Tensors and Tensor Analysis

A.F. Mashrafi
University of Dhaka
December 2023

Contents
0 Preface 2

1 Preliminaries 3
1.1 Vectors and Covectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Brief Introduction to Differentiable Manifolds . . . . . . . . . . . . . . . . . . . . . . . . 4

2 Tensors 7

3 Transformation Laws 10

4 Invariants 13
4.1 Trace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
4.2 Contraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

5 Symmetric and Antisymmetric Tensors 14

5.1 Symmetric Tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
5.2 Antiymmetric Tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
5.3 Tensor Density . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

6 Covariant Derivative 16

7 Metric Tensor 18
7.1 Lorentzian Signature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
7.2 Geodesic Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

1
0 Preface
This notes is for people who wants to study Tensor and Tensor analysis with minimal rigor. I have
tried to make the contents as less rigorous as I can but I’ll encourage the reader to look up the ac-
tual notes and books I have used to fill in the gaps. I have assumed that anyone who’s reading this
have a elementary knowledge about Linear Algebra. Even so I have discussed briefly about the things
that we’ll be needing the most. The differential geometry part is kind of vaguely written by emitting
every kind of proof and details on some extremely important points. Again the reader is encouraged
to look up the reference materials. If you find any typos or conceptual errors please mail me at af-
[email protected]. Books and Notes I have used to compile this note:

• Tensor Analysis on Manifolds by Richard L. Bishop Samuel I. Goldberg

• General Theory of Relativity (PHY413) Lecture notes by Atonu Roy Chowdhury Course taught
by Syed Hasibul Hassan Chowdhury

• The WE-Heraeus International Winter School on Gravity and Light Course taught by Dr. Fred-
eric P. Schuller
• Mathematical Methods for Physicists by George B. Arfken
• Introduction to manifolds by Loring W. Tu

2
1 Preliminaries

1.1 Vectors and Covectors

Definiton 1. A vector space V is a set with two operations, addition, + : V × V → V and
multiplication, · : R × V → V following with some additional properties given by

1. The commutative law for + : v + w = w + v.

2. The associative law for + : (v + w) + x = v + (w + x).
3. Existence of identity for + : v + 0 = v.
4. Existence of negatives: There is −v such that v + (−v) = 0.
5. a(v + w) = av + aw.
6. (a + b)v = av + by.
7. 1v = v.

The elements of V are called vectors.

We know ( so that means I am presuming some knowledge about vector spaces) that every Vector
v has a linearly independent set of basis. If we have a vector space V with dim N then let’s say it’s
basis vectors are {ϵ1 , ϵ2 , ...., ϵn } or in shorthand notation {ϵµ } where µ = 1, 2, ..., n. So if we have a
vector v ∈ V , then we can write it in terms of it’s basis. That is,
n
X
1 2 n
v = v ϵ1 + v ϵ2 + ... + v ϵn = v µ ϵµ
µ=0

Where v µ ’s are the components of the vector. In order to not always have the summation sym-
bols everywhere Einstein came up with a clever notation, known as Einstein summation convention.
Whenever we’ll have a repeated indices we’ll pretend there is a hidden summation ( I mean it actually
exists but we won’t write it for the simplicity of the calculation).
n
X
v µ ϵµ = v µ ϵµ
µ=0

Definiton 2. Let V be a vector space. Then it’s dual space V ∗ is a space of all the functions
f such that
f : V → R, where f ∈ V ∗

If {ϵµ } is a basis of V , then the dual basis is the set {ϵν }, where each ϵν ∈ V ∗ such that,

ν 1 if µ = ν
ϵ (ϵµ ) ==
0 if µ ̸= ν

An element of the dual basis is called a covector. In terms of the choosen basis a covector can be
written as

ω = ω ν ϵν
Where ων ’s are the components of the covector.

3
 Remark
We call a raised index a contravariant index and a lower index a covariant index and we stick
with the convention that a vector has components with contravariant indices and basis elements
with covariant indices and covectors the other way around.

