232A Lecture Notes

Representation Theory of Lorentz Group

1 Symmetries in Physics
Symmetries play crucial roles in physics. Noethers theorem relates symme-
tries of the system to conservation laws. In quantum mechanics, conserved
quantities then become the generators of the symmetry. The four forces
we know (gravity, strong, electromagnetic, and weak) are all based on gauge
symmetries. Therefore, it is very important to understand how particle states
and fields transform under the symmetries.
Symmetries always form mathematical groups. To figure out how the
abstract concepts of groups transform states and fields is an issue of rep-
resentation of the group. For representations of compact Lie groups (see
below), there are many excellent textbooks. Ive personally studied one by
Howard Georgi [1].
On the Hilbert space of quantum systems, all symmetries must be given
in terms of unitary or anti-unitary operators so that probabilities do not
change before and after the symmetry transformation. If the states |ai and
|bi transform to U |ai and U |bi under a unitary transformation, their inner
products do not change,

(ha|U )(U |bi) = ha|(U U |bi) = ha|bi. (1)

In the case of anti-unitary operators, we find instead1

(ha|U )(U |bi) = [ha|(U U |bi)] = (ha|bi) = hb|ai, (2)

which shows the initial and final states are interchanged, appropriate for time
reversal. In either case, the probabilities given by the absolute square of the
matrix elements remain the same.
However, the fields are not states. The quantum fields are the canonical
coordinates in the field-theory Lagrangian, and there is no probabilistic inter-
pretation for them. Therefore, fields may transform in non-unitary fashion
under symmetries.
1
To understand this unusual behavior of anti-unitary operators, I find the quantum
mechanics textbook by Albert Messiah [2] the most useful.

1
 In this lecture notes, I describe the representation of the Lorentz group,
a group of spatial rotations and Lorentz boosts, which is hard to find in
mathematics literature mostly focused on compact groups.2 When the fields
transform under the Lorentz group, we need to use non-unitary representa-
tions. On the other hand, when the states in the Hilbert space transform
under the Lorentz group, we use unitary representations.

2 Lorentz Group
The Lorentz group is a symmetry of the Minkowski spacetime that leaves
the origin unchanged. We adopt the metric

+1
1
g = . (3)
1
1

The proper time is defined by

(ds)2 = g dx dx = c2 (dt)2 (d~x)2 . (4)

(Whenever I put an arrow on top of a symbol, it refers to a three-vector

in space. x, y etc without an arrow refers to the whole spacetime coordi-
nate including time.) The Lorentz transformation is supposed to keep this
combination of time and space intervals invariant.
Therefore, the Lorentz group is a collection of matrices O that have the
property
OgOT = g. (5)
It consists of the regular rotations of the three-dimensional space such as

1
1
O= , (6)
cos sin
sin cos
2
The original works, however, are due to mathematicians Bartel Leendert van der
Waerden.

2
as well as invariance under Lorentz boosts such as

cosh sinh
sinh cosh
O= = (7)
1 1
1 1
p
Here, = v/c and = 1/ 1 2 , and we also introduced the rapidity
parameter = cosh , = sinh using the fact that 2 ()2 = 1.
The proper Lorentz group contains only transformations with unit deter-
minant called SO(3, 1), while improper Lorentz transformations are those
transformations with determinant 1 (either parity or time-reversal is in-
cluded.) Combined, they form the group O(3, 1). (For these notations, see
Appendix.) Therefore, a Lorentz transformation falls into one of the four
distinct categories:
categories no parity with parity
no time-reversal proper improper
with time-reversal improper proper

3 Generators
Lorentz transformations without parity nor time reversal are generated by
a set of operators (infinitesimal transformations), just like spatial rotations
are generated by angular momentum operators. They satisfy commutation
relations (Lie algebra). The easiest way to figure it out is to generalize the
orbital angular momentum operators (in the unit of ~) in coordinate space
1 1
Li = ijk xj pk = iijk xj k , (8)
~ ~
and define
M = i(x x ). (9)
Note that 1 = 1 = x etc because of the negative metric for spatial
coordinates. Here and below, we use the notation that Greek indices are
Lorentz indices = 0, 1, 2, 3 with distinction between upper and lower in-
dices, while the Latin indices are only space i = x, y, z without distinction
between upper and lower indices.

