Advanced General Relativity

and Quantum Field Theory

in Curved Spacetimes
Lecture Notes

Stefano Liberati

SISSA, IFPU, INFN

January 2020
 Preface

The following notes are made by students of the course of “Advanced General Relativity and Quantum
Field Theory in Curved Spacetimes”, which was held at the International School of Advanced Studies
(SISSA) of Trieste (Italy) in the year 2017 by professor Stefano Liberati.
Being the course directed to PhD students, this work and the notes therein are aimed to inter-
ested readers that already have basic knowledge of Special Relativity, General Relativity, Quantum
Mechanics and Quantum Field Theory; however, where possible the authors have included all the
definitions and concept necessary to understand most of the topics presented.
The course is based on di↵erent textbooks and papers; in particular, the first part, about Advanced
General Relativity, is based on:
• “General Relativity” by R. Wald [1]

• “Spacetime and Geometry” by S. Carroll [2]

• “A Relativist Toolkit” by E. Poisson [3]

• “Gravitation” by T. Padmanabhan [4]

while the second part, regarding Quantum Field Theory in Curved Spacetime, is based on
• “Quantum Fields in Curved Space” by N. C. Birrell and P. C. W. Davies [5]

• “Vacuum E↵ects in strong fields” by A. A. Grib, S. G. Mamayev and V. M. Mostepanenko [6]

• “Introduction to Quantum E↵ects in Gravity”, by V. F Mukhanov and S. Winitzki [7]

Where necessary, some other details could have been taken from paper and reviews in the standard
literature, that will be listed where needed and in the Bibliography.
Every possible mistake present in these notes are due to misunderstandings and imprecisions of
which only the authors of the following works must be considered responsible.

Credits
• Andrea Oddo (Chapters 1, 2, 3, 6, 7, Appendix A; revision)

• Giovanni Tambalo (Chapters 8, 9, 10, 11)

• Lumen Boco (Chapters 4, 5)

Also thanks to Giovanni Tricella and Paolo Campeti for useful scientific discussions.

1
 Contents

I Advanced General Relativity 5

1 Foundations of Relativity 6
1.1 Definitions and postulates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2 Axiomatic derivation of Special Relativity . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3 Equivalence Principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

2 Di↵erential Geometry 13
2.1 Manifolds, Vectors, Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.1.1 Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.1.2 Dual vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2 Tensor Densities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.3 Di↵erential forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.4 The Volume element . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.5 Curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.6 Intrinsic and Extrinsic curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.6.1 Example: Intrinsic-Extrinsic (Gaussian) curvature of two surfaces . . . . . . . . 21
2.6.2 Hypersurfaces and their Extrinsic curvature . . . . . . . . . . . . . . . . . . . . 21
2.7 Lie Derivative . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.8 Killing Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.9 Conserved quantities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

3 Kinematics 29
3.1 Geodesic Deviation Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.2 Raychaudhuri Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.2.1 Time-like geodesics congruence . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.2.2 Null geodesics congruence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.3 Energy Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

4 Variational Principle 37
4.1 Lagrangian Formulation of General Relativity . . . . . . . . . . . . . . . . . . . . . . . 37
4.1.1 Gibbons-Hawking-York counterterm . . . . . . . . . . . . . . . . . . . . . . . . 40
4.1.2 Schrödinger action . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.1.3 Palatini variation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

2
5 Alternative Theories of Gravity 44
5.1 Holding SEP: GR in D 6= 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
5.1.1 D < 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
5.1.2 D > 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
5.2 Holding SEP: Lanczos-Lovelock theories . . . . . . . . . . . . . . . . . . . . . . . . . . 46
5.2.1 More about the Euler characteristic . . . . . . . . . . . . . . . . . . . . . . . . 48
5.3 Relaxing SEP: F (R) Theories as a first example of higher curvature gravity . . . . . . 49
5.3.1 Palatini variation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
5.4 Relaxing SEP: Generalized Brans–Dicke theories . . . . . . . . . . . . . . . . . . . . . 51
5.4.1 F (R) theories as generalized Brans–Dicke theories . . . . . . . . . . . . . . . . 52
5.5 Relaxing SEP: Scalar-Tensor Theories . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
5.5.1 Einstein frame . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
5.5.2 Jordan frame . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
5.5.3 Horndeski theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
5.5.4 DHOST Theories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
5.6 Going further . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

6 Global Methods 57
6.1 Carter–Penrose Diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
6.1.1 Minkowski spacetime Carter–Penrose diagram . . . . . . . . . . . . . . . . . . . 58
6.1.2 Schwarzschild spacetime Carter–Penrose diagram . . . . . . . . . . . . . . . . . 60
6.2 Black Holes: Singularities and Event Horizon . . . . . . . . . . . . . . . . . . . . . . . 63
6.2.1 Singularities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
6.2.2 Horizons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
6.3 Killing Horizons and Surface Gravity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
6.3.1 Surface gravity: alternative definitions . . . . . . . . . . . . . . . . . . . . . . . 68
6.4 Cauchy Horizon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
6.5 Cosmological Horizon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
6.6 Local black hole horizons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
6.6.1 Trapped Surfaces and Trapped Regions . . . . . . . . . . . . . . . . . . . . . . 73
6.6.2 Apparent Horizon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
6.6.3 Trapping Horizon, Dynamical Horizon, Isolated Horizon . . . . . . . . . . . . . 75
6.7 Singularity Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77

7 From Black Hole Mechanics to

Black Hole Thermodynamics 81
7.1 Rotating Black Holes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
7.1.1 A ring Singularity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
7.1.2 Horizons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
7.1.3 Killing vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
7.2 Ergosphere and Penrose Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
7.2.1 Superradiance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
7.3 Black Hole Thermodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
7.3.1 Null hypersurfaces, Surface integrals and other tools . . . . . . . . . . . . . . . 90
7.3.2 Surface element . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
7.3.3 Komar Mass and the Smarr formula . . . . . . . . . . . . . . . . . . . . . . . . 92
7.3.4 On the necessity of Hawking radiation . . . . . . . . . . . . . . . . . . . . . . . 97
7.3.5 Hawking radiation and black hole entropy . . . . . . . . . . . . . . . . . . . . . 98

3
II Quantum Field Theory in Curved Spacetimes 101

8 Preliminaries 102
8.1 Second quantization in curved spacetime . . . . . . . . . . . . . . . . . . . . . . . . . . 102
8.2 Particle production by a time-varying potential . . . . . . . . . . . . . . . . . . . . . . 105
8.3 Coherent and Squeezed states . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

9 E↵ects of Quantum Field Theory in Curved Spacetime 111

9.1 The Casimir e↵ect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
9.1.1 1+1 dimensional case: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
9.2 The Unruh E↵ect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
9.2.1 The Rindler Wedge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
9.2.2 Quantum field theory on the Rindler Wedge . . . . . . . . . . . . . . . . . . . . 117

10 Evaporating Black Holes 121

10.1 Moving Mirrors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
10.1.1 Field theory with moving boundaries . . . . . . . . . . . . . . . . . . . . . . . . 121
10.1.2 Quantum field theory with moving boundaries . . . . . . . . . . . . . . . . . . 122
10.1.3 Special case: the exponentially receding mirror . . . . . . . . . . . . . . . . . . 123
10.2 Hawking Radiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
10.3 Open issues with Hawking radiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
10.3.1 Trans-planckian problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
10.3.2 The information loss problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
10.4 Black Hole Entropy Interpretations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137

11 Semiclassical Gravity 141

11.1 Semiclassical Einstein Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
11.1.1 Approximation Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
11.2 Renormalization of the Stress-Energy Tensor . . . . . . . . . . . . . . . . . . . . . . . 143
11.2.1 Point-splitting Renormalization . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
11.3 Semiclassical Collapse . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149

12 Time Machines and Chronology Protection 156

12.1 Time travel paradoxes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
12.2 Time Machines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
12.2.1 Time machines from globally rotating solutions . . . . . . . . . . . . . . . . . . 158
12.2.2 Wormholes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
12.2.3 Alcubierre Warp Drives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
12.3 Possible solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
12.3.1 Preemptive chronology protection? . . . . . . . . . . . . . . . . . . . . . . . . . 167
12.3.2 Superluminal warp drive instability . . . . . . . . . . . . . . . . . . . . . . . . . 167

Bibliography 172

4
 Part I

Advanced General Relativity

5
 1
Foundations of Relativity

1.1 Definitions and postulates

We often refer to spacetime as a metaphysical entity, but in physics this would never be the case;
therefore, our first task is to define both space and time in an operative fashion. We can replace the
concepts of time with duration measured by a clock and space with distance measured with a rod. This
is also because duration and distance are experimental quantities, that are directly meaningful, and
that don’t need any interpretation whatsoever. With this, the properties of space and time are now
directly related to the respective properties of rods and clocks. This gives us an operative definition
of spacetime, rather than a metaphysical definition.
We can then define an event. We will assume this to be a primitive notion in physics, in the same
way a point is a primitive notion in geometry. In particular, an event is something that happens for a
very small period time t in a very small region of space x; in this way, we can define our spacetime
as the collection of all possible events: given a generic event E, a spacetime M is
[
M⌘ E. (1.1)

We can postulate that every measurement is either local or reducible to local, and in this way
we can introduce the notion of an observer. We can imagine to place an observer in each point of
space, and hence define a set of observers Oi in our spacetime M, that measure in di↵erent locations
of spacetime. All these observers have trajectories in spacetime, and therefore we can define the set
of these worldlines as a congruence of observers, one for each point in M, in the sense that the
worldlines of all the observers never cross. The set of these worldines is U ⇢ M.

O1 O2 O3 O4 O5

Figure 1.1: Pictorial representation of a congruence of worldlines, each one representing one observer
(or a point in M).

One can go from an observer to another by performing a boost from the frame of the first observer
to the frame of the second observer through a coordinate transformation. We want now to relate

6
observations in di↵erent frames, and therefore we want to agree on a system of rulers and clocks.
Once we agree on a recipe to build a clock (e.g. quartz oscilation, period of radiation from atomic
transitions, etc...), we can give a clock to each observer in the set, thus introducing the concept of
local time.
To do kinematics, we need to relate measures made by di↵erent observers, so we need to synchronize
clocks. In order to do this, we want a way to relate measures of di↵erent observers: we introduce
a function t : U ! R, such that t(Oi ) gives the proper time on that worldline. Synchronizing the
clocks means to characterize the di↵erence in time, or the lapse of time, between two events: t =
t(E2 ) t(E1 ). Events E1 and E2 are said to be simultaneous if t(E1 ) = t(E2 ).
We can now separate space from spacetime and define it in the following way: given a time t0 we
can define the space St0 at t0 as the set of events such that their time is t0

St0 = {E | t(E) = t0 } , (1.2)

that is, a slice at some t0 that we can call “space”. Note that there is no independent notion of space
from the system of observers. We can also define co-local events as events that happen on the same
worldline of an observer.
If an experiment takes place in a sufficiently small region of spacetime, then we can introduce a
postulate: there is always a system of observers, along with a procedure of synchronization and choices
of length units, so that the distance x that exists between two observers does not depend on time,
and this distance respects the euclidean geometry axioms. In practice, it is always possible to find a
region of spacetime that is locally euclidean.
In general we can choose time arbitrarily, but not all choices are equivalent: we naturally prefer to
have a simple function defining time, for example a time function such that the equation of motion is
d2 ~x
= 0; (1.3)
dt2
a change of the time function implies the introduction of t0 , such that t = f (t0 ), and therefore

d2 ~x d2 f d~x
= ; (1.4)
dt02 dt02 dt
now, if we want to find another function in which the equation of motion is still valid, we need to
transform time in such a way that t0 (t) = at + b, where a and b are constants. With this function,
corresponding to a shift of the origin and a change of units, the law describing the equation of motion
is conserved.1
So, locally space is euclidean; this is telling us that space is homogeneous and isotropic: there is
no privileged point or preferred direction. For what time is concerned, we choose our clocks in a way
that time is homogeneous: every point in time is not preferred and laws don’t change if we perform
measurements at di↵erent times. There is no isotropy in time, however, since time is not reversible.
2
A system of observers in which ddt2x = 0 is called an inertial reference frame. At this point, we
introduce a second postulate: given an inertial reference frame, another frame moving with uniform
rectilinear motion with respect to the first one is again inertial. So, given an inertial reference frame
K, another reference frame K0 is inertial if the velocity v between the two frames is constant. This is
the kinematical formulation of relativity principle. There is also a stronger, dynamical, formulation
of the relativity principle: physical laws in any reference frame are unchanged, and dynamics appear
the same in all inertial frames.
Now we want to add the condition of precausality: changing the reference frame doesn’t change
the order of co-local events.
We now have all the ingredients to derive Special Relativity axiomatically, without knowing any-
thing about the speed of light.
1
Note also that changing units should not change physics: there should be a way to formulate physical laws so that
they are evidently unit independent.

7
1.2 Axiomatic derivation of Special Relativity
There is a theorem by von Ignatowski (1911), where it is showed that with all previous assumptions,
being
• Relativity principle

• Spatial and temporal homogeneity

• Spatial isotropy

• Precausality
we can derive Special Relativity. Let us introduce two reference frames, K(O, x, y, z, t) and K̄(Ō, x̄, ȳ, z̄, t̄)
the former moving with a velocity v with respect to the latter. We can always set t = x = y = z = 0
if t̄ = x̄ = ȳ = z̄ = 0 without loss of generality. We want to find coordinate transformations between
the two systems, xµ and x̄µ ; in particular: x̄µ = f µ (t, x, y, z, v).
Homogeneity implies that the time or space intervals measured in K̄ can depend only on the
corresponding intervals in K, not on the precise instants or positions in which the measurements are
performed.
µ @f µ
We can write dx̄µ = @f ⌫
@x⌫ dx , and due to homogeneity we also have @x⌫ = const since it cannot
depend on x; therefore f µ must be a linear function of the coordinates xµ . For simplicity, we can
assume v to be along one of the spatial directions, let’s say x: v = (v, 0, 0). We can then write
t̄ = A(v)t + B(v)x and x̄ = C(v)t + D(v)x.
A, B, C and D cannot be all independent. In fact, consider the motion of the K origin, O, w.r.t. K̄:
we are comoving with K̄, therefore at t̄ = 0, O is in x̄ = 0. Later O will be at x̄O = v t̄. But xO = 0,
therefore t̄ = A(v)t, x̄O = C(v)t, therefore x̄O = C(v)
A(v) t̄, and then

C(v) = vA(v) . (1.5)

Conversely, Ō moves in K with some velocity v̄; so xŌ = v̄t, x̄Ō = 0, therefore C(v)t + D(v)xŌ = 0,
and then
C(v) = v̄D(v) . (1.6)
By putting together the two previous boxed relations 1.5 and 1.6, we can find
C(v) v
D(v) = = A(v). (1.7)
v̄ v̄
Then, the transformation law can be written in matrix form as
✓ ◆ ✓ ◆✓ ◆
t̄ 1 ⇠(v) t
= (v) v , (1.8)
x̄ v v̄ x

where we defined ⇠(v) ⌘ B(v)

A(v) and renamed A(v) ⌘ (v).
Given the relativity principle, we only know that if v̄ = f (v), then v = f (v̄). However, we can use
the isotropy of space to ask for invariance after inversion of the spatial axis and velocity. (Reciprocity
principle) [8, 9]. In order to see this, let us start by defining the overall transformation as ⇤(v)
✓ ◆ ✓ ◆
t̄ t
= ⇤(v) , (1.9)
x̄ x

Note that for v = 0, we must have that the transformation matrix reduces to the identity, ⇤(0) = 1,
therefore we know that ⇠(0) = 0 and (0) = 1.
The above statement about isotropy implies that we should get the same transformation between
(t, x) and (t̄, x̄) coordinates if we invert the direction of the the x and x̄ axes (so that also v̄ ! v̄

8
and v ! v). This in turns implies that ( v) = (v), ⇠( v) = ⇠(v) and that v̄ = f ( v) as
well. However, we also know that v̄ = f (v), hence we conclude that f ( v) = f (v). Evidently we
must then have that f = ±I with the + option unphysical as it would lead to the conclusion that
x̄ = x for the v = 0 limit of our transformation. Hence we get in the end v̄ = v, and therefore our
transformation law becomes
✓ ◆ ✓ ◆✓ ◆
t̄ 1 ⇠(v) t
= (v) . (1.10)
x̄ v 1 x

Now, due to the relativity principle, ⇤ 1 (v) = ⇤(v̄), such that

✓ ◆ ✓ ◆
t t̄
= ⇤(v̄) . (1.11)
x x̄

But ⇤(v̄) = ⇤( v), therefore ⇤(v)⇤( v) = ⇤(v)⇤ 1 (v) = 1, where the inverse transformation matrix
is ✓ ◆
1 1 (v) (v)⇠(v)
⇤ (v) = , (1.12)
v (v) (v)
where = det ⇤ = 2 v 2 ⇠(v). Now, by imposing that ⇤(v)⇤( v) = 1 (and using that (v) is even
and ⇠(v) is odd under v ! v) we find 2 v 2 ⇠(v) = 1. This implies that = 1 and, that is
(discarding the negative solution as it would imply (0) = 1),
1
(v) = p , (1.13)
1 v⇠(v)

So that in the end ✓ ◆

1 1 ⇠(v)
⇤(v) = p . (1.14)
1 v⇠(v) v 1
We can now use the relativity principle to impose a group structure. Introduce a third reference
¯ which moves with constant velocity u with respect to K and ū with respect to K̄. Again,
frame K̄
without loss of generality, let us reduce the problem to the case in which u = (u, 0, 0) and ū = (ū, 0, 0).
Therefore, ū is a function of both v and u: ū ⌘ (u, v). Now, by simply applying the transformation
laws between couples of reference frames, we can write
✓ ◆ ✓ ¯ ◆ ✓ ◆ ✓ ◆
t̄ t̄ t̄ t
= ⇤(ū) ¯ = ⇤(v) , (1.15)
x̄ x̄ x̄ x

which in turn simply implies

✓ ◆ ✓ ¯ ◆ ✓ ¯ ◆
t̄ 1 t̄ t̄
=⇤ (v)⇤(ū) ¯ = ⇤(u) ¯ . (1.16)
x̄ x̄ x̄

Therefore, ⇤ 1 (v)⇤(ū) = ⇤(u), ⇤(ū) = ⇤(u)⇤(v), and we have found the composition of transforma-
tions:
⇤( (u, v)) = ⇤(u)⇤(v) , (1.17)
The above relation (together with the statement that if u⇤ is such that ū = (u⇤ , v) = 0, then
⇤(v)⇤(u⇤ ) = ⇤(0) = 1) implies that our transformations form a group.
From the previous boxed relation, we can also see that
✓ ◆ ✓ ◆
1 + u⇠(v) ⇠(u) + ⇠(v) 1 ⇠( )
(u) (v) = ( ) , (1.18)
v+u 1 + v⇠(u) 1

therefore
( ) = (u) (v) (1 + u⇠(v)) = (u) (v) (1 + v⇠(u)) , (1.19)

9
which implies
u⇠(v) = v⇠(u). (1.20)
This equation has two possible solutions. We could have ⇠(u) = ⇠(v) = 0: in this case,
✓ ◆
1 0 1
⇤(v) = (v) (v) = p = 1, (1.21)
v 1 1 v⇠(v)

and we find the Galilean transformations.

Another solution is given by:
u ⇠(u)
= ; (1.22)
v ⇠(v)
this implies ⇠(u) / u, ⇠(u) = ↵u, and therefore
✓ ◆
1 1 ↵v
⇤(v) = p . (1.23)
1 ↵v 2 v 1

The composition law implies (u) (v)(u + v) = ( ) ; by squaring both sides of the equation, we
find
2 1
(u) 2 (v)(u + v)2 = 2 ( ) 2 = 2
; (1.24)
1 ↵ 2
solving for gives
u+v
(u, v) = , (1.25)
1 + ↵uv
with ↵ > 0. It is evident that ↵ has the dimension of the inverse of the square of a velocity; also,
note, that if ↵ < 0, it would be possible to get a composite velocity equal to infinity even if the two
velocities are finite. We shall hence write ↵ ⌘ C12 , where C is a constant. Therefore,
✓ v
◆
1 1 C2
u+v
⇤(v) = q (u, v) = ; (1.26)
1 v2 v 1 1 + Cuv2
C2

if either u or v is equal to C, we find (u, v) = C.

In order to conclude our analysis we now need to impose the precausality condition:

@ t̄
> 0, (1.27)
@t
which gives (v) > 0 for any v, ↵ > 0, ↵v 2 < 1 and, in conclusion, v 2 < C 2 . So basically we have
recovered the Lorentz transformation and the relativistic composition law of velocities modulo the fact
that we have an undetermined velocity constant C appearing in our equations.
1 ⇣ v ⌘
t̄ = q t + 2x
1 Cv 2
2 C

1
x̄ = q (vt + x)
v2
1 C2

Experimentally, it is found that C is actually the speed of light, C = c. But in general, the value of
C depends on the physics behind the clocks and rods used to construct our spacetime: for example, for
a bat using sonar to probe its environment the corresponding C would naturally be the speed of sound.
(see eg. A. Trautman in Postepy Fizyki 45, 1 (1994) (in Polish). http://www.fuw.edu.pl/ amt/BATPOL.pdf).
What happens if we break some of the assumptions we made? Relaxing the relativity principle
implies the existence of a privileged frame, and this brings to all kinds of aether theories; breaking
isotropy [10, 11] of space implies finding a new symmetry group of transformations that depends both

10
on positions and velocities (Finsler geometry); breaking homogeneity means that spacetime locally is
no longer euclidean, we are losing the operative meaning of our definitions of space and time, and this
could require again for the geometry to be Finsler; breaking precausality is in general the more exotic
case.
For more details on von Ignatowski theorem look at references [12, 13, 14].

1.3 Equivalence Principles

Now, spacetime is locally Minkowski, so now we can go a step beyond and see what are the principles
that are governing physics.
A crucial role is played by the equivalence principle; actually there are a lot of equivalence principles
we can formulate. For a review on Equivalence Principles, see [15].

Newton Equivalence Principle (NEP) This states that in the newtonian limit the inertial mass
is equal to the gravitational mass, mi = mg . According to Galileo, masses shouldn’t enter in the laws
of free fall; in particular, take two di↵erent objects at any distance and let them fall; now join the two
objects: they fall as one object, but in the same way as the two distinct objects. Therefore, the mass
shouldn’t be involved in the laws of free fall.
This principle makes sense only in theories of gravity that admit a Newtonian limit. If (x(t), t)
is the Newtonian potential, then mi ẍ(t) = mg r (x(t), t), and thus mi ⌘ mg , since from Newton’s
third law we cannot distinguish between active and passive gravitational masses.
A possible test of this principle can be performed using pendula with di↵erent bobs of di↵erent
weight and nature but same length, that oscillate with the same period.

Weak Equivalence Principle (WEP) This is the universality of free fall: the motion under
gravitational forces of bodies does not depend on their weights or compositions.
More specifically, we want to consider bodies that are test particles and that are not self-gravitating:
a test particle is a particle that is not modifying the gravitational field in which it is falling in (so
it has no backreaction on the gravitational field); the self-gravitating nature of an object, instead, is
connected to its compactness , or the ratio between its gravitational energy and inertial energy

GN m2g 1
⌘ , (1.28)
r mi c2
where r is the scale dimension of the object. If NEP holds, mi = mg = m, and therefore

GN m
= ; (1.29)
c2 r
when ⌧ 1, the body is not self-gravitating.
Note that being a test particle depends on what background one considers, e.g. the Moon is a test
particle for the Sun, but it is not a test particle for the Earth.
Tests of this principle are, for example, torsion balances à la Eötvos.

Gravitational Weak Equivalence Principle (GWEP) This is exactly like the WEP plus the
self-gravity condition. In practice: does a test particle follow the same geodesic whether it has self-
gravity or not? Does having self-gravity make any di↵erence in the motion of the object? If not,
the GWEP holds. This principle turns out to be true only in theories of gravity with second order
field equations: General Relativity (with and without cosmological constant), Nordstrom gravity and
Lanczos-Lovelock theories for D > 4, [16].
A test for this equivalence principle is the Lunar Laser Ranging experiment.

11
Einstein Equivalence Principle (EEP) This states that fundamental non-gravitational test
physics is not a↵ected locally at any point of spacetime by the presence of gravitational field: you can
completely geometrise the action of the gravitational field. Why is it necessary to talk about funda-
mental non-gravitational test physics? Because, for example, we want to avoid composite systems,
e.g. two particles. Given two particles at a distance `, we would have

`¨ @2
/ ⇠ ⌧ij ,
` @xi @xj
where ⌧ij is the tidal field, that is zero in flat spacetime but di↵erent from zero in the presence of
gravitational fields.
This principle can be reformulated à la Cli↵ord Will: in particular,

EEP = WEP + LLI + LPI,

where WEP is the Weak Equivalence Principle, LLI stands for Local Lorentz Invariance and LPI
stands for Local Position Invariance. Basically, with the WEP you can describe gravity geometrically
(you can define an affine connection); with the LLI, it can be shown that local physical laws obey
Special Relativity, i.e. you can “undo” gravity and go in a free fall (local inertial frame) where non-
gravitational physics obey Special Relativity; LPI allows you to do this in every point of spacetime.
This equivalence principle is at the heart of metric theories of gravity.
A test of this principle could be to test the constancy of the fine-structure constant ↵, or of mass
ratios (mp /me ).

Strong Equivalence Principle (SEP) Essentially, this is the extension of the EEP to gravitational
physics: all fundamental test physics is not a↵ected locally by the presence of the gravitational field.
À la Will, SEP = GWEP + LLI + LPI: the SEP requires the GWEP, the universality of free-fall
extended to self-gravitating bodies, and the LLI and LPI; therefore, it requires that GN = const, and
this has been tested up to 10% of confidence.
This principle selects two theories of gravity: General Relativity and the Nordstrom gravity, which
is characterised by an equation, which is

R = 24⇡GN T ; (1.30)

however this Nordstrom gravity does not predict the deviation of light and has been therefore ruled
out.
All these equivalence principles are related. In particular, the WEP implies the NEP, but the
converse is not generally true, but only if masses enter in the equations of motion via their ratio,
mi /mg .
Also, the GWEP implies the WEP in the limit for = 0. Note that if the GNEP (which is the
NEP for self-gravitating bodies) fails, then also the GWEP fails.
Finally, the SEP implies both the GWEP and the EEP separately.
However, there are still some unsolved relations, the Schi↵’s conjecture and the Gravitational
Schi↵’s conjecture. Respectively, it is still unknown if the WEP implies the EEP or whether the
GWEP (with the possible addition of the EEP) implies the SEP.

12
 2
Di↵erential Geometry

2.1 Manifolds, Vectors, Tensors

The WEP implies that gravity is actually due to the geometry of spacetime. This pushes us to
introduce a connection, ⇢µ⌫ , and to promote the derivative operator @µ to a covariant derivative
operator rµ .
The EEP implies that, locally, spacetime is flat, a euclidean plane, gµ⌫ = ⌘µ⌫ , and therefore we
can compute the connection as a function of the metric g. Thus, spacetime must have the structure
of a di↵erential manifold.
We can define a manifold as a set of points M so that, in each point of M, there is an open
neighborhood which has a continuous one-to-one map onto an open set of Rn for some n, meaning
: M ! Rn .
In poor-man words, M might be a complicated space with non-trivial topology and curvature,
however locally it has to look like an n-dimensional euclidean space Rn .
The map is generally C p , and in this case the manifold is said to be C p di↵erentiable; if p ! 1,
then the manifold M is called smooth. Di↵erentiability is an important property, since it allows to
introduce tensor-like structures on the manifold.

2.1.1 Vectors
As a matter of fact, we can now define elementary objects on the manifold, like vectors. We can
imagine each vector as located at some point p 2 M. Conversely, all the vectors v at a point p 2 M
form a tangent space Tp . In general, the vector v is independent on the choice of coordinates, therefore
it should be possible to describe Tp without introducing any coordinate dependence.

Figure 2.1: Pictorial representation of the tangent space Tp to the point p 2 M

We could consider all the parametrized curves through p, : R ! M and their tangent vectors,
but this would be coordinate dependent, since we would write = xµ ( ) in some coordinate system.

13
 Let us try to consider, instead, F: the space of all the smooth functions on M, meaning that
f 2 F if f : M ! R and if it is C 1 . Then, we notice that each curve through p defines an operator on
F, the directional derivative, which maps f ! df d , where is an arbitrary parameter for the curve
p
through p.
So, the tangent space at p, Tp , can be identified with all the directional derivative operators along
curves through p, meaning that dd maps f in df d . It can be shown that Tp defines a vector space with
the same dimension as M, that the directional derivative acts linearly, and that it satisfy the Leibniz
rule [2, p. 63].
Now, given a coordinate system {xµ } in some open neighborhood U of p, these coordinates provide
an obvious set of n-directional derivatives at p, being this the set of partial derivatives at p, {@µ }:

d dxµ @f dxµ
f= = @µ f, (2.1)
d d @xµ d
µ
and therefore dxd are the components of the vector dd in the coordinate basis {xµ }.
Now, we can write the vector by expanding it in components through the choice of this natural
basis: v = v µ @µ , where v µ are the components and @µ is the basis of vectors.
Since the vector must be independent from the coordinate set, under a coordinate transformation
xµ ! x0 µ , the vector must behave as
0 0 @xµ
v µ @ µ = v = v µ @ µ0 = v µ @µ , (2.2)
@xµ0
therefore, the components of the vector transform as
0
0 @xµ
vµ = vµ . (2.3)
@xµ
Since a vector at a point can be thought as a directional derivative along a path through that point,
then a vector field defines a map from smooth functions to smooth functions over the manifold by
taking a derivative at each point. So, given two vector fields X and Y one can define their commutator
as
[X, Y ](f ) = X(Y (f )) Y (X(f )), (2.4)
which, in coordinate basis can be written as

[X, Y ]µ = X @ Y µ Y @ X µ; (2.5)

notice that the commutator is linear and obeys Leibniz rule; also, this is coordinate independent in
spite of the dependence on the partial derivative.

2.1.2 Dual vectors

Then, we can define a dual vector through the definition of cotangent space: a cotangent space Tp?
is the set of linear maps ! : Tp ! R, such that !(v) 2 R. Then, the map ! is a dual vector, ! 2 Tp? .
We can then define df 2 Tp? in such a way that df dd ⌘ df d
d , where d 2 Tp .
? µ
If ! 2 Tp , the natural choice of basis, in this case, is dx , therefore we can write

! = !µ dxµ ; (2.6)

as before, we want this object to be independent from the coordinate system, therefore, under coor-
dinate transformation, the components !µ must transform as

@xµ
! µ0 = !µ . (2.7)
@xµ0

14
 The basis of vectors and dual vectors, also, are defined in such a way that
@xµ
dxµ @⌫ = = µ
⌫, (2.8)
@x⌫
where µ ⌫ is the Kronecker delta.
If we apply a vector v to a dual vector !, we get

!µ dxµ v ⌫ @⌫ = !µ v ⌫ µ
⌫ = (! · v) 2 R, (2.9)

which is an inner product.

Now that we have vectors and dual vectors, we can also generate tensors, as multilinear maps
from a collection of k vectors and l dual vectors; thus, a generic tensor of rank (k, l) is given by

T = T µ1 ...µk⌫1 ...⌫l @µ1 ⌦ · · · ⌦ @µk ⌦ dx⌫1 ⌦ · · · ⌦ dx⌫l , (2.10)

where the symbol ⌦ indicates a tensor product. With this, similarly as we did before, we can write
the transformation of the components of T under a coordinate transformation as
0 0 0 0
µ0 ...µ0 @xµ1 @xµk @x⌫1 @x⌫l µ1 ...µk
T 1 k⌫ 0 ...⌫ 0 = . . . . . . T ⌫1 ...⌫l . (2.11)
1 l @xµ1 @xµk @x⌫1 @x⌫l
Notice that, with this representation, vectors are (1, 0) tensors, dual vectors are (0, 1) tensors, and
scalars are (0, 0) tensors.
Now, we can introduce the metric tensor, which is a symmetric, non-degenerate, rank (0, 2)
tensor g defined by its relation to the line element

ds2 = gµ⌫ dxµ dx⌫ , (2.12)

and such that det gµ⌫ 6= 0. The metric provides the notions of causality and light-cones. From the
metric g, we can also introduce the proper time (by setting all the spatial terms to zero), the proper
length, and introduce the concept of shortest distance between two points. This is sometimes called
geodesic, however beware that actually geodesics are straightest lines, that can be di↵erent from
shortest lines if the connection is di↵erent from the Christo↵el symbol of the metric.
Moreover, the metric tensor plays the role that the gravitational potential played in newtonian
dynamics; it also tells us how to compute inner products:

(u, v) = gµ⌫ uµ v ⌫ ; (2.13)

in metric theories of gravity, the metric tensor also determines the curvature through the connection
(see below).

2.2 Tensor Densities

Tensor densities, di↵erently from tensors, are objects that fail to transform as tensors, modulo the
Jacobian of the coordinate transformation taken into account
0
@xµ
J= . (2.14)
@xµ

An example is given by the Levi-Civita symbol:

8
>
<+1 even permutations of the indices
✏˜µ1 ...µn = 1 odd permutations of the indices . (2.15)
>
:
0 otherwise

15
Now, this object gives only 0, ±1, therefore it cannot transform as a tensor through an arbitrary
coordinate transformation. In general, we get
0 +1
@xµ @xµ1 @xµn
✏˜µ01 ...µ0n = ✏˜µ1 ...µn 0 . . . , (2.16)
@xµ @xµ1 @xµ0n
and the Jacobian is the object that spoils the tensor nature of the Levi-Civita symbol. The exponent
of the Jacobian is called weight.
Let us define determinant of the metric as
g ⌘ det g = |g|, (2.17)
and this, under a coordinate transformation, transforms as
0 2
0 @xµ
det g(x ) = det g(x), (2.18)
@xµ
with a weight equal to 2. Therefore, we can transform the Levi-Civita symbol into a tensor by
defining p
✏µ1 ...µn = |g|˜
✏µ1 ...µn . (2.19)

2.3 Di↵erential forms

A natural generalization of dual vectors is that of di↵erential forms. A di↵erential di↵erential p-form
is a (0, p) tensor which is completely anti-symmetric.
Scalars are in this sense 0-forms, dual vectors are 1-forms and the Levi-Civita tensor is a four-form.
The space of all the p-forms over a manifold is often denoted as ⇤p (M). Noticeable it is possible
to show that on a n-dimensional Manifold there are only p!(nn! p)! linearly independent p-forms. This
implies that for n = 4 there are at a point
• 4 one-form
• 6 two-forms
• 4 three-forms
• 1 four-form (the Levi-Civita tensor)
As we shall see the existence and uniqueness of the Levi-Civita tensor is important for an unambiguous
definition of the volume element.
Given a p-form A and a q-form B we can get a (P+q)-form by a completely anti-symmetrised
tensor product: the wedge product.
(p + q)!
(A ^ B)µ1 ...µ(p+q) = A[µ1 ...µp Bµp+1 ...µp+q ] (2.20)
p!q!
which for the case of two vectors Aµ and B⌫ , is simply
(A ^ B)µ⌫ = 2A[µ B ⌫] . (2.21)

The notation with the indices inside square brackets implies anti-symmetrization with respect to
the indices inside the brackets. In particular, one has
1 X
T[µ1 ...µn ] = ⇡ Tµ⇡(1) ...µ⇡(n) ,
n! ⇡

where the sum is taken over the permutations ⇡ of 1, . . . , n and ⇡ is +1 for even permutations

16
 and 1 for odd permutations. For example, for a two indices object, one has
1
T[µ⌫] = (Tµ⌫ T⌫µ ) .
2
Similarly for the symmetrization:
1 X
T(µ1 ...µn ) = Tµ⇡(1) ...µ⇡(n) ,
n! ⇡

and in the case of two indices, one again has

1
T(µ⌫) = (Tµ⌫ + T⌫µ ) .
2
We can also define the exterior derivative d mapping a p-form into a (p+1)-form. This is basically
equal to the wedge product of a p-form, say A, and the one-form @:

( dA)µ1 ...µ(p+1) = (p + 1)@[µ1 Aµ2 ...µp+1 ] (2.22)

and the simplest example of it is the gradient of a scalar field ( d )µ = @µ . It might seem strange
that an operation involving the simple partial derivative turns a tensor into a tensor. This is due to
the intrinsic antisymmetric structure of the exterior derivative.
Note that the Leibniz rule it is slightly modified for the exterior derivative. If ! is a p-form and ⌘
is a q-form, then
d(! ^ ⌘) = ( d!) ^ ⌘ + ( 1)p ! ^ ( d⌘) (2.23)

2.4 The Volume element

We previously noticed the important of the existence and uniqueness of a n-form in a n-dimensional
Manifold, the levi-Civita tensor. This is indeed criucial so to define volume elements (n-dimensional
infinitesimal regions which can be represented as a parallelepiped defined by an ordered set of n vectors
stemming from the same point in spacetime) in an integral like
Z
dxµ , (2.24)

Let us see how.

We want to defy an oriented volume element (if two vectors are interchanged we should get a
volume with same magnitude but opposite sign, or the volume should vanish when two of the vectors
are collinear) so the volume element should indeed be replaced by a wedge-product,

dn x = dx0 ^ · · · ^ dxn 1
, (2.25)

The problem is that the expression on the r.h.s of(2.25) is not a an n-form as it might seem.
Indeed it is a tensor density: while it is true that given two functions on the manifold say f and
g, then df ^dg is a coordinate independent two-form, this is not the case if the functions one choses
are the coordinates themselves as in our case. One can check this explicitly by making a coordinate
change after rewriting the element as
1
dx0 ^ · · · ^ dxn 1
= ✏˜µ ...µ dxµ1 ^ · · · ^ dxµn
n! 1 n
Again the fact that the volume element naively transforms as a density can be remedied by applying
|g| with the right weight p p
|g| dn x = |g| dx0 ^ · · · ^ dxn 1 , (2.26)

17
 Figure 2.2: Pictorial representation of a 3D volume element.

and indeed, by doing this, we recover a proper n-form which moreover can be easily recognised by
nothing else than the Levi-Civita tensor in coordinate basis
1
✏ = ✏µ1 ...µn dxµ1 ⌦ · · · ⌦ dxµn = ✏µ1 ...µn dxµ1 ^ · · · ^ dxµn =
n! (2.27)
1p p p
= ✏µ1 ...µn dxµ1 ^ · · · ^ dxµn = |g| dx0 . . . dxn 1 ⌘ |g| dn x.
|g|˜
n!

2.5 Curvature
We can now start to deal with curvature; a first way to introduce curvature is to seek for a general-
ization of partial derivatives @µ that: allows for a rank (k, l) tensor to become a rank (k, l + 1) tensor,
that acts linearly, and that obeys the Leibniz rule.
This actually implies the existence of an affine connection. In metric theories of gravity, the metric
enters in the definition of the connection µ⌫ :

1 ⇢
µ⌫ = g (@µ g⇢⌫ + @⌫ gµ⇢ @⇢ gµ⌫ ) ; (2.28)
2
the connection defined through the metric tensor is called the Christo↵el symbol, and this will enter
(as we will shortly see) in the computation of the Riemann tensor.
With this, it is natural to define a covariant derivative rµ so that it is coordinate invariant,
and in fact what we find is:

rµ v ⌫ = @ µ v ⌫ + ⌫
µ v rµ !⌫ = @ µ !⌫ µ⌫ ! . (2.29)

Now, rµ v ⌫ is a tensor; if we change coordinates, also r̂µ v ⌫ must be a tensor; therefore, also their
di↵erence must be a tensor, and it can be shown that
⇣ ⌘
rµ v ⌫ r̂µ v ⌫ = ˆ ⌫
µ⌫ µ⌫ v . (2.30)

Now, we can form a di↵erent connection also by permuting the two lower indices; therefore, in
general, we could have µ⌫ 6= ⌫µ . We then define the torsion as a rank (1, 2) tensor which is the
di↵erence of the two connections:

T µ⌫ ⌘ µ⌫ ⌫µ =2 [µ⌫] . (2.31)

In particular, if the connection is metric compatible1 , it inherits the symmetry of the metric, and
therefore one has that the torsion T µ⌫ = 0. Notice that, while the torsion is a tensor, the connection
is not a tensor, as it does not transform as a tensor.
1
Saying that the connection is metric compatible means that it is calculated through the metric, and therefore this
connection is the Christo↵el symbol of the metric.

18
 In general, assuming the EEP, one has that the torsion is zero, therefore in General Relativity the
torsion is automatically zero and also
r⇢ gµ⌫ = 0. (2.32)
We can define the divergence of a vector as
µ
div v ⌘ rµ v µ = @µ v µ + µ v . (2.33)

It can be proven that the following relation holds true:

µ 1 p
µ = p @ |g|, (2.34)
|g|

therefore the divergence of a vector reduces to

1 ⇣p ⌘
div v = p @µ |g|v µ . (2.35)
|g|

We can also introduce the D’Alembert operator, defined as

⇤ = rµ rµ , (2.36)

and it can be proven similarly that2

1 p
⇤ = p @µ ( gg µ⌫ @⌫ ) . (2.37)
g

In curved spacetime, one can also generalize the Stokes’ theorem:

Z Z
p p
rµ v µ g d n x = nµ v µ g dn 1 x, (2.38)
⌃ @⌃

where nµ is the vector orthogonal to the hypersurface @⌃, which is the boundary of the region ⌃.
We can also define the exterior derivative:

dA ⌘ r[µ A⌫] = @[µ A⌫] , (2.39)

and this is independent from the connection. Another quantity independent from the connection is

[x, y] = x⌫ r⌫ y µ y ⌫ r⌫ x µ . (2.40)

Now we want to apply the commutator to two covariant derivatives in order to find the Riemann
tensor. The Riemann tensor can be seen geometrically as the failure of parallel transporting a vector
in a closed curve. What one can find is that
⇣ ⌘
[rµ , r⌫ ] v ⇢ = @µ ⇢⌫ @ ⌫ ⇢µ + ⇢µ ⌫
⇢
µ v ⌘ R⇢ µ⌫ v . (2.41)

However, one can find that, in the case in which the connection is not symmetric, another term
emerges from the commutator of the two covariant derivatives:

[rµ , r⌫ ] v ⇢ = R⇢ µ⌫ v 2 [µ⌫] r v⇢; (2.42)

basically, if there is torsion, it is like the curve does not close, as if there is a defect in spacetime.
For the Riemann tensor a number of properties holds true:

1. R⇢ µ⌫ = R ⇢µ⌫ ; this holds true only if the connection is Levi–Civita one.

2
From now on, we will denote the absolute value of the determinant of the metric simply as g.

19
Figure 2.3: Pictorial representation of the failure of parallel transporting a vector in a closed curve,
giving rise to the Riemann tensor.

2. R⇢ µ⌫ = R⇢ ⌫µ ; this is always true by construction.

3. R⇢ µ⌫ = Rµ⌫⇢ ; this is true only if the connection is Levi–Civita one.
4. R⇢ µ⌫ + R⇢⌫ µ + R⇢µ⌫ = 0; this is true only if the torsion is zero.
5. Bianchi identity: r[ R⇢ ]µ⌫ = 0; this is true only if the torsion is zero, otherwise r[ R⇢ ]µ⌫ =
T[↵ ⇢ R ]↵µ⌫ [17].
Note that the Riemann tensor has n2 /12(n2 1) independent components, i.e. twenty in four dimen-
sions.
We can now define the Ricci tensor as the contraction of the Riemann tensor with respect to the
first and third indices, Rµ⌫ = R⇢µ⇢⌫ . Why exactly those indices?
Contracting the first index with the second gives zero if the connection is symmetric, Rµ⌫ 0 =
⇢
R ⇢µ⌫ = 0. However, if the connection is not metric compatible, the new Ricci tensor is not zero:
0 ↵ ↵
Rµ⌫ = @⌫ µ↵ + @µ ⌫↵ . (2.43)
Nonetheless, if the connection is not the Christo↵el symbol of the metric but it is symmetric, then we
0 = 2R
have Rµ⌫ [µ⌫] .
One can also define, at this point, the Ricci scalar as the trace of the Ricci tensor
R ⌘ g µ⌫ Rµ⌫ ; (2.44)
notice that, if we used the alternative Ricci tensor to compute the Ricci scalar, we would have found
R0 = 0, therefore R is unique even in metric-affine gravity.
At this point, the Einstein tensor can be introduced:
1
Gµ⌫ = Rµ⌫ gµ⌫ R. (2.45)
2
An important property that holds true for the Einstein tensor is rµ Gµ⌫ = 0, which is the contracted
Bianchi identity.
From the Riemann tensor, we can extract its completely traceless part, that has the name of Weyl
tensor; in n + 1 dimensions, this becomes:
2 2
C⇢ µ⌫ = R⇢ µ⌫ g⇢[µ R⌫] g [µ R⌫]⇢ + g g R, (2.46)
n 2 (n 1)(n 2) ⇢[µ ⌫]
and if the connection is metric compatible, then the following properties of the Weyl tensor hold true:
C⇢ µ⌫ = C[⇢ ][µ⌫] C⇢ µ⌫ = Cµ⌫⇢ C⇢[ µ⌫] = 0. (2.47)
One can also show that the Weyl tensor is invariant under conformal transformations, therefore doesn’t
convey informations about volume, but rather it conveys informations about perturbations, like grav-
itational waves, and about how the shape of a body changes due to tidal forces. In GR, in vacuum
solutions, the Weyl tensor is also the only non-zero part of the Riemann tensor.

20
2.6 Intrinsic and Extrinsic curvature
There are two di↵erent relevant concepts of curvature: the intrinsic curvature and the extrinsic
curvature.
• The Intrinsic curvature is an intrinsic property of the manifold, and therefore one does not need
to embed a manifold in a higher dimensional manifold in order to describe its intrinsic curvature.

• The Extrinsic curvature is a property of a submanifold embedded in a higher dimensional man-

ifold. So it does depend on the embedding as the latter induces a metric on the submanifold.

2.6.1 Example: Intrinsic-Extrinsic (Gaussian) curvature of two surfaces

For example, we can characterize the curvature of a two-sphere without embedding it in 3-dimensional
space: given the measured area of the sphere A, its radius a will be given by
r
A
a= . (2.48)
4⇡
If we take the line element on the two-sphere

ds2 = a2 d✓2 + sin2 ✓ d 2

, (2.49)

one finds that R✓ ✓ = a2 sin2 ✓, R = sin2 ✓, R✓✓ = 1, and therefore

2
R= , (2.50)
a2
which is twice the gaussian curvature of the sphere. So, note that the intrinsic curvature is R = 2KG ,
where KG = 1 2 is the gaussian curvature and 1,2 are the so called principal curvatures of the
surface which can be written as the inverses of the radii of curvature in the two principal directions on
the surface i = 1/ai . However, the extrinsic curvature turns out to be KExt = 1 + 2 [4, p. 232]. So
for example if one takes a flat sheet of paper and rolls it in one direction with radius R, it will appear
to have made a cylinder in three dimensional space. The extrinsic curvature will then be non-zero
KExt = 1/R + 0 = 1/R but the intrinsic (Gaussian) curvature will still be zero KG = 0 · 1/R = 0.

Note that if you do not change the metric, you cannot change the gaussian curvature: this is due
to the Theorema Egregium by Gauss, the gaussian curvature of a surface is unchanged when the
surface is bent but not stretched. This is also why the surface of the Earth cannot be mapped on
a plane without stretching or cutting.
This is also connected to a somewhat more common experience: eating pizza.
When on the plate, each slice of a pizza is flat, meaning that it has a zero gaussian curvature.
Picking it up, the slice curves downward, but in the perpendicular direction it still has zero
curvature. This means that, when we introduce a curvature along the direction of the crust, we
will force a zero gaussian curvature along the radius of the slice, which allows us to eat comfortably
the pizza slice!

2.6.2 Hypersurfaces and their Extrinsic curvature

All of this can be better formalized through the definition of hypersurfaces. A hypersurface is an
(n 1)-submanifold ⌃ of an n-dimensional manifold M.
We can define an hypersurface by setting a function of the coordinates f (x) to a constant value
f ⇤ on ⌃; we can then define a vector ⇣ µ ⌘ g µ⌫ r⌫ f (x). This vector is automatically orthogonal to the
hypersurface: take lµ 2 Tp (⌃), then

⇣ µ lµ = l⌫ r⌫ f (x) = 0, (2.51)

21
since on the hypersurface the function is constant; so, ⇣ µ is the normal vector to ⌃. This vector is
defined modulo some function, since given a function h(x), h(x)⇣ µ is still orthogonal to ⌃.
Now, by definition, if ⇣ µ is timelike, then ⌃ is spacelike; viceversa, if ⇣ µ is spacelike, then ⌃ is
timelike. Finally, if ⇣ µ is null, then ⌃ = N is null.
We can normalize the orthogonal vector by defining nµ as
⇣µ
nµ = " , (2.52)
|⇣ ⌫ ⇣⌫ |1/2
where " is either +1, 1 if ⌃ is spacelike or timelike respectively.

The null case is less trivial, since a null vector orthogonal to a null hypersurface is also parallel
to the hypersurface (its product with itself is zero). We can then take a parametrized family
µ
of curves xµ ( ) 2 N and write ⇣ µ = dx d i.e. as the tangent vector to these curves; it can be
shown that these curves are indeed (null) geodesics (to show this one just needs to show that
⇣ ⌫ r⌫ ⇣ µ / ⇣ µ [18]. Now, by definition, the null geodesics xµ ( ) with being an affine parameter,
for which the tangent vectors dxµ / d are normal to a null hypersurface N , are the “generators”
of N .
Now we want to define the extrinsic curvature of hypersurfaces. We can take an hypersurface ⌃ in
a manifold M; suppose ⌃ is spacelike, so that its normal vector nµ is timelike. Take the congruence of
all the timelike geodesics, so at each point nµ is the normal vector to ⌃. Associate an affine parameter
⌧ to the vector field and construct a basis y 1 , . . . y n 1 on ⌃: then we can construct an adapted set of
coordinates (⌧, y 1 , . . . , y n 1 ) such that

gµ⌫ dxµ dx⌫ = d⌧ 2 + ij dxi dxj . (2.53)

The induced metric ij has dimension n 1, and of course we can define the curvature associated to
this induced metric. However, we would like to be able to define the extrinsic curvature as a property
in n dimensions. For this reason, we construct the projector Pµ⌫ as

Pµ⌫ = gµ⌫ nµ n ⌫ , (2.54)

where = nµ nµ is the norm of the normal vector.

The projector is sometimes called the first fundamental form of the hypersurface; it has a number
of properties, for example, it projects any vector of the manifold M on the spacelike hypersurface ⌃,
as it can be checked by observing that any vector after the action of the projector is orthogonal to the
normal vector to the hypersurface

(Pµ⌫ v µ ) n⌫ = 0 8v µ 2 M; (2.55)

moreover, the projector acting on two vectors on ⌃ will behave like the metric:

Pµ⌫ v µ w⌫ = v · w 8v µ , w⌫ 2 ⌃; (2.56)

finally, the projector tensor is idempotent, P n = P , or alternatively

P µ P⌫ = P⌫µ . (2.57)

The extrinsic curvature is now defined as a rank (0, 2) tensor

1
Kµ⌫ ⌘ Ln Pµ⌫ , (2.58)
2
where the symbol Ln represents the Lie derivative along n, and it is the evolution of the object on
which the derivative is applied along n. We will discuss this Lie derivative better in the following
section.

22
 We can take the covariant derivative of an object on the hypersurface ⌃ by taking projections of
the whole object:
r̂ X µ⌫ = P ↵ P µ P ⌫ r↵ X . (2.59)
Then we can define a Riemann tensor on the hypersurface as
h i
r̂µ , r̂⌫ v ⇢ ⌘ R̂⇢ µ⌫ , (2.60)

which will be related to the Riemann tensor on the n-dimensional manifold by a suitable contraction
of the indices with appropriate projector operators
R̂⇢ µ⌫ = P ⇢↵ P P µ P ⌫ R↵ ; (2.61)
this is the so called Gauss-Codazzi equation; there is also the Codazzi equation that relates the Ricci
tensor to the extrinsic curvature:
µ 1
r̂[µ K⌫] = P ⌫ R⇢ n⇢ . (2.62)
2

2.7 Lie Derivative

The Lie derivative is a primitive connection-independent notion of derivative, and it is di↵erent from
the covariant derivative. In particular, it is the derivative with respect to the flow of a vector field.
Consider a congruence of curves xµ (s); we can define the set of vectors tangent to these curves as
µ
X(xµ (s)) = dxds . Consider an object of any rank T . At point P on the curve, we will have generally
T (P ). We can then drag T from P to the point Q along the curve and obtain T 0 (Q). We can then
define the Lie derivative of T along the curve defined by X as
T (Q) T 0 (Q)
LX T = lim , (2.63)
s!0 s
where T (Q) is just the same object T (P ), but at Q. Basically, the idea behind the Lie derivative is

Figure 2.4: Pictorial representation of the comparison of a dragged vector w.r.t. the vector taken at
the same point.

that we want to use the u vector field flow to drag objects of any form and compare with their dragged
value with their actual value at some point.
In order to understand the physical meaning of this derivative, we can restrict our definition to a
↵
curve with tangent u↵ = dx d , along which we want to calculate the Lie derivative of the vector field
A↵ .
The Lie-dragged vector A0 ↵ (x0 ) can be determined as

↵ @x0 ↵ @
A0 (x0 ) = A (x) = (x↵ + u↵ d ) A (x) = ↵
+ @ u↵ d A (x), (2.64)
@x @x

23
and therefore
↵
A0 (Q) = A↵ (P ) + @ u↵ A (P ) d ; (2.65)
conversely, by performing a Taylor expansion, we can write

A↵ (Q) = A↵ (P ) + u @ A↵ (P ) d . (2.66)

Therefore, the Lie derivative of the vector field A↵ along the curve of tangent u↵ is

Lu A↵ = @ A↵ u @ u↵ A . (2.67)

One can then generalize this to objects of any rank; for a scalar function f (x), one has

L u f = u↵ @ ↵ f ; (2.68)

for a (2, 0) rank tensor T µ⌫ one has

Lu T µ⌫ = u @ T µ⌫ T µ @ u⌫ T ⌫
@ uµ ; (2.69)

for a generic (k, l) rank tensor, we can write

X X
Lx T µ1 ,...,µk⌫1 ,...,⌫l = x @ T µ1 ,...,µk⌫1 ,...,⌫l (@⇢ xµi ) T µ1 ,...,⇢,...µ
⌫1 ,...,⌫l +
k
@⌫j x⇢ T µ1 ,...,µk⌫1 ,...,⇢,...,⌫l .
i j
(2.70)
The Lie derivative has several properties:

1. it is coordinate independent, and it maps a rank (k, l) tensor into a rank (k, l) tensor;

2. it is linear: LX (aT + bS) = aLX T + bLX S, with constant a and b;

3. it obeys the Leibniz rule: LX (T ⌦ S) = LX (T ) ⌦ S + T ⌦ LX (S).

