Lecture Notes on General Relativity (GR)

lazierthanthou
(https://github.com/lazierthanthou/Lecture_Notes_GR)

March 24, 2016

Contents
1 Topology 4
1.1 Topological Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2 Continuous maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Composition of continuous maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Inheriting a topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2 Manifolds 6
2.1 Topological manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2 Terminology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.3 Chart transition maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.4 Manifold philosophy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

3 Multilinear Algebra 8
3.1 Vector Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3.2 Linear Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.3 Vector Space of Homomorphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.4 Dual Vector Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.5 Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
3.6 Vectors and Covectors as Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
3.7 Bases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
3.8 Basis for the Dual Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3.9 Components of Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

4 Differential Manifolds 12
4.1 Compatible charts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
4.2 Diffeomorphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

5 Tangent Spaces 15
5.1 Velocities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
5.2 Tangent vector space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
5.3 Components of a vector w.r.t. a chart . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
5.4 Chart-induced basis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
5.5 Change of vector components under a change of chart . . . . . . . . . . . . . . . . . . . . . . . . 18
5.6 Cotangent spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
5.7 Change of components of a covector under a change of chart . . . . . . . . . . . . . . . . . . . . . 19

6 Fields 21
6.1 Bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

1
 6.2 Tangent bundle of smooth manifold . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
6.3 Vector fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
6.4 The C ∞ (M )-module Γ(T M ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
6.5 Tensor fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

7 Connections 25
7.1 Directional derivatives of tensor fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
7.2 New structure on (M, O, A) required to fix ∇ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
7.3 Change of Γ’s under change of chart . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
7.4 Normal Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

8 Parallel Transport & Curvature 29

8.1 Parallelity of vector fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
8.2 Autoparallely transported curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
8.3 Autoparallel equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
8.4 Torsion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
8.5 Curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

9 Newtonian spacetime is curved! 32

9.1 Laplace’s questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
9.2 The full wisdom of Newton I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
9.3 The foundations of the geometric formulation of Newton’s axiom . . . . . . . . . . . . . . . . . . 33

10 Metric Manifolds 36
10.1 Metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
10.2 Signature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
10.3 Length of a curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
10.4 Geodesics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

11 Symmetry 40
11.1 Push-forward map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
11.2 Pull-back map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
11.3 Flow of a complete vector field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
11.4 Lie subalgebras of the Lie algebra (Γ(T M ), [·, ·]) of vector fields . . . . . . . . . . . . . . . . . . . 42
11.5 Symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
11.6 Lie derivative . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

12 Integration 45
12.1 Review of integration on Rd . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
12.2 Integration on one chart . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
12.3 Volume forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
12.4 Integration on onechart domain U . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
12.5 Integration on the entire manifold . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

13 Lecture 13: Relativistic spacetime 49

13.1 Time orientation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
13.2 Observers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
13.3 Role of the Lorentz transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

14 Lecture 14: Matter 53

14.1 Point matter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
14.2 Field matter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
14.3 Energy-momentum tensor of matter fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

2
15 Einstein gravity 57
15.1 Hilbert . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
15.2 Variation of SHilbert . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
15.3 Solution of the ∇a T ab = 0 issue . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
15.4 Variants of the field equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

16 L18: Canonical Formulation of GR-I 60

16.1 Dynamical and Hamiltonian formulation of General Relativity . . . . . . . . . . . . . . . . . . . . 60

17 Lecture 22: Black Holes 61

17.1 Radial null geodesics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
17.2 Eddington-Finkelstein . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

Abstract

These are lecture notes on General Relativity.

They are based on the Central Lecture Course by Dr. Frederic P. Schuller (A thorough introduc-
tion to the theory of general relativity) introducing the mathematical and physical foundations of the
theory in 24 self-contained lectures at the International Winter School on Gravity and Light in Linz/Austria
for the WE Heraeus International Winter School of Gravity and Light, 2015 in Linz as part of the world-wide
celebrations of the 100th anniversary of Einstein’s theory of general relativity and the International Year of
Light 2015.
These lectures develop the theory from first principles and aim at an audience ranging from ambitious
undergraduate students to beginning PhD students in mathematics and physics. Satellite Lectures (see
other videos on this channel) by Bernard F Schutz (Gravitational Waves), Domenico Giulini (Canonical
Formulation of Gravity), Marcus C Werner (Gravitational Lensing) and Valeria Pettorino (Cosmic Microwave
Background) expand on the topics of this central lecture course and take students to the research frontier.
Spacetime is the physical key object, we shall be concerned about.

Spacetime is a 4-dimensional topological manifold with a smooth atlas carrying a torsion-

free connection compatible with a Lorentzian metric and a time orientation satisfying the Ein-
stein equations.

3
1 Topology

Motivation: At the coarsest level, spacetime is a set. But, a set is not enough to talk about continuity
of maps, which is required for classical physics notions such as trajectory of a particle. We do not want
jumps such as a particle disappearing at some point on its trajectory and appearing somewhere. So we
require continuity of maps. There could be many structures that allow us to talk about continuity, e.g.,
distance measure. But we need to be very minimal and very economic in order not to introduce undue
assumptions. So we are interested in the weakest structure that can be established on a set which allows a
good definition of continuity of maps. Mathematicians know that the weakest such structure is topology.
This is the reason for studying topological spaces.

1.1 Topological Spaces

Definition 1. Let M be a set and P(M ) be the power set of M , i.e., the set of all subsets of M .
A set O ⊆ P(M ) is called a topology, if it satisfies the following:

(i) ∅ ∈ O, M ∈ O

(ii) U ∈ O, V ∈ O =⇒ U ∩ V ∈ O
S

(iii) Uα ∈ O, α ∈ A (A is an index set =⇒ α∈A Uα ∈ O

Terminology:

1. the tuple (M, O) is a topological space.

2. U ∈ M is an open set if U ∈ O.

3. U ∈ M is a closed set if M \ U ∈ O.
Definition 2. (M, O), where O = {∅, M } is called the chaotic topology.
Definition 3. (M, O), where O = P(M ) is called the discrete topology.
Definition 4. A soft ball at the point p in Rd is the set
( d
)
X
Br (p) := (q1 , q2 , ..., qd ) | (qi − pi )2 < r2 where r ∈ R+ (1.1)
i=1

Definition 5. (Rd , Ostd ) is the standard topology, provided that U ∈ Ostd iff
∀p ∈ U, ∃r ∈ R+ : Br (p) ⊆ U

Proof. ∅ ∈ Ostd since ∀ p ∈ ∅, ∃r ∈ R+ : Br (p) ⊆ ∅ (i.e. satisfied “vacuously”)

Rd ∈ Ostd since ∀p ∈ Rd , ∃r = 1 ∈ R+ : Br (p) ⊆ Rd

Suppose U, V ∈ Ostd . Let p ∈ U ∩ V =⇒ ∃ r1 , r2 ∈ R+ s.t. Br1 (p) ⊆ U, Br2 (p) ⊆ V .

Let r = min {r1 , r2 } =⇒ Br (p) ⊆ U and Br (p) ⊆ V =⇒ Br (p) ⊆ U ∩ V =⇒ U ∩ V ∈ Ostd .
S
Suppose, Uα ∈ Ostd , ∀ α ∈ A. Let p ∈ α∈A Uα =⇒ ∃α ∈ A : p ∈ Uα
=⇒ ∃ r ∈ R+ : Br (p) ⊆ Uα ⊆ α∈A Uα =⇒ α∈A Uα ∈ Ostd .
S S

1.2 Continuous maps

A map f , f : M −→ N , connects each element of a set M (domain set) to an element of a set N (target
set).

Terminology:

4
 1. If f maps m ∈ M to n ∈ N , then we may say f (m) = n, or m maps to n, or m 7→ f (m) or m 7→ n.

2. If V ⊆ N, preimf (V ) := {m ∈ M |f (m) ∈ V }

3. If ∀n ∈ N, ∃m ∈ M : n = f (m), then f is surjective. Or, f : M  N .

4. If m1 , m2 ∈ M, m1 6= m2 =⇒ f (m1 ) 6= f (m2 ), then f is injective. Or, f : M ,→ N .

Definition 6. Let (M, OM ) and (N, ON ) be topological spaces. A map f : M −→ N is called continuous
w.r.t. OM and ON if V ∈ ON =⇒ (preimf (V )) ∈ OM .

Mnemonic: A map is continuous iff the preimages of all open sets are open sets.

1.3 Composition of continuous maps

Definition 7. If f : M −→ N and g : N −→ P , then

g ◦ f : M −→ P such that m 7→ (g ◦ f )(m) := g(f (m))

Theorem 1.1. If f : M −→ N is continuous w.r.t. OM and ON and g : N −→ P is continuous w.r.t. ON

and OP , then g ◦ f : M −→ P is continuous w.r.t. OM and OP .

Proof. Let W ∈ OP .

preimg◦f (W ) = {m ∈ M |g(f (m)) ∈ W } ∵ (g ◦ f )(m) = g(f (m))

= {m ∈ M |f (m) ∈ preimg (W )} preimg (W ) ∈ ON ∵ g is continuous
= preimf (preimg (W )) ∈ OM ∵ f is continuous
=⇒ g ◦ f is continuous

1.4 Inheriting a topology

Given a topological space (M, OM ), one way of inheriting a topology from it is the subspace topology.
Theorem 1.2. If (M, OM ) is a topological space and S ⊆ M , then the set O|S ⊆ P(S) such that O|S :=
{S ∩ U |U ∈ OM } is a topology. O|S is called the subspace topology inherited from OM .

Proof. 1. ∅, S ∈ O|S ∵ ∅ = S ∩ ∅, S = S ∩ M .

2. S1 , S2 ∈ O|S =⇒ ∃U1 , U2 ∈ OM : S1 = S ∩ U1 , S2 = S ∩ U2 =⇒ U1 ∩ U2 ∈ OM
=⇒ S ∩ (U1 ∩ U2 ) ∈ O|S =⇒ (S ∩ U1 ) ∩ (S ∩ U2 ) ∈ O|S =⇒ S1 ∩ S2 ∈ O|S .

3. Let α ∈ A, where A is an index set. Then Sα ∈ O|S =⇒ ∃Uα ∈ OM : Sα = S ∩ Uα .

S

Further, let U = α∈A Uα . Therefore, U ∈ OM .
S
S
S
S

Now, α∈A αS = α∈A (S ∩ U α ) = S ∩ α∈A Uα = S ∩ U =⇒ α∈A Sα ∈ O|S .

Theorem 1.3. If (M, OM ) and (N, ON ) are topological spaces, and f : M −→ N is continuous w.r.t OM and
ON , then the restriction of f to S ⊆ M, f |S : S −→ N s.t. f |S (s ∈ S) = f (s), is continuous w.r.t O|S and ON .

Proof. Let V ∈ ON . Then, preimf (V ) ∈ OM .

Now preimf |S (V ) = S ∩ preimf (V ) =⇒ preimf |S (V ) ∈ O|S =⇒ f |S is continuous.

5
2 Manifolds

Motivation: There exist so many topological spaces that mathematicians cannot even classify them. For
spacetime physics, we may focus on topological spaces (M, O) that can be charted, analogously to how the
surface of the earth is charted in an atlas.

2.1 Topological manifolds

Definition 8. A topological space (M, O) is called a d-dimensional topological manifold if

∀p ∈ M : ∃U ∈ O : p ∈ U, ∃x : U −→ x(U ) ⊆ Rd satisfying the following:

(i) x is invertible: x−1 : x(U ) −→ U

(ii) x is continuous w.r.t. (M, O) and (Rd , Ostd )

(iii) x−1 is continuous

2.2 Terminology

1. The tuple (U, x) is a chart of (M, O),

S
2. An atlas of (M, O) is a set A = {(Uα , xα )|α ∈ A, an index set} : α∈A Uα = M .

3. The map x : U −→ x(U ) ⊆ Rd is called the chart map.

4. The chart map x maps a point p ∈ U to a d-tuple of real numbers x(p) = (x1 (p), x2 (p), . . . , xd (p)). This
is equivalent to d-many maps xi (p) : U −→ R, which are called the coordinate maps.

5. If p ∈ U , then xi (p) is the ith coordinate of p w.r.t. the chart (U, x).

2.3 Chart transition maps

Imagine 2 charts (U, x) and (V, y) with overlapping regions, i.e., U ∩ V 6= ∅.

x−1 U ∩V
y
x
d
R ⊇ x(U ∩ V ) y(U ∩ V ) ⊆ Rd
y ◦ x−1

The map y ◦ x−1 is called the chart transition map, which maps an open set of Rd to another open set of
Rd . This map is continuous because it is composition of two continuous maps, Informally, these chart transition
maps contain instructions on how to glue together the charts of an atlas,

2.4 Manifold philosophy

Often it is desirable (or indeed the only way) to define properties (e.g., ‘continuity’) of real-world object (e.g.,
the curve γ : R −→ M ) by judging suitable coordinates not on the ‘real-world’ object itself, but on a chart-
representation of that real world object.

6
For example, in the picture below, we can use the map x◦γ to infer the continuity of the curve γ in U ⊆ M .

γ
R U ⊆M
x◦ x
γ
x(U ) ⊆ Rd

However, we need to ensure that the defined property does not change if we change our chosen chart. For
example, in the picture below, continuity in x ◦ γ should imply y ◦ γ. This is true, since y ◦ γ = y ◦ (x−1 ◦ x) ◦ γ =
(y ◦ x−1 ) ◦ (x ◦ γ) is continuous because it is a composition of two continuous functions, thanks to the continuity
of the chart transition map y ◦ x−1 .

y(U ) ⊆ Rd
γ
y◦ y

γ
R U ⊆M y ◦ x−1

x◦ x−1 x
γ

x(U ) ⊆ Rd

What about differentiability? Does differentiability of x ◦ γ guarantee differentiability of y ◦ γ? No. Since

composition of a differentiable map and a continuous map might only be continuous, The solution is to restrict
the atlas by removing those charts which are not differentiable. Thus, we have got rid of our problem. However,
we must remember that with the present structure, we cannot define differentiability at manifold level since we
do not know how to subtract or divide in U ⊆ M . Therefore, differentiability of γ : R −→ M makes no sense
yet.

7
3 Multilinear Algebra

Motivation: The essential object of study of linear algebra is vector space. However, a word of warning
here. We will not equip space(time) with vector space structure. This is evident since, unlike in vector
space, expressions such as 5 · Paris and Paris + Vienna do not make any sense. If multilinear algebra does
not further our aim of studying spacetime, then why do we study it? The tangent spaces Tp M (defined
in Lecture 5) at a point p of a smooth manifold M (defined in Lecture 4) carries a vector space structure
in a natural way even though the underlying position space(time) does not have a vector space structure.
Once we have a notion of tangent space, we have a derived notion of a tensor. Tensors are very important
in differential geometry.
It is beneficial to study vector spaces (and all that comes with it) abstractly for two reasons: (i) for
construction of Tp M , one needs an intermediate vector space C ∞ (M ), and (ii) tensor techniques are most
easily understood in an abstract setting.

3.1 Vector Spaces

Definition 9. A R-vector space is a triple (V, +, ·), where

i) V is a set,

ii) + : V × V −→ V (addition), and

iii) . : R × V −→ V (S-multiplication)

satisfying the following:

a) ∀u, v ∈ V : u + v = v + u (commutativity of +)

b) ∀u, v, w ∈ V : (u + v) + w = u + (v + w) (associativity of +)

c) ∃O ∈ V : ∀v ∈ V : O + v = v (neutral element in +)

d) ∀v ∈ V : ∃(−v) ∈ V : v + (−v) = 0 (inverse of element in +)

e) ∀λ, µ ∈ R, ∀v ∈ V : λ · (µ · v) = (λ · µ) · v (associativity in ·)

f) ∀λ, µ ∈ R, ∀v ∈ V : (λ + µ) · v = λ · v + µ · v (distributivity of ·)

g) ∀λ ∈ R, ∀u, v ∈ V : λ · u + λ · v = λ · (u + v) (distributivity of ·)

h) ∃1 ∈ R : ∀v ∈ V : 1 · v = v (unit element in ·)

Terminology: If (V, +, ·) is a vector space, an element of V is often referred to, informally, as a vector. But,
we should remember that it makes no sense to call an element of V a vector unless the vector space itself is
specified.

