M . There is a corresponding time-reversed

picture using coordinates (v, r). However, now we have that r = 2M is the past horizon being the
boundary v = −∞ of the future of I − . The full causal structure of the Schwarzschild black hole can
be summarised in its Penrose diagram, see Figure E.1.
Similar diagrams can be drawn for Reissner-Nordstrom, Kerr and the Kerr-Newman, see GR II,
although the latter have the novelty of having Cauchy horizons, hypersurfaces beyond which neither
fields nor space-time itself are determined by Cauchy data essentially as a consequence of naked
singularities, singularities in the past of observers. However, these cannot be seen from infinity. These

143
 Figure E.1: Penrose diagram for the Kruskal extension of the Schwarzschild spacetime. The
singularity at r = 0 (which is a genuine curvature singularity) is a black hole to the future of
every observer that crosses the future event horizon (or a white hole in the past).

black hole solutions are unique subject to various assumptions (like the existence of a stationary
Killing vector that looks like a time translation at large distances). Similarly, all of these spacetimes
have generalisations with non-vanishing cosmological constant.
This final state is tightly constrained as in four dimensions we have powerful uniqueness theorems.
Birkhoff’s theorem says that any spherically symmetric vacuum solution is static, which then implies
that it must be Schwarzschild. For Einstein-Maxwell system this extends to show that the only
spherically symmetric solution is Reissner-Nordstrom. But suppose we know only that the metric
exterior to a star is static. We further have:

Theorem E.1 (Israel). If (M, g) is an asymptotically-flat, static, vacuum space-time that is non-singular
on and outside an event horizon, then (M,g) is Schwarzschild.

More remarkably we have

Theorem E.2 (Carter-Robinson). If (M, g) is an asymptotically-flat stationary and axi-symmetric

vacuum spacetime that is non-singular on and outside an event horizon, then (M,g) is a member of the
two-parameter Kerr family. The parameters are the mass M an the angular momentum J.

The assumption of axi-symmetry has since been shown to be unnecessary by Hawking and Wald , i.e.,
for black holes, stationarity ⇒ axisymmetry.

A digression on null congruences and hypersurfaces

An important role is played by null hypersurfaces, i.e., hypersurfaces u(x) = constant such that ∇a u
is a null vector. It is a standard elementary exercise that l a = ∇a u is tangent to a family of null
geodesics.

144
More generally, a null congruence is a foliation of a region of space-time by null geodesics. It can be
defined by a null vector field l a whose integral curves are the null geodesics through each point. If it
is tangent to a congruence of affinely parametrised null geodesics, then

∇l l b := l a ∇a l b = 0 (E.68)

The geometry can be studied by introducing a 2-dimensional screen space consisting of tangent vectors
perpendicular to the direction of null geodesic. This space can be spanned by a pair of orthonormal
vectors X a , Y a , X · X = Y · Y = 1 and X · l = Y · l = 0. Given l, these are defined up to standard 2d
rotations. The general vector in this screen space can be parametrized as

V a = xX a + yY a . (E.69)
p
It is conventional to introduce the complex coordinate 2ζ = x + i y on screen space and associated
complex null vectors ma , m̄a so that

1
ma = p (X a − iY a ) , V a = ζm̄a + ζ̄ma . (E.70)
2

We can choose X a and Y a to be parallel propagated along the null geodesics, ∇l X a = ∇l Y a = ∇l ma =

0.
We wish to measure how images on the screen are distorted as they are propagated along the light
rays of the congruence. If V a connects nearby geodesics of the congruence, then it is Lie derived
along l a , i.e.,
[l, V ]a = ∇l V a − ∇V l a = 0 . (E.71)

It is easy to see that if l · V = 0 initially, then it remains zero. This gives

∇l ζ = −ρζ − σζ̄ , (E.72)

for some complex parameters ρ, σ. These can be interpreted as follows:

1. The imaginary part of ρ is the twist and generates rotations of the ζ plane. It vanishes iff the
congruence is hypersurface forming, l[a ∇ b l c] = 0 which implies that there is a rescaling of l a
so that l a = ∇a u for some function u.

2. The real part of ρ gives the expansion, ∇a l a = −2ρ and the area element of the orthogonal
transverse plane evolves by
A = −ima d x a ∧ m̄ b d x b ,

satisfies
Ll A = −2ρA (E.73)

3. The complex scalar σ is the shear in the sense that a circle in the ζ plane evolves into an ellipse.

4. Equation (E.71) implies the geodesic deviation equation

∇l ∇l V a = l b l c V d R bdc a (E.74)

145
 and this combines with (E.72) to give the Sachs equations

∇l ρ = ρ 2 + σσ̄ + Φ00 (E.75)

∇l σ = (ρ + ρ̄)σ + Ψ0 (E.76)

Here Ψ0 = Ca bcd l a m b l c md , Φ00 = Φab l a l b = − 21 R ab l a l b and is positive when the dominant

energy condition is satisfied. An important consequence for horizons and singularity theorems
is that the whole RHS of (E.75) is manifestly positive definite.

5. If a null hypersurface has vanishing shear, then it has the intrinsic geometry of a light cone
or null hyperplane in Minkowski space up to scaling (i.e. the metric restricts to a multiple of
dζd ζ̄ on R3 or S 2 × R where l a ∂a = ∂ v for a third coordinate v).

Horizons and black hole thermodynamics

For an asymptotically flat space-time, we define

Definition E.2. The event horizon H is the boundary of the past J − (I + ) of I + , that is, it is the
boundary of the region from which it is possible to escape to infinity along a causal curve.

Much is known about event horizons under reasonable assumptions appropriate to isolated systems
that settle down:

• H is a null hypersurface being the boundary of a past set (it clearly cannot be time-like as causal
paths could then cross both ways, and if it were space-like there would be regions to its past
that could not exit to I + ).

• H is ruled (or foliated) by complete null geodesics.

• If I has topology S 2 × R, as appropriate for the exterior of an isolated system, then so does H,
with the R factor being the null geodesics.

• The cross-sectional area is bounded above.

This is a rather excessively global definition that requires knowledge of the whole space-time. One
can also define with just local knowledge:

Definition E.3. a closed trapped surface is a two-surface of topology S 2 such that the outward pointing
null geodesics have nonpositive expansion (i.e., the area will drop or be constant in any outward
going null direction or ρ ≥ 0 where ρ is the spin coefficient in the definition of the Sachs equation).

