Physics 129b Lecture 1 Group Theory

Physics 129b Lecture 1 Caltech, 01/07/20
Introduction
What is a group? From Wikipedia: A group is an algebraic structure consisting of a set of elements
together with an operation that combines any two elements to form a third element.
Example: A set of elements: integers n. If we add two integers, we get a third integer!
The operation satisfies four conditions called the group axioms, namely closure, associativity, iden-
tity and invertibility.
• Closure: Combining two elements in the set always produces a third element in the same set.
n + m = (m + n).
• Associativity: When combining more than two elements, the result does not depend on the
order of combination. (n + m) + p = n + (m + p).
• Identity: There exists an identity element e in the set such that combining this element with
any other element a in the set gives a. 0 + n = n.
• Invertibility: For every element a in the set, there exist an element −a such that the combi-
nation of the two gives e. n + (−n) = 0.
Obviously, the set of integers, together with the addition operation, form a group, according to this
definition. But of course, the notion of group applies more generally.
Why is it interesting to physics?
In physics, group theory is used to describe the symmetry of the physical system.
Example: reflection symmetry of water molecule H2O.
As an electron moves around the atoms in the water molecule, the potential energy it sees would
be invariant under the left-right reflection operation (P ). If we want to find the distribution of the
electron ‘cloud’ around the atoms, we need to solve the Schroedinger equation of an electron moving
around one Oxygen and two Hydrogen atoms. That is a complicated task. However, without doing
1
any of the calculation, intuition tells us the distribution is more likely to look like the figure on the
right than the figure on the right because the figure on the right is more symmetric. That is, it is
more compatible with the reflection symmetry of the atomic configuration.
Combination of two reflection operations corresponds to doing one after another. Doing reflection
P twice is equivalent to doing nothing (the identity operation I). It is easy to check that the
set formed by {I, P } satisfies the above group axioms. Therefore, the set of symmetry operations
{I, P } form a group.
More generally, symmetry operations of a physical system always form a group and studying the
group structure will help us to simplify the physical analysis at hand and understand better the
dynamics of the system.
Actually, symmetry has become one of the guiding principles in formulating physical laws, because
physics should be ‘beautiful’. The following pictures contain examples of systems with different
types of symmetry.
Figure 1: Y reflection symmetry.
Syllabus
This course aims to introduce the basic concepts of group theory and discuss the application of
group theory in various fields of physics.
• General properties of groups (subgroup, coset, quotient group, homomorphism, direct prod-
uct)
2
Figure 2: X reflection symmetry.
Figure 3: Rotation symmetry.
• Group representations (Irreducible representation, properties, CG coefficient)
• Finite groups (permutation groups, cyclic groups, classification)
• Lie groups (SO(3), SU (2), Lie algebra, general properties of simple Lie groups)
• Application: atomic spectra
• Application: special relativity
• Application: standard model
• Application: condensed matter
3
Figure 4: Translation symmetry.
Figure 5: Scale invariance
Logistics
• Instructor: Xie Chen, Office: 163 W. Bridge, Mail Code: 149-33, Phone: x3793, Email:
xiechen@caltech.edu. Office hour: every Wednesday, 4pm-5pm at 163 W Bridge.
• TA: Chi Fang (Anthony) Chen, chifang@caltech.edu; Nabhah Shah, nnshah@caltech.edu.

Recitation and office hour to be determined.
• Prerequisites: Linear algebra, matrix; Differential equation; Basics of classical and quantum
physics.
• Recommended textbooks:
not required, can be used for reference reading and homework.
– A. Zee, Groups, Group Theory in a Nutshell for Physicists, Princeton University Press,
2016.
– H. F. Jones, Groups, Representations, and Physics, Institute of Physics Publishing, 2nd
ed., 1998.
– Wu-Ki Tung, Group Theory in Physics, World Scientific, 1985.
– J. F. Cornwell, Group Theory in Physics: An Introduction, Academic Press, 1997.
4
Figure 6: 3D lattice symmetry
– E. P. Wigner, Group Theory, Academic Press, 1959.

– J. Talman, Special Functions, a Group-Theoretic Approach, Benjamin, 1968.
– H. Georgi, Lie Algebras in Particle Physics, Benjamin, 1982.
– P. Ramond, Group Theory: A Physicist’s Survey, Cambridge Univ. Press, 2010.
http://caltech.tind.io/record/740769?ln=en [Cambridge ebook]
– M. Tinkham, Group Theory and Quantum Mechanics, 1964.
• Course website
http://www.its.caltech.edu/∼xcchen/courses/physics129.html
• Problem Sets
Posted on the website every Thursday, due on the following Thursday by 5pm. Please put
finished homework in a box marked ’Ph129b Inbox’ in Bridge Annex. Solutions will be posted
also on the website. Graded problem sets can be picked up in a box marked ’Ph129b Outbox’
in Bridge Annex one week after the due date.
• Exams
Final exam in the last week of the term.
• Grading
60% problem sets, 40% final exam.
1 Basic definition and properties of Groups
Definition: A group G is a set of elements {a, b, ...} with a law of composition which assigns to each
ordered pair a, b ∈ G another element, written as ab, of G. The law of composition satisfies the
following conditions:
5
1. Associativity: For all a, b, c ∈ G, a(bc) = (ab)c.
2. Existence of identity: G contains an element, the identity element, denoted by e, such that for
all a ∈ G, ae = ea = a.
3. Existence of inverse: For all a ∈ G there is an element, denoted by a−1 , such that aa−1 =
a−1 a = e.
Comments:
1. In many cases, the composition is also called ‘multiplication’, but it does not have to be
multiplication of numbers / matrices etc. It can be any operation that takes two elements and
maps to a third one.
2. In general ab 6= ba, i.e. the operation of ‘multiplication’ is non-commutative. That is why the
order of the elements of the pair a, b is important. If ab = ba for all a, b ∈ G the group is called
Abelian. Otherwise, it is called Non-Abelian.
3. The definition includes the fact that the product ab is also a member of the set G. This is the
closure axiom.
4. The number of elements in the group is called the order (cardinality) of the group.
Example:
Group G Composition Law Order Remarks

Zn : integers modulo n Addition (mod n) n abelian
Positive rational numbers multiplication ∞ abelian
What is the identity element for each of the group above? Does every element has an inverse?
Counter-Example:
Set S Composition Law Order Remarks

Z: integers multiplication ∞ 1/n ∈
/Z
R: real numbers (excluding ∞) multiplication ∞ 0 has no inverse
6
2 Examples
2.1 The cyclic group Cn
The symmetry group of rotations of a regular polygon with n directed sides.
Figure 1: Directed n-gon
The elementary rotation operation that maps a directed n-gon back to itself is rotation through
an angle 2π/n. Denote this elementary rotation operation as c. Applying c for r times (r =
0, 1, ..., n − 1), we can get the all the rotation symmetry operation of a directed n-gon. In this sense
c is called a ‘generator’ of the group. We write
Cn = gp{c}, cn = e (1)
which completely specifies the group.
Applying c for n times is the same as doing nothing and we have

cn = e (2)
We say that c is an order n element in the group. In general, the order of an element a in the
group is the smallest nonzero positive integer ka such that composing ka copies of a together gives
e, aka = e. ka depends on a.
The set of elements in the group are {e, c, c2 , ..., cn−1 }. The composition rule is
ct cs = ct+s(mod n) (3)
Obviously this is an abelian group
cs ct = ct cs = cs+t(mod n) (4)
Now consider the set of integers {0, 1, ..., n − 1} together with the operation of addition modulo n.
It forms the group Zn . In fact, there is a one to one correspondence between the elements of Cn
and the elements of Zn : cs ∼ s, s = 0, 1, ..., n − 1 and their composition rules match exactly. We
say that Cn is isomorphic to Zn , Cn ' Zn .
Another way of specifying the composition rule of a group is to use the multiplication table. That
is, to list the composition result for each pair of elements a and b ∈ G in a square table. For
example, for C3 , the multiplication table reads
1
a\b e c c2
e e c c2
c c c2 e
c2 c2 e c
2.2 The Dihedral Group Dn
The symmetry group of a regular polygon with n undirected sides.
Figure 2: Undirected n-gon
Dn contains all the elements of Cn , cr , r = 0, 1, ..., n − 1. The composition of elements in this subset
remains in this subset and satisfies the associativity, the existence of identify and the existence of
inverse axioms. We say that the subset of elements {cr , r = 0, 1, ..., n − 1} form a subgroup of Dn .
Moreover, because now the sides are undirected, the polygon can be mapped back to itself by a
reflection with respect to the reflection axes (dotted lines) shown in the figure. For a n polygon,
there are n reflection axis. Denote the reflection operations as b1 ,..., bn . Each bi is an order two
element of the group, because doing reflection twice is the same as doing nothing
b2i = e (5)
Each subset {e, bi } also forms a subgroup of Dn .
The full group contains 2n elements {cr (r = 0, 1, ..., n − 1), bi (i = 1, ..., n)}
What is the relation between the set of rotation operations cr and the set of reflection operations
bi ?
First note that their composition can be non-commutative.
Consider the case of D3 . The application of b1 followed by c is different from the application of
c followed by b1 , which can be seen by tracking the position of the vertices of the triangle (with
respect to the background labeling of A, B and C)
cb1 : A → B → C, B → A → B, C → C → A
(6)
b1 c : A → B → A, B → C → C, C → A → B
In fact, cb1 = b3 , b1 c = b2 . Therefore, D3 is nonabelian.
Secondly, all group elements can be generated by c and b1 . We can write D3 = gp{c, b1 }, c3 =
e, b21 = e. However, this is not a complete description of D3 . We also need to specify the relation
2
between b1 and c. We notice that (b1 c)2 = b22 = e. Adding this condition completes the description
of D3
D3 = gp{c, b1 }, c3 = e, b21 = e, (b1 c)2 = e (7)
In general, the composition rule of the elements in D3 is

ct cs = ct+s , cs bi = bi−s , bi cs = bi+s , bi bj = cj−i (8)
where t, s = 0, 1, 2; i, j = 1, 2, 3; the arithmetic t + s, i − s etc. are all defined mod 3; the power of
c takes value in 0, 1, 2 and the subscript of b takes value in 1, 2, 3.
Some useful relations in D3 are

cb1 c−1 = b2 , cb2 c−1 = b3 , cb3 c−1 = b1 , c−1 b1 c = b3 , c−1 b2 c = b1 , c−1 b3 c = b2 (9)
That is, if we conjugate a reflection operation by rotation, we get a different reflection operation.
This is intuitive to understand: conjugating reflection by rotation corresponds to rotating the
reflection axis and through direct observation we can see that the above relations should hold.
Now let’s consider the group of D4 . D4 is similar to D3 in that it consists of rotations and reflections.
The group is of order eight with group elements
{e, c, c2 , c3 , b1 , b2 , b3 , b4 } (10)
Rotations form an order 4 subgroup {e, c, c2 , c3 } while each reflection generates an order 2 subgroup
{e, bi }.
One different between D4 and D3 is that, not all reflection axes can be rotated into each other. In
particular
cb1 c−1 = b3 , cb2 c−1 = b4 , cb3 c−1 = b1 , cb4 c−1 = b2 (11)
So there are two different types of reflection operations.
The element c2 is special in D4 in that it commutes with all the other elements (please check),
while no such element exists in D3 . We call the subgroup generated by c2 – {e, c2 } – the center of
the group.
2.3 Permutation group Sn
The permutation of n objects.
Sn contains n! elements, which permutes the ordering of objects (1, ..., n) to (p1 , ..., pn ). Composi-
tion of group elements is defines as successive application of such permutations.
• n=2
The permutation of two objects 1 and 2 involves only one nontrivial operation: the exchange
of 1 and 2. Denote this element as a = (1, 2). This notation means that object 1 is moved to
position 2 and object 2 to position 1 in this operation. Note the redundancy in this notation
because (1, 2) and (2, 1) label the same operation. To remove this redundancy, we put the
object with the smallest number in the first position.
Obviously a2 = e. S2 is isomorphic to C2 and Z2 .
3
• n=3
The permutation of three objects 1, 2 and 3 involves three exchange operations (1, 2), (2, 3), (1, 3)
and two cyclic permutation operations (1, 2, 3), (1, 3, 2). Here (i, j) means that object i is
mapped to position j and object j is mapped to position i, the third object is left untouched.
(i, j, k) means that object i is mapped to position j, j to k and k to i. The exchange opera-
tions are of order 2 while the cyclic permutation operations are of order 3. Note that we have
used the same convention to remove redundancy in the notation.
By comparing to the effect of elements in D3 on the three vertices of the triangle, it is easy
to see that S3 ' D3 .
• n=4
S4 contains 4! = 24 elements. One is the identity e. Six of them are exchange of two objects
(i, j) (i to j and j to i, others untouched) and are of order 2. Three of them are exchanges
of two pairs of objects (i, j)(k, l) (i to j and j to i, k to l and l to k), still of order 2. Eight
of them are cyclic permutations of three objects (i, j, k) (i to j, j to k, k to i, the other one
untouched) and are of order 3. Six of them are cyclic permutations of four objects (i, j, k, l)
(i to j to k to l to i) and are of order 4.
Obviously this is a different group than D4 , but we can identify D4 as a subgroup of S4
corresponding to the subset of elements
{e, (A, C), (B, D), (A, B)(C, D), (A, D)(B, C), (A, C)(B, D), (A, B, C, D), (A, D, C, B)} (12)
That is, D4 can be identified as a subgroup of S4 .
4
2 Examples
2.3 The permutation group
Cayley’s Theorem
Every finite group of order n can be considered as (is isomorphic to) a subgroup of Sn .
To prove this theorem, consider the n group elements of group G as the objects that are being
permuted by Sn . We need to demonstrate the correspondence between the group elements of G
and that of a subgroup of Sn .
First, notice that left multiplication of an element g ∈ G on all group elements {h} of G corresponds
to a permutation of these n objects. This is because, firstly, if gh1 = gh2 , then h1 = h2 , which
can be obtained by multiplying g −1 on both sides of the first equation. This implies that after
left multiplication of g, there are n different elements in {gh, h ∈ G} and they all belong to G.
Therefore, left multiplication corresponds to permutation of the n group elements in G.
Secondly, the permutation operation Pg obtained with left multiplication of g ∈ G forms a group
where the composition of Pg1 and Pg2 is simply the permutation operation obtained with left
multiplication of g1 g2 . The identity operation is Pe . The inverse of Pg is Pg−1 . And it is easy to
verify that the composition of Pg ’s is associative.
* Note that this embedding of a group of order n into the permutation group Sn is different from
the previous embedding of D4 (which has 8 elements) into S4 .
2.4 The group of integers Z
At the beginning of the class, we mentioned that the set of all integers Z form a group. The
composition rule is addition and the identity element is 0. This group is different from all the
previous examples in that there are an infinite number of elements. The group is still discrete but
not finite. The Z group can be thought of as the n → ∞ limit of the Zn groups.
To take into account infinite groups like Z, the generating set of a group needs to be more rigorously
defined as a subset such that every element of the group can be expressed as the combination
(under the group operation) of finitely many elements of the subset and their inverses. Under this
definition, we can choose either {1} or {−1} as the generating set of the group of integers.
2.5 Circle group
The symmetry group of a directed circle.
1
Figure 1: Directed circle
A directed circle has a continuous rotation symmetry. The circle is invariant under rotation by any
angle θ ∈ [0, 2π). The composition of two rotation operations corresponds to the addition of two
angles θ1 + θ2 (mod 2π).
If we use the exponential eiθ to represent the group element, then the group elements correspond
to complex numbers of absolute value 1. The composition rule becomes multiplication eiθ1 eiθ2 =
ei(θ1 +θ2 ) . The circle group has an infinite number of elements and the elements are continuous.
Therefore, the circle group is said to be continuous.
The circle group can be thought of as another n → ∞ limit of the Cn , hence the Zn , group. This
is a different limit than the group of integers Z.
2.6 Matrix group
A set of matrices can form a group.
General Linear Matrix Group: the set of n × n invertible matrices with matrix multiplication as
the composition rule forms a group.
Comments:
(1) If the entries of the matrices are real numbers, the matrix group is said to be over R and denoted
as GL(n, R). If the entries of the matrices are complex numbers, the matrix group is said to be
over C and denoted as GL(n, C).
(2) The identity element in the group is the identity matrix.
(3) The matrix group is in general nonabelian.
(4) If we restrict to the set of matrices with determinant one, we get the special linear group
SL(n, R) or SL(n, C).
(5) We can also restrict to orthogonal or unitary matrices and get the orthogonal group O(n) or
the unitary group U (n).
2
3 Basic concepts in group theory
3.1 Conjugacy class
Conjugacy, definition: two elements a and b of a group G are conjugate if there exists an element
g ∈ G such that a = gbg −1 . The element g is called the conjugating element.
Example: in the Dihedral group D3 , two reflections b1 and b2 are conjugate because b2 = cb1 c−1 .
Properties:
(1) every element is conjugate to itself a = eae−1 .
(2) if a is conjugate to b (a = gbg −1 ), then b is conjugate to a (b = g −1 ag).
(3) if a is conjugate to b (a = gbg −1 ), and b is conjugate to c (b = hch−1 ), then a is conjugate to c

(a = ghc(gh)−1 ).
Conjugacy defines a particular kind of equivalence relation among group elements and conjugate
elements are similar to each other in some ways.
For example, if a and b are conjugate to each other, then they have the same order.
To show this, assume the order of a is ka and the order of b is kb and b = gag −1 . Then
bka = (gag −1 )ka = gaka g −1 = geg −1 = e (1)
Therefore, kb is a divisor of ka . Similarly we can show that ka is a divisor of kb . Therefore, ka = kb .

Conjugacy class: Elements of a group which are conjugate to each other are said to form a conjugacy
class.
Comments:
(1) Each element of a group belongs to one and only one conjugacy class. That is, different
conjugacy classes are disjoint. (If a is conjugate with a set of bi ’s and also conjugate with a set of
cj ’s, then the bi ’s and cj ’s are also conjugate with each other and they belong to the same conjugacy
class.)
(2) The identity element forms a class by itself. (For any g ∈ G, geg −1 = e.)
(3) Each group can be partitioned into a number of disjoint conjugacy classes.
Examples:
(1) Cyclic group Cn
Because gag −1 = a for any a and g in Cn , each group element forms a conjugacy class by itself.
The number of conjugacy classes is equal to the number of group elements.
3
This is a result common to all abelian groups.
(2) Dihedral group D3
b3 b2
c
A B
b1
Figure 2: D3 as the symmetry group of undirected regular triangle
D3 contains 6 group elements {e, c, c2 , b1 , b2 , b3 }, where c is rotation by 2π/3 and bi ’s are reflection
operations. Direct calculation shows
bi cb−1 2
i = c , cbi c
−1
= bi(mod 3)+1 , (2)
That is, c is conjugate to c2 , and the bi ’s are conjugate to each other.
Therefore these 6 elements can be partitioned into three conjugacy classes (e), (c, c2 ), (b1 , b2 , b3 ).
Elements in the same conjugacy class represent similar operations: doing nothing, rotation, reflec-
tion.
(3) Dihedral group D4
D C
b4
b3
c
A b2B
b1
Figure 3: D4 as the symmetry group of undirected square
The D4 group contains 8 elements
{e, c, c2 , c3 , b1 , b2 , b3 , b4 } (3)
where c is rotation by π/2, and bi is reflection across the corresponding axis.
Direct calculation shows that
cb1 c−1 = b3 , cb3 c−1 = b1 , cb2 c−1 = b4 , cb4 c−1 = b2 , bi cb−1

i =c
3
(4)
Therefore, the 8 elements are partitioned into five conjugacy classes (e), (b1 , b3 ), (b2 , b4 ), (c, c3 ),
(c2 ). Note that while elements in the same conjugacy class have the same order, the reverse is not
true. For example, b2 and b1 are both order 2 elements, but they are not conjugate to each other.
Conjugation has a very physical meaning. gag −1 is implementing the symmetry transformation of
g on the symmetry transformation of a. For example, cb1 c−1 corresponds to rotating the reflection
4
operation of b1 by π/2. As we can see, if we rotate the reflection axis of b1 by π/2, we get
b3 . Indeed, our previous calculation shows that cb1 c−1 = b3 . Another example is bi cb−1
i which
implements reflection on the elementary rotation operation and map it to the inverse rotation
operation. Indeed our calculation shows that bi cb−1 3
i =c .
5
3 Basic concepts in group theory
3.1 Subgroup
Definition: A subgroup H of G is a subset of G which itself forms a group under the composition
law of G.
Comments:
(1) The identity element e forms a subgroup by itself.
(2) The whole group G also forms a subgroup according to this definition.
(3) Any subgroup which is different from {e} and G is called a proper subgroup.
Example: C2 = {e, b1 } and C3 = {e, c, c2 } are both proper subgroups of D3 .
Coset, definition: given a subgroup H = {h1 , h2 , ..., hr } of a group G, the left coset of an element
g ∈ G, written as gH, is defined as the set of elements obtained by multiplying all elements of H
on the left by g:
gH := {gh1 , gh2 , ..., ghr } (1)
Comments:
(1) Each coset contains r distinct elements. (If h1 6= h2 , then gh1 =

6 gh2 . Because if gh1 = gh2 ,
then multiplying both sides from the left with g −1 , we get h1 = h2 , contradicting the original
assumption that h1 and h2 are distinct elements).
(2) For any g1 , g2 ∈ G, g1 H and g2 H either completely overlap or do not overlap with each other
at all. (Suppose that the two cosets contain one pair of identical elements g1 h1 = g2 h2 . Any
other element in g1 H can be obtained by right multiplication of some hk ∈ H with g1 h1 . As
g2 h2 hk = g1 h1 hk and g2 h2 hk belongs to g2 H, for every element in g1 H we can find a corresponding
element in g2 H and vice verse. Therefore, if g1 H and g2 H overlap, then they are completely the
same.)
(3) Because it is possible that g1 H and g2 H completely overlap with each other, the labeling of a
coset as gH is not unique.
(4) For h ∈ H, hH = H.
(5) Cosets provide a different way to partition a group into disjoint sets. This partition is different
from the conjugacy class partition. In particular, each disjoint set contains the same number of
elements r.
(6) One can similarly define the right coset Hg which in general gives a different partition than the
1
left cosets.
(7) A coset other than H itself does not form a group. In particular, it does not contain the identity
element.
Lagrange’s Theorem:
The order of any subgroup of G must be a divisor of the order of G.
Corollary: Any group of prime order has no proper subgroups (e.g. Cp for p prime).
Example: For the D3 group of order 6, H = {e, b1 } of order 2 forms a subgroup. Using the
composition rule b1 c = b2 , cb1 = b3 etc., we can see that the left cosets are eH = b1 H = {e, b1 },
cH = b3 H = {c, b3 }, c2 H = b2 H = {c2 , b2 }.
Normal subgroups: A subgroup H of G is said to be normal if it satisfies gHg −1 = H for any

g ∈ G.
Comments:
(1) H only has to be invariant under conjugation as a group. Each single element of H does not
have to be invariant. Instead they can be mapped into each other. But as long as ghi g −1 stays in
H, then H is a normal subgroup.
Example: The C3 subgroup {e, c, c2 } of D3 is a normal subgroup because bi cb−1 2

i = c . But the C2
−1
subgroup {e, b1 } is not a normal subgroup because cb1 c = b2 .
(2) An equivalent definition of normal subgroup is that the left coset gH is equal to the right coset
Hg.
Quotient group
The normal subgroup is special in that the set of cosets can be endowed with a group structure by
a suitable definition of the composition of two cosets. This is called the quotient group and denoted
as G/H.
Suppose that H is a normal subgroup of G. The set of disjoint cosets {gi H} forms a group if we
define the composition of two cosets g1 H and g2 H as g1 g2 H
(g1 H) ◦ (g2 H) := g1 g2 H (2)
First we need to show that this is a good definition. When we write gH, we have chosen a particular
g to label a coset, but in many cases a different g can be chosen to label the same coset. In the
definition above, we have used a particular choice of g to define the composition rule. We need to
show that the composition rule as defined does not depend on the choice of g.
Suppose that g1 H and g10 H are the same coset, and g2 H and g20 H are the same coset. Then we
can find a h1 ∈ H such that g1 h1 = g10 . Similarly we can find a h2 ∈ H such that g2 h2 = g20 . The
composition of g1 H and g2 H gives g1 g2 H. The composition of g10 H and g20 H gives g10 g20 H. g1 g2 H
2
and g10 g20 H are the same coset because
g10 g20 H = g1 h1 g2 h2 H = g1 h1 g2 H = g1 h1 Hg2 = g1 Hg2 = g1 g2 H (3)
where for the second and fourth = we have used the fact that hi H = H, for the third and fifth =
we have used the property of normal subgroup that gH = Hg.
Therefore, the composition rule given above is well defined.
Next, we need to check the closure, associativity, identity and inverse conditions of a group.
(1) closure: if g1 H and g2 H are both cosets of H, then g1 g2 H is also a coset because g1 g2 belongs
to G if g1 and g2 both belong to G.
(2) associativity: this follows from the associativity of G.
[(g1 H) ◦ (g2 H)] ◦ (g3 H) = (g1 g2 H) ◦ (g3 H) = (g1 g2 )g3 H = g1 (g2 g3 )H = g1 H ◦ [(g2 H) ◦ (g3 H)] (4)
(3) identity: H = eH is the identity element in the set of cosets because (eH) ◦ (gH) = gH and
(gH) ◦ (eH) = gH.
(4) inverse: the inverse element of gH is g −1 H because (gH) ◦ (g −1 H) = eH = (g −1 H) ◦ (gH).
Basically, these group properties follow from the group properties of G.
Example: Let’s take G to be D3 and H to be C3 = {e, c, c2 }. H is a normal subgroup as explained