3
 Using the fact x = g , it is easy to work out

[M , M ] = [i(x x ), i(x x )]
= (x g x g x g + x g
+x g x g x g + x g )
= i(g M + g M g M g M ). (10)

The generators of rotation are given by Jx = M 23 , Jy = M 31 , Jz = M 12 . To

see this, we can verify, e.g.,

[Jx , Jy ] = [M 23 , M 31 ] = ig 33 M 21 = iJz . (11)

In general,
[Ji , Jj ] = iijk Jk . (12)
The operators Ki = M 0i generate boost along the axis i = x, y, z. They
satisfy the commutation relations

[Jx , Ky ] = [M 23 , M 02 ] = ig 22 M 30 = iKz . (13)

In general,
[Ji , Kj ] = iijk Kk . (14)
It basically shows that boost generators rotate among themselves as a three-
vector. Finally, the commutation relations among the boost operators are

[Kx , Ky ] = [M 01 , M 02 ] = i(g 00 M 12 ) = iJz . (15)

In general,
[Ki , Kj ] = iijk Jk . (16)
One way to verify that M 0i above indeed generates Lorentz boost, see

M 01 t = i(x0 1 x1 0 )t = i(tz xt )t = ix, (17)

M 01 x = i(x0 1 x1 0 )x = i(tz xt )x = it. (18)

Therefore, I can write

      
01 t 0 1 t t
M = i = ix . (19)
x 1 0 x x

4
Then it is easy to compute the exponential of this generator,

iM 01 x
X n
e =e = (x )n . (20)
n=0
n!

Since (x )2 = I2 (two-by-two identity matrix), all the odd powers are x and
even powers I2 . Therefore,
"
# 
 
t X n n t
iM 01
X
e = I2 + x
x n! n! x
n=even n=odd
 
t
= [cosh I2 + sinh x ]
x
  
cosh sinh t
= . (21)
sinh cosh x

This is nothing but the boost Eq. (7). Note that this matrix is not unitary.
At this point, M (and hence J~ and K) ~ are abstract operators, just
~
like ~x, p~, L = ~x p~ in quantum mechanics. The task ahead of us is to
come up with something more concrete, just like states |jmi in quantum
mechanics and J , Jz acting on them. Namely, we need a column vector on
which operators act in forms of matrices. We then have to make sure that
these matrices satisfy the same commutation relations. The combination of
matrices and column vectors is called a representation of the Lie algebra
(commutation relations).
Note that the Lorentz group is non-compact. Imagine doing many same
small transformations successively. While the rotations eventually bring back
to where you started once the angle is 2 (compact), boost can go on for
ever and does not come back because above can go all the way to infinity
(non-compact). It is a mathematical theorem that says that non-compact
groups do not have finite-dimensional unitary representations. The reason is
very simple. A unitary matrix, multiplied many many times, will eventually
come back to unity. Unless the space is infinite-dimensional, so that you
may never come back to where you started. Indeed, the boost Eq. (7) in a
four-dimensional representation is not a unitary matrix.
For fields, we need to write their Lagrangians and do not want them to
have an infinite number of components. We assign fields to transform under
finite-dimensional representations of the Lorentz group. Therefore, the boost

5
generators are represented by non-unitary matrices, as you will see explicitly
later. Correspondingly, the boost generators are not hermitian.
On Hilbert space, however, we want all generators to be quantum me-
chanical observables and hence hermitian. Therefore, the Hilbert space must
be infinite-dimensional unitary representations. We will also see later how
to construct such representations.

4 Representation Theory of Lorentz Group

As mentioned earlier, we need to come up with finite-dimensional non-unitary
representations of the Lorentz group to be assigned for fields (not Hilbert
space). Here is how we can classify all such representations.
Using the commutation relations we derived, we can identify two sets of
mutually commuting operators. Let us define
L
i = (Ji iKi )/2. (22)
Then we find
[L
i , Lj ] = [Ji iKi , Jj iKj ]/4 = iijk (Jk iKk iKk Jk )/4 = iijk Lk ,
(23)
while the different sets commute

[L+
i , Lj ] = [Ji + iKi , Jj iKj ]/4 = iijk (Jk iKk + iKk Jk )/4 = 0. (24)