Penrose’s original singularity theorem deduces the existence of a singularity (in the form of geodesic
incompleteness) from the existence of such a closed trapped surface. It is easy to see from the signs
in the Sachs equations and following the outward going null geodesic normals off hte surface that a
closed trapped surface leads to:

Definition E.4. an apparent horizon is a null hypersurface of topology S 2 × R such that the expansion
of the outward going null rays is nonpositive (i.e., the area is non-increasing to the future).

146
The first of the Sachs equation for a null geodesic congruence generated by l gives

∇l ρ = ρ 2 + σσ̄ + Φ00 ≥ ρ 2 . (E.77)

Thus if ρ ≥ 0 then it cannot decrease. (Recall that if A is the area element, Ll A = −2ρA.)
However, Penrose’s theorem doesnt deduce the location of the singularity! In particular it is not clear
that an apparent horizon is hidden inside an event horizon and the following is open:
The cosmic censorship hypothesis: All singularities that arise from evolving from an initial data
hypersurface are hidden behind an event horizon and so cannot be seen from infinity.
Generally speaking we assume that a black hole settles down to being stationary or static. Then, the
event horizon must settle down to a null hypersurface with finite cross sectional area (otherwise
geodesics will be escaping to infinity). Once a black hole horizon settles down, its area is constant.
Assuming that the black hole is becoming stationary or static, it follows that it is (under suitable
analyticity assumptions) a Killing horizon:

Definition E.5. A Killing horizon is a null hypersurface on which a Killing vector ka becomes null, so
that the surface is defined by ka k a = 0 and k a ̸= 0. Thus k a is tangent to the null geodesic generators
of the horizon.

The fact that k a is Killing means that we must have ρ = σ = 0.

It follows from the black hole uniqueness theorems that, even if we started from some collapse
scenario, the final black hole, if essentially static or stationary, is Kerr Newman or Schwarzschild.
These all have a similar structure to Schwarzschild in that they can be continued analytically back to
a point where the standard future event horizon intersects a past one at a 2-surface C.

Definition E.6. Such a Killing Horizon is said to be a bifurcate Killing horizon if there exists cross-
section C of topology S 2 on which k a vanishes—it is bifurcate because then in a neighbourhood of
there is a transverse horizon such that k a = U∂U − V ∂V as for the crossover in Schwarzschild.

On a Killing horizon we can define

Definition E.7. The surface gravity κ is defined by

∇a k b k b = −2κka , or k a ∇a k b = κk b , (E.78)

where in the static case, ka is understood to be normalized to have ka k a = 1 at large distances. For
Schwarzschild κ = 1/4m.

Black hole thermodynamics starts with the Bekenstein bound on the entropy S: in a region of radius
R and mass-energy E the entropy is constrained by

2πkRE
S< (E.79)
hc
ħ

where we have not set the usual fundamental constants k, ħ

h, c to unity. It was arrived at by considera-
tion of throwing objects with entropy into black holes and trying to avoid violations of the second

147
law arising from the black hole eating entropy. In this view, the black does have entropy

kA
SBH = (E.80)
4G

and this is taken to be the maximal entropy state, i.e., the Bekenstein bound is saturated by the black
hole entropy.
Classically, one does not think of black holes as having microstates that could give rise to an entropy
in view of the black hole uniqueness theorems. These seem to imply that the black hole state is
unique, whereas an enropy suggests the existence of many equally likely microstates compatible with
given macroscopic observables. The black hole entropy is usually understood as having its origin in
quantum gravity.
This chain of reasoning subsequently led to the Holographic principle, that the maximum number N
of states in a spatial region of radius R satisfies

N < exp SBH (R) (E.81)

P
This comes from the definition of entropy of a system as S := − i pi log pi where pi is the probability
of the ith state. If the system is equidistributed, pi = 1/N , where N is the number of states (the
dimension of the Hilbert space of the system) we obtain S = log N . This is counter-intuitive without
general relativity because one thinks of the number of states in a region as being the exponential of
the volume rather than the area. However, gravitational collapse reduces this if there is too much
matter (too many particles) and indeed the vast bulk of the entropy is undertood to be gravitational.
The second law of thermodynamics: if the entropy is equated with the area of the event horizon in
black hole thermodynamics, the 2nd law states that it can only increase. We have the area theorem

Theorem E.3. The area of an event horizon is non-decreasing.

Proof: This is a simple consequence of the Sachs (or Raychauduri) equations

ρ̇ = ρ 2 + |σ|2 + Φ00 . (E.82)

This shows that in particular ρ̇ ≥ ρ 2 when the dominant energy condition is satisfied. Thus, if
ρ = ρ0 > 0 at some affine parameter value t = 0 on the generator, it is bounded below by the solution
ρ0
ρ0 (t) = , (E.83)
1 − ρ0 t

the solution to ρ̇ = ρ 2 with the same initial condition. Thus ρ → ∞ in finite time. This introduces a
cusp after which the null geodesic must then leave the horizon (see picture), contradicting its being a
generator of H. □
The first law of black hole thermodynamics: for a variation of a closed system with rotation and
charge can be stated as
d E = T dS + Ωd J + ΦH dQ (E.84)

Here E is the total energy, T the temerature, Ω the angular velocity, J the angular momentum, Q the

148
charge and φ the electrostatic potential. In the context of black holes, the total energy is the mass,
we identify the temperature with the surface gravity by

T = κ/2π (E.85)

and S with the area.

For Reissner Nordstrom, Φ = ΦH is the electric potential at the horizon and Q the total charge, and,
in the case of the Kerr solution, Ω is the angular velocity, and J the angular momentum.
There are a number of strategies for proving these formulae. The most basic is to simply establish
sufficient relations between the various quantities (M , A, ΩH , J, φH , Q) as can be read off from the
black hole metric, and then to differentiate it. The simplest example is for Schwarschild where the
area is that associated with the Schwarzschild radius r = 2M , A = 16πM 2 , upon which differentiation
yields
dA κ
dM = = dA, (E.86)
32πM 2π
giving the most basic version.
If we wish to introduce charge, we must consider the Reissner-Nordstrom solution

∆(r) 2 r2
ds2 = d t − d r 2 − r 2 dsS22 , ∆(r) = r 2 − 2M r + Q2 . (E.87)
r2 ∆(r)

This satisfies the Einstein equations with electromagnetic potential

Q
A= dt . (E.88)
r
The Killing horizons are where ∆ = 0 giving
p
r± = M ± M 2 − Q2 , (E.89)

assuming Q < M . The outer one is the event horizon and a short calculation shows that differentiating
the obvious relation A = πr+2 now gives
p
M 2 − Q2 Q
dM = dA + dQ . (E.90)
2πr+2 r+