Example: Consider a set of polynomials of fixed degree,

( N
)
 X
n
P := p : (−1, +1) −→ R  p(x) = pn · x , where pn ∈ R

n=0

with ⊕ : P × P −→ P with (p, q) 7→ p ⊕ q : (p ⊕ q)(x) = p(x) + q(x) and

 : R × P −→ P with (λ, p) →
7 λ  p : (λ  p)(x) = λ · p(x). (P, ⊕, ) is a vector space.

Caution: We are considering real vector spaces, that is S-multiplication with the elements of R. We shall often
use same symbols ‘+’ and ‘·’ for different vector spaces, but the context should make things clear. When R, R2 ,
etc. are used as vector spaces, the obvious (natural) operations shall be understood to be used.

8
3.2 Linear Maps

These are the structure-respecting maps between vector spaces.

Definition 10. If (V, +v , ·v ) and (W, +w , ·w ) are vector spaces, then φ : V −→ W is called a linear map if

i) ∀v, ṽ ∈ V : φ(v +v ṽ) = φ(v) +w φ(ṽ), and

ii) ∀λ ∈ R, v ∈ V : φ(λ ·v v) = λ ·w φ(v).

∼
Notation: φ : V −→ W is a linear map ⇐⇒ φ : V −→ W

Example: Consider the vector space (P, ⊕, ) from the above example,
Then, δ : P −→ P with p 7→ δ(p) := p0 is a linear map, because
∀p, q ∈ P : δ(p ⊕ q) = (p ⊕ q)0 = p0 ⊕ q 0 = δ(p) ⊕ δ(q) and
∀λ ∈ R, p ∈ P : δ(λ  p) = (λ  p)0 = λ  p0 .
∼ ∼ ∼
Theorem 3.1. If φ : U −→ V and ψ : V −→ W , then ψ ◦ φ : U −→ W .

φ ψ
U V W

ψ◦φ

Proof. ∀u, ũ ∈ U, (ψ ◦ φ)(u +u ũ) = ψ(φ(u +u ũ)) = ψ(φ(u) +v φ(ũ)) = ψ(φ(u)) +w ψ(φ(ũ)) = (ψ ◦ φ)(u) +w (ψ ◦
φ)(ũ).

∀λ ∈ R, u ∈ U, (ψ ◦ φ)(λ ·u u) = ψ(φ(λ ·u u)) = ψ(λ ·v φ(u)) = λ ·w ψ(φ(u)) = λ ·w (ψ ◦ φ)(u)

Example: Consider the vector space (P, ⊕, ) and the differential δ : P −→ P with p 7→ δ(p) := p0 from
previous example. Then, p00 , the second differential is also linear since it is composition of two linear maps, i.e.,
∼
δ ◦ δ : P −→ P .

3.3 Vector Space of Homomorphisms

n o
∼
Definition 11. If (V, +, ·) and (W, +, ·) are vector spaces, then Hom(V, W ) := φ : V −→ W .
Theorem 3.2. (Hom(V, W ), +, ·) is a vector space with
+ : Hom(V, W ) × Hom(V, W ) −→ Hom(V, W ) with (φ, ψ) 7→ φ + ψ : (φ + ψ)(v) = φ(v) + ψ(v) and
· : R × Hom(V, W ) −→ Hom(V, W ) with (λ, φ) 7→ λ · φ : (λ · φ)(v) = λ · φ(v).

Example: (Hom(P, P ), +, ·) is a vector space. δ ∈ Hom(P, P ), δ ◦ δ ∈ Hom(P, P ), δ ◦ δ ◦ δ ∈ Hom(P, P ), etc.

Therefore, maps such as 5 · δ + δ ◦ δ ∈ Hom(P, P ). Thus, mixed order derivatives are in Hom(P, P ), and hence
linear.

3.4 Dual Vector Spaces

n o
∼
Definition 12. If (V, +, ·) is a vector space, and V ∗ := φ : V −→ R = Hom(V, R) then
(V ∗ , +, ·) is called the dual vector space to V.

Terminology: ω ∈ V ∗ is called, informally, a covector.

∼ R1
Example: Consider I : P −→ R, i.e., I ∈ P ∗ . We define I(p) := 0 p(x) dx, which can be easily checked to be
linear with I(p + q) = I(p) + I(q) and I(λ · p) = λ · I(p). Thus I is a covector, which is the integration operator
R1
0
( ) dx which eats a function.

9
Remarks: We shall also see later that the gradient is a covector. In fact, lots of things in physicist’s life, which
are covectors, have been called vectors not to bother you with details. But covectors are neither esoteric nor
unnatural.

3.5 Tensors

We can think of tensors as multilinear maps.

Definition 13. Let (V, +, ·) be a vector space. An (r,s) -tensor T over V is a multilinear map

∼
T : V ∗ × V ∗ × · · · × V ∗ × V × V × · · · × V −→ R
| {z } | {z }
r times s times

Example: If T is a (1,1)-tensor, then

T (ω1 + ω2 , v) = T (ω1 , v) + T (ω2 , v),
T (ω, v1 + v2 ) = T (ω, v1 ) + T (ω, v2 ),
T (λ · ω, v) = λ · T (ω, v), and
T (ω, λ · v) = λ · T (ω, v).
Thus, T (ω1 + ω2 , v1 + v2 ) = T (ω1 , v1 ) + T (ω1 , v2 ) + T (ω2 , v1 ) + T (ω2 , v2 ).

Remarks: Sometimes it is said that a (1,1)-tensor is something that eats a vector and outputs a vector. Here
∼ ∼ ∼
is why. For T : V ∗ × V −→ R, define φT : V −→ (V ∗ )∗ with v 7→ T ((·), v). But, clearly T ((·), v) : V ∗ −→ R,
which eats a covector and spits a number. In other words, T ((·), v) ∈ (V ∗ )∗ . Although we are yet to define
dimension, let us just trust, for the time being, that for finite-dimensional vector spaces, (V ∗ )∗ = V . So,
∼
φT : V −→ V .
∼ R1
Example: Let g : P × P −→ R with (p, q) 7→ −1 p(x) · q(x) dx. Then, g is a (0,2)-tensor over P .

3.6 Vectors and Covectors as Tensors

Theorem 3.3. If (V, +, ·) is a vector space, ω ∈ V ∗ is a (0,1)-tensor.

n o
∼
Proof. ω ∈ V ∗ and, by definition, V ∗ := φ : V −→ R , which is a collection of (0,1)-tensors.

Theorem 3.4. If (V, +, ·) is a vector space, v ∈ V is a (1,0)-tensor.

∗ ∗
Proof. We have already stated, without proof and without defining dimensions,
n that V = (V
o ) for finite-
∼
dimensional vector spaces. Therefore, v ∈ V =⇒ v ∈ (V ∗ )∗ =⇒ v ∈ φ : V ∗ −→ R =⇒ v is a
(1,0)-tensor.

3.7 Bases

Definition 14. Let (V, +, ·) is a vector space. A subset B ⊆ V is called a basis if

n
X
∀v ∈ V, ∃!e1 , e2 , . . . , en ∈ B, ∃!v1 , v2 , . . . , vn ∈ R : v = vi · e i .
i=1
Definition 15. A vector space (V, +, ·) with a basis B is said to be d-dimensional if B has d elements. In
other words, dimV := d.

Remarks: The above definition is well-defined only if every basis of a vector space has the same number of
elements.

10
Remarks: Let (V, +, ·) is a vector space. Having chosen a basis e1 , e2 , . . . , en , we may uniquely associate
n
X
v 7→ (v1 , v2 , dotsc, vn ), these numbers being the components of v w.r.t. chosen basis where v = vi · e i .
i=1

3.8 Basis for the Dual Space

Let (V, +, ·) is a vector space. Having chosen a basis e1 , e2 , . . . , en for V , we can choose a basis 1 , 2 , . . . , n for
V ∗ entirely independent of basis of V . However, it is more economical to require that

1 if a = b
a (eb ) = δba =
0 if a 6= b

This uniquely determines 1 , 2 , . . . , n from choice of e1 , e2 , . . . , en .

Remarks: The reason for using indices as superscripts or subscripts is to be able to use the Einstein summation
P
convention, which will be helpful in dropping cumbersome symbols in several equations.
Definition 16. For a basis e1 , e2 , . . . , en of vector space (V, +, ·), 1 , 2 , . . . , n is called the dual basis of the
dual space, if a (eb ) = δba .

Example: Consider polynomials P of degree 3. Choose e0 , e1 , e2 , e3 ∈ P such that e0 (x) = 1, e1 (x) = x, e2 (x) =
1 
x2 ande3 (x) = x3 . Then, it can be easily verified that the dual basis is a = ∂ a  .
a! x=0

3.9 Components of Tensors

Definition 17. Let T be a (r, s)-tensor over a d-dimensional (finite) vector space (V, +, ·). Then, with respect
to some basis {e1 , . . . , er } and the dual basis {1 , . . . , s }, define (r + s)d real numbers

T i1 ...irj1 ...js := T (i1 , . . . , ir , ej1 , . . . , ejs )

such that the indices i1 , . . . , ir , j1 , . . . , js take all possible values in the set {1, . . . , d}. These numbers T i1 ...irj1 ...js
are called the components of the tensor T w.r.t. the chosen basis.

This is useful because knowing components (and the basis w.r.t which these components have been chosen),
one can reconstruct the entire tensor.

Example: If T is a (1, 1)-tensor, then T ij := T (i , ej ). Then

 
Xd d
X d X
X d d X
X d
T (ω, v) = T  ωi · i , v j · ej  = ωi v j T (i , ej ) = ωi v j T ij =: ωi v j T ij
i=1 j=1 i=1 j=1 i=1 j=1

11
4 Differential Manifolds

Motivation: So far we have dealt with topological manifolds which allow us to talk about continuity. But
to talk about smoothness of curves on manifolds, or velocities along these curves, we need something like
differentiability. Does the structure of topological manifold allow us to talk about differentiability? The
answer is a resounding no.

So this lecture is about figuring out what structure we need to add on a topological manifold M to start
talking about differentiability of curves (R −→ M ) on a manifold, or differentiability of functions (M −→ R)
on a manifold, or differentiability of maps (M −→ N ) from one manifold M to another manifold N .

γ:R U
x◦ x
γ
x(U ) ⊆ Rd

idea. try to “lift” the undergraduate notion of differentiability of a curve on Rd to a notion of differentiability
of a curve on M

Problem Can this be well-defined under change of chart?

y(U ∩ V ) ⊆ Rd
γ
y◦ y

γ:R U ∩ V 6= ∅ y ◦ x−1

x◦ x
γ

x(U ∩ V ) ⊆ Rd

x ◦ γ undergraduate differentiable (“as a map R −→ Rd ”)

Rd −→Rd R−→Rd
z }| { z }| {
y◦γ = (y ◦ x−1 ) ◦ (x ◦ γ) = y ◦ (x−1 ◦ x) ◦ γ
| {z } | {z } | {z }
maybe only continuous, but not undergraduate differentiable continuous undergrad differentiable

At first sight, strategy does not work out.

4.1 Compatible charts

In section 1, we used any imaginable charts on the top. mfd. (M, O).

To emphasize this, we may say that we took U and V from the maximal atlas A of (M, O).
Definition 18. Two charts (U, x) and (V, y) of a top. mfd. are called `-compatible if either

(a) U ∩ V = ∅ or

(b) U ∩ V 6= ∅

chart transition maps have undergraduate ` property.

12
EY : 20151109 e.g. since Rd −→ Rd , can use undergradate ` property such as continuity or differentiability.

y ◦ x−1 : x(U ∩ V ) ⊆ Rd −→ y(U ∩ V ) ⊆ Rd

x ◦ y −1 : y(U ∩ V ) ⊆ Rd −→ x(U ∩ V ) ⊆ Rd

Philosophy:
Definition 19. An atlas A` is a `-compatible atlas if any two charts in A` are `-compatible.
Definition 20. A `-manifold is a triple ( M, O , A` ) A` ⊆ Amaximal
| {z }
top. mfd.

` undergraduate `
C0 C 0 (Rd −→ Rd ) = continuous maps w.r.t. O
C1 C 1 (Rd −→ Rd ) = differentiable (once) and is continuous
Ck k-times continuously differentiable
Dk k-times differentiable
..
.
C∞ C ∞ (Rd −→ Rd )
⊇

Cω ∃ multi-dim. Taylor exp.

C∞ satisfy Cauchy-Riemann equations, pair-wise

EY : 20151109 Schuller says: C k is easy to work with because you can judge k-times cont. differentiability from
existence of all partial derivatives and their continuity. There are examples of maps that partial derivatives
exist but are not Dk , k-times differentiable.
Theorem 4.1 (Whitney). Any C k≥1 -atlas, AC k≥1 of a topological manifold contains a C ∞ -atlas.

Thus we may w.l.o.g. always consider C ∞ -manifolds, “smooth manifolds”, unless we wish to define Taylor
expandibility/complex differentiability . . .

EY : 20151109 Hassler Whitney 1

Definition 21. A smooth manifold ( M, O , A )
|{z}
| {z }
top. mfd. C ∞ −atlas

γ
R M
x◦γ
x

Rd
EY: 20151109 Schuller was explaining that the trajectory is real in M ; the coordinate
maps to obtain coordinates is x ◦ γ

4.2 Diffeomorphisms
φ
M−
→N

If M, N are naked sets, the structure preserving maps are the bijections (invertible maps).

e.g. {1, 2, 3} −→ {a, b}

Definition 22. M ∼ =set N (set-theoretically) isomorphic if ∃ bijection φ : M −→ N

Examples. N ∼
=set Z
N∼=set Q (EY: 20151109 Schuller says from diagonal counting)
1 http://mathoverflow.net/questions/8789/can-every-manifold-be-given-an-analytic-structure

13
 ∼
=set
N R

Now (M, OM ) ∼
=top (N, ON ) (topl.) isomorphic = “homeomorphic” ∃ bijection φ : M −→ N
φ, φ−1 are continuous.

(V, +, ·) ∼
=vec (W, +w , ·w ) (EY: 20151109 vector space isomorphism) if
∃ bijection φ : V −→ W linearly

finally
Definition 23. Two C ∞ -manifolds
(M, OM , AM ) and (N, ON , AN ) are said to be diffeomorphic if ∃ bijection φ : M −→ N s.t.

φ : M −→ N
φ−1 : N −→ M

are both C ∞ -maps

e−1
ye ◦ φ ◦ x
Rd Re

x
e ye
φ
C∞ M ⊇U V ⊆N

x y
−1
y◦φ◦x
undergraduate C ∞
d
R Re

Theorem 4.2. # = number of C ∞ -manifolds one can make out of a given C 0 -manifolds (if any) - up to
diffeomorphisms.

dimM #
1 1 Morse-Radon theorems
2 1 Morse-Radon theorems
3 1 Morse-Radon theorems
4 uncountably infinitely many
5 finite surgery theory
6 finite surgery theory
..
. finite surgery theory
EY : 20151109 cf. http://math.stackexchange.com/questions/833766/closed-4-manifolds-with-uncountably-many-
The wild world of 4-manifolds

14
5 Tangent Spaces

Lead question: “What is the velocity of a curve γ : R −→ M at the point p of the curve in M ?”

5.1 Velocities

Definition 24. Let (M, O, A) be a smooth manifold. Let there be a curve γ : R −→ M , which is at least C 1 .
Suppose γ(λ0 ) = p. The velocity of γ at the point p of the curve γ is the linear map

∼
vγ,p : C ∞ (M ) −→ R with f 7→ vγ,p (f ) := (f ◦ γ)0 (λ0 ) (5.1)

where C ∞ (M ) := {f : M −→ R | f is a smooth function } equipped with

(f ⊕ g)(p) := f (p) + g(p) and (λ ⊗ g)(p) := λ · g(p) is a vector space.

γ f Figure 5.1: f ◦ γ. Intuition: If the first R is thought of as time,

R M R and f as temperature, then f ◦γ relates time and temperature and
f ◦γ (f ◦ γ)0 is the rate of change of temperature as you run around the
curve.

past: “ vi (∂i f ) = ( vi ∂ )f
|{z} |{z}i
vector in past vector as map
In an imprecise way, we could say that we want vectors to survive as the directional derivatives they induce. This
is a very slight shift of perspective which is extremely powerful and leads to idea of tangent space in differential
geometry.

