Group Theory and Relativity: 8.1 What Is A Group?

8
Group theory and relativity
In previous chapters we have encountered frequently the concept of symmetry of

physical laws and symmetry of physical systems under mathematical or physical
transformations. The primary examples of symmetry transformations that we have
encountered were spatial rotations and Lorentz transformations. The concept of
symmetry is very important in physics, not only in the context of transformations
of space and time, but also in many others. As it happens, there is a beautiful branch
of mathematics, called group theory, which is ideally suited to the description of
symmetry. The application of group theory to physics, especially in the context of
quantum mechanics, was worked out by Wigner, Weyl, and others in the 1920s
and 1930s. By now it is part of the standard toolkit of an educated physicist. This
chapter will introduce basic group theory with an emphasis on the application
to the descriptions of rotational symmetry and Lorentz symmetry, which are the
examples of greatest relevance here.
8.1 What is a group?

Rotations and Lorentz transformations are examples of transformations that one
can imagine applying to physical objects, collections of objects, systems of equa-
tions (the Maxwell equations, for example), or even the entire Universe. When
they are symmetries of the system in question, as in the case of the Maxwell equa-
tions, the transformed system is identical to the original one. However, it is also
possible to transform a system that does not possess the symmetry, in which case
the transformed system looks different from the original one. The set of all trans-
formations that do describe symmetries form a group (which we will define later)
that is a characteristic property of the system in question. The more symmetry the
system possesses, the larger the symmetry group, and the more powerful are the
constraints implied by it.
260
Downloaded from https://www.cambridge.org/core. Teachers College Library - Columbia University, on 21 Aug 2018 at 14:15:37, subject to the
Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9780511755811.009
What is a group? 261
Let us begin by explaining the notion of a transformation group. For this

purpose, suppose that T1 and T2 denote two transformations (such as rotations,
Lorentz transformations, translations, etc.). Now consider first applying the trans-
formation T1 and then (to the resulting system) applying the transformation T2 . The
combined effect is a new composite transformation, which is denoted T2 T1 . Note
the convention that the transformation that is applied first is written to the right.
The sequential application of the transformations is represented mathematically
by juxtaposition or “multiplication.” Of course, this is not ordinary multiplication,
because T1 and T2 are not numbers. Rather, they are some sort of operators, and
this is an operator multiplication. In fact, it often happens that the multiplication is
noncommutative, which means that T2 T1 = T1 T2 .
If one starts with an arbitrary set of transformations and then enlarges the set
by including the transformations that undo them (inverse transformations) as well
as all transformations that can be obtained by multiplying these transformations
any number of times, eventually one reaches a closed algebraic system that is a
group. Group theory, which is a very mature branch of mathematics, is helpful for
analyzing the implications of symmetry. It is useful for classical systems, which
is the main concern in this book, but it really comes into its own in the context of
quantum mechanics (which was introduced in Chapter 7).
Simple examples of groups

Symmetrical objects or systems are familiar from everyday experience. For exam-
ple, a perfectly round ball has spherical symmetry. This means that it is invariant
under rotations about any axis that passes through the center of the ball. These
rotations form a group, called the group of rotations in three dimensions, denoted
by the symbol S O(3). (We will elaborate on this notation later.) An even simpler
example is the rotational symmetry of a circle. Rotations in the plane of the circle,
that leave the center fixed, are symmetry transformations that form the group of
rotations in two dimensions and are denoted S O(2). This group was discussed in
some detail in Chapter 1.
The notion of rotational symmetry generalizes to an arbitrary (positive integer)
number of dimensions. So, as you may have guessed, the group of rotations in D
dimensions is denoted S O(D). This group describes the symmetry of a sphere. For
example, the unit sphere in D dimensions is described in Euclidean coordinates by
the equation (x 1 )2 + (x 2 )2 + · · · + (x D )2 = 1. Such a sphere has D − 1 dimen-
sions. The fact that it is described by one equation means that it has one fewer
dimensions than the space in which it resides. The mathematical notation for such
a sphere is S D−1 .
262 Group theory and relativity
The spheres discussed in the preceding paragraphs have additional symmetries

beyond rotations. These are reflections. Since the reflections can be carried out in
an infinite number of different planes that pass through the center of the sphere,1
there are an infinite number of them. However, reflections about two different
planes are related by a rotation. Altogether, the complete symmetry of a sphere
is given by a group that contains reflections as well as rotations. In the case of D
dimensions, this group is denoted O(D). Clearly, this group contains the group
S O(D) as a subset, which is obtained by discarding the reflections. Whenever one
group is a subset of another group, it is called a subgroup.
As another example, let us consider the symmetries of the Euclidean plane, de-
scribed by Cartesian coordinates x and y (or x 1 and x 2 ). This plane is mapped
into itself by rotations by an arbitrary angle about any point in the plane. Also, it
is mapped to itself by a rigid shift (or translation) by an arbitrary amount in the x
or y directions, or by a combination of both. The set of all possible rotations and
translations of the plane form a group, called the Euclidean group in two dimen-
sions. It can be denoted E(2). There is a straightforward generalization to a space
of D Euclidean dimensions, for which the symmetry group is the Euclidean group
E(D).
The examples provided so far are all infinite groups. This means that there are
an infinite number of different symmetry transformations. For example, the rota-
tion angle is a continuous parameter that can take an arbitrary value (between 0
and 2π ). There are also examples of systems for which the number of symmetry
transformations is finite. For example, consider a regular polygon in the plane with
n sides. Such an object is invariant under rotations about the center by an angle that
is an integer multiple of 2π/n. Altogether, there are n inequivalent possibilities. So
this symmetry group, which is sometimes denoted Z n and sometimes denoted An ,
has n distinct elements. If one also adjoins the reflection symmetries one obtains a
group with 2n elements denoted Dn .
The four properties that define a group