1.2 Brief Introduction to Differentiable Manifolds

A manifold is something that locally looks like Rn . Now why do we need manifolds? While doing
maths and physics we often come across surfaces which are not flat. Even earth is not flat. So we can
not do calculus there globally. We have to transfer a portion of the surface ( open set) to a coordinate
system that is easy to work with and there we can do everything and we’ll be able to describe what’s
happening globally by doing things there. Lets say we have a manifold M with dimension n. Now we
choose a open set U . Now we map it to a coordinate system by a map φ. So, φ : U → Rn . And we call
this pair (U, φ) a chart. U is the coordinate neighbourhood and φ is the coordinate system. Now we

Figure 1: A portion of manifold is transferred into 2 dim cartesian coordinate.

have two charts (U, x) and (V, y). Now look at the following picture. The two maps between x(U ∩ V )
and y(U ∩ V ), y ◦ x−1 and x ◦ y −1 are called transition functions between the charts. ( Look at fig 2
and fig 3).

Figure 2: Transition functions between the charts

For each point p in M we define a vector space called tangent space Tp M . It will have the same
dimension as our manifold. Say our manifold is a sphere. Then at each point of the surface you can
draw a tangent plane. And this will be the tangent space for that point on that sphere. To geta better

∂
Idea look at Fig 4.. The basis for tangent space at a point p on the manifold M is given by ∂xµ
p
for µ = 1, ..., dimM .
Now we know that every vector space has a dual vector space. For Tp M , the dual vector space
n will obe
∗
Tp M . Which is called the cotangent space. The basis for the cotangent space are given by dxν p .

4
 Figure 3: Transition functions between the charts

Definiton 3. (Vector field) A vector field is a map X which assigns any point p ∈ M to a
tangent vector Xp at p. Given a vector field X and a function f we can define a new function

X(f ) : M → R

p 7→ Xp f
The vector field X is smooth ( C∞ ) if Xp f is a smooth function for any smooth function f .
One can also think of a vector field as a smooth map

X : C ∞ (M, R) → C ∞ (M, R)
for a given f ∈ C ∞ (M, R). It has the following properties:

1. X(f + g) = X(f ) + X(g), where f, g ∈ C ∞ (M )

2. X(λ · f ) = λ · X(f )
3. X(f •g) = f •X(g)+X(f )•g, where • : C ∞ (M )×C ∞ (M → C ∞ (M ) is the multiplication
on the ring C ∞ (M ). It is known as the Leibniz rule

Definiton 4. (1-form) The dual notion of vector field on a smooth manifold is called 1-form.
A 1-form ω assigns to each point p ∈ M a covector ωp at p. Given a smooth real-valued function
f on M , one can write down a 1-form df such that df |p ∈ Tp∗ M is given by

df |p (X|p ) = Xf (p) = X|p f

Note that we denote vector field and 1-form by X and df , respectively, and vector and covector
by X|p and df |p

5
 Figure 4: Different tangent plane of point p and q

∂
The action of dxµ |p on ∂xµ is given by
p
 
ν ∂
dx = δνµ
p ∂xµ p

6
2 Tensors

Definiton 5. Let V be a vector space. Then A (r, s) type tensor T is a multilinear map defined
as,

T : V ∗ × V ∗ × ... × V ∗ × V × V × ... × V → R
| {z } | {z }
r times s times

Remark
Multilinearity Let’s say we have vector space V of dim M and we have defines a tensor T
of type (1, 1) in that vector space. Then, multilinearity tells us that, say for ϕ, ψ ∈ V ∗ and
ν, λ ∈ V then,

T (ϕ + ψ, ν + λ) = T (ϕ, ν + λ) + T (ψ, ν + λ) = T (ϕ, ν) + T (ϕ, λ) + T (ψ, ν) + T (ψ, λ)

T (α · ϕ, ν) = α · T (ϕ, ν)

If we choose our vector space to be a tangent space on a point p of a manifold M . Then the tensor
defined there would be,

T |p : Tp∗ (M ) × Tp∗ (M )... × Tp∗ (M ) × Tp (M ) × .... × Tp (M ) → R

| {z } | {z }
r times s times

(If you don’t understand what Tp (M ) and Tp∗ (M ) means then please look up the Appendix A).

Example 1
A vector is a (1,0) type tensor. Because

T :V∗ →R

and we know that (V ∗ )∗ : V ∗ → R and (V ∗ )∗ = V .

Example 2
A covector is a (0,1) type tensor because

T :V →R

and we know that V ∗ : V → R.

