Appendix C

Lorentz group and the Dirac algebra

This appendix provides a review and summary of the Lorentz group, its
properties, and the properties of its innitesimal generators. It then reviews
representations of the Lorentz group and the Dirac algebra. This material is
intended to supplement Chapter 1, for those students who are not as familiar
with the Lorentz group and Dirac equation as they nd they need to be.
C.1 Lorentz group
According to special relativity, physical laws are unchanged by a linear
change of coordinates,
x

(C.1)
with and real, provided it leave unchanged the invariant separation
between points,
(x y)

(x y)

(x y)

(x y)

= [(x y)
0
]
2
+ [x y]
2
.
This condition does not constrain , since it cancels in the dierence, but it
imposes a constraint on ,
x

= x

, (C.2)
for all x

. A transformation of the form shown in Eq. (C.1) which satises

Eq. (C.2) is called a Poincare transformation. These transformations close
and form a group, called the Poincare group. The subgroup where is the
identity matrix and is arbitrary is a subgroup called the group of transla-
tions. We assume that this group and its implications, such as conservation
of energy and momentum, are familiar to the reader. Instead we concentrate
on the subgroup in which = 0, which is called the Lorentz group.
It is convenient to think of an element of the Lorentz group as a matrix
494
C.1 Lorentz group 495
which is operating on the coordinate x

. This is possible if we always write

with its rst index raised and second index lowered, so it carries x

to
x

, both with raised index. Writing in this way, repeated Lorentz trans-
formations are implemented via matrix multiplications of the respective
matrices:
x

and x

.
(C.3)
We see from Eq. (C.2) that the condition on

to be a Lorentz trans-
formation is,

(C.4)
for all x

. Since this must hold for all x

, we have

, (C.5)
or (writing

as when using matrix notation)

=
T
. (C.6)
The group of matrices satisfying Eq. (C.6) is called O(3, 1), and is a Lie
group. Therefore the same technology of Lie algebra generation may be
applied to it as to the groups of the previous appendix.
As we will discuss momentarily, not all elements of O(3, 1) can be built
innitesimally from the identity. Those elements which can, form a subgroup
written SO(3, 1), which we will now analyze. A Lorentz transformation

which is innitesimally close to the identity must be of form,

, (C.7)
with

a matrix of innitesimal coecients. The condition on

for

to be a valid Lorentz transformation is found by inserting Eq. (C.7)

into Eq. (C.6) and expanding to linear order in :

=
_

_
=

+ (

) +O(
2
) ,
0 =

. (C.8)
That is, the condition on

is that

be antisymmetric on its indices.

The space of real antisymmetric 4 4 matrices is 6 dimensional, so the
Lorentz group is 6 dimensional.
Now

is related to

as

. Since
00
= 1 and
ii
= 1
for i = 1, 2, 3, the sign of the space-time component of

must be the same

496 Lorentz group and the Dirac algebra
as the sign of the time-space component, while the space-space components
must be antisymmetric. Thus, the most general form of

is,

=
_
_
_
_
_
0 b
1
b
2
b
3
b
1
0 r
3
r
2
b
2
r
3
0 r
1
b
3
r
2
r
1
0
_
_
_
_
_
, (C.9)
symmetric in the space-time entries and anti-symmetric in the space-space
entries. The b
1
, b
2
, b
3
entries respectively cause innitesimal boosts in the
1, 2, 3 directions; the r
1
, r
2
, r
3
entries cause innitesimal rotations about the
1, 2, 3 axes. A general element of SO(3, 1) can be written as an exponential
of a nite

= (exp )

+
1
2

+
1
6

+. . . . (C.10)
When only the r
i
are nonzero, this gives a rotation by angle |r| about the
r axis. When only the b
i
are nonzero, this gives a boost by velocity tanh |

b|
along the

b axis. When both r and

b are nonzero the Lorentz transformation

cannot be described either solely as a rotation or as a boost. Note that, while
a rotation by angle |r| = 2 gives the identity , no nonzero magnitude of
boost |

b| returns the identity. Hence the group SO(3, 1) is non-compact.