Therefore there are two sets of Lie algebras which have the same form as those
of angular momentum operators su(2). Therefore any finite-dimensional rep-
resentations of Lorentz group are specified by two numbers, (j+ , j ), one for
3
the generators L+i and other for Li . The representation is (2j+ +1)(2j +1)
dimensional.
Note that under the rotation group, we discard the boost operators K ~
and there is no distinction between the two sets. Therefore, both j+ and j
come in for spatial rotations, and we add angular momenta j+ and j to find
angular momenta j = |j+ j |, |j+ j | + 1, j+ + j .
In order to obtain spin 1/2 under spatial rotations without anything else
we dont want, we find only two inequivalent possibilities, ( 21 , 0) and (0, 21 ).
Both of them are indeed j = 1/2 under rotations. The difference is in the
properties under the boost. Let us see how each of them transforms.
3
Mathematically, there is an isomorphism so(3, 1) C
= sl(2, C) sl(2, C), where each
sl(2, C) factor picks j representation.

6
4.1 ( 21 , 0) spinors
~ = J~ iK
For the ( 12 , 0) representation, j = 0 and therefore L ~ = 0. On
+
the other hand, the generators of L are represented by Pauli matrices

~ + = 1 ~ .
L (25)
2
Therefore we know all of them now,
1 ~ = i 1 ~ .
J~ = ~ , K (26)
2 2
Note that the boost generators K ~ are represented by non-Hermitian matrices,
~
and hence the representation is non-unitary (namely ei~K below is not a
unitary matrix).
A two-component ( 21 , 0)-spinor transforms under spatial rotation by
" #
~ J~
i ~ /2
i~ ~ ~
e =e = cos + i sin . (27)
2 2

Here, I used the notation ||. ~ The same spinor transforms under the
Lorentz boost as
 
i~ ~
K ~ ~
/2 ~ ~
e =e = cosh + sinh . (28)
2 2
Here, I used the notation |~ |.

4.2 (0, 21 ) spinors

In this case, we set L ~ + = J~ + iK ~ = 0. Again using the spin 1/2 nature,
~ = 1 ~ , and hence
L 2
1 ~ = +i 1 ~ .
J~ = ~ , K (29)
2 2
The only difference from the previous case is the sign of the boost operators.
Therefore we can immediately work out the transformations of (0, 21 )-spinor
" #
~ ~ ~ ~
~

eiJ = ei~/2 = cos + i sin (30)
2 2

7
under rotation (exactly the same as ) while
 
i~ ~
K ~ ~
/2 ~ ~
e =e = cosh sinh (31)
2 2

under Lorentz boosts (note the minus sign in the square bracket).
One point to note is that two spin 1/2 representations are related by
complex conjugation (up to unitary equivalence). Using the identity

(i2 )~ (i2 ) = ~ , (32)

it is easy to see that i2 transforms the same way as and vice versa.
Under rotations,
~ ~ /2 ~
i2 i2 (ei~/2 ) = i2 (ei~ (i2 )(i2 ) = ei~/2 (i2 ) , (33)

while under boosts,

/2
i2 i2 (e~~/2 ) = i2 e~~ (i2 )(i2 ) = e~~/2 (i2 ) . (34)

Therefore, two representations are practically complex conjugates to each

other (up to the unitary transformation i2 ).

4.3 Dirac spinors

There is a much more elegant way to obtain spin 1/2 representation using
Clifford algebra,4 as discussed in PeskinSchroeder. This construction works
in any dimensions of spactime, but it is not clear why and does not give
you any other spins. For the latter purpose, you need to go back to the
construction in the previous subsections.
It is a two-step process. The first step is to identify matrices that satisfy
the Clifford algebra
{ , } = 2g . (35)
In four-dimensional Minkowski spacetime, PeskinSchroeder picks
   
0 0 I2 i 0 i
= , = . (36)
I2 0 i 0
4
For Clifford algebra and spinors in arbitrary dimensions, see my 230A lecture notes.

8
It is easy to verify that they satisfy the required Clifford algebra. The main
point is that once -matrices are found, they will immediately give a repre-
sentation of the Lorentz group with the generators
i
M = [ , ]. (37)
4
Again, it is straightforward to verify that they satisfy the commutation re-
lations Eq. (10).