7
 Example 3 We can define a (1,1) tensor by

δp : Tp∗ (M ) × Tp (M ) → R, with δp (ωp , Xp ) = ωp (Xp )

for any ωp ∈ Tp∗ (M ) and Xp ∈ Tp (M ).

n o
∂
Definiton 6. Let T |p be a tensor of type (r, s) at p ∈ M . If the basis of Tp (M ) is ∂xν p and
the dual basis of Tp∗ (M ) µ
is {dx |p }. The components of T |p are then the (r + s) dim M
many
numbers
 
∂ ∂ ∂
T µ1 µ2 ...µr ν1 ν2 ...νs = T |p dxµ1 |p , dxµ2 |p , ..., dxµr |p , ν1 , ν2 , ..., νs 
∂x ∂x ∂x
p p p

So, when physicist talk about tensors they actually mean the components of a tensor.

Example 4 Let T be a (1, 1) tensor. Then Tji := T (ϵi , ϵj ). Then,

T (ϕ, υ) = T (ϕi ϵi , υ j ϵj ) = ϕi υ j T (ϵi , ϵj ) = ϕi υ j Tji

Definiton 7. (Tensor Product) The tensor product of tensor A of type (r, s) and tensor B
of type (t, u) is a tensor A ⊗ B of type (r + t, s + u). The components are then

(A ⊗ B)σ1 σ2 ...σr σr+1 ...σr+t ρ1 ρ2 ...ρs ρs+1 ...ρs+u = Aσ1 σ2 ...σr ρ1 ρ2 ...ρs B σr+1 ...σr+t ρs+1 ...ρs+u

Tensor Product have the following properties.

(A ⊗ B) ⊗ C = A ⊗ (B ⊗ C)

A ⊗ (B + C) = A ⊗ B + A ⊗ C
(A + B) ⊗ C = A ⊗ C + B ⊗ C
Don’t sit around. Verify them!
We can show that Tensors of same type form a vector space, which is called a tensor space. It’s
usually written as ℑ(r, s) if the tensors are (r, s) type. ( verify it). Let T ∈ ℑ(r, s) be a (r, s) type
tensor. And let the basis of the underlying vector space V to be {ϵµ } and the dual basis to be {ϵν }. a

T µ1 µ2 ...µr ν1 ν2 ...νs = T (ϵµ1 , ϵµ2 , ..., ϵµr , ϵν1 , ϵν2 , ..., ϵνs )
Then for any ω 1 , ω 2 , ..., ω r ∈ V ∗ and v1 , v2 , ..., vs ∈ V , we have ω p = ωµp ϵµ and vq = vqν ϵν . Then,

T (ω 1 , ..., ω r , v1 , ..., vs ) = T (ωµ1 ϵµ1 , ...., ωµr ϵµr , ...., v1ν ϵν1 , ...., vsν ϵνs )
= ωµ1 ...ωµr v1ν ...vsν T (ϵµ1 , ..., ϵµr , ϵν1 , ..., ϵνs )
= ωµ1 ...ωµr v1ν ...vsν T µ1 µ2 ...µr ν1 ν2 ...νs
= T µ1 µ2 ...µr ν1 ν2 ...νs ϵν1 ⊗ ... ⊗ ϵνs ⊗ ϵµ1 ⊗ ... ⊗ ϵµr (ω 1 , ..., ω r , v1 , ..., vs )

8
Remark

Here we haves used the fact that ω k = ωµk ϵµ =⇒ ω k ϵµ = ωµk ϵµ ϵµ =⇒ ωµk = ω k ϵµ Thus, T can
be written as direct product of the bases of V and V ∗ as

T = T µ1 µ2 ...µr ν1 ν2 ...νr ϵν1 ⊗ ... ⊗ ϵνs ⊗ ϵµ1 ⊗ ... ⊗ ϵµs

Definiton 8. (Tensor Field) An (r, s) tensor field is a map T that maps any point p ∈ M to
an (r, s) tensor T |p at p. Given r 1-forms η1 , ..., ηr and s vector fields X1 , ..., Xs we can define
a function M → R by
p 7→ T |p (η1 |p , ..., ηr |p , X1 |P , ..., Xs |p )
The tensor field T is called smooth if this funtion is smooth for any r 1-forms and s vector
fields.

9
3 Transformation Laws

Let V be a vector space with a basis {ϵı } and the dual basis be {ϵȷ }.