above. There are two cosets H = {e, c, c2 } and b1 H = {b1 , b2 , b3 }. They compose as
(H) ◦ (H) = H, (H) ◦ (b1 H) = b1 H, (b1 H) ◦ (H) = b1 H, (b1 H) ◦ (b1 H) = H (5)
Therefore, the quotient group G/H is isomorphic to the C2 group.
Counter-example: Consider the {e, b1 } subgroup of D3 . There are three cosets: H = b1 H = {e, b1 },
cH = b3 H = {c, b3 } and c2 H = b2 H = {c2 , b2 }. {e, b1 } is not a normal subgroup of D3 , therefore
the composition of the cosets is not well defined. Indeed we can check that
(H) ◦ (cH) = cH = H (6)
but
(b1 H) ◦ (cH) = b2 H 6= H (7)
3
3 Basic concepts of group theory
Direct Product:
The definition of the quotient group is like dividing the group G by its normal subgroup H. Is
there a way to multiply two groups together? The answer is yes.
The direct product of two groups G and H is again a group, with elements {(g, h), g ∈ G, h ∈ H}
and their composition rule is given by
(g1 , h1 )(g2 , h2 ) = (g1 g2 , h1 h2 ) (1)
It is straight forward to check that the group properties are satisfied. The direct product of G with
H is denoted as G × H.
Comments about direct product and quotient group:
(1) G × H contains subgroups G0 = {(g, e)} and H 0 = {(e, h)} which are isomorphic to G and H
respectively.
(2) Every element in G0 commute with every element in H 0 . ((g, e)(e, h) = (g, h) = (e, h)(g, e).)
(3) G0 and H 0 are both normal subgroups of G×H. ((g, h)(g1 , e)(g −1 , h−1 ) = (gg1 g −1 , e), (g, h)(e, h1 )(g −1 , h−1 ) =
(e, hh1 h−1 ).
(4) The quotient group (G × H)/G0 is isomorphic to H and the quotient group (G × H)/H 0 is
isomorphic to G.
(5) The order of G × H is the product of the order of G and the order of H.
Example: Take two C2 groups G = gp{a}, a2 = e, H = gp{b}, b2 = e. Their direct product G × H

contains four elements (e, e), (a, e), (e, b), (a, b) which forms a D2 group. Therefore, D2 = C2 × C2 .
Example: C4 = gp{a}, a4 = e also has a C2 normal subgroup {e, a2 } and the quotient group is
isomorphic to C2 . However, C4 6= C2 × C2 .
Example: D3 has a C3 subgroup and D3 /C3 = C2 . However, D3 6= C2 × C3 , because C2 is not a

normal subgroup of D3 . Instead, C2 × C3 = C6 .
Center
Definition: the center of a group, denoted as Z(G), is the set of elements that commute with every
element of G.
Z(G) = {a ∈ G | ag = ga, ∀g ∈ G} (2)
Comments:
1
(1) Z(G) is a normal subgroup of G. Actually, the requirement of the center is stronger than that
of a normal subgroup. Every element of Z(G) is invariant under conjugation.
(2) Z(G) is abelian.
(3) Z(G) = G if and only if G is abelian.
4 Group representations
4.1 Mapping between groups
A group homomorphism from group A to group B, is a mapping from group elements in A to group
elements in B, such that the group structure is preserved.
That is, the mapping f : A → B, takes each element a ∈ A and maps it to a unique element in
b = f (a) ∈ B. If a1 a2 = a3 , then f (a1 )f (a2 ) = f (a3 ), which implies that the identity element in A
is mapped to the identity element in B and f (a−1 ) = (f (a))−1 .
Comments:
(1) The set of elements {f (a), a ∈ A} in B is called the image of f . It may happen that not every
element in B is in the image of f . The image of f forms a subgroup of B.
(2) Different elements in A can be mapped to the same element in B.
(3) If different elements in A are mapped to different elements in B, then the mapping is said to
be faithful.
(4) The set of elements in A that are mapped to identity in B is said to be the kernel of f . The
kernel of f forms a subgroup of A. In fact, it is a normal subgroup of A (how to prove it?) and the
image in B form the corresponding quotient group.
(5) If f is a one to one mapping, then f is said to be an isomorphism. A faithful mapping is not
necessarily an isomorphism because it is possible that not all elements in B are in the image of f .
Example: If we choose A to be D3 , what group can the image be?
4.2 Group Representation: definition
A group is a very abstract concept. The same group can appear in many different ways. For
example, we can think of the cyclic group as the rotation symmetry of regular polygons. Or
equivalently, we can think of it as integer modulo n. To have a concrete handle when we try to
analyze properties of groups, it is useful to choose a nice way to write down the group elements and
their composition rules. It turns out to be particularly helpful to use matrices to represent groups
and that is what a group representation is.
2
Definition: a representation of a group G is a group homomorphism from G to GL(n, C), the
general linear group of invertible matrices of dimension n over complex numbers C.
Comments:
(1) Apart from GL(n, C), it is often useful to restrict the representation to general linear real
matrices GL(n, R), orthogonal matrices O(n) and unitary matrices U (n). The representation is
said to be complex, real, orthogonal or unitary respectively.
(2) The set of matrices used as representation can be thought of as linear transformations on an
underlying n dimensional vector space. Once we have chosen a basis for the vector space, the linear
representation can be written as matrices.
4.3 Examples
Example 1: D3 represented as 2 × 2 real matrices.
Consider the two dimensional real vector space. The elements of D3 can be taken to be linear
transformations of the real vector space.
b3 b2
c x
A B
b1
Figure 1: D3 as linear transformation of 2D vector space
For example, c represents √ counter clockwise rotation by 2π/3. A unit vector in the x direction
√
(1, 0) gets rotated to (− 2 , 23 ). A unit vector in the y direction (0, 1) gets rotated to (− 23 , − 12 ).
1
Therefore, the rotation can be represented by matrix

√ !
1 3
− −
D(c) = √32 2 (3)
2 − 12
c2 represents rotation by 4π/3. The corresponding matrix is

√ !
1 3
−
D(c2 ) = √2 2 (4)
− 23 − 12
3
It is straight forward to check that D(c2 ) = (D(c))2 .
b1 represents reflection across the y axis. The corresponding matrix is

−1 0
D(b1 ) = (5)
0 1
The matrix representing b2 and b3 are

√ ! √ !
3
1
√2 2
1
2√ − 23
D(b2 ) = 3
, D(b3 ) = 3
(6)
2 − 21 − 2 − 12
It is straight forward to check that this set of matrices give the same multiplication table as D3 .
In fact, all the matrices are real and orthogonal. Therefore, this is an orthogonal representation of
D3 .
4
4 Group representations
4.3 Examples
Example 1: D3 represented as 2 × 2 real matrices.
√ ! √ !
−1 − 23 −√12 3

1 0
D(e) = , D(c) = √32 , D(c2 ) = 2 ,
0 1
2 − 12 ! − 23 − 21
√
3
√ ! (1)
1 1
− 23

−1 0 √2 2 2√
D(b1 ) = , D(b2 ) = 3
, D(b3 ) = 3
0 1
2 − 21 − 2 − 12
Example 2: Circle group as rotation of 2D real vector space
The elements of the circle group can be taken to be continuous rotations of the 2D vector space.
The group element labeled by θ corresponds to the 2 × 2 matrix

cos θ − sin θ
D(θ) = (2)
sin θ cos θ
This is again a real orthogonal representation.
Example 3: Circle group as rotation of 3D real vector space
We can also imagine that the circle group represent rotations of 3D real vector space around the z
axis. The group element labeled by θ corresponds to 3 × 3 matrix
 
cos θ − sin θ 0
D(θ) =  sin θ cos θ 0 (3)
0 0 1
Example 4: Circle group as phase factors
With real matrices, the smallest faithful representation for the circle group is two dimensional.
However, if we are allowed to use complex numbers, the circle group has a one dimensional faithful
representation
D(θ) = eiθ (4)
4.4 Properties
1. Equivalent representations
1
Definition: Two n dimensional representations D(1) and D(2) of a group G are equivalent if all the
matrices D(1) (g) and D(2) (g) are related by the same similarity transformation
D(1) (g) = SD(2) (g)S −1 (5)
Comments:
(1) Equivalent representations have the same dimension.
(2) S is independent of g and this relation has to be satisfied for all g.
(3) This equivalence relation is consistent with the group property of the two representations. That
is, if D(2) (g1 )D(2) (g2 ) = D(2) (g1 g2 ), we have
D(1) (g1 )D(1) (g2 ) = SD(2) (g1 )S −1 SD(2) (g2 )S −1 = SD(2) (g1 g2 )S −1 = D(1) (g1 g2 ) (6)
(4) The relation between equivalent representations basically amounts to a basis change of the
underlying (real or complex) vector space. Equivalent representations are considered the same.
(5) Maschke’s Theorem: For finite (or more generally compact) groups, representations are always
equivalent to unitary representations.
2. Character
If we want to have a way to characterize representations such that equivalent representations are
characterized the same way, we can use the character.
Definition: The character of a representation D of a group G is the set χ = {χ(g)|g ∈ G}, where
χ(g) is the trace of the representation matrix D(g)
n
X
χ(g) = T r(D(g)) = D(g)i,i (7)
i=1
Comments:
(1) Matrices related by similarity transformation have the same trace
(2) Equivalent representations have the same character.
(3) Two representations with the same character are equivalent. This is a highly nontrivial and
highly useful fact. We are not going to prove it now, but we are going to see how it comes up when
we talk about irreducible representations.
(4) χ(e) = n
(5) χ(g) = χ(hgh−1 ), that is, conjugate elements have the same trace.
There was a question in class regarding whether different conjugacy classes always have different
traces in a faithful representation. The answer is no and here is a counterexample. For the
2
group D4 (the symmetry group of a regular square), there is a faithful two dimensional irreducible
representation generated by

0 1 −1 0 0 −1
D(c) = , D(b1 ) = , D(b2 ) = (8)
−1 0 0 1 −1 0
There are five conjugacy classes with traces D(e) = 2, D(c) = D(c3 ) = 0, D(b1 ) = D(b3 ) = 0,
D(b2 ) = D(b4 ) = 0, D(c2 ) = −2.
3. Reducibility
Example: circle group in three dimensional real vector space
Recall that the representation of the circle group in three dimensional real vector space reads
 
cos θ − sin θ 0
D(θ) =  sin θ cos θ 0 (9)
0 0 1
All the D(θ) matrices take a block diagonal form. When they multiply, different blocks do not talk
to each other. For the circle group, this is related to the fact that the z axis is invariant under the
rotation operation, while vectors in the x − y plane get rotated into each other. Therefore, this
representation effectively ‘decomposes’ into two separate representations, a 2D representation D(2)
acting on the x − y coordinates and the 1D representation D(1) acting on the z coordinate, which
are completely independent.
We write
D(θ) = D(1) (θ) ⊕ D(2) (θ) (10)
⊕ denotes direct sum, which is to combine matrices in block diagonal form.
Definition: A representation of dimension n+m is said to be (completely) reducible or decomposible

if there exists a basis transformation S such that SD(g)S −1 is of the form

−1 A(g) O
SD(g)S = (11)
O B(g)
for all g ∈ G, where A, B are sub-matrices of dimension m × m and n × n respectively. O is a null

matrix (all entries 0) of dimension n × m and m × n.
Comment: (1) A(g) and B(g) each form a representation of G. A(g1 )A(g2 ) = A(g1 g2 ), B(g1 )B(g2 ) =
B(g1 g2 ).
(2) A and B could be decomposible themselves and we can continue the process until every block
is irreducible.
(3) If we think of representation matrices as linear transformations of vector space, then blocks in
a reducible representation correspond to closed subspaces under the linear transformations.
The irreducible representations provide building blocks for general decomposible representations
and this is what we are going to focus on when we try to study in more detail about representations.
3
5 Irreducible Representations
Nick name: irreps.
5.1 Examples
(1) C2 = {e, c}: the cyclic group of order two has two irreducible representations.
D(e) = 1, D(c) = 1 and D(e) = 1, D(c) = −1 (12)
Comments:
a. The first one is trivial (everything mapped to 1) while the second one is nontrivial.
b. Both of them are one dimensional.
c. If we consider the two irreps as two two-component vectors, they are orthogonal to each other.
(2) D3 = {e, c, c2 , b1 , b2 , b3 }:
The dihedral group of order six has three irreps. The first one is trivial (everything mapped to 1).
The two nontrivial ones are:
√ !
1 3

− 2 − 2 −1 0
D(c) = 1, D(b1 ) = −1 and D(c) = √
3
, D(b1 ) = (13)
2 − 12 0 1
Comments:
a. the D3 group has two 1D irreps and one 2D irrep. In particular, the 2D representation cannot
be decomposed into two 1D representations, because no vector in the 2D vector space remains
invariant under the transformation of the whole group.
b. We have specified the irreps just by the matrices of their generators. If we write out the matrices
for all group elements, they are
irreps e c c2 b1 b2 b3
D(1) 1 1 1 1 1 1
D(2) 1 1 1 −1 −1 −1
√ ! √ ! √ ! √ !
1
− 23 −√12 3 1 3 1
− 23

1 0 − −1 0
D(3) √2
3
2 √2 2 2√
0 1
2 − 12 − 23 − 12 0 1
2
3
− 21 − 2
3
− 12
(14)
c. In D(1) and D(2) , different group elements can be mapped to the same matrix (number). In
D(3) , different group elements are always mapped to different matrices. D(3) is said to be a faithful
representation, while D(1) and D(2) are not faithful. The kernel of D(1) is the whole group, the
kernel of D(2) is the rotation subgroup (a normal subgroup), and the kernel of D(3) is the identity
element.
4
d. D(1) and D(2) are orthogonal six-component vectors. Is D(3) also orthogonal to them in some
way?
The answer is yes. Let’s take the element at position i, j (i, j = 1, 2) from each of the matrices in
D(3) .
e c c2 b1 b2 b3
(3) 1 1 1 1
D [1, 1] 1 −√2 −
√2
−1 √2 2√
D(3) [1, 2] 0 −√ 23 3
2√ 0 √23 − √23 (15)
(3) 3 3 3 3
D [2, 1] 0 2 − 2 0 2 − 2
1 1
(3)
D [2, 2] 1 − 2 −2 1 − 2 − 12
1
The D(3) [i, j]’s are pair-wise orthogonal. Moreover, they are orthogonal to D(1) and D(2) as well!
5
5 Irreducible Representations
Our observations regarding the above two examples actually apply more generally and are covered
by the following theorems.
5.2 Schur’s Lemma
Let D and D0 be two irreducible representations of a group G, of dimensions d and d0 respectively,

and suppose that there exists a d × d0 matrix A such that
D(g)A = AD0 (g), ∀g ∈ G. (1)
Then only the following situations are possible:
(a) if D(g) and D0 (g) are inequivalent representations, then A = 0;
(b) if D(g) and D0 (g) are equivalent representations, then A is the similarity transformation between
the two representations and det(A) 6= 0;
(c) In particular, if D(g) = D(g 0 ) (are identical representations), then A must be proportional to
the identity map.
Comments:
The Schur’s lemma applies only to irreps. It is not true for general reducible representations. For
example, let make a reducible representation of D3 by taking a direct sum of D(1) and D(2) above.

(1) (2) 2 1 0 1 0
D(g) = D (g) ⊕ D (g), D(e) = D(c) = D(c ) = , D(b1 ) = D(b2 ) = D(b3 ) =
0 1 0 −1
(2)
a 0
It is easy to see that AD(g)A−1 can be true for all g ∈ G with nonidentity matrix A = .
0 b
While we are not going to present the proof for the lemma (which can be found in many textbook
including Jones), we are going to discuss many of its important consequences.
All irreducible representations of an abelian group are one dimensional
Proof: Suppose that D(g) is an irrep for abelian group G. Pick one particular element g0 ∈ G. We
have
D(g0 )D(g) = D(g)D(g0 ), ∀g ∈ G (3)
because G is abelian. Because we have assumed that D(g) is irreducible, therefore, D(g0 ) must be
proportional to the identity matrix. Moreover, this conclusion applies for any g0 ∈ G. Therefore,
for D(g) to be irreducible, it has to be one dimensional.
1
5.3 The fundamental orthogonality theorem
Suppose that D(µ) and D(ν) are two irreducible representations of a finite group G which are not
equivalent if µ 6= ν (but which are identical if µ = ν). Then
1 X (µ) ∗ 1
D (g)[j, k] D(ν) (g)[s, t] = δµν δjs δkt (4)
|G| dµ
g∈G
This theorem is saying that the irreps of a group G are orthonormal in the following sense:
(1) For each irrep of G (in a fixed basis), take the [i, j] entry in the representation matrix for all g
and form a |G|-dimensional vector. (|G| is the order of the group).
2
P
(2) If there are nµ irreps of dµ dimension, we have a set of µ nµ dµ vectors.
(3) These vectors are all orthogonal to each other. That is, not only are vectors from different irreps
orthogonal, vectors from the same irrep but different positions in the matrix are also orthogonal.
(4) The norm (length) of each vector is |G|/dµ .
(Check that this is the case for the irreps of D3 .)
This theorem puts a strong limit on the number and dimension of irreps. In particular
X
nµ d2µ ≤ |G| (5)
µ
Actually, this inequality is saturated, as we will see later. (Check that this is the case for C2 and
D3 .)
5.4 Orthogonality of characters
Recall the definition and properties of character. The orthogonality relation for characters is
obtained by taking suitable traces of the fundamental orthogonality theorem. Tracing over indices
j, k and s, t, we get
1 P (µ) (g)[j, j] ∗ D (ν) (g)[s, s]
P P
|G| g∈G j s D ∗ P
1 P P (µ) (ν) (g)[s, s]

= |G| g∈G jD (g)[j, j] sD (6)
1 P (µ)
∗ (ν)
= |G| g χ (g) χ (g)
On the other hand, we have

1
∗
D(µ) (g)[j, j] D(ν) (g)[s, s]
P P P
|G| g∈G j s
P P 1 (7)
= j s dµ δµν δjs = δµν
Therefore,
1 X (µ) ∗ (ν)
χ (g) χ (g) = δµν (8)
|G| g
2
That is, the characters of different irreps are orthogonal to each other as |G| dimensional vectors.
Comments:
(1) Although this theorem is derived from the previous one, it has the nice property that it does
not depend on the particular basis chosen for the irreps.
(2) We can define an inner product between characters

1 X (µ) ∗ (ν)
hχ(µ) , χ(ν) i = χ (g) χ (g) = δµν (9)
|G| g
(3) A corollary of this theorem is that, the number of irreps must be smaller or equal to the number
of conjugacy classes.
Proof: Recall that the character of group elements in the same conjugacy class is the same. There-
fore, Eq. 8 can be re-written as
1 X (µ) ∗ (ν)
ki χi χi = δµν (10)
|G|
i
where i labels conjugacy classes, ki is the number of elements in the conjugacy class. Suppose that
there are p conjugacy classes in all. Therefore, the set of vectors
(µ) (µ)
p p
{ k1 χ1 , k2 χ2 , ..., kp χ(µ)
p
p }, ∀µ (11)
form an orthogonal set in the p-dimensional vector space. Therefore, the number of different irreps
is smaller or equal to the number of conjugacy classes. Actually, this inequality is again saturated
(verify this for C2 and D3 ).
3
5 Irreducible representations
5.5 Regular representation and its decomposition into irreps
To see that the inequality X

nµ d2µ ≤ |G| (1)
µ
is saturated, we need to consider the so-called regular representation.
The regular representation is the representation of a group on ‘itself’. Take all the group elements
g ∈ G and use them to label the orthonormal basis of a |G| dimensional complex vector space. (|G|
is the order of the group.) Let’s use the quantum mechanical notation and write the basis vectors
as states |gi. Notice that, now as basis state, the |gi can be put into linear combinations:
a1 |g1 i + a2 |g2 i + .... (2)
This |G| dimensional vector space is where the regular representation will act on.
Now we specify the linear transformation corresponding to each group element g ∈ G on this vector
space. As they are linear transformations, we only need to specify their action on the basis states:
D(g)|gi i = |ggi i (3)
That is, the linear transformation left multiplies the label of the basis state and permutes them.
And we can check that the D(g)’s form a representation of the group G because
D(g1 )D(g2 )|gi i = |g1 g2 gi i = D(g1 g2 )|gi i (4)
As this relation holds for every basis state |gi i, we have D(g1 )D(g2 ) = D(g1 g2 ).
Example: the regular representation of C2 is

1 0 0 1
D(e) = , D(c) = (5)
0 1 1 0
Q: is this irreducible? If so, why? If not, what are the irreducible blocks?
The regular representation is a very special representation. In the |gi i basis, the representation
matrices contains only 0’s and 1’s and each row and each column contains only one 1. In D(e),
all the 1’s are on the diagonal. In D(g), g 6= e, all the 1’s are off diagonal. Therefore we have the
following conclusion
χ(e) = |G|, χ(g) = 0 (6)
The regular representation is reducible (how to tell?) In fact, it contains every irreducible repre-
sentation of the group and each irrep µ appears dµ times where dµ is the dimension of the irrep µ.
Let’s try to see why this is the case.
1
Let’s suppose that we can decompose D(g) into irreps D(µ)
D(g) = ⊕µ aµ D(µ) (g) (7)
aµ is the integer representing the number of times each irrep appears.
How to find aµ ? Let’s take the trace on both sides

X
χ(g) = aµ χ(µ) (g) (8)
µ
where we have used the fact that the trace of the direct sum of matrices equals the sum of their
trace.
In order to find out each aµ from this sum, we take an inner product between the character of the
irrep µ and the character of the regular representation
X
hχ(µ) , χi = aµ hχ(µ) , χ(ν) i (9)
µ
As the χ(µ) are orthonormal, we have

hχ(µ) , χi = aµ (10)
On the other hand, χ = (|G|, 0, 0, 0, ...), therefore its inner product with χ(µ) gives
1
hχ(µ) , χi = |G|χ(µ) (e) = dµ (11)
|G|
Hence we have aµ = dµ . That is, the regular representation contains every irrep dµ times.
Now finally let’s look at Eq.8 again for the special case of g = e. We get
X X X
|G| = χ(e) = aµ χ(µ) (e) = aµ dµ = d2µ (12)
µ µ µ
That is, the inequality µ d2µ ≤ |G| is saturated. (In this expression, we are summing over all
P
irreps µ. If we count irreps of the same dimension once, then the relation becomes µ nµ d2µ = |G|,
P
where nµ is the number of irreps of dimension dµ .)
In fact, the inequality that the number of irreps is smaller or equal to the number of conjugacy
classes is also saturated (we don’t prove it here).
5.6 The character table
Now with many theorems at hand, one can try to find all the irreps of a group. This is a legitimate
question, but it has infinitely many different answers because the representation matrices depend
on the choice of basis. On the other hand, the character of irreps are independent of basis choice
and have a one to one correspondence with the equivalence class of irreps. Therefore, it suffices to
find all the orthonormal characters of the irreps of a group.
The characters of the irreps of a group are usually listed in a table where each row corresponds
to an irrep and each column corresponds to a conjugacy class (recall that elements in the same
2
conjugacy class have the same trace). Because the number of irreps equals the number of conjugacy
classes, this is a square table. To find the entries in the table, we make use of the following facts:
(1) µ d2µ = |G|. Because χµ (e) = dµ , the norm of the first column of the table (corresponding to
P
e) is |G|.
(2) hχ(µ) , χ(ν) i = δµν |G|.
As in the character table the character of elements in the same conjugacy class only appears once,
we have
(µ) (ν)
(3) i ki (χi )∗ χi = |G|. Here i labels different conjugacy classes and ki denotes the number of
P
elements in a conjugacy class.
Now let’s see some examples:
Character table of C3 :
The C3 group is an abelian group of order three. The three elements are {e, c, c2 } and each form
a conjugacy class. Therefore, there are three irreps and each of them being one dimensional. The
character table would be a 3 × 3 matrix.
The first irrep is the trivial one, with every element in the character table being 1. The second and
third irreps are nontrivial. The number corresponding to c needs to have order 3. Therefore, it can
only be ω = ei2π/3 and ω̄ = e−i2π/3 . The two possibilities correspond to the other two irreps.
C3 e c c2
χ(1) 1 1 1
(13)
χ(2) 1 ω ω̄
χ(3) 1 ω̄ ω
Comments:
(1) D(1) is a trivial representation while D(2) and D(3) are faithful representations.
(2) Their orthogonality can be checked in a straight forward way.
Once we have the character table, we can determine if any given representation is reducible and if
so what are the irreducible blocks. For example, consider the 2D representation of C3 as rotations
of a two dimensional plane. The representation matrices are
√ √
1 0 1 −1 3 1 −1 − 3
D(e) = , D(c) = √ 2
, D(c ) = √ (14)
0 1 2 − 3 −1 2 3 −1
Taking the trace of these matrices we get

