Mathematical Relativity

José Natário

Lisbon, 2021
 Contents

Preface 3
Chapter 1. Preliminaries 5
1. Special relativity 5
2. Differential geometry: mathematicians vs physicists 10
3. General relativity 13
4. Exercises 14
Chapter 2. Exact solutions 17
1. Minkowski spacetime 17
2. Penrose diagrams 18
3. The Schwarzschild solution 21
4. Friedmann-Lemaı̂tre-Robertson-Walker models 27
5. Matching 37
6. Oppenheimer-Snyder collapse 39
7. Exercises 43
Chapter 3. Causality 55
1. Past and future 55
2. Causality conditions 59
3. Exercises 60
Chapter 4. Singularity theorems 63
1. Geodesic congruences 63
2. Energy conditions 66
3. Conjugate points 67
4. Existence of maximizing geodesics 74
5. Hawking’s singularity theorem 78
6. Penrose’s singularity theorem 80
7. Exercises 86
Chapter 5. Cauchy problem 89
1. Divergence theorem 89
2. Klein-Gordon equation 91
3. Maxwell’s equations: constraints and gauge 96
4. Einstein’s equations 98
5. Constraint equations 102
6. Einstein equations with matter 104
1
2 CONTENTS

7. Exercises 104
Chapter 6. Mass in general relativity 109
1. Komar mass 109
2. Field theory 114
3. Einstein-Hilbert action 120
4. Gravitational waves 124
5. ADM mass 126
6. Positive mass theorem 129
7. Penrose inequality 130
8. Exercises 132
Chapter 7. Black holes 139
1. The Kerr solution 139
2. Killing horizons and the zeroth law 141
3. Smarr’s formula and the first law 145
4. Second law 147
5. Black hole thermodynamics and Hawking radiation 150
6. Exercises 151
Appendix: Mathematical concepts for physicists 155
Topology 155
Metric spaces 156
Hopf-Rinow theorem 156
Differential forms 156
Lie derivative 158
Cartan structure equations 160
Bibliography 163
Index 165
 Preface

These lecture notes were written for a one-semester course in mathemat-

ical relativity, aimed at mathematics and physics students, which has been
taught at Instituto Superior Técnico (Universidade de Lisboa) since 2010.
They are not meant as an introduction to general relativity, but rather as a
complementary, more advanced text, much like Part II of Wald’s textbook
[Wal84], on which they are loosely based. It is assumed that the reader
is familiar at least with special relativity, and has taken a course either in
Riemannian geometry (typically the mathematics students) or in general
relativity (typically the physics students). In other words, the reader is ex-
pected to be proficient in (some version of) differential geometry, and also
to be acquainted with the basic principles of relativity.
The aim of the course is to introduce the students to some of the most
important mathematical results of general relativity: the singularity theo-
rems of Hawking and Penrose, the Choquet-Bruhat theorem concerning the
Cauchy problem, the positive mass theorem of Schoen and Yau, and the the-
orems on black hole uniqueness and black hole thermodynamics. To be able
to do this, and also to provide familiarity with some simple examples, the
course begins with an initial discussion of exact solutions, Penrose diagrams
and causality.
It is, of course, impossible to give full proofs of all these results in a one-
semester course. Consequently, I was forced on many occasions to present
only proofs of simple particular cases, or even just brief sketches of proofs;
in all cases I have carefully indicated this. In writing these notes I strove for
brevity, and always opted for presenting the simplest version of the results:
this course should be regarded as panoramic rather than exhaustive. I have
also assumed some degree of maturity from the part of the reader, who is
trusted to look up possibly unfamiliar mathematical results used in the text
(some of which are, in anticipation, listed in the appendix). I have included
a number of exercises (about 10 per chapter), designed both to fill in some of
the calculations not presented in the main text and also to test and extend
the understanding of the material in that chapter.
Finally, I thank the many colleagues and students who read this text,
or parts of it, for their valuable comments and suggestions. Special thanks
are due to my colleague and friend Pedro Girão.

3
 CHAPTER 1

Preliminaries

In this initial chapter we give a very short introduction to special and

general relativity for mathematicians. In particular, we relate the index-free
differential geometry notation used in mathematics (e.g. [O’N83, dC93,
Boo03, GN14]) to the index notation used in physics (e.g. [MTW73,
Wal84, HE95]). As an exercise in index gymnastics, we derive the con-
tracted Bianchi identities.

1. Special relativity
Consider an inertial frame S ′ moving with velocity v with respect to
another inertial frame S along their common x-axis (Figure 1). According
to classical mechanics, the coordinate x′ of a point P in the frame S ′ is
related to its x coordinate in the frame S by
x′ = x − vt.
Moreover, a clock in S ′ initially synchronized with a clock in S is assumed
to keep the same time:
t′ = t.
Thus the spacetime coordinates of events are related by a so-called Galileo
transformation (
x′ = x − vt
.
t′ = t
If the point P is moving, its velocity in S ′ is related to its velocity in S
by
dx′ dx − vdt dx
′
= = − v.
dt dt dt
This is in conflict with the experimental fact that the speed of light is the
same in every inertial frame, indicating that classical mechanics is not cor-
rect. Einstein solved this problem in 1905 by replacing the Galileo transfor-
mation by the so-called Lorentz transformation:
(
x′ = γ(x − vt)
.
t′ = γ(t − vx)
Here
1
γ=√ ,
1 − v2
5
6 1. PRELIMINARIES

S S′

y y′ P

vt x′

Figure 1. Galileo transformation.

and we are using units such that the speed of light is c = 1 (for example
measuring time in years and distance in light-years). Note that if |v| is much
smaller than the speed of light, |v| ≪ 1, then γ ≃ 1, and we retrieve the
Galileo transformation (assuming v xt ≪ 1).
Under the Lorentz transformation velocities transform as
dx
dx′ γ(dx − vdt) dt − v
= = .
dt′ γ(dt − vdx) 1 − v dx
dt
In particular,
dx dx′ 1−v
=1⇒ ′ = = 1,
dt dt 1−v
that is, the speed of light is the same in the two inertial frames.
In 1908, Minkowski noticed that
−(dt′ )2 + (dx′ )2 = −γ 2 (dt − vdx)2 + γ 2 (dx − vdt)2 = −dt2 + dx2 ,
that is, the Lorentz transformations could be seen as isometries of R4 with
the indefinite metric
ds2 = −dt2 + dx2 + dy 2 + dz 2 = −dt ⊗ dt + dx ⊗ dx + dy ⊗ dy + dz ⊗ dz.
Definition 1.1. The pseudo-Riemannian manifold (R4 , ds2 ) ≡ (R4 , h·, ·i)
is called the Minkowski spacetime.
Note that the set of vectors with zero square form a cone (the so-called
light cone):
hv, vi = 0 ⇔ −(v 0 )2 + (v 1 )2 + (v 2 )2 + (v 3 )2 = 0.
Definition 1.2. A vector v ∈ R4 is said to be:
(1) timelike if hv, vi < 0;
(2) spacelike if hv, vi > 0;
(3) lightlike, or null, if hv, vi = 0.
 1. SPECIAL RELATIVITY 7

(4) causal if it is timelike or null;

∂
(5) future-pointing if it is causal and v, ∂t < 0.
(6) past-pointing if −v is future-pointing.
The same classification applies to (smooth) curves c : [a, b] → R4 according
to its tangent vector (Figure 2).

timelike future-pointing vector

∂
∂t

null vector ∂
∂y

p ∂
spacelike vector ∂x

Figure 2. Minkowski geometry (traditionally represented

with the t-axis pointing upwards).

1
The length |hv, vi| 2 of a timelike (resp. spacelike) vector v ∈ R4 repre-
sents the time (resp. distance) measured between two events p and p + v in
the inertial frame where these events happen in the same location (resp. are
simultaneous). If c : [a, b] → R4 is a timelike curve then its length
Z b
1
τ (c) = |hċ(s), ċ(s)i| 2 ds
a

represents the proper time measured by the particle between events c(a)
and c(b). We have:
Proposition 1.3. (Twin paradox) Of all timelike curves connecting
two events p, q ∈ R4 , the curve with maximal length is the line segment
(representing inertial motion).
Proof. We may assume p = (0, 0, 0, 0) and q = (T, 0, 0, 0) on some
inertial frame, and parameterize any timelike curve connecting p to q by the
time coordinate:
c(t) = (t, x(t), y(t), z(t)).
8 1. PRELIMINARIES

Therefore
Z T 1
Z T Z T

1
τ (c) = −1 + ẋ2 + ẏ 2 + ż 2 2
dt = 2 2
1 − ẋ − ẏ − ż 2 2
dt ≤ 1dt = T.
0 0 0

Most problems in special relativity can be recast as questions about the
geometry of the Minkowski spacetime.
Proposition 1.4. (Doppler effect) An observer moving with velocity v
away from a source of light of period T measures the period to be
r
′ 1+v
T =T .
1−v
Proof. Figure 3 represents two light signals emitted by an observer at
rest at x = 0 with a time difference T . These signals are detected by an
observer moving with velocity v, who measures a time difference T ′ between
them. Now, if the first signal is emitted at t = t0 , its history is the line
t = t0 + x. Consequently, the moving observer detects the signal at the
event with coordinates given by the solution of
t0

t = 1 − v

 t = t0 + x


 
⇔ .


x = vt


x =
 vt 0
1−v

t
x = vt

T′

Figure 3. Doppler effect.

Similarly, the second light signal is emitted at t = t0 + T , its history is

the line t = t0 + T + x, and it is detected by the moving observer at the
 1. SPECIAL RELATIVITY 9

event with coordinates given by

t0 + T

 t=
1−v



.
x = v(t0 + T )



1−v
Therefore the time difference between the signals as measured by the moving
observer is
s 2 
vt0 2

′ t0 + T t0 v(t0 + T )
T = − − −
1−v 1−v 1−v 1−v
s s r
T2 v2T 2 1 − v2 1+v
= 2
− 2
=T 2
=T .
(1 − v) (1 − v) (1 − v) 1−v

In particular, two observers at rest in an inertial frame measure the same

frequency for a light signal (Figure 4). However, because the gravitational
field couples to all forms of energy (as E = mc2 ), one expects that a photon
climbing in a gravitational field to lose energy, hence frequency. In 1912,
Einstein realized that this could be modelled by considering curved space-
time geometries, so that equal line segments in a (flat) spacetime diagram
do not necessarily correspond to the same length.

T′ = T

Figure 4. Minkowski geometry is incompatible with the

gravitational redshift.
10 1. PRELIMINARIES

2. Differential geometry: mathematicians vs physicists

Einstein’s idea to incorporate gravitation into relativity was to replace
the Minkowski spacetime (R4 , h·, ·i) by a curved four-dimensional Lorentzian
manifold (M, g) ≡ (M, h·, ·i). Here g is a Lorentzian metric, that is, a
symmetric 2-tensor field such that at each tangent space g = diag(−1, 1, 1, 1)
in an appropriate basis. Just like in Riemannian geometry, g determines a
Levi-Civita connection, the unique connection ∇ which is symmetric and
compatible with g:
∇X Y − ∇Y X = [X, Y ];
X · hY, Zi = h∇X Y, Zi + hY, ∇X Zi,
for all vector fields X, Y, Z. The curvature of this connection is then given
by the operator
R(X, Y )Z = ∇X ∇Y Z − ∇Y ∇X Z − ∇[X,Y ] Z.
The formulas above were written using the abstract notation usually
employed by mathematicians. It is often very useful (especially when dealing
with contractions) to use the more explicit notation usually adopted by
physicists, which emphasizes the indices of the various objects when written
in local coordinates:
Object Mathematicians Physicists
Vector field X Xµ
Covector field ω ωµ
Tensor product X ⊗ω X µ ων
Metric g ≡ h·, ·i gµν
Inner product g(X, Y ) ≡ hX, Y i gµν X µ Y ν
Associated covector X ♭ ≡ g(X, ·) Xν ≡ gµν X µ
Inverse metric g −1 g µν
Associated vector ω ≡ g −1 (ω, ·)
♯ ω ≡ g µν ωµ
ν

Derivative of a function df ∇µ f ≡ ∂ µ f
Covariant derivative ∇X Y X µ ∇µ Y ν
Covariant derivative tensor ∇X ∇µ X ≡ ∂µ X ν + Γνµα X α
ν

Here Γαµν are the Christoffel symbols of the Levi-Civita connection;

they can be computed from the components gµν of the metric tensor by the
formula1
1
Γαµν = g αβ (∂µ gνβ + ∂ν gµβ − ∂β gµν ) ,
2
and in turn be used to compute the components of the Riemann curvature
tensor:
Rαβ µν = dxµ (R(∂α , ∂β )∂ν ) = ∂α Γµβν − ∂β Γµαν + Γµαγ Γγβν − Γµβγ Γγαν .
1We use Einstein’s summation convention, with Greek indices running from 0 to 3
and Latin indices running from 1 to 3.
 2. DIFFERENTIAL GEOMETRY: MATHEMATICIANS VS PHYSICISTS 11

The covariant derivative tensor of a vector field X, not always empha-

sized in differential geometry courses for mathematicians, is simply the (1, 1)-
tensor field defined by
∇X(Y ) = ∇Y X.
Also not always emphasized in differential geometry courses for mathemati-
cians is the fact that any connection can be naturally extended to act on
tensor fields (via the Leibniz rule). For instance, if ω is a covector field
(1-form) then one defines
(∇X ω)(Y ) = X · [ω(Y )] − ω(∇X Y ).
In local coordinates, this is
(X µ ∇µ ων )Y ν = X µ ∂µ (ων Y ν ) − ων (X µ ∇µ Y ν )
= X µ (∂µ ων )Y ν + X µ ων ∂µ Y ν − ων (X µ ∂µ Y ν + X µ Γνµα Y α )
= (∂µ ων − Γαµν ωα )X µ Y ν ,
that is,
∇µ ων = ∂µ ων − Γαµν ωα .
The generalization for higher rank tensors is obvious: for instance, if T is a
(2, 1)-tensor then
β β
∇α Tµν = ∂α Tµν + Γβαγ Tµν
γ
− Γγαµ Tγν
β
− Γγαν Tµγ
β
.
Note that the condition of compatibility of the Levi-Civita connection with
the metric is simply
∇g = 0.
In particular, the operations of raising and lowering indices commute with
covariant differentiation.
As an exercise in index gymnastics, we will now derive a series of iden-
tities involving the Riemann curvature tensor. We start by rewriting its
definition in the notation of the physicists:
Rαβ µν X α Y β Z ν =
= X α ∇α (Y β ∇β Z µ ) − Y α ∇α (X β ∇β Z µ ) − (X α ∇α Y β − Y α ∇α X β )∇β Z µ
= (X α ∇α Y β )(∇β Z µ ) + X α Y β ∇α ∇β Z µ − (Y α ∇α X β )(∇β Z µ )
− Y α X β ∇α ∇β Z µ − (X α ∇α Y β )∇β Z µ + (Y α ∇α X β )∇β Z µ
= X α Y β (∇α ∇β − ∇β ∇α )Z µ .
In other words,
Rαβ µν Z ν = (∇α ∇β − ∇β ∇α )Z µ ,
or, equivalently,
(1) 2∇[α ∇β] Zµ = Rαβµν Z ν ,
12 1. PRELIMINARIES

where the square brackets indicate anti-symmetrization2. This is readily

generalized for arbitrary tensors: from
2∇[α ∇β] (Zµ Wν ) = (2∇[α ∇β] Zµ )Wν + (2∇[α ∇β] Wν )Zµ
= Rαβµσ Z σ Wν + Rαβνσ W σ Zµ
one concludes that
2∇[α ∇β] Tµν = Rαβµσ T σν + Rαβνσ Tµ σ .
Let us choose
Zµ = ∇µ f ≡ ∂ µ f
in equation (1). We obtain
R[αβµ]ν Z ν = 2∇[[α ∇β] Zµ] = 2∇[α ∇[β Zµ]] = 0,
because
∇[µ Zν] = ∂[µ Zν] − Γα[µν] Zα = ∂[µ ∂ν] f = 0.
Since we can choose Z arbitrarily at a given point, it follows that
R[αβµ]ν = 0 ⇔ Rαβµν + Rβµαν + Rµαβν = 0.
This is the so-called first Bianchi identity, and is key for obtaining the
full set of symmetries of the Riemann curvature tensor:
Rαβµν = −Rβαµν = −Rαβνµ = Rµναβ .
In the notation of the mathematicians, it is written as
R(X, Y )Z + R(Y, Z)X + R(Z, X)Y = 0
for all vector fields X, Y, Z.
Let us now take the covariant derivative of equation (1):
∇γ Rαβµν Z ν + Rαβµν ∇γ Z ν = 2∇γ ∇[α ∇β] Zµ .
At any given point we can choose Z such that
∇γ Z ν ≡ ∂γ Z ν + Γνγδ Z δ = 0.
Assuming this, we then obtain3
∇[γ Rαβ]µν Z ν = 2∇[γ ∇[α ∇β]] Zµ = 2∇[[γ ∇α] ∇β] Zµ
= R[γαβ]δ ∇δ Zµ + R[γα|µδ| ∇β] Z δ = 0.
Since we can choose Z arbitrarily at a given point, it follows that
(2) ∇[α Rβγ]µν = 0 ⇔ ∇α Rβγµν + ∇β Rγαµν + ∇γ Rαβµν = 0
This is the so-called second Bianchi identity. In the notation of the
mathematicians, it is written as
∇R(X, Y, Z, ·, ·) + ∇R(Y, Z, X, ·, ·) + ∇R(Z, X, Y, ·, ·) = 0
2Thus T = 1
(Tαβ − Tβα ), T[αβγ] = 1
(Tαβγ + Tβγα + Tγαβ − Tβαγ − Tαγβ − Tγβα ),
[αβ] 2 6
etc.
3In the formula below the indices between vertical bars are not anti-symmetrized.
 3. GENERAL RELATIVITY 13

for all vector fields X, Y, Z.

Recall that the Riemann curvature tensor has only one independent
contraction, called the Ricci tensor:
Rµν = Rαµ αν .
The trace of the Ricci tensor, in turn, is known as the scalar curvature:
R = g µν Rµν .
These quantities satisfy the so-called contracted Bianchi identity, which
is obtained from (2) by contracting the pairs of indices (β, µ) and (γ, ν):
 
β γ β 1 β 1
∇α R−∇ Rαβ −∇ Rαγ = 0 ⇔ ∇ Rαβ − ∇α R = 0 ⇔ ∇ Rαβ − Rgαβ = 0.
2 2
The contracted Bianchi identity is equivalent to the statement that the Ein-
stein tensor
1
Gµν = Rµν − Rgµν
2
is divergenceless:
∇µ Gµν = 0.

3. General relativity
Newtonian gravity is described by a scalar function φ, called the grav-
itational potential. The equation of motion for a free-falling particle of
mass m in Cartesian coordinates is
d2 xi d2 xi
m 2 = −m∂i φ ⇔ = −∂i φ.
dt dt2
Note that all free-falling particles follow the same trajectories (an observa-
tion dating back to Galileo). The gravitational potential is determined from
the matter mass density ρ by the Poisson equation
∆φ = 4πρ
(using units such that Newton’s gravitational constant is G = 1; this choice,
together with c = 1, defines the so-called geometrized units, where lengths,
time intervals and masses all have the same dimensions).
To implement his idea of describing gravity via a curved four-dimensional
Lorentzian manifold (M, g), Einstein had to specify (i) how free-falling par-
ticles would move on this manifold, and (ii) how to determine the curved
metric g. Since free particles move along straight lines in the Minkowski
spacetime, Einstein proposed that free falling particles should move along
timelike geodesics. In other words, he suggested replacing the Newtonian
equation of motion by the geodesic equation
ẍµ + Γµαβ ẋα ẋβ = 0.
Moreover, Einstein knew that it is possible to define the energy-momentum
tensor Tµν for the matter content of the Minkowski spacetime, so that the
14 1. PRELIMINARIES

conservation of energy and momentum is equivalent to the vanishing of its

divergence:
∇µ Tµν = 0.
This inspired Einstein to propose that g should satisfy the so-called Einstein
field equations:
Gµν + Λgµν = 8πTµν .
Here Λ is a constant, known as the cosmological constant. Note that the
Einstein field equations imply, via the contracted Bianchi identity, that the
energy-momentum tensor is divergenceless.
As a simple example, we consider a pressureless perfect fluid, known as
dust. Its energy-momentum tensor is
Tµν = ρUµ Uν ,
where ρ is the dust rest density and U is a unit timelike vector field tangent
to the histories of the dust particles. The equations of motion for the dust
can be found from
∇µ Tµν = 0 ⇔ [∇µ (ρUµ )] Uν + ρUµ ∇µ Uν = 0
⇔ div(ρU )U + ρ∇U U = 0.
Since U and ∇U U are orthogonal (because hU, U i = −1), we find
(
div(ρU ) = 0
∇U U = 0

in the support of ρ. These are, respectively, the equation of conservation

of mass and the geodesic equation. Thus the fact that free-falling particles
move along geodesics can be seen as a consequence of the Einstein field
equations (at least in this model).

4. Exercises
(1) Twin paradox: Two twins, Alice and Bob, are separated on their
20th birthday. While Alice remains on Earth (which is an inertial
frame to a very good approximation), Bob departs at 80% of the
speed of light towards Planet X, 8 light-years away from Earth.
Therefore Bob reaches his destination 10 years later (as measured
on the Earth’s frame). After a short stay, he returns to Earth,
again at 80% of the speed of light. Consequently Alice is 40 years
old when she sees Bob again.
(a) How old is Bob when they meet again?
(b) How can the asymmetry in the twins’ ages be explained? No-
tice that from Bob’s point of view he is at rest in his spaceship
and it is the Earth which moves away and then back again.
 4. EXERCISES 15

(c) Imagine that each twin watches the other trough a very pow-
erful telescope. What do they see? In particular, how much
time do they experience as they see one year elapse for their
twin?
(2) A particularly simple matter model is that of a smooth massless
scalar field φ : M → R, whose energy-momentum tensor is
1
Tµν = ∂µ φ ∂ν φ − (∂α φ ∂ α φ)gµν .
2
Show that if the Lorentzian manifold (M, g) satisfies the Einstein
equations with this matter model then φ satisfies the wave equa-
tion
 φ = 0 ⇔ ∇µ ∂µ φ = 0.
(3) The energy-momentum tensor for perfect fluid is
Tµν = (ρ + p)Uµ Uν + pgµν ,
where ρ is the fluid’s rest density, p is the fluid’s rest pressure, and
U is a unit timelike vector field tangent to the histories of the fluid
particles. Show that:
(a) (Tµν ) = diag(ρ, p, p, p) in any orthonormal frame including U ;
(b) the motion equations for the perfect fluid are
(
div(ρU ) + p div U = 0
,
(ρ + p)∇U U = −(grad p)⊥
where ⊥ represents the orthogonal projection on the spacelike
hyperplane orthogonal to U .
 CHAPTER 2

Exact solutions

In this chapter we present a number of exact solutions of the Einstein

field equations, construct their Penrose diagrams, and analyze their match-
ing across a timelike hypersurface. These solutions will be used as examples
or counter-examples when discussing the theorems in the subsequent chap-
ters. A different perspective on Penrose diagrams can be found in [HE95].

1. Minkowski spacetime
The simplest solution of the Einstein field equations with zero cosmo-
logical constant in vacuum (i.e. with vanishing energy-momentum tensor) is
the Minkowski spacetime, that is, R4 with the metric
ds2 = −dt2 + dx2 + dy 2 + dz 2 .
Since this metric is flat, its curvature vanishes, and so do its Ricci and
Einstein tensors. It represents a universe where there is no gravity whatso-
ever. Transforming the Cartesian coordinates (x, y, z) to spherical coordi-
nates (r, θ, ϕ) yields
ds2 = −dt2 + dr2 + r2 dθ2 + sin2 θdϕ2 .

Performing the additional change of coordinates

(
u=t−r (retarded time)
v =t+r (advanced time)
we obtain
ds2 = −du dv + r2 dθ2 + sin2 θdϕ2 ,

where
1
r(u, v) = (v − u).
2
The coordinates (u, v) are called null coordinates: their level sets are null
cones formed by outgoing/ingoing null geodesics emanating from the center.
Note that they are subject to the constraint
r ≥ 0 ⇔ v ≥ u.
Finally, the coordinate change
( (
ũ = tanh u u = arctanh ũ
(3) ⇔
ṽ = tanh v v = arctanh ṽ
17
18 2. EXACT SOLUTIONS

brings the metric into the form

1
ds2 = − 2 2 2 2


dũ dṽ + r dθ + sin θdϕ ,
(1 − ũ2 ) (1 − ṽ 2 )
where now
1
r (ũ, ṽ) = (arctanh ṽ − arctanh ũ)
2
and
(4) − 1 < ũ ≤ ṽ < 1.
Because (ũ, ṽ) are also null coordinates, it is common to represent their axes
tilted by 45◦ . The plane region defined by (4) is then represented in Figure 1.
ũ ṽ

Figure 1. Range of the coordinates (ũ, ṽ).

This region is usually called the Penrose diagram for the Minkowski
spacetime. If we take each point in the diagram to represent a sphere S 2 of
radius r (ũ, ṽ), the diagram itself represents the full spacetime manifold, in a
way that makes causality relations apparent: any causal curve is represented
in the diagram by a curve with tangent at most 45◦ from the vertical. In
Figure 2 we represent some level hypersurfaces of t and r in the Penrose
diagram. The former approach the point i0 in the boundary of the diagram,
called the spacelike infinity, whereas the later go from the boundary point
i− (past timelike infinity) to the boundary point i+ (future timelike
infinity). Finally, null geodesics start at the null boundary line I − (past
null infinity) and end at the null boundary line I + (future null infinity).
These boundary points and lines represent ideal points at infinity, and do
not correspond to actual points in the Minkowski spacetime.

2. Penrose diagrams
The concept of Penrose diagram can be easily generalized for any spher-
ically symmetric space-time. Such spacetimes have metric
ds2 = gAB dxA dxB + r2 dθ2 + sin2 θdϕ2 , r = r(x0 , x1 ),


 2. PENROSE DIAGRAMS 19

i+

I+

r=0 i0

I−

i−

Figure 2. Penrose diagram for the Minkowski spacetime,

with some level hypersurfaces of t and r represented.

where gAB dxA dxB is a Lorentzian metric on a 2-dimensional quotient man-

ifold with boundary (which we assume to be diffeomorphic to a region of
the plane). It turns out that any such metric is conformal to the Minkowski
metric:
(5) gAB dxA dxB = −Ω2 du dv, Ω = Ω(u, v).
This can be seen locally as follows: choose a spacelike line S, a coordinate
u along it, and a coordinate w along a family of null geodesics emanating
from S, so that S corresponds to w = 0 (Figure 3). Then near S we have
 
∂ ∂
guu = , >0
∂u ∂u
and  
∂ ∂
gww = , = 0.
∂w ∂w
Therefore the 2-dimensional metric is written in these coordinates
 
A B 2 2guw
gAB dx dx = guu du + 2guw dudw = guu du du + dw .
guu
As for any 1-form in a 2-dimensional manifold, we have
2guw
du + dw = f dv
guu
for suitable functions f and v. Note that f cannot vanish, because (u, w)
are local coordinates. Moreover, we can assume f < 0 by replacing v with
−v if necessary. Choosing Ω2 = −f guu then yields (5).
20 2. EXACT SOLUTIONS

Figure 3. Choice of the coordinates (u, w).

We then see that any spherically symmetric metric can be written as

ds2 = −Ω2 du dv + r2 dθ2 + sin2 θdϕ2


(6)
with Ω = Ω(u, v) and r = r(u, v). By rescaling u and v if necessary, we
can assume that the range of (u, v) is bounded, and hence obtain a Penrose
diagram depicting the causal geometry. As we will see, this is extremely
helpful in more complicated spherical symmetric solutions of the Einstein
field equations.
Remark 2.1. From
2
!
− Ω2 − Ω22
 
0 AB

0
(gAB ) = ⇒ g =
− Ω22
2
− Ω2 0 0
it is easily seen that
p 
∂A − det (gCD ) g AB ∂B u = 0 ⇔ ∇A ∇A u = 0.
and similarly for v. In other words, the null coordinates u and v are solutions
of the wave equation in the 2-dimensional Lorentzian manifold:
u = v = 0.
This is the Lorentzian analogue of the so-called isothermal coordinates
for Riemannian surfaces. The proof that the later exist locally is however
slightly more complicated: given a point p on the surface, one chooses a
local harmonic function with nonvanishing derivative,
∆u = 0, (du)p 6= 0,
and considers the equation
(7) dv = ⋆du.
Here ⋆ is the Hodge star, which for generic orientable n-dimensional pseudo-
Riemannian manifolds is defined as follows: if {ω 1 , . . . , ω n } is any positively
oriented orthonormal coframe then
⋆(ω 1 ∧ · · · ∧ ω k ) = hω 1 , ω 1 i · · · hω k , ω k i ω k+1 ∧ · · · ∧ ω n .
 3. THE SCHWARZSCHILD SOLUTION 21

By the Poincaré Lemma, equation (7) can be locally solved, since

d ⋆ du = ⋆ ⋆ d ⋆ du = ⋆(∆u) = 0.
Moreover, v is itself harmonic, because
∆v = ⋆d ⋆ dv = ⋆d ⋆ ⋆du = ⋆d(−du) = 0.
Finally,
1
kduk = kdvk =
Ω
for some local function Ω > 0, and so the metric is written is these coordi-
nates as
ds2 = Ω2 du2 + dv 2 .

3. The Schwarzschild solution

If we try to solve the vacuum Einstein field equations with zero cosmo-
logical constant for a spherically symmetric Lorentzian metric, we obtain,
after suitably rescaling the time coordinate, the Schwarzschild metric
2M −1 2
   
2 2M 2
dr + r2 dθ2 + sin2 θdϕ2


ds = − 1 − dt + 1 −
r r
(where M ∈ R is a constant). Note that for M = 0 we retrieve the Minkowski
metric in spherical coordinates. Note also that if M > 0 then the metric
is defined in two disconnected domains of coordinates, corresponding to
r ∈ (0, 2M ) and r ∈ (2M, +∞).
The physical interpretation of the Schwarzschild solution can be found
by considering the proper time of a timelike curve parameterized by the time
coordinate:
Z t 1 " #1
2M −1 2 2
  
2M 2 2 2 2 2
τ= 1− − 1− ṙ − r θ̇ − r sin θϕ̇ dt,
t0 r r
where ṙ = dr
dt , etc. The integrand LS is the Lagrangian for geodesic motion
in the Schwarzschild spacetime when parameterized by the time coordinate.
Now for motions with speeds much smaller than the speed of light we have
ṙ2 ≪ 1, etc. Assuming M r ≪ 1 as well we have
" #1
2M −1 2 2
 
2M 2 2 2 2 2
LS = 1 − − 1− ṙ − r θ̇ − r sin θϕ̇
r r
M 1 2 
≃1− − ṙ + r2 θ̇2 + sin2 θϕ̇2 = 1 − LN ,
r 2
where
1 2  M
LN = ṙ + r2 θ̇2 + r2 sin2 θϕ̇2 +
2 r
is precisely the Newtonian Lagrangian for the motion of a particle in the
gravitational field of a point mass M . The Schwarzschild solution should
therefore be considered the relativistic analogue of this field.
22 2. EXACT SOLUTIONS

To write the Schwarzschild metric in the form (6) we note that the
quotient metric is
2M −1 2
   
2 2M 2
ds = − 1 − dt + 1 − dr
r r
" #
2M −2 2
  
2M 2
=− 1− dt − 1 − dr
r r
" #" #
2M −1 2M −1
    
2M
=− 1− dt − 1 − dr dt + 1 − dr
r r r
 
2M
=− 1− du dv,
r
where we define
2M −1
Z  
u=t− 1− dr = t − r − 2M log |r − 2M |
r
and
2M −1
Z  
v =t+ 1− dr = t + r + 2M log |r − 2M |.
r
In the domain of coordinates r > 2M we have 1 − 2M r > 0, and so the
quotient metric is already in the required form. Note however that, unlike
what happened in the Minkowski spacetime, we now have
v − u = 2r + 4M log |r − 2M | ∈ (−∞, +∞).
Consequently, by applying the coordinate rescaling (3) we obtain the full
square, instead of a triangle (Figure 4). Besides the infinity points and null
boundaries also present in the Penrose diagram for the Minkowski spacetime,
there are two new null boundaries, H − (past event horizon) and H +
(future event horizon), where r = 2M .
It seems reasonable to expect that the metric can be extended across the
horizons, since r does not tend to zero nor to infinity there; this expectation
is confirmed by calculating the so-called Kretschmann scalar:
48M 2
Rαβµν Rαβµν = .
r6
This is perfectly well behaved as r → 2M , and seems to indicate that the
horizons are mere singularities of the coordinate system (t, r). To show that
this is indeed the case, note that in the (u, r) coordinate system the quotient
metric is written
 
2 2M
ds = − 1 − du2 − 2du dr.
r
Since
−1 + 2M
 
r −1
det = −1,
−1 0
 3. THE SCHWARZSCHILD SOLUTION 23

i+

H+ I+

i0

H− I−

i−

Figure 4. Penrose diagram for the region r > 2M of the

Schwarzschild spacetime, with some level hypersurfaces of t
and r represented.

we see that the metric is well defined across r = 2M in this coordinate

system. Moreover, we know that it solves the Einstein equations in the
coordinate domains r < 2M and r > 2M ; by continuity, it must solve it in
the whole domain r ∈ (0, +∞). Note that the coordinate domains r < 2M
and r > 2M are glued along r = 2M so that the outgoing null geodesics
u = constant go from r = 0 to r = +∞; in other words, the gluing is along
the past event horizon H − .
To obtain the Penrose diagram for the coordinate domain r < 2M we
note that the quotient metric can be written as
     
2 2M 2M 2M
ds = − 1 − du dv = − − 1 du (−dv) = − − 1 du dv ′ ,
r r r
where v ′ = −v. Since in this coordinate domain 2Mr − 1 > 0, the quotient
metric is in the required form. Note however that we now have
u + v ′ = −2r − 4M log |r − 2M | ∈ (−4M log(2M ), +∞),
and by setting
v ′′ = v ′ + 4M log(2M )
we obtain
u + v ′′ > 0.
Consequently, by applying the coordinate rescaling (3) we obtain a triangle
(Figure 5). There is now a spacelike boundary, where r = 0, and two null
boundaries H − , where r = 2M . The Penrose diagram for the domain of the
coordinates (u, r) can be obtained gluing the Penrose diagrams in Figures 4
and 5 along H − , so that the null geodesics u = constant match (Figure 6).
24 2. EXACT SOLUTIONS

H− H−

i− i−
r=0

Figure 5. Penrose diagram for a region r < 2M of the

Schwarzschild spacetime, with some level hypersurfaces of t
and r represented.

i+

H+ I+

i0

H− I−

i− r=0 i−

Figure 6. Penrose diagram for the domain of the coordi-

nates (u, r) in the Schwarzschild spacetime, with some level
hypersurfaces of t and r represented.

If instead we use the (v, r) coordinate system, the quotient metric is

written  
2 2M
ds = − 1 − dv 2 + 2dv dr.
r
Again the metric is well defined across r = 2M in this coordinate system,
since
−1 + 2M
 
r 1
det = −1,
1 0
and solves the Einstein equations in the whole coordinate domain r ∈
(0, +∞). The coordinate domains r < 2M and r > 2M are now glued along
r = 2M so that the ingoing null geodesics v = constant go from r = +∞ to
r = 0; in other words, the gluing is along the future event horizon H + .
 3. THE SCHWARZSCHILD SOLUTION 25

To obtain the Penrose diagram for the coordinate domain r < 2M we

note that the quotient metric can be written as
     
2 2M 2M 2M
ds = − 1 − du dv = − − 1 (−du) dv = − − 1 du′ dv,
r r r
2M
where u′ = −u. Since in this coordinate domain r − 1 > 0, the quotient
metric is in the required form. We have
u′ + v = 2r + 4M log |r − 2M | ∈ (−∞, 4M log(2M )),
and by setting
u′′ = u′ − 4M log(2M )
we obtain
u′′ + v < 0.
Consequently, by applying the coordinate rescaling (3) we obtain a triangle
(Figure 7). Again there is a spacelike boundary, where r = 0, and two null
boundaries H + , where r = 2M . The Penrose diagram for the domain of the
coordinates (v, r) can be obtained gluing the Penrose diagrams in Figures 4
and 7 along H + , so that the null geodesics v = constant match (Figure 8).

r=0
i+ i+

H+ H+

Figure 7. Penrose diagram for a region r < 2M of the

Schwarzschild spacetime, with some level hypersurfaces of t
and r represented.

Both regions r < 2M can of course be glued to the region r > 2M

simultaneously. Since they are invariant under reflections with respect to
t = 0 (the vertical line through their common vertex), it is then clear that
a mirror-reversed copy of the region r > 2M can be glued to the surviving
null boundaries H − and H + (Figure 9). The resulting spacetime, known
as the maximal analytical extension of the Schwarzschild solution, is a
solution of the Einstein equations which cannot be extended any further,
since r → 0 or r → +∞ on the boundary of its Penrose diagram. Note
that by continuity the Einstein equations hold at the point where the four
Penrose diagrams intersect (known as the bifurcate sphere), which can
be shown to be part of the manifold; this requires setting up yet another
coordinate system, and will not be pursued here.
26 2. EXACT SOLUTIONS

i+ r=0 i+

H+ I+

i0

H− I−

i−

Figure 8. Penrose diagram for the domain of the coordi-

nates (v, r) in the Schwarzschild spacetime, with some level
hypersurfaces of t and r represented.

i+ r=0 i+

I+ I+

i0

I− I−

i− r=0 i−

Figure 9. Penrose diagram for the maximal analytic exten-

sion of the Schwarzschild spacetime, with some level hyper-
surfaces of t and r represented.

