Chapter 1

Lorentz Group and Lorentz

Invariance
In studying Lorentz-invariant wave equations, it is essential that we put our under-
standing of the Lorentz group on rm ground. We rst dene the Lorentz transfor-
mation as any transformation that keeps the 4-vector inner product invariant, and
proceed to classify such transformations according to the determinant of the transfor-
mation matrix and the sign of the time component. We then introduce the generators
of the Lorentz group by which any Lorentz transformation continuously connected to
the identity can be written in an exponential form. The generators of the Lorentz
group will later play a critical role in nding the transformation property of the Dirac
spinors.
1.1 Lorentz Boost
Throughout this book, we will use a unit system in which the speed of light c is unity.
This may be accomplished for example by taking the unit of time to be one second
and that of length to be 2.99792458 10
10
cm (this number is exact
1
), or taking the
unit of length to be 1 cm and that of time to be (2.99792458 10
10
)
1
second. How
it is accomplished is irrelevant at this point.
Suppose an inertial frame K (space-time coordinates labeled by t, x, y, z) is mov-
ing with velocity in another inertial frame K

(space-time coordinates labeled by

t

, x

, y

, z

) as shown in Figure 1.1. The 3-component velocity of the origin of K

1
One cm is dened (1983) such that the speed of light in vacuum is 2.99792458 10
10
cm per
second, where one second is dened (1967) to be 9192631770 times the frequency of the hyper-ne
splitting of the Cs
133
ground state.
1
2 CHAPTER 1. LORENTZ GROUP AND LORENTZ INVARIANCE
K' K
y'
x'
y
x

K' K
(E',P') (E,P)
K' K frame frame
Figure 1.1: The origin of frame K is moving with velocity

= (, 0, 0) in frame K

,
and the origin of frame K

is moving with velocity

in frame K. The axes x and

x

are parallel in both frames, and similarly for y and z axes. A particle has energy
momentum (E,

P) in frame K and (E

,

P

) in frame K

.
measured in the frame K

K
, is taken to be in the +x

direction; namely,

K
(velocity of K in K

) = (, 0, 0)
def

. (1.1)
Assume that, in the frame K

, the axes x, y, z are parallel to the axes x

, y

, z

. Then,
the velocity of the origin of K

in K,

K
, is

K
=

K
= (, 0, 0) (velocity of K

in K). (1.2)
Note that

K
(

K
) is measured with respect to the axes of K

(K).
If a particle (or any system) has energy and momentum (E,

P) in the frame K,
then the energy and momentum (E

,

P

) of the same particle viewed in the frame K

are given by
E

=
E + P
x

1
2
, P

y
= P
y
,
P

x
=
E + P
x

1
2
, P

z
= P
z
, .
(1.3)
This can be written in a matrix form as
_
E

x
_
=
_


__
E
P
x
_
,
_
P

y
P

z
_
=
_
P
y
P
z
_
(1.4)
with

1

1
2
, =

1
2
. (1.5)
Note that and are related by

2
= 1 . (1.6)
1.1. LORENTZ BOOST 3
K' K
y'
x'
y
x
K' K
(E',P') (E,P)
K' frame K frame

Figure 1.2: Starting from the conguration of Figure 1.1, the same rotation is applied
to the axes in each frame. The resulting transformation represents a general Lorentz
boost.
Now start from Figure 1.1 and apply the same rotation to the axes of K and K

within each frame (Figure 1.2). Suppose the rotation is represented by a 33 matrix
R. Then, the velocity of K

in K,

K
, and and the velocity of K in K

K
, are
rotated by the same matrix R,

K
R

K
,

K
R

K
, (1.7)
and thus we still have

K
=

def

, (1.8)
where we have also redened the vector

which is well-dened in both K and K

frames in terms of

K
and

K
, respectively. The transformation in this case can
be obtained by noting that, in (1.4), the component of momentum transverse to

does not change and that P

x
, P

x
are the components of

P,

P

along

in each frame.
Namely, the tranformation can be written as
_
E

_
=
_


__
E
P

_
,

P

=

P

, (1.9)
where

and

denote components parallel and perpecdicular to

, repectively. Note
that

P

and

P

are 3-component quantities and the relation

P

=

P

holds compo-
nent by component because we have applied the same roation R in each frame.
The axes of K viewed in the frame K

are no longer perpendicular to each other

since they are contracted in the direction of

K
. Thus, the axes of K in general
are not parallel to the corresponding axes of K

at any time. However, since the

same rotation is applied in each frame, and since components transverse to

are the
same in both frames, the corresponding axes of K and K

are exactly parallel when

projected onto a plane perpendicular to

in either frames. The transformation (1.9)
4 CHAPTER 1. LORENTZ GROUP AND LORENTZ INVARIANCE
is thus correct for the specic relative oprientation of two frames as dened here,
and such transformation is called a Lorentz boost, which is a special case of Lorentz
transformation dened later in this chapter for which the relative orientation of the
two frames is arbitrary.
1.2 4-vectors and the metric tensor g

The quantity E
2

P
2
is invariant under the Lorentz boost (1.9); namely, it has the
same numerical value in K and K

:
E
2

P
2
= E
2
(P
2

+

P
2

)
= (E + P

)
2

_
(E + P

)
2
+

P
2

_
= (
2

2
)
. .
1
E
2
+ (
2

2
)
. .
1
P
2

P
2

= E
2

P
2
,
(1.10)
which is the invariant mass squared m
2
of the system. This invariance applies to any
number of particles or any object as long as E and

P refer to the same object.
The relative minus sign between E
2
and

P
2
above can be treated elegantly as
follows. Dene a 4-vector P

( = 0, 1, 2, 3) by
P

= (P
0
, P
1
, P
3
, P
4
)
def
(E, P
x
, P
y
, P
z
) = (E,

P) (1.11)
called an energy-momentum 4-vector where the index is called the Lorentz index
(or the space-time index). The = 0 component of a 4-vector is often called time
component, and the = 1, 2, 3 components space components.
Dene the inner product (or dot product) A B of two 4-vectors A

= (A
0
,

A)
and B

= (B
0
,

B) by
A B
def
A
0
B
0

A

B = A
0
B
0
A
1
B
1
A
2
B
2
A
3
B
3
. (1.12)
Then, P
2
P P is nothing but m
2
:
P
2
= P
0
2

P
2
= E
2

P
2
= m
2
(1.13)
which is invariant under Lorentz boost. This inner product P P is similar to the
norm x
2
of an ordinary 3-dimensional vector x, which is invariant under rotation,
except for the minus signs for the space components in the denition of the inner
product. In order to handle these minus signs conveniently, we dene subscripted
components of a 4-vector by
A
0
= A
0
, A
i
= A
i
(i = 1, 2, 3) . (1.14)
1.2. 4-VECTORS AND THE METRIC TENSOR G

5
Then the inner product (1.12) can be written as
A B = A
0
B
0
+ A
1
B
1
+ A
2
B
2
+ A
3
B
3
def
A

= A

, (1.15)
where we have used the convention that when a pair of the same index appears in the
same term, then summation over all possible values of the index ( = 0, 1, 2, 3 in this
case) is implied. In general, we will use Roman letters for space indices (take values
1,2,3) and greek letters for space-time (Lorentz) indices (take values 0,1,2,3). Thus,
x
i
y
i
=
3

i=1
x
i
y
i
(= x y), (A

+ B

)C

=
3

=0
(A

+ B

)C

, (1.16)
but no sum over or in
A

+ C

(, not in the same term). (1.17)

