Representations of the Symmetry Group of Spacetime

Kyle Drake, Michael Feinberg, David Guild, Emma Turetsky

March 11, 2009

Abstract
The Poincare group consists of the Lorentz isometries combined with Minkowski
spacetime translations. Its connected double cover, SL(2, C) n R1,3 , is the symmetry
group of spacetime. In 1939 Eugene Wigner discovered a stunning correspondence
between the elementary particles and the irreducible representations of this double
cover. We will classify these representations and explain their relationship to physical
phenomena such as spin.

Introduction

Minkowski spacetime is the mathematical model of flat (gravity-less) space and time. The
transformations on this space are the Lorentz transformations, known as O(1,3). The
identity component of O(1,3) is SO+ (1,3). The component SO+ (1,3) taken with the translations R1,3 is the Poincare group, which is the symmetry group of Minkowski spacetime.
However, physical experiments show that a connected double cover of the Poincare is more
appropriate in creating the symmetry group actual spacetime.
Why the double cover of the Poincare group? What is so special about this group in
particular that it describes the particles? The answer lies in the way the group is defined.
Simply, it is the set of all transformations that preserve the Minkowski metric of Minkowski
spacetime. In other words, it is the set of all transformations of spacetime that preserve
the speed of light.
SL(2,C) is the connected double cover of the identity component of SO+ (1,3). This is
the symmetry group of spacetime: the double cover of the Poincare group, SL(2,C) n R1,3 .
In 1939, Eugene Wigner classified the fundamental particles using the irreducible representations of the double cover of the Poincare group. Wigner was motivated by the idea
that symmetry underlies all physical laws. In particular, a physical experiment should come
up with the same results regardless of where, when, or what orientation the experiment
is done in. An experiments results should also be invariant whether the experiment is at
rest or moving at a constant velocity. It turns out the symmetries of physics go further
than this. It is common to combine several systems together such as when protons and
electrons combine to form atoms. When this is done, the overall symmetry of the system
should be related to the individual symmetries of its components. These components are
the irreducible representations [3].
The double cover of the Poincare group acts to classify the fundamental particles in
physics and explain patterns in their behaviors. In particular, the particles are most easily
classified by the irreducible representations of the double cover of the Poincare group. The
representations are determined by the different orbits in the group. These orbits serve to
classify types of particles.
The first two types of orbits correspond to the value m2 > 0. The value m2 is equivalent
to the idea of mass in physics. Thus elements in the first two orbits correspond to massive
particles which travel slower than the speed of light.
The particles that correspond to the light cones are those with m2 = 0, which are
particles that travel at the speed of light. Examples of these particles are photons, from
the which the light cone gains its name, and gravitons, which currently only exist in theory.
The particles on the single-sheet hyperboloid with values m2 < 0 are called tachyons.
They travel faster than the speed of light and have imaginary mass. As tachyons have
never been observed, we will not find the representations corresponding this orbit.
Finally, the orbit in which m2 = 0 and the particles are not moving through time
corresponds to vacuum. Vacuum is devoid of all particles, and thus is uninteresting to the
2

purposes of this paper.

It is somewhat surprising that the correspondences between the orbits of the double
cover of the Poincare group and types of particles line up as well as they do. The spin
values of particles predicted by physicists are exactly the values that come as a result of
finding the irreducible representations of the double cover of the Poincare group. This fact
is further explained by the fact that the Poincare groups covering space is connected and
has nice properties that emulate those observed in quantum spin.

The Poincar
e group

In order to describe the Poincare group, we first need to work through some preliminaries.
The orthogonal group O(n) is the group of n n real matrices whose transpose is equal
to their inverse. In other words, A O(n) if
AT

= A1 .

(1)

One can define an orthogonal group more generally, as follows. Given a vector space V
equipped with a symmetric bilinear form h , i, then the corresponding orthogonal group is
the group of all linear transformations of V that leave h , i invariant. So A O(V, h , i) if
hAx, Ayi = hx, yi
for all x, y V . By the Spectral Theorem [1, p. 397], there exists a basis in which we can
express h , i as a diagonal matrix g, where
hx, yi = xT gy.
Applying this to the definition of the orthogonal group gives us
(Ax)T g(Ay) = xT gy
= xT AT gAy = xT gy
= AT gA = g.

(2)

Recall that a real inner product is symmetric, bilinear, and positive definite. If V = (Rn , ),
the metric tensor g is the n n identity matrix and (2) simplifies to (1).
The signature of a symmetric bilinear form is the number of positive and negative
eigenvalues of its matrix, and is usually written (p, q). It turns out that the signature is
enough to define the orthogonal group; this group is usually written O(p, q). This group
is well-defined because given any two inner products with identical signature there is an
isomorphism between their respective groups. We will simplify things by choosing all
eigenvalues to have norm 1. In this notation, the standard orthogonal group in Rn is
O(n, 0).
3

2.1

Minkowski Spacetime

Much of our work will be done in Minkowski spacetime, a four dimensional real vector space
R1,3 . This space is distinguished by its inner product, which is a symmetric nondegenerate
bilinear form with signature (1, 3) or (3, 1). We will be using the former, and will be taking
the first dimension to be time. Note that this inner product does not match the normal
definition of an inner product because it is not positive-definite. This inner product, or
Minkowski metric, is
h~v , wi
~ = v0 w0 v1 w1 v2 w2 v3 w3 ,
where ~v and w
~ are four-dimensional vectors. Similarly, the Minkowski norm is
p
||~v || = h~v , ~v i = v02 v12 v22 v32 .
The group O(1, 3) is the set of linear transformations of Minkowski spacetime that
preserves the Minkowski metric. We can write the metric tensor

1 0
0
0
0 1 0
0

g =
0 0 1 0 .
0 0
0 1
Notice that g 1 = g. From (2), we have O(1, 3) = {A | AT gA = g}. This is not an easily
solvable equation, and we will not do so here. Instead, we will describe this group by its
generators.