Let us now analyze in detail the Penrose diagram for the maximal ana-
lytic extension of the Schwarzschild spacetime. There are two asymptoti-
cally flat regions r > 2M , corresponding to two causally disconnected uni-
verses, joined by a wormhole. There are also two regions where r < 2M : a
black hole region, bounded by the future event horizons H + , from which
no causal curve can escape; and a white hole region, bounded by the past
event horizons H − , from which every causal curve must escape. Note that
the horizons themselves correspond to spheres which are propagating at the
speed of light, but whose radius remains constant, r = 2M .
 4. FRIEDMANN-LEMAÎTRE-ROBERTSON-WALKER MODELS 27

The black hole in the maximal analytic extension of the Schwarzschild

spacetime is an eternal black hole, that is, a black hole which has always
existed (as opposed to having formed by some physical process). We will see
shortly how to use the Schwarzschild solution to model physically realistic
black holes.

4. Friedmann-Lemaı̂tre-Robertson-Walker models
The simplest models of cosmology, the study of the Universe as a whole,
are obtained from the assumption that space is homogeneous and isotropic
(which is true on average at very large scales). It is well known that the
only isotropic 3-dimensional Riemannian metrics are, up to scale, given by
dr2
dl2 = + r2 dθ2 + sin2 θdϕ2 ,


1 − kr 2

where 
 1
 for the standard metric on S 3
k= 0 for the standard metric on R3 .
for the standard metric on H 3

−1


Allowing for a time-dependent scale factor a(t) (also known as the “radius
of the Universe”), we arrive at the Friedmann-Lemaı̂tre-Robertson-
Walker (FLRW) family of Lorentzian metrics:
dr2
 
2 2 2 2 2 2 2


(8) ds = −dt + a (t) + r dθ + sin θdϕ .
1 − kr2
To interpret these metrics, we consider a general Lorentzian metric of
the form
ds2 = −e2φ dt2 + hij dxi dxj = −e2φ dt2 + dl2 .
The Riemannian metric dl2 is readily interpreted as giving the distances
measured between nearby observers with fixed space coordinates xi in radar
experiments: indeed, such observers measure proper time τ given by
dτ 2 = e2φ dt2 .
The null geodesics representing a radar signal bounced by a given observer
from a nearby observer (Figure 10) satisfy
ds2 = 0 ⇔ e2φ dt2 = dl2 ⇔ dτ 2 = dl2 ⇔ dτ = ±dl.
Since the speed of light is c = 1, the distance traveled between the
observers will be half the time between the emission and the reception of
the signal:
(τ + dl) − (τ − dl)
= dl.
2
Moreover, the unit timelike covector field tangent to the histories of the
observers with fixed space coordinates xi is
U = −eφ dt ⇔ Uµ = −eφ ∇µ t.
28 2. EXACT SOLUTIONS

2dl

Figure 10. Distance between nearby observers.

Therefore
∇U U µ = −U ν ∇ν (eφ ∇µ t) = (U · φ)U µ − eφ U ν ∇ν ∇µ t
= (U · φ)U µ − eφ U ν ∇µ ∇ν t = (U · φ)U µ + eφ U ν ∇µ (e−φ Uν )
= (U · φ)U µ − U ν Uν ∇µ φ + U ν ∇µ Uν
1
= ∇µ φ + (U ν ∇ν φ)U µ + ∇µ (U ν Uν ) = ∇µ φ + (U ν ∇ν φ)U µ ,
2
since U ν Uν = −1. In other words,
∇U U = (grad φ)⊥ ,
where ⊥ represents the orthogonal projection on the spacelike hyperplane
orthogonal to U .
Therefore the observers with fixed space coordinates in the FLRW mod-
els have zero acceleration, that is, they are free-falling (by opposition to the
corresponding observers in the Schwarzschild spacetime, who must acceler-
ate to remain at fixed r > 2M ). Moreover, the distance between two such
observers varies as
d0 d0 ȧ
d(t) = a(t) ⇒ d˙ = ȧ = d.
a0 a0 a
This relation, known as the Hubble law, is often written as
v = Hd,
where v is the relative velocity and
ȧ
H=
a
is the so-called Hubble constant (for historical reasons, since it actually
varies in time).
 4. FRIEDMANN-LEMAÎTRE-ROBERTSON-WALKER MODELS 29

We will model the matter content of the universe as an uniform dust of

galaxies placed at fixed space coordinates (hence free-falling):
(9) T = ρ(t)dt ⊗ dt.
Plugging the metric (8) and the energy-momentum tensor (9) into the Ein-
stein equations, and integrating once, results in the so-called Friedmann
equations
1 2 α Λ 2 k

 2 ȧ − a − 6 a = − 2




 4π ρa3 = α



3
(where α is an integration constant). The first Friedmann equation is a first
order ODE for a(t); it can be seen as the equation of conservation of energy
for a particle moving in the 1-dimensional effective potential
α Λ 2
V (a) = − − a
a 6
with energy − k2 . Once this equation has been solved, the second Friedmann
equation yields ρ(t) from a(t). We now examine in detail the FLRW models
arising from the solutions of these equations.

4.1. Milne universe. If we set α = Λ = 0 then the first Friedmann

equation becomes
ȧ2 = −k.
Therefore either k = 0 and ȧ = 0, which corresponds to the Minkowski
spacetime, or k = −1 and ȧ2 = 1, that is
ds2 = −dt2 + t2 dlH
2
3,

where dlH2 represents the metric of the unit hyperbolic 3-space; this is the
3
so-called Milne universe. It turns out that the Milne universe is isometric
to an open region of the Minkowski spacetime, namely the region limited
by the future (or past) light cone of the origin. This region is foliated by
hyperboloids St of the form
T 2 − X 2 − Y 2 − Z 2 = t2 ,
whose induced metric is that of a hyperbolic space of radius t (Figure 11).
Note that the light cone corresponds to a(t) = t = 0, that is, the Big Bang
of the Milne universe.

4.2. de Sitter universe. If α = 0 and Λ > 0 we can choose units such

that Λ = 3. The first Friedmann equation then becomes
ȧ2 − a2 = −k.
30 2. EXACT SOLUTIONS

St

Figure 11. Milne universe.

In this case all three values k = 1, k = 0 and k = −1 are possible; the

corresponding metrics are, respectively,
ds2 = −dt2 + cosh2 t dlS2 3 ;
ds2 = −dt2 + e2t dlR
2
3;

ds2 = −dt2 + sinh2 t dlH

2
3,

where dlS2 3 , dlR

2 and dl2
3 H 3 represent the metric of the unit 3-sphere, the
Euclidean 3-space and the unit hyperbolic 3-space.
It turns out that the last two models correspond to open regions of the
first, which is then called the de Sitter universe. It represents a spherical
universe which contracts to a minimum radius (1 in our units) and then
re-expands. It is easily seen to be isometric to the unit hyperboloid
−T 2 + X 2 + Y 2 + Z 2 + W 2 = 1
in the Minkowski 5-dimensional spacetime (Figure 12).
To obtain the Penrose diagram for the de Sitter universe we write its
metric as
ds2 = −dt2 + cosh2 t dψ 2 + sin2 ψ dθ2 + sin2 θdϕ2



= cosh2 t −dτ 2 + dψ 2 + sin2 ψ dθ2 + sin2 θdϕ2




= cosh2 t −dτ 2 + dψ 2 + r2 dθ2 + sin2 θdϕ2 ,

 

where ψ ∈ [0, π],

t
dt
Z
τ=
−∞ cosh t
 4. FRIEDMANN-LEMAÎTRE-ROBERTSON-WALKER MODELS 31

Figure 12. de Sitter universe.

and
r = cosh t sin ψ.
Since
+∞
dt
Z
= π,
−∞ cosh t
we see that the quotient metric is conformal to the square (0, π) × [0, π] of
the Minkowski 2-dimensional spacetime, and so the Penrose diagram is as
depicted in Figure 13. Note that there are two lines where r = 0, corre-
sponding to two antipodal points of the 3-sphere. A light ray emitted from
one of these points at t = −∞ has just enough time to reach the other point
at t = +∞ (dashed line in the diagram). Note also that in this case I −
and I + (defined as the past and future boundary points approached by null
geodesics along which r → +∞) are spacelike boundaries.
4.3. Anti-de Sitter universe. If α = 0 and Λ < 0 we can choose
units such that Λ = −3. The first Friedmann equation then becomes
ȧ2 + a2 = −k.
In this case only k = −1 is possible; the corresponding metric is
ds2 = −dt2 + cos2 t dlH
2
3.

It turns out (see Exercise 5 in Chapter 5) that this model is an open region
of the spacetime with metric
ds2 = − cosh2 ψdt2 + dψ 2 + sinh2 ψ dθ2 + sin2 θdϕ2

(where ψ ∈ [0, +∞)), called the anti-de Sitter universe. It represents a

static hyperbolic universe (with radius 1 in our units).
32 2. EXACT SOLUTIONS

I+

r=0 r=0

I−

Figure 13. Penrose diagram for the de Sitter universe.

To obtain the Penrose diagram for the anti-de Sitter universe we write
its metric as
dψ 2
 
2 2 2
+ sinh2 ψ dθ2 + sin2 θdϕ2


ds = cosh ψ −dt + 2
cosh ψ
= cosh ψ −dt + dx2 + r2 dθ2 + sin2 θdϕ2
2 2
 

where
ψ
dψ
Z
x=
0 cosh ψ
and
r = sinh ψ.
Since
+∞
dψ π
Z
= ,
0 cosh ψ 2
we see that the quotient metric is conformal to the strip R × [0, π2 ) of the
Minkowski 2-dimensional spacetime, and so the Penrose diagram is as de-
picted in Figure 14. The FLRW model above corresponds to the triangular
region in the diagram. Note also that in this case I − ≡ I + ≡ I is a
timelike boundary.
4.4. Universes with matter and Λ = 0. If α > 0 and Λ = 0, the
first Friedmann equation becomes
2α
ȧ2 − = −k.
a
In this case all three values k = 1, k = 0 and k = −1 are possible. Although
it is possible to obtain explicit formulas for the solutions of these equations,
it is simpler to analyze the graph of the effective potential V (a) (Figure 15).
Possibly by reversing and translating t, we can assume that all solutions
are defined for t > 0, with limt→0 a(t) = 0, implying limt→0 ρ(t) = +∞.
Therefore all three models have a true singularity at t = 0, known as the
 4. FRIEDMANN-LEMAÎTRE-ROBERTSON-WALKER MODELS 33

r=0 I

Figure 14. Penrose diagram for the anti-de Sitter universe.

Big Bang, where the scalar curvature R = 8πρ also blows up; this is not
true for the Milne universe or the open region in the anti-de Sitter universe,
which can be extended across the Big Bang. The spherical universe (k = 1)
reaches a maximum radius 2α and re-collapses, forming a second singularity
(the Big Crunch); the radius of the flat (k = 0) and hyperbolic (k = −1)
universes increases monotonically.
To obtain the Penrose diagram for the spherical universe we write its
metric as
ds2 = −dt2 + a2 (t) dψ 2 + sin2 ψ dθ2 + sin2 θdϕ2



= a2 (t) −dτ 2 + dψ 2 + sin2 ψ dθ2 + sin2 θdϕ2




= a2 (t) −dτ 2 + dψ 2 + r2 dθ2 + sin2 θdϕ2 ,

 

where ψ ∈ [0, π],

t
dt
Z
τ=
0 a(t)
and
r = a(t) sin ψ.
Since
tmax amax 2α
dt da da
Z Z Z
=2 =2 q = 2π,
0 a(t) 0 aȧ 0 a 2α
−1
a
34 2. EXACT SOLUTIONS

V (a)

k = −1

k=0 a

k=1

Figure 15. Effective potential for FLRW models with Λ = 0.

we see that the quotient metric is conformal to the rectangle (0, 2π) × [0, π]
of the Minkowski 2-dimensional spacetime, and so the Penrose diagram is
as depicted in Figure 16. Note that there are two lines where r = 0, corre-
sponding to two antipodal points of the 3-sphere. A light ray emitted from
one of these points at t = 0 has just enough time to circle once around the
universe and return at t = tmax (dashed line in the diagram). Note also that
the Big Bang and the Big Crunch are spacelike boundaries.
To obtain the Penrose diagram for the flat universe we write its metric
as
ds2 = −dt2 + a2 (t) dρ2 + ρ2 dθ2 + sin2 θdϕ2



= a2 (t) −dτ 2 + dρ2 + ρ2 dθ2 + sin2 θdϕ2




= a2 (t) −dτ 2 + dρ2 + r2 dθ2 + sin2 θdϕ2 ,

 

where ρ ∈ [0, +∞) and r = a(t)ρ. Since

Z +∞ Z +∞ Z +∞
dt da da
= = q = +∞,
0 a(t) 0 aȧ 0 a 2αa

we see that the quotient metric is conformal to the region (0, +∞) × [0, +∞)
of the Minkowski 2-dimensional spacetime, and so the Penrose diagram is
as depicted in Figure 17. Note that the Big Bang is a spacelike boundary.
The Penrose diagram for the hyperbolic universe turns out to be the
same as for the flat universe. To see this we write its metric as
ds2 = −dt2 + a2 (t) dψ 2 + sinh2 ψ dθ2 + sin2 θdϕ2



= a2 (t) −dτ 2 + dψ 2 + sinh2 ψ dθ2 + sin2 θdϕ2




= a2 (t) −dτ 2 + dψ 2 + r2 dθ2 + sin2 θdϕ2 ,

 

 4. FRIEDMANN-LEMAÎTRE-ROBERTSON-WALKER MODELS 35

Big Crunch

r=0 r=0

Big Bang

Figure 16. Penrose diagram for the spherical universe.

i+

I+
r=0

i0

Big Bang

Figure 17. Penrose diagram for the flat and hyperbolic universes.

where ψ ∈ [0, +∞) and r = a(t) sinh ψ, and note that

Z +∞ Z +∞ Z +∞
dt da da
= = q = +∞.
0 a(t) 0 aȧ 0 a 2α + 1
a
36 2. EXACT SOLUTIONS

4.5. Universes with matter and Λ > 0. If α > 0 and Λ > 0 we can
choose units such that Λ = 3. The first Friedmann equation then becomes
2α
ȧ2 − − a2 = −k.
a
In this case all three values k = 1, k = 0 and k = −1 are possible. As before,
we analyze the graph of the effective potential V (a) (Figure 18). The hyper-
bolic and flat universes behave qualitatively like when Λ = 0, although ȧ(t)
is now unbounded as t → +∞, instead of approaching some constant. The
spherical universe has a richer spectrum of possible behaviors, depending on
α, represented in Figure 18 by drawing the line of constant energy√−k = −1
at three different heights. The higher line (corresponding to α > 93 ) yields
a behaviour similar to that of the hyperbolic
√ and flat universes. The inter-
3
mediate line (corresponding to α = 9 ) gives rise to an unstable equilibrium
√
point a = 33 , where the attraction force of the matter is balanced by the
repulsion force of the cosmological constant; it corresponds to the so-called
Einstein universe, the first cosmological model ever proposed. The inter-
mediate line also yields two solutions asymptotic to the Einstein universe,
one containing a Big Bang and the√ other endless expansion. Finally, the
3
lower line (corresponding to α < 9 ) yields two different types of behaviour
(depending on the initial conditions): either similar to the spherical model
with Λ = 0, or to the de Sitter universe.

V (a)

k = −1

k=0 a

k=1

Figure 18. Effective potential for FLRW models with Λ > 0.

It is currently believed that the best model for our physical Universe is
the flat universe with Λ > 0. If we write the first Friedmann equation as
ȧ2 8π Λ
H2 ≡ 2
= ρ+
a 3 3
 5. MATCHING 37

then the terms on the right-hand side are in the proportion 2 : 5 at the
present time.

4.6. Universes with matter and Λ < 0. If α > 0 and Λ < 0 we can
choose units such that Λ = −3. The first Friedmann equation then becomes
2α
ȧ2 − + a2 = −k.
a
In this case all three values k = 1, k = 0 and k = −1 are possible. As
before, we analyze the graph of the effective potential V (a) (Figure 19).
The qualitative behaviour of the hyperbolic, flat and spherical universes is
the same as the spherical universe with Λ = 0, namely starting at a Big
Bang and ending at a Big Crunch.

V (a)

k = −1

k=0 a

k=1

Figure 19. Effective potential for FLRW models with Λ > 0.

5. Matching
Let (M1 , g1 ) and (M2 , g2 ) be solutions of the Einstein field equations
containing open sets U1 and U2 whose boundaries S1 and S2 are timelike
hypersurfaces, that is, hypersurfaces whose induced metric is Lorentzian
(or, equivalently, whose normal vector is spacelike). If S1 is diffeomorphic
to S2 then we can identify them to obtain a new manifold M , gluing U1 to
U2 along S1 ∼= S2 (Figure 20).
Let n be the unit normal vector to S1 pointing out of U1 , which we
identify with the unit normal vector to S2 pointing into U2 . If (x1 , x2 , x3 )
are local coordinates on S ≡ S1 ∼ = S2 , we can construct a system of local
1 2 3
coordinates (t, x , x , x ) in a neighbourhood of S by moving a distance
t along the geodesics with initial condition n. Note that U1 , S and U2
38 2. EXACT SOLUTIONS

g1

n
g2

S1 ∼
= S2

Figure 20. Matching two spacetimes.

∂
correspond to t < 0, t = 0 and t > 0 in these coordinates. Since ∂t is the
unit tangent vector to the geodesics, we have
     
∂ ∂ ∂ ∂ ∂ ∂ ∂
, = ∇∂ , i + ,∇ ∂
∂t ∂t ∂xi ∂t ∂t ∂x ∂t ∂t ∂xi
    
∂ ∂ ∂ 1 ∂ ∂
= ,∇ ∂ = i
, =0
∂t ∂xi ∂t ∂x 2 ∂t ∂t
(i = 1, 2, 3), where we used
 
∂ ∂ ∂ ∂
∇∂ − ∇ ∂ = , = 0.
∂t ∂xi ∂xi ∂t ∂t ∂xi
Since for t = 0 we have
   
∂ ∂ ∂
, = n, i = 0,
∂t ∂xi ∂x
∂
we see that ∂t remains orthogonal to the surfaces of constant t. This result
will be used repeatedly.
Lemma 5.1. (Gauss Lemma I) Let (M, g) be a Riemannian or a
Lorentzian manifold, and S ⊂ M a hypersurface whose normal vector field
n satisfies g(n, n) 6= 0. The hypersurfaces St obtained from S by moving
a distance t along the geodesics orthogonal to S remain orthogonal to the
geodesics.
The same ideas can be used to prove a closely related result.
 6. OPPENHEIMER-SNYDER COLLAPSE 39

Lemma 5.2. (Gauss Lemma II) Let (M, g) be a Riemannian or a

Lorentzian manifold and p ∈ M . The hypersurfaces St obtained from p by
moving a distance t along the geodesics through p remain orthogonal to the
geodesics.
In this coordinate system the metrics g1 and g2 are given on t ≤ 0 and
t ≥ 0, respectively, by
gA = dt2 + hA i
ij (t, x)dx dx
j

(A = 1, 2). Therefore we can define a continuous metric g on M if

h1ij (0, x)dxi dxj = h2ij (0, x)dxi dxj ,
that is, if
g1 |T S = g2 |T S .
This also guarantees continuity of all tangential derivatives of the metric,
but not of the normal derivatives. In order to have a C 1 metric we must
have
∂h1ij ∂h2ij
(0, x)dxi dxj = (0, x)dxi dxj ,
∂t ∂t
that is,
Ln g1 |T S = Ln g2 |T S .
Note that in this case the curvature tensor (hence the energy-momentum
tensor) is at most discontinuous across S. More importantly, as shown in
Exercise 8, the components Ttt and Tti of the energy-momentum tensor are
continuous across S (that is, the flow Tµν nµ of energy and momentum across
S is equal on both sides), implying that the energy-momentum tensor satis-
fies the integral version of the conservation equation ∇µ Tµν = 0. Therefore
we can consider (M, g) a solution of the Einstein equations.
Recall that
1
KA = Ln gA |T S
2 A

is known as the extrinsic curvature, or second fundamental form, of

SA . We can summarize the discussion above in the following statement.
Proposition 5.3. Two solutions (M1 , g1 ) and (M2 , g2 ) of the Einstein
field equations can be matched along diffeomorphic timelike boundaries S1
and S2 if and only if the induced metrics and second fundamental forms
coincide:
g1 = g2 and K1 = K2 .

6. Oppenheimer-Snyder collapse
We can use the matching technique to construct a solution of the Einstein
field equations which describes a spherical cloud of dust collapsing to a black
hole. This is a physically plausible model for a black hole, as opposed to the
eternal black hole.
40 2. EXACT SOLUTIONS

Let us take (M1 , g1 ) to be a flat collapsing FLRW universe:

g1 = −dτ 2 + a2 (τ ) dσ 2 + σ 2 dθ2 + sin2 θdϕ2 .



We choose S1 to be the hypersurface σ = σ0 , with normal vector

1 ∂
n= .
a ∂σ
The induced metric then is
g1 |T S = −dτ 2 + a2 (τ )σ0 2 dθ2 + sin2 θdϕ2 ,


1

and the second fundamental form

K1 = a(τ )σ0 dθ2 + sin2 θdϕ2 .

Here we used Cartan’s magic formula: if ω is a differential form and X

is a vector field then
LX ω = Xy dω + d(Xy ω).
Thus, for example,
Ln dτ = ny d2 τ + d(ny dτ ) = 0,
and analogously Ln dθ = Ln dϕ = 0; only Ln dσ 6= 0, but this is irrelevant
since dσ = 0 when restricted to σ = σ0 . Note that the function a(τ ) is
constrained by the first Friedmann equation:
2α
(10) ȧ2 =
a
(we assume Λ = 0).
We now take (M2 , g2 ) to be the Schwarzschild solution,
g2 = −V dt2 + V −1 dr2 + r2 dθ2 + sin2 θdϕ2 ,

where
2M
V =1− ,
r
and choose S2 to be a spherically symmetric timelike hypersurface given by
the parameterization (
t = t(τ )
.
r = r(τ )
The exact form of the functions t(τ ) and r(τ ) will be fixed by the matching
conditions; for the time being they are constrained only by the condition
that τ is the proper time along S2 :
−V ṫ2 + V −1 ṙ2 = −1.
The induced metric is then
g2 |T S = −dτ 2 + r2 (τ ) dθ2 + sin2 θdϕ2 ,


2

and so the two induced metrics coincide if and only if

(11) r(τ ) = a(τ )σ0 .
 6. OPPENHEIMER-SNYDER COLLAPSE 41

To simplify the calculation of the second fundamental form, we note that

since S1 is ruled by timelike geodesics, so is S2 , because the induced met-
rics and extrinsic curvatures are the same (therefore so are the Christoffel
symbols). Therefore t(τ ) and r(τ ) must be a solution of the radial geodesic
equations, which are equivalent to
( 
V ṫ = E V ṫ = E
(12) ⇔ 2M
−V ṫ2 + V −1 ṙ2 = −1 ṙ2 = E 2 − 1 +
r
(where E > 0 is a constant). Equations (10), (11) and (12) are compatible
if and only if (
E=1
,
M = ασ0 3
that is, if and only if S2 represents a spherical shell dropped from infinity
with zero velocity and the mass parameter of the Schwarzschild spacetime
is related to the density ρ(τ ) of the collapsing dust by
4π 3
M= r (τ )ρ(τ ).
3
To compute the second fundamental form of S2 we can then consider a
family of free-falling spherical shells which includes S2 . If s is the parameter
indexing the shells (with S2 corresponding to, say, s = s0 ) then by the Gauss
Lemma we can write the Schwarzschild metric in the form
g2 = −dτ 2 + A2 (τ, s)ds2 + r2 (τ, s) dθ2 + sin2 θdϕ2 .

The unit normal vector field n to the hypersurfaces of constant s is then

1 ∂
n= ,
A ∂s
and so we have
1
K2 = Ln g2 |T S = r(n · r) dθ2 + sin2 θdϕ2 .


2 2

On the other hand, in Schwarzschild coordinates

∂ ∂
n = V −1 ṙ + V ṫ ,
∂t ∂r
since n must be unit and orthogonal to
∂ ∂
ṫ + ṙ .
∂t ∂r
Therefore we have
K2 = rV ṫ dθ2 + sin2 θdϕ2 = rE dθ2 + sin2 θdϕ2 ,

or, using E = 1,
K2 = r(τ ) dθ2 + sin2 θdϕ2 .

In other words, K1 = K2 follows from the previous conditions, and so we

indeed have a solution of the Einstein equations.
42 2. EXACT SOLUTIONS

Big Crunch r=0 i+

i0
I+
S1

r=0 S2
i0
I−

I−

i− i−

Figure 21. Matching hypersurfaces in the collapsing flat

FLRW universe and the Schwarzschild solution.

To construct the Penrose diagram for this solution we represent S1 and

S2 in the Penrose diagrams of the collapsing flat FLRW universe (obtained
by reversing the time direction in the expanding flat FLRW universe) and the
Schwarzschild solution (Figure 21). Identifying these hypersurfaces results
in the Penrose diagram depicted in Figure 22.

r=0 i+

I+

i0

S
r=0

I−

i−

Figure 22. Penrose diagram for the Oppenheimer-Snyder collapse.

 7. EXERCISES 43

7. Exercises
(1) In this exercise we will solve the vacuum Einstein equations (with-
out cosmological constant) for the spherically symmetric Lorentzian
metric given by
ds2 = −(A(t, r))2 dt2 + (B(t, r))2 dr2 + r2 dθ2 + sin2 θdϕ2 ,

where A and B are positive smooth functions.

(a) Use Cartan’s first structure equations,
(
ωµν = −ωνµ
,
dω µ + ω µν ∧ ω ν = 0

to show that the nonvanishing connection forms for the or-

thonormal frame dual to
ω 0 = Adt, ω r = Bdr, ω θ = rdθ, ω ϕ = r sin θdϕ
∂ ∂
are (using the notation ˙ = ∂t and ′ = ∂r )

A′ Ḃ
ω0 r = ωr 0 = dt + dr ;
B A
1
ωθ r = −ω r θ = dθ ;
B
sin θ
ω ϕr r
= −ω ϕ = dϕ ;
B
ω ϕθ = −ω θ ϕ = cos θdϕ .

(b) Use Cartan’s second structure equations,

Ωµ ν = dω µν + ω µα ∧ ω αν ,
to show that the curvature forms on this frame are
!
0 r A′′ B − A′ B ′ ȦḂ − AB̈
Ω r=Ω 0= + ωr ∧ ω0 ;
AB 3 A3 B
A′ Ḃ
Ω0 θ = Ωθ 0 = 2
ωθ ∧ ω0 + ωθ ∧ ωr ;
rAB rAB 2
A′ Ḃ
Ω0 ϕ = Ωϕ0 = ωϕ ∧ ω0 + ωϕ ∧ ωr ;
rAB 2 rAB 2
B′ θ Ḃ
Ωθ r = −Ωr θ = 3
ω ∧ ωr + ωθ ∧ ω0 ;
rB rAB 2
B′ ϕ Ḃ
Ωϕr = −Ωr ϕ = 3
ω ∧ ωr + ωϕ ∧ ω0 ;
rB rAB 2
B2 − 1
Ωϕθ = −Ωθ ϕ = 2 2 ω ϕ ∧ ω θ .
r B
44 2. EXACT SOLUTIONS

(c) Using
Rαβ µν ω α ∧ ω β ,
X
Ωµ ν =
α<β

determine the components Rαβ µν of the curvature tensor in

this orthonormal frame, and show that the nonvanishing com-
ponents of the Ricci tensor in this frame are
A′′ B − A′ B ′ ȦḂ − AB̈ 2A′
R00 = + + ;
AB 3 A3 B rAB 2
2Ḃ
R0r = Rr0 = ;
rAB 2
A′ B ′ − A′′ B AB̈ − ȦḂ 2B ′
Rrr = + + ;
AB 3 A3 B rB 3
A′ B′ B2 − 1
Rθθ = Rϕϕ = − + + .
rAB 2 rB 3 r2 B 2
Conclude that the nonvanishing components of the Einstein
tensor in this frame are
2B ′ B2 − 1
G00 = + ;
rB 3 r2 B 2
2Ḃ
G0r = Gr0 = ;
rAB 2
2A′ B2 − 1
Grr = − ;
rAB 2 r2 B 2
A′′ B − A′ B ′ ȦḂ − AB̈ A′ B′
Gθθ = Gϕϕ = + + − .
AB 3 A3 B rAB 2 rB 3
(d) Show that if we write
 1
2m(t, r) − 2

B(t, r) = 1 −
r
for some smooth function m then
2m′
G00 = 2 .
r
Conclude that the Einstein equations G00 = G0r = 0 are equiv-
alent to
 1
2M − 2

B = 1− ,
r
where M ∈ R is an integration constant.
(e) Show that the Einstein equation G00 + Grr = 0 is equivalent
to A = α(t)
B for some positive smooth function α.
(f) Check that if A and B are as above then the remaining Einstein
equations Gθθ = Gϕϕ = 0 are automatically satisfied.
 7. EXERCISES 45

(g) Argue that it is always possible to rescale the time coordinate

t so that the metric is written

2M −1 2 2
   
2 2M 2
dr +r dθ2 + sin2 θdϕ2 .


ds = − 1 − dt + 1 −
r r

The statement that any spherically symmetric solution of the

vacuum Einstein equations without cosmological constant is of
this form is known as Birkhoff ’s theorem.
(2) Show that the Riemannian manifold obtained by gluing the hyper-
surfaces t = 0 of the two exterior regions in the maximally extended
Schwarzschild solution along the horizon r = 2M is isometric to the
Flamm paraboloid, that is, the hypersurface in R4 with equation

p w2
x2 + y 2 + z 2 = 2M +
8M

(Figure 23).

Figure 23. Two-dimensional analogue of the Flamm paraboloid.

(3) Recall that the nonvanishing components of the Einstein tensor of

the spherically symmetric Lorentzian metric

ds2 = −(A(t, r))2 dt2 + (B(t, r))2 dr2 + r2 dθ2 + sin2 θdϕ2

in the orthonormal frame dual to

ω 0 = Adt, ω r = Bdr, ω θ = rdθ, ω ϕ = r sin θdϕ,

46 2. EXACT SOLUTIONS

∂ ∂
are given by (using the notation ˙ = ∂t and ′ = ∂r )

2B ′ B2 − 1 2m′
G00 = + = ;
rB 3 r2 B 2 r2
2Ḃ
G0r = Gr0 = ;
rAB 2
2A′ B2 − 1
Grr = − ;
rAB 2 r2 B 2
A′′ B − A′ B ′ ȦḂ − AB̈ A′ B′
Gθθ = Gϕϕ = + + − ,
AB 3 A3 B rAB 2 rB 3
where
 − 1
2m(t, r) 2
B(t, r) = 1− .
r
(a) Assuming
• G0r = 0 (so that B, and hence m, do not depend on t);
• G00 +Grr = 0 (so that A = α(t)
B for some positive smooth
function α(t));
• α(t) = 1 (which can always be achieved by rescaling t),
show that
1 2
′′ 1 2
′
Gθθ = Gϕϕ = A + A .
2 r
(b) Prove that the general spherically symmetric solution of the
vacuum Einstein field equations with a cosmological constant
Λ is the Kottler metric
Λ 2 −1 2
   
2M Λ 2 2 2M
g =− 1− − r dt + 1 − − r dr
r 3 r 3
+ r2 dθ2 + sin2 θdϕ2 .

(c) Obtain the Penrose diagram for the maximal extension of the
Kottler solution with Λ > 0 and 0 < M < 3√1 Λ .
(d) Consider now the spherically symmetric electromagnetic field
F = E(t, r) ω r ∧ ω 0 .
Show that this field satisfies the vacuum Maxwell equations
dF = d ⋆ F = 0
(where ⋆ is the Hodge star) if and only if
e
E(t, r) = 2
r
for some constant e ∈ R (the electric charge in units for which
4πε0 = 1).
 7. EXERCISES 47

(e) As we shall see in Chapter 6, this electromagnetic field corre-

sponds to the energy-momentum tensor
E2  0 
T = ω ⊗ ω0 − ωr ⊗ ωr + ωθ ⊗ ωθ + ωϕ ⊗ ωϕ .
8π
Prove that the general spherically symmetric solution of the
Einstein field equations with an electromagnetic field of this
kind is the Reissner-Nordström metric
−1
e2 e2
  
2M 2 2M
g =− 1− + 2 dt + 1 − + 2 dr2
r r r r
+ r2 dθ2 + sin2 θdϕ2 .

(f) Obtain the Penrose diagram for the maximal extension of the
Reissner-Nordström solution with M > 0 and 0 < e2 < M 2 .
(4) Consider the spherically symmetric Lorentzian metric
 
2 2 2 1 2 2 2 2 2


ds = −dt + a (t) dr + r dθ + sin θdϕ ,
1 − kr2
where a is a positive smooth function.
(a) Use Cartan’s first structure equations,
(
ωµν = −ωνµ
,
dω µ + ω µν ∧ ω ν = 0

to show that the nonvanishing connection forms for the or-

thonormal frame dual to

− 1
ω 0 = dt, ω r = a(t) 1 − kr2 2 dr,
ω θ = a(t)rdθ, ω ϕ = a(t)r sin θdϕ

are

− 1
ω 0 r = ω r 0 = ȧ 1 − kr2 2
dr ;
0 θ
ω θ =ω 0 = ȧrdθ ;
ω0 ϕ = ω ϕ0 = ȧr sin θdϕ ;

1
ω θ r = −ω r θ = 1 − kr2 2
dθ ;

1
ω ϕr = −ω r ϕ = 1 − kr 2 2
sin θdϕ ;
ω ϕθ = −ω θ ϕ = cos θdϕ .

(b) Use Cartan’s second structure equations,

Ωµ ν = dω µν + ω µα ∧ ω αν ,
48 2. EXACT SOLUTIONS

to show that the curvature forms on this frame are

ä 0
Ω0 r = Ωr 0 = ω ∧ ωr ;
a
ä
Ω0 θ = Ωθ 0 = ω 0 ∧ ω θ ;
a
ä
Ω0 ϕ = Ωϕ0 = ω 0 ∧ ω ϕ ;
a
ȧ2

θ r k
Ω r = −Ω θ = + ωθ ∧ ωr ;
a2 a2
ȧ2
 
ϕ r k
Ω r = −Ω ϕ = + ωϕ ∧ ωr ;
a2 a2
ȧ2
 
ϕ θ k
Ω θ = −Ω ϕ = + ωϕ ∧ ωθ .
a2 a2

(c) Using

Rαβ µν ω α ∧ ω β ,
X
Ωµ ν =
α<β

determine the components Rαβ µν of the curvature tensor on

this orthonormal frame, and show that the nonvanishing com-
ponents of the Ricci tensor on this frame are

3ä
R00 = − ;
a
ä 2ȧ2 2k
Rrr = Rθθ = Rϕϕ = + 2 + 2.
a a a
Conclude that the nonvanishing components of the Einstein
tensor on this frame are

3ȧ2 3k
G00 = + 2;
a2 a
2ä ȧ2 k
Grr = Gθθ = Gϕϕ = − − 2− 2.
a a a
(d) Show that the Einstein equations with a cosmological constant
Λ for a comoving pressureless perfect fluid of nonnegative den-
sity ρ, G + Λg = 8πρ dt2 , are equivalent to the system
 2
ȧ k 8πρ Λ


a 2
+ 2
= +
 a 3 3
.
2
 2ä + ȧ + k = Λ



a a2 a2
 7. EXERCISES 49

Show that this system can be integrated to

4πρ 3


 a =α
 3

,
 1 α Λ
 ȧ2 − − a2 = −

 k
2 a 6 2
where α is a nonnegative integration constant.
(e) Draw the Penrose diagram of the solutions with α > 0, Λ > 0
and k = 0 (currently believed to model our physical Universe).
(5) Compute the metrics of the following manifolds in the local coor-
dinates indicated, and
√ sketch the corresponding Penrose diagrams:
(a) The region T > X 2 + Y 2 + Z 2 of the 4-dimensional Minkowski
spacetime using the parameterization


 T = t cosh ψ

X = t sinh ψ sin θ cos ϕ
.

 Y = t sinh ψ sin θ sin ϕ

Z = t sinh ψ cos θ


(b) The hyperboloid X 2 + Y 2 + Z 2 + W 2 = 1 + T 2 in the 5-

dimensional Minkowski spacetime using the parameterization

T = sinh t


X = cosh t sin ψ sin θ cos ϕ



Y = cosh t sin ψ sin θ sin ϕ .

Z = cosh t sin ψ cos θ





W = cosh t cos ψ

(c) The region W > 1 of the same hyperboloid using the parame-
terization



 T = sinh t cosh ψ
 = sinh t sinh ψ sin θ cos ϕ
X



Y = sinh t sinh ψ sin θ sin ϕ .

