Chapter 3

Tensors in Special Relativity

We need to express the equations of physics in a frame independent way. This can
be done for some equations by using four - vectors [ e.g. Newton’s Laws ] but what
about the rest of physics!!?
We now extend this idea by introducing more general coordinate indepen-
dent quantities called Tensors. In this section we will confine our discussion to
Minkowski spacetime, although the concepts will apply to curved spacetime and
General Relativity.

3.1 Metrics and forms

Recall that the scalar product of two four - vectors is A · B:
! "
A · B = (Aα eα ) B β eβ = Aα B β eα · eβ , (3.1)

so
A · B = Aα B β ηαβ , (3.2)

where
−1 0 0 0
 
 0 1 0 0 
ηαβ =  (3.3)
 
0 0 1 0

 
0 0 0 1
are the components of the metric tensor.
We can think of the metric tensor as a map which takes two four - vectors A
and B into the reals: g(A, B) = A · B.

29
CHAPTER 3: TENSORS IN SPECIAL RELATIVITY 30

The map is linear in the arguments, in the sense that

g(αA + βB, C) = αg(A, C) + βg(B, C) . (3.4)

We can generalize this by defining a tensor of type 0/N as a map which takes N
four - vectors into the reals which is linear in all its arguments, for example the
metric tensor is a type 0/2 tensor.

3.2 One - forms

These are tensors of type 0/1 and they map four - vectors into the reals. They are
denoted by p̃, so p̃(A) is a real number. One - forms form a vector space since

(p̃ + q̃) A = p̃(A) + q̃(A) , (αp̃) A = αp̃(A) . (3.5)

It is called the dual vector space to distinguish it from the space of four - vectors.
The components of p̃ are pα = p̃(eα ). We can write

p̃ →O {pα } . (3.6)

Generally we have

p̃(A) = p̃(Aα eα ) = Aα p̃(eα ) = Aα pα = A0 p0 + A1 p1 + A2 p2 + A3 p3 . (3.7)

Note that the signs are positive [ c.f A · B = −A0 B 0 + A1 B 1 + A2 B 2 + A3 B 3 ]. We

can think of vectors as columns and one - forms as rows:
A0
 
 A1 
p̃(A) = (p0 p1 p2 p3 )  
. (3.8)
A2
 
 
A3

p̃(A) is more fundamental than A · B since the latter is defined only if there is a
metric i.e. A · B = Aα B β ηαβ .
Now let us look at how one - forms transform:
! "
pβ̄ = p̃(eβ̄ ) = p̃ Λα β̄ eα

= Λα β̄ pα . (3.9)
CHAPTER 3: TENSORS IN SPECIAL RELATIVITY 31

A
~
p

Figure 3.1: p̃(A) is the number of surfaces the vector A pierces.

So one - forms transform like basis vectors, not like vector components.
Now since
! "
Aᾱ pᾱ = Λᾱ β Aβ (Λµ ᾱ pµ )

= δ µ β Aβ pµ , (3.10)

we have
Aᾱ pᾱ = Aµ pµ , (3.11)

so p̃(A) is frame independent.

We can introduce a one - form basis {w̃ α }, so that

p̃ = pα w̃α . (3.12)

Then
p̃(A) = pα w̃α (A) = pα w̃α (Aβ eβ ) = pα Aβ w̃α (eβ ) = pα Aα , (3.13)

so we must have
w̃α eβ = δ α β . (3.14)

This gives the basis for one - forms. It is said to be dual to {eα }.
One can show that
w̃ᾱ = Λᾱ β w̃β , (3.15)

so the basis one - forms transform like vector components [ as required notationally
].
Both vectors and one - forms have four components but they have different
geometrical interpretation. Vectors are like arrows but one - forms can be thought
CHAPTER 3: TENSORS IN SPECIAL RELATIVITY 32

ct
α
U
φ=φ( ct,x,y,x)

x
Figure 3.2: World line of particle with four - velocity U.

of as like three dimensional surfaces with the spacing between the surfaces defining
the magnitude of p̃ [ see Figure 3.1 ].