Terminology: If X is a vector seen as a map, then X acting on a function f , i.e. Xf is called the directional
derivative of f in the X direction.

5.2 Tangent vector space

Definition 25. For each point p ∈ M , the tangent space to M at the point p is the set

Tp M := {vγ,p | for all smooth curves γ through p} (5.2)

Figure 5.2: A pictorial representation of the tangent space Tx M of a single

point, x, on a manifold. A vector in this Tx M can represent a possible
velocity at x. After moving in that direction to a nearby point, one’s velocity
would then be given by a vector in the tangent space of that nearby point —
a different tangent space, not shown. By Alexwright at English Wikipedia
- Transferred from en.wikipedia to Commons by Ylebru., Public Domain
https: // commons. wikimedia. org/ w/ index. php? curid= 3941393

Figure 5.3: The tangent space Tx M and a tangent vector v ∈ Tx M ,

along a curve travelling through x ∈ M . By derivative work: McSush
(talk)Tangentialvektor.png: TNThe original uploader was TN at German
Wikipedia - Tangentialvektor.png, Public Domain, https: // commons.
wikimedia. org/ w/ index. php? curid= 4821938

Caution: Although the Fig. 5.2 and 5.3 refer to an ambient space in which M is embedded, the tangent space
has been defined intrinsically. There is a velocity corresponding to each curve along a different path in M passing

15
through p. Velocity along two different curves could be same, or curves along same paths but having different
parameter speeds would yield different velocities.
Theorem 5.1. (Tp M, ⊕, ⊗) is a vector space with

⊕ :Tp M × Tp M −→ Hom(C ∞ (M ), R)
(vγ,p ⊕ vδ,p )( f ) := vγ,p (f ) +R vδ,p (f )
|{z}
∈C ∞ (M )

 :R × Tp M −→ Hom(C ∞ (M ), R)
(α  vγ,p )(f ) := α ·R vγ,p (f )

Proof. Various conditions that must be satisfied by a vector space, are trivially satisfied. It remains to be shown
that

i) For product, ∃ τ curve : α  vγ,p = vτ,p

ii) For sum, ∃ σ curve : vγ,p ⊕ vδ,p = vσ,p

Product: Let τ : R −→ M with λ 7→ τ (λ) := γ(αλ + λ0 ) = (γ ◦ µα )(λ) where µα : R −→ R with r 7→ α · r + λ0 .

Then τ (0) = γ(λ0 ) = p, and

vτ,p = (f ◦ τ )0 (0) = (f ◦ γ ◦ µα )0 (0) = µ0α (0) · (f ◦ γ)0 (µα (0)) = α · (f ◦ γ)0 (λ0 ) = α · vγ,p

Sum: Choose a chart (U, x) and p ∈ U . (If the proof will depend on the choice of a chart, alarm bells should
ring. But we shall see that the result is finally independent of the chart.)
Let p = γ(λ0 ) = δ(λ1 ).
Now define σ : R −→ M with λ 7→ σ(λ) := x−1 ((x ◦ γ)(λ0 + λ) +(x ◦ δ)(λ1 + λ) − (x ◦ γ)(λ0 )).
| {z }
R−→Rd
Then, σx (0) = x−1 ((x ◦ γ)(λ0 ) + (x ◦ δ)(λ1 ) − (x ◦ γ)(λ0 )) = δ(λ1 ) = p.
Now

vσx ,p (f ) := (f ◦ σx )0 (0)
= ((f ◦ x−1 ) ◦ (x ◦ σx ))0 (0)
| {z } | {z }
Rd −→R R−→Rd
0
· ∂i (f ◦ x−1 ) (x(σ(0)))


= (x ◦ σx ) (0)
| {z } |{z}
(x◦γ)0 (λ0 )+(x◦δ)0 (λ1 ) p

= (x ◦ γ)0 (λ0 )(∂i (f ◦ x−1 ))(x(p)) + (x ◦ δ)(λ1 )(∂i (f ◦ x−1 ))(x(p))

= (f ◦ γ)0 (λ0 ) + (f ◦ δ)0 (λ1 )
= vγ,p (f ) + vδ,p (f ) ∀ f ∈ C ∞ (M )

picture: (cf. https://youtu.be/pepU_7NJSGM?t=39m5s)

If we push γ and δ to one chart, and add them there, then bring the sum back to M , we would get a curve
which would be different from the curve we would get if we used another chart. But it turns out, irrespective
of the charts selected, we get the same tangent/velocity. Conclusion: Adding trajectories is chart dependent;
hence, bad. Adding velocities is good because, whatever the charts, they yield the same derivative at the point
of intersection. Of course, you cannot add two curves (γ ⊕ δ)(λ) := γ(λ) +M δ(λ) because there is no addition
+M in M . Defining + through charts results in chart-dependent results, which is, therefore, not real.

16
5.3 Components of a vector w.r.t. a chart

γ f
R M R

x◦ x −
1
γ ◦x
f
Rd

Let (U, x) ∈ Asmooth , γ : R −→ U and γ(0) = p. Then

vγ,p (f ) := (f ◦ γ)0 (0)

= ((f ◦ x−1 ) ◦ (x ◦ γ))0 (0)
| {z } | {z }
Rd −→R R−→Rd
0
i
0
= (x ◦ γ) (0) · f ◦ x−1
i
(x(p))
0 i
= (x ◦ γ) (0) · (∂i (f ◦ x−1 ))(x(p))


| {z } | {z }
=:γ̇xi (0) =:( ∂fi )
∂x p
 
∂
= γ̇xi (0) · f ∀ f ∈ C ∞ (M ), f : M −→ R
∂xi p

Definition 26. For velocity vγ,p , as a map under use of a chart (U, x),
 
∂
vγ,p = γ̇xi (0) · (5.3)
∂xi p

where
0
i
γ̇xi = (x ◦ γ) (5.4)

are the components of the velocity vγ,p and

 
∂

0  i
= ∂i · ◦ x−1 = · ◦ x−1 (5.5)
∂xi

which eat a function, form a basis of Tp M w.r.t. which the components of the velocity need to be understood.

Note: The components of a vector are always w.r.t. a chart. In M , there is just the vector, no components.
Picture: https://youtu.be/pepU_7NJSGM?t=1h16s
Theorem 5.2. For a chart (U, x),
∂xi
= δji (5.6)
∂xj

Proof.

∂xi
= ∂j (xi ◦ x−1 )(x(p))
∂xj
= δji since xi ◦ x−1 : Rd −→ R s.t. (α1 , . . . , αd ) 7→ αi

17
5.4 Chart-induced basis
   
∂ ∂
Definition 27. If (U, x) ∈ Asmooth , then ,..., ∈ Tp U ⊆ Tp M constitute a chart-induced
∂x1 p ∂xd p
basis of Tp U .
 
∂
Proof. We have already shown that any vector in Tp U can be expressed in terms of . It remains to be
∂xi p
 
∂
shown that they are linearly independent. That is, we require λi = 0 =⇒ λi = 0 for all i = 1, . . . , d.
∂xi p
Or,
 
∂
0 = λi (xj ) xj : U −→ R is differentiable
∂xi p

= λi ∂i (xj ◦ x−1 )(x(p)) by Eq. 5.5

= λi δij by Theorem 5.2
j
=λ for all j = 1, . . . , d

Corollary 1. dim Tp M = d = dim M .

This follows from the fact d vectors are needed to express any vector in Tp M , and these d vectors arise from
the d coordinates of chart which shows that M has d dimensions.
 
∂
Terminology: X ∈ Tp M =⇒ ∃ γ : R −→ M : X = vγ,p and ∃ X 1 , . . . , X d : X = X i . X i are called
| {z } ∂xi p
∈R
components of the vector X w.r.t chart-induced basis.

5.5 Change of vector components under a change of chart

8 A vector does not change under change of chart. It is the vector components that transform under a change
of chart.

Let (U, x) and (V, y) be overlapping charts and p ∈ U ∩ V . Let X ∈ Tp M . Then, X can be expanded in terms
of chart-induced basis of the two charts as follows:
   
i ∂ i ∂
X(y) · = X = X · (5.7)
∂y i p |{z} |{z} (x) ∂xi p
(V,y) (U,x)

Now,
 
∂
f = ∂i (f ◦ x−1 )(x(p))
∂xi p

= ∂i ((f ◦ y −1 ) ◦(y ◦ x−1 )(x(p))

| {z } | {z }
Rd −→R Rd −→Rd
−1 j
= (∂i (y ◦ x ) )(x(p)) · (∂j (f ◦ y −1 ))(y(p))
= (∂i (y j ◦ x−1 ))(x(p)) · (∂j (f ◦ y −1 ))(y(p))
 j  
∂y ∂f
= ·
∂xi p ∂y j p

∂y j
     
∂ ∂
∴ = · (5.8)
∂xi p ∂xi p ∂y j p

18
Using Eq. 5.7 and Eq. 5.8,
∂y j
     
i ∂ j ∂
X(x) = X(y)
∂xi p ∂y j p ∂y j p

∂y j
 
j i
∴ X(y) = X(x) (5.9)
∂xi p

5.6 Cotangent spaces

Since Tp M is a vector space, therefore it is trivial to define cotangent space as follows.

Definition 28. For the tangent space Tp M at p ∈ M , cotangent space is defined as

∼
(Tp M )∗ := {ϕ : Tp M −→ R} (5.10)

Definition 29. If f ∈ C ∞ (M ), then the gradient of f at the point p ∈ M is defined as

∼
(df )p :Tp M −→ R
X 7→ (df )p (X) := Xf (5.11)

i.e. (df )p ∈ Tp M ∗

(df )p is a (0, 1)-tensor over the underlying vector space Tp M . We define the components of the gradient the
same way as we define the components of a tensor (refer section 3.9).
Definition 30. Components of gradient w.r.t. chart-induced basis of (U, x) are defined as
  !  
∂ ∂f
((df )p )j := (df )p = = ∂j (f ◦ x−1 )(x(p)) (5.12)
∂xj p ∂xj p

Theorem 5.3. A chart (U, x) =⇒ xi : U −→ R are smooth functions. Then, (dx1 )p , (dx2 )p , . . . , (dxd )p form
a basis of Tp∗ M .

Proof. In fact, (dxi )p form a dual basis since

 !
∂xa
  
a ∂
(dx )p = = δba (using Theorem 5.2) (5.13)
∂xb p ∂xb p

5.7 Change of components of a covector under a change of chart

8 A covector does not change under change of chart. It is the covector components that transform under a
change of chart.

Let (U, x) and (V, y) be overlapping charts and p ∈ U ∩ V . Let ω ∈ Tp∗ M . Then, ω can be expanded in terms
of chart-induced basis of the two charts as follows:

ω(y)j (dy j )p |{z} = ω(x)i (dxi )p

= ω |{z} (5.14)
(V,y) (U,x)

19
Now,

ω(y)j (dy j )p = ω(x)i (dxi )p by Eq. 5.14

   
∂ ∂
=⇒ ω(y)j (dy j )p = ω(x)i (dxi )p
∂y k p ∂y k p
∂xq
     
j ∂ i ∂
=⇒ ω(y)j (dy )p = ω(x)i (dx )p · by Eq. 5.8
∂y k p ∂y k p ∂xq p
 j  q  i
∂y ∂x ∂x
=⇒ ω(y)j = ω(x)i · by Eq. 5.11
∂y k p ∂y k p ∂xq p
 q
j ∂x
=⇒ ω(y)j δk = ω(x)i · δi by Theorem 5.2
∂y k p q
 i
∂x
=⇒ ω(y)k = ω(x)i
∂y k p

Or, with a change of indices,

∂xj
 
ω(y)i = ω(x)j (5.15)
∂y i p

∂xj
 
ω(y)i = ω(x)j
∂y i p
∂xj
 
i
=⇒ ω(y)i (dy )p = ω(x)j (dy i )p
∂y i p
∂xj
 
=⇒ ω= ω(x)j (dy i )p
∂y i p
∂xj
 
=⇒ ω(dxj )p = ω(x)j (dy i )p (dxj )p
∂y i p
∂xj
 
=⇒ ω(dxj )p = ω(dy i )p
∂y i p
∂xj
 
=⇒ (dxj )p = (dy i )p
∂y i p

∂xj
 
∴ (dxj )p = (dy i )p (5.16)
∂y i p

20
6 Fields

So far, we have focussed technically on a single tangent space and a vector/ covector in it, a basis if we chose a
chart. As physicists, we are interested in things such as vector fields such that at any point of a manifold, there
is a vector. The proper way to deal with it technically is theory of bundles.

6.1 Bundles
π
Definition 31. A bundle is a triple E −→ M , where
E is a smooth manifold, called the total space,
M is a smooth manifold, called the base space, and
π is a smooth map (surjective), called the projection map.

π
Definition 32. Let E −→ M be a bundle and p ∈ M . Then, fibre over p := preimπ ({p}).

π
Definition 33. A section σ of a bundle E −→ M is the map σ : M −→ E such that π ◦ σ = idM .

π
E M

Example: E is a cylinder, M a circle and π maps vertical lines on the cylinder to the point of intersection of
this line with the circle.

Example: If the fibre of p ∈ M is a tangent space, the section would pick one vector from the tangent
space.

Aside: In quantum mechanics, ψ : M −→ C is called a wavefunction, but it is actually a section which selects
one value from C for each p ∈ M .

6.2 Tangent bundle of smooth manifold

For this entire subsection, let (M, O, A) be a smooth manifold and let d := dim M .

Define the set,

˙
[
T M := Tp M (6.1)
p∈M

Now define a surjective map π as follows:

π :T M −→ M
(6.2)
X 7→ π(X) := p ∈ M such that X ∈ Tp M

π
Situation: T M
|{z} −→
|{z} M
|{z}
set surjective map smooth manifold

For a bundle, T M should be a smooth manifold and π a smooth map. Let us construct a topology on T M that
is the coarsest topology such that π is just continuous. (initial topology with respect to π). Define

OT M := {preimπ (U )|U ∈ O} (6.3)

21
It can be shown that (T M, OT M ) is a topological space. But we meed a smooth atlas.

Construction of a C ∞ -atlas on T M from the C ∞ -atlas A on M

Define
AT M :={(T U, ξx ) | (U, x) ∈ A} where
ξx :T U −→ R2d
(6.4)
 

X 7→ (x1 ◦ π)(X), . . . , (xd ◦ π)(X), (dx1 )π(X) (X), . . . , (dxd )π(X) (X)
 
| {z } | {z }
(U,x)− coords of π(X) (d-many) components of X w.r.t (U,x) (d-many)

In the above, (x1 ◦ π)(X) = x1 (π(X)) = x1 (p) = x1 -coordinate, and  

i ∂

j j i ∂

i j j
X ∈ Tπ(X) M =⇒ X = X(x) i
∂x π(X) =⇒ (dx )π(X) (X) = (dx )π(X) X(x) ∂x π(X) = X(x) δi = X(x) .
i

Thus ξx maps X to the coordinates of its base point π(X) under the chart (U, x) and the components of the
vector X w.r.t the basis induced by this chart.

We can write ξx−1 as follows:

ξx−1 : ξx (T U ) −→ T U
| {z }
⊆R2d
(6.5)
 
∂
(α1 , . . . , αd , β 1 , . . . , β d ) := β i
∂xi x−1 (α1 , . . . , αd )
| {z }
π(X)

Now we check, whether the atlas AT M smooth. That is, are the transitions between its charts smooth?
Theorem 6.1. AT M is a smooth atlas.

Proof. Let (U, ξx ) ∈ AT M , (V, ξy ) ∈ AT M and U ∩ V 6= ∅. Calculate the chart transition

  !
∂
(ξy ◦ ξx−1 )(α1 , . . . , αd , β 1 , . . . , β d ) = ξy βi by Eq. 6.5
∂xi x−1 (α1 ,...,αd )
  !   ! !
i m ∂ i m ∂
= . . . , (y ◦ π) β · , . . . , . . . , (dy )x−1 (α1 ,...,αd ) β ,... by Eq. 6.4
∂xm x−1 (α1 ,...,αd ) ∂xm x−1 (α1 ,...,αd )
 
 
 
    !   ! 

i
 m ∂  m i ∂ 
= . . . , y π β ·  , . . . , . . . , (β (dy )x−1 (α1 ,...,αd ) ,...
 