Z = sinh t sinh ψ cos θ




W = cosh t

(d) The region T > W of the same hyperboloid using the param-
eterization defined implicitly by
T = W + et



 X = et x

.
 Y = et y


Z = et z

50 2. EXACT SOLUTIONS

(e) The hyperboloid T 2 + U 2 = 1 + X 2 + Y 2 + Z 2 in R5 with the

pseudo-Riemannian metric
ds2 = −dT 2 − dU 2 + dX 2 + dY 2 + dZ 2
using the parameterization

T = cosh ψ cos t


U = cosh ψ sin t



X = sinh ψ sin θ cos ϕ .

Y = sinh ψ sin θ sin ϕ





Z = sinh ψ cos θ

(6) Show that the anti-de Sitter metric

ds2 = − cosh2 ψdt2 + dψ 2 + sinh2 ψ dθ2 + sin2 θdϕ2

is a solution of the vacuum Einstein field equations with cosmolog-

ical constant Λ = −3.
(7) The 3-dimensional anti-de Sitter space can be obtained by “un-
wrapping” the hyperboloid T 2 + U 2 = 1 + X 2 + Y 2 in R4 with the
pseudo-Riemannian metric
ds2 = −dT 2 − dU 2 + dX 2 + dY 2 .
In this exercise we identify R4 with the space of 2 × 2 matrices by
the map
 
T +X U +Y
(T, U, X, Y ) 7→ .
−U + Y T − X
(a) Show that the hyperboloid corresponds to the Lie group SL(2, R)
of 2 × 2 matrices with unit determinant.
(b) Check that the squared norm of a vector v ∈ R4 in the met-
ric above is hv, vi = − det V , where V is the 2 × 2 matrix
associated to v. Conclude that the metric induced on the hy-
perboloid is bi-invariant (that is, invariant under left and right
multiplication).
(c) Use the Penrose diagram in Figure 24 and the fact that the
one-parameter subgroups of a Lie group with a bi-invariant
metric are geodesics of that metric to conclude that the expo-
nential map exp : sl(2, R) → SL(2, R) is not surjective.
(d) Write explicitly the matrices in SL(2, R) which are not in the
image of exp : sl(2, R) → SL(2, R) by using the parameteriza-
tion 

 T = cosh ψ cos t

U = cosh ψ sin t
.

 X = sinh ψ cos ϕ

Y = sinh ψ sin ϕ

 7. EXERCISES 51

t=π

ψ=0 ψ = +∞

t=0

t = −π

Figure 24. Exponential map on SL(2, R).

(8) Consider a Riemannian or Lorentzian metric given in the Gauss

Lemma form
g = dt2 + hij (t, x)dxi dxj ,
so that the level sets of t are Riemannian or Lorentzian manifolds
with induced metric h(t) = hij dxi dxj and second fundamental form
1 ∂hij i j
K(t) = dx dx .
2 ∂t
Show that in these coordinates:
(a) The Christoffel symbols are
Γ0ij = −Kij ; Γijk = Γ̄ijk ; Γi0j = K ij ,
where Γ̄ijk are the Christoffel symbols of h.
(b) The components of the Riemann tensor are
∂ j
R0i0j = − K − Kil K lj ;
∂t i
Rij0 l ¯ i K lj + ∇
= −∇ ¯ j K li ;
Rijl m = R̄ijl m − Kil K mj + Kjl K mi ,
52 2. EXACT SOLUTIONS

¯ is the Levi-Civita connection of h and R̄ m are the

where ∇ ijl
components of the Riemann tensor of h.
(c) The time derivative of the inverse metric is given by the for-
mula
∂hij
= −2K ij .
∂t
(d) The components of the Ricci tensor are
∂ i
R00 = − K − Kij K ij ;
∂t i
¯ iK j + ∇
R0i = −∇ ¯ jKj ;
j i
∂
Rij = R̄ij − Kij + 2Kil K lj − K ll Kij ,
∂t
where R̄ij are the components of the Ricci tensor of h.
(e) The scalar curvature is
∂
2
R = R̄ − 2 K ii − K ii − Kij K ij ,
∂t
where R̄ is the scalar curvature of h.
(f) The component G00 of the Einstein tensor is
1
2 
G00 = −R̄ + K ii − Kij K ij .
2
This shows that the matching conditions guarantee the conti-
nuity of G00 and G0i = R0i .
(9) Recall that the nonvanishing components of the Einstein tensor of
the static, spherically symmetric Lorentzian metric
ds2 = −(A(r))2 dt2 + (B(r))2 dr2 + r2 dθ2 + sin2 θdϕ2

in the orthonormal frame dual to

ω 0 = Adt, ω r = Bdr, ω θ = rdθ, ω ϕ = r sin θdϕ,
are given by
2B ′ B2 − 1 2m′
G00 = + = ;
rB 3 r2 B 2 r2
2A′ B2 − 1
Grr = − ;
rAB 2 r2 B 2
A′′ B − A′ B ′ A′ B′
Gθθ = Gϕϕ = + − ,
AB 3 rAB 2 rB 3
where
 1
2m(r) − 2

B(r) = 1 − .
r
In this exercise we will solve the Einstein equations (without cos-
mological constant) for a static perfect fluid of constant rest density
ρ and rest pressure p, and match it to a Schwarzschild exterior.
 7. EXERCISES 53

(a) Show that

− 1
B(r) = 1 − kr2 2
,
where k = 8π3 ρ. Conclude that the spatial metric is that of a
sphere S with radius √1k .
3

(b) Solve the ordinary differential equation Grr = Gθθ to obtain

1
A(r) = C 1 − kr2 2 + D,
where C, D ∈ R are integration constants.
(c) Show that the matching conditions to a Schwarzschild exterior
of mass M > 0 across a surface r = R are
  1
 2M 2
A(R) = 1 −





 R
 1
2M − 2 .
 
M

′
A (R) = 2 1 −

 R R

  − 1

 2M 2
B(R) = 1 −


R
(d) Conclude that
1 1
2M r2 2
 
3 2M 2 1
A(r) = 1− − 1− .
2 R 2 R3
(e) Show that

1
k 1 − kr2 2
p(r) = 1 1 − k.
3
2 (1 − kR2 ) − 12 (1 − kr2 ) 2
2

What is the value of p(R)?

(f) Show that M and R must satisfy Buchdahl’s limit:
2M 8
< .
R 9
2M
What happens to p(0) as R → 98 ?
 CHAPTER 3

Causality

In this chapter we briefly discuss the causality theory of Lorentzian man-

ifolds, following [GN14]. We take a minimal approach; in particular, our
definition of global hyperbolicity is apparently stronger than, but equivalent
to, the usual one (as shown in [BS03, BS05, BS06]). More details can be
found in [O’N83, Wal84, Pen87, Nab88, HE95, Rin09].

1. Past and future

A spacetime (M, g) is said to be time-orientable if there exists a
timelike vector field, that is, a vector field X satisfying g(X, X) < 0. In
this case, we can define a time orientation on each tangent space Tp M by
declaring causal vectors v ∈ Tp M to be future-pointing if g(v, Xp ) ≤ 0.
It can be shown that any non-time-orientable spacetime admits a time-
orientable double covering (just like any non-orientable manifold admits
an orientable double covering).
Assume that (M, g) is time-oriented (i.e. time-orientable with a defi-
nite choice of time orientation). A timelike or causal curve c : I ⊂ R → M is
said to be future-directed if ċ is future-pointing (the definition of a past-
directed curve being the obvious analogue). The chronological future
of p ∈ M is the set I + (p) of all points to which p can be connected by a
future-directed timelike curve. The causal future of p ∈ M is the set J + (p)
of all points to which p can be connected by a future-directed causal curve.
Notice that I + (p) is simply the set of all events which are accessible to a
particle with nonzero mass at p, whereas J + (p) is the set of events which can
be causally influenced by p (as this causal influence cannot propagate faster
than the speed of light). Analogously, the chronological past of p ∈ M is
the set I − (p) of all points which can be connected to p by a future-directed
timelike curve, and the causal past of p ∈ M is the set J − (p) of all points
which can be connected to p by a future-directed causal curve.
In general, the chronological and causal pasts and futures can be quite
complicated sets, because of global features of the spacetime. Locally, how-
ever, causal properties are similar to those of Minkowski spacetime. More
precisely, we have the following statement:
Proposition 1.1. Let (M, g) be a time-oriented spacetime. Then each
point p0 ∈ M has an open neighborhood V ⊂ M such that the spacetime
(V, g) obtained by restricting g to V satisfies:
55
56 3. CAUSALITY

(1) V is geodesically convex, that is, V is a normal neighborhood

of each of its points such that given p, q ∈ V there exists a unique
geodesic (up to reparameterization) connecting p to q;
(2) q ∈ I + (p) if and only if there exists a future-directed timelike geo-
desic connecting p to q;
(3) J + (p) = I + (p);
(4) q ∈ J + (p) \ I + (p) if and only if there exists a future-directed null
geodesic connecting p to q.
Proof. Recall that the exponential map expp : U ⊂ Tp M → M is
the map given by
expp (v) = cv (1),
where cv is the geodesic with initial conditions cv (0) = p, ċv (0) = v; equiva-
lently, expp (tv) = cv (t) (since ctv (1) = cv (t)). Recall also that V is a normal
neighborhood of p if expp : U → V is a diffeomorphism. The existence of
geodesically convex neighborhoods is true for any affine connection and is
proved for instance in [KN96].
To prove assertion (2), we start by noticing that if there exists a future-
directed timelike geodesic connecting p to q then it is obvious that q ∈ I + (p).
Suppose now that q ∈ I + (p); then there exists a future-directed timelike
curve c : [0, 1] → V such that c(0) = p and c(1) = q. Choose normal
coordinates (x0 , x1 , x2 , x3 ), given by the parameterization
ϕ(x0 , x1 , x2 , x3 ) = expp (x0 E0 + x1 E1 + x2 E2 + x3 E3 ),
where {E0 , E1 , E2 , E3 } is an orthonormal basis of Tp M with E0 timelike and
future-pointing. These are global coordinates in V , since expp : U → V is a
diffeomorphism. Defining

2
2
2
2
Wp (q) = − x0 (q) + x1 (q) + x2 (q) + x3 (q)
3
X
= ηµν xµ (q)xν (q),
µ,ν=0

with (ηµν ) = diag(−1, 1, 1, 1), we have to show that Wp (q) < 0. Let Wp (t) =
Wp (c(t)). Since xµ (p) = 0, we have Wp (0) = 0. Setting xµ (t) = xµ (c(t)), we
obtain
X3
Ẇp (t) = 2 ηµν xµ (t)ẋν (t);
µ,ν=0
3
X 3
X
Ẅp (t) = 2 ηµν xµ (t)ẍν (t) + 2 ηµν ẋµ (t)ẋν (t),
µ,ν=0 µ,ν=0

and consequently
Ẇp (0) = 0;
Ẅp (0) = 2hċ(0), ċ(0)i < 0.
 1. PAST AND FUTURE 57

Therefore there exists ε > 0 such that Wp (t) < 0 for t ∈ (0, ε).
By the Gauss Lemma, the level surfaces of Wp are orthogonal to the
geodesics through p. Therefore, if cv (t) = expp (tv) is the geodesic with
initial condition v ∈ Tp M , we have
(grad Wp )cv (1) = a(v)ċv (1).
Now
d d
(grad Wp )cv (t) , ċv (t) = Wp (cv (t)) = htv, tvi
dt dt
d 2

= t hv, vi = 2thv, vi,
dt
and hence
(grad Wp )cv (1) , ċv (1) = 2hv, vi.
On the other hand,
(grad Wp )cv (1) , ċv (1) = ha(v)ċv (1), ċv (1)i = a(v)hv, vi.
We conclude that a(v) = 2, and therefore
(grad Wp )cv (1) = 2ċv (1).
Consequently, grad Wp is tangent to geodesics through p, being future-
pointing on future-directed geodesics.
Suppose that Wp (t) < 0. Then (grad Wp )c(t) is timelike future-pointing,
and so D E
Ẇ (t) = (grad Wp )c(t) , ċ(t) < 0,
as ċ(t) is also timelike future-pointing. We conclude that we must have
Wp (t) < 0 for all t ∈ [0, 1]. In particular, Wp (q) = Wp (1) < 0, and hence
there exists a future-directed timelike geodesic connecting p to q.
To prove assertion (3), let us see first that I + (p) ⊂ J + (p). If q ∈ I + (p),
then q is the limit of a sequence of points qn ∈ I + (p). By (2), qn = expp (vn )
with vn ∈ Tp M timelike future-pointing. Since expp is a diffeomorphism, vn
converges to a causal future-pointing vector v ∈ Tp M , and so q = expp (v)
can be reached from p by a future-directed causal geodesic. The converse
inclusion J + (p) ⊂ I + (p) holds in general (cf. Proposition 1.2).
Finally, (4) is obvious from (3) and the fact that expp is a diffeomorphism
onto V . 
This local behavior can be used to prove the following global result.
Proposition 1.2. Let (M, g) be a time-oriented spacetime and p ∈ M .
Then:
(1) I + (p) is open;
(2) J + (p) ⊂ I + (p);
(3) I + (p) = int J + (p)
(4) if r ∈ J + (p) and q ∈ I + (r) then q ∈ I + (p);
(5) if r ∈ I + (p) and q ∈ J + (r) then q ∈ I + (p).
58 3. CAUSALITY

Proof. Exercise. 
The twin paradox also holds locally for general spacetimes. More pre-
cisely, we have the following statement:
Proposition 1.3. (twin paradox) Let (M, g) be a time-oriented space-
time, p0 ∈ M and V ⊂ M a geodesically convex open neighborhood of p0 .
The spacetime (V, g) obtained by restricting g to V satisfies the following
property: if p, q ∈ V with q ∈ I + (p), c is the timelike geodesic connecting p
to q and γ is any timelike curve connecting p to q, then τ (γ) ≤ τ (c), with
equality if and only if γ is a reparameterization of c.
Proof. Any timelike curve γ : [0, 1] → V satisfying γ(0) = p, γ(1) = q
can be written as
γ(t) = expp (r(t)n(t)),
for t ∈ [0, 1], where r(t) ≥ 0 and hn(t), n(t)i = −1. We have
γ̇(t) = (expp )∗ (ṙ(t)n(t) + r(t)ṅ(t)) .
Since hn(t), n(t)i = −1, we have hṅ(t), n(t)i = 0, and consequently ṅ(t) is
tangent to the level surfaces of the function v 7→ hv, vi. We conclude that
γ̇(t) = ṙ(t)Xγ(t) + Y (t),
where X is the unit tangent vector field to timelike geodesics through p
and Y (t) = r(t)(expp )∗ ṅ(t) is tangent to the level surfaces of Wp (hence
orthogonal to Xγ(t) ). Consequently,
Z 1
1
τ (γ) = ṙ(t)Xγ(t) + Y (t), ṙ(t)Xγ(t) + Y (t) 2 dt
0
Z 1

1
= ṙ(t)2 − |Y (t)|2 2 dt
0
Z 1
≤ ṙ(t)dt = r(1) = τ (c),
0
where we have used the facts that γ is timelike, ṙ(t) > 0 for all t ∈ [0, 1]
(as γ̇ is future-pointing) and τ (c) = r(1) (as q = expp (r(1)n(1))). It should
be clear that τ (γ) = τ (c) if and only if |Y (t)| ≡ 0 ⇔ Y (t) ≡ 0 (Y (t) is
spacelike or zero) for all t ∈ [0, 1], implying that n is constant. In this case,
γ(t) = expp (r(t)n) is, up to reparameterization, the geodesic through p with
initial condition n ∈ Tp M . 
There is also a local property characterizing null geodesics.
Proposition 1.4. Let (M, g) be a time-oriented spacetime, p0 ∈ M and
V ⊂ M a geodesically convex open neighborhood of p0 . The spacetime (V, g)
obtained by restricting g to V satisfies the following property: if for p, q ∈ V
there exists a future-directed null geodesic c connecting p to q and γ is a
causal curve connecting p to q then γ is a reparameterization of c.
 2. CAUSALITY CONDITIONS 59

Proof. Since p and q are connected by a null geodesic, we conclude

from Proposition 1.1 that q ∈ J + (p) \ I + (p). Let γ : [0, 1] → V be a
causal curve connecting p to q. Then we must have γ(t) ∈ J + (p) \ I + (p)
for all t ∈ [0, 1], since γ(t0 ) ∈ I + (p) implies γ(t) ∈ I + (p) for all t > t0 (see
Proposition 1.2). Consequently, we have
D E
Wp (γ(t)) = 0 ⇒ (grad Wp )γ(t) , γ̇(t) = 0,
where Wp was defined in the proof of Proposition 1.1. The formula
(grad Wp )cv (1) = 2ċv (1),
which was proved for timelike geodesics cv with initial condition v ∈ Tp M ,
must also hold for null geodesics (by continuity). Hence grad Wp is tangent
to the null geodesics ruling J + (p) \ I + (p) and future-pointing. Since γ̇(t)
is also future-pointing, we conclude that γ̇ is proportional to grad Wp , and
therefore γ is a reparameterization of a null geodesic, which must be c. 
Corollary 1.5. Let (M, g) be a time-oriented spacetime and p ∈ M .
If q ∈ J + (p) \ I + (p) then any future-directed causal curve connecting p to q
must be a reparameterized null geodesic.

2. Causality conditions
For physical applications, it is important to require that the spacetime
satisfies reasonable causality conditions. The simplest of these conditions
excludes time travel, i.e. the possibility of a particle returning to an event
in its past history.
Definition 2.1. A spacetime (M, g) is said to satisfy the chronology
condition if it does not contain closed timelike curves.
This condition is violated by compact spacetimes:
Proposition 2.2. Any compact spacetime (M, g) contains closed time-
like curves.
Proof. Taking if necessary the time-orientable double covering, we can
assume that (M, g) is time-oriented. Since I + (p) is an open set for any
p ∈ M , it is clear that {I + (p)}p∈M is an open cover of M . If M is compact,
we can obtain a finite subcover {I + (p1 ), . . . , I + (pN )}. Now if p1 ∈ I + (pi ) for
i 6= 1 then I + (p1 ) ⊂ I + (pi ), and we can exclude I + (p1 ) from the subcover.
Therefore, we can assume without loss of generality that p1 ∈ I + (p1 ), and
hence there exists a closed timelike curve starting and ending at p1 . 
A stronger restriction on the causal behavior of the spacetime is the
following:
Definition 2.3. A spacetime (M, g) is said to be stably causal if there
exists a global time function, i.e. a smooth function t : M → R such that
grad(t) is timelike.
60 3. CAUSALITY

In particular, a stably causal spacetime is time-orientable. We choose

the time orientation defined by − grad(t), so that t increases along future-
directed timelike curves. Notice that this implies that no closed timelike
curves can exist, i.e. any stably causal spacetime satisfies the chronology
condition. In fact, any small perturbation of a stably causal spacetime still
satisfies the chronology condition (Exercise 4).
Let (M, g) be a time-oriented spacetime. A smooth future-directed
causal curve c : (a, b) → M (with possibly a = −∞ or b = +∞) is said
to be future-inextendible if limt→b c(t) does not exist. The definition of
a past-inextendible causal curve is analogous. The future domain of
dependence of S ⊂ M is the set D+ (S) of all events p ∈ M such that
any past-inextendible causal curve starting at p intersects S. Therefore any
causal influence on an event p ∈ D+ (S) had to register somewhere in S,
and one can expect that what happens at p can be predicted from data on
S. Similarly, the past domain of dependence of S is the set D− (S) of
all events p ∈ M such that any future-inextendible causal curve starting at
p intersects S. Therefore any causal influence of an event p ∈ D− (S) will
register somewhere in S, and one can expect that what happened at p can
be retrodicted from data on S. The domain of dependence of S is simply
the set D(S) = D+ (S) ∪ D− (S).
Let (M, g) be a stably causal spacetime with time function t : M →
R. The level sets Sa = t−1 (a) are said to be Cauchy hypersurfaces if
D(Sa ) = M . Spacetimes for which this happens have particularly good
causal properties.
Definition 2.4. A stably causal spacetime possessing a time function
whose level sets are Cauchy hypersurfaces is said to be globally hyper-
bolic.
Notice that the future and past domains of dependence of the Cauchy
hypersurfaces Sa are D+ (Sa ) = t−1 ([a, +∞)) and D− (Sa ) = t−1 ((−∞, a]).

3. Exercises
(1) Let (M, g) be the quotient of the 2-dimensional Minkowski space-
time by the discrete group of isometries generated by the map
f (t, x) = (−t, x + 1). Show that (M, g) is not time orientable.
(2) Let (M, g) be a time-oriented spacetime and p ∈ M . Show that:
(a) I + (p) is open;
(b) J + (p) is not necessarily closed;
(c) J + (p) ⊂ I + (p);
(d) I + (p) = int J + (p);
(e) if r ∈ J + (p) and q ∈ I + (r) then q ∈ I + (p);
(f) if r ∈ I + (p) and q ∈ J + (r) then q ∈ I + (p).
(3) Consider the 3-dimensional Minkowski spacetime (R3 , g), where
g = −dt2 + dx2 + dy 2 .
 3. EXERCISES 61

Let c : R → R3 be the curve c(t) = (t, cos t, sin t). Show that
although ċ(t) is null for all t ∈ R we have c(t) ∈ I + (c(0)) for all
t > 0. What kind of motion does this curve represent?
(4) Let (M, g) be a stably causal spacetime and h an arbitrary sym-
metric (2, 0)-tensor field with compact support. Show that for suf-
ficiently small |ε| the tensor field gε = g + εh is still a Lorentzian
metric on M , and (M, gε ) satisfies the chronology condition.
(5) Let (M, g) be the quotient of the 2-dimensional Minkowski space-
time by the discrete group of isometries generated by the map
f (t, x) = (t + 1, x + 1). Show that (M, g) satisfies the chronology
condition, but there exist arbitrarily small perturbations of (M, g)
(in the sense of Exercise 4) which do not.
(6) Let (M, g) be a time-oriented spacetime and S ⊂ M . Show that:
(a) S ⊂ D+ (S);
(b) D+ (S) is not necessarily open;
(c) D+ (S) is not necessarily closed.
(7) Let (M, g) be the 2-dimensional spacetime obtained by removing
the (closed) positive x-semi-axis of Minkowski 2-dimensional space-
time (cf. Figure 1). Show that:
(a) (M, g) is stably causal but not globally hyperbolic;
(b) there exist points p, q ∈ M such that J + (p) ∩ J − (q) is not
compact;
(c) there exist points p, q ∈ M with q ∈ I + (p) such that the
supremum of the lengths of timelike curves connecting p to q
is not attained by any timelike curve.
t t

D(S) J + (p)
x x

Figure 1. Stably causal but not globally hyperbolic spacetime.

(8) Let (Σ, h) be a 3-dimensional Riemannian manifold. Show that the

spacetime (M, g) = (R × Σ, −dt ⊗ dt + h) is globally hyperbolic if
and only if (Σ, h) is complete.
(9) Show that the following spacetimes are globally hyperbolic:
62 3. CAUSALITY

(a) the Minkowski spacetime;

(b) the FLRW spacetimes;
(c) the region {r > 2m} of the Schwarzschild spacetime;
(d) the region {r < 2m} of the Schwarzschild spacetime;
(e) the maximal analytical extension of the Schwarzschild space-
time.
(10) Let (M, g) be a global hyperbolic spacetime with Cauchy hyper-
surface S. Show that M is diffeomorphic to R × S.
 CHAPTER 4

Singularity theorems

Many solutions of the Einstein field equations, such as the Schwarzschild

solution or the FLRW cosmological models, display singularities, beyond
which timelike and null geodesics cannot be continued. The singularity
theorems of Hawking and Penrose, proved in this chapter, show that this is
a generic phenomenon: any sufficiently small perturbation of these singular
solutions will still be singular. We follow [Wal84] when discussing conjugate
points, and [GN14] when proving the existence of maximizing geodesics on
globally hyperbolic spacetimes. See also [O’N83, Pen87, Nab88, HE95].

1. Geodesic congruences
Let (M, g) be a Lorentzian manifold. A congruence of curves in an
open set U ⊂ M is the family of integral curves of a nonvanishing vector
field X in U . We will assume that X is unit timelike and geodesic, that is,
hX, Xi = −1 and ∇X X = 0.
The properties of the congruence determined by X are best analyzed by
considering its second fundamental form
Bµν = ∇ν Xµ .
This tensor is purely spatial, that is,
Bµν X µ = Bµν X ν = 0.
Indeed, since X is unit,
1
Bµν X µ = X µ ∇ν Xµ = ∇ν (Xµ X µ ) = 0.
2
On the other hand, because X is geodesic,
Bµν X ν = X ν ∇ν Xµ = ∇X Xµ = 0.
Proposition 1.1. The second fundamental form B satisfies
∇X Bµν = −Bµα B αν + Rανµβ X α X β .
Proof. We have
X α ∇α Bµν = X α ∇α ∇ν Xµ = X α ∇ν ∇α Xµ + X α Rανµβ X β
= ∇ν (X α ∇α Xµ ) − (∇ν X α )(∇α Xµ ) + Rανµβ X α X β
= −Bµα B αν + Rανµβ X α X β .

63
64 4. SINGULARITY THEOREMS


Let c(t, s) be a one-parameter family of geodesics of the congruence,
parameterized such that
∂c
=X
∂t
(Figure 1). The Jacobi vector field associated to c is
∂c
Y = .
∂s

Figure 1. Jacobi vector field.

Proposition 1.2. The Jacobi vector field satisfies

∇ X Y µ = B µν Y ν .
Proof. The definition of Y implies that
[X, Y ] = 0 ⇔ ∇X Y − ∇Y X = 0.
Consequently, we have
∇ X Y µ = ∇ Y X µ = Y ν ∇ ν X µ = B µν Y ν .

The equation for ∇X Bµν then yields the following famous result.
Proposition 1.3. The Jacobi vector field satisfies the Jacobi equation
∇X ∇X Y = R(X, Y )X.
 1. GEODESIC CONGRUENCES 65

Proof. We have
∇X ∇X Y α = ∇X (B αβ Y β ) = (∇X B αβ )Y β + B αβ ∇X Y β
= −B αµ Bµβ Y β + Rµβ αν X µ X ν Y β + B αβ B β µ Y µ
= Rαβµν X β X µ Y ν .
Alternatively, we can simply notice that
∇X ∇X Y = ∇X ∇Y X = ∇Y ∇X X + R(X, Y )X = R(X, Y )X.

Conversely, it can be shown that any solution of the Jacobi equation
along a timelike geodesic c is a Jacobi vector field for some geodesic congru-
ence including c.
We now define the kinematic quantities associated to the congruence.
Definition 1.4. The spatial metric associated to the congruence is
hµν = gµν + Xµ Xν .
The expansion, shear and vorticity are defined as1
θ = hµν Bµν = g µν Bµν ,
1
σµν = B(µν) − θhµν ,
3
ωµν = B[µν] ,
so that we have the decomposition
1
Bµν = θhµν + σµν + ωµν .
3
Note that all the tensors above are purely spatial:
hµν X ν = σµν X ν = ωµν X ν = 0,
Moreover, the trace of h is
hµν hµν = g µν hµν = g µν (gµν + Xµ Xν ) = 4 − 1 = 3,
and so the shear is traceless:
hµν σµν = g µν σµν = 0.
Fix a geodesic c, and let Y be a Jacobi field along c. If Y is initially
orthogonal to c then it will remain orthogonal:
X · (Xµ Y µ ) = (∇X Xµ )Y µ + Xµ ∇X Y µ = Xµ B µν Yν = 0.
In an orthonormal frame {X, E1 , E2 , E3 } parallel along c we then have
 
i j 1 1
Ẏ = Bij Y = θδij + σij + ωij Y j = θY i + σij Y j + ωij Y j
3 3
1Curved brackets indicate symmetrization: B 1
(µν) = 2
(Bµν + Bνµ ).
66 4. SINGULARITY THEOREMS

(i = 1, 2, 3). If we consider a small spacelike sphere in the hypersurface

orthogonal to c and let it be carried by the geodesics of the congruence, we
see that θ measures the rate at which the sphere’s volume grows, σ describes
the sphere’s volume-preserving shape deformations, and ω gives the sphere’s
angular velocity.
Proposition 1.5. The expansion of the congruence satisfies the Ray-
chaudhuri equation
1
X · θ = − θ2 − σµν σ µν + ωµν ω µν − Rµν X µ X ν .
3
Proof. Taking the trace of the equation for ∇X Bµν (that is, contracting
with g µν ) yields
X · θ = −Bµν B νµ − Rαβ X α X β
  
1 1 µν µν µν
=− θhµν + σµν + ωµν θh + σ − ω − Rµν X µ X ν
3 3
1
= − θ2 − σµν σ µν + ωµν ω µν − Rµν X µ X ν .
3


2. Energy conditions
Definition 2.1. A given energy-momentum tensor Tµν , with trace T ≡
g µν Tµν , is said to satisfy:
(1) the strong energy condition (SEC) if Tµν X µ X ν + 12 T ≥ 0 for
all unit timelike vectors X;
(2) the weak energy condition (WEC) if Tµν X µ X ν ≥ 0 for all
timelike vectors X;
(3) the null energy condition (NEC) if Tµν X µ X ν ≥ 0 for all null
vectors X;
(4) the dominant energy condition (DEC) if −T µν Xν is causal
and future-pointing for all causal future-pointing vectors X.
The weak energy condition is the reasonable requirement that any ob-
server should measure a non-negative energy density, and the null energy
condition can be thought of as the same requirement for observers moving
at the speed of light. The dominant energy condition, on the other hand,
demands that any observer should measure the flow of energy and momen-
tum to be causal. To understand the strong energy condition, we write the
Einstein equations as
1
Rµν − Rgµν = 8πTµν
2
(possibly including the cosmological constant in the energy-momentum ten-
sor). Note that the trace of this equation yields
−R = 8πT,
 3. CONJUGATE POINTS 67

and so the Einstein equations can also be written as

 
1
Rµν = 8π Tµν − T gµν .
2
Therefore the strong energy condition simply requires that the Ricci tensor
satisfies Rµν X µ X ν ≥ 0 for all timelike vectors X (given that the Einstein
equations are written as above).
Generically, the energy-momentum tensor is diagonalizable, that is, there
exists an orthonormal frame {E0 , E1 , E2 , E3 } in which the energy-momentum
tensor is diagonal,
(Tµν ) = diag(ρ, p1 , p2 , p3 ).
The timelike eigenvalue ρ is called the rest energy density, and the space-
like eigenvalues p1 , p2 , p3 are known as the principal pressures. In terms
of these eigenvalues, we have:
(1) SEC ⇔ ρ + 3i=1 pi ≥ 0 and ρ + pi ≥ 0 (i = 1, 2, 3).
P
(2) WEC ⇔ ρ ≥ 0 and ρ + pi ≥ 0 (i = 1, 2, 3).
(3) NEC ⇔ ρ + pi ≥ 0 (i = 1, 2, 3).
(4) DEC ⇔ ρ ≥ |pi | (i = 1, 2, 3).
Using this characterization, it is easy to see that the NEC is the weakest
energy condition, that is, it is implied by any of the other conditions. The
remaining three energy conditions are largely independent, except that the
DEC implies the WEC. Notice that in particular the SEC does not imply
the WEC.

3. Conjugate points
Definition 3.1. Let (M, g) be a Lorentzian manifold. A point q ∈ M is
said to be conjugate to p ∈ M along a timelike geodesic c if there exists a
nonvanishing solution Y of the Jacobi equation ∇X ∇X Y = R(X, Y )X such
that Yp = Yq = 0.
Informally, two points p and q are conjugate along c if there exists a
nearby timelike geodesic intersecting c at both p and q (Figure 2).
Chose an orthonormal frame {X, E1 , E2 , E3 } parallel along c, where X
is the unit tangent vector. Let Y be a Jacobi field Y which vanishes at
p = c(0). If we write
Y = Y 0 X + Y i Ei
then the Jacobi equation becomes
(
Ÿ 0 = 0
.
Ÿ i = Ri 00j Y j

Since Y 0 is an affine function of the proper time τ , it must vanish identically

if Y vanishes again along c. We will therefore assume that Y 0 = 0, that is,
68 4. SINGULARITY THEOREMS

Figure 2. Conjugate points.

that Y is orthogonal to c. Since the remaining components of Y satisfy a

linear ODE, we know that
Y i (τ ) = Aij (τ )Ẏ j (0),
where A(τ ) is the fundamental matrix solution vanishing at τ = 0:

A(0) = 0

Ȧ(0) = I .
i

Äij = R 00k Akj


Although A(0) = 0 is singular, A(τ ) is not singular for τ > 0 sufficiently

small, because Ȧ(0) = I. If A(τ ) becomes singular for some τ∗ > 0 then
q = c(τ∗ ) is conjugate to p (we just have to choose Ẏ (0) to be a nonvanishing
column vector in the kernel of A(τ∗ )).
Consider the congruence of timelike geodesics through p. Since on the
one hand
Ẏ i = Bij Y j
and on the other
Ẏ = ȦẎ (0) = ȦA−1 AẎ (0) = ȦA−1 Y,
we conclude that
B = ȦA−1 ,
and so
d
θ = tr B = tr(ȦA−1 ) = log(det A).
dτ
 3. CONJUGATE POINTS 69

Therefore the expansion of the congruence blows up if and only if the geo-
desic approaches the first conjugate point q.
Theorem 3.2. Let (M, g) be a 4-dimensional Lorentzian manifold sat-
isfying the SEC, c a timelike geodesic and p = c(0). Suppose that the ex-
pansion θ of the congruence of timelike geodesics through p takes a negative
value θ0 < 0 at some point r = c(τ0 ), with τ0 > 0. Then there exists a point
q conjugate to p along c at a distance at most |θ30 | from r.

Proof. By the Gauss Lemma, we can find local coordinates (t, x1 , x2 , x3 )

such that X ♭ = dt, and so
dX ♭ = 0 ⇔ ∇[µ Xν] = 0.
In other words, the congruence has no vorticity, and the Raychaudhuri equa-
tion becomes
dθ 1
= − θ2 − σµν σ µν − Rµν X µ X ν .
dτ 3
Because σ is purely spatial and (M, g) satisfies the SEC, we have
dθ 1 1 dθ 1 1 1 1
≤ − θ2 ⇔ − 2 ≥ ⇒ ≥ + (τ − τ0 ).
dτ 3 θ dτ 3 θ θ0 3
1
We conclude that θ vanishes, and so θ blows up, at proper time at most
τ0 + |θ30 | . 