2.2

O(1, 3)

The group O(1, 3) is known as the Lorentz group, and elements of this group are Lorentz
transformations. It has four connected components. Roughly speaking, a Lorentz transformation may or may not preserve the direction of time, and it may or may not preserve
the orientation of space; these choices correspond to the four connected components. The
two components that preserve time form the subgroup O+ (1, 3). The two components
that preserve the orientation of space form the subgroup SO(1, 3) which are the elements
with determinant 1. The identity component, which preserves both time and space, is the
subgroup SO+ (1, 3). This subgroup is the group of proper Lorentz transformations.
There are six types of transformations that generate SO+ (1, 3). Three of the generators
are simple spatial rotations, and the other three are time-space operators known as boosts.
The rotations Ri and boosts Bi are given below:

Rx =

Ry =

Rz =

Bx =

By =

Bz =

1 0
0
0
0 1
0
0

0 0 cos sin ,
0 0 sin cos

1
0
0
0
0 cos 0 sin
,

0
0
1
0
0 sin 0 cos

1
0
0
0
0 cos sin 0

0 sin cos 0 ,
0
0
0
1

cosh sinh 0 0
sinh cosh 0 0
,

0
0
1 0
0
0
0 1

cosh 0 sinh 0
0
1
0
0

sinh 0 cosh 0 ,
0
0
0
1

cosh 0 0 sinh
0
1 0
0

.
0
0 1
0
sinh 0 0 cosh

Special relativity says that two observers moving at constant velocity with respect to
one another will perceive a single event at different places and at different times. The
change from one observers reference frame to anothers is handled by boosts.
Assume that the two observers agree on the origin, and that their x, y, and z axes are
respectively parallel. Without loss of generality we will consider their relative velocity to
be in the x direction. In order to avoid conversion factors between units of measurement,
we will express time in seconds, distance in light-seconds, and velocity in light-seconds per
second (equivalently, as a fraction of the speed of light). Let
=

1
p
.
1 2

The transformation from one observers reference frame (t, x) to the second observers

reference frame (t0 , x0 ) is

t0 = (t + x)
x0 = (x + t)
y0 = y
z 0 = z.
Then
0

t
0 0
t
x0
0 0 x
=
.
y 0
0
0 1 0 y
z0
0
0 0 1
z
This is a symmetric matrix of one variable. Introducing the angle
1+
= ln p
1 2
allows us to rewrite this matrix as Bx . Notice how boosts are similar to rotations, but
involve the hyperbolic functions rather than the circular functions. This is due to the
difference in sign between space and time.
So far, we have not proven that these are generators of SO+ (1, 3). We will do so by
using the Lie algebra.

2.3

The Lie Algebra so(1, 3)

Recall that a Lie group, such as O(1, 3), is a smooth manifold as well as a group. The Lie
algebra of a Lie group is defined in terms of matrix exponentiation, which we will cover
later. For now we will use the fact that the Lie algebra is also the tangent space at the
identity equipped with a binary operator known as the Lie bracket. For matrix groups,
the bracket is the familiar matrix commutator [A, B] = AB BA.
We will proceed to find the Lie algebra corresponding to O(1, 3). Since we are only
considering the local properties of the identity element, the structure of the algebra will
only reflect the identity component of the group. Thus the Lie group O(1, 3) and the
subgroups discussed earlier all have the same Lie algebra, which we denote as so(1, 3).
Let A : R O(1, 3) be a continuous function with A(0) = I. Recalling the definition
of O(1, 3) from (2), we see that
A(t)T gA(t) = g
= A(t)T g = gA(t)1 .
Any element of the tangent space is the derivative of such a curve A at t = 0. Finding the
characteristic equation of the algebra is now simply a calculus problem:
6



d
A(t)T g
dt

T
d
=
A(t)
g
dt
 
T
d 
=
A(t)
g
dt t=0

d 
A(t)
=
dt t=0
Let B =

b11
b21

b31
b41

d
dt t=0 A(t).

b12
b22
b32
b42

b13
b23
b33
b43

d
gA(t)1
dt


d
dA(t) A(t)2
= g
dt
 

d 
= g
A(t) A(0)2
dt t=0
 
T
d 
= g
A(t)
g.
dt t=0
=

Then B = gB T g, which can be worked out as

b14
1 0 0 0 b11

b24
= 0 1 0 0 b12
0 0 1 0 b13
b34
0 0 0 1 b14
b44

b11 b21
b31
b12 b22 b32
=
b13 b23 b33
b14 b24 b34

b21
b22
b23
b24

b31
b32
b33
b34

b41
1 0
0
0

b42
0

0 1 0
b43 0 0 1 0
0 0
0 1
b44

b41
b42
.
b43
b44

Therefore B is of the form

0 a
b
a 0
d
B =
b d 0
c e f

c
e
.
f
0

(3)

Let Va be the matrix of this form where a = 1 and b = c = = 0, and define Vb ,

Vc , Vd , Ve , and Vf likewise. These six matrices are a basis for so(1, 3) as a vector space.
From this we know that so(1, 3) is 6-dimensional. Since a manifold has the same number
of dimensions as its tangent space, SO+ (1, 3) must be 6-dimensional as well. We claimed
earlier that the six Ri and Bi matrices were generators of SO+ (1, 3). Now we will show
how the basis elements Vi of so(1, 3) correspond to the generators of SO+ (1, 3).
The exponential map
The exponentiation operator exp is defined as

X
1 k
exp A =
A .
k!
k=0

The famous irrational e can be defined as exp 1. However, we are not limited to exponentiating scalars; the exp operator works equally well on matrices. For example,



k

X
1 0 a
0 a
exp
=
a 0
k! a 0
k=0

 