We can prove the first property, according to which the Lie derivative is coordinate invariant. Basically,
we want to prove that (LX T )0 = LX T 0 .
Assume to have a parameter s such that we have a congruence of curves xµ (s). Now, take s as
one of the coordinates, in a way that we have an adapted system of coordinates. In this way, we have
@
x= = x↵ @ ↵ . (2.71)
@s
@ 0
Now we want to write LX T = @s T ; specifically, we get (in order to not get a too messy notation, we
use indices creatively)
µ...
(LX T )0 ⌫... (2.72)
which proves the property.
Now we can determine explicitly the extrinsic curvature
1 1 ⇠ 1
r⇠g⇠
Kµ⌫ = Lu (gµ⌫ + uµ u⌫ ) = (u ⇠ µ⌫ + rµ u g ⌫ + r⌫ u g µ ) = (rµ u⌫ + r⌫ uµ ) = rµ u⌫ , (2.73)
2 2 2
where we have used Lu u = 0 and substituted partial derivatives with covariant derivatives and we
have used the fact that u⌫ = r⌫ f , and thus rµ u⌫ is symmetric. Therefore we have Kµ⌫ = rµ u⌫ and
K = K µµ = rµ uµ . Note that if the congruence is not geodesic, then the quantity aµ = u⌫ r⌫ uµ 6= 0,
and therefore Kµ⌫ = rµ u⌫ nµ a ⌫

24
2.8 Killing Vectors
Consider now the case in which LX gµ⌫ = 0; we can easily see from the steps in (2.73) that in this case
one gets
rµ x⌫ + r⌫ xµ = 0. (2.74)
Therefore, if we Lie-drag the metric along x, and we find that the metric is invariant under this drag,
than we get that x defines a symmetry of spacetime, an isometry. In this case, the vector x is called
a Killing vector and equation (2.74) is called the Killing equation.
How many symmetries do we expect to find in a four-dimensional spacetime?
If a spacetime possesses a maximal number of isometries, the spacetime is called maximally
symmetric. An example is given by Minkowski spacetime with the Poincaré group: translations plus
boosts and rotations.
In n-dimensions, if M has the maximum number of isometries, then M is maximally symmetric.
A property of such spacetimes is that they have constant curvature, and thus constant Ricci scalar.
Consider for example the Friedmann-Robertson-Walker metric. It can be obtained using
the Copernican principle. Assuming that space is homogeneous and isotropic, we can write that
M = R ⇥ ⌃3 , where ⌃3 is a maximally symmetric 3-surface. Now there are only three constant
curvature 3-surfaces: flat, hyperbolic and spherical. So, the metric will be of the form

ds2 = dt2 + R(t) d 2

.

We can put a space-independent scale factor in front of the three-metric d 2 which, since we want
a maximally-symmetric 3-surface, must be of the form
2
d = ij dxi dxj .

We can write an element that embodies all the three possibilities for the constant curvature:

2 dr̄2
d = + r̄2 d⌦2 ,
1 r̄2
where  = 0, ±1; if  = 0, we have the 3-plane, if  = +1 we have the 3-sphere, and if  = 1 we
have the constant-curvature hyperbolic 3-space.
One can then verify that the 3-dimensional Ricci scalar R(3) = const and that Rij = 2 ij .

Sometimes, we can find a Killing vector with the blink of an eye: assume to have an adapted system
of coordinates; then, if the metric is independent on one of these coordinates, we have a Killing vector
associated to that one coordinate.
Imagine we have a coordinate ⇤ such that @ ⇤ gµ⌫ = 0; then @@ ⇤ is a Killing vector. As an example,
consider the Schwarzschild metric
✓ ◆ ✓ ◆ 1
2 2M 2 2M
ds = 1 dt + 1 dr2 + r2 d⌦2 , (2.75)
r r
@
and we immediately realize that the metric is independent both on t and . Therefore both @t and
@
@ are Killing vectors.
Of course, this trick can not be complete: at most there are 10 Killing vectors but only 4 coordi-
nates, so by studying only the coordinates from which the metric is independent does not give us all
the Killing vectors.
A useful way to write a Killing vector is as follows: for example, take the Killing vector @@ ; then,
the components of this Killing vector are k µ = (0, 0, 0, 1) = µ ; in general, if the metric is independent
on the coordinate ⇤ , we can write the components of the Killing vector relative to ⇤ as
µ
kµ = ⇤ , (2.76)

25
and it is easy to check that the Killing vector itself then becomes
µ
k = k µ @µ = ⇤ @µ = @ ⇤ . (2.77)

However, in general you cannot find a coordinate system in which all the Killing vectors are
simultaneously of the form @ ⇤ .
Imagine to be in spacetime with 1 + 1 dimension:

ds2 = dt2 + dx2 ; (2.78)

we have two evident isometries, which are the translations along t and x, and therefore two Killing
@ @
vectors @t and @x . Let us perform these transformations:
(
t= sinh ⌘
; (2.79)
x= cosh ⌘

with this, the line element can be simply rewritten as

ds2 = 2
d⌘ 2 + d 2
. (2.80)
@
Then, we can see that the metric does not depend on the coordinate ⌘, therefore we have that @⌘ is
a Killing vector. This Killing vector is simply
@ @ @
=x +t , (2.81)
@⌘ @t @x
and this Killing vector is a generic boost, so this is the Killing vector associated to the boost of flat
spacetime. What is the norm of this Killing vector? It is easy to check that

@ µ ⌫ 2
= gµ⌫ ⌘ ⌘ = g⌘⌘ = < 0, (2.82)
@⌘

and therefore this Killing vector is timelike. This is an important comment, since it is possible that
the norm of a Killing vector is spacelike, and in that case, if we want to “see” the isometry defined
by it, we would have to follow a spacelike geodesic. Also, it is possible that the norm of a Killing
vector depends on the coordinates of spacetime, and therefore the behaviour of the Killing vector
could change.

By definition, we have
(rµ r⌫ r⌫ rµ )⇠⇢ = R⇢ µ⌫ ⇠ = Rµ⌫⇢ ⇠ , (2.83)
but using the Killing equation, rµ ⇠⇢ = r⇢ ⇠µ , we can also write

(rµ r⌫ ⇠⇢ + r⌫ r⇢ ⇠µ ) = Rµ⌫⇢ ⇠ . (2.84)

We can write this same equation by permuting cyclically the indices (µ, ⌫, ⇢), and in particular,
if we add the equation with (µ, ⌫, ⇢) to the equation with (⌫, ⇢, µ), and subtract the one with
(⇢, µ, ⌫), we get [1, p. 442], 2r⌫ r⇢ ⇠µ = 2Rµ⌫⇢ ⇠ , i.e. that for any Killing vector one has

rµ r⌫ ⇠⇢ = R⌫⇢µ ⇠ =) rµ r⇢ ⇠⌫ = R⌫⇢µ ⇠ . (2.85)

which contracted alos implies

rµ r⇢ ⇠ µ = R⇢ ⇠ . (2.86)

26
2.9 Conserved quantities
Due to Noether’s Theorem, given a symmetry, there is a conserved quantity. Let us write the action
for a point-like particle: Z
1
S= gµ⌫ ẋµ ẋ⌫ d ; (2.87)
2
extremizing the action, S = 0, gives
d 1
(gµ↵ ẋµ ) @↵ gµ⌫ ẋµ ẋ⌫ = 0; (2.88)
d 2
it can be proven that this is equivalent to the usual geodesic equation with a metric compatible .
Now, assume that there exists a coordinate ↵ ˆ such that @↵ˆ gµ⌫ = 0; then, the previous equation
reduces to
d
(gµ↵ˆ ẋµ ) = 0, (2.89)
d
and therefore gµ↵ˆ ẋµ is constant along the geodesic parametrized by . The Killing vector relative to
this symmetry is k µ = ↵µˆ ; we can also define ẋµ ⌘ uµ as the four-velocity along the geodesic, the
tangent vector to the curve. Therefore
gµ↵ˆ ẋµ = gµ⌫ k µ u⌫ = (k · u) = const., (2.90)
and this is the conserved quantity, written in a covariant way.
We can now introduce a notion of conserved energy if we have a globally defined timelike Killing
vector (we have a conserved momentum if the Killing vector is spacelike): take the stress-energy tensor
of the matter,
2 S
Tµ⌫ = p ; (2.91)
g g µ⌫
we can also define a current J µ = k⌫ T µ⌫ , where k⌫ is a timelike Killing field (for example, in a
stationary spacetime). The covariant derivative of this current is
rµ J µ = (rµ k⌫ )T µ⌫ + k⌫ (rµ T µ⌫ ) = 0, (2.92)
since the first term is zero by the Killing equation, while the second term is zero due to the conservation
of the stress-energy tensor. Since rµ J µ = const, we can apply Stokes’ theorem; we define the energy
as Z
p 3
E⌘ J µ nµ d x; (2.93)
⌃
since rµ Jµ = 0, we have Z
p
rµ J µ g d4 x = 0, (2.94)
M
and by Stokes’ theorem we have
Z Z Z
µ p 3 3 p µ p
0= J nµ d x= d x J nµ d3 x J µ nµ , (2.95)
@M ⌃2 ⌃1

where we assumed that the fields go to zero sufficiently rapidly in order to neglect the spatial infinity
boundary ◆0 of @M. Since the choice of ⌃1 and ⌃2 is arbitrary, we immediately see that the energy
E is a conserved quantity of our stationary spacetime.
Consider the previous example of a point-like particle; the energy of this particle is then given
by E = p · k; now p is always timelike, but in Schwarzschild we can verify that for r > 2M the
@
Killing vector k = @t is timelike, but inside r < 2M the same Killing vector is spacelike. Therefore,
E > 0 outside the horizon, but E < 0 inside r < 2M . Therefore, in Schwarzschild, it seems that
the energy of a particle falling in the black hole changes sign, and thus for an observer at infinity
it has negative energy: it seems that the particle brings energy away from the black hole. This

27
 is connected to the ergoregion, a region of spacetime where there can be negative-energy states.
We will delve more into details in the following.

Consider a maximally symmetric spacetime in n dimensions; what is the maximum number of

isometries? Take Rn : it will have n independent translation and n rotations, but not all of these
rotations are independent. The number of independent rotations is

n(n 1)
Nrot = ; (2.96)
2
the total number of isometries, then, is

n(n 1) n(n + 1)
Ntot = n + = ; (2.97)
2 2
in n = 4 we then have at most 10 isometries; in n = 3 we have at most 6 isometries.
In 4 dimensions, there are only three maximally symmetric spacetimes: Minkowski, de Sitter and
anti-de Sitter. Also, in a maximally symmetric spacetime there is a very simple formula for the
Riemann tensor:
R
R⇢ µ⌫ = (g⇢µ g ⌫ g⇢⌫ g µ ) , (2.98)
n(n 1)
where we remember that R is constant for maximally symmetric spacetimes.

28
 3
Kinematics

3.1 Geodesic Deviation Equation

The basic concept of General Relativity is that, in presence of gravity, geodesics would feel a kind of
tidal e↵ect, such that they would start to diverge or converge. Indeed, as we saw the WEP implies
that the gravitational field in a small region of size L can be exchanged with a suitably accelerating
frame only if l ⌧ L where L is the typical scale over which the gravitational field is varying. Actually,
if is the gravitational potential, the gravitational acceleration is g ⇡ @ and L 2 ⇡ @ 2 / , so one
has to require l ⌧ L ! [l2 @ 2 ]/ ⌧ 1.
If the above condition it is not satisfied the e↵ects of curvature will be observable by looking e.g.
at the relative acceleration between two geodesics. To be concrete, let us imagine we have a flat
two dimensional plane. If two geodesics cross at a point on this plane they will then diverge with a
constant angle. However, if the same two geodesics happen to lie on a two-sphere such angle will not
be constant (they will actually cross again). So the e↵ect of curvature clearly resides in the variation
of the rate of change of the two geodesic relative position.
Indeed, we can see this simply in the context of the motion of a set of non-relativistic particle in
Newtonian gravity [4, p. 197]. Imagine the set of their trajectories is parametrised as xµ (t, n) where t
is the absolute Newtonian time and n labels the trajectory. Then the separation between two nearby
trajectories is described by a vector n ⌘ (@xµ /@n). As we said, the e↵ect of curvature is not in the
rate of change of this vector but in its non-constancy. This, given two nearby trajectories, is
✓ ◆ ✓ ◆ ✓ ◆
@ 2 nµ @ 2 @xµ @ @ 2 xµ ⌫ @ @ ⌫ @
2
= = = n = n . (3.1)
@t2 @t2 @n @n @t2 @x⌫ @xµ @x⌫ @xµ
We see then that the e↵ect depends on the second derivative of the gravitational potential rather
(while the gravitational force is proportional to the first derivative). It is precisely this quantity that
measures the e↵ect of curvature and that cannot be made vanishing by moving to a freely falling
frame.
Let us now study the same phenomenon in full relativistic gravity. Define a set of geodesics s (t)
that are parametrized by some parameter s that varies from geodesics to geodesics, while t is the
affine parameter along the geodesics themselves. The collection of these curves define a smooth 2-
dimensional surface in spacetime, since we can assume that the parameter s changes continuously.
µ
We can now define two vectors, T µ and S µ : T µ ⌘ @x @t is the tangent vector to the geodesic, while
µ
S µ ⌘ @x@s is the deviation vector that goes from a geodesic to the other.
We are interested in understanding how geodesics move with respect to the others as t varies. We
can then define a vector describing the rate of change of the distance between two nearby geodesics
V µ , as the variation of S µ with respect to the parameter t
@S µ
Vµ = = (rT S)µ = T r S µ , (3.2)
@t

29
Figure 3.1: Pictorial representation of one parameter family of geodesics s (t) with its tangent and
deviation vectors.

and a “relative acceleration” of geodesics Aµ , given by the variation of V µ in t

@V µ
Aµ = = (rT V )µ = T r V µ . (3.3)
@t
Note that the acceleration along a given curve is a di↵erent concept, and it would be aµ = T r T µ ,
which is by definition zero for an affinely parametrized geodesic. Moreover, the fact that T and S are
basis vectors adapted to our coordinate system, implies that the their commutator vanishes, [S, T ] = 0
(one vector is Lie dragged by the other); this, by the definition of the commutator, translates as
S r T µ = T r Sµ. (3.4)
Now, let us compute explicitly Aµ :
⇣ ⌘ (3.4) ⇣ ⌘
Aµ = T ⇢ r⇢ V µ = T ⇢ r⇢ T r S µ = T ⇢ r⇢ S r T µ =
⇣ ⌘
= T ⇢ r⇢ S (r T µ ) + T ⇢ S (r⇢ r T µ ) =
⇣ ⌘ (3.5)
= T ⇢ r⇢ S (r T µ ) + T ⇢ S (r r⇢ T µ ) + T ⇢ S Rµ⌫⇢ T ⌫ =
⇣ ⌘
= T ⇢ r⇢ S (r T µ ) + S r (T ⇢ r⇢ T µ ) S (r T ⇢ ) (r⇢ T µ ) + Rµ⌫⇢ T ⇢ T ⌫ S .

Up to dummy indices, the first and third term in the last line are one the opposite of the other; the
second term is zero because we are considering affinely parametrized geodesics. Therefore, we obtain
as a final result
Aµ = Rµ⌫⇢ T ⇢ T ⌫ S . (3.6)
Note that one can derive the same result more simply using index-free notation. We have defined
S = @s and T = @t , and so that the commutator is [S, T ] = rS T rT S = 0. Now, the acceleration
is given simply by
A = rT rT S = rT rS T = (rT rS rS rT ) T = Rabcd T b T c S d @a .

The Hawking e↵ect can be described heuristically as the breaking apart of virtual pairs close to
a black hole. The Riemann tensor near the horizon is (in Schwarzschild coordinates) of the order
2M
r3
⇠ M12 (rBH = 2M for a Schwarzschild black hole), therefore this is the strength of the the
tidal forces at the horizon responsible for the breaking of the virtual pairs. We will further expand
on this later on in this lectures, however this simple observation makes it clear that the Hawking

30
 e↵ect should be expected to be more intense for small black holes and negligible for large ones.
More practically, given the dependence written above, one can conclude that an extended
object near the horizon will be stretched toward the center of the black hole, with a tidal force
which will be bigger the smaller is the mass of the black hole. This e↵ect is generically known as
spaghettification. The tensile force acting on a uniform rod of length ` in the radial direction,
mass m, and distance from the BH R, oriented radially, the tensile force is
GN MBH m
FT = ` (3.7)
4R3
For black holes with masses of order 1 100M , the spaghettification is such to be lethal to humans.
For black holes, above 104 105 M the tensile force should be weak enough for allowing horizon
crossing (but of course it will depends on the size and strength e.g. of your spaceship). Exercise:
compute the minimal mass of a black hole for which a 20 meters long spaceship weighting 15⇤103 kg
and with a tensile strength of 10,000 N can cross unharmed the horizon. (Answer M ⇡ 105 M ).

3.2 Raychaudhuri Equations

We said that the geodesics deviation equation is actually the more direct way to look at gravity.
This is however a a special case of as it keeps characterises only the behaviour of a one-parameter
family of geodesics. In general, one is instead interested in keeping track of the action of gravity on a
multi-dimensional set of neighbouring geodesics, a congruence, such as those describing the trajectory
of a bundle of photons or of a cloud of massive particles. The set of equations encapsulating such
behaviour are the so called Raychaudhuri equations.

3.2.1 Time-like geodesics congruence

Consider a set of particles moving through space; now we want to describe the motion of this whole set
of particles, more than every single particle. For concreteness let us analize first the case of massive
particles moving along time-like geodesics. Generalising the previous concept, take a congruence of
multiparameter timelike geodesics si (⌧ ); for every geodesic we define a normalized tangent vector
µ
uµ = dx µ µ
d⌧ , such that u uµ = 1, in a way that the geodesics are affinely parametrized, u rµ u = 0.
⌫

Take an adapted system of coordinates (⌧, s1 , . . . , sn 1 ), such that [Si , u] = 0. We want now to
consider the variation of Siµ along the geodesic:
dS µ
⌘ u⌫ r⌫ S µ = S ⌫ r⌫ uµ ⌘ B µ⌫ S ⌫ , (3.8)
d⌧
where we have defined the tensor B µ⌫ as the measure of the failure of S µ to being parallely transported
along the geodesic. We want to evaluate B µ⌫ along the congruence.
Now, B µ⌫ ⌘ r⌫ uµ , indeed lives on the subspace orthogonal to uµ ; in fact
r⌫ (uµ uµ ) r⌫ ( 1)
uµ Bµ⌫ = uµ r⌫ uµ = = = 0. (3.9)
2 2
Conversely, u⌫ Bµ⌫ = u⌫ r⌫ uµ = 0, since we are on an affinely parametrized geodesic.
Since Bµ⌫ belongs to the subspace orthogonal to uµ , it can be generally expanded in a part
containing the trace, a traceless symmetric part, and an antisymmetric part, each of them belonging
to the same orthogonal subspace:
✓
Bµ⌫ = Pµ⌫ + µ⌫ + !µ⌫ . (3.10)
3
We again the projector operator is P µ⌫ = µ
⌫ + uµ u⌫ for a timelike normal vector. and we have
introduced the trace of Bµ⌫ ,
✓ ⌘ P µ⌫ Bµ⌫ , (3.11)

31
 Figure 3.2: Pictorial representation of a geodesic congruence.

which is called the expansion of the congruence, and it tells us how much our cloud of particle
expands or contracts isotropically. In the above equation µ⌫ is he symmetric traceless part of Bµ⌫ ,

✓
µ⌫ = B(µ⌫) Pµ⌫ , (3.12)
3
and it is called the shear, which tells us how things get elongated or squashed. Finally, we had also
to introduce the antisymmetric part of Bµ⌫ ,

!µ⌫ = B[µ⌫] (3.13)

which is called the twist, and describes rotation of the particle of the cloud as they move forward
along the congruence. Note that whenever a geodesic congruence is hypersurface orthogonal (i.e its
tangent vector can be written as the gradient of a scalar function) then it must have !µ⌫ = 0 (see
e.g. [4, p. 226]).

Figure 3.3: Pictorial representation of the congreunce evolution associated to expansion, shear and
twist.

In order to derive the Raychaudhuri equations, we start by studying the variation of the tensor

32
Bµ⌫ with respect to ⌧ :
dBµ⌫
= u⇢ r⇢ Bµ⌫ = u⇢ r⇢ r⌫ uµ = u⇢ r⌫ r⇢ uµ + u⇢ R µ⌫⇢ u =
d⌧
⇢ ⇠⇠⇠
(u⇠
= r⌫ ⇠ r⇢ uµ ) (r⌫ u⇢ ) (r⇢ uµ ) + u⇢ R µ⌫⇢ u = (3.14)
= B ⇢⌫ Bµ⇢ + R ⇢
µ⌫⇢ u u =

= B ⇢⌫ Bµ⇢ Rµ ⌫⇢ u u⇢ .

To derive the Raychaudhuri equation for the expansion, we can simply take the trace of the
previous relation. By doing this, we get
d✓
= B ⇢⌫ B ⌫⇢ R⌫ ⌫⇢ u u⇢
d⌧ ✓ 2 ◆
✓ ⇢ ⌫ ✓ ⇢ ⌫ ✓ ⇢ ⌫ ⇢ ⌫
= P ⌫P ⇢ P P ! + + ! ⇢⌫ ! ⌫⇢ R ⇢ u u⇢ = (3.15)
9 3 ⌫ ⇢ 3 ⌫ ⇢ ⌫ ⇢

✓2 µ⌫
= µ⌫ + !µ⌫ ! µ⌫ Rµ⌫ uµ u⌫ .
3
If we take now the traceless symmetric part, we get the Raychaudhuri equation for the
shear [3]
d µ⌫ 2 1 ⇣ ⌘ 1
↵ ⇢ ↵ ↵
= ✓ µ⌫ µ↵ ⌫ ! µ⇢ ! ⌫ + P µ⌫ ↵ ! ↵ ! + C↵⌫µ u↵ u + R̄µ⌫ , (3.16)
d⌧ 3 3 2
where C↵⌫µ is the Weyl tensor, here related to the deformation of space due to gravitational waves,
and R̄µ⌫ is the trace-free Ricci tensor projected onto the three-dimensional subspace [3]
1
R̄µ⌫ = Pµ↵ P⌫ R↵ Pµ⌫ P ↵ R↵ . (3.17)
3
Taking the antisymmetric part now gives us the Raychaudhuri equation for the twist [3]
d!µ⌫ 2 ↵ ↵
= ✓!µ⌫ + µ !⌫↵ ⌫ !µ↵ . (3.18)
d⌧ 3

In general, if the congruence is not made of geodesics, an extra term aµ u⌫ appears in Bµ⌫ ,
d✓ µ
while in d⌧ will have to add a term rµ aµ , where aµ = du
d⌧ . If the congruence, however, is made
of geodesics that are not affinely parametrized, generally we have u⌫ r⌫ uµ = uµ and in the
expansion equation we have to add a term ✓.

3.2.2 Null geodesics congruence

If we instead consider a null congruence, we have slight di↵erences. Our timelike uµ is now substituted
by a null k µ . As a consequence the space orthogonal to the congruence is no more three dimensional
but rather bi-dimensional (if spacetime has four dimensions). As a way to see this, consider the
four-dimensional Minkowski metric in spherical coordinates:

ds2 = dt2 + dr2 + r2 d⌦2 .

We can define two types of null coordinates, u ⌘ t r and v ⌘ t + r, such that null vectors have either
u or v constant; with these two new coordinates, the metric becomes

ds2 = du dv + r2 (u, v) d⌦2 .

If we take a constant null surface at e.g. u = constant, the du dv part of the line element goes to zero,
and one can immediately see that the orthogonal subspace is thus two-dimensional. The problem is

33
that we cannot define a metric on such subspace e.g. via the usual projector g + kk as this would not
have a zero contraction with the tangent vector to the congruence (and in any case = k µ kµ = 0).
We can however define another, auxiliary, null vector lµ , which (in some frame) points in the
opposite spatial direction to k µ and is normalized so that lµ lµ = 0 and lµ kµ = 1. We also demand
that this auxiliary vector is parallel-transported, k µ rµ l⌫ = 0 (which is compatible with the previous
conditions as the inner product is preserved by parallel-transport). [For example, in flat spacetime if
k↵ = @↵ u in the local inertial frame, one can take l↵ = (1/2)@↵ v.] The problem with this “trick”
is that this auxiliary vector is by no means unique (as pointing in the opposite spatial direction is
frame-dependent as we said). But let us proceed for the moment and see what we get.
The projector operator for a subspace which is orthogonal to both k µ and lµ is then

Qµ⌫ = gµ⌫ + kµ l⌫ + k⌫ lµ . (3.19)

Now, all the relevant information we need are encoded in the projected version of B µ⌫ :

B̂ µ⌫ = Qµ↵ Q ⌫ B ↵ , (3.20)

where we have basically projected every index; with this, decomposing in all the components, we have
1ˆ
B̂µ⌫ = ✓Q µ⌫ + ˆµ⌫ + !
ˆ µ⌫ . (3.21)
2
It should be noted that now the equations will not depend on lµ ; in order to realize this, consider the
expansion
✓ˆ = Qµ⌫ B̂µ⌫ = B̂ µµ = g µ⌫ Bµ⌫ = ✓, (3.22)
where we used that fact that the projector Q is idempotent, and that any contraction of k with B is
zero. So, we see that in fact the expansion, that in principle could have had a dependence on lµ , does
not depend on lµ .
This is promising, however if we would compute explicitly ˆ and ! ˆ , we would find that they depend
on the choice of lµ ! Nonetheless, the Raychaudhuri equation for the expansion has still the familiar
form
d✓ ✓2
= ˆµ⌫ ˆ µ⌫ + ! ˆ µ⌫ Rµ⌫ k µ k ⌫ (+✓) ,
ˆ µ⌫ ! (3.23)
d 2
and the dependence on lµ cancels in the squares of ˆ and ! ˆ (while the projector tensors drop out of
the Ricci term when one takes the trace). Hence, we can conclude that in the end our equation is
safely frame independent.

One might wonder why in the above equation the expansion squared term has a two at denomi-
nator rather than the factor 3 we found for time-like geodesics congruence. The point is that the
factor at the denominator represent the dimension of the subspace orthogonal to uµ , and when uµ
is timelike, the orthogonal subspace has dimension 3 in a four dimensional spacetime. However,
as we saw, the subspace orthogonal to a null vector has dimension 2.

The expansion that we have defined here can be seen exactly as the variation of the volume of test
particles in n + 1 dimensions [4, p. 227]:

1 V (n)
✓= ;
V (n) ⌧
in four spacetime dimensions, the expansion of a timelike congruence is then

(4) 1 V (3)
✓timelike = ;
V (3) ⌧
for a null congruence, instead we have
(4) 1 A
✓null = .
A ⌧

34
3.3 Energy Conditions
While we just saw that the Raychaudhuri equation is completely geometric in its nature, the presence
of the term proportional to the Ricci tensor clearly suggest that it can be used together with the
Einstein equations (or their generalizations) so to tell us how the geometry behaves on the base of
conditions imposed on the matter stress-energy tensor.
Indeed, Einstein gravity is a particularly simple theory where the Ricci tensor is directly related
to the matter Stress Energy Tensor (and this is not the case for more general theories of gravity, as
we shall see later on) ✓ ◆
1
Rµ⌫ = 8⇡GN Tµ⌫ T gµ⌫ .
2
Hence it natural to substitute the Ricci tensor for the SET and depending on the nature of the
congruence introduce some reasonable conditions for predicting the behaviour of matter as described
by the Raychaudhuri equations. Let us see them in detail.

1. Null Energy Condition (NEC): we define this in a null Raychaudhuri equation; what we have
is a term Rµ⌫ k µ k ⌫ . Now, through Einstein’s equation, we replace this by Tµ⌫ k µ k ⌫ (the part with
T gµ⌫ is zero, since k µ is null). The NEC then states that

Tµ⌫ k µ k ⌫ 0 8k µ null.

If we consider a perfect fluid, Tµ⌫ = (⇢ + p)uµ u⌫ + pgµ⌫ , the NEC implies ⇢ + p 0.

2. Weak Energy Condition (WEC): this is sometimes used for constraining the behaviour of a
congruence of timelike geodesics; the WEC states that

Tµ⌫ v µ v ⌫ 0 8v µ timelike.

For a perfect fluid, the WEC implies both ⇢ + p 0 and ⇢ 0.

3. Dominant Energy Condition (DEC): the DEC requires the WEC and that the object Tµ⌫ v ⌫
(which is the flux measured by an observer along the congruence) is not spacelike for any time-
like v µ . For a perfect fluid the DEC implies ⇢ 0 and p 2 [ ⇢, ⇢]. In order to give a physical
interpretation, we can say that in this case the locally measured energy density is always pos-
itive for any timelike observer, and the energy flux si timelike or null; therefore, there are no
superluminal fluxes.

4. Strong Energy Condition (SEC): this is the natural one to use for a timelike congruence as
derives directly from using the above written simple relation the Ricci-SET in GR. It requires
that ✓ ◆
1
Tµ⌫ T gµ⌫ v µ v ⌫ 0 8v µ timelike.
2
For a perfect fluid the SEC takes the form: ⇢ + p 0 and ⇢ + 3p 0. The SEC implies the NEC
but not the WEC, and the DEC does not imply the SEC.

As a practical application, consider the copernican principle which, as we have already seen,
translates into the FRW metric. Take a timelike congruence. The tangent to this congruence, uµ ,
can be written as the gradient of a scalar, rµ (the cosmic time). Therefore, we can set the twist
to zero [4, p. 226], while by isotropy we can set the shear to zero.
Consider now a small spherical volume, V = 43 ⇡R3 , where R = R(t) = a(t)R0 . The expansion
is
1 V 1 4 ȧ
✓ = lim = 4 3 ⇡3R2 Ṙ = 3 = 3H.
V !0 V ⌧ 3 a
3 ⇡R

35
 Then, the Raychaudhuri equation for the expansion becomes

✓2
✓˙ = Rµ⌫ uµ u⌫ ;
3
Without loss of generality, one can choose uµ in a way that it is purely timelike; then the Ray-
chaudhuri equation becomes
ä
3Ḣ = 3H 2 R00 =) 3 = R00 .
a
Until now, we have not used Einstein’s equations (and per se the Raychaudhuri equations are
valid in any theory of gravity, being purely geometrical in nature) but we can do that now in
order to determine R00 . So in General relativity
✓ ◆
1
R00 = 8⇡GN T00 T g00 ,
2

which when used in the Raychaudhuri ends up provinding us with nothing else than the well
known second Friedmann equation
ä 4⇡GN
= (⇢ + 3p) ,
a 3
In particular, if for the cosmological fluid ⇢ + 3p > 0, then ä/a < 0, and the expansion of the
Universe decelerates; however, for ⇢ + 3p < 0, i.e. if the SEC is violated, then ä/a > 0, and the
expansion of the Universe accelerates. Current observations indeed seem to imply that if GR holds
at cosmological scales then the current energy-matter composition of the Universe is dominated
by a SEC-violating component. This is what we call dark energy.

36
 4
Variational Principle

4.1 Lagrangian Formulation of General Relativity

In the previous sections, we have seen a series of topics concerning kinematics, like the Raychaudhuri
equations; now we want to start studying the dynamics connecting gravity to the matter distribution
through the Einstein’s field equations:
Gµ⌫ = 8⇡GTµ⌫ . (4.1)
We will provide a formal derivation of the above equations using the variational principle: starting
from an action, we will vary it with respect to our dynamical field, arriving to the equations of motion
for that field. In our case the field is the metric, and we want to find a suitable Lagrangian density in
order to have second order equations of motion for it.
First of all we want to find a theory that satisfies the strong equivalence principle (universality
of free fall even for self gravitating bodies, local Lorentz invariance and local position invariance
for gravitational experiments). Usually, the Lagrangians we write depend on the fields and their
first derivatives, but, since for the strong equivalence principle we can always find a reference frame in
which the first derivatives of the metric vanish, we have to go to the next order and write a Lagrangian
depending also from the second derivatives of the metric (quite reasonably so, given that the latter
encodes the curvature of spacetime):

Lgrav = Lgrav (g, @g, @ 2 g). (4.2)

However, notice that the second derivatives dependence should be at most linear, otherwise by inte-
grating by parts we would end up with higher than the second order equations of motion.
Any non trivial tensor that can be made from the metric and its first and second derivatives is
a combination of the metric itself and the Riemann tensor (which takes into account all the possible
curvatures), and from these two we can build up only one possible scalar that is the Ricci scalar. We
saw in Chapter 2 that this is a good choice also if we are in non-metric theories, given that the Ricci
scalar is unique even if the are not the Christo↵el symbols.
Thus the action for gravity can be written as:
Z
1 p
SEH = d4 x gR (4.3)

where  = 16⇡G is the appropriate factor (which has to be constant for SEP to hold) fixed by the
requirement to recover standard non-relativistic gravity in the Newtonian limit. In the following, we
will sometime omit this for conciseness.
The action we have just written is called the Einstein–Hilbert action. We can also write a total
action
Stot = SEH + Sm , (4.4)

37
where Sm is the action for the matter fields
Z
p
Sm = d4 x gL i
, rµ i
. (4.5)

Notice that we are not allowing non-minimal gravity-matter couplings of the form f ( )R, since these
are equivalent to have a dynamical gravitational constant G(x) depending on the spacetime position,
as this would violate the requirement of local position invariance of all the interaction dictated by the
SEP. Similarly, we do not consider more general derivative couplings that would lead to even more
general classes of scalar tensor theories (more on this later).
The fact that the gravitational action contains up to second order derivatives of the metric, implies
that we cannot use straightforwardly the usual Euler-Lagrange equations. We can however still resort
to a minimal action principle and impose Stot = 0. 1 By performing the variation of the action with
respect to gµ⌫ and imposing it to be 0, we shall get the gravitational equations of motion.
Actually, it is more convenient to perform the variation with respect to the inverse metric g µ⌫
given that R = g µ⌫ Rµ⌫ . It can be shown that the following relation holds 2

gµ⌫ = gµ⇢ g⌫ g⇢ . (4.6)

Therefore, apart from some metric factors, the variation with respect to the metric or the inverse
metric leads to the same result.
Now, if we vary the Einstein–Hilbert action we get (for concreteness in what follows explicitly
adopt a ( , +, +, +) signature)
Z Z
1 4 p µ⌫ 1 ⇥p p p ⇤
SEH = d x gg Rµ⌫ = d4 x gg µ⌫ Rµ⌫ + gRµ⌫ g µ⌫ + R g (4.7)
k k
Since the functional derivative of a functional S with respect to a generic field is defined as:
Z
p S
S= d4 x g i i (4.8)

we would like to arrive to a form in which the term g µ⌫ is factorized. The second term in Equation
(4.7) is already in the right form, however for the other two we need some manipulations. Let us start
from the third one. The following property holds for a square matrix:
1 1
ln (det(A)) = tr( ln (A)) =) (det(A)) = tr(A A), (4.9)
det(A)

so that
1 p 1p
g = g µ⌫ gµ⌫ =) g= g(g⇢ g ⇢ ) =) g= ggµ⌫ g µ⌫ , (4.10)
g 2
where in the second step we have used Equation (4.6).
Finally, the first term in (4.7) can be shown (see the details below) to be a boundary term and
therefore one would be tempted to say that this term does not give any contribution to the variation
of the action. We will see in the following that this is not exactly the case, but for the moment we
can decide to neglect this term and postpone its discussion for a while.
1
For a Lagrangian function of i fields and their derivatives one can of course use generalised EL-equations in the form
EL EL
@ @ @2
i
L( i , @ i
, @2 i
,...) = 0 where i
= @µ + @µ @⌫ ...
@ i @(@µ i) @(@µ @⌫ i)

but this is not necessarily more efficient than doing the direct variation of the action.
2
This can be check starting from the identity g µ g ⌫ = ⌫µ , the variation then implies g µ g ⌫ + gµ g ⌫ = 0 which
of course can be rewritten has g 0 ⌫ = gµ 0 g ⌫ g µ .

38
 This said, using only the second and third term in Equation (4.7) and using the relation (4.10),
we immediately get the Einstein’s equations. To see this clearly we remind the definition of the stress
energy tensor:
2 Sm
Tµ⌫ ⌘ p . (4.11)
g g µ⌫
So we have
S SEH Sm
µ⌫
= µ⌫
+ µ⌫ = 0 (4.12)
g g g
where p ✓ ◆
SEH g 1
= Rµ⌫ gµ⌫ R (4.13)
g µ⌫ 16⇡G 2
and
Sm 1p
µ⌫
= gTµ⌫ (4.14)
g 2
Equation (4.12) expresses the Einstein’s equations.

Let us now try to motivate the choice of that definition of the stress-energy tensor.
Consider a scalar field with an action
Z 
4 p 1 µ⌫
S = d x g g (rµ )(r⌫ ) V ( ) ; (4.15)
2

if we vary it with respect to the inverse metric and apply the definition we used above, Equation
(4.11), we get

2 S 1
Tµ⌫ = p = rµ r⌫ gµ⌫ g ⇢ r⇢ r gµ⌫ V ( ), (4.16)
g g µ⌫ 2

and in flat spacetime this reduces exactly to the expected stress energy tensor for a scalar field.
We can also notice that the quantity defined in Equation (4.11) is a (0, 2) rank tensor, with
the dimension of an energy density, symmetric and covariantly conserved, since we can prove that
it satisfies the relation
rµ Tµ⌫ = 0,
where the covariant derivative is referred to the metric appearing in equation (4.11). This con-
servation would be related to the Bianchi identity of the Einstein tensor.

Now we can go back to the discussion of the first term in (4.7). In order to be more accurate, we
have to do the variation of the Ricci tensor. It can be proven that (see e.g. [2])
↵ ↵
Rµ⌫ = r↵ ( µ⌫ ) r⌫ ( ↵µ ). (4.17)
This equation provides the generic variation of the Ricci tensor and it is always true, independently
on the theory of gravity considered. Since we are in a metric theory, the affine connection is provided
by the Christo↵el symbols, so that
↵ 1h ↵
i
µ⌫ = g r
µ ⌫ ( g ) + g r
⌫ µ ( g ) g g
µ↵ ⌫ r ( g ) (4.18)
2
This can be seen as the di↵erence of two connections and hence it is a tensor and one can apply to it
the covariant derivative in the standard way:
↵ ↵ ↵ ↵ ↵
r⇢ ( µ⌫ ) = @⇢ ( µ⌫ ) + ⇢ µ⌫ ⇢µ ⌫ ⇢⌫ µ . (4.19)
So, putting everything in (4.17) gives in the end
Z Z Z
p p p
d4 x gg µ⌫ Rµ⌫ = d4 x grµ r⌫ ( g µ⌫ + g µ⌫ g↵ g↵ ) = d4 x grµ v µ , (4.20)
M M M

39
where the integral is performed over the whole manifold M and in the last step we have defined

v µ ⌘ r⌫ ( g µ⌫ + g µ⌫ g↵ g ↵ ). (4.21)

Since we have the divergence of a vector, we can use Stokes’ theorem to rewrite the integral as
Z I
4 p µ
p
d x g rµ v = d3 x h v µ nµ (4.22)
M @M

where the integral is performed over the boundary of the manifold @M, we have introduced the metric
h induced on such a boundary and again = n⇢ n⇢ . Usually, when we do variations of a field, we keep
it fixed at the boundary and thus integrals of this type vanish. However, in this case we do not have
only terms with the variation of the metric appears, but as well terms involving the variation of the
derivative of the metric, which can be generically non-zero. At this point, we can choose two di↵erent
approaches to solve this problem: we can either introduce a counter term in the action that cancels
this first term when we do the variation, or we recognise that the problematic terms are introduced
by the assumption that the connections are metric compatible. Let us consider below both paths.

4.1.1 Gibbons-Hawking-York counterterm

Let us analyse the product:

nµ v µ = nµ ( r↵ g µ↵ + g↵ rµ g ↵ ) = nµ g↵ (rµ g ↵ r g µ↵ ) = nµ P↵ (rµ g ↵ r g µ↵ ) (4.23)

where in the last passage we have substituted the projection operator P↵ = g↵ n↵ n in place of
the metric, since the second term of the projection operator applied to the parentheses vanishes.
Now, we can immediately see that the term P↵ r g µ↵ vanishes. Indeed, this terms corresponds
to take the covariant derivative projected on the hypersurface; this gives 0 because the field is held
fixed at the boundary in a variational principle and hence g vanishes on @M. So the integral in
(4.20) becomes Z
p
d3 x h (nµ P↵ rµ g ↵ ). (4.24)
@M
We can now prove that the term in parentheses is the variation of the trace of the extrinsic curvature
of the boundary. The trace of the extrinsic curvature is given by

K = r ↵ n↵ = P ↵ r ↵ n , (4.25)

where in the last step we introduced the projection operator because the directional derivative of n
along n is zero. If we now vary this quantity, we get (note that P↵ = 0 on the boundary)

1 1 1
K = n P↵ g [r↵ ( g ) + r ( g↵ ) r ( g↵ )] = n P ↵ g r ( g↵ ) = n P ↵ r ( g↵ );
2 2 2
(4.26)
where we used twice that the projected derivative applied on the variation of the metric gives zero.
So this shows that the term in equation (4.24) can be written as
Z
p
2 d3 x h K. (4.27)
@M

It is clear that in order to recover the standard Einstein equations, we need to add to the original
gravitational action a counter term, the so called Gibbons-Hawking-York term (GHY) and redefine
the gravitational action as
Z Z
1 4 p 2 p
Sgrav = SEH + SGHY = d x gR d3 x h K. (4.28)
16⇡G M 16⇡G @M

40
Indeed, if we perform the variation of this action, the additional term perfectly cancels the term
containing the variation of the Ricci tensor. p Note that, in principle, when we do the variation, we
should have also the variation with respect to h; still, since the full metric is fixed at the boundary
and its variation is zero, also the projected metric has null variation. In this way we can obtain again
the Einstein’s equations.
Thanks to this counternterm, we obtain the correct result for what concerns Einstein’s equations;
however, if we are interested in the numerical value of the action (e.g. in a path integral approach),
we still have to face some issues. As an example, consider a Minkowski spacetime with a boundary of
cylindrical shape (Figure 4.1). The Einstein–Hilbert part of the action is zero, since the spacetime is
flat. The boundary is made of two circular surfaces, respectively at t1 = const and t2 = const, where
the extrinsic curvature is zero, and a cylindrical surface at constant radius R with an induced metric
hij = dt2 + R2 d⌦2 . The vector orthogonal to this surface is n↵ = @↵ r and the p trace of the extrinsic
curvature can be computed to be K = r↵ n↵ = R2 . It is also easy to see that |h| = R2 |sin (✓)| and
so the GHY term turns out to be proportional to SGHY / R(t2 t1 ), which diverges as R ! 1. Thus
the action diverges.

Figure 4.1: Minkowski spacetime with a boundary cylindrical shape.

The way out is to regularize this in some way. First of all we remember that, even if the example
above is in flat spacetime, the action for gravity is written in a completely general way and it is valid
also when there is matter and the spacetime is curved. Now let us proceed to regularize the GHY
term as follows: Z
2 p
SGHY = d3 x h(K K0 ). (4.29)
16⇡G @M
Let us try to explain the meaning of this regularization. K0 is the extrinsic curvature of the boundary
@M computed in flat spacetime. Even if this term diverges, neither K nor K0 represent actual physical
quantities, the real physical quantity is the di↵erence between the surface term in our spacetime with
respect to the case in which there is no gravity; such a di↵erence is finite. In the case of Minkowski
spacetime we analysed before, K = K0 and the GHY term goes to zero; in the case of curved spacetime,
however, the di↵erence will be di↵erent from zero, but it will still be finite. This case is similar to the
Casimir e↵ect, where we do a subtraction of the zero point energy to obtain a finite e↵ect.
To be fair, the definition of K0 we gave is not precise: it is the extrinsic curvature of the boundary
in the maximally symmetric limit of the solution we are considering. For example, if we considered
the Schwarzschild solution, the K0 term of course would be the extrinsic curvature of the boundary
in flat spacetime, but if we considered a Schwarzschild-de Sitter solution, it would be the extrinsic
curvature computed in de Sitter.

41
 We conclude by stressing that, since K0 is a constant, its variation with respect to the metric gives
zero, and therefore it does not a↵ect the equation of motion for the metric. Hence, we recover the
Einstein’s equations.

4.1.2 Schrödinger action

Let us now try to understand the reason why the boundary term appears. Previously, we wrote the
Lagrangian for gravity as: Lgrav (g, @g, @ 2 g). The origin of the problem is in the second derivatives of
the metric, since they give rise to terms containing the variation of the derivative of the metric once
we integrate by parts. Indeed, one can write
p p
gR = g Qabcd Rabcd (4.30)

where Qabcd = 12 ( a c g bd d bc
a g ). Writing the Riemann tensor explicitly we get [4, p. 242]
p p p p
gR = 2 gQabcd adk kbc + 2@c [ gQabcd abd ] ⌘ gLquadratic + Lsurface . (4.31)

The Lquadratic part contains terms like gg or (@g)2 , and from there we can recover Einstein’s equations
by performing the metric variation, this is sometimes called the Schrödinger action; the surface part
p
Lsurface = @c [ gV c ], instead, has terms containing second derivatives of the metric and it is related
to the boundary term we have seen before. One could think that this term is exactly equal to the
GHY term, but this is true only if the chosen boundary @M has some specific coordinate ⇤ that
⇤ ⇤
remains constant on it. Therefore, on this boundary one has to have x = const and g µ = 0.
p
Remarkably, it can be noticed [4, p. 655] that Lbulk ⌘ gLquadratic and Lsurface are not inde-
pendent in GR but rather related in a simple way as (in D > 2 dimensions)
✓ ◆
1 @Lbulk
Lsurface = @i gab . (4.32)
[(D/2) 1] @(@i gab )
This “holographic” relation is typical of GR and of a class of theories in higher dimension still endowed
with secondo order field equations (the so called Lanczos-Lovelock theories) which we shall meet in
the next chapter.

4.1.3 Palatini variation

Palatini variation is another way to avoid the problems given by the boundary term in the gravitational
action and to derive the Einstein’s equations. The core of the Palatini variation is to not assume a
priori that the connection is provided by the Christo↵el symbols. In fact, problems started to arise
with the boundary term after equation (4.17), when we assumed that the connection is provided by
the Christo↵el symbols. The only assumptions we do in this case are:
⇢
1. the connection is symmetric (no torsion), [µ⌫] = 0;

2. matter couples to the metric and not to the connection, Sm = Sm (g, i , r i ).

The Ricci tensor will no longer depend on the metric but only on the connection: Rµ⌫ = Rµ⌫ ( ).
Thus, when we do the total variation of the gravitational action we get:
Sgrav µ⌫ Sgrav
Sgrav = g + . (4.33)
g µ⌫
From the first term, together with the variation of the matter action, we can directly rederive the
Einstein’s equations. The second term will lead us to the equation of motion for the connection. In
particular, we have
Z Z 
S 4
p µ⌫ Rµ⌫ 4
p µ⌫
= d x |g|g = d x |g|g r ( µ⌫ ) r⌫ ( µ) , (4.34)

42
where we have used equation (4.17). The symbol r indicates that now the covariant derivatives
are made from the connection, which are no more functions p of the metric (we are not in a metric
compatible theory). Thus we cannot safely bring the term |g|g µ⌫ inside the covariant derivative; we
have to use the Leibniz’s rule to do it. Specifically, we get
Z 
S p p p p
= d x r ( |g|g µ⌫
4
µ⌫ ) r ( |g|g )
µ⌫
µ⌫ r⌫ ( |g|g
µ⌫ µ⌫
µ )+ r⌫ ( |g|g ) µ .

(4.35)
The first and third terms inside the square brackets are the divergence of a vector therefore, for Stokes’
theorem, we can take the integral on the boundary of those terms, multiplied by a directional vector.
Fortunately, in this case is an independent field, and it can be kept fixed at the boundary; hence
the two terms vanish. What remains is
Z 
S p p
= d4 x µ⌫
r⌫ ( |g|g ) r ( |g|g )
µ
µ . (4.36)

Since the connection is symmetric, also the µ are symmetric in the exchange of µ and ; therefore,
it is sufficient to set to zero the symmetric part of the expression in the square brackets (in µ and )
in order to extremize the action. Through this requirement we can recover the equations of motion
p
for the connections: r |g| = 0 and r g µ⌫ = 0, so the connection is metric compatible.
(Exercise: show this. Hint: consider first the trace and then the symmetric traceless part of the
above term.
This is an elegant way to recover Einstein’s equations: a priori we do not consider a metric
compatible theory, and we vary with respect to the metric and the connections as independent fields;
the fact that the connection is exactly provided by the Christo↵el symbols of the metric is given by
the dynamics. This is not general; in 4 dimensions it happens only for GR. The fact that the Palatini
variation gives the same answer as the metric variation is a property only of theories with two degrees
of freedom: those associated a massless spin 2 particle, the graviton. If we have a theory with extra
degrees of freedom, this can always be made explicit by adoption a representation of the latter with
extra dynamical fields which we shall then need to set them constant at the boundary.
Before we move one, let us comment on the step that led us to (4.36): in equation p (4.35) we set to
zero the first and third terms applying the Stokes theorem even if we had no more |g| in the volume
element. However, in thosep terms we are performing the covariant derivative of a tensor density, and
not of a tensor, since |g| is inside the derivative, and this turns out to be equivalent to apply a
simple partial derivative. Indeed, when we perform a covariant derivative of a tensor density we have
something like:
r⇢ (A...... ) = @⇢ (A...... ) + (standard tensor terms) A...... ⇢, (4.37)
where we have left out the indices of A and the standard terms are the usual Christo↵el terms (see
e.g. [19, p. 69] for an example on the metric). Therefore, in our case we have
⇢
r⇢ V ⇢ = @ ⇢ V ⇢ + ⇢ V ⇢V
⇢
= @⇢ V ⇢ , (4.38)

the covariant derivative becomes the partial derivative, the Stokes theorem in curved space becomes
the usual one, and we can apply it in a integral with just d4 x without any problem.

43
 5
Alternative Theories of Gravity

In this Chapter, we will briefly review some of the most relevant alternative theories of gravity. While
their investigation started rather early (e.g. Brans–Dicke theory is dating back to 1961), it is undoubted
that the renewed in extension/alternatives to GR was strongly propelled by the need to explain in
recent times cosmological observations, and in particular Dark Energy. Bear in mind that this Chapter
is not exhaustive nor excessively detailed, its purpose is just to give the reader an overview, a glimpse
of how it is possible to formulate gravity theories di↵erent from General Relativity.

5.1 Holding SEP: GR in D 6= 4

If one wanted to create alternative theories of gravity, the first idea that could come to mind is to
change the spacetime dimension D = d + 1 (where d represents the space dimension). Let us give
a quick excursus on the main ideas and issues of such theories. We can divide them in two classes:
D < 4 and D > 4.

5.1.1 D<4
In the case of D = 2 we have that the Einstein tensor identically vanishes: Gµ⌫ = 0 in GR. The theory
trivialises: the Riemann tensor has only one independent component (remember that such number is
D2 (D2 1)/12 in D dimensions) which is indeed the Ricci scalar. Furthermore the integral of the
Ricci scalar is nothing else than a topological invariant (the Euler characteristic of the manifold, more
later). In such theories it is usually introduced a dilaton scalar field and the Lagrangian is written as:
L = R + X( ) + U ( ), where X( ) is the kinetic term of the scalar field and U ( ) the potential.
Also one can still derive a non-trivial 2D gravity action by the so called “dimensional reduction”:
e.g. in spherical symmetry one could integrate out the angular dependence so to get an e↵ective 2D
Lagrangian. R 3 p
1
More interesting is the case of D = 3. In fact, we can write the action: Sgrav = 16⇡G d x gR
1
and we arrive to the Einstein equations: Rij g
2 ij R = 8⇡GT ij . However in three dimensions the
Weyl tensor vanishes, so the curvature is completely encoded in the Ricci tensor and you cannot
have gravitational waves. We cannot have structure formation either, unless we introduce a negative
cosmological constant, ⇤ = l12 .

The BTZ black hole

A notable example of solution of the Einstein equations in 3 spacetime dimensions is the one of a
rotating black hole embedded in an anti-de Sitter background, the so called BTZ (Bañados–Teitelboim–

44
Zanelli) black hole
dr2
ds2 = F (r)dt2 + + r2 (d' ⌦dt)2 , (5.1)
F (r)
2
where ⌦(r) = 2rJ2 is the angular velocity and F (r) = M + rl2 + 4rJ2 (with M and J being the mass
and the angular momentum of the black hole with |J|  M ). In this case we have two horizons,
" ✓ ◆2 !#1/2
M l2 J
r± = 1± 1 1 . (5.2)
2 Ml

Since the BTZ solution is a vacuum solution with negative cosmological constant it is not surprising
that is locally, in the limit J ! 0, like an AdS one with the extra constraint = + 2⇡. I.e. the BTZ
BH is locally AdS but with a di↵erent global topology.

5.1.2 D>4
Let us move to the more interesting case, D > 4. We can have two kind of theories, compactified
extra dimensions and large extra dimensions.
In the first case, we imagine that there are compactified extra dimensions with a typical radius
R. Since the energy scales as E / R1 , having a compactified dimension means that, if the radius of
the extra dimension is small enough, we will not be able to probe it, since we would be probing it at
too low energies, and we would never realize that our spacetime has e.g. five dimension and topology
R ⇥ ⌃3 ⇥ S 1 However we are doing a theory of gravity, and this means that this structure cannot be
a background structure, but rather should have dynamics. So we have to promote the radius of this
extra dimension to a field R(x), and its dynamics is hard to study and implies to deal with technical
issues.
An alternative point of view stems from the following considerations. In D = 4 spacetime dimen-
sions we can write the Newtonian potential as
GN M M 1
V (4) = = (4)2 , (5.3)
r MP r
(4)
where in the last step MP is the Planck mass in 4 dimensions and we have written GN as the square
of the inverse of the Planck mass. We can try to generalize it in D dimensions. From a simple power-
counting of the usual Einstein–Hilbert action, we can
h find i that the dimensionality of the Newton
(D) (D)
constant in D spacetime dimensions in mass unit is GN = 2 D such that GN / 1/MPD 2 ; also,
the Newton potential in D dimension scales as r3 D, therefore,
(D)
GN M M 1 M 1 1
V (D) = ⇠⇣ ⌘D 2
=⇣ ⌘D 2
, (5.4)
rD 3 (D) rD 3 (D) RD 4 r
MP MP

where in the last step we introduced a characteristic scale of the extra dimensions R, and we rewrote
the potential in order to see explicitly the 1/r dependence experienced in the D = 4 case. Basically
we are trying to interpret the Newton potential (Equation 5.3) that we see in four dimensions as
described in Equation 5.4. So, equating the two potentials, we get:
⇣ ⌘
(4) 2
⇣ ⌘2 MP
(D)
MP =⇣ ⌘D 4 . (5.5)
(D)
MP R

It is a well known problem in theoretical physics that the coupling associated to gravity is very
di↵erent with respect to the coupling of the other Standard Model interactions (hierarchy problem).

45
However, the above reasoning shows that with increasing number of dimensions D, with a character-
istic length which is large enough, we can reduce the value of the Planck mass, in order to make it
comparable to the couplings of other forces. One can than conjecture that our Universe is actually
a 4-dimensional brane embedded in a higher-dimensional manifold, where the extra dimensions are
accessible only to gravity. Then, the standard model forces are stuck on the brane, while gravity is
the only interaction that can “communicate” with the other dimensions and has a characteristic scale
(
MP D) similar to those of the standard model.
However, all these theories are partially ruled out by experiments at LHC that show no signals
predicted by them (micro black holes from TeV collisions). Also the observation of GW170817 with
(GW) (EM)
its electromagnetic counterpart, tells us that dL ' dL pointing towards D = 4 [20].

