See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/360187899

Lecture Notes on General Relativity

Preprint · April 2022

CITATIONS READS

0 1,242

1 author:

Xiao-Kan Guo

33 PUBLICATIONS 74 CITATIONS

SEE PROFILE

All content following this page was uploaded by Xiao-Kan Guo on 23 August 2024.

The user has requested enhancement of the downloaded file.

 Lecture Notes on General Relativity
Xiao-Kan Guo (郭肖侃) ∗

August 24, 2024

Abstract

These are the lecture notes prepared for two introductory courses on General Rela-
tivity, one for graduate students and the other for undergraduate students, that I taught
at Bohai University in 2022. The purpose of these notes is twofold: i) to present the
fundamental mathematical structure of General Relativity in a concise but concrete way
so that the undergraduate students could follow the calculations; ii) to introduce some
useful advanced results, so that the graduate students would be ready for delving into
more advanced topics in Gravitation.
These notes are based mainly on the elegant book by Hughston & Tod [1], sup-
plemented by the books [2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14] and other articles.
The suggested textbooks in Chinese are: [11, 12] for graduate students and [15, 16] for
undergraduate students.

] Q

∗
E-mail: [email protected]
Contents
1 Introduction 4

2 Vectors and tensors in flat three-space 7

2.1 Electromagnetism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 Non-relativistic fluid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.3 Coordinate bases and tensors . . . . . . . . . . . . . . . . . . . . . . . . . . 11

3 Special relativistic geometry 14

3.1 Relativistic Kinematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.2 Lorentz contraction and time dilation . . . . . . . . . . . . . . . . . . . . . 16
3.3 Relativistic electrodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.4 Relativistic hydrodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

4 Tensors on manifolds 20
4.1 Covariant derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
4.2 Riemann Tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.3 Riemannian geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.4 Lie derivatives and Killing vectors . . . . . . . . . . . . . . . . . . . . . . . . 28

5 Geodesics 31
5.1 Geodesic deviation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
5.2 Raychaudhuri equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
5.3 Fermi-Walker transport and non-geodesic curves . . . . . . . . . . . . . . . 38
5.4 Spinning test particles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

6 Einstein’s field equations 41

6.1 Einstein’s equations from geodesic deviation . . . . . . . . . . . . . . . . . . 42
6.2 Einstein-Hilbert action . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
6.3 Energy conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
6.4 Hamiltonian formalism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

7 Cartan method and tetrad formalism 53

7.1 Differential forms and Cartan method . . . . . . . . . . . . . . . . . . . . . 53
7.2 Tetrad formalism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
7.3 Energy, momentum, and angular momentum . . . . . . . . . . . . . . . . . 58
7.4 A little bit Newman-Penrose formalism . . . . . . . . . . . . . . . . . . . . 60

8 Weak gravitational fields 61

8.1 Linearized gravity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
8.2 The slow-motion approximation . . . . . . . . . . . . . . . . . . . . . . . . 62
8.3 Gravitational waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
8.4 Gravitational radiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
8.5 Gravitoelectromagnetism . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
8.6 Gravitational wave memory . . . . . . . . . . . . . . . . . . . . . . . . . . 68

1
9 The Schwarzschild solution 70
9.1 Gravitational red-shift . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
9.2 Geodesics in the Schwarzschild spacetime . . . . . . . . . . . . . . . . . . . 74
9.3 The extended Schwarzschild solution . . . . . . . . . . . . . . . . . . . . . . 78
9.4 Interior solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
9.5 Birkhoff’s theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

10 The Kerr solution 87

10.1 Boyer-Lindquist coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . 88
10.2 Horizons, Penrose diagram and ergosphere . . . . . . . . . . . . . . . . . . 89
10.3 Geodesics in the Kerr spacetime . . . . . . . . . . . . . . . . . . . . . . . . . 93
10.4 Kerr-Eddington coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . 96
10.5 Kerr-Schild coordinates and Doran form . . . . . . . . . . . . . . . . . . . . 96
10.6 Newman-Penrose formalism for the Kerr solution . . . . . . . . . . . . . . 98

11 Charged solutions 98
11.1 The Reissner-Nordström solution . . . . . . . . . . . . . . . . . . . . . . . . 98
11.2 The Kerr-Newman solution . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

12 General properties of black holes 102

12.1 Causal structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
12.2 Horizons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
12.3 Surface gravity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
12.4 Near-horizon geometries . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
12.5 Gravitational charges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
12.6 Black hole thermodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
12.7 Hawking radiation: a tunneling glimpse . . . . . . . . . . . . . . . . . . . . 113
12.8 Geometric inequalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

13 Cosmology 117
13.1 Homogeneous and isotropic three-spaces . . . . . . . . . . . . . . . . . . . 117
13.2 Kinematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
13.3 Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
13.4 Inflation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
13.5 Anisotropic cosmologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

A More Newman-Penrose 134

A.1 Einstein equations in Newman-Penrose formalism . . . . . . . . . . . . . . 134
A.2 Asymptotic behavior and Bondi coordinates . . . . . . . . . . . . . . . . . . 137
A.3 BMS group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140

B Teleparallel equivalent of general relativity 143

B.1 Tetrads as reference frames . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
B.2 Lagrangian formulation of TEGR . . . . . . . . . . . . . . . . . . . . . . . 144
B.3 Symmetric teleparallel general relativity . . . . . . . . . . . . . . . . . . . . 145

2
C Black hole shadows 147
C.1 Angular size of the shadow . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
C.2 Shadow of a Kerr black hole . . . . . . . . . . . . . . . . . . . . . . . . . . . 150

Acknowledgments
Special thanks are given to Bekir Baytas who carefully read these notes and provided various
helpful suggestions. I also would like to thank Ping Li (李平), Chun-Yen Lin (林君彥),
Shupeng Song (宋術鵬) and Chun-Yan Wang (王春艷) for helpful discussions on the GR
courses.
I learned various topics relevant to GR from many professors, through courses (offline
or online) or conversations, and they are gratefully acknowledged here. They are: Bin
Chen (陳斌), Chao-Jun Feng (馮朝君), Xin-Zhou Li (李新洲) , Yongge Ma (馬永革),
Neil Turok, and Hongsheng Zhang (張宏升).

3
1 Introduction
General Relativity (GR) is Einstein’s theory of gravitation; it is also a theory of spacetime.
By spacetime we mean a single entity, a union of space and time; in mathematical terms,
spacetime is a Riemannian manifold. The important point is that GR incorporates the grav-
itational field into the structure of spacetime itself. Since the gravitation can vary from place
to place or from time to time, spacetime also changes, instead of being fixed.
(A quick impression of spacetime can be gained as follows. Recall Newton’s second law
of motion, F = ma, which divides the world into two parts, i.e forces and motions. The
motions are futhermore divided into acceleration a and the resistance to acceleration, the
inertial mass m. However, in the Newtonian gravity, we have ma = −GM m/r2 , where
the inertial and the gravitational parts cancel according to experiments. So in gravity, no
forces are needed; instead, one needs accelerations along trajectories, which in turn simply
means geometry! Moreover, the accelerations in a gravitational field should be same (Pisa!),
but we can give the particles different initial velocities. For example, we can drop a coin to
the floor, or throw it straight up and allow it to fall. In either case, it will start and end at
the same position. But the coin will reach the floor at different times, so they have different
trajectories in spacetime!)
The experimental verifications of GR have been so far very precise to a large extent,
thereby telling us that GR is (in most cases) a physically true theory of the real world.
However, several subtle problems, such as the quantization of gravity and the nature of
dark energy and dark matter, are still beyond the scope of GR. As a physical theory, GR
therefore represents our most successful understandings of gravitation and spacetime and at
the same time guiding us towards those new physics ahead. On the other hand, GR in itself
is a mathematical subject in differential geometry, which is of great mathematical interest.
The definitive form of GR were given in the ingenious work of Albert Einstein in 1915.
Since then, many brilliant physicists and mathematicians contribute to our further under-
standings of GR. A chronological (incomplete) list of important developments in GR is as
follows.

• 1915, Einstein formulated GR

• 1916, Schwarzschild solution

• 1919, Eddington observed the bending of light by Sun though solar eclipse

• 1921, Jebsen-Birkhoff theorem

• 1922, Friedmann equation

• 1929, Hubble observed the expansion of universe

• 1935, Einstein-Rosen bridge

• 1939, Oppenheimer-Snyder collapse

• 1948, Gamow: Big Bang

4
 • 1955, Raychaudhuri equation

• 1957, Regge-Wheeler perturbation

• 1959-1962, ADM formalism

• 1962, Newman-Penrose formalism

• 1963, Kerr solution; Penrose diagram

• 1965, Penzias and Wilson detected CMB; Penrose proved the singularity theorem

• 1967, Israel uniqueness theorem

• 1973-1976, Bekenstein, Hawking, Unruh, et al.: Black hole thermodynamics and

Hawking radiation

• 1975, Hulse and Taylor observed the energy loss of binary pulsar due to gravitational
radiation

• 1979, Shoen and Yau proved Positive Energy Theorem

• 1980, Bardeen: Cosmological perturbation

• 1981, Guth: Cosmic inflation

• 1986, Ashtekar formalism

• 2015, LIGO detected gravitational waves!

• 2019, Event Horizon Telescope released the first photo of black hole!

In this course we will be mainly concerned with the basic mathematical structures of
GR. Only after we are equipped with these mathematics can we truly understand the ob-
servational effects of GR listed above.
To start with, we take the spacetime in GR as a four-dimensional (4D) manifold M .
The 4D manifold M looks locally like a piece of R4 ; in fact, M is covered by coordinate
patches Ui in each of which we have local coordinates xa (a = 0, 1, 2, 3). The differential
structure is determined by the relations holding between systems of coordinates in overlap-
ping patches. To describe the differential structure, one exploits the tensor calculus, which
a sort of a many-index analogue of vector calculus. But unlike the vectors, the differen-
tiation of a tensor Abc... depends on the choice of the coordinate charts in an overlapping
patch. This dependence can be remedied by introducing a connection Γabc to each coor-
dinate patch, and the resulting differentiation is called the covariant differentiation, e.g.
∇a Abc = ∂a Abc + Γbap Apc + Γcaq Abq . With the covariant differentiation defined, M becomes
a differential manifold with connections.
In the case of special relativity, the Minkowski spacetime has a Lorentzian metric (with
c = 1)
ds2 = dt2 − dx2 − dy 2 − dz 2 .

5
(Note that the signature can also be chosen as (− + ++).) In special-relativistic particle
mechanics, the ds means the change in ’proper time’ experienced by the particle when it
undergoes a displacement in spacetime given by (dt, dx, dy, dz). Then apparently, if any of
the dx, dy, dz is non-zero, the proper time ds is less than coordinate time dt. In other words,
the proper time goes slower for a moving body! Let us rewrite the metric of Minkowski
spacetime as
ds2 = ηab dxa dxb
where a = 0, 1, 2, 3 and [ηab ] = diag(1, −1, −1, −1) is the flat Minkowski tensor. Sum-
mation over the repeated indices is always implied. In GR, since the gravitation and the
spacetime itself are allowed to change, we replace ηab by a generally variable tensor gab (x),
so that the metric becomes
ds2 = gab (x)dxa dxb .
So the spacetime in GR means a differential manifold with (Lorentzian) metric gab , i.e. a
Riemannian manifold. Now the proper time in GR is affected by the whole gravitational
field gab , e.g. the bending of light.
The variation of gab , in some sense, reflects the fact that the spacetime manifold is curved.
To see what the word ‘curved’ means, let us recall the fundamental theorem of Riemannian
geometry stating that
1
Γabc = g ad (∂b gcd + ∂c gbd − ∂d gbc ).
2
This Γbc is known as the Levi-Civita connection, which is metric compatible (∇a gbc = 0).
a

We see that, given a gab we can uniquely determine the Γabc , of course with respect to a
coordinate choice.
The changes in the gravitational field gab can now be described by pure geometry. How-
ever, the dynamical law of gab is expected to depend on the matter fields. One reason for
this expectation is that in classical mechanics, the Newton-Poisson equation ∇2 Φ = 4πGρ
relates the gravitational potential Φ to the matter density ρ. Ideally, we would like to have
a generalization of this Newton-Poisson equation, but a directly analogy fails for GR. To
further motivate the discussion, let us take one more step to recall that in the classical contin-
uum mechanics the equations of motion and the conservation laws for energy, momentum,
and angular momentum are embodied in the requirement that the stress tensor T ab should
be conserved (∇a T ab = 0). As and example, in the relativistic form, is the relativistic ideal
fluid with
T ab = (ρ + p)ua ub − pg ab
where ρ is the energy density and p is the pressure. By requiring ∇a T ab = 0 we obtain
Euler’s equations of motion for the fluid and the conservation equations for the energy.
Generalizing this example, we can account for the generic matter fields by a symmetric
tensor T ab with vanishing divergence.
On the geometry side, one starts from a metric tensor gab and obtains the Levi-Civita
connection Γabc , then from Γabc one hopes to construct a tensor Gab with vanishing divergence
like for the matter fields. Magically, it turns out that this Gab can be uniquely determined,
which leaded Einstein to propose his field equation

Gab = 8πGT ab .

6
This remarkable equation tells that matter sources geometry and geometry affects matter.
In other words, geometry is matter as matter is geometry!
Now, let the quest begin!......for the green valleys where dragons fly1 ......

2 Vectors and tensors in flat three-space

Before the journey to GR, let us first take a modest bow and pick up the basic vector algebra
in flat 3D space. We shall rewrite the 3D vector algebra in terms of the index notation,
which will train our muscle for later calculations in 4D.
Let xi , i = 1, 2, 3, be the Cartesian coordinates of 3D space with respect to a fixed origin;
we write Ai as the components of a vector A ⃗ with respect to this coordinate system. (A vector
is actually a coordinate-independent quantity specified by its magnitude |A| ⃗ and direction
⃗ The scalar product between two vectors can be written as A
⃗ A|.)
A/| ⃗·B ⃗ = Ai Bi .
Let us introduce the Kronecker delta δij defined by δij = 1 if i = j and zero otherwise.
One easily sees that δij = δji , δij δjk = δik , δii = 3, and δij Aj = Ai . Consider also the
permutation tensor ϵijk defined by ϵijk = 1 if ijk is an even permutation of 123, −1 for an
odd permutation of 123, and zero otherwise. One has ϵijk = ϵjki = ϵkij , ϵijk = −ϵjik , and
ϵiik = 0. The following identity is very useful:

ϵiab ϵipq = δap δbq − δaq δbp . (2.1)

With the help of ϵijk , we can write the wedge product as

⃗∧B
A ⃗ = ϵijk Aj Bk , (2.2)

and
⃗ · (A
C ⃗ ∧ B)
⃗ = [C,
⃗ A,
⃗ B]
⃗ = ϵijk Ci Aj Bk . (2.3)
By the cyclic property of ϵijk , we have immediately C
⃗ · (A
⃗ ∧ B)
⃗ =A
⃗ · (B
⃗ ∧ C)
⃗ =B
⃗ · (C
⃗ ∧ A).
⃗
We can easily calculate, for example,

⃗ ∧ (B
A ⃗ ∧ C)
⃗ = ϵijk Aj (ϵkpq Bp Cq ) = ϵijk ϵkpq Aj Bp Cq =
=(δip δjq − δiq δjp )Aj Bp Cq = Bi Aq Cq − Ci Ap Bp = B( ⃗ A⃗ · C)
⃗ − C(
⃗ A⃗ · B).
⃗ (2.4)

Consider next the algebra of 3×3 matrices. Let Uij be a 3×3 matrix, then its determinant
is
1
∆ = ϵijk Uip Ujq Ukr ϵpqr . (2.5)
6
In fact, ∆ϵijk = ϵpqr Uip Ujq Ukr and ∆ϵpqr = ϵijk Uip Ujq Ukr . For example, if ijk = 123 and
take Ap = U1p , Bq = U2q , Cr = U3r , we have

A1 A2 A3
∆ = ϵpqr Ap Bq Cr = B1 B2 B3 (2.6)
C1 C2 C3
1
This is taken from the lyrics of the song Emerald Sword by Rhapsody.

7
which is the conventional triple-product formula. The matrix multiplication of two matri-
ces, e.g. βij and γjk , is αjk = βij γjk . Using this, we have
ϵijk ∆α =αip αjq αkr ϵpqr = (βis γsp )(βjt γtq )(βku γur )ϵpqr =
=βis βjt βku γsp γtq γur ϵpqr = βis βjt βku ϵstu ∆γ = ϵijk ∆β ∆γ . (2.7)
The inverse matrix of Uij is
Vjk = (2∆)−1 ϵjpq ϵkab Uap Ubq ; (2.8)
indeed, we have
Uij Vjk = Uij (2∆)−1 ϵjpq ϵkab Uap Ubq = (2∆)−1 ϵkab Uij Uap Ubq ϵjpq = (2∆)−1 ϵkab ϵiab ∆ = δik .
(2.9)
Now the vector calculus in 3D can be greatly simplified. Firstly, by denoting ∇i =
∂/∂xi , we see that the commutation relation for the vector derivatives,
(∇j ∇k − ∇k ∇j )ϕ = 0 (2.10)
is equivalent to
ϵijk ∇j ∇k ϕ = 0. (2.11)
The divergence, gradient, and curl operations are now
divV⃗ =∇ · V⃗ = ∇i Vi , (2.12)
gradϕ =∇ϕ = ∇i ϕ, (2.13)
curlV⃗ =∇ ∧ V⃗ = ϵijk ∇j Vk . (2.14)
The Laplacian operator is then ∇2 = ∇ · ∇ = ∇i ∇i . From (2.11), we see that
div(curlV⃗ ) = ∇i (ϵijk ∇j Vk ) = 0, curl(gradϕ) = ϵijk ∇j (∇k ϕ) = 0. (2.15)
From these identities, we know that if a vector field ωi satisfies ∇i ωi = 0, then locally there
exists a vector field Vk such that ωi = ϵijk ∇j Vk , and if a vector field Vk satisfies ϵijk ∇j Vk = 0
then there exists locally a scalar ϕ such that Vk = ∇k ϕ.
We can further calculate
∇ ∧ (∇ ∧ V⃗ ) =ϵijk ∇j (ϵkpq ∇p Vq ) = ϵijk ϵkpq ∇j ∇p Vq = (δip δjq − δiq δjp )∇j ∇p Vq =
=∇i ∇j Vj − ∇j ∇j Vi = ∇(∇ · V⃗ ) − ∇2 V⃗ , (2.16)
⃗ ∧ B)
∇ · (A ⃗ =∇i (ϵijk Aj Bk ) = Bk (∇i ϵijk Aj ) + Aj (∇i ϵijk Bk ) =
⃗ · (∇ ∧ A)
=Bk (ϵijk ∇i Aj ) − Aj (ϵjik ∇i Bk ) = B ⃗ −A
⃗ · (∇ ∧ A).
⃗ (2.17)
The vector integral calculus can be summarized as
Z
dxi ∇i Q =QB − QA , (Newton-Leibnitz) (2.18)
γ
Z Z
dSϵijk Ni ∇j Q = dxk Q, (Stokes) (2.19)
S Z ZΓ
dV ∇i Q = dSNi Q, (Gauss) (2.20)
V Σ

8
where Q is a smooth field, γ is a smooth curve with endpoints A and B, S is a smooth surface
with boundary Γ, V is a region of space bounded by a closed surface Σ, and Ni is normal to
the surface or the outward normal to the sphere. If γ is given parametrically by xi (s) where
s is the arc length, then dxi = ti ds where ti is the unit tangent vector along γ. If the surface
S is represented parametrically as xi (u, v), then

Ni dS = ϵijk Uj Vk dudv (2.21)

with Uj = ∂xj /∂u and Vk = ∂xk /∂v.

2.1 Electromagnetism
Let Ei be the electric field, Bi the magnetic field, ρ the electric charge density, and Ji the
electric current density. Maxwell’s equations are then

∇i Bi =0 (2.22)
∇i Ei =4πρ (2.23)
ϵijk ∇j Ek = − Ḃi (2.24)
ϵijk ∇j Bk =Ėi + 4πJi (2.25)

where the dot means the time derivative ∂/∂t and we have set c = 1. From (2.22) and (2.24)
we obtain

Bi =ϵijk ∇j Ak (2.26)
Ei = − Ȧi − ∇i ϕ (2.27)

where the vector potential Ai and the scalar potential ϕ are determined up to gauge trans-
formations Ai → Ai + ∇i ψ, ϕ → ϕ − ψ̇ with ψ an arbitrary function. By differentiating
(2.23) with respect to time and taking the div of (2.25), we obtain via comparison that

∇i Ji + ρ̇ = 0, (2.28)

which is the equation of conservation for electric charge. Consider the net electric charge
Q in a region V , Z
Q= dV ρ; (2.29)
V
we have Z Z Z
Q̇ = dV ρ̇ = − dV ∇i Ji = − dSNi Ji (2.30)
V V Σ
which means that the rate of increase of Q is the net flux of charge across Σ into V .
In the vacuum (i.e. ρ = 0, Ji = 0), the electric and magnetic field both satisfy the wave
equations: Ëi = ∇2 Ei , B̈i = ∇2 Bi . Moreover, by contraction of (2.24) with Bi and (2.25)
with Ei , we obtain
∇i Pi + µ̇ = 0, (2.31)
where Pi = 4π1
ϵijk Ej Bk is the flux current of the electromagnetic energy and µ = 8π
1 ⃗2
(E +
⃗ ) is the density of electromagnetic energy. (2.31) means the energy conservation of the
B 2

vacuum electromagnetic field.

9
 For a particle of mass m and charge q moving in the electromagnetic field (Ei , Bi ), the
Lorentz equation of motion is

dxi
mV̇i = q(Ei + ϵijk Vj Bk ), Vk ≡ . (2.32)
dt

2.2 Non-relativistic fluid

Consider a perfect fluid with density ρ, pressure p, and velocity Vi . These parameters are all
functions of space and time. The equations of motion (EoM) of a perfect fluid are

ρ[V̇i + Vj ∇j Vi ] = −∇i p, (2.33)

which are called Euler’s equations. There is also the mass conservation equation

∇i (ρVi ) + ρ̇ = 0. (2.34)

The momentum conservation equation is in fact the Euler equations (2.33). To see this, let
us introduce the stress tensor
Tij = ρVi Vj + pδij , (2.35)
then (2.33) can be cast in the form of a conservation equation,

∂t (ρVi ) + ∇j Tij = 0. (2.36)

Define the vorticity ωi = ϵijk ∇j Vk , then

ϵijk Vj ωk =ϵijk Vj (ϵkpq ∇p Vq ) = ϵijk ϵkpq Vj ∇p Vq = (δip δjq − δiq δjp )Vj ∇p Vq =
1
=Vj ∇i Vj − Vj ∇j Vi = ∇i (Vj Vj ) − Vj ∇j Vi . (2.37)
2
By (2.37), the Euler’s equations become

1
ρV̇i = ρϵijk Vj ωk − ρ∇i V 2 − ∇i p. (2.38)
2
A steady flow is one for which ρ, p, and Vi are all time independent. If a steady flow has the
further property that ωi = 0 (or simply ϵijk Vj ωk = 0), we obtain from (2.38)

1
ρ∇i V 2 + ∇i p = 0 (2.39)
2
which means that ρ and p are related in some way. Suppose there exists a function h such
that ∇i (ρ−1 h) + ρ−1 ∇i p = 0, then (2.39) can be integrated to

1 2
ρV + h = E (2.40)
2
where E is an integration constant. If we take h = p and ρ =const., i.e. the incompressible
fluid, then (2.40) expresses the Bernoulli theorem. Since ω = 0, we have Vi = ∇i Φ where
Φ is the velocity potential, and (2.40) reduces to the Newton-Laplace’s equation ∇2 Φ = 0.

10
 Consider next a time-dependent incompressible fluid (ρ =const.). If we take the curl of
(2.38), it follows that

ω̇i =ϵijk ∇j (ϵkpq Vp ωq ) = ϵijk ϵkpq ∇j (Vp ωq ) = (δip δjq − δiq δjp )(ωq ∇j Vp + Vp ∇j ωq ) =
=ωj ∇j Vi − ωi ∇j Vj + Vi ∇j ωj − Vj ∇j ωi . (2.41)

We can see that the third term of (2.41) vanishes because of (2.15) and the second term
vanishes because of (2.34) and ρ =const.; thus we are left with

ω̇i + ζi = 0, ζi = Vj ∇j ωi − ωj ∇j Vi , (2.42)

which is the Helmholtz theorem.

Let Φ be the gravitational potential, then the Euler equations for the fluid moving under
the influence of a gravitational field are

∂t (ρVi ) + ∇j Tij = −ρ∇i Φ. (2.43)

A self gravitating fluid is one for which the mass density ρ of the fluid is itself the source of
the gravitational field. In other words, we have to supplement (2.43) with

∇2 Φ = 4πGρ. (2.44)

The (2.43) can be rewritten in the following form of conservation equation,

∂t (ρVi ) + ∇j (Tij + τij ) = 0. (2.45)

Indeed,

∇i Φ(4πGρ) =∇i Φ(∇2 Φ) = ∇i Φ(∇i ∇i Φ) + ∇i ∇j Φ∇j Φ − ∇i ∇j Φ∇j Φ =

1
=∇j (∇i Φ∇j Φ) − ∇i (∇j Φ∇j Φ) ≡ 4πG∇j τij . (2.46)
2

2.3 Coordinate bases and tensors

In the calculations presented above, the suffixes attached to the vectors (and tensors) are
all subscripts. In later sections, the difference between subscripts and superscripts becomes
important, so let us take a look at this difference.
Let us consider the 3D Cartesian coordinates (x, y, z) and an associated set of unit vectors
(i, j, k) along the coordinate axises. If we want to transform to another coordinate system
(u, v, w), e.g. from a rectangular one a spherical one, we could write

r = x(u, v, w)i + y(u, v, w)j + z(u, v, w)k (2.47)

By fixing u = u0 , the remaining (v, w) define a coordinate surface; likewise, we have other
two coordinate surfaces defined by fixing v = v0 or w = w0 . Two different coordinate
surface intersect at a coordinate curve, e.g.

r = x(u, v0 , w0 )i + y(u, v0 , w0 )j + z(u, v0 , w0 )k. (2.48)

11
If we differentiate (2.48) wrt u, we obtain a tangent vector ∂r/∂u to this coordinate curve.
This way, we obtain the following set of tangent vectors at the intersection point of three
coordinate curves:
∂r ∂r ∂r
eu ≡ , ev ≡ , ew ≡ . (2.49)
∂u ∂v ∂w
These vectors {eu , ev , ew } from the natural basis at the intersection point. But if we use the
normal vectors, instead of the tangent vectors, at the intersection point, we obtain another
set of basis vectors. To see this, let us invert the functions x(u, v, w), y(u, v, w), z(u, v, w) to
u(x, y, z), v(x, y, z), z(x, y, z), then the gradients of the new coordinates can be expanded
in terms of (i, j, k), e.g.
∂u ∂u ∂u
∇u = i+ j+ k. (2.50)
∂x ∂y ∂z
This ∇u is normal to the coordinate surface defined by u = u0 , and likewise for other
gradients. We therefore obtain another coordinate basis at the intersection point:

eu ≡ ∇u, ev ≡ ∇v, ew ≡ ∇w. (2.51)

We see from (2.49) and (2.51) that the places of suffixes are different.
Let us rewrite these relations using the index formalism as above, that is , we write the
basis (2.49) as ei and (2.51) as ei . For a vector A, we have that in the basis ei its components
are Ai such that A = Ai ei , where the superscripts are called the contravariant labels. Likewise,
in the basis ei the components are Ai such that A = Ai ei , and the subscripts are called the
covariant labels. (We see that the summation convention implicitly used above only applies to a
label/suffix appeared twice, once as a subscript and once as a superscript.)
A direct calculation shows that
∂r ∂ui ∂x ∂ui ∂y ∂ui ∂z ∂ui
ei · ej = ∇ui · = + + = = δji . (2.52)
∂uj ∂x ∂uj ∂y ∂uj ∂z ∂uj ∂uj

On the other hand, if we denote ei · ej ≡ gij , then

A · B = Ai ei · B j ej = gij Ai B j . (2.53)

Similarly, we have ei · ej ≡ g ij and

A · B = Ai ei · Bj ej = g ij Ai Bj . (2.54)

Since A · B = g ij Ai Bj = Ai B j , we obtain g ij Bj = B i , and likewise gij B j = Bi . As a

consequence, B i = g ij Bj = g ij gjk B k , which entails g ij gjk = δki .
Next, suppose we want to change to the coordinates (u(t), v(t), w(t)) depending on a
parameter t. Let r(t) be the (2.47) depending on t, then

dr ∂r ∂u ∂r ∂v ∂r ∂w
= + +
dt ∂u ∂t ∂v ∂t ∂w ∂t
which is equivalent to
ṙ(t) = u̇i (t)ei . (2.55)

12
The position vectors r(t) follows a curve γ. Obviously, the length of this curve γ can be
obtained by integrating |ṙ| wrt t. Since |ṙ|2 = ṙ · ṙ = u̇i (t)ei · u̇j (t)ej = gij u̇i u̇j , one has the
length of γ(t), t ∈ [a, b]
Z b p
L= dt gij u̇i u̇j . (2.56)
a

Meanwhile, the infinitesimal change dr = dui ei , so the line element

ds2 = dr · dr = dui ei · duj ej = gij dui duj . (2.57)

Example 2.1 (Spherical metric). In the spherical coordinates (r, θ, ϕ), the position vector
(2.47) becomes
r = r sin θ cos ϕi + r sin θ sin ϕj + r cos θk.
We can compute the natural basis

∂r
er = = sin θ cos ϕi + sin θ sin ϕj + cos θk,
∂r
∂r
eθ = = r cos θ cos ϕi + r cos θ sin ϕj − r sin θk,
∂θ
∂r
eϕ = = −r sin θ sin ϕi + r sin θ cos ϕj.
∂ϕ

Using (2.57), we obtain

ds2 = dr2 + r2 dθ2 + r2 sin2 θdϕ2 . (2.58)

The above formulas also formally hold for general tensors on general manifolds! On
a manifold, the natural basis {Ea } is in the tangent space Tp at a point, and the dual basis
{Ea } is in the cotangent space Tp∗ . The suffix notation for vectors still holds true. Now the
multi-indices quantities, such as ϵijk and gij , generalize vectors to tensors. As above, a tensor
(component) Aab11...b
...ar
r
of type (r, s) is defined with respect to the basis
∗
Ea1 ⊗ ... ⊗ Ear ⊗ Eb1 ⊗ ... ⊗ Ebs ∈ Tp(1) ⊗ ... ⊗ Tp(r) ⊗ Tp(1) ∗
⊗ ... ⊗ Tp(s) . (2.59)

An important point is that the (natural) basis {Ea } need not commute. But in order to
ensure that {Ea } stay in the tangent space Tp of the point p considered, these basis vectors
must form a Lie algebra, [Ea , Eb ] = Cab
c
Ec , or simply commute, [Ea , Eb ] = 0 (i.e. Frobenius
theorem). The above Cartesian case is just the commuting case with

Ea = ∂/∂xa , Ea = dxa , (2.60)

which is the coordinate basis used in most cases. On general manifolds the basis vectors will
form a Lie algebra, which is called a non-coordinate basis. Indeed, in GR we have locally a
flat Minkowski spacetime, so the moving basis at a point should be ortho-normalized to
the Minkowski metric, and such an orthonormal moving basis is a non-coordinate basis.
In other words, they cannot be transformed from a global Cartesian coordinate system, but
only locally expand a frame around a point on the manifold. (Cf. (7.8).)

13
3 Special relativistic geometry
Let us now turn to special relativity, still using the index notation. Now the spacetime
is a four-dimensional flat manifold with coordinates xa , a = 0, 1, 2, 3, i.e. a Minkowski
spacetime. The points of the Minkowski spacetime are called ’events’, and we are interested
in the relations of events to one another.
Since we are now in 4D, we write the 4-vectors relative to the coordinates xa as Aa , B a , ....
The inner product between two 4-vectors is gab Aa B b with gab = diag(1, −1, −1, −1) ≡ ηab
being the Minkowski metric. As gab is indefinite, we see that gab Aa Ab is not positive definite;
a nonzero 4-vector is called time-like, space-like, or null according as to whether gab Aa Ab
is positive, negative, or zero. A time-orientation is chosen by taking at will some time-like
vector, say U a = (1, 0, 0, 0), and designating it to be future-pointing. Any other time-like
or null vector V b such that gab U a V b > 0 is also future-pointing, whereas if gab U a V b < 0
then V b is past-pointing. Space-like vectors are neither future-pointing nor past-pointing.
The following propositions can be readily checked:

1. if P a is time-like and P a Sa = 0 then Sa = gab S b is space-like;

2. if P a and Qa are time-like and P a Qa > 0 then either both are future-pointing or both
are past-pointing;

3. if P a and Qa are null and P a Qa = 0 then P a and Qa are proportional;

4. if P a is null and P a Ra = 0 then either Ra is space-like or null;

5. if U a , V a and W a are time-like with U a Va > 0 and U a Wa > 0, then V a Wa > 0.

Note that these definitions and properties also hold locally in the neighborhood of a point
of a general manifold (cf. below).
Two events xa and y a are null-separated if gab (xa −y a )(xb −y b ) = 0, and for a given event
xa the set of all points null-separated from it is called the lightcone at xa . The lightcone of xa
is generated by a two-dimensional family of null geodesies emanating from xa . Indeed, for
any point y a on the null cone of xa , all points on the line xa + λ(xa − y a ) are null-separated
from xa .

3.1 Relativistic Kinematics

Consider a relativistic point particle with 4-momentum P a (which is required to be future-
pointing). The squared rest mass of the particle is then

M 2 = P a Pa . (3.1)

We denote by V a the unit future-pointing time-like 4-velocity of an observer. The total

relativistic energy of the particle, measured relative to the observer, is

E = P a Va . (3.2)

14
 If the particle is at rest w.r.t. the coordinate chosen and the observer is moving with velocity
⃗v , then P a = (M, 0, 0, 0) and V a = γ(1, ⃗v ) where γ = (1 − v 2 )−1/2 is the Lorentz factor,
and we can obtain the well-known result
M
E=√ . (3.3)
1 − v2

If P a is null, (3.3) fails but (3.2) is still applicable. So if a photon with 4-momentum P a
is emitted by a body with 4-velocity U a and received by an observer with 4-velocity V a ,
then the emitted energy measured in the frame of the emitting body is P a U a , whereas the
energy measured at reception is P a V a . If U a ̸= V a , so will the two measured energies,
which is the relativistic Doppler effect. As an example, take U a = (1, 0, 0, 0), V a = γ(1, ⃗v ),
and P a = (E, p⃗) with E 2 = p2 , then

E1 ≡ P a Ua = E, E2 ≡ P a Va = γE − γ⃗p · ⃗v .

Equivalently, E2 /E1 = γ(1 − v cos θ) where θ is the angle between p⃗ and ⃗v . Thus, if θ = 0,
i.e. p⃗ and ⃗v are parallel, then E2 /E1 = (1 − v)/(1 + v), so the received energy is measured
p

to be less than the emitted energy. On the other hand, if θ = π/2, then E2 /E1 = γ,
giving a received energy that is greater than that of the emitted energy. Considering the
approximation to first order in v, we see that E2 /E1 ≈ 1 − v when θ = 0, corresponding
to the non-relativistic red-shift formula; when θ = π/2, we have E2 /E1 ≈ 1, which means
that the transverse blue-shift effect (in higher orders in v) is a purely relativistic one.
When particles interact, the total 4-momentum is conserved. Thus if a particle with
4-momentum P a decays into two particles with 4-momenta Qa1 and Qa2 , we must have

P a = Qa1 + Qa2 . (3.4)

Likewise, if particles with four-momentum P a and Qa interact to form new particles with
four-momentum P̃ a and Q̃a , then we have

P a + Qa = P̃ a + Q̃a . (3.5)

As an example, consider a photon with 4-momentum P a scattering off an electron with

4-momentum Qa so as to produce a photon of 4-momentum P̃ a and an electron of 4-
momentum Q̃a . Let us write the 4-momentum of the electron before scattering as Qa =
mV a where V a is a unit vector, then the 3-momentum of the incident photon relative to
the rest frame is given by the space-like projection

P a − V a P b Vb . (3.6)

We can rewrite P a − V a P b Vb = ERa with E = P a Va being the energy of the incident pho-
ton and Ra the unit space-like vector in the direction of the incident photon’s 3-momentum
(V a Ra = 0). Similarly, we have P̃ a − V a P̃ b Vb = Ẽ R̃a for the scattered photon. Then
Ra R̃a = − cos θ defines the scattering angle θ. Thus,

(P a − V a P b Vb )(P̃a − Va P̃ b Vb ) = −E Ẽ cos θ,

15
which is the space-like part of the rest mass. Combining with the time-like part we obtain

P a P̃a = E Ẽ(1 − cos θ).

By squaring P a + Qa − P̃ a = Q̃a and that |Q| ≫ |P |, we have that 2m(E − Ẽ) = 2P a P̃a ,
so that
E Ẽ
E − Ẽ = (1 − cos θ) (3.7)
m
which is the Compton scattering formula.
For a free point particle, its trajectory in spacetime, the worldline, is a time-like line. We
call such lines the time-like geodesies

xa (s) = ba + sV a (3.8)

where ba is a point on the worldline and V a is a unit future-pointing time-like vector in the
direction of the worldline. Here, s is the proper time along the trajectory.

3.2 Lorentz contraction and time dilation

The Lorentz contraction effect arises out of a subtle ambiguity as to what is meant by the
’length’ of an extended body. If we think of the body as extended in time as well, so as to
form a kind of space-time entity analogous to a world line, then we may take space-like
cross-sections of this entity in various ways with space-like hyperplanes; and the relevant
length calculations vary in result according as to which hyperplane is chosen.
Consider a pair of parallel time-like geodesics. Suppose an observer based at point P
has unit 4-velocity U a parallel to these two time-like geodesics; the space-like hyperplane
orthogonal to U a intersects the geodesics at a pair of points α and β. Relative to the reference
−
→
frame defined by U a the rod is represented by the vector Ra = αβ, which has the squared
proper length L2 = −Ra Ra . Likewise, we suppose there exists another observer based at Q
with 4-velocity V a . The space-like hyperplane through Q orthogonal to V a intersects the
two parallel geodesics at the points ξ and ζ, and thus in the frame of V a the rod is represented
→
−
by R̃a = ξζ which has squared length L̃2 = −R̃a R̃a . Then, by transporting α to ξ, we see
that R̃a = Ra + ψU a . Since R̃a Va = 0, we find that ψ = −Ra Va /U b Vb . It follows that

L̃2 = − R̃a R̃a = −(Ra + ψU a )(Ra + ψUa ) =

= − Ra Ra − ψ 2 U a Ua = L2 − (Ra Va /U b Vb )2 . (3.9)

Therefore, in all circumstances L̃ < L, which is the Lorentz contraction effect.

If we take
V a = γ(U a + vX a ) (3.10)
where v is the speed, γ = (1 − v 2 )−1/2 , and Ra = LX a . One verifies that V a Va = 1. It
follows that Ra Va = γvL, U a Va = γ, and from (3.9),
√
L̃ = 1 − v 2 L. (3.11)

16
 Before turning to time dilation, we notice that observer’s concept of simultaneity is
defined by the space-like hyperplane; in other words, two events lying in the same space-
like hyperplane orthogonal to an observer’s worldline are regarded as simultaneous by that
observer.
Consider two worldlines cross at point A, one with 4-velocity U a and the other with
4-velocity V a . After passage of a duration of proper time τ̃ the second observer reaches the
point B̃. This point is simultaneous (i.e. in the same space-like hyperplane Π), in the frame
of the first observer, with a point B on the first observer’s worldline. By vector addition
rule, we have
τ U a + rX a = τ̃ V a (3.12)
where τ is the proper time elapsed along the trajectory AB, and X a is a unit space-like
vector in the direction B B̃. Substituting (3.10) into (3.12), we obtain r = γvτ̃ and
√
τ̃ = 1 − v 2 τ. (3.13)

This is the time dilation effect; in words, the first observer will claim the second observer’s
clock is running slow relative to his. Since the roles of U a and V a are symmetrical, the second
observer will think that, relative to his own frame, the first observer’s clock is running slow.

3.3 Relativistic electrodynamics

The electromagnetic field can be represented in relativistic terms by field strength tensor
Fab , (a, b = 0, 1, 2, 3) with Fab = −Fba , and having components

Ei =Fi0 , (i = 1, 2, 3) (3.14)
1
Bi = ϵijk Fjk . (3.15)
2
By (2.1), we also have
Fij = ϵijk Bk . (3.16)
The Maxwell’s equations in vacuum are (with ∇a ≡ ∂/∂xa and ∇a = gab ∇b )

∇a Fab =0, (3.17)

∇a Fbc + ∇b Fca + ∇c Fab =0. (3.18)

The (i, 0)-components of (3.17) give ∇i Ei = 0. Since ∇i Fij = −∇i Fij = −∇i ϵijk Bk =
ϵjpq ∇p Bq and ∇0 F0j = ∇0 F0j = −Ėj , it follows from ∇0 F0j +∇i Fij = 0 that ϵjpq ∇p Bq = Ėj .
By contraction of ∇i Fjk + ∇j Fki + ∇k Fij = 0 with ϵijk and using (3.15), we get ∇i Bi = 0.
Finally by contraction of ∇0 Fjk +∇i Fj0 +∇j F0i = 0 with 21 ϵijk we obtain Ḃk +ϵijk ∇i Ej = 0.
Therefore, these indeed recover the nonrelativistic form of the Maxwell’equations (2.22) to
(2.25).
If the sources are present, (3.17) becomes2

∇a Fab = −4πJb , ⃗
J a = (ρ, J), (3.19)
2
Note that this result need to be modified on a curved spacetime! (Ex.)

17
then naturally ∇a Ja = 0, the conservation of electromagnetic current. We can consider a
space-like hypersurface Σ with boundary ∂Σ and volume element dSa . We find the charge
Z Z I
1
Q= a
dSa j = ab
dSa ∇b F = dSab F ab (3.20)
Σ Σ 2 ∂Σ
where dSab is the volume element of ∂Σ and j a = −J a /(4π). When Σ is a space-like
hypersurface defined by constant t, dSab has the nonvanishing component dS0i = −dSi0 ≡
dSi , so that I
Q= dSi F 0i (3.21)
∂Σ

which recovers the non-relativistic result (2.30).

Furthermore, ∇a of (3.18) with the help of (3.19), we obtain

∇a ∇a Fab = −4π(∇a Jb − ∇b Ja ). (3.22)

We see that for the electromagnetic field to satisfy the wave equation, the current has to be
curl-free. (3.18) holds whether or not sources are present; it shows that there exists a vector
potential Aa such that
Fab = ∇a Ab − ∇b Aa , (3.23)
which is a relativistic form of (2.26)(2.27).
Before we proceed, let us introduce further notations. The anti-symmetric and sym-
metric products of two vectors can be written respectively as
1 1
A[a Bb] = (Aa Bb − Ab Ba ), A(a Bb) = (Aa Bb + Ab Ba ), (3.24)
2 2
so that Aa Bb = A[a Bb] + A(a Bb) . Such a decomposition holds for a general tensor Cab =
C[ab] + C(ab) with the anti-symmetric and symmetric products defined in the same way.
Furthermore in the case of three-index tensors we write
1
D(abc) = (Dabc + Dbca + Dcab + Dbac + Dcba + Dacb ), (3.25)
6
1
D[abc] = (Dabc + Dbca + Dcab − Dbac − Dcba − Dacb ). (3.26)
6
In particular, since Fbc is itself anti-symmetric, we have
1
E[a Fbc] = (Ea Fbc + Eb Fca + Ec Fab ). (3.27)
3
So the Maxwell’s equations (3.18) are simply ∇[a Fbc] = 0, and Fbc = 2∇[b Ac] , and so on.
The totally skew symmetric tensor in 4D is defined as in 3D (i,e, 1 for even permutations
of indices and −1 for odd permutations). We have following identities:

ϵabcd ϵabcd = − 24, (3.28)

abch
ϵabcd ϵ =− 6δdh , (3.29)
[g h]
ϵabcd ϵabgh = − 4δ[c δd] , (3.30)
[f h]
ϵabcd ϵaf gh = − 6δ[b δcg δd] , (3.31)
[e h]
ϵabcd ϵef gh = − 24δ[a δbf δcg δd] . (3.32)

18
 The stress-energy tensor of the electromagnetic field is
1 1
Tab = Fac Fbd g cd − gab Fde F de (3.33)
4π 16π
which is obviously symmetric. (Cf. the later §11.1 for a derivation.) In vacuum,

∇a Tab = 0, (3.34)

expressing the conservation of electromagnetic energy and momentum.

The dual of the field strength is
1 1
∗
Fab = gap gbq ϵpqrs Frs = ϵabcd F cd . (3.35)
2 2
By (3.30), we have
∗∗
Fab = −Fab . (3.36)
Using the dual field tensor, we can rewrite the (3.18) as

∇a∗ Fab = 0. (3.37)

Let us also consider the self-dual and the anti-self-dual parts of Fab :

(s.d.) + Fab = Fab − i ∗ Fab , (a.s.d.) − Fab = Fab + i ∗ Fab , (3.38)

satisfying respectively ∗+ Fab = i+ Fab and ∗− Fab = −i− Fab . We can see that the vacuum
Maxwell’s equations are expressed neatly by a single relation

∇a + F ab = 0. (3.39)

3.4 Relativistic hydrodynamics

A perfect relativistic fluid is characterized by its energy density ρ, pressure p, and the ev-
erywhere time-like and normalized velocity field ua . Its stress-energy tensor is

Tab = (ρ + p)ua ub − gab p (3.40)

with a general metric gab . The EoM are obtained from ∇a Tab = 0, i.e. the relativistic Euler
equation
(ρ + p)ua ∇a ub = (g bc − ub uc )∇c p ≡ Πbc ∇c p (3.41)
and the relativistic continuity equation

ua ∇a ρ + (ρ + p)∇a ua = 0. (3.42)

Indeed, ub ∇a Tab = 0 gives (3.42), and by substituting (3.42) into ∇a Tab one obtains (3.41).
Consider a simple perfect fluid composed of particles of a single type with number den-
sity n(xa ) = N/V = number/volume. For a simple fluid, the local conservation law of n is
also required:
∇a (nua ) = 0. (3.43)

19
The 4-vector N a = nua is called the number-flux 4-vector, as na = (nγ, nγv x , nγv y , nγv z )
if ua = (γ, γv x , γv y , γv z ) with the relativistic factor γ taken into account. If the fluid consists
of particles of different types, we can write the number-flux 4-vector as
XZ dxa
a
n = dτi i δ 4 (xa − xai (τi )) (3.44)
i
dτ i

where i labels the types of particles.