1 a2 0
1 0 a3
1 a4 0
1 0
0 a
=
+
+
+
+ ...
+
0 1
a 0
2 0 a2
6 a3 0
24 0 a4
 2k




X
1
1
a
0
0
a2k+1
+
=
0
(2k)! 0 a2k
(2k + 1)! a2k+1
k=0


cosh a sinh a
=
.
sinh a cosh a
This example shows how to compute exp Va . If we were to exponentiate the full 4 4
matrix, which merely adds rows and columns filled with zeroes, we would get e0 = 1 on
the diagonals. In other words,

cosh sinh 0 0
0 a 0 0
sinh cosh 0 0
a 0 0 0
,

exp
0 0 0 0 = 0
0
1 0
0
0
0 1
0 0 0 0
which is simply Bx .
Similar calculations show that the infinitesimal generators Vb and Vc exponentiate to
By and Bz , respectively. Furthermore, the exponentiation of the generators Vd , Ve , and Vf
yield Rz , Ry , and Rx .
This leads to an alternative definition of the Lie algebra g corresponding to a matrix
Lie group G:
g = {A | exp A G}.
Since SO+ (1, 3) is connected, the image of so(1, 3) under the exponential map is the entire
group [2, p. 116], and this definition matches the earlier one.

The symmetry group of Minkowsky spacetime

After exploring the Lorentz group, we turn to the full group of symmetries of Minkowski
spacetime: the Poincare group. Like the Lorentz group, elements of the Poincare group
preserve the Minkowsi metric; however, they need not be linear transformations. We further
restrict the Poincare group to those transformations that preserve both the direction of
time and the orientation of space. The Poincare group is SO+ (1, 3) n R1,3 , the semidirect product of the proper Lorentz transformations with the translations of Minkowski
spacetime.
8

Now that we have expressed the Poincare group, it would seem that we are in a position
to answer any questions about the symmetries of Minkowski spacetime. Unfortunately, the
universe is not completely defined by this 4-dimensional vector space, and we will have to
slightly expand our view.

The Double Cover of the Poincar

e Group

In 1922, Otto Stern and Walther Gerlach performed an experiment which showed the
existence of a physical phenomenon called quantum spin. The experiment involved shooting
silver atoms through a magnetic field and measuring where they hit a screen on the other
end. Under the classical model of physics, it would be expected that the atoms would
land somewhere on a spectrum of points on the screen. However, the atoms struck only
two discrete points. This is due to a feature called spin. The two paths the atoms take
correspond to the two spin states of electrons [6].
An electron can either have spin + 12 or spin 12 . This spin factor essentially tacks on
a fourth piece of information to the three velocity coordinates that already describe the
electrons motion. The existence of this spin information implies that SO+ (1, 3) is not a
perfect model for classifying the fundamental particles. Instead, we are interested in the
covering space of the proper Lorentz group, namely SL(2, C).
In order to account for spin, we must show that the special linear group SL(2, C) is
the double cover of the proper Lorentz group. SL(2,C) is the set of all 2 2 matrices A
with entries in C such that det A = 1. For any matrix A SL(2,C), since A has complex
entries, A has a Schur decomposition [1, p. 430]. This means that there exists a unitary
matrix B such that


a b
B 1 .
A = B
0 a1
We can now set a(t) to be a curve of non-zero complex numbers such that a(0) = 1 and
a(1) = a and similarly have b(t) be a curve with b(0) = 0 and b(1) = b. This means that


a(t) b(t)
A=B
B 1
0 a(t)1
is a curve of matrices from the identity to A in SL(2,C). So SL(2,C) is path-connected.
We will now describe a map from SL(2,C) onto SO+ (1,3). Recall that we are working in
the Minkowski space M and that Lorentz transformations can be applied to its elements.
For each x M , we identify it with a 2 2 complex matrix via

x0


x1
x0 + x3 x1 ix2 .
x2
x1 + ix2 x0 x3
x3
9

We can see that

 

x0 + x3 x1 ix2
x0 + x3 x1 ix2
=
x1 + ix2 x0 x3
x1 + ix2 x0 x3

so the matrix representing x is self-adjoint. Also, the determinant is the Minkowski norm
of x. This transformation allows us to take the product of elements of M with a 2 2
matrix.
Let A be a 2 2 matrix. Define a map
A : M M
by



x0 + x3 x1 ix2
A (x) = A
A .
x1 + ix2 x0 x3
Let A SL(2,C). Then

A=


a b
,
c d

where a, b, c, d C and det A = 1. So





a b x0 + x3 x1 ix2 a
c
A (x) =
c d x1 + ix2 x0 x3 b d


y0 + y3 y1 iy2
,
=
y1 + iy2 y0 y3
where

2y0 = x0 (||a||2 + ||b||2 + ||c||2 + ||d||2 ) + x1 (

ab + ab + cd + cd)
+ x3 (||a||2 ||b||2 + ||c||2 ||d||2 ),
+ix2 (
ab ab + cd cd)
2y1 = x0 (a
c+a
c + bd + bd) + x1 (ad + a
d + b
c + bc)
+ix2 (
ad + b
c ad bc) + x3 (a
c+a
c bd bd),
ix1 (ad + bc a
2y2 = ix0 (bd + a
c a
c bd)
d b
c)

+x2 (ad + a
d b
c bc) ix3 (
ac + bd a
c bd),
2y3 = x0 (||a||2 + ||b||2 ||c||2 ||d||2 ) + x1 (ab + a
b cd cd)
2
2
+ix2 (
ab + cd ab cd) + x3 (||a|| ||b|| ||c||2 + ||d||2 ).
This means that A x = A0 x where

||a||2 + ||b||2 + ||c||2 + ||d||2 a

b + ab + cd + cd

..
.
1
a
c+a
c + bd + bd
0

A =
.
2
..
2
2
||a|| + ||b|| ||c||2 ||d||2
...
10

||a||2 ||b||2 + ||c||2 ||d||2

..

.
.