∂xm  ∂xm
  −1 1 d
x (α ,...,α )  −1 1 d
x (α ,...,α ) 
 | {z } | {z } 
 
the base point, x−1 (α1 ,...,αd ) ∂y i
 
∂xm
x−1 (α1 ,...,αd )
! !
i
 
∂y
= . . . , (y i ◦ x−1 )(α1 , . . . , αd ), . . . , . . . , β m ,...
∂xm x−1 (α1 ,...,αd )

= . . . , (y i ◦ x−1 )(α1 , . . . , αd ), . . . , . . . , β m ∂m (y i ◦ x−1 )(x(x−1 (α1 , . . . , αd ))) , . . .

 

= . . . , (y i ◦ x−1 )(α1 , . . . , αd ) , . . . , . . . , β m ∂m (y i ◦ x−1 )(α1 , . . . , αd )

,...
 
| {z } | {z }
smooth ∵(A) is smooth atlas smooth ∵ chart transition map is C ∞ smooth

=⇒ (ξy ◦ ξx−1 ) is smooth =⇒ AT M is smooth

22
Further, the surjective map π is a smooth map because, in the chart representation, π takes the 2d components
of X ∈ T M to the d-coordinates of the base point in M , which can be seen to happen smoothly by seeing how
the components are mapped. Therefore, we have the following definition.
Definition 34. Then, using the smooth manifold (M, O, A) as the base space and the smooth manifold
(T M, OT M , AT M ) as the total space, the tangent bundle is the triple

π
T M −→ M (6.6)

6.3 Vector fields

Why did we put so much effort in making a smooth atlas on T M and defining a tangent bundle? The answer
is in the following definition of smooth vector field, not just any vector field.
π
Definition 35. For a tangent bundle T M −→ M , a smooth vector field χ is a smooth map such that
π ◦ χ = idM .
π
TM M

Remarks: χ is a section, which couldn’t have been a smooth map unless we had both M and T M as smooth
manifolds.

6.4 The C ∞ (M )-module Γ(T M )

We already know that C ∞ (M ), the collection of all smooth functions is a vector space with S-multiplication with
R. But we may also consider the structure (C ∞ (M ), +, ·) with point-wise addition between elements of C ∞ (M )
and point-wise multiplication between elements of C ∞ (M ). This structure satisfies all the requirements of a
field (commutativity, associativity, neutral element, inverse element under both operations, and distributivity)
except that there is no inverse for all non-zero elements under multiplication. This is so because a function that
is not zero everywhere, may be zero at some points and then point-wise multiplication with no function would
result in the value 1 everywhere. Such a structure is called a ring.

A module over a ring is a generalization of the notion of vector space over a field, wherein the corresponding
scalars are the elements of an arbitrary given ring.

Let us consider the module made from the set of all smooth vector fields over the ring C ∞ (M ). Define

Γ(T M ) = {χ : M −→ T M | χ is a smooth section} (6.7)

Definition 36. (Γ(T M ), ⊕, ) is a C ∞ (M )-module over the ring of C ∞ (M ) functions with χ, χ

e ∈ Γ(T M ) and
∞
g ∈ C (M ), such that
(χ ⊕ χ
e)(f ) := (χf ) + (e χf )
|{z}
C ∞ (M )
(g  χ)(f ) := g |{z}
· (χf )
C ∞ (M )

Facts: Besides other differences, there are following 2 important facts:

(1) Proving that every vector space has a basis depends upon the choice of set theory; in particular, on the
Axiom of Choice in ZFC theory.

(2) No such result exists for modules.

23
This is a shame, because otherwise, we could have chosen (for any manifold) vector fields, χ(1) , . . . , χ(d) ∈ Γ(T M )
and would be able to write every vector field χ in terms of component functions f i as χ = f i · χ(i) .

Simple counterexample: Take a sphere. Can we find a smooth vector field over the entire sphere. Can you
comb the sphere? No. For the field to be smooth, there is a problem. Morse Theory tells us that every smooth
vector field on a sphere must vanish at 2 points =⇒ basis cannot be chosen. We cannot choose a global basis.
Therefore, if required, we only expand a vector field in terms of a basis on a domain where it is possible.

Remarks: Although we cannot have a global basis for Γ(T M ), it is possible to do so locally. Thus, for the chart
(U, x) we can take the chart-induced basis of the vector field in the chart domain U as the map

∂ smooth
: U −−−−−→ T U
∂xi 
∂
 (6.8)
p 7→
∂xi p

6.5 Tensor fields

So far we have constructed the sections over the tangent bundle. That is, Γ(T M ) =”set of smooth vector fields”
as a C ∞ (M )-module.

Exactly along the same lines we can construct the cotangent bundle Γ(T ∗ M ) = “set of covector fields” as a
C ∞ (M )-module, by mapping a covector to the coordinates of its base point and components of the covector.
Γ(T M ) and Γ(T ∗ M ) are the basic building blocks for every tensor field.
Definition 37. An (r, s)-tensor field T is a multilinear map

∼
T : Γ(T ∗ M ) × · · · × Γ(T ∗ M ) × Γ(T M ) × · · · × Γ(T M ) −→ C ∞ (M ) (6.9)
| {z } | {z }
r s

Remarks: the multilinearity is in C ∞ (M ), in terms of addition in the modules and S-multiplication with func-
tions in C ∞ (M ).

Example: Let f ∈ C ∞ (M ). Then, define a (0, 1)-tensor field df as

∼
df :Γ(T M ) −→ C ∞ (M )
χ 7→ df (χ) := χf such that (χf )( p ) := χ(p) f
|{z} |{z}
∈M ∈Tp M

It can be checked that df is C ∞ −linear.

24
7 Connections

Motivation: So far, all we have dealt with (e.g., sets, topological manifolds, smooth manifolds, fields,
bundles, etc.) are structures that we have to provide by hand before we can start doing physics as we
know it. Why? Because we don’t have equations which determine what we have done so far. These are
assumptions you need to submit before you can do physics.

In this lecture we introduce yet another structure called connections which are determined by Einstein’s
equations. Everything from now on will be objects that are the subject of Einstein’s equations depending
on the matter in the Universe. Connections are also called covariant derivatives. Even though these are
different, for our purposes we shall not distinguish the two and use the more general connections.

So far, we saw that a vector field X can be used to provide a directional derivative of a function f ∈ C ∞ (M )
in the direction X

∇X f := Xf

Isn’t this a notational overkill? We already know

∇X f = Xf = (df )X

Actually, they are not quite the same because

X : C ∞ (M ) −→ C ∞ (M )
df : Γ(T M ) −→ C ∞ (M )
∇X : C ∞ (M ) −→ C ∞ (M )

where ∇X can be generalized to eat an arbitrary (p, q)-tensor field and yield a (p, q)-tensor field whereas X can
only eat functions.

∇X : C ∞ (M ) C ∞ (M )

.. ..
. .
∇X : (p, q)-tensor field (p, q)-tensor field

We need ∇X to provide the new structure to allow us to talk about directional derivatives of tensor fields and
vector fields. Of course, only in cases where ∇X acts on function f which is a (0, 0)-tensor, it is exactly the
same as Xf .

7.1 Directional derivatives of tensor fields

We formulate a wish list of properties which ∇X acting on a tensor field should have. We put this in form of
a definition. There may be many structures that satify this wish list. Any remaining freedom in choosing such
a ∇ will need to be provided as additional structure beyond the structure we already have. And we assume all
this takes place on a smooth manifold.
Definition 38. A connection ∇ on a smooth manifold (M, O, A) is a map that takes a pair consisting of a
vector (field) X and a (p, q)-tensor field T and sends them to a (p, q)-tensor (field) ∇X T satisfying

i) ∇X f = Xf ∀f ∈ C ∞ M

ii) ∇X (T + S) = ∇X T + ∇X S where T, S are (p, q)-tensors

25
iii) Leibnitz rule: ∇X T (ω1 , . . . , ωp , Y1 , . . . , Yq ) = (∇X T )(ω1 , . . . , ωp , Y1 , . . . , Yq )
+ T (∇X ω1 , . . . , ωp , Y1 , . . . , Yq ) + · · · + T (ω1 , . . . , ∇X ωp , Y1 , . . . , Yq )
+ T (ω1 , . . . , ωp , ∇X Y1 , . . . , Yq ) + · · · + T (ω1 , . . . , ωp , Y1 , . . . , ∇X Yq ) where T is a (p, q)-tensor

Note that for a (p, q)-tensor T and a (r, s)-tensor S, since:

(T ⊗ S)(ω(1) , . . . , ω(p+r) , Y(1) , . . . , Y(q+s) ) =
T (ω(1) , . . . , ω(p) , Y(1) , . . . , Y(q) ) · S(ω(p+1) , . . . , ω(p+r) , Y(q+1) , . . . , Y(q+s) ),
Leibnitz rule implies ∇X (T ⊗ S) = (∇X T ) ⊗ S + T ⊗ (∇X S).

iv) C ∞ -linearity: ∀f ∈ C ∞ (M ), ∇f X+Z T = f ∇X T + ∇Z T

C ∞ -linearity means that no matter how the function f scales the vectors at different points of the
manifold, the effect of the scaling at any point is independent of scaling in the neighbourhood and
depends only on how the scaling happens at that point.

A manifold with a connection ∇ is a quadruple (M, O, A, ∇), where M is a set, O is a topology and A is a
smooth atlas.

Remark: If ∇X (·) can be seen as an extension of X,

then ∇(·) (·) can be seen as an extension of d.

7.2 New structure on (M, O, A) required to fix ∇

How much freedom do we have in choosing such a structure?

Consider vector fields X, Y and chart (U, x) ∈ A. Then



∂ m
∇X Y = ∇( X i ∂ ) Y by expanding in chart-induced basis
∂xi ∂xm
 
m ∂
i
= X · ∇( ∂ ) Y by C ∞ -linearity
∂xi ∂xm
 ∂  
i

m i m ∂
= X ∇( ∂ ) Y m
+X ·Y · ∇( ∂ ) m using Leibnitz rule
} ∂x ∂xi ∂x
i
| ∂x{z | {z }
∂
= Ym a vector field, by defn.
∂xi
   
∂ m ∂ ∂
=X i
Y + X i · Y m · Γqmi q
∂xi ∂xm ∂x

Thus, by change of indices,

 
i ∂
(∇X Y ) = X m Yi + X m · Y n · Γinm (7.1)
∂xm

So we need (dim M )3 -many functions to define directional derivative of a vector field.

Definition 39. Given (M, O, A, ∇) and (U, x) ∈ A, then the connection coefficient functions (Γs) on M
of ∇ w.r.t (U, x) are (dim M )3 -many functions given by

Γijk : U −→ R
  
i i ∂
p 7→ Γ jk (p) := dx ∇( ∂
) (p) (7.2)
∂x k ∂xj

∂ ∂ i
Note: ∂x j is a vector field; ∴ ∇
( ∂x∂k ) ∂xj is a vector field, and dx is a covector which will result in a function
after acting on a vector field.

26
On a chart domain U , choice of the (dim M )3 -many functions Γijk suffices to fix the action of ∇ on a vector
field. What about the directional derivative of a covector field, or a tensor field? Will we have to provide
more and more coefficients? Fortunately, the same (dim M )3 -many functions fix the action of ∇ on any tensor
field.


We know that, for a covector, ∇ ∂m dxi = Σijm dxj , since dxi form a dual basis. Are these Σs independent
∂x
of Γs? Consider the following.
  
∂ ∂
∇ ∂ dxi
= ∇ ∂m δji = (δ i ) = 0
∂xm ∂xj ∂x ∂xm j
  ∂  
∂

=⇒ ∇ ∂m dxi + dx i
∇ ∂ =0
∂x ∂xj ∂xm ∂xj
| {z }
Γqjm ∂x
∂
q

  ∂  ∂
=⇒ ∇ ∂m dxi j
+ dxi Γqjm q = 0
∂x ∂x ∂x
  ∂  ∂ ∂
=⇒ ∇ ∂m dxi j
= −dxi Γqjm q = −Γqjm dxi q = −Γqjm δqi = −Γijm
∂x ∂x ∂x ∂x
  ∂ 
=⇒ ∇ ∂m dxi dxj = −Γijm dxj
∂x ∂xj
| {z }
=δjj =1

=⇒ ∇ ∂ dxi = −Γijm dxj

∂xm

 
=⇒ ∇ ∂ dxi = −Γijm
∂xm j

In summary,

i
(∇X Y ) = X(Y i ) + Γijm Y j X m (7.3)
j m
(∇X ω)i = X (ωi ) − Γ im ωj X (7.4)

Note that for the immediately above expression for (∇X Y )i , in the second term on the right hand side, Γijm
has the last entry at the bottom, m going in the direction of X, so that it matches up with X m . This is a good
mnemonic to memorize the index positions of Γ.

Similarly, as an example, by further application of Leibnitz rule, for a (1, 2)-tensor field T ,

i
(∇X T ) jk = X T ijk + Γism T sjk X m − Γsjm T isk X m − Γskm T ijs X m

7.3 Change of Γ’s under change of chart

Let (U, x), (V, y) ∈ A and U ∩ V 6= ∅.

 

i i ∂
Γ(y) := dy ∇ ∂
jk ∂y k ∂y j
∂y i q ∂xs ∂
 
= dx ∇ ∂xp ∂p j
∂xq ∂y k ∂x ∂y ∂x
s

i
 p  s
∂xs
  
∂y ∂x ∂x ∂ ∂
= dx q
∇ ∂p j + j ∇ ∂p s ∵ ∇ is C ∞ − linear
∂xq ∂y k ∂x ∂y ∂xs ∂y ∂x ∂x

∂y i ∂xp ∂ ∂xs q ∂y i ∂xp ∂xs
q

= q k p j
δs + q k j
Γ(x) sp
∂x ∂y ∂x ∂y ∂x ∂y ∂y
| {z }
∂
∂y k

27

i ∂y i ∂ 2 xq ∂y i ∂xs ∂xp
q
Γ(y) jk
= q k j
+ q j k
Γ(x) sp (7.5)
∂x ∂y ∂y ∂x ∂y ∂y

Eq. (7.5) is the change of connection coefficient function under the change of chart (U ∩ V, x) −→ (U ∩ V, y). Γ
is not a tensor due to the first term on left hand side in Eq. (7.5). However, for linear transformation between
2 q
coordinates in two charts, the term ∂y∂k x∂yj always vanishes and then, if Γs are zero in one chart, they will
be zero in the other chart too. However, there is no reason not to select a coordinate which is not a linear
transformation of another one.

7.4 Normal Coordinates

Can we find a coordinate system that makes the Γs vanish?

Theorem 7.1. Let p ∈ M of (M, O, A, ∇). Having chosen a point p, one can construct a chart (U, x) with
p ∈ U such that the symmetric part of Γs vanish at the point p (not necessarily in any neighbourhood). That is,

i
∀ p ∈ M, ∃ (U, x) ∈ A : p ∈ U and Γ(x) (jk) (p) = 0.
Such (U, x) is called a normal coordinate chart of ∇ at p ∈ M .

Proof. Let (V, y) ∈ A and p ∈ V . Then consider a new chart (U, x) to which one transits using the map (x◦y −1 )
whose ith component is given by

i
i
x ◦ y −1 α1 , . . . , αd := αi − Γ(y) (jk) αj αk


where the Γs are taken at the point p
∂xi
i
= ∂j xi ◦ y −1 = δji − Γ(y) (jm) αm


=⇒ j
∂y
∂ 2 xi
i
=⇒ k j
= − Γ(y) (jk)
∂y ∂y

TODO: Not understood by me.

28
8 Parallel Transport & Curvature

8.1 Parallelity of vector fields

Definition 40. Let (M, O, A, ∇) be a smooth manifold with connection ∇.

(1) A vector field X on M is said to be parallely transported along a smooth curve γ : R −→ M if

∇vγ X = 0 (8.1)

To make explicit, how this equation applies along the curve, we may state


∇vγ,γ(λ) X γ(λ)
=0

(2) A slightly weaker condition is “parallel” if, for µ : R −→ R,

∇vγ,γ(λ) X γ(λ)
= µ(λ)Xγ(λ) (8.2)

Remarks: Even though parallely transported sounds like an action, it is a property.