Example 5 Now let’s first take an (0,2) tensor A. And let’s choose a new basis of V , {fi }
which are related to the previous basis by

fi = aji ϵj

Then the components of A w.r.t the new basis are

Ajk = A(fj , fk )
= A(am l
j ϵm , ak ϵl )

= am l
j ak A(ϵm , ϵl )

= am l
j ak Aml

Example 6 Let A be a (1, 2) tensor. And let’s choose a new dual basis of V ∗ , ϕi which are


related to the previous basis by

ϕi = bim ϵm
Then the components of A w.r.t the new basis are

Af,k k
ij = A(ϕ , fi , fj )

= A(bkm ϵm , ani ϵn , alj ϵl )

= bkm ani alj A(ϵm , ϵn , ϵl )
= bkm ani alj Aϵ,m
nl

Note: The upper index written before comma of Af,k ij indicates that it’s in the f basis. Some-
times I will not write that index because it’d be clear from the context which basis it’s in.

10
 If V is a tangent space at a point p on the manifold M then the bases are obtained as coordinate
vector fields with respect to two system of coordinates (xi ) and (y i ).
∂
ϵi = , ϵi = dxi
∂xi
∂
fi = , ϕi = dy i
∂y i
Now we have to find the relationship between the bases. Let’s first work with the basis of our vector
space V .

 
∂
f = ∂i (f ◦ x−1 )
∂xi p
x(p)

= ∂i (f ◦ y −1 ◦ y ◦ x−1 ) x(p)

= ∂i (f ◦ y −1 ) ∂ (y ◦ x−1 )
y(p) j x(p)
j
∂f ∂y
=
∂y i p ∂xi p

So
∂ ∂y j ∂ ∂y j
ϵi = i
= i i
= fj
∂x ∂x ∂y ∂xi
We know that ϵi = aji fj . So,
∂y j
aji =
∂xi
Now for the dual basis. First we need a definition,

Definiton 9. (Gradient of a function) Let f be a function. Then,

(df )p : Tp M → R, (df )p X = X(f )

Where X ∈ Tp M and can be written in terms of basis of Tp M as,

i ∂
X = X(x)
∂xi p

Now we can write a (1, 0) tensor Ai w.r.t a new basis as,

ϕi = dy i = bij ϵj = bij dxj

Now we have to find the bj ’s, Now we’ll use the definition of gradiant. we take our function to be
∂
f = ∂x j.

((dy i )p )f = bij ((dxj )p )f

   
i ∂ i j ∂
((dy )p ) = bj ((dx )p )
∂xj ∂xj
∂y i ∂xj
= bij
∂xj p ∂xj p
i
∂y
bij =
∂xj p

11
So Now we can write the tensor from Example 6 in terms of the basis of the tangent space as,

Akij = bkm ani alj Am

nl
∂y k ∂y n ∂y l m
= A
∂xm ∂xi ∂xj nl

12
4 Invariants

4.1 Trace
Scalar-valued functions of tensors frequently are described in terms of the components of the tensors
with respect to a certain basis. If these values do not depend on the basis employed, the functions are
called invariants, or, more precisely, scalar invariants. One may also speak of tensor invariants when
the values are tensors themselves rather than scalars. As an illustration of these concepts we define
an invariant of tensors of type (1, 1), the trace, which is a well-known invariant of matrices. We have
already seen how these tensors may be considered as matrices. We definr

trace of A = tr A = Aii .

4.2 Contraction
Definiton 10. (Contraction) Suppose T is a (r, s) tensor. The contraction of T w.r.t a
contravariant index µa and a covariant index νb is given by setting µa = νb = u and evaluating

T ′ ≡ T µ1 ...u...µr ν1 ...u...νs

and T ′ now is a (r − 1, s − 1) tensor.

An example will make it more clear.

Example 7 Lets say we have a (2, 2) tensor A. The components of the tensor are given by
Aνµ11νµ22 . Contraction of A w.r.t µ2 , ν1 :

A′ = Aµuν1 2u
= Aµ1ν111 + Aµ2ν112 + Aµ3ν113

(for simplicity let’s say we are in R3 ).

Now A′ is a (1, 1) tensor now.