χ(e) = 2, χ(c) = −1, χ(c2 ) = −1 (15)
As a three dimensional vector, it can be decomposed as
χ = χ(2) + χ(3) (16)
3
Therefore, this 2D representation is reducible and contains two irrep blocks equivalent to D(2) and
D(3) .
Character table of D3 :
Now let’s discuss a more interesting example, the D3 group. The D3 group contains six group ele-
ments {e, c, c2 , b1 , b2 , b3 } which belong to three conjugacy classes {e}, {c, c2 }, {b1 , b2 , b3 }. Therefore,
the character table is also 3 × 3.
The first representation is trivial with every element mapped to 1. Therefore, the characters in the
first row are all 1.
2
P
Using the relation that µ dµ = |G|, we know that there are two 1D reps and one 2D rep in total.
Next we can try to solve for the other 1D representation. As c is an order 3 element, it can only be
mapped to 1, ω, ω̄. On the other hand, as the bi ’s are order 2 elements, they can only be mapped to
±1. Because cb1 = b2 , it is not possible for c to be ω or ω̄. It has to be 1. Then the bi ’ s are either
all 1 which corresponds to the trivial irrep or all −1 which corresponds to another 1D irrep. The
character of the second irrep is {1, 1, −1}. We can check that the first two irreps are orthogonal
(remember to count the multiplicity of the conjugacy classes).
We know a 2D representation which is obtained by thinking of D3 as linear transformations of the

two dimensional plane. This representation is generated by
√
(3) 1 −1 3 −1 0
D (c) = √ (3)
, D (b1 ) = (17)
2 − 3 −1 0 1
from which we find the corresponding character to be χ(3) (e) = 2, χ(3) (c) = −1, χ(3) (b) = 0. This
is orthogonal to the previous two characters. Therefore we know that this must be the third irrep
we are looking for.
If we do not know of the explicit form of the 2D irrep, we can still solve for its character. Because
it is two dimensional χ(e) = 2. Suppose that χ(c) = α and χ(b) = β. Due to the orthogonality
condition, we have
2 + 2α + 3β = 0, 2 + 2α − 3β = 0 (18)
from which we get α = −1 and β = 0. The norm of the character is 6 as it should be.
D3 {e} {c, c2 } {b1 , b2 , b3 }

χ(1) 1 1 1
(19)
χ(2) 1 1 −1
χ(3) 2 −1 0
5.7 Direct sum of representations
Starting from two representations D(1) (g) and D(2) (g) of dimensions m and n, we can form a bigger
representation of dimension m + n by taking a direct sum of them.
(1)
(1+2) (1) (2) D (g)
D (g) = D (g) ⊕ D (g) = (20)
D(2) (g)
4
If D(1) (g) and D(2) (g) are representations, D(1+2) (g) is also a representation and their characters
satisfy
χ(1+2) (g) = χ(1) (g) + χ(2) (g) (21)
5.8 Direct product of representations
Moreover, we can multiply two representations and get a new representation, as defined below.
Suppose that we have two representations D(µ) (g) and D(ν) (g) of dimensions m and n respectively,
the direct product of the two representations is a m × n dimensional representations and its matrix
elements are given by
D(µ×ν) (g)ab,cd = D(µ) (g)a,c × D(ν) (g)b,d (22)
This is denoted as D(µ×ν) = D(µ) ⊗ D(ν) . Can you show that D(µ×ν) (g) forms a representation if
D(µ) (g) and D(ν) both form representations?
Example: Consider the 2D irrep of D3 with generators

√
(3) 1 1 − 3 1 0
D (c) = √ (3)
, D (b1 ) = (23)
2 3 1 0 −1
The direct product of two copies of this 2D irrep is generated by

√ √
− −
   
√1 3 3 3
√ 1 0 0 0
(3×3) 1
√ 3 1 −3 −√ 3  , D(3×3) (b1 ) = 0 −1 0 0
 
D (c) = 
0 0 −1 0 (24)
4 3 √ −3 √1 3
3 3 3 1 0 0 0 1
It is easy to verify that this is indeed another representation of the group, but a reducible one. To
see what irrep blocks it contains, we calculate the character of this representation as
χ(e) = 4, χ(c) = 1, χ(b) = 0 (25)
which, according to the character table of D3 , can be decomposed into
χ = χ(1) + χ(2) + χ(3) (26)
Therefore, this direct product representation can be decomposed into three different irreps, each
appearing once.
The physical meaning of the direct product representation is very different from that of the direct
sum representation. The direct sum describes the symmetry transformation of a single object,
but with states in different subspaces; the direct product describes the symmetry of a composite
system, with more than one objects transforming under the symmetry. For example, consider
the electrons orbiting a nucleus. The s orbital and p orbital both transform under the rotation
symmetry of the system and each form a representation D(s) (g) and D(p) (g) of the group. If we
want to describe the symmetry transformation of an electron which can live both in s orbital or
p orbital (or some mixture of them), we use the direct sum D(s) (g) ⊕ D(p) (g); on the other hand,
if we want to describe the symmetry transformation of two electrons in the s orbital, we use the
direct product D(s) (g) ⊗ D(s) (g). (What if we want to describe two electrons that can live in both
s and p orbitals?)
5
5 Irreducible representations
5.9 Irreps of the circle group and charge
We have been talking mostly about finite groups. Continuous groups are different, but their rep-
resentation theory can be similar in many ways. Let’s consider the simplest case of a continuous
group: the circle group.
The circle group is abelian, therefore all of its irreps are one dimensional. The trivial one maps
every group element labeled by θ ∈ [0, 2π) to 1.
A nontrivial representation is given by D(1) (θ) = eiθ . Actually, the circle group has infinitely many
different irreps given by
D(n) (θ) = einθ , n ∈ Z (1)
While the group elements in the circle group are continuous, its irreps are discrete, labeled by
integers n, which is usually referred to as ‘charge’. This is because, consider a quantum mechanical
state containing n electrons and hence −n elementary charges, where n is an integer. Denote the
number operator of the electrons as N̂ and N̂ acting on the state gives eigenvalue n.
N̂ |ψi = n|ψi (2)
Now if we apply transformation eiN̂ θ to the state, the state remains invariant up to a global phase
factor
eiN̂ θ |ψi = einθ |ψi (3)
Therefore, |ψi transforms as a 1D representation of the circle group eiN̂ θ labeled by n where n is
the number of (negative) charges contained in the state. Similarly, we would call irreps of other
symmetry groups as symmetry charges.
In general, whenever a system has a fixed number of particles (the particles could be electrons,
atoms, ions, etc. independent of how much electronic charge they carry), the system has the
symmetry. This is usually called the ‘charge conservation symmetry’ in physics. If the charge
conservation symmetry is broken, i.e. the system does not have a fixed number of particles, there
are very serious consequences. The system will be a superconductor, a superfluid or a Bose-Einstein
condensate.
As the irreps are one dimensional, their character is simply given by χ(n) (θ) = einθ . The set of
characters, as continuous functions of θ, satisfy the following orthogonality condition
Z 2π
1
<χ (n1 )
,χ(n2 )
>= dθe−in1 θ ein2 θ = δn1 ,n2 (4)
2π 0
Among all the irreps, only the ones labeled by n = 1 and n = −1 are faithful. All others are
unfaithful. Under tensor product, the set of irreps form a group – the group of integers.
D(n1 ) ⊗ D(n2 ) = D(n1 +n2 ) (5)
1
6 Applications of finite groups
A. Zee, Group theory in a Nutshell for Physicists, chapter III.2
6.1 Vibration of coupled oscillators
Now let’s see how our understanding of symmetry allows us to obtain insight into the oscillation
eigenmodes of certain systems without knowing all the details of the dynamics of the system.
Imagine that we have a mechanical system where several mass blocks are bound together and in-
teract with each other. Suppose that the equilibrium configuration and the interaction of these
mass blocks have certain symmetry. Then if they undergo small vibrations around their equilib-
rium position, the dynamics of the system also has the same symmetry. In particular, we can
find elementary ‘modes’ of such vibrational motion which form irreducible representations of the
symmetry. Let’s see how this happens in the following examples.
2 1
x02 = a x01 = a
Figure 1: A coupled oscillator with reflection symmetry with respect to x = 0.
Let’s consider a system composed of two identical blocks connected with a spring in the middle.
The spring has Hook constant k. In equilibrium, the two blocks are at x01 = a and x02 = −a
respectively, as shown in Fig.1. This equilibrium configuration is invariant under reflection with
respect to x = 0. Moreover, the form of interaction (generated by the spring) also respects this
symmetry. What does this say about the dynamics of the system? From simple analysis (or purely
physical intuition), we know that there are two eigenmodes for the motion of the two blocks. One
is the center of mass motion, where the two blocks move together without changing their relative
position. The other is relative motion, where the center of mass (the middle point of the central
spring) does not move while the two blocks move relative to each other. These two modes of motion
both transform in a special way under reflection. In the center of mass motion, the displacement
are the same for block 1 and 2, therefore reflection maps the configuration to minus itself. In
the relative motion, the displacement of the two blocks are opposite to each other, therefore, the
configuration remains invariant under reflection. Let’s try to describe this in a more concrete way.
The motion of the oscillator is described by x1 (t) and x2 (t), the displacement of the two blocks
relative to their equilibrium position. The equation of motion is given by
2
m ddtx21 = −k(x1 − x2 )
2 (6)
m ddtx22 = −k(x2 − x1 )
Or in matrix form, we can write

d2 X
m = −KX (7)
dt2

x1 k −k
where X = ,K= .
x2 −k k
2
The eigenmodes are the vibrational motions which have a sin or cos dependence on time. That is,
we are looking for solutions of the form X = X̃ cos(ωt). Plug it into the above equation we get
mω 2 X̃ = K X̃ (8)
Therefore, X̃’s are eigenvectors of K with eigenvalue mω 2 . Now, without knowing the details of K,
it seems we cannot move forward. However, because we have the knowledge about the symmetry
of the system, we can understand a lot about these X̃’s without actually solving for them.
Recall that the system is

invariant under reflection across x = 0. This corresponds to the linear
0 −1
transformation of D(b) = . The fact that the dynamics of the system is invariant under
−1 0
this symmetry translates into the relation that
D−1 (b)KD(b) = K (9)
or equivalently
KD(b) = D(b)K (10)
What can we learn from this condition? Let’s think in a group theoretical way. Notice that the
reflection operation b generates a symmetry group. In this case, it is the C2 group with one other
element e. Of course, there can be more general cases where we havea bigger symmetry
group,
but
1 0 0 −1
let’s focus on this simple example first. The matrices D(e) = and D(b) = form
0 1 −1 0
a representation of the C2 group. Moreover, this representation is reducible. From
our discussion

1 1 1
in the last lecture, we know that with basis transformation generated by H = √2 , we can
1 −1
put this reducible representation into a block diagonal form

−1 1 0 −1 −1 0
HD(e)H = , HD(b)H = . (11)
0 1 0 1
That is, this reducible representation contains both irreps of C2 andeach
irrep
is contained once.

−1 1 1 −1 0
The two irreps are supported on the one dimensional vectors H ∝ and H ∝
0 1 1

1 1 1
respectively. That is, and each form a closed space under the symmetry trans-
−1 1 −1

1 1
formation. On , the group action is represented as D(e) = 1, D(b) = −1; on , the group
1 −1
action is represented as D(e) = 1, D(b) = 1.
On the other hand, we have the condition that D−1 (g)KD(g) = K, or equivalently KD(g) =
D(g)K. From Schur’s lemma, we can show that K has to take a block diagonal form in the irrep
basis with each block being proportional to identity. That is

−1 a 0
HKH = (12)
0 b

−1 a c
In order to show this, let’s suppose that HKH = . Because K is symmetric and H is
d b
orthogonal, we can set c = d. Then KD(g) = D(g)K translates into
(1) (1)
a c D 0 D 0 a c
= (13)
c b 0 D(2) 0 D(2) c b
3
Comparing the off diagonal entries we find cD(1) = cD(2) , which can only be satisfied when c = 0
as D(1) and D(2) are inequivalent irreps. On the other hand, there is no constraint on a and b.

1 1
Eq. 12 tells us that the eigenvectors of K are and respectively. Therefore, without
1 −1
knowing the exact form of K, but by simply considering the symmetry constraints on K, we can
find all its eigenvectors. They correspond to the support space of the irreps
contained in the
1
representation D(g). And this matches with our intuitive expectation: corresponds to the
1

1
center of mass motion where the displacement of 1 and 2 are the same; corresponds to the
−1
relative motion where the displacement of 1 and 2 are opposite to each other. Of course, without
knowing exactly the form of K, we will not be able to determine the eigenvalues a and b which in
this case correspond to the oscillation frequency of the eigenmodes.
In fact, this analysis applies not just to the particular configuration in Fig.1. It applies whenever
the two blocks are coupled in a way that is symmetric under reflection. For example, we can imagine
connecting the two blocks to walls on the two sides with springs of the same Hook constant (which
can be different from the middle block), as shown in Fig.2.
2 1
x02 = a x01 = a
Figure 2: Another coupled oscillator with reflection symmetry with respect to x = 0.
In this setup, the center of mass motion and the relative motion are still eigenmodes of this coupled
oscillator, but they are going to have different oscillation frequency than in the previous case. In
particular, now the center of mass motion is going to have a finite oscillation frequency, as compared
to the previous case where the center of mass motion is not oscillating.
Ok, let’s try to summarize what we are saying here:
(1) We have a physical situation (coupled harmonic oscillator) where we need to solve an eigen
equation KX = λX.
(2) The system has some symmetry, which acts on X as matrices D(g). The symmetry condition
translates into KD(g) = D(g)K.
(3) Under certain basis transformation H, D(g) can be put into a block diagonal form
(1)
−1 D (g)
HD(g)H = (14)
D(2) (g)
(4) If D(1) and D(2) are inequivalent one dimensional irreps, then under the same basis transfor-
mation, we have
−1 a
HKH = (15)
b
4

1 0
(5) The eigenmodes of K are H −1 and H −1 . Each of them transforms under the symmetry
0 1
as an irrep.
(6) There is not much we can say about the eigenvalues a and b which depends on the details of K.
5
6.2 A more complicated coupled harmonic oscillator
In the previous example, we considered a coupled harmonic oscillator with two degrees of freedom
and a reflection symmetry of group C2 . We find that the eigenmodes correspond to the support
space of the irreps of C2 so they are completely fixed by the symmetry while the eigenvalues cannot
be determined from symmetry consideration alone.
y
1
2 3
x
b1
Figure 1: A coupled oscillator with permutation symmetry among 1, 2 and 3.
Now let’s consider a slightly more complicated example of three blocks coupled together as shown
in Fig.1, where we will run into the situation of higher dimensional irreps and multiple copies
of equivalent irreps. Small (in plane) oscillation around this equilibrium position involves six
dynamical degrees of freedom: the displacement of 1, 2 and 3 in x and y directions respectively
(x1 , y1 , x2 , y2 , x3 , y3 ). The system has a D3 symmetry involving three fold rotation and reflection.
What can we tell about the oscillation eigenmodes from symmetry considerations?
First, let’s try to see how this D3 symmetry is represented on the six displacement coordinates.
By working out how the displacement coordinates transform into each other, we find that the
generators are represented as
 √ 
0 0 0 0 −1 − 3 −1 0 0 0 0 0
 
√
 0
 0
√ 0 0 3 −1    0 1 0 0 0 0
 
1 −1 − 3 0

0 0 0 
  0 0 0 0 −1 0
D(c) =  √  , D(b1 ) = 
  (1)
2  3 −1 0 0
√ 0 0   0 0 0 0 0 1


 0 0 −1 − 3 0 0 
  0 0 −1 0 0 0
√
0 0 3 −1 0 0 0 0 0 1 0 0
This is a reducible representation. The character of the representation is χ = {6, 0, 0, 0, 0, 0} from
1
which we see that
D = D(1) ⊕ D(2) ⊕ 2D(3) (2)
(D(1) , D(2) , D(3) as defined in the previous lectures.) That is, this six dimensional displacement
space decomposes into one trivial irrep, one nontrivial 1D irrep and two copies of the 2D irrep. We
can find a basis transformation S such that SDS −1 is in a block diagonal form with four blocks.
 (1)
D (g)

D(2) (g)
SD(g)S −1 = 
 
(3)
 (3)
 D (g) 
D(3) (g)
Note that because the last two blocks are the same, we can mix them by doing some basis trans-
formation between these two blocks without changing the structure of the blocks.
What do these blocks correspond to? D(1) corresponds to the mode where the triangle shrink or
√ √ T
expand as a whole, the corresponding vector is v1 = 0, −1, 23 , 12 , − 23 , 12 . In particular, we can
check that
D(g)v1 = v1 , ∀g ∈ G (4)
D(2) corresponds to the rotation of the triangle while expanding, the corresponding vector is v2 =
√ √ T
1, 0, − 21 , 23 , − 21 , − 23 , which satisfies
D(c)v2 = v2 , D(b1 )v2 = −v2 (5)
The basis for the next two blocks are not uniquely fixed. But we can see that the center of mass mo-
tion corresponding to the space spanned by vectors v3 = (1, 0, 1, 0, 1, 0)T and v4 = (0, 1, 0, 1, 0, 1)T
transform as D(3) while the remaining two dimensions transform as another D(3) .
What does this tell us about the eigenmodes of the oscillator? The fact that the system has a D3
symmetry implies that D(g)K = KD(g). We are going to use Schur’s lemma again to argue about
the structure of K. In particular, let’s suppose that under the basis transformation S, the matrix
K takes the form  
a b A B
 b c C D
SKS −1 =   AT C T E F 
 (6)
B T DT F T G
where a, b, c are numbers, A, B, C, D are 1 × 2 matrices and E, F, G are 2 × 2 matrix. Because D(g)
and K commute, so do SD(g)S −1 and SKS −1 . Therefore, we have
D(1) (g)
  
a b A B

 D(2) (g)  b
 c C D 
(3) A T C T E F
 D (g)  
D(3) (g) B T DT F T G
  (1) (7)
D (g)
 
a b A B
 b c C D  D(2) (g) 
=
 AT T

(3)

C E F   D (g) 
BT D T F T G (3)
D (g)
2
The left hand side is equal to
aD(1) (g) bD(1) (g) D(1) (g)A D(1) (g)B
 
 bD(2) (g) cD(2) (g) D(2) (g)C D(2) (g)D

D(3) (g)AT
 (8)
D(3) (g)C T D(3) (g)E D(3) (g)F 
D(3) (g)B T D(3) (g)DT D(3) (g)F T D(3) (g)G
and the right hand side is equal to
aD(1) (g) bD(2) (g) AD(3) (g) BD(3) (g)
 
 bD(1) (g) cD(2) (g) CD(3) (g) DD(3) (g)

A D (g) C D (g) ED(3) (g)
T (1) T (2)
 (9)
F D(3) (g) 
B T D(1) (g) DT D(2) (g) F T D(3) (g) GD(3) (g)
Now we are going to use Schur’s lemma which states that if AD(µ) (g) = D(ν) (g)A for inequivalent
irreps µ and ν, then (µ) (µ)
A= 0 and if AD (g) = D (g)A, A ∝ I. We get b = 0, A = B = C = D = 0,
1 0
E, F, G ∝ I2 = . Therefore,
0 1
 
a
 c
SKS −1 = 

 (10)
 eI2 f I2 
f I2 gI2
where a, c, e, f, g are numbers.
We can see that the support space of the two one dimensional irreps each form an eigenmode of
the coupled oscillator.
We still have four dimensions left, corresponding to thetwo copies of the two dimensional irrep.
e f
In the four dimensions, SKS −1 takes the form ⊗ I2 . We can further perform a basis
f g

e f
transformation and diagonalize the part. If we denote the total basis transformation as
f g
S 0 , then  
a
−1  c
S 0 KS 0 = 

0
 (11)
 e I2 
g 0 I2
Therefore, we conclude that in this four dimensional space, K can be decomposed into two diagonal
blocks, each of two dimensions. Symmetry transformation on each of the two dimensional subspace
form an irrep D(3) and the two dimensions have the same eigenvalue.
Let’s try to summarize what we learned from the previous examples.
(1) We have a physical problem which can be reduced to solving an eigenvalue equation of a
symmetric matrix K: KX = λX.
(2) The system has certain symmetry and the degrees of freedom transform under the symmetry
as D(g), g ∈ G. The matrix K satisfy D(g)KD(g)−1 = K.
3
(3) If we decompose D(g) into irreducible blocks, we have D(g) = ⊕nµ D(µ) (g).
Then we can make the following conclusions:
(4) Suppose that for some irrep µ of dimension dµ , nµ = 1. The corresponding dµ dimensional
vector space forms a degenerate subspace of eigenvectors. All the vectors in this vector space are
eigenvectors and they all have the same eigenvalue.
(5) For some other irrep µ of dimension dµ , nµ may be larger then 1. In the corresponding nµ dµ
dimensional vector space, we can find a number of nµ degenerate subspaces, each of dimension dµ .
From symmetry considerations alone, we cannot determine the basis for each subspace.
(6) From symmetry considerations alone, we cannot determine the eigenvalue of each degenerate
subspace. Among the degenerate subspaces, some of them may share the same eigenvalue, but this
is not guaranteed by symmetry and is said to be accidental.
(7) The more symmetry the system has, the more constraint we can put on the eigenvectors. For
example, suppose that with symmetry G vectors X a and X b belong to equivalent irreps while with
symmetry G0 (G ⊂ G0 ) they belong to inequivalent irreps. Therefore, with symmetry G0 we can
conclude that X a and X b cannot be mixed to form an eigenvector while with symmetry G alone
we cannot make the statement.
To summarize: The eigenvectors of K can be grouped into irreps. Eigenvectors in the same irrep
must be degenerate (have the same eigen-frequency) while eigenvectors in different irreps (which
may or may not be equivalent to each other) generically have different eigen-frequency. A short
take home message is: (Degenerate) eigen-spaces are labeled by irreps of the symmetry group.
4
6.2 A more complicated coupled harmonic oscillator
Let’s try to summarize what we learned from this example.
(1) We have a physical problem which can be reduced to solving an eigenvalue equation of a real
symmetric matrix K: KX = λX.
(2) The system has certain symmetry and the degrees of freedom transform under the symmetry
as D(g), g ∈ G. The matrix K satisfy D(g)KD(g)−1 = K.
(3) If we decompose D(g) into irreducible blocks, we have D(g) = ⊕nµ D(µ) (g).
Then we can make the following conclusions:
(4) Suppose that for some irrep µ of dimension dµ , nµ = 1. The corresponding dµ dimensional
vector space forms a degenerate subspace of eigenvectors. All the vectors in this vector space are
eigenvectors and they all have the same eigenvalue.
(5) For some other irrep µ of dimension dµ , nµ may be larger then 1. In the corresponding nµ dµ
dimensional vector space, we can find a number of nµ degenerate subspaces, each of dimension dµ .
From symmetry considerations alone, we cannot determine the basis for each subspace.
(6) From symmetry considerations alone, we cannot determine the eigenvalue of each degenerate
subspace. Among the degenerate subspaces, some of them may share the same eigenvalue, but this
is not guaranteed by symmetry and is said to be accidental.
(7) The more symmetry the system has, the more constraint we can put on the eigenvectors. For
example, suppose that with symmetry G vectors X a and X b belong to equivalent irreps while with
symmetry G0 (G ⊂ G0 ) they belong to inequivalent irreps. Therefore, with symmetry G0 we can
conclude that X a and X b cannot be mixed to form an eigenvector while with symmetry G alone
we cannot make the statement.
To summarize: The eigenvectors of K can be grouped into irreps. Eigenvectors in the same irrep
must be degenerate (have the same eigen-frequency) while eigenvectors in different irreps (which
may or may not be equivalent to each other) generically have different eigen-frequency. A short
take home message is: (Degenerate) eigen-spaces are labeled by irreps of the symmetry group and
transform under symmetry as the corresponding irrep.
6.3 Electron wave function in a lattice
A very similar line of argument can be applied to constrain quantum dynamics using symmetry.
1
Suppose that we want to study the motion of an electron, which must be formulated using quantum
mechanics. The fundamental equation of motion to use now is Schrödinger’s equation, instead of
Newton’s law.
∂
i~ Ψ(~r, t) = HΨ(~r, t) (1)
∂t
We look for states with a periodic time dependence of the form Ψ(~r, t) = Ψ0 (~r)e−iωt and the
equation reduces to
HΨ0 (~r) = ~ωΨ0 (~r) (2)
which is the eigenvalue equation for H. In the Hilbert space of the electron, H is a Hermitian
operator and has a complete set of orthogonal eigenstates with real eigenvalue (which corresponds
to the energy of the eigenstates).
If the system has certain symmetry which is represented on the Hilbert space as D(g), then
D−1 (g)HD(g) = H (3)
Now the algebra becomes exactly the same as in the classical case, and we can reach the same
conclusion:
(1) The eigenstates of H can be grouped into irreps. Eigenstates in the same irrep must be
degenerate (have the same energy) while states in different irreps (which may or may not be
equivalent to each other) generically have different energy.
(2) Once we have figured out all the irreps contained in D(g), the eigenstates are obtained through
linear combinations of equivalent irreps. Inequivalent irreps cannot be superposed to form eigen-
states.
Let’s consider the example of an electron moving in a one dimensional chain with discrete translation
symmetry. Instead of thinking about infinite system at the beginning, let’s start from a finite size
lattice and later take the limit of system size going to infinity. To preserve translation symmetry
in a finite system, we use the so-called ‘periodic boundary condition’ and put the system on a ring.
Suppose that the system contains N regularly spaced atoms and the electron moves in their poten-
tial. The hamiltonian
p2
H= + V (x) (4)
2m
2
is invariant under translation by lattice constant a, generated by eipa/~ .
Translation symmetry then forms a CN group. The hamiltonian of the electron being translation
invariant means that each of the eigenstates corresponds to one of the irreps of CN . CN has N
irreps, given by
1, ω, ω 2 , ω 3 , ...ω N −1 , ω = ei2πn/N , n = 0, 1, ..., N − 1 (5)
Each irrep is labeled by n, which is an integer mod N .
Define quantity
2πn
k= (6)
Na
where a is the lattice constant (physical distance between two lattice sites). k is called the wave
number of the wave function. ~k is called the crystal momentum of the electron. The irrep labeled
by n can then be labeled by k and the representation matrices (numbers) are give by
2π 4π 2π(N − 1)
1, ω, ω 2 , ω 3 , ...ω N −1 , ω = eika , k = 0, , , ..., (7)
Na Na Na
Now imagine taking the limit of N going to infinity. The possible values of k increase in number,
and eventually populate every point between 0 and 2π a . Equivalently, we can take k to take value
π π
between − a and a .
Recall the result we obtained previously.
The eigenstates of a Hamiltonian can be grouped into irreps of the symmetry group of the Hamil-
tonian.
Which means
The eigenstates transform under the symmetry, and the transformation matrix is exactly given by
the representation matrix in the corresponding irrep.
Let’s see how this conclusion applies to our case of lattice translation symmetry.
A consequence of translation symmetry is that, every eigenstate of the electron corresponds to an

irrep of the translation symmetry labeled by k. That is, the wave function Ψ(r) transforms under
translation as
T (a)Ψ(r) = eika Ψ(r) (8)
It can be shown that the wave function takes the form
Ψ(r) = eikr u(r) (9)
where r labels positions on the ring and u(r) is a periodic function with period a. That is u(r)
is invariant under translation along the lattice. Let’s check how it transforms under translation
symmetry
T (a)Ψ(r) = Ψ(r + a) = eik(r+a) u(r + a) = eika eikr u(r + a) = eika Ψ(r) (10)
That is, Ψ(r) indeed forms a 1D rep of the translation symmetry labeled by k, as we would have
expected.
3
Now let’s move on to two dimension and consider first the square lattice. Imagine a square lattice
on the surface of a torus with periodic boundary conditions in both directions. We have translation
symmetry in both x and y direction and it forms a CNx × CNy group. The irreps of the CNx × CNy
group is labeled by two integers nx and ny and following previous argument we can see that the
wave function of each eigenstate takes the form
~
Ψ(~r) = ei(kx rx +ky ry ) u(~r) = eik·~r u(~r) (11)
2πnx 2πny
where u(~r) is invariant under translation in both directions and kx = Nx ax
, ky = Ny ay gives the
momentum of the electron in x and y directions (when multiplied with ~).
This result can be generalized to other lattices in 2D and 3D and is summarized by Bloch’s theorem:
The energy eigenstates for an electron in a crystal takes the form