5.2 Holding SEP: Lanczos-Lovelock theories

We saw that the SEP can be seen as a combination of the GWEP, LLI and LPI where the latter is
applied also to gravitational experiments. It can be shown [16] that GWEP substantially requires that
in vacuum a theory does not admit extra dynamical degrees of freed apart from the graviton. In four
dimension GR is the only theory (with spin-2 gravitons) satisfying the SEP. 1
Nordström Gravity: We specified that GR is the only theory with spin-2 gravitons satisfying
SEP in four dimensions because indeed one can have a scalar theory of gravity which satisfies
the same principle. In 1913 such a theory was proposed by Gunnar Nordström as a way to have
a relativistic generalisation of the Poisson equation. After several iterations with Einstein and
the crucial contribution of a brilliant graduate student of Hendrik Lorentz, Adriaan Fokker, the
theory settled in its modern form R = 24⇡GN Tmatter (the so called Einstein–Fokker equation)
and Cabcd = 0 where R is the Ricci scalar, Tmatter is the trace of the SET and Cabcd is the Weyl
tensor. The theory is a scalar theory of gravitation because the condition on the Weyl implies
that all the solutions are conformally flat metrics gµ⌫ = 2 (x)⌘µ⌫ . The theory was superseded by
GR which among other things crucially predicts correct factor of light deflection by gravitating
bodies (angle which is trivially zero in Nortström gravity due to the constraint of conformally flat
solutions). See e.g. [25] for more details.
Lanczos-Lovelock theories are alternative gravity theories that have an equation of motion of the
second order in the metric. In order to have this feature, we expect the Lagrangian to be at most
linear in the second derivatives of the metric, which is the same feature we had in the Lagrangian
of the previous Chapter. Hence the Lagrangian should contain the curvature tensor at most linearly.
However, in such theories we can construct Lagrangians with even higher powers of the curvature
tensor, but in a special combination that allows us to eliminate all higher order terms in the field
equations. Let us see how to generalize the Lagrangian to include terms with higher power of the
curvature. We saw that we can rewrite the Einstein–Hilbert Lagrangian as
LEH = R = Q↵ R↵ , (5.6)
where we have defined
1
⌘ ( ↵g
Q↵ ↵ g ). (5.7)
2
This tensor is constructed only from the metric, has the same symmetries of the Riemann tensor
and it satisfies the property rµ Qµ = 0. It can also be proven that this is the only tensor constructed
from the metric alone that has such properties.
1 p
One can argue that also unimodular gravity [21] — which is like GR with the additional requirement g ⌘1–
satisfy the SEP. However, this theory can be rather seen as a generalisation of general relativity, in which the cosmological
constant appears as a single additional variable (not a field), but is then seen to be a constant of the motion. Indeed,
if one tries to reconstruct GR as a nonlinear field theory of gravitons over Minkowski spacetime (see e.g. [22] but also
[23]), one can get both GR and Unimodular gravity making di↵erent assumptions of the way di↵eomorphism invariance
is implemented, see e.g. [24].

46
 Actually let us write the above expression as

LEH = ↵ R↵ , (5.8)

where the object

1
↵⌘ ( ↵ ↵ ) (5.9)
2
is known as the alternating determinant, or alternating tensor. In general, we have

ab...j 1 m q n
mq...n ⌘ [a b ... j] , (5.10)
n!
that is the determinant of a matrix of deltas, constructed as follows,
0 m n
1
a ... a
det @. . . . . . . . .A . (5.11)
m ... n
j j

This procedure to write the Einstein–Hilbert lagrangian by the contraction of the Riemann tensor with
the divergence-free tensor Q leads to second order field equations [4]. To go beyond GR we would
like to maintain the same construction, but generalizing Q. We will allow it to be a linear function of
the curvature tensor. Hence, now Q will have the same symmetries of the Riemann tensor, it will be
divergence-free, and it will be a function of g ↵ and R↵ . Q can be built as:

Q↵ = R↵ G↵ g + G g ↵ + R↵ g R g↵ ; (5.12)

Such a form has again the right symmetries. This leads to a new Lagrangian, given by the contraction
between Q and the Riemann tensor, called the Gauss–Bonnet Lagrangian:
1
LGB = QR = [R↵ R↵ 4R↵ R↵ + R2 ] (5.13)
2
which will cancel on the terms of order higher than the second in the equations of motion. In 4
dimensions, the variation of this quantity is a pure divergence term and hence does not contribute
to the equations of motion. Furthermore, its integral on the four volume is equal to the Euler
characteristic of the manifold, which is a number characterising the topology of the considered
manifold: a topological invariant. However, in a higher number of dimensions, this is not equal to the
Euler characteristics anymore, and therefore it becomes relevant.
As it was done with the Einstein–Hilbert Lagrangian, also this term can be rewritten in a more
compact form:
LEH = 21 43 R12 34 , (5.14a)
1357 24 68
LGB = 2 4 6 8 R1 3 R5 7 . (5.14b)
Now we can easily extend it to higher orders in the Lagrangian; for example, at third order we
have
L (3) = 21 43 65 87 10
9 11 2 4 6 8 10 12
12 R1 3 R5 7 R9 11 , (5.15)
and in general at the m-th order we have
1 3 5 7...2k 1 2 4 6 8 2k 2 2k
L (m) = 2 4 6 8...2k R1 3 R5 7 . . . R2k 3 2k 1 , (5.16)

where k = 2m.
How we see explicitly that these Lagrangian are all leading to second order field equations? First of
all note that the splitting we showed for the EH action in a bulk action (quadratic in the connections)
and a surface part did not make explicit use of the form of Qabcd , just of its properties.
p p p p
gQabcd Rbcd
a
=2 gQabcd adk kbc + 2@c [ gQabcd abd ] ⌘ gLquadratic + Lsurface . (5.17)

47
 Alternative one can see that each L (m) is by construction a homogeneous function of the Riemann
tensor of degree m. !
(m) 1 @L(m) 1
L = Ra bcd ⌘ Pa bcd Ra bcd (5.18)
m @Ra bcd m
where evidently P abcd = mQabcd and hence, ra P abcd = 0 = ra Qabcd = 0.
The above form of the Lagrangian allows to show [4] that the EOM take in this case a simple form
✓ ◆
cde 1 (m) 1
Pb Racde gab L = Tab (5.19)
2 2
which does not involve more than the second derivatives of the metric (albeit for m 2 will be non
linear in the latters).
So in the end, the most general Lanczos–Lovelock (LL) Lagrangian is:
k
X
LLL = L (m) . (5.20)
m=1

Because of the antisymmetry of the deltas, we have that for 2m > D the term L (m) vanishes, so we
have a finite sum of di↵erent terms depending on the dimensionality of our manifold. If 2m = D the
term does not vanish, but it is a topological invariant (the Euler characteristic of that D-dimensional
manifold) and does not a↵ect the equations of motion, just like the Gauss–Bonnet term in D = 4.
Only terms for which D 2m + 1 introduce a non trivial dynamics.
Finally, one might wonder how in D = 4 dimension one could theoretically prefer GR over say its
LL counterpart given that SEP (at least in its formulation has GWEP+LLI+LPI) is satisfied by both.
Interestingly it is still true that these theories di↵ers in fundamental aspects such as gravitational waves
propagation (which in LL generically does not happen on the same background geometry and seems
connected to a problem of well-posedness of the Cauchy problem [26, 27]) or critical collapse [28]. The
fact that Lanczos–Lovelock theories comply with the GWEP albeit physically di↵erent from GR, seems
to indicate that there is potential room for improvement in the selection rules given by the equivalence
principles, perhaps by integrating them with some restrictions on the kind of self-interaction of the
gravitational degrees of freedom.

5.2.1 More about the Euler characteristic

The Euler characteristic is a topological invariant originally introduced for describing the topology of
polyhedra surfaces. It can be calculated according to the formula = V E +F where V, E, and F are
respectively the numbers of vertices (corners), edges and faces in the given polyhedron. Remarkably,
it is easy to see that any convex polyhedron’s surface has Euler characteristic = 2, which indeed is
also the Euler characteristic of a sphere.
More generally still, for any topological space, we can define the n-th Betti number bn as the rank
of the n-th singular homology group. The Euler characteristic can then be defined as the alternating
sum = b0 b1 + b2 b3 + · · · . This quantity is well-defined if the Betti numbers are all finite and if
they are zero beyond a certain index n⇤ .2
The Euler characteristic of a closed orientable surface can also be calculated from its genus g (the
number of tori in a connected sum decomposition of the surface; intuitively, the number of “handles”)
as = 2 2g as well as from the Gauss–Bonnet integral as shown before.
2
Homology itself was developed as a way to analyse and classify manifolds according to their cycles: closed loops
(or more generally submanifolds) that can be drawn on a given n dimensional manifold but not continuously deformed
into each other. Informally, the k-th Betti number refers to the number of k-dimensional holes on a topological surface.
The first few Betti numbers have the following definitions for 0-dimensional, 1-dimensional, and 2-dimensional simplicial
complexes: b0 is the number of connected components, b1 is the number of one-dimensional or “circular” holes, b2 is
the number of two-dimensional “voids” or “cavities”. Thus, for example, a torus has one connected surface component
so b0 = 1, two “circular” holes (one equatorial and one meridional) so b1 = 2, and a single cavity enclosed within the
surface so b2 = 1. Which give an Euler characteristic equal to zero.

48
 Figure 5.1: Illustration of a g = 3 surface, hence with = 4. (Image from en.wikipedia.org)

A funny application of these concepts is that of soccer balls. These are spheres covered by stitching
together pentagons and hexagons. If we call P the number of Pentagons and H that of hexagons then

Figure 5.2: Soccer ball and its pentagon-hexagon traditional tesselation. (Image from
en.wikipedia.org)

the number of faces is F = P + H, the number of vertexes is V = (5P + 6H)/3 (because at each vertex
three patches meet: two hexagons and one pentagon), and finally E = (5P + 6H)/2 edges (because
each edge is shared by two patches). This implies that V E + F = P/6 but we also know that
sphere = 2 = V E + F so we can deduce that any soccer ball, no matter how big it will be will
always have 12 pentagonal patches!

5.3 Relaxing SEP: F (R) Theories as a first example of higher

curvature gravity
If one decides to move away from SEP and fall back on requiring just the EEP then a whole set of
metric theories of gravity is explorable. A first obvious generalisation is to consider GR as the low
energy EFT action of a more complicated theory with a hierarchy of higher curvature terms. A simple
step in this direction would be to take the Gauss–Bonnet action with arbitrary coefficients (as all
its term have the same mass dimension). However, this is generally a dangerous path: the fourth
order equations of motions show that there are in general eight dynamical degrees of freedom that
in a suitable representation (choice of the field variables) can be associated to a massless graviton
(carrying two degrees of freedom), a scalar (carrying one degree of freedom) and a massive spin-2
particle (carrying five degree of freedom) where the latter is a ghost and hence is associated to an
instability.
While it is still possible to construct sane higher curvature theories involving terms in Riemann
and Ricci a very convenient framework to start with is to revert to theories which generalise GR by
changing the action from R ! F (R), where usually F (R) is taken to be an analytical function. The
gravitational action then becomes
Z Z
1 4 p
p
Sgrav = d x gF (R) 2 d3 x hF 0 (R)K (5.21)
16⇡G M @M

where F 0 (R) is the derivative of F (R) with respect to the Ricci scalar R. The second term, where K
is the trace of the extrinsic curvature of the manifold, is the analogous of the GHY term and we need

49
this term in order to cancel the part F 0 (R)g µ⌫ Rµ⌫ that one obtains when we perform the variation
of the first term.
However, from the variation of the second term we get two contributions: a term like F 0 (R) K
which cancels the term involving the variation of the Ricci shown above, and a term like K F 0 (R),
that will give the variation of the Ricci scalar. Hence, even if we started from a metric theory of
gravity without any new field, we would end up with a boundary term where F 0 (R) seems to be an
independent field, which needs to be fixed at the boundary.
Indeed if we write the equations of motion obtainable in this way we get
1
F 0 (R)Rµ⌫ F (R)gµ⌫ rµ r⌫ F 0 (R) + gµ⌫ 2F 0 (R) = 8⇡GTµ⌫ . (5.22)
2
As a sanity check, please note that if we put F (R) = R, the previous relation reduces to the Einstein
equations. These are fourth order partial di↵erential equations. The first part is similar to the Einstein
tensor, but the other two terms have no analogue in General Relativity. We can see that a 2F 0 (R)
appears, with a box operator in general implying a field propagation. We can also take the trace of
this expression and get
F 0 (R)R 2F (R) + 32F 0 (R) = 8⇡GT. (5.23)
This is a di↵erential equation, so, while in General Relativity R and T were algebraically related,
here we have a bigger set of solutions. Let us consider now a maximally symmetric solution, R = const.
If we consider the case T = 0, in General Relativity (in absence of a cosmological constant) we obtain
uniquely R = 0 (Minkowski spacetime); however, in this case, you can have also other solutions to
the equation (5.23), as R = C 6= 0, with Rµ⌫ = C4 gµ⌫ . This is interesting in a context related to
dark energy: as a matter of fact, depending on the choice of F 0 (R), we can have that the solution is
de Sitter or anti-de Sitter without introducing a cosmological constant which would violate the SEC,
since it would be readily provided by the theory without the need of exotic matter.

5.3.1 Palatini variation

We saw that in General Relativity, Palatini variation of the Einstein–Hilbert action gives the Einstein
equations, and tells us that the connection is provided by the Christo↵el symbols just from the
dynamics, without assuming it a priori. We now consider Palatini variation for F (R) theories, which
means that we are constructing a metric-affine F (R) theory, where the connection, again, is not fixed
to be the Christo↵el symbols, but has its own dynamics. We can write the Palatini action,
Z
1 p
SPal = d4 x gF (R) + Sm (g, i ) (5.24)
16⇡G
where with R we mean the Ricci scalar obtained considering symmetric connections that are not a
priori the Christo↵el symbols. If we vary with respect to the metric and with respect to the connection,
we respectively obtain the field equations
1
F 0 (R)Rµ⌫ gµ⌫ F (R) = 8⇡GTµ⌫ , (5.25a)
2
p
r( gF 0 (R)g µ⌫ ) = 0, (5.25b)
where, again, the covariant derivative is made from the connection. In analogy with what we did in
Palatini variation with the Einstein–Hilbert action, we see that the equation (5.25b) is forcing the
connection to be the Christo↵el symbols not of the metric gµ⌫ , but of a conformally related metric
hµ⌫ defined as p p
hhµ⌫ = gF 0 (R)gµ⌫ . (5.26)
Is it possible to eliminate the dependence of this new metric h and write the equations only in terms
of g? In principle yes. Let us write
0
µ⌫ = µ⌫ (h, @h) = µ⌫ (g, @g, F (R), @F 0 (R)); (5.27)

50
the connection can thus be defined in terms of g, F 0 (R) and their derivatives, and we have the relation

R = R( (h)) = R( (g, F 0 )). (5.28)

If we now take the trace of the field equation, (5.25a), we obtain

F 0 (R)R 2F (R) = 8⇡GT, (5.29)

that tells us that R is algebraically related to T (in analogy with the Einstein equations), hence
R = R(T, F 0 ). But in F 0 there is only the dependence from g and F 0 again, as we can see from
(5.28). Therefore we can in principle eliminate the independent connections from the field equations
and express them only in terms of the metric and the matter fields, once the shape of the function F
is known.
It can also be proven that the following relations between the Ricci scalars in the two di↵erent
conformally related metrics holds,
3 3
R=R+ (rµ F 0 (R))(rµ F 0 (R)) + 0 2F 0 (R), (5.30)
2F 0 (R)2 F (R)
which leads us to write the dynamical equation in the form:
✓ ◆
8⇡G 1 F (R) 1
Gµ⌫ = 0 Tµ⌫ gµ⌫ R 0
+ 0 (rµ r⌫ gµ⌫ 2)F 0 (R)
F (R) 2 F (R) F (R)

3 1 1
(rµ F 0 (R))(r⌫ F 0 (R)) gµ⌫ (rF 0 (R))2
2 F 0 (R)2 2
(5.31)
Knowing F (R) and the root of (5.29) R = R(T ) we have completely eliminated the dependence on
and everything is determined: all the terms are given as a function of g and F 0 . These are similar to
the Einstein equations, but with a big source term. Again, if F (R) = R we recover General Relativity.
However the source term contains derivatives of the stress energy tensor. This can lead to a
problem: if we have, for example, a star with a sharp boundary (a crust), the source term could
diverge; so the Palatini variation for F (R) theories does not work for discontinuous distributions of
matter.

5.4 Relaxing SEP: Generalized Brans–Dicke theories

In Brans–Dicke theories [29] we relax the SEP by promoting the Newton constant to be a field that
varies in spacetime (the term “generalised” that we use in the title of this section is due to the
fact that originally such theories did not include a potential term). In order to preserve background
independence of the theory we shall of course need to give a dynamics to such field by introducing a
kinetic term and a in general a potential as well. The action is of the form:
Z 
1 4 p ! µ⌫
Sgrav = d x g R g r µ r⌫ V( ) ; (5.32)
16⇡G0 M
as evident from the first term, G0 is a book-keeping constant, such that Newton’s constant is promoted
to a field GN = G0 that is a function of the spacetime point. The second term is a kinetic term that
gives the dynamics to the new field and the coupling depends on spacetime through ; the third
term is the potential.
The second term provides a long range interaction; using the Cassini probe, it has been proved that
the value of ! can be well constrained, ! > 40000. This in principle could be a good result because
with an high ! we have that the value of the field tends to be constant and it freezes-out (the field
becomes ultra-heavy and it is difficult to put it in motion); so for ! ! 1 we recover General Relativity
and the Newton constant appears to be constant. However, we can still make the theory compatible

51
with solar system tests of the GWEP if in the potential term we introduce an e↵ective mass to the
scalar field. This leads to a Yukawa potential term, so that an heavy field cannot e↵ectively propagate:
in this way, the problematic long range interaction is turned into short range, and therefore the tests
like the Cassini probe ones cannot constrain efficiently Brans–Dicke theories.

5.4.1 F (R) theories as generalized Brans–Dicke theories

The aim of this Subsection is to show that sometimes it is possible to map two theories of gravity one
in the other, even if in principle they appear to be very di↵erent. This can be explicitly seen for F (R)
theories, which were proposed in 1970 and discovered to be equal to special generalized Brans–Dicke
theories in 2006.

Metric F (R) Let us consider again metric F (R) theories (the connection is provided by the Christof-
fel symbols of the metric). We have seen above that F 0 (R) is like an extra degree of freedom (its
D’Alembert operator appears in the fields equations, Equation (5.23)). We use the auxiliary field
method to make this explicit. Let us rewrite the main part of the gravitational action in another form,
Z Z
1 p 1 p
Sgrav = d4 x gF (R) = d4 x g[F ( ) + F 0 ( )(R )], (5.33)
16⇡G M 16⇡G M
where we have introduced the auxiliary field . If we perform the variation with respect to , we
obtain = R if the condition F 00 ( ) 6= 0 holds. Therefore, imposing this last condition, by performing
the variation of the new action with respect to g, we obtain again the same equations of motion:
dynamically speaking, we have changed nothing. Let us now introduce a scalar field ⌘ F 0 ( ), and
define a potential V ( ) = ( ) F ( ( )); the action becomes
Z
1 p
Sgrav = d4 x g[ R V ( )]. (5.34)
16⇡G M
This equation makes explicit an extra degree of freedom that first was encoded in the function
F . Comparing (5.34) with (5.32), we see that metric F (R) theories with F 00 (R) 6= 0 are generalized
Brans–Dicke theories with ! = 0 and the potential V ( ) = ( ) F ( ( )).

Palatini F (R) If we instead start from the Palatini action and apply the auxiliary field method, we
obtain an action that can be written as
Z 
1 4 p 3 µ
SPal = d x g R+ r rµ V ( ) + Sm (g, i ), (5.35)
16⇡G M 2
where R is the Ricci scalar associated to the metric, expressed as a function of R using Equation
(5.30). This is a Brans–Dicke theory with ! = 32 , which can be proven to be singular. Indeed, in
general, Brans–Dicke theories have the property that, by rescaling the metric by a conformal factor
that is a function of , we can always get back to the Einstein gravity, with now coupled to the
matter. For ! = 32 this cannot be done. See also [30, 31].
Just to complete the picture of F (R) theories we show a picture that illustrates the link between
the di↵erent F (R) theories and the fact that they are mappable in particular cases of Brans–Dicke
theories.

5.5 Relaxing SEP: Scalar-Tensor Theories

Let us generalize the Brans–Dicke idea: we write down the most general scalar-tensor action compatible
with the Einstein equivalence principle,
Z  ⇣ ⌘
4 p A( )R B( ) µ⌫
SST = d x g g rµ r⌫ V ( ) + Sm e2↵( ) g, i , (5.36)
M 16⇡G 2

52
 f (R) GRAVITY

µ⌫ and gµ⌫ independent

t
METRIC-AFFINE f (R) µ⌫ = µ⌫

SM = SM (gµ⌫ , )
✏ %
PALATINI f (R) METRIC f (R)
KS KS
f (R) = R f (R) = R
* t
f 00 (R) 6= 0 GR f 00 (R) 6= 0

↵◆ ↵◆
3
BRANS–DICKE, !0 = 2 BRANS–DICKE, !0 = 0

Table 5.1: Classification of f (R) theories of gravity and equivalent Brans–Dicke theories (from [32]).

where we are allowing the scalar field to mediate the interaction between the metric and the matter
fields (no minimal coupling). Using the symmetries of the action, we can represent such theories in
two di↵erent but (classically) equivalent representations, called frames: the Einstein and the Jordan
frame. This will show once again that even if two gravity theories looks di↵erent they could have the
same physical content. As a matter of fact, we can always rescale the field without generating new
terms. Moreover we can always perform a conformal transformation (the action is invariant under
conformal symmetry) and rescale the metric of a function of (conformal symmetry). These two
transformations allow us to eliminate 2 of the 4 unknown functions in the action (A, B, ↵, V ).

5.5.1 Einstein frame

If we set A = B = 1 we get
Z " #
p R̃ 1 µ⌫ ⇣ ⌘
4
SST = d x g̃ g̃ r̃µ r̃⌫ V ( ) + Sm e2↵( ) g̃, i (5.37)
M 16⇡G 2

This is the Einstein frame representation: g̃ is the Einstein metric, and in the action we have an
Einstein–Hilbert part plus a standard kinetic and a potential term for the field . However, we do
not recover General Relativity, since the coupling of the matter fields with gravity is mediated by the
scalar field.3

5.5.2 Jordan frame

We can also set ↵ = 0, A = . In this case, we get
Z 
4 p R B( ) µ⌫
SST = d x g g rµ r⌫ V ( ) + Sm (g, i ). (5.38)
M 16⇡G 2

The gravity part is more complicated since there is a R coupling. The matter fields are coupled to
the Jordan metric in a minimal way, so the matter follows the geodesics of the Jordan metric, not of
the Einstein one. In fact, in the Einstein representation it is coupled to the Einstein metric through
the scalar field; the matter is never really free.
3
Of course, we still have the freedom to redefine our matter fields with a conformal transformation so that to reabsorb
the e2↵( ) . However our new fields, say ˜i = ⌦si ( ) i (where si its the i-field conformal weight) will be dependent on
(x), ˜i = ˜i ( ). This is tantamount to say that one does trade this simplification with spacetime position dependent
masses and coupling constants (which will still make the theory di↵erent from GR).

53
5.5.3 Horndeski theory
Now we can ask: is the action we wrote, depending on A, B, ↵, and V , the most general action that
can give second order field equations? The answer to this question is actually no: there is a more
general action, called Horndeski action [33].
Its general form is given by
5
X
L = Li ( , X), (5.39)
i=2

where we introduced X = @µ @µ /2 and the terms apearing in the sum are defined by:

L2 = K( , X) (5.40a)

L3 =
G3 ( , X)2 (5.40b)
⇥ ⇤
L4 = G4 ( , X)R + G4,X (2 )2 (rµ r⌫ )(rµ r⌫ ) (5.40c)
1 h i
L5 = G5 ( , X)Gµ⌫ (rµ r⌫ ) G5,X (2 )2 3(2 )(rµ r⌫ )(rµ r⌫ ) + 2(rµ r↵ )(r↵ r )(r rµ )
6
(5.40d)
Here, K and the Gi ’s are functions of and the kinetic term X, and Gi,X = @Gi /@X.
Horndeski theory admits an Einstein frame under some restrictions and is achieved via a disformal
transformation ḡµ⌫ = A( )gµ⌫ + B( )rµ r⌫ .

5.5.4 DHOST Theories

So far we blindly required for the modified theories to have second-order equations of motion. This
requirement is imposed for two reasons: first we want to fix the number of degrees-of-freedom of the
theory (higher-order equations need more initial conditions) and second we want to avoid instabilities
in the action. Indeed, in general higher-order equations imply the presence of additional degrees-of-
freedom which are ghost-unstable (meaning with the wrong sign for their kinetic term). For example, a
theorem guarantees that generically this sort of instabilities appears in the case of equations of motion
with higher than second order derivatives in time, the so called Ostrogradsky instability.4
However, it was suggested that — despite the appearance of higher order time derivatives — it is
still possible to avoid the Ostrogradsky ghost if the equations of motion can be recast into a second
order form by field redefinitions. This idea was firstly applied to Horndeski showing that it is possible
to add two more terms with one arbitrary function of the scalar field and its kinetic term entailing
third order derivatives which however do not lead to an Ostrogadsky instability. These terms and the
corresponding theory are called “beyond Horndeski”.
However, it was realised in recent years that even more generic Lagrangians with higher-order equa-
tions can be stable and propagate the desired number of degrees-of-freedom, provided some degeneracy
conditions are imposed on the theory. Such scalar-tensor theories go under the name of DHOST (De-
generate Higher Order Scalar-Tensor theories) and further generalise the Horndeski theory but still
propagate one scalar and one massless graviton.
We can elucidate the degeneracy-mechanism behind DHOST with an example from classical me-
chanics (for more details see e.g. [35]). Consider the Lagrangian for the variables (t) and h(t) defined
as
a c 1
L = ¨2 + b ¨ḣ + ḣ2 + ˙ 2 V ( , h) , (5.41)
2 2 2
4
Note that Ostrogadsky theorem does not straightforwardly apply to non-local EOMs (here meaning theories with
infinite higher derivatives): e.g. if one has a f (2) = 0 equation of motion. (Note however, that no general theorem guar-
antees the viability of any non-local theory, see e.g. [34]). Also, we shall see in what follows that constrained/degenerate
Lagrangians can evade the theorem, this is for example evidently the case for F (R) theories which have only an extra
scalar d.o.g. in addition to the graviton.

54
where a, b and c are generic coefficients and V ( , h) is a generic potential. The equations of motion
are ( .... ...
a + b h ¨ = @ V ( , h)
... . (5.42)
c ḧ b = @h V ( , h)
These equations are fourth-order for a 6= 0, third-order for a = 0 and b 6= 0 and second-order for
a = 0 = b. In total we need 6 initial conditions to find a solution, meaning we have 3 d.o.f. in general.
Despite the higher-order equations we can still reduce the number of d.o.f. by imposing a condition for
the coefficients. To show this, it is best to introduce an auxiliary variable Q and a Lagrange-multiplier
µ to put the system in a more familiar form:
a 2 c 1
L = Q̇ + b Q̇ḣ + ḣ2 + Q2 V ( , h) µ(Q ˙ ) . (5.43)
2 2 2
It is easy to see that this Lagrangian is equivalent to the previous one. At this point we can recast
the first three pieces in terms of the vector X T = (Q, h) as follows
✓ ◆
1 T 1 a b
L = Ẋ M Ẋ + Q2 V ( , h) µ(Q ˙ ) , M = . (5.44)
2 2 b c
From this equation we see that for det M 6= 0 we have 6 d.o.f., whereas for det M = 0 the kinetic
term acquires a zero-eigenvalue and the system only has 2 degrees-of-freedom. This is indeed the
degeneracy condition we were looking for; degenerate theories have parameters a, b and c satisfying
it. Furthermore one can also show that the resulting Hamiltonian is bounded-from-below, indicating
that indeed we avoid the presence of the Ostrogradsky ghost.
The same principle can be applied to scalar-tensor theories. Such setup can be made analogous to
this example by identifying (t) with the scalar field (x) and h(t) with the metric gµ⌫ (x). The most
general DHOST theory is given by [36]
µ⌫
L =f0 (X, ) + f1 (X, )⇤ + f2 (X, )R + f3 (X, )Gµ⌫ +
µ⌫⇢ µ⌫⇢ ↵
C(2) µ⌫ ⇢ + C(3) µ⌫ ⇢ ↵ , (5.45)
µ⌫⇢
where for convenience we define µ = rµ and the tensor C(2) can be decomposed as
5
X
µ⌫⇢ (2)
C(2) µ⌫ ⇢ = aA (X, )LA , (5.46)
A=1

with
(2) (2) (2)
L1 = µ⌫
µ⌫
, L2 = (⇤ )2 , L3 = (⇤ ) µ
µ⌫
⌫
,
(2) (2)
(5.47)
µ µ ⌫ 2
L4 = µ⇢ ⌫ , L5 =( µ⌫ ) ,
µ⌫⇢ ↵
and C(3) can be similarly decomposed as
10
X
µ⌫⇢ ↵ (3)
C(3) µ⌫ ⇢ ↵ = bA (X, )LA , (5.48)
A=1
(3)
with LA this time given by operators cubic in the second derivatives of (since it contains many
terms that are not essential for our discussion we do not write it explicitly).
For the theory (5.45), the degeneracy condition can be achieved by imposing some restrictions on
the functions f2 , f3 and aA and bA . If such condition is imposed the theory propagates one scalar and
one graviton and avoid the Ostrogradsky instability. Of course, appropriate choices for these functions
will give the Horndeski theory seen in the previous paragraph. We can imagine these DHOST theories
as the largest (so far) class of scalar-tensor theories propagating one scalar, as depicted in the diagram
of Figure 5.3.

55
 At this point one could be puzzled about the statement that higher-derivatives in the Lagrangian
lead to instabilities. Indeed, from the point of view of E↵ective Field Theories (EFTs) we know
that if we restrict to low energies and we integrate out some heavy field of mass M , we are going
@ 2n
to generate higher-derivative operators in the Lagrangian, containing schematically ⇠ M 2n .
Furthermore in general these operators do not have the structure appearing in Horndeski or
DHOST theories, and hence lead to higher-order equations (and additional d.o.f.). Usually in this
context one does not worry about Ostrogradsky instabilities, but why? The reason is that, as
one can easily check, the “additional” d.o.f. induced by these operators have masses ⇠ M , and
are therefore outside of the EFT. Whenever these operator become relevant (for instance when
we go to energies close to M ), the EFT breaks down and we need to consider the full theory.
Moreover, these operators are in general sub-dominant compared to other e↵ective operators with
a lower number of derivative and are therefore usually neglected. On the other hand, here we are
considering theories where higher-derivative operators can dominate the dynamics (e.g. in strong-
field regimes around Black Holes or in cosmology at late times), and in order to make sense of
them we need to impose the absence of instabilities. From this discussion we can also realize that
it is difficult to interpret such theories as standard EFTs where we integrate out weakly coupled
heavy fields. It is indeed an open problem to find possible UV completions for these scalar-tensor
theories.

5.6 Going further

This is far from an exhaustive list of generalized theories. Most noticeable we have not discussed
metric-affine theories of gravity such as Einstein–Cartan [37] (a generalization of GR including non
propagating torsion), Metric affine F (R) [38] or Lorentz breaking theories of gravity such as Einstein–
Aether [39] (in vacuum equivalent to GR plus a timelike vector setting a preferred frame) or Hořava–
Lifshitz gravity [40] (GR plus an hypersurface orthogonal vector which hence can be written as the
gradient of a scalar, the Kronon, providing a preferred foliation). We leave this to further study for
the interested student.
f.
ori ard

ski

.
ST
esk

d.o
rnd d
es

de

Ho eyon
thetand

O
rn

tra
DH
Ho

B
S

Ex

Figure 5.3: Schematic diagram of the various scalar theories discussed.

56
 6
Global Methods

6.1 Carter–Penrose Diagrams

Now we want to introduce global methods to discuss spacetime solutions and their dynamics, and it
would be useful if we could “see” the global structure of a spacetime, including infinities. Basically,
imagine that we want to describe the whole spacetime in such a way that we can draw it completely
on a piece of paper, and therefore we want to find a way to draw spatial and time infinities. The
starting point of this task is the light cone.
We can generally define one kind of spatial infinity, ◆0 , and two kind of time infinities: a future
time infinity ◆+ and a past time infinity, ◆ . Also, we can define two types of null infinities, reachable
by following the light-cone surface: we then have future null infinity I + and past null infinity I .
In total, we have 5 notions of infinity.

Figure 6.1: Representation of the di↵erent types of inifinity with respect to the light-cone.

The idea behind the so called Carter–Penrose diagrams is to start from this causal structure and
draw a diagram that conveys it in a compact form by bringing the infinities to a finite distance and
exposing it by a suitable conformal rescaling of the metric.

57
 Conformal transformations: Let (M, g) be a spacetime, and ⌦ : M ! R a nowhere-vanishing,
smooth function on M. This is called a conformal transformation, a scale transformation of the
metric g ! g̃ = ⌦2 g; this leaves invariant the angles between vectors, since

(x̃, ỹ) ⌦2 (x, y) (x, y)

cos ✓˜ = = 2 = = cos ✓, (6.1)
kx̃k kỹk ⌦ kxk kyk kxk kyk

but obviously not the norm of vectors. However, null vectors remain null:

g(x, x) = 0 ) g̃(x, x) = 0 (6.2)

In this way, we have introduced transformations that preserve the causal structure of spacetime.

6.1.1 Minkowski spacetime Carter–Penrose diagram

For example, let us take Minkowski spacetime; in spherical coordinates, the line element is

ds2 = dt2 + dr2 + r2 d⌦2 . (6.3)

As a first step, we reintroduce the null coordinates u ⌘ t r and v ⌘ t + r, that are such that u is
constant for outgoing photons and v is constant for ingoing photons; since r 0, then v u. With
this, the metric becomes
(v u)2
⌘ = du dv + d⌦2 . (6.4)
4
We now want to map infinities to a finite distance. We can do so by introducing suitable coordinates
of the form, U ⌘ arctan u, V = arctan v, which map the range 1 < u, v < +1, to the finite range
⇡/2 < U , V < ⇡/2.
By inverting the above coordinate transformation we also have u = tan U , v = tan V, du =
(1 + u2 ) dU , dv = (1 + v 2 ) dV, and thus
1
⌘= (1 + u2 )(1 + v 2 ) dU dV + (v u)2 d⌦2 . (6.5)
4
From this, we can change again coordinates by defining ⌧ ⌘ V + U , R ⌘ V U . We can see that ⌧ ± R
are limited to the real interval ( ⇡, ⇡). Moreover, from v u follows that V U and hence R 0.
The metric then becomes
1 1 ⇥ ⇤
⌘= 2 2
d⌧ 2 + dR2 + sin2 R d⌦2 . (6.6)
4 cos U cos V
We can make now two observations:

• In the above metric element R behaves as an angular variable.

• The Minkowski metric in this coordinates conformal appears to be conformal — with conformal
factor ⌦2 = 14 cos2 U1cos2 V — to the line element in square brackets, which is indeed (with the range
restrictions we have inherited) a portion of and Einstein static Universe which is homogeneous
and isotropic, and has a structure like R ⇥ S 3 .
⇡ ⇡
• we have now singularities at U = 2 and V = 2

The above points suggest an easy way forward: we define a fictitious metric ⌘˜ = ⌦ 2 ⌘ through
a conformal transformation, so to consider a portion of Einstein static Universe. Although distances
measured with the metric ⌘˜ will di↵er (by a possibly an infinite factor) from those measured with the
⌘ metric, the causal sructure will be the same in both metrics, with the advantage that the fictitious
metric ⌘˜ is well behaved at the values of (UV) correspondings to the asymptotic regions of ⌘.

58
 Figure 6.2: Spacetime diagram of Einstein static Universe.

If we draw this, by suppressing angular coordinates ✓ and ', and carefully taking into account the
ranges of our variables, we obtain what can be seen in Figure 6.2.
Therefore, each point of the cylinder is a 2-sphere of area 4⇡ sin2 R, timelike geodesics start at ◆
and end at ◆+ , null geodesics start at I and end at I + , while spacelike geodesics start and end at
◆0 .
If we cut this at R = ⇡ and unfold it, the shaded area will be the Carter-Penrose diagram of
Minkowski, Figure 6.3.

Figure 6.3: Carter-Penrose diagram of Minkowski spacetime.

The above example shows the basic procedure behind a causal diagram:
• Expose the causal structure by introducing a suitable set of null coordinates.
• Do a coordinate change that maps infinities to a finite range (normally using the arctan function
or something equivalent).
• Use a conformal transformation to get rid of annoying conformal factors and analyse the causal
structure of a simpler fictitious spacetime.
Let us then apply this technique now to a more interesting spacetime.

59
6.1.2 Schwarzschild spacetime Carter–Penrose diagram
One can perform a similar computation with Schwarzschild metric:
✓ ◆ ✓ ◆ 1
2M 2 2M
g= 1 dt + 1 dr2 + r2 d⌦2 . (6.7)
r r
First of all we want to introduce null coordinates. In order to find them in Schwarzschild spacetime,
let us study the behaviour of null rauys. One can set ds2 = 0 and ✓, = const.; then we immediately
find ✓ ◆2 ✓ ◆2
dt r
= , (6.8)
dr r 2M
from which we find the following relation between the time and radial coordinates
✓ ◆
r 2M
± t = r + 2M ln + const.; (6.9)
2M
we can now define the so-called Regge–Wheeler coordinate, or tortoise coordinate
✓ ◆
r 2M
r⇤ ⌘ r + 2M ln , (6.10)
2M
and then by performing a similar transformation as the one we made before,
(
u = t r⇤
(6.11)
v = t + r⇤ ,

which are Schwarzschild null coordinates, Schwarzschild metric becomes

✓ ◆
2M
g= 1 du dv + r2 d⌦2 , (6.12)
r
that still looks singular at r = 2M where r⇤ ! 1 and consequently u ! +1 and v ! 1.

Eddington–Finkelstein coordinates: The just introduced null coordinates for Schwarzschild

can be used to define a very useful set of coordinates the Eddington–Finkelstein ones. More
precisely we can write the metric in the Ingoing Eddington–Finkelstein coordinates (v, r)
✓ ◆
2M
g= 1 dv 2 + 2 dv dr + r2 d⌦2 , (6.13)
r

for which ingoing and outgoing radial light rays have trajectories respectively given by
dv dv 2
=0 and = , (6.14)
dr dr 1 2M/r

Alternatively one can have the so called Outgoing Eddington–Finkelstein coordinates (u, r)
✓ ◆
2M
g= 1 du2 + 2 du dr + r2 d⌦2 , (6.15)
r

for which outgoing and ingoing radial light rays have trajectories respectively given by
du du 2
=0 and = , (6.16)
dr dr 1 2M/r

See also Box 31.2 of [41].

60
 t*

r = 2M r

Figure 6.4: Eddington–Finkelstein plot showing the tilting of light cones close to the horizon of a black
hole (with the two angular dimensions suppressed). Please note that rays of constant v being ingoing
null ones are normally plotted at 45-degrees slant, just as they would be in flat spacetime. This is
equivalent to say that the vertical axis is defined w.r.t. the time t⇤ ⌘ v r = t + 2M ln |r/2M 1|.

The apparent singularity at r = 2M can be avoided by another change of coordinates. Let us then
introduce, U = e u/4M and V = ev/4M with implied ranges 1 < U < 0 and 0 < V < +1, so that
the metric becomes
32M 3 r/2M
g= e dU dV + r2 (u, v) d⌦2 ; (6.17)
r
These new coordinates are called Kruskal–Szekeres null coordinates. We can immediately see that, in
these coordinates, the metric is no longer singular in r = 2M or U, V = 0, not surprisingly given that we
already know that the singularity of the metric at the horizon is just a coordinate artefact. So nothing
forbids us from doing an analytical continuation and consider the whole range 1 < U, V < +1.
Finally, one can then again perform another coordinate transformation to get to Kruskal–Szekeres
cartesian coordinates: T ⌘ V +U2 , X ⌘
V U
2 , and the metric becomes

32M 3
ds2 = e r/2M
dT 2 + dX 2 + r2 d⌦2 . (6.18)
r
If we plot this metric, we get a Kruskal diagram, Figure 6.5.

T
II
X
IV I

III

Figure 6.5: Kruskal diagram of a Schwarzschild black hole.

However, this diagram is still not capturing the asymptotic behaviour of the geometry. For that
we shall need to introduce again a coordinate transformation based on the arctan function. Let us

61
then introduce a new set of coordinates
✓ ◆ ✓ ◆
U V
U = arctan p V = arctan p , (6.19)
M 2M
⇡ ⇡ ⇡ ⇡
we have coordinates that are, as before, in a finite range, 2 < V < 2 and 2 < U < 2. If we
perform one final coordinate transformation
V +U V U
T̄ = X̄ = , (6.20)
2 2
We get a metric that after a suitable conformal rescaling (Exercise: compute the conformal factor)
leads to the the conformal diagram shown in Figure 6.6.

‘
t = const
II
‘
IV I
III
‘

r* = const

Figure 6.6: Carter–Penrose diagram of the maximal analytical extension of a Schwarzschild spacetime.
This is sometime said to describe an eternal black hole.

Again, here each point is actually a 2-sphere. Di↵erently from Minkowski, however, now I ± , ◆± ,
and ◆0 are all made of disconnected pieces; moreover, not all null geodesics end on I + or begin from
I , and this can actually give us a hint on the notion of the complex causal structure associated to
a black hole.
Here we have four di↵erent regions: region I and region IV are asymptotically flat regions, which
represent the “outside” of a black hole, and thus here I ± [◆0 is identical to what we find in Minkowki.
Here we defined an asymptotically flat spacetime as a spacetime with some asymptotic structure of
Minkowski.
This actually implies that Schwarzschild metric possesses the same Killing vectors of Minkowski
at infinity, which will be important for allowing to define a mass M and an angular momentum J on
spacelike hypersurfaces.
Region II is the standard black hole region, with the wavy line representing the black hole singu-
larity of Schwarzschild spacetime in r = 0; notice that the only way to exit region II is either ending
into the singularity in r = 0 or escaping to region I or IV by travelling faster than light. Also notice
that the singularity is spacelike.
Region III is a region called white hole, which is the temporal inverse of a black hole: if in a
black hole nothing can exit and everything must remain inside, in a white hole nothing can enter and
everything must leave.

Remarkably, it is possible to create an analogue of a white hole in a sink [42]! This is connected
to shallow water waves: perturbations that propagate in a fluid under the influence of gravity
in a shallow water basin. Indeed it is possible to show that these perturbations propagate at long
wavelengths with a wave equation o a curved geometry determined by the fluid low [43].
1 h k2 k
i
ds2 = (c2 vB ) dt2 2vB · dx dt + dx · dx ,
c2

62
 p
where c ⌘ ghB where g is the gravitational acceleration and hB is the depth of the flow. So the
speed of these waves depends on the height of the basin in which they are flowing. Hence, in a
shallow water regime, these perturbations are slow. Let’s now consider the experiment where we
open out kitchen tap over a flat surface. The water forms a column that spreads radially as it
meets the plate and forms a circle centred around the column. This circle marks the transition
from supersonic to subsonic flow for the gravity waves. On the circle the water has the same
speed as the waves in that fluid. If we create perturbations in the water outside the circle so they
attempt to enter the circle, we would see that they would actually bounce o↵ it, in the same way
in which it is impossible to enter a white hole [42].

Is this maximally analytic extension realistic? The Carter-Penrose diagram of an eternal black hole
only reflects the fact that, per sé, General Relativity solutions are time reversible, so if there is a black
hole, there is also a white hole. However, initial conditions break time reversal, for example in the
case in which a black hole is generated by a collapse of star. If we have a collapse, the Carter–Penrose
diagram of Schwarzschild spacetime becomes like the one shown in Figure 6.7

Figure 6.7: Carter-Penrose diagram of a collapsing star generating a Schwarzschild black hole.

We see that in this case the white hole is no more present; nonetheless, we can also see that the
causal structure of black holes from collapse at late times would be undistinguishable from that of
eternal black holes (this is rigorously proven in [44]).

6.2 Black Holes: Singularities and Event Horizon

Black holes are, in general, objects that possess horizons and singularities. Let us start with the latter.
What is a singularity? Most often is identified as a set of events where the metric has some pathology,
or the curvature becomes infinity. However, the metric is a solution of our field equations and we
cannot assume it a priori. Hence an event in spacetime has a physical meaning only if (M, g) is well
defined at such event. This in turns implies that a singularity cannot be a set of events in spacetime
but has to be rather characterised as a set of missing events in (M, g), i.e. where spacetime itself is
not defined.

6.2.1 Singularities
More formally,an inextendible spacetime has a singularity if there is at least one incomplete geodesic.
Here, with inextendible spacetime we mean a spacetime which literally misses a point: for
example, if we take Minkowski spacetime and we trivially remove a point, that is not considered

63
inextendible. Therefore, we define an inextendible spacetime as a spacetime which is not isometric to
a proper subset of another spacetime.
An incomplete geodesic is a geodesic in which it is possible to reach the end of such geodesic
at a finite value of its affine parameter. I.e. a geodesic from which we cannot extend our geodesic
beyond a finite value of the affine parameter.
Note that albeit not completely bullet proof (there can be special cases when the first derivatives
of the metric are OK but the second derivatives blow up, and hence the geodesic are Ok but the
curvature is not) it is by far preferable to other definitions used in the past.

• Look for a singular behaviour in the metric: in that case we could have simply a coordinate
artefact. For example consider the metric g = (4|x|) 1 dx 2 2
p + dy which looks singular in x = 0.
2 2
However, is enough to intruduce the coordinate X = ± |x| to see that g = dX + dy i.e. that
the metric is just that of the Euclidean plane.

• Look for some curvature scalar blowing up. Unfortunately even if a scalar does blow up one can
still have a Riemann well behaved everywere except at infinity. Alternatively, one can have well
behaved scalars but a ill behaved Riemann.

• Also one can have curvature singularities without having a ill behaved Riemann tensor. For
example if we consider e a flat disk from which we excise a slice and glue the opposite side
together we still get a metric g = ⇢2 d✓2 + d⇢2 but with ✓ 2
/ [0, 2⇡] which corresponds to have a
conical singularity in ⇢ = 0 even if Riemann is zero everywhere else.

We can classify singularities in 3 di↵erent types: we could have future spacelike singularities (like
the singularity in r = 0 of a Schwarzschild black hole or the Big Crunch), past spacelike singularities
(like the physical singularity of a white hole, or the Big Bang) and timelike singularities (that are
present, for example, in rotating or charged black holes, or naked singularities).

6.2.2 Horizons
Now we want to define the notion of horizon. In order to do that, we need to define some concepts.
If at each point of a spacetime (M, g) we can locally distinguish a future and a past lightcone,
then we say that this spacetime is time orientable.
In this case, there must be a continuous and everywhere non-zero, globally defined, timelike vector
field v µ . A spacetime is space orientable if there is a three-form ✏[µ⌫ ] 6= 0 everywhere and continuous,
and if v µ ✏[µ⌫ ] = 0 for any v µ timelike [45].
A spacetime is orientable if and only if there is a continuous, everywhere non-null, globally defined
4-form ✏[µ⌫ ⇢] , and this is the volume form dx0 ^ · · · ^ x3 .
If a spacetime is both time orientable and space orientable, then ✏[µ⌫⇢ ] = v[µ ✏⌫⇢ ] , so saying that
a spacetime is both time orientable and space orientable is equivalent to say that it is orientable.

We can define chiral fermions in a spacetime if and only if ✏µ⌫ ⇢ exists. In fact, 5 =± and

0 1 2 3 sign([abcd]) a b c d
5 =i =i (6.21)
4!
in Minkowski, while in a curved spacetime it becomes
✏µ⌫ ⇢ µ ⌫
5 =i e e e e⇢ a b c d
(6.22)
4! a b c d

The chronological future of a point p, I + (p) is the set of points q of the manifold M such that
there exists a future directed timelike curve with (0) = p and (1) = q (with an appropriate affine
parametrization). Similarly, the chronological past of a point p, I (p) is the set of points q of the
manifold M such that there exists a past directed timelike curve with (0) = p and (1) = q.

64
 We can also define the causal future of a point p, denoted as J + (p) as the set of points q of
the manifold M such that there exists a future directed causal curve with (0) = p and (1) = q.
Similarly, the causal past of a point p, denoted as J (p) is the set of points q of the manifold M
such that there exists a past directed causal curve with (0) = p and (1) = q.
We can also define the chronological and causal pasts and futures of a subset of the manifold,
U ⇢ M, in a very intuitive way, as
[ [
I ± (U ) = I ± (p) J ± (U ) = J ± (p). (6.23)
p2U p2U

With these definitions in mind, we can define a black hole region Bbh as:

Bbh ⌘ M J (I + ), (6.24)

where we include the asymptotes in M. This definition characterises the fact that there is a portion
of events in our spacetime which will be never causally connected with I + . We can also define the
future event horizon as the border of this region [18]

H+ ⌘ J¯˙ (I + ); (6.25)

where we have introduced the border of the topological closure of the causal past of a set of points U
as J¯˙ (U ) = J¯ (U ) J (U ) with J¯ (U ) being the topological closure of the set (the intersection of
all closed sets containing J (U )) and J (U ) being the union of all the open sets containing J (U ),
which in the case of an open set coincides with the set itself. Note that the subtraction implicit in
the definition J¯˙ (I + ) is important as it gets rid of ◆0 and I — contained in J (I + ) — which
otherwise we would also be included in the event horizon.
Of course, all of this can also be extended to a white hole:

Bwh ⌘ M J + (I );

and analogously define a past event horizon

H ⌘ J¯˙+ (I ).

Notice that we have defined the event horizon of a black hole and the black hole region by using
future null infinity, I + . This means that, in order to properly define the event horizon, we need to
know a priori what is the infinite future evolution of our spacetime, and in particular what is the I +
of our manifold M. The event horizon defined in this way is, therefore, a teleological concept, but we
shall see that there are also other notions of horizon that are much more down-to-earth and e↵ective
when we want to describe physical processes and interactions with black holes.

Some final remarks. Event horizons are null hypersurfaces. Therefore they can be seen as a
collection of null geodesics which are the generators of the null hypersurface. Also we can always find
a null vector lµ which is affinely parametrized on the horizon such that lµ rµ l⌫ |H+ = 0.
No two points on the horizon can be timelike separated, it will be always null. Locally, this
is obvious for the very fact that the horizon is a null hypersurface. Globally, it can be proven by
contradiction: take two points, p and q, on the event horizon H+ . Suppose now that the curve
connecting q to p is timelike so that p 2 J (q). Consider two points: p0 close to p slightly inside
H+ and q 0 close to q slightly outside. The timelike curve between p and q could then be deformed
to a nearby timelike curve between p0 and q 0 with q 0 2 J (I + ) and p0 2 / J (I + ). However, this
0 0 0 0
also implies that p 2 J (q ) , but since q is outside the horizon, J (q ) ⇢ J (I + ), and therefore
p0 2 J (I + ). Hence, we get a contradiction.
There is also a theorem (due to Penrose) according to which the generators of the event horizon
can have a past end points but no future end points: a null light ray can become a generator of the

65
Figure 6.8: Illustration of the demonstration that two points on the horizon cannot be connected by
a timelike curve.

horizon, but the contrary is not possible. Accordingly given appropriate energy conditions, horizons
cannot bifurcate and black holes cannot split. Note that black hole evaporation by Hawking radiation
does imply that there can be a future end-point; however, in this case, all of the energy conditions are
violated.

Figure 6.9: Illustration of the implication of past and future end points for the horizon generators.

6.3 Killing Horizons and Surface Gravity

Let ⇠ a be a Killing vector field in region of spacetime which contains a null hypersurface S. If ⇠ a ? S,
then S is said to be a Killing horizon.
A theorem by S. W. Hawking states that in GR: the event horizon of a stationary asymptotically
flat spacetime is a Killing horizon.1 . Note:
• A stationary spacetime is a spacetime which admits a globally defined, non vanishing, Killing
vector which is timelike near infinity.
• A static spacetime is a stationary spacetime with a Killing vector which is also hypersurface
orthogonal, i.e. irrotational.
Hawking theorem implies that for stationary black holes it is possible to find a Killing vector
orthogonal to the horizon. But we also know that there is also another null vector along the surface,
whose orbits are the generators of the null hypersurface: i.e. there must be on H+ a vector lµ which
is both null and affinely parametrized, so lµ rµ l⌫ |H+ = 0.
Since ⇠ and l are both null vectors, we can write ⇠ = f (x)l, and also ⇠ µ rµ ⇠ ⌫ = ⇠ ⌫ |H+ , since
we cannot assume a priori that they are both affinely parametrized. The variable  is associated to
the failure of the geodesic to be affinely parametrized, and it is called surface gravity of the Killing
horizon. We can easily derive  by considering on the horizon
⇠ µ rµ ⇠ ⌫ = f lµ rµ (f l⌫ ) = f (lµ @µ f ) l⌫ + f 2⇠ ⇠
rµ⇠
lµ ⇠ l⌫ = f lµ @µ ln |f | f l⌫ = ⇠ µ @µ ln |f | ⇠ ⌫ ⌘ ⇠ ⌫ , (6.26)
1
The converse is not true: a Killing horizon does not need to be an event horizon

66
so that
 = ⇠ µ @µ ln |f | . (6.27)
In the case that the surface gravity is zero, the black hole is said to be extremal.

We can now apply the above definition to the surface gravity of a Schwarzschild black hole.
Consider the Killing vector ⇠ = @t @
, with components ⇠ µ = 0µ . This is also the Killing vector
orthogonal to the Killing horizon. However, this is not the generator of the Killing horizon.
@
Indeed one can check that it is l = @V which satisfies lµ rµ l⌫ |H+ = 0 as indeed U and V coincide
with the affine parameters along ingoing and outgoing null geodesics respectively (alternatively
one can directly proof this see e.g. [18] page 29).
@
The Killing vector ⇠ = @t can be written in terms of the exponential null coordinates U and
V as
@ @V @ @U @
⇠= = + ; (6.28)
@t @t @V @t @U
in these coordinates, the horizon is located in U = 0 (indeed, U / e u/4M , and at r = 2M , the
tortoise coordinate r⇤ ! 1, which means that u = (t r⇤ ) ! +1 and U = 0 ). Then at the
horizon we have
@ @V @
= , (6.29)
@t @t @V
⇤ )/4M
and since V = ev/4M = e(t+r , we get
@ V @
⇠= = . (6.30)
@t 4M @V
V @
This is tantamount to say that ⇠ = f l, with f = 4M and l = @V . So surface gravity is

V @ V V 1 1
 = ⇠ µ @µ ln |f | = ln = = , (6.31)
4M @V 4M 4M V 4M

What is its physical meaning? Why is this relevant?