The coefficient of dQ is indeed the value of the potential at the horizon. It is a more complicated task
to see that the surface gravity does indeed appropriately give the coefficient of dA (see the exercises).
Even more nontrivially, this works as stated above for the Kerr-Newman solution where there is also
rotation.
The zeroth law of Black hole thermodynamics: In the analogy with thermodynamics, κ plays the
role of temperature via T = κ/2π. The zeroth law is that the temperature is constant in equilibrium.
It is easy to see that the surface gravity κ is constant up the generators of H, because k a is Killing.
We will see in the problems that for a bifurcate Killing horizon κ is actually constant over the horizon.
Hence it is constant everywhere. The next result follows in greater generality but we will not prove it

149
here.
There is also a third law, that the entropy of an object at absolute zero is zero. This fails for black holes
for a number of reasons, but a vaguer version, that once cannot approach absolute zero temperature
with a finite number of processes does seem reasonable, as T → 0 corresponds to M → ∞.
The glaring omission in all this is of course that the temperature of a black hole classically would
seem to have to be zero. This will be seen to be resolved by Hawking radiation.

150
 Appendix F

Hypergeometric functions

In this appendix we review the definition and various properties of hypergeometric functions. For
further reference see for example [AAR99].
The hypergeometric function is a solution of Euler’s hypergeometric differential equation,

z(1 − z)∂z2 F + [γ − (α + β + 1)z] ∂z F − αβ F = 0 . (F.1)

which has three regular singular points at z = 0, 1 and ∞. Any second order linear equation with
three regular singular points can be converted to this equation through a change of variables.
For |z| < 1, the hypergeometric function can be defined through the following series expansion
∞
X (α)n (β)n z n
2 F1 (α, β; γ|z) = , (F.2)
n=0
(γ)n n!

where we introduced the (rising) Pochhammer symbol

Γ (q + n)
(q)n = . (F.3)
Γ (q)

when either α or β is a non-positive integer this series terminates in which case the hypergeometric
function reduces to a polynomial. For complex |z| ≥ 1 it can be analytically continued along any path
that avoids the branch points at z = 1 and z = ∞.
Depending on the sign of Re(γ − α − β) we find the following behaviour near z = 1,

Γ (γ)Γ (γ−α−β)
¨
z→1 Γ (γ−α)Γ (γ−β) , for Re(γ − α − β) > 0 ,
2 F1 (α, β; γ|z) ≃ Γ (γ)Γ (α+β−γ) γ−α−β
(F.4)
Γ (α)Γ (β) (1 − z) , for Re(γ − α − β) < 0 .

In addition,we have the following identities for the analytic continuation of the hypergeometric
functions,

Γ (γ)Γ (γ − α − β)
2 F1 (α, β; γ|z) = F1 (α, β; 1 + α + β − γ|1 − z) (F.5)
Γ (γ − α)Γ (γ − β) 2
Γ (γ)Γ (α + β − γ) γ−α−β
+ (1 − z)2 F1 (γ − α, γ − β; 1γ − α − β|1 − z) .
Γ (α)Γ (β)

and
γ−α−β
2 F1 (α, β; γ|z) = (1 − z)2 F1 (γ − α, γ − β; γ|z) . (F.6)

151
 Appendix G

A mini introduction to CFT

This section briefly describes the basic concepts necessary to formulate a non-perturbative definition
of CFT.

G.1 Conformal Transformations

For simplicity, in most formulas, we will consider Euclidean signature. We start by discussing conformal
transformations of Rd in Cartesian coordinates,

ds2 = δµν d x µ d x ν . (G.1)

A conformal transformation is a coordinate transformation that preserves the form of the metric
tensor up to a scale factor,
d x̃ µ d x̃ ν
δµν α = Ω2 (x)δαβ . (G.2)
d x d xβ
In other words, a conformal transformation is a local dilatation.

Exercise G.1. Show that, for d > 2, the most general infinitesimal conformal transformation is given by
x̃ µ = x µ + εµ (x) with

εµ (x) = aµ + λx µ + mµν x ν + x 2 bµ − 2x α bα x µ . (G.3)

In spacetime dimension d > 2, conformal transformations form the group SO(d +1, 1). The generators
Pµ and Mµν correspond to translation and rotations and they are present in any relativistic invariant
QFT. In addition, we have the generators of dilatations D and special conformal transformations Kµ .
It is convenient to think of the special conformal transformations as the composition of an inversion
followed by a translation followed by another inversion. Inversion is the conformal transformation1

xµ
xµ → . (G.4)
x2

Exercise G.2. Verify that inversion is a conformal transformation.

The form of the generators of the conformal algebra acting on functions can be obtained from
• ˜
i
φ (x µ + εµ (x)) = 1 + i aµ Pµ − λD + mµν Mµν + i bµ Kµ φ (x µ ) , (G.5)
2
1
Inversion is outside the component of the conformal group connected to the identity. Thus, it is possible to have CFTs
that are not invariant under inversion. In fact, CFTs that break parity also break inversion.

152
which leads to 2

Pµ = −i∂µ , D = −x µ ∂µ , (G.6)
ν 2


Mµν = −i x µ ∂ν − x ν ∂µ , Kµ = 2i x µ x ∂ν − i x ∂µ . (G.7)

Exercise G.3. Show that the generators obey the following commutation relations
     
D, Pµ = Pµ , D, Kµ = −Kµ , Kµ , Pν = 2δµν D − 2i Mµν ,
 
 

Mµν , Pα = i δµα Pν − δνα Pµ , Mµν , Kα = i δµα Kν − δνα Kµ ,
 

Mαβ , Mµν = i δαµ Mβν + δβν Mαµ − δβµ Mαν − δαν Mβµ . (G.8)

G.2 Local Operators

Local operators are divided into two types: primary and descendant. Descendant operators are
operators that can be written as (linear combinations of) derivatives of other local operators. Primary
operators can not be written as derivatives of other local operators. Primary operators at the origin are
annihilated by the generators of special conformal transformations. Moreover, they are eigenvectors
of the dilatation generator and form irreducible representations of the rotation group SO(d),
     B
Kµ , O(0) = 0 , [D, O(0)] = ∆ O(0) , Mµν , OA(0) = Mµν A OB (0) . (G.9)