Now we argue that O(3, 1) has 4 disconnected pieces, one of which is
SO(3, 1). To see this, take the determinant of Eq. (C.6):
Det = Det
T
= Det (Det )
2
. (C.11)
Since is nonsingular, we can divide by Det :
(Det )
2
= 1 Det = 1 . (C.12)
The determinant must vary continuously within a path connected region of
O(3, 1), but you cannot go discontinuously from 1 to 1, so any elements
of O(3, 1) with Det = 1 cannot be elements of the connected group
SO(3, 1). An element of O(3, 1) with Det = 1 is called proper, and an
element with Det = 1 is called improper.
Furthermore, if we write out the = 0, = 0 element of Eq. (C.6), it is

00
=

0
1 =
0
0

0
0
+

i=1,2,3

i
0

i
0
(
0
0
)
2
= 1 +

i=1,2,3
(
i
0
)
2
1 , (C.13)
so the square of the time-time component of any must always be at least
C.2 Generators of the Lorentz group 497
1, and
0
0
must be either 1 or 1. Again, you cannot go continuously
from 1 to 1, so no elements of SO(3, 1) have
0
0
< 0. An element of
O(3, 1) with
0
0
1 is called orthochronous, and an element with
0
0
1
is called non-orthochronous.
The canonical example of an improper (but orthochronous) element of
O(3, 1) is the parity transformation,
P =
_
_
_
_
_
1 0 0 0
0 1 0 0
0 0 1 0
0 0 0 1
_
_
_
_
_
, (C.14)
which satises Eq. (C.6) but has determinant 1. The canonical example of
a non-orthochronous (and also improper) transformation is the time reversal
transformation,
T =
_
_
_
_
_
1 0 0 0
0 1 0 0
0 0 1 0
0 0 0 1
_
_
_
_
_
, (C.15)
which also satises Eq. (C.6) but has T
0
0
= 1. It turns out that any
element of O(3, 1) must be an element of SO(3, 1), times either the iden-
tity (proper orthochronous), P (improper orthochronous), T (improper non-
orthochronous), or PT (proper non-orthochronous). The improper or non-
orthochronous Lorentz transformations need not be symmetries of naturein
fact, in the standard model, they are notbut it is an axiom of eld theory
that the elements of SO(3, 1) must be symmetries.
C.2 Generators of the Lorentz group
As discussed in Section B.1, each element SO(3, 1) must have associated
with it a unitary operator U() which implements it on the Hilbert space,
and which represents the group operation,
U(
1
)U(
2
) = U(
1

2
) . (C.16)
For an element innitesimally close to the identity, it must be possible to
expand these operators in a Lie algebra of generators,
U() = 1 +
i
2

+O(
2
) , (C.17)
498 Lorentz group and the Dirac algebra
for some operators

J

, antisymmetric in , . Similarly, there are genera-

tors for translations,
U() = 1 i

. (C.18)
The

P
i
are also called momentum operators, and the

J
ij
are called angular
momentum operators.
The commutation relations between the operators

P

,

J

can be worked
out by using Eq. (C.16). For instance, consider a translation by a small dis-
tance

, either preceded or followed by a Lorentz transformation involving

. We will evaluate the dierence between the two orders of operation,

in two ways.
First, if we translate rst and then rotate, then the coordinate is trans-
formed according to
x

= x

,
x

= (

)(x

)
= x

+ (

) + (

) , (C.19)
where the rst and second parenthesis represent a rotation and a translation.
The result is the same rotation, and a translation by

plus an extra piece

involving and . If the rotation is performed rst, we get
x

= x

,
x

= x

+ (

) + (

) , (C.20)
which is the rotation and the translation just by . The unitary operators
for these transformations are,
U() = 1 +
i
2

)

P

,
U() = 1 +
i
2

)

P

. (C.21)
The dierence of the operators, to second order in the innitesimals, is
U() U() = i

. (C.22)
(There is actually also a second order in piece, but it is the same for the
two Us and therefore cancels in this dierence.)
Alternately, we can say using Eq. (C.16) that
U() = U()U() =
_
1 +
i
2

__
1 i

_
,
U() = U()U() =
_
1 i

__
1 +
i
2

_
,
C.2 Generators of the Lorentz group 499
U() U() =
i
2
2

,

P

_
. (C.23)
Now equating Eq. (C.22) and Eq. (C.23), we learn what the commutator of

P with

J must be:

i
2
2

,

P

_
= i

. (C.24)
This must hold for any antisymmetric

and any

, so (antisymmetrizing
over the indices on ) the operators must satisfy
_

,

P

_
= i
_

_
. (C.25)
By a completely analogous but more involved procedure one can also show,
_

,

J

_
= i
_

_
, (C.26)
and (this is simpler)
_

,

P

_
= 0 . (C.27)
These commutation relations are called the Poincare algebra.
To make contact with the more familiar generators of rotations and boosts,
it is convenient to dene

J
i


ijk
2

J
jk
, (C.28)

K
i


J
0
i
, (C.29)
which are respectively the generator of rotations about the i axis and of
boosts along the i axis, so a rotation by

is exp(iJ
i

i
) and a boost by
v is exp(iK
i
v
i
). They satisfy the commutation relations, following from
Eq. (C.26),
_