Let c(t, s) be a one-parameter family of timelike curves connecting two

points p and q:
c(t0 , s) = p and c(t1 , s) = q
for all s. Then the connecting vector
∂c
Y = ,
∂s
which in general is not a Jacobi field, satisfies
Yp = Yq = 0.
We assume that c(t, s) has been parameterized in such a way that Y does
not vanish identically.
The tangent vector
∂c
X=
∂t
is timelike, and if we define
1
f (t, s) = (− hX, Xi) 2
then the length of each curve is
Z t1
τ (s) = f (t, s)dt.
t0
70 4. SINGULARITY THEOREMS

We have
Z t1 Z t1 Z t1
dτ ∂f 1 1
= dt = − hX, ∇Y Xi dt = − hX, ∇X Y i dt
ds t0 ∂s t0 f t0 f
Z t1   Z t1    
1 X
=− X· hX, Y i dt + ∇X , Y dt
t0 f t0 f
Z t1    
X
= ∇X , Y dt,
t0 f
where we used the Fundamental Theorem of Calculus and the fact that
Yp = Yq = 0. This shows that the timelike curve c defined as c(t) = c(t, 0)
has extremal length among all timelike curves in such one-parameter families
if and only if  
X
∇X = 0,
f
that is, if and only if it is a timelike geodesic. Assume this to be the case.
Then
Z t1 Z t1 
d2 τ
      
X X
2
(0) = Y · ∇X , Y dt = ∇Y ∇X , Y dt.
ds t0 f t0 f
Assuming that f (t, 0) = 1 (that is, c is parameterized by its proper time)
and hX, Y i = 0 for s = 0 (which is always possible by reparameterizing
c(t, s)) leads to
Z t1
d2 τ
(0) = h∇Y ∇X X, Y i dt.
ds2 t0
Finally, using
R(X, Y )Z = ∇X ∇Y Z − ∇Y ∇X Z − ∇[X,Y ] Z = ∇X ∇Y Z − ∇Y ∇X Z
we obtain
t1
d2 τ
Z
(0) = h∇X ∇Y X − R(X, Y )X, Y i dt
ds2 t0
Z t1
= h∇X ∇X Y − R(X, Y )X, Y i dt.
t0

Theorem 3.3. A timelike curve c connecting the points p, q ∈ M lo-

cally maximizes the proper time (along any one-parameter family of time-
like curves connecting the same points) if and only if it is a timelike geodesic
without conjugate points to p between p and q.
Proof. It is clear from what was done above that c being a geodesic is
a necessary condition.
Let us assume that c has no conjugate points (to p, say) between p and
q. In an orthonormal frame parallel along c we have
Z t1
d2 τ i

i i j

(0) = Y Ÿ − R 00j Y dt.
ds2 t0
 3. CONJUGATE POINTS 71

Because there are no conjugate points, the fundamental matrix solution A(t)
is nondegenerate for t ∈ (t0 , t1 ], and we can set
Y i = Aij Z j .
We have
Ÿ i = Aij Z̈ j + 2Ȧij Ż j + Äij Z j = Aij Z̈ j + 2Ȧij Ż j + Ri 00k Akj Z j ,
and so
Z t1
d2 τ j

k k

(0) = A ij Z Aik Z̈ + 2Ȧik Ż dt
ds2 t0
Z t1  
= Z t At AZ̈ + 2ȦŻ dt
t0
t1  
d  t t
Z 
t t t t t t
= Z A AŻ − Ż A AŻ − Z Ȧ AŻ + Z A ȦŻ dt
t0 dt
Z t1 Z t1  
t
=− (AŻ) AŻdt + Z t At Ȧ − Ȧt A Żdt.
t0 t0
Above we used the Fundamental Theorem of Calculus and the fact that
(AZ)t AŻ = Y t (Ẏ − ȦZ) = Y t (Ẏ − BY ) vanishes at t0 and t1 (although B
blows up as (t − t0 )−1 , Y vanishes as t − t0 or faster by Taylor’s formula).
From Ȧ = BA we have
At Ȧ − Ȧt A = At BA − At B t A = At B − B t A = 2At ωA = 0,

because the vorticity matrix ω vanishes for the congruence of timelike geodesics
through p. Therefore
Z t1
d2 τ
(0) = − (AŻ)t AŻdt ≤ 0,
ds2 t0
with equality if and only if
Ż ≡ 0 ⇒ Z ≡ 0 ⇒ Y ≡ 0
(note that if Z is constant then it must be zero because 0 = Yq = A(t1 )Z
and A(t1 ) is nonsingular). We conclude that c is indeed a maximum of the
proper time along any one-parameter family of timelike curves connecting p
and q.
On the other hand, if there exists a conjugate point along c between p
and q, say r = c(t∗ ), then let Ŷ be a nonvanishing Jacobi field such that
Ŷ (t0 ) = Ŷ (t∗ ) = 0 (in particular Ŷ is orthogonal to c), and let Y be the
vector field along c that coincides with Ŷ between p and r and is zero between
r and q. Similarly, let Ẑ be the (necessarily spacelike) vector field parallel
along c such that Ẑ(t∗ ) = −∇X Ŷ (t∗ ), and let Z(t) = θ(t)Ẑ(t), where θ is
a smooth function satisfying θ(t0 ) = θ(t1 ) = 0 and θ(t∗ ) = 1. Finally, let
Yε be the vector field along c defined by Yε = Y + εZ, and consider a one-
parameter family of curves cε (t, s) such that cε (t, 0) = c(t) and Yε = ∂c ∂s .
ε
72 4. SINGULARITY THEOREMS

Since Yε is not C 1 , we must write the formula for the second derivative of
the length as
Z t1 
d2 τ

(0) = − h∇X Yε , ∇X Yε i + hR(X, Yε )X, Yε i dt = I(Yε , Yε ),
ds2 t0

where the bilinear form I is clearly symmetric. Therefore

d2 τ
(0) = I(Y, Y ) + 2εI(Y, Z) + ε2 I(Z, Z).
ds2
Since Y is a Jacobi field between p and r, and is zero between r and q, we
have I(Y, Y ) = 0. On the other hand,
Z t∗  
I(Y, Z) = − h∇X Y, ∇X Zi + hR(X, Y )X, Zi dt
t0
 t ∗ Z t∗  
= − h∇X Y, Zi + h∇X ∇X Y, Zi − hR(X, Y )X, Zi dt
t0 t0
D E
∗ ∗
= ∇X Ŷ (t ), ∇X Ŷ (t ) > 0.

Therefore for ε > 0 sufficiently small the one-parameter family cε (t, s) con-
tains curves whose length is greater than the length of c.
Figure 3 illustrates the geometric idea behind the proof above: cs rep-
resents a generic curve of a one-parameter family corresponding to Y , and
has the same length as c; adding εZ changes cs between points u and v, say,
making it longer by the twin paradox.


The results above can be generalized for timelike geodesics orthogonal to

a spacelike hypersurface S (that is, a hypersurface whose induced metric
is Riemannian, or, equivalently, whose normal vector is timelike). If one
considers the congruence of such geodesics then at S
∇X Yµ = ∇Y Xµ = Y ν (∇ν Xµ ) = Y ν (∇(ν Xµ) + ∇[ν Xµ] )
1
= ∇(µ Xν) Y ν = LX gµν Y ν = Kµν Y ν .
2
Definition 3.4. Let (M, g) be a Lorentzian manifold and let S ⊂ M be
a spacelike hypersurface with second fundamental form K. A point q ∈ M
is said to be conjugate to S along a timelike geodesic c orthogonal to S
at some point p ∈ S if there exists a nonvanishing solution Y of the Jacobi
equation ∇X ∇X Y = R(X, Y )X such that Yp ∈ Tp S, (∇X Yµ )p = (Kµν Y ν )p
and Yq = 0.
In an orthonormal frame {X, E1 , E2 , E3 } parallel along c, again we can
assume that Y 0 = 0, and have for the remaining components
Y i (t) = Aij (t)Y j (0),
 3. CONJUGATE POINTS 73

v
∇X Ŷ (t∗ )

r
u
Y

cs
c

Figure 3. Proof of Theorem 3.3.

where A(t) is the fundamental matrix solution:


A(0) = I

Ȧ(0) = K .
Äij = Ri 00k Akj



Arguing as above, we have the following result.

Theorem 3.5. Let (M, g) be a 4-dimensional Lorentzian manifold sat-

isfying the SEC, S ⊂ M a spacelike hypersurface and c a timelike geodesic
orthogonal to S at some point p ∈ S. Suppose that the expansion θ of
the congruence of timelike geodesics orthogonal to S takes a negative value
θ0 < 0 at p. Then there exists a point q conjugate to S along c at a distance
at most |θ30 | from S.

Theorem 3.6. A timelike curve c connecting the spacelike hypersurface

S ⊂ M to the point q ∈ M locally maximizes the proper time (along any
one-parameter family of timelike curves connecting S to q) if and only if it
is a timelike geodesic orthogonal to S without conjugate points to S between
S and q.
74 4. SINGULARITY THEOREMS

Proof. The proof is basically the same as for curves connecting two
points. The main differences are the in formula
Z t1   Z t1    
dτ 1 X
=− X· hX, Y i dt + ∇X , Y dt
ds t0 f t0 f
Z t1    
1 X
= hXp , Yp i + ∇X , Y dt,
f (t0 , s) t0 f
which requires c to be orthogonal to S at p = c(t0 ); the fact that A(t0 ) does
not vanish, but instead
(AZ)t AŻ = Y t (Ẏ − ȦZ) = Y t (Ẏ − BAZ)
= Y t (Ẏ − BY ) = Y t (Ẏ − KY ) = 0
at t0 (where we assume without loss of generality that the curves c(t, s) are
orthogonal to S); and the integrated formula
Z t1 
d2 τ

µ ν
(0) = −(Kµν Y Y )(p) − h∇X Y, ∇X Y i + hR(X, Y )X, Y i dt
ds2 t0
= −(Kµν Y µ Y ν )(p) + I(Y, Y ),
which vanishes when Y is a Jacobi field. 

4. Existence of maximizing geodesics

Proposition 4.1. Let (M, g) be a globally hyperbolic spacetime, S a
Cauchy hypersurface and p ∈ D+ (S). Then D+ (S) ∩ J − (p) is compact.
Proof. Let us define a simple neighborhood U ⊂ M to be a geodesi-
cally convex open set diffeomorphic to an open ball whose boundary is a
compact submanifold of a larger geodesically convex open set (therefore ∂U
is diffeomorphic to S 3 and U is compact). It is clear that simple neighbor-
hoods form a basis for the topology of M . Also, it is easy to show that any
open cover {Vα }α∈A has a countable, locally finite refinement {Un }n∈N by
simple neighborhoods.
If A = D+ (S) ∩ J − (p) were not compact, there would exist a count-
able, locally finite open cover {Un }n∈N of A by simple neighborhoods not
admitting any finite subcover. Take qn ∈ A ∩ Un such that qm 6= qn for
m 6= n. The sequence {qn }n∈N cannot have accumulation points, since any
point in M has a neighborhood intersecting only finitely many simple neigh-
borhoods Un . In particular, each simple neighborhood Un contains only a
finite number of points in the sequence (as U n is compact).
Set p1 = p. Since p1 ∈ A, we have p1 ∈ Un1 for some n1 ∈ N. Let
qn 6∈ Un1 . Since qn ∈ J − (p1 ), there exists a future-directed causal curve
cn connecting qn to p1 . This curve will necessarily intersect ∂Un1 . Let r1,n
be an intersection point. Since Un1 contains only a finite number of points
in the sequence {qn }n∈N , there will exist infinitely many intersection points
r1,n . As ∂Un1 is compact, these will accumulate to some point p2 ∈ ∂Un1
(cf. Figure 4).
 4. EXISTENCE OF MAXIMIZING GEODESICS 75

Because U n1 is contained in a geodesically convex open set V , which can

be chosen so that v 7→ (π(v), exp(v)) is a diffeomorphism onto V × V , we
have p2 ∈ J − (p1 ): if γ1,n is the unique causal geodesic connecting p1 to r1,n ,
parameterized by the global time function t : M → R, then the subsequence
of {γ1,n } corresponding to a convergent subsequence of {r1,n } will converge
to a causal geodesic γ1 connecting p1 to p2 . If S = t−1 (0) then we have
t(r1,n ) ≥ 0, implying that t(p2 ) ≥ 0 and hence p2 ∈ A. Since p2 6∈ Un1 , there
must exist n2 ∈ N such that p2 ∈ Un2 .
Since Un2 contains only a finite number of points in the sequence {qn }n∈N ,
an infinite number of curves cn must intersect ∂Un2 to the past of r1,n . Let
r2,n be the intersection points. As ∂Un2 is compact, {r2,n } must accumulate
to some point p3 ∈ ∂Un2 . Because U n2 is contained in a geodesically con-
vex open set, p3 ∈ J − (p2 ): if γ2,n is the unique causal geodesic connecting
r1,n to r2,n , parameterized by the global time function, then the subse-
quence of {γ2,n } corresponding to convergent subsequences of both {r1,n }
and {r2,n } will converge to a causal geodesic connecting p2 to p3 . Since
J − (p2 ) ⊂ J − (p1 ) and t(r2,n ) ≥ 0 ⇒ t(p3 ) ≥ 0, we have p3 ∈ A.
Iterating the procedure above, we can construct a sequence {pi }i∈N of
points in A satisfying pi ∈ Uni with ni 6= nj if i 6= j, such that pi is connected
to pi+1 by a causal geodesic γi . It is clear that γi cannot intersect S, for
t(pi+1 ) > t(pi+2 ) ≥ 0. On the other hand, the piecewise smooth causal curve
obtained by joining the curves γi can easily be smoothed into a past-directed
causal curve starting at p1 which does not intersect S. Finally, such curve is
inextendible: it cannot converge to any point, as {pi }i∈N cannot accumulate.
But since p1 ∈ D+ (S), this curve would have to intersect S. Therefore A
must be compact. 

Corollary 4.2. Let (M, g) be a globally hyperbolic spacetime and p, q ∈

M . Then
(i) J + (p) is closed;
(ii) J + (p) ∩ J − (q) is compact.
Proof. Exercise. 
Proposition 4.1 is a key ingredient in establishing the following funda-
mental result.
Theorem 4.3. Let (M, g) be a globally hyperbolic spacetime with Cauchy
hypersurface S, and p ∈ D+ (S). Then, among all timelike curves connecting
p to S, there exists a timelike curve with maximal length. This curve is a
timelike geodesic, orthogonal to S.
Proof. Consider the set T (S, p) of all timelike curves connecting S to p.
Since we can always use the global time function t : M → R as a parameter,
these curves are determined by their images, which are compact subsets of
the compact set A = D+ (S) ∩ J − (p). As it is well known (see for instance
[Mun00]), the set C(A) of all compact subsets of A is a compact metric
76 4. SINGULARITY THEOREMS

U n1

p = p1

U n2

p2

p3

Figure 4. Proof of Proposition 4.1.

space for the Hausdorff metric dH , defined as follows: if d : M × M → R

is a metric yielding the topology of M ,
dH (K, L) = inf{ε > 0 | K ⊂ Uε (L) and L ⊂ Uε (K)},
where Uε (K) is a ε-neighborhood of K for the metric d. Therefore, the
closure C(S, p) = T (S, p) is a compact subset of C(A). It is not difficult
to show that C(S, p) can be identified with the set of continuous causal
curves connecting S to p (a continuous curve c : [0, t(p)] → M is said to be
causal if c(t2 ) ∈ J + (c(t1 )) whenever t2 > t1 ).
The length function τ : T (S, p) → R is defined by
Z t(p)
τ (c) = |ċ(t)|dt.
0
This function is upper semicontinuous, i.e. continuous for the topology
O = {(−∞, a) | −∞ ≤ a ≤ +∞}
in R. Indeed, let c ∈ T (S, p) be parameterized by its arclength u. For a suf-
ficiently small ε > 0, the function u can be extended to the ε-neighborhood
Uε (c) in such a way that its level hypersurfaces are spacelike and orthogonal
to c, that is, − grad u is timelike and coincides with ċ on c (cf. Figure 5). If
 4. EXISTENCE OF MAXIMIZING GEODESICS 77

γ ∈ T (S, p) is in the open ball Bε (c) ⊂ C(A) for the Hausdorff metric dH
then we can use u as a parameter, thus obtaining
du(γ̇) = 1 ⇔ hγ̇, grad ui = 1.
Therefore γ̇ can be decomposed as
1
γ̇ = grad u + X,
hgrad u, grad ui
where X is spacelike and orthogonal to grad u, and so
1
1 2
|γ̇| = + hX, Xi .
hgrad u, grad ui
Given δ > 0, we can choose ε > 0 sufficiently small so that
 2
1 δ
− < 1+
hgrad u, grad ui 2τ (c)
on the ε-neighborhood Uε (c) (as hgrad u, grad ui = −1 along c). We have
Z t(p) Z t(p) Z τ (c)
dγ du
τ (γ) = dt = |γ̇| dt = |γ̇| du,
0 dt 0 dt u(γ∩S)
where we have to allow for the fact that c is not necessarily orthogonal to
S, and so the initial point of γ is not necessarily at u = 0 (cf. Figure 5).
Consequently,
Z τ (c) 1
1 2
τ (γ) = − − hX, Xi du
u(γ∩S) hgrad u, grad ui
Z τ (c)    
δ δ
< 1+ du = 1 + (τ (c) − u(γ ∩ S)) .
u(γ∩S) 2τ (c) 2τ (c)
Choosing ε sufficiently small so that
2 −1
 
1
|u| < +
τ (c) δ
on S ∩ Uε (c), we obtain τ (γ) < τ (c) + δ, proving upper semicontinuity in
T (S, p). As a consequence, the length function can be extended to C(S, p)
through
τ (c) = lim sup{τ (γ) | γ ∈ Bε (c) ∩ T (S, p)}
ε→0
(as for ε > 0 sufficiently small the supremum will be finite). Also, it is clear
that if c ∈ T (S, p) then the upper semicontinuity of the length forces the
two definitions of τ (c) to coincide. The extension of the length function to
C(S, p) is trivially upper semicontinuous: given c ∈ C(S, p) and δ > 0, let
ε > 0 be such that τ (γ) < τ (c) + 2δ for any γ ∈ B2ε (c) ∩ T (S, p). Then it is
clear that τ (c′ ) ≤ τ (c) + 2δ < τ (c) + δ for any c′ ∈ Bε (c) ∩ C(S, p).
Finally, we notice that the compact sets of R for the topology O are the
sets with a maximum. Therefore, the length function attains a maximum
at some point c ∈ C(S, p). All that remains to be seen is that the maximum
78 4. SINGULARITY THEOREMS

is also attained at a smooth timelike curve γ. To do so, cover c with finitely

many geodesically convex neighborhoods and choose points p1 , . . . , pk in c
such that p1 ∈ S, pk = p and the portion of c between pi−1 and pi is contained
in a geodesically convex neighborhood for all i = 2, . . . , k. It is clear that
there exists a sequence cn ∈ T (S, p) such that cn → c and τ (cn ) → τ (c).
Let ti = t(pi ) and pi,n be the intersection of cn with t−1 (ti ). Replace cn
by the sectionally geodesic curve γn obtained by joining pi−1,n to pi,n in
the corresponding geodesically convex neighborhood. Then τ (γn ) ≥ τ (cn ),
and therefore τ (γn ) → τ (c). Since each sequence pi,n converges to pi , γn
converges to the sectionally geodesic curve γ obtained by joining pi−1 to pi
(i = 2, . . . , k), and it is clear that τ (γn ) → τ (γ) = τ (c). Therefore γ is a
point of maximum for the length. Finally, we notice that γ must be smooth
at the points pi , for otherwise we could increase its length by using the twin
paradox. Therefore γ must be a timelike geodesic. It is also clear that γ
must be orthogonal to S, for otherwise it would be possible to increase its
length by small deformations. 

Uε (c)
p
u = τ (c)

γ c

S u=0

Figure 5. Proof of Theorem 4.3.

5. Hawking’s singularity theorem

We have now all the necessary ingredients to prove the Hawking singu-
larity theorem.
Definition 5.1. A spacetime (M, g) is said to be singular if it is not
geodesically complete.
 5. HAWKING’S SINGULARITY THEOREM 79

Theorem 5.2. (Hawking [Haw66]) Let (M, g) be a globally hyperbolic

spacetime satisfying the strong energy condition, and suppose that the ex-
pansion of the congruence of future-pointing timelike geodesics orthogonal to
S satisfies θ ≤ θ0 < 0 on a Cauchy hypersurface S. Then (M, g) is singular.
Proof. We will show that no future-directed timelike geodesic orthog-
onal to S can be extended to proper time greater than τ0 = − θ30 to the
future of S. Suppose that this was not so. Then there would exist a future-
directed timelike geodesic c orthogonal to S, parameterized by proper time,
defined in an interval [0, τ0 + ε] for some ε > 0. Let p = c(τ0 + ε). Accord-
ing to Theorem 4.3, there would exist a timelike geodesic γ with maximal
length connecting S to p, orthogonal to S. Because τ (c) = τ0 + ε, we would
necessarily have τ (γ) ≥ τ0 + ε. Theorem 3.5 guarantees that γ would de-
velop a conjugate point at a distance of at most τ0 to the future of S, and
Theorem 3.6 states that γ would cease to be maximizing beyond this point.
Therefore we arrive at a contradiction. 
Remark 5.3. It should be clear that (M, g) is singular if the condition
θ ≤ θ0 < 0 on a Cauchy hypersurface S is replaced by the condition θ ≥
θ0 > 0 on S. In this case, no past-directed timelike geodesic orthogonal
to S can be extended to proper time greater than τ0 = θ30 to the past of S.
Example 5.4.
(1) The FLRW models with α > 0 and Λ = 0 are globally hyperbolic,
and satisfy the strong energy condition (as ρ > 0). Moreover, the
∂
expansion of the congruence tangent to ∂t is θ = 3aȧ . Assume that
the model is expanding at time t0 . Then θ = θ0 = 3a(t ȧ(t0 )
0)
> 0 on the
Cauchy hypersurface S = {t = t0 }, and hence Theorem 5.2 guar-
antees that this model is singular to the past of S (i.e. there exists
a big bang). Moreover, Theorem 5.2 implies that this singularity is
generic: any sufficiently small perturbation of an expanding FLRW
model satisfying the strong energy condition will also be singular.
Loosely speaking, any expanding universe must have begun at a
big bang.
(2) The region {r < 2m} of the Schwarzschild solution is globally hy-
perbolic, and satisfies the strong energy condition (as Rµν = 0).
The metric can be written in this region as
 
2 2m
− 1 dt2 + r2 dθ2 + sin2 θdϕ2 ,


g = −dτ +
r
where
2m  − 1
2m
Z 2
τ= −1 du.
r u
Therefore the inside of the black hole can be pictured as a cylinder
R × S 2 whose shape is evolving in time. As r → 0, the S 2 contracts
80 4. SINGULARITY THEOREMS

to a singularity, with the t-direction expanding. Since

dr  m 2 
K= − 2 dt + rdθ2 + r sin2 θdϕ2 ,
dτ r
we have
 − 1  
2m 2 2 3m
θ= −1 − 2 .
r r r
Therefore we have θ = θ0 < 0 on any Cauchy hypersurface S =
{r = r0 } with r0 < 3m 2 , and hence Theorem 5.2 guarantees that
the Schwarzschild solution is singular to the future of S. More-
over, Theorem 5.2 implies that this singularity is generic: any suf-
ficiently small perturbation of the Schwarzschild solution satisfying
the strong energy condition will also be singular. Loosely speaking,
once the collapse has advanced long enough, nothing can prevent
the formation of a singularity.
(3) It should be noted that Theorem 5.2 proves geodesic incomplete-
ness, not the existence of curvature singularities. For instance, it
applies to the Milne universe, or a globally hyperbolic region of the
anti-de Sitter universe, whose curvature is bounded (they are sim-
ply globally hyperbolic regions in larger, inextendible Lorentzian
manifolds).

6. Penrose’s singularity theorem

Let (M, g) be a globally hyperbolic spacetime, S a Cauchy hypersurface
with future-pointing unit normal vector field N , and Σ ⊂ S a compact 2-
dimensional submanifold with unit normal vector field n in S. Let cp be the
null geodesic with initial condition Np + np for each point p ∈ Σ. We define
a smooth map exp : (−ε, ε) × Σ → M for some ε > 0 as exp(r, p) = cp (r).

Definition 6.1. The critical values of exp are said to be conjugate

points to Σ.

Loosely speaking, conjugate points are points where geodesics starting

orthogonally at nearby points of Σ intersect.
Let q = exp(r0 , p) be a point not conjugate to Σ. If ϕ is a local parame-
terization of Σ around p, then we can construct a system of local coordinates
(u, r, x2 , x3 ) on some open set V ∋ q by using the map

(u, r, x2 , x3 ) 7→ exp(r, ψu (ϕ(x2 , x3 ))),

where ψu is the flow along the timelike geodesics orthogonal to S and the
map exp : (−ε, ε) × ψu (Σ) → M is defined as above.
 6. PENROSE’S SINGULARITY THEOREM 81

∂ ∂ ∂
Since ∂r is tangent to null geodesics, we have grr = ∂r , ∂r = 0. On
the other hand, we have
   
∂grµ ∂ ∂ ∂ ∂ ∂
= , µ = ,∇ ∂
∂r ∂r ∂r ∂x ∂r ∂r ∂xµ
   
∂ ∂ 1 ∂ ∂ ∂
= , ∇ ∂µ = , = 0,
∂r ∂x ∂r 2 ∂xµ ∂r ∂r
for µ = 0, 1, 2, 3. Since gru = −1 and gr2 = gr3 = 0 on ψu (Σ), we have
gru = −1 and gr2 = gr3 = 0 on V . Therefore the metric is written in this
coordinate system as
g = αdu2 − 2dudr + 2βA dudxA + γAB dxA dxB .
Since  
α −1 β2 β3  
−1 0 0 0  γ 22 γ 23
 β2 0 γ22 γ23  = − det γ32 γ33 ,
det  

β3 0 γ32 γ33
we see that the functions
 
∂ ∂
γAB = A
, B
∂x ∂x
form a positive definite matrix, and so g induces a Riemannian metric on
the 2-dimensional surfaces exp(r, ψu (Σ)), which are then spacelike. Since the
vector fields ∂x∂A can always be defined along cp , the matrix (γAB ) is also
well defined along cp , even at points where the coordinate system breaks
down, i.e. at points which are conjugate to Σ. These  ∂ ∂are ∂the ∂points for
which γ = det (γAB ) vanishes, since only then will ∂u , ∂r , ∂x2 , ∂x3 fail to
∂
be linearly independent. In fact the vector fields ∂xA are Jacobi fields along
cp .
It is easy to see that
Γuur = Γurr = ΓurA = Γrrr = ΓA
rr = 0 and ΓA
rB = γ
AC
βCB ,
1 ∂γAB
where (γ AB ) = (γAB )−1 and βAB = 2 ∂r .
Consequently,
∂ΓA
 
u A Ar B A
Rrr = Rurr + RArr = − − ΓAr ΓrB
∂r
∂
γ AB βAB − γ BC γ AD βCA βDB .


=−
∂r
The quantity
θ = γ AB βAB
appearing in this expression is called the expansion of the null geodesics,
and has an important geometric meaning:
 
1 ∂ 1 ∂ ∂ 1
θ = tr (γAB )−1 (γAB ) = log (det (γAB )) = log (det (γAB )) 2 .
2 ∂r 2 ∂r ∂r
82 4. SINGULARITY THEOREMS

Therefore the expansion yields the variation of the area element of the
spacelike 2-dimensional surfaces exp(r, ψu (Σ)). More importantly for our
purposes, we see that a singularity of the expansion indicates a zero of
det (γAB ), i.e. a conjugate point to ψu (Σ).
Proposition 6.2. Let (M, g) be a globally hyperbolic spacetime satisfy-
ing the null energy condition, S ⊂ M a Cauchy hypersurface with future-
pointing unit normal vector field N , Σ ⊂ S a compact 2-dimensional sub-
manifold with unit normal vector field n in S and p ∈ Σ a point where
θ = θ0 < 0. Then the null geodesic cp with initial condition Np + np con-
tains at least a point conjugate to Σ, at an affine parameter distance of at
most − θ20 to the future of Σ (assuming that it can be extended that far).
Proof. Since (M, g) satisfies the null energy condition, we have Rrr =
∂ µ ∂ ν



Rµν ∂r ∂r ≥ 0. Consequently,
∂θ
+ γ BC γ AD βCA βDB ≤ 0.
∂r
Choosing an orthonormal basis (where γ AB = δAB ), and using the inequality
(tr A)2 ≤ n tr(At A)
for square n × n matrices, it is easy to show that

1
γ BC γ AD βCA βDB = βBA βAB = tr (βAB )(βAB )t ≥ θ2 .
2
Consequently θ must satisfy
∂θ 1 2
+ θ ≤ 0.
∂r 2
Integrating this inequality yields
1 1 r
≥ + ,
θ θ0 2
and hence θ must blow up at a value of r no greater than − θ20 . 
We define the chronological future and the causal future of the
compact surface Σ as
[ [
I + (Σ) = I + (p) and J + (Σ) = J + (p)
p∈Σ p∈Σ

(with similar definitions for the chronological past and the causal past
of Σ). It is clear that I + (Σ), being the union of open sets, is itself open,
and also that J + (Σ) ⊂ I + (Σ) and I + (Σ) = int J + (Σ). On the other hand,
it is easy to generalize Proposition 4.1 (and consequently Corollary 4.2) to
the corresponding statements with compact surfaces replacing points. In
particular, J + (Σ) is closed. Therefore
∂J + (Σ) = ∂I + (Σ) = J + (Σ) \ I + (Σ),
 6. PENROSE’S SINGULARITY THEOREM 83

and so, by a straightforward generalization of Corollary 1.5 in Chapter 3,

every point in this boundary can be reached from a point in Σ by a future-
directed null geodesic. Moreover, this geodesic must be orthogonal to Σ.
Indeed, at Σ we have
∂ ∂
= N and = N + n,
∂u ∂r
and so the metric takes the form
g = −du2 − 2dudr + γAB dxA dxB .
If c : I ⊂ R → M is a future-directed null geodesic with c(0) ∈ Σ, its initial
tangent vector
∂ ∂ ∂ ∂
ċ(0) = u̇ + ṙ + ẋA A = (u̇ + ṙ)N + ṙn + ẋA A
∂u ∂r ∂x ∂x
must satisfy
u̇(u̇ + 2ṙ) = γAB ẋA ẋB .
Since c is future-directed we must have u̇ + ṙ > 0. On the other hand, by
choosing the unit normal to Σ on S to be either n or −n, we can assume
ṙ ≥ 0. If c is not orthogonal to Σ we then have
γAB ẋA ẋB > 0 ⇒ u̇(u̇ + 2ṙ) > 0 ⇒ u̇ > 0.
Now the region where u > 0 and r ≥ 0 is clearly a subset of I + (Σ), since its
points can be reached from Σ by a sectionally smooth curve composed of an
arc of timelike geodesic and an arc of null geodesic. Therefore, we see that
if c is not orthogonal to Σ then c(t) ∈ I + (Σ) for all t > 0.
Even future-directed null geodesics orthogonal to Σ may eventually enter
+
I (Σ). A sufficient condition for this to happen is given in the following
result.
Proposition 6.3. Let (M, g) be a globally hyperbolic spacetime, S a
Cauchy hypersurface with future-pointing unit normal vector field N , Σ ⊂ S
a compact 2-dimensional submanifold with unit normal vector field n in S,
p ∈ Σ, cp the null geodesic through p with initial condition Np + np and
q = cp (r) for some r > 0. If cp has a conjugate point between p and q then
q ∈ I + (Σ).
Proof. We will offer only a sketch of the proof. Let s be the first
conjugate point along cp between p and q. Since q is conjugate to p, there
exists another null geodesic γ starting at Σ which (approximately) intersects
cp at s (Figure 6). The piecewise smooth null curve obtained by following
γ between Σ and s, and cp between s and q is a causal curve but not a null
geodesic. This curve can be easily smoothed while remaining causal and
nongeodesic, and so by the generalization of Corollary 1.5 in Chapter 3 we
have q ∈ I + (Σ). 
84 4. SINGULARITY THEOREMS

q
γ

cp

p S

Figure 6. Proof of Proposition 6.3.

Definition 6.4. Let (M, g) be a globally hyperbolic spacetime and S

a Cauchy hypersurface with future-pointing unit normal vector field N . A
compact 2-dimensional submanifold Σ ⊂ S with unit normal vector field n
in S is said to be trapped if the expansions θ+ and θ− of the null geodesics
with initial conditions N + n and N − n are both negative everywhere on Σ.
We have now all the necessary ingredients to prove the Penrose singu-
larity theorem.
Theorem 6.5. (Penrose [Pen65]) Let (M, g) be a connected globally
hyperbolic spacetime with a noncompact Cauchy hypersurface S, satisfying
the null energy condition. If S contains a trapped surface Σ then (M, g) is
singular.
Proof. Let t : M → R be a global time function such that S = t−1 (0).
The integral curves of grad t, being timelike, intersect S exactly once, and
∂I + (Σ) at most once. This defines a continuous injective map π : ∂I + (Σ) →
S, whose image is open. Indeed, if q = π(p), then all points is some neigh-
borhood of q are images of points in ∂I + (Σ), as otherwise there would be a
sequence qk ∈ S with qk → q such that the integral curves of grad t through
qk would not intersect ∂I + (Σ). Letting rk be the intersections of these curves
with the Cauchy hypersurface t−1 (t(r)), for some point r to the future of p
along the integral line of grad t, we would have rk → r, and so rk ∈ I + (Σ)
for sufficiently large k (as I + (Σ) is open), leading to a contradiction.
 6. PENROSE’S SINGULARITY THEOREM 85

Since Σ is trapped (and compact), there exists θ0 < 0 such that the
expansions θ+ and θ− of the null geodesics orthogonal to Σ both satisfy
θ+ , θ− ≤ θ0 . We will show that there exists a future-directed null geodesic
orthogonal to Σ which cannot be extended to an affine parameter greater
than r0 = − θ20 to the future of Σ. Suppose that this was not so. Then,
according to Proposition 6.2, any null geodesic orthogonal to Σ would have
a conjugate point at an affine parameter distance of at most r0 to the future
of Σ, after which it would be in I + (Σ), by Proposition 6.3. Consequently,
∂I + (Σ) would be a (closed) subset of the compact set
exp+ ([0, r0 ] × Σ) ∪ exp− ([0, r0 ] × Σ)
(where exp+ and exp− refer to the exponential map constructed using the
unit normals n and −n), hence compact. Therefore the image of π would
also be compact, hence closed as well as open. Since M , and therefore S, are
connected, the image of π would be S, which would then be homeomorphic to
∂I + (Σ). But S is noncompact by hypothesis, and we reach a contradiction.

Remark 6.6. It should be clear that (M, g) is singular if the condition
of existence of a trapped surface is replaced by the condition of existence of
an anti-trapped surface, that is, a compact surface Σ ⊂ S such that the
expansions of null geodesics orthogonal to Σ are both positive. In this case,
there exists a past-directed null geodesic orthogonal to Σ which cannot be
extended to an affine parameter time greater than r0 = θ20 to the past of Σ.
Example 6.7.
(1) The region {r < 2m} of the Schwarzschild solution is globally hy-
perbolic, and satisfies the null energy condition (as Rµν = 0). Since
r (or −r) is clearly a time function (depending on the choice of
time orientation), it must increase (or decrease) along any future-
pointing null geodesic, and therefore any sphere Σ of constant (t, r)
is anti-trapped (or trapped). Since any Cauchy hypersurface is dif-
feomorphic to R × S 2 , hence noncompact, we conclude from Theo-
rem 6.5 that the Schwarzschild solution is singular to past (or fu-
ture) of Σ. Moreover, Theorem 6.5 implies that this singularity is
generic: any sufficiently small perturbation of the Schwarzschild
solution satisfying the null energy condition will also be singu-
lar. Loosely speaking, once the collapse has advanced long enough,
nothing can prevent the formation of a singularity.
(2) The FLRW models with α > 0 and Λ = 0 are globally hyper-
bolic, and satisfy the null energy condition. Moreover, radial null
geodesics satisfy
dr 1p
=± 1 − kr2 .
dt a
Therefore, if we start with a sphere Σ of constant (t, r) and follow
the orthogonal null geodesics along the direction of increasing or
86 4. SINGULARITY THEOREMS

decreasing r, we obtain spheres whose radii ar satisfy

d p
(ar) = ȧr + aṙ = ȧr ± 1 − kr2 .
dt
Assume that the model is expanding, with the big bang at t = 0,
and spatially noncompact (in particular k 6= 1). Then, for suffi-
ciently small t > 0, the sphere Σ is anti-trapped, and hence The-
orem 6.5 guarantees that this model is singular to the past of Σ
(i.e. there exists a big bang). Moreover, Theorem 6.5 implies that
this singularity is generic: any sufficiently small perturbation of
an expanding, spatially noncompact FLRW model satisfying the
null energy condition will also be singular. Loosely speaking, any
expanding universe must have begun at a big bang.

7. Exercises
(1) Let g be a Lorentzian metric given in the Gauss Lemma form,
g = −dt2 + hij dxi dxj ,
∂
and consider the geodesic congruence tangent to ∂t .
(a) Show that
1 ∂hij
Bij = Γ0ij = ,
2 ∂t
that is,
1
B = L ∂ g = K,
2 ∂t
where K is the second fundamental form of the hypersurfaces
of constant t.
(b) Conclude that the expansion of the congruence of the galaxies
in a FLRW model is
3ȧ
θ= = 3H.
a
(2) Let (M, g) be a Lorentzian manifold.
(a) Use the formula for the Lie derivative of a tensor,
(LX g)(Y, Z) = X · g(Y, Z) − g(LX Y, Z) − g(Y, LX Z)
to show that
(LX g)(Y, Z) = g(∇Y X, Z) + g(Y, ∇Z X).
(b) Show that this formula can be written as
(LX g)µν = ∇µ Xν + ∇ν Xµ .
(c) Suppose that X is a Killing vector field, i.e. LX g = 0. Use the
Killing equation
∇µ Xν + ∇ν Xµ = 0
 7. EXERCISES 87

to show that X is a solution of the Jacobi equation. Give a

geometric interpretation of this fact.
(3) Let T be a diagonalizable energy-momentum tensor, that is, (Tµν ) =
diag(ρ, p1 , p2 , p3 ) on some orthonormal frame {E0 , E1 , E2 , E3 }. Show
that:
(a) T satisfies the SEC if and only if ρ+ 3i=1 pi ≥ 0 and ρ+pi ≥ 0
P
(i = 1, 2, 3).
(b) T satisfies the WEC if and only if ρ ≥ 0 and ρ + pi ≥ 0
(i = 1, 2, 3).
(c) T satisfies the DEC if and only if ρ ≥ |pi | (i = 1, 2, 3).
(d) T satisfies the NEC if and only if ρ + pi ≥ 0 (i = 1, 2, 3).
(e) The first three conditions are independent except that the
DEC implies the WEC.
(f) The first three conditions imply the NEC.
(4) Let (M, g) be the globally hyperbolic Lorentzian manifold corre-
sponding to the  exterior region
 of the Schwarzschild solution, that
is, M = R × R3 \ B2m (0) and

2m −1 2
   
2m
dt2 + 1 − dr + r2 dθ2 + sin2 θdϕ2


g =− 1−
r r
(with m > 0).
(a) Show that for any r0 > 2m the curve
 r 
π m
c(t) = t, r0 , , t
2 r0 3
is a timelike, null or spacelike geodesic, according to whether
r0 > 3m, r0 = 3m or r0 < 3m.  q 
r0 3 π
(b) Argue that the point q = π m , r0 , 2 , π is conjugate to
the point p = 0, r0 , π2 , 0 along c (note that this can be done

without solving the Jacobi equation).