4.4 Four-vectors
4.5 Field Strengths

5 Poincar
e Group and Little Group
Lorentz transformations are not the only symmetries of the Minkowski space-
time. Translation in space and time are also symmetries. As you know al-
ready, translations are generated by energy and momentum operators P =
(E, p~). Therefore, it is useful to include them in our discussions as well.
Because all symmetries are supposed to be unitary (or anti-unitary) opera-
tors on the Hilbert space to preserve the probabilities, we now look for their
unitary representations.
Obviously, translations in different directions commute:

[P , P ] = 0. (38)

On the other hand, the energy-momentum four-vector should transform as a

four-vector, and hence

[M , P ] = i(g P g P ). (39)

Together with the commutation relations among the Lorentz generators M ,

they form the Pointcare algebra.
To see how Lorentz transformation act on the Hilbert space of Quantum
Field Theories, we need to represent the entire Poincare group, not just the
Lorentz group.

9
 If you specialize them to the spatial components only in three-dimensional
space, we recover familiar commutators from quantum mechanics, by identi-
fying J k = kl M Lem ,

[J 1 , J 2 ] = [M 23 , M 31 ] = ig 33 M 21 = i(1)(M 12 ) = iJ 3 , (40)
[J 1 , P 2 ] = [M 23 , P 2 ] = ig 22 P 3 = iP 3 . (41)

There are two Cosimo operators of the Poincaree symmetry. One is P 2 =

P P , which obviously commutes with P and is also Lorentz invariant and

hence commutes with M . A one-particle state is obviously its eigenstate

with the eigenvalue P 2 = m2 . The other Cosimo operator is made of Pauli
Lubanski pseudo-vector
1
W =  P M . (42)
2
Note that the orbital angular momentum drops out from M because of the
anti-symmetry with the momentum vector. Namely it picks up only the spin
part of the angular momentum.
Commutators of W with P leave another P or P contracted with the
LeviCivita symbol and hence vanish. It transforms as a Lorentz vector and
hence has the same commutator with M as the momentum vector, but has
the opposite transformation under parity. Clearly W 2 = W W is Lorentz
invariant and hence commutes with all generators.
For a massive particle P 2 = m2 > 0, we can always go to its rest frame

P = (m, 0, 0, 0). We define the little group of the Poincare group that pre-
serves this four-momentum. It is nothing but the group of three-dimensional
rotations SO(3). We can classify states according to the SO(3) eigenval-
ues, namely spin s and one of its component, say, sz . Because the only
non-vanishing component of the four-momentum is P 0 = m, only spatial
components of the PauliLubanski pseudo-vector survives, W ~ = m~s. There-
2 ~ 2 2
fore, W = W = m s(s + 1). All massive particles, therefore, can be
classified by their mass and spin eigenvalues.
The situation for massless particles P 2 = 0 are tricky. We can always go
to the frame where P = E(1, 0, 0, 1). This four-momentum does not change
under the rotation of the x y plane. However, note that

[M 0i , P ] = i(g 0 P i g i P 0 ) = ig i P 0 , (43)
[M 3i , P ] = i(g 3 P i g i P 1 ) = ig i P 3 . (44)

10
Since P 0 = P 3 = E, we find that M 0i M 3i remain symmetries, and hence a
part of the little group. Namely, Jz , Kx Jy , and Ky + Jx are the generators
of the little group with the commutation relations

[Kx Jy , Ky +Jx ] = 0, [Jz , Kx Jy ] = i(Ky +Jx ),

[Jz , Ky +Jx ] = i(Kx Jy ).
(45)
They are equivalent to the group of rotation of a plane and translations in
two directions, called Euclidean motion group in two dimensions E2 .
The problem here is that E2 is non-compact, and its unitary represen-
tations are in general infinite-dimensional. Once that we specified four-
momentum of the massless particle, we dont think there should be infinite
number of states left. The only way to find a finite-dimensional unitary rep-
resentation is to set all non-compact generators consistently to zero. We can
do so, by requiring Kx Jy = Ky + Jx = 0 with no contradictions with the
commutation relations. Then the only remaining generator is the rotation
around the z-axis Jz , which is nothing but the helicity of the particle,

J~ p~
h= = Jz (46)
|~p|

if p~ k z. It can have an arbitrary eigenvalue. Hence, any massless particle

carries the helicity eigenvalue h in addition to its four-momentum. Then the
CPT theorem says its anti-particle must carry the opposite helicity eigenvalue
h.
What are the allowed values of h? The discussion here does not give any
additional constraints. The real constraint is that the states must appear as
a result of the quantization of fields, that restricts the spin to be half-integers.
Therefore, the quantum field theory only allows for half-integer helicities for
massless particles.
If you want to know more about the Poincare symmetry, its little group,
and its unitary representation in arbitrary dimensions, consult my 230A lec-
ture notes here.