~
Ψ(~r) = eik·~r u(~r) (12)
where u(~r) is a periodic function with the same period as the lattice, ~~k is the crystal momentum of
the electron and ~k takes value in a so-called ’Brillouin zone’. In 1D, the Brilliouin zone is [− πa , πa ].
In 2D with square lattice, the Brillouin zone is a rectangle region between [− aπx , aπx ] in the kx
direction and [− aπy , aπy ] in the ky direction. For more complicated lattices, the Brillouin zone takes
more complicated shapes.
Ψ(~r) of this form is called a Bloch wave.
Each eigenstate will have a corresponding energy given by the Hamiltonian. Imagine that we move
along the k axis in the 1D Brillouin zone. As k is densely populated in the Brillouin zone, we
expect the energy to change with k in a smooth way. If we plot energy versus k, we get a smooth
line, forming a so called energy band. Moreover, there can be multiple states with the same k but
different energy. That means we can have multiple bands in the crystal. If we plot them all out,
we get the band structure of the crystal. This can be done for 2D and 3D lattice as well, although
it is harder to draw. The band structure is the most important notion in solid state physics.
6.4 Perturbation and degeneracy splitting
Now suppose that the Hamiltonian of the system is perturbed by a term V . How does the spectrum
change? In particular, we want to know if the degeneracy structure of the system change. This
4
Figure 1: An example of a 1D band structure.
will depend on the symmetry property of V . The short answer is: if V has the same symmetry
as H, then the degeneracy required by symmetry is all preserved while the accidental degeneracies
may be lifted (removed); on the other hand, if V breaks some of the symmetries of H, then it is
possible to lift non-accidental degeneracies as well. (Recall that non-accidental degeneracies exist
among different basis states of the same irrep while accidental degeneracies exist among different
irreps (which may or may not be equivalent).)
This can be shown by considering the symmetry of the perturbed Hamiltonian H 0 = H + V . If

V has the same symmetry as H, then so does H 0 . Our previous argument regarding the eigen
spectrum of H still applies. Therefore, inequivalent irreps of D(g) remain separate under H 0 . It
is possible that V mix equivalent irreps which originally belong to different eigen sectors of H.
This has two consequences. First, it can change the linear combination of the irreps that form
eigen sectors of the Hamiltonian; Secondly, it can change the eigenvalues of the sectors and thereby
remove or introduce accidental degeneracy. However, the degeneracy required by symmetry with
higher dimensional irreps remain intact.
On the other hand, if V has a smaller symmetry (or different symmetry) than H, then the symmetry
of H 0 will generally be a subgroup of that of H (G0 ⊂ G). In this case, some higher dimensional
irreps of G can break down to the direct sum of multiple irreps of G0 , therefore the degeneracy
is not protected by symmetry any more and can be lifted. (Can you think of an example of this
kind?)
5
6.6 Crystal tensor properties
Let’s consider some popular lattices and find out their symmetry.
The square lattice is invariant under the following transformations: a. rotation by π/2 (yellow
arrow around the yellow dot rotation center); b. reflection (with respect to the blue axises); c.
translation by unit distance in x or y directions (red arrows).
The triangular lattice is invariant under the following transformations: a. rotation by π/3 around
lattice points and rotation by 2π/3 around plaquette center; b. reflection (with respect to the blue
axises); c. translation by unit distance in the two directions specified by red arrows.
The cubic lattice in 3D is invariant under: a. translation by unit distance in i, j, k directions (red
arrows); b. reflection with respect to the ij, jk, ki planes; c. rotation around i, j, k axis by π/2;
d. rotation around the diagonal axis of the cube by 2π/3; e. inversion (i → −i, j → −j, k → −k)
centered at any lattice site.
The symmetry transformation of an infinite lattice falls into two types: point group symmetry,
transformations which keep at least one point fixed; translation symmetry, transformation which
1
moves every point in the same direction by the same amount. Put together, they form the full
space group symmetry of the lattice.
According to their symmetries, 2D lattices are classified into five Bravais lattices, including oblique,
rectangular, centered rectangular, hexagonal, and square lattices. 3D lattices are classified into 14
Bravais lattices, including for example primitive cubic, body-centered cubic, face-centered cubic,
hexagonal lattice, etc.
Moreover, each point in the lattice can have a structure of its own instead of just being a rotationally
invariant ball. For example, each point in the lattice can host a molecule with internal structures.
This will in general reduce the symmetry of the system and further distinguish among the lattices.
Taking this into consideration, there are 230 different lattices in 3D.
6.7 Constrains on macroscopic measurements
The macroscopic properties of a crystal is constrained by the underlying lattice symmetry. To

understand how such a constraint works, we need to distinguish between scalar properties, vector
properties and tensor properties.
Some properties of a crystal is just described by a number. For example, the mass, temperature,
specific heat or free energy of the system.
Some properties of a crystal is a three dimensional vector. For example, the electric polarization P
with three components (Px , Py , Pz ), and the current density (Jx , Jy , Jz ). The vectors have a single
index, labeling the three dimensions of space. We say that a vector is a tensor of rank 1.
Some properties are described by tensors, i.e. quantities with more than one index. For example,
the conductivity of a material is in general a two-index tensor. Usually, we think of conductivity as
a scalar which measures the proportionality constant between current density J and applied electric
field E. Both J and E are three dimensional vectors. Conductivity defined as the ratio between two
vectors is a scalar only when J and E points in the same direction and their ratio is independent
of the direction. But this is not necessarily true in a material. There are materials whose induced
current can lie in a different direction from the applied electric field. Then to describe conductivity,
we need to specify the proportionality constant between current density in every direction (x, y,
2
z) and applied electric field in every direction (x, y, z). Therefore, the conductivity becomes a two
index tensor
σij = Ji /Ej , i, j = x, y, z (1)
In the most general case, all nine entries in the tensor can be nonzero. Tensors with two indices
are of rank 2.
Some other useful example of two index tensor properties are the stress and strain. The strain tensor
describes the deformation (change in shape) of a body with respect to its original configuration.
Suppose that the original position of each particle is given by (rx , ry , rz ) and after deformation
each particle moves by (δx , δy , δz ). The displacement of each particle can be different, therefore,
(δx , δy , δz ) is in general a function of (rx , ry , rz ). If (δx , δy , δz ) is independent of (rx , ry , rz ), then
the deformation amounts to a global translation of the body and we are not interested in that. We
are interested in the case where every point has different displacement and hence the whole body
deforms. Therefore, the strain tensor is defined as
∂δi
ij = , i, j = x, y, z (2)
∂rj
The stress tensor on the other hand, is defined as the force acting in i direction on a unit surface
in the j direction
∂Fi
σij = , i, j = x, y, z (3)
∂Sj
Combining these tensors, we can get tensors of even higher rank. For example, stress can induce
electric polarization in piezoelectric materials. When the stress is small, the induced polarization
has a linear relation to the stress. Their proportionality constant, called the piezoelectric modulus,
is a rank three tensor and is defined as
∂Pi
dijk = , i, j, k = x, y, z (4)
∂σjk
Similarly, in an elastic material under small stress, strain and stress have a linear relation and their
proportionality constant, called the Young’s modulus, is a rank four tensor and is defined as
∂σij
λijkl = , i, j, k, l = x, y, z. (5)
∂kl
As these properties depends on the coordinate system x, y, z, if we rotate the coordinate system,
they should transform accordingly. In particular, in a quantum mechanical system at a thermal
equilibrium state of temperature T , the measured quantity is given by
< O >= Tr(e−H/kB T O) (6)
If the Hamiltonian of the system is invariant under certain symmetry, D(g)H = HD(g), then the
measured quantity should transform according to how O transforms.
In particular, suppose that we do atransformation Qij on the coordinate system. Writing Q

−1 0 0
in matrix form, for inversion, Q =  0 −1 0 ; for reflection across the y − z plane Q =
0 0 −1
   
−1 0 0 cos θ − sin θ 0
 0 1 0; for rotation around z axis, Q =  sin θ cos θ 0.
0 0 1 0 0 1
3
Under this transformation, the scalar property remains invariant. The vector properties transform
as X
D(Q)Pi D−1 (Q) = Q−1
ii0 Pi
0 (7)
i0
Note that D(Q) acts on the Hilbert space of the physical system while Q is a three dimensional
linear transformation acting on the spatial indices.
The rank two tensor properties transform as

X
0
σij = Q−1 −1
ii0 Qjj 0 σi j
0 0 (8)
i0 j 0
The rank three tensor properties transform as

X
d0ijk = Q−1 −1 −1
ii0 Qjj 0 Qkk0 di j k
0 0 0 (9)
i0 j 0 k 0
so on and so forth.
Now if the system has certain symmetry, it remains invariant under certain transformations of the
coordinate systems. Therefore, all the tensor properties should remain invariant. This puts a strong
constrain on which component of the tensor can be nonzero. Consider the following examples.
1. Vector property in systems with inversion symmetry.
Suppose
 that the
 system has certain vector property Pi , i = x, y, z. Under inversion Q =
−1 0 0
 0 −1 0 , P → −P . Therefore, systems with inversion symmetry, like cubic lattice, must
0 0 −1
have vanishing vector properties. Similarly, in systems with inversion symmetry, all tensor proper-
ties of odd rank must vanish. On the other hand, inversion symmetry does not constrain even rank
tensors in any way.
2. Vector property in systems with rotation symmetry.
Suppose that the system has certain vector property Pi , i = x, y, z. Under rotation the vector will
be rotated to a different direction unless it points along the rotation axis. Therefore, in systems
with rotation symmetry around a single axis, like the hexagonal lattice or the tetragonal lattice, it is
possible to have nonzero vector property (along the axis), while in systems with rotation symmetry
around multiple axes, like the cubic lattice, all the vector properties have to be zero.
3. Rank two tensor property in cubic lattice.
Suppose that we have a rank two tensor property. Let’s try to figure out how many independent
degrees of freedom there are of this property in a cubic lattice. A rank two tensor contains nine
entries, so originally there are nine degrees of freedom. A large class of these tensors, including
stress and strain, are symmetric (under transpose). That is,
σij = σji (10)
This is due to physics considerations, not symmetry, and it reduces the number of DOF to six. We
are left with σ11 , σ12 ,σ13 , σ22 , σ23 ,σ33 . Now let’s use the symmetry properties of the cubic lattice
to further reduce the number of DOF.
4
Inversion symmetry of the cubic lattice does not affect rank two tensors, but reflection and rotation
does. Take reflection across x − y plane for example. The transformation is
 
1 0 0
Q = Q−1 = 0 1 0  (11)
0 0 −1
Under this transformation

σ13 → −σ13 , σ23 → −σ23 (12)
while the other components remain invariant. Therefore, due to reflection symmetry across x − y
plane, σ13 = σ23 = 0. Similarly, using reflection symmetry across y − z plane, we get σ12 = 0.
Therefore, we are only left with the diagonal elements σ11 , σ22 , σ33 .
Now we use the rotation symmetry around the diagonal axis of the cube. Under a 2π/3 rotation
in this direction, x, y, z axes are cyclicly permuted. Therefore,
σ11 → σ22 → σ33 (13)
and have to be equal. That is, we can conclude that any rank two tensor property in a cubic lattice
has to be diagonal and the diagonal elements have to be equal.
5
7 Continuous Group
Now we are going to move on to continuous groups. We have seen the simplest example of a
continuous group, the circle group. Let’s first review how that works and see how the idea can be
generalized to more complicated groups.
7.1 SO(2)
Instead of saying “the circle group”, we are going to call it by a more popular name: the SO(2)
group. It is a matrix group of two dimensional orthogonal matrices with +1 determinant. It
represents rotation of a two dimensional vector space and is represented on this two dimensional
space as
cos θ − sin θ
R(θ) = (1)
sin θ cos θ
θ ∈ [0, 2π) and the group elements compose as
R(θ1 )R(θ2 ) = R(θ1 + θ2 mod 2π) (2)
Notice that here we are using a particular representation to define the group, but the group is
a more general abstract notion. In particular the group can have other kinds of representations.
This 2D representation is special though in that it is faithful. Other representations may not be
faithful. It is a slight abuse of terminology to call the group SO(2), but in most cases it should be
clear enough whether we are talking about the abstract group or this particular two dimensional
representation.
The continuity of the group elements comes from the continuity of the parameter θ. Moreover, the
group has a nice property called compact, which roughly means that the parameter takes value in
the bounded region of [0, 2π).
This is an abelian group and the2D representation

actually decomposes into two 1D irreps through
1 i
unitary transformation S = √12
1 −i
iθ
−1 e 0
SR(θ)S = (3)
0 e−iθ
Of course there are an infinite number of irreps given by {einθ }, n ∈ Z.
Because all irreps are 1D, the character of the representation is just given by the irrep itself.
χ(n) = einθ , θ ∈ [0, 2π) (4)
These characters satisfy an orthogonality condition similar to the finite group case. However,
instead of summing over individual group elements, we need to perform an integration over them.
Z 2π
(n0 ) 1 0
(n)
hχ , χ i = dθe−inθ ein θ = δnn0 (5)
2π 0
1
Notice that I have chosen a normalization for the inner product of characters so that each character
have length 1.
We can use this orthogonality condition of characters in the same way as we have used it for finite
groups. For example, we can check that the 2D rep given above decomposes into two 1D irreps.
The character of the 2D irrep
χ = 2 cos θ = eiθ + e−iθ (6)
Therefore
R(θ) = D(1) (θ) ⊕ D(−1) (θ) (7)
as we have seen above.
The direct product of irreps goes as

0 0
D(n) ⊗ D(n ) = D(n+n ) (8)
Therefore, under direct product, the irreps form a group which is isomorphic to the group of
integers.
For finite groups, a useful notion is the generator of the group. Once we have identified the
generators of a group and the relations between them, we knows which group it is. For continuous
group, can we similarly find such generators? For the SO(2) group, intuition says that the generator
of the group is an infinitesimal rotation by a very small angle θ. But of course, no θ is small enough,
we can always find a smaller one. What we define instead, is an infinitesimal generator
dR(θ)
X=i (9)
dθ θ=0

for any representation R. Any group elements in the continuous group can then be obtained by
taking the exponential of this infinitesimal generator.
R(θ) = e−iθX (10)

(−iθX)k
The exponential of an operator is defined as e−iθX = ∞
P
k=0 k! . If we can diagonalize X into
V XV −1 = D, where D = diag(d1 , d2 , ...), then e −iθX −1
= V diag(e −iθd 1 , e−iθd2 , ...)V .
For the irrep labeled by (n) inθ (n) = −n. For the 2D orthogonal representation,
n, D (θ) =e , X
cos θ − sin θ 0 −i
R(θ) = ,X= . Note that X is Hermitian because R is unitary.
sin θ cos θ i 0
In physics, the Hermitian generator X is sometimes identified as the orbital angular momentum
Jz around the rotation axis z (e.g. for electron orbits around a nucleus). Suppose that a wave
function forms an irrep of the SO(2) group. That is,
R(θ)|ψi = e−inθ |ψi (11)
The state is said to have orbital angular momentum Jz = n. In other situations, X maybe identified
with the number of particles N in the system (e.g. for electrons in metals or insulators) and in this
particular state N = n.
2
7.2 SO(3)
This is the group of three dimensional orthogonal matrices with +1 determinant. This set of
matrices describe rotation of a three dimensional real vector space. Even though it is just one
dimension up from SO(2), it is much more complicated but also much more interesting!
First, SO(3) is not abelian any more. Imagine we perform a rotation around the z axis first by an
angle θ, the transformation of the 3D vector space is given by
 
cos θ − sin θ 0
Rz (θ) =  sin θ cos θ 0 (12)
0 0 1
Next let’s perform a rotation around x axis by an angle θ0 , the transformation of the 3D vector
space is given by  
1 0 0
Rx (θ0 ) = 0 cos θ0 − sin θ0  (13)
0 sin θ0 cos θ0
Direct calculation shows that Rz (θ)Rx (θ0 ) 6= Rx (θ0 )Rz (θ) for general θ and θ0 . Similar to the case of
SO(2), these three dimensional matrices provide one particular representation of the SO(3) group,
but the group may have other representations. This three dimensional representation is special
in that it is faithful and irreducible. As a nonabelian group, SO(3) can have higher dimensional
irreps, which we are going to discuss later.
Infinitesimal generators
First, let us try to understand what are the infinitesimal generators of SO(3). Following the
discussion of SO(2), it is easy to see that to generate rotation around z axis, the infinitesimal
generator is  
0 −i 0
X3 =  i 0 0 (14)
0 0 0
such that Rz (θ) = e−iθX3 . Similarly, we find that
   
0 0 0 0 0 i
X1 = 0 0 −i , X2 =  0 0 0 (15)
0 i 0 −i 0 0
such that Rx (θ) = e−iθX1 , Ry (θ) = e−iθX2 . X1 , X2 , X3 each generate a subgroup of rotation around
x, y and z axes respectively.
Are X1 , X2 and X3 enough to generate all SO(3) transformations?
Euler’s rotation theorem says: any transformation in SO(3) is equivalent to a single rotation about
some axis for a certain angle. It can be shown (Jones page 102-103) that such rotation operation
can be written in the form
R~n (θ) = e−iθ(nx X1 +ny X2 +nz X3 ) = e−iθX~n (16)
3
Therefore, the linear combination of X1 , X2 and X3 gives the infinitesimal generator of all trans-
formations in SO(3). The inverse of this operation corresponds to rotation around the same axis
but with opposite angle R~n−1 (θ) = eiθX~n .
In quantum mechanics, X1 , X2 and X3 correspond to angular momentum operator Jx , Jy , Jz and