(c) Show explicitly
q that if r0 > 3m then c stops being maximizing
3
for t > π rm 0
.
(5) Let (M, g) be a globally hyperbolic spacetime and p, q ∈ M with
q ∈ I + (p). Show that among all timelike curves connecting p to
q there exists a timelike curve with maximal length, which is a
timelike geodesic.
(6) Use ideas similar to those leading to the proof Hawking’s singularity
theorem to prove Myers’s theorem: if (M, h·, ·i) is a complete
Riemannian manifold whose Ricci curvature satisfies Rµν X µ X ν ≥
εgµν X µ X ν for some ε > 0 then M is compact. Can these ideas be
used to prove a singularity theorem in Riemannian geometry?
(7) Explain why Hawking’s singularity theorem does not apply to each
of the following geodesically complete Lorentzian manifolds:
88 4. SINGULARITY THEOREMS

(a) Minkowski’s spacetime;

(b) Einstein’s universe;
(c) de Sitter’s universe;
(d) Anti-de Sitter spacetime.
(8) Consider the metric
ds2 = α du2 − 2 du dr + 2βA du dxA + γAB dxA dxB .
(a) Show that the Christoffel symbols satisfy
Γuur = Γurr = ΓurA = Γrrr = ΓA
rr = 0 and ΓA
rB = γ
AC
βCB ,
where (γ AB ) = (γAB )−1 and βAB = 12 ∂γ∂r AB
.
(b) Conclude that
∂
γ AB βAB − γ AB γ CD βAC βBD .


Rrr = −
∂r
(9) Let (M, g) be a globally hyperbolic spacetime with Cauchy hyper-
surfaces S0 and S1 satisfying S1 ⊂ D+ (S0 ), and Σ ⊂ S1 a compact
surface. Show that:
(a) D+ (S0 ) ∩ J − (Σ) is compact;
(b) J − (Σ) is closed.
(10) Explain why Penrose’s singularity theorem does not apply to each
of the following geodesically complete Lorentzian manifolds:
(a) Minkowski’s spacetime;
(b) Einstein’s universe;
(c) de Sitter’s universe;
(d) Anti-de Sitter spacetime.
 CHAPTER 5

Cauchy problem

A fundamental fact about the Einstein equations is that they are hy-
perbolic in nature. In this chapter we discuss the Cauchy problem for the
Einstein field equations, following [Wal84]. We start by studying the Klein-
Gordon equation, as a prototypical wave equation, and the Maxwell equa-
tions, where the issues of constraints on the inital data and gauge freedom
also arise. We then sketch the proof of the Choquet-Bruhat theorem, and
discuss the Lichnerowicz method for solving the constraint equations. See
[Rin09] for a more complete discussion.

1. Divergence theorem
It is possible to define the divergence of a vector field on any orientable
manifold where a volume form has been chosen.
Definition 1.1. Let M be an orientable n-dimensional manifold with
volume form ǫ, and let X be a vector field on M . The divergence of X is
the function div X such that
d(Xy ǫ) = (div X)ǫ.
The following result is a straightforward application of the Stokes theo-
rem.
Theorem 1.2. (Divergence theorem) If M is a compact orientable
n-dimensional manifold with boundary ∂M then
Z Z
(div X)ǫ = Xy ǫ.
M ∂M

Proposition 1.3. If (M, g) is a pseudo-Riemannian manifold with Levi-

Civita connection ∇ then
div X = ∇µ X µ .
Proof. Take local coordinates such that X = ∂1 around each point
where X does not vanish. Since
q
ǫ = | det(gµν )| dx1 ∧ . . . ∧ dxn ,
we have
1
d(Xy ǫ) = LX ǫ − Xy dǫ = ∂1 log |det(gµν )| 2 ǫ,
89
90 5. CAUCHY PROBLEM

where we used dǫ = 0. Since

∂1 log |det A| = tr(A−1 ∂1 A)
for any matrix-valued function A in M , we have
1 1
d(Xy ǫ) = (g µν ∂1 gµν ) ǫ = (g µν (LX g)µν ) ǫ
2 2
1 µν
= g (∇µ Xν + ∇ν Xµ )ǫ = (∇µ X µ )ǫ.
2
This formula can be easily extended to the set of zeros of X: by continuity
on the boundary, and trivially in the interior. 
Assume now that (M, g) is a compact orientable n-dimensional Lorentzian
manifold with boundary ∂M , and let {E1 , . . . , En } be a positive orthonor-
mal frame with E1 = N the outward unit normal vector on ∂M . The volume
element is
ǫ = −E1♭ ∧ . . . ∧ En♭ ,
so that
ǫ(E1 , . . . , En ) = 1,
and the volume element of ∂M with the induced orientation is
σ = E1 y ǫ = ±E2♭ ∧ . . . ∧ En♭
(according to whether N is timelike or spacelike). Therefore we have on ∂M
Xy ǫ = ∓hX, N iσ,
implying that the divergence theorem can be written as
Z Z
(div X)ǫ = hX, niσ,
M ∂M
where n is the outward unit normal vector in the points where it is space-
like, and is the inward unit normal vector in the points where it is timelike
(Figure 1).
It may happen that ∂M has points where the normal is null, hence
tangent to ∂M . In that case we choose the orthonormal frame such that
N = E1 + E2
is a null normal, with E1 timelike and pointing outwards, and E2 spacelike
and (necessarily) pointing inwards. Then
ǫ = −N ♭ ∧ E2♭ ∧ . . . ∧ En♭
and so
Xy ǫ = −hX, N iσ,
where
σ = E2♭ ∧ . . . ∧ En♭
is a volume element on ∂M (compatible with the induced orientation, as
E1 points outwards). So we must choose in this case n = −N , that is, we
must use the normal whose timelike component points inwards and whose
 2. KLEIN-GORDON EQUATION 91

n timelike

n null

n spacelike

Figure 1. Normal vector for the divergence theorem on a

Lorentzian manifold.

spacelike component points outwards (Figure 1). Note that the magnitude
(but not the sign) of the volume element σ will depend on the choice of n.

2. Klein-Gordon equation
Let (M, g) be a Lorentzian manifold. A smooth function φ : M → R is
a solution of the Klein-Gordon equation if it satisfies
φ − m2 φ = 0 ⇔ ∇µ ∂µ φ − m2 φ = 0
for some constant m > 0. For reasons that will become clear in Chapter 6,
we define the energy-momentum tensor associated to this equation as
1
Tµν = ∂µ φ ∂ν φ − gµν ∂α φ ∂ α φ + m2 φ2 .


2
If φ is a solution of the Klein-Gordon equation then
∇µ Tµν = φ ∂ν φ + ∂µ φ ∇µ ∂ν φ − ∂α φ ∇ν ∂ α φ − m2 φ ∂ν φ
= (φ − m2 φ)∂ν φ = 0
Moreover, if (M, g) is time-oriented then it is possible to prove that T sat-
isfies the dominant energy condition, that is, Tµν X ν corresponds to a past-
pointing causal vector whenever X is a future-pointing causal vector. As-
sume that X is a future-pointing timelike Killing vector field, and define
Y µ = T µν Xν . Then Y is a past-pointing causal vector field satisfying
∇µ Y µ = T µν ∇µ Xν = 0
(where we used the Killing equation ∇(µ Xν) = 0 and the symmetry of T ).
92 5. CAUCHY PROBLEM

Let us now focus on the case when (M, g) is flat Minkowski spacetime
∂
and X = ∂t . Consider the Cauchy hypersurface S0 = {t = t0 }, and let B0
be the ball of radius R in that hypersurface:
B0 = {t = t0 , x2 + y 2 + z 2 ≤ R2 }.
Let S1 = {t = t1 }, with t1 > t0 , be another Cauchy hypersurface, and
consider the ball B1 = D(B0 ) ∩ S1 in that hypersurface (Figure 2). By the
divergence theorem we have
Z Z Z
hY, Xi + hY, ni + hY, −Xi = 0,
B0 C B1

where C is the null portion of ∂D(B0 ) between S0 and S1 and n is a past-

pointing normal. Since Y is a past-pointing causal vector we have hY, ni ≤ 0,
and so
Z Z
(13) hY, Xi ≤ hY, Xi.
B1 B0

B1

n
C

B0

S0

Figure 2. Proof of Proposition 2.1.

Note that
1
hY, Xi = Tµν X µ X ν = (X · φ)2 + ∂α φ ∂ α φ + m2 φ2


2
1
(∂0 φ)2 + (∂x φ)2 + (∂y φ)2 + (∂z φ)2 + m2 φ2 .

=
2
We conclude immediately that if φ is a solution of the Klein-Gordon equation
and φ = ∂0 φ = 0 in B0 then φ = 0 in B1 , and indeed in D(B0 ) (since t1 is
arbitrary). Because B0 can be taken arbitrarily large and the Klein-Gordon
equation is linear we can then deduce the following result.
Proposition 2.1. Given two smooth functions φ0 , ψ0 : S0 → R, there
exists at most a solution φ of the Klein-Gordon equation satisfying φ = φ0
and ∂0 φ = ψ0 on S0 .
 2. KLEIN-GORDON EQUATION 93

Given a smooth function φ : M → R and some set A ⊂ M we define the

Sobolev norm
Z
kφk2H 1 (A) =
 2
φ + (∂0 φ)2 + (∂x φ)2 + (∂y φ)2 + (∂z φ)2 .

A
In this definition A can either be an open set or a submanifold (in which
case we use the induced volume form in the integral).
More generally, for each k ∈ N0 we define the Sobolev norms
Z X
2
kφkH k (A) = (∂ α φ)2 ,
A |α|≤k

where α = (α0 , α1 , α2 , α3 ) ∈ N0 4 , |α| = α0 + α1 + α2 + α3 , ∂ = (∂0 , ∂x , ∂y , ∂z )

and ∂ α = ∂0α0 ∂xα1 ∂yα2 ∂zα3 . Similarly, given a smooth function φ0 : S0 → R we
define Z X
2
kφ0 kH k (B0 ) = (Dα φ0 )2 ,
B0 |α|≤k

where now D = (∂x , ∂y , ∂z ). Inequality (13) can then be written as

kφk2H 1 (B1 ) ≤ Ckφk2H 1 (B0 ) ≤ Ckφ0 k2H 1 (B0 ) + Ckψ0 k2H 0 (B0 ) ,
where C > 0 is a generic positive constant (not always the same). Integrating
this inequality in t from t0 − R to t0 + R we obtain
kφk2H 1 (D(B0 )) ≤ Ckφ0 k2H 1 (B0 ) + Ckψ0 k2H 0 (B0 ) .
On S0 , all spatial partial derivatives of φ and ∂0 φ are given by partial deriva-
tives of φ0 and ψ0 . On the other hand, from the Klein-Gordon equation we
have
∂02 φ|S0 = ∂x2 φ0 + ∂y2 φ0 + ∂z2 φ0 − m2 φ0 ,
and so, by partial differentiation, we can obtain all spatial partial derivatives
of ∂02 φ on S0 from partial derivatives of φ0 . Differentiating the Klein-Gordon
equation with respect to t yields
∂03 φ|S0 = ∂x2 ψ0 + ∂y2 ψ0 + ∂z2 ψ0 − m2 ψ0 ,
and it should be clear that all partial derivatives of φ on S0 are given by
partial derivatives of φ0 and ψ0 . Note from the general partial derivative of
the Klein-Gordon equation,
∂02 ∂ α φ = ∂x2 ∂ α φ + ∂y2 ∂ α φ + ∂z2 ∂ α φ − m2 ∂ α φ,
that ∂ α φ also satisfies the Klein-Gordon equation, and so
k∂ α φk2H 1 (B1 ) ≤ Ck∂ α φk2H 1 (B0 ) ⇒
kφk2H k (B1 ) ≤ Ckφ0 k2H k (B0 ) + Ckψ0 k2H k−1 (B0 ) ⇒
kφk2H k (D(B0 )) ≤ Ckφ0 k2H k (B0 ) + Ckψ0 k2H k−1 (B0 ) ,
whence
kφkH k (D(B0 )) ≤ Ckφ0 kH k (B0 ) + Ckψ0 kH k−1 (B0 ) .
94 5. CAUCHY PROBLEM

Another important norm is the supremum norm,

kφkC 0 (A) = sup |φ(p)|.
p∈A
More generally, we define the norms
X
kφkC k (A) = sup |∂ α φ(p)|.
p∈A
|α|≤k

Definition 2.2. A set A ⊂ Rn is said to satisfy the interior cone

condition if there exists a (closed) cone of height h > 0 and solid angle
Ω > 0 at the vertex such that for each point p ∈ A it is possible to map the
cone isometrically into A in such a way that the vertex is mapped to p.
Theorem 2.3. (Sobolev inequality) If A ⊂ Rn satisfies the interior
cone condition and k > n2 then for any smooth function f : A → R
kf kC 0 (A) ≤ Ckf kH k (A) .
Proof. See [AF03], or Exercise 2. 
For a solution of the Klein-Gordon equation we then have
kφkC 0 (D(B0 )) ≤ kφkH 3 (D(B0 )) ≤ Ckφ0 kH 3 (B0 ) + Ckψ0 kH 2 (B0 ) .
More generally,
k∂ α φkC 0 (D(B0 )) ≤ k∂ α φkH 3 (D(B0 )) ≤ Ckφ0 kH m+3 (B0 ) + Ckψ0 kH m+2 (B0 ) ,
where m = |α|. We conclude that
kφkC m (D(B0 )) ≤ Ckφ0 kH m+3 (B0 ) + Ckψ0 kH m+2 (B0 ) ,
whence
kφkC m (D(B0 )) ≤ Ckφ0 kC m+3 (B0 ) + Ckψ0 kC m+2 (B0 ) .
Theorem 2.4. Given smooth initial data φ0 , ψ0 : S0 → R for the Klein-
Gordon equation, there exists a unique smooth solution φ satisfying φ = φ0
and ∂0 φ = ψ0 on S0 . Moreover, if B0 ⊂ S0 is a ball then the solution in
D(B0 ) depends only on the initial data in B0 , and the map
C m+3 (B0 ) × C m+2 (B0 ) ∋ (φ0 , ψ0 ) 7→ φ ∈ C m (D(B0 ))
is continuous.
Proof. We just have to prove existence of solution. To do that, we
note that if φ is a solution then we know all its partial derivatives on S0 .
Therefore, we can construct a power series for φ around each point of S0 . If
φ0 and ψ0 are analytic then the Cauchy-Kowalewski theorem guarantees
that these series converge, and so there exists an analytic solution φ, which
can be shown to be globally defined. If φ0 and ψ0 are smooth then there exist
sequences φ0,n and ψ0,n of analytic functions which converge to φ0 and ψ0
in all spaces C m (B0 ). The corresponding analytic solutions φn of the Klein-
Gordon equation thus form a Cauchy sequence in all spaces C m (D(B0 )), and
hence must converge in all these spaces to some function φ. This function is
 2. KLEIN-GORDON EQUATION 95

therefore smooth, and, passing to the limit, must satisfy the Klein-Gordon
equation in all sets of the form D(B0 ). 
Using similar tecnhiques, it is possible to prove a much stronger result
(see for instance [Rin09]).
Theorem 2.5. Let (M, g) be a globally hyperbolic spacetime with a Cauchy
hypersurface S and future unit normal N . Then the linear, diagonal sec-
ond order hyperbolic system
g µν ∇µ ∂ν φi + Aµij ∂µ φj + Bij φj + Ci = 0 (i = 1, . . . , n),
where Aµij ,Bij and Ci are smooth and ∇ is any connection, yields a well-
posed Cauchy problem with initial data in S. More precisely, given smooth
initial data (φ1 , . . . , φn , N · φ1 , . . . , N · φn ) on S there exists a unique smooth
solution of the system, defined in M . Moreover, the solutions depend con-
tinuously on the initial data, and if two initial data sets coincide on some
closed subset B ⊂ S then the corresponding solutions coincide on D(B).
To solve the Cauchy problem for the Einstein equations, we will need to
solve more complicated systems of hyperbolic equations.
Theorem 2.6. Consider the quasi-linear, diagonal second order
hyperbolic system
g µν (x, φ, ∂φ)∇µ ∂ν φi = Fi (x, φ, ∂φ) (i = 1, . . . , n),
where g µν and Fi are smooth and ∇ is any connection on some manifold
M . Let (φ0 )1 , . . . , (φ0 )n be a solution of this system, and define (g0 )µν =
g µν (x, φ0 , ∂φ0 ). Assume that (M, g0 ) is globally hyperbolic, and let S be a
Cauchy hypersurface. Then the system above yields a well-posed Cauchy
problem with initial data in S, in the following sense: given initial data
in S sufficiently close to the initial data for (φ0 )1 , . . . , (φ0 )n there exists
an open neighborhood V of S such that the system has a unique solution in
V , and (V, g(x, φ, ∂φ)) is globally hyperbolic. Moreover, the solutions depend
continuously on the initial data, and if two initial data sets coincide on some
closed subset B ⊂ S then the corresponding solutions coincide in D(B).
Proof. The idea of the proof is to start with the linear hyperbolic
system
g µν (x, φ0 , ∂φ0 )∇µ ∂ν φi = Fi (x, φ0 , ∂φ0 ) (i = 1, . . . , n),
which by the previous theorem has a unique solution φ1 close to φ0 . Because
of this, there exists a neighborhood V1 of S such that (V1 , g(x, φ1 , ∂φ1 )) is
globally hyperbolic with Cauchy hypersurface S, and so the system
g µν (x, φ1 , ∂φ1 )∇µ ∂ν φi = Fi (x, φ1 , ∂φ1 ) (i = 1, . . . , n),
again has a unique solution φ2 close to φ1 . Iterating this procedure we obtain
a sequence φn which can then be shown to converge to the unique solution
of the quasi-linear hyperbolic system (see [Rin09] for more details). 
96 5. CAUCHY PROBLEM

3. Maxwell’s equations: constraints and gauge

As a warm-up problem to solving the Einstein field equations we consider
the considerably easier problem of solving the Maxwell equations without
sources in flat Minkowski spacetime. These equations can be split into what
we shall call constraint equations,
(
div E = 0
,
div B = 0
and evolution equations:
∂E

 ∂t = curl B



.
∂B


= − curl E


∂t
One expects that the evolution equations completely determine E(t, x) and
B(t, x) from initial data
(
E(0, x) = E0 (x)
.
B(0, x) = B0 (x)
This initial data, however, is not completely free: it must satisfy the con-
straint equations
div E0 = div B0 = 0.
This suffices to guarantee that the constraint equations are satisfied by
E(t, x) and B(t, x), since they are preserved by the evolution: for instance,
 
∂ ∂E
(div E) = div = div(curl B) = 0.
∂t ∂t
As we will see, solving the Einstein field equations will also require splitting
them into constraint equations and evolution equations. Another issue that
will have to be dealt with, gauge freedom, also occurs when solving the
Maxwell equations by using the electromagnetic gauge potentials. To do
so, we note that two of the Maxwell equations (those which do not admit
sources) are equivalent to the existence of a vector potential A and a scalar
potential φ in terms of which B and E can be written:
 
div B = 0 B = curl A
∂B ⇔ ∂A .
curl E = − E = − grad φ −
∂t ∂t
These potentials, however, are nonunique: given any smooth function χ, the
potentials 
A′ = A + grad χ
∂χ
 φ′ = φ −
∂t
 3. MAXWELL’S EQUATIONS: CONSTRAINTS AND GAUGE 97

yield the same fields B and E (A′ and φ′ are said to be related to A and φ
by a gauge transformation). The remaining Maxwell equations can now
be written as
∂
 
div E = 0 ∆φ + (div A) = 0

∂E ⇔ ∂t 2 .
curl B = grad(div A) − ∆A = − grad ∂φ − ∂ A

∂t ∂t ∂t2
Therefore, if there exist gauge potentials satisfying
∂φ
div A = −
∂t
(the so-called Lorentz gauge) then these equations reduce to uncoupled
wave equations:
(
φ = 0
A = 0
To solve the Maxwell equations using the gauge potentials we then solve
these wave equations with initial data φ0 , ∂φ ∂A


∂t and A 0 , ∂t 0 satisfying:
0
(1) curl A0 = B0 (possible because div B0 = 0);
(2) φ0 =
0 (by choice);
(3) ∂A = −E0 (giving the correct initial electric field);
 ∂t 0
∂φ
(4) ∂t = − div A0 (so that the Lorentz gauge condition holds).
0
These potentials will determine a solution of the Maxwell equations with
the correct initial data if the Lorentz gauge condition holds for all
time. Now  
∂φ ∂
 div A + = div(A) + (φ) = 0
∂t ∂t
and by (4)
 
∂φ
div A + = 0.
∂t 0
Moreover,
∂2φ
     
∂ ∂φ ∂A ∂A
div A + = div + 2 = div + ∆φ,
∂t ∂t ∂t ∂t ∂t
and so   
∂ ∂φ
div A + = − div E0 = 0.
∂t ∂t 0
By uniqueness of solution of the wave equation, we conclude that the Lorentz
gauge condition does hold for all time, and so the potentials obtained by
solving the wave equation with the initial conditions above do determine the
solution of the Maxwell equations with initial data E0 and B0 .
98 5. CAUCHY PROBLEM

Remark 3.1. The electromagnetic potentials can be seen as the com-

ponents of the electromagnetic potential one-form
A = −φ dt + A1 dx + A2 dy + A3 dz.
Note that a gauge transformation can be written as
A′ = A + dχ.
The electric and magnetic fields can in turn be seen as the components of
the Faraday tensor
F = dA = E 1 dx ∧ dt + E 2 dy ∧ dt + E 3 dz ∧ dt
+ B 1 dy ∧ dz + B 2 dz ∧ dx + B 3 dx ∧ dy.
It should be obvious that F remains invariant under a gauge transformation.
The Maxwell equations can be written as
dF = 0 ⇔ F = dA
and
d ⋆ F = 0,
since
⋆F = E 1 dy ∧ dz + E 2 dz ∧ dx + E 3 dx ∧ dy
− B 1 dx ∧ dt − B 2 dy ∧ dt − B 3 dz ∧ dt
(that is, the Hodge star replaces B with E and E with −B).

4. Einstein’s equations
Let (M, g) be a globally hyperbolic spacetime and S ⊂ M a Cauchy
hypersurface. Let us write g in the Gauss Lemma form near S,
g = −dt2 + hij (t, x)dxi dxj ,
so that the level sets of t are Riemannian manifolds with induced metric
h(t) = hij dxi dxj and second fundamental form
1 ∂hij i j
K(t) = dx dx .
2 ∂t
For this choice of coordinates (gauge), finding the metric is equivalent to
finding a time-dependent Riemannian metric h(t) on S. The vacuum Ein-
stein field equations Gµν = 0 can be split into constraint equations
( (
2
G00 = 0 R̄ + K ii − Kij K ij = 0
⇔
G0i = 0 ¯ iK j − ∇
∇ ¯ jKj = 0
j i
and evolution equations
∂
Gij = 0 ⇔ Kij = −R̄ij + 2Kil K lj − K ll Kij ,
∂t
¯
where ∇, R̄ and R̄ij are the Levi-Civita connection, the scalar curvature
and the Ricci tensor of h. Note that the evolution equations allow us to
 4. EINSTEIN’S EQUATIONS 99

evolve h(t) and K(t), whereas the constraint equations restrict their initial
values h(0) and K(0). If the initial data satisfy the constraint equations, so
does the solution of the evolution equations. Indeed, the contracted Bianchi
identities give us “for free” the equations
∇α Gαβ = 0 ⇔ ∇0 G0β + ∇i Giβ = 0 ⇔ −∇0 G0β + hij ∇j Giβ = 0
⇔ ∂0 G0β − Γα00 Gαβ − Γα0β G0α = hij (∂j Giβ − Γαji Gαβ − Γαjβ Giα ).
If the evolution equations hold, we have Gij = ∂α Gij = 0, and so the
contracted Bianchi identities become
(
∂0 G00 = Γα00 Gα0 + Γα00 G0α + hij (∂j Gi0 − Γαji Gα0 − Γ0j0 Gi0 )
∂0 G0k = Γ000 G0k + Γα0k G0α − hij (Γ0ji G0k + Γ0jk Gi0 )
(
∂0 G00 = hij (∂j Gi0 − Kji G00 − Γ̄kji Gk0 )
⇔ .
∂0 G0k = K ik G0i − hij (Kji G0k + Kjk Gi0 )

This is a system of linear first order partial differential equations on G0β ;

integrating the last three equations, and then the first, it is easy to see that,
since the initial data vanishes at t = 0, the solution vanishes for all t.
Remark 4.1. In general, any time function t : M → R whose level
sets St are Cauchy hypersurfaces can be completed into a system of local
coordinates (t, x1 , x2 , x3 ). If
grad t
N =−
| grad t|
is the future-pointing unit normal to St , we have the orthogonal decompo-
sition
∂
= αN + β,
∂t
where the positive function α is known as the lapse function and the vector
field β, tangent to St , is known as the shift vector (Figure 3). In these
coordinates, the metric is written
g = (−α2 + βi β i )dt2 + 2βi dtdxi + hij dxi dxj
= −α2 dt2 + hij (dxi + β i dt)(dxj + β j dt).
Note that the Riemannian metric of the Cauchy hypersurfaces St is still
h(t) = hij dxi dxj ; the lapse function and the shift vector merely specify how
points with the same coodinates xi in different Cauchy hypersurfaces St are
related, which is a matter of choice, that is, a gauge freedom. The Gauss
Lemma form of the metric, for instance, corresponds α = 1 and β = 0, but
this is not necessarily the best choice.
To prove the existence and uniqueness result for the vacuum Einstein
field equations it is best to choose so-called harmonic coordinates, that
100 5. CAUCHY PROBLEM

∂
∂t
αN

N
β

St

Figure 3. Lapse function and shift vector.

is, coordinates xµ satisfying the wave equation, xµ = 0. This equation can
be written as
1
H µ ≡ ∂α g αµ + g αµ g ρσ ∂α gρσ = 0.
2
We define the reduced Ricci tensor to be
H 1
Rµν ≡ Rµν + gα(µ ∂ν) H α = − g αβ ∂α ∂β gµν + Fµν (g, ∂g),
2
and the reduced Einstein equations to be
H
Rµν = 0.
Note that Rµν H = R µ
µν if the coordinates x are harmonic; in this case, the
reduced Einstein equations coincide with the Einstein equations. Moreover,
the reduced Einstein equations are a quasi-linear, diagonal second order
hyperbolic system for the components of the metric. Given initial data
(hij , Kij ) satisfying the constraint equations, consider the following initial
data for the reduced Einstein equations:
(1) gij = hij (forced);
(2) gi0 = 0 (by choice);
(3) g00 = −1 (by choice);
∂g
(4) ∂tij = 2Kij (forced);
∂g
(5) ∂t0µ such that H µ = 0 in S.
If (hij , Kij ) is close to the trivial data (δij , 0) for the Minkowski spacetime,
Theorem 2.6 guarantees that we can solve the reduced Einstein equations
in some open neighborhood V of S. From
H 1 1
Gµν = Rµν − RH gµν − gα(µ ∂ν) H α + gµν ∂α H α
2 2
 4. EINSTEIN’S EQUATIONS 101

H =0
it is easily seen that if gµν satisfies the reduced Einstein equations Rµν
then the contracted Bianchi identities yield
∇µ Gµν = 0 ⇒ g µν ∂µ ∂ν H α + Aαµ β
β ∂µ H = 0.

Moreover, we have from the constraint equations

Gµ0 = 0 ⇒ ∂0 H µ = 0
on S. Therefore H µ is the solution of a linear, diagonal second order hyper-
bolic system with vanishing initial conditions. We conclude that H µ = 0 in
V , and therefore gµν solves the Einstein equations in V .
We can always assume that our initial data is close to the trivial data
by rescaling: if xµ are local coordinates such that gµν = ηµν at some point
p ∈ S, where ηµν is the Minkowski metric, we define new coordinates x̄µ by
the formula
1
x̄µ = xµ ,
λ
where λ > 0 is a constant. In these new coordinates the metric is
∂xα ∂xβ
ḡµν = gαβ = λ2 gµν .
∂ x̄µ ∂ x̄ν
If this metric is a solution of the vacuum Einstein field equations, then so is
1
g̃µν = ḡµν = gµν .
λ2
Note that this metric satisfies
∂g̃µν ∂gµν ∂t ∂gµν
= =λ .
∂ t̄ ∂t ∂ t̄ ∂t
∂g̃
Therefore for λ sufficiently small the initial data (g̃µν , ∂µν t̄ ) will be close to
(ηµν , 0).
In this way we can obtain a local solution of the Einstein field equations
in a neighborhood of each point p ∈ S. By uniqueness of solution of a
quasi-linear, diagonal second order hyperbolic system we can glue these local
solutions to obtain a global solution defined on an open neighborhood of S.
In other words, given initial data (h, K) satisfying the constraint equations,
there exists a globally hyperbolic spacetime (M, g) satisfying the Einstein
field equations such that S is a Cauchy hypersurface with induced metric h
and second fundamental form K.
Finally, now that we have proved the existence of such solutions, it is pos-
sible to prove the existence of a maximal solution. The proof is as follows:
if (M1 , g1 ) and (M2 , g2 ) are two solutions of the Einstein field equations con-
taining S with the same initial data (h, K), we say that (M1 , g1 ) ≤ (M2 , g2 )
if there is an isometric embedding ψ : M1 → M2 preserving (S, h, K). Note
that it is possible that neither (M1 , g1 ) ≤ (M2 , g2 ) nor (M2 , g2 ) ≤ (M1 , g1 ).
A set of solutions with the property that any two are related by ≤ is called
a chain; it is clear that every chain has an upper bound (the union up to
102 5. CAUCHY PROBLEM

isometric embeddings). Under these conditions, Zorn’s Lemma guaran-

tees that there is a maximal element (M, g) in the set of all solutions, that
is, a solution which cannot be isometric embedded into any other solution.
It is possible to prove that this element is unique (if there were two such
maximal solutions it would be possible to patch them together to construct
a larger solution). We then have the following fundamental result.
Theorem 4.2. (Choquet-Bruhat [FB55, CBG69]) Let (S, h) be a
3-dimensional Riemannian manifold and K a symmetric tensor field in S
satisfying the constraint equations
(
2
R̄ + K ii − Kij K ij = 0
,
¯ iK j − ∇
∇ ¯ jKj = 0
j i

where ∇ ¯ and R̄ are the Levi-Civita connection and the scalar curvature of
h. Then there exists a unique (up to isometry) 4-dimensional Lorentzian
manifold (M, g), called the maximal Cauchy development of (S, h, K),
satisfying:
(i) (M, g) is a solution of the vacuum Einstein equations;
(ii) (M, g) is globally hyperbolic with Cauchy hypersurface S;
(iii) The induced metric and second fundamental forms of S are h and K;
(iv) Any 4-dimensional Lorentzian manifold satisfying (i) − (iii) can be
isometrically embedded into (M, g).
Moreover, if (S, h, K) and (S̄, h̄, K̄) coincide on some closed subset B ∼
= B̄
then D(B) and D(B̄) are isometric. Finally, g depends continuously on the
initial data (h, K) (for appropriate topologies).
It should be stressed that the sketch of proof above glosses over many of
the technical difficulties, such as showing that the upper bound is a Haus-
dorff manifold, or that any solution can be isometrically embedded into the
maximal element; see [Rin09] for the full details. A recent proof of this
result which avoids Zorn’s Lemma can be found in [Sbi16].

5. Constraint equations
To obtain initial data for the Einstein equations it is necessary to solve
the nonlinear constraint equations
(
2
R̄ + K ii − Kij K ij = 0
.
¯ iK j − ∇
∇ ¯ jKj = 0
j i

The Lichnerowicz method for solving these equations is as follows: one

starts by choosing an arbitrary Riemannian metric h and an arbitrary sym-
metric tensor K satisfying
(
K ii = 0
¯ jKj = 0
∇ i
 5. CONSTRAINT EQUATIONS 103

(that is, K is traceless and divergenceless). These choices satisfy the second,
but not the first, constraint equations. On then defines the conformally
rescaled metric
h̃ = u4 h
and the rescaled symmetric tensor
K̃ = u−2 K.
Clearly K̃ is still traceless, and it is easily seen that it is also divergenceless:
˜ j K̃ j = 0,
∇ i

where ∇ ˜ is the Levi-Civita connection of h̃ (see Exercise 6). The first con-
straint equation for the metric h̃ and the symmetric tensor K̃, on the other
hand, becomes

¯ − 1 R̄u + 1 u−7 Kij K ij = 0.

R̃ − K̃ij K̃ ij = 0 ⇔ ∆u
8 8
This is a nonlinear elliptic equation in one variable, much simpler than the
original system. If one chooses the so-called time symmetric case K = 0
(the reason for this designation being that in the Gauss Lemma coordinates
t → −t is clearly an isometry of the solution), this equation becomes linear:

¯ − 1 R̄u = 0.
∆u
8
As an example, choose h to be the Euclidean metric, hij = δij ; then R̄ = 0
and the equation above is simply the Laplace equation
∆u = 0.
A simple solution, related to the gravitational field of a point mass, is
M
u=1+ .
2r
This solution leads exactly to the Schwarzschild solution, which in isotropic
coordinates is written
!2
1− M M 4
 
2 2
ds = − 2r
dt + 1 + (dr2 + r2 (dθ2 + sin2 θdϕ2 )).
1+ M2r
2r

Since the Laplace equation is linear, one can superimpose solutions. Thus
an initial data set for a set of N black holes initially at rest can be obtained
by choosing
N  
X Mi
u= 1+ ,
2ri
i=1

where Mi > 0 and ri is the Euclidean distance to a fixed point pi ∈ R3

(i = 1, . . . , N ).
104 5. CAUCHY PROBLEM

6. Einstein equations with matter

So far we have analyzed only the vacuum Einstein field equations. To in-
clude matter (and also a cosmological constant Λ) we must introduce matter
fields, generically represented by ψ, and consider a system of the form
(
Gµν + Λgµν = 8πTµν (g, ψ)
.
Field equations for ψ
If the equations for ψ form a hyperbolic system then Choquet-Bruhat’s
theorem still applies, with the constraint equations
(
2
R̄ + K ii − Kij K ij − 2Λ = 16πρ
.
∇¯ iK j − ∇
¯ j K j = 8πJi
j i
Here ρ = T00 and Ji = −T0i are computed from h and the initial data for
ψ. As an example, the Einstein-Klein-Gordon system is given by
(
Gµν + Λgµν = 8π ∂µ φ ∂ν φ − 12 gµν ∂α φ ∂ α φ + m2 φ2



,
∇µ ∂µ φ − m2 φ = 0
and we have (
ρ = 21 ψ0 2 + hij ∂i φ0 ∂j φ0 + m2 φ0 2


,
Ji = −ψ0 ∂i φ0
where φ0 and ψ0 are the initial data for φ.