2
2
2
2
|a|| ||b|| ||c|| + ||d||

One can check that A0 SO+ (1,3). So we have a map : SL(2,C) SO+ (1,3) defined by
(A) = A . To see that is a homomorphism, let A, B SL(2,C) and x M . Then
(AB)x = AB (x)
= ABx(AB)
= ABxB A
= A(BxB )A
= A (BxB )
= A B x
= (A)(B)x.
Lemma 3.1 The homomorphism is a surjective 2 1 map.
Because SL(2,C) is connected and is continuous, its image is connected. This, however,
does not prove that its image is all of SO+ (1,3). For that, it suffices to show that there are
elements in SL(2,C) which map to the six generators of SO+ (1,3) under . Here they are:

cosh sinh 0 0


sinh cosh 0 0
cosh 2 sinh 2
,

0
0
1 0
sinh 2 cosh 2
0
0
0 1

cosh

0
sinh

0


0
1
0
0
cosh 2
i sinh 2

sinh 0 cosh 0 ,
i sinh 2 cosh 2
0
0
0
1

cosh 0 0 sinh
"
#!
0
e2
0
1 0
0
,

2
0
0 1
0
0 e
sinh 0 0 cosh

1
0
0
0
" i
#!
0 cos sin 0
e2
0

=
i
0 sin cos 0 ,
0 e 2
0
0
0
1

11


1


0
cos( 2 ) sin( 2 )

=
0
sin( 2 ) cos( 2 )
0

1



cos( 2 ) i sin( 2 )
0

=
0
i sin( 2 ) cos( 2 )
0

0
0
0 sin
,
1
0
0 cos

0
0
0
1
0
0
.
0 cos sin
0 sin cos
0
cos
0
sin

Therefore the image of SL(2,C) under is all of SO+ (1,3). Notice how the angles in the
elements in SL(2,C) are doubled under the map to SO+ (1,3).
Since is a homomorphism, the identity of SL(2,C) maps to the identity of SO+ (1,3)
under it. It is also the case that

2 0 0 0
1 0 2 0 0
= I.
(I) =
2 0 0 2 0
0 0 0 2
To show that only I and I in SL(2,C) map to the identity, we set the diagonal entries in
A0 = A to 1. This yields the equations
||a||2 + ||b||2 + ||c||2 + ||d||2 = 2,
ad + a
d + b
c + bc = 2,
ad + a
d b
c bc = 2,
||a||2 ||b||2 ||c||2 + ||d||2 = 2.
This system has only two solutions, a = 1, b = 0, c = 0, d = 1 and a = 1, b = 0, c = 0,
d = 1. These are then the only elements which map to I.
Now, if A, B SL(2,C) and A = B , then
AB 1 = A B 1 = A 1
B =I
so
AB 1 = I
and therefore A = B. So every element in SO+ (1,3) has exactly two preimages, which
are negatives of each other. Therefore SL(2,C) double covers SO+ (1, 3)under . 
As SL(2,C) is the connected double cover of SO+ (1, 3), we can find the double cover of
the Poincare group, SL(2,C) n R1,3 .

12

Representation Theory

A representation [2, p. 3] of a group G in a vector space V is a homomorphism :

G GL(V ), where GL(V ) is the group of invertible linear transformations V V with
composition as the group operator. For a given group G, there may be many different
choices for V and . The regular representation for finite groups assigns each element of
G its own basis element ~eh in V, so : G Rn , where n = |G|. Define the representation
by (g)(~eh ) = ~egh .
Recall that the characteristic of a field is the smallest integer m such that m 1 = 0 if
it exists. If there is no such x, the characteristic of the field is 0. For both the complex
numbers and the real numbers, the characteristic is 0. A subspace W is called invariant
under if (G)(W ) W .
Any nontrivial vector spaces V has {~0} and V as invariant subspaces. It is easy to
check that these subspaces are invariant. A representation : G GL(V ) is irreducible if
the only invariant subspaces of V are {~0} and V . A useful tool for finding representations
is Maschkes Theorem[7, p. 320].
Theorem 4.1 (Maschkes Theorem) Let G be a finite group and let : G GL (V )
where V is finite dimensional, dim(V ) > 0, and V is over a field F such that char F = 0
or char F = p where p divides |G|. Then can be decomposed into the sum of one or more
irreducible representations.

4.1

Characters

A reference for this material is Representation Theory: A First Course[2].

Often, rather than looking at the representation itself, we instead look at the trace of
the representation. The trace of a matrix is the sum of the elements on the main diagonal.
We then define the character of to be : G C,
(g) = T r ((g)) .
Character have useful properties that makes working with them worthwhile.
1. (g) = (aga1 )
(a) is constant on each conjugacy class.
(b) is invariant under change of basis.
2. (g 1 ) = (g).
3. 1 2 (g) = 1 (g) + 2 (g).

13

(4)

For finite G, there is an inner product from C C C

h1 , 2 i =

1 X
1 (g)2 (g).
|G|

(5)

gG

If is an irreducible representation then h , i = 1, while if 1 and 2 are different

irreducible representations then h1 , 2 i = 0. Thus the characters of the irreducible
representations form an orthonormal basis of some function space. Using the properties of
character, we can show that the function space is the space of functions that map conjugates
to the same values. These functions are called the conjugacy class functions of G [2, p.
22].

Semidirect Products

Given two groups N and H such that N H = {e}, the semidirect product of N and H is
a group G. The result is that each g G
g = nh,
where n N and h H. Under the semidirect product N is normal in G and H is a
subgroup of G. The notation is G = H nN . Our focus is on the group created by N = R1,3
and H = SL(2, C).

