You are on page 1of 30

Group Theory

Robert Gilmore
Physics Department, Drexel University, Philadelphia, Pennsylvania 19104, USA
(Dated: October 3, 2012)
Printed from:Wiley-Mathematical/GroupChapter/group.tex on October 3, 2012

I. INTRODUCTION in Physics. As a latest tour de force in the develop-


ment of Physics, they play a central roll in the creation
Symmetry has sung its siren song to Physicists since of Gauge Theories, which describe the interactions be-
the beginning of time, or at least since before there were tween Fermions and the Bosons that are responsible for
Physicists. Today the ideas of symmetry are incorpo- the interactions among the Fermions, and which lies at
rated into a subject with the less imaginative and sug- the heart of the Standard Model. We provide the sim-
gestive name of Group Theory. This Chapter introduces plest example of a gauge theory, based on the simplest
many of the ideas of group theory that are important in compact one parameter Lie group U (1), in Sec. VIII.
the natural sciences. For an encore, in Sec. IX we show how the theory of
Natural philosophers in the past have come up with the special functions of mathematical physics (Legendre
many imaginative arguments for estimating physical and associated Legendre functions, Laguerre and associ-
quantities. They have often used out-of-the-box methods ated Laguerre functions, Gegenbauer, Chebyshev, Her-
that were proprietary to pull rabbits out of hats. When mite, Bessel functions, and others) are subsumed under
these ideas were made available to a wider audience they the theory of representations of some low-dimensional Lie
were often improved upon in unexpected and previously groups. The classical theory of special functions came to
unimaginable ways. A number of these methods are pre- fruition in the mid 19th century, long before Lie groups
cursors of group theory. These are Dimensional Analysis, and their representations were even invented.
Scaling Theory, and Dynamical Similarity. We review
these three methods in Sec. II.
In Sec. III we get down to the business at hand, intro- II. PRECURSORS TO GROUP THEORY
ducing the definition of group and giving a small set of
important examples. These range from finite groups to Barenblatt has given a beautiful derivation of Pythago-
discrete groups to Lie groups. These also include trans- ras Theorem that is out-of-the-box and suggests some of
formation groups, which played an important if under- the ideas behind Dimensional Analysis. The area of the
recognized roll in the development of classical physics, in right triangle (a, b, c) is 12 ab (Fig. 1). Dimensionally,
particular the theories of Special and General Relativity. the area is proportional to square of any of the sides, mul-
The relation between these theories and group theory is tiplied by some factor. We make a unique choice of side
indicated in Sec. IV. by choosing the hypotenuse, so that (a, b, c) = c2 f (),
Despite this important roll in the development of is one of the two acute angles, and f () 6= 0 unless = 0
Physics, groups existed at the fringe of the Physics of or /2. Equating the two expressions
the early 20th century. It was not until the theory of
the linear matrix representations of groups was invented    
that the theory of groups migrated from the outer fringes 1 a b 1 b  a  symmetry
f () = = = f ( )
to play a more central roll in Physics. Important points 2 c c 2 c c 2
in the theory of representations are introduced in Sec. (1)
V. Representations were used in an increasingly imag- This shows (a) that the same function f () applies for
inative number of ways in Physics throughout the 20th all similar triangles and (b) f () = f ( 2 ). The latter
century. Early on they were used label states in Quantum result is due to reflection symmetry of the triangle about
systems with a symmetry group: for example, the rota- the bisector of the right angle: the triangle changes but
tion group SO(3). Once states were named, degeneracies its area does not. We need (a) alone to prove Pythagoras
could be predicted and computations simplified. Such ap- Theorem. The proof is in the figure caption.
plications are indicated in Sec. VI. Later, they were used websearch: Pythagoras theorem Barenblatt
when symmetry was not present, or just the remnant of a
broken symmetry was present. When used in this sense,
they are often called dynamical groups. This type of
use grossly extended the importance of Group Theory
2

We can invert this matrix to find


1
1 1 1 1 1 1
0 3 2 = 0 1 2 (4)
0 2 1 0 2 3

This allows us to determine the values of the exponents


which provide the appropriate combinations of impor-
tant physical parameters to construct the characteristic
atomic length:

a 1 1 1 0 1
FIG. 1: The area of the large right triangle is the sum b = 0 1 2 1 = 1 (5)
of the areas of the two similar smaller right triangles:
c 0 2 3 0 2
(a, b, c) = (d, f, a) + (f, e, b), so that c2 f () =
a2 f () + b2 f (). Since f () 6= 0 for a nondegenerate This result tells us that
right triangle, a2 + b2 = c2 .
a0 m1 (e2 )1 (~)2 = ~2 /me2 108 cm (6)
A Dimensional Analysis
To construct a characteristic atomic time, we can re-
place the vector col[0, 1, 0] in Eq. (5) by the vector
How big is a hydrogen atom? col[0, 0, 1], giving us the result 0 ~3 /m(e2 )2 . Fi-
The size of the electron orbit around the proton in nally, to get a characteristic energy, we can form the
the hydrogen atom ought to depend on the electron mass combination E M L2 T 2 = m(~2 /me2 )2 (~3 /me4 )2 =
me , or more precisely the electron-proton reduced mass me4 /~2 . Another, and more systematic, way to get this
= me MP /(me + MP ). It should also depend on the result is to substitute the vector col[1, 2, 2] in Eq. (5).
value of Plancks constant h or reduced Plancks constant Note that our estimate would be somewhat different if
~ = h/2. Since the interaction between the proton and we had used h instead of ~ = h/2 in these arguments.
electron is electromagnetic, of the form V (r) = e2 /r We point out that this method is very useful for estimat-
(Gaussian units), it should depend on e2 . ing the order of magnitude of physical parameters and
Mass is measured in gm. The dimensions of the charge usually gets the prefactor within a factor of 10.
coupling e2 are determined by recognizing that e2 /r is a websearch: dimensional analysis
(potential) energy, with dimensions M 1 L2 T 2 . We will
use capital letters M , L, and T to characterize the three
independent dimensional directions. As a result, the B Scaling
charge coupling e2 has dimensions M L3 T 2 and is mea-
sured in gm(cm)3 /sec2 . The quantum of action ~ has Positronium is a bound state of an electron e with a
dimensions [~] = M L2 T 1 . positron e, its antiparticle. How big is positronium?
Constant Dimensions Value Units To address this question we could work very hard and
M 9.10442 1028 gm solve the positronium Hamiltonian. Or we could be lazy
~ M L2 T 1 1.05443 1027 gm cm2 sec1
and observe that the hydrogen atom radius is inversely
e2
M L3 T 2 2.30655 1019 gm cm3 sec2
proportional to the reduced electron-proton mass, so the
a0 L ? cm
positronium radius should be inversely proportional to
Can we construct something with the dimensions of
the reduced electron-positron mass: pos. = me me/(me +
length from m, e2 , and ~? To do this, we introduce three
me) = 21 me since the electron and positron have equal
unknown exponents a, b, and c and write
masses me = me. Since the reduced electron-proton mass
a0 ' ma (e2 )b ~c = (M )a (M L3 T 2 )b (M L2 T 1 )c is effectively the electron mass, the positronium atom is
= (M )a+b+c L0a+3b+2c T 0a2bc approximately twice as large as the hydrogen atom.
(2) In a semiconductor it is possible to excite an electron
and set this result equal to the dimensions of whatever from an almost filled (valence) band into an almost empty
we would like to compute, in this case the Bohr orbit (conduction) band. This leaves a hole behind in the
a0 (characteristic atomic length), with [a0 ] = L. This valence band. The positively charged hole in the valence
results in a matrix equation band interacts with the excited electron in the conduction
band through a reduced Coulomb interaction: V (r) =

1 1 1

a 0 e2 /r. The strength of the interaction is reduced by
0 3 2 b = 1 (3) screening effects which are swept into a phenomenological
0 2 1 c 0 dielectric constant . In addition, the effective masses
me of the excited electron and the left-behind hole mh
3

are modified from the free-space electron mass values by


many-particle effects. TABLE I: Four important results in the historical devel-
How big is an exciton in Gallium Arsenide (GaAs)? opment of science are consequences of scaling arguments.
For this semiconductor the phenomenological parameters
are  = 12.5, me = 0.07me , mh = 0.4me .
We extend the scaling argument above by comput- k Scaling Law
ing the reduced mass of the electron hole pair: eh = 1 T 2 ' D3 Kepler #3
(0.07me )(0.4me )/(0.07 + 0.4)me = 0.06me and replacing 0 D'T Newton #1
e2 in the expression Eq. (4) for the Bohr radius a0 by +1 z ' t2 Galileo : Rolling Stones
e2 /. The effect is to multiply a0 by 12.5/0.06 = 208. +2 T ' D0 Hooke
The ground state radius of the exciton formed in GaAs
is about 106 cm. The ground state binding energy is
lower than the hydrogen atom binding energy of 13.6 eV
by a factor of 0.06/12.52 = 3.8 104 so it is 5.2 meV .
Scaling arguments such as these are closely related to Z  2 !
renormalization group arguments as presented in Chap- dx
ter XXX. I= m V (x) dt (9)
dt

then the scaling relations x x0 = x, t t0 = t lead


C Dynamical Similarity to a modified action

Jupiter is about five times further (5.2AU) from our 2 !


2
Z 
Sun than the Earth. How many earth years does it take 0 dx k2 2
for Jupiter to orbit the Sun? I = m V (x) dt (10)
dt
Landau and Lifshitz provide an elegant solution to
this simple question using similarity (scaling) arguments. The Action I 0 is proportional to the original Action I,
The equation of motion for the Earth around the Sun is and therefore leads to the same equations of motion,
only when k2 2 = 1. That is, the time elapsed, T ,
d2 xE x
E is proportional to the distance traveled, D, according to
mE = GmE MS (7) T ' D(1k/2) . Four cases are of interest.
dt2E |xE |
2
k=1 (Coulomb/Gravitational Potential) The period
of a planetary orbit scales like the 3/2 power of the dis-
where xE is a vector from the sun to the earth and x E
tance from the Sun (Keplers Third Law).
the unit vector in this direction. If Jupiter is in a geo-
k= 0 (No forces) The distance traveled is proportional
metrically similar orbit, then xJ = xE , with = 5.2.
to the time elapsed (essentially Newtons First Law). To
Similarly, time will evolve along the Jupiter trajectory
recover Newtons first law completely it is only necessary
in a scaled version of its evolution along the Earths tra-
to carry
 out the variation in Eq. (10), which leads to
jectory: tJ = tE . Substituting these scaled expressions d dx
into the equation of motion for Jupiter, and cancelling dt dt = 0.
k=+1 (Free fall in a homogeneous gravitational field)
out mJ from both sides, we find
The potential V (z) = mgz describes free fall in a ho-
mogeneous graviational field. Galileo is reputed to have
d2 x E 1 x
E dropped rocks off the Leaning Tower of Pisa to determine
2 2 = 2 GMS 2 (8) that the distance fallen was proportional to the square of
dtE |xE |
the time elapsed. The story is apocryphal: in fact, he
This scaled equation for Jupiters orbit can only be rolled stones down an inclined plane to arrive at the re-
equated to the equation for the Earths trajectory (the or- sult z ' t2 .
bits are similar) provided 3 / 2 = 1. That is, = 3/2 , k=+2 (Harmonic oscillator potential) The period is
so that the time-scaling factor is 5.23/2 = 12.5. independent of displacement: = 1 independent of .
We have derived Keplers Third Law without even solv- Hookes law, F = kx, V (x) = 21 kx2 leads to oscillatory
ing the equations of motion! Landau and Lifshitz point motion whose frequency is independent of the amplitude
out that you can do even better than that. You dont of motion. This was particularly useful for constructing
even need to know the equations of motion to construct robust clocks.
scaling relations when motion is described by a poten- These four historical results in the development of early
tial V (x) which is homogeneous of degree k. This means science are summarized in Table I and Fig. 2.
that V (x) = k V (x). When the equations of motion websearch: dynamical similarity
are derivable from a Variational Principle I = 0, where
4

Group multiplication has two inputs and one output.


The two inputs must be members of the set. The first
axiom (Closure) requires that the output must also be
a member of the set.
The composition rule does not allow us to multiply
three input arguments. Rather, two can be combined to
one, and that output can be combined with the third.
This can be done in two different ways that preserves
the order (i, j, k). The second axiom (Associativity)
requires that these two different ways give the same final
output.
The third axiom (Identity) requires that a special
group operation exists. This, combined with any opera-
tion, gives back exactly that operation.
The fourth axiom (Unique Inverse) guarantees that
FIG. 2: Four substantial advances in the development for each group element, there is another uniquely defined
of early Physics are summarized. Each is a consequence group element, with the property that the product of the
of using a homogeneous potential with a different degree two is the unique identity element.
k in a variational description of the dynamics. Scaling
relates the size scale of the trajectory to the time scale Remark 1 Indexes: The notation (subscripts
= p , p = 1 12 k of the motion. i, j, k, ) may suggest that the indices are inte-
gers. This is not generally true: for continuous
groups the indices are points in some subspace of
III. GROUPS a Euclidean space or more complicated manifold.
Remark 2 Commutativity: In general the output
In this Section we finally get to the point of defining
of the group multiplication depends on the order
what a group is by stating the group axioms. These are
of the inputs: gi gj 6= gj gi . If the result is
illustrated with a number of examples: the two-element
independent of the order the group is said to be
group, the group of the equilateral triangle, finite groups,
commutative.
the permutation group, discrete groups, point groups and
space groups. Then we introduce groups of transforma- websearch: group theory physics
tions in space as matrix groups. Lie groups are intro- The group operations in some previous examples
duced and examples of matrix Lie groups are presented. (Pythagoras, Scaling, Dynamical Similarity) involve mul-
Lie groups are linearized to form their Lie algebras, and tiplication by one or more real numbers (continuous
groups are recovered from their algebras by reversing the groups). For the dimensional analysis example it was
linearization procedure using the exponential mapping. multiplication by integer powers of fundamental physical
Many of the important properties of Lie algebras are in- constants (discrete group).
troduced, including isomorphisms among different rep- It is not entirely obvious that the Unique Inverse ax-
resentations of a Lie algebra. A powerful disentangling iom is needed. It is included among the axioms because
theorem is presented and illustrated in a very simple case many of our uses involve relating measurements made by
that plays a prominent role in the field of Quantum Op- two observers. For example, if Allyson on the Earth can
tics. We will use this result in the Section on Special predict something about the length of a year on Jupiter,
Functions. then Bob on Jupiter should just as well be able to pre-
dict the length of Allysons year on Earth. Basically,
this axiom is an implementation of Galileos Principle of
A Group Axioms
Relativity.

