1 The Lorentz transformation

Under a special Lorentz transformation

1 the space-time coordinates of an event,
as observed by two dierent inertial observers, are related by

t vx/c2
t0 = p
1 v 2 /c2
x vt
x0 = p
1 v 2 /c2
y0 = y
z0 = z (1)

Using the notation

2 3

x0 ct, , x1 = x, x2 = y, x3 = z, x = (ct, ~r)

(the reason why the indices are written as superscripts - rather than the usual
subscripts, is that we will soon meet entities which have to be written with
subscripts) we can rewrite these equations in the form

x00 = x0 x1

x01 = x0 + x1

x02 = x2
x03 = x3 (2)

v 1
where = , = p . One important feature of this transformation is
c 1 2
that the quantity, called the interval
4 :

2
2
2
2
c2 t2 + x2 + y 2 + z 2 x0 + x1 + x2 + x3 = x x (3)

is an invariant - it remains the same in all frames. The quantity on the

right are elements of a 44 matrix given by

1 0 0 0
0 1 0 0
( )
0
(4)
0 1 0
0 0 0 1
1 To refresh your memory, a SLT is the case where two inertial observers share a common
space-time origin, use mutually parallel coordinate axes, where the relative velocity is parallel
to the common X X 0 axes.
2 t is replaced byct so that all four components have the same dimension.
3 We will often abuse notation by writing x for both the -th component of x and the
4-vector as a whole. It should be clear from the context which one is being meant.
4 technically, this should be called the squared interval - but almost everybody calls it the
interval - and so shall we.

1
As we will see later, is actually a covariant second rank tensor - but right
now it is perhaps more protable to think of this as a matrix that is nearly -
but not quite - the 44 identity matrix. Of course, we have also made use of
the summation convention here.
Of course, the transformation is more complicated in the case of a general
Lorentz transformation. But in general, the Lorentz transformation is a linear
transformation of coordinates

x0 = x (5)

which leaves the interval invariant

x0 x0 = x x (6)

Note that the coecients of the Lorentz transformation can be written as partial
derivatives
x0
= (7)
x
For the special Lorentz transformation, the transformation matrix is given by

0 0

0 0
( ) =
0
(8)
0 1 0
0 0 0 1

Equations (5) and (6) imply

5

= (9)

In terms of the matrix L whose -th element is this is

LT L = (10)

where ( ). Taking the determinant of both sides of (10) yields

det L = 1

while using ==0 in (9) gives

 0 
 0  1

We dene a transformation to be a proper Lorentz transformation

6 if it obeys
0
det L = +1 and 0 1. You can easily check that the special Lorentz trans-
formation (8) satises (9) and also that it is a proper Lorentz transformation.

5 Actually, substituting (5) in (6) leads to x x = x x . Although, this

is true for arbitrary x , this does not directly lead to (9) - it only says that its two sides dier
by an entity that must be antisymmetric in and . Since it can be easily checked that both
sides are symmetric, (9) follows.
6 It is an important mathematical property that the set of all proper Lorentz transforma-
tions form a subgroup of the group formed by all Lorentz transformations - although we will
not be using group theoretical techniques here.

2
Another very important subset of Lorentz transformations
p are rotations. Note
that under a rotation t stays unchanged and the length x2 + y 2 + z 2 stays un-
changed - so that it is a linear transformation that leaves the interval invariant
- hence, a genuine Lorentz transformation. Of course, for a rotation the Lorentz
transformation matrix takes the form

1 0 0 0

0
(R ) =
0

R
0

where R is a 33 orthogonal matrix denoting the particular rotation.

2 Scalars, vectors, tensors ...

2.1 Scalars

Things like the interval, which are described by a single number and are invariant
under a Lorentz transformation,

s s0 = s (11)

are called scalars. Examples are ds2 , the proper time d = dt, the mass m of
a particle, its charge q etc.

2.2 4-Vectors

 
An object described by 4 components ~
A0 , A1 , A2 , A3 A0 , A is called a 4-

vector if it transforms in the same way as the coordinates do under a Lorentz

transformation:
A A0 = A (12)

Some rather obvious examples are

dx
 
0

E
x = x = ct, ~x , = (c, ~u) , p m = (mc, m~u) = , p~
d c

Note that under a rotation, the 3-vector part gets acted upon by R, while A0
stays unchanged. Thus, the 3-vector part is what we used to call a vector in
pre-relativity days!