Correlation functions of scalar primary operators obey

∆1 ∆n
− −
∂ x̃ d
∂ x̃ d
〈O1 (x̃ 1 ) . . . On (x̃ n )〉 = ... 〈O1 (x 1 ) . . . On (x n )〉 (G.10)
∂x x1 ∂x xn

for all conformal transformations x → x̃. As explained above, it is sufficient to impose Poincaré
invariance and this transformation rule under inversion,
    
x1 xn
∆1
∆ n
O1 . . . On = x 12 . . . x n2 〈O1 (x 1 ) . . . On (x n )〉 .
x 12 x n2

This implies that vacuum one-point functions 〈O(x)〉 vanish except for the identity operator (which
is the unique operator with ∆ = 0). It also fixes the form of the two and three point functions,

δi j
Oi (x)O j ( y) = , (G.11)
(x − y)2∆i
C123
〈O1 (x 1 )O2 (x 2 )O3 (x 3 )〉 = ∆1 +∆2 −∆3 ∆1 +∆3 −∆2
,
|x 12 | |x 13 | |x 23 |∆2 +∆3 −∆1

where we have normalized the operators to have unit two point function.
The four-point function is not fixed by conformal symmetry because with four points one can construct
2
We define the dilatation generator D in a non-standard fashion so that it has real eigenvalues in unitary CFTs.

153
two independent conformal invariant cross-ratios

2 2 2 2
x 12 x 34 x 14 x 23
u = zz̄ = 2 2
, v = (1 − z)(1 − z̄) = 2 2
. (G.12)
x 13 x 24 x 13 x 24

The general form of the four point function is

A(u, v)
O(x 1 ) . . . O(x 4 ) =
. (G.13)
2 2 ∆
x 13 x 24

G.3 Ward identities

To define the stress-energy tensor it is convenient to consider the theory in a general background
metric gµν . Formally, we can write
Z
1
〈O1 (x 1 ) . . . On (x n )〉 g = [dφ]e−S[φ,g] O1 (x 1 ) . . . On (x n ) , (G.14)
Z[g]

where Z[g] = [dφ]e−S[φ,g] is the partition function for the background metric gµν . Recalling the
R

classical definition
2 δS
T µν (x) = − p , (G.15)
g δgµν (x)
it is natural to define the quantum stress-energy tensor operator via the equation

Z[g + δg]
Z
1 p
=1+ dx gδgµν (x) 〈T µν (x)〉 g + O(δg 2 ) , (G.16)
Z[g] 2

and

〈O1 (x 1 ) . . . On (x n )〉 g+δg − 〈O1 (x 1 ) . . . On (x n )〉 g

Z
1 p
d x gδgµν (x) 〈T µν (x)O1 (x 1 ) . . . On (x n )〉 g

= (G.17)
2
− 〈T µν (x)〉 g 〈O1 (x 1 ) . . . On (x n )〉 g + O(δg 2 ) .


Under an infinitesimal coordinate transformation x̃ µ = x µ + εµ (x), the metric tensor changes g̃µν =
gµν − ∇µ εν − ∇ν εµ but the physics should remain invariant. In particular, the partition function
Z[g] = Z[g̃] and the correlation functions 3

〈O1 (x̃ 1 ) . . . On (x̃ n )〉 g̃ = 〈O1 (x 1 ) . . . On (x n )〉 g , (G.18)

3
If the operators are not scalars (e.g. if they are vector operators) then one also needs to take into account the rotation
of their indices.

154
do not change. This leads to the conservation equation ∇µ T µν (x) g
and

n
X ∂
εµ (x i ) µ 〈O1 (x 1 ) . . . On (x n )〉 g (G.19)
i=1
∂ xi
Z
p
=− dx gεν (x) ∇µ T µν (x)O1 (x 1 ) . . . On (x n ) g

for all εµ (x) that decays sufficiently fast at infinity. Thus ∇µ T µν = 0 up to contact terms.
Correlation functions of primary operators transform homogeneously under Weyl transformations of
the metric 4
〈O1 (x 1 ) . . . On (x n )〉 g
〈O1 (x 1 ) . . . On (x n )〉Ω2 g = . (G.20)
[Ω(x 1 )]∆1 . . . [Ω(x n )]∆n
Exercise G.4. Show that this transformation rule under local rescalings of the metric (together with
coordinate invariance) implies (G.10) under conformal transformations.

Consider now an infinitesimal Weyl transformation Ω = 1+ω, which corresponds to a metric variation
δgµν = 2ωgµν . From (G.17) and (G.20) we conclude that

n
X
∆i ω(x i ) 〈O1 (x 1 ) . . . On (x n )〉 g
i=1
Z
p
g ω(x)gµν 〈T µν (x)O1 (x 1 ) . . . On (x n )〉 g

=− dx (G.21)

− 〈T µν (x)〉 g 〈O1 (x 1 ) . . . On (x n )〉 g .


Consider the following codimension 1 integral over the boundary of a region B, 5

Z
dSµ εν (x) 〈T µν (x)O1 (x 1 ) . . . On (x n )〉 g

I= (G.22)
∂B
− 〈T µν (x)〉 g 〈O1 (x 1 ) . . . On (x n )〉 g .


One can think of this as the total flux of the current εν T µν , where εν (x) is an infinitesimal conformal
transformation. Gauss law tells us that this flux should be equal to the integral of the divergence of
the current
1
∇µ (εν T µν ) = εν ∇µ T µν + ∇µ εν T µν = εν ∇µ T µν + ∇α εα gµν T µν , (G.23)
d
where we used the symmetry of the stress-energy tensor T µν = T νµ and the definition of an infinites-
imal conformal transformation ∇µ εν + ∇ν εµ = d2 ∇α εα gµν . Using Gauss law and (G.19) and (G.21)
we conclude that
X ∂ ∆i

µ α
I =− ε (x i ) µ + ∇α ε (x i ) 〈O1 (x 1 ) . . . On (x n )〉 g . (G.24)
x ∈B ∂ xi d
i

4
In general, the partition fungion is not invariant in even dimensions. This is the Weyl anomaly Z[Ω2 g] = Z[g]e−SWe y l [Ω,g] .
5
In the notation of the Conformal Bootstrap chapter [?] this is the topological operator Q ε [∂ B] inserted in the correlation
function 〈O1 (x 1 ) . . . On (x n )〉 g .

155
The equality of (G.22) and (G.24) for any infinitesimal conformal transformation (G.3) is the most
useful form of the conformal Ward identities.