A group G consists of a set of group elements


g1 , g2 , g3 , G together with an operation , called B Finite Groups
group multiplication that satisfy the following four
axions: 1 The Two-Element Group Z2
Closure: gi G, gj G gi gj G
The simplest nontrivial group has one additional oper-
Associativity: (gi gj ) gk = gi (gj gk ) ation beyond the identity e: G = {e, f } with f f = e.
This group can act in our three-dimensional space R3 in
Identity: g1 gi = gi = gi g1
several different ways:
Unique Inverse: gk gl = g1 = gl gk
5

f =Z
Reflection: (x, y, z) (+x, +y, z) Reflections: There are three reflections i , each in a
straight line through the center of the triangle and
f =RZ () the vertex i (i = 1, 2, 3, c.f. Fig. 3).
Rotation: (x, y, z) (x, y, +z)
f =P These operations can be defined by their action on the
Inversion: (x, y, z) (x, y, z)
vertices of the triangle. For example
These three different actions of the order-two group on
R3 describe: reflections in the x-y plane, Z ; rotations  

0 1 0

around the Z axis through radians, RZ (); and inver- 1 2 3
C3+ 0 0 1 (12)
sion in the origin, the parity operation, P. They can be 2 3 1
1 0 0
distinguished by their matrix representations, which are
The first description (( )) says that the rotation C3+ maps
Z P vertex 1 to vertex 2, 2 3 and 3 1. The second
RZ () description ([ ]) can be understood as follows:
+1 0 0 1 0 0 1 0 0
0 +1 0 0 1 0 0 1 0
0 0 1 0 0 +1 0 0 1
2

0 1 0

1
(11) 3 = 0 0 1 2 (13)
1 1 0 0 3
2 Group of Equilateral Triangle C3v The group multiplication law can be represented through
a 6 6 matrix that describes the output of gi gj , with
gi listed by column and gj by row:

gj e C3+ C3 1 2 3
gi
e e C3+ C3 1 2 3
C3+ C3+ C3 e 2 3 1
(14)
C3 C3 e C3+ 3 1 2
1 1 3 2 e C3 C3+
2 2 1 3 C3+ e C3
3 3 2 1 C3 C3+ e

This table makes clear that the group is not commuta-


tive: C3 = 1 2 6= 2 1 = C3+ .
The partition of the six elements in this group into
three subsets of geometrically equivalent transformations
is typical of any group. These subsets are called classes.
Classes are defined by the condition

FIG. 3: The group of the equilateral triangle consists of: Class : {h1 , h2 , } g hi g 1 = hj all g G (15)
(a) the identify group operation e; (b) two rotations C3
by 23 about the centroid of the triangle; and (c) three
All elements in the same class have essentially the same
reflections i in straight lines between the centroid and properties. They are equivalent under a group transfor-
each of the vertices i. mation.
 The three classes for the finite group C3v are:
{e} , C3+ , C3 , {1 , 2 , 3 }.
The six operations that map the equilateral triangle to
itself constitute the group C3v (c.f. Fig. 3). There are
three distinct types of operations: 3 Cyclic Groups Cn

Identity: e This operation does nothing: it maps each The cyclic group consists of all rotations of the circle
vertex into itself. into itself through the angle 2/n radians, and integer
Rotations: Two rotations C3 about the center of the multiples of this angle. There are n such operations.
triangle through 2/3 radians. The rotation through 2k/n radians is obtained by ap-
plying the smallest rotation (also called Cn ) k times.
This smallest rotation is called a generator of the group.
6

The group is commutative. There are therefore as many In this representation group composition is equivalent to
classes as group elements. The group operations can be matrix multiplication. The group is commutative. The
put in 1 : 1 correspondence with the complex numbers generator for the group and this matrix representation is
and also with real 2 2 matrices: obtained by setting k = 1. There is also an entire set
of 1 1 complex matrix representations with generator
Ta eipa . The representations with p0 = p + 2/a are

cos 2k sin 2k
h i  
11 22 equivalent, so all the inequivalent complex representa-
ei2k/n Cnk n n (16)
sin n cos 2k
2k
n tions can be parameterized by real values of p in the range
0 p < 2/a or, more symmetrically /a p /a,
with k = 0, 1, 2 , n 1 or k = 1, 2, n. The group with the endpoints identified. The real parameter p is
element, 1 1 complex matrix, 2 2 real matrix with in the dual space to the lattice, called the first Brillouin
k = 1 is the generator for the group, the 1 1 matrix zone.
representation, and the 2 2 matrix representation.

2 Translation Groups: 2 Dimensions


4 Permutation Groups Sn

Now imagine a series of lattice points in a plane at


Permutation groups act to interchange things. For ex- positions x = i1 f1 + i2 f2 . Here i1 , i2 are integers and
ample, if we have n numbers 1, 2, 3, n, each permuta- the vectors f1 , f2 are not colinear but otherwise arbitrary.
tion group operation will act to scramble the order of the Then the set of rigid displacements (j1 , j2 ) move lattice
integers differently. Two useful ways to describe elements points x to new locations as per
in the permutation group are shown in Eq. (12) for the
permutation group on three vertices of an equilateral tri-
angle. In the first case, the extension of this notation for Tj1 f1 +j2 f2 (i1 f1 + i2 f2 ) = (i1 + j1 )f1 + (i2 + j2 )f2 (18)
individual group elements consists of a matrix with two
rows, the top showing the ordering before the operation is Generalizing Eq. (17), there is a simple 1 : 1 (or faithful)
applied, the bottom showing the ordering after the group matrix representation for this group of rigid translations:
operation has been applied. In the second case shown in
Eq. (12) the extension consists of n n matrices with
exactly one +1 in each row and each column. The order 1 0 j1 |f1 |
(number of group operations) of Sn is n!. Permutation Tj1 f1 +j2 f2 0 1 j2 |f2 | (19)
groups are noncommutative for n > 2. S3 = C3v . 0 0 1
The permutation group plays a fundamental role in
both mathematics and physics. In mathematics it is used Extension to groups of rigid displacements of lattices in
to label the irreducible tensor representations of all Lie higher dimensions is straightforward.
groups of interest. In physics it is required to distinguish
among different states that many identical particles (ei-
ther bosons or fermions) can assume. 3 Spacegroups
websearch: permutation group, Hamermesh
When |f1 | = |f2 | and the two vectors are orthogonal,
rotations through k/2 (k = 1, 2, 3) radians about any
C Infinite Discrete Groups lattice point map the lattice into itself. So also do re-
flections in lines perpendicular to |f1 | and |f2 | as well as
1 Translation Groups: 1 Dimension lines perpendicular to |f1 ||f2 |. This set of group oper-
ations contains displacements, rotations, and reflections.
It is an example of a two-dimensional space group. There
Imagine a series of points at locations na along the
are many other space groups in two dimensions and very
straight line, where a is a physical parameter with di-
many more in three dimensions. These groups were first
mensions of length ([a] = L) and n is an integer. The
used to enumerate the types of regular lattices that Na-
group that leaves this set invariant consists of rigid dis-
ture allows in two and three dimensions. After the devel-
placements through integer multiples of the fundamental
opment of Quantum Mechanics they were used in another
length. The operation Tka displaces the point at na to
way (depending of the theory of representations): to give
position (n + k)a. This group has a single generator Ta ,
names to wavefunctions that describe electrons (and also
and Tka = Ta Ta Ta = Tak . It is convenient to
phonons) in these crystal lattices.
represent these group operations by 2 2 matrices

 
1 ka
Tka (17)
0 1
7

D Matrix Groups Here L is a set of three 33 angular momentum matrices

1 Translation Groups
0 0 0 0 0 1 0 1 0
Lx = 0 0 1 Ly = 0 0 0 Lz = 1 0 0
The group of rigid translations of points in R3 through 0 1 0 1 0 0 0 0 0
distances a1 in the x-direction, a2 in the y-direction, and (24)
a3 in the z-direction can be described by simple block We will show later how this marvelous expression has
4 4 (4 = 3 + 1) matrices: been derived.
There is a 1:1 correspondence between points in the

1 0 0 a1
interior of a sphere of radius and rotations through
0 1 0 a2 an angle in the range 0 < . Two points on the
Ta1 ,a2 ,a3 (20) sphere surface (
n, ) and (n, ) describe the same rota-
0 0 1 a3
0 0 0 1 tion. The parameter space describing this group is not a
simply connected submanifold of R3 : it is a doubly con-
If the a belong to a lattice the group is discrete. If they nected manifold. The relation between continuous groups
are continuous (a = (x, y, z)) the group is continuous and and their underlying parameter space involves some fas-
has dimension three. cinating topology.

2 Heisenberg Group H3 4 Lorentz Group SO(3, 1)

The Heisenberg group H3 plays a fundamental role The Lorentz group is the group of linear transforma-
in quantum mechanics. As it appears in the quantum tions that leave invariant the square of the distance be-
theory it is described by infinite-dimensional matrices. tween two nearby points in spacetime: (cdt, dx, dy, dz)
However, the group itself is three dimensional. In fact, it and (cdt0 , dx0 , dy 0 , dz 0 ). The distance can be written in
has a simple faithful description in terms of 33 matrices matrix form:
depending on three parameters:

(cd )2 = (cdt)2 (dx2 + dy 2 + dz 2 ) =


1 a c
h(a, b, c) = 0 1 b (21)
0 0 1
+1 0 0 0 cdt

The group is not commutative. The group multiplication   0 1 0 0 dx
cdt dx dy dz (25)
law can be easily worked out via matrix multiplication: 0 0 1 0 dy
0 0 0 1 dz

1 a1 + a2 c2 + a1 b2 If the infinitesimals in the primed coordinate system are
h1 h2 = h3 = h(a3 , b3 , c3 ) = 0 1 b1 + b2 related to those in the unprimed coordinate system by
0
0 0 1 a linear transformation dx = M dx then the
(22) matrices M must satisfy the constraint (t means matrix
The result c3 = a1 b2 +c2 leads to remarkable noncommu- transpose)
tativity properties among canonically conjugate variables
in the quantum theory: [p, x] = ~/i.
M t I1,3 M = I1,3 (26)

3 Rotation Group SO(3) where I1,3 is the diagonal matrix diag(+1, 1, 1, 1).
The matrices M belong to the orthogonal group O(1, 3).
The set of rigid rotations of R3 into itself forms a This is a six-parameter group. Clearly the rotations
group. It is conveniently represented by a faithful 3 3 (three dimensions worth) form a subgroup,
 represented
matrix. The 3 3 matrix describing rotations about an 1 0
by matrices of the form , where R(n, )
through an angle , 0 is
axis of unit length n 0 R( n, )
is given in Eq.(23). This group has four disconnected
components, each   to a 4 4 matrix of the
 connected
n
1

1 0 1 0 1 0 1 0
3 (1cos )form 0 I3 , 0 I3 , , . We
 
n, ) I3 cos +
( n L sin + n
2 n1 n
2 n 0 I3 0 I3
n
3 choose the component connected to the identity I4 . This
(23)
8

is the special Lorentz group SO(1, 3). A general matrix


in this group can be written in the form   
1 1 1 2 2 2
B(1 )B(2 ) = =
1 1 1 2 2 2
SO(1, 3) = B()R() (27)   (31)
tot tot tot
where the matrices B() describe boost transformations. tot tot tot
A boost transformation maps a coordinate system at rest
to a coordinate moving with velocity v = c and with 1 +2
Simple matrix multiplication shows tot = 1+ 1 2
, which
axes parallel to the stationary coordinate system. We is the relativistic velocity addition formula for parallel
will describe these transformations in more detail below. velocity transformations.
Since every group operation in SO(1, 3) can be ex- When the boosts are not parallel, their product is a
pressed as the product of a rotation operation with a transformation in SO(1, 3) that can be written as the
boost, we can formally write B() = SO(1, 3)/SO(3). product of a boost with a rotation:
In fact, such an expression makes sense in this case. A
similar result makes sense for any group G and subgroup
H G. Every element g G can be expressed as B( 1 )B( 2 ) = B( tot )R() (32)
the product of an h H and another group operation
c G. In fact, it is a theorem that the set c belongs to Multiplying two boost matrices of the form given in
a small subset C G called a coset. For finite groups, Eq. (29) leads to a simple expression for tot and a more
|C| = |G|/|H|. That is, the order, or number of elements complicated expression for tot
in C, is equal to the number of elements in G divided by
the number of elements in the subgroup H. For infinite
and continuous groups something similar holds (in the tot = h 1 2 (1 + 1 2 )
(33)
i
1 2
sense of dimensions or measures). Such a decomposition, tot tot = 1 2 + (1 1)2 1 + 2 2
1 1
generally written
This shows what is intuitively obvious: the boost direc-
tion is in the plane of the two boosts. Less obvious is the
Coset Decomposition G = (G/H)H = C H (28) rotation required by noncollinear boosts. It is around an
axis parallel to the crossproduct of the two boosts. When
is called a coset decomposition. In general, the set of the two boosts are perpendicular the result is
group elements in C do not satisfy the axioms for a
group. Not all coset decompositions carry as much phys-
ical sense as the boost - rotation decomposition of the 1 2
sin() = 1 2
n (34)
Lorentz group. 1 + 1 2
A general boost transformation can be written in the
When one of the boosts is infinitesimal we find
form

"

# B()B() = B( + d)R(
nd) (35)
B() = i j (29)
I3 + ( 1) Multiplying out these matrices and comparing the two
sides gives:
For example, a boost with v/c = (, 0, 0) has the follow-
ing effect on coordinates:
1 1
 
d = 1 + ( )
0 2
ct 0 0 ct (ct + x)