3.2.1 Gradients
For a scalar field φ(x) and a world line of some particle x(τ ), we have

φ(τ ) = φ (ct(τ ), x(τ ), y(τ ), z(τ )) . (3.16)

If U is the tangent to the curve [ the four - velocity of the particle, see Figure 3.2
] then:
dφ ∂φ dt ∂φ dx ∂φ dy ∂φ dz
=c + + + (3.17)
dτ ∂ct dτ ∂x dτ ∂y dτ ∂z dτ
so
dφ ∂φ 0 ∂φ 1 ∂φ 2 ∂φ 3
= U + U + U + U , (3.18)
dτ ∂ct ∂x ∂y ∂z
dx
since U = dτ
[ see section 2.2 ].
This defines a one - form since it maps U into real numbers and represents the
rate of change of φ along a curve with tangent U.
) *
˜ →O ∂φ ∂φ ∂φ ∂φ
dφ , , , . (3.19)
∂ct ∂x ∂y ∂z
In three dimensions one thinks of a gradient as a vector [ normal to surfaces of
constant φ ] but d̃ is a one - form and specifies a vector only if there is a metric.
Now how do the components of d̃φ transform?
! " ! "
d̃φ = Λβ ᾱ d̃φ . (3.20)
ᾱ β
CHAPTER 3: TENSORS IN SPECIAL RELATIVITY 33

But we also have by the chain rule:

∂φ ∂φ ∂xβ
= , (3.21)
∂xᾱ ∂xβ ∂xᾱ
which means that
! " ∂xβ ! "
d̃φ = d̃φ , (3.22)
ᾱ ∂xᾱ β

so
β ∂xβ
Λ ᾱ = ᾱ , (3.23)
∂x
and since xα ,β = δ α β we have
∂φ
d̃xα = w̃α , d̃φ = d̃xα . (3.24)
∂xα
This is a useful result, that the basis one - form is just d̃xα .

3.2.2 The metric as a mapping of vectors onto one - forms

We now introduce the idea that the metric acts as a mapping between vectors and
one - forms. To see how this works, consider g and a single vector V. Since the
metric g requires two vectorial arguments, the expression g (V, ) still lacks one:
when another one is supplied, it becomes a number. Therefore g (V, ) considered
is a linear function of vectors producing real numbers: a one - form. We call it
Ṽ ( ).
So we have
g (V, ) = Ṽ( ) . (3.25)

Then Ṽ is a one - form whose value on a vector A is V · A:

Ṽ (A) = g (V, A) = V · A . (3.26)

What are the components of Ṽ. They are

Vα = Ṽ (eα ) = V · eα = eα · V
! "
= eα · V β eβ

= V β (eα · eβ )

= ηαβ V β . (3.27)
CHAPTER 3: TENSORS IN SPECIAL RELATIVITY 34

Thus
V0 = V β η0β = V 0 η00 = −V 0 (3.28)

and
V1 = V β η1β = V 1 η11 = V1 . (3.29)

This can be summarized as follows: If

V →O (a, b, c, d) , (3.30)

then
Ṽ →O (−a, b, c, d) . (3.31)

The components of Ṽ are obtained from those of V by changing the sign of the
time component.
Since det ηαβ = −1 [ i.e. non zero ], there exists an inverse metric which we can
write as η αβ [ In Special Relativity the components of ηαβ are the same as η αβ , but
this will not be true in the curved spacetime of General Relativity ].
The inverse metric defines a map from one - forms to vectors

Aα = η αβ Aβ . (3.32)

In particular, we can map the gradient one - form d̃φ into a vector gradient:
) *
∂φ ∂φ ∂φ ∂φ
dφ →O − , , , . (3.33)
∂ct ∂x ∂y ∂z
We can regard a vector as a 1/0 tensor i.e. a map from one - forms into the
reals, so
A (p̃) = Aα eα pβ w̃β = Aα pβ eα w̃β = Aα pα = p̃ (A) (3.34)

and
Aα = A (w̃α ) . (3.35)

The inverse metric η αβ can be used to define the magnitude of a one - form:

p̃2 = pα pβ η αβ = −p20 + p21 + p22 + p23 , (3.36)

and the scalar product of two one - forms:

p̃ · q̃ = pα qβ η αβ = −p0 q0 + p1 q1 + p2 q2 + p3 q3 . (3.37)
CHAPTER 3: TENSORS IN SPECIAL RELATIVITY 35

These are identical to the corresponding quantities for vectors, i.e.

p̃2 = p2 , p̃ · q̃ = p · q . (3.38)

Thus one - forms are timelike/spacelike/null if the corresponding vectors are.