Let us start again with ⇠ µ rµ ⇠ ⌫ = in ⇠ ⌫ |H+ , where the subscript stands for “inaffinity”. Now,
introduce a point splitting so that we can compute the same quantity in the coincidence limit of a
point p that lies on the horizon and a point q slightly outside of it

2 (⇠ µ rµ ⇠ )(⇠ ⌫ r⌫ ⇠ )
in = lim . (6.32)
q!p ⇠ ⇢ ⇠⇢ q

This formula is useful because it allows a physical interpretation of the surface gravity. Let us start
by recognising that the normalized acceleration along the orbits of ⇠ µ can be written as

(⇠ µ rµ ⇠ )
a ⌘ , (6.33)
⇠ ⌫ ⇠⌫

and that we can make a scalar out of it by a ⌘ (a a )1/2 . We also know the redshift factor is just
⌘ ( ⇠ µ ⇠µ )1/2 . Therefore, we can see that the surface gravity in this limiting procedure is given by

in = lim ( a), (6.34)

q!p

and we notice that basically the surface gravity is the acceleration per unit mass needed to remain
static and hovering just over the horizon as measured from infinity (since there is a redshift factor).
If we lower an object attached with a rope, this is the acceleration we have to give the rope in order
to keep the object hovering above the horizon. At the horizon, the acceleration a ! 1 while ! 0,
so that their product remains finite.

67
 It can be proven that  is constant on the horizon H+ (we shall show this explicitly later). Also,
if H+ is the event horizon of a Killing vector ⇠, it is also the event horizon of a vector ⇠¯ = C⇠: in
this case, we would have ̄ = C, and this means that the normalization of the surface gravity is
arbitrary. However, in general the normalization is set in such a way that at infinity the Killing vector
⇠ becomes exactly the one of Minkowski spacetime, that has norm 1. We would be left then with an
arbitrary choice on the sign of the Killing vector, but this can be resolved by the condition that the
Killing vector is future-directed.

6.3.1 Surface gravity: alternative definitions

In general, there are many other definitions of surface gravity, and they coincide with the ones given
here only in a stationary spacetime and when we are considering Killing horizons (see e.g. [46] for a
more detailed review).
For example, we can introduce the peeling surface gravity, peeling : consider a generic spherically
symmetric metric (not necessarily stationary)
✓ ◆ ✓ ◆
2 2 (r,t) 2M (r, t) 2 2M (r, t) 1 2
ds = e 1 dt + 1 dr + r2 d⌦2 . (6.35)
r r
We will define an evolving horizon as the place where g00 = 0. This is not exactly an event horizon,
but it is a dynamical horizon rH such that
2M (rH , t)
= 1. (6.36)
rH
Radial null geodesics in this metric will satisfy the relation
✓ ◆
dr 2M
= ±e 1 . (6.37)
dt r
Now, consider a geodesic infinitesimally close to the horizon, and expands around it. We get
✓ ◆
dr 1 2M 0
= ±e (r(t) rH (t)) + O(r rH )2 ; (6.38)
dt rH
we define peeling as the coefficient
✓ ◆
1 1 2M 0
peeling = e , (6.39)
2 rH

and therefore dr
dt = ±2p (t) (r(t) rH (t)) + . . . . One can now get that the distance of the geodesic
from the horizon as a function of time is
 Z
|r rH |(t) ' |r rH | (t0 ) exp 2 p (t) dt . (6.40)

Similarly for two nearby null geodesics r1 (t), r2 (t) one gets
 Z
|r1 r2 |(t) ' |r1 r2 | (t0 ) exp 2 p (t) dt . (6.41)

1
In Schwarzschild limit, ! 0 and M (r, t) = const = M , so that p = 4M = in . More generally,
two notion of surface gravity always coincides in stationary situations, when the horzion is also a
Killing horizon. However, p better captures how null geodesics “peel” away from the horizon. If we
plot u = const and v = const lines in a (t⇤ , r) plane, we would see that the u = const lines would peel
away exponentially from the horizon in a way ruled by this factor.

68
 This is also related to Hawking radiation: one has this exponential peeling, ruled by p , both
inside and outside the horizon. In the already discussed heuristic picture for Hawking radiation
(see the section of geodesic deviation equation) a virtual pair forms at the two sides of the horizon,
it will break apart with a law similar to the previous equation, and therefore one particle will end
in the black hole, while the other one will escape to infinity. The above mentioned peeling null
rays on the two sides of the horizon are those followed by these Hawking pairs. This is why when
we have an evaporating black hole, its temperature depends really on the peeling surface gravity
and the latter will coincides with in only in stationary spacetimes (see e.g. [47]).

Another definition for surface gravity we can give is the so-called normal surface gravity, normal ,
⇣ ⌘
ra ⇠ b ⇠b ⌘ 2normal ⇠ a . (6.42)

Indeed, we know that H+ is null and that ⇠ µ is normal to H+ , but being a null vector, we can
write (⇠ µ ⇠µ ) |H+ = const = 0, so that this norm can be taken as the function defining the horizon
hypersurface. Now this implies that if we consider the vector r⌫ (⇠ µ ⇠µ ) this will be orthogonal to H+
as ⇠ ⌫ is. Consequently they should be proportional to each other

r⌫ (⇠ µ ⇠µ ) / ⇠ ⌫ . (6.43)

with the proportionality factor defining the normal surface gravity as above.
Again also this notion of surface gravity is equivalent to the inaffinity surface gravity when dealing
with stationary black holes, i.e. with Killing horizons. This can be seen using the Killing equation: in
fact, we can write
1 ⇣ ⌘
normal ⇠ a ⌘ ra ⇠ b ⇠b = ⇠b ra ⇠ b = ⇠b rb ⇠ a ⌘ in ⇠ a , (6.44)
2
I.e. normal = in .
Another definition of surface gravity is the generator surface gravity g :
1 ⇣ [a b] ⌘
2g = r ⇠ r[a ⇠b] . (6.45)
2
If the Killing equation holds, this is still equivalent to in ; in order to prove this (Exercise), one
has to use the Frobenius theorem, that holds whenever we have a hypersurface orthogonal vector and
assures that ⇠[µ r⌫ ⇠⇢] |H = 0.

6.4 Cauchy Horizon

There are situation in which one would like to deal with the evolution of spacetimes which are neither
static nor stationary. For example if we want to describe black holes mergers, evolution, and so on.
In order to do that, we need to introduce some important concepts needed to be able to describe
the evolution of the spacetime from initial conditions: i.e. one needs to solve a Cauchy problem. In
General Relativity, solving a Cauchy problem requires that we have a good hypersurface on which our
initial conditions can be provided. This hypersurface is called Cauchy hypersurface.
In order to define these hypersurfaces formally, we need some preliminary concepts. First of all,
we need to define an acronal set. An acronal set is a subset of the manifold S ⇢ M such that, for
any two points p, q 2 S we have that q 2/ I ± (p); in practice, this means that there cannot be a timelike
curve that links them, and therefore I ± (S) \ S = ?.
Another concept is that of domain of dependence. The future domain of dependence is defined
as the set of points p such that every past inextensible causal curve intersects S.
Here, with inextensible we mean that there are no future or past endpoints: a past/future endpoint
is a point p such that, for any neighbourhood O(p) of p there is a t0 so that for any (t < t0 ) or

69
(t > t0 ) respectively, one has (t) 2 O(p). (Note also that this definition is di↵erent from the case of
incomplete geodesics for which there are no t > t0 6= 1.) Similarly, the past domain of dependence is
the set of points q such that every future inextensible causal curve intersects S.
We denote the future and past domain of dependence of a point p as D± (p), and similarly for a
submanifold S, D± (S). We then define a Cauchy hypersurface as an acronal, non-compact surface
⌃ for which D(⌃) ⌘ D (⌃) [ D+ (⌃) coincides with the whole manifold, D(⌃) = M.2 Given some
initial data on a Cauchy hypersurface, one can evolve backward and forward the whole manifold. A
spacetime that admits a Cauchy hypersurface is called globally hyperbolic.
An example of a globally hyperbolic spacetime is given by the spacetime of a collapsing star, since
we can construct Cauchy hypersurfaces covering the whole manifold.
There might be limits to the Cauchy evolution: these are called Cauchy horizons. If a spacetime
is not globally hyperbolic, it happens when D+ or D has a boundary; there is no surface we can find
such that D+ and D fill the whole spacetime.
There are two types of Cauchy horizons: there are Cauchy horizons from timelike singularities and
Cauchy horizons from time machines (or causality violations, or closed timelike curves).

‘ region I

Figure 6.10: Carter-Penrose diagram of the Reissner-Nordström spacetime, highlighting the Cauchy
horizon.

An example of Cauchy horizon from timelike singularities can be found, for example, in the
Reissner-Nordström metric, which is the metric around a static and stationary black hole with mass
M and electric charge Q; this is given by simply substituting in Schwarzschild metric
✓ ◆ ✓ ◆
2M 2M Q2
1 ! 1 + 2 ; (6.46)
r r r

the Cauchy horizon in this case is the one visible in Figure 6.10 for the non extremal case, Q2 < M 2
(the extremal RN black hole being the solution with Q2 = M 2 and the naked singularity case being
the one with Q2 > M 2 ). It is clear that the presence of the timelike singularity prevents us to evolve
from a set of initial data given on the drawn hypersurface to the region beyond the Cauchy horizon.
A simple example of a Cauchy horizon from causality violation can be given for a 1 + 1 geometry
of a cylindrical spacetime with (t, x) = (t, x + 1) with metric

ds2 = cos( ) dt2 sin( ) ( dt dx + dx dt) + cos( ) dx2 ; (6.47)

here, = cot 1 (t), and thus (t = 1) = 0 and (t = +1) = ⇡; the structure of the spacetime is
then similar to the one visible in Figure 6.11. In this case, the Cauchy horizon is called a chronological
horizon.
Finally, let us comment that for a future Cauchy horizon the entire infinite history of the external
spacetime in region I, and its counterpart on the right side of figure 6.11, is in its causal past, i.e.
2
There is also a weaker concept of Cauchy surface namely that of a partial Cauchy surface which for for a spacetime
M is a hypersurface which no causal curve intersects more than once.

70
Figure 6.11: Causal structure of a causality-violating spacetime. The Cauchy horizon is the point
where the light-cones tilt enough to permit the existence of a closed null curve.

visible, so signals from I must undergo an infinite blueshift as they approach the Cauchy horizon. For
this reason, Cauchy horizons are generally considered to be unstable under classical perturbations no
matter how small. So, for any physically realistic collapse, the Cauchy horizon is often seen as a singular
null hypersurface for which new physics beyond GR is needed. Let us just add for completeness that
this issue is not completely settle as recently it was claimed that in spacetimes with a cosmological
constant stable Cauchy horizons might exist (see [48] and follow up literature) so violating the Cosmic
censorship conjecture in its strong form.

The Cosmic Censorship Conjecture: After the realisation that singularities are a generic
prediction of GR (see singularity theorems later on in this lecture notes), it was understood that
if they could be observed to outside observers then they could mark a sharp breakdown of the
predictive power of physics (at least the classical one). For this reason was conjectured by Penrose
that singularities should always be screened by horizons (non naked singularities). This concept
has later on evolved in a distinction between a weak cosmic censorship and a strong cosmic
censorship conjectures to take also int account the breakdown of Cauchy evolution introduced by
Cauchy horizons

Weak Cosmic censorship conjecture : No singularity can be visible from future null infinity.
I.e. for generic initial data, the maximal Cauchy development possesses a complete future
null infinity.

Strong Cosmic censorship conjecture : General relativity is a deterministic theory, in the

same sense that classical mechanics is a deterministic theory. In other words, physical
solutions of the theory should always be globally hyperbolic spacetimes, which implies that
the maximal Cauchy development of generic compact or asymptotically flat initial data is
locally inextendible as a regular Lorentzian manifold.

It is evident that solutions with Cauchy horizons violate the strong cosmic censorship while they
are compatible with the weak one.

71
6.5 Cosmological Horizon
Sometimes the term Hubble horizon is used in cosmology, however what it is then defined as a horizon
is not really a horizon in the causal sense. For a more detailed discussion see [49, 50].
Consider the FRW metric
8
>
<sin k=1
2 2 2 2
⇥ 2 2 2
⇤
ds = c dt + R (t) d + S ( ) d⌦ , where S ( ) = k=0 (6.48)
>
:
sinh k = 1.

In general, we can write the proper distance of an object as a function of time as D(t) = R(t) (z).
If we di↵erentiate with respect to proper time we get that the velocity with which an object, observed
at redshift z now, gets far from us is
V (t, z) = Ḋ(t) = Ṙ(t) (z) + R(t) ˙ (z), (6.49)
where the first term is the recession velocity Vrec and the second term is the peculiar velocity Vpec .3
Even neglecting peculiar motion ( ˙ = 0), one can easily see that the recession velocity e.g. of a
galaxy from us is a time dependent quantity as the expansion of the universe varies in time. The
current recession velocity of a galaxy observed now at t = t0 will then be V = Ṙ0 (z).
The Hubble sphere is defined as the radius when the velocity of recession is equal to c; therefore,
neglecting the peculiar velocity for the moment we get (z) = c/Ṙ(t) and hence the Hubble radius is
c
DH = ; (6.50)
H
where H(t) = Ḋ(t)/D(t). This is not an horizon, it is the speed of light sphere (SLS) radius and
corresponds to a redshift of roughly zSLS = 1.46 in ⇤CDM; therefore, since we can see beyond this
distance, this is definitely not an horizon.
For photons, we can write c dt = R(t) d , and from this we find a peculiar velocity vp = R ˙ = c.
This also implies that the total velocity for photons travelling towards us is V = Vrec c which is still
positive (away from us) if Vrec > c, i.e. if they are outside our Hubble sphere. However, the Hubble
sphere expands in time (e.g. in a matter dominated universe DH = 3ct/2) and so it can overtake these
receding incoming photons that will then find themselves in a region with Vrec < c and hence will be
able to get to us.
Also, with photons, we can define the so called particle horizon, which again is not an event horizon.
This is defined in comoving distance as
Z t
dt0
PH (t) = c 0
, (6.51)
0 R(t )

and this tells us that the coordinate position of a particle after some time t from when it was emitted
goes like the integral above. This defines the region that can be causally connected at time t after the
Big Bang. In physical distance, this is of course DPH (t) = R(t) PH (t).
The event horizon tells us the boundary of the region that can be causally connected from the
start to the end of the Universe; therefore, the only event horizon we can define in cosmology is
Z 1
dt0
EH = c . (6.52)
0 R(t0 )
p
Now, if our Universe were asymptotically de Sitter, we would get asymptotically R(t) = exp ⇤/3t,
so we would find that this horizon would be finite and equal top the size of the de Sitter Universe,
which scales with the cosmological constant as DEH = c/H = c 3/⇤ and coincides in this special
case with the Hubble sphere.
3
Note that we can always “trade” redshift for time if we know the universe content, i.e. R(t), via the relation
(1 + z) = R0 /R(t).

72
6.6 Local black hole horizons
As we discussed above the characterization of a black hole horizon as event horizon is very straight-
forward and tightly linked to the causal structure of spacetime. We shalls ee later on that it plays a
crucial role in the definition of the four laws of black hole mechanics. It is however, as we also stressed,
an intrinsically non-local definition which is also teleological in nature: defining the EH of a black hole
requires the full knowledge of the spacetime in the infinite future. Hence, per se, these are not useful
concepts when dealing e.g. with dynamical processes or astrophysical observation (see e.g. [51]). For
this reason alternative notions of a horizon, more intrinsic and local, have been proposed. But in order
to understand them we shall need first to add a few notions to our toolkit.

6.6.1 Trapped Surfaces and Trapped Regions

We now want to introduce now the concept of trapped surfaces. Consider Figure 6.12 were we have
time in the vertical axis and radius and one angle on a t = constant slice (i.e. we have suppressed the
extra space dimension). There are two possible null rays that can be emitted from each point of the

Figure 6.12:

initial spacelike 2-surface S 2 : ingoing light rays, along vector n, and outgoing light rays, along vector
l, with l and n null vectors. Being both l and n null vectors, they are both orthogonal to S 2 .
In Minkowski, all the ingoing light rays will converge in a point and form a kind of cone with
a“bowl”, made out of outgoing light rays, around it (Figure 6.13). We define F 4 the chronological
future of S 2 , F 4 ⌘ I + (S 2 ) and B 3 the null hypersurface bounding it B 3 = @F 4 . It is easy to see that
this is a C 0 submanifold of M which is also acronal.

Figure 6.13:

Now, let us introduce gravity. In an Eddington–Finkelstein plot, we can see that light-cones are
straight far away from the horizon and begin to tilt as they get close to the horizon, Figure 6.4. This
notion can be used to define the notion of a no-escape, a trapped, region without resorting to non-local
concepts like in the case of the event horizon. Indeed, time-like observers are forced to travel inside
their own light cones and then if in a gravitational collapse they become enough tilted, after some
time the region S 2 will be causally connected with a smaller and smaller region (see Figure 6.14).

73
Figure 6.14: Representation of how gravity can tilt the light cones and lead to the formation of a
trapped surface.

This is the idea behind a trapped region. Now we want to give a more rigorous definition. Re-
member that, for a null congruence, the expansion can be written as
1 A
✓ = lim . (6.53)
A!0 A
It is easy to see that the expansion of the ingoing light rays is always negative, ✓n < 0, even in
Minkowsky. In order to characterise the behaviour of a trapped region we can however notice that in
this case also the expansion of the outgoing null congruence, ✓l , is negative, ✓l < 0.
Hence, we can finally give the following definition: a trapped surface is a T 2 , closed4 , spacelike,
2-surface such that both ingoing and outgoing generators are converging; this means that the expansion
of both ingoing and outgoing congruences orthogonal to the 2-surface are negative: ✓n < 0, ✓l < 0.
The limit case between having a trapped surface and a non-trapped surface, with ✓l = 0, is called a
marginally trapped surface.

Figure 6.15: (a) Untrapped region; (b) future trapped region (from [52]).
4
A closed manifold is a manifold which is compact and boundary-less.

74
 A trapped region on a 3-D hypersurface ⌃ is the set of all the points in ⌃ through which a trapped
surface passes. In general, we could have disconnected trapped regions, therefore we can define a total
trapped region on a Cauchy hypersurface as the closure of the union of all the trapped region Ti on
⌃.

6.6.2 Apparent Horizon

Now we introduce a new concept of horizon: an apparent horizon A is the boundary F˙ of a
total trapped region on ⌃. Therefore, A is an outer marginally trapped surface with ✓out = 0. For
a stationary black hole, the apparent horizon coincides with the event horizon, A = H. During a
collapse, however, the apparent horizon is included inside the event horizon, A ⇢ H.
The idea is then to regard the apparent horizon as the instantaneous surface of a black hole. Notice
that, di↵erently from the concept of event horizon, where we need to know our J (I + ) in order to be
able to define it, the apparent horizon is a local notion, and as such in principle observable. Basically,
the event horizon is drawn by the light rays entering at 45 , bended by gravity, that then stick to the
same radius. The apparent horizon, however, has another shape, following the one of B 3 as it can be
seen in the following Figure 6.16.

Figure 6.16: Formation of an event horizon (red) and an apparent horizon (blue and red) during a
stellar collapse. The dashed lines represent the incoming light rays of the future horizon generators.
(2 space dimensions are suppressed).

In practice, the di↵erence between apparent and event horizon can be summarized as follows: the
apparent horizon is the boundary of a region through which at some time anything can’t escape; the
event horizon is the boundary of a region through which at any time anything can’t escape.

6.6.3 Trapping Horizon, Dynamical Horizon, Isolated Horizon

The concept of apparent horizon has recently become a little inadequate, since in this case the notion
of a trapped region relies on the choice of ⌃: of course one can calculate the expansion ✓l of T 2
knowing only the intrinsic 3-metric qab and the extrinsic curvature Kab of ⌃ and one does not need
to evolve these quantities away from ⌃ even locally. In this sense the notion is local to M. However,
this locality is achieved at the price of restricting T 2 to lie in ⌃ and to strongly depends on its choice.
However, a foliation of the spacetime in ⌃ surfaces can be made in di↵erent ways. So the existence of
a trapped region, not just its shape, can depend on the choice of foliation. This is not so desirable,
and in modern times there has been an evolution of the concept of horizon (see [53, 54] for an in depth
discussion).
Instead of giving a slicing, we can focus on the structure of the geodesics stemming from the
horizon. A future outer trapping horizon (FOTH) is a smooth 3-submanifold H of spacetime M
foliated by closed, 2-surfaces S such that:

1. the expansion of one future-directed null normal to the foliation la is zero, ✓l = 0;

75
 2. the expansion of the other future-directed null normal to the surface is negative, ✓n < 0;
3. the Lie derivative along n of the expansion of l is negative, Ln ✓l < 0.
The first condition is analogous to having marginally trapped surface; the second condition tells us
that we are considering a black hole instead of a white hole; the third condition implies the notion of
an outer surface.
Consider now the Raychaudhuri equation supplemented by the condition of hypersurface orthog-
onality of l
d✓ ✓2
= |!|⇢
| |2 + ⇢ 2
Rab la lb ; (6.54)
d 2
using the Einstein equation, Rab la lb = Tab la lb = 0, if = 0 (no GW) and initially ✓ = 0, then also
d✓
d = 0, which means that this remains null. However, if the black hole is accreting the horizon will
be growing, and therefore will be spacelike.

Let us see that an accreting horizon will be locally spacelike [53]. Denote by V a a vector field
which is tangential to H, everywhere orthogonal to the foliation by marginally trapped surfaces
and preserves this foliation. We can take it as a linear combination of the two null vectors
V = l f n for some constant f . Let us normalize l and n so that l · n = 1. This implies
that |V |2 = 2f and hence H is respectively, space-like, null or time-like, depending on whether
f is positive, zero or negative. Now, even during an accretion event V lies by definition always
on H which is also defined by the requirement that ✓l = 0. Hence LV ✓(l) = 0 which implies
Ll ✓(l) = f Ln ✓(l). But (6.54) tell us that in absence of exotic matter Ll ✓(l)  0. At the same, ✓l
becomes negative as one moves along n to the interior of the marginally trapped surfaces, whence
Ln ✓(l) < 0 also. So we conclude that f is non-negative, i.e. that the time-like case is ruled out.
Finally, if the flux of non-exotic energy across H is non-zero on some leafs of the foliation of H,
the right side of (6.54) cannot vanish identically on that leaf. Thus, f will be strictly positive
somewhere on each of these leafs, and the horizon will be space-like there.

The definition of a FOTH is not completely intrinsic to H: we need to know the flow of a congruence
along the other congruence; if we relax this third condition, retaining the rest of the FOTH definition,
we have a dynamical horizon which is a smooth, three-dimensional, space-like H of spacetime M.
Basically a dynamical horizon is a space-like 3-manifold which is foliated by closed, marginally trapped
2-surfaces and whose definition is now foliation independent, and intrinsic to H (while in FOTH the
condition 3 requires knowledge of the geometry in the n direction away from H) and they can be
present even in spacetime which do not admit an event horizon (e.g. also in spatially compact space-
times). Note however, that this is payed at the price to have a a less stringent definition. Indeed, in
time dependent situations, if the dominant energy condition holds and the space-time is asymptotically
predictable, it is easy to see that dynamical horizons will lie inside the event horizon. However, in the
interior of an expanding event horizon, there may be many dynamical horizons. Nonetheless, under
fairly general conditions one can associate with each evolving black hole an outer-most or canonical
dynamical horizon and it is then natural to focus just on this canonical one.
Finally we can also have an isolated horizon, which have the local structure of an event horizon.
More precisely an isolated horizon, is a 3D submanifold which satisfy
1. is topologically S 2 ⇥ R and null
2. along any null normal field tangent to the outgoing expansion rate ✓l vanishes (the horizon
area is constant in time)
3. All field equations holds on and the SET Tab on is such that T a b lb is a future directed
causal vector (it is not spacelike) for any null norm l a

4. the commutator [Ll ,Da ]V b = 0 for any V b tangent to where Da is the induced derivative
operator on .

76
A trapping horizon for which there are no flows, no gravitational waves, and so on, becomes an isolated
horizon.
Finally, note that, FOTHs, dynamical horizons, isolated horizons are intrinsic “spacetime notions”,
defined quasi-locally. They are not defined relatively to special space-like surfaces, as in the case of
apparent horizons, nor do they require to know the future null infinity of spacetime, as in the case of
event horizons. I.e., being quasi-local, they are not teleological.

6.7 Singularity Theorems

Singularity theorems represented a crucial step in our understanding of gravity. In the early sixties
of the XX century indeed it was far from clear if the singularity chacracterzing stationary black hole
solutions could not be avoided in realistic stellar collapse scenarios for example due to large deviation
from spherical or axis symmetry. The first application of topological methods to gravity by Penrose
in 1965 was indeed a breakthrough that cannot be overstated and which lead in the following to a
new way of proceeding at the base of the subsequent revolution that culminated in the seventies with
the discovery of black hole thermodynamics.
As we said, the first of the singularity theorems is due to Penrose [55, 56], in layman words it states
that a singularity is the unavoidable endpoint of any gravitational collapse if one does not have exotic
matter. To prove his theorem, Penrose starts by assuming classical General Relativity, that energy is
always positive, and that there is a star that starts collapsing, all relatively general assumptions. He
proves and deduces that if the star collapses too much, then the collapse becomes unstoppable, that
black holes are common and that they contain a singularity.
Another type of singularity, however, is represented by the one preceding a Big Bang scenario and
also in this case singularity theorems adapted to cosmology were developed [57, 58]; according to these
theorems, singularities in cosmological settings arise at the start of the Big Bang scenario, only by
assuming that the energy is always positive and that the Universe is expanding right now. From this
it can be deduced that there is a singularity in our past and that time had actually a beginning.
Let us focus on the singularity theorem by Penrose. During a collapse, is a singularity inevitable?
If the collapse is not isotropic, can a singularity be avoided? All this problems are tackled and solved
by the singularity theorems. Penrose used topological tools in General Relativity.
Recapping, our hypothesis will include the fact that we use classical General Relativity, that we
assume appropriate energy conditions, and that we have stellar collapse, which means that we will
eventually have a trapped region T 2 . The thesis is that a singularity forms.

Penrose’s Singularity theorem. Assume the following:

1. we have a trapped surface, and there is a maximum expansion associated to this trapped surface,
✓max = ✓0 < 0;

2. the NEC condition holds: Tab k a k b 0;

3. General Relativity holds: in this way, we can write

✓ ◆
a b 1
Rab k k = 8⇡ Tab T gab k a k b = 8⇡Tab k a k b ;
2

4. the manifold M is a globally hyperbolic spacetime, therefore M = R ⇥ ⌃3 , and therefore no

Cauchy horizon forms.

Then, a singularity must form.

77
Figure 6.17: A schematic depiction of a spherically symmetric collapse with the essential elements of
the singularity theorem demonstration.

Proof: we have F 4 = I + (T 2 ), B 3 = @F 4 = I˙+ (T 2 ) 5 . In a stellar, spherically symmetric, collapse

there will be a stage in which all the congruences will converge in a point ⌦.
Consider the Raychaudhuri equation

d✓ ✓2
= Rab k a k b ; (6.55)
d 2
using the NEC, we can then write
d✓ ✓2
 , (6.56)
d 2
and in the best case scenario, this becomes

d✓ ✓2
= . (6.57)
d 2
By solving this di↵erential equation, we simply get

✓ 1
= ✓0 1 + , (6.58)
2
and evidently ✓ diverges to infinity as approaches 2✓0 1 . Therefore, in a finite , the expansion
reaches 1, and thus all congruences converge to the point ⌦ (this is strictly true for a spherically
symmetric collapse, in general there could be a extended region where the focussing happens). The
question now becomes: is ⌦ part of M?
Assume that ⌦ 2 M, which means that ⌦ is not a singularity. Then B 3 is compact and boundary-
less, being generated by a compact system of closed segments, and therefore is a closed submanifold
of M.6
5
Note, we adopt here [55] notation but one can alternatively find in textbooks F 4 = J + (T 2 ), B 3 = @F 4 = J˙+ (T 2 ).
6
By a compact manifold we mean that it is topologically compact. A sphere, torous or a disk are topologically
compact. A closed manifold is a manifold which is compact but also boundariless. A disk is compact, and it is a closed
subset of the plane, but it is not a closed manifold as it has a boundary in the Euclidean plane.

78
 Now, consider a timelike vector field tµ . It can be shown that such a congruence of timelike curves
must exists in a spacetime with a Cauchy hypersurface. Since one and only one curve of the congruence
passes through each point of space, and since a timelike curve cannot intersect a null surface such as
B 3 more than once (i.e. B 3 is acronal), each integral curve of tµ will pinch B 3 only once and the same
will be true for ⌃3 (which is also acronal).
So tµ induces a continuous one to one map, a homeomorphism, from points on B 3 to points on ⌃3 .
Let us call S 3 , the image of B 3 generated via this map on ⌃3 (notice that we are assuming that there
are no causality violations, since we assumed M = R ⇥ ⌃3 ). In particular, given that we are dealing
with a homeomorphism, this image of B 3 over ⌃3 must be compact.
However, ⌃3 is non-compact, and since S 3 is compact, we have that S 3 ⇢ ⌃3 , and therefore S 3
must have a boundary in ⌃3 . But if this is the case, S 3 is not boundary-less, and even if it is compact,
it is not closed.
But, S 3 is homeomorphic to B 3 , which is closed by assumption, so S 3 cannot have a boundary.
Therefore we have reached a contradiction, which means that ⌦ 2 / M, and thus ⌦ is a singularity. ⌅
The above theorem was later extended to singularity theorems in cosmology (see e.g. [58]).

Wald provides us with a similar and alternative proof of this theorem.

In particular, assume that B 3 is compact and boundary-less, because of geodesic completeness;
S 3 is an image with a one-to-one homeomorphic map of B 3 on ⌃3 .
B 3 is compact, hence S 3 is compact; but it is also closed on ⌃3 , since B 3 is compact and
boundary-less, therefore S 3 must be too. S 3 must be a closed subset of ⌃3 . But it is also true
that on each point of B 3 there is an open neighbourhood homeomorphic to an open ball. But this
must be true also for S 3 , because the two are homeomorphic; but this implies that S 3 , as a subset
of ⌃3 , is also open. The only way in which S 3 can be both closed and open on ⌃3 is that S 3 is
the same set of ⌃3 (Wald assumes that the spacetime is connected, and therefore only M and ?
are both open and closed). Therefore the image of B 3 must be the whole Cauchy hypersurface.
However S 3 is also supposed to be compact, but ⌃3 is not compact by definition, so it is
not possible that S 3 coincides with ⌃3 . Then, S 3 ⇢ ⌃3 , but then B 3 cannot be compact and
boundary-less.

Penrose’s singularity theorem can be seen as morally similar to Gödel incompleteness theorems in
mathematics, in the sense that it tells us that General Relativity is an incomplete theory. This
does not necessarily mean that the theory must be quantized, but just that we have a theory, General
Relativity, that predicts its own demise, since singularities form behind the event horizon of black
holes, and we have very strong, albeit non conclusive, evidences of the existence of General Relativity
black holes, with the detection of gravitational waves emitted by sources compatible with black hole
binaries (e.g. [59]), and with the image of the black hole M87*, published in April 2019 by the Event
Horizon Telescope collaboration, [60].
Note that a more general theorem one was proposed by Hawking and Penrose later on in 1970.
According to this theorem, the spacetime M necessarily contains incomplete timelike or null geodesics
which cannot be continued, provided the following conditions are satisfied:

1. The spacetime contains no Closed Timelike Curves (which implies that the spacetime has no
problems causality-wise, and therefore there are no Cauchy Horizons)

2. For any timelike vector uµ , Rµ⌫ uµ u⌫ 0 (meaning that the Strong Energy Conditions applies)

3. For each timelike or null geodesics with tangential vector uµ there is a point in which u[↵ R ] [✏ u⇢] u u 6=
0 (spacetime is of general type with special symmetries )

4. A trapped surface exists.

This theorem applies also when a trapped surface is generated e.g. in a closed universes, where there
are no non-compact ⌃3 , so that the Penrose theorem can be straightforwardly be applied.

79
 Can the singularity be resolved by quantum gravity? Of course yes, this is after all what we do
expect from a quantum theory of gravitation. It is interesting to not that if one also requires this
resolution to respect the strong cosmic censorship conjecture a finite number of possibilities are left
(see e.g. [61] and reference therein).

80
 7
From Black Hole Mechanics to
Black Hole Thermodynamics

Birkho↵’s theorem states that any spherically symmetric vacuum solution is static, which e↵ectively
implies that it must be Schwarzschild (independently if it is describing the geometry of a black hole or
that of the exterior of a spherical object). A generalisation of this theorem to the Einstein–Maxwell
system shows that the only spherically symmetric solution is Reissner–Nordström.
However, if the geometry is not spherically symmetric, we have no analogue of Birkho↵’s theorem;
for example, around the Earth we do not know a priori what is the metric. In principle, we can still
write the metric in a weak field approximation – where the metric contains terms like (1 2 ) – and
essentially expand the gravitational potential in spherical harmonics
1 X
X 1 X
l m
Djl
= Ylm (✓, '). (7.1)
rj
j=1 l=0 m= l

For a general potential, we shall have to determine all these coefficients; however, black holes are
simple as they obey the No Hair Theorem: if a black hole forms after a sufficient long time all
higher multipoles are radiated away, only the lower multipole coefficients will be relevant.
In particular a theorem, due to Carter–Robinson (plus Hawking and Wald), states: If (M, g) is
an asymptotically-flat, stationary, vacuum spacetime that is non-singular on and outside an event
horizon, then (M, g) is a member of the two-parameter family of axisymmetric black hole solution 1 .
The parameters being the mass M an the angular momentum J
Consider gravitational waves: the conservation of the SET implies the conservation of mass, linear
and angular momentum so the emission happens from the quadrupole moment up. Basically, this tells
us that the mass and the angular momentum of the black hole cannot be the source of gravitational
waves and are therefore conserved once the black hole is stationary. During the formation of a black
hole (in a general collapse or in a merging event), mass and angular momentum are partially shed
away (not just by matter loss but also by gravitational waves that carry away both energy and angular
momentum), but once the black hole forms they becomes the conserved charges characterising its
structure.
What about electric and magnetic fields? The electric field is conserved, since the charge Q is
conserved. According to the membrane paradigm [62], the charge sticks on the horizon which from
outside can be seems as a thin membrane made up by the infinite overlapping layers of matter (a
stretched horizon) which the outside observer can never see crossing the horizon. This matter can
carry charge which then becomes for the ouside observer a black hole charge. The magnetic field is
1
An asymptotically flat spacetime is axisymmetric if there exists a Killing vector field (the “axial” Killing vector
field) that is spacelike near 1 and for which all orbits are closed.

81
Figure 7.1: Artist impression of the gravitational waves emission by a black hole merger. Credits
(Nobel prize pages)

another story: the magnetic monopole P (if it exists) can be attached to the stretched horizon of
the black hole as the electric charge, but the magnetic field B itself is di↵erent; it cannot exist in
a black hole surrounded by vacuum, as nothing would prevent fields lines from being radiated away
as closed loops in the collapse. Hoverer, for a black hole surrounded by a disk of matter, a torus
(what sometimes it unceremoniously called a “dirty black hole”) field lines can attach on the stretched
horizon on one side and the matter on the other. Something very relevant for astrophysical phenomena
like the Blandford–Znajek process.

Figure 7.2: Artist impression of the magnetic field lines from the stretched horizon to the accretion
disk. Credits Dana Berry (NASA)

7.1 Rotating Black Holes

The generic metric describing a rotating, charged black hole in a vacuum is the Kerr–Newman
metric and it was found only in 1965, just two years after Roy Kerr had provided the metric for a
rotating black hole which is the same as below with e = 0:
✓ ◆ ✓ 2 ◆
2 a2 sin2 ✓ 2 2 r + a2
ds = dt 2a sin ✓ dt d'+
⌃ ⌃
 2 (7.2)
(r + a2 )2 a2 sin2 ✓ 2 2 ⌃ 2 2
+ sin ✓ d' + dr + ⌃ d✓ ,
⌃
where
⌃ ⌘ r2 + a2 cos2 ✓,
⌘ r2 2M r + a2 + e2 ,
J
e2 ⌘ Q2 + P 2 and a⌘ ,
M

82
Note that in astrophysics it is often used a dimensionless spin parameter ā = a/M = J/M 2 (which is
often presented simply as a so to confuse relativists!).
The above solution is an electrovac (vacuum apart from electromagnetic field) solution of the
Einstein equations. The associated electromagnetic one form potential non-vanishing components are

Qr P a cos ✓ Qar sin2 ✓ + P (r2 + a2 ) cos ✓

At = and A' =
⌃ ⌃
This is an axisymmetric stationary but not static solution (e.g. you can check it is not symmetric
under time reversal due to the dt d term), it admits a globally defined, non-vanishing, timelike Killing
vector outside of the horizon; it is also asymptotically flat, since at infinity this reduces to Minkowski.
The coordinate we used are called Boyer–Lindquist coordinates.
Setting a = Q = P = 0, we recover the Schwarzschild solution; setting a = 0 we recover the
Reissner–Nordström solution; setting P = Q = 0 we recover the standard Kerr solution for a rotating
black hole. All of them are Ricci flat solutions. It is sometimes argued that Kerr–Newman black
holes cannot exist in nature: an electrically charged black hole can selectively attract charge of the
opposite sign, and this can neutralize the electric charge of the black hole. Therefore, Kerr–Newman
black holes should not exist (even though this is still disputed) and astrophysical black holes should
be Kerr ones. From now one we shall focus on this solution then, albeit we shall provide sometimes
the generalisation of important results to the full Kerr-Newman case.

7.1.1 A ring Singularity

Let us try to see if there are singularities of the metric.

• In ✓ = 0 (i.e. on the axis of rotation of the solution) we have a singularity but this is obviously
the standard singularity at the poles of our polar coordinate set: changing coordinates we can
easily eliminate it.

• Another interesting point is at ⌃ = 0. In this case, there are two possible solutions; at r = 0
and ✓ = ⇡2 we could have a problem. What is the shape of this singularity? It looks like a point,
but only in the equator

Let us try to analyse this in another set of coordinates, the Kerr–Schild coordinates, in which
the metric acquires the form gµ⌫ = ⌘µ⌫ + f (r, z)kµ k⌫ , where kµ is a null vector. The coordinate
transformation is defined as
8 ⇥ R ⇤
>
<x + iy = (r + ia) sin ✓ exp i
> d' + a dr
z = r cos ✓ (7.3)
>
> R⇣ 2 +a2
⌘
:t̂ = dt + r
dr r

with the radius given implicitly by the equation

r4 x2 + y 2 + z 2 a2 r 2 a2 z 2 = 0.

In these coordinates the Kerr metric has the form


2 2 2 2 2 2M r3 r a z dz 2
ds = dt̂ + dx + dy + dz + 4 2 2
dt̂ + 2 (x dx + y dy) (x dy y dx) + ,
r +a z r + a2 r + a2
2 r
(7.4)
With this line element one can recognize that
✓ ◆
2M r3 rx + ay ry ax z
f (r, z) = 4 and (kt̂ , kx , ky , kz ) = 1, 2 , , . (7.5)
r + a2 z 2 r + a2 r 2 + a2 r

83
 Let us start noticing that in the limit M ! 0 we recover Minkowski as expected (this is true in
any coordinate system but in the Kerr–Schild one is straightforward to recognise the left over metric).
Moreover, in this coordinate systems it is evident that surfaces of constant r are cofocal ellipsoids,
indeed from the coordinates definitions it is easy to see that

x2 + y 2 = (r2 + a2 ) sin2 ✓ and z 2 = r2 cos2 ✓ (7.6)

so that ✓ ◆
z2 x2 + y 2 z 2
x2 + y 2 = (r2 + a2 ) 1 ) + 2 = 1, (7.7)
r2 r 2 + a2 r
We can easily see from (7.6) that as r ! 0 these surfaces degenerate to a disk: at r = 0 we have
z = 0 and x2 + y 2  a2 (as sin2 ✓  1). At the singular point r = 0, ✓ = ⇡2 , we get

x2 + y 2 = a2 , (7.8)

which is the equation of a circle: the singular point is actually resolved in these coordinates into a ring
on the equatorial plane. An investigation of the causal structure of Kerr shows that this is indeed a
timelike singularity, we hence expect to find an associated Cauchy horizon.

7.1.2 Horizons
Consider now the singularity at = 0: this implies (in the case of astrophysical black holes, e = 0)

r2 2M r + a2 = 0, (7.9)

which admits solutions p

r± = M ± M2 a2 . (7.10)
At a = 0, we get r+ = 2M , so we recover the event horizon of a Schwarzschild black hole. However,
di↵erently from the Schwarzschild case, here we have in general two horizons. In particular, we have
some special cases.
If a2 > M 2 , we would have no horizon, since r± would acquire an imaginary part; however, we
would still have the singularity at ⌃ = 0; this means we would have a naked singularity. According to
the Cosmic Censorship Conjecture, naked singularities should not exist in the Universe: any singularity
should be protected by an event horizon. In case a naked singularity existed, we would have strong
e↵ects of quantum gravity exposed to us, and even worse, because we would have serious problems
with causality.
Another possibility is given by a2 = M 2 : then we would have r+ = r = M , and this is called
an extremal black hole, the maximally rotating black hole. They are also considered unphysical: in
astrophysics there is the so-called Thorne limit, a . 0.998M 2 , that is close to the extremal limit.
Extremal black holes have also zero surface gravity and, therefore, have a zero Hawking temperature;
black hole thermodynamics, we shall see later on, implies via its third law that it should be impossible
to turn a non-extremal black hole into an extremal one in a finite number of thermodynamical steps.
Consider, then, the standard Kerr black hole case a2 < M 2 . We now want to see if the singularity
at r± is really a coordinate artefact. We can introduce so-called Kerr coordinates, horizon penetrating
coordinates, similar to ingoing Eddington–Finkelstein coordinates, defined w.r.t. BL coordinates as
(see Se.g. Box 33.2 of [41]. See also [63] for an extensive pedagogical introduction to the Kerr
geometry). ( 2 2
dv = dt + r a dr
(7.11)
d'˜ = d' + a dr,
2
Thorne in 1974 recognized that thin discs radiate and as a result, the black hole spin would be upper bounded
because the black hole would preferentially swallow negative angular momentum photons emitted by the accretion flow.
A number of studies since then have suggested that this limit may not hold (or that the limiting spin may be higher) if
accretion takes place via a geometrically thick flow.

84
and with these we can rewrite the Kerr metric as
( a2 sin2 ✓) 2 r 2 + a2
ds2 = dv + 2 dv dr 2✓ sin2 ✓ dv d'+
˜
⌃ ⌃ (7.12)
2 (r2 + a2 )2 a2 sin2 ✓ 2 2 2
2a sin ✓ d'˜ dv + sin ✓ d'˜ + ⌃ d✓ ,
⌃
which is perfectly fine at = 0. Note also that for a ! 0 gives the Schwarzschild metric in ingoing
(v, r, ✓, ')
˜ EF coordinates
✓ ◆
2M
ds2 = 1 dv 2 + 2 dv dr + r2 (d✓2 + sin2 ✓ d'˜2 ). (7.13)
r
27

+
r± I+
r+ I+
0 0

r+ I -

- r± I-
r-
r=0

r=0

r-
+
r± I+
r+ I+
0 0

r+ I -

- r± I-

FIG. 2. Carter-Penrose diagrams for the equatorial plane of near-extreme Kerr [left], extreme Kerr [middle] and NHEK [right].
Figure 7.3: Carter–Penrose diagrams
(The diagrams depend onof✓ because
non-extremal (left)
these geometries are and symmetric
not spherically extremal (right)
and, away from the Kerr
equatorialblack
plane, it hole (from
±
possible to continue past the singularity.) In the diagrams for (near-)extreme Kerr, I denotes future/past null infinity
[64]). For further detailsis(rabout the conformal 0 structure of Kerr, see also [58, p. 165].
= +1 with t = ±1), is spacelike infinity (r = +1), r is the outer/inner horizon, and the singularity lies at r = 0. The
±

thin lines correspond to hypersurfaces of constant r, which are timelike outside the black hole [red] and spacelike inside [gray].

A. Geodesics in global NHEK

7.1.3 Killing vectors
Recall from Sec. III E that in global coordinates, the NHEK line element is
 @
Consider now the symmetries of the metric: ds̃ = 2M
this admits dy
2
1 + y d⌧ +
two+ self-evident
2 2Killing vectors,
d✓ + ⇤ (d' + y d⌧ ) , 2
(141)
⇠⌘ 2
2 2 2
@t and
1+y 2
@
⌘ @' , since the metricandistheindependent from t
generators of SL(2, R) ⇥ U(1) are
and '. The norm of ⇠ at r = r + is given by
" #
y p 1
L0 = i @ , L± = e±i i 12+✓y @ ± p
a@2 sin
±p 2
y @ , W0 = @ . (142)
|⇠|2r=r+ = g00 |r+ = 1 +2y
2 2
6=1 0;
+y
(7.14)
r+ +
The Casimir of SL(2, R) is the (manifestly reducible) a2 cosKilling
symmetric
2 ✓ tensor

µ⌫ µ ⌫ 1 µ ⌫ µ ⌫
C = L L + L L +L L .
however, the surfaces at both r = r± are null surfaces, but
2 since ⇠ is not a null Killing (143)
0 0 +
vector field at +

µ⌫
the horizon, ⇠ is not a generator of the horizons. Moreover, we see that at the horizon ⇠ is spacelike,
It is related to K̃ , the NHEK limit (99) of the irreducible Killing tensor on Kerr (2), by
µ⌫ µ⌫ µ ⌫ 2 µ⌫
while it is timelike at infinity (Kerr is asymptotically K̃ = C Minkowski),
+ W W + M g̃ . which means that there 0 0(144) must be a
µ
region where ⇠ becomes null. Finally,
The motion sinceof massisµ and
of a free particle spacelike
four-momentum spacelike
P is describedatby rthe+ ,geodesic
therefore
equation, neither can be
µ ⌫ µ⌫ 2
a generator of the horizon. P r̃ P = 0, g̃ P P = µµ . (145)
µ ⌫

We can nonetheless introduce a Killing vector ⌅± which is a linear combination of ⇠ and which
is null on r+ :
a @ @
⌅± ⌘ ⇠ + 2 2
= + ⌦± H . (7.15)
r± + a @t @'
One can easily check that |⌅± |2r± = 0; also, it can be proven that the surface gravity of the Kerr black
hole for both the horizons is given by
r± r⌥
± = 2 + a2 ) . (7.16)
2(r±

85
 a
The ⌦H defined above, ⌦H ⌘ 2 +a2 ,
r±
is the angular velocity of the horizon. In fact, we have

@' @t @' @t
⌅µ rµ (' ⌦H t) = (⇠ µ @µ + ⌦H µ
@µ ) (' ⌦H t) = ⌦H + ⌦H ⌦2H = 0, (7.17)
@t @t @' @'
which means that ' ⌦H t = const along the flow defined by ⌅µ and ' = ⌦H t are orbits of ⌅. We
deduce that particles following these orbits rotate with angular velocity ⌦H w.r.t. static observers at
infinity (whose worldlines are instead orbits of ⇠). Since null generators of the EH follows the orbits
of ⌅+ then we can say that the EH rotates w.r.t. to asymptotic observers at infinity with angular
velocity ⌦H

7.2 Ergosphere and Penrose Process

Let us now find the region where |⇠|2 = 0; this happens where
✓ ◆
2M r
1 =0 =) r2 + a2 cos2 ✓ 2M r = 0, (7.18)
r2 + a2 cos2 ✓
p
which has solution rE = M ± M 2 a2 cos2 ✓; since we are interested into values of r outside r+ , we
consider only the solution with the plus sign, and therefore we find that the radius of the region in
which ⇠ becomes spacelike, called ergosphere, is
p
rE = M + M 2 a2 cos2 ✓. (7.19)

Figure 7.4: Representation of the structure of a Kerr black hole with a = 0.8M in Kerr–Schild
coordinates at constant time; the left portion represents the slice at z = 0 (face-on, top view), while
the right portion represents the slice at y = 0 (edge-on, side view); rE is the border of the outer
ergoregion, r+ is the outer horizon, r is the inner horizon. In orange, the ring singularity of Kerr.
Notice that, in the side view, both r+ and r are ellipses in Kerr–Schild coordinates, while rE is not.

The fact that ⇠ becomes spacelike means that, inside the ergosphere, one cannot remain still:
@ @
usually, we follow trajectories along @t , but we can do that only if @t is timelike. Therefore, inside the
ergosphere, we would be forced to rotate with the black hole, we would be dragged by the spacetime.
Notwithstanding the strange nature of the surface at rE , this is not an horizon, since it is always
possible to spiral out of the ergosphere.
If we have a Killing vector and we contract it with the four-momentum, we find a constant along
geodesics, pµ µ = const. If = ⇠, this constant is the energy
E= pµ ⇠ µ ; (7.20)

86
Figure 7.5: Location of the horizons, ergospheres and the ring singularity of the Kerr spacetime in
Cartesian Kerr-Schild coordinates. From wikipedia.org

if = , the constant conserved along the geodesic motion is the angular momentum

L = pµ µ. (7.21)

Now, since ⇠ is spacelike inside the ergosphere, it is possible to have negative energy states in this
region, E < 0. Now we can realize that, since negative energy states exist outside the horizon, we can
sort of “mine out” energy from the black hole. This is called Penrose process.
Consider a material particle with energy E0 approaching a Kerr black hole. Once it goes inside
the ergoshpere, suppose that this particle splits into two particles, one of energy E1 escaping the black
hole and one of energy E2 going to the black hole. Is it possible to have a situation in which E1 > E0 ?
We have
E2 = p2 · ⇠|r<rE , E1 = p1 · ⇠ > 0, E0 > 0. (7.22)
The fact that ⇠ becomes spacelike inside the ergoregion implies that for r < rE we can choose orbits
where E2 < 0; in this case we have that, once the particle goes inside the black hole, we have
EBH ! EBH |E2 |, and therefore E1 > E0 . How is this possible? Who “pays the bill”?
Consider the Killing vector generating the horizon contracted with the momentum of the infalling
particle at H+ . Being ⌅ timelike outside the horizon one has

p2 · ⌅ 0
(7.23)
= p·⇠ ⌦H p · = E⇠ ⌦H L ,
E
and therefore L  ⌦H⇠ . Hence, if and when the Penrose process is realized, we have E⇠ < 0, which
in turn implies L < 0, and therefore the infalling particle carries both negative energy and negative
angular momentum.
Now, suppose that the infalling particle makes the black hole change its mass from M to M + M
and its angular momentum from J to J + J. We have M = E2 < 0 and J = L2 < 0. Also,
J  ⌦MH
, and if we write explicitly the angular momentum of the black hole

J
⌦H = p , (7.24)
2M 3 + 2M M 4 J2
we immediately get
⇣ p ⌘ M
J  2M M 2 + M 4 . J2 (7.25)
J
This must be true for each step of the Penrose process, and can be rearranged as
⇣ p ⌘
M2 + M4 J2 0; (7.26)

87
we can then define the irreducible mass of a rotating black hole as

2 1⇣ 2 p 4 ⌘
Mirr ⌘ M + M J2 . (7.27)
2
We immediately see that if J ! 0, the irreducible mass reduces to the mass M of the black hole; also,
by inverting the relation, we can see that

J2
M 2 = Mirr
2
+ 2
Mirr . (7.28)
4Mirr
Therefore, the Penrose process continues as long as there is angular momentum, and when this is no
longer the case, the black hole mass equates the irreducible mass (from here the name “irreducible”).
The maximum energy we can extract using the Penrose process is therefore Emax = M0 Mirr ;
therefore, we can determine the efficiency of the Penrose process:
p !1/2
M0 Mirr Mirr M02 + M04 J2 a=1 1
=1 =1 = 1 p ' 29%. (7.29)
M0 M0 2M02 2

Let us now try to determine the area of a black hole; this is given by
Z
p 2
ABH = g✓✓ g'' d✓ d' = 4⇡ r+ + a2 = 16⇡Mirr 2
; (7.30)
r+

therefore, since we know that Mirr 0, we also know that ABH 0. Hence, the Penrose process
can go on as long as the black hole area keeps on increasing.
This is a special case of a theorem by Hawking – known as the Area Law – that states that, if
matter satisfies the WEC, in any physical process, ABH 0. And this is in analogy with thermody-
namics: energy extraction, in thermodynamics, is related to entropy which, in any physical process,
cannot decrease, while here it is the area of a black hole that cannot decrease. We will delve deeper
into this concept in later on.

7.2.1 Superradiance
Is there an analogue of the Penrose process in a field-theoretical context? The answer is still yes, and
in this case it is called superradiance [65]. We send a wave toward a black hole, and this wave is
partly transmitted with a coefficient I and partly reflected with a coefficient R. What happens is that
|R|2 > |I|2 and the wave gets amplified. If we have a scalar field wave of equation = 0 cos (!t m'),
superradiance is realised when
0 < ! < m⌦H (+e H ) . (7.31)
where in parenthesis we have added also the contribution in the case a Kerr–Neuman black hole The
average power loss is, then
1 2
P = ABH ! (! m⌦H ) . (7.32)
2 0
Note that, like the Penrose process, superradiance has to be there in order not to decrease the area
of a black hole in the above described scattering process. Indeed, the existence of superradiance can
also be derived from Hawking’s Area Theorem (see e.g. page 329 [1]).
Another thing to keep in mind is that fermions do not superradiate [66]: this is a consequence of
the fermions SET violating the WEC which is a basic assumption necessary to demonstrate the Area
theorem. However, superradiance applies to scalar fields, electromagnetic field and tensor fields, like
gravitational waves.
A final remarks. Is there an ergoregion in a Schwarzschild geometry? Of course there is, but it
is inside the event horizon. In fact, we already saw that, inside the event horizon of a Schwarzschild
black hole, the Killing vector associated to time translations becomes spacelike. Also, notice that in

88
the limit for J ! 0, the radius of the ergosphere rE approaches the radius of the event horizon of
a Schwarzschild black hole. Therefore, the negative energy states are also present in Schwarzschild
geometry, however they cannot be extracted from behind the horizon, at least via classical processes.
This is another hint suggesting to look into quantum e↵ects in curved spacetime.

89
7.3 Black Hole Thermodynamics
In 1970, Wheeler posed a peculiar question: what happens if we throw something into a black hole?
Is the second law of thermodynamics violated? How do the laws of thermodynamics work in presence
of a black hole?
In 1971-72, we have the realization of the Penrose process, Christodoulou and Ruffini introduce
the concept of irreducible mass and Hawking formulates the Area Law.
In 1973, Bekenstein was the first to make the conjecture that there was a connection between
entropy of a black hole and area of a black hole, as well as between the surface gravity of the black
hole and the temperature of the black hole.
All this led to the formulation of a generalized second law, which basically states that
✓ ◆
ABH
Sout + ⌘ 2 0, (7.33)
`P

with ⌘ a constant.
After Bekenstein, in a seminal paper Bardeen, Carter and Hawking (1973) [67] conjectured that the
four laws of black hole mechanics had a direct link with the laws of thermodynamics. 3 Nonetheless, up
to this point, there was no clear notion of a temperature for black holes, albeit the laws were strongly
suggesting that the surface gravity  had to be seen as the analogue of the black hole temperature T .
In 1974-75, Hawking finally found that by cleverly applying quantum field theory on black holes
spacetimes 4 indeed they would radiate with a temperature determined by their surface gravity

TH = , (7.34)
2⇡
and from there everything was consistent. However, let us stress that the four laws are classical laws,
while Hawking radiation is a quantum e↵ect. It looks like, someway, somehow, general relativity knows
about quantum field theory...