8.2 Autoparallely transported curves

Definition 41. A curve γ : R −→ M is called autoparallely transported if

∇v γ vγ = 0 (8.3)

Remarks: Sometimes, this curve is called an autoparallel curve. But we wish to call a curve autoparallel if
∇vγ vγ = µvγ .

8.3 Autoparallel equation

Express ∇vγ vγ = 0 in terms of chart representation.

0 = ∇v γ vγ
 
 γ̇ n
∂ m
= ∇γ̇ m ∂ (x) remember that γ(x) := xm ◦ γ
(x) ∂x m ∂xn
 ∂  
 ∂
= γ̇ m ∇( ∂m ) γ̇ n + γ̇ m n
γ̇ ∇ ( ∂x∂m ) ∂xn x index is understood, hence suppressed
∂x ∂xn
   
∂ n ∂ ∂
= γ̇ m γ̇ + γ̇ m n
γ̇ ∇ ( ∂x∂m ) ∂xn
∂xm ∂xn
   
∂ q ∂ ∂
= γ̇ m γ̇ + γ̇ m n
γ̇ Γ q
nm change of index in 1st term
∂xm ∂xq ∂xq
 
∂ ∂
= γ̇ m m γ̇ q + γ̇ m γ̇ n Γqnm
∂x ∂xq
∂
= (γ̈ q + γ̇ m γ̇ n Γqnm ) q TODO: show that 1st term is 2nd derivative
∂x

In summary:
q
γ̈(x) (λ) + (Γ(x) )qmn (γ(λ))γ̇(x)
m n
(λ)γ̇(x) (λ) = 0 (8.4)

29
Eq. (8.4) is the chart expression of the condition that γ be autoparallely transported.

i
Example: (a) In Euclidean plane having a chart (U = R2 , x = idR2 ), Γ(x) jk = 0
m m
=⇒ γ̈(x) = 0 =⇒ γ(x) (λ) = am λ + bm , where a, b ∈ Rd .

(b) Consider the round sphere (S 2 , O, A, ∇round ), i.e., the sphere (S 2 , O, A) with the connection ∇round . Con-
sider the chart x(p) = (θ, φ) where θ ∈ (0, π) and φ ∈ (0, 2π). In this chart ∇round is given by

1
x−1 (θ, φ) := − sin θ cos θ


Γ(x) 22

2
2
x−1 (θ, φ) = x−1 (θ, φ) := cot θ



Γ(x) 12
Γ(x) 21

All other Γs vanish. Then, using the sloppy notation (familiar to us from classical mechanics) i.e., x1 (p) = θ(p)
and x2 (p) = φ(p), the autoparallel equation is
)
θ̈ + Γ122 φ̇φ̇ = 0 θ̈ − sin θ cos θφ̇φ̇ = 0
=⇒
φ̈ + 2Γ212 θ̇φ̇ = 0 φ̈ + 2 cot θθ̇φ̇ = 0

It can be seen that the above equations are satisfied at the equator where θ(λ) = π/2, and φ(λ) = ωλ + φ0
(running around the equator at constant speed ω). Thus, this curve is autoparallel. However, φ(λ) = ωλ2 + φ0
wouldn’t be autoparallel.

8.4 Torsion

Can we use ∇ to define tensors on (M, O, A, ∇)?

Definition 42. The torsion of a connection ∇ is the (1, 2)-tensor field

T (ω, X, Y ) := ω(∇X Y − ∇Y X − [X, Y ]) (8.5)

where [X, Y ], called the commutator of X and Y is a vector field defined by [X, Y ]f := X(Y f ) − Y (Xf ).

Proof. We shall check that T is C ∞ -linear in each entry.

T (f ω, f X, Y ) = f ω(∇X Y − ∇Y (X) − [X, Y ])

= f T (ω, X, Y )
T (ω + ψ, X, Y ) = (ω + ψ)(∇X Y − ∇Y (X) − [X, Y ])
= T (ω, X, Y ) + T (ψ, X, Y )
T (ω, f X, Y ) = ω(∇f X Y − ∇Y (f X) − [f X, Y ])
= ω(f ∇X Y − (∇Y (f ))X − f (∇Y X) − [f X, Y ])
= ω(f ∇X Y − (Y f )X − f (∇Y X) − [f X, Y ])
But [f X, Y ]g = f X(Y g) − Y (f X)g = f X(Y g) − (Y f )(Xg) − f Y (Xg) =⇒ [f X, Y ] = f [X, Y ] − (Y f )X
∴ T (ω, f X, Y ) = ω(f ∇X Y − (Y f )X − f (∇Y X) − f [X, Y ] + (Y f )X)
= ω(f ∇X Y − f (∇Y X) − f [X, Y ])
= f ω(∇X Y − (∇Y X) − [X, Y ]) = f T (ω, X, Y )
Further, T (ω, X, Y ) = −T (ω, Y, X), which means scaling in the last factor need not be checked separately.

Additivity in the last two factors can also be checked.

Definition 43. A (M, O, A, ∇) is called torsion-free if the torsion of its connection is zero. That is, T = 0.

30
In a chart
 
∂ ∂
T iab := T dxi , a , b = dxi (. . . ) = Γiab − Γiba = 2Γi[ab]
∂x ∂x

From now on, in these lectures, we only use torsion-free connections.

8.5 Curvature

Definition 44. Riemann curvature of a connection ∇ is the (1, 3)-tensor field

Riem(ω, Z, X, Y ) := ω(∇X ∇Y Z − ∇Y ∇X Z − ∇[X,Y ] Z) (8.6)

Proof. It can be shown that C ∞ -linear in each slot. This has been left as an exercise to the reader.

Algebraic relevance of Riem: We ask whether there is difference in applying the two directional derivatives
in different order, i.e.
∇X ∇Y Z − ∇Y ∇X Z = Riem(·, Z, X, Y ) + ∇[X,Y ] Z
In one chart (U, x), denoting ∇ ∂a by ∇a ,
∂x
m m
(∇a ∇b Z) − (∇b ∇a Z) = Riemmnab Z n + ∇  ∂ ∂
 Z
,
∂xa ∂xb
| {z }
=0, since they commute
As the last term vanishes, we can see how the Riem tensor components contain all the information about how
the ∇a and ∇b fail to commute if they act on a vector field. If they act on a tensor field, there are several
terms on RHS like the one term above; if they act on a function, of course they commute. Being a tensor, Riem
vanishes in all coordinate systems if it vanishes in one coordinate system, as it does in flat spaces.

Geometric significance of Riem:

Figure 8.1: If we parallel transport a vector u at p to q along two dif-

ferent paths vw and wv, the resulting vectors at q are different in gen-
eral. If, however, we parallel transport a vector in a Euclidean space,
where the parallel transport is defined in our usual sense, the resulting
vector does not depend on the path along which it has been parallel trans-
ported. We expect that this non-integrability of parallel transport char-
acterizes the intrinsic notion of curvature, which does not depend on the
special coordinates chosen. From an answer of Sepideh Bakhoda [1] on
http: // math. stackexchange. com/ q/ 465672

For small v and w, if T = 0, (δu)m = Riemmnab v a wb un + O(v 2 w, vw2 ).

31
9 Newtonian spacetime is curved!

Axiom 1 (Newton I:). A body on which no force acts moves uniformly along a straight line.
Axiom 2 (Newton II:). Deviation of a body’s motion from such uniform straight motion is effected by a force,
reduced by a factor of the body’s reciprocal mass.

Remarks:

(1) 1st axiom - in order to be relevant - must be read as a measurement prescription for the geometry of space.
If somehow, we know that no force acts on a particle, we know that the path it takes is a straight line –
thus, we learn about the geometry of space. After all, unlike in maths, there is no obvious way to tell what
is a straight line. Remember, if we don’t know what a straight line is, we don’t know what a deviation
from a straight line is.

(2) Since gravity universally acts on every particle, in a universe with at least two particles, gravity must not
be considered a force if Newton I is supposed to remain applicable.

9.1 Laplace’s questions

Question: Can gravity be encoded in a curvature of space, such that its effects show if particles under the
influence of (no other) force we postulated to more along straight lines in this curved space?

Answer: No!

Proof. Gravity, as a force point of view:

mẍα (t) = mf α (x(t))

| {z }
f orce:F α

where −∂α f α = 4πGρ (Poisson); ρ = mass density of matter.

The same m appearing on both sides of the equation is an experimental fact, also known as the weak equiva-
lence principle.
∴ ẍα (t) − f α (x(t)) = 0

Laplace asks: Is this (ẍ(t)) of the form ẍα (t) + Γαβγ (x(t))ẋβ (t)ẋγ (t) = 0? That is, does it take the form of
autoparallel equation?

No. Because the Γ can only depend on the point x where you are, but the velocities ẋβ (t) and ẋγ (t) can take
any value and therefore the Γs cannot take care of the f α in the preceding equation. Had there been such Γs,
we would be able to find the notion of straight line that could have absorbed the effect that we usually attribute
to a force.

Conclusion: One cannot find Γ s such that Newton’s equation takes the form of an autoparallel equation.

9.2 The full wisdom of Newton I

Laplace asked: Can we find a curvature of space such that particles move along straight lines?

Use the information from Newton’s first law that particles (under influence of no force) move not just in straight
line, but also uniformly. A curve, after all, is not just a set of points, but also how their parameter is associated
with the points.

32
Introduce the appropriate setting to talk about the difference easily. How? We use spacetime instead of just
space. By using the extra coordinate viz. time, we do not need to keep track of the curve parameter since we
can just refer to time to ascertain uniformity of the motion.

Insight: Uniform & straight motion in space is simply straight motion in spacetime. We do not need to say
uniform. This can be seen by drawing the path of the particle in a t-x graph, wherein straight line results only
when the motion is uniform. So let’s try in spacetime:


Let x : R −→ R3   worldline (history) of the particle

be a particle’s ←→ { X :R −→ R4

t 7→ (t, x1 (t), x2 (t), x3 (t)) := (X 0 (t), X 1 (t), X 2 (t), X 3 (t))

trajectory in space


That’s all it takes. Let us assume that x : R −→ R3 satisfies Newton’s law concerning gravitational force, i.e.
we can omit m on both sides of the equation ẍα = −f α (x(t)).
Trivial rewritings:
Ẋ 0 = 1
Ẍ 0 =0 a = 0, 1, 2, 3

=⇒ Ẍ α − f α (X(t)) · Ẋ 0 · Ẋ 0 =0 =⇒ Ẍ a + Γabc Ẋ b Ẋ c = 0
| {z }
(α=1,2,3) autoparallel eqn. in spacetime

!
Yes, choosing Γ0ab = 0, Γαβγ = 0 = Γα0β = Γαβ0 . Only Γα00 = −f α .

Question: Is this a coordinate-choice artifact?

No, since Rα0β0 = − ∂x∂ β f α (only non-vanishing components) (tidal force tensor, − the Hessian of the force
component)

Ricci tensor =⇒ R00 = Rm0m0 = −∂α f α = 4πGρ

Poisson: −∂α f α = 4πG · ρ

writing: T00 = 21 s
=⇒ R00 = 8πGT00
hhh((( (
Einstein in 1912 (
Rab
(= 8πGT
(h hhab
h
Conclusion: Laplace’s idea works in spacetime

Remark
Γα00 = −f α
Rαβγδ = 0 α, β, γ, δ = 1, 2, 3
R00 = 4πGρ

Q: What about transformation behavior of LHS of

ẍa + Γabc Ẋ b Ẋ c = 0
| {z }
(∇vX vX )a
| {z }
:=aa “acceleration vector”

9.3 The foundations of the geometric formulation of Newton’s axiom

Definition 45. A Newtonian spacetime is a quintuple (M, O, A, ∇, t) where (M, O, A) is a 4-dimensional

smooth manifold, and

33
t : M −→ R smooth function

(i) “There is an absolute space” (dt)p 6= 0 ∀p ∈ M

(ii) “Absolute time flows uniformly”

∇dt
|{z} =0 everywhere
(0,2)-tensor field

(iii) add to axioms of Newtonian spacetime ∇ = 0 torsion free

Definition 46. Absolute space at time τ

dt6=0 a
Sτ := {p ∈ M |t(p) = τ } −−−→ M = Sτ

Definition 47. A vector X ∈ Tp M is called

(a) future-directed, if dt(X) > 0

(b) spatial, if dt(X) = 0

(c) past-directed, if dt(X) < 0

Picture Newton I: The worldline of a particle under the influence of no force (gravity isn’t one, anyway) is a
future-directed autoparallel i.e.
∇v X vX = 0
dt(vX ) > 0

Newton II:
F
∇v X vX = m ⇐⇒ m · a = F
where F is a spatial vector field: dt(F ) = 0.

Convention: restrict attention to atlases Astratif ied whose charts (U, x) have the property

x0 : U −→ R
x1 : U −→ R 0
“absolute time flows uniformly”
= ∇dt
.. .. x0 = t|U =⇒
0
. . 0=∇ ∂ dx = −Γ0ba a = 0, 1, 2, 3
∂xa

x3

Let’s evaluate in a chart (U, x) of a stratified atlas Asheet : Newton II:

F
∇v X vX = m
in a chart.

(X 0 )00 + (
Γ0( )0 (X b )0stratified atlas = 0
a ( ((
cd (X
((
0 0 0 0 0 0 Fα
(X α )00 + Γαγδ X γ X δ + Γα00 X 0 X 0 + 2Γαγ0 X γ X 0 = α = 1, 2, 3
m

=⇒ (X 0 )00 (λ) = 0 =⇒ X 0 (λ) = aλ + b constants a, b with

stratified
X 0 (λ) = (x0 ◦ X)(λ) = (t ◦ X)(λ)
convention parametrize worldline by absolute time

d d
=a
dλ dt

34
 Fα
a2 Ẍ α + a2 Γαγδ Ẋ γ Ẋ δ + a2 Γα00 Ẋ 0 Ẋ 0 + 2Γαγ0 Ẋ γ Ẋ 0 =
m
α α γ δ α 0 0 α γ 0 1 Fα
=⇒ Ẍ + Γ γδ Ẋ Ẋ + Γ 00 Ẋ Ẋ + 2Γ γ0 Ẋ Ẋ = 2
| {z } a m
aα

35
10 Metric Manifolds

We establish a structure on a smooth manifold that allows one to assign vectors in each tangent space a length
(and an angle between vectors in the same tangent space). From this structure, one can then define a notion of
length of a curve. Then we can look at shortest curves (which will be called geodesics).

Requiring then that the shortest curves coincide with the straight curves (w.r.t. ∇) will result in ∇ being
determined by the metric structure g. ∇, in turn determines the curvature given by Riem. Thus

straight=shortest/
longest/ stationary curves
T =0
g ∇ Riem

10.1 Metrics

Definition 48. A metric g on a smooth manifold (M, O, A) is a (0, 2)-tensor field satisfying

(i) symmetry: g(X, Y ) = g(Y, X) ∀ X, Y vector fields

(ii) non-degeneracy: the musical map

“flat” [ : Γ(T M ) −→ Γ(T ∗ M )

X 7→ [(X)
where [(X)(Y ) := g(X, Y )
[(X) ∈ Γ(T ∗ M )

In thought bubble: [(X) = g(X, ·)

. . . is a C ∞ -isomorphism in other words, it is invertible.

Remark: ([(X))a or
Xa
([(X))a := gam X m

Thought bubble: [−1 = ]

[−1 (ω)a := g am ωm
00
[−1 (ω)a := (g “−1 )am ωm =⇒ not needed. (all of this is not needed)
00
Definition 49. The (2, 0)-tensor field g “−1 with respect to a metric g is the symmetric
00
g “−1 : Γ(T ∗ M ) × Γ(T ∗ M ) →
− C ∞ (M )
(ω, σ) 7→ ω([−1 (σ)) [−1 (σ) ∈ Γ(T M ))

chart: gab = gba (g −1 )am gmb = δba

Example: Consider (S 2 , O, A) and the chart (U, x)

ϕ ∈ (0, 2π), θ ∈ (0, π)

Define the metric

" #
−1 R2 0
gij (x (θ, ϕ)) =
0 R2 sin2 θ
ij

36
R ∈ R+

“the metric of the round sphere of radius R”

10.2 Signature
 
λ1 0
Aam v m = λv a
 .. 