7. Exercises
(1) Let (M, g) be a time-oriented Lorentzian manifold and X a future-
pointing timelike vector field. Given a smooth function φ : M → R
let
1
Tµν = ∂µ φ∂ν φ − gµν ∂α φ∂ α φ + m2 φ2


2
be the energy-momentum tensor associated to the Klein-Gordon
equation for φ, and let Y be the vector field defined by
Yµ = Tµν X ν .
Show that:
(a) Y = (X · φ) grad φ − 12 hgrad φ, grad φi + m2 φ2 X;

(b) Y is causal.
(c) Y is past-pointing.
(2) (Sobolev inequality) Let Q be a closed solid cone in Rn with
height h, solid angle Ω and vertex at the origin. Let ψ : R → R
be a smooth nonincreasing function with ψ(r) = 1 for r < h3 and
ψ(r) = 0 for r > 2h
3 . Show that:
(a) For any smooth function f : Q → R and any k ∈ N we have
Z R(θ)
(−1)k ∂k
f (0) = rk−1 k (ψ(r)f (r, θ)) dr,
(k − 1)! 0 ∂r
 7. EXERCISES 105

where (r, θ) are the usual spherical coordinates in Rn and r =

R(θ) is the equation for the base of the cone (here f (r, θ)
represents the function f written in spherical coordinates).
(b) There exists a constant C, depending on k and Ω, such that
∂k
Z
f (0) = C rk−n k (ψf ) .
Q ∂r
(c) For k > n2 we have |f (0)| ≤ C ′ kf kH k (Q) , where the constant C ′
depends on k, h and Ω only (you will need to use the Cauchy-
Schwarz inequality for multiple integrals).
(3) Consider a Lorentzian metric given in the Gauss Lemma form
g = −dt2 + hij (t, x)dxi dxj ,
so that the level sets of t are Riemannian manifolds with induced
metric h(t) = hij dxi dxj and second fundamental form
1 ∂hij i j
K(t) = dx dx .
2 ∂t
Show that in these coordinates:
(a) The Christoffel symbols are
Γ0ij = Kij ; Γijk = Γ̄ijk ; Γi0j = K ij ,
where Γ̄ijk are the Christoffel symbols of h.
(b) The components of the Riemann tensor are
∂ j
R0i0j = − K − Kil K lj ;
∂t i
Rij0 l ¯ i K lj + ∇
= −∇ ¯ j K li ;
Rijl m = R̄ijl m + Kil K mj − Kjl K mi ,
¯ is the Levi-Civita connection of h and R̄ m are the
where ∇ ijl
components of the Riemann tensor of h.
(c) The time derivative of the inverse metric is given by the for-
mula
∂hij
= −2K ij .
∂t
(d) The components of the Ricci tensor are
∂ i
R00 = − K − Kij K ij ;
∂t i
¯ iK j + ∇
R0i = −∇ ¯ jKj ;
j i
∂
Rij = R̄ij + Kij − 2Kil K lj + K ll Kij ,
∂t
where R̄ij are the components of the Ricci tensor of h.
106 5. CAUCHY PROBLEM

(e) The scalar curvature is

∂ i
2
R = R̄ + 2 K i + K ii + Kij K ij ,
∂t
where R̄ is the scalar curvature of h.
(f) The component G00 of the Einstein tensor is
1
2 
G00 = R̄ + K ii − Kij K ij .
2
(4) Let (M, g) be an (n + 1)-dimensional Lorentzian manifold and
(x0 , . . . , xn ) local coordinates on M . Show that:
(a) The condition for these coordinates to be harmonic is written
1
∇α ∇α xµ = 0 ⇔ H µ ≡ ∂α g αµ + g αµ g ρσ ∂α gρσ = 0
2
(µ = 0, . . . , n).
(b) The reduced Ricci tensor is
H 1
Rµν ≡ Rµν + gα(µ ∂ν) H α = − g αβ ∂α ∂β gµν + Fµν (g, ∂g).
2
(5) Denoting by h0 the standard constant curvature metric on R3 , S 3 or
H 3 , compute the cosmological constant and determine the maximal
globally hyperbolic developments of the following sets of initial data
for the vacuum Einstein equations (possibly with a cosmological
constant):
(a) (R3 , h0 , 0);
(b) (R3 , h0 , h0 );
(c) (S 3 , h0 , 0);
(d) (H 3 , h0 , 0);
(e) (H 3 , h0 , h0 );
(f) (R3 , h0 , diag(p1 , p2 , p3 )) with p1 + p2 + p3 = p1 2 + p2 2 + p3 2 = 1.
(6) Let (S, h, K) be an initial data set for the vacuum Einstein equa-
tions, with K traceless and divergenceless. Given a smooth posi-
tive function u : S → R, consider the conformally rescaled metric
h̃ = u4 h and the symmetric tensor K̃ = u−2 K. By using normal
coordinates when convenient, show that:
(a) The Christoffel symbols Γ̃ijk of h̃ are related to the Christoffel
symbols Γ̄ijk of h by

Γ̃ijk = Γ̄ijk + 2∂j (log u)hi k + 2∂k (log u)hi j − 2∂ i (log u)hjk .
(b) K̃ is divergenceless for the Levi-Civita connection of h̃.
(c) The Ricci tensor R̃ij of h̃ is related to the Ricci tensor R̄ij of
h by
¯ i ∂j (log u) − 2∆(log
R̃ij = R̄ij − 2∇ ¯ u)hij
+ 4∂i (log u)∂j (log u) − 4 |grad(log u)|2 hij .
 7. EXERCISES 107

(d) The scalar curvature R̃ of h̃ is related to the scalar curvature

R̄ of h by
¯
R̃ = u−4 R̄ − 8u−5 ∆u.
(7) Check that the metric
!2
1− M M 4
 
2 2
ds = − 2r
M
dt + 1 + (dr2 + r2 (dθ2 + sin2 θdϕ2 ))
1 + 2r 2r
is indeed the Schwarzschild metric by making the coordinate change
M 2
 
R=r 1+ .
2r
 CHAPTER 6

Mass in general relativity

In this chapter we address the important but subtle concept of mass in

general relativity. Following [Wal84], we start by defining the Komar mass
for stationary spacetimes. We then discuss field theory and introduce the
Einstein-Hilbert action as a means of motivating the definition of the ADM
mass. Finally, we prove the (Riemannian) positive mass theorem and the
(Riemannian) Penrose inequality for graphs, following [Lam10]. For more
details see [Mar09, Bra11].

1. Komar mass
Recall that the Newtonian gravitational field satisfies
div G = −4πρ,
where ρ is the mass density of the matter generating the field. Therefore
the total mass of a given system is given by
1 1
Z Z Z
M= ρ=− div G = − hG, ni ,
R3 4π R3 4π Σ
where Σ is any surface enclosing all the matter and n is the outward unit
normal. The fact that M does not depend on Σ is equivalent to the statement
that div G = 0 in the region between any two such surfaces.
For a static Lorentzian metric, that is, a metric of the form
ds2 = −e2φ dt2 + hij dxi dxj
with φ and h not depending on t, the analogue of the gravitational field
is minus the acceleration of the observers with constant space coordinates
xi , that is, − grad φ (see Chapter 2). On the other hand, since the metric
is static, we expect that the energy computed at a given surface should be
multiplied by the redshift factor eφ to obtain its reference value at infinity.
The relativistic analogue of the formula above is then
1 1
Z Z
M= eφ (∂µ φ)nµ = (∂µ eφ )nµ .
4π Σ 4π Σ
We have
1
∂µ eφ = ∂µ (−K ν Kν ) 2 = −e−φ K ν ∇µ Kν = −N ν ∇µ Kν ,
∂ ∂
where K = ∂t is the timelike Killing vector field and N = e−φ ∂t is the unit
timelike vector with the same direction, that is, the future-pointing unit
109
110 6. MASS IN GENERAL RELATIVITY

normal to the hypersurfaces St of constant t (see Figure 1). Therefore, we

can write
1 1
Z Z
µ ν
M =− (∇µ Kν )n N = (∇µ Kν )N µ nν .
4π Σ 4π Σ

K
N
E1
n

− grad φ

Σ
St

Figure 1. Computing the Komar mass on a static spacetime.

Because K is a Killing vector field, ∇µ Kν is a 2-form; more precisely,

(dK ♭ )µν = ∇µ Kν − ∇ν Kµ = 2∇µ Kν .
If E1 and E2 are two unit vector fields tangent to Σ such that {N, n, E1 , E2 }
is a positive orthonormal frame (see Figure 1), and so {−N ♭ , n♭ , E1♭ , E2♭ } is
a positive orthonormal coframe, then we can expand
∇K ♭ = −∇K ♭ (N, n)N ♭ ∧ n♭ + . . . ,
and so
1 1
Z Z
♭
M= ∇K (N, n) = ∇K ♭ (N, n)E1♭ ∧ E2♭
4π Σ 4π Σ
1 1
Z Z
♭ ♭ ♭
=− −∇K (N, n) ⋆ (N ∧ n ) = − ⋆∇K ♭ ,
4π Σ 4π Σ
that is
1
Z
M =− ⋆dK ♭ .
8π Σ
This expression is the so-called Komar mass. Although we arrived at this
expression by considering a static spacetime, it actually can be defined for
any stationary spacetime. In other words, the timelike Killing vector field
K does not have to be hypersurface-orthogonal.
To show that the Komar mass is well defined, that is, that ⋆dK ♭ is a
closed 2-form in vacuum, we start by noticing that if X is any vector field
then
(div X)ǫ = d(Xy ǫ) = d ⋆ X ♭ ,
whence
div X = − ⋆ d ⋆ X ♭ .
 1. KOMAR MASS 111

In particular, for any smooth function φ

φ = div grad φ = − ⋆ d ⋆ (grad φ)♭ = − ⋆ d ⋆ dφ.
In local coordinates, we have
q
⋆ dφ = ǫ(grad φ, ·, ·, ·) = | det(gµν )| dx0 ∧ dx1 ∧ dx2 ∧ dx3 (grad φ, ·, ·, ·)
q q
= | det(gµν )| ∂ 0 φ dx1 ∧ dx2 ∧ dx3 − | det(gµν )| ∂ 1 φ dx0 ∧ dx2 ∧ dx3 + . . .
Therefore
q 
d ⋆ dφ = ∂0 | det(gµν )| ∂ φ dx0 ∧ dx1 ∧ dx2 ∧ dx3
0

q 
+ ∂1 | det(gµν )| ∂ φ dx0 ∧ dx1 ∧ dx2 ∧ dx3 + . . .
1

and we obtain the useful formula

q 
1 α
φ = p ∂α | det(gµν )| ∂ φ .
| det(gµν )|
It is natural to try and generalize this formula for arbitrary k-forms.
Definition 1.1. If ω is a k-form then its Hodge d’Alembertian is
the k-form
H ω = − ⋆ d ⋆ dω − d ⋆ d ⋆ ω.
Note that d ⋆ φ = 0 for 0-forms φ, and so we obtain the formula above.
It turns out that in general the Hodge d’Alembertian does not coincide with
the usual d’Alembertian
ω = ∇µ ∇µ ω
(sometimes called the rough d’Alembertian). The relation between these
two operators for 1-forms is given by the following result.
Theorem 1.2. (Weitzenbock formula) If ω is a 1-form then
H ω − ω = −Ric(ω ♯ , ·).
Proof. We have
3·2 1
(⋆d ⋆ dω)δ = ǫγαβδ ∇γ (ǫµναβ ∇µ ων ) = ǫαβγδ ǫµναβ ∇γ ∇µ ων
3! 2! 2!
= (−g µγ g ν δ + g µδ g ν γ )∇γ ∇µ ων = −∇µ ∇µ ωδ + ∇ν ∇δ ων
and
4 1
(d ⋆ d ⋆ ω)δ = ∇δ (ǫγναβ ∇γ (ǫµναβ ωµ )) = ǫγναβ ǫµναβ ∇δ ∇γ ωµ
4! 3!
= −g µγ ∇δ ∇γ ωµ = −∇δ ∇µ ωµ .
Therefore
(⋆d ⋆ dω + d ⋆ d ⋆ ω)δ = −(ω)δ + ∇µ ∇δ ωµ − ∇δ ∇µ ωµ .
112 6. MASS IN GENERAL RELATIVITY

The Weitzenbock formula now follows from

∇µ ∇δ ω µ − ∇δ ∇µ ω µ = Rµδ µν ω ν = Rδν ω ν .

∂
Now let K = ∂t be the timelike Killing vector field in a stationary
spacetime. Then
 q 
∂
d ⋆ K♭ = d y | det(gµν )| dt ∧ dx1 ∧ dx2 ∧ dx3
∂t
q 
1 2 3
=d | det(gµν )| dx ∧ dx ∧ dx

∂
q
= | det(gµν )| dx1 ∧ dx2 ∧ dx3 = 0,
∂t
since the metric coefficients do not depend on t. From the Weitzenbock
formula we then have
⋆d ⋆ dK ♭ = −K ♭ + Ric(K, ·).
Now, adding or subtracting cyclic permutations of the fundamental identity
for the Riemann tensor,
∇µ ∇ν Kα − ∇ν ∇µ Kα = Rµνα β Kβ ,
∇ν ∇α Kµ − ∇α ∇ν Kµ = Rναµ β Kβ ,
∇α ∇µ Kν − ∇µ ∇α Kν = Rαµν β Kβ ,
and using the Killing equation and the first Bianchi identity, we have
∇µ ∇ν Kα = −Rναµ β Kβ ,
whence
Kα = −Rαβ Kβ .
We conclude that
⋆d ⋆ dK ♭ = 2Ric(K, ·) ⇔ d ⋆ dK ♭ = 2 ⋆ Ric(K, ·),
implying that ⋆dK ♭ is indeed closed in vacuum, that is, the Komar mass is
well defined: any two homologous compact orientable surfaces Σ1 and Σ2
which enclose the matter content of the stationary spacetime can be used
to compute it (Figure 2).
If Σ is the boundary of a spacelike 3-dimensional manifold B whose
future-pointing unit normal is N then the Komar mass can be written as
1 1 1
Z Z Z
♭ ♭
M =− ⋆dK = − d ⋆ dK = − ⋆Ric(K, ·)
8π Σ 8π B 4π B
1 1
Z Z D E
=− ǫ(Ric(K, ·)♯ , ·, ·, ·) = − − (Ric(K, ·)♯ , N
4π B 4π B
Z  
1 1
Z
= Rµν K µ N ν = 2 Tµν − T gµν K µ N ν
4π B B 2
 1. KOMAR MASS 113

vacuum matter vacuum

Σ1

Σ2

Figure 2. Computing the Komar mass with two homolo-

gous surfaces.

(we used ǫ = −N ♭ ∧ σ in the second line, where σ is the volume element

for B). Note that the Komar mass is not, as one might have guessed, the
integral Z
M′ = Tµν K µ N ν .
B
This integral is also well defined in a stationary spacetime, since
∇µ (Tµν K ν ) = (∇µ Tµν )K ν + Tµν ∇µ K ν = 0,
due to the contracted Bianchi identity and the Killing equation. However,
we also have
∇µ (T Kµ ) = K µ ∇µ T + T ∇µ Kµ = 0,
because T is constant along K, which has zero divergence (by contracting
the Killing equation). Therefore we also have
  
µ 1 ν
∇ Tµν − T gµν K = 0.
2
To have an idea of what exactly is measured by the Komar mass, consider
a static spacetime, where it is possible to choose B such that N = e−φ K.
In this case, for a perfect fluid (whose flow lines are necessarily the integral
curves of K), we have
Z   Z  
1 1
Z
φ
M =2 T (N, N ) + T e = 2 ρ + (−ρ + 3p) eφ = (ρ + 3p) eφ .
B 2 B 2 B
Thus we see that the Komar mass also includes the pressure; this is remi-
niscent of the Newtonian formula for the internal energy of a monoatomic
114 6. MASS IN GENERAL RELATIVITY

gas,
3
U = pV.
2

2. Field theory
Let us consider the problem of how to define the energy of a field ψ in
flat Minkowski space. The field equations for the field ψ are usually the
equations for the critical points of an action
Z
S= L(ψ, ∂ψ) dt dx1 dx2 dx3 ,
R4

obtained by integrating a Lagrangian density L (where we assume that

the field decays fast enough so that S is well defined). The field equations
are then the Euler-Lagrange equations
 
∂L ∂L
∂µ − = 0.
∂(∂µ ψ) ∂ψ
Indeed, if ψ(λ) is a one-parameter family of fields such that ψ(0) is a critical
d
point of the action, and δ ≡ dλ |λ=0
, then
  Z  
∂L ∂L ∂L ∂L
Z
δS = δψ + δ(∂µ ψ) = δψ + ∂µ (δψ)
R4 ∂ψ ∂(∂µ ψ) R4 ∂ψ ∂(∂µ ψ)
  Z   
∂L ∂L ∂L
Z
= ∂µ δψ + − ∂µ δψ.
R4 ∂(∂µ ψ) R4 ∂ψ ∂(∂µ ψ)
For field variations δψ with compact support we then have
Z   
∂L ∂L
δS = − ∂µ δψ,
R4 ∂ψ ∂(∂µ ψ)
and so δS = 0 for al such variations if and only if the Euler-Lagrange
equations hold.
The canonical energy-momentum tensor is defined as
∂L
T µν = ∂ν ψ − Lδ µν
∂(∂µ ψ)
and satisfies
 
∂L ∂L
∂µ T µν = ∂µ ∂ν ψ + ∂µ ∂ν ψ
∂(∂µ ψ) ∂(∂µ ψ)
∂L ∂L
− (∂µ ψ)δ µν − (∂µ ∂α ψ)δ µν = 0.
∂ψ ∂(∂α ψ)
Defining the Hamiltonian density to be
∂L
H = −T 00 = − ∂0 ψ + L,
∂(∂0 ψ)
 2. FIELD THEORY 115

it is then clear from the divergence theorem that the Hamiltonian

Z
H= H dx1 dx2 dx3
St

is independent of the hypersurface St of constant t chosen to compute the

integral (assuming that the field decays fast enough so that H is well defined
and the boundary integral corresponding to the divergence term vanishes).
The Hamiltonian can be identified with the total energy of the field ψ, which
is therefore constant in time.
If we have N fields ψ1 , . . . , ψN instead of a single field ψ, then it is easily
seen that the field equations are
 
∂L ∂L
∂µ − =0 (i = 1, . . . , N ),
∂(∂µ ψi ) ∂ψi
the canonical energy-momentum tensor is
∂L
T µν = ∂ν ψi − Lδ µν
∂(∂µ ψi )
(summed over i), and the Hamiltonian density is
∂L
H = −T 00 = − ∂0 ψi + L.
∂(∂0 ψi )
Note carefully that up until this point the metric was not used to raise
or lower indices, or even to apply the divergence theorem. This will be
important in Section 5.
If we use Cartesian coordinates then the tensor
∂L
T µν = ∂ ν ψi − Lg µν
∂(∂µ ψi )
satisfies the conservation equation
∇µ T µν = 0.
This provides a method for obtaining the energy-momentum tensor that is
used in the Einstein field equations for the various matter models. We now
list some simple examples.

2.1. Klein-Gordon field. The Lagrangian density for the Klein-Gordon

field is
1
L = (g µν ∂µ φ ∂ν φ + m2 φ2 ).
2
Consequently the canonical energy-momentum tensor for the Klein-Gordon
field is
1
T µν = ∂ µ φ ∂ ν φ − g µν ∂α φ ∂ α φ + m2 φ2 ,


2
as claimed in Chapter 5.
116 6. MASS IN GENERAL RELATIVITY

2.2. Electromagnetic field. In this case we can take as our fields the
electromagnetic potentials A0 , A1 , A2 , A3 . The Lagrangian density (in units
where 4πε0 = 1) is
1 αµ βν
L= g g Fαβ Fµν ,
16π
where
Fµν = ∂µ Aν − ∂ν Aµ .
Using
∂Fαβ
= δ µα δ ν β − δ ν α δ µβ ,
∂(∂µ Aν )
it is easily seen that
 
∂L ∂L
∂µ − = 0 ⇔ ∂µ F µν = 0,
∂(∂µ Aν ) ∂Aν

which are indeed the Maxwell equations d ⋆ F = 0, as

ǫµγδν ∇µ ǫαβγδ Fαβ = ǫµνγδ ǫαβγδ ∇µ Fαβ

= −2(δ µα δ ν β − δ ν α δ µβ )∇µ Fαβ = −2∇µ Fµν .

Note that the equations dF = 0 follow automatically from the definition

F = dA. The canonical energy-momentum tensor for the electromagnetic
field is
 
µν ∂L ν µν 1 µα ν 1 αβ µν
Tcan = ∂ Aα − Lg = F ∂ Aα − Fαβ F g .
∂(∂µ Aα ) 4π 4

This tensor is neither symmetric nor gauge-invariant. However,

F µα ∂ ν Aα = F µα F να + F µα ∂α Aν ,

where the first term is symmetric and gauge-invariant, and the second term
is divergenceless:

∂µ (F µα ∂α Aν ) = F µα ∂µ ∂α Aν = 0

(because F is antisymmetric and the partial derivatives commute). There-

fore the tensor
 
µν 1 µα ν 1 αβ µν
T = F F α − Fαβ F g
4π 4

is symmetric, gauge-invariant and divergenceless. This is the true energy-

momentum tensor for the electromagnetic field.
 2. FIELD THEORY 117

R4

↓π

k
S

Figure 3. A continuous medium in Minkowski’s spacetime.

2.3. Relativistic elasticity. A continuous medium can be described

by a Riemannian 3-manifold (S, k) (the relaxed configuration) and pro-
jection map π : R4 → S whose level sets are timelike curves (the worldlines
of the medium particles), as shown in Figure 3.
If we choose local coordinates (x̄1 , x̄2 , x̄3 ) on S then we can think of π
as a set of three scalar fields x̄1 , x̄2 , x̄3 : R4 → R. For a given worldline, we
can complete this set of scalar fields into local coordinates (t̄, x̄1 , x̄2 , x̄3 ) for
R4 such that t̄ is the proper time along that worldline and its level sets are
orthogonal to it:
g = −dt̄2 + hij dx̄i dx̄j (on the worldline).
Notice that the orthogonal metric
h = hij dx̄i dx̄j
can be thought of as a time-dependent Riemannian metric on S, describing
the local deformations of the medium along each worldline, that is, the
deviations from the natural metric
k = kij dx̄i dx̄j .
We can compute the (inverse) metric h from
∂ x̄i ∂ x̄j
hij = g µν ,
∂xµ ∂xν
118 6. MASS IN GENERAL RELATIVITY

which does not depend on the choice of t̄. In other words, the metric h is
a quadratic function of the partial derivatives of the fields x̄1 , x̄2 , x̄3 . An
elastic Lagrangian density L for these fields is obtained by assuming that
L = L(x̄i , hij ). The canonical energy-momentum tensor is
∂L
T µν = ∂ ν x̄i − Lg µν ,
∂(∂µ x̄i )
and so in the coordinate system (t̄, x̄1 , x̄2 , x̄3 ) we have
T 0̄0̄ = L,
that is, the elastic Lagrangian density is just the rest energy density ρ = T 0̄0̄
measured by each particle in the medium. The choice of ρ = ρ(x̄i , hij ) is
called the elastic law of the continuous medium.
We define homogeneous and isotropic materials to be those for
which ρ depends only on the eigenvalues (s1 2 , s2 2 , s3 2 ) of hij with respect to
kij (that is, the eigenvalues of the matrix (hij ) in a frame where kij = δij ).
Note that (s1 , s2 , s3 ) are the stretch factors along the principal directions
given by the eigenvectors of hij , that is, the distance between two nearby
points along a principal direction in the current configuration, as measured
by the metric h, divided by the distance between the same points in the
relaxed configuration, as measured by the metric k.
Assume that kij = δij , that is, that (S, k) is the Euclidean space. We
define the more convenient variables
1
λ0 = det(hij ) = ;
(s1 s2 s3 )2

1 1 1
λ1 = tr(hij ) = 2
+ 2 + 2;
s1 s2 s3

1 1 1
λ2 = tr cof(hij ) = 2
+ 2
+ .
(s1 s2 ) (s2 s3 ) (s3 s1 )2
Note that
1
1 2
s1 s2 s3 =
λ0
is the volume occupied in the deformed state by a unit volume of material
in the relaxed configuration. Equivalently,
1
n = (λ0 ) 2
is the number density of particles of the medium in the deformed state, if we
normalize the number density in the relaxed configuration to be 1 particle
per unit volume. Elastic media whose elastic law depends only on n,
ρ = ρ(λ0 ),
 2. FIELD THEORY 119

are simply perfect fluids. To check this, we note that from the formula for
the inverse of a matrix we have
1 ji 1 ij
hij = A = A ,
λ0 λ0
where Aij is the (i, j)-cofactor of (hij ). On the other hand, from the Laplace
expansion for determinants we have
3
X
λ0 = hij Aij (no sum over i),
j=1
and so
∂λ0
= Aij = λ0 hij .
∂hij
Therefore
dρ ∂λ0 ∂hij
T µν = ∂ ν x̄k − ρg µν
dλ0 ∂hij ∂(∂µ x̄k )
dρ
= λ0 hij (g µα δ i k ∂α x̄j + g µα ∂α x̄i δ j k )∂ ν x̄k − ρg µν
dλ0
dρ
= 2λ0 hij ∂ µ x̄i ∂ ν x̄j − ρg µν .
dλ0
Since
hµν = hij ∂µ x̄i ∂ν x̄j
is simply the metric on the hyperplanes orthogonal to the worldlines, that
is,
hµν = gµν + Uµ Uν ,
where U is the unit tangent vector to the worldlines, we obtain
 
µν dρ µ ν dρ
T = 2λ0 U U + 2λ0 − ρ g µν
dλ0 dλ0
= (ρ + p)U µ U ν + pg µν
with
dρ
p = 2λ0 − ρ,
dλ0
which is indeed the energy-momentum tensor √ of a perfect fluid. For example,
dust corresponds to the elastic law ρ = ρ0 λ0 (for some positive constant
ρ0 ), yielding p = 0, and a stiff fluid, with equation of state p = ρ, is given
by the choice ρ = ρ0 λ0 .
To obtain elastic materials that are not fluids we must choose elastic
laws that also depend on λ1 and λ2 . For instance, an elastic law is said to
be quasi-Hookean if it is of the form
ρ = p̂(n) + µ̂(n)σ,
where σ is a shear scalar, that is, a non-negative function of the stretch
factors such that σ = 0 if and only if s1 = s2 = s3 . The functions ρ̂ and
µ̂ are called the unsheared energy density and the rigidity modulus
120 6. MASS IN GENERAL RELATIVITY

of the elastic material. Examples of these are the John quasi-Hookean

material, corresponding to the shear scalar
s1 2 + s2 2 + s3 2
σ= 1 − 3,
(s1 2 s2 2 s3 2 ) 3
and the Karlovini-Samuelsson quasi-Hookean material, corresponding
to the shear scalar
"  #
s1 s2 2 s1 s3 2 s2 s3 2
   
1
σ= − + − + − .
12 s2 s1 s3 s1 s3 s2

It is easily seen that the first elastic law is of the form

ρ = f (λ0 ) + g(λ0 )λ2 ,
whereas the second is of the form
ρ = f (λ0 ) + g(λ0 )λ1 λ2 .

3. Einstein-Hilbert action
The variational formulation of field theories is coordinate-free, and so it
gives a simple method to write the field equations on an arbitrary coordinate
system. We must be careful, however, to note that the action written in
the new coordinate system must include the Jacobian of the coordinate
transformation:
Z q
S= L(ψ, ∂ψ) | det(gµν )| dx0 dx1 dx2 dx3 .
R4

This suggests that to generalize the field equations to an arbitrary curved

spacetime (M, g) one should consider actions of this form, where L should be
invariant under coordinate changes. In particular, one may wonder if there
is a Lagrangian action for the metric itself which yields the Einstein field
equations. The answer to this question is affirmative, and the corresponding
action is known as the Einstein-Hilbert action:
Z q Z
0 1 2 3
S= R | det(gµν )| dx dx dx dx = Rǫ,
M M

where R and ǫ are the scalar curvature and the volume element of the met-
ric g, and the integral is over an arbitrary (oriented) manifold M . Note
that R depends on g and its first and second partial derivatives, unlike the
Lagrangian densities that we encountered before. Instead of deriving the
Euler-Lagrange equations for this case, we will proceed in a more geometric
way. We note here, however, that this action is exceptional in that it leads to
second-order equations for the metric, as opposed to the fourth order equa-
tions that are typical of Lagrangian densities depending on second partial
derivatives.
 3. EINSTEIN-HILBERT ACTION 121

To obtain the Euler-Lagrange equations for the Einstein-Hilbert action

we start by considering two affine connections ∇ and ∇˜ on M . Because
˜ X − ∇X )(f Y ) = f (∇
(∇ ˜ X − ∇X )Y,
there exists a tensor C such that
(∇˜ X − ∇X )Y = C(X, Y ) ⇔ ∇
˜ µ Y ν = ∇µ Y ν + Cµα
ν
Y α.
If both ∇ and ∇˜ are symmetric then
˜ XY − ∇
0 = (∇ ˜ Y X) − (∇X Y − ∇Y X) = C(X, Y ) − C(Y, X),
that is, C is symmetric:
α α
Cµν = Cνµ .
Using the Leibniz rule, it is easy to determine the relation between the
covariant derivatives of any tensor using the two connections: for example,
∇˜ α Tµν
β β
= ∇α Tµν β
+ Cαγ γ
Tµν γ
− Cαµ β
Tγν γ
− Cαν β
Tµγ .
Assume now that ∇ ˜ is the Levi-Civita connection for the metric g. Then
we have
0=∇ ˜ α gµν = ∇α gµν − Cαµ
β β
gβν − Cαν gµβ = ∇α gµν − Cναµ − Cµαν .
By subtracting this identity from its cyclic permutations,
∇µ gνα = Cαµν + Cνµα ,
∇ν gαµ = Cµνα + Cανµ ,
we readily obtain
2Cαµν = ∇µ gνα + ∇ν gαµ − ∇α gµν ⇔
α 1
(14) Cµν = g αβ (∇µ gνβ + ∇ν gµβ − ∇β gµν ) .
2
Moreover, we have
˜ µ∇
∇ ˜ νXα = ∇
˜ µ (∇ν X α + C α X β ) = ∇µ ∇ν X α + C α ∇ν X β − Cµν
β
∇β X α
νβ µβ
α γ
+ ∇µ Cνβ X β + Cνβ
α
∇µ X β + Cµγ
α
Cνβ X β − Cµν
γ α
Cγβ Xβ,
whence
˜ µ∇
R̃µν αβ X β = (∇ ˜ν − ∇˜ ν∇
˜ µ )X α = R α X β
µν β
 
α α α γ α γ
+ ∇µ Cνβ − ∇ν Cµβ + Cµγ Cνβ − Cνγ Cµβ Xβ,
that is,
γ γ
(15) R̃µν αβ = Rµν αβ + ∇µ Cνβ
α α
− ∇ν Cµβ α
+ Cµγ Cνβ α
− Cνγ Cµβ .
Note that we retrieve the usual formulae for the Christoffel symbols and the
Riemann tensor from equations (14) and (15) in the case when ∇µ = ∂µ .
Let g(λ) be a one-parameter family of Lorentzian metrics on M , and
choose ∇ and ∇ ˜ to be the Levi-Civita connections of g(0) and g(λ). The
122 6. MASS IN GENERAL RELATIVITY

difference between these connections is a tensor C(λ) with C(0) = 0. Again

d
setting δ ≡ dλ |
, we have from (14)
λ=0

α 1
δCµν = g αβ (∇µ δgνβ + ∇ν δgµβ − ∇β δgµν )
2
1
∇µ δgν α + ∇ν δgµ α − ∇α δgµν ,


=
2
where gµν means gµν (0) and all indices are raised or lowered with this metric.
Note carefully that δg means the tensor δgµν , possible with some indices
raised. Thus for instance
δ(g µν ) = −g µα g νβ δgαβ = −δg µν .
From (15) we have

δRµν αβ = ∇µ δCνβ
α α
− ∇ν δCµβ ,
whence
α α
δRµν = ∇α δCµν − ∇µ δCαν
1
1
= ∇α ∇µ δgν α + ∇ν δgµ α − ∇α δgµν − ∇µ ∇ν δgα α .
2 2
Note that

g µν δRµν = −∇µ ∇µ δgν ν + ∇µ ∇ν δgµν = ∇µ (−∇µ δgν ν + ∇ν δgµν )

is a divergence with respect to the metric g(0), and will vanish when inte-
grated for variations of the metric with compact support.
The variation of the Einstein-Hilbert action is
Z Z Z
δS = δ Rǫ = δ(Rǫ) = (δRǫ + Rδǫ).
M M M

We have
δR = δ(g µν Rµν ) = −δg µν Rµν + ∇µ (−∇µ δgν ν + ∇ν δgµν )
and, using the identity
δ det A = (det A) tr(A−1 δA),

for any matrix-valued function A,

q
δǫ = δ − det(gµν ) dx0 ∧ dx1 ∧ dx2 ∧ dx3
1 1
= − (det(gµν )) g µν δgµν (− det(gµν ))− 2 dx0 ∧ dx1 ∧ dx2 ∧ dx3
2
1
= gµν δg µν ǫ.
2
 3. EINSTEIN-HILBERT ACTION 123

We conclude that
Z  
1
Z
δS = − Rµν − Rgµν δg µν ǫ + ∇µ (−∇µ δgν ν + ∇ν δgµν ) ǫ
2
ZM M

=− Gµν δg µν ǫ.
M
for variations of the metric with compact support. We conclude that the
Euler-Lagrange equations for the Einstein-Hilbert action are the Einstein
equations Gµν = 0.
Remark 3.1. It is easy to see that the Einstein tensor of a compact
surface (M, g) vanishes identically. Therefore we have
Z
δ Rǫ = 0
M
automatically. If g1 and g2 are any two Riemannian metrics on M then
g(λ) = (1 − λ)g1 + λg2 is a Riemannian metric which interpolates between
g1 and g2 . We then have
d
Z Z Z
R(λ)ǫ(λ) = 0 ⇒ R1 ǫ1 = R2 ǫ 2 ,
dλ M M M
that is, the integral of the scalar curvature does not depend on the metric.
This statement is known as the Gauss-Bonnet theorem.
To include matter fields ψ and a cosmological constant Λ in the Einstein
equations we consider the action
Z  
1
S= L(g, ψ) − (R − 2Λ) ǫ.
M 16π
It is clear that
1
Z Z
δS = E(ψ)δψ ǫ + (Gµν + Λgµν − 8πTµν ) δg µν ǫ,
M 16π M
where we have defined
1
Z Z Z
δ L(g, ψ)ǫ = E(ψ)δψ ǫ − Tµν δg µν ǫ.
M M 2 M
The Euler-Lagrange equations are then the Einstein equations with sources
plus the field equations for ψ:
(
Gµν + Λgµν = 8πTµν
.
E(ψ) = 0
The energy-momentum tensor is then
δL
Tµν = −2 µν − Lgµν ,
δg
where the second term comes from the variation of the volume element and
we have set
δL
δL = µν δg µν + E(ψ)δψ.
δg
124 6. MASS IN GENERAL RELATIVITY

It is interesting to note that the energy-momentum tensor as defined here of-

ten agrees with what one would expect from the canonical energy-momentum
tensor associated to the Lagrangian density L in Minkowski’s spacetime. See
[Sop76] for the exact relation.

4. Gravitational waves
Since we have computed the variation of the Ricci tensor, we make a
short digression to discuss the linearized Einstein vacuum equations, which
describe the propagation of gravitational waves on a fixed solution (M, g) of
the full nonlinear vacuum equations Rµν = Λgµν . The linearized equations
are simply
δRµν = Λδgµν ,
that is,
∇α ∇µ δgν α + ∇ν δgµ α − ∇α δgµν − ∇µ ∇ν δgα α = 2Λδgµν .


(16)
Note that some variations of the metric are trivial, in that they arise from
the diffeomorphism invariance of the Einstein equations: if ψλ is a one-
parameter family of diffeomorphisms, then the variation
g(λ) = ψλ ∗ g
yields metrics which are isometric to g = g(0), but expressed in different
coordinates. In this case we have
δg = LV g,
that is
δgµν = ∇µ Vν + ∇ν Vµ ,
where V is the vector field defined at each point p ∈ M by
d
Vp = ψλ (p).
dλ |λ=0
Therefore, the linearized Einstein equations have gauge freedom: we can
always add a Lie derivative of g to any given variation of the metric without
altering its physical meaning. This corresponds to gauge transformations of
the form
δgµν → δgµν + ∇µ Vν + ∇ν Vµ .
We will now construct a gauge where equations (16) look particularly simple.
To do that, we consider the trace-reversed metric perturbation
1
δg µν = δgµν − (δgα α )gµν ,
2
which transforms under a gauge transformation as
δg µν → δg µν + ∇µ Vν + ∇ν Vµ − (∇α V α )gµν ,
 4. GRAVITATIONAL WAVES 125

so that its divergence transforms as

∇µ δg µν →∇µ δg µν + Vν + ∇µ ∇ν V µ − ∇ν ∇α V α
= ∇µ δg µν + Vν + Rµν µα V α
= ∇µ δg µν + Vν + Rνα V α .
Assume that (M, g) is globally hyperbolic. By solving the wave equation
Vν + Rνα V α + ∇µ δg µν = 0,
we can then change to a gauge where
∇µ δg µν = 0,
that is,
1
∇µ δgµν = ∇ν δgα α .
2
Taking the trace of equation (16) we then obtain
(17)  δgα α + 2Λδgα α = 0.
Note that we still have a residual gauge freedom, corresponding to vector
fields V such that
Vν + Rνα V α = 0.
Choosing V and its derivatives on some Cauchy hypersurface S such that
δgα α + 2∇α V α = N · (δgα α + 2∇α V α ) = 0,
where N is the future-pointing unit normal to S, and solving the wave
equation for V above, we guarantee the existence of a gauge where
δgα α = N · (δgα α ) = 0
on S. The wave equation (17) then guarantees that
δgα α = 0
on M , and so
∇µ δgµν = 0.
This is the so-called transverse traceless gauge. In this gauge, the lin-
earized Einstein equation (16) can be written as
∇α ∇µ δgν α + ∇α ∇ν δgµ α − δgµν = 2Λδgµν .
Using
∇α ∇µ δgν α = ∇µ ∇α δgν α + Rαµνβ δg βα + Rαµ αβ δgν β
= Rαµνβ δg αβ + Rµβ δgν β
= Rαµνβ δg αβ + Λδgµν ,
we finally obtain
δgµν − 2Rαµνβ δg αβ = 0.
126 6. MASS IN GENERAL RELATIVITY

5. ADM mass
If we write the metric in the Gauss Lemma form,
g = −dt2 + hij (t, x)dxi dxj
then from the exercises in Chapter 5 we have
∂
2
K i i + K i i + Kij K ij .


R = R̄ + 2
∂t
Setting
∂
X = Ki i ,
∂t
we have
div X = ∇0 X 0 + ∇i X i = ∂0 X 0 + Γii0 X 0 = ∂0 X 0 + K i i X 0 ,
that is,
∂
2
K i i = div X − K i i .


∂t
We conclude that

2
R = R̄ − K i i + Kij K ij + 2 div X,
and so the Einstein-Hilbert action corresponds to the Lagrangian density
q 
2 
L = det(hij ) R̄ − K i i + Kij K ij .