7.3.1 Null hypersurfaces, Surface integrals and other tools

Black hole thermodynamics is also called horizon thermodynamics. Horizons of stationary black holes
are 3-d null hypersurfaces. As such, let us elaborate better about how to describe their structure more
precisely.
First of all, any hypersurface can be defined or by requiring the constancy of some function of
the coordinates over it, or equivalently by describing it by some parametric equations of the form
x↵ = x↵ (y a ) where y a (with a = 1, 2, 3 = v, ✓, for the specific null case we are considering). 5
Noticeably, we can easily construct in this case a set of tangent vectors to the hypersurface. Indeed,
for the vectors
@x↵
e↵a = (7.35)
@y a
one has e↵a k↵ = 0 where k↵ is the orthogonal vector to the hypersurface.
Also, the metric intrinsic to a hypersurface ⌃ can be obtained by restricting the line element to
displacements along the hypersurface.
✓ ↵ ◆✓ ◆
@x @x
ds2⌃ = g↵ dx↵ dx = g↵ dy ↵
dy = hab dy a dy b (7.36)
@y a @y b
3
Note, however, that the third law was rigorously proved by Werner Israel only in 1986 [68].
4
Let us stress that, as we shall see soon, in order to recover Hawking radiation that there is no need for quantum
gravity, only QFT in curved spacetime
5
For example, a two-sphere in a 3d flat space can be described either by the condition (x, y, z) : x2 + y 2 + z 2 = R2
where R is the sphere radius, or we can define it by providing the relation between the x↵ coordinates and those on the
surface, i.e. x = R sin ✓ cos , y = R sin ✓ sin , z = R cos ✓.

90
where hab ⌘ g↵ e↵a eb is of course the induced metric (first fundamental form) of ⌃.
Let us now specialise to a null hypersurface. When dealing with null hypersurfaces it is often
convenient to introduce a system of coordinates adapted to their generators: let us then introduce as
one of the coordinates on the horizon a (not necessarily affine) parameter v along a generator. For
each v = const, we have a 2-d spatial section of the horizon S 2 . We shall then choose the other two
coordinates as ✓A , with A = 2, 3, i.e. the two spatial coordinates on S 2 . These are constant on each
generator of H+ and span the two dimensional surface orthogonal to it. Basically, moving along v
corresponds to move along a generator of H+ while moving along a ✓A corresponds to move from one
generator to another. Let us now consider the vectors tangent to H+

Figure 7.6: Pictorial representation of the null congruence generating H+ and a section.

✓ ◆ ✓ ◆
↵ @x↵ @x↵
⌅ ⌘ , `↵A ⌘ , (7.37)
@v ✓A @✓A v

and indeed ⌅↵ `↵A = 0 and furthermore L⌅ `↵A = 0 in stationary black hole solutions. In a Kerr
black hole, we already know that ⌅↵ = ⇠ ↵ + ⌦H ↵ . Note that in this case e↵1 = ⌅↵ and hence
h11 = g↵ e↵1 e1 = g↵ ⌅↵ ⌅ = 0. So that in this case the induced metric reduces to

ds2⌃ = AB d✓A d✓B , (7.38)

where AB ⌘ g↵ `↵A `B .
However, there are two bundles of null geodesics orthogonal to each horizon slice S 2 . If ⌅µ is the
tangent vector to the first null congruence (the one associated with H+ ), we can define an auxiliary
null vector N ⌫ and choose the normalization in such a way that ⌅µ Nµ = 1. With this base of vectors
we can in general decompose the inverse metric in 4 dimensions in a way such that 6 .
⇣ ⌘
g↵ = ⌅↵ N + N ↵ ⌅ + AB `↵A `B . (7.39)

Remember, that the conditions N ⌫ N⌫ = 0 and ⌅µ Nµ = 1 are not sufficient to uniquely determine N ⌫
and hence the above decomposition. However, as in the case of the Raychaudhury equation physical
quantities will come out to be independent of it.

7.3.2 Surface element

Normally, when dealing with the integral of a total divergence we apply the Stokes theorem
Z I p I
p
d 4 x g r↵ K↵ = d3 x hn↵ K↵ = d⌃↵ K↵ . (7.40)

p
where d⌃↵ ⌘ d3 x h n↵ = n↵ d⌃ is called the directional/vectorial surface element that points
in the direction of increasing values of the function whose constancy determines each hypersurface.
However, in the case of a null hypersurface, this is problematic, because h = 0.
Hence, in order to generalize the above discussion to null surfaces we have to revert to a more
fundamental description of the directional (hyper)-surface element: indeed, using the Levi–Civita
6
For a non-null hypersurface the analogue decomposition would be g ↵ = n↵ n + AB ↵
`A `B

91
tensor ✏µ↵ and the previously introduced basis of tangent vectors (7.35) one can write the surface
element as
d⌃µ = ✏µ↵ e↵1 e2 e3 d3 y; (7.41)
It can be shown (see e.g. [3] pages 64-65) that this is the correct volume element independently from
the nature of the hypersurface. In particular, for timelike or spacelike hypersurfaces it is equivalent to
d⌃µ = nµ d⌃, where again = nµ nµ = ±1 if the hypersurface is timelike or spacelike respectively.
Now, remember that in our, null, case, we have that e↵1 = ⌅↵ , d3 y = dv d2 ✓, and therefore the
directional line element becomes
d⌃µ = ⌅⌫ dSµ⌫ dv, (7.42)
where dSµ⌫ = ✏µ⌫ e2 e3 d2 ✓ is the bi-directional surface element for the two-surface.
Using the fact that ✏µ⌫ e2 e3 is orthogonal to any e↵A and that it is antisymmetric in µ, ⌫, one can
p 2
show [3] that dSµ⌫ can be rewritten as the 2-d volume element dS ⌘ d ✓ times the antisymmetrized
product of the two normals to this 2-surface, i.e. ⌅ and N . Hence
p
dSµ⌫ = 2 ⌅[µ N⌫] d2 ✓ . (7.43)

Finally, combining (7.42) with (7.43), one then gets that the directional 3d surface element for a
null hypersurface is (use ⌅⌫ ⌅⌫ = 0 and ⌅⌫ N⌫ = 1)
p
d⌃µ = ⌅µ d2 ✓ dv (7.44)

Let us now apply these concepts to a proper definition of mass and angular momentum in stationary
black hole spacetimes.

7.3.3 Komar Mass and the Smarr formula

In general, defining an energy in General Relativity is not simple; however, we have seen that the
presence of Killing vectors in a spacetime is associated to conserved quantities.
Consider then the vector Qa defined as Qa ⌘ ⇠b Rab where ⇠ is some Killing vector. We can see
that ✓ ◆
⇠ ⇠⇠ 1 a Bianchi 1 b
a
ra Q = ⇠ R ⇠ra ⇠b + ⇠ ra R b =⇠ ra G b + g b R
ab b a b a
= ⇠ @b R = 0, (7.45)
2 2
where the first term cancels due to the Killing equation, we made use of the contracted Bianchi identity
ra Ga b = 0, and the last equality is given by the fact that we are basically performing a Lie derivative
along the isometry defined by ⇠ of a scalar built from the metric gµ⌫ , which is invariant under ⇠.
Therefore, ra Qa = 0, and Qa is a conserved current.: i.e we have for non-null hypersurfaces
Z I p 3 Z Z
ap 4 a a
0= ra Q g d x = Q na h d y = d⌃a Q d⌃a Qa , (7.46)
M @M ⌃2 ⌃1

as long as Qa goes sufficiently fast to zero at ◆0 . Therefore, we can define a conserved quantity I
proportional to the above integral Z
I/ d⌃a Qa (7.47)
⌃
Let us now specialise to the, for us, more relevant case in which we take a family of spacelike
hypersurfaces ⌃i to foliate a stationary spacetime with a horizon. In this case we can write (remember
the Killing vectors’ property Rµ ⇠ = r⌫ rµ ⇠ ⌫ (see (2.86))
Z Z I q
⌫ µ µ ⌫ 2 (2) n r rµ ⇠ ⌫ ;
I/ d⌃µ ⇠ R ⌫ = d⌃µ r⌫ (r ⇠ ) = d ✓ [µ ⌫] (7.48)
⌃ ⌃

where nµ is the timelike vector orthogonal to the three-dimensional spacelike hypersurface setting
the directionality of d⌃µ , while rµ is a spacelike vector orthogonal to the two-sphere S 2 at the

92
boundary of ⌃ introduced by the application of the Stokes’ theorem to remove the total derivative.
The antisymmetrization is induced by the contraction with rµ ⇠ ⌫ (Killing vector).
If we take such boundary S 2 to be a section of H+ it is easy to see that the two null vectors ⌅↵
and N ↵ we defined before can related to n↵ and r↵ as
1 1
⌅↵ = p (n↵ + r↵ ), N ↵ = p (n↵ r↵ ). (7.49)
2 2
With this prescription, we have that I is proportional to
I q
I/ d2 ✓ (2) ⌅ N rµ ⇠ ⌫ ,
[µ ⌫] (7.50)
S

and by Equation (7.43), we can rewrite the integral in the much simpler form
I
1
I/ rµ ⇠ ⌫ dSµ⌫ . (7.51)
2 S
@ @
Now, consider the Kerr geometry; we have two Killing vectors, ⇠ = @t and = @' and therefore
we have two associated conserved quantities. In particular, we can define the Komar conserved
mass I
1
M= rµ ⇠ ⌫ dSµ⌫ , (7.52)
8⇡GN S
which is the mass conserved on a two-surface boundary of the spacetime, therefore the black hole mass
associated to a 2-d slice of the event horizon; and the Komar conserved angular momentum
I
1
J= rµ ⌫ dSµ⌫ . (7.53)
16⇡GN S
Note that in the above formulas the proportionality factors were fixed by the requirement that these
quantities coincide with those obtained from Hamiltonian based definitions.

In general we can foliate a spacetime as x↵ = (t, y a ) by providing a lapse function N and a shift
function N a so that the metric can be written as
⇣ ⌘
ds2 = N 2 dt2 + hab ( dy a + N a dt) dy b + N b dt (7.54)

Now one can choose initial values for the first and second fundamental forms on a given spacelike
hypersurface ⌃, i.e. for hab and the extrinsic curvature Kab , and evolve them using the Einstein
field equations (which also imply a set of constraints to be preserved during the evolution).
Remarkably, the constraint equations imply that when considering the gravitational Hamilto-
nian in the case in which hab and Kab satisfy the Einstein vacuum field equations, one finds that
it can be expressed as a boundary term i.e.
I h ⇣ ⌘ ip
sol 1
HG = N (k k0 ) Na K ab Khab rb d2 ✓ (7.55)
8⇡ St

where k is the trace of the extrinsic curvature of the 2-surface, k0 is the trace of the extrinsic
curvature of the 2-surface when embedded in flat space, and rb is the normal to St . Note also
that this Hamiltonian is is non-zero only for non-compact manifolds.
We can then define the gravitational mass of an asymptotically-flat spacetime as the limit
value of HGsol when S is a 2-sphere at ◆0 evaluate by choosing N = 1 and N a = 0 (so that this
t
asymptotic quantity is associated to pure time translations). This is called the ADM mass.
I
1 p 2
MADM = lim (k k0 ) d ✓ (7.56)
8⇡ St!1 St

93
 Similarly one can define the angular momentum of an asymptotically-flat spacetime by taking the
same limit and choosing N = 0 and N a = a where a is the vector associated to asymptotic
rotations in the cartesian plane.
Of course, we do know that we can consider alternative way to reach infinity and these in
general will be associated to di↵erent notions of mass. In particular, we can consider to characterise
the gravitational mass on a point of I + . This can be achieved by going to null coordinates u, v
and consider the gravitational mass in the limit of v ! 1. This defines the Bondi–Sachs mass as
I
1 p 2
MBS = (k k0 ) d ✓ (7.57)
8⇡ S(u,v!1)

which is mostly relevant when one needs to describe the mass of a radiating body.
Actually, the above example, an isolated body emitting a constant flux of radiation, is useful
in showing the di↵erence between the ADM and Bondi–Sachs masses. In fact in this case, one
has that on a t = constant slice the ADM mass observed at infinity does not change as the sum
of the mass of the body and that of the emitted radiation remains constant albeit both of them
change. On the contrary the Bondi–Sachs mass describes the mass present on a given null slice
at u = constant as v ! 1 hence it fails to intersect and take note of all the radiation emitted
previously of the retarded time u. So in general, it will decrease at u increases.
For a more detail discussion about the 3 + 1 decomposition, the gravitational Hamiltonian and
the ADM and Bondi-Sachs masses see e.g. [[3] chapter 4]

Of course, in a black hole spacetime mass and angular momentum do not need to be associated
only to the spacelike leaves of H+ . In this case we can stick to the initial expression for I as
Z
I/ d⌃µ ⇠ ⌫ Rµ⌫ (7.58)
⌃
p
where again d⌃µ ⌘ d3 x h nµ , and by making use of Einstein field equations
✓ ◆
1
Rµ⌫ = 8⇡GN Tµ⌫ gµ⌫ T , (7.59)
2

we can define a mass Z ✓ ◆

1 p
M =2 Tµ⌫ gµ⌫ T nµ ⇠ ⌫ h d 3 y , (7.60)
⌃ 2
and an angular momentum
Z ✓ ◆
p
1
J= Tµ⌫ gµ⌫ T nµ ⌫
h d3 y , (7.61)
⌃ 2

associated to matter fluxes across the hypersurface ⌃.

Figure 7.7: Maximal analytical extension of Schwarzschild with the chose hypersurface ⌃.

94
 Then, for example, in a black hole spacetime, across a spacelike Cauchy hypersurface foliating it ,
one can see that the total mass is given by
I Z ✓ ◆
p
1 µ ⌫ 1
M= r ⇠ dSµ⌫ + Tµ⌫ gµ⌫ T ⌘ µ ⇠ ⌫ h d3 y , (7.62)
8⇡GN H+ (v) ⌃ 2
| {z } | {z }
black hole mass ⌘MH flux

So the energy on a given hypersurface could be defined by the black hole mass plus the contribution
of matter fluxes outside. A similar relation is valid for the total angular momentum J.
With the above conserved quantities associated to the horizon we can also derive a very useful
formula. Consider the combination MH 2⌦H JH
I Z
1 1 p 2
MH 2⌦H JH = rµ (⇠ ⌫ + ⌦H ⌫ ) dSµ⌫ = rµ ⌅⌫ · 2 ⌅[µ N⌫] d ✓=
8⇡GN H+ (v) | {z } 8⇡GN H+ (v) | {z }
⌅⌫ ⌘ dS
I I I
1 1  
= ⌅µ rµ ⌅⌫ N⌫ dS =  ⌅⌫ N⌫ dS = dS = ABH .
4⇡G | {z } 4⇡GN 4⇡GN H+ (v) 4⇡G
⌅⌫
(7.63)
So we finally get the Smarr formula

MH = 2⌦H JH + ABH . (7.64)
4⇡GN
This completes our minimal toolkit necessary for dealing with the four laws of black hole mechanics.

Zero-th Law
The zero-th law of thermodynamics states that the temperature of a body in thermal equilibrium is
constant all over the body. In the case of black hole mechanics, this reduces to the fact that the surface
gravity is constant over the event horizon of a stationary black hole:  = const over H+ .
We want to prove that  is constant both along each generator of the congruence and also from
one generator to another. Consider the generator surface gravity
1
2 = r ⇠ ↵ r ⇠↵ ; (7.65)
2
also remember that for a Killing vector the relation (2.85) holds r⌫ rµ ⇠↵ = R↵µ⌫ ⇠ . Consider now
the directional derivative of 2 along a generator of the horizon
1 ⇣ ⌘ ⇣ ⌘ ⇣ ⌘
⇠ r 2 = ⇠ r r ⇠ ↵ (r ⇠↵ ) = ⇠ (r r ⇠↵ ) r ⇠ ↵ = ⇠ R↵ ⇠ r ⇠ ↵ = 0, (7.66)
2
and this is zero since we have ⇠ ⇠ which is symmetric in and , but the Riemann tensor is
antisymmetric in the same indices. So the surface gravity is constant along each generator of the
horizon.
What about the transverse direction on a leaf at constant v? I.e when we move from a generator
to another one? The above derivation also implies that in general

r 2 = 2@  = R↵ ⇠ r ⇠↵, (7.67)

so if we multiply both sides by `A :

`A r 2 = 2@ `A = R↵ `A ⇠ r ⇠ ↵ . (7.68)

We would like to show that the right hand side of this equation has to be zero for a stationary black
hole.

95
 Let us specialise first to the case of a non-extremal (so  6= 0), eternal, black hole. The two-surface
where the future event horizon meets the past event horizon is called the bifurcation surface, B. This
is a fixed point in the flow of the Killing vector field, in the sense that the Killing vector that generates
the horizon vanishes there, ⇠ a |B = 0. Then, via Eq. (7.68), we can conclude that (@ ) `A |B = 0; now,
since `A is Lie-dragged along v, and since we already proved that  does not change along v, then we
can deduce that (@ ) `A = 0 on all of H+ .
In the case where we are not dealing with an eternal black hole, but e.g. we have a black hole
created by a collapse, we know that the late-time region in the Carter–Penrose diagram is equal to
the eternal case; therefore, there exists a v̄ such that, for any v > v̄, the two situations are identical,
and hence the surface gravity will be again constant for all the leaves at v = const with v > v̄.
A more rigorous proof, making use of the DEC, can be found in [67].

First Law
In thermodynamics, the first law basically states the conservation of energy; in the case of black hole
mechanics, it can be show that, under an infinitesimal transformation, the following holds:

dM = dA + ⌦H dJ. (7.69)
8⇡
Let us prove it by comparing two close Kerr black holes. ⇣ ⌘
2 ⌘ 8⇡ M 2 + M 4
p
Let us remember that for a Kerr black hole ABH = 16⇡Mirr J 2 . Hence
A = A(M, J) and inverting M = M (A, J); both A and J have the same dimension, which is [A] =
[J] = M 2 , therefore M must be an homogeneous function of degree 12 of J and A:
1/2
M ( A, J) = M (A, J); (7.70)
according to Euler’s theorem, for an homogeneous function f of degree n one can always write,
@f
nf = xi , (7.71)
@xi
then, in our case this becomes
1 @M @M
M =A +J . (7.72)
2 @A @J
Using the Smarr formula (7.64) we substitute for M so that
 @M @M
A + ⌦H J = A +J , (7.73)
8⇡G @A @J
which implies ✓ ◆ ✓ ◆
@M  @M
A +J ⌦H = 0. (7.74)
@A 8⇡G @J
For this to be true for any A and J one has to have
@M  @M
= = ⌦H . (7.75)
@A 8⇡G @J
But now, the total variation of the mass can be written as
@M @M
dM = dA + dJ, (7.76)
@A @J
and substituting the previous relations, we immediately find the first law of black hole mechanics

dM = dA + ⌦H dJ . (7.77)
8⇡G
Note that this derivation heavily relied on properties of the Kerr geometry, a vacuum solution. It is
nonetheless possible to extend the law to situations including (infinitesimal) matter fluxes (see e.g. [3]
page 212).

96
Second Law
In thermodynamics, the second law states that in a thermodynamical process, the entropy of an
isolated system cannot decrease; the analogous in black hole mechanics is the area law, which we have
already seen: if the NEC holds, then A 0. The surface area of a black hole can never decrease
(Area Theorem, Hawking 1971).
The area theorem rests on two important results from Penrose. First, the focusing theorem: if at
some time ✓ = ✓0 < 0 and if the NEC holds, than we shall reach a focusing point in a finite time,
✓ ! 1 (we applied this in the singularity theorem). Second, the event horizon H+ of a black hole
is generated by null geodesics which can have past but no future end-points: a null-ray from I can
get into the horizon and become part of the generators of the horizon, but no generator can leave H+ ,
which in turn implies that it cannot run into a caustic in the future once it has entered H+ . (We
already met this theorem explaining why black holes cannot bifurcate).
Then, with these two results, it is obvious that the area of the horizon cannot decrease: if it
decreased at some point so that ✓ < 0 for some generators, then the focussing theorem guarantees
that these generator would run into a caustic at which ✓ = 1 and then light-rays would leave H+
implying a violation of the “no end-points theorem”.

Third Law
In standard thermodynamics, the third law takes two forms:

• Nerst formulation or isoentropic formulation: S = const at T = 0, that means that the

entropy does not depend on the macroscopic variables of the system; zero temperature states of
a system are isoentropic. Whatever is the starting point, there is a universal value of the entropy
at zero temperature which is furthermore equal to zero accordingly with the so called Planck
postulate, S(T = 0) = 0.

• Unattainability formulation: it is impossible to reach the absolute zero temperature state

with a finite number of thermodynamical steps; we can go to zero temperature asymptotically,
but we cannot reach it.

It is interesting to notice that, for black holes, the third law is violated in the Nerst formulation.
Consider an extremal black hole: the entropy is still proportional to the area, and the area still
depends on the mass of the black hole. However, the third law in the unattainability formulation is
realised: the surface gravity of a black hole cannot reduced to zero within a finite advanced time. This
di↵erence w.r.t. standard thermodynamics seems to hint towards the possibility that extremal black
hole have to be considered a di↵erent class of objects w.r.t. non-extremal ones (see e.g. [69]).
This law was rigorously proved by Werner Israel only in 1986 [68]. Israel proved that the topological
structure of the trapped surfaces of a non-extremal and an extremal black hole is di↵erent, and one
cannot be deformed into the other if WEC holds. For a partial demonstration of this law and its
connection with WEC see e.g. [3].

7.3.4 On the necessity of Hawking radiation

Now that we have shown the four laws of black hole mechanics, it is easy to see that we incur into
paradoxes if we do not assume that black holes can emit Hawking radiation.
First of all, let us notice that black holes can apparently violate the second law of thermodynamics
if they are not endowed with an entropy: indeed in this case it would be enough to drop some entropic
system (e.g. a cup of tea) into a black hole to reduce the total entropy of the universe. In response
to this Bekenstein conjectured that black hole have to had an entropy and the Area law suggested it
had to be proportional to their area
SBH A
= ⌘¯ 2 , (7.78)
kB `P

97
(where ⌘¯ was some unknown proportionality factor and the Area is normalised to the Planck area for
dimensional reasons) and that this entropy had to satisfy a generalised second law by which Stot =
Sout + SBH never decreases.
Similarly, one is led to conjecture by the zeroth and first laws that there had also to be a temper-
ature of associated to the black hole given by
~ 
kB TBH = . (7.79)
8⇡c ⌘¯
However, classically the black hole cannot radiate. This leads to a paradox.
Let us consider a black hole of temperature TBH in a box with a thermal bath of temperature Tout .
Imagine that some radiation E falls into the black hole; the box will have to decrease its entropy by
an amount
E
Sout = , (7.80)
Tout
while the black hole will increase its entropy by the amount
E
SBH = . (7.81)
TBH
If Tout < TBH , then we are in trouble, because
✓ ◆
1 1
Stot = E < 0, (7.82)
TBH Tout
and this would violate the generalized second law.
Now, let us instead conjecture that somehow the black hole radiates black body radiation instead

4 4 3 4 U
U= SB Trad S= SB T = , (7.83)
3 3 Trad
where U is the internal energy. Let us consider again the above ideal experiment, this time in its
worst case scenario, i.e. Tout ⌧ TBH so that we can take Tout ! 0; now, if the black hole emits energy
dE > 0 we have
dE
SBH = , (7.84)
TBH
and for the box
4 dE
Sout = ; (7.85)
3 Trad
but we assumed Trad = TBH so the total variation of entropy is
1 dE
dStot = > 0, (7.86)
3 TBH
So the generalized second law this time holds if somehow black holes can radiate.

7.3.5 Hawking radiation and black hole entropy

As we all know, the resolution of this puzzle came in 1975 with the discovery by Hawking of quantum
black hole radiation which takes his name. By a masterful calculation of quantum field theory on a
black hole spacetime, Hawking found
~ ~c3
k B TH = = , (7.87)
2⇡c 8⇡GM
and hence the previously introduced ⌘¯ turned out to be ⌘¯ = 14 . This implies that the entropy of a
black hole is indeed
SBH ABH
= 2 . (7.88)
kB 4`P

98
 The temperature scales with the inverse of the mass of the black hole
✓ ◆
8 M
T ' 6 ⇥ 10 K. (7.89)
M

So black holes are very cold! (Exercise Compute the mass of a black hole that has a temperature
equal to the temperature of the Cosmic Microwave Background, TCMB = 2.725 K.)
Black holes are unstable: the more they evaporate, the more their mass decreases, the more their
temperature increases, the more they emit; this is a sign that black holes possess a negative heat
capacity, and therefore are really thermodynamically unstable.
Also, the lifetime of a black hole scales like the cube of the mass, ⌧BH / M 3 , since
dM
= T 4 A. (7.90)
dt
While a rigorous derivation of Hawking radiation will require us a full QFT in curved spacetime
treatment, in what follows we shall provide an heuristic derivation based on the physical intuition
that Hawking pair are produced from the vacuum by the tidal forces acting on virtual pairs in the
vacuum in proximity of a black hole horizon.
The basic idea is that in a stationary spacetime, the creation of particles is due to the strong tidal
forces nearby the event horizon of a black hole, breaking virtual pairs apart. Notice that such tidal
forces are stronger the less massive the black hole is, since at the horizon
GM m 1
Ftidal ⇠ 3
r / 2 r.
r M
~
Now, consider pairs with a Compton length c = mc , and use the geodesic deviation equation:

D2 S µ
Aµ = = Rµ⌫⇢ T ⌫ T ⇢ S ,
dt2
where T µ is the vector tangent to the set of geodesics and S µ is the vector going from one geodesic
to the other. Consider now the reference frame at rest with respect to the particle, T µ = (1, 0);
the condition for pair breaking is that the gravitational field must do a work 2m over a distance c ,
therefore we can take S µ = (0, c ). With this, the acceleration becomes
F W 2m 2
A= = = = .
m md m c c

Now, in the reference frame we defined, we have Ai = Ri00j c , so we need |Ri00j | & 2c , meaning a
curvature & 1c .
Now, once the virtual pair is broken apart, the two particles can be considered “on-shell”, and if
they are massive they will travel on timelike geodesics. If the black hole is stationary, than we know
that energy is conserved; because of the vacuum outside of the black hole, the initial energy is zero,
Ein = 0; after the pair creation, the two particles will have energy E1 = p1 · ⇠ and E2 = p2 · ⇠. By
energy conservation:
Ein = 0 = E1 + E2 = (p1 · ⇠ + p2 · ⇠).
If we assume that, for example, particle 1 escapes to infinity, E1 > 0, since ⇠ in this case is the usual
Minkowski timelike Killing vector, and then E2 < 0, meaning particle 2 must have a negative energy
in some local inertial frame.
However, in quantum mechanics, a real particle must have a positive energy in any local inertial
frame: the only way out of this apparent paradox is that particle 2 reaches a region where ⇠ becomes
spacelike: the ergoregion beyond the horizon.
In the region where |⇠|2 > 0 and ⇠ is spacelike, saying that p2 · ⇠ < 0 is equivalent to say that
the three-momentum of particle 2 is negative in some local inertial frame, and it is here that the

99
existence of particle 2 is allowed. This is also the reason why the pair breaking must happen close to
the horizon.
After particle 2 enters the black hole, the mass as measured at infinity must decrease:

M 0 = M + E2 < M, E2 < 0,

and therefore MBH = |E2 | = E1 , being this the energy radiated at infinity.

100
 Part II

Quantum Field Theory

in Curved Spacetimes

101
 8
Preliminaries

8.1 Second quantization in curved spacetime

Motivated by the results obtained for black hole dynamics, we are interested in the quantization of
field theories in non-trivial metric backgrounds. In this section of the notes we shall use the particle
physics signature (+, , , ) (metric determinants in square roots are alway assumed in absolute
value even if not explicitly written).
In particular, let us focus on a real scalar field (x) minimally coupled to gravity. The minimal
coupling procedure consists in replacing all non-generically covariant quantities in the action with the
generically covariant counterparts
⌘ µ⌫ ! g µ⌫
@ µ ! rµ (8.1)
p
d4 x ! g d4 x ,
In this way one is assured that the equation of motion reduces to the usual Klein-Gordon equation in
flat space. Therefore, the non-interacting Lagrangian density for (x) becomes
p
g ⇥ µ⌫ ⇤
L = g rµ r⌫ m2 2 . (8.2)
2
Notice that the term ⇠ 2 R goes to zero in flat space as well, so it is a possible interaction term.
However, such term violates the equivalence principle, since it is peculiar of scalar fields (indeed
fermions and vector fields would couple di↵erently to gravity). Moreover, in d spacetime dimensions
for m = 0 and ⇠ = (d 2)/(4(d 1)) the action for would be conformally invariant.
We now move to the canonical quantization of , that now becomes an operator (or better, an
operator valued distribution). Of course, the equation of motion is
(⇤ + m2 ) (x) = 0 , (8.3)
where the box operator is the usual d’Alembertian operator ⇤ = rµ @µ . First, we can consider a
spatial foliation of M and write the metric gµ⌫ in ADM form. Then, in the canonical quantization we
need to define a conjugate momentum ⇡ as
@L p
⇡(x) ⌘ = g g µt rµ . (8.4)
@(rt )
Then the commutation relations between the fields are
i
[ (t, x), ⇡(t, y)] = p (3) (x y) ,
g (8.5)
[ (t, x), (t, y)] = [⇡(t, x), ⇡(t, y)] = 0 .

102
 Again, within the ADM formalism we can consider a foliation at time t that we call ⌃t then the
scalar product between two solution of the Klein-Gordon (KG) equation 1 , 2 is defined as
Z Z ⇣
⇤ ⇤ µ
p 3 ! ⇤⌘ µ p 3
( 1, 2) ⌘ i ( 1 rµ 2 2 r µ 1 ) n h d x = i 1 r µ 2 n hd x . (8.6)
⌃t ⌃t

This product is independent on the specific ⌃t we pick, indeed:

✓Z ⇣ Z ⇣ ◆
! ⇤⌘ µ p 3 ! ⇤⌘ µ p 3
( 1 , 2 ) ⌃2 ( 1 , 2 ) ⌃1 = i 1rµ 2 n hd x 1rµ 2 n hd x
⌃2 ⌃1
Z ⇣ ! ⌘p
= i rµ 1 r µ ⇤2 gd4 x
ZV
µ ⇠⇠⇠ ⇠ µ ⇠⇠⇠ ⇠⇤ p 4
1 rµ 2 + 1 ⇤ 2 2⇤ 1
⇤ ⇤ ⇤
= i r⇠
⇠ r⇠
⇠ 1 rµ 2 gd x
V
=0,

where in the second line we used Gauss’s theorem (assuming that the fields go to zero fast enough at
◆0 ). In the fourth line we obtain zero since the fields satisfy the KG equation.
The scalar product (., .) defines the positive/negative-norm modes which form a complete orthonor-
mal basis {fi } of solutions of the KG equation

(fi , fj ) = ij (fi⇤ , fj⇤ ) = ij (fi , fj⇤ ) = 0 . (8.7)

These modes also satisfy @t f = i!f and @t f ⇤ = +i!f ⇤ (from this they are called sometime posi-
tive/negative energy modes, albeit ! > 0 in both cases).
The field can then be expanded in such a basis
Xh i
(x) = âi fi (x) + â†i fi⇤ (x) , (8.8)
i

where the index i spans the whole basis and the operators âi , â†i are the creation and annihilation
operators which inherit the commutation relations

[âi , â†j ] = ij , [âi , âj ] = [â†i , â†j ] = 0 . (8.9)

In this basis, these operators define a vacuum state |0if which is annihilated by all the âi ’s:

âi |0if = 0 8i . (8.10)

Then the Fock space is built by n-particles states for the mode i as
1 ⇣ † ⌘n
|ni if ⌘ p âi |0if . (8.11)
n!

Finally, we can define the number operator N̂i|f as

N̂i|f ⌘ â†i âi . (8.12)

Remark 1: the so defined vacuum state is invariant under Poincaré transformations. For rotations
this is trivial; for boosts of velocity v < 1 we have that a frequency !k > 0 is transformed into
!k0 = (v)(!k v|k|) but this is still a positive quantity.
Remark 2: also the normal ordering is a Poincaré invariant procedure.
However, when general coordinate transformations are allowed, all these prescriptions are not
assured to hold. For instance, in this situation the choice of another basis is not necessarily equivalent.

103
Consider a new set of solutions {gA }, (whose indexes are in principle di↵erent from i, j, .. and we call
A, B, ...). For this choice, the field now reads
Xh i
ˆ(x) = b̂A gA (x) + b̂†A gA
⇤
(x) . (8.13)
A

and the vacuum is now |0ig 6= |0if as it is defined as the state annihilated by the new set of operators
b̂A for any A. Although the two basis are di↵erent, the field (x) must be the same so we expect a
relation between the bases. Since we are dealing with linear functional spaces there must be a linear
relation between gA and fi . Then it must be the case that
X
gA = [↵Ai fi + Ai fi⇤ ] . (8.14a)
i
X
⇤ ⇤
fi = [↵Ai gA Ai gA ] . (8.14b)
A

These relations between the two sets of modes are called Bogolyubov transformations and the
coefficients ↵Ai and Ai are called Bogolyubov coefficients. 1
These are explicitly given by the following scalar products:

↵Ai = (gA , fi ), Ai = (gA , fi⇤ ) . (8.15)

From these result it is also easy to obtain the relation between the two operators âi and b̂A .
Xh † ⇤
i
âi = b̂A ↵Ai + b̂A Ai , (8.16a)
A
Xh i
b̂A = ⇤
âi ↵Ai â†i ⇤
Ai . (8.16b)
i

The Bogolyubov coefficients satisfy the following normalisation conditions

X
⇤ ⇤
(↵Ak ↵ik Ak ik ) = Ai , (8.17a)
k
X
(↵Ak ik Ak ↵ik ) =0. (8.17b)
k

A very natural question arises now. Given that the two bases are in principle inequivalent, what is
the expectation value of the number operator N̂A|g over the vacuum |0if ? In other words, how many
particles does an observer in the basis {gA } sees in the vacuum of the observer in the basis {fi }?

X
f h0|N̂A|g |0if = f h0|b̂†A b̂A |0if = †
f h0|(↵Ai âi
⇤
Ai âi )(↵Aj âj
⇤ †
Aj âj )|0if
ij
X †
X †
X
⇤ ⇤ 2
= Ai Aj f h0|âi âj |0if = Ai Aj f h0|(âj âi + ij )|0if = | Ai | . (8.18)
ij ij i

The conclusion is that the mixing between the positive and negative energy modes appearing in
the change of basis (8.14) implies that the observed associated to the quantization in the gA basis
percives the vacuum defined by on observer adopting the fi basis as filled with particles. The concept
of “particle” is Lorentz invariant, however it is not generally covariant.
The Bogolyubov transformations are relevant in several physical situations
1
Note that in the extant literature one can also find the alternative spellings Bogoliubov and (more rarely) Bogolubov.

104
 • They can be used to relate the observations of di↵erent observers in curved spacetimes. E.g. in a
Schwarzschild spacetime they can relate the vacua of an observer static at infinity and of another
one freely falling into the black hole.

• They can relate the vacua in the asymptotic past and future of a universe undergoing an expan-
sion at some time, or in Minkowski space with a time varying potential.

• They are used in condensed matter physics in relating the inequivalent vacua of the atoms and
of the quasi-particle in a Bose–Einstein condensate.

Let’s see a simple example below.

8.2 Particle production by a time-varying potential

The above mentioned possible non-equivalence of particle vacua can be easily seen in the simple
situation of a massive scalar (x) in flat spacetime obeying the usual KG equation but with an
external time-dependent potential U (t):

⇤ + m2 + U (t) (x) = 0 , (8.19)

Since the potential is only time dependent, the equation is linear in space, and a basis orthonormal
modes will have in general the form uk = Nk gk (t)eikx Then the classical field can be written as
Z
d3 k
(x) = Nk gk (t)eikx . (8.20)
(2⇡)3

where the Nk is some normalization factor.

Inserting this ansatz in the KG equation (8.19) we can separate the time and space dependent
equations. In particular for the time dependent modes we get:

g̈k (t) + !k (t)gk (t) = 0 , (8.21)

with the time-dependent frequency !k2 ⌘ |k|2 + m2 + U (t).

Remark: If the function U (t) is periodic, then (8.21) is in the form of a Flouquet equation (or
called Mathieu equation for U (t) = A sin(⌦t)). Such a di↵erential equation has a characteristic band-
like structure for its solutions. For instance, for certain values of the parameters m and ⌦ (generally
within bands) the solution is unstable and behaves as gk ⇠ eµk t , where µk is called Flouquet index.
This parametric amplification is very important for example in the pre-heating stage after inflation
(see e.g. [70] for a review), when the oscillations of the inflaton around the minimum of its potential
lead to huge amplifications for the standard model fields.
We are instead interested in a situation for which the function !k (t) varies over a finite time lapse
and asymptotically takes constant values both in the far past and future

lim !k (t) = !out ,

t!+1
(8.22)
lim !k (t) = !in .
t! 1

Corresponding to the two di↵erent limits, we can expand the field in terms of plane-waves with
(in) (out)
frequency !out or !in . Let us indicate the two di↵erent basis as {uk } and {uk0 }. From the
discussion of the previous section the relation between the “in” and “out” basis is
(out)
Xh (in) ⇤(in)
i
uk0 = ↵kk0 uk + kk0 uk . (8.23)
k

105
Figure 8.1: Instability bands associated to the Matthieu equation, i.e. an harmonic oscillator with a
periodic frequency. In particular here the latter has been parametrized as !(t) = [a(k) 2q cos(2z)]
with z = mt. Figure from [71].

where the isotropy and homogeneity in space implies

⇣ ⌘ ⇣ ⌘
(out) (in) (out) ⇤(in)
↵kk0 = uk0 , uk = ↵k (3) (k k0 ) = ↵k k,k0 , kk 0 = u k0
, u k = k
(3)
(k+k0 ) = k -k,k0 .
(8.24)
Formulas (8.24) are telling us is that particles can be created out of the vacuum in pairs and the overall
momentum is conserved (obviously, given the invariance of the problem under space translations).
Indeed, if the initial quantum state is the vacuum for the “in” observer, then in the final state we
are going to have production of particles out of the vacuum. From (8.18) we have:
(out)
X
h0in |N̂k0 |0in i = | kk0 |2 . (8.25)
k

Now, we would like to characterise the nature of the in vacuum state for a late time observer. For
general U (t) the above problem cannot be analytically solved but one can alternatively look for two
opposite limits: indeed if U̇ /U ⇠ 1/⌧ is the typical time over which the transition between the two
asymptotic values of U happens, and ! is the typical frequency associated to the excited modes, then
we can have two regimes [72]

• Sudden regime !⌧ ⌧ 1: modes for which this is realised have a small energy compared to the
energy that U brings into the system. Hence, they can efficiently be excited and we expect a
large particle production;

• Adiabatic regime !⌧ 1: modes in this regime corresponds to highly energetic particles whose
production we then expect to be heavily (exponentially) suppressed.

Note that sometimes these two above regimes are characterised by the equivalent conditions !/!
˙ !
for the sudden regime, and !/!
˙ ⌧ ! for the adiabatic one.
In particular one can see that in the sudden regime, assuming an instantaneous transition from
!in to !out (e.g. at t = 0 due to a potential U (t) / (t)), one gets

2 1 (!in + !out )2
| k| ⇡ , (8.26)
4 !in !out

106
Note that in this case Bogolyubov coefficient is invariant under the exchange of !in and !out as only
their relative value does matter. In the case of an adiabatic regime one has instead
2
| k| ⇡ exp ( 2⇡ !out ⌧ ) , (8.27)

Finally, note that in the above results we have used (8.24).

For the adiabatic case, one can say, in the Schrödinger picture, that during an adiabatic change of
!(t) the state continually adjusts to remain close to the instantaneous adiabatic ground state. This
state at some time t = t0 is the one annihilated by the lowering operator a(ft0 ) defined by the solution
f (t) satisfying the initial conditions at t0 and corresponding to the “instantaneous positive frequency
solution” at some time t. This instantaneous adiabatic ground state is sometimes called the “lowest
order adiabatic ground state” at t0 .
We can write the total number of out-particles present at late times as (introducing the proper
volume normalization) Z 1
V
N= 2 | kk0 |2 !out
2
d!out , (8.28)
2⇡ 0
and hence the number spectrum will be
dN 1 2 2
= 2| kk0 | !out (8.29)
V d!out 2⇡
The full spectrum will interpolate between these two regimes and will be characterised by a peak at
frequencies !out ⇠ 1/⌧ . The resulting shape is shown below in Figure 8.2. Let us conclude that the

0.008

0.006

dN

0.004

0.002

0 0.5 1 1.5 2
Frequency

Figure 8.2: Number spectrum (photons per unit volume). The horizontal axis is !out in arbitrary units.
The typical timescale ⌧ is set equal to one. The vertical axis is in arbitrary units. The spectrum might
resemble a black body but it is di↵erent. At low frequencies, where the sudden approximation holds, it
2 instead of linearly. At high frequency instead, it does show an exponential Planck-like
scales like !out
suppression.

case of particle creation in an expanding homogeneous and isotropic universe is a simple generalisation
of the above case. Also in this setting we shall have everywhere in space back-to-back pair creation
due to conservation of 3-momentum and overall picture remains basically unchanged.

8.3 Coherent and Squeezed states

We want now to describe more precisely the relation between the in and out vacua which we expect
to be related via a unitary operator. A quick – although not rigorous – way to obtain such mapping
is the following.

107
 For concreteness, let us consider again the example of particle production associated to a time-
varying potential. Recalling (8.16a) we can write
(in)
Xh (out) †(out)
i (8.24)
h
(out) †(out)
i
⇤ ⇤
0 = âk |0in i = ↵kk0 âk0 + kk 0 â k0
|0 in i = ↵ â
k k + k â k |0in i . (8.30)
k0

Hence ⇤
(out) k †(out)
âk |0in i = â k |0in i . (8.31)
↵k
Now let’s note that from h i
(out) †(out)
âk0 , âk = (3)
(k k0 ) , (8.32)
@
the operator â can be seen formally as (think about the relation between x and p = @/@x) as â = @â†
.
Then the above relation can be written as
⇤
@|0in i k †(out)
= â k |0in i (8.33)
@a†k(out) ↵k

Formally, one could “integrate” both sides to get

✓ ◆ ⇤
|0in i k †(out) †(out)
log = â âk . (8.34)
|0out i 2↵k k
Then, by summing over all k’s this implies
Y  ⇤
k †(out) †(out)
|0in i = Nk exp
â â k |0out i
2↵k k
k
+1 p ✓ ⇤ ◆n
Y X
n 2n! k
= Nk ( 1) |nk , n-k i , (8.35)
n! 2↵k
k n=0

where Nk are normalization coefficients and |nk , n-k i is a collection of 2n particles of which n have
momentum k and the other n momentum k.
Therefore, the in vacuum appears as what is called a squeezed state for the out observer. Indeed,
as we will outline in a moment, equation (8.35) is nothing but a particular form of a squeezed state.
Generally speaking, for single-mode states the most generic squeezed state is defined as [73]

|z, ⇣i = D̂(z)Ŝ(⇣)|0i, z, ⇣ 2 C , (8.36)

where D̂(z) is the displacement operator whilst Ŝ(⇣) is the squeezing operator and are defined as
h i
D̂(z) ⌘ exp z↠z ⇤ â , (8.37)
 ✓ ⇣ ⌘ ◆
1 2
Ŝ(⇣) ⌘ exp ⇣ ⇤ (â)2 ⇣ ↠. (8.38)
2
Further, these additional relations hold (for the sake of clarity, we suppress the explicit dependence of
the operators):
⇣⇤
Ŝ † ↠Ŝ = (cosh |⇣|) ↠+ (sinh |⇣|) â , (8.39)
|⇣|
⇣
Ŝ † âŜ = (cosh |⇣|) â + (sinh |⇣|) ↠, (8.40)
|⇣|
Ŝ † Ŝ = 1 , (8.41)
D̂† âD̂ = â + z , (8.42)
† † † ⇤
D̂ â D̂ = â + z . (8.43)

108
 A coherent state is an eigenstate of the annihilation operator, or equivalently can be seen as a
squeezed state (8.36) with ⇣ = 0. This type of state represents a “classical” solution, in the sense that
its expectation value is
h i = zf + z ⇤ f ⇤ , (8.44)
and its variance minimizes the quantum uncertainty:
2
h i = h i2 . (8.45)

On the other hand, a squeezed state is obtained for z = 0 and is the “least classical” state. They
are associated with the increase (w.r.t. the coherent case) of the uncertainty in one quantum variable
w.r.t. its conjugate. Also this kind of states are associated with violations of energy conditions.
If we go back to (8.30) and confront it with (8.40) we can see that they have the same form if we
identify
⇤ ⇣
↵k = cosh |⇣| , k = sinh |⇣| , (8.46)
|⇣|
(out) (in)
then, clearly, the relation between in and out annihilation operators becomes Ŝ † âk Ŝ = âk .
ˆ out i we get:
Thence, if we write |0in i = ⌅|0
(in) (out) ˆ out i .
0 = âk |0in i = Ŝ † âk Ŝ ⌅|0 (8.47)
ˆ = Ŝ † , so that
This relation is satisfied only for ⌅

|0in i = Ŝ † |0out i (8.48)

I.e. particle production from the quantum vacuum via Bogolyubov coefficients is always realised by a
squeezing operation. Of course, (8.35) has not exactly the same form as (8.38) but this is just because
in the case of pair production from a time-varying potential we do not have a single-mode state but
rather a two-mode squeezed state.
In general, a two-modes, k and l, squeezing operator takes the form
 ⇣ ⌘
1
Ŝkl = exp ⇣â†k â†l ⇣ ⇤ âk âl . (8.49)
2

and generates a squeezed state when applied on the two modes vacuum |⇣kl i = Ŝkl |0k , 0l i. In quantum
optics the two modes are sometime called as the “idler ”and the “signal”.
It is interesting to consider the expectation value of the number operator of one of the two modes
of the pair in such a squeezed state

h⇣kl |N̂k |⇣kl i = h0k , 0l |Ŝ † â†k Ŝ Ŝ † âk Ŝ|0k , 0l i = sinh2 |⇣|h0k , 0l |âk â†k |0k , 0l i = sinh2 (⇣) = | k|
2
, (8.50)

where we have used the relations (8.39) and (8.40).

A natural question arises at this point: can a squeezed state produce a thermal distribution? A
brief manipulation implies
sinh2 |⇣| 1
sinh2 |⇣| = 2 2 = 2
(8.51)
cosh |⇣| sinh |⇣| [tanh |⇣|] 1
~!k /(2kB T )
= Planck distribution if =) tanh |⇣| ⇠ e , (8.52)

where indeed
| |2
tanh2 |⇣| = . (8.53)
|↵|2
So, in this case we get a black body spectrum at an e↵ective temperature
~!k
kB Tsqueezing = . (8.54)
2 ln coth |⇣|

109
Note the dependence on !k , this implies that in general “signal” and “idler” do not need to have the
same squeezing temperature.
Some comments are in order at this point. The above results about thermality and squeezing
have a strong formal analogy with thermofield dynamics (TFD) [74], where a doubling of the physical
Hilbert space of states is invoked in order to be able to rewrite the usual Gibbs (mixed state) thermal
average of an observable as an expectation value with respect to a temperature dependent “vacuum”
state (the thermofield vacuum, a pure state).
In the TFD approach, a trace over the unphysical (fictitious) states of the fictitious Hilbert space
gives rise to thermal averages for physical observables completely analogous to the one in (8.50) if we
make the identification (8.52).
The formal analogy with TFD allows us to conclude that, if we measure only one photon mode,
the two-mode squeezed-state acts as a thermofield vacuum and the single-mode expectation values
acquire a thermal character corresponding to a “temperature” Tsqueezing related with the squeezing
parameter ⇣
From this analysis, the conclusion is that squeezed states can be related to thermality if the
special condition (8.52) holds. This implies an “exponential” behaviour of the Bogolyubov coefficients,
needed for such a thermal spectrum can naturally arise in de Sitter space, where the scale factor goes
exponentially a(t) ⇠ eHt , or in spacetimes with black holes. In the latter case the “peeling” of geodesics
close to the horizon turns out to be related to the temperature of the black hole itself.
Additionally, it is important to realise that by going from one basis to another, unitarity is never
lost because the squeezing operator is unitary: Ŝ † Ŝ = 1.
Finally, a Planck distribution is not equivalent to thermality. In fact, at thermal equilibrium all
the n-point correlation functions of a field theory become trivially related to the 2-point correlation
function. However, in a squeezed vacuum this is in general not the case [75].

110
 9
E↵ects of Quantum Field Theory in
Curved Spacetime

9.1 The Casimir e↵ect

In 19th century P.C. Causseè described in his “L’ Album du Marin” a mysterious phenomenon which
was the cause of maritime disasters [76, 77]. Figure 9.1 shows the situation which he called “Calme
avec grosse houle”: no wind but still with a big swell running. In this situation he stressed that if two
ships end up lying parallel at a close distance then often “une certaine force attractive” was appearing,
pulling the two ships towards each other and possibly leading to a collision.

Figure 9.1: From Causseè “The Mariners Album” [76], two ships heavily rolling in a situation where
there are still long waves but no wind.

This phenomenon was for many years considered as just another superstition of sailors because it
was far from clear what force should be at work in these situations. It is only recently that this e↵ect
was given the name of the “Maritime Casimir E↵ect” [77].
The real Casimir e↵ect, however, was discovered in 1948 by Casimir himself. His idea was to
consider a system of two conducting plates facing each other, and to study the vacuum energy of the
system.
First, let us consider a Cartesian system of coordinates (x1 , x2 , x3 ) and a single thin conducting
plate (a foil) positioned along x3 = 0. What is the electromagnetic stress-energy tensor Tµ⌫ for this
vacuum configuration?

111
 • Due to Lorentz invariance in the coordinates (t, x1 , x2 ) we expect the stress energy tensor to be
proportional to the Minkowski metric in these coordinates: Tµ⌫ / diag( 1, 1, 1, ?)

• Conformal invariance of the Electromagnetic Lagrangian, implies that our SET must also be
traceless. This fix the last component to be Tµ⌫ / diag( 1, 1, 1, 3)

• Finally, the proportionality factor must be depdent only on the direction of the symmetry break-
ing, i.e. x3 , so: Tµ⌫ = f (x3 ) diag( 1, 1, 1, 3)

However, being an energy density our SET must also have dimension 4 in energy. Unfortunately we
have at our disposal only ~ (dimensions E ⇥ t) and c (a velocity) and we miss a length scale (or mass
or time). Hence we conclude that f = 0 and therefore Tµ⌫ = 0 everywhere.
Now, consider two plates positioned at a distance a between each other and along the plane x3 = 0.
Such a configuration is depicted in Fig 9.2.

Plates
Vacuum
Fluctuations

Figure 9.2: Casimir configuration for two conducting plates, facing each other at a distance a.

With the same argument as before one can show that the form of hTµ⌫ i in this case we can give to
f (x3 ) the dimension of an energy density (E/m3 ) by requiring f (x3 ) = ↵(~c)/a4 so that finally

~c
hTµ⌫ i = ↵ diag( 1, 1, 1, 3), (9.1)
a4
where ↵ is a real constant to be determined. From a QFT calculation, Casimir was able to show that
↵ = ⇡ 2 /720 so the energy density in the vacuum state at zero temperature is

⇡ 2 ~c
⇢Cas = hTtt i = . (9.2)
720a4
We can then calculate the energy between the plates as

⇡ 2 ~c
ECas = a⇢ = . (9.3)
720a3
Which implies that it is energetically convenient to reduce the distance between the plates (minus
sign). Consequently, there will be an attractive force F between the plates given by

@ECas ⇡ 2 ~c
F = = . (9.4)
@a 240a4
A direct measurement in 1958 at the Philips Labs found indeed that F ⇡ 0.2 · 10 5 N/cm2 for plates
separated by a distance a = 0.5 µm.

112
9.1.1 1+1 dimensional case:
Let us know compute explicitly the Casimir energy density in the simpler case of a two dimensional
spacetime. First, we note that in the Coulomb gauge r ~ ·A ~ = 0 the electromagnetic potential Aµ is
equivalent to two independent real scalar fields. For convenience, we can quantise a single scalar and
multiply our result by 2. The equation of motion for this scalar (t, x) is the Klein-Gordon equation
⇤ (t, x) = 0.
It is clear that the eventual di↵erence in energy will not be due to the outside modes but to the
di↵erent distribution of the field modes confined to between the plates (actually in this 1 + 1 case
limited to a segment of length a) w.r.t the one present in the absence of the plates. So, we shall need
to find our solutions both with and without (free space) the two plates.
When the two plates are absent, we are e↵ectively in Minkowski space, so the solution is
Z h i
dp
(t, x) = p âp e ip·x + â†p eip·x , (9.5)
(2⇡)2!p

where p · x = t!p xk is the usual scalar product, !p = |k| is the frequency, and âp , â†p are respectively
the annihilation and creation operators satisfying the oscillator algebra [âp , â†k ] = (p k). The vacuum
state of the outside is defined as the state annihilated by all the âp so that âp |0i = 0. Therefore the
vacuum energy in the free region is (from now on we take ~ = c = 1)
Z
1 +1 L
Efree = h0|Hˆ |0i ⇥ Volume = |k| dk , (9.6)
2 1 2⇡
where Hˆ is the Hamiltonian density for the single scalar field and we have introduced a normalisation
length L for the one dimensional volume. Of course, this quantity diverges, so we decide to regularize
it with an exponential function e ks . In the end we will eventually take the limit s ! 0. Also we want
to compute the energy in a volume a so we shall multiply our expression by a/L
Z Z +1
a a 1 +1 |k|s a 2a @ a
Efree = Efree = |k|e dk = e ks dk = (9.7)
L 2 1 2⇡ 2(2⇡) @s 0 2⇡s2
In the inside region we must fulfil the Dirichlet boundary conditions on the plates:
(t, 0) = (t, a) = 0 8t 2 R (9.8)
hence, the wave-number assumes only discrete values kn = ⇡n/a and the solution of the wave equation
is
1 h i
+1
X
(t, x) = p ân e it!n + â†n eit!n sin (kn x) , (9.9)
a!n
n=1
and the vacuum energy is
+1
1 X ⇡n ⇡
Eclosed = h0|Hˆ |0i ⇥ Volume = = ⇣( 1), (9.10)
2 a 2a
n=1

where ⇣( 1) is the Riemann zeta function in 1, which formally diverges. 1 We regularize the infinity
as before:
+1
1 X ⇡n s⇡n/a ⇡ a ⇡
Eclosed|s = e = 2 ' 2
+ O(s). (9.11)
2 a 8a sinh (⇡s/(2a)) 2⇡s 24a
n=1
1
The Riemann’s zeta function is defined as
X1
1
⇣(x) = x
n=1
n
which converges for real x > 1. It is possible to do an analytical continuation of this function to all x 2 C with the
exception of x = 1 where ⇣(x) has a pole. While the e↵ectiveness of the Riemann zeta function to provide straightfor-
wardly the correct result as a proper renormalisation albeit apparent miraculous it appears to be ubiquitous. See [5] for
a more complete discussion.