When a pair of Lorentz indices is summed over, usually one index is subscript and
the other is superscript. Such indices are said to be contracted. Also, it is important
that there is only one pair of a given index per term. We do not consider implied
summations such as A

to be well-dened. [(A

+ B

)C

is well-dened since
it is equal to A

+ B

.]
Now, dene the metric tensor g

by
g
00
= g
11
= g
22
= g
33
= 1, g

= 0 ( ,= ) (1.18)
which is symmetric:
g

= g

. (1.19)
The corresponding matrix G is dened as
g

def

0 1 2 3 =
0
1
2
3
_
_
_
_
_
_
1 0 0 0
0 1 0 0
0 0 1 0
0 0 0 1
_
_
_
_
_
_
||

def
G (1.20)
When we form a matrix out of a quantity with two indices, by denition we take the
rst index to increase downward, and the second to increase to the right.
As dened in (1.14) for a 4-vector, switching an index between superscript and
subscript results in a sign change when the index is 1,2, or 3, while the sign is
unchanged when the index is zero. We adopt the same rule for the indices of g

. In
6 CHAPTER 1. LORENTZ GROUP AND LORENTZ INVARIANCE
fact, from now on, we enforce the same rule for all space-time indices (unless otherwise
stated, such as for the Kronecker delta below). Then we have
g

= g

, g

= g

(1.21)
where

is the Kroneckers delta (

= 1 if = , 0 otherwise) which we dene to

have only subscripts. Then, g

can be used together with contraction to lower or

raise indices:
A

= g

, A

= g

(1.22)
which are equivalent to the rule (1.14).
The inner product of 4-vectors A and B (1.12) can also be written in matrix form
as
A B = A

=
(A
0
A
1
A
2
A
3
)
_
_
_
_
_
_
1 0 0 0
0 1 0 0
0 0 1 0
0 0 0 1
_
_
_
_
_
_
_
_
_
_
_
_
B
0
B
1
B
2
B
3
_
_
_
_
_
_
= A
T
GB . (1.23)
When we use 4-vectors in matrix form, they are understood to be column vectors
with superscripts, while their transpose are row vectors:
A
def

_
_
_
_
_
_
A
0
A
1
A
2
A
3
_
_
_
_
_
_
=
_
_
_
_
_
_
A
0
A
x
A
y
A
z
_
_
_
_
_
_
, A
T
= (A
0
, A
1
, A
2
, A
3
) (in matrix form). (1.24)
1.3 Lorentz group
The Lorentz boost (1.4) can be written in matrix form as
P

= P (1.25)
with
P

=
_
_
_
_
_
_
E

x
P

y
P

z
_
_
_
_
_
_
, =
_
_
_
_
_
_
0 0
0 0
0 0 1 0
0 0 0 1
_
_
_
_
_
_
, P =
_
_
_
_
_
_
E
P
x
P
y
P
z
_
_
_
_
_
_
. (1.26)
In terms of components, this can be written as
P

, (1.27)
1.3. LORENTZ GROUP 7
where we have dened the components of the matrix by taking the rst index to be
superscript and the second to be subscript (still the rst index increases downward
and the second index increases to the right):

def

=
0 1 2 3 =
0
1
2
3
_
_
_
_
_
_
0 0
0 0
0 0 1 0
0 0 0 1
_
_
_
_
_
_
||

(1.28)
For example,
0
1
= and thus
01
= , etc. The superscript and subscript in
(1.27) were chosen such that the index is contracted and that the index on both
sides of the equality has consistent position, namely, both are superscript.
We have seen that P
2
= E
2

P
2
is invariant under the Lorentz boost given by
(1.4) or (1.9). We will now nd the necessary and sucient condition for a 4 4
matrix to leave the inner product of two 4-vectors invariant. Suppose A

and B

transform by the same matrix :

A

, B

. (1.29)
Then the inner products A

and A B can be written using (1.22) as

A

= A

..
g

..

..

= (g

)A

A B = A

..
g

= g

.
(1.30)
In order for A

= A B to hold for any A and B, the coecients of A

should
be the same term by term (To see this, set A

= 1 for = and 0 for all else, and

B

= 1 for = and 0 for all else.):

g

= g

. (1.31)
On the other hand, if satises this condition, the same derivation above can be
traced backward to show that the inner product A B dened by (1.12) is invariant.
Thus, (1.31) is the necessary and sucient condition.
8 CHAPTER 1. LORENTZ GROUP AND LORENTZ INVARIANCE
What does the condition (1.31) tell us about the nature of the matrix ? Using
(1.22), we have g

, then the condition becomes

= g

raise on both sides

= g

(=

) . (1.32)
Comparing this with the denition of the inverse transformation
1
:
(
1
)

, (1.33)
we see that the inverse matrix of is obtained by
(
1
)

, (1.34)
which means that one simply has to change the sign of the components for which
only one of the indices is zero (namely,
0
i
and
i
0
) and then transpose it:
=
_
_
_
_
_
_

0
0

0
1

0
2

0
3

1
0

1
1

1
2

1
3

2
0

2
1

2
2

2
3

3
0

3
1

3
2

3
3
_
_
_
_
_
_
,
1
=
_
_
_
_
_
_

0
0

1
0

2
0

3
0

0
1

1
1

2
1

3
1

0
2

1
2

2
2

3
2

0
3

1
3

2
3

3
3
_
_
_
_
_
_
. (1.35)
Thus, the set of matrices that keep the inner product of 4-vectors invariant is made of
matrices that become their own inverse when the signs of components with one time
index are ipped and then transposed. As we will see below, such set of matrices
forms a group, called the Lorentz group, and any such transformation [namely, one
that keeps the 4-vector inner product invariant, or equivalently that satises the
condition (1.31)] is dened as a Lorentz transformation.
To show that such set of matrices forms a group, it is convenient to write the
condition (1.31) in matrix form. Noting that when written in terms of components,
we can change the ordering in any way we want, the condition can be written as

= g

, or
T
G = G. (1.36)
A set forms a group when for any two elements of the set x
1
and x
2
, a product
x
1
x
2
can be dened such that
1. (Closure) The product x
1
x
2
also belongs to the set.
2. (Associativity) For any elements x
1
, x
2
and x
3
, (x
1
x
2
)x
3
= x
1
(x
2
x
3
).
3. (Identity) There exists an element I in the set that satises Ix = xI = x for
any element x.
1.3. LORENTZ GROUP 9
4. (Inverse) For any element x, there exists an element x
1
in the set that satises
x
1
x = xx
1
= I.
In our case at hand, the set is all 4 4 matrices that satisfy
T
G = G, and
we take the ordinary matrix multiplication as the product which denes the group.
The proof is straightforward:
1. Suppose
1
and
2
belong to the set (i.e.
T
1
G
1
= G and
T
2
G
2
= G). Then,
(
1

2
)
T
G(
1

2
) =
T
2

T
1
G
1
. .
G

2
=
T
2
G
2
= G. (1.37)
Thus, the product
1

2
also belongs to the set.
2. The matrix multiplication is of course associative: (
1

2
)
3
=
1
(
2

3
).
3. The identity matrix I (I

) belongs to the set (I

T
GI = G), and satises
I = I = for any element.
4. We have already seen that if a 4 4 matrix satises
T
G = G, then its
inverse exists as given by (1.34). It is instructive, however, to prove it more
formally. Taking the determinant of
T
G = G,
det
T
. .
det
det G
. .
1
det = det G
. .
1
(det )
2
= 1 , (1.38)
where we have used the property of determinant
det(MN) = det M det N (1.39)
with M and N being square matrices of same rank. Thus, det ,= 0 and
therefore its inverse
1
exists. Also, it belongs to the set: multiplying
T
G =
G by (
1
)
T
from the left and by
1
from the right,
(
1
)
T