Exercise G.5. Conformal symmetry fixes the three-point function of a spin 2 primary operator and two
scalars up to an overall constant, 6

H µν (x 1 , x 2 , x 3 )
〈O(x 1 )O(x 2 )T µν (x 3 )〉 = C12T , (G.25)
|x 12 |2∆−d+2 |x 13 |d−2 |x 23 |d−2

where µ µ
µν 1 x 13 x 23
H µ ν
= V V − Vα V α δµν , µ
V = 2
− 2
. (G.26)
d x 13 x 23
Write the conformal Ward identity (G.22)=(G.24) for the three point function 〈T µν (x)O(0)O( y)〉 for
the case of an infinitesimal dilation εµ (x) = λx µ and with the surface ∂ B being a sphere centred at the
origin and with radius smaller than | y|. Use this form of the conformal Ward identity in the limit of an
infinitesimally small sphere ∂ B and formula (G.25) for the three point function to derive

d∆ 1
COO T = − , (G.27)
d − 1 Sd

2πd/2
where Sd = Γ (d/2) is the volume of a (d − 1)-dimensional unit sphere.

G.4 State-Operator Map

Consider Rd in spherical coordinates. Writing the radial coordinate as r = eτ we find

ds2 = d r 2 + r 2 dΩ2d−1 = e2τ dτ2 + dΩ2d−1 .

(G.28)

Thus, the cylinder R × S d−1 can be obtained as a Weyl transformation of euclidean space Rd .

Exercise G.6. Compute the two-point function of a scalar primary operator on the cylinder using the
Weyl transformation property (G.20).

A local operator inserted at the origin of Rd prepares a state at τ = −∞ on the cylinder. On the
other hand, a state on a constant time slice of the cylinder can be propagated backwards in time until
it corresponds to a boundary condition on a arbitrarily small sphere around the origin of Rd , which
defines a local operator. Furthermore, time translations on the cylinder correspond to dilatations on
Rd . This teaches us that the spectrum of the dilatation generator on Rd is the same as the energy
spectrum for the theory on R × S d−1 . 7
6
You can try to derive this formula using the embedding space formalism of section G.7.
7
More precisely, there can be a constant shift equal to the Casimir energy
of the vacuum on S d−1 , which is related with
1
the Weyl anomaly. In d = 2, this gives the usual energy spectrum ∆ − 12 c
L where c is the central charge and L is the
radius of S 1 .

156
G.5 Operator Product Expansion

The Operator Product Expansion (OPE) between two scalar primary operators takes the following
form  
X
Oi (x)O j (0) = Ci jk |x|∆k −∆i −∆ j Ok (0) + β x µ ∂µ Ok (0) + . . . (G.29)
k
| {z }
descendants

where β denotes a number determined by conformal symmetry. For simplicity we show only the
contribution of a scalar operator Ok . In general, in the OPE of two scalars there are primary operators
of all spins.

Exercise G.7. Compute β by using this OPE inside a three-point function.

The OPE has a finite radius of convergence inside correlation functions. This follows from the state
operator map with an appropriate choice of origin for radial quantization.

G.6 Conformal Bootstrap

Using the OPE successively one can reduce any n−point function to a sum of one-point functions,
which all vanish except for the identity operator. Thus, knowing the operator content of the theory,
i.e. the scaling dimensions ∆ and SO(d) irreps R of all primary operators, and the OPE coefficients
Ci jk ,8 one can determine all correlation functions of local operators. This set of data is called CFT
data because it essentially defines the theory. 9 The CFT data is not arbitrary, it must satisfy several
constraints:

• OPE associativity - Different ways of using the OPE to compute a correlation function must
give the same result. This leads to the conformal bootstrap equations described below.

• Existence of stress-energy tensor - The stress-energy tensor Tµν is a conserved primary

operator (with ∆ = d) whose correlation functions obey the conformal Ward identities.

• Unitarity - In our Euclidean context this corresponds to reflection positivity and it implies lower
bounds on the scaling dimensions. It also implies that one can choose a basis of real operators
where all OPE coefficients are real. In the context of statistical physics, there are interesting
non-unitary CFTs.

It is sufficient to impose OPE associativity for all four-point functions of the theory. For a four-point
function of scalar operators, the bootstrap equation reads

(12)(34) (13)(24)
X X
C12k Ck34 G∆ (x 1 , . . . , x 4 ) = C13q Cq24 G∆ (x 1 , . . . , x 4 ) ,
k ,l k q ,l q
k q

8
For primary operators O1 , O2 , O3 transforming in non-trivial irreps of SO(d) there are several OPE coefficients C123 .
The number of OPE coefficients C123 is given by the number of symmetric traceless tensor representations that appear in
the tensor product of the 3 irreps of SO(d) associated to O1 , O2 and O3 .
9
However, there are observables besides the vacuum correlation functions of local operators. It is also interesting to
study non-local operators (line operators, surface operators, boundary conditions, etc) and correlation functions in spaces
with non-trivial topology (for example, correlators at finite temperature).

157
where G∆,l are conformal blocks, which encode the contribution from a primary operator of dimension
∆ and spin l and all its descendants.

G.7 Embedding Space Formalism

The conformal group SO(d + 1, 1) acts naturally on the space of light rays through the origin of
Rd+1,1 ,

2
2
2
− P 0 + P 1 + · · · + P d+1 = 0 . (G.30)

A section of this light-cone is a d−dimensional manifold where the CFT lives. For example, it is easy
to see that the Poincaré section P 0 + P d+1 = 1 is just Rd . To see this parametrize this section using

1 + x2 1 − x2
P 0 (x) = , P µ (x) = x µ , P d+1 (x) = , (G.31)
2 2

with µ = 1, . . . , d and x µ ∈ Rd and compute the induced metric. In fact, any conformally flat
manifold can be obtained as a section of the light-cone in the embedding space Rd+1,1 . Using the
parametrization P A = Ω(x)P A(x) with x µ ∈ Rd , one can easily show that the induced metric is simply
given by ds2 = Ω2 (x)δµν d x µ d x ν . With this is mind, it is natural to extend a primary operator from
the physical section to the full light-cone with the following homogeneity property

O(λP) = λ−∆ O(P) , λ∈R. (G.32)

This implements the Weyl transformation property (G.20). One can then compute correlation functions
directly in the embedding space, where the constraints of conformal symmetry are just homogeneity
and SO(d + 1, 1) Lorentz invariance. Physical correlators are simply obtained by restricting to the
section of the light-cone associated with the physical space of interest. This idea goes back to Dirac [?]
and has been further develop by many authors [?, ?, ?, ?, ?, ?, ?].

Exercise G.8. Rederive the form of two and three point functions of scalar primary operators in Rd
using the embedding space formalism.