(36)
x 0 x (x + ct)
0 1 1
 
y = =

0 0 1 y
0 y n
d =
2
z 0 0 0 1 z z
(30) In the nonrelativistic limit, when is also small, 1
1 / 2 12 . This (in)famous factor of 1/2 is known as
p
Here 2 ()2 = 1 so = 1/ 1 2 . In the non-
relativistic limit x0 = (x + ct) x + vt, so has an the Thomas factor in atomic physics.
interpretation of = v/c. websearch: Lorentz group, boost, rotation
The product of two boosts in the same direction is websearch: Thomas precession
obtained by matrix multiplication. This can be carried
out on a 2 2 submatrix of that given in Eq. (30):
9

E Lie Groups O(p,q), U(p,q), Sp(p,q): These groups are defined


by replacing In in the
 definitions
 for O(n), U (n), Sp(n) by
The group operations g in a Lie group are parame- +Ip 0
the matrix Ip,q = . These groups preserve
terized by points x in a manifold Mn of dimension n: 0 Iq
g = g(x), x Mn . The product of two group opera- an indefinite nonsingular metric in linear vector spaces
tions g(x) and g(y) is parameterized by a point z in the of dimension (p + q). The groups O(n), U (n), Sp(n) are
manifold: g(x) g(y) = g(z), where z = z(x, y). This compact (a useful topological concept) and so are rel-
composition law can be very complicated. It is necessar- atively easy to deal with. This means effectively that
ily nonlinear unless the group is commutative (c.f., Eq. only a finite volume of parameter space is required to
(22) for H3 ). parameterize every element in the group. The groups
Almost all of the Lie groups of use to physicists exist O(p, q), U (p, q), Sp(p, q) are not compact if both p and
as matrix groups. For this reason it is possible for us to q are nonzero. Further, O(p, q) ' O(q, p) by a simple
skip over the fundamental details of whether the com- similarity transformation, and similarly for the others.
position law must be analytic and the elegant details of websearch: Lie groups of matrices
their definition and derivations. We list several types of
matrix groups below.
GL(n;R), GL(n;C), GL(n;Q): These groups consist F Lie Algebras
of n n matrices, each of whose n2 matrix elements are
real numbers, complex numbers, or quaternions. In order Lie algebras are constructed for a Lie group by lineariz-
for inverses to exist, noninvertible matrices do not occur ing the Lie group in the neighborhood of the identity e.
in these matrix groups. The number of real parameters Matrix Lie algebras are obtained for n n matrix Lie
required to specify an element in these groups is: n2 , 2 groups by linearizing the matrix group in the neighbor-
n2 , 4 n2 . hood of the unit matrix In . A Lie group and its Lie
SL(n;R), SL(n;C): These groups are subgroups of algebra have the same dimension.
GL(n; R) and GL(n; C) containing the subset of matri- In the neighborhood of the identity the groups
ces with determinant +1. The real dimensions of these GL(n; R), GL(n; C), GL(n, Q) have the form
groups are (n2 1) dim(F ) where dim(F ) = (1, 2) for
F = (R, C).
O(n), U(n), Sp(n): Three important classes of GL(n; F ) In + M (37)
groups are defined by placing quadratic constraints on
matrices. The orthogonal group O(n) is the subgroup where M is an nn matrix, all of whose matrix elements
of GL(n; R) containing only matrices M that satisfy are small. Over the real, complex, and quaternion fields
M t In M = In . Here In is the unit n n matrix and t sig- the matrix elements are small real or complex numbers
nifies the transpose of the matrix. This constraint arises or small quaternions. Quaternions q can be expressed as
in a natural way when requiring that linear transforma- 2 2 complex matrices using the Pauli spin matrices :
tions in a real n-dimensional linear vector space preserve
a positive definite inner product. The unitary group 3
X
U (n) is the subgroup of GL(n; C) for which the matrices q (c0 , c1 ) = (r0 , r1 , r2 , r3 ) = r =
M satisfy M In M = In , where signifies the adjoint, =0
or complex conjugate transpose matrix. The symplectic
group Sp(n) is defined similarly for the quaternions. In
this case signifies quaternion conjugate transpose. The
 
r0 + ir3 r1 ir2
real dimensions of these groups are n(n 1)/2, n2 , and (38)
r1 + ir2 r0 ir3
n(2n + 1), respectively.
SO(n), SU(n): The S stands for special and special The Lie algebras gl(n; F ) of GL(n; F ) have dimensions
means the determinant is +1. For the group O(n) the de- f n2 , f = 1, 2, 4 for F = R, C, Q.
terminant of any group operation is a real number whose For the special linear groups, the determinant is
modulus is +1: i.e., 1. Placing the special constraint
on the group of orthogonal transformations reduces the
number of elements in the group by one half (in a mea- det(In + M ) = 1 + trM + h.o.t. (39)
sure theoretic sense) but does not reduce the dimension
of the space required to parameterize the elements in this The Lie algebras are defined by the traceless condition.
group. For the group U (n) the determinant of any group The Lie algebra sl(n; R) of SL(n; R) consists of real trace-
operation is a complex number whose modulus is +1: i.e., less n n matrices. It has dimension n2 1. The Lie
ei . Placing the special constraint on U (n) reduces the algebra sl(n; C) of SL(n; C) consists of traceless complex
dimension by one: dim SU (n) = n2 1. The symplectic n n matrices. It has real dimension 2n2 2.
group Sp(n) has determinant +1. Many Lie groups are defined by a metric-preserving
condition: M GM = G, where G = is some suitable
10

metric matrix (see Sec. III E). The linearization of this Two Lie algebras with the same set of structure con-
condition is stants are isomorphic. The Lie algebra of 2 2 matrices
obtained from su(2) is spanned by three operators that
can be chosen as proportional ( 2i ) to the Pauli spin ma-
M GM (In + M ) G(In + M ) = G + M G + GM trices (c.f., Eq. (38)):

     
i 0 1 i 0 i
1 0 i

+h.o.t = G M G + GM = 0 (40) S1 = S2 = S3 =
2 1 0 2 0 1
+i 0 2
(44)
When G is the identify matrix, the lie algebras
These three operators satisfy the commutation relations
so(n; R), su(n; C), sp(n; Q) consists of real antisymmet-
ric matrices M t = M , complex traceless antihermitian
matrices M = M , and quaternion antihermitian ma- [Si , Sj ] = ijk Sk (45)
trices M = M , respectively.
The elements in a Lie group must obey one single group As a result, the Lie algebra for the group so(3) of ro-
multiplication law. The elements in the groups Lie al- tations in R3 is isomorphic to the Lie algebra su(2) for
gebra have simpler properties, as they are obtained by the group of unimodular metric-preserving rotations in
linearization about the identity group operation. In or- a complex two dimnensional space, SU (2). Spin and or-
der to reflect the nonlinear group multiplication proper- bital rotations are intimately connected.
k
ties they must satisfy more conditions. Construction by The notation for the structure constants Cij for a Lie
linearization means that the Lie algebra is a linear vec- algebra gives the appearance of being components of a
tor space. If X and Y are elements in a Lie algebra, tensor. In fact, they are: the tensor is first order con-
X + Y is also in the Lie algebra. Here and are travariant (in k) and second order covariant, and anti-
arbitrary real scalars. Closure under multiplication in symmetric, in i, j. It is possible to form a second order
the group leads to the requirement that the commuta- covariant tensor (Cartan-Killing metric) from the com-
tor [X, Y ] = XY Y X = [Y, X] is also an element in ponents of the structure constant by double contraction:
the Lie algebra. There is only one other condition that
the elements in a Lie algebra are required to satisfy: the X
s r
Jacobi identity: gij = Cir Cjs = gji (46)
rs

This real symmetric tensor looks like a metric tensor.


[X, [Y, Z]] + [Y, [Z, X]] + [Z, [X, Y ]] = 0 (41) In fact, it has very powerful properties. If g is non-
singular the Lie algebra, and its Lie group, is simple
For matrix Lie algebras this relation is indeed an identity,
or semisimple. If g is negative definite, the group is
but for abstract Lie algebras where composition XY is
compact. It is quite remarkable that an algebraic struc-
not naturally defined, the commutator [X, Y ] is always
ture gives such powerful topological information.
defined, and the identity is satisfied.
As an example, for SO(3) and SU (2) the Cartan-
The Lie algebra so(3) of the rotation group SO(3) =
Killing metric Eq. (46) is
SO(3; R) consists of real 3 3 antisymmetric matrices.
This group and its algebra are three dimensional. The Lie
algebra (it is a linear vector space) is spanned by three
X
gij = (irs )(jsr ) = ij (47)
basis vectors. These are 3 3 antisymmetric matrices. r,s
A standard choice for these basis vectors is given in Eq.
(24). Their commutation relations are given by For the real forms SO(2, 1) of SO(3) and SU (1, 1) of
SU (2) the Cartan-Killing metric tensor is

[Li , Lj ] = ijk Lk (42)


+1 0 0
More generally, for an arbitrary n-dimensional Lie alge- g ((so(2, 1), su(1, 1)) = 0 +1 0 (48)
bra it is possible to choose a set of n basis vectors (ma- 0 0 1
trices, operators) Xi . The commutation relations are en-
capsulated by a set of structure constants Cij k
that are The structure of this metric tensor (two positive diagonal
defined by elements or eigenvalues, and one negative) tells us about
the topology of the groups: they have two noncompact di-
rections and one compact direction. The compact direc-
k tion describes the compact subgroups SO(2) and U (1),
[Xi , Xj ] = Cij Xk (43)
respectively.
k
The structure constants for so(3) are Cij = ijk , 1 websearch: Lie groups Lie algebras
i, j, k 3. websearch: EXPonential map Lie algebras
11

G Operator Realizations of Lie Algebras



0 l d
Each Lie algebra has three useful operator realizations. 0 0 r = lL + rR + dD (53)
They are given in terms of boson operators, fermion op- 0 0 0
erators, and differential operators.
Boson annihilation operators bi and creation operators Here L, R, D are 3 3 matrices. The only nonzero com-
bj for independent modes i, j = 1, 2, , their fermion mutator is [L, R] = D. The corresponding differential
counterparts fi , fj , and the operators i , xj satisfy the operator algebra is
following commutation or anticommutation relations

 0 l d
 x
h i x y z 0 0 r y = lL + rR + dD (54)
bi , bj
= bi bj bj bi = ij 0 0 0 z
n o
fi , fj = fi fj + fj fi = ij (49)
The three differential operators are
[i , xj ] = i xj xj i = ij

In spite of the fact that bosons and differential opera- L = xy R = yz D = xz (55)


tors satisfy commutation relations and fermion operators
satisfy anticommutation (see the + sign in Eq. (49)) re- Among these operators: none depends on z (so z has
lations, bilinear combinations Zij = bi bj , fi fj , xi j of nothing to operate on) and none contains x , so that in
these operators satisfy commutation relations: essence x is an irrelevant variable. A more economical
representation of this algebra is obtained by zeroing out
the cyclic variables z, x and replacing their conjugate
[Zij , Zrs ] = Zis jr Zrj si (50) variables z , x by +1.

These commutation relations can be used to associate


operator algebras to matrix Lie algebras. The procedure 0 l d 0
1 y 0 0 0 r y = lL0 + rR0 + dD0
 
is simple. We illustrate for boson operators. Assume (56)
A, B, C = [A, B] are n n matrices in a matrix Lie alge- 0 0 0 1
bra. Associate operator A to matrix A by means of
L0 = y R0 = y D0 = 1 (57)
A A = bi Aij bj (51) In essence, we have zeroed out the operators coupled to
the vanishing rows and columns of the matrix Lie alge-
and similarly for other matrices. Then bra and replaced their conjugate variables by 1. The
representation given is essentially the Hiesenberg repre-
h i sentation of the position (y) and conjugate momentum
[A, B] = bi Aij bj , br Brs bs = bi [A, B]is bs = bi Cis bs = C (py ' y ) operators in Quantum Mechanics.
(52) websearch: Lie algebra operator algebra mapping
This result holds if the bilinear combinations of boson
creation and annihilation operators are replaced by bi-
linear combinations of fermion creation and annihilation H Exponentiation
operators or products of multiplication (by xi ) and dif-
ferentiation (by j ) operators. The mapping of a Lie group, with a complicated non-
One consequence of this matrix Lie algebra to operator linear composition, down to a Lie algebra with a simple
algebra isomorphism is that any Hamiltonian that can be linear combinatorial structure plus a commutator, would
expressed in terms of bilinear products of creation and not be so useful if it were not possible to undo this map-
annihilation operators for either bosons or fermions can ping. In effect, the linearization is undone by the ex-
be studied in a simpler matrix form. ponential map. For an operator X the exponential is
The operator algebra constructed from the spin op- defined in the usual way:
erators in Eq. (44) has been used by Schwinger for an
elegant construction of all the irreducible representations
of the Lie group SU (2) (c.f., Sec. V E and Fig. 4). X2 X3 X Xk
EXP (X) = eX = I +X + + + = (58)
We use the matrix-to-operator mapping now to con- 2! 3! k!
k=0
struct a differential operator realization of the Heisenberg
group, given in Eq. (21). Linearizing about the identify The radius of convergence of the exponential function is
gives a three-dimensional Lie algebra of the form infinite. This means that we can map a Lie algebra back
to its parent Lie group in an algorithmic way.
12

We illustrate with two important examples. For the give the same group operation in SO(3) but they de-
first, we construct a simple parameterization of the group scribe group operations in SU (2) that differ by sign:
SU (2) by exponentiating its Lie algebra. The Lie alge- z, 2 + ) = I2 (
( z, ). In short, two group operations
bra is given in Eq. (44). Define M = 2i n . Then M and M in SU (2) map to the same group opration
M 2 = (/2)2 I2 is a diagonal matrix. The exponential in SO(3). In words, SU (2) is a double cover of SO(3).
expansion can be rearranged to contain even powers in For SU (2) all points inside a sphere of radius 2 in the
one sum and odd powers in another: Lie albegra map to different group operations, and all
points on the sphere surface map to one group operation
I2 . The group SU (2) is simply connected. Any path
(/2)2 (/2)4
 
e M
= I2 1 + + starting and ending at the same point (for example, the
2! 4! identity) can be continuously contracted to the identity.
By contrast, for SO(3) all points inside a sphere of
radius in the Lie algebra map to different group opera-
(/2)2 (/2)4
 