Define the specific energy as ρ/n and the specific volume as 1/n, then we have a local
form of the first law of thermodynamics
ρ 1
kT dS = d + pd (3.45)
n n
where S is the specific heat and T is the temperature. Simple rearrangement gives
1 ρ+p
dS = dρ + 2 dn, (3.46)
nkT n kT
where (ρ + p)/n ≡ f is the specific enthalpy. Thus if we regard ρ and n as independent
variables on which S, T and p depend, we can define a fundamental complete equation of
state for a simple perfect fluid as a specification of the entropy function S(ρ, n). A simple
perfect fluid is locally adiabatic. To see this, we rewrite (3.46) as knT ∇a S = ∇a ρ − f ∇a n,
then

knT ua ∇a S = ua ∇a ρ − f ua ∇a n = −f n∇a ua − f ua ∇a n = −f ∇a (nua ) = 0 (3.47)

where the continuity equation of the form ua ∇a ρ + f n∇a ua = 0 is used.

4 Tensors on manifolds
An n-dimensional topological manifold is a Hausdorff topological space M with the prop-
erty that every point in M has an open neighborhood homeomorphic to an open set in Rn .
Let M be covered by a family of open sets Ui , the coordinate patches, such that each open
set of M can be mapped to an open set of Rn : xa : Ui 7→ Rn ; p 7→ xa [p].
Let us write Uij = Ui ∩ Uj for the various intersection regions of the patches. Suppose
a point p lies in the overlap region U12 with coordinates xa [p] in U1 and y a [p] in U2 . For all
p in U12 we require the existence of a set of functions xa (y b ) such that xa [p] = xa (y b [p]) ≡
a
f12 (y b [p]) with f12
a
being the coordinate transition function. We require the coordinate
transition functions to be compatible with one another in triple overlap regions. Thus if

xa [p] = xa (y b [p]) in U12 ; y a [p] = y a (z b [p]) in U23 ; xa [p] = xa (z b [p]) in U13

we require xa (y b (z c [p])) = xa (z c [p]) in U123 .

For a differentiable manifold we require the transition functions to be differentiable
functions. In particular, if the transition functions have continuous partial derivatives of
order r then we say the manifold is differentiable of class r. If the transition functions admit
derivatives to all orders then we say that the manifold is smooth.

20
 Let ϕ(xa ) be a scalar field on U1 and ϕ′ (x′a ) a scalar field in U2 . If ϕ(xa ) = ϕ′ (x′a ) on
U12 , then together they define a scalar field on U1 ∪ U2 . If the scalar fields on those patches
all agree as above in the overlap regions, we obtain thereby a globally defined scalar field
on the manifold. Consider next the following linear differential operators acting on scalar
fields on a differentiable manifold (cf. (2.60)):
∂ ∂
V a (x) on U, V ′a (x′ ) ′a
on U ′ . (4.1)
∂xa ∂x
For scalar fields ϕ(x) on U and ϕ′ (x′ ) on U ′ agreeing on U ∩ U ′ , these differential operators
satisfy V a (x) ∂x∂ a ϕ(x) = V ′a (x′ ) ∂x∂′a ϕ′ (x′ ); but

∂ ∂x′b ∂ ′ ∂x′a ∂ ′ ′
V a (x) a b


ϕ(x) = V (x) ϕ x(x ) = V (x) ϕ (x ),
∂xa ∂xa ∂x′b ∂xb ∂x′a
so we require in U ∩ U ′
∂ ′ ′ ∂x′a ∂ ′ ′
V ′a (x′ ) ϕ (x ) = V b
(x) ϕ (x ). (4.2)
∂x′a ∂xb ∂x′a
We therefore obtain the transition law for a contravariant vector field in U ∩ U ′
∂x′a
V ′a (x′ ) = V b (x) . (4.3)
∂xb
Dual to the contravariant vector fields, the covariant vector fields have the following
transition law in U ∩ U ′
∂xb
A′a (x′ ) = Ab (x) ′a . (4.4)
∂x
To see this, consider the grad of scalar fields:
∂ϕ(x) ∂ϕ′ (x′ )
Aa (x) = , A′a (x′ ) = , (4.5)
∂xa ∂x′a
then, by the definition of ϕ′ ,
∂ϕ(x) ∂x′c ∂ϕ(x(x′ )) ∂x′c ∂ϕ′ (x′ ) ∂x′c ′ ′
Ab (x) = = = = A (x ). (4.6)
∂xb ∂xb ∂x′c ∂xb ∂x′c ∂xb c
b ∂xb ∂x′c
Multiplying each side of (4.6) by ∂x
∂x
′a , and using the identity ∂x′a ∂xb
= δac , one obtains (4.4).
For general scalar fields, we can have locally
X
Aa = Θr ∇a ϕr (4.7)
r

where ϕr are the scalars such that ∇a ϕr constitute a basis for covariant vector (covector) fields
over the relevant region of the manifold. This is in fact a consequence of Pfaff’s theorem.
For general tensor fields, we denote the valence by (p, q), i.e. p contravariant indices and
q covariant indices. The rank of a tensor is then (p + q). The transition law has for each
index either a Jacobian matrix, or the corresponding inverse Jacobian matrix, depending on
whether the index is of contravariant or covariant type. For example, for (1, 1) tensor Cba ,
we have the transition law
∂x′a ∂xq p
Cb′a = C . (4.8)
∂xp ∂x′b q

21
The metric tensor gab therefore transforms as

∂xc ∂xd
′
gab = gcd . (4.9)
∂x′a ∂x′b
Denote by g the determinant of gab , then we see from above that g ′ = det| ∂x ′ | g, and for a
∂x 2
√
scalar ϕ, we see that −gϕ is a scalar density of weight −1, i.e. having the usual Jacobian
√
rule, and can be integrated as usual. The integration measure −gdx4 is called the volume
element. Note that, a manifold M is orientable if the Jacobian det| ∂x
∂x
′ | > 0.

The tensor rank can be changed by multiplication, but not by addition:

1. addition, e.g. Apqr + Bqr

p p
= Cqr ;

2. outer multiplication, e.g. Apq Bst

r pr
= Cqst ;

3. inner multiplication or contraction, e.g. Lm

n Mop = Nop .
n m

4.1 Covariant derivatives

We have seen that differentiation of scalar fields is definable on a differentiable manifold,
provided that the transition functions are themselves differentiable. Here, we would like
to extend the differentiation to general tensor fields. To this end, we need to introduce
connections.
Suppose the manifold M is covered by a family of open sets U, U ′ , and so on. On each
patch U we define a three-index array of functions Γabc , the connection. In transition regions,
they transform as
∂x′a ∂xq ∂xr p ∂x′a ∂ 2 xp
Γ′a
bc = Γ + (4.10)
∂xp ∂x′b ∂x′c qr ∂xp ∂x′b ∂x′c
which is different from the tensor transformations. The second term of (4.10) can be rewrit-
ten as
∂x′a ∂ 2 xp ∂  ∂x′a ∂xp  ∂xp ∂ 2 x′a ∂xq ∂xr ∂ 2 x′a
= − ′c q p = − ′b ′c q r (4.11)
∂xp ∂x′b ∂x′c ∂x′b ∂xp ∂x′c ∂x ∂x ∂x ∂x ∂x ∂x ∂x
b ′c
where we have used ∂x ′a ∂xb = δa and the transformation of covariant vector labels. Thus if
∂x ∂x c
′a a
we write Jba ≡ ∂x
∂xb
and J˜ba ≡ ∂x
∂x
′b , then (4.10) becomes

Γ′a a ˜q ˜r p a ′ ˜p
bc = Jp Jb Jc Γqr + Jp ∂b Jc . (4.12)

Or equivalently, by virtue of (4.11),

Γ′a a ˜q ˜r p ˜q ˜r a
bc = Jp Jb Jc Γqr − Jb Jc ∂q Jr . (4.13)

Now we want to show that the second term of (4.10) ensures the compatibility of the
connections in triple overlap regions. Indeed, we suppose that in the triple overlap region
U ∩ U ′ ∩ U ′′ , Γabc transforms to Γ′a
bc and Γbc transforms to Γbc . Then it turns out that Γbc
′a ′′a a

automatically transforms to Γ′′a bc .

22
 Take Jba , J˜ba in U as above, and let us write the counterparts in U ′ and U ′′ as Kba , K̃ba and
Lab , L̃ab respectively. Then, By simple substitution, we obtain
a q r p ˜v ˜w u p ′ ˜u
Γ′′a a ′′ p
bc =Kp K̃b K̃c (Ju Jq Jr Γvw + Ju ∂q Jr ) + Kp ∂b K̃c =
=La L̃v L̃w Γu + K a K̃ q K̃ r J p ∂ ′ J˜u + K a ∂ ′′ K̃ p =
u b c vw p b c u q r p b c
r ′′ ˜u a q ˜u ′′ r
=Lau L̃vb L̃w u
c Γvw + a
Lu (K̃c ∂b Jr ) + Kq Ju (Jr ∂b K̃c ) =
=Lau L̃vb L̃w u
c Γvw + Lau ∂b′′ (K̃cr J˜ru ) = Lau L̃vb L̃w u a ′′ u
c Γvw + Lu ∂b L̃c (4.14)

where use has been made of the relations Kba Jcb = Lac and K̃ba ∂q′ = (∂b′′ x′q )∂b′ = ∂b′′ .
Thus, a manifold with connection is a differentiable manifold endowed with a field Γabc
subject to transition laws of the form (4.10).
The connections helps us to define a general rule for the differentiation of tensor fields.
To see this note first that if V a is a contravariant vector field then ∂b V a fails to transform as
a tensor field of valence (1, 1) in transition regions:

∂b′ V ′a = J˜bp ∂p (Jqa V q ) = J˜bp Jqa ∂p V q + (J˜bp ∂p Jqa )V q . (4.15)

Now define the covariant differentiation on vector fields of valence (1, 0) as

∇b V a = ∂b V a + Γabc V c . (4.16)

Then in a transition region U ∩ U ′ , we have by simple substitutions,

∇′b V ′a =∂b′ V ′a + Γ′a bc V

′c
=
=J˜bq Jpa (∂b V p ) + (J˜bq ∂q Jpa )V p + (Jpa J˜bq J˜cr Γpqr − J˜bq J˜cr ∂q Jra )Jsc V s =
=J˜q Jpa (∂b V p ) + Jpa J˜q (J˜cr Jsc Γpqr V s ) + (J˜q ∂q Jpa )V p − (J˜q ∂q Jra )(J˜cr Jsc V s ) =
b b b b
=J˜bq Jpa (∂b V p ) + Jpa J˜bq (Γpqr V r ) = J˜bq Jpa (∇b V p ), (4.17)

which is a tensor transformation. Clearly, ∇b V a is tensor of valence (1, 1). The covariant
differentiation on covector fields is

∇b Wa = ∂b Wa − Γcba Wc (4.18)

which is obviously of valence (0, 2). To see (4.18), let us consider the differentiation on a
particular scalar, i.e. ∇a ϕ = ∂a ϕ with ϕ = V b Wb , which expands to

∇a ϕ =∇a (V b Wb ) = (∇a V b )Wb + V b (∇a Wb ) = (∂a V b + Γbac V c )Wb + V b (∂b Wa − Γcba Wc ) =

=(∂a V b )Wb + V b ∂a Wb = ∂a (V b Wb ). (4.19)

We therefore see that (4.18) is a consequence of the consistency between the Leibnitz rule
and (4.16). The covariant differentiation on tensors with more indices can be similarly
formulated, e.g.
∇a Vdbc = ∂a Vdbc + Γbap Vdpc + Γcap Vdbp − Γqad Vqbc . (4.20)
Let ϕ be a smooth scalar field, then clearly in each patch we have ∂a ∂b ϕ = ∂b ∂a ϕ. How-
ever, ∇a ∇b ϕ does not necessarily commute with ∇b ∇a ϕ. In fact, we have

∇a ∇b ϕ = ∂a ∂b ϕ − Γcab ∂c ϕ, ∇b ∇a ϕ = ∂b ∂a ϕ − Γcba ∂c ϕ

23
and hence
∇a ∇b ϕ − ∇b ∇a ϕ = (−Γcab + Γcba )∇c ϕ. (4.21)
We emphasize that the tensor in the bracket is the torsion tensor
c
Tab ≡ −Γcab + Γcba = −2Γc[ab] (4.22)

Einstein’s GR is torsion-free, i.e. Tbca = 0; but in other modified theories of gravity, it could
be non-vanishing, e.g. in Riemann-Cartan gravity.

4.2 Riemann Tensor

The Riemann tensor arises as a measure of the extent to which commutativity fails for the
second covariant derivative of a vector field.

Theorem 4.1. Let M be a manifold with a connection Γabc that is at least once differentiable in
each coordinate patch. Then there exists a unique tensor field Rabcd such that
d
(∇a ∇b − ∇b ∇a )Vc − Tab ∇d Vc = −Rabcd Vd (4.23)
d
for any smooth vector field Vc , where Tab is the torsion associated with ∇a .

Proof. Since

∇a ∇b Vc = ∂a (∇b Vc ) − Γpab ∇p Vc − Γpac ∇b Vp , ∇b ∇a Vc = ∂b (∇a Vc ) − Γpba ∇p Vc − Γpbc ∇a Vp ,

and Tab
c
= −Γcab + Γcba , we obtain
d
(∇a ∇b − ∇b ∇a )Vc − Tab ∇d Vc =(∇a ∇b − ∇b ∇a )Vc + (Γpab ∇p Vc − Γpba ∇p Vc ) =
=∂a (∇b Vc ) − ∂b (∇a Vc ) − Γpac ∇b Vp + Γpbc ∇a Vp . (4.24)

Let us expand

∂a (∇b Vc ) =∂a ∂b Vc − ∂a (Γdbc Vd ) = ∂a ∂b Vc − (∂a Γdbc )Vd − Γdbc (∂a Vd )

∂b (∇a Vc ) =∂b ∂a Vc − (∂b Γdac )Vd − Γdac (∂b Vd )
Γpac ∇b Vp =Γpac ∂b Vp − Γpac Γdbp Vd
Γpbc ∇a Vp =Γpbc ∂a Vp − Γpbc Γdap Vd .

After inserting these expressions into (4.24), we obtain (4.23) with

1
R d = ∂[a Γdb]c − Γp[a|c| Γdb]p (4.25)
2 abc

where A[a|b|c]d = 21 (Aabcd − Acbad ).

The action of ∇[a ∇b] on the contravariant vector fields can be obtained from the con-
sistency between (4.23) and (4.21) with ϕ = Vc W c , and the result is
1 c 1
∇[a ∇b] W d = Tab ∇c W d + Rabcd W c . (4.26)
2 2

24
The action of ∇[a ∇b] on tensors with higher valence can be obtained by repeating these
two basic actions, e.g.

1 f 1 1 1
∇[a ∇b] Ucde = Tab ∇f Ucde + Rabpd Ucpe + Rabq e Ucdq − Rabcr Urde . (4.27)
2 2 2 2
The Riemann tensor Rabcd has some symmetries. The obvious one, Rabcd = −Rbacd , can
be seen from the definition. Another symmetry is given below:
d
Proposition 4.2. If ∇a is torsion-free, then R[abc] = 0.

Proof. Since ∇a is torsion-free, we have ∇[a ∇b] Vc = − 12 Rabcd Vd . Let Vc = ∇c ϕ according to

(4.7), then
1 d
− R[abc] ∇d ϕ = ∇[a ∇b ∇c] ϕ = ∇[a ∇[b ∇c]] ϕ = 0
2
for any ϕ.
d
Note that R[abc] = 31 (Rabcd + Rbcad + Rcabd ), as Rabcd = −Rbacd .

d
Proposition 4.3 (Bianchi identity). If ∇a is torsion-free, then ∇[a Rbc]e = 0 .

Proof. For arbitrary V d , we have (by (4.27))

2∇[a ∇b] ∇c V d = −Rabcp ∇p V d + Rabq d ∇c V q .

d
Since R[abc] = 0, we have

d
2∇[a ∇b ∇c] V d = R[ab|q| ∇c] V q . (4.28)

On the other hand,

2∇a ∇[b ∇c] V d = (∇a Rbcq d )∇p V q + Rbcq d ∇a V q

whence
d
2∇[a ∇b ∇c] V d = (∇[a Rbc]qd )∇p V q + R[bc|q| ∇a] V q . (4.29)
Equating (4.28) and (4.29) we obtain the desired result.

4.3 Riemannian geometry

It is convenient in gravitation theory to distinguish three levels of geometry, with increasing
structure and complexity: (a) the underlying differentiable manifold, (b) the connection, (c)
the metric gab . For Riemannian geometry (b) and (c) are made compatible by ∇a gbc = 0
with ∇a torsion-free. In fact, ∇a can be uniquely determined from gab :

Theorem 4.4 (The fundamental theorem of Riemannian geometry). Let gab be the metric,
a symmetric, non-degenerate, differentiable tensor field, on a manifold M . There exists a unique
torsion-free connection ∇a on M such that ∇a gbc = 0.

25
Proof. ∇a gbc = 0 means
′ ′
∂a gbc − Γbab gb′ c − Γcac gbc′ = 0. (4.30)
By cyclic permutation of indices we also have
′ ′
∂b gca − Γcbc gc′ a − Γaba gca′ =0 (4.31)
′ ′
∂c gab − Γaca ga′ b − Γbcb gab′ =0.. (4.32)

Then (4.30)−(4.31)−(4.32) gives

′
2Γcbc gc′ a + ∂a gcb − ∂b gca − ∂c gab = 0. (4.33)

Since gab is non-degenerate there exists a unique symmetric tensor g ab such that gab g bc = δac ,
so g ad ×(4.33) gives the Levi-Civita connection (or Christoffel symbol)

1
Γdbc = g da (∂b gca + ∂c gba − ∂a gbc ). (4.34)
2
Thus if ∇a gbc = 0 then Γcab must certainly be of this form. Conversely if Γcab is given as
above, then ∇a gbc = 0, as desired.

When the Levi-Civita connection (4.34) vanishes, i.e. Γabc = 0 at P , we have ∂c gab
′
|P = 0,
i.e. the metric is a constant metric. Then after diagonalization, this metric at P can be
transformed into the flat ηab . The corresponding coordinates for such a flat metric to hold
at P are called the Riemann normal coordinates.
Remark 4.5. As a side remark, we discuss some useful formulas derived from the Levi-Civita
connection (4.34). Recall the matrix identity δ(ln det M ) = Tr(M −1 δM ), then if we take
M = (gab ), we have
g −1 δg = Tr[(gab )−1 δ(gab )] = g ab δgab , (4.35)
so
√ 1 1√
δ −g = (−g)−1/2 δg = −gg ab δgab . (4.36)
2 2
The Levi-Civita connection contracted on one index gives
1 1 √
Γbab = g bc ∂a gbc = √ ∂a −g. (4.37)
2 −g
This way, the covariant divergence is simplified to
1 √
∇c v c = ∂c v c + Γbab v a = √ ∂a ( −gv a ) (4.38)
−g
which allows us to integrate by parts even with integrands containing covariant derivatives.
For example,
√ i√
Z Z h
ab 4
S ∇a wb −gd x = ∇a (S wb ) − (∇a S )wb −gd4 x
ab ab
M MZ
√ √
Z
=− ab
(∇a S )wb −gd x +4
∂a ( −gS ab wb )d4 x (4.39)
M M

where the last term is a boundary term.

26
 We call the following contraction the curvature tensor:
′
Rabcd = Rabcd gdd′ . (4.40)

The curvature tensor possesses the following index-symmetries:

Rabcd =R[ab]cd (4.41)

R[abc]d =0 (4.42)
Rabcd =Rab[cd] (4.43)
Rabcd =Rcdab (4.44)

The first two relations are already known from above. For (4.43) we deduce from ∇a gcd = 0
that
′ ′
0 = 2∇[a ∇b] gcd = −Rabcc gc′ d − Rabdd gd′ d = −(Rabcd + Rabdc ).
To see (4.44), we use (4.42) and (4.43) to get

Rabcd = − Rbcad − Rcabd = Rcbad − Rcabd = −2Rc[ab]d

Rabcd = − Rbcad − Rcabd = Rbcda − Racdb = −2R[a|cd|b] ;

by comparing R[c|[ab]|d] and R[a|[cd]|b] , we obtain the symmetry (4.44).

We can count the independent variables in the curvature tensor as follows. Suppose
for the moment we are in n dimensions, then we have n4 components to start with. The
symmetry (4.41) restricts the number to n2 × n(n−1) 2
independent components; (4.44) restricts
further it to
1 n(n − 1)  n(n − 1)  1
× × + 1 = n(n − 1)(n(n − 1) + 2).
2 2 2 8
Finally, if two of the four indices in the Bianchi identity (4.42) are the same, then one of
three terms in (4.42) must be zero and (4.42) reduces to (4.41) or (4.43), so we only need
to consider (4.42) with four different indices. This way the Bianchi identity simply gives
1
4!
n(n − 1)(n − 2)(n − 3) constraints. Thus, the number of independent variables is
1 1 1
n(n − 1)(n(n − 1) + 2) − n(n − 1)(n − 2)(n − 3) = n2 (n2 − 1). (4.45)
8 4! 12
In 4D, the number is 20.
Two more basic tensors can be defined in terms of the curvature tensor:

Ricci tensor : Rab = Racbc = Racbd g cd (4.46)

Ricci scalar : R= Raa = Rab g ab ab cd
= Rabcd g g (4.47)

More explicitly,
Rab = ∂c Γcab − ∂a Γcbc + Γcab Γdcd − Γcad Γdbc (4.48)
By (4.44), we see that the Ricci tensor for a Riemannian geometry is necessarily symmetric,
Rab = Rba . The Einstein tensor is
1
Gab = Rab − gab R. (4.49)
2

27
 d
By contracting the Bianchi identity as g ad g ce ∇[a Rbc]e = 0, we get 2∇a Rab − ∇b R = 0, or

∇a Gab = 0. (4.50)

The Ricci tensor is the trace part of the curvature. In 4D, it carries half the total information;
the full curvature tensor has twenty independent components, while the Ricci tensor has
ten. The remaining components constitute the Weyl tensor

1
Cabcd = Rabcd − (ga[c Rd]b − gb[c Rd]a ) + Rga[c gd]b . (4.51)
3

It can be verified that Cabcd is traceless; also, the Cab cd is invariant under local rescalings
gab → e2ω gab , i.e. conformal transformations (Ex.). A metric is called conformally flat if its
Weyl tensor is zero.
The Weyl tensor can be identified with spin-2 field because of its algebraic symmetries.
From (4.51), we can check for Cabcd the same symmetries (4.41)–(4.44), which, together
with the traceless condition, define the spin-2 fields. The field equations for the spin-2 fields
Cabcd can be obtained from the Bianchi identity for the Weyl tensor in the following way.
We first invert (4.51) to express the Riemann curvature tensor in terms of the Weyl tensor
and the Einstein tensors,
1 1
Rabcd = Cabcd + (gac Gbd − gad Gbc − gbc Gad + gbd Gac ) + (gac gbd − gad gbc )R. (4.52)
2 3
Then the Bianchi identity becomes

2
∇[e Cab]cd = −gc[a ∇e Gb]d + gd[a ∇e Gb]c − gc[a ∇e Rgb]d . (4.53)
3
We will see in the next section that the RHS of (4.53) vanishes, so that ∇[a Cbc]de = 0, which
is also equivalent to Rab = 0. This clearly shows that the gravitons are spin-2 fields!

4.4 Lie derivatives and Killing vectors

Let P a and Qa be vector fields, then the Lie derivative of Qa with respect to P a is

LP Qa = P b ∇b Qa − Qb ∇b P a (4.54)

where ∇a is torsion-free. Since

P b ∇b Qa − Qb ∇b P a = P b ∂b Qa + P b Γabc Qc − Qb ∂b P a − Qb Γabc P c = P b ∂b Qa − Qb ∂b P a , (4.55)

we see that the Lie derivatives on vector fields are independent of the choice of connections.
The action of Lie derivative on covector fields is

LP Sa = P b ∇b Sa + Sb ∇a P b , (4.56)

This is also independent of the choice of connections.

28
 The action on scalars is simply LP ϕ = P a ∇a ϕ. Indeed, using the the Leibniz property,

LP (Qa Sa ) = (LP Qa )Sa + Qa (LP Sa ) =

=(P b ∇b Qa )Sa − (Qb ∇b P a )Sa + Qa (P b ∇b Sa ) + Qa (Sb ∇a P b ) = P b ∇b (Qa Sa ). (4.57)

The action of Lie derivatives on tensors of higher rank is given according to the follow-
ing scheme:

LP Qab =P c ∇c Qab − Qcb ∇c P a − Qac ∇c P b , (4.58)

c a
LP Sab =P ∇c Sab + Scb ∇c P + Sac ∇c P . b
(4.59)

Proposition 4.6. L[U,V ] = LU LV − LV LU .

This proposition tells us that the Lie derivatives form a Lie algebra. Recalling the discus-
sion about the non-coordinate basis in §2.3, we see that the basis vectors form a Lie algebra
means the Lie derivatives of the basis vectors do not drag the basis vectors out of the tangent
spaces!
As an application, we can use Lie derivatives to give an alternative proof of Thm. 4.4:

Alternative proof of Theorem 4.4. The following expression is connection independent:

LV gbc + 2∇[c V a gb]a = 2gab ∇c V a + V a (∇a gbc + ∇c gab − ∇b gac ).

Then, after multiplying by 12 g bd , we know from the connection independence that

1 1
∇c V d + V a g bd (∇a gbc + ∇c gab − ∇b gac ) = ∂c V d + V a g bd (∇a gbc + ∇c gab − ∇b gac ) .
2 2
Thus if ∇a gbc = 0, it then follows that

∇c V d = ∂c V d + V a Γdac

which is then uniquely determined.

Symmetries of a spacetime are described by Killing vectors K a , which are directions in

which one can move without the ’geometry changing’, i.e. an isometry:

LK gab = 0. (4.60)

To see that the geometry is not changing, we first write out the Lie derivative

LK gab = K c ∇c gab + gcb ∇c K a + gac ∇c K b = 0. (4.61)

Consider then a shift in the basis vectors, xa → xa + K a , then

dxa → dxa +dK a = dxa +∂b K a dxb +O(K 2 ), gab (x) → gab (x+K) = gab +K c ∂c gab +O(K 2 ).

Hence,

gab (x)dxa dxb → (gab + K c ∇c gab + gcb ∇c K a + gac ∇c K b + O(K 2 ))dxa dxb ,

29
then (4.61) means gab is unchanged. Obviously, this Killing equation can also be re-expressed
as
∇(a Kb) = 0. (4.62)

We can always choose the suitable coordinates such that the Killing vector is K 0 =const.
and K i = 0. Indeed, given the coordinates xa and the Killing vector K a , we can change to
a
new coordinates x̃a through K̃ a = ∂∂xx̃ b K b such that

∂ x̃0 b ∂ x̃i b
K = const., K =0
∂xb ∂xb
which is a set of differential equations that are solvable. Then from (4.61) we see that ∂0 gab =
0. In words, if the metric is independent on a particular coordinate, then the translation
along this direction is a Killing vector.
n(n+1)
Theorem 4.7. An n-dimensional spacetime has at most 2
Killing vectors.

Proof. From the definitions of Riemann tensor and of Killing vector, we know that

∇a ∇d Kc = −Rcdab Kb .

(From the property (4.42), we must have ∇[c ∇d Ka] = 0 if such a Ka exists. So, −Rcdab Kb =
(∇c ∇d − ∇d ∇c )Ka = −∇c ∇a Kd − ∇d ∇c Ka = ∇a ∇d Kc .) Then in the Taylor expansion of
Ka , only those terms involving Ka and ∇c Ka remain:

Ka = Aab Kb + Ba bc ∇c Kb

for some functions Aab and Ba bc . Therefore, there are n Ka , and n(n−1)
2
∇c Ka as constrained
by ∇(c Ka) = 0.

We say the space(time) with the maximal n(n+1)

2
Killing vectors a maximally symmetric
space. A maximally symmetric space is homogeneous and isotropic at each point of it.
More generally, we can define Killing tensors Ka...b satisfying

Ka...b = K(a...b) , ∇(c Ka...b) = 0, (4.63)

and similarly we have pc ∇c (Ka...b pa ...pb ) = 0. A Killing tensor can have “square root” in the
following sense: We define the principal Killing-Yano tensor Yab by

1
∇c Yab = 2gc[b ξa] , ξa = − ∇c Y ca . (4.64)
3
In 4D, the Killing tensor can be constructed from the Killing-Yano tensor as

1
Kab = ((∗Y ) · (∗Y ))ab = Qab − gab Qcc , (4.65)
2
where Qab = Yac Yb c and (ω · ω)ab = 1
ω
p! aγ1 ...γp
ω aγ1 ...γp for (p + 1)-form ω.

30
5 Geodesics
Let Ux and Ux′ be a pair of overlapping coordinate patches in a differentiable manifold M .
A curve γ in M is parametrized with parameter u according to xa ≡ xa (u) in the Ux patch.
a ′b
Consider the tangent vectors, ξ a (u) = dx du
in Ux and ξ ′b (u) = dx
du
in Ux′ . Then ξ a transforms
′b
as a contravariant vector, ξ ′b = ∂xa ξ a as can be seen from the chain rule for derivatives.
∂x

Let Γabc (xd (u)) be the restriction of that connection to the curve γ in the patch Ux . Sup-
pose V a (x(u)) is a contravariant vector field given along the curve γ such that V ′b (u) =
∂x′b a dV a (u)
∂x a V (u) in U x ∩ U x ′ . Then
du
along γ does not transform properly as a tensor between
patches; however the absolute derivative,

DV a dV a
= + Γabc ξ b V c (5.1)
Du du
does transform properly. This is simply the change-of variable version of the covariant
derivative. We can also denote the above absolute derivative along the tangent vector ξ b as
ξ b ∇b V a .
A geodesic on a manifold is defined to be a differentiable curve γ such that its tangent
vector ξ a (u) satisfies
Dξ a dxa (u)
= λ(u)ξ a , ξ a (u) = (5.2)
Du du
for some function λ(u). Thus for a geodesic the absolute derivative of the tangent vector
always points in the direction of the tangent vector. The geodesic equation (5.2) can be
rewritten in the more familiar form
d2 xa (u) b
a dx dx
c
dxa
+ Γ bc = λ(u) . (5.3)
du2 du du du
In GR, small freely-falling bodies move along geodesic trajectories. The curvature of space-
time makes itself felt through the connection term appearing in the geodesic equation.
Let us consider the effect of a change in parameter u 7→ u(σ) along the curve. Then

dxa dσ dxa d2 xa d2 σ dxa  dσ 2 d2 xa

= , = + ;
du du dσ du2 du2 dσ du dσ 2
consequently, the geodesic equation (5.3) becomes

d2 σ dxa  dσ 2 d2 xa b
a dx dx
c
dσ 2 dxa dσ
+ + Γ bc = λ(u) .
du2 dσ du dσ 2 dσ dσ du dσ du
d2 σ
If σ(u) is chosen as an affine parameter along the geodesic such that du2
= λ(u) dσ
du
, then we
can solve for Z Z u β
σ(u) = eλ(α) dαdβ.

With this choice we obtain

d 2 xa b
a dx dx
c
+ Γ bc = 0. (5.4)
dσ 2 dσ dσ

31
When the affine parameter σ = τ is the proper time, it is straightforward to show (Ex.) that
the geodesic equation (5.4) is the Euler-Lagrange equation of the Lagrangian3

L = gab (xc )ẋa ẋb (5.5)

where the dot means the differentiation wrt the proper time.
A curve γ with tangent vector ξ a is a geodesic if

ξ b ∇b ξ a = 0 (5.6)

for some affine parametrization σ. If ξ a ξa ̸= 0, then by a suitable affine parametrization we

can arrange always that either ξ a ξa = 1 or ξ a ξa = −1, depending on whether the geodesic
is time-like or space-like. A geodesic for which ξ a ξa = 0 is called a null geodesic. We can
further define the exponential map exp : Tp → M ; ξ a 7→ exp ξ a . If exp can be defined in
every tangent space Tp of p ∈ M , then M is geodesic complete.
A Killing vector Ka generates a conserved quantity along the geodesics. For example,
consider a particle with 4-momentum pa moving along a timelike geodesic pa ∇a pb = 0,
then
pa ∇a (Kb pb ) = pa Kb ∇a pb + pa pb ∇a Kb = pa pb ∇a Kb = pa pb ∇(a Kb) = 0, (5.7)
so Kb pb is a conserved quantity. What happens when this particle is not moving on a
geodesic? In this non-geodesic case, let us write pa = mξ a for a timelike ξ a tangent to
a curve γ and we have

ξ a ∇a (pb Kb ) = m(Kb ξ a ∇a ξ b + ξ b ξ a ∇a Kb ) = mKb ξ a ∇a ξ b = mKa αa (5.8)

where we have defined the acceleration field, αa = ξ b ∇b ξ a . Let us denote the magnitude of
αa as α = (−αn αn )1/2 . Then we have the following interesting inequality [17, 7]

|ξ b ∇b (pa Ka )| =|mKa αa | = |mhab αa K b | ⩽

p p
(5.9)
p
⩽ −hab αa αb −m2 hab K a K b = α (pa Ka )2 − m2 (K a Ka )

where hab = gab − ξa ξb and we have used the Schwarz inequality. Since ξ a is timelike, we
know that pa Ka > 0, so this inequality yields

|ξ b ∇b (pa Ka )| ⩽ α(pa Ka ). (5.10)

Using this result, we can deduce that the total integrated acceleration TA(γ) along a curve
γ in the following sense is bounded:
Z b
|ξ ∇b (pa Ka )|
Z Z
TA(γ) = αds ⩾ aK
⩾ | ξ b ∇b (ln(pa Ka ))ds|. (5.11)
γ γ p a γ

In other words, if γ passes through two points p1 and p2 , the total integrated acceleration is at
least |ln(pa Ka )|p2 −ln(pa Ka )|p1 |. This inequality is important for restricting some unphysical
behaviors of particle motion in GR!
3
However, we cannot directly interpret this as the proof for a shortest path, because the dimension of this
Lagrangian is [length]2 .

32
 When the tangent vector la of a geodesic is null, we need to consider the conformal
isometry gab → Ω2 gab . The infinitesimal generator ζa of such a conformal isometry satisfies
the conformal Killing equation
∇a ζa + ∇b ζa = ωgab (5.12)
where ω can be determined by simply taking the trace of (5.12), i.e. ω = 21 ∇a ζa . Then the
conformal Killing equation gives
lb ∇b (ζa la ) = 0. (5.13)
The conformal Killing vectors ζ a can be used to construct the conserved quantities along
the null geodesics.

5.1 Geodesic deviation

Consider a one-parameter family of geodesies xa (σ, ρ), the geodesic congruence, where ρ
labels the various geodesics. Let γ1 ∼ xa (σ, ρ) and γ2 ∼ xa (σ, ρ + ϵ) be two neighboring
curves, with ϵ small. Define
xa (σ, ρ + ϵ) − xa (σ, ρ)
η a (σ, ρ, ϵ) = ,
ϵ
then we have the deviation vectors
∂xa (σ, ρ)
η a (σ, ρ) = lim η a (σ, ρ, ϵ) = . (5.14)
ϵ→0 ∂ρ

Let γ be the geodesic corresponding to some fixed ρ0 , then η a (σ) ≡ η a (σ, ρ0 ) is a con-
travariant vector field along γ which measures the infinitesimal displacement of neighboring
geodesics from γ as σ varies. In fact, we have the equation of geodesic deviation:

D2 η a (σ)
= Rbcda (x(σ))ξ b η c ξ d . (5.15)
Dσ 2

To show this, we start from the differentiation of the geodesic equations:

∂2 ∂ a ∂ h a ∂xb ∂xc i
− 2 x (ρ, σ) = Γ (x(σ, ρ)) =
∂σ ∂ρ ∂ρ bc ∂σ ∂σ
∂xb ∂xc ∂ a ∂xb ∂ 2 xc
= Γbc (x(σ, ρ)) + 2Γabc (x(σ, ρ)) .
∂σ ∂σ ∂ρ ∂σ ∂σ∂ρ

By the definitions of η a and ξ a , we obtain

d2 η a (σ) a b dη
c
a b c d
= −∂c Γbd ξ ξ η − 2Γbc ξ . (5.16)
dσ 2 dσ

33
Next, we compute the absolute derivative:
Dη a dη a
= + Γapq ξ p η q ,
Dσ dσ
D2 η a d  Dη a  r
a Dη s d  dη a a p q

a
 dη r
r p q

2
= + Γ rs ξ = + Γ pq ξ η + Γ rs + Γ pq ξ η ξs =
Dσ dσ Dσ Dσ dσ dσ dσ
d2 η a dξ p
dη q
dη r
= 2 + ∂d Γapq ξ d ξ p η q + Γapq η q + Γapq ξ p + Γars ξ s + Γars Γrpq ξ p η q ξ s =
dσ dσ dσ dσ
d2 η a dη q
= 2 + ∂d Γapq ξ d ξ p η q + Γapq (−Γars ξ r ξ s )η q + 2Γapq ξ p + Γars Γrpq ξ p η q ξ s =
dσ dσ
d2 η a dη c
= 2 + ∂b Γacd ξ b η c ξ d + 2Γabc ξ b + (Γabr Γrcd − Γacr Γrbd )ξ b η c ξ d . (5.17)
dσ dσ
Combining (5.16) and (5.17) entails (5.15).
There exists a much simpler derivation of the equation of geodesic deviation, by using
the commutators of absolute derivatives. Let xa (α, β) be a 2-surface S parametrized by α
and β, and let V a (α, β) = V a (x(α, β)) be a vector field defined on the surface S. Then we
can define the absolute derivatives
DV a dV a DV a dV a
= + Γabc V b Ac , = + Γabc V b B c (5.18)
Dα dα Dβ dβ
where Ac = ∂xc /∂α, B c = ∂xc /∂β.
D 2 D2
Theorem 5.1. ( DαDβ − DβDα
)V d (α, β) = Rabcs Aa B b V c .

Instead of calculating by brute force, let us take a geometric path to prove this theorem.
DV b
Lemma 5.2. If V a (x(α)) is a vector on the curve xa = xa (α), then Dα
= Aa ∇a V b .

Proof.
DV b dV b dV b
= + Γbad V a Ad = Aa a + Γbad V a Ad = Aa ∇a V b .
Dα dα dx

Lemma 5.3. Aa ∇a B b − B a ∇a Ab = 0.

Proof. We know that Aa ∇a B b − B a ∇a Ab is connection independent, so that

∂B b ∂Ab ∂xb ∂xb

Aa ∇a B b − B a ∇a Ab = Aa ∂a B b − B a ∂a Ab = − = − = 0.
∂α ∂β ∂α∂β ∂β∂α

Proof of Theorem 5.1. Using the Lemma 5.2, we can rewrite the geodesic deviation equation
as
Aa ∇a (B b ∇b V d ) − B a ∇a (Ab ∇b V d ) = Rabcd Aa B b V c .
By the Leibniz rule, however, we have

(Aa ∇a B b ∇b − B a ∇a Ab ∇b )V d = Aa B b (∇a ∇b − ∇b ∇a )V d + (Aa ∇a B b − B a ∇a Ab )∇b V d .

Then the result follows by using Lemma 5.3.

34
 Now we can give a simplified derivation of the geodesic deviation equation. Denote
a (σ,ρ) a (σ,ρ)
ξ ≡ ∂x ∂σ
a
and η a ≡ ∂x ∂ρ so that Dξ a /Dσ = 0. Then by Theorem 5.1, we have
Dη a a

Dσ
= Dξ
Dη
, and hence

D2 η a D2 ξ a D2 ξ a D2 ξ a
= = − = Rbcda ξ b η c ξ d .
Dσ 2 DσDη DσDη DηDσ
This proof strategy can be applied for the deviations in the congruence of non-geodesic
curves (cf. §5.3).
In the case of special time-like geodesics with ξ a = (1, 0, 0, 0), we see that the deviation
vector orthogonal to ξ a is a 3D vector, η a = (0, η i ), and the geodesic deviation equation
becomes
D2 η i
2
= R0j0i η j . (5.19)
Dσ
This equation is called the Jacobi equation and its solutions η i are called the Jacobi fields.
Two points p, q on a geodesic γ are conjugate if there exists a Jacobi field along γ that vanishes
nowhere but at both p and q. There is a famous result:

Proposition 5.4 ([2]). Let q and p be a pair of conjugate points on a non-spacelike geodesic γ. If
there is no other conjugate points of q between q and p, then the non-spacelike geodesic between q
and p is maximal, i.e. with maximal length.

The upshot of this proposition is that in a Lorentz geometry, a geodesics is the longest
path!
For the conjugate points to occur, one naturally expects some kind of converging behav-
ior of the geodesics, which can be made rigorous in the next subsection using Raychaudhuri
equations.

5.2 Raychaudhuri equation

a
Consider a geodesic congruence xa (σ, ρ). Let ξ a = dx dσ
be the normalized time-like tangent
vector of a geodesic in this congruence. We see from ∇b (ξ a ξa ) = 0 that ξ b ∇b ξa = 0, so if
we define Bab = ∇b ξa , then Bab ξ a = Bab ξ b = 0, i.e. Bab is space-like. On the other hand,
a
the vector η a = dx
dρ
is the deviation vector which is chosen to be orthogonal to ξ a . This
choice is always possible, for if we choose ηa ξ a |σ=0 = 0, because obviously dσ
d
(ηa ξ a ) = 0, we
see that the orthogonality is preserved for all λ. This orthogonality then implies

[ξ, η]a = ξ b ∂b η a − η b ∂b ξ a = ξ b ∇b η a − η b ∇b ξ a = 0 (5.20)

where the substitution ∂ → ∇ is guaranteed by the torsion-free condition. Thus, we find

ξ b ∇b η a = η b ∇b ξ a = η b B ab (5.21)

which means the the tensor Bab characterize the ways of transports in the deviation vectors.
In the way of last paragraph, we can choose ξ a = (1, 0, 0, 0), and then the deviation
vectors are space-like. More generally, let us introduce the projector

hab = gab + ξa ξb (5.22)

35
to pick out the tensor components orthogonal to ξ a . With these, we can define

expansion : θ = B ab hab (5.23)

1
shear : σab = B(ab) − θhab (5.24)
3
twist : ωab = B[ab] (5.25)

thereby,
1
Bab = θhab + σab + ωab . (5.26)
3
The expansion θ measures the expansion (θ > 0) or contraction (θ < 0) of the geodesics
nearby. More precisely, it measures the changing rate of the sectional volume of the geodesic
congruence:
1 d
θ= δV. (5.27)
δV dσ
To see this, we can choose a geodesic γ0 in the congruence as the reference geodesic and
consider a particular parameter σp at the point p on this γ0 . Then the 3-submanifold xa =
xa (σ0 , y i ) with constant σ0 of the geodesic congruence is called the section near p. Now the
a
deviation vectors ηia = ∂x ∂y i
, (i = 1, 2, 3, ) are tangent to the section, and the induced metric
on the section is
hij = gab ηia ηjb = hab ηia ηjb .
√
So the sectional volume is just δV√= hd3 y. Since d3 y should be invariant in the comoving
dh
frame, we have δV1 dσ d
δV = √1h dσ d
h = 21 hij dσij , where

dhij
= ξ a ∇a (gab ηia ηjb ) = ηia ηjb (∇a ξb + ∇b ξa ) = ηia ηjb (Bab + Bba )
dσ
where we have used (5.20). We therefore obtain (5.27).
The shear σab characterizes the deformation of a ball with its volume kept constant. The
twist ωab characterizes the twist of the geodesic congruence about the reference geodesic.
These quantities are the geometric analogs of those in fluid mechanics.
Let Σ be a 3D submanifold of M defined by setting f =const.; Σ is also called a codimension-
1 hypersurface Σ of M . We say a vector ξ a is hypersurface-orthogonal to some Σ iff

ξ[a ∇b ξc] = 0. (5.28)

Indeed, the normal vector ξ a = ∇a f to Σ satisfies the above (5.28). We can readily see that if
the twist ωab = ∇[b ξa] = 0, then the geodesic congruence is locally hypersurface-orthogonal
to some Σ.
Now we can consider the changes in the expansion θ along a geodesic. First, we can
compute

ξ c ∇c Bab = ξ c ∇c ∇b ξa =ξ c ∇b ∇c ξa − Rcbad ξ c ξd =
=∇b (ξ c ∇c )ξa − (∇b ξ c )∇c ξa − Rcbad ξ c ξd = −B cb Bac − Rcbad ξ c ξd .

Multiplying by hab , we obtain from above

dθ
ξ c ∇c θ = = −B cb Bac hab − Rcbad ξ c ξd hab = −B ca Bac − Rcd ξ c ξ d
dσ

36
where
1 1 1
B ca Bac =( θhca + σ ca + ω ca )( θhac + σac + ωac ) = θ2 + σca σ ca − ωca ω ca
3 3 3
where we have used the fact that hab , σab , ωab are orthogonal to each other (because σab , ωab
are space-like and their traces vanish, they are orthogonal to hab ; σ ab is symmetric, while
ωab is antisymmetric, ao σ ab ωab = 0.) We thus obtain the Raychaudhuri equation

dθ 1
= − θ2 − σab σ ab + ωab ω ab − Rcd ξ c ξ d . (5.29)
dσ 3

Note that similar calculations can be performed for ξ c ∇c σab and ξ c ∇c ωab . A remark is that
in ξ c ∇c ωab the Weyl tensor replaces the curvature tensor. In other words, the Weyl tensor
causes shear, and indirectly influences the expansion through the shear term in the Ray-
chaudhuri equation.
For null geodesic congruence, let us introduce the double-null decomposition. For a
null vector la , since la la = 0, we see that la is orthogonal to la ; in other words, some part
na of la lies in the 3D submanifold orthogonal to la . We can choose the ‘gauge’ condition
na na = 0, na la = 1. This way, the deviation vectors η defined by ηa la = 0 and ηa na = 0
span a 2D submanifold T⊥ of M , and we can consider the projector on T⊥ :

P ab = δ ab + na lb + la nb . (5.30)

Given a deviation vector η a = Pba η b in T⊥ , we have

ξ b ∇b η a = ξ b ∇b (P ac η c ) = P ac ξ b ∇a η c = P ac η c ∇a ξ b = P ac B bc η c = P ac B bc P cd η d ≡ B̃ ad η d .