Here are some useful example of contracted tensors:

a · b = ai bi = δij ai bj
∂2 ∂2
∇2 ϕ = = δ ij
∂xi ∂xi ∂xi ∂xj
k
∂v
(∇ × v) = ϵijk j
∂x
(a × b) · c = δij ci ϵijkjkl ak bl = ϵijk ci ak bl

13
5 Symmetric and Antisymmetric Tensors

5.1 Symmetric Tensor

Definiton 11. (Symmetric Tensor) It’ll be a lot easier to give you an example of a symmetric
tensor of type (0,3) and (0,2) tensor.

1. Let T be a (0, 2) tensor. Then the components of the tensor is given by Tij . Now it’ll be a
symmetric tensor if
Tij = Tji

2. Let T be a (0, 3) tensor. Then the components of the tensor is given by Tijk . Now it’ll be a
symmetric tensor if
Tijk = Tjik = Tikj = ..
An example of a symmetric tensor that we always use is the dot product.

5.2 Antiymmetric Tensor

Definiton 12. (Antisymmetric Tensor) Again I’ll present an example and it’ll be clear from
the example what it means. Let T be a (0, 3) tensor. Then the components of the tensor is given
by Tijk . Now it’ll be a antisymmetric tensor if

Tijk = −Tjik = Tjki = −Tkji = ...

If a tensor changes sign under exchange of each pair of its indices, then the tensor is completely
(or totally) antisymmetric. Cross product is an example of antisymmetric tensor.

Let T be a tensor of type (0, s) with components Tijk... and a pair of indicies i and j T has a
symmetric and antisymmetric parts defined as
1
T(ij)k... = (Tijk.. + Tjik... ) (Symmetric part)
2
1
(Tijk.. − Tjik... ) (Antiymmetric part)
T[ij]k... =
2
Every tensor of rank 2 can be decomposed into a symmetric and anti-symmetric pair as:
1 1
Tij = (Tij + Tjk ) + (Tij − Tjk )
2 2
This decomposition is not in general true for tensors of rank 3 or more.

5.3 Tensor Density

We define the Levi-Civita symbol as


+1 if µ1 µ2 ...µn is an even permutation of 012...(n − 1)
ϵµ1 µ2 ...µn = −1 if µ1 µ2 ...µn is an odd permutation of 012...(n − 1)
0 otherwise


14
 Note that the Levi-Civita symbol does not transform as a tensor even though it looks like a tensor
because of the indices. It is defined not to change under coordinate transformation. But we can make
a tensor out of it. We know that it’s behaviours can be related to that of an ordinary tensor by first
noting that given any n × n matrix M µ ′µ the determinant |M | obeys

ϵµ′1 µ′2 ...µ′n |M | = ϵµ1 µ2 ...µn M1µ µ′ M2µ µ′ ...Mnµ µ′n

1 2

µ
∂x
setting M µ µ′ = ∂xµ′
we have
′
∂xµ ∂xµ1 ∂xµ2 ∂xµn
ϵµ′1 µ′2 ...µ′n = ϵ µ µ ...µ ′ ′ ... ′
∂xµ 1 2 n
∂xµ1 ∂xµ2 ∂xµn
µ µ′
∂x ∂x
where we have used the fact that the inverse of ∂x µ′ is ∂xµ . So the Levi-Civita symbol transforms in a
way close to the tensor transformation law except fot the determinant out front. Objects transforming
in this way are known as Tensor densities. Another example is given by the determinate of the
metric g = |gµν | ( you’ll learn about it in section 7. It’s easy to check, by taking the determinant of
∂xµ ∂xν
both side of gµ′ ν ′ = ∂x µ′ ∂xµ′ gµν that under coordinate transformation we get

′
∂xµ −2
g(xµ′ ) = g(xµ )
∂xµ
Therefore g is also not a tensor. It transforms in a way similar to the Levi-Civita symbol, except that
the Jacobin is raised to the −2 power. The power to which the Jacobin is raised is known as the weight
of the tensor density. The Lev-Civita symbol is a density of weight, whole g is a density of weight −2.
However there is a simple way to convert a density to an honest tensor- multiplying by |g|w/2 where w
is the weight of the density ( the absolute value signs are there because g < 0 for Lorentzian metrics)/
The result will transform according to the tensor transformation law. Therefore we can define the
Levi-Civita Tensor as
ϵµ1 µ2 ...µn M1µ µ′
p
ϵµ′1 µ′2 ...µ′n = |g|e
1

we can also define it covariant indices as

′ ′ ′ 1 µ1 µ2 ...µn µ
ϵµ1 µ2 ...µn = p e ϵ M1 µ′
|g| 1

contracting it’s n − 1 indices we can get

ϵµ1 µ2 ...µn−1 α ϵµ1 µ2 ...µn−1 β = (−1)s (n − 1)!δβα

Here s = 1 for Lorentzian signature.