113
 Now, the Casimir energy inside the plates is the di↵erence between Equation (9.11) and Equation
(9.7):
⇡
E= . (9.12)
24a
Note that the same result would have been obtained equivalently by utilizing the analytical continu-
1
ation of the Riemann function in (9.10) that gives ⇣( 1) = 12 .
Using the last observation it is now easy to obtain the real result in 4 dimensions. We just
regularize the vacuum energy in the closed region and analytically continue the divergent result. From
the previous calculation it is easy to get directly the form of the energy, that we multiply by a factor
2 for the electromagnetic case:
Z +1 r ⇣ ⇡n ⌘2
2 d2 k X 2+
E=a 2
k
R2 (2⇡) n=1 a
Z +1 
d2 k X 2 ⇣ ⇡n ⌘2 s
= lim a2 2
k +
s! 1/2 R2 (2⇡) n=1
a

1 (s 1) ⇣ ⇡n ⌘2
+1
X 2s
= lim
s! 1/2 4⇡ (s) a
n=1
⇡2 ⇡2
= ⇣( 3) = ,
6a 720a
which is the required result.
Also non-trivial topologies can have a Casimir energy, and in this case we talk about topological
Casimir e↵ect. For instance consider a spatially closed manifold (a circle) of length L. Then a scalar
field defined over the circle has to obey periodic boundary conditions:
(t, 0) = (t, L)
@x (t, 0) = @x (t, L)
With the same procedure as in the Casimir case, one can obtain the vacuum energy for the scalar. In
this situation the energy is E = ⇡/(6L). If we would have chosen anti-periodic boundary conditions
we would have got E = +⇡/(12L). In general the topology of spacetime is not fixed by the metric
(neither by the Einstein equation). A natural but still open question arising is: why do we live in a
simple-connected geometry?
The above results highlight an important point: one cannot see the Casimir e↵ect as a produced
by a reduction of available modes inside the plates: the allowed modes are still infinite. One has
rather to think to the e↵ect as something induced by a mode redistribution caused by the presence of
the plates/boundary conditions. Also the required Casimir subtraction highlights the fact that this
redistribution is di↵erent from the free case mainly at wavelengths set by the size of the confined region.
This is intrinsically an infrared e↵ect that leaves unmodified the structure of the UV divergences of
the field which hence are the same between the free and the bounded case and can be removed via
the Casimir subtraction.
Finally, let us point out an important feature common to a wide class of cases of vacuum polar-
ization in external fields. We can return to our example of 1 + 1 Casimir e↵ect. Consider the
energy density which in this case is simply given by Eq. (9.10) divided by the interval length a.
1
1 X ⇡n
" = h0|T00 |0i = !n , !n = (9.13)
2a a
n=1

In order to compute it we can adopt a slightly di↵erent procedure for making the Casimir sub-
traction. It is in fact sometimes very useful in two dimensional problems to use (especially for
problems much more complicated that the one at hand) the so called Abel–Plana formula which

114
 is generically
1
X Z 1 Z 1 1
X
F (0) F (i ) F ( i )
F (n) F ( )d = +i d ⌘ reg F (n) (9.14)
0 2 0 exp (2⇡ ) 1
n=0 0

where F (z) is an analytic function at integer points and is a dimensionless variable. This
formula is a powerful tool in calculating spectra because the exponentially fast convergence of the
integral in F (±i ) removes the need for explicitly inserting a cut-o↵.
In our specific case, F (n) = !n and = ⇡t/a, and so Eq.(9.14) applied to the former energy
density takes the form Z
1 1 !d!
"= (9.15)
⇡ 0 exp (2a!) 1
Although it may appear surprising, we have found that the spectral density of the Casimir energy
of a scalar field on a line interval does indeed coincide, apart from the sign, with a thermal spectral
density at temperature T = 1/a.

9.2 The Unruh E↵ect

The Unruh e↵ect takes place in a Minkowski spacetime when one considers the response of a detector
which is uniformly accelerated. The upshot is that Minkowski vacuum appears as a thermal state to
uniformly accelerated observers with proper acceleration ap at a temperature equal to

~ap
KB TU nruh = .
2⇡c
In order to see how this is realise let’s understand first the causal structure perceived by these accel-
erated observers.

9.2.1 The Rindler Wedge

For our purposes it is sufficient to study the case of 1 + 1 dimensions. Let us consider a flat spacetime
with metric gµ⌫ = diag(+1, 1). The spacetime interval is simply ds2 = dt2 dx2 . The velocity uµ
of an observer moving with constant proper acceleration ap satisfies the following equation

uµ rµ u⌫ = a⌫p . (9.16)

One can easily check by integrating this equation that, in Cartesian coordinates (t, x), such an observer
would lie along the hyperbola x2 t2 = ap 2 . Inspired by this, one can consider a new set of coordinates
(⌘, ⇠) under which this motion is just along the curve ⇠ = const. Of course, the required change of
coordinates is an hyperbolic parametrization:

ea⇠
t= sinh(a⌘)
a (9.17)
ea⇠
x= cosh(a⌘),
a
where the parameter a is just a bookkeeping parameter, not a physical acceleration like ap , which
in principle we could put equal to one. In this system of coordinates (9.17) it easy to see that the
⇠=constant lines are the hyperbola
x2 t2 = [ea⇠ /a]2 ,
which implies
a⇠
ap = ae .

115
Figure 9.3: Eternally accelerating observers in Minkowski space. Their worldlines are shown in blue
and labelled by ⇠. Events in the shaded region such as the black dot are hidden to them. The Rindler
horizon is the boundary between the shaded and unshaded regions. Rindler space corresponds to the
right wedge outlined by the dashed black null lines. The straight dotted lines are lines of constant
Rindler time ⌘. Picture taken from “Black Holes notes”, by Fay Dowker.

Also these hyperbola intersect the t = 0 axis at a distance d = (exp ⇠a)/a = 1/ap . The ⌘=constant
surfaces are straight lines stemming from the origin at di↵erent angles as ⌘ varies. This is why ⌘ is
also called the hyperbolic opening angle of the Rindler wedge x > |t|. See Fig. 9.3.
The change of coordinates (9.17) puts the metric in the conformal form

ds2 = e2⇠a d⌘ 2 d⇠ 2 . (9.18)

Of course we can also introduce null coordinates related both to the Cartesian coordinates say U = t x
and V = t + x as well as to the Rindler ones, u = ⌘ ⇠ and v = ⌘ + ⇠. In these coordinates the metric
is respectively
ds2 = dU dV = e2a⇠ du dv.
The coordinates (9.17) cover only a patch of the full Minkowski space (the part on the right of
Fig 9.3, x > |t|). Hence, an eternally accelerating observer sees an horizon (see Fig. 9.3): the Rindler
horizon at t = x, or ⌘ ! +1. More specifically H+ = {(x, t) 2 M|x = t}. This horizon is generated
by a Killing vector field, which is also the generator of boosts,
✓ ◆
@ @ @
boost ⌘ =a x +t . (9.19)
@⌘ @t @x
and is elsewhere the tangent to the aforementioned hyperbola of constant ⇠.
In Cartesian coordinates this vector has contravarian components µ = a(x, t) and hence µ =
a(x, t) covariant ones. Also 2 = a2 (x2 t2 ). The surface gravity can be obtained easily from the
“normal” definition,
@µ 2 = 2 µ |H+ . (9.20)
Indeed by direct substitution and choosing the x component of the above equation

@x a2 (x2 t2 ) = 2at , (9.21)

2
(2a x = 2at)|H+ . (9.22)

Hence  = a.

116
 From the previous comment, one can realize that such a quantity is not the one observed by one of
the Rindler observer. The physical, or proper, surface gravity is the one measured by an accelerating
observer (at constant ⇠). Along its curve, the proper time is d⌧ = ea⇠ d⌘ so the vector of interest is
@⌧ = e a⇠ @⌘ . Therefore the acceleration per unit mass that we obtain is proper = e a⇠ = ap .
At this point we are ready to perform the quantisation in the two frames, and see if accelerated
observers measure a di↵erent number of particles.

9.2.2 Quantum field theory on the Rindler Wedge

In Minkowski coordinates, the quantization for a massless scalar is, as usual, done in terms of plane
waves: Z ⇣ ⌘
(t, x) = âk ūk + â†k ū⇤k dk, (9.23)
R
where the plane waves are defined as
eikx i$t
ūk = p , (9.24)
(2⇡)2$
where $ = |k| and 1 < k < +1. These are positive norm modes with respect to the global timelike
Killing vector @t so that @t ūk = i$uk . The Minkowski vacuum |0m i is defined as âk |0m i = 0.
In the Rindler coordinates, however, we need to be careful. Indeed, the coordinates (9.17) cover
only the right patch of spacetime (see Fig 9.3). The other side can be covered by modifying the
definition of x ! x (9.17). For this reason, a complete basis for the scalar field in Rindler

t x t x

Figure 9.4: Minkowski space divided in Left and Right Rindler wedges

coordinates has to contain both right and left modes 2 . In particular, the field has to be expanded as
Z ⇣h i h i⌘
L† ⇤L R† ⇤R
(⌘, ⇠) = b̂L u
k k
L
+ b̂ u
k k + b̂ R R
u
k k + b̂ k u k dk, (9.25)
R

where in this case the modes are

eikx i!⌘ eikx+i!⌘
uR
k = p , uL
k = p , (9.26)
(2⇡)2! (2⇡)2!
2
We do not include modes in the future and past regions of Fig. 9.4 as these region are causally connected with the
Rindler wedges and modes there can be hence decomposed in combinations of the wedges ones.

117
while they are zero when evaluated on the opposite wedge of the Rindler space (uR L
k |L = uk |R = 0).
Also in this case ! = |k| and 1 < k < +1
Note also that uR L
k and uk are positive-norm modes with respect to future pointing timelike Killing
vectors which have opposite signs, i.e. +@⌘ in the right wedge and @⌘ on the left wedge (think at the
fact that lines of increasing ⌘ on the right side correspond to lower an lower lines on the left wedge).
Hence the positive norm modes are

@ ⌘ uR
k = i!uR
k, @ ⌘ uL
k = i!uL
k. (9.27)
R/L
Finally, the Rindler vacuum |0iRin is defined as b̂k |0iRin = 0.
Our goal is to compute the number of particles measured by an accelerating observer in the
Minkowski vacuum. I.e. to compute $,! = (uR ⇤
! , ū$ ) That this Bogolyubov coefficient will be non
zero is already evident from the fact that uR L
k does not go over smoothly to uk at the origin, given
that there the exponent has to jump between ±i!⌘, while ūK is obviously analytic at the same point.
I.e. Rindler positive norm modes cannot be made of a simple superposition of positive norm modes of
the Minkowski base.
It was Unruh to first realise that a smart way to get around this problem at the origin was to
introduce a suitable combination of the left and right Rindler modes which are indeed always analytical
and bounded for all real (U, V ) and everywhere in the lower half complex (U, V ) planes (see e.g. [5]
page 115)
(1) ⇡! ⇡!
⇤L
fk Nk ⌘ e 2a uR k +e
2a u
k
(2) ⇡! ⇡!
(9.28)
⇤R L
fk Nk ⌘ e 2a u k + e 2a uk ,
where Nk are just normalization coefficients.
We can then use these combinations for expanding our scalar field so to find the following scalar
products with the old basis
(1) 1 2a ⇡! (2) ⇤ 1 ⇡!
(fk , uR
k 0 ) = Nk e k,k0 (fk , uR
k 0 ) = Nk e
2a
k,k0
(9.29)
(1) ⇤ (2)
(fk , uk 0 R ) = 0 (fk , uk0 R ) = 0.

In terms of these new functions the field can be written as

Z h ⇣ ⇡! ⌘i 1/2 n⇣ ⌘ ⇣ ⌘o
(1) (1) (1)† ⇤(1) (2) (2) (2)† ⇤(2)
(⌘, ⇠) = 2 sinh dˆk fk + dˆk fk + dˆk fk + dˆk fk dk (9.30)
R a
(i)
where the dˆk are the operators associated to the new basis. Given that these modes share the same
analyticity properties of the Minkowski modes, they must also share the same vacuum state. Hence
the operators dˆk also annihilate the Minkowski vacuum.
From the scalar products (9.29) we also get the Bogolyubov relations between the b̂k and the dˆk
h ⇣ ⇡! ⌘i 1/2 n ⇡! (2) ⇡!
o
(1)†
b̂L
k = 2 sinh e 2a dˆk + e 2a dˆ k
a (9.31)
h ⇣ ⇡! ⌘i 1/2 n ⇡! ⇡!
o
(1) (2)†
R
b̂k = 2 sinh e 2a dˆk + e 2a dˆ k .
a

At this point, since we know how the dˆk acts on the Minkowski vacuum |0iM we are ready to
evaluate the expectation value of the number operator on the right wedge of the Rindler space:
⇡! ⇡!
†R R e a ˆ(2) ˆ(2)† e a 1
M h0|b̂k b̂k |0iM = M h0|d k d k |0iM = ⇡! = 2⇡! (9.32)
2 sinh ⇡!
a 2 sinh a e a 1
a
This corresponds to a thermal distribution of particles with temperature T = 2⇡ . This is sometime
called the Rindler wedge temperature, as it not associated to a special observer, and it can always be

118
rescaled to be Twedge = 1/2⇡. However, the physical temperature is experienced by an observer along
⇠ = const. Therefore, as we did for the proper surface gravity, we can obtain a proper temperature
Tproper as
T a a⇠ p
Tproper = p = T e a⇠ = e = (9.33)
g00 2⇡ 2⇡
This corresponds to the temperature measured by the accelerating observer.
Comments:

• A non zero Bogolyubov coefficient does not necessarily imply a non zero SET as there can be
↵ interference terms. Nonetheless, in the special case when one gets a thermal distribution,
the quantum correlations are lost and interference terms vanish. So on a single Rindler wedge
the SET will be the one corresponding to a thermal bath at the wedge temperature.

• The total SET is covariant under coordinate transformation and so if it is equal to zero in
Minkowski (as it is set by definition), it should be zero also for accelerated observers. In fact
it can be shown that the expectation value of the SET in the (Rindler) vacuum state of the
accelerated observer is non zero but corresponds to a vacuum polarization which is equivalent
to subtracting from the Minkowski vacuum a thermal bath at the Unruh temperature [75].
This contribution exactly cancels out the other one coming from the fact that a thermal
bath of real particles is actually experienced from the Rindler observer. So in each wedge

h0|Tµ⌫ |0iMink = 0 = h0|Tµ⌫ |0iRind + Thermal bath (9.34)

Physically this tells us that we can regard the non-equivalence of the Minkowski and Rindler
vacua as an example of vacuum polarization. Moreover this result can easily be understood
in a thermofield dynamics framework. In this case the SET of Minkowski can be written
as the sum of its corresponding values in the left and right Rindler wedges (which play the
role of the two “twin spaces” of TFD [74])

h0|Tµ⌫ |0iMink = 0 = h0|Tµ⌫ |0iL + h0|Tµ⌫ |0iR (9.35)

Each of the terms on the right hand side contributes a thermal bath at the Unruh temper-
ature but the sign of the timelike Killing vector is opposite on the opposite wedges. Hence
h0|Tµ⌫ |0iL = h0|Tµ⌫ |0iR . So we see how the TFD construction can help in rapidly arriving
at otherwise non-intuitive results.

• The thermal distribution can be seen as an e↵ect of the apparent horizon at ū = v̄ = 0

experienced by the Rindler observer. This apparent horizon is similar to the event horizon
of a black hole although it has a less objective reality: it is dependent on the state of motion
of some observers and is not observed, for example, by Minkowski observers.
Moreover, this apparent horizon can be globally defined for observers who have undergone
a uniform acceleration for an infinite time. In the case of uniform acceleration for a finite
time, the thermal spectrum is replaced by a more complicated and general distribution.

• Remarkably, one can readily understand that the found result is very general, independent
on the details of the field, the detector and of how the constant acceleration is reached.
There is indeed a very general result due to Bisognano and Wichmann which states that
when restricted to a Rindler wedge the vacuum state of a relativistic quantum field is a
canonical thermal state with density matrix [78]

⇢ˆ / exp[ 2⇡H⌘ /a],

119
where H⌘ is the so called boost or Rindler Hamiltonian given by
Z Z
a b
H⌘ = Tab d⌃ = a x Ttt dx dy dz.
⌃t

I.e. this is the Hamiltonian associated to the generator of boosts @⌘ . From ⇢R we can easily
~a
see then that the “temperature” of the Rindler wedge is always going to be TR = 2⇡
The above results are rooted in the fact that the Minkowski vacuum can be, not surprisingly,
written as an entangled state [75]
YX
|0iM / e ⇡!n |nL i ⌦ |nR i (9.36)
! n

Where !n are the eigenvalues of H⌘ . Hence, if we trace over the states of the left (or right)
side we obtain a thermal distribution.
X
⇢ˆR / TrL |0iM h0| / e 2⇡!n |nR ihnR | (9.37)
n

120
 10
Evaporating Black Holes

10.1 Moving Mirrors

Before delving into black hole radiation it is interesting and instructive to analyse the behaviour
of quantum radiation in presence of a moving boundaries in a flat gravitational background. More
precisely, let us focus on the idealised situation in 1 + 1 dimensions of a massless scalar field on a
a flat metric with a moving boundary over which imposes on the field Dirichlet boundary conditions,
so that this boundary can be considered a mirror in motion.
In light cone coordinates, u = t x and v = t + x, the trajectory of the mirror can be parametrised
as v = p(u) or u = q(v), where p(u), q(v) are functions satisfying u = q(v) = q(p(u)) (i.e. q = p 1 ).

10.1.1 Field theory with moving boundaries

The boundary condition for (u, v) takes the form

(u, p(u)) = 0 8 u 2 R. (10.1)

Moreover, the wave equation has the very simple form

@2
⇤ (u, v) = (u, v) = 0. (10.2)
@u@v
A general solution for the above equation is of the form (u, v) = f (u) + g(v).
However, the boundary conditions (10.1) imply that

(u, p(u)) = 0 = f (u) + g(p(u)) =) f (u) = g(p(u)) (10.3)

(u, p(u)) = 0 = f (q(v)) + g(v) =) g(v) = f (q(v)) (10.4)

So we can write the solutions in the presence of the moving mirror alternatively as

(u, v) = f (u) f (q(v)) = g(v) g(p(u)). (10.5)

We can interpret f (u) and g(v) as the right and left moving waves respectively, while f (q(v)), g(p(u))
represent waves reflected by the mirror.
Let us first notice, that the motion of the mirror naturally induces a Doppler shift for the incoming
rays. To see this let us consider two rays, at advanced times v and v + v, approaching the mirror.
The corresponding bounced solution is characterized by a retarded time u and u + u. From the
parametric equation of the mirror we have

v + v = p(u + u) ' p(u) + p0 (u) u + O( u2 ). (10.6)

121
Thus v = p0 (u) u. The frequency of the wave is inversely proportional to u, therefore

1 p0 (u)
!/ ' . (10.7)
u v
In the case of a static mirror p(u) = u there is no Doppler shift in the frequency, however e.g. for a
moving mirror in uniform rectilinear motion for p(u) = u (with  =const) we get redshift or blueshift
for  < 1 or  > 1 respectively (as the mirror moves indeed with velocity V = ( 1)/( + 1)).

10.1.2 Quantum field theory with moving boundaries

The quantization of the field (u, v) can be achieved as usual. First, we recall that, without the mirror,
the left and right going modes are
i!v
g⇠e , ! > 0 (left going)
i!u
f ⇠e , ! > 0 (right going)

Similarly, in the presence of the mirror from (10.5) we have

1 h i
u(in)
! =p e i!v
e i!p(u)
(10.8)
4⇡!
1 h i
u(out)
! =p e i!u
e i!q(v)
. (10.9)
4⇡!
These are two possible bases for the mode expansion of (u, v) and in general we do expect that by
bouncing o↵ the mirror the reflected components can be a mixture of positive and negative norm
modes of the free Minkowski base. This mode mixing, as we have seen, then leads to the conclusion
that particle creation must generically occur.
In 1+1 dimensions it is possible to show [79] that the energy flux Tuv is given by the so-called
Schwarzian derivative of p(u)
" ✓ ◆ #
1 p000 3 p00 2
F (u) = hTuu i = hTxt i = . (10.10)
24⇡ p0 2 p0

In Cartesian coordinates the trajectory can be equivalently written as

8
>
<x = z(t)
umirror = t z(t) (10.11)
>
:
vmirror = t + z(t)

thus p(u) = tmirror (u) + z(tmirror (u)). In terms of z(t) the flux becomes
... 2 s
z (z̈ 1) 3ż z̈ 2 ↵˙ 1 + ż
F (u) = 4 2
= , (10.12)
12⇡(ż 1) (ż + 1) tmirror 12 (1 + ż)3

z̈
with ↵ ⌘ (1 ż) 3/2 being the proper acceleration of the mirror.

This result is also interesting because it also shows that that for trajectories of constant acceleration
↵ the flux of particles vanishes, F (u) = 0 [79]. This is a case where the Bogolyubov coefficient is in
general non-zero but nonetheless the stress energy tensor vanishes [80].

122
Figure 10.1: Penrose diagram representing the motion of a mirror in Minkowski space. The curve
delimiting the shaded region represents the trajectory of the mirror. The two blue curves represent
two light rays reflected by the mirror and ending up on IR+ .

10.1.3 Special case: the exponentially receding mirror

We specialize now in the case depicted in Fig 10.1, where the mirror starts at ◆ and reaches asymp-
totically IL+ with advanced time v0 .
An appropriate basis for an observer on I is given by the incoming modes of (10.8) that for the
reader convenience we rewrite below
1 h i!v i
u(in)
! = p e e i!p(u)
. (10.13)
4⇡!

As depicted in Fig 10.1, the modes that can reach IR+ are those leaving I with v < v0 and are
given by purely right going modes plus reflected modes

(out) 1 h i!u i
u!,R = p e ✓(v0 v)e i!q(v) , (10.14)
4⇡!

whilst on IL+ we have only those modes that left I at v > v0 and hence were not reflected by the
mirror
(out) 1
u!,L = p e i!v ✓(v v0 ). (10.15)
4⇡!
Therefore, the appropriate expansions for the field are:
Z +1 h i
†(in) ⇤(in)
(u, v) = d! â(in)
! u (in)
! + â ! u ! (10.16)
0
Z +1 h i
(out) (out) †(out) ⇤(out)
(u, v) = d! â!,L u!,L + â!,L u!,L + [L ! R] . (10.17)
0

(in)
The vacuum state |0in i on I is by definition the state annihilated by all the operators â! , while
+ (out)
|0out i on IL,R is annihilated by all the operators â!,L,R . In the Heisenberg picture, let us impose that
the quantum state associated to the whole spacetime is |0in i at I . We want then to describe the
evolution of the destruction and creator operators via the Bogolyubov coefficients so to see how the
same state appears on I + .

123
 The last statement can be translated in the condition that there will be a particle creation by the
moving mirror if any of the Bogolyubov coefficients will be non zero. I.e. we need to check if
⇣ ⌘ 2
L,R 2 ⇤(out)
| !,! 0 | = u!0 (L,R) , u(in)
! 6= 0

L |2 = 0. This is a direct
Let us start by noticing that it is quite straightforward to see that | !,! 0
+
consequence of the fact that all the rays reaching IL have not been scattered o↵ the mirror, hence
there is not reason for mode mixing to occur. On the contrary, on IR+ things are more interesting,
R |2 we should try to calculate.
since also reflected modes can arrive there. So it is definitely | !,! 0

To do so, we need to consider a specific form for the trajectory. So let us choose
⇥ ⇤
p(u) = ✓( u)u + ✓(u) v0 Ae u , (10.18)

meaning that for u < 0 we have astatic mirror while for u > 0 we have an exponential recession. In
this specific case the coefficients in the Bogoliubov transformation can be worked out explicitly, and
the final result turns out to be
⇣ ⌘ ✓ ◆
R 2 ⇤(out) (in) 2 e ⇡!/ i! 2
| !!0 | = u!R , u!0 = 2 0 1+ , (10.19)
4⇡ !! k

where (z) is the Gamma function which is defined for a complex number z 2 C whose real part is
positive R(z) > 0 Z 1
(z) = e x xz 1 dx . (10.20)
0
⇡
Let us recall the properties (1 + z) = z (z) and (z) (1 z) = sin(⇡z) we obtain
◆✓
R 2 e ⇡!/ i! 2
| !! 0 | = 2 0 1+
4⇡ !! 
✓ ◆ ✓ ◆
e ⇡!/ i! i!
= 2 0 1+ 1
4⇡ !!  
✓ ◆ ✓ ◆
e ⇡!/ i! i! i!
= 2 0 1
4⇡ !!   
e ⇡!/ i! ⇡ e ⇡!/ ! ⇡
= =
2 0
4⇡ !!  sin  i!⇡ 4⇡ !!  sinh !⇡
2 0

1 1
= . (10.21)
2⇡k! 0 e2!⇡/ 1
where we have used sin(ix) = i sinh(x) and the expression of sinh in terms of exponentials.
Therefore the number of particles measured by an observer on IR+ is
Z +1
(out) 1 1 d! 0
h0in |N̂!R |0in i = . (10.22)
2⇡ e2!⇡/ 1 0 !0

The presence of a logarithmically divergent integral is due to the ill normalization property of Fourier
modes (the divergence could be cured by using properly normalizable wave-packets).
We conclude that an exponentially receding mirror induces a thermal spectrum with a temperature
T = 2⇡
on IR+ . The parameter  represents the acceleration of the mirror, and is the analogous of
the proper acceleration of a Rindler observer.

124
10.2 Hawking Radiation
In the collapse of a star into a black hole one would expect the particle production to be restricted
into a transitory time window associated with the time-varying geometry. As soon as the black hole
has settled down, it would be reasonable for such radiation to disappear. Surprisingly enough, this
consideration turned out to be wrong. Hawking proved [81] that a thermal spectrum continues to
radiate for an indefinitely long time from any stationary black hole.
By recalling the results of the previous lectures about black holes thermodynamics, this fact should
come as no surprise. Indeed, we already argued that black holes should radiate, at a quantum level,
with a characteristic temperature related to their surface gravity. In this section we want to show
explicitly this relation also by exploiting the analogy with the just seen case of an exponentially
receding mirrors.
We will work in 3 + 1 dimensions with a massless scalar field (representing radiation)1 in a
spherically symmetric collapsing star geometry. The most general form of the metric in this context
is (note that we are back to the relativists’ signature ( , +, +, +))
dr2
ds2 = C(r)dt2 + + r2 d⌦2 , (10.23)
C(r)
where C(r) is a function of r only (and possibly of the conserved quantities M, Q, ⇤) and the angular
variables are ✓ and '. In this geometry, the Klein-Gordon equation takes the form
  ✓ ◆
1 @2 1 @ 2 @ 1 1 @ @ 1 @2
⇤ = + r C + sin ✓ + = 0. (10.24)
C @t2 r2 @r @r r2 sin ✓ @✓ @✓ sin2 ✓ @'2
A
Pgeneric solution can be split into a (t, r) and (✓, ') in a spherical harmonics decomposition (t, r, ⌦) =
m m
ml l (t, r)Yl (⌦), where Yl (✓, ) are eigenfunctions of the angular momentum operator (second
squared bracket in (10.24))
✓ ◆
1 @ @Ylm 1 @ 2 Ylm
sin ✓ + = l(l + 1)Ylm (10.25)
sin ✓ @✓ @✓ sin2 ✓ @'2
Obviously, we are mostly interested in the equation for l (t, x)

1 @2 l 1 @ @ l(l + 1)
2
+ 2 r2 C l = 0. (10.26)
C @t r @r @r r2
It is convenient to introduce, as previously done for the Schwarzschild solution, the Regge-Wheeler
coordinate x from the di↵erential relation
dr
dx = . (10.27)
C(r)
@ @r @ @
In this way we have @x = @x @r = C @r and, by defining l = l r, we can simplify equation (10.26)
further. Indeed we have for the term in square brackets
@ l @ l @ l
r2 C = r2 = C l +r . (10.28)
@r @x @x
Hence equation (10.26) is written as

1 @2 l 1 @ @ l l(l + 1)
2
+ 2 C l+r l =
rC @t r C @x @x r3
" #
1 @2 l 1 @C @r @ l @ l @2 l l(l + 1)
+ 2 l C +C +r 2 l =
rC @t2 r C @r @x @x @x @x r3

1 @2 l 1 @2 l C 0 l(l + 1)
= 2
+ 2
C l l = 0,
rC @t rC @x r r3
1
The final result can be proved to hold even in presence of any type of field (vector and spinor fields for example).

125
 @C
where C 0 = @r . Thus, the above equation in equivalent to a wave equation in a 1 + 1 flat-space

@2 l @2 l
+ + Vl (r) l = 0, (10.29)
@t2 @x2
with potential Vl (r) defined as 
C 0 l(l + 1)
Vl (r) ⌘ C + . (10.30)
r r2
This function approaches zero for r ! +1 and also close to the horizon H+ (recall the form of C(r)
in Schwarzschild). Hence, Vl (r) represents the potential barrier a particle has to climb in order to
escape from the gravitational attraction of the star.
The analogy with the moving mirror situation is now clearer: for Vl (r) = 0, equation (10.29) is
equivalent to (10.2), furthermore the analogy holds also in terms of the global structure of the two
spacetimes (e.g. see Fig 10.1 and Fig 10.2). In the case of the black hole geometry, the mirror is to be
thought at the origin r = 0 and our problem is to find the analogue of the relation v = p(u) relating
coordinates on I and I + .
To do so, we need to match the coordinates inside and outside the collapsing star. Inside, the
metric takes the generic form for a time dependent spherically symmetric metric

ds2 = (⌧, ) d⌧ 2 + d 2
+ ⇢2 (⌧, ) d⌦2 , (10.31)

where and ⇢ can be chosen to be regular everywhere including H+ .

Additionally, we can also introduce a set of null coordinates
(
U =⌧
(10.32)
V =⌧+

and fix the center at = 0 (i.e. U = V ).

On the outside we also have the usual set of null coordinates u = t x and v = t + x. However,
we need to work with coordinates which are everywhere well behaved (includinhg H+ ) in order to
perform the matching, and in this case these are the usual Kruskal coordinates:
(
U = e u
(10.33)
V = ev

where  is the surface gravity of the black hole. This set of coordinates is regular everywhere.
To perform the matching between (U, V ) and (U , V ) we proceed as follows. We take two null rays
H and . The first, H , is the generator of the horizon H+ and is extended back into the past where
it hits I at some value of v that we fix being v = v0 . The second geodesic, , on the other hand is
“infinitesimally close” to H and starts from I at some v < v0 . See Fig 10.2. Equivalently, these
two rays can be seen as propagating back from I + (and thus being described by u = +1 and u = u
respectively). We assume they are free plane waves on I + but, possibly, a collection of modes on
I .
The distance between H and can be written, in terms of U , as dU . This must correspond to
a dU = (U )dU in terms of the inside coordinates, with (U ) is determined by the details of the
collapse. Additionally, in terms of V and V we have dV = ⇣(V )dV = ⇣(v)dv where, again, ⇣(V ) is
determined by the details of the collapse. The second step in the last equation follows from the fact
that the coordinate v is also well behaved on the horizon and thus V can be replaced with it.
Close to H+ we have x ! 1 thus u = (t x) ! +1 and U ! 0. At the same time, v
approaches v0 . Hence: (
dU ' (0)dU
(10.34)
dV ' ⇣(v0 )dv

126
Figure 10.2: Penrose diagram representing a collapsing, spherically symmetric, star. The geodesic H
lies on the future event horizon H+ while the geodesic is infinitesimally close to H and starts from
I with advanced time v.

Finally, we recall that at the center of the star = 0, so U = V and thus dU = dV (by the fact
that the geometry is stationary). This makes us conclude that

⇣(v0 )dv ' (0)dU . (10.35)

This relation can be integrated:

Z v Z U
(0) (0)
⇣(v0 )dv 0 = (0)dU 0 =) v = v0 + U =) v = v0 e u
⌘ p(u). (10.36)
v0 0 ⇣(v0 ) ⇣(v0 )

Hence, v = p(u) takes the same form as in the case of the exponentially receding mirror (10.18) with
A = (0)/⇣(v0 )!
Therefore, by following the previous steps we can conclude that for Vl (r) = 0 we must have
Z +1
+ 1 1 d! 0
h0I |N̂!I |0I i = , (10.37)
2⇡ e2!⇡/ 1 0 !0
+
where |0I i is the vacuum on I and N̂!I the number operator on I + . Let us note again that this
this thermal behaviour of the black hole is rooted in  which here controls the amount of “peeling” of
geodesics close to H+ , as shown in (10.36).
Up to now we have neglected the potential. The presence of a potential barrier inevitably points
toward a reflection of some of the rays, while others can manage to tunnel through. In such a case
the wave has the form
(out) i!u (1) (2)
u!l (u, v) ⇠ e + u!l + u!l (10.38)
(1)
Here the function u!l ⇠ ✓(v0 v)e i!q(v) Kl (!) describes transmitted modes that crossed the potential
barrier and reach I + after bouncing/passing through the center of the star, with Kl (!) being an

127
 (2)
attenuation coefficient. The function u!l contains instead the reflected part of the wave, i.e. modes
that bounced o↵ the potential barrier and got to I + without entering the collapsing star . Obviously,
these are unprocessed modes which do not contribute to the Bogolyubov coefficients. So in the end
(in) (out)⇤ (in) (1)⇤
| !!0 |2 = |(u!0 l , u!l )|2 = |Kl |2 |(u!0 l , u!l )|2 , only the transmitted modes contributes.
The upshot is that the correct number of particles observed on I + gets corrected by a function
of the attenuation coefficient
Z +1
I+ 1 l (!) d! 0
h0I |N̂! |0I i = , (10.39)
2⇡ e2!⇡/ 1 0 !0

where l (!) is the so-called gray factor, associated to the attenuation Kl (!), that allows for the black
hole to be at thermal equilibrium. Indeed, to have detailed balance we need the coefficient

radiation emitted in (!, l)

R= , (10.40)
radiation absorbed in (!, l)

to be independent on l (!). Since the absorbed radiation does depend on this function, it must be the
case also for the emitted radiation to have the same dependence, in such a way that l (!) is factored
out in R.
Before moving to a brief descriptions of the issues raised by the discovery of Hawking radiation it
is perhaps worth discussing in some detail the nature of the quantum states that can be associated
with a black hole spacetime.

• Unruh state: this is the state we just discussed above. It is defined as the that state which is
vacuum on I and at H+ for a freely-falling. For an observer at infinity this is a thermal state
due to Hawking radiation. It is the only non-singular state, vacuum on I , on a spacetime
where a black hole is formed by a collapse (like that shown in Fig 10.2). However, the case of
an eternal black hole (Fig. 6.6) this state is singular on the past horizon H (we shall see why
when we shall discuss explicitly the semiclassical collapse).

• Hartle–Hawking state: This is a state describing a thermal equilibrium between a black hole and
a surrounding thermal bath at the Hawking temperature. As such it is defined as the state which
is thermal at the Hawking temperature both on I + and I . Being a state of equilibrium it is
the natural state for eternal black holes (Fig. 6.6) and indeed is the only one which is non-singular
everywhere in these spacetimes.

• Boulware state: This is defined as the global state which is vacuum both on I + and I . It is
a state appropriate for the exterior of stationary spherically symmetric stars. The annihilation
operator is defined with respect to the timelike Killing vector @t of this spacetime. Noticeably
the expectation-value of Tµ⌫ on this state diverges at the horizon [45]
Z
⌫ 1 ! 2 d!
B h0| Tµ |0iB ⇠ diag( 1, 1/3, 1/3, 1/3) (10.41)
2M 2 e8⇡! 1
2⇡ 2 1 r

We see that such configuration due to the pre-factor (1 2M/r) 2 develops a huge vacuum polarisation
close to the horizon and that the expectation value of the stress energy tensor appears to correspond
to the subtraction from the vacuum of a black body radiation at the Hawking temperature [75].

128
10.3 Open issues with Hawking radiation
Although apparently surprising, Hawking result follows from minimalistic assumptions about General
Relativity and quantum theory for free fields in curved spacetime and the presence of interactions
does not a↵ect much the conclusion drawn in the previous section (actually, we shall see later that the
thermality at the Hawking temperature can be deduced on purely geometrical grounds). However, as
it is often the case in physics, the discovery of Hawking radiation led to new questions which we are
going to pinpoint below.

10.3.1 Trans-planckian problem

Let us work in ingoing Eddington–Finkelstein coordinates. Let us evolve back in advanced time v a
wave packet along a u = const. ray, from I + to I . For a freely falling observer such wave packet
peak frequency goes as the inverse of the advanced time in Kruskal coordinates, !peak = V 1 = e v . 2

Figure 10.3: Ray-tracying backward in time a Hawking packet at u = const.

Therefore, in the interval v = v2 v1 > 0 we have a blueshift

!p (v1 )
' e v
. (10.42)
!p (v2 )
For a black hole with the mass of the Sun,  = (4M ) 1 and for a time interval v ⇠ 2sec we obtain
!p (v1 )/!p (v2 ) ⇠ 105 . Hence while ray tracing the wave packet backward in time towards the horizon,
the peak frequency undergoes an exponentially fast blueshift and, eventually, it reaches frequencies
above the Planck mass ! 1019 GeV, i.e. in principle beyond the regime of validity of out QFT in
CS framework.Alternatively, we can say that the transplanckian problem consist in the fact that the
IR Hawking quanta reaching I + seems to originate from far UV modes in the vacuum at H+ or
equivalently on I .
Of course, frequencies are not per se covariant objects and one might think that boosting to
a suitable Lorentz frame can remove the issue. However, we actually talking about UV s-waves
emanating from the black hole and there is no way those can be boosted away.
While, in Hawking’s original derivation this issue was not directly touched it was soon recognised
by a non-negligible community, led by Bill Unruh and Ted Jacobson. It was partially the need to
test the robustness of Hawking radiation that gave since the early nineties a strong impulse to an
alternative and quite original framework, that of Analogue Gravity [43].

Analogue Gravity
And I cherish more than anything else the Analogies, my most trustworthy masters. They know all
the secrets of Nature, and they ought least to be neglected in Geometry. Johannes Kepler.

2
The Kruskal coordinates strictly speaking are not everywhere associated to freely falling observers. However, in
proximity of the horizon of a stationary black hole and at infinity T and X do describe proper time and distance for
geodesic observers.

129
 In 1981, Unruh put forward a powerful analogy between curved spacetime and classical hydrody-
namics, showing how di↵erent physics can be described mathematically in the same way. Consider for
instance a perfect fluid (irrotational and non-viscous) satisfying the continuity and Euler equations
(
⇢˙ + r · (⇢v) = 0
⇥ ⇤ (10.43)
⇢ @v @t + (v · r) v = rp

where ⇢ is the density, p the pressure, v the velocity, and r ⇥ v = 0 =) v = r being the so called
velocity potential of the fluid. For a barotropic fluid, we also have the equation of state ⇢ = ⇢(p)
dp
with c2s ⌘ d⇢ . Albeit we considered the forced induced just by pressure, it is straightforward to add
external forces, at least conservative body forces such as Newtonian gravity ⇢r or dissipative e↵ects
induced by viscosity.
If we perturb this set of equations around a background ( 0 , ⇢0 , p0 , v0 ), then in terms of the
perturbations ( 1 , ⇢1 , p1 , v1 ) the equations (10.43) can be recast into a single second order di↵erential
equation  ✓ ◆  ✓ ◆
@ ⇢0 @ 1 v0 ⇢0 @ 1
+ v 0 · r 1 = r · ⇢0 r 1 + v 0 · r 1 . (10.44)
@t c2s @t c2s @t
Equivalently, this can be recast into the simple form:

⇤ 1 = 0, (10.45)

where ⇤ is the D’Alembertian computed with the auxiliary metric


⇢0 (c2s v20 ) v0j
gµ⌫ = 2 . (10.46)
cs v0i ij

Furthermore, one can realize that the physics of the system can mimic the one of a black hole. For
instance, we can reach g00 = 0 for c2s = v02 meaning that there is a trapped region for the waves.
To make this example more tangible, we can consider the hole in a bathtub. The background
fluid has a larger velocity v02 as we go closer to the hole. Therefore eventually we reach a point where
indeed g00 = 0 and this equation delimits a region below which waves propagating on top of the fluid
must flow downstream (see Fig 10.4). The analogy can be made even more precise by considering the
Schwarzschild metric in Painlevè-Gullstrand coordinates, that are defined as
8  ✓ q ◆
> p
<tPG = t ± 4M arctanh 2 2M r 2 2M r rPG = r
p (10.47)
>
:dtPG = dt ± 2M/r dr +=outgoing PG, -=ingoing PG
(1 2Mr )

The metric takes the form

✓ ◆ r
2 2M 2M
ds = 1 dt2PG ± dr tPG + dr2 + r2 d⌦2 , (10.48)
r r

It should be noted that the + sign corresponds to a coordinate patch that covers the usual asymptotic
region plus the region containing the future singularity of the maximally-extended Schwarzschild
spacetime. Thus, it covers the future horizon and the black hole singularity. On the other hand the
sign corresponds to a coordinate patch that covers the usual asymptotic region plus the region
containing the past singularity. Thus it covers the past horizon and the white hole singularity.
We can now see that cast in this form the Schwarzschild metric is indeed of the acoustic form
(modulo a conformal factor) and the analogy with fluid case can be made evident if we make the
choice of the flow r
2M
|v0 | ! , ⇢ ! r 3/2 , (10.49)
r

130
Figure 10.4: Representation of a hole with a trapped region (red circle). As the throat shrinks, wave
packets need larger and larger group velocities to reach the flat region on top (from [43]).

where the condition on the density is imposed by consistency with the continuity equation. 3 Let us
stress that many other black hole-like geometries can be designed by choosing suitable fluid flows. The
Schwarzschild example above is just one among many (and not the most practical one to realise in a
laboratory). All one needs is a fluid flow with an acoustic horizon as shown in Fig. 10.5

Figure 10.5: A moving fluid can form an “acoustic horizon” when supersonic flow prevents upstream
motion of sound waves (from [43]).

To push the analogy even further, one could ask if there is an analogous of the Hawking radiation for
the fluid. The answer is yes, and involves the phonons: the quantum fluctuations of the fluid. However,
testing experimentally the quantum properties of a fluid is quite a hard task. Nonetheless, it is possible
to simulate hHawking radiation is some systems like superfluids or Bose-Einstein condensates (BEC),
where the coherence time is much longer than in a conventional fluid.
Within this analogy, the transplanckian problem can be seen in the following way. Below a char-
acteristic energy scale K, the collection of atoms behaves as a fluid (or BEC), but as soon as we probe
higher energies we start to resolve the atomic degrees of freedom of the system. This very fact is
modelled by modifying the dispersion relation of the phonons to
✓ ◆
2 2 2 kn
! = cs k + ⌘ n 2 . (10.50)
K

where ⌘ s a dimensionless constant which can be negative or positive depending on the analogue
system used. This modified dispersion relation (10.50) would introduce a Lorentz violating term in
the equation of motion that have significant impact on the solutions and on their behaviour close to
the horizon H+ . Actually, if d! +
dk > 1, then the group velocity of a wave close to H can exceed the
+
speed of light/sound, and a particle back-traced starting from I could be seen to come from behind
3
In order to satisfy the Euler equation one needs in this case also to add a suitable external potential, but this is not
relevant for our discussion here.

131
the horizon. If, however, the modification (10.50) is such that d!
dk < 1 then the particle would change
its group velocity sign close to the horizon and when back-traced it would be seen to come from I
instead of piling up at H+ . See Fig. 10.6.

Figure 10.6: Modes associated to analogue Hawking radiation for ⌘ positive (solid lines) or negatives
(not all the solutions for n > 2 dispersion relations are showed).

It has been shown that Hawking radiation is robust against di↵erent UV completions of General
Relativity. More precisely, possible corrections to the Hawking temperature would be of order ⇤ ⌧ 1
where ⇤ is parametrically related to K.

A worth-noting result in this context is the Lorentz violating theory by Petr Hořava formulated
in 2009. He proposed a Lorentz violating UV completion of GR with the following dispersion
relation for gravitons ✓ ◆
k4 k6
! 2 = c2 k 2 + 2 + 4 , (10.51)
M⇤ M⇤
which gives a power counting renormalizable theory of gravity. In this formulation, black holes
solutions are endowed with a new impenetrable horizon called universal horizon. As the name
suggest, not even particles with speeds approaching infinity are able to cross such horizon.

Finally, let us stress that the lesson from Analogue gravity is not that one needs UV Lorentz
breaking in order to solve the transplanckian problem. On the contrary it shows, in a well defined
system where the UV completion is under control, that the Hawking radiation is rather robust, or if
you wish insensitive, about the details of the UV completion of your theory. This is good news for its
reliability as a universal prediction.

10.3.2 The information loss problem

Hawking radiation predicts that a black hole will evaporate in a finite time. Therefore, a black hole
cannot be eternal, and the Penrose diagram of a spacetime in which a black hole is formed at early
times and later on completely evaporates will be of the form shown in Fig 10.7. In this spacetime, after
a certain time, the singularity disappears. At the quantum level, on an early Cauchy hypersurface ⌃1 ,
we can start from a state | i and after the black hole is completely gone we end up on a late Cauchy
hypersurface ⌃2 with a thermal state |Ti. This is a problem because quantum evolution is unitary,
thence
| (t)i = eiĤt | (0)i, (10.52)

132
Figure 10.7: Spacetime of an evaporating black hole. After a while the singularity disappears. The
quantum states | i and |Ti are defined over spacelike slices of the manifold.

where Ĥ is the Hamiltonian operator of the system but the above argument seems to imply that a
pure state | i can in principle evolve in a thermal, mixed, state |Ti. Let’s try to formalize this more
precisely.
The density matrix of a system is an operator
X
⇢ˆ = Pn | n ih n | (10.53)
n
P
satisfying Tr ⇢ˆ = n Pn = 1. The above general state is what we call a mixed state.
P In general for a given observable  one has that the expectation value is <  >= Tr[ˆ ⇢Â] =
P
n n n h |A| n i. I.e. a quantum system is in general a superposition of several quantum states and a
measurement will find it a particular state | n i with probability Pn . The outcomes of measurements
on mixed states are di↵erent pure quantum states eigenvalues with probability distribution Pn .
On the other hand a pure state is always of the form ⇢ˆ = | n̄ ih n̄ | and hence for these states the
outcome of a given measurement is pre-determined: a complete measurement of the system provides
always the same answer. Moreover, for pure states P we also have ⇢ˆ2 = | n̄ ih n̄ || n̄ ih n̄ | = ⇢ˆ so that
Tr ⇢ˆ2 = Tr ⇢ˆ = 1 while for a mixed state Tr ⇢ˆ2 = n Pn2 < 1.
The concept of “information” in the system is captured by the von Neumann entropy S ⌘
Tr (ˆ⇢ log ⇢ˆ). For pure states S = 0, whilst for mixed states we could have a situation of the sort:
 1
0
⇢ˆ = 2 1 , (10.54)
0 2

with states |0i and |1i. In this case S = log 12 = log 2 > 0.
In quantum mechanics, starting from a pure state there is no way, through unitary evolution, to
end up with a mixed state. Indeed, for Û Û † = 1 and | i0 = U | i then ⇢ˆ0 = U | ih |U † , meaning that
2
⇢ˆ0 = U | ih |U † U | ih |U † = U | ih |U † = ⇢ˆ0 . (10.55)

So if | i is a pure state, then | i0 = U | i will also be a pure state. If we are not making measurements
on the system (no wave-function collapse), nor we are tracing over part of the quantum system, then

133
there is no physical process in standard quantum mechanics that could bring a pure state | i to a
thermal (mixed) state |Ti without violating unitarity. Nonetheless, as we already stressed, it seems
that black holes can exactly do this.
Here we would like to highlight some notable attempts to solve this problem.

• Hawking tried to solve this problem by introducing a non-unitary super-scattering matrix [82],
/̂ an operator such that ⇢ˆ0 = S
S, /̂ ⇢ˆ with S
/̂ 6= S S̄ where S is the usual scattering matrix and S̄
is complex conjugate. This idea, however, was soon abandoned due to the difficulty to built a
consistent quantum mechanics consistent with it [83].

• Another possibility is that information is released in the quantum gravity regime M ⇠ MP l at

the end of the evaporation. However, if the black hole is this small there seems to be not enough
energy left to encode all the missing information. Indeed, it was noticed by Don Page [84]
that any decrease in the entanglement entropy of the Hawking radiation (a process often called
“purification”) has to start happening before half of the original black hole entropy is left (i.e.
the black hole area is halved) otherwise there would be no more enough internal degrees for
purifying the total Hawking flux by the end of the evaporation process at ⌧ ⇠ M 3 . The time
at which this is exactly realised is tPage ⇠ Rh3 /lP2 l and is called the Page-time. In this approach,

Figure 10.8: Behaviour of the Hawking radiation entanglement entropy in the standard (Hawking)
scenario and in Page’s scenario.

the entropy of the black hole S is interpreted as Boltzmann entropy, and then there is a time
(the Page-time) tpage at which

S(t0 ) A(t0 )
S(tPage ) = , A(tPage ) = , (10.56)
2 2
where A is the area and t0 is the time of formation of the black hole. For t < tpage , a perturbative
analysis should not be able to find any information leaking out; then, for t > tpage , as the black
hole starts to shrink and entropy decreases, we have lots of information coming out.

• A third, albeit no more very popular idea, is that of a remnant: the evaporation could stop
at a certain point leaving a residual object with a huge entropy. A common objection is that
since there are infinite numbers of ways of forming black holes and letting them evaporate, this
remnant must have an infinite number of quantum states in order to encode the information in
the initial state. Since the remnants carry mass, it must be possible to pair-produce them in a
gravitational field and the total pair-production rate should be proportional to the number of
remnant species, and therefore infinite. Thus, it would seem that remnants can be experimentally
ruled out by the observed absence of copious pair-production.

134
 However, it is clear that the standard argument that infinite pair-production is inevitable for all
types of remnants is too naive, and arguments/calculations (e.g. in Euclidean quantum gravity)
have been given that in some theories the pair production rate is finite.
Nonetheless, it seems hard to devise a conservation law forbidding remnant decay since that
would also forbid remnant formation while, in the absence of a conservation law, it is hard to
understand why matrix elements connecting a massive remnant to the vacuum plus outgoing
radiation should be exactly zero. Alternatively one could say that a process leading to the
formation of an eternal remnant would violate CPT symmetry.

• Another option is the so-called firewall : one may try to solve the information loss by allowing
a slow leaking of information through the Hawking radiation itself. Here the argument is based
on the fact that the following four, seemingly solid, assumptions

1. Black holes form and evaporate, as seen by observers close to I + , in a way fully compatible
with standard QFT (no unitarity violation)
2. Outside the (stretched) horizon of a massive black hole, physics can be described by a set
of semiclassical equations
3. Bekenstein–Hawking entropy does determine the black hole density of states for observers
outside the stretched horizon
4. A freely falling observer experiences nothing out of the ordinary (no new physics) at horizon
crossing and until the singularity is met.

Albeit each of these assumptions is per se quite conservative, one can show that they are not
reciprocally compatible [85]. The basic argument is that for the outside evolution to be unitary
one has to require that the Hawking radiation starts to be purified no later than the Page time.
However, such a purification (decrease in the Hawking radiation entanglement entropy) requires
an entanglement between late and early Hawking quanta (more precisely each late Hawking
quanta will have to be entangled with the whole system of early Hawking quanta). This is
required if we want that the entablement entropy of the sum of Early and Late Hawking quanta
say SEL is smaller that SE (which is what it should happen if one is on the descending branch of
the Page curve in Fig. 10.8). However, we have seen before when we analysed the Unruh e↵ect
that in order to have a regular state everywhere, and in particular at the horizon, the vacuum
state should be of the form (9.36).
YX
|0iM / e ⇡!n |nL i ⌦ |nR i
! n

The same consideration hold for the Unruh vacuum associated to a black hole formed by a
collapse. Hence the late Hawking quanta have to be entangled with the early ones (for any hope
to preserve unitarity) and at the same time to their ingoing partners (request of “no drama” at
the horizon). But here it comes the problem: this violates the so called monogamy of quantum
entanglement for the late Hawking quanta.
We can bring this to focus by using the property of strong sub-additivity: if one divides a system
in three subsystems A, B and C (so that the Hilbert space factorises) then sub-additivity of
entropy requires that SAB + SBC SA + SC . In our case let us take A as the early Hawking
quanta, B as the late Hawking quanta and C as the late Hawking partners. In this case we know
that SBC = 0 as the Unruh state is just a squeezed state of the initial Minkowski vacuum on I
(and hence it is pure if one takes both B and C). But we also know that SB ⇠ O(1) given that
Hawking quanta are thermally populated. Also SC = SB (they form a pure state). However,
strong sub-additivity would then imply SAB SA + SC > SA , i.e. exactly the contrary of what
we need for decreasing the entanglement entropy! (and it can be shown that small corrections

135
 a black hole geometry to a white hole geometry can explain the change in the causal structure of the
space-time by allowing for quantum effects to ‘tilt’ the direction of the light-cones, effectively turning
the black hole horizon into a white hole horizon (see Figure 2).
As we shall see later, the lifespan of the horizon is model dependent but, for distant observers,
is usually long enough to ensure compatibility with astrophysical observation of black hole candidates.
At early times the models are well described by a classical GR collapse solution. Quantum corrections
exist for a finite time, thus restoring classicality at late times.

QG tB
{ r

rb

Figure 2. Finkelstein diagram for the black hole to white hole transition. The grey area enclosed within
dashed lines represents the region where quantum effects are important (QG). The grey area within
solid lines represents the trapped region in the exterior space-time. The solid thick line rb represents the
Figure 10.10: Bouncing black hole spacetime by
boundary of the cloud. The solid thin vertical line represents the horizon in the exterior region. In this
case the transition is completely symmetric in time. The bounce occurs at the same time t B for all shells
quantum gravity: the horizon is replaced by a
(as in the homogeneous case). An horizon grazing photon (thin curved line), stays in the vicinity of the
trapped region plus a region where classical grav-
horizon until right after t B . The lifetime of the white hole is the same as the lifetime of the black hole.

Figure 10.9: Regular spacetime: the black hole ity does not hold and quantum e↵ects dominate.
horizon is replaced by a trapped region (shaded Such region has to extend outside the trapping
area) and the singularity is removed. horizon. (From [87])

cannot fix this problem). It is then clear that if we want to insist that the black hole evaporation
has o be a unitary process at least one of the above assumptions must be ditched.
One popular conjecture is that it is the fourth assumption the one to be wrong — alternative one
can give up one of the other assumptions (e.g. introducing non-locality, or UV Lorentz violation
or violation of standard EFT at large distances). The violation of the fourth assumption implies
that B and C do not combine in a pure state (the Unruh vacuum). However, this is the only
non-singular state that can be compatible with the black hole casual structure, deviation from it
will lead to arbitrary large excitations on the horizon which would take the form of a “firewall”
(a highly energetic “wall”) at the horizon of any black hole (even very large ones after a sufficient
long time).