Thus if the deviation vectors η a are in T⊥ at the outset, then they will stay in T⊥ along the
geodesic congruence. As in the time-like case, we can decompose B̃ ab into
1
B̃ ab = θ̃P ab + σ̃ ab + ω̃ ab (5.31)
2
where the expansion, shear and twist are respectively

θ̃ =B̃ aa (5.32)
1
σ̃ab =B̃(ab) − θ̃Pab (5.33)
2
ω̃ab =B̃[ab] (5.34)

As in the time-like case, we immediately see that if ω̃ab = 0, then la is hypersurface-

orthogonal to a class of null hypersurfaces (parameterized by the transport of na ).
Unlike the time-like congruence, now the space-like submanifold T⊥ orthogonal to la
and na is 2D. Let η (1) , η (2) be two linearly independent vectors spanning a basis of T⊥ , and
(1) (2)
define an area element a = ϵabcd la nb ηc ηd . Then
Da e (2) (2) (2)
=l ∇e a = ϵabcd la nb le ∇e (ηc(1) ηd ) = ϵabcd la nb [(le ∇e ηc(1) )ηd + ηc(1) le ∇e ηd ] =
Dσ
=ϵabcd la nb (B̃c e ηe(1) ηe(2) + ηc(1) B̃d e ηe(2) ) = 2θ̃a. (5.35)

37
Thus θ̃ characterizes the change in the area elements.
The Raychaudhuri equation for null congruence can be similarly calculated, and the
result is
dθ̃ 1
= − θ̃2 − σ̃ab σ̃ ab + ω̃ab ω̃ ab − Rcd lc ld . (5.36)
dσ 2

5.3 Fermi-Walker transport and non-geodesic curves

Let S ⊂ M be a 3D hypersurface.4 At a point p ∈ S, we consider a vector na normal to
S and the coordinates X a = (t, X i ) near p. Then, along a geodesic with na as a tangent
vector, we have
1
nb ∇b (na X a ) = na X b ∇b na = X b ∇b (na na ) = 0. (5.37)
2
This means that na is always orthogonal to X i . Such a coordinate system is called the
Gaussian normal coordinates.
Another useful coordinate system is the Fermi coordinates taking the rotation effects
into account. Let ea(b) be the orthonormal tetrad of an observer such that ea(0) = ξ a is the
unit vector tangent to the observer’s worldline. The tetrad satisfy ea(b) ea(c) = η(b)(c) . Now
a
consider an observer with 4-velocity ua and 4-acceleration aa = du dτ
, and suppose that

ξ a ∇a eb(c) = −Ωab eb(c) . (5.38)

Since ξ a ∇a (ea(b) ea(c) ) = ξ a ∇a (η(b)(c) ) = 0, we see that Ωab = −Ωba , and hence we can take
Ωab as a 4-rotation of the tetrad. We can decompose Ωab into temporal and spatial parts:

Ωab = ua ξ b − ξ a ub + ω ab

where ω ab = −ω ba stands for spatial 3-rotation (not the twist!) and ω ab ξb = 0. Because
ξ a ∇a ξ b = ab , we have Ωab ξb = −aa = ua , so that
Dea(c)  
= ξ a ab − aa ξ b − ω ab eb(c) ≡ −Ωab eb(c) . (5.39)
Dσ
The condition Ωab = 0 is the condition for the parallel transport of the vector ea along the
curve. Obviously, an affinely parametrized geodesic has the property that its tangent vector
is parallelly transported along the curve. If ω ab = 0, aa ̸= 0, (i.e. no spatial rotations) then
the tetrad is Fermi-Walker transported. The absolute derivative DF ()/Dσ defining the Fermi-
Walker transport is called the Fermi derivative,

DF V a DV a
= + (aa ξ b − ξ a ab )Vb . (5.40)
Dσ Dσ
Clearly, if V a is orthogonal to ξ a , then DF ()/Dσ = D()/Dσ.
Consider then a spinning observer measuring time intervals according to his proper
time, i.e. T = τ . If we take σ = τ , then the unit tangent vector na to the space-like geodesic
4
A 3D hypersurface S is a 3D manifold embedded in the 4D manifold M . Let i : S ,→ M be such an
embedding, then the metric i∗ gab induced on S is called the first fundamental form. Cf. §6.4.

38
at a particular τ can be decomposed in terms of the tetrad ea(b) into n(i) = na ea(i) , n(0) = 0,
and we define the Fermi coordinates by

(T, X, Y, Z) = (τ, sn(1) , sn(2) , sn(2) ) (5.41)

where s is the length parameter along the spatial geodesic.

If the basis vectors are Fermi-Walker transported, these basis vectors are then transported
along a (possibly non-geodesic) curve and keep the basis property. This way, one can use the
Fermi derivatives to study the non-geodesic curves just like the geodesics. Let us consider
then a time-like curve γ with a unit time-like tangent vector V a , and a nearby curve with
tangent vector Z a . From the definition of absolute derivative along γ, we know

DZ b
= V a ∇a Z b = (∇b V a )Z b (5.42)
Dσ
where we have assumed LV Z a = V b ∇b Z a − Z b ∇b V a = 0. Let hab = δba + V a Vb be the
projection to the horizontal subspace Hp ⊂ Tp normal to V a , and let Z⊥a = hab Z b . We have
therefore
DZ⊥b
hab = (∇b V a )Z⊥b , (5.43)
Dσ
and by using the absolute derivatives we have

D  b DZ⊥c 
hab hc = hab V d ∇d [(∇c V b )hce Z e ] = hab V d ∇d [(∇c V b )(δec + V c Ve )Z e ] =
Dσ Dσ
=hab (∇d ∇c V b Z⊥c V d + ∇c V b ∇d V c Ve Z e V d + ∇c V b V c ∇d Ve Z e V d + ∇c V b hce ∇d Z e V d ) =
=hab (∇d ∇c V b Z⊥c V d − ∇c ∇d V b Z⊥c V d + ∇c ∇d V b Z⊥c V d +
+ ∇c V b ∇d V c Ve Z e V d + ∇c V b hce ∇d Z e V d + V c ∇c V b V d ∇d Ve Z e ) =
=Rdcba V d Z c V b + hab ∇c V̇ b Z⊥c + V̇ a V̇b Z⊥b (5.44)

where V̇ a = V b ∇b V a . This (5.44) is the deviation equation for general curves; clearly, for
geodesics V̇ a = 0 and this equation reduces to the geodesic deviation equation (5.15).
In terms of Fermi derivatives, the above equations (5.43) and (5.46) can be written as
DF Z⊥a
=∇b V a Z⊥b , (5.45)
Dσ
DF2 Z⊥a
2
=Rdcba Z c V b V d + hab ∇c V̇ b Z⊥c + V̇ a V̇b Z⊥b . (5.46)
Dσ
If we work in the Fermi coordinates (5.41), then these equations can be further reduced to
their spatial components and the Fermi derivatives become ordinary derivatives:

dZ i
=∇j V i Z j , (5.47)
ds
d2 Z i
=R0j0i Z j + ∇j V̇ i Z j + V̇ i V̇j Z j . (5.48)
ds2
Notice that (5.47) is a first-order ODE, so we can write the general solution as

Z i (s) = Aij (s)Z j |q (5.49)

39
where Z j |q is the value at a point p and Aij satisfies dAij /ds = ∇k Vi Akj . In terms of the
matrix Aij , the deviation equation can be rewritten as

d2 Aij
2
= R0i0k Akj + ∇k V̇ i Akj + V̇ i V̇k Akj . (5.50)
ds
Likewise, the expansion, shear and twist can be respectively express in the Fermi coor-
dinates as
d
θ =(det A)−1 (det A), (5.51)
ds
d 1
σij =A−1
k(j Ai)k − δij θ, (5.52)
ds 3
d
ωij =A−1
k[j Ai]k . (5.53)
ds
In deriving these expressions, we have used ωab = hac hbd ∇[d Vv] and Bab = ∇b Va = hac hbd ∇(d Vv) .
By taking the trace of the symmetric part of A−1 kj ×(5.50), we obtain the Raychaudhuri equa-
tion
dθ 1
= − θ2 − σab σ ab + ωab ω ab − Rab V a V b + ∇a V̇ a . (5.54)
ds 3
where (i, j, k, ...) is generalized to (a, b, c, ...), but multiplying −gij V i V j contributes a minus
sign.
Next, let us take a second look at the Jacobi equation (5.19) for geodesics deviations.
Suppose Aij |q = 0, then at the conjugate point p of q we must have Aij |p = 0; by (5.51), this
happens when θ tends to ±∞, as d(det A)/ds is finite. For example, if θ < 0 initially and
dθ/ds < 0, then by the Raychaudhuri equation θ will tend to −∞, and hence there must
be a conjugate point provided that the parameter s is large enough.
The similar analysis can of course be done for general null curves [2].

5.4 Spinning test particles

The Fermi coordinates are chosen for a spinning observer, while the Fermi-Walker trans-
port requires vanishing spatial rotations. It is desirable to understand also the relevant equa-
tions for spinning test particles, as the spinning motion is a very common and actual phe-
nomenon. The geodesic equations and the geodesic deviation equations we have obtained
so far can be combined to derive such equations.
We first write down the equations of motion for spinning test particles, now known as
Mathisson-Papapetrou equations,

DP a 1
= Rabcd S cd V b . (5.55)
Ds 2
ab
In this equation, V a is a unit time-like tangent vector and P a = mV a + Vb DS
Ds
is the 4-
momentum of a spinning test particle with spin tensor S defined by
ab

DS ab
= P aV b − P bV a. (5.56)
Ds

40
The above two equations hold in the so-called pole-dipole approximation in which only the
dipole components of the energy-momentum tensor of the particle are taken into account.
If P a = mV a , then (5.55) and (5.56) become respectively

DV a 1 a DS ab
= R S cd V b , = 0. (5.57)
Ds 2m bcd Ds
Proof of (5.57). Consider the following transformation of parameters from s to τ ,
dxa dxa Dη a
= +β
ds dτ Dτ
where η a is the deviation vector. Performing a covariant derivative on both sides gives

DV a DV a D2 η a
= +β .
Ds Dτ Dτ 2
a
By the geodesic equations, we see that DV
Dτ
= 0, while to the last term the geodesic equations
can be applied. By setting S = s̄(V η − V b η a ) and β = s̄/m, we get the first equation of
ab a b

(5.57). The second equation can be obtained by taking covariant derivative of the Frenkel
condition
Sab V a = 0 (5.58)
and using again the geodesic equations.
ab
We therefore see that a spinning particle moves along a geodesic if the condition DS Ds
=
0 or Sab V = 0 is satisfied. Obviously, a spinning test particle would surely deviate from
a
ab
the the geodesics if DS
Ds
̸= 0. So we need to ask for the generalized deviation equations
and Raychaudhuri equations for spinning test particles. Indeed, adding only the spinning
degrees of freedom, such generalizations are possible, cf. [18, 19]. The strategy is simply to
perform the similar derivation as above with the geodesic equations replaced by Mathisson-
Papapetrou equations (Ex.).

6 Einstein’s field equations

We want to construct a theory of gravity that is consistent with the demands of special
relativity and that reduces to the Newtonian theory of gravity in the appropriate limit, i.e.
v/c→0
GR? / Newtonian Gravity

G→0 G→0

 v/c→0 
Special Relativity / Newtonian Mechanics

To this end, we try to preserve desirable features from Newtonian gravity and special rela-
tivity.
First, in special relativity we have the 4D Minkowski space-time with a metric ηab . By
use of ηab we can define proper-time along any curve and distinguish, at each point, time-
like, null and space-like vectors. The matter content of space-time is specified by the energy

41
momentum tensor, Tab , and the conservation of energy-momentum is guaranteed by the
conservation equation ∂Tab /∂xa = 0. In special relativity a free particle subject to no forces
moves on a time-like straight line.
In GR, a particle cannot be subject to no forces, as gravity is universally present; we
consider instead the freely falling particles, particles subject to no force except gravity. In
Newtonian gravity, a freely falling particle in a gravitational field with potential Φ moves
according to
mai = Fi = −m∇i Φ (6.1)
where m in the first term is the inertial mass, while in the third therm is the gravitational
mass. Canceling the masses (i.e. the equivalence principle), we obtain the equation for the
path of a freely falling particle,
d 2 xi
= −∇i Φ. (6.2)
dt2
We may therefore interpret Newtonian gravity as saying that a gravitational field deter-
mines a preferred class of paths which sufficiently small objects (small enough that their
own gravitational field does not interfere) must follow. In GR, we should retain this feature
by requiring that the new theory should have an affine connection that defines the preferred
class of paths as its geodesics. If the connection is to be related to ∇i , then the curvature, being
the derivative of the connection, must be related to second derivatives of Φ.
We must also seek an analogue of the field equation of Newtonian gravity,

∇2 Φ = 4πGρ (6.3)

relating the field to its source, the matter density ρ. The analogue in GR must relate the
curvature to the matter content specified by Tab .
In summary, GR is expected be based on a 4D manifold with a non-degenerate metric
tensor gab and an affine connection Γabc . The metric determines the proper time as measured
by physical clocks. We shall assume that connections are metric-compatible, ∇a gbc = 0,
and torsion-free for simplicity. By the fundamental theorem of Riemannian geometry we
know that in this case the connection is uniquely determined by the metric. The matter
content will be specified by a symmetric tensor Tab , which is assumed to be conserved,

∇a T ab = 0. (6.4)

Since this equation involves the connection, it embodies an assumption about the interaction
of matter and the gravitational field. Finally, we must seek a field equation of the form

{tensor constructed from curvature} = G × {tensor constructed from matter} (6.5)

that reduces to (6.3) in the Newtonian limit.

6.1 Einstein’s equations from geodesic deviation

More explicitly, let us consider the relative motion of freely falling particles. In Newtonian
gravity, suppose we have a family of paths xi (σ, ρ), i = 1, 2, 3 where ρ labels the paths and

42
σ is time along the paths. Let V i = ∂xi /∂σ be the velocities and let η i = ∂xi /∂ρ be a vector
connecting infinitesimally neighboring paths. Then equation of the paths (6.2) becomes

∂V i
= −∇i Φ. (6.6)
∂σ
Now if ρ2 = ρ1 + δρ with small δρ, then
∂V i ∂V i ∂ 2V i ∂V i ∂ 2ηi
≈ +δρ = +δρ
∂σ ρ2 ∂σ ρ1 ∂σ∂ρ ρ1 ∂σ ρ1 ∂σ 2 ρ1

and
∂V i ∂V i
= −∇i Φ|ρ1 , = −∇i Φ|ρ1 − δρη j ∇i ∇j Φ|ρ1
∂σ ρ1 ∂σ ρ2

from which we have that

∂ 2ηi
= −η j ∇i ∇j Φ. (6.7)
∂σ 2
Recall that in a 4D manifold M with curvature, the analogue of (6.7) is the equation of
geodesic deviation
D2 η a
= Rbcda V b η c V d ≡ Φac η c . (6.8)
Dσ 2
We see that role of ∇i ∇j Φ is now played by Φca = Rbcda V b V d . In Newtonian gravity, the
field equation (6.3) relates the trace of ∇i ∇j Φ to the matter density ρ. Here,

Φaa = Rbada V b V d = Rbd V b V d (6.9)

which suggests that the field equations of GR should relate the Ricci tensor to the matter
density. We know from special relativity that the matter density as measured by an observer
with four-velocity V a is Tab V a V b , so we might try

Rab V a V b ∝ Tab V a V b .

If this were to hold for all V a we would be led to

Rab = κTab (6.10)

for some constant κ to be determined by the Newtonian limit.

However, since
1
∇a Rab = g ab ∇a R, (6.11)
2
the assumption ∇a T ab = 0 is not fulfilled. To remedy this, let us recall the Einstein tensor
Gab = Rab − 12 gab R, and
∇a Gab = 0. (6.12)
Therefore, we can guess
Gab = κTab , (6.13)
which is equivalent to
1
Rab = κ(Tab − gab T ), T = g ab Tab . (6.14)
2

43
To fix κ, we recall from special relativity that
3
X
T =ρ+ pi (6.15)
i=1

where the pi are the ’principal pressures’. Then

3
1 1 X
Φaa a b a b
= Rab V V = κ(Tab V V − T ) = κ(ρ − pi ). (6.16)
2 2 i=1

The Newtonian limit means the sum of the principal pressures should be much less than ρ
(which means that it is dominated by dust, as the pressure is p/c2 if c ̸= 1), so by comparing
the field equations (6.3) we obtain
κ = 8πG. (6.17)
Thus, we obtain Einstein’s field equations
1
Rab − gab R = 8πGTab , (6.18)
2
or equivalently, after taking the trace,
1
Rab = 8πG(Tab − gab T ). (6.19)
2
We can see from (6.19) that in the absence of matter, the vacuum field equations are
simply
Rab = 0. (6.20)
Mathematically speaking, if the Ricci tensor of an n-dimensional manifold is proportional
to its metric tensor in the following way,
2Λ
Rab = gab , (6.21)
n−2
we say this manifold is an Einstein manifold. In 4D, we see that Rab = Λgab , and the extended
vacuum Einstein’ field equations consistent with this condition are
1
Rab − gab R + Λgab = 0, (6.22)
2
as can be easily shown by contracting this with g ab . We call Λ the cosmological constant.
After adding matter contributions, we have the full Einstein’s field equations with a cosmo-
logical constant
1
Rab − gab R + Λgab = 8πGTab . (6.23)
2
The cosmological-constant term in (6.22) can be alternatively interpreted as a ‘matter’
contribution, with energy-momentum tensor Tab = −Λgab /(8πG). If we take a static ob-
server in spacetime with 4-velocity ua = ((g00 )−1/2 , 0, 0, 0), we can read out
Λ Λ
ρΛ = , pΛ = − , (6.24)
8πG 8πG
as the energy density and pressure of a perfect fluid.

44
6.2 Einstein-Hilbert action
To motivate an action for spacetimes, we require (i) the fundamental variables are the metric
g and the connection Γ and there should be no other background fields or coordinates; (ii)
the geometry should be Riemannian; (iii) the action should involve no terms containing
more than two derivatives of the metric and should be linear in second derivatives. The
requirement (iii) is more general than the usual actions for particle mechanics which only
involve no more than first derivatives. To follow this tradition of particle mechanics, let
R √
I = L −gdx4 be the action with a scalar L that only involves first derivatives of gab .
In Riemannian geometry, we have constructed a scalar R, the Ricci scalar, but R contains
two derivatives of gab . To proceed, we can use the Gauss theorem to reduce the second
derivatives of gab in R to first derivatives, that is,
√
√ √ ∂( −gω c ) 4
Z Z Z
R −gd x =4
L −gd x + 4
dx (6.25)
M M M ∂xc

for some ωa containing the second derivatives of gab . From the definition of R in terms of
Γ,
√ √ √  
−gR = −gg ab Rab = −g g ab ∂c Γcab − g ab ∂b Γcac + g ab Γcab Γdcd − g ab Γdac Γcbd ,
√ √
we see that ω c = −gg ab Γcab − −gg ac Γdad , and by using (4.36) and the covariant derivatives
(in the underlined terms),
√ √ √ √  
−gL = − ∂c ( −gg ab )Γcab + ∂b ( −gg ab )Γcac + −g g ab Γcab Γdcd − g ab Γdac Γcbd =
√ ab c
√ ab c
√ bd a c
√  ab c d ab d c

= − ∂c −gg Γab − −g∂c g Γab − −gg Γbd Γac + −g g Γab Γcd − g Γac Γbd =
√ √ √
= − −gg ab Γdcd Γcab − −g(−Γacd g db − Γbcd g ad )Γcab − −gg bd Γabd Γcac +
√  
+ −g g ab Γcab Γdcd − g ab Γdac Γcbd =
√  
= −g g ab Γdac Γcbd − g ab Γcab Γdcd . (6.26)

Now, the variation wrt L amounts to the variation wrt R,

√ √
Z Z
δ 4
R −gd x = δ L −gd4 x. (6.27)
M M

Therefore, we have the general action that meets the above three criteria, known as the
Einstein-Hilbert action,

√
Z
1
I= (R − 2Λ) −gd4 x + Imatter (6.28)
2κ M

where Λ is the cosmological constant. Let us now derive Einstein’s equations using action
principle.
Firstly, we assume that the variation of the matter action can be written as

√
Z
1
δImatter = − Tab −gδg ab d4 x (6.29)
2 M

45
where Tab is the stress-energy tensor of the matter fields. That is, the stress-energy tensor
should be
2 δImatter
Tab = − √ . (6.30)
−g δg ab
For the gravitational term in the action, we have
√ √ √ i
Z Z h
δ (R − 2Λ) −gd x =4
(R − 2Λ)δ −g + ((δg ab )Rab + g ab (δRab )) −g d4 x. (6.31)
M M
√
The δ −g can be obtained from (4.36) and by the fact that δ(gab g ab ) = 0,
√ 1√ 1√
δ −g = −gg ab δgab = − −ggab δg ab . (6.32)
2 2
For the variation of Rab = ∂c Γcab − ∂a Γcbc + Γcab Γdcd − Γcad Γdbc is
δRab = ∂c δΓcab − ∂a δΓcbc + δΓcab Γdcd + Γcab δΓdcd − δΓcad Γdbc − Γcad δΓdbc = ∇c δΓcab − ∇a δΓcbc . (6.33)
Using (4.38), we have
√ √ √
−gg ab δRab = −g∇c v c = ∂c ( −gv c ), v c = g ab δΓcab − g cd δΓede . (6.34)
We therefore see that δRab only contributes to a total derivative. Such boundary terms
are important for defining global quantities such as mass, but irrelevant for the bulk field
equations. In summary, we obtain
1
Z h
1 i√
δIgravity = − (R − 2Λ)gab + Rab −gδg ab d4 x. (6.35)
2κ M 2
Thus, the total variation of the Einstein-Hilbert action entails Einstein’s field equations
1
Rab − gab (R − 2Λ) = κTab . (6.36)
2
Let us take a closer look at the boundary term from (6.34),
√ √
Z Z
1 1
B= 4
d x −g∇c v =c
d3 x −hnc v c (6.37)
2κ M 2κ ∂M
where nc is normal to the boundary ∂M . Since in vc there are variations δgab and also
δ(∂c gab ), it is far from obvious that this boundary term could be ignored if we only fix the
metric at the boundary. But fixing both variations δgab and δ(∂c gab ) at the boundary would
be too strong. The right strategy is to introduce a counter term to cancel the boundary
term. An obvious choice is simply
√
Z
1
SB = d3 x −hnc (−v c ). (6.38)
2κ ∂M
A nice point is that R + ∇c (−v c ) recovers exactly the Lagrangian (6.26). Notice that the
choice of boundary term is not unique, as the normal components along nc are irrelevant.
Indeed, −nc v c = 2K − 2hcd ∂d nc + na hcd ∂c gda , and consequently the following Gibbons-
Hawking-York term
√
Z
1
SGHY = d3 x −hK. (6.39)
2κ ∂M
also cancels the boundary term, where K = ∇c nc is the trace of the extrinsic curvature of
∂M (cf. (6.70)).

46
6.3 Energy conditions
The Bianchi identity ∇a Gab = 0 then forces the conservation of the matter stress-energy
tensor, ∇a Tmatter
ab
= 0, which in turn forces the Noether’s theorem in field theory. On the
other hand, by another (first) Bianchi identity (4.42), we find
1  1 
∇a Cabcd = ∇[c Rd]b + gb[c ∇d] R = 8πG ∇[c Td]c + ga[c ∇d] T . (6.40)
6 3
Thus, once the stress-energy tensor is given, we can solve for the Weyl tensor Cabcd from
the above differential equation.
Another important aspect of the stress-energy tensors is whether they are physically
reasonable. This is studied through the energy conditions:

• weak energy condition: Tab ξ a ξ b ⩾ 0, where ξ a is the 4-velocity of an observer (hence

non-space-like);

• null energy condition: Tab la lb ⩾ 0, where la is null;

• dominant energy condition: Tab ξ a ξ b ⩾ 0, and T ab ξb must be non-space-like;

• strong energy condition: Tab ξ a ξ b ⩾ 12 T .

To see the physical implications of these conditions, let us consider a perfect fluid with
energy momentum tensor Tab = (ρ + p)ua ub − gab p. By setting ξ a = aua + bla where la is
null, we have

Tab ξ a ξ b = Tab (aua + bla )(aub + blb ) = a2 Tab ua ub + abTab la nb + abTab ua lb + b2 Tab la lb .

Noticing that (ua ua = 1)

Tab ua ub = (ρ + p)ua ub − gab p ua ub = (ρ + p) − p = ρ

Tab la ub = (ρ + p)ua ub − gab p la ub = ρua la

Tab la lb = (ρ + p)ua ub − gab p la lb = (ρ + p)(ua la )2 (6.41)

we obtain

Tab ξ a ξ b = a2 ρ + 2abρua la + b2 (ρ + p)(ua la )2 = ρ(a2 + 2abua la ) + b2 (ρ + p)(ua la )2 .

Since we know that ξ a is time-like, i.e. (aua + bla )(aua + bla ) = a2 + 2abua la > 0, the weak
energy condition then gives
ρ ⩾ 0, ρ + p ⩾ 0. (6.42)
From (6.41), we see that the null energy condition gives (ρ + p)(ua la )2 ⩾ 0, that is

ρ + p ⩾ 0. (6.43)

The dominant energy condition further requires |T ab ξb | ⩾ 0, that is,

Tab T bc ξ a ξ c =((ρ + p)ua ub − gab p)((ρ + p)ub uc − δcb p)ξ a ξ c =

=((ρ + p)2 ua uc − 2p(ρ + p)ua uc + p2 gac )ξ a ξ c = ((ρ2 − p2 )ua uc + p2 gac )ξ a ξ c ⩾ 0.

47
Thus,
ρ ⩾ |p|. (6.44)
The energy conditions are useful for several purposes. As a first application, we can prove
the energy-momentum conservation using the dominant energy condition.

Theorem 6.1. Let U be a compact spacetime region with a past non-timelike boundary ∂U1 , a
future non-timelike boundary ∂U2 and a timelike boundary ∂U3 . If the energy-momentum tensor
Tab satisfies the dominant energy condition and vanishes on ∂U1 and ∂U3 , then Tab = 0 on U.

Proof. Let t be a smooth function on U with ∇a t being timelike. Consider the following
integral
Z Z Z Z
ab ab ab
I(t) = ∇b (T ∇a t)dv = T ∇b ∇a tdv + ∇b T ∇a tdv = T ab ∇a tdσb
U U U ∂U

where dv denotes the volume form on U and dσa = na dσ on ∂U. Now use the condition
that T ab = 0 on ∂U1 and ∂U3 ; the boundary integral can be decomposed into
Z Z Z
= +
∂U U ∩∂U2 U ∩Ht

where Ht is the surface with fixed t. By assumption, ∇a t is future-pointing on ∂U2 , so the

second integral is positive, and one has
Z Z Z
ab
T ∇a tdσb ⩽ T ∇b ∇a tdv + ∇b T ab ∇a tdv.
ab
U ∩Ht U U

The dominant energy condition tells us that T ab ∇a t∇b t ⩾ 0 and T ab ∇a t is non-spacelike.

Because U is compact, ∇b ∇a t should have a upper bound along the timelike direction of
∇a t, otherwise ∇a t is possible to violate the compactness. In other words, there exists a
constant P > 0 such that T ab ∇b ∇a t ⩽ P T ab ∇a t∇b t. We can therefore consider
Z Z Z t nZ o
ab ab ′ ab ab
P T ∇a t∇b tdv + ∇b T ∇a tdv = dt (P T ∇b t + ∇b T )dσa ,
U U U ∩Ht

thereby
Z Z t nZ o
ab
T ∇a tdσb ⩽ dt ′ ab
(P T ∇b t + ∇b T )dσa . ab
(6.45)
U ∩Ht U ∩Ht

Note that Z Z t nZ o
ab ′ ab
x(t) ≡ T ∇a t∇b tdv = dt T ∇b t ⩾ 0
U U ∩Ht

and from (6.45) we have dx/dt ⩽ P x. For t earlier enough (i.e. outside of U), we have
U ∩ Ht = ∅, thus T ab = 0 and does not change.

Next, let us revisit the Raychaudhuri equation.

Theorem 6.2. If a geodesic congruence satisfies the following properties: i) the twist ωab = 0; ii)
the stress-energy tensor satisfies the strong energy condition, i.e. Tab ξ a ξ b ⩾ T /2; iii) the cosmological
dθ
constant Λ < 0, then dσ ⩽ 0.

48
Proof. Consider Einstein’s equations with cosmological constant Λ in the form (6.19)
1
Rab = 8πG(Tab − gab T ) − gab Λ.
2
Then,
1
Rab ξ a ξ b = 8πG(Tab ξ a ξ b − T ) − Λ.
2
Thus, using the assumed conditions, we see from the Raychaudhuri equation (5.29) that
dθ
dσ
⩽ 0.

This result dσdθ

⩽ 0 means that the geodesics become more and more contracted. Such
geometric reasoning is very effective in understanding the physics. For instance, consider
a small region in spacetime in which all the points are static, i.e. ωab = 0, σab = 0, ξ a =
(1, 0, 0, 0). Locally, we have gab ≈ ηab , T ≈ ρ+ 3i=1 pi , then using again Einstein’s equations
P

in the form (6.19), we can write the Raychaudhuri equation as

dθ X
= −R00 = −4π(ρ + pi ). (6.46)
dσ i

Thus if ρ, p > 0, then dσ

dθ
< 0, thereby attractive. Furthermore, if the initially θ0 < 0, then
the universe will keep contracting in this case, and result in a singularity. Under reasonable
conditions, one can find the conjugate point of the initial singularity (cf. Jacobi field), so
physically speaking, if the Big Bang singularity exists, then the Big Crunch singularity is
also possible.

6.4 Hamiltonian formalism

Einstein’s field equations can also be formulated in various different forms. An immediate
thought is to find the Hamiltonian formalism based on the Lagrangian used in the Einstein-
Hilbert action. To this end, consider the (3 + 1)-decomposition of the spacetime manifold,
M∼ = R × Σ, where Σ is a space-like hypersurface in a one-parameter foliation of M .
At this point, it is appropriate to introduce the hypersurface theory. We say that a
codimension-1 embedded submanifold Σ of M is a hypersurface. It is then necessary to
ask whether the geometric quantities, such the tensor fields and the covariant derivatives
defined on Σ is consistent with those defined on M . It turns out that the tensor algebra on
Σ is isomorphic to the algebra of tenors on M that are tangent to Σ. Consequently, we can
decompose tensors (or vectors) on M into two parts: those tangent to Σ and those normal
to Σ.
Let i : Σ ,→ M be the embedding map. Then the pullback i∗ (gab ) ≡ h̃ab of the metric
gab on M is called the induced metric, or first fundamental form, on Σ. h̃ab can be mapped
to the tensor hab defined on M , but withthe constraint

gab λa η b , if both λa and η a tangent to Σ


a b
hab λ η = (6.47)
0, if λa or η a normal to Σ

In most cases, it is safe to be casual about the above distinction between h̃ab and hab and
simply call hab the induced metric on Σ. Let us write the part of the metric normal to Σ as

49
kab = gab − hab . We can think of hab and k ab as the projections onto the components tangent
and normal to Σ respectively. For example, a tensor Aab defined on Σ can be obtained from
the (tangential) projection of the tensor Amn defined on M as

Aab = ham hnb Amn . (6.48)

Likewise, we can project the covariant derivative ∇m on M to ham ∇a ≡ ∇Σ

m on S. Then
we have the non-metricity condition

∇Σ m n p
a hbc = h a h b h c ∇m hnp = 0, (6.49)

and similarly by orthogonality, we have hma k nb k pc ∇m hnp = 0. However, the following

quantity is non-vanishing
Kabc = hma hnb k pc ∇m hnp (6.50)
which is the extrinsic curvature of the submanifold Σ. The following properties of Kabc can
be checked by direct calculations (Ex.).
Proposition 6.3. Let Kabc be the extrinsic curvature of a hypersurface Σ embedded in M , and let
ξ a be a vector tangent to a geodesic on Σ. Then
1. K[ab]c = 0;
c
2. ξ d ∇d ξ c = Kab ξ a ξ b .
The extrinsic curvature Kabc of Σ satisfies the Gauss-Codazzi equation relating it to the
intrinsic curvatures of M and Σ. To see this, consider first the curvature of Σ defined by
∇Σ ,
1 (Σ) p p
− Rabcd λb =∇Σ Σ a r m a r a m s n
[c ∇d] λ = h [c h d] ∇r (h p ∇m λ ) = h [c h d] h s ∇p (h r h n ∇m λ ) =
2
=hp[c hrd] has (∇p hmr )hsn ∇m λn + hmr (∇p hsn )∇m λn + hmr hsn ∇p ∇m λn .
 

(6.51)

By (6.49) and (6.50), we know that

hpc hrd ∇p hmr = hpc hrd g mq ∇p hqr = hpc hrd (k mq + hmq )∇p hqr = Kcd m (6.52)

Since K[ab]c = 0, we see that the first term in (6.51) vanishes. The remaining terms can be
further simplified as
1
− (Σ)
Rabcd λb =hp[c hmd] has (∇p hsn )∇m λn + hp[c hmd] han ∇p ∇m λn =
2
an
=K[c hmd] ∇m λn + hpc hmd han ∇[p ∇m] λn =
an 1
=K[c hmd] ∇m λn − hpc hmd han Rnbpm λn . (6.53)
2
By the definition of kabc (6.50), we know the third index is a normal component, so that
Kc an hnb = 0. Therefore,

Kc an hmd ∇m λn =Kc an hmd ∇m (hnb λb ) = Kc an hmd λb ∇m hnb = Kc an hmd hbr λr ∇m hnb =

=Kc an Kdrn λr = K a c n Kdrn λr .

50
Consequently, (6.53) becomes
 1 1 
− (Σ) Rarcd − K a [c n Kd]rn λr + hpc hmd han Rnrpm λn λr = 0. (6.54)
2 2
For any smooth vector η b , we can always project it to hrb η b ≡ λr , thereby
 
R bcd + 2K [c Kd]bn λ − h c h d h n h b R rpm λ η b = 0
(Σ) a a n r p m a r n n
(6.55)

which is the desired Gauss-Codazzi equation.

Let na be normal to Σ with the normalization property na na = 1, then the induced
metric hab can be expresses as
hab = gab − na nb (6.56)
which has been used a lot when discussing the Raychaudhuri equation. We also define the
second fundamental form

Kab = − Kabc nc = −hma hnb k pc nc ∇m hnp = −hma hnb np ∇m hnp

= − hma hnb [∇m (np hnp ) − hnp ∇n np ] = hma hnb ∇m nn (6.57)

which is also called the extrinsic curvature for hypersurfaces. In terms of Kab , the Gauss-
Codazzi equation (6.55) becomes
(Σ)
Rabcd = −2K a [c Kd]b + hpc hmd han hrb Rnrpm . (6.58)

Contacting the a, d indices of (6.58), we obtain

(Σ)
Rbc = − K a c Kab + K aa Kcb + hpc hmn hrb Rnrpm =
=KKbc − Kab K ac + hnb hpc Rnp − Rmbcr nm nr (6.59)

where K = K aa . Contracting once more (with hbc ) yields

(Σ)
R = K 2 − Kab K ab + hnp Rnp − Rmr nm nr = K 2 − Kab K ab + R − 2Rmr nm nr . (6.60)

It is interesting to observe at this point that if Σ has a vanishing extrinsic curvature, Kab = 0,
then the RHS of (6.60) takes the form of Einstein tensor, and we have
1  1 
− (Σ) R = Rab − gab R na nb . (6.61)
2 2
In other word, if Einstein’s equations hold, then in this case Σ has an extrinsic curvature
(Σ)
R = −16πTab na nb . Sometimes, such Σ’s are the geodesic generated hypersurfaces.
Now we can turn to the Hamiltonian formulation. Let us we write the coordinates as
x = (t, xi ), then the one-parameter space-like hypersurfaces can be represented by the
a
a
function X a = X a (t, xi ). The tangent vector Xia = ∂X
∂xi
and the normal vector na at a point
(t, xi ) on Σ should satisfy the following conditions:

gab Xia nb = 0, (6.62)

gab Xia Xjb = −hij , (6.63)
gab na nb = 1, (6.64)

51
which simply mean that Σ is space-like and na is time-like.
Now if (t, xi ) is changed to (t + dt, xi ), then Σ is deformed. Define the deformation
vector by N a = ∂t X a and consider the (3 + 1)-decomposition

N a = N na + N i Xia (6.65)

where N is called the lapse and N i is called the shift. Since N a is time-like, we have

gab dX a dX b = dX a dXb = N a Na dt2 = (N 2 − N i Ni )dt2 = gtt dt2

with Ni = hij N j , that is,

gtt = N 2 − N i Ni . (6.66)
Similarly, we have

git = gab Xia N b = gab Xia (N nb + N j Xjb ) = −N j hji = −Ni . (6.67)

Combining (6.62), (6.66) and (6.67), we obtain the ADM decomposition of the line element

ds2 = (N 2 − N i Ni )dt2 − 2Ni dxi dt − hij dxi dxj = N 2 dt2 − hij (dxi + N i dt)(dxj + N j dt).
(6.68)
Based on this decomposed metric, we find its determinant

−g = (N 2 − N i Ni )h + hhij Ni Nj = N 2 h. (6.69)

Recall the Gauss-Codazzi equation (6.60). If we take na = (N, 0, 0, 0), then we have
1
Kij = − (∂t hij −(3) ∇i Nj −(3) ∇j Ni ), (6.70)
2N
and the Gauss-Codazzi equation

R = −(3) R − Kij K ij + K 2 (6.71)

where K = hij Kij . By using (6.69), (6.70) and (6.71), we obtain the ADM action (up to a
surface term)
Z Z √
IADM = dt d3 x hN [(3) R + Kij K ij − K 2 − 2Λ]. (6.72)
R Σ

which is Hamiltonian counterpart of the Einstein-Hilbert action (6.28).

We can calculate the momentum conjugate to hij
δLADM √ h δKlm δK i √
π ij = = 2 hN K lm −K = h(K ij − Khij ). (6.73)
δ(∂t hij ) δ(∂t hij ) δ(∂t hij )

We therefore see that the Hamiltonian corresponding to the ADM action is (Ex.)
Z Z
HADM = 3 ij
d x(π ∂0 hij − LADM ) = d3 x(N H + Ni Hi ). (6.74)
Σ Σ

But obviously δLADM

δ(∂t N )
= 0 and δLADM
δ(∂t N i )
= 0, which give rise to the Hamiltonian constraint
√
H= h((3) R − Kij K ij + K 2 − 2Λ) = 0, (6.75)

52
and the momentum constraint

Hi =(3) ∇j (h−1/2 π ij ) = 0. (6.76)

So the Hamiltonian corresponding to the ADM action is a sum of constraints.

(Ex. Rewrite the geodesic deviation equation and the Raychaudhuri equation in the
ADM formalism. )

7 Cartan method and tetrad formalism

So far, we have worked with tensors, but in fact, the algebra, geometry, and even Einstein’s
equations can be reformulated in an index-free notation. The index-free approach is based
on the calculus of differential forms and it is known as the Cartan method. This method is
not just a mathematical reformulation, and it becomes necessary when we consider coupling
fermions to gravity.

7.1 Differential forms and Cartan method

A differential form is a totally skew covariant tensor, ωa...b = ω[a...b] . The type of a form is
the number of its indices, which should be smaller than n, the dimension of spacetime. A
form of type p is called a p-form; by convention, a scalar function is a 0-form.
If α is a p-form and β is a q-form, then their wedge product is

γa...d = α[a...b βc...d] ; γ =α∧β (7.1)

is a (p + q)-form. One can easily verified that

1. (α1 + α2 ) ∧ β = α1 ∧ β + α2 ∧ β;

2. (α ∧ β) ∧ γ = α ∧ (β ∧ γ);

3. α ∧ β = (−1)pq β ∧ α where α is a p-form and β is a p-form.

The algebra of forms defined by the wedge product is a Grassmann algebra. We define the
exterior derivative of the p-form α as

dα ≡ ∇[a αb...c] (7.2)

which is (p + 1)-form. The exterior derivative is part of the manifold structure and is given
prior to any choice of connection. In this sense, the forms are more primitive than connec-
tions. Using the exterior derivatives, we find the following properties:

1. d(dα) = d2 α = 0 for all α;

2. d(α ∧ β) = dα ∧ β + (−1)p α ∧ dβ where α is a p-form.

53
A form α for which dα = 0 is said to be closed, while if α = dβ for some β, we say that α
is exact.
The exterior derivatives dxa of the coordinate functions xa on a region of a manifold
form a basis for the 1-forms, and all possible wedge products of these give a coordinate basis
for all forms. Thus any p-form α can be written as

α = αa...b dxa ∧ ... ∧ dxb , (7.3)

and the exterior derivative is

∂αa...b c
dα = dx ∧ dxa ∧ ... ∧ dxb . (7.4)
∂xc
Given an n-dimensional manifold M , if there exists a continuous and non-vanishing n-form
h(x)dxa1 ∧ ... ∧ dxan , then M is said to be oriented. The volume form is special orientation
n-form
√
ϵ = −gdx1 ∧ ... ∧ dxn , (7.5)
which is invariant under the change of coordinate basis. All the n-forms on an n-dimensional
manifold form an 1-dimensional representation of the symmetric group Sn ; therefore any
n-form can be obtained from the volume form ϵ by multiplying a scalar function.
Recall the dual field strength (3.35) which is obtained by the ‘∗-operation’. This ‘∗-
operation’ on differential forms is called the Hodge-∗ and maps a p-form to an (n − p)-form
1 b1 b2 ...bp
(∗α)a1 a2 ...an−p = ϵ a1 a2 an−p αb1 b2 ...bp , (7.6)
p!
b1 b2 ...bp √
where ϵ a1 a2 an−p are components of the volume form (with the factor −g!). More-
over,
∗ ∗ α = (−1)s+p(n−p) α (7.7)
where s is the number of (+1) in the metric.
If a manifold has a choice of metric then we may choose a non-coordinate basis of 1-
forms {θi , i = 0, 1, 2, 3}, so that the metric is

ds2 = µij θi θj (7.8)

where µij is a matrix of constants. Here the index i, j, ... labels which basis element is
involved, instead of the spacetime index a, b, .... If µij coincides with the Minkowski metric,
θi is called a tetrad or vierbein. There is a dual basis of vector fields {ei , i = 0, 1, 2, 3} such
that θbj ecj = δbc . The differentiation of θi along ej is

ecj ∇c θai = −Γijk θak (7.9)

where the Γijk are the Ricci rotation coefficients. If we contract (7.9) with g ab θbm , we find
(i
Γ kj µ
m)k
= 0. (7.10)

We define the (spin) connection 1-form by

ω ik = Γikj θj . (7.11)

54
By (7.10) we have
(i
ω k µm)k = 0. (7.12)

After contraction with θbj , (7.9) becomes

∇b θai = −ω ikb θak . (7.13)

And if we skew this on [ab], then we have (Cartan’s first structure equation)

dθi = −ω ik ∧ θk . (7.14)

Given the basis θi explicitly, we can solve (7.12) and (7.14) for ω ik . If there is non-vanishing
torsion, then we have the torsion 1-form

T i = dθi + ω ik ∧ θk . (7.15)

If we change the original choice of basis by a linear transformation, θi → Lij θj , then

from (7.14) we find
ω ij → (Lim ω mn − dLin )L̃nj (7.16)
where L̃nj is the inverse of Lin (so that Lin L̃nj = δji ). The inhomogeneous term indicates that
the quantity involved is not a tensor. To retain the tensor transformation, we introduce the
curvature 2-form (Cartan’s second structure equation)

1
Ωij = dω ij + ω ik ∧ ω kj = Rmnj i θm ∧ θn , (7.17)
2

so that
Ωij → Lim Ωmn L̃nj . (7.18)
Since d2 = 0, from d of (7.14), we find

Ωij ∧ θj = 0; (7.19)

from d of (7.17), we have

dΩij − Ωik ∧ ω kj + ω ik ∧ Ωkj = 0, (7.20)

which is the Bianchi identity. Likewise, from d of (7.15), we find

dT i + ω ij ∧ T j = Ωij ∧ θj . (7.21)

Example 7.1 (Spherically symmetric geometries). Consider the spherically symmetric met-
ric in the general form (cf. later sections):

ds2 = e2λ dt2 − e2µ dr2 − r2 (dθ2 + sin2 θdϕ2 ) (7.22)

where λ and µ are functions of r. We take the basis as

θ0 = eλ dt, θ1 = eµ dr, θ2 = rdθ, θ3 = r sin θdϕ, (7.23)

55
so that µij = diag(1, −1, −1, −1). From (7.12), we see the symmetries

ω 0i =ω i0 , i = 1, 2, 3 (7.24)
ω ij = − ω i , j
i, j = 1, 2, 3, i ̸= j (7.25)

Firstly we calculate the exterior derivatives of the basis

dθ0 =λ′ eλ dr ∧ dt = λ′ e−µ θ1 ∧ θ0 (7.26)

1
dθ =0 (7.27)
1
dθ2 =dr ∧ dθ = e−µ θ1 ∧ θ2 (7.28)
r
1 1
dθ3 = sin θdr ∧ dϕ + r cos θdθ ∧ dϕ = e−µ θ1 ∧ θ3 + cot θθ2 ∧ θ3 (7.29)
r r
where a prime stands for differentiation with respect to r. From (7.14) we can solve for

ω 01 =ω 10 = λ′ e−µ θ0 (7.30)
ω 02 =ω 03 = ω 20 = ω 30 = 0 (7.31)
1
ω 12 = − ω 21 = − e−µ θ2 (7.32)
r
1
ω 13 = − ω 31 = − e−µ θ3 (7.33)
r
1
ω 13 = − ω 32 = − cot θθ3 (7.34)
r
Then the curvature forms are obtained from (7.17)

Ω01 = − (λ′′ − (λ′ )2 − λ′ µ′ )e−2µ θ0 ∧ θ1 (7.35)

1
Ω02 = − λ′ e−2µ θ0 ∧ θ2 (7.36)
r
1
Ω03 = − λ′ e−2µ θ0 ∧ θ3 (7.37)
r
1
Ω12 = µ′ e−2µ θ1 ∧ θ2 (7.38)
r
1
Ω13 = µ′ e−2µ θ1 ∧ θ3 (7.39)
r
1
Ω23 = 2 (1 − e−2µ )θ2 ∧ θ3 (7.40)
r
from which we have

R0110 = − (λ′′ − (λ′ )2 − λ′ µ′ )e−2µ (7.41)

1
R0220 =R0330 = − λ′ e−2µ (7.42)
r
1
R1221 =R1331 = µ′ e−2µ (7.43)
r
1
R2332 = 2 (1 − e−2µ ) (7.44)
r

56
The Ricci tensor is given by

λ′
R00 =e−2µ (λ′′ − (λ′ )2 − λ′ µ′ + 2 ) (7.45)
r
λ′
R11 =e−2µ (λ′′ − (λ′ )2 − λ′ µ′ − 2 ) (7.46)
r
1 1 ′
R22 =R33 = − 2 (1 − e ) + (λ − µ′ )e−2µ .
−2µ
(7.47)
r r
Notice that these expressions are written wrt the basis θi , and we can transform them to the
spherical coordinate basis to get the familiar results.