Now that we have explained the concept of a group, we can give the formal defini-
tion. First of all, a group is a set. More specifically, at least in physical applications,
it is a set of transformations, which often correspond to symmetries of a physical
system. This set can be either finite or infinite, as we illustrated above. When four
additional properties, described below, are satisfied the set becomes a group. The
group is called finite or infinite according to whether the set is finite or infinite.
1 We imagine here that the planes have D − 1 dimensions. A reflection in a plane maps a vector to a reflected
vector by reversing the sign of the part of the vector that is perpendicular to the plane.
Let us list the four basic defining properties of a group first, and then following
that we will explore them in greater detail.
• A group has a binary composition rule (often called group multiplication) that
assigns an element of the group to each ordered pair of elements of the group.
• Group multiplication is required to be associative.
• A group contains a unique identity element.
• Every element of a group has a unique inverse.
The first property (or axiom) that turns a set into a group is that there is a rule
for combining any two elements of the group that gives a third element of the
group. Thus, if a and b are any two elements of a group G, then the group prod-
uct ab is also an element of G. Here we have represented the group composition
rule as multiplication. This is usually a convenient convention, though sometimes
it is more natural to represent it by addition. This property is often referred to as
closure. This simply means the set G is closed under group multiplication. Math-
ematically, group multiplication can be viewed as a map from G × G to G.
The group composition rule is not required to be commutative. This means that
we do not require that ab and ba are the same elements of G. The special class
of groups for which ab = ba for every pair of elements are called Abelian groups.
Those for which this is not the case are called nonabelian groups. (The word de-
rives the name of the mathematician Abel.) The addition notation for group com-
position is often used for abelian groups, that is, a + b = b + a.
The second defining property of a group is associativity. This is the requirement
that for any three elements a, b, c of a group G, (ab)c = a(bc). The parentheses
serve to indicate the order in which the group multiplications are carried out. Thus,
to be very explicit, (ab)c means that the multiplication ab should be carried out
first, and the resulting element of the group should then multiply c from the left.
A consequence of associativity is that the parentheses are unnecessary. One can
simply write abc. The meaning is unambiguous, because either of the two possible
interpretations gives the same result.
Multiplication of square n × n matrices is also associative and need not be com-
mutative. This fact at least raises the possibility that the multiplication table for a
group G can be realized by appropriately chosen matrices. Certainly, if we had a
multiplication rule that was not associative, we would know that this was not possi-
ble. The representation of groups by sets of matrices is an important issue to which
we will return later. As an aside, let us remark that nonassociative mathematical
systems are sometimes considered by mathematicians (and even by physicists), but
they are certainly not groups.
The third defining property of a group is the existence of an identity element.
This means that every group G contains a unique element, often called e, with
the property that ea = ae = a for every element a ∈ G. Note that e is both a left
identity and a right identity. Another possible notation for the identity element
is 1, but using the more abstract symbol e serves to emphasize that we are not
dealing with ordinary numbers. In an additive notation for group composition, it
is natural to represent the identity element by 0, though a more abstract symbol
could again be used. Thus, for an additive group, 0 + a = a + 0 = a. In terms of
transformations of a physical system, the identity transformation is the trivial one
that leaves the system untouched.
The final defining property of a group is the existence of an inverse. Specifically,
for every element a ∈ G, we require the existence of another element of G, de-
noted a −1 , with the property that aa −1 = a −1 a = e. In terms of transformations,
a −1 is the transformation that undoes the transformation a. Note that the inverse
of a −1 is a itself, since a satisfies the defining equations for such an inverse. For
an Abelian group that is described by an additive notation, the natural notation for
the inverse of a is −a.
We have now described all four of the defining properties of a group. Any set
that has a composition rule with these properties is a group. A group might satisfy
additional properties that turns it into an example of a mathematical category with
additional structure. However, in such a case, it is still a group.
Group representations
In the preceding we have given the abstract definition of a group. As physicists,
we are mainly concerned with specific physical realizations of groups, where they
act as transformations of specific physical systems. Such systems usually are de-
scribed in some coordinate basis, and so the effect of the transformation is to make
a linear change of the coordinates. This can be described as multiplication by a ma-
trix. Thus, it is natural from a physical point of view to represent group elements
by matrices. This also turns out to be useful for understanding the mathematical
structure of groups.
By definition, a group representation is an explicit realization of the group multi-
plication rule in terms of matrices made out of ordinary numbers (real or complex).
Let us consider square matrices with n rows and n columns. The integer n is called
the dimension of the group representation. In the special case of a one-dimensional
representation, the group elements are represented by ordinary numbers, which can
be viewed as matrices with just one row and one column. Given a group G, sup-
pose that we have a rule for assigning a matrix to every element of the group.
Specifically, to each element a ∈ G we assign an n × n matrix D(a). More explic-
itly, D(a) consists of n 2 numbers Di j (a), where i, j = 1, 2, . . . , n label the rows
and columns.
The matrices D(a) are not completely arbitrary. Rather, they are supposed to
be chosen in such a way as to capture the group multiplication rule. This means
that multiplying the matrices should correspond to multiplying the group elements.
Expressed as an equation, the rule is
D(a)D(b) = D(ab). (8.1)
Displaying the row and column labels, this equation takes the form

n
Dik (a)Dk j (b) = Di j (ab). (8.2)
k =1
Such a rule for assigning matrices to group elements is called a group representa-
tion. More specifically, when the matrices have n rows and columns, it is called an
n-dimensional representation.
There are a couple of simple facts about group representations that should be
noted. First of all, since D(e)D(a) = D(ea) = D(a), the identity element is rep-
resented by the unit matrix I :
Di j (e) = δi j = Ii j i, j = 1, 2, . . . , n. (8.3)
Similarly, since D(a −1 )D(a) = D(a −1 a) = D(e) = I , we deduce that the inverse
element is represented by the inverse matrix
D(a −1 ) = [D(a)]− 1 . (8.4)
Also, since multiplication of finite-dimensional matrices is associative, the asso-

ciativity rule is essential for the existence of a matrix representation.
Representations are allowed to be infinite-dimensional. For such representations
the matrices have an infinite number of rows and columns. Although they do exist,
there is no point in considering them if the group itself is finite. The reason is
that all the essential structure of the group is captured by a certain collection of
finite-dimensional representations. However, for certain types of infinite groups,
they can be important. Then associativity is a nontrivial requirement.
The defining equation D(a)D(b) = D(ab) maps one mathematical system
(a group) into another one (a collection of matrices) so as to preserve its defining
structure (the group composition rule). This is an example of what mathematicians
call a homomorphism. It should be emphasized that there need not be a one-to-one
correspondence between the group elements and the matrices. D(a) = D(b) need
not imply that a = b. For some group representations this is true and for others
it is not. When there is a one-to-one correspondence, the group representation is
called faithful. In this case, the homomorphism is also an isomorphism.
There are families of representations that, in a sense to be defined, are equiva-

lent. So, in classifying representations of a group, one would only want to count
this family once. Put differently, we could choose any convenient representative of
this family. To be explicit, let a → D(a) be an N -dimensional representation of
a group G. Then another N -dimensional representation a → D̃(a) is considered
equivalent to the first one if and only if there exists a nonsingular (that is, invert-
ible) N × N matrix A such that D̃(a) = A−1 D(a)A for all a ∈ G. This definition
makes sense because
D̃(a) D̃(b) = A−1 D(a)A A−1 D(b)A
= A−1 D(a)D(b)A = A−1 D(ab)A = D̃(ab). (8.5)
A representation D is called unitary if all the representation matrices D(a) are
unitary matrices.2 In general, group representations need not be unitary. However,
for all finite groups, and for large classes of infinite ones, it turns out that every
representation is equivalent to a unitary one.
Suppose we are given two representations D (1) and D (2) of a group G, having
dimensions N1 and N2 , respectively. Then we can form a new representation D of
dimension N1 + N2 as a direct sum as follows.
(1)
D (a) 0
D(a) = . (8.6)
0 D (2) (a)
Here we have written an (N1 + N2 ) × (N1 + N2 ) matrix in block form. These

matrices can be verified easily to satisfy the definition of a representation. A
representation of this type is called reducible. Clearly, it does not add knowl-
edge about the group to what we already had from knowing D (1) and D (2) . So
the important representations are those that are irreducible. By definition, they are
not of this block form, nor are they equivalent to a representation of this block
form.
To conclude this brief introduction to group representations, the representa-
tions that are usually of greatest interest are the unitary irreducible representa-
tions. For finite groups and a large class of infinite ones that is pretty much
the whole story. However, for certain types of infinite groups (noncompact Lie
groups) the story is more complicated. Rotation groups are of the simpler (com-
pact) type, but the Lorentz group is of the more complicated (noncompact)
type.
2 Let us recall the definitions. The adjoint of a matrix M, denoted M † , is the transpose and complex conju-
gate, that is, (M † )i j = (M ji )∗ . A unitary matrix is one for which the adjoint is the inverse, that is, M M † =
M† M = I .
Finite and infinite groups 267
8.2 Finite and infinite groups