M 1 + (59) tions, and two points ( n, ) and (n, ) at opposite ends
3! 5! of a straight line through the origin map to the same
group operation. The group SO(3) is not simply con-
The even terms sum to cos(/2) and the odd terms sum nected. Any closed path from the origin that cuts the
to sin(/2)/(/2). The result is surface = once (or an odd number of times) can-
not be continuously deformed to the identity. The group

i

SO(3) is doubly connected.
EXP
n n sin
= cos I2 + i (60) This is the simplest example of a strong theorem by
2 2 2 Cartan. There is a 1 : 1 relation between Lie algebras and
A similar power series expansion involving the angular simply connected Lie groups. Every Lie group with the
momentum matrices in Eq. (24) leads to the parameteri- same (isomorphic) Lie algebra is either simply connected
zation of the rotation group operations given in Eq. (23). or else the quotient (coset) of the simply connected Lie
Specifically, EXP ( n L) = group by a discrete invariant subgroup.
For matrix Lie groups, discrete invariant subgroups
consist of scalar multiples of the unit matrix. For
 n
1  the the isomorphic Lie algebras su(2) = so(3) the Lie
L sin + n
I3 cos + n 1 n
2 n3 n 2 (1 cos ) group SU (2) is simply connected. Its discrete invari-
n
3 ant subgroup consists of multiples of the identity ma-
(61) trix: {I2 , I2 }. Cartans theorem states SO(3) =
The Lie groups SO(3) and SU (2) possess isomorphic SU (2)/ {I2 , I2 }. This makes explicit the 2 1 nature
Lie algebras. The Lie algebra is three-dimensional. The of the relation between SU (2) and SO(3).
basis vectors in so(3) can be chosen as the angular mo- The group SU (3) is simply  connected. It discrete in-
mentum matrices given in Eq. (24) and the basis vectors variant subgroup consists of I3 , I3 , 2 I3 , with 3 = 1.
for su(2) as i/2 times the Pauli spin matrices, as in Eq. The only other Lie group  with the Lie algebra su(3) is
(44). A point in the Lie algebra (e.g., R3 ) can be iden- the 3 1 image SU (3)/ I3 , I3 , 2 I3 . This group has
tified by a unit vector n and a radial distance from the a description in terms of real eight dimensional matrices
origin . Under exponentiation, the point ( n, ) maps to (the eightfold way).
the group operation given in Eq. (61) for SO(3) and in websearch: EXP Lie algebra Lie group
Eq. (60) for SU (2).
The simplest way to explore how the Lie algebra pa-
rameterizes the two groups is to look at how points along I Riemannian Symmetric Spaces
a straight line through the origin of the Lie algebra map
to operations in the two groups. For simplicity we choose The Lie algebra for the Lorentz group consists of 4 4
the z-axis. Then ( z, ) maps to matrices:


cos sin 0 0 1 2 3
 
i/2
e 0
SU (2), sin cos 0 SO(3) 0 3 2
0 ei/2 so(1, 3) = 1 (63)
0 0 1 2 3 0 1
(62) 3 2 1 0
As increases from 0 to 2 the SU (2) group operation
varies from +I2 to I2 while the SO(3) group opera- The Cartan-Killing metric for SO(1, 3) is
tion starts at I3 and returns to +I3 . The SU (2) group
operation returns to the identity +I2 only after in-
creases from 2 to 4. The rotations by and + 2 g(so(1, 3), so(1, 3)) = 12 + 22 + 32 12 22 32 (64)
13

The subalgebra of rotations so(3) describes the compact 4. Compute this result in the faithful matrix represen-
subgroup SO(3). The remaining three infinitesimal gen- tation.
erators parameterized by i span the noncompact part of 5. Lift this result back to the operator form.
this group, the coset SO(1, 3)/SO(3), and exponentiate We now show how this algorithm works. The opera-
to boost operations. tors x and x have one nonzero commutator [x , x] = 1.
Cartan has pointed out that it is often possible to de- These three operators close under commutation. They
compose the Lie algebra g for Lie group G into two com- therefore form a Lie algebra. This is the algebra h3 of
plementary subspaces: k, which exponentiates to a sub- the Heisenberg group, Eq. (57). We also have a faithful
group K G; and a complementary subspace p, where matrix representation of this Lie algebra, given in Eq.
the two subspaces satisfy these commutation relations: (56). We then make the identification


[k, k] k [k, p] p [p, p] k (65) 0 b 0 1 b ab
2
ax+bx
e EXP 0 0 a = 0 1 a (66)
When this is possible, EXP (p) = P = G/K. Further, if 0 0 0 0 0 1
the Cartan-Killing metric is negative definite on k, and
positive definite on p, then K is a maximal compact sub- Now we identify this matrix with
group of G and the coset P = G/K is a Riemannian
symmetric space. For the Lorentz group SO(1, 3), by
1 0 0 1 0 d 1 l 0
Cartans criterion SO(3) is the maximal compact sub-
erx edI elx 0 1 r 0 1 0 0 1 0 (67)
group and the coset of boost transformations B() =
0 0 1 0 0 1 0 0 1
SO(1, 3)/SO(3) is a three-dimensional Riemannian space
with positive-definite metric. In this case the space is a By multiplying out the three matrices in Eq. (67) and
3-hyperboloid (ct)2 x2 y 2 z 2 = cst. embedded in comparing with the matrix elements of the 3 3 matrix
R4 . The metric on this space is obtained by moving the in Eq. (66) we learn that l = b, r = a, d = ab 2 . Porting
metric (1, 1, 1) at the origin (x, y, z) = (0, 0, 0) over the the results of this matrix calculation back to the land of
embedded space using the set of Lorenz group transfor- operator algebras, we find
mations in the quotient space SO(1, 3)/SO(3). Cartan
also showed that all Riemannian symmetric spaces arise
ab
as quotients of (simple) Lie groups by maximal compact eax+bx = eax e 2 ebx (68)
subgroups.
websearch: Riemannian symmetric spaces, Car- We will use this expression in Sec. IX G below to con-
tan struct a generating function for the Hermite polynomials.
websearch: Lie group disentangling theorems

J Disentangling Results
IV. APPLICATIONS IN CLASSICAL PHYSICS
It happens surprisingly often in distantly related fields
of physics that expressions of the form ex+x are encoun- Group Theorys first important role in physics came
tered. Needless to say, these are not necessarily endear- even before Quantum Mechanics was discovered. The
ing to work with. One approach to simplifying computa- two pillars of classical deterministic physics are Mechan-
tions involving such operators is to rewrite the operator ics and Electrodynamics. Group Theory played a fun-
in such a way that all differential operators x act first, damental role in rectifying the difficulties in describing
and all multiplications by x act last. One way to effect the interactions between these two fields. The principle
this decomposition is to cross ones fingers and write this tool used, besides Group Theory, was Galileos Princi-
operator as eax+bx ' eax ebx and hope for the best. Of ple of Relativity and an assumption about the undrlying
course this doesnt work, since the operators x and x do elegance of physical theories.
not commute.
Exponential operator rearrangements are called disen-
tangling theorems. Since the exponential mapping is in- A Principle of Relativity
volved, powerful methods are available when the opera-
tors in the exponential belong to a finite-dimensional Lie The Principle of Relativity posits: Two observers, S
algebra. Here is the algorithm: and S 0 , observe the same physical system. Each knows
1. Determine the Lie algebra. how his coordinate system is related to the others
2. Find a faithful finite-dimensional matrix represen- that is, the transformation of coordinates that maps one
tation of this Lie algebra. coordinate system into the other. Assume both observers
3. Identify how you want the operators ordered in the collect data on the same physical system. Given the data
final product of exponentials. that S takes, and the coordinate transformation from S
14

to S 0 , S can predict the data that S 0 has recorded. And determine how fast the earth was moving in the Universe.
he will be correct. As ever more precise measurements in the late nineteenth
Essentially, without this ability to communicate data and early twentieth century lead to greater disappoint-
among observers, there would be little point in pursuing ment, more and more bizarre explanations were created
the scientific method. to explain this null result. Finally Einstein and Poincare
A second assumption that is used is usually not stated returned to the culprit Eq.(72) and asserted what the ex-
explicitly. This assumption is: The quantitative formu- periments showed: c0 = c, so that
lation of physical laws is simple and elegant (whatever
that means).
websearch: Galileo relativity (c dt)0 c dt0 dt0 = linear comb. dx, dy, dz, dt (73)

The condition that the distance function


B Mechanics and Electrodynamics
(c d )2 = (c dt)2 (dx2 + dy 2 + dz 2 ) =
(74)
The quantitative formulation of Mechanics for a single (c dt0 )2 (dx02 + dy 02 + dz 02 )
particle in an inertial frame is
is invariant leads directly to the transformation law for
infinitesimals dx0 = x , where the 4 4 matrices be-
dp 0
long to the Lorentz group SO(3, 1), = x /x .
= F(x) (69)
dt The transformation laws taking inertial frames S to iner-
tial frames S 0 involves inhomogeneous coordinate trans-
where p is defined by p = m dx dt . The transformations formations
from one inertial coordinate system to another consist of:
displacements in space d, displacements in time d, Rigid
x0
    
rotations R, and boosts with constant velocity v that SO(3, 1) d x
keep the axis parallel. The space and time coordinates = (75)
1 0 1 1
in S 0 are related to those in S by
While Maxwells equations remain unchanged in form
0
under this set of coordinate transformations (inhomoge-
x = Rx + vt + d neous Lorentz group), Newtons force law no longer pre-
(70)
t0 = t + d serves its form.
In order to find the proper form for the laws of classi-
In inertial coordinate system S 0 Newtons equations are
cal mechanics under this new set of transformations the
following two-step process was adopted:
dp0 x0 = Rx
= F0 (x0 ) p0 = Rp (71) 1. Find an equation that has the proper transforma-
dt F0 = RF tion properties under the inhomogeneous Lorentz
group;
The equations of motion have the same vectorial form in
both inertial coordinate systems (the simple and elegant 2. If the equation reduces to Newtons equation of mo-
assumption). tion in the nonrelativistic limit 0 it is the
Newtons Laws were incredibly successful in describing proper generalization of Newtons equation of mo-
planetary motion in our solar system, so when Maxwell tion.
developed his laws for electrodynamics, it was natural to Appliction of the procedure leads to the relativistic equa-
assume that they also retained their form under the set tion for particle motion
of inertial coordinate transformations given in Eq. (70).
Applying these transformations to Maxwells equations
creates a big mess. But if one looks only at signals prop- dp
= f (76)
agating along or opposite the direction of the velocity v, d
these assumptions predict
where p is defined by p = m dx d . The components of
the relativistic four vector f are related to the three-
(c dt)0 c0 dt c0 = c |v| (72) vector force F by

The round trip time for a light signal in S is 2L/c while in


2L/c F
S 0 the time lapse was predicted to be c+vL L
+ cv = 1 2,
f = F + ()
(77)

with = |v|/c. f 0 = F
This predicted difference in elapsed round trip times
in rest and moving frames was thought to enable us to (c.f., Eq. (29)).
15

C Gravitation
0 0 0 0
!
d2 x
x d2 x

Einstein wondered how it could be possible to deter- 0
dx dx dx dx
2
+, = + ,
mine if you were in an inertial frame. He decided that the d d d x d 2 d d
algorithm for responding to this question, Newtons First (80)
Law (In an inertial frame, an object at rest remains at This means that the set of terms on the left, or those
rest and an object in motion remains in motion with the within the brackets on the right, have the simple transfor-
same velocity unless acted upon by external forces.) was mation properties of a four-vector. In a freely falling co-
circularly defined (How do you know there are no forces? ordinate system the Christoffel symbols vanish and what
2
When you are sufficiently far away from the fixed stars. remains is ddx2 . This Special Relativity expression is zero
How do you know you are sufficiently far away? When in the absence of forces, so the equation that describes
there are no forces.) the trajectory of a particle in a gravitational field is
He therefore set out to formulate the laws of mechanics
in such a way that they were invariant in form under an
arbitrary coordinate transformation. While the Lorentz d2 x dx dx
+ , =0 (81)
group is six dimensional, general coordinate transforma- d 2 d d
tions form an infinite dimensional group. The transfor-
mation hproperties i at any point are defined by a Jacobian
0 D Reflections
matrix x x (x) . Whereas for the Lorentz this matrix is
constant throughout space, for general coordinate trans- Two lines of reasoning have entered the reconciliation
formations this 4 4 matrix is position dependent. of the two pillars of classical deterministic physics and
Nevertheless, he was able to modify the algorithm de- the creation of a theory of gravitation. One is group the-
scribed above to formulate laws that are invariant under ory and is motivated by Galileos Principle of Relativity.
all coordinate transformations. This two-step process is The other is more vague. It is a Principle of Elegance:
a powerful formulation of the Equivalence Principle. It is there is the mysterious assumption that the structure of
called the Principle of General Covariance. It states that the real equations of physics are simple, elegant, and
a law of physics holds in the presence of a gravitational invariant under a certain class of coordinate transforma-
field provided: tions. The groups are the 10 parameter Inhomogeneous
Lorentz Group in the case of the Special Theory of Rel-
1. The equation is invariant in form under an arbitrary
ativity and the much larger groups of general cordinate
coordinate transformation x x0 (x).
transformations in the case of the General Theory of Rel-
2. In a locally free-falling coordinate system, or the ativity. There is every liklihood that Martians from an-
absence of a gravitational field, the equation as- other galaxy will recognize the Principle of Relativity but
sumes the form of a law within the Special Theory no guarantee that their sense of simplicity and elegance
of Relativity. will be anything like our own.