15
6 Covariant Derivative

Definiton 13. (Covariant Derivative) Let X(M ) be the space of all the vector fields on M .
Then a covariant derivative on map M is a map

∇ : X(M ) × X(M ) → X(M )

(X, Y ) 7→ ∇X Y
which satisfy the following properties:

1. ∇f X+gY Z = f ∇X Z + g∇Y Z, where f, g ∈ C ∞ (M ) and X, Y, Z ∈ X(M ).

2. ∇X (Y + Z) = ∇X Y + ∇X Z
3. ∇X (f Y ) = f ∇X Y + (∇X f )Y , where ∇X f is defined as ∇X f = X(f )
∇ is also called the connection.

Remark

If Y is a (1,0) tensor field ( a vector field), then ∇Y can be considered a (1, 1) tensor field in
the following way
(∇Y )(η, X) := η(∇X Y ) ∈ C ∞ (M )
In other words, ∇ increases the covariant index by 1. Then (1, 1) tensor field ∇Y has components
µ
(∇)µ n u, also denoted by Y,nu
µ
≡ ∂Y
∂xν . Now consider again,

(∇Y )(η, X) := η(∇X Y ) ∈ C ∞ (M ) =⇒ (∇)µ ν ηµ X ν = ηµ (∇X Y )µ

∴ (∇X Y )µ = (∇Y )µ ν X ν = Y,νµ X ν .

Now consider two vector fields X = X µ ∂x∂ µ and Y = Y ν ∂x∂ ν . Then,

∂
∇X Y = ∇X µ Yν ν
∂

 ∂x
∂xµ

∂
= X ∇ ∂µ Y ν ν
µ
∂x ∂x
ν
 
∂Y ∂ ∂
= Xµ µ + X µ ν
Y ∇ ∂
∂x ∂xν ∂xµ ∂xν

∂
Since ∇ ∂
∂xν is a vector field we can expand it in terms of our basis w.r.t the chart (U, x) as,
∂xµ

∂ ∂
∇ ν
∂:= Γσνµ σ
∂x ∂xµ ∂x
where Γσνµ are called the Connection Coefficient functions of ∇ on the manifold M w.r.t to chart
(U, x). But be careful. Γσνµ ’s doesn’t transform as a tensor. So,

∂Y σ ∂ ∂
∇X Y = X µ µ σ
+ X µ Y ν Γσνµ σ
∂x ∂x ∂x
σ
 
µ ∂Y µ ν σ ∂
= X µ
+ X Y Γνµ
∂x ∂xσ

16
 σ ∂Y σ
∴ (∇X Y ) = X µ + X µ Y ν Γσνµ
∂xµ
Furthermore,

∂Y σ ∂Y σ
 
σ ν σ µ
(∇X Y ) = µ
+ Y Γ νµ X =⇒ Y;µσ = (∇Y )σ µ = + Y ν Γσνµ
∂x ∂xµ

In other words,
Y;µσ = Y,µ
σ
+ Γσνµ Y ν

17
7 Metric Tensor

Definiton 14. Definition (Metric Tensor). A metric tensor at p ∈ M is a tensor of type

(0, 2), i.e. it is a map
g|p : Tp M × Tp M → R
with the following properties:
1. (symmetry) For Xp , Yp ∈ Tp M , g|p (Xp , Yp ) = g|p (Yp , Xp ).
2. (Non-degeneracy) If g|p (Xp , Yp ) = 0 for all Yp ∈ Tp M , then Xp = 0.