Permutation groups
The groups of greatest relevance to this book are groups containing rotations and
Lorentz transformations. However, finite groups are also important – and the gen-
eral theory is somewhat simpler. They arise, for example, in characterizing the
symmetries of crystals. An important class of finite groups are groups of permu-
tations. In fact, every finite group can be realized as a subgroup of a permutation
group. The group of permutations of N objects is usually called the symmetric
group S N . The idea is that given N objects, in a definite order, we can consider
changing (or permuting) the order of the objects. Each such reordering is a group
operation. The number of distinct elements in the group S N is the same as the
number of distinct orderings, which is N !.
There are various ways of describing a specific permutation, and we won’t ex-
plore every one of them here. One possibility is to write the integers 1 through N
in some order inside square brackets as follows [ p1 p2 . . . p N ]. The idea is that the
permutation p takes the first object and puts it in the p1 th place, it puts the second
object in the p2 th place, and so forth. It is essential that all the pi s are distinct, so
that each of the N integers occurs exactly once. The identity element in this nota-
tion is [12 . . . N ]. The inverse permutation p −1 simply undoes the original one. So
the object in the p1 th place is moved to the first place, and so forth.
A special case of a permutation is one that simply interchanges two of the
objects and leaves the other N − 2 unaffected. Such simple interchanges gen-
erate the entire group. This means that by making a suitable sequence of inter-
changes one can build up an arbitrary permutation. The way one does this is not
unique, of course. However, we can divide the group of permutations into two
sets, called the even permutations and the odd permutations. Even permutations
are built up by an even number of interchanges and odd permutations are built
up by an odd number of interchanges. Despite the non-uniqueness pointed out a
moment ago, this is a meaningful distinction. Even permutations can only be built
up from an even number of interchanges, and odd ones can only be built up by
an odd number of interchanges. The identity element is an even permutation, of
course.
The set of even permutations consists of exactly half the elements of the group
(for N > 1), and the odd permutations are the other half. For example, S3 has
six elements of which three are even and three are odd. The even permutations
form a subgroup, called the alternating group. This is obvious because the product
of two even elements can also be described as a product of an even number of
interchanges.
The two simplest irreducible representations of the symmetric group S N play

an important role in quantum mechanics. The two representations in question are
both one-dimensional. In fact, they are the only one-dimensional representations,
and all other irreducible representations of S N have more than one dimension.
The first one is the trivial representation, which assigns the number 1 to every
group element. The second one is a bit more interesting. It assigns the number 1
to every even permutation and the number −1 to every odd permutation. This is a
representation because even and odd permutations satisfy the rules: odd × odd =
even, odd × even = odd, and even × even = even.
In quantum theory there are two classes of particles called bosons and fermions.
(This is discussed in much greater detail in Chapter 9.) A system of N identical
bosons or fermions is described by a wave function (x1 , x2 , . . . , x N ), where xi
represents all the coordinates that are required to describe the ith particle. As usual
in quantum mechanics, ||2 is a probability density. One of the remarkable facts
in quantum mechanics is that identical particles are indistinguishable. All that it
is meaningful to ask is how the wave function transforms under a permutation of
the N identical particles. Nature is kind, and only the two simple representations
described above are utilized. Applying a permutation p, the wave function trans-
forms as follows:
(x p1 , x p2 , . . . , x pN ) = D( p)(x1 , x2 , . . . , x N ). (8.7)
In the case of bosons, D is the trivial representation, so that D( p) = 1 in all
cases. This means that the wave function has total symmetry. This simple but pro-
found truth is at the heart of many remarkable phenomena such as superconduc-
tivity, lasers, and Bose condensation.
In the case of fermions, the other one-dimensional representation is utilized.
This means that the wave function has total antisymmetry. One consequence of
this is that the wave function vanishes when the coordinates of two of the fermions
approach one another. This is the origin of the Pauli exclusion principle for elec-
trons, which is responsible for the stability of ordinary matter.
Unitary and orthogonal groups

In the previous section we discussed a class of finite groups. Let us now turn to
some examples of infinite groups. The groups we will discuss here are examples
of Lie groups, named for the mathematician Sophus Lie.
In the discussion of group representations we explained that the multiplication
rule for a group can be represented by associating a square matrix to each group
element and matrix multiplication to group multiplication. Here we will turn the
story around and consider groups that are defined in terms of matrices and their
Finite and infinite groups 269
multiplication in the first place. For such a group we get one faithful representation
for free – namely, the defining representation, which simply assigns each matrix to
itself. There will also be other representations, of course.
Let us begin by defining the Lie group O(N ). This is the group formed by
all N × N real orthogonal matrices. By definition, an orthogonal matrix A, with
matrix elements denoted Ai j , is one that satisfies the equation

N
Aik A jk = δi j . (8.8)
k =1
In matrix notation this takes the more compact form A A T = I . Here I denotes the
unit matrix (Ii j = δi j ), and A T denotes the transpose matrix ((A T )k j = A jk ).
Let us prove that orthogonal matrices form a group. First we note that the asso-
ciativity axiom is satisfied automatically, since it is always valid for multiplication
of finite-dimensional matrices made from numbers (real or complex). Then we
note that the N × N matrix I is an orthogonal matrix, and it is obviously the iden-
tity element of the group. Next we must show that the product of two orthogonal
matrices is itself an orthogonal matrix. This requires checking whether, for any
two N × N orthogonal matrices A and B, it is true that (AB)(AB)T = I . This
is verified easily using the identity (AB)T = B T A T and associativity to rearrange
the parentheses. The final requirement that needs to be checked is that every or-
thogonal matrix has an inverse, and that this inverse is itself an orthogonal matrix.
To prove this we note that the inverse of an orthogonal matrix A is A T . Therefore
we need to check whether A T (A T )T = I . This is true, because (A T )T = A.
As we have said, any element A of the Lie group O(N ) satisfies the equation
A A T = I . Let us take the determinant of each side of this equation. To do this we
recall the following facts: det I = 1, det (AB) = det A det B, and det A T = det A.
Applying these facts to the equation A A T = I , one deduces that (det A)2 = 1.
Hence det A = ±1. The orthogonal matrices that satisfy det A = 1 form a sub-
group of O(N ), which is denoted S O(N ). The letter S stands for “special,” mean-
ing that the determinant is unity, or the matrix is unimodular.
The physical interpretation is as follows: orthogonal matrices with det A = 1
can be interpreted as rotations of an N -dimensional Cartesian space. (The N-
dimensional rotation group was discussed previously in Chapter 5.) Those with
det A = −1 can be interpreted as reflections of the same N -dimensional Cartesian
space.
This construction has a straightforward generalization to N × N complex ma-
trices. As mentioned in an earlier footnote, a complex N × N matrix A is unitary
if and only if it satisfies the equation A A† = 1. Here A† is called the adjoint (or
Hermitian conjugate) of A. It is the complex conjugate of the transpose. In terms
of components (A† )i j = (A ji ) . The claim is that the set of all such matrices form
a group called U (N ). The proof that the axioms of a group are satisfied works the
same way as before, and we leave it as an exercise. Note that, as a special case,
the matrix A is allowed to be real. For such a matrix, adjoint and transpose are the
same thing. Therefore O(N ) is a subgroup of U (N ).
Let us identify an important subgroup of U (N ) by the same method we used
for O(N ). We form the determinant of both sides of the equation A A† = 1. By
the same reasoning as before, this leads to the equation |det A|2 = 1. Therefore
|det A| = 1, and det A is a phase. As before, the unimodular (det A = 1) unitary
matrices form a subgroup denoted SU (N ).
8.3 Rotations form a group