Using these arguments, he was able to show that the


equation describing the trajectory of a particle is V. LINEAR REPRESENTATIONS

The theory of representations of groups more pre-


d2 x
dx dx

= , (78) cisely the linear representations of groups by matrices
d 2 d d was actively studied by mathematicians while physicists
The Christoffel symbols are defined in terms of the metric actively ignored these results. This picture changed dra-
tensor g , (x) by matically with the development of the Quantum Theory,
the understanding that the appropriate phase space
  was the Hilbert space describing a quantum system, and
1 g, g, g, that group operations acted in these spaces through their
, = g , + (79)
2 x x x linear matrix representations.
A linear matrix representation is a mapping of group
They are not components of a tensor, as coordinate elements g to matrices g (g) that preserves the group
systems can be found (freely falling, as in an elevator) in operation:
which they vanish and the metric tensor reduces to its
form in Special Relativity form g = diag(1, 1, 1, 1).
Neither the left hand side nor the right hand side of gi gj = gk (gi ) (gj ) = (gk ) (82)
Eq. (78) is invariant under arbitrary coordinate changes
(extra terms creep in), but the following transformation Here is the composition in the group and indicates
law is valid: matrix multiplication. Often the mapping is one-way:
16

many different group operations can map to the same reasonable that the number of matrix elements in all the
matrix. If the mapping is 1 : 1 the representation is irreducible representations of a group provide a set of ba-
called faithful. sis functions for the function space defined on the set of
group operations. This is true: it is a powerful theorem.
There is a far-reaching duality between the elements in
A Maps to Matrices a group and the set of matrix elements in its set of irre-
ducible representations. Therefore if (g), = 1, 2,
We have already seen many matrix representations. are the irreducible representations of a group G and the
We have seen representations of the two-element group dimension of (g) is d (i.e., (g) consists of d d
Z2 as reflection, rotation, and inversion matrices acting matrices), then the total number of matrix elements is
in R3 (c.f. Eq. (11)). the order of the group G:
So: how many representations does a group have? It
is clear from the example of Z2 that we can create an allX
irreps
infinite number of representations. However, if we squint d2 = |G| (85)
carefully at the three representations presented in Eq.
(11) we see that all these representations are diagonal:
direct sums of essentially two distinct representations:
q
d
Further, the set of functions |G| rs (g) form a complete
orthonormal set of functions on the group space. The
Z2 e f orthogonality relation is
1 1 1 (83)
2 1 1 s s
X d0 0 d
Each of the three matrix representations of Z2 in Eq. (11) 0 0 (g) (g) = (0 , )(r0 s0 , rs)
|G| r s |G| rs
is a direct sum of these two irreducible representations: gG
(86)
and the completeness relation is
Z = 1 1 2
RZ () = 1 2 2 (84) s s
P = 2 2 2 XX d 0 d
(g ) (g) = (g 0 , g) (87)
|G| rs |G| rs
A basic result of representation theory is that for large rs
classes of groups (finite, discrete, compact Lie groups)
These complicated expressions can be considerably
every representation can be written as a direct sum of
simplified when written in the Dirac notation. Define
the special class of representations called irreducible rep-
resentations. The procedure for constructing this direct
sum proceeds by matrix diagonalization. Irreducible rep-
s s
d d
resentations are those that cannot be further diagonal- hg| i= (g), h |gi = (g) (88)
ized. In particular, one dimensional representations can-
rs |G| rs rs |G| rs
not be further diagonalized. Rather than enumerating
all possible representations of a group, it is sufficient to For convenience, we have assumed that the irreducible
enumerate only the much smaller set of irreducible rep- representations are unitary: (g) = (g 1 ) and =t .
resentations. In Dirac notation, the orthogonality and completeness
websearch: group representations relations are

0 0
B Group Element - Matrix Element Duality Orthogonality : h 0 0 |gihg| i = h 0 0| i
rs rs r s rs
(89)

Completeness : hg 0 | ih |gi = hg 0 |gi
The members of a group can be treated as a set of rs rs
points. It then becomes possible to define a set of func-
tions on this set of points. How many independent func- As usual, doubled dummy indices are summed over.
tions are needed to span this function space? A not too
particularly convenient choice of basis functions are the
delta functions fi (g) = (g, gi ). For example, for C3v C Classes and Characters
there are six group operations and therefore six basis
functions for the linear vector space of functions defined The group element - matrix element duality is elegant
on this group. and powerful. It leads to yet another duality, somewhat
Each matrix element in any representation is a func- less elegant but, in compensation, even more powerful.
tion defined on the members of a group. It would seem This is the character - class duality.
17

We have already encountered classes in Eq. (15). Two


elements c1 , c2 are in the same class if there is a group  +
operation, g, for which gc1 g 1 = c2 . The character of a {e} C3 , C3 {1 , 2 , 3 }
matrix is its trace. All elements in the same class have 1 2 3
the same character in any representation, for 1 1 1 1 (94)
2 1 1 1
3 2 1 0
Tr (c2 ) = Tr (gc1 g 1 ) = Tr (g)(c1 )(g 1 ) = Tr (c1 )
(90) The first line shows how the group operations are appor-
The last result comes from invariance of the trace under tioned to the three classes. The second shows the number
cyclic permutation of the argument matrices. of group operations in each class. The remaining lines
With relatively little work, the powerful orthogonal- show the trace of the matrix representatives of the ele-
ity and completeness relations for the group elements - ments in each class in each representation. For example,
matrix elements can be transformed to corresponding or- the 1 in the middle of the last line is 1 = 12 21 .
thogonality and completeness relations for classes and The character of the identity group operation e is the
characters. If (i) is the character for elements in class dimension of the matrix representation, d .
i in irreducible representation and ni is the number of We use this character table to perform a Fourier anal-
group operations in that class, the character-class duality ysis on representations of this group. For example, the
is described by the following relations: representation of C3v = S3 in terms of 3 3 permuta-
tion matrices is not irreducible. For various reasons we
might like to know which irreducible representations of
X 0
Orthogonality : ni (i) (i) = |G|(0 , ) (91) C3v are contained in this reducible representation. The
i characters of the matrices describing each class are (c.f.,
Eq. (13)):
X
Completeness : ni (i) (i0 ) = |G|(i0 , i) (92)  +
{e} C3 , C3 {1 , 2 , 3 }
(95)
33 3 0 1
websearch: group characters and classes
To determine the irreducible content of this representa-
tion we take the inner product of Eq. (95) with the rows
D Fourier Analysis on Groups of Eq. (94) using Eq. (91) with the results

The group C3v has six elements. Its set of irreducible h33 |1 i = 1 3 1 + 2 0 1 + 3 1 1 = 6
representations has a total of six matrix elements. There- h33 |2 i = 1 3 1 + 2 0 1 + 3 1 1 = 0
fore d21 + d22 + = 6. This group has three classes. By h33 |3 i = 1 3 2 + 2 0 1 + 3 1 0 = 6
the character-class duality, it has three irreducible repre- (96)
sentations. As a result, d1 = d2 = 1 and d3 = 2. The As a result, the permutation representation is reducible
matrices of the six group operations in the three irre- and 33 ' 1 3 .
ducible representations are: websearch: character tables
websearch: character analysis
1 2 3
 
1 0
e [1] [1] E Irreps of SU (2)
 0 1 
a b
C3+ [1] [1] The unitary irreducible representations of Lie groups
 b a  can be constructed following two routes. One route be-
a b
C3 [1] [1] gins with the group. The second begins with its Lie alge-
 b a (93) bra. The second method is simpler to implement, so we
1 0
1 [1] [1] use it here to construct the hermitian irreps of su(2) and
 0 1 then exponentiate them to the unitary irreps of SU (2).
a b
2 [1] [1] The first step is to construct shift operators from the
 b a  basis vectors in su(2):
a b
3 [1] [1]

b a
1 3
a= 2 b= 2

The character table for this group is


18

 
0 1
S+ = Sx + iSy =
0 0
0 0 [Sz , S ] = S
S = Sx iSy =
  1 0 [S+ , S ] = 2Sz
1 0
Sz = 12
0 1
(97)
Next, we use the matrix algebra to operator algebra map-
ping (c.f., Sec. III G) to construct a useful boson operator
realization of this Lie algebra:

S+ S+ = b1 b2 1 
Sz Sz = b1 b1 b2 b2 (98)
2
S S = b2 b1
The next step introduces representations. Introduce a FIG. 4: Angular momentum operators J have isomorphic
state space on which the boson operators b1 , b1 act, with commutation relations with specific biliner combinations
basis vectors |n1 i, n1 = 0, 1, 2, with the action given bi bj of boson creation and annihilation operators for two
as usual by modes. The occupation number for the first mode is
plotted along the x axis and that for the second mode is
plotted along the y axis. The number-conserving opera-
b1 |n1 i = |n1 + 1i n1 + 1 b1 |n1 i = |n1 1i n1 (99)
tors act along diagonals like the operators J+ , J , Jz to
Introduce a second state space for the operators b2 , b2 and easily provide states and matrix elements for the su(2)
basis vectors |n2 i, n2 = 0, 1, 2, . In order to construct operators.
the irreducible representations of su(2) we introduce a
grid, or lattice, of states |n1 , n2 i = |n1 i |n2 i. The
and similarly for the shift-down operator
operators S , Sz are number-conserving and move along
the diagonal n1 + n2 = cnst. (c.f., Fig. 4). It is very
useful to relabel the basis vectors in this lattice by two j j p
integers. One (j) identifies the diagonal, the other (m) S | i=| i (j m + 1)(j + m) (103)
m m1
specifies position along a diagonal:
In this representation of the (spin) angular momentum
2j = n1 + n2 n1 = j + m j algebra su(2), Sz = Jz is diagonal and S = J have
|n1 , n2 i |
i one nonzero diagonal row just above (below) the main
2m = n1 n2 n2 = j m m
(100) diagonal. The hermitian irreducible representations of
The spectrum is 2j = 0, 1, 2, and m = j, j + su(2) with j = 0, 21 , 1, 32 , 2, 52 form a complete set of
1, , +j. irreducible representations for this Lie algebra.
The matrix elements of the operators S with respect to The unitary irreducible representations of SU (2) are
j obtained by exponentiating i times the hermitian repre-
the basis | i are constructed from the matrix elements sentations of su(2):
m
of the operators bi bj on the basis vectors |n1 , n2 i. For Sz
we find DJ [SU (2)] = EXP i
n J (104)

j with Jx = (J+ +J )/2 and Jy = (J+ J )/2i, and J are


Sz | i = 12 (b1 b1 b2 b2 )|n1 , n2 i = the (2j + 1) (2j + 1) matrices whose matrix elements
m
(101) are given in Eqs. (101-103). For many purposes only
j
|n1 , n2 i 12 (n1 n2 ) = | im the character of an irreducible representation is needed.
m
The character depends only on the class and the class is
For the shift-up operator uniquely determined by the rotation angle (rotations by
angle about any axis n are geometrically equivalent).
j It is sufficient to compute the trace of any rotation, for
S+ | i = b1 b2 |n1 , n2 i = |n1 + 1, n2 1i n1 + 1 n2 =
m example the rotation
 about the z axis. This matrix is
j p diagonal: eiJz m0 ,m = eim m0 ,m and its trace is
| i (j + m + 1)(j m)
m+1
(102)
19

by 2/2 radians about the same midpoints of opposite


TABLE II: (top) Character table for the cubic group Oh . faces, and six rotations C2 about the midpoints of oppo-
The functions in the right-hand column are some of the site edges. The characters for these five classes in the five
basis vectors that carry the corresponding representa- irreducible representations are given at the top in Table
tion. (bottom) Characters for rotations about indicated II. At the bottom of the table are the characters of the
angle in the irreducible representations of the rotation irreducible representations of the rotation group SO(3)
group. in the irreducible representations of dimension 2L + 1.
These are obtained from Eq. (105). A character analysis
Oh E 8C3 3C42 6C2 6C4 Basis (c.f., Eq. (96)) leads to the Oh irreducible content of each
of the lowest six irreducible representations of SO(3).
A1 1 1 1 1 1 r 2 = x2 + y 2 + z 2
websearch: crystal field theory
A2 1 1 1 1 1
websearch: ligand field theory
E 2 1 2 0 0 (x2 y 2 , 3z 2 r2 )
websearch: N.B.S. tables of crystal point groups
T1 3 0 1 1 1 (x, y, z), (Lx , Ly , Lz )
T2 3 0 1 1 1 (yz, zx, xy)
2 2 2 2
L: 0 3 2 2 4 Reduction VI. SYMMETRY GROUPS
0S 1 1 1 1 1 A1
1P 3 0 1 1 1 T1 Groups first appeared in the Quantum Theory as a
2D 5 1 1 1 1 E T2 tool for labelling eigenstates of a Hamiltonian with useful
3F 7 1 1 1 1 A2 T1 T2 quantum numbers. If a Hamiltonian H is invariant under
4G 9 0 1 1 1 A1 E T1 T2 the action of a group G, then gHG1 = H, g G.
5 H 11 1 1 1 1 E 2T1 T2 If | i satisfies Schrodingers time-independent equation
H| i E| i = 0, so that

g(H E)| i = g(H E)g 1 g| i =




+j
sin(j + 21 )
(H E)H| ih |g| i = (H E)H| iD,
X
j () = eim = (105) (g)
m=j
sin 12 (106)
All states | i related to each other by a group trans-
These characters are orthonormal with respect to the formation g G have the same energy eigenvalue. The
weight w() = 1 sin2 2 .