T
. .
(
1
. .
I
)
T
G
1
. .
I
= (
1
)
T
G
1
(
1
)
T
G
1
= G. (1.40)
This completes the proof that s that satisfy (1.36) form a group.
Since the inverse of a Lorentz transformation is also a Lorentz transformation, it
should satisfy the condition (1.31)
g

= g

(
1
)

(
1
)

= g

= g

, (1.41)
10 CHAPTER 1. LORENTZ GROUP AND LORENTZ INVARIANCE
where we have used the inversion rule (1.34). The formulas (1.31), (1.41), and their
variations are then summarized as follows: on the left hand side of the form g = g,
an index of g (call it ) is contracted with an index of a and the other index of
g (call it ) with an index of the other . As long as and are both rst or
both second indices on the s, and as long as the rest of the indices are the same
(including superscript/subscript) on both sides of the equality, any possible way of
indexing gives a correct formula. Similarly, on the left hand side of the form = g,
an index of a and an index of the other are contracted. As long as the contracted
indices are both rst or both second indices on s, and as long as the rest of the
indices are the same on both sides of the equality, any possible way of indexing gives
a correct formula.
A natural question at this point is whether the Lorentz group dened in this
way is any larger than the set of Lorentz boosts dened by (1.9). The answer is
yes. Clearly, any rotation in the 3-dimensional space keeps

A

B invariant while it
does not change the time components A
0
and B
0
. Thus, it keeps the 4-vector inner
product A B = A
0
B
0

A

B invariant, and as a result it belongs to the Lorentz
group by denition. On the other hand, the only way the boost (1.9) does not change
the time component is to set = 0 in which case the transformation is the identity
transformation. Thus, any nite rotation in the 3-dimensional space is not a boost.
Furthermore, the time reversal T and the space inversion P dened by
T
def
T

def

_
_
_
_
_
_
1
1
1
1
_
_
_
_
_
_
, P
def
P

def

_
_
_
_
_
_
1
1
1
1
_
_
_
_
_
_
(1.42)
satisfy
T
T
GT = G, P
T
GP = G, (1.43)
and thus belong to the Lorentz group. Even though the matrix P has the same
numerical form as G, it should be noted that P is a Lorentz transformation but G
is not (it is a metric). The dierence is also reected in the fact that the matrix
P is dened by the rst index being superscript and the second subscript (because
it is a Lorentz transformation), while the matrix G is dened by both indices being
subscript (or both superscript).
As we will see later, boosts and rotations can be formed by consecutive innitesi-
mal transformations starting from identity I (they are continuously connected to I),
while T and P cannot (they are disconnected from I, or said to be discrete trans-
formations). Any product of boosts, rotation, T, and P belongs to the Lorentz group,
and it turns out that they saturate the Lorentz group. Thus, we write symbolically
Lorentz group = boost + rotation + T + P . (1.44)
1.4. CLASSIFICATION OF LORENTZ TRANSFORMATIONS 11
Later, we will see that any Lorentz transformation continuously connected to I is a
boost, a rotation, or a combination thereof.
If the origins of the inertial frames K and K

touch at t = t

= 0 and x = x

= 0,
the coordinate x

= (t, x) of any event transforms in the same way as P

:
x

. (1.45)
This can be extended to include space-time translation between the two frames:
x

+ a

, (1.46)
where a

is a constant 4-vector. The transformation of energy-momentum is not

aected by the space-time translation, and is still given by P

= P. Such transfor-
mations that include space-time translation also form a group and called the inho-
mogeneous Lorentz group or the Poincare group. The group formed by the trans-
formations with a

= 0 is sometimes called the homogeneous Lorentz group. Unless

otherwise stated, we will deal with the homogeneous Lorentz group; namely without
space-time translation.
1.4 Classication of Lorentz transformations
Up to this point, we have not specied that Lorentz transformations are real (namely,
all the elements are real). In fact, Lorentz transformations as dened by (1.31) in
general can be complex and the complex Lorentz transformations plays an impor-
tant role in a formal proof of an important symmetry theorem called CPT theorem
which states that the laws of physics are invariant under the combination of particle-
antiparticle exchange (C), mirror inversion (P), and time reversal (T) under certain
natural assumptions. In this book, however, we will assume that Lorentz transfor-
mations are real.
As seen in (1.38), all Lorentz transformation satisfy (det )
2
= 1, or equivalently,
det = +1 or 1. We dene proper and improper Lorentz transformations as
_
det = +1 : proper
det = 1 : improper
. (1.47)
Since det(
1

2
) = det
1
det
2
, the product of two proper transformations or two
improper transformations is proper, while the product of a proper transformation and
a improper transformation is improper.
Next, look at the (, ) = (0, 0) component of the dening condition g

=
g

:
g

0
= g
00
= 1 , (
0
0
)
2

i=1
(
i
0
)
2
= 1 (1.48)
12 CHAPTER 1. LORENTZ GROUP AND LORENTZ INVARIANCE

0
0
1
orthochronous

0
0
1
non-orthochronous
det =+1
proper

(po)
TP
(po)
det =1
improper
P
(po)
T
(po)
Table 1.1: Classication of the Lorentz group.
(po)
is any proper and orthochronous
Lorentz transformation.
or
(
0
0
)
2
= 1 +
3

i=1
(
i
0
)
2
1 , (1.49)
which means
0
0
1 or
0
0
1, and this denes the orthochronous and non-
orthochronous Lorentz transformations:
_

0
0
1 : orthochronous

0
0
1 : non-orthochronous
. (1.50)
It is easy to show that the product of two orthochronous transformations or two non-
orthochronous transformations is orthochronous, and the product of an orthochronous
transformation and a non-orthochronous transformation is non-orthochronous.
From the denitions (1.42) and I

, we have
det I = det(TP) = +1, det T = det P = 1,
I
0
0
= P
0
0
= +1, T
0
0
= (TP)
0
0
= 1
(1.51)
Thus, the identity I is proper and orthochronous, P is improper and orthochronous,
T is improper and non-orthochronous, and TP is proper and non-orthochronous. Ac-
cordingly, we can multiply any proper and orthochronous transformations by each of
these to form four sets of transformations of given properness and orthochronousness
as shown in Table 1.1. Any Lorentz transformation is proper or improper (i.e. det =
1) and orthochronous or non-orthochronous (i.e. [
0
0
[
2
1). Since any transforma-
tion that is not proper and orthochronous can be made proper and orthochronous by
multiplying T, P or TP, the four forms of transformations in Table 1.1 saturate the
Lorentz group. For example, if is improper and orthochronous, then P
def