3.3 More general tensors

3.3.1 Tensors of type 0/2
The product of two one - forms, written as p̃ ⊗ q̃ defines a linear map which takes
two vectors into the reals:

p̃ ⊗ q̃ (A, B) = p̃ (A) q̃ (B) . (3.39)

It is therefore a 0/2 tensor. ⊗ denotes the outer product. It is the formal notation
to show how the 0/2 tensor is formed from two one - forms.
Note that this product is non - commutative since p̃(B)q̃(A) gives a different
result [ EXERCISE 3.4 ] i.e.
p̃ ⊗ q̃ $= q̃ ⊗ p̃ . (3.40)

The most general 0/2 tensor is a linear sum of such outer products. So

f (A, B) = f (Aα eα , B β eβ )

= Aα B β f (eα eβ )

= Aα B β fαβ , (3.41)

where fαβ are the components of the map f and we have used linearity.
If we take a basis for f as w̃ αβ [ 16 components ], then

f = fαβ w̃αβ . (3.42)

But

fµν = f (eµ , eν ) = fαβ w̃αβ (eµ , eν )

⇒ w̃αβ (eµ , eν ) = δ α µ δ β ν , (3.43)

CHAPTER 3: TENSORS IN SPECIAL RELATIVITY 36

so we have
w̃αβ = w̃α ⊗ w̃β . (3.44)

Under a Lorentz transformation, the components of f become:

fᾱβ̄ = Λµ ᾱ Λν β̄ fµν . (3.45)

A 0/2 tensor is symmetric if

f (A, B) = f (B, A) ⇔ fαβ = fβα , (3.46)

and anti - symmetric if

f (A, B) = −f (B, A) ⇔ fαβ = −fβα . (3.47)

It follows that any 0/2 tensor can be uniquely decomposed into a symmetric and
anti - symmetric part.
fαβ = f(αβ) + f[αβ] , (3.48)

with the symmetric and anti - symmetric parts given by

f(αβ) = 12 (fαβ + fβα ) (3.49)

and
1
f[αβ] = 2
(fαβ − fβα ) . (3.50)

3.3.2 Tensors of type 0/N

We define a tensor of type 0/N as a linear map from N vectors into the reals:

T = Tαβ... w̃α ⊗ w̃β ⊗ . . . , (3.51)

and it’s components transform according to

Tᾱβ̄... = Λµ ᾱ Λν β̄ . . . Tµν... (3.52)

It can be split into symmetric and anti - symmetric parts on any two indices.
CHAPTER 3: TENSORS IN SPECIAL RELATIVITY 37

3.3.3 Tensors of type 2/0

Recall that a vector can be regarded as a 1/0 tensor [ i.e. a map of one - forms into
the reals ]. Therefore the outer product of two vectors defines a 2/0 tensor, i.e. it
is a linear map from two one - forms into the reals:

V ⊗ W(p̃, q̃) = V(p̃)W(q̃) . (3.53)

It follows that the most general 2/0 tensor is a linear sum of such outer products:

f (p̃, q̃) = pα qβ f αβ , (3.54)

where
f αβ = f (w̃α, w̃β ) , f = f αβ eα eβ . (3.55)

Under a Lorentz transformation, the components of f are:

f ᾱβ̄ = Λᾱ µ Λβ̄ ν f µν . (3.56)

As before we can separate f into symmetric and anti - symmetric parts:

! "
f (αβ) = 1
2
f αβ + f βα , (3.57)

and
! "
f [αβ] = 1
2
f αβ − f βα . (3.58)

3.3.4 Tensors of type M/0

We define a tensor of type M/0 as a linear map that takes M one - forms into the
reals. So we can write
T = T αβ... eα ⊗ eβ ⊗ . . . , (3.59)

whose components transform according to

T ᾱβ̄... = Λᾱ µ Λβ ν . . . T µν... . (3.60)

CHAPTER 3: TENSORS IN SPECIAL RELATIVITY 38

3.3.5 Tensors of type M/N

Finally we define a tensor of type M/N as a linear function that maps M one - forms
and N vectors into the reals:

T = T α... β... eα ⊗ . . . w̃β ⊗ . . . . (3.61)

The components of T transform according to

T ᾱ... β̄... = Λᾱ µ . . . Λν β̄ . . . T µ... ν... . (3.62)

These results just follow from linearity.