5.1

Irreducible Representations of Semidirect Products

Let G = H n N . Recall that this means every element g of G can be uniquely written as
g = nh, where n N and h H. Now we will build the irreducible representations of G
from the irreducible representations of its normal subgroup, N . Since N is Abelian, each
conjugacy class has only one element [4, p. 68], and each irreducible representation can be
shown to be one dimensional in N .
Define N as the space of all one dimensional characters of N . Also define Nj as the
orbit of G acting on N containing j . Now follow these steps [9, p. 138]:
1. Break N into orbits under G and pick a representative character j from each orbit.
2. For each j , find the isotropy group Lj of j ; meaning the subgroup of H that fixes
j .
3. Choose an irreducible representation j of Lj and extend it to Gj = Lj n N by
j (hn)v = j (n)(h)v.
4. Create the vector bundle E, and let G act on (E).
5. Have Vj be identified with the -section concentrated on j such that Vj = ((E))j .
14

Once we have constructed the irreducible representations of G from the irreducible

representations of N . While some of the steps we havent totally explained, the method
will become clear when we apply it to the double cover of the Poincare group.

6
6.1

Representations of the Poincar

e Group
Poincar
e Group as Semi-Sirect Product

From here, we can describe the orbits of the double cover of the Poincare Group. Note
that every element in the double cover of the Poincare group acting on R1,3 preserves m2 .
Therefore each orbit is contained in a level set of m2 . It turns out that there are six families
of orbits. They have the following values for m2 and x0 . Note that there is not a single
orbit for all vectors with m2 > 0, x0 > 0 or m2 > 0, x0 < 0, but an infinite family. In fact,
there is a unique orbit for each value of m2 , given either x < 0 or x > 0. The six types
listed above correspond to the six families [9, p. 32].
The first two orbit types are those corresponding to the values m2 > 0, x0 > 0 and
m2 > 0, x0 < 0. To find the representative point of an orbit, we simply consider the general
form of a point in that orbit. In the case of m2 > 0, x0 > 0 or m2 > 0, x0 < 0, these orbits
have representative points of

m


0
m 0

0
0 m
0
with m 6= 0, because as can clearly be seen, the time coordinate is nonzero and real, so
m2 > 0. The orbits are represented by a two-sheet three-dimensional hyperboloid in R1,3 .
It is not at all difficult to see that a linear transformation that fixes the representative
point would look like

1 0
0
0
m
m
0 a22 a23 a24 0 0

0 a32 a33 a34 0 = 0

0 a42 a43 a44
0
0
Thus the matrix that fixes the representative point of our first two orbits is a member
of the rotation subgroup of the connected double cover of the proper Lorentz group. This
double cover is SU(2).
Now we consider the isotropy groups of the third orbit. This orbit corresponds to the
values m2 = 0, x0 > 0. This is the forward looking light cone (which is three-dimensional)
in R1,3 respectively.

15


T
The representative points of this orbit will be 0 0 where 6= 0. We use this
point because the value m2 for this vector in R1,3 is 0, and because the time coordinate of
this vector, namely is not 0.
In the case of these two orbits, however, there are two types of linear transformations
that keep this representative point fixed. Rotations about the y-axis will fix the point:

1
0
0
0
0 cos 0 sin 0 0
.
=

0
0
1
0
0
0
0 sin 0 cos
By similar argument, we see the fourth orbit is also fixed by transformations of this form.

T
Now consider just the third orbits representative point 0 0 , then the transformation

1 + (u2 + v 2 )/2 v (u2 + v 2 )/2 u

v
1
v
0
SL(2, C)
T (u, v) =
2
2
(u2 + v 2 )/2
v 1 (u + v )/2 u
u
0
u
1
will also leave the point fixed as shown by the following calculations.
1 + (u2 + v 2 )/2 v (u2 + v 2 )/2

v
1
v

(u2 + v 2 )/2
v 1 (u2 + v 2 )/2
u
0
u

2
2
+ (u + v )/2 (u2 + v 2 )/2

vv
=
+ (u2 + v 2 )/2 (u2 + v 2 )/2
uu

=
.
0

u
0
0

u
0
1

Thus T (u, v) also leaves the fourth orbits representative point invariant [5, p. 58].
Let D(u, v, ) = T (u, v)R(). Then any transformation D(u, v, ) will fix the representative point of the light cones as well. Note, then, that any such translation can be
written as a composition of a translation by some 2-vector with a rotation of some angle.
In other words, the transformation D(u, v, ) is an arbitrary transformation from the isometry group of R2 , namely E(2). This is the isotropy group for the third and fourth orbits
of the Poincare group.
16

As a result of these previous transformations we do not end up with the Euclidean

group E(2), but its connected double cover. This double cover is E(2).
It is isomorphic to

E(2), except that rotations by in E(2) correspond to rotations of 2 in E(2).

2
The other two orbits of SU(2) correspond to the values of m < 0 and m2 = p0 = 0.
Their isotropy groups are SL(2, R) and SL(2, C) respectively, but we will not prove these
facts. These two orbits are not as relevant to the study of physics as the first four are,
and as such will not be studied as much in this paper. This is because the m2 < 0 orbit
corresponds to tachyons, which are faster than light particles. These have never been
observed and only recently have been theorized. The m2 = p0 = 0 orbit corresponds to
vacuum, and as such is entirely uninteresting to study [9].

6.2

The Representations of the Isotropy Groups

Let us consider the isotropy group of a representative point in one of the light cone orbits.

Recall that it is the double cover, E(2),

of the set of all Euclidean transformations of the

plane. It is clear to see from the way we constructed this isotropy group that E(2)
is, itself,
a semi-direct product of the connected double cover of SO(2) and the set of translations R2 .
So we can find the groups representations using the fact that it is a semi-direct product.
The orbits of SO(2) acting on R2 are circles, and the one-point orbit consisting of the
origin. It turns out that representations corresponding to the circle (or its connected double
cover) orbit do not occur in physics [9, p. 147], so we will restrict our focus to the origin.
The isotropy group of {0} is exactly SO(2), the set of rotations of the plane.
Theorem 6.1 The irreducible representations of the double cover of SO(2) are classified
by
1
3
s = 0, , 1, , . . . , where n Z
2
2
The irreducible representations of SO(2)s double cover are all one-dimensional, as the
group is abelian. The representations are given by
ein , n Z
These representations correspond to rotations of S1 . It must be the case that n Z as,
in order for the representation, which can be thought of as a map : S1 S1 to be
continuous, the function needs to map the circle to itself some integer number of times.
However, in this representation, corresponds to a rotation by 2 as given by the double

covering nature of E(2).