(po)
is proper and orthochronous, and can be written as = PP = P
(po)
.
It is straightforward to show that the set of proper transformations and the set
of orthochronous transformations separately form a group, and that proper and or-
thochronous transformations by themselves form a group. Also, the set of proper
1.4. CLASSIFICATION OF LORENTZ TRANSFORMATIONS 13
and orthochronous transformations and the set of improper and non-orthochronous
transformations together form a group.
Exercise 1.1 Classication of Lorentz transformations.
(a) Suppose = AB where , A, and B are Lorentz transformations. Prove that
is orthochronous if A and B are both orthochronous or both non-orthochronous,
and that is non-orthochronous if one of A and B is orthochronous and the other is
non-orthochronous.
[hint: Note that we can write
0
0
= A
0
0
B
0
0
+ a

b with a (A
0
1
, A
0
2
, A
0
3
) and

b (B
1
0
, B
2
0
, B
3
0
). Then use [a

b[ [a[[

b[. Also, one can derive a

2
= A
0
0
2
1 and

b
2
= B
0
0
2
1. ]
(b) Show that the following sets of Lorentz transformations each form a group:
1. proper transformations
2. orthochronous transformations
3. proper and orthochronous transformations
4. proper and orthochronous transformations plus improper and non-orthochronous
transformations
As mentioned earlier (and as will be shown later) boosts and rotations are con-
tinuously connected to the identity. Are they then proper and orthochronous? To
show that this is the case, it suces to prove that an innitesimal transformation
can change det and
0
0
only innitesimally, since then multiplying an innitesimal
transformation cannot jump across the gap between det = +1 and det = 1 or
the gap between
0
0
1 and
0
0
1.
An innitesimal transformation is a transformation that is very close to the iden-
tity I and any such transformation can be written as
= I + dH (1.52)
where d is a small number and H is a 44 matrix of order unity meaning the maximum
of the absolute values of its elements is about 1. To be specic, we could dene it such
that max
,
[H

[ = 1 and d 0, which uniquely denes the decomposition above.

We want to show that for any Lorentz transformation , multiplying I +dH changes
the determinant or the (0, 0) component only innitesimally; namely, the dierences
vanish as we take d to zero.
The determinant of a n n matrix A is dened by
det A
def

permutations
s
i
1
,i
2
,...in
A
i
1
1
A
i
2
2
. . . A
inn
(1.53)
14 CHAPTER 1. LORENTZ GROUP AND LORENTZ INVARIANCE
where the sum is taken over (i
1
, i
2
, . . . i
n
) which is any permutation of (1, 2, . . . n),
and s
i
1
,i
2
,...in
is 1(1) if (i
1
, i
2
, . . . i
n
) is an even(odd) permutation. When applied to
4 4 Lorentz transformations, this can be written as
det A
def

0
A

1
A

2
A

3
, (1.54)
where the implicit sum is over , , , = 0, 1, 2, 3 and

is the totally anti-

symmetric 4-th rank tensor dened by

def

_
=
_
+1
1
_
if () is an
_
even
odd
_
permutation of (0, 1, 2, 3)
= 0 if any of are equal
(1.55)
The standard superscript/subscript rule applies to the indices of

; namely,
0123
=

0123
= 1, etc. Then, it is easy to show that
det(I + dH) = 1 + d TrH + (higher orders in d) , (1.56)
where the trace of a matrix A is dened as the sum of the diagonal elements:
TrA
def

=0
A

. (1.57)
Exercise 1.2 Determinant and trace.
Determinant of a n n matrix is dened by
det A
def
s
i
1
i
2
,...in
A
i
1
1
A
i
2
2
. . . A
inn
where sum over i
1
, i
2
. . . i
n
is implied (each taking values 1 through n) and s(i
1
, i
2
. . . i
n
)
is the totally asymmetric n-th rank tensor:
s
i
1
i
2
...in

_
+1(1) if (i
1
, i
2
. . . i
n
) is an even (odd ) permutation of (1, 2, . . . n).
0 if any of i
1
, i
2
. . . i
n
are equal.
Show that to rst order of a small number d, the determinant of a matrix that is
innitesimally close to the identity matrix I is given by
det(I + dH) = 1 + dTrH + (higher orders in d) ,
where H is a certain matrix whose size is of order 1, and the trace (Tr) of a matrix
is dened by
TrH
n

i=1
H
ii
.
1.5. TENSORS 15
Since all diagonal elements of H are of order unity or smaller, (1.56) tells us that
det 1 as we take d 0. In fact, the innitesimal transformation is a Lorentz
transformation, so we know that det = 1. Thus, we see that the determinant of an
innitesimal transformation is strictly +1. It then follows from det() = det det
that multiplying an innitesimal transformation to any transformation does not
change the determinant of the transformation.
The (0, 0) component of is
()
0
0
= [(I + dH)]
0
0
= [ + dH]
0
0
=
0
0
+ d (H)
0
0
. (1.58)
Since (H)
0
0
is a nite number for a nite , the change in the (0, 0) component tends
to zero as we take d 0. Thus, no matter how many innitesimal transformations
are multiplied to , the (0, 0) component cannot jump across the gap between +1
and 1.
Thus, continuously connected Lorentz transformations have the same properness
and orthochronousness. Therefore, boosts and rotations, which are continuously con-
nected to the identity, are proper and orthochronous.
Do Lorentz boosts form a group?
A natural question is whether Lorentz boosts form a group by themselves. The answer
is no, and this is because two consecutive boosts in dierent directions turn out to
be a boost plus a rotation as we will see when we study the generators of the Lorentz
group. Thus, boosts and rotations have to be combined to form a group. On the
other hand, rotations form a group by themselves.
1.5 Tensors
Suppose A

and B

are 4-vectors. Each is a set of 4 numbers that transform under

a Lorentz transformation as
A

, B

. (1.59)
Then, the set of 16 numbers A

(, = 0, 1, 2, 3) transforms as
A

. (1.60)
Anything that has 2 Lorentz indices, which is a set of 16 numbers, and transforms as
( )

( )

(1.61)
is called a second rank tensor (or simply a tensor). It may be real, complex, or even
a set of operators. Similarly, a quantity that has 3 indices and transforms as
( )

( )

(1.62)
16 CHAPTER 1. LORENTZ GROUP AND LORENTZ INVARIANCE
is called a third-rank tensor, and so on. A 4-vector (or simply a vector) is a rst-
rank tensor. A Lorentz-invariant quantity, sometimes called a scalar, has no Lorentz
index, and thus it is a zero-th rank tensor:
( )

= ( ) (scalar) . (1.63)
Contracted indices do not count in deciding the rank of a tensor. For example,
A

: (scalar), A

: (vector), F

: (tensor), etc. (1.64)

The metrix g

has two Lorentz indices and thus can be considered a second-rank

tensor (thus, the metric tensor), then it should transform as
g

= g

(1.65)
where the second equality is due to (1.31). Namely, the metric tensor is invariant
under Lorentz transformations.
In order for some equation to be Lorentz-invariant, the Lorentz indices have to be
the same on both sides of the equality, including the superscript/subscript distinction.
By Lorentz-invariant, we mean that if an equation holds in one frame, then it holds
in any other frame after all the quantities that appear in the equation are evaluated
in the new frame. In the literature, such equations are sometimes called Lorentz
covariant: both sides of the equality change values but the form stays the same. For
example, if an equation A

= B

(which is actually a set of 16 equations) holds in

a frame, then it also holds in any other frame:
A

= B

. (1.66)
Thus, equations such as
m
2
= P

,
P

= A

+ B

,
F

= A

(1.67)
are all Lorentz-invariant, assuming of course that the quantities transform in the
well-dened ways as described above.
1.6 Fields (classical)
A eld is a quantity that is a function of space-time point x

= (t, x) (or event).