Rotations in two dimensions were discussed in Chapter 1, where the group prop-
erties were introduced (without mentioning groups). Specifically, we discussed the
2 × 2 matrix R(θ) describing a rotation by an angle θ. It was shown that R(0) is
the identity matrix I , and that R(θ)R(− θ) = I , so that R(− θ) is the inverse of
R(θ). In an exercise you showed the group closure rule R(θ1 )R(θ2 ) = R(θ1 + θ2 ).
In this section we wish to generalize these facts to rotations in an arbitrary number
of dimensions with special emphasis on the particular case of three dimensions.
As we discussed in the beginning of this chapter, one way of thinking about ro-
tation groups is as symmetries of spheres. Since an n-dimensional sphere S n is nat-

urally embedded in n + 1 Euclidean dimensions (by the equation i (x i )2 = R 2 ),
the rotations of S n can be thought of as acting in n + 1 dimensions. The
group of symmetries is O(n + 1) if we include reflections as well as rotations,
or S O(n + 1) if we restrict to rotations only. Recall that the (n + 1) × (n + 1)
matrices that represent the group elements have determinant −1 for reflections
and +1 for rotations. The group S O(2), which describes the rotations of a circle,
is relatively trivial since it is Abelian. In this section, we wish to explore rotations
in three dimensions, given by the group S O(3), which is decidedly less trivial.
Needless to say, it is the example of greatest physical interest. We will explore this
Lie group by focusing on infinitesimal transformations – ones that differ infinites-
imally from the identity. Following that, we will explore a closely related group,
SU (2), which is required to describe rotations of fermions in quantum mechanics.
Finally, we will make some brief remarks about rotations in higher dimensions.
Infinitesimal rotations and the SO(3) algebra

In Section 1.4 we considered infinitesimal rotations in two dimensions. The type of
analysis we described there is very powerful, and it will be generalized here to the
Rotations form a group 271
case of rotations in three dimensions. The basic conceptual fact is that in the case
of a continuous group the structure of the group in an infinitesimal neighborhood
of the identity element contains all the essential information required to understand
arbitrary finite group transformations. The only caveat is that different groups can
be identical in the neighborhood of the identity. When this happens, the two groups
are distinguished by their global topology. As we will see, the two groups have the
same Lie algebra. A pair of groups that are related in this way, which we will
describe in detail, are S O(3) and SU (2).
Recall that rotation matrices are orthogonal matrices with determinants of +1,
and an orthogonal matrix is one that satisfies A A T = A T A = I . Here I denotes
the unit matrix. Since we wish to focus on rotations in three dimensions, we will
be interested in the case that A is a 3 × 3 matrix. For example, a rotation by an
angle θ in the x–y plane is given by the matrix
 
cos θ −sin θ 0
Rz (θ) =  sin θ cos θ 0. (8.9)
0 0 1
Now suppose that the rotation angle θ is very small. Expanding to first order we
have
 
1 −θ 0
Rz (θ) = θ 1 0 + O(θ 2 ) = I + θ L̃ z + O(θ 2 ), (8.10)
0 0 1
where
 
0 −1 0
L̃ z = 1 0 0. (8.11)
0 0 0
This matrix encodes infinitesimal rotations about the z axis. It is referred to as an

infinitesimal generator, because finite transformations can be built up by repeated
action of infinitesimal ones.
Frequently in the physics literature, one includes an extra factor of i in the def-
inition of the infinitesimal generators in order that the matrices are Hermitian. Of
course, if one does this, compensating factors of −i have to be added to their co-
efficients so that the formulas are unchanged. Therefore we define L̃ z = i L z , and
it follows that Rz (θ) = I + iθ L z + · · · and
 
0 −1 0
L z = −i 1 0 0. (8.12)
0 0 0
A rotation by an angle θ about a specified axis can be obtained by rotating N

times by an angle θ/N . Infinitesimal rotations enter in the limit N → ∞. This
limit gives an exponential of the infinitesimal generator, that is,
 
cos θ −sin θ 0
eθ L̃ z = eiθ L z =  sin θ cos θ 0 = Rz (θ). (8.13)
0 0 1
To see that this is correct, note that it has the correct small θ expansion, and it
has the group property eiθ1 L z eiθ2 L z = ei(θ1 + θ2 )L z . Also, it is an orthogonal matrix,
because
(eiθ L z )T = eiθ L z = e−iθ L z ,

T
(8.14)
which is the inverse of the original matrix. In a completely analogous way we can
derive infinitesimal generators for rotations about the x and y axes:
 
0 0 0
L x = −i 0 0 −1, (8.15)
0 1 0
 
0 0 1
L y = −i  0 0 0. (8.16)
−1 0 0
The infinitesimal generators of a Lie group form an algebraic system called a
Lie algebra. The closure of the group under multiplication translates into closure
of the vector space spanned by the infinitesimal generators under commutation.
Explicitly, for the example at hand, one has the commutation relation
[L x , L y ] = L x L y − L y L x = i L z , (8.17)
as is easily verified. Two similar formulas are obtained by cyclically permuting the
indices x, y, z.
Representation by differential operators

Let us consider what the preceding analysis implies for the Cartesian coordinates
x, y, z. For a rotation by an infinitesimal angle θ about the z axis, we have
      
x 1 −θ 0 x x − θy
 y  =  θ 1 0   y  + O(θ 2 ) =  y + θ x  + O(θ 2 ) (8.18)
z 0 0 1 z z
This can again be written in the form

   
x x
 y  = (1 + iθ L z + O(θ 2 ))  y  (8.19)
z z
but now with L z expressed as a differential operator

∂ ∂
L z = −i x −y . (8.20)
∂y ∂x
Similar expressions for L x and L y are obtained by cyclically permuting the coordi-
nates x, y, z. Formulas of this type are familiar to students of quantum mechanics
as operators that (aside from a factor of h̄) represent orbital angular momentum.
The infinitesimal generators, when represented as differential operators, satisfy
the same commutation relations as before. Representations of infinitesimal gener-
ators by differential operators play an important role in the general theory of Lie
groups and Lie algebras.
The relationship between SO(3) and SU(2)

The three-dimensional rotation group S O(3) is related closely to the group SU (2),
which consists of all two-dimensional unitary matrices with unit determinants. In
fact, there is a two-to-one relationship between the two groups, with two distinct
elements of SU (2) corresponding to each element of S O(3). One says that SU (2)
is a covering group of S O(3). There are a variety ways of understanding this re-
lationship, and we will mention a couple of them here. One crucial fact is that the
two groups are identical in the vicinity of the identity element, and therefore they
have the same Lie algebra.
Let us begin by giving a couple of explicit descriptions of the matrices that form
the group SU (2). One possible parametrization is

a b
U= (8.21)
−b∗ a ∗
subject to the restriction that det U = |a|2 + |b|2 = 1. As the reader can verify,
this specific parametrization ensures that the matrices are unitary (U †U = 1). If
one separates real and imaginary parts by writing a = a1 + ia2 and b = b1 + ib2 ,
then the determinant condition takes the form (a1 )2 + (a2 )2 + (b1 )2 + (b2 )2 = 1,
which is the equation for a three-dimensional sphere S 3 . Therefore, one says that
the group manifold of SU (2) is S 3 .
Let us now consider SU (2) matrices that differ from the identity matrix by an in-
finitesimal amount. Such matrices can be written in terms of three real infinitesimal
parameters αx , α y , αz as follows