Therefore the Hamiltonian density is

∂L ∂L
H=− ∂0 hij + L = − Kij + L
∂(∂0 hij ) ∂(Kij )
q  
= − det(hij ) 2K ij − 2K l l hij Kij + L
q   q
i 2 ij


= det(hij ) R̄ + K i − Kij K = 2 det(hij ) G00 = 0

for any solution of the vacuum Einstein field equations. That is, the total
energy associated to the fields hij is simply zero.
To try to obtain a nonzero quantity we reconsider the boundary terms
that were discarded when varying the Einstein-Hilbert action:
Z Z
µ ν ν
∇ (−∇µ δgν + ∇ δgµν ) ǫ = (−∇µ δgν ν + ∇ν δgµν ) nµ ω,
M ∂M
where we take M to be a manifold with a timelike boundary ∂M at infinity,
∂
tangent to ∂t , with unit normal n and volume element ω (Figure 4). Note
that it is not necessary to consider the flux of the vector field X which we
discarded above as it is orthogonal to n. The boundary integral can be
written as Z Z  
¯ k δhij ni σdt,
−∂i (hjk δhjk ) + hjk ∇
R Σ
 5. ADM MASS 127

∂
∂t
∂M

Figure 4. Computing the boundary terms for the Einstein-

Hilbert action.

∂
where we take ∂M to be the flow by ∂t of a spacelike surface Σ and ω = dt∧σ.
Suppose that h approaches the Euclidean metric at infinity, so that in an
appropriate coordinate system (x1 , x2 , x3 ) we have
hij = δij + O r−p , δhij = O r−p , ∂k hij , ∂k δhij , Γkij = O r−p−1

2 2 2
as r → +∞, with r2 = (x1 ) + (x2 ) + (x3 ) and p > 21 . Then the boundary
integral can be written as
xi
Z Z
lim (−∂i δhjj + ∂j δhij ) dt
r→∞ R S
r
r
xi
Z Z
= δ lim (−∂i hjj + ∂j hij ) dt = δI,
r→∞ R S
r
r
where Sr is a coordinate sphere of radius r. Note that δ can be brought
outside the integral because we have replaced the metric-dependent terms
¯ n and σ.
∇,
If S is the Einstein-Hilbert action, we then have
Z
δS = − Gµν δg µν + δI,
M
and so the Einstein equations hold if and only if
δ(S − I) = 0.
One can think of I as the integral of a singular Lagrangian density I. The
Einstein-Hilbert Lagrangian density L should therefore be replaced by L−I,
and so, since I does not depend on Kij , the corresponding Hamiltonian
128 6. MASS IN GENERAL RELATIVITY

density H should be replaced by H − I = −I. The Hamiltonian should then

be
xi
Z
H = − lim (∂j hij − ∂i hjj ) .
r→∞ S
r
r
This Hamiltonian suggests a definition of the total energy of the gravitational
field at a given time slice, provided that h approaches the Euclidean metric
at infinity.
Definition 5.1. A 3-dimensional Riemannian manifold (S, h) is said
to be asymptotically flat if there exist:
(i) A compact set K ⊂ S such that S \ K is diffeomorphic to R3 \ B1 (0)
(Figure 5);
(ii) A chart (x1 , x2 , x3 ) on S \ K (called a chart at infinity) such that
|hij − δij | + r|∂k hij | + r2 |∂k ∂l hij | = O(r−p ) and R̄ = O(r−q )
2 2 2
for some p > 21 and q > 3, where r2 = (x1 ) + (x2 ) + (x3 ) and R̄ is
the scalar curvature of h.

Figure 5. Asymptotically flat manifold.

Definition 5.2. (Arnowitt-Deser-Misner [ADM61]) The ADM

mass of an asymptotically flat Riemannian manifold (S, h) is
1 xi
Z
M = lim (∂j hij − ∂i hjj ) ,
r→+∞ 16π S r
r

where Sr is a sphere of radius r in the chart at infinity (x1 , x2 , x3 ).

Note that in this definition the Hamiltonian has been multiplied by the
1
factor − 16π that is used when coupling to matter fields.
Theorem 5.3. (Ashtekar [AMA79]) If (S, h) is asymptotically flat
and the maximal Cauchy development of (S, h, K) is stationary then the
Komar mass of the maximal Cauchy development coincides with the ADM
mass of (S, h).
 6. POSITIVE MASS THEOREM 129

Theorem 5.4. (Bartnik [Bar86]) The ADM mass is well defined, that
is, it does not depend on the choice of the chart at infinity.

6. Positive mass theorem

Since the gravitational field is attractive, its energy is presumably neg-
ative, at least for bound states. On the other hand, we expect gravitational
waves to carry positive energy. It is therefore an important question to de-
cide whether there is a lower bound for the ADM mass (the inexistence of
which would signal an instability). In the simplest case of time-symmetric
initial data (K = 0) the restriction equations reduce to

R̄ = 16πρ,

and we expect ρ ≥ 0 ⇔ R̄ ≥ 0 for reasonable matter fields. In this case, we

have the following famous result.

Theorem 6.1. (Schoen and Yau [SY81]) Let (S, h) be a complete

asymptotically flat Riemannian 3-manifold with scalar curvature R̄ ≥ 0.
Then:
(i) Its ADM mass is nonnegative, M ≥ 0.
(ii) If M = 0 then S = R3 and h is the Euclidean metric.

Proof. Following [Lam10], we give a proof of (i) only for asymptot-

ically flat Riemannian 3-manifolds S that are graphs of smooth functions
f : R3 → R with the metric h induced by the Euclidean metric of R4 . Us-
ing the Cartesian coordinates (x1 , x2 , x3 ) of R3 as global coordinates on the
graph we have
hij = δij + ∂i f ∂j f.
From that expression one can show that the scalar curvature of the graph
is the divergence of a vector field on R3 :
 
1
R̄ = ∂i (∂i f ∂j ∂j f − ∂i ∂j f ∂j f ) .
1 + | grad f |2

On the other hand,

1 xi
Z
M = lim (∂j hij − ∂i hjj )
r→+∞ 16π Sr r
1 xi
Z
= lim (∂i ∂j f ∂j f + ∂i f ∂j ∂j f − 2∂i ∂j f ∂j f )
r→+∞ 16π Sr r
1 xi
Z
= lim (∂i f ∂j ∂j f − ∂i ∂j f ∂j f ) .
r→+∞ 16π Sr r
130 6. MASS IN GENERAL RELATIVITY

Since (S, h) is asymptotically flat, the derivatives of f approach zero with

certain decays as r → +∞. One can then easily show that
Z   i
1 1 x
M = lim (∂ f ∂j ∂j f − ∂i ∂j f ∂j f )
2 i
r→+∞ 16π S 1 + | grad f | r
r 
1 1
Z
= ∂i (∂i f ∂j ∂j f − ∂i ∂j f ∂j f )
16π R3 1 + | grad f |2
1
Z
= R̄ ≥ 0.
16π R3
We do not prove the rigidity statement (ii). It is interesting to note that
this statement, together with the formula above for the ADM mass, implies
that any graph with zero scalar curvature is flat. 
The volume element of the graph is
ǫ = 1 + | grad f |2 dx1 ∧ dx2 ∧ dx3 ,
p

since the eigenvalues of the matrix (hij ) are 1 + | grad f |2 for the eigenvector
grad f and 1 for the eigenvectors orthogonal to grad f . Consequently the
ADM mass is
1 R̄ ρ
Z Z Z
M= p ǫ= p ǫ< ρ ǫ.
16π S 1 + | grad f |2 S 1 + | grad f |2 S

The difference Z
M− ρǫ < 0
S
can be thought of as the (negative) gravitational binding energy.

7. Penrose inequality
The positive mass theorem admits a refinement in the case when black
holes are present, known as the Penrose inequality. The idea is that black
hole horizons correspond to minimal surfaces Σ on the Riemannian manifold
(S, h), each contributing with a mass M at least as big as the mass of a
Schwarzschild black hole with the same event horizon area A:
r
A
M≥ .
16π
To understand the precise motivation for this inequality, recall that in the
proof of the Penrose singularity theorem we defined the outward null expan-
sion of a 2-surface Σ on a Cauchy hypersurface S as
1 ∂γAB 1  
θ = γ AB = tr L ∂ g .
2 ∂r 2 ∂r |T Σ
If N is the future-pointing unit normal to S and n is the outward unit normal
to Σ on S we then have
1 1
θ = tr (LN +n g)|T Σ = tr K|T Σ + tr (Ln g)|T Σ ,
2 2
 7. PENROSE INEQUALITY 131

∂
where we used the fact that ∂r = N + n on Σ, and also that if X, Y are
tangent to Σ then
(LZ g) (X, Y ) = hX, ∇Y Zi + hY, ∇X Zi
depends only on Z along Σ. For time-symmetric initial data (K = 0) this
becomes
1 1
θ = tr (Ln g)|T Σ = tr (Ln h)|T Σ = tr κ,
2 2
where h is the metric induced by g on S and κ is the second fundamental form
of Σ on S. We conclude that Σ is marginally trapped, that is, has zero
outward null expansion, if and only tr κ = 0, which is precisely the condition
for Σ to be a minimal surface. Now a marginally trapped surface anticipates
the formation of trapped surfaces, which will lead to geodesic incompleteness
of the resulting spacetime, presumably due to singularities. If one believes
the weak cosmic censorship conjecture, these singularities can only
occur inside black holes, and so there must be a black hole event horizon
enveloping any marginally trapped surface, presumably with greater area.
We shall see in Chapter 7 that the area of a black hole horizon cannot
decrease towards the future, and also that, for a given mass, the area of the
horizon is maximal for a Schwarzschild black hole. If we assume that the
final state of the maximal globally hyperbolic development of (S, h, 0) is a
black hole spacetime, then the final black hole mass must be bounded by the
ADM mass (the difference being the energy carried away by gravitational
radiation). Combining all these observations leads to the Penrose inequality.
Theorem 7.1. (Huisken and Ilmanen [HI01], Bray [Bra01]) Let
(S, h) be a complete asymptotically flat Riemannian manifold with scalar
curvature R̄ ≥ 0. Then:
q
A
(i) Its ADM mass satisfies M ≥ 16π , where A is the sum of the areas
of the outermost
q minimal surfaces (Figure 6).
A
(ii) If M = 16π then the restriction of (S, h, 0) to the exterior of the outer
minimal surfaces coincides with the initial data for the Schwarzschild
solution of mass M outside the event horizon.
Proof. Following [Lam10], we give a proof of only for asymptotically
flat Riemannian 3-manifolds S that are graphs of smooth functions f : U ⊂
R3 → R, with the metric h induced by the Euclidean metric of R4 . Here
N
[
3
U =R \ Ka ,
a=1

where K1 , . . . , KN are disjoint convex compact sets with smooth boundaries

∂K1 , . . . , ∂KN to which f extends as a constant and where | grad f | → +∞
(so that they are minimal surfaces of the graph). Applying the divergence
theorem as in the proof of the positive mass theorem for graphs we now
132 6. MASS IN GENERAL RELATIVITY

obtain
N Z  
1 1 X 1
Z
M= R̄ + (∂
2 i
f ∂ ∂
j j f − ∂ ∂
i j f ∂ j f ) ni ,
16π U 16π ∂Ka 1 + | grad f |
a=1
where n is the outward unit normal to ∂Ka . Now, as is well known, the
¯ of f in ∂Ka by
Laplacian ∆f of f in U is related to the Laplacian ∆f
∆f = ∆f¯ + Hf (n, n) + (n · f ) tr κ,
where Hf is the Hessian of f and κ is the second fundamental form of ∂Ka
¯ = 0, and so
in R3 . Since f is constant on ∂Ka we have ∆f
(∂i f ∂j ∂j f − ∂i ∂j f ∂j f )ni = (n · f )∆f − Hf (n, grad f )
= (n · f )Hf (n, n) + (n · f )2 tr κ − hgrad f, niHf (n, n)
= hgrad f, ni2 tr κ = | grad f |2 tr κ,
where we used grad f = hgrad f, nin. Therefore
N Z
| grad f |2
 
1 1 X
Z
M= R̄ + tr κ
16π U 16π 1 + | grad f |2
a=1 ∂Ka
N Z
1 1 X
Z
= R̄ + tr κ.
16π U 16π ∂Ka a=1
Now Minkowski’s inequality for the smooth boundaries of compact con-
vex sets states that Z p
tr κ ≥ 16πAa ,
∂Ka
where Aa is the area of ∂Ka . Since R̄ ≥ 0 we conclude that
v
N r u N
X Aa u 1 X
M≥ ≥ t Aa ,
16π 16π
a=1 a=1
√ √ √
where we used a + b ≤ a + b for a, b > 0.
We do not prove the rigidity statement (ii). It is interesting to note that
this statement, together with the formula above for the ADM mass, implies
that any graph of the type considered above with zero scalar curvature is a
Flamm paraboloid. 

8. Exercises
(1) Let g be a static spherically symmetric Lorentzian metric on R4
whose matter fields have spatially compact support and satisfy the
dominant energy condition. There exist smooth functions φ = φ(r)
and m = m(r) such that
dr2
g = −e2φ(r) dt2 + 2 2 2 2


+ r dθ + sin θdϕ .
1 − 2m(r)
r
 8. EXERCISES 133

A1
A2

Figure 6. Penrose inequality; the sum of the areas of the

outer minimal surfaces is A = A1 + A2 .

Show that:
(a) The Einstein equations imply
dm
= 4πr2 ρ;
dr
dφ m + 4πr3 p
= ,
dr r(r − 2m)
where ρ and p are the energy density and the radial pressure
as measured by the static observers.
(b) There exist constants M ≥ 0 and Φ ∈ R such that
lim m(r) = M and lim φ(r) = Φ.
r→+∞ r→+∞

(c) If we choose the coordinate t such that Φ = 0 then M is the

Komar mass of g with respect to the timelike Killing vector
∂
field ∂t .
(d) The constant M satisfies M ≤ E, where
Z
E= ρ,
{t=0}

with equality exactly when g is the Minkowski metric.

(2) Starting with the appropriate Lagrangian density, show that the
Klein-Gordon equation can be written as
q 
1
p ∂α | det(gµν )| ∂ φ − m2 φ = 0.
α
| det(gµν )|
(3) Starting with the Einstein-Hilbert-Klein-Gordon action
Z  
1 µν 1
g ∂µ φ ∂ν φ + m2 φ2 −


S= R ǫ
M 2 16π
134 6. MASS IN GENERAL RELATIVITY

obtain the energy-momentum tensor for φ:

1
Tµν = ∂µ φ ∂ν φ − gµν ∂α φ ∂ α φ + m2 φ2 .


2
Note that this agrees with what one would expect from the canon-
ical energy-momentum tensor.
(4) Show that the Einstein-Hilbert action is equivalent to the (first
order, non-geometric) Einstein action
Z  q
S= g αβ Γδαγ Γγβδ − Γγαβ Γδγδ | det(gµν )| dx0 dx1 dx2 dx3 ,
M
by showing that they differ by a term of the form ∂α Qα , where
 q
Qα = g βγ Γαβγ − g αβ Γγβγ | det(gµν )|.
The following formulae will be useful:
1
q q
∂α | det(gµν )| = − gβγ ∂α g βγ | det(gµν )|
2
and
∇α g βγ = 0 ⇔ ∂α g βγ = −Γβαδ g δγ − Γγαδ g βδ .
(5) Let (M, g) be a globally hyperbolic solution of the vacuum Einstein
equations Rµν = Λgµν . Show that:
(a) The transverse traceless gauge is preserved by vector fields V
satisfying
Vν + Rνα V α = 0 and ∇µ V µ = 0.
(b) If V and its derivatives are chosen on some Cauchy hypersur-
face S such that
∇µ V µ = N · (∇µ V µ ) = 0,
where N is the future-pointing unit normal to S, then the wave
equation for V guarantees that V is divergenceless on M .
(c) Given Cartesian coordinates xµ in Minkowski spacetime, there
exists a transverse traceless gauge where δg0µ = 0 (assume that
the Poisson equation can be solved in R3 as needed).
(6) Let f : R3 → R be a smooth function and consider the metric h
induced on its graph S by the Euclidean metric in R4 ,
hij = δij + ∂i f ∂j f.
Show that:
(a) The inverse metric is
∂i f ∂j f
hij = δij − .
1 + | grad f |2
(b) The Christoffel symbols are
∂k f ∂i ∂j f
Γ̄kij = .
1 + | grad f |2
 8. EXERCISES 135

(c) The Ricci tensor is

∂i ∂j f ∂k ∂k f − ∂i ∂k f ∂j ∂k f
R̄ij =
1 + | grad f |2
∂k f ∂l f ∂i ∂k f ∂j ∂l f − ∂k f ∂l f ∂i ∂j f ∂k ∂l f
+ .
(1 + | grad f |2 )2
(d) The scalar curvature is
∂i ∂i f ∂j ∂j f − ∂i ∂j f ∂i ∂j f
R̄ =
1 + | grad f |2
2∂j f ∂k f (∂i ∂i f ∂j ∂k f − ∂i ∂j f ∂i ∂k f )
− .
(1 + | grad f |2 )2
(e) The scalar curvature can be written as
 
1
R̄ = ∂i (∂i f ∂j ∂j f − ∂i ∂j f ∂j f ) .
1 + | grad f |2
(7) Let h be the spherically symmetric Riemannian metric defined in
R3 by
dr2
+ r2 dθ2 + sin2 θdϕ2 ,


h= 2m(r)
1− r
where m is a smooth function whose derivative has compact sup-
port.
(a) Check that in Cartesian coordinates we have
2m(r)
r3
hij = δij + xi xj .
1 − 2m(r)
r

(b) Show that the limit

M = lim m(r)
r→+∞

exists, and h is asymptotically flat with ADM mass M (which

in particular coincides with the Komar mass when appropri-
ate).
(c) Check that h has scalar curvature
4 dm
R̄ = ,
r2 dr
and use this to prove the Riemannian positive mass theorem
for h.
(d) Show that r = r0 is a minimal surface if and only if m(r0 ) = r20
(in which case r is a well-defined coordinate only for r > r0 ),
and use this to prove the Riemannian Penrose inequality for
h.
136 6. MASS IN GENERAL RELATIVITY

(8) Consider a Riemannian metric given in the Gauss Lemma form

g = dt2 + hij (t, x)dxi dxj ,
so that the hypersurface t = 0 is a Riemannian manifold with
induced metric h = hij dxi dxj and second fundamental form
1 ∂hij i j
K= dx dx .
2 ∂t
Show that:
(a) The Laplacian operators ∆ and ∆ ¯ of g and h are related by
¯ + Hf (∂0 , ∂0 ) + (∂0 f ) tr K,
∆f = ∆f
where Hf is the Hessian of f .
(b) The metric induced on the hypersurface t = λf (x) is
h(λ) = hij (λf (x), x) + λ2 ∂i f ∂j f dxi dxj .
 

(c) The first variation of this metric is

d
δh ≡ h(λ) = 2f Kij dxi dxj .
dλ |λ=0
(d) The first variation of the volume element is
d
δσ ≡ σ(λ) = f tr Kσ.
dλ |λ=0
(e) The second variation of the volume element is
d2
δ2σ ≡ σ(λ)
dλ2 |λ=0
 
1 2 i 2

ij

2
= f R̄ − R + K i − Kij K + | grad f | σ,
2
where R and R̄ are the scalar curvatures of g and h.
(f) There is no metric on the 3-torus with positive scalar curva-
ture. (You will need to use the fact that any metric in the
3-torus admits a minimizing 2-torus; this type of idea is used
in the proof of the rigidity statement in the positive mass the-
orem.)
(9) Consider the surfaces Σt obtained from the boundary ∂K ≡ Σ0 of
a compact convex set K ⊂ R3 by flowing a distance t along the
unit normal. Let A(t) and V (t) be the area of Σt and the volume
bounded by Σt .
(a) Show that Z
Ȧ(0) = tr κ,
∂K
where κ is the second fundamental form of Σ.
(b) Prove that Ä(t) = 8π, implying A(t) = 4πt2 + Ȧ(0)t + A(0).
4π 3 Ȧ(0) 2
(c) Conclude that V (t) = 3 t + 2 t + A(0)t + V (0).
 8. EXERCISES 137

(d) Use the isoperimetric inequality A(t)3 ≥ 36πV (t)2 to prove

Minkowski’s inequality:
p
Ȧ(0) ≥ 16πA(0).
 CHAPTER 7

Black holes

This final chapter is devoted to one of the most fascinating predictions

of general relativity: black holes and black hole thermodynamics. Follow-
ing [Tow97] (see also [BCH73, Poi07]), we introduce the Kerr solution,
whose importance is highlighted in the black hole uniqueness theorem. We
then discuss abstract Killing horizons and prove the zeroth, first and second
laws of black hole thermodynamics. Finally, we briefly mention Hawking
radiation and its role in fixing the black hole temperature and entropy.

1. The Kerr solution

General rotating black holes are described by the Kerr metric, given
in the so-called Boyer-Lindquist coordinates by
4M ar sin2 θ ρ2 2
 
2 2M r
ds = − 1 − 2 dt2 − dtdϕ + dr
ρ ρ2 ∆
2M a2 r sin2 θ
 
2 2 2 2
+ ρ dθ + r + a + sin2 θdϕ2 ,
ρ2
where
ρ2 = r2 + a2 cos2 θ,
∆ = r2 − 2M r + a2 ,
and M, a ∈ R are constants. Note that the Schwarzschild metric is a particu-
lar case, corresponding to a = 0. It is possible to prove that the Kerr metric
solves the vacuum Einstein field equations (see for instance [O’N95]).
The Kerr metric is not spherically symmetric, but admits a two-dimensional
∂
group of isometries, generated by the Killing vector fields X = ∂t and
∂
Y = ∂ϕ . The Komar mass associated to X is
1
Z
MKomar = − ⋆dX ♭ ,
8π Σ
where Σ is a 2-surface of constant (t, r), and can be computed to be
MKomar = M.
The expression for the Komar mass in terms of the energy-momentum
tensor, given in Chapter 6, suggests the definition of the Komar angular
139
140 7. BLACK HOLES

momentum as
1
Z
JKomar = ⋆dY ♭
16π Σ
(note the change in sign and absolute value of the constant, due to the
fact that Y is now spacelike and essentially orthogonal to the timelike unit
normal N ). The same exact argument used the Komar mass shows that
JKomar does not depend on the choice of Σ. Performing the calculation for
the Kerr metric yields
JKomar = M a,
and so the parameter a can be interpreted as the angular momentum per
unit mass.
The Killing vector X becomes null on the hypersurface given by the
equation
p
r = M + M 2 − a2 cos2 θ,
known as the ergosphere. However, it is easy to show that the metric
induced on this hypersurface is Lorentzian, and so it cannot be the black
hole event horizon (since it can be crossed both ways by timelike curves).
The event horizon corresponds to the hypersurface r = r+ , where
p
r+ = M + M 2 − a 2 .
Indeed, the function ∆ changes sign on this hypersurface, and so grad r
becomes timelike (meaning that r must decrease along causal curves). Note
that the ergosphere encloses the event horizon, touching it only at the poles,
as shown in Figure 1. The region in between, where X is spacelike, is
called the ergoregion, because matter fields satisfying the dominant energy
condition can have negative energy there, and so extract energy from the
black hole when absorbed (a mechanism known as the Penrose process
in the case of particles or superradiance in the case of fields). Note also
that the existence of an event horizon requires that |a| ≤ M . If |a| < M the
black hole is said to be subextremal, and if |a| = M it is called extremal.

ergosphere event horizon

ergoregion black hole

Figure 1. Spacelike cross-section of the Kerr solution.

 2. KILLING HORIZONS AND THE ZEROTH LAW 141

A time-oriented spacetime (M, g) is said to be asymptotically flat if it

contains an open set I where the metric is well approximated (in a certain
sense which we will not make precise) by the Minkowski metric in the region
{x2 + y 2 + z 2 > R2 }, for sufficiently large R > 0. For such spacetimes we
can define the black hole region as B = M \ J − (I), so that B consists of
the events which cannot send signals to I. If B = 6 ∅, the spacetime is called
a black hole spacetime, and H + = ∂B is called the event horizon.
Finally, an asymptotically flat spacetime is called stationary if it admits a
Killing vector field which is timelike on I.
The importance of the Kerr solution stems from the following result.
Theorem 1.1. (Israel [Isr67], Carter [Car71], Hawking [Haw72],
Robinson [Rob75], Chruściel and Costa [CC08]) The Kerr solution is
the only real-analytic, stationary black hole spacetime satisfying the vacuum
Einstein equations.

2. Killing horizons and the zeroth law

A hypersurface whose normal vectors are null is called a null hyper-
surface. Notice that, unlike what happens for spacelike or timelike hyper-
surfaces (whose normal vectors are timelike or spacelike), the restriction of
the metric to null hypersurfaces is degenerate.
Definition 2.1. A Killing horizon is a null hypersurface which is
orthogonal to a nonvanishing Killing vector field.
Note that the Killing vector field is therefore null on the corresponding
Killing horizon, and tangent to it. Killing horizons are important due to the
following result.
Theorem 2.2. (Hawking [Haw72]) The event horizon of a stationary
black hole spacetime whose matter fields satisfy hyperbolic equations and the
weak energy condition is a Killing horizon.
Proposition 2.3. The integral curves of a nonvanishing normal to a
null hypersurface (e.g. the Killing vector field orthogonal to a Killing hori-
zon) are reparameterized null geodesics.
Proof. Let Z be nonvanishing and normal to a null hypersurface S,
and let p ∈ S. For any tangent vector v ∈ Tp S it is easy to construct a local
vector field V which is tangent to S, commutes with Z and satisfies Vp = v.
We have
1
hV, ∇Z Zi = − h∇Z V, Zi = − h∇V Z, Zi = − V · hZ, Zi = 0.
2
Since p and v are arbitrary, we conclude that ∇Z Z is orthogonal to S, i.e.
∇Z Z = kZ,
for some function k : S → R. 
142 7. BLACK HOLES

Definition 2.4. If H is a Killing horizon associated to the Killing

vector field Z then the function k : H → R such that
∇Z Z = kZ,
on H is called the surface gravity of H relative to Z.
Theorem 2.5. (Zeroth law of black hole thermodynamics) The
surface gravity of a Killing horizon on a spacetime satisfying the dominant
energy condition is a constant function.
Proof. Let Z be a Killing vector field associated to a Killing horizon
H . Since Z is orthogonal to H , we have from the Frobenius theorem that
Z ♭ ∧ dZ ♭ = 0
on H . Because Z ♭ is nonvanishing, and can therefore be completed into a
coframe, we necessarily have
dZ ♭ = 2Z ♭ ∧ U ♭ ⇔ ∇α Zβ = Zα Uβ − Uα Zβ
on H , for some vector field U . If the vector fields X and Y are tangent to
H then
(18) X α Y β ∇α Zβ = 0,
and consequently
(19) Y α Z β ∇α Xβ = −Y α X β ∇α Zβ = 0.
on H . Taking the derivative of (18) along any vector field V tangent to H
yields
V µ X α Y β ∇µ ∇α Zβ = −V µ (∇µ X α )Y β ∇α Zβ − V µ X α (∇µ Y β )∇α Zβ
= −V µ (Y β ∇µ X α + X α ∇µ Y β )(Zα Uβ − Uα Zβ ) = 0,
in view of (19). On the other hand, we saw in Chapter 6 that any Killing
vector field Z satisfies
(20) ∇µ ∇α Zβ = −Rαβµ ν Zν ,
whence
(21) Rαβµν X α Y β V µ Z ν = 0
for any three vector fields X, Y, V tangent to H . If X and Y are orthonormal
then they can always be completed into a local frame {X, Y, Z, W } such that
W is null, orthogonal to X and Y , and normalized against Z,
Zµ W µ = −1.
In this frame the metric is written
gµν = Xµ Xν + Yµ Yν − Zµ Wν − Wµ Zν ,
 2. KILLING HORIZONS AND THE ZEROTH LAW 143

and so we obtain, for any vector field V tangent to H ,

(22) Rαβ V α Z β = Rαµβν V α Z β g µν = −Rαµβν V α Z β Z µ W ν
= V α Z β W ν ∇α ∇β Zν ,
where we used (21) and (20).
We have
(23) k = − hW, ∇Z Zi = −W ν Z µ ∇µ Zν .
Differentiating (23) along a vector field V tangent H yields
V · k = −(V α ∇α W ν )Z µ ∇µ Zν − W ν (V α ∇α Z µ )∇µ Zν − W ν Z µ V α ∇α ∇µ Zν .
The first term in the right-hand side of this equation is
−(V α ∇α W ν )kZν = kW ν V α ∇α Zν = kW ν V α (Zα Uν − Uα Zν ) = kV α Uα ,
whereas the second term is
−W ν V α (Zα U µ − Uα Z µ )∇µ Zν = V α Uα W ν Z µ ∇µ Zν = −kV α Uα .
Therefore these terms cancel out, and, using (22), we obtain
V · k = −Rαβ V α Z β .
From (22) it is clear that
Rαβ Z α Z β = 0,
implying that the vector field
Iα = Rαβ Z β
is tangent to H . By Einstein’s equation, we have
 
1 1
Iα = Rgαβ − Λgαβ + 8πTαβ Z β = RZα − ΛZα + 8πTαβ Z β ,
2 2
where R is the scalar curvature, Λ is the cosmological constant and Tαβ is
the energy-momentum tensor. Therefore the vector field
Jα = Tαβ Z β
is also tangent to H . Since Tαβ satisfies the dominant energy condition, the
vector field J must be causal, and since it is tangent to H it can only be null
and parallel to Z. We conclude that the vector field I is also proportional
to Z, and so
V · k = −Iα V α = 0,
that is, k is constant along H . 
Let λ be a local coordinate that is an affine parameter for the null
geodesics along H , that is,
 
∂ ∂ ∂
, =0 and ∇∂ =0
∂λ ∂λ ∂λ ∂λ
144 7. BLACK HOLES

on H . Then it is easy to see from ∇Z Z = kZ that

∂
Z = k(λ − λ0 )
∂λ
on H , where λ0 may depend on the null geodesic. In other words, Z vanishes
∂
on some cross-section of H . If Z and ∂λ are future-pointing then Z vanishes
to the past when k is positive (that is, λ > λ0 ), and to the future if k is
negative (that is, λ < λ0 ).
If t is a local coordinate in a neighborhood of H such that
∂
Z=
∂t
then we have on H
dλ ∂ ∂
Z= = k(λ − λ0 ) ,
dt ∂λ ∂λ
implying that
λ − λ0 = Cekt ,
where C may depend on the null geodesic. By rescaling λ conveniently we
may assume that
∂
Z = ekt
∂λ
on H .
Now consider a timelike congruence crossing H , with tangent unit time-
like vector field U satisfying
[Z, U ] = 0.
This can be accomplished by taking a timelike hypersurface transverse to
Z, ruled by timelike curves, and moving each curve by the flow of Z. The
quantity
 
∂ D E
E=− , U = − e−kt Z, U
∂λ
represents the energy of a given null geodesic as measured by an observer
of the congruence when crossing the Killing horizon, and is related to the
frequency of the associated wave. Since Z is a Killing field, we have
D E D E D E
Z · E = −LZ e−kt Z, U = − LZ (e−kt Z), U − e−kt Z, LZ U
D E D E D E
= − [Z, e−kt Z], U − e−kt Z, [Z, U ] = ke−kt Z, U = −kE,

implying that
E = E0 e−kt .
In other words, the energy of the null geodesic as measured by the observers
of the congruence decreases exponentially if k > 0 (redshift effect), and
increases exponentially if k < 0 (blueshift effect).
 3. SMARR’S FORMULA AND THE FIRST LAW 145

3. Smarr’s formula and the first law

Unlike the case of the Schwarzschild black hole, where the event horizon
∂
is the Killing horizon corresponding to X = ∂t , the event horizon of the
Kerr black hole is a Killing horizon for the Killing vector field
Z = X + ΩY,
∂
where Y = ∂ϕ and Ω ∈ R is an appropriate constant.
Definition 3.1. Ω is called the angular velocity of the event hori-
zon.
To find Ω, we write the quadratic equation in ω for the vector X + ωY
to be null at some point with r > r+ (so that we can use Boyer-Lindquist
coordinates):
4M ar sin2 θ 2M a2 r sin2 θ
   
2M r 2 2
− 1− 2 − ω+ r +a + sin2 θ ω 2 = 0.
ρ ρ2 ρ2
The discriminant of this equation has the simple form ∆ sin2 θ, and conse-
quently vanishes when we take r = r+ , in which case the quadratic equation
has the single solution
a a
Ω= 2 2
= .
r+ + a 2M r+
This solution is the limit as r → r+ of the values of ω = ω(r, θ) such that
X + ωY is null, and so it must coincide with the angular velocity of the
event horizon.
From the expressions of the Komar mass and angular momentum, it is
clear that
1
Z
M − 2ΩJ = − ⋆dZ ♭
8π Σ
for any compact orientable 2-surface Σ enclosing the event horizon H . Let
us consider the case when Σ is a spacelike cross-section of H (Figure 2).
We can uniquely define a future-pointing unit timelike vector field N and
an unit spacelike vector field n, both orthogonal to Σ, such that Z = N + n.
Because Z is a Killing vector field, ∇µ Zν is a 2-form; more precisely,
(dZ ♭ )µν = ∇µ Zν − ∇ν Zµ = 2∇µ Zν .
If E1 and E2 are two unit vector fields tangent to Σ such that {N, n, E1 , E2 }
is a positive orthonormal frame, and so {−N ♭ , n♭ , E1♭ , E2♭ } is a positive or-
thonormal coframe, then we can expand
∇Z ♭ = −∇Z ♭ (N, n)N ♭ ∧ n♭ + . . . .
Therefore,
1 1
Z Z
M − 2ΩJ = − ⋆∇Z ♭ = ∇Z ♭ (N, n)E1♭ ∧ E2♭ .
4π Σ 4π Σ
146 7. BLACK HOLES

Since
∇Z ♭ (N, n) = ∇Z ♭ (Z, n) = h∇Z Z, ni = hkZ, ni = k,
we finally obtain the Smarr formula:
kA
M= + 2ΩJ,
4π
where A is the area of the cross-section Σ (which in particular is the same
for all cross-sections of H ).

N
Z
H
E1

Figure 2. Spacelike cross-section of the event horizon.

A cross section of the event horizon can be obtained by taking the limit
as r → r+ of a surface of constant (t, r). The induced metric is then
2M a2 r+ sin2 θ
 
2 2 2 2 2 2 2
ds = (r+ + a cos θ)dθ + r+ + a + 2 sin2 θdϕ2 ,
r+ + a2 cos2 θ
and its area is
p
2
A = 4π(r+ + a2 ) = 8π(M 2 + M M 2 − a2 ),
whence r
A 4πJ 2
M= + .
16π A
Noting that M = M (A, J) is homogeneous of degree 1/2, we know from
Euler’s homogeneous function theorem that
1 ∂M ∂M
M= A+ J.
2 ∂A ∂J
On the other hand, it is easy to check that
∂M
= Ω.
∂J
Smarr’s formula then implies the following result:
 4. SECOND LAW 147

Theorem 3.2. (First law of black hole thermodynamics) The

function M = M (A, J) giving the mass of a Kerr black hole as a func-
tion of the area of (spacelike cross-sections of ) its horizon and its angular
momentum satisfies
k
dM = dA + ΩdJ.
8π
This formula provides an easy way to compute the surface gravity of a
Kerr black hole horizon:
1
k= − M Ω2 .
4M
In particular, for an extremal black hole (r+ = a = M ) we have
1
Ω= ⇒ k = 0.
2M
4. Second law
We consider arbitrary test fields propagating on a Kerr background.
Apart from ignoring their gravitational backreaction, we make no further
hypotheses on the fields: they could be any combination of scalar or elec-
tromagnetic fields, fluids, elastic media, or other types of matter. By the
Einstein equation, their combined energy-momentum tensor T must satisfy
∇µ T µν = 0.
Using the symmetry of T and the Killing equation,
∇µ Xν + ∇ν Xµ = 0,
we have
∇µ (T µν Xν ) = 0.
This conservation law suggests that the total field energy on a given spacelike
hypersurface S extending from the black hole event horizon H + to infinity
(Figure 3) should be Z
E= T µν Xν Nµ ,
S
where N is the future-pointing unit normal to S.
Analogously, the total field angular momentum on a spacelike hypersur-
face S extending from the event horizon to infinity is
Z
(24) L = − T µν Yν Nµ ,
S
where the minus sign accounts for the timelike unit normal.
Consider now two such spacelike hypersurfaces, S0 and S1 , with S1 to
the future of S0 (Figure 3). The energy absorbed by the black hole across
the subset H of H + between S0 and S1 is then
Z Z
∆M = T µν Xν Nµ − T µν Xν Nµ ,
S0 S1
148 7. BLACK HOLES

i+

H+ I+
S1
H
i0

S0

H− I−

i−

Figure 3. Penrose diagrams for the region of outer commu-

nication of the Kerr spacetime.

whereas the angular momentum absorbed by the black hole across H is

Z Z
µν
∆J = − T Y ν Nµ + T µν Yν Nµ .
S0 S1
Therefore, we have
Z Z
µν
∆M − Ω∆J = T Zν N µ − T µν Zν Nµ .
S0 S1
Because Z is also a Killing vector field,
∇µ (T µν Zν ) = 0,
and so the divergence theorem, applied to the region bounded by S0 , S1 and
H, yields Z
∆M − Ω∆J = T µν Zν Zµ
H
(we use −Z as the null normal on H). Therefore we have the following
result.
Theorem 4.1. (Second law of black hole thermodynamics, test
field version) If the energy-momentum tensor T corresponding to any col-
lection of test fields propagating on a Kerr background satisfies the null en-
ergy condition at the event horizon then the energy ∆M and the angular
momentum ∆J absorbed by the black hole satisfy
∆M ≥ Ω∆J.
If we think of Kerr black holes as stationary states and imagine that the
interaction of a Kerr black hole with test fields results in a new Kerr black
 4. SECOND LAW 149

hole, then, in view of the first law of black hole thermodynamics, we can
rewrite the result above as

8π
dA = (dM − ΩdJ) ≥ 0,
k

that is, the area of the event horizon of a Kerr black hole can only increase
as a result of its interaction with test fields.
In fact, it is possible to prove a general result about the area of the event
horizon of a black hole spacetime.