existence of a symmetry group G for a Hamiltonian H
websearch: Schwinger on angular momentum provides representation labels for the quantum states and
also describes the degeneracy patterns that can be ob-
served. If the symmetry group G is a Lie group, so that
F Crystal Field Theory g = eX , then eX HeX = H [X, H] = 0. The ex-
istence of operators X that commute with the Hamilto-
The type of Fourier analysis outlined above has found nian H is a clear signal that the physics described by the
a useful role in Crystal (or Ligand) Field Theory. We Hamiltonian is invariant under a Lie group.
illustrate with a simple example. For example, for a particle in a spherically symmetric
A many-electron atom with total angular momentum potential V (r) Schrodingers time-independent equation
L is placed in a crystal field with cubic symmetry. How is
do the 2L + 1-fold degenerate levels split?
Before immersion in the crystal field, the atom has
p2
 
spherical symmetry. Its symmetry group is the rotation + V (r) = E (107)
group, the irreducible representations DL have dimension 2m
2L + 1, the classes are rotations through angle , and the
character for the class in representation DL is given with p = (~/i). The Hamiltonian operator is invari-
in Eq. (105) with j L (integer). When the atom is ant under rotations. Equivalently, it commutes with the
placed in an electric field with cubic symmetry Oh , the angular momentum operators L = r p: [L, H] = 0.
irreducible representations of SO(3) become reducible. The wavefunctions can be partly labeled by rotation
l
The irreducible content is obtained through a character group quantum numbers, l and m: m (r, , ).
analysis. In fact, by standard separation of variables arguments
The group has 24 operations partitioned into five this description can be made more precise: (r, , ) =
1 l
classes. These include the identity E, eight rotations C3 r Rnl (r)Ym (, ). Here Rnl (r) are radial wavefunctions
by 2/3 radians about the diagonals through the oppo- that depend on the potential V (r) but the angular func-
site vertices of the cube, six rotations C4 by 2/4 radians tion Yml (, ) is a piece of geometry: it depends only
about the midpoints of opposite faces, three rotations C42 on the existence of rotation symmetry. It is the same no
20

matter what the potential is. In fact, these functions can Selection rules derived from SO(3) simplify the angular
be constructed from the matrix representations of the integral. First, the integral vanishes unless M = M 0
group SO(3). The action of a rotation group operation M = 0. It also vanishes unless L = 1, 0. By parity, it
g on the angular functions is vanishes if L = 0, and by time reversal its value for M
and M are the same. The nonzero values are
l
gYml (, ) = Yml 0 (, )Dm 0 m (g) (108) r Z
4 L1
where the construction of the Wigner D matrices has A(L, M ) = Y L0 ()Y01 ()YM ()d =
3 M
been described in Sec. V E.
If the symmetry group is reduced, as in the case of s
(L + M )(L M )
SO(3) Oh described in Sec. V F, the eigenstates are M 0 M
identified by the labels of the irreducible representations (2L + 1)(2L 1)
of Oh : A1 , A2 , E, T1 , T2 . (111)
Once the states have been labeled, computations must The radial integrals also satisfy selection rules: they
be done. At this point the power of group theory becomes vanish unless L = 1. The nonzero integrals are all
apparent. Matrices must be computed for example, related:
matrix elements of a Hamiltonian. Typically, most ma-
trix elements vanish (by group-theoretic selection rules).

N N 2 L2
Z
Of the small number that do not vanish, many are sim- R(N, L) = RN,L (r)rRN,L1 dr = R(2, 1)
ply related to a small number of the others. In short, 0 2 3
using group theory as a guide, only a small number of (112)
computations must actually be done. with 1 L N 1. All integrals are proportional to
This feature of group theory is illustrated by comput- the single integral R(2, 1). This comes from yet another
ing the eigenstates and their energy eigenvalues for an symmetry that the Coulomb potential exhibits (c.f. Sec.
electron in the N = 4 multiplet of the hydrogen atom VII below), not shared by other spherically symmetric
under the influence of a constant external field E. The potentials. The single integral to be evaluated is
Hamiltonian to be diagonalized is

R(2, 1) = 3 3a0 (113)
L0 p2 e2 L
hN 0 0 | + eE r|N i (109) This integral is proportional to the Bohr radius a0 of the
M 2m r M
hydrogen atom, whose value was estimated in Eq. (6).
The first two terms in the Hamiltonian describe the elec- The auguments above show drastic simplifications in
tron in a Coulomb potential, the last is the Stark per- the computational load for the Stark problem. Of the
turbation, which describes the interaction of a dipole 256 = 162 matrix elements to compute only 18 are
d = er with a constant external electric field: HSt. = nonzero. All are real. Since the Hamiltonian is hermitian
d E. In the N = 4 multiple we set N 0 = N = 4, (symmetric if real) there are in fact only 9 nonzero matrix
so that L0 , L = 0, 1, 2, 3 and M ranges from L to elements to construct. Each is a product of two factors,
+L and L0 M 0 +L0 . The matrix elements of so only 6 (angular) plus 1 (radial) quantities need be
the Coulomb Hamiltonian are EN N 0 N L0 L M 0 M , with computed. These numbers must be stuffed into a 16 16
E4 = 13.6/42 eV . matrix to be diagonalized. But there are no nonzero ma-
P3=41
There are L=0 (2L + 1) = 16 states in the N = 4 trix elements between states with M 0 6= M . This means
multiplet, so 162 matrix elements of the 16 16 ma- that by organizing the row and columns appropriately
trix must be computed. We simplify the computation by the matrix can be written in block diagonal form. The
choosing the z axis in the direction of the applied uni- block diagonal form consists of a 1 1 matrix for M = 3,
a 2 2 matrix for M = 2, a 3 3 matrix for M = 1,
q eE r eEz (E = |E|). In
form electric field, so that
a 4 4 matrix for M = 0, a 3 3 matrix for M = 1,
4 1
addition we write z = 3 rY0 (, ). The matrix ele- etc. The 1 1 matrices are already diagonal. The 2 2
ments factor (separation of variables) into a radial part matrices are identical, so only one needs to be diagonal-
and an angular part, as follows ized. Similarly for the two 3 3 matrices. There is only
one 4 4 matrix. The computational load for diagonal-
izaing this matrix has been reduced from T ' 162 log 16
h4L0 M 0 |eEz|4LM i eE Radial Angular
to T ' 22 log 2 + 32 log 3 + 42 log 4, a factor of 20 (assum-
R
(110) ing the effort required for diagonalizing an n n matrix
Radial = 0qR4L0 (r)r1 R4L (r)dr
R L0 goes like n2 log n)!
Angular = 4
3 YM 0 ()Y01 ()YM
L
()d It gets even better. For the N = 5 multiplet the
1 1, 2 2, 3 3, 4 4 matrices are all proportional to
where = (, ) and d = sin dd. the matrices of the corresponding size for N = 4. The
21

proportionality factor is 5/4. Only one new matrix needs group is larger than the obvious symmetry group SO(3).
to be constructed the 5 5 matrix. This symmetry On the bound states this Lie algebra describes the Lie
extends to all values of N . group SO(4). The irreducible representation labels for
This is a rather simple example that can be carried out the quantum states are N , N = 1, 2, 3..., . The three
by hand. This was done when Quantum Mechanics was nested groups SO(2) SO(3) SO(4) and their repre-
first developed, when the fastest computer was a greased sentation labels and branching rules are:
abacus. Today time savings of a factor of 20 on such
a simple problem would hardly be noticed. But calcu-
lations have also inflated in size. Reducing a 106 106 Group Rep.Label Degeneracy Branching Rules
matrix to about 1000 103 103 reduces the computational SO(4) N N2
effort by a factor of 2000. For example, a computation SO(3) L 2L + 1 0, 1, 2, N 1
that would take 6 years without such methods could be SO(2) M 1 L M +L
done in a day with these methods. (114)
Symmetry groups play several roles in Quantum Me- Branching rules identify the irreducible representations
chanics. of a subgroup that any representation of a larger group
branches to under group - subgroup reduction. We have
They provide group representation labels to iden- seen branching rules in Table II.
tify the energy eigenstates of a Hamiltonian with One advantage of using the larger group is that there
symmetry. are more shift operators in the Lie algebra. The shift
operators, acting on one state, moves it to another (c.f.
They provide selection rules that save us the effort L+
|LM i |L, M + 1i). This means that there are well-
of computing matrix elements whose values are zero
defined algebraic relations among states that belong to
(by symmetry!).
the same N multiplet. This means that more of any
And they allow transformation of a Hamiltonian computation can be pushed from the physical domain to
matrix to block diagonal form, so that the compu- the geometric domain, and simplifications accrete.
tational load can be drastically reduced. Why stop there? In the hydrogen atom the energy dif-
ference between the most tightly bound state, the ground
websearch: symmetry groups in physics state, and the most weekly bound state (N ) is
websearch: Stark effect N=4 13.6 eV . When this difference is compared with the elec-
tron rest energy of 511 KeV , the symemtry breaking is
about 13.6/511000 ' 0.000027 or 2.7 103 %. This sug-
VII. DYNAMICAL GROUPS gests that there is a yet larger group that accounts for this
near degeneracy. Searches eventually lead to the non-
A widely accepted bit of wisdom among Physicists is compact conformal group SO(4, 2) SO(4) as the
that symmetry implies degeneracy, and the larger the all-inclusive symmetry group of the hydrogen atom.
symmetry, the larger the degeneracy. What works for- The virtue of using this larger group is that states in dif-
ward ought to work backward (Newtons Third Law): ferent multiplets N, N 1 can be connected by shift op-
if the degeneracy is greater than expected, the symmetry erators within the algebra so(4, 2), and ultimately there
is greater than apparent. is only one number to compute. Including this larger
group in Eq. (114) would include inserting it in the row
above SO(4), showing there is only one representation
A Conformal Symmetry label for bound states, indicating its degeneracy is ,
and adding branching rules N = 1, 2 , to the SO(4)
The hydrogen atom has rotational symmetry SO(3), row.
and this requires 2L + 1 fold degeneracy. But the states websearch: conformal group, hydrogen atom
with the same principal quantum number N are all de- websearch: A. O. Barut
generate in the absence of spin and other relativistic ef-
fects, and nearly degenerate in the presence of these ef-
B Atomic Shell Structure
fects. It would make sense to look for a larger than ap-
parent symmetry. It exists in the form of the Runge-
Lenz vector M = 2m 1
(p L L p) e2 r/r, where Broken symmetry beautifully accounts for the system-
r, p, L = r p are the position, momentum, and orbital atics of the chemical elements. It accounts for the filling
angular momentum operators for the electron. The three scheme as electrons enter a screened Coulomb potential
orbital angular momentum operators Li and three com- around a nuclear charge +Ze as the nuclear charge in-
ponents of the Runge-Lenz vector close under commu- creases from Z = 1 to Z > 92. The screening is caused
tation to form a Lie algebra. The six operators com- by inner electrons. The filling scheme accounts for the
mute with the Hamiltonian, so the hidden symmetry magic numbers among the chemical elements: these
22

are the nuclear charges of exceptionally stable chemical


elements He, Ne, Ar, Kr, Xe, Rn with atomic numbers
2, 10, 18, 36, 54, 86.
When more than one electron is present around a
nuclear charge +Ze then the outer electrons see a
screened central charge and the SO(4) symmetry aris-
ing from the Coulomb nature of the potential is lost.
There is a reduction in symmetry, a broken symmetry:
SO(4) SO(3). The quantum numbers (N, L) can be
used to label states and energies, EN,L and these energy
levels are (2L + 1)-fold degenerate. The SO(4) multi-
plet with quantum number N splits into orbital angular
momentum multiplets with L values ranging from 0 to a
maximum of L = N 1.
Each additional electron must enter an orbital that is
not already occupied by the Pauli Exclusion Principle.
This Principle is enforced by the requirement that the
total electron wavefunction transform under the unique
antisymmetric representation anti. (Sk ) on the permuta-
tion group Sk for k electrons.
Generally, the larger the L value the further the outer
electron is from the central charge, on average. And the
further it is, the larger is the negative charge density con-
tributed by inner electrons that reduces the strength of
the central nuclear attraction. As a result EN,0 < EN,1 <
< EN,L=N 1 . There is mixing among levels with dif-
ferent values of N and L. The following energy order-
ing scheme, ultimately justified by detailed calculations,
accounts for the systematics of the chemical elements,
including the magic numbers:

FIG. 5: Broken SO(4) dynamical symmetry due to


1S|2S 2P |3S 3P |4S 3D 4P |5S 4D 5P |6S 4F 5D 6P |7S screening of the central Coulomb potential by inner elec-
(115) trons successfully accounts for the known properties of
Each level can hold 2(2L + 1) electrons. The first factor the chemical elements, as reflected in Mendeleevs Peri-
of 2 = (2s + 1) with s = 12 is due to electron spin. The odic Table of the Chemical Elements.
vertical bar | indicates a large energy gap. The cumu-
lative occupancy reproduces the magic numbers of the
chemical elements: 2, 10, 18, 36, 54, 86. The filling order have a degeneracy (N + 2)(N + 1)/2. Under a spherically
is shown in Fig. 5. Broken symmetry is consistent with symmetric perturbation these highly degenerate multi-
Mendeleevs periodic table of the chemical elements. plets would split into mutliplets identified by an angular
websearch: Periodic Table momentum index. A character analysis gives this branch-
ing result