their linear combination X~n = nx X1 + ny X2 + nz X3 corresponds to angular momentum in the ~n
direction.
The vector space of linear combinations of X1 , X2 and X3 leads to the most important concept in
describing the set of continuous groups we are interested in.
SO(3) (and also SO(2)) is an example of a Lie group. Its infinitesimal generators form a Lie
algebra. The Lie algebra is usually denoted with lower case letters of the name of the group. For
example, the Lie algebra of the SO(2) group is so(2) and the Lie algebra of SO(3) is so(3).
There are two important structures of this algebra:
(1) it is a (real) vector space. That is, the linear combination of two infinitesimal generators is
(linearly proportional to) an infinitesimal generator;
(2) The commutator of two infinitesimal generators [Xi , Xj ] = Xi Xj −Xj Xi is (linearly proportional
to) an infinitesimal generator.
Comment:
1. By focusing on the infinitesimal generators, we reduce the study of a continuous group with an
infinite number of elements to the study of a finite set, the basis of the Lie algebra.
2. In SO(3), X1 , X2 , X3 form the basis of the vector space. We have shown above that (1) is true
for SO(3). Let’s now see that (2) is also true. First [Xi , Xi ] = 0 which is the infinitesimal generator
for doing nothing because eiθ0 = I.
[X1 , X2 ] = iX3 , [X2 , X3 ] = iX1 , [X3 , X1 ] = iX2 (17)
From which we can show that for any two linear combinations of X1 , X2 , X3 , we have
~ ~nb · X]
[~na · X, ~ = i(~na × ~nb ) · X
~ (18)
P
3. Notice that because in general Xi ’s do not commute, eiθ ni Xi iθni Xi .
Q
i 6= ie
4. The commutator can be thought of as a composition rule between the infinitesimal generators,
mapping two such generators to a third one. This composition rule is anti-commuting, [Xi , Xj ] =
−[Xj , Xi ]. It satisfies the Jacobi Identity
[[X, Y ], Z] + [[Y, Z], X] + [[Z, X], Y ] = 0 (19)
5. The commutator between the infinitesimal generators is very useful in determining the conjugacy
classes of the group. For SO(3) we are going to find that rotation operations around different axes
with the same angle are conjugate to each other. That is,
~ ~ ~ ~
eiφ~n2 ·X eiθ~n1 ·X e−iφ~n2 ·X = eiθ~n3 ·X (20)
4
~ ~
Physically this is very intuitive. We can imagine that eiφ~n2 ·X and e−iφ~n2 ·X maps the vector ~n1 to ~n3
and back. Then under this mapping, the rotation around axis ~n1 for angle θ is mapped to rotation
around axis ~n3 for angle θ.
We are not going to derive this result in class, but we are going to work it out in the homework.
6. The conjugacy classes of SO(3) then consist of rotation through the same angle about different
axes and can be labelled simply by that angle θ. Correspondingly, characters are just a function of
θ. In the three dimensional special orthogonal representation, we have
χ = 2 cos(θ) + 1 (21)
which can be verified directly for Rx (θ), Ry (θ) and Rz (θ).
Irreducible representations
Now let’s see what irreps the group SO(3) has. There is an infinite number of them, as you might
have expected from the continuous nature of this group. Instead of trying to find irreps of the group,
we can just find irreps of the infinitesimal generators (the algebra). Then by taking exponentiation,
we can recover the group elements.
That is, we are looking for irreducible representations of X1 , X2 , X3 such that
1. X1 , X2 , X3 are finite dimensional Hermitian operators.
2. they satisfy the relation [Xa , Xb ] = iabc Xc , where abc = 1 if {ijk} can be obtained from {xyz}
by a cyclic permutation, abc = −1 if {abc} can be obtained from {xyz} by a cyclic permutation
and an exchange and abc = 0 otherwise.
3. once exponentiated, they give rise to the SO(3) group. (This requirement may seem redundant,
if the previous two are satisfied. But in fact it is not, as we are going to see later on.)
In physics, this exercise is called ‘finding the orbit of an electron in a Hydrogen atom’. Rotation
invariance around z axis in the Hydrogen atom implies that every orbit is labeled by a particular
value of angular momentum Jz in the z direction. If we take into account the full rotation symmetry
of the Hydrogen atom in three dimensional space, then the orbits should be labelled by irreps of
the SO(3) group, not just the SO(2) group, and they transform under SO(3) rotation as an irrep.
5
8 Continuous Group
8.2 SO(3)
Irreducible representations
Now let’s see what irreps the group SO(3) has. There is an infinite number of them, as you might
have expected from the continuous nature of this group. Instead of trying to find irreps of the
group, we can just find irreps of the infinitesimal generators (the algebra). Then by taking the
exponetial, we can recover the group elements.
That is, we are looking for irreducible representations of X1 , X2 , X3 such that
1. X1 , X2 , X3 are finite dimensional Hermitian operators.
2. they satisfy the relation [Xa , Xb ] = iabc Xc , where abc = 1 if {ijk} can be obtained from {xyz}
by a cyclic permutation, abc = −1 if {abc} can be obtained from {xyz} by a cyclic permutation
and an exchange and abc = 0 otherwise.
3. once exponentiated, they give rise to the SO(3) group. (This requirement may seem redundant,
if the previous two are satisfied. But in fact it is not, as we are going to see later on.)
In physics, this exercise is called ‘finding the orbit of an electron in a Hydrogen atom’. Rotation
invariance around z axis in the Hydrogen atom implies that every orbit is labeled by a particular
value of angular momentum Jz in the z direction. If we take into account the full rotation symmetry
of the Hydrogen atom in three dimensional space, then the orbits should be labelled by irreps of
the SO(3) group, not just the SO(2) group, and they transform under SO(3) rotation as an irrep.
So let’s follow the physicists notation. We write Jx , Jy and Jz for X1 , X2 , X3 and look for their
irreps.
The first thing to define for an irrep is the ‘Casimir Operator’: An operator which commutes with
all the elements of a Lie group is said to be a ‘Casimir Operator’ of that group.
Comments:
1. According to Schur’s Lemma, the Casimir Operator is proportional to identity on the irrep.
2. An equivalent requirement is that, the Casimir Operator commute with all infinitesimal gener-
ators.
3. For SO(3), the Casimir Operator is denoted as J 2 and one can check that it can be obtained
from J 2 = Jx2 + Jy2 + Jz2 . For the 3D special orthogonal representation, J 2 = 2I3 . Physically, it has
the meaning of the total magnitude of the orbital angular momentum of the electron. All vectors
of an irrep are eigenvectors of J 2 with the same eigenvalue. This eigenvalue provides a one to one
1
labelling of the equivalence class of irreps of SO(3).
Previously when we talked about irreps, we always talked about the equivalence class of them
without choosing a particular basis and hence a particular form of the representation matrices.
With SO(3), physicist prefer to use a special basis and we are going to write everything down using
this basis. Recall that Jx , Jy and Jz do not commute. Therefore, we cannot find a common basis
for all three of them. Instead, we just use the eigenstates of Jz as the basis to write down irreps.
That is, the basis states are common eigenvectors of J 2 and Jz . This is possible because J 2 and Jz
commute.
So what are the eigenstates of Jz in a finite dimensional irrep? Suppose that state |mi is one such
eigenstate satisfying
Jz |mi = m|mi (1)
Then under rotation around z axis, this state |mi acquires a phase factor
e−iθJz |mi = e−iθm |mi (2)
Because θ = 2π rotation is the same as the identity transformation, we have e−i2πm = 1. That is,
the eigenvalues of Jz are integers. Similarly the eigenvalues of Jx and Jy are also integers.
In a finite dimensional representation, m has an upper bound. Let’s suppose that this maximum
value is j ∈ Z+ (or j = 0). From here, we can derive the whole representation as follows.
Starting from |ji, we can go to all other eigenstates of Jz by applying Jx and Jy . In particular,
define
J± = Jx ± iJy (3)
as the raising and lowering operators. J± † = J∓ . They earned these names because applying J±
on |mi maps it to |m ± 1i.
Jz (J± |mi) = Jz (Jx ± iJy )|mi = (Jx Jz + iJy ± iJy Jz ± Jx )|mi = (m ± 1)J± |mi (4)
Now if we apply J+ to |ji the resulting state should have 0 norm because we assumed that |ji is
already the eigenvector with the largest Jz eigenvalue.
J+ |ji = (Jx + iJy )|ji = 0 (5)
Because of this, we can see that |ji is an eigenstate of J 2 with eigenvalue j(j + 1) because
J 2 |ji = (Jz2 + Jx2 + Jy2 )|ji = (Jz2 + J− J+ + Jz )|ji = j(j + 1)|ji (6)
Because we know that J 2 is proportional to identity in a particular irrep. Therefore, in this irrep,
all states (the |mis and their superpositions) are eigenstates of J 2 with eigenvalue j(j + 1).
Now let’s apply J− to |ji and obtain all other eigenstates of Jz with smaller eigenvalues.
J− |ji ∝ |j − 1i, J− |j − 1i ∝ |j − 2i, ... (7)
There are two questions we need to answer regarding this procedure: 1. what is the normalization
of states J− |mi? 2. when does this procedure stop? That is, there is a minimum m in every finite
dimensional representation. What is this minimum m given j?
To answer this question, we observe that
hm|J+ J− |mi = hm|J 2 − Jz2 + Jz |mi = j(j + 1) − m(m − 1) (8)
2
Therefore, p
J− |mi = j(j + 1) − m(m − 1)|m − 1i (9)
and this procedure stops when j(j + 1) − m(m − 1) = 0, which happens if m = j + 1 or m = −j.
Because we know that m ≤ j, therefore, the only solution is actually m = −j, which is the smallest
eigenvalue of Jz in this irrep. Similar calculation shows
p
J+ |mi = j(j + 1) − m(m + 1)|m + 1i (10)
In this way, we have found a representation of the infinitesimal generators of SO(3) on a 2j + 1

dimensional vector space with basis vectors
|ji, |j − 1i, ..., | − ji (11)
which transforms under J+ , J− , Jz as

p p
Jz |mi = m|mi, J+ |mi = j(j + 1) − m(m + 1)|m + 1i, J− |mi = j(j + 1) − m(m − 1)|m − 1i
(12)
Correspondingly
Jx |mi = 12 ( pj(j + 1) − m(m + 1)|m + 1i + pj(j + 1) − m(m − 1)|m − 1i),

p p
1 (13)
Jy |mi = 2i ( j(j + 1) − m(m + 1)|m + 1i − j(j + 1) − m(m − 1)|m − 1i),
2
Jz |mi = m|mi, J |mi = j(j + 1)|mi
Let’s see how this looks like in some simple cases.
First, consider the case of j = 0. This irrep is one dimensional and the operators are all represented
as numbers
Jz = 0, Jx = 0, Jy = 0, J 2 = 0 (14)
If we exponentiate them, the rotation operators we get are all trivial
R~n (θ) = e−iθ(nx Jx +ny Jy +nz Jz ) = 1 (15)
Next, let’s move on to the case of j = 1. This irrep is three dimensional with basis states |1i, |0i,
| − 1i. In matrix form, Jx , Jy , Jz and J 2 read
   √   √   
1 0 0 √0 2 √0 √0 − 2 0
√ 2 0 0
1 i
Jz = 0 0 0  , Jx =  2 √0 2 , Jy =  2 √0 − 2 , J 2 = 0 2 0 
2 2
0 0 −1 0 2 0 0 2 0 0 0 2
(16)
We know of another three dimensional representation of SO(3) which is given by the three di-
mensional special orthogonal matrices. How are these two representations related? If we list the
generators of the special orthogonal matrices, we can see that
     
0 0 0 0 0 i 0 −i 0
X1 = 0 0 −i , X2 =  0 0 0 , X3 =  i 0 0 (17)
0 i 0 −i 0 0 0 0 0
3
This is related to the Jx , Jy , Jz given above by a basis transformation
 
1 −i √0
1 
S=√ 0 0 2 (18)
2 1 i 0
Therefore, the two three dimensional representations we have seen so far, are equivalent to each
other.
Similarly, we can build up representations of five, seven, ... dimensions. Each of them correspond to
a different irrep of SO(3). That is, SO(3) has one equivalence class of irrep in every odd dimension,
labeled by integer j.
Characters
For an irrep labeled by j as derived above, we can find the character of a conjugacy class labeled
by θ by taking the trace of Rzj (θ). In particular
Jz = diag(j, j − 1, ..., −j) (19)
Therefore
Rzj (θ) = diag(eijθ , ei(j−1)θ , ..., e−ijθ ) (20)
and the character is
sin(j + 1/2)θ
χ(j) (θ) = (21)
sin(θ/2)
(take j = 1 and compare it to the case of the three dimensional special orthogonal representation.)
Are these characters orthogonal to each other? In order to answer the question, we need to define
an integration for the group, which integrates over its parameter space. The parameter space of
SO(3) is highly nontrivial. First, we can specify every group element by a rotation axis direction
~n and an angle θ. ~n is a unit vector and we can take it to correspond to points on the surface of
a unit ball. θ takes value from −π to π and we can take it to correspond to the radial direction
of the solid ball. However, the parameter space is not the ball because a π rotation is the same
as a −π rotation. Therefore, the two ends of the same diameter should be identified. Therefore, a
solid ball with this identification gives the parameter space of SO(3). The geometry and topology
of this space is too complicated to discuss here. Instead I will just claim that the integration we
want to use is Z 2π
dθ
(1 − cos(θ)) (22)
0 2π
and the characters are orthogonal under this integration
sin(j + 1/2)θ sin(j 0 + 1/2)θ

Z 2π
(j) (j 0 ) dθ
hχ , χ i = (1 − cos(θ)) = δjj 0 (23)
0 2π sin(θ/2) sin(θ/2)
4
7 Continuous Group
7.2 SU (2): the special unitary matrices of dimension two
Before I move on to talk about how the irreps of SO(3) combine with each other (in direct product),
I would like to digress and talk about SU (2) first. SU (2) is very similar to SO(3) but also different
in very important ways. It turns out that the irreps of SO(3) is a subset of irreps of SU (2) and
when physicist study ‘addition of angular momentum’, what they really do is to study the direct
product of irreps of SU (2) instead of SO(3). So let’s first understand what SU (2) is.
Instead of starting from the definition of SU (2), let’s start by considering the three Pauli matrices

1 0 1 1 0 −i 1 1 0
σx = , σy = , σz = (1)
2 1 0 2 i 0 2 0 −1
It is easy to check that (1) they are Hermitian finite dimensional matrices (2) they satisfy the com-
mutation rule [σa , σb ] = iabc σc . It seems that they fulfill the requirement of being the infinitesimal
generator of SO(3). Actually, not quite. You may notice that once exponentiated, they do not
quite give rise to the SO(3) group. In particular, consider the 2π rotation around a particular axis,
say z
Rz(1/2) (2π) = e−i2πσz = −I2 (2)
That is, 2π rotation is not exactly doing nothing. Instead it adds a global phase factor of −1.
Therefore, the group generated by σx , σy , and σz is not quite the SO(3) group, in which doing 2π
rotation should be the same as doing nothing. Instead, it generates the SU (2) group: the group of
special unitary matrices of dimension two.
Let’s make linear superpositions of the infinitesimal generators and take their exponential.
R~n (θ) = e−iθσ~n = e−iθ(nx σx +ny σy +nz σz ) (3)
Because (σ~n )2 = (nx σx + ny σy + nz σz )2 = I2 /4, we have

∞
(−iθσ~n )k X (−iθ/2)k X (−iθ/2)k
X θ θ
R~n (θ) = = + 2σ~n = cos − i sin 2σ~n (4)
k! k! k! 2 2
k=0 even k odd k
This represents all possible 2 × 2 unitary matrices with determinant 1. (homework)
Although we started from three generators with the same commutation relation as those for SO(3),
the major difference between SU (2) and SO(3) is that the parameter takes value in [0, 4π), not
[0, 2π). Only when θ = 4π does R~n (θ) equal identity.
While R~n (2π) is not equal to identity, it is proportional to identity. Therefore, it commutes with
all other group elements and generates the center of the group (recall the definition of the center),
which is a C2 group. As the center of a group is a normal subgroup as well, we can take the quotient
1
of SU (2) with respect to this C2 group and we recover the SO(3) group as the quotient group. We
say that SU (2) is a double cover of SO(3).
You may wonder why we care about SU (2) so much in physics. As it turns out, a very important
property of electrons (and other fundamental particles) is their internal spin. This is not related
to the orbital motion of the electron around a nucleus. Instead, it is something intrinsic to the
electron. People realized that electron spin lives in a two dimensional Hilbert space. We can choose
the basis state of this two dimensional Hilbert space as the eigenstates of σz .

1 1 1 1 1 1
σz
= , σz −
= − − (5)
2 2 2 2 2 2
The raising and lowering operator maps between the two

0 1 0 0 1 1 1 1
σ+ = σx + iσy = , σ− = σx − iσy = , σ+ −
=
, σ− = − (6)
0 0 1 0 2 2 2 2
This is exactly the same relation as those given in the previous lecture if we set j = 21 . Therefore,
this spin 1/2 behaves in every way like its integer angular momentum cousins, with one difference.
Under spatial rotation around axis ~n through angle θ, it transforms as
R~n (θ) = e−iθσ~n = e−iθ(nx σx +ny σy +nz σz ) (7)
which does not form a representation of the SO(3) group but the SU (2) group. This is a spe-
cial property of quantum mechanics. That is, we can have quantum mechanical wave functions
transforming under symmetry operations up to a phase factor. Here the phase factor shows up as
R~n (2π) = −I. This is ok in quantum mechanics because global phase factor is not measurable. It
is in some sense a redundancy of the wave function representation, but through this example we
can see that this redundancy is absolutely crucial because without it, there cannot be a spin 1/2
representation of rotation symmetry!
In fact, SU (2) has other irreducible representations as well. All the irreps of SO(3) are irreps of
SU (2) as well, even though they are not quite faithful. Moreover, SU (2) has one irrep in every
even dimension which can be obtained in exactly the same way as the odd dimensional irreps for
SO(3) but starting from half integer j, j = 1/2, 3/2, .... Of course, only the odd dimension irreps
are irreps of SO(3). The even dimension ones are irreps of SO(3) only up to a phase factor, and
we say that they are projective representations of SO(3).
In physics, people often mix the notion of SO(3) and SU (2). It happens because in quantum
mechanics both projective and nonprojective representations are allowed and they both show up in
physical situations (like electron spin and orbital angular momentum) and can interact with each
other.
7.3 Clebsch-Gordon Coefficients
Now let’s take all the irreps of SU (2) (or all the nonprojective and projective irreps of SO(3)) and
see how they interact with each other. In particular, we want to know if we take the direct product
between two irreps labeled by j and j 0 (both can be either integer or half integer), how does the
composite representation decompose into a direct sum of irreducible blocks? In physics, this is
called the ‘addition of angular momentum’ and it was known that
D(j1 ) ⊗ D(j2 ) = ⊕jj=|j
1 +j2
1 −j2 |
D(j) (8)
2
That is, the addition of angular momentum j1 with angular momentum j2 gives rise to angular
momentum from |j1 − j2 | to j1 + j2 .
One comment on terminology. When we talk about addition of angular momentum, we are actually
taking the direct product of the corresponding irreps. We say that their angular momentum add,
because the angular momentum operator of the direct product representation is the sum of the
angular momentum operator of each of the irrep.
(j ) (j ) 1 2 1 2
D~n 1 (θ) ⊗ D~n 2 (θ) = eiθJ~n ⊗ eiθJ~n = eiθ(J~n ⊗Ij2 (j2 +1) +Ij1 (j1 +1) ⊗J~n ) (9)
Therefore, J~ntot = J~n1 ⊗ Ij2 (j2 +1) + Ij1 (j1 +1) ⊗ J~n2 , hence the name ‘addition’. Note that J~n1 ⊗ Ij2 (j2 +1)
and Ij1 (j1 +1) ⊗ J~n2 commute, therefore we can simply add them when multiplying their exponential.
Usually we just use the short-hand notation J~n1 for J~n1 ⊗ Ij2 (j2 +1) and J~n2 for Ij1 (j1 +1) ⊗ J~n2 . Hence
we have the relation
Jxtot = Jx1 + Jx2 , Jytot = Jy1 + Jy2 , Jztot = Jz1 + Jz2 (10)
Now let’s show the relation in Eq. 8 using the character of the irreps. Suppose that j1 ≥ j2 . The
character of the direct product of D(j1 ) and D(j2 ) is
sin(j1 + 1/2)θ sin(j2 + 1/2)θ

χ(j1 ) (θ)χ(j2 ) (θ) = (11)
sin(θ/2) sin(θ/2)
Using equivalent expressions for χ(j1 ) (θ) and χ(j2 ) (θ), we get
ei(j1 +1/2)θ −e−i(j1 +1/2)θ Pj2

χ(j1 ) (θ)χ(j2 ) (θ) = 2i sin(θ/2) m=−j2 e
imθ
1 P j 2 i(j1 +m+1/2)θ − e−i(j1 −m+1/2)θ

(12)
= 2i sin(θ/2) m=−j2 e
Because the summation over m is over −j2 to j2 , we can change m to −m in the second term so that
j1 and m always appear as a combination j = j1 + m. Therefore, we can replace the summation as
over j = j1 − j2 to j = j1 + j2 .
1 Pj1 +j2 i(j+1/2)θ − e−i(j+1/2)θ
χ(j1 ) (θ)χ(j2 ) (θ) = 2i sin(θ/2) j=j1 −j2 e
Pj1 +j2 sin(j+1/2)θ Pj1 +j2 (j)
(13)
= j=j1 −j2 sin(θ/2) = j=j1 −j2 χ (θ)
In physics, people are not only interested in what D(j) are contained in the direct product of
D(j1 ) ⊗ D(j2 ) , but also how the basis states of D(j) can be obtained from the basis states of D(j1 )
and D(j2 ) . As we have completely fixed the choice of basis for each of the irrep, there is a concrete
process to obtain this relation.
Let’s consider the simplest example of the direct product of two j = 1/2 irreps D(1/2) ⊗ D(1/2) .
Each D(1/2) is two dimensional and the direct product of two copies of them is four dimensional
with basis states
1 1 1 1 1 1 1 1
2 2 , 2 − 2 , − 2 2 , − 2 − 2 (14)

the number in the |i labels the angular momentum m in z direction. In order to make it explicit
that these states belong to the j = 1/2 irrep, we label the states as |j1 , m1 ; j2 , m2 i

1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
2 2 2 2 , 2, 2; 2, −2 , 2, −2; 2, 2 , 2, −2; 2, −2
, ; , (15)
3
According to the previous discussion
D(1/2) ⊗ D(1/2) = D(0) + D(1) (16)
D(0) is one dimensional and has one basis state with angular momentum 0. D(1) is three dimensional
and has three basis states with z direction angular momentum 1, 0 and −1 respectively. To specify
that they comes from the composition of two angular momentum 1/2 irreps, we label the state as
|(j1 , j2 )j, mi
1 1 1 1 1 1 1 1
2 2 0, 0 , 2 , 2 1, 1 , 2 , 2 1, 0 , 2 , 2 1, −1
, (17)
The basis states listed in Eq. 14 and those listed in Eq. 17 are the basis states of the same four
dimensional Hilbert space. Therefore, they are related by a unitary transformation. In particular,
each state in Eq. 17 can be decomposed into a linear superposition of the basis states given in
Eq. 14 and we want to find out the coefficient of these linear superpositions.
Let’s start from the state 12 , 21 1, 1 . This state is special in that it has the largest angular

momentum m = 1 in z direction among the four states in Eq. 17. Because
Jztot = Jz1 + Jz2 (18)
we have
m = m1 + m2 (19)
1 1 1 1
Among the four states listed in Eq. 14, only one satisfies m1 + m2 = 1: 2 , 2 ; 2 , 2 . Therefore,
these two states must be the same up to a phase factor. In physics, the convention is to choose the
phase factor to be 1. That is
1 1 1 1 1 1
,
2 2 1, 1 = 2 , 2 ; 2 , 2 (20)

Starting from here, we can obtain all other states in D(1) by applying the lowering operator for the
total angular momentum
tot
J− = Jxtot − iJytot = Jx1 + Jx2 − iJy1 − iJy2 = J−
1 2
+ J− (21)
We find
√ 1 1

tot 1 1 2 1 1 1 1
1
1 1 1 1 1 1 1 1
2 ,
1, 0 = J− , 1, 1 = J− + J− , ; , = ,− ; ,
+ , ; ,−

2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
(22)
That is
1 1 1 1 1 1 1 1 1 1 1
2 2 1, 0 = √2 2 , − 2 ; 2 , 2 + 2 , 2 ; 2 , − 2
, (23)
We can see that this relation is consistent because m = m1 + m2 = 0.
If we apply the lowering operator again, we get

√ 1 1 tot 1 , 1 1, 0 = J 1 + J 2 √1 1 , − 1 ; 1 , 1 + 1 , 1 ; 1 , − 1
√ 1 1 1 1
2 2 , 2 1, −1 = J− 2 2 − − 2 2 2 2 2 2 2 2 2 = 2 2, −2; 2, −2
(24)
That is
1 1 1 1 1 1
2 2 1, −1 = 2 , − 2 ; 2 , − 2
, (25)
which can also be obtained by comparing m with m1 + m2 in the basis states.
4
Now we have found all the basis states for D(1) , we are left with only the basis state for D(0) . In
order
1 1 to determine its decomposition in the basis states of Eq. 14, we only need to observe that
, 0, 0 is a state with m = 0, therefore, it has to be equal to some kind of superposition
2 2
of 12 , 12 ; 12 , − 21 and 12 , − 21 ; 12 , 21 which both have m1 + m2 = 0. We have already seen one such

superposition which gave rise to

1 1 1 1 1 1 1 1 1 1 1
,
2 2 1, 0 = √ , − ; , + , ; ,− (26)
2 2 2 2 2 2 2 2 2
As 12 , 12 0, 0 is orthogonal to 12 , 21 1, 0 , it can only be

1 1 1 1 1 1 1 1 1 1 1
2 2 0, 0 = √2 2 , − 2 ; 2 , 2 − 2 , 2 ; 2 , − 2
, (27)
Now let’s put our findings in a table
j1 = 12 , j2 = 12 m1 = 21 , m2 = 1
2 m1 = − 12 , m2 = 1
2 m1 = 21 , m2 = − 12 m1 = − 12 , m2 = − 12
j = 1, m = 1 1 0 0 0
j = 1, m = 0 0 √1 √1 0
2 2
j = 1, m = −1 0 0 0 1
j = 0, m = 0 0 √1 − √12 0
2
The numbers in this table is called the Clebsch-Gordon coefficient or CG coefficient for short. It is
the coefficient in the decomposition of total angular momentum basis in terms of the tensor product
of component angular momentum basis. The CG coefficient of other j1 , j2 and j can be obtained
in a very similar way.
5
9 Electron orbits in atoms
Now let’s see how our understanding of the irreps of SO(3) (SU (2)) can help us understand the
structure of electron orbits in atoms.
9.1 Central potential
Consider an electron moving around a nucleus. The Hamiltonian of the electron is
P2
H= + V (r) (1)
2m
V (r) is the Coulomb potential of the electron in the field of the nucleus. The crucial property of
this central potential is that it depends only on r, the distance of the electron from the nucleus,
and not on the direction of the electron. Therefore, the Hamiltonian is invariant under rotation in
full three dimensional space. That is
R~n (θ)HR~n (−θ) = H (2)
The wave function of the electron is a (normalized) complex function over space ψ(x, y, z). In
spherical coordinates, it is written as ψ(r, θ, φ), where r is the radial distance from the origin, θ is
the angle between the vector and the positive z direction, φ is the angle between the projection on
the xy plane and the positive x direction. It lives in an infinite dimensional Hilbert space. In this
Hilbert space, the angular momentum operators take the form