A Groups
A mathematical group is defined by a set of rules for multiplications. For
any element g1 , g2 G, there is a product g1 g2 G. It is called abelian if
the multiplication is commutative g1 g2 = g2 g1 for any elements, while it is

11
non-abelian if some of the elements do not commute. There has to be one
element called identity e which satisfies ge = eg = g. In addition, there must
be the inverse g 1 for any element gg 1 = g 1 g = e. Then G is called a
group.
Groups with smooth continuous parameters are Lie groups. Translations
can be done by a little, or a lot. A succession of little translations will get to
a big translation. This is an example of Lie groups.
Generators of the group correspond to infinitesimal transformations, namely
group elements very close to the identity element. The multiplication rules
of a group are in one-to-one correspondence to the commutation relations
of the generators. The familiar examples are the momentum operator p~
generating the spatial translation by a finite distance ~a as U (~a) = ei~p~a , or
~ = eiJ ~ ~
angular momentum operators J~ generating the spatial rotation U ()
around the axis ~ by the angle ||.
~ The momentum operators commute among
~
each other, and hence two translations simply add: U (~a)U (~b) = ei~p~a ei~pb =
~
ei~p(~a+b) = U (~a + ~b). Namely the vanishing commutators [pi , pj ] = 0 specify
this property of spatial translations. On the other hand, two successive rota-
tions are quite complicated in general. When U (~1 )U (~2 ) = U (~3 ), it is not
easy to write ~3 in terms of ~1,2 . This is a consequence of the commutation
relations.
Therefore, the commutation relations among generators basically define
the group. The commutation relations are called Lie algebra.

B Names of continuous groups

Groups with smooth continuous parameters are called Lie groups. Among
them, compact Lie groups are the most studied. These are groups where
doing a small transformation many many times will bring you back to where
you started. Rotation of the Earth is a good example. On the other hand,
non-compact Lie groups allow for a transformation to keep going indefinitely.
Lorentz boost and translation in space are good examples.
There are three types of classical groups.

B.1 General and Special Linear Groups

The collection of all N N real matrices with non-zero determinant (so that
inverse exists) is called GL(N, R), general linear. If you require the matri-

12
 GL(N, C) GL(N, R) SL(N, C) SL(N, R)
2N 2 N2 2N 2 2 N2 1
SO(N ) U (N ) SU (N ) U Sp(N ) Sp(N, R)
1 2 2
2
N (N 1) N N 1 2N (2N + 1) N (2N + 1)

Table 1: The number of generators for representative classical groups.

ces to have determinant one, they form a subgroup SL(N, R) called special
linear. The word special normally means determinant is one. Exactly the
same works for complex numbers, defining GL(N, C) and SL(N, C). The
names of their Lie algebras are usually written with small German letters,
sl(N, C), gl(N, R) etc.

B.2 Orthogonal Groups

An orthogonal group O(N ) is a group of N N orthogonal matrices that
satisfy OOT = I. The determinant of an orthogonal matrix can be either +1
or 1. The orthogonal matrices with determinant +1 form a subgroup of
O(N ), called SO(N ). S stands for special, which is supposed to mean unit
determinant. In most physics applications, orthogonal groups appear only
with real coefficients, O(N, R) or SO(N, R), where the elements correspond
to rotations of space. The only difference between the two is if you all
parity-like transformations. Therefore, the Lie algebras, which by definition
can refer only to the neighborhood of the origin, do not know the difference
between the two. Therefore, we refer to them as so(N ) to make it clear that
we cannot reach the parity-like transformations in O(N ) by exponentiating
the generators.5
A generalization is indefinite orthogonal groups. Instead of requiring
OIOT = I, we introduce an indefinite metric

Ip,q = diag(1, , 1, 1, , 1), (47)

| {z } | {z }
p q

5
Fields with half-odd spins change their signs under 2 rotations. Therefore, strictly
speaking, the half-odd spins are not representations of the rotation groups, but of their
double cover . Namely that you define an enlarged group Spin(N ) where the 2 rotations
form a Z2 subgroup containing only two elements +1 and 1. The rotation group is
then a coset group SO(N ) = Spin(N )/Z2 . The familiar situation is the three-dimensional
rotation group SO(3) = Spin(3)/Z2 .