Proposition 4.2. The event horizon of a black hole spacetime is ruled

by future-inextendible null geodesics.

Proof. Given the causal structure of Minkowski’s spacetime, it is clear

that J − (I) coincides with I − (I), an open set. Let p ∈ H be any point in the
event horizon and let U ∋ p be a simple neighborhood (see Proposition 4.1
in Chapter 4). Given a sequence {pn } ⊂ U ∩ J − (I) converging to p, let
cn be a future-pointing causal curve connecting pn to I, let qn be the first
intersection of cn with ∂U , and let γn be the future-pointing causal geodesic
connecting pn to qn in U (see Figure 4). Since the exponential map centered
on any point in U is a diffeomorphism, these geodesics converge to a future-
pointing causal geodesic γ in U with initial point p. Note that γ cannot enter
J − (I), because then we would have p ∈ J − (I) ≡ int J − (I), in contradiction
with p ∈ H ≡ ∂J − (I). Moreover, every point in γ is the limit of points
in γn , hence points in J − (I). We conclude that γ is a curve on J − (I) \
int J − (I) = H . Finally, γ cannot be a timelike geodesic, because then
the sequence qn would enter the open set I + (p), and we would again have
p ∈ J − (I). We conclude that γ is a null geodesic. If we extend γ maximally
towards the future we obtain a future-inextendible null geodesic; covering
this curve with simple neighborhoods and applying similar arguments to the
above, one can easily show that it never leaves H . 
150 7. BLACK HOLES

qn

pn

Figure 4. Proof of Proposition 4.2.

Theorem 4.3. (Second law of black hole thermodynamics, Hawk-

ing’s version [Haw72]) If the energy-momentum tensor of a black hole
spacetime satisfies the null energy condition at the event horizon and the
null geodesics ruling the event horizon are complete towards the future then
their expansion is never negative. In particular, the area of any spacelike
cross-section of the event horizon cannot decrease towards the future.
Proof. Note that the null geodesics ruling the event horizon are orthog-
onal to any spacelike cross-section Σ, so that the discussion preceding the
proof of Penrose’s singularity theorem in Chapter 4 applies. Suppose that
the expansion of some null geodesic were negative at some point p ∈ H .
Then, by the analogue of Proposition 6.2 in Chapter 4, that null geodesic
would have a conjugate point to the future of p, after which, by the ana-
logue of Proposition 6.3 in Chapter 4, it would leave H . Since this would
contradict Proposition 4.2, the expansion can never be negative. 

5. Black hole thermodynamics and Hawking radiation

Recall the three laws of thermodynamics for, say, a gas:
• Zeroth law: The temperature is constant throughout the gas when
thermal equilibrium has been reached.
• First law: The internal energy U as a function of the gas entropy
S and volume V satisfies
dU = T dS − pdV,
where T is the temperature and p is the pressure.
 6. EXERCISES 151

• Second law: The entropy of the gas cannot decrease towards the
future.
These are remarkably similar to the three laws of black hole thermody-
namics, if we identify the black hole mass M with the internal energy, (some
multiple of) the horizon’s surface gravity with the black hole’s temperature
and (some multiple of) the horizon’s area with the black hole’s entropy. In-
spired by this analogy, Bekenstein [Bek72] proposed in 1972 that black holes
are indeed thermodynamic systems. Hawking initially resisted this sugges-
tion, since objects in thermal equilibrium at a given temperature must emit
black body radiation, which black holes cannot (classically) do. In 1974,
however, Hawking [Haw74] applied methods of quantum field theory in
curved spacetime to show that black holes do indeed emit particles with a
thermal spectrum corresponding to the temperature1
k
T =
2π
(Hawking radiation). In particular, this fixed the black hole entropy as
A
S= .
4
An elementary discussion of quantum field theory in curved spacetime and
Hawking radiation can be found in [Car03].

6. Exercises
(1) To compute the Komar mass and angular momentum of the Kerr
solution we consider the region r ≫ M, a.
(a) Show that a positive orthonormal coframe is approximately
given in this region by
ω 0 ∼ dt, ω r ∼ dr, ω θ ∼ rdθ,
2M a sin θ
ω ϕ ∼ r sin θdϕ − dt.
r2
(b) Establish the following asymptotic formulas:
2M a sin2 θ
 
♭ 2M
X ∼− 1− dt − dϕ;
r r
2M a sin2 θ
Y♭ ∼− dt + r2 sin2 θdϕ;
r
2M
dX ♭ ∼ 2 ω 0 ∧ ω r + . . . ;
r
6M a sin2 θ 0
dY ♭ ∼ − ω ∧ ωr + . . . .
r2
(c) Prove that MKomar = M and JKomar = M a.
1In units where Boltzmann’s constant and the reduced Planck constant are both set
to 1.
152 7. BLACK HOLES

(2) Show that the metric induced on the ergoshpere is

2M 3 r
ds2 = − 2a sin2 θdtdϕ + dθ2
(r − M )2
+ r2 + a2 + a2 sin2 θ sin2 θdϕ2 ,


√
where r = M + M 2 − a2 cos2 θ. Prove that this metric is Lorentzian.
(3) The symmetry semiaxis θ = 0 is a totally geodesic submanifold of
the Kerr solution, with metric
∆ r 2 + a2 2
ds2 = − dt 2
+ dr .
r 2 + a2 ∆
Obtain the maximal analytical extension of this submanifold. Note
that r = 0 is not a singularity, and so the metric can be continued
for negative values of r.
(4) Consider the static and spherically symmetric metric given in local
coordinates by
ds2 = −V (r)dt2 + V (r)dr2 + r2 dθ2 + sin2 θdϕ2 .

(a) Show that this metric can be written in the form

ds2 = −V (r)dv 2 + 2dvdr + r2 dθ2 + sin2 θdϕ2 ,

where
dr
Z
v =t+ ,
V (r)
(b) Assume that there V has an isolated zero at r = rH , so that
this hypersurface is a Killing horizon. Show that the corre-
∂
sponding surface gravity relative to ∂v is
1
k = V ′ (rH ).
2
(5) Consider a one-parameter family of null geodesics γ(τ, λ) connect-
ing two timelike curves c0 (τ ) = γ(τ, 0) and c1 (τ ) = γ(τ, 1). Show
that  
∂ ∂γ ∂γ
, = 0,
∂λ ∂τ ∂λ
and use this to prove that if τ is the proper time for the curve c0
and Z τ
τ ′ (τ ) = |ċ1 (t)| dt
τ0
is the proper time for the curve c1 then
dτ ′ E0
= ,
dτ E1
where
   
∂γ ċ1 ∂γ
E0 = − ċ0 , and E1 = − ,
∂λ |ċ1 | ∂λ
 6. EXERCISES 153

are the energies of the null geodesic γ(τ, λ) as measured by the

observers corresponding to c0 and c1 .
(6) Compute the area, the angular velocity and the surface gravity of a
Kerr black hole horizon (you may find the relation r+ 2 + a2 = 2M r
+
to be useful here).
(7) Prove that test fields satisfying the null energy condition at the
event horizon cannot destroy an extremal Kerr black hole. More
precisely, prove that if an extremal black hole is characterized by
the physical quantities (M, J), and absorbs energy and angular
momentum (∆M, ∆J) by interacting with the test fields, then the
metric corresponding to the physical quantities (M + ∆M, J + ∆J)
represents, to first order in ∆M and ∆J, either a subextremal or
an extremal Kerr black hole.
(8) Show that if a test field satisfying the null energy condition at the
event horizon extracts energy from a Kerr black hole then a = J/M
always decreases. What fraction of the black hole’s mass can be
extracted?
(9) Check that the second law holds for the black hole resulting from
an Oppenheimer-Snyder collapse.
(10) Use the second law of black hole thermodynamics to:
(a) Prove that a Schwarzschild black hole cannot split into two
Kerr black holes;
(b) Give an upper bound to the energy released in the form of
gravitational waves when two Kerr black holes coalesce to form
a Schwarzschild black hole. Can the efficiency of this process
ever exceed 50%?
 Appendix: Mathematical concepts for physicists

In this appendix we list some mathematical concepts which will be used

in the main text, for the benefit of readers whose background is in physics.

Topology
Definition 6.1. A topological space is a set M with a topology,
that is, a list of the open subsets of M , satisfying:
(1) Both ∅ and M are open;
(2) Any union of open sets is open;
(3) Any finite intersection of open sets is open.
All the usual topological notions can now be defined. For instance, a
closed set is a set whose complement is open. The interior int A of a
subset A ⊂ M is the largest open set contained in A, its closure A is the
smallest closed set containing A, and its boundary is ∂A = A \ int A.
The main object of topology is the study of limits and continuity.
Definition 6.2. A sequence {pn } is said to converge to p ∈ M if for
any open set U ∋ p there exists k ∈ N such that pn ∈ U for all n ≥ k.
Definition 6.3. A map f : M → N between two topological spaces is
said to be continuous if for each open set U ⊂ N the preimage f −1 (U ) is
an open subset of M . A bijection f is called a homeomorphism if both f
and its inverse f −1 are continuous.
A system of local coordinates on a manifold is an example of a homeo-
morphism between the coordinate neighborhood and an open set of Rn .
Two fundamental concepts in topology are compactness and connected-
ness.
Definition 6.4. A subset A ⊂ M is said to be compact if every cover
of A by open sets admits a finite subcover. It is said to be connected if it
is impossible to write A = (A ∩ U ) ∪ (A ∩ V ) with U, V disjoint open sets
and A ∩ U, A ∩ V 6= ∅.
The following result generalizes the theorems of Weierstrass and Bolzano.
Theorem 6.5. Continuous maps carry compact sets to compact sets,
and connected sets to connected sets.
155
156 APPENDIX: MATHEMATICAL CONCEPTS FOR PHYSICISTS

Metric spaces
Definition 6.6. A metric space is a set M and a distance function
d : M × M → [0, +∞) satisfying:
(1) Positivity: d(p, q) ≥ 0 and d(p, q) = 0 if and only if p = q;
(2) Symmetry: d(p, q) = d(q, p);
(3) Triangle inequality: d(p, r) ≤ d(p, q) + d(q, r),
for all p, q, r ∈ M .
The open ball with center p and radius ε is the set
Bε (p) = {q ∈ M | d(p, q) < ε}.
Any metric space has a natural topology, whose open sets are unions of open
balls. In this topology pn → p if and only if d(pn , p) → 0, F ⊂ M is closed
if and only if every convergent sequence in F has limit in F , and K ⊂ M is
compact if and only if every sequence in K has a sublimit in K.
A fundamental notion for metric spaces is completeness.
Definition 6.7. A sequence {pn } in M is said to be a Cauchy se-
quence if for all ε > 0 there exists N ∈ N such that d(pn , pm ) < ε for all
m, n > N . A metric space is said to be complete if all its Cauchy sequences
converge.
In particular, any compact metric space is complete.

Hopf-Rinow theorem
Definition 6.8. A Riemannian manifold is said to be geodesically
complete if any geodesic is defined for every value of its parameter.
Definition 6.9. Let (M, g) be a connected Riemannian manifold and
p, q ∈ M . The distance between p and q is defined as
d(p, q) = inf{l(γ) | γ is a smooth curve connecting p to q}.
It is easily seen that (M, d) is a metric space. Remarkably, the com-
pleteness of this metric space is equivalent to geodesic completeness.
Theorem 6.10. (Hopf-Rinow) A connected Riemannian manifold (M, g)
is geodesically complete if and only if (M, d) is a complete metric space.

Differential forms
Definition 6.11. A differential form ω of degree k is simply a com-
pletely anti-symmetric k-tensor: ωα1 ···αk = ω[α1 ···αk ] .
For instance, covector fields are differential forms of degree 1. Differential
forms are useful because of their rich algebraic and differential structure.
 DIFFERENTIAL FORMS 157

Definition 6.12. If ω is a k-form and η is an l-form then their exterior

product is the (k + l)-form
(k + l)!
(ω ∧ η)α1 ···αk β1 ···βl =ω[α1 ···αk ηβ1 ···βl ] ,
k! l!
and the exterior derivative of ω is the (k + 1)-form
(dω)αα1 ···αk = (k + 1)∇[α ωα1 ···αk ] ,
where ∇ is any symmetric connection.
It is easy to see that any k-form ω is given in local coordinates by
X
(25) ω= ωα1 ···αk (x)dxα1 ∧ · · · ∧ dxαk ,
α1 <···<αk
n


and therefore has k independent components on an n-dimensional mani-
fold.
Proposition 6.13. If ω, η and θ are differential forms then:
(1) ω ∧ (η ∧ θ) = (ω ∧ η) ∧ θ;
(2) ω ∧ η = (−1)(deg ω)(deg η) η ∧ ω;
(3) ω ∧ (η + θ) = ω ∧ η + ω ∧ θ;
(4) d(ω + η) = dω + dη;
(5) d(ω ∧ η) = dω ∧ η + (−1)deg ω ω ∧ dη;
(6) d2 ω = 0.
It is clear from these properties that if the k-form ω is given in local
coordinates by (25) above then
X X
dω = ∂α ωα1 ···αk dxα ∧ dxα1 ∧ · · · ∧ dxαk .
α1 <···<αk α

The last property in Proposition 6.13 has a converse, known as the

Poincaré Lemma.
Lemma 6.14. (Poincaré) If dω = 0 then locally ω = dη.
A related result is the Frobenius theorem. Here we present a partic-
ular case of this result.
Theorem 6.15. (Frobenius) The nonvanishing 1-form ω is locally or-
thogonal to a family of hypersurfaces if and only if ω ∧ dω = 0.
To prove the easy direction in this equivalence it suffices to note that ω
is locally orthogonal to a family of hypersurfaces {f = constant} if and only
if ω = gdf for some nonvanishing function g. Note that in particular this is
always true for 1-forms in 2-dimensional manifolds.
We will now assume that our n-dimensional manifold is oriented, that
is, that an orientation can be, and has been, consistently chosen on every
tangent space. Any n-form ω is written in local coordinates as
ω = a(x)dx1 ∧ · · · ∧ dxn .
158 APPENDIX: MATHEMATICAL CONCEPTS FOR PHYSICISTS

If the coordinate system is positive, that is, if the coordinate basis {∂1 , . . . , ∂n }
has positive orientation at all points, we define
Z Z
ω= a(x)dx1 . . . dxn ,
U x(U )

where U is the coordinate neighborhood. This formula does not depend on

the choice of local coordinates because dx1 ∧ · · · ∧ dxn transforms by the
determinant of the change of variables.
Theorem 6.16. (Stokes) If M is an oriented n-dimensional manifold
with boundary ∂M and ω is an (n − 1)-form then
Z Z
dω = ω.
M ∂M
In this theorem the orientations of M and ∂M are related as follows:
if ∂M is a level set of x1 in a positive coordinate system and ∂1 points
outwards then the coordinate system (x2 , . . . , xn ) on ∂M is positive.
If M has a metric g then its volume element is the n-form ǫ which
is 1 when contracted with a positive orthonormal frame. In positive local
coordinates we have
q
ǫ = | det(gµν )| dx1 ∧ · · · ∧ dxn .
It is easily seen that ∇ǫ = 0, where ∇ is the Levi-Civita connection. If ω is
a k-form then its Hodge dual is the (n − k)-form ⋆ω given by
1 α1 ···αk
(⋆ω)β1 ···βn−k = ω ǫα1 ···αk β1 ···βn−k .
k!
The operator ⋆, called the Hodge star, can alternatively be defined as follows:
if {ω 1 , . . . , ω n } is any positively oriented orthonormal coframe (so that the
volume element is ǫ = ω 1 ∧ . . . ∧ ω n ) then
⋆(ω 1 ∧ . . . ∧ ω k ) = g(ω 1 , ω 1 ) · · · g(ω k , ω k ) ω k+1 ∧ . . . ∧ ω n
= ǫ((ω 1 )♯ , . . . , (ω k )♯ , . . .).

Lie derivative
A vector field X can be identified with the differential operator that cor-
responds to taking derivatives along X. In local coordinates, this operator
is given by
X · f = X µ ∂µ f.
It turns out that the commutator of two vector fields X and Y , regarded as
differential operators, is also a vector field:
[X, Y ] · f = X · (Y µ ∂µ f ) − Y · (X µ ∂µ f )
= (X · Y µ )∂µ f + Y µ X ν ∂ν ∂µ f − (Y · X µ )∂µ f − X µ Y ν ∂ν ∂µ f
= (X · Y µ − Y · X µ )∂µ f.
 LIE DERIVATIVE 159

Definition 6.17. The Lie bracket of two vector fields X and Y is the
vector field
[X, Y ] = (X · Y µ − Y · X µ )∂µ .
This operation is intimately related with the exterior derivative.
Proposition 6.18. If ω is a 1-form then
dω(X, Y ) = X · ω(Y ) − Y · ω(X) − ω([X, Y ])
for all vector fields X and Y .
Proof. In local coordinates we have
dω(X, Y ) = (∂µ ων − ∂ν ωµ )X µ Y ν = Y ν X · ων − X ν Y · ων
= X · (ων Y ν ) − ων X · Y ν − Y · (ων X ν ) + ων Y · X ν
= X · ω(Y ) − Y · ω(X) − ων (X · Y ν − Y · X ν ).

If the vector field X is nonzero at some point p then there exists a
coordinate system defined in a neighborhood of p such that X = ∂1 . In
fact, we just have to fix local coordinates (x2 , . . . , xn ) on a hypersurface Σ
transverse to X at p and let x1 be the parameter for the flow of X starting
at Σ. If T is any tensor, we define its Lie derivative along X as the tensor
with components
(LX T )αβ11...β
···αk
l
= ∂1 Tβα11...β
···αk
l
.
This can be extended to points where X vanishes by continuity. Although
this definition seems to depend on the coordinate system, it is actually
invariant. To check this, we just have to find an invariant expression for the
Lie derivative of functions, vector fields and 1-forms and then notice that
the Leibniz rule applies.
Proposition 6.19. If X is a vector field then:
(1) LX f = X · f for functions f ;
(2) LX Y = [X, Y ] for vector fields Y ;
(3) LX ω = Xy dω + d(Xy ω) for 1-forms ω
(where y means contraction in the first index).
Proof. The formula for functions is immediate. In the coordinate sys-
tem where X = ∂1 ,
LX Y = ∂1 Y µ ∂µ = (X · Y µ − Y · X µ )∂µ = [X, Y ].
Finally, we have
(Xy dω + d(Xy ω))(Y ) = dω(X, Y ) + Y · ω(X) = X · ω(Y ) − ω([X, Y ])
= LX (ω(Y )) − ω(LX Y ) = (LX ω)(Y ),
where we used the Leibniz rule. This formula is sometimes called Cartan’s
magic formula. 
160 APPENDIX: MATHEMATICAL CONCEPTS FOR PHYSICISTS

Cartan structure equations

Let {Eµ } be an orthonormal frame, and {ω µ } the corresponding or-
thonormal coframe, so that
ω µ (Eν ) = δ µν .
Note that the metric can be written as
ds2 = ηµν ω µ ⊗ ω ν ,
where (ηµν ) = diag(−1, 1, 1, 1) is the flat space metric (which we will use to
raise and lower indices).
Definition 6.20. The connection forms associated to the orthonor-
mal frame {Eµ } are the 1-forms ω µν such that
∇X Eν = ω µν (X)Eµ
for all vector fields X. The curvature forms associated this frame are the
2-forms Ωµ ν such that
R(X, Y )Eν = Ωµ ν (X, Y )Eµ
for all vector fields X, Y .
Note that the components of the Riemann tensor in the orthonormal
frame can be retrieved from the curvature forms by noticing that
Ωµ ν = Rαβ µν ω α ⊗ ω β = Rαβ µν ω α ∧ ω β .
X

α<β

These forms can be computed by using the so-called Cartan structure

equations. This is by far the most efficient way to compute the curvature.
Theorem 6.21. The connection forms are the unique solution of Car-
tan’s first structure equations
(
ωµν = −ωνµ
,
dω µ + ω µν ∧ ω ν = 0
and the curvature forms are given by Cartan’s second structure equa-
tions
Ωµ ν = dω µν + ω µα ∧ ω αν .
Proof. The first condition is equivalent to
X · hEµ , Eν i = 0
for all vector fields X, which in turn is equivalent to the compatibility of
the connection with the metric. Using
dω µ (X, Y ) = X · ω µ (Y ) − Y · ω µ (X) − ω µ ([X, Y ])
for all vector fields X and Y , it is easy to see that the second condition is
equivalent to
[Eα , Eβ ] = ∇Eα Eβ − ∇Eβ Eα ,
 CARTAN STRUCTURE EQUATIONS 161

which in turn is equivalent to the symmetry of the connection. Since the

Levi-Civita connection is the only connection which is symmetric and com-
patible with the metric, we conclude that Cartan’s first structure equations
have a unique solution.
Finally, the third condition can be derived by writing
Ωµ ν (X, Y )Eµ = R(X, Y )Eν = ∇X ∇Y Eν − ∇Y ∇X Eν − ∇[X,Y ] Eν
in terms of the connection forms. 
 Bibliography

[AF03] R. Adams and J. Fournier, Sobolev Spaces, Academic Press, 2003.

[ADM61] R. Arnowitt, S. Deser, and C. Misner, Coordinate invariance and energy ex-
pressions in general relativity, Phys. Rev. 122 (1961), 997–1006.
[AMA79] A. Ashtekar and A. Magnon-Ashtekar, On conserved quantities in general rel-
ativity, J. Math. Phys. 20 (1979), 793–800.
[Bar86] R. Bartnik, The mass of an asymptotically flat manifold, Commun. Pure Appl.
Math. 39 (1986), 661–693.
[BCH73] J. Bardeen, B. Carter, and S. Hawking, The four laws of black hole mechanics,
Commun. Math. Phys. 31 (1973), 161–170.
[Bek72] J. Bekenstein, Black holes and the second law, Nuovo Cim. Lett. 4 (1972),
737–740.
[BS03] A. Bernal and M. Sánchez, On smooth Cauchy hypersurfaces and Geroch’s
splitting theorem, Commun. Math. Phys. 243 (2003), 461–470.
[BS05] A. Bernal and M. Sánchez, Smoothness of time functions and the metric split-
ting of globally hyperbolic spacetimes, Commun. Math. Phys. 257 (2005), 43–50.
[BS06] A. Bernal and M. Sánchez, Further results on the smoothability of Cauchy hy-
persurfaces and Cauchy time functions, Lett. Math. Phys. 77 (2006), 183–197.
[Boo03] W. Boothby, An introduction to differentiable manifolds and Riemannian ge-
ometry, Academic Press, 2003.
[Bra01] H. Bray, Proof of the Riemannian Penrose inequality using the positive mass
theorem, J. Diff. Geom. 59 (2001), 177–267.
[Bra11] , On the positive mass, Penrose, and ZAS inequalities in general dimen-
sion, arXiv:1101.2230 (2011).
[Car71] B. Carter, Axisymmetric black hole has only two degrees of freedom, Phys. Rev.
Lett. 26 (1971), 331–.
[Car03] S. Carroll, Spacetime and geometry: An introduction to general relativity, Pear-
son, 2003.
[CBG69] Y. Choquet-Bruhat and R. Geroch, Global aspects of the Cauchy problem in
general relativity, Commun. Math. Phys. 14 (1969), 329–335.
[CC08] P. Chruściel and J. Costa, On uniqueness of stationary vacuum black holes,
Astérisque 321 (2008), 195–265.
[dC93] M. do Carmo, Riemannian geometry, Birkhäuser, 1993.
[FB55] Y. Fourès-Bruhat, Théorème d’existence pour certains systèmes d’équations aux
derivées partielles non linéaires, Acta Math. 88 (1955), 141–225.
[GN14] L. Godinho and J. Natário, An introduction to Riemannian geometry: With
applications to mechanics and relativity, Springer, 2014.
[Haw66] S. Hawking, The occurrence of singularities in cosmology, Proc. Roy. Soc. Lon.
A 294 (1966), 511–521.
[Haw72] , Black holes in general relativity, Commun. Math. Phys. 25 (1972),
152–166.
[Haw74] , Black holes explosions?, Nature 248 (1974), 30–31.
[HE95] S. Hawking and G. Ellis, The large scale structure of space-time, Cambridge
University Press, 1995.
163
164 BIBLIOGRAPHY

[HI01] G. Huisken and T. Ilmanen, The inverse mean curvature flow and the Rie-
mannian Penrose inequality, J. Diff. Geom. 59 (2001), 353–437.
[Isr67] W. Israel, Event horizons in static vacuum space-times, Phys. Rev. 164 (1967),
1776–1779.
[KN96] S. Kobayashi and K. Nomizu, Foundations of differential geometry, vol. I and
II, Wiley, 1996.
[Lam10] M.-K. Lam, The graphs cases of the Riemannian positive mass and Penrose
inequalities in all dimensions, arXiv:1010.4256 (2010).
[Mar09] M. Mars, Present status of the Penrose inequality, Class. Quant. Grav. 26
(2009), 193001.
[MTW73] C. Misner, K. Thorne, and J. A. Wheeler, Gravitation, Freeman, 1973.
[Mun00] J. Munkres, Topology, Prentice-Hall, 2000.
[Nab88] G. Naber, Spacetime and singularities – An introduction, Cambridge University
Press, 1988.
[O’N83] B. O’Neill, Semi-Riemannian geometry, Academic Press, 1983.
[O’N95] , The geometry of Kerr black holes, A K Peters, 1995.
[Pen65] R. Penrose, Gravitational collapse and space-time singularities, Phys. Rev. Lett.
14 (1965), 57–59.
[Pen87] , Techniques of differential topology in relativity, Society for Industrial
and Applied Mathematics, 1987.
[Poi07] E. Poisson, A relativist’s toolkit: The mathematics of black-hole mechanics,
Cambridge University Press, 2007.
[Rin09] H. Ringström, The Cauchy problem in general relativity, European Mathemat-
ical Society, 2009.
[Rob75] D. Robinson, Uniqueness of the Kerr black hole, Phys. Rev. Lett. 34 (1975),
905–906.
[Sbi16] J. Sbierski, On the existence of a maximal Cauchy development for the Einstein
equations: A dezornification, Ann. Henri Poincaré 17 (2016), 301–329.
[SY81] R. Schoen and S.-T. Yau, Proof of the positive mass theorem ii, Commun. Math.
Phys. 79 (1981), 231–260.
[Sop76] D. Soper, Classical Field Theory, Wiley, 1976.
[Tow97] P. Townsend, Black holes, arXiv:gr-qc/9707012 (1997).
[Wal84] R. Wald, General relativity, University of Chicago Press, 1984.
 Index

action, 114 Cauchy

Einstein, 134 hypersurface, 60, 95
Einstein-Hilbert, 120 maximal development, 102
ADM mass, 128 problem, 95
angular momentum, 140 sequence, 94, 156
angular velocity of event horizon, 145 Cauchy-Kowalewski theorem, 94
anti-de Sitter universe, 31, 50 causal
anti-trapped surface, 85 continuous curve, 76
asymptotically flat future, 55, 82
region, 26 past, 55, 82
Riemannian manifold, 128 vector, 7
spacetime, 141 chain, 101
chart at infinity, 128
Bianchi identity Choquet-Bruhat theorem, 102
contracted, 13, 99, 101 Christoffel symbols, 10
first, 12 chronological
second, 12 future, 55, 82
bifurcate sphere, 25 past, 55, 82
Big Bang, 29, 33 chronology condition, 59
Big Crunch, 33 closed set, 155
Birkhoff theorem, 45 closure, 155
black hole, 26, 141 compact set, 75, 155
eternal, 27 complete metric space, 156
extremal, 140 condition
spacetime, 141 chronology, 59
subextremal, 140 interior cone, 94
uniqueness theorem, 141 congruence, 63
black hole thermodynamics conjugate point, 67
first law, 147 to a 2-surface, 80
second law, 148, 150 to a spacelike hypersurface, 72
zeroth law, 142 connected set, 155
blueshift effect, 144 connection
boundary, 155 forms, 160
Boyer-Lindquist coordinates, 139 Levi-Civita, 10
Buchdahl’s limit, 53 constraint equations, 96, 98, 102
continuous
canonical energy-momentum tensor, 114 causal curve, 76
Cartan map, 155
magic formula, 40, 159 contracted Bianchi identity, 13, 99, 101
structure equations, 160 convergent sequence, 155
165
166 INDEX

coordinates energy-momentum tensor, 13, 123

Boyer-Lindquist, 139 canonical, 114
harmonic, 99 dust, 14
isothermal, 20 electromagnetic field, 116
normal, 56 perfect fluid, 15, 119
null, 17 scalar field, 15, 91, 115, 134
cosmological constant, 14 equation
cosmology, 27 Jacobi, 64
curvature, 10 Klein-Gordon, 91
extrinsic, 39 Poisson, 13
forms, 160 Raychaudhuri, 66
scalar, 13 equations
curve constraint, 96, 98, 102
continuous causal, 76 Einstein, 14
future-directed, 55 Euler-Lagrange, 114
future-inextendible, 60 evolution, 96, 98
past-directed, 55 Friedmann, 29
past-inextendible, 60 Maxwell, 46, 96, 116
ergoregion, 140
d’Alembertian ergosphere, 140
Hodge, 111 Euler-Lagrange equations, 114
rough, 111 event horizon, 22, 141
de Sitter universe, 30 angular velocity, 145
density area, 146
Hamiltonian, 114 surface gravity, 142
Lagrangian, 114 evolution equations, 96, 98
differential form, 156 expansion
distance, 156 of a null congruence, 81
divergence, 89 of a timelike congruence, 65
theorem, 89 exponential map, 56, 80
domain of dependence, 60 exterior derivative, 157
future, 60 exterior product, 157
past, 60 extremal black hole, 140
dominant energy condition, 66 extrinsic curvature, 39, 98
dust, 14
Faraday tensor, 98
Einstein field
action, 134 electromagnetic, 46, 98, 116
field equations, 14 Jacobi, 64
reduced equations, 100 scalar, 15, 91, 115
tensor, 13 first Bianchi identity, 12
universe, 36 first law, 147
Einstein-Hilbert action, 120 Flamm paraboloid, 45, 132
elastic law, 118 FLRW metrics, 27
quasi-Hookean, 119 form
electromagnetic conection, 160
field, 46, 98, 116 curvature, 160
potential one-form, 98 formula
energy condition Cartan, 40, 159
dominant, 66 Smarr, 146
null, 66 Weitzenbock, 111
strong, 66 Friedmann equations, 29
weak, 66 Frobenius theorem, 142, 157
 INDEX 167

function quasi-linear, 95
distance, 156 hypersurface
lapse, 99 Cauchy, 60
upper semicontinuous, 76 null, 141
future spacelike, 72
causal, 55, 82 timelike, 37
chronological, 55, 82
domain of dependence, 60 inequality
future-directed curve, 55 Minkowski, 132, 137
future-inextendible curve, 60 Sobolev, 94
future-pointing vector, 7 infinity
null, 18
Galileo transformation, 5 spacelike, 18
gauge timelike, 18
Lorentz, 97 interior, 155
potentials, 96 interior cone condition, 94
transformation, 97 isothermal coordinates, 20
transverse traceless, 125
Jacobi
Gauss Lemma
equation, 64
first version, 38
field, 64
second version, 39
John material, 120
Gauss-Bonnet theorem, 123
geodesically complete manifold, 156 Karlovini-Samuelsson material, 120
geodesically convex neighborhood, 56 Kerr metric, 139
geometrized units, 13 Killing horizon, 141
global time function, 59 Klein-Gordon equation, 91
globally hyperbolic spacetime, 60 Komar
gravitational angular momentum, 140
potential, 13 mass, 110
wave, 124 Kottler metric, 46
Kretschmann scalar, 22
Hamiltonian, 115
density, 114 Lagrangian density, 114
harmonic coordinates, 99 elastic, 118
Hausdorff electromagnetic field, 116
manifold, 102 scalar field, 115
metric, 76 lapse function, 99
Hawking Lemma
radiation, 151 Gauss, 38, 39
singularity theorem, 79 Poincaré, 157
Hodge Zorn, 102
d’Alembertian, 111 Levi-Civita connection, 10
dual, 158 Lichnerowicz method, 102
homeomorphism, 155 Lie bracket, 159
horizon light cone, 6
event, 22, 141 lightlike vector, 6
Killing, 141 linear hyperbolic system, 95
Hubble Lorentz
constant, 28 gauge, 97
law, 28 transformation, 5
Huisken-Ilmanen theorem, 131 Lorentzian
hyperbolic system manifold, 10
linear, 95 metric, 10
168 INDEX

manifold energy condition, 66

geodesically complete, 156 hypersurface, 141
Hausdorff, 102 infinity, 18
Lorentzian, 10 vector, 6
oriented, 157
map open
continuous, 155 ball, 156
exponential, 56, 80 set, 155
marginally trapped surface, 131 oriented manifold, 157
mass
past
ADM, 128
causal, 55, 82
Komar, 110
chronological, 55, 82
material
domain of dependence, 60
homogeneous and isotropic, 118
past-directed curve, 55
John, 120
past-inextendible curve, 60
Karlovini-Samuelsson, 120
past-pointing vector, 7
maximal
Penrose
analytical extension, 25
diagram, 18
Cauchy development, 102
process, 140
Maxwell equations, 46, 96, 116
singularity theorem, 84
metric
perfect fluid, 15
anti-de Sitter, 31, 50
Poincaré Lemma, 157
de Sitter, 30
Poisson equation, 13
Einstein, 36
positive mass theorem, 129
FLRW, 27
potential
Hausdorff, 76
gauge, 96
Kerr, 139
gravitational, 13
Kottler, 46
one-form, 98
Lorentzian, 10
principal pressures, 67
Reissner-Nordström, 47
proper time, 7
Schwarzschild, 21
purely spatial tensor, 63
spatial, 65
metric space, 156 quasi-Hookean elastic law, 119
compact, 76 quasi-linear hyperbolic system, 95
complete, 156
Milne universe, 29 Raychaudhuri equation, 66
Minkowski redshift effect, 144
inequality, 132, 137 reduced
spacetime, 6 Einstein equations, 100
Myers theorem, 87 Ricci tensor, 100
Reissner-Nordström metric, 47
neighborhood relaxed configuration, 117
geodesically convex, 56 rest energy density, 67, 118
normal, 56 Ricci tensor, 13
simple, 74 reduced, 100
norm rigidity modulus, 119
Sobolev, 93 rough d’Alembertian, 111
supremum, 94
normal scalar
coordinates, 56 curvature, 13
neighborhood, 56 field, 15, 91, 115
null Kretschmann, 22
coordinates, 17 shear, 119
 INDEX 169

Schwarzschild metric, 21 Einstein, 13

second Bianchi identity, 12 energy-momentum, 13, 123
second fundamental form, 39, 98 Faraday, 98
of a congruence, 63 Ricci, 13
second law, 148, 150 theorem
set Birkhoff, 45
closed, 155 black hole uniqueness, 141
compact, 75, 155 Cauchy-Kowalewski, 94
connected, 155 Choquet-Bruhat, 102
open, 155 divergence, 89
shear, 65 Frobenius, 142, 157
scalar, 119 Gauss-Bonnet, 123
shift vector, 99 Hawking, 79
simple neighborhood, 74 Huisken-Ilmanen, 131
singular spacetime, 78 Myers, 87
singularity theorem Penrose, 84
Hawking, 79 positive mass, 129
Penrose, 84 time-orientable
Smarr formula, 146 double covering, 55
Sobolev spacetime, 55
inequality, 94 time-oriented spacetime, 55
norm, 93 timelike
space hypersurface, 37
metric, 156 infinity, 18
topological, 155 vector, 6
spacelike topological space, 155
hypersurface, 72 topology, 155
infinity, 18 transformation
vector, 6 Galileo, 5
spacetime Lorentz, 5
asymptotically flat, 141 transverse traceless gauge, 125
black hole, 141 trapped surface, 84
globally hyperbolic, 60 twin paradox, 14, 58
Minkowski, 6
singular, 78 universe
stably causal, 59 anti-de Sitter, 31, 50
static, 109 de Sitter, 30
stationary, 110, 141 Einstein, 36
spatial metric, 65 Milne, 29
stably causal spacetime, 59 unsheared energy density, 119
static spacetime, 109 upper semicontinuous function, 76
stationary spacetime, 110, 141
strong energy condition, 66 vector
subextremal black hole, 140 causal, 7
superradiance, 140 future-pointing, 7
supremum norm, 94 lightlike, 6
surface null, 6
anti-trapped, 85 past-pointing, 7
marginally trapped, 131 shift, 99
trapped, 84 spacelike, 6
surface gravity, 142 timelike, 6
volume element, 158
tensor vorticity, 65
170 INDEX

wave
equation, 15
gravitational, 124
weak cosmic censorship conjecture, 131
weak energy condition, 66
Weitzenbock formula, 111
white hole, 26
wormhole, 26

zeroth law, 142

Zorn Lemma, 102