3.3.6 Index “raising” and “lowering”

In the same way that the metric maps a vector V into a one - form, it maps a M/N
tensor into a M − 1/N + 1 tensor, i.e. it lowers an index. Similarly the inverse
metric maps a M/N tensor into a M + 1/N − 1 tensor, i.e. it raises an index. So
for example

T α βγ = ηβµ T αµ γ ,

Tα β γ = ηαµ T µβ γ ,

T αβγ = η γµ T αβ µ . (3.63)

In Special Relativity, raising or lowering a 0 component changes the sign of the

component; raising or lowering 1, 2, or 3 components has no effect.
We can operate the inverse metric on the metric to get the Kroneker delta [
EXERCISE 3.5 ]:
δ α β = η αµ ηµβ = η α β . (3.64)

So far we have confined our attention to Lorentz frames [ i.e. inertial frames
]. We can also allow more general coordinate transformations in a more general
space i.e. xᾱ = xᾱ (xβ ). We then define

∂xᾱ
Λᾱ β = . (3.65)
∂xβ
CHAPTER 3: TENSORS IN SPECIAL RELATIVITY 39

Tensors will then transform as before, for example

Aᾱ... β̄... = Λᾱ µ . . . Λν β̄ Aµ... ν... . (3.66)

Old fashioned texts regard the above as the definition of a tensor. Raised indices
are called contravariant because they transform “contrary” to basis vectors:
∂xᾱ β
Aᾱ = A , (3.67)
∂xβ
while lowered indices are called covariant:

∂xβ
pᾱ = ᾱ . (3.68)
∂x
In particular one - forms are sometimes called covariant vectors, while ordinary
vectors are called contravariant vectors.

3.4 General properties of tensors

Tensors have the following general properties:

• If Aα... β... and B α... β... are M/N tensors, so are λAα... β... , µB α... β... and
λAα... β... + µB α... β... for any λ and µ. Thus tensors of a given type form a
vector space.

• If Aα... β... is a M/N tensor and B γ... δ... is of type M ! /N ! , then the outer
product Aα... β... B γ... δ... is a tensor of type M + M ! /N + N ! .

• For any tensor of type M/N, one can construct a tensor of type M − 1/N −
1 by contracting an upper index with a lower index for example Aαβ αγ is
constructed from Aαβ δγ [ δ → α or α → δ ]. Note however that there may
be several ways of construction and they give different tensors.

• Tensors can be symmetric on pairs of upper or lower indices for example

T αβ... γ... = T βα γ... and T α γδ = T α... δγ , or anti - symmetric. However symmetric
or anti - symmetric pairs of indices with one index up and the other down
cannot be defined since the relationship need not hold in all frames.
CHAPTER 3: TENSORS IN SPECIAL RELATIVITY 40

3.5 Tensor derivatives and gradients

The partial derivative of a tensor is not generally itself a tensor although it is for
∂φ
a scalar: ∂xα
is a 0/1 tensor.
For a vector A, we have
∂Aᾱ ∂xᾱ γ
) *
∂
= A
∂xβ̄ ∂xβ̄ ∂xγ
∂xδ ∂xᾱ ∂Aγ ∂ 2 xᾱ ∂xδ γ
= + A . (3.69)
∂xβ̄ ∂xγ ∂xδ ∂xδ ∂xγ ∂xβ̄
∂Aα
It follows that ∂xβ
transforms like a 1/1 tensor only if the second term is zero, i.e.

∂ 2 xᾱ ∂xδ
= 0
∂xδ ∂xγ ∂xβ̄
⇒ xᾱ = B ᾱ β xβ + C ᾱ , (3.70)

where B ᾱ β and C ᾱ are constants. This is true for flat space with coordinates
(ct, x, y, z) under the Poincaré transformations and in this case B ᾱ β may be inter-
preted as a 4D rotation matrix.

• Note however that it is not true for flat space with a general coor-
dinate system or in the curved spacetime of General Relativity.

We can define the derivative of a general M/N tensor T along a curve param-
eterized by the proper time τ as follows:
+ ,
dT T (τ + δτ ) − T (τ )
= lim . (3.71)
dτ δτ →0 δτ
If the basis vectors and one - forms are the same everywhere then:
dT α...
) *
dT
= eα ⊗ . . . ⊗ w̃β ⊗ . . . (3.72)
dτ dτ
where
dT α... β... ∂T α... β... δ
= U , (3.73)
dτ ∂xδ
and U is the tangent to the curve. Thus dT dτ
is also like a M/N tensor, written as

∇U T →O {T α... β...,γ U γ } , (3.74)

CHAPTER 3: TENSORS IN SPECIAL RELATIVITY 41

where , γ means ∂/∂xγ . We can then define a M/N + 1 tensor ∇T:

∇T ≡ T α... β...,γ eα ⊗ . . . ⊗ w̃β ⊗ . . . ⊗ w̃γ . (3.75)

This is the tensor gradient [ remember the gradient of a scalar d̃φ ].