A scalar quantity that is a function of space time is called a scalar eld, a vector
quantity that is a function of space time is called a vector eld, etc. The rank of
a eld and the Lorentz transformation properties (scalar, vector, tensor, etc.) are
1.6. FIELDS (CLASSICAL) 17
dened in the same way as before, provided that the quantities are evaluated at the
same event point before and after a Lorentz transformation; namely,
Scalar eld :

(x

) = (x)
Vector eld : A

(x

) =

Tensor eld : T

(x

) =

(x)
(1.68)
where x

and x are related by

x

. (1.69)
For example, a vector eld associates a set of 4 numbers A

(x) to an event point

x, say when an ant sneezes. In another frame, there are a set of 4 numbers A

(x

)
associated with the same event x

, namely, when the ant sneezes in that frame, and

they are related to the 4 numbers A

(x) in the original frame by the matrix . The

functional shape of a primed eld is in general dierent from that of the corresponding
unprimed eld. Namely, if one plots (x) as a function of x and

(x

) as a function
of x

, they will look dierent.

When a quantity is a function of x, we naturally encounter space-time derivatives
of such quantity. Then a question arises as to how they transform under a Lorentz
transformation. Take a scalar eld f(x), and form a set of 4 numbers (elds) by
taking space-time derivatives:
f
x

(x) =
_
f
x
0
(x),
f
x
1
(x),
f
x
2
(x),
f
x
3
(x)
_
=
_
f
t
(x),
f
x
(x),
f
y
(x),
f
z
(x)
_
.
(1.70)
Then pick two space-time points x
1
and x
2
which are close in space and in time. The
argument below is based on the observation that the dierence between the values
of the scalar eld at the two event points is the same in any frame. Since f(x) is
a scalar eld, the values at a given event is the same before and after a Lorentz
transformation:
f

(x

1
) = f(x
1
), f

(x

2
) = f(x
2
) , (1.71)
or
f

(x

1
) f

(x

2
) = f(x
1
) f(x
2
) . (1.72)
Since x
1
and x
2
are close, this can be written as
dx

(x

1
) = dx

f
x

(x
1
) , (1.73)
where summation over is implied, and
dx

def
x

1
x

2
, dx

def
x

1
x

2
. (1.74)
18 CHAPTER 1. LORENTZ GROUP AND LORENTZ INVARIANCE
which tells us that the quantity dx

(f/x

) is Lorentz-invariant. Since dx

= x

1
x

2
is a superscripted 4-vector, it follows that f/x

should transform as a subscripted

4-vector (which transforms as A

):
f

(x

) =

f
x

(x) . (1.75)
In fact, together with dx

dx

, we have
dx

(x

) =
_

dx

_
_

f
x

(x)
_
=

. .
g

by (1.32)
dx

f
x

(x)
= dx

f
x

(x) ,
(1.76)
showing that it is indeed Lorentz-invariant.
Thus, the index in the dierential operator /x

acts as a subscript even

though it is a superscript on x. To make this point clear, /x

is often written using

a subscript as

def

=
_

t
,

x
,

y
,

z
_
=
_

t
,

_
. (1.77)
Once

is dened, the standard subscript/superscript rule applies; namely,

=
/x

, etc. Symbolically, the operator

then transforms as a superscripted 4-vector:

, (1.78)
with

/x

.
Example: Consider the 4-component charge current density j

(x) = ((x),

j(x)).
We can see that this is indeed a Lorentz 4-vector as follows: Suppose the charge is
carried by some medium, such as gas of ions, then pick a space-time point x and let

0
be the charge density in the rest frame of the medium and

be the velocity of the
medium at that point. Then the charge density in the frame in question is larger
than
0
by the factor = 1/

1
2
due to Lorentz contraction
=
0
. (1.79)
Since

j =

, j

can be written as
j

= (,

j) = (
0
,
0

) =
0
(, ) =
0

, (1.80)
1.7. GENERATORS OF THE LORENTZ GROUP 19
where we have dened the 4-velocity

by

def
(,

) . (1.81)
On the other hand, the 4-momentum of a particle with mass m can be written as
P

= (m, m

) = m

(1.82)
which means that the 4-velocity

is a Lorentz 4-vector, and therefore so is j

. When
the charge is carried by more than one dierent media, unique rest frame of the media
where
0
is dened does not exist. The total j

, however, is the sum of j

for each
medium. Since j

for each medium is a 4-vector, the sum is also a 4-vector.

Then

= 0 is a Lorentz-invariant equation; namely, if it is true in one frame,

then it is true in any frame. Using (1.77), we can write

= 0 as

=
0
j
0
+
1
j
1
+
2
j
2
+
3
j
3
=

t
+

x
j
x
+

y
j
y
+

z
j
z
=

t
+

note the sign!

j = 0 ,
(1.83)
which is nothing but the charge conservation equation. Thus, we see that if charge is
conserved in one frame it is conserved in any frame.
1.7 Generators of the Lorentz group
In this section, we will focus on the proper and orthochronous Lorentz group. Other
elements of the Lorentz group can be obtained by multiplying T, P, and TP to the
elements of this group. The goal is to show that any element that is continuously
connected to the identity can be written as
2
= e

i
K
i
+
i
L
i
, (i = 1, 2, 3) (1.84)
where
i
and
i
are real numbers and K
i
and L
i
are 44 matrices. Such group whose
elements can be parametrized by a set of continuous real numbers (in our case they
2
In the literature, it is often dened as expi(
i
K
i
+
i
L
i
), which would make the operators
hermitian if the transformation were unitary (e.g. representations of the Lorentz group in the
Hilbert space). The Lorentz transformation matrices in space-time are in general not unitary, and
for now, we will dene without the i so that the expressions become simpler.
20 CHAPTER 1. LORENTZ GROUP AND LORENTZ INVARIANCE
are
i
and
i
) is called a Lie group. The operators K
i
and L
i
are called the generators
of the Lie group.
Any element of the proper and orthochronous Lorentz group is continuously con-
nected to the identity. Actually we have not proven this, but we will at least show
that all boosts, rotations and combinations thereof are continuously connected to the
identity (and vice versa).
1.7.1 Innitesimal transformations
Lets start by looking at a Lorentz transformation which is innitesimally close to the
identity:

= g

(1.85)
where

is a set of small (real) numbers. Inserting this to the dening condition

(1.31) or equivalently

= g

(1.32), we get
g

= (g

)(g

)
= g

+ g

= g

.
(1.86)
Keeping terms to the rst order in , we then obtain

. (1.87)
Namely,

is anti-symmetric (which is true when the indices are both subscript or

both superscript; in fact,

is not anti-symmetric under ), and thus it has 6

independent parameters:

_
_
_
_
_
_
_
0
01

02

03

01
0
12

13

02

12
0
23

03

13

23
0
_
_
_
_
_
_
_
(1.88)
This can be conveniently parametrized using 6 anti-symmetric matrices as

=
01
(M
01
)

+
02
(M
02
)

+
03
(M
03
)

+
23
(M
23
)

+
13
(M
13
)

+
12
(M
12
)

(1.89)
=

<

(M

1.7. GENERATORS OF THE LORENTZ GROUP 21

with
(M
01
)

=
_
_
_
_
_
_
0 1 0 0
1 0 0 0
0 0 0 0
0 0 0 0
_
_
_
_
_
_
, (M
23
)

=
_
_
_
_
_
_
0 0 0 0
0 0 0 0
0 0 0 1
0 0 1 0
_
_
_
_
_
_
,
(M
02
)

=
_
_
_
_
_
_
0 0 1 0
0 0 0 0
1 0 0 0
0 0 0 0
_
_
_
_
_
_
, (M
13
)