∂ ∂ ∂ ∂ ∂
Jx = i~ sin φ + cot θ cos φ , Jy = i~ − cos φ + cot θ sin φ , Jz = −i~ (3)
∂θ ∂φ ∂θ ∂φ ∂φ
Because of this, we can conclude that the the eigenstates of H (the electron orbitals) form irreducible
representations of SO(3). That is, the eigenstates can be grouped into subsets labelled by j,
j = 0, 1, 2, .... The set labeled by j contains 2j + 1 states, which can be further labeled by m =
−j, −j + 1, ..., j. The 2j + 1 states in the same irrep have the same energy (are degenerate). The
irrep labeled by j corresponds to orbits with total angular momentum J 2 = j(j + 1). In atomic
physics, the usual notation for angular momentum is l instead of j and the orbits with l = 0, 1, 2, 3...
are called s, p, d, f... orbits respectively. The state labeled by m in the l irrep corresponds to orbits
with z direction angular momentum m.
l and m are not sufficient to uniquely label different electron orbits in atoms, as they describe only
the angular distribution of the electron wave function and does not contain any radial information.
We need one more label n which describes roughly how far away the electron is from the nucleus.
It turns out for each n, the possible choice of l is limited to be from 0 to n − 1. Therefore, we can
have orbits 1s, 2s, 2p, 3s, 3p, 3d etc.
In Hydrogen atom, the energy of the orbits depends solely on n. However, in more general atoms,
the energy of the orbits depends both on n and j. Of course, the energy does not depend on m
1
which labels different states of the same irrep, due to the rotation symmetry of the atom. The
Hydrogen atom contains a lot of ‘accidental’ degeneracy if we consider only the SO(3) symmetry.
In fact, it was realized that the Hydrogen atom has a higher SO(4) symmetry which guaran-
tees the degeneracy of different orbitals with the same n. Here is an article that explains this
http://hep.uchicago.edu/ rosner/p342/projs/weinberg.pdf.
9.2 Spin orbit interaction
If we examine the electron orbits around the Hydrogen atom really carefully, we will find that some
electron orbits (e.g. 2p) spits into two with a small energy difference. How could that happen?
People realized that this is due to the interaction between the orbital and spin degrees of freedom
of the electron. That is, the Hamiltonian contains an extra term of the form
~ · J~ = f (r) (Sx Jx + Sy Jy + Sz Jz )
∆H = f (r)S (4)
where S~ denotes the spin angular momentum of the electron and J~ denotes the orbital angular
momentum of the electron. Due to this coupling, the Hamiltonian is no longer invariant under
rotation on the orbital part or the spin part alone. Instead, it is invariant if the orbital angular
momentum and spin angular momentum rotates together so that their dot product remains the
same.
In this case, the spin angular momentum and orbital angular momentum are no longer good labels
~ 2 = s(s + 1) and J~2 = j(j + 1) still commute with the
of the eigenstates of the Hamiltonian. S
Hamiltonian but Sz and Jz do not. Instead, the total angular momentum operator L ~ = S~ + J~
commute with the Hamiltonian
~2 = S
L ~ 2 + J~2 + 2S
~ · J,
~ Lz = Sz + Jz , [L
~ 2 , H] = 0, [Lz , H] = 0 (5)
and we can use their eigenvalue to label eigenstates of the Hamiltonian. That is, due to spin
orbit interaction, the eigenstates of the Hamiltonian are no longer tensor product of eigenstates
of spin and orbital angular momentum |j, mj ; s, ms i. Instead they are eigenstates of total angular
momentum |(j, s) l, ml i, which is the result of adding spin and orbital angular momentum together.
From here we can understand how the atomic spectrum changes with spin orbit interaction. On an
s orbital, spin orbital interaction corresponds to the addition of j = 0 with s = 1/2 which results in
a single irrep l = 1/2. Therefore, spin orbit interaction does not split the degeneracy of the j = 0
orbital. On a p orbital, on the other hand, spin orbital interaction corresponds to the addition of
j = 1 with s = 1/2 which results in two irreps, one with l = 1/2 one with l = 3/2. These two
irreps have different energy (states within each irrep have the same energy). Therefore, spin orbit
interaction splits the energy of a p orbital.
If we now look at the spectral line generated by the transition of an electron between a p orbital and
an s orbital, a single line will split into a doublet. This is the fine structure of Hydrogen spectrum.
2
9.3 Transition selection rule
Now imagine we apply an external perturbation to the atom in order to drive transition of the
electron from one orbital to another. How does the transition rate depend on the form of the
perturbation? Quantum mechanics tells us that the transition amplitude is give by
Tif = hψi |O|ψf i (6)
where O is the perturbation operator and |Tif |2 gives the transition probability.
In the study of atomic spectrum, the most common perturbation is electric dipole operator E ~ · ~r
(by e.g. shining a laser on the atom). Suppose that we choose the electric field in the laser to point
in the z direction, can we drive the transition between two s orbitals? That is, we want to calculate
the probability amplitude
Tif = hψis |Ez z|ψfs i (7)
What we find is that, Tif actually has to be zero. To see this, we insert operator e−iπJx eiπJx
between the operator and the initial and final states
Tif = hψis |e−iπJx eiπJx Ez ze−iπJx eiπJx |ψfs i (8)
We combine these operators so that the initial and final state are both acted upon by the operator
eiπJx which does a π rotation around the x direction and the dipole operator in the middle gets
conjugated by eiπJx which also does the x axis π rotation.
As the initial and final states are both s orbitals, they remain invariant under this rotation
eiπJx |ψfs i = |ψfs i, eiπJx |ψis i = |ψis i (9)
The dipole operator on the other hand, gets a − sign
eiπJx Ez ze−iπJx = −Ez z (10)
Because of this extra sign

Tif = −Tif (11)
and it has to be zero.
In deriving this result, we have used only the symmetry property of the initial, final state and that
of the perturbation operator.
Here we encounter an important concept. That is, operators, just like states, can transform under
symmetry and form representations. The way symmetry acts on operators is by conjugation
O → D(g)OD−1 (g) (12)
Under this conjugation, an operator, represented as an n × n matrix, is mapped to a different

operator, represented as another n × n matrix. If we find a basis for the set of operators which
3
transform into each other under symmetry conjugation, we can decompose the initial and final
operator in this basis and find the representation of the symmetry in this basis.
In the above discussion, we are considering a C2 symmetry of the total rotation group (π rotation
around x axis). The initial and final states both form a (1, 1) representation of the C2 group. The
dipole operator, on the other hand, forms (1, −1) representation of the C2 group. The transition
probability is zero because the direct product of a (1, 1) representation (the initial state) with a
(1, −1) representation (the perturbation operator) cannot give rise to a (1, 1) representation (the
final state). Therefore, it is not possible to make transitions between two s orbitals with a electric
dipole perturbation, hence a selection rule.
If we use the full SO(3) rotation symmetry of the group, we can obtain stronger selection rules. In
particular, operators can also form j = 0, j = 1, j = 2... representations of SO(3). The transition
probability from a state with angular momentum j1 , m1 to a state with angular momentum j2 , m2
through an operator with angular momentum j, m is proportional to the CG coefficient of composing
j1 , m1 with j, m and obtaining j2 , m2 . This is the content of the so called Wigner-Eckart theorem.
j
Tensor Operator: An irreducible tensor operator Tm is a set of operators labelled by fixed integer
j and m = −j, −j + 1, ..., j which transform under rotation according to
X j j
U (R)Tmj
U (R)−1 = Dmm0 (R)Tm 0 (13)
m0
j
where U (R) = eiθJ~n represents rotation around axis ~n through angle θ and Dmm 0 is the 2j + 1
dimensional irrep of SO(3).
Examples:
(1) j = 0
The operator J 2 commute with all J~n , therefore it remains invariant under the above transformation
U (R)J 2 U (R)−1 = J 2 (14)
We say that J 2 is a scalar operator. In particular, if we regard the commutation relation [J~n , J 2 ] = 0
as the action of J~n on J 2 , we see that J 2 forms a one dimensional vector space where all the
generators of SO(3) acts as 0. Therefore, all the group elements of SO(3) acts as 1 and J 2 forms
the j = 0 singlet representation of SO(3).
Similarly r2 = x2 + y 2 + z 2 which measures the distance of a particle from the origin is also a scalar
operator. This is easy to understand intuitively because the distance of the particle from the origin
does not change if we rotate the system. To see how the math works more explicitly, notice that
in quantum mechanics the angular momentum operator is given by
J~ = ~r × p~ = (ypz − zpy , zpx − xpz , xpy − ypx ) (15)
Using the commutation relation between position and momentum ([x, px ] = [y, py ] = [z, pz ] = i,
otherwise they commute), we get
[r2 , Jx ] = [x2 , Jx ] + [y 2 , Jx ] + [z 2 , Jx ] = [y 2 , −zpy ] + [z 2 , ypz ] = −2iyz + 2iyz = 0 (16)
Similarly we find [r2 , Jy ] = [r2 , Jz ] = 0. Therefore, r2 is also a scalar operator, as expected.
4
(2) j = 1
The operators Jx , Jy , Jz transforms as a three dimensional vector under rotation and forms a j = 1
representation. In homework 7 problem 1, you have shown this in questions 1-4. We say that J~n is
a vector operator.
Similarly, ~r = (x, y, z) is also a vector operator. Again this is true intuitively because ~r describes
the position of the particle from the origin. To see why this is true mathematically, we need to work
out the commutation relation between ~r and J~n . It can be checked that [Ja , rb ] = iabc rc . This is
analogous to the commutation relation [Ja , Jb ] = iabc Jc . If we think of the commutation relation
as action of Ja on rb or Jb , then we can see that {rb } and {Jb } both form a three dimensional
vector space on which {Ja } acts as the generator of 3 dimensional special orthogonal matrices,
which forms the j = 1 representation of SO(3).
(3) j = 2
We will see an example of tensor operator in homework. We will not discuss it in detail here.
Wigner-Eckart Theorem
The Wigner-Eckart theorem states that: the transition amplitude from one angular momentum
state |jmi to another angular momentum state |j 0 m0 i via the application of a tensor operator TM
J
0 0
is proportional to the CG coefficient C(j Jj, m M m) and the proportionality constant depends only
on j 0 Jj and not on m0 M m.
hj 0 m0 |TM
J
|jmi = C(j 0 Jj; m0 M m)hj 0 ||T J ||ji (17)
We are not going to present the proof for the Wigner-Eckard theorem but only discuss its conse-
quences. For proof, see Jones, page 114-115.
One of the most important consequence of the Wigner-Eckard theorem is the dipole selection rules,
which tells us how electrons can move from one orbital |nlmi to another orbital |n0 l0 m0 i due to the
application of an external electric field. The electric field operator E~r is a vector operator. From
Wigner-Eckart theorem, we know that
hn0 l0 m0 |ErM |nlmi = C(l0 1l, m0 M m)hn0 l0 ||~r||nli (18)
Hence we have the selection rules: (1) δl = l0 − l = −1, 0, 1.
(2) If the electric field is pointing in the z direction, M = 0, δm = m0 − m = 0. If the electric field
is pointing in x or y direction, δm = ±1.
(Actually δl = 0 is not allowed due to spatial inversion symmetry (x → −x, y → −y, z → −z).
Can you see why?)
5
Reference: Jones, “Groups, Representations, and Physics”, Chapter 10.
9 Lorentz Group and Special Relativity
Special relativity says, physics laws should look the same for different observers in different inertial
reference frames.
In the non-relativistic setting, the coordinates of different reference frames are related by the
Euclidean transformation. In particular, if two reference frames S and S 0 coincide at t = 0 and are
moving with relative velocity ~v = (v, 0, 0), then the relation between the coordinates of an event in
the two reference frames is
t0 = t, ~r0 = ~r − ~v t x0 = x − vt, y 0 = y, z 0 = z

(1)
Under such a transformation the spatial distance between two points (at the same time) remains
invariant
(~r10 − ~r20 )2 = (~r1 − ~r2 )2 (2)
In special relativity however, we need to use the Lorentz transformation and replace the above
relation with
t0 = γ(t − xv/c2 ), x0 = γ(x − vt), y 0 = y, z 0 = z (3)
where γ = √ c . This particular transformation induced by a relative velocity is called a boost.
c2 −~v 2
The transformation may look complicated, but it is designed so that the velocity of light remains
invariant in all inertial reference frames. Suppose that we send out a light signal from the origin
at t = 0 in the x direction. In reference frame S, at a later time t, the signal has travelled to point
x = ct, y = 0, z = 0. Transformed to the reference frame of S 0 , we find that the coordinate of the
corresponding signal is r r
0 c−v 0 c−v 0
t =t , x = ct , y = 0, z 0 = 0 (4)
c+v c+v
Therefore the velocity of light in the frame of S 0 is also c.
This is just one particular example of the whole group of Lorentz transformation, which we are
going to study in detail below.
9.1 Coordinate four vector
The fact that time and space gets mixed together under Lorentz transformation motivates the
definition of the space-time four vector x̃µ , µ = 0, 1, 2, 3 for the Minkowski space
x̃0 = ct, x̃1 = x, x̃2 = y, x̃3 = z (5)
Notice that the components of the four vector all have the dimension of length.
1
Lorentz transformation leaves the ‘interval’
|x̃|2 = c2 t2 − x2 − y 2 − z 2 (6)
invariant. (This ensures that the speed of light remains invariant in all reference frames.)
We define a metric tensor, g µν such that g µν = 0 if µ 6= ν and g 00 = −g 11 = −g 22 = −g 33 = 1.

That is,  
1 0 0 0
0 −1 0 0
g=  (7)
0 0 −1 0 
0 0 0 −1
Then we can define the ‘length’ of the four vector as
|x̃|2 = x̃T gx̃ (8)

 
1 0 0
This is very different from the metric we are used to. In the usual Euclidean space, g = 0 1 0
0 0 1
is positive definite. The metric for the Minkowski space is not positive definite and will result in
some special properties of the Lorentz group.
Suppose that under a Lorentz transformation, the four vector transforms as
x̃0 = Λx̃ (9)
Then the invariance of the length of the vector requires that
|x̃0 |2 = x̃0T gx̃0 = x̃ΛT gΛx̃ = x̃T gx̃ = |x̃|2 (10)
Because this is true for all x, we have ΛT gΛ = g.
Notice that if g = I3 , this condition reduces to the orthogonality condition of three dimensional
rotation transformations which form the group SO(3). The Lorentz group can be thought of as the
group of ‘orthogonal’ transformations on a space with metric g = diag(−1, 1, 1, 1) and it is denoted
as SO(3, 1).
9.2 Lorentz Transformations
What kind of Λ satisfies the above condition?
First, any spatial rotation involving x̃1 , x̃2 , x̃3 keeps the length of the four vector invariant. There-
fore, the spatial rotation transformations ∈ SO(3) forms a subgroup of the Lorentz group. The
transformation matrices take the form
 
1 0 0 0
0
Λ~rn (θ) = 

0
 (11)
R~n (θ) 
0
2
where R~n (θ) is the three dimensional special orthogonal matrix representing the spatial rotation.
This subgroup of transformations is generated by
     
0 0 0 0 0 0 0 0 0 0 0 0
 , X2 = 0 0 0 i  , X3 = 0 0 −i 0
0 0 0 0     
X1 =  0 0 0 −i 0 0 0 0 0 i 0 0 (12)
0 0 i 0 0 −i 0 0 0 0 0 0
A different kind of Lorentz transformations which do involve time are the ‘boosts’. Boost in the x
direction gives rise to the transformation
 
γ −γv/c 0 0
−γv/c γ 0 0
Λbx (v) = 
 0
 (13)
0 1 0
0 0 0 1
ζ −ζ sinh ζ
where γ = √c2c−v2 . If we define γ = cosh ζ = e +e 2 , so that tanh ζ = cosh ζ = vc , then Λbx (v) can
be re-written as  
cosh ζ − sinh ζ 0 0
− sinh ζ cosh ζ 0 0
Λbx (ζ) =   0
 (14)
0 1 0
0 0 0 1
T
One can explicitly check that Λbx (ζ) gΛbx (ζ) = g.

The infinitesimal generator for x direction boost is

 
0 −i 0 0
dΛbx (ζ) −i 0 0 0
Y1 = i =  (15)
dζ ζ=0  0 0 0 0

0 0 0 0
By exponentiating Y1 , we can recover Λbx (ζ)
Λbx (ζ) = e−iζY1 (16)
However, notice one important difference with the generator for SO(3): ζ is not bounded. As v
approaches c, ζ approaches +∞. Therefore, the Lorentz group SO(3, 1) is not compact. This has
a series of consequence. One of them being that the finite dimensional representations of SO(3, 1)
are no longer unitary.
Similarly, boosts in y and z directions are generated by

   
0 0 −i 0 0 0 0 −i
 0 0 0 0 0 0 0 0
Y2 = 
−i 0 0 0 , Y3 =  0
   (17)
0 0 0
0 0 0 0 −i 0 0 0
Boosts in an arbitrary direction ~n can be obtained first by rotating ~n to x axis, applying the boost
in x direction and then rotating back.
In general, an arbitrary Lorentz transformation contains both spatial rotation and boost. The
whole group is generated from X1 , X2 , X3 and Y1 , Y2 , Y3 . The Lie algebra of SO(3, 1) is a six
dimensional real vector space with commutators
[Xa , Xb ] = iabc Xc , [Xa , Yb ] = iabc Yc , [Ya , Yb ] = −iabc Xc (18)
3
Comments:
(1) While X1 , X2 , X3 is closed under commutation, Y1 , Y2 , Y3 are not. Therefore, the boosts do
not form a subgroup.
(2) While X1 , X2 , X3 are Hermitian, Y1 , Y2 , Y3 are anti-Hermitian. Therefore, the representation

is not unitary (the boost transformation is not unitary).
(3) The first and second commutation relation says that X1 , X2 , X3 transform as a vector under
SO(3), so do Y1 , Y2 , Y3 .
(4) We can make linear combinations between X and Y

1
Xa(±) = (Xa ± iYa ) (19)
2
In terms of Xa± , the commutation relations become
(+) (−) (−)
[Xa(+) , Xb ] = iabc Xc(+) , [Xa(−) , Xb ] = iabc Xc(−) , [Xa(+) , Xb ] = 0 (20)
That is, the set of six generators break up into two subsets, such that each subset is equivalent to
the Lie algebra of SU (2) and the two subsets are independent of each other.
9.3 Irreducible representations
The four dimensional matrices Λ provides one possible representation of SO(3, 1) while the group is
abstractly defined as that satisfying the same composition rule as the Λ’s. In terms of Lie algebra,
the group is defined as that with a six dimensional Lie algebra, satisfying the commutation relation
[Xa , Xb ] = iabc Xc , [Xa , Yb ] = iabc Yc , [Ya , Yb ] = −iabc Xc (21)
or
(+) (−) (−)
[Xa(+) , Xb ] = iabc Xc(+) , [Xa(−) , Xb ] = iabc Xc(−) , [Xa(+) , Xb ] = 0 (22)
Now we can try to see what irreducible representations do SO(3, 1) have. Following our analysis
of SO(3), in order to find irreps for a Lie group, we can try to find the irreps for its Lie algebra,
but with the danger that we get the irrep of the covering group (SU (2) for SO(3)). Things work
in a very similar way for SO(3, 1).
We see that the Lie algebra of SO(3, 1) contains two SU (2) part. Therefore, its irrep can be labelled
by (j1 , j2 ), where j1 , j2 are integer or half-integer. The representation is then (2j1 + 1)(2j2 + 1)
dimensional. The generators are
Xa+ = Jaj1 ⊗ I2j2 +1 , Xa− = I2j1 +1 ⊗ Jaj2 (23)
From which we get
Xa = Jaj1 ⊗ I2j2 +1 + I2j1 +1 ⊗ Jaj2 , Ya = −i(Jaj1 ⊗ I2j2 +1 − I2j1 +1 ⊗ Jaj2 ) (24)
If we then take the exponential, we can recover the group (or its covering group).
4
Let’s see some example irreps.
(1) j1 = 0, j2 = 0
This is the trivial representation. It is one dimensional, with all the generators being 0 and all the
group elements being represented by 1. In quantum field theory, this representation is carried by a
relativistic scalar field (e.g. Higgs field).
(2) j1 = 1/2, j2 = 0
This is called a spinor representation. It is two dimensional.
X1+ = σx , X2+ = σy , X3+ = σz , X1− = 0, X2− = 0, X3− = 0 (25)
Correspondingly
X1 = σx , X2 = σy , X3 = σz , Y1 = −iσx , Y2 = −iσy , Y3 = −iσz (26)
The Lorentz transformations are then parameterized by six real numbers θ1 , θ2 , θ3 , φ1 , φ2 , φ3
~ φ) ~ ~ ~ ~ ~ ~
~ = ei(θ·X+φ·Y ) = ei(θ−iφ)·~σ
Λ(θ, (27)
~
Note that this is different from the SU (2) group which contains matrices eiθ·~σ . In fact, the set of
matrices generated are the group of special (determinant 1) linear (invertible) matrices of dimension
2, SL(2, C). SL(2, C) is a covering group of the SO(3, 1) group as 2π spatial rotation results in −I
instead of I.
In quantum field theory, this representation is carried by the Weyl fermion.
(3) j1 = 0, j2 = 1/2
This is another spinor representation with generators
X1 = σx , X2 = σy , X3 = σz , Y1 = iσx , Y2 = iσy , Y3 = iσz (28)
This is inequivalent to the previous representation because they cannot be related by a basis
transformation.
In quantum field theory, the direct sum of the j1 = 1/2, j2 = 0 representation and the j1 = 0,
j2 = 1/2 representation is carried by the Dirac fermion.
(4) j1 = 1/2, j2 = 1/2
This is a four dimensional representation. Actually, it is exactly the four dimensional representa-
tion which we used to define SO(3, 1). That is, the space time four vector transforms with this
representation.
5
10 Group Theory and Standard Model
Group theory played a big role in the development of the Standard model, which explains the origin
of all fundamental particles we see in nature. In order to understand how that works, we need to
learn about a new Lie group: SU (3).
10.1 SU (3) and more about Lie groups
SU (3) is the group of special (det U = 1) unitary (U U † = I) matrices of dimension three. What are
the generators of SU (3)? If we want three dimensional matrices X such that U = eiθX is unitary
(eigenvalues of absolute value 1), then X need to be Hermitian (real eigenvalue). Moreover, if
U has determinant 1, X has to be traceless. Therefore, the generators of SU (3) are the set of
traceless Hermitian matrices of dimension 3. Let’s count how many independent parameters we
need to characterize this set of matrices (what is the dimension of the Lie algebra). 3 × 3 complex
matrices contains 18 real parameters. If it is to be Hermitian, then the number of parameters
reduces by a half to 9. If we further impose traceless-ness, then the number of parameter reduces
to 8. Therefore, the generator of SU (3) forms an 8 dimensional vector space.
We can choose a basis for this eight dimensional vector space as

       
0 1 0 0 −i 0 1 0 0 0 0 1
λ1 = 1 0 0 , λ2 = i 0 0 , λ3 = 0 −1
     0 , λ4 = 0 0 0 (1)
0 0 0 0 0 0 0 0 0 1 0 0
       
0 0 −i 0 0 0 0 0 0 1 0 0
1
λ5 = 0 0 0  , λ6 = 0 0 1 , λ7 = 0 0 −i , λ8 = √ 0 1 0  (2)
i 0 0 0 1 0 0 i 0 3 0 0 −2
They are called the Gell-Mann matrices. Among them λ3 and λ8 are diagonal and the other six
correspond to the off-diagonal part of the Hermitian matrices. If we recall the structure of the Lie
algebra of SU (2), λ3 and λ8 are similar to Jz and others are similar to Jx and Jy .
Note that λ1 , λ2 and λ3 are exactly the Jx , Jy and Jz operators acting on the first two dimensions.
Therefore, they form a su(2) sub-algebra and generate an SU (2) subgroup of SU (3).
By convention, we define
1
Ta = λa (3)
2
Such that
1
T r(Ta Tb ) = δab (4)
2
From the Ta ’s we can determine the commutator of the Lie algebra
[Ta , Tb ] = ifabc Tc (5)
1
where fabc is called the structure constant of the group and completely antisymmetric with respect
to the exchange of any two indices. Recall for SU (2), the structure constant was abc which has a
similar antisymmetric property.
The three dimensional special unitary matrices provide only one possible representation of SU (3).
This is called the fundamental representation. How to find the other representations? We can
proceed in a similar way as SU (2). Remember that for SU (2), what we did was
1. find the Casimir operator which commute with the whole algebra and use its eigenvalue to label
different representations
2. within each representation, use the eigenstates of Jz to label different basis states
3. define the raising and lowering operators J± to map from one basis state to another and determine
the largest and smallest Jz eigenvalue given j.
Let’s try to do something similar for SU (3). First, the su(3) algebra has two Casimir operators
8
X X
C1 = Ti2 , C2 = dijk Ti Tj Tk (6)
i=1 ijk
where dijk can be obtained from the anti-commutation relation for the generators
1
{Ti , Tj } = Ti Tj + Tj Ti = δij + dijk Tk (7)
3
Direct calculation shows that C1 = 43 , C2 = 10
9 for the fundamental representation. In fact, the
value for C1 and C2 can be obtained from two integers p and q.
1 1
C1 = (p2 + q 3 + 3p + 3q + pq), C2 = (p − q)(3 + p + 2q)(3 + q + 2p) (8)
3 18
Therefore, instead of using C1 and C2 to label representations, we can use p and q and denote each
irrep as D(p, q). The fundamental representation is then labeled as D(1, 0).
Now we want to find a set of basis states for each irrep. Similar to SU (2), we can use the eigenstates
of T3 as basis states. In fact, because T8 commute with T3 , we can use their common eigenstates
as basis states of the irrep. There are no other independent generator which commutes with both.
Of course, we can make other choices of generators to define basis states, but it can be shown that
different choices are all equivalent to each other and give the same results.
In a Lie algebra, the subset of commuting hermitian generators which is as large as possible is
called the Cartan subalgebra. We can choose a basis for this subalgebra Hi such that
Hi = Hi† , [Hi , Hj ] = 0, T r(Hi Hj ) = kD δij (kD is representation and normalization dependent) (9)
The Hi , i = 1, ..., m, are called the Cartan generators and m is called the rank of the algebra.
The basis states of an irreducible representation can then be labeled by eigenvalues of Hi