2.3 The inner product

Given two 4-vectors A and B the quantity ~B

A B = A0 B 0 + A ~ is a
scalar :

A B A0 B 0 = A B = A B = A B
(13)

3
This is the 4-dimensional inner product - a direct generalization of the dot
product of 3-vectors .
7
Using the invariant inner product, we can classify a 4-vector into three
classes:

2
timelike if A A = A0~A
A ~<0

2
lightlike or null if A A = A0 A~A~=0

2
spacelike if A A = A0 A ~A ~>0

2.4 Second rank tensors

The quantity A B has 16 components which transform in the following way

A B A0 B 0 = A A = A B

Any object described by 16 components that transform in a similar manner is

called a second rank (contravariant)
8 tensor (also called a (2, 0) tensor) :

t t0 = t (14)

Note that if the original tensor is symmetric t = t , the transformed

tensor obeys

t0 = t = t = t = t0

so that the symmetry is preserved under a Lorentz transformation. Let me

emphasize once again that even though the individual components of a vector
may be very dierent in another frame, the symmetry of a tensor (if present) is a
deeper property - one that is the same in all frames. So, while the 00- component
of a tensor being equal to its 11-component is a statement that is true (perhaps)
for a particular inertial observer, the tensor is symmetric is a statement that
is true for all observers! Similarly you can check that antisymmetry is also a
property that is independent of the particular frame. Also note that any second
rank tensor can be written as a sum of a symmetric and an antisymmetric part
in an obvious manner

1 1
t = (t + t ) + (t t ) = S + A
2 2
1 1
where S = 2 (t + t ) and A = 2 (t t ) are the symmettric and
antisymmetric parts of t , respectively.

7 Of course, there are major dierences : for starters, in 3D ~a ~a is alwaysnon-negative, and

a zero value for this implies that ~a = 0. Because of the sign in the metric, this is no longer
the case for 4-vectors.
8 The names contravariant or covariant (a term we will soon meet) are not in great favor
nowadays - especially given the somewhat confusing meanings of the terms (contra - of course
means opposite, while co means with), however most textbooks still use them.

4
2.5 Higher rank tensors

The notion of tensors can easily be generalised to higher ranks. For example a
fourth rank contravariant tensor is described by 44 = 256 components. which
transform according to

T T 0 = T (15)

2.6 Covectors

Let us now consider the 4 dimensional analogue of the gradient of a scalar

. How does this transform under a Lorentz transformation? It is
x
easy to see that
0 x
0 = =
x0 x0 x
x
here we meet a new coecient : - which, though very similar to (you
x0 0
x
should recall from (7) that this is ), is actually the -component of the
x
inverse of the matrix L. We denote this coecient by

x
(16)
x0
note that the indices on this symbol slant from bottom left to top right - as
opposed to top left to bottom right - which was the case for the original Lorentz
coecients. Note that , , being the element of L1 , is actually the

1
-element of LT . From this, it should be obvious that

= (17)

= (18)

1 T
For example, the top line is actually the -th element of LT L =
1
L L = I. If you get confused, you can always fall back on

x x0 x
= = =
x0 x x
With this notation, we have

0 = (19)

Thus, the four components of the 4-gradient transforms in a similar, but not
quite the same manner as the four coordinates - the coecient matrix is dierent!
By the way, for the special Lorentz transformation dened by (2) these new
coecients end up looking very similar

0 0
0 0
=


(20)
0 0 1 0
0 0 0 1

5
- compare (8). This fact that the two matrices in (8) and (20) have components
diering at most by a sign is not an accident, as we will soon see.
Now, the four components of the coordinates served as the prototype for
vectors - objects with four components that transform like the coordinates. As
we have just seen, the 4-gradient also has four components, but they transform
(slightly) dierently. In fact the 4-gradient is the prototype for a dierent kind
of object - the covector. We dene a covector as an object with four components
which transform as
B B0 = B (21)

The fact that the covector is a dierent kind of object

9 than the vector is
depicted by the positioning of the index - we write a covector by subscripts. This
nally clears up the mystery about why we chose to write vector components
with superscripts - we need to distinguish the two types!

2.7 Covariant and mixed tensors

The concept of higher rank tensors generalize to the covariant case. A sec-
ond rank covariant tensor (a (0, 2)tensor) is denoted by two lower indices, and
transforms like
t t0 = t (22)

Now, of course, we can have mixed tensors - objects with both upper and lower
indices, which each index providing its own kind of coecient in the transfor-
mation law. As an example, let us look at the transformation law for a rank
(2, 1) tensor :
T T 0 = T (23)

2.8 Contractions

Given a 4-vector A and a 4-covector B we can form the quantity

A B = B A A0 B0 + A1 B1 + A2 B2 + A3 B3

This, of course, is a single number with no free index. How does this behave
under a Lorentz transformation?

A B A0 B0 = A B = A B = A B = A B

Thus we see that this single component object is, indeed, a scalar! This kind
of an operation, where an upstairs and a downstairs index in an expression
are set equal (and hence summed over) is called a contraction. You could have

9 Why didn't we meet this stu before? Well, the major dierence between vectors and
covectors is that while for the former, the transformation matrix L is the same as the co-

1
ordinates, for the latter it is LT . Earlier, in the 3D world, we were concerned with

1
how things transform under a rotation of axes - and for rotations RT =R - there is no
dierence between the two!