J
i
,

J
j
_
= i
ijk

J
k
, (C.30)
_

J
i
,

K
j
_
= i
ijk

K
k
, (C.31)
_

K
i
,

K
j
_
= i
ijk

J
k
. (C.32)
The rst expression is the familiar commutator between rotations. The
second means that, if a rotation is performed before a boost, the boost will
be in a dierent direction than before the rotation is performed, which is
intuitively clear. The third result is more surprising; the commutator of two
boosts is a rotation. More importantly, the sign is opposite on the last result
than on the previous two.
500 Lorentz group and the Dirac algebra
C.3 Representations of the Lorentz group
Just as for an internal symmetry, an SO(3, 1) transformation will carry a
eld to a linear combination of elds, so the elds must transform under
representations of the group. The dierence is that the transformed eld
will be at the Lorentz transformed point:
U()
a
(x)U

() = D
1
ab
()
b
(x) , (C.33)
with D
1
() = D() an dependent matrix in the space of elds. The
elds can be chosen to block-diagonalize the matrix D into irreducible rep-
resentations of the Lorentz group. For instance, in QED, the components of
the gauge potential A

mix with each other under Lorentz transformations,

but they never mix with the dierent spin components of the electron e
i
;
so there is one block of D which mixes the A

and an independent block

mixing the e
i
. Our goal now is to nd the possible structures D can take.
Just as for internal symmetries, there are two very simple irreducible rep-
resentations, which are also physically important. One is the trivial (singlet)
representation,
U()(x)U

() = (x) , (C.34)
which for SO(3, 1) is called the scalar representation. Lorentz symmetry
demands that the Lagrangian be a Lorentz scalar. The other is the vec-
tor representation, for which the eld index is a four-vector index and the
representation matrix is itself:
U()A

(x)U

() = (
1
)

(x) . (C.35)
A representation is determined by a set of matrices J

with the same

commutation relations as the

J

. That is, the matrix D

ab
must be of the
form
D
ab
() = exp
_
i
2

ab
_
, (C.36)
with the exponentiation interpreted as matrix exponentiation with a, b the
matrix indices, and J

satisfying
_
J

, J

_
= i
_

_
. (C.37)
The problem of nding representations is the problem of nding all sets of
matrices with this algebra.
It is believed that only eld theories containing nite numbers of elds are
well dened. Therefore we need only look for nite-dimensional representa-
tions of SO(3, 1). The classication of the representations is made easier by
C.3 Representations of the Lorentz group 501
the following convenient property of the group. Using Eq. (C.30)Eq. (C.32),
one can show that the operators

L
i

J
i
+i

K
i
2
,

R
i

J
i
i

K
i
2
, (C.38)
satisfy the commutation relations,
_

L
i
,

L
j
_
= i
ijk

L
k
, (C.39)
_

R
i
,

R
j
_
= i
ijk

R
k
, (C.40)
_

L
i
,

R
j
_
= 0 . (C.41)
Therefore the generators of SO(3, 1) can be split into two subsets which
commute with each other, and each satisfy the same commutation relations
as the group SU(2). This group is familiar as the group of rotations and its
representations are well known; they are the spin-zero representation, the
spin-half representation, the spin-one representation, and so forth. A general
irreducible representation can be described by its transformation properties
under

L and under

R, eg, spin m/2 under

L and spin n/2 under

R.
Only four representations will be of any interest to us in studying the
standard model, because it turns out that only four representations can
participate in renormalizable interactions in a theory satisfying the basic
principles laid out in Section 1.2.
The rst of these is the scalar representation already introduced. The
scalar representation transforms as (0, 0), that is, as spin-zero under

L and
spin-zero under

R. The Lie algebra representations are

J

= 0 and the
transformation matrix D = 1 is the identity.
The second common representation is the vector representation, which
transforms as (
1
2
,
1
2
). The Lie algebra is represented as J

= i(

), and D = , as already discussed. Note that for both of these

representations, the matrix D is always real; therefore it is consistent to
consider real valued scalar or vector elds.
The other two interesting representations are called spinor representa-
tions, and consist of two elds which mix with each other under Lorentz
transformations. Since these are probably less familiar to the reader and
are in some ways more complicated than the scalar and vector representa-
tions, we will discuss them at length in the next section.
502 Lorentz group and the Dirac algebra
C.4 Spinors and the Dirac algebra
We now introduce the other two physically important representations of the
Lorentz group, the left and right handed spinor representations. A eld
transforming in one of these can be rewritten in terms of the other, and it
is convenient to combine them together using Majorana notation, which we
will also introduce and which we use throughout this book.
C.4.1 Spinor representations
The simplest nontrivial matrices which satisfy Eq. (C.39) are the Pauli ma-
trices,