C Nuclear Shell Structure N L Values Spectroscopic


0 0 S
Magic numbers among nuclei suggested that, here also, 1 1 P
one could possibly describe many different nuclei with a (116)
2 2, 0 D, S
single simple organizational structure. The magic num- 3 3, 1 F, P
bers are: 2, 8, 20, 28, 40, 50, 82, 126, both for protons and 4 4, 2, 0 G, D, S
for neutrons. The following model was used to organize
this information. For example, the N = 4 harmonic oscillator multiplet
Assume that the effective nuclear potential for protons splits into an L = 4 multiplet, an L = 2 multiplet, and
(or neutrons) is that of a a three-dimensional isotropic an L = 0 multiplet. The larger the angular momentum,
harmonic oscillator. The basis states can be described the lower the energy. After this splitting, the spin of
by |n1 , n2 , n3 i. One excitation would be three-fold de- the proton (or neutron) is coupled to the orbital angular
generate: |1, 0, 0i, |0, 1, 0i, |0, 0, 1i, two excitations would momentum to give values of the total angular momentum
be 6-fold degenerate, and states with N excitations would J = L 12 , except that for S states only the J = 12 state
23

occurs. Again, the larger angular momentum occurs at a is 9 + 2 3 + 1 = 16 = 42 dimensional, and is closely
lower energy than the smaller angular momentum. The related to the noncompact Lie algebra u(3, 1). This em-
resulting filling order, analogous to Eq. (115), is bedding in a larger group is analogous to the inclusion
SO(4) SO(4, 2) for the hydrogen atom.
0S1/2 |1P3/2 1P1/2 |2D5/2 2S1/2 2D3/2 |3F7/2 | websearch: nuclear shell model
3P3/2 3F5/2 P1/2 4G9/2 |4D5/2 4G7/2 4S1/2 4D3/2 5H11/2 |
5H9/2 5F7/2 5F5/2 5P3/2 5P1/2 6I13/2 |
D Dynamical Models
(117)
Each shell with angular momentum j can hold up to
2j + 1 nucleons. Broken symmetry is also consistent with In this Section so far we have described the hydro-
the periodic table associated with nuclear shell models. gen atom using a very large group SO(4, 2) and breaking
The filling order is shown in Fig. 6. down the symmetry to SO(4) and further to SO(3) when
there are Coulomb-breaking perturbations that maintain
their spherical symmetry. We have also introduced a se-
quence of groups and subgroups U (3, 1) U (3) SO(3)
to provide a basis for the nuclear shell model.
Nuclear computations are very difficult because there
is no nuclear force. The force acting between nucleons
is a residual force from the quark-quark interaction. This
is analogous to the absence of a molecular force. There
is none - the force that binds together atoms in molecules
is the residual electromagnetic force after exchange and
other interactions have been taken into account.
For this reason it would be very useful to develop a sys-
tematic way for making nuclear models and carrying out
calculations within the context of these models. Group
theory to the rescue!
The first step in creating a simple environment for
quantitative nuclear models is to assume that pairs of
nucleons bind tightly into boson-like excitations. The
leading assumption is that of all the nuclear-pair de-
grees of freedom, the most important were those with
scalar (S, L = 0) and quadrupole (D, L = 2) trans-
formation properties under the rotation group SO(3).
States in a Hilbert space describing 2 protons (neu-
trons, nucleons) can be produced by creation operators
FIG. 6: The filling order describing very many proper- s , dm acting on the vacuum |0; 0, 0, 0, 0, 0i. For n pairs
ties of nuclear ground states is described by the levels of of nucleons, n creation operators act
an isotropic harmonic oscillator potential with multiplets P to produce states
|ns ; n2 , n1 , n0 , n1 , n2 i with ns + m nm = n. There
having N excitations and degeneracy (N + 1)(N + 2)/2. are (n + 6 1)!/n!(6 1)! states in this Hilbert space.
The degeneracy is broken by a spherically symmetric For computational convenience they can be arranged by
perturbation and broken further by spin-orbit coupling. their transformation properties under rotations SO(3).
For both perturbations energy increases as angular mo- For example, the two boson Hilbert space has 21 states
mentum decreases. The filling order shown successfully consisting of an L = 0 state from s s , an L = 2 mul-
accounts for the known properties of the ground states tiplet from s dm , and multiplets with L = 0, 2, 4 from
of most even-even nuclei, including the magic numbers.
dm0 dm .
In the higher levels the largest spin angular momentum
The Hamiltonian acts within the space with a fixed
state (e.g., 5H11/2 ) is pushed down into the next lower
number of bosons. It must therefore be constructed from
multiplet, containing all the remaining N = 4 states,
number-conserving operators: bi bj ,where the boson op-
with the exception of the 4G9/2 .
erators include the s and d excitations. These opera-
tors must be rotationally invariant. At the linear level
At a group theoretical level, our starting point has been
only two such operators exist: s s and dm dm . At the
the Lie algebra u(3) with basis vectors bi bj (1 i, j 3) quadratic level there are a small number of additional
whose representations are labeled by an integer index N , rotationally invariant operators. The n boson Hamilto-
the number of excitations present. This algebra can be nian can therefore be systematically parameterized by a
embedded in a larger Lie algebra containing in addition relatively small number of terms. The parameters can be

shift up operators bi , their counterpart annihilation op- varied in attempts to fit models to nuclear spectra and
erators bj , and the identity operator I. The Lie algebra
24

transition rates. In the two boson example with 21 states, Weyl came back to this general idea following
it is sufficient to diagonalize this Hamiltonian in the two- Schrodingers development (1926) of Wave Mechanics. In
dimensional subspace of L = 0 multiplets, in another two this case a modified objective was achieved: he succeeded
dimensional subspace with the two states with L = 2 and in describing how light interacts with charged matter.
ML = 2 (all other ML values will give the same result), The original theory (GR) involved a real scaling trans-
and the one-dimensional suspace with L = 4, ML = 4. formation that was space-time dependent. As a result,
The Interacting Boson Model outlined above has it is in the same spirit as the discussion about scaling in
deeply extended our understanding of nuclear physics. In Sec. II B, but more general. His modified theory (QM)
fact, some Hamiltonians can be solved by hand. These involved a complex phase transformation. In some sense
involve a group-subgroup chain. The chain of groups is this would be an analytic continuation of the scaling ar-
shown in Fig. 7. This model incorporates in a magnifi- guments, but the spirit of the discussion given in Sec.
cent way the use of groups in their capacity as symmetry II B does not in any sense suggest phase changes.
groups, implying degeneracy, and dynamical groups, im- The starting point of this work is the observation that
plying relations among multiplets of different energies. if (x, t) satisfies Schrodingers time-dependent equation,
so also does ei (x, t), for

   
i i
H i~ e (x, t) = e H i~ (x, t) = 0
t t
(118)
This fails to be true if the phase depends on space-time
coordinates, for then the derivative terms act on this
phase when we try to pull it through the Hamiltonian
and time-derivative operators:

p2
  

+ q(x, t) i~ ei(x,t) (x, t) =
2m t

(p + ~)2
 
i(x,t)
e + q(x, t) + ~ i~ (x, t)
2m t t
(119)
Symmetry is not preserved! What to do?
FIG. 7: States with 2N nucleons outside a closed shell It had long been known that the electric and magnetic
are described by N bosons in the Interating Boson Model. fields E, B could be represented by fictitious poten-
The basis states carry a symmetric representation of the tials that served to simplify Maxwells equations but were
Lie group U (6). Various limiting Hamiltonians that ex- otherwise not real. The vector potential A and scalar
hibit a group-subgroup symmetry can be diagonalized by potential are related to the real fields by
hand. The three group-subgroup chains for which this is
possible are shown here.
B = A
1 A (120)
websearch: Interacting Boson Model E =
c t
This simplification is not unique. The vector potential
VIII. GAUGE THEORY can be changed by the addition of the gradient of a
scalar field (x, t), and the scalar potential correspond-
Gauge transformations were introduced by Weyl fol- ingly changed:
lowing Einsteins development (1916) of the Theory of
General Relativity. In crude terms, Weyls original idea
was to introduce a ruler (the gauge of gauge theory) A A0 = A + B0 = B
whose length was an arbitrary function of position. His 1 (121)
0 = E0 = E
original objective was to unify the two then-known forces c t
of Nature: gravitation and electromagnetism. His theory The resolution of the difficulty is to assume that the
is quite beautiful but Einstein raised serious objections, electrostatic part of the interaction is described by the
and Weyl eventually relinquished it. Einsteins objection term q(x, t) in the Hamiltonian and the magnetic part
was that if Weyls theory were correct then the results of is represented by replacing p by p qc A(x, t) wherever it
laboratory experiments would depend on the history of appears in the Hamiltonian. Under these conditions
the material being investigated.
25

( , 0 , analogous to photons that carry the electromag-


 q   q netic interaction) responsible for the interaction among
p A(x, t) ei(x,t) = ei(x,t) p A(x, t) + ~(x, t) the fermions.
c c
(122) This program was carried out by Yang and Mills.
and They succeeded in determining the nature of the inter-
action. But we now understand that nuclear interactions
    are residual forces left over from the strong interactions
i(x,t) i(x,t) among the quarks.
q i~ e =e q + ~ i~
t t t Nevertheless, the program persisted. The gauge pro-
(123) gram can be phrased as follows.
q
If we choose (x, t) = ~c (x, t), then the added terms 1. Suppose there is a set of n fermion fields that are
on the right in Eq. (122) are invariant under a g-parameter Lie group.
2. Assume that the Hamiltonian (Lagrangian, Ac-
q q q 0 tion Integral) for these fields, without any interaction,
p A(x, t) (x, t) = p A (x, t) (124) is known.
c c c
3. Now assume that the Lie group parameters are
and those on the right in Eq. (123) are allowed to be functions on spacetime. What additional
terms occur in the Hamiltonian (c.f., Eq. (119) above).
q (x, t) 4. How many boson fields must be introduced in order
q(x, t) i~ = q0 (x, t) i~ (125) to leave the structure of the Hamiltonian invariant?
c t t t
5. How must they be introduced into the Hamiltonian.
The result is that the structure of the interaction be- That is: what is the structure of the Minimal Coupling
tween the electromagnetic fields and charged particles is in terms of the Lie algebra parameters (its structure con-
invariant provided the interaction is given in terms of the stants)?
fictitious fields A, by 6. How do these new fields transform under the Lie
group and its spacetime extension?
q 7. What field equations do the new fields satisfy?
p p A(x, t) i~ i~ + q(x, t) (126) These questions have all been answered. The number
c t t of new fields required is exactly the number of generators
There are several other ways to couple the electromag- of the Lie group (i.e., its dimension). Each field is a
netic field with charged particles that are allowed by sym- four component field. Their dynamical equations are a
metry. But the structure of the interaction described by consequence of this theory. All new fields are massless.
Eq. (126) is sufficient to account for all known measure- This theory has been applied to describe the elec-
ments. It turns out that Maxwells equations are also a troweak interaction U (2) ' U (1) SU (2) to predict
consequence of the structure of this interaction. the massless electromagnetic field and three boson fields
This principle is called the Principle of Minimal Elec- called W , Z 0 that transmit the weak interaction. This
tromagnetic Coupling. theory was also applied to describe three quarks. The
The phase transformation introduced in Eq. (118) Lie group used was SU (3) and the theory predicted the
belongs to the Lie group U (1). Its generalization to existence of eight (thats the dimension of the Lie group
position-dependent phase ei(x,t) does not belong to a SU (3)) gluon fields, all massless. The gluon fields trans-
Lie group. mit the strong interaction. In the case of the gluons,
Questions soon surfaced if the same process could be the mass seems to be small enough to be consistent with
used to describe the interaction between more compli- zero but that is definitely not the case of the very mas-
cated charged particles and the fields that cause inter- sive weak gauge bosons W , Z 0 . A new mechanism was
actions among them. It seemed that the proton-neutron called for, and proposed, to describe how these mass-
pair was a good candidate for such a treatment. These less particles acquire such a heavy mass. This mecha-
two particles seemed to be essentially the same, except nism was proposed by Higgs, among others, and is called
that one was charged and the other not. Neglecting the Higgs mechanism. The discovery of the Higgs boson
charge, these two particles could be treated as an isospin has recently been announced.
doublet. The nucleon wavefunction could be treated websearch: gauge theory lie groups
p
as a two-state system: |i = | i and the Hamiltonian
n
describing nuclear interactions should be invariant un-
der a global SU (2) transformation, analogous to a global
U (1) transformation ei in Eq. (118). If the SU (2)
rotation were allowed to vary with position, perhaps it
would be possible to determine the nature of the inter-
action between the nucleons (fermions) and the bosons
26

IX. GROUP THEORY AND SPECIAL B Relation with Lie Groups


FUNCTIONS
A Lie group lives on a manifold Mn of dimension n.
A Summary of Some Properties Each group operation is a function of position in the man-
ifold: g = g(x), x M. The product of two group oper-
The classical special functions of mathematical physics ations is defined by an analytic composition law on the
were developed in the 19th century in response to a va- manifold:
riety of specific physical problems. They include the
Legendre and associated Legendre functions, the La- g(x) g(y) = g(z) z = z(x, y) (129)
guerre and associated Laguerre functions, the Gegen-
bauer, Chebyshev, Hermite, and Bessel functions. They It is not until we construct representations for the
are for the most part orthogonal polynomials. They are group, or on top of the manifold, that really interesting
constructed by choosing a basis set f0 , f1 , f2 , that are things begin to happen. Representations
monomials in the position representation (Dirac nota-
tion): hx|f0 i = x0 , hx|f1 i = x1 , hx|f2 i = x2 , and g(x)
ij (g(x)) (130)
then creating an orthogonal set by successive Gramm-
Schmidt orthogonalization by means of an inner product are functions defined on the manifold. Suitably normal-
Rb
hf |gi = a f (x)g(x)w(x)dx with various weights w(x) ized, the set of matrix elements for the complete set of
for the different functions: UIR (unitary irreducible representations) form a com-
plete orthonormal set of functions on the manifold Mn .
By duality (the miracles of Hilbert space theory), the
|0 ih0 | triplet of indices , i, j is described by as many integers as
|0 i = |f0 i, |1 i = |f1 i |f0 i
h0 |0 i the dimension of Mn . For example, for three dimensional
j1 (127) Lie groups, such as SO(3), SU (2), SO(2, 1), ISO(2), H3
X |k ihk |
|j i = |fj i |fj i the matrix elements are indexed by three integers and
hk |k i
k=0
can be represented in the form ij (g(x)) = h i |g(x)| j i.
The Bessel functions are the exception to this rule, as Including the appropriate normalization factor, they can
they are not polynomials. be expressed as
These functions obey a common variety of properties
s
Differential Equation: dim()
(g(x)) = hg(x)| i (131)
g2 (x)y 00 + g1 (x)y 0 + g0 (x)y = 0 (128a) V ol(G) ij i, j
Representations are powerful because they lie in two
Recurrence Relations: worlds: geometric and algebraic. They have one foot
in the manifold (hg(x)| ' hx| above) and the other in
a1n fn+1 (x) = (a2n + a3n x)fn (x) a4n fn1 (x)
(128b) algebra ( | i ' |ni above).
ij
All special functions are specific matrix elements, eval-
Differential Relations:
uated on specific submanifolds, of specific irreducible rep-
dfn (x) resentations of some Lie group.
g2 (x) = g1 (x)fn (x) + g0 (x)fn1 (x) (128c) We illustrate these ideas with a few examples without
dx
pretending weve even scratched the surface of this vast
Generating Functions: and fascinating field.