13
and require OIp,q OT = Ip,q . In other words, it is an invariance of the
Minkowski-like spacetime with p space and q time directions (or vice versa).
It defines again O(p, q) or SO(p, q) groups. The proper Lorentz group of our
four-dimensional Minkowski spacetime is SO(3, 1), and together with parity
and time-reversal, O(3, 1).

B.3 Unitary Groups

A unitary group U (N ) is a group of N N unitary matrices that satisfy
U U = I. The determinant of a unitary matrix can be any phase. The
unitary matrices with unit determinant +1 form a subgroup of U (N ), called
SU (N ). In most physics applications, unitary groups appear only with com-
plex coefficients, U (N, C) or SU (N, C). The difference between U (N ) and
SU (N ) is that U (N ) contains an additional freedom of an overall phase
change. Since the phases form an abelian group of 1 1 unitary matrices
U (1), we see U (N ) SU (N ) U (1).6
A generalization is indefinite unitary groups. Instead of requiring U IU =
I, we introduce an indefinite metric

Ip,q = diag(1, , 1, 1, , 1), (48)

| {z } | {z }
p q

and require U Ip,q U = Ip,q . It defines U (p, q) or SU (p, q) groups.

B.4 Symplectic Groups

This is a group that shows up in classical Hamiltonian mechanics when you
discuss canonical transformations. We look for linear transformations among
the canonical coordinates qi and their conjugate momenta pj while preserving
Poisson brackets {qi , pj } = ij for i, j = 1, , N . Namely, we define a big
6
I used because this is not precise. The element e2i/N belongs to both SU (N ) and
U (1), and generate a ZN subgroup. Therefore, I need to avoid double counting and hence
U (N ) = [SU (N ) U (1)]/ZN to be precise.

14
2N -component column vector
q1

..
.
qN

x= . (49)
p1
.
..
pN
The Poisson brackets then form a big 2N 2N matrix,
 
0 IN
J = ({xi , xj }) = . (50)
IN 0
Here, IN is an N N identity matrix.
Linear transformations on the vector x Sx preserve the Poisson brack-
ets if
S T JS = J. (51)
The matrices S form the symplectic group Sp(2N, R).7
The generators appear for infinitesimal transformations S = I + i +
O()2 . Eq. (51) implies
T J + J = 0. (52)
Using the fact J T = J, we see (J)T = T J T = T J = J. Therefore,
J is a general 2N 2N symmetric real matrix with N (2N + 1) independent
components. If orthogonality is further imposed, there remain N 2 generators,
and the group is equivalent to U (N ).
If you allow for complex coefficients, Sp(2N, C) has 2N (2N + 1) gen-
erators. What appears often in the physics literature is U Sp(2N ), where
unitarity is also required. It gives an additional requirement of hermiticity
on the generators,
= . (53)
Then
J J = J T J = J(J) = . (54)
This cuts down the number of generators to its half N (2N + 1).
7
There are two camps with this notation. Since the vector space is 2N -dimensional,
some people prefer to use 2N . On the other hand, the phase space is always even-
dimensional because of (q, p) pairs and the factor of two is not a useful piece of information.
Some feel N is sufficient. Just be careful which notation is being used.

15
B.5 Equivalence of Small Groups
Some of the classical groups are equivalent when they are small. The well-
known example is between the rotation group SO(3) and group of spin 1/2
SU (2). To be precise, SO(3) comes back to +1 for 2 rotation, while
2 rotation is 1 for SU (2), and hence SO(3) = SU (2)/Z2 . Similarly,
U Sp(2) = SU (2), SO(4) = [SU (2) SU (2)]/Z2 , SO(5)
= U Sp(4)/Z2 ,

SO(6) = SU (4)/Z2 .

B.6 Simple Compact Groups

Simple compact Lie algebras have been completely classified by Dynkin. The
word simple means that the Lie algebra does not contain a generator that
commutes with all the other generators, and the generators cannot be split
into more than one mutually commuting sets. There are four series. AN ,
BN , CN , DN generates SU (N + 1), SO(2N + 1), U Sp(2N ), and SO(2N ),
respectively. There are several simple compact Lie algebras that do not
belong to any of the series and hence called exceptional, E6 , E7 , E8 , G2 ,
and F4 . E6 and E8 appear as candidates of unified theories, G2 in string
compactifications.

References
[1] H. Georgi, Lie Algebras in Particle Physics: from Isospin to Unified
Theories, 344 pages, Westview Press, 1999.

[2] A. Messiah, Quantum Mechanics, 1152 pages, Dover Publications,

1999.

16