1
=
_
0 1
1 0
_
,
2
=
_
0 i
i 0
_
,
3
=
_
1 0
0 1
_
, (C.42)
which satisfy the commutation relation
_

i
2
,

j
2
_
= i
ijk

k
2
. (C.43)
Therefore, if the matrices representing

L
i
and

R
i
are
i
/2 and 0 respec-
tively, we get a representation of the Lorentz algebra. Inverting Eq. (C.38),
rotations and boosts are implemented by the matrices,
J
i
=

i
2
, K
i
= i

i
2
, (Left handed spinor) (C.44)
which it is easy to show satisfy Eq. (C.30) through Eq. (C.32).
Therefore, a pair of elds
a
, a = 1, 2 can transform under Lorentz trans-
formations according to
U()
a
U

() = D
ab
()
b
, D() = [exp(i(r
i
ib
i
)
i
/2)] , (C.45)
where r
i
, b
i
are the amount of rotation and boost performed, as introduced
in Eq. (C.9). The two elds
a
are generally referred to as the components
of a single spinor eld with a the spinor index, which is almost always
suppressed by writing and D in matrix notation ( as a column vector,
D as a matrix). Such a spinor eld is called a left handed Weyl spinor
L
.
Alternately, R
i
could be represented by the Pauli matrices and L
i
by 0s,
J
i
=

i
2
, K
i
= i

i
2
, (Right handed spinor) (C.46)
in which case a Lorentz transform acts on via
U()
a
U

() = D
ab
()
b
, D() = [exp(i(r
i
+ib
i
)
i
/2)] . (C.47)
A eld transforming this way is called a right handed Weyl spinor
R
.
C.4 Spinors and the Dirac algebra 503
Since the matrices D we just constructed are in general complex, a spinor

L
or
R
must be a pair of complex elds. We can ask how the complex
conjugate of
L
transforms. Because complex conjugation ips the i in front
of K in Eq. (C.44), the answer is that it transforms as a right handed Weyl
spinor. More properly, dening the matrix
i
2
=
_
0 1
1 0
_
satisfying

i
=
i
, (C.48)
we see that times the conjugate of
L
transforms according to,
U()

L
U

() =
_
exp
_
i(r
i
ib
i
)

i
2
_

L
_

= exp
_
+i(r
i
+ib
i
)

i
2
_

L
= exp
_
i(r
i
+ib
i
)

i
2
_

L
, (C.49)
which is precisely the transformation rule for a right handed Weyl spinor.
Similarly,

R
transforms as a left handed Weyl spinor, and (

L
)

=
L
transforms as a left handed Weyl spinor again. Both the eld and its complex
conjugate will typically appear in the Lagrangian so it is important to have
a notation which can deal with each. Whether we consider the left or right
handed version as the eld rather than the conjugated object is a matter of
convention.
C.4.2 Weyl, Majorana, Dirac
There are two common notational ways of dealing with the fact that a eld
can be written either as a left or a right handed spinor.
One, called Weyl notation, expresses the elds as two component objects,
and then species whether one is referring to
L
or to its right handed
conjugate

L
by using either an undotted or a dotted index:

=
L
and

L
. (Indices are raised and lowered using and dotted and undotted
according to whether they are conjugated.) This notation is common in the
supersymmetry and string theory literature.
An alternative which we will use, Majorana notation, writes a single four
component eld
M
, dened as

M
=
_

L

L
_
, (C.50)
that is,
M
redundantly records both the left handed and the right handed
504 Lorentz group and the Dirac algebra
ways of writing the eld. The individual pieces can be accessed separately
by using the projection operators
P
L

_
1 0
0 0
_
and P
R

_
0 0
0 1
_
. (C.51)
The action of rotations and boosts on
M
are respectively,
J
i
=
_

i
2
0
0

i
2
_
, K
i
=
_
i
i
2
0
0
i
i
2
_
. (C.52)
If a left-handed spinor transforms nontrivially under an internal symmetry
group, then since the right-handed version involves complex conjugation, the
right-handed version

L
transforms under the conjugate representation.
In particular, if
L
has charge q under a U(1) symmetry and is in the
fundamental representation of an SU(N) symmetry, then

L
has charge
q and transforms under the anti-fundamental representation of SU(N).
One must keep this in mind when constructing Lagrangians out of Majorana
spinors.
In QED and QCD, if we write the spinor elds as left-handed objects, the
elds form pairs with conjugate symmetry transformation properties. For
instance, in QED, there is a eld E
L
which is charge -1 under U
em
(1), called
the left-handed electron, and a eld E