1 + iαz iαx + α y
U= + O(α 2 ) (8.22)
iαx − α y 1 − iαz
This corresponds to a1 = 1, a2 = αz , b1 = α y , and b2 = αx , which satisfies the
determinant condition to order (α)2 . This expression can be recast in terms of a
basis of three 2 × 2 matrices σ called Pauli matrices:
U = 1 + i σ · α + O(α 2 ), (8.23)
where (as in Chapter 7)

0 1 0 −i 1 0
σx = σy = σz = . (8.24)
1 0 i 0 0 −1
The Pauli matrices are infinitesimal generators of SU (2). Let us therefore ex-
amine their commutation relations. One finds that
[σx , σ y ] = 2iσz , (8.25)
and two more equations given by cyclic permutation of indices. Aside from the
factor of 2, these are the same commutation relations that we found in the preced-
ing subsections. Thus the three matrices σ /2 have the same commutation relations
as the three operators L that generate S O(3). This demonstrates that the two Lie
groups have the same Lie algebra.
Let us now consider finite SU (2) transformations. We described one way of pa-
rameterizing them above, but let us now introduce another one. Consider a matrix
of the following form
U = exp (i σ · α ), (8.26)
where now the vector α represents three finite real numbers. This agrees with the
previous expression to first order in α, but it is exactly unitary. To show this, note
that the Pauli matrices are Hermitian, and therefore the adjoint of U is equal to its
inverse.
The relationship between SU (2) and S O(3) can be obtained by considering the
traceless Hermitian matrix

z x − iy
X = σ · x = . (8.27)
x + iy −z
If one performs a unitary transformation, by an element of the group SU (2) one
obtains a transformed matrix
X = U −1XU = σ · x . (8.28)
Note that X is also Hermitian and traceless, and therefore x consists of three
real numbers that are related to x by a real linear transformation. Furthermore
this linear transformation is a rotation. To see this, we note that −det X = x 2 +
y 2 + z 2 , and det X = det X . Therefore the vectors x and x have the same length.
In this way one associates a unique rotation to each element of the group SU
(2).
To understand this better, let us specialize to the case of transformations gener-
ated by σz , which will correspond to rotations about the z axis. One has
iα
e 0
U = exp (iασz ) = . (8.29)
0 e−iα
Moreover, for this matrix one finds that
X = U −1XU = σ · x (8.30)
corresponds to a rotation about the z axis by an angle 2α. In particular, the matrix
U = −I , corresponding to α = π, corresponds to a rotation by an angle of 2π ,
which is the identity element of S O(3).
We can now understand the two-to-one relationship between the two groups.
Given an SU (2) matrix U , there is a another distinct SU (2) matrix −U , which
corresponds to the same S O(3) rotation. A fancier way of saying this is that the
two SU (2) matrices I and −I form an invariant subgroup of SU (2), and S O(3) is
isomorphic to SU (2) modulo this invariant subgroup. Another way of saying the
same thing is that the map from SU (2) to S O(3) described above is a two-to-one
homomorphism.
Spinor representations
Because of the two-to-one relationship between the two groups, the three-
dimensional matrices that define S O(3) provide a three-dimensional representa-
tion of SU (2), whereas the two-dimensional matrices of SU (2) do not provide a
two-dimensional representation of S O(3). More generally, it turns out that SU (2)
has irreducible representations of every positive dimension, whereas only the ones
of odd dimension are also representations of S O(3). The even-dimensional repre-
sentations of SU (2) are sometimes called spinor representations.
It is customary to label the irreducible representations by their “spin” j. The
spin is related to the dimension n of the representation by n = 2 j + 1. Thus
the irreducible representations of S O(3) are labeled by j = 0, 1, 2, . . . , and the
spinor representations of SU (2), which are not representations of S O(3), have
j = 1/2, 3/2, 5/2, . . . .
Rotations in higher dimensions

Rotations in D dimensions are described by the Lie group S O(D). This can be
analyzed in much the same way that we analyzed S O(3). However, to describe this
generalization it is convenient to make some minor notational changes. First of all,
we label the Cartesian coordinates by x 1 , x 2 , . . . , x D , rather than using distinct
letters x, y, z. Secondly, in D dimensions specific rotations should be described
by reference to the two-dimensional plane that is rotated. In three dimensions such
a rotation can be characterized by the third direction, which does not transform,
and can be identified as the axis of rotation. In D dimensions, the generalization of
this statement is that there are (D − 2) orthogonal directions that do not transform
under a specific rotation. Thus, what we have been calling a rotation about the
z axis, we would now characterize as a rotation in the x 1 –x 2 plane.
Let us now consider the generalization of the infinitesimal generators of rota-
tions obtained for S O(3) in Section 8.3. Specifically we wish to generalize the
formula in Eq. (8.12) for L z , which we would now call L 12 . Altogether, there are
D(D − 1)/2 infinitesimal generators L i j = −L ji . This number is the dimension
of Lie group S O(D). In other words, it is the number of real parameters that is
required to characterize an arbitrary rotation. The infinitesimal generators L i j de-
scribe infinitesimal rotations of two of the D coordinates (x i and x j ). Therefore
they are given by D × D matrices in which the only nonzero entries occur in the
i j and ji positions. There is only one possible generalization of Eq. (8.12) with
these properties, namely
(L i j )kl = i(δik δ jl − δil δ jk ). (8.31)
The Lie algebra of S O(D) is obtained by computing the commutators of these

matrices, giving formulas that generalize (8.17), for example. The result of this
computation is
[L i j , L kl ] = i(δik L jl − δil L jk + δ jl L ik − δ jk L il ). (8.32)
The infinitesimal generators of S O(D) can also be represented by differential

operators that act on the coordinates, generalizing (8.20). The formula is derived
in the same way as before, and one finds

∂ ∂
L i j = −i x i j − x j i . (8.33)
∂x ∂x
One can verify that these differential operators satisfy commutation relations given
above.
It is natural to ask whether the existence of a covering group with spinor rep-
resentations also generalizes from three dimensions to higher dimensions. Not
Lorentz transformations form a group 277
surprisingly, the answer is yes. The covering group of S O(D) is called Spin(D).
In particular, in the case of three dimensions, Spin(3) = 4SU (2). In general, there
is not another equivalent name for Spin(D), though it does turn out that Spin(5) is
isomorphic to the symplectic group USp(4) and Spin(6) is isomorphic to SU (4).
Thus, in these two cases the minimal spinor representations are four-dimensional.
8.4 Lorentz transformations form a group

In Chapter 2 we discussed Lorentz transformations and showed that they could be
described by 2 × 2 matrices, which characterize linear transformations of the coor-
dinates x and t. We also discussed the fact that successive Lorentz transformations
give a Lorentz transformation, which we now recognize as the group property. In
particular, in Exercise 2.19 you showed that L(ξ1 )L(ξ2 ) = L(ξ1 + ξ2 ) corresponds
to the velocity addition rule. The case of one spatial dimension is somewhat spe-
cial, however. For the more relevant case of three spatial dimensions, the possibili-
ties are richer. For example, one can imagine multiplying a Lorentz transformation
in one direction by a Lorentz transformation in another. The result turns out not
to be a pure Lorentz boost, but rather a combination of a boost and a rotation.
Therefore, in order to describe the group that contains Lorentz transformations, it
is necessary to view it as an extension of the group of rotations. In other words the
rotation group is a subgroup of the Lorentz group. This does not contradict what
we found when there was only one spatial direction, since in that case there are no
rotations.
Lorentz transformations can be viewed, from a mathematical point of view,
as generalized rotations in a spacetime with an indefinite metric, as was al-
ready discussed in Chapter 5. A rotation about the origin of a D-dimensional
Euclidean space transforms the coordinates according to the rule x → x̃, where
l˜ j
x̃ = j R j x and R is an orthogonal matrix satisfying R R = R R = I , or
l T T
equivalently R T I R = I . This can be written in terms of components as