X
g(x, z) = an fn (x)z n (128d) C Spherical Harmonics and SO(3)
n=0

For the group SU (2) the underlying manifold is a solid


Rodrigues Formula:
three-dimensional sphere. There are many ways to pa-
1 dn rameterize an operation in this group. We use an Euler-
fn (x) = {(x)(g(x))n } (128e) angle like parameterization introduced by Wigner:
an (x) dxn

The coefficients and functions can be found in standard j j iJz iJy iJz j
Dmk (, , ) = h |e e e | i=
tabulations (e.g., Abramowicz and Stegun). The Bessel m k
(132)
functions have similar properties.
websearch: Abramowicz and Stegun eim djmk ()eik
27

The orthogonality properties of the matrix elements are hx0 |O|xi or the discrete basis hn0 |O|ni. In the first basis
the coordinates x describe a submanifold in the group
Z 2 Z Z 2 0
manifold Mn , and the operator is a differential operator.
j j
d sin d d Dm 0 k 0 (, , )Dmk (, , ) In the second basis the indices n are an appropriate sub-
0 0 0 set of the group representation and row/column (i, j)
index set and operator is a matrix with entries in the real
8 2 j 0 j or complex field.
= m0 m k0 k
2j + 1 It is also possible to compute the matrix elements in a
(133) mixed basis hx|O|ni. It is in this basis that really exciting
The volume of the group in this parameterization is 8 2 . things happen, for
The normalization factor, converting
p the matrix elements
to a complete orthonormal set, is (2j + 1)/8 2 .
In order to find a complete set of functions on the hx|O|ni
sphere (, ) we search for those matrix elements above . & (137)
that are independent of the angle . These only occur for hx|O|x0 ihx0 |ni = hx|n0 ihn0 |O|ni
k = 0, which occurs only among the subset of irreducible
representations with j = l (integer). Integrating out the On the left hand side a differential operator hx|O|x0 i acts
d dependence in Eq. (132) leads to a definition of the on the special function hx0 |ni, while on the right-hand
spherical harmonics in terms of some Wigner D matrix side a matrix hn0 |O|ni multiplies the special functions
elements (c.f., Eq. (133): hx|n0 i.
For the rotation group acting on the sphere surface
l
r (, ) and the choice O = L we find for h|L | i
2l + 1 l m
Yml (, ) = Dm0 (, , ) (134) computed as on the left in Eq.(137)
4
These functions on the two-dimensional unit sphere sur-  
face (, ) inherit their orthogonality and compelteness i cos
e +i (cos 0 cos )(0 )Yml (0 , 0 )
properties from the corresponding properties of the UIR sin
j
matrix elements Dmk on the three-dimensional solid  
sphere of radius 2. i cos
Other special functions are similarly related to these =e +i Yml (, )
sin
matrix elements. The associated Legendre polynomials (138)
are and as computed on the right
s
(l + m)! l l0 l0 l l
p
Plm (cos ) = d () (135) h| 0 ih |L | i = Ym1 (, ) (l m + 1)(l m)
(l m)! 0,0 m m0 m
(139)
and the Legendre polynomials are There are a number of Lie groups that can be de-
fined to act on a one dimensional space. In such cases
l
the infinitesimal generators take the form of functions of
Pl (cos ) = D0,0 (, , ) = dl0,0 () (136) the coordinate x and the derivative d/dx. We illustrate
These functions inherit the their measure w() from the the ideas behind differential and recursion relations in
measure on SU (2) and their orthogonality and complete- the context of the Hiesenberg group H3 . Its algebra h3
ness properties from those of the Wigner rotation matrix is spanned by three operators, universally  identified as


elements Dmkj
[SU (2)]. a, a
  , I with commutation relations a, a = I, [a, I] =
We emphasize again that these functions are specific a , I = 0. These operators have matrix elements as fol-
j lows in the continuous basis (geometric) representation:
matrix elements Dmk , evaluated on specific submanifolds
(sphere, line), of specific irreducible representations (j =
l) of SU (2). hx0 |a|xi = (x0 x) 12 (x + D)
hx0 |a |xi = (x0 x) 12 (x D) (140)
D Differential and Recursion Relations hx0 |I|xi = (x0 x)

and discrete basis (algebraic) representation:


We can understand the wide variety of relations that
exist among the special functions (e.g., recursion rela-
tions, etc.) in terms of group theory/representation the- hn0 |a|ni = n0 ,n1 n
ory as follows. It is possible to compute the matrix el- hn0 |a |ni = n0 ,n+1 n0 (141)
ements of an operator O in either the continuous basis hn0 |I|ni = n0 ,n
28

d
Here D = dx . it also commutes with every element in the Lie algebra.
The special functions are the mixed basis matrix el- Such operators can always be constructed for semisimple
ements hx|ni. We can compute these starting with the Lie algebras where the Cartan metric gij (c.f., Eq. (46))
ground, or lowest, state |0i. is nonsingular. The operator g ij Xi Xj has this property.
The construction of nontrivial quadratic operators with
hx|a|0i this property is even possible for many Lie algebras that
. & are not semisimple. When it is possible, the left-hand
hx|a|0i = hx|nihn|a|0i (142) side of Eq. (137) is a second order differential operator
1 (x + D)hx|0i
and the right hand side is a constant. This constant is
= 0
2 the eigenvalue in the differential equation (first property
2 listed above).
This equation has a unique
solution N hx|0i = ex /2 up
For the three-dimensional nonsemisimple group
to scale factor, N = 1/ .
4
ISO(2) of length-preserving translations and rotations
The remaining normalized basis states are constructed
of the plane to itself, the three infinitesimal generators
by applying the raising operator:
are L3 which generates rotations around the z axis and
T1 , T2 which generate displacements in the x and y direc-
(a )n 0 0 (x D)n x2 /2 tions. The operators T1 and T2 commute. The operators
hx|ni = hx| |x ihx |0i = p e
n! 2n n! L3 , T = T1 iT2 satisfy commutation relations
2
Hn (x)ex /2
= p  3 
2n n! = T T +, T = 0
 
L ,T (148)
(143)
The Hermite polynomials in Eq. (143) are defined by When acting on the plane, the three can be expressed in
terms of a radial (r) and angular () variable.
2 2
Hn (x) = e+x /2
(x D)n ex /2
(144)  
1 i
The states hx|ni are normalized to +1. L3 = T = ei + (149)
i r r
In order to construct the recursion relations for
the
Hermite polynomials, choose O = x = (a + a )/ 2 in Basis vectors |mi are introduced that satisfy the condi-
Eq. (137). Then tion

2
Hn (x)ex /2 1
hx|O|ni = x p = hx|n0 ihn0 |(a + a )|ni L3 |mi = m|mi hr|mi = gm (r)eim (150)
2n n! 2
(145) Single-valuedness requires m is an integer. Adjacent ba-
The two nonzero matrix elements on the right are given in sis vectors are defined by
Eq. (141). They couple xHn (x) on the left with Hn1 (x)
on the right. When the expression is cleaned up the  
d m
standard recursion relation is obtained: T |mi = |m 1i gm (r) = gm1 (r)
dr r
(151)
2x Hn (x) = Hn+1 (x) + 2n Hn1 (x) (146) Finally, the identity T + T |mi = |mi gives Bessels equa-
tion
The differential relation
is obtained in the same
way,
replacing x = (a + a )/ 2 by D = (a a )/ 2 in Eq.
(145). On the left hand side we find the derivative of m2
 
1 d d
2 r +1 2 gm (r) = 0 (152)
Hn (x) as well as the derivative of ex /2 , and on the r dr dr r
right hand side a linear combination of Hn1 (x). When
the expression is cleaned up there results the standard
differential relation F Addition Theorems

0 Addition theorems reflect the group composition prop-


Hn (x) = 2n Hn1 (x) (147)
erty through the matrix multiplication property of rep-
resentations:
E Differential Equation
hn|g(x)g(y)|n0 i
It happens often that an operator can be formed that . & (153)
0 0
P
is quadratic in the basis vectors of the Lie algebra and k hn|g(x)|kihk|g(y)|n i = hn|g [z(x, y)] |n i
29

The special function at argument z is expressed as a pair- X. SUMMARY


wise product of special functions evaluated at the oper-
ations x and y for which x y = z. The bestknown of The study of symmetry has had a profound influence
these addition results is on the development of the natural sciences. Group The-
ory has been used in constructive ways before groups
D00l l
() = D0m (g11 )Dm0l
(g2 ) even existed. We have given a flavor of what can be
l l l
D00 () = Dm0 (g1 )Dm0 (g2 ) done with symmetry and related arguments in Sec. II,
2l + 1 X (154) which describes three types of arguments that live in
Pl (cos ) = Yml (1 , 1 )Yml (2 , 2 ) the same ballpark as Group Theory. Groups were for-
4 m mally introduced in Sec. III and a number of examples
Here we have taken g1 = (1 , 1 , ) and g2 = (2 , 2 , ), given, ranging from finite groups to Lie groups. Trans-
and is the angle between these two points on the sphere formation groups played a big role in the development
surface, defined by of classical physics: mechanics and electrodynamics. In
fact, it was the need to formulate these two theories so
that their structure remained unchanged under transfor-
cos = cos 1 cos 2 + sin 1 sin 2 cos(2 1 ) (155) mations from the same group that led to the Theory of
Special Relativity. The group at hand was the Inhomo-
geneous Lorentz Group, the 10-parameter Lie group of
G Generating Functions Lorentz transformations and translations acting on fields
defined over space-time. Sec. IV describes how group
Generating functions are constructed by computing theory played a role in the development of Special Rel-
the exponential of an operator O in the Lie algebra in ativity. The next step beyond requiring invariance un-
two different ways and then equating

the results. We il- der the same Lorentz transformation at every space-time

lustrate for H3 by computing hx|e 2ta |0i. We first com- point involved allowing the Lorentz transformation to
pute the brute strength Taylor series expansion of the vary from point to point in a continuous way and still
exponential: requiring some kind of invariance (c.f., Gauge Theories
as discussed in Sec. VIII). This extended aesthetic led
to the Theory of General Relativity.
2 1 X tn Hn (x)
hx|et(xD) |x0 ihx0 |0i = ex /2
(156) Up to this point Physicists could have done without
4
n=0 n! all the particular intricacies of Group Theory. The next
step in the growth of this subject was the intensive study
Eq. (144) was used to obtain this result. of the linear representations of groups. The growth was
Next, we observe that exponentials of differential op- indispensible when Quantum Theory was developed, be-
erators are closely related to Taylor series expansions, for cause groups acted in Hilbert spaces through their linear
d
instance et dx f (x) = f (x t). To exploit this we use the matrix representations. We provided an overview of rep-
result of the disentangling theorem Eq. (68) to write resentation theory in Sec. V. At first, groups were ap-
plied in the Quantum Theory as symmetry groups (c.f.,
et(xD) = etx et
2
/2 tD
e (157) Sec. VI. In this capacity they were used to describe
the degeneracies in energy levels that were required by
Then symmetry. Shortly afterward they were used in a more
daring way to describe nondegenerate levels related to
2 each other either by a broken symmetry or simply by op-
hx|et(xD) |x0 ihx0 |0i = etx et /2 tD
e hx|0i = erators that had little to do with symmetry but had the
(158) good sense to close under commutation with each other.
1 tx t2 /2 (xt)2 /2
e e e Some applications of Dynamical Groups are described in
4
Sec. VII.
By comparing the two calculations, Eq. (156) with Eq. Gauge theories were briefly treated in Sec. VIII. In
(158), we find the standard generating function for the such theories one begins with a symmetry group and re-
Hermite polynomials. quires that a Hamiltonina, Lagrangian, or Action remain
invariant under the transformation when the parame-
n ters of the transformation group are allowed to be func-
2 X t Hn (x) tions over space-time. It is remarkable that this require-
e2xtt = (159)
n=0
n! ment leads to the prediction of new fields, the nature of
the interaction of the new fields with the original fields,
websearch: Lie groups and special functions the structure of the equations of the new fields, and the
websearch: wigner and talman mass spectrum of these new fields: all new masses are
websearch: vilenkin and special functions zero. This problem was overcome by proposing that a
websearch: w. miller and special functions
30

new particle, now called the Higgs Boson, exists. Its knowing at least a little bit about group theory. Group
discovery was announced in 2012. theory has migrated from the outer fringes of physics
As a closing tribute to the theory of groups and their (gruppenpest, 1928) to the central player, even the lin-
linear matrix representations, we hint how the entire the- gua franca, of modern physics.
ory of the Special Functions of Mathematical Physics,
which was created long before the Lie groups were in-
vented, is a study of the properties of specific matrix ele- FURTHER READING
ments of specific matrix representations of particular Lie
groups acting over special submanifolds of the differen- Gilmore, R. (2008), Lie Groups, Physics, and Ge-
tiable manifold that parameterizes the Lie group. These ometry, An Introduction for Physicists, Engineers, and
ideas are sketched by simple examples in Sec. IX. Chemists, Cambridge: Cambridge University Press.
Group theory has progressed from the outer fringes Griffiths, D. (2008), Introduction to Elementary Par-
of theoretical physics in 1928, when it was referred to ticles, Weinheim, Germany: Wiley-VCH.
as the gruppenpest (1928 Weyl to Dirac at Princeton), Hamermesh, M. (1962), Group Theory and its Ap-
through the mainstream of modern physics, to wind up plication to Physical Problems, Reading, MA: Addison-
playing the central role in the development of physical Wesley; reprint (1989), New York: Dover.
theory. Theoretical Physicists now believe that if a the- Sternberg, S. (1994), Group Theory and Physics, Cam-
ory of fundamental interactions is not a gauge theory it bridge: University Press.
doesnt have the right to be considered a theory of inter- Talman, J. D. (1968), Special Functions: A Group
actions at all. Gauge theory is the new version of simple Theoretic Approach (Based on Lectures by Eugene P.
and elegant. Wigner), New York: Benjamin.
We learn that Nature was not tamed until Adam was Tinkham, M. (1964), Group Theory and Quantum Me-
able to give names to all the animals. Just so, we cant chanics, New York: McGraw Hill.
even give names to particles and their states without