Lorentz tensor redux∗

Emily Nardoni

Contents

1 Introduction 1

2 The Lorentz transformation 2

3 The metric 4

4 General properties 5

5 The Lorentz group 5

1 Introduction

A Lorentz tensor is, by definition, an object whose indices transform like a tensor under Lorentz
transformations; what we mean by this precisely will be explained below. A 4-vector is a tensor with
one index (a first rank tensor), but in general we can construct objects with as many Lorentz indices
as we like.

Recall that we write a 4-vector in components as

A0
 
 A1 
Aµ =  
 A2  (1)
A3

where we use Greek indices to run over all the spacetime indices, µ ∈ [0, 3]. We’ve already seen many
examples of 4-vectors in this course; for instance1 ,
∗
As discussed in 4D recitation section on 3/2/15 @UCSD.
1
We think of 4-vectors with upper indices as column vectors, and 4-vectors with lower indices as row vectors, but when
listing the components here we don’t worry about organizing them into rows vs. columns.

1
 xµ = (ct, x, y, z)
 
1
∂µ = ∂t , ∂x , ∂y , ∂z
c
ω 
kµ = , kx , ky , kx the wave-vector;
c
µ
j = (cρ, J)~ for ρ the electric charge density,
J~ the electric current density;
dxµ dxµ
uµ = =γ = γ(c, ~v )
dτ  dt 
E
pµ = muµ = , p~ , E = γmc2 ;
c
dpµ
fµ =
dτ
d2 xµ
aµ =
dτ 2

We’ve also seen many conservation equations and examples of Lorentz invariant quantities, like

∂µ j µ = 0 : current conservation;
 
2 1 2 ~2
∂ A= ∂ − ∇ A : the wave equation, whose form
c2 t
is invariant under Lorentz transformations;
∆s2 = (c∆t)2 − (∆~x)2 the invariant length squared,
√
⇒ c∆τ ≡ ∆s2 is a Lorentz scalar, the proper time;
u2 = c2
 2
2 2 E
p = (mc) = − p~2
c
X
pµ = (constant)µ : relativistic energy/momentum conservation
particles

for a closed system.

2 The Lorentz transformation

First, we write the components of the Lorentz transformation matrix in index notation. Recall that to
transform the components of a 4-vector (let’s for now just consider the 4-vector ∆xµ ) from an unprimed
frame to a frame which is moving at speed v in the +x̂ direction relative to F (call it the primed frame),
we use the Lorentz transformation

∆x00 ∆x0
    
γ −γβ 0 0
 ∆x01   −γβ γ 0 0   ∆x1 
 ∆x02  =  0 (2)
    
0 1 0   ∆x2 
∆x03 0 0 0 1 ∆x3

2
where γ = √ 1
, β = vc . Now we write the components of the Lorentz transformation matrix as Λµν ,
1−β 2
where µ is a row index and ν is a column index, such that

Λ00 Λ01 Λ02 Λ03

 
 Λ10 Λ11 Λ12 Λ13 
Λ=
 Λ20
 (3)
Λ21 Λ22 Λ23 
Λ30 Λ31 Λ32 Λ33

Then, the Lorentz transformation for ∆xµ can be written in the compact notation

3
X
(∆x0 )µ = Λµν ∆xν
ν=0
(4)
≡ Λµν ∆xν

We use the Einstein summation convention, meaning that whenever you see two of the same index on
one side of an equation, you sum over all the values of that index. The index ν which is summed over
lives in the unprimed frame, while the free index µ lives in the primed frame. In particular, by Λµν
we mean the components of the Lorentz transformation matrix which transforms the components of a
4-vector in a frame associated with the bottom index ν to a frame associated with the top index µ.

For example, for the case of the transformation in Eq. 2, we have Λ00 = Λ11 = γ, Λ01 = Λ10 =
−γβ, Λ22 = Λ33 = 1, and the rest of the components are zero. Then

(∆x0 )µ = Λµ0 ∆x0 + Λµ1 ∆x1 + Λµ2 ∆x2 + Λµ3 ∆x3

= Λµ0 c∆t + Λµ1 ∆x + Λµ2 ∆y + Λµ3 ∆z

⇒(∆x0 )0 = (c∆t0 ) = γ(c∆t) − γβ∆x (5)

0 1 0
(∆x ) = (∆x ) = −γβ(c∆t) + γ∆x
(∆x0 )2 = (∆y 0 ) = ∆y
(∆x0 )3 = (∆z 0 ) = ∆z

is the usual Lorentz transformation to a frame moving in the +x̂ direction.

The inverse Lorentz transformation should satisfy (Λ−1 )βµ Λµν = δ βν , where δ βν ≡ diag(1, 1, 1, 1) is the
Kronecker delta. Then, multiply by the inverse on both sides of Eq. 4 to find

(Λ−1 )βµ (∆x0 )µ = δ βν ∆xν

(6)
= ∆xβ

The inverse (Λ−1 )βµ is also written as Λµβ . The notation is as follows: the left index denotes a row
while the right index denotes a column, while the top index denotes the frame we’re transforming to
and the bottom index denotes the frame we’re transforming from. Then, the operation Λµβ Λµν means

3
sum over the index µ which lives in the primed frame, leaving unprimed indices β and ν (so that the
RHS of Eq. 6 is unprimed as it should be), where the sum is over a row of Λµβ and a column of Λµν ,
which is precisely the operation of matrix multiplication.