˜
Rik̃ δk̃l˜ R lj = δi j . (8.34)
k̃ l˜
The reason we have written the formulas this way is to emphasize that the trans-
formation preserves the “metric” I , which corresponds to the matrix δi j . These
formulas just represent the fact that the vectors x and x̃ have the same length s,

where the length squared is given by the usual formula, namely s 2 = i (x i )2 =

i j δi j x x .
i j
Lorentz transformations have a similar rule, namely they are linear

transformations of the coordinates that preserve the invariant interval

s 2 = −(x 0 )2 + i (x
i )2 . We can rewrite this in the convenient form

s2 = ηµν x µ x ν , (8.35)
µν
where the matrix η is the Lorentz metric. We use lower case Greek indices to
label spacetime directions, with µ = 0 corresponding to the time direction and
µ = i corresponding to the spatial directions. Specifically, the Lorentz metric η
is a diagonal matrix, whose diagonal components are η00 = −1 and ηii = 1 for
i = 1, 2, . . . D. The usual case is D = 3, of course.
Let us now consider a linear transformation of spacetime given by x̃ λ =
λ̃ ν
ν L ν x , as before. The only difference from before is that the indices now range
from 0 to D rather than from 1 to D. We require that this transformation preserve
the invariant interval s 2 , since this is the defining property of a Lorentz transfor-
mation. Expressed in terms of matrices this yields the condition L T ηL = η. This
can be written in terms of coordinates as

L ρ̃µ ηρ̃ λ̃ L λ̃ν = ηµν . (8.36)
ρ̃ λ̃
Any (D + 1) × (D + 1) matrix that satisfies this equation corresponds to a

Lorentz transformation. The converse is also true: each Lorentz transformation
corresponds to a matrix A satisfying this equation. Therefore Eq. (8.36) is the
defining representation of the Lorentz group. This representation is (D + 1)-
dimensional when there are D spatial dimensions.
We can now show that the set of all real matrices L satisfying Eq. (8.36) form a
group. It is quite obvious that L = I , which is the identity element, is a solution.
Also, if L is a solution, then so is the inverse matrix L −1 = ηL T η. It is only a little
more work, which we leave to the reader, to verify the closure property: namely,
if L 1 and L 2 are solutions, then so is their product L 1 L 2 . The mathematical name
for this group of matrices is O(D, 1), since it preserves a metric with D plus signs
and one minus sign.
The Lorentz group has four components

The group O(D, 1) includes components that describe spatial reflections and that
reverse the direction of time. To illustrate the latter, note that L = η is an element
of the group that reverses the sign of x 0 without changing the spatial coordinates.
Similarly, there is a diagonal L that changes the sign of a spatial coordinate, which
describes a reflection. Altogether, the group has four components: (1) no reflection
or time reversal, (2) reflection but no time reversal, (3) time reversal but no reflec-
tion, and (4) reflection and time reversal. The determinant of the matrix L is +1 for
components (1) and (4), and −1 for components (2) and (3). The “proper Lorentz
group” is defined to consist of transformations belonging to component (1) only.
They form a subgroup of O(D, 1). This subgroup is usually denoted S O(D, 1).
Note that this is a more severe restriction than requiring that the determinant is
+1, since that would also allow transformations of the fourth type.3 Components
(1) and (2), taken together, form the subgroup of the full Lorentz group that pre-
serves the direction of time. The Lorentz transformations with this property are
sometimes called orthochronous Lorentz transformations.
In thinking about Lorentz transformations, it is sometimes helpful to try to visu-
alize their effect on light cones. For example, time reversal clearly interchanges the
forward and backward light cones, whereas spatial reflections preserve light cones.
The difference between space and spacetime rotations

Both the Lorentz group O(D, 1) and the proper Lorentz group S O(D, 1) contain
the group of spatial rotations S O(D) as a subgroup. Note that this means that spa-
tial rotations are considered to be special cases of Lorentz transformations. In more
elementary discussions one often distinguishes rotations from boosts. However,
from a more general viewpoint, this is awkward, since the generic transformation
is a combination of a boost and a rotation. So it is much more natural to call all of
these transformations Lorentz transformations and to identify spatial rotations as
those Lorentz transformations that are contained in the rotation subgroup.
A much more profound distinction between the rotation group S O(D) and the
proper Lorentz group S O(D, 1) is that the former is a compact Lie group and the
latter is a noncompact Lie group. This is an important topological distinction that
has far-reaching consequences for the structure of the groups and their representa-
tions.
A group is compact if the group manifold has finite volume. To make this state-
ment precise one ought to explain how one measures the volume of a group. Here
let us just describe it intuitively. A rotation is necessarily bounded in size: after
you reach an angle of 2π, you’re back where you started. As a consequence any
group of rotations has finite volume. Lorentz boosts, on the other hand, can be arbi-
trarily large. The boost factor γ = (1 − v 2 /c2 )−1/2 approaches infinity as v → c.
Therefore Lorentz groups have infinite volume. Another fact that illustrates the
same point concerns the surfaces that are preserved by the respective groups. Ro-
tation groups preserve spheres, which have finite volume, whereas Lorentz groups
preserve hyperboloids, which have infinite volume.
3 An alternative notation in which this does include the fourth component is also used. In that case, some addi-
tional symbol, such as a vertical arrow, is needed to represent the restriction to the first component.
Infinitesimal Lorentz transformations

Let us now construct the infinitesimal generators for Lorentz transformations. The
formulas are given by straightforward generalization of those for rotations that
were obtained in Section 8.3. We just have to be sure to replace unit matrices by
the Lorentz metric in the appropriate places. In that section we found the matrix
that describes the infinitesimal generator of a rotation in the x i –x j plane is given
by
(L i j )kl = i(δik δ jl − δil δ jk ). (8.37)
We need a generalization to the Lorentz group generators that reduces to this

formula when we restrict all indices to spatial directions. The natural (and correct)
generalization to spacetime is
(L µν )λρ = i(ηµρ δνλ − ηνρ δµλ ). (8.38)
Here we find it convenient to write the row index ρ as a subscript and the column
index λ as a superscript. This ensures that when we multiply matrices the summa-
tion combines a subscript with a superscript. This is a convenient way to ensure
that the Lorentz metric is properly taken into account. Note that for this choice
of L µν the only nonzero elements of the matrix L i j are those that are given in
Eq. (8.37). However, this equation does more. It also gives the infinitesimal gen-
erators for Lorentz boosts. A boost in the x i direction is generated by the matrix
L 0i . From Eq. (8.38) we can read off that the only nonzero elements of the matrix
L 0i are (L 0i )i0 = (L 0i )0i = −i.
The Lie algebra of S O(D, 1) is obtained by computing the commutators of the
matrices that describe the infinitesimal transformations. The result of this compu-
tation is
[L µν , L ρλ ] = i(ηµρ L νλ − ηµλ L νρ + ηνλ L µρ − ηνρ L µλ ). (8.39)
Again generalizing what we did previously for rotations, the infinitesimal gen-
erators of S O(D, 1) can be represented by differential operators that act on the
coordinates. This yields