Thus, it is conventional in physics to let
1
s= n
2
so that the irreducible representations are classified by
3
1
s = 0, , 1, , . . . . 
2
2
17

Unfortunately, finding the irreducible representations of SU(2) is more difficult.

Haar measure

In order to complete some of the work that follows, we will need to integrate a function f
over a Lie group G. To that end, we will define an invariant integral.
A function is compactly supported if it is identically zero outside some compact set. Let
F0 (G) be the space of compactly supported continuous functions on G. Then integration
da is left invariant if
Z
Z
f (a)da
(6)
(r(b)f )(a)da =
G

for all f F0 (G) and b G. Here r is the regular representation of G on F0 (G), namely
r(b)f (a) = f (b1 a).
If G is a compact group, any function on a compact group is compactly supported. Furthermore, the invariant integral is unique up to scalar multiples and the total volume of
the group is finite [9, p. 174]. Then we can fix an integral by simply declaring
Z
1 da = 1.
G

This integral is known as the Haar measure.

It remains to find an invariant integral on SU(2) and normalize it.

7.1

Differential forms

A differential k-form on a Lie group G is an alternating k-tensor, which can be used in a

k-dimensional integral [8, p. 88]. A 0-form on G is the same as a function f : G K. We
will not go into the details of differential forms in this paper.
Differential forms are combined using the alternating wedge operator
(k-form) (l-form) (k + l)-form
= .
We will occasionally drop the symbol when its use is clearly implied.
If : R S is differentiable and is a differential form on S, then the pullback
is a differential form on R. Furthermore, given certain assumptions we have
Z
Z

=
.
R

18

(The necessary assumptions are that is a diffeomorphism, is an n-form where n = dim S,

R and S are oriented, and is orientation-preserving. For the specific uses that follow, all
of these are true.)
If does not have compact support, this integral might not be defined. However, we
can multiply it by a compactly supported function f to get
Z
Z

f .
(7)
f =
S

In order to find a Haar measure, we are going to set R = S = G. Now the pullback
is not moving the differential form from one group to another, but it might still move
it around within our group G. Choose some b G and define b : G G to be left
multiplication by b1 . For the 0-form f , we find that f = r(b)f [9, p. 174]. Now we can
simplify (7) to
Z
Z
r(b)f b =
f .
(8)
G

Our task is to find an such that b = for all b G. If we can do that, then (8)
simplifies to (6), aside from changes in notation. Suppose we have a representation A of G
on Ck :
A : G GL(k, C).
We can consider A as k 2 component functions Aij : G C, which are also 0-forms on G.
Then define dA component-wise to get a k k matrix of 1-forms.
Theorem 7.1 Each entry of A1 dA is a left invariant differential form.
Because A is a homomorphism,
b A(g) = A(b g)
= A(b1 g)
= A(b)1 A(g)
and
b A = B 1 A
where B = A(b). Furthermore, because B is constant,
b (dA) = d(b A)
= d(B 1 A)
= B 1 dA.
19

Now it is a simple computation to show that A1 dA is left invariant:

b (A1 dA) = b (A1 )b (dA)
= b (A1 )B 1 dA
= b (A)1 B 1 dA
= (B 1 A)1 B 1 dA
= A1 BB 1 dA
= A1 dA.

The n-form that we want is a wedge product of n of these 1-forms. Since any such n-form
is an invariant integral, and there is only one such integral up to scalar multiples, any n
choices will give us something that can be normalized to the Haar measure.

7.2

Haar measure of SU(2)

Any element of SU(2) is of the form


A =

where ||2 + ||2 = 1. So

A



d
d
dA =
d
d


d
+ d d + d
.
=
+
+
d
d d
d


By theorem 7.1, each entry is a left invariant form. From here, we choose to take the wedge
product of the second, third and fourth entries:

(
d + d)
(d + d) (d
d)

= (
)d
d (d
d)

= d d (d
d).
In order to simplify this, we can express d in terms of the other forms by differentiating:

+ = 1

= d
+
d + d + d
= 0

= d = 1 (d
+
d + d).

20

Remember that dx dx = 0, so two of the terms in d will cancel out. The invariant
3-form is now

= d d (d

d d (d
d)
( 1 (d
+
d + d)))
+
= d d (d
1 d
)
+
= d d ((
) 1 d
)
= d d ( 1 d
)
= 1 d d d
.
Thus 1 d d d
is a left invariant form.
It will be useful later to express this in spherical coordinates. Let
= cos + i sin cos ,
= sin sin ei ,
with 0 , 0 , and 0 2. This parameterizes SU(2); we will use these
coordinates later to simplify the Haar measure to a single variable.
Let u and v be real functions. Then
d(u + iv) d(u iv) = 2i du dv.
Therefore
d d
= 2i d(cos ) d(sin cos )
=

2i sin d (sin d(cos ) + cos cos d)

2i sin d sin d(cos )

2i sin d ( sin sin d)

= 2i sin2 sin d d.
When expanding d we can safely ignore any d and d terms, since they will drop out

21

when wedged with d d

. Then
1
1
d d d
= d d
d

1
=
2i sin2 sin d d d

1
=
2i sin2 sin d d d(sin sin ei )