Remark

In the chart (U, x), g|p = gµν (p)dxµ p ⊗ dxν p . g, as a (0, 2) tensor field can also be written
as g = gµν dxµ dxν . Since the components gµν (p) of g|p form a symmetric matrix ( from the
symmetry of the metric tensor) such a matrix can be diagonalized following an appropriate
choice of basis vector. Non-degeneracy implies that none of the diagonal element will be 0.
Because, if gii = 0 after diagonalization, we take Xp ∈ Tp M such that X i = 1 and X j = 0 for
i ̸= j in that basis. Then

g|p (Xp , Yp ) = gµν X µ Y ν = giν X i Y ν = gii X i Y i = 0

for any Yp ∈ Tp M. This contradicts non=degeneracy. So none of the diagonal elements are
0. Then after scaling the basis vectors approproately we can make the diagonal elements ±1.
Such a basis is then called an orthonormal basis.

In differential geometry, one is interested in Riemannian metrics. For such metrics, the signature
is + + ...+, i.e., all diagonal elements are +1 in an orthonormal basis. But in general relativity,
we are interested in Lorentzian metrics, i.e., those are with signature − + +...+.

Definiton 15. (Inverse Metric) The inverse metric g −1 |p w.r.t the metric g|p is a (2, 0)
tensor given by the map
g −1 |P : Tp∗ M × Tp∗ M → R

Remark

The inverse metric g −1 is not the inverse of the metric g in the sense of a map, but in the matrix
sense. That is the inverse of g|p would be map like R → Tp M × Tp M which is not the same as
the map of g −1 |p . If we denote the component of g −1 |p as g ab |p then what we mean by inverse
is that the following holds:
g ab gbc = δcb

We can raise and lower the indicies of a tensor using the metric tensor. Such as,

Xa = gam X m

or
X m = g ma Xa

18
 Example 8 In Rn , the Euclidean metric is

g = dx1 ⊗ dx2 + .... + dxn otimesdxn

In the chart (x1 , ..., xn ), we have gµν =diag(1, 1, ..., 1)

Example 9 On S 2 let our chart be (θ, ϕ) denote the spherical polar coordinate chart. The
round unit metric on S 2 is
g = dθ ⊗ dθ + sin2 θ dϕ ⊗ dϕ
in the chart (θ, ϕ), we have gµν =diag(1, sin2 θ)

7.1 Lorentzian Signature

On a Lorentzian manifold, we take basis indices µ, ν to fun from 0 to n − 1. At p ∈ M we choose an
orhonormal basis {eµ } so that the matrix with entries of the metric components gets diagonalized, i.e.
 
g|p eµ p , eν p = ηµν = diag(−1, 1, 1, ..., 1)
n o
such a basis is not unique. Let e′µ p be any other such basis with

eµ |p = (A−1 )ν µ eν p

Then we have,
   
ηmuν = g p p
e′µ p , e′ν
= (A −1 σ
) µ e σ p
, (A −1 ρ
) ν e ρ p
 
−1 σ −1 ρ −1 σ
= (A ) µ (A ) ν g p eσ p , eρ p = (A ) µ (A−1 )ρ ν ησρ

∴ ηµν = Aµ σ Aν ρ ηµν
which is precisely the defining equations of a Lorentz transformation in special relativity. Hence,
different orthonormal basis at p ∈ M are related by Lorentz transformations. The A’s are actually the
Λ’s we see in special relativity.

Definiton 16. A Riemannian (Lorentzian) manifold is a pair (M, g) where M is a Rie-

mannian (Lorentzian) manifold and g is a Riemannian (Lorentzian) metric tensor field.

On a Riemannian manifold, one can now define the length of a curve as in R3 : if λ : (a, b) → M is
a smooth curve with tangent vector X|λ(t) then its length is
Z b

Z b
dt g|λ(t) X λ (t), X λ (t) = dt g(X, X)(λ(t)).
a a

Definiton 17. On a Lorentzian manifold (M,g), a non-zero vector Xp ∈ Tp M is timelike if

g|p (X p , X p ) < 0, lightlike if g|p (X p , X p ) = 0, and spacelike if g|p (X p , X p ) > 0

19
7.2 Geodesic Equation
Let p and q be points connected by a timelike curve, i.e. the tangent vectors at all point of the curve
are timelike. There are infinitely many timelike curves between p and q. The proper time between p
and q will be different for different curves. It’s a natural question to ask which curve extremizes proper
time. Consider timelike curves from p to q with parameter u such that λ(0) = p and λ(1) = q. The
proper time between p and q along sucg a curve is given by the funtional
Z 1
τ [λ] = dG(x(u), ẋ(u)),
0
p
where G(x(u), ẋ(u)) = −gσν We are writing xµ (u) as a shorthand for xµ (λ(u)).
(λ(u)) ẋµ (u)ẋν (u).
The curve that extrimizes proper time must satisfy the Euler-Lagrange equation.
 