 . 

0 λn
 
1
 .. 

 . 

 
Linear algebra:

 1 


 −1 

gam v m a
=λ·v ?
 .. 

 . 

 

 −1 


 0 

 .. 

 . 

0

(1, 1) tensor has eigenvalues

(0, 2) has signature (p, q) (well-defined)

(+ + +) 


(+ + −)

d + 1 if p + q = dimV
(+ − −) 



(− − −)
Definition 50. A metric is called Riemannian if its signature is (++· · · +), and Lorentzian if it is (+−· · · −).

10.3 Length of a curve

Let γ be a smooth curve. Then we know its veloctiy vγ,γ(λ) at each γ(λ) ∈ M .
Definition 51. On a Riemannian metric manifold (M, O, A, g), the speed of a curve at γ(λ) is the number
q
s(λ) = ( g(vγ , vγ ))γ(λ) (10.1)

(I feel the need for speed, then I feel the need for a metric)

Aside: [v a ] = T1
[gab ] = L2 q
2
p
[ gab v a v b ] = TL2 = TL
Definition 52. Let γ : (0, 1) −→ M a smooth curve. Then the length of γ, L[γ] ∈ R is the number

Z 1 Z 1 q
L[γ] := dλs(λ) = dλ (g(vγ , vγ ))γ(λ) (10.2)
0 0

F. Schuller: “velocity is more fundamental than speed, speed is more fundamental than length”

37
Example: Reconsider the round sphere of radius R. Consider its equator:
π
θ(λ) := (x1 ◦ γ)(λ) = , ϕ(λ) := (x2 ◦ γ)(λ) = 2πλ3
2
=⇒ θ0 (λ) = 0, ϕ0 (λ) = 6πλ2

" #
R2
On the same chart gij =
R2 sin2 θ

Do everything in this chart

Z 1 q
L[γ] = dλ gij (x−1 (θ(λ), ϕ(λ)))(xi ◦ γ)0 (λ)(xj ◦ γ)0 (λ)
0
Z 1 q
= dλ R2 · 0 + R2 sin2 (θ(λ))36π 2 λ4
0
Z 1
1
= 6πR dλλ2 = 6πR[ λ3 ]10 = 2πR
0 3

Theorem 10.1. γ : (0, 1) −→ M and σ : (0, 1) −→ (0, 1) smooth bijective and increasing “reparametrization”
L[γ] = L[γ ◦ σ]

Proof. in Tutorials

10.4 Geodesics

Definition 53. A curve γ : (0, 1) −→ M is called a geodesic on a Riemannian manifold (M, O, A, g) if it is a

stationary curve with respect to a length functional L.

Thought bubble: In classical mechanics, deform the curve a little,  times this deformation, to first order, it
agrees with L[γ].
Theorem 10.2. γ is geodesic iff it satisfies the Euler-Lagrange equations for the Lagrangian

L :T M −→ R
p
X 7→ g(X, X)

In a chart, the Euler Lagrange equations take the form:

 ·
∂L ∂L
− =0
∂ ẋm ∂xm

F.Schuller: this is a chart dependent formulation

here: q
L(γ i , γ̇ i ) = gij (γ(λ))γ̇ i (λ)γ̇ j (λ)

Euler-Lagrange equations:

∂L 1
= √ gmj (γ(λ))γ̇ j (λ)
∂ γ̇ m ...
 · !·
∂L 1 1
gmj (γ(λ)) · γ̇ j (λ) + √ gmj (γ(λ))γ̈ j (λ) + γ̇ s (∂s gmj )γ̇ j (λ)


= √
∂ γ̇ m ... ...

Thought bubble: reparametrize g(γ̇, γ̇) = 1 (it’s a condition on my reparametrization)

38
By a clever choice of reparametrization ( √1... )· = 0

∂L 1
= √ ∂m gij (γ(λ))γ̇ i (λ)γ̇ j (λ)
∂γ m 2 ...

putting this together as Euler-Lagrange equations:

1
gmj γ̈ j + ∂s gmj γ̇ s γ̇ j − ∂m gij γ̇ i γ̇ j = 0
2
−1 1
γ¨ + (g ) (∂i gmj − ∂m gij )γ̇ i γ̇ j = 0
q qm
(multiply on both sides (g −1 )qm )
2
1
γ¨q + (g −1 )qm (∂i gmj + ∂j gmi − ∂m gij )γ̇ i γ̇ j = 0
2

geodesic equation for γ in a chart.

1
(g −1 )qm (∂i gmj + ∂j gmi − ∂m gij ) =: Γqij (γ(λ))
2
 ·  
∂L ∂L
Thought bubble: a+dimM
∂ξx
− ∂xia =0
σ(x) x σ(x)
L.C.
Definition 54. Christoffel symbol Γ are the connection coefficient functions of the so-called Levi-Civita
connection L.C. ∇

We usually make this choice of ∇ if g is given.

L.C.
(M, O, A, g) −→ (M, O, A, g, ∇)

abstract way: ∇g = 0 and T = 0 (torsion)

=⇒ ∇ = L.C. ∇
Definition 55. (a) The Riemann-Christoffel curvature is defined by

Rabcd := gam Rmbcd (10.3)

(b) Ricci
Rab = Rmamb (10.4)
L.C.
Thought bubble: with a metric, ∇

(c) (Ricci) scalar curvature:

R = g ab Rab (10.5)

Thought bubble: L.C. ∇

Definition 56. Einstein curvature of (M, O, A, g) is defined as

1
Gab := Rab − gab R (10.6)
2

00
Convention: g ab := (g “−1 )ab

F. Schuller: these indices are not being pulled up, because what would you pull them up with

(student) Question: Does the Einstein curvature yield new information?

Answer:
g ab Gab = Rab g ab − 12 gab g ab R = R − δaa R = R − 12 dimM R = (1 − d2 )R

39
11 Symmetry

This lecture is about symmetry but we will pick a number of elementary techniques in differential geometry
that we will need in Einstein’s theory. We shall motivate these techniques by appealing to the feeling
that the round sphere (S 2 , O, A, g round ) has rotational symmetry, while the potato (S 2 , O, A, g potato ) does
not.

So far we have considered symmetry by having inner product first, and then demanding that w.r.t. that
inner product we classify linear maps A acting on vectors X and Y such that inner product of AX and AY
results in inner product XY .

Here we talk about an altogether different idea. Firstly, since the distinction between the two is entirely
contained in g, we are talking about the rotational symmetry of g round . Secondly, while an inner product is
on one tangent space, there are many different tangent spaces with different inner products. g talks about
the distribution of these inner products over the sphere, and that distribution in some sense is rotationally
invariant or not.

Therefore, the question is: How to describe the symmetries of a metric? This is important because nobody
has solved Einstein’s Equations without assuming some sort of additional assumptions such as symmetry
of the solution.

11.1 Push-forward map

Definition 57. Let M and N be smooth manifolds with tangent bundles T M and T N respectively. Let
φ : M −→ N be a smooth map. Then, the push-forward map of φ is the map

φ∗ :T M −→ T N
X 7→ φ∗ (X)
where φ∗ (X)f := X(f ◦ φ) (∀ f ∈ C ∞ (N )) (11.1)

φ∗
TM TN

πT M πT N
Figure 11.1: Push-forward map: φ∗ takes a vector X ∈ Tp M
φ f in the tangent space at the point p ∈ M to the vector φ∗ (X) ∈
M N R Tq N in the tangent space at the point φ(p) = q ∈ N , such that
f ◦φ the action of φ∗ (X) on any smooth function f ∈ C ∞ (N ) results
in the same value as the action of X on the function (f ◦ φ).

Note: If we take an entire fibre at the point p ∈ M , applying φ∗ on it remains within the fibre at the point
φ(p) ∈ N . That is

φ∗ (Tp M ) ⊆ Tφ(p) N

Mnemonic: “vectors are pushed forward” across tangent bundles in a manner dictated by the underlying
map.

Components of push-forward φ∗ w.r.t charts (U, x) ∈ AM and (V, y) ∈ AN : Let p ∈ U and φ(p) ∈ V .
∂ ∂
Since ∂x i
p
is a vector, we have φ∗ ( ∂x i ) as a vector in N. Then we can select a component of this vector by
p
using dy a as follows:

40
  !!  ! !
∂ φ̂a
     
a ∂ ∂ a ∂ a ∂ a
dy φ∗ = φ∗ y = (y ◦ φ) = (y ◦ φ) := (11.2)
∂xi p ∂xi p ∂xi p ∂xi p ∂xi
p
| {z }
:=φa
∗i

φ
M ⊇U V ⊆N
y◦
φ=
x : φˆ y

x(U ) y(V )
| {z } | {z } Figure 11.2: Components of push-forward map w.r.t charts (U, x) ∈ AM
⊆Rdim M
y ◦ φ ◦ x−1 ⊆Rdim N
and (V, y) ∈ AN .
Theorem 11.1. If γ : R −→ M is a curve in M , then φ ◦ γ : R −→ N is a curve in N . Then, φ∗ pushes the
tangent to a curve γ (velocity) to the tangent to the curve (φ ◦ γ), i.e.,

φ∗ (vγ,p ) = v(φ◦γ),φ(p) (11.3)

Proof. Let p = γ(λ0 ). Then ∀ f ∈ C ∞ (N ),

φ∗ (vγ,p ) f = vγ,p (f ◦ φ) (by Eq. 11.1)

0
= ((f ◦ φ) ◦ γ) (λ0 ) (by Eq. 5.1)
0
= (f ◦ (φ ◦ γ)) (λ0 ) (by associativity of composition)
= v(φ◦γ),φ(γ(λ0 )) f (by Eq. 5.1)
= v(φ◦γ),φ(p) f (γ(λ0 ) = p)
=⇒ φ∗ (vγ,p ) = v(φ◦γ),φ(p)

11.2 Pull-back map

Definition 58. Let M and N be smooth manifolds with cotangent bundles T ∗ M and T ∗ N respectively. Let
φ : M −→ N be a smooth map. Then, the pull-back map of φ is the map

φ∗ :T ∗ N −→ T ∗ M
ω 7→ φ∗ (ω)
where φ∗ (ω)(X) := ω(φ∗ (X)) (∀ X ∈ Tp M ) (11.4)

Components of pull-back φ∗ w.r.t charts (U, x) ∈ AM and (V, y) ∈ AN : Let p ∈ U and φ(p) ∈ V . Since
dy a is a covector, we have φ∗ (dypa ) as a covector in N. Then we can select a component of this covector by using
∂
∂xi as follows:

!
   ∂ 
∗a ∗ a
φ := φ (dy )φ(p)
∂xi p
  !
a ∂
= (dy )φ(p) φ∗ (by Eq. 11.4)
∂xi p
!
∂ φ̂a
= = φa∗i (by Eq. 11.2)
∂xi
p

41
Thus, the components of the push-forward and pull-back maps are exactly the same.

a
(φ∗ (X)) = φa∗i X i
(φ∗ (ω))i = φa∗i ωa

Remember, a = (1, . . . , dimN ) and i = (1, . . . , dimM ).

Claim: φ∗ (df ) = d(f ◦ φ).

Mnemonic: “covectors are pulled back” across tangent bundles in a manner dictated by the underlying map.

Important application:
Definition 59. Let M and N be smooth manifolds. Let φ : M ,→ N be an injective map. If we know a metric
g on N , then the induced metric gM in M is defined using the push-forward map φ∗ as follows:

gM (X, Y ) := g (φ∗ (X), φ∗ (Y )) (∀ X, Y ∈ Tp M ) (11.5)

In terms of components, ! !
  ∂ φ̂a ∂ φ̂b
(gM )ij = (gab )φ(p) (11.6)
p ∂xi ∂xj
φ(p) φ(p)

Example: N = (R , Ostd , A) and M = (S , O, A), then we can have several injective maps φ : S 2 ,→ R3 . For
3 2

example, S 2 could live in R3 either as a potato or a round sphere. However, suppose R3 is equipped with the
Euclidean metric gE ...(TODO complete this example)

11.3 Flow of a complete vector field

Definition 60. Let X be a vector field on a smooth manifold (M, O, A). A curve γ : I ⊆ R −→ M is called
an integral curve of X if
vγ,γ(λ) = Xγ(λ)

Definition 61. A vector field X is complete if all integral curves have I = R (i.e. domain is all of R).
Theorem 11.2. Compactly supported smooth vector field is complete.
Definition 62. The flow of a complete vector field X on a manifold M is a 1-parameter family

hX :R × M −→ M
(λ, p) 7→ γp (λ)

where γp : R −→ M is the integral curve of X with γ(0) = p.

Then for fixed λ ∈ R, hX

λ : M −→ M is smooth.

Picture: If S is a set of points in M , then hX

λ (S) can be seen as the new position of these points under the
flow h after the passage of λ units of parameter. In general, hX
X
λ (S) 6= S( if X 6= 0).

11.4 Lie subalgebras of the Lie algebra (Γ(T M ), [·, ·]) of vector fields

We know that Γ(T M ) = { set of all vector fields }, which can be seen as a C ∞ (M )-module, or as a R-vector
space.

(Γ(T M ), [·, ·]) with [X, Y ] defined by its action on a function f by [X, Y ]f := X(Y f ) − Y (Xf ) is a Lie algebra
since X, Y ∈ Γ(T M ) =⇒ [X, Y ] ∈ Γ(T M ) and the following properties are satisfied:

(i) Anticommutativity: [X, Y ] = −[Y, X]

42
 (ii) Linearity: [λX + Z, Y ] = λ[X, Y ] + [Z, Y ] where λ ∈ R

(iii) Jacobi identity: [X, [Y, Z]] + [Z, [X, Y ]] + [Y, [Z, X]] = 0

Let X1 , . . . , Xs be s (many) vector fields on M , such that

k k
∀i, j ∈ {1, . . . , s} [Xi , Xj ] = Cij Xk where Cij ∈R
| {z }
linear combination of Xk s

k
Cij are called structure constants.

Let spanR {X1 , . . . , Xs } := {all linear combinations of Xk }. Then (spanR {X1 , . . . , Xs }, [·, ·]) is a Lie subalgebra
of (Γ(T M ), [·, ·]).

Example: In S 2 , assume that the vector fields X1 , X2 , X3 satisfy [X1 , X2 ] = X3 , [X2 , X3 ] = X1 and [X3 , X1 ] =
X2 . Then (spanR {X1 , X2 , X3 }, [·, ·]) (= SO(3)) is a Lie subalgebra. An instance of vector fields satisfying these
conditions is (with Xi , θ, φ all taken at a point p, and x1 = θ, x2 = φ)

∂ ∂
X1 = − sin φ − cot θ cos φ
∂θ ∂φ
∂ ∂
X2 = cos φ − cot θ cos φ
∂θ ∂φ
∂
X3 =
∂φ

Note that the above is defined on a merely smooth manifold without any additional structure like metric.

11.5 Symmetry

Definition 63. A finite-dimensional Lie subalgebra (L, [·, ·]) is said to be a symmetry of a metric tensor field
g if

hX (A), hX




g λ ∗ λ ∗
(B) = g(A, B) (∀X (complete vector field) ∈ L, ∀λ ∈ R, ∀A, B ∈ Tp M )

∗
In another formulation (using pullback), hX λ g = g . The pullback of φ : M −→ M on g, itself, is defined as
follows:
(φ∗ g)(A, B) := g(φ∗ (A), φ∗ (B))

11.6 Lie derivative

It can be shown that the following expression is precisely the Lie derivative of g w.r.t a vector field

∗
hX
λ g−g
LX g := lim (11.7)
λ−→0 λ

Clearly, L is a symmetry of g, iff LX g = 0.

Definition 64. The Lie derivative L on a smooth manifold (M, O, A) a pair of a vector field X and a
(p, q)-tensor field T to a (p, q)-tensor field such that

(i) LX f = Xf ∀f ∈ C ∞ M

(ii) LX Y = [X, Y ] where X, Y are vector fields

43
 This condition sucks in information about the vector field X. It is not C ∞ -linear in the lower index.
If it were, the derivative would be independent of values of X at nearby points to the point where
derivative is evaluated. This is an important difference between the covariant derivative ∇ and the
Lie derivative.