R
which is charge 1 under U
em
(1),
called the left-handed positron. In this case it is most convenient to think
of the latter as the conjugate of a right-handed eld with charge -1, E
R
,
called the right-handed electron, and to combine them together in a single
4-component object called a Dirac spinor, e = [E
L
E
R
]
T
.
The Lorentz transformation properties of Majorana and Dirac spinors are
the same. The two distinctions are that the upper and lower components
of a Dirac spinor generally have the same transformation properties under
internal symmetries, while for Majorana spinors they have conjugate trans-
formation properties; and the upper and lower components of a Dirac spinor
are independent, while for a Majorana spinor they are redundant notations
for the same eld.
C.4.3 Tensor products of spinors
Since the Lagrangian must be a Lorentz scalar, it must be a sum of terms
even in spinorial elds. Therefore we need to know how products of two
spinor elds transform. We will only consider the combination of a spinor
eld
1
with the complex conjugate of another,

2
. This is sucient for
Majorana spinors because
T
2

1
can be re-expressed in terms of

1
, and
C.4 Spinors and the Dirac algebra 505
it suces for Dirac spinors with internal symmetries because only such com-
binations are invariant under the internal symmetries.
The Hermitian conjugate of a spinor eld transforms as
U()

() = (D())

() . (C.53)
The J
i
are Hermitian, but the K
i
are anti-Hermitian, so D() is not in
general unitary. Therefore

does not have the inverse transformation

property of . However, there is a Hermitian, unit determinant matrix ,

_
0 1
1 0
_
, J
i
= J
i
, K
i
= K
i
, (C.54)
which ips the sign of K but not J when commuted across D

, so D

=
D
1
. Therefore, dening =

, called the Dirac conjugate of ,

U()U

() =

() =

D
1
() = D
1
() , (C.55)
so has the inverse transformation property of .
Since has 4 components, there are sixteen independent 4 4 matrices
which can be used to combine spinors,
2

1
. These can all be gotten
from four such matrices, called the gamma matrices

, given in Eq. (1.87).

These satisfy anticommutation relations called the Cliord algebra,
_

_
= 2

1. (C.56)
The matrices J

can be expressed in terms of the gamma matrices:

J

=
i
4
_

_
, (C.57)
which together with Eq. (C.56) is enough to prove that J

satises the
Lorentz algebra, Eq. (C.37). Further, these relations ensure that
_
J

_
= i (

) , (C.58)
from which it follows that
D
1
()

D() =

. (C.59)
Therefore, while the combination
2

1
is a scalar,
U()
2

1
U
1
() =
2
D
1
()D() =
2

1
is a scalar , (C.60)
the combination
2

1
is a vector,
U()
2

1
U
1
() =
2
D
1
()

D()
1
=

1
is a vector .
(C.61)
506 Lorentz group and the Dirac algebra
Similarly, dening

= 2iJ

, the combination
U()
2

1
U
1
() =

1
(C.62)
is a rank-2 antisymmetric tensor.
Next, dene

5
=
5

i
24

= i
0

3
, (C.63)
where the latter follows from the anti-commutation of the distinct gamma
matrices. We have that
U()
2

1
U
1
() =
i
24

2
D
1

D
1
(C.64)
=
i
24

1
= (Det )
i
24

1
= (Det )
2

1
.
Therefore
2

1
is a pseudoscalar, a scalar under SO(3, 1) which ips sign
under parity transformations. Finally, the quantity
2

5
transforms as
a pseudovector,
U()
2

1
U
1
() =
2
D
1

5
D
1
= (Det )

1
.
(C.65)
Since this gives 1 + 4 + 6 + 4 + 1 = 16 independent contractions, the above
are exhaustive; any other matrix sandwiched between
2
and
1
must be a
linear combination of 1,

5
, and
5
.
The choice of matrices made above is called the chiral basis and is con-
venient because the right and left handed components of factorize. How-
ever, multiplying by an arbitrary unitary matrix S and all matrices by
SS
1
leaves the theory unchanged. While the explicit expressions for
the matrices are obviously changed, certain relations are not, and are there-
fore particularly valuable. In particular, the Cliord algebra, Eq. (C.56),
the relations Eq. (C.57), Eq. (C.58), Eq. (C.59), the denition Eq. (C.63) of

5
in terms of the other matrices, and the relations between the projection
operators and
5
,
P
L
=
1+
5
2
, P
R
=
1
5
2
, (C.66)
are basis independent and should therefore be sucient to evaluate any
invariant quantities.