1
=
2i sin2 sin d d (i sin sin ei )d

1
=
2i sin2 sin d d id

1
2i(i) sin2 sin d d d
=

= 2 sin2 sin d d d.
So sin2 sin d d d is a left invariant 3-form on SU(2).
We want the total volume of the group to be 1. We can integrate in the normal way:
Z Z Z
Z 2 Z Z
2
sin sin d d d =
sin2 sin d d d
SU(2)
0
0
0
Z Z
= 2
sin2 sin d d
0
0
Z
= 4
sin2 d
0

= 2 2 .
Since we can multiply this by any constant, we end up with the Haar measure
=

1
sin2 sin d d d.
2 2

The irreducible representations of SU(2)

In order to find the representations of SU(2), we will show that each finite-dimensional
irreducible representation of a topological group occurs as a subrepresentation of the regular
representation of the group on the space of continuous functions of that group.
Let G be a topological group and V be a finite dimensional irreducible Hilbert space.
We say (, V ) is a subrepresentation of (, W ) if there exists an injective map : V W
such that for all b in G, (b)(x) = ((b)x).

22

Let : G Hom(V, V ) be any representation. For any V , representation , ` V ,

x V , let a G, let x : G C be
x (a) = `((a)1 x).
For any V , we will define the topology on Hom(V, V ) to be the weakest one such that for
all representations , for all ` V , for all x V , the map x : G C is continuous.
Recall that F(G) is the space of continuous functions on G and there exists a regular
representation of G on F(G). Then we have the map : V F(G) where for x V ,
(x) = x .
The regular representation acts on x F(G) by
r(b)(x)(a) = (x)(b1 a).
It is simple to show that r is well defined and a homomorphism. As (x) F(G),
a space of continuous functions on G, it is continuous on G. Since G is a topological
group, multiplication by an element in G is continuous. This means that r(b)(x) is the
composition of continuous functions and is therefore continuous, so it is in F(G).
Also, we can see that r(b) : F(G) F(G) is a linear map for any b G. Let `1 , `2 V
and x V . Define 1 (x) = `1 ((a)1 x) and 2 (x) = `2 ((a)1 x). Then
r(b)(1 (x) + 2 (x))(a) = (1 (x) + 2 (x))(b1 a)
= r(b)(1 (x) + 2 (x))(a)
= 1 (x)(b1 a) + 2 (x)(b1 a)
= r(b)1 (x)(a) + r(b)2 (x)(a).
For some constant k,
r(b)(k(x)) = k(x) b1 = k(r(b)(x))
so r is well-defined.
We can also see that r is a homomorphism:
r(ab)(x) = (x) (ab)1
= (x) b1 a1
= r(a)((x) b1 )
= r(a)(r(b)(x)).
Theorem 8.1 Any finite-dimensional irreducible representation of G occurs as a subrepresentation of r. In other words, let (, V ) be a finite dimensional irreducible representation
of G. Then there exists a subspace W F(G) with an injective linear transformation
: V W such that for all b G, r(b)(x) = ((b)x).
23

First we will show is a G-morphism, meaning r(b)(x) = ((b)x). Let x V and

a, b G. Then
((b)x)(a) = `((a)1 (b)x)
by the definition of . Also
r(b)(x)(a) = (a)(b1 a)
= `((b1 a)1 x)
= `(((b1 )(a))1 x)
= `((a)1 (b)x).
So ((b)x)(a) = r(b)(x)(a).
Now we will show that is injective. We say V is topologically irreducible if the only
closed invariant subspaces of V are V itself and {0}. We will take for granted that since
V is finite dimensional and irreducible, it is also topologically irreducible [9, p. 178]. We
can show that ker is a closed invariant subspace of V . It is fixed by the action of G on
V , namely . Let x ker , then
((b)(x)) = r(b)(x) = r(b)0 = 0.
So (b)x is also in ker .
To show that is injective, let ` V \{0}. Then there exists x V such that `(x) 6= 0.
This means that
(x)I = l((I)1 x) = l(x) 6= 0.
Therefore if ` 6= 0, then 6= 0 Hom(V, F(G)). So then ker = {0} as it cannot be V .
We now have a finite dimensional irreducible representation of G on V , , with an
injective map : V W F(G) and for all b G, r(b)(x) = ((b)x). So every finite
dimensional irreducible representation of G occurs as a subrepresentation of r on F(G). 
If we define the scalar product on F(G) by
Z
(f1 , f2 )G = (f1 (a)f2 (a)
G

then we make F(G) into a pre-Hilbert space. Notice that this uses the Haar measure
discussed in the previous section. This allows us to obtain a Hilbert space called L2 (G).
The regular representation is now a unitary representation of G on L2 (G) [9, p. 179].
Having now created L2 (G), take G = SU(2), which is a compact group. The Peter-Weyl
theorem says the space L2 (G) decomposes into a Hilbert space direct sum of irreducible
representations of G, each of which is finite dimensional. This result is similar to Maschkes
Theorem, as any representation in L2 (G) can be decomposed into a direct sum of irreducible

24

representations. An important consequence of this theorem is that the irreducible characters form an orthonormal basis of the Hilbert space of square integrable central functions
[9, p. 179]. This is exactly the same statement we had for finite G, only then we were
considering conjugacy class functions. But this requires an inner product, which is the
inner product on F(G) derived from the Haar measure. To find these representations, we
use the Peter-Weyl theorem and the fact that irreducible characters have unit length under
the Haar inner product. From here, our task is to determine the form of the possibilities
for the characters of SU(2).
Let (, V ) be an irreducible representation of SU(2). Then


ei
0
0 ei

is a unitary matrix. This is because a representation of SU(2) must preserve the operation
of SU(2). Thus any element to which the representation maps decomposes to be a direct
sum of eigenspaces. Using this fact, we can then define a basis v1 , . . . , vn of V such that
for j = 1, . . . , n

 i
e
0
vj = j ()vj

0 ei
for some function j .
Because a representation is a homomorphism, we have
j (1 + 2 ) = j (1 )j (2 )
and so
j = eicj
for some cj . What this means is that if we consider the representations character , we
have
X
() =
eicj .
Now our problem is reduced to finding what the possibilities for cj are. By definition,
() is a continuous map. In order for this to be the case, cj must be an integer. Two
more facts will help us determine the possibilities for cj . First off, we know that, since is
an even function of , if cj occurs, then cj must occur as well. Secondly, we know that,
as was stated earlier, h, i = 1. So
2
1 = h, i =

()() sin2 ()d

25

1
=

()() sin2 ()d.