7.2 Tetrad formalism

We consider the tetrad θi and the dual vector field ej . To rewrite Einstein’s equations in
terms of tetrads, we first consider the scalar curvature 4-form

∗R = ϵij ∧ Ωij (7.48)

where ϵij = ∗(θi ∧ θj ). To see this, we first have that

1
ϵij = ∗(θi ∧ θj ) = ϵklmn µki µlj θm ∧ θn , (7.49)
2
and hence ϵij = 21 ϵijkl θk ∧ θl which implies that

1
ϵij ∧ θk ∧ θl = ϵijmn θm ∧ θn ∧ θk ∧ θl = (δik δjl − δjk δil )ϵ. (7.50)
2
Then from (7.49) and (7.50) and Cartan’s second structure equation we get
1 1
ϵij ∧ Ωij = ϵij Rij kl ∧ θk ∧ θl = (δik δjl − δjk δil )Rij kl ϵ = Rϵ = ∗R. (7.51)
2 2
With ∗R, we can derive Einstein’s equations by varying the Einstein-Hilbert action. To
this end, we first look at the variation

δ(ϵij ∧ Ωij ) = δϵij ∧ Ωij + ϵij ∧ δΩij . (7.52)

From (7.49), we have that

1
δϵij = δ(ϵijkl θk ∧ θl ) = δθk ∧ ϵijk (7.53)
2
where ϵijk = ϵijkl θl . On the other hand,

δΩij = dδω ij + δω ik ∧ ω kj + ω ik ∧ δω kj . (7.54)

Therefore,

δ(ϵij ∧ Ωij ) =δθk ∧ (ϵijk ∧ Ωij ) + d(ϵij ∧ δω ij ) − dϵij ∧ δω αβ + ηij ∧ (δω ik ∧ ω kj + ω ik ∧ δω kj )

=δθk ∧ (ϵijk ∧ Ωij ) + d(ϵij ∧ δω ij ) − Dϵij ∧ δω αβ . (7.55)

57
For Levi-Civita connection with vanishing torsion, we have
1
Dϵij = D(ϵijkl θk ∧ θl ) = ϵijkl Dθk ∧ θl = ϵijkl T k ∧ θl = 0, (7.56)
2
and hence
δ(ϵij ∧ Ωij ) = δθk ∧ (ϵijk ∧ Ωij ) + d(ϵij ∧ δω ij ). (7.57)
Next, we rewrite the Einstein-Hilbert action as
1
L= ∗ R + Lmatter , (7.58)
16πG
and suppose that the variation of the matter Lagrangian Lmatter results in

δLmatter = δθi ∧ ∗Ti (7.59)

where Ti is the energy-momentum 3-form. Then using (7.57) and (7.59), we have
 1 
δL = δθi ∧ ϵijk ∧ Ωjk + ∗Ti + d(...), (7.60)
16πG
so the field equation is
1
− ϵijk ∧ Ωjk = 8πG ∗ Ti . (7.61)
2
We show that this equation (7.61) is equivalent to Einstein’ equation in metric form. Indeed,
by using Cartan’s second structure equation and the Hodge star operation, (7.61) becomes
1
− ϵmij ∧ θk ∧l Rij kl = 8πGTmn ϵn . (7.62)
4
Due to the following two identities (ex.)

θi ∧ ϵjk =δki ϵj − δji ϵk , (7.63)

i
θ ∧ ϵjkl =δli ϵjk + δki ϵlj + δji ϵkl , (7.64)

we see that
1 1
− ϵmij ∧ θk ∧l Rij kl = − Rij kl δjl (δik ϵm − δm
k
ϵi ) + δil (δm
k
ϵj − δjk ϵm ) + δm
l
(δjk ϵi − δik ϵi ) =


4 4
1 ij  1 i 
ij i
= − R kl ϵm + R mj ϵi = R m − δm R ϵi . (7.65)
2 2
Then the Einstein equations in metric form follows immediately.

7.3 Energy, momentum, and angular momentum

The Einstein equation (7.61) in the tetrad form naturally entails the definitions of energy,
momentum, and angular momentum for an isolated gravitational system.
Note first that the curvature 2-form is given by Cartan’s second structure equation (7.17).
Then the first term dωij contributes to (7.62) as

dωij ∧ ϵij m =d(ωij ∧ ϵij m ) − ωij ∧ dϵij m =

=d(ωij ∧ ϵij m ) + ωij ∧ (−ω ik ∧ ϵkj m − ω jk ∧ ϵik m + ω km ∧ ϵij k ), (7.66)

58
where in the second line w e have used the relation Dϵij m = 0. Consequently, Cartan’s
second structure equation becomes

1 1
− Ωij ∧ ϵij m = − dωij ∧ ϵij m − 8πG ∗ tm (7.67)
2 2
where
1
∗tm = ωij ∧ (ωkm ∧ ϵijk − ω jk ∧ ϵikm ). (7.68)
16πG
The Einstein equation (7.61) becomes

1
− d(ωij ∧ ϵij m ) = 8πG ∗ (Tm + tm ). (7.69)
2
This tm is not symmetric with respect to a natural basis and does not transform as a tensor,
and consequently it cannot be used as an energy-momentum tensor.
To make tm symmetric, we rewrite the (7.61) by inserting ϵijk = ϵijkl θl and Cartan’s
second structure equation as

1
− ϵijkl θl ∧ (dωjk − ωmj ∧ ω mk ) = 8πG ∗ T i . (7.70)
2
The first term on the LHS of (7.70) can be rewritten using the integration by part,

d(ωjk ∧ θl ) = θl ∧ ωjk − ωjk ∧ dθl = θl ∧ ωjk − ωjk ∧ (ωnl ∧ θn ), (7.71)

by which the (7.70) can be expressed as

1
− ϵijkl d(ωjk ∧ θl ) = 8πG ∗ (Ti + tiLL ), (7.72)
2
where
1 ijkl
∗tiLL = − ϵ (ωmj ∧ ω mk ∧ θl − ωjk ∧ ωml ∧ θm ) (7.73)
16πG
is the Landau-Lifshitz 3-form. Clearly, this ∗tiLL is symmetric in a coordinate basis θi = dxi .
Namely, we have
dxi ∧ ∗tjLL − dxj ∧ ∗tiLL = 0. (7.74)
This entails
d(∗M ij ) = 0 (7.75)
with ∗M ij = xi ∗ τ j − xj ∗ τ i and τ i = Ti + tiLL .
Now consider an isolated gravitational system with asymptotic flat geometry. Let Σ be
a spacelike hypersurface that is also asymptotically flat. Then the (4-)momentum of this
system is
√
Z
i
P = −g ∗ τ i , (7.76)
Σ

and the angular momentum is

√
Z
ij
J = −g ∗ M ij . (7.77)
Σ

59
If we use again the field equation in the form of (7.72), we get

√ √
Z Z
1
i
P = i
−g ∗ τ = − −gω jk ∧ ϵijk . (7.78)
Σ 16πG 2
S∞

The boundary ∂Σ is taken to be the surface at infinity, i.e. S∞ 2

. Similarly, we have the
asymptotic expression for angular momentum

√
Z
1
ij
−g (xi ϵj kl − xj ϵikl ) ∧ ω kl + ϵij . (7.79)


J =
16πG S∞ 2

To pass from (7.77) to (7.79), one needs to use Cartan’s second structure equation, the
torsionless condition, the definition of connection coefficients, and the transform between
tetrads and metrics (ex. or cf. [8]).
Based on the total energy P 0 defined above, the positive energy theorem for GR can be
proved with the help of spinors.

Theorem 7.2 (Positive mass theorem). Let Σ be an asymptotically flat spacelike hypersurface.
Assume that the dominant energy condition holds on Σ. Then the total energy P 0 = 0 iff the
spacetime is flat in a neighborhood of Σ.

7.4 A little bit Newman-Penrose formalism

Here, we introduce an alternative form of Einstein’s equations known as the Newman-
Penrose formalism.
At each point of the spacetime manifold M , we introduce the null tetrad la , na , ma , m̄a ,
where la , na are real null vectors and ma , m̄a are complex null vectors formed from two
real orthonormal space-like vectors aa , ba by ma = √12 (aa − iba ). The tetrad satisfies the
pseudo-orthogonality relations

la na = −ma m̄a = 1. (7.80)

Let us denote collectively Zma

= (la , na , ma , m̄a ), e.g. Z1a = la , then the metric tensor can
be expressed by the null tetrad as

g ab = Zm
a b mn
Zn η = 2[l(a nb) − m(a m̄b) ] (7.81)

where, unlike the Minkowski metric,

 
0 1 0 0
1 0 0 0
 
η mn = .
0 0 0 −1
0 0 1 0

Einstein equations rewritten in this formalism are quite complicated (cf. Appendix A.1),
but extremely useful, for example, in describing isolated systems and analyze the gravita-
tional radiation emitted by compact sources.

60
8 Weak gravitational fields
The Einstein’s field equations in four dimensions are ten coupled nonlinear partial differ-
ential equations. If the summations were written out in full, each equation would have
hundreds of terms. Given this complexity, it is unlikely that we will find the general solu-
tion any time soon. We can consider the weak gravitational fields that are slightly different
from the flat Minkowski spacetime.

8.1 Linearized gravity

Let us assume that locally we can have the perturbation form of the metric

gab = ηab + ϵhab , g ab = η ab − ϵhab , (8.1)

where ηab is the Minkowski metric and hab = η ac η bd hcd . Then, we have
1 1
Γabc = g ad (∂c gbd + ∂b gcd − ∂d gbc ) = ϵη ad (∂c hbd + ∂b hcd − ∂d hbc ) + O(ϵ2 ). (8.2)
2 2
Letting ϵ = 1 for the moment, the curvature tensor is
1
Rabcd = (∂a ∂d hbc + ∂b ∂c had − ∂b ∂d hac − ∂a ∂c hbd ) (8.3)
2
and the Ricci tensor is
1
Rab = (∂a ∂ c hbc + ∂b ∂ c hac − □hab − ∂a ∂b h) (8.4)
2
where □ = η cd ∂c ∂d . To write down the Einstein tensor, we define the trace-reversed metric
perturbation h̄ab = hab − 12 ηab h, so that R = ∂a ∂b h̄ab + 12 h̄ and
1
Gab = (−□h̄ab + ∂a ∂ c h̄bc + ∂b ∂ c h̄ac − ηab ∂ c ∂ d h̄cd ). (8.5)
2
Notice that the splitting (8.1) is coordinate-dependent; we can therefore choose a coor-
dinate condition, or gauge condition, to further simplify the results. Under the coordinate
change xa → xa + ξ a , we know from the Killing equation that

δgab = ∇a ξb + ∇b ξa = ∂a ξb + ∂b ξa + O(h2 ) ⇒ hab → hab + ∂a ξb + ∂b ξa

so that
∂ c h̄ac → ∂ c h̄ac + □ξa .
Let us choose the coordinate condition −∂ c h̄ac = □ξa , or equivalently (de Donder gauge,
or harmonic coordinate condition)
∂ c h̄ac = 0 (8.6)
then
1
Gab = − □h̄ab + O(h2 ). (8.7)
2
Finally, Einstein’s field equations reduce to the wave equations

□h̄ab = −2κTab + O(h2 ). (8.8)

61
8.2 The slow-motion approximation
The slow-motion approximation means v/c ≪ 1 and we also assume that the time deriva-
tives of all quantities are much less than the space derivatives (in other words, the rate of
change per year is much less than the rate of change per light year).
Our aim is to relate the geodesic equation to the Newtonian force law (6.2), and relate
the Einstein equations (6.18) to the Newton-Poisson equation (6.3). A direct approach is to
consider the limit □ → −∇2 as c becomes very large, and choose suitable Tab . Here, let us
take a different way without using the coordinate conditions.
For the geodesic equation we first consider a slowly moving particle with v = O(ϵ)
(retaining ϵ and c), then

dxi dt dxi dx0 dt

= = O(ϵ), i = 1, 2, 3, = = 1 + O(ϵ).
ds ds cdt ds ds
Thus, the geodesic equation becomes

d2 xi 0
i dx dx
0

2
+ Γ00 = O(ϵ2 ) (8.9)
dt ds ds
where (by the assumption ∂0 = O(ϵ)∂i )
1 1
Γi00 = − ϵ(∂0 h0i + ∂0 h0i − ∂i h00 ) + O(ϵ2 ) = ϵ∂i h00 + O(ϵ2 ).
2 2
If we identify
g00 = 1 + ϵh00 = 1 + 2ϕ + O(ϵ2 ) (8.10)
we see that the geodesic equation agrees with (6.2).
For the Ricci tensor we have

Rab = ∂b Γcac − ∂c Γcab + O(ϵ2 )

so that
1
R00 = −∂i Γi00 + O(ϵ2 ) = − ϵ∂ii h00 + O(ϵ2 ) = −∇2 Φ + O(ϵ2 ). (8.11)
2
Then the 00-component of Einstein’s field equations reduce to the Newton-Poisson equa-
tion (6.3).

8.3 Gravitational waves

By solving the wave equations (8.8) of the weak gravitational fields, we obtain the gravita-
tional waves.
For the propagation of gravitational waves in empty space, let us choose the radiation
gauge (or transverse traceless (TT) gauge)

ha0 = 0, h = 0. (8.12)

The radiation gauge can be chosen just like the radiation (i.e. Coulomb) gauge in electro-
magnetism. This gauge freedom is due to the residual dof in hab + ∂a ξb + ∂b ξa if □ξ a = 0.

62
Note that h = 0 gives hab = h̄ab . Since h0i = 0, we obtain from the de Donder gauge
condition (8.6) that
∂0 h00 = 0, (8.13)
which means h00 is a constant and can be further gauge transformed to h00 = 0. In other
words, h00 does not contain information about gravitational waves. As a consequence, a
plane wave in radiation gauge will take the form
a
hij = ϵij eika x (8.14)

with ka k a = 0, k i ϵij = 0 (transverse condition). We see that the wave vector k a is null,
implying that gravitational waves move at the speed of light.
Let us count the propagating degrees of freedom (dof ) of gravitational waves in 4D. The
tensor ϵij is symmetric, so we have 10 dof to start with. The transverse condition k i ϵij = 0
give rise to 4 constraints; the traceless condition h = 0 gives 1 constraint; the condition
ha0 = 0 gives 4 constraints, one of which coincides with k i ϵi0 = 0. Consequently, two dof
or polarizations remain: for propagation in the z direction, for instance, the amplitudes are

ϵ+ ϵ× 0!
ϵij = ϵ× ϵ+ 0 . (8.15)
0 0 0

The + and × polarizations differ by a π/2 rotation. The general form of the perturbed
metric with gravitational waves propagating along the z-axis is

ds2 = dt2 − dz 2 − (1 − ϵ+ h+ )dx2 − (1 + ϵ+ h+ )dy 2 − 2ϵ× h× dxdy. (8.16)

Now suppose a weak gravitational plane wave hij = ϵij sin(ka xa ) with a + polarization
moving along the z-axis. As hab is a perturbation to ηab , we can write the line element as

ds2 = dt2 − dz 2 − (1 − ϵ+ sin ω(t − z))dx2 − (1 + ϵ+ sin ω(t − z))dy 2 (8.17)

The proper distance between two test particles at a fixed time is then dℓ with
R

dℓ2 = dz 2 + (1 − ϵ+ sin ω(t − z))dx2 + (1 + ϵ+ sin ω(t − z))dy 2 .

If ∆x = ℓ0 cos θ, ∆y = ℓ0 sin θ (on a circle), then

h 1 i
ℓ = ℓ0 1 − ϵ+ cos 2θ sin ω(t − z0 ) + O(ϵ2+ ). (8.18)
2
Such a oscillating behavior is the basis for the observational detection of gravitational waves!
It is also possible to obtain these behaviors from the geodesic deviation equation. To this
end, consider two freely falling particles following nearby geodesics. Let us work in the
local inertial frame of these two nearby particles, then in this frame the geodesic deviation
equation (5.15) simplifies to

d2 η a (σ)
= Rbcda (x(σ))ξ b η c ξ d . (8.19)
dσ 2

63
without further connection terms. We identify η a as the proper length of the components
of the deviation vector; this way, the above equation becomes gauge-invariant. So we can
work in the TT gauge and obtain the Jacobi equation (5.19). For a wave traveling in the
z-direction, the Jacobi equations in the TT gauge are

∂ 2ηx ϵ ∂ 2 TT ∂ 2ηy ϵ ∂ 2 TT
= h , = h (8.20)
∂t2 2 ∂t2 xx ∂t2 2 ∂t2 yy
where have chosen the initial η a = (0, ϵ, 0, 0) in the x-direction. The similar equations can
be obtained for more general η a . These equations are of fundamental importance; they tell
us that in the local inertial frame of one particle, say A, one can see the acceleration of the
other nearby particle B, caused by a force known as the tidal force! The Jacobi equation
therefore tells us the acceleration caused by the tidal force. In practice, in (8.19) we need
to add other forces, e.g. electromagnetic forces, to get a complete equation of motion. But
since other interactions are much stronger than gravity, to detect the tidal displacements
caused by the incident gravitational waves is not easy.

8.4 Gravitational radiation

We turn to the gravitational radiation generated by a source. In analogy to the electrody-
namics, we find the quadrupolar solution (of Liénard-Wiechert type) to the wave equation

Tab (t − |⃗x − ⃗y |, ⃗y ) 3
Z
h̄ab = 4G d ⃗y (8.21)
|⃗x − ⃗y |

This means that waves travel at the speed of light: the perturbation h̄ab at time t at a distance
L from the source is determined by the source at the earlier time (t − L/c).
Suppose the source is confined to a small region U of size r, which we observe at a much
larger distance |⃗x − ⃗y | ≈ R ≫ r. The spatial components of the perturbation are then
Z
4G
h̄ij ≈ Tij (t − R, ⃗y )d3 ⃗y (8.22)
R U

(The other components can be solved from the coordinate condition.) Form the conserva-
tion equation, we find

∂a T ab = 0 ⇒ ∂0 T 00 = −∂k T k0 , ∂0 T 0k = −∂k T ki .

Let us define Qij = xi xj T 00 , then by taking the derivative wrt coordinates, we have

∂0 Qij =xi xj ∂0 T 00 = −xi xj ∂k T k0 = −∂k (xi xj T k0 ) + xj T i0 + xi T j0

∂02 Qij = − ∂k (xi xj ∂0 T k0 ) + xj ∂k T ik − xi T j0 − xi ∂k T jk
= − ∂k (xi xj ∂0 T k0 + xj T ik + xi T jk ) + 2T ij .

Thus, we find, after discarding the total divergence,

2G d2
Z Z
2G
h̄ij ≈ ∂02 Qij (t 3
− R, ⃗y )d ⃗y = yi yj T 00 (t − R, ⃗y )d3 ⃗y (8.23)
R U R dt2 U

64
which is the quadrupole moment of the source. The reason for the quadrupole moments can
also been seen by using the multi-pole moment expansion in analogy to the electromagnetic
case, cf. [11].
The above presentation is a little tricky. To gain more insights, let us consider the Fourier
transform of (8.21). Take the time coordinate t as that in the local inertial frame define by
ηab , then the (spacetime) Fourier transformed result is
Z
ˆ = 4G T̂ab (ω, ⃗y ) iω|⃗x−⃗y| 3
h̄ab e d ⃗y . (8.24)
|⃗x − ⃗y |
The harmonic gauge condition is transformed to
3 ˆ
∂ h̄
ˆ =
X aj
−iω h̄0a . (8.25)
j=1
∂xj

With this condition, we only need to know the 0a components. At a large distance R ≫
ω −1 ,
eiω|⃗x−⃗y| eiωR
≈
|⃗x − ⃗y | R
and hence can be pulled out of the integration. The remaining integration is
Z s hZ Z Z
3
X ∂ T̂ib i
T̂ab d ⃗y = (T̂ib ya )d3 y − ya d y = −iω T̂0b ya d3 y =
3

i=1
∂yi ∂y i
Z 3 Z Z
iω 3 iω Xh ∂ 3 ∂ T̂0b i
=− (T̂0b ya + T̂0a xb )d y = − (T̂0j ya yb )d y − ya yb d3 y =
2 2 j=1 ∂yj ∂tj
ω2
Z
=− T̂00 ya yb d3 y. (8.26)
2
In total, we have
2 iωR
ˆ = − 2Gω e q̂
h̄ (8.27)
ij ij
R
where q̂ij is the Fourier transform of quadrupole tensor
Z
qij = T00 xi xj d3 x. (8.28)

Performing the reversed Fourier transform, then (8.46) gives

2G d2 qij
h̄ij = (8.29)
R dt2
in consistency with (8.23).
The gravitational radiation carries energy. In GR, the energies are hard to be localized,
as the geometry and dynamics are intimately correlated. However, the total energy for an
isolated system is well-defined. In linear gravity, we have obtained the vacuum Einstein
equation to the first order in hab , (cf. (8.5)),
(1)
Gab [hcd ] = 0. (8.30)

65
When the computation is done to the second order in hab , we have the second-order terms
in the Ricci tensor
(2) 1 1
Rab = hcd ∂a ∂b hcd − hcd ∂c ∂(a hb)d + (∂a hcd )∂b hcd + (∂ d hcb )∂[d hc]a +
2 4
1 1 1
+ ∂d (hdc ∂c hab ) − (∂ c h)∂c hab − (∂d hcd − ∂ c h)∂(a hb)c . (8.31)
2 4 2
(2)
We can write hab + hab . To ensure that the vacuum Einstein equations hold up to the
(2)
second-order terms in hab , we require hab to satisfy

(1) (2) (2) (2) (2) 1

Gab [hcd ] + Gab [hcd ] = 0, Gab = Rab − ηab R(2) . (8.32)
2
This condition can be recast into
(1) (2) 1 (2)
Gab [hcd ] = 8πtab , tab = − G [hcd ]. (8.33)
8π ab
In other words the O(h2 ) terms contribute an effective energy-momentum tensor tab . No-
tice that we should not interpret this tab as a meaningful local gravitational energy-momentum
tensor, as it is not gauge-invariant (diffeomorphism symmetric). But the total energy on a
space-like hypersurface Σ, Z
E= t00 d3 x (8.34)
Σ

is gauge-invariant under the asymptotically flat conditions,

hab ∼ O(r−1 ), ∂c hab ∼ O(r−2 ), ∂c ∂d hab ∼ O(r−3 ). (8.35)

Similarly, the total radiation energy

Z
∆E = − ta0 dS a (8.36)
S

is also gauge-invariant. Here, S is a time-like hypersurface within two future light-cones

(i.e. two null hypersurfaces). Using (8.29), we obtain
3
1 X  d3 Qij 2
Z
∆E = P dt, P = (8.37)
45 i,j=1 dt3

where Qij = qij − 13 δij q is the trace-free quadrupole tensor.

The above quantities are not directly detected in a gravitational detector. The more
observation-oriented quantities, such as the energy flux and gravitational wave luminosity
can be found in [6].

8.5 Gravitoelectromagnetism
Let us take T 00 = ρ and T 0i = j i , where ρ is the mass density and j i the mass current of the
source, then the conservation equation ∇a T ab = 0 take the familiar form: ∂ρ/∂t+ ∇i j i = 0.

66
We have seen that, in the slow-motion approximation, h̄00 (of O(c2 )) and h̄0i (of O(c3 )) are
much larger than h̄ij (of O(c4 )). Therefore, from (8.21), we find the dominant components

ρ(t − |⃗x − ⃗y |, ⃗y ) 3 j i (t − |⃗x − ⃗y |, ⃗y ) 3

Z Z
h̄00 = 4G d ⃗y , h̄0i = −4G d ⃗y . (8.38)
|⃗x − ⃗y | |⃗x − ⃗y |

Now let us introduce 5 the gravitoelectric potential Φ and the gravitomagnetic potential
Ai such that
h̄00 = 4Φ, h̄0i = −2Ai (8.39)
then
ρ(t − |⃗x − ⃗y |, ⃗y ) 3 j i (t − |⃗x − ⃗y |, ⃗y ) 3
Z Z
Φ=G d ⃗y , Ai = 2G d ⃗y . (8.40)
|⃗x − ⃗y | |⃗x − ⃗y |
In terms of the gravitoelectromagnetic potentials Φ and Ai , the de Donder gauge condition
(8.6) for h0a becomes
1
∂0 h̄00 + ∂i h̄0i = 0 ⇒ ∂t Φ + ∂i Ai = 0 (8.41)
2
which is the same as the Lorentz gauge condition for electromagnetic fields. Pushing the
analogy with electromagnetism further, we define the gravitoelectric Bi in the same way as
(2.26) and gravitomagnetic Ei = − 12 Ȧi −∇i Φ. Then we can rewrite the weak field equation
(8.8) in the gravitoelectromagnetic form

1
∇i Bi =0 (8.42)
2
∇i Ei =4πGρ (8.43)
1
ϵijk ∇j Ek = − Ḃi (8.44)
2
1
ϵijk ∇j Bk =Ėi + 4πGJi (8.45)
2
Now the perturbed metric takes the general form

ds2 = (1 − 2Φ)dt2 + 4Ai dxi dt − (1 + 2Ψ)δij dxi dxj . (8.46)

where Ψ = h̄ij /2. This form of metric can be obtained in the full GR, if we use the Fermi
coordinate (T, X, Y, Z) for geodesic observers (cf. (5.41)):

1 1
Φ(T, X i ) = − R0i0j (T )X i X j , Ai (T, X i ) = R0ijk (T )X j X k ,
2 3
2
Ψij (T, X ) = − Rijkl (T )X k X l
i
(8.47)
3
where Rabcd (T ) is the curvature tensor evaluated along the geodesic (where X i = 0). Then,
keeping terms to the first order in X i , we find

1
Ei (T, X i ) = R0i0j (T )X j , Bi (T, X i ) = − ϵijk Rjk0l (T )X l . (8.48)
2
5
This part is borrowed from [20].

67
For the metric (8.46) with (8.47), the spatial components of the geodesic equation are, to
the linear order in ⃗v /c, (Ex.)

dvi
= ∂i Φ − 2(ϵijk vj Bk ) + 2∂t Ai − v i ∂t (2Ψ + Φ). (8.49)
dt
We see that the first and the third term combine to form −Ei plus a term proportional
to ∂t Ai which is second order in X i and hence can be discarded. Thus, up to O(X i ), the
geodesic equation takes the form of the Lorentz force law:

mv̇i = −m(Ei + 2ϵijk vj Bk ). (8.50)

From this perspective, the so-called Lense-Thirring precession is the gravitoelectromag-

netic counterpart of the Thomas precession of particles with spins.

8.6 Gravitational wave memory

The electromagnetic analogs of gravitational waves exist not only in the transverse traceless
gauge of weak fields, but also in the full GR, as we have seen in the gravitomagnetism in the
Fermi coordinates. In fact, we can understand the gravitational wave effect using again the
electromagnetic analog of gravitational wave [21]. The gravitational wave memory refers
to the residual effect on the gravitational wave detector left after the gravitational wave has
passed.
The electromagnetic analog of gravitational wave memory is the kick received by a test
electric charge. A test particle of charge q and mass m follows the equations of motion
m⃗x¨ = q E,
⃗ then a kick is a change in the velocity

q ∞ ⃗
Z
∆⃗v = Edt. (8.51)
m −∞

Recall the wave equation (3.22) for the electromagnetic field strength, and consider the
corresponding equation for the spatial i-components Ei = Fi0 :

∇a ∇a Ei = −4π(∂i J0 − ∂0 Ji ). (8.52)

From the above equation we can read the current

∇a Ei = Qa = δ ai j0 − δ a0 ji . (8.53)

Now let us integrate the current Qa over the boundary ∂M of a spacetime region M with
normal vector na . If we choose the na to be n0 = 1, ni = −r̂i , then Qa na = −(r̂i j0 + ji ). In
fact, we can define the projector
Pij = δij − r̂i r̂j (8.54)
onto the space orthogonal to r̂. Then

Qa na = r̂i (−j0 − r̂ · ⃗j) − Pi k jk = r̂i (j a na ) − Pi k jk (8.55)

68
We see that the first term of (8.55) is the total charge which is conserved, so only the second
term contributes to the memory effect
Z ∞ Z Z
1h i
Ei dt = − k 3
Pi jk d y + k 3
Pi jk d y ̸= 0 (8.56)
−∞ r P∞ P−∞

where we have assumed that r ≫ |⃗y | in the retarded solution Ei as in (8.21) such that Pu is
the null plane at u = t − r.
To consider the gravitational wave memory, we recall the Jacobi equation (5.19). In the
asymptotic region, we have the memory tensor expressed in terms of the curvature tensor
as Z ∞ Z u
Mij (r, r̂) = duVij (u, r, r̂), Vij (u, r, r̂) = dũRi00j (ũ, r, r̂). (8.57)
−∞ −∞

To find a wave equation for Rabcd , let us first contract the Bianchi identity ∇[a Rbc]de = 0
with the Minkowski metric η ae to obtain

∇a Rbcda = ∇c Rbd − ∇b Rcd . (8.58)

Then
∇a ∇a Rbcde = ∇b (∇d Rec − ∇e Rdc ) + ∇c (∇e Rdb − ∇d Reb ), (8.59)
so the particular component

∇a ∇a Ri00j = ∂i (∂0 R0j − ∂j R00 ) + ∂0 (∂j R0i − ∂0 Rij ), (8.60)

from which we can read out

1 a
Qa = [δ (∂0 Rij − ∂j R0i ) + δ ai (∂j R00 − ∂0 R0j )]. (8.61)
4π 0
As in the electromagnetic case, we have
Z u2 Z
1 1
duVij (u, r, r̂) = d4 y [∂0 Rij − ∂j R0i + r̂i (∂0 R0j − ∂j R00 )]. (8.62)
u1 4 M 4π

The integrand of (8.62) is a total divergence ∇a Q̃a , if we set

1 a
Q̃a = [δ Rij − δ aj R0i + r̂i (δ a0 R0j − δ aj R00 )].
4π 0
Therefore,
1
Q̃a na = [Rij + R0i r̂j + R0j r̂i + R00 r̂i r̂j )] ≡ S1 + S2 (8.63)
4π
where

S1 =(T 0a na + Tb a na r̂b )(δij + r̂i r̂j ) − 2Ti a na r̂j − 2Tj a na r̂i

1
S2 =2(Pia Pjb − P mn Pij )Tab .
2
Obviously, the terms in S1 are the total energy and angular momentum (cf. later sections),
and hence S2 contributes to the memory tensor as
Z Z
1h i
Mij (r, r̂) = 3
d yS2 − 3
d yS2 ̸= 0. (8.64)
r P∞ P−∞

69
9 The Schwarzschild solution
Let us consider the solutions to Einstein’s field equations in the simplified situation of time-
independent and spherically symmetric metric. The time-independence means that ∂gab /∂t =
0 where we have chosen x0 = t and further that the metric is unchanged if we change the
sign of t. Then the (static) metric must necessarily take the form

ds2 = gab dxa dxb = g00 dt2 + gij dxi dxj . (9.1)

Spherical symmetry means that on a surface of constant t we can find a radial coordinate R
such that the surfaces of constant R are spheres. If we introduce polar coordinates θ and ϕ
on each sphere in the standard way, then the metric of one such sphere must take the form
−C 2 (R)(dθ2 + sin2 θdϕ2 ) for some function C(R). We can ‘line up’ every radius as a line of
constant θ and ϕ and use the same angular coordinates on every sphere, so that the entire
three-dimensional metric take the form

gij dxi dxj = −B 2 (R)dR2 − C 2 (R)(dθ2 + sin2 θdϕ2 ).

Then the general metric consistent with our assumptions takes the form

ds2 = A2 (R)dt2 − B 2 (R)dR2 − C 2 (R)(dθ2 + sin2 θdϕ2 ). (9.2)

Let us define r = C(R), then (9.2) can be re-expressed in the following form

ds2 = e2λ(r) dt2 − e2µ(r) dr2 − r2 (dθ2 + sin2 θdϕ2 ). (9.3)

Thus the general spherically-symmetric time-independent metric depends on just two func-
tions λ and µ of the radial coordinate. If λ = µ = 0 then we recognize (9.3) as the Minkowski
metric in spherical polar coordinates. Thus if (9.3) is to be the metric of an isolated system,
we must demand that it approaches the Minkowski metric at large distances, i.e. asymptot-
ically flat.
Let us do some computations according to the metric (9.3). Firstly, we know that

g00 = e2λ , g11 = −e2µ , g22 = −r2 , g33 = −r2 sin2 θ,

g 00 = e−2λ , g 11 = −e−2µ , g 22 = −r−2 , g 33 = −r−2 sin−2 θ. (9.4)

70
With the help of time-independence, the nonvanishing Christoffel symbols are

1 1
Γ001 = g 00 (∂1 g00 ) = e−2λ (∂1 e2λ ) = λ′
2 2
1 1
Γ100 = g 11 (−∂1 g00 ) = − e−2µ (−∂1 e2λ ) = λ′ e−2(µ−λ)
2 2
1 1
Γ111 = g 11 (∂1 g11 ) = − e−2µ (−∂1 e2µ ) = µ′
2 2
1 1
Γ122 = g 11 (−∂1 g22 ) = − e−2µ (∂1 r2 ) = −re−2µ
2 2
1 1
Γ133 = g 11 (−∂1 g33 ) = − e−2µ (∂1 r2 sin2 θ) = −re−2µ sin2 θ
2 2
1 1 1
Γ212 = g 22 (∂1 g22 ) = − r−2 (−∂1 r2 ) =
2 2 r
1 1
Γ233 = g 22 (−∂2 g33 ) = − r−2 (∂2 r2 sin2 θ) = − sin θ cos θ
2 2
1 1 1
Γ313 = g 33 (∂1 g33 ) = − r−2 sin−2 θ(−∂1 r2 sin2 θ) =
2 2 r
1 1
Γ323 = g 33 (∂2 g33 ) = − r−2 sin−2 θ(−∂2 r2 sin2 θ) = cot θ.
2 2
Next, the components of the Riemann tensor are

R0110 =∂0 Γ011 − ∂1 Γ001 − Γp01 Γ01p + Γp11 Γ00p = −∂1 Γ001 − Γ001 Γ010 + Γ111 Γ001 =
= − ∂1 (λ′ ) − (λ′ )2 + (µ′ )(λ′ ) = µ′ λ′ − λ′′ − λ′2
R0220 =∂0 Γ022 − ∂2 Γ002 − Γp02 Γ02p + Γp22 Γ00p = Γ122 Γ001 = −λ′ re−2µ
R0330 =∂0 Γ033 − ∂3 Γ003 − Γp03 Γ03p + Γp33 Γ00p = Γ133 Γ001 = −λ′ re−2µ sin2 θ
R1221 =∂1 Γ122 − ∂2 Γ112 − Γp12 Γ12p + Γp22 Γ11p = ∂1 Γ122 − Γ212 Γ122 + Γ122 Γ111 =
1
=∂1 (−re−2µ ) − (−re−2µ ) + (−re−2µ )µ′ = µ′ re−2µ
r
R133 =∂1 Γ33 − ∂3 Γ13 − Γp13 Γ13p + Γp33 Γ11p = ∂1 Γ133 − Γ313 Γ133 + Γ133 Γ111 =
1 1 1

1
=∂1 (−re−2µ sin2 θ) − (−re−2µ sin2 θ) + (−re−2µ sin2 θ)µ′ = µ′ re−2µ sin2 θ
r
R233 =∂2 Γ33 − ∂3 Γ23 − Γ23 Γ23p + Γp33 Γ22p = ∂2 Γ233 − Γ323 Γ233 + Γ133 Γ221 =
2 2 2 p

1
=∂2 (− sin θ cos θ) − (cot θ)(− sin θ cos θ) + (−re−2µ sin2 θ)( ) = (1 − e−2µ ) sin2 θ.
r
Then,

R0101 = − R0110 = −g00 R0110 = −e2λ (µ′ λ′ − λ′′ − λ′2 ) = R1010

R0202 = − R0220 = −g00 R0220 = λ′ re−2(µ−λ) = R2020
R0303 = − R0330 = −g00 R0330 = λ′ re−2(µ−λ) sin2 θ
R1212 = − R1221 = −g11 R1221 = µ′ r = R2121
R1313 = − R1331 = −g11 R1331 = µ′ r sin2 θ
R2323 = − R2332 = −g22 R2332 = r2 (1 − e−2µ ) sin2 θ.

71
So the Ricci tensor of (9.3) has components

R00 =g cd R0c0d = g 11 R0101 + g 22 R0202 + g 33 R0303 =

1 1
= − e−2µ [−e2λ (µ′ λ′ − λ′′ − λ′2 )] − 2 [λ′ re−2(µ−λ) ] − 2 2 [λ′ re−2(µ−λ) sin2 θ] =
r r sin θ
′
λ
=e2λ−2µ (−λ′′ − (λ′ )2 + λ′ µ′ − 2 ) (9.5)
r
R11 =g cd R1c1d = g 00 R1010 + g 22 R1212 + g 33 R1313 =
1 1
= − (µ′ λ′ − λ′′ − λ′2 ) − 2 (µ′ r) − 2 2 (µ′ r sin2 θ)
r r sin θ
′
µ
=λ′′ + (λ′ )2 − λ′ µ′ − 2 (9.6)
r
R22 =g 00 R2020 + g 11 R2121 + g 33 R2323 =
1
=e−2λ (λ′ re−2(µ−λ) ) − e−2µ (µ′ r) − 2 2 (r2 (1 − e−2µ ) sin2 θ) =
r sin θ
−2µ 2µ
=e (1 − e + r(λ − µ )) ′ ′
(9.7)
R33 =R22 sin2 θ. (9.8)

and zero otherwise. In fact, we can also directly compute Rab from Γabc using (4.48).
Consider then the vacuum Einstein’s field equations Rab = 0. From (9.5) and (9.6) we
obtain λ′ + µ′ = 0, and hence λ + µ = 0 for simplicity. (9.7) then gives

e−2µ (1 − e2µ − 2rµ′ ) = (re−2µ )′ − 1 = 0

so, for integration constant k,

k
e−2µ = 1 +
= e2λ
r
As g00 = e ≈ 1 + 2Φ + O(ϵ ), we must have k = 2Φ = −2GM/r. We finally obtain the
2λ 2

Schwarzschild solution
 2GM  2  2GM −1 2
ds2 = 1 − dt − 1 − dr − r2 (dθ2 + sin2 θdϕ2 ). (9.9)
r r

From (9.9) we readily see that the infinitesimal radial distance and the infinitesimal
proper time intervals are respectively
 2GM −1/2  2GM 1/2
dR = 1 − dr, dτ = 1 − dt. (9.10)
r r
The ’critical radius’ where the metric (9.9) becomes singular is called the Schwarzschild
radius
r = rs = 2GM. (9.11)
If we retain the cosmological constant as in (6.36), then by similar calculations we obtain
the Schwaschild-de Sitter solution
 rs Λ   rs Λ −1 2
ds2 = 1 − − r2 dt2 − 1 − − r2 dr − r2 (dθ2 + sin2 θdϕ2 ). (9.12)
r 3 r 3

72
Consider then the equation 1 − rrs − Λ3 r2 = 0. When Λ−1 > 9rs2 /4, the equation has three
real roots. Around r = rs , the Λ can be ignored, in which case the Schwaschild-de Sitter
solution reduces to the Schwaschild
q solution (in this particular choice of coordinates). On
the other hand, around r = 3
Λ
, the rs can be ignored, then the Schwaschild-de Sitter
solution reduces to the de Sitter metric
 r2  2  r2 −1 2
2
ds = 1 − 2 dt − 1 − 2 dr − r2 (dθ2 + sin2 θdϕ2 ) (9.13)
R R
q
where R = Λ3 .

9.1 Gravitational red-shift

Suppose the emitter of a light ray has world line E with unit 4-velocity ua and the observer
has world line O with 4-velocity v a . If the light ray is an affinely parametrized null geodesic
L with tangent vector ℓa , then the emitted frequency νE and observed frequency νO are
respectively
νE = gab ua ℓb , νO = gab v a ℓb . (9.14)
If the worldline E is at radius rE and the worldline O at radius rO , then
ka ka
ua = p , va = p , (9.15)
v(rE ) v(rO )

where k a is the time-like Killing vector of the Schwarzschild solution and

2GM
v(r) = gab k a k b ≈ 1 − .
r
Since L is affinely parametrized and k a is a Killing vector, gab k a ℓb is constant along L. Thus
p
νO gab k a ℓb v(rE )  2GM  21  2GM − 12
=p = 1 − 1 − . (9.16)
νE v(rO ) gab k a ℓb rE rO

This entails a time dilation effect between E and O due to the gravitational field: In a small
interval δSE of proper time at E, n wave crests will be emitted where n = νE δSE ; these n
wave crests will arrive at O in an interval δSO where n = vO δSO . Thus,
 2GM  12  2GM − 12
δSO = 1 − 1− δSE . (9.17)
rO rE
In words, the clock at the smaller value of r, i.e. at the smaller (more negative) gravitational
potential runs slower. Importantly, when r = rs = 2GM , the clock runs infinitely slow!
If rE and rO are both large compared to rs , then we have the following approximate
expression for the red-shift:
∆ν νO − νE GM GM
= ≈− + = ∆Φ. (9.18)
νE νE rO rE
Again, if rE = rs , the red-shift diverges, so the observer cannot see any light rays from the
emitter— a black hole!

73
9.2 Geodesics in the Schwarzschild spacetime
Now we consider the geodesics in the Schwarzschild spacetime. Let us consider the La-
grangian (with G = 1)
 2M  2  2M −1 2
L= 1− ṫ − 1 − ṙ − r2 (θ̇2 + sin2 θϕ̇2 ). (9.19)
r r
The corresponding Euler-Lagrange equations are
 ∂L • ∂L  2M
•
− = 2 1− ṫ = 0 (9.20)
∂ ṫ ∂t r
 ∂L • ∂L  2M
−1 • 2M 2M
−2 2
− = −2 1 − ṙ − 2 1 − ṙ +
∂ ṙ ∂r r r r
2M
+ 2r(θ̇2 + sin2 θϕ̇2 ) − 2 ṫ2 = 0 (9.21)
r
 ∂L • ∂L
− =(−2r2 θ̇)• + 2r2 sin θ cos θϕ̇2 = 0 (9.22)
∂ θ̇ ∂θ
 ∂L • ∂L
− =(−2r2 sin2 θϕ̇)• = 0 (9.23)
∂ ϕ̇ ∂ϕ

where ()• means d()/dσ with σ being the affine parameter for the geodesics.
L, which is zero or ±1 for null, time-like or space-like geodesics, is a first integral of
these equations. Two other first integrals
 2M 
E = 1− ṫ (9.24)
r
J =r2 sin2 θϕ̇ (9.25)

can be easily seen from (9.20) and (9.23). Taking ξ a = ẋa , since Q = gab ξ a k b is constant
along γ, we obtain from the first integrals the Killing vectors of the Schwarzschild solution

kEa =(1, 0, 0, 0) (9.26)

kJa =(0, 0, 0, 1). (9.27)

The three first integrals L, E, J are sufficient for solving these equations, because any
geodesic of the Schwarzschild solution must, by the spherical symmetry, remain in a plane
going through the center. Therefore, we are left with, for time-like geodesies, the once-
integrated geodesic equations

J =r2 ϕ̇ (9.28)
 2M 
E = 1− ṫ (9.29)
r
 2M  2  2M −1 2
L= 1− ṫ − 1 − ṙ − r2 ϕ̇2 = 1. (9.30)
r r
We can obtain from these equations that

2M J 2 2M J 2
ṙ2 = (E 2 − 1) + − 2 + . (9.31)
r r r3

74
Let u = r−1 , then
 ∂u 2 u̇2 ṙ2 1
= = = 2 f (u) (9.32)
∂ϕ ϕ̇2 r4 ϕ̇2 J
where
f (u) = (E 2 − 1) + 2M u − J 2 u2 + 2M J 2 u3 .
We integrate (9.32) and set u = u0 at ϕ = ϕ0 , to find
Z u Z ϕ ′
du′ dϕ
p = (9.33)
u0 f (u′ ) ϕ0 J

which is an elliptic integral. To find approximate solutions, we first observe the following
behavior of f (u): for 2M ⩽ r < ∞,
 1   1  J2
f = E 2 > 0, f (0) = E 2 − 1, f˙(0) = 2M > 0, f˙ = 2M + > 0.
2M 2M M
Since f (u) is a cubic, several distinct cases arise:

1. f (u) is positive throughout. The orbit has no turning points, so the orbiting body
plunges into the central body.

2. f (u) is positive from r = ∞ = 1/u down to a certain value of u, so u is monotonic

in ϕ to here. Then there is a turning point. This is a hyperbolic orbit: the orbiting
body comes in, reaches a closest approach and goes back out. There is also a region
near r = 2M where f (u) is positive. This corresponds to the orbits of bodies rising
from the central body and falling back.

3. f (u) is only positive in a region of u not including 0, that is for r− < r < r+ for
some finite r± . The orbiting body is confined between these values of r which are
therefore perihelion and aphelion. This is an elliptic orbit. Note that E 2 < 1 for this
case. Again there are orbits arising from the central body.