Vector primary operators can also be extended to the embedding space. In this case, we impose

P AOA(P) = 0 , OA(λP) = λ−∆ OA(P) , λ∈R, (G.33)

and the physical operator is obtained by projecting the indices to the section,

∂ PA
Oµ (x) = OA(P) . (G.34)
∂ xµ P A=P A(x)

Notice that this implies a redundancy: OA(P) → OA(P) + PAΛ(P) gives rise to the same physical
operator O(x), for any scalar function Λ(P) such that Λ(λP) = λ−∆−1 Λ(P). This redundancy together
with the constraint P AOA(P) = 0 remove 2 degrees of freedom of the (d + 2)-dimensional vector OA.

158
Exercise G.9. Show that the two-point function of vector primary operators is given by

ηAB (P1 · P2 ) − P2A P1B

OA(P1 )O B (P2 ) = const , (G.35)
(−2P1 · P2 )∆+1

up to redundant terms.

Exercise G.10. Consider the parametrization P A = P 0 , P µ , P d+1 = (cosh τ, Ωµ , − sinh τ) of the global



2
2
section P 0 − P d+1 = 1, where Ωµ (µ = 1, . . . , d) parametrizes a unit (d − 1)−dimensional sphere,
Ω · Ω = 1. Show that this section has the geometry of a cylinder exactly like the one used for the
state-operator map.

Conformal correlation functions extended to the light-cone of R1,d+1 are annihilated by the generators
of SO(1, d + 1)
n
(i)
X
JAB 〈O1 (P1 ) . . . On (Pn )〉 = 0 , (G.36)
i=1
(i)
where JAB is the generator
∂ ∂
 ‹
JAB = −i PA − PB , (G.37)
∂ PB ∂ PA
acting on the point Pi . For a given choice of light cone section, some generators will preserve the
section and some will not. The first are Killing vectors (isometry generators) and the second are
conformal Killing vectors. The commutation relations give the usual Lorentz algebra

[JAB , JC D ] = i (ηAC JBD + ηBD JAC − ηBC JAD − ηAD JBC ) . (G.38)

Exercise G.11. Check that the conformal algebra (G.8) follows from (G.38) and

D = −i J0,d+1 , Pµ = Jµ0 − Jµ,d+1 ,

Mµν = Jµν , Kµ = Jµ0 + Jµ,d+1 . (G.39)

Exercise G.12. Show that equation (G.36) for JAB = J0,d+1 implies time translation invariance on the
cylinder
n
X ∂
〈O1 (τ1 , Ω1 ) . . . On (τn , Ωn )〉 = 0 , (G.40)
i=1
∂ τi

and dilatation invariance on Rd

n 
µ ∂
X 
∆i + x i µ 〈O1 (x 1 ) . . . On (x n )〉 = 0 . (G.41)
i=1
∂ x i

In this case, you will need to use the differential form of the homogeneity property P A ∂ ∂P A Oi (P) =
−∆i Oi (P). It is instructive to do this exercise for the other generators as well.

159
 V =2 V =4
E=3 E=6
F =3 F =2
g=0 g=1

Figure G.1: Vacuum diagrams in the double line notation. Interaction vertices are marked with
a small blue dot. The left diagram is planar while the diagram on the right has the topology of
a torus (genus 1 surface).

G.8 Large N Factorization

Consider a U(N ) gauge theory with fields V = 2in the adjoint

valued V representation.
=4 Schematically, we can
write the action as
N
Z E = 3 E=6 
S= d x Tr (DΦ)2 + c3 Φ3 + c4 Φ4 + . . . (G.42)
λ F =3 F =2
where we introduced the ’t Hooft coupling λ = g Y2 M N and ci are other coupling constants independent
of N . Following ’t Hooft [?], we consider the limit of large N with λ kept fixed. The propagator of an
adjoint field obeys
¬ ¶ λ
Φij Φkl ∝ δli δ kj (G.43)
N
where we used the fact that the adjoint representation can be represented as the direct product of
the fundamental and the anti-fundamental representation. This suggests that one can represent a
propagator by a double line, where each line denotes the flow of a fundamental index. Start by
considering the vacuum diagrams in this language. A diagram with V vertices, E propagators (or
edges) and F lines (or faces) scales as
‹V  ‹E ‹χ
λ
 
N N
F
N = λF , (G.44)
λ N λ

where χ = V + F − E = 2−2g is the minimal Euler character of the two dimensional surface where the
double line diagram can be embedded and g is the number of handles of this surface. Therefore, the
large N limit is dominated by diagrams that can be drawn on a sphere (g = 0). These diagrams are
called planar diagrams. For a given topology, there is an infinite number of diagrams that contribute
with increasing powers of the coupling λ, corresponding to tesselating the surface with more and
more faces. Figure G.1 shows two examples of vacuum diagrams in the double line notation. This
topological expansion has the structure of string perturbation theory with λ/N playing the role of the
string coupling. As we shall see this is precisely realized in maximally supersymmetric Yang-Mills
theory (SYM).


Let us now consider single-trace local operators of the form O = cJ Tr ΦJ , where cJ is a normalization
constant independent of N . Adapting the argument above, it is easy to conclude that the connected

160
correlators are given by a large N expansion of the form
∞
X
〈O1 . . . On 〉c = N 2−n−2g f g (λ) , (G.45)
g=0

which is dominated by the planars diagrams (g = 0). Moreover, we see that the planar two-point
function is independent of N while connected higher point functions are suppressed by powers of
N . This is large N factorization. In particular it implies that the two-point function of a multi-trace
operator Õ(x) =: O1 (x) . . . Ok (x) : is dominated by the product of the two-point functions of its
single-trace constituents
Y 1
Õ(x)Õ( y) ≈ 〈Oi (x)Oi ( y)〉 = , (G.46)
∆i
P
i (x − y)2 i

where we assumed that the single-trace operators were scalar conformal primaries properly normalized.
We conclude that the scaling dimension of the multi-trace operator Õ is given by i ∆i + O(1/N 2 ) .
P

In other words, the space of local operators in a large N CFT has the structure of a Fock space with
single-trace operators playing the role of single particle states of a weakly coupled theory. This is the
form of large N factorization relevant for AdS/CFT. However, notice that conformal invariance was
not important for the argument. It is well known that large N factorization also occurs in confining
gauge theories. Physically, it means that colour singlets (like glueballs or mesons) interact weakly in
large N gauge theories (see [?] for a clear summary).
The stress tensor has a natural normalization that follows from the action, Tµν ∼ Nλ Tr ∂µ Φ∂ν Φ . This

leads to the large N scaling

Tµ1 ν1 (x 1 ) . . . Tµn νn (x n ) c ∼ N 2 , (G.47)

which will be important below. This normalization of Tµν is also fixed by the Ward identities.