=
_
_
_
_
_
_
0 0 0 0
0 0 0 1
0 0 0 0
0 1 0 0
_
_
_
_
_
_
,
(M
03
)

=
_
_
_
_
_
_
0 0 0 1
0 0 0 0
0 0 0 0
1 0 0 0
_
_
_
_
_
_
, (M
12
)

=
_
_
_
_
_
_
0 0 0 0
0 0 1 0
0 1 0 0
0 0 0 0
_
_
_
_
_
_
(1.90)
Note that for a given pair of and , (M

is a 4 4 matrix, while

is a real
number. The elements (M

can be written in a concise form as follows: rst, we

note that in the upper right half of each matrix (i.e. for < ), the element with
(, ) = (, ) is 1 and all else are zero, which can be written as g

. For the
lower half, all we have to do is to ip and and add a minus sign. Combining the
two halves, we get
(M

= g

. (1.91)
This is dened only for < so far. For > , we will use this same expression as
the denition; then, (M

is anti-symmetric with respect to ( ):

(M

= (M

, (1.92)
which also means (M

= 0 if = . Together with

, (1.89) becomes

<

(M

>

(M

=
1
2

(M

, (1.93)
where in the last expression, sum over all values of and is implied. The innites-
imal transformation (1.85) can then be written as

= g

+
1
2

(M

, (1.94)
or in matrix form,
= I +
1
2

. (1.95)
22 CHAPTER 1. LORENTZ GROUP AND LORENTZ INVARIANCE
where the rst indices of M

, which is a 4 4 matrix for given and , is taken

to be superscript and the second subscript; namely, in the same way as Lorentz
transformation. Namely, when no explicit indexes for elements are given, the 4 4
matrix M

is dened as
M

def
(M

. (1.96)
It is convenient to divide the six matrices to two groups as
K
i
def
M
0i
, L
i
def
M
jk
(i, j, k : cyclic) . (1.97)
We always use subscripts for K
i
and L
i
since only possible values are i = 1, 2, 3, and
similarly to M

, elements of the matrices K

i
s and L
i
s are dened by taking the
rst Lorentz index to be superscript and the second subscript:
K
i
def
(K
i
)

, L
i
def
(L
i
)

. (1.98)
Later, we will see that Ks generate boosts and Ls generate rotations. Explicitly,
they can be obtained by raising the index in (1.90) (note also the the minus sign
in L
2
= M
13
):
K
1
=
_
_
_
_
_
_
0 1 0 0
1 0 0 0
0 0 0 0
0 0 0 0
_
_
_
_
_
_
, K
2
=
_
_
_
_
_
_
0 0 1 0
0 0 0 0
1 0 0 0
0 0 0 0
_
_
_
_
_
_
, K
3
=
_
_
_
_
_
_
0 0 0 1
0 0 0 0
0 0 0 0
1 0 0 0
_
_
_
_
_
_
(1.99)
L
1
=
_
_
_
_
_
_
0 0 0 0
0 0 0 0
0 0 0 1
0 0 1 0
_
_
_
_
_
_
, L
2
=
_
_
_
_
_
_
0 0 0 0
0 0 0 1
0 0 0 0
0 1 0 0
_
_
_
_
_
_
, L
3
=
_
_
_
_
_
_
0 0 0 0
0 0 1 0
0 1 0 0
0 0 0 0
_
_
_
_
_
_
(1.100)
By inspection, we see that the elements of Ks and Ls can be written as
(K
i
)
j
k
= 0 , (K
i
)
0

= (K
i
)

0
= g
i

,
(L
i
)
j
k
=
ijk
, (L
i
)
0

= (L
i
)

0
= 0 ,
(i, j, k = 1, 2, 3; = 0, 1, 2, 3) (1.101)
where
ijk
is a totally anti-symmetric quantity dened for i, j, k = 1, 2, 3:

ijk
def

_
=
_
+1
1
_
if (i, j, k) is an
_
even
odd
_
permutation of (1, 2, 3),
= 0 if any of i, j, k are equal.
(1.102)
1.7. GENERATORS OF THE LORENTZ GROUP 23
An explicit calculation shows that Ks and Ls satisfy the following commutation
relations:
[K
i
, K
j
] =
ijk
L
k
[L
i
, L
j
] =
ijk
L
k
(1.103)
[L
i
, K
j
] =
ijk
K
k
,
where sum over k = 1, 2, 3 is implied, and the commutator of two operators A, B is
dened as
[A, B]
def
AB BA. (1.104)
Note that the relation [K
i
, K
j
] =
ijk
L
k
can also be written as [K
i
, K
j
] = L
k
(i, j, k: cyclic), etc.
Exercise 1.3 Verify the commutation relations (1.103 ). You may numerically verify
them, or you may try proving generally by using the general formula for the elements
of the matrixes.
Exercise 1.4 Boost in a general direction.
Start from the formula for boost (1.9 ) where P

is the component of

P parallel to

,
and

P

is the component perpendicular to

; namely,
P

=

P n, and

P

=

P P

n
with n =

/ (and similarly for

P

). Note that

is well-dened in the primed frame
also by the particular relative orientation of the two frames chosen.
(a) Show that the corresponding Lorentz transformation matrix is given by
=
_
_
_
_
_

x

y

z

x
1 +
2
x

x

y

x

y

x

y
1 +
2
y

y

z

x

z

y

z
1 +
2
z
_
_
_
_
_
, with
1

2
.
(b) Show that when is small, the Lorentz transformation matrix for a boost is given
to the rst order in by
= 1 +
i
K
i
. (summed over i = 1, 2, 3)
(c) In the explicit expression of given above, one notes that the top row [
0

( =
0, 1, 2, 3)] and the left-most column [

0
( = 0, 1, 2, 3)] are nothing but the velocity
4-vector

= (,

). Lets see how it works for general Lorentz transformations
(proper and orthochronous). Suppose the relative orientation of the two frames K
and K

is not given by

K
=

K
, where

K
is the velocity of the origin of K
measured in K

, and

K
is the velocity of the origin of K

measured in K. Let be
the corresponding Lorentz transformation. Express
0

and

0
in terms of

K
and

K
. (hint: Place a mass m at the origin of K and view it from K

, and place a mass

at the origin of K

and view it from K.)

24 CHAPTER 1. LORENTZ GROUP AND LORENTZ INVARIANCE
1.7.2 Finite transformations
Now we will show that any nite (namely, not innitesimal) rotation can be written as
e

i
L
i
, and any nite boost can be written as e

i
K
i
, where
i
and
i
(i = 1, 2, 3) are some
nite real numbers. First, however, let us review some mathematical background:
Matrix exponentiations
The exponential of a mm matrix A is also a mm matrix dened by
e
A
def
lim
n
_
I +
A
n
_
n
, (1.105)
which can be expanded on the right hand side as
e
A
= lim
n
n

k=0
n(n 1) . . . (n k + 1)
k !
A
k
n
k
. (1.106)
Since the sum is a rapidly converging series, one can sum only the terms with k n
for which n(n 1) . . . (n k + 1) n
k
. It then leads to
e
A
=

k=0
A
k
k !
, (1.107)
which can also be regarded as a denition of e
A
.
Using the denition (1.105) or (1.107), we see that
_
e
A
_