ρ ∂ ρ ∂
L µν = −i ηµρ x − ηνρ x . (8.40)
∂xν ∂xµ
In this equation it is understood that the repeated index ρ is summed. In particular,
for the generators of Lorentz boosts we have

i ∂ 0 ∂
L i0 = −i x +x . (8.41)
∂x0 ∂xi
One can verify that these differential operators satisfy commutation relations given
above.
To illustrate these equations, consider an infinitesimal boost in the x 1 direction
generated by L 10 :
0 0 0
x x x + θ x1
= (1 + iθ L 10 + O(θ ))2
= + O(θ 2 ). (8.42)
x 1 x1 x1 + θ x0
Exponentiating this to give a finite Lorentz boost gives
0 0
x x cosh θ x 0 + sinh θ x 1
= exp [iθ L 10 ] = . (8.43)
x 1 x1 cosh θ x 1 + sinh θ x 0
We recognize these as the standard formulas for a Lorentz
boost. The velocity
v= βc associated to the boost is given by cosh θ = 1/ 1 − β and sin h θ =
2
β/ 1 − β .
2
SL(2, C)
The proper Lorentz group for four-dimensional spacetime, S O(3, 1), contains the
rotation group S O(3) as a subgroup. As we explained in Section 8.3, S O(3) has a
covering group, SU (2), which contains two elements (of opposite sign) for every
element of S O(3). This extension of the rotation group is required to describe
fermions in quantum theory. Suppose now we want to construct relativistic (that is,
Lorentz-invariant) quantum theories that contain fermions. What is the appropriate
symmetry group in such a case? Clearly, we require a covering group of S O(3, 1)
that contains SU (2) as a subgroup. Fortunately, such a group exists, and it is called
S L(2, C).
The group S L(2, C) consists of all 2 × 2 complex matrices M with det M =
1 (unimodular matrices). This group has three complex dimensions (or six real
dimensions). A generic element has the form

a b
M= ad − bc = 1, a, b, c, d ∈ C. (8.44)
c d
Since there are six independent real parameters, there are six independent infinites-
imal generators. One can show that they give precisely the same Lie algebra as
S O(3, 1).
The relationship between S L(2, C) and S O(3, 1) can be described by an exten-
sion of the reasoning that we used to explain the relationship between SU (2) and
S O(3). For this purpose, we associate the Hermitian matrix
0
x + x3 x1 − i x2
X= (8.45)
x1 + i x2 x0 − x3
to the spacetime point x µ . If one performs a transformation, by an element M of

the group S L(2, C) one obtains a transformed matrix
X = M † XM. (8.46)
Note that X is also Hermitian, and therefore it corresponds to a spacetime point
x µ consisting of four real numbers that are related to x µ by a real linear trans-
formation. Furthermore, this linear transformation is a Lorentz transformation.
To see this, we note that −det X = s 2 = −(x 0 )2 + (x 1 )2 + (x 2 )2 + (x 3 )2 , and
det X = det X . Therefore the vectors x µ and x µ have the same invariant length.
In this way one associates a unique Lorentz transformation to each element of
the group S L(2, C). In fact, the mapping is still two-to-one, because the matrices
M and −M correspond to the same Lorentz transformation.
The group S L(2, C) has two inequivalent two-dimensional irreducible represen-
tations.4 One of them is the defining representation: D (L) (M) = M and the other
is the complex conjugate representation: D (R) (M) = M ∗ . As representations of
S L(2, C) these are inequivalent, though they become equivalent upon restriction
to the SU (2) subgroup.
Two-component vectors that are transformed by these representation matrices
are called Weyl spinors. Ones that transform by D (L) are called “left-handed” or
“positive chirality,” whereas the ones that transform by D (R) are called “right-
handed” or “negative chirality.” Explicitly, if ψα is a left-handed Weyl spinor, un-
der a Lorentz transformation M,
ψα → ψα = Mα β ψβ α, β = 1, 2. (8.47)
The Hermitian conjugate ψ † is a right-handed Weyl spinor.
8.5 The Poincaré group

A relativistic theory is required to be invariant under the group of Lorentz trans-
formations. However, the Lorentz group is only a subgroup of the whole sym-
metry group. Additional symmetries, which are not part of the Lorentz group,
are shifts (or “translations”) in space and time. Such translations are given by
x µ → x µ + a µ , where the a µ s are D + 1 arbitrary real constants that characterize
the translations. Note that a 0 /c is the time translation. Adjoining these transfor-
mations to the Lorentz group yields a Lie group called the Poincaré group. It is
a nontrivial extension in the sense that translations do not commute with Lorentz
transformations. The subgroup generated by spatial translations and rotations is the
4 These representations are not unitary, but that’s okay, since the group is noncompact.
The Poincaré group 283
Euclidean group E(D), which was discussed earlier. It does not require relativity,
so it is also a symmetry of nonrelativistic theories.
Space and time translations

Space and time translations can be represented by differential operators. To see
how this works consider a scalar function f (x µ ). Performing a translation gives
f (x µ + a µ ). For an infinitesimal translation we can expand to first order
∂
f (x ρ + a ρ ) = f (x ρ ) + a µ
f (x ρ ) + O(a 2 ) (8.48)
∂xµ
Therefore ∂/∂ x µ generates a translation of x µ .
Translations form a noncompact abelian Lie group. One can exponentiate the
infinitesimal transformations to obtain finite ones. Specifically,

ρ ρ µ ∂
f (x + a ) = exp a f (x ρ ). (8.49)
∂xµ
To prove this, expand the exponential in a power series. Then one sees that this
gives the standard formula for the power series expansion of a function about a
shifted point. All that is required is that the function f should have sufficient ana-
lyticity to ensure the convergence of the series.
Let us rewrite the preceding formula in the form
f (x ρ + a ρ ) = exp (ia µ Pµ ) f (x ρ ), (8.50)
where
∂
. Pµ = −i (8.51)
∂xµ
Aside from a factor of h̄, the spatial part of Pµ is the operator that represents
momentum in quantum mechanics, as was discussed in Chapter 7. The Lie algebra
for translations is simply
[Pµ , Pν ] = 0, (8.52)
so the group of translations is Abelian.
Commutators of translations and rotations

In the preceding subsection, we have introduced the generator of translations, Pµ ,
and presented a representation in terms of differential operators that act on the
spacetime coordinates x µ . In Section 8.4 we did the same thing for the generators
of Lorentz transformations L µν . In each case we could show that these operators

satisfied the appropriate Lie algebras, namely
[Pµ , Pν ] = 0 (8.53)
and
[L µν , L ρλ ] = i(ηµρ L νλ − ηµλ L νρ + ηνλ L µρ − ηνρ L µλ ). (8.54)
To complete the characterization of the Poincaré algebra, we must also specify the
commutators of the translation generators with the Lorentz generators. Using the
representation in terms of differential operators, it is straightforward to compute
this algebra. One finds that
[Pµ , L νρ ] = i(ηµν Pρ − ηµρ Pν ). (8.55)
This formula encodes the fact that Pµ transforms as a spacetime vector.