4. There are only orbits arising from the central body.

Recall the Newtonian case, (in the Solar system mu = m/r ≪ 1)

 ∂u 2 1
= (h + 2M u − J 2 u2 ), h = E2 − 1 (9.34)
∂ϕ J2
the solution of which is
M J2
u = 2 (1 + e cos ϕ), 2
e =1+h 2 (9.35)
J M
where the integration constant has been chosen so that du/dϕ = 0 at ϕ = 0. This orbit is
then an ellipse or hyperbola with focus at the origin according as to whether e2 is less than
or greater than one. Next, suppose that the cubic term is small, which is indeed the case in
reality. We then assume an approximation of the form
M
u= (1 + e cos ϕ) + V (9.36)
J2

75
where V is a perturbation term. Then (9.32) becomes, keeping only the O(V ) terms,

∂V M3
sin ϕ = V cos ϕ − 4 (1 + e cos ϕ)3 (9.37)
∂ϕ eJ

whence
M 3 h 1 + 3e2 e2 e2
i
V = 4 cos ϕ − cos 2ϕ + 3 1 + + 3eϕ sin ϕ . (9.38)
J e 2 2
We see that the first three terms are periodic, so that they may change the shape of the orbit
slightly but the orbit will remain closed. The fourth term is not periodic. If we think of the
period of the orbit as the angle in ϕ between two successive perihelia, then we must look at
the turning points to see what this is. To this end, consider

∂u Me M3 h 1 i
2
= − 2 sin ϕ + 4 − sin ϕ + e sin 2ϕ + 3eϕ cos ϕ (9.39)
∂ϕ J J e

which equals 0 at ϕ = 0, and then the next zero is not at ϕ = π, as it would be in Newtonian
gravitation, but at ϕ = π + ϵ. From (9.39)= 0 we find

3M 2
ϵ= π. (9.40)
J2
Thus in one complete revolution, the perihelion moves on from where it ’would be’ in
Newtonian theory by an amount 2ϵ, the ’perihelion advance’.
Next, we turn to the null geodesics. In this case, (9.30) is replaced by
 2M  2  2M −1 2
L= 1− ṫ − 1 − ṙ − r2 ϕ̇2 = 0. (9.41)
r r
Then we have as before  ∂u 2 E2
= 2 − u2 + 2M u3 . (9.42)
∂ϕ J
The Newtonian equation is found by omitting the last term, and the solution is

E
u= sin(ϕ − ϕ0 ) (9.43)
J
which is a straight line.
Suppose ϕ0 = 0 so that the straight line is just y = J/E and let

E
u= sin ϕ + V (9.44)
J
Keeping only linear terms in V we find from (9.42) that

dV M E2
cos ϕ = − sin ϕV + sin3 ϕ (9.45)
dϕ J2

whence
3M E 2 1
V = 2
(1 + cos 2ϕ). (9.46)
2J 3

76
Consider the limit of ϕ as r → ∞. If ϕ → ϕ∞ ≪ 1, then from (9.44) and (9.46) we find

E 3M E 2 1
ϕ∞ + 2
(1 + ) = 0, (9.47)
J 2J 3
so
2M E
ϕ∞ = − . (9.48)
J
The total deflection will be 2ϕ∞ .
Interestingly, the light from a distant star or galaxy would be deflected by various strong
gravitational fields, thereby producing multiple images for us to observe. This phenomenon
is called the gravitational lensing. The recently observed imaged of blacks by Event Horizon
Telescope are also effects of the light deflection, cf. Appendix C.
The radial null geodesics are null geodesics with J = 0, so from (9.29) and (9.41), we
obtain

ṙ2 =E 2 (9.49)
dt ṫ  2M −1
= =± 1− (9.50)
dr ṙ r
From (9.49) for an ingoing null geodesic we have

r = r0 − Es (9.51)

where s is an affine parameter, and from (9.50)

t = −r − 2M ln(r − 2M ) + const. (9.52)

For an outgoing null geodesic the signs are switched. An important point is that when
r < 2M the time-like and space-like vectors are switched. As we can see from (9.52), when
r < 2M , t decreases as r increases. In other words, a photon (and also those particles inside
a lightcone) falling into a black will inevitably fall to the singularity.
The radial motion of a massive particle is quite different. For a massive particle which is
originally static (E = 1) at infinity r → ∞, its radial (J = 0) free-fall follows (9.31):
r
2M dr rs
2
ṙ = ⇒ =± (9.53)
r dτ r

where τ is the proper time, and the minus (−) stands for radially ingoing motion. Mean-
while, the condition E = 1 gives
 2M  dt dt  rs −1
1− =1 ⇒ = 1− . (9.54)
r dτ dτ r
Thus, the 4-velocity of this type of particle is
r r
 rs −1 rs    rs −1 rs 
a
u = 1− ,− , 0, 0 , and ua = 1, 1 − , 0, 0 . (9.55)
r r r r

77
9.3 The extended Schwarzschild solution
So far, we have only worked with a particular patch of the Schwarzschild solution. To find
other patches of the manifold, we observe that the geodesics on a manifold with metric are
intrinsic, so that once we have the metric in one patch, we can find the geodesics in that
patch and use them as a guide to lead us to other patches.
We derived the Schwarzschild metric in the patch specified by

−∞ < t < ∞, 2M < r < ∞, 0 ⩽ θ ⩽ π, 0 ⩽ ϕ < 2π (9.56)

To extend to the patches with r < 2M , let us introduce the ingoing Eddington-Finkelstein
coordinate
v = t + r + 2M ln(r − 2M ), (9.57)

then the radially incoming null geodesic (9.52) is given by

v = const., r = r0 − Es. (9.58)

As we can see, unlike t, v is now finite at r = 2M . The Schwarzschild metric in these

coordinates becomes
 2M  2
ds2 = 1 − dv − 2dvdr − r2 (dθ2 + sin2 θdϕ2 ), (9.59)
r
and the allowed range of coordinates is now

−∞ < v < ∞, 0 < r < ∞, 0 < θ < π, 0 ⩽ ϕ < 2π (9.60)

Likewise, we can introduce the outgoing Eddington-Finkelstein coordinate

u = t − r − 2M ln(r − 2M ) (9.61)

and obtain  2M  2
2
ds = 1 − du + 2dudr − r2 (dθ2 + sin2 θdϕ2 ), (9.62)
r
with
−∞ < u < ∞, 0 < r < ∞, 0 < θ < π, 0 ⩽ ϕ < 2π (9.63)
In these patches, the metric is still singular at r = 0. We can calculate the Kretschmann
scalar of the Schwarzschild solution:
48M 2
K = Rabcd Rabcd = . (9.64)
r6
We therefore see that r = 0 is an intrinsic singularity.
Next, we also consider the radial free-fall of a massive particle with 4-velocity given by
(9.55). Accordingly, we can define a new time function T such that
r
 rs −1 rs
dT = dt + 1 − dr. (9.65)
r r

78
Thus, we have r
 rs 2
2 2
ds = dT − dr + dT − r2 dΩ2 (9.66)
r
This set of coordinates (T, r, θ, ϕ) are called the Painlevé-Gullstrand coordinates. As one can
easily check, we have in these coordinates Ṫ = 1, θ̇ = 0 = ϕ̇. These Painlevé-Gullstrand
coordinates are horizon-penetrating and regular at the Schwarzschild radius. In these co-
ordinates, one can calculate (Ex.)
rs rs rs rs
R0101 = − , R0202 = , R2323 = , R1212 = R1313 = − , (9.67)
r3 2r3 r3 2r3
as a consequence of which the Jacobi equations for the spatial η i become

D2 η 1 rs D2 η 2 rs D2 η 3 rs
2
= 3 η1, 2
= − 3 η2, 2
= − 3 η3. (9.68)
Dτ r Dτ 2r Dτ 2r
We see that in the radial r-direction the tidal force is a stretching force, while in the angular
directions it is a compressing force.
Alternatively, if we consider the following set of coordinate transformations
r
r  r  rs
dt = dτ + 1 − dρ, dr = (−dτ + dρ), (9.69)
r − rs r − rs r

the Schwarzschild solution in terms of (τ, ρ, θ, ϕ) is

rs
ds2 = dτ 2 − dρ2 − r2 (τ, ρ)(dθ2 + sin2 θdϕ2 ) (9.70)
r(τ, ρ)

which are known as the Lemaitre coordinates. Note that the coordinate ρ is comoving,
i.e. a constant along the time-like geodesics, and τ is the proper time of all the comoving
observers.
For one more coordinate choice, let

U = −e−κs u , V = eκs v (9.71)

where κs = 1/4M is the surface gravity. (See later sections.) The coordinates (U, V, θ, ϕ)
are known as the Kruskal-Szekeres coordinates, in which the Schwarzschild metric is

32M 3 − r
ds2 = e 2M dU dV − r2 (dθ2 + sin2 θdϕ2 ). (9.72)
r
It is clear that nothing singular happens at r = 2M : a coordinate U or V goes to zero, but the
metric is completely well-behaved. Now the coordinate range is simply −∞ < U, V < ∞;
in other words, the Schwarzschild metric is maximally extended in the Kruskal-Szekeres
coordinates.
Consider an alternative form of the Kruskal-Szekeres coordinates, X = (V − U )/2, T =
(V + U )/2, then (9.72) becomes

64M 3 −r/2M
ds2 = e (dT 2 − dX 2 ) − r2 dΩ2 . (9.73)
r

79
 i+ i+

I+ II I+

i0 III I i0

I− IV I−

i− i−

Figure 1: Penrose diagram for the Schwarzschild metric.

If we further set T = 0, then te remaining metric is

64M 3 −r/2M
ds2 = − e dX 2 − r2 (dθ2 + sin2 θdϕ2 ). (9.74)
r
To see what this metric is, we fix θ = π/2, then (9.74) becomes

64M 3 −r/2M  rs −1 2

2
ds = − e 2 2 2
dX − r dϕ = − 1 − dr − r2 dϕ2 (9.75)
r r
 −1
which is actually the metric of a cylinder, that is wormhole! Indeed, if we let 1 − rs
r
=
(dz/dr) + 1, then the above metric becomes the standard cylindrical metric ds = dz +
2 2 2

dr2 + r2 dϕ2 .
Let us define new coordinates Ũ , Ṽ by U = tan Ũ , V = tan Ṽ , then Ũ , Ṽ have finite
ranges. If we rescale the metric by

gab → (cos2 Ũ cos2 Ṽ )gab , (9.76)

then we have in particular −π/2 < U, V < π/2. The resulting spacetime diagram, with
coordinates X̃ = (Ṽ − Ũ )/2, T̃ = (Ṽ + Ũ )/2, is called the Penrose diagram, cf. Fig. 1.
The infinity is now sorted into five regions: past time-like infinity i− , the starting point of
infinite time-like geodesics; future time-like infinity i+ , their endpoint; space-like infinity
i0 , the endpoint of space-like geodesics; and past and future null infinity, I − and I + , the
initial and final points of infinite null geodesics.
We remark that, null geodesics are represented by the 45-degree lines, making it clear
that light can escape to I + from regions I, III, and IV but not from region II. Therefore,the
hypersurface that separates regions I, III, and IV from region II is the event horizon of the
Schwarzschild black hole.

9.4 Interior solution

Recall the calculation of the gravitational potential of a uniform sphere of mass in Newtonian
gravity. Suppose the sphere is of radius R and constant density ρ0 for r < R, then the
Newton-Poisson equation
∇2 Φ = 4πGρ (9.77)

80
with Φ(r) → 0 for large r, has solution
GM GM 2
Φ=− , (r > R); Φ= (r − 3R2 ), (r < R) (9.78)
r 2R3
where M = 4πR3 ρ0 /3. Now let us consider the similar situation in GR.
Before proceeding, let us consider first the non-relativistic self-gravitating fluid in hy-
drostatic equilibrium in Newtonian theory. Recall the EoM of a self-gravitating fluid

ρ[V̇i + Vj ∇j Vi ] = −∇i p − ρ∇i Φ. (9.79)

For a fluid in static equilibrium Vi = 0, so we are left with

∇i p = −ρ∇i Φ (9.80)

representing the balance of pressure and gravitation. In the spherically symmetric case for
any function f (r) we have
xi ′ 2
∇i f = f, ∇i ∇i f = f ′′ + f ′
r r
where ()′ = d()/dr. Then (9.77) and (9.80) are just
2
Φ′′ + Φ = 4πGρ, p′ = −ρΦ′ . (9.81)
r
We can (multiply by r2 ) integrate the first equation to get
4πG r
Z
k
Φ= 2 ρ(ξ)ξ 2 dξ + 2 .
r 0 r
By further assuming that the derivatives of Φ, ρ, p are nonsingular, we see that k = 0, as
Φ(0) should be nonsingular. Thus if we write
Z r
m(r) = 4π ρ(ξ)ξ 2 dξ (9.82)
0

we have
Gm(r)
Φ= , (9.83)
r
and
Gρm(r)
p′ = − . (9.84)
r2
Now suppose we have a spherically symmetric star of radius R and mass M in GR. For
r > R, the gravitational field is given by the Schwarzschild solution (9.9), while for r ⩽ R
we expect a solution of the form

ds2 = e2λ dt2 − e2µ dr2 − r2 dΩ2 (9.85)

with a stress-energy tensor of the form (3.40) with the fluid velocity
ξa
ua = p , (9.86)
ξb ξ b

81
where ξa = (1, 0, 0, 0) is the time-like Killing vector. This way, the components of Tab are

T00 = e2λ ρ, T11 = e2µ p, T22 = r2 p, T33 = r2 p sin2 θ. (9.87)

To write down the full Einstein’s equations, we need the Ricci scalar in addition to the Ricci
tensors (9.5)-(9.8):

R =g ab Rab = g 00 R00 + g 11 R11 + g 22 R22 + g 33 R33 =

λ′
µ′

=e−2λ e2λ−2µ (−λ′′ − (λ′ )2 + λ′ µ′ − 2 ) − e−2µ λ′′ + (λ′ )2 − λ′ µ′ − 2 −
r r
1 −2µ ′ ′ 1 −2µ ′ ′
− 2 e (1 − e + r(λ − µ )) − 2 2 e (1 − e + r(λ − µ )) sin2 θ =
2µ 2µ



r r sin θ
′ ′

λ − µ 2
=2e−2µ −λ′′ − (λ′ )2 + λ′ µ′ − − 2 e−2µ (1 − e2µ + r(λ′ − µ′ )). (9.88)
r r
Then
1 1
G00 =R00 − g00 R = 2 e2λ−2µ (1 − e2µ − 2rµ′ ) (9.89)
2 r
1 1
G11 =R11 − g11 R = − 2 (1 − e2µ + 2rλ′ ) (9.90)
2 r
1 λ′ − µ′

G22 =R22 − g22 R = r2 e−2µ −λ′′ − (λ′ )2 + λ′ µ′ − (9.91)
2 r
2
G33 =G33 sin θ. (9.92)

Thus, Einstein’s field equations become

1 −2µ
8πGρ = e (1 − e2µ − 2rµ′ ) (9.93)
r2
1
8πGp = − 2 e−2µ (1 − e2µ + 2rλ′ ) (9.94)
r
λ′ − µ′

8πGp =e−2µ −λ′′ − (λ′ )2 + λ′ µ′ − . (9.95)
r
Let us assume that various quantities are nonsingular at the origin, and define
 2Gm(r) 
e−2µ = 1 − . (9.96)
r
Then (9.93) reduces to
1 −2µ 1 1
8πGρ = 2
(e − 1 − 2rµ′ e−2µ ) = 2 [(re−2µ )′ − 1] = 2 [(r − 2Gm(r))′ − 1],
r r r
so we have
dm(r)
= 4πr2 ρ (9.97)
dr
and Z r
m(r) = 4π r′2 ρ(r′ )dr′ , with m(rs ) = M, (9.98)
0

which is an analog of (9.82).

82
 We also have from the other equations that

Gm(r)  4πr3 p  2Gm(r) −1

′
λ = 1+ 1− (9.99)
r2 m(r) r
3 
−Gρm(r)  p  4πr p 2Gm(r) −1
′
p = 1+ 1+ 1− (9.100)
r2 ρ m(r) r

which correspond to (9.83) and (9.84). (9.97) and (9.100) are known as the Tolman-Oppenheimer-
Volkoff (TOV) equations.
To solve the TOV equations, we can consider for simplicity the special and very idealized
case of a star of uniform density ρ0 . Then from (9.82), we have that

M r3
m(r) = M (r > R); m(r) = (r ⩽ R). (9.101)
R3
With the boundary condition p(R) = 0, i.e. pressureless at the boundary of the star, (9.100)
can be integrated to
h p1 − 2GM/R − p1 − 2GM r2 /R3 i
p(r) = ρ0 p p , r<R (9.102)
1 − 2GM r2 /R3 − 3 1 − 2GM/R

and 0 for r > R. Also, (9.99) gives

h3r r
2GM 1 2GM r2 i2
e2λ = 1− − 1− , r<R (9.103)
2 R 2 R3
and unchanged (Schwarzschild) for r > R. From (9.102) we see that p(r) reaches a maxi-
mum at r = 0,
h p1 − 2GM/R − 1 i
pc = p(r = 0) = ρ0 p . (9.104)
1 − 3 1 − 2GM/R
If we consider pc as a function of M/R, then pc is monotonically increasing, and to keep pc
finite,it is required that
M 4
⩽ . (9.105)
R 9
In other words, we cannot arbitrarily increase the mass of an object while keeping the radius
fixed. Compact stars saturating this bound is called Buchdahl star. Stars in realistic astro-
physical cases are of course more complicated, for example, one needs to consider quantum
mechanical pressures.
Interestingly, we can easily find an exterior metric for the spherically symmetric space-
time with a pressureless fluid. Recall the radial null geodesics (9.50), and now introduce a
new coordinate ṽ such that (9.50) is deformed to
 2GM −1
dt = dṽ + 1 − dr. (9.106)
r
If we change M to m(ṽ), then the Schwarzschild metric becomes
 2Gm(ṽ)  2
2
ds = 1 − dṽ + 2dṽdr − r2 dΩ2 (9.107)
r

83
which is called the Vaidya metric. We can find that (Ex.)
2m′
Rṽṽ = − , R = 0,
r2
where m′ = dm/dṽ. So, from Einstein’s equations we know that the stress-energy tensor is
that of pressureless fluid:
m′
Tab = − ka kb , ka = ∂a ṽ. (9.108)
4πr2

9.5 Birkhoff’s theorem

So far the metric is assumed to be static from the outset. In fact, we have a strong result,
known as Birkhoff’s theorem, saying that the spherically symmetric spacetime with T ab = 0
is static. (When there is a perfect fluid stress-energy tensor, as for the interior solution, the
TOV equations are then the necessary condition for the metric being static.)
Theorem 9.1 (Birkhoff ). If a spacetime has spherical spatial symmetry and vanishing Ricci
curvature, then it must be static. If the metric is singular as r → r0 = 2GM > 0 then the spacetime
metric is the Schwarzchild metric.
Proof. We present a proof using the Cartan method (cf. [22]). To start with, let M be a
spherically symmetric spacetime, which means that there exists a fiber bundle E : M → B
with local trivialization M = B ×S 2 , i.e. with fibers being the 2-spheres S 2 . Then, through
each event of M , there is a round 2-sphere with curvature 1/r2 , where r : B → R is
restricted to be a regular function on B. This way, the r can be chosen as a coordinate
function on B, and consequently we have another coordinate function t on B such that
dr and dt are orthogonal wrt the metric g on B. Here, the metric g on B is obtained by
projecting the metric ḡ to B. If we denote by h the metric on the round 2-sphere with
radius 1, then the metric (9.3) is simply

ḡ = g + r2 h.

Let ap , (p = 0, 1) be an othonormal frame field on B with a0 time-like, and denote its

dual 1-forms by φp . Let bu , (u = 2, 3) be an orthonormal frame field on S 2 with dual 1-forms
ψ u . Then there exist connection 1-forms φpq on B, as in (7.14), such that dφp = −φpq ∧ φq .
We denote the ‘curvature’ 2-forms Φpq = dφpq + φpr ∧ φrq . On S 2 we have dψ u = −ψ uv ∧ ψ v
and Ψuv = dψ uv + ψ uw ∧ ψ wv . Let us define the following vector fields on M :
1
āp = ap , φ̄p = φp , b̄u = bu , ψ̄ u = rψ u ,
r
and introduce the connection 1-forms φ̄pq , ψ̄ uv , ω̄ pu , ω̄ up on M , so that the yet unknown
differentials of these vectors fields on M will take the following general form:

dφ̄p = −φ̄pq ∧ φ̄q − ω̄ pu ∧ ψ̄ u , dψ̄ u = −ψ̄ uv ∧ φ̄v − ω̄ up ∧ ψ̄ p .

In fact, we can calculate

dφ̄p =dφp = −φpq ∧ φq = −φpq ∧ φ̄q ,

dψ̄ u =d(rψ u ) = rp φp ∧ ψ u − rψ uv ∧ ψ v = −ψ uv ∧ ψ̄ v − rp ψ u ∧ φ̄p

84
where we have assumed that dr = rp φp , so ω̄ pu ∧ ψ̄ u = 0 and

p rp rp
φ̄ q = φpq , ψ̄ uv = ψ uv , ω̄ up u
= rp ψ = ψ̄ u , ω̄ pu = −ḡ pq
ḡuv ω̄ vq = − ψ̄u .
r r
Next, let us introduce some notation. Suppose M has frame field ei and dual 1-form ω i .
Given a smooth function f : M → R, we write its Hessian as Hf = fij ω i ω j where fij is a
symmetric tensor on M such that fij ω j = dfi − fj ω ji . We denote (Hf )i = fij ω j . Also, if
Ωij are the curvature 2-forms of ei , we can write the Ricci form as Rj = ei ⌟Ωij . With the
notation explained, let us compute the curvature 2-forms:
rq u
Φ̄pq =dφ̄pq + φ̄pr ∧ φ̄rq + ω̄ pu ∧ ω̄ uq = Φpq + ω̄ pu ∧ ψ̄ = Φpq
r
rr rr u
Ψ̄uv =dψ̄ uv+ ∧ ψ̄ uw
+ ∧ ψ̄ wv= ω̄ ur
− 2 ψ̄ ∧ ψ̄v ω̄ rv Ψuv
r
rr
Ω̄up =dω̄ p + ψ̄ v ∧ ω̄ p + ω̄ r ∧ φ̄ p = d(rp ψ ) + ψ uv ∧ rp ψ v + ψ̄ u ∧ φrp =
u u v u r u
r
1 1
=drp ∧ ψ u + rp dψ u + rp ψ uv ∧ ψ v − rr φrp ∧ ψ u = rpq φ̄q ∧ ψ̄ u = −ψ̄ u ∧ (Hr)p
r r
1
Ω̄pu = − (Hr)p ∧ ψ̄u
r
The Ricci forms are then
1  2 2
R̄p =āp ⌟Φ̄pq + b̄u ⌟Ω̄up = ap ⌟Φpq + b̄u ⌟ rpq φ̄q ∧ ψ̄ u = Rp − rpq φ̄q = Rp − (Hr)p
r r r
1 r
  r r r 
R̄u =āp ⌟Ω̄pu + b̄v ⌟Ψ̄vu = āp ⌟ ψ̄u ∧ (Hr)p + b̄v ⌟ Ψvu − 2 ψ̄ v ∧ ψ̄u =
r r
r  ∆r r rr 
1 1 r r r 1 r
= − ap ⌟(Hr)p ψ̄u + bv ⌟Ψvu − 2 (2ψ̄u − h̄uv ψ̄ v ) = Ru − + 2 ψ̄u
r r r r r r
where ∆r = ḡ pq (Hr)pq .
To explicitly compute these terms, we set the metric on B by (N.B. the sign difference
from the rest part of the notes)

g = −e2 dt2 + g 2 dr2 ,

then φ0 = edt, φ0 = −edt, φ1 = φ1 = gdr. Consequently, dr = φ1 /g and hence r0 =

0, r1 = g −1 and rr rr = g −2 . In addition,
1 1
(Hr)0 = dr0 − r1 φ10 = − φ10 , (Hr)1 = dr1 − r0 φ01 = − dg.
g g2

We need to compute φ10 = φ01 (cf.(7.24)). Since

er 0 gt 1
dφ0 = d(edt) = er dr ∧ dt = − φ ∧ φ1 , dφ1 = d(gdr) = gt dt ∧ dr = − φ ∧ φ0 ,
eg eg

where er = ∂e/∂r and so on, we find φ01 = er 0

eg
φ + gt 1
eg
φ and φ00 = φ11 = 0. Noting that
gt 0 gr 1 et 0 er 1
dg = gt dt + gr dr = φ + θ , de = et dt + er dr = φ + θ
e g e g

85
we have the following partial derivatives g0 = gt /e, g1 = gr /g, e0 = et /e, e1 = er /g. Hence
er 0 gt 1 e1 0 g0 1
φ01 = φ + φ = φ + φ = (ln e)1 φ0 + (ln g)0 φ1 ,
eg eg e g

and consequently

1  1 
(Hr)0 = − (ln e)1 φ0 + (ln g)0 φ1 , (Hr)1 = − (ln g)0 φ0 + (ln g)1 φ1 .
g g
Moreover,
1 1 e 1  e
∆r = −(Hr)00 + (Hr)11 = [(ln e)1 − (ln g)1 ] = ln = 2 ln .
g g g 1 g g r

We write Rp = KB φp where KB is the sectional curvature of B. To compute KB , we

consider

Φ̄01 =dφ̄01 + φ̄0r ∧ φ̄r1 = dφ̄01 = d (ln e)1 φ0 + (ln g)0 φ1 =

=(ln e)11 φ1 ∧ φ0 − (ln e)1 φ01 ∧ φ1 + (ln g)00 φ0 ∧ φ1 − (ln g)0 φ10 ∧ φ0 =
 e g00  0
11
= −(ln e)11 − [(ln e)1 ]2 + (ln g)00 − [(ln g)0 ]2 φ0 ∧ φ1 = −


+ φ ∧ φ1
e g

where we have used φ01 = (−er /(eg))1 φ0 ∧ φ1 . Therefore,

e11 g00 1 h er  g  i
t
KB = − + =− − .
e g eg g r e t

Putting these altogether, we obtain the Ricci forms

 2  2
R̄0 = −KB + (ln e)1 φ̄0 + (ln g)0 φ̄1
rg rg
2  2 
R̄1 = (ln g)0 φ̄0 + −KB + (ln g)1 φ̄1
rg rg
h1 1 1  e i
R̄u = 2 1 − 2 − ln ψ̄u
r g rg g 1

where we have used Ru = ψu which can be checked by explicit calculations (cf. §7.4 of
[11]).
In the vacuum, Einstein’s field equations tell us that the Ricci tensor is vanishing, so the
vanishing Ricci forms entail

2 2
KB = (ln e)1 = − (ln g)1 , (9.109)
rg rg
2
(ln g)0 = 0, (9.110)
rg
1 1 1  e
1 − − ln = 0. (9.111)
r2 g2 rg g 1

From (9.110) we see that g = g(r). (9.109) implies that (ln eg)1 = 0, that is, eg(r) = f (t).
By introducing τ such that dτ = f (t)dt, then g = −g −2 dτ 2 + g 2 dr2 . Noting that e = g −1

86
in the (τ, r) coordinates and denoting h = e2 , we can rewrite (9.111) into (rh)r = 1 which
can be solved for h = 1 + c/r for some constant c. Since rg 2
(ln e)1 = rg12 (ln e2 )r = hr /r,
the (9.109) also can be written as (rh)rr = 1. Thus, the (9.109),(9.110) and (9.111) can be
solved for the metric of M as
r+c 2 r
ḡ = − dτ + dr2 + r2 dΣ2
r r+c
which is static. It is clear that by setting c = −2GM , the metric ḡ reduces to the Schwarzschild
metric.
The reverse statement that in GR a static gravitational field in vacuum must be spheri-
cally symmetric, is first proved by Israel and is now known as Israel’s theorem.

10 The Kerr solution

The Kerr metric is a solution of the vacuum field equations in the case of rotating mass.
The Kerr solution depends on two parameters: the total mass M and the total angular
momentum J. This simple fact holds due to the following result:
Theorem 10.1 (Carter,1971). If a vacuum spacetime manifold is an asymptotically flat, station-
ary, axially symmetric, with the further property that the spacetime is nonsingular at one of the event
horizons and in the region outside that horizon, then this vacuum spacetime belongs to the class of
Kerr solutions with two parameters M and J.
Here, the stationarity means a Killing vector ∂t , and axial symmetry means another
Killing vector ∂ϕ . As a consequence, the metric gab is only a function of (x1 , x2 ). If we
further require the inversion symmetry t → −t, ϕ → −ϕ, then g01 = g02 = g13 = g23 = 0,
so the metric takes the general form
ds2 = g00 dt2 + 2g03 dtdϕ + g33 dϕ2 + g11 (dx1 )2 + 2g12 dx1 dx2 + g22 (dx2 )2 .
 

Since a 2D Riemannian manifold is conformally flat, the x1 − x2 part is equivalent to the

Ω2 ηij . Thus we can write
ds2 = Adt2 − B(dϕ − ωdt)2 + Cdr2 + Ddθ2 (10.1)
where A, B, C, D, ω, are functions of (r, θ). The vacuum Einstein’s equations alone cannot
determine all these functions, which explains the need for those presumed properties in
Thm.10.1.
On the other hand, by fixing the coordinates (x1 , x2 ), we can solve ds2 = 0 for
r 
dϕ g03 g03 2 g00
=− ± − (10.2)
dt g33 g33 g33
We can see that the sign of g00 affects the behavior of (10.2): When g00 > 0, dϕ dt
and −g03
will have the same sign, so the direction of motion of a null particle must be same as the
direction of the rotation of spacetime. For g00 < 0, there is no such behaviors. We call the
surface defined by g00 = 0 the stationary limit surface or the ergosphere.
In later sections, we will see that the event horizon is a null hypersurface. A hypersurface
defined by f (xa ) = 0 is a null if g ab ∇a f ∇b f = 0. In the present case, if f (r) only depends
on r, then the null condition means g rr = 0.

87
10.1 Boyer-Lindquist coordinates
Let us present an elementary derivation. Consider the following form of metric ansatz in
the coordinates (x0 , x1 , x2 , x3 ) = (τ, ζ, θ, ϕ),
q
ds2 = γdτ 2 − Σ dζ 2 + dθ2 + dϕ2 + 2qdτ dϕ (10.3)


a
where γ, Σ are functions of (ζ, θ), q is a function of θ, a is a constant, and dτ = dt − qdϕ.
Let
Σ = a(p − q), ∆ = γp + (1 − γ)q
with p a function of ζ only. Then, we can compute the Levi-Civita connections:

pγ1 pγ2 + (1 − γ)2 q2

Γ001 = , Γ002 = ,
2∆ 2∆
q[pγ1 + (1 − γ)p1 ] q[pγ2 + (1 − γ)2q2 ]
Γ013 =− , Γ023 = − ,
2∆ 2∆
γ1 p1 q(qγ1 − p1 )
Γ100 = , Γ111 = −Γ122 = , Γ133 = ,
2Σ 2(p − q) 2Σ
q2 qγ1
Γ112 =− , Γ103 = − ,
2(p − q) 2Σ
γ2 q2 q 2 γ2 − [p + 2(1 − γ)q]q2
Γ200 = , Γ211 = −Γ222 = , Γ233 = ,
2Σ 2(p − q) 2Σ
p1 (1 − γ)q2 − qγ2
Γ212 = , Γ203 = ,
2(p − q) 2Σ
γ1 qγ2 − γ(1 − γ)q2
Γ301 = , Γ302 = ,
2∆ 2q∆
γp1 − qγ1 [pγ + (1 − γ 2 )q] − q 2 γ2
Γ313 = , Γ323 = , (10.4)
2∆ 2q∆
where the subscripts 1 and 2 denote respectively the derivatives wrt ζ and θ. The Ricci
tensor Rab has nonvanishing components R00 , R03 , R11 , R12 , R22 , R33 . We further assume
that ∆2 = 0, or equivalently

(1 − γ)q2 2f
γ2 = − ⇒ γ =1− . (10.5)
p−q p−q
R√
If we let σ = ∆dζ, then f is a function of σ only. Then, the relation qR00 +R03 becomes
 f˙ṗ   f˙ṗ 
2 2 2 2 2 2
p 2 q − 2qq22 + q2 − 2q ṗ − q 2 q + 3q2 − 2qq22 = 0 (10.6)
f f
where the dot means the derivative wrt σ. Since p, f are functions of σ, but q is a function of
θ, one has f˙ṗ/f = K and ṗ2 = cp + n for constants K, c, n. Then, we have the consistency
condition
2(K − c)q 2 − 2qq22 + q22 = 0 = 2nq + 2Kq 2 + 3q22 − 2qq22 (10.7)
which entails c = 2K, and

q22 = −2Kq 2 − nq, ṗ2 = 2Kp + n. (10.8)

88
Together with
ṗ2 = 2Kp + n, (10.9)
(10.5) (10.8) ensure that the vacuum Einstein’s field equations Rab = 0 are satisfied. When
K > 0, n < 0 and p is not bounded above, one can solve these equations for
√
n 2K
K 2σ2 − n
q=− sin2
θ , p= , f = µ(p − a)1/2 (10.10)
2K 2 2K
where µ is a constant. By a suitable choice of parameters,
m r 2 + a2
K = 2, n = −4a, µ= √ , p= ,
a a
we can transform the above solution into the form of Kerr solution in the Boyer-Lindquist
coordinates. Let a = J/M be the specific angular momentum, so that the Kerr metric is

ds2 = dt2 − 2M rΣ−1 (dt − a sin2 θdϕ)2 − Σ(∆−1 dr2 + dθ2 ) − (r2 + a2 ) sin2 θdϕ2 .
(10.11)
where
Σ = r2 + a2 cos2 θ, ∆ = r2 − 2M r + a2 .
After some arrangement, we have equivalently

ds2 = Σ−1 ∆(dt − a sin2 θdϕ)2 − Σ(∆−1 dr2 + dθ2 ) − Σ−1 sin2 θ[(r2 + a2 )dϕ − adt]2 .
(10.12)
If we further take A = (r + a ) − ∆a sin θ, then the Kerr metric takes the generic form
2 2 2 2 2

of (10.1):
Σ∆ 2 A sin2 θ Σ
ds2 = dt − (dϕ − ωdt)2 − dr2 − Σdθ2 (10.13)
A Σ ∆
where ω = 2M ra/A is the dragging angular velocity. Obviously, if we redefine dϕ′ =
dϕ − ωdt, then (10.13) is a diagonal in the rotating frame (t, r, θ, ϕ′ ).

10.2 Horizons, Penrose diagram and ergosphere

It is obvious that in the limit a → 0 we recover the Schwarzschild metric. The Kerr metric
(10.12) is singular at Σ = 0, i.e. at r = 0 and θ = π/2, meaning a ’ring’ singularity. Cf. Fig.
2. On the other hand, the real roots of ∆ = (r − r+ )(r − r− ) = 0 exist when a2 ⩽ M 2 :
√
r± = M ± M 2 − a2 (10.14)

indicating the inner(-) and outer(+) horizons. Clearly, the condition a2 < M 2 ensures that
the singularity is hidden behind the horizon; when a2 = M 2 , we have the extremal Kerr
black hole. Cf. Fig. 3.
In depicting these Penrose diagrams, we have in fact made the coordinate transformation
of the t − r part of the Kerr metric to the Eddington-Finkelstein-type null coordinates.
When θ = 0, the t − r part of the Kerr metric reduces to
 2M r  2 2M r  2M r  2
ds2 = 1 − 2 dt − 2 drdt − 1 + dr . (10.15)
r + a2 r 2 + a2 r 2 + a2

89
 r−

r+
I

Figure 2: Penrose diagram for the Kerr metric. The dashed lines are the singular rings.

Now consider the null coordinates v = t + r, then (10.15) becomes

 2M r  2
ds2 = 1 − 2 dv − 2drdv. (10.16)
r + a2
This submanifold is not geodesic complete, as can be see from solving the geodesic equa-
tions. The geodesic equations derived from (10.20) are
 2M r  dv dr  2M r  dv 2 dv dr
1− 2 2
− = E, 1 − 2 2
−2 = L, (10.17)
r + a dλ dv r +a dλ dλ dv
where E = gvv v̇ + gvr ṙ/2 and L are the constants of motion. Since we require a2 < M 2 ,
there exists a solution
1 1
r = λ, v = c + F (r), F (r) = 2r + ln|r − r+ | − ln|r − r+ | (10.18)
B C
with h a2 i h a2 i
B = 2M (1 − 2 )1/2 + 1 , C = 2M (1 − 2 )1/2 − 1 .
M M
Clearly, F (r) → ∞ as r → r± . To ensure the geodesic completeness, we can introduce a
new set of null coordinates w = v − F (r), then (10.15) becomes
 2M r  2
ds2 = 1 − 2 dw − 2drdw, (10.19)
r + a2
and (10.20) becomes
 2M r 
ds2 = 1 − 2 dvdw (10.20)
r + a2
which is geodesic complete. In terms of u, v, we can construct the Kruskal-like coordinates
and depict the Penrose diagrams. Notice that these Penrose diagrams are for the t − r part
of the Kerr metric! For general null coordinates, see the next subsection.

90
 I

Figure 3: Penrose diagram for the extremal Kerr metric. Two horizons r± coincide.

The Penrose diagram thus constructed deserves a closer look. Let us consider a time-
like particle falling into a Kerr black hole along the axis θ = 0, then its motion in the t − r
directions is determined by the metric (10.15). From (10.17) we obtain

dr √ 2M r
= ± E 2 − V 2, V 2 = 1 − 2 . (10.21)
dλ r + a2

As r → ∞, we see that Vp → 1 − 2M/rpin consistency with the Schwarzschild result.

p

But as r → −∞, V → 1 + 2M/|r| = 1 − 2(−M )/|r|, which is effectively an anti-

gravity sourced by −M . In other words, gravity near the singular ring of the Kerr black
hole becomes repulsive! Let us discuss this situation in more detail. The potential function
V 2 = 1+M/|a| reaches its maximum at r = −a, so the problem becomes a potential-barrier
one:

1. E 2 ⩾ 1 + M/|a|: The particle starting at r = ∞ can pass through r = 0 and reaches

r = −∞. (The solid arrow in Fig. 4.)

2. 1 ⩽ E 2 ⩽ 1 + M/|a|: The particle starting at r = ±∞ cannot penetrate r = −a, while

the particle starting at r = ∞ can reach r = ∞ again as shown in Fig. 4 by the wavy
arrows.

3. E 2 < 1, there is no penetration whatsoever, and the particle oscillated between r± .

(The dotted arrows in Fig. 4.)

Notice that, in the region r < 0, it is possible for gϕϕ = −[(r2 +a2 ) sin2 θ+2M ra2 sin4 θ/Σ] >
0, so that the Killing vector ξϕ becomes space-like. In other words, in the anti-gravity
region, it is possible to have closed time-like curves!
The stationary limit surface can be obtained from gtt = 0:
√
rS ± = M ± M 2 − a2 cos2 θ. (10.22)

The region between r+ and rS + is called the ergoregion or ergosphere. See, e.g. Fig. 5.

91
 ] Q

Figure 4: Penrose diagram for the time-like geodesics falling into the black hole. The solid,
wavy and dotted arrows are explained in the main text.

+
C
E r+ b rS +
Bf
A

Figure 5: E is the ergoregion of a Kerr black hole. The Penrose process A → B + C is also
depicted.

The famous Penrose process tells us it is possible to extract energy from the ergoregion!
Suppose that an observer at rest at the infinity, say E, sends a particle A into a Kerr black
hole. Let ua = (1, 0, 0, 0) be the 4-velocity of the observer, then this observer sees the
energy of the particle A as
E A = PaA (E)ua = P0A (E).
Suppose the particle A decays into two particles B and C at the position D, then by the
conservation of 4-momentum (3.4), we know that P A (D) = P B (D)+P C (D). If the particle
C returns to the infinity, the observer would find that

E C = P0C (D) = E A − P0B (D).

If the particle B could escape rS + along a time-like curve, then we have E C < E A . But if
the particle B cannot escape rS + , P0B = PaB ea0 could be negative, since inside the ergoregion
g00 = e0 e0 > 0. In this particular case, we have

E C > E A. (10.23)

92
Meanwhile, the angular momentum is also reduced. Indeed, the decrease in the mass is
accompanied by the change in the angular momentum

M 7→ M − P0B , J 7→ J + P4B .

Suppose the 4-velocity of the observer is ua = (1, 0, 0, Ω), then from the negative-mass
condition used above
E B = PaB ua = P0B + P4B Ω < 0,
we obtain that if P0B > 0 for the observer, then

P4B < 0. (10.24)

In other words, energy has been extracted from the black hole; at the same time, the M and
J of the Kerr black hole are reduced.

10.3 Geodesics in the Kerr spacetime

Consider as before the Lagrangian L = 21 gab ∂0 xa ∂0 xb , where gab is now given by the Kerr
metric (10.13). Since gab is now independent of t and ϕ, we see immediately two Killing
vectors ξt , ξϕ . The contraction of the Killing vector ξ a with the tangent vector γ̇ a to a
geodesic gives rise to a constant of motions, ξ a γ̇a =const.. For time-like geodesics, we can
a
choose γ̇ a = ua as the particle 4-velocity, where ua = dx dτ
is tangent to the geodesic. Thus
if we take ξ = P = mu be the 4-momentum of a test particle moving along the geodesic,
a a a

then we can obtain four constants of motion: i) the particle mass m; ii) the azimuthal angular
momentum Φ = ξϕ uϕ = −pϕ ; iii) the energy E = ξt ut = pt ; iv) Carter’s constant K. It is
more convenient to express these constants of motion as the following specific quantities,

Φ E K p2  λ2  2
λ= = −gϕu uu , γ= = θ2 + a2 (1 − γ 2 ) +
= gtu uu , Q= cos θ.
m m m m sin2 θ
(10.25)
We can consider the simple equation (3.1) for the particle mass and express the constant
motion m as

g tt pt pt + g tϕ pt pϕ + g ϕϕ pϕ pϕ + g ϕt pt pθ + gθθ pθ pθ + grr pr pr = m2 (10.26)

where pθ = dθ/dτ and pr = dr/dτ . By substituting relevant terms into the above equation,
we can obtain the equation of motion for a free test particle in the Kerr spacetime:

1 2 2 (r2 + a2 )2  2 1  1 a2  2 2a  r 2 + a2 
a sin θ − E + − Φ − 1− EΦ+
Σ ∆ Σ sin2 θ ∆ Σ ∆
 dθ 2 Σ  dr 2
+Σ + = −m2 . (10.27)
dτ ∆ dτ
For simplicity, we consider the geodesic motion in the equatorial plane (θ = π2 , dτ
dθ
= 0, Σ =
r ). In this case, the EoM reduces to
2

 dr 2
4
Vr = r [r(r2 +a2 )+2a2 M ]E 2 −4aM EΦ−(r−2M )Φ2 −∆rm2 . (10.28)

r = Vr ,
dτ

93
The circular motion corresponds to the condition Vr = 0 and dVr /dr = 0, from which we
obtain
r3/2 − 2M r1/2 ± aM 1/2 m1/2 (r2 ∓ 2aM 1/2 r1/2 + a2 )
γ= , λ=± .
r3/4 (r3/2 − 3M r1/2 ± 2aM 1/2 )1/2 r3/4 (r3/2 − 3M r1/2 ± 1aM 1/2 )1/2
(10.29)
When γ < 1, the particle is in a a bounded orbit. We see that for large r, γ tends to 1 and
hence the orbits are always bounded; when r is small, the ± distinguishes the cases with
Φ > 0 and Φ < 0. The minimal bounded orbits are obtained by setting γ = 1, and the
result is
rmb = 2m ∓ a + 2m1/2 (m ∓ a)1/2 (10.30)
Moreover, the condition d2 Vr /dr2 ⩾ 0 is a stability condition. Then d2 Vr /dr2 = 0 gives the
innermost stable circular orbit (ISCO), with

rISCO = m{3 + Z2 ∓ [(3 − Z1 )(3 + Z1 + 2Z2 )]1/2 } (10.31)

where Z1 = 1 + (1 − a2 /M 2 )1/3 [(1 + a/M )1/3 + (1 − a/M )1/3 ], Z2 = (3a2 /M 2 + Z12 )1/2 .
Another important case is the geodesic motion of a radial freely falling particle. In terms
of the constants λ, γ, Q, the radial geodesic equations are

dt A
= (γ − λω), (10.32)
dτ ∆Σ
dϕ A h Σ − 2M r  λ i
= + ωγ , (10.33)
dτ ∆Σ A sin2 θ
 dθ 2 1 h i
2 2 2 2
= 2 Q − λ cot θ + (γ − 1)a cos θ , 2
(10.34)
dτ Σ
 dr 2 1 h
2
i
2 2 2 2
= 2 γ(r + a ) − aλ − ∆ r + (λ − aγ) + Q . (10.35)
dτ Σ
If the free-falling particle starts at the spatial infinity with zero initial velocity and zero total
angular momentum, then γ = 1, λ = 0, pθ = 0 and Q = 0, so that the radial geodesic
equations are simplified to

dt A dϕ 2M ar dθ  dr 2 A − ∆Σ
= , = , = 0, = . (10.36)
dτ ∆Σ dτ ∆Σ dτ dτ Σ2
We can read from these equations the 4-velocity of the infalling particle,
√
 A A − ∆Σ 2M ar 
a
u = ,− , 0, . (10.37)
∆Σ Σ ∆Σ
The radial geodesic motions allow the introduction of Lemaitre coordinates [23]. To
this end, let us work with the form (10.13). Firstly note that the dr/dτ in (10.37) can be
integrated to
Z τ Z r r Z r
′ ′ Σ2 1 r′2 + a2 cos2 θ
− dτ = dr =√ dr′ p
c r0 A − Σ∆ 2M r0 r′ (r′2 + a2 )

94
where c is an integration constant. In stead of directly integrating the elliptic integral, we
consider the following: Let z = r/a, then
Z τ Z z
′ a3/2 z ′2 + a2 cos2 θ a3/2
− dτ = √ dz ′ p ≡√ [f (z, θ) − f (z0 , θ)].
c 2M z0 z ′ (z ′2 + a2 ) 2M
If we set z0 = 0 and c = ρ, we have that
a3/2 df dz df z 2 + a2 cos2 θ
dρ = dτ + √ dr, with = p .
2M dz dr dz z(z 2 + a2 )
Therefore we can choose the following coordinate transformation
√
 A  A A − ∆Σ
dt = 1 − dρ + dτ, dr = (dρ − dτ ). (10.38)
∆Σ ∆Σ Σ
Then (10.13) becomes synchronous and diagonal
 ∆Σ  2 A
ds2 = dτ 2 − 1 − dρ − Σdθ2 − sin2 θdϕ′2 (10.39)
A Σ
where dϕ′ = dϕ − ωdt. One can readily show that the tetrad adapted to (10.39) is geodesic
and Fermi-Walker transported (Ex.).
The radial null geodesics, in particular, give rise to the double null coordinates [24].
The analysis is, as expected, quite complicated. Briefly stated, since the null geodesics in the
Kerr spacetime are ‘twisted’ by the rotation, we need firstly to find null geodesics that are
radial, i.e. in a hypersurface orthogonal null geodesic congruence. To do this this we need
secondly to solve for the constants of motions K∗ (with E = 1, Φ = 0 fixed), so as to define
the null coordinate functions u, v in terms of K∗ . Suppose such a K had been solved, then
the outgoing and ingoing coordinate functions are respectively given by
p
(r2 + a2 )2 − K∗ ∆ p
du =dt − dr − ±|h K∗ − a2 sin2 θdθ, (10.40)
p ∆
(r2 + a2 )2 − K∗ ∆ p
dv =dt + dr ± |h K∗ − a2 sin2 θdθ, (10.41)
∆
where the symbol ±|h means + for the northern hemisphere and − for the southern hemi-
sphere at the future null infinity. The coordinate transformation is
1 h1 i ∆
dt = (dv + du), dr = (dv − du) − kdθ √ , (10.42)
2 2 R
where R = (r2 + a2 )2 − K∗ ∆. Then the Kerr solution (in the Boyer-Lindquist coordinates)
is transformed to
2 1 2M r Σ∆  2 2 1 2M r Σ∆ 
ds = 1− − (du + dv ) + 1− + dudv+
4 Σ R 2 Σ R
 2M ar sin2 θ Σ∆   2M ar sin2 θ Σ∆  AΣ 2 A 2
+ + kdθ dv + − kdθ du − dθ − sin θdϕ2 .
Σ R Σ R R Σ
(10.43)

The extended null coordinates can now be defined as before: U = −e−κ+ u , V = eκ+ v . The
resulting metric in the coordinated (U, V, θ, ϕ) can be readily obtained (Ex.).