• Another possible way out is conjecturing that the problem lies in the scenario depicted by
Figure 10.7 . Indeed it is easy to see that our problem are stemming from the causal structure
of the full evaporation scenario. In particular by the presence of a full fledged event horizon.
Can we coincive scenarios where only a trapping horizons form? One possibility is that quantum
gravity regularise the inner singularity in black hole solutions so leading to scenarios like regular
black holes (see e.g [86] and references therein) as depicted in Fig 10.9 or a bounce (see e.g. [87]
and references therein) as in Fig 10.10. It might be the case that, somehow, quantum gravity
removes the singularity from spacetime. Additionally, since J (I + ) covers now the whole
manifold, we don’t have an event horizon anymore, but rather a transient trapped region. In
this case of regular black holes the main issue seems to reside in the stability of these objects
given that the presence of a inner horizon seems to be associated to the so called mass inflation
phenomenon by which the horizon could destabilise on very short timescales (at least without
fine tuning the solution) [86]. In the case of black hole bounces one is facing instead the necessity
to be compatible with the astrophysical observation of long living black hole like objects. This
can be achieved by requiring a large di↵erence between the timescale of the bounce for freely
falling observers w.r.t that of static observers at infinity but still shorter of the evaporating time
(⌧ ⇠ O(M 2 ) rather then O(M 3 )). This is a model dependent feature which is still unclear if it
is preferred by computable quantum gravity models.

• Finally, another interesting possibility is the presence of additional symmetries on the horizon
that are not obviously realised. For instance, Bondi, Metzner and Sachs (BMS) have shown that

136
 asymptotically flat spacetimes have a very large, infinite dimensional symmetry group (BMS
group) which is a generalisation of the Poincarè group and acts non-trivially on I + . Associ-
ated to this symmetries (called super-translations and super-rotations) are a series of conserved
charges. Remarkably, it was recently shown that the same symmetries arise also close to the
event horizon of a black hole and that they are associated there to“soft hairs”. It was then
conjectured that this realisation might help resolving the information paradox by encoding in
such conserved “soft” gravitons the information associated to infalling matter and Hawking part-
ners. Upon evaporation these soft hairs conservation would then imply a necessary entanglement
between the black hole and the radiation.
While attractive also this resolution seems not fully consistent yet. To start with it can be
shown that the soft gravitons/hairs decouple from finite energy states and can be factored out.
It is then unclear how these IR feature could end up encoding the UV correlations which seems
needed to reduce the entanglement entropy of the radiation. This scenarios heavily relies on
the symmetries of four dimensional asymptotically flat spacetimes, but we already know that
the information problem can be as well formulated in higher dimensions and in more general
scenarios like for example AdS black holes.

10.4 Black Hole Entropy Interpretations

As already shown, the Bekenstein–Hawking entropy for a black holes is SBH = 4LA2 . What degrees of
P
freedom are responsible for it? Do we have a microscopic interpretation for this entropy?
Let us have a panoramic view of proposals people consider in order to address these questions.

• Boltzmann entropy:
One hypothesis is that SBH corresponds to a “Boltzmann” entropy, meaning that it can be
interpreted as arising from the counting of the miscrostates responsible for the black hole config-
uration. In this proposal SBH = kB log W, where W is the number of microscopic configurations
giving the same M , J and Q. However, where are these degrees of freedom? How can this
proposal address the loss of information problem? These are still open issues in this approach.

• Entanglement entropy:
A second approach is to relate the Bekenstein-Hawking entropy to the entropy of entanglement,
or Von-Neumann entropy Sent = kB Tr ⇢ log ⇢. Indeed, due to the Hawking radiation emitted
by the black hole, the density matrix is mixed and it could yield the right value for the entropy.
With more advanced methods that are beyond the scope of the course, one can compute this
entanglement entropy, and finds Sent = ⇤A2 , where A is indeed the area of the black hole and
⇤ is a UV cuto↵ needed in order to make the result finite. The reason for this divergence is
that null-rays starting from I accumulate on the horizon, in such a way that the overall result
diverges. The aforementioned cuto↵ can be interpreted as the thickness of the horizon: due to
quantum gravity fluctuations one cannot localize the horizon with a precision below the Planck
length. Hence, this motivated to take ⇤ ⇠ MPl so to recover the right value for the entropy.
Unfortunately this is a very rough estimate and more rigorous results are needed to claim an
equivalence with SBH . Indeed, this is the result for a single field living around the black hole,
but what about other SM fields? The idea is that these fields provide additional divergent terms
to Sent . Requiring the result to be finite provides a renormalization condition for the Newton
constant and the possible other constants for extended theories of gravity [88].

• AdS/CFT duality:
The AdS/CFT duality relates a gravity theory living on the bulk AdS-space to a lower dimen-
sional CFT on the boundary. In this context there is a di↵erent notion of entanglement entropy

137
 CFT
°A

Area

AdS d+2

Figure 10.11: Schematic representation of the surface A in AdS.

for a region of area A on the boundary. Given the region A, one takes the minimal surface A
on the bulk subtending A. The entanglement entropy is then found to be related to the bulk
area:
Area( A )
Sent = .
4Gd+2
N
This idea was recenty proposed by S. Ryu and T. Takayanagi, but it is still unclear how it could
be extended beyond the AdS?CFT scenario.

• Black hole entropy as Noether charge:

An alternative interpretation of the black hole entropy is due to Wald and collaborators. Al-
beit it does not per se explain the microscopic origin of the entropy, it has the advantage to
provide a computational tool for generalising the Bekenstein–Hawking entropy to more general
gravitational theories than GR.
Let us start by considering a collection of dynamical fields in D spacetime dimensions, collectively
denoted by , including a metric tensor gab plus other possible matter fields, whose dynamics
is determined by a di↵eomorphism-invariant Lagrangian D-form L = L✏, with ✏ the spacetime
volume element.
Under a generic variation of the fields, we saw that the variation of L can be expressed as a
sum of a bulk term plus a boundary one:

L=E + d⇥( , ) (10.57)

where the (D 1)-form ⇥ is locally constructed out of and . From (10.57) we read that the
.
e.o.m. are E = 0 for each 4 .
In particular one can consider infinitesimal variations along a vector field ⇠, = L⇠ . By
di↵eomorphism invariance, to any vector field ⇠ corresponds a Nother current (D 1)-form

J[⇠] = ⇥( , L⇠ ) ⇠·L (10.58)

which is conserved on shell as indeed

.
dJ[⇠] = d⇥( , L⇠ ) d⇠ · L = E L⇠ = 0. (10.59)

Here we have used that L is exact by construction, and hence also closed dL = 0 and we have
made use of (10.57) and that L = ⇠ · dL + d(⇠ · L) = d(⇠ · L) = d⇠ · L.
4
From now on the dot indicates equalities holding on shell.

138
 The conservation of J for any ⇠ implies the existence of a (D 2)-form Q[⇠] [89]
.
J[⇠] = dQ[⇠] . (10.60)

This is the so called “Noether potential” associated to ⇠.

Q enters in the definition of the conserved charges: indeed the Hamiltonian variation, associated
with the flow of ⇠, over an initial value surface ⌃ with boundary @⌃, is given by [90]
Z
⇥ ⇤
H[⇠] = Q[⇠] ⇠ · ⇥( , ) ; (10.61)
@⌃

it is then natural to identify the variations of the energy E and the angular momentum J~ at
infinity as5
Z
⇥ ⇤
E= Q[t] t · ⇥( , ) , (10.62)
S1
Z Z
⇥ ⇤
J~ = Q[ ~ ] ~ · ⇥( , ) = Q[ ~ ], (10.63)
S1 S1

where S1 is the outer boundary of @⌃, and the last equality of (10.63) follows from the fact
that ~ is tangential to S1 . (Notice that, as usual, the angular charges are defined up to a
conventional minus sign.).
R R
Now, our theory admits a D 1 form B( ) such that ⇠ · ⇥( , ) = ⇠ · B( ) (like e.g. in
GR), then one can define the conserved Hamiltonian charge as
Z
H[⇠] = Q[⇠] ⇠ · B; (10.64)
S1

in particular the angular momentum is exactly the Noether charge at infinity, modulo a sign:
Z
~
J= Q[ ~ ]. (10.65)
S1

Let us now specialise to a generic stationary black hole spacetime. If the field ⇠ is taken to be
the general Killing field generating the horizon (and null on it) ⇠ = t + ⌦ ~ H · ~ , then equation
(10.61) implies the first law of black hole mechanics: let (i) ⇠ be a dynamical symmetry, meaning
.
that L⇠ = 0 for all the ’s, and (ii) be a variation of the dynamical fields around the BH
.
solution, such that solves the linearized e.o.m.; then H[⇠] = 0, iii) we take an initial value
surface with boundary @⌃ = S1 [ B, with B the bifurcation surface of the black hole. Then
given that Z
H[⇠] = E[t] ⌦ ~ H · ~J + Q[⇠] ⇠⇥( , ) (10.66)
B
.
It follows [90, 91] that, being H[⇠] = 0 and ⇠⇥( , ) = 0 on B, one gets
.  ~ H · ~J
E= S+⌦ (10.67)
2⇡
where S is 2⇡/ times the integral of Q over the bifurcation surface:
Z
2⇡
S= Q[⇠], (10.68)
 B
5
E contains also work term contributions from long range fields, such as gauge fields.

139
 Since /2⇡ is the Hawking temperature, one interprets S as the thermodynamical entropy of
the BH6 .
Finally, for a general gravitational Lagrangian eq.(10.68) can be egeneralized as [91]:
Z
abcd abcd L
S = 2⇡ ER ✏ˆab ✏ˆcd ✏¯ , ER = (10.69)
B Rabcd

where ✏¯ is the area element of B and ✏ˆab is the binormal to B.

6
Note however that this identification fails if the dynamical fields have divergent components at the bifurcation surface.
This circumstance occours, for example, in the case of gauge fields, but one can see that in this case the divergences at
the horizon can be gauged out by an appropriate gauge fixing, thus recovering the correct expression for the entropy.

140
 11
Semiclassical Gravity

11.1 Semiclassical Einstein Equations

Phenomena like Hawking radiation are computed assuming quantum matter as test field on the classi-
cal gravitational background. Of course, this is just an approximation and for example the description
of the black hole evaporation requires to take in to account the back-reaction of quantum fields on the
spacetime. Unfortunately, missing a full fledged quantum theory of matter and gravitation one has to
resort to di↵erent kind of approximations to treat this problem.
For instance, if matter is described by some field (x) with stress-energy tensor Tµ⌫ [ , g] in a
quantum state | i, and the metric is considered classical, then there are two interesting objects to
consider:
Gµ⌫ [g] , h |T̂µ⌫ [ , g]| i. (11.1)
Provided that it makes any sense, we can postulate the so called semiclassical Einstein equations

Gµ⌫ [g] = 8⇡GN h |T̂µ⌫ [ , g]| i . (11.2)

Whenever this is a good approximation for the dynamics of the metric gµ⌫ , we say that we are in
a semi-classical approximation. Note that classical solutions are obtained from (11.2) in the limit
~ ! 0, where the classical stress-energy tensor is given by
cl
Tµ⌫ ⌘ lim h |T̂µ⌫ [ , g]| i. (11.3)
~!0

It makes sense to define a quantum stress-energy tensor as

Q cl
Tµ⌫ ⌘ h |T̂µ⌫ [ , g]| i Tµ⌫ ; (11.4)

this quantity is clearly already of order ⇠ O(~).

But is this a justified ansatz? For example, is it consistent to quantise matter while neglecting
quantum fluctuations in the gravitational sector? If we do not do so, we have to consider the metric
as an operator, ĝµ⌫ , and the wave-function (or better wave-functional) as a tensor product | , gi ⌘
| i ⌦ |gi. The operatorial version of the Einstein equations would then have the form

h , g|Gµ⌫ [ĝ]| , gi = 8⇡GN h , g|T̂µ⌫ [ , g]| , gi. (11.5)

Complications arise since Gµ⌫ [ĝ] is a non-linear function of the metric operator, so in general hGµ⌫ [ĝ]i ⌘
h , g|Gµ⌫ [ĝ]| , gi =
6 Gµ⌫ [hĝi]. Nonetheless, we can always expand the metric around a classical solu-
tion as
cl
ĝµ⌫ = gµ⌫ 1 + ˆµ⌫ . (11.6)

141
Formally, one can then obtain (keeping only terms up to quadratic in µ⌫ )

Gµ⌫ [hĝi] hGµ⌫ [ĝ]i = 8⇡GN ht̂µ⌫ [ˆ ]i , (11.7)

where t̂µ⌫ can be seen as the contribution to the stress-energy tensor due to quantum fluctuations of
the metric.
Hence, we can write the operatorial version of the Einstein equations (11.5) as
⇣ ⌘
Gµ⌫ [hĝi] = 8⇡GN hT̂µ⌫ i + ht̂µ⌫ [ˆ ]i . (11.8)

This procedure suggests that, in the limit where the matter/gravitational state is highly populated
(e.g. macroscopic values of the fields that can be approximated by coherent states), the term htµ⌫ [ˆ ]i
becomes negligible and hĝµ⌫ i behave as a classical metric. So formally, in this limit, equation (11.8)
does reduce to (11.2).
There are, however, few remarkable problems with this approach. First of all, we are not even
sure that a full quantum theory admits an operatorial form like (11.8). A second natural objection
appears when de-localized quantum states are considered. For instance, in a semi-classical approach,
a matter state of the form
1
| i = p (|M1 i + |M2 i) , (11.9)
2
where |Mi i is a generic mass state, would not make sense at all in the “averaged” equation (11.2).
Even when these issues are left aside, expression (11.2) remains an extremely hard equation to
solve. Indeed, only trivial exact solutions are known at the present time and one usually needs to use
approximation techniques in order to handle such a complicate problem.

11.1.1 Approximation Techniques

In dealing with Einstein semiclassical equation several approximation techniques have been devised.

• Linearization: The metric can be expanded around flat space as gµ⌫ = ⌘µ⌫ + hµ⌫ , with hµ⌫
considered as a small perturbation. By keeping only linear terms in the field equation, and by
cl = 0, one finds that the perturbation satisfies
taking a vacuum state for matter such that Tµ⌫
✓ ◆
1 16⇡GN Q
⇤ hµ⌫ h ⌘µ⌫ = Tµ⌫ + O(h2 ). (11.10)
2 c2

In this way, the graviton hµ⌫ appears just as a usual quantum field in flat space. For instance,
graviton-matter scattering can be computed in this method (very time consuming though). The
only drawback is that General Relativity is non-renormalizable, hence not every Green function
is well defined.

• Test field limit: this is tantamount to quantum field theory in curved spacetime. One takes
a fixed background geometry, a field in its vacuum state (so that the classical SET is zero) and
Q
then perform a quantization of this field of the background so to calculate Tµ⌫ (which being of
order O(~) is naturally suppressed) neglecting its back-reaction on the background metric.
cl and computes T Q in a vacuum.
• Full back-reaction: First, one starts with a classical metric gµ⌫ µ⌫
cl + h , and fluctuations h
At this point, the metric is allowed to fluctuate gµ⌫ = gµ⌫ µ⌫ µ⌫ are to
Q
be computed form the linearised Einstein equation with Tµ⌫ as a source on the right-hand side.
Q
When this is done, one can re-compute Tµ⌫ in presence of the computed gµ⌫ and re-iterate the
previous calculations until convergence is (hopefully) obtained. In practical situations, back-
reactions are considered non-relevant, so they are usually neglected (at most the first iteration
is performed).

142
 Q
This is all nice and well, but so far we have no prescription on how to compute Tµ⌫ . The most
prominent issue in its calculation is the fact that generally it diverges. This is not a surprise, indeed
this is the case also in flat space (recall the divergent Casimir energy). These divergences, even for
non-interacting field theories, can all be traced back to coincident-point problems: the stress-energy
tensor is made of products of field evaluated at the same space-time point. For example, Tµ⌫ for a
scalar is
1 1
Tµ⌫ (x) = rµ (x)r⌫ (x) gµ⌫ V ( (x)). (11.11)
2 2
In particular, products of the sort ⇠ (x) (x) produce UV divergences. In a mathematical language,
quantum fields are operator-valued distributions, but a product between two distributions is not a
well-defined mathematical concept (for instance ⇠ (x) (x) has no meaning mathematically speaking).
This UV problem in usual Quantum Field Theory is dealt with by the use of either normal ordering
or Casimir subtraction. It is easy to realize that these procedures cannot work in curved spacetime.

• First of all, normal ordering is the procedure of moving all the creation operator to the left and
annihilation operators to the right of a given expression. However, we saw that the operators â
and ↠can be mixed by a Bogolyubov transformation in curved space, so normal order is not
“conserved” in a generic change of basis.

• For what regards the Casimir subtraction scheme it is also easy to see that problems arise. Con-
sider the FRW metric. As it is well known, this metric is conformally equivalent to Minkowski,
so it can be brought to the form
⇥ ⇤
ds2FRW = C(⌘) d⌘ 2 d 2 , (11.12)

where d⌘ = a(t) 1 dt. A particular solution can be given by C(⌘) = cos2 (A⌘) so that the scale
⇥ ⇤1/2
factor is a(t) = 1 (At)2 .
Also in this case, the stress-energy tensor for a scalar field in vacuum formally diverges (as
expected) hTµ⌫ iFRW ! 1. However, like in the case of the Casimir energy, we can regularize
its components with the introduction of a regulator ⇠ exp[ ↵(! 2 + A2 )1/2 ], and by then letting
↵ ! 0 once we have removed the divergences. An option is to remove them by subtracting the
divergent expectation value in Minkowski hTµ⌫ iM . For example, the regularized energy density
would look like
⇢ ✓ 2 ◆
48 2 1 2 D (⌘)
hT00 iFRW = [32⇡ 2 C 2 (⌘)] 1 + D 2
(⌘) 8A + A A 2
log ↵ + O(↵0 ),
↵2 ↵2 2
(11.13)
@
with D(⌘) ⌘ @⌘ log C. The Minkowski limit would correspond to assigning C = 1, and hence
D = A = 0. Indeed, in Minkowski one finds
48
hT00 iM = + O(↵0 ). (11.14)
32⇡ 2 ↵2
So we immediately realise that by subtracting (11.14) from (11.13) does not save the day. What
we learnt is that in curved spaces divergences are also curvature dependent so the analogue of
te Casimir subtraction can easily fail.

11.2 Renormalization of the Stress-Energy Tensor

As we have seen, UV divergences in curved spacetime appear in subtle ways and naive renormalization
procedures usually fail. Thus, we need more solid procedures in order to make sense of hTµ⌫ i.
There are three main ways to regularize and renormalize the stress-energy tensor, all of which can
also be used in flat space:

143
 • Dimensional regularization;

• Riemann-zeta function renormalization;

• Point-splitting renormalization.
Given that he first and the second renormalization procedures are rigorous only in Riemannian
manifolds 1 ; for this reason, we focus on the point-splitting procedure.

11.2.1 Point-splitting Renormalization

To solve the problem of terms like (x) (x), we introduce an auxiliary point y and a geodesics 0
connecting y with x. Then the stress-energy tensor in a vacuum state can be written as

h0|T̂µ⌫ (x)|0i ⌘ lim h0|T̂µ⌫ (x, y; 0 )|0i, (11.15)

y!x

where it is understood that, since Tµ⌫ is quadratic in the fields, one of them is evaluated at x while
the other at y. It can be shown that for small distances (between y and x) this procedure does not
depend on the particular choice of 0 . Another option would be to take an average over all possible
paths connecting the two points, but we are not going to dwell into this.
The result of this procedure has the following structure [45]
(1)
h0|T̂µ⌫ (x)|0iren = Dµ⌫ (x, y; 0) G (x, y; 0 )ren
(11.16)
G(1) (x, y; 0 )ren ⌘ h0| { (x), (y)} |0iren .

The object Dµ⌫ (x, y; 0 ) is a second-order di↵erential operator that depends solely on the background
geometry (and is therefore regular), whilst G(1) (x, y; 0 )ren is the renormalized Hadamard Green func-
tion for the field (x). The form (11.16) is written on the assumption that ultimately Tµ⌫ (x) contains
second-order derivatives and that it is constructed with products of two fields.

Green functions
In general, there is a plethora of interesting Green functions.
• Hadamard Green function:

G(1) (x, y; 0) ⌘ h0| { (x), (y)} |0i (11.17)

• Pauli-Jordan or Schwinger Green function:

iG(x, y) ⌘ h0|[ (x), (y)]|0i (11.18)

• Wightman Green functions:

G+ (x, y) ⌘ h0| (x) (y)|0i (11.19a)

G (x, y) ⌘ h0| (y) (x)|0i (11.19b)

The above Green functions are related as

iG(x, y) = G+ (x, y) G (x, y)
G(1) (x, y) = G+ (x, y) + G (x, y)

1
It has also been argued that Zeta regularization and dimensional regularization are equivalent since they use the
same principle of using analytic continuation in order for a series or integral to converge.

144
 • Feynman Green function:
iGF (x, y) ⌘ h0|T (x), (y)|0i
(11.20)
= ✓(y0 x0 )G (x, y) + ✓(x0 y0 )G+ (x, y)

• Retarded Green function:

GR (x, y) ⌘ ✓(x0 y0 )G(x, y) (11.21)

• Advanced Green function:

GA (x, y) ⌘ ✓(y0 x0 )G(x, y) (11.22)

Green functions for thermal states

In curved spacetimes, not only vacuum states are relevant, but also thermal states are important. A
thermal state |Ti is a statistical distribution of pure states. To deal with them we have to specify the
probability distribution for the thermal bath. In the Gran-canonical ensemble the density matrix is,
as usual, given by
⇢ˆT = e (⌦+µN̂ Ĥ) , (11.23)
where = (kB T ) 1 , Ĥ is the Hamiltonian, µ is the chemical potential, N̂ is the number operator,
and the gran-potential ⌦ fixes the normalization for ⇢ˆT . For any operator Â, the expectation value in
a thermal state is X h i
hÂiT = ⇢Ti h i |Â| i i = Tr ⇢ˆÂ . (11.24)
i
where ⇢Ti = h i |ˆ
⇢T | i i.
The Wightman functions evaluated over such a thermal state (with µ = 0) have the following
striking property:
G± ±
T (x0 , x; y0 , y) = GT (x0 + i , x; y0 , y), (11.25)
so a thermal environment induces a periodicity along the imaginary time-axis. Moreover the Hadamard
Green functions satisfy X
(1)
GT (x0 , x; y0 , y) = G(1) (x0 + ik , x; y0 , y), (11.26)
k2Z
i.e. the thermal Green functions can be written as an infinite imaginary sum of the corresponding
T = 0 Green functions. This last result can be used to show [5] the useful fact that periodicity in
Euclidean time implies thermality in Lorentzian time. From this very fact one could be able to derive
thermally in stat ionary back hole spacetimes in a geometric way.
Semiclassical Euclidean quantum gravity techniques [92] play a key role in the investigation of the
thermodynamics of black holes [93]. Given the classical Einstein–Hilbert action plus Gibbons–
Hawking term for gravity and the action of classical matter fields, one can formulats the Euclidean
path integral by means of a Wick rotation t ! i⌧ a .
In particular, one finds that the dominant contribution to the Euclidean path integral is given
by gravitational instantons (i.e. non-singular solutions of the Euclidean Einstein equations).
In spacetimes with event horizons, metrics extremizing the Euclidean action are gravitational
instantons only after removal of a conical singularity at the horizon [92]. A period must therefore
be fixed in the imaginary time, ⌧ ! ⌧ + , which becomes a sort of angular coordinate.
To be concrete we can take the standard Schwarzschild metric in metric signature:
⇣ rh ⌘ 2 ⇣ rh ⌘ 1
ds2E = 1 d⌧ + 1 dr2 + r2 d⌦2 (11.27)
r r
This solution still shows singularities at r = rh = 2M and r = 0. If we now define a new radial

145
 p
variable, ⇢ = 2 r rh , the metric acquires the form
✓ ◆ 1 ✓ ◆ ✓ ◆
⇢2 ⇢2 2 ⇢2 ⇢2
ds2E = + rh d⌧ + + rh d⇢2 + + rh d⌦2 (11.28)
4 4 4 4

For ⇢ ! 0 one then gets:

1 ⇢2 2
ds2E = d⌧ + rh d⇢2 + rh2 d⌦2 (11.29)
rh 4
p
If one now defines ⇢¯ ⌘ ⇢ rh and ⌧¯ ⌘ ⌧ /(2rh ) the metric takes the so called conical form:

ds2E = ⇢¯2 d¯
⌧ 2 + d¯
⇢2 + rh2 d⌦2 (11.30)

Smoothness at ⇢¯ = 0 (no conical singularity) requires that the rescaled imaginary time behaves
like an angle. I.e. ⌧¯ = ⌧¯ + 2⇡ which in turns imply ⌧ = ⌧ + 2⇡/kappa. So, the required period
corresponds exactly to the inverse Hawking temperature H = 2⇡/ and because of the relation
between the periodicity of the Euclidean Green’s functions and the thermal character of the
corresponding Green’s functions in Lorentzian signature, this implies that the state of fields on
this spacetime will have to be thermal at the Hawking temperature.
In such a way thermodynamics appears as a requirement of consistency of quantum field theory
on spacetimes with Killing horizons, and in this sense we shall define such a thermodynamics as
“intrinsic” to the spacetime rather than the property of a specific field. (Note however, that the
situation changes dramatically in the case of extremal black holes where one can see that there is
no way to fix the period of the imaginary time, and hence the intrinsic temperature. b )
Hence, in these cases, the e↵ective action is truly a free energy function F . Indeed, we can
consider the partition function for solutions where a temperature can be fixed in the way described
above.
The partition function in a canonical ensemble Z can be written as
Z
Z = Tr exp ( H) = D [ , gµ⌫ ] exp [ I ( , gµ⌫ )], (11.34)

Using the fact that a black hole solution is an extremum of the (Euclidean) gravitational action,
at the tree level of the semiclassical expansion one then obtains
Z Z
1 1
Z ⇠ exp [ IE ] , IE = [( R + 2⇤) + Lmatter )] + [K] , (11.35)
16⇡ M 8⇡ @M

where IE is the on-shell Euclidean action and [K] = K K0 is the di↵erence between the extrinsic
curvature of the manifold and that of a reference background. The Euclidean action is then related
to the free energy F as IE = H F and the corresponding entropy can be computed as

2 @F
S⌘ (11.36)
@

and similarly the internal energy and the heat capacity are respectively given by U = @ ( F ) and
C = 2 @ 2 ( F ). This formulas provide the black hole thermodynamical quantities at the tree

level while in the case IE is the one loop e↵ective action of the fields on the black hole background
the same formula can be used to derive the analogue quantities of the radiation of the background
(and in this case the entropy coincides with the entanglement entropy of the radiation).
a
Working in the Euclidean signature is a standard technique in QFT in flat backgrounds. It has the advantages
of making the path integral “weighted” by a real factor and in general transforms the equation of motion of free
fields from hyperbolic to elliptical, a feature that often leads to just one Green function satisfying the boundary
conditions.

146
 b
Indeed for an extremal black hole general form of the metric is
⇣ rh ⌘2 2 ⇣ rh ⌘ 2
ds2E = 1 d⌧ + 1 dr2 + r2 d⌦2 (11.31)
r r
Performing the same transformations as before this can be cast in the form:
✓ 2 ◆ 2 ✓ 2 ◆2 ✓ 2 ◆2 ✓ 2 ◆
⇢ ⇢ ⇢ 4 ⇢
ds2E = + rh d⌧ 2 + + rh d⇢ 2
+ + r h d⌦2 (11.32)
4 4 4 ⇢2 4
which in the ⇢ ! 0 limit gives
✓ ◆2
1 ⇢4 4r2
ds2E = d⌧ 2 + 2h d⇢2 + rh2 d⌦2 (11.33)
rh 16 ⇢
It is now easy to see that in this case passing to a system of coordinates equivalent to the previous one given by ⌧¯
and ⇢¯ does not give rise to any “conical-like” metric. This implies that, for extremal black holes, there is no way to
fix the period of the imaginary time and hence the intrinsic temperature.

Let us go back to (11.16). Although the problem of renormalizing Tµ⌫ is reformulated in terms of
Green functions, we still need to remove the infinities from them.
In general, Dµ⌫ is UV free and contains only geometrical quantities. One has instead to deal with
the Green function infinities in the coincidence limit. The basic idea behind the renormalization of
these infinities is that they are universal and related to the UV structure of the vacuum. As such, this
divergence structure of the Green functions should show a universal behaviour similar to the one of
Minkowski vacuum if the state is well defined all over our spacetime. State that satisfy this assumption
are called Hadamard states.
In Minkowski, the divergent structure of the Hadamard function is
1
lim G(1) (x, y; 0) = + ↵ log 0 (x, y) + $(x, y), (11.37)
y!x 4⇡ 2 0 (x, y)

where 0 (x, y) is half of the squared geodesics distance

1
0 (x, y) = (x y)µ (x y)µ + ✏2 + 2i✏(x0 y0 ) , ✏ ! 0+
2
Note that the terms dependent on ✏ have to be introduced to deal with null distances. Note that the
terms depending on 0 (x, y) contain the divergent parts, while the function $(x, y) is in general a
finite quantity.
In curved space, the divergent structure of the Hadamard function, evaluated on a Hadamard
state, does reproduce the same UV behaviour, and indeed one finds
p 
(1) (x, y) 0 1
lim G (x, y; 0 ) = + v 0 (x, y) log | 0 (x, y)| + $ 0 (x, y) ; (11.38)
y!x 4⇡ 2 0 (x, y)
where it is self-evident that the divergent structure of (11.38) is the same as the one in flat spacetime
(11.37). The only di↵erence is the appearance of p the (regular) spacetime dependent functions v 0 (x, y)
and $ 0 (x, y) and most of all the overall factor (x, y) 0 , where
" #
rxµ ry⌫ 0 (x, y)
(x, y) 0 ⌘ det p (11.39)
g(x)g(y)

is the so-called van Vleck determinant (these result change in dimensions di↵erent from 4). This is an
exquisitely geometrical object which takes into account the focussing e↵ects due to tidal forces (see
e.g. [45]).
Finally, the renormalized Green function can be obtained by subtracting the divergent part of
(11.38):
p 
(1) (1) (x, y) 0 1
Gren (x, y) = lim G (x, y; 0 ) 2
+ v 0 (x, y) log | 0 (x, y)| . (11.40)
y!x 4⇡ 0 (x, y)

147
 The form of Dµ⌫ is in general a complicated expression. For the particular case of a conformally-
coupled massless scalar field can be found for example in [45, p. 285] (too long and clumsy to be worth
reporting here). However, once it is determined and applied to (11.40) the RSET is finally derived.
Note that there is nothing “new” in this renormalization. Indeed the first term in (11.38) is related
to a wave-function renormalization whereas the logarithmic term represents a mass renormalization.
One can see this is the case for example by considering the Green function in Minkowski for a massive
scalar field: in order to have a regular x ! y limit one needs to subtract a term ⇠ 1/x2 and a
term ⇠ log(x). Even in curved spacetimes, because of local Lorentz invariance, the structure of the
divergences is the same as in Minkowski and Equation (11.38) is only a covariant generalization of the
flat-space case (if the state we consider is well-behaved i.e. is Hadamard).
However, in curve spacetime this renormalization procedure is not completely unambiguous. In-
deed, the Hadamard subtraction scheme is always ambiguous up to further finite renoprmalization::
in principle it is always possible to subtract the divergent part plus some arbitrary finite function so
changing the resulting finite part $(x, y).
In order to address this ambiguity, there are some useful physical criteria that can be laid down
so to fix hT µ⌫ (x)iren These are often named the Wald’s criteria for a well-defined renormalized
stress-energy tensor:

1. Conservation equation: it must be the case that the renormalized stress-energy tensor remains
conserved rµ hT µ⌫ (x)iren = 0 (compatibility with Contracted Bainchi identity in GR);

2. Causality: hT µ⌫ (x)iren for a fixed “in” state hT µ⌫ (x)iren at a point p must depend only on J¯ (p)
(causal past of p); (the time reversal of this statement must be true for any fixed “out” state).

3. Standard results for o↵-diagonal elements: for positive-norm states the elements h |Tµ⌫ | i must
behave as classical objects; This is simply the observation that as h |Tµ⌫ | i is anyway finite for
orthonormal states h | i = 0 2 , then its value must be formally the same one as in the classical
case.

4. Minkowski limit: for vanishing curvature, the stress-energy tensor must reduce consistently to
its normal ordered version in Minkowski.

Given these requirements it is possible to proof that albeit they cannot pinpoint the corrected
RSET to use, any two SET will di↵er just by a local, conserved, geometrical tensor

Hµ⌫ ⌘ Tµ⌫ T̄µ⌫ . (11.41)

Let us stress that this tensor depends only on the spacetime geometry and not on the state, which
indeed it can be always decomposed as
✓ ◆
1 p 2
Hµ⌫ = c1 gµ⌫ + c2 Gµ⌫ + c3 p gR + . . . . (11.42)
g g µ⌫

Clearly, the coefficients ci appearing in this expression can be interpreted as providing a renormal-
ization for the bare quantities appearing on the left-hand-side of the Einstein equations. For instance,
c1 renormalizes the Cosmological constant ⇤, c2 renormalizes the Newton constant, and so on. There-
fore, the renormalization of Tµ⌫ immediately provides a renormalization for matter and gravitation
couplings (and can induce higher order curvature terms even if they are not present in the bare action).

P ⇣ ⌘ P ⇣ ⌘
2
Think for example (formally) h |H| i = k !k h |a†k ak | i + 1/2 h |1| i = k ! h |a†k ak | i .

148
11.3 Semiclassical Collapse
In most situations, gravitational quantum e↵ects are negligible: gravitational interactions are techni-
cally irrelevant at low energies and become strong only around Planckian energies. This means that in
a scattering experiments we can neglect gravity when we deal with energies below E ⇠ G 1/2 ⇠ MPl .
On the other hand classical non-linearities of General Relativity arise in many situations, for
instance close to the formation of an horizon. In these scenarios one could wonder whether quantum
e↵ects can become important e.g. due to large blueshift e↵ects close to the horizon.

Indeed, one could see that the transplanckian problem discussed in section 10.3.1 originates already
during the collapse albeit being usually formulated in static spacetimes.
We can, as usual, encode the dynamics of the geometry in the relation U = p(u) between the
affine null coordinates U and u, regular on I and I + , respectively. Neglecting back-scattering,
a mode of the form
1 i!u
! (r, t) ⇡ e , (11.43)
(2⇡)3/2 (2!)1/2 r

near I takes, near I + , the form

1 i⌦p(u)
'⌦ (r, t) ⇡ e . (11.44)
(2⇡)3/2 (2⌦)1/2 r

This can be regarded, approximately, as a mode of the type presented in equation (11.43), but now
with u-dependent frequency !(u, ⌦) = ṗ(u) ⌦, where the dot denotes di↵erentiation with respect
to u. (Of course, this formula just expresses the redshift undergone by a signal in travelling from
I to I + .)
In general we can expect a mode to be excited if the standard adiabatic condition

|!(u,
˙ ⌦)|/! 2 ⌧ 1 (11.45)

does not hold. It is not difficult to see that this happens for frequencies smaller than

⌦0 (u) ⇠ |p̈(u)|/ṗ(u)2 . (11.46)

One can then think of ⌦0 (u) as a frequency marking, at each instant of retarded time u, the
separation between the modes that have been excited (⌦ ⌧ ⌦0 ) and those that are still unexcited
(⌦ ⌦0 ).
Moreover, Planck-scale modes (as defined on I ) are excited in a finite amount of time, even
before the actual formation of any trapped region. Indeed, they start to be excited when the
surface of the star is above the classical location of the horizon by a proper distance D of about
one Planck length, as measured by Schwarzschild static observers. We can see this by observing
that the red-shift factor satisfies close to the horizon (where p(u) ' UH A1 e u )

(1 2M/r)1/2 ⇠ !/⌦ = ṗ(u) ⇠ /⌦0 , (11.47)

where  = (4M ) 1 is the surface gravity. This then implies (r 2M ) ⇠ /⌦20 , where we have
used  M ⇠ 1. Hence
D ⇠ (r 2M ) (1 2M/r) 1/2 ⇠ 1/⌦0 . (11.48)
Hence, the trans–Planckian problem has its roots at the very onset of the formation of the trapping
horizon.

Should we have to worry about this? Indeed, the resurfacing of the transplanckain issue also in a
semiclassical description of the gravitational collapse rise the question of wether quantum mechanical

149
 Figure 11.1: Penrose diagram for the collapse of a star in 1 + 1-dimensions.

e↵ects can a↵ect the formation of an horizon and appears to be of great relevance if we want to assess
the robustness of Hawking radiation. After all, the gravitational collapse of a star is expected to
happen in a regime in which we can apply semiclassical gravity: we have quantum fields living in a
classical curved space and quantum e↵ects are expected to be relatively small. Let us then consider
this problem in more detail (this discussion is based on [94]).
Consider a collapsing star. Outside of the star the vacuum state is Boulware and it is regular outside
the horizon. We want to assess whether backreaction becomes important during the formation of the
horizon (when classical non-linearities are large).
Since we are dealing with quantum fields it is better to work in Heisenberg representation, where
the quantum state is fixed during the evolution and is globally defined over our spacetime. We call
this state |Ci (as for Collapse). At early times this is chosen to be the vacuum state for an observer
at I (i.e. the state is annihilated by the annihilation operator associated to free waves at I ).
Having an Unruh state at late times would imply that a free-falling observer at H+ would not detect
particles and that hTµ⌫ i = 0. Moreover, for a far-away observer at I + such state would be at late
times a thermal state at the Hawking temperature.
Let us work in a manageable 1 + 1 geometry for a spherical collapse, and consider a scalar field on
such background. Albeit quite simple this set-up is sufficiently general for drawing lessons about the
possible relevance of quantum backreaction.
Before starting the calculation we can make some guesses on what we expect to happen based
on what we learned in the previous lectures. Since we are considering a spectator free-field in the
background geometry of the collapse, its renormalized stress-energy tensor (RSET) can only depend
on the geometrical quantities (invariants) and physical parameters (such as the mass of the field). At
the horizon we only have coordinate singularities, and invariant quantities such as the Riemann tensor
squared are finite on H+ . Therefore we expect the RSET to be finite as well. It is not trivial however
to see how the Boulware state (appropriate for a spacetime with a static start) could be turned into
the Unruh one by the onset of horizon formation.
More formally there is a theorem due to Fulling, Sweeny and Wald [95] that guarantees that in a
globally hyperbolic spacetime (i.e. no Cauchy horizons) if a state is Hadamard on I (or any given
Cauchy hypersurface) it will remain so everywhere. In a semiclassical description of a collapse under
consideration, this can be used to predict that the state |Ci, which coincides with vacuum state for
observers on I , cannot also be a vacuum state for observers on I + as this would imply that |Ci is
the Boulware state, which however it is not regular at H+ .
We can choose coordinates U and W defined as the coordinates respectively on on IL and IR
as in Fig. 11.1. The metric in these null-coordinates reads
g= C(U, W ) dU dW . (11.49)
Note that outside of the star, because of Birkho↵’s theorem, we have a static geometry. We are also
going to assume there is a relation between the coordinate U and the null coordinate on IR+ , that

150
we call u, of the form U = p(u). This function p(u) tells us about the dynamics of the collapse: rays
going from IL to IR+ are blue/red shifted and this information is contained in p(u). Also, we can
write the metric as a function of the coordinates (u, W ) as

g= ¯ W ) du dW =
C(u, C ṗ(u) du dW . (11.50)

We therefore have the relations

¯ ṗ , @U
C = C/ @u = @U = ṗ@U . (11.51)
@u
The RSET for any massless quantum field (corresponding to a vacuum on I ) is

TU U / C 1/2 @U2 C 1/2

, (11.52)
TW W / C 1/2 @W
2
C 1/2 , (11.53)
TU W / R . (11.54)

This relations are given explicitly in [5] and are related to the so called trace anomaly.

Trace anomaly. For a conformal invariant field the trace of the classical SET is zero. Upon
quantization one in general has that the total one loop e↵ective action W will also be in general
conformal invariant so that if g ! ⌦(x)2 g then

⌦(x) W
hTµµ i = p =0 (11.55)
g ⌦ ⌦=1

In order to renomalize our action will shall still need to separate the finite and divergent parts
of the e↵ective action W = Wren + Wdiv , however the resulting two terms are not separately
conformal invariant and the both have a non-zero trace. The origin of this is that the action is
conformal invariant in d = 4 but not exactly so e.g. if one performs a dimensional regularisation
to handle the divergent part d = 4 + ✏. Even in the limit ✏ ! 0 this leaves an imprint in the form
of the trace anomaly. For a conformal invariant field on a conformally flat spacetime one can show
that all of the renormalised SET will be accounted for by the trace anomaly coming from Wren .

For us the prefactors are not important, what is relevant is the dependence of the geometrical
quantities. The components TU W and TW W are manifestly well-behaved everywhere: they are the
same as in an always static geometry and do not contain the dynamical quantity ṗ(u). On the other
hand TU U can be problematic, so we are going to check whether it can become singular.
First, using (11.50) and (11.51) we can write
1 h ¯1/2 2 ¯ 1/2 i
TU U / C 1/2 @U2 C 1/2 = C @u C ṗ1/2 @u ṗ 1/2 . (11.56)
ṗ
The first term on the right-hand side is a term that would be present outside of a static star, and
contains the divergent term of the Boulware vacuum ⇠ (1 2M/r)2 . The second term (containing ṗ)
is instead related to the dynamics of the collapse and describes the related particle flux.
To track accurately quantum e↵ects at the onset of the horizon it is obviously best to use system
of horizon-crossing coordinates. Therefore we introduce a horizon penetrating system: the Painlevè-
Gullstrand coordinates (x, t) defined in (10.47).3 The metric takes the form (omitting the P G sub-
scripts for simplicty)
g = c2 (x, t) dt2 + [ dx v(x, t) dt]2 . (11.57)
3
For the reader convenience we recall them here
( h ⇣ q ⌘ p i
tPG = tsch ± 4M arctanh 2 r2M
sch
2 2M rsch
xPG = rsch

151
The horizon is located at c+v = 0. The null-rays trajectories are obtained from the relation dx = (±c+
v) dt respectively for left- and right-moving rays (the dependence on x, t will be always understood).
Outside the star the metric is of course static, so c and v become time-independent. When we
reach infinity the metric has to be asymptotically the flat metric, so we have limx!1 c(x) = 1,
limx!1 v(x) = 0.
We need to connect the two systems of coordinates. To do so we integrate the equation for the
right-going null-rays starting from an event (t, x) just outside the star to another one (tf , xf ) at IR+
Z xf
dx dx0
dt = =) tf t = . (11.58)
c(x) + v(x) x c(x0 ) + v(x0 )

Assuming asymptotic flatness, c(+1) = 1 and v(+1) = 0, the null coordinate u is then given by 4 .
Z xf Z x
dx0 dx0
u ⌘ lim (tf xf ) = t xf + = t = . (11.59)
tf !1 x c(x0 ) + v(x0 ) c(x0 ) + v(x0 )

In contrast, along an incoming ray leaving asymptotic past infinity IR at an event (ti , xi ) and
remaining outside the star, we have instead
Z x
dx dx0
dt = =) t ti = 0
. (11.60)
c(x) + v(x) xi c(x ) v(x0 )

Then the null coordinate W is defined as

Z x Z x
dx0 dx0
W ⌘ lim (ti + xi ) = t + xi + =t+ . (11.61)
ti ! 1 xi c(x0 ) v(x0 ) c(x0 ) v(x0 )

The expressions (11.59) and (11.61) for u and W give us useful expressions outside the star:
1
@x u = ,
c+v
ṗ(u)
@x U = @x p(u) = ṗ(u)@x u = ,
c+v
@t U = ṗ(u)@t u = ṗ(u) , (11.62)
1
@x W = ,
c v
@t W = 1 .

We see indeed that the dynamics enters only in U and not in W . We will also need the change of
variables for the partial derivative

@t @x c+v c2 v2
@u = @t + @x = @t @x . (11.63)
@u @u 2c 2c
Other useful relations, which can be derived by substituting and comparing the line elements (11.49)
and (11.57), are
1
@t U = (c + v)@x U , @t W = (c v)@x W , C(t, x) = . (11.64)
@x U @x W
4
R xf dx0
Note that xf = c(x0 )+v(x0 )

152
 At this point we can change coordinates in the RSET (using eq. (11.62) and (11.64))

Ttt = (@t U )2 TU U + 2(@t U )(@t W )TU W + (@t W )2 TW W

= ṗ2 TU U 2ṗTU W + TW W , (11.65)
2 2
Txx = (@x U ) TU U + 2(@x U )(@x W )TU W + (@x W ) TW W
ṗ2 2ṗ 1
= TU U TU W + TW W , (11.66)
(c + v)2 c2 v 2 (c v)2
Ttx = (@t U )(@x U )TU U + (@t U @x W + @x U @t W )TU W + (@t W )(@x W )TW W
ṗ2 2ṗv 1
= TU U + 2 TU W + TW W . (11.67)
c+v c v2 c v
The problematic terms for the backreaction are those involving c + v, that goes to zero at H+ . We
can see that Ttt is well behaved at the horizon while Ttx is always less divergent than Txx there. Also
no divergences are coming from the parts proportional to TW W .
So, let us focus on the region near the horizon, and also assume c(x) = 1 so that v encodes all
the x dependence of the problem (this choice is equivalent in a specific choice for the collapse). The
horizon can be taken at x = 0, and therefore at H+ we have v = 1. Let us expand v around this
point
v(x) ' 1 + x + x2 + . . . . (11.68)
@v
The coefficient  = @x |x=0 represent the surface gravity of the black hole (this can be checked for
instance in Schwarzschild).
At this point we can focus on TU U :
1 h ¯1/2 2 ¯ 1/2 i
TU U / C @u C ṗ1/2 @u ṗ1/2 . (11.69)
ṗ
By expanding around the horizon we get
ṗ 1
C¯ = ṗ C = = =1 v 2 (x) ⇡ 2x . (11.70)
@x U @x W @x u@x W
So that h i 2
C¯1/2 @u2 C¯ 1/2
' (2x)1/2 x@x x@x (2x) 1/2
= + O(x2 ) . (11.71)
4
Note that if this was the full result, we would have Ttx , Txx diverging as 1/x and 1/x2 for x ! 0
(implying a large backreaction).
Let us however consider now also the contribution of the dynamical term. At late times (u ! 1),
we already know that
p(u) ' UH A1 e u (11.72)
but what about earlier times? One can show that in this case one generically gets extra sub-leading
corrections to the previous formula [94].

u A2 2u A3 3u
p(u) ' UH A1 e + e e + ... . (11.73)
2 3!
Therefore we can see that
" ... ✓ ◆ # " ✓ ◆2 #
1/2 2 1/2 1 p 3 p̈ 2 2 A3 3 A2
ṗ @u ṗ = = + + 2 e 2u
+ O(e 3u
) . (11.74)
2 ṗ 2 ṗ 4 2A1 4 A1

After this calculation we can make some comments.

• not surprisingly the flux term in this 1+1 problem is ruled by the Schwarzian derivative.

153
 • In the end, by summing the two contributions (11.71) and (11.74) in (11.69) we obtain a cancel-
lation of the leading terms (those in 2 /4) so that the components Ttx , Txx are indeed regular.
This is the “Unruh state miracle” and it is what makes this the only regular state compatible
with the spacetime associated to a gravitational collapse leading to the formation of a future
event horizon H+ . The vacuum polarization static term divergence at H+ (the same of the Boul-
ware state) is exactly balanced by an equal and opposite divergence coming from the Hawking
flux term.

• For what regards the subleading contributions, it is easy to see that they do not introduce other
divergences. Indeed note that for small x (so that 1 + v ⇠ x) we have
Z
dx 1
u=t 't log x + const , (11.75)
c+v 

so that e u / xe t and so e 2u ⇠ (x2 ). Therefore the subleading terms in (11.74) are able
to regulate the singularities ⇠ 1/x2 .

• The subleading terms that survive are exponentially small in time. For example, we get Txx ⇠
O(1)e 2t . So after the formation of the horizon these terms rapidly decay and no information
about the collapse remains encoded (via a residual REST) on the horizon. This is an another
manifestation of the no-hair theorem for black holes as any collapse related feature is rapidly
shed away.

• Moreover, if at some t we write x = (r(t) rh ) ' (t tH ) + . . . (where tH is the time of

formation of the horizon) then one can show that

A2 1 A3 2
/ , / . (11.76)
A1 A1
The constant represent the velocity of matter during the collapse. For a freely falling collapse
this is very close to the speed of light. However, if some unknown new physics allows the
collapse proceeds slowly, then one could start witnessing large quantum contributions. This
suggests the guess made at the beginning is right: collapsing stars with no relevant support
at horizon-formation experience negligible quantum back-reaction (since as we have just seen
quantum e↵ects are negligible if not allowed to grow in first place by something else slowing
down the collapse). This could have been foreseeable. If we start with Minkowski vacuum on
I and we have to remain in a Hadamard state during a Cauchy evolution then we can keep
renormalising the SET. As long as the star collapse in near free falling, we can always go in
the corresponding local inertial frame where the renormalisation of the SET should not show
large deviation from the Minkowski vacuum. Once a horizon is formed the RSET at the horizon
should rapidly become zero as needed for the onset the Unruh vacuum being the latter the only
regular vacuum compatible with this geometry.

• Our deductions do not change if instead of an event horizon one consider simply a trapping
horizon existing for a finite time. Simply in this case they would be relevant for observations on
a specific region of spacetime (e.g. the finite portion of I + associated to light rays which arrive
there after having experienced an exponential redshift.

So in the end it seems very unlikely then that quantum physics can prevent the formation of an
horizon in standard gravitational collapse. Nonetheless, this does not guarantee that the regularisation
of spacetime singularities imposed by quantum gravity could not dramatically change the black hole
structure and fate at later time (see e.g. the scenarios listed in [61]). For example, one could explore
bouncing scenarios where a series of multiple bounces could lead conditions where the above mentioned
corrections might not be negligible.

154
 Anyway, modulo the above mentioned scenarios resorting to transient quantum gravity regimes, we
can safely conclude that the above analysis corroborates the conclusion that gravitational collapse will
generically lead to the formation of a trapping horizon even when taking into account the semiclassical
back-reaction, and that the onset of a thermal at the Hawking temperature on I + at late times will
be unavoidable, just by regularity of the quantum state, once such a horizon formes.

155
 12
Time Machines and Chronology
Protection

Sometimes, this is seen as something related to science-fiction, but in reality this is a serious problem
of General Relativity, since the theory allows time travels. The fact that this is a problem in a certain
sense is good, because represents a point to investigate in order to extend the theory, as time travel
solutions can be seen of further signs that General Relativity is an incomplete theory.

12.1 Time travel paradoxes

Another good reason to explore time machine solutions in General relativity is the fact that they are
intrinsically paradoxical (at least without entailing radical departures from our vision of how physics
works). Note that by paradoxical we mean here true logical inconsistencies rather that the kind of
apparent paradoxes that sometime we encounter in discussing the implications of physical theories.
To be clear and example of the latter type is the so called twin paradox in Special Relativity,
where one of two twins is sent in an interstellar travel at relativistic speed while the other stay on
Earth. According to the one that stayed on Earth, the twin who left, once back to Earth, should be
younger. However, because of the relativity of reference frames, according to the travelling twin, the
reverse should apply as well. However, this is not a real paradox, since there is a way to solve the
problem: in order to return, the travelling twin has to decelerate and accelerate back, thus breaking
the symmetry of the problem, and therefore the travelling twin ends up being the only one remaining
younger.
The real paradoxes associated to time machines are in situations involving travelling backward in
time; travelling forward in time is not a problem: this is the situation in the aforementioned twin
paradox or in science-fiction movies, like Interstellar where one hour for people sitting on a close orbit
to a black hole could corresponds to several years for someone farther away. More precisely, there
are two types of real paradoxes associated to backward time-travel: the grandfather paradox and the
bootstrap paradox.

In the grandfather paradox, a man travels back in time and kills his own grandfather before he
meets his grandmother. In this way, the time traveller can never be born, and in this way he
will not be able to travel back in time. Therefore the grandfather lives, the time traveller is born
and he has the chance to get back in time, and so on. Thus, it is not clear from which present
the time traveller can come from, and this causes a paradox.

In the bootstrap paradox, for example, you can go back in time to give your past self the winning
numbers of a lottery. At this point, you have generated information from nowhere. This type of
paradoxes can then be also transformed in a sort of grandfather paradox: imagine for example

156
 that, once you won the lottery, later in your future you conclude that money do not buy happiness
and you decide to not travel back in time to give yourself the winning numbers...

Actually, there are numerous examples one can formulate for the bootstrap paradox; the science-
fiction British TV Show Doctor Who provides one such example in one episode:
So, there’s this man, he has a time machine. Up and down history he goes — zip, zip, zip,
zip, zip — getting into scrapes. Another thing he has is a passion for the works of Ludwig van
Beethoven. Then, one day, he thinks, “What’s the point in having a time machine if you don’t
get to meet your heroes?” So, o↵ he goes to 18th century Germany, but he can’t find Beethoven
anywhere. No one’s heard of him. Not even his family have any idea who the time traveller is
talking about. Beethoven literally doesn’t exist. [...] The time traveller panics. He can’t bear the
thought of a world without the music of Beethoven. Luckily, he’d brought all of his Beethoven
sheet music for Ludwig to sign. So, he copies out all the concertos and the symphonies, and he
gets them published. He becomes Beethoven. And history continues with barely a feather ru✏ed.
My question is this: who put those notes and phrases together? Who really composed Beethoven’s
Fifth?
(Doctor Who, Before the Flood, written by Toby Whithouse, 2015)

12.2 Time Machines

After this preliminary discussion about the paradoxes of time travel let us now try to formulate more
precisely what we meant by a time machine in physics.
A time machine is a region of spacetime which allows to travel back in time. This region can be
an integral part of a given solution of the Einstein field equations or it can be dynamically generated
in an otherwise perfectly sane spacetime by using some sort of device.
For a technical definition of what we mean by travel back in time we have to consider world-
lines associated to light or material observers. From the concepts of causal curve and chronological
curves, we can in fact introduce the concepts of closed causal curve (CCC) and closed timelike
curve (CTC). So we have time machines in a spacetime if in at least so some region of it one finds
CCC or CTC. Closed causal curves produce causality violations, while closed timelike curves produce
chronology violations.
Notice that chronology violation imply causality violation, but the converse is not true. Indeed, if
one has only CCC the above mentioned paradoxes cannot be encountered: nonetheless it is possible
to show, and rather intuitive, that such spacetimes if perturbed can easily lead to CTC. This is why
time machines are associated to the existence of a chronology violation region (i.e. a region endowed
with CTC).
In order to define a latter we can start by noticing that:
A point p is in a chronology violating region if the intersection of the chronological past and the
chronological future of p is not zero:

I 0 (p) ⌘ I + (p) \ I (p) 6= ;; (12.1)

the total chronology violating region in a manifold M is then defined as

[
I 0 (M) = I 0 (p); (12.2)
p2M

Obviously, the same concepts can be defined for causality violation, where it is sufficient to replace I
with J in the definitions above.
Then we can define a chronological horizon as
⇥ ⇤
H+ (I) = @ I + (I 0 (M)) ; (12.3)

157
chronological horizons are, as we have already said, a type of Cauchy horizons; they mark the onset
of a region in which we can have chronology violation, therefore they mark the boundary between a
“sane” region of the manifold and a region where closed timelike curves are allowed.
But is time travel really allowed in General Relativity? Is there any spacetime which is solution
of the Einstein field equations, in which closed timelike curves are present? The answer is yes.
There are three categories of solutions of the field equations that allow time travels: globally rotat-
ing solutions, wormholes solutions, and faster-than-light travel solutions (warp drives and Krasnikov
tubes).