(iii) LX (T + S) = LX T + LX S where T, S are (p, q)-tensors

(iv) Leibnitz rule: LX T (ω1 , . . . , ωp , Y1 , . . . , Yq ) = (LX T )(ω1 , . . . , ωp , Y1 , . . . , Yq )

+ T (LX ω1 , . . . , ωp , Y1 , . . . , Yq ) + · · · + T (ω1 , . . . , LX ωp , Y1 , . . . , Yq )
+ T (ω1 , . . . , ωp , LX Y1 , . . . , Yq ) + · · · + T (ω1 , . . . , ωp , Y1 , . . . , LX Yq ) where T is a (p, q)-tensor

Note that for a (p, q)-tensor T and a (r, s)-tensor S, since:

(T ⊗ S)(ω(1) , . . . , ω(p+r) , Y(1) , . . . , Y(q+s) ) =
T (ω(1) , . . . , ω(p) , Y(1) , . . . , Y(q) ) · S(ω(p+1) , . . . , ω(p+r) , Y(q+1) , . . . , Y(q+s) ),
Leibnitz rule implies LX (T ⊗ S) = (LX T ) ⊗ S + T ⊗ (LX S).

(v) LX+Y T = LX T + LY T

Observe that, in chart (U, x)

i ∂ ∂
(LX Y ) = X m (Y i ) − (X i ) Ys
∂xm ∂x s
| {z }
requires knowing X around the point

However, for covariant derivative

i ∂
(∇X Y ) = X m (Y i ) + Γism X m Y s
∂xm
In general, for a (1, 1)-tensor T

i ∂ ∂X i ∂X s
(LX T ) j = X m (T i ) − s T xj + j T is
∂xm j | ∂x{z } | ∂x{z }
(0 −0 for lower index) (0 +0 for upper index)

Application: As above, it is easy to calculate components of Lie derivative of metric g, LX g. Thus, by checking
whether the derivative equals 0 or not, it can be determined whether a metric features a symmetry.

44
12 Integration

This lecture will be the completion of our “lift” of analysis on charts to the manifold level. We want to be
R
able to integrate a function f over a manifold M . This M f will be an important tool for writing down
the action on Einstein Equations.

However, to define integral we need a mild new structure on the smooth manifold (M, O, A). It requires
(i) a choice of a certain tensor field, the so-called volume form and
(ii) a restriction on the atlas A, which is called ’orientation’.

12.1 Review of integration on Rd

We review this because this is what, after all, happens on charts; and we want to use this knowledge to have a
well-defined integration on manifolds.

a) If F : R −→ R, we assume a notion of integration is known. We define an integral over an interval (a, b) as

follows:
Z Z b
F := dx F (x) (which is understood in terms of, say, Riemann’s integral).
(a,b) a

b) If F : Rk −→ R, then

(i) on a box-shaped domain, Box = (a, b) × (c, d) × · · · × (u, v) ⊆ Rk , the integral on the box is a series of
integrals which have to be evaluated one after the other as follows
Z Z b Z d Z v
1 2
F := dx dx · · · dxk F (x1 , x2 , . . . , xk )
Box a c u

(ii) for other domains, G ⊆ Rk , we first introduce an indicator function µG : Rk −→ R such that

1, x∈G
µG (x) =
0, x 6∈ G

and then define

Z Z +∞ Z +∞ Z +∞
F := dx1 dx2 · · · dxk µG (x) · F (x1 , x2 , . . . , xk )
G −∞ −∞ −∞

While this may not be a practical definition, it tells us what we mean by an integral over a function
from Rk to R over an arbitrary domain G.

Note: All of the above comes with the disclaimer ’if the integral exists’ since there could be many issues that
do not allow the existence of the integral as defined above.

Change of variables, which may also be called integration by substitution.

Theorem 12.1. If F : Rk 3 G −→ R and φ : preimφ (G)(∈ Rk ) −→ G, then
Z Z
F (x) = |det(∂x)(y)| ·(F ◦ φ)(y)
G preimφ (G) | {z }
Jacobian of φ

45
 φ
Rk 3 preimφ (G) G ∈ Rk

F◦ F
φ
R

Example: Consider the domain G ⊂ R2 , which includes the entire R2 except the x-axis. Let

φ :R+ × {(0, π) ∪ (π, 2π)} −→ G

(r, ϕ) 7→ (r cos ϕ, r sin ϕ)

Thus, G is in Cartesian coordinates and preimφ (G) is in polar coordinates. Let us calculate the Jacobian.
 
 cos ϕ sin ϕ 
b
(∂a x )(r, ϕ) = 

−r sin ϕ

r cos ϕ
det(∂a xb )(r, ϕ) = r
Z Z ∞ Z 2π
1 2 1 2
=⇒ dx dx
| {z } F (x , x ) = dr dϕ r F (r cos ϕ, r sin ϕ)
G 0 0 | {z }
volume element volume element

12.2 Integration on one chart

Let (M, O, A) be a smooth manifold, f : M −→ R and choose charts (U, x), (U, y) ∈ A

y(U ) ∈ Rk
f(
y)
:=
f◦
y y −1

f
y ◦ x−1 U R

−1
x x
f◦
)
:=
f (x
x(U ) ∈ Rk

R
Consider U
f Z Z
f := dk α f(x) (α)
U x(U )

12.3 Volume forms

Definition 65. On a smooth manifold (M, O, A)

a (0, dimM )-tensor field Ω is called a volume form if

(a) Ω vanishes nowhere (i.e. Ω 6= 0 ∀ p ∈ M )

(b) totally antisymmetric

Ω(. . . , |{z} Y . . . ) = −Ω(. . . , |{z}

X , . . . , |{z} Y , . . . , |{z}
X ...)
ith jth ith jth

46
In a chart:
Ωi1 ...id = Ω[i1 ...id ]

Example (M, O, A, g) metric manifold

construct volume form Ω from g

In any chart: (U, x)

q
Ωi1 ...id := det(gij (x))i1 ...id

where Levi-Civita symbol i1 ...id is defined as 123...d = +1

1...d = [i1 ...id ]

Proof. (well-defined) Check: What happens under a change of charts

q
det(g(y)ij )i1 ...id =
Ω(y)i1 ...id =
s
∂xm ∂xn ∂y m1 ∂y md
= det(gmn (x) i j
) i1 . . . [m1 ...md ] =
∂y ∂y ∂x ∂xid
       
∂x  ∂y ∂x
q  q
= |detgij (x)| det
 det i1 ...id = detgij (x)i1 ...id sgn det
∂y  ∂x ∂y

EY : 20150323

Consider the following:

i1 id
Ω(y)(Y(1) . . . Y(d) ) = Ω(y)i1 ...id Y(1) . . . Y(d) =
q
i1 id
= det(gij (y))i1 ...id Y(1) . . . Y(d) =
s
∂xm ∂xn ∂y i1 ∂y id
= det(gmn (x)) i j
i1 ...id m1 . . . m X m1 . . . X md =
∂y ∂y ∂x ∂x d
s
∂xm ∂xn
 
∂y
= det(gmn (x)) i j
det m1 ...md X m1 . . . X md =
∂y ∂y ∂x
    
p ∂x  ∂y
m1 ...md X m1 . . . X md =

= det(gmn (x)) det det
∂y  ∂x
    
p ∂x ∂x
= det(gmn (x))m1 ...md sgn det X m1 . . . X md = sgn(det )Ωm1 ...md (x)X m1 . . . X md
∂y ∂y

 
∂y
If det ∂x > 0,
Ω(y)(Y(1) . . . Y(d) ) = Ω(x)(X(1) . . . X(d) )

This works also if Levi-Civita symbol i1 ...id doesn’t change at all under a change of charts. (around 42:43
https://youtu.be/2XpnbvPy-Zg)

Alright, let’s require,

restrict the smooth atlas A

47
 to a subatlas (A↑ still an atlas)
A↑ ⊆ A

s.t. ∀ (U, x), (V, y) have chart transition maps y ◦ x−1

x ◦ y −1
 
∂y
s.t. det ∂x >0
↑
such A called an oriented atlas

(M, O, A, g) =⇒ (M, O, A↑ , g)

Note: associated bundles.

 
∂y b ∂y b
Note also: det ∂xa = det(∂a (y b x−1 )) ∂xa is an endomorphism on vector space V . ϕ : V −→ V
detϕ independent of choice of b
p
g is a (0, 2) tensor field, not endomorphism (not independent of choice of basis) |det(gij (y))|
Definition 66. Ω be a volume form on (M, O, A↑ ) and consider chart (U, x)

Definition 67. ω(X) := Ωi1 ...id i1 ...id same way 12...d = +1
[... ]
one can show

 
∂x
ω(y) = det ω(x) scalar density
∂y

12.4 Integration on one chart domain U

Definition 68. Z Z
(U,y)
f: = dd βω(y) (y −1 (β))f(y) (β) (12.1)
U y(U )

Proof. : Check that it’s (well-defined), how it changes under change of charts
Z Z Z Z     
(U,y) d −1 d
 ∂y  −1 ∂x
f: = d βω(y) (y (β))f(y) (β) = = d α det
 f(x) (α)ω(x) (x (α)det =
U y(U ) (U,y) x(U ) ∂x  ∂y
Z
= dd αω(x) (x−1 (x))f(x) (α)
x(U )

On an oriented metric manifold (M, O, A↑ , g)

Z Z q
f := dd α det(gij (x))(x−1 (α)) f(x) (α)
U x(U ) | {z }
√
g

12.5 Integration on the entire manifold

48
13 Lecture 13: Relativistic spacetime

Recall, from Lecture 9, the definition of Newtonian spacetime

∇ torsion free
t ∈ C ∞ (M )
(M, O, A, ∇, t)
dt 6= 0
∇dt = 0 (uniform time)

and the definition of relativistic spacetime (before Lecture )

∇ torsion-free
↑
(M, O, A , ∇, g, T ) g Lorentzian metric(+ − −−)
T time-orientation

13.1 Time orientation

Definition 69. (M, O, A↑ , g) a Lorentzian manifold. Then a time-orientation is given by a vector field T that

(i) does not vanish anywhere

(ii) g(T, T ) > 0

Newtonian vs. relativistic

Newtonian
X was called future-directed if
dt(X) > 0

∀ p ∈ M , take half plane, half space of Tp M

also stratified atlas so make planes of constant t straight
relativistic
half cone ∀ p, q ∈ M , half-cone ⊆ Tp M

This definition of spacetime

Question
I see how the cone structure arises from the new metric. I don’t understand however, how the T , the time
orientation, comes in

Answer
(
(M, O, A, g) g ←
− + − −−)

requiring g(X, X) > 0, select cones

T chooses which cone

This definition of spacetime has been made to enable the following physical postulates:

(P1) The worldline γ of a massive particle satisfies

(i) gγ(λ) (vγ,γ(lambda) , vγ,γ(λ) ) > 0

(ii) gγ(λ) (T, vγ,γ(λ) ) > 0

49
(P2) Worldlines of massless particles satisfy

(i) gγ(λ) (vγ,γ(λ) , vγ,γ(λ) ) = 0

(ii) gγ(λ) (T, vγ,γ(λ) ) > 0

picture: spacetime:

Answer (to a question) T is a smooth vector field, T determines future vs. past, “general relativity: we have
such a time orientation; smoothness makes it less arbitrary than it seems” -FSchuller,

Claim: 9/10 of a metric are determined by the cone

spacetime determined by distribution, only one-tenth error

13.2 Observers

(M, O, A↑ , ∇, g, T )
Definition 70. An observer is a worldline γ with

g(vγ , vγ ) > 0
g(T, vγ ) > 0

together with a choice of basis

vγ,γ(λ) ≡ e0 (λ), e1 (λ), e2 (λ), e3 (λ)
 
1
 −1 
of each Tγ(λ) M where the observer worldline passes, if g(ea (λ), eb (λ)) = ηab =
 
−1

 
−1 ab

precise: observer = smooth curve in the frame bundle LM over M

13.2.1 Two physical postulates

(P3) A clock carried by a specific observer (γ, e) will measure a time

Z λ1 q
τ := dλ gγ(λ) (vγ,γ(λ) , vγ,γ(λ) )
λ0

between the two “events”

γ(λ0 ) “start the clock”

and
γ(λ1 ) “stop the clock”

Compare with Newtonian spacetime:

t(p) = 7

Thought bubble: proper time/eigentime τ

50
 M = R4
O = Ost
Application/Example.
A 3 (R4 , idR4 )
i
g : g(x)ij = ηij ; T(x) = (1, 0, 0, 0)i

=⇒ Γi(x) jk = 0 everywhere

=⇒ (M, O, A↑ , g, T, ∇) Riemm = 0
=⇒ spacetime is flat

This situation is called special relativity.

Consider two observers:

γ : (0, 1) −→ M
i
γ(x) = (λ, 0, 0, 0)i
δ : (0, 1) −→ M

(λ, αλ, 0, 0)i 1
i
λ≤ 2
α ∈ (0, 1) :δ(x) =
(λ, (1 − λ)α, 0, 0)i 1
λ> 2

let’s calculate:
Z 1 q Z 1
τγ := i γ̇ j =
g(x)ij γ̇(x) dλ1 = 1
(x)
0 0
Z 1/2 p Z 1 p Z 1 p p
τδ := dλ 1− α2 + 12 − (−α)2 = 1 − α2 = 1 − α2
0 1/2 0

Note: piecewise integration

Taking the clock postulate (P3) seriously, one better come up with a realistic clock design that supports
the postulate. idea.

2 little mirrors

(P4) Postulate

Let (γ, e) be an observer, and

δ be a massive particle worldline that is parametrized s.t. g(vγ , vγ ) = 1 (for parametrization/normalization
convenience)

Suppose the observer and the particle meet somewhere (in spacetime)

δ(τ2 ) = p = γ(τ1 )

This observer measures the 3-velocity (spatial velocity) of this particle as

vδ : α (vδ,δ(τ2 ) )eα α = 1, 2, 3 (13.1)

where 0 , 1 , 2 , 3 is the unique dual basis of e0 , e1 , e2 , e3

EY:20150407

There might be a major correction to Eq. (13.1) from the Tutorial 14 : Relativistic spacetime, matter, and
Gravitation, see the second exercise, Exercise 2, third question:

α (vδ )
v := eα (13.2)
0 (vδ )

51
Consequence: An observer (γ, e) will extract quantities measurable in his laboratory from objective spacetime
quantities always like that.

Ex: F Faraday (0, 2)-tensor of electromagnetism:

 
0 E1 E2 E3
−E 0 B3 −B2 
1
F (ea , eb ) = Fab =
 
−E2 −B3

0 B1 
−E3 B2 −B1 0
observer frame ea , eb

Eα := F (e0 , eα )
B γ := F (eα , eρ )αβγ where 123 = +1 totally antisymmetric

13.3 Role of the Lorentz transformations

Lorentz transformations emerge as follows:

Let (γ, e) and (e
γ , ee) be observers with γ(τ1 ) = γ
e(τ2 )

(for simplicity γ(0) = γ

e(0)

Now
e0 , . . . , e 1 at τ = 0
and ee0 , . . . , ee1 at τ = 0
both bases for the same Tγ(0) M

Thus: eea = Λba eb Λ ∈ GL(4)

Now:

ea , eeb ) = g(Λma em , Λnb en ) =

ηab = g(e
= Λma Λnb g(em , en )
| {z }
ηmn

i.e. Λ ∈ O(1, 3)

Result: Lorentz transformations relate the frames of any two observers at the same point.

xµ − Λµν xν ” is utter nonsense

“e

Tutorial

I didn’t see a tutorial video for this lecture, but I saw that the Tutorial sheet number 14 had the relevant topics.
Go there.

52
14 Lecture 14: Matter

two types of matter

point matter

field matter

point matter

massive point particle

more of a phenomenological importance

field matter

electromagnetic field

more fundamental from the GR point of view

both classical matter types

14.1 Point matter

Our postulates (P1) and (P2) already constrain the possible particle worldlines.