95
10.4 Kerr-Eddington coordinates
To eliminate the singularities at r± , let us introduce the new coordinates (T, R, Θ, Φ) defined
by the following relations

dT = dt + 2M rdr/∆, dR = dr, dΘ = dθ, dΦ = dϕ − adr/∆ (10.44)

In these coordinates, the Kerr metric is

ds2 =dT 2 − dR2 − 2a sin2 ΘdRdΦ − ΣdΘ2 − (R2 + a2 ) sin2 ΘdΦ2

− 2M RΣ−1 (dR + a sin2 ΘdΦ + dT )2 (10.45)

where Σ = R2 +a2 cos2 Θ. Now the metric in this form is manifestly non-singular at ∆ = 0,
but retains the singularity at Σ = 0.
In both the Boyer-Lindquist and Kerr-Eddington coordinates, we can read off the time
and angular Killing vectors:

T a = (1, 0, 0, 0), Φa = (0, 0, 0, 1). (10.46)

Let us compute the norms of the Killing vectors:

2M  r2 
T a Ta =1 − (10.47)
r r2 + a2 cos2 θ
 2M r sin4 θ 
Φa Φa = − r2 sin2 θ − a2 sin2 θ + 2 . (10.48)
r + a2 cos2 θ
An interesting point immediately arises: neither of these expressions necessarily has a def-
inite sign, as long as r is allowed to be negative (N.B. We are extending the manifold into
another ’sheet’ where r is negative!).
In the Kerr-Eddington coordinates, we obtain the Killing vector at the horizons r± :
 a 
T± = ∂T + 2 ∂Φ . (10.49)
r ± + a2

10.5 Kerr-Schild coordinates and Doran form

Consider the transformation from the Kerr-Eddington coordinates

x + iy = (R − ia)eiΦ sin Θ, z = R cos Θ, τ = T. (10.50)

Then the Kerr metric in these coordinates (τ, x, y, z) is

ds2 =dτ 2 − dx2 − dy 2 − dz 2 −

2M R3  R(xdx + ydy) + a(xdy − ydx) zdz 2
− 4 + + dτ (10.51)
R + a2 z 2 R 2 + a2 R
where the function R(x, y, z) is determined by

R4 − (x2 + y 2 + z 2 − a2 )R2 − a2 z 2 = 0. (10.52)

96
At R = 0, we have a disk
z = 0, x = −a sin Φ sin Θ y = a cos Φ sin Θ,
and Θ = π/2 corresponds to the singular ring. Notice only the ring is singular, and r can
be analytically extended to the interior of the disk! The surfaces of constant R for R ̸= 0,
on the other hand, form a family of confocal ellipsoids. To see this, consider (10.52),
x2 + y 2 z2 x2 + y 2 z2
+ = − =1
R 2 + a2 R 2 a2 sin2 Θ a2 cos2 Θ
which is a family of hyperboloids.
The Kerr-Schild metric (10.51) belongs to the general form
gab = ηab + hℓa ℓb (10.53)
where ηab is a flat metric and ℓa is null. For the Kerr solution, ℓa is the principal null vector
of the Riemann tensor defined by
ℓc ℓ[a Rb]cd[e ℓf ] ℓd = 0. (10.54)
This form of Kerr metric is very suitable for analyzing the asymptotic flat structures (cf.
(8.35)), as we can see through bare eyes.
A very useful special case of the Kerr-Schild coordinates of the Kerr solution is the Doran
form [25]. In the coordinates (τ, x, y, z), the Doran form of the Kerr solution is
 2α 
2 a
ds = ηab dx dx −b 2
aa vb + α va vb dxa dxb , (10.55)
ρ
where
(2M R)1/2 a2 z 2
α= , ρ2 = R2 + 2 . (10.56)
ρ R
Again, R is a function implicitly defined by (10.52). The two vectors va , aa are
 ay −ay   rx ry z
va = 1, 2 , 2 2 1/2
, 0 , aa = (r + a ) 0, 2 , , ; (10.57)
a + r 2 a2 + r 2 a + r 2 a2 + r 2 r
they are related to the two principal null directions ℓa,± by ℓa,± = (r2 +a2 )1/2 va ±(αρva +aa ).
When transformed to the spherical coordinates (t, r, θ, ϕ), the Doran form (10.55) becomes
 ρ 2
ds2 = dt2 − 2 2 1/2
dr + α(dt − a sin2
θdϕ) − ρ2 dθ2 − (r2 + a2 ) sin2 θdϕ2 . (10.58)
(r + a )
In this case, we note that ρ = Σ = r2 + a2 cos2 θ. Like in the case of Painlevé-Gullstrand
coordinates, one can check that in these Doran coordinates we also have ṫ = 1, θ̇ = 0 = ϕ̇.
The important point of the Doran form of the Kerr solution is that the time coordinate
t now is just the local proper time for a radially free-falling observer. When θ = 0 = ϕ. the
radial geodesics are simply
 α(r2 + a2 )1/2 
ẋa = 1, − , 0, 0 , ẋa = (1, 0, 0, 0). (10.59)
ρ
In the case of Schwarzschild solution, the radial geodesics give rise to the Painlevé-Gullstrand
coordinates. Similarly, by a further coordinate transformation (Ex. not easy!), one can ob-
tain the Painlevé-Gullstrand form of the Kerr solution.

97
10.6 Newman-Penrose formalism for the Kerr solution
We present a derivation of the Kerr metric from the Schwarzschild metric using the Newman-
Penrose formalism. From the Schwarzchild solution in the ingoing Eddington-Finkelstan
coordinates (9.59), we can find its null tetrad
1 2M 1  1 
a a
l =∂r , n = ∂v + (1 − a
)∂r , m = √ ∂θ − i ∂ϕ . (10.60)
2 r 2r sin θ
Now let
M M
na = ∂v + (1 − − )
r r̄
where
r = r′ − ia cos θ, v = v ′ − ia cos θ, r′ , v ′ ∈ R.
Then the following null tetrad in the coordinate (r′ , v ′ .θ, ϕ)

′a ′a 1
 M r′ 
l =(0, 1, 0, 0), n = 1, − ′a , 0, 0 ,
2 r + a2 cos2 θ
1  −i 
m′a = √ ia sin θ, ia sin θ, 1, (10.61)
2(r′ + ia cos θ) sin θ

recover the Kerr metric. This method for deriving the Kerr metric is known as the Newman-
Janis algorithm.
In addition to the Killing vectors ∂t and ∂ϕ , there are also Killing tensors for the Kerr
geometry:
Kab = 2Σl(a nb) + r2 gab , (10.62)
where we have taken the Kinnersley null tetrad
1 2 1 2
la = (r + a2 , ∆, 0, a), na = (r + a2 , −∆, 0, a)
∆ 2Σ
(Ex. Solve for the ma .) It is worth noting that the linear combination of the Kinnersley
null tetrad can slove the Mathisson-Papapetrou equations for spinning particles in in the
equatorial plane of a Kerr black hole with the spin vector perpendicular to it, cf. [26].

11 Charged solutions
The charged solutions to Einsteins’ field equations are important for many reasons. We
have introduced the vacuum solutions and the solutions with a perfect fluid; in the cases of
charged solutions gravity is coupled to gauge fields.

11.1 The Reissner-Nordström solution

We have seen in Sec. 3 that the electromagnetic strength Fab = ∂a Ab −∂b Aa = ∇a Ab −∇b Aa
satisfy the field equation ∇a F ab = 0, as electromagnetic fields are abelian. The matter action
is
g2 √
Z  
Imatter = − −gdx4 F ab Fab . (11.1)
8κ

98
The Einstein’ equations are now
1 (F )
Rab − R = 8πGg 2 Tab (11.2)
2
where
√
(F ) 2 δ  1 cd
 1 h δ −g 2 √ δ  ec f d
i
Tab =− √ − Fcd F = √ F + −g F F
ef cd g g =
−g δgab 4 2 −g δg ab δg ab
1 1 1 1
= − gab F 2 + Fcd Fab (δae δbc g f d + δaf δbd g ec ) = − gab F 2 + (Fa f Fbf + Fea Fb e ) =

4 2 4 2
1 1 
= Fac Fb c − gab Fcd F cd . (11.3)
4 4
The spherically symmetric ansätz for the field strength can be made as

F̂ = Ftr (r)dt ∧ dr + Fθϕ (θ)dθ ∧ dϕ (11.4)

with dF̂ = 0. Recalling the Hodge dual of Fab , (3.35) and the spherically symmetric metric
(9.3), we have
√
−g 1 1 eλ+µ
∗Ftr = ϵ̃trθϕ g θθ g ϕϕ Fθϕ = ϵ̃trθϕ (eλ+µ r2 sin θ) 2 2 2 Fθϕ = 2 Fθϕ
2√ r r sin θ r sin θ
−g
∗Fθϕ = − ϵ̃θϕtr g tt g rr Ftr = −ϵ̃θϕtr (eλ+µ r2 sin θ)e−2λ e−2µ Ftr = −e−(λ+µ) r2 sin θFtr ,
2
that is,
eλ+µ
∗F̂ = −e−(λ+µ) r2 sin θFtr dθ ∧ dϕ + Fθϕ dt ∧ dr. (11.5)
r2 sin θ
Thus the field equations become

eλ+µ
d(∗F̂ ) = −∂r (e−(λ+µ) r2 Ftr ) sin θdr ∧ dθ ∧ dϕ + ∂θ (sin−1 θFθϕ ) dθ ∧ dt ∧ dr = 0, (11.6)
r2
which can be solved for
Q eλ+β P
Ftr (r) = , Fθϕ = sin θ (11.7)
4g r2 4g
where the constants Q and P are the electric charge and magnetic charge respectively.
With the electromagnetic fields explained, we can solve the Einstein’s field equations to
obtain (Ex.)
 2M Q2 + P 2  2  2M Q2 + P 2 −1 2
ds2 = 1 − + dt − 1 − + dr + r2 dΩ2 (11.8)
r r2 r r2

which is known as the Reissner-Nordström (RN) solution. By letting g00 = 0, we see that
there two horizons:
(11.9)
p
r± = M ± M 2 − (Q2 + P 2 ).
Now the RN metric can be rewritten as
∆ 2 r2 2
ds2 = 2
dt − dr − r2 dΩ2 , ∆ = (r − r+ )(r − r− ). (11.10)
r ∆

99
 T
I′

III r=0
r−
II
r+
I

Figure 6: Penrose diagram for the RN metric. The dashed line is the time-like escaping
path.

Obviously, when Q = 0 = P , the RN solution reduces to the Schwarzschild solution

(in this set of coordinates). A few words about the case with P = 0, Q ̸= 0: If Q2 > M 2 ,
then g00 = r∆2 > 0. In other words, there is no horizon to protect the singularity r = 0,
thereby making it naked, which is unphysical! So we only consider the case with Q2 < M 2 .
2
Let us introduce the Regge-Wheeler coordinates r∗ such that dr∗ = r∆ dr, and define the
Eddington-Finkelstein-like coordinates by dv = dt − dr∗ , du = dt + dr∗ , then, for example,
the ingoing metric is
∆
ds2 = 2 dv 2 − 2dvdr − r2 dΩ2 , (11.11)
r
and likewise for the outgoing one. We see that in these coordinates the metric is nonsingular
at r± .
Now the Kruskal coordinates cannot cover the whole RN spacetime. In fact, the Kruskal
coordinates are only well-defined wrt a particular horizon, r+ or r− . An important point
is that the intrinsic singularity is a time-like surface. So, it is possible for a moving parti-
cle with time-like 4-velocity throughout the in-fall to avoid the singularity and leave an
outer horizon enter a new asymptotically flat region. The Penrose diagram then consists of
various asymptotically flat regions connected by the black-hole tunnels. See Fig. 6.
In the case r+ = r− the RN black hole is called extremal. The Penrose diagram for an
extremal RH black hole is shown in Fig. 7.
As a final remark, we argue that the singular region of a RN black hole is physically inac-
cessible. To see this, recall the inequality (5.11) concerning the total integrated acceleration.
From this inequality, we know that if a timelike curve γ has finite integrated acceleration,
then the quantity pa Ka remains finite along the curve. But for a RN black hole, we have
 2M Q2 
pa Ka ⩾ mK a Ka = m 1 − + (11.12)
r r

100
 I′

III

Figure 7: Penrose diagram for the extremal RN metric. Now there is no region II, and the
time-like geodesics (dashed lines) directly enter region I′ .

which is divergent in the singular region. Then, the inequality (5.11) forbids a physical
particle with finite total integrated acceleration from accessing this singular region!

11.2 The Kerr-Newman solution

The Kerr-Newman solution for charged rotating massive bodies can be seen by replacing
the ∆ in the Kerr solution by

∆ = r2 − 2M r + a2 + e2 , e2 = Q2 + P 2 . (11.13)

In this case, the Maxwell 1-form is

1
Qr(dt − a sin2 θdϕ) − P cos θ[adt − (r2 + a2 )sϕ] . (11.14)

A=
Σ
The Kerr-Newman metric in the Boyer-Lindquist coordinates is then the same as (10.12).
The Killing tensor for the Kerr-Newman spacetime, as in (10.62), can be obtained from
the Killing-Yano tensors [27]. The principal Killing-Yano tensors in this case are the 2-
forms h:
h = db, b = r2 (dt − a sin2 θdϕ) − a2 cos2 θ(dt − adϕ) (11.15)
By using the formula (4.65), we obtain the explicit form of the Killing tensor,

a2 cos2 θ h 2 i2 a2 cos2 θ∆ r2 r2
K= (r + a2 )∂t + a∂ϕ − (∂r )2 + (a sin 2
θ∂ t + ∂ϕ )2
+ (∂θ )2 .
∆Σ Σ Σ sin2 θ Σ
(11.16)
Now we consider the expansion in a of the (electrically charged) Kerr-Newman metric.
To the order O(a),

dt2
ds2 = f dt2 − − 2a sin2 θ(f − 1)dtdϕ − r2 sin2 θdϕ2 − r2 dθ2 + O(a2 ) (11.17)
f
Q2
where f = 1 − 2M
r
− r2
, and the vector potential is

Q
A= (dt − a sin2 θdϕ) + O(a2 ). (11.18)
r

101
The corresponding approximate Killing-Yano tensor and the Killing tensor become respec-
tively

2b =r2 dt − ar2 sin2 θdϕ + O(a2 ), (11.19)

1
K =2a∂t ∂ϕ + (∂ϕ )2 + (∂θ )2 + O(a2 ). (11.20)
sin2 θ
We see that the first term in (11.20) consists of the two Killing vectors of the Kerr(-Newman)
spacetime. To make these Killing symmetries exact, i.e. corresponding to some solution to
Einstein’s equations, one can consider a slightly modified metric known as the (generalized)
Lense-Thirring metric

dt2  a(f − 1) 2
2
ds =f dt −2 2 2
− r sin θ dϕ + 2
dt − r2 dθ2 , (11.21)
f r
Qh  a i
A = dt − a sin2 θ dϕ + 2 dr . (11.22)
r fr

This metric admits the Killing symmetries (11.20) as exact symmetries!

12 General properties of black holes

12.1 Causal structure
Let us first introduce the causality structure on a general spacetime M . A causal curve is
a curve in M such that every pair of events/points on this curve are connected by either
time-like or null curves. Given a submanifold S of M , we define the causal future J + (S)
of S as the set of spaceitme events connected to S by those future-pointing causal curves
emanating from S, and define the chronological future I + (S) of S as the events in J + (S)
that are connected to S by time-like curves. Likewise we can define the causal past J − (S)
by past-pointing causal curves. Then E + (S) = J + (S) − I + (S) is the future horismos of S
which consists of the null curves from S, and E − (S) = J − (S) − I + (S) is the past horismos.
If two events cannot be connected by a time-like curve, they are called achronal. A
submanifold S is achronal if I + (S)∩S = 0. If S ⊃ I + (S), we say S is a future set. Obviously,
the boundary Ṡ = S̄ ∩ M − S of a future set S is achronal, which is called the achronal
boundary. For example, the future or past horismos of a point is an achronal boundary, as
there are only null curves connecting points on the horismos (cf. Fig. 8).
A curve γ : F → M is a future-directed non-spacelike curve if for any t ∈ F there exists
a neighborhood G ⊂ F of t and a convex neighborhood U ⊂ M of γ(t) such that, for any
t1 ∈ G, γ(t1 ) ∈ J − (γ(t1 ), U ) − γ(t) if t1 < t and γ(t1 ) ∈ J + (γ(t1 ), U ) − γ(t) if t1 > t.
The past-directed non-spacelike curves, future-directed time-like curves and past-directed
time-like curves are defined similarly. If for every neighborhood V of a point p ∈ M there
exists t ∈ F such that for t1 ∈ F with t1 ⩾ t one has γ(t1 ) ∈ V , then p is a future endpoint
of the future-directed non-spacelike curve γ. A future-directed non-spacelike curve γ is
future-inextendible if it has no future endpoint. The past-inextendible curve is defined
similarly. The inextendible curves are actually endless, i.e. they cannot be further extended.

102
 S

•q

M −S

Figure 8: A depiction of the future or past horismos as the achronal boundary. S is the
future set. The dashed line is a time-like curve.

An irreducible future set (IF) is a future set that cannot be expressed as the union of
two future sets, except that one future set is included in another future set. The irreducible
past sets (IP) are defined similarly. An IF (resp. IP) is the chronological future (resp. past)
of a timelike curve in M , or equivalently an IF (resp. IP) is generated by such a timelike
curve. IF can be further classified as follows: A proper IF (PIF) is the chronological future
of a point in M , for example I + (p); a terminal IF (TIF) is generated by past-inextendible
timelike curves. We have dually the PIP and TIP. TIP and TIF define the ideal points,
or hypersurfaces, of M : TIP defines the future boundary points and TIF defines the past
boundary points. We should distinguish the singular boundary points from the regular
boundary points at infinity. We say a TIP is an ∞-TIP if a generating curve for TIP has
infinitely long proper length (which is actually the proper time for a timelike curve). A
TIP that is not an ∞-TIP is called a singular TIP. The singular TIF is defined similarly. In
particular, a null-singular TIP is a light ray with finite affine length.
Given a set S of achronal events, we define the future domain of dependence D+ (S) of
S as the set of events such that each past-directed inextendible non-spacelike curve passing
through a point p ∈ D+ (S) will intersect S. Similarly, the past domain of dependence
D− (S) can be defined by replacing the past-directed curves by the future-directed curves.
The null boundary of D+ (S) is called the future Cauchy horizon H + (S), and likewise the
null boundary of D− (S) is the past Cauchy horizon H − (S). (For example, the r− in the RN
Penrose diagram!) Clearly, the Cauchy horizons are achronal. The causal region relevant to
S is D(S) ≡ D+ (S) ∪ D− (S). In general, D(S) cannot cover an entire spacetime manifold
M ; if D(Σ) for some 3-hypersurface Σ covers the entire M , we say this Σ is a Cauchy surface
and this M is globally hyperbolic.
The global hyperbolicity of a set or manifold is related to the uniqueness of solution to
the wave equations with a δ-function source. Given a point p ∈ M , if p can be determined
by its chronological past (future), i.e. I − (q) = I − (p) (I + (q) = I + (p)) entails q = p, then
the strong causality condition is said to be satisfied at p. Now suppose the strong causality
condition is satisfied in a set N ; if for any two points p, q ∈ N , J + (p) ∩ J − (q) is compact and
within N , then N is globally hyperbolic. In particular, the interior of a closed achronal set,
int(D(S)) = D(S) − Ḋ(S), is globally hyperbolic, so a manifold M is globally hyperbolic
if the Cauchy surfaces Σ exists. A particularly useful result is that for two points p, q in
a globally hyperbolic M with the property that q ∈ J + (p), there exists a (longest) non-
spacelike geodesics from p to q. As a consequence, every inextendible non-spacelike curve in

103
 i+ i+

I+ H+ I+

i0 B i0

I− H− I−

i− i−

Figure 9: Penrose diagram for the Kruskal-Szekeres extension of Schwarzschild metric.

The future and past horizons are shown. B means the bifurcating ball.

M intersects a Cauchy surface Σ exactly once. Topologically, a globally hyperbolic manifold

M is homeomorphic to R × Σ.
The Penrose diagram is based on a conformal compactification c : (M, gab ) → (M̃ , g̃ab )
where M̃ is compact manifold with a boundary and g̃ab = Ω2 gab . If the spacetime manifold
M has a conformal compactification M̃ such that each null geodesic in M is mapped to a
geodesic in M̃ that starts at I − and ends at I + , then we say M is asymptotically simple.
However, the black hole spacetimes are not asymptotically simple, as the the null geodesics
might fall into a black hole and cannot reach I + . We need weaker conditions: M is weakly
asymptotically simple if there is an open subset U ⊂ M that can be mapped by the conformal
compactification c to an open neighborhood U ′ of the boundary ∂ M̃ of M̃ such that U ′ ∩
M̃ is isometric to U . If furthermore the Ricci curvature Rab on U vanishes, M is then
asymptotically flat.

12.2 Horizons
Now suppose the infinities i− , i+ , i0 , I − and I + as in the Penrose diagram are well defined
for the spacetime manifold M . For S ⊂ M , we consider the topological closure J¯− (S) of
the causal past J − (S) of S, then its boundary is J˙− (S) = J¯− (S) − J − (S). If we take S as the
future null infinity I + of M , we define the following quantity as the future event horizon of
M:
H + = J˙− (I + ) (12.1)
Likewise, we can also define the past event horizon of M by

H − = J˙+ (I − ). (12.2)

For example, these event horizons are shown in the Penrose diagram of the Kruskal-Szekeres
extension, Fig.9. H + and H − are null hypersurfaces.
Let us consider the case of H + : As a null hypersurface, two points on H + cannot be
connected by a time-like curve. Indeed, suppose there is a time-like curve connecting two
points p, q with q ∈ J − (p); we can slightly deform this time-like curve into a new time-like
curve connecting p′ , q ′ with p′ ∈ J − (I + ) but q ′ ∈
/ J − (I + ), but q ′ ∈ J − (p′ ) ∈ J − (I + ), a

104
 i+

H+ I+

caustic • i0

I−

i−

Figure 10: Penrose diagram for a collapsing star.

contradiction. We therefore call a null geodesic the generator of the event horizon H + . The
point at which a generator leaves H + is called a caustic.

Theorem 12.1 (Penrose). The generator of H + can have caustics in its causal past, but not in
its causal future.

Proof. The past caustics exist because the generator could meet the singularity. (Cf. Fig. 10.)
For the future case, consider a point p ∈ J˙− (I + ) = H + . By definition, there exists a null
geodesic passing through p. Let us consider the points {pi ∈ J − (I + )} in a neighborhood
Up of p and the time-like geodesics {γi } passing through pi . Then these γi intersect the
boundary of Up at points qi ; these qi have a limit point q on J˙− (I + ). The curve connecting
p and q must be null. This process can be iterated, so that the point q is always on H + .

In other words, null geodesics can enter H + but cannot leave it, and null geodesics can
leave H − but cannot enter it.
More generally, a null hypersurface N is the Killing horizon of a bull Killing vector
ξ if ξ a is tangent to N . Hawking proved a theorem saying that the event horizon of an
a

asymptotically flat, stationary spacetime must be a Killing horizon [28].

Let Σ be s space-like hypersurface. A closed 2-surface T in Σ is called a trapped surface if
for all future-pointing null geodesics normal (or equivalently, tangent) to T the expansion
θ̃ < 0 (i.e. converging). Let T be part of Σ such that T contains the trapped surface T , then
T is called the trapped region. The boundary ∂T of T is the apparent horizon of Σ.

Theorem 12.2 (Penrose). Let M be a spacetime with metric gab , where gab is the solution to the
Einstein’s field equations with the stress-energy tensor satisfying the strong energy condition. If M
is globally hyperbolic with a noncompact Cauchy surface Σ and there is a trapped surface T in the
future of Σ, then there must be a null-singular TIP on M .

Proof. We follow a simplified proof given in [29]. Since M is globally hyperbolic, the interior
intD+ (Σ) = I + (Σ). Let B = ∂I + (T ) be the null boundary surface. By the same arguments
leading to Thm 12.1 and the fact that I + (T ) ⊂ I + (Σ) = intD+ (Σ), we know that the

105
null generators on B can only have past endpoint on T . (The past-inextendible curves in
intD+ (Σ) must intersect Σ. But now B ⊂ I + (Σ) = intD+ (Σ) needs not to interset Σ. Thus
the null generators on B being past-inextendible would lead to a contradiction.) Now that
T is a trapped surface, we have that θ̃ < 0 at the past endpoint on T . The strong energy
condition ensures that dθ̃/dτ ⩽ 0 (cf. Thm 6.2); together with the initial condition θ̃ < 0,
this shows the light rays starting at the past endpoint on T must contract to a conjugate
point in the future, if the parameter range is large enough. If the parameter range is not
large enough to reach this conjugate point, then the light rays with finite affine-length
will form a null-singular TIP. To see this, suppose such a TIP would not appear. Then
the light rays must leave B and enter intI + (T ) becoming timelike. In this situation, B
consists of finite affine-length null curves that converge to different endpoints on T ; they
could converge to their past endpoints on T , or to the caustics of their future endpoints, or
simply to the intersection point. That is, the future set I + (T ) as a 3D topological manifold
must be closed, i.e. compact and without boundary. But this will contradict the following
result for a time-orientable Lorentz manifold: There exists a globally smooth timelike vector field
ha , such that its integral curves can map injectively the compact achronal B to a homeomorphic image
B ′ in Σ. Because Σ is a noncompact 3D topological manifold, we see that this impossible.
Thus, there must exist a null-singular TIP.

Notice that the non-compactness consition can be relaxed, leading to the Penrose-
Hawking singularity theorem. The proof is more involved, which exploits the techniques
of geodesic incompleteness. Cf. [2].

12.3 Surface gravity

Let la be the null vector normal to the null hypersurface N satisfying the null geodesic
equation lb ∇b la = 0. Since ξa la = 0 and la la = 0, we see that the Killing vector can be
written as ξ a = f la for some function f . Then, the null geodesic equation becomes

ξ b ∇b ξ a = κs ξ a (12.3)

where κs = ξa ∂ a ln|f | is called the surface gravity. In terms of the time-like Killing vector
ξ = ∂/∂t, κs may be given a geometrical form
 1  12
a b
κs = − ∇ ξ ∇a ξb (12.4)
2 rs

(which can be seen by contracting the Frobenius integrability condition ξ[a ∇b ξc] = 0 with
∇a ξ b ).

Lemma 12.3. κs is a constant along the orbit of the Killing vector ξ a = f la .

Proof. By the definitions of Killing vector and the Riemann tensor, we have the affine
collineation condition6
∇a ∇b ξ c = Rcbad ξ d .
6
Notice the lack of minus sign due to the place of the contravariant index.

106
So,

ξ c ∇c κ2s = −(∇a ξ b )ξ c ∇c ∇a ξb = −(∇a ξ b )ξ c Rabcd ξd = −(∇a ξ b )Rabcd ξ c ξd = 0.

Thus if κs ̸= 0 on the orbit of ξ a , we could choose coordinates on N such that

∂ dλ d
ξa = =
∂α dα dλ
so that f = dλ/dα, la = d/dλ. Since κs is a constant, we can integrate κs = ∂α ln|f | to get
f = f0 eκs α . If we take f0 = ±κs , then
dλ
= ±κs eκs α , ⇒ λ = ±rκs α (12.5)
dα
where α is arbitrary. We therefore cover the λ > 0 and λ < 0 parts of N , while the λ = 0
part is called the bifurcating horizon B.
The surface gravity is the GR analog of the Newtonian gravity F = M/REarth 2
at the
surface of Earth. To see this, consider a general metric

ds2 = f (r)dt2 − f (r)−1 dr2 − r2 dΩ2 . (12.6)

A particle with rest mass m0 stays still at r, then its 4-velocity is

dt
ua = (ṫ, 0, 0, 0), ṫ = = f −1/2 , (12.7)
dτ
and hence its 4-acceleration is

aa = ub ∇b ua = u0 ∇0 ua = u0 (Γa00 u0 ) = f −1 Γa00 . (12.8)

Since Γ100 = 21 g 11 ∂1 (−g00 ) = 21 f f ′ , we have

1 p 1 f′
a1 = f ′ , a = |a1 | = −g11 a1 a1 = √ . (12.9)
2 2 f
This acceleration, when observed by an observer at infinity, will be red-shifted to be
dτ 1
a∞ = a = f 1/2 a = f ′ . (12.10)
dt 2
At the horizon rH , we say this a∞ (rH ) as the surface gravity. For the Schwarzschild metric,
we obtain
M 1 1
a= 2p , κs = . (12.11)
r 1 − 2M/r 4M
For the RN metric, we find
1
κ± = 2
(r± − r∓ ) (12.12)
2r±
for two horizons. And also, for the Kerr metric,
1
κ± = 2
(r± − r∓ ). (12.13)
2(r± + a2 )

107
 i+

V I+

i0L W i0R

I−

i−

Figure 11: Penrose diagram for the Minkowski spacetime. W is the Rindler wedge. The
dashed line is the orbit of the Killing vector ∂t .

12.4 Near-horizon geometries

Let us consider the near-horizon geometry of the Schwarzschild metric. Define a new
coordinate x by r − 2M = x2 /8M , then
2M (κs x)2
1− = → (κs x)2 , dr2 → (κs x)2 dx2 as r → 2M.
r 1 + (κs x)2
So the near-horizon geometry of the Schwarzschild metric is described by
1
ds2 = (κs x)2 dt2 − dx2 − 2
dΩ2 . (12.14)
4κ
This spacetime is the flat spacetime viewed by an accelerating observer with a constant
acceleration; its Penrose diagram is the same as that of the Minkowski spacetime (Fig. 11).
Focusing on the t − x part of (12.14), we can define as before the Kruskal coordinates
U = −xe−κs t , V ′ = xeκs t such that the t − x part of metric becomes
′

ds2 = dU ′ dV ′ (12.15)

which is called the Rindler spacetime. The x > 0 part corresponds to U ′ < 0, V ′ > 0, which
is called the Rindler wedge; x = 0 or r = 2M corresponds to U ′ = 0, V ′ = 0, which is the
Killing horizon wrt ξ a = ∂t with surface gravity κ. In general, the surface gravity κs ̸= 0
in the Rindler wedge.
To see the physical meaning of surface gravity κ in Rindler spacetime, let us consider an
observer with 4-velocity ua given by the Killing vector ξ a as in (9.86), then its 4-acceleration
is
ξ b ∇b ξ a ξ b ∇b (ξc ξ c ) a ξ b ∇b ξ a
aa = ub ∇b ua = − ξ = (12.16)
ξa ξ a 2ξa ξ a ξa ξ a

108
As ξ a = (1, 0, 0, 0) in the (t, x) coordinates, we can see from the normalization that for an
observer at rest at x, ξa ξ a = κs x and ua = ((κs x)−1 , 0, 0, 0). As before, we can calculate

1 ξ b ∇b ξ a
Γ100 = g 11 ∂0 (−g00 ) = xκ2s , a1 = = κs .
2 κs x

Thus, we have
a = κs . (12.17)
So κs is the proper acceleration of an observer in Rindler spacetime.
If we consider the observer at proper time t, and suppose she sees a particle with accel-
eration at , then
dτ 1
at = κs ⇒ at = . (12.18)
dt x
In other words, a particle at x = a−1 has constant acceleration a.
The near-horizon geometry of an extremal RN black hole is also very interesting. The
metric of the extremal RN is
 M 2 2  M −2 2
2
ds = 1 − dt − 1 − dr − r2 dΩ2 . (12.19)
r r
Let us define new a coordinate λ by r = M (1 + λ), then the metric (12.19) can be expanded
in λ, to the leading order in λ, to result in

ds2 ∝ (λ2 dt2 + M 2 λ−2 dλ2 ) + M 2 dΩ2 . (12.20)

This is an AdS2 × S 2 spacetime, which is called a Robinson-Bertotti spacetime. This AdS2

part of the near-horizon geometry exists for many extremal black holes, and paves the way
for more advanced topics such as the Kerr/CFT correspondence.

12.5 Gravitational charges

In an asymptotically flat spacetime, we have hab = gab − ηab = O( 1r ) as r → ∞. From
Einstein’s field equation, we obtain the Fierz-Pauli equations for the perturbation hab ,

1
∂ a ∂a hab + ∂a ∂b h + 2∂(b ha) = −2κ(Tab − ηab T ) (12.21)
2
where the labels are raised or lowered by ηab , and ha = η bc ∂b hca . By taking the trace of
(12.21), we have
∂ a ∂a h + ∂a ha = κT. (12.22)
Consider the weak static dust, i.e. the nonvanishing components are T00 = ρ, ρ̇ = 0 and
4πGρ ≪ 1, then (12.21) becomes

∂ i ∂i h00 = −κT00 , (12.23)

and (12.22) becomes

∂ i ∂i h00 − ∂ i ∂i hjj + ∂i ∂j hij = κT00 . (12.24)

109
From (12.24)−(12.23), we obtain
1
T00 = ∂i (∂j hij − ∂i hjj ), (12.25)
2κ
so that we can define the ADM mass for an asymptotically flat spacetime
Z Z
1
EADM = 3
d xT00 = dSi (∂j hij − ∂i hjj ). (12.26)
t=const. 2κ ∞
If we consider (12.24)+(12.23), then we get

∂i (∂j hij − ∂i hjj ) = −2∂ i ∂i h00 , (12.27)

so Z
1
EADM =− dSi ∂i h00 . (12.28)
κ ∞

(Note that the κ in the above formulas is the gravitational constant!) Since g ij Γ00j = − 12 ∂i h00 +
O(r−3 ) and ∂t is a Killing vector, we see that
Z Z Z
1 1 1
EADM = ij 0
dSi g Γ0j = i 0
dS0i ∇ ξ = − dSab ∇a ξ b . (12.29)
4πG ∞ 4πG ∞ 8πG ∞

where we have used the Killing equation in the last equality.

Let V be the volume of a space-like hypersurface. Given a Killing vector ξ a , we can
define the Komar integral I
c
Qξ (V ) = dSab ∇a ξ b (12.30)
2κ ∂V
where c is constant. By the definition of curvature and the Killing equation, we have
Z Z Z
c  1 a
Qξ (V ) = dSa ∇b ∇ ξ = c dSa T b ξ − T ξ ≡ dSa J a (ξ).
a b a b
(12.31)
2κ V 2
We can readily see that
 1  c c c
∇a J a (ξ) = c T ab ∇a ξ b − T ∇a ξ a − ξ · ∂T = − ξ · ∂T = ξ · ∂R = 0 (12.32)
2 2 2 2
which means that J a is a conserved current, and hence Qξ (V ) is a conserved charge.
When ξ a is a time-like Killing vector, Qξ (V ) gives rise to the ADM mass (12.29). When
ξ = ∂ϕ ≡ η, we obtain the angular momentum
I
1
J(V ) = dSab ∇a η b . (12.33)
2κ ∞
Next, let us turn to the relation between mass and angular momentum. To this end,
consider a space-like hypersurface Σ outside a stationary black hole that has boundaries on
i0 and a 2-sphere H on the event horizon. Cf. Fig. 12. Now the angular momentum is
Z I Z
1 a b 1 a b 1
J= dSa ∇b ∇ η + dSab ∇ η = dSa Rab η b + JH =
κ Σ 2κ H 8πG Σ
Z  1 a
a b
= dSa T b η − T η + JH . (12.34)
Σ 2

110
 i+
H
•
H+ I+

Σ i0

I−

i−

Figure 12: Σ (dashed line) is the space-like hypersurface for considering the Smarr formula.
The black dot means the 2-sphere H.

If there is only electromagnetic fields in the matter part, then we have by definition T = 0,
so Z
J= dSa T ab η b + JH . (12.35)
Σ

Similarly, the total mass can be expressed as

Z I
a b 1
M = − 2 dSa T b ξ + 2ΩH JH − dSab ∇a k b =
8πG
ZΣ I H
1
= − 2 dSa T ab k b + 2ΩH J − dSab ∇a k b (12.36)
Σ 8πG H

where ξ = ∂t and k = ξ + ΩH η with ΩH = 2Mar+ . We use the result that on H, dSab =

(ka nb − kb na )dA, where ka = √12 (1, 1, 0, 0), na = √12 (1, −1, 0, 0), that is
I I I
1 1 κ κs
− a b
dSab ∇ k = − a b
dA(k ∇a k )nb = − dAk b nb = A (12.37)
8πG H 4πG H 4πG H 4πG
where we have used the definition of surface gravity, (12.3). When there is no matter, we
obtain the Smarr formula (for a Kerr black hole)
κs
M= A + 2ΩH J. (12.38)
4πG

12.6 Black hole thermodynamics

Theorem 12.4 (Zeroth Law). If the stress-energy tensor satisfy the dominant energy condition,
i.e. Tab ξ a ξ b ⩾ 0 and T ab ξb must be non-spacelike, then the surface gravity κs on the future event
horizon H + is a constant.

To prove this, we need the following lemma:

Lemma 12.5. If N is a Killing horizon, then B̃ab = 0 for null geodesics on N and dθ̃
dσ
= 0 on
N.

111
Proof. Let ξ a = f la be a Killing vector on N with la ∇a lb = 0. We have known that ω̃ = 0
in this case, so

B̃ab = B̃(ab) = Pa c B̃(cd) P db = Pa c ∇(d lc) P db = Pa c (∂(d f )ξc) P db = 0

where in the last equality we have used the Killing equation. Since θ̃ = 0 on N , we also
have dσ
dθ̃
= 0 on N .

Corollary 12.6. If N is a Killing horizon, then Rab la lb = 0 on N .

This is an immediate consequence of the Raychaudhuri equation and Lmm. 12.5.

Proof of Zeroth Law. Let ξ a be a Killing vector orthogonal to H + , then we know from the
above results that Rab ξ a ξ b = 0 and ξ 2 = 0 on H + . By Einstein’s field equations, we have
that
0 = Tab ξ a ξ b |H + ≡ Ja ξ a |H + .
We expand J a = T ab ξ b by the basis of the tangent space of H + :

J a = aξ a + b1 η(1)
a a
+ b2 η(2)

where ξ a is null and η(i)

a
are space-like. By the dominant energy condition, we know that
it cannot be space-like. Thus, J a ∝ ξ a should be null, thereby
c c
0 = ξ[a Jb] |H + = ξ[a Tb] ξc |H + = ξ[a Rb] ξc |H + = ξ[a ∂b] κs |H + .

This means ∂a κs ∝ ξa ; we also see that there always exists a tangent vector t such that
t · ∂κs = 0.

Theorem 12.7 (First Law). If a stationary black hole with mass M , electric charge Q, angular
momentum J, future event horizon H + , surface gravity κ, electric surface potential ΦH , and angular
velocity ΩH , is perturbed and then relaxes to a new stationary black hole with mass M + δM , electric
charge Q + δQ and angular momentum J + δJ, then we have
κs
dM = dA + ΩH dJ + ΦH dQ (12.39)
8π
Proof of First Law with Q = 0. Recall that the outer horizon of a Kerr black hole is r+ =
√
M + M 2 + a2 , so the area of the outer horizon is
√
2
A = 4π(r+ + a2 ) = 8πM (M + M 2 − a2 ). (12.40)

Then a small change in the area is given by

8π a
δA = (δM − ΩδJ), ΩH = ,
κs 2M r+

or equivalently,
κs
δM = δA + ΩH δJ. (12.41)
8π

112
By the Carter theorem, we know that M is a function of (A, J). Since A and J both have
the dimensionality of [M ]2 , M (A, J) must be a homogeneous function of degree 1/2, then
using the Smarr formula, we have

1 ∂M ∂M κs  ∂M κs   ∂M 
M =A +J = A + ΩH J ⇒ A − A +J − ΩH = 0.
2 ∂A ∂J 8π ∂A 8π ∂J
Since A and J are free parameters, we obtain ∂M
∂A
= κs
8π
A, ∂M
∂J
= ΩH .

Theorem 12.8 (Second Law; Hawking [28]). If the weak energy condition holds for the energy
momentum tensor, i.e. Tab ξ a ξ b ⩾ 0, and if the cosmic censorship conjecture is true, then the future
event horizon in an asymptotically flat spacetime is a non-decreasing function of time.

Proof. Recall that for a null geodesic congruence, we have dσ

da
= 2aθ̃. If we can prove θ̃ ⩾ 0 on
H , then the theorem is proved. Suppose θ̃ < 0, then the geodesic congruence will always
+

contract to a point, including the future caustics, which contradicts Penrose’s theorem.

Conjecture 12.9 (Third Law). If the stress-energy tensor is bounded and satisfies the weak energy
condition, then the surface gravity of the black hole cannot reach zero in finite advanced time.

Werner Israel once provided a proof of the third law in [30]. However, recently Christoph
Kehle and Ryan Unger strongly disprove Israel’s argument in [31]. Basically, the above
“third law” is currently dead.

12.7 Hawking radiation: a tunneling glimpse

Black holes can evaporate through Hawking radiation. Indeed, it was the discovery of
Hawking radiation that makes the black-hole thermodynamical laws consistent. However,
a proper understanding of Hawking radiation requires the knowledge of quantum field
theories on curved spacetimes, which are beyond the scope of this course. Nevertheless, we
can take a glimpse at the process of Hawking radiation still using Penrose diagrams [32].
We consider a nonrotating, spherically symmetric black hole. Suppose such a black is
formed by the accretion of null matter shell, and then evaporates by emitting null matter
shell. We model the Hawking radiation process as a pair of null matter shell just outside the
horizon, consisting of an outgoing positive-mass shell and an ingoing negative-mass shell.
The simplest case with a single accretion and a single burst of Hawking radiation is depicted
in Fig. 13. More detailed Penrose diagrams can be found in [32].
There is in fact a simple way of calculating the Hawking temperature, known as the
tunneling method, without referring to the quantum field theories on curved spacetimes.
The ideal is to consider the quantum mechanical tunneling of the positive-mass particles
out of the horizon (or the black hole region) just like the potential tunneling in quantum
mechanics, while the negative-mass particles fall into the center of the black hole. This
physical picture is consistent with the Penrose diagram Fig. 13 if the gravitational potential
to be tunneled is outside H.
In quantum mechanics, the potential tunneling predicts a general form of the potential
tunneling amplitude Ψ ∝ eiI/ℏ , so the tunneling rate is Γ ∝ |Ψ|2 ∝ e2ImI/ℏ . It is hoped that
this rate should be also proportional to the Boltzmann weight, Γ ∝ eβE , so as to ensure

113
 _ •_ ?
T
H

Figure 13: A simplified Penrose diagram of an evaporating black hole. The dashed arrow
stands for the positive-mass matter shell, and the straight arrow stands for the negative-mass
shell. The region T is the trapped region!

a thermal radiation. Therefore, by the identification, e2ImI/ℏ ∼ eβE , we can obtain the
Hawking temperature 1/β. To implement this procedure, let us consider a general black
hole metric
ds2 = Adt2 − B −1 dr2 − gϕϕ (dϕ − N ϕ dt)2 − gij dxi dxj (12.42)
where A, B, N ϕ are functions of r, xi . Here, xi denotes other coordinates, for example,
θ in (10.13). We further suppose that at the horizon r = rH , A(rH ) = 0 = B(rH ) and
N ϕ (rH ) = ΩH . The black hole is possibly charged, with gauge potential Aa = (Φ, 0, Aϕ , 0)
and charge Q. To find I, suppose I satisfies the Hamilton-Jacobi equation

m2
g ab (∂a I + QAa )(∂b I + QAb ) + =0 (12.43)
ℏ2
where m is the rest mass of the tunneling particle. In the near-horizon region, we can
approximate ϕ by ϕ − ΩH t ≡ χ by assumption. Then we make the ansatz

I = Et + Jχ − W (r, xi ) (12.44)

where E is the energy of the tunneling particle, J is the angular momentum canonically
conjugate to χ. Substituting this ansatz (12.44) back into the Hamilton-Jacobi equation
(12.43), we obtain

∂W 1 h  m2 i1/2
≈ −√ (E − ΩJ − QΦ)2 − A 2 + Ψ (12.45)
∂r AB ℏ

with Ψ = g ij ∂i W ∂j W + gϕϕ−1
(J + QAϕ )2 . In the near-horizon region, since A ≈ 0, the
second term in [ ] can be ignored to obtain
Z ro Z ro
∂W 1
W ≈ dr =− dr √ (E − ΩJ − QΦ). (12.46)
ri ∂r ri AB

114
Note that this integration is wrt the the coordinate r, and we need to change to an inte-
gration over the proper spatial distance! When the angular coordinates are fixed, the proper
spatial distance σ is found by
dr
dσ = √ . (12.47)
B
The functions A, B can be Taylor expanded in the near-horizon region,
∂A ∂B
A= (r − rH ) + ..., B= (r − rH ) + ...
∂r ∂r
so that
Z  ∂B −1/2
dr 1 2  ∂A ∂B −1/2
σ= √ =2 (r − rH )1/2 + ..., √ = + ....
B ∂r A σ ∂r ∂r
Thus,
 ∂A ∂B −1/2 Z dσ
ImI = −ImW =2 Im (E − ΩJ − QΦ) =
∂r ∂r σ
 ∂A ∂B −1/2 1  E − Ω J − QΦ 
H
=2 × Res|σ=0 =
∂r ∂r 2 σ
 ∂A ∂B −1/2
=2π (E − ΩH J − QΦ) (12.48)
∂r ∂r
where the integration contour is chosen counter-clockwise around σ = 0. By identifying
e2ImI/ℏ = eβE , we read out the Hawking temperature (in the units ℏ = c = kB = 1)
1  ∂A ∂B 1/2
TH = . (12.49)
4π ∂r ∂r
In particular, comparing this to the first law of black hole thermodynamics (12.39), we
see that the T dS term in the conventional thermodynamics is recovered. In particular, for
a Schwarzschild black hole, the corresponding entropy takes a simple form
A
SBH = (12.50)
4
which is the well-known Bekenstein-Hawking entropy.

12.8 Geometric inequalities

In addition to the black hole thermodynamics, the geometric quantities of black holes can
also be combined in other interesting ways. (This subsection follows [33].)
We have seen that the Kerr solutions are characterized by the mass M and the p angular
momentum J. But for the cosmic censorship conjecture to hold, we require that |J| ⩽ M
to avoid naked singularities; the equality holds for extremal Kerr black holes. Recall also the
area of the outer horizon of a Kerr black hole, (12.40), from which and the above inequality
we obtain the following inequalities for a Kerr black hole:
r
A
⩽ M, (=, Schwarzschild), (12.51)
16π
|J| ⩽ M, (=, Extremal Kerr), (12.52)
p

8π|J| ⩽ A, (=, Extremal Kerr), (12.53)

115
 (M0 , J? 0 , A0 )

• H (M0 ⩽ M, J0 = J)

(J, M )
Σ

Figure 14: Penrose diagram of gravitational collapse. The wavy line means that there exist
gravitational waves. The area increases along the horizon; the mass decreases along the null
infinity.

where in the brackets we have indicated the situations where the equalities hold. (12.51) is
the well-known Penrose inequality. Recall the surface gravity at the outer horizon of a Kerr
black hole, (12.13), which is
1  (8πJ)2 
κ= 1− , (12.54)
4M A2
then (12.53) means that the surface gravity should be non-negative.
In fact, these inequalities also hold for non-stationary, or dynamical black holes. In
general, from (12.40) we can solve for
r
A 4πJ 2
MBH = + . (12.55)
16π A
Consider its variation in the form of (12.41); in axial symmetry with conserved angular
momentum (e.g. defined by Komar integral), δJ = 0, we see from Hawking’s area theorem
δA > 0 that δMBH > 0. But due to the form of the surface gravity κ as in (12.54), we see
that δMBH > 0 is equivalent to the inequality (12.53), which also indicates that the mass
term (12.55) can indeed be used as quasi-local mass for the black hole solely.
Next, consider a general scenario of gravitational collapse with the assumptions that i)
the cosmic censorship conjecture holds and ii) the spacetime settles down to a stationary
Kerr black hole with parameters (M0 , J0 , A0 ). Consider the situation as in Fig. 14: The
Cauchy surface Σ intersects the horizon at H whose area is A. By Hawking’s area theorem,
we have A0 ⩾ A. As the gravitational waves carry out positive energy, the mass decreases,
M0 ⩽ M . Moreover, suppose the axial symmetry is always preserved, so that the angular
momentum is conserved, J0 = J. In this situation, we have from above that
r
A 4πJ 2
+ = MBH ⩽ M0 ⩽ M, (=, Kerr) (12.56)
16π A
which means that in the dynamical situation the mass of the black should be smaller than
the total gravitational mass. When J = 0, this inequality reduces to (12.51).