6
more complicated examples like F or T S etc. For example, the latter
transforms as

T S T 0 S 0 =  T S 

The two cocients colored red yield , while the blue ones give  so that

T 0 S 0 =  T S  = T S

and this transformation shows that this is a (2, 0) tensor. The nal rule is simple
- in an expression with contractions the free indices (the ones not being summed
over) and their positions determine the kind and transformation of the resulting
object! Of course, for this it is essential that the indices being contracted are of
opposite types - this means that they contribute inverse factors to the overall
transformation and hence cancel out. Note that contracting two indices of the
same kind will not lead to cancelling factors - this is why in the summation
convention we insist that the dummy indices being summed over must be one
superx and one sux.
Let us now explore a slightly dierent question - suppose we know that
an expression likeT S is a 4-vector (as the free index suggests). If T
is an arbitrary to be(2, 0)tensor , then what can we say about the Lorentz
transformation properties of the four component object S ? The index position
suggests 4-covector - to see that this is, indeed, true, notice

T 0 S0 = T S = T S0 = T S

since this is valid for all T , we must have S0 = S or S0 =

S . This is the inverse transformation - the old components S as a linear
combination of the new S0 . To get direct transformation, lets multiply (and
sum over the dummy index) both sides by :

S0 = S

Now, the on the right is just so that we nd

S0 = S

Thus - what the position of the index suggested is borne out - S is a 4-covector.

2.9 Raising or lowering indices

Let me point out that though directly contracting two vectors or two covectors
do not yield anything meaningful, we already know a way of making a scalar
out of two vectors. The quantity A B is a scalar! But this must imply that
the quantity A is a covector since it gives a scalar on contraction with a
vector. We can directly check this, since

A A0 = A

7
It is easy to see that equation (9) can be rewritten in the form =
so that
A0 = A = ( A )
which is exactly the covector transformation law! Hence we see that to every
vector (raised index) we can naturally associate a covector (lowered index) object
- by simply contracting it with .

A A (24)

Of course, to go the other way - from lower to higher index - we will need the
inverse of the matrix ( ) - whose elements we denote by .

A = A (25)

In a similar fashion we can use and to raise and lower, respectively,

the indices of a higher rank tensor. For example

T = T

Because of the simple form that ( ) has - it is actually its own inverse - so
that the are numerically the same as their lower-index counterparts. Again,
because the metric ( ) is so simple - raising and lowering of indices simply
boils down to this simple rule - whenever a temporal index (0) is moved up or
down, the sign changes - there is no change of sign accompanying the shifting
of a spatial index. Thus for a 4-vector

   
~ ,
A A0 , A we have ~
A = A0 , A (26)

Note that the raising/lowering of index also works for and - which
are not tensors! This can be easily seen in the case of the special Lorentz
transformation by directly comparing the explicit forms in (8) and (20) - only
the mixed components (one spatial and one temporal) changes sign in this case
- the only non-zero ones that change sign are the 0,1 and 1,0 elements. That
this also works for general Lorentz transformations follows from the equation
= we had derived a while ago - this can be directly rewritten
in the form
=
This should explain the convention in positioning of the indices of the vector
and covector transformation components!

2.10 Invariant tensors

Just what kind of an object is the metric ? The positioning of the indices
seem to suggest that it is a covariant second rank tensor. On the other hand, we
use exactly the same numbers (1, 1, 1, 1) in all inertial frames - isn't a tensor

8
supposed to transform when you change frames. Now, the components in the
new frame, taking to be a second rank tensor, should change into

0
=

1
In matrix form (remembering that is the -element of the matrix LT )
T 1 1


, this is the -element of L L , which according to (10) is the same as
0
! Hence = . In we have the example of a covariant second rank
tensor which is numerically invariant under a Lorentz transformation. Other
examples of numerically invariant tensors are and , which, as their index
positions suggest, are (2, 0) and (1, 1) tensors, respectively.
Another numerically invariant tensor is the 4-dimensional version of the
Levi-Civita symbol  . Remember that this is completely antisymmetric in
the indices (and hence vanishes whenever two indices match) and is dened by
0123 = +1. Now, if we treat this as a (4, 0) tensor as the indices suggest, its
value in a new inertial frame should be

0 =  = det ( ) 

where we have used the basic denition of determinants. Thus, for a proper
Lorentz transformation (remember that this means det L = +1)  is nu-
merically invariant
10 .

2.11 Manifest covariance

Why should we care about all this? For one thing, as we saw in class we can
rewrite the Maxwell equations in the form

F = 0 j , G = 0 (27)

Apart from the fact that this is a very compact form (just the fact these are two
equations in place of the usual four should be enough to indicate that)
11 , the
real advantage of this form is that it makes it obvious at one glance that these
equations are equally valid for all inertial observers - which is the chief demand
of relativity. Note that both sides of (27) are 4-vectors (as indicated by the po-
sitioning of the free indices - so that they both transform in the same way under
a Lorentz transformation. The two sides vary under a Lorentz transformation
- but they co-vary - which means they vary in the same manner. This means
that if they are equal in one inertial frame, they will have to be equal in all
frames. Thus, equations that can be written in a tensorial form, with matching
free indices on both sides, are manifestly Lorentz covariant!

10 Strictly speaking,  is actually a slightly dierent kind of object -it is a (4, 0) tensor
density of weight -1. This actually ensures that it is numerically invariant - even for more
general transformations.
11 Of course, two of Maxwell's equations are scalar equations - the other two are vector
equations - this means that there are eight equations in all (each vector equation counts as
three - one for each component). The compact form looks like two equations - but each is a
4-vector equation. So you have eight equations is this version as well!