Using the identity

1
1
sin2 () = (1 cos(2)) = (2 e2i e2i )
2
4
we have
n
n Z
1 1 X X 2 icj icj
e
e
(2 e2i e2i )d.
1 = h, i =
4
0

(9)

j=1 k=1

Notice that
Z

emi d = 0

for integers m 6= 0; if m = 0, the integral is equal to 2. So in solving for (9), we mulitply

out and keep only the terms that multiply to 1 while discarding the ones that multiply to
zero. With this logic in mind, we expand (9) to get

h, i =

1
4

X
n X
n  Z
2
j=1 k=1

e(cj ck )i d +

Z
0

e(cj ck +2)i d +


e(cj ck 2)i d .(10)

So we have n terms in which j = k, each of which yields a term of the form 14 1 2 2 = 1.

Therefore the integral (9) equals 1 on at least n occasions. Furthermore, we have 2a terms
where j 6= k where cj = ck yielding 2a occasions where (9) equals 1 for some unknown
quantity a. There are also 2b terms in which cj = ck + 2 for some unknown number b,
which yields as a solution for (9) 41 1 2 (1) 2b times. Note 14 1 2 (1) 2b = b.
So h, i = n + 2a b. It is the case that b can at most be n 1. Therefore,
n b + 2a > 1 + 2a. However, since h, i = 1, it must be the case that a = 0 and
b = n 1.
Thus we have found all of the possibilities for the cj s. For each n, {cj } has the properties
that the cj s are symmetric about the origin and that if cj {cj }, then cj {cj }.
So, for a given n, {cj } = {(n + 1, n + 3, . . . , n 3, n 1}. It is the convention to label
the characters and representations, which we will prove exist shortly, by s = (n 1)/2.

26

Theorem 8.2 The irreducible characters of SU(2) are the following

0
1/2
1

= 1,
= ei + ei ,
= e2i + 1 + e2i ,

3/2 = e3i + ei + ei + e3i ,

..
.
s

= e2si + e(22s)i + + e(2s2)i + e2si , . . . .

It is not difficult to find the representations explicitly. The group SU(2) acts on C2 .
Thus the group has representations in the space of functions on C2 . Let Vs be the space
of all homogeneous polynomials of degree 2s. A basis for Vs is


z12s , z12s1 z2 , z12s2 z22 , . . . , z22s .

To develop a method for computing these irreducible representations, we will first find
the trivial representation 0 .
Let A SU(2). Thus A is of the form


a b
A=
,
b a
and


a b
b a

2
2
where |a|
 + |b|
 = 1. So we can treat |a| as cos() and |b| as sin() for some .
z1
Let
C2 . Consider
z2
1

z1
z2


=

az1 + bz2
bz1 + az2


.

Note that the basis for V0 is {1}, which can be thought of as {z10 }. To find the matrix
representation, we simply plug az1 + bz2 into z10 , which is the first element of V0 s basis.
For the s = 0 case, we end up with the matrix
0 = [1] .

27

For another example, consider s = 12 . The basis for V1 is {z1 , z2 }. Replacing the basis
elements z1 and z2 with az1 + bz2 and bz1 + az2 yields the polynomials
az1 + bz2
bz1 + az2 .
Note that these two polynomials are linear transformations of the basis elements z1 and
z2 . The coordinates of the polynomials are the rows of the representation matrix. So the
matrix for this representation is exactly


a b
1/2 =
.
b a
These first two cases seemed a bit basic. So let us now compute 1 . The basis for V2 is
{z12 , z1 z2 , z22 }. Using the same algorithm used in computing the previous representations,
we find that

2
2
b
a
ba
1 = 2ab aa bb 2ab .
b2
ab
a2
Note that the trace of this matrix, after the a and b values are replaced with their respective
cos() and sin() values, is
e2i + 1 + e2i = 1 .
Using this method and solving for the traces of the representations we get does, in fact,
confirm that there exists a representation of trace
s

= e2si + e(22s)i + + e(2s2)i + e2si

for any value of s.

Thus, we have found the irreducible representations of SU(2). 
Having found these representations, we now have found the irreducible representations
of the double cover of the Poincare group which correspond to particles and. Here are a
few examples of fundamental particles that with spin values that correspond to spin values
found using the method developed in this paper.
Spin
s=0
s = 1/2
s=1
s = 3/2
s=2

m2 > 0
Higgs boson
neutrino, electron, quark
W boson, Z boson
baryon, gravitino

m2 = 0

photon, gluon
graviton

28

References
[1] J. Fraleigh and R. Beauregard, Linear Algebra, Second Edition, Addison-Wesley Publishing Company, 1990.
[2] W. Fulton and J. Harris, Representation Theory: A First Course, Springer-Verlag,
1991.
[3] David J. Gross Physics Today 46 - 50, (1995).
[4] I.N Hernstein, Abstract Algebra, Second Edition, Macmillan Publishing Company,
1990.
[5] Y.S. Kim and M.E. Noz, Theory and Applications of the Poincare Group, D. Reidel
Publishing Company, 1986.
[6] M. Reid and B. Szendr
oi, Geometry and Topology, Cambridge University Press, 2005.
[7] W.R. Scott, Group Theory Prentice-Hall Inc., 1964.
[8] Michael Spivak, Calculus on Manifolds, Westview Press, 1998.
[9] S. Sternberg, Group theory and physics, Cambridge University Press, 1994.

29