161
 Bibliography

[AAR99] George E. Andrews, Richard Askey, and Ranjan Roy. Special Functions. Encyclopedia of
Mathematics and its Applications. Cambridge University Press, 1999.

[AEMM19] Ahmed Almheiri, Netta Engelhardt, Donald Marolf, and Henry Maxfield. The entropy of
bulk quantum fields and the entanglement wedge of an evaporating black hole. JHEP,
12:063, 2019.

[AGM+ 00] Ofer Aharony, Steven S. Gubser, Juan Martin Maldacena, Hirosi Ooguri, and Yaron Oz.
Large N field theories, string theory and gravity. Phys. Rept., 323:183–386, 2000.

[AIS78] S. J. Avis, C. J. Isham, and D. Storey. Quantum Field Theory in anti-De Sitter Space-Time.
Phys. Rev. D, 18:3565, 1978.

[All85] Bruce Allen. Vacuum States in de Sitter Space. Phys. Rev. D, 32:3136, 1985.

[Ann12] Dionysios Anninos. De Sitter Musings. Int. J. Mod. Phys. A, 27:1230013, 2012.

[BBB+ 09] Christian Bär, Christian Becker, Romeo Brunetti, Klaus Fredenhagen, Nicolas Ginoux,
Frank Pfäffle, and Alexander Strohmaier. Quantum Field Theory on Curved Spacetimes.
Lecture Notes in Physics. Springer, 2009.

[BCH73] James M. Bardeen, B. Carter, and S. W. Hawking. The Four laws of black hole mechanics.
Commun. Math. Phys., 31:161–170, 1973.

[BD77] Claude W. Bernard and Anthony Duncan. Regularization and Renormalization of

Quantum Field Theory in Curved Space-Time. Annals Phys., 107:201, 1977.

[BD78] T. S. Bunch and P. C. W. Davies. Quantum Field Theory in de Sitter Space: Renormalization
by Point Splitting. Proc. Roy. Soc. Lond. A, 360:117–134, 1978.

[BD84] N. D. Birrell and P. C. W. Davies. Quantum Fields in Curved Space. Cambridge Monographs
on Mathematical Physics. Cambridge Univ. Press, Cambridge, UK, 2 1984.

[Bek73] Jacob D. Bekenstein. Black holes and entropy. Phys. Rev., D7:2333–2346, 1973.

[BF82] Peter Breitenlohner and Daniel Z. Freedman. Stability in Gauged Extended Supergravity.
Annals Phys., 144:249, 1982.

[BMS02] Raphael Bousso, Alexander Maloney, and Andrew Strominger. Conformal vacua and
entropy in de Sitter space. Phys. Rev. D, 65:104039, 2002.

[BS05] Antonio N. Bernal and Miguel Sanchez. Smoothness of time functions and the metric
splitting of globally hyperbolic space-times. Commun. Math. Phys., 257:43–50, 2005.

162
[BT13] R. Bott and L.W. Tu. Differential Forms in Algebraic Topology. Graduate Texts in Mathem-
atics. Springer New York, 2013.

[CGP14] Miguel S. Costa, Vasco Gonçalves, and João Penedones. Spinning AdS Propagators.
JHEP, 09:064, 2014.

[CJ86] P. Candelas and B. P. Jensen. Feynman green function inside a schwarzschild black hole.
Phys. Rev. D, 33:1596–1603, Mar 1986.

[CT68] N. A. Chernikov and E. A. Tagirov. Quantum theory of scalar fields in de Sitter space-time.
Ann. Inst. H. Poincare Phys. Theor. A, 9:109, 1968.

[DFM+ 99a] Eric D’Hoker, Daniel Z. Freedman, Samir D. Mathur, Alec Matusis, and Leonardo Rastelli.
Graviton and gauge boson propagators in AdS(d+1). Nucl. Phys. B, 562:330–352, 1999.

[DFM+ 99b] Eric D’Hoker, Daniel Z. Freedman, Samir D. Mathur, Alec Matusis, and Leonardo Rastelli.
Graviton exchange and complete four point functions in the AdS / CFT correspondence.
Nucl. Phys. B, 562:353–394, 1999.

[DO01] F. A. Dolan and H. Osborn. Conformal four point functions and the operator product
expansion. Nucl. Phys. B, 599:459–496, 2001.

[Dod03] Scott Dodelson. Modern Cosmology. Academic Press, Amsterdam, 2003.

[dRNS15] Adrián del Río and José Navarro-Salas. Equivalence of adiabatic and dewitt-schwinger
renormalization schemes. Phys. Rev. D, 91:064031, Mar 2015.

[EL10] Chas A. Egan and Charles H. Lineweaver. A Larger Estimate of the Entropy of the
Universe. Astrophys. J., 710:1825–1834, 2010.

[ESP12] Sheer El-Showk and Kyriakos Papadodimas. Emergent Spacetime and Holographic CFTs.
JHEP, 10:106, 2012.

[FG85] C. Fefferman and C.R. Graham. Conformal invariants. Societé Mathématique de France,
1985.

[Fri22] Alexander Friedman. über die Krümmung des Raumes. Z. Physik, 10:377–386, 1922.

[Fri24] Alexander Friedman. Über die Möglichkeit einer Welt mit konstanter negativer Krüm-
mung des Raumes. Z. Physik, 21:326–332, 1924.

[FS11] Guido Festuccia and Nathan Seiberg. Rigid Supersymmetric Theories in Curved Super-
space. JHEP, 06:114, 2011.

[FVP12] D.Z. Freedman and A. Van Proeyen. Supergravity. Cambridge University Press, 2012.

[Ger70] Robert P. Geroch. The domain of dependence. J. Math. Phys., 11:437–439, 1970.

[GH77a] G. W. Gibbons and S. W. Hawking. Action Integrals and Partition Functions in Quantum
Gravity. Phys. Rev. D, 15:2752–2756, 1977.

163
[GH77b] G. W. Gibbons and S. W. Hawking. Cosmological Event Horizons, Thermodynamics, and
Particle Creation. Phys. Rev. D, 15:2738–2751, 1977.

[GH94] G. W. Gibbons and S. W. Hawking, editors. Euclidean quantum gravity. 1994.