d ∂G ∂G
− = 0.
du ∂ ẋµ ∂xµ
Recall that,
p
G(x(λ(u)), ẋ(λ(u))) = −gσν (λ(u)) ẋµ (u)(λ(u))ẋν (u)(λ(u)) (1)
∂G 1
µ
= [−gσν δµσ ẋν − gσν ẋσ δµν ] (2)
∂ ẋ 2G
1
= [−gµν ẋν − gσµ ẋσ ] (3)
2G
1
= [−gµν ẋν − gµσ ẋσ ] (4)
2G
gµν ν
=− ẋ (5)
G
∂G 1
= [−gσν,µ ẋσ ẋν ] (6)
∂xµ 2G
Now recall r
dτ dxµ (λ(u)) dxν (λ(u))
= −gµν (λ(u)) =G
du du du
d d
so, = G . Now Euler-Lagrange equation reads
du dτ
d  gµν ν  1
− ẋ + gσν,µ ẋσ ẋν = 0
du G 2G

dxσ dxν
 
d 1 d ν 1
=⇒ − G gµν G x + gσν,µ G ·G =0 (7)
dτ G dτ 2G dτ dτ
dxν dxσ dxν
 
d G
=⇒ − G gµν + gσν,µ =0 (8)
dτ dτ 2 dτ dτ
dxν dxσ dxν
 
d 1
∴ gµν − gσν,µ =0
dτ dτ 2 dτ dτ
Hence,
d2 xν d dxν 1 dxσ dxν
gmuν + µν − gσν,µ =0 (9)
dτ 2 dτ dτ 2 dτ dτ
d2 xν ∂gµν dxρ dxν 1 dxσ dxν
=⇒ gmuν 2
+ ρ
− gσν,µ =0 (10)
dτ ∂x dτ dτ 2 dτ dτ
d2 xν dxρ dxν 1 dxρ dxν
=⇒ gmuν + gµν,ρ − gρν,µ =0 (11)
dτ 2 dτ dτ 2 dτ dτ

20
Now we observe that
∂gµν dxρ dxν ∂gµρ dxν dxρ
=
∂xρ dτ dτ ∂xν dτ dτ
Therefor we write

dxρ dxν 1 ∂gµν dxρ dxν 1 ∂gµρ dxν dxρ

gµν,ρ = ρ
+ (12)
dτ dτ 2 ∂x dτ dτ 2 ∂xν dτ dτ
1 dxν dxρ
= (gµν,ρ + gµρ,ν ) (13)
2 dτ dτ
Now (11) reads

d2 xν dxρ dxν 1 dxρ dxν

gmuν + gµν,ρ − gρν,µ =0 (14)
dτ 2 dτ dτ 2 dτ dτ
d2 xν 1 dxν dxρ
=⇒ gmuν 2
+ (gµν,ρ + gµρ,ν − gνρ,µ ) =0 (15)
dτ 2 dτ dτ
Contraction with g σµ on both sides, we get

d2 xν 1 dxν dxρ
g σµ gmuν 2
+ g σµ (gµν,ρ + gµρ,ν − gνρ,µ ) =0 (16)
dτ 2 dτ dτ
d2 xν 1 σµ dxν dxρ
=⇒ δνσ + g (gµν,ρ + g µρ,ν − gνρ,µ ) =0 (17)
dτ 2 |2 {z } dτ dτ
Γσ
νρ

(18)

d2 xσ ν
σ dx dx
ρ
∴ + Γ νρ =0 (19)
dτ 2 dτ dτ
Here the Γσνρ ’s are the connection coefficients for the Levi-Civita Connection and they have a special
name. They are called the Christoffel symbols. And equation (18) is called the geodesic equation.
Remark
In Minkowski spacetime, metric in an inertial reference frame is constant. Therefore Γσνρ ’s are
0. Then we’ll have
d2 xσ
=0
dτ 2
Which is the equation of motion of a free particle.

21