116
13 Cosmology
We aim to construct solutions of Einstein’s equations with sources that will provide a model
for the large scale features of the universe. We assume that our model should be homoge-
neous and isotropic on average, which is consistent with the observed symmetries of the
universe to a to a high degree.
Roughly speaking, homogeneity means that at any given time, every points of the space
look the same. More precisely, let us consider an one-parameter family of space-like hyper-
surfaces Σt slicing the spacetime manifold M . If for any t and p, q ∈ Σt there exists a metric
gab defining an isometry from p to q, then we say M is (spatially) homogeneous.
As for isotropy, consider a congruence of time-like curves filling the spacetime. Given
a point p on one of such time-like curves, we denote the tangent vector to the curve at p as
ξ a . For two unit space-like vectors S1a , S2a at P orthogonal to ξ a , if there is such an isometry
given by the metric gab that fixes ξ a but moves S1a to S2a , we say M is (spatially) isotropic.

13.1 Homogeneous and isotropic three-spaces

In a 3D Riemannian manifold, the curvature tensor can be expressed in terms of the Ricci
tensor, Ricci scalar and metric tensor via
1
Rabcd = Rac gbd − Rbc gad + Rbd gac − Rad gbc − R(gac gbd − gbc gad ). (13.1)
2
Thus any anisotropy in the Riemann tensor must in this sense arise directly from the Ricci
tensor. We may consider a linear transformation on vectors at any point by ℓa → Rba ℓb .
This transformation will have eigenvectors, which are therefore preferred directions, unless
it is proportional to the identity. Then the homogeneity and isotropy about a point taken
together must entail that there are no preferred vectors at any point, that is, Rab = kgab for
some constant k. By taking the trace, the consistency implies that
1
Rab = Rgab , (13.2)
3
and then
1
Rabcd = R(gac gbd − gbc gad ). (13.3)
6
Applying the Bianchi identity to (13.3), we find that R must be a constant, i.e. the homoge-
neous and isotropic 3D manifold must be a constant curvature manifold.
If the metric gab has constant curvature, then clearly so has the metric ĝab = λ2 gab for
some constant λ. One can verify that R̂abcd = Rabcd and hence

R̂ = λ−2 R. (13.4)

This means that after rescaling the sign of R will not be changed, so there are only three
essentially distinct cases to consider: k = 0, 1, −1.
Since an isotropic metric should be spherically symmetric, we may have, after rescaling,

ds2 = dχ2 + f 2 (χ)(dθ2 + sin2 θdϕ2 ). (13.5)

117
The Ricci scalar of this metric is (Ex.)
 f ′′ 1 (f ′ )2 
R=− 2 − 2 + 2 . (13.6)
f f f
Since  f ′′ (f ′ )2 
′
f (f ′ )2 = f 2 f ′ 2 + 2 = −Rf 2 f ′ + f ′

f f
we have
1
f (f ′ )2 = f − Rf 3 . (13.7)
3
Now with R = 3k, the solutions to (13.7) are

1. k = 0, f = χ (flat space)

2. k = 1, f = sin χ (three-sphere S 3 )

3. k = −1, f = sinh χ (hyperboloid)

Let us discuss the case of k = 1. The metric

ds2 = dχ2 + sin2 χ(dθ2 + sin2 θdϕ2 ) (13.8)

is of S 3 . To see this, suppose S 3 is given by X 2 + Y 2 + Z 2 + W 2 = 1 in an Euclidean space

with metric ds2 = dX 2 + dY 2 + dZ 2 + dW 2 . We may introduce the polar coordinates

X = R sin θ cos ϕ sin χ, Y = R sin θ sin ϕ sin χ, Z = R cos ϕ sin χ, W = R cos χ.

Then the Euclidean metric becomes

ds2 = dR2 + R2 dχ2 + sin2 χ(dθ2 + sin2 θdϕ2 ) . (13.9)

Imposing the sphere condition R = 1 entails the metric (13.8).

Similarly, for k = −1, the metric

ds2 = dχ2 + sinh2 χ(dθ2 + sin2 θdϕ2 ) (13.10)

is the metric induced on the hyperboloid of unit future pointing time-like vectors in Minkowski
space. Suppose, as above, the hyperboloid is defined by T 2 − X 2 − Y 2 − Z 2 = 1, T > 0 in
the Minkowski space ds2 = dT 2 − dX 2 − dY 2 − dZ 2 . By introducing

X = R sin θ cos ϕ sinh χ, Y = R sin θ sin ϕ sinh χ, Z = R cos ϕ sinh χ, T = R cosh χ

we have
ds2 = dR2 − R2 dχ2 + sinh2 χ(dθ2 + sin2 θdϕ2 ) . (13.11)

Again, imposing the sphere condition R = 1 entails the metric (13.10).

Group-theoretically, isotropy about a point means the stabilizer of that point is isomor-
phic to the 3D rotation group SO(3). By homogeneity we mean that any point can be
taken to any other point by an isometry, i.e. that the isometry group is transitive. In fact,

118
a 3D homogeneous and isotropic space has the largest possible isometry group (6D) and is
therefore a maximally symmetric space. (Cf. §9.1.2 of [12].)
For k = 0, the isometry group of the flat space is the semi-direct product of three
translations and the rotation group SO(3). This is transitive and the stabilizer of each point
is SO(3). For k = 1, the isometry group of S 3 is SO(4), which is transitive but the stabilizer
of a point is the subgroup SO(3). For k = −1, the isometry group of the hyperboloid
is SO(1,3), the Lorentz group, which is transitive as there exists Lorentz transformations
relating points. The stabilizer subgroup is again SO(3).

13.2 Kinematics
Let us distinguish a preferred class of observers, namely those that actually see the universe
as isotropic, with a preferred time-like vector field ua , tangent to their worldlines. We
assume that there is a preferred family of hypersurfaces orthogonal to the worldlines of the
fundamental observers, i.e. the instantaneous rest spaces of the fundamental observers; these
hypersurfaces are required to be homogeneous and isotropic three-spaces.
We shall work with the comoving coordinates where the instantaneous rest spaces are la-
beled by the proper time t along the fundamental observers, and the three-space coordinates
are constant along the worldlines of the fundamental observers. Note that the fundamental
velocity ua is just the gradient of t: ua = ∇a t = (1, 0, 0, 0). Consequently, the metric on
the 3-hypersurface is known from the last subsection and can be expressed as
 dr2 
ds2 = dt2 − a2 (t)gij dxi dxj = dt2 − a2 (t) + r 2
(dθ 2
+ sin 2
θdϕ 2
) (13.12)
1 − kr2
which is known as the Friedmann-Robertson-Walker (FRW) metric. The a(t) is called the
scale factor. Here, r is the χ in (13.5).
Notice that in this form of spatial metric we have R̃ij = −2kg̃ij where g̃ij is the metric
of 3-space.
On the other hand, by the same considerations from symmetry (namely, the fundamen-
tal velocity ua plus the h.i. gab (t)), we can determine the stress-energy tensor to be that of
a perfect fluid (3.40)
Tab = (ρ + p)ua ub − gab p. (13.13)
In a sense, this fluid is a ‘gas’ of clusters of galaxies. From the conservation equation ∇a T ab =
0, we obtain the relativistic Euler equation (3.41) and the relativistic continuity equation
(3.42), which are reproduced here:
ua ∇a ub = 0, ∇a ((ρ + p)ua ) = ub ∇b p. (13.14)
The first equation is trivial, as ua is a unit vector and ua ∇a ub = ua ∇b ua = ∇b (ua ua )/2 = 0.
For the second , we need
1 √ 2ȧ
∇a ua = ∇a ∇a t = √ ∂b ( −gg ab ∂a t) =
−g a
where the dot means ∂/∂t. Then the second equation of (13.14) becomes
3ȧ
ρ̇ + (ρ + p) = 0, (13.15)
a

119
which can be rewritten, for later uses, as
d
(ρa3 ) = − 3pa2 (13.16)
da
(ρa3 )• = − p(a3 )• (13.17)

The (13.17) takes the form of a energy conservation relation: dM/dt = −pdV /dt.
Next, we consider the null geodesics in this geometry. Suppose that we, as observers,
are situated at the origin of the spatial coordinates, then the only null geodesics reaching or
leaving us are radial and given by
a2 (t)ṙ2
L = ṫ2 − = 0. (13.18)
1 − kr2
Integrating this, we obtain Z Z
dt dr
=± √ (13.19)
a(t) 1 − kr2
for incoming(-) and outgoing(+) null geodesics. Suppose a galaxy p at coordinate r1 releases
a flash of light at t1 that is received by us at r = 0 at t0 . Then by (13.19), we have
Z t0 Z r1
dt dr
= √ (13.20)
t1 a(t) 0 1 − kr2
and for a later flash Z t0 +δt0
dt
. (13.21)
t1 +δt1 a(t)
Thus, δt1 /a(t1 ) = δt0 /a(t0 ) as a cosmological time dilation, and in terms of frequencies,
λ0 /λ1 = a(t0 )/a(t1 ). The cosmological red-shift is

δλ a(t0 ) − a(t1 )
z= = . (13.22)
λ a(t1 )

For an expanding universe z > 0.

Of course, (13.20) defines both the past light cones of the origin at t0 and the future light
cones of the origin at t1 . If (13.20) has a finite limit as t0 recedes into the future, then there
will be events which are never visible to the fundamental observer at the origin, and we
say there is a cosmological (event) horizon. The cosmological horizon is observer dependent,
unlike the black hole horizon. Another possibility is that (13.20) could have a finite limit as
the lower limit of integration recedes into the past to some initial value. In this case there will
be fundamental observers who neither have seen nor have been seen by the fundamental
observer at the origin since infinite past, and we say the limiting boundary worldline of the
visible particles. is a particle horizon.

13.3 Dynamics
Now we consider Einstein’s equations in the FRW geometry. We can read from (13.12)
that the nonvanishing metric components are

gtt = 1, gij = −a2 g̃ij (13.23)

120
where g̃ij is the metric for 3-space:
1
g̃rr = , g̃θθ = r2 , g̃ϕϕ = r2 sin2 θ.
1 − kr2
The nonvanishing Christoffel symbols are

1
Γtij = g tt (−∂t gij ) = aȧg̃ij ,
2
1
Γitj = g ik (∂t gjk ) = a−1 ȧδji ,
2
1
Γijk = g̃ il (∂j g̃kl + ∂k g̃lj − ∂l g̃jk ) ≡ Γ̃ijk (13.24)
2
Next, the components of the Ricci tensor are

Rtt =Rtiti = ∂t Γiit + Γiit Γiti = 3a−1 ä

Rij =Riaj a = ∂i Γaaj − ∂a Γaij − Γpij Γaap + Γpaj Γaip =
= − ∂t Γtij + Γjtj Γtij + ∂i Γkkj − ∂k Γkij − Γpij Γkkp + Γpkj Γkip = R̃ij − (aä + 2ȧ2 )g̃ij (13.25)

As R̃ij = −2kg̃ij , we obtain the tt- and ij-components:

3a−1 ä = − 4πG(ρ + 3p) (13.26)

−1 −2 2
a ä + 2a ȧ + 2ka −2
=4πG(ρ − p) (13.27)

After eliminating ä between these we obtain

8
ȧ2 + k = πGρa2 (13.28)
3

which is known as the Friedmann equation. Also, the conservation equation gives

ρ̇ + 3a−1 ȧ(ρ + p) = 0. (13.29)

We still need the equation of state relating ρ and p to completely solve these equations.
Consider first the dust, i.e. p = 0. Then (13.16) becomes da d
(ρa3 ) = 0, so

ρ = M a−3 (13.30)

for some constant M . The Friedmann equation becomes

1
ȧ2 + k = κM a−1 , κ ≡ 8πG. (13.31)
3
To discuss its solutions, let us introduce a new variable ψ such that dt/dψ = a. We consider
the solutions for three cases of k:

1. k = 0. We can solve for

2/3
a = a(t − t0 ) . (13.32)
Then a2 = 3κM/4, thereby ρ = 34 κ−1 (t − t0 )−2 .

121
 2. k = 1. We have the Friedmann equation in the following form
 da 2 1
= κM a − ka2 . (13.33)
dψ 3

We obtain
1 1
a = κM (1 − cos(ψ − ψ0 )), t − t0 = κM (ψ − sin(ψ − ψ0 )) (13.34)
6 6
which is a cycloid, thereby closed.

3. k = −1. We obtain
1 1
a = κM (cosh ψ − 1), t − t0 = κM (sinh ψ − 1). (13.35)
6 6

Next, we consider electromagnetic radiation with equation of state, ρ = 3p. Again by

(13.16) we have
ρ = M a−4 , (13.36)
and the Friedmann equation becomes

1  da 2 1
ȧ2 + k = κM a−2 , or = κM − ka2 . (13.37)
3 dψ 3

The solutions are

1. k = 0. 4 1/4
a= κM t1/2 . (13.38)
3
2. k = 1. 1 1/2 1 1/2
a= κM sin ψ, t= κM (1 − cos ψ). (13.39)
3 3
3. k = −1. 1 1/2 1 1/2
a= κM sinh ψ, t= κM (cosh ψ − 1). (13.40)
3 3
These solutions behave qualitatively the same as the dust case.
In both cases (i.e. matter and radiation), there is an initial curvature singularity. This is
because from (13.26) we see that ä < 0 if ρ + 3p > 0 which generically holds, so the scale
factor is a function of t that is concave downwards. Once a is positive, it will eventually reach
0. Notice that such an argument is consistent with that using the Raychaudhuri equation,
as the (13.26) is equivalent to the Raychaudhuri equation. Indeed, if we set the expansion as

ȧ
θ=3 , (13.41)
a
then (13.26) becomes
1
θ̇ = −4πG(ρ + 3p) − θ2 (13.42)
3
which is just the Raychudhuri equation for the irrotational, shear-free and geodesic matter.

122
 If there is no matter but only a cosmological constant, the Friedmann equation becomes
 ȧ 2 Λ k
= − 2. (13.43)
a 3 a
For k = 0, the solution is the de Sitter solution
√
a(t) = a0 e Λ/3(t−t0 ) (13.44)
where a0 is a constant. Unlike previous solutions, the de Sitter solution cannot have started
with a big bang at finite times.
Now we consider two cosmological parameters that can be observationally determined.
These are the Hubble parameter H (with dimensions of inverse of time) and the dimen-
sionless deceleration parameter q:
ȧ aä
H = , q = − 2. (13.45)
a ȧ
These can be seen by expanding the scale factor a(t) around t0 , the time of today’s universe,
h ȧ
0 ä0 (t − t0 )2 i h i
a(t) = a0 1+ (t−t0 )+ +... = a0 1+H0 (t−t0 )−q0 H02 (t−t0 )2 +... . (13.46)
a0 a0 2
Roughly, knowing the H0 at present time, we know the age of the universe as H0−1 . We
introduce the critical density
3H 2
ρc = 3H 2 κ−1 = , (13.47)
8πG
then from the Friedmann equation we find
1 ρ 
k = κρa2 − ȧ2 = ȧ2 −1 (13.48)
3 ρc
so that the sign of k depends on the ratio of the actual density at any time to the critical
density at that time. For instance, if ρ > ρc , one might say that there is then sufficient matter
to ’close up the universe’. Still, the density ρ0 at present is not easy to determine. Next, from
(13.26) we can write the deceleration parameter as
1 3p  ρ
q= 1+ . (13.49)
2 ρ ρc
For dust, qd = ρ/(2ρc ), and for radiation, qr = ρ/ρc . Thus a direct measurement of q0 will
give the sign of k directly.
To determine H0 and q0 , we recall the relations (13.20) and (13.22).Then the apparent
area distance
1 1 q0 − 3 2
dA = r1 a(t1 ) = z+ z + ... (13.50)
H0 2 H0
which is the Hubble law. Here the apparent area distance dA can be introduced as follows.
The past light cone of the observer at t0 meets the surface t1 in a sphere of coordinate radius
r1 with area 4πr12 a2 (t1 ). We imagine an object, say a galaxy, covering an area A on this
sphere, and subtending a solid angle δω at the earth. In flat space we would have A = r2 δω,
so in curved space we define the apparent area distance by d2A = A/(δω). Since
δω A
= 2 2
4π 4πr1 a (t1 )
so that dA = r1 a(t1 ). The useful luminosity distance is simply dL = (1 + z)2 dA .

123
13.4 Inflation
Inflation is the central topic of the modern physical cosmology. Inflation generates the
initial irregularities of the gravitational and matter perturbations in the early universe and
amplifying them to the formation of the larger scale structure of the universe.7 Inflation
also solves the so-called flatness problem and the horizon problem.
By (cosmic) inflation we mean the acceleration condition ä > 0, as in the de Sitter
solution (13.44). This condition, when expressed in terms of the Hubble parameter H =
ȧ/a, is
ä Ḣ
= Ḣ − H 2 > 0 ⇔ − 2 < 1. (13.51)
a H
which satisfied for positive Ḣ > 0. The inflation condition immediately implies that the
Friedmann equation without cosmological constant (13.26)

ä 4πG
=− (ρ + 3p) > 0 ⇒ ρ + 3p < 0. (13.52)
a 3
Since the energy density ρ > 0, we conclude that inflation requires matter to have negative
pressure. (Cf. (6.24) where the cosmological term also contributes a negative pressure!) For
a linear equation of state p = ωρ, this means that ω < −1/3.
To understand various inflation models, it is necessary to know some (classical) scalar
field theories that satisfy ρ + 3p < 0. Consider the following action for scalar matter fields8

√ √
Z Z h 1 i
Sϕ = 4
−gd xLϕ = 4 ab
−gd x − g ∇a ϕ∇b ϕ − V (ϕ) , (13.53)
2

with a unspecified potential V (ϕ). Different inflation models are characterized by different
V (ϕ), and the ϕ is called an inflaton. The energy-momentum tensor for ϕ is

Tab = ∇a ϕ∇b ϕ + gab Lϕ (13.54)

from which one reads the energy density and pressure for ϕ(t):

1 1
ρϕ = ϕ̇2 + V (ϕ), pϕ = ϕ̇2 − V (ϕ). (13.55)
2 2

We can see that if ϕ̇2 /2 is much smaller than V (ϕ), we have ωϕ = pϕ /ρϕ ≈ −1, as desired.
From the conservation equation (13.15), and ρ̇ϕ = ϕ̈ϕ̇ + ϕ̇dV /dϕ, ρ + pϕ = ϕ̇2 , we have
 dV 
ϕ̇ ϕ̈ + 3H ϕ̇ + = 0. (13.56)
dϕ

If ϕ̇ ̸= 0, this means the Klein-Gordon equation for ϕ,

dV
ϕ̈ + 3H ϕ̇ + = 0. (13.57)
dϕ
7
The theory of cosmological perturbations necessarily involves quantum field theories on curved space-
times, which is beyond the scope of this course.
8
Do not confuse this scalar field ϕ with the angular coordinate in the rest part of these notes!

124
 q
Let us introduce the reduced Planck mass MPl = ℏc
8πG
, or simply MPl−2 = 8π in the
units G = c = ℏ = 1. This way, the cosmological field equations are
1 1 
H2 = ϕ̇ + V (ϕ) , (13.58)
3MPl2 2
ä 1  
− = ϕ̇ − V (ϕ) , (13.59)
a 3MPl2
dV
0 =ϕ̈ + 3H ϕ̇ + . (13.60)
dϕ

From (13.59) we see that for ä/a > 0, we require ϕ̇ < V (ϕ). In other words, we should
consider the case of slow motions. More formally, we can consider the slow-roll approxi-
mation
ϕ̇2 ≪ V (ϕ), ϕ̈ ≪ 1. (13.61)
In this approximation, (13.58) and (13.60) become first-order ODEs:
1
H2 ≈ V (ϕ), (13.62)
3MPl2
dV
3H ϕ̇ ≈ − . (13.63)
dϕ
If we further introduce the slow-roll parameters

MPl2  V ′ 2 V ′′
ϵ= , η = MPl2 , (13.64)
2 V V
where ()′ = d()/dϕ, then the slow-roll approximation is equivalent to saying that ϵ ≪
1, η ≪ 1.
During inflation the scale of the universe typically increases by about 1026 , so it is more
convenient to use exponents. Define the number of e-foldings by
Z tend
a(tend )
N (t) = log = H(t)dt, (13.65)
a(t) t

where tend is the time at which the inflation ends. The number of e-folding measures the
amount of inflation; to solve the flatness and horizon problem, it is required that N (t) is
about 50-70. In the slow-roll approximation, we can directly compute N (t). To this end,
we first find from (13.63) that dt = −(3H/V ′ )dϕ, so that
Z tend Z ϕend Z ϕ(t)
3H 2 1 V
N (t) = H(t)dt ≈ − ′
dϕ ≈ 2 dϕ. (13.66)
t ϕ(t) V MPl ϕend V ′

So we only need to know the potential V (ϕ) to check the inflation behavior, without solving
the field equations.
As an example, consider the massive potential, V (ϕ) = 12 m2 ϕ2 . We have V ′ = m2 ϕ and
V ′′ = m2 , so that V ′ /V = 2/ϕ and V ′′ /V = 2/ϕ2 . Namely, the slow-roll parameters are
2 2
ϵ = MPl2 , η = MPl2 . (13.67)
ϕ2 ϕ2

125
 √
By the condition for slow roll, ϵ ≪ 1, we obtain ϕ ≫ 2MPl . If we define the end of
√
inflation by ϵ = 1, we have then ϕend = 2MPl . The number of e-folding is
Z ϕi Z ϕi
1 V 1 ϕ 1  ϕ 2 1
i
N ≈ 2 dϕ = dϕ = (ϕ2
− ϕ2
end ) = − . (13.68)
MPl ϕend V ′ MPl2 ϕend 2 4MPl2 i 2MPl 2
So we need ϕi = 15MPl to obtain over 60 e-foldings. Of course, in this example the field
equations (13.62) and (13.63) can be exactly solved. Firstly, (13.62) can be solved to get
1 m
H≈√ ϕ (13.69)
6 MPl
By substituting (13.69) into (13.63), we get
r r
3 m 3
ϕϕ̇ ≈ −m2 ϕ ⇒ ϕ(t) = − M mt + C.
2 MPl 2 Pl
By setting the time when the inflation begins ti = 0, and taking into account ϕi = 15MPl
√
and ϕend = 2MPl , we determine
r
 3 
ϕ(t) = MP l 15 − mt . (13.70)
2
To obtain a, we use H = ȧ/a = d log a/dt, then from the solutions (13.69) and (13.70) we
can obtain r
m2 2 3
log a(t) = − t + 5 mt + log D.
6 2
By further setting a(ti = 0) = D, we finally determine
n m2 r
3 o
a(t) = ai exp − t2 + 5 mt . (13.71)
6 2
Another typical example is the power-law inflation with an exponential potential
 r2 ϕ 
V (ϕ) = V0 exp − , V0 , p > 0. (13.72)
p MP l
The slow-roll parameters can be readily computed to be ϵ = 1/p, η = 2/p; therefore the
slow-roll condition requires that p > 1. But now the end of inflation is not obvious to see.
As p is a constant, it is possible for the inflation to be eternal. Let us solve the field equations
(13.62) and (13.63) to complete the discussion. Firstly, (13.62) gives
r  r1 ϕ 
1 V0
H≈ exp − . (13.73)
MPl 3 2p MP l
Substituting this into (13.63) entails
p  r1 ϕ  r
2  r2 ϕ 
3V0 exp − ϕ̇ ≈ V0 exp −
2p MP l p p MP l
r s
p  1 ϕ  2V0
⇒ MPl 2p exp = (t − t0 ).
2p MP l 3p
This way, (13.73) becomes
p
H= , ⇒ a(t) = ai (t − t0 )p , (13.74)
t − t0
hence the name ‘power-law’.

126
13.5 Anisotropic cosmologies
Anisotropic cosmologies constitute a class of cosmological models that are homogeneous
but anisotropic. In particular, this allows these models to have non-zero Weyl tensors, that
may in turn change the character of the initial singularities that typically arise.
We want spacetimes that are foliated by homogeneous space-like hypersurfaces of con-
stant time. Homogeneity for the hypersurfaces means that there is a 3D group of isometries
that is transitive, so we need to know the classification of 3D Lie algebras, i.e. the Bianchi
classification [34].
Let Xi , i = 1, 2, 3 form a basis of a 3D Lie algebra with structure constant Cijk . Pick a
totally antisymmetric tensor ϵijk = ϵ[ijk] and define

tij = ϵimn Cmn

j
= nij + ϵijk ak

where nij = n(ij) . With this decomposition, we can reduce the Jacobi identity, Cijk Clk
m
+
Cjl Cik + Cli Cjk = 0, to
k m k m

nij aj = 0. (13.75)
Indeed, if we choose i = 1, j = 2, l = 3, the Jacobi identity becomes

ϵ123 t3k ϵ3kn tnm + ϵ231 t1k ϵ1kn tnm + ϵ312 t2k ϵ2kn tnm = 0

which implies that ϵijk tjk til = 0. Then, using the identities ϵijk njk = 0, ϵijk ϵjkl = 2δil , ϵijk aj ak =
0, one obtains (13.75).
Under a change of basis we have

nij → L̃ip L̃jq npq , aj → Lkj ak (13.76)

where Lji L̃kj = δik and det L = 1 to preserve ϵijk . A rescaling of ϵijk induces the following
changes
ϵijk → λϵijk , nij → λnij , aj → aj . (13.77)
Now the classification can be split into two classes: Class A with ai = 0 and Class B with
ai ̸= 0. In Class A, we may use orthogonal Lji to diagonalize nij and then diagonalize Lji to
set all the non-zero diagonal elements of nij equal in magnitude. Finally, a suitable choice
of λ will set all the non-zero diagonal elements ni , 1 = 1, 2, 3 equal to ±1. This classification
is given in Tab. 1.

Table 1: The Bianchi types for class A.

n1 n2 n3 Bianchi Type
0 0 0 I
1 0 0 II
0 1 -1 VI0
0 1 1 VII0
1 1 -1 VIII
1 1 1 IX

127
 In Class B, we may use the orthogonal Lji to set

a! 0 0 0!
ai = 0 , nij = 0 n1 0
0 0 0 n2

Now diagonal Lji will set n1 , n2 equal to ±1 or 0 and will change ai in such a way that the
quantity h = a2 /(n1 n2 ) is invariant. This classification is given in Tab. 2.

Table 2: The Bianchi types for class B.

a n1 n2 n3 Bianchi Type
1 0 0 0 V
1 0 0 1 IV
1 0 1 -1 III or VI−1
√
√−h 0 1 -1 VIh , h < 0
h 0 1 1 VIIh , h > 0

Given the classification, we seek for each Bianchi type a 3-manifold with metric and
isometry group of that type. Suppose that Xi can be realized as vector fields on a 3-manifold
M with correct commutators. At a point p ∈ M , choose another basis Yi of vectors with
corresponding dual basis ωi of 1-forms and propagate these around M by LXi ω j = 0. The
commutator of two Yi is [Yi , Yj ] = Ĉijk Yk , and we can calculate

1
Yja Ykb ∇[a ωb]i = Y[ja Yk]b ∇a ωbi = −ωbi Y[ja ∇|a| Yk]b = − ωbi Ĉjk
m b
Ym ,
2
so
1 i j
dω i = − Ĉjk ω ∧ ωk . (13.78)
2
If Yi = aji Xj , then we known from the above properties that
p q m
Xi (akj ) = Cijm akm , Ĉijk am
k = ai aj Cpq

which can be solved for Yi if the Xi are given. In particular if we choose aji = −δij at p, then
we see that Ĉijk = −Cijk , and we obtain the invariant 1-form

1 i j
dω i = Cjk ω ∧ ωk . (13.79)
2
To find suitable coordinates for M for a given type, we solve (13.79) for ω i given the Cijk for
that type; with the ω i known, the metric dσ 2 = aij ω i ω j , where aij is a symmetric matrix,
will have the those Xi as Killing vectors. The corresponding spacetime metric is then

ds2 = dt2 − aij ω i ω j (13.80)

where t measures proper time along the curves normal to the surfaces of constant t which
are homogeneous.

128
 As an example, consider the Bianchi I case. From Tab. 1, we know that the structure
constants are all zero, so the invariant 1-forms are all exact: ω 1 = dx, ω 2 = dy, ω 3 = dz.
There correspond three Killing vectors

∂ ∂ ∂
X1 = − , X2 = − , X3 = − (13.81)
∂x ∂y ∂z

meaning three translational symmetries. We may diagonalize the aij , so that the spacetime
metric can be written as

ds2 = dt2 − X 2 (t)dx2 − Y 2 (t)dy 2 − Z 2 (t)dz 2 . (13.82)

Using the Cartan method, we read off the basis

θ0 = dt, θ1 = Xdx, θ2 = Y dy, θ3 = Zdz,

and we find the Ricci tensor components

R00 = θ̇ + A2 + B 2 + C 2 , R11 = −Ȧ − θA, R22 = −Ḃ − θB, R33 = −Ċ − θC

where A = Ẋ/X, B = Ẏ /Y, C = Ż/Z and θ = A + B + C.

We shall seek perfect fluid solutions with the fluid flow lines parallel to θ0 , i.e. orthogonal
to the surfaces of homogeneity. This means the following Einstein’s equations
1 1
R00 = − κ(ρ + 3p), R11 = R22 = R33 = − κ(ρ − p) (13.83)
2 2
Let us look first for vacuum solutions; the vacuum Einstein’s equations give

2R00 − R = −2(AB + BC + CA) = 0 (13.84)

so that θ2 = A2 + B 2 + C 2 . Then R00 = θ̇ + θ2 = 0 further entails that θ = (t − t0 )−1 or

simply θ = 1/t. Then from R11 = 0 we find A = p/t, and similarly B = q/t, C = r/t. In
summary, we obtain the constraints

p + q + r = 1, p2 + q 2 + r2 = 1 (13.85)

and from A = Ẋ/X = p/t we find X = X0 tp . Thus, the metric is

ds2 = dt2 − t2p dx2 − t2q dy 2 − t2r dz 2 (13.86)

which is known as the Kasner metric.

Consider next the dust (p = 0) cosmology of the Kasner metric (13.86). In this case,
Einstein’s equations entail
3
R11 + R22 + R33 = −θ̇ − θ2 = − κρ. (13.87)
2
By use of the conservation equation ρ̇ + θ(ρ + p) = 0, we get
3 3
(θ̇ + θ2 )• = κρ̇ = − κθρ = θ(θ̇ + θ2 ).
2 2

129
 ...
By setting θ = v̇/v, then we have ( v̈v )• = − v̇v̈
v2
, that is, v = 0, so v = αt2 + 2βt + γ.
Consequently,
2(αt + β) 4α
θ= 2 , κρ = .
αt + 2βt + γ 3v
Suppose α = 1, β = 0, then from R11 = 0 we have (vA)• = 2/3, so that
2 t + p0 2 t + q0 2 t + r0
A= , B= , C= .
3 t2 + γ 3 t2 + γ 3 t2 + γ
We see that A + B + C = θ and AB + BC + CA = κρ entail the constraints

p0 + q0 + r0 = 0, p0 q0 + q0 r0 + r0 p0 = 3γ.

Finally, we can solve for the metric components, e.g. X = (t − k)p (t + k)2/3−p with p =
(1+p0 /k)/3, and similarly for Y, Z. This gives the dust cosmology counterpart to the Kasner
metric. There will be a singularity of the metric at t = k. Near the singularity, the dust
cosmology approaches the Kasner metric, so there should be no curvature, and hence the
singularity is in the Weyl tensor!
The simplest case with nonvanishing structure constant is the Bianchi II cosmology. In
this case, the Lie algebra can be chosen as

[X2 , X3 ] = X1 , [X1 , X2 ] = [X3 , X1 ] = 0, (13.88)

and one can check that the following Killing vectors

∂ ∂ ∂ ∂
X1 = , X2 = −x , X3 = (13.89)
∂z ∂y ∂z ∂x

satisfy the commutation relations in (13.88). If we choose the basis as

∂ ∂ ∂ ∂
e1 = , e2 = , e3 = −y , (13.90)
∂z ∂y ∂x ∂z
then the dual basis is
ω 1 = dz + ydx, ω 2 = dy, ω 3 = dx. (13.91)
Correspondingly, the metric of Bianchi II cosmology can be written as

ds2 = dt2 − a2 (t)(ω 1 )2 − b2 (t)(ω 2 )2 − c2 (t)(ω 3 )2 . (13.92)

If we change the time parameter by dt = abcdτ , we can introduce the following parametriza-
tion
a = eα(τ ) , b = eβ(τ ) , c = eγ(τ ) . (13.93)
Using this parametrization (13.93), the spatial components of the vacuum Einstein’s equa-
tions are
1 1 1
α̈ = − e4α , β̈ = e4α , γ̈ = e4α , (13.94)
2 2 2
and the temporal component is

α̈ + β̈ + γ̈ = 2(α̇β̇ + α̇γ̇ + β̇ γ̇) (13.95)

130
where the dot means derivative wrt τ . The first equation in (13.94) can be integrated to

1
α̇2 = H 2 − e4α , (H : a const.)
4
which gives
1
α(τ ) = ∓Hτ − ln(e∓4Hτ + 1). (13.96)
2
On the other hand, from the first two equations in (13.94), we have α̈+ β̈ = 0, or α̇+ β̇ = B,
for some constant B. Therefore, we can solve it for

β(τ ) = Bτ + B0 − α(τ ), (13.97)

and similarly,
γ(τ ) = Cτ + C0 − α(τ ). (13.98)
Finally, substituting these α, β, γ into (13.95) entails the relation, BC = H 2 , between con-
stants.
If we choose the + sign in the solution (13.96), then in the limit τ → −∞ the solutions
becomes
α(τ ) = Hτ, β(τ ) = (B − H)τ, γ(τ ) = (C − H)τ, (13.99)
and hence the spatial volume of the universe behaves as

abc = eα+β+γ = e(B+C−H)τ . (13.100)

In the opposite limit τ → ∞, we have

α(τ ) = −Hτ, β(τ ) = (B + H)τ, γ(τ ) = (C + H)τ, (13.101)

and
abc = eα+β+γ = e(B+C+H)τ . (13.102)
When B, C, H > 0, we see that the limit τ → −∞ is closer to the singularity than τ → 0,
as the spatial volume is smaller. We also see that these two limits both tends to the Kasner
solution. The transition between these two limits is known as the Kasner transition.
Let us introduce the Lifshitz-Khalatnikov parameter u = H/B, so that in the limit
τ → ∞ the asymptotic Kasner behavior is

a → tp1 , b → tp2 , c → tp3 (13.103)

where the Kasner indices are now given by

u 1+u u(1 + u)
p1 = − , p2 = , p3 = . (13.104)
1 + u + u2 1 + u + u2 1 + u + u2
In the limit τ → −∞, we have instead
′ ′ ′
a → tp1 , b → tp2 , c → tp3 (13.105)

131
with
u 1−u u(u − 1)
p′1 = = p2 (u−1), p′2 = = p1 (u−1), p′3 = = p3 (u−1).
1 − u + u2 1 − u + u2 1 − u + u2
(13.106)
Therefore, the Kasner transition from (τ → ∞) to (τ → −∞) is effectuated by changing
the scale factor a ↔ b and shifting the Lifshitz-Khalatnikov parameter by

u → u − 1. (13.107)

Another important case is the Bianchi IX cosmology. From Tab. 1, we know that in
this case the Lie algebra should be an su(2) Lie algebra

[X2 , X3 ] = X1 , [X1 , X2 ] = X3 , [X3 , X1 ] = X2 . (13.108)

Without going into too much details, we remark that the solution of Bianchi IX cosmol-
ogy is kind of (triple) generalization of the Bianchi II case, with the scale factors a, b, c all
becoming important. The result on Kasner transition is that, after a single Kasner transi-
tion, one of the scale factors becomes small, but others are still large. In other words, there
sill be multiple Kasner transitions, each of which is characterized by the shift (13.107). If
we start with u > 1, then after an integer number of Kasner transitions, we will reach the
regime in which u < 1. But the surprising thing is that the Kasner indices have the prop-
erty pi (u) = pi ( u1 ), whereby we have that after u < 1 it is true that u1 > 1. Namely, a new
series of Kasner transitions begin! This oscillatory behavior of Kasner transitions near the
singularity is now known as the Belinski-Khalatnikov-Lifshitz transition.
A famous Bianchi IX cosmological model is the Mixmaster universe proposed by Misner.
The Hamiltonian for the Mixmaster universe contains a potential having the interpretation
of a ball moving in a billiard with moving walls, which is a typical model of classical chaos.
This interpretation then naturally introduces chaos into the BKL transition, a surprise from
the anisotropic cosmology!
Each of the FRW cosmologies is a particular case of a spatially homogeneous, anisotropic
cosmology, and one question that must be answered is why the actual universe is so much
less anisotropic than the equations in principle allow it to be. One might speculate that
the ‘initial conditions’ on the structure of the universe in some way entail isotropy—or,
alternatively, a vanishing Weyl tensor, i.e. the Weyl curvature hypothesis.
When there is a positive cosmological constant Λ, this problem is partly understood due
to the following strong result (also known as the cosmic no-hair theorem):

Theorem 13.1 (Wald [35]). An initially expanding Bianchi cosmological model, except for type
IX, with a cosmological constant Λ > 0, non-positive spatial curvature (3) R ⩽ 0, and matter
satisfying the dominant and strong energy conditions, must continue to expand to a de Sitter spacetime
asymptotically.

Proof. Let Σ be the homogeneous spatial hypersurface projected from a Bianchi model by
hab = gab + na nb . Then the extrinsic curvature Kab of Σ can be decomposed into its trace
K and trace-free part as
1
Kab = Khab + σab (13.109)
3

132
where σab is the shear of time-like geodesics, as can be easily checked by the definitions. To
proceed, recall Einstein’s equations (6.19) and (6.23), and we rewrite them as

Gab na nb − Λ − 8πGTab na nb =0, (13.110)

1
Rab na nb + Λ − 8πG(Tab − gab T )na nb =0. (13.111)
2
Combining (13.109)-(13.111) yields
3 3
K 2 =3Λ + σab σ ab − (3) R + 24πGTab na nb , (13.112)
2 2
1 2 1
K̇ =Λ − K − σab σ ab − 8πG(Tab − gab T )na nb , (13.113)
3 2
where the dot means the derivative wrt proper time. For Bianchi models, the spatial spin
connections are given by the structure constants of the 3D Lie algebra 13.79. Therefore,
we have
(3) 1 1
R = − C aab C c c b + C abc C c a b − Cabc C abc =
2 4
3 j 1 2
−1 ij
= − aj a − h (nij n − n ) (13.114)
2 2
where we have used (13.75) and h = det|hab | which can be calculated with the help of ϵijk
using (2.5). From (13.114) we see that (3) R > 0 if nij nij < 12 n2 which indicates that nij ̸= 0.
Then, by (13.75) we have aj = 0, i.e. Bianchi type IX. Namely, we have
(3)
R⩽0 (13.115)

for other Bainchi models than type IX. For these models with (13.115), if we use the dom-
inant and strong energy conditions in (13.112) and (13.113), we obtain
1
K̇ ⩽ Λ − K 2 ⩽ 0. (13.116)
3
Moreover, (13.112) implies that
K 2 ⩾ 3Λ. (13.117)
So if initially K > 0, then we always have K > 0 for all time. (13.116) can be rewritten as
1 dK 1
⩽−
K2 − 3Λ dτ 3
which can be integrated to yield an upper bound
(3Λ)1/2
K⩽ , α = (3/Λ)1/2 . (13.118)
tanh(τ /α)
From (13.117) and (13.118), we see that as time τ increases K will tend to (3/Λ)1/2 . Com-
bining (13.112) and (13.118) we also have
2 2Λ
σab σ ab ⩽ (K 2 − 3Λ) ⩽ , (13.119)
3 sinh2 (τ /α)
Λ 1
Tab na nb ⩽ 2 . (13.120)
8π sinh (τ /α)

133
Therefore, as time increases both the shear and Tab na nb tend to zero. The above large time
(τ ≫ α) behaviors imply that the spatical metric should take the de Sitter form

hab (τ ) = e2(τ −τ0 )/α hab (τ0 ). (13.121)

This completes the proof.

This theorem is a good award for our journey! In the future, may you ride the winds of
GR!

A More Newman-Penrose
In this appendix, we recollect more results in the Newman-Penrose formalism. The Einstein
equations are taken from [3], and the asymptotic properties follows [36].

A.1 Einstein equations in Newman-Penrose formalism

From the null tetrad we can define the complex Ricci rotation coefficients

γ mnp = ∇b Zam Z na Z pb (A.1)

which leads to the twelve spin coefficients

κ =γ 131 = ∇b la ma lb (A.2)
ρ =γ 134 = ∇b la ma m̄b (A.3)
σ =γ 133 = ∇b la ma mb (A.4)
τ =γ 132
= ∇b la m n a b
(A.5)
ν =−γ 242
= −∇b na m̄ n a b
(A.6)
µ = − γ 243 = −∇b na m̄a mb (A.7)
π = − γ 241 = −∇b na m̄a lb (A.8)
1 1
α = (γ 124 − γ 344 ) = (∇b la na m̄b − ∇b ma m̄a m̄b ) (A.9)
2 2
1 123 1
β = (γ − γ 343 ) = (∇b la na mb − ∇b ma m̄a mb ) (A.10)
2 2
1 122 1
γ = (γ − γ 342 ) = (∇b la na nb − ∇b ma m̄a nb ) (A.11)
2 2
1 121 1
ϵ = (γ − γ 341 ) = (∇b la na lb − ∇b ma m̄a lb ). (A.12)
2 2
Since the spacetime is assumed to be empty in a neighborhood of the null infinity, the
gravitational field is given by the Weyl tensor there. The five independent complex tetrad

134
components of the Weyl tensor are

Ψ0 = − Cabcd la mb lc md (A.13)
Ψ1 = − Cabcd la nb lc md (A.14)
1
Ψ2 = − Cabcd (la na lc nd − la nb mc m̄d ) (A.15)
2
Ψ3 = − Cabcd m̄a nb lc nd (A.16)
a b
Ψ4 = − Cabcd m̄ n m̄ n . c d
(A.17)

The six independent, three real and three complex, tetrad components of the traceless Ricci
tensor Sab = Rab − 14 gab R are

1
Φ00 = Sab la lb = Φ̄00 (A.18)
2
1
Φ01 = Sab la mb = Φ̄10 (A.19)
2
1
Φ02 = Sab ma mb = Φ̄20 (A.20)
2
1
Φ11 = Sab (la nb + ma m̄a ) = Φ̄11 (A.21)
4
1
Φ12 = Sab na mb = Φ̄21 (A.22)
2
1
Φ22 = Sab na nb = Φ̄22 . (A.23)
2
Finally, the intrinsic derivatives acting on a scalar can be defined as

Dϕ =∇a ϕla = la ∂a ϕ (A.24)

∇ϕ =∇a ϕna = na ∂a ϕ (A.25)
δϕ =∇a ϕma = ma ∂a ϕ (A.26)
a
δ̄ϕ =∇a ϕm̄ = m̄ ∂a ϕ. a
(A.27)

With these definitions, Einstein’s equations can be recast into the following sets of equa-
tions. Let us present these tedious equations here, merely for the purpose of exhibition. The
first set contains the commutator equations for the commutator of two intrinsic derivatives
acting on a scalar ϕ:

(A.28)
 
(∆D − D∆)ϕ = (γ + γ̄)D + (ϵ + ϵ̄)∆ − (τ + π̄)δ̄ − (τ̄ + π)δ ϕ
(A.29)
 
(δD − Dδ)ϕ = (ᾱ + β − π̄)D + κ∆ − σ δ̄ − (ρ̄ + ϵ − ϵ̄)δ ϕ
(A.30)
 
(δ∆ − ∆δ)ϕ = −ν̄D + (τ − ᾱ − β)∆ + λ̄δ̄ + (µ − γ + γ̄)δ ϕ
(A.31)
 
(δ̄δ − δ δ̄)ϕ = (µ̄ − µ)D + (ρ̄ − ρ)∆ − (ᾱ − β)δ̄ − (β̄ − α)δ ϕ.