In particular, one can show that in terms of the components of Λ given in Eq. 3, the components of
Λ−1 will be given by

Λ00 −Λ10 −Λ20 −Λ30

 
 −Λ01 Λ11 Λ21 Λ31 
Λ−1 =
 −Λ02 Λ12
. (7)
Λ22 Λ32 
−Λ03 Λ13 Λ23 Λ33

As a special case, the inverse to the transformation in Eq. 2 gives

∆x00 ∆x0
    
γ γβ 0 0
 γβ γ 0 0   ∆x01   ∆x1 
= (8)
0 1 0   ∆x02
  
 0   ∆x2 
0 0 0 1 ∆x03 ∆x3

as we expect it to.

3 The metric

The metric ηµν is a special Lorentz tensor, which for Minkowski spacetime in our convention is given
by

 
1 0 0 0
 0 −1 0 0 
 0 0 −1 0  ≡ diag(1, −1, −1, −1).
η= (9)


0 0 0 −1

The other convention is to use ηµν = diag(−1, 1, 1, 1), which will change around minus signs in various
places. Different spacetime geometries have correspondingly different metrics, but since we live in
Minkowski spacetime we only use this form for the metric here.

We use the metric to raise and lower Lorentz indices. By definition Aµ ≡ ηµν Aν given a 4-vector Aµ
with an upstairs index. Think of Aµ as a column vector, and Aµ as a row vector.

The inverse metric η µα with upstairs indices satisfies η µα ηαν = δ µν ; then, one can show that η µα =
diag(1, −1, −1, −1). In other words, the Minkowski metric is its own inverse. We can then use the
inverse metric to raise indices, as in Aµ = η µα Aα given a 4-vector with a lower index. Note that a
general metric will not be its own inverse, but the Minkowski metric is a special case.

4
4 General properties

• We’ve already seen how to use the metric to raise and lower indices. In general, we use the
metric to raise and lower any good Lorentz index. For instance, given a tensor Tαβ , we can raise
an index by T µβ = η µα Tαβ ; or given a third-rank tensor F αβγ we could lower two indices by
F αµν = ηµβ ηνγ F αβγ .

• The number of free (unsummed) indices must match on the left-hand side and right-hand side of
an equation, while the labels used for indices which are summed over (repeated indices, or “dummy
indices”) are irrelevant. For instance, ∆xµ ∆xµ means exactly the same thing as ∆xα ∆xα since in
both cases we’re summing over the index values [0, 3]. Also, we’ve already seen that whether an
index appears upstairs or downstairs is important, so in particular the free indices should match
in both label and up vs. down placement on both sides of an equation.

• You are guaranteed that an object made up of tensors and 4-vectors with no free indices is Lorentz
invariant. We’ve seen many examples of this before; for instance the invariant interval defined by
∆s2 = ∆sµ ∆sµ = ηµν ∆sµ ∆sν = (∆s0 )2 − (∆s1 )2 − (∆s2 )2 − (∆s3 )2 = (c∆t)2 − (∆~x)2 gives the
same number when calculated in any Lorentz frame. In general, we call an object with no free
indices a Lorentz scalar.

5 The Lorentz group

With all these ingredients, we can write down the condition for an object Λ to be a Lorentz transfor-
mation:

ηµν = Λαµ Λβν ηαβ (10)

In the language of matrices, this translates to ΛT ηΛ = η, for ΛT the matrix transpose of Λ. As an

explicit example, consider the Lorentz transformation of Eq. 2. We note that ΛT = Λ. We can do out
the matrix multiplication, and explicitly verify that:

   
γ −γβ 0 0 1 0 0 0 γ −γβ 0 0
−γβ γ 0 0 0 −1 0 0   −γβ γ 0 0 
ΛT ηΛ = 
  
  
 0 0 1 0   0 0 −1 0  0 0 1 0 
0 0 0 1 0 0 0 −1 0 0 0 1
  (11)
1 0 0 0
 0 −1 0 0 
= 0 0 −1 0  = η.


0 0 0 −1

This condition is both necessary and sufficient for a 4 × 4 matrix Λ to leave the inner product of any
two 4-vectors invariant. One part of this statement is to show that given any two 4-vectors Aµ and B µ ,
A · B = A0 · B 0 :

5
 A · B = Aα B α = ηαβ Aα B β ; (12)
0 0
A ·B = A0α B 0α 0α
= ηαβ A B 0β

= ηαβ (Λαµ Aµ )(Λβν B ν )

= (ηαβ Λαµ Λβν ) Aµ B ν
= ηµν Aµ B ν = Eq. 12.

Again, by definition a Lorentz tensor is something whose free indices transform under a Lorentz trans-
formation defined in Eq. 10; for instance, given some tensor Rαβγ , in a primed frame one can write the
components of R as Rµνρ0 = Λαµ Λβν Λγρ Rαβγ .

In words what Eq. 10 says is that Lorentz transformations are transformations of spacetime that
preserve the Minkowski metric. These form a group, the group of all distance-preserving maps—
isometries—of Minkowski spacetime that leave the origin fixed2 .

A group is a set of elements with an operation that combines any two elements to form a third, which
satisfies certain properties (closure, associativity, identity, and inverse). Here, the elements are the Λ’s
and the group operation is matrix multiplication. Then, we have that

• The product of any 2 Lorentz transformations is another Lorentz transformation (closure).

• Associativity of Lorentz transformations (which follows from the properties of matrix multiplica-
tion).

• The identity is Λµν = δ µν as we’ve already discussed.

• The inverse of Λµν is (Λ−1 )µν = Λνµ , as already discussed3 .

Eq. 10 ensures that the laws of physics take the same form in all inertial frames of reference.

2
The bigger group of isometries of Minkowski spacetime that don’t leave the origin fixed is called the Poincaré group.
For this group, a coordinate transformation would be given by x0µ = Λµν xν + aµ for aµ a vector of arbitrary constants.
3
Recall that a square matrix has an inverse if and only if its determinant is nonzero. One can use Eq. 10 to show that
the determinant of any Lorentz transformation is nonzero, and thus the inverse always exists.