= e
A

, (1.108)
where the hermitian conjugate of a matrix A is dened by (A

)
ij
A

ji
. The deter-
minant of e
A
can be written using (1.105) as
det e
A
= lim
n
_
det
_
I +
A
n
__
n
= lim
n
_
1 +
TrA
n
+ . . . +
c
k
n
k
+ . . .
_
n
,
(1.109)
where we have used (1.56). This does not depend on c
k
(k > 1) since the derivative
with respect to c
k
vanishes as can be readily veried. Thus, c
k
(k > 1) can be set to
zero and we have
det e
A
= e
TrA
. (1.110)
1.7. GENERATORS OF THE LORENTZ GROUP 25
The derivative of e
xA
(x is a number, while A is a constant matrix) with respect to x
can be obtained using (1.107),
d
dx
e
xA
=

k=1
(k x
k1
)A
k
k !
= A

k=1
x
k1
A
k1
(k 1) !
; (1.111)
thus,
d
dx
e
xA
= Ae
xA
. (1.112)
There is an important theorem that expresses a product of two exponentials in
terms of single exponential, called the Campbell-Baker-Hausdor (CBH) theorem
(presented here without proof):
e
A
e
B
= e
A+B+
1
2
[A,B]+
, (1.113)
where denotes the higher-order commutators of A and B such as [A, [A, B]],
[A, [[A, B], B]] etc. with known coecients. Note that the innermost commutator is
always [A, B] since otherwise it is zero ([A, A] = [B, B] = 0), and thus if [A, B] is a
commuting quantity (a c-number), then is zero. Applying (1.113) to B = A,
we get
e
A
e
A
= e
AA
= I , (1.114)
or
(e
A
)
1
= e
A
. (1.115)
Rotation
An innitesimal rotation around the z-axis by [Figure 1.3(a)] can be written as
_
x

_
=
_
x y
y + x
_
=
_
1
1
__
x
y
_
= (I + L
z
)
_
x
y
_
, (1.116)
with
L
z
=
_
x y
x 0 1
y 1 0
_
. (1.117)
Then, a rotation by a nite angle is constructed as n consecutive rotations by /n
each and taking the limit n . Using (1.116), it can be written as
_
x

_
= lim
n
_
I +

n
L
z
_
n _
x
y
_
= e
Lz
_
x
y
_
, (1.118)
26 CHAPTER 1. LORENTZ GROUP AND LORENTZ INVARIANCE
where we have used the denition (1.105).
From the explicit expression of L
z
(1.117), we have L
2
z
= I, L
3
z
= L
z
, L
4
z
= I,
etc. In general,
L
4n
z
= I, L
4n+1
z
= L
z
, L
4n+2
z
= I, L
4n+3
z
= L
z
, (1.119)
where n is an integer. Using the second denition of e
A
(1.107), the rotation matrix
e
Lz
can then be written in terms of the trigonometirc functions as
e
Lz
= I + L
z
+

2
2 !
L
2
z
..
I
+

3
3 !
L
3
z
..
L
z
+. . . (1.120)
=
_
1

2
2 !
+ . . .
_
. .
cos
I +
_


3
3 !
+ . . .
_
. .
sin
L
z
(1.121)
=
_
cos sin
sin cos
_
, (1.122)
which is probably a more familiar form of a rotation around the z-axis by an angle .
Similarly, rotations around x and y axes are generated by L
x
and L
y
as obtained
by cyclic permutations of (x, y, z) in the derivation above. Switching to numerical
indices [(L
x
, L
y
, L
z
) (L
1
, L
2
, L
3
)],
L
1
=
_
2 3
2 0 1
3 1 0
_
, L
2
=
_
3 1
3 0 1
1 1 0
_
, L
3
=
_
1 2
1 0 1
2 1 0
_
. (1.123)
Are these identical to the denition (1.100) which was given in 4 4 matrix form,
or equivalently (1.101)? Since L
i
is the change of coordinates by the rotation, the
x

n
x x

(x,y)
(x',y')
x
y

x
y
(a) (b)
Figure 1.3: Innitesimal rotation around the z-axis by an angle (a), and around a
general direction

by an angle /n (b).
1.7. GENERATORS OF THE LORENTZ GROUP 27
elements of a 4 4 matrix corresponding to unchanged coordinates should be zero.
We then see that the Ls given above are indeed identical to (1.100).
A general rotation is then given by
e

i
L
i
= e

L
, (1.124)
where

def
(
1
,
2
,
3
),

L
def
(L
1
, L
2
, L
3
) . (1.125)
As we will see below, this is a rotation around the direction

by an angle [

[. To
see this, rst we write e

i
L
i
using the denition (1.105):
e

i
L
i
= lim
n
_
I +

i
L
i
n
_
n
, (1.126)
which shows that it is a series of small rotations each given by I +
i
L
i
/n. The
action of such an innitesimal transformation [Figure 1.3(b)] on x is (writing the
space components only)
x
j
=
_
I +

i
L
i
n
_
j
k
x
k
= g
j
k
x
k
+
1
n

i
(L
i
)
j
k
. .

ijk
by (1.101)
x
k
= x
j

1
n

ijk

i
x
k
= x
j
+
1
n
(

x)
j
(1.127)
where we have used the denition of the three-dimensional cross product
(a

b)
i
=
ijk
a
j
b
k
. (1.128)
Thus, I +
i
L
i
/n is nothing but a small rotation around

by an angle /n (Figure 1.3).
Then n such rotations applied successively will result in a rotation by an angle
around the same axis

.
Boosts
A boost in x direction by a velocity is given by (1.26):
=
_
t x
t
x
_
,
_
=
1

1
2
, =
_
. (1.129)
28 CHAPTER 1. LORENTZ GROUP AND LORENTZ INVARIANCE
When is small (= ), 1 and to the rst order in ; then, the innitesimal
boost can be written as
=
_
1
1
_
= I + K
x
, (1.130)
with
K
x
=
_
t x
t 0 1
x 1 0
_
. (1.131)
Suppose we apply n such boosts consecutively, where we take n to innity while n
is xed to a certain value :
n = . (1.132)
Then the resulting transformation is
= lim
n
_
I +

n
K
x
_
n
= e
Kx
, (1.133)
where we have used the denition (1.105). Is the velocity of this boost? The answer
is no, even though it is a function of the velocity. Lets expand the exponential above
by the second denition (1.105) and use K
2
x
= I:
= e
Kx
(1.134)
= I + K
x
+

2
2 !
K
2
x
..
I
+

3
3 !
K
3
x
..
K
x
+. . .
=
_
1 +

2
2 !
+ . . .
_
. .
cosh
I +
_
+

2
2 !
+ . . .
_
. .
sinh
K
x
=
_
cosh sinh
sinh cosh
_
. (1.135)
Comparing with (1.130), we see that this is a boost of a velocity given by
= cosh , = sinh (1.136)
or
=

= tanh . (1.137)
Note that the relation
2

2
= 1 (1.6) is automatically satised since cosh
2

sinh
2
= 1.
Thus, n consecutive boosts by a velocity /n each did not result in a boost of a
velocity ; rather, it was a boost of a velocity = tanh . This breakdown of the
1.7. GENERATORS OF THE LORENTZ GROUP 29
simple addition rule of velocity is well known: the relativistic rule of velocity addition
states that two consecutive boosts, by
1
and by
2
, do not result in a boost of
1
+
2
,
but in a boost of a velocity
0
given by