The preceding formulas are a complete description of the Poincaré algebra. Ar-
bitrary finite Poincaré group transformations can be written as the sum of a trans-
lation and a Lorentz transformation. That is what one obtains by exponentiating a
linear combination of the infinitesimal transformations.
Representations of the Poincaré group

The Poincaré group can be represented in terms of differential operators and in
terms of matrices. We have already given the representation in terms of differential
operators, so we now want to describe a representation in terms of matrices. We
know such a representation for the Lorentz subgroup, namely in terms of (D + 1)-
dimensional matrices in the defining representation of S O(D, 1). So the problem
is to extend this representation to the entire Poincaré group.
If we are to find such a matrix representation, we must be able to represent
translations by matrices. To do that we must confront the fact that translations
are inhomogeneous transformations. This means that the transformation x µ →
x µ + a µ has a term that is not a multiple of x µ . Therefore the right-hand side
cannot possibly be written as an x-independent matrix acting on the vector x µ .
The trick for overcoming this obstacle is to use matrices with an extra row and
column as follows:
µ µ µ ν µ
x δν a x x + aµ
→ = . (8.56)
1 0 1 1 1
To generalize this construction to the entire Poincaré group, let us write an ar-
µ̃ µ̃
bitrary Poincaré group transformation in the form x̃ µ = L ν x ν + a µ , where L ν
Exercises 285
is a S O(D, 1) matrix. Clearly this transformation is reproduced by the matrix for-

mula
µ
µ̃ µ̃ ν
µ̃ ν
x̃ Lν a x L ν x + a µ̃
= = . (8.57)
1 0 1 1 1
The Poincaré group has two classes of particle representations – massless

and massive. This is important since both kinds of particles are known to exist,
for example, electrons and protons are massive, but photons and gravitons are
massless. Massless particles travel with the speed of light in any inertial refer-
ence frame. In the massive case the Lorentz-invariant combination of momen-
tum and energy P 2 c2 − E 2 = −M 2 c4 = 0, and one can choose a frame of ref-
erence, called the rest frame, such that E = Mc2 and P = 0. Note that the energy
of a particle is understood here to include its rest energy as well as its kinetic
energy.
Let us consider a massive particle at rest in four-dimensional spacetime. The
subgroup of the Poincaré group (called the little group) that keeps the particle at
rest consists of all rotations and translations. In quantum theory, the particle is
represented by a multicomponent wave function that transforms according to a
particular irreducible representation of SU (2), which corresponds to the spin s of
the particle.
In the massless case it is not possible to bring the particle to rest, and the anal-
ysis is different. One can restrict the choice of Lorentz frame by requiring that
the particle travels in a particular direction, for example the z direction. Then
we have Pz c = E and Px = Py = 0. In this case the little group includes only
those rotations that preserve the direction of motion, that is, rotations about the z
axis. The representations of this group play the role of spin in this case. These
representations are one-dimensional, since the group S O(2) is Abelian. These
one-dimensional representations are labeled by a quantum number called helic-
ity. However, if spatial inversion (parity) is also a symmetry the group is enlarged
to O(2), which is nonabelian. Then a positive and a negative helicity can belong
to the same irreducible representation.
Exercises
8.1 Describe the transformation of the plane given by an arbitrary element
g of the group E 2 (described in Section 8.1). Let g1 and g2 denote two
elements of the group E 2 . Compute the transformation that is described
by the group product g1 g2 .
8.2 Work out the multiplication table for the group Dn described in
Section 8.1.
8.3 Which of the following form a group under ordinary multiplication: the
nonzero integers, the nonzero real numbers, the nonzero complex num-
bers? Why have we excluded zero?
8.4 Enumerate the elements of the group S3 . Is this group Abelian? Which
elements are even and which are odd? Verify that the even elements form
a subgroup.
8.5 Prove that the set of all unitary N × N matrices form a group.
8.6 Prove that the mapping A → det A is a one-dimensional representation of
the group U (N ).
8.7 Show explicitly that the matrix in Eq. (8.12) satisfies Eq. (8.13).
8.8 Verify that the differential operators L i = ii jk x j ∂k satisfy the same Lie
algebra as the 3 × 3 matrices in Eqs. (8.12), (8.15) and (8.16).
8.9 Verify that the map from SU (2) to S O(3) described in Eqs. (8.27) and
(8.28) is a two-to-one homomorphism.
8.10 Verify that the differential operators in Eq. (8.33) satisfy the commutation
relations in Eq. (8.32).
8.11 Using the representation in Eq. (8.41), show that the commutator of two
boost generators [L i0 , L j0 ] gives a rotation generator.
8.12 Construct the infinitesimal generators of SL(2, C). Show that they give the
same Lie algebra as S O(3, 1).
8.13 Verify the commutation relations in Eq. (8.55).

Group Theory and Relativity: 8.1 What Is A Group?

Uploaded by

Copyright:

Available Formats

Group Theory and Relativity: 8.1 What Is A Group?

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Group Theory and Relativity: 8.1 What Is A Group?

Uploaded by

Copyright:

Available Formats

8

Group theory and relativity

In previous chapters we have encountered frequently the concept of symmetry of

8.1 What is a group?

Let us begin by explaining the notion of a transformation group. For this

Simple examples of groups

The spheres discussed in the preceding paragraphs have additional symmetries

The four properties that define a group

D(a)D(b) = D(ab). (8.1)

D(a −1 ) = [D(a)]− 1 . (8.4)

Also, since multiplication of finite-dimensional matrices is associative, the asso-

There are families of representations that, in a sense to be defined, are equiva-

Here we have written an (N1 + N2 ) × (N1 + N2 ) matrix in block form. These

8.2 Finite and infinite groups

The two simplest irreducible representations of the symmetric group S N play

Unitary and orthogonal groups

8.3 Rotations form a group

Infinitesimal rotations and the SO(3) algebra

This matrix encodes infinitesimal rotations about the z axis. It is referred to as an

A rotation by an angle θ about a specified axis can be obtained by rotating N

(eiθ L z )T = eiθ L z = e−iθ L z ,

Representation by differential operators

This can again be written in the form

The relationship between SO(3) and SU(2)

where (as in Chapter 7)

[σx , σ y ] = 2iσz , (8.25)

X  = U −1XU = σ · x . (8.28)

Moreover, for this matrix one finds that

X  = U −1XU = σ · x (8.30)

Rotations in higher dimensions

(L i j )kl = i(δik δ jl − δil δ jk ). (8.31)

The Lie algebra of S O(D) is obtained by computing the commutators of these

[L i j , L kl ] = i(δik L jl − δil L jk + δ jl L ik − δ jk L il ). (8.32)

The infinitesimal generators of S O(D) can also be represented by differential

8.4 Lorentz transformations form a group

equivalently R T I R = I . This can be written in terms of components as

Lorentz transformations have a similar rule, namely they are linear

Any (D + 1) × (D + 1) matrix that satisfies this equation corresponds to a

The Lorentz group has four components

The difference between space and spacetime rotations

Infinitesimal Lorentz transformations

(L i j )kl = i(δik δ jl − δil δ jk ). (8.37)

We need a generalization to the Lorentz group generators that reduces to this

(L µν )λρ = i(ηµρ δνλ − ηνρ δµλ ). (8.38)

[L µν , L ρλ ] = i(ηµρ L νλ − ηµλ L νρ + ηνλ L µρ − ηνρ L µλ ). (8.39)

to the spacetime point x µ . If one performs a transformation, by an element M of

8.5 The Poincaré group

Space and time translations

Commutators of translations and rotations

of Lorentz transformations L µν . In each case we could show that these operators

[L µν , L ρλ ] = i(ηµρ L νλ − ηµλ L νρ + ηνλ L µρ − ηνρ L µλ ). (8.54)

[Pµ , L νρ ] = i(ηµν Pρ − ηµρ Pν ). (8.55)

This formula encodes the fact that Pµ transforms as a spacetime vector.

Representations of the Poincaré group

is a S O(D, 1) matrix. Clearly this transformation is reproduced by the matrix for-

The Poincaré group has two classes of particle representations – massless

You might also like

X = U −1XU = σ · x . (8.28)

X = U −1XU = σ · x (8.30)