Hi |p, q, µ
~ i = µi |p, q, µ
~i (10)
In the case of fundamental representation of SU (3), we have three basis states labeled by (apart
from p = 1, q = 0)
√ ! √ ! √ !
1 3 1 3 3
µ
~= , , − , , 0, − (11)
2 6 2 6 3
2
The vectors µ
~ are called the weight vectors of the algebra.
Now we can compose ‘raising’ and ‘lowering’ operators out of the other six generators of su(3).
Similar to SU (2), the raising and lowering operators can be chosen to have only one off-diagonal
element. Define
1 1 1
E±1,0 = √ (T1 ± iT2 ) , E±1/2,±√3/2 = √ (T4 ± iT5 ) , E∓1/2,±√3/2 = √ (T6 ± iT7 ) (12)
2 2 2
Their subscript comes from their commutation relation with T3 and T8 and is related to how they
change the weight vector. For example
1 1
[T3 , E±1,0 ] = √ ([T3 , T1 ] ± i[T3 , T2 ]) = (±1)E±1,0 , [T8 , E±1,0 ] = √ ([T8 , T1 ] ± i[T8 , T2 ]) = 0E±1,0
2 2
(13)
Correspondingly, application of E±1,0 to a state |~ µi maps it to state |~ µ + (±1, 0)i. The vectors
√ √
~ = (±1, 0), (±1/2, ± 3/2), (∓1/2, ± 3/2)
α (14)
are called roots of the algebra. Moreover, we can find
[Eα~ , E−~α ] = α ~
~ ·H (15)
which is a generalization of SU (2) relation [J+ , J− ] = 2Jz (if we rescale J+ , J− by √12 , we get
[J+ , J− ] = Jz ). Note that while the weight vector is specific to each representation, the root vector
is the same for all representations.
Now we can repeat the exercise of SU (2) and try to find the set of weight vectors which make up a
representation. In particular, for a particular p and q, we can start from a particular weight vector,
change it using the raising and lowering operators. By calculating how the norm of the basis state
changes under such mapping (as a function of p, q, µ ~ and α
~ ), we can find the ‘highest weight vector’
and the ‘lowest weight vector’ and hence the dimension of the representation and the action of all
generators in this basis. In this way, we can determine all irreducible representations of SU (3).
However, this calculation if too complicated and we are not going to do it explicitly.
Instead, we only mention here a few important representation of SU (3). It turns out the dimension
of a particular representation D(p, q) is
1
d(p, q) = (p + 1)(q + 1)(p + q + 2). (16)
2
1. D(0, 0)
The dimension of this representation is 1 and this is the trivial representation. Every generator
is represented as 0 and every group element is represented as 1. Still this is a very important
representation in physics and it is called the singlet representation.
2. D(1, 0)
This is the 3 dimensional representation given by the special unitary matrices. It is called the
fundamental representation. We have discussed a lot about this representation above.
3. D(0, 1)
3
This is again 3 dimensional. The infinitesimal generators are related to those in D(1, 0) by Ta0 = −Ta∗
and the group elements are related by complex conjugation.
4. D(1, 1)
This representation is 8 dimensional. This is an important representation called the adjoint repre-
sentation. The adjoint representation is a representation of a Lie group on the vector space of its Lie
algebra. The SU (3) group has 8 generators, therefore, its adjoint representation is 8 dimensional.
The adjoint representation is obtained by interpreting the commutation relation
[T̂a , Tb ] = ifabc Tc (17)
as the action of T̂a on the basis Tb of the Lie algebra, mapping them to different linear combinations
of the basis. For example, because
1 1 1 1
[T1 , T2 ] = iT3 , [T1 , T3 ] = −iT2 , [T1 , T4 ] = i T7 , [T1 , T7 ] = −i T4 , [T1 , T5 ] = −i T6 , [T1 , T6 ] = i T5
2 2 2 2
(18)
Therefore, T1 is represented by the 8 × 8 matrix
0 0 0 0 0 0 0 0
 
0
 0 −i 0 0 0 0 0

0 i 0 0 0 0 0 0
0 − 2i
 
0 0 0 0 0 0
T1adjoint =
0 i
 (19)
 0 0 0 0 2 0 0

0
 0 0 0 − 2i 0 0 0

0 0 0 2i 0 0 0 0
0 0 0 0 0 0 0 0
We can see that if we change the basis of the Lie algebra to that of the Carton operator and the
raising and lowering operator, the matrix corresponding to T3 and T8 becomes diagonal with the
diagonal vector being the root of the algebra.
4
10 Group Theory and Standard Model
10.2 Gauge Theory
Electromagnetic field
Before we present the standard model, we need to explain what a gauge symmetry is.
The idea of gauge symmetry started from the study of electromagnetism. Recall that Maxwell’s
equation reads
~ ~
~ = − ∂B , ∇ · E
~ = 0, ∇ × E
∇·B ~ = ρ, ∇×B
~ = µ0~j + 0 µ0 ∂ E (1)
∂t 0 ∂t
The first equation is automatically satisfied if
~ =∇×A
B ~ (2)
~ = 0 for any A.
because ∇ · (∇ × A) ~ Substituting in the second equation we get
!
∂ ~
A
∇× E ~+ =0 (3)
∂t
which can be automatically satisfied if

~
~ = −∇ϕ − ∂ A
E (4)
∂t
because ∇ × (∇ϕ) = 0 for any ϕ.
~ and B,
The important observation here is that: given E ~ A ~ and ϕ are not uniquely determined.
~ ~ ~
Instead, we get the same E and B if we change A and ϕ as
A ~ + ∇f, ϕ → ϕ − ∂f
~→A (5)
∂t
here f can be any real function of space and time. This is called a gauge transformation of
the electromagnetic potentials ϕ and A ~ and ϕ and A ~ are called the gauge fields. Under the
~ and B
gauge transformation, the E ~ fields remain invariant, hence the Maxwell’s equation remains
invariant. Therefore, the gauge transformation is a symmetry of the Maxwell’s equation. Note
that the gauge transformation is different from a global symmetry transformation in that the
transformation can be different at different space time point. In this sense, it is called a local
symmetry of E&M.
~ and try to solve for them,

Now , we can re-write the last two Maxwell’s equations using ϕ and A
while keeping in mind that any two solution which differ by
∂f
(ϕ0 , A
~ 0 ) = (ϕ, A)
~ + (− , ∇f ) (6)
∂t
1
are equivalent solutions for the EM field.
Recall that we saw in homework 8 that the electromagnetic field (E, ~ B)~ forms a representation
of the Lorentz symmetry. This is now easier to see if we realize that (ϕ, A) ~ forms a four vector
~ ~
representation of Lorentz symmetry. The (E, B) is then a rank two tensor representation of Lorentz
symmetry because it is obtained by taking derivative of a four vector (ϕ, A) ~ with respect to another
four vector (t, ~r). (Recall the definition of tensor from lecture 10.) Note that the gauge symmetry we
have been talking about is independent from the Lorentz symmetry and the (E, ~ B)
~ fields transform
under both.
Interaction with electrons
~ fields to study electromagnetism given that they are redun-

But why do we want to use the (ϕ, A)
~ B)
dant? Why not just use the (E, ~ fields? It may seem that (ϕ, A)~ are just objects that we made
~ ~ ~ are indeed real
up while (E, B) are the real physical fields. However, it was realized that (ϕ, A)
(even though they are redundant).
It took a lot of effort and a lot of confusion and what people realized was that we need to write
our theory with (ϕ, A) ~ instead of (E,
~ B)
~ if we are to keep the theory local. This comes from the
observation of the Aharonov-Bohm effect, where electron wave function undergoes a phase shift if
the electron moves around a solenoid (manifested as a shift of interference pattern in a double slit
experiment). A solenoid has nonzero magnetic field B ~ only on its inside and zero magnetic field on
the outside, so if we are to describe this effect in terms of the interaction of the electron with the B~
field, the interaction becomes nonlocal. On the other hand, the A ~ field is nonzero even outside the
solenoid, therefore, the AB effect can be attributed to the local interaction between the electron
and A.~ We introduced gauge redundancy in order to recover locality of the theory.
In particular, the equation of motion for an electron in an electromagnetic field is given by

 2 
~
∂ψ  p~ − eA
i~ = + eϕ ψ (7)

∂t 2m
p 2
Without the electromagnetic field, the 2m term describes the kinetic energy of the electron. The
eϕ term describes the electric potential experienced by the electron. By changing p~ to p~ − eA~ we
couple the electron to the magnetic field as well and in classical mechanics one can check that this
is the correct form for the Hamiltonian to reproduce the Lorentz force (F~ = e~v × B)~ experienced
by the electron. For quantum mechanics, we take this form of Hamiltonian and make position and
momentum into operators.
2
Therefore, the basic equation describing the motion of an electron in an electromagnetic field is writ-
~ instead of (E,
ten in terms of (ϕ, A) ~ B).
~ If ψ(~r, t) is the wave function satisfying the Schroedinger
~
equation without the A field, then the wave function satisfying the Schroedinger equation with the
~ field is
A
e ~r ~ 0
Z
0 iφ
ψ (~r, t) = e ψ(~r, t), φ = A(~r ) · d~r0 (8)
~ 0
That is, the A~ field changes the phase factor of the electron wave function which leads to the shift
of the interference pattern.
~ is redundant. How does the gauge transformation of (ϕ, A)

But (ϕ, A) ~ change the equation? If we
~ with
replace (ϕ, A)
~0 = A
A ~ + ∇f, ϕ0 = ϕ − ∂f (9)
∂t
The equation becomes
 2 
~
∂ψ  p~ − eA − e∇f ∂f 
i~ = + eϕ − e  ψ (10)
∂t 2m ∂t
It looks like a different equation, but these extra terms of gauge transformation can actually be
absorbed if we re-define
ψ 0 = eief /~ ψ (11)
where f is the real function on space time.
p − e∇f )ψ 0 = −i~eief /~ ∇ψ + e∇f eief /~ ψ − e∇f eief /~ ψ = −i~eief /~ ∇ψ

(~ (12)
therefore the right hand side becomes

 2 
~
 p~ − eA ∂f 
eief /~  + eϕ − e  ψ (13)
2m ∂t
While the left hand side becomes

∂ψ 0 ∂ψ ∂f
i~ = i~eief /~ − e eief /~ ψ (14)
∂t ∂t ∂t
3
Comparing the term we see that we get back to the original equation
 2 
p
~ − eA~
∂ψ 
i~ = + eϕ ψ (15)

∂t 2m
Therefore, the Schrodinger’s equation is invariant under the following transformation
A ~ + ∇f, ϕ0 = ϕ − ∂f , ψ 0 = eief /~ ψ
~0 = A (16)
∂t
The transformation on the wave function of ψ is by adding a phase factor that is space and time
dependent. Therefore, it is a local U (1) transformation. The (ϕ, A) ~ field is correspondingly called
a U (1) gauge field. The phase factor change to ψ is proportional to e, the charge carried by the
electron. If we have a particle with charge 2e, the phase factor involved in the transformation would
double.
The way we put EM fields into the equation for an electron is with the following changes:
∂ ∂ ∂ ∂ ~
i~ → i~ − eϕ, −i~ → −i~ − eA (17)
∂t ∂t ∂~r ∂~r
This is called minimal coupling.
Nonabelian gauge field
While the concept of a gauge field may already look highly nontrivial, we need to generalize it to
nonabelian gauge groups in order to formulate the theory underlying standard model. In particular,
we need not only U (1) gauge field, but also SU (2) and SU (3) gauge field.
Recall that a particle coupled to the U (1) gauge field transform under the gauge transformation as
ψ → einθ ψ (18)
Here we are setting constants e and ~ as 1. n is the charge carried by the particle. Or in terms of
representation theory, n labels the irrep of U (1) carried by the particle. The wave function of the
particle then transforms under the gauge transformation as the corresponding irrep labeled by n.
The group element of the transformation, labeled by θ here, is space time dependent.
If a particle is coupled to a SU (2) gauge field, it first needs to form a representation of SU (2)
labeled by j. If j > 0, then the wave function has multiple (2j + 1) components. Then under SU (2)
gauge transformation, the wave function transforms
P3
ψ → U (g(x, t))ψ = ei k=1 θk Jk ψ (19)
Here ψ is a 2j + 1 component vector, Jk ’s are the three generators of SU (2) in the irrep labeled by
j, θk ’s label the group element of the transformation and can be space time dependent.
The gauge field of SU (2) takes value in its Lie algebra and can be written as a linear combination
of the generators.
W = W k Jk (20)
(The gauge field of U (1) has only one component corresponding to its only one generator.) Note
that similar to the U (1) gauge field A, each W also has a µ label, where µ = 0, 1, 2, 3. Moreover,
these fields can have space time dependence (through the space time dependence of W k ).
4
The gauge fields then transform under the gauge transformation as
Wµ → U Wµ U −1 − i(∂µ U )U −1 (21)
Notice how this reduces to the gauge transformation for the U (1) gauge field if W and U is one
dimensional. (Here we have used the metric (−1, 1, 1, 1) so that ∂0 = −∂/∂t, ∂1 = ∂/∂x, ∂2 = ∂/∂y,
∂3 = ∂/∂z.)
Then if in the Hamiltonian the particle wave function and the gauge field couple as
∂µ → ∂µ − iWµ (22)
Then the form of the Schrodinger’s equation remains invariant (homework).
We said that the particle forms an irrep labeled by j. What representation does the gauge field
form for SU (2)? This is a strange question because the gauge field transforms under the gauge
transformation which is space time dependent, so in principle the transformation group is SU (2)
to the power of number of space time points. But if we assume we apply the same transformation
to all space time points, i.e. W k has no space time dependence, then the gauge symmetry becomes
a global symmetry and the symmetry group is SU (2). Then we can see that the gauge field
transformations as
Wµ → U Wµ U −1 (23)
which is exactly how the Lie algebra transforms under the group. Therefore, the gauge field forms
the adjoint representation of the group.
All of the discussion for SU (2) can be straight forwardly generalized to the case of SU (3). The
gauge field has then 8 components and transforms as the adjoint representation of SU (3) while the
particle coupled to it can form an arbitrary representation labeled by p, q.
10.3 Standard Model
How does all this relate to the particles and forces of the Standard Model? The standard model,
in its most commonly accepted formulation, contains a U (1) gauge field, a SU (2) gauge field,
and a SU (3) gauge field. The fundamental particles, like the quarks and the electron, couple to
these gauge fields. They transform under the gauge symmetries as certain representations of the
SU (3) × SU (2) × U (1) group.
The elementary particles in the standard model fall into two major classes: gauge bosons and
fermions. The Quarks and Leptons are fermions, with half integer spin and the Gauge bosons are
bosons, with integer spin.
5
The gauge bosons come from the quantization of the gauge fields. That is, we can take the gauge
fields, find their Fourier modes, quantize them so that each Fourier mode becomes a quantum
harmonic oscillator like degree of freedom. The gauge bosons are then the quantized excitations of
these harmonic oscillators. There is one type of gauge boson corresponding to every dimension of
the Lie algebra of the gauge group.
For SU (3), its Lie algebra is eight dimensional. Correspondingly, there are eight gauge bosons called
the ’Gluons’. For the remaining SU (2) × U (1) gauge group, the Lie algebra is four dimensional,
corresponding to one photon, two types of W boson and one type of Z boson. Note that photon
does not correspond to the Lie algebra of the U (1) part of the gauge group. Instead, the four
generators of SU (2) × U (1) mix together and form some linear combinations that correspond to
the photon and the W and Z bosons. The SU (2) × U (1) part is called the ’electroweak’ subgroup
of the total gauge group.
The other type of elementary particles are the fermions. The fermions form some spinor represen-
tation of the Lorentz group (with half integer spins) and at the same time form some representation
of the gauge group. The quarks forms a nontrivial (the fundamental) representation of SU (3). The
three components of the quark field are called ‘red’, ‘green’, and ‘blue’ – the color charge of the
quarks. The quarks hence couple to the SU (3) gauge field and can interact with each other by
exchanging gluons. The force mediated by the SU (3) gauge field is called the strong interaction.
Leptons, on the other hand, do not transform under SU (3) and therefore couple only to the
electroweak part of the gauge group and experience only electroweak interaction. Among them
the electron, muon and tau carry electromagnetic charge while the neutrinos are neutral. The
neutrinos only interact through the weak interaction.
Finally, there is the Higgs boson, which has a different origin. The Higgs boson is responsible for
the symmetry breaking of the electroweak part of the gauge group and gives mass to the W and Z
bosons.
6
Physics 129b Lecture 20 Caltech
12 Time Reversal Symmetry in Quantum Mechanics
In classical mechanics, time reversal has the following effect on physical quantities:
t(time) → −t, x(position) → x, p(momentum) → −p, E(electric field) → E, B(magnetic field) → −B

(1)
In quantum mechanics, we would expect similar relations to hold under time reversal. However,
there is a problem. Consider the canonical commutation relation between position and momentum
[x, p] = i~ (2)
If x → x, p → −p under time reversal, then this commutation relation no longer holds.
How to solve this problem?
It was realized by Wigner that in quantum mechanics, time reversal has to be defined in a very
special way different from all other symmetries. Time reversal operator is anti-unitary: it maps i
to −i. If this is the case, then the canonical commutation relation between x and p will remain
invariant under time reversal.
The action of time reversal on a wave function then contains two parts: first take complex conju-
gation in a certain basis, then apply a linear transformation in this basis. That is, we can write
T = UK (3)
where K denotes complex conjugation and U denotes some unitary transformation.
Then time reversal acts on operators as
T OT −1 = U KOKU † = U O∗ U † (4)
That is, the action of time reversal on operators contains two parts: first take complex conjugation
of the operator written in certain basis, then conjugate the operator by some unitary transformaiton
in this basis.
Let’s see what this anti-unitary property of time reversal implies.
0. Time reversal does not commute with complex numbers.
T cT −1 = c∗ (5)
1. Time reversal is not a linear operator on wave functions.
T (α1 ψ1 + α2 ψ2 ) = α1∗ T (ψ1 ) + α2∗ T (ψ2 ) (6)
Instead it is said to be anti-linear.
1
Are anti-linear operators allowed as symmetry operators in quantum mechanics? The answer is
yes, because:
2. Under time reversal, the absolute value of wave function inner product remains invariant.
To see this, notice that

T ψ1 = U ψ1∗ , T ψ2 = U ψ2∗ (7)
Therefore,
hT ψ1 |T ψ2 i = hψ1∗ |U † U |ψ2∗ i = hψ1 |ψ2 i∗ (8)
The inner product between any two wave functions does not remain invariant, but that is not a
problem because the absolute value of the inner product is and only the absolute value is experi-
mentally measurable. In fact, Wigner showed that in order for the absolute value of inner product
be invariant, the symmetry operation must either be unitary or anti-unitary.
3. Representations of time reversal symmetry.
Time reversal, as its name suggests, forms a C2 symmetry group, T 2 = 1. However, the represen-
tations of time reversal symmetry are very different from that of usual C2 symmetry groups, which
have unitary representations.
Let’s first look for one dimensional representations of time reversal symmetry. That is, we are
looking for quantum states which transform under time reversal as
T |ψi = eiθ |ψi (9)
Applying time reversal again, we get
T 2 |ψi = e−iθ (T |ψi) = e−iθ eiθ |ψi = |ψi (10)
Therefore, the ‘quantum number’ of time reversal symmetry eiθ can be any complex number of
absolute value 1. It may seem that time reversal has an infinite number of one dimensional rep-
resentations, but in fact they are all equivalent to each other. If we apply a basis transformation
|ψ 0 i = eiα |ψi, then
T |ψ 0 i = e−iα eiθ |ψi = e−i2α eiθ |ψ 0 i (11)
By tuning α, we can change from one θ to any other θ0 . Therefore, in this sense, time reversal has
only one (trivial) one dimensional representation. In fact, this is the only linear representation of
time reversal.
On the other hand, time reversal has an interesting projective representation. Consider the action
of
0 1
T = 2iσy K = K (12)
−1 0
on the Hilbert space of a spin 1/2. T applies the expected operation to all spin operators
T σx T −1 = −σx , T σy T −1 = −σy , T σz T −1 = −σz (13)
That is, it reverses the direction of spin – a magnetic dipole moment. However, this representation
is not linear
T 2 = −1 (14)
It is a projective representation of time reversal and more interestingly it has to be at least two
dimension (can you show it?). As fermions all have spin 1/2, we say that fermions transform under
time reversal as T 2 = −1.
2
13 Spontaneous Symmetry Breaking
Even when a physical system has certain symmetry, the lowest energy state may spontaneously
break that symmetry. This has a lot of implication for the low energy excitations in the system.
The simplest illustration of this idea is the Mexican hat. The shape of the hat (hence the potential
energy) is invariant under the rotation around the vertical axis. If we put a ball at the top of
the hat, that is an unstable equilibrium point and under small perturbation the ball would go
downhill (downhat) into the valley. Each point in the valley has the lowest potential energy so is a
stable equilibrium point. The ball may roll into any of these points but once it does, the rotation
symmetry of the system is broken. Exactly which point the ball rolls into depends on the details
of the initial perturbation.
Another example is the chiral molecule. Chiral molecules are ones that break mirror reflection
symmetry. In 3D space, if we choose the mirror plane to be the xz plane, then mirror reflection
operates as x → x, y → −y, z → z. It generates a Z2 group. Chiral molecules are not invariant
under such transformations, although the equation of motion – the Schrodinger’s equation for the
nucleus and electrons – is invariant. Therefore, the wave-function of a chiral molecule does not
form a 1D representation of the Z2 group. Instead, the ‘left-handed’ molecule and the ‘right-
handed’ molecule map into each other under the symmetry transformation and together form a
two dimensional reducible representation of the symmetry and the two configurations have the
same energy. The fact that chiral molecules break mirror reflection symmetry means that they
can interact with light in a different way. Suppose that we shine a light linearly polarized in the
z direction onto the molecule. The electric field in the light points in the z direction and is hence
invariant under mirror reflection. If the molecule is reflection symmetric, then the light coming out
of the molecule would also be reflection symmetric. On the other hand, if the molecule is chiral, the
outgoing light may also be chiral. That is, the outgoing light may be circularly polarized, described
by a light vector z + iy. Under mirror reflection, the light vector becomes z − iy.
The prototypical spontaneous symmetry breaking in a many-body system is a magnet. Consider a

system of classical magnetic moments. Let’s start in 1D and consider the case where the magnetic
3
moments are constrained to point either in the positive or negative z direction with the moment
being Sz = ±1. The magnetic moments interact with their neighbors: each pair has higher energy
if their moments are opposite to each other and has lower energy if their moments point in the
same direction. X
E=− Si Si+1 (15)
i
There are then two lowest energy states, one with all the moments pointing up and one with all
the moments pointing down.
Now let’s talk about symmetry. The system has a Z2 symmetry of flipping magnetic moments.
That is, the energy of the system remains invariant if we flip all the moments at the same time
Si → −Si (16)
However, each of the lowest energy states break this symmetry. Under the symmetry transforma-
tion, they map into each other and they always have the same energy.
Low energy excitations then exist in the form of a domain wall between the two states. That is,
we can have the left hand side of the system in the up state and the right hand side of the system
in the down state. Within each half system, the energy of the system is minimized, but interface
between them has a finite energy cost of 2. In a system with periodic boundary condition, there are
always an even number of domain walls. If we create a pair of domain walls, let them propagate
through the system and re-annihilate with each other, we can flip the up domain into the domain
domain and vice verse.
In two dimensions, we can have a similar model with moments restricted to the z direction and
interaction energy being X
E=− Si Si+1 (17)
i
Again the system has the Z2 symmetry of flipping moments and the two lowest energy states break
this symmetry. The difference from the 1D case is that the domain wall between two domains now
takes the form of a loop. Magnetic moments on the two sides of the loop are opposite to each other
and the energy cost of the domain wall grows with its length. Therefore, if we want to flip one
lowest energy state into the other by creating a domain wall and letting it sweep across the whole
system, the energy barrier in the whole process is proportional to the linear size of the system.
Compare to the care of 1D where the energy cost of flipping domains is only finite.
Now back to 1D and let’s consider the more interesting case where the moments are constrained to
the xy plane. S~i is now a vector and can point in any direction in the two dimensional plane. The
interaction between the moments still prefer to have them aligned and the total energy is given by
X
E=− S ~i+1
~i · S (18)
i
4
The system has a U (1) symmetry of (global) rotation around the z axis while the lowest energy
state breaks this symmetry. In the lowest energy state, the moments all point in the same direction.
The direction can be arbitrary (within the 2D plane) and different lowest energy states are mapped
into each other under the global rotation symmetry.
The low energy excitation of the system now comes in the form of a spin wave. Suppose that we
start from a lowest energy state. Instead of doing global rotation, we can apply small rotations in
a space time dependent way. If we choose the rotation angle to be
θ(i) = θ0 cos(k · i) (19)
then each nearest neighbor pair of moments are slightly mis-aligned, giving rise to a total energy
of X X
E = −|S|2 cos(θi+1 − θi ) ≈ −|S|2 [−N + θ02 (cos(k · i + 1) − cos(k · i))2 ] (20)
i i
where the approximation depends on the fact that θ0 is small. In the long wavelength limit where
k << 1, we can see that the energy difference between such spin wave states and the lowest energy
states scales as k 2 . That is, the excitation energy goes to zero in the long wavelength limit and the
spin wave has a quadratic dispersion.
A more interesting type of low energy excitation is a topological one. Suppose that in a 1D system
with periodic boundary condition, starting from the lowest energy state, we rotate the moments
with angle
2πn
θn = (21)
N
where N is the total number of moments. If we trace the direction of the moments around the
ring, we see that they rotate by 2π. This is called the winding number of the moments. It can
5
only take values of integer multiples of 2π which is a discrete set of numbers. This configuration is
fundamentally different from the spin wave one in that it cannot be smoothly deformed back to the
lowest energy state. If we smoothly deform a state, we cannot change its winding number. This
excitations is said to be topological and costs energy
1
∆E ∼ (22)
N
6
Physics 129b Homework 1 Due 01/16/20 by 5pm
0. Prove that every element in a group has a unique inverse. (Hint: the associativity condition of
the composition rule may be helpful.)
1. Which of the following define groups? For those that do not explain the reasons. For those that
do, find the identity and the inverse element.
• The set of all complex n × n matrices when the group operation is matrix multiplication.
• The set of all complex n × n matrices, with a nonzero determinant, when the group operation
is matrix multiplication.
• The set of all Hermitian n × n matrices M (M † = M ), with a nonzero determinant, when

the group operation is matrix multiplication.
• The set of all unitary n × n matrices (M † M = I) when the group operation is matrix
multiplication.
2. Find a minimal generating set of the D4 group, show that every element of the group can
be generated from this generating set.“Minimal” means that if any element is taken out of the
generating set, the set no longer generates the whole group.
3. A set of real-valued functions fi of a real variable x can also define a group if we define
multiplication as follows: given fi and fj , the product of fi ·fj is defined as function fi (fj (x)). Show
that the functions I(x) = x and A(x) = (1 − x)−1 generate a three element group. Furthermore,
including a function C(x) = x−1 generates a six-element group.
4. Which of the following groups are isomorphic to each other. Give the explicit correspondence
where an isomorphism exists:
• the complex numbers (1,i,-1,-i) with respect to multiplication;
• the integers (2,4,6,8) with respect to multiplication modulo 10;
• the permutations of four objects: no permutation (denoted as ()), exchanging object 1 with
2 (denoted as (1, 2)), exchanging object 3 with 4 (denoted as (3, 4)), exchanging 1 with 2 and
3 with 4 (denoted as (1, 2)(3, 4));
• the permutations of four objects: no permutation (denoted as ()), cyclic permutation 1 to 2

to 3 to 4 (denoted as (1, 2, 3, 4)), cyclic permutation 1 to 4 to 3 to 2 (denoted as (1, 4, 3, 2)),
exchanging 1 with 3 and 2 with 4 (denoted as (1, 3)(2, 4)).