12.2.1 Time machines from globally rotating solutions

In this category are included the van Stockum spacetimes, the Gödel Universe and Kerr Black Holes.
The basic mechanism generating time machine regions in spinning spacetimes can be summarised
with the fact that for a sufficiently high velocity, there are generically regions where the tilting of the
lightcones can lead to the formation of closed timelike curves.

The van Stockum spacetime

In this solution of General Relativity, we have an infinite cylinder of dust of density ⇢(r) that satisfies
the condition
d⇢
> 0; (12.4)
dr
the solution is given by the following metric:

dz 2 + dr2
ds2 = dt2 2ar2 dt d' + (1 a2 r2 )r2 d'2 + , (12.5)
exp(a2 r2 /2)

where a > 0 is a constant such that a2 is proportional to the mass density of the dust at r = 0. The
coordinates range in the intervals

t 2 ( 1, +1), ' 2 [ ⇡, ⇡], r 2 (0, +1), z 2 ( 1, +1). (12.6)

We can see that there are closed timelike curves if we consider an azimuthal curve at fixed (r, z, t),

Figure 12.1: Graphical representation of the tilting of lightcones in a van Stockum spacetime. At a
particular value of the radius, the tilting is such that closed null curves are possible and, beyond that
radius, also closed timelike curves are allowed.

which is an integral curve of the angle and hence is always closed. If we compute the invariant
length Z
s = ds =) s2 = (1 a2 r2 )r2 (2⇡)2 , (12.7)

158
we can see that s2 becomes negative (timelike) for r > a1 . I.e. for sufficiently high values of the radius,
we have closed timelike curves.
Finally, let us note that also Tipler cylinders are included in this family (with a massive cylinder
replacing the cylindrical distribution of dust) as well as Gott’s spacetimes (these are spacetimes with
two infinitely long, parallel, cosmic strings longitudinally spinning around each other).

The Gödel Universe

The Gödel universe [96], is very similar to Einstein static Universe, in the sense that it contains a
homogeneous distribution of dust and a cosmological constant, but the dust is swirling so it describes
a universe endowed with a global rotation. Therefore, this solution is not isotropic but only homoge-
neous, in the sense that the axis of rotation is the same at any point and appears the same at any
location.
The stress-energy tensor of this universe is

µ⌫ ⇤ µ⌫
Ttot = ⇢v µ v ⌫ ⇢ + p̄)v µ v ⌫ + p̄g µ⌫ ,
g = (¯ (12.8)
8⇡G
with
⇤ ⇤ ⌦2
p̄ = ⇢¯ = ⇢ + and ⇢= with ⌦2 = ⇤; (12.9)
8⇡G 4⇡G 4⇡GN
we can pick the velocity of the fluid in a way that it is always timelike, v µ = (1, 0, 0, 0), and in addition
we can define a vorticity ! µ = (0, 0, 0, ⌦), where ⌦ is the angular velocity of the universe [45, page
271]. Then, the solution is given by
p 1 2p2⌦y 2
ds2 = dt2 2e 2⌦y
dt dx e dx + dy 2 + dz 2 . (12.10)
2
Gödel universe has many interesting features that would deserve a much more in depth discussion (see
e.g. [97]), but for us its main characteristic is that it also entails light-cones tilting and CTC. In order
to see this, it is more convenient to go to cylindrical coordinates, where the metric becomes [45]
p p h p i
ds2 = dt̄2 + 4⌦ 1 sinh2 ( 2⌦r) d' dt̄ + 2⌦ 2 sinh2 ( 2⌦r) 1 sinh2 ( 2⌦r) d'2 + dr2 + dz 2 ;
(12.11)
as we did before, we can compute the invariant length along ' and see where this can become negative,
h i p
2 2
p p ln(1 + 2)
s' / 1 sinh ( 2⌦r) < 0 , if sinh( 2⌦r) > 1 i.e. ⌦r > p , (12.12)
2
So, around each point of Gödel spacetime (being this universe homogeneous) one can find sufficiently
far away orbits there are closed timelike curves. See Figure 12.2.
This solution is considered unphysical, since the rotation of the whole Universe seems physically
unmotivated, and also because a fine tuning of the cosmological constant is needed. Indeed, the van
Stockum and Gödel solutions are sometime dismissively called “GIGO solutions”, which stands for
“Garbage In - Garbage Out”, in the sense that with a very strange/unphysical input in the Einstein
equations, it is obvious to have a very strange/unphysical output. Finally, it is also worth noticing that
this universe also a non-Machian aspect of GR: it has the same content as Einstein static Universe,
still it has a completely di↵erent behaviour, since there is a rotation not “visible” in the distribution
of the matter content.

159
Figure 12.2: Left: graphical representation of the tilting of the lightcones in a Gödel Universe. The
grey cylinder represents the Gödel horizon. Right: example of a Closed Timelike Curve in a Gödel
spacetime. From [98].

Kerr Black Holes

We have already seen Kerr black holes, described by a metric given by
✓ ◆ ✓ 2 ◆
2 a2 sin2 ✓ 2 2 r + a2
ds = dt 2a sin ✓ dt d'+
⌃ ⌃
 2
(r + a2 )2 a2 sin2 ✓ ⌃
+ sin2 ✓ d'2 + dr2 + ⌃ d✓2 ,
⌃
= r2 2M r + a2 , ⌃ = r2 + a2 cos2 ✓

We have seen that in r = 0 there is a ring singularity; however, this singularity is only at ✓ = ⇡2 .
Therefore, approaching r = 0 at an angle di↵erent than ⇡/2 will not result in encountering the
singularity. It can be shown that traversing the plane on which there is the singularity through the
ring is analogous to extend the Kerr metric to negative values of r (e.g. [58]).
Now, consider the Killing vector of Kerr related to the conservation of angular momentum along
@
', = @' . Its norm is given by

(r2 + a2 )2 a2 sin2 ✓
| |2 = g'' = sin2 ✓ , (12.13)
⌃
Now it is easy to see that such norm becomes negative, whenever

(r2 + a2 )2 a2 sin2 ✓ < 0

(12.14)
) r4 + (2 sin2 ✓)a2 r2 + 2M a2 r sin2 ✓ + a4 cos2 ✓ < 0.

This is realised for small negative values of r – which means close to the ring singularity after one
has travelled through the ring – and for ✓ close to ⇡2 (not equal to ⇡2 in order to avoid the ring
singularity); since describes closed curves, when the previous condition is realized we are in presence
of closed timelike curves and hence of chronology violation conditions. Also in this case, people is
often unconcerned about this unpleasant feature because it happens well within the Cauchy horizon,
which, as said, is conjectured to be unstable anyway.

160
Figure 12.3: Schematic representation of the basic structure of a Kerr black hole. The region de-
nominated “Sisytube” is the toroidal region close to the singularity where closed timelike curves are
allowed. (Credits: Andrew Hamilton, http://jila.colorado.edu/~ajsh/insidebh/index.html)

12.2.2 Wormholes
Wormholes are shortcuts through spacetime, connecting or points of teh same universe (intra-universe
wormhole) or points of di↵erent, disconnected universes (inter-universe wormholes). From a topological
point of view, an intra-universe wormhole (a wormhole between two points in the same universe) is
a compact region of spacetime whose boundary is topologically trivial, but whose interior is not
simply connected. Formalising this idea leads to definitions such as the following [45]: If a Minkowski
spacetime contains a compact region W , and if the topology of W is of the form W ⇠ R ⇥ ⌃3 where
⌃3 is a three-manifold of nontrivial topology, whose boundary has topology of the form @⌃ ⇠ S 2 and
if, furthermore, the hypersurfaces ⌃3 are all spacelike, then the region W contains a quasi-permanent
intra-universe wormhole.
Wormholes can come in di↵erent shape and sizes. Let us hence do here a little bit of taxonomy.

Wheeler Wormholes
These are structures at the Planck scale, where you imagine to have a foam, where geometry is not
uniform and can have topologically complex structure. Since we do not really know what happens at
the Planck scale, it is generally believed that they are not a problem as far as chronology violations
are concerned.

Schwarzschild wormholes — Einstein–Rosen Bridges

Schwarzschild wormholes, also known as Einstein–Rosen bridges, are connections between the two
asymptotically flat regions the maximally extended version of the Schwarzschild metric (describing an
eternal black hole with no charge and no rotation). In this spacetime, it is possible to come up with
coordinate systems such that if a hypersurface of constant time is chosen (e.g. T = 0 and ✓ = ⇡/2)
then the “embedding diagram”, drawn depicting the curvature of space at that time deducible from
the Schwarzschild line element, will look like a tube connecting the two exterior eulclidean regions,
known as an “Einstein–Rosen bridge”. See Fig. 12.4. These form of wormholes is not traversable (as
it is referring to a space-like surface) and even worse not even stable: in 1962 it was shown by Wheeler
and Fuller [99] that if the Einstein–Rosen bridge connects two parts of the same universe, then it will
pinch o↵ too quickly for light (or any particle moving slower than light) that falls in from one exterior
region to make it to the other exterior region.

161
 Figure 12.4: Embedding of a Schwarzschild wormhole (From Wikipedia).

Morris-Thorne Traversable Wormholes

In an attempt to improve over the the Einstein–Rosen bridge and construct a traversable wormhole
Morris and Thorne proposed in 1998 a new set of solution of Einstein equations [100, 101]. 1 In
principle, these traversable wormholes are not forbidden or unstable, however they need to violate
energy conditions and would be very difficult to produce. Let us add that the above comment does
not apply only to the Morris–Thorne solutions: these are just a special case of traversable wormholes
for which the same point would apply. 2 .
The Morris–Thorne solution in General Relativity is given by a time-independent, non-rotating,
spherically symmetric metric:
⇥ ⇤
ds2 = e 2 (`) dt2 + d`2 + r2 (`) d✓2 + sin2 ✓ d'2 . (12.15)

This metric needs some requirements in order to describe a wormhole:

1. the range for the coordinate ` should be ` 2 ( 1, +1);

2. the structure of r(`) must be such that there are no event horizons, since we want a traversable
solution;

3. this must be an asymptotically flat and regular solution at infinity, which implies that
r(`)
lim = 1, (12.16)
`!±1 |`|

and therefore r(`) ⇠ |`| + O(1) at large `. Moreover, since we want to recover Minkowski at
infinity, we also want lim`!±1 (`) = ± = const which can be then set to 1 by a proper
redefinition of the time coordinate.

4. The metric component have to be at least C 2 in `

Solving the whole solution, one finds the radius of the wormhole to be

rwormhole = r0 ⌘ min{r(`)}; (12.17)

1
Noticeably, these are the structures appearing in Interstellar or in Contact by Carl Sagan, and apparently Kip
Thorne was stimulated in his research by a question of Carl Sagan which was at the time writing the novel Contact.
2
For example a wormhole solution can be built by cutting the Schwarzschild solution at some r very close to rH and
gluing it with an identical metric on the other side: inserting the so obtained metric the Einstein equations would still
lead to a SET associated to large violations of the energy conditions)

162
basically, ` is a coordinate going through the wormhole, being 1 at one side and +1 at the other
side; anyway, ` can be set in a way that r0 = r(` = 0), and therefore ` = 0 at the center of the
wormhole, at the “throat”.
Given the above requirements the spacetime described by this solution is then a sort of shortcut
between two asymptotically flat regions of the same universe or of disconnected universes. See Fig. 12.5.

Figure 12.5: A traversable Lorentzian wormhole. From Web.

In order to see that this metric is associated, as anticipated, to large violation of the energy
conditions, all one needs to do is to inserte the solution into the Einstein equations and find the
stress-energy tensor. The latter takes the form

Tµ⌫ = diag (⇢(r), ⌧ (r), p✓ (r), p' (r)) , (12.18)

where ⇢ is the density, ⌧ (r) = pr (r) is the tension, p✓ is the pressure along the ✓ component and p'
the pressure along the ' component. One can show that there is always a radius r⇤ such that, for any
r between r0 and r⇤ , ⇢ ⌧ < 0, which means a violation of the NEC, which in turn implies a violation
of WEC, SEC and DEC.
Notice that all these aforementioned energy condition violations do not forbid a priori the existence
of such a solution. As a matter of fact, we already know that our universe violates at least the SEC and
not only that: quantum phenomenal like the Hawking radiation or even the Casimir e↵ect can violate
most if not all of energy conditions, therefore it might be that quantum e↵ects could be engineered
by an advanced civilisations so to realise this type of structures. Nonetheless, let us stress that this
will be very difficult as generically large amount of exotic matter will be required.
Indeed let us estimate the value of the energy density and of the tension at the throat. These
are [45]
✓ ◆2
1 1 36 10 m
⇢(r0 )  ⌧ (r0 ) = = 5 ⇥ 10 N cm2 , (12.19)
8⇡Gr02 8⇡Gr02 r0
which implies that, for example, to have a traversable wormhole with r0 ⇠ 1 m, one has to have

r0 c 2
Mthroat ' ⇠ MX , (12.20)
G
which means that we need at the wormhole throat something of the order of a Jupiter’s mass of exotic
matter, in order to keep open a wormhole about one meter wide.
But how could a wormhole become a time machine? For example, we could set the two ends of
the wormhole in a way that, by entering at a time t, you exit in the same place at a time t t, and
in this way you would have basically travelled back in time.

163
 In order to achieve a situation like this, one possibil-
ity, consist in placing one of the ends of the worm-
hole close to a black hole (or some other very com-
pact object) and the other basically in Minkowski
space, in order to have a time dilation between the
two clocks (shown in the picture on the left by the
dashed grey lines). If we now send a light ray (solid
black lines) through the wormhole, every time it
passes through the wormhole it will feel a greater
e↵ect of time dilation, as illustrated here.
After some time, the light ray will be on the left end
(the exit) of the wormhole at the same time as it
will be on the right end (the entering) of the worm-
hole. From that point on, it will be possible to exit
the wormhole before entering it on the other side,
and we would then have a chronology violation.

Figure 12.6: Wormhole time machine constructed

CAL REVIEW LETTERS
by delaying time at one mouth. 26 SEPTEMBER 1988

—2M/ ~
I ~,
ases from
um value
onward to
ere finite
is violated An alternative, could be to Lorentz move away at
an be ex-
relativistic speeds one of the two ends for some time
G. Here p
n, p=e
and the boost it back close to the initial mouth.
= Je dl The tow mouths in this case are like the twin in
"Casimir the famous twin “paradox” of special relativity.
netic field The non-inertial path of the boosted mouth mo-
ition") to tion generates a time shift between the two mouths
wo identi-
(the right mouth to ”age” less than the left as seen
equal elec-
y out our from the exterior) and then getting them closer and
s/rv. Be- closer can generate first a CCC (the chronological
is in the =Z horizon) and then CTCs.
he lateral
FIG. 2. Spacetime diagram for conversion of a wormhole
into a time
Figure machine.
12.7: Spacetime diagram for the conversion
oinb field
of a spherical, traversable wormhole into a time ma-
Reissner- chine by one mouth
parametrized boost. Figure from [100, 101]
by a time coordinate t introduced below.
nce at the At T =0, the wormhole's mouths are at rest near each
field equa- other. Subsequently, the left mouth remains at rest
f (i) zc To be more precise let us see how a CTC can come into e↵ects with these sort of expedients.
while the right mouth accelerates to near-light speed,
[s —l0 Consider the its
then reverses worldlines
motion and of returns
the twoto wormhole
its original mouths,
loca- l1 and l2 at distance l; suppose that l1µ (t) =
; (ii) the (t,tion.
0, 0,The
0) and l2µ (t) =
advanced (t +can
beings T, 0, 0, l). The
produce invariant
this motion by length of the geodesic connecting the mouths at
d by dis- the sameont the
pulling canright
be calculated as before:or electrically.
mouth gravitationally
fraction- This motion causes the right mouth to "age" less than
unit area kl2µ (t) l1µ (t)k
s2 (t) = Consequently,
the left as seen from the exterior. at2 late
= k(T, 0, 0, l)k2 = T 2 + l2 , (12.21)
'
cr & —, zcs, times by traversing the wormhole from right mouth to
and
left,it one
can can
be seen
travelthat if l2  T
backward in 2 ,time 2 (t)one
then(i.se. ,  0can and therefore we would have a closed timelike curve.
h a total traverse
Whena sclosed
2 0 it iscurve)
(t) >timelike alwayandpossible
thereby, toperhaps,
find, via some Lorentz boost, a frame where the two
d, 2o+ps violate causality.
theory re- The metric inside the accelerating wormhole and out-
xceed that side but near its mouths is 164
zcs corre-
e electron ds —(I+glFcos8) e dt +dl2+r (d8 +sin 8' ).
Here 4 @(1) and r =r(l) are the same functions as for
wormhole mouths are at equal times. The required rapidity of the boost would be = T /l However,
as `2 ! T 2 one can easily see that such rapidity would approach one, and for l < T , one would need
to go faster than the speed of light in order to undo the time machine.

12.2.3 Alcubierre Warp Drives

These Alcubierre warp drive [102] describes a metric allowing for faster-than-light (FTL) travel. Note
that classically the speed of the bubble can exceed the speed of light albeit locally this is still the
limit speed for the propagation of signals (we saw an example of this in cosmology when discussion
the meaning of the Hubble “horizon”).
In particular, it is characterised by are bubble-like structure where there is an expansion of the
spacetime in the back of the bubble and a contraction at its front. The inside of the bubble is basically
flat Minkowski spacetime, and the bubble can travel at an arbitrary speed. 3 Also in this case large
violations of the energy condition are required. Finally, the Alcubierre warp drive is just a special
case of a family of solutions with the same characteristics (i.e. there are several other types of warp
drives).
The structure of such a pacetime is schematically shown Figure 12.8: one has a “bubble” of flat
spacetime, such that behind it there is an expansion of spacetime and in front of it a contraction, that
makes the bubble to move forward. In order for this to be stable, one needs areas with large violation
of energy conditions localised on the sides at the transition between expansion and contraction.

Figure 12.8: Spacetime structure in Alcubierre warp drive (from Wikipedia user AllenMcC.) ).

The metric describing this solution is

ds2 = dt2 + ( dx vc f (r) dt)2 + dy 2 + dz 2 , (12.22)

where vc is the speed of the center of the bubble, and f (r) is a function describing the shape of the
bubble. Here, r is the distance from the center of the bubble,
q
r = [x xc (t)]2 + y 2 + z 2 ; (12.23)

the function f (r) is usually taken to be f (0) = 1 and f (1) = 0. We can also see that the worldline
{xc (t), 0, 0, 0} is a geodesic, therefore at the centre of the bubble one does not feel any acceleration.
Now, while having a superluminal warp drive does not violate any physical law per se (as long as
one admits EC violating matter) it is easy to say that it can be easily turned into a time machine.
Consider Figure 12.9. If you can travel faster than light in any reference frame – for example,
along the red spacelike trajectory from the origin, and than following the blue trajectory – you can
always pick a Lorentz frame where the arrival point belongs to the past of the starting point. In this
3
Also in this case there is a connection with science fiuction as apparently the investigation on teh possibility to built
such a structure was inspired by the famous warp drives used in the TV series Star Trek.

165
Figure 12.9: Light-cone representation of a situation where superluminal travel could generate a
chronology violation. The green reference frame travels at subluminal speed with respect to the
orange one. The orange frame sends a superluminal signal in its future, from O to A. The green frame
sends a superluminal signal in its future from A to B. This causes B to arrive in the orange frame
before OA is sent.

way, you can create a closed timelike curve. However this is not true if we have a particle that in
a given reference frame is superluminal: in this way we would have a preferred system of reference
and, in this case, Lorentz invariance would not hold anymore. If Lorentz invariance is removed, this
paradox is removed from the onset. So keep it mind it is superluminal travel plus Lorentz invariance
that make a time machine possible.
Once you have built a warp drive, this cannot communicate with the outside world, if it is travelling
faster than light. Then, in order to travel to, say, Alpha Centauri, one should build first a kind of
railway, similar to a magnetic monorail, with synchronised devices to create EC violations at the right
time in the right place (e.g. by generating localized, large, vacuum polarization). Then you can come
back using this “railway” at an arbitrary speed. There have been alternative ideas that allow the
spaceship itself to generate its own faster than light “railway”. Most noticeably, there is the so called
Krasnikov tube [103], where the light cones are opened and tilted during the trip so to allow FTL
in one direction. Also in this case the tube has to be created with a first trip on the destination but
even worse the amount of required NEC/DEC violation is huge as it increases with the length of the
tube that needs to be kept permanently opened.

12.3 Possible solutions

Now, how can we solve this problems?
As far as warp drives are concerned, it seems that quantum e↵ects in curved spacetime forbid this
possibility. What can we say about wormholes, for example? Are there possible way out to avoid time
machines? It is thought that the solution could be one of these:

1. Splitting Timelines: once a paradox happens, the timeline bifurcates; in this case, one has
to do physics on a non-Hausdor↵ manifold, that have a disconnected structure with branches.
More precisely, a topology is said to be Hausdor↵ if and only if for any two di↵erent points x1
and x2 , there exist opens sets O1 and O2 such that x1 2 O1 , x2 2 O2 , and O1 \ O2 = ;. For
example, a non-Hausdor↵ set is what one obtains by removing the zero from the real numbers
axis and replacing it with two values 01 and 02 . In this case, consider a small interval ( ✏, +✏)

166
 with ✏ > 0. Then the two sets ( ✏, 0) [ {01 } [ (0, ✏) and ( ✏, 0) [ {02 } [ (0, ✏) are both open sets
containing respectively just 01 and 02 but their intersection is not empty.

2. Novikov Conjecture: in this case the idea is that time paradoxes are just apparent ones
because histories are indeed consistent and periodic. This is equivalent to say if you are about
to create a time travel paradox you will not not succeed, but on the contrary, your time travel
and your actions in the past had to happen for history to unfold the way it has [104, 105].
This solution is very popular among philosophers of science, less so among physicists because of
in some toy models (e.g. Minkowski spacetime with periodic boundary conditions in time and
space plus a massless scalar field) it was shown that not always the required periodic solution
do exists. Moreover, such conjecture seems also difficult to accomodate with the probabilistic
nature of quantum mechanics.

3. Multiverse Solution, where there is an infinite number of Universes, and in any of them any-
thing can happen, and there could be a sort of path integral that is connected to the probability
of something happening or not happening with respect to other Universes. This is also related
to the Many World Interpretations of Quantum Mechanics by Everett.

4. Hawking’s Chronology protection conjecture, that states that quantum physics will forbid
time machines to form; in the case of the warp drive, we have already seen that quantum e↵ects
forbid superluminal warp drives; in the case of the wormhole, we can see that we have an
increasing concentration of photons near the critical point, an accumulation, and in proximity
of the formation of the chronological horizon, we would see a diverging stress-energy tensor.
However, there is a theorem by Kay, Radzikowski and Wald [106] according to which a quantum
state fails to be Hadamard on chronological horizons. Since we need an Hadamard state to
renormalize the stress-energy tensor, this implies that close to a chronological horizon the stress-
energy tensor cannot be renormalized, and the final word cannot be set. The final answer should
be Quantum Gravity, that should forbid the formation of time machines.

12.3.1 Preemptive chronology protection?

While we have only circumstantial evidence for the validity of the chronology protection conjecture,
there is at least in the case of the Alcubierre warp drive some evidence this can be indeed true. In
fact, in the case of the warp drive it can be shown that even before attempting to use it to form CTC
an QFT driven instability sets in as long as the propagation is superluminal [107, 108]).
Consider the Carter–Penrose diagram of a warp drive as shown in Figure 12.10. Once the superlu-
minal warp drive is established, there are two apparent horizons, where the back wall acts like a black
hole and the front wall acts like a white hole: inside a superluminal bubble, nothing coming from the
back can come in, since it is moving faster than light to every observer outside, and therefore it acts
like a black horizon; however, if someone tries to send a signal outside of the bubble from the front,
the bubble from the outside is receding faster than light, and therefore the front wall acts like a white
horizon.
Now one can understand that an instability appears: the back wall, being a black horizon, emits
Hawking radiation, that flows to the white horizon, which then becomes an accumulator of this
radiation, and this gives a growing instability. Let us see this in more detail.

12.3.2 Superluminal warp drive instability

In the actual computation we shall restrict our attention to the 1 + 1 dimensions case (since in this
case one can carry out a complete analytic treatment) and follow [107]. Changing coordinates to those
associated with an observer at the center of the bubble, the warp-drive metric (12.22) becomes

ds2 = c2 dt2 + [dr v̄(r)dt]2 , v̄ = v vc , (12.24)

167
 i+

IL
)
1

+ (t
=

=
,r

+
1

1
+

,r
=
st

=
con

(t
+
r=

1
L
I

)
i0L

HC
+ (t
HL

=
r1
(t

+
1
=

(r
t=

,r
1 +
1
con

=
,r
st

r2
H

)
=
r1
)

r=c

HR
HC

+
)

(t
r2
onst
(t

=
=
=

+
(r

1
1

,r
,r

=
H
=

r2
r1

)
)
i0R

IR
st

)
con

1
r=

(t

+
=

=
,r
1

1
,r
=

=
+

(t
1

R
I
)
i

Figure 12.10: Left panel, structure of Alcubierre spacetime (from [109]), right panel corresponding
Carter-Penrose diagram of an eternal Alcubierre warp drive (from [107]).

where r ⌘ x xc (t) is now the signed distance from the center. Let us consider a dynamical situation
in which the warp-drive geometry interpolates between an initial Minkowski spacetime [v̂(t, r) ! 0, for
t ! 1] and a final stationary superluminal (vc > c) bubble [v̂(t, r) ! v̄(r), for t ! +1]. The causal
structure will be the one shown in Fig. 12.11 To an observer living inside the bubble this geometry has
two horizons, a black horizon H+ located at some r = r1 and a white horizon H located at r = r2 .
Here let us just add that from the point of view of the Cauchy development of I these spacetimes
+
posses Cauchy horizons (see HC ). Note that these are an artefact of the idealised case in which the
warp drive keeps on going at superluminal speed forever. As such we shall not deem them physically
relevant in what follow.

Light-ray propagation
Let us now consider light-ray propagation in the above described geometry. Only the behaviour of
right-going rays determines the universal features of the RSET, just like outgoing modes do in the
case of a black hole collapse. Therefore, we need essentially the relation between the past and future
null coordinates U and u respectively on I and I + , labelling right-going light rays.
There are two special right-going rays defining, respectively, the asymptotic location of the black
and white horizons. In terms of the right-going past null coordinate U let us denote these two rays
by UBH and UWH , respectively. The finite interval U 2 (UWH , UBH ) is mapped to the infinite interval
u 2 ( 1, +1) covering all the rays traveling inside the bubble. For rays which are close to the black
horizon, the relation between U and u can be approximated as a series of the form[107]

1 u A2 21 u
U (u ! +1) ' UBH + A1 e + e + ... . (12.25)
2
Here An are constants (with A1 < 0) and 1 > 0 represents the surface gravity of the black horizon.
This relation is the standard result for the formation of a black hole through gravitational collapse.
As a consequence, the quantum state which is vacuum on I will show, for an observer inside the
warp-drive bubble, Hawking radiation with temperature TH = 1 /2⇡.

168
 i+

IL
1

+ (t
=

=
,r

+
1

1
+

,r
=
t

=
ns

(t
+
co

1
L
I
r=

)
i0L

1 +

HC
+
H

(t
=
+
1
t=

,r
co

=
ns

r2
t

)
st
r=

r = con
r1

HR
+
2
H
IL

(t
=
(t

+
=

1
,r
1
,r

=
r2
=

)
r = r2
1 i0R
)

o ns t

)
r=c

1
+
=
,r
1
=
(t
R
I
i

Figure 12.11: Carter-Penrose diagram of a dynamically formed Alcubierre warp drive (from [107]).

Equivalently, we find that the corresponding expansion in proximity of the white horizon is [107]
D2 22 u
U (u ! 1) ' UWH + D1 e2 u + e + ... , (12.26)
2
where D2 > 0 and 2 is the white hole surface gravity and is also defined to be positive. The
interpretation of this relation in terms of particle production is not as clear as in the black horizon
case and a full study of the renormalised stress energy tensor (RSET) is required.

Renormalized stress-energy tensor

In past null coordinates U and W the metric can be written as

ds2 = C(U, W )dU dW . (12.27)

In the stationary region at late times, we can use the previous future null coordinate u and a new
coordinate w̃, defined as
Z r
dr
w̃(t, r) = t + . (12.28)
0 c v̄(r)
In these coordinates the metric is expressed as
C̄(u, w̃)
ds2 = C̄(u, w̃)dudw̃ , with C(U, W ) = , (12.29)
ṗ(u)q̇(w̃)

where U = p(u) and W = q(w̃). In this way, C̄ depends only on r through u, w̃.

169
 For concreteness, we refer to the RSET associated with a quantum massless scalar field living
on the spacetime. The RSET components for this case are again given in Ref. [5] and using the
relationships U = p(u), W = q(w̃) and the time-independence of u and w̃, one can calculate the RSET
components in the stationary (late times) region.[107]
Let us now focus on the energy density inside the bubble, in particular at the energy ⇢ as measured
by a set of free-falling observers, whose four velocity is uµc = (1, v̄) in (t, r) components. For these
observers neglecting transient terms one obtains [107] ⇢ = Tµ⌫ uµc u⌫c = ⇢st + ⇢dyn , where we define a
static term ⇢st , depending only on the r coordinate through v̄(r),
" #
1 v̄ 4 v̄ 2 + 2 0 2 2v̄ 00
⇢st ⌘ v̄ + v̄ , (12.30)
24⇡ (1 v̄ 2 )2 1 v̄ 2

and a, time-dependent, dynamic term

...
1 F(u) 3p̈2 (u) 2ṗ(u) p (u)
⇢dyn ⌘ , where F(u) ⌘ . (12.31)
48⇡ (1 + v̄)2 ṗ2 (u)

Physical interpretation
Let us start by looking at behavior of the RSET in the center of the bubble at late times. Here
⇢st = 0, because v̄(r = 0) = v̄ 0 (r = 0) = 0. One can evaluate ⇢dyn from Eq. (12.31) by using a
late-time expansion for F(u), which gives F(u) ⇡ 21 , so that ⇢(r = 0) ⇡ 21 /(48⇡) = ⇡TH2 /12, where
TH ⌘ 1 /(2⇡) is the usual Hawking temperature. This result confirms that an observer inside the
bubble measures a thermal flux of radiation at temperature TH .
Let us now study ⇢ on the horizons H+ and H . Here, both ⇢st and ⇢dyn are divergent because
of the (1 + v̄) factors in the denominators. Using the late time expansion of F(u) in the proximity of
the black horizon one gets
( " ✓ ◆ # )
A 2
A ⇣ ⌘
2 3
lim F(u) = 21 1 + 3 2 e 21 t (r r1 )2 + O (r r1 )3 , (12.32)
r!r1 A1 A1

and expanding both the static and the dynamic terms up to order O(r r1 ), one obtains that the
diverging terms (/ (r r1 ) 2 and / (r r1 ) 1 ) in ⇢st and ⇢dyn exactly cancel each other exactly
as it happens in the semiclassical collapse [107]. We hence get an Unruh-like state setting up in
correspondence to the formation of the warp drive. It is now clear that the total ⇢ is O(1) on the
horizon and does not diverge at any finite time. By looking at the subleading terms,
" ✓ ◆ #
e 21 t A2 2 A3
⇢= 3 2 + A + O (r r1 ) , (12.33)
48⇡ A1 A1

where A is a constant, we see that on the black horizon the contribution of the transient radiation
(di↵erent from Hawking radiation) dies o↵ exponentially with time, on a time scale ⇠ 1/1 .
Close to the white horizon, the divergences in the static and dynamical contributions cancel each
other, as in the black horizon case. However, something distinctive occurs with the subleading contri-
butions. In fact, they now becomes
" ✓ ◆ #
e22 t D2 2 D3
⇢= 3 2 + D + O (r r1 ) . (12.34)
48⇡ D1 D1

This expression shows an exponential increase of the energy density with time. This means that ⇢
grows exponentially and eventually diverges along H . This is just the statement that a Unruh-like
state cannot be regular also on a white horizon.

170
 In a completely analogous way, one can study ⇢ close to the Cauchy horizon. Performing an
expansion at late times (t ! +1) one finds that the RSET diverges also there [107]. Note that the
above mentioned divergences are very di↵erent in nature. The divergence at late times on H+ stems
from the untamed growth of the transient disturbances produced by the white horizon formation. The
RSET divergence on the Cauchy horizon is due instead to the well known infinite blue-shift su↵ered by
light rays while approaching this kind of horizon. While the second can be deemed inconclusive because
of the Kay–Radikowski–Wald theorem, the first one is inescapable. Summarising: the backreaction
of the RSET will doom the warp drive making it semiclassically unstable. This seems to suggest a
preemptive implementation of Hawking’s chronology protection conjecture.

171
 Bibliography

[1] Robert M Wald. General relativity. Chicago, IL: Chicago Univ. Press, 1984.
[2] Sean M. Carroll. Spacetime and geometry. An introduction to general relativity. 2004.
[3] Eric Poisson. A Relativist’s Toolkit: The Mathematics of Black-Hole Mechanics. Cambridge
University Press, 2004. doi: 10.1017/CBO9780511606601.
[4] T. Padmanabhan. Gravitation: Foundations and Frontiers. Cambridge University Press, 2010.
doi: 10.1017/CBO9780511807787.
[5] N. D. Birrell and P. C. W. Davies. Quantum Fields in Curved Space. Cambridge Monographs
on Mathematical Physics. Cambridge, UK: Cambridge Univ. Press, 1984. doi: 10 . 1017 /
CBO9780511622632.
[6] A A Grib, S G Mamayev, and V M Mostepanenko. Vacuum quantum e↵ects in strong fields.
St. Petersburg: Fridmann Lab., 1994.
[7] Viatcheslav Mukhanov and Sergei Winitzki. Introduction to quantum e↵ects in gravity. Cam-
bridge University Press, 2007.
[8] Vittorio Berzi and Vittorio Gorini. “Reciprocity Principle and the Lorentz Transformations”.
In: Journal of Mathematical Physics (1969).
[9] R. Torretti. Relativity and geometry. 1983.
[10] Sebastiano Sonego and Massimo Pin. “Foundations of anisotropic relativistic mechanics”. In:
Journal of Mathematical Physics (2009).
[11] George Bogoslovsky. “Some physical displays of the space anisotropy relevant to the feasibility
of its being detected at a laboratory”. In: arXiv e-prints (2007).
[12] Harvey R. Brown. Physical relativity. Space-time structure from a dynamical perspective. 2005.
[13] S. Liberati. “Tests of Lorentz invariance: a 2013 update”. In: Classical and Quantum Gravity
(2013).
[14] Andrea Pelissetto and Massimo Testa. “Getting the Lorentz transformations without requiring
an invariant speed”. In: American Journal of Physics (2015).
[15] Eolo Di Casola, Stefano Liberati, and Sebastiano Sonego. “Nonequivalence of equivalence prin-
ciples”. In: American Journal of Physics (2015).
[16] Eolo Di Casola, Stefano Liberati, and Sebastiano Sonego. “Weak equivalence principle for self-
gravitating bodies: A sieve for purely metric theories of gravity”. In: Phys. Rev. D (2014).
[17] In: (2005). url: http://www.slimy.com/~steuard/teaching/tutorials/GRtorsion.pdf.

172
[18] P. K. Townsend. Black Holes. 1997. arXiv: gr-qc/9707012 [gr-qc].
[19] Erwin Schrödinger. Space-Time Structure. Cambridge Science Classics. Cambridge University
Press, 1985. doi: 10.1017/CBO9780511586446.
[20] Kris Pardo et al. “Limits on the number of spacetime dimensions from GW170817”. In: Journal
of Cosmology and Astro-Particle Physics (2018).
[21] R. Bufalo, M. Oksanen, and A. Tureanu. “How unimodular gravity theories di↵er from general
relativity at quantum level”. In: Eur. Phys. J. C75.10 (2015), p. 477. doi: 10.1140/epjc/
s10052-015-3683-3. arXiv: 1505.04978 [hep-th].
[22] Stanley Deser. “Selfinteraction and gauge invariance”. In: Gen. Rel. Grav. 1 (1970), pp. 9–18.
doi: 10.1007/BF00759198. arXiv: gr-qc/0411023 [gr-qc].
[23] T. Padmanabhan. “From gravitons to gravity: Myths and reality”. In: Int. J. Mod. Phys. D17
(2008), pp. 367–398. doi: 10.1142/S0218271808012085. arXiv: gr-qc/0409089 [gr-qc].
[24] Carlos Barceló, Raúl Carballo-Rubio, and Luis J. Garay. “Unimodular gravity and general
relativity from graviton self-interactions”. In: Phys. Rev. D89.12 (2014), p. 124019. doi: 10.
1103/PhysRevD.89.124019. arXiv: 1401.2941 [gr-qc].
[25] Domenico Giulini. “What is (not) wrong with scalar gravity?” In: Stud. Hist. Phil. Sci. B39.1
(2008), pp. 154–180. doi: 10.1016/j.shpsb.2007.09.001. arXiv: gr-qc/0611100 [gr-qc].
[26] Yvonne Choquet-Bruhat. General Relativity and the Einstein Equations. United Kingdom:
Oxford University Press, 2009.
[27] Steven Willison. “Lovelock gravity and Weyl’s tube formula”. In: Phys. Rev. D80 (2009),
p. 064018. doi: 10.1103/PhysRevD.80.064018. arXiv: 0904.3224 [gr-qc].
[28] Sveta Golod and Tsvi Piran. “Choptuik’s Critical Phenomenon in Einstein-Gauss-Bonnet Grav-
ity”. In: Phys. Rev. D85 (2012), p. 104015. doi: 10 . 1103 / PhysRevD . 85 . 104015. arXiv:
1201.6384 [gr-qc].
[29] C. Brans and R. H. Dicke. “Mach’s principle and a relativistic theory of gravitation”. In: Phys.
Rev. 124 (1961). [,142(1961)], pp. 925–935. doi: 10.1103/PhysRev.124.925.
[30] Thomas P. Sotiriou and Valerio Faraoni. “f(R) theories of gravity”. In: Reviews of Modern
Physics (2010).
[31] Antonio De Felice and Shinji Tsujikawa. “f( R) Theories”. In: Living Reviews in Relativity
(2010).
[32] Thomas P. Sotiriou. “f(R) gravity and scalar tensor theory”. In: Classical and Quantum Gravity
(2006).
[33] Gregory Walter Horndeski. “Second-Order Scalar-Tensor Field Equations in a Four-Dimensional
Space”. In: International Journal of Theoretical Physics (1974).
[34] Neil Barnaby and Niky Kamran. “Dynamics with infinitely many derivatives: The Initial value
problem”. In: JHEP 02 (2008), p. 008. doi: 10 . 1088 / 1126 - 6708 / 2008 / 02 / 008. arXiv:
0709.3968 [hep-th].
[35] David Langlois and Karim Noui. “Degenerate higher derivative theories beyond Horndeski:
evading the Ostrogradski instability”. In: JCAP 1602.02 (2016), p. 034. doi: 10.1088/1475-
7516/2016/02/034. arXiv: 1510.06930 [gr-qc].
[36] Jibril Ben Achour et al. “Degenerate higher order scalar-tensor theories beyond Horndeski up to
cubic order”. In: JHEP 12 (2016), p. 100. doi: 10.1007/JHEP12(2016)100. arXiv: 1608.08135
[hep-th].
[37] I. L. Shapiro. “Physical aspects of the space-time torsion”. In: Phys. Rept. 357 (2002), p. 113.
doi: 10.1016/S0370-1573(01)00030-8. arXiv: hep-th/0103093 [hep-th].

173
[38] Thomas P. Sotiriou and Stefano Liberati. “Metric-affine f(R) theories of gravity”. In: Annals
Phys. 322 (2007), pp. 935–966. doi: 10.1016/j.aop.2006.06.002. arXiv: gr- qc/0604006
[gr-qc].
[39] Ted Jacobson. “Einstein-aether gravity: A Status report”. In: PoS QG-PH (2007), p. 020. doi:
10.22323/1.043.0020. arXiv: 0801.1547 [gr-qc].
[40] Thomas P. Sotiriou. “Horava-Lifshitz gravity: a status report”. In: J. Phys. Conf. Ser. 283
(2011), p. 012034. doi: 10.1088/1742-6596/283/1/012034. arXiv: 1010.3218 [hep-th].
[41] Charles W. Misner, K. S. Thorne, and J. A. Wheeler. Gravitation. San Francisco: W. H. Free-
man, 1973. isbn: 9780716703440, 9780691177793.
[42] G. Jannes et al. “Experimental demonstration of the supersonic-subsonic bifurcation in the
circular jump: A hydrodynamic white hole”. In: Phys. Rev. E83 (2011), p. 056312. doi: 10.
1103/PhysRevE.83.056312. arXiv: 1010.1701 [physics.flu-dyn].
[43] Carlos Barceló, Stefano Liberati, and Matt Visser. “Analogue Gravity”. In: Living Reviews in
Relativity (2005).
[44] I. Racz and R. M. Wald. “Extensions of spacetimes with Killing horizons”. In: Classical and
Quantum Gravity (1992).
[45] Matt Visser. Lorentzian wormholes. From Einstein to Hawking. 1995.
[46] Bethan Cropp, Stefano Liberati, and Matt Visser. “Surface gravities for non-Killing horizons”.
In: Class. Quant. Grav. 30 (2013), p. 125001. doi: 10.1088/0264-9381/30/12/125001. arXiv:
1302.2383 [gr-qc].
[47] Carlos Barcelo et al. “Hawking-like radiation from evolving black holes and compact horizonless
objects”. In: JHEP 02 (2011), p. 003. doi: 10.1007/JHEP02(2011)003. arXiv: 1011.5911
[gr-qc].
[48] Vitor Cardoso et al. “Quasinormal modes and Strong Cosmic Censorship”. In: Phys. Rev. Lett.
120.3 (2018), p. 031103. doi: 10.1103/PhysRevLett.120.031103. arXiv: 1711.10502 [gr-qc].
[49] G. F. R. Ellis and T. Rothman. “Lost horizons”. In: American Journal of Physics 61.10 (1993),
pp. 883–893. doi: 10.1119/1.17400. eprint: https://doi.org/10.1119/1.17400. url:
https://doi.org/10.1119/1.17400.
[50] Tamara M. Davis and Charles H. Lineweaver. “Expanding confusion: common misconceptions
of cosmological horizons and the superluminal expansion of the universe”. In: Proc. Astron.
Soc. Austral. (2003). [Publ. Astron. Soc. Austral.21,97(2004)]. doi: 10.1071/AS03040. arXiv:
astro-ph/0310808 [astro-ph].
[51] Matt Visser. “Physical observability of horizons”. In: Phys. Rev. D90.12 (2014), p. 127502. doi:
10.1103/PhysRevD.90.127502. arXiv: 1407.7295 [gr-qc].
[52] Koushik Balasubramanian and Sujan P. Dabholkar. “Time-Dependent Warping and Non-
Singular Bouncing Cosmologies”. In: (2014). arXiv: 1401.7015 [hep-th].
[53] Abhay Ashtekar and Badri Krishnan. “Dynamical horizons and their properties”. In: Phys.
Rev. D68 (2003), p. 104030. doi: 10 . 1103 / PhysRevD . 68 . 104030. arXiv: gr - qc / 0308033
[gr-qc].
[54] Abhay Ashtekar and Badri Krishnan. “Isolated and Dynamical Horizons and Their Applica-
tions”. In: Living Reviews in Relativity 7.1 (2004), p. 10. doi: 10.12942/lrr-2004-10. url:
https://doi.org/10.12942/lrr-2004-10.
[55] Roger Penrose. “Gravitational Collapse and Space-Time Singularities”. In: Phys. Rev. Lett. 14
(3 1965), pp. 57–59.
[56] I. D. Novikov and Valeri P. Frolov. PHYSICS OF BLACK HOLES. Vol. 27. Dordrecht, Nether-
lands: Kluwer Academic, 1989. doi: 10.1007/978-94-017-2651-1.

174
[57] S. W. Hawking. “The Occurrence of Singularities in Cosmology. III. Causality and Singulari-
ties”. In: Proceedings of the Royal Society of London Series A (1967).
[58] S. W. Hawking and G. F. R. Ellis. “The Cosmic Black-Body Radiation and the Existence of
Singularities in Our Universe”. In: Astrophys. J. (1968).
[59] B. P. Abbott et al. “GWTC-1: A Gravitational-Wave Transient Catalog of Compact Binary
Mergers Observed by LIGO and Virgo during the First and Second Observing Runs”. In: Phys.
Rev. X 9 (3 2019), p. 031040.
[60] The EHT Collaboration et al. “First M87 Event Horizon Telescope Results. I. The Shadow of
the Supermassive Black Hole”. In: ApJL 875 (2019), p. 1. url: https://iopscience.iop.
org/article/10.3847/2041-8213/ab0ec7.
[61] Raúl Carballo-Rubio et al. “Geodesically complete black holes”. In: (2019). arXiv: 1911.11200
[gr-qc].
[62] Kip S. Thorne, Richard H. Price, and Douglas A. MacDonald. Black holes: The membrane
paradigm. 1986.
[63] Matt Visser. “The Kerr spacetime: A Brief introduction”. In: Kerr Fest: Black Holes in Astro-
physics, General Relativity and Quantum Gravity Christchurch, New Zealand, August 26-28,
2004. 2007. arXiv: 0706.0622 [gr-qc].
[64] Daniel Kapec and Alexandru Lupsasca. “Particle motion near high-spin black holes”. In: Class.
Quant. Grav. 37.1 (2020), p. 015006. doi: 10.1088/1361-6382/ab519e. arXiv: 1905.11406
[hep-th].
[65] Richard Brito, Vitor Cardoso, and Paolo Pani. “Superradiance”. In: Lect. Notes Phys. 906
(2015), pp.1–237. doi: 10.1007/978-3-319-19000-6. arXiv: 1501.06570 [gr-qc].
[66] W. Unruh. “Separability of the Neutrino Equations in a Kerr Background”. In: Phys. Rev. Lett.
31 (20 1973), pp. 1265–1267.
[67] J. M. Bardeen, B. Carter, and S. W. Hawking. “The four laws of black hole mechanics”. In:
Communications in Mathematical Physics (1973).
[68] W. Israel. “Third Law of Black-Hole Dynamics: A Formulation and Proof”. In: Phys. Rev. Lett.
57.4 (1986), p. 397. doi: 10.1103/PhysRevLett.57.397.
[69] F. Belgiorno and M. Martellini. “Black holes and the third law of thermodynamics”. In: Int.
J. Mod. Phys. D13 (2004), pp. 739–770. doi: 10.1142/S0218271804004876. arXiv: gr- qc/
0210026 [gr-qc].
[70] Rouzbeh Allahverdi et al. “Reheating in Inflationary Cosmology: Theory and Applications”. In:
Annual Review of Nuclear and Particle Science 60.1 (2010), pp. 27–51. doi: 10.1146/annurev.
nucl.012809.104511. eprint: https://doi.org/10.1146/annurev.nucl.012809.104511.
url: https://doi.org/10.1146/annurev.nucl.012809.104511.
[71] Peter Roper, Paul Bresslo↵, and Andre Longtin. In: (2000).
[72] Ted Jacobson. “Introduction to quantum fields in curved space-time and the Hawking e↵ect”.
In: Lectures on quantum gravity. Proceedings, School of Quantum Gravity, Valdivia, Chile,
January 4-14, 2002. 2003, pp. 39–89. doi: 10.1007/0-387-24992-3_2. arXiv: gr-qc/0308048
[gr-qc].
[73] L. H. Ford. “Quantum field theory in curved space-time”. In: Particles and fields. Proceedings,
9th Jorge Andre Swieca Summer School, Campos do Jordao, Brazil, February 16-28, 1997. 1997,
pp. 345–388. arXiv: gr-qc/9707062 [gr-qc].
[74] YASUSHI TAKAHASHI and HIROOMI UMEZAWA. “THERMO FIELD DYNAMICS”. In:
International Journal of Modern Physics B 10.13n14 (1996), pp. 1755–1805. doi: 10.1142/
S0217979296000817. eprint: https://doi.org/10.1142/S0217979296000817. url: https:
//doi.org/10.1142/S0217979296000817.

175
[75] D. W. Sciama, P. Candelas, and D. Deutsch. “Quantum Field Theory, Horizons and Thermo-
dynamics”. In: Adv. Phys. 30 (1981), pp. 327–366. doi: 10.1080/00018738100101457.
[76] P.C. Causseè. l’Album du Marin. Charpentier, Nantes, 1836.
[77] Sipko L. Boersma. “A maritime analogy of the Casimir e↵ect”. In: American Journal of Physics
64.5 (1996), pp. 539–541. doi: 10.1119/1.18150. eprint: https://doi.org/10.1119/1.18150.
url: https://doi.org/10.1119/1.18150.
[78] Ted Jacobson. “Black holes and Hawking radiation in spacetime and its analogues”. In: Lect.
Notes Phys. 870 (2013), pp. 1–29. doi: 10.1007/978-3-319-00266-8_1. arXiv: 1212.6821
[gr-qc].
[79] S. A. Fulling and P. C. W. Davies. “Radiation from a Moving Mirror in Two Dimensional
Space-Time: Conformal Anomaly”. In: Proceedings of the Royal Society of London Series A
(1976).
[80] Stefano Liberati, Tony Rothman, and Sebastiano Sonego. “Nonthermal nature of incipient
extremal black holes”. In: Phys. Rev. D 62 (2 2000), p. 024005.
[81] S. W. Hawking. “Particle creation by black holes”. In: Communications in Mathematical Physics
(1975).
[82] S. W. Hawking. “Superscattering matrix for two-dimensional black holes”. In: Phys. Rev. D 50
(6 1994), pp. 3982–3986.
[83] Andrew Strominger. “Les Houches lectures on black holes”. In: NATO Advanced Study Institute:
Les Houches Summer School, Session 62: Fluctuating Geometries in Statistical Mechanics and
Field Theory Les Houches, France, August 2-September 9, 1994. 1994. arXiv: hep-th/9501071
[hep-th].
[84] Don N. Page. “Information in black hole radiation”. In: Phys. Rev. Lett. 71 (23 1993), pp. 3743–
3746.
[85] Ahmed Almheiri et al. “Black Holes: Complementarity or Firewalls?” In: JHEP 02 (2013),
p. 062. doi: 10.1007/JHEP02(2013)062. arXiv: 1207.3123 [hep-th].
[86] Raúl Carballo-Rubio et al. “On the viability of regular black holes”. In: JHEP 07 (2018), p. 023.
doi: 10.1007/JHEP07(2018)023. arXiv: 1805.02675 [gr-qc].
[87] Daniele Malafarina. “Classical collapse to black holes and quantum bounces: A review”. In:
Universe 3.2 (2017), p. 48. doi: 10.3390/universe3020048. arXiv: 1703.04138 [gr-qc].
[88] Joshua H. Cooperman and Markus A. Luty. “Renormalization of Entanglement Entropy and
the Gravitational E↵ective Action”. In: JHEP 12 (2014), p. 045. doi: 10.1007/JHEP12(2014)
045. arXiv: 1302.1878 [hep-th].
[89] Robert M. Wald. “On identically closed forms locally constructed from a field”. In: Journal
of Mathematical Physics 31.10 (1990), pp. 2378–2384. doi: 10.1063/1.528839. url: https:
//doi.org/10.1063/1.528839.
[90] Robert M. Wald. “Black hole entropy is the Noether charge”. In: Phys. Rev. D 48 (8 1993),
R3427–R3431.
[91] Vivek Iyer and Robert M. Wald. “Some properties of the Noether charge and a proposal for
dynamical black hole entropy”. In: Phys. Rev. D 50 (2 1994), pp. 846–864.
[92] G W Gibbons and S W Hawking. Euclidean Quantum Gravity. WORLD SCIENTIFIC, 1993.
doi: 10.1142/1301. eprint: https://www.worldscientific.com/doi/pdf/10.1142/1301.
url: https://www.worldscientific.com/doi/abs/10.1142/1301.
[93] G. W. Gibbons and S. W. Hawking. “Action Integrals and Partition Functions in Quantum
Gravity”. In: Phys. Rev. D15 (1977), pp. 2752–2756. doi: 10.1103/PhysRevD.15.2752.

176
 [94] Carlos Barceló et al. “Fate of gravitational collapse in semiclassical gravity”. In: Phys. Rev. D
(2008).
[95] Stephen A. Fulling, Mark Sweeny, and Robert M. Wald. “Singularity structure of the two-
point function quantum field theory in curved spacetime”. In: Comm. Math. Phys. 63.3 (1978),
pp. 257–264. url: https://projecteuclid.org:443/euclid.cmp/1103904566.
[96] Kurt Gödel. “An Example of a New Type of Cosmological Solutions of Einstein’s Field Equa-
tions of Gravitation”. In: Reviews of Modern Physics (1949).
[97] In: (2013). doi: 10.1088/1367-2630/15/1/013063. url: https://doi.org/10.1088%2F1367-
2630%2F15%2F1%2F013063.
[98] Frank Grave. PhD thesis. 2010.
[99] Robert W. Fuller and John A. Wheeler. “Causality and Multiply Connected Space-Time”. In:
Phys. Rev. 128 (2 1962), pp. 919–929.
[100] Michael S. Morris, Kip S. Thorne, and Ulvi Yurtsever. “Wormholes, Time Machines, and the
Weak Energy Condition”. In: Phys. Rev. Lett. 61 (13 1988), pp. 1446–1449.
[101] Michael S. Morris and Kip S. Thorne. “Wormholes in spacetime and their use for interstellar
travel: A tool for teaching general relativity”. In: American Journal of Physics 56.5 (1988),
pp. 395–412. doi: 10.1119/1.15620. eprint: https://doi.org/10.1119/1.15620. url:
https://doi.org/10.1119/1.15620.
[102] In: (1994). doi: 10.1088/0264-9381/11/5/001. url: https://doi.org/10.1088%2F0264-
9381%2F11%2F5%2F001.
[103] S. V. Krasnikov. “Hyperfast travel in general relativity”. In: Phys. Rev. D (1998).
[104] I. D. Novikov. Evolution of the universe. 1983.
[105] John Friedman et al. “Cauchy problem in spacetimes with closed timelike curves”. In: Phys.
Rev. D (1990).
[106] Bernard S. Kay, Marek J. Radzikowski, and Robert M. Wald. “Quantum Field Theory on
Spacetimes with a Compactly Generated Cauchy Horizon”. In: Communications in Mathemat-
ical Physics (1997).
[107] Stefano Finazzi, Stefano Liberati, and Carlos Barceló. “Semiclassical instability of dynamical
warp drives”. In: Phys. Rev. D (2009).
[108] Carlos Barceló, Stefano Finazzi, and Stefano Liberati. “On the impossibility of superluminal
travel: the warp drive lesson”. In: arXiv e-prints (2010).
[109] Eric Davis. In: (2004).

177