0
=

1
+
2
1 +
1

2
. (1.138)
Due to the identity tanh(
1
+
2
) = (tanh
1
+tanh
1
)/(1 +tanh
1
tanh
2
), however,
it becomes additive when velocities are transformed by
i
= tanh
i
(i = 0, 1, 2);
namely,
0
=
1
+
2
holds.
The matrix K
x
( K
1
) given in (1.131) is identical to the 44 form given in (1.99)
when all other elements that correspond to unchanged coordinates are set to zero.
The boosts along y and z directions are obtained by simply replacing x with y or z
in the derivation above. Thus, we see that K
2
and K
3
given in (1.99) indeed generate
boosts in y and z directions, respectively.
A boost in a general direction would then be given by
= e

i
K
i
, (1.139)
where

(
1
,
2
,
3
) are the parameters of the boost. In order to see what kind of
transformation this represents, lets write it as a series of innitesimal transformations
using (1.105):
e

i
K
i
= lim
n
_
I +

i
n
K
i
_
n
. (1.140)
From the explicit forms of K
i
(1.99), we can write the innitesimal transformation as
I +

i
n
K
i
= I +
1
n
_
_
_
_
_
0
1

2

3

3
0
_
_
_
_
_
. (1.141)
On the other hand, a boost in a general direction by a small velocity

is given by
(1.9) with 1, and [

[:
_
E

_
=
_
1
1
__
E
P

_
,

P

=

P

, (1.142)
or
_

_
E

= E + P

= E +

P
P

= P

+ E

=

P

_


P

=

P + E

(1.143)
30 CHAPTER 1. LORENTZ GROUP AND LORENTZ INVARIANCE
where we have used

P
()
= P
()

+

P
()

/). This can be written in 4 4 matrix

form as
_
_
_
_
_
E

x
P

y
P

z
_
_
_
_
_
=
_

_
I +
_
_
_
_
_
0
1

2

3

3
0
_
_
_
_
_
_

_
_
_
_
_
_
E
P
x
P
y
P
z
_
_
_
_
_
. (1.144)
Comparing this with (1.141), we can identify that I +
i
K
i
/n as a boost in

direction
by a velocity /n ( [

[). Then n consecutive such boosts will result in a boost

in the same direction. Since the rule of addition of velocity (1.138) is valid in any
direction as long as the boosts are in the same direction, the n boosts by velocity
/n each will result in a single boost of velocity = tanh as before. Thus, e

i
K
i
represents a boost in

direction by a velocity = tanh .
Boost + rotation
First, we show that a rotation followed by a rotation is a rotation, but a boost followed
by a boost is not in general a boost. Consider a rotation e

i
L
i
followed by another
rotation e

i
L
i
where

and

are arbitrary vectors. Using the CBH theorem (1.113),
we can write the product of the two transformations as
e

i
L
i
e

j
L
j
= e

i
L
i
+
j
L
j
+
1
2
[
i
L
i
,
j
L
j
] +...
, (1.145)
where . . . represents terms with higher-order commutators such as [
i
L
i
, [
j
L
j
,
k
L
k
]]
etc. Now we can use the commutation relations (1.103) to remove all commutators
in the exponent on the right hand side. The result will be a linear combination of Ls
with well-dened coecients (call them
i
) since the coecients in . . . in the CBH
theorem are known. Here, there will be no Ks appearing in the linear combination
because of the commutation relation [L
i
, L
j
] =
ijk
L
k
. Thus, the product is written
as
e

i
L
i
e

j
L
j
= e

i
L
i
( : a function of

,

), (1.146)
which is just another rotation.
Next, consider a boost e

i
K
i
followed by another boost e

i
K
i
:
e

i
K
i
e

j
K
j
= e

i
K
i
+
j
K
j
+
1
2
[

i
K
i
,
j
K
j
] +...
. (1.147)
Again, the brackets can be removed by the commutation relations (1.103) reducing
the exponent to a linear combination of Ks and Ls. This time, there will be Ls
appearing through the relation [K
i
, K
j
] =
ijk
L
k
which are in general not cancelled
among dierent terms. Thus, a boost followed by a boost is not in general another
boost; rather, it is a combination of boost and rotation:
e

i
K
i
e

j
K
j
= e

i
K
i
+
i
L
i
( ,

: functions of

,

). (1.148)
1.7. GENERATORS OF THE LORENTZ GROUP 31
It is easy to show, however, that if two boosts are in the same direction, then the
product is also a boost.
Any combinations of boosts and rotations can then be written as
= e

i
K
i
+
i
L
i
= e
1
2
aM

, (1.149)
where we have dened the anti-symmetric tensor a

by
a
0i
def

i
, a
ij
def

k
(i, j, k : cyclic), a

= a

, (1.150)
and the factor 1/2 arises since terms with > as well as < are included in
the sum. The expression of an innitesimal transformation (1.95) is nothing but this
expression in the limit of small a

. Since we now know that any product of such

transformations can also be written as (1.149) by the CBH theorem, we see that the
set of Lorentz transformations connected to the identity is saturated by boosts and
rotations.
We have seen that the generators Ks and Ls and their commutation relations
(called the Lie algebra) play critical roles in understanding the Lorentz group. In
fact, generators and their commutation relations completely determine the structure
of the Lie group, as described briey below.
Structure constants
When the commutators of generators of a Lie group are expressed as linear com-
binations of the generators themselves, the coecients of the linear expressions are
called the structure constants of the Lie group. For example, the coeents
ijk
in
(1.103) are the structure constants of the Lorentz group. We will now show that the
structure constants completely dene the structure of a Lie group. To see this, we
have to dene what we mean by same structure. Two sets T( f) and (( g) are
said to have the same structure if there is a mapping between T and ( such that if
f
1
, f
2
T and g
1
, g
2
( are mapped to each other:
f
1
g
1
, f
2
g
2
(1.151)
then, the products f
1
f
2
and g
1
g
1
are also mapped to each other by the same mapping:
f
1
f
2
g
1
g
2
; (1.152)
namely, the mapping preserves the product rule.
Suppose the sets T and ( are Lie groups with the same number of generators F
i
and G
i
and that they have the same set of structure constants c
ijk
[F
i
, F
j
] = c
ijk
F
k
, [G
i
, G
j
] = c
ijk
G
k
. (1.153)
32 CHAPTER 1. LORENTZ GROUP AND LORENTZ INVARIANCE
Any element of T and ( can be expressed in exponential form using the corresponding
generators and a set of real parameters. We dene the mapping between T and ( by
the same set of the real parameters:
f = e

i
F
i
g = e

i
G
i
. (1.154)
If f
1
g
1
and f
2
g
2
, then they can be written as
f
1
= e

i
F
i
g
1
= e

i
G
i
(1.155)
f
2
= e

i
F
i
g
2
= e

i
G
i
, (1.156)
where
i
and
i
are certain sets of real parameters. Then the question is whether
the products f
1
f
2
and g
1
g
2
are mapped to each other by the same mapping. The
products f
1
f
2
and g
1
g
2
can be written using the CBH theorem as
f
1
f
2
= e

i
F
i
e

j
F
j
= e

i
F
i
+
j
F
j
+
1
2
[
i
F
i
,
j
F
j
] +...
= e

i
F
i
, (1.157)
g
1
g
2
= e

i
G
i
e

j
G
j
= e

i
G
i
+
j
G
j
+
1
2
[
i
G
i
,
j
G
j
] +...
= e

i
G
i
, (1.158)
The numbers
i
and
i
are obtained by removing the commutators using the commu-
tation relations (1.153), and thus completely determined by
i
,
i
and c
ijk
; namely,

i
=
i
, and thus f
1
f
2
and g
1
g
2
are mapped to each other by (1.154). Thus, if two Lie
groups have the same set of structure constants, then they have the same structure.