[Har17] Thomas Hartman. Lecture notes on classical de sitter space. 2017.

[Haw76] S. W. Hawking. Breakdown of Predictability in Gravitational Collapse. Phys. Rev. D,

14:2460–2473, 1976.

[HH83] J. B. Hartle and S. W. Hawking. Wave Function of the Universe. Phys. Rev. D, 28:2960–
2975, 1983.

[Kro23] Juan A. Valiente Kroon. Conformal Methods in General Relativity. Cambridge University
Press, 2023.

[KS11] Zohar Komargodski and Adam Schwimmer. On Renormalization Group Flows in Four
Dimensions. JHEP, 12:099, 2011.

[Kub57] R. Kubo. Statistical-mechanical theory of irreversible processes i. general theory and

simple applications to magnetic and conduction problems. Journal of the Physical Society
of Japan, 12:570–586, 1957.

[Lab00] J. M. F. Labastida. Knot invariants and Chern-Simons theory. In 3rd European Congress
of Mathematics: Shaping the 21st Century, 7 2000.

[Lem31] Abbé Georges Lemaître. A Homogeneous Universe of Constant Mass and Increasing
Radius accounting for the Radial Velocity of Extra-galactic Nebulae. Monthly Notices of
the Royal Astronomical Society, 91(5):483–490, 03 1931.

[Lem33] Abbé Georges Lemaître. L’Univers en expansion. Annales de la Société Scientifique de

Bruxelles, 53:51, January 1933.

[LM05] Jose Labastida and Marcos Mariño. Topological Quantum Field Thoery and Four Manifolds.
Springer, 2005.

[LT98] Hong Liu and Arkady A. Tseytlin. D = 4 superYang-Mills, D = 5 gauged supergravity,

and D = 4 conformal supergravity. Nucl. Phys. B, 533:88–108, 1998.

[LT99] Hong Liu and Arkady A. Tseytlin. On four point functions in the CFT / AdS correspond-
ence. Phys. Rev. D, 59:086002, 1999.

[Mar01] M. Marino. An introduction to Donaldson-Witten theory. In Summer School 2001 on

Geometric and Topological Methods for Quantum Field Theory, pages 265–311, 7 2001.

[Mar05] Marcos Mariño. Chern-Simons theory and knot invariants. In Chern-Simons Theory,
Matrix Models, and Topological Strings. Oxford University Press, 09 2005.

[Moo12] G. Moore. A very long lecture on the physical approach to donaldson and seiberg-witten
invariants of four-manifolds. 2012.

164
[Mot85] E. Mottola. Particle Creation in de Sitter Space. Phys. Rev. D, 31:754, 1985.

[MS59] Paul C. Martin and Julian Schwinger. Theory of many-particle systems. i. Physical Review,
115:1342–1373, 1959.

[Nab10] G.L. Naber. Topology, Geometry and Gauge fields: Foundations. Texts in Applied Mathem-
atics. Springer New York, 2010.

[Nak90] Mikio Nakahara. Geometry, topology and physics. Graduate student series in physics.
Hilger, Bristol, 1990.

[O’N83] Barrett O’Neill. Semi-Riemannian Geometry With Applications to Relativity, 103, Volume
103 (Pure and Applied Mathematics). Academic Press, 1983.

[Paj20] Enrico Pajer. Field Theory in Cosmology. Part III, 2020.

[Pen20] Geoffrey Penington. Entanglement Wedge Reconstruction and the Information Paradox.
JHEP, 09:002, 2020.

[Poi04] Eric Poisson. A Relativist’s Toolkit: The Mathematics of Black-Hole Mechanics. Cambridge
University Press, 2004.

[PR84] Roger Penrose and Wolfgang Rindler. Spinors and Space-Time. Cambridge Monographs
on Mathematical Physics. Cambridge University Press, 1984.

[PS95] M.E. Peskin and D.V. Schroeder. An Introduction To Quantum Field Theory. Frontiers in
Physics. Avalon Publishing, 1995.

[PT09] Leonard E. Parker and D. Toms. Quantum Field Theory in Curved Spacetime: Quant-
ized Field and Gravity. Cambridge Monographs on Mathematical Physics. Cambridge
University Press, 8 2009.

[Rob35] H. P. Robertson. Kinematics and World-Structure. Astrophysical Journal, 82:284, Novem-

ber 1935.

[Rob36a] H. P. Robertson. Kinematics and World-Structure II. Astrophysical Journal, 83:187, April
1936.

[Rob36b] H. P. Robertson. Kinematics and World-Structure III. Astrophysical Journal, 83:257, May
1936.

[Sre07] M. Srednicki. Quantum Field Theory. Cambridge University Press, 2007.

[SS76] Christiane Schomblond and Philippe Spindel. Conditions d’unicité pour le propagateur
δ1 (x, y) du champ scalaire dans l’univers de de Sitter. Annales de l’institut Henri Poincaré.
Section A, Physique Théorique, 25(1):67–78, 1976.

[SSV01] Marcus Spradlin, Andrew Strominger, and Anastasia Volovich. Les Houches lectures on
de Sitter space. In Les Houches Summer School: Session 76: Euro Summer School on Unity
of Fundamental Physics: Gravity, Gauge Theory and Strings, pages 423–453, 10 2001.

165
[STY95] Misao Sasaki, Takahiro Tanaka, and Kazuhiro Yamamoto. Euclidean vacuum mode
functions for a scalar field on open de Sitter space. Phys. Rev. D, 51:2979–2995, 1995.

[SV96] Andrew Strominger and Cumrun Vafa. Microscopic origin of the Bekenstein-Hawking
entropy. Phys. Lett., B379:99–104, 1996.

[Unr76] W. G. Unruh. Notes on black hole evaporation. Phys. Rev. D, 14:870, 1976.

[VP99] Antoine Van Proeyen. Tools for supersymmetry. Ann. U. Craiova Phys., 9(I):1–48, 1999.

[Wal37] A. G. Walker. On milne’s theory of world-structure. Proceedings of the London Mathemat-

ical Society, s2-42:90–127, 1937.

[Wal84] Robert M. Wald. General Relativity. The University od Chicago Press, 1984.

[Wei95] Steven Weinberg. The Quantum Theory of Fields, volume 1-3. Cambridge University
Press, 1995.

[Zam86] A. B. Zamolodchikov. Irreversibility of the Flux of the Renormalization Group in a 2D

Field Theory. JETP Lett., 43:730–732, 1986. [Pisma Zh. Eksp. Teor. Fiz.43,565(1986)].

166