135
The second set contains the spin-coefficient equations:

Dρ − δ̄κ =(ρ2 + σσ̄) + (ϵ + ϵ̄)ρ − κ̄τ − κ(3α + β̄ − π) + Φ00 (A.32)

Dσ − δκ =(ρ + ρ̄)σ + (3ϵ − ϵ̄)σ − (τ − π̄ + ᾱ + 3β)κ + Ψ0 (A.33)
Dτ − ∆κ =(τ + π̄)ρ + (τ̄ + π)σ + (ϵ − ϵ̄)τ − (3γ + γ̄)κ + Ψ1 + Φ01 (A.34)
Dα − δ̄ϵ =(ρ + ϵ̄ − 2ϵ)α + β σ̄ − β̄ϵ − κλ − κ̄γ + (ϵ + ρ)π + Φ10 (A.35)
Dβ − δϵ =(α + π)σ + (ρ̄ − ϵ̄)β − (µ + γ)κ − (ᾱ − π̄)ϵ + Ψ1 (A.36)
Dγ − ∆ϵ =(τ + π̄)α + (τ̄ + π)β − (ϵ + ϵ̄)γ − (γ + γ̄)ϵ + τ π − νκ + Ψ2 − Λ + Φ11
(A.37)
Dλ − δ̄π =(ρλ + σ̄µ) + π 2 + (α − β̄)π − ν κ̄ − (3ϵ − ϵ̄)λ + Φ20 (A.38)
Dµ − δπ =(ρ̄µ + σλ) + ππ̄ − (ϵ + ϵ̄)µ − π(ᾱ − β) − νκ + Ψ2 + Λ (A.39)
Dν − ∆π =(π + τ̄ )µ + (π̄ + τ )λ + (γ − γ̄)π − (3ϵ + ϵ̄)ν + Ψ3 + Ψ21 (A.40)
∆λ − δ̄ν = − (µ + µ̄)λ − (3γ − γ̄)λ + (3α + β̄ + π − τ̄ )ν − Ψ4 (A.41)
δρ − δ̄σ =ρ(ᾱ + β) − σ(3α − β̄) + (ρ − ρ̄)τ + (µ − µ̄)κ − Ψ1 + Φ01 (A.42)
δα − δ̄β =(µρ − λσ) + αᾱ + β β̄ − 2αβ + γ(ρ − ρ̄) + ϵ(µ − µ̄) − Ψ2 + Λ + Φ11 (A.43)
δλ − δ̄µ =(ρ − ρ̄)ν + (µ − µ̄)π + µ(α + β̄) + λ(ᾱ − 3β) − Ψ3 + Φ21 (A.44)
δν − ∆µ =(µ2 + λλ̄) + (γ + γ̄)µ − ν̄π + (τ − 3β − ᾱ)ν + Φ22 (A.45)
δγ − ∆β =(τ − ᾱ − β)γ + µτ − σν − ϵν̄ − β(γ − γ̄ − µ) + αλ̄ + Φ12 (A.46)
δτ − ∆σ =(µσ + λ̄ρ) + (τ + β − ᾱ)τ − (3γ − γ̄)σ − κν̄ + Φ02 (A.47)
∆ρ − δ̄τ = − (ρµ̄ + σλ) + (β̄ − α − τ̄ )τ + (γ + γ̄)ρ + νκ − Ψ2 − 2Λ (A.48)
∆α − δ̄γ =(ρ + ϵ)ν − (τ + β)λ + (γ̄ − µ̄)α + (β̄ − τ̄ )γ − Ψ3 (A.49)

where the Λ is defined by the Ricci scalar R = −24Λ. The third set contains the Bianchi

136
identities of spin coefficients
δ̄Ψ0 −DΨ1 + DΦ01 − δΦ00 = (4α − π)Ψ0 − 2(2ρ + ϵ)Ψ1 + 3κΨ2 + (π̄ − 2ᾱ − 2β)Φ00 +
+ 2(ϵ + ρ̄)Φ01 + 2σΦ10 − 2κΦ11 − κ̄Φ02 , (A.50)
∆Ψ0 −δΨ1 + DΦ02 − δΦ01 = (4γ − µ)Ψ0 − 2(2τ + β)Ψ1 + 3σΨ2 − λ̄Φ00 + 2(π̄ − β)Φ01 +
+ 2σΦ11 + (2ϵ − 2ϵ̄ + ρ̄)Φ02 − 2κΦ12 , (A.51)
3(δ̄Ψ1 −DΨ2 ) + 2(DΦ11 − δΦ10 ) + δ̄Φ01 − ∆Φ00 = 3λΨ0 − 9ρΨ2 + 6(α − π)Ψ1 + 6κΨ3 +
+ (µ̄ − 2µ − 2γ − 2γ̄)Φ00 + (2α + 2π + 2τ̄ )Φ01 + 2(τ − 2ᾱ + π̄)Φ10 +
+ 2(2ρ̄ − ρ)Φ11 + 2σΦ20 − σ̄Φ02 − 2κ̄Φ12 − 2κΦ21 , (A.52)
3(∆Ψ1 −δΨ2 ) + 2(DΦ12 − δΦ11 ) + δ̄Φ02 − ∆Φ01 = 3νΨ0 − 9τ Ψ2 + 6(γ − µ)Ψ1 + 6σΨ3 −
− ν̄Φ00 + (2µ̄ − 2µ − 2γ)Φ01 − 2λ̄Φ10 + 2(τ + 2π̄)Φ11 + (2α + 2π + τ̄ − 2β̄)Φ02 +
+ (2ρ̄ − 2ρ − 4ϵ̄)Φ12 + 2σΦ21 − 2κΦ22 , (A.53)
3(δ̄Ψ2 −DΨ3 ) + DΦ21 − δΦ20 + 2(δ̄Φ11 − ∆Φ10 ) = 6λΨ1 − 9πΨ2 + 6(ϵ − ρ)Ψ3 + 3κΨ4 −
− 2νΦ00 + 2λΦ01 + 2(µ̄ − µ − 2γ̄)Φ10 + 2(π + 2τ̄ )Φ11 + (2β + 2τ + π̄ − 2ᾱ)Φ20 +
+ (2ρ̄ − 2ρ − 2ϵ̄)Φ21 − 2σ̄Φ12 − κ̄Φ22 , (A.54)
3(∆Ψ2 −δΨ3 ) + DΦ22 − δΦ21 + 2(δ̄Φ12 − ∆Φ11 ) = 6νΨ1 − 9µΨ2 + 6(β − τ )Ψ3 + 3σΨ4 −
− 2νΦ01 − 2ν̄Φ10 + 2(2µ̄ − µ)Φ11 + 2λΦ02 − λ̄Φ20 + 2(π + τ̄ − 2β̄)Φ12 +
+ (2β + 2τ + 2π̄)Φ21 + (ρ̄ − 2ϵ − 2ϵ̄ − 2ρ)Φ22 , (A.55)
3δ̄Ψ3 −DΨ4 + δ̄Φ21 − ∆Φ20 = 3λΨ2 − 2(α + 2π)Ψ3 + (4ϵ − ρ)Ψ4 − 2νΦ10 + 2λΦ11 +
+ (2γ − 2γ̄ + µ̄)Φ20 + (2τ̄ − 2α)Φ21 − σ̄Φ22 , (A.56)
3∆Ψ3 −δΨ4 + δ̄Φ22 − ∆Φ21 = 3νΨ2 − 2(γ + 2µ)Ψ3 + (4β − τ )Ψ4 − 2νΦ11 + 2λΦ12 −
− ν̄Φ20 + (2µ̄ + 2γ)Φ21 + (τ̄ − 2β̄ − 2α)Φ22 , (A.57)
DΦ11 −δΦ10 − δ̄Φ01 + ∆Φ00 + 3DΛ = (2γ − µ + 2γ̄ − µ̄)Φ00 + (π − 2α − 2τ̄ )Φ01 +
+ (π̄ − 2ᾱ − 2τ )Φ10 + 2(ρ + ρ̄)Φ11 + σ̄Φ02 + σΦ20 − κ̄Φ12 − κΦ21 , (A.58)
DΦ12 −δΦ11 − δ̄Φ02 + ∆Φ01 + 3δΛ = (2γ − µ − 2µ̄)Φ01 + ν̄Φ00 − λ̄Φ10 + (2π̄ − τ )Φ11 +
+ (π + 2β̄ − 2α − τ̄ )Φ02 + (2ρ + ρ̄ − 2ϵ̄)Φ12 + σΦ12 − κΦ22 , (A.59)
DΦ22 −δΦ21 − δ̄Φ12 + ∆Φ11 + 3∆Λ = νΦ01 + ν̄Φ10 − 2(µ + µ̄)Φ11 − λΦ02 − λ̄Φ20 +
+ (2π − τ̄ + 2β̄)Φ12 + (2β − τ + 2π̄)Φ21 + (ρ + ρ̄ − 2ϵ − 2ϵ̄)Φ22 . (A.60)
When an electromagnetic field with the Maxwell tensor Fab = ∂a Ab − ∂b Aa is present,
we have also three complex Maxwell scalars
1
ϕ0 = Fab la mb , ϕ1 = Fab (la nb + ma m̄b ), ϕ2 = Fab na m̄b . (A.61)
2

A.2 Asymptotic behavior and Bondi coordinates

The null infinity properties can be studied by introducing a new coordinate system (u, r, ζ, ζ̄)
in the neighborhood of I + . In this system, the time u represents a family of null surfaces, r
is the affine parameter along the null geodesics of the constant-u surfaces, and ζ = eiϕ cot 2θ
is the complex stereographic angle labeling the null geodesics of I + . We reach I + by tak-
ing r → ∞, thus the null infinity I + is described by the remaining coordinates (u, ζ, ζ̄).

137
Now, the two-surface metric becomes,

4r2 dζdζ̄
ds2 = − , (A.62)
P2
making the usual choice of Ω = r−1 as the conformal factor, eq. (A.62) gives the induced
metric on I + ,
4dζdζ̄
dŝ2 = − . (A.63)
P2
Here, P (u, ζ, ζ̄) is a strictly positive function related to the conformal factor Ω. If we restrict
the conformal transformation by imposing the condition

P = P0 = 1 + ζ ζ̄. (A.64)

With this choice, the two-surface metric becomes a sphere. These coordinates are then
called Bondi coordinates.
Note that the definition of null tetrad given in §7.4 has certain freedoms such as

1. Null rotations about la :

la → la ,
ma → ma + Bla ,
na → na + B̄ma + B m̄a + B B̄la . (A.65)

2. Null rotations about na :

la → la + B̄ma + B m̄a + B B̄na ,

ma → ma + Bna ,
na → na . (A.66)

3. Rotate ma while keeping la and na fixed:

ma → eiλ ma . (A.67)

The last freedom introduces the notion of spin weight, a quantity η that under (A.67) trans-
forms as,
η → eisλ η, (A.68)
is said to have spin weight s. One can also define the spin weighted differential operators ð
and ð̄ as,
∂(P s f ) ∂(P −s f )
ðf = P 1−s , ð̄f = P 1+s , (A.69)
∂ζ ∂ ζ̄
where P is the function in (A.63). The operators ð and ð̄ raise and lower the spin weight
by one respectively. Since the choice of the coordinate system is not unique, we can make
a different choice from the original u =const. cut, i.e.

u → û = T (u, ζ, ζ̄). (A.70)

138
Under this choice, the null tetrad system (la , na , ma ) transforms as
 
ðT ð̄T ðT ð̄T
a
l → Ṫ la + m̄a + ma + na , (A.71)
Ṫ r Ṫ r Ṫ 2 r2
ðT a
ma → ma + n , (A.72)
Ṫ r
na → Ṫ −1 na . (A.73)

By Sachs’ “peeling” assumption, one can obtain the asymptotic behavior of the Weyl
and Maxwell scalars, and the spin coefficients for any asymptotically flat spacetime:

Ψ0 = ψ00 r−5 + O(r−6 ), Ψ1 = ψ10 r−4 + O(r−5 ),

Ψ2 = ψ20 r−3 + O(r−4 ), Ψ3 = ψ30 r−2 + O(r−3 ),
Ψ4 = ψ40 r−1 + O(r−2 ), ϕ0 = ϕ00 r−3 + O(r−4 ), (A.74)
ϕ1 = ϕ01 r−2 + O(r−3 ), ϕ2 = ϕ02 r−1 + O(r−2 ).

The asymptotic spin coefficients are given by

κ = π = ε = 0, ρ = ρ̄, τ = ᾱ + β, ρ = −r−1 − σ 0 σ̄ 0 r−3 + O(r−5 ),

σ = σ 0 r−2 + [(σ 0 )2 σ̄ 0 − ψ00 /2]r−4 + O(r−5 ), α = α0 r−1 + O(r−2 ),
β = β 0 r−1 + O(r−2 ), γ = γ 0 − ψ20 (2r2 )−1 + O(r−3 ), µ = µ0 r −1 + O(r−2 ),
λ = λ0 r−1 + O(r−2 ), ν = ν 0 + O(r−1 ), (A.75)

where
1 Ṗ
α0 = −β̄ 0 = − ð̄ ln P, γ0 = − ,
2 2P
Ṗ
ν 0 = −2ð̄γ 0 , λ0 = σ̄˙ 0 − σ̄ 0 , µ0 = U 0 = −ðð̄ ln P. (A.76)
P
These components satisfy
 Ṗ  Ṗ
ψ20 − ψ̄20 = ð̄2 σ 0 − ð2 σ̄ 0 + σ̄ 0 λ0 − σ 0 λ̄0 , ψ30 = ð̄ðð̄ ln P + ðλ0 , ψ40 = −ð̄2 − λ̇0 + λ0 .
P P
(A.77)
Finally, the Bianchi identities become,

Ṗ
ψ̇20 − 3ψ20 = −ðψ30 + σ 0 ψ40 + 2ϕ02 ϕ̄02 ,
P
Ṗ
ψ̇10 − 3ψ10 = −ðψ20 + 2σ 0 ψ30 + 4ϕ01 ϕ̄02
P
Ṗ
ψ̇00 − 3ψ00 = −ðψ10 + 3σ 0 ψ20 + 6ϕ02 ϕ̄02 ,
P
Ṗ
ϕ̇01 − 2ϕ01 = −ðϕ02 ,
P
Ṗ
ϕ̇00 − 2ϕ00 = −ðϕ01 + σ 0 ϕ02 , (A.78)
P
where the dots represent ∂u .

139
A.3 BMS group
The choice of the Bondi coordinate system is still not unique, and the coordinate transfor-
mations between two Bondi systems is called the Bondi-Metzner-Sachs (BMS) transforma-
tions.
Now, since Ṗ = 0 in the Bondi coordinates, (A.77) become

ψ20 − ψ̄20 = ð̄2 σ 0 − ð2 σ̄ 0 + σ̄ 0 σ̇ 0 − σ 0 σ̄˙ 0 ,

ψ30 = ðσ̄˙ 0 ,
¨0.
ψ40 = −σ̄ (A.79)

In many applications, it is convenient to define the so-called Mass Aspect Ψ from the Weyl
scalar ψ20 ,
Ψ = ψ20 + ð2 σ̄ 0 + σ 0 σ̄˙ 0 , (A.80)
which satisfies the reality condition Ψ = Ψ̄. Finally the Bianchi identities in the Bondi
system take the form,

ψ̇20 = −ðψ30 + σ 0 ψ40 + 2ϕ02 ϕ̄02 , (A.81)

ψ̇10 = −ðψ20 + 2σ 0 ψ30 + 4ϕ01 ϕ̄02 , (A.82)
ψ̇00 = −ðψ10 + 3σ 0 ψ20 + 6ϕ02 ϕ̄02 , (A.83)
ϕ̇01 = −ðϕ02 , (A.84)
ϕ̇00 = −ðϕ01 + σ 0 ϕ02 . (A.85)

Also, it is possible to write eq. (A.81) in terms of Ψ as follows,

Ψ̇ = σ̇ 0 σ̄˙ 0 + 2ϕ02 ϕ̄02 . (A.86)

In a Bondi system, one can define the 4-momentum vector for an asymptotically flat
spacetime as, Z
1
a
P =− √ Ψ˜la dΩ, (A.87)
8π 2 S
where
˜la = 1 (1 + ζ ζ̄, ζ + ζ̄, −i(ζ − ζ̄), 1 − ζ ζ̄), (A.88)
1 + ζ ζ̄
and dΩ = 4idζ∧d
P02
ζ̄
is the area of the unit sphere. Immediately from this definition it follows
that the Bondi mass can be written as
c2
Z
MBondi = − √ ΨdΩ. (A.89)
8π 2G S
Now, taking the time derivative of (A.89), and using eq. (A.86), we have

c2
Z
Ṁ = − √ (σ̇ 0 σ̄˙ 0 + 2ϕ02 ϕ̄02 )dΩ; (A.90)
8π 2G S
since the integral is always positive, the RHS of the last equation is negative, i.e.

Ṁ < 0 if σ̇ 0 ̸= 0. (A.91)

140
Thus Ṁ measures the amount of mass loss carried away as gravitational radiation!
The set of coordinate transformations at I + preserving the conditions in the Bondi
coordinates is called the BMS Group. This group is the same as the asymptotic symmetry
group that arises from the infinitesimal generators, i.e. from the asymptotic Killing vectors.
To construct the BMS group, we start by considering the following mapping

û = T (u, ζ, ζ̄), (A.92)

r̂ = Ṫ r,−1
(A.93)
aζ + b
ζ̂ = (ad − bc = 1), (A.94)
cζ + d
where u, ζ, and r are the standard Bondi-type coordinates. The fraction linear transforma-
tions (A.94) are the only one-to-one mapping of the sphere to itself. Then, the BMS group
is defined by (A.94) and the following restriction on the transformations (A.92),

û = K(u + α), (A.95)

where α(ζ, ζ̄) is a regular function on the sphere. From eq. (A.95) and (A.94) one can show
that that the spherical metric transforms as,
¯
4dζ̂dζ̂ 4dζdζ̄
= K2 , (A.96)
P̂02 P02

where the conformal factor K associated with this transformation is given by

1 P0 ¯
K = J−2 = (1 + ζ ζ̄)[(aζ + b)(ā + ζ̄ + b̄) + (cζ + d)(c̄ζ̄ + d)],
P̂0

with J = ∂ζ ∂ ζ̄
∂ ζ̂ ∂ ζ̂¯
and
¯
P0 = 1 + ζ ζ̄, P̂0 = 1 + ζ̂ ζ̂.
The subgroup of the BMS group obtained by setting

K = 1, ζ̂ = ζ, , û = u + α, (A.97)

is known as the supertranslation subgroup. A supertranslation α(ζ, ζ̄) moves points on each
generator by an amount α(ζ, ζ̄), which can be expressed in terms of infinite constants using
a tensorial spin-s harmonic expansion,

α = α0 + αi Y1i0 (ζ, ζ̄) + αij Y2ij

0
(ζ, ζ̄) + .... (A.98)

In this expansion, α0 , and αi represents the ordinary translations.

Given the BMS transformed set of coordinates (û, r̂, ζ, ζ̄), one can build a new null tetrad
system (ˆla , n̂a , m̂a , m̂
¯ a ), by using (A.71)-(A.73),
 
L
ˆl = Ṫ la − m̄a − ma +
a L̄ L L̄
na ,
r r r2
n̂a = Ṫ −1 na ,
L
m̂a = ma − na , (A.99)
r

141
where
ð(u) T
L=− , (A.100)
Ṫ
with ð(u) standing for applying the ð operator keeping u as a constant.
At this point, we can explain the physical interpretation of some of the complex scalars
by the dominant terms in a tensorial spin-s harmonic expansion of these functions. That is,

ψ10 = ψ10i (u)Y1i1 (ζ, ζ̄) + ψ10ij (u)Y2ij

1
(ζ, ζ̄),
ψ20 = ψ200 + ψ20i Y1i0 (ζ, ζ̄) + ψ20ij (u)Y2ij
0
(ζ, ζ̄),
σ 0 = σ ij (u)Y2ij
2
(ζ, ζ̄),
ϕ00 = ϕ0i 1
0 (u)Y1i (ζ, ζ̄),
ϕ01 = ϕ00 0i 0
1 + ϕ1 (u)Y1i (ζ, ζ̄). (A.101)

Assuming that the axis of symmetry is the z-axis, we can write the non-zero component of
the angular momentum as
√
6 2G
Im[ψ10 − σ 0 ðσ̄ 0 ]z = − 3 J z , (A.102)
c
where J z is the Komar angular momentum. Since the real contribution of [σ 0 ðσ̄ 0 ]i is zero
for any axisymmetric spacetime, one only needs to use the real part of ψ10i to define the
dipole mass moment, √
6 2G z
Re[ψ1 ] = − 3 D .
0 z
(A.103)
c
On the other hand, for a Bondi system, we have defined a precise notion of mass and
linear momentum. So, we can use the ℓ = 0, 1 components of the tensorial expansion, and
relate ψ20 to the Bondi 4-momentum (M, P i ), by
√
2 2G
(A.104)
 0 2 0 0˙0

ψ2 + ð σ̄ + σ σ̄ |ℓ=0 = − 2 M,
c
i 6G
(A.105)
 0
ψ2 + ð2 σ̄ 0 + σ 0 σ̄˙ 0 = − 3 P i.
c
The ℓ ≥ 2 terms of the l.h.s of eqs. (A.104) or (A.105), are the so called supermomentum at
null infinity.
The scalar Ψ4 measures the gravitational radiation received at null infinity. This scalar
is related to the gravitational wave modes as follows,

Ψ4 = ḧ+ − iḧ× , (A.106)

where h+ , h× are the two modes of the gravitational wave in the TT gauge. Thus, the
complex function σ̈ 0 , introduced in eq. (A.79), yields the gravitational radiation reaching
at null infinity.

142
B Teleparallel equivalent of general relativity
The teleparallel equivalent of general relativity (TEGR) is an alternative geometrical for-
mulation of Einstein’s GR. In this appendix, we follow the presentation of [37, 38].
Teleparallelism means distant parallelism: In a spacetime endowed with a set of tetrad
fields, two vectors at distant points are called parallel if they have identical tetrad components
at the two points considered.

B.1 Tetrads as reference frames

Consider the tetrads eia as defined in §7. By definition, eia transforms as covariant vector fields
under coordinate transformations, and as contravariant vector fields under SO(3,1) Lorentz
transformations, ẽia = Λi j eja , where {Λi j } are the representation matrices of SO(3,1) group.
Note that the tetrads establish the local reference frame of an observer that moves along
a trajectory C, represented by the worldline xa (τ ). In order to measure field quantities with
magnitude and direction, an observer must project these quantities on the frame carried by
the observer. The projection of a vector V a (x) at a position xa , on a particular frame eia (x),
is V i (x) = eia (x) V a (x). The velocity of the observer along C is ua (τ ) = dxa /dτ . If we
identify the observer’s velocity with the i = 0 component of the cotetrad eai , ea0 = ua (τ ),
the acceleration aa of the observer is given by the absolute derivative of ua along C,

Dua Dea0
aa = = = ub ∇b ea0 , (B.1)
dτ dτ
where the covariant derivative is constructed out of the metrical torsion-free Christoffel
symbols 0 Γabc . We see that if ua = ea0 represents a geodesic trajectory, then the frame is in
free fall and aa = 0. Therefore, a non-vanishing aa do represent inertial accelerations of the
frame.
As for the acceleration of the frame along an arbitrary path xa (τ ) of the observer. the
acceleration of the frame along the path is

Deai
= ϕi j eaj . (B.2)
dτ
The tensor ϕij is the antisymmetric acceleration tensor, which can be identified with the
4-rotation in the Fermi-Walker transport frame (cf. (5.38)). Obviously,

Dei a
ϕi j = eja = eja ub ∇b eai = −eai ub ∇b eja . (B.3)
dτ
Recall the identity (7.13), i.e. ∇b eia = −ω ikb eka , where the spin connection can be decom-
posed into
ωija = ω̊ija + Kaij . (B.4)
The torsion-free part ω̊ija can be calculated from the tetrads as

1
ω̊ija = − eka (Ωijk − Ωjik − Ωkij ), Ωijk = eib (eaj ∂a ebk − eak ∂a ebj ). (B.5)
2

143
And the contortion tensor Kaij is defined by
1
Kaij = eci ebj (Tcab + Tbca + Tacb ), Tcab = eic Tiab , (B.6)
2
with Tiab = ∂a eib − ∂b eia . We consider the Weitzenböck gauge ωija = 0, then we have the
identity ω̊ija = −Kija . After simple manipulations we finally obtain
1
ϕij = [T0ij + Ti0j − Tj0i ], (B.7)
2
which is clearly not invariant under local SO(3,1) transformations but invariant under co-
ordinate transformations. We interpret ϕij as the inertial accelerations along the trajectory
xa (τ ).
Therefore, given any set of tetrads for an arbitrary space-time, its geometrical interpre-
tation may be obtained i) either by identifying the velocity ua = ea0 of the field of observers,
together with the orientation of the other three components of the tetrad, or ii) by the val-
ues of the acceleration tensor ϕij , which characterize the inertial state of the frame. In both
cases, the fixation of the frame requires the fixation of 6 components of the tetrads. Note
that, given an arbitrary frame, it is possible to rotatethe frame, by local Lorentz transforma-
tions, to obtain a Fermi-Walker transported frame in which ϕij = 0.

B.2 Lagrangian formulation of TEGR

Teleparallel theories of gravity are constructed out of the tetrads eia and the torsion tensor
Tiab . The class of theories whose field equations are second order differential equations
are established from a Lagrangian density constructed out of the Weitzenböck invariants
T ijk Tijk , T ijk Tjik and T i Ti , where Ti = T j ji . The equivalence of a particular teleparallel
theory with Einstein’s general relativity is verified by means of algebraic identities between
the tetrad fields, the torsion tensor, the contorsion tensor and the SO(3,1) spin connection.
The most important identity is

ω̊ija = −Kija . (B.8)

This identity holds in the Weitzenböck gauge, ωija = 0, which means that

∇a eib = ∂a eib + ω̊ i ja ejb − Γ̊cab eic = 0. (B.9)

Or equivalently,
eic ∂a eib = Γ̊cab − eic (ω̊ija )ejb . (B.10)
The left hand side is called the Weitzenböck connection, Wab c
. Taking into account (B.8),
we obtain
Wabc
= Γ̊cab + eic Kaij ejb . (B.11)
The curvature tensor of the Weitzenböck connection vanishes, Rabcd (W ) = 0. Thus, the
scalar curvature density eR(e) = e eia ejb Rijab (0 ω) becomes
1 1
eR(e) = −e( T ijk Tijk + T ijk Tjik − T i Ti ) + 2∂a (eT a ), (B.12)
4 2

144
where Ta = T b ba . Let us also introduce the tensor
1 1
Σijk = (T ijk + T jik − T kij ) + (η ik T j − η ij T k ), (B.13)
4 2
then (B.12) may be rewritten as

eR(e) = −eΣijk Tijk + 2∂a (eT a ) . (B.14)

We see that, except for the total divergence, the quadratic scalar density eΣijk Tijk is equiva-
lent to the scalar curvature density eR(e). Therefore, the Lagrangian density of TEGR can
be chosen as
L(e) = −keΣijk Tijk − LM (B.15)
up to surfaces terms, where LM stands for the Lagrangian density of the matter fields, and
k = 1/(16πG).
The field equations derived from L(e) are
1 1
eic eja ∂b (eΣjcb ) − e(Σjb i Tjba − eia Tjkl Σjkl ) = eTia , (B.16)
4 4k
where Tia is defined by δLM /δeia = eTaµ . These field equations are covariant under the
local SO(3,1) transformation. Let us define ω̊ a ≡ ω̊c ca , then
1
Ta = −ω̊a , Tacb = ω̊cba − ω̊bca , Σajk = − (ω̊ajk − eja ω̊k + eka ω̊j ).
2
By further algebraic manipulation, one obtains that the LHS of (B.16) can be identically
rewritten as 12 e Ria (e) − 21 eia R(e) . The indices in (B.16) can be converted into spacetime
 

indices, and hence the LHS of the latter equation becomes proportional to (Rab − 12 gab R).
Consequently, a metric tensor gµν that is a solution of Einstein’s equations is also a solution
of (B.16).
Ex. Work out the ADM formulation for TEGR.

B.3 Symmetric teleparallel general relativity

The symmetric teleparallel general relativity (STGR) is also equivalent to GR. In STGR,
we require the curvature and torsion are equal to zero, and the gravity is attributed to
the non-metricity tensor Qabc . In this theory the metric gab and affine connection Γabc are
fundamental. Similar to TEGR, here under the teleparallelism constraint, one can always
choose the so-called coincident gauge in which the affine connection vanishes.
The non-metricity tensor is defined as

Qabc = ∇a gbc = ∂a gbc − Γdab gdc − Γdac gbd , (B.17)

then the affine connection can be represented as

Γdab = Γ̊dab + Sab

d
, (B.18)

where Γ̊dab is the usual Levi-Civita connection from the metric and
1 1
Sdab = − (Tabd + Tbad − Tdab ) − (Qabd + Qbad − Qdab ) (B.19)
2 2

145
is the distortion tensor. Here, all quantities denoted by a ring over the heads are associated
with the Levi-Civita connection. The relation between curvatures corresponding to Γdab
and Γ̊dab is then
Rabc d = R̊abcd − 2∇˚ [a S d b]c + 2S e [a|c| S d b]e , (B.20)
and the scalar curvature relation is
˚ [a S d d] a + 2S e [a a S d d]e .
R = R̊ − 2∇ (B.21)

Next, the teleparallelism constrains the curvature tensor Rabc d (Γ) to be zero. Besides, the
symmetrical teleparallelism further poses a torsionless constraint on the connection, T c ab =
0, so that in STGR gravitation is totally attributed to the non-metricity. Indeed, the action
of STGR is constructed as
√ √
Z Z
1 1
SSTGR = 4
d x −gQ = d4 x −gQabc P abc , (B.22)
2 2
where we have set 8πG = 1, and
1 1 1 1
P abc = − Qabc + Q(bc)a + (Qa − Q̃a )g bc − g a(b Qc) . (B.23)
4 2 4 4
The vectors Qa , Q̃a are two different traces of the non-metricity tensor

Qa = g bc Qabc , Q̃a = g bc Qbac .

This way, one can prove that the STGR action (B.22) is identical the the Einstein-Hilbert
action of GR up to a surface term,
√ h
Z
1 i
SSTGR = 4 ˚ a a
d x −g R̊ + ∇µ (Q − Q̃ ) , (B.24)
2
So this theory is equivalent to GR.
The constraints Rabc d (Γ) = 0 and T a bc = 0 allow us to choose a coordinate system {y a }
such that Γabc (y) = 0 (so that Qabc = ∂a gbc ). This is called the coincident gauge. Then in any
other coordinate system {xa }, the affine connection will have the following form,
∂xa
Γabc (x) = d ∂b ∂c y d . (B.25)
∂y
We see that the affine connection in STGR is purely inertial, and only the metric is the
fundamental variable.
In STGR, if we choose any other gauge in which the connection does not vanish but
has the general form of (B.25), only the surface term in the action (B.24) is changed. So we
can always take the coincident gauge in STGR to obtain the EoM for metrics,
2 √
1
√ ∂d −gP dab − Qg ab + Qacd P b cd + 2Qcda Pcd b = T ab , (B.26)


−g 2
where T ab is energy-momentum tensor of matter fields, and the EoM for connections is
simply
√
∇a ∇b ( −gP dab ) = 0. (B.27)
Ex. Work out the ADM formulation for STGR. :)

146
C Black hole shadows
The observations by Event Horizon Telescope of the black shadows of the black hole at
the center of the galaxy M87 (2019) and the black hole Sagittarius A* at the center of our
galaxy (2022) are great astronomical breakthroughs in recent days. In this appendix, we
present the analytic calculation of black shadows using simply null geodesics in black hole
spacetimes following [39].
By black shadows we mean the following: If light rays pass close to a black hole, they
can be deflected very strongly and even travel on circular orbits so that no light comes out of
a black hole, and consequently a black hole is seen as a dark disk in the sky, i.e. a backdrop
of light sources (or a dark silhouette of a body which occurs when we look at it against
a bright background). We therefore consider the idealized situation where there are light
sources densely distributed everywhere in the universe but not in the region between the
observer and the black hole. As each light ray corresponds to a point on the observer’s sky,
we would then assign brightness to a point on the observer’s sky if the corresponding light
ray goes to infinity and darkness otherwise. The dark part of the observer’s sky is what we
call the shadow. Its boundary corresponds to light rays that go neither to infinity nor to
the horizon but are trapped within the space-time. In the Schwarzschild spacetime, these
light rays asymptotically approach an unstable photon sphere, i.e., a sphere that is filled with
circular lightlike geodesics that are unstable with respect to radial perturbations.

C.1 Angular size of the shadow

Now let us compute the shadow in a spherically symmetric and static spacetime. (Here we
will not consider objects without a photon sphere.)
Consider a spherically symmetric and static metric

ds2 = A(r)dt2 − B(r)dr2 − D(r) dθ2 + sin2 θ dϕ2 , (C.1)

where A(r), B(r) and D(r) are positive functions of r. In this metric, the Lagrangian for
the geodesics is
L = A(r)ṫ2 − B(r)ṙ2 − D(r)(θ̇2 + sin2 θϕ̇2 ). (C.2)
Because of the symmetry, it suffices to consider geodesics in the equatorial plane with θ =
π/2, sin θ = 1. The t and ϕ components of the Euler-Lagrange equation give us two
constants of motion, (cf. (9.24) and (9.25))

E = A(r)ṫ, J = D(r)ϕ̇. (C.3)

The null geodesic equation is

− A(r)ṫ2 + B(r) ṙ2 + D(r) ϕ̇2 = 0 . (C.4)

After inserting (C.3), this equation can be solved for ṙ2 /ϕ̇2 = (dr/dϕ)2 which gives us the
orbit equation for lightlike geodesics,
 2
D(r) D(r) E 2
 
dr
= −1 . (C.5)
dϕ B(r) A(r) J 2

147
We see that the orbit equation for a given metric depends only on one constant of motion,
for example, on the impact parameter J/E = b. Note that (C.5) is of the same form as an
energy conservation law in one-dimensional classical mechanics, (dr/dϕ)2 + Veff (r) = 0,
where the effective potential depends on the impact parameter b = J/E, with ϕ playing the
role of the time variable. If the impact parameter has been fixed, we may thus visualize the
radial motion by a motion in the classical potential Veff (r). The circular orbits can then be
determined by solving the equations Veff = 0 and dVeff /dr = 0 with respect to r and b.
In situations where the light ray approaches the center and then goes out again after
reaching a minimum radius R, it is convenient to rewrite the orbit equation (C.5) using
R instead of b. As R corresponds to the turning point of the trajectory, the condition
dr/dϕ|R = 0 has to hold. Using (C.5), we obtain the relation between R and the con-
stant of motion E/L:
E2 A(R) 1 A(R)
2
= , or 2
= . (C.6)
J D(R) b D(R)
Let us also introduce the function
D(r)
h(r)2 = . (C.7)
A(r)

It is easy to see that b = h(R). Using (C.6) and (C.7) in (C.5), we get
2
h(r)2
  
dr D(r)
= −1 . (C.8)
dϕ B(r) h(R)2

For constructing the shadow we assume that a static observer at radius coordinate rO
sends light rays into the past. As can be seen from Fig. 15, the angle α between such a light
ray and the radial direction is given by
√ p
grr dr B(r) dr
cot α = √ = p . (C.9)
gϕϕ dϕ r=rO D(r) dϕ r=rO

With the help of (C.8), we obtain

h(rO )2
cot2 α = − 1. (C.10)
h(R)2

By elementary trigonometry, we get

h(R)2
sin2 α = . (C.11)
h(rO )2

The boundary curve of the shadow corresponds to past-oriented light rays that asymptoti-
cally approach one of the unstable circular light orbits at radius rph . Therefore we have to
consider the limit R → rph in (C.11) for getting the angular radius αsh of the shadow,

h(rph )2
sin2 αsh = . (C.12)
h(rO )2

148
Figure 15: Example of light ray emitted from the observer’s position (green disk) into the
past under an angle α. rph is the photon sphere radius. (This picture is taken from [39]
without authors’ permission. Sorry!)

Here h(r) is given by the formula (C.7). Note that the critical value bcr of the impact pa-
rameter is connected with rph by bcr = h(rph ) Therefore we can also write eq.(C.12) as
b2cr b2cr A(rO )
sin2 αsh = , or sin2 αsh = . (C.13)
h(rO )2 D(rO )

The only thing we still have to find is the radius of the photon sphere rph . Along a circular
light orbit the two conditions dr/dϕ = 0 and d2 r/dϕ2 = 0 have to hold simultaneously. The
condition of d2 r/dφ2 = 0 can be obtained by differentiation of eq. (C.5). Solving the two
equations together, we find the equation for the radius of a circular light orbit in the form
d
0 = h(r)2 . (C.14)
dr
This equation may be satisfied for several radius values r, so there may be several photon
spheres.
Let us now consider a spherically symmetric and static metric (C.1) that is asymptotically
flat, i.e. A(r) → 1, B(r) → 1 and D(r)/r2 → 1 for r → ∞. Then for an observer at a large
distance the angular size (C.12) of the shadow can be approximated by
h(rph ) bcr
αsh = , or αsh = . (C.15)
rO rO
Therefore, for calculating the shadow for large distances in an asymptotically flat spherically
symmetric and static space-time, it is sufficient to determine the function h(r) atpr = rph .
If
pwe additionally assume that D(r) = r and introduce the function w(r) = A(r) =
2

−gtt (r), then the equation (C.14) can be simplified to

rw′ (r) = w(r). (C.16)

In the approximations stated above it can be further simplified to give

rph
rsh ≡ bcr = . (C.17)
w(rph )
rsh has the dimension of a length.

149
Example C.1 (Shadow in the Schwarzschild spacetime for a static observer). For the Schwarzschild
space-time, the function h(r) and the radius of the photon sphere rph specify to

r2 √
h(r)2 = , rph = 3M , bcr = 3 3M , (C.18)
1 − 2M/r

and, after substitution into (C.12), we obtain Synge’s formula

√ p
27M 2 (1 − 2M/rO ) 3 3M 1 − 2M/rO
2
sin αsh = 2
, or sin αsh = . (C.19)
rO rO

For large distances we have:

√
3 3M
αsh = , rO ≫ M . (C.20)
rO

Example C.2 (Shadow in the Reissner-Nordström space-time for a static observer). In the
Reissner-Nordström space-time, the function h(r) takes the form

r2
h(r) = 2
. (C.21)
1 − 2M/r + Q2 /r2

(C.14) gives
3M 1p
rph = + 9M 2 − 8Q2 . (C.22)
2 2
For the critical value of the impact parameter we obtain
 p 4
3M + − 9M 2 8Q2
b2cr =  p . (C.23)
8 3M 2 − 2Q2 + M 9M 2 − 8Q2

With known bcr , the angular radius αsh of the shadow can now be found for an arbitrary
position of the observer using (C.13). For observers at large distances, (C.15) can be used.

C.2 Shadow of a Kerr black hole

What happens to the photon sphere if the black hole is rotating? To that end we now
consider a Kerr black hole in Boyer-Lindquist coordinates (10.12). In this case, there are 4
constants of motion which, for lightlike geodesics, are L = gab ẋa ẋb = 0, the energy E, the
z component of the angular momentum J and the Carter constant K. (Cf. (10.25).)
To begin with, let us consider light rays in the equatorial plane. In the domain of outer
communication (i.e., between the outer horizon at r+ and infinity) there are always exactly
two circular light orbits; one is corotating with the black hole (prograde), the other one
counterrotating (retrograde). Both of them are unstable with respect to radial perturbations,
i.e. they can serve as limit curves for lightlike geodesics. Now the value of the radius
decreases from 3m to m for the prograde orbit and it increases from 3m to 4m for the
retrograde one, if the BH spin parameter |a| varies from 0 to m.

150
 Now let us consider non-equatorial light rays. In the Kerr metric there is no photon sphere
but rather a photon region, also sometimes referred to as photon shell. Whereas a photon sphere
is filled with circular light rays, the Kerr photon region is filled with spherical, in general non-
planar, light rays. As all spherical lightlike geodesics in the domain of outer communication
turn out to be unstable, they can serve as limit curves to which past-oriented light rays
from an observer position spiral asymptotically. This way, the photon region determines
the boundary curve of the shadow.
For all light rays issuing from the observer position into the past the initial direction is
determined by two angles in the observer’s sky, a colatitude angle ϑ and an azimuthal angle
ψ. To define these, consider the orthonormal tetrad

(r2 + a2 ) ∂t + a ∂ϕ ∂θ
e0 = √ √ , e1 = √ ,
r2 + a2 cos2 θ r2 − 2M r + a2 r2 + a2 cos2 θ
√
−∂ϕ − a sin2 θ ∂t − r2 − 2M r + a2 ∂r
e2 = √ √ , e3 = √ (C.24)
r2 + a2 cos2 θ r2 − 2M r + a2 r2 + a2 cos2 θ
at the position of the observer. This tetrad is chosen such that e3 − e0 is tangent to the
past-oriented ingoing principal null ray whereas −e3 − e0 is tangent to the past-oriented
outgoing principal null ray.
The pair (ϑ, ψ) gives us one point of the shadow boundary curve. If the azimuthal angle
ψ varies from −π/2 to π/2, the corresponding radius coordinate rp varies from its maximal
value to its minimal value. This gives the lower half of the boundary curve of the shadow to
which the upper half is symmetric. When working this out for an observer at the position
(rO , θO ), one finds
p 2 p
JE (rp ) − a sin2 θO rO + a2 − 2M rO KE (rp )
sin ψ(rp ) = p , sin ϑ(rp ) = 2
, (C.25)
KE (rp ) sin θO rO − aJE (rp ) + a2

where
4rp2 (rp2 + a2 − 2M rp ) −rp2 (rp − 3M ) − rp a2 − a2 M
KE (rp ) = , aJE (rp ) = (C.26)
(rp − M )2 rp − M

are the constants of motion, KE = K/E 2 and LE = J/E, of the spherical light ray at rp .
The minimal and maximal values, rp,min and rp,max , of the parameter rp can be be found
from the equations

sin ψ(rp ) = 1 for rp,min , and sin ψ(rp ) = −1 for rp,max . (C.27)

(C.25) give us the celestial coordinates of the boundary curve of the shadow explicitly in
dependence of the parameter rp .
For plotting the shadow, we use a stereographic projection which maps the celestial
sphere of the observer (except the pole at θ = π) onto a plane that is tangent to this sphere
at the pole θ = 0. In this plane we introduce (dimensionless) Cartesian coordinates,
   
ϑ(rp ) ϑ(rp )
(C.28)



x(rp ) = −2 tan sin ψ(rp ) , y(rp ) = −2 tan cos ψ(rp ) .
2 2

151
 nearby equatorial observer

Figure 16: Kerr shadow as seen by an equatorial observer (θO = π/2). This is taken from
[39], without permission. LEFT: Shadow as seen by an observer at rO = 5m for different val-
ues of spin (from the leftmost to the rightmost): a = 0.02m, 0.4m, 0.75m, 0.999m. RIGHT:
Shadow with fixed spin a = 0.999m as seen by an observer at different positions (from the
innermost to the outermost): rO = 5m, 10m, 20m, 50m.

Examples of Kerr shadows are presented in Fig.16.

For a distant observer, rO ≫ m, the formulas for the shadow can be simplified. Firstly,
we can solve the second equation in (C.25) for ϑ(rp ) and then linearize with respect to m/rO ,
p
KE (rp )
ϑ(rp ) = . (C.29)
rO
Secondly, to within the same approximation (C.28) simplify to
s
2
2
a sin θO − JE (rp ) 1 JE (rp ) − a sin2 θO
x(rp ) = , y(rp ) = ± KE (rp ) − . (C.30)
rO sin θO rO sin2 θO

Examples of shadow calculations based on (C.30), with (C.26), are presented in Fig.17.

References
[1] L. P. Hughston and K. P. Tod, An Introduction to General Relativity, Cambridge Uni-
versity Press, 1990.

[2] S. W. Hawking and G. F. R. Ellis, The Large Scale Structure of Space-Time, Cambridge
University Press, 1973.

[3] M. Carmeli, Classical Fields: General Relativity and Gauge Theory, John Wiley & Sons,
1982.

[4] R. M. Wald, General Relativity, The University of Chicago Press, 1984.

152
 distant equatorial observer

Figure 17: Kerr shadow as seen by a distant equatorial observer (rO ≫ m, θO = π/2).
This is again taken from [39], without permission. The axes are in units of m/rO . LEFT:
Kerr shadow for different values of spin (from the leftmost to the rightmost): a = 0, 0.5m,
0.9999m. RIGHT: Shadow of nearly extreme Kerr BH for different values of spin (from the
leftmost to the rightmost): a = 0.97m, 0.99m, 0.9999m.

[5] J. Foster and J. D. Nightingale, A Short Course in General Relativity, 3rd Edition,
Springer, 2006.

[6] B. Schutz, A First Course in General Relativity, 2nd Edition, Cambridge University Press,
2009.

[7] D. B. Malament, Topics in the Foundations of General Relativity and Newtonian Gravitation
Theory, The University of Chicago Press, 2012.

[8] N. Straumann, General Relativity, 2nd Edition, Springer, 2013.

[9] C. G. Böhmer, Introduction to General Relativity and Cosmology, World Scientific, 2017.

[10] S. Carlip, General Relativity: A Concise Introduction, Oxford University Press, 2019.

[11] B. Chen (陳斌), General Relativity, Peking University Press, 2018.

[12] L. Liu (劉遼) and Z. Zhao (趙崢), General Relativity, 2nd Edition, Higher Education
Press, 2004.
(N.B. The metric signature is not uniformly chosen in this book......)

[13] Z.-J. Zhang (張鎮九), Modern Relativity and Black Hole Physics, Central China Normal
University Press, 1986.

[14] L. Zhao (趙柳), Introduction to Relativity and Gravitation, Science Press, 2017.

[15] B. Chen (陳斌), Introduction to General Relativity, Peking University Press, 2022.
(This is the simplified version of [11].)

153
[16] Z. Zhao (趙崢) and W.-B. Liu (劉文彪), Fundamentals of General Relativity, Tsinghua
University Press, 2010.
(This is the simplified version of [12].)

[17] S. K. Chakrabarti, R. Geroch and C.-b. Liang (梁燦彬), Timelike curves of limited
acceleration in general relativity, J. Math. Phys. 24, 597 (1983).

[18] M. Mohseni, World-line deviation and spinning particles, Phys. Lett. B 587, 133
(2004).

[19] M. Mohseni, The Raychaudhuri equation for spinning test particles, Gen. Relativ.
Gravit. 47, 24 (2015).

[20] M. L. Ruggiero and A. Ortolan, Gravito-electromagnetic approach for the space-time

of a plane gravitational wave, J. Phys. Commun. 4, 055013 (2020).

[21] D. Garfinkle, Gravitational wave memory and the wave equation, Class. Quantum
Grav. 39 135010 (2022).

[22] J. L. Weiner, Generalizing Birkhoff, Gen. Relativ. Gravit. 53, 34 (2021).

[23] F. Sorge, Kerr spacetime in Lemaitre coordinates, arXiv:2112.15441.

[24] M. A. Arganaraz and O. M. Moreschi, Double null coordinates for Kerr spacetime,
Phys. Rev. D 104, 024049 (2021).

[25] C. Doran, New form of the Kerr solution, Phys. Rev. D 61, 067503 (2000).

[26] K. P. Tod, F. de Flice, and M. Calvani, Spinning test particles in the field of a black
hole, Nuovo Cimento B 34, 365 (1976).

[27] F. Gray and D. Kubizňák, Slowly rotating black holes with exact Killing tensor sym-
metries, Phys. Rev. D 105, 064017 (2022).

[28] S. W. Hawking, Black holes in general relativity, Commun. Math. Phys. 25, 152
(1972).

[29] R. Penrose, Singularity theorems, in C.-N. Yang (楊振寧), M.-L. Ge (葛墨林), Y.-H.
He (何楊輝), (ed.) Topology and Physics, World Scientific, 2019.

[30] W. Israel, Third law of black-hole dynamics: A formulation and proof, Phys. Rev.
Lett. 57, 397 (1986).

[31] C. Kehle and R. Unger, Gravitational collapse to extremal black holes and the third law
of black hole thermodynamics, arXiv:2211.15742v2; Extremal black hole formation as
a critical phenomenon, arXiv:2402.10190.

[32] J. C. Schindler, A. Aguirre, and A. Kuttner, Understanding black hole evaporation

using explicitly computed Penrose diagrams, Phys. Rev. D 101, 024010 (2020).

[33] S. Dain, Geometric inequalities for black holes, Gen. Relativ. Gravit. 46, 1715 (2014).

154
 [34] A. Yu. Kamenshchik, The Bianchi classification of the three-dimensional Lie algebras
and homogeneous cosmologies and the Mixmaster universe, in S. Cacciatori et al. (ed.),
Einstein Equations: Physical and Mathematical Aspects of General Relativity, Springer,
2019.

[35] R. M. Wald, Asymptotic behavior of homogeneous cosmological models in the pres-

ence of a positive cosmological constant, Phys. Rev. D 28, 2118 (1983).

[36] L. A. Gómez López and G. D. Quiroga, Asymptotic structure of spacetime and the
Newman-Penrose formalism: A brief review, Rev. Mex. Fis. 63, 275 (2017).

[37] J. W. Maluf, The teleparallel equivalent of general relativity, Ann. Phys. (Berlin) 525,
339 (2013).

[38] D.-h. Zhao (趙德浩), Covariant formulation of f (Q) theory, Eur. Phys. J. C 82, 303
(2022).

[39] V. Perlick and O. Yu. Tsupko, Calculating black hole shadows: Review of analytical
studies, Phys. Rep. 947, 1 (2022).

155

View publication stats