But what is their precise law of motion, possibly in the presence of “forces”,

(a) without external forces Z q

Smassive [γ] := m dλ gγ(λ) (vγ,γ(λ) , vγ,γ(λ) )

with:
gγ(λ) (Tγ(λ) , vγ,γ(λ) ) > 0

dynamical law Euler-Lagrange equation

similarly Z
Smassless [γ, µ] = dλµg(vγ,γ(λ) , vγ,γ(λ) )

δµ g(vγ,γ(λ) , vγ,γ(λ) ) = 0
δγ e.o.m.

Reason for describing equations of motion by actions is that composite systems have an action that is the
sum of the actions of the parts of that system, possibly including “interaction terms.”

Example.
S[γ] + S[δ] + Sint [γ, δ]

(b) presence of external forces

or rather presence of fields to which a particle “couples”

Example Z q
S[γ; A] = dλm gγ(λ) (vγ,γ(λ) , vγ,γ(λ) ) + qA(vγ,γ(λ) )

where A is a covector field on M . A fixed (e.g. the electromagnetic potential)

53
 m
Consider Euler-Lagrange eqns. Lint = qA(x) γ̇(x)

˙ !
∂Lint ∂Lint a
m(∇vγ vγ )a + γ− m = 0 =⇒ m(∇vγ vγ ) = −qF am γ̇ m
∂ ˙m(x) ∂γ(x) | {z }
| {z } Lorentz force on a charged particle in an electromagnetic field
∗

 ˙ 
∂L ∂L ∂ m
= qA(x)a , γ=q· (A(x)m ) · γ̇(x)
∂ γ̇ a ∂ ˙m ∂xm
∂L ∂
=q· (A(x)m )γ̇ m
∂γ a ∂xa
 
∂Aa ∂Am m m
∗=q − γ̇(x) = q · F(x)am γ̇(x)
∂xm ∂xa
F ← Faraday

Z q
S[γ] = (m g(vγ , vγ ) + qA(vγ ))dλ

14.2 Field matter

Definition 71. Classical (non-quantum) field matter is any tensor field on spacetime where equations of motion
derive from an action.

Example:
√
Z
1
SMaxwell [A] = d4 x −gFab Fcd g ac g bd
4 M

A (0, 1)-tensor field

= thought cloud: for simplicity one chart covers all of M
√
− for −g (+ − −−)

Fab := 2∂[a Ab] = 2(∇[a A)b]

Euler-Lagrange equations for fields

∂2L
 
∂L ∂ ∂L ∂ ∂
0= − +
∂Am ∂xs ∂∂s Am ∂xs ∂xt ∂∂t ∂s Am

Example . . .
(∇ ∂ F )ma = j a
∂xm

inhomogeneous Maxwell

thought bubble j = qvγ

∂[a Fb] − ()

homogeneous Maxwell

Other example well-liked by textbooks

√
Z
SKlein-Gordon [φ] := d4 x −g[g ab (∂a φ)(∂b φ) − m2 φ2 ]
M

φ (0, 0)-tensor field

54
14.3 Energy-momentum tensor of matter fields

At some point, we want to write down an action for the metric tensor field itself.

But then, this action Sgrav [g] will be added to any Smatter [A, φ, . . . ] in order to describe the total system.

Stotal [g, A] = Sgrav [g] + SMaxwell [A, g]

δA :=⇒ Maxwell’s equations

1
δgab : Gab + (−2T ab ) = 0
16πG
G Newton’s constant

Gab = 8πGN T ab

Definition 72. Smatter [Φ, g] is a matter action, the so-called energy-momentum tensor is
 
−2 ∂Lmatter ∂Lmatter
T ab := √ − ∂s + ...
−g ∂gab ∂∂s gab

−2
− of √
g is Schrödinger minus (EY : 20150408 F.Schuller’s joke? but wise)

choose all sign conventions s.t.

T (0 , 0 ) > 0

Example: For SMaxwell :

1
Tab = Fam Fbn g mn − Fmn F mn gab
4
Tab ≡ TMaxwellab

T (e0 , e0 ) = E 2 + B 2

T (e0 , eα ) = (E × B)α

Fact: One often does not specify the fundamental action for some matter, but one is rather satisfied to assume
certain properties / forms of
Tab

Example Cosmology: (homogeneous & isotropic)

perfect fluid

of pressure p and density ρ modelled by

T ab = (ρ + p)ua ub − pg ab

radiative fluid

What is a fluid of photons:

55
 ab
TMaxwell gab = 0
ab !
observe: Tp.f. gab = 0
= (ρ + p)ua ub gab − p g ab gab
| {z }
4

↔ρp 04p = 0
ρ = 3p

p = 31 ρ

Reconvene at 3 pm? (EY : 20150409 I sent a Facebook (FB) message to the International Winter School on
Gravity and Light: there was no missing video; it continues on Lecture 15 immediately)

Tutorial 14: Relativistic Spacetime, Matter and Gravitation

Exercise 2: Lorentz force law.

Question electromagnetic potential.

56
15 Einstein gravity

Recall that in Newtonian spacetime, we were able to reformulate the Poisson law ∆φ = 4πGN ρ in terms of the
Newtonian spacetime curvature as
R00 = 4πGN ρ

R00 with respect to ∇Newton , and GN = Newtonian gravitational constant.

This prompted Einstein to postulate that the relativistic field equations for the Lorentzian metric g of (rela-
tivistic) spacetime
Rab = 8πGN Tab

However, this equation suffers from a problem. We know from matter theory that in RHS, (∇a T )ab = 0 since
this has been formulated from an action. But in LHS, (∇a R)ab 6= 0 generically. Einstein tried to argue this
problem away. Nevertheless, the equations cannot be upheld.

15.1 Hilbert

Hilbert was a specialist for variational principles. To find the appropriate LHS of the gravitational field equa-
tions, Hilbert suggested to start from an action

√
Z
SHilbert [g] = −gRab g ab
M

which, in a sense, is formulated in terms of “simplest action”.

Aim: varying this w.r.t. metric gab will result in some tensor Gab .

15.2 Variation of SHilbert

√ √ √
Z
!
0 = |{z}
δ SHilbert [g] = [δ −g g ab Rab + −g δg ab Rab + −g g ab δRab ]
M | {z } |{z} | {z }
gi 1 2 3

√ mn √
ad 1: δ −g = −(detg)g
√
2 −g
δgmn
= 12 −gg mn δgmn
the above comes from δdet(g) = det(g)g mn δgmn e.g. from det(g) = exp trln g

ad 2: g ab gbc = δca =⇒ (δg ab )gbc + g ab (δgbc ) = 0 =⇒ δg ab = −g am g bn δgmn

ad 3:

∆Rab =
|{z} δ∂b Γmam − δ∂m Γmab + ΓΓ − ΓΓ
normal coords at point

= ∂b δΓmam − ∂m δΓmab = ∇b (δΓ)mam − ∇m (δΓ)mab

√ √
=⇒ −gg ab δRab = −g

“if you formulate the variation properly, you’ll see the variation δ commute with ∂b ”

i i
Γ(x) jk − Γg(x) jk are the components of a (1, 2)-tensor.
Let us use the notation: (∇b A)ij =: Aij;b

√ √ √ √ √
∴ −gg ab δRab |{z}
= −g(g ab δΓmam );b − −g(g ab δΓmab );m = −g Ab;b − −g B m,m
∇g=0

57
Question: Why is the difference of coefficients a tensor?

Answer:

i ∂y i ∂xm ∂xq
m ∂y i ∂ 2 xm
Γ(y) = Γ(x) +
jk ∂xm ∂y j ∂y k nq ∂xm ∂y j ∂y k

Collecting terms, one obtains

1√ √ √ √
Z
!
0 = δSHilbert = [ −g g mn δgmn g ab Rab − −g g am g bn δgmn Rab + ( −g Aa ) ,a − ( −g B b ) ,b ]
M 2 | {z } | {z }
surface surface term
√
Z
1 1
= −g δ gmn [ g mn R − Rmn ] =⇒ Gmn = Rmn − g mn R
M |{z} 2 2
arbitrary variation

Hence Hilbert, from this “mathematical” argument, concluded that one may take

1
Rab − gab R = 8πGN Tab
2

Einstein equations
√
Z
SE−H [g] = −g R
M

15.3 Solution of the ∇a T ab = 0 issue

One can show (−→ Tutorials) that the Einstein curvature

1
Gab = Rab − gab R
2

satisfy the so-called contracted differential Bianchi identity (∇a G)ab = 0.

15.4 Variants of the field equations

(a) a simple rewriting:

1 1
Rab − gab R = 8πGN Tab = Tab (GN = )
2 8π
1
Rab − gab R = Tab || g ab (contract on both sides with g ab )
2
R − 2R = T := Tab g ab
=⇒ R = −T
1
=⇒ Rab + gab T = Tab
2
1
⇐⇒Rab = (Tab − T gab ) =: Tbab
2
∴ Rab = Tbab

R √
(b) SE−H [g] := M
−g(R + 2Λ) (Λ is called cosmological constant)

History:
1915: Λ < 0 (Einstein) in order to get a non-expanding universe
>1915: Λ = 0 (Hubble)

58
 today: Λ > 0 to account for an accelerated expansion
Λ 6= 0 can be interpreted as a contribution − 21 Λg to the energy-momentum of matter in spacetime. This
energy, which does not interact with anything but contributes to the curvature is called “dark energy”.

Question: surface terms scalar?

Answer: for a careful treatment of the surface terms which we discarded, see, e.g. E. Poisson, “A relativist’s
toolkit” C.U.P. “excellent book”

Question: What is a constant on a manifold?

R√ R√
Answer: −g Λ = Λ −g 1

[back to dark energy]

[Weinberg used QCD to calculate Λ using the idea that Λ could arise as the vacuum energy of the standard
model fields. It turns out that
Λcalculated = 10120 × Λobs
which is called the “worst prediction of physics”.

Tutorials: check that

• Schwarzscheld metric (1916)

• FRW metric

• pp-wave metric

• Reisner-Nordstrom

=⇒ are solutions to Einstein’s equations

in high school

mẍ + mω 2 x2 = 0

x(t) = cos (ωt)

ET: [elementary tutorials]

study motion of particles & observers in Schwarzschild S.T.

Satellite lectures:
Marcus C. Werner: Gravitational lensing

odd number of pictures Morse theory (EY:20150408 Morse Theory !!!)

Domenico Giulini: Canonical Formulations of GR

Hamiltonian form

Key to Quantum Gravity

59
16 L18: Canonical Formulation of GR-I

16.1 Dynamical and Hamiltonian formulation of General Relativity

Purpose:

1) formulate and solve initial-value problems

2) integrate Einstein’s Equations by numerical codes

3) characterise degrees of freedom

4) characterise isolated systems, associated symmetry groups and conserved quantities like Energy/Mass, Mo-
menta (linear and angular), Poincare charges

5) starting point for “canonical quantisation” program

How do we achieve this goal? We will rewrite Einstein’s Equations in form of a constrained Hamiltonian
system.

1
Rµν − gµν R + Λ gµν = |{z}
k Tµν
2 |{z}
| {z } cosmological constant 8πG
Gµν c4

k = 8πG
c4 is an important quantity as it turns the energy density Tµν into curvature.
Physical dimensions:

1
for curvature, [Gµν ] = ,
m2
Joule
for energy density [Tµν ] =
m3
1
m2 m
∴ [k] = J
=
m3
J

Convention (for this lecture):

Greek indices run from 0 to 3 and latin indices from 1 to 3
signature is (−, +, +, +) as it makes space positive definite in 3 + 1-decomposition
T00 is positive energy density.

60
17 Lecture 22: Black Holes

Only depends on Lectures 1-15, so does lecture on “Wednesday”

Schwarzschild solution also vacuum solution (from tutorial EY : oh no, must do tutorial)

Study the Schwarzschild as a vacuum solution of the Einstein equation:

m = GN M where M is the “mass”

 
2m 1
g = 1− dt ⊗ dt − dr ⊗ dr − r2 (dθ ⊗ dθ + sin2 θdϕ ⊗ dϕ
r 1 − 2m
r

in the so-called Schwarzschild coordinates t r θ ϕ

(−∞, ∞) (0, ∞) (0, π) (0, 2π)
What staring at this metric for a while, two questions naturally pose themselves:

(i) What exactly happens r = 2m?

t r θ ϕ
(−∞, ∞) (0, 2m) ∪ (2m, ∞) (0, π) (0, 2π)

(ii) Is there anything (in the real world) beyond t −→ −∞?

t −→ +∞
idea: Map of Linz, blown up

Insight into these two issues is afforded by stopping to stare.

Look at geodesic of g, instead.

17.1 Radial null geodesics

null - g(vγ , vγ ) = 0

Consider null geodesic in “Schd”

Z "   −1 #
2m 2 2m 2 2 2 2 2
S[γ] = dλ 1− ṫ − 1 − ṙ − r (θ̇ + sin θϕ̇ )
r r

with [. . . ] = 0

and one has, in particular, the t-eqn. of motion:

  .
2m
1− ṫ = 0
r
=⇒  
2m
1− ṫ = k = const.
r

Consider radial null geodesics

!
θ = const. ϕ = const.

61
From 2 and 2
=⇒ ṙ2 = k 2 ↔ ṙ = ±k

=⇒ r(λ) = ±k · λ

Hence, we may consider

t(r) := t(±kλ)
e

Case A: ⊕

de
t t˙
e k r
= =
=
dr ṙ 1 − 2m
r k r − 2m

t+ (r) = r + 2m ln |r − 2m|
=⇒ e

(outgoing null geodesics)

Case b. ± (Circle around −, consider −):

t− (r) = −r − 2m ln |r − 2m|
e

(ingoing null geodesics)

Picture

17.2 Eddington-Finkelstein

Brilliantly simple idea:

change (on the domain of the Schwarzschild coordinates) to different coordinates, s.t.
in those new coordinates,
ingoing null geodesics appear as straight lines, of slope −1

This is achieved by

t̄(t, r, θ, ϕ) := t + 2m ln |r − 2m|

Recall: ingoing null geodesic has

t(r) = −(r + 2m ln |r − 2m|)

e (Schdcoords)

⇐⇒ t̄ − 2m ln |r − 2m| = −r − 2m ln |r − 2m| + const.

∴ t̄ = −r + const.

(Picture)

outgoing null geodesics

t̄ = r + 4m ln |r − 2m| + const.

Consider the new chart (V, g) while (U, x) was the Schd chart.

62
 [
U
|{z} { horizon } = V
Schd

“chart image of the horizon”

Now calculate the Schd metric g w.r.t. Eddington-Finkelstein coords.

t̄(t, r, θ, ϕ) = t + 2m ln |r − 2m|
r̄(t, r, θ, ϕ) = r
θ̄(t, r, θ, ϕ) = θ
ϕ̄(t, r, θ, ϕ) = ϕ

EY : 20150422 I would suggest that after seeing this, one would calculate the metric by your favorite CAS. I
like the Sage Manifolds package for Sage Math.

Schwarzschild BH.sage on github

Schwarzschild BH.sage on Patreon

Schwarzschild BH.sage on Google Drive

sage : load ( ‘ ‘ Schw a r z s c h i l d _ B H . sage ’ ’)
4 - dimensional manifold ’M ’
expr = expr . sim p l i f y _ r a d i c a l ()
Levi - Civita connection ’ nabla_g ’ associated with the Lorentzian metric ’g ’ on the 4 - dimensional manifold ’M ’
Launched png viewer for Graphics object consisting of 4 graphics primitives

Then calculate the Schwarzschild metric g but in Eddington-Finkelstein coordinates. Keep in mind to calculate
the set of coordinates that uses t̄, not e
t:
sage : gI . display ()
gI = (2* m - r )/ r dt * dt - r /(2* m - r ) dr * dr + r ^2 dth * dth + r ^2* sin ( th )^2 dph * dph
sage : gI . display ( X_EF_I_null . frame ())
gI = (2* m - r )/ r dtbar * dtbar + 2* m / r dtbar * dr + 2* m / r dr * dtbar + (2* m + r )/ r dr * dr + r ^2 dth * dth + r ^2* sin ( th )^2 dph * dph

63
References
[1] Sepideh Bakhoda (http://math.stackexchange.com/users/36591/sepideh-bakhoda), Are there simple examples
of Riemannian manifolds with zero curvature and nonzero torsion, Mathematics Stack Exchange. URL:
http://math.stackexchange.com/q/465672 (version: 2013-08-12).

64