±1 0
• the four matrices with respect to multiplication.
0 ±1
1
1. Suppose that G is a finite group of order N .
• Show that for any element g ∈ G, the order of g must be a divisor of N . (Hint: consider the
subgroup of G generated by g.)
• If the order of a group is a prime number p, show that the group has to be a cyclic group.
2. The center of a group, denoted as Z(G), is the set of elements that commute with every element
of G.
Z(G) = {a ∈ G | ag = ga, ∀g ∈ G} (1)
Consider the Dihedral group D4 (the symmetry of a square).
• What is the center Z of D4 ?
• What is the group D4 /Z?
3. Suppose that H is a subgroup of G. Show that if the set of left cosets {gi H|gi ∈ G} is the same
as the set of right cosets {Hgj |gj ∈ G}, then H is a normal subgroup satisfying gH = Hg for all
g ∈ G.
4. Consider the quaternion group Q = gp{x, y}, x4 = e, x2 = y 2 , y −1 xy = x−1 .
• What is the order of the group? (Hint: list all possible distinct compositions of powers of x
and powers of y.)
• Decompose Q into conjugacy classes.
• Find one (proper) normal subgroup of Q and the corresponding quotient group.
• Is Q isomorphic to D4 ? Explain your answer.
1
1. Consider the cyclic group Cn .
• If n is the product of two distinct primes p1 and p2 , is Cn isomorphic to Cp1 × Cp2 ? Explain
your answer.
• If n is the square of a prime p, is Cn isomorphic to Cp × Cp ? Explain your answer.
2. Consider the quaternion group Q = gp{x, y}, x4 = e, x2 = y 2 , y −1 xy = x−1 .
Find a faithful matrix representation of Q. (hint: the simplest one is two dimensional. In a faithful
representation, different group elements are mapped to different matrices.)
3. Given a normal subgroup N of a group G, a representation DG/N of the quotient group G/N
can be lifted to give a representation DG of the full group G by the following definition:
DG (g) := DG/N (gN ) (1)
That is, each element of the group is assigned the matrix DG/N of the coset to which it belongs.
• Verify that DG (g) indeed provides a representation of G, i.e. DG (g1 )DG (g2 ) = DG (g1 g2 ).
• What is the kernel of this representation?
4. Find n different one dimensional representations of the cyclic group Cn . Verify that they are
orthogonal to each other. Are there higher dimensional irreducible representations of Cn ?
1
1. Show that the character of the direct product representation equals the product of the characters
of the component representations. That is, if
D(µ×ν) = D(µ) ⊗ D(ν) (1)
then
χ(µ×ν) (g) = χ(µ) (g) × χ(ν) (g) (2)
2. Recall all the properties we have derived for the quaternion group in the previous homeworks.
a. If we are to write down the character table for the quaternion group, what is the size of the
table? why?
b. What are the dimensions of the irreps?
c. Construct the character table for the quaternion group. You can use the following steps: 1.
Write down the first row (corresponding to the trivial irrep) 2. For the rows corresponding to
nontrivial 1D irreps, directly solve for the nontrivial 1D irreps and fill in the characters 3. use the
orthogonality property to determine the character for the last row (corresponding to the 2D irrep).
Suppose that we take the direct product of two copies of the 2D irrep of the quaternion group.
d. what is the character of this direct product representation?
e. Decompose this direct product representation into irreps.
Answer these questions without using the explicit form of the 2D irrep, but only its character.
(hint: use result in problem 1.)
3. In this problem, we are going to prove the one to one correspondence between characters and
equivalence classes of representations.
a. Suppose that D is a reducible representation of G. With certain invertible transformation S, we

can put D into a block diagonal form. SD(g)S −1 = ⊕i Di (g). Show that the character of D equals
the sum of the characters of Di .
b. Suppose that D and D0 are two representations of the same dimension and they have the same
character. Use the orthogonality property of characters to show that D and D0 must decompose
into the same set of irreps.
c. Show that D and D0 are related by an invertible transformation AD(g)A−1 = D0 (g), hence are
equivalent.
4. Consider the symmetry group of a regular triangle (D3 ) as some linear transformation of the two
dimensional vector space where the triangle is embedded. When the coordinates x and y undergo
the linear transformation of the group, the quadratic form
x2 y2
a √ + bxy + c √
2 2
1
is transformed into another quadratic form with different a0 , b0 , c0 . (a0 , b0 , c0 ) as a three dimensional
vector is related to (a, b, c) by a 3 × 3 matrix. The 3 × 3 matrices obtained in this way form
a representation of the D3 group. Find the character of this representation. If the representation
is not irreducible decompose it to its irreducible representations. (You don’t need to find the
basis transformation to put the matrix into block diagonal form. Just state which irreps it should
contain.)
2
1. Coupled oscillator with translation symmetry.
Consider a system of N mass blocks coupled to each other with springs along a ring, as shown in
the following figure. The mass blocks are constrained to move along the ring. The masses and
springs are all the same. At equilibrium, all the springs are relaxed and they have the same length.
The system has discrete translation symmetry.
Now consider small oscillation around the equilibrium configuration shown in the figure.
(1) How many degrees of freedom are there in the system? What are they?
(2) Write down the matrix T which represents elementary translation transformation on the degrees
of freedom. (Hint: if the system is in equilibrium, the elementary translation maps one mass block
to the next.)
(3) What symmetry group does translation form? List all the irreps of the symmetry group.
(4) What is the character of the representation found in (2)? When decomposed into irreducible
blocks, what irreps does it contain?
(5) Find the eigenvector corresponding to each irrep. (Use the fact that the eigenvector transform
under translation as the corresponding irrep.)
(6) If springs are added to connect second nearest-neighbor blocks while preserving translation
symmetry, how do the eigenvectors change?
(7) Now take into consideration that the system is also symmetric under reflection. How do the
eigenvectors transform under the reflection symmetry?
(8) What can we conclude about the eigen-frequencies when both reflection and translation sym-
metries are preserved?
2. Coupled oscillator under perturbation
Consider the coupled oscillator as shown in the figure. The blue mass blocks all have the same
mass and can move in the 2D xy plane. The black bars represent walls. Consider small oscillations
around this equilibrium configuration.
(1) When k1 = 0, the system has full D3 symmetry. According to what we discussed in class, what
conclusion can we draw about the eigenmodes of the system?
1
(2) When k1 6= 0, reflection symmetry is broken while rotational symmetry is still preserved. How
could the eigenmodes of the system change? (Use symmetry arguments. Do not solve the equation
of motion directly)
2
1. Non-symmorphic space group
Consider the one dimensional pattern as shown in the following figure (extending from left infinity
to right infinity). All the triangles in this pattern are of the same shape.
(a) What are the translation operations that keep the pattern invariant?
(b) What are the point group operations that keep the pattern invariant?
(c) Are there symmetry transformations that cannot be obtained by composing the transformations
in (a) and (b)?
2. The point group C3v is, among other things, the symmetry group of the ammonia molecule N H3 ,
which forms a right pyramid on an equilateral triangle base, as shown below. AB = BC = CA. O
is the center of the triangle and OD is perpendicular to ∆ABC.
A B
The symmetry group is generated by a 3-fold rotation c around the axis OD and a reflection σv
with respect to plane OAD.
(a) Show that C3v ' (is isomorphic to)D3 .
(b) Show that the molecule can possess a permanent electric dipole moment P~ . What is the
direction of this electric dipole moment?
(c) A magnetic dipole moment, like a magnetic field, is an axial vector. Under reflection symmetry,
it gets an extra minus sign compared to a regular vector (such as electric dipole). For example, under
reflection with respect to the xy plane, the magnetic field (Mx , My , Mz ) goes to (−Mx , −My , Mz ).
Its transformation under rotation is the same as a regular vector. Show that the N H3 molecule
cannot possess a permanent magnetic moment.
3. The tetragonal lattice can be obtained from the cubic lattice by stretching in the z direction
1
such that the lattice constant in the z direction is not equal to that in the x and y direction.
(a) what kind of rotation symmetry does the lattice have?
(b) what kind of reflection symmetry does the lattice have?
(c) Consider a macroscopic property of the material described by a rank two tensor σij such that
σij = σji . How many independent degrees of freedom does the tensor have in a material with
tetragonal lattice symmetry?
4. The O(2) Group.
The group O(2) can be represented as a linear transformation on a two dimensional real vector
space. In addition to rotation operations R(ψ) in the x − y plane, it also contains the reflection
operation S which maps x → −x, y → y.
(1) Write down the two dimensional representation of the rotation operation and the reflection
operation. What other operations are contained in the group? (hint: R(ψ) corresponds to all 2 × 2
real orthogonal matrices with determinant 1. The reflection operation S corresponds to one 2 × 2
real orthogonal matrix with determinant −1. When we combine two matrices, their determinant
multiply.)
(2) Show that SR(ψ)S −1 = R(−ψ), so that the group is no longer abelian. What are the conjugacy
classes of this group?
(3) What is the character for each conjugacy class of the two dimensional representation obtained
above?
2
1. Conjugacy classes of SO(3)
In this problem we are going to show that the conjugacy classes of SO(3) consist of rotations around
different axes but with the same angle. That is we are going to demonstrate the relation
~ ~ ~ ~
e−iφ~n2 ·J eiθ~n1 ·J eiφ~n2 ·J = eiθ~n3 ·J (1)
where ~n3 = R~n2 (φ)~n1 . R~n2 (φ) is the three dimensional real orthogonal representation of rotation
around ~n2 through angle φ and it acts on the three dimensional real vector space of ~n.
(1) Define Syx (φ) = e−iφJx Jy eiφJx , Szx (φ) = e−iφJx Jz eiφJx . Take the derivative of Syx (φ) with respect
∂ x ∂ iφJx
to φ and show that ∂φ Sy (φ) = Szx (φ). (hint: ∂φ e = eiφJx iJx ).
∂ x
(2) Similarly show that ∂φ Sz (φ) = −Syx (φ)
(3) Combining these two equations and use the boundary condition that Syx (0) = Jy , Szx (0) = Jz
to obtain solutions Syx (φ) = cos φJy + sin φJz , Szx (φ) = cos φJz − sin φJy . That is, Jy and Jz rotate
into each other under the conjugation of e−iφJx as if they are the y and z component of a three
dimensional vector transforming under rotation around x axis through angle φ.

(4) Taking into account the fact that Sxx (φ) = e−iφJx Jx eiφJx = Jx , show that S~nx (φ) = e−iφJx ~n · J~ eiφJx =
~ where Rx (φ) is the three dimensional orthogonal matrix representing rotation around
(Rx (φ)~n) · J,
x axis through angle φ and it acts on the three dimensional vector ~n.
Now convince yourself that this relation works not only for x axis rotation but for rotation around
−iφ~ · ~ ~ ~ ~ You don’t need to show
arbitrary axis as well. That is, e n 2 J ~n1 · J eiφ~n2 ·J = (R~n2 (φ)~n1 ) · J.
work for this part, but if you want to work everything out, it is helpful to choose three vectors ~n2 ,
~na2 , ~nb2 which are orthogonal to each other and build up the relation from there.
(5) Now use the above relation and show that

~ ~ ~ ~
e−iφ~n2 ·J eiθ~n1 ·J eiφ~n2 ·J = eiθ~n3 ·J (2)
~
where ~n3 = R~n2 (φ)~n1 . (hint: decompose eiθ~n1 ·J as a polynomial series.)
2. The SU (2) Group
In this problem, you will show that

1 0 1 1 0 −i 1 1 0
σx = , σy = , σz = (3)
2 1 0 2 i 0 2 0 −1
are the infinitesimal generators of the group of special unitary matrices of dimension two.

a b
(1) For a two dimension complex matrix , if it is unitary and has determinant 1, what
c d
conditions do a, b, c, d have to satisfy?
(2) Find a parameterization for all special unitary matrices of dimension two.
1
(3) Now take a linear combination of σx , σy and σz as σ~n = nx σx + ny σy + nz σz , where ~n =
(nx , ny , nz ) is a real unit vector. Show that (σ~n )2 = 14 I2 .
(4) Show that

iθσ~n θ θ
R~n (θ) = e = cos + i2 sin σ~n (4)
2 2
Hint: use the Taylor expansion of eiθσ~n .
(5) Show that the set of matrices you obtain in (2) coincide with the set of matrices you obtain in
(4).
(6) Show that σx , σy and σz satisfy the same commutation relation as the infinitesimal generators
Jx , Jy and Jz for SO(3).
(7) Is the group SU (2) isomorphic to SO(3)?
3. P orbital
The p orbital of an electron has three components px , py , pz . The angular part of the wave function
in these orbitals are given by
ψx (θ, φ) = N sin θ cos φ, ψy (θ, φ) = N sin θ sin φ, ψz (θ, φ) = N cos θ (5)
where θ and φ are the usual polar coordinates and N is some pre-factor independent of θ and φ.
(1) Show that the three wave functions areR orthogonal to each other. Hint: inner product of wave
π R 2π ∗
functions in polar coordinates is given by θ=0 φ=0 ψ (θ, φ)ψ̃(θ, φ) sin θdθdφ
∂
(2) The angular momentum operator in z direction is given by Jz = −i ∂φ . How is it represented
as a three dimensional matrix in the Hilbert space spanned by ψx , ψy and ψz ?
(3) Show that if the electron moves on a spherical shell of radius r, then ψx , ψy and ψz are
proportional to the x, y, z coordinates of the electron respectively.
(4) Which irreducible representation do px , py and pz orbitals form under SO(3) rotation?
2
1. Spin 3/2 representation
Spin 3/2 forms a four dimensional irreducible representation of SU (2). Use basis states
Jz |3/2i = 3/2|3/2i, Jz |1/2i = 1/2|1/2i, Jz | − 1/2i = −1/2| − 1/2i, Jz | − 3/2i = −3/2| − 3/2i, (1)
Write down the angular momentum operator Jx , Jy , Jz as 4 × 4 matrices. (Use the formula in the
lecture note for Jx and Jy .)
2. Addition of angular momentum
Consider the addition of a j1 = 1/2 and a j2 = 1 angular momentum.
(1) What angular momentum component is contained in this addition? That is, if we take the
direct product of a j1 = 1/2 irrep and a j = 1 irrep and decompose the composite representation
into the direct sum of irreps, what irreps do we obtain? (Just apply the conclusion we obtained in
class. You do not need to derive it from the character of the irreps.)
(2) Determine the Clebsch-Gordon coefficient for the addition of j1 = 1/2 and j2 = 1. Your
answer may depend on some arbitrary choice of phase factors. Make the choice so that the the CG
coefficients are all real. (There is still an ambiguity of ± signs. Either choice is ok.)
3. Electric dipole as vector operator
The electric dipole operator represents the potential energy of an electron in a uniform electric field
−eE ~ ·~r = −e(Ex x + Ey y + Ez z). While the dipole operator pointing in a particular direction breaks
the SO(3) rotation symmetry, the set of all dipole operators (of the same |E|) forms a representation
of the SO(3) rotation group. Rotation is generated by the angular momentum operator J~ = ~r × p~,
where ~r = (x, y, z) is the vector operator of spatial location (we also write it as (rx , ry , rz )) and
p~ = (px , py , pz ) is the vector operator of momentum.
(1) Use the commutation relation [ra , pb ] = iδab to show that [Ja , rb ] = iabc rc .
This is similar to the commutation relation between angular momentum operators [Ja , Jb ] = iabc Jc .
Using this commutation relation between angular momentum operators, we showed in homework
7 problem 1 that J~ rotates as a three dimensional vector under rotation generated by J.
~ Convince
yourself that the same conclusion applies to ~r. That is, ~r rotates as a three dimensional vector
under rotation generated by J.~ You do not need to show work for this part.
(2) In fact we can interpret the commutation relation [Ja , rb ] = iabc rc as the action of Ja on
rb . Consider the three dimensional operator space with basis operators rx , ry and rz . Ja can be
interpreted as a linear transformation in this space. Write down the action of Jx , Jy , Jz as three
dimensional matrices in this operator space.
(3) Which dipole operator remains invariant under rotation around z axis?
(4) Which dipole operator remains invariant under rotation around z axis up to a phase factor?
(You may need to use complex Ex , Ey , Ez .)
1
(5) Which irrep of SO(3) does the operator space form?
4. Electric quadrupole selection rule
The electric quadrupole operators contain a set of nine operators defined as
Dij = 3ri rj − r2 δij (2)
where ri , i = 1, 2, 3, are the x, y and z spatial coordinates of the electron. Notice that these nine
operators are not independent. In particular,
D11 + D22 + D33 = 0, D12 = D21 , D23 = D32 , D31 = D13 (3)
Therefore, only five of them are independent.
(1) Define r+ = − √12 (r1 + ir2 ), r− = √1 (r1

2
− ir2 ). Denote r0 ≡ rz , J0 ≡ Jz . Calculate [Ja , rb ],
where a, b = 0, +, −.
(2) Consider the set of five quadrupole operators

2 √ √
F0 = √ r02 + r+ r− , F1 = 2r+ r0 , F−1 = 2r− r0 , F2 = r+
2 2

, F−2 = r− (4)
6
Show that by interpreting the commutator [Ja , Fb ] as the action of Ja on the five dimensional
operator space of Fb , the quadrupole operators Fb transform under SO(3) rotation as the j = 2
representation. (hint: calculate [J0 , Fb ], [J+ , Fb ], [J− , Fb ] using the relation [A, BC] = B[A, C] +
[A, B]C. You can use Mathematica to help with the computation.)
(3) If a single electron transits from one atomic orbital (labeled by nlm) to another atomic orbital
(labeled by n0 l0 m0 ) due to the application of the electric quadrupole operator, what is the selection
rule for δl = l0 − l and δm = m0 − m?
2
Physics 129b Homework 9
(You don’t need to turn in this homework. Solution to this homework has been posted.)
1. SL(2, C) as double cover of SO(3, 1)
Consider the (1/2, 0) representation of the Lie algebra of SO(3, 1).
(1) What are the generators Xa+ and Xa− which form two separate su(2) sub-algebra?
(2) What are the spatial rotation and boost generators? (Xa = Xa+ + Xa− , Ya = −i(Xa+ − Xa− ))
(3) Now take the exponential of linear combinations of Xa and Ya with real coefficients. Show that
the transformations generated in this way are special linear complex matrices of dimension two.
(hint: the Pauli matrices σx , σy and σz form a basis of all traceless 2 × 2 matrices.) Comments: It
is also true that all special linear complex matrices of dimension two can be generated in this way,
but it is more complicated to show and not required.
2. Electromagnetic tensor
The electromagnetic tensor F µν is antisymmetric. Written in matrix form

 
0 −Ex /c −Ey /c −Ez /c
Ex /c 0 −Bz By 
F =  (1)
Ey /c Bz 0 −Bx 
Ez /c −By Bx 0
Under Lorentz transformation, F transforms as
F 0 = ΛF ΛT (2)
(1) We can take out the six independent nonzero components of F : Ex /c, Ey /c, Ez /c, Bx , By ,
Bz and write the Lorentz transformation as 6 × 6 matrices. What is the form of the matrix for
spatial rotation around x, y and z direction? What is the form of the matrix for Lorentz boost
in x, y and z direction? (You can use Mathematica to help with the calculation. Also you can
do the calculation for one direction only and then guess the form of the transformation in other
directions.)
(2) What are the 6 × 6 infinitesimal generators for spatial rotation around x, y and z direction and
Lorentz boost in x, y and z direction?
(3) Is this representation of SO(3, 1) reducible? If so, which irreducible blocks does it contain?
(Find the Casimir operator of the two su(2) sub-algebras and diagonalize them.)
3. Nonabelian gauge theory
Consider a particle forming a representation j under SU (2) and coupled to the corresponding gauge
field Wµ . The Schrodinger equation for the particle is
3 2 !
∂ 1 X ∂
i ψ= −i − Ws + W0 ψ (3)
∂t 2m ∂xs
s=1
1
wk Jk .
P
Here ψ is a 2j + 1 dimensional vector and W are (2j + 1) × (2j + 1) matrices W = k
Show that the form of the equation remains invariant under transformation
ψ → U ψ, Wµ → U Wµ U −1 − i(∂µ U )U −1 (4)
P
where µ = 0, 1, 2, 3, ∂0 = −∂/∂t, ∂1 = ∂/∂x, ∂2 = ∂/∂y, ∂3 = ∂/∂z, U = ei k ak Jk .

Physics 129b Lecture 1 Group Theory

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Physics 129b Lecture 1 Group Theory

Uploaded by

Copyright:

Available Formats

Physics 129b Lecture 1 Caltech, 01/07/20

Why is it interesting to physics?

Example: reflection symmetry of water molecule H2O.

Figure 1: Y reflection symmetry.

Figure 3: Rotation symmetry.

• Group representations (Irreducible representation, properties, CG coefficient)

• Finite groups (permutation groups, cyclic groups, classification)

• Application: atomic spectra

• Application: special relativity

• Application: standard model

• Application: condensed matter

Figure 5: Scale invariance

• TA: Chi Fang (Anthony) Chen, chifang@caltech.edu; Nabhah Shah, nnshah@caltech.edu.

– E. P. Wigner, Group Theory, Academic Press, 1959.

1 Basic definition and properties of Groups

Group G Composition Law Order Remarks

Set S Composition Law Order Remarks

2.1 The cyclic group Cn

The symmetry group of rotations of a regular polygon with n directed sides.

Figure 1: Directed n-gon

Applying c for n times is the same as doing nothing and we have

2.2 The Dihedral Group Dn

The symmetry group of a regular polygon with n undirected sides.

Figure 2: Undirected n-gon

Each subset {e, bi } also forms a subgroup of Dn .

First note that their composition can be non-commutative.

In fact, cb1 = b3 , b1 c = b2 . Therefore, D3 is nonabelian.

In general, the composition rule of the elements in D3 is

Some useful relations in D3 are

2.3 Permutation group Sn

The permutation of n objects.

That is, D4 can be identified as a subgroup of S4 .

2.3 The permutation group

2.4 The group of integers Z

2.5 Circle group

The symmetry group of a directed circle.

2.6 Matrix group

A set of matrices can form a group.

(2) The identity element in the group is the identity matrix.

(3) The matrix group is in general nonabelian.

3.1 Conjugacy class

(1) every element is conjugate to itself a = eae−1 .

(2) if a is conjugate to b (a = gbg −1 ), then b is conjugate to a (b = g −1 ag).

(3) if a is conjugate to b (a = gbg −1 ), and b is conjugate to c (b = hch−1 ), then a is conjugate to c

bka = (gag −1 )ka = gaka g −1 = geg −1 = e (1)

Therefore, kb is a divisor of ka . Similarly we can show that ka is a divisor of kb . Therefore, ka = kb .

(1) Cyclic group Cn

(2) Dihedral group D3

That is, c is conjugate to c2 , and the bi ’s are conjugate to each other.

(3) Dihedral group D4

The D4 group contains 8 elements

where c is rotation by π/2, and bi is reflection across the corresponding axis.

Direct calculation shows that

cb1 c−1 = b3 , cb3 c−1 = b1 , cb2 c−1 = b4 , cb4 c−1 = b2 , bi cb−1

3 Basic concepts in group theory

(1) The identity element e forms a subgroup by itself.

Example: C2 = {e, b1 } and C3 = {e, c, c2 } are both proper subgroups of D3 .

(1) Each coset contains r distinct elements. (If h1 6= h2 , then gh1 =

The order of any subgroup of G must be a divisor of the order of G.

Normal subgroups: A subgroup H of G is said to be normal if it satisfies gHg −1 = H for any

Example: The C3 subgroup {e, c, c2 } of D3 is a normal subgroup because bi cb−1 2

(g1 H) ◦ (g2 H) := g1 g2 H (2)